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 spin_lock(&dir->i_lock); 82 if (list_empty(&nfsi->open_files) && 83 (nfsi->cache_validity & NFS_INO_DATA_INVAL_DEFER)) 84 nfs_set_cache_invalid(dir, 85 NFS_INO_INVALID_DATA | 86 NFS_INO_REVAL_FORCED); 87 list_add(&ctx->list, &nfsi->open_files); 88 spin_unlock(&dir->i_lock); 89 return ctx; 90 } 91 return ERR_PTR(-ENOMEM); 92 } 93 94 static void put_nfs_open_dir_context(struct inode *dir, struct nfs_open_dir_context *ctx) 95 { 96 spin_lock(&dir->i_lock); 97 list_del(&ctx->list); 98 spin_unlock(&dir->i_lock); 99 kfree(ctx); 100 } 101 102 /* 103 * Open file 104 */ 105 static int 106 nfs_opendir(struct inode *inode, struct file *filp) 107 { 108 int res = 0; 109 struct nfs_open_dir_context *ctx; 110 111 dfprintk(FILE, "NFS: open dir(%pD2)\n", filp); 112 113 nfs_inc_stats(inode, NFSIOS_VFSOPEN); 114 115 ctx = alloc_nfs_open_dir_context(inode); 116 if (IS_ERR(ctx)) { 117 res = PTR_ERR(ctx); 118 goto out; 119 } 120 filp->private_data = ctx; 121 out: 122 return res; 123 } 124 125 static int 126 nfs_closedir(struct inode *inode, struct file *filp) 127 { 128 put_nfs_open_dir_context(file_inode(filp), filp->private_data); 129 return 0; 130 } 131 132 struct nfs_cache_array_entry { 133 u64 cookie; 134 u64 ino; 135 const char *name; 136 unsigned int name_len; 137 unsigned char d_type; 138 }; 139 140 struct nfs_cache_array { 141 u64 last_cookie; 142 unsigned int size; 143 unsigned char page_full : 1, 144 page_is_eof : 1, 145 cookies_are_ordered : 1; 146 struct nfs_cache_array_entry array[]; 147 }; 148 149 struct nfs_readdir_descriptor { 150 struct file *file; 151 struct page *page; 152 struct dir_context *ctx; 153 pgoff_t page_index; 154 u64 dir_cookie; 155 u64 last_cookie; 156 u64 dup_cookie; 157 loff_t current_index; 158 loff_t prev_index; 159 160 __be32 verf[NFS_DIR_VERIFIER_SIZE]; 161 unsigned long dir_verifier; 162 unsigned long timestamp; 163 unsigned long gencount; 164 unsigned long attr_gencount; 165 unsigned int cache_entry_index; 166 signed char duped; 167 bool plus; 168 bool eof; 169 }; 170 171 static void nfs_readdir_array_init(struct nfs_cache_array *array) 172 { 173 memset(array, 0, sizeof(struct nfs_cache_array)); 174 } 175 176 static void nfs_readdir_page_init_array(struct page *page, u64 last_cookie) 177 { 178 struct nfs_cache_array *array; 179 180 array = kmap_atomic(page); 181 nfs_readdir_array_init(array); 182 array->last_cookie = last_cookie; 183 array->cookies_are_ordered = 1; 184 kunmap_atomic(array); 185 } 186 187 /* 188 * we are freeing strings created by nfs_add_to_readdir_array() 189 */ 190 static 191 void nfs_readdir_clear_array(struct page *page) 192 { 193 struct nfs_cache_array *array; 194 int i; 195 196 array = kmap_atomic(page); 197 for (i = 0; i < array->size; i++) 198 kfree(array->array[i].name); 199 nfs_readdir_array_init(array); 200 kunmap_atomic(array); 201 } 202 203 static struct page * 204 nfs_readdir_page_array_alloc(u64 last_cookie, gfp_t gfp_flags) 205 { 206 struct page *page = alloc_page(gfp_flags); 207 if (page) 208 nfs_readdir_page_init_array(page, last_cookie); 209 return page; 210 } 211 212 static void nfs_readdir_page_array_free(struct page *page) 213 { 214 if (page) { 215 nfs_readdir_clear_array(page); 216 put_page(page); 217 } 218 } 219 220 static void nfs_readdir_array_set_eof(struct nfs_cache_array *array) 221 { 222 array->page_is_eof = 1; 223 array->page_full = 1; 224 } 225 226 static bool nfs_readdir_array_is_full(struct nfs_cache_array *array) 227 { 228 return array->page_full; 229 } 230 231 /* 232 * the caller is responsible for freeing qstr.name 233 * when called by nfs_readdir_add_to_array, the strings will be freed in 234 * nfs_clear_readdir_array() 235 */ 236 static const char *nfs_readdir_copy_name(const char *name, unsigned int len) 237 { 238 const char *ret = kmemdup_nul(name, len, GFP_KERNEL); 239 240 /* 241 * Avoid a kmemleak false positive. The pointer to the name is stored 242 * in a page cache page which kmemleak does not scan. 243 */ 244 if (ret != NULL) 245 kmemleak_not_leak(ret); 246 return ret; 247 } 248 249 /* 250 * Check that the next array entry lies entirely within the page bounds 251 */ 252 static int nfs_readdir_array_can_expand(struct nfs_cache_array *array) 253 { 254 struct nfs_cache_array_entry *cache_entry; 255 256 if (array->page_full) 257 return -ENOSPC; 258 cache_entry = &array->array[array->size + 1]; 259 if ((char *)cache_entry - (char *)array > PAGE_SIZE) { 260 array->page_full = 1; 261 return -ENOSPC; 262 } 263 return 0; 264 } 265 266 static 267 int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page) 268 { 269 struct nfs_cache_array *array; 270 struct nfs_cache_array_entry *cache_entry; 271 const char *name; 272 int ret; 273 274 name = nfs_readdir_copy_name(entry->name, entry->len); 275 if (!name) 276 return -ENOMEM; 277 278 array = kmap_atomic(page); 279 ret = nfs_readdir_array_can_expand(array); 280 if (ret) { 281 kfree(name); 282 goto out; 283 } 284 285 cache_entry = &array->array[array->size]; 286 cache_entry->cookie = entry->prev_cookie; 287 cache_entry->ino = entry->ino; 288 cache_entry->d_type = entry->d_type; 289 cache_entry->name_len = entry->len; 290 cache_entry->name = name; 291 array->last_cookie = entry->cookie; 292 if (array->last_cookie <= cache_entry->cookie) 293 array->cookies_are_ordered = 0; 294 array->size++; 295 if (entry->eof != 0) 296 nfs_readdir_array_set_eof(array); 297 out: 298 kunmap_atomic(array); 299 return ret; 300 } 301 302 static struct page *nfs_readdir_page_get_locked(struct address_space *mapping, 303 pgoff_t index, u64 last_cookie) 304 { 305 struct page *page; 306 307 page = grab_cache_page(mapping, index); 308 if (page && !PageUptodate(page)) { 309 nfs_readdir_page_init_array(page, last_cookie); 310 if (invalidate_inode_pages2_range(mapping, index + 1, -1) < 0) 311 nfs_zap_mapping(mapping->host, mapping); 312 SetPageUptodate(page); 313 } 314 315 return page; 316 } 317 318 static u64 nfs_readdir_page_last_cookie(struct page *page) 319 { 320 struct nfs_cache_array *array; 321 u64 ret; 322 323 array = kmap_atomic(page); 324 ret = array->last_cookie; 325 kunmap_atomic(array); 326 return ret; 327 } 328 329 static bool nfs_readdir_page_needs_filling(struct page *page) 330 { 331 struct nfs_cache_array *array; 332 bool ret; 333 334 array = kmap_atomic(page); 335 ret = !nfs_readdir_array_is_full(array); 336 kunmap_atomic(array); 337 return ret; 338 } 339 340 static void nfs_readdir_page_set_eof(struct page *page) 341 { 342 struct nfs_cache_array *array; 343 344 array = kmap_atomic(page); 345 nfs_readdir_array_set_eof(array); 346 kunmap_atomic(array); 347 } 348 349 static void nfs_readdir_page_unlock_and_put(struct page *page) 350 { 351 unlock_page(page); 352 put_page(page); 353 } 354 355 static struct page *nfs_readdir_page_get_next(struct address_space *mapping, 356 pgoff_t index, u64 cookie) 357 { 358 struct page *page; 359 360 page = nfs_readdir_page_get_locked(mapping, index, cookie); 361 if (page) { 362 if (nfs_readdir_page_last_cookie(page) == cookie) 363 return page; 364 nfs_readdir_page_unlock_and_put(page); 365 } 366 return NULL; 367 } 368 369 static inline 370 int is_32bit_api(void) 371 { 372 #ifdef CONFIG_COMPAT 373 return in_compat_syscall(); 374 #else 375 return (BITS_PER_LONG == 32); 376 #endif 377 } 378 379 static 380 bool nfs_readdir_use_cookie(const struct file *filp) 381 { 382 if ((filp->f_mode & FMODE_32BITHASH) || 383 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api())) 384 return false; 385 return true; 386 } 387 388 static int nfs_readdir_search_for_pos(struct nfs_cache_array *array, 389 struct nfs_readdir_descriptor *desc) 390 { 391 loff_t diff = desc->ctx->pos - desc->current_index; 392 unsigned int index; 393 394 if (diff < 0) 395 goto out_eof; 396 if (diff >= array->size) { 397 if (array->page_is_eof) 398 goto out_eof; 399 return -EAGAIN; 400 } 401 402 index = (unsigned int)diff; 403 desc->dir_cookie = array->array[index].cookie; 404 desc->cache_entry_index = index; 405 return 0; 406 out_eof: 407 desc->eof = true; 408 return -EBADCOOKIE; 409 } 410 411 static bool 412 nfs_readdir_inode_mapping_valid(struct nfs_inode *nfsi) 413 { 414 if (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA)) 415 return false; 416 smp_rmb(); 417 return !test_bit(NFS_INO_INVALIDATING, &nfsi->flags); 418 } 419 420 static bool nfs_readdir_array_cookie_in_range(struct nfs_cache_array *array, 421 u64 cookie) 422 { 423 if (!array->cookies_are_ordered) 424 return true; 425 /* Optimisation for monotonically increasing cookies */ 426 if (cookie >= array->last_cookie) 427 return false; 428 if (array->size && cookie < array->array[0].cookie) 429 return false; 430 return true; 431 } 432 433 static int nfs_readdir_search_for_cookie(struct nfs_cache_array *array, 434 struct nfs_readdir_descriptor *desc) 435 { 436 int i; 437 loff_t new_pos; 438 int status = -EAGAIN; 439 440 if (!nfs_readdir_array_cookie_in_range(array, desc->dir_cookie)) 441 goto check_eof; 442 443 for (i = 0; i < array->size; i++) { 444 if (array->array[i].cookie == desc->dir_cookie) { 445 struct nfs_inode *nfsi = NFS_I(file_inode(desc->file)); 446 447 new_pos = desc->current_index + i; 448 if (desc->attr_gencount != nfsi->attr_gencount || 449 !nfs_readdir_inode_mapping_valid(nfsi)) { 450 desc->duped = 0; 451 desc->attr_gencount = nfsi->attr_gencount; 452 } else if (new_pos < desc->prev_index) { 453 if (desc->duped > 0 454 && desc->dup_cookie == desc->dir_cookie) { 455 if (printk_ratelimit()) { 456 pr_notice("NFS: directory %pD2 contains a readdir loop." 457 "Please contact your server vendor. " 458 "The file: %s has duplicate cookie %llu\n", 459 desc->file, array->array[i].name, desc->dir_cookie); 460 } 461 status = -ELOOP; 462 goto out; 463 } 464 desc->dup_cookie = desc->dir_cookie; 465 desc->duped = -1; 466 } 467 if (nfs_readdir_use_cookie(desc->file)) 468 desc->ctx->pos = desc->dir_cookie; 469 else 470 desc->ctx->pos = new_pos; 471 desc->prev_index = new_pos; 472 desc->cache_entry_index = i; 473 return 0; 474 } 475 } 476 check_eof: 477 if (array->page_is_eof) { 478 status = -EBADCOOKIE; 479 if (desc->dir_cookie == array->last_cookie) 480 desc->eof = true; 481 } 482 out: 483 return status; 484 } 485 486 static int nfs_readdir_search_array(struct nfs_readdir_descriptor *desc) 487 { 488 struct nfs_cache_array *array; 489 int status; 490 491 array = kmap_atomic(desc->page); 492 493 if (desc->dir_cookie == 0) 494 status = nfs_readdir_search_for_pos(array, desc); 495 else 496 status = nfs_readdir_search_for_cookie(array, desc); 497 498 if (status == -EAGAIN) { 499 desc->last_cookie = array->last_cookie; 500 desc->current_index += array->size; 501 desc->page_index++; 502 } 503 kunmap_atomic(array); 504 return status; 505 } 506 507 /* Fill a page with xdr information before transferring to the cache page */ 508 static int nfs_readdir_xdr_filler(struct nfs_readdir_descriptor *desc, 509 __be32 *verf, u64 cookie, 510 struct page **pages, size_t bufsize, 511 __be32 *verf_res) 512 { 513 struct inode *inode = file_inode(desc->file); 514 struct nfs_readdir_arg arg = { 515 .dentry = file_dentry(desc->file), 516 .cred = desc->file->f_cred, 517 .verf = verf, 518 .cookie = cookie, 519 .pages = pages, 520 .page_len = bufsize, 521 .plus = desc->plus, 522 }; 523 struct nfs_readdir_res res = { 524 .verf = verf_res, 525 }; 526 unsigned long timestamp, gencount; 527 int error; 528 529 again: 530 timestamp = jiffies; 531 gencount = nfs_inc_attr_generation_counter(); 532 desc->dir_verifier = nfs_save_change_attribute(inode); 533 error = NFS_PROTO(inode)->readdir(&arg, &res); 534 if (error < 0) { 535 /* We requested READDIRPLUS, but the server doesn't grok it */ 536 if (error == -ENOTSUPP && desc->plus) { 537 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS; 538 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags); 539 desc->plus = arg.plus = false; 540 goto again; 541 } 542 goto error; 543 } 544 desc->timestamp = timestamp; 545 desc->gencount = gencount; 546 error: 547 return error; 548 } 549 550 static int xdr_decode(struct nfs_readdir_descriptor *desc, 551 struct nfs_entry *entry, struct xdr_stream *xdr) 552 { 553 struct inode *inode = file_inode(desc->file); 554 int error; 555 556 error = NFS_PROTO(inode)->decode_dirent(xdr, entry, desc->plus); 557 if (error) 558 return error; 559 entry->fattr->time_start = desc->timestamp; 560 entry->fattr->gencount = desc->gencount; 561 return 0; 562 } 563 564 /* Match file and dirent using either filehandle or fileid 565 * Note: caller is responsible for checking the fsid 566 */ 567 static 568 int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry) 569 { 570 struct inode *inode; 571 struct nfs_inode *nfsi; 572 573 if (d_really_is_negative(dentry)) 574 return 0; 575 576 inode = d_inode(dentry); 577 if (is_bad_inode(inode) || NFS_STALE(inode)) 578 return 0; 579 580 nfsi = NFS_I(inode); 581 if (entry->fattr->fileid != nfsi->fileid) 582 return 0; 583 if (entry->fh->size && nfs_compare_fh(entry->fh, &nfsi->fh) != 0) 584 return 0; 585 return 1; 586 } 587 588 static 589 bool nfs_use_readdirplus(struct inode *dir, struct dir_context *ctx) 590 { 591 if (!nfs_server_capable(dir, NFS_CAP_READDIRPLUS)) 592 return false; 593 if (test_and_clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags)) 594 return true; 595 if (ctx->pos == 0) 596 return true; 597 return false; 598 } 599 600 /* 601 * This function is called by the lookup and getattr code to request the 602 * use of readdirplus to accelerate any future lookups in the same 603 * directory. 604 */ 605 void nfs_advise_use_readdirplus(struct inode *dir) 606 { 607 struct nfs_inode *nfsi = NFS_I(dir); 608 609 if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) && 610 !list_empty(&nfsi->open_files)) 611 set_bit(NFS_INO_ADVISE_RDPLUS, &nfsi->flags); 612 } 613 614 /* 615 * This function is mainly for use by nfs_getattr(). 616 * 617 * If this is an 'ls -l', we want to force use of readdirplus. 618 * Do this by checking if there is an active file descriptor 619 * and calling nfs_advise_use_readdirplus, then forcing a 620 * cache flush. 621 */ 622 void nfs_force_use_readdirplus(struct inode *dir) 623 { 624 struct nfs_inode *nfsi = NFS_I(dir); 625 626 if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) && 627 !list_empty(&nfsi->open_files)) { 628 set_bit(NFS_INO_ADVISE_RDPLUS, &nfsi->flags); 629 invalidate_mapping_pages(dir->i_mapping, 630 nfsi->page_index + 1, -1); 631 } 632 } 633 634 static 635 void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry, 636 unsigned long dir_verifier) 637 { 638 struct qstr filename = QSTR_INIT(entry->name, entry->len); 639 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 640 struct dentry *dentry; 641 struct dentry *alias; 642 struct inode *inode; 643 int status; 644 645 if (!(entry->fattr->valid & NFS_ATTR_FATTR_FILEID)) 646 return; 647 if (!(entry->fattr->valid & NFS_ATTR_FATTR_FSID)) 648 return; 649 if (filename.len == 0) 650 return; 651 /* Validate that the name doesn't contain any illegal '\0' */ 652 if (strnlen(filename.name, filename.len) != filename.len) 653 return; 654 /* ...or '/' */ 655 if (strnchr(filename.name, filename.len, '/')) 656 return; 657 if (filename.name[0] == '.') { 658 if (filename.len == 1) 659 return; 660 if (filename.len == 2 && filename.name[1] == '.') 661 return; 662 } 663 filename.hash = full_name_hash(parent, filename.name, filename.len); 664 665 dentry = d_lookup(parent, &filename); 666 again: 667 if (!dentry) { 668 dentry = d_alloc_parallel(parent, &filename, &wq); 669 if (IS_ERR(dentry)) 670 return; 671 } 672 if (!d_in_lookup(dentry)) { 673 /* Is there a mountpoint here? If so, just exit */ 674 if (!nfs_fsid_equal(&NFS_SB(dentry->d_sb)->fsid, 675 &entry->fattr->fsid)) 676 goto out; 677 if (nfs_same_file(dentry, entry)) { 678 if (!entry->fh->size) 679 goto out; 680 nfs_set_verifier(dentry, dir_verifier); 681 status = nfs_refresh_inode(d_inode(dentry), entry->fattr); 682 if (!status) 683 nfs_setsecurity(d_inode(dentry), entry->fattr, entry->label); 684 goto out; 685 } else { 686 d_invalidate(dentry); 687 dput(dentry); 688 dentry = NULL; 689 goto again; 690 } 691 } 692 if (!entry->fh->size) { 693 d_lookup_done(dentry); 694 goto out; 695 } 696 697 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr, entry->label); 698 alias = d_splice_alias(inode, dentry); 699 d_lookup_done(dentry); 700 if (alias) { 701 if (IS_ERR(alias)) 702 goto out; 703 dput(dentry); 704 dentry = alias; 705 } 706 nfs_set_verifier(dentry, dir_verifier); 707 out: 708 dput(dentry); 709 } 710 711 /* Perform conversion from xdr to cache array */ 712 static int nfs_readdir_page_filler(struct nfs_readdir_descriptor *desc, 713 struct nfs_entry *entry, 714 struct page **xdr_pages, 715 unsigned int buflen, 716 struct page **arrays, 717 size_t narrays) 718 { 719 struct address_space *mapping = desc->file->f_mapping; 720 struct xdr_stream stream; 721 struct xdr_buf buf; 722 struct page *scratch, *new, *page = *arrays; 723 int status; 724 725 scratch = alloc_page(GFP_KERNEL); 726 if (scratch == NULL) 727 return -ENOMEM; 728 729 xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen); 730 xdr_set_scratch_page(&stream, scratch); 731 732 do { 733 if (entry->label) 734 entry->label->len = NFS4_MAXLABELLEN; 735 736 status = xdr_decode(desc, entry, &stream); 737 if (status != 0) 738 break; 739 740 if (desc->plus) 741 nfs_prime_dcache(file_dentry(desc->file), entry, 742 desc->dir_verifier); 743 744 status = nfs_readdir_add_to_array(entry, page); 745 if (status != -ENOSPC) 746 continue; 747 748 if (page->mapping != mapping) { 749 if (!--narrays) 750 break; 751 new = nfs_readdir_page_array_alloc(entry->prev_cookie, 752 GFP_KERNEL); 753 if (!new) 754 break; 755 arrays++; 756 *arrays = page = new; 757 } else { 758 new = nfs_readdir_page_get_next(mapping, 759 page->index + 1, 760 entry->prev_cookie); 761 if (!new) 762 break; 763 if (page != *arrays) 764 nfs_readdir_page_unlock_and_put(page); 765 page = new; 766 } 767 status = nfs_readdir_add_to_array(entry, page); 768 } while (!status && !entry->eof); 769 770 switch (status) { 771 case -EBADCOOKIE: 772 if (entry->eof) { 773 nfs_readdir_page_set_eof(page); 774 status = 0; 775 } 776 break; 777 case -ENOSPC: 778 case -EAGAIN: 779 status = 0; 780 break; 781 } 782 783 if (page != *arrays) 784 nfs_readdir_page_unlock_and_put(page); 785 786 put_page(scratch); 787 return status; 788 } 789 790 static void nfs_readdir_free_pages(struct page **pages, size_t npages) 791 { 792 while (npages--) 793 put_page(pages[npages]); 794 kfree(pages); 795 } 796 797 /* 798 * nfs_readdir_alloc_pages() will allocate pages that must be freed with a call 799 * to nfs_readdir_free_pages() 800 */ 801 static struct page **nfs_readdir_alloc_pages(size_t npages) 802 { 803 struct page **pages; 804 size_t i; 805 806 pages = kmalloc_array(npages, sizeof(*pages), GFP_KERNEL); 807 if (!pages) 808 return NULL; 809 for (i = 0; i < npages; i++) { 810 struct page *page = alloc_page(GFP_KERNEL); 811 if (page == NULL) 812 goto out_freepages; 813 pages[i] = page; 814 } 815 return pages; 816 817 out_freepages: 818 nfs_readdir_free_pages(pages, i); 819 return NULL; 820 } 821 822 static int nfs_readdir_xdr_to_array(struct nfs_readdir_descriptor *desc, 823 __be32 *verf_arg, __be32 *verf_res, 824 struct page **arrays, size_t narrays) 825 { 826 struct page **pages; 827 struct page *page = *arrays; 828 struct nfs_entry *entry; 829 size_t array_size; 830 struct inode *inode = file_inode(desc->file); 831 size_t dtsize = NFS_SERVER(inode)->dtsize; 832 int status = -ENOMEM; 833 834 entry = kzalloc(sizeof(*entry), GFP_KERNEL); 835 if (!entry) 836 return -ENOMEM; 837 entry->cookie = nfs_readdir_page_last_cookie(page); 838 entry->fh = nfs_alloc_fhandle(); 839 entry->fattr = nfs_alloc_fattr(); 840 entry->server = NFS_SERVER(inode); 841 if (entry->fh == NULL || entry->fattr == NULL) 842 goto out; 843 844 entry->label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT); 845 if (IS_ERR(entry->label)) { 846 status = PTR_ERR(entry->label); 847 goto out; 848 } 849 850 array_size = (dtsize + PAGE_SIZE - 1) >> PAGE_SHIFT; 851 pages = nfs_readdir_alloc_pages(array_size); 852 if (!pages) 853 goto out_release_label; 854 855 do { 856 unsigned int pglen; 857 status = nfs_readdir_xdr_filler(desc, verf_arg, entry->cookie, 858 pages, dtsize, 859 verf_res); 860 if (status < 0) 861 break; 862 863 pglen = status; 864 if (pglen == 0) { 865 nfs_readdir_page_set_eof(page); 866 break; 867 } 868 869 verf_arg = verf_res; 870 871 status = nfs_readdir_page_filler(desc, entry, pages, pglen, 872 arrays, narrays); 873 } while (!status && nfs_readdir_page_needs_filling(page)); 874 875 nfs_readdir_free_pages(pages, array_size); 876 out_release_label: 877 nfs4_label_free(entry->label); 878 out: 879 nfs_free_fattr(entry->fattr); 880 nfs_free_fhandle(entry->fh); 881 kfree(entry); 882 return status; 883 } 884 885 static void nfs_readdir_page_put(struct nfs_readdir_descriptor *desc) 886 { 887 put_page(desc->page); 888 desc->page = NULL; 889 } 890 891 static void 892 nfs_readdir_page_unlock_and_put_cached(struct nfs_readdir_descriptor *desc) 893 { 894 unlock_page(desc->page); 895 nfs_readdir_page_put(desc); 896 } 897 898 static struct page * 899 nfs_readdir_page_get_cached(struct nfs_readdir_descriptor *desc) 900 { 901 return nfs_readdir_page_get_locked(desc->file->f_mapping, 902 desc->page_index, 903 desc->last_cookie); 904 } 905 906 /* 907 * Returns 0 if desc->dir_cookie was found on page desc->page_index 908 * and locks the page to prevent removal from the page cache. 