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 (switched) { 1998 d_lookup_done(dentry); 1999 if (!res) 2000 res = dentry; 2001 else 2002 dput(dentry); 2003 } 2004 if (IS_ERR(res)) 2005 return PTR_ERR(res); 2006 return finish_no_open(file, res); 2007 } 2008 EXPORT_SYMBOL_GPL(nfs_atomic_open); 2009 2010 static int 2011 nfs4_do_lookup_revalidate(struct inode *dir, struct dentry *dentry, 2012 unsigned int flags) 2013 { 2014 struct inode *inode; 2015 2016 if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY)) 2017 goto full_reval; 2018 if (d_mountpoint(dentry)) 2019 goto full_reval; 2020 2021 inode = d_inode(dentry); 2022 2023 /* We can't create new files in nfs_open_revalidate(), so we 2024 * optimize away revalidation of negative dentries. 2025 */ 2026 if (inode == NULL) 2027 goto full_reval; 2028 2029 if (nfs_verifier_is_delegated(dentry)) 2030 return nfs_lookup_revalidate_delegated(dir, dentry, inode); 2031 2032 /* NFS only supports OPEN on regular files */ 2033 if (!S_ISREG(inode->i_mode)) 2034 goto full_reval; 2035 2036 /* We cannot do exclusive creation on a positive dentry */ 2037 if (flags & (LOOKUP_EXCL | LOOKUP_REVAL)) 2038 goto reval_dentry; 2039 2040 /* Check if the directory changed */ 2041 if (!nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU)) 2042 goto reval_dentry; 2043 2044 /* Let f_op->open() actually open (and revalidate) the file */ 2045 return 1; 2046 reval_dentry: 2047 if (flags & LOOKUP_RCU) 2048 return -ECHILD; 2049 return nfs_lookup_revalidate_dentry(dir, dentry, inode); 2050 2051 full_reval: 2052 return nfs_do_lookup_revalidate(dir, dentry, flags); 2053 } 2054 2055 static int nfs4_lookup_revalidate(struct dentry *dentry, unsigned int flags) 2056 { 2057 return __nfs_lookup_revalidate(dentry, flags, 2058 nfs4_do_lookup_revalidate); 2059 } 2060 2061 #endif /* CONFIG_NFSV4 */ 2062 2063 struct dentry * 2064 nfs_add_or_obtain(struct dentry *dentry, struct nfs_fh *fhandle, 2065 struct nfs_fattr *fattr) 2066 { 2067 struct dentry *parent = dget_parent(dentry); 2068 struct inode *dir = d_inode(parent); 2069 struct inode *inode; 2070 struct dentry *d; 2071 int error; 2072 2073 d_drop(dentry); 2074 2075 if (fhandle->size == 0) { 2076 error = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr); 2077 if (error) 2078 goto out_error; 2079 } 2080 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 2081 if (!(fattr->valid & NFS_ATTR_FATTR)) { 2082 struct nfs_server *server = NFS_SB(dentry->d_sb); 2083 error = server->nfs_client->rpc_ops->getattr(server, fhandle, 2084 fattr, NULL); 2085 if (error < 0) 2086 goto out_error; 2087 } 2088 inode = nfs_fhget(dentry->d_sb, fhandle, fattr); 2089 d = d_splice_alias(inode, dentry); 2090 out: 2091 dput(parent); 2092 return d; 2093 out_error: 2094 d = ERR_PTR(error); 2095 goto out; 2096 } 2097 EXPORT_SYMBOL_GPL(nfs_add_or_obtain); 2098 2099 /* 2100 * Code common to create, mkdir, and mknod. 2101 */ 2102 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle, 2103 struct nfs_fattr *fattr) 2104 { 2105 struct dentry *d; 2106 2107 d = nfs_add_or_obtain(dentry, fhandle, fattr); 2108 if (IS_ERR(d)) 2109 return PTR_ERR(d); 2110 2111 /* Callers don't care */ 2112 dput(d); 2113 return 0; 2114 } 2115 EXPORT_SYMBOL_GPL(nfs_instantiate); 2116 2117 /* 2118 * Following a failed create operation, we drop the dentry rather 2119 * than retain a negative dentry. This avoids a problem in the event 2120 * that the operation succeeded on the server, but an error in the 2121 * reply path made it appear to have failed. 2122 */ 2123 int nfs_create(struct user_namespace *mnt_userns, struct inode *dir, 2124 struct dentry *dentry, umode_t mode, bool excl) 2125 { 2126 struct iattr attr; 2127 int open_flags = excl ? O_CREAT | O_EXCL : O_CREAT; 2128 int error; 2129 2130 dfprintk(VFS, "NFS: create(%s/%lu), %pd\n", 2131 dir->i_sb->s_id, dir->i_ino, dentry); 2132 2133 attr.ia_mode = mode; 2134 attr.ia_valid = ATTR_MODE; 2135 2136 trace_nfs_create_enter(dir, dentry, open_flags); 2137 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags); 2138 trace_nfs_create_exit(dir, dentry, open_flags, error); 2139 if (error != 0) 2140 goto out_err; 2141 return 0; 2142 out_err: 2143 d_drop(dentry); 2144 return error; 2145 } 2146 EXPORT_SYMBOL_GPL(nfs_create); 2147 2148 /* 2149 * See comments for nfs_proc_create regarding failed operations. 