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 = le32_to_cpu(info->uid); 616 inode->i_gid = le32_to_cpu(info->gid); 617 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode, 618 inode->i_uid, inode->i_gid); 619 } 620 621 if ((issued & CEPH_CAP_LINK_EXCL) == 0) 622 set_nlink(inode, le32_to_cpu(info->nlink)); 623 624 /* be careful with mtime, atime, size */ 625 ceph_decode_timespec(&atime, &info->atime); 626 ceph_decode_timespec(&mtime, &info->mtime); 627 ceph_decode_timespec(&ctime, &info->ctime); 628 queue_trunc = ceph_fill_file_size(inode, issued, 629 le32_to_cpu(info->truncate_seq), 630 le64_to_cpu(info->truncate_size), 631 le64_to_cpu(info->size)); 632 ceph_fill_file_time(inode, issued, 633 le32_to_cpu(info->time_warp_seq), 634 &ctime, &mtime, &atime); 635 636 /* only update max_size on auth cap */ 637 if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) && 638 ci->i_max_size != le64_to_cpu(info->max_size)) { 639 dout("max_size %lld -> %llu\n", ci->i_max_size, 640 le64_to_cpu(info->max_size)); 641 ci->i_max_size = le64_to_cpu(info->max_size); 642 } 643 644 ci->i_layout = info->layout; 645 inode->i_blkbits = fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1; 646 647 /* xattrs */ 648 /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */ 649 if ((issued & CEPH_CAP_XATTR_EXCL) == 0 && 650 le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) { 651 if (ci->i_xattrs.blob) 652 ceph_buffer_put(ci->i_xattrs.blob); 653 ci->i_xattrs.blob = xattr_blob; 654 if (xattr_blob) 655 memcpy(ci->i_xattrs.blob->vec.iov_base, 656 iinfo->xattr_data, iinfo->xattr_len); 657 ci->i_xattrs.version = le64_to_cpu(info->xattr_version); 658 xattr_blob = NULL; 659 } 660 661 inode->i_mapping->a_ops = &ceph_aops; 662 inode->i_mapping->backing_dev_info = 663 &ceph_sb_to_client(inode->i_sb)->backing_dev_info; 664 665 switch (inode->i_mode & S_IFMT) { 666 case S_IFIFO: 667 case S_IFBLK: 668 case S_IFCHR: 669 case S_IFSOCK: 670 init_special_inode(inode, inode->i_mode, inode->i_rdev); 671 inode->i_op = &ceph_file_iops; 672 break; 673 case S_IFREG: 674 inode->i_op = &ceph_file_iops; 675 inode->i_fop = &ceph_file_fops; 676 break; 677 case S_IFLNK: 678 inode->i_op = &ceph_symlink_iops; 679 if (!ci->i_symlink) { 680 int symlen = iinfo->symlink_len; 681 char *sym; 682 683 BUG_ON(symlen != inode->i_size); 684 spin_unlock(&ci->i_ceph_lock); 685 686 err = -ENOMEM; 687 sym = kmalloc(symlen+1, GFP_NOFS); 688 if (!sym) 689 goto out; 690 memcpy(sym, iinfo->symlink, symlen); 691 sym[symlen] = 0; 692 693 spin_lock(&ci->i_ceph_lock); 694 if (!ci->i_symlink) 695 ci->i_symlink = sym; 696 else 697 kfree(sym); /* lost a race */ 698 } 699 break; 700 case S_IFDIR: 701 inode->i_op = &ceph_dir_iops; 702 inode->i_fop = &ceph_dir_fops; 703 704 ci->i_dir_layout = iinfo->dir_layout; 705 706 ci->i_files = le64_to_cpu(info->files); 707 ci->i_subdirs = le64_to_cpu(info->subdirs); 708 ci->i_rbytes = le64_to_cpu(info->rbytes); 709 ci->i_rfiles = le64_to_cpu(info->rfiles); 710 ci->i_rsubdirs = le64_to_cpu(info->rsubdirs); 711 ceph_decode_timespec(&ci->i_rctime, &info->rctime); 712 break; 713 default: 714 pr_err("fill_inode %llx.%llx BAD mode 0%o\n", 715 ceph_vinop(inode), inode->i_mode); 716 } 717 718 no_change: 719 spin_unlock(&ci->i_ceph_lock); 720 721 /* queue truncate if we saw i_size decrease */ 722 if (queue_trunc) 723 ceph_queue_vmtruncate(inode); 724 725 /* populate frag tree */ 726 /* FIXME: move me up, if/when version reflects fragtree changes */ 727 nsplits = le32_to_cpu(info->fragtree.nsplits); 728 mutex_lock(&ci->i_fragtree_mutex); 729 for (i = 0; i < nsplits; i++) { 730 u32 id = le32_to_cpu(info->fragtree.splits[i].frag); 731 struct ceph_inode_frag *frag = __get_or_create_frag(ci, id); 732 733 if (IS_ERR(frag)) 734 continue; 735 frag->split_by = le32_to_cpu(info->fragtree.splits[i].by); 736 dout(" frag %x split by %d\n", frag->frag, frag->split_by); 737 } 738 mutex_unlock(&ci->i_fragtree_mutex); 739 740 /* were we issued a capability? */ 741 if (info->cap.caps) { 742 if (ceph_snap(inode) == CEPH_NOSNAP) { 743 ceph_add_cap(inode, session, 744 le64_to_cpu(info->cap.cap_id), 745 cap_fmode, 746 le32_to_cpu(info->cap.caps), 747 le32_to_cpu(info->cap.wanted), 748 le32_to_cpu(info->cap.seq), 749 le32_to_cpu(info->cap.mseq), 750 le64_to_cpu(info->cap.realm), 751 info->cap.flags, 752 caps_reservation); 753 } else { 754 spin_lock(&ci->i_ceph_lock); 755 dout(" %p got snap_caps %s\n", inode, 756 ceph_cap_string(le32_to_cpu(info->cap.caps))); 757 ci->i_snap_caps |= le32_to_cpu(info->cap.caps); 758 if (cap_fmode >= 0) 759 __ceph_get_fmode(ci, cap_fmode); 760 spin_unlock(&ci->i_ceph_lock); 761 } 762 } else if (cap_fmode >= 0) { 763 pr_warning("mds issued no caps on %llx.