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 duration %lu ms ttl %lu\n", 809 dentry, 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 (duration == 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 le32_to_cpu(rinfo->dlease->duration_ms); 1026 if (!have_lease) 1027 dout("fill_trace no dentry lease or dir cap\n"); 1028 1029 /* rename? */ 1030 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) { 1031 dout(" src %p '%.*s' dst %p '%.*s'\n", 1032 req->r_old_dentry, 1033 req->r_old_dentry->d_name.len, 1034 req->r_old_dentry->d_name.name, 1035 dn, dn->d_name.len, dn->d_name.name); 1036 dout("fill_trace doing d_move %p -> %p\n", 1037 req->r_old_dentry, dn); 1038 1039 d_move(req->r_old_dentry, dn); 1040 dout(" src %p '%.*s' dst %p '%.*s'\n", 1041 req->r_old_dentry, 1042 req->r_old_dentry->d_name.len, 1043 req->r_old_dentry->d_name.name, 1044 dn, dn->d_name.len, dn->d_name.name); 1045 1046 /* ensure target dentry is invalidated, despite 1047 rehashing bug in vfs_rename_dir */ 1048 ceph_invalidate_dentry_lease(dn); 1049 1050 /* 1051 * d_move() puts the renamed dentry at the end of 1052 * d_subdirs. We need to assign it an appropriate 1053 * directory offset so we can behave when holding 1054 * I_COMPLETE. 1055 */ 1056 ceph_set_dentry_offset(req->r_old_dentry); 1057 dout("dn %p gets new offset %lld\n", req->r_old_dentry, 1058 ceph_dentry(req->r_old_dentry)->offset); 1059 1060 dn = req->r_old_dentry; /* use old_dentry */ 1061 in = dn->d_inode; 1062 } 1063 1064 /* null dentry? */ 1065 if (!rinfo->head->is_target) { 1066 dout("fill_trace null dentry\n"); 1067 if (dn->d_inode) { 1068 dout("d_delete %p\n", dn); 1069 d_delete(dn); 1070 } else { 1071 dout("d_instantiate %p NULL\n", dn); 1072 d_instantiate(dn, NULL); 1073 if (have_lease && d_unhashed(dn)) 1074 d_rehash(dn); 1075 update_dentry_lease(dn, rinfo->dlease, 1076 session, 1077 req->r_request_started); 1078 } 1079 goto done; 1080 } 1081 1082 /* attach proper inode */ 1083 ininfo = rinfo->targeti.in; 1084 vino.ino = le64_to_cpu(ininfo->ino); 1085 vino.snap = le64_to_cpu(ininfo->snapid); 1086 in = dn->d_inode; 1087 if (!in) { 1088 in = ceph_get_inode(sb, vino); 1089 if (IS_ERR(in)) { 1090 pr_err("fill_trace bad get_inode " 1091 "%llx.%llx\n", vino.ino, vino.snap); 1092 err = PTR_ERR(in); 1093 d_delete(dn); 1094 goto done; 1095 } 1096 dn = splice_dentry(dn, in, &have_lease, true); 1097 if (IS_ERR(dn)) { 1098 err = PTR_ERR(dn); 1099 goto done; 1100 } 1101 req->r_dentry = dn; /* may have spliced */ 1102 ihold(in); 1103 } else if (ceph_ino(in) == vino.ino && 1104 ceph_snap(in) == vino.snap) { 1105 ihold(in); 1106 } else { 1107 dout(" %p links to %p %llx.%llx, not %llx.%llx\n", 1108 dn, in, ceph_ino(in), ceph_snap(in), 1109 vino.ino, vino.snap); 1110 have_lease = false; 1111 in = NULL; 1112 } 1113 1114 if (have_lease) 1115 update_dentry_lease(dn, rinfo->dlease, session, 1116 req->r_request_started); 1117 dout(" final dn %p\n", dn); 1118 i++; 1119 } else if (req->r_op == CEPH_MDS_OP_LOOKUPSNAP || 1120 req->r_op == CEPH_MDS_OP_MKSNAP) { 1121 struct dentry *dn = req->r_dentry; 1122 1123 /* fill out a snapdir LOOKUPSNAP dentry */ 1124 BUG_ON(!dn); 1125 BUG_ON(!req->r_locked_dir); 1126 BUG_ON(ceph_snap(req->r_locked_dir) != CEPH_SNAPDIR); 1127 ininfo = rinfo->targeti.in; 1128 vino.ino = le64_to_cpu(ininfo->ino); 1129 vino.snap = le64_to_cpu(ininfo->snapid); 1130 in = ceph_get_inode(sb, vino); 1131 if (IS_ERR(in)) { 1132 pr_err("fill_inode get_inode badness %llx.