2150 */ 2151 int 2152 nfs_mknod(struct user_namespace *mnt_userns, struct inode *dir, 2153 struct dentry *dentry, umode_t mode, dev_t rdev) 2154 { 2155 struct iattr attr; 2156 int status; 2157 2158 dfprintk(VFS, "NFS: mknod(%s/%lu), %pd\n", 2159 dir->i_sb->s_id, dir->i_ino, dentry); 2160 2161 attr.ia_mode = mode; 2162 attr.ia_valid = ATTR_MODE; 2163 2164 trace_nfs_mknod_enter(dir, dentry); 2165 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev); 2166 trace_nfs_mknod_exit(dir, dentry, status); 2167 if (status != 0) 2168 goto out_err; 2169 return 0; 2170 out_err: 2171 d_drop(dentry); 2172 return status; 2173 } 2174 EXPORT_SYMBOL_GPL(nfs_mknod); 2175 2176 /* 2177 * See comments for nfs_proc_create regarding failed operations. 2178 */ 2179 int nfs_mkdir(struct user_namespace *mnt_userns, struct inode *dir, 2180 struct dentry *dentry, umode_t mode) 2181 { 2182 struct iattr attr; 2183 int error; 2184 2185 dfprintk(VFS, "NFS: mkdir(%s/%lu), %pd\n", 2186 dir->i_sb->s_id, dir->i_ino, dentry); 2187 2188 attr.ia_valid = ATTR_MODE; 2189 attr.ia_mode = mode | S_IFDIR; 2190 2191 trace_nfs_mkdir_enter(dir, dentry); 2192 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr); 2193 trace_nfs_mkdir_exit(dir, dentry, error); 2194 if (error != 0) 2195 goto out_err; 2196 return 0; 2197 out_err: 2198 d_drop(dentry); 2199 return error; 2200 } 2201 EXPORT_SYMBOL_GPL(nfs_mkdir); 2202 2203 static void nfs_dentry_handle_enoent(struct dentry *dentry) 2204 { 2205 if (simple_positive(dentry)) 2206 d_delete(dentry); 2207 } 2208 2209 static void nfs_dentry_remove_handle_error(struct inode *dir, 2210 struct dentry *dentry, int error) 2211 { 2212 switch (error) { 2213 case -ENOENT: 2214 d_delete(dentry); 2215 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 2216 break; 2217 case 0: 2218 nfs_d_prune_case_insensitive_aliases(d_inode(dentry)); 2219 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 2220 } 2221 } 2222 2223 int nfs_rmdir(struct inode *dir, struct dentry *dentry) 2224 { 2225 int error; 2226 2227 dfprintk(VFS, "NFS: rmdir(%s/%lu), %pd\n", 2228 dir->i_sb->s_id, dir->i_ino, dentry); 2229 2230 trace_nfs_rmdir_enter(dir, dentry); 2231 if (d_really_is_positive(dentry)) { 2232 down_write(&NFS_I(d_inode(dentry))->rmdir_sem); 2233 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name); 2234 /* Ensure the VFS deletes this inode */ 2235 switch (error) { 2236 case 0: 2237 clear_nlink(d_inode(dentry)); 2238 break; 2239 case -ENOENT: 2240 nfs_dentry_handle_enoent(dentry); 2241 } 2242 up_write(&NFS_I(d_inode(dentry))->rmdir_sem); 2243 } else 2244 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name); 2245 nfs_dentry_remove_handle_error(dir, dentry, error); 2246 trace_nfs_rmdir_exit(dir, dentry, error); 2247 2248 return error; 2249 } 2250 EXPORT_SYMBOL_GPL(nfs_rmdir); 2251 2252 /* 2253 * Remove a file after making sure there are no pending writes, 2254 * and after checking that the file has only one user. 2255 * 2256 * We invalidate the attribute cache and free the inode prior to the operation 2257 * to avoid possible races if the server reuses the inode. 2258 */ 2259 static int nfs_safe_remove(struct dentry *dentry) 2260 { 2261 struct inode *dir = d_inode(dentry->d_parent); 2262 struct inode *inode = d_inode(dentry); 2263 int error = -EBUSY; 2264 2265 dfprintk(VFS, "NFS: safe_remove(%pd2)\n", dentry); 2266 2267 /* If the dentry was sillyrenamed, we simply call d_delete() */ 2268 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { 2269 error = 0; 2270 goto out; 2271 } 2272 2273 trace_nfs_remove_enter(dir, dentry); 2274 if (inode != NULL) { 2275 error = NFS_PROTO(dir)->remove(dir, dentry); 2276 if (error == 0) 2277 nfs_drop_nlink(inode); 2278 } else 2279 error = NFS_PROTO(dir)->remove(dir, dentry); 2280 if (error == -ENOENT) 2281 nfs_dentry_handle_enoent(dentry); 2282 trace_nfs_remove_exit(dir, dentry, error); 2283 out: 2284 return error; 2285 } 2286 2287 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode 2288 * belongs to an active ".nfs..." file and we return -EBUSY. 