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 /* 933 * Set the cookie verifier if the page cache was empty 934 */ 935 if (desc->page_index == 0) 936 memcpy(nfsi->cookieverf, verf, 937 sizeof(nfsi->cookieverf)); 938 } 939 res = nfs_readdir_search_array(desc); 940 if (res == 0) { 941 nfsi->page_index = desc->page_index; 942 return 0; 943 } 944 nfs_readdir_page_unlock_and_put_cached(desc); 945 return res; 946 } 947 948 static bool nfs_readdir_dont_search_cache(struct nfs_readdir_descriptor *desc) 949 { 950 struct address_space *mapping = desc->file->f_mapping; 951 struct inode *dir = file_inode(desc->file); 952 unsigned int dtsize = NFS_SERVER(dir)->dtsize; 953 loff_t size = i_size_read(dir); 954 955 /* 956 * Default to uncached readdir if the page cache is empty, and 957 * we're looking for a non-zero cookie in a large directory. 958 */ 959 return desc->dir_cookie != 0 && mapping->nrpages == 0 && size > dtsize; 960 } 961 962 /* Search for desc->dir_cookie from the beginning of the page cache */ 963 static int readdir_search_pagecache(struct nfs_readdir_descriptor *desc) 964 { 965 int res; 966 967 if (nfs_readdir_dont_search_cache(desc)) 968 return -EBADCOOKIE; 969 970 do { 971 if (desc->page_index == 0) { 972 desc->current_index = 0; 973 desc->prev_index = 0; 974 desc->last_cookie = 0; 975 } 976 res = find_and_lock_cache_page(desc); 977 } while (res == -EAGAIN); 978 return res; 979 } 980 981 /* 982 * Once we've found the start of the dirent within a page: fill 'er up... 983 */ 984 static void nfs_do_filldir(struct nfs_readdir_descriptor *desc, 985 const __be32 *verf) 986 { 987 struct file *file = desc->file; 988 struct nfs_cache_array *array; 989 unsigned int i = 0; 990 991 array = kmap(desc->page); 992 for (i = desc->cache_entry_index; i < array->size; i++) { 993 struct nfs_cache_array_entry *ent; 994 995 ent = &array->array[i]; 996 if (!dir_emit(desc->ctx, ent->name, ent->name_len, 997 nfs_compat_user_ino64(ent->ino), ent->d_type)) { 998 desc->eof = true; 999 break; 1000 } 1001 memcpy(desc->verf, verf, sizeof(desc->verf)); 1002 if (i < (array->size-1)) 1003 desc->dir_cookie = array->array[i+1].cookie; 1004 else 1005 desc->dir_cookie = array->last_cookie; 1006 if (nfs_readdir_use_cookie(file)) 1007 desc->ctx->pos = desc->dir_cookie; 1008 else 1009 desc->ctx->pos++; 1010 if (desc->duped != 0) 1011 desc->duped = 1; 1012 } 1013 if (array->page_is_eof) 1014 desc->eof = true; 1015 1016 kunmap(desc->page); 1017 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %llu\n", 1018 (unsigned long long)desc->dir_cookie); 1019 } 1020 1021 /* 1022 * If we cannot find a cookie in our cache, we suspect that this is 1023 * because it points to a deleted file, so we ask the server to return 1024 * whatever it thinks is the next entry. We then feed this to filldir. 1025 * If all goes well, we should then be able to find our way round the 1026 * cache on the next call to readdir_search_pagecache(); 1027 * 1028 * NOTE: we cannot add the anonymous page to the pagecache because 1029 * the data it contains might not be page aligned. Besides, 1030 * we should already have a complete representation of the 1031 * directory in the page cache by the time we get here. 1032 */ 1033 static int uncached_readdir(struct nfs_readdir_descriptor *desc) 1034 { 1035 struct page **arrays; 1036 size_t i, sz = 512; 1037 __be32 verf[NFS_DIR_VERIFIER_SIZE]; 1038 int status = -ENOMEM; 1039 1040 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %llu\n", 1041 (unsigned long long)desc->dir_cookie); 1042 1043 arrays = kcalloc(sz, sizeof(*arrays), GFP_KERNEL); 1044 if (!arrays) 1045 goto out; 1046 arrays[0] = nfs_readdir_page_array_alloc(desc->dir_cookie, GFP_KERNEL); 1047 if (!arrays[0]) 1048 goto out; 1049 1050 desc->page_index = 0; 1051 desc->last_cookie = desc->dir_cookie; 1052 desc->duped = 0; 1053 1054 status = nfs_readdir_xdr_to_array(desc, desc->verf, verf, arrays, sz); 1055 1056 for (i = 0; !desc->eof && i < sz && arrays[i]; i++) { 1057 desc->page = arrays[i]; 1058 nfs_do_filldir(desc, verf); 1059 } 1060 desc->page = NULL; 1061 1062 1063 for (i = 0; i < sz && arrays[i]; i++) 1064 nfs_readdir_page_array_free(arrays[i]); 1065 out: 1066 kfree(arrays); 1067 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __func__, status); 1068 return status; 1069 } 1070 1071 /* The file offset position represents the dirent entry number. A 1072 last cookie cache takes care of the common case of reading the 1073 whole directory. 1074 */ 1075 static int nfs_readdir(struct file *file, struct dir_context *ctx) 1076 { 1077 struct dentry *dentry = file_dentry(file); 1078 struct inode *inode = d_inode(dentry); 1079 struct nfs_inode *nfsi = NFS_I(inode); 1080 struct nfs_open_dir_context *dir_ctx = file->private_data; 1081 struct nfs_readdir_descriptor *desc; 1082 int res; 1083 1084 dfprintk(FILE, "NFS: readdir(%pD2) starting at cookie %llu\n", 1085 file, (long long)ctx->pos); 1086 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS); 1087 1088 /* 1089 * ctx->pos points to the dirent entry number. 1090 * *desc->dir_cookie has the cookie for the next entry. We have 1091 * to either find the entry with the appropriate number or 1092 * revalidate the cookie. 1093 */ 1094 if (ctx->pos == 0 || nfs_attribute_cache_expired(inode)) { 1095 res = nfs_revalidate_mapping(inode, file->f_mapping); 1096 if (res < 0) 1097 goto out; 1098 } 1099 1100 res = -ENOMEM; 1101 desc = kzalloc(sizeof(*desc), GFP_KERNEL); 1102 if (!desc) 1103 goto out; 1104 desc->file = file; 1105 desc->ctx = ctx; 1106 desc->plus = nfs_use_readdirplus(inode, ctx); 1107 1108 spin_lock(&file->f_lock); 1109 desc->dir_cookie = dir_ctx->dir_cookie; 1110 desc->dup_cookie = dir_ctx->dup_cookie; 1111 desc->duped = dir_ctx->duped; 1112 desc->attr_gencount = dir_ctx->attr_gencount; 1113 memcpy(desc->verf, dir_ctx->verf, sizeof(desc->verf)); 1114 spin_unlock(&file->f_lock); 1115 1116 do { 1117 res = readdir_search_pagecache(desc); 1118 1119 if (res == -EBADCOOKIE) { 1120 res = 0; 1121 /* This means either end of directory */ 1122 if (desc->dir_cookie && !desc->eof) { 1123 /* Or that the server has 'lost' a cookie */ 1124 res = uncached_readdir(desc); 1125 if (res == 0) 1126 continue; 1127 if (res == -EBADCOOKIE || res == -ENOTSYNC) 1128 res = 0; 1129 } 1130 break; 1131 } 1132 if (res == -ETOOSMALL && desc->plus) { 1133 clear_bit(NFS_INO_ADVISE_RDPLUS, &nfsi->flags); 1134 nfs_zap_caches(inode); 1135 desc->page_index = 0; 1136 desc->plus = false; 1137 desc->eof = false; 1138 continue; 1139 } 1140 if (res < 0) 1141 break; 1142 1143 nfs_do_filldir(desc, nfsi->cookieverf); 1144 nfs_readdir_page_unlock_and_put_cached(desc); 1145 } while (!desc->eof); 1146 1147 spin_lock(&file->f_lock); 1148 dir_ctx->dir_cookie = desc->dir_cookie; 1149 dir_ctx->dup_cookie = desc->dup_cookie; 1150 dir_ctx->duped = desc->duped; 1151 dir_ctx->attr_gencount = desc->attr_gencount; 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 struct nfs4_label *label; 1498 unsigned long dir_verifier; 1499 int ret; 1500 1501 ret = -ENOMEM; 1502 fhandle = nfs_alloc_fhandle(); 1503 fattr = nfs_alloc_fattr(); 1504 label = nfs4_label_alloc(NFS_SERVER(inode), GFP_KERNEL); 1505 if (fhandle == NULL || fattr == NULL || IS_ERR(label)) 1506 goto out; 1507 1508 dir_verifier = nfs_save_change_attribute(dir); 1509 ret = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr, label); 1510 if (ret < 0) { 1511 switch (ret) { 1512 case -ESTALE: 1513 case -ENOENT: 1514 ret = 0; 1515 break; 1516 case -ETIMEDOUT: 1517 if (NFS_SERVER(inode)->flags & NFS_MOUNT_SOFTREVAL) 1518 ret = 1; 1519 } 1520 goto out; 1521 } 1522 ret = 0; 1523 if (nfs_compare_fh(NFS_FH(inode), fhandle)) 1524 goto out; 1525 if (nfs_refresh_inode(inode, fattr) < 0) 1526 goto out; 1527 1528 nfs_setsecurity(inode, fattr, label); 1529 nfs_set_verifier(dentry, dir_verifier); 1530 1531 /* set a readdirplus hint that we had a cache miss */ 1532 nfs_force_use_readdirplus(dir); 1533 ret = 1; 1534 out: 1535 nfs_free_fattr(fattr); 1536 nfs_free_fhandle(fhandle); 1537 nfs4_label_free(label); 1538 1539 /* 1540 * If the lookup failed despite the dentry change attribute being 1541 * a match, then we should revalidate the directory cache. 1542 */ 1543 if (!ret && nfs_verify_change_attribute(dir, dentry->d_time)) 1544 nfs_mark_dir_for_revalidate(dir); 1545 return nfs_lookup_revalidate_done(dir, dentry, inode, ret); 1546 } 1547 1548 /* 1549 * This is called every time the dcache has a lookup hit, 1550 * and we should check whether we can really trust that 1551 * lookup. 1552 * 1553 * NOTE! The hit can be a negative hit too, don't assume 1554 * we have an inode! 1555 * 1556 * If the parent directory is seen to have changed, we throw out the 1557 * cached dentry and do a new lookup. 1558 */ 1559 static int 1560 nfs_do_lookup_revalidate(struct inode *dir, struct dentry *dentry, 1561 unsigned int flags) 1562 { 1563 struct inode *inode; 1564 int error; 1565 1566 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE); 1567 inode = d_inode(dentry); 1568 1569 if (!inode) 1570 return nfs_lookup_revalidate_negative(dir, dentry, flags); 1571 1572 if (is_bad_inode(inode)) { 1573 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n", 1574 __func__, dentry); 1575 goto out_bad; 1576 } 1577 1578 if (nfs_verifier_is_delegated(dentry)) 1579 return nfs_lookup_revalidate_delegated(dir, dentry, inode); 1580 1581 /* Force a full look up iff the parent directory has changed */ 1582 if (!(flags & (LOOKUP_EXCL | LOOKUP_REVAL)) && 1583 nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU)) { 1584 error = nfs_lookup_verify_inode(inode, flags); 1585 if (error) { 1586 if (error == -ESTALE) 1587 nfs_mark_dir_for_revalidate(dir); 1588 goto out_bad; 1589 } 1590 nfs_advise_use_readdirplus(dir); 1591 goto out_valid; 1592 } 1593 1594 if (flags & LOOKUP_RCU) 1595 return -ECHILD; 1596 1597 if (NFS_STALE(inode)) 1598 goto out_bad; 1599 1600 trace_nfs_lookup_revalidate_enter(dir, dentry, flags); 1601 error = nfs_lookup_revalidate_dentry(dir, dentry, inode); 1602 trace_nfs_lookup_revalidate_exit(dir, dentry, flags, error); 1603 return error; 1604 out_valid: 1605 return nfs_lookup_revalidate_done(dir, dentry, inode, 1); 1606 out_bad: 1607 if (flags & LOOKUP_RCU) 1608 return -ECHILD; 1609 return nfs_lookup_revalidate_done(dir, dentry, inode, 0); 1610 } 1611 1612 static int 1613 __nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags, 1614 int (*reval)(struct inode *, struct dentry *, unsigned int)) 1615 { 1616 struct dentry *parent; 1617 struct inode *dir; 1618 int ret; 1619 1620 if (flags & LOOKUP_RCU) { 1621 parent = READ_ONCE(dentry->d_parent); 1622 dir = d_inode_rcu(parent); 1623 if (!dir) 1624 return -ECHILD; 1625 ret = reval(dir, dentry, flags); 1626 if (parent != READ_ONCE(dentry->d_parent)) 1627 return -ECHILD; 1628 } else { 1629 parent = dget_parent(dentry); 1630 ret = reval(d_inode(parent), dentry, flags); 1631 dput(parent); 1632 } 1633 return ret; 1634 } 1635 1636 static int nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags) 1637 { 1638 return __nfs_lookup_revalidate(dentry, flags, nfs_do_lookup_revalidate); 1639 } 1640 1641 /* 1642 * A weaker form of d_revalidate for revalidating just the d_inode(dentry) 1643 * when we don't really care about the dentry name. This is called when a 1644 * pathwalk ends on a dentry that was not found via a normal lookup in the 1645 * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals). 1646 * 1647 * In this situation, we just want to verify that the inode itself is OK 1648 * since the dentry might have changed on the server. 1649 */ 1650 static int nfs_weak_revalidate(struct dentry *dentry, unsigned int flags) 1651 { 1652 struct inode *inode = d_inode(dentry); 1653 int error = 0; 1654 1655 /* 1656 * I believe we can only get a negative dentry here in the case of a 1657 * procfs-style symlink. Just assume it's correct for now, but we may 1658 * eventually need to do something more here. 1659 */ 1660 if (!inode) { 1661 dfprintk(LOOKUPCACHE, "%s: %pd2 has negative inode\n", 1662 __func__, dentry); 1663 return 1; 1664 } 1665 1666 if (is_bad_inode(inode)) { 1667 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n", 1668 __func__, dentry); 1669 return 0; 1670 } 1671 1672 error = nfs_lookup_verify_inode(inode, flags); 1673 dfprintk(LOOKUPCACHE, "NFS: %s: inode %lu is %s\n", 1674 __func__, inode->i_ino, error ? "invalid" : "valid"); 1675 return !error; 1676 } 1677 1678 /* 1679 * This is called from dput() when d_count is going to 0. 1680 */ 1681 static int nfs_dentry_delete(const struct dentry *dentry) 1682 { 1683 dfprintk(VFS, "NFS: dentry_delete(%pd2, %x)\n", 1684 dentry, dentry->d_flags); 1685 1686 /* Unhash any dentry with a stale inode */ 1687 if (d_really_is_positive(dentry) && NFS_STALE(d_inode(dentry))) 1688 return 1; 1689 1690 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { 1691 /* Unhash it, so that ->d_iput() would be called */ 1692 return 1; 1693 } 1694 if (!