1 #include <linux/ceph/ceph_debug.h> 2 3 #include <linux/module.h> 4 #include <linux/fs.h> 5 #include <linux/slab.h> 6 #include <linux/string.h> 7 #include <linux/uaccess.h> 8 #include <linux/kernel.h> 9 #include <linux/namei.h> 10 #include <linux/writeback.h> 11 #include <linux/vmalloc.h> 12 #include <linux/pagevec.h> 13 14 #include "super.h" 15 #include "mds_client.h" 16 #include <linux/ceph/decode.h> 17 18 /* 19 * Ceph inode operations 20 * 21 * Implement basic inode helpers (get, alloc) and inode ops (getattr, 22 * setattr, etc.), xattr helpers, and helpers for assimilating 23 * metadata returned by the MDS into our cache. 24 * 25 * Also define helpers for doing asynchronous writeback, invalidation, 26 * and truncation for the benefit of those who can't afford to block 27 * (typically because they are in the message handler path). 28 */ 29 30 static const struct inode_operations ceph_symlink_iops; 31 32 static void ceph_invalidate_work(struct work_struct *work); 33 static void ceph_writeback_work(struct work_struct *work); 34 static void ceph_vmtruncate_work(struct work_struct *work); 35 36 /* 37 * find or create an inode, given the ceph ino number 38 */ 39 static int ceph_set_ino_cb(struct inode *inode, void *data) 40 { 41 ceph_inode(inode)->i_vino = *(struct ceph_vino *)data; 42 inode->i_ino = ceph_vino_to_ino(*(struct ceph_vino *)data); 43 return 0; 44 } 45 46 struct inode *ceph_get_inode(struct super_block *sb, struct ceph_vino vino) 47 { 48 struct inode *inode; 49 ino_t t = ceph_vino_to_ino(vino); 50 51 inode = iget5_locked(sb, t, ceph_ino_compare, ceph_set_ino_cb, &vino); 52 if (inode == NULL) 53 return ERR_PTR(-ENOMEM); 54 if (inode->i_state & I_NEW) { 55 dout("get_inode created new inode %p %llx.%llx ino %llx\n", 56 inode, ceph_vinop(inode), (u64)inode->i_ino); 57 unlock_new_inode(inode); 58 } 59 60 dout("get_inode on %lu=%llx.%llx got %p\n", inode->i_ino, vino.ino, 61 vino.snap, inode); 62 return inode; 63 } 64 65 /* 66 * get/constuct snapdir inode for a given directory 67 */ 68 struct inode *ceph_get_snapdir(struct inode *parent) 69 { 70 struct ceph_vino vino = { 71 .ino = ceph_ino(parent), 72 .snap = CEPH_SNAPDIR, 73 }; 74 struct inode *inode = ceph_get_inode(parent->i_sb, vino); 75 struct ceph_inode_info *ci = ceph_inode(inode); 76 77 BUG_ON(!S_ISDIR(parent->i_mode)); 78 if (IS_ERR(inode)) 79 return inode; 80 inode->i_mode = parent->i_mode; 81 inode->i_uid = parent->i_uid; 82 inode->i_gid = parent->i_gid; 83 inode->i_op = &ceph_dir_iops; 84 inode->i_fop = &ceph_dir_fops; 85 ci->i_snap_caps = CEPH_CAP_PIN; /* so we can open */ 86 ci->i_rbytes = 0; 87 return inode; 88 } 89 90 const struct inode_operations ceph_file_iops = { 91 .permission = ceph_permission, 92 .setattr = ceph_setattr, 93 .getattr = ceph_getattr, 94 .setxattr = ceph_setxattr, 95 .getxattr = ceph_getxattr, 96 .listxattr = ceph_listxattr, 97 .removexattr = ceph_removexattr, 98 }; 99 100 101 /* 102 * We use a 'frag tree' to keep track of the MDS's directory fragments 103 * for a given inode (usually there is just a single fragment). We 104 * need to know when a child frag is delegated to a new MDS, or when 105 * it is flagged as replicated, so we can direct our requests 106 * accordingly. 107 */ 108 109 /* 110 * find/create a frag in the tree 111 */ 112 static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci, 113 u32 f) 114 { 115 struct rb_node **p; 116 struct rb_node *parent = NULL; 117 struct ceph_inode_frag *frag; 118 int c; 119 120 p = &ci->i_fragtree.rb_node; 121 while (*p) { 122 parent = *p; 123 frag = rb_entry(parent, struct ceph_inode_frag, node); 124 c = ceph_frag_compare(f, frag->frag); 125 if (c < 0) 126 p = &(*p)->rb_left; 127 else if (c > 0) 128 p = &(*p)->rb_right; 129 else 130 return frag; 131 } 132 133 frag = kmalloc(sizeof(*frag), GFP_NOFS); 134 if (!frag) { 135 pr_err("__get_or_create_frag ENOMEM on %p %llx.%llx " 136 "frag %x\n", &ci->vfs_inode, 137 ceph_vinop(&ci->vfs_inode), f); 138 return ERR_PTR(-ENOMEM); 139 } 140 frag->frag = f; 141 frag->split_by = 0; 142 frag->mds = -1; 143 frag->ndist = 0; 144 145 rb_link_node(&frag->node, parent, p); 146 rb_insert_color(&frag->node, &ci->i_fragtree); 147 148 dout("get_or_create_frag added %llx.%llx frag %x\n", 149 ceph_vinop(&ci->vfs_inode), f); 150 return frag; 151 } 152 153 /* 154 * find a specific frag @f 155 */ 156 struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f) 157 { 158 struct rb_node *n = ci->i_fragtree.rb_node; 159 160 while (n) { 161 struct ceph_inode_frag *frag = 162 rb_entry(n, struct ceph_inode_frag, node); 163 int c = ceph_frag_compare(f, frag->frag); 164 if (c < 0) 165 n = n->rb_left; 166 else if (c > 0) 167 n = n->rb_right; 168 else 169 return frag; 170 } 171 return NULL; 172 } 173 174 /* 175 * Choose frag containing the given value @v. If @pfrag is 176 * specified, copy the frag delegation info to the caller if 177 * it is present. 178 */ 179 u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v, 180 struct ceph_inode_frag *pfrag, 181 int *found) 182 { 183 u32 t = ceph_frag_make(0, 0); 184 struct ceph_inode_frag *frag; 185 unsigned nway, i; 186 u32 n; 187 188 if (found) 189 *found = 0; 190 191 mutex_lock(&ci->i_fragtree_mutex); 192 while (1) { 193 WARN_ON(!ceph_frag_contains_value(t, v)); 194 frag = __ceph_find_frag(ci, t); 195 if (!frag) 196 break; /* t is a leaf */ 197 if (frag->split_by == 0) { 198 if (pfrag) 199 memcpy(pfrag, frag, sizeof(*pfrag)); 200 if (found) 201 *found = 1; 202 break; 203 } 204 205 /* choose child */ 206 nway = 1 << frag->split_by; 207 dout("choose_frag(%x) %x splits by %d (%d ways)\n", v, t, 208 frag->split_by, nway); 209 for (i = 0; i < nway; i++) { 210 n = ceph_frag_make_child(t, frag->split_by, i); 211 if (ceph_frag_contains_value(n, v)) { 212 t = n; 213 break; 214 } 215 } 216 BUG_ON(i == nway); 217 } 218 dout("choose_frag(%x) = %x\n", v, t); 219 220 mutex_unlock(&ci->i_fragtree_mutex); 221 return t; 222 } 223 224 /* 225 * Process dirfrag (delegation) info from the mds. Include leaf 226 * fragment in tree ONLY if ndist > 0. Otherwise, only 227 * branches/splits are included in i_fragtree) 228 */ 229 static int ceph_fill_dirfrag(struct inode *inode, 230 struct ceph_mds_reply_dirfrag *dirinfo) 231 { 232 struct ceph_inode_info *ci = ceph_inode(inode); 233 struct ceph_inode_frag *frag; 234 u32 id = le32_to_cpu(dirinfo->frag); 235 int mds = le32_to_cpu(dirinfo->auth); 236 int ndist = le32_to_cpu(dirinfo->ndist); 237 int i; 238 int err = 0; 239 240 mutex_lock(&ci->i_fragtree_mutex); 241 if (ndist == 0) { 242 /* no delegation info needed. */ 243 frag = __ceph_find_frag(ci, id); 244 if (!frag) 245 goto out; 246 if (frag->split_by == 0) { 247 /* tree leaf, remove */ 248 dout("fill_dirfrag removed %llx.%llx frag %x" 249 " (no ref)\n", ceph_vinop(inode), id); 250 rb_erase(&frag->node, &ci->i_fragtree); 251 kfree(frag); 252 } else { 253 /* tree branch, keep and clear */ 254 dout("fill_dirfrag cleared %llx.