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_wb_ref = 0; 359 ci->i_wrbuffer_ref = 0; 360 ci->i_wrbuffer_ref_head = 0; 361 ci->i_shared_gen = 0; 362 ci->i_rdcache_gen = 0; 363 ci->i_rdcache_revoking = 0; 364 365 INIT_LIST_HEAD(&ci->i_unsafe_writes); 366 INIT_LIST_HEAD(&ci->i_unsafe_dirops); 367 spin_lock_init(&ci->i_unsafe_lock); 368 369 ci->i_snap_realm = NULL; 370 INIT_LIST_HEAD(&ci->i_snap_realm_item); 371 INIT_LIST_HEAD(&ci->i_snap_flush_item); 372 373 INIT_WORK(&ci->i_wb_work, ceph_writeback_work); 374 INIT_WORK(&ci->i_pg_inv_work, ceph_invalidate_work); 375 376 INIT_WORK(&ci->i_vmtruncate_work, ceph_vmtruncate_work); 377 378 return &ci->vfs_inode; 379 } 380 381 static void ceph_i_callback(struct rcu_head *head) 382 { 383 struct inode *inode = container_of(head, struct inode, i_rcu); 384 struct ceph_inode_info *ci = ceph_inode(inode); 385 386 INIT_LIST_HEAD(&inode->i_dentry); 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, implemented; 564 struct timespec mtime, atime, ctime; 565 u32 nsplits; 566 struct ceph_buffer *xattr_blob = NULL; 567 int err = 0; 568 int queue_trunc = 0; 569 570 dout("fill_inode %p ino %llx.%llx v %llu had %llu\n", 571 inode, ceph_vinop(inode), le64_to_cpu(info->version), 572 ci->i_version); 573 574 /* 575 * prealloc xattr data, if it looks like we'll need it. only 576 * if len > 4 (meaning there are actually xattrs; the first 4 577 * bytes are the xattr count). 578 */ 579 if (iinfo->xattr_len > 4) { 580 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS); 581 if (!xattr_blob) 582 pr_err("fill_inode ENOMEM xattr blob %d bytes\n", 583 iinfo->xattr_len); 584 } 585 586 spin_lock(&inode->i_lock); 587 588 /* 589 * provided version will be odd if inode value is projected, 590 * even if stable. skip the update if we have newer stable 591 * info (ours>=theirs, e.g. due to racing mds replies), unless 592 * we are getting projected (unstable) info (in which case the 593 * version is odd, and we want ours>theirs). 594 * us them 595 * 2 2 skip 596 * 3 2 skip 597 * 3 3 update 598 */ 599 if (le64_to_cpu(info->version) > 0 && 600 (ci->i_version & ~1) >= le64_to_cpu(info->version)) 601 goto no_change; 602 603 issued = __ceph_caps_issued(ci, &implemented); 604 issued |= implemented | __ceph_caps_dirty(ci); 605 606 /* update inode */ 607 ci->i_version = le64_to_cpu(info->version); 608 inode->i_version++; 609 inode->i_rdev = le32_to_cpu(info->rdev); 610 611 if ((issued & CEPH_CAP_AUTH_EXCL) == 0) { 612 inode->i_mode = le32_to_cpu(info->mode); 613 inode->i_uid = le32_to_cpu(info->uid); 614 inode->i_gid = le32_to_cpu(info->gid); 615 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode, 616 inode->i_uid, inode->i_gid); 617 } 618 619 if ((issued & CEPH_CAP_LINK_EXCL) == 0) 620 inode->i_nlink = le32_to_cpu(info->nlink); 621 622 /* be careful with mtime, atime, size */ 623 ceph_decode_timespec(&atime, &info->atime); 624 ceph_decode_timespec(&mtime, &info->mtime); 625 ceph_decode_timespec(&ctime, &info->ctime); 626 queue_trunc = ceph_fill_file_size(inode, issued, 627 le32_to_cpu(info->truncate_seq), 628 le64_to_cpu(info->truncate_size), 629 le64_to_cpu(info->size)); 630 ceph_fill_file_time(inode, issued, 631 le32_to_cpu(info->time_warp_seq), 632 &ctime, &mtime, &atime); 633 634 /* only update max_size on auth cap */ 635 if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) && 636 ci->i_max_size != le64_to_cpu(info->max_size)) { 637 dout("max_size %lld -> %llu\n", ci->i_max_size, 638 le64_to_cpu(info->max_size)); 639 ci->i_max_size = le64_to_cpu(info->max_size); 640 } 641 642 ci->i_layout = info->layout; 643 inode->i_blkbits = fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1; 644 645 /* xattrs */ 646 /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */ 647 if ((issued & CEPH_CAP_XATTR_EXCL) == 0 && 648 le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) { 649 if (ci->i_xattrs.blob) 650 ceph_buffer_put(ci->i_xattrs.blob); 651 ci->i_xattrs.blob = xattr_blob; 652 if (xattr_blob) 653 memcpy(ci->i_xattrs.blob->vec.iov_base, 654 iinfo->xattr_data, iinfo->xattr_len); 655 ci->i_xattrs.version = le64_to_cpu(info->xattr_version); 656 xattr_blob = NULL; 657 } 658 659 inode->i_mapping->a_ops = &ceph_aops; 660 inode->i_mapping->backing_dev_info = 661 &ceph_sb_to_client(inode->i_sb)->backing_dev_info; 662 663 switch (inode->i_mode & S_IFMT) { 664 case S_IFIFO: 665 case S_IFBLK: 666 case S_IFCHR: 667 case S_IFSOCK: 668 init_special_inode(inode, inode->i_mode, inode->i_rdev); 669 inode->i_op = &ceph_file_iops; 670 break; 671 case S_IFREG: 672 inode->i_op = &ceph_file_iops; 673 inode->i_fop = &ceph_file_fops; 674 break; 675 case S_IFLNK: 676 inode->i_op = &ceph_symlink_iops; 677 if (!ci->i_symlink) { 678 int symlen = iinfo->symlink_len; 679 char *sym; 680 681 BUG_ON(symlen != inode->i_size); 682 spin_unlock(&inode->i_lock); 683 684 err = -ENOMEM; 685 sym = kmalloc(symlen+1, GFP_NOFS); 686 if (!sym) 687 goto out; 