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 entry->fattr->label); 687 goto out; 688 } else { 689 d_invalidate(dentry); 690 dput(dentry); 691 dentry = NULL; 692 goto again; 693 } 694 } 695 if (!entry->fh->size) { 696 d_lookup_done(dentry); 697 goto out; 698 } 699 700 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr, entry->fattr->label); 701 alias = d_splice_alias(inode, dentry); 702 d_lookup_done(dentry); 703 if (alias) { 704 if (IS_ERR(alias)) 705 goto out; 706 dput(dentry); 707 dentry = alias; 708 } 709 nfs_set_verifier(dentry, dir_verifier); 710 out: 711 dput(dentry); 712 } 713 714 /* Perform conversion from xdr to cache array */ 715 static int nfs_readdir_page_filler(struct nfs_readdir_descriptor *desc, 716 struct nfs_entry *entry, 717 struct page **xdr_pages, 718 unsigned int buflen, 719 struct page **arrays, 720 size_t narrays) 721 { 722 struct address_space *mapping = desc->file->f_mapping; 723 struct xdr_stream stream; 724 struct xdr_buf buf; 725 struct page *scratch, *new, *page = *arrays; 726 int status; 727 728 scratch = alloc_page(GFP_KERNEL); 729 if (scratch == NULL) 730 return -ENOMEM; 731 732 xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen); 733 xdr_set_scratch_page(&stream, scratch); 734 735 do { 736 if (entry->fattr->label) 737 entry->fattr->label->len = NFS4_MAXLABELLEN; 738 739 status = xdr_decode(desc, entry, &stream); 740 if (status != 0) 741 break; 742 743 if (desc->plus) 744 nfs_prime_dcache(file_dentry(desc->file), entry, 745 desc->dir_verifier); 746 747 status = nfs_readdir_add_to_array(entry, page); 748 if (status != -ENOSPC) 749 continue; 750 751 if (page->mapping != mapping) { 752 if (!--narrays) 753 break; 754 new = nfs_readdir_page_array_alloc(entry->prev_cookie, 755 GFP_KERNEL); 756 if (!new) 757 break; 758 arrays++; 759 *arrays = page = new; 760 } else { 761 new = nfs_readdir_page_get_next(mapping, 762 page->index + 1, 763 entry->prev_cookie); 764 if (!new) 765 break; 766 if (page != *arrays) 767 nfs_readdir_page_unlock_and_put(page); 768 page = new; 769 } 770 status = nfs_readdir_add_to_array(entry, page); 771 } while (!status && !entry->eof); 772 773 switch (status) { 774 case -EBADCOOKIE: 775 if (entry->eof) { 776 nfs_readdir_page_set_eof(page); 777 status = 0; 778 } 779 break; 780 case -ENOSPC: 781 case -EAGAIN: 782 status = 0; 783 break; 784 } 785 786 if (page != *arrays) 787 nfs_readdir_page_unlock_and_put(page); 788 789 put_page(scratch); 790 return status; 791 } 792 793 static void nfs_readdir_free_pages(struct page **pages, size_t npages) 794 { 795 while (npages--) 796 put_page(pages[npages]); 797 kfree(pages); 798 } 799 800 /* 801 * nfs_readdir_alloc_pages() will allocate pages that must be freed with a call 802 * to nfs_readdir_free_pages() 803 */ 804 static struct page **nfs_readdir_alloc_pages(size_t npages) 805 { 806 struct page **pages; 807 size_t i; 808 809 pages = kmalloc_array(npages, sizeof(*pages), GFP_KERNEL); 810 if (!pages) 811 return NULL; 812 for (i = 0; i < npages; i++) { 813 struct page *page = alloc_page(GFP_KERNEL); 814 if (page == NULL) 815 goto out_freepages; 816 pages[i] = page; 817 } 818 return pages; 819 820 out_freepages: 821 nfs_readdir_free_pages(pages, i); 822 return NULL; 823 } 824 825 static int nfs_readdir_xdr_to_array(struct nfs_readdir_descriptor *desc, 826 __be32 *verf_arg, __be32 *verf_res, 827 struct page **arrays, size_t narrays) 828 { 829 struct page **pages; 830 struct page *page = *arrays; 831 struct nfs_entry *entry; 832 size_t array_size; 833 struct inode *inode = file_inode(desc->file); 834 size_t dtsize = NFS_SERVER(inode)->dtsize; 835 int status = -ENOMEM; 836 837 entry = kzalloc(sizeof(*entry), GFP_KERNEL); 838 if (!entry) 839 return -ENOMEM; 840 entry->cookie = nfs_readdir_page_last_cookie(page); 841 entry->fh = nfs_alloc_fhandle(); 842 entry->fattr = nfs_alloc_fattr_with_label(NFS_SERVER(inode)); 843 entry->server = NFS_SERVER(inode); 844 if (entry->fh == NULL || entry->fattr == NULL) 845 goto out; 846 847 array_size = (dtsize + PAGE_SIZE - 1) >> PAGE_SHIFT; 848 pages = nfs_readdir_alloc_pages(array_size); 849 if (!pages) 850 goto out; 851 852 do { 853 unsigned int pglen; 854 status = nfs_readdir_xdr_filler(desc, verf_arg, entry->cookie, 855 pages, dtsize, 856 verf_res); 857 if (status < 0) 858 break; 859 860 pglen = status; 861 if (pglen == 0) { 862 nfs_readdir_page_set_eof(page); 863 break; 864 } 865 866 verf_arg = verf_res; 867 868 status = nfs_readdir_page_filler(desc, entry, pages, pglen, 869 arrays, narrays); 870 } while (!status && nfs_readdir_page_needs_filling(page)); 871 872 nfs_readdir_free_pages(pages, array_size); 873 out: 874 nfs_free_fattr(entry->fattr); 875 nfs_free_fhandle(entry->fh); 876 kfree(entry); 877 return status; 878 } 879 880 static void nfs_readdir_page_put(struct nfs_readdir_descriptor *desc) 881 { 882 put_page(desc->page); 883 desc->page = NULL; 884 } 885 886 static void 887 nfs_readdir_page_unlock_and_put_cached(struct nfs_readdir_descriptor *desc) 888 { 889 unlock_page(desc->page); 890 nfs_readdir_page_put(desc); 891 } 892 893 static struct page * 894 nfs_readdir_page_get_cached(struct nfs_readdir_descriptor *desc) 895 { 896 return nfs_readdir_page_get_locked(desc->file->f_mapping, 897 desc->page_index, 898 desc->last_cookie); 899 } 900 901 /* 902 * Returns 0 if desc->dir_cookie was found on page desc->page_index 903 * and locks the page to prevent removal from the page cache. 904 */ 905 static int find_and_lock_cache_page(struct nfs_readdir_descriptor *desc) 906 { 907 struct inode *inode = file_inode(desc->file); 908 struct nfs_inode *nfsi = NFS_I(inode); 909 __be32 verf[NFS_DIR_VERIFIER_SIZE]; 910 int res; 911 912 desc->page = nfs_readdir_page_get_cached(desc); 913 if (!desc->page) 914 return -ENOMEM; 915 if (nfs_readdir_page_needs_filling(desc->page)) { 916 res = nfs_readdir_xdr_to_array(desc, nfsi->cookieverf, verf, 917 &desc->page, 1); 918 if (res < 0) { 919 nfs_readdir_page_unlock_and_put_cached(desc); 920 if (res == -EBADCOOKIE || res == -ENOTSYNC) { 921 invalidate_inode_pages2(desc->file->f_mapping); 922 desc->page_index = 0; 923 return -EAGAIN; 924 } 925 return res; 926 } 927 /* 928 * Set the cookie verifier if the page cache was empty 929 */ 930 if (desc->page_index == 0) 931 memcpy(nfsi->cookieverf, verf, 932 sizeof(nfsi->cookieverf)); 933 } 934 res = nfs_readdir_search_array(desc); 935 if (res == 0) 936 return 0; 937 nfs_readdir_page_unlock_and_put_cached(desc); 938 return res; 939 } 940 941 static bool nfs_readdir_dont_search_cache(struct nfs_readdir_descriptor *desc) 942 { 943 struct address_space *mapping = desc->file->f_mapping; 944 struct inode *dir = file_inode(desc->file); 945 unsigned int dtsize = NFS_SERVER(dir)->dtsize; 946 loff_t size = i_size_read(dir); 947 948 /* 949 * Default to uncached readdir if the page cache is empty, and 950 * we're looking for a non-zero cookie in a large directory. 951 */ 952 return desc->dir_cookie != 0 && mapping->nrpages == 0 && size > dtsize; 953 } 954 955 /* Search for desc->dir_cookie from the beginning of the page cache */ 956 static int readdir_search_pagecache(struct nfs_readdir_descriptor *desc) 957 { 958 int res; 959 960 if (nfs_readdir_dont_search_cache(desc)) 961 return -EBADCOOKIE; 962 963 do { 964 if (desc->page_index == 0) { 965 desc->current_index = 0; 966 desc->prev_index = 0; 967 desc->last_cookie = 0; 968 } 969 res = find_and_lock_cache_page(desc); 970 } while (res == -EAGAIN); 971 return res; 972 } 973 974 /* 975 * Once we've found the start of the dirent within a page: fill 'er up... 976 */ 977 static void nfs_do_filldir(struct nfs_readdir_descriptor *desc, 978 const __be32 *verf) 979 { 980 struct file *file = desc->file; 981 struct nfs_cache_array *array; 982 unsigned int i = 0; 983 984 array = kmap(desc->page); 985 for (i = desc->cache_entry_index; i < array->size; i++) { 986 struct nfs_cache_array_entry *ent; 987 988 ent = &array->array[i]; 989 if (!dir_emit(desc->ctx, ent->name, ent->name_len, 990 nfs_compat_user_ino64(ent->ino), ent->d_type)) { 991 desc->eof = true; 992 break; 993 } 994 memcpy(desc->verf, verf, sizeof(desc->verf)); 995 if (i < (array->size-1)) 996 desc->dir_cookie = array->array[i+1].cookie; 997 else 998 desc->dir_cookie = array->last_cookie; 999 if (nfs_readdir_use_cookie(file)) 1000 desc->ctx->pos = desc->dir_cookie; 1001 else 1002 desc->ctx->pos++; 1003 if (desc->duped != 0) 1004 desc->duped = 1; 1005 } 1006 if (array->page_is_eof) 1007 desc->eof = true; 1008 1009 kunmap(desc->page); 1010 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %llu\n", 1011 (unsigned long long)desc->dir_cookie); 1012 } 1013 1014 /* 1015 * If we cannot find a cookie in our cache, we suspect that this is 1016 * because it points to a deleted file, so we ask the server to return 1017 * whatever it thinks is the next entry. We then feed this to filldir. 