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/namei.h> 10 #include <linux/writeback.h> 11 #include <linux/vmalloc.h> 12 #include <linux/posix_acl.h> 13 #include <linux/random.h> 14 15 #include "super.h" 16 #include "mds_client.h" 17 #include "cache.h" 18 #include <linux/ceph/decode.h> 19 20 /* 21 * Ceph inode operations 22 * 23 * Implement basic inode helpers (get, alloc) and inode ops (getattr, 24 * setattr, etc.), xattr helpers, and helpers for assimilating 25 * metadata returned by the MDS into our cache. 26 * 27 * Also define helpers for doing asynchronous writeback, invalidation, 28 * and truncation for the benefit of those who can't afford to block 29 * (typically because they are in the message handler path). 30 */ 31 32 static const struct inode_operations ceph_symlink_iops; 33 34 static void ceph_invalidate_work(struct work_struct *work); 35 static void ceph_writeback_work(struct work_struct *work); 36 static void ceph_vmtruncate_work(struct work_struct *work); 37 38 /* 39 * find or create an inode, given the ceph ino number 40 */ 41 static int ceph_set_ino_cb(struct inode *inode, void *data) 42 { 43 ceph_inode(inode)->i_vino = *(struct ceph_vino *)data; 44 inode->i_ino = ceph_vino_to_ino(*(struct ceph_vino *)data); 45 return 0; 46 } 47 48 struct inode *ceph_get_inode(struct super_block *sb, struct ceph_vino vino) 49 { 50 struct inode *inode; 51 ino_t t = ceph_vino_to_ino(vino); 52 53 inode = iget5_locked(sb, t, ceph_ino_compare, ceph_set_ino_cb, &vino); 54 if (inode == NULL) 55 return ERR_PTR(-ENOMEM); 56 if (inode->i_state & I_NEW) { 57 dout("get_inode created new inode %p %llx.%llx ino %llx\n", 58 inode, ceph_vinop(inode), (u64)inode->i_ino); 59 unlock_new_inode(inode); 60 } 61 62 dout("get_inode on %lu=%llx.%llx got %p\n", inode->i_ino, vino.ino, 63 vino.snap, inode); 64 return inode; 65 } 66 67 /* 68 * get/constuct snapdir inode for a given directory 69 */ 70 struct inode *ceph_get_snapdir(struct inode *parent) 71 { 72 struct ceph_vino vino = { 73 .ino = ceph_ino(parent), 74 .snap = CEPH_SNAPDIR, 75 }; 76 struct inode *inode = ceph_get_inode(parent->i_sb, vino); 77 struct ceph_inode_info *ci = ceph_inode(inode); 78 79 BUG_ON(!S_ISDIR(parent->i_mode)); 80 if (IS_ERR(inode)) 81 return inode; 82 inode->i_mode = parent->i_mode; 83 inode->i_uid = parent->i_uid; 84 inode->i_gid = parent->i_gid; 85 inode->i_op = &ceph_dir_iops; 86 inode->i_fop = &ceph_dir_fops; 87 ci->i_snap_caps = CEPH_CAP_PIN; /* so we can open */ 88 ci->i_rbytes = 0; 89 return inode; 90 } 91 92 const struct inode_operations ceph_file_iops = { 93 .permission = ceph_permission, 94 .setattr = ceph_setattr, 95 .getattr = ceph_getattr, 96 .setxattr = ceph_setxattr, 97 .getxattr = ceph_getxattr, 98 .listxattr = ceph_listxattr, 99 .removexattr = ceph_removexattr, 100 .get_acl = ceph_get_acl, 101 .set_acl = ceph_set_acl, 102 }; 103 104 105 /* 106 * We use a 'frag tree' to keep track of the MDS's directory fragments 107 * for a given inode (usually there is just a single fragment). We 108 * need to know when a child frag is delegated to a new MDS, or when 109 * it is flagged as replicated, so we can direct our requests 110 * accordingly. 111 */ 112 113 /* 114 * find/create a frag in the tree 115 */ 116 static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci, 117 u32 f) 118 { 119 struct rb_node **p; 120 struct rb_node *parent = NULL; 121 struct ceph_inode_frag *frag; 122 int c; 123 124 p = &ci->i_fragtree.rb_node; 125 while (*p) { 126 parent = *p; 127 frag = rb_entry(parent, struct ceph_inode_frag, node); 128 c = ceph_frag_compare(f, frag->frag); 129 if (c < 0) 130 p = &(*p)->rb_left; 131 else if (c > 0) 132 p = &(*p)->rb_right; 133 else 134 return frag; 135 } 136 137 frag = kmalloc(sizeof(*frag), GFP_NOFS); 138 if (!frag) { 139 pr_err("__get_or_create_frag ENOMEM on %p %llx.%llx " 140 "frag %x\n", &ci->vfs_inode, 141 ceph_vinop(&ci->vfs_inode), f); 142 return ERR_PTR(-ENOMEM); 143 } 144 frag->frag = f; 145 frag->split_by = 0; 146 frag->mds = -1; 147 frag->ndist = 0; 148 149 rb_link_node(&frag->node, parent, p); 150 rb_insert_color(&frag->node, &ci->i_fragtree); 151 152 dout("get_or_create_frag added %llx.%llx frag %x\n", 153 ceph_vinop(&ci->vfs_inode), f); 154 return frag; 155 } 156 157 /* 158 * find a specific frag @f 159 */ 160 struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f) 161 { 162 struct rb_node *n = ci->i_fragtree.rb_node; 163 164 while (n) { 165 struct ceph_inode_frag *frag = 166 rb_entry(n, struct ceph_inode_frag, node); 167 int c = ceph_frag_compare(f, frag->frag); 168 if (c < 0) 169 n = n->rb_left; 170 else if (c > 0) 171 n = n->rb_right; 172 else 173 return frag; 174 } 175 return NULL; 176 } 177 178 /* 179 * Choose frag containing the given value @v. If @pfrag is 180 * specified, copy the frag delegation info to the caller if 181 * it is present. 182 */ 183 static u32 __ceph_choose_frag(struct ceph_inode_info *ci, u32 v, 184 struct ceph_inode_frag *pfrag, int *found) 185 { 186 u32 t = ceph_frag_make(0, 0); 187 struct ceph_inode_frag *frag; 188 unsigned nway, i; 189 u32 n; 190 191 if (found) 192 *found = 0; 193 194 while (1) { 195 WARN_ON(!ceph_frag_contains_value(t, v)); 196 frag = __ceph_find_frag(ci, t); 197 if (!frag) 198 break; /* t is a leaf */ 199 if (frag->split_by == 0) { 200 if (pfrag) 201 memcpy(pfrag, frag, sizeof(*pfrag)); 202 if (found) 203 *found = 1; 204 break; 205 } 206 207 /* choose child */ 208 nway = 1 << frag->split_by; 209 dout("choose_frag(%x) %x splits by %d (%d ways)\n", v, t, 210 frag->split_by, nway); 211 for (i = 0; i < nway; i++) { 212 n = ceph_frag_make_child(t, frag->split_by, i); 213 if (ceph_frag_contains_value(n, v)) { 214 t = n; 215 break; 216 } 217 } 218 BUG_ON(i == nway); 219 } 220 dout("choose_frag(%x) = %x\n", v, t); 221 222 return t; 223 } 224 225 u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v, 226 struct ceph_inode_frag *pfrag, int *found) 227 { 228 u32 ret; 229 mutex_lock(&ci->i_fragtree_mutex); 230 ret = __ceph_choose_frag(ci, v, pfrag, found); 231 mutex_unlock(&ci->i_fragtree_mutex); 232 return ret; 233 } 234 235 /* 236 * Process dirfrag (delegation) info from the mds. Include leaf 237 * fragment in tree ONLY if ndist > 0. Otherwise, only 238 * branches/splits are included in i_fragtree) 239 */ 240 static int ceph_fill_dirfrag(struct inode *inode, 241 struct ceph_mds_reply_dirfrag *dirinfo) 242 { 243 struct ceph_inode_info *ci = ceph_inode(inode); 244 struct ceph_inode_frag *frag; 245 u32 id = le32_to_cpu(dirinfo->frag); 246 int mds = le32_to_cpu(dirinfo->auth); 247 int ndist = le32_to_cpu(dirinfo->ndist); 248 int i; 249 int err = 0; 250 251 mutex_lock(&ci->i_fragtree_mutex); 252 if (ndist == 0) { 253 /* no delegation info needed. */ 254 frag = __ceph_find_frag(ci, id); 255 if (!frag) 256 goto out; 257 if (frag->split_by == 0) { 258 /* tree leaf, remove */ 259 dout("fill_dirfrag removed %llx.%llx frag %x" 260 " (no ref)\n", ceph_vinop(inode), id); 261 rb_erase(&frag->node, &ci->i_fragtree); 262 kfree(frag); 263 } else { 264 /* tree branch, keep and clear */ 265 dout("fill_dirfrag cleared %llx.