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#include "crypto.h" 13 14/* 15 * Directory operations: readdir, lookup, create, link, unlink, 16 * rename, etc. 17 */ 18 19/* 20 * Ceph MDS operations are specified in terms of a base ino and 21 * relative path. Thus, the client can specify an operation on a 22 * specific inode (e.g., a getattr due to fstat(2)), or as a path 23 * relative to, say, the root directory. 24 * 25 * Normally, we limit ourselves to strict inode ops (no path component) 26 * or dentry operations (a single path component relative to an ino). The 27 * exception to this is open_root_dentry(), which will open the mount 28 * point by name. 29 */ 30 31const struct dentry_operations ceph_dentry_ops; 32 33static bool __dentry_lease_is_valid(struct ceph_dentry_info *di); 34static int __dir_lease_try_check(const struct dentry *dentry); 35 36/* 37 * Initialize ceph dentry state. 38 */ 39static int ceph_d_init(struct dentry *dentry) 40{ 41 struct ceph_dentry_info *di; 42 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(dentry->d_sb); 43 44 di = kmem_cache_zalloc(ceph_dentry_cachep, GFP_KERNEL); 45 if (!di) 46 return -ENOMEM; /* oh well */ 47 48 di->dentry = dentry; 49 di->lease_session = NULL; 50 di->time = jiffies; 51 dentry->d_fsdata = di; 52 INIT_LIST_HEAD(&di->lease_list); 53 54 atomic64_inc(&mdsc->metric.total_dentries); 55 56 return 0; 57} 58 59/* 60 * for f_pos for readdir: 61 * - hash order: 62 * (0xff << 52) | ((24 bits hash) << 28) | 63 * (the nth entry has hash collision); 64 * - frag+name order; 65 * ((frag value) << 28) | (the nth entry in frag); 66 */ 67#define OFFSET_BITS 28 68#define OFFSET_MASK ((1 << OFFSET_BITS) - 1) 69#define HASH_ORDER (0xffull << (OFFSET_BITS + 24)) 70loff_t ceph_make_fpos(unsigned high, unsigned off, bool hash_order) 71{ 72 loff_t fpos = ((loff_t)high << 28) | (loff_t)off; 73 if (hash_order) 74 fpos |= HASH_ORDER; 75 return fpos; 76} 77 78static bool is_hash_order(loff_t p) 79{ 80 return (p & HASH_ORDER) == HASH_ORDER; 81} 82 83static unsigned fpos_frag(loff_t p) 84{ 85 return p >> OFFSET_BITS; 86} 87 88static unsigned fpos_hash(loff_t p) 89{ 90 return ceph_frag_value(fpos_frag(p)); 91} 92 93static unsigned fpos_off(loff_t p) 94{ 95 return p & OFFSET_MASK; 96} 97 98static int fpos_cmp(loff_t l, loff_t r) 99{ 100 int v = ceph_frag_compare(fpos_frag(l), fpos_frag(r)); 101 if (v) 102 return v; 103 return (int)(fpos_off(l) - fpos_off(r)); 104} 105 106/* 107 * make note of the last dentry we read, so we can 108 * continue at the same lexicographical point, 109 * regardless of what dir changes take place on the 110 * server. 111 */ 112static int note_last_dentry(struct ceph_dir_file_info *dfi, const char *name, 113 int len, unsigned next_offset) 114{ 115 char *buf = kmalloc(len+1, GFP_KERNEL); 116 if (!buf) 117 return -ENOMEM; 118 kfree(dfi->last_name); 119 dfi->last_name = buf; 120 memcpy(dfi->last_name, name, len); 121 dfi->last_name[len] = 0; 122 dfi->next_offset = next_offset; 123 dout("note_last_dentry '%s'\n", dfi->last_name); 124 return 0; 125} 126 127 128static struct dentry * 129__dcache_find_get_entry(struct dentry *parent, u64 idx, 130 struct ceph_readdir_cache_control *cache_ctl) 131{ 132 struct inode *dir = d_inode(parent); 133 struct dentry *dentry; 134 unsigned idx_mask = (PAGE_SIZE / sizeof(struct dentry *)) - 1; 135 loff_t ptr_pos = idx * sizeof(struct dentry *); 136 pgoff_t ptr_pgoff = ptr_pos >> PAGE_SHIFT; 137 138 if (ptr_pos >= i_size_read(dir)) 139 return NULL; 140 141 if (!cache_ctl->page || ptr_pgoff != page_index(cache_ctl->page)) { 142 ceph_readdir_cache_release(cache_ctl); 143 cache_ctl->page = find_lock_page(&dir->i_data, ptr_pgoff); 144 if (!cache_ctl->page) { 145 dout(" page %lu not found\n", ptr_pgoff); 146 return ERR_PTR(-EAGAIN); 147 } 148 /* reading/filling the cache are serialized by 149 i_rwsem, no need to use page lock */ 150 unlock_page(cache_ctl->page); 151 cache_ctl->dentries = kmap(cache_ctl->page); 152 } 153 154 cache_ctl->index = idx & idx_mask; 155 156 rcu_read_lock(); 157 spin_lock(&parent->d_lock); 158 /* check i_size again here, because empty directory can be 159 * marked as complete while not holding the i_rwsem. */ 160 if (ceph_dir_is_complete_ordered(dir) && ptr_pos < i_size_read(dir)) 161 dentry = cache_ctl->dentries[cache_ctl->index]; 162 else 163 dentry = NULL; 164 spin_unlock(&parent->d_lock); 165 if (dentry && !lockref_get_not_dead(&dentry->d_lockref)) 166 dentry = NULL; 167 rcu_read_unlock(); 168 return dentry ? : ERR_PTR(-EAGAIN); 169} 170 171/* 172 * When possible, we try to satisfy a readdir by peeking at the 173 * dcache. We make this work by carefully ordering dentries on 174 * d_child when we initially get results back from the MDS, and 175 * falling back to a "normal" sync readdir if any dentries in the dir 176 * are dropped. 177 * 178 * Complete dir indicates that we have all dentries in the dir. It is 179 * defined IFF we hold CEPH_CAP_FILE_SHARED (which will be revoked by 180 * the MDS if/when the directory is modified). 181 */ 182static int __dcache_readdir(struct file *file, struct dir_context *ctx, 183 int shared_gen) 184{ 185 struct ceph_dir_file_info *dfi = file->private_data; 186 struct dentry *parent = file->f_path.dentry; 187 struct inode *dir = d_inode(parent); 188 struct dentry *dentry, *last = NULL; 189 struct ceph_dentry_info *di; 190 struct ceph_readdir_cache_control cache_ctl = {}; 191 u64 idx = 0; 192 int err = 0; 193 194 dout("__dcache_readdir %p v%u at %llx\n", dir, (unsigned)shared_gen, ctx->pos); 195 196 /* search start position */ 197 if (ctx->pos > 2) { 198 u64 count = div_u64(i_size_read(dir), sizeof(struct dentry *)); 199 while (count > 0) { 200 u64 step = count >> 1; 201 dentry = __dcache_find_get_entry(parent, idx + step, 202 &cache_ctl); 203 if (!dentry) { 204 /* use linar search */ 205 idx = 0; 206 break; 207 } 208 if (IS_ERR(dentry)) { 209 err = PTR_ERR(dentry); 210 goto out; 211 } 212 di = ceph_dentry(dentry); 213 spin_lock(&dentry->d_lock); 214 if (fpos_cmp(di->offset, ctx->pos) < 0) { 215 idx += step + 1; 216 count -= step + 1; 217 } else { 218 count = step; 219 } 220 spin_unlock(&dentry->d_lock); 221 dput(dentry); 222 } 223 224 dout("__dcache_readdir %p cache idx %llu\n", dir, idx); 225 } 226 227 228 for (;;) { 229 bool emit_dentry = false; 230 dentry = __dcache_find_get_entry(parent, idx++, &cache_ctl); 231 if (!dentry) { 232 dfi->file_info.flags |= CEPH_F_ATEND; 233 err = 0; 234 break; 235 } 236 if (IS_ERR(dentry)) { 237 err = PTR_ERR(dentry); 238 goto out; 239 } 240 241 spin_lock(&dentry->d_lock); 242 di = ceph_dentry(dentry); 243 if (d_unhashed(dentry) || 244 d_really_is_negative(dentry) || 245 di->lease_shared_gen != shared_gen || 246 ((dentry->d_flags & DCACHE_NOKEY_NAME) && 247 fscrypt_has_encryption_key(dir))) { 248 spin_unlock(&dentry->d_lock); 249 dput(dentry); 250 err = -EAGAIN; 251 goto out; 252 } 253 if (fpos_cmp(ctx->pos, di->offset) <= 0) { 254 __ceph_dentry_dir_lease_touch(di); 255 emit_dentry = true; 256 } 257 spin_unlock(&dentry->d_lock); 258 259 if (emit_dentry) { 260 dout(" %llx dentry %p %pd %p\n", di->offset, 261 dentry, dentry, d_inode(dentry)); 262 ctx->pos = di->offset; 263 if (!dir_emit(ctx, dentry->d_name.name, 264 dentry->d_name.len, ceph_present_inode(d_inode(dentry)), 265 d_inode(dentry)->i_mode >> 12)) { 266 dput(dentry); 267 err = 0; 268 break; 269 } 270 ctx->pos++; 271 272 if (last) 273 dput(last); 274 last = dentry; 275 } else { 276 dput(dentry); 277 } 278 } 279out: 280 ceph_readdir_cache_release(&cache_ctl); 281 if (last) { 282 int ret; 283 di = ceph_dentry(last); 284 ret = note_last_dentry(dfi, last->d_name.name, last->d_name.len, 285 fpos_off(di->offset) + 1); 286 if (ret < 0) 287 err = ret; 288 dput(last); 289 /* last_name no longer match cache index */ 290 if (dfi->readdir_cache_idx >= 0) { 291 dfi->readdir_cache_idx = -1; 292 dfi->dir_release_count = 0; 293 } 294 } 295 return err; 296} 297 298static bool need_send_readdir(struct ceph_dir_file_info *dfi, loff_t pos) 299{ 300 if (!dfi->last_readdir) 301 return true; 302 if (is_hash_order(pos)) 303 return !ceph_frag_contains_value(dfi->frag, fpos_hash(pos)); 304 else 305 return dfi->frag != fpos_frag(pos); 306} 307 308static int ceph_readdir(struct file *file, struct dir_context *ctx) 309{ 310 struct ceph_dir_file_info *dfi = file->private_data; 311 struct inode *inode = file_inode(file); 312 struct ceph_inode_info *ci = ceph_inode(inode); 313 struct ceph_fs_client *fsc = ceph_inode_to_client(inode); 314 struct ceph_mds_client *mdsc = fsc->mdsc; 315 int i; 316 int err; 317 unsigned frag = -1; 318 struct ceph_mds_reply_info_parsed *rinfo; 319 320 dout("readdir %p file %p pos %llx\n", inode, file, ctx->pos); 321 if (dfi->file_info.flags & CEPH_F_ATEND) 322 return 0; 323 324 /* always start with . and .. */ 325 if (ctx->pos == 0) { 326 dout("readdir off 0 -> '.'