688 memcpy(sym, iinfo->symlink, symlen); 689 sym[symlen] = 0; 690 691 spin_lock(&inode->i_lock); 692 if (!ci->i_symlink) 693 ci->i_symlink = sym; 694 else 695 kfree(sym); /* lost a race */ 696 } 697 break; 698 case S_IFDIR: 699 inode->i_op = &ceph_dir_iops; 700 inode->i_fop = &ceph_dir_fops; 701 702 ci->i_dir_layout = iinfo->dir_layout; 703 704 ci->i_files = le64_to_cpu(info->files); 705 ci->i_subdirs = le64_to_cpu(info->subdirs); 706 ci->i_rbytes = le64_to_cpu(info->rbytes); 707 ci->i_rfiles = le64_to_cpu(info->rfiles); 708 ci->i_rsubdirs = le64_to_cpu(info->rsubdirs); 709 ceph_decode_timespec(&ci->i_rctime, &info->rctime); 710 711 /* set dir completion flag? */ 712 if (ci->i_files == 0 && ci->i_subdirs == 0 && 713 ceph_snap(inode) == CEPH_NOSNAP && 714 (le32_to_cpu(info->cap.caps) & CEPH_CAP_FILE_SHARED) && 715 (issued & CEPH_CAP_FILE_EXCL) == 0 && 716 (ci->i_ceph_flags & CEPH_I_COMPLETE) == 0) { 717 dout(" marking %p complete (empty)\n", inode); 718 /* ci->i_ceph_flags |= CEPH_I_COMPLETE; */ 719 ci->i_max_offset = 2; 720 } 721 break; 722 default: 723 pr_err("fill_inode %llx.%llx BAD mode 0%o\n", 724 ceph_vinop(inode), inode->i_mode); 725 } 726 727 no_change: 728 spin_unlock(&inode->i_lock); 729 730 /* queue truncate if we saw i_size decrease */ 731 if (queue_trunc) 732 ceph_queue_vmtruncate(inode); 733 734 /* populate frag tree */ 735 /* FIXME: move me up, if/when version reflects fragtree changes */ 736 nsplits = le32_to_cpu(info->fragtree.nsplits); 737 mutex_lock(&ci->i_fragtree_mutex); 738 for (i = 0; i < nsplits; i++) { 739 u32 id = le32_to_cpu(info->fragtree.splits[i].frag); 740 struct ceph_inode_frag *frag = __get_or_create_frag(ci, id); 741 742 if (IS_ERR(frag)) 743 continue; 744 frag->split_by = le32_to_cpu(info->fragtree.splits[i].by); 745 dout(" frag %x split by %d\n", frag->frag, frag->split_by); 746 } 747 mutex_unlock(&ci->i_fragtree_mutex); 748 749 /* were we issued a capability? */ 750 if (info->cap.caps) { 751 if (ceph_snap(inode) == CEPH_NOSNAP) { 752 ceph_add_cap(inode, session, 753 le64_to_cpu(info->cap.cap_id), 754 cap_fmode, 755 le32_to_cpu(info->cap.caps), 756 le32_to_cpu(info->cap.wanted), 757 le32_to_cpu(info->cap.seq), 758 le32_to_cpu(info->cap.mseq), 759 le64_to_cpu(info->cap.realm), 760 info->cap.flags, 761 caps_reservation); 762 } else { 763 spin_lock(&inode->i_lock); 764 dout(" %p got snap_caps %s\n", inode, 765 ceph_cap_string(le32_to_cpu(info->cap.caps))); 766 ci->i_snap_caps |= le32_to_cpu(info->cap.caps); 767 if (cap_fmode >= 0) 768 __ceph_get_fmode(ci, cap_fmode); 769 spin_unlock(&inode->i_lock); 770 } 771 } else if (cap_fmode >= 0) { 772 pr_warning("mds issued no caps on %llx.%llx\n", 773 ceph_vinop(inode)); 774 __ceph_get_fmode(ci, cap_fmode); 775 } 776 777 /* update delegation info? */ 778 if (dirinfo) 779 ceph_fill_dirfrag(inode, dirinfo); 780 781 err = 0; 782 783 out: 784 if (xattr_blob) 785 ceph_buffer_put(xattr_blob); 786 return err; 787 } 788 789 /* 790 * caller should hold session s_mutex. 791 */ 792 static void update_dentry_lease(struct dentry *dentry, 793 struct ceph_mds_reply_lease *lease, 794 struct ceph_mds_session *session, 795 unsigned long from_time) 796 { 797 struct ceph_dentry_info *di = ceph_dentry(dentry); 798 long unsigned duration = le32_to_cpu(lease->duration_ms); 799 long unsigned ttl = from_time + (duration * HZ) / 1000; 800 long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000; 801 struct inode *dir; 802 803 /* only track leases on regular dentries */ 804 if (dentry->d_op != &ceph_dentry_ops) 805 return; 806 807 spin_lock(&dentry->d_lock); 808 dout("update_dentry_lease %p mask %d duration %lu ms ttl %lu\n", 809 dentry, le16_to_cpu(lease->mask), duration, ttl); 810 811 /* make lease_rdcache_gen match directory */ 812 dir = dentry->d_parent->d_inode; 813 di->lease_shared_gen = ceph_inode(dir)->i_shared_gen; 814 815 if (lease->mask == 0) 816 goto out_unlock; 817 818 if (di->lease_gen == session->s_cap_gen && 819 time_before(ttl, dentry->d_time)) 820 goto out_unlock; /* we already have a newer lease. */ 821 822 if (di->lease_session && di->lease_session != session) 823 goto out_unlock; 824 825 ceph_dentry_lru_touch(dentry); 826 827 if (!di->lease_session) 828 di->lease_session = ceph_get_mds_session(session); 829 di->lease_gen = session->s_cap_gen; 830 di->lease_seq = le32_to_cpu(lease->seq); 831 di->lease_renew_after = half_ttl; 832 di->lease_renew_from = 0; 833 dentry->d_time = ttl; 834 out_unlock: 835 spin_unlock(&dentry->d_lock); 836 return; 837 } 838 839 /* 840 * Set dentry's directory position based on the current dir's max, and 841 * order it in d_subdirs, so that dcache_readdir behaves. 