1018 * If all goes well, we should then be able to find our way round the 1019 * cache on the next call to readdir_search_pagecache(); 1020 * 1021 * NOTE: we cannot add the anonymous page to the pagecache because 1022 * the data it contains might not be page aligned. Besides, 1023 * we should already have a complete representation of the 1024 * directory in the page cache by the time we get here. 1025 */ 1026 static int uncached_readdir(struct nfs_readdir_descriptor *desc) 1027 { 1028 struct page **arrays; 1029 size_t i, sz = 512; 1030 __be32 verf[NFS_DIR_VERIFIER_SIZE]; 1031 int status = -ENOMEM; 1032 1033 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %llu\n", 1034 (unsigned long long)desc->dir_cookie); 1035 1036 arrays = kcalloc(sz, sizeof(*arrays), GFP_KERNEL); 1037 if (!arrays) 1038 goto out; 1039 arrays[0] = nfs_readdir_page_array_alloc(desc->dir_cookie, GFP_KERNEL); 1040 if (!arrays[0]) 1041 goto out; 1042 1043 desc->page_index = 0; 1044 desc->last_cookie = desc->dir_cookie; 1045 desc->duped = 0; 1046 1047 status = nfs_readdir_xdr_to_array(desc, desc->verf, verf, arrays, sz); 1048 1049 for (i = 0; !desc->eof && i < sz && arrays[i]; i++) { 1050 desc->page = arrays[i]; 1051 nfs_do_filldir(desc, verf); 1052 } 1053 desc->page = NULL; 1054 1055 1056 for (i = 0; i < sz && arrays[i]; i++) 1057 nfs_readdir_page_array_free(arrays[i]); 1058 out: 1059 kfree(arrays); 1060 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __func__, status); 1061 return status; 1062 } 1063 1064 /* The file offset position represents the dirent entry number. A 1065 last cookie cache takes care of the common case of reading the 1066 whole directory. 1067 */ 1068 static int nfs_readdir(struct file *file, struct dir_context *ctx) 1069 { 1070 struct dentry *dentry = file_dentry(file); 1071 struct inode *inode = d_inode(dentry); 1072 struct nfs_inode *nfsi = NFS_I(inode); 1073 struct nfs_open_dir_context *dir_ctx = file->private_data; 1074 struct nfs_readdir_descriptor *desc; 1075 pgoff_t page_index; 1076 int res; 1077 1078 dfprintk(FILE, "NFS: readdir(%pD2) starting at cookie %llu\n", 1079 file, (long long)ctx->pos); 1080 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS); 1081 1082 /* 1083 * ctx->pos points to the dirent entry number. 1084 * *desc->dir_cookie has the cookie for the next entry. We have 1085 * to either find the entry with the appropriate number or 1086 * revalidate the cookie. 1087 */ 1088 if (ctx->pos == 0 || nfs_attribute_cache_expired(inode)) { 1089 res = nfs_revalidate_mapping(inode, file->f_mapping); 1090 if (res < 0) 1091 goto out; 1092 } 1093 1094 res = -ENOMEM; 1095 desc = kzalloc(sizeof(*desc), GFP_KERNEL); 1096 if (!desc) 1097 goto out; 1098 desc->file = file; 1099 desc->ctx = ctx; 1100 desc->plus = nfs_use_readdirplus(inode, ctx); 1101 1102 spin_lock(&file->f_lock); 1103 desc->dir_cookie = dir_ctx->dir_cookie; 1104 desc->dup_cookie = dir_ctx->dup_cookie; 1105 desc->duped = dir_ctx->duped; 1106 page_index = dir_ctx->page_index; 1107 desc->attr_gencount = dir_ctx->attr_gencount; 1108 memcpy(desc->verf, dir_ctx->verf, sizeof(desc->verf)); 1109 spin_unlock(&file->f_lock); 1110 1111 if (test_and_clear_bit(NFS_INO_FORCE_READDIR, &nfsi->flags) && 1112 list_is_singular(&nfsi->open_files)) 1113 invalidate_mapping_pages(inode->i_mapping, page_index + 1, -1); 1114 1115 do { 1116 res = readdir_search_pagecache(desc); 1117 1118 if (res == -EBADCOOKIE) { 1119 res = 0; 1120 /* This means either end of directory */ 1121 if (desc->dir_cookie && !desc->eof) { 1122 /* Or that the server has 'lost' a cookie */ 1123 res = uncached_readdir(desc); 1124 if (res == 0) 1125 continue; 1126 if (res == -EBADCOOKIE || res == -ENOTSYNC) 1127 res = 0; 1128 } 1129 break; 1130 } 1131 if (res == -ETOOSMALL && desc->plus) { 1132 clear_bit(NFS_INO_ADVISE_RDPLUS, &nfsi->flags); 1133 nfs_zap_caches(inode); 1134 desc->page_index = 0; 1135 desc->plus = false; 1136 desc->eof = false; 1137 continue; 1138 } 1139 if (res < 0) 1140 break; 1141 1142 nfs_do_filldir(desc, nfsi->cookieverf); 1143 nfs_readdir_page_unlock_and_put_cached(desc); 1144 } while (!desc->eof); 1145 1146 spin_lock(&file->f_lock); 1147 dir_ctx->dir_cookie = desc->dir_cookie; 1148 dir_ctx->dup_cookie = desc->dup_cookie; 1149 dir_ctx->duped = desc->duped; 1150 dir_ctx->attr_gencount = desc->attr_gencount; 1151 dir_ctx->page_index = desc->page_index; 1152 memcpy(dir_ctx->verf, desc->verf, sizeof(dir_ctx->verf)); 1153 spin_unlock(&file->f_lock); 1154 1155 kfree(desc); 1156 1157 out: 1158 dfprintk(FILE, "NFS: readdir(%pD2) returns %d\n", file, res); 1159 return res; 1160 } 1161 1162 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int whence) 1163 { 1164 struct nfs_open_dir_context *dir_ctx = filp->private_data; 1165 1166 dfprintk(FILE, "NFS: llseek dir(%pD2, %lld, %d)\n", 1167 filp, offset, whence); 1168 1169 switch (whence) { 1170 default: 1171 return -EINVAL; 1172 case SEEK_SET: 1173 if (offset < 0) 1174 return -EINVAL; 1175 spin_lock(&filp->f_lock); 1176 break; 1177 case SEEK_CUR: 1178 if (offset == 0) 1179 return filp->f_pos; 1180 spin_lock(&filp->f_lock); 1181 offset += filp->f_pos; 1182 if (offset < 0) { 1183 spin_unlock(&filp->f_lock); 1184 return -EINVAL; 1185 } 1186 } 1187 if (offset != filp->f_pos) { 1188 filp->f_pos = offset; 1189 if (nfs_readdir_use_cookie(filp)) 1190 dir_ctx->dir_cookie = offset; 1191 else 1192 dir_ctx->dir_cookie = 0; 1193 if (offset == 0) 1194 memset(dir_ctx->verf, 0, sizeof(dir_ctx->verf)); 1195 dir_ctx->duped = 0; 1196 } 1197 spin_unlock(&filp->f_lock); 1198 return offset; 1199 } 1200 1201 /* 1202 * All directory operations under NFS are synchronous, so fsync() 1203 * is a dummy operation. 1204 */ 1205 static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end, 1206 int datasync) 1207 { 1208 dfprintk(FILE, "NFS: fsync dir(%pD2) datasync %d\n", filp, datasync); 1209 1210 nfs_inc_stats(file_inode(filp), NFSIOS_VFSFSYNC); 1211 return 0; 1212 } 1213 1214 /** 1215 * nfs_force_lookup_revalidate - Mark the directory as having changed 1216 * @dir: pointer to directory inode 1217 * 1218 * This forces the revalidation code in nfs_lookup_revalidate() to do a 1219 * full lookup on all child dentries of 'dir' whenever a change occurs 1220 * on the server that might have invalidated our dcache. 1221 * 1222 * Note that we reserve bit '0' as a tag to let us know when a dentry 1223 * was revalidated while holding a delegation on its inode. 1224 * 1225 * The caller should be holding dir->i_lock 1226 */ 1227 void nfs_force_lookup_revalidate(struct inode *dir) 1228 { 1229 NFS_I(dir)->cache_change_attribute += 2; 1230 } 1231 EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate); 1232 1233 /** 1234 * nfs_verify_change_attribute - Detects NFS remote directory changes 1235 * @dir: pointer to parent directory inode 1236 * @verf: previously saved change attribute 1237 * 1238 * Return "false" if the verifiers doesn't match the change attribute. 1239 * This would usually indicate that the directory contents have changed on 1240 * the server, and that any dentries need revalidating. 1241 */ 1242 static bool nfs_verify_change_attribute(struct inode *dir, unsigned long verf) 1243 { 1244 return (verf & ~1UL) == nfs_save_change_attribute(dir); 1245 } 1246 1247 static void nfs_set_verifier_delegated(unsigned long *verf) 1248 { 1249 *verf |= 1UL; 1250 } 1251 1252 #if IS_ENABLED(CONFIG_NFS_V4) 1253 static void nfs_unset_verifier_delegated(unsigned long *verf) 1254 { 1255 *verf &= ~1UL; 1256 } 1257 #endif /* IS_ENABLED(CONFIG_NFS_V4) */ 1258 1259 static bool nfs_test_verifier_delegated(unsigned long verf) 1260 { 1261 return verf & 1; 1262 } 1263 1264 static bool nfs_verifier_is_delegated(struct dentry *dentry) 1265 { 1266 return nfs_test_verifier_delegated(dentry->d_time); 1267 } 1268 1269 static void nfs_set_verifier_locked(struct dentry *dentry, unsigned long verf) 1270 { 1271 struct inode *inode = d_inode(dentry); 1272 struct inode *dir = d_inode(dentry->d_parent); 1273 1274 if (!nfs_verify_change_attribute(dir, verf)) 1275 return; 1276 if (inode && NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) 1277 nfs_set_verifier_delegated(&verf); 1278 dentry->d_time = verf; 1279 } 1280 1281 /** 1282 * nfs_set_verifier - save a parent directory verifier in the dentry 1283 * @dentry: pointer to dentry 1284 * @verf: verifier to save 1285 * 1286 * Saves the parent directory verifier in @dentry. If the inode has 1287 * a delegation, we also tag the dentry as having been revalidated 1288 * while holding a delegation so that we know we don't have to 1289 * look it up again after a directory change. 