%llx\n", 764 ceph_vinop(inode)); 765 __ceph_get_fmode(ci, cap_fmode); 766 } 767 768 /* set dir completion flag? */ 769 if (S_ISDIR(inode->i_mode) && 770 updating_inode && /* didn't jump to no_change */ 771 ci->i_files == 0 && ci->i_subdirs == 0 && 772 ceph_snap(inode) == CEPH_NOSNAP && 773 (le32_to_cpu(info->cap.caps) & CEPH_CAP_FILE_SHARED) && 774 (issued & CEPH_CAP_FILE_EXCL) == 0 && 775 !ceph_dir_test_complete(inode)) { 776 dout(" marking %p complete (empty)\n", inode); 777 ceph_dir_set_complete(inode); 778 ci->i_max_offset = 2; 779 } 780 781 /* update delegation info? */ 782 if (dirinfo) 783 ceph_fill_dirfrag(inode, dirinfo); 784 785 err = 0; 786 787 out: 788 if (xattr_blob) 789 ceph_buffer_put(xattr_blob); 790 return err; 791 } 792 793 /* 794 * caller should hold session s_mutex. 795 */ 796 static void update_dentry_lease(struct dentry *dentry, 797 struct ceph_mds_reply_lease *lease, 798 struct ceph_mds_session *session, 799 unsigned long from_time) 800 { 801 struct ceph_dentry_info *di = ceph_dentry(dentry); 802 long unsigned duration = le32_to_cpu(lease->duration_ms); 803 long unsigned ttl = from_time + (duration * HZ) / 1000; 804 long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000; 805 struct inode *dir; 806 807 /* only track leases on regular dentries */ 808 if (dentry->d_op != &ceph_dentry_ops) 809 return; 810 811 spin_lock(&dentry->d_lock); 812 dout("update_dentry_lease %p duration %lu ms ttl %lu\n", 813 dentry, duration, ttl); 814 815 /* make lease_rdcache_gen match directory */ 816 dir = dentry->d_parent->d_inode; 817 di->lease_shared_gen = ceph_inode(dir)->i_shared_gen; 818 819 if (duration == 0) 820 goto out_unlock; 821 822 if (di->lease_gen == session->s_cap_gen && 823 time_before(ttl, dentry->d_time)) 824 goto out_unlock; /* we already have a newer lease. */ 825 826 if (di->lease_session && di->lease_session != session) 827 goto out_unlock; 828 829 ceph_dentry_lru_touch(dentry); 830 831 if (!di->lease_session) 832 di->lease_session = ceph_get_mds_session(session); 833 di->lease_gen = session->s_cap_gen; 834 di->lease_seq = le32_to_cpu(lease->seq); 835 di->lease_renew_after = half_ttl; 836 di->lease_renew_from = 0; 837 dentry->d_time = ttl; 838 out_unlock: 839 spin_unlock(&dentry->d_lock); 840 return; 841 } 842 843 /* 844 * Set dentry's directory position based on the current dir's max, and 845 * order it in d_subdirs, so that dcache_readdir behaves. 846 * 847 * Always called under directory's i_mutex. 848 */ 849 static void ceph_set_dentry_offset(struct dentry *dn) 850 { 851 struct dentry *dir = dn->d_parent; 852 struct inode *inode = dir->d_inode; 853 struct ceph_inode_info *ci; 854 struct ceph_dentry_info *di; 855 856 BUG_ON(!inode); 857 858 ci = ceph_inode(inode); 859 di = ceph_dentry(dn); 860 861 spin_lock(&ci->i_ceph_lock); 862 if (!ceph_dir_test_complete(inode)) { 863 spin_unlock(&ci->i_ceph_lock); 864 return; 865 } 866 di->offset = ceph_inode(inode)->i_max_offset++; 867 spin_unlock(&ci->i_ceph_lock); 868 869 spin_lock(&dir->d_lock); 870 spin_lock_nested(&dn->d_lock, DENTRY_D_LOCK_NESTED); 871 list_move(&dn->d_u.d_child, &dir->d_subdirs); 872 dout("set_dentry_offset %p %lld (%p %p)\n", dn, di->offset, 873 dn->d_u.d_child.prev, dn->d_u.d_child.next); 874 spin_unlock(&dn->d_lock); 875 spin_unlock(&dir->d_lock); 876 } 877 878 /* 879 * splice a dentry to an inode. 880 * caller must hold directory i_mutex for this to be safe. 881 * 882 * we will only rehash the resulting dentry if @prehash is 883 * true; @prehash will be set to false (for the benefit of 884 * the caller) if we fail. 885 */ 886 static struct dentry *splice_dentry(struct dentry *dn, struct inode *in, 887 bool *prehash, bool set_offset) 888 { 889 struct dentry *realdn; 890 891 BUG_ON(dn->d_inode); 892 893 /* dn must be unhashed */ 894 if (!d_unhashed(dn)) 895 d_drop(dn); 896 realdn = d_materialise_unique(dn, in); 897 if (IS_ERR(realdn)) { 898 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n", 899 PTR_ERR(realdn), dn, in, ceph_vinop(in)); 900 if (prehash) 901 *prehash = false; /* don't rehash on error */ 902 dn = realdn; /* note realdn contains the error */ 903 goto out; 904 } else if (realdn) { 905 dout("dn %p (%d) spliced with %p (%d) " 906 "inode %p ino %llx.