%llx frag %x" 266 " referral\n", ceph_vinop(inode), id); 267 frag->mds = -1; 268 frag->ndist = 0; 269 } 270 goto out; 271 } 272 273 274 /* find/add this frag to store mds delegation info */ 275 frag = __get_or_create_frag(ci, id); 276 if (IS_ERR(frag)) { 277 /* this is not the end of the world; we can continue 278 with bad/inaccurate delegation info */ 279 pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n", 280 ceph_vinop(inode), le32_to_cpu(dirinfo->frag)); 281 err = -ENOMEM; 282 goto out; 283 } 284 285 frag->mds = mds; 286 frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP); 287 for (i = 0; i < frag->ndist; i++) 288 frag->dist[i] = le32_to_cpu(dirinfo->dist[i]); 289 dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n", 290 ceph_vinop(inode), frag->frag, frag->ndist); 291 292 out: 293 mutex_unlock(&ci->i_fragtree_mutex); 294 return err; 295 } 296 297 static int ceph_fill_fragtree(struct inode *inode, 298 struct ceph_frag_tree_head *fragtree, 299 struct ceph_mds_reply_dirfrag *dirinfo) 300 { 301 struct ceph_inode_info *ci = ceph_inode(inode); 302 struct ceph_inode_frag *frag; 303 struct rb_node *rb_node; 304 int i; 305 u32 id, nsplits; 306 bool update = false; 307 308 mutex_lock(&ci->i_fragtree_mutex); 309 nsplits = le32_to_cpu(fragtree->nsplits); 310 if (nsplits) { 311 i = prandom_u32() % nsplits; 312 id = le32_to_cpu(fragtree->splits[i].frag); 313 if (!__ceph_find_frag(ci, id)) 314 update = true; 315 } else if (!RB_EMPTY_ROOT(&ci->i_fragtree)) { 316 rb_node = rb_first(&ci->i_fragtree); 317 frag = rb_entry(rb_node, struct ceph_inode_frag, node); 318 if (frag->frag != ceph_frag_make(0, 0) || rb_next(rb_node)) 319 update = true; 320 } 321 if (!update && dirinfo) { 322 id = le32_to_cpu(dirinfo->frag); 323 if (id != __ceph_choose_frag(ci, id, NULL, NULL)) 324 update = true; 325 } 326 if (!update) 327 goto out_unlock; 328 329 dout("fill_fragtree %llx.%llx\n", ceph_vinop(inode)); 330 rb_node = rb_first(&ci->i_fragtree); 331 for (i = 0; i < nsplits; i++) { 332 id = le32_to_cpu(fragtree->splits[i].frag); 333 frag = NULL; 334 while (rb_node) { 335 frag = rb_entry(rb_node, struct ceph_inode_frag, node); 336 if (ceph_frag_compare(frag->frag, id) >= 0) { 337 if (frag->frag != id) 338 frag = NULL; 339 else 340 rb_node = rb_next(rb_node); 341 break; 342 } 343 rb_node = rb_next(rb_node); 344 rb_erase(&frag->node, &ci->i_fragtree); 345 kfree(frag); 346 frag = NULL; 347 } 348 if (!frag) { 349 frag = __get_or_create_frag(ci, id); 350 if (IS_ERR(frag)) 351 continue; 352 } 353 frag->split_by = le32_to_cpu(fragtree->splits[i].by); 354 dout(" frag %x split by %d\n", frag->frag, frag->split_by); 355 } 356 while (rb_node) { 357 frag = rb_entry(rb_node, struct ceph_inode_frag, node); 358 rb_node = rb_next(rb_node); 359 rb_erase(&frag->node, &ci->i_fragtree); 360 kfree(frag); 361 } 362 out_unlock: 363 mutex_unlock(&ci->i_fragtree_mutex); 364 return 0; 365 } 366 367 /* 368 * initialize a newly allocated inode. 369 */ 370 struct inode *ceph_alloc_inode(struct super_block *sb) 371 { 372 struct ceph_inode_info *ci; 373 int i; 374 375 ci = kmem_cache_alloc(ceph_inode_cachep, GFP_NOFS); 376 if (!ci) 377 return NULL; 378 379 dout("alloc_inode %p\n", &ci->vfs_inode); 380 381 spin_lock_init(&ci->i_ceph_lock); 382 383 ci->i_version = 0; 384 ci->i_time_warp_seq = 0; 385 ci->i_ceph_flags = 0; 386 atomic_set(&ci->i_release_count, 1); 387 atomic_set(&ci->i_complete_count, 0); 388 ci->i_symlink = NULL; 389 390 memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout)); 391 392 ci->i_fragtree = RB_ROOT; 393 mutex_init(&ci->i_fragtree_mutex); 394 395 ci->i_xattrs.blob = NULL; 396 ci->i_xattrs.prealloc_blob = NULL; 397 ci->i_xattrs.dirty = false; 398 ci->i_xattrs.index = RB_ROOT; 399 ci->i_xattrs.count = 0; 400 ci->i_xattrs.names_size = 0; 401 ci->i_xattrs.vals_size = 0; 402 ci->i_xattrs.version = 0; 403 ci->i_xattrs.index_version = 0; 404 405 ci->i_caps = RB_ROOT; 406 ci->i_auth_cap = NULL; 407 ci->i_dirty_caps = 0; 408 ci->i_flushing_caps = 0; 409 INIT_LIST_HEAD(&ci->i_dirty_item); 410 INIT_LIST_HEAD(&ci->i_flushing_item); 411 ci->i_cap_flush_seq = 0; 412 ci->i_cap_flush_last_tid = 0; 413 memset(&ci->i_cap_flush_tid, 0, sizeof(ci->i_cap_flush_tid)); 414 init_waitqueue_head(&ci->i_cap_wq); 415 ci->i_hold_caps_min = 0; 416 ci->i_hold_caps_max = 0; 417 INIT_LIST_HEAD(&ci->i_cap_delay_list); 418 INIT_LIST_HEAD(&ci->i_cap_snaps); 419 ci->i_head_snapc = NULL; 420 ci->i_snap_caps = 0; 421 422 for (i = 0; i < CEPH_FILE_MODE_NUM; i++) 423 ci->i_nr_by_mode[i] = 0; 424 425 mutex_init(&ci->i_truncate_mutex); 426 ci->i_truncate_seq = 0; 427 ci->i_truncate_size = 0; 428 ci->i_truncate_pending = 0; 429 430 ci->i_max_size = 0; 431 ci->i_reported_size = 0; 432 ci->i_wanted_max_size = 0; 433 ci->i_requested_max_size = 0; 434 435 ci->i_pin_ref = 0; 436 ci->i_rd_ref = 0; 437 ci->i_rdcache_ref = 0; 438 ci->i_wr_ref = 0; 439 ci->i_wb_ref = 0; 440 ci->i_wrbuffer_ref = 0; 441 ci->i_wrbuffer_ref_head = 0; 442 ci->i_shared_gen = 0; 443 ci->i_rdcache_gen = 0; 444 ci->i_rdcache_revoking = 0; 445 446 INIT_LIST_HEAD(&ci->i_unsafe_writes); 447 INIT_LIST_HEAD(&ci->i_unsafe_dirops); 448 spin_lock_init(&ci->i_unsafe_lock); 449 450 ci->i_snap_realm = NULL; 451 INIT_LIST_HEAD(&ci->i_snap_realm_item); 452 INIT_LIST_HEAD(&ci->i_snap_flush_item); 453 454 INIT_WORK(&ci->i_wb_work, ceph_writeback_work); 455 INIT_WORK(&ci->i_pg_inv_work, ceph_invalidate_work); 456 457 INIT_WORK(&ci->i_vmtruncate_work, ceph_vmtruncate_work); 458 459 ceph_fscache_inode_init(ci); 460 461 return &ci->vfs_inode; 462 } 463 464 static void ceph_i_callback(struct rcu_head *head) 465 { 466 struct inode *inode = container_of(head, struct inode, i_rcu); 467 struct ceph_inode_info *ci = ceph_inode(inode); 468 469 kmem_cache_free(ceph_inode_cachep, ci); 470 } 471 472 void ceph_destroy_inode(struct inode *inode) 473 { 474 struct ceph_inode_info *ci = ceph_inode(inode); 475 struct ceph_inode_frag *frag; 476 struct rb_node *n; 477 478 dout("destroy_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode)); 479 480 ceph_fscache_unregister_inode_cookie(ci); 481 482 ceph_queue_caps_release(inode); 483 484 /* 485 * we may still have a snap_realm reference if there are stray 486 * caps in i_snap_caps. 487 */ 488 if (ci->i_snap_realm) { 489 struct ceph_mds_client *mdsc = 490 ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc; 491 struct ceph_snap_realm *realm = ci->i_snap_realm; 492 493 dout(" dropping residual ref to snap realm %p\n", realm); 494 spin_lock(&realm->inodes_with_caps_lock); 495 list_del_init(&ci->i_snap_realm_item); 496 spin_unlock(&realm->inodes_with_caps_lock); 497 ceph_put_snap_realm(mdsc, realm); 498 } 499 500 kfree(ci->i_symlink); 501 while ((n = rb_first(&ci->i_fragtree)) != NULL) { 502 frag = rb_entry(n, struct ceph_inode_frag, node); 503 rb_erase(n, &ci->i_fragtree); 504 kfree(frag); 505 } 506 507 __ceph_destroy_xattrs(ci); 508 if (ci->i_xattrs.blob) 509 ceph_buffer_put(ci->i_xattrs.blob); 510 if (ci->i_xattrs.prealloc_blob) 511 ceph_buffer_put(ci->i_xattrs.prealloc_blob); 512 513 call_rcu(&inode->i_rcu, ceph_i_callback); 514 } 515 516 int ceph_drop_inode(struct inode *inode) 517 { 518 /* 519 * Positve dentry and corresponding inode are always accompanied 520 * in MDS reply. So no need to keep inode in the cache after 521 * dropping all its aliases. 522 */ 523 return 1; 524 } 525 526 /* 527 * Helpers to fill in size, ctime, mtime, and atime. We have to be 528 * careful because either the client or MDS may have more up to date 529 * info, depending on which capabilities are held, and whether 530 * time_warp_seq or truncate_seq have increased. (Ordinarily, mtime 531 * and size are monotonically increasing, except when utimes() or 532 * truncate() increments the corresponding _seq values.) 