842 */ 843 static void ceph_set_dentry_offset(struct dentry *dn) 844 { 845 struct dentry *dir = dn->d_parent; 846 struct inode *inode = dn->d_parent->d_inode; 847 struct ceph_dentry_info *di; 848 849 BUG_ON(!inode); 850 851 di = ceph_dentry(dn); 852 853 spin_lock(&inode->i_lock); 854 if ((ceph_inode(inode)->i_ceph_flags & CEPH_I_COMPLETE) == 0) { 855 spin_unlock(&inode->i_lock); 856 return; 857 } 858 di->offset = ceph_inode(inode)->i_max_offset++; 859 spin_unlock(&inode->i_lock); 860 861 spin_lock(&dir->d_lock); 862 spin_lock_nested(&dn->d_lock, DENTRY_D_LOCK_NESTED); 863 list_move(&dn->d_u.d_child, &dir->d_subdirs); 864 dout("set_dentry_offset %p %lld (%p %p)\n", dn, di->offset, 865 dn->d_u.d_child.prev, dn->d_u.d_child.next); 866 spin_unlock(&dn->d_lock); 867 spin_unlock(&dir->d_lock); 868 } 869 870 /* 871 * splice a dentry to an inode. 872 * caller must hold directory i_mutex for this to be safe. 873 * 874 * we will only rehash the resulting dentry if @prehash is 875 * true; @prehash will be set to false (for the benefit of 876 * the caller) if we fail. 877 */ 878 static struct dentry *splice_dentry(struct dentry *dn, struct inode *in, 879 bool *prehash, bool set_offset) 880 { 881 struct dentry *realdn; 882 883 BUG_ON(dn->d_inode); 884 885 /* dn must be unhashed */ 886 if (!d_unhashed(dn)) 887 d_drop(dn); 888 realdn = d_materialise_unique(dn, in); 889 if (IS_ERR(realdn)) { 890 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n", 891 PTR_ERR(realdn), dn, in, ceph_vinop(in)); 892 if (prehash) 893 *prehash = false; /* don't rehash on error */ 894 dn = realdn; /* note realdn contains the error */ 895 goto out; 896 } else if (realdn) { 897 dout("dn %p (%d) spliced with %p (%d) " 898 "inode %p ino %llx.%llx\n", 899 dn, dn->d_count, 900 realdn, realdn->d_count, 901 realdn->d_inode, ceph_vinop(realdn->d_inode)); 902 dput(dn); 903 dn = realdn; 904 } else { 905 BUG_ON(!ceph_dentry(dn)); 906 dout("dn %p attached to %p ino %llx.%llx\n", 907 dn, dn->d_inode, ceph_vinop(dn->d_inode)); 908 } 909 if ((!prehash || *prehash) && d_unhashed(dn)) 910 d_rehash(dn); 911 if (set_offset) 912 ceph_set_dentry_offset(dn); 913 out: 914 return dn; 915 } 916 917 /* 918 * Incorporate results into the local cache. This is either just 919 * one inode, or a directory, dentry, and possibly linked-to inode (e.g., 920 * after a lookup). 921 * 922 * A reply may contain 923 * a directory inode along with a dentry. 924 * and/or a target inode 925 * 926 * Called with snap_rwsem (read). 927 */ 928 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req, 929 struct ceph_mds_session *session) 930 { 931 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; 932 struct inode *in = NULL; 933 struct ceph_mds_reply_inode *ininfo; 934 struct ceph_vino vino; 935 struct ceph_fs_client *fsc = ceph_sb_to_client(sb); 936 int i = 0; 937 int err = 0; 938 939 dout("fill_trace %p is_dentry %d is_target %d\n", req, 940 rinfo->head->is_dentry, rinfo->head->is_target); 941 942 #if 0 943 /* 944 * Debugging hook: 945 * 946 * If we resend completed ops to a recovering mds, we get no 947 * trace. Since that is very rare, pretend this is the case 948 * to ensure the 'no trace' handlers in the callers behave. 949 * 950 * Fill in inodes unconditionally to avoid breaking cap 951 * invariants. 952 */ 953 if (rinfo->head->op & CEPH_MDS_OP_WRITE) { 954 pr_info("fill_trace faking empty trace on %lld %s\n", 955 req->r_tid, ceph_mds_op_name(rinfo->head->op)); 956 if (rinfo->head->is_dentry) { 957 rinfo->head->is_dentry = 0; 958 err = fill_inode(req->r_locked_dir, 959 &rinfo->diri, rinfo->dirfrag, 960 session, req->r_request_started, -1); 961 } 962 if (rinfo->head->is_target) { 963 rinfo->head->is_target = 0; 964 ininfo = rinfo->targeti.in; 965 vino.ino = le64_to_cpu(ininfo->ino); 966 vino.snap = le64_to_cpu(ininfo->snapid); 967 in = ceph_get_inode(sb, vino); 968 err = fill_inode(in, &rinfo->targeti, NULL, 969 session, req->r_request_started, 970 req->r_fmode); 971 iput(in); 972 } 973 } 974 #endif 975 976 if (!rinfo->head->is_target && !rinfo->head->is_dentry) { 977 dout("fill_trace reply is empty!\n"); 978 if (rinfo->head->result == 0 && req->r_locked_dir) 979 ceph_invalidate_dir_request(req); 980 return 0; 981 } 982 983 if (rinfo->head->is_dentry) { 984 struct inode *dir = req->r_locked_dir; 985 986 err = fill_inode(dir, &rinfo->diri, rinfo->dirfrag, 987 session, req->r_request_started, -1, 988 &req->r_caps_reservation); 989 if (err < 0) 990 return err; 991 } 992 993 /* 994 * ignore null lease/binding on snapdir ENOENT, or else we 995 * will have trouble splicing in the virtual snapdir later 996 */ 997 if (rinfo->head->is_dentry && !req->r_aborted && 998 (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name, 999 fsc->mount_options->snapdir_name, 1000 req->r_dentry->d_name.len))) { 1001 /* 1002 * lookup link rename : null -> possibly existing inode 1003 * mknod symlink mkdir : null -> new inode 1004 * unlink : linked -> null 1005 */ 1006 struct inode *dir = req->r_locked_dir; 1007 struct dentry *dn = req->r_dentry; 1008 bool have_dir_cap, have_lease; 1009 1010 BUG_ON(!dn); 1011 BUG_ON(!dir); 1012 BUG_ON(dn->d_parent->d_inode != dir); 1013 BUG_ON(ceph_ino(dir) != 1014 le64_to_cpu(rinfo->diri.in->ino)); 1015 BUG_ON(ceph_snap(dir) != 1016 le64_to_cpu(rinfo->diri.in->snapid)); 1017 1018 /* do we have a lease on the whole dir? */ 1019 have_dir_cap = 1020 (le32_to_cpu(rinfo->diri.in->cap.caps) & 1021 CEPH_CAP_FILE_SHARED); 1022 1023 /* do we have a dn lease? */ 1024 have_lease = have_dir_cap || 1025 (le16_to_cpu(rinfo->dlease->mask) & 1026 CEPH_LOCK_DN); 1027 1028 if (!have_lease) 1029 dout("fill_trace no dentry lease or dir cap\n"); 1030 1031 /* rename? */ 1032 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) { 1033 dout(" src %p '%.