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, const struct cred *cred) 72 { 73 struct nfs_inode *nfsi = NFS_I(dir); 74 struct nfs_open_dir_context *ctx; 75 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL); 76 if (ctx != NULL) { 77 ctx->duped = 0; 78 ctx->attr_gencount = nfsi->attr_gencount; 79 ctx->dir_cookie = 0; 80 ctx->dup_cookie = 0; 81 ctx->cred = get_cred(cred); 82 spin_lock(&dir->i_lock); 83 if (list_empty(&nfsi->open_files) && 84 (nfsi->cache_validity & NFS_INO_DATA_INVAL_DEFER)) 85 nfsi->cache_validity |= 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 put_cred(ctx->cred); 100 kfree(ctx); 101 } 102 103 /* 104 * Open file 105 */ 106 static int 107 nfs_opendir(struct inode *inode, struct file *filp) 108 { 109 int res = 0; 110 struct nfs_open_dir_context *ctx; 111 112 dfprintk(FILE, "NFS: open dir(%pD2)\n", filp); 113 114 nfs_inc_stats(inode, NFSIOS_VFSOPEN); 115 116 ctx = alloc_nfs_open_dir_context(inode, current_cred()); 117 if (IS_ERR(ctx)) { 118 res = PTR_ERR(ctx); 119 goto out; 120 } 121 filp->private_data = ctx; 122 out: 123 return res; 124 } 125 126 static int 127 nfs_closedir(struct inode *inode, struct file *filp) 128 { 129 put_nfs_open_dir_context(file_inode(filp), filp->private_data); 130 return 0; 131 } 132 133 struct nfs_cache_array_entry { 134 u64 cookie; 135 u64 ino; 136 struct qstr string; 137 unsigned char d_type; 138 }; 139 140 struct nfs_cache_array { 141 int size; 142 int eof_index; 143 u64 last_cookie; 144 struct nfs_cache_array_entry array[]; 145 }; 146 147 typedef struct { 148 struct file *file; 149 struct page *page; 150 struct dir_context *ctx; 151 unsigned long page_index; 152 u64 *dir_cookie; 153 u64 last_cookie; 154 loff_t current_index; 155 loff_t prev_index; 156 157 unsigned long dir_verifier; 158 unsigned long timestamp; 159 unsigned long gencount; 160 unsigned int cache_entry_index; 161 bool plus; 162 bool eof; 163 } nfs_readdir_descriptor_t; 164 165 static 166 void nfs_readdir_init_array(struct page *page) 167 { 168 struct nfs_cache_array *array; 169 170 array = kmap_atomic(page); 171 memset(array, 0, sizeof(struct nfs_cache_array)); 172 array->eof_index = -1; 173 kunmap_atomic(array); 174 } 175 176 /* 177 * we are freeing strings created by nfs_add_to_readdir_array() 178 */ 179 static 180 void nfs_readdir_clear_array(struct page *page) 181 { 182 struct nfs_cache_array *array; 183 int i; 184 185 array = kmap_atomic(page); 186 for (i = 0; i < array->size; i++) 187 kfree(array->array[i].string.name); 188 array->size = 0; 189 kunmap_atomic(array); 190 } 191 192 /* 193 * the caller is responsible for freeing qstr.name 194 * when called by nfs_readdir_add_to_array, the strings will be freed in 195 * nfs_clear_readdir_array() 196 */ 197 static 198 int nfs_readdir_make_qstr(struct qstr *string, const char *name, unsigned int len) 199 { 200 string->len = len; 201 string->name = kmemdup_nul(name, len, GFP_KERNEL); 202 if (string->name == NULL) 203 return -ENOMEM; 204 /* 205 * Avoid a kmemleak false positive. The pointer to the name is stored 206 * in a page cache page which kmemleak does not scan. 207 */ 208 kmemleak_not_leak(string->name); 209 string->hash = full_name_hash(NULL, name, len); 210 return 0; 211 } 212 213 static 214 int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page) 215 { 216 struct nfs_cache_array *array = kmap(page); 217 struct nfs_cache_array_entry *cache_entry; 218 int ret; 219 220 cache_entry = &array->array[array->size]; 221 222 /* Check that this entry lies within the page bounds */ 223 ret = -ENOSPC; 224 if ((char *)&cache_entry[1] - (char *)page_address(page) > PAGE_SIZE) 225 goto out; 226 227 cache_entry->cookie = entry->prev_cookie; 228 cache_entry->ino = entry->ino; 229 cache_entry->d_type = entry->d_type; 230 ret = nfs_readdir_make_qstr(&cache_entry->string, entry->name, entry->len); 231 if (ret) 232 goto out; 233 array->last_cookie = entry->cookie; 234 array->size++; 235 if (entry->eof != 0) 236 array->eof_index = array->size; 237 out: 238 kunmap(page); 239 return ret; 240 } 241 242 static inline 243 int is_32bit_api(void) 244 { 245 #ifdef CONFIG_COMPAT 246 return in_compat_syscall(); 247 #else 248 return (BITS_PER_LONG == 32); 249 #endif 250 } 251 252 static 253 bool nfs_readdir_use_cookie(const struct file *filp) 254 { 255 if ((filp->f_mode & FMODE_32BITHASH) || 256 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api())) 257 return false; 258 return true; 259 } 260 261 static 262 int nfs_readdir_search_for_pos(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc) 263 { 264 loff_t diff = desc->ctx->pos - desc->current_index; 265 unsigned int index; 266 267 if (diff < 0) 268 goto out_eof; 269 if (diff >= array->size) { 270 if (array->eof_index >= 0) 271 goto out_eof; 272 return -EAGAIN; 273 } 274 275 index = (unsigned int)diff; 276 *desc->dir_cookie = array->array[index].cookie; 277 desc->cache_entry_index = index; 278 return 0; 279 out_eof: 280 desc->eof = true; 281 return -EBADCOOKIE; 282 } 283 284 static bool 285 nfs_readdir_inode_mapping_valid(struct nfs_inode *nfsi) 286 { 287 if (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA)) 288 return false; 289 smp_rmb(); 290 return !test_bit(NFS_INO_INVALIDATING, &nfsi->flags); 291 } 292 293 static 294 int nfs_readdir_search_for_cookie(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc) 295 { 296 int i; 297 loff_t new_pos; 298 int status = -EAGAIN; 299 300 for (i = 0; i < array->size; i++) { 301 if (array->array[i].cookie == *desc->dir_cookie) { 302 struct nfs_inode *nfsi = NFS_I(file_inode(desc->file)); 303 struct nfs_open_dir_context *ctx = desc->file->private_data; 304 305 new_pos = desc->current_index + i; 306 if (ctx->attr_gencount != nfsi->attr_gencount || 307 !nfs_readdir_inode_mapping_valid(nfsi)) { 308 ctx->duped = 0; 309 ctx->attr_gencount = nfsi->attr_gencount; 310 } else if (new_pos < desc->prev_index) { 311 if (ctx->duped > 0 312 && ctx->dup_cookie == *desc->dir_cookie) { 313 if (printk_ratelimit()) { 314 pr_notice("NFS: directory %pD2 contains a readdir loop." 315 "Please contact your server vendor. " 316 "The file: %.