%llx\n", 907 dn, dn->d_count, 908 realdn, realdn->d_count, 909 realdn->d_inode, ceph_vinop(realdn->d_inode)); 910 dput(dn); 911 dn = realdn; 912 } else { 913 BUG_ON(!ceph_dentry(dn)); 914 dout("dn %p attached to %p ino %llx.%llx\n", 915 dn, dn->d_inode, ceph_vinop(dn->d_inode)); 916 } 917 if ((!prehash || *prehash) && d_unhashed(dn)) 918 d_rehash(dn); 919 if (set_offset) 920 ceph_set_dentry_offset(dn); 921 out: 922 return dn; 923 } 924 925 /* 926 * Incorporate results into the local cache. This is either just 927 * one inode, or a directory, dentry, and possibly linked-to inode (e.g., 928 * after a lookup). 929 * 930 * A reply may contain 931 * a directory inode along with a dentry. 932 * and/or a target inode 933 * 934 * Called with snap_rwsem (read). 935 */ 936 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req, 937 struct ceph_mds_session *session) 938 { 939 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; 940 struct inode *in = NULL; 941 struct ceph_mds_reply_inode *ininfo; 942 struct ceph_vino vino; 943 struct ceph_fs_client *fsc = ceph_sb_to_client(sb); 944 int i = 0; 945 int err = 0; 946 947 dout("fill_trace %p is_dentry %d is_target %d\n", req, 948 rinfo->head->is_dentry, rinfo->head->is_target); 949 950 #if 0 951 /* 952 * Debugging hook: 953 * 954 * If we resend completed ops to a recovering mds, we get no 955 * trace. Since that is very rare, pretend this is the case 956 * to ensure the 'no trace' handlers in the callers behave. 957 * 958 * Fill in inodes unconditionally to avoid breaking cap 959 * invariants. 960 */ 961 if (rinfo->head->op & CEPH_MDS_OP_WRITE) { 962 pr_info("fill_trace faking empty trace on %lld %s\n", 963 req->r_tid, ceph_mds_op_name(rinfo->head->op)); 964 if (rinfo->head->is_dentry) { 965 rinfo->head->is_dentry = 0; 966 err = fill_inode(req->r_locked_dir, 967 &rinfo->diri, rinfo->dirfrag, 968 session, req->r_request_started, -1); 969 } 970 if (rinfo->head->is_target) { 971 rinfo->head->is_target = 0; 972 ininfo = rinfo->targeti.in; 973 vino.ino = le64_to_cpu(ininfo->ino); 974 vino.snap = le64_to_cpu(ininfo->snapid); 975 in = ceph_get_inode(sb, vino); 976 err = fill_inode(in, &rinfo->targeti, NULL, 977 session, req->r_request_started, 978 req->r_fmode); 979 iput(in); 980 } 981 } 982 #endif 983 984 if (!rinfo->head->is_target && !rinfo->head->is_dentry) { 985 dout("fill_trace reply is empty!\n"); 986 if (rinfo->head->result == 0 && req->r_locked_dir) 987 ceph_invalidate_dir_request(req); 988 return 0; 989 } 990 991 if (rinfo->head->is_dentry) { 992 struct inode *dir = req->r_locked_dir; 993 994 err = fill_inode(dir, &rinfo->diri, rinfo->dirfrag, 995 session, req->r_request_started, -1, 996 &req->r_caps_reservation); 997 if (err < 0) 998 return err; 999 } 1000 1001 /* 1002 * ignore null lease/binding on snapdir ENOENT, or else we 1003 * will have trouble splicing in the virtual snapdir later 1004 */ 1005 if (rinfo->head->is_dentry && !req->r_aborted && 1006 (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name, 1007 fsc->mount_options->snapdir_name, 1008 req->r_dentry->d_name.len))) { 1009 /* 1010 * lookup link rename : null -> possibly existing inode 1011 * mknod symlink mkdir : null -> new inode 1012 * unlink : linked -> null 1013 */ 1014 struct inode *dir = req->r_locked_dir; 1015 struct dentry *dn = req->r_dentry; 1016 bool have_dir_cap, have_lease; 1017 1018 BUG_ON(!dn); 1019 BUG_ON(!dir); 1020 BUG_ON(dn->d_parent->d_inode != dir); 1021 BUG_ON(ceph_ino(dir) != 1022 le64_to_cpu(rinfo->diri.in->ino)); 1023 BUG_ON(ceph_snap(dir) != 1024 le64_to_cpu(rinfo->diri.in->snapid)); 1025 1026 /* do we have a lease on the whole dir? */ 1027 have_dir_cap = 1028 (le32_to_cpu(rinfo->diri.in->cap.caps) & 1029 CEPH_CAP_FILE_SHARED); 1030 1031 /* do we have a dn lease? */ 1032 have_lease = have_dir_cap || 1033 le32_to_cpu(rinfo->dlease->duration_ms); 1034 if (!have_lease) 1035 dout("fill_trace no dentry lease or dir cap\n"); 1036 1037 /* rename? */ 1038 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) { 1039 dout(" src %p '%.*s' dst %p '%.*s'\n", 1040 req->r_old_dentry, 1041 req->r_old_dentry->d_name.len, 1042 req->r_old_dentry->d_name.name, 1043 dn, dn->d_name.len, dn->d_name.name); 1044 dout("fill_trace doing d_move %p -> %p\n", 1045 req->r_old_dentry, dn); 1046 1047 d_move(req->r_old_dentry, dn); 1048 dout(" src %p '%.*s' dst %p '%.