1 // SPDX-License-Identifier: GPL-2.0 2 #include <linux/ceph/ceph_debug.h> 3 4 #include <linux/spinlock.h> 5 #include <linux/namei.h> 6 #include <linux/slab.h> 7 #include <linux/sched.h> 8 #include <linux/xattr.h> 9 10 #include "super.h" 11 #include "mds_client.h" 12 13 /* 14 * Directory operations: readdir, lookup, create, link, unlink, 15 * rename, etc. 16 */ 17 18 /* 19 * Ceph MDS operations are specified in terms of a base ino and 20 * relative path. Thus, the client can specify an operation on a 21 * specific inode (e.g., a getattr due to fstat(2)), or as a path 22 * relative to, say, the root directory. 23 * 24 * Normally, we limit ourselves to strict inode ops (no path component) 25 * or dentry operations (a single path component relative to an ino). The 26 * exception to this is open_root_dentry(), which will open the mount 27 * point by name. 28 */ 29 30 const struct dentry_operations ceph_dentry_ops; 31 32 static bool __dentry_lease_is_valid(struct ceph_dentry_info *di); 33 static int __dir_lease_try_check(const struct dentry *dentry); 34 35 /* 36 * Initialize ceph dentry state. 37 */ 38 static int ceph_d_init(struct dentry *dentry) 39 { 40 struct ceph_dentry_info *di; 41 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(dentry->d_sb); 42 43 di = kmem_cache_zalloc(ceph_dentry_cachep, GFP_KERNEL); 44 if (!di) 45 return -ENOMEM; /* oh well */ 46 47 di->dentry = dentry; 48 di->lease_session = NULL; 49 di->time = jiffies; 50 dentry->d_fsdata = di; 51 INIT_LIST_HEAD(&di->lease_list); 52 53 atomic64_inc(&mdsc->metric.total_dentries); 54 55 return 0; 56 } 57 58 /* 59 * for f_pos for readdir: 60 * - hash order: 61 * (0xff << 52) | ((24 bits hash) << 28) | 62 * (the nth entry has hash collision); 63 * - frag+name order; 64 * ((frag value) << 28) | (the nth entry in frag); 65 */ 66 #define OFFSET_BITS 28 67 #define OFFSET_MASK ((1 << OFFSET_BITS) - 1) 68 #define HASH_ORDER (0xffull << (OFFSET_BITS + 24)) 69 loff_t ceph_make_fpos(unsigned high, unsigned off, bool hash_order) 70 { 71 loff_t fpos = ((loff_t)high << 28) | (loff_t)off; 72 if (hash_order) 73 fpos |= HASH_ORDER; 74 return fpos; 75 } 76 77 static bool is_hash_order(loff_t p) 78 { 79 return (p & HASH_ORDER) == HASH_ORDER; 80 } 81 82 static unsigned fpos_frag(loff_t p) 83 { 84 return p >> OFFSET_BITS; 85 } 86 87 static unsigned fpos_hash(loff_t p) 88 { 89 return ceph_frag_value(fpos_frag(p)); 90 } 91 92 static unsigned fpos_off(loff_t p) 93 { 94 return p & OFFSET_MASK; 95 } 96 97 static int fpos_cmp(loff_t l, loff_t r) 98 { 99 int v = ceph_frag_compare(fpos_frag(l), fpos_frag(r)); 100 if (v) 101 return v; 102 return (int)(fpos_off(l) - fpos_off(r)); 103 } 104 105 /* 106 * make note of the last dentry we read, so we can 107 * continue at the same lexicographical point, 108 * regardless of what dir changes take place on the 109 * server. 110 */ 111 static int note_last_dentry(struct ceph_dir_file_info *dfi, const char *name, 112 int len, unsigned next_offset) 113 { 114 char *buf = kmalloc(len+1, GFP_KERNEL); 115 if (!buf) 116 return -ENOMEM; 117 kfree(dfi->last_name); 118 dfi->last_name = buf; 119 memcpy(dfi->last_name, name, len); 120 dfi->last_name[len] = 0; 121 dfi->next_offset = next_offset; 122 dout("note_last_dentry '%s'\n", dfi->last_name); 123 return 0; 124 } 125 126 127 static struct dentry * 128 __dcache_find_get_entry(struct dentry *parent, u64 idx, 129 struct ceph_readdir_cache_control *cache_ctl) 130 { 131 struct inode *dir = d_inode(parent); 132 struct dentry *dentry; 133 unsigned idx_mask = (PAGE_SIZE / sizeof(struct dentry *)) - 1; 134 loff_t ptr_pos = idx * sizeof(struct dentry *); 135 pgoff_t ptr_pgoff = ptr_pos >> PAGE_SHIFT; 136 137 if (ptr_pos >= i_size_read(dir)) 138 return NULL; 139 140 if (!cache_ctl->page || ptr_pgoff != page_index(cache_ctl->page)) { 141 ceph_readdir_cache_release(cache_ctl); 142 cache_ctl->page = find_lock_page(&dir->i_data, ptr_pgoff); 143 if (!cache_ctl->page) { 144 dout(" page %lu not found\n", ptr_pgoff); 145 return ERR_PTR(-EAGAIN); 146 } 147 /* reading/filling the cache are serialized by 148 i_rwsem, no need to use page lock */ 149 unlock_page(cache_ctl->page); 150 cache_ctl->dentries = kmap(cache_ctl->page); 151 } 152 153 cache_ctl->index = idx & idx_mask; 154 155 rcu_read_lock(); 156 spin_lock(&parent->d_lock); 157 /* check i_size again here, because empty directory can be 158 * marked as complete while not holding the i_rwsem. */ 159 if (ceph_dir_is_complete_ordered(dir) && ptr_pos < i_size_read(dir)) 160 dentry = cache_ctl->dentries[cache_ctl->index]; 161 else 162 dentry = NULL; 163 spin_unlock(&parent->d_lock); 164 if (dentry && !lockref_get_not_dead(&dentry->d_lockref)) 165 dentry = NULL; 166 rcu_read_unlock(); 167 return dentry ? : ERR_PTR(-EAGAIN); 168 } 169 170 /* 171 * When possible, we try to satisfy a readdir by peeking at the 172 * dcache. We make this work by carefully ordering dentries on 173 * d_child when we initially get results back from the MDS, and 174 * falling back to a "normal" sync readdir if any dentries in the dir 175 * are dropped. 176 * 177 * Complete dir indicates that we have all dentries in the dir. It is 178 * defined IFF we hold CEPH_CAP_FILE_SHARED (which will be revoked by 179 * the MDS if/when the directory is modified). 180 */ 181 static int __dcache_readdir(struct file *file, struct dir_context *ctx, 182 int shared_gen) 183 { 184 struct ceph_dir_file_info *dfi = file->private_data; 185 struct dentry *parent = file->f_path.dentry; 186 struct inode *dir = d_inode(parent); 187 struct dentry *dentry, *last = NULL; 188 struct ceph_dentry_info *di; 189 struct ceph_readdir_cache_control cache_ctl = {}; 190 u64 idx = 0; 191 int err = 0; 192 193 dout("__dcache_readdir %p v%u at %llx\n", dir, (unsigned)shared_gen, ctx->pos); 194 195 /* search start position */ 196 if (ctx->pos > 2) { 197 u64 count = div_u64(i_size_read(dir), sizeof(struct dentry *)); 198 while (count > 0) { 199 u64 step = count >> 1; 200 dentry = __dcache_find_get_entry(parent, idx + step, 201 &cache_ctl); 202 if (!dentry) { 203 /* use linar search */ 204 idx = 0; 205 break; 206 } 207 if (IS_ERR(dentry)) { 208 err = PTR_ERR(dentry); 209 goto out; 210 } 211 di = ceph_dentry(dentry); 212 spin_lock(&dentry->d_lock); 213 if (fpos_cmp(di->offset, ctx->pos) < 0) { 214 idx += step + 1; 215 count -= step + 1; 216 } else { 217 count = step; 218 } 219 spin_unlock(&dentry->d_lock); 220 dput(dentry); 221 } 222 223 dout("__dcache_readdir %p cache idx %llu\n", dir, idx); 224 } 225 226 227 for (;;) { 228 bool emit_dentry = false; 229 dentry = __dcache_find_get_entry(parent, idx++, &cache_ctl); 230 if (!dentry) { 231 dfi->file_info.flags |= CEPH_F_ATEND; 232 err = 0; 233 break; 234 } 235 if (IS_ERR(dentry)) { 236 err = PTR_ERR(dentry); 237 goto out; 238 } 239 240 spin_lock(&dentry->d_lock); 241 di = ceph_dentry(dentry); 242 if (d_unhashed(dentry) || 243 d_really_is_negative(dentry) || 244 di->lease_shared_gen != shared_gen) { 245 spin_unlock(&dentry->d_lock); 246 dput(dentry); 247 err = -EAGAIN; 248 goto out; 249 } 250 if (fpos_cmp(ctx->pos, di->offset) <= 0) { 251 __ceph_dentry_dir_lease_touch(di); 252 emit_dentry = true; 253 } 254 spin_unlock(&dentry->d_lock); 255 256 if (emit_dentry) { 257 dout(" %llx dentry %p %pd %p\n", di->offset, 258 dentry, dentry, d_inode(dentry)); 259 ctx->pos = di->offset; 260 if (!dir_emit(ctx, dentry->d_name.name, 261 dentry->d_name.len, ceph_present_inode(d_inode(dentry)), 262 d_inode(dentry)->i_mode >> 12)) { 263 dput(dentry); 264 err = 0; 265 break; 266 } 267 ctx->pos++; 268 269 if (last) 270 dput(last); 271 last = dentry; 272 } else { 273 dput(dentry); 274 } 275 } 276 out: 277 