2289 * 2290 * If sillyrename() returns 0, we do nothing, otherwise we unlink. 2291 */ 2292 int nfs_unlink(struct inode *dir, struct dentry *dentry) 2293 { 2294 int error; 2295 int need_rehash = 0; 2296 2297 dfprintk(VFS, "NFS: unlink(%s/%lu, %pd)\n", dir->i_sb->s_id, 2298 dir->i_ino, dentry); 2299 2300 trace_nfs_unlink_enter(dir, dentry); 2301 spin_lock(&dentry->d_lock); 2302 if (d_count(dentry) > 1) { 2303 spin_unlock(&dentry->d_lock); 2304 /* Start asynchronous writeout of the inode */ 2305 write_inode_now(d_inode(dentry), 0); 2306 error = nfs_sillyrename(dir, dentry); 2307 goto out; 2308 } 2309 if (!d_unhashed(dentry)) { 2310 __d_drop(dentry); 2311 need_rehash = 1; 2312 } 2313 spin_unlock(&dentry->d_lock); 2314 error = nfs_safe_remove(dentry); 2315 nfs_dentry_remove_handle_error(dir, dentry, error); 2316 if (need_rehash) 2317 d_rehash(dentry); 2318 out: 2319 trace_nfs_unlink_exit(dir, dentry, error); 2320 return error; 2321 } 2322 EXPORT_SYMBOL_GPL(nfs_unlink); 2323 2324 /* 2325 * To create a symbolic link, most file systems instantiate a new inode, 2326 * add a page to it containing the path, then write it out to the disk 2327 * using prepare_write/commit_write. 2328 * 2329 * Unfortunately the NFS client can't create the in-core inode first 2330 * because it needs a file handle to create an in-core inode (see 2331 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the 2332 * symlink request has completed on the server. 2333 * 2334 * So instead we allocate a raw page, copy the symname into it, then do 2335 * the SYMLINK request with the page as the buffer. If it succeeds, we 2336 * now have a new file handle and can instantiate an in-core NFS inode 2337 * and move the raw page into its mapping. 2338 */ 2339 int nfs_symlink(struct user_namespace *mnt_userns, struct inode *dir, 2340 struct dentry *dentry, const char *symname) 2341 { 2342 struct page *page; 2343 char *kaddr; 2344 struct iattr attr; 2345 unsigned int pathlen = strlen(symname); 2346 int error; 2347 2348 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s)\n", dir->i_sb->s_id, 2349 dir->i_ino, dentry, symname); 2350 2351 if (pathlen > PAGE_SIZE) 2352 return -ENAMETOOLONG; 2353 2354 attr.ia_mode = S_IFLNK | S_IRWXUGO; 2355 attr.ia_valid = ATTR_MODE; 2356 2357 page = alloc_page(GFP_USER); 2358 if (!page) 2359 return -ENOMEM; 2360 2361 kaddr = page_address(page); 2362 memcpy(kaddr, symname, pathlen); 2363 if (pathlen < PAGE_SIZE) 2364 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen); 2365 2366 trace_nfs_symlink_enter(dir, dentry); 2367 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr); 2368 trace_nfs_symlink_exit(dir, dentry, error); 2369 if (error != 0) { 2370 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s) error %d\n", 2371 dir->i_sb->s_id, dir->i_ino, 2372 dentry, symname, error); 2373 d_drop(dentry); 2374 __free_page(page); 2375 return error; 2376 } 2377 2378 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 2379 2380 /* 2381 * No big deal if we can't add this page to the page cache here. 2382 * READLINK will get the missing page from the server if needed. 2383 */ 2384 if (!add_to_page_cache_lru(page, d_inode(dentry)->i_mapping, 0, 2385 GFP_KERNEL)) { 2386 SetPageUptodate(page); 2387 unlock_page(page); 2388 /* 2389 * add_to_page_cache_lru() grabs an extra page refcount. 2390 * Drop it here to avoid leaking this page later. 