*s' dst %p '%.*s'\n", 1034 req->r_old_dentry, 1035 req->r_old_dentry->d_name.len, 1036 req->r_old_dentry->d_name.name, 1037 dn, dn->d_name.len, dn->d_name.name); 1038 dout("fill_trace doing d_move %p -> %p\n", 1039 req->r_old_dentry, dn); 1040 1041 d_move(req->r_old_dentry, dn); 1042 dout(" src %p '%.*s' dst %p '%.*s'\n", 1043 req->r_old_dentry, 1044 req->r_old_dentry->d_name.len, 1045 req->r_old_dentry->d_name.name, 1046 dn, dn->d_name.len, dn->d_name.name); 1047 1048 /* ensure target dentry is invalidated, despite 1049 rehashing bug in vfs_rename_dir */ 1050 ceph_invalidate_dentry_lease(dn); 1051 1052 /* 1053 * d_move() puts the renamed dentry at the end of 1054 * d_subdirs. We need to assign it an appropriate 1055 * directory offset so we can behave when holding 1056 * I_COMPLETE. 1057 */ 1058 ceph_set_dentry_offset(req->r_old_dentry); 1059 dout("dn %p gets new offset %lld\n", req->r_old_dentry, 1060 ceph_dentry(req->r_old_dentry)->offset); 1061 1062 dn = req->r_old_dentry; /* use old_dentry */ 1063 in = dn->d_inode; 1064 } 1065 1066 /* null dentry? */ 1067 if (!rinfo->head->is_target) { 1068 dout("fill_trace null dentry\n"); 1069 if (dn->d_inode) { 1070 dout("d_delete %p\n", dn); 1071 d_delete(dn); 1072 } else { 1073 dout("d_instantiate %p NULL\n", dn); 1074 d_instantiate(dn, NULL); 1075 if (have_lease && d_unhashed(dn)) 1076 d_rehash(dn); 1077 update_dentry_lease(dn, rinfo->dlease, 1078 session, 1079 req->r_request_started); 1080 } 1081 goto done; 1082 } 1083 1084 /* attach proper inode */ 1085 ininfo = rinfo->targeti.in; 1086 vino.ino = le64_to_cpu(ininfo->ino); 1087 vino.snap = le64_to_cpu(ininfo->snapid); 1088 in = dn->d_inode; 1089 if (!in) { 1090 in = ceph_get_inode(sb, vino); 1091 if (IS_ERR(in)) { 1092 pr_err("fill_trace bad get_inode " 1093 "%llx.%llx\n", vino.ino, vino.snap); 1094 err = PTR_ERR(in); 1095 d_delete(dn); 1096 goto done; 1097 } 1098 dn = splice_dentry(dn, in, &have_lease, true); 1099 if (IS_ERR(dn)) { 1100 err = PTR_ERR(dn); 1101 goto done; 1102 } 1103 req->r_dentry = dn; /* may have spliced */ 1104 ihold(in); 1105 } else if (ceph_ino(in) == vino.ino && 1106 ceph_snap(in) == vino.snap) { 1107 ihold(in); 1108 } else { 1109 dout(" %p links to %p %llx.%llx, not %llx.%llx\n", 1110 dn, in, ceph_ino(in), ceph_snap(in), 1111 vino.ino, vino.snap); 1112 have_lease = false; 1113 in = NULL; 1114 } 1115 1116 if (have_lease) 1117 update_dentry_lease(dn, rinfo->dlease, session, 1118 req->r_request_started); 1119 dout(" final dn %p\n", dn); 1120 i++; 1121 } else if (req->r_op == CEPH_MDS_OP_LOOKUPSNAP || 1122 req->r_op == CEPH_MDS_OP_MKSNAP) { 1123 struct dentry *dn = req->r_dentry; 1124 1125 /* fill out a snapdir LOOKUPSNAP dentry */ 1126 BUG_ON(!dn); 1127 BUG_ON(!req->r_locked_dir); 1128 BUG_ON(ceph_snap(req->r_locked_dir) != CEPH_SNAPDIR); 1129 ininfo = rinfo->targeti.in; 1130 vino.ino = le64_to_cpu(ininfo->ino); 1131 vino.snap = le64_to_cpu(ininfo->snapid); 1132 in = ceph_get_inode(sb, vino); 1133 if (IS_ERR(in)) { 1134 pr_err("fill_inode get_inode badness %llx.%llx\n", 1135 vino.ino, vino.snap); 1136 err = PTR_ERR(in); 1137 d_delete(dn); 1138 goto done; 1139 } 1140 dout(" linking snapped dir %p to dn %p\n", in, dn); 1141 dn = splice_dentry(dn, in, NULL, true); 1142 if (IS_ERR(dn)) { 1143 err = PTR_ERR(dn); 1144 goto done; 1145 } 1146 req->r_dentry = dn; /* may have spliced */ 1147 ihold(in); 1148 rinfo->head->is_dentry = 1; /* fool notrace handlers */ 1149 } 1150 1151 if (rinfo->head->is_target) { 1152 vino.ino = le64_to_cpu(rinfo->targeti.in->ino); 1153 vino.snap = le64_to_cpu(rinfo->targeti.in->snapid); 1154 1155 if (in == NULL || ceph_ino(in) != vino.ino || 1156 ceph_snap(in) != vino.snap) { 1157 in = ceph_get_inode(sb, vino); 1158 if (IS_ERR(in)) { 1159 err = PTR_ERR(in); 1160 goto done; 1161 } 1162 } 1163 req->r_target_inode = in; 1164 1165 err = fill_inode(in, 1166 &rinfo->targeti, NULL, 1167 session, req->r_request_started, 1168 (le32_to_cpu(rinfo->head->result) == 0) ? 1169 req->r_fmode : -1, 1170 &req->r_caps_reservation); 1171 if (err < 0) { 1172 pr_err("fill_inode badness %p %llx.