\n"); 327 if (!dir_emit(ctx, ".", 1, ceph_present_inode(inode), 328 inode->i_mode >> 12)) 329 return 0; 330 ctx->pos = 1; 331 } 332 if (ctx->pos == 1) { 333 u64 ino; 334 struct dentry *dentry = file->f_path.dentry; 335 336 spin_lock(&dentry->d_lock); 337 ino = ceph_present_inode(dentry->d_parent->d_inode); 338 spin_unlock(&dentry->d_lock); 339 340 dout("readdir off 1 -> '..'\n"); 341 if (!dir_emit(ctx, "..", 2, ino, inode->i_mode >> 12)) 342 return 0; 343 ctx->pos = 2; 344 } 345 346 err = ceph_fscrypt_prepare_readdir(inode); 347 if (err < 0) 348 return err; 349 350 spin_lock(&ci->i_ceph_lock); 351 /* request Fx cap. if have Fx, we don't need to release Fs cap 352 * for later create/unlink. */ 353 __ceph_touch_fmode(ci, mdsc, CEPH_FILE_MODE_WR); 354 /* can we use the dcache? */ 355 if (ceph_test_mount_opt(fsc, DCACHE) && 356 !ceph_test_mount_opt(fsc, NOASYNCREADDIR) && 357 ceph_snap(inode) != CEPH_SNAPDIR && 358 __ceph_dir_is_complete_ordered(ci) && 359 __ceph_caps_issued_mask_metric(ci, CEPH_CAP_FILE_SHARED, 1)) { 360 int shared_gen = atomic_read(&ci->i_shared_gen); 361 362 spin_unlock(&ci->i_ceph_lock); 363 err = __dcache_readdir(file, ctx, shared_gen); 364 if (err != -EAGAIN) 365 return err; 366 } else { 367 spin_unlock(&ci->i_ceph_lock); 368 } 369 370 /* proceed with a normal readdir */ 371more: 372 /* do we have the correct frag content buffered? */ 373 if (need_send_readdir(dfi, ctx->pos)) { 374 struct ceph_mds_request *req; 375 int op = ceph_snap(inode) == CEPH_SNAPDIR ? 376 CEPH_MDS_OP_LSSNAP : CEPH_MDS_OP_READDIR; 377 378 /* discard old result, if any */ 379 if (dfi->last_readdir) { 380 ceph_mdsc_put_request(dfi->last_readdir); 381 dfi->last_readdir = NULL; 382 } 383 384 if (is_hash_order(ctx->pos)) { 385 /* fragtree isn't always accurate. choose frag 386 * based on previous reply when possible. */ 387 if (frag == (unsigned)-1) 388 frag = ceph_choose_frag(ci, fpos_hash(ctx->pos), 389 NULL, NULL); 390 } else { 391 frag = fpos_frag(ctx->pos); 392 } 393 394 dout("readdir fetching %llx.%llx frag %x offset '%s'\n", 395 ceph_vinop(inode), frag, dfi->last_name); 396 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS); 397 if (IS_ERR(req)) 398 return PTR_ERR(req); 399 400 err = ceph_alloc_readdir_reply_buffer(req, inode); 401 if (err) { 402 ceph_mdsc_put_request(req); 403 return err; 404 } 405 /* hints to request -> mds selection code */ 406 req->r_direct_mode = USE_AUTH_MDS; 407 if (op == CEPH_MDS_OP_READDIR) { 408 req->r_direct_hash = ceph_frag_value(frag); 409 __set_bit(CEPH_MDS_R_DIRECT_IS_HASH, &req->r_req_flags); 410 req->r_inode_drop = CEPH_CAP_FILE_EXCL; 411 } 412 if (dfi->last_name) { 413 struct qstr d_name = { .name = dfi->last_name, 414 .len = strlen(dfi->last_name) }; 415 416 req->r_path2 = kzalloc(NAME_MAX + 1, GFP_KERNEL); 417 if (!req->r_path2) { 418 ceph_mdsc_put_request(req); 419 return -ENOMEM; 420 } 421 422 err = ceph_encode_encrypted_dname(inode, &d_name, 423 req->r_path2); 424 if (err < 0) { 425 ceph_mdsc_put_request(req); 426 return err; 427 } 428 } else if (is_hash_order(ctx->pos)) { 429 req->r_args.readdir.offset_hash = 430 cpu_to_le32(fpos_hash(ctx->pos)); 431 } 432 433 req->r_dir_release_cnt = dfi->dir_release_count; 434 req->r_dir_ordered_cnt = dfi->dir_ordered_count; 435 req->r_readdir_cache_idx = dfi->readdir_cache_idx; 436 req->r_readdir_offset = dfi->next_offset; 437 req->r_args.readdir.frag = cpu_to_le32(frag); 438 req->r_args.readdir.flags = 439 cpu_to_le16(CEPH_READDIR_REPLY_BITFLAGS); 440 441 req->r_inode = inode; 442 ihold(inode); 443 req->r_dentry = dget(file->f_path.dentry); 444 err = ceph_mdsc_do_request(mdsc, NULL, req); 445 if (err < 0) { 446 ceph_mdsc_put_request(req); 447 return err; 448 } 449 dout("readdir got and parsed readdir result=%d on " 450 "frag %x, end=%d, complete=%d, hash_order=%d\n", 451 err, frag, 452 (int)req->r_reply_info.dir_end, 453 (int)req->r_reply_info.dir_complete, 454 (int)req->r_reply_info.hash_order); 455 456 rinfo = &req->r_reply_info; 457 if (le32_to_cpu(rinfo->dir_dir->frag) != frag) { 458 frag = le32_to_cpu(rinfo->dir_dir->frag); 459 if (!rinfo->hash_order) { 460 dfi->next_offset = req->r_readdir_offset; 461 /* adjust ctx->pos to beginning of frag */ 462 ctx->pos = ceph_make_fpos(frag, 463 dfi->next_offset, 464 false); 465 } 466 } 467 468 dfi->frag = frag; 469 dfi->last_readdir = req; 470 471 if (test_bit(CEPH_MDS_R_DID_PREPOPULATE, &req->r_req_flags)) { 472 dfi->readdir_cache_idx = req->r_readdir_cache_idx; 473 if (dfi->readdir_cache_idx < 0) { 474 /* preclude from marking dir ordered */ 475 dfi->dir_ordered_count = 0; 476 } else if (ceph_frag_is_leftmost(frag) && 477 dfi->next_offset == 2) { 478 /* note dir version at start of readdir so 479 * we can tell if any dentries get dropped */ 480 dfi->dir_release_count = req->r_dir_release_cnt; 481 dfi->dir_ordered_count = req->r_dir_ordered_cnt; 482 } 483 } else { 484 dout("readdir !did_prepopulate\n"); 485 /* disable readdir cache */ 486 dfi->readdir_cache_idx = -1; 487 /* preclude from marking dir complete */ 488 dfi->dir_release_count = 0; 489 } 490 491 /* note next offset and last dentry name */ 492 if (rinfo->dir_nr > 0) { 493 struct ceph_mds_reply_dir_entry *rde = 494 rinfo->dir_entries + (rinfo->dir_nr-1); 495 unsigned next_offset = req->r_reply_info.dir_end ? 496 2 : (fpos_off(rde->offset) + 1); 497 err = note_last_dentry(dfi, rde->name, rde->name_len, 498 next_offset); 499 if (err) { 500 ceph_mdsc_put_request(dfi->last_readdir); 501 dfi->last_readdir = NULL; 502 return err; 503 } 504 } else if (req->r_reply_info.dir_end) { 505 dfi->next_offset = 2; 506 /* keep last name */ 507 } 508 } 509 510 rinfo = &dfi->last_readdir->r_reply_info; 511 dout("readdir frag %x num %d pos %llx chunk first %llx\n", 512 dfi->frag, rinfo->dir_nr, ctx->pos, 513 rinfo->dir_nr ? rinfo->dir_entries[0].offset : 0LL); 514 515 i = 0; 516 /* search start position */ 517 if (rinfo->dir_nr > 0) { 518 int step, nr = rinfo->dir_nr; 519 while (nr > 0) { 520 step = nr >> 1; 521 if (rinfo->dir_entries[i + step].offset < ctx->pos) { 522 i += step + 1; 523 nr -= step + 1; 524 } else { 525 nr = step; 526 } 527 } 528 } 529 for (; i < rinfo->dir_nr; i++) { 530 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i; 531 532 if (rde->offset < ctx->pos) { 533 pr_warn("%s: rde->offset 0x%llx ctx->pos 0x%llx\n", 534 __func__, rde->offset, ctx->pos); 535 return -EIO; 536 } 537 538 if (WARN_ON_ONCE(!rde->inode.in)) 539 return -EIO; 540 541 ctx->pos = rde->offset; 542 dout("readdir (%d/%d) -> %llx '%.*s' %p\n", 543 i, rinfo->dir_nr, ctx->pos, 544 rde->name_len, rde->name, &rde->inode.in); 545 546 if (!dir_emit(ctx, rde->name, rde->name_len, 547 ceph_present_ino(inode->i_sb, le64_to_cpu(rde->inode.in->ino)), 548 le32_to_cpu(rde->inode.in->mode) >> 12)) { 549 /* 550 * NOTE: Here no need to put the 'dfi->last_readdir', 551 * because when dir_emit stops us it's most likely 552 * doesn't have enough memory, etc. So for next readdir 553 * it will continue. 