*s has duplicate cookie %llu\n", 317 desc->file, array->array[i].string.len, 318 array->array[i].string.name, *desc->dir_cookie); 319 } 320 status = -ELOOP; 321 goto out; 322 } 323 ctx->dup_cookie = *desc->dir_cookie; 324 ctx->duped = -1; 325 } 326 if (nfs_readdir_use_cookie(desc->file)) 327 desc->ctx->pos = *desc->dir_cookie; 328 else 329 desc->ctx->pos = new_pos; 330 desc->prev_index = new_pos; 331 desc->cache_entry_index = i; 332 return 0; 333 } 334 } 335 if (array->eof_index >= 0) { 336 status = -EBADCOOKIE; 337 if (*desc->dir_cookie == array->last_cookie) 338 desc->eof = true; 339 } 340 out: 341 return status; 342 } 343 344 static 345 int nfs_readdir_search_array(nfs_readdir_descriptor_t *desc) 346 { 347 struct nfs_cache_array *array; 348 int status; 349 350 array = kmap(desc->page); 351 352 if (*desc->dir_cookie == 0) 353 status = nfs_readdir_search_for_pos(array, desc); 354 else 355 status = nfs_readdir_search_for_cookie(array, desc); 356 357 if (status == -EAGAIN) { 358 desc->last_cookie = array->last_cookie; 359 desc->current_index += array->size; 360 desc->page_index++; 361 } 362 kunmap(desc->page); 363 return status; 364 } 365 366 /* Fill a page with xdr information before transferring to the cache page */ 367 static 368 int nfs_readdir_xdr_filler(struct page **pages, nfs_readdir_descriptor_t *desc, 369 struct nfs_entry *entry, struct file *file, struct inode *inode) 370 { 371 struct nfs_open_dir_context *ctx = file->private_data; 372 const struct cred *cred = ctx->cred; 373 unsigned long timestamp, gencount; 374 int error; 375 376 again: 377 timestamp = jiffies; 378 gencount = nfs_inc_attr_generation_counter(); 379 desc->dir_verifier = nfs_save_change_attribute(inode); 380 error = NFS_PROTO(inode)->readdir(file_dentry(file), cred, entry->cookie, pages, 381 NFS_SERVER(inode)->dtsize, desc->plus); 382 if (error < 0) { 383 /* We requested READDIRPLUS, but the server doesn't grok it */ 384 if (error == -ENOTSUPP && desc->plus) { 385 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS; 386 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags); 387 desc->plus = false; 388 goto again; 389 } 390 goto error; 391 } 392 desc->timestamp = timestamp; 393 desc->gencount = gencount; 394 error: 395 return error; 396 } 397 398 static int xdr_decode(nfs_readdir_descriptor_t *desc, 399 struct nfs_entry *entry, struct xdr_stream *xdr) 400 { 401 struct inode *inode = file_inode(desc->file); 402 int error; 403 404 error = NFS_PROTO(inode)->decode_dirent(xdr, entry, desc->plus); 405 if (error) 406 return error; 407 entry->fattr->time_start = desc->timestamp; 408 entry->fattr->gencount = desc->gencount; 409 return 0; 410 } 411 412 /* Match file and dirent using either filehandle or fileid 413 * Note: caller is responsible for checking the fsid 414 */ 415 static 416 int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry) 417 { 418 struct inode *inode; 419 struct nfs_inode *nfsi; 420 421 if (d_really_is_negative(dentry)) 422 return 0; 423 424 inode = d_inode(dentry); 425 if (is_bad_inode(inode) || NFS_STALE(inode)) 426 return 0; 427 428 nfsi = NFS_I(inode); 429 if (entry->fattr->fileid != nfsi->fileid) 430 return 0; 431 if (entry->fh->size && nfs_compare_fh(entry->fh, &nfsi->fh) != 0) 432 return 0; 433 return 1; 434 } 435 436 static 437 bool nfs_use_readdirplus(struct inode *dir, struct dir_context *ctx) 438 { 439 if (!nfs_server_capable(dir, NFS_CAP_READDIRPLUS)) 440 return false; 441 if (test_and_clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags)) 442 return true; 443 if (ctx->pos == 0) 444 return true; 445 return false; 446 } 447 448 /* 449 * This function is called by the lookup and getattr code to request the 450 * use of readdirplus to accelerate any future lookups in the same 451 * directory. 452 */ 453 void nfs_advise_use_readdirplus(struct inode *dir) 454 { 455 struct nfs_inode *nfsi = NFS_I(dir); 456 457 if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) && 458 !list_empty(&nfsi->open_files)) 459 set_bit(NFS_INO_ADVISE_RDPLUS, &nfsi->flags); 460 } 461 462 /* 463 * This function is mainly for use by nfs_getattr(). 464 * 465 * If this is an 'ls -l', we want to force use of readdirplus. 466 * Do this by checking if there is an active file descriptor 467 * and calling nfs_advise_use_readdirplus, then forcing a 468 * cache flush. 469 */ 470 void nfs_force_use_readdirplus(struct inode *dir) 471 { 472 struct nfs_inode *nfsi = NFS_I(dir); 473 474 if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) && 475 !list_empty(&nfsi->open_files)) { 476 set_bit(NFS_INO_ADVISE_RDPLUS, &nfsi->flags); 477 invalidate_mapping_pages(dir->i_mapping, 478 nfsi->page_index + 1, -1); 479 } 480 } 481 482 static 483 void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry, 484 unsigned long dir_verifier) 485 { 486 struct qstr filename = QSTR_INIT(entry->name, entry->len); 487 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 488 struct dentry *dentry; 489 struct dentry *alias; 490 struct inode *inode; 491 int status; 492 493 if (!