1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * linux/fs/nfs/dir.c 4 * 5 * Copyright (C) 1992 Rick Sladkey 6 * 7 * nfs directory handling functions 8 * 9 * 10 Apr 1996 Added silly rename for unlink --okir 10 * 28 Sep 1996 Improved directory cache --okir 11 * 23 Aug 1997 Claus Heine claus@momo.math.rwth-aachen.de 12 * Re-implemented silly rename for unlink, newly implemented 13 * silly rename for nfs_rename() following the suggestions 14 * of Olaf Kirch (okir) found in this file. 15 * Following Linus comments on my original hack, this version 16 * depends only on the dcache stuff and doesn't touch the inode 17 * layer (iput() and friends). 18 * 6 Jun 1999 Cache readdir lookups in the page cache. -DaveM 19 */ 20 21 #include <linux/compat.h> 22 #include <linux/module.h> 23 #include <linux/time.h> 24 #include <linux/errno.h> 25 #include <linux/stat.h> 26 #include <linux/fcntl.h> 27 #include <linux/string.h> 28 #include <linux/kernel.h> 29 #include <linux/slab.h> 30 #include <linux/mm.h> 31 #include <linux/sunrpc/clnt.h> 32 #include <linux/nfs_fs.h> 33 #include <linux/nfs_mount.h> 34 #include <linux/pagemap.h> 35 #include <linux/pagevec.h> 36 #include <linux/namei.h> 37 #include <linux/mount.h> 38 #include <linux/swap.h> 39 #include <linux/sched.h> 40 #include <linux/kmemleak.h> 41 #include <linux/xattr.h> 42 #include <linux/xxhash.h> 43 44 #include "delegation.h" 45 #include "iostat.h" 46 #include "internal.h" 47 #include "fscache.h" 48 49 #include "nfstrace.h" 50 51 /* #define NFS_DEBUG_VERBOSE 1 */ 52 53 static int nfs_opendir(struct inode *, struct file *); 54 static int nfs_closedir(struct inode *, struct file *); 55 static int nfs_readdir(struct file *, struct dir_context *); 56 static int nfs_fsync_dir(struct file *, loff_t, loff_t, int); 57 static loff_t nfs_llseek_dir(struct file *, loff_t, int); 58 static void nfs_readdir_clear_array(struct page*); 59 60 const struct file_operations nfs_dir_operations = { 61 .llseek = nfs_llseek_dir, 62 .read = generic_read_dir, 63 .iterate_shared = nfs_readdir, 64 .open = nfs_opendir, 65 .release = nfs_closedir, 66 .fsync = nfs_fsync_dir, 67 }; 68 69 const struct address_space_operations nfs_dir_aops = { 70 .freepage = nfs_readdir_clear_array, 71 }; 72 73 #define NFS_INIT_DTSIZE PAGE_SIZE 74 75 static struct nfs_open_dir_context * 76 alloc_nfs_open_dir_context(struct inode *dir) 77 { 78 struct nfs_inode *nfsi = NFS_I(dir); 79 struct nfs_open_dir_context *ctx; 80 81 ctx = kzalloc(sizeof(*ctx), GFP_KERNEL_ACCOUNT); 82 if (ctx != NULL) { 83 ctx->attr_gencount = nfsi->attr_gencount; 84 ctx->dtsize = NFS_INIT_DTSIZE; 85 spin_lock(&dir->i_lock); 86 if (list_empty(&nfsi->open_files) && 87 (nfsi->cache_validity & NFS_INO_DATA_INVAL_DEFER)) 88 nfs_set_cache_invalid(dir, 89 NFS_INO_INVALID_DATA | 90 NFS_INO_REVAL_FORCED); 91 list_add_tail_rcu(&ctx->list, &nfsi->open_files); 92 memcpy(ctx->verf, nfsi->cookieverf, sizeof(ctx->verf)); 93 spin_unlock(&dir->i_lock); 94 return ctx; 95 } 96 return ERR_PTR(-ENOMEM); 97 } 98 99 static void put_nfs_open_dir_context(struct inode *dir, struct nfs_open_dir_context *ctx) 100 { 101 spin_lock(&dir->i_lock); 102 list_del_rcu(&ctx->list); 103 spin_unlock(&dir->i_lock); 104 kfree_rcu(ctx, rcu_head); 105 } 106 107 /* 108 * Open file 109 */ 110 static int 111 nfs_opendir(struct inode *inode, struct file *filp) 112 { 113 int res = 0; 114 struct nfs_open_dir_context *ctx; 115 116 dfprintk(FILE, "NFS: open dir(%pD2)\n", filp); 117 118 nfs_inc_stats(inode, NFSIOS_VFSOPEN); 119 120 ctx = alloc_nfs_open_dir_context(inode); 121 if (IS_ERR(ctx)) { 122 res = PTR_ERR(ctx); 123 goto out; 124 } 125 filp->private_data = ctx; 126 out: 127 return res; 128 } 129 130 static int 131 nfs_closedir(struct inode *inode, struct file *filp) 132 { 133 put_nfs_open_dir_context(file_inode(filp), filp->private_data); 134 return 0; 135 } 136 137 struct nfs_cache_array_entry { 138 u64 cookie; 139 u64 ino; 140 const char *name; 141 unsigned int name_len; 142 unsigned char d_type; 143 }; 144 145 struct nfs_cache_array { 146 u64 change_attr; 147 u64 last_cookie; 148 unsigned int size; 149 unsigned char page_full : 1, 150 page_is_eof : 1, 151 cookies_are_ordered : 1; 152 struct nfs_cache_array_entry array[]; 153 }; 154 155 struct nfs_readdir_descriptor { 156 struct file *file; 157 struct page *page; 158 struct dir_context *ctx; 159 pgoff_t page_index; 160 pgoff_t page_index_max; 161 u64 dir_cookie; 162 u64 last_cookie; 163 loff_t current_index; 164 165 __be32 verf[NFS_DIR_VERIFIER_SIZE]; 166 unsigned long dir_verifier; 167 unsigned long timestamp; 168 unsigned long gencount; 169 unsigned long attr_gencount; 170 unsigned int cache_entry_index; 171 unsigned int buffer_fills; 172 unsigned int dtsize; 173 bool clear_cache; 174 bool plus; 175 bool eob; 176 bool eof; 177 }; 178 179 static void nfs_set_dtsize(struct nfs_readdir_descriptor *desc, unsigned int sz) 180 { 181 struct nfs_server *server = NFS_SERVER(file_inode(desc->file)); 182 unsigned int maxsize = server->dtsize; 183 184 if (sz > maxsize) 185 sz = maxsize; 186 if (sz < NFS_MIN_FILE_IO_SIZE) 187 sz = NFS_MIN_FILE_IO_SIZE; 188 desc->dtsize = sz; 189 } 190 191 static void nfs_shrink_dtsize(struct nfs_readdir_descriptor *desc) 192 { 193 nfs_set_dtsize(desc, desc->dtsize >> 1); 194 } 195 196 static void nfs_grow_dtsize(struct nfs_readdir_descriptor *desc) 197 { 198 nfs_set_dtsize(desc, desc->dtsize << 1); 199 } 200 201 static void nfs_readdir_page_init_array(struct page *page, u64 last_cookie, 202 u64 change_attr) 203 { 204 struct nfs_cache_array *array; 205 206 array = kmap_atomic(page); 207 array->change_attr = change_attr; 208 array->last_cookie = last_cookie; 209 array->size = 0; 210 array->page_full = 0; 211 array->page_is_eof = 0; 212 array->cookies_are_ordered = 1; 213 kunmap_atomic(array); 214 } 215 216 /* 217 * we are freeing strings created by nfs_add_to_readdir_array() 218 */ 219 static void nfs_readdir_clear_array(struct page *page) 220 { 221 struct nfs_cache_array *array; 222 unsigned int i; 223 224 array = kmap_atomic(page); 225 for (i = 0; i < array->size; i++) 226 kfree(array->array[i].name); 227 array->size = 0; 228 kunmap_atomic(array); 229 } 230 231 static void nfs_readdir_page_reinit_array(struct page *page, u64 last_cookie, 232 u64 change_attr) 233 { 234 nfs_readdir_clear_array(page); 235 nfs_readdir_page_init_array(page, last_cookie, change_attr); 236 } 237 238 static struct page * 239 nfs_readdir_page_array_alloc(u64 last_cookie, gfp_t gfp_flags) 240 { 241 struct page *page = alloc_page(gfp_flags); 242 if (page) 243 nfs_readdir_page_init_array(page, last_cookie, 0); 244 return page; 245 } 246 247 static void nfs_readdir_page_array_free(struct page *page) 248 { 249 if (page) { 250 nfs_readdir_clear_array(page); 251 put_page(page); 252 } 253 } 254 255 static void nfs_readdir_array_set_eof(struct nfs_cache_array *array) 256 { 257 array->page_is_eof = 1; 258 array->page_full = 1; 259 } 260 261 static bool nfs_readdir_array_is_full(struct nfs_cache_array *array) 262 { 263 return array->page_full; 264 } 265 266 /* 267 * the caller is responsible for freeing qstr.name 268 * when called by nfs_readdir_add_to_array, the strings will be freed in 269 * nfs_clear_readdir_array() 270 */ 271 static const char *nfs_readdir_copy_name(const char *name, unsigned int len) 272 { 273 const char *ret = kmemdup_nul(name, len, GFP_KERNEL); 274 275 /* 276 * Avoid a kmemleak false positive. The pointer to the name is stored 277 * in a page cache page which kmemleak does not scan. 278 */ 279 if (ret != NULL) 280 kmemleak_not_leak(ret); 281 return ret; 282 } 283 284 static size_t nfs_readdir_array_maxentries(void) 285 { 286 return (PAGE_SIZE - sizeof(struct nfs_cache_array)) / 287 sizeof(struct nfs_cache_array_entry); 288 } 289 290 /* 291 * Check that the next array entry lies entirely within the page bounds 292 */ 293 static int nfs_readdir_array_can_expand(struct nfs_cache_array *array) 294 { 295 if (array->page_full) 296 return -ENOSPC; 297 if (array->size == nfs_readdir_array_maxentries()) { 298 array->page_full = 1; 299 return -ENOSPC; 300 } 301 return 0; 302 } 303 304 static int nfs_readdir_page_array_append(struct page *page, 305 const struct nfs_entry *entry, 306 u64 *cookie) 307 { 308 struct nfs_cache_array *array; 309 struct nfs_cache_array_entry *cache_entry; 310 const char *name; 311 int ret = -ENOMEM; 312 313 name = nfs_readdir_copy_name(entry->name, entry->len); 314 315 array = kmap_atomic(page); 316 if (!name) 317 goto out; 318 ret = nfs_readdir_array_can_expand(array); 319 if (ret) { 320 kfree(name); 321 goto out; 322 } 323 324 cache_entry = &array->array[array->size]; 325 cache_entry->cookie = array->last_cookie; 326 cache_entry->ino = entry->ino; 327 cache_entry->d_type = entry->d_type; 328 cache_entry->name_len = entry->len; 329 cache_entry->name = name; 330 array->last_cookie = entry->cookie; 331 if (array->last_cookie <= cache_entry->cookie) 332 array->cookies_are_ordered = 0; 333 array->size++; 334 if (entry->eof != 0) 335 nfs_readdir_array_set_eof(array); 336 out: 337 *cookie = array->last_cookie; 338 kunmap_atomic(array); 339 return ret; 340 } 341 342 #define NFS_READDIR_COOKIE_MASK (U32_MAX >> 14) 343 /* 344 * Hash algorithm allowing content addressible access to sequences 345 * of directory cookies. Content is addressed by the value of the 346 * cookie index of the first readdir entry in a page. 347 * 348 * The xxhash algorithm is chosen because it is fast, and is supposed 349 * to result in a decent flat distribution of hashes. 350 * 351 * We then select only the first 18 bits to avoid issues with excessive 352 * memory use for the page cache XArray. 18 bits should allow the caching 353 * of 262144 pages of sequences of readdir entries. Since each page holds 354 * 127 readdir entries for a typical 64-bit system, that works out to a 355 * cache of ~ 33 million entries per directory. 356 */ 357 static pgoff_t nfs_readdir_page_cookie_hash(u64 cookie) 358 { 359 if (cookie == 0) 360 return 0; 361 return xxhash(&cookie, sizeof(cookie), 0) & NFS_READDIR_COOKIE_MASK; 362 } 363 364 static bool nfs_readdir_page_validate(struct page *page, u64 last_cookie, 365 u64 change_attr) 366 { 367 struct nfs_cache_array *array = kmap_atomic(page); 368 int ret = true; 369 370 if (array->change_attr != change_attr) 371 ret = false; 372 if (array->size > 0 && array->array[0].cookie != last_cookie) 373 ret = false; 374 kunmap_atomic(array); 375 return ret; 376 } 377 378 static void nfs_readdir_page_unlock_and_put(struct page *page) 379 { 380 unlock_page(page); 381 put_page(page); 382 } 383 384 static void nfs_readdir_page_init_and_validate(struct page *page, u64 cookie, 385 u64 change_attr) 386 { 387 if (PageUptodate(page)) { 388 if (nfs_readdir_page_validate(page, cookie, change_attr)) 389 return; 390 nfs_readdir_clear_array(page); 391 } 392 nfs_readdir_page_init_array(page, cookie, change_attr); 393 SetPageUptodate(page); 394 } 395 396 static struct page *nfs_readdir_page_get_locked(struct address_space *mapping, 397 u64 cookie, u64 change_attr) 398 { 399 pgoff_t index = nfs_readdir_page_cookie_hash(cookie); 400 struct page *page; 401 402 page = grab_cache_page(mapping, index); 403 if (!page) 404 return NULL; 405 nfs_readdir_page_init_and_validate(page, cookie, change_attr); 406 return page; 407 } 408 409 static u64 nfs_readdir_page_last_cookie(struct page *page) 410 { 411 struct nfs_cache_array *array; 412 u64 ret; 413 414 array = kmap_atomic(page); 415 ret = array->last_cookie; 416 kunmap_atomic(array); 417 return ret; 418 } 419 420 static bool nfs_readdir_page_needs_filling(struct page *page) 421 { 422 struct nfs_cache_array *array; 423 bool ret; 424 425 array = kmap_atomic(page); 426 ret = !nfs_readdir_array_is_full(array); 427 kunmap_atomic(array); 428 return ret; 429 } 430 431 static void nfs_readdir_page_set_eof(struct page *page) 432 { 433 struct nfs_cache_array *array; 434 435 array = kmap_atomic(page); 436 