909 */ 910 static int find_and_lock_cache_page(struct nfs_readdir_descriptor *desc) 911 { 912 struct inode *inode = file_inode(desc->file); 913 struct nfs_inode *nfsi = NFS_I(inode); 914 __be32 verf[NFS_DIR_VERIFIER_SIZE]; 915 int res; 916 917 desc->page = nfs_readdir_page_get_cached(desc); 918 if (!desc->page) 919 return -ENOMEM; 920 if (nfs_readdir_page_needs_filling(desc->page)) { 921 res = nfs_readdir_xdr_to_array(desc, nfsi->cookieverf, verf, 922 &desc->page, 1); 923 if (res < 0) { 924 nfs_readdir_page_unlock_and_put_cached(desc); 925 if (res == -EBADCOOKIE || res == -ENOTSYNC) { 926 invalidate_inode_pages2(desc->file->f_mapping); 927 desc->page_index = 0; 928 return -EAGAIN; 929 } 930 return res; 931 } 932 memcpy(nfsi->cookieverf, verf, sizeof(nfsi->cookieverf)); 933 } 934 res = nfs_readdir_search_array(desc); 935 if (res == 0) { 936 nfsi->page_index = desc->page_index; 937 return 0; 938 } 939 nfs_readdir_page_unlock_and_put_cached(desc); 940 return res; 941 } 942 943 static bool nfs_readdir_dont_search_cache(struct nfs_readdir_descriptor *desc) 944 { 945 struct address_space *mapping = desc->file->f_mapping; 946 struct inode *dir = file_inode(desc->file); 947 unsigned int dtsize = NFS_SERVER(dir)->dtsize; 948 loff_t size = i_size_read(dir); 949 950 /* 951 * Default to uncached readdir if the page cache is empty, and 952 * we're looking for a non-zero cookie in a large directory. 953 */ 954 return desc->dir_cookie != 0 && mapping->nrpages == 0 && size > dtsize; 955 } 956 957 /* Search for desc->dir_cookie from the beginning of the page cache */ 958 static int readdir_search_pagecache(struct nfs_readdir_descriptor *desc) 959 { 960 int res; 961 962 if (nfs_readdir_dont_search_cache(desc)) 963 return -EBADCOOKIE; 964 965 do { 966 if (desc->page_index == 0) { 967 desc->current_index = 0; 968 desc->prev_index = 0; 969 desc->last_cookie = 0; 970 } 971 res = find_and_lock_cache_page(desc); 972 } while (res == -EAGAIN); 973 return res; 974 } 975 976 /* 977 * Once we've found the start of the dirent within a page: fill 'er up... 978 */ 979 static void nfs_do_filldir(struct nfs_readdir_descriptor *desc) 980 { 981 struct file *file = desc->file; 982 struct nfs_inode *nfsi = NFS_I(file_inode(file)); 983 struct nfs_cache_array *array; 984 unsigned int i = 0; 985 986 array = kmap(desc->page); 987 for (i = desc->cache_entry_index; i < array->size; i++) { 988 struct nfs_cache_array_entry *ent; 989 990 ent = &array->array[i]; 991 if (!dir_emit(desc->ctx, ent->name, ent->name_len, 992 nfs_compat_user_ino64(ent->ino), ent->d_type)) { 993 desc->eof = true; 994 break; 995 } 996 memcpy(desc->verf, nfsi->cookieverf, sizeof(desc->verf)); 997 if (i < (array->size-1)) 998 desc->dir_cookie = array->array[i+1].cookie; 999 else 1000 desc->dir_cookie = array->last_cookie; 1001 if (nfs_readdir_use_cookie(file)) 1002 desc->ctx->pos = desc->dir_cookie; 1003 else 1004 desc->ctx->pos++; 1005 if (desc->duped != 0) 1006 desc->duped = 1; 1007 } 1008 if (array->page_is_eof) 1009 desc->eof = true; 1010 1011 kunmap(desc->page); 1012 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %llu\n", 1013 (unsigned long long)desc->dir_cookie); 1014 } 1015 1016 /* 1017 * If we cannot find a cookie in our cache, we suspect that this is 1018 * because it points to a deleted file, so we ask the server to return 1019 * whatever it thinks is the next entry. We then feed this to filldir. 1020 * If all goes well, we should then be able to find our way round the 1021 * cache on the next call to readdir_search_pagecache(); 1022 * 1023 * NOTE: we cannot add the anonymous page to the pagecache because 1024 * the data it contains might not be page aligned. Besides, 1025 * we should already have a complete representation of the 1026 * directory in the page cache by the time we get here. 1027 */ 1028 static int uncached_readdir(struct nfs_readdir_descriptor *desc) 1029 { 1030 struct page **arrays; 1031 size_t i, sz = 512; 1032 __be32 verf[NFS_DIR_VERIFIER_SIZE]; 1033 int status = -ENOMEM; 1034 1035 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %llu\n", 1036 (unsigned long long)desc->dir_cookie); 1037 1038 arrays = kcalloc(sz, sizeof(*arrays), GFP_KERNEL); 1039 if (!arrays) 1040 goto out; 1041 arrays[0] = nfs_readdir_page_array_alloc(desc->dir_cookie, GFP_KERNEL); 1042 if (!arrays[0]) 1043 goto out; 1044 1045 desc->page_index = 0; 1046 desc->last_cookie = desc->dir_cookie; 1047 desc->duped = 0; 1048 1049 status = nfs_readdir_xdr_to_array(desc, desc->verf, verf, arrays, sz); 1050 1051 for (i = 0; !desc->eof && i < sz && arrays[i]; i++) { 1052 desc->page = arrays[i]; 1053 nfs_do_filldir(desc); 1054 } 1055 desc->page = NULL; 1056 1057 1058 for (i = 0; i < sz && arrays[i]; i++) 1059 nfs_readdir_page_array_free(arrays[i]); 1060 out: 1061 kfree(arrays); 1062 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __func__, status); 1063 return status; 1064 } 1065 1066 /* The file offset position represents the dirent entry number. A 1067 last cookie cache takes care of the common case of reading the 1068 whole directory. 1069 */ 1070 static int nfs_readdir(struct file *file, struct dir_context *ctx) 1071 { 1072 struct dentry *dentry = file_dentry(file); 1073 struct inode *inode = d_inode(dentry); 1074 struct nfs_open_dir_context *dir_ctx = file->private_data; 1075 struct nfs_readdir_descriptor *desc; 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 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 do { 1111 res = readdir_search_pagecache(desc); 1112 1113 if (res == -EBADCOOKIE) { 1114 res = 0; 1115 /* This means either end of directory */ 1116 if (desc->dir_cookie && !desc->eof) { 1117 /* Or that the server has 'lost' a cookie */ 1118 res = uncached_readdir(desc); 1119 if (res == 0) 1120 continue; 1121 if (res == -EBADCOOKIE || res == -ENOTSYNC) 1122 res = 0; 1123 } 1124 break; 1125 } 1126 if (res == -ETOOSMALL && desc->plus) { 1127 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags); 1128 nfs_zap_caches(inode); 1129 desc->page_index = 0; 1130 desc->plus = false; 1131 desc->eof = false; 1132 continue; 1133 } 1134 if (res < 0) 1135 break; 1136 1137 nfs_do_filldir(desc); 1138 nfs_readdir_page_unlock_and_put_cached(desc); 1139 } while (!desc->eof); 1140 1141 spin_lock(&file->f_lock); 1142 dir_ctx->dir_cookie = desc->dir_cookie; 1143 dir_ctx->dup_cookie = desc->dup_cookie; 1144 dir_ctx->duped = desc->duped; 1145 dir_ctx->attr_gencount = desc->attr_gencount; 1146 memcpy(dir_ctx->verf, desc->verf, sizeof(dir_ctx->verf)); 1147 spin_unlock(&file->f_lock); 1148 1149 kfree(desc); 1150 1151 out: 1152 dfprintk(FILE, "NFS: readdir(%pD2) returns %d\n", file, res); 1153 return res; 1154 } 1155 1156 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int whence) 1157 { 1158 struct nfs_open_dir_context *dir_ctx = filp->private_data; 1159 1160 dfprintk(FILE, "NFS: llseek dir(%pD2, %lld, %d)\n", 1161 filp, offset, whence); 1162 1163 switch (whence) { 1164 default: 1165 return -EINVAL; 1166 case SEEK_SET: 1167 if (offset < 0) 1168 return -EINVAL; 1169 spin_lock(&filp->f_lock); 1170 break; 1171 case SEEK_CUR: 1172 if (offset == 0) 1173 return filp->f_pos; 1174 spin_lock(&filp->f_lock); 1175 offset += filp->f_pos; 1176 if (offset < 0) { 1177 spin_unlock(&filp->f_lock); 1178 return -EINVAL; 1179 } 1180 } 1181 if (offset != filp->f_pos) { 1182 filp->f_pos = offset; 1183 if (nfs_readdir_use_cookie(filp)) 1184 dir_ctx->dir_cookie = offset; 1185 else 1186 dir_ctx->dir_cookie = 0; 1187 if (offset == 0) 1188 memset(dir_ctx->verf, 0, sizeof(dir_ctx->verf)); 1189 dir_ctx->duped = 0; 1190 } 1191 spin_unlock(&filp->f_lock); 1192 return offset; 1193 } 1194 1195 /* 1196 * All directory operations under NFS are synchronous, so fsync() 1197 * is a dummy operation. 1198 */ 1199 static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end, 1200 int datasync) 1201 { 1202 dfprintk(FILE, "NFS: fsync dir(%pD2) datasync %d\n", filp, datasync); 1203 1204 nfs_inc_stats(file_inode(filp), NFSIOS_VFSFSYNC); 1205 return 0; 1206 } 1207 1208 /** 1209 * nfs_force_lookup_revalidate - Mark the directory as having changed 1210 * @dir: pointer to directory inode 1211 * 1212 * This forces the revalidation code in nfs_lookup_revalidate() to do a 1213 * full lookup on all child dentries of 'dir' whenever a change occurs 1214 * on the server that might have invalidated our dcache. 1215 * 1216 * Note that we reserve bit '0' as a tag to let us know when a dentry 1217 * was revalidated while holding a delegation on its inode. 1218 * 1219 * The caller should be holding dir->i_lock 1220 */ 1221 void nfs_force_lookup_revalidate(struct inode *dir) 1222 { 1223 NFS_I(dir)->cache_change_attribute += 2; 1224 } 1225 EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate); 1226 1227 /** 1228 * nfs_verify_change_attribute - Detects NFS remote directory changes 1229 * @dir: pointer to parent directory inode 1230 * @verf: previously saved change attribute 1231 * 1232 * Return "false" if the verifiers doesn't match the change attribute. 1233 * This would usually indicate that the directory contents have changed on 1234 * the server, and that any dentries need revalidating. 1235 */ 1236 static bool nfs_verify_change_attribute(struct inode *dir, unsigned long verf) 1237 { 1238 return (verf & ~1UL) == nfs_save_change_attribute(dir); 1239 } 1240 1241 static void nfs_set_verifier_delegated(unsigned long *verf) 1242 { 1243 *verf |= 1UL; 1244 } 1245 1246 #if IS_ENABLED(CONFIG_NFS_V4) 1247 static void nfs_unset_verifier_delegated(unsigned long *verf) 1248 { 1249 *verf &= ~1UL; 1250 } 1251 #endif /* IS_ENABLED(CONFIG_NFS_V4) */ 1252 1253 static bool nfs_test_verifier_delegated(unsigned long verf) 1254 { 1255 return verf & 1; 1256 } 1257 1258 static bool nfs_verifier_is_delegated(struct dentry *dentry) 1259 { 1260 return nfs_test_verifier_delegated(dentry->d_time); 1261 } 1262 1263 static void nfs_set_verifier_locked(struct dentry *dentry, unsigned long verf) 1264 { 1265 struct inode *inode = d_inode(dentry); 1266 1267 if (!nfs_verifier_is_delegated(dentry) && 1268 !nfs_verify_change_attribute(d_inode(dentry->d_parent), verf)) 1269 goto out; 1270 if (inode && NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) 1271 nfs_set_verifier_delegated(&verf); 1272 out: 1273 dentry->d_time = verf; 1274 } 1275 1276 /** 1277 * nfs_set_verifier - save a parent directory verifier in the dentry 1278 * @dentry: pointer to dentry 1279 * @verf: verifier to save 1280 * 1281 * Saves the parent directory verifier in @dentry. If the inode has 1282 * a delegation, we also tag the dentry as having been revalidated 1283 * while holding a delegation so that we know we don't have to 1284 * look it up again after a directory change. 