533 */ 534 int ceph_fill_file_size(struct inode *inode, int issued, 535 u32 truncate_seq, u64 truncate_size, u64 size) 536 { 537 struct ceph_inode_info *ci = ceph_inode(inode); 538 int queue_trunc = 0; 539 540 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 || 541 (truncate_seq == ci->i_truncate_seq && size > inode->i_size)) { 542 dout("size %lld -> %llu\n", inode->i_size, size); 543 inode->i_size = size; 544 inode->i_blocks = (size + (1<<9) - 1) >> 9; 545 ci->i_reported_size = size; 546 if (truncate_seq != ci->i_truncate_seq) { 547 dout("truncate_seq %u -> %u\n", 548 ci->i_truncate_seq, truncate_seq); 549 ci->i_truncate_seq = truncate_seq; 550 551 /* the MDS should have revoked these caps */ 552 WARN_ON_ONCE(issued & (CEPH_CAP_FILE_EXCL | 553 CEPH_CAP_FILE_RD | 554 CEPH_CAP_FILE_WR | 555 CEPH_CAP_FILE_LAZYIO)); 556 /* 557 * If we hold relevant caps, or in the case where we're 558 * not the only client referencing this file and we 559 * don't hold those caps, then we need to check whether 560 * the file is either opened or mmaped 561 */ 562 if ((issued & (CEPH_CAP_FILE_CACHE| 563 CEPH_CAP_FILE_BUFFER)) || 564 mapping_mapped(inode->i_mapping) || 565 __ceph_caps_file_wanted(ci)) { 566 ci->i_truncate_pending++; 567 queue_trunc = 1; 568 } 569 } 570 } 571 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 && 572 ci->i_truncate_size != truncate_size) { 573 dout("truncate_size %lld -> %llu\n", ci->i_truncate_size, 574 truncate_size); 575 ci->i_truncate_size = truncate_size; 576 } 577 578 if (queue_trunc) 579 ceph_fscache_invalidate(inode); 580 581 return queue_trunc; 582 } 583 584 void ceph_fill_file_time(struct inode *inode, int issued, 585 u64 time_warp_seq, struct timespec *ctime, 586 struct timespec *mtime, struct timespec *atime) 587 { 588 struct ceph_inode_info *ci = ceph_inode(inode); 589 int warn = 0; 590 591 if (issued & (CEPH_CAP_FILE_EXCL| 592 CEPH_CAP_FILE_WR| 593 CEPH_CAP_FILE_BUFFER| 594 CEPH_CAP_AUTH_EXCL| 595 CEPH_CAP_XATTR_EXCL)) { 596 if (timespec_compare(ctime, &inode->i_ctime) > 0) { 597 dout("ctime %ld.%09ld -> %ld.%09ld inc w/ cap\n", 598 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec, 599 ctime->tv_sec, ctime->tv_nsec); 600 inode->i_ctime = *ctime; 601 } 602 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) { 603 /* the MDS did a utimes() */ 604 dout("mtime %ld.%09ld -> %ld.%09ld " 605 "tw %d -> %d\n", 606 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec, 607 mtime->tv_sec, mtime->tv_nsec, 608 ci->i_time_warp_seq, (int)time_warp_seq); 609 610 inode->i_mtime = *mtime; 611 inode->i_atime = *atime; 612 ci->i_time_warp_seq = time_warp_seq; 613 } else if (time_warp_seq == ci->i_time_warp_seq) { 614 /* nobody did utimes(); take the max */ 615 if (timespec_compare(mtime, &inode->i_mtime) > 0) { 616 dout("mtime %ld.%09ld -> %ld.%09ld inc\n", 617 inode->i_mtime.tv_sec, 618 inode->i_mtime.tv_nsec, 619 mtime->tv_sec, mtime->tv_nsec); 620 inode->i_mtime = *mtime; 621 } 622 if (timespec_compare(atime, &inode->i_atime) > 0) { 623 dout("atime %ld.%09ld -> %ld.%09ld inc\n", 624 inode->i_atime.tv_sec, 625 inode->i_atime.tv_nsec, 626 atime->tv_sec, atime->tv_nsec); 627 inode->i_atime = *atime; 628 } 629 } else if (issued & CEPH_CAP_FILE_EXCL) { 630 /* we did a utimes(); ignore mds values */ 631 } else { 632 warn = 1; 633 } 634 } else { 635 /* we have no write|excl caps; whatever the MDS says is true */ 636 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) { 637 inode->i_ctime = *ctime; 638 inode->i_mtime = *mtime; 639 inode->i_atime = *atime; 640 ci->i_time_warp_seq = time_warp_seq; 641 } else { 642 warn = 1; 643 } 644 } 645 if (warn) /* time_warp_seq shouldn't go backwards */ 646 dout("%p mds time_warp_seq %llu < %u\n", 647 inode, time_warp_seq, ci->i_time_warp_seq); 648 } 649 650 /* 651 * Populate an inode based on info from mds. May be called on new or 652 * existing inodes. 653 */ 654 static int fill_inode(struct inode *inode, 655 struct ceph_mds_reply_info_in *iinfo, 656 struct ceph_mds_reply_dirfrag *dirinfo, 657 struct ceph_mds_session *session, 658 unsigned long ttl_from, int cap_fmode, 659 struct ceph_cap_reservation *caps_reservation) 660 { 661 struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc; 662 struct ceph_mds_reply_inode *info = iinfo->in; 663 struct ceph_inode_info *ci = ceph_inode(inode); 664 int issued = 0, implemented, new_issued; 665 struct timespec mtime, atime, ctime; 666 struct ceph_buffer *xattr_blob = NULL; 667 struct ceph_cap *new_cap = NULL; 668 int err = 0; 669 bool wake = false; 670 bool queue_trunc = false; 671 bool new_version = false; 672 673 dout("fill_inode %p ino %llx.%llx v %llu had %llu\n", 674 inode, ceph_vinop(inode), le64_to_cpu(info->version), 675 ci->i_version); 676 677 /* prealloc new cap struct */ 678 if (info->cap.caps && ceph_snap(inode) == CEPH_NOSNAP) 679 new_cap = ceph_get_cap(mdsc, caps_reservation); 680 681 /* 682 * prealloc xattr data, if it looks like we'll need it. only 683 * if len > 4 (meaning there are actually xattrs; the first 4 684 * bytes are the xattr count). 685 */ 686 if (iinfo->xattr_len > 4) { 687 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS); 688 if (!xattr_blob) 689 pr_err("fill_inode ENOMEM xattr blob %d bytes\n", 690 iinfo->xattr_len); 691 } 692 693 spin_lock(&ci->i_ceph_lock); 694 695 /* 696 * provided version will be odd if inode value is projected, 697 * even if stable. skip the update if we have newer stable 698 * info (ours>=theirs, e.g. due to racing mds replies), unless 699 * we are getting projected (unstable) info (in which case the 700 * version is odd, and we want ours>theirs). 701 * us them 702 * 2 2 skip 703 * 3 2 skip 704 * 3 3 update 705 */ 706 if (ci->i_version == 0 || 707 ((info->cap.flags & CEPH_CAP_FLAG_AUTH) && 708 le64_to_cpu(info->version) > (ci->i_version & ~1))) 709 new_version = true; 710 711 issued = __ceph_caps_issued(ci, &implemented); 712 issued |= implemented | __ceph_caps_dirty(ci); 713 new_issued = ~issued & le32_to_cpu(info->cap.caps); 714 715 /* update inode */ 716 ci->i_version = le64_to_cpu(info->version); 717 inode->i_version++; 718 inode->i_rdev = le32_to_cpu(info->rdev); 719 inode->i_blkbits = fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1; 720 721 if ((new_version || (new_issued & CEPH_CAP_AUTH_SHARED)) && 722 (issued & CEPH_CAP_AUTH_EXCL) == 0) { 723 inode->i_mode = le32_to_cpu(info->mode); 724 inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid)); 725 inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid)); 726 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode, 727 from_kuid(&init_user_ns, inode->i_uid), 728 from_kgid(&init_user_ns, inode->i_gid)); 729 } 730 731 if ((new_version || (new_issued & CEPH_CAP_LINK_SHARED)) && 732 (issued & CEPH_CAP_LINK_EXCL) == 0) 733 set_nlink(inode, le32_to_cpu(info->nlink)); 734 735 if (new_version || (new_issued & CEPH_CAP_ANY_RD)) { 736 /* be careful with mtime, atime, size */ 737 ceph_decode_timespec(&atime, &info->atime); 738 ceph_decode_timespec(&mtime, &info->mtime); 739 ceph_decode_timespec(&ctime, &info->ctime); 740 ceph_fill_file_time(inode, issued, 741 le32_to_cpu(info->time_warp_seq), 742 &ctime, &mtime, &atime); 743 } 744 745 if (new_version || 746 (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) { 747 ci->i_layout = info->layout; 748 queue_trunc = ceph_fill_file_size(inode, issued, 749 le32_to_cpu(info->truncate_seq), 750 le64_to_cpu(info->truncate_size), 751 le64_to_cpu(info->size)); 752 /* only update max_size on auth cap */ 753 if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) && 754 ci->i_max_size != le64_to_cpu(info->max_size)) { 755 dout("max_size %lld -> %llu\n", ci->i_max_size, 756 le64_to_cpu(info->max_size)); 757 ci->i_max_size = le64_to_cpu(info->max_size); 758 } 759 } 760 761 /* xattrs */ 762 /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */ 763 if ((issued & CEPH_CAP_XATTR_EXCL) == 0 && 764 le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) { 765 if (ci->i_xattrs.blob) 766 ceph_buffer_put(ci->i_xattrs.blob); 767 ci->i_xattrs.blob = xattr_blob; 768 if (xattr_blob) 769 memcpy(ci->i_xattrs.blob->vec.iov_base, 770 iinfo->xattr_data, iinfo->xattr_len); 771 ci->i_xattrs.version = le64_to_cpu(info->xattr_version); 772 ceph_forget_all_cached_acls(inode); 773 xattr_blob = NULL; 774 } 775 776 inode->i_mapping->a_ops = &ceph_aops; 777 inode->i_mapping->backing_dev_info = 778 &ceph_sb_to_client(inode->i_sb)->backing_dev_info; 779 780 switch (inode->i_mode & S_IFMT) { 781 case S_IFIFO: 782 case S_IFBLK: 783 case S_IFCHR: 784 case S_IFSOCK: 785 init_special_inode(inode, inode->i_mode, inode->i_rdev); 786 inode->i_op = &ceph_file_iops; 787 break; 788 case S_IFREG: 789 inode->i_op = &ceph_file_iops; 790 inode->i_fop = &ceph_file_fops; 791 break; 792 case S_IFLNK: 793 inode->i_op = &ceph_symlink_iops; 794 if (!ci->i_symlink) { 795 u32 symlen = iinfo->symlink_len; 796 char *sym; 797 798 spin_unlock(&ci->i_ceph_lock); 799 800 err = -EINVAL; 801 if (WARN_ON(symlen != inode->i_size)) 802 goto out; 803 804 err = -ENOMEM; 805 sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS); 806 if (!sym) 807 goto out; 808 809 spin_lock(&ci->i_ceph_lock); 810 if (!ci->i_symlink) 811 ci->i_symlink = sym; 812 else 813 kfree(sym); /* lost a race */ 814 } 815 break; 816 case S_IFDIR: 817 inode->i_op = &ceph_dir_iops; 818 inode->i_fop = &ceph_dir_fops; 819 820 ci->i_dir_layout = iinfo->dir_layout; 821 822 ci->i_files = le64_to_cpu(info->files); 823 ci->i_subdirs = le64_to_cpu(info->subdirs); 824 ci->i_rbytes = le64_to_cpu(info->rbytes); 825 ci->i_rfiles = le64_to_cpu(info->rfiles); 826 ci->i_rsubdirs = le64_to_cpu(info->rsubdirs); 827 ceph_decode_timespec(&ci->i_rctime, &info->rctime); 828 break; 829 default: 830 pr_err("fill_inode %llx.%llx BAD mode 0%o\n", 831 ceph_vinop(inode), inode->i_mode); 832 } 833 834 /* set dir completion flag? */ 835 if (S_ISDIR(inode->i_mode) && 836 ci->i_files == 0 && ci->i_subdirs == 0 && 837 ceph_snap(inode) == CEPH_NOSNAP && 838 (le32_to_cpu(info->cap.caps) & CEPH_CAP_FILE_SHARED) && 839 (issued & CEPH_CAP_FILE_EXCL) == 0 && 840 !__ceph_dir_is_complete(ci)) { 841 dout(" marking %p complete (empty)\n", inode); 842 __ceph_dir_set_complete(ci, atomic_read(&ci->i_release_count)); 843 } 844 845 /* were we issued a capability? */ 846 if (info->cap.caps) { 847 if (ceph_snap(inode) == CEPH_NOSNAP) { 848 ceph_add_cap(inode, session, 849 le64_to_cpu(info->cap.cap_id), 850 cap_fmode, 851 le32_to_cpu(info->cap.caps), 852 le32_to_cpu(info->cap.wanted), 853 le32_to_cpu(info->cap.seq), 854 le32_to_cpu(info->cap.mseq), 855 le64_to_cpu(info->cap.realm), 856 info->cap.flags, &new_cap); 857 wake = true; 858 } else { 859 dout(" %p got snap_caps %s\n", inode, 860 ceph_cap_string(le32_to_cpu(info->cap.caps))); 861 ci->i_snap_caps |= le32_to_cpu(info->cap.caps); 862 if (cap_fmode >= 0) 863 __ceph_get_fmode(ci, cap_fmode); 864 } 865 } else if (cap_fmode >= 0) { 866 pr_warning("mds issued no caps on %llx.%llx\n", 867 ceph_vinop(inode)); 868 __ceph_get_fmode(ci, cap_fmode); 869 } 870 spin_unlock(&ci->i_ceph_lock); 871 872 if (wake) 873 wake_up_all(&ci->i_cap_wq); 874 875 /* queue truncate if we saw i_size decrease */ 876 if (queue_trunc) 877 ceph_queue_vmtruncate(inode); 878 879 /* populate frag tree */ 880 if (S_ISDIR(inode->i_mode)) 881 ceph_fill_fragtree(inode, &info->fragtree, dirinfo); 882 883 /* update delegation info? */ 884 if (dirinfo) 885 ceph_fill_dirfrag(inode, dirinfo); 886 887 err = 0; 888 out: 889 if (new_cap) 890 ceph_put_cap(mdsc, new_cap); 891 if (xattr_blob) 892 ceph_buffer_put(xattr_blob); 893 return err; 894 } 895 896 /* 897 * caller should hold session s_mutex. 898 */ 899 static void update_dentry_lease(struct dentry *dentry, 900 struct ceph_mds_reply_lease *lease, 901 struct ceph_mds_session *session, 902 unsigned long from_time) 903 { 904 struct ceph_dentry_info *di = ceph_dentry(dentry); 905 long unsigned duration = le32_to_cpu(lease->duration_ms); 906 long unsigned ttl = from_time + (duration * HZ) / 1000; 907 long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000; 908 struct inode *dir; 909 910 /* only track leases on regular dentries */ 911 if (dentry->d_op != &ceph_dentry_ops) 912 return; 913 914 spin_lock(&dentry->d_lock); 915 dout("update_dentry_lease %p duration %lu ms ttl %lu\n", 916 dentry, duration, ttl); 917 918 /* make lease_rdcache_gen match directory */ 919 dir = dentry->d_parent->d_inode; 920 di->lease_shared_gen = ceph_inode(dir)->i_shared_gen; 921 922 if (duration == 0) 923 goto out_unlock; 924 925 if (di->lease_gen == session->s_cap_gen && 926 time_before(ttl, dentry->d_time)) 927 goto out_unlock; /* we already have a newer lease. */ 928 929 if (di->lease_session && di->lease_session != session) 930 goto out_unlock; 931 932 ceph_dentry_lru_touch(dentry); 933 934 if (!di->lease_session) 935 di->lease_session = ceph_get_mds_session(session); 936 di->lease_gen = session->s_cap_gen; 937 di->lease_seq = le32_to_cpu(lease->seq); 938 di->lease_renew_after = half_ttl; 939 di->lease_renew_from = 0; 940 dentry->d_time = ttl; 941 out_unlock: 942 spin_unlock(&dentry->d_lock); 943 return; 944 } 945 946 /* 947 * splice a dentry to an inode. 948 * caller must hold directory i_mutex for this to be safe. 949 * 950 * we will only rehash the resulting dentry if @prehash is 951 * true; @prehash will be set to false (for the benefit of 952 * the caller) if we fail. 953 */ 954 static struct dentry *splice_dentry(struct dentry *dn, struct inode *in, 955 bool *prehash) 956 { 957 struct dentry *realdn; 958 959 BUG_ON(dn->d_inode); 960 961 /* dn must be unhashed */ 962 if (!d_unhashed(dn)) 963 d_drop(dn); 964 realdn = d_materialise_unique(dn, in); 965 if (IS_ERR(realdn)) { 966 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n", 967 PTR_ERR(realdn), dn, in, ceph_vinop(in)); 968 if (prehash) 969 *prehash = false; /* don't rehash on error */ 970 dn = realdn; /* note realdn contains the error */ 971 goto out; 972 } else if (realdn) { 973 dout("dn %p (%d) spliced with %p (%d) " 974 "inode %p ino %llx.%llx\n", 975 dn, d_count(dn), 976 realdn, d_count(realdn), 977 realdn->d_inode, ceph_vinop(realdn->d_inode)); 978 dput(dn); 979 dn = realdn; 980 } else { 981 BUG_ON(!ceph_dentry(dn)); 982 dout("dn %p attached to %p ino %llx.%llx\n", 983 dn, dn->d_inode, ceph_vinop(dn->d_inode)); 984 } 985 if ((!prehash || *prehash) && d_unhashed(dn)) 986 d_rehash(dn); 987 out: 988 return dn; 989 } 990 991 /* 992 * Incorporate results into the local cache. This is either just 993 * one inode, or a directory, dentry, and possibly linked-to inode (e.g., 994 * after a lookup). 