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