1285 */ 1286 void nfs_set_verifier(struct dentry *dentry, unsigned long verf) 1287 { 1288 1289 spin_lock(&dentry->d_lock); 1290 nfs_set_verifier_locked(dentry, verf); 1291 spin_unlock(&dentry->d_lock); 1292 } 1293 EXPORT_SYMBOL_GPL(nfs_set_verifier); 1294 1295 #if IS_ENABLED(CONFIG_NFS_V4) 1296 /** 1297 * nfs_clear_verifier_delegated - clear the dir verifier delegation tag 1298 * @inode: pointer to inode 1299 * 1300 * Iterates through the dentries in the inode alias list and clears 1301 * the tag used to indicate that the dentry has been revalidated 1302 * while holding a delegation. 1303 * This function is intended for use when the delegation is being 1304 * returned or revoked. 1305 */ 1306 void nfs_clear_verifier_delegated(struct inode *inode) 1307 { 1308 struct dentry *alias; 1309 1310 if (!inode) 1311 return; 1312 spin_lock(&inode->i_lock); 1313 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) { 1314 spin_lock(&alias->d_lock); 1315 nfs_unset_verifier_delegated(&alias->d_time); 1316 spin_unlock(&alias->d_lock); 1317 } 1318 spin_unlock(&inode->i_lock); 1319 } 1320 EXPORT_SYMBOL_GPL(nfs_clear_verifier_delegated); 1321 #endif /* IS_ENABLED(CONFIG_NFS_V4) */ 1322 1323 /* 1324 * A check for whether or not the parent directory has changed. 1325 * In the case it has, we assume that the dentries are untrustworthy 1326 * and may need to be looked up again. 1327 * If rcu_walk prevents us from performing a full check, return 0. 1328 */ 1329 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry, 1330 int rcu_walk) 1331 { 1332 if (IS_ROOT(dentry)) 1333 return 1; 1334 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE) 1335 return 0; 1336 if (!nfs_verify_change_attribute(dir, dentry->d_time)) 1337 return 0; 1338 /* Revalidate nfsi->cache_change_attribute before we declare a match */ 1339 if (nfs_mapping_need_revalidate_inode(dir)) { 1340 if (rcu_walk) 1341 return 0; 1342 if (__nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0) 1343 return 0; 1344 } 1345 if (!nfs_verify_change_attribute(dir, dentry->d_time)) 1346 return 0; 1347 return 1; 1348 } 1349 1350 /* 1351 * Use intent information to check whether or not we're going to do 1352 * an O_EXCL create using this path component. 1353 */ 1354 static int nfs_is_exclusive_create(struct inode *dir, unsigned int flags) 1355 { 1356 if (NFS_PROTO(dir)->version == 2) 1357 return 0; 1358 return flags & LOOKUP_EXCL; 1359 } 1360 1361 /* 1362 * Inode and filehandle revalidation for lookups. 1363 * 1364 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL, 1365 * or if the intent information indicates that we're about to open this 1366 * particular file and the "nocto" mount flag is not set. 1367 * 1368 */ 1369 static 1370 int nfs_lookup_verify_inode(struct inode *inode, unsigned int flags) 1371 { 1372 struct nfs_server *server = NFS_SERVER(inode); 1373 int ret; 1374 1375 if (IS_AUTOMOUNT(inode)) 1376 return 0; 1377 1378 if (flags & LOOKUP_OPEN) { 1379 switch (inode->i_mode & S_IFMT) { 1380 case S_IFREG: 1381 /* A NFSv4 OPEN will revalidate later */ 1382 if (server->caps & NFS_CAP_ATOMIC_OPEN) 1383 goto out; 1384 fallthrough; 1385 case S_IFDIR: 1386 if (server->flags & NFS_MOUNT_NOCTO) 1387 break; 1388 /* NFS close-to-open cache consistency validation */ 1389 goto out_force; 1390 } 1391 } 1392 1393 /* VFS wants an on-the-wire revalidation */ 1394 if (flags & LOOKUP_REVAL) 1395 goto out_force; 1396 out: 1397 return (inode->i_nlink == 0) ? -ESTALE : 0; 1398 out_force: 1399 if (flags & LOOKUP_RCU) 1400 return -ECHILD; 1401 ret = __nfs_revalidate_inode(server, inode); 1402 if (ret != 0) 1403 return ret; 1404 goto out; 1405 } 1406 1407 static void nfs_mark_dir_for_revalidate(struct inode *inode) 1408 { 1409 spin_lock(&inode->i_lock); 1410 nfs_set_cache_invalid(inode, NFS_INO_REVAL_PAGECACHE); 1411 spin_unlock(&inode->i_lock); 1412 } 1413 1414 /* 1415 * We judge how long we want to trust negative 1416 * dentries by looking at the parent inode mtime. 1417 * 1418 * If parent mtime has changed, we revalidate, else we wait for a 1419 * period corresponding to the parent's attribute cache timeout value. 1420 * 1421 * If LOOKUP_RCU prevents us from performing a full check, return 1 1422 * suggesting a reval is needed. 1423 * 1424 * Note that when creating a new file, or looking up a rename target, 1425 * then it shouldn't be necessary to revalidate a negative dentry. 1426 */ 1427 static inline 1428 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry, 1429 unsigned int flags) 1430 { 1431 if (flags & (LOOKUP_CREATE | LOOKUP_RENAME_TARGET)) 1432 return 0; 1433 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG) 1434 return 1; 1435 return !nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU); 1436 } 1437 1438 static int 1439 nfs_lookup_revalidate_done(struct inode *dir, struct dentry *dentry, 1440 struct inode *inode, int error) 1441 { 1442 switch (error) { 1443 case 1: 1444 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is valid\n", 1445 __func__, dentry); 1446 return 1; 1447 case 0: 1448 /* 1449 * We can't d_drop the root of a disconnected tree: 1450 * its d_hash is on the s_anon list and d_drop() would hide 1451 * it from shrink_dcache_for_unmount(), leading to busy 1452 * inodes on unmount and further oopses. 1453 */ 1454 if (inode && IS_ROOT(dentry)) 1455 return 1; 1456 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is invalid\n", 1457 __func__, dentry); 1458 return 0; 1459 } 1460 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) lookup returned error %d\n", 1461 __func__, dentry, error); 1462 return error; 1463 } 1464 1465 static int 1466 nfs_lookup_revalidate_negative(struct inode *dir, struct dentry *dentry, 1467 unsigned int flags) 1468 { 1469 int ret = 1; 1470 if (nfs_neg_need_reval(dir, dentry, flags)) { 1471 if (flags & LOOKUP_RCU) 1472 return -ECHILD; 1473 ret = 0; 1474 } 1475 return nfs_lookup_revalidate_done(dir, dentry, NULL, ret); 1476 } 1477 1478 static int 1479 nfs_lookup_revalidate_delegated(struct inode *dir, struct dentry *dentry, 1480 struct inode *inode) 1481 { 1482 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1483 return nfs_lookup_revalidate_done(dir, dentry, inode, 1); 1484 } 1485 1486 static int 1487 nfs_lookup_revalidate_dentry(struct inode *dir, struct dentry *dentry, 1488 struct inode *inode) 1489 { 1490 struct nfs_fh *fhandle; 1491 struct nfs_fattr *fattr; 1492 struct nfs4_label *label; 1493 unsigned long dir_verifier; 1494 int ret; 1495 1496 ret = -ENOMEM; 1497 fhandle = nfs_alloc_fhandle(); 1498 fattr = nfs_alloc_fattr(); 1499 label = nfs4_label_alloc(NFS_SERVER(inode), GFP_KERNEL); 1500 if (fhandle == NULL || fattr == NULL || IS_ERR(label)) 1501 goto out; 1502 1503 dir_verifier = nfs_save_change_attribute(dir); 1504 ret = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr, label); 1505 if (ret < 0) { 1506 switch (ret) { 1507 case -ESTALE: 1508 case -ENOENT: 1509 ret = 0; 1510 break; 1511 case -ETIMEDOUT: 1512 if (NFS_SERVER(inode)->flags & NFS_MOUNT_SOFTREVAL) 1513 ret = 1; 1514 } 1515 goto out; 1516 } 1517 ret = 0; 1518 if (nfs_compare_fh(NFS_FH(inode), fhandle)) 1519 goto out; 1520 if (nfs_refresh_inode(inode, fattr) < 0) 1521 goto out; 1522 1523 nfs_setsecurity(inode, fattr, label); 1524 nfs_set_verifier(dentry, dir_verifier); 1525 1526 /* set a readdirplus hint that we had a cache miss */ 1527 nfs_force_use_readdirplus(dir); 1528 ret = 1; 1529 out: 1530 nfs_free_fattr(fattr); 1531 nfs_free_fhandle(fhandle); 1532 nfs4_label_free(label); 1533 1534 /* 1535 * If the lookup failed despite the dentry change attribute being 1536 * a match, then we should revalidate the directory cache. 1537 */ 1538 if (!ret && nfs_verify_change_attribute(dir, dentry->d_time)) 1539 nfs_mark_dir_for_revalidate(dir); 1540 return nfs_lookup_revalidate_done(dir, dentry, inode, ret); 1541 } 1542 1543 /* 1544 * This is called every time the dcache has a lookup hit, 1545 * and we should check whether we can really trust that 1546 * lookup. 1547 * 1548 * NOTE! The hit can be a negative hit too, don't assume 1549 * we have an inode! 1550 * 1551 * If the parent directory is seen to have changed, we throw out the 1552 * cached dentry and do a new lookup. 1553 */ 1554 static int 1555 nfs_do_lookup_revalidate(struct inode *dir, struct dentry *dentry, 1556 unsigned int flags) 1557 { 1558 struct inode *inode; 1559 int error; 1560 1561 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE); 1562 inode = d_inode(dentry); 1563 1564 if (!inode) 1565 return nfs_lookup_revalidate_negative(dir, dentry, flags); 1566 1567 if (is_bad_inode(inode)) { 1568 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n", 1569 __func__, dentry); 1570 goto out_bad; 1571 } 1572 1573 if (nfs_verifier_is_delegated(dentry)) 1574 return nfs_lookup_revalidate_delegated(dir, dentry, inode); 1575 1576 /* Force a full look up iff the parent directory has changed */ 1577 if (!