%llx frag %x" 255 " referral\n", ceph_vinop(inode), id); 256 frag->mds = -1; 257 frag->ndist = 0; 258 } 259 goto out; 260 } 261 262 263 /* find/add this frag to store mds delegation info */ 264 frag = __get_or_create_frag(ci, id); 265 if (IS_ERR(frag)) { 266 /* this is not the end of the world; we can continue 267 with bad/inaccurate delegation info */ 268 pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n", 269 ceph_vinop(inode), le32_to_cpu(dirinfo->frag)); 270 err = -ENOMEM; 271 goto out; 272 } 273 274 frag->mds = mds; 275 frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP); 276 for (i = 0; i < frag->ndist; i++) 277 frag->dist[i] = le32_to_cpu(dirinfo->dist[i]); 278 dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n", 279 ceph_vinop(inode), frag->frag, frag->ndist); 280 281 out: 282 mutex_unlock(&ci->i_fragtree_mutex); 283 return err; 284 } 285 286 287 /* 288 * initialize a newly allocated inode. 289 */ 290 struct inode *ceph_alloc_inode(struct super_block *sb) 291 { 292 struct ceph_inode_info *ci; 293 int i; 294 295 ci = kmem_cache_alloc(ceph_inode_cachep, GFP_NOFS); 296 if (!ci) 297 return NULL; 298 299 dout("alloc_inode %p\n", &ci->vfs_inode); 300 301 ci->i_version = 0; 302 ci->i_time_warp_seq = 0; 303 ci->i_ceph_flags = 0; 304 ci->i_release_count = 0; 305 ci->i_symlink = NULL; 306 307 memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout)); 308 309 ci->i_fragtree = RB_ROOT; 310 mutex_init(&ci->i_fragtree_mutex); 311 312 ci->i_xattrs.blob = NULL; 313 ci->i_xattrs.prealloc_blob = NULL; 314 ci->i_xattrs.dirty = false; 315 ci->i_xattrs.index = RB_ROOT; 316 ci->i_xattrs.count = 0; 317 ci->i_xattrs.names_size = 0; 318 ci->i_xattrs.vals_size = 0; 319 ci->i_xattrs.version = 0; 320 ci->i_xattrs.index_version = 0; 321 322 ci->i_caps = RB_ROOT; 323 ci->i_auth_cap = NULL; 324 ci->i_dirty_caps = 0; 325 ci->i_flushing_caps = 0; 326 INIT_LIST_HEAD(&ci->i_dirty_item); 327 INIT_LIST_HEAD(&ci->i_flushing_item); 328 ci->i_cap_flush_seq = 0; 329 ci->i_cap_flush_last_tid = 0; 330 memset(&ci->i_cap_flush_tid, 0, sizeof(ci->i_cap_flush_tid)); 331 init_waitqueue_head(&ci->i_cap_wq); 332 ci->i_hold_caps_min = 0; 333 ci->i_hold_caps_max = 0; 334 INIT_LIST_HEAD(&ci->i_cap_delay_list); 335 ci->i_cap_exporting_mds = 0; 336 ci->i_cap_exporting_mseq = 0; 337 ci->i_cap_exporting_issued = 0; 338 INIT_LIST_HEAD(&ci->i_cap_snaps); 339 ci->i_head_snapc = NULL; 340 ci->i_snap_caps = 0; 341 342 for (i = 0; i < CEPH_FILE_MODE_NUM; i++) 343 ci->i_nr_by_mode[i] = 0; 344 345 ci->i_truncate_seq = 0; 346 ci->i_truncate_size = 0; 347 ci->i_truncate_pending = 0; 348 349 ci->i_max_size = 0; 350 ci->i_reported_size = 0; 351 ci->i_wanted_max_size = 0; 352 ci->i_requested_max_size = 0; 353 354 ci->i_pin_ref = 0; 355 ci->i_rd_ref = 0; 356 ci->i_rdcache_ref = 0; 357 ci->i_wr_ref = 0; 358 ci->i_wrbuffer_ref = 0; 359 ci->i_wrbuffer_ref_head = 0; 360 ci->i_shared_gen = 0; 361 ci->i_rdcache_gen = 0; 362 ci->i_rdcache_revoking = 0; 363 364 INIT_LIST_HEAD(&ci->i_unsafe_writes); 365 INIT_LIST_HEAD(&ci->i_unsafe_dirops); 366 spin_lock_init(&ci->i_unsafe_lock); 367 368 ci->i_snap_realm = NULL; 369 INIT_LIST_HEAD(&ci->i_snap_realm_item); 370 INIT_LIST_HEAD(&ci->i_snap_flush_item); 371 372 INIT_WORK(&ci->i_wb_work, ceph_writeback_work); 373 INIT_WORK(&ci->i_pg_inv_work, ceph_invalidate_work); 374 375 INIT_WORK(&ci->i_vmtruncate_work, ceph_vmtruncate_work); 376 377 return &ci->vfs_inode; 378 } 379 380 static void ceph_i_callback(struct rcu_head *head) 381 { 382 struct inode *inode = container_of(head, struct inode, i_rcu); 383 struct ceph_inode_info *ci = ceph_inode(inode); 384 385 INIT_LIST_HEAD(&inode->i_dentry); 386 kmem_cache_free(ceph_inode_cachep, ci); 387 } 388 389 void ceph_destroy_inode(struct inode *inode) 390 { 391 struct ceph_inode_info *ci = ceph_inode(inode); 392 struct ceph_inode_frag *frag; 393 struct rb_node *n; 394 395 dout("destroy_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode)); 396 397 ceph_queue_caps_release(inode); 398 399 /* 400 * we may still have a snap_realm reference if there are stray 401 * caps in i_cap_exporting_issued or i_snap_caps. 402 */ 403 if (ci->i_snap_realm) { 404 struct ceph_mds_client *mdsc = 405 ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc; 406 struct ceph_snap_realm *realm = ci->i_snap_realm; 407 408 dout(" dropping residual ref to snap realm %p\n", realm); 409 spin_lock(&realm->inodes_with_caps_lock); 410 list_del_init(&ci->i_snap_realm_item); 411 spin_unlock(&realm->inodes_with_caps_lock); 412 ceph_put_snap_realm(mdsc, realm); 413 } 414 415 kfree(ci->i_symlink); 416 while ((n = rb_first(&ci->i_fragtree)) != NULL) { 417 frag = rb_entry(n, struct ceph_inode_frag, node); 418 rb_erase(n, &ci->i_fragtree); 419 kfree(frag); 420 } 421 422 __ceph_destroy_xattrs(ci); 423 if (ci->i_xattrs.blob) 424 ceph_buffer_put(ci->i_xattrs.blob); 425 if (ci->i_xattrs.prealloc_blob) 426 ceph_buffer_put(ci->i_xattrs.prealloc_blob); 427 428 call_rcu(&inode->i_rcu, ceph_i_callback); 429 } 430 431 432 /* 433 * Helpers to fill in size, ctime, mtime, and atime. We have to be 434 * careful because either the client or MDS may have more up to date 435 * info, depending on which capabilities are held, and whether 436 * time_warp_seq or truncate_seq have increased. (Ordinarily, mtime 437 * and size are monotonically increasing, except when utimes() or 438 * truncate() increments the corresponding _seq values.) 439 */ 440 int ceph_fill_file_size(struct inode *inode, int issued, 441 u32 truncate_seq, u64 truncate_size, u64 size) 442 { 443 struct ceph_inode_info *ci = ceph_inode(inode); 444 int queue_trunc = 0; 445 446 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 || 447 (truncate_seq == ci->i_truncate_seq && size > inode->i_size)) { 448 dout("size %lld -> %llu\n", inode->i_size, size); 449 inode->i_size = size; 450 inode->i_blocks = (size + (1<<9) - 1) >> 9; 451 ci->i_reported_size = size; 452 if (truncate_seq != ci->i_truncate_seq) { 453 dout("truncate_seq %u -> %u\n", 454 ci->i_truncate_seq, truncate_seq); 455 ci->i_truncate_seq = truncate_seq; 456 /* 457 * If we hold relevant caps, or in the case where we're 458 * not the only client referencing this file and we 459 * don't hold those caps, then we need to check whether 460 * the file is either opened or mmaped 461 */ 462 if ((issued & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_RD| 463 CEPH_CAP_FILE_WR|CEPH_CAP_FILE_BUFFER| 464 CEPH_CAP_FILE_EXCL| 465 CEPH_CAP_FILE_LAZYIO)) || 466 mapping_mapped(inode->i_mapping) || 467 __ceph_caps_file_wanted(ci)) { 468 ci->i_truncate_pending++; 469 queue_trunc = 1; 470 } 471 } 472 } 473 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 && 474 ci->i_truncate_size != truncate_size) { 475 dout("truncate_size %lld -> %llu\n", ci->i_truncate_size, 476 truncate_size); 477 ci->i_truncate_size = truncate_size; 478 } 479 return queue_trunc; 480 } 481 482 void ceph_fill_file_time(struct inode *inode, int issued, 483 u64 time_warp_seq, struct timespec *ctime, 484 struct timespec *mtime, struct timespec *atime) 485 { 486 struct ceph_inode_info *ci = ceph_inode(inode); 487 int warn = 0; 488 489 if (issued & (CEPH_CAP_FILE_EXCL| 490 CEPH_CAP_FILE_WR| 491 CEPH_CAP_FILE_BUFFER| 492 CEPH_CAP_AUTH_EXCL| 493 CEPH_CAP_XATTR_EXCL)) { 494 if (timespec_compare(ctime, &inode->i_ctime) > 0) { 495 dout("ctime %ld.