%llx\n", 1133 vino.ino, vino.snap); 1134 err = PTR_ERR(in); 1135 d_delete(dn); 1136 goto done; 1137 } 1138 dout(" linking snapped dir %p to dn %p\n", in, dn); 1139 dn = splice_dentry(dn, in, NULL, true); 1140 if (IS_ERR(dn)) { 1141 err = PTR_ERR(dn); 1142 goto done; 1143 } 1144 req->r_dentry = dn; /* may have spliced */ 1145 ihold(in); 1146 rinfo->head->is_dentry = 1; /* fool notrace handlers */ 1147 } 1148 1149 if (rinfo->head->is_target) { 1150 vino.ino = le64_to_cpu(rinfo->targeti.in->ino); 1151 vino.snap = le64_to_cpu(rinfo->targeti.in->snapid); 1152 1153 if (in == NULL || ceph_ino(in) != vino.ino || 1154 ceph_snap(in) != vino.snap) { 1155 in = ceph_get_inode(sb, vino); 1156 if (IS_ERR(in)) { 1157 err = PTR_ERR(in); 1158 goto done; 1159 } 1160 } 1161 req->r_target_inode = in; 1162 1163 err = fill_inode(in, 1164 &rinfo->targeti, NULL, 1165 session, req->r_request_started, 1166 (le32_to_cpu(rinfo->head->result) == 0) ? 1167 req->r_fmode : -1, 1168 &req->r_caps_reservation); 1169 if (err < 0) { 1170 pr_err("fill_inode badness %p %llx.%llx\n", 1171 in, ceph_vinop(in)); 1172 goto done; 1173 } 1174 } 1175 1176 done: 1177 dout("fill_trace done err=%d\n", err); 1178 return err; 1179 } 1180 1181 /* 1182 * Prepopulate our cache with readdir results, leases, etc. 1183 */ 1184 int ceph_readdir_prepopulate(struct ceph_mds_request *req, 1185 struct ceph_mds_session *session) 1186 { 1187 struct dentry *parent = req->r_dentry; 1188 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; 1189 struct qstr dname; 1190 struct dentry *dn; 1191 struct inode *in; 1192 int err = 0, i; 1193 struct inode *snapdir = NULL; 1194 struct ceph_mds_request_head *rhead = req->r_request->front.iov_base; 1195 u64 frag = le32_to_cpu(rhead->args.readdir.frag); 1196 struct ceph_dentry_info *di; 1197 1198 if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) { 1199 snapdir = ceph_get_snapdir(parent->d_inode); 1200 parent = d_find_alias(snapdir); 1201 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n", 1202 rinfo->dir_nr, parent); 1203 } else { 1204 dout("readdir_prepopulate %d items under dn %p\n", 1205 rinfo->dir_nr, parent); 1206 if (rinfo->dir_dir) 1207 ceph_fill_dirfrag(parent->d_inode, rinfo->dir_dir); 1208 } 1209 1210 for (i = 0; i < rinfo->dir_nr; i++) { 1211 struct ceph_vino vino; 1212 1213 dname.name = rinfo->dir_dname[i]; 1214 dname.len = rinfo->dir_dname_len[i]; 1215 dname.hash = full_name_hash(dname.name, dname.len); 1216 1217 vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino); 1218 vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid); 1219 1220 retry_lookup: 1221 dn = d_lookup(parent, &dname); 1222 dout("d_lookup on parent=%p name=%.*s got %p\n", 1223 parent, dname.len, dname.name, dn); 1224 1225 if (!dn) { 1226 dn = d_alloc(parent, &dname); 1227 dout("d_alloc %p '%.