nfs_readdir_array_set_eof(array); 437 kunmap_atomic(array); 438 } 439 440 static struct page *nfs_readdir_page_get_next(struct address_space *mapping, 441 u64 cookie, u64 change_attr) 442 { 443 pgoff_t index = nfs_readdir_page_cookie_hash(cookie); 444 struct page *page; 445 446 page = grab_cache_page_nowait(mapping, index); 447 if (!page) 448 return NULL; 449 nfs_readdir_page_init_and_validate(page, cookie, change_attr); 450 if (nfs_readdir_page_last_cookie(page) != cookie) 451 nfs_readdir_page_reinit_array(page, cookie, change_attr); 452 return page; 453 } 454 455 static inline 456 int is_32bit_api(void) 457 { 458 #ifdef CONFIG_COMPAT 459 return in_compat_syscall(); 460 #else 461 return (BITS_PER_LONG == 32); 462 #endif 463 } 464 465 static 466 bool nfs_readdir_use_cookie(const struct file *filp) 467 { 468 if ((filp->f_mode & FMODE_32BITHASH) || 469 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api())) 470 return false; 471 return true; 472 } 473 474 static void nfs_readdir_seek_next_array(struct nfs_cache_array *array, 475 struct nfs_readdir_descriptor *desc) 476 { 477 if (array->page_full) { 478 desc->last_cookie = array->last_cookie; 479 desc->current_index += array->size; 480 desc->cache_entry_index = 0; 481 desc->page_index++; 482 } else 483 desc->last_cookie = array->array[0].cookie; 484 } 485 486 static void nfs_readdir_rewind_search(struct nfs_readdir_descriptor *desc) 487 { 488 desc->current_index = 0; 489 desc->last_cookie = 0; 490 desc->page_index = 0; 491 } 492 493 static int nfs_readdir_search_for_pos(struct nfs_cache_array *array, 494 struct nfs_readdir_descriptor *desc) 495 { 496 loff_t diff = desc->ctx->pos - desc->current_index; 497 unsigned int index; 498 499 if (diff < 0) 500 goto out_eof; 501 if (diff >= array->size) { 502 if (array->page_is_eof) 503 goto out_eof; 504 nfs_readdir_seek_next_array(array, desc); 505 return -EAGAIN; 506 } 507 508 index = (unsigned int)diff; 509 desc->dir_cookie = array->array[index].cookie; 510 desc->cache_entry_index = index; 511 return 0; 512 out_eof: 513 desc->eof = true; 514 return -EBADCOOKIE; 515 } 516 517 static bool nfs_readdir_array_cookie_in_range(struct nfs_cache_array *array, 518 u64 cookie) 519 { 520 if (!array->cookies_are_ordered) 521 return true; 522 /* Optimisation for monotonically increasing cookies */ 523 if (cookie >= array->last_cookie) 524 return false; 525 if (array->size && cookie < array->array[0].cookie) 526 return false; 527 return true; 528 } 529 530 static int nfs_readdir_search_for_cookie(struct nfs_cache_array *array, 531 struct nfs_readdir_descriptor *desc) 532 { 533 unsigned int i; 534 int status = -EAGAIN; 535 536 if (!nfs_readdir_array_cookie_in_range(array, desc->dir_cookie)) 537 goto check_eof; 538 539 for (i = 0; i < array->size; i++) { 540 if (array->array[i].cookie == desc->dir_cookie) { 541 if (nfs_readdir_use_cookie(desc->file)) 542 desc->ctx->pos = desc->dir_cookie; 543 else 544 desc->ctx->pos = desc->current_index + i; 545 desc->cache_entry_index = i; 546 return 0; 547 } 548 } 549 check_eof: 550 if (array->page_is_eof) { 551 status = -EBADCOOKIE; 552 if (desc->dir_cookie == array->last_cookie) 553 desc->eof = true; 554 } else 555 nfs_readdir_seek_next_array(array, desc); 556 return status; 557 } 558 559 static int nfs_readdir_search_array(struct nfs_readdir_descriptor *desc) 560 { 561 struct nfs_cache_array *array; 562 int status; 563 564 array = kmap_atomic(desc->page); 565 566 if (desc->dir_cookie == 0) 567 status = nfs_readdir_search_for_pos(array, desc); 568 else 569 status = nfs_readdir_search_for_cookie(array, desc); 570 571 kunmap_atomic(array); 572 return status; 573 } 574 575 /* Fill a page with xdr information before transferring to the cache page */ 576 static int nfs_readdir_xdr_filler(struct nfs_readdir_descriptor *desc, 577 __be32 *verf, u64 cookie, 578 struct page **pages, size_t bufsize, 579 __be32 *verf_res) 580 { 581 struct inode *inode = file_inode(desc->file); 582 struct nfs_readdir_arg arg = { 583 .dentry = file_dentry(desc->file), 584 .cred = desc->file->f_cred, 585 .verf = verf, 586 .cookie = cookie, 587 .pages = pages, 588 .page_len = bufsize, 589 .plus = desc->plus, 590 }; 591 struct nfs_readdir_res res = { 592 .verf = verf_res, 593 }; 594 unsigned long timestamp, gencount; 595 int error; 596 597 again: 598 timestamp = jiffies; 599 gencount = nfs_inc_attr_generation_counter(); 600 desc->dir_verifier = nfs_save_change_attribute(inode); 601 error = NFS_PROTO(inode)->readdir(&arg, &res); 602 if (error < 0) { 603 /* We requested READDIRPLUS, but the server doesn't grok it */ 604 if (error == -ENOTSUPP && desc->plus) { 605 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS; 606 desc->plus = arg.plus = false; 607 goto again; 608 } 609 goto error; 610 } 611 desc->timestamp = timestamp; 612 desc->gencount = gencount; 613 error: 614 return error; 615 } 616 617 static int xdr_decode(struct nfs_readdir_descriptor *desc, 618 struct nfs_entry *entry, struct xdr_stream *xdr) 619 { 620 struct inode *inode = file_inode(desc->file); 621 int error; 622 623 error = NFS_PROTO(inode)->decode_dirent(xdr, entry, desc->plus); 624 if (error) 625 return error; 626 entry->fattr->time_start = desc->timestamp; 627 entry->fattr->gencount = desc->gencount; 628 return 0; 629 } 630 631 /* Match file and dirent using either filehandle or fileid 632 * Note: caller is responsible for checking the fsid 633 */ 634 static 635 int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry) 636 { 637 struct inode *inode; 638 struct nfs_inode *nfsi; 639 640 if (d_really_is_negative(dentry)) 641 return 0; 642 643 inode = d_inode(dentry); 644 if (is_bad_inode(inode) || NFS_STALE(inode)) 645 return 0; 646 647 nfsi = NFS_I(inode); 648 if (entry->fattr->fileid != nfsi->fileid) 649 return 0; 650 if (entry->fh->size && nfs_compare_fh(entry->fh, &nfsi->fh) != 0) 651 return 0; 652 return 1; 653 } 654 655 #define NFS_READDIR_CACHE_USAGE_THRESHOLD (8UL) 656 657 static bool nfs_use_readdirplus(struct inode *dir, struct dir_context *ctx, 658 unsigned int cache_hits, 659 unsigned int cache_misses) 660 { 661 if (!nfs_server_capable(dir, NFS_CAP_READDIRPLUS)) 662 return false; 663 if (ctx->pos == 0 || 664 cache_hits + cache_misses > NFS_READDIR_CACHE_USAGE_THRESHOLD) 665 return true; 666 return false; 667 } 668 669 /* 670 * This function is called by the getattr code to request the 671 * use of readdirplus to accelerate any future lookups in the same 672 * directory. 673 */ 674 void nfs_readdir_record_entry_cache_hit(struct inode *dir) 675 { 676 struct nfs_inode *nfsi = NFS_I(dir); 677 struct nfs_open_dir_context *ctx; 678 679 if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) && 680 S_ISDIR(dir->i_mode)) { 681 rcu_read_lock(); 682 list_for_each_entry_rcu (ctx, &nfsi->open_files, list) 683 atomic_inc(&ctx->cache_hits); 684 rcu_read_unlock(); 685 } 686 } 687 688 /* 689 * This function is mainly for use by nfs_getattr(). 690 * 691 * If this is an 'ls -l', we want to force use of readdirplus. 692 */ 693 void nfs_readdir_record_entry_cache_miss(struct inode *dir) 694 { 695 struct nfs_inode *nfsi = NFS_I(dir); 696 struct nfs_open_dir_context *ctx; 697 698 if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) && 699 S_ISDIR(dir->i_mode)) { 700 rcu_read_lock(); 701 list_for_each_entry_rcu (ctx, &nfsi->open_files, list) 702 atomic_inc(&ctx->cache_misses); 703 rcu_read_unlock(); 704 } 705 } 706 707 static void nfs_lookup_advise_force_readdirplus(struct inode *dir, 708 unsigned int flags) 709 { 710 if (nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE)) 711 return; 712 if (flags & (LOOKUP_EXCL | LOOKUP_PARENT | LOOKUP_REVAL)) 713 return; 714 nfs_readdir_record_entry_cache_miss(dir); 715 } 716 717 static 718 void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry, 719 unsigned long dir_verifier) 720 { 721 struct qstr filename = QSTR_INIT(entry->name, entry->len); 722 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 723 struct dentry *dentry; 724 struct dentry *alias; 725 struct inode *inode; 726 int status; 727 728 if (!(entry->fattr->valid & NFS_ATTR_FATTR_FILEID)) 729 return; 730 if (!(entry->fattr->valid & NFS_ATTR_FATTR_FSID)) 731 return; 732 if (filename.len == 0) 733 return; 734 /* Validate that the name doesn't contain any illegal '\0' */ 735 if (strnlen(filename.name, filename.len) != filename.len) 736 return; 737 /* ...or '/' */ 738 if (strnchr(filename.name, filename.len, '/')) 739 return; 740 if (filename.name[0] == '.') { 741 if (filename.len == 1) 742 return; 743 if (filename.len == 2 && filename.name[1] == '.') 744 return; 745 } 746 filename.hash = full_name_hash(parent, filename.name, filename.len); 747 748 dentry = d_lookup(parent, &filename); 749 again: 750 if (!dentry) { 751 dentry = d_alloc_parallel(parent, &filename, &wq); 752 if (IS_ERR(dentry)) 753 return; 754 } 755 if (!d_in_lookup(dentry)) { 756 /* Is there a mountpoint here? If so, just exit */ 757 if (!nfs_fsid_equal(&NFS_SB(dentry->d_sb)->fsid, 758 &entry->fattr->fsid)) 759 goto out; 760 if (nfs_same_file(dentry, entry)) { 761 if (!entry->fh->size) 762 goto out; 763 nfs_set_verifier(dentry, dir_verifier); 764 status = nfs_refresh_inode(d_inode(dentry), entry->fattr); 765 if (!status) 766 nfs_setsecurity(d_inode(dentry), entry->fattr); 767 trace_nfs_readdir_lookup_revalidate(d_inode(parent), 768 dentry, 0, status); 769 goto out; 770 } else { 771 trace_nfs_readdir_lookup_revalidate_failed( 772 d_inode(parent), dentry, 0); 773 d_invalidate(dentry); 774 dput(dentry); 775 dentry = NULL; 776 goto again; 777 } 778 } 779 if (!entry->fh->size) { 780 d_lookup_done(dentry); 781 goto out; 782 } 783 784 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr); 785 alias = d_splice_alias(inode, dentry); 786 d_lookup_done(dentry); 787 if (alias) { 788 if (IS_ERR(alias)) 789 goto out; 790 dput(dentry); 791 dentry = alias; 792 } 793 nfs_set_verifier(dentry, dir_verifier); 794 trace_nfs_readdir_lookup(d_inode(parent), dentry, 0); 795 out: 796 dput(dentry); 797 } 798 799 static int nfs_readdir_entry_decode(struct nfs_readdir_descriptor *desc, 800 struct nfs_entry *entry, 801 struct xdr_stream *stream) 802 { 803 int ret; 804 805 if (entry->fattr->label) 806 entry->fattr->label->len = NFS4_MAXLABELLEN; 807 ret = xdr_decode(desc, entry, stream); 808 if (ret || !desc->plus) 809 return ret; 810 nfs_prime_dcache(file_dentry(desc->file), entry, desc->dir_verifier); 811 return 0; 812 } 813 814 /* Perform conversion from xdr to cache array */ 815 static int nfs_readdir_page_filler(struct nfs_readdir_descriptor *desc, 816 struct nfs_entry *entry, 817 struct page **xdr_pages, unsigned int buflen, 818 struct page **arrays, size_t narrays, 819 u64 change_attr) 820 { 821 struct address_space *mapping = desc->file->f_mapping; 822 struct xdr_stream stream; 823 struct xdr_buf buf; 824 struct page *scratch, *new, *page = *arrays; 825 u64 cookie; 826 int status; 827 828 scratch = alloc_page(GFP_KERNEL); 829 if (scratch == NULL) 830 return -ENOMEM; 831 832 xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen); 833 xdr_set_scratch_page(&stream, scratch); 834 835 do { 836 status = nfs_readdir_entry_decode(desc, entry, &stream); 837 if (status != 0) 838 break; 839 840 status = nfs_readdir_page_array_append(page, entry, &cookie); 841 if (status != -ENOSPC) 842 continue; 843 844 if (page->mapping != mapping) { 845 if (!