554 */ 555 dout("filldir stopping us...\n"); 556 return 0; 557 } 558 559 /* Reset the lengths to their original allocated vals */ 560 ctx->pos++; 561 } 562 563 ceph_mdsc_put_request(dfi->last_readdir); 564 dfi->last_readdir = NULL; 565 566 if (dfi->next_offset > 2) { 567 frag = dfi->frag; 568 goto more; 569 } 570 571 /* more frags? */ 572 if (!ceph_frag_is_rightmost(dfi->frag)) { 573 frag = ceph_frag_next(dfi->frag); 574 if (is_hash_order(ctx->pos)) { 575 loff_t new_pos = ceph_make_fpos(ceph_frag_value(frag), 576 dfi->next_offset, true); 577 if (new_pos > ctx->pos) 578 ctx->pos = new_pos; 579 /* keep last_name */ 580 } else { 581 ctx->pos = ceph_make_fpos(frag, dfi->next_offset, 582 false); 583 kfree(dfi->last_name); 584 dfi->last_name = NULL; 585 } 586 dout("readdir next frag is %x\n", frag); 587 goto more; 588 } 589 dfi->file_info.flags |= CEPH_F_ATEND; 590 591 /* 592 * if dir_release_count still matches the dir, no dentries 593 * were released during the whole readdir, and we should have 594 * the complete dir contents in our cache. 595 */ 596 if (atomic64_read(&ci->i_release_count) == 597 dfi->dir_release_count) { 598 spin_lock(&ci->i_ceph_lock); 599 if (dfi->dir_ordered_count == 600 atomic64_read(&ci->i_ordered_count)) { 601 dout(" marking %p complete and ordered\n", inode); 602 /* use i_size to track number of entries in 603 * readdir cache */ 604 BUG_ON(dfi->readdir_cache_idx < 0); 605 i_size_write(inode, dfi->readdir_cache_idx * 606 sizeof(struct dentry*)); 607 } else { 608 dout(" marking %p complete\n", inode); 609 } 610 __ceph_dir_set_complete(ci, dfi->dir_release_count, 611 dfi->dir_ordered_count); 612 spin_unlock(&ci->i_ceph_lock); 613 } 614 dout("readdir %p file %p done.\n", inode, file); 615 return 0; 616} 617 618static void reset_readdir(struct ceph_dir_file_info *dfi) 619{ 620 if (dfi->last_readdir) { 621 ceph_mdsc_put_request(dfi->last_readdir); 622 dfi->last_readdir = NULL; 623 } 624 kfree(dfi->last_name); 625 dfi->last_name = NULL; 626 dfi->dir_release_count = 0; 627 dfi->readdir_cache_idx = -1; 628 dfi->next_offset = 2; /* compensate for . and .. */ 629 dfi->file_info.flags &= ~CEPH_F_ATEND; 630} 631 632/* 633 * discard buffered readdir content on seekdir(0), or seek to new frag, 634 * or seek prior to current chunk 635 */ 636static bool need_reset_readdir(struct ceph_dir_file_info *dfi, loff_t new_pos) 637{ 638 struct ceph_mds_reply_info_parsed *rinfo; 639 loff_t chunk_offset; 640 if (new_pos == 0) 641 return true; 642 if (is_hash_order(new_pos)) { 643 /* no need to reset last_name for a forward seek when 644 * dentries are sotred in hash order */ 645 } else if (dfi->frag != fpos_frag(new_pos)) { 646 return true; 647 } 648 rinfo = dfi->last_readdir ? &dfi->last_readdir->r_reply_info : NULL; 649 if (!rinfo || !rinfo->dir_nr) 650 return true; 651 chunk_offset = rinfo->dir_entries[0].offset; 652 return new_pos < chunk_offset || 653 is_hash_order(new_pos) != is_hash_order(chunk_offset); 654} 655 656static loff_t ceph_dir_llseek(struct file *file, loff_t offset, int whence) 657{ 658 struct ceph_dir_file_info *dfi = file->private_data; 659 struct inode *inode = file->f_mapping->host; 660 loff_t retval; 661 662 inode_lock(inode); 663 retval = -EINVAL; 664 switch (whence) { 665 case SEEK_CUR: 666 offset += file->f_pos; 667 break; 668 case SEEK_SET: 669 break; 670 case SEEK_END: 671 retval = -EOPNOTSUPP; 672 goto out; 673 default: 674 goto out; 675 } 676 677 if (offset >= 0) { 678 if (need_reset_readdir(dfi, offset)) { 679 dout("dir_llseek dropping %p content\n", file); 680 reset_readdir(dfi); 681 } else if (is_hash_order(offset) && offset > file->f_pos) { 682 /* for hash offset, we don't know if a forward seek 683 * is within same frag */ 684 dfi->dir_release_count = 0; 685 dfi->readdir_cache_idx = -1; 686 } 687 688 if (offset != file->f_pos) { 689 file->f_pos = offset; 690 file->f_version = 0; 691 dfi->file_info.flags &= ~CEPH_F_ATEND; 692 } 693 retval = offset; 694 } 695out: 696 inode_unlock(inode); 697 return retval; 698} 699 700/* 701 * Handle lookups for the hidden .snap directory. 702 */ 703struct dentry *ceph_handle_snapdir(struct ceph_mds_request *req, 704 struct dentry *dentry) 705{ 706 struct ceph_fs_client *fsc = ceph_sb_to_client(dentry->d_sb); 707 struct inode *parent = d_inode(dentry->d_parent); /* we hold i_rwsem */ 708 709 /* .snap dir? */ 710 if (ceph_snap(parent) == CEPH_NOSNAP && 711 strcmp(dentry->d_name.name, fsc->mount_options->snapdir_name) == 0) { 712 struct dentry *res; 713 struct inode *inode = ceph_get_snapdir(parent); 714 715 res = d_splice_alias(inode, dentry); 716 dout("ENOENT on snapdir %p '%pd', linking to snapdir %p. Spliced dentry %p\n", 717 dentry, dentry, inode, res); 718 if (res) 719 dentry = res; 720 } 721 return dentry; 722} 723 724/* 725 * Figure out final result of a lookup/open request. 726 * 727 * Mainly, make sure we return the final req->r_dentry (if it already 728 * existed) in place of the original VFS-provided dentry when they 729 * differ. 730 * 731 * Gracefully handle the case where the MDS replies with -ENOENT and 732 * no trace (which it may do, at its discretion, e.g., if it doesn't 733 * care to issue a lease on the negative dentry). 734 */ 735struct dentry *ceph_finish_lookup(struct ceph_mds_request *req, 736 struct dentry *dentry, int err) 737{ 738 if (err == -ENOENT) { 739 /* no trace? */ 740 err = 0; 741 if (!req->r_reply_info.head->is_dentry) { 742 dout("ENOENT and no trace, dentry %p inode %p\n", 743 dentry, d_inode(dentry)); 744 if (d_really_is_positive(dentry)) { 745 d_drop(dentry); 746 err = -ENOENT; 747 } else { 748 d_add(dentry, NULL); 749 } 750 } 751 } 752 if (err) 753 dentry = ERR_PTR(err); 754 else if (dentry != req->r_dentry) 755 dentry = dget(req->r_dentry); /* we got spliced */ 756 else 757 dentry = NULL; 758 return dentry; 759} 760 761static bool is_root_ceph_dentry(struct inode *inode, struct dentry *dentry) 762{ 763 return ceph_ino(inode) == CEPH_INO_ROOT && 764 strncmp(dentry->d_name.name, ".ceph", 5) == 0; 765} 766 767/* 768 * Look up a single dir entry. If there is a lookup intent, inform 769 * the MDS so that it gets our 'caps wanted' value in a single op. 770 */ 771static struct dentry *ceph_lookup(struct inode *dir, struct dentry *dentry, 772 unsigned int flags) 773{ 774 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb); 775 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(dir->i_sb); 776 struct ceph_mds_request *req; 777 int op; 778 int mask; 779 int err; 780 781 dout("lookup %p dentry %p '%pd'\n", 782 dir, dentry, dentry); 783 784 if (dentry->d_name.len > NAME_MAX) 785 return ERR_PTR(-ENAMETOOLONG); 786 787 if (IS_ENCRYPTED(dir)) { 788 bool had_key = fscrypt_has_encryption_key(dir); 789 790 err = fscrypt_prepare_lookup_partial(dir, dentry); 791 if (err < 0) 792 return ERR_PTR(err); 793 794 /* mark directory as incomplete if it has been unlocked */ 795 if (!had_key && fscrypt_has_encryption_key(dir)) 796 ceph_dir_clear_complete(dir); 797 } 798 799 /* can we conclude ENOENT locally? */ 800 if (d_really_is_negative(dentry)) { 801 struct ceph_inode_info *ci = ceph_inode(dir); 802 struct ceph_dentry_info *di = ceph_dentry(dentry); 803 804 spin_lock(&ci->i_ceph_lock); 805 dout(" dir %p flags are 0x%lx\n", dir, ci->i_ceph_flags); 806 if (strncmp(dentry->d_name.name, 807 fsc->mount_options->snapdir_name, 808 dentry->d_name.len) && 809 !is_root_ceph_dentry(dir, dentry) && 810 ceph_test_mount_opt(fsc, DCACHE) && 811 __ceph_dir_is_complete(ci) && 812 __ceph_caps_issued_mask_metric(ci, CEPH_CAP_FILE_SHARED, 1)) { 813 __ceph_touch_fmode(ci, mdsc, CEPH_FILE_MODE_RD); 814 spin_unlock(&ci->i_ceph_lock); 815 dout(" dir %p complete, -ENOENT\n", dir); 816 d_add(dentry, NULL); 817 di->lease_shared_gen = atomic_read(&ci->i_shared_gen); 818 return NULL; 819 } 820 spin_unlock(&ci->i_ceph_lock); 821 } 822 823 op = ceph_snap(dir) == CEPH_SNAPDIR ? 