(entry->fattr->valid & NFS_ATTR_FATTR_FILEID)) 494 return; 495 if (!(entry->fattr->valid & NFS_ATTR_FATTR_FSID)) 496 return; 497 if (filename.len == 0) 498 return; 499 /* Validate that the name doesn't contain any illegal '\0' */ 500 if (strnlen(filename.name, filename.len) != filename.len) 501 return; 502 /* ...or '/' */ 503 if (strnchr(filename.name, filename.len, '/')) 504 return; 505 if (filename.name[0] == '.') { 506 if (filename.len == 1) 507 return; 508 if (filename.len == 2 && filename.name[1] == '.') 509 return; 510 } 511 filename.hash = full_name_hash(parent, filename.name, filename.len); 512 513 dentry = d_lookup(parent, &filename); 514 again: 515 if (!dentry) { 516 dentry = d_alloc_parallel(parent, &filename, &wq); 517 if (IS_ERR(dentry)) 518 return; 519 } 520 if (!d_in_lookup(dentry)) { 521 /* Is there a mountpoint here? If so, just exit */ 522 if (!nfs_fsid_equal(&NFS_SB(dentry->d_sb)->fsid, 523 &entry->fattr->fsid)) 524 goto out; 525 if (nfs_same_file(dentry, entry)) { 526 if (!entry->fh->size) 527 goto out; 528 nfs_set_verifier(dentry, dir_verifier); 529 status = nfs_refresh_inode(d_inode(dentry), entry->fattr); 530 if (!status) 531 nfs_setsecurity(d_inode(dentry), entry->fattr, entry->label); 532 goto out; 533 } else { 534 d_invalidate(dentry); 535 dput(dentry); 536 dentry = NULL; 537 goto again; 538 } 539 } 540 if (!entry->fh->size) { 541 d_lookup_done(dentry); 542 goto out; 543 } 544 545 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr, entry->label); 546 alias = d_splice_alias(inode, dentry); 547 d_lookup_done(dentry); 548 if (alias) { 549 if (IS_ERR(alias)) 550 goto out; 551 dput(dentry); 552 dentry = alias; 553 } 554 nfs_set_verifier(dentry, dir_verifier); 555 out: 556 dput(dentry); 557 } 558 559 /* Perform conversion from xdr to cache array */ 560 static 561 int nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry, 562 struct page **xdr_pages, struct page *page, unsigned int buflen) 563 { 564 struct xdr_stream stream; 565 struct xdr_buf buf; 566 struct page *scratch; 567 struct nfs_cache_array *array; 568 unsigned int count = 0; 569 int status; 570 571 scratch = alloc_page(GFP_KERNEL); 572 if (scratch == NULL) 573 return -ENOMEM; 574 575 if (buflen == 0) 576 goto out_nopages; 577 578 xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen); 579 xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE); 580 581 do { 582 if (entry->label) 583 entry->label->len = NFS4_MAXLABELLEN; 584 585 status = xdr_decode(desc, entry, &stream); 586 if (status != 0) { 587 if (status == -EAGAIN) 588 status = 0; 589 break; 590 } 591 592 count++; 593 594 if (desc->plus) 595 nfs_prime_dcache(file_dentry(desc->file), entry, 596 desc->dir_verifier); 597 598 status = nfs_readdir_add_to_array(entry, page); 599 if (status != 0) 600 break; 601 } while (!entry->eof); 602 603 out_nopages: 604 if (count == 0 || (status == -EBADCOOKIE && entry->eof != 0)) { 605 array = kmap(page); 606 array->eof_index = array->size; 607 status = 0; 608 kunmap(page); 609 } 610 611 put_page(scratch); 612 return status; 613 } 614 615 static 616 void nfs_readdir_free_pages(struct page **pages, unsigned int npages) 617 { 618 unsigned int i; 619 for (i = 0; i < npages; i++) 620 put_page(pages[i]); 621 } 622 623 /* 624 * nfs_readdir_alloc_pages() will allocate pages that must be freed with a call 625 * to nfs_readdir_free_pages() 626 */ 627 static 628 int nfs_readdir_alloc_pages(struct page **pages, unsigned int npages) 629 { 630 unsigned int i; 631 632 for (i = 0; i < npages; i++) { 633 struct page *page = alloc_page(GFP_KERNEL); 634 if (page == NULL) 635 goto out_freepages; 636 pages[i] = page; 637 } 638 return 0; 639 640 out_freepages: 641 nfs_readdir_free_pages(pages, i); 642 return -ENOMEM; 643 } 644 645 static 646 int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t *desc, struct page *page, struct inode *inode) 647 { 648 struct page *pages[NFS_MAX_READDIR_PAGES]; 649 struct nfs_entry entry; 650 struct file *file = desc->file; 651 struct nfs_cache_array *array; 652 int status = -ENOMEM; 653 unsigned int array_size = ARRAY_SIZE(pages); 654 655 nfs_readdir_init_array(page); 656 657 entry.prev_cookie = 0; 658 entry.cookie = desc->last_cookie; 659 entry.eof = 0; 660 entry.fh = nfs_alloc_fhandle(); 661 entry.fattr = nfs_alloc_fattr(); 662 entry.server = NFS_SERVER(inode); 663 if (entry.fh == NULL || entry.fattr == NULL) 664 goto out; 665 666 entry.label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT); 667 if (IS_ERR(entry.label)) { 668 status = PTR_ERR(entry.label); 669 goto out; 670 } 671 672 array = kmap(page); 673 674 status = nfs_readdir_alloc_pages(pages, array_size); 675 if (status < 0) 676 goto out_release_array; 677 do { 678 unsigned int pglen; 679 status = nfs_readdir_xdr_filler(pages, desc, &entry, file, inode); 680 681 if (status < 0) 682 break; 683 pglen = status; 684 status = nfs_readdir_page_filler(desc, &entry, pages, page, pglen); 685 if (status < 0) { 686 if (status == -ENOSPC) 687 status = 0; 688 break; 689 } 690 } while (array->eof_index < 0); 691 692 nfs_readdir_free_pages(pages, array_size); 693 out_release_array: 694 kunmap(page); 695 nfs4_label_free(entry.label); 696 out: 697 nfs_free_fattr(entry.fattr); 698 nfs_free_fhandle(entry.fh); 699 return status; 700 } 701 702 /* 703 * Now we cache directories properly, by converting xdr information 704 * to an array