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