--narrays) 846 break; 847 new = nfs_readdir_page_array_alloc(cookie, GFP_KERNEL); 848 if (!new) 849 break; 850 arrays++; 851 *arrays = page = new; 852 } else { 853 new = nfs_readdir_page_get_next(mapping, cookie, 854 change_attr); 855 if (!new) 856 break; 857 if (page != *arrays) 858 nfs_readdir_page_unlock_and_put(page); 859 page = new; 860 } 861 desc->page_index_max++; 862 status = nfs_readdir_page_array_append(page, entry, &cookie); 863 } while (!status && !entry->eof); 864 865 switch (status) { 866 case -EBADCOOKIE: 867 if (!entry->eof) 868 break; 869 nfs_readdir_page_set_eof(page); 870 fallthrough; 871 case -EAGAIN: 872 status = 0; 873 break; 874 case -ENOSPC: 875 status = 0; 876 if (!desc->plus) 877 break; 878 while (!nfs_readdir_entry_decode(desc, entry, &stream)) 879 ; 880 } 881 882 if (page != *arrays) 883 nfs_readdir_page_unlock_and_put(page); 884 885 put_page(scratch); 886 return status; 887 } 888 889 static void nfs_readdir_free_pages(struct page **pages, size_t npages) 890 { 891 while (npages--) 892 put_page(pages[npages]); 893 kfree(pages); 894 } 895 896 /* 897 * nfs_readdir_alloc_pages() will allocate pages that must be freed with a call 898 * to nfs_readdir_free_pages() 899 */ 900 static struct page **nfs_readdir_alloc_pages(size_t npages) 901 { 902 struct page **pages; 903 size_t i; 904 905 pages = kmalloc_array(npages, sizeof(*pages), GFP_KERNEL); 906 if (!pages) 907 return NULL; 908 for (i = 0; i < npages; i++) { 909 struct page *page = alloc_page(GFP_KERNEL); 910 if (page == NULL) 911 goto out_freepages; 912 pages[i] = page; 913 } 914 return pages; 915 916 out_freepages: 917 nfs_readdir_free_pages(pages, i); 918 return NULL; 919 } 920 921 static int nfs_readdir_xdr_to_array(struct nfs_readdir_descriptor *desc, 922 __be32 *verf_arg, __be32 *verf_res, 923 struct page **arrays, size_t narrays) 924 { 925 u64 change_attr; 926 struct page **pages; 927 struct page *page = *arrays; 928 struct nfs_entry *entry; 929 size_t array_size; 930 struct inode *inode = file_inode(desc->file); 931 unsigned int dtsize = desc->dtsize; 932 unsigned int pglen; 933 int status = -ENOMEM; 934 935 entry = kzalloc(sizeof(*entry), GFP_KERNEL); 936 if (!entry) 937 return -ENOMEM; 938 entry->cookie = nfs_readdir_page_last_cookie(page); 939 entry->fh = nfs_alloc_fhandle(); 940 entry->fattr = nfs_alloc_fattr_with_label(NFS_SERVER(inode)); 941 entry->server = NFS_SERVER(inode); 942 if (entry->fh == NULL || entry->fattr == NULL) 943 goto out; 944 945 array_size = (dtsize + PAGE_SIZE - 1) >> PAGE_SHIFT; 946 pages = nfs_readdir_alloc_pages(array_size); 947 if (!pages) 948 goto out; 949 950 change_attr = inode_peek_iversion_raw(inode); 951 status = nfs_readdir_xdr_filler(desc, verf_arg, entry->cookie, pages, 952 dtsize, verf_res); 953 if (status < 0) 954 goto free_pages; 955 956 pglen = status; 957 if (pglen != 0) 958 status = nfs_readdir_page_filler(desc, entry, pages, pglen, 959 arrays, narrays, change_attr); 960 else 961 nfs_readdir_page_set_eof(page); 962 desc->buffer_fills++; 963 964 free_pages: 965 nfs_readdir_free_pages(pages, array_size); 966 out: 967 nfs_free_fattr(entry->fattr); 968 nfs_free_fhandle(entry->fh); 969 kfree(entry); 970 return status; 971 } 972 973 static void nfs_readdir_page_put(struct nfs_readdir_descriptor *desc) 974 { 975 put_page(desc->page); 976 desc->page = NULL; 977 } 978 979 static void 980 nfs_readdir_page_unlock_and_put_cached(struct nfs_readdir_descriptor *desc) 981 { 982 unlock_page(desc->page); 983 nfs_readdir_page_put(desc); 984 } 985 986 static struct page * 987 nfs_readdir_page_get_cached(struct nfs_readdir_descriptor *desc) 988 { 989 struct address_space *mapping = desc->file->f_mapping; 990 u64 change_attr = inode_peek_iversion_raw(mapping->host); 991 u64 cookie = desc->last_cookie; 992 struct page *page; 993 994 page = nfs_readdir_page_get_locked(mapping, cookie, change_attr); 995 if (!page) 996 return NULL; 997 if (desc->clear_cache && !nfs_readdir_page_needs_filling(page)) 998 nfs_readdir_page_reinit_array(page, cookie, change_attr); 999 return page; 1000 } 1001 1002 /* 1003 * Returns 0 if desc->dir_cookie was found on page desc->page_index 1004 * and locks the page to prevent removal from the page cache. 1005 */ 1006 static int find_and_lock_cache_page(struct nfs_readdir_descriptor *desc) 1007 { 1008 struct inode *inode = file_inode(desc->file); 1009 struct nfs_inode *nfsi = NFS_I(inode); 1010 __be32 verf[NFS_DIR_VERIFIER_SIZE]; 1011 int res; 1012 1013 desc->page = nfs_readdir_page_get_cached(desc); 1014 if (!desc->page) 1015 return -ENOMEM; 1016 if (nfs_readdir_page_needs_filling(desc->page)) { 1017 /* Grow the dtsize if we had to go back for more pages */ 1018 if (desc->page_index == desc->page_index_max) 1019 nfs_grow_dtsize(desc); 1020 desc->page_index_max = desc->page_index; 1021 trace_nfs_readdir_cache_fill(desc->file, nfsi->cookieverf, 1022 desc->last_cookie, 1023 desc->page->index, desc->dtsize); 1024 res = nfs_readdir_xdr_to_array(desc, nfsi->cookieverf, verf, 1025 &desc->page, 1); 1026 if (res < 0) { 1027 nfs_readdir_page_unlock_and_put_cached(desc); 1028 trace_nfs_readdir_cache_fill_done(inode, res); 1029 if (res == -EBADCOOKIE || res == -ENOTSYNC) { 1030 invalidate_inode_pages2(desc->file->f_mapping); 1031 nfs_readdir_rewind_search(desc); 1032 trace_nfs_readdir_invalidate_cache_range( 1033 inode, 0, MAX_LFS_FILESIZE); 1034 return -EAGAIN; 1035 } 1036 return res; 1037 } 1038 /* 1039 * Set the cookie verifier if the page cache was empty 1040 */ 1041 if (desc->last_cookie == 0 && 1042 memcmp(nfsi->cookieverf, verf, sizeof(nfsi->cookieverf))) { 1043 memcpy(nfsi->cookieverf, verf, 1044 sizeof(nfsi->cookieverf)); 1045 invalidate_inode_pages2_range(desc->file->f_mapping, 1, 1046 -1); 1047 trace_nfs_readdir_invalidate_cache_range( 1048 inode, 1, MAX_LFS_FILESIZE); 1049 } 1050 desc->clear_cache = false; 1051 } 1052 res = nfs_readdir_search_array(desc); 1053 if (res == 0) 1054 return 0; 1055 nfs_readdir_page_unlock_and_put_cached(desc); 1056 return res; 1057 } 1058 1059 /* Search for desc->dir_cookie from the beginning of the page cache */ 1060 static int readdir_search_pagecache(struct nfs_readdir_descriptor *desc) 1061 { 1062 int res; 1063 1064 do { 1065 res = find_and_lock_cache_page(desc); 1066 } while (res == -EAGAIN); 1067 return res; 1068 } 1069 1070 /* 1071 * Once we've found the start of the dirent within a page: fill 'er up... 1072 */ 1073 static void nfs_do_filldir(struct nfs_readdir_descriptor *desc, 1074 const __be32 *verf) 1075 { 1076 struct file *file = desc->file; 1077 struct nfs_cache_array *array; 1078 unsigned int i; 1079 1080 array = kmap(desc->page); 1081 for (i = desc->cache_entry_index; i < array->size; i++) { 1082 struct nfs_cache_array_entry *ent; 1083 1084 ent = &array->array[i]; 1085 if (!dir_emit(desc->ctx, ent->name, ent->name_len, 1086 nfs_compat_user_ino64(ent->ino), ent->d_type)) { 1087 desc->eob = true; 1088 break; 1089 } 1090 memcpy(desc->verf, verf, sizeof(desc->verf)); 1091 if (i == array->size - 1) { 1092 desc->dir_cookie = array->last_cookie; 1093 nfs_readdir_seek_next_array(array, desc); 1094 } else { 1095 desc->dir_cookie = array->array[i + 1].cookie; 1096 desc->last_cookie = array->array[0].cookie; 1097 } 1098 if (nfs_readdir_use_cookie(file)) 1099 desc->ctx->pos = desc->dir_cookie; 1100 else 1101 desc->ctx->pos++; 1102 } 1103 if (array->page_is_eof) 1104 desc->eof = !desc->eob; 1105 1106 kunmap(desc->page); 1107 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %llu\n", 1108 (unsigned long long)desc->dir_cookie); 1109 } 1110 1111 /* 1112 * If we cannot find a cookie in our cache, we suspect that this is 1113 * because it points to a deleted file, so we ask the server to return 1114 * whatever it thinks is the next entry. We then feed this to filldir. 1115 * If all goes well, we should then be able to find our way round the 1116 * cache on the next call to readdir_search_pagecache(); 1117 * 1118 * NOTE: we cannot add the anonymous page to the pagecache because 1119 * the data it contains might not be page aligned. Besides, 1120 * we should already have a complete representation of the 1121 * directory in the page cache by the time we get here. 1122 */ 1123 static int uncached_readdir(struct nfs_readdir_descriptor *desc) 1124 { 1125 struct page **arrays; 1126 size_t i, sz = 512; 1127 __be32 verf[NFS_DIR_VERIFIER_SIZE]; 1128 int status = -ENOMEM; 1129 1130 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %llu\n", 1131 (unsigned long long)desc->dir_cookie); 1132 1133 arrays = kcalloc(sz, sizeof(*arrays), GFP_KERNEL); 1134 if (!arrays) 1135 goto out; 1136 arrays[0] = nfs_readdir_page_array_alloc(desc->dir_cookie, GFP_KERNEL); 1137 if (!arrays[0]) 1138 goto out; 1139 1140 desc->page_index = 0; 1141 desc->cache_entry_index = 0; 1142 desc->last_cookie = desc->dir_cookie; 1143 desc->page_index_max = 0; 1144 1145 trace_nfs_readdir_uncached(desc->file, desc->verf, desc->last_cookie, 1146 -1, desc->dtsize); 1147 1148 status = nfs_readdir_xdr_to_array(desc, desc->verf, verf, arrays, sz); 1149 if (status < 0) { 1150 trace_nfs_readdir_uncached_done(file_inode(desc->file), status); 1151 goto out_free; 1152 } 1153 1154 for (i = 0; !desc->eob && i < sz && arrays[i]; i++) { 1155 desc->page = arrays[i]; 1156 nfs_do_filldir(desc, verf); 1157 } 1158 desc->page = NULL; 1159 1160 /* 1161 * Grow the dtsize if we have to go back for more pages, 1162 * or shrink it if we're reading too many. 1163 */ 1164 if (!desc->eof) { 1165 if (!desc->eob) 1166 nfs_grow_dtsize(desc); 1167 else if (desc->buffer_fills == 1 && 1168 i < (desc->page_index_max >> 1)) 1169 nfs_shrink_dtsize(desc); 1170 } 1171 out_free: 1172 for (i = 0; i < sz && arrays[i]; i++) 1173 nfs_readdir_page_array_free(arrays[i]); 1174 out: 1175 if (!nfs_readdir_use_cookie(desc->file)) 1176 nfs_readdir_rewind_search(desc); 1177 desc->page_index_max = -1; 1178 kfree(arrays); 1179 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __func__, status); 1180 return status; 1181 } 1182 1183 #define NFS_READDIR_CACHE_MISS_THRESHOLD (16UL) 1184 1185 static bool nfs_readdir_handle_cache_misses(struct inode *inode, 1186 struct nfs_readdir_descriptor *desc, 1187 unsigned int cache_misses, 1188 bool force_clear) 1189 { 1190 if (desc->ctx->pos == 0 || !desc->plus) 1191 return false; 1192 if (cache_misses <= NFS_READDIR_CACHE_MISS_THRESHOLD && !force_clear) 1193 return false; 1194 trace_nfs_readdir_force_readdirplus(inode); 1195 return true; 1196 } 1197 1198 /* The file offset position represents the dirent entry number. A 1199 last cookie cache takes care of the common case of reading the 1200 whole directory. 1201 */ 1202 static int nfs_readdir(struct file *file, struct dir_context *ctx) 1203 { 1204 struct dentry *dentry = file_dentry(file); 1205 struct inode *inode = d_inode(dentry); 1206 struct nfs_inode *nfsi = NFS_I(inode); 1207 struct nfs_open_dir_context *dir_ctx = file->private_data; 1208 struct nfs_readdir_descriptor *desc; 1209 unsigned int cache_hits, cache_misses; 1210 bool force_clear; 1211 int res; 1212 1213 dfprintk(FILE, "NFS: readdir(%pD2) starting at cookie %llu\n", 1214 file, (long long)ctx->pos); 1215 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS); 1216 1217 /* 1218 * ctx->pos points to the dirent entry number. 1219 * *desc->dir_cookie has the cookie for the next entry. We have 1220 * to either find the entry with the appropriate number or 1221 * revalidate the cookie. 