*s'\n", 1049 req->r_old_dentry, 1050 req->r_old_dentry->d_name.len, 1051 req->r_old_dentry->d_name.name, 1052 dn, dn->d_name.len, dn->d_name.name); 1053 1054 /* ensure target dentry is invalidated, despite 1055 rehashing bug in vfs_rename_dir */ 1056 ceph_invalidate_dentry_lease(dn); 1057 1058 /* 1059 * d_move() puts the renamed dentry at the end of 1060 * d_subdirs. We need to assign it an appropriate 1061 * directory offset so we can behave when holding 1062 * D_COMPLETE. 1063 */ 1064 ceph_set_dentry_offset(req->r_old_dentry); 1065 dout("dn %p gets new offset %lld\n", req->r_old_dentry, 1066 ceph_dentry(req->r_old_dentry)->offset); 1067 1068 dn = req->r_old_dentry; /* use old_dentry */ 1069 in = dn->d_inode; 1070 } 1071 1072 /* null dentry? */ 1073 if (!rinfo->head->is_target) { 1074 dout("fill_trace null dentry\n"); 1075 if (dn->d_inode) { 1076 dout("d_delete %p\n", dn); 1077 d_delete(dn); 1078 } else { 1079 dout("d_instantiate %p NULL\n", dn); 1080 d_instantiate(dn, NULL); 1081 if (have_lease && d_unhashed(dn)) 1082 d_rehash(dn); 1083 update_dentry_lease(dn, rinfo->dlease, 1084 session, 1085 req->r_request_started); 1086 } 1087 goto done; 1088 } 1089 1090 /* attach proper inode */ 1091 ininfo = rinfo->targeti.in; 1092 vino.ino = le64_to_cpu(ininfo->ino); 1093 vino.snap = le64_to_cpu(ininfo->snapid); 1094 in = dn->d_inode; 1095 if (!in) { 1096 in = ceph_get_inode(sb, vino); 1097 if (IS_ERR(in)) { 1098 pr_err("fill_trace bad get_inode " 1099 "%llx.%llx\n", vino.ino, vino.snap); 1100 err = PTR_ERR(in); 1101 d_delete(dn); 1102 goto done; 1103 } 1104 dn = splice_dentry(dn, in, &have_lease, true); 1105 if (IS_ERR(dn)) { 1106 err = PTR_ERR(dn); 1107 goto done; 1108 } 1109 req->r_dentry = dn; /* may have spliced */ 1110 ihold(in); 1111 } else if (ceph_ino(in) == vino.ino && 1112 ceph_snap(in) == vino.snap) { 1113 ihold(in); 1114 } else { 1115 dout(" %p links to %p %llx.%llx, not %llx.%llx\n", 1116 dn, in, ceph_ino(in), ceph_snap(in), 1117 vino.ino, vino.snap); 1118 have_lease = false; 1119 in = NULL; 1120 } 1121 1122 if (have_lease) 1123 update_dentry_lease(dn, rinfo->dlease, session, 1124 req->r_request_started); 1125 dout(" final dn %p\n", dn); 1126 i++; 1127 } else if (req->r_op == CEPH_MDS_OP_LOOKUPSNAP || 1128 req->r_op == CEPH_MDS_OP_MKSNAP) { 1129 struct dentry *dn = req->r_dentry; 1130 1131 /* fill out a snapdir LOOKUPSNAP dentry */ 1132 BUG_ON(!dn); 1133 BUG_ON(!req->r_locked_dir); 1134 BUG_ON(ceph_snap(req->r_locked_dir) != CEPH_SNAPDIR); 1135 ininfo = rinfo->targeti.in; 1136 vino.ino = le64_to_cpu(ininfo->ino); 1137 vino.snap = le64_to_cpu(ininfo->snapid); 1138 in = ceph_get_inode(sb, vino); 1139 if (IS_ERR(in)) { 1140 pr_err("fill_inode get_inode badness %llx.%llx\n", 1141 vino.ino, vino.snap); 1142 err = PTR_ERR(in); 1143 d_delete(dn); 1144 goto done; 1145 } 1146 dout(" linking snapped dir %p to dn %p\n", in, dn); 1147 dn = splice_dentry(dn, in, NULL, true); 1148 if (IS_ERR(dn)) { 1149 err = PTR_ERR(dn); 1150 goto done; 1151 } 1152 req->r_dentry = dn; /* may have spliced */ 1153 ihold(in); 1154 rinfo->head->is_dentry = 1; /* fool notrace handlers */ 1155 } 1156 1157 if (rinfo->head->is_target) { 1158 vino.ino = le64_to_cpu(rinfo->targeti.in->ino); 1159 vino.snap = le64_to_cpu(rinfo->targeti.in->snapid); 1160 1161 if (in == NULL || ceph_ino(in) != vino.ino || 1162 ceph_snap(in) != vino.snap) { 1163 in = ceph_get_inode(sb, vino); 1164 if (IS_ERR(in)) { 1165 err = PTR_ERR(in); 1166 goto done; 1167 } 1168 } 1169 req->r_target_inode = in; 1170 1171 err = fill_inode(in, 1172 &rinfo->targeti, NULL, 1173 session, req->r_request_started, 1174 (le32_to_cpu(rinfo->head->result) == 0) ? 1175 req->r_fmode : -1, 1176 &req->r_caps_reservation); 1177 if (err < 0) { 1178 pr_err("fill_inode badness %p %llx.%llx\n", 1179 in, ceph_vinop(in)); 1180 goto done; 1181 } 1182 } 1183 1184 done: 1185 dout("fill_trace done err=%d\n", err); 1186 return err; 1187 } 1188 1189 /* 1190 * Prepopulate our cache with readdir results, leases, etc. 1191 */ 1192 int ceph_readdir_prepopulate(struct ceph_mds_request *req, 1193 struct ceph_mds_session *session) 1194 { 1195 struct dentry *parent = req->r_dentry; 1196 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; 1197 struct qstr dname; 1198 struct dentry *dn; 1199 struct inode *in; 1200 int err = 0, i; 1201 struct inode *snapdir = NULL; 1202 struct ceph_mds_request_head *rhead = req->r_request->front.iov_base; 1203 u64 frag = le32_to_cpu(rhead->args.readdir.frag); 1204 struct ceph_dentry_info *di; 1205 1206 if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) { 1207 snapdir = ceph_get_snapdir(parent->d_inode); 1208 parent = d_find_alias(snapdir); 1209 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n", 1210 rinfo->dir_nr, parent); 1211 } else { 1212 dout("readdir_prepopulate %d items under dn %p\n", 1213 rinfo->dir_nr, parent); 1214 if (rinfo->dir_dir) 1215 ceph_fill_dirfrag(parent->d_inode, rinfo->dir_dir); 1216 } 1217 1218 for (i = 0; i < rinfo->dir_nr; i++) { 1219 struct ceph_vino vino; 1220 1221 dname.name = rinfo->dir_dname[i]; 1222 dname.len = rinfo->dir_dname_len[i]; 1223 dname.hash = full_name_hash(dname.name, dname.len); 1224 1225 vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino); 1226 vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid); 1227 1228 retry_lookup: 1229 dn = d_lookup(parent, &dname); 1230 dout("d_lookup on parent=%p name=%.