(dentry->d_sb->s_flags & SB_ACTIVE)) { 1695 /* Unhash it, so that ancestors of killed async unlink 1696 * files will be cleaned up during umount */ 1697 return 1; 1698 } 1699 return 0; 1700 1701 } 1702 1703 /* Ensure that we revalidate inode->i_nlink */ 1704 static void nfs_drop_nlink(struct inode *inode) 1705 { 1706 spin_lock(&inode->i_lock); 1707 /* drop the inode if we're reasonably sure this is the last link */ 1708 if (inode->i_nlink > 0) 1709 drop_nlink(inode); 1710 NFS_I(inode)->attr_gencount = nfs_inc_attr_generation_counter(); 1711 nfs_set_cache_invalid( 1712 inode, NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_CTIME | 1713 NFS_INO_INVALID_NLINK); 1714 spin_unlock(&inode->i_lock); 1715 } 1716 1717 /* 1718 * Called when the dentry loses inode. 1719 * We use it to clean up silly-renamed files. 1720 */ 1721 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode) 1722 { 1723 if (S_ISDIR(inode->i_mode)) 1724 /* drop any readdir cache as it could easily be old */ 1725 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA); 1726 1727 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { 1728 nfs_complete_unlink(dentry, inode); 1729 nfs_drop_nlink(inode); 1730 } 1731 iput(inode); 1732 } 1733 1734 static void nfs_d_release(struct dentry *dentry) 1735 { 1736 /* free cached devname value, if it survived that far */ 1737 if (unlikely(dentry->d_fsdata)) { 1738 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) 1739 WARN_ON(1); 1740 else 1741 kfree(dentry->d_fsdata); 1742 } 1743 } 1744 1745 const struct dentry_operations nfs_dentry_operations = { 1746 .d_revalidate = nfs_lookup_revalidate, 1747 .d_weak_revalidate = nfs_weak_revalidate, 1748 .d_delete = nfs_dentry_delete, 1749 .d_iput = nfs_dentry_iput, 1750 .d_automount = nfs_d_automount, 1751 .d_release = nfs_d_release, 1752 }; 1753 EXPORT_SYMBOL_GPL(nfs_dentry_operations); 1754 1755 struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags) 1756 { 1757 struct dentry *res; 1758 struct inode *inode = NULL; 1759 struct nfs_fh *fhandle = NULL; 1760 struct nfs_fattr *fattr = NULL; 1761 struct nfs4_label *label = NULL; 1762 unsigned long dir_verifier; 1763 int error; 1764 1765 dfprintk(VFS, "NFS: lookup(%pd2)\n", dentry); 1766 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP); 1767 1768 if (unlikely(dentry->d_name.len > NFS_SERVER(dir)->namelen)) 1769 return ERR_PTR(-ENAMETOOLONG); 1770 1771 /* 1772 * If we're doing an exclusive create, optimize away the lookup 1773 * but don't hash the dentry. 1774 */ 1775 if (nfs_is_exclusive_create(dir, flags) || flags & LOOKUP_RENAME_TARGET) 1776 return NULL; 1777 1778 res = ERR_PTR(-ENOMEM); 1779 fhandle = nfs_alloc_fhandle(); 1780 fattr = nfs_alloc_fattr(); 1781 if (fhandle == NULL || fattr == NULL) 1782 goto out; 1783 1784 label = nfs4_label_alloc(NFS_SERVER(dir), GFP_NOWAIT); 1785 if (IS_ERR(label)) 1786 goto out; 1787 1788 dir_verifier = nfs_save_change_attribute(dir); 1789 trace_nfs_lookup_enter(dir, dentry, flags); 1790 error = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr, label); 1791 if (error == -ENOENT) 1792 goto no_entry; 1793 if (error < 0) { 1794 res = ERR_PTR(error); 1795 goto out_label; 1796 } 1797 inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label); 1798 res = ERR_CAST(inode); 1799 if (IS_ERR(res)) 1800 goto out_label; 1801 1802 /* Notify readdir to use READDIRPLUS */ 1803 nfs_force_use_readdirplus(dir); 1804 1805 no_entry: 1806 res = d_splice_alias(inode, dentry); 1807 if (res != NULL) { 1808 if (IS_ERR(res)) 1809 goto out_label; 1810 dentry = res; 1811 } 1812 nfs_set_verifier(dentry, dir_verifier); 1813 out_label: 1814 trace_nfs_lookup_exit(dir, dentry, flags, error); 1815 nfs4_label_free(label); 1816 out: 1817 nfs_free_fattr(fattr); 1818 nfs_free_fhandle(fhandle); 1819 return res; 1820 } 1821 EXPORT_SYMBOL_GPL(nfs_lookup); 1822 1823 #if IS_ENABLED(CONFIG_NFS_V4) 1824 static int nfs4_lookup_revalidate(struct dentry *, unsigned int); 1825 1826 const struct dentry_operations nfs4_dentry_operations = { 1827 .d_revalidate = nfs4_lookup_revalidate, 1828 .d_weak_revalidate = nfs_weak_revalidate, 1829 .d_delete = nfs_dentry_delete, 1830 .d_iput = nfs_dentry_iput, 1831 .d_automount = nfs_d_automount, 1832 .d_release = nfs_d_release, 1833 }; 1834 EXPORT_SYMBOL_GPL(nfs4_dentry_operations); 1835 1836 static fmode_t flags_to_mode(int flags) 1837 { 1838 fmode_t res = (__force fmode_t)flags & FMODE_EXEC; 1839 if ((flags & O_ACCMODE) != O_WRONLY) 1840 res |= FMODE_READ; 1841 if ((flags & O_ACCMODE) != O_RDONLY) 1842 res |= FMODE_WRITE; 1843 return res; 1844 } 1845 1846 static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags, struct file *filp) 1847 { 1848 return alloc_nfs_open_context(dentry, flags_to_mode(open_flags), filp); 1849 } 1850 1851 static int do_open(struct inode *inode, struct file *filp) 1852 { 1853 nfs_fscache_open_file(inode, filp); 1854 return 0; 1855 } 1856 1857 static int nfs_finish_open(struct nfs_open_context *ctx, 1858 struct dentry *dentry, 1859 struct file *file, unsigned open_flags) 1860 { 1861 int err; 1862 1863 err = finish_open(file, dentry, do_open); 1864 if (err) 1865 goto out; 1866 if (S_ISREG(file->f_path.dentry->d_inode->i_mode)) 1867 nfs_file_set_open_context(file, ctx); 1868 else 1869 err = -EOPENSTALE; 1870 out: 1871 return err; 1872 } 1873 1874 int nfs_atomic_open(struct inode *dir, struct dentry *dentry, 1875 struct file *file, unsigned open_flags, 1876 umode_t mode) 1877 { 1878 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 1879 struct nfs_open_context *ctx; 1880 struct dentry *res; 1881 struct iattr attr = { .ia_valid = ATTR_OPEN }; 1882 struct inode *inode; 1883 unsigned int lookup_flags = 0; 1884 bool switched = false; 1885 int created = 0; 1886 int err; 1887 1888 /* Expect a negative dentry */ 1889 BUG_ON(d_inode(dentry)); 1890 1891 dfprintk(VFS, "NFS: atomic_open(%s/%lu), %pd\n", 1892 dir->i_sb->s_id, dir->i_ino, dentry); 1893 1894 err = nfs_check_flags(open_flags); 1895 if (err) 1896 return err; 1897 1898 /* NFS only supports OPEN on regular files */ 1899 if ((open_flags & O_DIRECTORY)) { 1900 if (!d_in_lookup(dentry)) { 1901 /* 1902 * Hashed negative dentry with O_DIRECTORY: dentry was 1903 * revalidated and is fine, no need to perform lookup 1904 * again 1905 */ 1906 return -ENOENT; 1907 } 1908 lookup_flags = LOOKUP_OPEN|LOOKUP_DIRECTORY; 1909 goto no_open; 1910 } 1911 1912 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) 1913 return -ENAMETOOLONG; 1914 1915 if (open_flags & O_CREAT) { 1916 struct nfs_server *server = NFS_SERVER(dir); 1917 1918 if (!(server->attr_bitmask[2] & FATTR4_WORD2_MODE_UMASK)) 1919 mode &= ~current_umask(); 1920 1921 attr.ia_valid |= ATTR_MODE; 1922 attr.ia_mode = mode; 1923 } 1924 if (open_flags & O_TRUNC) { 1925 attr.ia_valid |= ATTR_SIZE; 1926 attr.ia_size = 0; 1927 } 1928 1929 if (!