1222 */ 1223 nfs_revalidate_mapping(inode, file->f_mapping); 1224 1225 res = -ENOMEM; 1226 desc = kzalloc(sizeof(*desc), GFP_KERNEL); 1227 if (!desc) 1228 goto out; 1229 desc->file = file; 1230 desc->ctx = ctx; 1231 desc->page_index_max = -1; 1232 1233 spin_lock(&file->f_lock); 1234 desc->dir_cookie = dir_ctx->dir_cookie; 1235 desc->page_index = dir_ctx->page_index; 1236 desc->last_cookie = dir_ctx->last_cookie; 1237 desc->attr_gencount = dir_ctx->attr_gencount; 1238 desc->eof = dir_ctx->eof; 1239 nfs_set_dtsize(desc, dir_ctx->dtsize); 1240 memcpy(desc->verf, dir_ctx->verf, sizeof(desc->verf)); 1241 cache_hits = atomic_xchg(&dir_ctx->cache_hits, 0); 1242 cache_misses = atomic_xchg(&dir_ctx->cache_misses, 0); 1243 force_clear = dir_ctx->force_clear; 1244 spin_unlock(&file->f_lock); 1245 1246 if (desc->eof) { 1247 res = 0; 1248 goto out_free; 1249 } 1250 1251 desc->plus = nfs_use_readdirplus(inode, ctx, cache_hits, cache_misses); 1252 force_clear = nfs_readdir_handle_cache_misses(inode, desc, cache_misses, 1253 force_clear); 1254 desc->clear_cache = force_clear; 1255 1256 do { 1257 res = readdir_search_pagecache(desc); 1258 1259 if (res == -EBADCOOKIE) { 1260 res = 0; 1261 /* This means either end of directory */ 1262 if (desc->dir_cookie && !desc->eof) { 1263 /* Or that the server has 'lost' a cookie */ 1264 res = uncached_readdir(desc); 1265 if (res == 0) 1266 continue; 1267 if (res == -EBADCOOKIE || res == -ENOTSYNC) 1268 res = 0; 1269 } 1270 break; 1271 } 1272 if (res == -ETOOSMALL && desc->plus) { 1273 nfs_zap_caches(inode); 1274 desc->plus = false; 1275 desc->eof = false; 1276 continue; 1277 } 1278 if (res < 0) 1279 break; 1280 1281 nfs_do_filldir(desc, nfsi->cookieverf); 1282 nfs_readdir_page_unlock_and_put_cached(desc); 1283 if (desc->page_index == desc->page_index_max) 1284 desc->clear_cache = force_clear; 1285 } while (!desc->eob && !desc->eof); 1286 1287 spin_lock(&file->f_lock); 1288 dir_ctx->dir_cookie = desc->dir_cookie; 1289 dir_ctx->last_cookie = desc->last_cookie; 1290 dir_ctx->attr_gencount = desc->attr_gencount; 1291 dir_ctx->page_index = desc->page_index; 1292 dir_ctx->force_clear = force_clear; 1293 dir_ctx->eof = desc->eof; 1294 dir_ctx->dtsize = desc->dtsize; 1295 memcpy(dir_ctx->verf, desc->verf, sizeof(dir_ctx->verf)); 1296 spin_unlock(&file->f_lock); 1297 out_free: 1298 kfree(desc); 1299 1300 out: 1301 dfprintk(FILE, "NFS: readdir(%pD2) returns %d\n", file, res); 1302 return res; 1303 } 1304 1305 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int whence) 1306 { 1307 struct nfs_open_dir_context *dir_ctx = filp->private_data; 1308 1309 dfprintk(FILE, "NFS: llseek dir(%pD2, %lld, %d)\n", 1310 filp, offset, whence); 1311 1312 switch (whence) { 1313 default: 1314 return -EINVAL; 1315 case SEEK_SET: 1316 if (offset < 0) 1317 return -EINVAL; 1318 spin_lock(&filp->f_lock); 1319 break; 1320 case SEEK_CUR: 1321 if (offset == 0) 1322 return filp->f_pos; 1323 spin_lock(&filp->f_lock); 1324 offset += filp->f_pos; 1325 if (offset < 0) { 1326 spin_unlock(&filp->f_lock); 1327 return -EINVAL; 1328 } 1329 } 1330 if (offset != filp->f_pos) { 1331 filp->f_pos = offset; 1332 dir_ctx->page_index = 0; 1333 if (!nfs_readdir_use_cookie(filp)) { 1334 dir_ctx->dir_cookie = 0; 1335 dir_ctx->last_cookie = 0; 1336 } else { 1337 dir_ctx->dir_cookie = offset; 1338 dir_ctx->last_cookie = offset; 1339 } 1340 dir_ctx->eof = false; 1341 } 1342 spin_unlock(&filp->f_lock); 1343 return offset; 1344 } 1345 1346 /* 1347 * All directory operations under NFS are synchronous, so fsync() 1348 * is a dummy operation. 1349 */ 1350 static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end, 1351 int datasync) 1352 { 1353 dfprintk(FILE, "NFS: fsync dir(%pD2) datasync %d\n", filp, datasync); 1354 1355 nfs_inc_stats(file_inode(filp), NFSIOS_VFSFSYNC); 1356 return 0; 1357 } 1358 1359 /** 1360 * nfs_force_lookup_revalidate - Mark the directory as having changed 1361 * @dir: pointer to directory inode 1362 * 1363 * This forces the revalidation code in nfs_lookup_revalidate() to do a 1364 * full lookup on all child dentries of 'dir' whenever a change occurs 1365 * on the server that might have invalidated our dcache. 1366 * 1367 * Note that we reserve bit '0' as a tag to let us know when a dentry 1368 * was revalidated while holding a delegation on its inode. 1369 * 1370 * The caller should be holding dir->i_lock 1371 */ 1372 void nfs_force_lookup_revalidate(struct inode *dir) 1373 { 1374 NFS_I(dir)->cache_change_attribute += 2; 1375 } 1376 EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate); 1377 1378 /** 1379 * nfs_verify_change_attribute - Detects NFS remote directory changes 1380 * @dir: pointer to parent directory inode 1381 * @verf: previously saved change attribute 1382 * 1383 * Return "false" if the verifiers doesn't match the change attribute. 1384 * This would usually indicate that the directory contents have changed on 1385 * the server, and that any dentries need revalidating. 1386 */ 1387 static bool nfs_verify_change_attribute(struct inode *dir, unsigned long verf) 1388 { 1389 return (verf & ~1UL) == nfs_save_change_attribute(dir); 1390 } 1391 1392 static void nfs_set_verifier_delegated(unsigned long *verf) 1393 { 1394 *verf |= 1UL; 1395 } 1396 1397 #if IS_ENABLED(CONFIG_NFS_V4) 1398 static void nfs_unset_verifier_delegated(unsigned long *verf) 1399 { 1400 *verf &= ~1UL; 1401 } 1402 #endif /* IS_ENABLED(CONFIG_NFS_V4) */ 1403 1404 static bool nfs_test_verifier_delegated(unsigned long verf) 1405 { 1406 return verf & 1; 1407 } 1408 1409 static bool nfs_verifier_is_delegated(struct dentry *dentry) 1410 { 1411 return nfs_test_verifier_delegated(dentry->d_time); 1412 } 1413 1414 static void nfs_set_verifier_locked(struct dentry *dentry, unsigned long verf) 1415 { 1416 struct inode *inode = d_inode(dentry); 1417 struct inode *dir = d_inode(dentry->d_parent); 1418 1419 if (!nfs_verify_change_attribute(dir, verf)) 1420 return; 1421 if (inode && NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) 1422 nfs_set_verifier_delegated(&verf); 1423 dentry->d_time = verf; 1424 } 1425 1426 /** 1427 * nfs_set_verifier - save a parent directory verifier in the dentry 1428 * @dentry: pointer to dentry 1429 * @verf: verifier to save 1430 * 1431 * Saves the parent directory verifier in @dentry. If the inode has 1432 * a delegation, we also tag the dentry as having been revalidated 1433 * while holding a delegation so that we know we don't have to 1434 * look it up again after a directory change. 1435 */ 1436 void nfs_set_verifier(struct dentry *dentry, unsigned long verf) 1437 { 1438 1439 spin_lock(&dentry->d_lock); 1440 nfs_set_verifier_locked(dentry, verf); 1441 spin_unlock(&dentry->d_lock); 1442 } 1443 EXPORT_SYMBOL_GPL(nfs_set_verifier); 1444 1445 #if IS_ENABLED(CONFIG_NFS_V4) 1446 /** 1447 * nfs_clear_verifier_delegated - clear the dir verifier delegation tag 1448 * @inode: pointer to inode 1449 * 1450 * Iterates through the dentries in the inode alias list and clears 1451 * the tag used to indicate that the dentry has been revalidated 1452 * while holding a delegation. 1453 * This function is intended for use when the delegation is being 1454 * returned or revoked. 1455 */ 1456 void nfs_clear_verifier_delegated(struct inode *inode) 1457 { 1458 struct dentry *alias; 1459 1460 if (!inode) 1461 return; 1462 spin_lock(&inode->i_lock); 1463 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) { 1464 spin_lock(&alias->d_lock); 1465 nfs_unset_verifier_delegated(&alias->d_time); 1466 spin_unlock(&alias->d_lock); 1467 } 1468 spin_unlock(&inode->i_lock); 1469 } 1470 EXPORT_SYMBOL_GPL(nfs_clear_verifier_delegated); 1471 #endif /* IS_ENABLED(CONFIG_NFS_V4) */ 1472 1473 static int nfs_dentry_verify_change(struct inode *dir, struct dentry *dentry) 1474 { 1475 if (nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE) && 1476 d_really_is_negative(dentry)) 1477 return dentry->d_time == inode_peek_iversion_raw(dir); 1478 return nfs_verify_change_attribute(dir, dentry->d_time); 1479 } 1480 1481 /* 1482 * A check for whether or not the parent directory has changed. 1483 * In the case it has, we assume that the dentries are untrustworthy 1484 * and may need to be looked up again. 1485 * If rcu_walk prevents us from performing a full check, return 0. 1486 */ 1487 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry, 1488 int rcu_walk) 1489 { 1490 if (IS_ROOT(dentry)) 1491 return 1; 1492 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE) 1493 return 0; 1494 if (!nfs_dentry_verify_change(dir, dentry)) 1495 return 0; 1496 /* Revalidate nfsi->cache_change_attribute before we declare a match */ 1497 if (nfs_mapping_need_revalidate_inode(dir)) { 1498 if (rcu_walk) 1499 return 0; 1500 if (__nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0) 1501 return 0; 1502 } 1503 if (!nfs_dentry_verify_change(dir, dentry)) 1504 return 0; 1505 return 1; 1506 } 1507 1508 /* 1509 * Use intent information to check whether or not we're going to do 1510 * an O_EXCL create using this path component. 1511 */ 1512 static int nfs_is_exclusive_create(struct inode *dir, unsigned int flags) 1513 { 1514 if (NFS_PROTO(dir)->version == 2) 1515 return 0; 1516 return flags & LOOKUP_EXCL; 1517 } 1518 1519 /* 1520 * Inode and filehandle revalidation for lookups. 1521 * 1522 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL, 1523 * or if the intent information indicates that we're about to open this 1524 * particular file and the "nocto" mount flag is not set. 1525 * 1526 */ 1527 static 1528 int nfs_lookup_verify_inode(struct inode *inode, unsigned int flags) 1529 { 1530 struct nfs_server *server = NFS_SERVER(inode); 1531 int ret; 1532 1533 if (IS_AUTOMOUNT(inode)) 1534 return 0; 1535 1536 if (flags & LOOKUP_OPEN) { 1537 switch (inode->i_mode & S_IFMT) { 1538 case S_IFREG: 1539 /* A NFSv4 OPEN will revalidate later */ 1540 if (server->caps & NFS_CAP_ATOMIC_OPEN) 1541 goto out; 1542 fallthrough; 1543 case S_IFDIR: 1544 if (server->flags & NFS_MOUNT_NOCTO) 1545 break; 1546 /* NFS close-to-open cache consistency validation */ 1547 goto out_force; 1548 } 1549 } 1550 1551 /* VFS wants an on-the-wire revalidation */ 1552 if (flags & LOOKUP_REVAL) 1553 goto out_force; 1554 out: 1555 if (inode->i_nlink > 0 || 1556 (inode->i_nlink == 0 && 1557 test_bit(NFS_INO_PRESERVE_UNLINKED, &NFS_I(inode)->flags))) 1558 return 0; 1559 else 1560 return -ESTALE; 1561 out_force: 1562 if (flags & LOOKUP_RCU) 1563 return -ECHILD; 1564 ret = __nfs_revalidate_inode(server, inode); 1565 if (ret != 0) 1566 return ret; 1567 goto out; 1568 } 1569 1570 static void nfs_mark_dir_for_revalidate(struct inode *inode) 1571 { 1572 spin_lock(&inode->i_lock); 1573 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE); 1574 spin_unlock(&inode->i_lock); 1575 } 1576 1577 /* 1578 * We judge how long we want to trust negative 1579 * dentries by looking at the parent inode mtime. 1580 * 1581 * If parent mtime has changed, we revalidate, else we wait for a 1582 * period corresponding to the parent's attribute cache timeout value. 1583 * 1584 * If LOOKUP_RCU prevents us from performing a full check, return 1 1585 * suggesting a reval is needed. 1586 * 1587 * Note that when creating a new file, or looking up a rename target, 1588 * then it shouldn't be necessary to revalidate a negative dentry. 1589 */ 1590 static inline 1591 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry, 1592 unsigned int flags) 1593 { 1594 if (flags & (LOOKUP_CREATE | LOOKUP_RENAME_TARGET)) 1595 return 0; 1596 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG) 1597 return 1; 1598 /* Case insensitive server? Revalidate negative dentries */ 1599 if (nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE)) 1600 return 1; 1601 return !nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU); 1602 } 1603 1604 static int 1605 nfs_lookup_revalidate_done(struct inode *dir, struct dentry *dentry, 1606 struct inode *inode, int error) 1607 { 1608 switch (error) { 1609 case 1: 1610 break; 1611 case 0: 1612 /* 1613 * We can't d_drop the root of a disconnected tree: 1614 * its d_hash is on the s_anon list and d_drop() would hide 1615 * it from shrink_dcache_for_unmount(), leading to busy 1616 * inodes on unmount and further oopses. 