*s got %p\n", 1231 parent, dname.len, dname.name, dn); 1232 1233 if (!dn) { 1234 dn = d_alloc(parent, &dname); 1235 dout("d_alloc %p '%.*s' = %p\n", parent, 1236 dname.len, dname.name, dn); 1237 if (dn == NULL) { 1238 dout("d_alloc badness\n"); 1239 err = -ENOMEM; 1240 goto out; 1241 } 1242 err = ceph_init_dentry(dn); 1243 if (err < 0) { 1244 dput(dn); 1245 goto out; 1246 } 1247 } else if (dn->d_inode && 1248 (ceph_ino(dn->d_inode) != vino.ino || 1249 ceph_snap(dn->d_inode) != vino.snap)) { 1250 dout(" dn %p points to wrong inode %p\n", 1251 dn, dn->d_inode); 1252 d_delete(dn); 1253 dput(dn); 1254 goto retry_lookup; 1255 } else { 1256 /* reorder parent's d_subdirs */ 1257 spin_lock(&parent->d_lock); 1258 spin_lock_nested(&dn->d_lock, DENTRY_D_LOCK_NESTED); 1259 list_move(&dn->d_u.d_child, &parent->d_subdirs); 1260 spin_unlock(&dn->d_lock); 1261 spin_unlock(&parent->d_lock); 1262 } 1263 1264 di = dn->d_fsdata; 1265 di->offset = ceph_make_fpos(frag, i + req->r_readdir_offset); 1266 1267 /* inode */ 1268 if (dn->d_inode) { 1269 in = dn->d_inode; 1270 } else { 1271 in = ceph_get_inode(parent->d_sb, vino); 1272 if (IS_ERR(in)) { 1273 dout("new_inode badness\n"); 1274 d_delete(dn); 1275 dput(dn); 1276 err = PTR_ERR(in); 1277 goto out; 1278 } 1279 dn = splice_dentry(dn, in, NULL, false); 1280 if (IS_ERR(dn)) 1281 dn = NULL; 1282 } 1283 1284 if (fill_inode(in, &rinfo->dir_in[i], NULL, session, 1285 req->r_request_started, -1, 1286 &req->r_caps_reservation) < 0) { 1287 pr_err("fill_inode badness on %p\n", in); 1288 goto next_item; 1289 } 1290 if (dn) 1291 update_dentry_lease(dn, rinfo->dir_dlease[i], 1292 req->r_session, 1293 req->r_request_started); 1294 next_item: 1295 if (dn) 1296 dput(dn); 1297 } 1298 req->r_did_prepopulate = true; 1299 1300 out: 1301 if (snapdir) { 1302 iput(snapdir); 1303 dput(parent); 1304 } 1305 dout("readdir_prepopulate done\n"); 1306 return err; 1307 } 1308 1309 int ceph_inode_set_size(struct inode *inode, loff_t size) 1310 { 1311 struct ceph_inode_info *ci = ceph_inode(inode); 1312 int ret = 0; 1313 1314 spin_lock(&ci->i_ceph_lock); 1315 dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size); 1316 inode->i_size = size; 1317 inode->i_blocks = (size + (1 << 9) - 1) >> 9; 1318 1319 /* tell the MDS if we are approaching max_size */ 1320 if ((size << 1) >= ci->i_max_size && 1321 (ci->i_reported_size << 1) < ci->i_max_size) 1322 ret = 1; 1323 1324 spin_unlock(&ci->i_ceph_lock); 1325 return ret; 1326 } 1327 1328 /* 1329 * Write back inode data in a worker thread. (This can't be done 1330 * in the message handler context.) 1331 */ 1332 void ceph_queue_writeback(struct inode *inode) 1333 { 1334 ihold(inode); 1335 if (queue_work(ceph_inode_to_client(inode)->wb_wq, 1336 &ceph_inode(inode)->i_wb_work)) { 1337 dout("ceph_queue_writeback %p\n", inode); 1338 } else { 1339 dout("ceph_queue_writeback %p failed\n", inode); 1340 iput(inode); 1341 } 1342 } 1343 1344 static void ceph_writeback_work(struct work_struct *work) 1345 { 1346 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info, 1347 i_wb_work); 1348 struct inode *inode = &ci->vfs_inode; 1349 1350 dout("writeback %p\n", inode); 1351 filemap_fdatawrite(&inode->i_data); 1352 iput(inode); 1353 } 1354 1355 /* 1356 * queue an async invalidation 1357 */ 1358 void ceph_queue_invalidate(struct inode *inode) 1359 { 1360 ihold(inode); 1361 if (queue_work(ceph_inode_to_client(inode)->pg_inv_wq, 1362 &ceph_inode(inode)->i_pg_inv_work)) { 1363 dout("ceph_queue_invalidate %p\n", inode); 1364 } else { 1365 dout("ceph_queue_invalidate %p failed\n", inode); 1366 iput(inode); 1367 } 1368 } 1369 1370 /* 1371 * Invalidate inode pages in a worker thread. (This can't be done 1372 * in the message handler context.) 