%09ld -> %ld.%09ld inc w/ cap\n", 496 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec, 497 ctime->tv_sec, ctime->tv_nsec); 498 inode->i_ctime = *ctime; 499 } 500 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) { 501 /* the MDS did a utimes() */ 502 dout("mtime %ld.%09ld -> %ld.%09ld " 503 "tw %d -> %d\n", 504 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec, 505 mtime->tv_sec, mtime->tv_nsec, 506 ci->i_time_warp_seq, (int)time_warp_seq); 507 508 inode->i_mtime = *mtime; 509 inode->i_atime = *atime; 510 ci->i_time_warp_seq = time_warp_seq; 511 } else if (time_warp_seq == ci->i_time_warp_seq) { 512 /* nobody did utimes(); take the max */ 513 if (timespec_compare(mtime, &inode->i_mtime) > 0) { 514 dout("mtime %ld.%09ld -> %ld.%09ld inc\n", 515 inode->i_mtime.tv_sec, 516 inode->i_mtime.tv_nsec, 517 mtime->tv_sec, mtime->tv_nsec); 518 inode->i_mtime = *mtime; 519 } 520 if (timespec_compare(atime, &inode->i_atime) > 0) { 521 dout("atime %ld.%09ld -> %ld.%09ld inc\n", 522 inode->i_atime.tv_sec, 523 inode->i_atime.tv_nsec, 524 atime->tv_sec, atime->tv_nsec); 525 inode->i_atime = *atime; 526 } 527 } else if (issued & CEPH_CAP_FILE_EXCL) { 528 /* we did a utimes(); ignore mds values */ 529 } else { 530 warn = 1; 531 } 532 } else { 533 /* we have no write|excl caps; whatever the MDS says is true */ 534 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) { 535 inode->i_ctime = *ctime; 536 inode->i_mtime = *mtime; 537 inode->i_atime = *atime; 538 ci->i_time_warp_seq = time_warp_seq; 539 } else { 540 warn = 1; 541 } 542 } 543 if (warn) /* time_warp_seq shouldn't go backwards */ 544 dout("%p mds time_warp_seq %llu < %u\n", 545 inode, time_warp_seq, ci->i_time_warp_seq); 546 } 547 548 /* 549 * Populate an inode based on info from mds. May be called on new or 550 * existing inodes. 551 */ 552 static int fill_inode(struct inode *inode, 553 struct ceph_mds_reply_info_in *iinfo, 554 struct ceph_mds_reply_dirfrag *dirinfo, 555 struct ceph_mds_session *session, 556 unsigned long ttl_from, int cap_fmode, 557 struct ceph_cap_reservation *caps_reservation) 558 { 559 struct ceph_mds_reply_inode *info = iinfo->in; 560 struct ceph_inode_info *ci = ceph_inode(inode); 561 int i; 562 int issued, implemented; 563 struct timespec mtime, atime, ctime; 564 u32 nsplits; 565 struct ceph_buffer *xattr_blob = NULL; 566 int err = 0; 567 int queue_trunc = 0; 568 569 dout("fill_inode %p ino %llx.%llx v %llu had %llu\n", 570 inode, ceph_vinop(inode), le64_to_cpu(info->version), 571 ci->i_version); 572 573 /* 574 * prealloc xattr data, if it looks like we'll need it. only 575 * if len > 4 (meaning there are actually xattrs; the first 4 576 * bytes are the xattr count). 577 */ 578 if (iinfo->xattr_len > 4) { 579 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS); 580 if (!xattr_blob) 581 pr_err("fill_inode ENOMEM xattr blob %d bytes\n", 582 iinfo->xattr_len); 583 } 584 585 spin_lock(&inode->i_lock); 586 587 /* 588 * provided version will be odd if inode value is projected, 589 * even if stable. skip the update if we have newer stable 590 * info (ours>=theirs, e.g. due to racing mds replies), unless 591 * we are getting projected (unstable) info (in which case the 592 * version is odd, and we want ours>theirs). 593 * us them 594 * 2 2 skip 595 * 3 2 skip 596 * 3 3 update 597 */ 598 if (le64_to_cpu(info->version) > 0 && 599 (ci->i_version & ~1) >= le64_to_cpu(info->version)) 600 goto no_change; 601 602 issued = __ceph_caps_issued(ci, &implemented); 603 issued |= implemented | __ceph_caps_dirty(ci); 604 605 /* update inode */ 606 ci->i_version = le64_to_cpu(info->version); 607 inode->i_version++; 608 inode->i_rdev = le32_to_cpu(info->rdev); 609 610 if ((issued & CEPH_CAP_AUTH_EXCL) == 0) { 611 inode->i_mode = le32_to_cpu(info->mode); 612 inode->i_uid = le32_to_cpu(info->uid); 613 inode->i_gid = le32_to_cpu(info->gid); 614 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode, 615 inode->i_uid, inode->i_gid); 616 } 617 618 if ((issued & CEPH_CAP_LINK_EXCL) == 0) 619 inode->i_nlink = le32_to_cpu(info->nlink); 620 621 /* be careful with mtime, atime, size */ 622 ceph_decode_timespec(&atime, &info->atime); 623 ceph_decode_timespec(&mtime, &info->mtime); 624 ceph_decode_timespec(&ctime, &info->ctime); 625 queue_trunc = ceph_fill_file_size(inode, issued, 626 le32_to_cpu(info->truncate_seq), 627 le64_to_cpu(info->truncate_size), 628 le64_to_cpu(info->size)); 629 ceph_fill_file_time(inode, issued, 630 le32_to_cpu(info->time_warp_seq), 631 &ctime, &mtime, &atime); 632 633 /* only update max_size on auth cap */ 634 if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) && 635 ci->i_max_size != le64_to_cpu(info->max_size)) { 636 dout("max_size %lld -> %llu\n", ci->i_max_size, 637 le64_to_cpu(info->max_size)); 638 ci->i_max_size = le64_to_cpu(info->max_size); 639 } 640 641 ci->i_layout = info->layout; 642 inode->i_blkbits = fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1; 643 644 /* xattrs */ 645 /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */ 646 if ((issued & CEPH_CAP_XATTR_EXCL) == 0 && 647 le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) { 648 if (ci->i_xattrs.blob) 649 ceph_buffer_put(ci->i_xattrs.blob); 650 ci->i_xattrs.blob = xattr_blob; 651 if (xattr_blob) 652 memcpy(ci->i_xattrs.blob->vec.iov_base, 653 iinfo->xattr_data, iinfo->xattr_len); 654 ci->i_xattrs.version = le64_to_cpu(info->xattr_version); 655 xattr_blob = NULL; 656 } 657 658 inode->i_mapping->a_ops = &ceph_aops; 659 inode->i_mapping->backing_dev_info = 660 &ceph_sb_to_client(inode->i_sb)->backing_dev_info; 661 662 switch (inode->i_mode & S_IFMT) { 663 case S_IFIFO: 664 case S_IFBLK: 665 case S_IFCHR: 666 case S_IFSOCK: 667 init_special_inode(inode, inode->i_mode, inode->i_rdev); 668 inode->i_op = &ceph_file_iops; 669 break; 670 case S_IFREG: 671 inode->i_op = &ceph_file_iops; 672 inode->i_fop = &ceph_file_fops; 673 break; 674 case S_IFLNK: 675 inode->i_op = &ceph_symlink_iops; 676 if (!ci->i_symlink) { 677 int symlen = iinfo->symlink_len; 678 char *sym; 679 680 BUG_ON(symlen != inode->i_size); 681 spin_unlock(&inode->i_lock); 682 683 err = -ENOMEM; 684 sym = kmalloc(symlen+1, GFP_NOFS); 685 if (!sym) 686 goto out; 687 memcpy(sym, iinfo->symlink, symlen); 688 sym[symlen] = 0; 689 690 spin_lock(&inode->i_lock); 691 if (!ci->i_symlink) 692 ci->i_symlink = sym; 693 else 694 kfree(sym); /* lost a race */ 695 } 696 break; 697 case S_IFDIR: 698 inode->i_op = &ceph_dir_iops; 699 inode->i_fop = &ceph_dir_fops; 700 701 ci->i_dir_layout = iinfo->dir_layout; 702 703 ci->i_files = le64_to_cpu(info->files); 704 ci->i_subdirs = le64_to_cpu(info->subdirs); 705 ci->i_rbytes = le64_to_cpu(info->rbytes); 706 ci->i_rfiles = le64_to_cpu(info->rfiles); 707 ci->i_rsubdirs = le64_to_cpu(info->rsubdirs); 708 ceph_decode_timespec(&ci->i_rctime, &info->rctime); 709 710 /* set dir completion flag? */ 711 if (ci->i_files == 0 && ci->i_subdirs == 0 && 712 ceph_snap(inode) == CEPH_NOSNAP && 713 (le32_to_cpu(info->cap.caps) & CEPH_CAP_FILE_SHARED) && 714 (issued & CEPH_CAP_FILE_EXCL) == 0 && 715 (ci->i_ceph_flags & CEPH_I_COMPLETE) == 0) { 716 dout(" marking %p complete (empty)\n", inode); 717 /* ci->i_ceph_flags |= CEPH_I_COMPLETE; */ 718 ci->i_max_offset = 2; 719 } 720 break; 721 default: 722 pr_err("fill_inode %llx.