ceph_readdir_cache_release(&cache_ctl); 278 if (last) { 279 int ret; 280 di = ceph_dentry(last); 281 ret = note_last_dentry(dfi, last->d_name.name, last->d_name.len, 282 fpos_off(di->offset) + 1); 283 if (ret < 0) 284 err = ret; 285 dput(last); 286 /* last_name no longer match cache index */ 287 if (dfi->readdir_cache_idx >= 0) { 288 dfi->readdir_cache_idx = -1; 289 dfi->dir_release_count = 0; 290 } 291 } 292 return err; 293 } 294 295 static bool need_send_readdir(struct ceph_dir_file_info *dfi, loff_t pos) 296 { 297 if (!dfi->last_readdir) 298 return true; 299 if (is_hash_order(pos)) 300 return !ceph_frag_contains_value(dfi->frag, fpos_hash(pos)); 301 else 302 return dfi->frag != fpos_frag(pos); 303 } 304 305 static int ceph_readdir(struct file *file, struct dir_context *ctx) 306 { 307 struct ceph_dir_file_info *dfi = file->private_data; 308 struct inode *inode = file_inode(file); 309 struct ceph_inode_info *ci = ceph_inode(inode); 310 struct ceph_fs_client *fsc = ceph_inode_to_client(inode); 311 struct ceph_mds_client *mdsc = fsc->mdsc; 312 int i; 313 int err; 314 unsigned frag = -1; 315 struct ceph_mds_reply_info_parsed *rinfo; 316 317 dout("readdir %p file %p pos %llx\n", inode, file, ctx->pos); 318 if (dfi->file_info.flags & CEPH_F_ATEND) 319 return 0; 320 321 /* always start with . and .. */ 322 if (ctx->pos == 0) { 323 dout("readdir off 0 -> '.'\n"); 324 if (!dir_emit(ctx, ".", 1, ceph_present_inode(inode), 325 inode->i_mode >> 12)) 326 return 0; 327 ctx->pos = 1; 328 } 329 if (ctx->pos == 1) { 330 u64 ino; 331 struct dentry *dentry = file->f_path.dentry; 332 333 spin_lock(&dentry->d_lock); 334 ino = ceph_present_inode(dentry->d_parent->d_inode); 335 spin_unlock(&dentry->d_lock); 336 337 dout("readdir off 1 -> '..'\n"); 338 if (!dir_emit(ctx, "..", 2, ino, inode->i_mode >> 12)) 339 return 0; 340 ctx->pos = 2; 341 } 342 343 spin_lock(&ci->i_ceph_lock); 344 /* request Fx cap. if have Fx, we don't need to release Fs cap 345 * for later create/unlink. */ 346 __ceph_touch_fmode(ci, mdsc, CEPH_FILE_MODE_WR); 347 /* can we use the dcache? */ 348 if (ceph_test_mount_opt(fsc, DCACHE) && 349 !ceph_test_mount_opt(fsc, NOASYNCREADDIR) && 350 ceph_snap(inode) != CEPH_SNAPDIR && 351 __ceph_dir_is_complete_ordered(ci) && 352 __ceph_caps_issued_mask_metric(ci, CEPH_CAP_FILE_SHARED, 1)) { 353 int shared_gen = atomic_read(&ci->i_shared_gen); 354 355 spin_unlock(&ci->i_ceph_lock); 356 err = __dcache_readdir(file, ctx, shared_gen); 357 if (err != -EAGAIN) 358 return err; 359 } else { 360 spin_unlock(&ci->i_ceph_lock); 361 } 362 363 /* proceed with a normal readdir */ 364 more: 365 /* do we have the correct frag content buffered? */ 366 if (need_send_readdir(dfi, ctx->pos)) { 367 struct ceph_mds_request *req; 368 int op = ceph_snap(inode) == CEPH_SNAPDIR ? 369 CEPH_MDS_OP_LSSNAP : CEPH_MDS_OP_READDIR; 370 371 /* discard old result, if any */ 372 if (dfi->last_readdir) { 373 ceph_mdsc_put_request(dfi->last_readdir); 374 dfi->last_readdir = NULL; 375 } 376 377 if (is_hash_order(ctx->pos)) { 378 /* fragtree isn't always accurate. choose frag 379 * based on previous reply when possible. */ 380 if (frag == (unsigned)-1) 381 frag = ceph_choose_frag(ci, fpos_hash(ctx->pos), 382 NULL, NULL); 383 } else { 384 frag = fpos_frag(ctx->pos); 385 } 386 387 dout("readdir fetching %llx.%llx frag %x offset '%s'\n", 388 ceph_vinop(inode), frag, dfi->last_name); 389 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS); 390 if (IS_ERR(req)) 391 return PTR_ERR(req); 392 err = ceph_alloc_readdir_reply_buffer(req, inode); 393 if (err) { 394 ceph_mdsc_put_request(req); 395 return err; 396 } 397 /* hints to request -> mds selection code */ 398 req->r_direct_mode = USE_AUTH_MDS; 399 if (op == CEPH_MDS_OP_READDIR) { 400 req->r_direct_hash = ceph_frag_value(frag); 401 __set_bit(CEPH_MDS_R_DIRECT_IS_HASH, &req->r_req_flags); 402 req->r_inode_drop = CEPH_CAP_FILE_EXCL; 403 } 404 if (dfi->last_name) { 405 req->r_path2 = kstrdup(dfi->last_name, GFP_KERNEL); 406 if (!req->r_path2) { 407 ceph_mdsc_put_request(req); 408 return -ENOMEM; 409 } 410 } else if (is_hash_order(ctx->pos)) { 411 req->r_args.readdir.offset_hash = 412 cpu_to_le32(fpos_hash(ctx->pos)); 413 } 414 415 req->r_dir_release_cnt = dfi->dir_release_count; 416 req->r_dir_ordered_cnt = dfi->dir_ordered_count; 417 req->r_readdir_cache_idx = dfi->readdir_cache_idx; 418 req->r_readdir_offset = dfi->next_offset; 419 req->r_args.readdir.frag = cpu_to_le32(frag); 420 req->r_args.readdir.flags = 421 cpu_to_le16(CEPH_READDIR_REPLY_BITFLAGS); 422 423 req->r_inode = inode; 424 ihold(inode); 425 req->r_dentry = dget(file->f_path.dentry); 426 err = ceph_mdsc_do_request(mdsc, NULL, req); 427 if (err < 0) { 428 ceph_mdsc_put_request(req); 429 return err; 430 } 431 dout("readdir got and parsed readdir result=%d on " 432 "frag %x, end=%d, complete=%d, hash_order=%d\n", 433 err, frag, 434 (int)req->r_reply_info.dir_end, 435 (int)req->r_reply_info.dir_complete, 436 (int)req->r_reply_info.hash_order); 437 438 rinfo = &req->r_reply_info; 439 if (le32_to_cpu(rinfo->dir_dir->frag) != frag) { 440 frag = le32_to_cpu(rinfo->dir_dir->frag); 441 if (!rinfo->hash_order) { 442 dfi->next_offset = req->r_readdir_offset; 443 /* adjust ctx->pos to beginning of frag */ 444 ctx->pos = ceph_make_fpos(frag, 445 dfi->next_offset, 446 false); 447 } 448 } 449 450 dfi->frag = frag; 451 dfi->last_readdir = req; 452 453 if (test_bit(CEPH_MDS_R_DID_PREPOPULATE, &req->r_req_flags)) { 454 dfi->readdir_cache_idx = req->r_readdir_cache_idx; 455 if (dfi->readdir_cache_idx < 0) { 456 /* preclude from marking dir ordered */ 457 dfi->dir_ordered_count = 0; 458 } else if (ceph_frag_is_leftmost(frag) && 459 dfi->next_offset == 2) { 460 /* note dir version at start of readdir so 461 * we can tell if any dentries get dropped */ 462 dfi->dir_release_count = req->r_dir_release_cnt; 463 dfi->dir_ordered_count = req->r_dir_ordered_cnt; 464 } 465 } else { 466 dout("readdir !did_prepopulate\n"); 467 /* disable readdir cache */ 468 dfi->readdir_cache_idx = -1; 469 /* preclude from marking dir complete */ 470 dfi->dir_release_count = 0; 471 } 472 473 /* note next offset and last dentry name */ 474 if (rinfo->dir_nr > 0) { 475 struct ceph_mds_reply_dir_entry *rde = 476 rinfo->dir_entries + (rinfo->dir_nr-1); 477 unsigned next_offset = req->r_reply_info.dir_end ? 478 2 : (fpos_off(rde->offset) + 1); 479 err = note_last_dentry(dfi, rde->name, rde->name_len, 480 next_offset); 481 if (err) { 482 ceph_mdsc_put_request(dfi->last_readdir); 483 dfi->last_readdir = NULL; 484 return err; 485 } 486 } else if (req->r_reply_info.dir_end) { 487 dfi->next_offset = 2; 488 /* keep last name */ 489 } 490 } 491 492 rinfo = &dfi->last_readdir->r_reply_info; 493 dout("readdir frag %x num %d pos %llx chunk first %llx\n", 494 dfi->frag, rinfo->dir_nr, ctx->pos, 495 rinfo->dir_nr ? rinfo->dir_entries[0].offset : 0LL); 496 497 i = 0; 498 /* search start position */ 499 if (rinfo->dir_nr > 0) { 500 int step, nr = rinfo->dir_nr; 501 while (nr > 0) { 502 step = nr >> 1; 503 if (rinfo->dir_entries[i + step].offset < ctx->pos) { 504 i += step + 1; 505 nr -= step + 1; 506 } else { 507 nr = step; 508 } 509 } 510 } 511 for (; i < rinfo->dir_nr; i++) { 512 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i; 513 514 BUG_ON(rde->offset < ctx->pos); 515 516 ctx->pos = rde->offset; 517 dout("readdir (%d/%d) -> %llx '%.*s' %p\n", 518 i, rinfo->dir_nr, ctx->pos, 519 rde->name_len, rde->name, &rde->inode.in); 520 521 BUG_ON(!rde->inode.in); 522 523 if (!dir_emit(ctx, rde->name, rde->name_len, 524 ceph_present_ino(inode->i_sb, le64_to_cpu(rde->inode.in->ino)), 525 le32_to_cpu(rde->inode.in->mode) >> 12)) { 526 /* 527 * NOTE: Here no need to put the 'dfi->last_readdir', 528 * because when dir_emit stops us it's most likely 529 * doesn't have enough memory, etc. So for next readdir 530 * it will continue. 