*s' = %p\n", parent, 1228 dname.len, dname.name, dn); 1229 if (dn == NULL) { 1230 dout("d_alloc badness\n"); 1231 err = -ENOMEM; 1232 goto out; 1233 } 1234 err = ceph_init_dentry(dn); 1235 if (err < 0) { 1236 dput(dn); 1237 goto out; 1238 } 1239 } else if (dn->d_inode && 1240 (ceph_ino(dn->d_inode) != vino.ino || 1241 ceph_snap(dn->d_inode) != vino.snap)) { 1242 dout(" dn %p points to wrong inode %p\n", 1243 dn, dn->d_inode); 1244 d_delete(dn); 1245 dput(dn); 1246 goto retry_lookup; 1247 } else { 1248 /* reorder parent's d_subdirs */ 1249 spin_lock(&parent->d_lock); 1250 spin_lock_nested(&dn->d_lock, DENTRY_D_LOCK_NESTED); 1251 list_move(&dn->d_u.d_child, &parent->d_subdirs); 1252 spin_unlock(&dn->d_lock); 1253 spin_unlock(&parent->d_lock); 1254 } 1255 1256 di = dn->d_fsdata; 1257 di->offset = ceph_make_fpos(frag, i + req->r_readdir_offset); 1258 1259 /* inode */ 1260 if (dn->d_inode) { 1261 in = dn->d_inode; 1262 } else { 1263 in = ceph_get_inode(parent->d_sb, vino); 1264 if (IS_ERR(in)) { 1265 dout("new_inode badness\n"); 1266 d_delete(dn); 1267 dput(dn); 1268 err = PTR_ERR(in); 1269 goto out; 1270 } 1271 dn = splice_dentry(dn, in, NULL, false); 1272 if (IS_ERR(dn)) 1273 dn = NULL; 1274 } 1275 1276 if (fill_inode(in, &rinfo->dir_in[i], NULL, session, 1277 req->r_request_started, -1, 1278 &req->r_caps_reservation) < 0) { 1279 pr_err("fill_inode badness on %p\n", in); 1280 goto next_item; 1281 } 1282 if (dn) 1283 update_dentry_lease(dn, rinfo->dir_dlease[i], 1284 req->r_session, 1285 req->r_request_started); 1286 next_item: 1287 if (dn) 1288 dput(dn); 1289 } 1290 req->r_did_prepopulate = true; 1291 1292 out: 1293 if (snapdir) { 1294 iput(snapdir); 1295 dput(parent); 1296 } 1297 dout("readdir_prepopulate done\n"); 1298 return err; 1299 } 1300 1301 int ceph_inode_set_size(struct inode *inode, loff_t size) 1302 { 1303 struct ceph_inode_info *ci = ceph_inode(inode); 1304 int ret = 0; 1305 1306 spin_lock(&inode->i_lock); 1307 dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size); 1308 inode->i_size = size; 1309 inode->i_blocks = (size + (1 << 9) - 1) >> 9; 1310 1311 /* tell the MDS if we are approaching max_size */ 1312 if ((size << 1) >= ci->i_max_size && 1313 (ci->i_reported_size << 1) < ci->i_max_size) 1314 ret = 1; 1315 1316 spin_unlock(&inode->i_lock); 1317 return ret; 1318 } 1319 1320 /* 1321 * Write back inode data in a worker thread. (This can't be done 1322 * in the message handler context.) 1323 */ 1324 void ceph_queue_writeback(struct inode *inode) 1325 { 1326 if (queue_work(ceph_inode_to_client(inode)->wb_wq, 1327 &ceph_inode(inode)->i_wb_work)) { 1328 dout("ceph_queue_writeback %p\n", inode); 1329 ihold(inode); 1330 } else { 1331 dout("ceph_queue_writeback %p failed\n", inode); 1332 } 1333 } 1334 1335 static void ceph_writeback_work(struct work_struct *work) 1336 { 1337 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info, 1338 i_wb_work); 1339 struct inode *inode = &ci->vfs_inode; 1340 1341 dout("writeback %p\n", inode); 1342 filemap_fdatawrite(&inode->i_data); 1343 iput(inode); 1344 } 1345 1346 /* 1347 * queue an async invalidation 1348 */ 1349 void ceph_queue_invalidate(struct inode *inode) 1350 { 1351 if (queue_work(ceph_inode_to_client(inode)->pg_inv_wq, 1352 &ceph_inode(inode)->i_pg_inv_work)) { 1353 dout("ceph_queue_invalidate %p\n", inode); 1354 ihold(inode); 1355 } else { 1356 dout("ceph_queue_invalidate %p failed\n", inode); 1357 } 1358 } 1359 1360 /* 1361 * invalidate any pages that are not dirty or under writeback. this 1362 * includes pages that are clean and mapped. 