that can be used for lookups later. This results in 705 * fewer cache pages, since we can store more information on each page. 706 * We only need to convert from xdr once so future lookups are much simpler 707 */ 708 static 709 int nfs_readdir_filler(void *data, struct page* page) 710 { 711 nfs_readdir_descriptor_t *desc = data; 712 struct inode *inode = file_inode(desc->file); 713 int ret; 714 715 ret = nfs_readdir_xdr_to_array(desc, page, inode); 716 if (ret < 0) 717 goto error; 718 SetPageUptodate(page); 719 720 if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) { 721 /* Should never happen */ 722 nfs_zap_mapping(inode, inode->i_mapping); 723 } 724 unlock_page(page); 725 return 0; 726 error: 727 nfs_readdir_clear_array(page); 728 unlock_page(page); 729 return ret; 730 } 731 732 static 733 void cache_page_release(nfs_readdir_descriptor_t *desc) 734 { 735 put_page(desc->page); 736 desc->page = NULL; 737 } 738 739 static 740 struct page *get_cache_page(nfs_readdir_descriptor_t *desc) 741 { 742 return read_cache_page(desc->file->f_mapping, desc->page_index, 743 nfs_readdir_filler, desc); 744 } 745 746 /* 747 * Returns 0 if desc->dir_cookie was found on page desc->page_index 748 * and locks the page to prevent removal from the page cache. 749 */ 750 static 751 int find_and_lock_cache_page(nfs_readdir_descriptor_t *desc) 752 { 753 struct inode *inode = file_inode(desc->file); 754 struct nfs_inode *nfsi = NFS_I(inode); 755 int res; 756 757 desc->page = get_cache_page(desc); 758 if (IS_ERR(desc->page)) 759 return PTR_ERR(desc->page); 760 res = lock_page_killable(desc->page); 761 if (res != 0) 762 goto error; 763 res = -EAGAIN; 764 if (desc->page->mapping != NULL) { 765 res = nfs_readdir_search_array(desc); 766 if (res == 0) { 767 nfsi->page_index = desc->page_index; 768 return 0; 769 } 770 } 771 unlock_page(desc->page); 772 error: 773 cache_page_release(desc); 774 return res; 775 } 776 777 /* Search for desc->dir_cookie from the beginning of the page cache */ 778 static inline 779 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc) 780 { 781 int res; 782 783 if (desc->page_index == 0) { 784 desc->current_index = 0; 785 desc->prev_index = 0; 786 desc->last_cookie = 0; 787 } 788 do { 789 res = find_and_lock_cache_page(desc); 790 } while (res == -EAGAIN); 791 return res; 792 } 793 794 /* 795 * Once we've found the start of the dirent within a page: fill 'er up... 796 */ 797 static 798 int nfs_do_filldir(nfs_readdir_descriptor_t *desc) 799 { 800 struct file *file = desc->file; 801 int i = 0; 802 int res = 0; 803 struct nfs_cache_array *array = NULL; 804 struct nfs_open_dir_context *ctx = file->private_data; 805 806 array = kmap(desc->page); 807 for (i = desc->cache_entry_index; i < array->size; i++) { 808 struct nfs_cache_array_entry *ent; 809 810 ent = &array->array[i]; 811 if (!dir_emit(desc->ctx, ent->string.name, ent->string.len, 812 nfs_compat_user_ino64(ent->ino), ent->d_type)) { 813 desc->eof = true; 814 break; 815 } 816 if (i < (array->size-1)) 817 *desc->dir_cookie = array->array[i+1].cookie; 818 else 819 *desc->dir_cookie = array->last_cookie; 820 if (nfs_readdir_use_cookie(file)) 821 desc->ctx->pos = *desc->dir_cookie; 822 else 823 desc->ctx->pos++; 824 if (ctx->duped != 0) 825 ctx->duped = 1; 826 } 827 if (array->eof_index >= 0) 828 desc->eof = true; 829 830 kunmap(desc->page); 831 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n", 832 (unsigned long long)*desc->dir_cookie, res); 833 return res; 834 } 835 836 /* 837 * If we cannot find a cookie in our cache, we suspect that this is 838 * because it points to a deleted file, so we ask the server to return 839 * whatever it thinks is the next entry. We then feed this to filldir. 840 * If all goes well, we should then be able to find our way round the 841 * cache on the next call to readdir_search_pagecache(); 842 * 843 * NOTE: we cannot add the anonymous page to the pagecache because 844 * the data it contains might not be page aligned. Besides, 845 * we should already have a complete representation of the 846 * directory in the page cache by the time we get here. 847 */ 848 static inline 849 int uncached_readdir(nfs_readdir_descriptor_t *desc) 850 { 851 struct page *page = NULL; 852 int status; 853 struct inode *inode = file_inode(desc->file); 854 struct nfs_open_dir_context *ctx = desc->file->private_data; 855 856 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n", 857 (unsigned long long)*desc->dir_cookie); 858 859 page = alloc_page(GFP_HIGHUSER); 860 if (!page) { 861 status = -ENOMEM; 862 goto out; 863 } 864 865 desc->page_index = 0; 866 desc->last_cookie = *desc->dir_cookie; 867 desc->page = page; 868 ctx->duped = 0; 869 870 status = nfs_readdir_xdr_to_array(desc, page, inode); 871 if (status < 0) 872 goto out_release; 873 874 status = nfs_do_filldir(desc); 875 876 out_release: 877 nfs_readdir_clear_array(desc->page); 878 cache_page_release(desc); 879 out: 880 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", 881 __func__, status); 882 return status; 883 } 884 885 /* The file offset position represents the dirent entry number. A 886 last cookie cache takes care of the common case of reading the 887 whole directory. 