531 */ 532 dout("filldir stopping us...\n"); 533 return 0; 534 } 535 ctx->pos++; 536 } 537 538 ceph_mdsc_put_request(dfi->last_readdir); 539 dfi->last_readdir = NULL; 540 541 if (dfi->next_offset > 2) { 542 frag = dfi->frag; 543 goto more; 544 } 545 546 /* more frags? */ 547 if (!ceph_frag_is_rightmost(dfi->frag)) { 548 frag = ceph_frag_next(dfi->frag); 549 if (is_hash_order(ctx->pos)) { 550 loff_t new_pos = ceph_make_fpos(ceph_frag_value(frag), 551 dfi->next_offset, true); 552 if (new_pos > ctx->pos) 553 ctx->pos = new_pos; 554 /* keep last_name */ 555 } else { 556 ctx->pos = ceph_make_fpos(frag, dfi->next_offset, 557 false); 558 kfree(dfi->last_name); 559 dfi->last_name = NULL; 560 } 561 dout("readdir next frag is %x\n", frag); 562 goto more; 563 } 564 dfi->file_info.flags |= CEPH_F_ATEND; 565 566 /* 567 * if dir_release_count still matches the dir, no dentries 568 * were released during the whole readdir, and we should have 569 * the complete dir contents in our cache. 570 */ 571 if (atomic64_read(&ci->i_release_count) == 572 dfi->dir_release_count) { 573 spin_lock(&ci->i_ceph_lock); 574 if (dfi->dir_ordered_count == 575 atomic64_read(&ci->i_ordered_count)) { 576 dout(" marking %p complete and ordered\n", inode); 577 /* use i_size to track number of entries in 578 * readdir cache */ 579 BUG_ON(dfi->readdir_cache_idx < 0); 580 i_size_write(inode, dfi->readdir_cache_idx * 581 sizeof(struct dentry*)); 582 } else { 583 dout(" marking %p complete\n", inode); 584 } 585 __ceph_dir_set_complete(ci, dfi->dir_release_count, 586 dfi->dir_ordered_count); 587 spin_unlock(&ci->i_ceph_lock); 588 } 589 590 dout("readdir %p file %p done.\n", inode, file); 591 return 0; 592 } 593 594 static void reset_readdir(struct ceph_dir_file_info *dfi) 595 { 596 if (dfi->last_readdir) { 597 ceph_mdsc_put_request(dfi->last_readdir); 598 dfi->last_readdir = NULL; 599 } 600 kfree(dfi->last_name); 601 dfi->last_name = NULL; 602 dfi->dir_release_count = 0; 603 dfi->readdir_cache_idx = -1; 604 dfi->next_offset = 2; /* compensate for . and .. */ 605 dfi->file_info.flags &= ~CEPH_F_ATEND; 606 } 607 608 /* 609 * discard buffered readdir content on seekdir(0), or seek to new frag, 610 * or seek prior to current chunk 611 */ 612 static bool need_reset_readdir(struct ceph_dir_file_info *dfi, loff_t new_pos) 613 { 614 struct ceph_mds_reply_info_parsed *rinfo; 615 loff_t chunk_offset; 616 if (new_pos == 0) 617 return true; 618 if (is_hash_order(new_pos)) { 619 /* no need to reset last_name for a forward seek when 620 * dentries are sotred in hash order */ 621 } else if (dfi->frag != fpos_frag(new_pos)) { 622 return true; 623 } 624 rinfo = dfi->last_readdir ? &dfi->last_readdir->r_reply_info : NULL; 625 if (!rinfo || !rinfo->dir_nr) 626 return true; 627 chunk_offset = rinfo->dir_entries[0].offset; 628 return new_pos < chunk_offset || 629 is_hash_order(new_pos) != is_hash_order(chunk_offset); 630 } 631 632 static loff_t ceph_dir_llseek(struct file *file, loff_t offset, int whence) 633 { 634 struct ceph_dir_file_info *dfi = file->private_data; 635 struct inode *inode = file->f_mapping->host; 636 loff_t retval; 637 638 inode_lock(inode); 639 retval = -EINVAL; 640 switch (whence) { 641 case SEEK_CUR: 642 offset += file->f_pos; 643 break; 644 case SEEK_SET: 645 break; 646 case SEEK_END: 647 retval = -EOPNOTSUPP; 648 goto out; 649 default: 650 goto out; 651 } 652 653 if (offset >= 0) { 654 if (need_reset_readdir(dfi, offset)) { 655 dout("dir_llseek dropping %p content\n", file); 656 reset_readdir(dfi); 657 } else if (is_hash_order(offset) && offset > file->f_pos) { 658 /* for hash offset, we don't know if a forward seek 659 * is within same frag */ 660 dfi->dir_release_count = 0; 661 dfi->readdir_cache_idx = -1; 662 } 663 664 if (offset != file->f_pos) { 665 file->f_pos = offset; 666 file->f_version = 0; 667 dfi->file_info.flags &= ~CEPH_F_ATEND; 668 } 669 retval = offset; 670 } 671 out: 672 inode_unlock(inode); 673 return retval; 674 } 675 676 /* 677 * Handle lookups for the hidden .snap directory. 678 */ 679 struct dentry *ceph_handle_snapdir(struct ceph_mds_request *req, 680 struct dentry *dentry) 681 { 682 struct ceph_fs_client *fsc = ceph_sb_to_client(dentry->d_sb); 683 struct inode *parent = d_inode(dentry->d_parent); /* we hold i_rwsem */ 684 685 /* .snap dir? */ 686 if (ceph_snap(parent) == CEPH_NOSNAP && 687 strcmp(dentry->d_name.name, fsc->mount_options->snapdir_name) == 0) { 688 struct dentry *res; 689 struct inode *inode = ceph_get_snapdir(parent); 690 691 res = d_splice_alias(inode, dentry); 692 dout("ENOENT on snapdir %p '%pd', linking to snapdir %p. Spliced dentry %p\n", 693 dentry, dentry, inode, res); 694 if (res) 695 dentry = res; 696 } 697 return dentry; 698 } 699 700 /* 701 * Figure out final result of a lookup/open request. 702 * 703 * Mainly, make sure we return the final req->r_dentry (if it already 704 * existed) in place of the original VFS-provided dentry when they 705 * differ. 706 * 707 * Gracefully handle the case where the MDS replies with -ENOENT and 708 * no trace (which it may do, at its discretion, e.g., if it doesn't 709 * care to issue a lease on the negative dentry). 710 */ 711 struct dentry *ceph_finish_lookup(struct ceph_mds_request *req, 712 struct dentry *dentry, int err) 713 { 714 if (err == -ENOENT) { 715 /* no trace? */ 716 err = 0; 717 if (!req->r_reply_info.head->is_dentry) { 718 dout("ENOENT and no trace, dentry %p inode %p\n", 719 dentry, d_inode(dentry)); 720 if (d_really_is_positive(dentry)) { 721 d_drop(dentry); 722 err = -ENOENT; 723 } else { 724 d_add(dentry, NULL); 725 } 726 } 727 } 728 if (err) 729 dentry = ERR_PTR(err); 730 else if (dentry != req->r_dentry) 731 dentry = dget(req->r_dentry); /* we got spliced */ 732 else 733 dentry = NULL; 734 return dentry; 735 } 736 737 static bool is_root_ceph_dentry(struct inode *inode, struct dentry *dentry) 738 { 739 return ceph_ino(inode) == CEPH_INO_ROOT && 740 strncmp(dentry->d_name.name, ".ceph", 5) == 0; 741 } 742 743 /* 744 * Look up a single dir entry. If there is a lookup intent, inform 745 * the MDS so that it gets our 'caps wanted' value in a single op. 746 */ 747 static struct dentry *ceph_lookup(struct inode *dir, struct dentry *dentry, 748 unsigned int flags) 749 { 750 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb); 751 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(dir->i_sb); 752 struct ceph_mds_request *req; 753 int op; 754 int mask; 755 int err; 756 757 dout("lookup %p dentry %p '%pd'\n", 758 dir, dentry, dentry); 759 760 if (dentry->d_name.len > NAME_MAX) 761 return ERR_PTR(-ENAMETOOLONG); 762 763 if (IS_ENCRYPTED(dir)) { 764 err = __fscrypt_prepare_readdir(dir); 765 if (err) 766 return ERR_PTR(err); 767 if (!fscrypt_has_encryption_key(dir)) { 768 spin_lock(&dentry->d_lock); 769 dentry->d_flags |= DCACHE_NOKEY_NAME; 770 spin_unlock(&dentry->d_lock); 771 } 772 } 773 774 /* can we conclude ENOENT locally? */ 775 if (d_really_is_negative(dentry)) { 776 struct ceph_inode_info *ci = ceph_inode(dir); 777 struct ceph_dentry_info *di = ceph_dentry(dentry); 778 779 spin_lock(&ci->i_ceph_lock); 780 dout(" dir %p flags are 0x%lx\n", dir, ci->i_ceph_flags); 781 if (strncmp(dentry->d_name.name, 782 fsc->mount_options->snapdir_name, 783 dentry->d_name.len) && 784 !is_root_ceph_dentry(dir, dentry) && 785 ceph_test_mount_opt(fsc, DCACHE) && 786 __ceph_dir_is_complete(ci) && 787 __ceph_caps_issued_mask_metric(ci, CEPH_CAP_FILE_SHARED, 1)) { 788 __ceph_touch_fmode(ci, mdsc, CEPH_FILE_MODE_RD); 789 spin_unlock(&ci->i_ceph_lock); 790 dout(" dir %p complete, -ENOENT\n", dir); 791 d_add(dentry, NULL); 792 di->lease_shared_gen = atomic_read(&ci->i_shared_gen); 793 return NULL; 794 } 795 spin_unlock(&ci->i_ceph_lock); 796 } 797 798 op = ceph_snap(dir) == CEPH_SNAPDIR ? 