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