888 */ 889 static int nfs_readdir(struct file *file, struct dir_context *ctx) 890 { 891 struct dentry *dentry = file_dentry(file); 892 struct inode *inode = d_inode(dentry); 893 struct nfs_open_dir_context *dir_ctx = file->private_data; 894 nfs_readdir_descriptor_t my_desc = { 895 .file = file, 896 .ctx = ctx, 897 .dir_cookie = &dir_ctx->dir_cookie, 898 .plus = nfs_use_readdirplus(inode, ctx), 899 }, 900 *desc = &my_desc; 901 int res = 0; 902 903 dfprintk(FILE, "NFS: readdir(%pD2) starting at cookie %llu\n", 904 file, (long long)ctx->pos); 905 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS); 906 907 /* 908 * ctx->pos points to the dirent entry number. 909 * *desc->dir_cookie has the cookie for the next entry. We have 910 * to either find the entry with the appropriate number or 911 * revalidate the cookie. 912 */ 913 if (ctx->pos == 0 || nfs_attribute_cache_expired(inode)) 914 res = nfs_revalidate_mapping(inode, file->f_mapping); 915 if (res < 0) 916 goto out; 917 918 do { 919 res = readdir_search_pagecache(desc); 920 921 if (res == -EBADCOOKIE) { 922 res = 0; 923 /* This means either end of directory */ 924 if (*desc->dir_cookie && !desc->eof) { 925 /* Or that the server has 'lost' a cookie */ 926 res = uncached_readdir(desc); 927 if (res == 0) 928 continue; 929 } 930 break; 931 } 932 if (res == -ETOOSMALL && desc->plus) { 933 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags); 934 nfs_zap_caches(inode); 935 desc->page_index = 0; 936 desc->plus = false; 937 desc->eof = false; 938 continue; 939 } 940 if (res < 0) 941 break; 942 943 res = nfs_do_filldir(desc); 944 unlock_page(desc->page); 945 cache_page_release(desc); 946 if (res < 0) 947 break; 948 } while (!desc->eof); 949 out: 950 if (res > 0) 951 res = 0; 952 dfprintk(FILE, "NFS: readdir(%pD2) returns %d\n", file, res); 953 return res; 954 } 955 956 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int whence) 957 { 958 struct nfs_open_dir_context *dir_ctx = filp->private_data; 959 960 dfprintk(FILE, "NFS: llseek dir(%pD2, %lld, %d)\n", 961 filp, offset, whence); 962 963 switch (whence) { 964 default: 965 return -EINVAL; 966 case SEEK_SET: 967 if (offset < 0) 968 return -EINVAL; 969 spin_lock(&filp->f_lock); 970 break; 971 case SEEK_CUR: 972 if (offset == 0) 973 return filp->f_pos; 974 spin_lock(&filp->f_lock); 975 offset += filp->f_pos; 976 if (offset < 0) { 977 spin_unlock(&filp->f_lock); 978 return -EINVAL; 979 } 980 } 981 if (offset != filp->f_pos) { 982 filp->f_pos = offset; 983 if (nfs_readdir_use_cookie(filp)) 984 dir_ctx->dir_cookie = offset; 985 else 986 dir_ctx->dir_cookie = 0; 987 dir_ctx->duped = 0; 988 } 989 spin_unlock(&filp->f_lock); 990 return offset; 991 } 992 993 /* 994 * All directory operations under NFS are synchronous, so fsync() 995 * is a dummy operation. 996 */ 997 static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end, 998 int datasync) 999 { 1000 struct inode *inode = file_inode(filp); 1001 1002 dfprintk(FILE, "NFS: fsync dir(%pD2) datasync %d\n", filp, datasync); 1003 1004 inode_lock(inode); 1005 nfs_inc_stats(inode, NFSIOS_VFSFSYNC); 1006 inode_unlock(inode); 1007 return 0; 1008 } 1009 1010 /** 1011 * nfs_force_lookup_revalidate - Mark the directory as having changed 1012 * @dir: pointer to directory inode 1013 * 1014 * This forces the revalidation code in nfs_lookup_revalidate() to do a 1015 * full lookup on all child dentries of 'dir' whenever a change occurs 1016 * on the server that might have invalidated our dcache. 1017 * 1018 * Note that we reserve bit '0' as a tag to let us know when a dentry 1019 * was revalidated while holding a delegation on its inode. 1020 * 1021 * The caller should be holding dir->i_lock 1022 */ 1023 void nfs_force_lookup_revalidate(struct inode *dir) 1024 { 1025 NFS_I(dir)->cache_change_attribute += 2; 1026 } 1027 EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate); 1028 1029 /** 1030 * nfs_verify_change_attribute - Detects NFS remote directory changes 1031 * @dir: pointer to parent directory inode 1032 * @verf: previously saved change attribute 1033 * 1034 * Return "false" if the verifiers doesn't match the change attribute. 1035 * This would usually indicate that the directory contents have changed on 1036 * the server, and that any dentries need revalidating. 1037 */ 1038 static bool nfs_verify_change_attribute(struct inode *dir, unsigned long verf) 1039 { 1040 return (verf & ~1UL) == nfs_save_change_attribute(dir); 1041 } 1042 1043 static void nfs_set_verifier_delegated(unsigned long *verf) 1044 { 1045 *verf |= 1UL; 1046 } 1047 1048 #if IS_ENABLED(CONFIG_NFS_V4) 1049 static void nfs_unset_verifier_delegated(unsigned long *verf) 1050 { 1051 *verf &= ~1UL; 1052 } 1053 #endif /* IS_ENABLED(CONFIG_NFS_V4) */ 1054 1055 static bool nfs_test_verifier_delegated(unsigned long verf) 1056 { 1057 return verf & 1; 1058 } 1059 1060 static bool nfs_verifier_is_delegated(struct dentry *dentry) 1061 { 1062 return nfs_test_verifier_delegated(dentry->d_time); 1063 } 1064 1065 static void nfs_set_verifier_locked(struct dentry *dentry, unsigned long verf) 1066 { 1067 struct inode *inode = d_inode(dentry); 1068 1069 if (!nfs_verifier_is_delegated(dentry) && 1070 !nfs_verify_change_attribute(d_inode(dentry->d_parent), verf)) 1071 goto out; 1072 if (inode && NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) 1073 nfs_set_verifier_delegated(&verf); 1074 out: 1075 dentry->d_time = verf; 1076 } 1077 1078 /** 1079 * nfs_set_verifier - save a parent directory verifier in the dentry 1080 * @dentry: pointer to dentry 1081 * @verf: verifier to save 1082 * 1083 * Saves the parent directory verifier in @dentry. If the inode has 1084 * a delegation, we also tag the dentry as having been revalidated 1085 * while holding a delegation so that we know we don't have to 1086 * look it up again after a directory change. 