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