(open_flags & O_CREAT) && !d_in_lookup(dentry)) { 1930 d_drop(dentry); 1931 switched = true; 1932 dentry = d_alloc_parallel(dentry->d_parent, 1933 &dentry->d_name, &wq); 1934 if (IS_ERR(dentry)) 1935 return PTR_ERR(dentry); 1936 if (unlikely(!d_in_lookup(dentry))) 1937 return finish_no_open(file, dentry); 1938 } 1939 1940 ctx = create_nfs_open_context(dentry, open_flags, file); 1941 err = PTR_ERR(ctx); 1942 if (IS_ERR(ctx)) 1943 goto out; 1944 1945 trace_nfs_atomic_open_enter(dir, ctx, open_flags); 1946 inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr, &created); 1947 if (created) 1948 file->f_mode |= FMODE_CREATED; 1949 if (IS_ERR(inode)) { 1950 err = PTR_ERR(inode); 1951 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err); 1952 put_nfs_open_context(ctx); 1953 d_drop(dentry); 1954 switch (err) { 1955 case -ENOENT: 1956 d_splice_alias(NULL, dentry); 1957 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1958 break; 1959 case -EISDIR: 1960 case -ENOTDIR: 1961 goto no_open; 1962 case -ELOOP: 1963 if (!(open_flags & O_NOFOLLOW)) 1964 goto no_open; 1965 break; 1966 /* case -EINVAL: */ 1967 default: 1968 break; 1969 } 1970 goto out; 1971 } 1972 1973 err = nfs_finish_open(ctx, ctx->dentry, file, open_flags); 1974 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err); 1975 put_nfs_open_context(ctx); 1976 out: 1977 if (unlikely(switched)) { 1978 d_lookup_done(dentry); 1979 dput(dentry); 1980 } 1981 return err; 1982 1983 no_open: 1984 res = nfs_lookup(dir, dentry, lookup_flags); 1985 if (!res) { 1986 inode = d_inode(dentry); 1987 if ((lookup_flags & LOOKUP_DIRECTORY) && inode && 1988 !S_ISDIR(inode->i_mode)) 1989 res = ERR_PTR(-ENOTDIR); 1990 else if (inode && S_ISREG(inode->i_mode)) 1991 res = ERR_PTR(-EOPENSTALE); 1992 } else if (!IS_ERR(res)) { 1993 inode = d_inode(res); 1994 if ((lookup_flags & LOOKUP_DIRECTORY) && inode && 1995 !S_ISDIR(inode->i_mode)) { 1996 dput(res); 1997 res = ERR_PTR(-ENOTDIR); 1998 } else if (inode && S_ISREG(inode->i_mode)) { 1999 dput(res); 2000 res = ERR_PTR(-EOPENSTALE); 2001 } 2002 } 2003 if (switched) { 2004 d_lookup_done(dentry); 2005 if (!res) 2006 res = dentry; 2007 else 2008 dput(dentry); 2009 } 2010 if (IS_ERR(res)) 2011 return PTR_ERR(res); 2012 return finish_no_open(file, res); 2013 } 2014 EXPORT_SYMBOL_GPL(nfs_atomic_open); 2015 2016 static int 2017 nfs4_do_lookup_revalidate(struct inode *dir, struct dentry *dentry, 2018 unsigned int flags) 2019 { 2020 struct inode *inode; 2021 2022 if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY)) 2023 goto full_reval; 2024 if (d_mountpoint(dentry)) 2025 goto full_reval; 2026 2027 inode = d_inode(dentry); 2028 2029 /* We can't create new files in nfs_open_revalidate(), so we 2030 * optimize away revalidation of negative dentries. 2031 */ 2032 if (inode == NULL) 2033 goto full_reval; 2034 2035 if (nfs_verifier_is_delegated(dentry)) 2036 return nfs_lookup_revalidate_delegated(dir, dentry, inode); 2037 2038 /* NFS only supports OPEN on regular files */ 2039 if (!S_ISREG(inode->i_mode)) 2040 goto full_reval; 2041 2042 /* We cannot do exclusive creation on a positive dentry */ 2043 if (flags & (LOOKUP_EXCL | LOOKUP_REVAL)) 2044 goto reval_dentry; 2045 2046 /* Check if the directory changed */ 2047 if (!nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU)) 2048 goto reval_dentry; 2049 2050 /* Let f_op->open() actually open (and revalidate) the file */ 2051 return 1; 2052 reval_dentry: 2053 if (flags & LOOKUP_RCU) 2054 return -ECHILD; 2055 return nfs_lookup_revalidate_dentry(dir, dentry, inode); 2056 2057 full_reval: 2058 return nfs_do_lookup_revalidate(dir, dentry, flags); 2059 } 2060 2061 static int nfs4_lookup_revalidate(struct dentry *dentry, unsigned int flags) 2062 { 2063 return __nfs_lookup_revalidate(dentry, flags, 2064 nfs4_do_lookup_revalidate); 2065 } 2066 2067 #endif /* CONFIG_NFSV4 */ 2068 2069 struct dentry * 2070 nfs_add_or_obtain(struct dentry *dentry, struct nfs_fh *fhandle, 2071 struct nfs_fattr *fattr, 2072 struct nfs4_label *label) 2073 { 2074 struct dentry *parent = dget_parent(dentry); 2075 struct inode *dir = d_inode(parent); 2076 struct inode *inode; 2077 struct dentry *d; 2078 int error; 2079 2080 d_drop(dentry); 2081 2082 if (fhandle->size == 0) { 2083 error = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr, NULL); 2084 if (error) 2085 goto out_error; 2086 } 2087 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 2088 if (!(fattr->valid & NFS_ATTR_FATTR)) { 2089 struct nfs_server *server = NFS_SB(dentry->d_sb); 2090 error = server->nfs_client->rpc_ops->getattr(server, fhandle, 2091 fattr, NULL, NULL); 2092 if (error < 0) 2093 goto out_error; 2094 } 2095 inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label); 2096 d = d_splice_alias(inode, dentry); 2097 out: 2098 dput(parent); 2099 return d; 2100 out_error: 2101 d = ERR_PTR(error); 2102 goto out; 2103 } 2104 EXPORT_SYMBOL_GPL(nfs_add_or_obtain); 2105 2106 /* 2107 * Code common to create, mkdir, and mknod. 2108 */ 2109 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle, 2110 struct nfs_fattr *fattr, 2111 struct nfs4_label *label) 2112 { 2113 struct dentry *d; 2114 2115 d = nfs_add_or_obtain(dentry, fhandle, fattr, label); 2116 if (IS_ERR(d)) 2117 return PTR_ERR(d); 2118 2119 /* Callers don't care */ 2120 dput(d); 2121 return 0; 2122 } 2123 EXPORT_SYMBOL_GPL(nfs_instantiate); 2124 2125 /* 2126 * Following a failed create operation, we drop the dentry rather 2127 * than retain a negative dentry. This avoids a problem in the event 2128 * that the operation succeeded on the server, but an error in the 2129 * reply path made it appear to have failed. 2130 */ 2131 int nfs_create(struct user_namespace *mnt_userns, struct inode *dir, 2132 struct dentry *dentry, umode_t mode, bool excl) 2133 { 2134 struct iattr attr; 2135 int open_flags = excl ? O_CREAT | O_EXCL : O_CREAT; 2136 int error; 2137 2138 dfprintk(VFS, "NFS: create(%s/%lu), %pd\n", 2139 dir->i_sb->s_id, dir->i_ino, dentry); 2140 2141 attr.ia_mode = mode; 2142 attr.ia_valid = ATTR_MODE; 2143 2144 trace_nfs_create_enter(dir, dentry, open_flags); 2145 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags); 2146 trace_nfs_create_exit(dir, dentry, open_flags, error); 2147 if (error != 0) 2148 goto out_err; 2149 return 0; 2150 out_err: 2151 d_drop(dentry); 2152 return error; 2153 } 2154 EXPORT_SYMBOL_GPL(nfs_create); 2155 2156 /* 2157 * See comments for nfs_proc_create regarding failed operations. 