799 CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP; 800 req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS); 801 if (IS_ERR(req)) 802 return ERR_CAST(req); 803 req->r_dentry = dget(dentry); 804 req->r_num_caps = 2; 805 806 mask = CEPH_STAT_CAP_INODE | CEPH_CAP_AUTH_SHARED; 807 if (ceph_security_xattr_wanted(dir)) 808 mask |= CEPH_CAP_XATTR_SHARED; 809 req->r_args.getattr.mask = cpu_to_le32(mask); 810 811 ihold(dir); 812 req->r_parent = dir; 813 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags); 814 err = ceph_mdsc_do_request(mdsc, NULL, req); 815 if (err == -ENOENT) { 816 struct dentry *res; 817 818 res = ceph_handle_snapdir(req, dentry); 819 if (IS_ERR(res)) { 820 err = PTR_ERR(res); 821 } else { 822 dentry = res; 823 err = 0; 824 } 825 } 826 dentry = ceph_finish_lookup(req, dentry, err); 827 ceph_mdsc_put_request(req); /* will dput(dentry) */ 828 dout("lookup result=%p\n", dentry); 829 return dentry; 830 } 831 832 /* 833 * If we do a create but get no trace back from the MDS, follow up with 834 * a lookup (the VFS expects us to link up the provided dentry). 835 */ 836 int ceph_handle_notrace_create(struct inode *dir, struct dentry *dentry) 837 { 838 struct dentry *result = ceph_lookup(dir, dentry, 0); 839 840 if (result && !IS_ERR(result)) { 841 /* 842 * We created the item, then did a lookup, and found 843 * it was already linked to another inode we already 844 * had in our cache (and thus got spliced). To not 845 * confuse VFS (especially when inode is a directory), 846 * we don't link our dentry to that inode, return an 847 * error instead. 848 * 849 * This event should be rare and it happens only when 850 * we talk to old MDS. Recent MDS does not send traceless 851 * reply for request that creates new inode. 852 */ 853 d_drop(result); 854 return -ESTALE; 855 } 856 return PTR_ERR(result); 857 } 858 859 static int ceph_mknod(struct mnt_idmap *idmap, struct inode *dir, 860 struct dentry *dentry, umode_t mode, dev_t rdev) 861 { 862 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(dir->i_sb); 863 struct ceph_mds_request *req; 864 struct ceph_acl_sec_ctx as_ctx = {}; 865 int err; 866 867 if (ceph_snap(dir) != CEPH_NOSNAP) 868 return -EROFS; 869 870 err = ceph_wait_on_conflict_unlink(dentry); 871 if (err) 872 return err; 873 874 if (ceph_quota_is_max_files_exceeded(dir)) { 875 err = -EDQUOT; 876 goto out; 877 } 878 879 dout("mknod in dir %p dentry %p mode 0%ho rdev %d\n", 880 dir, dentry, mode, rdev); 881 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_MKNOD, USE_AUTH_MDS); 882 if (IS_ERR(req)) { 883 err = PTR_ERR(req); 884 goto out; 885 } 886 887 req->r_new_inode = ceph_new_inode(dir, dentry, &mode, &as_ctx); 888 if (IS_ERR(req->r_new_inode)) { 889 err = PTR_ERR(req->r_new_inode); 890 req->r_new_inode = NULL; 891 goto out_req; 892 } 893 894 req->r_dentry = dget(dentry); 895 req->r_num_caps = 2; 896 req->r_parent = dir; 897 ihold(dir); 898 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags); 899 req->r_args.mknod.mode = cpu_to_le32(mode); 900 req->r_args.mknod.rdev = cpu_to_le32(rdev); 901 req->r_dentry_drop = CEPH_CAP_FILE_SHARED | CEPH_CAP_AUTH_EXCL | 902 CEPH_CAP_XATTR_EXCL; 903 req->r_dentry_unless = CEPH_CAP_FILE_EXCL; 904 905 ceph_as_ctx_to_req(req, &as_ctx); 906 907 err = ceph_mdsc_do_request(mdsc, dir, req); 908 if (!err && !req->r_reply_info.head->is_dentry) 909 err = ceph_handle_notrace_create(dir, dentry); 910 out_req: 911 ceph_mdsc_put_request(req); 912 out: 913 if (!err) 914 ceph_init_inode_acls(d_inode(dentry), &as_ctx); 915 else 916 d_drop(dentry); 917 ceph_release_acl_sec_ctx(&as_ctx); 918 return err; 919 } 920 921 static int ceph_create(struct mnt_idmap *idmap, struct inode *dir, 922 struct dentry *dentry, umode_t mode, bool excl) 923 { 924 return ceph_mknod(idmap, dir, dentry, mode, 0); 925 } 926 927 static int ceph_symlink(struct mnt_idmap *idmap, struct inode *dir, 928 struct dentry *dentry, const char *dest) 929 { 930 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(dir->i_sb); 931 struct ceph_mds_request *req; 932 struct ceph_acl_sec_ctx as_ctx = {}; 933 umode_t mode = S_IFLNK | 0777; 934 int err; 935 936 if (ceph_snap(dir) != CEPH_NOSNAP) 937 return -EROFS; 938 939 err = ceph_wait_on_conflict_unlink(dentry); 940 if (err) 941 return err; 942 943 if (ceph_quota_is_max_files_exceeded(dir)) { 944 err = -EDQUOT; 945 goto out; 946 } 947 948 dout("symlink in dir %p dentry %p to '%s'\n", dir, dentry, dest); 949 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SYMLINK, USE_AUTH_MDS); 950 if (IS_ERR(req)) { 951 err = PTR_ERR(req); 952 goto out; 953 } 954 955 req->r_new_inode = ceph_new_inode(dir, dentry, &mode, &as_ctx); 956 if (IS_ERR(req->r_new_inode)) { 957 err = PTR_ERR(req->r_new_inode); 958 req->r_new_inode = NULL; 959 goto out_req; 960 } 961 962 req->r_path2 = kstrdup(dest, GFP_KERNEL); 963 if (!req->r_path2) { 964 err = -ENOMEM; 965 goto out_req; 966 } 967 req->r_parent = dir; 968 ihold(dir); 969 970 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags); 971 req->r_dentry = dget(dentry); 972 req->r_num_caps = 2; 973 req->r_dentry_drop = CEPH_CAP_FILE_SHARED | CEPH_CAP_AUTH_EXCL | 974 CEPH_CAP_XATTR_EXCL; 975 req->r_dentry_unless = CEPH_CAP_FILE_EXCL; 976 977 ceph_as_ctx_to_req(req, &as_ctx); 978 979 err = ceph_mdsc_do_request(mdsc, dir, req); 980 if (!err && !req->r_reply_info.head->is_dentry) 981 err = ceph_handle_notrace_create(dir, dentry); 982 out_req: 983 ceph_mdsc_put_request(req); 984 out: 985 if (err) 986 d_drop(dentry); 987 ceph_release_acl_sec_ctx(&as_ctx); 988 return err; 989 } 990 991 static int ceph_mkdir(struct mnt_idmap *idmap, struct inode *dir, 992 struct dentry *dentry, umode_t mode) 993 { 994 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(dir->i_sb); 995 struct ceph_mds_request *req; 996 struct ceph_acl_sec_ctx as_ctx = {}; 997 int err; 998 int op; 999 1000 err = ceph_wait_on_conflict_unlink(dentry); 1001 if (err) 1002 return err; 1003 1004 if (ceph_snap(dir) == CEPH_SNAPDIR) { 1005 /* mkdir .snap/foo is a MKSNAP */ 1006 op = CEPH_MDS_OP_MKSNAP; 1007 dout("mksnap dir %p snap '%pd' dn %p\n", dir, 1008 dentry, dentry); 1009 } else if (ceph_snap(dir) == CEPH_NOSNAP) { 1010 dout("mkdir dir %p dn %p mode 0%ho\n", dir, dentry, mode); 1011 op = CEPH_MDS_OP_MKDIR; 1012 } else { 1013 err = -EROFS; 1014 goto out; 1015 } 1016 1017 if (op == CEPH_MDS_OP_MKDIR && 1018 ceph_quota_is_max_files_exceeded(dir)) { 1019 err = -EDQUOT; 1020 goto out; 1021 } 1022 1023 1024 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS); 1025 if (IS_ERR(req)) { 1026 err = PTR_ERR(req); 1027 goto out; 1028 } 1029 1030 mode |= S_IFDIR; 1031 req->r_new_inode = ceph_new_inode(dir, dentry, &mode, &as_ctx); 1032 if (IS_ERR(req->r_new_inode)) { 1033 err = PTR_ERR(req->r_new_inode); 1034 req->r_new_inode = NULL; 1035 goto out_req; 1036 } 1037 1038 req->r_dentry = dget(dentry); 1039 req->r_num_caps = 2; 1040 req->r_parent = dir; 1041 ihold(dir); 1042 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags); 1043 req->r_args.mkdir.mode = cpu_to_le32(mode); 1044 req->r_dentry_drop = CEPH_CAP_FILE_SHARED | CEPH_CAP_AUTH_EXCL | 1045 CEPH_CAP_XATTR_EXCL; 1046 req->r_dentry_unless = CEPH_CAP_FILE_EXCL; 1047 1048 ceph_as_ctx_to_req(req, &as_ctx); 1049 1050 err = ceph_mdsc_do_request(mdsc, dir, req); 1051 if (!err && 1052 !req->r_reply_info.head->is_target && 1053 !req->r_reply_info.head->is_dentry) 1054 err = ceph_handle_notrace_create(dir, dentry); 1055 out_req: 1056 ceph_mdsc_put_request(req); 1057 out: 1058 if (!err) 1059 ceph_init_inode_acls(d_inode(dentry), &as_ctx); 1060 else 1061 d_drop(dentry); 1062 ceph_release_acl_sec_ctx(&as_ctx); 1063 return err; 1064 } 1065 1066 static int ceph_link(struct dentry *old_dentry, struct inode *dir, 1067 struct dentry *dentry) 1068 { 1069 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(dir->i_sb); 1070 struct ceph_mds_request *req; 1071 int err; 1072 1073 if (dentry->d_flags & DCACHE_DISCONNECTED) 1074 return -EINVAL; 1075 1076 err = ceph_wait_on_conflict_unlink(dentry); 1077 if (err) 1078 return err; 1079 1080 if (ceph_snap(dir) != CEPH_NOSNAP) 1081 return -EROFS; 1082 1083 dout("link in dir %p %llx.