1087 */ 1088 void nfs_set_verifier(struct dentry *dentry, unsigned long verf) 1089 { 1090 1091 spin_lock(&dentry->d_lock); 1092 nfs_set_verifier_locked(dentry, verf); 1093 spin_unlock(&dentry->d_lock); 1094 } 1095 EXPORT_SYMBOL_GPL(nfs_set_verifier); 1096 1097 #if IS_ENABLED(CONFIG_NFS_V4) 1098 /** 1099 * nfs_clear_verifier_delegated - clear the dir verifier delegation tag 1100 * @inode: pointer to inode 1101 * 1102 * Iterates through the dentries in the inode alias list and clears 1103 * the tag used to indicate that the dentry has been revalidated 1104 * while holding a delegation. 1105 * This function is intended for use when the delegation is being 1106 * returned or revoked. 1107 */ 1108 void nfs_clear_verifier_delegated(struct inode *inode) 1109 { 1110 struct dentry *alias; 1111 1112 if (!inode) 1113 return; 1114 spin_lock(&inode->i_lock); 1115 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) { 1116 spin_lock(&alias->d_lock); 1117 nfs_unset_verifier_delegated(&alias->d_time); 1118 spin_unlock(&alias->d_lock); 1119 } 1120 spin_unlock(&inode->i_lock); 1121 } 1122 EXPORT_SYMBOL_GPL(nfs_clear_verifier_delegated); 1123 #endif /* IS_ENABLED(CONFIG_NFS_V4) */ 1124 1125 /* 1126 * A check for whether or not the parent directory has changed. 1127 * In the case it has, we assume that the dentries are untrustworthy 1128 * and may need to be looked up again. 1129 * If rcu_walk prevents us from performing a full check, return 0. 1130 */ 1131 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry, 1132 int rcu_walk) 1133 { 1134 if (IS_ROOT(dentry)) 1135 return 1; 1136 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE) 1137 return 0; 1138 if (!nfs_verify_change_attribute(dir, dentry->d_time)) 1139 return 0; 1140 /* Revalidate nfsi->cache_change_attribute before we declare a match */ 1141 if (nfs_mapping_need_revalidate_inode(dir)) { 1142 if (rcu_walk) 1143 return 0; 1144 if (__nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0) 1145 return 0; 1146 } 1147 if (!nfs_verify_change_attribute(dir, dentry->d_time)) 1148 return 0; 1149 return 1; 1150 } 1151 1152 /* 1153 * Use intent information to check whether or not we're going to do 1154 * an O_EXCL create using this path component. 1155 */ 1156 static int nfs_is_exclusive_create(struct inode *dir, unsigned int flags) 1157 { 1158 if (NFS_PROTO(dir)->version == 2) 1159 return 0; 1160 return flags & LOOKUP_EXCL; 1161 } 1162 1163 /* 1164 * Inode and filehandle revalidation for lookups. 1165 * 1166 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL, 1167 * or if the intent information indicates that we're about to open this 1168 * particular file and the "nocto" mount flag is not set. 1169 * 1170 */ 1171 static 1172 int nfs_lookup_verify_inode(struct inode *inode, unsigned int flags) 1173 { 1174 struct nfs_server *server = NFS_SERVER(inode); 1175 int ret; 1176 1177 if (IS_AUTOMOUNT(inode)) 1178 return 0; 1179 1180 if (flags & LOOKUP_OPEN) { 1181 switch (inode->i_mode & S_IFMT) { 1182 case S_IFREG: 1183 /* A NFSv4 OPEN will revalidate later */ 1184 if (server->caps & NFS_CAP_ATOMIC_OPEN) 1185 goto out; 1186 fallthrough; 1187 case S_IFDIR: 1188 if (server->flags & NFS_MOUNT_NOCTO) 1189 break; 1190 /* NFS close-to-open cache consistency validation */ 1191 goto out_force; 1192 } 1193 } 1194 1195 /* VFS wants an on-the-wire revalidation */ 1196 if (flags & LOOKUP_REVAL) 1197 goto out_force; 1198 out: 1199 return (inode->i_nlink == 0) ? -ESTALE : 0; 1200 out_force: 1201 if (flags & LOOKUP_RCU) 1202 return -ECHILD; 1203 ret = __nfs_revalidate_inode(server, inode); 1204 if (ret != 0) 1205 return ret; 1206 goto out; 1207 } 1208 1209 /* 1210 * We judge how long we want to trust negative 1211 * dentries by looking at the parent inode mtime. 1212 * 1213 * If parent mtime has changed, we revalidate, else we wait for a 1214 * period corresponding to the parent's attribute cache timeout value. 1215 * 1216 * If LOOKUP_RCU prevents us from performing a full check, return 1 1217 * suggesting a reval is needed. 1218 * 1219 * Note that when creating a new file, or looking up a rename target, 1220 * then it shouldn't be necessary to revalidate a negative dentry. 1221 */ 1222 static inline 1223 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry, 1224 unsigned int flags) 1225 { 1226 if (flags & (LOOKUP_CREATE | LOOKUP_RENAME_TARGET)) 1227 return 0; 1228 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG) 1229 return 1; 1230 return !nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU); 1231 } 1232 1233 static int 1234 nfs_lookup_revalidate_done(struct inode *dir, struct dentry *dentry, 1235 struct inode *inode, int error) 1236 { 1237 switch (error) { 1238 case 1: 1239 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is valid\n", 1240 __func__, dentry); 1241 return 1; 1242 case 0: 1243 nfs_mark_for_revalidate(dir); 1244 if (inode && S_ISDIR(inode->i_mode)) { 1245 /* Purge readdir caches. */ 1246 nfs_zap_caches(inode); 1247 /* 1248 * We can't d_drop the root of a disconnected tree: 1249 * its d_hash is on the s_anon list and d_drop() would hide 1250 * it from shrink_dcache_for_unmount(), leading to busy 1251 * inodes on unmount and further oopses. 