1290 */ 1291 void nfs_set_verifier(struct dentry *dentry, unsigned long verf) 1292 { 1293 1294 spin_lock(&dentry->d_lock); 1295 nfs_set_verifier_locked(dentry, verf); 1296 spin_unlock(&dentry->d_lock); 1297 } 1298 EXPORT_SYMBOL_GPL(nfs_set_verifier); 1299 1300 #if IS_ENABLED(CONFIG_NFS_V4) 1301 /** 1302 * nfs_clear_verifier_delegated - clear the dir verifier delegation tag 1303 * @inode: pointer to inode 1304 * 1305 * Iterates through the dentries in the inode alias list and clears 1306 * the tag used to indicate that the dentry has been revalidated 1307 * while holding a delegation. 1308 * This function is intended for use when the delegation is being 1309 * returned or revoked. 1310 */ 1311 void nfs_clear_verifier_delegated(struct inode *inode) 1312 { 1313 struct dentry *alias; 1314 1315 if (!inode) 1316 return; 1317 spin_lock(&inode->i_lock); 1318 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) { 1319 spin_lock(&alias->d_lock); 1320 nfs_unset_verifier_delegated(&alias->d_time); 1321 spin_unlock(&alias->d_lock); 1322 } 1323 spin_unlock(&inode->i_lock); 1324 } 1325 EXPORT_SYMBOL_GPL(nfs_clear_verifier_delegated); 1326 #endif /* IS_ENABLED(CONFIG_NFS_V4) */ 1327 1328 /* 1329 * A check for whether or not the parent directory has changed. 1330 * In the case it has, we assume that the dentries are untrustworthy 1331 * and may need to be looked up again. 1332 * If rcu_walk prevents us from performing a full check, return 0. 1333 */ 1334 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry, 1335 int rcu_walk) 1336 { 1337 if (IS_ROOT(dentry)) 1338 return 1; 1339 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE) 1340 return 0; 1341 if (!nfs_verify_change_attribute(dir, dentry->d_time)) 1342 return 0; 1343 /* Revalidate nfsi->cache_change_attribute before we declare a match */ 1344 if (nfs_mapping_need_revalidate_inode(dir)) { 1345 if (rcu_walk) 1346 return 0; 1347 if (__nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0) 1348 return 0; 1349 } 1350 if (!nfs_verify_change_attribute(dir, dentry->d_time)) 1351 return 0; 1352 return 1; 1353 } 1354 1355 /* 1356 * Use intent information to check whether or not we're going to do 1357 * an O_EXCL create using this path component. 1358 */ 1359 static int nfs_is_exclusive_create(struct inode *dir, unsigned int flags) 1360 { 1361 if (NFS_PROTO(dir)->version == 2) 1362 return 0; 1363 return flags & LOOKUP_EXCL; 1364 } 1365 1366 /* 1367 * Inode and filehandle revalidation for lookups. 1368 * 1369 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL, 1370 * or if the intent information indicates that we're about to open this 1371 * particular file and the "nocto" mount flag is not set. 1372 * 1373 */ 1374 static 1375 int nfs_lookup_verify_inode(struct inode *inode, unsigned int flags) 1376 { 1377 struct nfs_server *server = NFS_SERVER(inode); 1378 int ret; 1379 1380 if (IS_AUTOMOUNT(inode)) 1381 return 0; 1382 1383 if (flags & LOOKUP_OPEN) { 1384 switch (inode->i_mode & S_IFMT) { 1385 case S_IFREG: 1386 /* A NFSv4 OPEN will revalidate later */ 1387 if (server->caps & NFS_CAP_ATOMIC_OPEN) 1388 goto out; 1389 fallthrough; 1390 case S_IFDIR: 1391 if (server->flags & NFS_MOUNT_NOCTO) 1392 break; 1393 /* NFS close-to-open cache consistency validation */ 1394 goto out_force; 1395 } 1396 } 1397 1398 /* VFS wants an on-the-wire revalidation */ 1399 if (flags & LOOKUP_REVAL) 1400 goto out_force; 1401 out: 1402 return (inode->i_nlink == 0) ? -ESTALE : 0; 1403 out_force: 1404 if (flags & LOOKUP_RCU) 1405 return -ECHILD; 1406 ret = __nfs_revalidate_inode(server, inode); 1407 if (ret != 0) 1408 return ret; 1409 goto out; 1410 } 1411 1412 static void nfs_mark_dir_for_revalidate(struct inode *inode) 1413 { 1414 spin_lock(&inode->i_lock); 1415 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE); 1416 spin_unlock(&inode->i_lock); 1417 } 1418 1419 /* 1420 * We judge how long we want to trust negative 1421 * dentries by looking at the parent inode mtime. 1422 * 1423 * If parent mtime has changed, we revalidate, else we wait for a 1424 * period corresponding to the parent's attribute cache timeout value. 1425 * 1426 * If LOOKUP_RCU prevents us from performing a full check, return 1 1427 * suggesting a reval is needed. 1428 * 1429 * Note that when creating a new file, or looking up a rename target, 1430 * then it shouldn't be necessary to revalidate a negative dentry. 1431 */ 1432 static inline 1433 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry, 1434 unsigned int flags) 1435 { 1436 if (flags & (LOOKUP_CREATE | LOOKUP_RENAME_TARGET)) 1437 return 0; 1438 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG) 1439 return 1; 1440 return !nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU); 1441 } 1442 1443 static int 1444 nfs_lookup_revalidate_done(struct inode *dir, struct dentry *dentry, 1445 struct inode *inode, int error) 1446 { 1447 switch (error) { 1448 case 1: 1449 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is valid\n", 1450 __func__, dentry); 1451 return 1; 1452 case 0: 1453 /* 1454 * We can't d_drop the root of a disconnected tree: 1455 * its d_hash is on the s_anon list and d_drop() would hide 1456 * it from shrink_dcache_for_unmount(), leading to busy 1457 * inodes on unmount and further oopses. 1458 */ 1459 if (inode && IS_ROOT(dentry)) 1460 return 1; 1461 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is invalid\n", 1462 __func__, dentry); 1463 return 0; 1464 } 1465 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) lookup returned error %d\n", 1466 __func__, dentry, error); 1467 return error; 1468 } 1469 1470 static int 1471 nfs_lookup_revalidate_negative(struct inode *dir, struct dentry *dentry, 1472 unsigned int flags) 1473 { 1474 int ret = 1; 1475 if (nfs_neg_need_reval(dir, dentry, flags)) { 1476 if (flags & LOOKUP_RCU) 1477 return -ECHILD; 1478 ret = 0; 1479 } 1480 return nfs_lookup_revalidate_done(dir, dentry, NULL, ret); 1481 } 1482 1483 static int 1484 nfs_lookup_revalidate_delegated(struct inode *dir, struct dentry *dentry, 1485 struct inode *inode) 1486 { 1487 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1488 return nfs_lookup_revalidate_done(dir, dentry, inode, 1); 1489 } 1490 1491 static int 1492 nfs_lookup_revalidate_dentry(struct inode *dir, struct dentry *dentry, 1493 struct inode *inode) 1494 { 1495 struct nfs_fh *fhandle; 1496 struct nfs_fattr *fattr; 1497 unsigned long dir_verifier; 1498 int ret; 1499 1500 ret = -ENOMEM; 1501 fhandle = nfs_alloc_fhandle(); 1502 fattr = nfs_alloc_fattr_with_label(NFS_SERVER(inode)); 1503 if (fhandle == NULL || fattr == NULL) 1504 goto out; 1505 1506 dir_verifier = nfs_save_change_attribute(dir); 1507 ret = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr); 1508 if (ret < 0) { 1509 switch (ret) { 1510 case -ESTALE: 1511 case -ENOENT: 1512 ret = 0; 1513 break; 1514 case -ETIMEDOUT: 1515 if (NFS_SERVER(inode)->flags & NFS_MOUNT_SOFTREVAL) 1516 ret = 1; 1517 } 1518 goto out; 1519 } 1520 ret = 0; 1521 if (nfs_compare_fh(NFS_FH(inode), fhandle)) 1522 goto out; 1523 if (nfs_refresh_inode(inode, fattr) < 0) 1524 goto out; 1525 1526 nfs_setsecurity(inode, fattr, fattr->label); 1527 nfs_set_verifier(dentry, dir_verifier); 1528 1529 /* set a readdirplus hint that we had a cache miss */ 1530 nfs_force_use_readdirplus(dir); 1531 ret = 1; 1532 out: 1533 nfs_free_fattr(fattr); 1534 nfs_free_fhandle(fhandle); 1535 1536 /* 1537 * If the lookup failed despite the dentry change attribute being 1538 * a match, then we should revalidate the directory cache. 1539 */ 1540 if (!ret && nfs_verify_change_attribute(dir, dentry->d_time)) 1541 nfs_mark_dir_for_revalidate(dir); 1542 return nfs_lookup_revalidate_done(dir, dentry, inode, ret); 1543 } 1544 1545 /* 1546 * This is called every time the dcache has a lookup hit, 1547 * and we should check whether we can really trust that 1548 * lookup. 1549 * 1550 * NOTE! The hit can be a negative hit too, don't assume 1551 * we have an inode! 1552 * 1553 * If the parent directory is seen to have changed, we throw out the 1554 * cached dentry and do a new lookup. 1555 */ 1556 static int 1557 nfs_do_lookup_revalidate(struct inode *dir, struct dentry *dentry, 1558 unsigned int flags) 1559 { 1560 struct inode *inode; 1561 int error; 1562 1563 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE); 1564 inode = d_inode(dentry); 1565 1566 if (!inode) 1567 return nfs_lookup_revalidate_negative(dir, dentry, flags); 1568 1569 if (is_bad_inode(inode)) { 1570 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n", 1571 __func__, dentry); 1572 goto out_bad; 1573 } 1574 1575 if (nfs_verifier_is_delegated(dentry)) 1576 return nfs_lookup_revalidate_delegated(dir, dentry, inode); 1577 1578 /* Force a full look up iff the parent directory has changed */ 1579 if (!