1363 */ 1364 static void ceph_invalidate_nondirty_pages(struct address_space *mapping) 1365 { 1366 struct pagevec pvec; 1367 pgoff_t next = 0; 1368 int i; 1369 1370 pagevec_init(&pvec, 0); 1371 while (pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) { 1372 for (i = 0; i < pagevec_count(&pvec); i++) { 1373 struct page *page = pvec.pages[i]; 1374 pgoff_t index; 1375 int skip_page = 1376 (PageDirty(page) || PageWriteback(page)); 1377 1378 if (!skip_page) 1379 skip_page = !trylock_page(page); 1380 1381 /* 1382 * We really shouldn't be looking at the ->index of an 1383 * unlocked page. But we're not allowed to lock these 1384 * pages. So we rely upon nobody altering the ->index 1385 * of this (pinned-by-us) page. 1386 */ 1387 index = page->index; 1388 if (index > next) 1389 next = index; 1390 next++; 1391 1392 if (skip_page) 1393 continue; 1394 1395 generic_error_remove_page(mapping, page); 1396 unlock_page(page); 1397 } 1398 pagevec_release(&pvec); 1399 cond_resched(); 1400 } 1401 } 1402 1403 /* 1404 * Invalidate inode pages in a worker thread. (This can't be done 1405 * in the message handler context.) 1406 */ 1407 static void ceph_invalidate_work(struct work_struct *work) 1408 { 1409 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info, 1410 i_pg_inv_work); 1411 struct inode *inode = &ci->vfs_inode; 1412 u32 orig_gen; 1413 int check = 0; 1414 1415 spin_lock(&inode->i_lock); 1416 dout("invalidate_pages %p gen %d revoking %d\n", inode, 1417 ci->i_rdcache_gen, ci->i_rdcache_revoking); 1418 if (ci->i_rdcache_revoking != ci->i_rdcache_gen) { 1419 /* nevermind! */ 1420 spin_unlock(&inode->i_lock); 1421 goto out; 1422 } 1423 orig_gen = ci->i_rdcache_gen; 1424 spin_unlock(&inode->i_lock); 1425 1426 ceph_invalidate_nondirty_pages(inode->i_mapping); 1427 1428 spin_lock(&inode->i_lock); 1429 if (orig_gen == ci->i_rdcache_gen && 1430 orig_gen == ci->i_rdcache_revoking) { 1431 dout("invalidate_pages %p gen %d successful\n", inode, 1432 ci->i_rdcache_gen); 1433 ci->i_rdcache_revoking--; 1434 check = 1; 1435 } else { 1436 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n", 1437 inode, orig_gen, ci->i_rdcache_gen, 1438 ci->i_rdcache_revoking); 1439 } 1440 spin_unlock(&inode->i_lock); 1441 1442 if (check) 1443 ceph_check_caps(ci, 0, NULL); 1444 out: 1445 iput(inode); 1446 } 1447 1448 1449 /* 1450 * called by trunc_wq; take i_mutex ourselves 1451 * 1452 * We also truncate in a separate thread as well. 1453 */ 1454 static void ceph_vmtruncate_work(struct work_struct *work) 1455 { 1456 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info, 1457 i_vmtruncate_work); 1458 struct inode *inode = &ci->vfs_inode; 1459 1460 dout("vmtruncate_work %p\n", inode); 1461 mutex_lock(&inode->i_mutex); 1462 __ceph_do_pending_vmtruncate(inode); 1463 mutex_unlock(&inode->i_mutex); 1464 iput(inode); 1465 } 1466 1467 /* 1468 * Queue an async vmtruncate. If we fail to queue work, we will handle 1469 * the truncation the next time we call __ceph_do_pending_vmtruncate. 