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 13 #include "super.h" 14 #include "mds_client.h" 15 #include "cache.h" 16 #include <linux/ceph/decode.h> 17 18 /* 19 * Ceph inode operations 20 * 21 * Implement basic inode helpers (get, alloc) and inode ops (getattr, 22 * setattr, etc.), xattr helpers, and helpers for assimilating 23 * metadata returned by the MDS into our cache. 24 * 25 * Also define helpers for doing asynchronous writeback, invalidation, 26 * and truncation for the benefit of those who can't afford to block 27 * (typically because they are in the message handler path). 28 */ 29 30 static const struct inode_operations ceph_symlink_iops; 31 32 static void ceph_invalidate_work(struct work_struct *work); 33 static void ceph_writeback_work(struct work_struct *work); 34 static void ceph_vmtruncate_work(struct work_struct *work); 35 36 /* 37 * find or create an inode, given the ceph ino number 38 */ 39 static int ceph_set_ino_cb(struct inode *inode, void *data) 40 { 41 ceph_inode(inode)->i_vino = *(struct ceph_vino *)data; 42 inode->i_ino = ceph_vino_to_ino(*(struct ceph_vino *)data); 43 return 0; 44 } 45 46 struct inode *ceph_get_inode(struct super_block *sb, struct ceph_vino vino) 47 { 48 struct inode *inode; 49 ino_t t = ceph_vino_to_ino(vino); 50 51 inode = iget5_locked(sb, t, ceph_ino_compare, ceph_set_ino_cb, &vino); 52 if (inode == NULL) 53 return ERR_PTR(-ENOMEM); 54 if (inode->i_state & I_NEW) { 55 dout("get_inode created new inode %p %llx.%llx ino %llx\n", 56 inode, ceph_vinop(inode), (u64)inode->i_ino); 57 unlock_new_inode(inode); 58 } 59 60 dout("get_inode on %lu=%llx.%llx got %p\n", inode->i_ino, vino.ino, 61 vino.snap, inode); 62 return inode; 63 } 64 65 struct inode *ceph_lookup_inode(struct super_block *sb, struct ceph_vino vino) 66 { 67 struct inode *inode; 68 ino_t t = ceph_vino_to_ino(vino); 69 inode = ilookup5_nowait(sb, t, ceph_ino_compare, &vino); 70 return inode; 71 } 72 73 /* 74 * get/constuct snapdir inode for a given directory 75 */ 76 struct inode *ceph_get_snapdir(struct inode *parent) 77 { 78 struct ceph_vino vino = { 79 .ino = ceph_ino(parent), 80 .snap = CEPH_SNAPDIR, 81 }; 82 struct inode *inode = ceph_get_inode(parent->i_sb, vino); 83 struct ceph_inode_info *ci = ceph_inode(inode); 84 85 BUG_ON(!S_ISDIR(parent->i_mode)); 86 if (IS_ERR(inode)) 87 return inode; 88 inode->i_mode = parent->i_mode; 89 inode->i_uid = parent->i_uid; 90 inode->i_gid = parent->i_gid; 91 inode->i_op = &ceph_dir_iops; 92 inode->i_fop = &ceph_dir_fops; 93 ci->i_snap_caps = CEPH_CAP_PIN; /* so we can open */ 94 ci->i_rbytes = 0; 95 return inode; 96 } 97 98 const struct inode_operations ceph_file_iops = { 99 .permission = ceph_permission, 100 .setattr = ceph_setattr, 101 .getattr = ceph_getattr, 102 .setxattr = ceph_setxattr, 103 .getxattr = ceph_getxattr, 104 .listxattr = ceph_listxattr, 105 .removexattr = ceph_removexattr, 106 }; 107 108 109 /* 110 * We use a 'frag tree' to keep track of the MDS's directory fragments 111 * for a given inode (usually there is just a single fragment). We 112 * need to know when a child frag is delegated to a new MDS, or when 113 * it is flagged as replicated, so we can direct our requests 114 * accordingly. 115 */ 116 117 /* 118 * find/create a frag in the tree 119 */ 120 static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci, 121 u32 f) 122 { 123 struct rb_node **p; 124 struct rb_node *parent = NULL; 125 struct ceph_inode_frag *frag; 126 int c; 127 128 p = &ci->i_fragtree.rb_node; 129 while (*p) { 130 parent = *p; 131 frag = rb_entry(parent, struct ceph_inode_frag, node); 132 c = ceph_frag_compare(f, frag->frag); 133 if (c < 0) 134 p = &(*p)->rb_left; 135 else if (c > 0) 136 p = &(*p)->rb_right; 137 else 138 return frag; 139 } 140 141 frag = kmalloc(sizeof(*frag), GFP_NOFS); 142 if (!frag) { 143 pr_err("__get_or_create_frag ENOMEM on %p %llx.%llx " 144 "frag %x\n", &ci->vfs_inode, 145 ceph_vinop(&ci->vfs_inode), f); 146 return ERR_PTR(-ENOMEM); 147 } 148 frag->frag = f; 149 frag->split_by = 0; 150 frag->mds = -1; 151 frag->ndist = 0; 152 153 rb_link_node(&frag->node, parent, p); 154 rb_insert_color(&frag->node, &ci->i_fragtree); 155 156 dout("get_or_create_frag added %llx.%llx frag %x\n", 157 ceph_vinop(&ci->vfs_inode), f); 158 return frag; 159 } 160 161 /* 162 * find a specific frag @f 163 */ 164 struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f) 165 { 166 struct rb_node *n = ci->i_fragtree.rb_node; 167 168 while (n) { 169 struct ceph_inode_frag *frag = 170 rb_entry(n, struct ceph_inode_frag, node); 171 int c = ceph_frag_compare(f, frag->frag); 172 if (c < 0) 173 n = n->rb_left; 174 else if (c > 0) 175 n = n->rb_right; 176 else 177 return frag; 178 } 179 return NULL; 180 } 181 182 /* 183 * Choose frag containing the given value @v. If @pfrag is 184 * specified, copy the frag delegation info to the caller if 185 * it is present. 186 */ 187 u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v, 188 struct ceph_inode_frag *pfrag, 189 int *found) 190 { 191 u32 t = ceph_frag_make(0, 0); 192 struct ceph_inode_frag *frag; 193 unsigned nway, i; 194 u32 n; 195 196 if (found) 197 *found = 0; 198 199 mutex_lock(&ci->i_fragtree_mutex); 200 while (1) { 201 WARN_ON(!ceph_frag_contains_value(t, v)); 202 frag = __ceph_find_frag(ci, t); 203 if (!frag) 204 break; /* t is a leaf */ 205 if (frag->split_by == 0) { 206 if (pfrag) 207 memcpy(pfrag, frag, sizeof(*pfrag)); 208 if (found) 209 *found = 1; 210 break; 211 } 212 213 /* choose child */ 214 nway = 1 << frag->split_by; 215 dout("choose_frag(%x) %x splits by %d (%d ways)\n", v, t, 216 frag->split_by, nway); 217 for (i = 0; i < nway; i++) { 218 n = ceph_frag_make_child(t, frag->split_by, i); 219 if (ceph_frag_contains_value(n, v)) { 220 t = n; 221 break; 222 } 223 } 224 BUG_ON(i == nway); 225 } 226 dout("choose_frag(%x) = %x\n", v, t); 227 228 mutex_unlock(&ci->i_fragtree_mutex); 229 return t; 230 } 231 232 /* 233 * Process dirfrag (delegation) info from the mds. Include leaf 234 * fragment in tree ONLY if ndist > 0. Otherwise, only 235 * branches/splits are included in i_fragtree) 236 */ 237 static int ceph_fill_dirfrag(struct inode *inode, 238 struct ceph_mds_reply_dirfrag *dirinfo) 239 { 240 struct ceph_inode_info *ci = ceph_inode(inode); 241 struct ceph_inode_frag *frag; 242 u32 id = le32_to_cpu(dirinfo->frag); 243 int mds = le32_to_cpu(dirinfo->auth); 244 int ndist = le32_to_cpu(dirinfo->ndist); 245 int i; 246 int err = 0; 247 248 mutex_lock(&ci->i_fragtree_mutex); 249 if (ndist == 0) { 250 /* no delegation info needed. */ 251 frag = __ceph_find_frag(ci, id); 252 if (!frag) 253 goto out; 254 if (frag->split_by == 0) { 255 /* tree leaf, remove */ 256 dout("fill_dirfrag removed %llx.%llx frag %x" 257 " (no ref)\n", ceph_vinop(inode), id); 258 rb_erase(&frag->node, &ci->i_fragtree); 259 kfree(frag); 260 } else { 261 /* tree branch, keep and clear */ 262 dout("fill_dirfrag cleared %llx.