1252 */ 1253 if (IS_ROOT(dentry)) 1254 return 1; 1255 } 1256 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is invalid\n", 1257 __func__, dentry); 1258 return 0; 1259 } 1260 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) lookup returned error %d\n", 1261 __func__, dentry, error); 1262 return error; 1263 } 1264 1265 static int 1266 nfs_lookup_revalidate_negative(struct inode *dir, struct dentry *dentry, 1267 unsigned int flags) 1268 { 1269 int ret = 1; 1270 if (nfs_neg_need_reval(dir, dentry, flags)) { 1271 if (flags & LOOKUP_RCU) 1272 return -ECHILD; 1273 ret = 0; 1274 } 1275 return nfs_lookup_revalidate_done(dir, dentry, NULL, ret); 1276 } 1277 1278 static int 1279 nfs_lookup_revalidate_delegated(struct inode *dir, struct dentry *dentry, 1280 struct inode *inode) 1281 { 1282 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1283 return nfs_lookup_revalidate_done(dir, dentry, inode, 1); 1284 } 1285 1286 static int 1287 nfs_lookup_revalidate_dentry(struct inode *dir, struct dentry *dentry, 1288 struct inode *inode) 1289 { 1290 struct nfs_fh *fhandle; 1291 struct nfs_fattr *fattr; 1292 struct nfs4_label *label; 1293 unsigned long dir_verifier; 1294 int ret; 1295 1296 ret = -ENOMEM; 1297 fhandle = nfs_alloc_fhandle(); 1298 fattr = nfs_alloc_fattr(); 1299 label = nfs4_label_alloc(NFS_SERVER(inode), GFP_KERNEL); 1300 if (fhandle == NULL || fattr == NULL || IS_ERR(label)) 1301 goto out; 1302 1303 dir_verifier = nfs_save_change_attribute(dir); 1304 ret = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr, label); 1305 if (ret < 0) { 1306 switch (ret) { 1307 case -ESTALE: 1308 case -ENOENT: 1309 ret = 0; 1310 break; 1311 case -ETIMEDOUT: 1312 if (NFS_SERVER(inode)->flags & NFS_MOUNT_SOFTREVAL) 1313 ret = 1; 1314 } 1315 goto out; 1316 } 1317 ret = 0; 1318 if (nfs_compare_fh(NFS_FH(inode), fhandle)) 1319 goto out; 1320 if (nfs_refresh_inode(inode, fattr) < 0) 1321 goto out; 1322 1323 nfs_setsecurity(inode, fattr, label); 1324 nfs_set_verifier(dentry, dir_verifier); 1325 1326 /* set a readdirplus hint that we had a cache miss */ 1327 nfs_force_use_readdirplus(dir); 1328 ret = 1; 1329 out: 1330 nfs_free_fattr(fattr); 1331 nfs_free_fhandle(fhandle); 1332 nfs4_label_free(label); 1333 return nfs_lookup_revalidate_done(dir, dentry, inode, ret); 1334 } 1335 1336 /* 1337 * This is called every time the dcache has a lookup hit, 1338 * and we should check whether we can really trust that 1339 * lookup. 1340 * 1341 * NOTE! The hit can be a negative hit too, don't assume 1342 * we have an inode! 1343 * 1344 * If the parent directory is seen to have changed, we throw out the 1345 * cached dentry and do a new lookup. 1346 */ 1347 static int 1348 nfs_do_lookup_revalidate(struct inode *dir, struct dentry *dentry, 1349 unsigned int flags) 1350 { 1351 struct inode *inode; 1352 int error; 1353 1354 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE); 1355 inode = d_inode(dentry); 1356 1357 if (!inode) 1358 return nfs_lookup_revalidate_negative(dir, dentry, flags); 1359 1360 if (is_bad_inode(inode)) { 1361 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n", 1362 __func__, dentry); 1363 goto out_bad; 1364 } 1365 1366 if (nfs_verifier_is_delegated(dentry)) 1367 return nfs_lookup_revalidate_delegated(dir, dentry, inode); 1368 1369 /* Force a full look up iff the parent directory has changed */ 1370 if (!(flags & (LOOKUP_EXCL | LOOKUP_REVAL)) && 1371 nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU)) { 1372 error = nfs_lookup_verify_inode(inode, flags); 1373 if (error) { 1374 if (error == -ESTALE) 1375 nfs_zap_caches(dir); 1376 goto out_bad; 1377 } 1378 nfs_advise_use_readdirplus(dir); 1379 goto out_valid; 1380 } 1381 1382 if (flags & LOOKUP_RCU) 1383 return -ECHILD; 1384 1385 if (NFS_STALE(inode)) 1386 goto out_bad; 1387 1388 trace_nfs_lookup_revalidate_enter(dir, dentry, flags); 1389 error = nfs_lookup_revalidate_dentry(dir, dentry, inode); 1390 trace_nfs_lookup_revalidate_exit(dir, dentry, flags, error); 1391 return error; 1392 out_valid: 1393 return nfs_lookup_revalidate_done(dir, dentry, inode, 1); 1394 out_bad: 1395 if (flags & LOOKUP_RCU) 1396 return -ECHILD; 1397 return nfs_lookup_revalidate_done(dir, dentry, inode, 0); 1398 } 1399 1400 static int 1401 __nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags, 1402 int (*reval)(struct inode *, struct dentry *, unsigned int)) 1403 { 1404 struct dentry *parent; 1405 struct inode *dir; 1406 int ret; 1407 1408 if (flags & LOOKUP_RCU) { 1409 parent = READ_ONCE(dentry->d_parent); 1410 dir = d_inode_rcu(parent); 1411 if (!dir) 1412 return -ECHILD; 1413 ret = reval(dir, dentry, flags); 1414 if (parent != READ_ONCE(dentry->d_parent)) 1415 return -ECHILD; 1416 } else { 1417 parent = dget_parent(dentry); 1418 ret = reval(d_inode(parent), dentry, flags); 1419 dput(parent); 1420 } 1421 return ret; 1422 } 1423 1424 static int nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags) 1425 { 1426 return __nfs_lookup_revalidate(dentry, flags, nfs_do_lookup_revalidate); 1427 } 1428 1429 /* 1430 * A weaker form of d_revalidate for revalidating just the d_inode(dentry) 1431 * when we don't really care about the dentry name. This is called when a 1432 * pathwalk ends on a dentry that was not found via a normal lookup in the 1433 * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals). 1434 * 1435 * In this situation, we just want to verify that the inode itself is OK 1436 * since the dentry might have changed on the server. 