%llx\n", 1173 in, ceph_vinop(in)); 1174 goto done; 1175 } 1176 } 1177 1178 done: 1179 dout("fill_trace done err=%d\n", err); 1180 return err; 1181 } 1182 1183 /* 1184 * Prepopulate our cache with readdir results, leases, etc. 1185 */ 1186 int ceph_readdir_prepopulate(struct ceph_mds_request *req, 1187 struct ceph_mds_session *session) 1188 { 1189 struct dentry *parent = req->r_dentry; 1190 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; 1191 struct qstr dname; 1192 struct dentry *dn; 1193 struct inode *in; 1194 int err = 0, i; 1195 struct inode *snapdir = NULL; 1196 struct ceph_mds_request_head *rhead = req->r_request->front.iov_base; 1197 u64 frag = le32_to_cpu(rhead->args.readdir.frag); 1198 struct ceph_dentry_info *di; 1199 1200 if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) { 1201 snapdir = ceph_get_snapdir(parent->d_inode); 1202 parent = d_find_alias(snapdir); 1203 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n", 1204 rinfo->dir_nr, parent); 1205 } else { 1206 dout("readdir_prepopulate %d items under dn %p\n", 1207 rinfo->dir_nr, parent); 1208 if (rinfo->dir_dir) 1209 ceph_fill_dirfrag(parent->d_inode, rinfo->dir_dir); 1210 } 1211 1212 for (i = 0; i < rinfo->dir_nr; i++) { 1213 struct ceph_vino vino; 1214 1215 dname.name = rinfo->dir_dname[i]; 1216 dname.len = rinfo->dir_dname_len[i]; 1217 dname.hash = full_name_hash(dname.name, dname.len); 1218 1219 vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino); 1220 vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid); 1221 1222 retry_lookup: 1223 dn = d_lookup(parent, &dname); 1224 dout("d_lookup on parent=%p name=%.*s got %p\n", 1225 parent, dname.len, dname.name, dn); 1226 1227 if (!dn) { 1228 dn = d_alloc(parent, &dname); 1229 dout("d_alloc %p '%.*s' = %p\n", parent, 1230 dname.len, dname.name, dn); 1231 if (dn == NULL) { 1232 dout("d_alloc badness\n"); 1233 err = -ENOMEM; 1234 goto out; 1235 } 1236 err = ceph_init_dentry(dn); 1237 if (err < 0) { 1238 dput(dn); 1239 goto out; 1240 } 1241 } else if (dn->d_inode && 1242 (ceph_ino(dn->d_inode) != vino.ino || 1243 ceph_snap(dn->d_inode) != vino.snap)) { 1244 dout(" dn %p points to wrong inode %p\n", 1245 dn, dn->d_inode); 1246 d_delete(dn); 1247 dput(dn); 1248 goto retry_lookup; 1249 } else { 1250 /* reorder parent's d_subdirs */ 1251 spin_lock(&parent->d_lock); 1252 spin_lock_nested(&dn->d_lock, DENTRY_D_LOCK_NESTED); 1253 list_move(&dn->d_u.d_child, &parent->d_subdirs); 1254 spin_unlock(&dn->d_lock); 1255 spin_unlock(&parent->d_lock); 1256 } 1257 1258 di = dn->d_fsdata; 1259 di->offset = ceph_make_fpos(frag, i + req->r_readdir_offset); 1260 1261 /* inode */ 1262 if (dn->d_inode) { 1263 in = dn->d_inode; 1264 } else { 1265 in = ceph_get_inode(parent->d_sb, vino); 1266 if (IS_ERR(in)) { 1267 dout("new_inode badness\n"); 1268 d_delete(dn); 1269 dput(dn); 1270 err = PTR_ERR(in); 1271 goto out; 1272 } 1273 dn = splice_dentry(dn, in, NULL, false); 1274 if (IS_ERR(dn)) 1275 dn = NULL; 1276 } 1277 1278 if (fill_inode(in, &rinfo->dir_in[i], NULL, session, 1279 req->r_request_started, -1, 1280 &req->r_caps_reservation) < 0) { 1281 pr_err("fill_inode badness on %p\n", in); 1282 goto next_item; 1283 } 1284 if (dn) 1285 update_dentry_lease(dn, rinfo->dir_dlease[i], 1286 req->r_session, 1287 req->r_request_started); 1288 next_item: 1289 if (dn) 1290 dput(dn); 1291 } 1292 req->r_did_prepopulate = true; 1293 1294 out: 1295 if (snapdir) { 1296 iput(snapdir); 1297 dput(parent); 1298 } 1299 dout("readdir_prepopulate done\n"); 1300 return err; 1301 } 1302 1303 int ceph_inode_set_size(struct inode *inode, loff_t size) 1304 { 1305 struct ceph_inode_info *ci = ceph_inode(inode); 1306 int ret = 0; 1307 1308 spin_lock(&inode->i_lock); 1309 dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size); 1310 inode->i_size = size; 1311 inode->i_blocks = (size + (1 << 9) - 1) >> 9; 1312 1313 /* tell the MDS if we are approaching max_size */ 1314 if ((size << 1) >= ci->i_max_size && 1315 (ci->i_reported_size << 1) < ci->i_max_size) 1316 ret = 1; 1317 1318 spin_unlock(&inode->i_lock); 1319 return ret; 1320 } 1321 1322 /* 1323 * Write back inode data in a worker thread. (This can't be done 1324 * in the message handler context.) 1325 */ 1326 void ceph_queue_writeback(struct inode *inode) 1327 { 1328 if (queue_work(ceph_inode_to_client(inode)->wb_wq, 1329 &ceph_inode(inode)->i_wb_work)) { 1330 dout("ceph_queue_writeback %p\n", inode); 1331 ihold(inode); 1332 } else { 1333 dout("ceph_queue_writeback %p failed\n", inode); 1334 } 1335 } 1336 1337 static void ceph_writeback_work(struct work_struct *work) 1338 { 1339 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info, 1340 i_wb_work); 1341 struct inode *inode = &ci->vfs_inode; 1342 1343 dout("writeback %p\n", inode); 1344 filemap_fdatawrite(&inode->i_data); 1345 iput(inode); 1346 } 1347 1348 /* 1349 * queue an async invalidation 1350 */ 1351 void ceph_queue_invalidate(struct inode *inode) 1352 { 1353 if (queue_work(ceph_inode_to_client(inode)->pg_inv_wq, 1354 &ceph_inode(inode)->i_pg_inv_work)) { 1355 dout("ceph_queue_invalidate %p\n", inode); 1356 ihold(inode); 1357 } else { 1358 dout("ceph_queue_invalidate %p failed\n", inode); 1359 } 1360 } 1361 1362 /* 1363 * invalidate any pages that are not dirty or under writeback. this 1364 * includes pages that are clean and mapped. 