2391 */ 2392 put_page(page); 2393 } else 2394 __free_page(page); 2395 2396 return 0; 2397 } 2398 EXPORT_SYMBOL_GPL(nfs_symlink); 2399 2400 int 2401 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) 2402 { 2403 struct inode *inode = d_inode(old_dentry); 2404 int error; 2405 2406 dfprintk(VFS, "NFS: link(%pd2 -> %pd2)\n", 2407 old_dentry, dentry); 2408 2409 trace_nfs_link_enter(inode, dir, dentry); 2410 d_drop(dentry); 2411 if (S_ISREG(inode->i_mode)) 2412 nfs_sync_inode(inode); 2413 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name); 2414 if (error == 0) { 2415 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 2416 ihold(inode); 2417 d_add(dentry, inode); 2418 } 2419 trace_nfs_link_exit(inode, dir, dentry, error); 2420 return error; 2421 } 2422 EXPORT_SYMBOL_GPL(nfs_link); 2423 2424 /* 2425 * RENAME 2426 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a 2427 * different file handle for the same inode after a rename (e.g. when 2428 * moving to a different directory). A fail-safe method to do so would 2429 * be to look up old_dir/old_name, create a link to new_dir/new_name and 2430 * rename the old file using the sillyrename stuff. This way, the original 2431 * file in old_dir will go away when the last process iput()s the inode. 2432 * 2433 * FIXED. 2434 * 2435 * It actually works quite well. One needs to have the possibility for 2436 * at least one ".nfs..." file in each directory the file ever gets 2437 * moved or linked to which happens automagically with the new 2438 * implementation that only depends on the dcache stuff instead of 2439 * using the inode layer 2440 * 2441 * Unfortunately, things are a little more complicated than indicated 2442 * above. For a cross-directory move, we want to make sure we can get 2443 * rid of the old inode after the operation. This means there must be 2444 * no pending writes (if it's a file), and the use count must be 1. 2445 * If these conditions are met, we can drop the dentries before doing 2446 * the rename. 2447 */ 2448 int nfs_rename(struct user_namespace *mnt_userns, struct inode *old_dir, 2449 struct dentry *old_dentry, struct inode *new_dir, 2450 struct dentry *new_dentry, unsigned int flags) 2451 { 2452 struct inode *old_inode = d_inode(old_dentry); 2453 struct inode *new_inode = d_inode(new_dentry); 2454 struct dentry *dentry = NULL, *rehash = NULL; 2455 struct rpc_task *task; 2456 int error = -EBUSY; 2457 2458 if (flags) 2459 return -EINVAL; 2460 2461 dfprintk(VFS, "NFS: rename(%pd2 -> %pd2, ct=%d)\n", 2462 old_dentry, new_dentry, 2463 d_count(new_dentry)); 2464 2465 trace_nfs_rename_enter(old_dir, old_dentry, new_dir, new_dentry); 2466 /* 2467 * For non-directories, check whether the target is busy and if so, 2468 * make a copy of the dentry and then do a silly-rename. If the 2469 * silly-rename succeeds, the copied dentry is hashed and becomes 2470 * the new target. 2471 */ 2472 if (new_inode && !S_ISDIR(new_inode->i_mode)) { 2473 /* 2474 * To prevent any new references to the target during the 2475 * rename, we unhash the dentry in advance. 2476 */ 2477 if (!d_unhashed(new_dentry)) { 2478 d_drop(new_dentry); 2479 rehash = new_dentry; 2480 } 2481 2482 if (d_count(new_dentry) > 2) { 2483 int err; 2484 2485 /* copy the target dentry's name */ 2486 dentry = d_alloc(new_dentry->d_parent, 2487 &new_dentry->d_name); 2488 if (!dentry) 2489 goto out; 2490 2491 /* silly-rename the existing target ... */ 2492 err = nfs_sillyrename(new_dir, new_dentry); 2493 if (err) 2494 goto out; 2495 2496 new_dentry = dentry; 2497 rehash = NULL; 2498 new_inode = NULL; 2499 } 2500 } 2501 2502 if (S_ISREG(old_inode->i_mode)) 2503 nfs_sync_inode(old_inode); 2504 task = nfs_async_rename(old_dir, new_dir, old_dentry, new_dentry, NULL); 2505 if (IS_ERR(task)) { 2506 error = PTR_ERR(task); 2507 goto out; 2508 } 2509 2510 error = rpc_wait_for_completion_task(task); 2511 if (error != 0) { 2512 ((struct nfs_renamedata *)task->tk_calldata)->cancelled = 1; 2513 /* Paired with the atomic_dec_and_test() barrier in rpc_do_put_task() */ 2514 smp_wmb(); 2515 } else 2516 error = task->tk_status; 2517 rpc_put_task(task); 2518 /* Ensure the inode attributes are revalidated */ 2519 if (error == 0) { 2520 spin_lock(&old_inode->i_lock); 2521 NFS_I(old_inode)->attr_gencount = nfs_inc_attr_generation_counter(); 2522 nfs_set_cache_invalid(old_inode, NFS_INO_INVALID_CHANGE | 2523 NFS_INO_INVALID_CTIME | 2524 NFS_INO_REVAL_FORCED); 2525 spin_unlock(&old_inode->i_lock); 2526 } 2527 out: 2528 if (rehash) 2529 d_rehash(rehash); 2530 trace_nfs_rename_exit(old_dir, old_dentry, 2531 new_dir, new_dentry, error); 2532 if (!error) { 2533 if (new_inode != NULL) 2534 nfs_drop_nlink(new_inode); 2535 /* 2536 * The d_move() should be here instead of in an async RPC completion 2537 * handler because we need the proper locks to move the dentry. If 2538 * we're interrupted by a signal, the async RPC completion handler 2539 * should mark the directories for revalidation. 