%llx frag %x" 263 " referral\n", ceph_vinop(inode), id); 264 frag->mds = -1; 265 frag->ndist = 0; 266 } 267 goto out; 268 } 269 270 271 /* find/add this frag to store mds delegation info */ 272 frag = __get_or_create_frag(ci, id); 273 if (IS_ERR(frag)) { 274 /* this is not the end of the world; we can continue 275 with bad/inaccurate delegation info */ 276 pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n", 277 ceph_vinop(inode), le32_to_cpu(dirinfo->frag)); 278 err = -ENOMEM; 279 goto out; 280 } 281 282 frag->mds = mds; 283 frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP); 284 for (i = 0; i < frag->ndist; i++) 285 frag->dist[i] = le32_to_cpu(dirinfo->dist[i]); 286 dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n", 287 ceph_vinop(inode), frag->frag, frag->ndist); 288 289 out: 290 mutex_unlock(&ci->i_fragtree_mutex); 291 return err; 292 } 293 294 295 /* 296 * initialize a newly allocated inode. 297 */ 298 struct inode *ceph_alloc_inode(struct super_block *sb) 299 { 300 struct ceph_inode_info *ci; 301 int i; 302 303 ci = kmem_cache_alloc(ceph_inode_cachep, GFP_NOFS); 304 if (!ci) 305 return NULL; 306 307 dout("alloc_inode %p\n", &ci->vfs_inode); 308 309 spin_lock_init(&ci->i_ceph_lock); 310 311 ci->i_version = 0; 312 ci->i_time_warp_seq = 0; 313 ci->i_ceph_flags = 0; 314 atomic_set(&ci->i_release_count, 1); 315 atomic_set(&ci->i_complete_count, 0); 316 ci->i_symlink = NULL; 317 318 memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout)); 319 320 ci->i_fragtree = RB_ROOT; 321 mutex_init(&ci->i_fragtree_mutex); 322 323 ci->i_xattrs.blob = NULL; 324 ci->i_xattrs.prealloc_blob = NULL; 325 ci->i_xattrs.dirty = false; 326 ci->i_xattrs.index = RB_ROOT; 327 ci->i_xattrs.count = 0; 328 ci->i_xattrs.names_size = 0; 329 ci->i_xattrs.vals_size = 0; 330 ci->i_xattrs.version = 0; 331 ci->i_xattrs.index_version = 0; 332 333 ci->i_caps = RB_ROOT; 334 ci->i_auth_cap = NULL; 335 ci->i_dirty_caps = 0; 336 ci->i_flushing_caps = 0; 337 INIT_LIST_HEAD(&ci->i_dirty_item); 338 INIT_LIST_HEAD(&ci->i_flushing_item); 339 ci->i_cap_flush_seq = 0; 340 ci->i_cap_flush_last_tid = 0; 341 memset(&ci->i_cap_flush_tid, 0, sizeof(ci->i_cap_flush_tid)); 342 init_waitqueue_head(&ci->i_cap_wq); 343 ci->i_hold_caps_min = 0; 344 ci->i_hold_caps_max = 0; 345 INIT_LIST_HEAD(&ci->i_cap_delay_list); 346 ci->i_cap_exporting_mds = 0; 347 ci->i_cap_exporting_mseq = 0; 348 ci->i_cap_exporting_issued = 0; 349 INIT_LIST_HEAD(&ci->i_cap_snaps); 350 ci->i_head_snapc = NULL; 351 ci->i_snap_caps = 0; 352 353 for (i = 0; i < CEPH_FILE_MODE_NUM; i++) 354 ci->i_nr_by_mode[i] = 0; 355 356 mutex_init(&ci->i_truncate_mutex); 357 ci->i_truncate_seq = 0; 358 ci->i_truncate_size = 0; 359 ci->i_truncate_pending = 0; 360 361 ci->i_max_size = 0; 362 ci->i_reported_size = 0; 363 ci->i_wanted_max_size = 0; 364 ci->i_requested_max_size = 0; 365 366 ci->i_pin_ref = 0; 367 ci->i_rd_ref = 0; 368 ci->i_rdcache_ref = 0; 369 ci->i_wr_ref = 0; 370 ci->i_wb_ref = 0; 371 ci->i_wrbuffer_ref = 0; 372 ci->i_wrbuffer_ref_head = 0; 373 ci->i_shared_gen = 0; 374 ci->i_rdcache_gen = 0; 375 ci->i_rdcache_revoking = 0; 376 377 INIT_LIST_HEAD(&ci->i_unsafe_writes); 378 INIT_LIST_HEAD(&ci->i_unsafe_dirops); 379 spin_lock_init(&ci->i_unsafe_lock); 380 381 ci->i_snap_realm = NULL; 382 INIT_LIST_HEAD(&ci->i_snap_realm_item); 383 INIT_LIST_HEAD(&ci->i_snap_flush_item); 384 385 INIT_WORK(&ci->i_wb_work, ceph_writeback_work); 386 INIT_WORK(&ci->i_pg_inv_work, ceph_invalidate_work); 387 388 INIT_WORK(&ci->i_vmtruncate_work, ceph_vmtruncate_work); 389 390 ceph_fscache_inode_init(ci); 391 392 return &ci->vfs_inode; 393 } 394 395 static void ceph_i_callback(struct rcu_head *head) 396 { 397 struct inode *inode = container_of(head, struct inode, i_rcu); 398 struct ceph_inode_info *ci = ceph_inode(inode); 399 400 kmem_cache_free(ceph_inode_cachep, ci); 401 } 402 403 void ceph_destroy_inode(struct inode *inode) 404 { 405 struct ceph_inode_info *ci = ceph_inode(inode); 406 struct ceph_inode_frag *frag; 407 struct rb_node *n; 408 409 dout("destroy_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode)); 410 411 ceph_fscache_unregister_inode_cookie(ci); 412 413 ceph_queue_caps_release(inode); 414 415 /* 416 * we may still have a snap_realm reference if there are stray 417 * caps in i_cap_exporting_issued or i_snap_caps. 418 */ 419 if (ci->i_snap_realm) { 420 struct ceph_mds_client *mdsc = 421 ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc; 422 struct ceph_snap_realm *realm = ci->i_snap_realm; 423 424 dout(" dropping residual ref to snap realm %p\n", realm); 425 spin_lock(&realm->inodes_with_caps_lock); 426 list_del_init(&ci->i_snap_realm_item); 427 spin_unlock(&realm->inodes_with_caps_lock); 428 ceph_put_snap_realm(mdsc, realm); 429 } 430 431 kfree(ci->i_symlink); 432 while ((n = rb_first(&ci->i_fragtree)) != NULL) { 433 frag = rb_entry(n, struct ceph_inode_frag, node); 434 rb_erase(n, &ci->i_fragtree); 435 kfree(frag); 436 } 437 438 __ceph_destroy_xattrs(ci); 439 if (ci->i_xattrs.blob) 440 ceph_buffer_put(ci->i_xattrs.blob); 441 if (ci->i_xattrs.prealloc_blob) 442 ceph_buffer_put(ci->i_xattrs.prealloc_blob); 443 444 call_rcu(&inode->i_rcu, ceph_i_callback); 445 } 446 447 /* 448 * Helpers to fill in size, ctime, mtime, and atime. We have to be 449 * careful because either the client or MDS may have more up to date 450 * info, depending on which capabilities are held, and whether 451 * time_warp_seq or truncate_seq have increased. (Ordinarily, mtime 452 * and size are monotonically increasing, except when utimes() or 453 * truncate() increments the corresponding _seq values.) 454 */ 455 int ceph_fill_file_size(struct inode *inode, int issued, 456 u32 truncate_seq, u64 truncate_size, u64 size) 457 { 458 struct ceph_inode_info *ci = ceph_inode(inode); 459 int queue_trunc = 0; 460 461 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 || 462 (truncate_seq == ci->i_truncate_seq && size > inode->i_size)) { 463 dout("size %lld -> %llu\n", inode->i_size, size); 464 inode->i_size = size; 465 inode->i_blocks = (size + (1<<9) - 1) >> 9; 466 ci->i_reported_size = size; 467 if (truncate_seq != ci->i_truncate_seq) { 468 dout("truncate_seq %u -> %u\n", 469 ci->i_truncate_seq, truncate_seq); 470 ci->i_truncate_seq = truncate_seq; 471 472 /* the MDS should have revoked these caps */ 473 WARN_ON_ONCE(issued & (CEPH_CAP_FILE_EXCL | 474 CEPH_CAP_FILE_RD | 475 CEPH_CAP_FILE_WR | 476 CEPH_CAP_FILE_LAZYIO)); 477 /* 478 * If we hold relevant caps, or in the case where we're 479 * not the only client referencing this file and we 480 * don't hold those caps, then we need to check whether 481 * the file is either opened or mmaped 482 */ 483 if ((issued & (CEPH_CAP_FILE_CACHE| 484 CEPH_CAP_FILE_BUFFER)) || 485 mapping_mapped(inode->i_mapping) || 486 __ceph_caps_file_wanted(ci)) { 487 ci->i_truncate_pending++; 488 queue_trunc = 1; 489 } 490 } 491 } 492 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 && 493 ci->i_truncate_size != truncate_size) { 494 dout("truncate_size %lld -> %llu\n", ci->i_truncate_size, 495 truncate_size); 496 ci->i_truncate_size = truncate_size; 497 } 498 499 if (queue_trunc) 500 ceph_fscache_invalidate(inode); 501 502 return queue_trunc; 503 } 504 505 void ceph_fill_file_time(struct inode *inode, int issued, 506 u64 time_warp_seq, struct timespec *ctime, 507 struct timespec *mtime, struct timespec *atime) 508 { 509 struct ceph_inode_info *ci = ceph_inode(inode); 510 int warn = 0; 511 512 if (issued & (CEPH_CAP_FILE_EXCL| 513 CEPH_CAP_FILE_WR| 514 CEPH_CAP_FILE_BUFFER| 515 CEPH_CAP_AUTH_EXCL| 516 CEPH_CAP_XATTR_EXCL)) { 517 if (timespec_compare(ctime, &inode->i_ctime) > 0) { 518 dout("ctime %ld.