%llx BAD mode 0%o\n", 723 ceph_vinop(inode), inode->i_mode); 724 } 725 726 no_change: 727 spin_unlock(&inode->i_lock); 728 729 /* queue truncate if we saw i_size decrease */ 730 if (queue_trunc) 731 ceph_queue_vmtruncate(inode); 732 733 /* populate frag tree */ 734 /* FIXME: move me up, if/when version reflects fragtree changes */ 735 nsplits = le32_to_cpu(info->fragtree.nsplits); 736 mutex_lock(&ci->i_fragtree_mutex); 737 for (i = 0; i < nsplits; i++) { 738 u32 id = le32_to_cpu(info->fragtree.splits[i].frag); 739 struct ceph_inode_frag *frag = __get_or_create_frag(ci, id); 740 741 if (IS_ERR(frag)) 742 continue; 743 frag->split_by = le32_to_cpu(info->fragtree.splits[i].by); 744 dout(" frag %x split by %d\n", frag->frag, frag->split_by); 745 } 746 mutex_unlock(&ci->i_fragtree_mutex); 747 748 /* were we issued a capability? */ 749 if (info->cap.caps) { 750 if (ceph_snap(inode) == CEPH_NOSNAP) { 751 ceph_add_cap(inode, session, 752 le64_to_cpu(info->cap.cap_id), 753 cap_fmode, 754 le32_to_cpu(info->cap.caps), 755 le32_to_cpu(info->cap.wanted), 756 le32_to_cpu(info->cap.seq), 757 le32_to_cpu(info->cap.mseq), 758 le64_to_cpu(info->cap.realm), 759 info->cap.flags, 760 caps_reservation); 761 } else { 762 spin_lock(&inode->i_lock); 763 dout(" %p got snap_caps %s\n", inode, 764 ceph_cap_string(le32_to_cpu(info->cap.caps))); 765 ci->i_snap_caps |= le32_to_cpu(info->cap.caps); 766 if (cap_fmode >= 0) 767 __ceph_get_fmode(ci, cap_fmode); 768 spin_unlock(&inode->i_lock); 769 } 770 } else if (cap_fmode >= 0) { 771 pr_warning("mds issued no caps on %llx.%llx\n", 772 ceph_vinop(inode)); 773 __ceph_get_fmode(ci, cap_fmode); 774 } 775 776 /* update delegation info? */ 777 if (dirinfo) 778 ceph_fill_dirfrag(inode, dirinfo); 779 780 err = 0; 781 782 out: 783 if (xattr_blob) 784 ceph_buffer_put(xattr_blob); 785 return err; 786 } 787 788 /* 789 * caller should hold session s_mutex. 790 */ 791 static void update_dentry_lease(struct dentry *dentry, 792 struct ceph_mds_reply_lease *lease, 793 struct ceph_mds_session *session, 794 unsigned long from_time) 795 { 796 struct ceph_dentry_info *di = ceph_dentry(dentry); 797 long unsigned duration = le32_to_cpu(lease->duration_ms); 798 long unsigned ttl = from_time + (duration * HZ) / 1000; 799 long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000; 800 struct inode *dir; 801 802 /* only track leases on regular dentries */ 803 if (dentry->d_op != &ceph_dentry_ops) 804 return; 805 806 spin_lock(&dentry->d_lock); 807 dout("update_dentry_lease %p mask %d duration %lu ms ttl %lu\n", 808 dentry, le16_to_cpu(lease->mask), duration, ttl); 809 810 /* make lease_rdcache_gen match directory */ 811 dir = dentry->d_parent->d_inode; 812 di->lease_shared_gen = ceph_inode(dir)->i_shared_gen; 813 814 if (lease->mask == 0) 815 goto out_unlock; 816 817 if (di->lease_gen == session->s_cap_gen && 818 time_before(ttl, dentry->d_time)) 819 goto out_unlock; /* we already have a newer lease. */ 820 821 if (di->lease_session && di->lease_session != session) 822 goto out_unlock; 823 824 ceph_dentry_lru_touch(dentry); 825 826 if (!di->lease_session) 827 di->lease_session = ceph_get_mds_session(session); 828 di->lease_gen = session->s_cap_gen; 829 di->lease_seq = le32_to_cpu(lease->seq); 830 di->lease_renew_after = half_ttl; 831 di->lease_renew_from = 0; 832 dentry->d_time = ttl; 833 out_unlock: 834 spin_unlock(&dentry->d_lock); 835 return; 836 } 837 838 /* 839 * Set dentry's directory position based on the current dir's max, and 840 * order it in d_subdirs, so that dcache_readdir behaves. 841 */ 842 static void ceph_set_dentry_offset(struct dentry *dn) 843 { 844 struct dentry *dir = dn->d_parent; 845 struct inode *inode = dn->d_parent->d_inode; 846 struct ceph_dentry_info *di; 847 848 BUG_ON(!inode); 849 850 di = ceph_dentry(dn); 851 852 spin_lock(&inode->i_lock); 853 if ((ceph_inode(inode)->i_ceph_flags & CEPH_I_COMPLETE) == 0) { 854 spin_unlock(&inode->i_lock); 855 return; 856 } 857 di->offset = ceph_inode(inode)->i_max_offset++; 858 spin_unlock(&inode->i_lock); 859 860 spin_lock(&dir->d_lock); 861 spin_lock_nested(&dn->d_lock, DENTRY_D_LOCK_NESTED); 862 list_move(&dn->d_u.d_child, &dir->d_subdirs); 863 dout("set_dentry_offset %p %lld (%p %p)\n", dn, di->offset, 864 dn->d_u.d_child.prev, dn->d_u.d_child.next); 865 spin_unlock(&dn->d_lock); 866 spin_unlock(&dir->d_lock); 867 } 868 869 /* 870 * splice a dentry to an inode. 871 * caller must hold directory i_mutex for this to be safe. 872 * 873 * we will only rehash the resulting dentry if @prehash is 874 * true; @prehash will be set to false (for the benefit of 875 * the caller) if we fail. 876 */ 877 static struct dentry *splice_dentry(struct dentry *dn, struct inode *in, 878 bool *prehash, bool set_offset) 879 { 880 struct dentry *realdn; 881 882 BUG_ON(dn->d_inode); 883 884 /* dn must be unhashed */ 885 if (!d_unhashed(dn)) 886 d_drop(dn); 887 realdn = d_materialise_unique(dn, in); 888 if (IS_ERR(realdn)) { 889 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n", 890 PTR_ERR(realdn), dn, in, ceph_vinop(in)); 891 if (prehash) 892 *prehash = false; /* don't rehash on error */ 893 dn = realdn; /* note realdn contains the error */ 894 goto out; 895 } else if (realdn) { 896 dout("dn %p (%d) spliced with %p (%d) " 897 "inode %p ino %llx.%llx\n", 898 dn, dn->d_count, 899 realdn, realdn->d_count, 900 realdn->d_inode, ceph_vinop(realdn->d_inode)); 901 dput(dn); 902 dn = realdn; 903 } else { 904 BUG_ON(!ceph_dentry(dn)); 905 dout("dn %p attached to %p ino %llx.%llx\n", 906 dn, dn->d_inode, ceph_vinop(dn->d_inode)); 907 } 908 if ((!prehash || *prehash) && d_unhashed(dn)) 909 d_rehash(dn); 910 if (set_offset) 911 ceph_set_dentry_offset(dn); 912 out: 913 return dn; 914 } 915 916 /* 917 * Incorporate results into the local cache. This is either just 918 * one inode, or a directory, dentry, and possibly linked-to inode (e.g., 919 * after a lookup). 