1617 */ 1618 if (inode && IS_ROOT(dentry)) 1619 error = 1; 1620 break; 1621 } 1622 trace_nfs_lookup_revalidate_exit(dir, dentry, 0, error); 1623 return error; 1624 } 1625 1626 static int 1627 nfs_lookup_revalidate_negative(struct inode *dir, struct dentry *dentry, 1628 unsigned int flags) 1629 { 1630 int ret = 1; 1631 if (nfs_neg_need_reval(dir, dentry, flags)) { 1632 if (flags & LOOKUP_RCU) 1633 return -ECHILD; 1634 ret = 0; 1635 } 1636 return nfs_lookup_revalidate_done(dir, dentry, NULL, ret); 1637 } 1638 1639 static int 1640 nfs_lookup_revalidate_delegated(struct inode *dir, struct dentry *dentry, 1641 struct inode *inode) 1642 { 1643 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1644 return nfs_lookup_revalidate_done(dir, dentry, inode, 1); 1645 } 1646 1647 static int nfs_lookup_revalidate_dentry(struct inode *dir, 1648 struct dentry *dentry, 1649 struct inode *inode, unsigned int flags) 1650 { 1651 struct nfs_fh *fhandle; 1652 struct nfs_fattr *fattr; 1653 unsigned long dir_verifier; 1654 int ret; 1655 1656 trace_nfs_lookup_revalidate_enter(dir, dentry, flags); 1657 1658 ret = -ENOMEM; 1659 fhandle = nfs_alloc_fhandle(); 1660 fattr = nfs_alloc_fattr_with_label(NFS_SERVER(inode)); 1661 if (fhandle == NULL || fattr == NULL) 1662 goto out; 1663 1664 dir_verifier = nfs_save_change_attribute(dir); 1665 ret = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr); 1666 if (ret < 0) { 1667 switch (ret) { 1668 case -ESTALE: 1669 case -ENOENT: 1670 ret = 0; 1671 break; 1672 case -ETIMEDOUT: 1673 if (NFS_SERVER(inode)->flags & NFS_MOUNT_SOFTREVAL) 1674 ret = 1; 1675 } 1676 goto out; 1677 } 1678 1679 /* Request help from readdirplus */ 1680 nfs_lookup_advise_force_readdirplus(dir, flags); 1681 1682 ret = 0; 1683 if (nfs_compare_fh(NFS_FH(inode), fhandle)) 1684 goto out; 1685 if (nfs_refresh_inode(inode, fattr) < 0) 1686 goto out; 1687 1688 nfs_setsecurity(inode, fattr); 1689 nfs_set_verifier(dentry, dir_verifier); 1690 1691 ret = 1; 1692 out: 1693 nfs_free_fattr(fattr); 1694 nfs_free_fhandle(fhandle); 1695 1696 /* 1697 * If the lookup failed despite the dentry change attribute being 1698 * a match, then we should revalidate the directory cache. 1699 */ 1700 if (!ret && nfs_dentry_verify_change(dir, dentry)) 1701 nfs_mark_dir_for_revalidate(dir); 1702 return nfs_lookup_revalidate_done(dir, dentry, inode, ret); 1703 } 1704 1705 /* 1706 * This is called every time the dcache has a lookup hit, 1707 * and we should check whether we can really trust that 1708 * lookup. 1709 * 1710 * NOTE! The hit can be a negative hit too, don't assume 1711 * we have an inode! 1712 * 1713 * If the parent directory is seen to have changed, we throw out the 1714 * cached dentry and do a new lookup. 1715 */ 1716 static int 1717 nfs_do_lookup_revalidate(struct inode *dir, struct dentry *dentry, 1718 unsigned int flags) 1719 { 1720 struct inode *inode; 1721 int error; 1722 1723 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE); 1724 inode = d_inode(dentry); 1725 1726 if (!inode) 1727 return nfs_lookup_revalidate_negative(dir, dentry, flags); 1728 1729 if (is_bad_inode(inode)) { 1730 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n", 1731 __func__, dentry); 1732 goto out_bad; 1733 } 1734 1735 if (nfs_verifier_is_delegated(dentry)) 1736 return nfs_lookup_revalidate_delegated(dir, dentry, inode); 1737 1738 /* Force a full look up iff the parent directory has changed */ 1739 if (!(flags & (LOOKUP_EXCL | LOOKUP_REVAL)) && 1740 nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU)) { 1741 error = nfs_lookup_verify_inode(inode, flags); 1742 if (error) { 1743 if (error == -ESTALE) 1744 nfs_mark_dir_for_revalidate(dir); 1745 goto out_bad; 1746 } 1747 goto out_valid; 1748 } 1749 1750 if (flags & LOOKUP_RCU) 1751 return -ECHILD; 1752 1753 if (NFS_STALE(inode)) 1754 goto out_bad; 1755 1756 return nfs_lookup_revalidate_dentry(dir, dentry, inode, flags); 1757 out_valid: 1758 return nfs_lookup_revalidate_done(dir, dentry, inode, 1); 1759 out_bad: 1760 if (flags & LOOKUP_RCU) 1761 return -ECHILD; 1762 return nfs_lookup_revalidate_done(dir, dentry, inode, 0); 1763 } 1764 1765 static int 1766 __nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags, 1767 int (*reval)(struct inode *, struct dentry *, unsigned int)) 1768 { 1769 struct dentry *parent; 1770 struct inode *dir; 1771 int ret; 1772 1773 if (flags & LOOKUP_RCU) { 1774 parent = READ_ONCE(dentry->d_parent); 1775 dir = d_inode_rcu(parent); 1776 if (!dir) 1777 return -ECHILD; 1778 ret = reval(dir, dentry, flags); 1779 if (parent != READ_ONCE(dentry->d_parent)) 1780 return -ECHILD; 1781 } else { 1782 parent = dget_parent(dentry); 1783 ret = reval(d_inode(parent), dentry, flags); 1784 dput(parent); 1785 } 1786 return ret; 1787 } 1788 1789 static int nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags) 1790 { 1791 return __nfs_lookup_revalidate(dentry, flags, nfs_do_lookup_revalidate); 1792 } 1793 1794 /* 1795 * A weaker form of d_revalidate for revalidating just the d_inode(dentry) 1796 * when we don't really care about the dentry name. This is called when a 1797 * pathwalk ends on a dentry that was not found via a normal lookup in the 1798 * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals). 1799 * 1800 * In this situation, we just want to verify that the inode itself is OK 1801 * since the dentry might have changed on the server. 1802 */ 1803 static int nfs_weak_revalidate(struct dentry *dentry, unsigned int flags) 1804 { 1805 struct inode *inode = d_inode(dentry); 1806 int error = 0; 1807 1808 /* 1809 * I believe we can only get a negative dentry here in the case of a 1810 * procfs-style symlink. Just assume it's correct for now, but we may 1811 * eventually need to do something more here. 1812 */ 1813 if (!inode) { 1814 dfprintk(LOOKUPCACHE, "%s: %pd2 has negative inode\n", 1815 __func__, dentry); 1816 return 1; 1817 } 1818 1819 if (is_bad_inode(inode)) { 1820 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n", 1821 __func__, dentry); 1822 return 0; 1823 } 1824 1825 error = nfs_lookup_verify_inode(inode, flags); 1826 dfprintk(LOOKUPCACHE, "NFS: %s: inode %lu is %s\n", 1827 __func__, inode->i_ino, error ? "invalid" : "valid"); 1828 return !error; 1829 } 1830 1831 /* 1832 * This is called from dput() when d_count is going to 0. 1833 */ 1834 static int nfs_dentry_delete(const struct dentry *dentry) 1835 { 1836 dfprintk(VFS, "NFS: dentry_delete(%pd2, %x)\n", 1837 dentry, dentry->d_flags); 1838 1839 /* Unhash any dentry with a stale inode */ 1840 if (d_really_is_positive(dentry) && NFS_STALE(d_inode(dentry))) 1841 return 1; 1842 1843 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { 1844 /* Unhash it, so that ->d_iput() would be called */ 1845 return 1; 1846 } 1847 if (!(dentry->d_sb->s_flags & SB_ACTIVE)) { 1848 /* Unhash it, so that ancestors of killed async unlink 1849 * files will be cleaned up during umount */ 1850 return 1; 1851 } 1852 return 0; 1853 1854 } 1855 1856 /* Ensure that we revalidate inode->i_nlink */ 1857 static void nfs_drop_nlink(struct inode *inode) 1858 { 1859 spin_lock(&inode->i_lock); 1860 /* drop the inode if we're reasonably sure this is the last link */ 1861 if (inode->i_nlink > 0) 1862 drop_nlink(inode); 1863 NFS_I(inode)->attr_gencount = nfs_inc_attr_generation_counter(); 1864 nfs_set_cache_invalid( 1865 inode, NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_CTIME | 1866 NFS_INO_INVALID_NLINK); 1867 spin_unlock(&inode->i_lock); 1868 } 1869 1870 /* 1871 * Called when the dentry loses inode. 1872 * We use it to clean up silly-renamed files. 1873 */ 1874 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode) 1875 { 1876 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { 1877 nfs_complete_unlink(dentry, inode); 1878 nfs_drop_nlink(inode); 1879 } 1880 iput(inode); 1881 } 1882 1883 static void nfs_d_release(struct dentry *dentry) 1884 { 1885 /* free cached devname value, if it survived that far */ 1886 if (unlikely(dentry->d_fsdata)) { 1887 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) 1888 WARN_ON(1); 1889 else 1890 kfree(dentry->d_fsdata); 1891 } 1892 } 1893 1894 const struct dentry_operations nfs_dentry_operations = { 1895 .d_revalidate = nfs_lookup_revalidate, 1896 .d_weak_revalidate = nfs_weak_revalidate, 1897 .d_delete = nfs_dentry_delete, 1898 .d_iput = nfs_dentry_iput, 1899 .d_automount = nfs_d_automount, 1900 .d_release = nfs_d_release, 1901 }; 1902 EXPORT_SYMBOL_GPL(nfs_dentry_operations); 1903 1904 struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags) 1905 { 1906 struct dentry *res; 1907 struct inode *inode = NULL; 1908 struct nfs_fh *fhandle = NULL; 1909 struct nfs_fattr *fattr = NULL; 1910 unsigned long dir_verifier; 1911 int error; 1912 1913 dfprintk(VFS, "NFS: lookup(%pd2)\n", dentry); 1914 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP); 1915 1916 if (unlikely(dentry->d_name.len > NFS_SERVER(dir)->namelen)) 1917 return ERR_PTR(-ENAMETOOLONG); 1918 1919 /* 1920 * If we're doing an exclusive create, optimize away the lookup 1921 * but don't hash the dentry. 1922 */ 1923 if (nfs_is_exclusive_create(dir, flags) || flags & LOOKUP_RENAME_TARGET) 1924 return NULL; 1925 1926 res = ERR_PTR(-ENOMEM); 1927 fhandle = nfs_alloc_fhandle(); 1928 fattr = nfs_alloc_fattr_with_label(NFS_SERVER(dir)); 1929 if (fhandle == NULL || fattr == NULL) 1930 goto out; 1931 1932 dir_verifier = nfs_save_change_attribute(dir); 1933 trace_nfs_lookup_enter(dir, dentry, flags); 1934 error = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr); 1935 if (error == -ENOENT) { 1936 if (nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE)) 1937 dir_verifier = inode_peek_iversion_raw(dir); 1938 goto no_entry; 1939 } 1940 if (error < 0) { 1941 res = ERR_PTR(error); 1942 goto out; 1943 } 1944 inode = nfs_fhget(dentry->d_sb, fhandle, fattr); 1945 res = ERR_CAST(inode); 1946 if (IS_ERR(res)) 1947 goto out; 1948 1949 /* Notify readdir to use READDIRPLUS */ 1950 nfs_lookup_advise_force_readdirplus(dir, flags); 1951 1952 no_entry: 1953 res = d_splice_alias(inode, dentry); 1954 if (res != NULL) { 1955 if (IS_ERR(res)) 1956 goto out; 1957 dentry = res; 1958 } 1959 nfs_set_verifier(dentry, dir_verifier); 1960 out: 1961 trace_nfs_lookup_exit(dir, dentry, flags, PTR_ERR_OR_ZERO(res)); 1962 nfs_free_fattr(fattr); 1963 nfs_free_fhandle(fhandle); 1964 return res; 1965 } 1966 EXPORT_SYMBOL_GPL(nfs_lookup); 1967 1968 void nfs_d_prune_case_insensitive_aliases(struct inode *inode) 1969 { 1970 /* Case insensitive server? Revalidate dentries */ 1971 if (inode && nfs_server_capable(inode, NFS_CAP_CASE_INSENSITIVE)) 1972 d_prune_aliases(inode); 1973 } 1974 EXPORT_SYMBOL_GPL(nfs_d_prune_case_insensitive_aliases); 1975 1976 #if IS_ENABLED(CONFIG_NFS_V4) 1977 static int nfs4_lookup_revalidate(struct dentry *, unsigned int); 1978 1979 const struct dentry_operations nfs4_dentry_operations = { 1980 .d_revalidate = nfs4_lookup_revalidate, 1981 .d_weak_revalidate = nfs_weak_revalidate, 1982 .d_delete = nfs_dentry_delete, 1983 .d_iput = nfs_dentry_iput, 1984 .d_automount = nfs_d_automount, 1985 .d_release = nfs_d_release, 1986 }; 1987 EXPORT_SYMBOL_GPL(nfs4_dentry_operations); 1988 1989 static fmode_t flags_to_mode(int flags) 1990 { 1991 fmode_t res = (__force fmode_t)flags & FMODE_EXEC; 1992 if ((flags & O_ACCMODE) != O_WRONLY) 1993 res |= FMODE_READ; 1994 if ((flags & O_ACCMODE) != O_RDONLY) 1995 res |= FMODE_WRITE; 1996 return res; 1997 } 1998 1999 static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags, struct file *filp) 2000 { 2001 return alloc_nfs_open_context(dentry, flags_to_mode(open_flags), filp); 2002 } 2003 2004 static int do_open(struct inode *inode, struct file *filp) 2005 { 2006 nfs_fscache_open_file(inode, filp); 2007 return 0; 2008 } 2009 2010 static int nfs_finish_open(struct nfs_open_context *ctx, 2011 struct dentry *dentry, 2012 struct file *file, unsigned open_flags) 2013 { 2014 int err; 2015 2016 err = finish_open(file, dentry, do_open); 2017 if (err) 2018 goto out; 2019 if (S_ISREG(file->f_path.dentry->d_inode->i_mode)) 2020 nfs_file_set_open_context(file, ctx); 2021 else 2022 err = -EOPENSTALE; 2023 out: 2024 return err; 2025 } 2026 2027 int nfs_atomic_open(struct inode *dir, struct dentry *dentry, 2028 struct file *file, unsigned open_flags, 2029 umode_t mode) 2030 { 2031 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 2032 struct nfs_open_context *ctx; 2033 struct dentry *res; 2034 struct iattr attr = { .ia_valid = ATTR_OPEN }; 2035 struct inode *inode; 2036 unsigned int lookup_flags = 0; 2037 unsigned long dir_verifier; 2038 bool switched = false; 2039 int created = 0; 2040 int err; 2041 2042 /* Expect a negative dentry */ 2043 BUG_ON(d_inode(dentry)); 2044 2045 dfprintk(VFS, "NFS: atomic_open(%s/%lu), %pd\n", 2046 dir->i_sb->s_id, dir->i_ino, dentry); 2047 2048 err = nfs_check_flags(open_flags); 2049 if (err) 2050 return err; 2051 2052 /* NFS only supports OPEN on regular files */ 2053 if ((open_flags & O_DIRECTORY)) { 2054 if (!d_in_lookup(dentry)) { 2055 /* 2056 * Hashed negative dentry with O_DIRECTORY: dentry was 2057 * revalidated and is fine, no need to perform lookup 2058 * again 2059 */ 2060 return -ENOENT; 2061 } 2062 lookup_flags = LOOKUP_OPEN|LOOKUP_DIRECTORY; 2063 goto no_open; 2064 } 2065 2066 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) 2067 return -ENAMETOOLONG; 2068 2069 if (open_flags & O_CREAT) { 2070 struct nfs_server *server = NFS_SERVER(dir); 2071 2072 if (!