995 * 996 * A reply may contain 997 * a directory inode along with a dentry. 998 * and/or a target inode 999 * 1000 * Called with snap_rwsem (read). 1001 */ 1002 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req, 1003 struct ceph_mds_session *session) 1004 { 1005 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; 1006 struct inode *in = NULL; 1007 struct ceph_vino vino; 1008 struct ceph_fs_client *fsc = ceph_sb_to_client(sb); 1009 int err = 0; 1010 1011 dout("fill_trace %p is_dentry %d is_target %d\n", req, 1012 rinfo->head->is_dentry, rinfo->head->is_target); 1013 1014 #if 0 1015 /* 1016 * Debugging hook: 1017 * 1018 * If we resend completed ops to a recovering mds, we get no 1019 * trace. Since that is very rare, pretend this is the case 1020 * to ensure the 'no trace' handlers in the callers behave. 1021 * 1022 * Fill in inodes unconditionally to avoid breaking cap 1023 * invariants. 1024 */ 1025 if (rinfo->head->op & CEPH_MDS_OP_WRITE) { 1026 pr_info("fill_trace faking empty trace on %lld %s\n", 1027 req->r_tid, ceph_mds_op_name(rinfo->head->op)); 1028 if (rinfo->head->is_dentry) { 1029 rinfo->head->is_dentry = 0; 1030 err = fill_inode(req->r_locked_dir, 1031 &rinfo->diri, rinfo->dirfrag, 1032 session, req->r_request_started, -1); 1033 } 1034 if (rinfo->head->is_target) { 1035 rinfo->head->is_target = 0; 1036 ininfo = rinfo->targeti.in; 1037 vino.ino = le64_to_cpu(ininfo->ino); 1038 vino.snap = le64_to_cpu(ininfo->snapid); 1039 in = ceph_get_inode(sb, vino); 1040 err = fill_inode(in, &rinfo->targeti, NULL, 1041 session, req->r_request_started, 1042 req->r_fmode); 1043 iput(in); 1044 } 1045 } 1046 #endif 1047 1048 if (!rinfo->head->is_target && !rinfo->head->is_dentry) { 1049 dout("fill_trace reply is empty!\n"); 1050 if (rinfo->head->result == 0 && req->r_locked_dir) 1051 ceph_invalidate_dir_request(req); 1052 return 0; 1053 } 1054 1055 if (rinfo->head->is_dentry) { 1056 struct inode *dir = req->r_locked_dir; 1057 1058 if (dir) { 1059 err = fill_inode(dir, &rinfo->diri, rinfo->dirfrag, 1060 session, req->r_request_started, -1, 1061 &req->r_caps_reservation); 1062 if (err < 0) 1063 goto done; 1064 } else { 1065 WARN_ON_ONCE(1); 1066 } 1067 1068 if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME) { 1069 struct qstr dname; 1070 struct dentry *dn, *parent; 1071 1072 BUG_ON(!rinfo->head->is_target); 1073 BUG_ON(req->r_dentry); 1074 1075 parent = d_find_any_alias(dir); 1076 BUG_ON(!parent); 1077 1078 dname.name = rinfo->dname; 1079 dname.len = rinfo->dname_len; 1080 dname.hash = full_name_hash(dname.name, dname.len); 1081 vino.ino = le64_to_cpu(rinfo->targeti.in->ino); 1082 vino.snap = le64_to_cpu(rinfo->targeti.in->snapid); 1083 retry_lookup: 1084 dn = d_lookup(parent, &dname); 1085 dout("d_lookup on parent=%p name=%.*s got %p\n", 1086 parent, dname.len, dname.name, dn); 1087 1088 if (!dn) { 1089 dn = d_alloc(parent, &dname); 1090 dout("d_alloc %p '%.*s' = %p\n", parent, 1091 dname.len, dname.name, dn); 1092 if (dn == NULL) { 1093 dput(parent); 1094 err = -ENOMEM; 1095 goto done; 1096 } 1097 err = ceph_init_dentry(dn); 1098 if (err < 0) { 1099 dput(dn); 1100 dput(parent); 1101 goto done; 1102 } 1103 } else if (dn->d_inode && 1104 (ceph_ino(dn->d_inode) != vino.ino || 1105 ceph_snap(dn->d_inode) != vino.snap)) { 1106 dout(" dn %p points to wrong inode %p\n", 1107 dn, dn->d_inode); 1108 d_delete(dn); 1109 dput(dn); 1110 goto retry_lookup; 1111 } 1112 1113 req->r_dentry = dn; 1114 dput(parent); 1115 } 1116 } 1117 1118 if (rinfo->head->is_target) { 1119 vino.ino = le64_to_cpu(rinfo->targeti.in->ino); 1120 vino.snap = le64_to_cpu(rinfo->targeti.in->snapid); 1121 1122 in = ceph_get_inode(sb, vino); 1123 if (IS_ERR(in)) { 1124 err = PTR_ERR(in); 1125 goto done; 1126 } 1127 req->r_target_inode = in; 1128 1129 err = fill_inode(in, &rinfo->targeti, NULL, 1130 session, req->r_request_started, 1131 (!req->r_aborted && rinfo->head->result == 0) ? 1132 req->r_fmode : -1, 1133 &req->r_caps_reservation); 1134 if (err < 0) { 1135 pr_err("fill_inode badness %p %llx.%llx\n", 1136 in, ceph_vinop(in)); 1137 goto done; 1138 } 1139 } 1140 1141 /* 1142 * ignore null lease/binding on snapdir ENOENT, or else we 1143 * will have trouble splicing in the virtual snapdir later 1144 */ 1145 if (rinfo->head->is_dentry && !req->r_aborted && 1146 req->r_locked_dir && 1147 (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name, 1148 fsc->mount_options->snapdir_name, 1149 req->r_dentry->d_name.len))) { 1150 /* 1151 * lookup link rename : null -> possibly existing inode 1152 * mknod symlink mkdir : null -> new inode 1153 * unlink : linked -> null 1154 */ 1155 struct inode *dir = req->r_locked_dir; 1156 struct dentry *dn = req->r_dentry; 1157 bool have_dir_cap, have_lease; 1158 1159 BUG_ON(!dn); 1160 BUG_ON(!dir); 1161 BUG_ON(dn->d_parent->d_inode != dir); 1162 BUG_ON(ceph_ino(dir) != 1163 le64_to_cpu(rinfo->diri.in->ino)); 1164 BUG_ON(ceph_snap(dir) != 1165 le64_to_cpu(rinfo->diri.in->snapid)); 1166 1167 /* do we have a lease on the whole dir? */ 1168 have_dir_cap = 1169 (le32_to_cpu(rinfo->diri.in->cap.caps) & 1170 CEPH_CAP_FILE_SHARED); 1171 1172 /* do we have a dn lease? */ 1173 have_lease = have_dir_cap || 1174 le32_to_cpu(rinfo->dlease->duration_ms); 1175 if (!have_lease) 1176 dout("fill_trace no dentry lease or dir cap\n"); 1177 1178 /* rename? */ 1179 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) { 1180 struct inode *olddir = req->r_old_dentry_dir; 1181 BUG_ON(!olddir); 1182 1183 dout(" src %p '%.*s' dst %p '%.*s'\n", 1184 req->r_old_dentry, 1185 req->r_old_dentry->d_name.len, 1186 req->r_old_dentry->d_name.name, 1187 dn, dn->d_name.len, dn->d_name.name); 1188 dout("fill_trace doing d_move %p -> %p\n", 1189 req->r_old_dentry, dn); 1190 1191 d_move(req->r_old_dentry, dn); 1192 dout(" src %p '%.*s' dst %p '%.*s'\n", 1193 req->r_old_dentry, 1194 req->r_old_dentry->d_name.len, 1195 req->r_old_dentry->d_name.name, 1196 dn, dn->d_name.len, dn->d_name.name); 1197 1198 /* ensure target dentry is invalidated, despite 1199 rehashing bug in vfs_rename_dir */ 1200 ceph_invalidate_dentry_lease(dn); 1201 1202 /* d_move screws up sibling dentries' offsets */ 1203 ceph_dir_clear_complete(dir); 1204 ceph_dir_clear_complete(olddir); 1205 1206 dout("dn %p gets new offset %lld\n", req->r_old_dentry, 1207 ceph_dentry(req->r_old_dentry)->offset); 1208 1209 dn = req->r_old_dentry; /* use old_dentry */ 1210 } 1211 1212 /* null dentry? */ 1213 if (!rinfo->head->is_target) { 1214 dout("fill_trace null dentry\n"); 1215 if (dn->d_inode) { 1216 dout("d_delete %p\n", dn); 1217 d_delete(dn); 1218 } else { 1219 dout("d_instantiate %p NULL\n", dn); 1220 d_instantiate(dn, NULL); 1221 if (have_lease && d_unhashed(dn)) 1222 d_rehash(dn); 1223 update_dentry_lease(dn, rinfo->dlease, 1224 session, 1225 req->r_request_started); 1226 } 1227 goto done; 1228 } 1229 1230 /* attach proper inode */ 1231 if (!dn->d_inode) { 1232 ceph_dir_clear_complete(dir); 1233 ihold(in); 1234 dn = splice_dentry(dn, in, &have_lease); 1235 if (IS_ERR(dn)) { 1236 err = PTR_ERR(dn); 1237 goto done; 1238 } 1239 req->r_dentry = dn; /* may have spliced */ 1240 } else if (dn->d_inode && dn->d_inode != in) { 1241 dout(" %p links to %p %llx.%llx, not %llx.