824 CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP; 825 req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS); 826 if (IS_ERR(req)) 827 return ERR_CAST(req); 828 req->r_dentry = dget(dentry); 829 req->r_num_caps = 2; 830 831 mask = CEPH_STAT_CAP_INODE | CEPH_CAP_AUTH_SHARED; 832 if (ceph_security_xattr_wanted(dir)) 833 mask |= CEPH_CAP_XATTR_SHARED; 834 req->r_args.getattr.mask = cpu_to_le32(mask); 835 836 ihold(dir); 837 req->r_parent = dir; 838 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags); 839 err = ceph_mdsc_do_request(mdsc, NULL, req); 840 if (err == -ENOENT) { 841 struct dentry *res; 842 843 res = ceph_handle_snapdir(req, dentry); 844 if (IS_ERR(res)) { 845 err = PTR_ERR(res); 846 } else { 847 dentry = res; 848 err = 0; 849 } 850 } 851 dentry = ceph_finish_lookup(req, dentry, err); 852 ceph_mdsc_put_request(req); /* will dput(dentry) */ 853 dout("lookup result=%p\n", dentry); 854 return dentry; 855} 856 857/* 858 * If we do a create but get no trace back from the MDS, follow up with 859 * a lookup (the VFS expects us to link up the provided dentry). 860 */ 861int ceph_handle_notrace_create(struct inode *dir, struct dentry *dentry) 862{ 863 struct dentry *result = ceph_lookup(dir, dentry, 0); 864 865 if (result && !IS_ERR(result)) { 866 /* 867 * We created the item, then did a lookup, and found 868 * it was already linked to another inode we already 869 * had in our cache (and thus got spliced). To not 870 * confuse VFS (especially when inode is a directory), 871 * we don't link our dentry to that inode, return an 872 * error instead. 873 * 874 * This event should be rare and it happens only when 875 * we talk to old MDS. Recent MDS does not send traceless 876 * reply for request that creates new inode. 877 */ 878 d_drop(result); 879 return -ESTALE; 880 } 881 return PTR_ERR(result); 882} 883 884static int ceph_mknod(struct mnt_idmap *idmap, struct inode *dir, 885 struct dentry *dentry, umode_t mode, dev_t rdev) 886{ 887 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(dir->i_sb); 888 struct ceph_mds_request *req; 889 struct ceph_acl_sec_ctx as_ctx = {}; 890 int err; 891 892 if (ceph_snap(dir) != CEPH_NOSNAP) 893 return -EROFS; 894 895 err = ceph_wait_on_conflict_unlink(dentry); 896 if (err) 897 return err; 898 899 if (ceph_quota_is_max_files_exceeded(dir)) { 900 err = -EDQUOT; 901 goto out; 902 } 903 904 dout("mknod in dir %p dentry %p mode 0%ho rdev %d\n", 905 dir, dentry, mode, rdev); 906 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_MKNOD, USE_AUTH_MDS); 907 if (IS_ERR(req)) { 908 err = PTR_ERR(req); 909 goto out; 910 } 911 912 req->r_new_inode = ceph_new_inode(dir, dentry, &mode, &as_ctx); 913 if (IS_ERR(req->r_new_inode)) { 914 err = PTR_ERR(req->r_new_inode); 915 req->r_new_inode = NULL; 916 goto out_req; 917 } 918 919 if (S_ISREG(mode) && IS_ENCRYPTED(dir)) 920 set_bit(CEPH_MDS_R_FSCRYPT_FILE, &req->r_req_flags); 921 922 req->r_dentry = dget(dentry); 923 req->r_num_caps = 2; 924 req->r_parent = dir; 925 ihold(dir); 926 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags); 927 req->r_args.mknod.mode = cpu_to_le32(mode); 928 req->r_args.mknod.rdev = cpu_to_le32(rdev); 929 req->r_dentry_drop = CEPH_CAP_FILE_SHARED | CEPH_CAP_AUTH_EXCL | 930 CEPH_CAP_XATTR_EXCL; 931 req->r_dentry_unless = CEPH_CAP_FILE_EXCL; 932 933 ceph_as_ctx_to_req(req, &as_ctx); 934 935 err = ceph_mdsc_do_request(mdsc, dir, req); 936 if (!err && !req->r_reply_info.head->is_dentry) 937 err = ceph_handle_notrace_create(dir, dentry); 938out_req: 939 ceph_mdsc_put_request(req); 940out: 941 if (!err) 942 ceph_init_inode_acls(d_inode(dentry), &as_ctx); 943 else 944 d_drop(dentry); 945 ceph_release_acl_sec_ctx(&as_ctx); 946 return err; 947} 948 949static int ceph_create(struct mnt_idmap *idmap, struct inode *dir, 950 struct dentry *dentry, umode_t mode, bool excl) 951{ 952 return ceph_mknod(idmap, dir, dentry, mode, 0); 953} 954 955#if IS_ENABLED(CONFIG_FS_ENCRYPTION) 956static int prep_encrypted_symlink_target(struct ceph_mds_request *req, 957 const char *dest) 958{ 959 int err; 960 int len = strlen(dest); 961 struct fscrypt_str osd_link = FSTR_INIT(NULL, 0); 962 963 err = fscrypt_prepare_symlink(req->r_parent, dest, len, PATH_MAX, 964 &osd_link); 965 if (err) 966 goto out; 967 968 err = fscrypt_encrypt_symlink(req->r_new_inode, dest, len, &osd_link); 969 if (err) 970 goto out; 971 972 req->r_path2 = kmalloc(CEPH_BASE64_CHARS(osd_link.len) + 1, GFP_KERNEL); 973 if (!req->r_path2) { 974 err = -ENOMEM; 975 goto out; 976 } 977 978 len = ceph_base64_encode(osd_link.name, osd_link.len, req->r_path2); 979 req->r_path2[len] = '\0'; 980out: 981 fscrypt_fname_free_buffer(&osd_link); 982 return err; 983} 984#else 985static int prep_encrypted_symlink_target(struct ceph_mds_request *req, 986 const char *dest) 987{ 988 return -EOPNOTSUPP; 989} 990#endif 991 992static int ceph_symlink(struct mnt_idmap *idmap, struct inode *dir, 993 struct dentry *dentry, const char *dest) 994{ 995 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(dir->i_sb); 996 struct ceph_mds_request *req; 997 struct ceph_acl_sec_ctx as_ctx = {}; 998 umode_t mode = S_IFLNK | 0777; 999 int err; 1000 1001 if (ceph_snap(dir) != CEPH_NOSNAP) 1002 return -EROFS; 1003 1004 err = ceph_wait_on_conflict_unlink(dentry); 1005 if (err) 1006 return err; 1007 1008 if (ceph_quota_is_max_files_exceeded(dir)) { 1009 err = -EDQUOT; 1010 goto out; 1011 } 1012 1013 dout("symlink in dir %p dentry %p to '%s'\n", dir, dentry, dest); 1014 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SYMLINK, USE_AUTH_MDS); 1015 if (IS_ERR(req)) { 1016 err = PTR_ERR(req); 1017 goto out; 1018 } 1019 1020 req->r_new_inode = ceph_new_inode(dir, dentry, &mode, &as_ctx); 1021 if (IS_ERR(req->r_new_inode)) { 1022 err = PTR_ERR(req->r_new_inode); 1023 req->r_new_inode = NULL; 1024 goto out_req; 1025 } 1026 1027 req->r_parent = dir; 1028 ihold(dir); 1029 1030 if (IS_ENCRYPTED(req->r_new_inode)) { 1031 err = prep_encrypted_symlink_target(req, dest); 1032 if (err) 1033 goto out_req; 1034 } else { 1035 req->r_path2 = kstrdup(dest, GFP_KERNEL); 1036 if (!req->r_path2) { 1037 err = -ENOMEM; 1038 goto out_req; 1039 } 1040 } 1041 1042 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags); 1043 req->r_dentry = dget(dentry); 1044 req->r_num_caps = 2; 1045 req->r_dentry_drop = CEPH_CAP_FILE_SHARED | CEPH_CAP_AUTH_EXCL | 1046 CEPH_CAP_XATTR_EXCL; 1047 req->r_dentry_unless = CEPH_CAP_FILE_EXCL; 1048 1049 ceph_as_ctx_to_req(req, &as_ctx); 1050 1051 err = ceph_mdsc_do_request(mdsc, dir, req); 1052 if (!err && !req->r_reply_info.head->is_dentry) 1053 err = ceph_handle_notrace_create(dir, dentry); 1054out_req: 1055 ceph_mdsc_put_request(req); 1056out: 1057 if (err) 1058 d_drop(dentry); 1059 ceph_release_acl_sec_ctx(&as_ctx); 1060 return err; 1061} 1062 1063static int ceph_mkdir(struct mnt_idmap *idmap, struct inode *dir, 1064 struct dentry *dentry, umode_t mode) 1065{ 1066 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(dir->i_sb); 1067 struct ceph_mds_request *req; 1068 struct ceph_acl_sec_ctx as_ctx = {}; 1069 int err; 1070 int op; 1071 1072 err = ceph_wait_on_conflict_unlink(dentry); 1073 if (err) 1074 return err; 1075 1076 if (ceph_snap(dir) == CEPH_SNAPDIR) { 1077 /* mkdir .snap/foo is a MKSNAP */ 1078 op = CEPH_MDS_OP_MKSNAP; 1079 dout("mksnap dir %p snap '%pd' dn %p\n", dir, 1080 dentry, dentry); 1081 } else if (ceph_snap(dir) == CEPH_NOSNAP) { 1082 dout("mkdir dir %p dn %p mode 0%ho\n", dir, dentry, mode); 1083 op = CEPH_MDS_OP_MKDIR; 1084 } else { 1085 err = -EROFS; 1086 goto out; 1087 } 1088 1089 if (op == CEPH_MDS_OP_MKDIR && 1090 ceph_quota_is_max_files_exceeded(dir)) { 1091 err = -EDQUOT; 1092 goto out; 1093 } 1094 if ((op == CEPH_MDS_OP_MKSNAP) && IS_ENCRYPTED(dir) && 1095 !fscrypt_has_encryption_key(dir)) { 1096 err = -ENOKEY; 1097 goto out; 1098 } 1099 1100 1101 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS); 