2158 */ 2159 int 2160 nfs_mknod(struct user_namespace *mnt_userns, struct inode *dir, 2161 struct dentry *dentry, umode_t mode, dev_t rdev) 2162 { 2163 struct iattr attr; 2164 int status; 2165 2166 dfprintk(VFS, "NFS: mknod(%s/%lu), %pd\n", 2167 dir->i_sb->s_id, dir->i_ino, dentry); 2168 2169 attr.ia_mode = mode; 2170 attr.ia_valid = ATTR_MODE; 2171 2172 trace_nfs_mknod_enter(dir, dentry); 2173 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev); 2174 trace_nfs_mknod_exit(dir, dentry, status); 2175 if (status != 0) 2176 goto out_err; 2177 return 0; 2178 out_err: 2179 d_drop(dentry); 2180 return status; 2181 } 2182 EXPORT_SYMBOL_GPL(nfs_mknod); 2183 2184 /* 2185 * See comments for nfs_proc_create regarding failed operations. 2186 */ 2187 int nfs_mkdir(struct user_namespace *mnt_userns, struct inode *dir, 2188 struct dentry *dentry, umode_t mode) 2189 { 2190 struct iattr attr; 2191 int error; 2192 2193 dfprintk(VFS, "NFS: mkdir(%s/%lu), %pd\n", 2194 dir->i_sb->s_id, dir->i_ino, dentry); 2195 2196 attr.ia_valid = ATTR_MODE; 2197 attr.ia_mode = mode | S_IFDIR; 2198 2199 trace_nfs_mkdir_enter(dir, dentry); 2200 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr); 2201 trace_nfs_mkdir_exit(dir, dentry, error); 2202 if (error != 0) 2203 goto out_err; 2204 return 0; 2205 out_err: 2206 d_drop(dentry); 2207 return error; 2208 } 2209 EXPORT_SYMBOL_GPL(nfs_mkdir); 2210 2211 static void nfs_dentry_handle_enoent(struct dentry *dentry) 2212 { 2213 if (simple_positive(dentry)) 2214 d_delete(dentry); 2215 } 2216 2217 int nfs_rmdir(struct inode *dir, struct dentry *dentry) 2218 { 2219 int error; 2220 2221 dfprintk(VFS, "NFS: rmdir(%s/%lu), %pd\n", 2222 dir->i_sb->s_id, dir->i_ino, dentry); 2223 2224 trace_nfs_rmdir_enter(dir, dentry); 2225 if (d_really_is_positive(dentry)) { 2226 down_write(&NFS_I(d_inode(dentry))->rmdir_sem); 2227 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name); 2228 /* Ensure the VFS deletes this inode */ 2229 switch (error) { 2230 case 0: 2231 clear_nlink(d_inode(dentry)); 2232 break; 2233 case -ENOENT: 2234 nfs_dentry_handle_enoent(dentry); 2235 } 2236 up_write(&NFS_I(d_inode(dentry))->rmdir_sem); 2237 } else 2238 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name); 2239 trace_nfs_rmdir_exit(dir, dentry, error); 2240 2241 return error; 2242 } 2243 EXPORT_SYMBOL_GPL(nfs_rmdir); 2244 2245 /* 2246 * Remove a file after making sure there are no pending writes, 2247 * and after checking that the file has only one user. 2248 * 2249 * We invalidate the attribute cache and free the inode prior to the operation 2250 * to avoid possible races if the server reuses the inode. 2251 */ 2252 static int nfs_safe_remove(struct dentry *dentry) 2253 { 2254 struct inode *dir = d_inode(dentry->d_parent); 2255 struct inode *inode = d_inode(dentry); 2256 int error = -EBUSY; 2257 2258 dfprintk(VFS, "NFS: safe_remove(%pd2)\n", dentry); 2259 2260 /* If the dentry was sillyrenamed, we simply call d_delete() */ 2261 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { 2262 error = 0; 2263 goto out; 2264 } 2265 2266 trace_nfs_remove_enter(dir, dentry); 2267 if (inode != NULL) { 2268 error = NFS_PROTO(dir)->remove(dir, dentry); 2269 if (error == 0) 2270 nfs_drop_nlink(inode); 2271 } else 2272 error = NFS_PROTO(dir)->remove(dir, dentry); 2273 if (error == -ENOENT) 2274 nfs_dentry_handle_enoent(dentry); 2275 trace_nfs_remove_exit(dir, dentry, error); 2276 out: 2277 return error; 2278 } 2279 2280 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode 2281 * belongs to an active ".nfs..." file and we return -EBUSY. 2282 * 2283 * If sillyrename() returns 0, we do nothing, otherwise we unlink. 2284 */ 2285 int nfs_unlink(struct inode *dir, struct dentry *dentry) 2286 { 2287 int error; 2288 int need_rehash = 0; 2289 2290 dfprintk(VFS, "NFS: unlink(%s/%lu, %pd)\n", dir->i_sb->s_id, 2291 dir->i_ino, dentry); 2292 2293 trace_nfs_unlink_enter(dir, dentry); 2294 spin_lock(&dentry->d_lock); 2295 if (d_count(dentry) > 1) { 2296 spin_unlock(&dentry->d_lock); 2297 /* Start asynchronous writeout of the inode */ 2298 write_inode_now(d_inode(dentry), 0); 2299 error = nfs_sillyrename(dir, dentry); 2300 goto out; 2301 } 2302 if (!d_unhashed(dentry)) { 2303 __d_drop(dentry); 2304 need_rehash = 1; 2305 } 2306 spin_unlock(&dentry->d_lock); 2307 error = nfs_safe_remove(dentry); 2308 if (!error || error == -ENOENT) { 2309 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 2310 } else if (need_rehash) 2311 d_rehash(dentry); 2312 out: 2313 trace_nfs_unlink_exit(dir, dentry, error); 2314 return error; 2315 } 2316 EXPORT_SYMBOL_GPL(nfs_unlink); 2317 2318 /* 2319 * To create a symbolic link, most file systems instantiate a new inode, 2320 * add a page to it containing the path, then write it out to the disk 2321 * using prepare_write/commit_write. 2322 * 2323 * Unfortunately the NFS client can't create the in-core inode first 2324 * because it needs a file handle to create an in-core inode (see 2325 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the 2326 * symlink request has completed on the server. 2327 * 2328 * So instead we allocate a raw page, copy the symname into it, then do 2329 * the SYMLINK request with the page as the buffer. If it succeeds, we 2330 * now have a new file handle and can instantiate an in-core NFS inode 2331 * and move the raw page into its mapping. 2332 */ 2333 int nfs_symlink(struct user_namespace *mnt_userns, struct inode *dir, 2334 struct dentry *dentry, const char *symname) 2335 { 2336 struct page *page; 2337 char *kaddr; 2338 struct iattr attr; 2339 unsigned int pathlen = strlen(symname); 2340 int error; 2341 2342 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s)\n", dir->i_sb->s_id, 2343 dir->i_ino, dentry, symname); 2344 2345 if (pathlen > PAGE_SIZE) 2346 return -ENAMETOOLONG; 2347 2348 attr.ia_mode = S_IFLNK | S_IRWXUGO; 2349 attr.ia_valid = ATTR_MODE; 2350 2351 page = alloc_page(GFP_USER); 2352 if (!page) 2353 return -ENOMEM; 2354 2355 kaddr = page_address(page); 2356 memcpy(kaddr, symname, pathlen); 2357 if (pathlen < PAGE_SIZE) 2358 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen); 2359 2360 trace_nfs_symlink_enter(dir, dentry); 2361 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr); 2362 trace_nfs_symlink_exit(dir, dentry, error); 2363 if (error != 0) { 2364 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s) error %d\n", 2365 dir->i_sb->s_id, dir->i_ino, 2366 dentry, symname, error); 2367 d_drop(dentry); 2368 __free_page(page); 2369 return error; 2370 } 2371 2372 /* 2373 * No big deal if we can't add this page to the page cache here. 2374 * READLINK will get the missing page from the server if needed. 2375 */ 2376 if (!add_to_page_cache_lru(page, d_inode(dentry)->i_mapping, 0, 2377 GFP_KERNEL)) { 2378 SetPageUptodate(page); 2379 unlock_page(page); 2380 /* 2381 * add_to_page_cache_lru() grabs an extra page refcount. 2382 * Drop it here to avoid leaking this page later. 