%09ld -> %ld.%09ld inc w/ cap\n", 519 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec, 520 ctime->tv_sec, ctime->tv_nsec); 521 inode->i_ctime = *ctime; 522 } 523 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) { 524 /* the MDS did a utimes() */ 525 dout("mtime %ld.%09ld -> %ld.%09ld " 526 "tw %d -> %d\n", 527 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec, 528 mtime->tv_sec, mtime->tv_nsec, 529 ci->i_time_warp_seq, (int)time_warp_seq); 530 531 inode->i_mtime = *mtime; 532 inode->i_atime = *atime; 533 ci->i_time_warp_seq = time_warp_seq; 534 } else if (time_warp_seq == ci->i_time_warp_seq) { 535 /* nobody did utimes(); take the max */ 536 if (timespec_compare(mtime, &inode->i_mtime) > 0) { 537 dout("mtime %ld.%09ld -> %ld.%09ld inc\n", 538 inode->i_mtime.tv_sec, 539 inode->i_mtime.tv_nsec, 540 mtime->tv_sec, mtime->tv_nsec); 541 inode->i_mtime = *mtime; 542 } 543 if (timespec_compare(atime, &inode->i_atime) > 0) { 544 dout("atime %ld.%09ld -> %ld.%09ld inc\n", 545 inode->i_atime.tv_sec, 546 inode->i_atime.tv_nsec, 547 atime->tv_sec, atime->tv_nsec); 548 inode->i_atime = *atime; 549 } 550 } else if (issued & CEPH_CAP_FILE_EXCL) { 551 /* we did a utimes(); ignore mds values */ 552 } else { 553 warn = 1; 554 } 555 } else { 556 /* we have no write|excl caps; whatever the MDS says is true */ 557 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) { 558 inode->i_ctime = *ctime; 559 inode->i_mtime = *mtime; 560 inode->i_atime = *atime; 561 ci->i_time_warp_seq = time_warp_seq; 562 } else { 563 warn = 1; 564 } 565 } 566 if (warn) /* time_warp_seq shouldn't go backwards */ 567 dout("%p mds time_warp_seq %llu < %u\n", 568 inode, time_warp_seq, ci->i_time_warp_seq); 569 } 570 571 /* 572 * Populate an inode based on info from mds. May be called on new or 573 * existing inodes. 574 */ 575 static int fill_inode(struct inode *inode, 576 struct ceph_mds_reply_info_in *iinfo, 577 struct ceph_mds_reply_dirfrag *dirinfo, 578 struct ceph_mds_session *session, 579 unsigned long ttl_from, int cap_fmode, 580 struct ceph_cap_reservation *caps_reservation) 581 { 582 struct ceph_mds_reply_inode *info = iinfo->in; 583 struct ceph_inode_info *ci = ceph_inode(inode); 584 int i; 585 int issued = 0, implemented; 586 struct timespec mtime, atime, ctime; 587 u32 nsplits; 588 struct ceph_buffer *xattr_blob = NULL; 589 int err = 0; 590 int queue_trunc = 0; 591 592 dout("fill_inode %p ino %llx.%llx v %llu had %llu\n", 593 inode, ceph_vinop(inode), le64_to_cpu(info->version), 594 ci->i_version); 595 596 /* 597 * prealloc xattr data, if it looks like we'll need it. only 598 * if len > 4 (meaning there are actually xattrs; the first 4 599 * bytes are the xattr count). 600 */ 601 if (iinfo->xattr_len > 4) { 602 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS); 603 if (!xattr_blob) 604 pr_err("fill_inode ENOMEM xattr blob %d bytes\n", 605 iinfo->xattr_len); 606 } 607 608 spin_lock(&ci->i_ceph_lock); 609 610 /* 611 * provided version will be odd if inode value is projected, 612 * even if stable. skip the update if we have newer stable 613 * info (ours>=theirs, e.g. due to racing mds replies), unless 614 * we are getting projected (unstable) info (in which case the 615 * version is odd, and we want ours>theirs). 616 * us them 617 * 2 2 skip 618 * 3 2 skip 619 * 3 3 update 620 */ 621 if (le64_to_cpu(info->version) > 0 && 622 (ci->i_version & ~1) >= le64_to_cpu(info->version)) 623 goto no_change; 624 625 issued = __ceph_caps_issued(ci, &implemented); 626 issued |= implemented | __ceph_caps_dirty(ci); 627 628 /* update inode */ 629 ci->i_version = le64_to_cpu(info->version); 630 inode->i_version++; 631 inode->i_rdev = le32_to_cpu(info->rdev); 632 633 if ((issued & CEPH_CAP_AUTH_EXCL) == 0) { 634 inode->i_mode = le32_to_cpu(info->mode); 635 inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid)); 636 inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid)); 637 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode, 638 from_kuid(&init_user_ns, inode->i_uid), 639 from_kgid(&init_user_ns, inode->i_gid)); 640 } 641 642 if ((issued & CEPH_CAP_LINK_EXCL) == 0) 643 set_nlink(inode, le32_to_cpu(info->nlink)); 644 645 /* be careful with mtime, atime, size */ 646 ceph_decode_timespec(&atime, &info->atime); 647 ceph_decode_timespec(&mtime, &info->mtime); 648 ceph_decode_timespec(&ctime, &info->ctime); 649 queue_trunc = ceph_fill_file_size(inode, issued, 650 le32_to_cpu(info->truncate_seq), 651 le64_to_cpu(info->truncate_size), 652 le64_to_cpu(info->size)); 653 ceph_fill_file_time(inode, issued, 654 le32_to_cpu(info->time_warp_seq), 655 &ctime, &mtime, &atime); 656 657 /* only update max_size on auth cap */ 658 if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) && 659 ci->i_max_size != le64_to_cpu(info->max_size)) { 660 dout("max_size %lld -> %llu\n", ci->i_max_size, 661 le64_to_cpu(info->max_size)); 662 ci->i_max_size = le64_to_cpu(info->max_size); 663 } 664 665 ci->i_layout = info->layout; 666 inode->i_blkbits = fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1; 667 668 /* xattrs */ 669 /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */ 670 if ((issued & CEPH_CAP_XATTR_EXCL) == 0 && 671 le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) { 672 if (ci->i_xattrs.blob) 673 ceph_buffer_put(ci->i_xattrs.blob); 674 ci->i_xattrs.blob = xattr_blob; 675 if (xattr_blob) 676 memcpy(ci->i_xattrs.blob->vec.iov_base, 677 iinfo->xattr_data, iinfo->xattr_len); 678 ci->i_xattrs.version = le64_to_cpu(info->xattr_version); 679 xattr_blob = NULL; 680 } 681 682 inode->i_mapping->a_ops = &ceph_aops; 683 inode->i_mapping->backing_dev_info = 684 &ceph_sb_to_client(inode->i_sb)->backing_dev_info; 685 686 switch (inode->i_mode & S_IFMT) { 687 case S_IFIFO: 688 case S_IFBLK: 689 case S_IFCHR: 690 case S_IFSOCK: 691 init_special_inode(inode, inode->i_mode, inode->i_rdev); 692 inode->i_op = &ceph_file_iops; 693 break; 694 case S_IFREG: 695 inode->i_op = &ceph_file_iops; 696 inode->i_fop = &ceph_file_fops; 697 break; 698 case S_IFLNK: 699 inode->i_op = &ceph_symlink_iops; 700 if (!ci->i_symlink) { 701 u32 symlen = iinfo->symlink_len; 702 char *sym; 703 704 spin_unlock(&ci->i_ceph_lock); 705 706 err = -EINVAL; 707 if (WARN_ON(symlen != inode->i_size)) 708 goto out; 709 710 err = -ENOMEM; 711 sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS); 712 if (!sym) 713 goto out; 714 715 spin_lock(&ci->i_ceph_lock); 716 if (!ci->i_symlink) 717 ci->i_symlink = sym; 718 else 719 kfree(sym); /* lost a race */ 720 } 721 break; 722 case S_IFDIR: 723 inode->i_op = &ceph_dir_iops; 724 inode->i_fop = &ceph_dir_fops; 725 726 ci->i_dir_layout = iinfo->dir_layout; 727 728 ci->i_files = le64_to_cpu(info->files); 729 ci->i_subdirs = le64_to_cpu(info->subdirs); 730 ci->i_rbytes = le64_to_cpu(info->rbytes); 731 ci->i_rfiles = le64_to_cpu(info->rfiles); 732 ci->i_rsubdirs = le64_to_cpu(info->rsubdirs); 733 ceph_decode_timespec(&ci->i_rctime, &info->rctime); 734 break; 735 default: 736 pr_err("fill_inode %llx.