(flags & (LOOKUP_EXCL | LOOKUP_REVAL)) && 1578 nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU)) { 1579 error = nfs_lookup_verify_inode(inode, flags); 1580 if (error) { 1581 if (error == -ESTALE) 1582 nfs_mark_dir_for_revalidate(dir); 1583 goto out_bad; 1584 } 1585 nfs_advise_use_readdirplus(dir); 1586 goto out_valid; 1587 } 1588 1589 if (flags & LOOKUP_RCU) 1590 return -ECHILD; 1591 1592 if (NFS_STALE(inode)) 1593 goto out_bad; 1594 1595 trace_nfs_lookup_revalidate_enter(dir, dentry, flags); 1596 error = nfs_lookup_revalidate_dentry(dir, dentry, inode); 1597 trace_nfs_lookup_revalidate_exit(dir, dentry, flags, error); 1598 return error; 1599 out_valid: 1600 return nfs_lookup_revalidate_done(dir, dentry, inode, 1); 1601 out_bad: 1602 if (flags & LOOKUP_RCU) 1603 return -ECHILD; 1604 return nfs_lookup_revalidate_done(dir, dentry, inode, 0); 1605 } 1606 1607 static int 1608 __nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags, 1609 int (*reval)(struct inode *, struct dentry *, unsigned int)) 1610 { 1611 struct dentry *parent; 1612 struct inode *dir; 1613 int ret; 1614 1615 if (flags & LOOKUP_RCU) { 1616 parent = READ_ONCE(dentry->d_parent); 1617 dir = d_inode_rcu(parent); 1618 if (!dir) 1619 return -ECHILD; 1620 ret = reval(dir, dentry, flags); 1621 if (parent != READ_ONCE(dentry->d_parent)) 1622 return -ECHILD; 1623 } else { 1624 parent = dget_parent(dentry); 1625 ret = reval(d_inode(parent), dentry, flags); 1626 dput(parent); 1627 } 1628 return ret; 1629 } 1630 1631 static int nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags) 1632 { 1633 return __nfs_lookup_revalidate(dentry, flags, nfs_do_lookup_revalidate); 1634 } 1635 1636 /* 1637 * A weaker form of d_revalidate for revalidating just the d_inode(dentry) 1638 * when we don't really care about the dentry name. This is called when a 1639 * pathwalk ends on a dentry that was not found via a normal lookup in the 1640 * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals). 1641 * 1642 * In this situation, we just want to verify that the inode itself is OK 1643 * since the dentry might have changed on the server. 1644 */ 1645 static int nfs_weak_revalidate(struct dentry *dentry, unsigned int flags) 1646 { 1647 struct inode *inode = d_inode(dentry); 1648 int error = 0; 1649 1650 /* 1651 * I believe we can only get a negative dentry here in the case of a 1652 * procfs-style symlink. Just assume it's correct for now, but we may 1653 * eventually need to do something more here. 1654 */ 1655 if (!inode) { 1656 dfprintk(LOOKUPCACHE, "%s: %pd2 has negative inode\n", 1657 __func__, dentry); 1658 return 1; 1659 } 1660 1661 if (is_bad_inode(inode)) { 1662 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n", 1663 __func__, dentry); 1664 return 0; 1665 } 1666 1667 error = nfs_lookup_verify_inode(inode, flags); 1668 dfprintk(LOOKUPCACHE, "NFS: %s: inode %lu is %s\n", 1669 __func__, inode->i_ino, error ? "invalid" : "valid"); 1670 return !error; 1671 } 1672 1673 /* 1674 * This is called from dput() when d_count is going to 0. 1675 */ 1676 static int nfs_dentry_delete(const struct dentry *dentry) 1677 { 1678 dfprintk(VFS, "NFS: dentry_delete(%pd2, %x)\n", 1679 dentry, dentry->d_flags); 1680 1681 /* Unhash any dentry with a stale inode */ 1682 if (d_really_is_positive(dentry) && NFS_STALE(d_inode(dentry))) 1683 return 1; 1684 1685 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { 1686 /* Unhash it, so that ->d_iput() would be called */ 1687 return 1; 1688 } 1689 if (!(dentry->d_sb->s_flags & SB_ACTIVE)) { 1690 /* Unhash it, so that ancestors of killed async unlink 1691 * files will be cleaned up during umount */ 1692 return 1; 1693 } 1694 return 0; 1695 1696 } 1697 1698 /* Ensure that we revalidate inode->i_nlink */ 1699 static void nfs_drop_nlink(struct inode *inode) 1700 { 1701 spin_lock(&inode->i_lock); 1702 /* drop the inode if we're reasonably sure this is the last link */ 1703 if (inode->i_nlink > 0) 1704 drop_nlink(inode); 1705 NFS_I(inode)->attr_gencount = nfs_inc_attr_generation_counter(); 1706 nfs_set_cache_invalid( 1707 inode, NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_CTIME | 1708 NFS_INO_INVALID_OTHER | NFS_INO_REVAL_FORCED); 1709 spin_unlock(&inode->i_lock); 1710 } 1711 1712 /* 1713 * Called when the dentry loses inode. 1714 * We use it to clean up silly-renamed files. 1715 */ 1716 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode) 1717 { 1718 if (S_ISDIR(inode->i_mode)) 1719 /* drop any readdir cache as it could easily be old */ 1720 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA); 1721 1722 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { 1723 nfs_complete_unlink(dentry, inode); 1724 nfs_drop_nlink(inode); 1725 } 1726 iput(inode); 1727 } 1728 1729 static void nfs_d_release(struct dentry *dentry) 1730 { 1731 /* free cached devname value, if it survived that far */ 1732 if (unlikely(dentry->d_fsdata)) { 1733 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) 1734 WARN_ON(1); 1735 else 1736 kfree(dentry->d_fsdata); 1737 } 1738 } 1739 1740 const struct dentry_operations nfs_dentry_operations = { 1741 .d_revalidate = nfs_lookup_revalidate, 1742 .d_weak_revalidate = nfs_weak_revalidate, 1743 .d_delete = nfs_dentry_delete, 1744 .d_iput = nfs_dentry_iput, 1745 .d_automount = nfs_d_automount, 1746 .d_release = nfs_d_release, 1747 }; 1748 EXPORT_SYMBOL_GPL(nfs_dentry_operations); 1749 1750 struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags) 1751 { 1752 struct dentry *res; 1753 struct inode *inode = NULL; 1754 struct nfs_fh *fhandle = NULL; 1755 struct nfs_fattr *fattr = NULL; 1756 struct nfs4_label *label = NULL; 1757 unsigned long dir_verifier; 1758 int error; 1759 1760 dfprintk(VFS, "NFS: lookup(%pd2)\n", dentry); 1761 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP); 1762 1763 if (unlikely(dentry->d_name.len > NFS_SERVER(dir)->namelen)) 1764 return ERR_PTR(-ENAMETOOLONG); 1765 1766 /* 1767 * If we're doing an exclusive create, optimize away the lookup 1768 * but don't hash the dentry. 1769 */ 1770 if (nfs_is_exclusive_create(dir, flags) || flags & LOOKUP_RENAME_TARGET) 1771 return NULL; 1772 1773 res = ERR_PTR(-ENOMEM); 1774 fhandle = nfs_alloc_fhandle(); 1775 fattr = nfs_alloc_fattr(); 1776 if (fhandle == NULL || fattr == NULL) 1777 goto out; 1778 1779 label = nfs4_label_alloc(NFS_SERVER(dir), GFP_NOWAIT); 1780 if (IS_ERR(label)) 1781 goto out; 1782 1783 dir_verifier = nfs_save_change_attribute(dir); 1784 trace_nfs_lookup_enter(dir, dentry, flags); 1785 error = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr, label); 1786 if (error == -ENOENT) 1787 goto no_entry; 1788 if (error < 0) { 1789 res = ERR_PTR(error); 1790 goto out_label; 1791 } 1792 inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label); 1793 res = ERR_CAST(inode); 1794 if (IS_ERR(res)) 1795 goto out_label; 1796 1797 /* Notify readdir to use READDIRPLUS */ 1798 nfs_force_use_readdirplus(dir); 1799 1800 no_entry: 1801 res = d_splice_alias(inode, dentry); 1802 if (res != NULL) { 1803 if (IS_ERR(res)) 1804 goto out_label; 1805 dentry = res; 1806 } 1807 nfs_set_verifier(dentry, dir_verifier); 1808 out_label: 1809 trace_nfs_lookup_exit(dir, dentry, flags, error); 1810 nfs4_label_free(label); 1811 out: 1812 nfs_free_fattr(fattr); 1813 nfs_free_fhandle(fhandle); 1814 return res; 1815 } 1816 EXPORT_SYMBOL_GPL(nfs_lookup); 1817 1818 #if IS_ENABLED(CONFIG_NFS_V4) 1819 static int nfs4_lookup_revalidate(struct dentry *, unsigned int); 1820 1821 const struct dentry_operations nfs4_dentry_operations = { 1822 .d_revalidate = nfs4_lookup_revalidate, 1823 .d_weak_revalidate = nfs_weak_revalidate, 1824 .d_delete = nfs_dentry_delete, 1825 .d_iput = nfs_dentry_iput, 1826 .d_automount = nfs_d_automount, 1827 .d_release = nfs_d_release, 1828 }; 1829 EXPORT_SYMBOL_GPL(nfs4_dentry_operations); 1830 1831 static fmode_t flags_to_mode(int flags) 1832 { 1833 fmode_t res = (__force fmode_t)flags & FMODE_EXEC; 1834 if ((flags & O_ACCMODE) != O_WRONLY) 1835 res |= FMODE_READ; 1836 if ((flags & O_ACCMODE) != O_RDONLY) 1837 res |= FMODE_WRITE; 1838 return res; 1839 } 1840 1841 static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags, struct file *filp) 1842 { 1843 return alloc_nfs_open_context(dentry, flags_to_mode(open_flags), filp); 1844 } 1845 1846 static int do_open(struct inode *inode, struct file *filp) 1847 { 1848 nfs_fscache_open_file(inode, filp); 1849 return 0; 1850 } 1851 1852 static int nfs_finish_open(struct nfs_open_context *ctx, 1853 struct dentry *dentry, 1854 struct file *file, unsigned open_flags) 1855 { 1856 int err; 1857 1858 err = finish_open(file, dentry, do_open); 1859 if (err) 1860 goto out; 1861 if (S_ISREG(file->f_path.dentry->d_inode->i_mode)) 1862 nfs_file_set_open_context(file, ctx); 1863 else 1864 err = -EOPENSTALE; 1865 out: 1866 return err; 1867 } 1868 1869 int nfs_atomic_open(struct inode *dir, struct dentry *dentry, 1870 struct file *file, unsigned open_flags, 1871 umode_t mode) 1872 { 1873 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 1874 struct nfs_open_context *ctx; 1875 struct dentry *res; 1876 struct iattr attr = { .ia_valid = ATTR_OPEN }; 1877 struct inode *inode; 1878 unsigned int lookup_flags = 0; 1879 bool switched = false; 1880 int created = 0; 1881 int err; 1882 1883 /* Expect a negative dentry */ 1884 BUG_ON(d_inode(dentry)); 1885 1886 dfprintk(VFS, "NFS: atomic_open(%s/%lu), %pd\n", 1887 dir->i_sb->s_id, dir->i_ino, dentry); 1888 1889 err = nfs_check_flags(open_flags); 1890 if (err) 1891 return err; 1892 1893 /* NFS only supports OPEN on regular files */ 1894 if ((open_flags & O_DIRECTORY)) { 1895 if (!d_in_lookup(dentry)) { 1896 /* 1897 * Hashed negative dentry with O_DIRECTORY: dentry was 1898 * revalidated and is fine, no need to perform lookup 1899 * again 1900 */ 1901 return -ENOENT; 1902 } 1903 lookup_flags = LOOKUP_OPEN|LOOKUP_DIRECTORY; 1904 goto no_open; 1905 } 1906 1907 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) 1908 return -ENAMETOOLONG; 1909 1910 if (open_flags & O_CREAT) { 1911 struct nfs_server *server = NFS_SERVER(dir); 1912 1913 if (!(server->attr_bitmask[2] & FATTR4_WORD2_MODE_UMASK)) 1914 mode &= ~current_umask(); 1915 1916 attr.ia_valid |= ATTR_MODE; 1917 attr.ia_mode = mode; 1918 } 1919 if (open_flags & O_TRUNC) { 1920 attr.ia_valid |= ATTR_SIZE; 1921 attr.ia_size = 0; 1922 } 1923 1924 if (!(open_flags & O_CREAT) && !d_in_lookup(dentry)) { 1925 d_drop(dentry); 1926 switched = true; 1927 dentry = d_alloc_parallel(dentry->d_parent, 1928 &dentry->d_name, &wq); 1929 if (IS_ERR(dentry)) 1930 return PTR_ERR(dentry); 1931 if (unlikely(!d_in_lookup(dentry))) 1932 return finish_no_open(file, dentry); 1933 } 1934 1935 ctx = create_nfs_open_context(dentry, open_flags, file); 1936 err = PTR_ERR(ctx); 1937 if (IS_ERR(ctx)) 1938 goto out; 1939 1940 trace_nfs_atomic_open_enter(dir, ctx, open_flags); 1941 inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr, &created); 1942 if (created) 1943 file->f_mode |= FMODE_CREATED; 1944 if (IS_ERR(inode)) { 1945 err = PTR_ERR(inode); 1946 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err); 1947 put_nfs_open_context(ctx); 1948 d_drop(dentry); 1949 switch (err) { 1950 case -ENOENT: 1951 d_splice_alias(NULL, dentry); 1952 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1953 break; 1954 case -EISDIR: 1955 case -ENOTDIR: 1956 goto no_open; 1957 case -ELOOP: 1958 if (!