%llx\n", 1242 dn, dn->d_inode, ceph_vinop(dn->d_inode), 1243 ceph_vinop(in)); 1244 have_lease = false; 1245 } 1246 1247 if (have_lease) 1248 update_dentry_lease(dn, rinfo->dlease, session, 1249 req->r_request_started); 1250 dout(" final dn %p\n", dn); 1251 } else if (!req->r_aborted && 1252 (req->r_op == CEPH_MDS_OP_LOOKUPSNAP || 1253 req->r_op == CEPH_MDS_OP_MKSNAP)) { 1254 struct dentry *dn = req->r_dentry; 1255 struct inode *dir = req->r_locked_dir; 1256 1257 /* fill out a snapdir LOOKUPSNAP dentry */ 1258 BUG_ON(!dn); 1259 BUG_ON(!dir); 1260 BUG_ON(ceph_snap(dir) != CEPH_SNAPDIR); 1261 dout(" linking snapped dir %p to dn %p\n", in, dn); 1262 ceph_dir_clear_complete(dir); 1263 ihold(in); 1264 dn = splice_dentry(dn, in, NULL); 1265 if (IS_ERR(dn)) { 1266 err = PTR_ERR(dn); 1267 goto done; 1268 } 1269 req->r_dentry = dn; /* may have spliced */ 1270 } 1271 done: 1272 dout("fill_trace done err=%d\n", err); 1273 return err; 1274 } 1275 1276 /* 1277 * Prepopulate our cache with readdir results, leases, etc. 1278 */ 1279 static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req, 1280 struct ceph_mds_session *session) 1281 { 1282 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; 1283 int i, err = 0; 1284 1285 for (i = 0; i < rinfo->dir_nr; i++) { 1286 struct ceph_vino vino; 1287 struct inode *in; 1288 int rc; 1289 1290 vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino); 1291 vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid); 1292 1293 in = ceph_get_inode(req->r_dentry->d_sb, vino); 1294 if (IS_ERR(in)) { 1295 err = PTR_ERR(in); 1296 dout("new_inode badness got %d\n", err); 1297 continue; 1298 } 1299 rc = fill_inode(in, &rinfo->dir_in[i], NULL, session, 1300 req->r_request_started, -1, 1301 &req->r_caps_reservation); 1302 if (rc < 0) { 1303 pr_err("fill_inode badness on %p got %d\n", in, rc); 1304 err = rc; 1305 continue; 1306 } 1307 } 1308 1309 return err; 1310 } 1311 1312 int ceph_readdir_prepopulate(struct ceph_mds_request *req, 1313 struct ceph_mds_session *session) 1314 { 1315 struct dentry *parent = req->r_dentry; 1316 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; 1317 struct qstr dname; 1318 struct dentry *dn; 1319 struct inode *in; 1320 int err = 0, ret, i; 1321 struct inode *snapdir = NULL; 1322 struct ceph_mds_request_head *rhead = req->r_request->front.iov_base; 1323 struct ceph_dentry_info *di; 1324 u64 r_readdir_offset = req->r_readdir_offset; 1325 u32 frag = le32_to_cpu(rhead->args.readdir.frag); 1326 1327 if (rinfo->dir_dir && 1328 le32_to_cpu(rinfo->dir_dir->frag) != frag) { 1329 dout("readdir_prepopulate got new frag %x -> %x\n", 1330 frag, le32_to_cpu(rinfo->dir_dir->frag)); 1331 frag = le32_to_cpu(rinfo->dir_dir->frag); 1332 if (ceph_frag_is_leftmost(frag)) 1333 r_readdir_offset = 2; 1334 else 1335 r_readdir_offset = 0; 1336 } 1337 1338 if (req->r_aborted) 1339 return readdir_prepopulate_inodes_only(req, session); 1340 1341 if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) { 1342 snapdir = ceph_get_snapdir(parent->d_inode); 1343 parent = d_find_alias(snapdir); 1344 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n", 1345 rinfo->dir_nr, parent); 1346 } else { 1347 dout("readdir_prepopulate %d items under dn %p\n", 1348 rinfo->dir_nr, parent); 1349 if (rinfo->dir_dir) 1350 ceph_fill_dirfrag(parent->d_inode, rinfo->dir_dir); 1351 } 1352 1353 /* FIXME: release caps/leases if error occurs */ 1354 for (i = 0; i < rinfo->dir_nr; i++) { 1355 struct ceph_vino vino; 1356 1357 dname.name = rinfo->dir_dname[i]; 1358 dname.len = rinfo->dir_dname_len[i]; 1359 dname.hash = full_name_hash(dname.name, dname.len); 1360 1361 vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino); 1362 vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid); 1363 1364 retry_lookup: 1365 dn = d_lookup(parent, &dname); 1366 dout("d_lookup on parent=%p name=%.*s got %p\n", 1367 parent, dname.len, dname.name, dn); 1368 1369 if (!dn) { 1370 dn = d_alloc(parent, &dname); 1371 dout("d_alloc %p '%.*s' = %p\n", parent, 1372 dname.len, dname.name, dn); 1373 if (dn == NULL) { 1374 dout("d_alloc badness\n"); 1375 err = -ENOMEM; 1376 goto out; 1377 } 1378 ret = ceph_init_dentry(dn); 1379 if (ret < 0) { 1380 dput(dn); 1381 err = ret; 1382 goto out; 1383 } 1384 } else if (dn->d_inode && 1385 (ceph_ino(dn->d_inode) != vino.ino || 1386 ceph_snap(dn->d_inode) != vino.snap)) { 1387 dout(" dn %p points to wrong inode %p\n", 1388 dn, dn->d_inode); 1389 d_delete(dn); 1390 dput(dn); 1391 goto retry_lookup; 1392 } else { 1393 /* reorder parent's d_subdirs */ 1394 spin_lock(&parent->d_lock); 1395 spin_lock_nested(&dn->d_lock, DENTRY_D_LOCK_NESTED); 1396 list_move(&dn->d_u.d_child, &parent->d_subdirs); 1397 spin_unlock(&dn->d_lock); 1398 spin_unlock(&parent->d_lock); 1399 } 1400 1401 /* inode */ 1402 if (dn->d_inode) { 1403 in = dn->d_inode; 1404 } else { 1405 in = ceph_get_inode(parent->d_sb, vino); 1406 if (IS_ERR(in)) { 1407 dout("new_inode badness\n"); 1408 d_drop(dn); 1409 dput(dn); 1410 err = PTR_ERR(in); 1411 goto out; 1412 } 1413 } 1414 1415 if (fill_inode(in, &rinfo->dir_in[i], NULL, session, 1416 req->r_request_started, -1, 1417 &req->r_caps_reservation) < 0) { 1418 pr_err("fill_inode badness on %p\n", in); 1419 if (!dn->d_inode) 1420 iput(in); 1421 d_drop(dn); 1422 goto next_item; 1423 } 1424 1425 if (!dn->d_inode) { 1426 dn = splice_dentry(dn, in, NULL); 1427 if (IS_ERR(dn)) { 1428 err = PTR_ERR(dn); 1429 dn = NULL; 1430 goto next_item; 1431 } 1432 } 1433 1434 di = dn->d_fsdata; 1435 di->offset = ceph_make_fpos(frag, i + r_readdir_offset); 1436 1437 update_dentry_lease(dn, rinfo->dir_dlease[i], 1438 req->r_session, 1439 req->r_request_started); 1440 next_item: 1441 if (dn) 1442 dput(dn); 1443 } 1444 if (err == 0) 1445 req->r_did_prepopulate = true; 1446 1447 out: 1448 if (snapdir) { 1449 iput(snapdir); 1450 dput(parent); 1451 } 1452 dout("readdir_prepopulate done\n"); 1453 return err; 1454 } 1455 1456 int ceph_inode_set_size(struct inode *inode, loff_t size) 1457 { 1458 struct ceph_inode_info *ci = ceph_inode(inode); 1459 int ret = 0; 1460 1461 spin_lock(&ci->i_ceph_lock); 1462 dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size); 1463 inode->i_size = size; 1464 inode->i_blocks = (size + (1 << 9) - 1) >> 9; 1465 1466 /* tell the MDS if we are approaching max_size */ 1467 if ((size << 1) >= ci->i_max_size && 1468 (ci->i_reported_size << 1) < ci->i_max_size) 1469 ret = 1; 1470 1471 spin_unlock(&ci->i_ceph_lock); 1472 return ret; 1473 } 1474 1475 /* 1476 * Write back inode data in a worker thread. (This can't be done 1477 * in the message handler context.) 1478 */ 1479 void ceph_queue_writeback(struct inode *inode) 1480 { 1481 ihold(inode); 1482 if (queue_work(ceph_inode_to_client(inode)->wb_wq, 1483 &ceph_inode(inode)->i_wb_work)) { 1484 dout("ceph_queue_writeback %p\n", inode); 1485 } else { 1486 dout("ceph_queue_writeback %p failed\n", inode); 1487 iput(inode); 1488 } 1489 } 1490 1491 static void ceph_writeback_work(struct work_struct *work) 1492 { 1493 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info, 1494 i_wb_work); 1495 struct inode *inode = &ci->vfs_inode; 1496 1497 dout("writeback %p\n", inode); 1498 filemap_fdatawrite(&inode->i_data); 1499 iput(inode); 1500 } 1501 1502 /* 1503 * queue an async invalidation 1504 */ 1505 void ceph_queue_invalidate(struct inode *inode) 1506 { 1507 ihold(inode); 1508 if (queue_work(ceph_inode_to_client(inode)->pg_inv_wq, 1509 &ceph_inode(inode)->i_pg_inv_work)) { 1510 dout("ceph_queue_invalidate %p\n", inode); 1511 } else { 1512 dout("ceph_queue_invalidate %p failed\n", inode); 1513 iput(inode); 1514 } 1515 } 1516 1517 /* 1518 * Invalidate inode pages in a worker thread. (This can't be done 1519 * in the message handler context.) 