1102 if (IS_ERR(req)) { 1103 err = PTR_ERR(req); 1104 goto out; 1105 } 1106 1107 mode |= S_IFDIR; 1108 req->r_new_inode = ceph_new_inode(dir, dentry, &mode, &as_ctx); 1109 if (IS_ERR(req->r_new_inode)) { 1110 err = PTR_ERR(req->r_new_inode); 1111 req->r_new_inode = NULL; 1112 goto out_req; 1113 } 1114 1115 req->r_dentry = dget(dentry); 1116 req->r_num_caps = 2; 1117 req->r_parent = dir; 1118 ihold(dir); 1119 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags); 1120 req->r_args.mkdir.mode = cpu_to_le32(mode); 1121 req->r_dentry_drop = CEPH_CAP_FILE_SHARED | CEPH_CAP_AUTH_EXCL | 1122 CEPH_CAP_XATTR_EXCL; 1123 req->r_dentry_unless = CEPH_CAP_FILE_EXCL; 1124 1125 ceph_as_ctx_to_req(req, &as_ctx); 1126 1127 err = ceph_mdsc_do_request(mdsc, dir, req); 1128 if (!err && 1129 !req->r_reply_info.head->is_target && 1130 !req->r_reply_info.head->is_dentry) 1131 err = ceph_handle_notrace_create(dir, dentry); 1132out_req: 1133 ceph_mdsc_put_request(req); 1134out: 1135 if (!err) 1136 ceph_init_inode_acls(d_inode(dentry), &as_ctx); 1137 else 1138 d_drop(dentry); 1139 ceph_release_acl_sec_ctx(&as_ctx); 1140 return err; 1141} 1142 1143static int ceph_link(struct dentry *old_dentry, struct inode *dir, 1144 struct dentry *dentry) 1145{ 1146 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(dir->i_sb); 1147 struct ceph_mds_request *req; 1148 int err; 1149 1150 if (dentry->d_flags & DCACHE_DISCONNECTED) 1151 return -EINVAL; 1152 1153 err = ceph_wait_on_conflict_unlink(dentry); 1154 if (err) 1155 return err; 1156 1157 if (ceph_snap(dir) != CEPH_NOSNAP) 1158 return -EROFS; 1159 1160 err = fscrypt_prepare_link(old_dentry, dir, dentry); 1161 if (err) 1162 return err; 1163 1164 dout("link in dir %p %llx.%llx old_dentry %p:'%pd' dentry %p:'%pd'\n", 1165 dir, ceph_vinop(dir), old_dentry, old_dentry, dentry, dentry); 1166 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LINK, USE_AUTH_MDS); 1167 if (IS_ERR(req)) { 1168 d_drop(dentry); 1169 return PTR_ERR(req); 1170 } 1171 req->r_dentry = dget(dentry); 1172 req->r_num_caps = 2; 1173 req->r_old_dentry = dget(old_dentry); 1174 /* 1175 * The old_dentry maybe a DCACHE_DISCONNECTED dentry, then we 1176 * will just pass the ino# to MDSs. 1177 */ 1178 if (old_dentry->d_flags & DCACHE_DISCONNECTED) 1179 req->r_ino2 = ceph_vino(d_inode(old_dentry)); 1180 req->r_parent = dir; 1181 ihold(dir); 1182 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags); 1183 req->r_dentry_drop = CEPH_CAP_FILE_SHARED | CEPH_CAP_XATTR_EXCL; 1184 req->r_dentry_unless = CEPH_CAP_FILE_EXCL; 1185 /* release LINK_SHARED on source inode (mds will lock it) */ 1186 req->r_old_inode_drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL; 1187 err = ceph_mdsc_do_request(mdsc, dir, req); 1188 if (err) { 1189 d_drop(dentry); 1190 } else if (!req->r_reply_info.head->is_dentry) { 1191 ihold(d_inode(old_dentry)); 1192 d_instantiate(dentry, d_inode(old_dentry)); 1193 } 1194 ceph_mdsc_put_request(req); 1195 return err; 1196} 1197 1198static void ceph_async_unlink_cb(struct ceph_mds_client *mdsc, 1199 struct ceph_mds_request *req) 1200{ 1201 struct dentry *dentry = req->r_dentry; 1202 struct ceph_fs_client *fsc = ceph_sb_to_client(dentry->d_sb); 1203 struct ceph_dentry_info *di = ceph_dentry(dentry); 1204 int result = req->r_err ? req->r_err : 1205 le32_to_cpu(req->r_reply_info.head->result); 1206 1207 if (!test_bit(CEPH_DENTRY_ASYNC_UNLINK_BIT, &di->flags)) 1208 pr_warn("%s dentry %p:%pd async unlink bit is not set\n", 1209 __func__, dentry, dentry); 1210 1211 spin_lock(&fsc->async_unlink_conflict_lock); 1212 hash_del_rcu(&di->hnode); 1213 spin_unlock(&fsc->async_unlink_conflict_lock); 1214 1215 spin_lock(&dentry->d_lock); 1216 di->flags &= ~CEPH_DENTRY_ASYNC_UNLINK; 1217 wake_up_bit(&di->flags, CEPH_DENTRY_ASYNC_UNLINK_BIT); 1218 spin_unlock(&dentry->d_lock); 1219 1220 synchronize_rcu(); 1221 1222 if (result == -EJUKEBOX) 1223 goto out; 1224 1225 /* If op failed, mark everyone involved for errors */ 1226 if (result) { 1227 int pathlen = 0; 1228 u64 base = 0; 1229 char *path = ceph_mdsc_build_path(dentry, &pathlen, 1230 &base, 0); 1231 1232 /* mark error on parent + clear complete */ 1233 mapping_set_error(req->r_parent->i_mapping, result); 1234 ceph_dir_clear_complete(req->r_parent); 1235 1236 /* drop the dentry -- we don't know its status */ 1237 if (!d_unhashed(dentry)) 1238 d_drop(dentry); 1239 1240 /* mark inode itself for an error (since metadata is bogus) */ 1241 mapping_set_error(req->r_old_inode->i_mapping, result); 1242 1243 pr_warn("async unlink failure path=(%llx)%s result=%d!\n", 1244 base, IS_ERR(path) ? "<<bad>>" : path, result); 1245 ceph_mdsc_free_path(path, pathlen); 1246 } 1247out: 1248 iput(req->r_old_inode); 1249 ceph_mdsc_release_dir_caps(req); 1250} 1251 1252static int get_caps_for_async_unlink(struct inode *dir, struct dentry *dentry) 1253{ 1254 struct ceph_inode_info *ci = ceph_inode(dir); 1255 struct ceph_dentry_info *di; 1256 int got = 0, want = CEPH_CAP_FILE_EXCL | CEPH_CAP_DIR_UNLINK; 1257 1258 spin_lock(&ci->i_ceph_lock); 1259 if ((__ceph_caps_issued(ci, NULL) & want) == want) { 1260 ceph_take_cap_refs(ci, want, false); 1261 got = want; 1262 } 1263 spin_unlock(&ci->i_ceph_lock); 1264 1265 /* If we didn't get anything, return 0 */ 1266 if (!got) 1267 return 0; 1268 1269 spin_lock(&dentry->d_lock); 1270 di = ceph_dentry(dentry); 1271 /* 1272 * - We are holding Fx, which implies Fs caps. 1273 * - Only support async unlink for primary linkage 1274 */ 1275 if (atomic_read(&ci->i_shared_gen) != di->lease_shared_gen || 1276 !(di->flags & CEPH_DENTRY_PRIMARY_LINK)) 1277 want = 0; 1278 spin_unlock(&dentry->d_lock); 1279 1280 /* Do we still want what we've got? */ 1281 if (want == got) 1282 return got; 1283 1284 ceph_put_cap_refs(ci, got); 1285 return 0; 1286} 1287 1288/* 1289 * rmdir and unlink are differ only by the metadata op code 1290 */ 1291static int ceph_unlink(struct inode *dir, struct dentry *dentry) 1292{ 1293 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb); 1294 struct ceph_mds_client *mdsc = fsc->mdsc; 1295 struct inode *inode = d_inode(dentry); 1296 struct ceph_mds_request *req; 1297 bool try_async = ceph_test_mount_opt(fsc, ASYNC_DIROPS); 1298 int err = -EROFS; 1299 int op; 1300 1301 if (ceph_snap(dir) == CEPH_SNAPDIR) { 1302 /* rmdir .snap/foo is RMSNAP */ 1303 dout("rmsnap dir %p '%pd' dn %p\n", dir, dentry, dentry); 1304 op = CEPH_MDS_OP_RMSNAP; 1305 } else if (ceph_snap(dir) == CEPH_NOSNAP) { 1306 dout("unlink/rmdir dir %p dn %p inode %p\n", 1307 dir, dentry, inode); 1308 op = d_is_dir(dentry) ? 1309 CEPH_MDS_OP_RMDIR : CEPH_MDS_OP_UNLINK; 1310 } else 1311 goto out; 1312retry: 1313 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS); 1314 if (IS_ERR(req)) { 1315 err = PTR_ERR(req); 1316 goto out; 1317 } 1318 req->r_dentry = dget(dentry); 1319 req->r_num_caps = 2; 1320 req->r_parent = dir; 1321 ihold(dir); 1322 req->r_dentry_drop = CEPH_CAP_FILE_SHARED | CEPH_CAP_XATTR_EXCL; 1323 req->r_dentry_unless = CEPH_CAP_FILE_EXCL; 1324 req->r_inode_drop = ceph_drop_caps_for_unlink(inode); 1325 1326 if (try_async && op == CEPH_MDS_OP_UNLINK && 1327 (req->r_dir_caps = get_caps_for_async_unlink(dir, dentry))) { 1328 struct ceph_dentry_info *di = ceph_dentry(dentry); 1329 1330 dout("async unlink on %llu/%.*s caps=%s", ceph_ino(dir), 1331 dentry->d_name.len, dentry->d_name.name, 1332 ceph_cap_string(req->r_dir_caps)); 1333 set_bit(CEPH_MDS_R_ASYNC, &req->r_req_flags); 1334 req->r_callback = ceph_async_unlink_cb; 1335 req->r_old_inode = d_inode(dentry); 1336 ihold(req->r_old_inode); 1337 1338 spin_lock(&dentry->d_lock); 1339 di->flags |= CEPH_DENTRY_ASYNC_UNLINK; 1340 spin_unlock(&dentry->d_lock); 1341 1342 spin_lock(&fsc->async_unlink_conflict_lock); 1343 hash_add_rcu(fsc->async_unlink_conflict, &di->hnode, 1344 dentry->d_name.hash); 1345 spin_unlock(&fsc->async_unlink_conflict_lock); 1346 1347 err = ceph_mdsc_submit_request(mdsc, dir, req); 1348 if (!err) { 1349 /* 1350 * We have enough caps, so we assume that the unlink 1351 * will succeed. Fix up the target inode and dcache. 