2383 */ 2384 put_page(page); 2385 } else 2386 __free_page(page); 2387 2388 return 0; 2389 } 2390 EXPORT_SYMBOL_GPL(nfs_symlink); 2391 2392 int 2393 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) 2394 { 2395 struct inode *inode = d_inode(old_dentry); 2396 int error; 2397 2398 dfprintk(VFS, "NFS: link(%pd2 -> %pd2)\n", 2399 old_dentry, dentry); 2400 2401 trace_nfs_link_enter(inode, dir, dentry); 2402 d_drop(dentry); 2403 if (S_ISREG(inode->i_mode)) 2404 nfs_sync_inode(inode); 2405 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name); 2406 if (error == 0) { 2407 ihold(inode); 2408 d_add(dentry, inode); 2409 } 2410 trace_nfs_link_exit(inode, dir, dentry, error); 2411 return error; 2412 } 2413 EXPORT_SYMBOL_GPL(nfs_link); 2414 2415 /* 2416 * RENAME 2417 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a 2418 * different file handle for the same inode after a rename (e.g. when 2419 * moving to a different directory). A fail-safe method to do so would 2420 * be to look up old_dir/old_name, create a link to new_dir/new_name and 2421 * rename the old file using the sillyrename stuff. This way, the original 2422 * file in old_dir will go away when the last process iput()s the inode. 2423 * 2424 * FIXED. 2425 * 2426 * It actually works quite well. One needs to have the possibility for 2427 * at least one ".nfs..." file in each directory the file ever gets 2428 * moved or linked to which happens automagically with the new 2429 * implementation that only depends on the dcache stuff instead of 2430 * using the inode layer 2431 * 2432 * Unfortunately, things are a little more complicated than indicated 2433 * above. For a cross-directory move, we want to make sure we can get 2434 * rid of the old inode after the operation. This means there must be 2435 * no pending writes (if it's a file), and the use count must be 1. 2436 * If these conditions are met, we can drop the dentries before doing 2437 * the rename. 2438 */ 2439 int nfs_rename(struct user_namespace *mnt_userns, struct inode *old_dir, 2440 struct dentry *old_dentry, struct inode *new_dir, 2441 struct dentry *new_dentry, unsigned int flags) 2442 { 2443 struct inode *old_inode = d_inode(old_dentry); 2444 struct inode *new_inode = d_inode(new_dentry); 2445 struct dentry *dentry = NULL, *rehash = NULL; 2446 struct rpc_task *task; 2447 int error = -EBUSY; 2448 2449 if (flags) 2450 return -EINVAL; 2451 2452 dfprintk(VFS, "NFS: rename(%pd2 -> %pd2, ct=%d)\n", 2453 old_dentry, new_dentry, 2454 d_count(new_dentry)); 2455 2456 trace_nfs_rename_enter(old_dir, old_dentry, new_dir, new_dentry); 2457 /* 2458 * For non-directories, check whether the target is busy and if so, 2459 * make a copy of the dentry and then do a silly-rename. If the 2460 * silly-rename succeeds, the copied dentry is hashed and becomes 2461 * the new target. 2462 */ 2463 if (new_inode && !S_ISDIR(new_inode->i_mode)) { 2464 /* 2465 * To prevent any new references to the target during the 2466 * rename, we unhash the dentry in advance. 2467 */ 2468 if (!d_unhashed(new_dentry)) { 2469 d_drop(new_dentry); 2470 rehash = new_dentry; 2471 } 2472 2473 if (d_count(new_dentry) > 2) { 2474 int err; 2475 2476 /* copy the target dentry's name */ 2477 dentry = d_alloc(new_dentry->d_parent, 2478 &new_dentry->d_name); 2479 if (!dentry) 2480 goto out; 2481 2482 /* silly-rename the existing target ... */ 2483 err = nfs_sillyrename(new_dir, new_dentry); 2484 if (err) 2485 goto out; 2486 2487 new_dentry = dentry; 2488 rehash = NULL; 2489 new_inode = NULL; 2490 } 2491 } 2492 2493 if (S_ISREG(old_inode->i_mode)) 2494 nfs_sync_inode(old_inode); 2495 task = nfs_async_rename(old_dir, new_dir, old_dentry, new_dentry, NULL); 2496 if (IS_ERR(task)) { 2497 error = PTR_ERR(task); 2498 goto out; 2499 } 2500 2501 error = rpc_wait_for_completion_task(task); 2502 if (error != 0) { 2503 ((struct nfs_renamedata *)task->tk_calldata)->cancelled = 1; 2504 /* Paired with the atomic_dec_and_test() barrier in rpc_do_put_task() */ 2505 smp_wmb(); 2506 } else 2507 error = task->tk_status; 2508 rpc_put_task(task); 2509 /* Ensure the inode attributes are revalidated */ 2510 if (error == 0) { 2511 spin_lock(&old_inode->i_lock); 2512 NFS_I(old_inode)->attr_gencount = nfs_inc_attr_generation_counter(); 2513 nfs_set_cache_invalid(old_inode, NFS_INO_INVALID_CHANGE | 2514 NFS_INO_INVALID_CTIME | 2515 NFS_INO_REVAL_FORCED); 2516 spin_unlock(&old_inode->i_lock); 2517 } 2518 out: 2519 if (rehash) 2520 d_rehash(rehash); 2521 trace_nfs_rename_exit(old_dir, old_dentry, 2522 new_dir, new_dentry, error); 2523 if (!error) { 2524 if (new_inode != NULL) 2525 nfs_drop_nlink(new_inode); 2526 /* 2527 * The d_move() should be here instead of in an async RPC completion 2528 * handler because we need the proper locks to move the dentry. If 2529 * we're interrupted by a signal, the async RPC completion handler 2530 * should mark the directories for revalidation. 2531 */ 2532 d_move(old_dentry, new_dentry); 2533 nfs_set_verifier(old_dentry, 2534 nfs_save_change_attribute(new_dir)); 2535 } else if (error == -ENOENT) 2536 nfs_dentry_handle_enoent(old_dentry); 2537 2538 /* new dentry created? */ 2539 if (dentry) 2540 dput(dentry); 2541 return error; 2542 } 2543 EXPORT_SYMBOL_GPL(nfs_rename); 2544 2545 static DEFINE_SPINLOCK(nfs_access_lru_lock); 2546 static LIST_HEAD(nfs_access_lru_list); 2547 static atomic_long_t nfs_access_nr_entries; 2548 2549 static unsigned long nfs_access_max_cachesize = 4*1024*1024; 2550 module_param(nfs_access_max_cachesize, ulong, 0644); 2551 MODULE_PARM_DESC(nfs_access_max_cachesize, "NFS access maximum total cache length"); 2552 2553 static void nfs_access_free_entry(struct nfs_access_entry *entry) 2554 { 2555 put_cred(entry->cred); 2556 kfree_rcu(entry, rcu_head); 2557 smp_mb__before_atomic(); 2558 atomic_long_dec(&nfs_access_nr_entries); 2559 smp_mb__after_atomic(); 2560 } 2561 2562 static void nfs_access_free_list(struct list_head *head) 2563 { 2564 struct nfs_access_entry *cache; 2565 2566 while (!list_empty(head)) { 2567 cache = list_entry(head->next, struct nfs_access_entry, lru); 2568 list_del(&cache->lru); 2569 nfs_access_free_entry(cache); 2570 } 2571 } 2572 2573 static unsigned long 2574 nfs_do_access_cache_scan(unsigned int nr_to_scan) 2575 { 2576 LIST_HEAD(head); 2577 struct nfs_inode *nfsi, *next; 2578 struct nfs_access_entry *cache; 2579 long freed = 0; 2580 2581 spin_lock(&nfs_access_lru_lock); 2582 list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) { 2583 struct inode *inode; 2584 2585 if (nr_to_scan-- == 0) 2586 break; 2587 inode = &nfsi->vfs_inode; 2588 spin_lock(&inode->i_lock); 2589 if (list_empty(&nfsi->access_cache_entry_lru)) 2590 goto