1365 */ 1366 static void ceph_invalidate_nondirty_pages(struct address_space *mapping) 1367 { 1368 struct pagevec pvec; 1369 pgoff_t next = 0; 1370 int i; 1371 1372 pagevec_init(&pvec, 0); 1373 while (pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) { 1374 for (i = 0; i < pagevec_count(&pvec); i++) { 1375 struct page *page = pvec.pages[i]; 1376 pgoff_t index; 1377 int skip_page = 1378 (PageDirty(page) || PageWriteback(page)); 1379 1380 if (!skip_page) 1381 skip_page = !trylock_page(page); 1382 1383 /* 1384 * We really shouldn't be looking at the ->index of an 1385 * unlocked page. But we're not allowed to lock these 1386 * pages. So we rely upon nobody altering the ->index 1387 * of this (pinned-by-us) page. 1388 */ 1389 index = page->index; 1390 if (index > next) 1391 next = index; 1392 next++; 1393 1394 if (skip_page) 1395 continue; 1396 1397 generic_error_remove_page(mapping, page); 1398 unlock_page(page); 1399 } 1400 pagevec_release(&pvec); 1401 cond_resched(); 1402 } 1403 } 1404 1405 /* 1406 * Invalidate inode pages in a worker thread. (This can't be done 1407 * in the message handler context.) 1408 */ 1409 static void ceph_invalidate_work(struct work_struct *work) 1410 { 1411 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info, 1412 i_pg_inv_work); 1413 struct inode *inode = &ci->vfs_inode; 1414 u32 orig_gen; 1415 int check = 0; 1416 1417 spin_lock(&inode->i_lock); 1418 dout("invalidate_pages %p gen %d revoking %d\n", inode, 1419 ci->i_rdcache_gen, ci->i_rdcache_revoking); 1420 if (ci->i_rdcache_revoking != ci->i_rdcache_gen) { 1421 /* nevermind! */ 1422 spin_unlock(&inode->i_lock); 1423 goto out; 1424 } 1425 orig_gen = ci->i_rdcache_gen; 1426 spin_unlock(&inode->i_lock); 1427 1428 ceph_invalidate_nondirty_pages(inode->i_mapping); 1429 1430 spin_lock(&inode->i_lock); 1431 if (orig_gen == ci->i_rdcache_gen && 1432 orig_gen == ci->i_rdcache_revoking) { 1433 dout("invalidate_pages %p gen %d successful\n", inode, 1434 ci->i_rdcache_gen); 1435 ci->i_rdcache_revoking--; 1436 check = 1; 1437 } else { 1438 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n", 1439 inode, orig_gen, ci->i_rdcache_gen, 1440 ci->i_rdcache_revoking); 1441 } 1442 spin_unlock(&inode->i_lock); 1443 1444 if (check) 1445 ceph_check_caps(ci, 0, NULL); 1446 out: 1447 iput(inode); 1448 } 1449 1450 1451 /* 1452 * called by trunc_wq; take i_mutex ourselves 1453 * 1454 * We also truncate in a separate thread as well. 1455 */ 1456 static void ceph_vmtruncate_work(struct work_struct *work) 1457 { 1458 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info, 1459 i_vmtruncate_work); 1460 struct inode *inode = &ci->vfs_inode; 1461 1462 dout("vmtruncate_work %p\n", inode); 1463 mutex_lock(&inode->i_mutex); 1464 __ceph_do_pending_vmtruncate(inode); 1465 mutex_unlock(&inode->i_mutex); 1466 iput(inode); 1467 } 1468 1469 /* 1470 * Queue an async vmtruncate. If we fail to queue work, we will handle 1471 * the truncation the next time we call __ceph_do_pending_vmtruncate. 1472 */ 1473 void ceph_queue_vmtruncate(struct inode *inode) 1474 { 1475 struct ceph_inode_info *ci = ceph_inode(inode); 1476 1477 if (queue_work(ceph_sb_to_client(inode->i_sb)->trunc_wq, 1478 &ci->i_vmtruncate_work)) { 1479 dout("ceph_queue_vmtruncate %p\n", inode); 1480 ihold(inode); 1481 } else { 1482 dout("ceph_queue_vmtruncate %p failed, pending=%d\n", 1483 inode, ci->i_truncate_pending); 1484 } 1485 } 1486 1487 /* 1488 * called with i_mutex held. 1489 * 1490 * Make sure any pending truncation is applied before doing anything 1491 * that may depend on it. 1492 */ 1493 void __ceph_do_pending_vmtruncate(struct inode *inode) 1494 { 1495 struct ceph_inode_info *ci = ceph_inode(inode); 1496 u64 to; 1497 int wrbuffer_refs, wake = 0; 1498 1499 retry: 1500 spin_lock(&inode->i_lock); 1501 if (ci->i_truncate_pending == 0) { 1502 dout("__do_pending_vmtruncate %p none pending\n", inode); 1503 spin_unlock(&inode->i_lock); 1504 return; 1505 } 1506 1507 /* 1508 * make sure any dirty snapped pages are flushed before we 1509 * possibly truncate them.. so write AND block! 1510 */ 1511 if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) { 1512 dout("__do_pending_vmtruncate %p flushing snaps first\n", 1513 inode); 1514 spin_unlock(&inode->i_lock); 1515 filemap_write_and_wait_range(&inode->i_data, 0, 1516 inode->i_sb->s_maxbytes); 1517 goto retry; 1518 } 1519 1520 to = ci->i_truncate_size; 1521 wrbuffer_refs = ci->i_wrbuffer_ref; 1522 dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode, 1523 ci->i_truncate_pending, to); 1524 spin_unlock(&inode->i_lock); 1525 1526 truncate_inode_pages(inode->i_mapping, to); 1527 1528 spin_lock(&inode->i_lock); 1529 ci->i_truncate_pending--; 1530 if (ci->i_truncate_pending == 0) 1531 wake = 1; 1532 spin_unlock(&inode->i_lock); 1533 1534 if (wrbuffer_refs == 