%llx BAD mode 0%o\n", 737 ceph_vinop(inode), inode->i_mode); 738 } 739 740 /* set dir completion flag? */ 741 if (S_ISDIR(inode->i_mode) && 742 ci->i_files == 0 && ci->i_subdirs == 0 && 743 ceph_snap(inode) == CEPH_NOSNAP && 744 (le32_to_cpu(info->cap.caps) & CEPH_CAP_FILE_SHARED) && 745 (issued & CEPH_CAP_FILE_EXCL) == 0 && 746 !__ceph_dir_is_complete(ci)) { 747 dout(" marking %p complete (empty)\n", inode); 748 __ceph_dir_set_complete(ci, atomic_read(&ci->i_release_count)); 749 ci->i_max_offset = 2; 750 } 751 no_change: 752 spin_unlock(&ci->i_ceph_lock); 753 754 /* queue truncate if we saw i_size decrease */ 755 if (queue_trunc) 756 ceph_queue_vmtruncate(inode); 757 758 /* populate frag tree */ 759 /* FIXME: move me up, if/when version reflects fragtree changes */ 760 nsplits = le32_to_cpu(info->fragtree.nsplits); 761 mutex_lock(&ci->i_fragtree_mutex); 762 for (i = 0; i < nsplits; i++) { 763 u32 id = le32_to_cpu(info->fragtree.splits[i].frag); 764 struct ceph_inode_frag *frag = __get_or_create_frag(ci, id); 765 766 if (IS_ERR(frag)) 767 continue; 768 frag->split_by = le32_to_cpu(info->fragtree.splits[i].by); 769 dout(" frag %x split by %d\n", frag->frag, frag->split_by); 770 } 771 mutex_unlock(&ci->i_fragtree_mutex); 772 773 /* were we issued a capability? */ 774 if (info->cap.caps) { 775 if (ceph_snap(inode) == CEPH_NOSNAP) { 776 ceph_add_cap(inode, session, 777 le64_to_cpu(info->cap.cap_id), 778 cap_fmode, 779 le32_to_cpu(info->cap.caps), 780 le32_to_cpu(info->cap.wanted), 781 le32_to_cpu(info->cap.seq), 782 le32_to_cpu(info->cap.mseq), 783 le64_to_cpu(info->cap.realm), 784 info->cap.flags, 785 caps_reservation); 786 } else { 787 spin_lock(&ci->i_ceph_lock); 788 dout(" %p got snap_caps %s\n", inode, 789 ceph_cap_string(le32_to_cpu(info->cap.caps))); 790 ci->i_snap_caps |= le32_to_cpu(info->cap.caps); 791 if (cap_fmode >= 0) 792 __ceph_get_fmode(ci, cap_fmode); 793 spin_unlock(&ci->i_ceph_lock); 794 } 795 } else if (cap_fmode >= 0) { 796 pr_warning("mds issued no caps on %llx.%llx\n", 797 ceph_vinop(inode)); 798 __ceph_get_fmode(ci, cap_fmode); 799 } 800 801 /* update delegation info? */ 802 if (dirinfo) 803 ceph_fill_dirfrag(inode, dirinfo); 804 805 err = 0; 806 807 out: 808 if (xattr_blob) 809 ceph_buffer_put(xattr_blob); 810 return err; 811 } 812 813 /* 814 * caller should hold session s_mutex. 815 */ 816 static void update_dentry_lease(struct dentry *dentry, 817 struct ceph_mds_reply_lease *lease, 818 struct ceph_mds_session *session, 819 unsigned long from_time) 820 { 821 struct ceph_dentry_info *di = ceph_dentry(dentry); 822 long unsigned duration = le32_to_cpu(lease->duration_ms); 823 long unsigned ttl = from_time + (duration * HZ) / 1000; 824 long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000; 825 struct inode *dir; 826 827 /* only track leases on regular dentries */ 828 if (dentry->d_op != &ceph_dentry_ops) 829 return; 830 831 spin_lock(&dentry->d_lock); 832 dout("update_dentry_lease %p duration %lu ms ttl %lu\n", 833 dentry, duration, ttl); 834 835 /* make lease_rdcache_gen match directory */ 836 dir = dentry->d_parent->d_inode; 837 di->lease_shared_gen = ceph_inode(dir)->i_shared_gen; 838 839 if (duration == 0) 840 goto out_unlock; 841 842 if (di->lease_gen == session->s_cap_gen && 843 time_before(ttl, dentry->d_time)) 844 goto out_unlock; /* we already have a newer lease. */ 845 846 if (di->lease_session && di->lease_session != session) 847 goto out_unlock; 848 849 ceph_dentry_lru_touch(dentry); 850 851 if (!di->lease_session) 852 di->lease_session = ceph_get_mds_session(session); 853 di->lease_gen = session->s_cap_gen; 854 di->lease_seq = le32_to_cpu(lease->seq); 855 di->lease_renew_after = half_ttl; 856 di->lease_renew_from = 0; 857 dentry->d_time = ttl; 858 out_unlock: 859 spin_unlock(&dentry->d_lock); 860 return; 861 } 862 863 /* 864 * Set dentry's directory position based on the current dir's max, and 865 * order it in d_subdirs, so that dcache_readdir behaves. 866 * 867 * Always called under directory's i_mutex. 868 */ 869 static void ceph_set_dentry_offset(struct dentry *dn) 870 { 871 struct dentry *dir = dn->d_parent; 872 struct inode *inode = dir->d_inode; 873 struct ceph_inode_info *ci; 874 struct ceph_dentry_info *di; 875 876 BUG_ON(!inode); 877 878 ci = ceph_inode(inode); 879 di = ceph_dentry(dn); 880 881 spin_lock(&ci->i_ceph_lock); 882 if (!__ceph_dir_is_complete(ci)) { 883 spin_unlock(&ci->i_ceph_lock); 884 return; 885 } 886 di->offset = ceph_inode(inode)->i_max_offset++; 887 spin_unlock(&ci->i_ceph_lock); 888 889 spin_lock(&dir->d_lock); 890 spin_lock_nested(&dn->d_lock, DENTRY_D_LOCK_NESTED); 891 list_move(&dn->d_u.d_child, &dir->d_subdirs); 892 dout("set_dentry_offset %p %lld (%p %p)\n", dn, di->offset, 893 dn->d_u.d_child.prev, dn->d_u.d_child.next); 894 spin_unlock(&dn->d_lock); 895 spin_unlock(&dir->d_lock); 896 } 897 898 /* 899 * splice a dentry to an inode. 900 * caller must hold directory i_mutex for this to be safe. 901 * 902 * we will only rehash the resulting dentry if @prehash is 903 * true; @prehash will be set to false (for the benefit of 904 * the caller) if we fail. 905 */ 906 static struct dentry *splice_dentry(struct dentry *dn, struct inode *in, 907 bool *prehash, bool set_offset) 908 { 909 struct dentry *realdn; 910 911 BUG_ON(dn->d_inode); 912 913 /* dn must be unhashed */ 914 if (!d_unhashed(dn)) 915 d_drop(dn); 916 realdn = d_materialise_unique(dn, in); 917 if (IS_ERR(realdn)) { 918 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n", 919 PTR_ERR(realdn), dn, in, ceph_vinop(in)); 920 if (prehash) 921 *prehash = false; /* don't rehash on error */ 922 dn = realdn; /* note realdn contains the error */ 923 goto out; 924 } else if (realdn) { 925 dout("dn %p (%d) spliced with %p (%d) " 926 "inode %p ino %llx.%llx\n", 927 dn, d_count(dn), 928 realdn, d_count(realdn), 929 realdn->d_inode, ceph_vinop(realdn->d_inode)); 930 dput(dn); 931 dn = realdn; 932 } else { 933 BUG_ON(!ceph_dentry(dn)); 934 dout("dn %p attached to %p ino %llx.%llx\n", 935 dn, dn->d_inode, ceph_vinop(dn->d_inode)); 936 } 937 if ((!prehash || *prehash) && d_unhashed(dn)) 938 d_rehash(dn); 939 if (set_offset) 940 ceph_set_dentry_offset(dn); 941 out: 942 return dn; 943 } 944 945 /* 946 * Incorporate results into the local cache. This is either just 947 * one inode, or a directory, dentry, and possibly linked-to inode (e.g., 948 * after a lookup). 949 * 950 * A reply may contain 951 * a directory inode along with a dentry. 952 * and/or a target inode 953 * 954 * Called with snap_rwsem (read). 