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