0) 1535 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL); 1536 if (wake) 1537 wake_up_all(&ci->i_cap_wq); 1538 } 1539 1540 1541 /* 1542 * symlinks 1543 */ 1544 static void *ceph_sym_follow_link(struct dentry *dentry, struct nameidata *nd) 1545 { 1546 struct ceph_inode_info *ci = ceph_inode(dentry->d_inode); 1547 nd_set_link(nd, ci->i_symlink); 1548 return NULL; 1549 } 1550 1551 static const struct inode_operations ceph_symlink_iops = { 1552 .readlink = generic_readlink, 1553 .follow_link = ceph_sym_follow_link, 1554 }; 1555 1556 /* 1557 * setattr 1558 */ 1559 int ceph_setattr(struct dentry *dentry, struct iattr *attr) 1560 { 1561 struct inode *inode = dentry->d_inode; 1562 struct ceph_inode_info *ci = ceph_inode(inode); 1563 struct inode *parent_inode = dentry->d_parent->d_inode; 1564 const unsigned int ia_valid = attr->ia_valid; 1565 struct ceph_mds_request *req; 1566 struct ceph_mds_client *mdsc = ceph_sb_to_client(dentry->d_sb)->mdsc; 1567 int issued; 1568 int release = 0, dirtied = 0; 1569 int mask = 0; 1570 int err = 0; 1571 int inode_dirty_flags = 0; 1572 1573 if (ceph_snap(inode) != CEPH_NOSNAP) 1574 return -EROFS; 1575 1576 __ceph_do_pending_vmtruncate(inode); 1577 1578 err = inode_change_ok(inode, attr); 1579 if (err != 0) 1580 return err; 1581 1582 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR, 1583 USE_AUTH_MDS); 1584 if (IS_ERR(req)) 1585 return PTR_ERR(req); 1586 1587 spin_lock(&inode->i_lock); 1588 issued = __ceph_caps_issued(ci, NULL); 1589 dout("setattr %p issued %s\n", inode, ceph_cap_string(issued)); 1590 1591 if (ia_valid & ATTR_UID) { 1592 dout("setattr %p uid %d -> %d\n", inode, 1593 inode->i_uid, attr->ia_uid); 1594 if (issued & CEPH_CAP_AUTH_EXCL) { 1595 inode->i_uid = attr->ia_uid; 1596 dirtied |= CEPH_CAP_AUTH_EXCL; 1597 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || 1598 attr->ia_uid != inode->i_uid) { 1599 req->r_args.setattr.uid = cpu_to_le32(attr->ia_uid); 1600 mask |= CEPH_SETATTR_UID; 1601 release |= CEPH_CAP_AUTH_SHARED; 1602 } 1603 } 1604 if (ia_valid & ATTR_GID) { 1605 dout("setattr %p gid %d -> %d\n", inode, 1606 inode->i_gid, attr->ia_gid); 1607 if (issued & CEPH_CAP_AUTH_EXCL) { 1608 inode->i_gid = attr->ia_gid; 1609 dirtied |= CEPH_CAP_AUTH_EXCL; 1610 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || 1611 attr->ia_gid != inode->i_gid) { 1612 req->r_args.setattr.gid = cpu_to_le32(attr->ia_gid); 1613 mask |= CEPH_SETATTR_GID; 1614 release |= CEPH_CAP_AUTH_SHARED; 1615 } 1616 } 1617 if (ia_valid & ATTR_MODE) { 1618 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode, 1619 attr->ia_mode); 1620 if (issued & CEPH_CAP_AUTH_EXCL) { 1621 inode->i_mode = attr->ia_mode; 1622 dirtied |= CEPH_CAP_AUTH_EXCL; 1623 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || 1624 attr->ia_mode != inode->i_mode) { 1625 req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode); 1626 mask |= CEPH_SETATTR_MODE; 1627 release |= CEPH_CAP_AUTH_SHARED; 1628 } 1629 } 1630 1631 if (ia_valid & ATTR_ATIME) { 1632 dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode, 1633 inode->i_atime.tv_sec, inode->i_atime.tv_nsec, 1634 attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec); 1635 if (issued & CEPH_CAP_FILE_EXCL) { 1636 ci->i_time_warp_seq++; 1637 inode->i_atime = attr->ia_atime; 1638 dirtied |= CEPH_CAP_FILE_EXCL; 1639 } else if ((issued & CEPH_CAP_FILE_WR) && 1640 timespec_compare(&inode->i_atime, 1641 &attr->ia_atime) < 0) { 1642 inode->i_atime = attr->ia_atime; 1643 dirtied |= CEPH_CAP_FILE_WR; 1644 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || 1645 !timespec_equal(&inode->i_atime, &attr->ia_atime)) { 1646 ceph_encode_timespec(&req->r_args.setattr.atime, 1647 &attr->ia_atime); 1648 mask |= CEPH_SETATTR_ATIME; 1649 release |= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_RD | 1650 CEPH_CAP_FILE_WR; 1651 } 1652 } 1653 if (ia_valid & ATTR_MTIME) { 1654 dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode, 1655 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec, 1656 attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec); 1657 if (issued & CEPH_CAP_FILE_EXCL) { 1658 ci->i_time_warp_seq++; 1659 inode->i_mtime = attr->ia_mtime; 1660 dirtied |= CEPH_CAP_FILE_EXCL; 1661 } else if ((issued & CEPH_CAP_FILE_WR) && 1662 timespec_compare(&inode->i_mtime, 1663 &attr->ia_mtime) < 0) { 1664 inode->i_mtime = attr->ia_mtime; 1665 dirtied |= CEPH_CAP_FILE_WR; 1666 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || 1667 !timespec_equal(&inode->i_mtime, &attr->ia_mtime)) { 1668 ceph_encode_timespec(&req->r_args.setattr.mtime, 1669 &attr->ia_mtime); 1670 mask |= CEPH_SETATTR_MTIME; 1671 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD | 1672 CEPH_CAP_FILE_WR; 1673 } 1674 } 1675 if (ia_valid & ATTR_SIZE) { 1676 dout("setattr %p size %lld -> %lld\n", inode, 1677 inode->i_size, attr->ia_size); 1678 if (attr->ia_size > inode->i_sb->s_maxbytes) { 1679 err = -EINVAL; 