(open_flags & O_NOFOLLOW)) 1959 goto no_open; 1960 break; 1961 /* case -EINVAL: */ 1962 default: 1963 break; 1964 } 1965 goto out; 1966 } 1967 1968 err = nfs_finish_open(ctx, ctx->dentry, file, open_flags); 1969 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err); 1970 put_nfs_open_context(ctx); 1971 out: 1972 if (unlikely(switched)) { 1973 d_lookup_done(dentry); 1974 dput(dentry); 1975 } 1976 return err; 1977 1978 no_open: 1979 res = nfs_lookup(dir, dentry, lookup_flags); 1980 if (switched) { 1981 d_lookup_done(dentry); 1982 if (!res) 1983 res = dentry; 1984 else 1985 dput(dentry); 1986 } 1987 if (IS_ERR(res)) 1988 return PTR_ERR(res); 1989 return finish_no_open(file, res); 1990 } 1991 EXPORT_SYMBOL_GPL(nfs_atomic_open); 1992 1993 static int 1994 nfs4_do_lookup_revalidate(struct inode *dir, struct dentry *dentry, 1995 unsigned int flags) 1996 { 1997 struct inode *inode; 1998 1999 if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY)) 2000 goto full_reval; 2001 if (d_mountpoint(dentry)) 2002 goto full_reval; 2003 2004 inode = d_inode(dentry); 2005 2006 /* We can't create new files in nfs_open_revalidate(), so we 2007 * optimize away revalidation of negative dentries. 2008 */ 2009 if (inode == NULL) 2010 goto full_reval; 2011 2012 if (nfs_verifier_is_delegated(dentry)) 2013 return nfs_lookup_revalidate_delegated(dir, dentry, inode); 2014 2015 /* NFS only supports OPEN on regular files */ 2016 if (!S_ISREG(inode->i_mode)) 2017 goto full_reval; 2018 2019 /* We cannot do exclusive creation on a positive dentry */ 2020 if (flags & (LOOKUP_EXCL | LOOKUP_REVAL)) 2021 goto reval_dentry; 2022 2023 /* Check if the directory changed */ 2024 if (!nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU)) 2025 goto reval_dentry; 2026 2027 /* Let f_op->open() actually open (and revalidate) the file */ 2028 return 1; 2029 reval_dentry: 2030 if (flags & LOOKUP_RCU) 2031 return -ECHILD; 2032 return nfs_lookup_revalidate_dentry(dir, dentry, inode); 2033 2034 full_reval: 2035 return nfs_do_lookup_revalidate(dir, dentry, flags); 2036 } 2037 2038 static int nfs4_lookup_revalidate(struct dentry *dentry, unsigned int flags) 2039 { 2040 return __nfs_lookup_revalidate(dentry, flags, 2041 nfs4_do_lookup_revalidate); 2042 } 2043 2044 #endif /* CONFIG_NFSV4 */ 2045 2046 struct dentry * 2047 nfs_add_or_obtain(struct dentry *dentry, struct nfs_fh *fhandle, 2048 struct nfs_fattr *fattr, 2049 struct nfs4_label *label) 2050 { 2051 struct dentry *parent = dget_parent(dentry); 2052 struct inode *dir = d_inode(parent); 2053 struct inode *inode; 2054 struct dentry *d; 2055 int error; 2056 2057 d_drop(dentry); 2058 2059 if (fhandle->size == 0) { 2060 error = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr, NULL); 2061 if (error) 2062 goto out_error; 2063 } 2064 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 2065 if (!(fattr->valid & NFS_ATTR_FATTR)) { 2066 struct nfs_server *server = NFS_SB(dentry->d_sb); 2067 error = server->nfs_client->rpc_ops->getattr(server, fhandle, 2068 fattr, NULL, NULL); 2069 if (error < 0) 2070 goto out_error; 2071 } 2072 inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label); 2073 d = d_splice_alias(inode, dentry); 2074 out: 2075 dput(parent); 2076 return d; 2077 out_error: 2078 d = ERR_PTR(error); 2079 goto out; 2080 } 2081 EXPORT_SYMBOL_GPL(nfs_add_or_obtain); 2082 2083 /* 2084 * Code common to create, mkdir, and mknod. 2085 */ 2086 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle, 2087 struct nfs_fattr *fattr, 2088 struct nfs4_label *label) 2089 { 2090 struct dentry *d; 2091 2092 d = nfs_add_or_obtain(dentry, fhandle, fattr, label); 2093 if (IS_ERR(d)) 2094 return PTR_ERR(d); 2095 2096 /* Callers don't care */ 2097 dput(d); 2098 return 0; 2099 } 2100 EXPORT_SYMBOL_GPL(nfs_instantiate); 2101 2102 /* 2103 * Following a failed create operation, we drop the dentry rather 2104 * than retain a negative dentry. This avoids a problem in the event 2105 * that the operation succeeded on the server, but an error in the 2106 * reply path made it appear to have failed. 2107 */ 2108 int nfs_create(struct user_namespace *mnt_userns, struct inode *dir, 2109 struct dentry *dentry, umode_t mode, bool excl) 2110 { 2111 struct iattr attr; 2112 int open_flags = excl ? O_CREAT | O_EXCL : O_CREAT; 2113 int error; 2114 2115 dfprintk(VFS, "NFS: create(%s/%lu), %pd\n", 2116 dir->i_sb->s_id, dir->i_ino, dentry); 2117 2118 attr.ia_mode = mode; 2119 attr.ia_valid = ATTR_MODE; 2120 2121 trace_nfs_create_enter(dir, dentry, open_flags); 2122 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags); 2123 trace_nfs_create_exit(dir, dentry, open_flags, error); 2124 if (error != 0) 2125 goto out_err; 2126 return 0; 2127 out_err: 2128 d_drop(dentry); 2129 return error; 2130 } 2131 EXPORT_SYMBOL_GPL(nfs_create); 2132 2133 /* 2134 * See comments for nfs_proc_create regarding failed operations. 2135 */ 2136 int 2137 nfs_mknod(struct user_namespace *mnt_userns, struct inode *dir, 2138 struct dentry *dentry, umode_t mode, dev_t rdev) 2139 { 2140 struct iattr attr; 2141 int status; 2142 2143 dfprintk(VFS, "NFS: mknod(%s/%lu), %pd\n", 2144 dir->i_sb->s_id, dir->i_ino, dentry); 2145 2146 attr.ia_mode = mode; 2147 attr.ia_valid = ATTR_MODE; 2148 2149 trace_nfs_mknod_enter(dir, dentry); 2150 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev); 2151 trace_nfs_mknod_exit(dir, dentry, status); 2152 if (status != 0) 2153 goto out_err; 2154 return 0; 2155 out_err: 2156 d_drop(dentry); 2157 return status; 2158 } 2159 EXPORT_SYMBOL_GPL(nfs_mknod); 2160 2161 /* 2162 * See comments for nfs_proc_create regarding failed operations. 2163 */ 2164 int nfs_mkdir(struct user_namespace *mnt_userns, struct inode *dir, 2165 struct dentry *dentry, umode_t mode) 2166 { 2167 struct iattr attr; 2168 int error; 2169 2170 dfprintk(VFS, "NFS: mkdir(%s/%lu), %pd\n", 2171 dir->i_sb->s_id, dir->i_ino, dentry); 2172 2173 attr.ia_valid = ATTR_MODE; 2174 attr.ia_mode = mode | S_IFDIR; 2175 2176 trace_nfs_mkdir_enter(dir, dentry); 2177 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr); 2178 trace_nfs_mkdir_exit(dir, dentry, error); 2179 if (error != 0) 2180 goto out_err; 2181 return 0; 2182 out_err: 2183 d_drop(dentry); 2184 return error; 2185 } 2186 EXPORT_SYMBOL_GPL(nfs_mkdir); 2187 2188 static void nfs_dentry_handle_enoent(struct dentry *dentry) 2189 { 2190 if (simple_positive(dentry)) 2191 d_delete(dentry); 2192 } 2193 2194 int nfs_rmdir(struct inode *dir, struct dentry *dentry) 2195 { 2196 int error; 2197 2198 dfprintk(VFS, "NFS: rmdir(%s/%lu), %pd\n", 2199 dir->i_sb->s_id, dir->i_ino, dentry); 2200 2201 trace_nfs_rmdir_enter(dir, dentry); 2202 if (d_really_is_positive(dentry)) { 2203 down_write(&NFS_I(d_inode(dentry))->rmdir_sem); 2204 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name); 2205 /* Ensure the VFS deletes this inode */ 2206 switch (error) { 2207 case 0: 2208 clear_nlink(d_inode(dentry)); 2209 break; 2210 case -ENOENT: 2211 nfs_dentry_handle_enoent(dentry); 2212 } 2213 up_write(&NFS_I(d_inode(dentry))->rmdir_sem); 2214 } else 2215 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name); 2216 trace_nfs_rmdir_exit(dir, dentry, error); 2217 2218 return error; 2219 } 2220 EXPORT_SYMBOL_GPL(nfs_rmdir); 2221 2222 /* 2223 * Remove a file after making sure there are no pending writes, 2224 * and after checking that the file has only one user. 2225 * 2226 * We invalidate the attribute cache and free the inode prior to the operation 2227 * to avoid possible races if the server reuses the inode. 2228 */ 2229 static int nfs_safe_remove(struct dentry *dentry) 2230 { 2231 struct inode *dir = d_inode(dentry->d_parent); 2232 struct inode *inode = d_inode(dentry); 2233 int error = -EBUSY; 2234 2235 dfprintk(VFS, "NFS: safe_remove(%pd2)\n", dentry); 2236 2237 /* If the dentry was sillyrenamed, we simply call d_delete() */ 2238 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { 2239 error = 0; 2240 goto out; 2241 } 2242 2243 trace_nfs_remove_enter(dir, dentry); 2244 if (inode != NULL) { 2245 error = NFS_PROTO(dir)->remove(dir, dentry); 2246 if (error == 0) 2247 nfs_drop_nlink(inode); 2248 } else 2249 error = NFS_PROTO(dir)->remove(dir, dentry); 2250 if (error == -ENOENT) 2251 nfs_dentry_handle_enoent(dentry); 2252 trace_nfs_remove_exit(dir, dentry, error); 2253 out: 2254 return error; 2255 } 2256 2257 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode 2258 * belongs to an active ".nfs..." file and we return -EBUSY. 2259 * 2260 * If sillyrename() returns 0, we do nothing, otherwise we unlink. 2261 */ 2262 int nfs_unlink(struct inode *dir, struct dentry *dentry) 2263 { 2264 int error; 2265 int need_rehash = 0; 2266 2267 dfprintk(VFS, "NFS: unlink(%s/%lu, %pd)\n", dir->i_sb->s_id, 2268 dir->i_ino, dentry); 2269 2270 trace_nfs_unlink_enter(dir, dentry); 2271 spin_lock(&dentry->d_lock); 2272 if (d_count(dentry) > 1) { 2273 spin_unlock(&dentry->d_lock); 2274 /* Start asynchronous writeout of the inode */ 2275 write_inode_now(d_inode(dentry), 0); 2276 error = nfs_sillyrename(dir, dentry); 2277 goto out; 2278 } 2279 if (!d_unhashed(dentry)) { 2280 __d_drop(dentry); 2281 need_rehash = 1; 2282 } 2283 spin_unlock(&dentry->d_lock); 2284 error = nfs_safe_remove(dentry); 2285 if (!error || error == -ENOENT) { 2286 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 2287 } else if (need_rehash) 2288 d_rehash(dentry); 2289 out: 2290 trace_nfs_unlink_exit(dir, dentry, error); 2291 return error; 2292 } 2293 EXPORT_SYMBOL_GPL(nfs_unlink); 2294 2295 /* 2296 * To create a symbolic link, most file systems instantiate a new inode, 2297 * add a page to it containing the path, then write it out to the disk 2298 * using prepare_write/commit_write. 2299 * 2300 * Unfortunately the NFS client can't create the in-core inode first 2301 * because it needs a file handle to create an in-core inode (see 2302 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the 2303 * symlink request has completed on the server. 2304 * 2305 * So instead we allocate a raw page, copy the symname into it, then do 2306 * the SYMLINK request with the page as the buffer. If it succeeds, we 2307 * now have a new file handle and can instantiate an in-core NFS inode 2308 * and move the raw page into its mapping. 