(flags & (LOOKUP_EXCL | LOOKUP_REVAL)) && 1580 nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU)) { 1581 error = nfs_lookup_verify_inode(inode, flags); 1582 if (error) { 1583 if (error == -ESTALE) 1584 nfs_mark_dir_for_revalidate(dir); 1585 goto out_bad; 1586 } 1587 nfs_advise_use_readdirplus(dir); 1588 goto out_valid; 1589 } 1590 1591 if (flags & LOOKUP_RCU) 1592 return -ECHILD; 1593 1594 if (NFS_STALE(inode)) 1595 goto out_bad; 1596 1597 trace_nfs_lookup_revalidate_enter(dir, dentry, flags); 1598 error = nfs_lookup_revalidate_dentry(dir, dentry, inode); 1599 trace_nfs_lookup_revalidate_exit(dir, dentry, flags, error); 1600 return error; 1601 out_valid: 1602 return nfs_lookup_revalidate_done(dir, dentry, inode, 1); 1603 out_bad: 1604 if (flags & LOOKUP_RCU) 1605 return -ECHILD; 1606 return nfs_lookup_revalidate_done(dir, dentry, inode, 0); 1607 } 1608 1609 static int 1610 __nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags, 1611 int (*reval)(struct inode *, struct dentry *, unsigned int)) 1612 { 1613 struct dentry *parent; 1614 struct inode *dir; 1615 int ret; 1616 1617 if (flags & LOOKUP_RCU) { 1618 parent = READ_ONCE(dentry->d_parent); 1619 dir = d_inode_rcu(parent); 1620 if (!dir) 1621 return -ECHILD; 1622 ret = reval(dir, dentry, flags); 1623 if (parent != READ_ONCE(dentry->d_parent)) 1624 return -ECHILD; 1625 } else { 1626 parent = dget_parent(dentry); 1627 ret = reval(d_inode(parent), dentry, flags); 1628 dput(parent); 1629 } 1630 return ret; 1631 } 1632 1633 static int nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags) 1634 { 1635 return __nfs_lookup_revalidate(dentry, flags, nfs_do_lookup_revalidate); 1636 } 1637 1638 /* 1639 * A weaker form of d_revalidate for revalidating just the d_inode(dentry) 1640 * when we don't really care about the dentry name. This is called when a 1641 * pathwalk ends on a dentry that was not found via a normal lookup in the 1642 * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals). 1643 * 1644 * In this situation, we just want to verify that the inode itself is OK 1645 * since the dentry might have changed on the server. 1646 */ 1647 static int nfs_weak_revalidate(struct dentry *dentry, unsigned int flags) 1648 { 1649 struct inode *inode = d_inode(dentry); 1650 int error = 0; 1651 1652 /* 1653 * I believe we can only get a negative dentry here in the case of a 1654 * procfs-style symlink. Just assume it's correct for now, but we may 1655 * eventually need to do something more here. 1656 */ 1657 if (!inode) { 1658 dfprintk(LOOKUPCACHE, "%s: %pd2 has negative inode\n", 1659 __func__, dentry); 1660 return 1; 1661 } 1662 1663 if (is_bad_inode(inode)) { 1664 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n", 1665 __func__, dentry); 1666 return 0; 1667 } 1668 1669 error = nfs_lookup_verify_inode(inode, flags); 1670 dfprintk(LOOKUPCACHE, "NFS: %s: inode %lu is %s\n", 1671 __func__, inode->i_ino, error ? "invalid" : "valid"); 1672 return !error; 1673 } 1674 1675 /* 1676 * This is called from dput() when d_count is going to 0. 1677 */ 1678 static int nfs_dentry_delete(const struct dentry *dentry) 1679 { 1680 dfprintk(VFS, "NFS: dentry_delete(%pd2, %x)\n", 1681 dentry, dentry->d_flags); 1682 1683 /* Unhash any dentry with a stale inode */ 1684 if (d_really_is_positive(dentry) && NFS_STALE(d_inode(dentry))) 1685 return 1; 1686 1687 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { 1688 /* Unhash it, so that ->d_iput() would be called */ 1689 return 1; 1690 } 1691 if (!(dentry->d_sb->s_flags & SB_ACTIVE)) { 1692 /* Unhash it, so that ancestors of killed async unlink 1693 * files will be cleaned up during umount */ 1694 return 1; 1695 } 1696 return 0; 1697 1698 } 1699 1700 /* Ensure that we revalidate inode->i_nlink */ 1701 static void nfs_drop_nlink(struct inode *inode) 1702 { 1703 spin_lock(&inode->i_lock); 1704 /* drop the inode if we're reasonably sure this is the last link */ 1705 if (inode->i_nlink > 0) 1706 drop_nlink(inode); 1707 NFS_I(inode)->attr_gencount = nfs_inc_attr_generation_counter(); 1708 nfs_set_cache_invalid( 1709 inode, NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_CTIME | 1710 NFS_INO_INVALID_NLINK); 1711 spin_unlock(&inode->i_lock); 1712 } 1713 1714 /* 1715 * Called when the dentry loses inode. 1716 * We use it to clean up silly-renamed files. 1717 */ 1718 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode) 1719 { 1720 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { 1721 nfs_complete_unlink(dentry, inode); 1722 nfs_drop_nlink(inode); 1723 } 1724 iput(inode); 1725 } 1726 1727 static void nfs_d_release(struct dentry *dentry) 1728 { 1729 /* free cached devname value, if it survived that far */ 1730 if (unlikely(dentry->d_fsdata)) { 1731 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) 1732 WARN_ON(1); 1733 else 1734 kfree(dentry->d_fsdata); 1735 } 1736 } 1737 1738 const struct dentry_operations nfs_dentry_operations = { 1739 .d_revalidate = nfs_lookup_revalidate, 1740 .d_weak_revalidate = nfs_weak_revalidate, 1741 .d_delete = nfs_dentry_delete, 1742 .d_iput = nfs_dentry_iput, 1743 .d_automount = nfs_d_automount, 1744 .d_release = nfs_d_release, 1745 }; 1746 EXPORT_SYMBOL_GPL(nfs_dentry_operations); 1747 1748 struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags) 1749 { 1750 struct dentry *res; 1751 struct inode *inode = NULL; 1752 struct nfs_fh *fhandle = NULL; 1753 struct nfs_fattr *fattr = NULL; 1754 unsigned long dir_verifier; 1755 int error; 1756 1757 dfprintk(VFS, "NFS: lookup(%pd2)\n", dentry); 1758 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP); 1759 1760 if (unlikely(dentry->d_name.len > NFS_SERVER(dir)->namelen)) 1761 return ERR_PTR(-ENAMETOOLONG); 1762 1763 /* 1764 * If we're doing an exclusive create, optimize away the lookup 1765 * but don't hash the dentry. 1766 */ 1767 if (nfs_is_exclusive_create(dir, flags) || flags & LOOKUP_RENAME_TARGET) 1768 return NULL; 1769 1770 res = ERR_PTR(-ENOMEM); 1771 fhandle = nfs_alloc_fhandle(); 1772 fattr = nfs_alloc_fattr_with_label(NFS_SERVER(dir)); 1773 if (fhandle == NULL || fattr == NULL) 1774 goto out; 1775 1776 dir_verifier = nfs_save_change_attribute(dir); 1777 trace_nfs_lookup_enter(dir, dentry, flags); 1778 error = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr); 1779 if (error == -ENOENT) 1780 goto no_entry; 1781 if (error < 0) { 1782 res = ERR_PTR(error); 1783 goto out; 1784 } 1785 inode = nfs_fhget(dentry->d_sb, fhandle, fattr, fattr->label); 1786 res = ERR_CAST(inode); 1787 if (IS_ERR(res)) 1788 goto out; 1789 1790 /* Notify readdir to use READDIRPLUS */ 1791 nfs_force_use_readdirplus(dir); 1792 1793 no_entry: 1794 res = d_splice_alias(inode, dentry); 1795 if (res != NULL) { 1796 if (IS_ERR(res)) 1797 goto out; 1798 dentry = res; 1799 } 1800 nfs_set_verifier(dentry, dir_verifier); 1801 out: 1802 trace_nfs_lookup_exit(dir, dentry, flags, PTR_ERR_OR_ZERO(res)); 1803 nfs_free_fattr(fattr); 1804 nfs_free_fhandle(fhandle); 1805 return res; 1806 } 1807 EXPORT_SYMBOL_GPL(nfs_lookup); 1808 1809 #if IS_ENABLED(CONFIG_NFS_V4) 1810 static int nfs4_lookup_revalidate(struct dentry *, unsigned int); 1811 1812 const struct dentry_operations nfs4_dentry_operations = { 1813 .d_revalidate = nfs4_lookup_revalidate, 1814 .d_weak_revalidate = nfs_weak_revalidate, 1815 .d_delete = nfs_dentry_delete, 1816 .d_iput = nfs_dentry_iput, 1817 .d_automount = nfs_d_automount, 1818 .d_release = nfs_d_release, 1819 }; 1820 EXPORT_SYMBOL_GPL(nfs4_dentry_operations); 1821 1822 static fmode_t flags_to_mode(int flags) 1823 { 1824 fmode_t res = (__force fmode_t)flags & FMODE_EXEC; 1825 if ((flags & O_ACCMODE) != O_WRONLY) 1826 res |= FMODE_READ; 1827 if ((flags & O_ACCMODE) != O_RDONLY) 1828 res |= FMODE_WRITE; 1829 return res; 1830 } 1831 1832 static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags, struct file *filp) 1833 { 1834 return alloc_nfs_open_context(dentry, flags_to_mode(open_flags), filp); 1835 } 1836 1837 static int do_open(struct inode *inode, struct file *filp) 1838 { 1839 nfs_fscache_open_file(inode, filp); 1840 return 0; 1841 } 1842 1843 static int nfs_finish_open(struct nfs_open_context *ctx, 1844 struct dentry *dentry, 1845 struct file *file, unsigned open_flags) 1846 { 1847 int err; 1848 1849 err = finish_open(file, dentry, do_open); 1850 if (err) 1851 goto out; 1852 if (S_ISREG(file->f_path.dentry->d_inode->i_mode)) 1853 nfs_file_set_open_context(file, ctx); 1854 else 1855 err = -EOPENSTALE; 1856 out: 1857 return err; 1858 } 1859 1860 int nfs_atomic_open(struct inode *dir, struct dentry *dentry, 1861 struct file *file, unsigned open_flags, 1862 umode_t mode) 1863 { 1864 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 1865 struct nfs_open_context *ctx; 1866 struct dentry *res; 1867 struct iattr attr = { .ia_valid = ATTR_OPEN }; 1868 struct inode *inode; 1869 unsigned int lookup_flags = 0; 1870 bool switched = false; 1871 int created = 0; 1872 int err; 1873 1874 /* Expect a negative dentry */ 1875 BUG_ON(d_inode(dentry)); 1876 1877 dfprintk(VFS, "NFS: atomic_open(%s/%lu), %pd\n", 1878 dir->i_sb->s_id, dir->i_ino, dentry); 1879 1880 err = nfs_check_flags(open_flags); 1881 if (err) 1882 return err; 1883 1884 /* NFS only supports OPEN on regular files */ 1885 if ((open_flags & O_DIRECTORY)) { 1886 if (!d_in_lookup(dentry)) { 1887 /* 1888 * Hashed negative dentry with O_DIRECTORY: dentry was 1889 * revalidated and is fine, no need to perform lookup 1890 * again 1891 */ 1892 return -ENOENT; 1893 } 1894 lookup_flags = LOOKUP_OPEN|LOOKUP_DIRECTORY; 1895 goto no_open; 1896 } 1897 1898 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) 1899 return -ENAMETOOLONG; 1900 1901 if (open_flags & O_CREAT) { 1902 struct nfs_server *server = NFS_SERVER(dir); 1903 1904 if (!(server->attr_bitmask[2] & FATTR4_WORD2_MODE_UMASK)) 1905 mode &= ~current_umask(); 1906 1907 attr.ia_valid |= ATTR_MODE; 1908 attr.ia_mode = mode; 1909 } 1910 if (open_flags & O_TRUNC) { 1911 attr.ia_valid |= ATTR_SIZE; 1912 attr.ia_size = 0; 1913 } 1914 1915 if (!(open_flags & O_CREAT) && !d_in_lookup(dentry)) { 1916 d_drop(dentry); 1917 switched = true; 1918 dentry = d_alloc_parallel(dentry->d_parent, 1919 &dentry->d_name, &wq); 1920 if (IS_ERR(dentry)) 1921 return PTR_ERR(dentry); 1922 if (unlikely(!d_in_lookup(dentry))) 1923 return finish_no_open(file, dentry); 1924 } 1925 1926 ctx = create_nfs_open_context(dentry, open_flags, file); 1927 err = PTR_ERR(ctx); 1928 if (IS_ERR(ctx)) 1929 goto out; 1930 1931 trace_nfs_atomic_open_enter(dir, ctx, open_flags); 1932 inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr, &created); 1933 if (created) 1934 file->f_mode |= FMODE_CREATED; 1935 if (IS_ERR(inode)) { 1936 err = PTR_ERR(inode); 1937 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err); 1938 put_nfs_open_context(ctx); 1939 d_drop(dentry); 1940 switch (err) { 1941 case -ENOENT: 1942 d_splice_alias(NULL, dentry); 1943 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1944 break; 1945 case -EISDIR: 1946 case -ENOTDIR: 1947 goto no_open; 1948 case -ELOOP: 1949 if (!(open_flags & O_NOFOLLOW)) 1950 goto no_open; 1951 break; 1952 /* case -EINVAL: */ 1953 default: 1954 break; 1955 } 1956 goto out; 1957 } 1958 1959 err = nfs_finish_open(ctx, ctx->dentry, file, open_flags); 1960 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err); 1961 put_nfs_open_context(ctx); 1962 out: 1963 if (unlikely(switched)) { 1964 d_lookup_done(dentry); 1965 dput(dentry); 1966 } 1967 return err; 1968 1969 no_open: 1970 res = nfs_lookup(dir, dentry, lookup_flags); 1971 if (switched) { 1972 d_lookup_done(dentry); 1973 if (!res) 1974 res = dentry; 1975 else 1976 dput(dentry); 1977 } 1978 if (IS_ERR(res)) 1979 return PTR_ERR(res); 1980 return finish_no_open(file, res); 1981 } 1982 EXPORT_SYMBOL_GPL(nfs_atomic_open); 1983 1984 static int 1985 nfs4_do_lookup_revalidate(struct inode *dir, struct dentry *dentry, 1986 unsigned int flags) 1987 { 1988 struct inode *inode; 1989 1990 if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY)) 1991 goto full_reval; 1992 if (d_mountpoint(dentry)) 1993 goto full_reval; 1994 1995 inode = d_inode(dentry); 1996 1997 /* We can't create new files in nfs_open_revalidate(), so we 1998 * optimize away revalidation of negative dentries. 1999 */ 2000 if (inode == NULL) 2001 goto full_reval; 2002 2003 if (nfs_verifier_is_delegated(dentry)) 2004 return nfs_lookup_revalidate_delegated(dir, dentry, inode); 2005 2006 /* NFS only supports OPEN on regular files */ 2007 if (!S_ISREG(inode->i_mode)) 2008 goto full_reval; 2009 2010 /* We cannot do exclusive creation on a positive dentry */ 2011 if (flags & (LOOKUP_EXCL | LOOKUP_REVAL)) 2012 goto reval_dentry; 2013 2014 /* Check if the directory changed */ 2015 if (!nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU)) 2016 goto reval_dentry; 2017 2018 /* Let f_op->open() actually open (and revalidate) the file */ 2019 return 1; 2020 reval_dentry: 2021 if (flags & LOOKUP_RCU) 2022 return -ECHILD; 2023 return nfs_lookup_revalidate_dentry(dir, dentry, inode); 2024 2025 full_reval: 2026 return nfs_do_lookup_revalidate(dir, dentry, flags); 2027 } 2028 2029 static int nfs4_lookup_revalidate(struct dentry *dentry, unsigned int flags) 2030 { 2031 return __nfs_lookup_revalidate(dentry, flags, 2032 nfs4_do_lookup_revalidate); 2033 } 2034 2035 #endif /* CONFIG_NFSV4 */ 2036 2037 struct dentry * 2038 nfs_add_or_obtain(struct dentry *dentry, struct nfs_fh *fhandle, 2039 struct nfs_fattr *fattr, 2040 struct nfs4_label *label) 2041 { 2042 struct dentry *parent = dget_parent(dentry); 2043 struct inode *dir = d_inode(parent); 2044 struct inode *inode; 2045 struct dentry *d; 2046 int error; 2047 2048 d_drop(dentry); 2049 2050 if (fhandle->size == 0) { 2051 error = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr); 2052 if (error) 2053 goto out_error; 2054 } 2055 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 2056 if (!(fattr->valid & NFS_ATTR_FATTR)) { 2057 struct nfs_server *server = NFS_SB(dentry->d_sb); 2058 error = server->nfs_client->rpc_ops->getattr(server, fhandle, 2059 fattr, NULL); 2060 if (error < 0) 2061 goto out_error; 2062 } 2063 inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label); 2064 d = d_splice_alias(inode, dentry); 2065 out: 2066 dput(parent); 2067 return d; 2068 out_error: 2069 d = ERR_PTR(error); 2070 goto out; 2071 } 2072 EXPORT_SYMBOL_GPL(nfs_add_or_obtain); 2073 2074 /* 2075 * Code common to create, mkdir, and mknod. 2076 */ 2077 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle, 2078 struct nfs_fattr *fattr) 2079 { 2080 struct dentry *d; 2081 2082 d = nfs_add_or_obtain(dentry, fhandle, fattr, fattr->label); 2083 if (IS_ERR(d)) 2084 return PTR_ERR(d); 2085 2086 /* Callers don't care */ 2087 dput(d); 2088 return 0; 2089 } 2090 EXPORT_SYMBOL_GPL(nfs_instantiate); 2091 2092 /* 2093 * Following a failed create operation, we drop the dentry rather 2094 * than retain a negative dentry. This avoids a problem in the event 2095 * that the operation succeeded on the server, but an error in the 2096 * reply path made it appear to have failed. 2097 */ 2098 int nfs_create(struct user_namespace *mnt_userns, struct inode *dir, 2099 struct dentry *dentry, umode_t mode, bool excl) 2100 { 2101 struct iattr attr; 2102 int open_flags = excl ? O_CREAT | O_EXCL : O_CREAT; 2103 int error; 2104 2105 dfprintk(VFS, "NFS: create(%s/%lu), %pd\n", 2106 dir->i_sb->s_id, dir->i_ino, dentry); 2107 2108 attr.ia_mode = mode; 2109 attr.ia_valid = ATTR_MODE; 2110 2111 trace_nfs_create_enter(dir, dentry, open_flags); 2112 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags); 2113 trace_nfs_create_exit(dir, dentry, open_flags, error); 2114 if (error != 0) 2115 goto out_err; 2116 return 0; 2117 out_err: 2118 d_drop(dentry); 2119 return error; 2120 } 2121 EXPORT_SYMBOL_GPL(nfs_create); 2122 2123 /* 2124 * See comments for nfs_proc_create regarding failed operations. 2125 */ 2126 int 2127 nfs_mknod(struct user_namespace *mnt_userns, struct inode *dir, 2128 struct dentry *dentry, umode_t mode, dev_t rdev) 2129 { 2130 struct iattr attr; 2131 int status; 2132 2133 dfprintk(VFS, "NFS: mknod(%s/%lu), %pd\n", 2134 dir->i_sb->s_id, dir->i_ino, dentry); 2135 2136 attr.ia_mode = mode; 2137 attr.ia_valid = ATTR_MODE; 2138 2139 trace_nfs_mknod_enter(dir, dentry); 2140 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev); 2141 trace_nfs_mknod_exit(dir, dentry, status); 2142 if (status != 0) 2143 goto out_err; 2144 return 0; 2145 out_err: 2146 d_drop(dentry); 2147 return status; 2148 } 2149 EXPORT_SYMBOL_GPL(nfs_mknod); 2150 2151 /* 2152 * See comments for nfs_proc_create regarding failed operations. 