920 * 921 * A reply may contain 922 * a directory inode along with a dentry. 923 * and/or a target inode 924 * 925 * Called with snap_rwsem (read). 926 */ 927 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req, 928 struct ceph_mds_session *session) 929 { 930 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; 931 struct inode *in = NULL; 932 struct ceph_mds_reply_inode *ininfo; 933 struct ceph_vino vino; 934 struct ceph_fs_client *fsc = ceph_sb_to_client(sb); 935 int i = 0; 936 int err = 0; 937 938 dout("fill_trace %p is_dentry %d is_target %d\n", req, 939 rinfo->head->is_dentry, rinfo->head->is_target); 940 941 #if 0 942 /* 943 * Debugging hook: 944 * 945 * If we resend completed ops to a recovering mds, we get no 946 * trace. Since that is very rare, pretend this is the case 947 * to ensure the 'no trace' handlers in the callers behave. 948 * 949 * Fill in inodes unconditionally to avoid breaking cap 950 * invariants. 951 */ 952 if (rinfo->head->op & CEPH_MDS_OP_WRITE) { 953 pr_info("fill_trace faking empty trace on %lld %s\n", 954 req->r_tid, ceph_mds_op_name(rinfo->head->op)); 955 if (rinfo->head->is_dentry) { 956 rinfo->head->is_dentry = 0; 957 err = fill_inode(req->r_locked_dir, 958 &rinfo->diri, rinfo->dirfrag, 959 session, req->r_request_started, -1); 960 } 961 if (rinfo->head->is_target) { 962 rinfo->head->is_target = 0; 963 ininfo = rinfo->targeti.in; 964 vino.ino = le64_to_cpu(ininfo->ino); 965 vino.snap = le64_to_cpu(ininfo->snapid); 966 in = ceph_get_inode(sb, vino); 967 err = fill_inode(in, &rinfo->targeti, NULL, 968 session, req->r_request_started, 969 req->r_fmode); 970 iput(in); 971 } 972 } 973 #endif 974 975 if (!rinfo->head->is_target && !rinfo->head->is_dentry) { 976 dout("fill_trace reply is empty!\n"); 977 if (rinfo->head->result == 0 && req->r_locked_dir) 978 ceph_invalidate_dir_request(req); 979 return 0; 980 } 981 982 if (rinfo->head->is_dentry) { 983 struct inode *dir = req->r_locked_dir; 984 985 err = fill_inode(dir, &rinfo->diri, rinfo->dirfrag, 986 session, req->r_request_started, -1, 987 &req->r_caps_reservation); 988 if (err < 0) 989 return err; 990 } 991 992 /* 993 * ignore null lease/binding on snapdir ENOENT, or else we 994 * will have trouble splicing in the virtual snapdir later 995 */ 996 if (rinfo->head->is_dentry && !req->r_aborted && 997 (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name, 998 fsc->mount_options->snapdir_name, 999 req->r_dentry->d_name.len))) { 1000 /* 1001 * lookup link rename : null -> possibly existing inode 1002 * mknod symlink mkdir : null -> new inode 1003 * unlink : linked -> null 1004 */ 1005 struct inode *dir = req->r_locked_dir; 1006 struct dentry *dn = req->r_dentry; 1007 bool have_dir_cap, have_lease; 1008 1009 BUG_ON(!dn); 1010 BUG_ON(!dir); 1011 BUG_ON(dn->d_parent->d_inode != dir); 1012 BUG_ON(ceph_ino(dir) != 1013 le64_to_cpu(rinfo->diri.in->ino)); 1014 BUG_ON(ceph_snap(dir) != 1015 le64_to_cpu(rinfo->diri.in->snapid)); 1016 1017 /* do we have a lease on the whole dir? */ 1018 have_dir_cap = 1019 (le32_to_cpu(rinfo->diri.in->cap.caps) & 1020 CEPH_CAP_FILE_SHARED); 1021 1022 /* do we have a dn lease? */ 1023 have_lease = have_dir_cap || 1024 (le16_to_cpu(rinfo->dlease->mask) & 1025 CEPH_LOCK_DN); 1026 1027 if (!have_lease) 1028 dout("fill_trace no dentry lease or dir cap\n"); 1029 1030 /* rename? */ 1031 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) { 1032 dout(" src %p '%.*s' dst %p '%.*s'\n", 1033 req->r_old_dentry, 1034 req->r_old_dentry->d_name.len, 1035 req->r_old_dentry->d_name.name, 1036 dn, dn->d_name.len, dn->d_name.name); 1037 dout("fill_trace doing d_move %p -> %p\n", 1038 req->r_old_dentry, dn); 1039 1040 d_move(req->r_old_dentry, dn); 1041 dout(" src %p '%.*s' dst %p '%.*s'\n", 1042 req->r_old_dentry, 1043 req->r_old_dentry->d_name.len, 1044 req->r_old_dentry->d_name.name, 1045 dn, dn->d_name.len, dn->d_name.name); 1046 1047 /* ensure target dentry is invalidated, despite 1048 rehashing bug in vfs_rename_dir */ 1049 ceph_invalidate_dentry_lease(dn); 1050 1051 /* 1052 * d_move() puts the renamed dentry at the end of 1053 * d_subdirs. We need to assign it an appropriate 1054 * directory offset so we can behave when holding 1055 * I_COMPLETE. 1056 */ 1057 ceph_set_dentry_offset(req->r_old_dentry); 1058 dout("dn %p gets new offset %lld\n", req->r_old_dentry, 1059 ceph_dentry(req->r_old_dentry)->offset); 1060 1061 dn = req->r_old_dentry; /* use old_dentry */ 1062 in = dn->d_inode; 1063 } 1064 1065 /* null dentry? */ 1066 if (!rinfo->head->is_target) { 1067 dout("fill_trace null dentry\n"); 1068 if (dn->d_inode) { 1069 dout("d_delete %p\n", dn); 1070 d_delete(dn); 1071 } else { 1072 dout("d_instantiate %p NULL\n", dn); 1073 d_instantiate(dn, NULL); 1074 if (have_lease && d_unhashed(dn)) 1075 d_rehash(dn); 1076 update_dentry_lease(dn, rinfo->dlease, 1077 session, 1078 req->r_request_started); 1079 } 1080 goto done; 1081 } 1082 1083 /* attach proper inode */ 1084 ininfo = rinfo->targeti.in; 1085 vino.ino = le64_to_cpu(ininfo->ino); 1086 vino.snap = le64_to_cpu(ininfo->snapid); 1087 in = dn->d_inode; 1088 if (!in) { 1089 in = ceph_get_inode(sb, vino); 1090 if (IS_ERR(in)) { 1091 pr_err("fill_trace bad get_inode " 1092 "%llx.%llx\n", vino.ino, vino.snap); 1093 err = PTR_ERR(in); 1094 d_delete(dn); 1095 goto done; 1096 } 1097 dn = splice_dentry(dn, in, &have_lease, true); 1098 if (IS_ERR(dn)) { 1099 err = PTR_ERR(dn); 1100 goto done; 1101 } 1102 req->r_dentry = dn; /* may have spliced */ 1103 igrab(in); 1104 } else if (ceph_ino(in) == vino.ino && 1105 ceph_snap(in) == vino.snap) { 1106 igrab(in); 1107 } else { 1108 dout(" %p links to %p %llx.%llx, not %llx.%llx\n", 1109 dn, in, ceph_ino(in), ceph_snap(in), 1110 vino.ino, vino.snap); 1111 have_lease = false; 1112 in = NULL; 1113 } 1114 1115 if (have_lease) 1116 update_dentry_lease(dn, rinfo->dlease, session, 1117 req->r_request_started); 1118 dout(" final dn %p\n", dn); 1119 i++; 1120 } else if (req->r_op == CEPH_MDS_OP_LOOKUPSNAP || 1121 req->r_op == CEPH_MDS_OP_MKSNAP) { 1122 struct dentry *dn = req->r_dentry; 1123 1124 /* fill out a snapdir LOOKUPSNAP dentry */ 1125 BUG_ON(!dn); 1126 BUG_ON(!req->r_locked_dir); 1127 BUG_ON(ceph_snap(req->r_locked_dir) != CEPH_SNAPDIR); 1128 ininfo = rinfo->targeti.in; 1129 vino.ino = le64_to_cpu(ininfo->ino); 1130 vino.snap = le64_to_cpu(ininfo->snapid); 1131 in = ceph_get_inode(sb, vino); 1132 if (IS_ERR(in)) { 1133 pr_err("fill_inode get_inode badness %llx.