2540 */ 2541 d_move(old_dentry, new_dentry); 2542 nfs_set_verifier(old_dentry, 2543 nfs_save_change_attribute(new_dir)); 2544 } else if (error == -ENOENT) 2545 nfs_dentry_handle_enoent(old_dentry); 2546 2547 /* new dentry created? */ 2548 if (dentry) 2549 dput(dentry); 2550 return error; 2551 } 2552 EXPORT_SYMBOL_GPL(nfs_rename); 2553 2554 static DEFINE_SPINLOCK(nfs_access_lru_lock); 2555 static LIST_HEAD(nfs_access_lru_list); 2556 static atomic_long_t nfs_access_nr_entries; 2557 2558 static unsigned long nfs_access_max_cachesize = 4*1024*1024; 2559 module_param(nfs_access_max_cachesize, ulong, 0644); 2560 MODULE_PARM_DESC(nfs_access_max_cachesize, "NFS access maximum total cache length"); 2561 2562 static void nfs_access_free_entry(struct nfs_access_entry *entry) 2563 { 2564 put_group_info(entry->group_info); 2565 kfree_rcu(entry, rcu_head); 2566 smp_mb__before_atomic(); 2567 atomic_long_dec(&nfs_access_nr_entries); 2568 smp_mb__after_atomic(); 2569 } 2570 2571 static void nfs_access_free_list(struct list_head *head) 2572 { 2573 struct nfs_access_entry *cache; 2574 2575 while (!list_empty(head)) { 2576 cache = list_entry(head->next, struct nfs_access_entry, lru); 2577 list_del(&cache->lru); 2578 nfs_access_free_entry(cache); 2579 } 2580 } 2581 2582 static unsigned long 2583 nfs_do_access_cache_scan(unsigned int nr_to_scan) 2584 { 2585 LIST_HEAD(head); 2586 struct nfs_inode *nfsi, *next; 2587 struct nfs_access_entry *cache; 2588 long freed = 0; 2589 2590 spin_lock(&nfs_access_lru_lock); 2591 list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) { 2592 struct inode *inode; 2593 2594 if (nr_to_scan-- == 0) 2595 break; 2596 inode = &nfsi->vfs_inode; 2597 spin_lock(&inode->i_lock); 2598 if (list_empty(&nfsi->access_cache_entry_lru)) 2599 goto remove_lru_entry; 2600 cache = list_entry(nfsi->access_cache_entry_lru.next, 2601 struct nfs_access_entry, lru); 2602 list_move(&cache->lru, &head); 2603 rb_erase(&cache->rb_node, &nfsi->access_cache); 2604 freed++; 2605 if (!list_empty(&nfsi->access_cache_entry_lru)) 2606 list_move_tail(&nfsi->access_cache_inode_lru, 2607 &nfs_access_lru_list); 2608 else { 2609 remove_lru_entry: 2610 list_del_init(&nfsi->access_cache_inode_lru); 2611 smp_mb__before_atomic(); 2612 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags); 2613 smp_mb__after_atomic(); 2614 } 2615 spin_unlock(&inode->i_lock); 2616 } 2617 spin_unlock(&nfs_access_lru_lock); 2618 nfs_access_free_list(&head); 2619 return freed; 2620 } 2621 2622 unsigned long 2623 nfs_access_cache_scan(struct shrinker *shrink, struct shrink_control *sc) 2624 { 2625 int nr_to_scan = sc->nr_to_scan; 2626 gfp_t gfp_mask = sc->gfp_mask; 2627 2628 if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL) 2629 return SHRINK_STOP; 2630 return nfs_do_access_cache_scan(nr_to_scan); 2631 } 2632 2633 2634 unsigned long 2635 nfs_access_cache_count(struct shrinker *shrink, struct shrink_control *sc) 2636 { 2637 return vfs_pressure_ratio(atomic_long_read(&nfs_access_nr_entries)); 2638 } 2639 2640 static void 2641 nfs_access_cache_enforce_limit(void) 2642 { 2643 long nr_entries = atomic_long_read(&nfs_access_nr_entries); 2644 unsigned long diff; 2645 unsigned int nr_to_scan; 2646 2647 if (nr_entries < 0 || nr_entries <= nfs_access_max_cachesize) 2648 return; 2649 nr_to_scan = 100; 2650 diff = nr_entries - nfs_access_max_cachesize; 2651 if (diff < nr_to_scan) 2652 nr_to_scan = diff; 2653 nfs_do_access_cache_scan(nr_to_scan); 2654 } 2655 2656 static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head) 2657 { 2658 struct rb_root *root_node = &nfsi->access_cache; 2659 struct rb_node *n; 2660 struct nfs_access_entry *entry; 2661 2662 /* Unhook entries from the cache */ 2663 while ((n = rb_first(root_node)) != NULL) { 2664 entry = rb_entry(n, struct nfs_access_entry, rb_node); 2665 rb_erase(n, root_node); 2666 list_move(&entry->lru, head); 2667 } 2668 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS; 2669 } 2670 2671 void nfs_access_zap_cache(struct inode *inode) 2672 { 2673 LIST_HEAD(head); 2674 2675 if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0) 2676 return; 2677 /* Remove from global LRU init */ 2678 spin_lock(&nfs_access_lru_lock); 2679 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) 2680 list_del_init(&NFS_I(inode)->access_cache_inode_lru); 2681 2682 spin_lock(&inode->i_lock); 2683 __nfs_access_zap_cache(NFS_I(inode), &head); 2684 spin_unlock(&inode->i_lock); 2685 spin_unlock(&nfs_access_lru_lock); 2686 nfs_access_free_list(&head); 2687 } 2688 EXPORT_SYMBOL_GPL(nfs_access_zap_cache); 2689 2690 static int access_cmp(const struct cred *a, const struct nfs_access_entry *b) 2691 { 2692 struct group_info *ga, *gb; 2693 int g; 2694 2695 if (uid_lt(a->fsuid, b->fsuid)) 2696 return -1; 2697 if (uid_gt(a->fsuid, b->fsuid)) 2698 return 1; 2699 2700 if (gid_lt(a->fsgid, b->fsgid)) 2701 return -1; 2702 if (gid_gt(a->fsgid, b->fsgid)) 2703 return 1; 2704 2705 ga = a->group_info; 2706 gb = b->group_info; 2707 if (ga == gb) 2708 return 0; 2709 if (ga == NULL) 2710 return -1; 2711 if (gb == NULL) 2712 return 1; 2713 if (ga->ngroups < gb->ngroups) 2714 return -1; 2715 if (ga->ngroups > gb->ngroups) 2716 return 1; 2717 2718 for (g = 0; g < ga->ngroups; g++) { 2719 if (gid_lt(ga->gid[g], gb->gid[g])) 2720 return -1; 2721 if (gid_gt(ga->gid[g], gb->gid[g])) 2722 return 1; 2723 } 2724 return 0; 2725 } 2726 2727 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, const struct cred *cred) 2728 { 2729 struct rb_node *n = NFS_I(inode)->access_cache.rb_node; 2730 2731 while (n != NULL) { 2732 struct nfs_access_entry *entry = 2733 rb_entry(n, struct nfs_access_entry, rb_node); 2734 int cmp = access_cmp(cred, entry); 2735 2736 if (cmp < 0) 2737 n = n->rb_left; 2738 else if (cmp > 0) 2739 n = n->rb_right; 2740 else 2741 return entry; 2742 } 2743 return NULL; 2744 } 2745 2746 static int nfs_access_get_cached_locked(struct inode *inode, const struct cred *cred, u32 *mask, bool may_block) 2747 { 2748 struct nfs_inode *nfsi = NFS_I(inode); 2749 struct nfs_access_entry *cache; 2750 bool retry = true; 2751 int err; 2752 2753 spin_lock(&inode->i_lock); 2754 for(;;) { 2755 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS) 2756 goto out_zap; 2757 cache = nfs_access_search_rbtree(inode, cred); 2758 err = -ENOENT; 2759 if (cache == NULL) 2760 goto out; 2761 /* Found an entry, is our attribute cache valid? */ 2762 if (!nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS)) 2763 break; 2764 if (!retry) 2765 break; 2766 err = -ECHILD; 2767 if (!may_block) 2768 goto out; 2769 spin_unlock(&inode->i_lock); 2770 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode); 2771 if (err) 2772 return err; 2773 spin_lock(&inode->i_lock); 2774 retry = false; 2775 } 2776 *mask = cache->mask; 2777 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru); 2778 err = 0; 2779 out: 2780 spin_unlock(&inode->i_lock); 2781 return err; 2782 out_zap: 2783 spin_unlock(&inode->i_lock); 2784 nfs_access_zap_cache(inode); 2785 return -ENOENT; 2786 } 2787 2788 static int nfs_access_get_cached_rcu(struct inode *inode, const struct cred *cred, u32 *mask) 2789 { 2790 /* Only check the most recently returned cache entry, 2791 * but do it without locking. 