1352 */ 1353 drop_nlink(inode); 1354 d_delete(dentry); 1355 } else { 1356 spin_lock(&fsc->async_unlink_conflict_lock); 1357 hash_del_rcu(&di->hnode); 1358 spin_unlock(&fsc->async_unlink_conflict_lock); 1359 1360 spin_lock(&dentry->d_lock); 1361 di->flags &= ~CEPH_DENTRY_ASYNC_UNLINK; 1362 spin_unlock(&dentry->d_lock); 1363 1364 if (err == -EJUKEBOX) { 1365 try_async = false; 1366 ceph_mdsc_put_request(req); 1367 goto retry; 1368 } 1369 } 1370 } else { 1371 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags); 1372 err = ceph_mdsc_do_request(mdsc, dir, req); 1373 if (!err && !req->r_reply_info.head->is_dentry) 1374 d_delete(dentry); 1375 } 1376 1377 ceph_mdsc_put_request(req); 1378out: 1379 return err; 1380} 1381 1382static int ceph_rename(struct mnt_idmap *idmap, struct inode *old_dir, 1383 struct dentry *old_dentry, struct inode *new_dir, 1384 struct dentry *new_dentry, unsigned int flags) 1385{ 1386 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(old_dir->i_sb); 1387 struct ceph_mds_request *req; 1388 int op = CEPH_MDS_OP_RENAME; 1389 int err; 1390 1391 if (flags) 1392 return -EINVAL; 1393 1394 if (ceph_snap(old_dir) != ceph_snap(new_dir)) 1395 return -EXDEV; 1396 if (ceph_snap(old_dir) != CEPH_NOSNAP) { 1397 if (old_dir == new_dir && ceph_snap(old_dir) == CEPH_SNAPDIR) 1398 op = CEPH_MDS_OP_RENAMESNAP; 1399 else 1400 return -EROFS; 1401 } 1402 /* don't allow cross-quota renames */ 1403 if ((old_dir != new_dir) && 1404 (!ceph_quota_is_same_realm(old_dir, new_dir))) 1405 return -EXDEV; 1406 1407 err = ceph_wait_on_conflict_unlink(new_dentry); 1408 if (err) 1409 return err; 1410 1411 err = fscrypt_prepare_rename(old_dir, old_dentry, new_dir, new_dentry, 1412 flags); 1413 if (err) 1414 return err; 1415 1416 dout("rename dir %p dentry %p to dir %p dentry %p\n", 1417 old_dir, old_dentry, new_dir, new_dentry); 1418 req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS); 1419 if (IS_ERR(req)) 1420 return PTR_ERR(req); 1421 ihold(old_dir); 1422 req->r_dentry = dget(new_dentry); 1423 req->r_num_caps = 2; 1424 req->r_old_dentry = dget(old_dentry); 1425 req->r_old_dentry_dir = old_dir; 1426 req->r_parent = new_dir; 1427 ihold(new_dir); 1428 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags); 1429 req->r_old_dentry_drop = CEPH_CAP_FILE_SHARED | CEPH_CAP_XATTR_EXCL; 1430 req->r_old_dentry_unless = CEPH_CAP_FILE_EXCL; 1431 req->r_dentry_drop = CEPH_CAP_FILE_SHARED | CEPH_CAP_XATTR_EXCL; 1432 req->r_dentry_unless = CEPH_CAP_FILE_EXCL; 1433 /* release LINK_RDCACHE on source inode (mds will lock it) */ 1434 req->r_old_inode_drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL; 1435 if (d_really_is_positive(new_dentry)) { 1436 req->r_inode_drop = 1437 ceph_drop_caps_for_unlink(d_inode(new_dentry)); 1438 } 1439 err = ceph_mdsc_do_request(mdsc, old_dir, req); 1440 if (!err && !req->r_reply_info.head->is_dentry) { 1441 /* 1442 * Normally d_move() is done by fill_trace (called by 1443 * do_request, above). If there is no trace, we need 1444 * to do it here. 1445 */ 1446 d_move(old_dentry, new_dentry); 1447 } 1448 ceph_mdsc_put_request(req); 1449 return err; 1450} 1451 1452/* 1453 * Move dentry to tail of mdsc->dentry_leases list when lease is updated. 1454 * Leases at front of the list will expire first. (Assume all leases have 1455 * similar duration) 1456 * 1457 * Called under dentry->d_lock. 1458 */ 1459void __ceph_dentry_lease_touch(struct ceph_dentry_info *di) 1460{ 1461 struct dentry *dn = di->dentry; 1462 struct ceph_mds_client *mdsc; 1463 1464 dout("dentry_lease_touch %p %p '%pd'\n", di, dn, dn); 1465 1466 di->flags |= CEPH_DENTRY_LEASE_LIST; 1467 if (di->flags & CEPH_DENTRY_SHRINK_LIST) { 1468 di->flags |= CEPH_DENTRY_REFERENCED; 1469 return; 1470 } 1471 1472 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc; 1473 spin_lock(&mdsc->dentry_list_lock); 1474 list_move_tail(&di->lease_list, &mdsc->dentry_leases); 1475 spin_unlock(&mdsc->dentry_list_lock); 1476} 1477 1478static void __dentry_dir_lease_touch(struct ceph_mds_client* mdsc, 1479 struct ceph_dentry_info *di) 1480{ 1481 di->flags &= ~(CEPH_DENTRY_LEASE_LIST | CEPH_DENTRY_REFERENCED); 1482 di->lease_gen = 0; 1483 di->time = jiffies; 1484 list_move_tail(&di->lease_list, &mdsc->dentry_dir_leases); 1485} 1486 1487/* 1488 * When dir lease is used, add dentry to tail of mdsc->dentry_dir_leases 1489 * list if it's not in the list, otherwise set 'referenced' flag. 1490 * 1491 * Called under dentry->d_lock. 1492 */ 1493void __ceph_dentry_dir_lease_touch(struct ceph_dentry_info *di) 1494{ 1495 struct dentry *dn = di->dentry; 1496 struct ceph_mds_client *mdsc; 1497 1498 dout("dentry_dir_lease_touch %p %p '%pd' (offset 0x%llx)\n", 1499 di, dn, dn, di->offset); 1500 1501 if (!list_empty(&di->lease_list)) { 1502 if (di->flags & CEPH_DENTRY_LEASE_LIST) { 1503 /* don't remove dentry from dentry lease list 1504 * if its lease is valid */ 1505 if (__dentry_lease_is_valid(di)) 1506 return; 1507 } else { 1508 di->flags |= CEPH_DENTRY_REFERENCED; 1509 return; 1510 } 1511 } 1512 1513 if (di->flags & CEPH_DENTRY_SHRINK_LIST) { 1514 di->flags |= CEPH_DENTRY_REFERENCED; 1515 di->flags &= ~CEPH_DENTRY_LEASE_LIST; 1516 return; 1517 } 1518 1519 mdsc = ceph_sb_to_client(dn->d_sb)->mdsc; 1520 spin_lock(&mdsc->dentry_list_lock); 1521 __dentry_dir_lease_touch(mdsc, di), 1522 spin_unlock(&mdsc->dentry_list_lock); 1523} 1524 1525static void __dentry_lease_unlist(struct ceph_dentry_info *di) 1526{ 1527 struct ceph_mds_client *mdsc; 1528 if (di->flags & CEPH_DENTRY_SHRINK_LIST) 1529 return; 1530 if (list_empty(&di->lease_list)) 1531 return; 1532 1533 mdsc = ceph_sb_to_client(di->dentry->d_sb)->mdsc; 1534 spin_lock(&mdsc->dentry_list_lock); 1535 list_del_init(&di->lease_list); 1536 spin_unlock(&mdsc->dentry_list_lock); 1537} 1538 1539enum { 1540 KEEP = 0, 1541 DELETE = 1, 1542 TOUCH = 2, 1543 STOP = 4, 1544}; 1545 1546struct ceph_lease_walk_control { 1547 bool dir_lease; 1548 bool expire_dir_lease; 1549 unsigned long nr_to_scan; 1550 unsigned long dir_lease_ttl; 1551}; 1552 1553static unsigned long 1554__dentry_leases_walk(struct ceph_mds_client *mdsc, 1555 struct ceph_lease_walk_control *lwc, 1556 int (*check)(struct dentry*, void*)) 1557{ 1558 struct ceph_dentry_info *di, *tmp; 1559 struct dentry *dentry, *last = NULL; 1560 struct list_head* list; 1561 LIST_HEAD(dispose); 1562 unsigned long freed = 0; 1563 int ret = 0; 1564 1565 list = lwc->dir_lease ? &mdsc->dentry_dir_leases : &mdsc->dentry_leases; 1566 spin_lock(&mdsc->dentry_list_lock); 1567 list_for_each_entry_safe(di, tmp, list, lease_list) { 1568 if (!lwc->nr_to_scan) 1569 break; 1570 --lwc->nr_to_scan; 1571 1572 dentry = di->dentry; 1573 if (last == dentry) 1574 break; 1575 1576 if (!spin_trylock(&dentry->d_lock)) 1577 continue; 1578 1579 if (__lockref_is_dead(&dentry->d_lockref)) { 1580 list_del_init(&di->lease_list); 1581 goto next; 1582 } 1583 1584 ret = check(dentry, lwc); 1585 if (ret & TOUCH) { 1586 /* move it into tail of dir lease list */ 1587 __dentry_dir_lease_touch(mdsc, di); 1588 if (!last) 1589 last = dentry; 1590 } 1591 if (ret & DELETE) { 1592 /* stale lease */ 1593 di->flags &= ~CEPH_DENTRY_REFERENCED; 1594 if (dentry->d_lockref.count > 0) { 1595 /* update_dentry_lease() will re-add 1596 * it to lease list, or 1597 * ceph_d_delete() will return 1 when 1598 * last reference is dropped */ 1599 list_del_init(&di->lease_list); 1600 } else { 1601 di->flags |= CEPH_DENTRY_SHRINK_LIST; 1602 list_move_tail(&di->lease_list, &dispose); 1603 dget_dlock(dentry); 1604 } 1605 } 1606next: 1607 spin_unlock(&dentry->d_lock); 1608 if (ret & STOP) 1609 break; 1610 } 1611 spin_unlock(&mdsc->dentry_list_lock); 1612 1613 while (!list_empty(&dispose)) { 1614 di = list_first_entry(&dispose, struct