1470 */ 1471 void ceph_queue_vmtruncate(struct inode *inode) 1472 { 1473 struct ceph_inode_info *ci = ceph_inode(inode); 1474 1475 if (queue_work(ceph_sb_to_client(inode->i_sb)->trunc_wq, 1476 &ci->i_vmtruncate_work)) { 1477 dout("ceph_queue_vmtruncate %p\n", inode); 1478 ihold(inode); 1479 } else { 1480 dout("ceph_queue_vmtruncate %p failed, pending=%d\n", 1481 inode, ci->i_truncate_pending); 1482 } 1483 } 1484 1485 /* 1486 * called with i_mutex held. 1487 * 1488 * Make sure any pending truncation is applied before doing anything 1489 * that may depend on it. 1490 */ 1491 void __ceph_do_pending_vmtruncate(struct inode *inode) 1492 { 1493 struct ceph_inode_info *ci = ceph_inode(inode); 1494 u64 to; 1495 int wrbuffer_refs, wake = 0; 1496 1497 retry: 1498 spin_lock(&inode->i_lock); 1499 if (ci->i_truncate_pending == 0) { 1500 dout("__do_pending_vmtruncate %p none pending\n", inode); 1501 spin_unlock(&inode->i_lock); 1502 return; 1503 } 1504 1505 /* 1506 * make sure any dirty snapped pages are flushed before we 1507 * possibly truncate them.. so write AND block! 1508 */ 1509 if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) { 1510 dout("__do_pending_vmtruncate %p flushing snaps first\n", 1511 inode); 1512 spin_unlock(&inode->i_lock); 1513 filemap_write_and_wait_range(&inode->i_data, 0, 1514 inode->i_sb->s_maxbytes); 1515 goto retry; 1516 } 1517 1518 to = ci->i_truncate_size; 1519 wrbuffer_refs = ci->i_wrbuffer_ref; 1520 dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode, 1521 ci->i_truncate_pending, to); 1522 spin_unlock(&inode->i_lock); 1523 1524 truncate_inode_pages(inode->i_mapping, to); 1525 1526 spin_lock(&inode->i_lock); 1527 ci->i_truncate_pending--; 1528 if (ci->i_truncate_pending == 0) 1529 wake = 1; 1530 spin_unlock(&inode->i_lock); 1531 1532 if (wrbuffer_refs == 0) 1533 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL); 1534 if (wake) 1535 wake_up_all(&ci->i_cap_wq); 1536 } 1537 1538 1539 /* 1540 * symlinks 1541 */ 1542 static void *ceph_sym_follow_link(struct dentry *dentry, struct nameidata *nd) 1543 { 1544 struct ceph_inode_info *ci = ceph_inode(dentry->d_inode); 1545 nd_set_link(nd, ci->i_symlink); 1546 return NULL; 1547 } 1548 1549 static const struct inode_operations ceph_symlink_iops = { 1550 .readlink = generic_readlink, 1551 .follow_link = ceph_sym_follow_link, 1552 }; 1553 1554 /* 1555 * setattr 1556 */ 1557 int ceph_setattr(struct dentry *dentry, struct iattr *attr) 1558 { 1559 struct inode *inode = dentry->d_inode; 1560 struct ceph_inode_info *ci = ceph_inode(inode); 1561 struct inode *parent_inode = dentry->d_parent->d_inode; 1562 const unsigned int ia_valid = attr->ia_valid; 1563 struct ceph_mds_request *req; 1564 struct ceph_mds_client *mdsc = ceph_sb_to_client(dentry->d_sb)->mdsc; 1565 int issued; 1566 int release = 0, dirtied = 0; 1567 int mask = 0; 1568 int err = 0; 1569 int inode_dirty_flags = 0; 1570 1571 if (ceph_snap(inode) != CEPH_NOSNAP) 1572 return -EROFS; 1573 1574 __ceph_do_pending_vmtruncate(inode); 1575 1576 err = inode_change_ok(inode, attr); 1577 if (err != 0) 1578 return err; 1579 1580 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR, 1581 USE_AUTH_MDS); 1582 if (IS_ERR(req)) 1583 return PTR_ERR(req); 1584 1585 spin_lock(&inode->i_lock); 1586 issued = __ceph_caps_issued(ci, NULL); 1587 dout("setattr %p issued %s\n", inode, ceph_cap_string(issued)); 1588 1589 if (ia_valid & ATTR_UID) { 1590 dout("setattr %p uid %d -> %d\n", inode, 1591 inode->i_uid, attr->ia_uid); 1592 if (issued & CEPH_CAP_AUTH_EXCL) { 1593 inode->i_uid = attr->ia_uid; 1594 dirtied |= CEPH_CAP_AUTH_EXCL; 1595 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || 1596 attr->ia_uid != inode->i_uid) { 1597 req->r_args.setattr.uid = cpu_to_le32(attr->ia_uid); 1598 mask |= CEPH_SETATTR_UID; 1599 release |= CEPH_CAP_AUTH_SHARED; 1600 } 1601 } 1602 if (ia_valid & ATTR_GID) { 1603 dout("setattr %p gid %d -> %d\n", inode, 1604 inode->i_gid, attr->ia_gid); 1605 if (issued & CEPH_CAP_AUTH_EXCL) { 1606 inode->i_gid = attr->ia_gid; 1607 dirtied |= CEPH_CAP_AUTH_EXCL; 1608 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || 1609 attr->ia_gid != inode->i_gid) { 1610 req->r_args.setattr.gid = cpu_to_le32(attr->ia_gid); 1611 mask |= CEPH_SETATTR_GID; 1612 release |= CEPH_CAP_AUTH_SHARED; 1613 } 1614 } 1615 if (ia_valid & ATTR_MODE) { 1616 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode, 1617 attr->ia_mode); 1618 if (issued & CEPH_CAP_AUTH_EXCL) { 1619 inode->i_mode = attr->ia_mode; 1620 dirtied |= CEPH_CAP_AUTH_EXCL; 1621 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || 1622 attr->ia_mode != inode->i_mode) { 1623 req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode); 1624 mask |= CEPH_SETATTR_MODE; 1625 release |= CEPH_CAP_AUTH_SHARED; 1626 } 1627 } 1628 1629 if (ia_valid & ATTR_ATIME) { 1630 dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode, 1631 inode->i_atime.tv_sec, inode->i_atime.tv_nsec, 1632 attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec); 1633 if (issued & CEPH_CAP_FILE_EXCL) { 1634 ci->i_time_warp_seq++; 1635 inode->i_atime = attr->ia_atime; 1636 dirtied |= CEPH_CAP_FILE_EXCL; 1637 } else if ((issued & CEPH_CAP_FILE_WR) && 1638 timespec_compare(&inode->i_atime, 1639 &attr->ia_atime) < 0) { 1640 inode->i_atime = attr->ia_atime; 1641 dirtied |= CEPH_CAP_FILE_WR; 1642 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || 1643 !timespec_equal(&inode->i_atime, &attr->ia_atime)) { 1644 ceph_encode_timespec(&req->r_args.setattr.atime, 1645 &attr->ia_atime); 1646 mask |= CEPH_SETATTR_ATIME; 1647 release |= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_RD | 1648 CEPH_CAP_FILE_WR; 1649 } 1650 } 1651 if (ia_valid & ATTR_MTIME) { 1652 dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode, 1653 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec, 1654 attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec); 1655 if (issued & CEPH_CAP_FILE_EXCL) { 1656 ci->i_time_warp_seq++; 1657 inode->i_mtime = attr->ia_mtime; 1658 dirtied |= CEPH_CAP_FILE_EXCL; 1659 } else if ((issued & CEPH_CAP_FILE_WR) && 1660 timespec_compare(&inode->i_mtime, 1661 &attr->ia_mtime) < 0) { 1662 inode->i_mtime = attr->ia_mtime; 1663 dirtied |= CEPH_CAP_FILE_WR; 1664 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || 1665 !timespec_equal(&inode->i_mtime, &attr->ia_mtime)) { 1666 ceph_encode_timespec(&req->r_args.setattr.mtime, 1667 &attr->ia_mtime); 1668 mask |= CEPH_SETATTR_MTIME; 1669 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD | 1670 CEPH_CAP_FILE_WR; 1671 } 1672 } 1673 if (ia_valid & ATTR_SIZE) { 1674 dout("setattr %p size %lld -> %lld\n", inode, 1675 inode->i_size, attr->ia_size); 1676 if (attr->ia_size > inode->i_sb->s_maxbytes) { 1677 err = -EINVAL; 1678 goto out; 1679 } 1680 if ((issued & CEPH_CAP_FILE_EXCL) && 1681 attr->ia_size > inode->i_size) { 1682 inode->i_size = attr->ia_size; 1683 inode->i_blocks = 1684 (attr->ia_size + (1 << 9) - 1) >> 9; 1685 inode->i_ctime = attr->ia_ctime; 1686 ci->i_reported_size = attr->ia_size; 1687 dirtied |= CEPH_CAP_FILE_EXCL; 1688 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || 1689 attr->ia_size != inode->i_size) { 1690 req->r_args.setattr.size = cpu_to_le64(attr->ia_size); 1691 req->r_args.setattr.old_size = 1692 cpu_to_le64(inode->i_size); 1693 mask |= CEPH_SETATTR_SIZE; 1694 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD | 1695 CEPH_CAP_FILE_WR; 1696 } 1697 } 1698 1699 /* these do nothing */ 1700 if (ia_valid & ATTR_CTIME) { 1701 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME| 1702 ATTR_MODE|ATTR_UID|ATTR_GID)) == 0; 1703 dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode, 1704 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec, 1705 attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec, 1706 only ? "ctime only" : "ignored"); 1707 inode->i_ctime = attr->ia_ctime; 1708 if (only) { 1709 /* 1710 * if kernel wants to dirty ctime but nothing else, 1711 * we need to choose a cap to dirty under, or do 1712 * a almost-no-op setattr 1713 */ 1714 if (issued & CEPH_CAP_AUTH_EXCL) 1715 dirtied |= CEPH_CAP_AUTH_EXCL; 1716 else if (issued & CEPH_CAP_FILE_EXCL) 1717 dirtied |= CEPH_CAP_FILE_EXCL; 1718 else if (issued & CEPH_CAP_XATTR_EXCL) 1719 dirtied |= CEPH_CAP_XATTR_EXCL; 1720 else 1721 mask |= CEPH_SETATTR_CTIME; 1722 } 1723 } 1724 if (ia_valid & ATTR_FILE) 1725 dout("setattr %p ATTR_FILE ... hrm!\n", inode); 1726 1727 if (dirtied) { 1728 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied); 1729 inode->i_ctime = CURRENT_TIME; 1730 } 1731 1732 release &= issued; 1733 spin_unlock(&inode->i_lock); 1734 1735 if (inode_dirty_flags) 1736 __mark_inode_dirty(inode, inode_dirty_flags); 1737 1738 if (mask) { 1739 req->r_inode = inode; 1740 ihold(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 = inode; 1782 ihold(inode); 1783 req->r_num_caps = 1; 1784 req->r_args.getattr.mask = cpu_to_le32(mask); 1785 err = ceph_mdsc_do_request(mdsc, NULL, req); 1786 ceph_mdsc_put_request(req); 1787 dout("do_getattr result=%d\n", err); 1788 return err; 1789 } 1790 1791 1792 /* 1793 * Check inode permissions. We verify we have a valid value for 1794 * the AUTH cap, then call the generic handler. 1795 */ 1796 int ceph_permission(struct inode *inode, int mask, unsigned int flags) 1797 { 1798 int err; 1799 1800 if (flags & IPERM_FLAG_RCU) 1801 return -ECHILD; 1802 1803 err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED); 1804 1805 if (!err) 1806 err = generic_permission(inode, mask, flags, NULL); 1807 return err; 1808 } 1809 1810 /* 1811 * Get all attributes. Hopefully somedata we'll have a statlite() 1812 * and can limit the fields we require to be accurate. 1813 */ 1814 int ceph_getattr(struct vfsmount *mnt, struct dentry *dentry, 1815 struct kstat *stat) 1816 { 1817 struct inode *inode = dentry->d_inode; 1818 struct ceph_inode_info *ci = ceph_inode(inode); 1819 int err; 1820 1821 err = ceph_do_getattr(inode, CEPH_STAT_CAP_INODE_ALL); 1822 if (!err) { 1823 generic_fillattr(inode, stat); 1824 stat->ino = ceph_translate_ino(inode->i_sb, inode->i_ino); 1825 if (ceph_snap(inode) != CEPH_NOSNAP) 1826 stat->dev = ceph_snap(inode); 1827 else 1828 stat->dev = 0; 1829 if (S_ISDIR(inode->i_mode)) { 1830 if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), 1831 RBYTES)) 1832 stat->size = ci->i_rbytes; 1833 else 1834 stat->size = ci->i_files + ci->i_subdirs; 1835 stat->blocks = 0; 1836 stat->blksize = 65536; 1837 } 1838 } 1839 return err; 1840 } 1841