1373 */ 1374 static void ceph_invalidate_work(struct work_struct *work) 1375 { 1376 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info, 1377 i_pg_inv_work); 1378 struct inode *inode = &ci->vfs_inode; 1379 u32 orig_gen; 1380 int check = 0; 1381 1382 spin_lock(&ci->i_ceph_lock); 1383 dout("invalidate_pages %p gen %d revoking %d\n", inode, 1384 ci->i_rdcache_gen, ci->i_rdcache_revoking); 1385 if (ci->i_rdcache_revoking != ci->i_rdcache_gen) { 1386 /* nevermind! */ 1387 spin_unlock(&ci->i_ceph_lock); 1388 goto out; 1389 } 1390 orig_gen = ci->i_rdcache_gen; 1391 spin_unlock(&ci->i_ceph_lock); 1392 1393 truncate_inode_pages(&inode->i_data, 0); 1394 1395 spin_lock(&ci->i_ceph_lock); 1396 if (orig_gen == ci->i_rdcache_gen && 1397 orig_gen == ci->i_rdcache_revoking) { 1398 dout("invalidate_pages %p gen %d successful\n", inode, 1399 ci->i_rdcache_gen); 1400 ci->i_rdcache_revoking--; 1401 check = 1; 1402 } else { 1403 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n", 1404 inode, orig_gen, ci->i_rdcache_gen, 1405 ci->i_rdcache_revoking); 1406 } 1407 spin_unlock(&ci->i_ceph_lock); 1408 1409 if (check) 1410 ceph_check_caps(ci, 0, NULL); 1411 out: 1412 iput(inode); 1413 } 1414 1415 1416 /* 1417 * called by trunc_wq; take i_mutex ourselves 1418 * 1419 * We also truncate in a separate thread as well. 1420 */ 1421 static void ceph_vmtruncate_work(struct work_struct *work) 1422 { 1423 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info, 1424 i_vmtruncate_work); 1425 struct inode *inode = &ci->vfs_inode; 1426 1427 dout("vmtruncate_work %p\n", inode); 1428 mutex_lock(&inode->i_mutex); 1429 __ceph_do_pending_vmtruncate(inode); 1430 mutex_unlock(&inode->i_mutex); 1431 iput(inode); 1432 } 1433 1434 /* 1435 * Queue an async vmtruncate. If we fail to queue work, we will handle 1436 * the truncation the next time we call __ceph_do_pending_vmtruncate. 1437 */ 1438 void ceph_queue_vmtruncate(struct inode *inode) 1439 { 1440 struct ceph_inode_info *ci = ceph_inode(inode); 1441 1442 ihold(inode); 1443 if (queue_work(ceph_sb_to_client(inode->i_sb)->trunc_wq, 1444 &ci->i_vmtruncate_work)) { 1445 dout("ceph_queue_vmtruncate %p\n", inode); 1446 } else { 1447 dout("ceph_queue_vmtruncate %p failed, pending=%d\n", 1448 inode, ci->i_truncate_pending); 1449 iput(inode); 1450 } 1451 } 1452 1453 /* 1454 * called with i_mutex held. 1455 * 1456 * Make sure any pending truncation is applied before doing anything 1457 * that may depend on it. 1458 */ 1459 void __ceph_do_pending_vmtruncate(struct inode *inode) 1460 { 1461 struct ceph_inode_info *ci = ceph_inode(inode); 1462 u64 to; 1463 int wrbuffer_refs, wake = 0; 1464 1465 retry: 1466 spin_lock(&ci->i_ceph_lock); 1467 if (ci->i_truncate_pending == 0) { 1468 dout("__do_pending_vmtruncate %p none pending\n", inode); 1469 spin_unlock(&ci->i_ceph_lock); 1470 return; 1471 } 1472 1473 /* 1474 * make sure any dirty snapped pages are flushed before we 1475 * possibly truncate them.. so write AND block! 1476 */ 1477 if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) { 1478 dout("__do_pending_vmtruncate %p flushing snaps first\n", 1479 inode); 1480 spin_unlock(&ci->i_ceph_lock); 1481 filemap_write_and_wait_range(&inode->i_data, 0, 1482 inode->i_sb->s_maxbytes); 1483 goto retry; 1484 } 1485 1486 to = ci->i_truncate_size; 1487 wrbuffer_refs = ci->i_wrbuffer_ref; 1488 dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode, 1489 ci->i_truncate_pending, to); 1490 spin_unlock(&ci->i_ceph_lock); 1491 1492 truncate_inode_pages(inode->i_mapping, to); 1493 1494 spin_lock(&ci->i_ceph_lock); 1495 ci->i_truncate_pending--; 1496 if (ci->i_truncate_pending == 0) 1497 wake = 1; 1498 spin_unlock(&ci->i_ceph_lock); 1499 1500 if (wrbuffer_refs == 0) 1501 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL); 1502 if (wake) 1503 wake_up_all(&ci->i_cap_wq); 1504 } 1505 1506 1507 /* 1508 * symlinks 1509 */ 1510 static void *ceph_sym_follow_link(struct dentry *dentry, struct nameidata *nd) 1511 { 1512 struct ceph_inode_info *ci = ceph_inode(dentry->d_inode); 1513 nd_set_link(nd, ci->i_symlink); 1514 return NULL; 1515 } 1516 1517 static const struct inode_operations ceph_symlink_iops = { 1518 .readlink = generic_readlink, 1519 .follow_link = ceph_sym_follow_link, 1520 }; 1521 1522 /* 1523 * setattr 1524 */ 1525 int ceph_setattr(struct dentry *dentry, struct iattr *attr) 1526 { 1527 struct inode *inode = dentry->d_inode; 1528 struct ceph_inode_info *ci = ceph_inode(inode); 1529 struct inode *parent_inode; 1530 const unsigned int ia_valid = attr->ia_valid; 1531 struct ceph_mds_request *req; 1532 struct ceph_mds_client *mdsc = ceph_sb_to_client(dentry->d_sb)->mdsc; 1533 int issued; 1534 int release = 0, dirtied = 0; 1535 int mask = 0; 1536 int err = 0; 1537 int inode_dirty_flags = 0; 1538 1539 if (ceph_snap(inode) != CEPH_NOSNAP) 1540 return -EROFS; 1541 1542 __ceph_do_pending_vmtruncate(inode); 1543 1544 err = inode_change_ok(inode, attr); 1545 if (err != 0) 1546 return err; 1547 1548 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR, 1549 USE_AUTH_MDS); 1550 if (IS_ERR(req)) 1551 return PTR_ERR(req); 1552 1553 spin_lock(&ci->i_ceph_lock); 1554 issued = __ceph_caps_issued(ci, NULL); 1555 dout("setattr %p issued %s\n", inode, ceph_cap_string(issued)); 1556 1557 if (ia_valid & ATTR_UID) { 1558 dout("setattr %p uid %d -> %d\n", inode, 1559 inode->i_uid, attr->ia_uid); 1560 if (issued & CEPH_CAP_AUTH_EXCL) { 1561 inode->i_uid = attr->ia_uid; 1562 dirtied |= CEPH_CAP_AUTH_EXCL; 1563 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || 1564 attr->ia_uid != inode->i_uid) { 1565 req->r_args.setattr.uid = cpu_to_le32(attr->ia_uid); 1566 mask |= CEPH_SETATTR_UID; 1567 release |= CEPH_CAP_AUTH_SHARED; 1568 } 1569 } 1570 if (ia_valid & ATTR_GID) { 1571 dout("setattr %p gid %d -> %d\n", inode, 1572 inode->i_gid, attr->ia_gid); 1573 if (issued & CEPH_CAP_AUTH_EXCL) { 1574 inode->i_gid = attr->ia_gid; 1575 dirtied |= CEPH_CAP_AUTH_EXCL; 1576 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || 1577 attr->ia_gid != inode->i_gid) { 1578 req->r_args.setattr.gid = cpu_to_le32(attr->ia_gid); 1579 mask |= CEPH_SETATTR_GID; 1580 release |= CEPH_CAP_AUTH_SHARED; 1581 } 1582 } 1583 if (ia_valid & ATTR_MODE) { 1584 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode, 1585 attr->ia_mode); 1586 if (issued & CEPH_CAP_AUTH_EXCL) { 1587 inode->i_mode = attr->ia_mode; 1588 dirtied |= CEPH_CAP_AUTH_EXCL; 1589 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || 1590 attr->ia_mode != inode->i_mode) { 1591 req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode); 1592 mask |= CEPH_SETATTR_MODE; 1593 release |= CEPH_CAP_AUTH_SHARED; 1594 } 1595 } 1596 1597 if (ia_valid & ATTR_ATIME) { 1598 dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode, 1599 inode->i_atime.tv_sec, inode->i_atime.tv_nsec, 1600 attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec); 1601 if (issued & CEPH_CAP_FILE_EXCL) { 1602 ci->i_time_warp_seq++; 1603 inode->i_atime = attr->ia_atime; 1604 dirtied |= CEPH_CAP_FILE_EXCL; 1605 } else if ((issued & CEPH_CAP_FILE_WR) && 1606 timespec_compare(&inode->i_atime, 1607 &attr->ia_atime) < 0) { 1608 inode->i_atime = attr->ia_atime; 1609 dirtied |= CEPH_CAP_FILE_WR; 1610 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || 1611 !timespec_equal(&inode->i_atime, &attr->ia_atime)) { 1612 ceph_encode_timespec(&req->r_args.setattr.atime, 1613 &attr->ia_atime); 1614 mask |= CEPH_SETATTR_ATIME; 1615 release |= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_RD | 1616 CEPH_CAP_FILE_WR; 1617 } 1618 } 1619 if (ia_valid & ATTR_MTIME) { 1620 dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode, 1621 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec, 1622 attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec); 1623 if (issued & CEPH_CAP_FILE_EXCL) { 1624 ci->i_time_warp_seq++; 1625 inode->i_mtime = attr->ia_mtime; 1626 dirtied |= CEPH_CAP_FILE_EXCL; 1627 } else if ((issued & CEPH_CAP_FILE_WR) && 1628 timespec_compare(&inode->i_mtime, 1629 &attr->ia_mtime) < 0) { 1630 inode->i_mtime = attr->ia_mtime; 1631 dirtied |= CEPH_CAP_FILE_WR; 1632 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || 1633 !timespec_equal(&inode->i_mtime, &attr->ia_mtime)) { 1634 ceph_encode_timespec(&req->r_args.setattr.mtime, 1635 &attr->ia_mtime); 1636 mask |= CEPH_SETATTR_MTIME; 1637 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD | 1638 CEPH_CAP_FILE_WR; 1639 } 1640 } 1641 if (ia_valid & ATTR_SIZE) { 1642 dout("setattr %p size %lld -> %lld\n", inode, 1643 inode->i_size, attr->ia_size); 1644 if (attr->ia_size > inode->i_sb->s_maxbytes) { 1645 err = -EINVAL; 1646 goto out; 1647 } 1648 if ((issued & CEPH_CAP_FILE_EXCL) && 1649 attr->ia_size > inode->i_size) { 1650 inode->i_size = attr->ia_size; 1651 inode->i_blocks = 1652 (attr->ia_size + (1 << 9) - 1) >> 9; 1653 inode->i_ctime = attr->ia_ctime; 1654 ci->i_reported_size = attr->ia_size; 1655 dirtied |= CEPH_CAP_FILE_EXCL; 1656 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || 1657 attr->ia_size != inode->i_size) { 1658 req->r_args.setattr.size = cpu_to_le64(attr->ia_size); 1659 