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