%llx old_dentry %p:'%pd' dentry %p:'%pd'\n", 1084 dir, ceph_vinop(dir), old_dentry, old_dentry, dentry, dentry); 1085 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LINK, USE_AUTH_MDS); 1086 if (IS_ERR(req)) { 1087 d_drop(dentry); 1088 return PTR_ERR(req); 1089 } 1090 req->r_dentry = dget(dentry); 1091 req->r_num_caps = 2; 1092 req->r_old_dentry = dget(old_dentry); 1093 /* 1094 * The old_dentry maybe a DCACHE_DISCONNECTED dentry, then we 1095 * will just pass the ino# to MDSs. 1096 */ 1097 if (old_dentry->d_flags & DCACHE_DISCONNECTED) 1098 req->r_ino2 = ceph_vino(d_inode(old_dentry)); 1099 req->r_parent = dir; 1100 ihold(dir); 1101 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags); 1102 req->r_dentry_drop = CEPH_CAP_FILE_SHARED | CEPH_CAP_XATTR_EXCL; 1103 req->r_dentry_unless = CEPH_CAP_FILE_EXCL; 1104 /* release LINK_SHARED on source inode (mds will lock it) */ 1105 req->r_old_inode_drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL; 1106 err = ceph_mdsc_do_request(mdsc, dir, req); 1107 if (err) { 1108 d_drop(dentry); 1109 } else if (!req->r_reply_info.head->is_dentry) { 1110 ihold(d_inode(old_dentry)); 1111 d_instantiate(dentry, d_inode(old_dentry)); 1112 } 1113 ceph_mdsc_put_request(req); 1114 return err; 1115 } 1116 1117 static void ceph_async_unlink_cb(struct ceph_mds_client *mdsc, 1118 struct ceph_mds_request *req) 1119 { 1120 struct dentry *dentry = req->r_dentry; 1121 struct ceph_fs_client *fsc = ceph_sb_to_client(dentry->d_sb); 1122 struct ceph_dentry_info *di = ceph_dentry(dentry); 1123 int result = req->r_err ? req->r_err : 1124 le32_to_cpu(req->r_reply_info.head->result); 1125 1126 if (!test_bit(CEPH_DENTRY_ASYNC_UNLINK_BIT, &di->flags)) 1127 pr_warn("%s dentry %p:%pd async unlink bit is not set\n", 1128 __func__, dentry, dentry); 1129 1130 spin_lock(&fsc->async_unlink_conflict_lock); 1131 hash_del_rcu(&di->hnode); 1132 spin_unlock(&fsc->async_unlink_conflict_lock); 1133 1134 spin_lock(&dentry->d_lock); 1135 di->flags &= ~CEPH_DENTRY_ASYNC_UNLINK; 1136 wake_up_bit(&di->flags, CEPH_DENTRY_ASYNC_UNLINK_BIT); 1137 spin_unlock(&dentry->d_lock); 1138 1139 synchronize_rcu(); 1140 1141 if (result == -EJUKEBOX) 1142 goto out; 1143 1144 /* If op failed, mark everyone involved for errors */ 1145 if (result) { 1146 int pathlen = 0; 1147 u64 base = 0; 1148 char *path = ceph_mdsc_build_path(dentry, &pathlen, 1149 &base, 0); 1150 1151 /* mark error on parent + clear complete */ 1152 mapping_set_error(req->r_parent->i_mapping, result); 1153 ceph_dir_clear_complete(req->r_parent); 1154 1155 /* drop the dentry -- we don't know its status */ 1156 if (!d_unhashed(dentry)) 1157 d_drop(dentry); 1158 1159 /* mark inode itself for an error (since metadata is bogus) */ 1160 mapping_set_error(req->r_old_inode->i_mapping, result); 1161 1162 pr_warn("async unlink failure path=(%llx)%s result=%d!\n", 1163 base, IS_ERR(path) ? "<<bad>>" : path, result); 1164 ceph_mdsc_free_path(path, pathlen); 1165 } 1166 out: 1167 iput(req->r_old_inode); 1168 ceph_mdsc_release_dir_caps(req); 1169 } 1170 1171 static int get_caps_for_async_unlink(struct inode *dir, struct dentry *dentry) 1172 { 1173 struct ceph_inode_info *ci = ceph_inode(dir); 1174 struct ceph_dentry_info *di; 1175 int got = 0, want = CEPH_CAP_FILE_EXCL | CEPH_CAP_DIR_UNLINK; 1176 1177 spin_lock(&ci->i_ceph_lock); 1178 if ((__ceph_caps_issued(ci, NULL) & want) == want) { 1179 ceph_take_cap_refs(ci, want, false); 1180 got = want; 1181 } 1182 spin_unlock(&ci->i_ceph_lock); 1183 1184 /* If we didn't get anything, return 0 */ 1185 if (!got) 1186 return 0; 1187 1188 spin_lock(&dentry->d_lock); 1189 di = ceph_dentry(dentry); 1190 /* 1191 * - We are holding Fx, which implies Fs caps. 1192 * - Only support async unlink for primary linkage 1193 */ 1194 if (atomic_read(&ci->i_shared_gen) != di->lease_shared_gen || 1195 !(di->flags & CEPH_DENTRY_PRIMARY_LINK)) 1196 want = 0; 1197 spin_unlock(&dentry->d_lock); 1198 1199 /* Do we still want what we've got? */ 1200 if (want == got) 1201 return got; 1202 1203 ceph_put_cap_refs(ci, got); 1204 return 0; 1205 } 1206 1207 /* 1208 * rmdir and unlink are differ only by the metadata op code 1209 */ 1210 static int ceph_unlink(struct inode *dir, struct dentry *dentry) 1211 { 1212 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb); 1213 struct ceph_mds_client *mdsc = fsc->mdsc; 1214 struct inode *inode = d_inode(dentry); 1215 struct ceph_mds_request *req; 1216 bool try_async = ceph_test_mount_opt(fsc, ASYNC_DIROPS); 1217 int err = -EROFS; 1218 int op; 1219 1220 if (ceph_snap(dir) == CEPH_SNAPDIR) { 1221 /* rmdir .snap/foo is RMSNAP */ 1222 dout("rmsnap dir %p '%pd' dn %p\n", dir, dentry, dentry); 1223 op = CEPH_MDS_OP_RMSNAP; 1224 } else if (ceph_snap(dir) == CEPH_NOSNAP) { 1225 dout("unlink/rmdir dir %p dn %p inode %p\n", 1226 dir, dentry, inode); 1227 op = d_is_dir(dentry) ? 1228 CEPH_MDS_OP_RMDIR : CEPH_MDS_OP_UNLINK; 1229 } else 1230 goto out; 1231 retry: 1232 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS); 1233 if (IS_ERR(req)) { 1234 err = PTR_ERR(req); 1235 goto out; 1236 } 1237 req->r_dentry = dget(dentry); 1238 req->r_num_caps = 2; 1239 req->r_parent = dir; 1240 ihold(dir); 1241 req->r_dentry_drop = CEPH_CAP_FILE_SHARED | CEPH_CAP_XATTR_EXCL; 1242 req->r_dentry_unless = CEPH_CAP_FILE_EXCL; 1243 req->r_inode_drop = ceph_drop_caps_for_unlink(inode); 1244 1245 if (try_async && op == CEPH_MDS_OP_UNLINK && 1246 (req->r_dir_caps = get_caps_for_async_unlink(dir, dentry))) { 1247 struct ceph_dentry_info *di = ceph_dentry(dentry); 1248 1249 dout("async unlink on %llu/%.*s caps=%s", ceph_ino(dir), 1250 dentry->d_name.len, dentry->d_name.name, 1251 ceph_cap_string(req->r_dir_caps)); 1252 set_bit(CEPH_MDS_R_ASYNC, &req->r_req_flags); 1253 req->r_callback = ceph_async_unlink_cb; 1254 req->r_old_inode = d_inode(dentry); 1255 ihold(req->r_old_inode); 1256 1257 spin_lock(&dentry->d_lock); 1258 di->flags |= CEPH_DENTRY_ASYNC_UNLINK; 1259 spin_unlock(&dentry->d_lock); 1260 1261 spin_lock(&fsc->async_unlink_conflict_lock); 1262 hash_add_rcu(fsc->async_unlink_conflict, &di->hnode, 1263 dentry->d_name.hash); 1264 spin_unlock(&fsc->async_unlink_conflict_lock); 1265 1266 err = ceph_mdsc_submit_request(mdsc, dir, req); 1267 if (!err) { 1268 /* 1269 * We have enough caps, so we assume that the unlink 1270 * will succeed. Fix up the target inode and dcache. 