955 */ 956 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req, 957 struct ceph_mds_session *session) 958 { 959 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; 960 struct inode *in = NULL; 961 struct ceph_mds_reply_inode *ininfo; 962 struct ceph_vino vino; 963 struct ceph_fs_client *fsc = ceph_sb_to_client(sb); 964 int i = 0; 965 int err = 0; 966 967 dout("fill_trace %p is_dentry %d is_target %d\n", req, 968 rinfo->head->is_dentry, rinfo->head->is_target); 969 970 #if 0 971 /* 972 * Debugging hook: 973 * 974 * If we resend completed ops to a recovering mds, we get no 975 * trace. Since that is very rare, pretend this is the case 976 * to ensure the 'no trace' handlers in the callers behave. 977 * 978 * Fill in inodes unconditionally to avoid breaking cap 979 * invariants. 980 */ 981 if (rinfo->head->op & CEPH_MDS_OP_WRITE) { 982 pr_info("fill_trace faking empty trace on %lld %s\n", 983 req->r_tid, ceph_mds_op_name(rinfo->head->op)); 984 if (rinfo->head->is_dentry) { 985 rinfo->head->is_dentry = 0; 986 err = fill_inode(req->r_locked_dir, 987 &rinfo->diri, rinfo->dirfrag, 988 session, req->r_request_started, -1); 989 } 990 if (rinfo->head->is_target) { 991 rinfo->head->is_target = 0; 992 ininfo = rinfo->targeti.in; 993 vino.ino = le64_to_cpu(ininfo->ino); 994 vino.snap = le64_to_cpu(ininfo->snapid); 995 in = ceph_get_inode(sb, vino); 996 err = fill_inode(in, &rinfo->targeti, NULL, 997 session, req->r_request_started, 998 req->r_fmode); 999 iput(in); 1000 } 1001 } 1002 #endif 1003 1004 if (!rinfo->head->is_target && !rinfo->head->is_dentry) { 1005 dout("fill_trace reply is empty!\n"); 1006 if (rinfo->head->result == 0 && req->r_locked_dir) 1007 ceph_invalidate_dir_request(req); 1008 return 0; 1009 } 1010 1011 if (rinfo->head->is_dentry) { 1012 struct inode *dir = req->r_locked_dir; 1013 1014 if (dir) { 1015 err = fill_inode(dir, &rinfo->diri, rinfo->dirfrag, 1016 session, req->r_request_started, -1, 1017 &req->r_caps_reservation); 1018 if (err < 0) 1019 return err; 1020 } else { 1021 WARN_ON_ONCE(1); 1022 } 1023 } 1024 1025 /* 1026 * ignore null lease/binding on snapdir ENOENT, or else we 1027 * will have trouble splicing in the virtual snapdir later 1028 */ 1029 if (rinfo->head->is_dentry && !req->r_aborted && 1030 req->r_locked_dir && 1031 (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name, 1032 fsc->mount_options->snapdir_name, 1033 req->r_dentry->d_name.len))) { 1034 /* 1035 * lookup link rename : null -> possibly existing inode 1036 * mknod symlink mkdir : null -> new inode 1037 * unlink : linked -> null 1038 */ 1039 struct inode *dir = req->r_locked_dir; 1040 struct dentry *dn = req->r_dentry; 1041 bool have_dir_cap, have_lease; 1042 1043 BUG_ON(!dn); 1044 BUG_ON(!dir); 1045 BUG_ON(dn->d_parent->d_inode != dir); 1046 BUG_ON(ceph_ino(dir) != 1047 le64_to_cpu(rinfo->diri.in->ino)); 1048 BUG_ON(ceph_snap(dir) != 1049 le64_to_cpu(rinfo->diri.in->snapid)); 1050 1051 /* do we have a lease on the whole dir? */ 1052 have_dir_cap = 1053 (le32_to_cpu(rinfo->diri.in->cap.caps) & 1054 CEPH_CAP_FILE_SHARED); 1055 1056 /* do we have a dn lease? */ 1057 have_lease = have_dir_cap || 1058 le32_to_cpu(rinfo->dlease->duration_ms); 1059 if (!have_lease) 1060 dout("fill_trace no dentry lease or dir cap\n"); 1061 1062 /* rename? */ 1063 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) { 1064 dout(" src %p '%.*s' dst %p '%.*s'\n", 1065 req->r_old_dentry, 1066 req->r_old_dentry->d_name.len, 1067 req->r_old_dentry->d_name.name, 1068 dn, dn->d_name.len, dn->d_name.name); 1069 dout("fill_trace doing d_move %p -> %p\n", 1070 req->r_old_dentry, dn); 1071 1072 d_move(req->r_old_dentry, dn); 1073 dout(" src %p '%.*s' dst %p '%.*s'\n", 1074 req->r_old_dentry, 1075 req->r_old_dentry->d_name.len, 1076 req->r_old_dentry->d_name.name, 1077 dn, dn->d_name.len, dn->d_name.name); 1078 1079 /* ensure target dentry is invalidated, despite 1080 rehashing bug in vfs_rename_dir */ 1081 ceph_invalidate_dentry_lease(dn); 1082 1083 /* 1084 * d_move() puts the renamed dentry at the end of 1085 * d_subdirs. We need to assign it an appropriate 1086 * directory offset so we can behave when dir is 1087 * complete. 1088 */ 1089 ceph_set_dentry_offset(req->r_old_dentry); 1090 dout("dn %p gets new offset %lld\n", req->r_old_dentry, 1091 ceph_dentry(req->r_old_dentry)->offset); 1092 1093 dn = req->r_old_dentry; /* use old_dentry */ 1094 in = dn->d_inode; 1095 } 1096 1097 /* null dentry? */ 1098 if (!rinfo->head->is_target) { 1099 dout("fill_trace null dentry\n"); 1100 if (dn->d_inode) { 1101 dout("d_delete %p\n", dn); 1102 d_delete(dn); 1103 } else { 1104 dout("d_instantiate %p NULL\n", dn); 1105 d_instantiate(dn, NULL); 1106 if (have_lease && d_unhashed(dn)) 1107 d_rehash(dn); 1108 update_dentry_lease(dn, rinfo->dlease, 1109 session, 1110 req->r_request_started); 1111 } 1112 goto done; 1113 } 1114 1115 /* attach proper inode */ 1116 ininfo = rinfo->targeti.in; 1117 vino.ino = le64_to_cpu(ininfo->ino); 1118 vino.snap = le64_to_cpu(ininfo->snapid); 1119 in = dn->d_inode; 1120 if (!in) { 1121 in = ceph_get_inode(sb, vino); 1122 if (IS_ERR(in)) { 1123 pr_err("fill_trace bad get_inode " 1124 "%llx.%llx\n", vino.ino, vino.snap); 1125 err = PTR_ERR(in); 1126 d_drop(dn); 1127 goto done; 1128 } 1129 dn = splice_dentry(dn, in, &have_lease, true); 1130 if (IS_ERR(dn)) { 1131 err = PTR_ERR(dn); 1132 goto done; 1133 } 1134 req->r_dentry = dn; /* may have spliced */ 1135 ihold(in); 1136 } else if (ceph_ino(in) == vino.ino && 1137 ceph_snap(in) == vino.snap) { 1138 ihold(in); 1139 } else { 1140 dout(" %p links to %p %llx.%llx, not %llx.%llx\n", 1141 dn, in, ceph_ino(in), ceph_snap(in), 1142 vino.ino, vino.snap); 1143 have_lease = false; 1144 in = NULL; 1145 } 1146 1147 if (have_lease) 1148 update_dentry_lease(dn, rinfo->dlease, session, 1149 req->r_request_started); 1150 dout(" final dn %p\n", dn); 1151 i++; 1152 } else if ((req->r_op == CEPH_MDS_OP_LOOKUPSNAP || 1153 req->r_op == CEPH_MDS_OP_MKSNAP) && !req->r_aborted) { 1154 struct dentry *dn = req->r_dentry; 1155 1156 /* fill out a snapdir LOOKUPSNAP dentry */ 1157 BUG_ON(!dn); 1158 BUG_ON(!req->r_locked_dir); 1159 BUG_ON(ceph_snap(req->r_locked_dir) != CEPH_SNAPDIR); 1160 ininfo = rinfo->targeti.in; 1161 vino.ino = le64_to_cpu(ininfo->ino); 1162 vino.snap = le64_to_cpu(ininfo->snapid); 1163 in = ceph_get_inode(sb, vino); 1164 if (IS_ERR(in)) { 1165 pr_err("fill_inode get_inode badness %llx.%llx\n", 1166 vino.ino, vino.snap); 1167 err = PTR_ERR(in); 1168 d_delete(dn); 1169 goto done; 1170 } 1171 dout(" linking snapped dir %p to dn %p\n", in, dn); 1172 dn = splice_dentry(dn, in, NULL, true); 1173 if (IS_ERR(dn)) { 1174 err = PTR_ERR(dn); 1175 goto done; 1176 } 1177 req->r_dentry = dn; /* may have spliced */ 1178 ihold(in); 1179 rinfo->head->is_dentry = 1; /* fool notrace handlers */ 1180 } 1181 1182 if (rinfo->head->is_target) { 1183 vino.ino = le64_to_cpu(rinfo->targeti.in->ino); 1184 vino.snap = le64_to_cpu(rinfo->targeti.in->snapid); 1185 1186 if (in == NULL || ceph_ino(in) != vino.ino || 1187 ceph_snap(in) != vino.snap) { 1188 in = ceph_get_inode(sb, vino); 1189 if (IS_ERR(in)) { 1190 err = PTR_ERR(in); 1191 goto done; 1192 } 1193 } 1194 req->r_target_inode = in; 1195 1196 err = fill_inode(in, 1197 &rinfo->targeti, NULL, 1198 session, req->r_request_started, 1199 (le32_to_cpu(rinfo->head->result) == 0) ? 