ceph_dentry_info, 1615 lease_list); 1616 dentry = di->dentry; 1617 spin_lock(&dentry->d_lock); 1618 1619 list_del_init(&di->lease_list); 1620 di->flags &= ~CEPH_DENTRY_SHRINK_LIST; 1621 if (di->flags & CEPH_DENTRY_REFERENCED) { 1622 spin_lock(&mdsc->dentry_list_lock); 1623 if (di->flags & CEPH_DENTRY_LEASE_LIST) { 1624 list_add_tail(&di->lease_list, 1625 &mdsc->dentry_leases); 1626 } else { 1627 __dentry_dir_lease_touch(mdsc, di); 1628 } 1629 spin_unlock(&mdsc->dentry_list_lock); 1630 } else { 1631 freed++; 1632 } 1633 1634 spin_unlock(&dentry->d_lock); 1635 /* ceph_d_delete() does the trick */ 1636 dput(dentry); 1637 } 1638 return freed; 1639} 1640 1641static int __dentry_lease_check(struct dentry *dentry, void *arg) 1642{ 1643 struct ceph_dentry_info *di = ceph_dentry(dentry); 1644 int ret; 1645 1646 if (__dentry_lease_is_valid(di)) 1647 return STOP; 1648 ret = __dir_lease_try_check(dentry); 1649 if (ret == -EBUSY) 1650 return KEEP; 1651 if (ret > 0) 1652 return TOUCH; 1653 return DELETE; 1654} 1655 1656static int __dir_lease_check(struct dentry *dentry, void *arg) 1657{ 1658 struct ceph_lease_walk_control *lwc = arg; 1659 struct ceph_dentry_info *di = ceph_dentry(dentry); 1660 1661 int ret = __dir_lease_try_check(dentry); 1662 if (ret == -EBUSY) 1663 return KEEP; 1664 if (ret > 0) { 1665 if (time_before(jiffies, di->time + lwc->dir_lease_ttl)) 1666 return STOP; 1667 /* Move dentry to tail of dir lease list if we don't want 1668 * to delete it. So dentries in the list are checked in a 1669 * round robin manner */ 1670 if (!lwc->expire_dir_lease) 1671 return TOUCH; 1672 if (dentry->d_lockref.count > 0 || 1673 (di->flags & CEPH_DENTRY_REFERENCED)) 1674 return TOUCH; 1675 /* invalidate dir lease */ 1676 di->lease_shared_gen = 0; 1677 } 1678 return DELETE; 1679} 1680 1681int ceph_trim_dentries(struct ceph_mds_client *mdsc) 1682{ 1683 struct ceph_lease_walk_control lwc; 1684 unsigned long count; 1685 unsigned long freed; 1686 1687 spin_lock(&mdsc->caps_list_lock); 1688 if (mdsc->caps_use_max > 0 && 1689 mdsc->caps_use_count > mdsc->caps_use_max) 1690 count = mdsc->caps_use_count - mdsc->caps_use_max; 1691 else 1692 count = 0; 1693 spin_unlock(&mdsc->caps_list_lock); 1694 1695 lwc.dir_lease = false; 1696 lwc.nr_to_scan = CEPH_CAPS_PER_RELEASE * 2; 1697 freed = __dentry_leases_walk(mdsc, &lwc, __dentry_lease_check); 1698 if (!lwc.nr_to_scan) /* more invalid leases */ 1699 return -EAGAIN; 1700 1701 if (lwc.nr_to_scan < CEPH_CAPS_PER_RELEASE) 1702 lwc.nr_to_scan = CEPH_CAPS_PER_RELEASE; 1703 1704 lwc.dir_lease = true; 1705 lwc.expire_dir_lease = freed < count; 1706 lwc.dir_lease_ttl = mdsc->fsc->mount_options->caps_wanted_delay_max * HZ; 1707 freed +=__dentry_leases_walk(mdsc, &lwc, __dir_lease_check); 1708 if (!lwc.nr_to_scan) /* more to check */ 1709 return -EAGAIN; 1710 1711 return freed > 0 ? 1 : 0; 1712} 1713 1714/* 1715 * Ensure a dentry lease will no longer revalidate. 1716 */ 1717void ceph_invalidate_dentry_lease(struct dentry *dentry) 1718{ 1719 struct ceph_dentry_info *di = ceph_dentry(dentry); 1720 spin_lock(&dentry->d_lock); 1721 di->time = jiffies; 1722 di->lease_shared_gen = 0; 1723 di->flags &= ~CEPH_DENTRY_PRIMARY_LINK; 1724 __dentry_lease_unlist(di); 1725 spin_unlock(&dentry->d_lock); 1726} 1727 1728/* 1729 * Check if dentry lease is valid. If not, delete the lease. Try to 1730 * renew if the least is more than half up. 1731 */ 1732static bool __dentry_lease_is_valid(struct ceph_dentry_info *di) 1733{ 1734 struct ceph_mds_session *session; 1735 1736 if (!di->lease_gen) 1737 return false; 1738 1739 session = di->lease_session; 1740 if (session) { 1741 u32 gen; 1742 unsigned long ttl; 1743 1744 gen = atomic_read(&session->s_cap_gen); 1745 ttl = session->s_cap_ttl; 1746 1747 if (di->lease_gen == gen && 1748 time_before(jiffies, ttl) && 1749 time_before(jiffies, di->time)) 1750 return true; 1751 } 1752 di->lease_gen = 0; 1753 return false; 1754} 1755 1756static int dentry_lease_is_valid(struct dentry *dentry, unsigned int flags) 1757{ 1758 struct ceph_dentry_info *di; 1759 struct ceph_mds_session *session = NULL; 1760 u32 seq = 0; 1761 int valid = 0; 1762 1763 spin_lock(&dentry->d_lock); 1764 di = ceph_dentry(dentry); 1765 if (di && __dentry_lease_is_valid(di)) { 1766 valid = 1; 1767 1768 if (di->lease_renew_after && 1769 time_after(jiffies, di->lease_renew_after)) { 1770 /* 1771 * We should renew. If we're in RCU walk mode 1772 * though, we can't do that so just return 1773 * -ECHILD. 1774 */ 1775 if (flags & LOOKUP_RCU) { 1776 valid = -ECHILD; 1777 } else { 1778 session = ceph_get_mds_session(di->lease_session); 1779 seq = di->lease_seq; 1780 di->lease_renew_after = 0; 1781 di->lease_renew_from = jiffies; 1782 } 1783 } 1784 } 1785 spin_unlock(&dentry->d_lock); 1786 1787 if (session) { 1788 ceph_mdsc_lease_send_msg(session, dentry, 1789 CEPH_MDS_LEASE_RENEW, seq); 1790 ceph_put_mds_session(session); 1791 } 1792 dout("dentry_lease_is_valid - dentry %p = %d\n", dentry, valid); 1793 return valid; 1794} 1795 1796/* 1797 * Called under dentry->d_lock. 1798 */ 1799static int __dir_lease_try_check(const struct dentry *dentry) 1800{ 1801 struct ceph_dentry_info *di = ceph_dentry(dentry); 1802 struct inode *dir; 1803 struct ceph_inode_info *ci; 1804 int valid = 0; 1805 1806 if (!di->lease_shared_gen) 1807 return 0; 1808 if (IS_ROOT(dentry)) 1809 return 0; 1810 1811 dir = d_inode(dentry->d_parent); 1812 ci = ceph_inode(dir); 1813 1814 if (spin_trylock(&ci->i_ceph_lock)) { 1815 if (atomic_read(&ci->i_shared_gen) == di->lease_shared_gen && 1816 __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 0)) 1817 valid = 1; 1818 spin_unlock(&ci->i_ceph_lock); 1819 } else { 1820 valid = -EBUSY; 1821 } 1822 1823 if (!valid) 1824 di->lease_shared_gen = 0; 1825 return valid; 1826} 1827 1828/* 1829 * Check if directory-wide content lease/cap is valid. 1830 */ 1831static int dir_lease_is_valid(struct inode *dir, struct dentry *dentry, 1832 struct ceph_mds_client *mdsc) 1833{ 1834 struct ceph_inode_info *ci = ceph_inode(dir); 1835 int valid; 1836 int shared_gen; 1837 1838 spin_lock(&ci->i_ceph_lock); 1839 valid = __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1); 1840 if (valid) { 1841 __ceph_touch_fmode(ci, mdsc, CEPH_FILE_MODE_RD); 1842 shared_gen = atomic_read(&ci->i_shared_gen); 1843 } 1844 spin_unlock(&ci->i_ceph_lock); 1845 if (valid) { 1846 struct ceph_dentry_info *di; 1847 spin_lock(&dentry->d_lock); 1848 di = ceph_dentry(dentry); 1849 if (dir == d_inode(dentry->d_parent) && 1850 di && di->lease_shared_gen == shared_gen) 1851 __ceph_dentry_dir_lease_touch(di); 1852 else 1853 valid = 0; 1854 spin_unlock(&dentry->d_lock); 1855 } 1856 dout("dir_lease_is_valid dir %p v%u dentry %p = %d\n", 1857 dir, (unsigned)atomic_read(&ci->i_shared_gen), dentry, valid); 1858 return valid; 1859} 1860 1861/* 1862 * Check if cached dentry can be trusted. 1863 */ 1864static int ceph_d_revalidate(struct dentry *dentry, unsigned int flags) 1865{ 1866 int valid = 0; 1867 struct dentry *parent; 1868 struct inode *dir, *inode; 1869 struct ceph_mds_client *mdsc; 1870 1871 valid = fscrypt_d_revalidate(dentry, flags); 1872 if (valid <= 0) 1873 return valid; 1874 1875 if (flags & LOOKUP_RCU) { 1876 parent = READ_ONCE(dentry->d_parent); 1877 dir = d_inode_rcu(parent); 1878 if (!dir) 1879 return -ECHILD; 1880 inode = d_inode_rcu(dentry); 1881 } else { 1882 parent = dget_parent(dentry); 1883 dir = d_inode(parent); 1884 inode = d_inode(dentry); 1885 } 1886 1887 dout("d_revalidate %p '%pd' inode %p offset 0x%llx nokey %d\n", dentry, 1888 dentry, inode, ceph_dentry(dentry)->offset, 1889 !!