1271 */ 1272 drop_nlink(inode); 1273 d_delete(dentry); 1274 } else { 1275 spin_lock(&fsc->async_unlink_conflict_lock); 1276 hash_del_rcu(&di->hnode); 1277 spin_unlock(&fsc->async_unlink_conflict_lock); 1278 1279 spin_lock(&dentry->d_lock); 1280 di->flags &= ~CEPH_DENTRY_ASYNC_UNLINK; 1281 spin_unlock(&dentry->d_lock); 1282 1283 if (err == -EJUKEBOX) { 1284 try_async = false; 1285 ceph_mdsc_put_request(req); 1286 goto retry; 1287 } 1288 } 1289 } else { 1290 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags); 1291 err = ceph_mdsc_do_request(mdsc, dir, req); 1292 if (!err && !req->r_reply_info.head->is_dentry) 1293 d_delete(dentry); 1294 } 1295 1296 ceph_mdsc_put_request(req); 1297 out: 1298 return err; 1299 } 1300 1301 static int ceph_rename(struct mnt_idmap *idmap, struct inode *old_dir, 1302 struct dentry *old_dentry, struct inode *new_dir, 1303 struct dentry *new_dentry, unsigned int flags) 1304 { 1305 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(old_dir->i_sb); 1306 struct ceph_mds_request *req; 1307 int op = CEPH_MDS_OP_RENAME; 1308 int err; 1309 1310 if (flags) 1311 return -EINVAL; 1312 1313 if (ceph_snap(old_dir) != ceph_snap(new_dir)) 1314 return -EXDEV; 1315 if (ceph_snap(old_dir) != CEPH_NOSNAP) { 1316 if (old_dir == new_dir && ceph_snap(old_dir) == CEPH_SNAPDIR) 1317 op = CEPH_MDS_OP_RENAMESNAP; 1318 else 1319 return -EROFS; 1320 } 1321 /* don't allow cross-quota renames */ 1322 if ((old_dir != new_dir) && 1323 (!ceph_quota_is_same_realm(old_dir, new_dir))) 1324 return -EXDEV; 1325 1326 err = ceph_wait_on_conflict_unlink(new_dentry); 1327 if (err) 1328 return err; 1329 1330 dout("rename dir %p dentry %p to dir %p dentry %p\n", 1331 old_dir, old_dentry, new_dir, new_dentry); 1332 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS); 1333 if (IS_ERR(req)) 1334 return PTR_ERR(req); 1335 ihold(old_dir); 1336 req->r_dentry = dget(new_dentry); 1337 req->r_num_caps = 2; 1338 req->r_old_dentry = dget(old_dentry); 1339 req->r_old_dentry_dir = old_dir; 1340 req->r_parent = new_dir; 1341 ihold(new_dir); 1342 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags); 1343 req->r_old_dentry_drop = CEPH_CAP_FILE_SHARED | CEPH_CAP_XATTR_EXCL; 1344 req->r_old_dentry_unless = CEPH_CAP_FILE_EXCL; 1345 req->r_dentry_drop = CEPH_CAP_FILE_SHARED | CEPH_CAP_XATTR_EXCL; 1346 req->r_dentry_unless = CEPH_CAP_FILE_EXCL; 1347 /* release LINK_RDCACHE on source inode (mds will lock it) */ 1348 req->r_old_inode_drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL; 1349 if (d_really_is_positive(new_dentry)) { 1350 req->r_inode_drop = 1351 ceph_drop_caps_for_unlink(d_inode(new_dentry)); 1352 } 1353 err = ceph_mdsc_do_request(mdsc, old_dir, req); 1354 if (!err && !req->r_reply_info.head->is_dentry) { 1355 /* 1356 * Normally d_move() is done by fill_trace (called by 1357 * do_request, above). If there is no trace, we need 1358 * to do it here. 1359 */ 1360 d_move(old_dentry, new_dentry); 1361 } 1362 ceph_mdsc_put_request(req); 1363 return err; 1364 } 1365 1366 /* 1367 * Move dentry to tail of mdsc->dentry_leases list when lease is updated. 1368 * Leases at front of the list will expire first. (Assume all leases have 1369 * similar duration) 1370 * 1371 * Called under dentry->d_lock. 1372 */ 1373 void __ceph_dentry_lease_touch(struct ceph_dentry_info *di) 1374 { 1375 struct dentry *dn = di->dentry; 1376 struct ceph_mds_client *mdsc; 1377 1378 dout("dentry_lease_touch %p %p '%pd'\n", di, dn, dn); 1379 1380 di->flags |= CEPH_DENTRY_LEASE_LIST; 1381 if (di->flags & CEPH_DENTRY_SHRINK_LIST) { 1382 di->flags |= CEPH_DENTRY_REFERENCED; 1383 return; 1384 } 1385 1386 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc; 1387 spin_lock(&mdsc->dentry_list_lock); 1388 list_move_tail(&di->lease_list, &mdsc->dentry_leases); 1389 spin_unlock(&mdsc->dentry_list_lock); 1390 } 1391 1392 static void __dentry_dir_lease_touch(struct ceph_mds_client* mdsc, 1393 struct ceph_dentry_info *di) 1394 { 1395 di->flags &= ~(CEPH_DENTRY_LEASE_LIST | CEPH_DENTRY_REFERENCED); 1396 di->lease_gen = 0; 1397 di->time = jiffies; 1398 list_move_tail(&di->lease_list, &mdsc->dentry_dir_leases); 1399 } 1400 1401 /* 1402 * When dir lease is used, add dentry to tail of mdsc->dentry_dir_leases 1403 * list if it's not in the list, otherwise set 'referenced' flag. 1404 * 1405 * Called under dentry->d_lock. 1406 */ 1407 void __ceph_dentry_dir_lease_touch(struct ceph_dentry_info *di) 1408 { 1409 struct dentry *dn = di->dentry; 1410 struct ceph_mds_client *mdsc; 1411 1412 dout("dentry_dir_lease_touch %p %p '%pd' (offset 0x%llx)\n", 1413 di, dn, dn, di->offset); 1414 1415 if (!list_empty(&di->lease_list)) { 1416 if (di->flags & CEPH_DENTRY_LEASE_LIST) { 1417 /* don't remove dentry from dentry lease list 1418 * if its lease is valid */ 1419 if (__dentry_lease_is_valid(di)) 1420 return; 1421 } else { 1422 di->flags |= CEPH_DENTRY_REFERENCED; 1423 return; 1424 } 1425 } 1426 1427 if (di->flags & CEPH_DENTRY_SHRINK_LIST) { 1428 di->flags |= CEPH_DENTRY_REFERENCED; 1429 di->flags &= ~CEPH_DENTRY_LEASE_LIST; 1430 return; 1431 } 1432 1433 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc; 1434 spin_lock(&mdsc->dentry_list_lock); 1435 __dentry_dir_lease_touch(mdsc, di), 1436 spin_unlock(&mdsc->dentry_list_lock); 1437 } 1438 1439 static void __dentry_lease_unlist(struct ceph_dentry_info *di) 1440 { 1441 struct ceph_mds_client *mdsc; 1442 if (di->flags & CEPH_DENTRY_SHRINK_LIST) 1443 return; 1444 if (list_empty(&di->lease_list)) 1445 return; 1446 1447 mdsc = ceph_sb_to_client(di->dentry->d_sb)->mdsc; 1448 spin_lock(&mdsc->dentry_list_lock); 1449 list_del_init(&di->lease_list); 1450 spin_unlock(&mdsc->dentry_list_lock); 1451 } 1452 1453 enum { 1454 KEEP = 0, 1455 DELETE = 1, 1456 TOUCH = 2, 1457 STOP = 4, 1458 }; 1459 1460 struct ceph_lease_walk_control { 1461 bool dir_lease; 1462 bool expire_dir_lease; 1463 unsigned long nr_to_scan; 1464 unsigned long dir_lease_ttl; 1465 }; 1466 1467 static unsigned long 1468 __dentry_leases_walk(struct ceph_mds_client *mdsc, 1469 struct ceph_lease_walk_control *lwc, 1470 int (*check)(struct dentry*, void*)) 1471 { 1472 struct ceph_dentry_info *di, *tmp; 1473 struct dentry *dentry, *last = NULL; 1474 struct list_head* list; 1475 LIST_HEAD(dispose); 1476 unsigned long freed = 0; 1477 int ret = 0; 1478 1479 list = lwc->dir_lease ? &mdsc->dentry_dir_leases : &mdsc->dentry_leases; 1480 spin_lock(&mdsc->dentry_list_lock); 1481 list_for_each_entry_safe(di, tmp, list, lease_list) { 1482 if (!lwc->nr_to_scan) 1483 break; 1484 --lwc->nr_to_scan; 1485 1486 dentry = di->dentry; 1487 if (last == dentry) 1488 break; 1489 1490 if (!spin_trylock(&dentry->d_lock)) 1491 continue; 1492 1493 if (__lockref_is_dead(&dentry->d_lockref)) { 1494 list_del_init(&di->lease_list); 1495 goto next; 1496 } 1497 1498 ret = check(dentry, lwc); 1499 if (ret & TOUCH) { 1500 /* move it into tail of dir lease list */ 1501 __dentry_dir_lease_touch(mdsc, di); 1502 if (!last) 1503 last = dentry; 1504 } 1505 if (ret & DELETE) { 1506 /* stale lease */ 1507 di->flags &= ~CEPH_DENTRY_REFERENCED; 1508 if (dentry->d_lockref.count > 0) { 1509 /* update_dentry_lease() will re-add 1510 * it to lease list, or 1511 * ceph_d_delete() will return 1 when 1512 * last reference is dropped */ 1513 list_del_init(&di->lease_list); 1514 } else { 1515 di->flags |= CEPH_DENTRY_SHRINK_LIST; 1516 list_move_tail(&di->lease_list, &dispose); 1517 dget_dlock(dentry); 1518 } 1519 } 1520 next: 1521 spin_unlock(&dentry->d_lock); 1522 if (ret & STOP) 1523 break; 1524 } 1525 spin_unlock(&mdsc->dentry_list_lock); 1526 1527 while (!list_empty(&dispose)) { 1528 di = list_first_entry(&dispose, struct ceph_dentry_info, 1529 lease_list); 1530 dentry = di->dentry; 1531 spin_lock(&dentry->d_lock); 1532 1533 list_del_init(&di->lease_list); 1534 di->flags &= ~CEPH_DENTRY_SHRINK_LIST; 1535 if (di->flags & CEPH_DENTRY_REFERENCED) { 1536 spin_lock(&mdsc->dentry_list_lock); 1537 if (di->flags & CEPH_DENTRY_LEASE_LIST) { 1538 list_add_tail(&di->lease_list, 1539 &mdsc->dentry_leases); 1540 } else { 1541 __dentry_dir_lease_touch(mdsc, di); 1542 } 1543 spin_unlock(&mdsc->dentry_list_lock); 1544 } else { 1545 freed++; 1546 } 1547 1548 