(server->attr_bitmask[2] & FATTR4_WORD2_MODE_UMASK)) 2073 mode &= ~current_umask(); 2074 2075 attr.ia_valid |= ATTR_MODE; 2076 attr.ia_mode = mode; 2077 } 2078 if (open_flags & O_TRUNC) { 2079 attr.ia_valid |= ATTR_SIZE; 2080 attr.ia_size = 0; 2081 } 2082 2083 if (!(open_flags & O_CREAT) && !d_in_lookup(dentry)) { 2084 d_drop(dentry); 2085 switched = true; 2086 dentry = d_alloc_parallel(dentry->d_parent, 2087 &dentry->d_name, &wq); 2088 if (IS_ERR(dentry)) 2089 return PTR_ERR(dentry); 2090 if (unlikely(!d_in_lookup(dentry))) 2091 return finish_no_open(file, dentry); 2092 } 2093 2094 ctx = create_nfs_open_context(dentry, open_flags, file); 2095 err = PTR_ERR(ctx); 2096 if (IS_ERR(ctx)) 2097 goto out; 2098 2099 trace_nfs_atomic_open_enter(dir, ctx, open_flags); 2100 inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr, &created); 2101 if (created) 2102 file->f_mode |= FMODE_CREATED; 2103 if (IS_ERR(inode)) { 2104 err = PTR_ERR(inode); 2105 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err); 2106 put_nfs_open_context(ctx); 2107 d_drop(dentry); 2108 switch (err) { 2109 case -ENOENT: 2110 d_splice_alias(NULL, dentry); 2111 if (nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE)) 2112 dir_verifier = inode_peek_iversion_raw(dir); 2113 else 2114 dir_verifier = nfs_save_change_attribute(dir); 2115 nfs_set_verifier(dentry, dir_verifier); 2116 break; 2117 case -EISDIR: 2118 case -ENOTDIR: 2119 goto no_open; 2120 case -ELOOP: 2121 if (!(open_flags & O_NOFOLLOW)) 2122 goto no_open; 2123 break; 2124 /* case -EINVAL: */ 2125 default: 2126 break; 2127 } 2128 goto out; 2129 } 2130 2131 err = nfs_finish_open(ctx, ctx->dentry, file, open_flags); 2132 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err); 2133 put_nfs_open_context(ctx); 2134 out: 2135 if (unlikely(switched)) { 2136 d_lookup_done(dentry); 2137 dput(dentry); 2138 } 2139 return err; 2140 2141 no_open: 2142 res = nfs_lookup(dir, dentry, lookup_flags); 2143 if (!res) { 2144 inode = d_inode(dentry); 2145 if ((lookup_flags & LOOKUP_DIRECTORY) && inode && 2146 !(S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))) 2147 res = ERR_PTR(-ENOTDIR); 2148 else if (inode && S_ISREG(inode->i_mode)) 2149 res = ERR_PTR(-EOPENSTALE); 2150 } else if (!IS_ERR(res)) { 2151 inode = d_inode(res); 2152 if ((lookup_flags & LOOKUP_DIRECTORY) && inode && 2153 !(S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))) { 2154 dput(res); 2155 res = ERR_PTR(-ENOTDIR); 2156 } else if (inode && S_ISREG(inode->i_mode)) { 2157 dput(res); 2158 res = ERR_PTR(-EOPENSTALE); 2159 } 2160 } 2161 if (switched) { 2162 d_lookup_done(dentry); 2163 if (!res) 2164 res = dentry; 2165 else 2166 dput(dentry); 2167 } 2168 if (IS_ERR(res)) 2169 return PTR_ERR(res); 2170 return finish_no_open(file, res); 2171 } 2172 EXPORT_SYMBOL_GPL(nfs_atomic_open); 2173 2174 static int 2175 nfs4_do_lookup_revalidate(struct inode *dir, struct dentry *dentry, 2176 unsigned int flags) 2177 { 2178 struct inode *inode; 2179 2180 if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY)) 2181 goto full_reval; 2182 if (d_mountpoint(dentry)) 2183 goto full_reval; 2184 2185 inode = d_inode(dentry); 2186 2187 /* We can't create new files in nfs_open_revalidate(), so we 2188 * optimize away revalidation of negative dentries. 2189 */ 2190 if (inode == NULL) 2191 goto full_reval; 2192 2193 if (nfs_verifier_is_delegated(dentry)) 2194 return nfs_lookup_revalidate_delegated(dir, dentry, inode); 2195 2196 /* NFS only supports OPEN on regular files */ 2197 if (!S_ISREG(inode->i_mode)) 2198 goto full_reval; 2199 2200 /* We cannot do exclusive creation on a positive dentry */ 2201 if (flags & (LOOKUP_EXCL | LOOKUP_REVAL)) 2202 goto reval_dentry; 2203 2204 /* Check if the directory changed */ 2205 if (!nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU)) 2206 goto reval_dentry; 2207 2208 /* Let f_op->open() actually open (and revalidate) the file */ 2209 return 1; 2210 reval_dentry: 2211 if (flags & LOOKUP_RCU) 2212 return -ECHILD; 2213 return nfs_lookup_revalidate_dentry(dir, dentry, inode, flags); 2214 2215 full_reval: 2216 return nfs_do_lookup_revalidate(dir, dentry, flags); 2217 } 2218 2219 static int nfs4_lookup_revalidate(struct dentry *dentry, unsigned int flags) 2220 { 2221 return __nfs_lookup_revalidate(dentry, flags, 2222 nfs4_do_lookup_revalidate); 2223 } 2224 2225 #endif /* CONFIG_NFSV4 */ 2226 2227 struct dentry * 2228 nfs_add_or_obtain(struct dentry *dentry, struct nfs_fh *fhandle, 2229 struct nfs_fattr *fattr) 2230 { 2231 struct dentry *parent = dget_parent(dentry); 2232 struct inode *dir = d_inode(parent); 2233 struct inode *inode; 2234 struct dentry *d; 2235 int error; 2236 2237 d_drop(dentry); 2238 2239 if (fhandle->size == 0) { 2240 error = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr); 2241 if (error) 2242 goto out_error; 2243 } 2244 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 2245 if (!(fattr->valid & NFS_ATTR_FATTR)) { 2246 struct nfs_server *server = NFS_SB(dentry->d_sb); 2247 error = server->nfs_client->rpc_ops->getattr(server, fhandle, 2248 fattr, NULL); 2249 if (error < 0) 2250 goto out_error; 2251 } 2252 inode = nfs_fhget(dentry->d_sb, fhandle, fattr); 2253 d = d_splice_alias(inode, dentry); 2254 out: 2255 dput(parent); 2256 return d; 2257 out_error: 2258 d = ERR_PTR(error); 2259 goto out; 2260 } 2261 EXPORT_SYMBOL_GPL(nfs_add_or_obtain); 2262 2263 /* 2264 * Code common to create, mkdir, and mknod. 2265 */ 2266 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle, 2267 struct nfs_fattr *fattr) 2268 { 2269 struct dentry *d; 2270 2271 d = nfs_add_or_obtain(dentry, fhandle, fattr); 2272 if (IS_ERR(d)) 2273 return PTR_ERR(d); 2274 2275 /* Callers don't care */ 2276 dput(d); 2277 return 0; 2278 } 2279 EXPORT_SYMBOL_GPL(nfs_instantiate); 2280 2281 /* 2282 * Following a failed create operation, we drop the dentry rather 2283 * than retain a negative dentry. This avoids a problem in the event 2284 * that the operation succeeded on the server, but an error in the 2285 * reply path made it appear to have failed. 2286 */ 2287 int nfs_create(struct user_namespace *mnt_userns, struct inode *dir, 2288 struct dentry *dentry, umode_t mode, bool excl) 2289 { 2290 struct iattr attr; 2291 int open_flags = excl ? O_CREAT | O_EXCL : O_CREAT; 2292 int error; 2293 2294 dfprintk(VFS, "NFS: create(%s/%lu), %pd\n", 2295 dir->i_sb->s_id, dir->i_ino, dentry); 2296 2297 attr.ia_mode = mode; 2298 attr.ia_valid = ATTR_MODE; 2299 2300 trace_nfs_create_enter(dir, dentry, open_flags); 2301 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags); 2302 trace_nfs_create_exit(dir, dentry, open_flags, error); 2303 if (error != 0) 2304 goto out_err; 2305 return 0; 2306 out_err: 2307 d_drop(dentry); 2308 return error; 2309 } 2310 EXPORT_SYMBOL_GPL(nfs_create); 2311 2312 /* 2313 * See comments for nfs_proc_create regarding failed operations. 2314 */ 2315 int 2316 nfs_mknod(struct user_namespace *mnt_userns, struct inode *dir, 2317 struct dentry *dentry, umode_t mode, dev_t rdev) 2318 { 2319 struct iattr attr; 2320 int status; 2321 2322 dfprintk(VFS, "NFS: mknod(%s/%lu), %pd\n", 2323 dir->i_sb->s_id, dir->i_ino, dentry); 2324 2325 attr.ia_mode = mode; 2326 attr.ia_valid = ATTR_MODE; 2327 2328 trace_nfs_mknod_enter(dir, dentry); 2329 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev); 2330 trace_nfs_mknod_exit(dir, dentry, status); 2331 if (status != 0) 2332 goto out_err; 2333 return 0; 2334 out_err: 2335 d_drop(dentry); 2336 return status; 2337 } 2338 EXPORT_SYMBOL_GPL(nfs_mknod); 2339 2340 /* 2341 * See comments for nfs_proc_create regarding failed operations. 2342 */ 2343 int nfs_mkdir(struct user_namespace *mnt_userns, struct inode *dir, 2344 struct dentry *dentry, umode_t mode) 2345 { 2346 struct iattr attr; 2347 int error; 2348 2349 dfprintk(VFS, "NFS: mkdir(%s/%lu), %pd\n", 2350 dir->i_sb->s_id, dir->i_ino, dentry); 2351 2352 attr.ia_valid = ATTR_MODE; 2353 attr.ia_mode = mode | S_IFDIR; 2354 2355 trace_nfs_mkdir_enter(dir, dentry); 2356 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr); 2357 trace_nfs_mkdir_exit(dir, dentry, error); 2358 if (error != 0) 2359 goto out_err; 2360 return 0; 2361 out_err: 2362 d_drop(dentry); 2363 return error; 2364 } 2365 EXPORT_SYMBOL_GPL(nfs_mkdir); 2366 2367 static void nfs_dentry_handle_enoent(struct dentry *dentry) 2368 { 2369 if (simple_positive(dentry)) 2370 d_delete(dentry); 2371 } 2372 2373 static void nfs_dentry_remove_handle_error(struct inode *dir, 2374 struct dentry *dentry, int error) 2375 { 2376 switch (error) { 2377 case -ENOENT: 2378 d_delete(dentry); 2379 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 2380 break; 2381 case 0: 2382 nfs_d_prune_case_insensitive_aliases(d_inode(dentry)); 2383 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 2384 } 2385 } 2386 2387 int nfs_rmdir(struct inode *dir, struct dentry *dentry) 2388 { 2389 int error; 2390 2391 dfprintk(VFS, "NFS: rmdir(%s/%lu), %pd\n", 2392 dir->i_sb->s_id, dir->i_ino, dentry); 2393 2394 trace_nfs_rmdir_enter(dir, dentry); 2395 if (d_really_is_positive(dentry)) { 2396 down_write(&NFS_I(d_inode(dentry))->rmdir_sem); 2397 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name); 2398 /* Ensure the VFS deletes this inode */ 2399 switch (error) { 2400 case 0: 2401 clear_nlink(d_inode(dentry)); 2402 break; 2403 case -ENOENT: 2404 nfs_dentry_handle_enoent(dentry); 2405 } 2406 up_write(&NFS_I(d_inode(dentry))->rmdir_sem); 2407 } else 2408 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name); 2409 nfs_dentry_remove_handle_error(dir, dentry, error); 2410 trace_nfs_rmdir_exit(dir, dentry, error); 2411 2412 return error; 2413 } 2414 EXPORT_SYMBOL_GPL(nfs_rmdir); 2415 2416 /* 2417 * Remove a file after making sure there are no pending writes, 2418 * and after checking that the file has only one user. 2419 * 2420 * We invalidate the attribute cache and free the inode prior to the operation 2421 * to avoid possible races if the server reuses the inode. 2422 */ 2423 static int nfs_safe_remove(struct dentry *dentry) 2424 { 2425 struct inode *dir = d_inode(dentry->d_parent); 2426 struct inode *inode = d_inode(dentry); 2427 int error = -EBUSY; 2428 2429 dfprintk(VFS, "NFS: safe_remove(%pd2)\n", dentry); 2430 2431 /* If the dentry was sillyrenamed, we simply call d_delete() */ 2432 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { 2433 error = 0; 2434 goto out; 2435 } 2436 2437 trace_nfs_remove_enter(dir, dentry); 2438 if (inode != NULL) { 2439 error = NFS_PROTO(dir)->remove(dir, dentry); 2440 if (error == 0) 2441 nfs_drop_nlink(inode); 2442 } else 2443 error = NFS_PROTO(dir)->remove(dir, dentry); 2444 if (error == -ENOENT) 2445 nfs_dentry_handle_enoent(dentry); 2446 trace_nfs_remove_exit(dir, dentry, error); 2447 out: 2448 return error; 2449 } 2450 2451 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode 2452 * belongs to an active ".nfs..." file and we return -EBUSY. 2453 * 2454 * If sillyrename() returns 0, we do nothing, otherwise we unlink. 