2792 */ 2793 struct nfs_inode *nfsi = NFS_I(inode); 2794 struct nfs_access_entry *cache; 2795 int err = -ECHILD; 2796 struct list_head *lh; 2797 2798 rcu_read_lock(); 2799 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS) 2800 goto out; 2801 lh = rcu_dereference(list_tail_rcu(&nfsi->access_cache_entry_lru)); 2802 cache = list_entry(lh, struct nfs_access_entry, lru); 2803 if (lh == &nfsi->access_cache_entry_lru || 2804 access_cmp(cred, cache) != 0) 2805 cache = NULL; 2806 if (cache == NULL) 2807 goto out; 2808 if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS)) 2809 goto out; 2810 *mask = cache->mask; 2811 err = 0; 2812 out: 2813 rcu_read_unlock(); 2814 return err; 2815 } 2816 2817 int nfs_access_get_cached(struct inode *inode, const struct cred *cred, 2818 u32 *mask, bool may_block) 2819 { 2820 int status; 2821 2822 status = nfs_access_get_cached_rcu(inode, cred, mask); 2823 if (status != 0) 2824 status = nfs_access_get_cached_locked(inode, cred, mask, 2825 may_block); 2826 2827 return status; 2828 } 2829 EXPORT_SYMBOL_GPL(nfs_access_get_cached); 2830 2831 static void nfs_access_add_rbtree(struct inode *inode, 2832 struct nfs_access_entry *set, 2833 const struct cred *cred) 2834 { 2835 struct nfs_inode *nfsi = NFS_I(inode); 2836 struct rb_root *root_node = &nfsi->access_cache; 2837 struct rb_node **p = &root_node->rb_node; 2838 struct rb_node *parent = NULL; 2839 struct nfs_access_entry *entry; 2840 int cmp; 2841 2842 spin_lock(&inode->i_lock); 2843 while (*p != NULL) { 2844 parent = *p; 2845 entry = rb_entry(parent, struct nfs_access_entry, rb_node); 2846 cmp = access_cmp(cred, entry); 2847 2848 if (cmp < 0) 2849 p = &parent->rb_left; 2850 else if (cmp > 0) 2851 p = &parent->rb_right; 2852 else 2853 goto found; 2854 } 2855 rb_link_node(&set->rb_node, parent, p); 2856 rb_insert_color(&set->rb_node, root_node); 2857 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); 2858 spin_unlock(&inode->i_lock); 2859 return; 2860 found: 2861 rb_replace_node(parent, &set->rb_node, root_node); 2862 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); 2863 list_del(&entry->lru); 2864 spin_unlock(&inode->i_lock); 2865 nfs_access_free_entry(entry); 2866 } 2867 2868 void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set, 2869 const struct cred *cred) 2870 { 2871 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL); 2872 if (cache == NULL) 2873 return; 2874 RB_CLEAR_NODE(&cache->rb_node); 2875 cache->fsuid = cred->fsuid; 2876 cache->fsgid = cred->fsgid; 2877 cache->group_info = get_group_info(cred->group_info); 2878 cache->mask = set->mask; 2879 2880 /* The above field assignments must be visible 2881 * before this item appears on the lru. We cannot easily 2882 * use rcu_assign_pointer, so just force the memory barrier. 2883 */ 2884 smp_wmb(); 2885 nfs_access_add_rbtree(inode, cache, cred); 2886 2887 /* Update accounting */ 2888 smp_mb__before_atomic(); 2889 atomic_long_inc(&nfs_access_nr_entries); 2890 smp_mb__after_atomic(); 2891 2892 /* Add inode to global LRU list */ 2893 if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) { 2894 spin_lock(&nfs_access_lru_lock); 2895 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) 2896 list_add_tail(&NFS_I(inode)->access_cache_inode_lru, 2897 &nfs_access_lru_list); 2898 spin_unlock(&nfs_access_lru_lock); 2899 } 2900 nfs_access_cache_enforce_limit(); 2901 } 2902 EXPORT_SYMBOL_GPL(nfs_access_add_cache); 2903 2904 #define NFS_MAY_READ (NFS_ACCESS_READ) 2905 #define NFS_MAY_WRITE (NFS_ACCESS_MODIFY | \ 2906 NFS_ACCESS_EXTEND | \ 2907 NFS_ACCESS_DELETE) 2908 #define NFS_FILE_MAY_WRITE (NFS_ACCESS_MODIFY | \ 2909 NFS_ACCESS_EXTEND) 2910 #define NFS_DIR_MAY_WRITE NFS_MAY_WRITE 2911 #define NFS_MAY_LOOKUP (NFS_ACCESS_LOOKUP) 2912 #define NFS_MAY_EXECUTE (NFS_ACCESS_EXECUTE) 2913 static int 2914 nfs_access_calc_mask(u32 access_result, umode_t umode) 2915 { 2916 int mask = 0; 2917 2918 if (access_result & NFS_MAY_READ) 2919 mask |= MAY_READ; 2920 if (S_ISDIR(umode)) { 2921 if ((access_result & NFS_DIR_MAY_WRITE) == NFS_DIR_MAY_WRITE) 2922 mask |= MAY_WRITE; 2923 if ((access_result & NFS_MAY_LOOKUP) == NFS_MAY_LOOKUP) 2924 mask |= MAY_EXEC; 2925 } else if (S_ISREG(umode)) { 2926 if ((access_result & NFS_FILE_MAY_WRITE) == NFS_FILE_MAY_WRITE) 2927 mask |= MAY_WRITE; 2928 if ((access_result & NFS_MAY_EXECUTE) == NFS_MAY_EXECUTE) 2929 mask |= MAY_EXEC; 2930 } else if (access_result & NFS_MAY_WRITE) 2931 mask |= MAY_WRITE; 2932 return mask; 2933 } 2934 2935 void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result) 2936 { 2937 entry->mask = access_result; 2938 } 