remove_lru_entry; 2591 cache = list_entry(nfsi->access_cache_entry_lru.next, 2592 struct nfs_access_entry, lru); 2593 list_move(&cache->lru, &head); 2594 rb_erase(&cache->rb_node, &nfsi->access_cache); 2595 freed++; 2596 if (!list_empty(&nfsi->access_cache_entry_lru)) 2597 list_move_tail(&nfsi->access_cache_inode_lru, 2598 &nfs_access_lru_list); 2599 else { 2600 remove_lru_entry: 2601 list_del_init(&nfsi->access_cache_inode_lru); 2602 smp_mb__before_atomic(); 2603 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags); 2604 smp_mb__after_atomic(); 2605 } 2606 spin_unlock(&inode->i_lock); 2607 } 2608 spin_unlock(&nfs_access_lru_lock); 2609 nfs_access_free_list(&head); 2610 return freed; 2611 } 2612 2613 unsigned long 2614 nfs_access_cache_scan(struct shrinker *shrink, struct shrink_control *sc) 2615 { 2616 int nr_to_scan = sc->nr_to_scan; 2617 gfp_t gfp_mask = sc->gfp_mask; 2618 2619 if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL) 2620 return SHRINK_STOP; 2621 return nfs_do_access_cache_scan(nr_to_scan); 2622 } 2623 2624 2625 unsigned long 2626 nfs_access_cache_count(struct shrinker *shrink, struct shrink_control *sc) 2627 { 2628 return vfs_pressure_ratio(atomic_long_read(&nfs_access_nr_entries)); 2629 } 2630 2631 static void 2632 nfs_access_cache_enforce_limit(void) 2633 { 2634 long nr_entries = atomic_long_read(&nfs_access_nr_entries); 2635 unsigned long diff; 2636 unsigned int nr_to_scan; 2637 2638 if (nr_entries < 0 || nr_entries <= nfs_access_max_cachesize) 2639 return; 2640 nr_to_scan = 100; 2641 diff = nr_entries - nfs_access_max_cachesize; 2642 if (diff < nr_to_scan) 2643 nr_to_scan = diff; 2644 nfs_do_access_cache_scan(nr_to_scan); 2645 } 2646 2647 static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head) 2648 { 2649 struct rb_root *root_node = &nfsi->access_cache; 2650 struct rb_node *n; 2651 struct nfs_access_entry *entry; 2652 2653 /* Unhook entries from the cache */ 2654 while ((n = rb_first(root_node)) != NULL) { 2655 entry = rb_entry(n, struct nfs_access_entry, rb_node); 2656 rb_erase(n, root_node); 2657 list_move(&entry->lru, head); 2658 } 2659 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS; 2660 } 2661 2662 void nfs_access_zap_cache(struct inode *inode) 2663 { 2664 LIST_HEAD(head); 2665 2666 if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0) 2667 return; 2668 /* Remove from global LRU init */ 2669 spin_lock(&nfs_access_lru_lock); 2670 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) 2671 list_del_init(&NFS_I(inode)->access_cache_inode_lru); 2672 2673 spin_lock(&inode->i_lock); 2674 __nfs_access_zap_cache(NFS_I(inode), &head); 2675 spin_unlock(&inode->i_lock); 2676 spin_unlock(&nfs_access_lru_lock); 2677 nfs_access_free_list(&head); 2678 } 2679 EXPORT_SYMBOL_GPL(nfs_access_zap_cache); 2680 2681 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, const struct cred *cred) 2682 { 2683 struct rb_node *n = NFS_I(inode)->access_cache.rb_node; 2684 2685 while (n != NULL) { 2686 struct nfs_access_entry *entry = 2687 rb_entry(n, struct nfs_access_entry, rb_node); 2688 int cmp = cred_fscmp(cred, entry->cred); 2689 2690 if (cmp < 0) 2691 n = n->rb_left; 2692 else if (cmp > 0) 2693 n = n->rb_right; 2694 else 2695 return entry; 2696 } 2697 return NULL; 2698 } 2699 2700 static int nfs_access_get_cached_locked(struct inode *inode, const struct cred *cred, struct nfs_access_entry *res, bool may_block) 2701 { 2702 struct nfs_inode *nfsi = NFS_I(inode); 2703 struct nfs_access_entry *cache; 2704 bool retry = true; 2705 int err; 2706 2707 spin_lock(&inode->i_lock); 2708 for(;;) { 2709 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS) 2710 goto out_zap; 2711 cache = nfs_access_search_rbtree(inode, cred); 2712 err = -ENOENT; 2713 if (cache == NULL) 2714 goto out; 2715 /* Found an entry, is our attribute cache valid? */ 2716 if (!nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS)) 2717 break; 2718 if (!retry) 2719 break; 2720 err = -ECHILD; 2721 if (!may_block) 2722 goto out; 2723 spin_unlock(&inode->i_lock); 2724 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode); 2725 if (err) 2726 return err; 2727 spin_lock(&inode->i_lock); 2728 retry = false; 2729 } 2730 res->cred = cache->cred; 2731 res->mask = cache->mask; 2732 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru); 2733 err = 0; 2734 out: 2735 spin_unlock(&inode->i_lock); 2736 return err; 2737 out_zap: 2738 spin_unlock(&inode->i_lock); 2739 nfs_access_zap_cache(inode); 2740 return -ENOENT; 2741 } 2742 2743 static int nfs_access_get_cached_rcu(struct inode *inode, const struct cred *cred, struct nfs_access_entry *res) 2744 { 2745 /* Only check the most recently returned cache entry, 2746 * but do it without locking. 2747 */ 2748 struct nfs_inode *nfsi = NFS_I(inode); 2749 struct nfs_access_entry *cache; 2750 int err = -ECHILD; 2751 struct list_head *lh; 2752 2753 rcu_read_lock(); 2754 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS) 2755 goto out; 2756 lh = rcu_dereference(list_tail_rcu(&nfsi->access_cache_entry_lru)); 2757 cache = list_entry(lh, struct nfs_access_entry, lru); 2758 if (lh == &nfsi->access_cache_entry_lru || 2759 cred_fscmp(cred, cache->cred) != 0) 2760 cache = NULL; 2761 if (cache == NULL) 2762 goto out; 2763 if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS)) 2764 goto out; 2765 res->cred = cache->cred; 2766 res->mask = cache->mask; 2767 err = 0; 2768 out: 2769 rcu_read_unlock(); 2770 return err; 2771 } 2772 2773 int nfs_access_get_cached(struct inode *inode, const struct cred *cred, struct 2774 nfs_access_entry *res, bool may_block) 2775 { 2776 int status; 2777 2778 status = nfs_access_get_cached_rcu(inode, cred, res); 2779 if (status != 0) 2780 status = nfs_access_get_cached_locked(inode, cred, res, 2781 may_block); 2782 2783 return status; 2784 } 2785 EXPORT_SYMBOL_GPL(nfs_access_get_cached); 2786 2787 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set) 2788 { 2789 struct nfs_inode *nfsi = NFS_I(inode); 2790 struct rb_root *root_node = &nfsi->access_cache; 2791 struct rb_node **p = &root_node->rb_node; 2792 struct rb_node *parent = NULL; 2793 struct nfs_access_entry *entry; 2794 int cmp; 2795 2796 spin_lock(&inode->i_lock); 2797 while (*p != NULL) { 2798 parent = *p; 2799 entry = rb_entry(parent, struct nfs_access_entry, rb_node); 2800 cmp = cred_fscmp(set->cred, entry->cred); 2801 2802 if (cmp < 0) 2803 p = &parent->rb_left; 2804 else if (cmp > 0) 2805 p = &parent->rb_right; 2806 else 2807 goto found; 2808 } 2809 rb_link_node(&set->rb_node, parent, p); 2810 rb_insert_color(&set->rb_node, root_node); 2811 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); 2812 spin_unlock(&inode->i_lock); 2813 return; 2814 found: 2815 rb_replace_node(parent, &set->rb_node, root_node); 2816 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); 2817 list_del(&entry->lru); 2818 spin_unlock(&inode->i_lock); 2819 nfs_access_free_entry(entry); 2820 } 2821 2822 void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set) 2823 { 2824 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL); 2825 if (cache == NULL) 2826 return; 2827 RB_CLEAR_NODE(&cache->rb_node); 2828 cache->cred = get_cred(set->cred); 2829 cache->mask = set->mask; 2830 2831 /* The above field assignments must be visible 2832 * before this item appears on the lru. We cannot easily 2833 * use rcu_assign_pointer, so just force the memory barrier. 