(dentry->d_flags & DCACHE_NOKEY_NAME)); 1890 1891 mdsc = ceph_sb_to_client(dir->i_sb)->mdsc; 1892 1893 /* always trust cached snapped dentries, snapdir dentry */ 1894 if (ceph_snap(dir) != CEPH_NOSNAP) { 1895 dout("d_revalidate %p '%pd' inode %p is SNAPPED\n", dentry, 1896 dentry, inode); 1897 valid = 1; 1898 } else if (inode && ceph_snap(inode) == CEPH_SNAPDIR) { 1899 valid = 1; 1900 } else { 1901 valid = dentry_lease_is_valid(dentry, flags); 1902 if (valid == -ECHILD) 1903 return valid; 1904 if (valid || dir_lease_is_valid(dir, dentry, mdsc)) { 1905 if (inode) 1906 valid = ceph_is_any_caps(inode); 1907 else 1908 valid = 1; 1909 } 1910 } 1911 1912 if (!valid) { 1913 struct ceph_mds_request *req; 1914 int op, err; 1915 u32 mask; 1916 1917 if (flags & LOOKUP_RCU) 1918 return -ECHILD; 1919 1920 percpu_counter_inc(&mdsc->metric.d_lease_mis); 1921 1922 op = ceph_snap(dir) == CEPH_SNAPDIR ? 1923 CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP; 1924 req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS); 1925 if (!IS_ERR(req)) { 1926 req->r_dentry = dget(dentry); 1927 req->r_num_caps = 2; 1928 req->r_parent = dir; 1929 ihold(dir); 1930 1931 mask = CEPH_STAT_CAP_INODE | CEPH_CAP_AUTH_SHARED; 1932 if (ceph_security_xattr_wanted(dir)) 1933 mask |= CEPH_CAP_XATTR_SHARED; 1934 req->r_args.getattr.mask = cpu_to_le32(mask); 1935 1936 err = ceph_mdsc_do_request(mdsc, NULL, req); 1937 switch (err) { 1938 case 0: 1939 if (d_really_is_positive(dentry) && 1940 d_inode(dentry) == req->r_target_inode) 1941 valid = 1; 1942 break; 1943 case -ENOENT: 1944 if (d_really_is_negative(dentry)) 1945 valid = 1; 1946 fallthrough; 1947 default: 1948 break; 1949 } 1950 ceph_mdsc_put_request(req); 1951 dout("d_revalidate %p lookup result=%d\n", 1952 dentry, err); 1953 } 1954 } else { 1955 percpu_counter_inc(&mdsc->metric.d_lease_hit); 1956 } 1957 1958 dout("d_revalidate %p %s\n", dentry, valid ? "valid" : "invalid"); 1959 if (!valid) 1960 ceph_dir_clear_complete(dir); 1961 1962 if (!(flags & LOOKUP_RCU)) 1963 dput(parent); 1964 return valid; 1965} 1966 1967/* 1968 * Delete unused dentry that doesn't have valid lease 1969 * 1970 * Called under dentry->d_lock. 1971 */ 1972static int ceph_d_delete(const struct dentry *dentry) 1973{ 1974 struct ceph_dentry_info *di; 1975 1976 /* won't release caps */ 1977 if (d_really_is_negative(dentry)) 1978 return 0; 1979 if (ceph_snap(d_inode(dentry)) != CEPH_NOSNAP) 1980 return 0; 1981 /* vaild lease? */ 1982 di = ceph_dentry(dentry); 1983 if (di) { 1984 if (__dentry_lease_is_valid(di)) 1985 return 0; 1986 if (__dir_lease_try_check(dentry)) 1987 return 0; 1988 } 1989 return 1; 1990} 1991 1992/* 1993 * Release our ceph_dentry_info. 1994 */ 1995static void ceph_d_release(struct dentry *dentry) 1996{ 1997 struct ceph_dentry_info *di = ceph_dentry(dentry); 1998 struct ceph_fs_client *fsc = ceph_sb_to_client(dentry->d_sb); 1999 2000 dout("d_release %p\n", dentry); 2001 2002 atomic64_dec(&fsc->mdsc->metric.total_dentries); 2003 2004 spin_lock(&dentry->d_lock); 2005 __dentry_lease_unlist(di); 2006 dentry->d_fsdata = NULL; 2007 spin_unlock(&dentry->d_lock); 2008 2009 ceph_put_mds_session(di->lease_session); 2010 kmem_cache_free(ceph_dentry_cachep, di); 2011} 2012 2013/* 2014 * When the VFS prunes a dentry from the cache, we need to clear the 2015 * complete flag on the parent directory. 2016 * 2017 * Called under dentry->d_lock. 2018 */ 2019static void ceph_d_prune(struct dentry *dentry) 2020{ 2021 struct ceph_inode_info *dir_ci; 2022 struct ceph_dentry_info *di; 2023 2024 dout("ceph_d_prune %pd %p\n", dentry, dentry); 2025 2026 /* do we have a valid parent? */ 2027 if (IS_ROOT(dentry)) 2028 return; 2029 2030 /* we hold d_lock, so d_parent is stable */ 2031 dir_ci = ceph_inode(d_inode(dentry->d_parent)); 2032 if (dir_ci->i_vino.snap == CEPH_SNAPDIR) 2033 return; 2034 2035 /* who calls d_delete() should also disable dcache readdir */ 2036 if (d_really_is_negative(dentry)) 2037 return; 2038 2039 /* d_fsdata does not get cleared until d_release */ 2040 if (!d_unhashed(dentry)) { 2041 __ceph_dir_clear_complete(dir_ci); 2042 return; 2043 } 2044 2045 /* Disable dcache readdir just in case that someone called d_drop() 2046 * or d_invalidate(), but MDS didn't revoke CEPH_CAP_FILE_SHARED 2047 * properly (dcache readdir is still enabled) */ 2048 di = ceph_dentry(dentry); 2049 if (di->offset > 0 && 2050 di->lease_shared_gen == atomic_read(&dir_ci->i_shared_gen)) 2051 __ceph_dir_clear_ordered(dir_ci); 2052} 2053 2054/* 2055 * read() on a dir. This weird interface hack only works if mounted 2056 * with '-o dirstat'. 2057 */ 2058static ssize_t ceph_read_dir(struct file *file, char __user *buf, size_t size, 2059 loff_t *ppos) 2060{ 2061 struct ceph_dir_file_info *dfi = file->private_data; 2062 struct inode *inode = file_inode(file); 2063 struct ceph_inode_info *ci = ceph_inode(inode); 2064 int left; 2065 const int bufsize = 1024; 2066 2067 if (!ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), DIRSTAT)) 2068 return -EISDIR; 2069 2070 if (!dfi->dir_info) { 2071 dfi->dir_info = kmalloc(bufsize, GFP_KERNEL); 2072 if (!dfi->dir_info) 2073 return -ENOMEM; 2074 dfi->dir_info_len = 2075 snprintf(dfi->dir_info, bufsize, 2076 "entries: %20lld\n" 2077 " files: %20lld\n" 2078 " subdirs: %20lld\n" 2079 "rentries: %20lld\n" 2080 " rfiles: %20lld\n" 2081 " rsubdirs: %20lld\n" 2082 "rbytes: %20lld\n" 2083 "rctime: %10lld.%09ld\n", 2084 ci->i_files + ci->i_subdirs, 2085 ci->i_files, 2086 ci->i_subdirs, 2087 ci->i_rfiles + ci->i_rsubdirs, 2088 ci->i_rfiles, 2089 ci->i_rsubdirs, 2090 ci->i_rbytes, 2091 ci->i_rctime.tv_sec, 2092 ci->i_rctime.tv_nsec); 2093 } 2094 2095 if (*ppos >= dfi->dir_info_len) 2096 return 0; 2097 size = min_t(unsigned, size, dfi->dir_info_len-*ppos); 2098 left = copy_to_user(buf, dfi->dir_info + *ppos, size); 2099 if (left == size) 2100 return -EFAULT; 2101 *ppos += (size - left); 2102 return size - left; 2103} 2104 2105 2106 2107/* 2108 * Return name hash for a given dentry. This is dependent on 2109 * the parent directory's hash function. 2110 */ 2111unsigned ceph_dentry_hash(struct inode *dir, struct dentry *dn) 2112{ 2113 struct ceph_inode_info *dci = ceph_inode(dir); 2114 unsigned hash; 2115 2116 switch (dci->i_dir_layout.dl_dir_hash) { 2117 case 0: /* for backward compat */ 2118 case CEPH_STR_HASH_LINUX: 2119 return dn->d_name.hash; 2120 2121 default: 2122 spin_lock(&dn->d_lock); 2123 hash = ceph_str_hash(dci->i_dir_layout.dl_dir_hash, 2124 dn->d_name.name, dn->d_name.len); 2125 spin_unlock(&dn->d_lock); 2126 return hash; 2127 } 2128} 2129 2130WRAP_DIR_ITER(ceph_readdir) // FIXME! 2131const struct file_operations ceph_dir_fops = { 2132 .read = ceph_read_dir, 2133 .iterate_shared = shared_ceph_readdir, 2134 .llseek = ceph_dir_llseek, 2135 .open = ceph_open, 2136 .release = ceph_release, 2137 .unlocked_ioctl = ceph_ioctl, 2138 .compat_ioctl = compat_ptr_ioctl, 2139 .fsync = ceph_fsync, 2140 .lock = ceph_lock, 2141 .flock = ceph_flock, 2142}; 2143 2144const struct file_operations ceph_snapdir_fops = { 2145 .iterate_shared = shared_ceph_readdir, 2146 .llseek = ceph_dir_llseek, 2147 .open = ceph_open, 2148 .release = ceph_release, 2149}; 2150 2151const struct inode_operations ceph_dir_iops = { 2152 .lookup = ceph_lookup, 2153 .permission = ceph_permission, 2154 .getattr = ceph_getattr, 2155 .setattr = ceph_setattr, 2156 .listxattr = ceph_listxattr, 2157 .get_inode_acl = ceph_get_acl, 2158 .set_acl = ceph_set_acl, 2159 .mknod = ceph_mknod, 2160 .symlink = ceph_symlink, 2161 .mkdir = ceph_mkdir, 2162 .link = ceph_link, 2163 .unlink = ceph_unlink, 2164 .rmdir = ceph_unlink, 2165 .rename = ceph_rename, 2166 .create = ceph_create, 2167 .atomic_open = ceph_atomic_open, 2168}; 2169 2170const struct inode_operations ceph_snapdir_iops = { 2171 .lookup = ceph_lookup, 2172 .permission = ceph_permission, 2173 .getattr = ceph_getattr, 2174 .mkdir = ceph_mkdir, 2175 .rmdir = ceph_unlink, 2176 .rename = ceph_rename, 2177}; 2178 2179const struct dentry_operations ceph_dentry_ops = { 2180 .d_revalidate = ceph_d_revalidate, 2181 .d_delete = ceph_d_delete, 2182 .d_release = ceph_d_release, 2183 .d_prune = ceph_d_prune, 2184 .d_init = ceph_d_init, 2185}; 2186