1520 */ 1521 static void ceph_invalidate_work(struct work_struct *work) 1522 { 1523 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info, 1524 i_pg_inv_work); 1525 struct inode *inode = &ci->vfs_inode; 1526 u32 orig_gen; 1527 int check = 0; 1528 1529 mutex_lock(&ci->i_truncate_mutex); 1530 spin_lock(&ci->i_ceph_lock); 1531 dout("invalidate_pages %p gen %d revoking %d\n", inode, 1532 ci->i_rdcache_gen, ci->i_rdcache_revoking); 1533 if (ci->i_rdcache_revoking != ci->i_rdcache_gen) { 1534 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE)) 1535 check = 1; 1536 spin_unlock(&ci->i_ceph_lock); 1537 mutex_unlock(&ci->i_truncate_mutex); 1538 goto out; 1539 } 1540 orig_gen = ci->i_rdcache_gen; 1541 spin_unlock(&ci->i_ceph_lock); 1542 1543 truncate_inode_pages(inode->i_mapping, 0); 1544 1545 spin_lock(&ci->i_ceph_lock); 1546 if (orig_gen == ci->i_rdcache_gen && 1547 orig_gen == ci->i_rdcache_revoking) { 1548 dout("invalidate_pages %p gen %d successful\n", inode, 1549 ci->i_rdcache_gen); 1550 ci->i_rdcache_revoking--; 1551 check = 1; 1552 } else { 1553 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n", 1554 inode, orig_gen, ci->i_rdcache_gen, 1555 ci->i_rdcache_revoking); 1556 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE)) 1557 check = 1; 1558 } 1559 spin_unlock(&ci->i_ceph_lock); 1560 mutex_unlock(&ci->i_truncate_mutex); 1561 out: 1562 if (check) 1563 ceph_check_caps(ci, 0, NULL); 1564 iput(inode); 1565 } 1566 1567 1568 /* 1569 * called by trunc_wq; 1570 * 1571 * We also truncate in a separate thread as well. 1572 */ 1573 static void ceph_vmtruncate_work(struct work_struct *work) 1574 { 1575 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info, 1576 i_vmtruncate_work); 1577 struct inode *inode = &ci->vfs_inode; 1578 1579 dout("vmtruncate_work %p\n", inode); 1580 __ceph_do_pending_vmtruncate(inode); 1581 iput(inode); 1582 } 1583 1584 /* 1585 * Queue an async vmtruncate. If we fail to queue work, we will handle 1586 * the truncation the next time we call __ceph_do_pending_vmtruncate. 1587 */ 1588 void ceph_queue_vmtruncate(struct inode *inode) 1589 { 1590 struct ceph_inode_info *ci = ceph_inode(inode); 1591 1592 ihold(inode); 1593 1594 if (queue_work(ceph_sb_to_client(inode->i_sb)->trunc_wq, 1595 &ci->i_vmtruncate_work)) { 1596 dout("ceph_queue_vmtruncate %p\n", inode); 1597 } else { 1598 dout("ceph_queue_vmtruncate %p failed, pending=%d\n", 1599 inode, ci->i_truncate_pending); 1600 iput(inode); 1601 } 1602 } 1603 1604 /* 1605 * Make sure any pending truncation is applied before doing anything 1606 * that may depend on it. 1607 */ 1608 void __ceph_do_pending_vmtruncate(struct inode *inode) 1609 { 1610 struct ceph_inode_info *ci = ceph_inode(inode); 1611 u64 to; 1612 int wrbuffer_refs, finish = 0; 1613 1614 mutex_lock(&ci->i_truncate_mutex); 1615 retry: 1616 spin_lock(&ci->i_ceph_lock); 1617 if (ci->i_truncate_pending == 0) { 1618 dout("__do_pending_vmtruncate %p none pending\n", inode); 1619 spin_unlock(&ci->i_ceph_lock); 1620 mutex_unlock(&ci->i_truncate_mutex); 1621 return; 1622 } 1623 1624 /* 1625 * make sure any dirty snapped pages are flushed before we 1626 * possibly truncate them.. so write AND block! 1627 */ 1628 if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) { 1629 dout("__do_pending_vmtruncate %p flushing snaps first\n", 1630 inode); 1631 spin_unlock(&ci->i_ceph_lock); 1632 filemap_write_and_wait_range(&inode->i_data, 0, 1633 inode->i_sb->s_maxbytes); 1634 goto retry; 1635 } 1636 1637 /* there should be no reader or writer */ 1638 WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref); 1639 1640 to = ci->i_truncate_size; 1641 wrbuffer_refs = ci->i_wrbuffer_ref; 1642 dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode, 1643 ci->i_truncate_pending, to); 1644 spin_unlock(&ci->i_ceph_lock); 1645 1646 truncate_inode_pages(inode->i_mapping, to); 1647 1648 spin_lock(&ci->i_ceph_lock); 1649 if (to == ci->i_truncate_size) { 1650 ci->i_truncate_pending = 0; 1651 finish = 1; 1652 } 1653 spin_unlock(&ci->i_ceph_lock); 1654 if (!finish) 1655 goto retry; 1656 1657 mutex_unlock(&ci->i_truncate_mutex); 1658 1659 if (wrbuffer_refs == 0) 1660 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL); 1661 1662 wake_up_all(&ci->i_cap_wq); 1663 } 1664 1665 /* 1666 * symlinks 1667 */ 1668 static void *ceph_sym_follow_link(struct dentry *dentry, struct nameidata *nd) 1669 { 1670 struct ceph_inode_info *ci = ceph_inode(dentry->d_inode); 1671 nd_set_link(nd, ci->i_symlink); 1672 return NULL; 1673 } 1674 1675 static const struct inode_operations ceph_symlink_iops = { 1676 .readlink = generic_readlink, 1677 .follow_link = ceph_sym_follow_link, 1678 .setattr = ceph_setattr, 1679 .getattr = ceph_getattr, 1680 .setxattr = ceph_setxattr, 1681 .getxattr = ceph_getxattr, 1682 .listxattr = ceph_listxattr, 1683 .removexattr = ceph_removexattr, 1684 }; 1685 1686 /* 1687 * setattr 1688 */ 1689 int ceph_setattr(struct dentry *dentry, struct iattr *attr) 1690 { 1691 struct inode *inode = dentry->d_inode; 1692 struct ceph_inode_info *ci = ceph_inode(inode); 1693 const unsigned int ia_valid = attr->ia_valid; 1694 struct ceph_mds_request *req; 1695 struct ceph_mds_client *mdsc = ceph_sb_to_client(dentry->d_sb)->mdsc; 1696 int issued; 1697 int release = 0, dirtied = 0; 1698 int mask = 0; 1699 int err = 0; 1700 int inode_dirty_flags = 0; 1701 1702 if (ceph_snap(inode) != CEPH_NOSNAP) 1703 return -EROFS; 1704 1705 err = inode_change_ok(inode, attr); 1706 if (err != 0) 1707 return err; 1708 1709 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR, 1710 USE_AUTH_MDS); 1711 if (IS_ERR(req)) 1712 return PTR_ERR(req); 1713 1714 spin_lock(&ci->i_ceph_lock); 1715 issued = __ceph_caps_issued(ci, NULL); 1716 dout("setattr %p issued %s\n", inode, ceph_cap_string(issued)); 1717 1718 if (ia_valid & ATTR_UID) { 1719 dout("setattr %p uid %d -> %d\n", inode, 1720 from_kuid(&init_user_ns, inode->i_uid), 1721 from_kuid(&init_user_ns, attr->ia_uid)); 1722 if (issued & CEPH_CAP_AUTH_EXCL) { 1723 inode->i_uid = attr->ia_uid; 1724 dirtied |= CEPH_CAP_AUTH_EXCL; 1725 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || 1726 !uid_eq(attr->ia_uid, inode->i_uid)) { 1727 req->r_args.setattr.uid = cpu_to_le32( 1728 from_kuid(&init_user_ns, attr->ia_uid)); 1729 mask |= CEPH_SETATTR_UID; 1730 release |= CEPH_CAP_AUTH_SHARED; 1731 } 1732 } 1733 if (ia_valid & ATTR_GID) { 1734 dout("setattr %p gid %d -> %d\n", inode, 1735 from_kgid(&init_user_ns, inode->i_gid), 1736 from_kgid(&init_user_ns, attr->ia_gid)); 1737 if (issued & CEPH_CAP_AUTH_EXCL) { 1738 inode->i_gid = attr->ia_gid; 1739 dirtied |= CEPH_CAP_AUTH_EXCL; 1740 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || 1741 !gid_eq(attr->ia_gid, inode->i_gid)) { 1742 req->r_args.setattr.gid = cpu_to_le32( 1743 from_kgid(&init_user_ns, attr->ia_gid)); 1744 mask |= CEPH_SETATTR_GID; 1745 release |= CEPH_CAP_AUTH_SHARED; 1746 } 1747 } 1748 if (ia_valid & ATTR_MODE) { 1749 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode, 1750 attr->ia_mode); 1751 if (issued & CEPH_CAP_AUTH_EXCL) { 1752 inode->i_mode = attr->ia_mode; 1753 