2834 */ 2835 smp_wmb(); 2836 nfs_access_add_rbtree(inode, cache); 2837 2838 /* Update accounting */ 2839 smp_mb__before_atomic(); 2840 atomic_long_inc(&nfs_access_nr_entries); 2841 smp_mb__after_atomic(); 2842 2843 /* Add inode to global LRU list */ 2844 if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) { 2845 spin_lock(&nfs_access_lru_lock); 2846 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) 2847 list_add_tail(&NFS_I(inode)->access_cache_inode_lru, 2848 &nfs_access_lru_list); 2849 spin_unlock(&nfs_access_lru_lock); 2850 } 2851 nfs_access_cache_enforce_limit(); 2852 } 2853 EXPORT_SYMBOL_GPL(nfs_access_add_cache); 2854 2855 #define NFS_MAY_READ (NFS_ACCESS_READ) 2856 #define NFS_MAY_WRITE (NFS_ACCESS_MODIFY | \ 2857 NFS_ACCESS_EXTEND | \ 2858 NFS_ACCESS_DELETE) 2859 #define NFS_FILE_MAY_WRITE (NFS_ACCESS_MODIFY | \ 2860 NFS_ACCESS_EXTEND) 2861 #define NFS_DIR_MAY_WRITE NFS_MAY_WRITE 2862 #define NFS_MAY_LOOKUP (NFS_ACCESS_LOOKUP) 2863 #define NFS_MAY_EXECUTE (NFS_ACCESS_EXECUTE) 2864 static int 2865 nfs_access_calc_mask(u32 access_result, umode_t umode) 2866 { 2867 int mask = 0; 2868 2869 if (access_result & NFS_MAY_READ) 2870 mask |= MAY_READ; 2871 if (S_ISDIR(umode)) { 2872 if ((access_result & NFS_DIR_MAY_WRITE) == NFS_DIR_MAY_WRITE) 2873 mask |= MAY_WRITE; 2874 if ((access_result & NFS_MAY_LOOKUP) == NFS_MAY_LOOKUP) 2875 mask |= MAY_EXEC; 2876 } else if (S_ISREG(umode)) { 2877 if ((access_result & NFS_FILE_MAY_WRITE) == NFS_FILE_MAY_WRITE) 2878 mask |= MAY_WRITE; 2879 if ((access_result & NFS_MAY_EXECUTE) == NFS_MAY_EXECUTE) 2880 mask |= MAY_EXEC; 2881 } else if (access_result & NFS_MAY_WRITE) 2882 mask |= MAY_WRITE; 2883 return mask; 2884 } 2885 2886 void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result) 2887 { 2888 entry->mask = access_result; 2889 } 2890 EXPORT_SYMBOL_GPL(nfs_access_set_mask); 2891 2892 static int nfs_do_access(struct inode *inode, const struct cred *cred, int mask) 2893 { 2894 struct nfs_access_entry cache; 2895 bool may_block = (mask & MAY_NOT_BLOCK) == 0; 2896 int cache_mask = -1; 2897 int status; 2898 2899 trace_nfs_access_enter(inode); 2900 2901 status = nfs_access_get_cached(inode, cred, &cache, may_block); 2902 if (status == 0) 2903 goto out_cached; 2904 2905 status = -ECHILD; 2906 if (!may_block) 2907 goto out; 2908 2909 /* 2910 * Determine which access bits we want to ask for... 2911 */ 2912 cache.mask = NFS_ACCESS_READ | NFS_ACCESS_MODIFY | NFS_ACCESS_EXTEND; 2913 if (nfs_server_capable(inode, NFS_CAP_XATTR)) { 2914 cache.mask |= NFS_ACCESS_XAREAD | NFS_ACCESS_XAWRITE | 2915 NFS_ACCESS_XALIST; 2916 } 2917 if (S_ISDIR(inode->i_mode)) 2918 cache.mask |= NFS_ACCESS_DELETE | NFS_ACCESS_LOOKUP; 2919 else 2920 cache.mask |= NFS_ACCESS_EXECUTE; 2921 cache.cred = cred; 2922 status = NFS_PROTO(inode)->access(inode, &cache); 2923 if (status != 0) { 2924 if (status == -ESTALE) { 2925 if (!S_ISDIR(inode->i_mode)) 2926 nfs_set_inode_stale(inode); 2927 else 2928 nfs_zap_caches(inode); 2929 } 2930 goto out; 2931 } 2932 nfs_access_add_cache(inode, &cache); 2933 out_cached: 2934 cache_mask = nfs_access_calc_mask(cache.mask, inode->i_mode); 2935 if ((mask & ~cache_mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) != 0) 2936 status = -EACCES; 2937 out: 2938 trace_nfs_access_exit(inode, mask, cache_mask, status); 2939 return status; 2940 } 2941 2942 static int nfs_open_permission_mask(int openflags) 2943 { 2944 int mask = 0; 2945 2946 if (openflags & __FMODE_EXEC) { 2947 /* ONLY check exec rights */ 2948 mask = MAY_EXEC; 2949 } else { 2950 if ((openflags & O_ACCMODE) != O_WRONLY) 2951 mask |= MAY_READ; 2952 if ((openflags & O_ACCMODE) != O_RDONLY) 2953 mask |= MAY_WRITE; 2954 } 2955 2956 return mask; 2957 } 2958 2959 int nfs_may_open(struct inode *inode, const struct cred *cred, int openflags) 2960 { 2961 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags)); 2962 } 2963 EXPORT_SYMBOL_GPL(nfs_may_open); 2964 2965 static int nfs_execute_ok(struct inode *inode, int mask) 2966 { 2967 struct nfs_server *server = NFS_SERVER(inode); 2968 int ret = 0; 2969 2970 if (S_ISDIR(inode->i_mode)) 2971 return 0; 2972 if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_MODE)) { 2973 if (mask & MAY_NOT_BLOCK) 2974 return -ECHILD; 2975 ret = __nfs_revalidate_inode(server, inode); 2976 } 2977 if (ret == 0 && !execute_ok(inode)) 2978 ret = -EACCES; 2979 return ret; 2980 } 2981 2982 int nfs_permission(struct user_namespace *mnt_userns, 2983 struct inode *inode, 2984 int mask) 2985 { 2986 const struct cred *cred = current_cred(); 2987 int res = 0; 2988 2989 nfs_inc_stats(inode, NFSIOS_VFSACCESS); 2990 2991 if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) 2992 goto out; 2993 /* Is this sys_access() ? */ 2994 if (mask & (MAY_ACCESS | MAY_CHDIR)) 2995 goto force_lookup; 2996 2997 switch (inode->i_mode & S_IFMT) { 2998 case S_IFLNK: 2999 goto out; 3000 case S_IFREG: 3001 if ((mask & MAY_OPEN) && 3002 nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN)) 3003 return 0; 3004 break; 3005 case S_IFDIR: 3006 /* 3007 * Optimize away all write operations, since the server 3008 * will check permissions when we perform the op. 3009 */ 3010 if ((mask & MAY_WRITE) && !(mask & MAY_READ)) 3011 goto out; 3012 } 3013 3014 force_lookup: 3015 if (!NFS_PROTO(inode)->access) 3016 goto out_notsup; 3017 3018 res = nfs_do_access(inode, cred, mask); 3019 out: 3020 if (!res && (mask & MAY_EXEC)) 3021 res = nfs_execute_ok(inode, mask); 3022 3023 dfprintk(VFS, "NFS: permission(%s/%lu), mask=0x%x, res=%d\n", 3024 inode->i_sb->s_id, inode->i_ino, mask, res); 3025 return res; 3026 out_notsup: 3027 if (mask & MAY_NOT_BLOCK) 3028 return -ECHILD; 3029 3030 res = nfs_revalidate_inode(inode, NFS_INO_INVALID_MODE | 3031 NFS_INO_INVALID_OTHER); 3032 if (res == 0) 3033 res = generic_permission(&init_user_ns, inode, mask); 3034 goto out; 3035 } 3036 EXPORT_SYMBOL_GPL(nfs_permission); 3037