%llx\n", 1134 vino.ino, vino.snap); 1135 err = PTR_ERR(in); 1136 d_delete(dn); 1137 goto done; 1138 } 1139 dout(" linking snapped dir %p to dn %p\n", in, dn); 1140 dn = splice_dentry(dn, in, NULL, true); 1141 if (IS_ERR(dn)) { 1142 err = PTR_ERR(dn); 1143 goto done; 1144 } 1145 req->r_dentry = dn; /* may have spliced */ 1146 igrab(in); 1147 rinfo->head->is_dentry = 1; /* fool notrace handlers */ 1148 } 1149 1150 if (rinfo->head->is_target) { 1151 vino.ino = le64_to_cpu(rinfo->targeti.in->ino); 1152 vino.snap = le64_to_cpu(rinfo->targeti.in->snapid); 1153 1154 if (in == NULL || ceph_ino(in) != vino.ino || 1155 ceph_snap(in) != vino.snap) { 1156 in = ceph_get_inode(sb, vino); 1157 if (IS_ERR(in)) { 1158 err = PTR_ERR(in); 1159 goto done; 1160 } 1161 } 1162 req->r_target_inode = in; 1163 1164 err = fill_inode(in, 1165 &rinfo->targeti, NULL, 1166 session, req->r_request_started, 1167 (le32_to_cpu(rinfo->head->result) == 0) ? 1168 req->r_fmode : -1, 1169 &req->r_caps_reservation); 1170 if (err < 0) { 1171 pr_err("fill_inode badness %p %llx.%llx\n", 1172 in, ceph_vinop(in)); 1173 goto done; 1174 } 1175 } 1176 1177 done: 1178 dout("fill_trace done err=%d\n", err); 1179 return err; 1180 } 1181 1182 /* 1183 * Prepopulate our cache with readdir results, leases, etc. 1184 */ 1185 int ceph_readdir_prepopulate(struct ceph_mds_request *req, 1186 struct ceph_mds_session *session) 1187 { 1188 struct dentry *parent = req->r_dentry; 1189 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; 1190 struct qstr dname; 1191 struct dentry *dn; 1192 struct inode *in; 1193 int err = 0, i; 1194 struct inode *snapdir = NULL; 1195 struct ceph_mds_request_head *rhead = req->r_request->front.iov_base; 1196 u64 frag = le32_to_cpu(rhead->args.readdir.frag); 1197 struct ceph_dentry_info *di; 1198 1199 if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) { 1200 snapdir = ceph_get_snapdir(parent->d_inode); 1201 parent = d_find_alias(snapdir); 1202 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n", 1203 rinfo->dir_nr, parent); 1204 } else { 1205 dout("readdir_prepopulate %d items under dn %p\n", 1206 rinfo->dir_nr, parent); 1207 if (rinfo->dir_dir) 1208 ceph_fill_dirfrag(parent->d_inode, rinfo->dir_dir); 1209 } 1210 1211 for (i = 0; i < rinfo->dir_nr; i++) { 1212 struct ceph_vino vino; 1213 1214 dname.name = rinfo->dir_dname[i]; 1215 dname.len = rinfo->dir_dname_len[i]; 1216 dname.hash = full_name_hash(dname.name, dname.len); 1217 1218 vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino); 1219 vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid); 1220 1221 retry_lookup: 1222 dn = d_lookup(parent, &dname); 1223 dout("d_lookup on parent=%p name=%.*s got %p\n", 1224 parent, dname.len, dname.name, dn); 1225 1226 if (!dn) { 1227 dn = d_alloc(parent, &dname); 1228 dout("d_alloc %p '%.*s' = %p\n", parent, 1229 dname.len, dname.name, dn); 1230 if (dn == NULL) { 1231 dout("d_alloc badness\n"); 1232 err = -ENOMEM; 1233 goto out; 1234 } 1235 err = ceph_init_dentry(dn); 1236 if (err < 0) { 1237 dput(dn); 1238 goto out; 1239 } 1240 } else if (dn->d_inode && 1241 (ceph_ino(dn->d_inode) != vino.ino || 1242 ceph_snap(dn->d_inode) != vino.snap)) { 1243 dout(" dn %p points to wrong inode %p\n", 1244 dn, dn->d_inode); 1245 d_delete(dn); 1246 dput(dn); 1247 goto retry_lookup; 1248 } else { 1249 /* reorder parent's d_subdirs */ 1250 spin_lock(&parent->d_lock); 1251 spin_lock_nested(&dn->d_lock, DENTRY_D_LOCK_NESTED); 1252 list_move(&dn->d_u.d_child, &parent->d_subdirs); 1253 spin_unlock(&dn->d_lock); 1254 spin_unlock(&parent->d_lock); 1255 } 1256 1257 di = dn->d_fsdata; 1258 di->offset = ceph_make_fpos(frag, i + req->r_readdir_offset); 1259 1260 /* inode */ 1261 if (dn->d_inode) { 1262 in = dn->d_inode; 1263 } else { 1264 in = ceph_get_inode(parent->d_sb, vino); 1265 if (IS_ERR(in)) { 1266 dout("new_inode badness\n"); 1267 d_delete(dn); 1268 dput(dn); 1269 err = PTR_ERR(in); 1270 goto out; 1271 } 1272 dn = splice_dentry(dn, in, NULL, false); 1273 if (IS_ERR(dn)) 1274 dn = NULL; 1275 } 1276 1277 if (fill_inode(in, &rinfo->dir_in[i], NULL, session, 1278 req->r_request_started, -1, 1279 &req->r_caps_reservation) < 0) { 1280 pr_err("fill_inode badness on %p\n", in); 1281 goto next_item; 1282 } 1283 if (dn) 1284 update_dentry_lease(dn, rinfo->dir_dlease[i], 1285 req->r_session, 1286 req->r_request_started); 1287 next_item: 1288 if (dn) 1289 dput(dn); 1290 } 1291 req->r_did_prepopulate = true; 1292 1293 out: 1294 if (snapdir) { 1295 iput(snapdir); 1296 dput(parent); 1297 } 1298 dout("readdir_prepopulate done\n"); 1299 return err; 1300 } 1301 1302 int ceph_inode_set_size(struct inode *inode, loff_t size) 1303 { 1304 struct ceph_inode_info *ci = ceph_inode(inode); 1305 int ret = 0; 1306 1307 spin_lock(&inode->i_lock); 1308 dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size); 1309 inode->i_size = size; 1310 inode->i_blocks = (size + (1 << 9) - 1) >> 9; 1311 1312 /* tell the MDS if we are approaching max_size */ 1313 if ((size << 1) >= ci->i_max_size && 1314 (ci->i_reported_size << 1) < ci->i_max_size) 1315 ret = 1; 1316 1317 spin_unlock(&inode->i_lock); 1318 return ret; 1319 } 1320 1321 /* 1322 * Write back inode data in a worker thread. (This can't be done 1323 * in the message handler context.) 1324 */ 1325 void ceph_queue_writeback(struct inode *inode) 1326 { 1327 if (queue_work(ceph_inode_to_client(inode)->wb_wq, 1328 &ceph_inode(inode)->i_wb_work)) { 1329 dout("ceph_queue_writeback %p\n", inode); 1330 igrab(inode); 1331 } else { 1332 dout("ceph_queue_writeback %p failed\n", inode); 1333 } 1334 } 1335 1336 static void ceph_writeback_work(struct work_struct *work) 1337 { 1338 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info, 1339 i_wb_work); 1340 struct inode *inode = &ci->vfs_inode; 1341 1342 dout("writeback %p\n", inode); 1343 filemap_fdatawrite(&inode->i_data); 1344 iput(inode); 1345 } 1346 1347 /* 1348 * queue an async invalidation 1349 */ 1350 void ceph_queue_invalidate(struct inode *inode) 1351 { 1352 if (queue_work(ceph_inode_to_client(inode)->pg_inv_wq, 1353 &ceph_inode(inode)->i_pg_inv_work)) { 1354 dout("ceph_queue_invalidate %p\n", inode); 1355 igrab(inode); 1356 } else { 1357 dout("ceph_queue_invalidate %p failed\n", inode); 1358 } 1359 } 1360 1361 /* 1362 * invalidate any pages that are not dirty or under writeback. this 1363 * includes pages that are clean and mapped. 1364 */ 1365 static void ceph_invalidate_nondirty_pages(struct address_space *mapping) 1366 { 1367 struct pagevec pvec; 1368 pgoff_t next = 0; 1369 int i; 1370 1371 pagevec_init(&pvec, 0); 1372 while (pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) { 1373 for (i = 0; i < pagevec_count(&pvec); i++) { 1374 struct page *page = pvec.pages[i]; 1375 pgoff_t index; 1376 int skip_page = 1377 (PageDirty(page) || PageWriteback(page)); 1378 1379 if (!skip_page) 1380 skip_page = !trylock_page(page); 1381 1382 /* 1383 * We really shouldn't be looking at the ->index of an 1384 * unlocked page. But we're not allowed to lock these 1385 * pages. So we rely upon nobody altering the ->index 1386 * of this (pinned-by-us) page. 1387 */ 1388 index = page->index; 1389 if (index > next) 1390 next = index; 1391 next++; 1392 1393 if (skip_page) 1394 continue; 1395 1396 generic_error_remove_page(mapping, page); 1397 unlock_page(page); 1398 } 1399 pagevec_release(&pvec); 1400 cond_resched(); 1401 } 1402 } 1403 1404 /* 1405 * Invalidate inode pages in a worker thread. (This can't be done 1406 * in the message handler context.) 