1680 goto out; 1681 } 1682 if ((issued & CEPH_CAP_FILE_EXCL) && 1683 attr->ia_size > inode->i_size) { 1684 inode->i_size = attr->ia_size; 1685 inode->i_blocks = 1686 (attr->ia_size + (1 << 9) - 1) >> 9; 1687 inode->i_ctime = attr->ia_ctime; 1688 ci->i_reported_size = attr->ia_size; 1689 dirtied |= CEPH_CAP_FILE_EXCL; 1690 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || 1691 attr->ia_size != inode->i_size) { 1692 req->r_args.setattr.size = cpu_to_le64(attr->ia_size); 1693 req->r_args.setattr.old_size = 1694 cpu_to_le64(inode->i_size); 1695 mask |= CEPH_SETATTR_SIZE; 1696 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD | 1697 CEPH_CAP_FILE_WR; 1698 } 1699 } 1700 1701 /* these do nothing */ 1702 if (ia_valid & ATTR_CTIME) { 1703 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME| 1704 ATTR_MODE|ATTR_UID|ATTR_GID)) == 0; 1705 dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode, 1706 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec, 1707 attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec, 1708 only ? "ctime only" : "ignored"); 1709 inode->i_ctime = attr->ia_ctime; 1710 if (only) { 1711 /* 1712 * if kernel wants to dirty ctime but nothing else, 1713 * we need to choose a cap to dirty under, or do 1714 * a almost-no-op setattr 1715 */ 1716 if (issued & CEPH_CAP_AUTH_EXCL) 1717 dirtied |= CEPH_CAP_AUTH_EXCL; 1718 else if (issued & CEPH_CAP_FILE_EXCL) 1719 dirtied |= CEPH_CAP_FILE_EXCL; 1720 else if (issued & CEPH_CAP_XATTR_EXCL) 1721 dirtied |= CEPH_CAP_XATTR_EXCL; 1722 else 1723 mask |= CEPH_SETATTR_CTIME; 1724 } 1725 } 1726 if (ia_valid & ATTR_FILE) 1727 dout("setattr %p ATTR_FILE ... hrm!\n", inode); 1728 1729 if (dirtied) { 1730 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied); 1731 inode->i_ctime = CURRENT_TIME; 1732 } 1733 1734 release &= issued; 1735 spin_unlock(&inode->i_lock); 1736 1737 if (inode_dirty_flags) 1738 __mark_inode_dirty(inode, inode_dirty_flags); 1739 1740 if (mask) { 1741 req->r_inode = inode; 1742 ihold(inode); 1743 req->r_inode_drop = release; 1744 req->r_args.setattr.mask = cpu_to_le32(mask); 1745 req->r_num_caps = 1; 1746 err = ceph_mdsc_do_request(mdsc, parent_inode, req); 1747 } 1748 dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err, 1749 ceph_cap_string(dirtied), mask); 1750 1751 ceph_mdsc_put_request(req); 1752 __ceph_do_pending_vmtruncate(inode); 1753 return err; 1754 out: 1755 spin_unlock(&inode->i_lock); 1756 ceph_mdsc_put_request(req); 1757 return err; 1758 } 1759 1760 /* 1761 * Verify that we have a lease on the given mask. If not, 1762 * do a getattr against an mds. 1763 */ 1764 int ceph_do_getattr(struct inode *inode, int mask) 1765 { 1766 struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb); 1767 struct ceph_mds_client *mdsc = fsc->mdsc; 1768 struct ceph_mds_request *req; 1769 int err; 1770 1771 if (ceph_snap(inode) == CEPH_SNAPDIR) { 1772 dout("do_getattr inode %p SNAPDIR\n", inode); 1773 return 0; 1774 } 1775 1776 dout("do_getattr inode %p mask %s mode 0%o\n", inode, ceph_cap_string(mask), inode->i_mode); 1777 if (ceph_caps_issued_mask(ceph_inode(inode), mask, 1)) 1778 return 0; 1779 1780 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS); 1781 if (IS_ERR(req)) 1782 return PTR_ERR(req); 1783 req->r_inode = inode; 1784 ihold(inode); 1785 req->r_num_caps = 1; 1786 req->r_args.getattr.mask = cpu_to_le32(mask); 1787 err = ceph_mdsc_do_request(mdsc, NULL, req); 1788 ceph_mdsc_put_request(req); 1789 dout("do_getattr result=%d\n", err); 1790 return err; 1791 } 1792 1793 1794 /* 1795 * Check inode permissions. We verify we have a valid value for 1796 * the AUTH cap, then call the generic handler. 1797 */ 1798 int ceph_permission(struct inode *inode, int mask, unsigned int flags) 1799 { 1800 int err; 1801 1802 if (flags & IPERM_FLAG_RCU) 1803 return -ECHILD; 1804 1805 err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED); 1806 1807 if (!err) 1808 err = generic_permission(inode, mask, flags); 1809 return err; 1810 } 1811 1812 /* 1813 * Get all attributes. Hopefully somedata we'll have a statlite() 1814 * and can limit the fields we require to be accurate. 1815 */ 1816 int ceph_getattr(struct vfsmount *mnt, struct dentry *dentry, 1817 struct kstat *stat) 1818 { 1819 struct inode *inode = dentry->d_inode; 1820 struct ceph_inode_info *ci = ceph_inode(inode); 1821 int err; 1822 1823 err = ceph_do_getattr(inode, CEPH_STAT_CAP_INODE_ALL); 1824 if (!err) { 1825 generic_fillattr(inode, stat); 1826 stat->ino = ceph_translate_ino(inode->i_sb, inode->i_ino); 1827 if (ceph_snap(inode) != CEPH_NOSNAP) 1828 stat->dev = ceph_snap(inode); 1829 else 1830 stat->dev = 0; 1831 if (S_ISDIR(inode->i_mode)) { 1832 if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), 1833 RBYTES)) 1834 stat->size = ci->i_rbytes; 1835 else 1836 stat->size = ci->i_files + ci->i_subdirs; 1837 stat->blocks = 0; 1838 stat->blksize = 65536; 1839 } 1840 } 1841 return err; 1842 } 1843