spin_unlock(&dentry->d_lock); 1549 /* ceph_d_delete() does the trick */ 1550 dput(dentry); 1551 } 1552 return freed; 1553 } 1554 1555 static int __dentry_lease_check(struct dentry *dentry, void *arg) 1556 { 1557 struct ceph_dentry_info *di = ceph_dentry(dentry); 1558 int ret; 1559 1560 if (__dentry_lease_is_valid(di)) 1561 return STOP; 1562 ret = __dir_lease_try_check(dentry); 1563 if (ret == -EBUSY) 1564 return KEEP; 1565 if (ret > 0) 1566 return TOUCH; 1567 return DELETE; 1568 } 1569 1570 static int __dir_lease_check(struct dentry *dentry, void *arg) 1571 { 1572 struct ceph_lease_walk_control *lwc = arg; 1573 struct ceph_dentry_info *di = ceph_dentry(dentry); 1574 1575 int ret = __dir_lease_try_check(dentry); 1576 if (ret == -EBUSY) 1577 return KEEP; 1578 if (ret > 0) { 1579 if (time_before(jiffies, di->time + lwc->dir_lease_ttl)) 1580 return STOP; 1581 /* Move dentry to tail of dir lease list if we don't want 1582 * to delete it. So dentries in the list are checked in a 1583 * round robin manner */ 1584 if (!lwc->expire_dir_lease) 1585 return TOUCH; 1586 if (dentry->d_lockref.count > 0 || 1587 (di->flags & CEPH_DENTRY_REFERENCED)) 1588 return TOUCH; 1589 /* invalidate dir lease */ 1590 di->lease_shared_gen = 0; 1591 } 1592 return DELETE; 1593 } 1594 1595 int ceph_trim_dentries(struct ceph_mds_client *mdsc) 1596 { 1597 struct ceph_lease_walk_control lwc; 1598 unsigned long count; 1599 unsigned long freed; 1600 1601 spin_lock(&mdsc->caps_list_lock); 1602 if (mdsc->caps_use_max > 0 && 1603 mdsc->caps_use_count > mdsc->caps_use_max) 1604 count = mdsc->caps_use_count - mdsc->caps_use_max; 1605 else 1606 count = 0; 1607 spin_unlock(&mdsc->caps_list_lock); 1608 1609 lwc.dir_lease = false; 1610 lwc.nr_to_scan = CEPH_CAPS_PER_RELEASE * 2; 1611 freed = __dentry_leases_walk(mdsc, &lwc, __dentry_lease_check); 1612 if (!lwc.nr_to_scan) /* more invalid leases */ 1613 return -EAGAIN; 1614 1615 if (lwc.nr_to_scan < CEPH_CAPS_PER_RELEASE) 1616 lwc.nr_to_scan = CEPH_CAPS_PER_RELEASE; 1617 1618 lwc.dir_lease = true; 1619 lwc.expire_dir_lease = freed < count; 1620 lwc.dir_lease_ttl = mdsc->fsc->mount_options->caps_wanted_delay_max * HZ; 1621 freed +=__dentry_leases_walk(mdsc, &lwc, __dir_lease_check); 1622 if (!lwc.nr_to_scan) /* more to check */ 1623 return -EAGAIN; 1624 1625 return freed > 0 ? 1 : 0; 1626 } 1627 1628 /* 1629 * Ensure a dentry lease will no longer revalidate. 1630 */ 1631 void ceph_invalidate_dentry_lease(struct dentry *dentry) 1632 { 1633 struct ceph_dentry_info *di = ceph_dentry(dentry); 1634 spin_lock(&dentry->d_lock); 1635 di->time = jiffies; 1636 di->lease_shared_gen = 0; 1637 di->flags &= ~CEPH_DENTRY_PRIMARY_LINK; 1638 __dentry_lease_unlist(di); 1639 spin_unlock(&dentry->d_lock); 1640 } 1641 1642 /* 1643 * Check if dentry lease is valid. If not, delete the lease. Try to 1644 * renew if the least is more than half up. 1645 */ 1646 static bool __dentry_lease_is_valid(struct ceph_dentry_info *di) 1647 { 1648 struct ceph_mds_session *session; 1649 1650 if (!di->lease_gen) 1651 return false; 1652 1653 session = di->lease_session; 1654 if (session) { 1655 u32 gen; 1656 unsigned long ttl; 1657 1658 gen = atomic_read(&session->s_cap_gen); 1659 ttl = session->s_cap_ttl; 1660 1661 if (di->lease_gen == gen && 1662 time_before(jiffies, ttl) && 1663 time_before(jiffies, di->time)) 1664 return true; 1665 } 1666 di->lease_gen = 0; 1667 return false; 1668 } 1669 1670 static int dentry_lease_is_valid(struct dentry *dentry, unsigned int flags) 1671 { 1672 struct ceph_dentry_info *di; 1673 struct ceph_mds_session *session = NULL; 1674 u32 seq = 0; 1675 int valid = 0; 1676 1677 spin_lock(&dentry->d_lock); 1678 di = ceph_dentry(dentry); 1679 if (di && __dentry_lease_is_valid(di)) { 1680 valid = 1; 1681 1682 if (di->lease_renew_after && 1683 time_after(jiffies, di->lease_renew_after)) { 1684 /* 1685 * We should renew. If we're in RCU walk mode 1686 * though, we can't do that so just return 1687 * -ECHILD. 1688 */ 1689 if (flags & LOOKUP_RCU) { 1690 valid = -ECHILD; 1691 } else { 1692 session = ceph_get_mds_session(di->lease_session); 1693 seq = di->lease_seq; 1694 di->lease_renew_after = 0; 1695 di->lease_renew_from = jiffies; 1696 } 1697 } 1698 } 1699 spin_unlock(&dentry->d_lock); 1700 1701 if (session) { 1702 ceph_mdsc_lease_send_msg(session, dentry, 1703 CEPH_MDS_LEASE_RENEW, seq); 1704 ceph_put_mds_session(session); 1705 } 1706 dout("dentry_lease_is_valid - dentry %p = %d\n", dentry, valid); 1707 return valid; 1708 } 1709 1710 /* 1711 * Called under dentry->d_lock. 1712 */ 1713 static int __dir_lease_try_check(const struct dentry *dentry) 1714 { 1715 struct ceph_dentry_info *di = ceph_dentry(dentry); 1716 struct inode *dir; 1717 struct ceph_inode_info *ci; 1718 int valid = 0; 1719 1720 if (!di->lease_shared_gen) 1721 return 0; 1722 if (IS_ROOT(dentry)) 1723 return 0; 1724 1725 dir = d_inode(dentry->d_parent); 1726 ci = ceph_inode(dir); 1727 1728 if (spin_trylock(&ci->i_ceph_lock)) { 1729 if (atomic_read(&ci->i_shared_gen) == di->lease_shared_gen && 1730 __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 0)) 1731 valid = 1; 1732 spin_unlock(&ci->i_ceph_lock); 1733 } else { 1734 valid = -EBUSY; 1735 } 1736 1737 if (!valid) 1738 di->lease_shared_gen = 0; 1739 return valid; 1740 } 1741 1742 /* 1743 * Check if directory-wide content lease/cap is valid. 1744 */ 1745 static int dir_lease_is_valid(struct inode *dir, struct dentry *dentry, 1746 struct ceph_mds_client *mdsc) 1747 { 1748 struct ceph_inode_info *ci = ceph_inode(dir); 1749 int valid; 1750 int shared_gen; 1751 1752 spin_lock(&ci->i_ceph_lock); 1753 valid = __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1); 1754 if (valid) { 1755 __ceph_touch_fmode(ci, mdsc, CEPH_FILE_MODE_RD); 1756 shared_gen = atomic_read(&ci->i_shared_gen); 1757 } 1758 spin_unlock(&ci->i_ceph_lock); 1759 if (valid) { 1760 struct ceph_dentry_info *di; 1761 spin_lock(&dentry->d_lock); 1762 di = ceph_dentry(dentry); 1763 if (dir == d_inode(dentry->d_parent) && 1764 di && di->lease_shared_gen == shared_gen) 1765 __ceph_dentry_dir_lease_touch(di); 1766 else 1767 valid = 0; 1768 spin_unlock(&dentry->d_lock); 1769 } 1770 dout("dir_lease_is_valid dir %p v%u dentry %p = %d\n", 1771 dir, (unsigned)atomic_read(&ci->i_shared_gen), dentry, valid); 1772 return valid; 1773 } 1774 1775 /* 1776 * Check if cached dentry can be trusted. 1777 */ 1778 static int ceph_d_revalidate(struct dentry *dentry, unsigned int flags) 1779 { 1780 int valid = 0; 1781 struct dentry *parent; 1782 struct inode *dir, *inode; 1783 struct ceph_mds_client *mdsc; 1784 1785 valid = fscrypt_d_revalidate(dentry, flags); 1786 if (valid <= 0) 1787 return valid; 1788 1789 if (flags & LOOKUP_RCU) { 1790 parent = READ_ONCE(dentry->d_parent); 1791 dir = d_inode_rcu(parent); 1792 if (!dir) 1793 return -ECHILD; 1794 inode = d_inode_rcu(dentry); 1795 } else { 1796 parent = dget_parent(dentry); 1797 dir = d_inode(parent); 1798 inode = d_inode(dentry); 1799 } 1800 1801 dout("d_revalidate %p '%pd' inode %p offset 0x%llx nokey %d\n", dentry, 1802 dentry, inode, ceph_dentry(dentry)->offset, 1803 !!(dentry->d_flags & DCACHE_NOKEY_NAME)); 1804 1805 mdsc = ceph_sb_to_client(dir->i_sb)->mdsc; 1806 1807 /* always trust cached snapped dentries, snapdir dentry */ 1808 if (ceph_snap(dir) != CEPH_NOSNAP) { 1809 dout("d_revalidate %p '%pd' inode %p is SNAPPED\n", dentry, 1810 dentry, inode); 1811 valid = 1; 1812 } else if (inode && ceph_snap(inode) == CEPH_SNAPDIR) { 1813 valid = 1; 1814 } else { 1815 valid = dentry_lease_is_valid(dentry, flags); 1816 if (valid == -ECHILD) 1817 return valid; 1818 if (valid || dir_lease_is_valid(dir, dentry, mdsc)) { 1819 if (inode) 1820 valid = ceph_is_any_caps(inode); 1821 else 1822 valid = 1; 1823 } 1824 } 1825 1826 if (!valid) { 1827 struct ceph_mds_request *req; 1828 int op, err; 1829 u32 mask; 1830 1831 if (flags & LOOKUP_RCU) 1832 return -ECHILD; 1833 1834 percpu_counter_inc(&mdsc->metric.d_lease_mis); 1835 1836 op = ceph_snap(dir) == CEPH_SNAPDIR ? 