1407 */ 1408 static void ceph_invalidate_work(struct work_struct *work) 1409 { 1410 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info, 1411 i_pg_inv_work); 1412 struct inode *inode = &ci->vfs_inode; 1413 u32 orig_gen; 1414 int check = 0; 1415 1416 spin_lock(&inode->i_lock); 1417 dout("invalidate_pages %p gen %d revoking %d\n", inode, 1418 ci->i_rdcache_gen, ci->i_rdcache_revoking); 1419 if (ci->i_rdcache_revoking != ci->i_rdcache_gen) { 1420 /* nevermind! */ 1421 spin_unlock(&inode->i_lock); 1422 goto out; 1423 } 1424 orig_gen = ci->i_rdcache_gen; 1425 spin_unlock(&inode->i_lock); 1426 1427 ceph_invalidate_nondirty_pages(inode->i_mapping); 1428 1429 spin_lock(&inode->i_lock); 1430 if (orig_gen == ci->i_rdcache_gen && 1431 orig_gen == ci->i_rdcache_revoking) { 1432 dout("invalidate_pages %p gen %d successful\n", inode, 1433 ci->i_rdcache_gen); 1434 ci->i_rdcache_revoking--; 1435 check = 1; 1436 } else { 1437 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n", 1438 inode, orig_gen, ci->i_rdcache_gen, 1439 ci->i_rdcache_revoking); 1440 } 1441 spin_unlock(&inode->i_lock); 1442 1443 if (check) 1444 ceph_check_caps(ci, 0, NULL); 1445 out: 1446 iput(inode); 1447 } 1448 1449 1450 /* 1451 * called by trunc_wq; take i_mutex ourselves 1452 * 1453 * We also truncate in a separate thread as well. 1454 */ 1455 static void ceph_vmtruncate_work(struct work_struct *work) 1456 { 1457 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info, 1458 i_vmtruncate_work); 1459 struct inode *inode = &ci->vfs_inode; 1460 1461 dout("vmtruncate_work %p\n", inode); 1462 mutex_lock(&inode->i_mutex); 1463 __ceph_do_pending_vmtruncate(inode); 1464 mutex_unlock(&inode->i_mutex); 1465 iput(inode); 1466 } 1467 1468 /* 1469 * Queue an async vmtruncate. If we fail to queue work, we will handle 1470 * the truncation the next time we call __ceph_do_pending_vmtruncate. 1471 */ 1472 void ceph_queue_vmtruncate(struct inode *inode) 1473 { 1474 struct ceph_inode_info *ci = ceph_inode(inode); 1475 1476 if (queue_work(ceph_sb_to_client(inode->i_sb)->trunc_wq, 1477 &ci->i_vmtruncate_work)) { 1478 dout("ceph_queue_vmtruncate %p\n", inode); 1479 igrab(inode); 1480 } else { 1481 dout("ceph_queue_vmtruncate %p failed, pending=%d\n", 1482 inode, ci->i_truncate_pending); 1483 } 1484 } 1485 1486 /* 1487 * called with i_mutex held. 1488 * 1489 * Make sure any pending truncation is applied before doing anything 1490 * that may depend on it. 1491 */ 1492 void __ceph_do_pending_vmtruncate(struct inode *inode) 1493 { 1494 struct ceph_inode_info *ci = ceph_inode(inode); 1495 u64 to; 1496 int wrbuffer_refs, wake = 0; 1497 1498 retry: 1499 spin_lock(&inode->i_lock); 1500 if (ci->i_truncate_pending == 0) { 1501 dout("__do_pending_vmtruncate %p none pending\n", inode); 1502 spin_unlock(&inode->i_lock); 1503 return; 1504 } 1505 1506 /* 1507 * make sure any dirty snapped pages are flushed before we 1508 * possibly truncate them.. so write AND block! 1509 */ 1510 if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) { 1511 dout("__do_pending_vmtruncate %p flushing snaps first\n", 1512 inode); 1513 spin_unlock(&inode->i_lock); 1514 filemap_write_and_wait_range(&inode->i_data, 0, 1515 inode->i_sb->s_maxbytes); 1516 goto retry; 1517 } 1518 1519 to = ci->i_truncate_size; 1520 wrbuffer_refs = ci->i_wrbuffer_ref; 1521 dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode, 1522 ci->i_truncate_pending, to); 1523 spin_unlock(&inode->i_lock); 1524 1525 truncate_inode_pages(inode->i_mapping, to); 1526 1527 spin_lock(&inode->i_lock); 1528 ci->i_truncate_pending--; 1529 if (ci->i_truncate_pending == 0) 1530 wake = 1; 1531 spin_unlock(&inode->i_lock); 1532 1533 if (wrbuffer_refs == 0) 1534 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL); 1535 if (wake) 1536 wake_up_all(&ci->i_cap_wq); 1537 } 1538 1539 1540 /* 1541 * symlinks 1542 */ 1543 static void *ceph_sym_follow_link(struct dentry *dentry, struct nameidata *nd) 1544 { 1545 struct ceph_inode_info *ci = ceph_inode(dentry->d_inode); 1546 nd_set_link(nd, ci->i_symlink); 1547 return NULL; 1548 } 1549 1550 static const struct inode_operations ceph_symlink_iops = { 1551 .readlink = generic_readlink, 1552 .follow_link = ceph_sym_follow_link, 1553 }; 1554 1555 /* 1556 * setattr 1557 */ 1558 int ceph_setattr(struct dentry *dentry, struct iattr *attr) 1559 { 1560 struct inode *inode = dentry->d_inode; 1561 struct ceph_inode_info *ci = ceph_inode(inode); 1562 struct inode *parent_inode = dentry->d_parent->d_inode; 1563 const unsigned int ia_valid = attr->ia_valid; 1564 struct ceph_mds_request *req; 1565 struct ceph_mds_client *mdsc = ceph_sb_to_client(dentry->d_sb)->mdsc; 1566 int issued; 1567 int release = 0, dirtied = 0; 1568 int mask = 0; 1569 int err = 0; 1570 int inode_dirty_flags = 0; 1571 1572 if (ceph_snap(inode) != CEPH_NOSNAP) 1573 return -EROFS; 1574 1575 __ceph_do_pending_vmtruncate(inode); 1576 1577 err = inode_change_ok(inode, attr); 1578 if (err != 0) 1579 return err; 1580 1581 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR, 1582 USE_AUTH_MDS); 1583 if (IS_ERR(req)) 1584 return PTR_ERR(req); 1585 1586 spin_lock(&inode->i_lock); 1587 issued = __ceph_caps_issued(ci, NULL); 1588 dout("setattr %p issued %s\n", inode, ceph_cap_string(issued)); 1589 1590 if (ia_valid & ATTR_UID) { 1591 dout("setattr %p uid %d -> %d\n", inode, 1592 inode->i_uid, attr->ia_uid); 1593 if (issued & CEPH_CAP_AUTH_EXCL) { 1594 inode->i_uid = attr->ia_uid; 1595 dirtied |= CEPH_CAP_AUTH_EXCL; 1596 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || 1597 attr->ia_uid != inode->i_uid) { 1598 req->r_args.setattr.uid = cpu_to_le32(attr->ia_uid); 1599 mask |= CEPH_SETATTR_UID; 1600 release |= CEPH_CAP_AUTH_SHARED; 1601 } 1602 } 1603 if (ia_valid & ATTR_GID) { 1604 dout("setattr %p gid %d -> %d\n", inode, 1605 inode->i_gid, attr->ia_gid); 1606 if (issued & CEPH_CAP_AUTH_EXCL) { 1607 inode->i_gid = attr->ia_gid; 1608 dirtied |= CEPH_CAP_AUTH_EXCL; 1609 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || 1610 attr->ia_gid != inode->i_gid) { 1611 req->r_args.setattr.gid = cpu_to_le32(attr->ia_gid); 1612 mask |= CEPH_SETATTR_GID; 1613 release |= CEPH_CAP_AUTH_SHARED; 1614 } 1615 } 1616 if (ia_valid & ATTR_MODE) { 1617 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode, 1618 attr->ia_mode); 1619 if (issued & CEPH_CAP_AUTH_EXCL) { 1620 inode->i_mode = attr->ia_mode; 1621 dirtied |= CEPH_CAP_AUTH_EXCL; 1622 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || 1623 attr->ia_mode != inode->i_mode) { 1624 req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode); 1625 mask |= CEPH_SETATTR_MODE; 1626 release |= CEPH_CAP_AUTH_SHARED; 1627 } 1628 } 1629 