req->r_args.setattr.old_size = 1660 cpu_to_le64(inode->i_size); 1661 mask |= CEPH_SETATTR_SIZE; 1662 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD | 1663 CEPH_CAP_FILE_WR; 1664 } 1665 } 1666 1667 /* these do nothing */ 1668 if (ia_valid & ATTR_CTIME) { 1669 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME| 1670 ATTR_MODE|ATTR_UID|ATTR_GID)) == 0; 1671 dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode, 1672 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec, 1673 attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec, 1674 only ? "ctime only" : "ignored"); 1675 inode->i_ctime = attr->ia_ctime; 1676 if (only) { 1677 /* 1678 * if kernel wants to dirty ctime but nothing else, 1679 * we need to choose a cap to dirty under, or do 1680 * a almost-no-op setattr 1681 */ 1682 if (issued & CEPH_CAP_AUTH_EXCL) 1683 dirtied |= CEPH_CAP_AUTH_EXCL; 1684 else if (issued & CEPH_CAP_FILE_EXCL) 1685 dirtied |= CEPH_CAP_FILE_EXCL; 1686 else if (issued & CEPH_CAP_XATTR_EXCL) 1687 dirtied |= CEPH_CAP_XATTR_EXCL; 1688 else 1689 mask |= CEPH_SETATTR_CTIME; 1690 } 1691 } 1692 if (ia_valid & ATTR_FILE) 1693 dout("setattr %p ATTR_FILE ... hrm!\n", inode); 1694 1695 if (dirtied) { 1696 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied); 1697 inode->i_ctime = CURRENT_TIME; 1698 } 1699 1700 release &= issued; 1701 spin_unlock(&ci->i_ceph_lock); 1702 1703 if (inode_dirty_flags) 1704 __mark_inode_dirty(inode, inode_dirty_flags); 1705 1706 if (mask) { 1707 req->r_inode = inode; 1708 ihold(inode); 1709 req->r_inode_drop = release; 1710 req->r_args.setattr.mask = cpu_to_le32(mask); 1711 req->r_num_caps = 1; 1712 parent_inode = ceph_get_dentry_parent_inode(dentry); 1713 err = ceph_mdsc_do_request(mdsc, parent_inode, req); 1714 iput(parent_inode); 1715 } 1716 dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err, 1717 ceph_cap_string(dirtied), mask); 1718 1719 ceph_mdsc_put_request(req); 1720 __ceph_do_pending_vmtruncate(inode); 1721 return err; 1722 out: 1723 spin_unlock(&ci->i_ceph_lock); 1724 ceph_mdsc_put_request(req); 1725 return err; 1726 } 1727 1728 /* 1729 * Verify that we have a lease on the given mask. If not, 1730 * do a getattr against an mds. 1731 */ 1732 int ceph_do_getattr(struct inode *inode, int mask) 1733 { 1734 struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb); 1735 struct ceph_mds_client *mdsc = fsc->mdsc; 1736 struct ceph_mds_request *req; 1737 int err; 1738 1739 if (ceph_snap(inode) == CEPH_SNAPDIR) { 1740 dout("do_getattr inode %p SNAPDIR\n", inode); 1741 return 0; 1742 } 1743 1744 dout("do_getattr inode %p mask %s mode 0%o\n", inode, ceph_cap_string(mask), inode->i_mode); 1745 if (ceph_caps_issued_mask(ceph_inode(inode), mask, 1)) 1746 return 0; 1747 1748 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS); 1749 if (IS_ERR(req)) 1750 return PTR_ERR(req); 1751 req->r_inode = inode; 1752 ihold(inode); 1753 req->r_num_caps = 1; 1754 req->r_args.getattr.mask = cpu_to_le32(mask); 1755 err = ceph_mdsc_do_request(mdsc, NULL, req); 1756 ceph_mdsc_put_request(req); 1757 dout("do_getattr result=%d\n", err); 1758 return err; 1759 } 1760 1761 1762 /* 1763 * Check inode permissions. We verify we have a valid value for 1764 * the AUTH cap, then call the generic handler. 1765 */ 1766 int ceph_permission(struct inode *inode, int mask) 1767 { 1768 int err; 1769 1770 if (mask & MAY_NOT_BLOCK) 1771 return -ECHILD; 1772 1773 err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED); 1774 1775 if (!err) 1776 err = generic_permission(inode, mask); 1777 return err; 1778 } 1779 1780 /* 1781 * Get all attributes. Hopefully somedata we'll have a statlite() 1782 * and can limit the fields we require to be accurate. 1783 */ 1784 int ceph_getattr(struct vfsmount *mnt, struct dentry *dentry, 1785 struct kstat *stat) 1786 { 1787 struct inode *inode = dentry->d_inode; 1788 struct ceph_inode_info *ci = ceph_inode(inode); 1789 int err; 1790 1791 err = ceph_do_getattr(inode, CEPH_STAT_CAP_INODE_ALL); 1792 if (!err) { 1793 generic_fillattr(inode, stat); 1794 stat->ino = ceph_translate_ino(inode->i_sb, inode->i_ino); 1795 if (ceph_snap(inode) != CEPH_NOSNAP) 1796 stat->dev = ceph_snap(inode); 1797 else 1798 stat->dev = 0; 1799 if (S_ISDIR(inode->i_mode)) { 1800 if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), 1801 RBYTES)) 1802 stat->size = ci->i_rbytes; 1803 else 1804 stat->size = ci->i_files + ci->i_subdirs; 1805 stat->blocks = 0; 1806 stat->blksize = 65536; 1807 } 1808 } 1809 return err; 1810 } 1811