2939 EXPORT_SYMBOL_GPL(nfs_access_set_mask); 2940 2941 static int nfs_do_access(struct inode *inode, const struct cred *cred, int mask) 2942 { 2943 struct nfs_access_entry cache; 2944 bool may_block = (mask & MAY_NOT_BLOCK) == 0; 2945 int cache_mask = -1; 2946 int status; 2947 2948 trace_nfs_access_enter(inode); 2949 2950 status = nfs_access_get_cached(inode, cred, &cache.mask, may_block); 2951 if (status == 0) 2952 goto out_cached; 2953 2954 status = -ECHILD; 2955 if (!may_block) 2956 goto out; 2957 2958 /* 2959 * Determine which access bits we want to ask for... 2960 */ 2961 cache.mask = NFS_ACCESS_READ | NFS_ACCESS_MODIFY | NFS_ACCESS_EXTEND; 2962 if (nfs_server_capable(inode, NFS_CAP_XATTR)) { 2963 cache.mask |= NFS_ACCESS_XAREAD | NFS_ACCESS_XAWRITE | 2964 NFS_ACCESS_XALIST; 2965 } 2966 if (S_ISDIR(inode->i_mode)) 2967 cache.mask |= NFS_ACCESS_DELETE | NFS_ACCESS_LOOKUP; 2968 else 2969 cache.mask |= NFS_ACCESS_EXECUTE; 2970 status = NFS_PROTO(inode)->access(inode, &cache, cred); 2971 if (status != 0) { 2972 if (status == -ESTALE) { 2973 if (!S_ISDIR(inode->i_mode)) 2974 nfs_set_inode_stale(inode); 2975 else 2976 nfs_zap_caches(inode); 2977 } 2978 goto out; 2979 } 2980 nfs_access_add_cache(inode, &cache, cred); 2981 out_cached: 2982 cache_mask = nfs_access_calc_mask(cache.mask, inode->i_mode); 2983 if ((mask & ~cache_mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) != 0) 2984 status = -EACCES; 2985 out: 2986 trace_nfs_access_exit(inode, mask, cache_mask, status); 2987 return status; 2988 } 2989 2990 static int nfs_open_permission_mask(int openflags) 2991 { 2992 int mask = 0; 2993 2994 if (openflags & __FMODE_EXEC) { 2995 /* ONLY check exec rights */ 2996 mask = MAY_EXEC; 2997 } else { 2998 if ((openflags & O_ACCMODE) != O_WRONLY) 2999 mask |= MAY_READ; 3000 if ((openflags & O_ACCMODE) != O_RDONLY) 3001 mask |= MAY_WRITE; 3002 } 3003 3004 return mask; 3005 } 3006 3007 int nfs_may_open(struct inode *inode, const struct cred *cred, int openflags) 3008 { 3009 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags)); 3010 } 3011 EXPORT_SYMBOL_GPL(nfs_may_open); 3012 3013 static int nfs_execute_ok(struct inode *inode, int mask) 3014 { 3015 struct nfs_server *server = NFS_SERVER(inode); 3016 int ret = 0; 3017 3018 if (S_ISDIR(inode->i_mode)) 3019 return 0; 3020 if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_MODE)) { 3021 if (mask & MAY_NOT_BLOCK) 3022 return -ECHILD; 3023 ret = __nfs_revalidate_inode(server, inode); 3024 } 3025 if (ret == 0 && !execute_ok(inode)) 3026 ret = -EACCES; 3027 return ret; 3028 } 3029 3030 int nfs_permission(struct user_namespace *mnt_userns, 3031 struct inode *inode, 3032 int mask) 3033 { 3034 const struct cred *cred = current_cred(); 3035 int res = 0; 3036 3037 nfs_inc_stats(inode, NFSIOS_VFSACCESS); 3038 3039 if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) 3040 goto out; 3041 /* Is this sys_access() ? */ 3042 if (mask & (MAY_ACCESS | MAY_CHDIR)) 3043 goto force_lookup; 3044 3045 switch (inode->i_mode & S_IFMT) { 3046 case S_IFLNK: 3047 goto out; 3048 case S_IFREG: 3049 if ((mask & MAY_OPEN) && 3050 nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN)) 3051 return 0; 3052 break; 3053 case S_IFDIR: 3054 /* 3055 * Optimize away all write operations, since the server 3056 * will check permissions when we perform the op. 3057 */ 3058 if ((mask & MAY_WRITE) && !(mask & MAY_READ)) 3059 goto out; 3060 } 3061 3062 force_lookup: 3063 if (!NFS_PROTO(inode)->access) 3064 goto out_notsup; 3065 3066 res = nfs_do_access(inode, cred, mask); 3067 out: 3068 if (!res && (mask & MAY_EXEC)) 3069 res = nfs_execute_ok(inode, mask); 3070 3071 dfprintk(VFS, "NFS: permission(%s/%lu), mask=0x%x, res=%d\n", 3072 inode->i_sb->s_id, inode->i_ino, mask, res); 3073 return res; 3074 out_notsup: 3075 if (mask & MAY_NOT_BLOCK) 3076 return -ECHILD; 3077 3078 res = nfs_revalidate_inode(inode, NFS_INO_INVALID_MODE | 3079 NFS_INO_INVALID_OTHER); 3080 if (res == 0) 3081 res = generic_permission(&init_user_ns, inode, mask); 3082 goto out; 3083 } 3084 EXPORT_SYMBOL_GPL(nfs_permission); 3085