2455 */ 2456 int nfs_unlink(struct inode *dir, struct dentry *dentry) 2457 { 2458 int error; 2459 int need_rehash = 0; 2460 2461 dfprintk(VFS, "NFS: unlink(%s/%lu, %pd)\n", dir->i_sb->s_id, 2462 dir->i_ino, dentry); 2463 2464 trace_nfs_unlink_enter(dir, dentry); 2465 spin_lock(&dentry->d_lock); 2466 if (d_count(dentry) > 1 && !test_bit(NFS_INO_PRESERVE_UNLINKED, 2467 &NFS_I(d_inode(dentry))->flags)) { 2468 spin_unlock(&dentry->d_lock); 2469 /* Start asynchronous writeout of the inode */ 2470 write_inode_now(d_inode(dentry), 0); 2471 error = nfs_sillyrename(dir, dentry); 2472 goto out; 2473 } 2474 if (!d_unhashed(dentry)) { 2475 __d_drop(dentry); 2476 need_rehash = 1; 2477 } 2478 spin_unlock(&dentry->d_lock); 2479 error = nfs_safe_remove(dentry); 2480 nfs_dentry_remove_handle_error(dir, dentry, error); 2481 if (need_rehash) 2482 d_rehash(dentry); 2483 out: 2484 trace_nfs_unlink_exit(dir, dentry, error); 2485 return error; 2486 } 2487 EXPORT_SYMBOL_GPL(nfs_unlink); 2488 2489 /* 2490 * To create a symbolic link, most file systems instantiate a new inode, 2491 * add a page to it containing the path, then write it out to the disk 2492 * using prepare_write/commit_write. 2493 * 2494 * Unfortunately the NFS client can't create the in-core inode first 2495 * because it needs a file handle to create an in-core inode (see 2496 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the 2497 * symlink request has completed on the server. 2498 * 2499 * So instead we allocate a raw page, copy the symname into it, then do 2500 * the SYMLINK request with the page as the buffer. If it succeeds, we 2501 * now have a new file handle and can instantiate an in-core NFS inode 2502 * and move the raw page into its mapping. 2503 */ 2504 int nfs_symlink(struct user_namespace *mnt_userns, struct inode *dir, 2505 struct dentry *dentry, const char *symname) 2506 { 2507 struct page *page; 2508 char *kaddr; 2509 struct iattr attr; 2510 unsigned int pathlen = strlen(symname); 2511 int error; 2512 2513 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s)\n", dir->i_sb->s_id, 2514 dir->i_ino, dentry, symname); 2515 2516 if (pathlen > PAGE_SIZE) 2517 return -ENAMETOOLONG; 2518 2519 attr.ia_mode = S_IFLNK | S_IRWXUGO; 2520 attr.ia_valid = ATTR_MODE; 2521 2522 page = alloc_page(GFP_USER); 2523 if (!page) 2524 return -ENOMEM; 2525 2526 kaddr = page_address(page); 2527 memcpy(kaddr, symname, pathlen); 2528 if (pathlen < PAGE_SIZE) 2529 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen); 2530 2531 trace_nfs_symlink_enter(dir, dentry); 2532 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr); 2533 trace_nfs_symlink_exit(dir, dentry, error); 2534 if (error != 0) { 2535 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s) error %d\n", 2536 dir->i_sb->s_id, dir->i_ino, 2537 dentry, symname, error); 2538 d_drop(dentry); 2539 __free_page(page); 2540 return error; 2541 } 2542 2543 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 2544 2545 /* 2546 * No big deal if we can't add this page to the page cache here. 2547 * READLINK will get the missing page from the server if needed. 2548 */ 2549 if (!add_to_page_cache_lru(page, d_inode(dentry)->i_mapping, 0, 2550 GFP_KERNEL)) { 2551 SetPageUptodate(page); 2552 unlock_page(page); 2553 /* 2554 * add_to_page_cache_lru() grabs an extra page refcount. 2555 * Drop it here to avoid leaking this page later. 2556 */ 2557 put_page(page); 2558 } else 2559 __free_page(page); 2560 2561 return 0; 2562 } 2563 EXPORT_SYMBOL_GPL(nfs_symlink); 2564 2565 int 2566 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) 2567 { 2568 struct inode *inode = d_inode(old_dentry); 2569 int error; 2570 2571 dfprintk(VFS, "NFS: link(%pd2 -> %pd2)\n", 2572 old_dentry, dentry); 2573 2574 trace_nfs_link_enter(inode, dir, dentry); 2575 d_drop(dentry); 2576 if (S_ISREG(inode->i_mode)) 2577 nfs_sync_inode(inode); 2578 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name); 2579 if (error == 0) { 2580 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 2581 ihold(inode); 2582 d_add(dentry, inode); 2583 } 2584 trace_nfs_link_exit(inode, dir, dentry, error); 2585 return error; 2586 } 2587 EXPORT_SYMBOL_GPL(nfs_link); 2588 2589 /* 2590 * RENAME 2591 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a 2592 * different file handle for the same inode after a rename (e.g. when 2593 * moving to a different directory). A fail-safe method to do so would 2594 * be to look up old_dir/old_name, create a link to new_dir/new_name and 2595 * rename the old file using the sillyrename stuff. This way, the original 2596 * file in old_dir will go away when the last process iput()s the inode. 2597 * 2598 * FIXED. 2599 * 2600 * It actually works quite well. One needs to have the possibility for 2601 * at least one ".nfs..." file in each directory the file ever gets 2602 * moved or linked to which happens automagically with the new 2603 * implementation that only depends on the dcache stuff instead of 2604 * using the inode layer 2605 * 2606 * Unfortunately, things are a little more complicated than indicated 2607 * above. For a cross-directory move, we want to make sure we can get 2608 * rid of the old inode after the operation. This means there must be 2609 * no pending writes (if it's a file), and the use count must be 1. 2610 * If these conditions are met, we can drop the dentries before doing 2611 * the rename. 2612 */ 2613 int nfs_rename(struct user_namespace *mnt_userns, struct inode *old_dir, 2614 struct dentry *old_dentry, struct inode *new_dir, 2615 struct dentry *new_dentry, unsigned int flags) 2616 { 2617 struct inode *old_inode = d_inode(old_dentry); 2618 struct inode *new_inode = d_inode(new_dentry); 2619 struct dentry *dentry = NULL, *rehash = NULL; 2620 struct rpc_task *task; 2621 int error = -EBUSY; 2622 2623 if (flags) 2624 return -EINVAL; 2625 2626 dfprintk(VFS, "NFS: rename(%pd2 -> %pd2, ct=%d)\n", 2627 old_dentry, new_dentry, 2628 d_count(new_dentry)); 2629 2630 trace_nfs_rename_enter(old_dir, old_dentry, new_dir, new_dentry); 2631 /* 2632 * For non-directories, check whether the target is busy and if so, 2633 * make a copy of the dentry and then do a silly-rename. If the 2634 * silly-rename succeeds, the copied dentry is hashed and becomes 2635 * the new target. 2636 */ 2637 if (new_inode && !S_ISDIR(new_inode->i_mode)) { 2638 /* 2639 * To prevent any new references to the target during the 2640 * rename, we unhash the dentry in advance. 2641 */ 2642 if (!d_unhashed(new_dentry)) { 2643 d_drop(new_dentry); 2644 rehash = new_dentry; 2645 } 2646 2647 if (d_count(new_dentry) > 2) { 2648 int err; 2649 2650 /* copy the target dentry's name */ 2651 dentry = d_alloc(new_dentry->d_parent, 2652 &new_dentry->d_name); 2653 if (!dentry) 2654 goto out; 2655 2656 /* silly-rename the existing target ... */ 2657 err = nfs_sillyrename(new_dir, new_dentry); 2658 if (err) 2659 goto out; 2660 2661 new_dentry = dentry; 2662 rehash = NULL; 2663 new_inode = NULL; 2664 } 2665 } 2666 2667 if (S_ISREG(old_inode->i_mode)) 2668 nfs_sync_inode(old_inode); 2669 task = nfs_async_rename(old_dir, new_dir, old_dentry, new_dentry, NULL); 2670 if (IS_ERR(task)) { 2671 error = PTR_ERR(task); 2672 goto out; 2673 } 2674 2675 error = rpc_wait_for_completion_task(task); 2676 if (error != 0) { 2677 ((struct nfs_renamedata *)task->tk_calldata)->cancelled = 1; 2678 /* Paired with the atomic_dec_and_test() barrier in rpc_do_put_task() */ 2679 smp_wmb(); 2680 } else 2681 error = task->tk_status; 2682 rpc_put_task(task); 2683 /* Ensure the inode attributes are revalidated */ 2684 if (error == 0) { 2685 spin_lock(&old_inode->i_lock); 2686 NFS_I(old_inode)->attr_gencount = nfs_inc_attr_generation_counter(); 2687 nfs_set_cache_invalid(old_inode, NFS_INO_INVALID_CHANGE | 2688 NFS_INO_INVALID_CTIME | 2689 NFS_INO_REVAL_FORCED); 2690 spin_unlock(&old_inode->i_lock); 2691 } 2692 out: 2693 if (rehash) 2694 d_rehash(rehash); 2695 trace_nfs_rename_exit(old_dir, old_dentry, 2696 new_dir, new_dentry, error); 2697 if (!error) { 2698 if (new_inode != NULL) 2699 nfs_drop_nlink(new_inode); 2700 /* 2701 * The d_move() should be here instead of in an async RPC completion 2702 * handler because we need the proper locks to move the dentry. If 2703 * we're interrupted by a signal, the async RPC completion handler 2704 * should mark the directories for revalidation. 2705 */ 2706 d_move(old_dentry, new_dentry); 2707 nfs_set_verifier(old_dentry, 2708 nfs_save_change_attribute(new_dir)); 2709 } else if (error == -ENOENT) 2710 nfs_dentry_handle_enoent(old_dentry); 2711 2712 /* new dentry created? */ 2713 if (dentry) 2714 dput(dentry); 2715 return error; 2716 } 2717 EXPORT_SYMBOL_GPL(nfs_rename); 2718 2719 static DEFINE_SPINLOCK(nfs_access_lru_lock); 2720 static LIST_HEAD(nfs_access_lru_list); 2721 static atomic_long_t nfs_access_nr_entries; 2722 2723 static unsigned long nfs_access_max_cachesize = 4*1024*1024; 2724 module_param(nfs_access_max_cachesize, ulong, 0644); 2725 MODULE_PARM_DESC(nfs_access_max_cachesize, "NFS access maximum total cache length"); 2726 2727 static void nfs_access_free_entry(struct nfs_access_entry *entry) 2728 { 2729 put_group_info(entry->group_info); 2730 kfree_rcu(entry, rcu_head); 2731 smp_mb__before_atomic(); 2732 atomic_long_dec(&nfs_access_nr_entries); 2733 smp_mb__after_atomic(); 2734 } 2735 2736 static void nfs_access_free_list(struct list_head *head) 2737 { 2738 struct nfs_access_entry *cache; 2739 2740 while (!list_empty(head)) { 2741 cache = list_entry(head->next, struct nfs_access_entry, lru); 2742 list_del(&cache->lru); 2743 nfs_access_free_entry(cache); 2744 } 2745 } 2746 2747 static unsigned long 2748 nfs_do_access_cache_scan(unsigned int nr_to_scan) 2749 { 2750 LIST_HEAD(head); 2751 struct nfs_inode *nfsi, *next; 2752 struct nfs_access_entry *cache; 2753 long freed = 0; 2754 2755 spin_lock(&nfs_access_lru_lock); 2756 list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) { 2757 struct inode *inode; 2758 2759 if (nr_to_scan-- == 0) 2760 break; 2761 inode = &nfsi->vfs_inode; 2762 spin_lock(&inode->i_lock); 2763 if (list_empty(&nfsi->access_cache_entry_lru)) 2764 goto remove_lru_entry; 2765 cache = list_entry(nfsi->access_cache_entry_lru.next, 2766 struct nfs_access_entry, lru); 2767 list_move(&cache->lru, &head); 2768 rb_erase(&cache->rb_node, &nfsi->access_cache); 2769 freed++; 2770 if (!list_empty(&nfsi->access_cache_entry_lru)) 2771 list_move_tail(&nfsi->access_cache_inode_lru, 2772 &nfs_access_lru_list); 2773 else { 2774 remove_lru_entry: 2775 list_del_init(&nfsi->access_cache_inode_lru); 2776 smp_mb__before_atomic(); 2777 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags); 2778 smp_mb__after_atomic(); 2779 } 2780 spin_unlock(&inode->i_lock); 2781 } 2782 spin_unlock(&nfs_access_lru_lock); 2783 nfs_access_free_list(&head); 2784 return freed; 2785 } 2786 2787 unsigned long 2788 nfs_access_cache_scan(struct shrinker *shrink, struct shrink_control *sc) 2789 { 2790 int nr_to_scan = sc->nr_to_scan; 2791 gfp_t gfp_mask = sc->gfp_mask; 2792 2793 if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL) 2794 return SHRINK_STOP; 2795 return nfs_do_access_cache_scan(nr_to_scan); 2796 } 2797 2798 2799 unsigned long 2800 nfs_access_cache_count(struct shrinker *shrink, struct shrink_control *sc) 2801 { 2802 return vfs_pressure_ratio(atomic_long_read(&nfs_access_nr_entries)); 2803 } 2804 2805 static void 2806 nfs_access_cache_enforce_limit(void) 2807 { 2808 long nr_entries = atomic_long_read(&nfs_access_nr_entries); 2809 unsigned long diff; 2810 unsigned int nr_to_scan; 2811 2812 if (nr_entries < 0 || nr_entries <= nfs_access_max_cachesize) 2813 return; 2814 nr_to_scan = 100; 2815 diff = nr_entries - nfs_access_max_cachesize; 2816 if (diff < nr_to_scan) 2817 nr_to_scan = diff; 2818 nfs_do_access_cache_scan(nr_to_scan); 2819 } 2820 2821 static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head) 2822 { 2823 struct rb_root *root_node = &nfsi->access_cache; 2824 struct rb_node *n; 2825 struct nfs_access_entry *entry; 2826 2827 /* Unhook entries from the cache */ 2828 while ((n = rb_first(root_node)) != NULL) { 2829 entry = rb_entry(n, struct nfs_access_entry, rb_node); 2830 rb_erase(n, root_node); 2831 list_move(&entry->lru, head); 