1200 req->r_fmode : -1, 1201 &req->r_caps_reservation); 1202 if (err < 0) { 1203 pr_err("fill_inode badness %p %llx.%llx\n", 1204 in, ceph_vinop(in)); 1205 goto done; 1206 } 1207 } 1208 1209 done: 1210 dout("fill_trace done err=%d\n", err); 1211 return err; 1212 } 1213 1214 /* 1215 * Prepopulate our cache with readdir results, leases, etc. 1216 */ 1217 static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req, 1218 struct ceph_mds_session *session) 1219 { 1220 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; 1221 int i, err = 0; 1222 1223 for (i = 0; i < rinfo->dir_nr; i++) { 1224 struct ceph_vino vino; 1225 struct inode *in; 1226 int rc; 1227 1228 vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino); 1229 vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid); 1230 1231 in = ceph_get_inode(req->r_dentry->d_sb, vino); 1232 if (IS_ERR(in)) { 1233 err = PTR_ERR(in); 1234 dout("new_inode badness got %d\n", err); 1235 continue; 1236 } 1237 rc = fill_inode(in, &rinfo->dir_in[i], NULL, session, 1238 req->r_request_started, -1, 1239 &req->r_caps_reservation); 1240 if (rc < 0) { 1241 pr_err("fill_inode badness on %p got %d\n", in, rc); 1242 err = rc; 1243 continue; 1244 } 1245 } 1246 1247 return err; 1248 } 1249 1250 int ceph_readdir_prepopulate(struct ceph_mds_request *req, 1251 struct ceph_mds_session *session) 1252 { 1253 struct dentry *parent = req->r_dentry; 1254 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; 1255 struct qstr dname; 1256 struct dentry *dn; 1257 struct inode *in; 1258 int err = 0, i; 1259 struct inode *snapdir = NULL; 1260 struct ceph_mds_request_head *rhead = req->r_request->front.iov_base; 1261 u64 frag = le32_to_cpu(rhead->args.readdir.frag); 1262 struct ceph_dentry_info *di; 1263 1264 if (req->r_aborted) 1265 return readdir_prepopulate_inodes_only(req, session); 1266 1267 if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) { 1268 snapdir = ceph_get_snapdir(parent->d_inode); 1269 parent = d_find_alias(snapdir); 1270 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n", 1271 rinfo->dir_nr, parent); 1272 } else { 1273 dout("readdir_prepopulate %d items under dn %p\n", 1274 rinfo->dir_nr, parent); 1275 if (rinfo->dir_dir) 1276 ceph_fill_dirfrag(parent->d_inode, rinfo->dir_dir); 1277 } 1278 1279 for (i = 0; i < rinfo->dir_nr; i++) { 1280 struct ceph_vino vino; 1281 1282 dname.name = rinfo->dir_dname[i]; 1283 dname.len = rinfo->dir_dname_len[i]; 1284 dname.hash = full_name_hash(dname.name, dname.len); 1285 1286 vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino); 1287 vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid); 1288 1289 retry_lookup: 1290 dn = d_lookup(parent, &dname); 1291 dout("d_lookup on parent=%p name=%.*s got %p\n", 1292 parent, dname.len, dname.name, dn); 1293 1294 if (!dn) { 1295 dn = d_alloc(parent, &dname); 1296 dout("d_alloc %p '%.*s' = %p\n", parent, 1297 dname.len, dname.name, dn); 1298 if (dn == NULL) { 1299 dout("d_alloc badness\n"); 1300 err = -ENOMEM; 1301 goto out; 1302 } 1303 err = ceph_init_dentry(dn); 1304 if (err < 0) { 1305 dput(dn); 1306 goto out; 1307 } 1308 } else if (dn->d_inode && 1309 (ceph_ino(dn->d_inode) != vino.ino || 1310 ceph_snap(dn->d_inode) != vino.snap)) { 1311 dout(" dn %p points to wrong inode %p\n", 1312 dn, dn->d_inode); 1313 d_delete(dn); 1314 dput(dn); 1315 goto retry_lookup; 1316 } else { 1317 /* reorder parent's d_subdirs */ 1318 spin_lock(&parent->d_lock); 1319 spin_lock_nested(&dn->d_lock, DENTRY_D_LOCK_NESTED); 1320 list_move(&dn->d_u.d_child, &parent->d_subdirs); 1321 spin_unlock(&dn->d_lock); 1322 spin_unlock(&parent->d_lock); 1323 } 1324 1325 di = dn->d_fsdata; 1326 di->offset = ceph_make_fpos(frag, i + req->r_readdir_offset); 1327 1328 /* inode */ 1329 if (dn->d_inode) { 1330 in = dn->d_inode; 1331 } else { 1332 in = ceph_get_inode(parent->d_sb, vino); 1333 if (IS_ERR(in)) { 1334 dout("new_inode badness\n"); 1335 d_drop(dn); 1336 dput(dn); 1337 err = PTR_ERR(in); 1338 goto out; 1339 } 1340 dn = splice_dentry(dn, in, NULL, false); 1341 if (IS_ERR(dn)) 1342 dn = NULL; 1343 } 1344 1345 if (fill_inode(in, &rinfo->dir_in[i], NULL, session, 1346 req->r_request_started, -1, 1347 &req->r_caps_reservation) < 0) { 1348 pr_err("fill_inode badness on %p\n", in); 1349 goto next_item; 1350 } 1351 if (dn) 1352 update_dentry_lease(dn, rinfo->dir_dlease[i], 1353 req->r_session, 1354 req->r_request_started); 1355 next_item: 1356 if (dn) 1357 dput(dn); 1358 } 1359 req->r_did_prepopulate = true; 1360 1361 out: 1362 if (snapdir) { 1363 iput(snapdir); 1364 dput(parent); 1365 } 1366 dout("readdir_prepopulate done\n"); 1367 return err; 1368 } 1369 1370 int ceph_inode_set_size(struct inode *inode, loff_t size) 1371 { 1372 struct ceph_inode_info *ci = ceph_inode(inode); 1373 int ret = 0; 1374 1375 spin_lock(&ci->i_ceph_lock); 1376 dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size); 1377 inode->i_size = size; 1378 inode->i_blocks = (size + (1 << 9) - 1) >> 9; 1379 1380 /* tell the MDS if we are approaching max_size */ 1381 if ((size << 1) >= ci->i_max_size && 1382 (ci->i_reported_size << 1) < ci->i_max_size) 1383 ret = 1; 1384 1385 spin_unlock(&ci->i_ceph_lock); 1386 return ret; 1387 } 1388 1389 /* 1390 * Write back inode data in a worker thread. (This can't be done 1391 * in the message handler context.) 1392 */ 1393 void ceph_queue_writeback(struct inode *inode) 1394 { 1395 ihold(inode); 1396 if (queue_work(ceph_inode_to_client(inode)->wb_wq, 1397 &ceph_inode(inode)->i_wb_work)) { 1398 dout("ceph_queue_writeback %p\n", inode); 1399 } else { 1400 dout("ceph_queue_writeback %p failed\n", inode); 1401 iput(inode); 1402 } 1403 } 1404 1405 static void ceph_writeback_work(struct work_struct *work) 1406 { 1407 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info, 1408 i_wb_work); 1409 struct inode *inode = &ci->vfs_inode; 1410 1411 dout("writeback %p\n", inode); 1412 filemap_fdatawrite(&inode->i_data); 1413 iput(inode); 1414 } 1415 1416 /* 1417 * queue an async invalidation 1418 */ 1419 void ceph_queue_invalidate(struct inode *inode) 1420 { 1421 ihold(inode); 1422 if (queue_work(ceph_inode_to_client(inode)->pg_inv_wq, 1423 &ceph_inode(inode)->i_pg_inv_work)) { 1424 dout("ceph_queue_invalidate %p\n", inode); 1425 } else { 1426 dout("ceph_queue_invalidate %p failed\n", inode); 1427 iput(inode); 1428 } 1429 } 1430 1431 /* 1432 * Invalidate inode pages in a worker thread. (This can't be done 1433 * in the message handler context.) 