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