2153 */ 2154 int nfs_mkdir(struct user_namespace *mnt_userns, struct inode *dir, 2155 struct dentry *dentry, umode_t mode) 2156 { 2157 struct iattr attr; 2158 int error; 2159 2160 dfprintk(VFS, "NFS: mkdir(%s/%lu), %pd\n", 2161 dir->i_sb->s_id, dir->i_ino, dentry); 2162 2163 attr.ia_valid = ATTR_MODE; 2164 attr.ia_mode = mode | S_IFDIR; 2165 2166 trace_nfs_mkdir_enter(dir, dentry); 2167 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr); 2168 trace_nfs_mkdir_exit(dir, dentry, error); 2169 if (error != 0) 2170 goto out_err; 2171 return 0; 2172 out_err: 2173 d_drop(dentry); 2174 return error; 2175 } 2176 EXPORT_SYMBOL_GPL(nfs_mkdir); 2177 2178 static void nfs_dentry_handle_enoent(struct dentry *dentry) 2179 { 2180 if (simple_positive(dentry)) 2181 d_delete(dentry); 2182 } 2183 2184 static void nfs_dentry_remove_handle_error(struct inode *dir, 2185 struct dentry *dentry, int error) 2186 { 2187 switch (error) { 2188 case -ENOENT: 2189 d_delete(dentry); 2190 fallthrough; 2191 case 0: 2192 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 2193 } 2194 } 2195 2196 int nfs_rmdir(struct inode *dir, struct dentry *dentry) 2197 { 2198 int error; 2199 2200 dfprintk(VFS, "NFS: rmdir(%s/%lu), %pd\n", 2201 dir->i_sb->s_id, dir->i_ino, dentry); 2202 2203 trace_nfs_rmdir_enter(dir, dentry); 2204 if (d_really_is_positive(dentry)) { 2205 down_write(&NFS_I(d_inode(dentry))->rmdir_sem); 2206 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name); 2207 /* Ensure the VFS deletes this inode */ 2208 switch (error) { 2209 case 0: 2210 clear_nlink(d_inode(dentry)); 2211 break; 2212 case -ENOENT: 2213 nfs_dentry_handle_enoent(dentry); 2214 } 2215 up_write(&NFS_I(d_inode(dentry))->rmdir_sem); 2216 } else 2217 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name); 2218 nfs_dentry_remove_handle_error(dir, dentry, error); 2219 trace_nfs_rmdir_exit(dir, dentry, error); 2220 2221 return error; 2222 } 2223 EXPORT_SYMBOL_GPL(nfs_rmdir); 2224 2225 /* 2226 * Remove a file after making sure there are no pending writes, 2227 * and after checking that the file has only one user. 2228 * 2229 * We invalidate the attribute cache and free the inode prior to the operation 2230 * to avoid possible races if the server reuses the inode. 2231 */ 2232 static int nfs_safe_remove(struct dentry *dentry) 2233 { 2234 struct inode *dir = d_inode(dentry->d_parent); 2235 struct inode *inode = d_inode(dentry); 2236 int error = -EBUSY; 2237 2238 dfprintk(VFS, "NFS: safe_remove(%pd2)\n", dentry); 2239 2240 /* If the dentry was sillyrenamed, we simply call d_delete() */ 2241 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { 2242 error = 0; 2243 goto out; 2244 } 2245 2246 trace_nfs_remove_enter(dir, dentry); 2247 if (inode != NULL) { 2248 error = NFS_PROTO(dir)->remove(dir, dentry); 2249 if (error == 0) 2250 nfs_drop_nlink(inode); 2251 } else 2252 error = NFS_PROTO(dir)->remove(dir, dentry); 2253 if (error == -ENOENT) 2254 nfs_dentry_handle_enoent(dentry); 2255 trace_nfs_remove_exit(dir, dentry, error); 2256 out: 2257 return error; 2258 } 2259 2260 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode 2261 * belongs to an active ".nfs..." file and we return -EBUSY. 2262 * 2263 * If sillyrename() returns 0, we do nothing, otherwise we unlink. 2264 */ 2265 int nfs_unlink(struct inode *dir, struct dentry *dentry) 2266 { 2267 int error; 2268 int need_rehash = 0; 2269 2270 dfprintk(VFS, "NFS: unlink(%s/%lu, %pd)\n", dir->i_sb->s_id, 2271 dir->i_ino, dentry); 2272 2273 trace_nfs_unlink_enter(dir, dentry); 2274 spin_lock(&dentry->d_lock); 2275 if (d_count(dentry) > 1) { 2276 spin_unlock(&dentry->d_lock); 2277 /* Start asynchronous writeout of the inode */ 2278 write_inode_now(d_inode(dentry), 0); 2279 error = nfs_sillyrename(dir, dentry); 2280 goto out; 2281 } 2282 if (!d_unhashed(dentry)) { 2283 __d_drop(dentry); 2284 need_rehash = 1; 2285 } 2286 spin_unlock(&dentry->d_lock); 2287 error = nfs_safe_remove(dentry); 2288 nfs_dentry_remove_handle_error(dir, dentry, error); 2289 if (need_rehash) 2290 d_rehash(dentry); 2291 out: 2292 trace_nfs_unlink_exit(dir, dentry, error); 2293 return error; 2294 } 2295 EXPORT_SYMBOL_GPL(nfs_unlink); 2296 2297 /* 2298 * To create a symbolic link, most file systems instantiate a new inode, 2299 * add a page to it containing the path, then write it out to the disk 2300 * using prepare_write/commit_write. 2301 * 2302 * Unfortunately the NFS client can't create the in-core inode first 2303 * because it needs a file handle to create an in-core inode (see 2304 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the 2305 * symlink request has completed on the server. 2306 * 2307 * So instead we allocate a raw page, copy the symname into it, then do 2308 * the SYMLINK request with the page as the buffer. If it succeeds, we 2309 * now have a new file handle and can instantiate an in-core NFS inode 2310 * and move the raw page into its mapping. 2311 */ 2312 int nfs_symlink(struct user_namespace *mnt_userns, struct inode *dir, 2313 struct dentry *dentry, const char *symname) 2314 { 2315 struct page *page; 2316 char *kaddr; 2317 struct iattr attr; 2318 unsigned int pathlen = strlen(symname); 2319 int error; 2320 2321 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s)\n", dir->i_sb->s_id, 2322 dir->i_ino, dentry, symname); 2323 2324 if (pathlen > PAGE_SIZE) 2325 return -ENAMETOOLONG; 2326 2327 attr.ia_mode = S_IFLNK | S_IRWXUGO; 2328 attr.ia_valid = ATTR_MODE; 2329 2330 page = alloc_page(GFP_USER); 2331 if (!page) 2332 return -ENOMEM; 2333 2334 kaddr = page_address(page); 2335 memcpy(kaddr, symname, pathlen); 2336 if (pathlen < PAGE_SIZE) 2337 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen); 2338 2339 trace_nfs_symlink_enter(dir, dentry); 2340 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr); 2341 trace_nfs_symlink_exit(dir, dentry, error); 2342 if (error != 0) { 2343 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s) error %d\n", 2344 dir->i_sb->s_id, dir->i_ino, 2345 dentry, symname, error); 2346 d_drop(dentry); 2347 __free_page(page); 2348 return error; 2349 } 2350 2351 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 2352 2353 /* 2354 * No big deal if we can't add this page to the page cache here. 2355 * READLINK will get the missing page from the server if needed. 2356 */ 2357 if (!add_to_page_cache_lru(page, d_inode(dentry)->i_mapping, 0, 2358 GFP_KERNEL)) { 2359 SetPageUptodate(page); 2360 unlock_page(page); 2361 /* 2362 * add_to_page_cache_lru() grabs an extra page refcount. 2363 * Drop it here to avoid leaking this page later. 2364 */ 2365 put_page(page); 2366 } else 2367 __free_page(page); 2368 2369 return 0; 2370 } 2371 EXPORT_SYMBOL_GPL(nfs_symlink); 2372 2373 int 2374 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) 2375 { 2376 struct inode *inode = d_inode(old_dentry); 2377 int error; 2378 2379 dfprintk(VFS, "NFS: link(%pd2 -> %pd2)\n", 2380 old_dentry, dentry); 2381 2382 trace_nfs_link_enter(inode, dir, dentry); 2383 d_drop(dentry); 2384 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name); 2385 if (error == 0) { 2386 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 2387 ihold(inode); 2388 d_add(dentry, inode); 2389 } 2390 trace_nfs_link_exit(inode, dir, dentry, error); 2391 return error; 2392 } 2393 EXPORT_SYMBOL_GPL(nfs_link); 2394 2395 /* 2396 * RENAME 2397 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a 2398 * different file handle for the same inode after a rename (e.g. when 2399 * moving to a different directory). A fail-safe method to do so would 2400 * be to look up old_dir/old_name, create a link to new_dir/new_name and 2401 * rename the old file using the sillyrename stuff. This way, the original 2402 * file in old_dir will go away when the last process iput()s the inode. 2403 * 2404 * FIXED. 2405 * 2406 * It actually works quite well. One needs to have the possibility for 2407 * at least one ".nfs..." file in each directory the file ever gets 2408 * moved or linked to which happens automagically with the new 2409 * implementation that only depends on the dcache stuff instead of 2410 * using the inode layer 2411 * 2412 * Unfortunately, things are a little more complicated than indicated 2413 * above. For a cross-directory move, we want to make sure we can get 2414 * rid of the old inode after the operation. This means there must be 2415 * no pending writes (if it's a file), and the use count must be 1. 2416 * If these conditions are met, we can drop the dentries before doing 2417 * the rename. 2418 */ 2419 int nfs_rename(struct user_namespace *mnt_userns, struct inode *old_dir, 2420 struct dentry *old_dentry, struct inode *new_dir, 2421 struct dentry *new_dentry, unsigned int flags) 2422 { 2423 struct inode *old_inode = d_inode(old_dentry); 2424 struct inode *new_inode = d_inode(new_dentry); 2425 struct dentry *dentry = NULL, *rehash = NULL; 2426 struct rpc_task *task; 2427 int error = -EBUSY; 2428 2429 if (flags) 2430 return -EINVAL; 2431 2432 dfprintk(VFS, "NFS: rename(%pd2 -> %pd2, ct=%d)\n", 2433 old_dentry, new_dentry, 2434 d_count(new_dentry)); 2435 2436 trace_nfs_rename_enter(old_dir, old_dentry, new_dir, new_dentry); 2437 /* 2438 * For non-directories, check whether the target is busy and if so, 2439 * make a copy of the dentry and then do a silly-rename. If the 2440 * silly-rename succeeds, the copied dentry is hashed and becomes 2441 * the new target. 