1434 */ 1435 static void ceph_invalidate_work(struct work_struct *work) 1436 { 1437 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info, 1438 i_pg_inv_work); 1439 struct inode *inode = &ci->vfs_inode; 1440 u32 orig_gen; 1441 int check = 0; 1442 1443 mutex_lock(&ci->i_truncate_mutex); 1444 spin_lock(&ci->i_ceph_lock); 1445 dout("invalidate_pages %p gen %d revoking %d\n", inode, 1446 ci->i_rdcache_gen, ci->i_rdcache_revoking); 1447 if (ci->i_rdcache_revoking != ci->i_rdcache_gen) { 1448 /* nevermind! */ 1449 spin_unlock(&ci->i_ceph_lock); 1450 mutex_unlock(&ci->i_truncate_mutex); 1451 goto out; 1452 } 1453 orig_gen = ci->i_rdcache_gen; 1454 spin_unlock(&ci->i_ceph_lock); 1455 1456 truncate_inode_pages(inode->i_mapping, 0); 1457 1458 spin_lock(&ci->i_ceph_lock); 1459 if (orig_gen == ci->i_rdcache_gen && 1460 orig_gen == ci->i_rdcache_revoking) { 1461 dout("invalidate_pages %p gen %d successful\n", inode, 1462 ci->i_rdcache_gen); 1463 ci->i_rdcache_revoking--; 1464 check = 1; 1465 } else { 1466 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n", 1467 inode, orig_gen, ci->i_rdcache_gen, 1468 ci->i_rdcache_revoking); 1469 } 1470 spin_unlock(&ci->i_ceph_lock); 1471 mutex_unlock(&ci->i_truncate_mutex); 1472 1473 if (check) 1474 ceph_check_caps(ci, 0, NULL); 1475 out: 1476 iput(inode); 1477 } 1478 1479 1480 /* 1481 * called by trunc_wq; 1482 * 1483 * We also truncate in a separate thread as well. 1484 */ 1485 static void ceph_vmtruncate_work(struct work_struct *work) 1486 { 1487 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info, 1488 i_vmtruncate_work); 1489 struct inode *inode = &ci->vfs_inode; 1490 1491 dout("vmtruncate_work %p\n", inode); 1492 __ceph_do_pending_vmtruncate(inode); 1493 iput(inode); 1494 } 1495 1496 /* 1497 * Queue an async vmtruncate. If we fail to queue work, we will handle 1498 * the truncation the next time we call __ceph_do_pending_vmtruncate. 1499 */ 1500 void ceph_queue_vmtruncate(struct inode *inode) 1501 { 1502 struct ceph_inode_info *ci = ceph_inode(inode); 1503 1504 ihold(inode); 1505 1506 if (queue_work(ceph_sb_to_client(inode->i_sb)->trunc_wq, 1507 &ci->i_vmtruncate_work)) { 1508 dout("ceph_queue_vmtruncate %p\n", inode); 1509 } else { 1510 dout("ceph_queue_vmtruncate %p failed, pending=%d\n", 1511 inode, ci->i_truncate_pending); 1512 iput(inode); 1513 } 1514 } 1515 1516 /* 1517 * Make sure any pending truncation is applied before doing anything 1518 * that may depend on it. 1519 */ 1520 void __ceph_do_pending_vmtruncate(struct inode *inode) 1521 { 1522 struct ceph_inode_info *ci = ceph_inode(inode); 1523 u64 to; 1524 int wrbuffer_refs, finish = 0; 1525 1526 mutex_lock(&ci->i_truncate_mutex); 1527 retry: 1528 spin_lock(&ci->i_ceph_lock); 1529 if (ci->i_truncate_pending == 0) { 1530 dout("__do_pending_vmtruncate %p none pending\n", inode); 1531 spin_unlock(&ci->i_ceph_lock); 1532 mutex_unlock(&ci->i_truncate_mutex); 1533 return; 1534 } 1535 1536 /* 1537 * make sure any dirty snapped pages are flushed before we 1538 * possibly truncate them.. so write AND block! 1539 */ 1540 if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) { 1541 dout("__do_pending_vmtruncate %p flushing snaps first\n", 1542 inode); 1543 spin_unlock(&ci->i_ceph_lock); 1544 filemap_write_and_wait_range(&inode->i_data, 0, 1545 inode->i_sb->s_maxbytes); 1546 goto retry; 1547 } 1548 1549 /* there should be no reader or writer */ 1550 WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref); 1551 1552 to = ci->i_truncate_size; 1553 wrbuffer_refs = ci->i_wrbuffer_ref; 1554 dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode, 1555 ci->i_truncate_pending, to); 1556 spin_unlock(&ci->i_ceph_lock); 1557 1558 truncate_inode_pages(inode->i_mapping, to); 1559 1560 spin_lock(&ci->i_ceph_lock); 1561 if (to == ci->i_truncate_size) { 1562 ci->i_truncate_pending = 0; 1563 finish = 1; 1564 } 1565 spin_unlock(&ci->i_ceph_lock); 1566 if (!finish) 1567 goto retry; 1568 1569 mutex_unlock(&ci->i_truncate_mutex); 1570 1571 if (wrbuffer_refs == 0) 1572 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL); 1573 1574 wake_up_all(&ci->i_cap_wq); 1575 } 1576 1577 /* 1578 * symlinks 1579 */ 1580 static void *ceph_sym_follow_link(struct dentry *dentry, struct nameidata *nd) 1581 { 1582 struct ceph_inode_info *ci = ceph_inode(dentry->d_inode); 1583 nd_set_link(nd, ci->i_symlink); 1584 return NULL; 1585 } 1586 1587 static const struct inode_operations ceph_symlink_iops = { 1588 .readlink = generic_readlink, 1589 .follow_link = ceph_sym_follow_link, 1590 .setattr = ceph_setattr, 1591 .getattr = ceph_getattr, 1592 .setxattr = ceph_setxattr, 1593 .getxattr = ceph_getxattr, 1594 .listxattr = ceph_listxattr, 1595 .removexattr = ceph_removexattr, 1596 }; 1597 1598 /* 1599 * setattr 1600 */ 1601 int ceph_setattr(struct dentry *dentry, struct iattr *attr) 1602 { 1603 struct inode *inode = dentry->d_inode; 1604 struct ceph_inode_info *ci = ceph_inode(inode); 1605 struct inode *parent_inode; 1606 const unsigned int ia_valid = attr->ia_valid; 1607 struct ceph_mds_request *req; 1608 struct ceph_mds_client *mdsc = ceph_sb_to_client(dentry->d_sb)->mdsc; 1609 int issued; 1610 int release = 0, dirtied = 0; 1611 int mask = 0; 1612 int err = 0; 1613 int inode_dirty_flags = 0; 1614 1615 if (ceph_snap(inode) != CEPH_NOSNAP) 1616 return -EROFS; 1617 1618 err = inode_change_ok(inode, attr); 1619 if (err != 0) 1620 return err; 1621 1622 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR, 1623 USE_AUTH_MDS); 1624 if (IS_ERR(req)) 1625 return PTR_ERR(req); 1626 1627 spin_lock(&ci->i_ceph_lock); 1628 issued = __ceph_caps_issued(ci, NULL); 1629 dout("setattr %p issued %s\n", inode, ceph_cap_string(issued)); 1630 1631 if (ia_valid & ATTR_UID) { 1632 dout("setattr %p uid %d -> %d\n", inode, 1633 from_kuid(&init_user_ns, inode->i_uid), 1634 from_kuid(&init_user_ns, attr->ia_uid)); 1635 if (issued & CEPH_CAP_AUTH_EXCL) { 1636 inode->i_uid = attr->ia_uid; 1637 dirtied |= CEPH_CAP_AUTH_EXCL; 1638 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || 1639 !uid_eq(attr->ia_uid, inode->i_uid)) { 1640 req->r_args.setattr.uid = cpu_to_le32( 1641 from_kuid(&init_user_ns, attr->ia_uid)); 1642 mask |= CEPH_SETATTR_UID; 1643 release |= CEPH_CAP_AUTH_SHARED; 1644 } 1645 } 1646 if (ia_valid & ATTR_GID) { 1647 dout("setattr %p gid %d -> %d\n", inode, 1648 from_kgid(&init_user_ns, inode->i_gid), 1649 from_kgid(&init_user_ns, attr->ia_gid)); 1650 if (issued & CEPH_CAP_AUTH_EXCL) { 1651 inode->i_gid = attr->ia_gid; 1652 dirtied |= CEPH_CAP_AUTH_EXCL; 1653 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || 1654 !gid_eq(attr->ia_gid, inode->i_gid)) { 1655 req->r_args.setattr.gid = cpu_to_le32( 1656 from_kgid(&init_user_ns, attr->ia_gid)); 1657 mask |= CEPH_SETATTR_GID; 1658 release |= CEPH_CAP_AUTH_SHARED; 1659 } 1660 } 1661 if (ia_valid & ATTR_MODE) { 1662 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode, 1663 attr->ia_mode); 1664 if (issued & CEPH_CAP_AUTH_EXCL) { 1665 inode->i_mode = attr->ia_mode; 1666 dirtied |= CEPH_CAP_AUTH_EXCL; 1667 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || 1668 attr->ia_mode != inode->i_mode) { 1669 req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode); 1670 mask |= CEPH_SETATTR_MODE; 1671 release |= CEPH_CAP_AUTH_SHARED; 1672 } 1673 } 1674 1675 if (ia_valid & ATTR_ATIME) { 1676 dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode, 1677 inode->i_atime.tv_sec, inode->i_atime.tv_nsec, 1678 attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec); 1679 if (issued & CEPH_CAP_FILE_EXCL) { 1680 ci->i_time_warp_seq++; 1681 inode->i_atime = attr->ia_atime; 1682 dirtied |= CEPH_CAP_FILE_EXCL; 1683 } else if ((issued & CEPH_CAP_FILE_WR) && 1684 timespec_compare(&inode->i_atime, 1685 &attr->ia_atime) < 0) { 1686 inode->i_atime = attr->ia_atime; 1687 dirtied |= CEPH_CAP_FILE_WR; 1688 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || 1689 !timespec_equal(&inode->i_atime, &attr->ia_atime)) { 1690 ceph_encode_timespec(&req->r_args.setattr.atime, 1691 &attr->ia_atime); 1692 mask |= CEPH_SETATTR_ATIME; 1693 release |= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_RD | 1694 CEPH_CAP_FILE_WR; 1695 } 1696 } 1697 if (ia_valid & ATTR_MTIME) { 1698 dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode, 1699 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec, 1700 attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec); 1701 if (issued & CEPH_CAP_FILE_EXCL) { 1702 ci->i_time_warp_seq++; 1703 inode->i_mtime = attr->ia_mtime; 1704 dirtied |= CEPH_CAP_FILE_EXCL; 1705 } else if ((issued & CEPH_CAP_FILE_WR) && 1706 timespec_compare(&inode->i_mtime, 1707 &attr->ia_mtime) < 0) { 1708 inode->i_mtime = attr->ia_mtime; 1709 dirtied |= CEPH_CAP_FILE_WR; 1710 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || 1711 !timespec_equal(&inode->i_mtime, &attr->ia_mtime)) { 1712 ceph_encode_timespec(&req->r_args.setattr.mtime, 1713 &attr->ia_mtime); 1714 mask |= CEPH_SETATTR_MTIME; 1715 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD | 1716 CEPH_CAP_FILE_WR; 1717 } 1718 } 1719 if (ia_valid & ATTR_SIZE) { 1720 dout("setattr %p size %lld -> %lld\n", inode, 1721 inode->i_size, attr->ia_size); 1722 if (attr->ia_size > inode->i_sb->s_maxbytes) { 1723 err = -EINVAL; 1724 goto out; 1725 } 1726 if ((issued & CEPH_CAP_FILE_EXCL) && 1727 attr->ia_size > inode->i_size) { 1728 inode->i_size = attr->ia_size; 1729 inode->i_blocks = 1730 (attr->ia_size + (1 << 9) - 1) >> 9; 1731 inode->i_ctime = attr->ia_ctime; 1732 ci->i_reported_size = attr->ia_size; 1733 dirtied |= CEPH_CAP_FILE_EXCL; 1734 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || 1735 attr->ia_size != inode->i_size) { 1736 req->r_args.setattr.size = cpu_to_le64(attr->ia_size); 1737 req->r_args.setattr.old_size = 1738 cpu_to_le64(inode->i_size); 1739 mask |= CEPH_SETATTR_SIZE; 1740 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD | 1741 CEPH_CAP_FILE_WR; 1742 } 1743 } 1744 1745 /* these do nothing */ 1746 if (ia_valid & ATTR_CTIME) { 1747 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME| 1748 ATTR_MODE|ATTR_UID|ATTR_GID)) == 0; 1749 dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode, 1750 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec, 1751 attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec, 1752 only ? "ctime only" : "ignored"); 1753 inode->i_ctime = attr->ia_ctime; 1754 if (only) { 1755 /* 1756 * if kernel wants to dirty ctime but nothing else, 1757 * we need to choose a cap to dirty under, or do 1758 * a almost-no-op setattr 1759 */ 1760 if (issued & CEPH_CAP_AUTH_EXCL) 1761 dirtied |= CEPH_CAP_AUTH_EXCL; 1762 else if (issued & CEPH_CAP_FILE_EXCL) 1763 dirtied |= CEPH_CAP_FILE_EXCL; 1764 else if (issued & CEPH_CAP_XATTR_EXCL) 1765 dirtied |= CEPH_CAP_XATTR_EXCL; 1766 else 1767 mask |= CEPH_SETATTR_CTIME; 1768 } 1769 } 1770 if (ia_valid & ATTR_FILE) 1771 dout("setattr %p ATTR_FILE ... hrm!\n", inode); 1772 1773 if (dirtied) { 1774 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied); 1775 inode->i_ctime = CURRENT_TIME; 1776 } 1777 1778 release &= issued; 1779 spin_unlock(&ci->i_ceph_lock); 1780 1781 if (inode_dirty_flags) 1782 __mark_inode_dirty(inode, inode_dirty_flags); 1783 1784 if (mask) { 1785 req->r_inode = inode; 1786 ihold(inode); 1787 req->r_inode_drop = release; 1788 req->r_args.setattr.mask = cpu_to_le32(mask); 1789 req->r_num_caps = 1; 1790 parent_inode = ceph_get_dentry_parent_inode(dentry); 1791 err = ceph_mdsc_do_request(mdsc, parent_inode, req); 1792 iput(parent_inode); 1793 } 1794 dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err, 1795 ceph_cap_string(dirtied), mask); 1796 1797 ceph_mdsc_put_request(req); 1798 if (mask & CEPH_SETATTR_SIZE) 1799 __ceph_do_pending_vmtruncate(inode); 1800 return err; 1801 out: 1802 spin_unlock(&ci->i_ceph_lock); 1803 ceph_mdsc_put_request(req); 1804 return err; 1805 } 1806 1807 /* 1808 * Verify that we have a lease on the given mask. If not, 1809 * do a getattr against an mds. 1810 */ 1811 int ceph_do_getattr(struct inode *inode, int mask) 1812 { 1813 struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb); 1814 struct ceph_mds_client *mdsc = fsc->mdsc; 1815 struct ceph_mds_request *req; 1816 int err; 1817 1818 if (ceph_snap(inode) == CEPH_SNAPDIR) { 1819 dout("do_getattr inode %p SNAPDIR\n", inode); 1820 return 0; 1821 } 1822 1823 dout("do_getattr inode %p mask %s mode 0%o\n", inode, ceph_cap_string(mask), inode->i_mode); 1824 if (ceph_caps_issued_mask(ceph_inode(inode), mask, 1)) 1825 return 0; 1826 1827 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS); 1828 if (IS_ERR(req)) 1829 return PTR_ERR(req); 1830 req->r_inode = inode; 1831 ihold(inode); 1832 req->r_num_caps = 1; 1833 req->r_args.getattr.mask = cpu_to_le32(mask); 1834 err = ceph_mdsc_do_request(mdsc, NULL, req); 1835 ceph_mdsc_put_request(req); 1836 dout("do_getattr result=%d\n", err); 1837 return err; 1838 } 1839 1840 1841 /* 1842 * Check inode permissions. We verify we have a valid value for 1843 * the AUTH cap, then call the generic handler. 1844 */ 1845 int ceph_permission(struct inode *inode, int mask) 1846 { 1847 int err; 1848 1849 if (mask & MAY_NOT_BLOCK) 1850 return -ECHILD; 1851 1852 err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED); 1853 1854 if (!err) 1855 err = generic_permission(inode, mask); 1856 return err; 1857 } 1858 1859 /* 1860 * Get all attributes. Hopefully somedata we'll have a statlite() 1861 * and can limit the fields we require to be accurate. 1862 */ 1863 int ceph_getattr(struct vfsmount *mnt, struct dentry *dentry, 1864 struct kstat *stat) 1865 { 1866 struct inode *inode = dentry->d_inode; 1867 struct ceph_inode_info *ci = ceph_inode(inode); 1868 int err; 1869 1870 err = ceph_do_getattr(inode, CEPH_STAT_CAP_INODE_ALL); 1871 if (!err) { 1872 generic_fillattr(inode, stat); 1873 stat->ino = ceph_translate_ino(inode->i_sb, inode->i_ino); 1874 if (ceph_snap(inode) != CEPH_NOSNAP) 1875 stat->dev = ceph_snap(inode); 1876 else 1877 stat->dev = 0; 1878 if (S_ISDIR(inode->i_mode)) { 1879 if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), 1880 RBYTES)) 1881 stat->size = ci->i_rbytes; 1882 else 1883 stat->size = ci->i_files + ci->i_subdirs; 1884 stat->blocks = 0; 1885 stat->blksize = 65536; 1886 } 1887 } 1888 return err; 1889 } 1890