1630 if (ia_valid & ATTR_ATIME) { 1631 dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode, 1632 inode->i_atime.tv_sec, inode->i_atime.tv_nsec, 1633 attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec); 1634 if (issued & CEPH_CAP_FILE_EXCL) { 1635 ci->i_time_warp_seq++; 1636 inode->i_atime = attr->ia_atime; 1637 dirtied |= CEPH_CAP_FILE_EXCL; 1638 } else if ((issued & CEPH_CAP_FILE_WR) && 1639 timespec_compare(&inode->i_atime, 1640 &attr->ia_atime) < 0) { 1641 inode->i_atime = attr->ia_atime; 1642 dirtied |= CEPH_CAP_FILE_WR; 1643 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || 1644 !timespec_equal(&inode->i_atime, &attr->ia_atime)) { 1645 ceph_encode_timespec(&req->r_args.setattr.atime, 1646 &attr->ia_atime); 1647 mask |= CEPH_SETATTR_ATIME; 1648 release |= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_RD | 1649 CEPH_CAP_FILE_WR; 1650 } 1651 } 1652 if (ia_valid & ATTR_MTIME) { 1653 dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode, 1654 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec, 1655 attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec); 1656 if (issued & CEPH_CAP_FILE_EXCL) { 1657 ci->i_time_warp_seq++; 1658 inode->i_mtime = attr->ia_mtime; 1659 dirtied |= CEPH_CAP_FILE_EXCL; 1660 } else if ((issued & CEPH_CAP_FILE_WR) && 1661 timespec_compare(&inode->i_mtime, 1662 &attr->ia_mtime) < 0) { 1663 inode->i_mtime = attr->ia_mtime; 1664 dirtied |= CEPH_CAP_FILE_WR; 1665 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || 1666 !timespec_equal(&inode->i_mtime, &attr->ia_mtime)) { 1667 ceph_encode_timespec(&req->r_args.setattr.mtime, 1668 &attr->ia_mtime); 1669 mask |= CEPH_SETATTR_MTIME; 1670 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD | 1671 CEPH_CAP_FILE_WR; 1672 } 1673 } 1674 if (ia_valid & ATTR_SIZE) { 1675 dout("setattr %p size %lld -> %lld\n", inode, 1676 inode->i_size, attr->ia_size); 1677 if (attr->ia_size > inode->i_sb->s_maxbytes) { 1678 err = -EINVAL; 1679 goto out; 1680 } 1681 if ((issued & CEPH_CAP_FILE_EXCL) && 1682 attr->ia_size > inode->i_size) { 1683 inode->i_size = attr->ia_size; 1684 inode->i_blocks = 1685 (attr->ia_size + (1 << 9) - 1) >> 9; 1686 inode->i_ctime = attr->ia_ctime; 1687 ci->i_reported_size = attr->ia_size; 1688 dirtied |= CEPH_CAP_FILE_EXCL; 1689 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || 1690 attr->ia_size != inode->i_size) { 1691 req->r_args.setattr.size = cpu_to_le64(attr->ia_size); 1692 req->r_args.setattr.old_size = 1693 cpu_to_le64(inode->i_size); 1694 mask |= CEPH_SETATTR_SIZE; 1695 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD | 1696 CEPH_CAP_FILE_WR; 1697 } 1698 } 1699 1700 /* these do nothing */ 1701 if (ia_valid & ATTR_CTIME) { 1702 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME| 1703 ATTR_MODE|ATTR_UID|ATTR_GID)) == 0; 1704 dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode, 1705 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec, 1706 attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec, 1707 only ? "ctime only" : "ignored"); 1708 inode->i_ctime = attr->ia_ctime; 1709 if (only) { 1710 /* 1711 * if kernel wants to dirty ctime but nothing else, 1712 * we need to choose a cap to dirty under, or do 1713 * a almost-no-op setattr 1714 */ 1715 if (issued & CEPH_CAP_AUTH_EXCL) 1716 dirtied |= CEPH_CAP_AUTH_EXCL; 1717 else if (issued & CEPH_CAP_FILE_EXCL) 1718 dirtied |= CEPH_CAP_FILE_EXCL; 1719 else if (issued & CEPH_CAP_XATTR_EXCL) 1720 dirtied |= CEPH_CAP_XATTR_EXCL; 1721 else 1722 mask |= CEPH_SETATTR_CTIME; 1723 } 1724 } 1725 if (ia_valid & ATTR_FILE) 1726 dout("setattr %p ATTR_FILE ... hrm!\n", inode); 1727 1728 if (dirtied) { 1729 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied); 1730 inode->i_ctime = CURRENT_TIME; 1731 } 1732 1733 release &= issued; 1734 spin_unlock(&inode->i_lock); 1735 1736 if (inode_dirty_flags) 1737 __mark_inode_dirty(inode, inode_dirty_flags); 1738 1739 if (mask) { 1740 req->r_inode = igrab(inode); 1741 req->r_inode_drop = release; 1742 req->r_args.setattr.mask = cpu_to_le32(mask); 1743 req->r_num_caps = 1; 1744 err = ceph_mdsc_do_request(mdsc, parent_inode, req); 1745 } 1746 dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err, 1747 ceph_cap_string(dirtied), mask); 1748 1749 ceph_mdsc_put_request(req); 1750 __ceph_do_pending_vmtruncate(inode); 1751 return err; 1752 out: 1753 spin_unlock(&inode->i_lock); 1754 ceph_mdsc_put_request(req); 1755 return err; 1756 } 1757 1758 /* 1759 * Verify that we have a lease on the given mask. If not, 1760 * do a getattr against an mds. 1761 */ 1762 int ceph_do_getattr(struct inode *inode, int mask) 1763 { 1764 struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb); 1765 struct ceph_mds_client *mdsc = fsc->mdsc; 1766 struct ceph_mds_request *req; 1767 int err; 1768 1769 if (ceph_snap(inode) == CEPH_SNAPDIR) { 1770 dout("do_getattr inode %p SNAPDIR\n", inode); 1771 return 0; 1772 } 1773 1774 dout("do_getattr inode %p mask %s mode 0%o\n", inode, ceph_cap_string(mask), inode->i_mode); 1775 if (ceph_caps_issued_mask(ceph_inode(inode), mask, 1)) 1776 return 0; 1777 1778 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS); 1779 if (IS_ERR(req)) 1780 return PTR_ERR(req); 1781 req->r_inode = igrab(inode); 1782 req->r_num_caps = 1; 1783 req->r_args.getattr.mask = cpu_to_le32(mask); 1784 err = ceph_mdsc_do_request(mdsc, NULL, req); 1785 ceph_mdsc_put_request(req); 1786 dout("do_getattr result=%d\n", err); 1787 return err; 1788 } 1789 1790 1791 /* 1792 * Check inode permissions. We verify we have a valid value for 1793 * the AUTH cap, then call the generic handler. 1794 */ 1795 int ceph_permission(struct inode *inode, int mask, unsigned int flags) 1796 { 1797 int err; 1798 1799 if (flags & IPERM_FLAG_RCU) 1800 return -ECHILD; 1801 1802 err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED); 1803 1804 if (!err) 1805 err = generic_permission(inode, mask, flags, NULL); 1806 return err; 1807 } 1808 1809 /* 1810 * Get all attributes. Hopefully somedata we'll have a statlite() 1811 * and can limit the fields we require to be accurate. 1812 */ 1813 int ceph_getattr(struct vfsmount *mnt, struct dentry *dentry, 1814 struct kstat *stat) 1815 { 1816 struct inode *inode = dentry->d_inode; 1817 struct ceph_inode_info *ci = ceph_inode(inode); 1818 int err; 1819 1820 err = ceph_do_getattr(inode, CEPH_STAT_CAP_INODE_ALL); 1821 if (!err) { 1822 generic_fillattr(inode, stat); 1823 stat->ino = ceph_translate_ino(inode->i_sb, inode->i_ino); 1824 if (ceph_snap(inode) != CEPH_NOSNAP) 1825 stat->dev = ceph_snap(inode); 1826 else 1827 stat->dev = 0; 1828 if (S_ISDIR(inode->i_mode)) { 1829 if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), 1830 RBYTES)) 1831 stat->size = ci->i_rbytes; 1832 else 1833 stat->size = ci->i_files + ci->i_subdirs; 1834 stat->blocks = 0; 1835 stat->blksize = 65536; 1836 } 1837 } 1838 return err; 1839 } 1840