2309 */ 2310 int nfs_symlink(struct user_namespace *mnt_userns, struct inode *dir, 2311 struct dentry *dentry, const char *symname) 2312 { 2313 struct page *page; 2314 char *kaddr; 2315 struct iattr attr; 2316 unsigned int pathlen = strlen(symname); 2317 int error; 2318 2319 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s)\n", dir->i_sb->s_id, 2320 dir->i_ino, dentry, symname); 2321 2322 if (pathlen > PAGE_SIZE) 2323 return -ENAMETOOLONG; 2324 2325 attr.ia_mode = S_IFLNK | S_IRWXUGO; 2326 attr.ia_valid = ATTR_MODE; 2327 2328 page = alloc_page(GFP_USER); 2329 if (!page) 2330 return -ENOMEM; 2331 2332 kaddr = page_address(page); 2333 memcpy(kaddr, symname, pathlen); 2334 if (pathlen < PAGE_SIZE) 2335 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen); 2336 2337 trace_nfs_symlink_enter(dir, dentry); 2338 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr); 2339 trace_nfs_symlink_exit(dir, dentry, error); 2340 if (error != 0) { 2341 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s) error %d\n", 2342 dir->i_sb->s_id, dir->i_ino, 2343 dentry, symname, error); 2344 d_drop(dentry); 2345 __free_page(page); 2346 return error; 2347 } 2348 2349 /* 2350 * No big deal if we can't add this page to the page cache here. 2351 * READLINK will get the missing page from the server if needed. 2352 */ 2353 if (!add_to_page_cache_lru(page, d_inode(dentry)->i_mapping, 0, 2354 GFP_KERNEL)) { 2355 SetPageUptodate(page); 2356 unlock_page(page); 2357 /* 2358 * add_to_page_cache_lru() grabs an extra page refcount. 2359 * Drop it here to avoid leaking this page later. 2360 */ 2361 put_page(page); 2362 } else 2363 __free_page(page); 2364 2365 return 0; 2366 } 2367 EXPORT_SYMBOL_GPL(nfs_symlink); 2368 2369 int 2370 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) 2371 { 2372 struct inode *inode = d_inode(old_dentry); 2373 int error; 2374 2375 dfprintk(VFS, "NFS: link(%pd2 -> %pd2)\n", 2376 old_dentry, dentry); 2377 2378 trace_nfs_link_enter(inode, dir, dentry); 2379 d_drop(dentry); 2380 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name); 2381 if (error == 0) { 2382 ihold(inode); 2383 d_add(dentry, inode); 2384 } 2385 trace_nfs_link_exit(inode, dir, dentry, error); 2386 return error; 2387 } 2388 EXPORT_SYMBOL_GPL(nfs_link); 2389 2390 /* 2391 * RENAME 2392 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a 2393 * different file handle for the same inode after a rename (e.g. when 2394 * moving to a different directory). A fail-safe method to do so would 2395 * be to look up old_dir/old_name, create a link to new_dir/new_name and 2396 * rename the old file using the sillyrename stuff. This way, the original 2397 * file in old_dir will go away when the last process iput()s the inode. 2398 * 2399 * FIXED. 2400 * 2401 * It actually works quite well. One needs to have the possibility for 2402 * at least one ".nfs..." file in each directory the file ever gets 2403 * moved or linked to which happens automagically with the new 2404 * implementation that only depends on the dcache stuff instead of 2405 * using the inode layer 2406 * 2407 * Unfortunately, things are a little more complicated than indicated 2408 * above. For a cross-directory move, we want to make sure we can get 2409 * rid of the old inode after the operation. This means there must be 2410 * no pending writes (if it's a file), and the use count must be 1. 2411 * If these conditions are met, we can drop the dentries before doing 2412 * the rename. 2413 */ 2414 int nfs_rename(struct user_namespace *mnt_userns, struct inode *old_dir, 2415 struct dentry *old_dentry, struct inode *new_dir, 2416 struct dentry *new_dentry, unsigned int flags) 2417 { 2418 struct inode *old_inode = d_inode(old_dentry); 2419 struct inode *new_inode = d_inode(new_dentry); 2420 struct dentry *dentry = NULL, *rehash = NULL; 2421 struct rpc_task *task; 2422 int error = -EBUSY; 2423 2424 if (flags) 2425 return -EINVAL; 2426 2427 dfprintk(VFS, "NFS: rename(%pd2 -> %pd2, ct=%d)\n", 2428 old_dentry, new_dentry, 2429 d_count(new_dentry)); 2430 2431 trace_nfs_rename_enter(old_dir, old_dentry, new_dir, new_dentry); 2432 /* 2433 * For non-directories, check whether the target is busy and if so, 2434 * make a copy of the dentry and then do a silly-rename. If the 2435 * silly-rename succeeds, the copied dentry is hashed and becomes 2436 * the new target. 2437 */ 2438 if (new_inode && !S_ISDIR(new_inode->i_mode)) { 2439 /* 2440 * To prevent any new references to the target during the 2441 * rename, we unhash the dentry in advance. 2442 */ 2443 if (!d_unhashed(new_dentry)) { 2444 d_drop(new_dentry); 2445 rehash = new_dentry; 2446 } 2447 2448 if (d_count(new_dentry) > 2) { 2449 int err; 2450 2451 /* copy the target dentry's name */ 2452 dentry = d_alloc(new_dentry->d_parent, 2453 &new_dentry->d_name); 2454 if (!dentry) 2455 goto out; 2456 2457 /* silly-rename the existing target ... */ 2458 err = nfs_sillyrename(new_dir, new_dentry); 2459 if (err) 2460 goto out; 2461 2462 new_dentry = dentry; 2463 rehash = NULL; 2464 new_inode = NULL; 2465 } 2466 } 2467 2468 task = nfs_async_rename(old_dir, new_dir, old_dentry, new_dentry, NULL); 2469 if (IS_ERR(task)) { 2470 error = PTR_ERR(task); 2471 goto out; 2472 } 2473 2474 error = rpc_wait_for_completion_task(task); 2475 if (error != 0) { 2476 ((struct nfs_renamedata *)task->tk_calldata)->cancelled = 1; 2477 /* Paired with the atomic_dec_and_test() barrier in rpc_do_put_task() */ 2478 smp_wmb(); 2479 } else 2480 error = task->tk_status; 2481 rpc_put_task(task); 2482 /* Ensure the inode attributes are revalidated */ 2483 if (error == 0) { 2484 spin_lock(&old_inode->i_lock); 2485 NFS_I(old_inode)->attr_gencount = nfs_inc_attr_generation_counter(); 2486 nfs_set_cache_invalid(old_inode, NFS_INO_INVALID_CHANGE | 2487 NFS_INO_INVALID_CTIME | 2488 NFS_INO_REVAL_FORCED); 2489 spin_unlock(&old_inode->i_lock); 2490 } 2491 out: 2492 if (rehash) 2493 d_rehash(rehash); 2494 trace_nfs_rename_exit(old_dir, old_dentry, 2495 new_dir, new_dentry, error); 2496 if (!error) { 2497 if (new_inode != NULL) 2498 nfs_drop_nlink(new_inode); 2499 /* 2500 * The d_move() should be here instead of in an async RPC completion 2501 * handler because we need the proper locks to move the dentry. If 2502 * we're interrupted by a signal, the async RPC completion handler 2503 * should mark the directories for revalidation. 2504 */ 2505 d_move(old_dentry, new_dentry); 2506 nfs_set_verifier(old_dentry, 2507 nfs_save_change_attribute(new_dir)); 2508 } else if (error == -ENOENT) 2509 nfs_dentry_handle_enoent(old_dentry); 2510 2511 /* new dentry created? */ 2512 if (dentry) 2513 dput(dentry); 2514 return error; 2515 } 2516 EXPORT_SYMBOL_GPL(nfs_rename); 2517 2518 static DEFINE_SPINLOCK(nfs_access_lru_lock); 2519 static LIST_HEAD(nfs_access_lru_list); 2520 static atomic_long_t nfs_access_nr_entries; 2521 2522 static unsigned long nfs_access_max_cachesize = 4*1024*1024; 2523 module_param(nfs_access_max_cachesize, ulong, 0644); 2524 MODULE_PARM_DESC(nfs_access_max_cachesize, "NFS access maximum total cache length"); 2525 2526 static void nfs_access_free_entry(struct nfs_access_entry *entry) 2527 { 2528 put_cred(entry->cred); 2529 kfree_rcu(entry, rcu_head); 2530 smp_mb__before_atomic(); 2531 atomic_long_dec(&nfs_access_nr_entries); 2532 smp_mb__after_atomic(); 2533 } 2534 2535 static void nfs_access_free_list(struct list_head *head) 2536 { 2537 struct nfs_access_entry *cache; 2538 2539 while (!list_empty(head)) { 2540 cache = list_entry(head->next, struct nfs_access_entry, lru); 2541 list_del(&cache->lru); 2542 nfs_access_free_entry(cache); 2543 } 2544 } 2545 2546 static unsigned long 2547 nfs_do_access_cache_scan(unsigned int nr_to_scan) 2548 { 2549 LIST_HEAD(head); 2550 struct nfs_inode *nfsi, *next; 2551 struct nfs_access_entry *cache; 2552 long freed = 0; 2553 2554 spin_lock(&nfs_access_lru_lock); 2555 list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) { 2556 struct inode *inode; 2557 2558 if (nr_to_scan-- == 0) 2559 break; 2560 inode = &nfsi->vfs_inode; 2561 spin_lock(&inode->i_lock); 2562 if (list_empty(&nfsi->access_cache_entry_lru)) 2563 goto remove_lru_entry; 2564 cache = list_entry(nfsi->access_cache_entry_lru.next, 2565 struct nfs_access_entry, lru); 2566 list_move(&cache->lru, &head); 2567 rb_erase(&cache->rb_node, &nfsi->access_cache); 2568 freed++; 2569 if (!list_empty(&nfsi->access_cache_entry_lru)) 2570 list_move_tail(&nfsi->access_cache_inode_lru, 2571 &nfs_access_lru_list); 2572 else { 2573 remove_lru_entry: 2574 list_del_init(&nfsi->access_cache_inode_lru); 2575 smp_mb__before_atomic(); 2576 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags); 2577 smp_mb__after_atomic(); 2578 } 2579 spin_unlock(&inode->i_lock); 2580 } 2581 spin_unlock(&nfs_access_lru_lock); 2582 nfs_access_free_list(&head); 2583 return freed; 2584 } 2585 2586 unsigned long 2587 nfs_access_cache_scan(struct shrinker *shrink, struct shrink_control *sc) 2588 { 2589 int nr_to_scan = sc->nr_to_scan; 2590 gfp_t gfp_mask = sc->gfp_mask; 2591 2592 if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL) 2593 return SHRINK_STOP; 2594 return nfs_do_access_cache_scan(nr_to_scan); 2595 } 2596 2597 2598 unsigned long 2599 nfs_access_cache_count(struct shrinker *shrink, struct shrink_control *sc) 2600 { 2601 return vfs_pressure_ratio(atomic_long_read(&nfs_access_nr_entries)); 2602 } 2603 2604 static void 2605 nfs_access_cache_enforce_limit(void) 2606 { 2607 long nr_entries = atomic_long_read(&nfs_access_nr_entries); 2608 unsigned long diff; 2609 unsigned int nr_to_scan; 2610 2611 if (nr_entries < 0 || nr_entries <= nfs_access_max_cachesize) 2612 return; 2613 nr_to_scan = 100; 2614 diff = nr_entries - nfs_access_max_cachesize; 2615 if (diff < nr_to_scan) 2616 nr_to_scan = diff; 2617 nfs_do_access_cache_scan(nr_to_scan); 2618 } 2619 2620 static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head) 2621 { 2622 struct rb_root *root_node = &nfsi->access_cache; 2623 struct rb_node *n; 2624 struct nfs_access_entry *entry; 2625 2626 /* Unhook entries from the cache */ 2627 while ((n = rb_first(root_node)) != NULL) { 2628 entry = rb_entry(n, struct nfs_access_entry, rb_node); 2629 rb_erase(n, root_node); 2630 list_move(&entry->lru, head); 2631 } 2632 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS; 2633 } 2634 2635 void nfs_access_zap_cache(struct inode *inode) 2636 { 2637 LIST_HEAD(head); 2638 2639 if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0) 2640 return; 2641 /* Remove from global LRU init */ 2642 spin_lock(&nfs_access_lru_lock); 2643 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) 2644 list_del_init(&NFS_I(inode)->access_cache_inode_lru); 2645 2646 spin_lock(&inode->i_lock); 2647 __nfs_access_zap_cache(NFS_I(inode), &head); 2648 spin_unlock(&inode->i_lock); 2649 spin_unlock(&nfs_access_lru_lock); 2650 nfs_access_free_list(&head); 2651 } 2652 EXPORT_SYMBOL_GPL(nfs_access_zap_cache); 2653 2654 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, const