2442 */ 2443 if (new_inode && !S_ISDIR(new_inode->i_mode)) { 2444 /* 2445 * To prevent any new references to the target during the 2446 * rename, we unhash the dentry in advance. 2447 */ 2448 if (!d_unhashed(new_dentry)) { 2449 d_drop(new_dentry); 2450 rehash = new_dentry; 2451 } 2452 2453 if (d_count(new_dentry) > 2) { 2454 int err; 2455 2456 /* copy the target dentry's name */ 2457 dentry = d_alloc(new_dentry->d_parent, 2458 &new_dentry->d_name); 2459 if (!dentry) 2460 goto out; 2461 2462 /* silly-rename the existing target ... */ 2463 err = nfs_sillyrename(new_dir, new_dentry); 2464 if (err) 2465 goto out; 2466 2467 new_dentry = dentry; 2468 rehash = NULL; 2469 new_inode = NULL; 2470 } 2471 } 2472 2473 task = nfs_async_rename(old_dir, new_dir, old_dentry, new_dentry, NULL); 2474 if (IS_ERR(task)) { 2475 error = PTR_ERR(task); 2476 goto out; 2477 } 2478 2479 error = rpc_wait_for_completion_task(task); 2480 if (error != 0) { 2481 ((struct nfs_renamedata *)task->tk_calldata)->cancelled = 1; 2482 /* Paired with the atomic_dec_and_test() barrier in rpc_do_put_task() */ 2483 smp_wmb(); 2484 } else 2485 error = task->tk_status; 2486 rpc_put_task(task); 2487 /* Ensure the inode attributes are revalidated */ 2488 if (error == 0) { 2489 spin_lock(&old_inode->i_lock); 2490 NFS_I(old_inode)->attr_gencount = nfs_inc_attr_generation_counter(); 2491 nfs_set_cache_invalid(old_inode, NFS_INO_INVALID_CHANGE | 2492 NFS_INO_INVALID_CTIME | 2493 NFS_INO_REVAL_FORCED); 2494 spin_unlock(&old_inode->i_lock); 2495 } 2496 out: 2497 if (rehash) 2498 d_rehash(rehash); 2499 trace_nfs_rename_exit(old_dir, old_dentry, 2500 new_dir, new_dentry, error); 2501 if (!error) { 2502 if (new_inode != NULL) 2503 nfs_drop_nlink(new_inode); 2504 /* 2505 * The d_move() should be here instead of in an async RPC completion 2506 * handler because we need the proper locks to move the dentry. If 2507 * we're interrupted by a signal, the async RPC completion handler 2508 * should mark the directories for revalidation. 2509 */ 2510 d_move(old_dentry, new_dentry); 2511 nfs_set_verifier(old_dentry, 2512 nfs_save_change_attribute(new_dir)); 2513 } else if (error == -ENOENT) 2514 nfs_dentry_handle_enoent(old_dentry); 2515 2516 /* new dentry created? */ 2517 if (dentry) 2518 dput(dentry); 2519 return error; 2520 } 2521 EXPORT_SYMBOL_GPL(nfs_rename); 2522 2523 static DEFINE_SPINLOCK(nfs_access_lru_lock); 2524 static LIST_HEAD(nfs_access_lru_list); 2525 static atomic_long_t nfs_access_nr_entries; 2526 2527 static unsigned long nfs_access_max_cachesize = 4*1024*1024; 2528 module_param(nfs_access_max_cachesize, ulong, 0644); 2529 MODULE_PARM_DESC(nfs_access_max_cachesize, "NFS access maximum total cache length"); 2530 2531 static void nfs_access_free_entry(struct nfs_access_entry *entry) 2532 { 2533 put_cred(entry->cred); 2534 kfree_rcu(entry, rcu_head); 2535 smp_mb__before_atomic(); 2536 atomic_long_dec(&nfs_access_nr_entries); 2537 smp_mb__after_atomic(); 2538 } 2539 2540 static void nfs_access_free_list(struct list_head *head) 2541 { 2542 struct nfs_access_entry *cache; 2543 2544 while (!list_empty(head)) { 2545 cache = list_entry(head->next, struct nfs_access_entry, lru); 2546 list_del(&cache->lru); 2547 nfs_access_free_entry(cache); 2548 } 2549 } 2550 2551 static unsigned long 2552 nfs_do_access_cache_scan(unsigned int nr_to_scan) 2553 { 2554 LIST_HEAD(head); 2555 struct nfs_inode *nfsi, *next; 2556 struct nfs_access_entry *cache; 2557 long freed = 0; 2558 2559 spin_lock(&nfs_access_lru_lock); 2560 list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) { 2561 struct inode *inode; 2562 2563 if (nr_to_scan-- == 0) 2564 break; 2565 inode = &nfsi->vfs_inode; 2566 spin_lock(&inode->i_lock); 2567 if (list_empty(&nfsi->access_cache_entry_lru)) 2568 goto remove_lru_entry; 2569 cache = list_entry(nfsi->access_cache_entry_lru.next, 2570 struct nfs_access_entry, lru); 2571 list_move(&cache->lru, &head); 2572 rb_erase(&cache->rb_node, &nfsi->access_cache); 2573 freed++; 2574 if (!list_empty(&nfsi->access_cache_entry_lru)) 2575 list_move_tail(&nfsi->access_cache_inode_lru, 2576 &nfs_access_lru_list); 2577 else { 2578 remove_lru_entry: 2579 list_del_init(&nfsi->access_cache_inode_lru); 2580 smp_mb__before_atomic(); 2581 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags); 2582 smp_mb__after_atomic(); 2583 } 2584 spin_unlock(&inode->i_lock); 2585 } 2586 spin_unlock(&nfs_access_lru_lock); 2587 nfs_access_free_list(&head); 2588 return freed; 2589 } 2590 2591 unsigned long 2592 nfs_access_cache_scan(struct shrinker *shrink, struct shrink_control *sc) 2593 { 2594 int nr_to_scan = sc->nr_to_scan; 2595 gfp_t gfp_mask = sc->gfp_mask; 2596 2597 if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL) 2598 return SHRINK_STOP; 2599 return nfs_do_access_cache_scan(nr_to_scan); 2600 } 2601 2602 2603 unsigned long 2604 nfs_access_cache_count(struct shrinker *shrink, struct shrink_control *sc) 2605 { 2606 return vfs_pressure_ratio(atomic_long_read(&nfs_access_nr_entries)); 2607 } 2608 2609 static void 2610 nfs_access_cache_enforce_limit(void) 2611 { 2612 long nr_entries = atomic_long_read(&nfs_access_nr_entries); 2613 unsigned long diff; 2614 unsigned int nr_to_scan; 2615 2616 if (nr_entries < 0 || nr_entries <= nfs_access_max_cachesize) 2617 return; 2618 nr_to_scan = 100; 2619 diff = nr_entries - nfs_access_max_cachesize; 2620 if (diff < nr_to_scan) 2621 nr_to_scan = diff; 2622 nfs_do_access_cache_scan(nr_to_scan); 2623 } 2624 2625 static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head) 2626 { 2627 struct rb_root *root_node = &nfsi->access_cache; 2628 struct rb_node *n; 2629 struct nfs_access_entry *entry; 2630 2631 /* Unhook entries from the cache */ 2632 while ((n = rb_first(root_node)) != NULL) { 2633 entry = rb_entry(n, struct nfs_access_entry, rb_node); 2634 rb_erase(n, root_node); 2635 list_move(&entry->lru, head); 2636 } 2637 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS; 2638 } 2639 2640 void nfs_access_zap_cache(struct inode *inode) 2641 { 2642 LIST_HEAD(head); 2643 2644 if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0) 2645 return; 2646 /* Remove from global LRU init */ 2647 spin_lock(&nfs_access_lru_lock); 2648 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) 2649 list_del_init(&NFS_I(inode)->access_cache_inode_lru); 2650 2651 spin_lock(&inode->i_lock); 2652 __nfs_access_zap_cache(NFS_I(inode), &head); 2653 spin_unlock(&inode->i_lock); 2654 spin_unlock(&nfs_access_lru_lock); 2655 nfs_access_free_list(&head); 2656 } 2657 EXPORT_SYMBOL_GPL(nfs_access_zap_cache); 2658 2659 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, const struct cred *cred) 2660 { 2661 struct rb_node *n = NFS_I(inode)->access_cache.rb_node; 2662 2663 while (n != NULL) { 2664 struct nfs_access_entry *entry = 2665 rb_entry(n, struct nfs_access_entry, rb_node); 2666 int cmp = cred_fscmp(cred, entry->cred); 2667 2668 if (cmp < 0) 2669 n = n->rb_left; 2670 else if (cmp > 0) 2671 n = n->rb_right; 2672 else 2673 return entry; 2674 } 2675 return NULL; 2676 } 2677 2678 static int nfs_access_get_cached_locked(struct inode *inode, const struct cred *cred, struct nfs_access_entry *res, bool may_block) 2679 { 2680 struct nfs_inode *nfsi = NFS_I(inode); 2681 struct nfs_access_entry *cache; 2682 bool retry = true; 2683 int err; 2684 2685 spin_lock(&inode->i_lock); 2686 for(;;) { 2687 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS) 2688 goto out_zap; 2689 cache = nfs_access_search_rbtree(inode, cred); 2690 err = -ENOENT; 2691 if (cache == NULL) 2692 goto out; 2693 /* Found an entry, is our attribute cache valid? */ 2694 if (!nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS)) 2695 break; 2696 if (!retry) 2697 break; 2698 err = -ECHILD; 2699 if (!may_block) 2700 goto out; 2701 spin_unlock(&inode->i_lock); 2702 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode); 2703 if (err) 2704 return err; 2705 spin_lock(&inode->i_lock); 2706 retry = false; 2707 } 2708 res->cred = cache->cred; 2709 res->mask = cache->mask; 2710 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru); 2711 err = 0; 2712 out: 2713 spin_unlock(&inode->i_lock); 2714 return err; 2715 out_zap: 2716 spin_unlock(&inode->i_lock); 2717 nfs_access_zap_cache(inode); 2718 return -ENOENT; 2719 } 2720 2721 static int nfs_access_get_cached_rcu(struct inode *inode, const struct cred *cred, struct nfs_access_entry *res) 2722 { 2723 /* Only check the most recently returned cache entry, 2724 * but do it without locking. 