2832 } 2833 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS; 2834 } 2835 2836 void nfs_access_zap_cache(struct inode *inode) 2837 { 2838 LIST_HEAD(head); 2839 2840 if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0) 2841 return; 2842 /* Remove from global LRU init */ 2843 spin_lock(&nfs_access_lru_lock); 2844 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) 2845 list_del_init(&NFS_I(inode)->access_cache_inode_lru); 2846 2847 spin_lock(&inode->i_lock); 2848 __nfs_access_zap_cache(NFS_I(inode), &head); 2849 spin_unlock(&inode->i_lock); 2850 spin_unlock(&nfs_access_lru_lock); 2851 nfs_access_free_list(&head); 2852 } 2853 EXPORT_SYMBOL_GPL(nfs_access_zap_cache); 2854 2855 static int access_cmp(const struct cred *a, const struct nfs_access_entry *b) 2856 { 2857 struct group_info *ga, *gb; 2858 int g; 2859 2860 if (uid_lt(a->fsuid, b->fsuid)) 2861 return -1; 2862 if (uid_gt(a->fsuid, b->fsuid)) 2863 return 1; 2864 2865 if (gid_lt(a->fsgid, b->fsgid)) 2866 return -1; 2867 if (gid_gt(a->fsgid, b->fsgid)) 2868 return 1; 2869 2870 ga = a->group_info; 2871 gb = b->group_info; 2872 if (ga == gb) 2873 return 0; 2874 if (ga == NULL) 2875 return -1; 2876 if (gb == NULL) 2877 return 1; 2878 if (ga->ngroups < gb->ngroups) 2879 return -1; 2880 if (ga->ngroups > gb->ngroups) 2881 return 1; 2882 2883 for (g = 0; g < ga->ngroups; g++) { 2884 if (gid_lt(ga->gid[g], gb->gid[g])) 2885 return -1; 2886 if (gid_gt(ga->gid[g], gb->gid[g])) 2887 return 1; 2888 } 2889 return 0; 2890 } 2891 2892 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, const struct cred *cred) 2893 { 2894 struct rb_node *n = NFS_I(inode)->access_cache.rb_node; 2895 2896 while (n != NULL) { 2897 struct nfs_access_entry *entry = 2898 rb_entry(n, struct nfs_access_entry, rb_node); 2899 int cmp = access_cmp(cred, entry); 2900 2901 if (cmp < 0) 2902 n = n->rb_left; 2903 else if (cmp > 0) 2904 n = n->rb_right; 2905 else 2906 return entry; 2907 } 2908 return NULL; 2909 } 2910 2911 static int nfs_access_get_cached_locked(struct inode *inode, const struct cred *cred, u32 *mask, bool may_block) 2912 { 2913 struct nfs_inode *nfsi = NFS_I(inode); 2914 struct nfs_access_entry *cache; 2915 bool retry = true; 2916 int err; 2917 2918 spin_lock(&inode->i_lock); 2919 for(;;) { 2920 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS) 2921 goto out_zap; 2922 cache = nfs_access_search_rbtree(inode, cred); 2923 err = -ENOENT; 2924 if (cache == NULL) 2925 goto out; 2926 /* Found an entry, is our attribute cache valid? */ 2927 if (!nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS)) 2928 break; 2929 if (!retry) 2930 break; 2931 err = -ECHILD; 2932 if (!may_block) 2933 goto out; 2934 spin_unlock(&inode->i_lock); 2935 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode); 2936 if (err) 2937 return err; 2938 spin_lock(&inode->i_lock); 2939 retry = false; 2940 } 2941 *mask = cache->mask; 2942 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru); 2943 err = 0; 2944 out: 2945 spin_unlock(&inode->i_lock); 2946 return err; 2947 out_zap: 2948 spin_unlock(&inode->i_lock); 2949 nfs_access_zap_cache(inode); 2950 return -ENOENT; 2951 } 2952 2953 static int nfs_access_get_cached_rcu(struct inode *inode, const struct cred *cred, u32 *mask) 2954 { 2955 /* Only check the most recently returned cache entry, 2956 * but do it without locking. 2957 */ 2958 struct nfs_inode *nfsi = NFS_I(inode); 2959 struct nfs_access_entry *cache; 2960 int err = -ECHILD; 2961 struct list_head *lh; 2962 2963 rcu_read_lock(); 2964 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS) 2965 goto out; 2966 lh = rcu_dereference(list_tail_rcu(&nfsi->access_cache_entry_lru)); 2967 cache = list_entry(lh, struct nfs_access_entry, lru); 2968 if (lh == &nfsi->access_cache_entry_lru || 2969 access_cmp(cred, cache) != 0) 2970 cache = NULL; 2971 if (cache == NULL) 2972 goto out; 2973 if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS)) 2974 goto out; 2975 *mask = cache->mask; 2976 err = 0; 2977 out: 2978 rcu_read_unlock(); 2979 return err; 2980 } 2981 2982 int nfs_access_get_cached(struct inode *inode, const struct cred *cred, 2983 u32 *mask, bool may_block) 2984 { 2985 int status; 2986 2987 status = nfs_access_get_cached_rcu(inode, cred, mask); 2988 if (status != 0) 2989 status = nfs_access_get_cached_locked(inode, cred, mask, 2990 may_block); 2991 2992 return status; 2993 } 2994 EXPORT_SYMBOL_GPL(nfs_access_get_cached); 2995 2996 static void nfs_access_add_rbtree(struct inode *inode, 2997 struct nfs_access_entry *set, 2998 const struct cred *cred) 2999 { 3000 struct nfs_inode *nfsi = NFS_I(inode); 3001 struct rb_root *root_node = &nfsi->access_cache; 3002 struct rb_node **p = &root_node->rb_node; 3003 struct rb_node *parent = NULL; 3004 struct nfs_access_entry *entry; 3005 int cmp; 3006 3007 spin_lock(&inode->i_lock); 3008 while (*p != NULL) { 3009 parent = *p; 3010 entry = rb_entry(parent, struct nfs_access_entry, rb_node); 3011 cmp = access_cmp(cred, entry); 3012 3013 if (cmp < 0) 3014 p = &parent->rb_left; 3015 else if (cmp > 0) 3016 p = &parent->rb_right; 3017 else 3018 goto found; 3019 } 3020 rb_link_node(&set->rb_node, parent, p); 3021 rb_insert_color(&set->rb_node, root_node); 3022 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); 3023 spin_unlock(&inode->i_lock); 3024 return; 3025 found: 3026 rb_replace_node(parent, &set->rb_node, root_node); 3027 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); 3028 list_del(&entry->lru); 3029 spin_unlock(&inode->i_lock); 3030 nfs_access_free_entry(entry); 3031 } 3032 3033 void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set, 3034 const struct cred *cred) 3035 { 3036 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL); 3037 if (cache == NULL) 3038 return; 3039 RB_CLEAR_NODE(&cache->rb_node); 3040 cache->fsuid = cred->fsuid; 3041 cache->fsgid = cred->fsgid; 3042 cache->group_info = get_group_info(cred->group_info); 3043 cache->mask = set->mask; 3044 3045 /* The above field assignments must be visible 3046 * before this item appears on the lru. We cannot easily 3047 * use rcu_assign_pointer, so just force the memory barrier. 3048 */ 3049 smp_wmb(); 3050 nfs_access_add_rbtree(inode, cache, cred); 3051 3052 /* Update accounting */ 3053 smp_mb__before_atomic(); 3054 atomic_long_inc(&nfs_access_nr_entries); 3055 smp_mb__after_atomic(); 3056 3057 /* Add inode to global LRU list */ 3058 if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) { 3059 spin_lock(&nfs_access_lru_lock); 3060 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) 3061 list_add_tail(&NFS_I(inode)->access_cache_inode_lru, 3062 &nfs_access_lru_list); 3063 spin_unlock(&nfs_access_lru_lock); 3064 } 3065 nfs_access_cache_enforce_limit(); 3066 } 3067 EXPORT_SYMBOL_GPL(nfs_access_add_cache); 3068 3069 #define NFS_MAY_READ (NFS_ACCESS_READ) 3070 #define NFS_MAY_WRITE (NFS_ACCESS_MODIFY | \ 3071 NFS_ACCESS_EXTEND | \ 3072 NFS_ACCESS_DELETE) 3073 #define NFS_FILE_MAY_WRITE (NFS_ACCESS_MODIFY | \ 3074 NFS_ACCESS_EXTEND) 3075 #define NFS_DIR_MAY_WRITE NFS_MAY_WRITE 3076 #define NFS_MAY_LOOKUP (NFS_ACCESS_LOOKUP) 3077 #define NFS_MAY_EXECUTE (NFS_ACCESS_EXECUTE) 3078 static int 3079 nfs_access_calc_mask(u32 access_result, umode_t umode) 3080 { 3081 int mask = 0; 3082 3083 if (access_result & NFS_MAY_READ) 3084 mask |= MAY_READ; 3085 if (S_ISDIR(umode)) { 3086 if ((access_result & NFS_DIR_MAY_WRITE) == NFS_DIR_MAY_WRITE) 3087 mask |= MAY_WRITE; 3088 if ((access_result & NFS_MAY_LOOKUP) == NFS_MAY_LOOKUP) 3089 mask |= MAY_EXEC; 3090 } else if (S_ISREG(umode)) { 3091 if ((access_result & NFS_FILE_MAY_WRITE) == NFS_FILE_MAY_WRITE) 3092 mask |= MAY_WRITE; 3093 if ((access_result & NFS_MAY_EXECUTE) == NFS_MAY_EXECUTE) 3094 mask |= MAY_EXEC; 3095 } else if (access_result & NFS_MAY_WRITE) 3096 mask |= MAY_WRITE; 3097 return mask; 3098 } 3099 3100 void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result) 3101 { 3102 entry->mask = access_result; 3103 } 3104 EXPORT_SYMBOL_GPL(nfs_access_set_mask); 3105 3106 static int nfs_do_access(struct inode *inode, const struct cred *cred, int mask) 3107 { 3108 struct nfs_access_entry cache; 3109 bool may_block = (mask & MAY_NOT_BLOCK) == 0; 3110 int cache_mask = -1; 3111 int status; 3112 3113 trace_nfs_access_enter(inode); 3114 3115 status = nfs_access_get_cached(inode, cred, &cache.mask, may_block); 3116 if (status == 0) 3117 goto out_cached; 3118 3119 status = -ECHILD; 3120 if (!may_block) 3121 goto out; 3122 3123 /* 3124 * Determine which access bits we want to ask for... 3125 */ 3126 cache.mask = NFS_ACCESS_READ | NFS_ACCESS_MODIFY | NFS_ACCESS_EXTEND | 3127 nfs_access_xattr_mask(NFS_SERVER(inode)); 3128 if (S_ISDIR(inode->i_mode)) 3129 cache.mask |= NFS_ACCESS_DELETE | NFS_ACCESS_LOOKUP; 3130 else 3131 cache.mask |= NFS_ACCESS_EXECUTE; 3132 status = NFS_PROTO(inode)->access(inode, &cache, cred); 3133 if (status != 0) { 3134 if (status == -ESTALE) { 3135 if (!S_ISDIR(inode->i_mode)) 3136 nfs_set_inode_stale(inode); 3137 else 3138 nfs_zap_caches(inode); 3139 } 3140 goto out; 3141 } 3142 nfs_access_add_cache(inode, &cache, cred); 3143 out_cached: 3144 cache_mask = nfs_access_calc_mask(cache.mask, inode->i_mode); 3145 if ((mask & ~cache_mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) != 0) 3146 status = -EACCES; 3147 out: 3148 trace_nfs_access_exit(inode, mask, cache_mask, status); 3149 return status; 3150 } 3151 3152 static int nfs_open_permission_mask(int openflags) 3153 { 3154 int mask = 0; 3155 3156 if (openflags & __FMODE_EXEC) { 3157 /* ONLY check exec rights */ 3158 mask = MAY_EXEC; 3159 } else { 3160 if ((openflags & O_ACCMODE) != O_WRONLY) 3161 mask |= MAY_READ; 3162 if ((openflags & O_ACCMODE) != O_RDONLY) 3163 mask |= MAY_WRITE; 3164 } 3165 3166 return mask; 3167 } 3168 3169 int nfs_may_open(struct inode *inode, const struct cred *cred, int openflags) 3170 { 3171 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags)); 3172 } 3173 EXPORT_SYMBOL_GPL(nfs_may_open); 3174 3175 static int nfs_execute_ok(struct inode *inode, int mask) 3176 { 3177 struct nfs_server *server = NFS_SERVER(inode); 3178 int ret = 0; 3179 3180 if (S_ISDIR(inode->i_mode)) 3181 return 0; 3182 if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_MODE)) { 3183 if (mask & MAY_NOT_BLOCK) 3184 return -ECHILD; 3185 ret = __nfs_revalidate_inode(server, inode); 3186 } 3187 if (ret == 0 && !execute_ok(inode)) 3188 ret = -EACCES; 3189 return ret; 3190 } 3191 3192 int nfs_permission(struct user_namespace *mnt_userns, 3193 struct inode *inode, 3194 int mask) 3195 { 3196 const struct cred *cred = current_cred(); 3197 int res = 0; 3198 3199 nfs_inc_stats(inode, NFSIOS_VFSACCESS); 3200 3201 if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) 3202 goto out; 3203 /* Is this sys_access() ? */ 3204 if (mask & (MAY_ACCESS | MAY_CHDIR)) 3205 goto force_lookup; 3206 3207 switch (inode->i_mode & S_IFMT) { 3208 case S_IFLNK: 3209 goto out; 3210 case S_IFREG: 3211 if ((mask & MAY_OPEN) && 3212 nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN)) 3213 return 0; 3214 break; 3215 case S_IFDIR: 3216 /* 3217 * Optimize away all write operations, since the server 3218 * will check permissions when we perform the op. 3219 */ 3220 if ((mask & MAY_WRITE) && !(mask & MAY_READ)) 3221 goto out; 3222 } 3223 3224 force_lookup: 3225 if (!NFS_PROTO(inode)->access) 3226 goto out_notsup; 3227 3228 res = nfs_do_access(inode, cred, mask); 3229 out: 3230 if (!res && (mask & MAY_EXEC)) 3231 res = nfs_execute_ok(inode, mask); 3232 3233 dfprintk(VFS, "NFS: permission(%s/%lu), mask=0x%x, res=%d\n", 3234 inode->i_sb->s_id, inode->i_ino, mask, res); 3235 return res; 3236 out_notsup: 3237 if (mask & MAY_NOT_BLOCK) 3238 return -ECHILD; 3239 3240 res = nfs_revalidate_inode(inode, NFS_INO_INVALID_MODE | 3241 NFS_INO_INVALID_OTHER); 3242 if (res == 0) 3243 res = generic_permission(&init_user_ns, inode, mask); 3244 goto out; 3245 } 3246 EXPORT_SYMBOL_GPL(nfs_permission); 3247