struct cred *cred) 2655 { 2656 struct rb_node *n = NFS_I(inode)->access_cache.rb_node; 2657 2658 while (n != NULL) { 2659 struct nfs_access_entry *entry = 2660 rb_entry(n, struct nfs_access_entry, rb_node); 2661 int cmp = cred_fscmp(cred, entry->cred); 2662 2663 if (cmp < 0) 2664 n = n->rb_left; 2665 else if (cmp > 0) 2666 n = n->rb_right; 2667 else 2668 return entry; 2669 } 2670 return NULL; 2671 } 2672 2673 static int nfs_access_get_cached_locked(struct inode *inode, const struct cred *cred, struct nfs_access_entry *res, bool may_block) 2674 { 2675 struct nfs_inode *nfsi = NFS_I(inode); 2676 struct nfs_access_entry *cache; 2677 bool retry = true; 2678 int err; 2679 2680 spin_lock(&inode->i_lock); 2681 for(;;) { 2682 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS) 2683 goto out_zap; 2684 cache = nfs_access_search_rbtree(inode, cred); 2685 err = -ENOENT; 2686 if (cache == NULL) 2687 goto out; 2688 /* Found an entry, is our attribute cache valid? */ 2689 if (!nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS)) 2690 break; 2691 if (!retry) 2692 break; 2693 err = -ECHILD; 2694 if (!may_block) 2695 goto out; 2696 spin_unlock(&inode->i_lock); 2697 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode); 2698 if (err) 2699 return err; 2700 spin_lock(&inode->i_lock); 2701 retry = false; 2702 } 2703 res->cred = cache->cred; 2704 res->mask = cache->mask; 2705 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru); 2706 err = 0; 2707 out: 2708 spin_unlock(&inode->i_lock); 2709 return err; 2710 out_zap: 2711 spin_unlock(&inode->i_lock); 2712 nfs_access_zap_cache(inode); 2713 return -ENOENT; 2714 } 2715 2716 static int nfs_access_get_cached_rcu(struct inode *inode, const struct cred *cred, struct nfs_access_entry *res) 2717 { 2718 /* Only check the most recently returned cache entry, 2719 * but do it without locking. 2720 */ 2721 struct nfs_inode *nfsi = NFS_I(inode); 2722 struct nfs_access_entry *cache; 2723 int err = -ECHILD; 2724 struct list_head *lh; 2725 2726 rcu_read_lock(); 2727 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS) 2728 goto out; 2729 lh = rcu_dereference(list_tail_rcu(&nfsi->access_cache_entry_lru)); 2730 cache = list_entry(lh, struct nfs_access_entry, lru); 2731 if (lh == &nfsi->access_cache_entry_lru || 2732 cred_fscmp(cred, cache->cred) != 0) 2733 cache = NULL; 2734 if (cache == NULL) 2735 goto out; 2736 if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS)) 2737 goto out; 2738 res->cred = cache->cred; 2739 res->mask = cache->mask; 2740 err = 0; 2741 out: 2742 rcu_read_unlock(); 2743 return err; 2744 } 2745 2746 int nfs_access_get_cached(struct inode *inode, const struct cred *cred, struct 2747 nfs_access_entry *res, bool may_block) 2748 { 2749 int status; 2750 2751 status = nfs_access_get_cached_rcu(inode, cred, res); 2752 if (status != 0) 2753 status = nfs_access_get_cached_locked(inode, cred, res, 2754 may_block); 2755 2756 return status; 2757 } 2758 EXPORT_SYMBOL_GPL(nfs_access_get_cached); 2759 2760 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set) 2761 { 2762 struct nfs_inode *nfsi = NFS_I(inode); 2763 struct rb_root *root_node = &nfsi->access_cache; 2764 struct rb_node **p = &root_node->rb_node; 2765 struct rb_node *parent = NULL; 2766 struct nfs_access_entry *entry; 2767 int cmp; 2768 2769 spin_lock(&inode->i_lock); 2770 while (*p != NULL) { 2771 parent = *p; 2772 entry = rb_entry(parent, struct nfs_access_entry, rb_node); 2773 cmp = cred_fscmp(set->cred, entry->cred); 2774 2775 if (cmp < 0) 2776 p = &parent->rb_left; 2777 else if (cmp > 0) 2778 p = &parent->rb_right; 2779 else 2780 goto found; 2781 } 2782 rb_link_node(&set->rb_node, parent, p); 2783 rb_insert_color(&set->rb_node, root_node); 2784 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); 2785 spin_unlock(&inode->i_lock); 2786 return; 2787 found: 2788 rb_replace_node(parent, &set->rb_node, root_node); 2789 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); 2790 list_del(&entry->lru); 2791 spin_unlock(&inode->i_lock); 2792 nfs_access_free_entry(entry); 2793 } 2794 2795 void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set) 2796 { 2797 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL); 2798 if (cache == NULL) 2799 return; 2800 RB_CLEAR_NODE(&cache->rb_node); 2801 cache->cred = get_cred(set->cred); 2802 cache->mask = set->mask; 2803 2804 /* The above field assignments must be visible 2805 * before this item appears on the lru. We cannot easily 2806 * use rcu_assign_pointer, so just force the memory barrier. 2807 */ 2808 smp_wmb(); 2809 nfs_access_add_rbtree(inode, cache); 2810 2811 /* Update accounting */ 2812 smp_mb__before_atomic(); 2813 atomic_long_inc(&nfs_access_nr_entries); 2814 smp_mb__after_atomic(); 2815 2816 /* Add inode to global LRU list */ 2817 if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) { 2818 spin_lock(&nfs_access_lru_lock); 2819 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) 2820 list_add_tail(&NFS_I(inode)->access_cache_inode_lru, 2821 &nfs_access_lru_list); 2822 spin_unlock(&nfs_access_lru_lock); 2823 } 2824 nfs_access_cache_enforce_limit(); 2825 } 2826 EXPORT_SYMBOL_GPL(nfs_access_add_cache); 2827 2828 #define NFS_MAY_READ (NFS_ACCESS_READ) 2829 #define NFS_MAY_WRITE (NFS_ACCESS_MODIFY | \ 2830 NFS_ACCESS_EXTEND | \ 2831 NFS_ACCESS_DELETE) 2832 #define NFS_FILE_MAY_WRITE (NFS_ACCESS_MODIFY | \ 2833 NFS_ACCESS_EXTEND) 2834 #define NFS_DIR_MAY_WRITE NFS_MAY_WRITE 2835 #define NFS_MAY_LOOKUP (NFS_ACCESS_LOOKUP) 2836 #define NFS_MAY_EXECUTE (NFS_ACCESS_EXECUTE) 2837 static int 2838 nfs_access_calc_mask(u32 access_result, umode_t umode) 2839 { 2840 int mask = 0; 2841 2842 if (access_result & NFS_MAY_READ) 2843 mask |= MAY_READ; 2844 if (S_ISDIR(umode)) { 2845 if ((access_result & NFS_DIR_MAY_WRITE) == NFS_DIR_MAY_WRITE) 2846 mask |= MAY_WRITE; 2847 if ((access_result & NFS_MAY_LOOKUP) == NFS_MAY_LOOKUP) 2848 mask |= MAY_EXEC; 2849 } else if (S_ISREG(umode)) { 2850 if ((access_result & NFS_FILE_MAY_WRITE) == NFS_FILE_MAY_WRITE) 2851 mask |= MAY_WRITE; 2852 if ((access_result & NFS_MAY_EXECUTE) == NFS_MAY_EXECUTE) 2853 mask |= MAY_EXEC; 2854 } else if (access_result & NFS_MAY_WRITE) 2855 mask |= MAY_WRITE; 2856 return mask; 2857 } 2858 2859 void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result) 2860 { 2861 entry->mask = access_result; 2862 } 2863 EXPORT_SYMBOL_GPL(nfs_access_set_mask); 2864 2865 static int nfs_do_access(struct inode *inode, const struct cred *cred, int mask) 2866 { 2867 struct nfs_access_entry cache; 2868 bool may_block = (mask & MAY_NOT_BLOCK) == 0; 2869 int cache_mask = -1; 2870 int status; 2871 2872 trace_nfs_access_enter(inode); 2873 2874 status = nfs_access_get_cached(inode, cred, &cache, may_block); 2875 if (status == 0) 2876 goto out_cached; 2877 2878 status = -ECHILD; 2879 if (!may_block) 2880 goto out; 2881 2882 /* 2883 * Determine which access bits we want to ask for... 2884 */ 2885 cache.mask = NFS_ACCESS_READ | NFS_ACCESS_MODIFY | NFS_ACCESS_EXTEND; 2886 if (nfs_server_capable(inode, NFS_CAP_XATTR)) { 2887 cache.mask |= NFS_ACCESS_XAREAD | NFS_ACCESS_XAWRITE | 2888 NFS_ACCESS_XALIST; 2889 } 2890 if (S_ISDIR(inode->i_mode)) 2891 cache.mask |= NFS_ACCESS_DELETE | NFS_ACCESS_LOOKUP; 2892 else 2893 cache.mask |= NFS_ACCESS_EXECUTE; 2894 cache.cred = cred; 2895 status = NFS_PROTO(inode)->access(inode, &cache); 2896 if (status != 0) { 2897 if (status == -ESTALE) { 2898 if (!S_ISDIR(inode->i_mode)) 2899 nfs_set_inode_stale(inode); 2900 else 2901 nfs_zap_caches(inode); 2902 } 2903 goto out; 2904 } 2905 nfs_access_add_cache(inode, &cache); 2906 out_cached: 2907 cache_mask = nfs_access_calc_mask(cache.mask, inode->i_mode); 2908 if ((mask & ~cache_mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) != 0) 2909 status = -EACCES; 2910 out: 2911 trace_nfs_access_exit(inode, mask, cache_mask, status); 2912 return status; 2913 } 2914 2915 static int nfs_open_permission_mask(int openflags) 2916 { 2917 int mask = 0; 2918 2919 if (openflags & __FMODE_EXEC) { 2920 /* ONLY check exec rights */ 2921 mask = MAY_EXEC; 2922 } else { 2923 if ((openflags & O_ACCMODE) != O_WRONLY) 2924 mask |= MAY_READ; 2925 if ((openflags & O_ACCMODE) != O_RDONLY) 2926 mask |= MAY_WRITE; 2927 } 2928 2929 return mask; 2930 } 2931 2932 int nfs_may_open(struct inode *inode, const struct cred *cred, int openflags) 2933 { 2934 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags)); 2935 } 2936 EXPORT_SYMBOL_GPL(nfs_may_open); 2937 2938 static int nfs_execute_ok(struct inode *inode, int mask) 2939 { 2940 struct nfs_server *server = NFS_SERVER(inode); 2941 int ret = 0; 2942 2943 if (S_ISDIR(inode->i_mode)) 2944 return 0; 2945 if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_OTHER)) { 2946 if (mask & MAY_NOT_BLOCK) 2947 return -ECHILD; 2948 ret = __nfs_revalidate_inode(server, inode); 2949 } 2950 if (ret == 0 && !execute_ok(inode)) 2951 ret = -EACCES; 2952 return ret; 2953 } 2954 2955 int nfs_permission(struct user_namespace *mnt_userns, 2956 struct inode *inode, 2957 int mask) 2958 { 2959 const struct cred *cred = current_cred(); 2960 int res = 0; 2961 2962 nfs_inc_stats(inode, NFSIOS_VFSACCESS); 2963 2964 if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) 2965 goto out; 2966 /* Is this sys_access() ? */ 2967 if (mask & (MAY_ACCESS | MAY_CHDIR)) 2968 goto force_lookup; 2969 2970 switch (inode->i_mode & S_IFMT) { 2971 case S_IFLNK: 2972 goto out; 2973 case S_IFREG: 2974 if ((mask & MAY_OPEN) && 2975 nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN)) 2976 return 0; 2977 break; 2978 case S_IFDIR: 2979 /* 2980 * Optimize away all write operations, since the server 2981 * will check permissions when we perform the op. 2982 */ 2983 if ((mask & MAY_WRITE) && !(mask & MAY_READ)) 2984 goto out; 2985 } 2986 2987 force_lookup: 2988 if (!NFS_PROTO(inode)->access) 2989 goto out_notsup; 2990 2991 res = nfs_do_access(inode, cred, mask); 2992 out: 2993 if (!res && (mask & MAY_EXEC)) 2994 res = nfs_execute_ok(inode, mask); 2995 2996 dfprintk(VFS, "NFS: permission(%s/%lu), mask=0x%x, res=%d\n", 2997 inode->i_sb->s_id, inode->i_ino, mask, res); 2998 return res; 2999 out_notsup: 3000 if (mask & MAY_NOT_BLOCK) 3001 return -ECHILD; 3002 3003 res = nfs_revalidate_inode(NFS_SERVER(inode), inode); 3004 if (res == 0) 3005 res = generic_permission(&init_user_ns, inode, mask); 3006 goto out; 3007 } 3008 EXPORT_SYMBOL_GPL(nfs_permission); 3009 3010 /* 3011 * Local variables: 3012 * version-control: t 3013 * kept-new-versions: 5 3014 * End: 3015 */ 3016