2725 */ 2726 struct nfs_inode *nfsi = NFS_I(inode); 2727 struct nfs_access_entry *cache; 2728 int err = -ECHILD; 2729 struct list_head *lh; 2730 2731 rcu_read_lock(); 2732 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS) 2733 goto out; 2734 lh = rcu_dereference(list_tail_rcu(&nfsi->access_cache_entry_lru)); 2735 cache = list_entry(lh, struct nfs_access_entry, lru); 2736 if (lh == &nfsi->access_cache_entry_lru || 2737 cred_fscmp(cred, cache->cred) != 0) 2738 cache = NULL; 2739 if (cache == NULL) 2740 goto out; 2741 if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS)) 2742 goto out; 2743 res->cred = cache->cred; 2744 res->mask = cache->mask; 2745 err = 0; 2746 out: 2747 rcu_read_unlock(); 2748 return err; 2749 } 2750 2751 int nfs_access_get_cached(struct inode *inode, const struct cred *cred, struct 2752 nfs_access_entry *res, bool may_block) 2753 { 2754 int status; 2755 2756 status = nfs_access_get_cached_rcu(inode, cred, res); 2757 if (status != 0) 2758 status = nfs_access_get_cached_locked(inode, cred, res, 2759 may_block); 2760 2761 return status; 2762 } 2763 EXPORT_SYMBOL_GPL(nfs_access_get_cached); 2764 2765 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set) 2766 { 2767 struct nfs_inode *nfsi = NFS_I(inode); 2768 struct rb_root *root_node = &nfsi->access_cache; 2769 struct rb_node **p = &root_node->rb_node; 2770 struct rb_node *parent = NULL; 2771 struct nfs_access_entry *entry; 2772 int cmp; 2773 2774 spin_lock(&inode->i_lock); 2775 while (*p != NULL) { 2776 parent = *p; 2777 entry = rb_entry(parent, struct nfs_access_entry, rb_node); 2778 cmp = cred_fscmp(set->cred, entry->cred); 2779 2780 if (cmp < 0) 2781 p = &parent->rb_left; 2782 else if (cmp > 0) 2783 p = &parent->rb_right; 2784 else 2785 goto found; 2786 } 2787 rb_link_node(&set->rb_node, parent, p); 2788 rb_insert_color(&set->rb_node, root_node); 2789 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); 2790 spin_unlock(&inode->i_lock); 2791 return; 2792 found: 2793 rb_replace_node(parent, &set->rb_node, root_node); 2794 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); 2795 list_del(&entry->lru); 2796 spin_unlock(&inode->i_lock); 2797 nfs_access_free_entry(entry); 2798 } 2799 2800 void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set) 2801 { 2802 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL); 2803 if (cache == NULL) 2804 return; 2805 RB_CLEAR_NODE(&cache->rb_node); 2806 cache->cred = get_cred(set->cred); 2807 cache->mask = set->mask; 2808 2809 /* The above field assignments must be visible 2810 * before this item appears on the lru. We cannot easily 2811 * use rcu_assign_pointer, so just force the memory barrier. 2812 */ 2813 smp_wmb(); 2814 nfs_access_add_rbtree(inode, cache); 2815 2816 /* Update accounting */ 2817 smp_mb__before_atomic(); 2818 atomic_long_inc(&nfs_access_nr_entries); 2819 smp_mb__after_atomic(); 2820 2821 /* Add inode to global LRU list */ 2822 if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) { 2823 spin_lock(&nfs_access_lru_lock); 2824 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) 2825 list_add_tail(&NFS_I(inode)->access_cache_inode_lru, 2826 &nfs_access_lru_list); 2827 spin_unlock(&nfs_access_lru_lock); 2828 } 2829 nfs_access_cache_enforce_limit(); 2830 } 2831 EXPORT_SYMBOL_GPL(nfs_access_add_cache); 2832 2833 #define NFS_MAY_READ (NFS_ACCESS_READ) 2834 #define NFS_MAY_WRITE (NFS_ACCESS_MODIFY | \ 2835 NFS_ACCESS_EXTEND | \ 2836 NFS_ACCESS_DELETE) 2837 #define NFS_FILE_MAY_WRITE (NFS_ACCESS_MODIFY | \ 2838 NFS_ACCESS_EXTEND) 2839 #define NFS_DIR_MAY_WRITE NFS_MAY_WRITE 2840 #define NFS_MAY_LOOKUP (NFS_ACCESS_LOOKUP) 2841 #define NFS_MAY_EXECUTE (NFS_ACCESS_EXECUTE) 2842 static int 2843 nfs_access_calc_mask(u32 access_result, umode_t umode) 2844 { 2845 int mask = 0; 2846 2847 if (access_result & NFS_MAY_READ) 2848 mask |= MAY_READ; 2849 if (S_ISDIR(umode)) { 2850 if ((access_result & NFS_DIR_MAY_WRITE) == NFS_DIR_MAY_WRITE) 2851 mask |= MAY_WRITE; 2852 if ((access_result & NFS_MAY_LOOKUP) == NFS_MAY_LOOKUP) 2853 mask |= MAY_EXEC; 2854 } else if (S_ISREG(umode)) { 2855 if ((access_result & NFS_FILE_MAY_WRITE) == NFS_FILE_MAY_WRITE) 2856 mask |= MAY_WRITE; 2857 if ((access_result & NFS_MAY_EXECUTE) == NFS_MAY_EXECUTE) 2858 mask |= MAY_EXEC; 2859 } else if (access_result & NFS_MAY_WRITE) 2860 mask |= MAY_WRITE; 2861 return mask; 2862 } 2863 2864 void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result) 2865 { 2866 entry->mask = access_result; 2867 } 2868 EXPORT_SYMBOL_GPL(nfs_access_set_mask); 2869 2870 static int nfs_do_access(struct inode *inode, const struct cred *cred, int mask) 2871 { 2872 struct nfs_access_entry cache; 2873 bool may_block = (mask & MAY_NOT_BLOCK) == 0; 2874 int cache_mask = -1; 2875 int status; 2876 2877 trace_nfs_access_enter(inode); 2878 2879 status = nfs_access_get_cached(inode, cred, &cache, may_block); 2880 if (status == 0) 2881 goto out_cached; 2882 2883 status = -ECHILD; 2884 if (!may_block) 2885 goto out; 2886 2887 /* 2888 * Determine which access bits we want to ask for... 2889 */ 2890 cache.mask = NFS_ACCESS_READ | NFS_ACCESS_MODIFY | NFS_ACCESS_EXTEND; 2891 if (nfs_server_capable(inode, NFS_CAP_XATTR)) { 2892 cache.mask |= NFS_ACCESS_XAREAD | NFS_ACCESS_XAWRITE | 2893 NFS_ACCESS_XALIST; 2894 } 2895 if (S_ISDIR(inode->i_mode)) 2896 cache.mask |= NFS_ACCESS_DELETE | NFS_ACCESS_LOOKUP; 2897 else 2898 cache.mask |= NFS_ACCESS_EXECUTE; 2899 cache.cred = cred; 2900 status = NFS_PROTO(inode)->access(inode, &cache); 2901 if (status != 0) { 2902 if (status == -ESTALE) { 2903 if (!S_ISDIR(inode->i_mode)) 2904 nfs_set_inode_stale(inode); 2905 else 2906 nfs_zap_caches(inode); 2907 } 2908 goto out; 2909 } 2910 nfs_access_add_cache(inode, &cache); 2911 out_cached: 2912 cache_mask = nfs_access_calc_mask(cache.mask, inode->i_mode); 2913 if ((mask & ~cache_mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) != 0) 2914 status = -EACCES; 2915 out: 2916 trace_nfs_access_exit(inode, mask, cache_mask, status); 2917 return status; 2918 } 2919 2920 static int nfs_open_permission_mask(int openflags) 2921 { 2922 int mask = 0; 2923 2924 if (openflags & __FMODE_EXEC) { 2925 /* ONLY check exec rights */ 2926 mask = MAY_EXEC; 2927 } else { 2928 if ((openflags & O_ACCMODE) != O_WRONLY) 2929 mask |= MAY_READ; 2930 if ((openflags & O_ACCMODE) != O_RDONLY) 2931 mask |= MAY_WRITE; 2932 } 2933 2934 return mask; 2935 } 2936 2937 int nfs_may_open(struct inode *inode, const struct cred *cred, int openflags) 2938 { 2939 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags)); 2940 } 2941 EXPORT_SYMBOL_GPL(nfs_may_open); 2942 2943 static int nfs_execute_ok(struct inode *inode, int mask) 2944 { 2945 struct nfs_server *server = NFS_SERVER(inode); 2946 int ret = 0; 2947 2948 if (S_ISDIR(inode->i_mode)) 2949 return 0; 2950 if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_MODE)) { 2951 if (mask & MAY_NOT_BLOCK) 2952 return -ECHILD; 2953 ret = __nfs_revalidate_inode(server, inode); 2954 } 2955 if (ret == 0 && !execute_ok(inode)) 2956 ret = -EACCES; 2957 return ret; 2958 } 2959 2960 int nfs_permission(struct user_namespace *mnt_userns, 2961 struct inode *inode, 2962 int mask) 2963 { 2964 const struct cred *cred = current_cred(); 2965 int res = 0; 2966 2967 nfs_inc_stats(inode, NFSIOS_VFSACCESS); 2968 2969 if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) 2970 goto out; 2971 /* Is this sys_access() ? */ 2972 if (mask & (MAY_ACCESS | MAY_CHDIR)) 2973 goto force_lookup; 2974 2975 switch (inode->i_mode & S_IFMT) { 2976 case S_IFLNK: 2977 goto out; 2978 case S_IFREG: 2979 if ((mask & MAY_OPEN) && 2980 nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN)) 2981 return 0; 2982 break; 2983 case S_IFDIR: 2984 /* 2985 * Optimize away all write operations, since the server 2986 * will check permissions when we perform the op. 2987 */ 2988 if ((mask & MAY_WRITE) && !(mask & MAY_READ)) 2989 goto out; 2990 } 2991 2992 force_lookup: 2993 if (!NFS_PROTO(inode)->access) 2994 goto out_notsup; 2995 2996 res = nfs_do_access(inode, cred, mask); 2997 out: 2998 if (!res && (mask & MAY_EXEC)) 2999 res = nfs_execute_ok(inode, mask); 3000 3001 dfprintk(VFS, "NFS: permission(%s/%lu), mask=0x%x, res=%d\n", 3002 inode->i_sb->s_id, inode->i_ino, mask, res); 3003 return res; 3004 out_notsup: 3005 if (mask & MAY_NOT_BLOCK) 3006 return -ECHILD; 3007 3008 res = nfs_revalidate_inode(inode, NFS_INO_INVALID_MODE | 3009 NFS_INO_INVALID_OTHER); 3010 if (res == 0) 3011 res = generic_permission(&init_user_ns, inode, mask); 3012 goto out; 3013 } 3014 EXPORT_SYMBOL_GPL(nfs_permission); 3015