1837 CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP; 1838 req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS); 1839 if (!IS_ERR(req)) { 1840 req->r_dentry = dget(dentry); 1841 req->r_num_caps = 2; 1842 req->r_parent = dir; 1843 ihold(dir); 1844 1845 mask = CEPH_STAT_CAP_INODE | CEPH_CAP_AUTH_SHARED; 1846 if (ceph_security_xattr_wanted(dir)) 1847 mask |= CEPH_CAP_XATTR_SHARED; 1848 req->r_args.getattr.mask = cpu_to_le32(mask); 1849 1850 err = ceph_mdsc_do_request(mdsc, NULL, req); 1851 switch (err) { 1852 case 0: 1853 if (d_really_is_positive(dentry) && 1854 d_inode(dentry) == req->r_target_inode) 1855 valid = 1; 1856 break; 1857 case -ENOENT: 1858 if (d_really_is_negative(dentry)) 1859 valid = 1; 1860 fallthrough; 1861 default: 1862 break; 1863 } 1864 ceph_mdsc_put_request(req); 1865 dout("d_revalidate %p lookup result=%d\n", 1866 dentry, err); 1867 } 1868 } else { 1869 percpu_counter_inc(&mdsc->metric.d_lease_hit); 1870 } 1871 1872 dout("d_revalidate %p %s\n", dentry, valid ? "valid" : "invalid"); 1873 if (!valid) 1874 ceph_dir_clear_complete(dir); 1875 1876 if (!(flags & LOOKUP_RCU)) 1877 dput(parent); 1878 return valid; 1879 } 1880 1881 /* 1882 * Delete unused dentry that doesn't have valid lease 1883 * 1884 * Called under dentry->d_lock. 1885 */ 1886 static int ceph_d_delete(const struct dentry *dentry) 1887 { 1888 struct ceph_dentry_info *di; 1889 1890 /* won't release caps */ 1891 if (d_really_is_negative(dentry)) 1892 return 0; 1893 if (ceph_snap(d_inode(dentry)) != CEPH_NOSNAP) 1894 return 0; 1895 /* vaild lease? */ 1896 di = ceph_dentry(dentry); 1897 if (di) { 1898 if (__dentry_lease_is_valid(di)) 1899 return 0; 1900 if (__dir_lease_try_check(dentry)) 1901 return 0; 1902 } 1903 return 1; 1904 } 1905 1906 /* 1907 * Release our ceph_dentry_info. 1908 */ 1909 static void ceph_d_release(struct dentry *dentry) 1910 { 1911 struct ceph_dentry_info *di = ceph_dentry(dentry); 1912 struct ceph_fs_client *fsc = ceph_sb_to_client(dentry->d_sb); 1913 1914 dout("d_release %p\n", dentry); 1915 1916 atomic64_dec(&fsc->mdsc->metric.total_dentries); 1917 1918 spin_lock(&dentry->d_lock); 1919 __dentry_lease_unlist(di); 1920 dentry->d_fsdata = NULL; 1921 spin_unlock(&dentry->d_lock); 1922 1923 ceph_put_mds_session(di->lease_session); 1924 kmem_cache_free(ceph_dentry_cachep, di); 1925 } 1926 1927 /* 1928 * When the VFS prunes a dentry from the cache, we need to clear the 1929 * complete flag on the parent directory. 1930 * 1931 * Called under dentry->d_lock. 1932 */ 1933 static void ceph_d_prune(struct dentry *dentry) 1934 { 1935 struct ceph_inode_info *dir_ci; 1936 struct ceph_dentry_info *di; 1937 1938 dout("ceph_d_prune %pd %p\n", dentry, dentry); 1939 1940 /* do we have a valid parent? */ 1941 if (IS_ROOT(dentry)) 1942 return; 1943 1944 /* we hold d_lock, so d_parent is stable */ 1945 dir_ci = ceph_inode(d_inode(dentry->d_parent)); 1946 if (dir_ci->i_vino.snap == CEPH_SNAPDIR) 1947 return; 1948 1949 /* who calls d_delete() should also disable dcache readdir */ 1950 if (d_really_is_negative(dentry)) 1951 return; 1952 1953 /* d_fsdata does not get cleared until d_release */ 1954 if (!d_unhashed(dentry)) { 1955 __ceph_dir_clear_complete(dir_ci); 1956 return; 1957 } 1958 1959 /* Disable dcache readdir just in case that someone called d_drop() 1960 * or d_invalidate(), but MDS didn't revoke CEPH_CAP_FILE_SHARED 1961 * properly (dcache readdir is still enabled) */ 1962 di = ceph_dentry(dentry); 1963 if (di->offset > 0 && 1964 di->lease_shared_gen == atomic_read(&dir_ci->i_shared_gen)) 1965 __ceph_dir_clear_ordered(dir_ci); 1966 } 1967 1968 /* 1969 * read() on a dir. This weird interface hack only works if mounted 1970 * with '-o dirstat'. 1971 */ 1972 static ssize_t ceph_read_dir(struct file *file, char __user *buf, size_t size, 1973 loff_t *ppos) 1974 { 1975 struct ceph_dir_file_info *dfi = file->private_data; 1976 struct inode *inode = file_inode(file); 1977 struct ceph_inode_info *ci = ceph_inode(inode); 1978 int left; 1979 const int bufsize = 1024; 1980 1981 if (!ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), DIRSTAT)) 1982 return -EISDIR; 1983 1984 if (!dfi->dir_info) { 1985 dfi->dir_info = kmalloc(bufsize, GFP_KERNEL); 1986 if (!dfi->dir_info) 1987 return -ENOMEM; 1988 dfi->dir_info_len = 1989 snprintf(dfi->dir_info, bufsize, 1990 "entries: %20lld\n" 1991 " files: %20lld\n" 1992 " subdirs: %20lld\n" 1993 "rentries: %20lld\n" 1994 " rfiles: %20lld\n" 1995 " rsubdirs: %20lld\n" 1996 "rbytes: %20lld\n" 1997 "rctime: %10lld.%09ld\n", 1998 ci->i_files + ci->i_subdirs, 1999 ci->i_files, 2000 ci->i_subdirs, 2001 ci->i_rfiles + ci->i_rsubdirs, 2002 ci->i_rfiles, 2003 ci->i_rsubdirs, 2004 ci->i_rbytes, 2005 ci->i_rctime.tv_sec, 2006 ci->i_rctime.tv_nsec); 2007 } 2008 2009 if (*ppos >= dfi->dir_info_len) 2010 return 0; 2011 size = min_t(unsigned, size, dfi->dir_info_len-*ppos); 2012 left = copy_to_user(buf, dfi->dir_info + *ppos, size); 2013 if (left == size) 2014 return -EFAULT; 2015 *ppos += (size - left); 2016 return size - left; 2017 } 2018 2019 2020 2021 /* 2022 * Return name hash for a given dentry. This is dependent on 2023 * the parent directory's hash function. 2024 */ 2025 unsigned ceph_dentry_hash(struct inode *dir, struct dentry *dn) 2026 { 2027 struct ceph_inode_info *dci = ceph_inode(dir); 2028 unsigned hash; 2029 2030 switch (dci->i_dir_layout.dl_dir_hash) { 2031 case 0: /* for backward compat */ 2032 case CEPH_STR_HASH_LINUX: 2033 return dn->d_name.hash; 2034 2035 default: 2036 spin_lock(&dn->d_lock); 2037 hash = ceph_str_hash(dci->i_dir_layout.dl_dir_hash, 2038 dn->d_name.name, dn->d_name.len); 2039 spin_unlock(&dn->d_lock); 2040 return hash; 2041 } 2042 } 2043 2044 WRAP_DIR_ITER(ceph_readdir) // FIXME! 2045 const struct file_operations ceph_dir_fops = { 2046 .read = ceph_read_dir, 2047 .iterate_shared = shared_ceph_readdir, 2048 .llseek = ceph_dir_llseek, 2049 .open = ceph_open, 2050 .release = ceph_release, 2051 .unlocked_ioctl = ceph_ioctl, 2052 .compat_ioctl = compat_ptr_ioctl, 2053 .fsync = ceph_fsync, 2054 .lock = ceph_lock, 2055 .flock = ceph_flock, 2056 }; 2057 2058 const struct file_operations ceph_snapdir_fops = { 2059 .iterate_shared = shared_ceph_readdir, 2060 .llseek = ceph_dir_llseek, 2061 .open = ceph_open, 2062 .release = ceph_release, 2063 }; 2064 2065 const struct inode_operations ceph_dir_iops = { 2066 .lookup = ceph_lookup, 2067 .permission = ceph_permission, 2068 .getattr = ceph_getattr, 2069 .setattr = ceph_setattr, 2070 .listxattr = ceph_listxattr, 2071 .get_inode_acl = ceph_get_acl, 2072 .set_acl = ceph_set_acl, 2073 .mknod = ceph_mknod, 2074 .symlink = ceph_symlink, 2075 .mkdir = ceph_mkdir, 2076 .link = ceph_link, 2077 .unlink = ceph_unlink, 2078 .rmdir = ceph_unlink, 2079 .rename = ceph_rename, 2080 .create = ceph_create, 2081 .atomic_open = ceph_atomic_open, 2082 }; 2083 2084 const struct inode_operations ceph_snapdir_iops = { 2085 .lookup = ceph_lookup, 2086 .permission = ceph_permission, 2087 .getattr = ceph_getattr, 2088 .mkdir = ceph_mkdir, 2089 .rmdir = ceph_unlink, 2090 .rename = ceph_rename, 2091 }; 2092 2093 const struct dentry_operations ceph_dentry_ops = { 2094 .d_revalidate = ceph_d_revalidate, 2095 .d_delete = ceph_d_delete, 2096 .d_release = ceph_d_release, 2097 .d_prune = ceph_d_prune, 2098 .d_init = ceph_d_init, 2099 }; 2100