dirtied |= CEPH_CAP_AUTH_EXCL; 1754 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || 1755 attr->ia_mode != inode->i_mode) { 1756 inode->i_mode = attr->ia_mode; 1757 req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode); 1758 mask |= CEPH_SETATTR_MODE; 1759 release |= CEPH_CAP_AUTH_SHARED; 1760 } 1761 } 1762 1763 if (ia_valid & ATTR_ATIME) { 1764 dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode, 1765 inode->i_atime.tv_sec, inode->i_atime.tv_nsec, 1766 attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec); 1767 if (issued & CEPH_CAP_FILE_EXCL) { 1768 ci->i_time_warp_seq++; 1769 inode->i_atime = attr->ia_atime; 1770 dirtied |= CEPH_CAP_FILE_EXCL; 1771 } else if ((issued & CEPH_CAP_FILE_WR) && 1772 timespec_compare(&inode->i_atime, 1773 &attr->ia_atime) < 0) { 1774 inode->i_atime = attr->ia_atime; 1775 dirtied |= CEPH_CAP_FILE_WR; 1776 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || 1777 !timespec_equal(&inode->i_atime, &attr->ia_atime)) { 1778 ceph_encode_timespec(&req->r_args.setattr.atime, 1779 &attr->ia_atime); 1780 mask |= CEPH_SETATTR_ATIME; 1781 release |= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_RD | 1782 CEPH_CAP_FILE_WR; 1783 } 1784 } 1785 if (ia_valid & ATTR_MTIME) { 1786 dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode, 1787 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec, 1788 attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec); 1789 if (issued & CEPH_CAP_FILE_EXCL) { 1790 ci->i_time_warp_seq++; 1791 inode->i_mtime = attr->ia_mtime; 1792 dirtied |= CEPH_CAP_FILE_EXCL; 1793 } else if ((issued & CEPH_CAP_FILE_WR) && 1794 timespec_compare(&inode->i_mtime, 1795 &attr->ia_mtime) < 0) { 1796 inode->i_mtime = attr->ia_mtime; 1797 dirtied |= CEPH_CAP_FILE_WR; 1798 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || 1799 !timespec_equal(&inode->i_mtime, &attr->ia_mtime)) { 1800 ceph_encode_timespec(&req->r_args.setattr.mtime, 1801 &attr->ia_mtime); 1802 mask |= CEPH_SETATTR_MTIME; 1803 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD | 1804 CEPH_CAP_FILE_WR; 1805 } 1806 } 1807 if (ia_valid & ATTR_SIZE) { 1808 dout("setattr %p size %lld -> %lld\n", inode, 1809 inode->i_size, attr->ia_size); 1810 if (attr->ia_size > inode->i_sb->s_maxbytes) { 1811 err = -EINVAL; 1812 goto out; 1813 } 1814 if ((issued & CEPH_CAP_FILE_EXCL) && 1815 attr->ia_size > inode->i_size) { 1816 inode->i_size = attr->ia_size; 1817 inode->i_blocks = 1818 (attr->ia_size + (1 << 9) - 1) >> 9; 1819 inode->i_ctime = attr->ia_ctime; 1820 ci->i_reported_size = attr->ia_size; 1821 dirtied |= CEPH_CAP_FILE_EXCL; 1822 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || 1823 attr->ia_size != inode->i_size) { 1824 req->r_args.setattr.size = cpu_to_le64(attr->ia_size); 1825 req->r_args.setattr.old_size = 1826 cpu_to_le64(inode->i_size); 1827 mask |= CEPH_SETATTR_SIZE; 1828 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD | 1829 CEPH_CAP_FILE_WR; 1830 } 1831 } 1832 1833 /* these do nothing */ 1834 if (ia_valid & ATTR_CTIME) { 1835 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME| 1836 ATTR_MODE|ATTR_UID|ATTR_GID)) == 0; 1837 dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode, 1838 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec, 1839 attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec, 1840 only ? "ctime only" : "ignored"); 1841 inode->i_ctime = attr->ia_ctime; 1842 if (only) { 1843 /* 1844 * if kernel wants to dirty ctime but nothing else, 1845 * we need to choose a cap to dirty under, or do 1846 * a almost-no-op setattr 1847 */ 1848 if (issued & CEPH_CAP_AUTH_EXCL) 1849 dirtied |= CEPH_CAP_AUTH_EXCL; 1850 else if (issued & CEPH_CAP_FILE_EXCL) 1851 dirtied |= CEPH_CAP_FILE_EXCL; 1852 else if (issued & CEPH_CAP_XATTR_EXCL) 1853 dirtied |= CEPH_CAP_XATTR_EXCL; 1854 else 1855 mask |= CEPH_SETATTR_CTIME; 1856 } 1857 } 1858 if (ia_valid & ATTR_FILE) 1859 dout("setattr %p ATTR_FILE ... hrm!\n", inode); 1860 1861 if (dirtied) { 1862 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied); 1863 inode->i_ctime = CURRENT_TIME; 1864 } 1865 1866 release &= issued; 1867 spin_unlock(&ci->i_ceph_lock); 1868 1869 if (inode_dirty_flags) 1870 __mark_inode_dirty(inode, inode_dirty_flags); 1871 1872 if (ia_valid & ATTR_MODE) { 1873 err = posix_acl_chmod(inode, attr->ia_mode); 1874 if (err) 1875 goto out_put; 1876 } 1877 1878 if (mask) { 1879 req->r_inode = inode; 1880 ihold(inode); 1881 req->r_inode_drop = release; 1882 req->r_args.setattr.mask = cpu_to_le32(mask); 1883 req->r_num_caps = 1; 1884 err = ceph_mdsc_do_request(mdsc, NULL, req); 1885 } 1886 dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err, 1887 ceph_cap_string(dirtied), mask); 1888 1889 ceph_mdsc_put_request(req); 1890 if (mask & CEPH_SETATTR_SIZE) 1891 __ceph_do_pending_vmtruncate(inode); 1892 return err; 1893 out: 1894 spin_unlock(&ci->i_ceph_lock); 1895 out_put: 1896 ceph_mdsc_put_request(req); 1897 return err; 1898 } 1899 1900 /* 1901 * Verify that we have a lease on the given mask. If not, 1902 * do a getattr against an mds. 1903 */ 1904 int ceph_do_getattr(struct inode *inode, int mask) 1905 { 1906 struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb); 1907 struct ceph_mds_client *mdsc = fsc->mdsc; 1908 struct ceph_mds_request *req; 1909 int err; 1910 1911 if (ceph_snap(inode) == CEPH_SNAPDIR) { 1912 dout("do_getattr inode %p SNAPDIR\n", inode); 1913 return 0; 1914 } 1915 1916 dout("do_getattr inode %p mask %s mode 0%o\n", inode, ceph_cap_string(mask), inode->i_mode); 1917 if (ceph_caps_issued_mask(ceph_inode(inode), mask, 1)) 1918 return 0; 1919 1920 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS); 1921 if (IS_ERR(req)) 1922 return PTR_ERR(req); 1923 req->r_inode = inode; 1924 ihold(inode); 1925 req->r_num_caps = 1; 1926 req->r_args.getattr.mask = cpu_to_le32(mask); 1927 err = ceph_mdsc_do_request(mdsc, NULL, req); 1928 ceph_mdsc_put_request(req); 1929 dout("do_getattr result=%d\n", err); 1930 return err; 1931 } 1932 1933 1934 /* 1935 * Check inode permissions. We verify we have a valid value for 1936 * the AUTH cap, then call the generic handler. 1937 */ 1938 int ceph_permission(struct inode *inode, int mask) 1939 { 1940 int err; 1941 1942 if (mask & MAY_NOT_BLOCK) 1943 return -ECHILD; 1944 1945 err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED); 1946 1947 if (!err) 1948 err = generic_permission(inode, mask); 1949 return err; 1950 } 1951 1952 /* 1953 * Get all attributes. Hopefully somedata we'll have a statlite() 1954 * and can limit the fields we require to be accurate. 1955 */ 1956 int ceph_getattr(struct vfsmount *mnt, struct dentry *dentry, 1957 struct kstat *stat) 1958 { 1959 struct inode *inode = dentry->d_inode; 1960 struct ceph_inode_info *ci = ceph_inode(inode); 1961 int err; 1962 1963 err = ceph_do_getattr(inode, CEPH_STAT_CAP_INODE_ALL); 1964 if (!err) { 1965 generic_fillattr(inode, stat); 1966 stat->ino = ceph_translate_ino(inode->i_sb, inode->i_ino); 1967 if (ceph_snap(inode) != CEPH_NOSNAP) 1968 stat->dev = ceph_snap(inode); 1969 else 1970 stat->dev = 0; 1971 if (S_ISDIR(inode->i_mode)) { 1972 if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), 1973 RBYTES)) 1974 stat->size = ci->i_rbytes; 1975 else 1976 stat->size = ci->i_files + ci->i_subdirs; 1977 stat->blocks = 0; 1978 stat->blksize = 65536; 1979 } 1980 } 1981 return err; 1982 } 1983