1437 */ 1438 static int nfs_weak_revalidate(struct dentry *dentry, unsigned int flags) 1439 { 1440 struct inode *inode = d_inode(dentry); 1441 int error = 0; 1442 1443 /* 1444 * I believe we can only get a negative dentry here in the case of a 1445 * procfs-style symlink. Just assume it's correct for now, but we may 1446 * eventually need to do something more here. 1447 */ 1448 if (!inode) { 1449 dfprintk(LOOKUPCACHE, "%s: %pd2 has negative inode\n", 1450 __func__, dentry); 1451 return 1; 1452 } 1453 1454 if (is_bad_inode(inode)) { 1455 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n", 1456 __func__, dentry); 1457 return 0; 1458 } 1459 1460 error = nfs_lookup_verify_inode(inode, flags); 1461 dfprintk(LOOKUPCACHE, "NFS: %s: inode %lu is %s\n", 1462 __func__, inode->i_ino, error ? "invalid" : "valid"); 1463 return !error; 1464 } 1465 1466 /* 1467 * This is called from dput() when d_count is going to 0. 1468 */ 1469 static int nfs_dentry_delete(const struct dentry *dentry) 1470 { 1471 dfprintk(VFS, "NFS: dentry_delete(%pd2, %x)\n", 1472 dentry, dentry->d_flags); 1473 1474 /* Unhash any dentry with a stale inode */ 1475 if (d_really_is_positive(dentry) && NFS_STALE(d_inode(dentry))) 1476 return 1; 1477 1478 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { 1479 /* Unhash it, so that ->d_iput() would be called */ 1480 return 1; 1481 } 1482 if (!(dentry->d_sb->s_flags & SB_ACTIVE)) { 1483 /* Unhash it, so that ancestors of killed async unlink 1484 * files will be cleaned up during umount */ 1485 return 1; 1486 } 1487 return 0; 1488 1489 } 1490 1491 /* Ensure that we revalidate inode->i_nlink */ 1492 static void nfs_drop_nlink(struct inode *inode) 1493 { 1494 spin_lock(&inode->i_lock); 1495 /* drop the inode if we're reasonably sure this is the last link */ 1496 if (inode->i_nlink > 0) 1497 drop_nlink(inode); 1498 NFS_I(inode)->attr_gencount = nfs_inc_attr_generation_counter(); 1499 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_CHANGE 1500 | NFS_INO_INVALID_CTIME 1501 | NFS_INO_INVALID_OTHER 1502 | NFS_INO_REVAL_FORCED; 1503 spin_unlock(&inode->i_lock); 1504 } 1505 1506 /* 1507 * Called when the dentry loses inode. 1508 * We use it to clean up silly-renamed files. 1509 */ 1510 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode) 1511 { 1512 if (S_ISDIR(inode->i_mode)) 1513 /* drop any readdir cache as it could easily be old */ 1514 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA; 1515 1516 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { 1517 nfs_complete_unlink(dentry, inode); 1518 nfs_drop_nlink(inode); 1519 } 1520 iput(inode); 1521 } 1522 1523 static void nfs_d_release(struct dentry *dentry) 1524 { 1525 /* free cached devname value, if it survived that far */ 1526 if (unlikely(dentry->d_fsdata)) { 1527 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) 1528 WARN_ON(1); 1529 else 1530 kfree(dentry->d_fsdata); 1531 } 1532 } 1533 1534 const struct dentry_operations nfs_dentry_operations = { 1535 .d_revalidate = nfs_lookup_revalidate, 1536 .d_weak_revalidate = nfs_weak_revalidate, 1537 .d_delete = nfs_dentry_delete, 1538 .d_iput = nfs_dentry_iput, 1539 .d_automount = nfs_d_automount, 1540 .d_release = nfs_d_release, 1541 }; 1542 EXPORT_SYMBOL_GPL(nfs_dentry_operations); 1543 1544 struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags) 1545 { 1546 struct dentry *res; 1547 struct inode *inode = NULL; 1548 struct nfs_fh *fhandle = NULL; 1549 struct nfs_fattr *fattr = NULL; 1550 struct nfs4_label *label = NULL; 1551 unsigned long dir_verifier; 1552 int error; 1553 1554 dfprintk(VFS, "NFS: lookup(%pd2)\n", dentry); 1555 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP); 1556 1557 if (unlikely(dentry->d_name.len > NFS_SERVER(dir)->namelen)) 1558 return ERR_PTR(-ENAMETOOLONG); 1559 1560 /* 1561 * If we're doing an exclusive create, optimize away the lookup 1562 * but don't hash the dentry. 1563 */ 1564 if (nfs_is_exclusive_create(dir, flags) || flags & LOOKUP_RENAME_TARGET) 1565 return NULL; 1566 1567 res = ERR_PTR(-ENOMEM); 1568 fhandle = nfs_alloc_fhandle(); 1569 fattr = nfs_alloc_fattr(); 1570 if (fhandle == NULL || fattr == NULL) 1571 goto out; 1572 1573 label = nfs4_label_alloc(NFS_SERVER(dir), GFP_NOWAIT); 1574 if (IS_ERR(label)) 1575 goto out; 1576 1577 dir_verifier = nfs_save_change_attribute(dir); 1578 trace_nfs_lookup_enter(dir, dentry, flags); 1579 error = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr, label); 1580 if (error == -ENOENT) 1581 goto no_entry; 1582 if (error < 0) { 1583 res = ERR_PTR(error); 1584 goto out_label; 1585 } 1586 inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label); 1587 res = ERR_CAST(inode); 1588 if (IS_ERR(res)) 1589 goto out_label; 1590 1591 /* Notify readdir to use READDIRPLUS */ 1592 nfs_force_use_readdirplus(dir); 1593 1594 no_entry: 1595 res = d_splice_alias(inode, dentry); 1596 if (res != NULL) { 1597 if (IS_ERR(res)) 1598 goto out_label; 1599 dentry = res; 1600 } 1601 nfs_set_verifier(dentry, dir_verifier); 1602 out_label: 1603 trace_nfs_lookup_exit(dir, dentry, flags, error); 1604 nfs4_label_free(label); 1605 out: 1606 nfs_free_fattr(fattr); 1607 nfs_free_fhandle(fhandle); 1608 return res; 1609 } 1610 EXPORT_SYMBOL_GPL(nfs_lookup); 1611 1612 #if IS_ENABLED(CONFIG_NFS_V4) 1613 static int nfs4_lookup_revalidate(struct dentry *, unsigned int); 1614 1615 const struct dentry_operations nfs4_dentry_operations = { 1616 .d_revalidate = nfs4_lookup_revalidate, 1617 .d_weak_revalidate = nfs_weak_revalidate, 1618 .d_delete = nfs_dentry_delete, 1619 .d_iput = nfs_dentry_iput, 1620 .d_automount = nfs_d_automount, 1621 .d_release = nfs_d_release, 1622 }; 1623 EXPORT_SYMBOL_GPL(nfs4_dentry_operations); 1624 1625 static fmode_t flags_to_mode(int flags) 1626 { 1627 fmode_t res = (__force fmode_t)flags & FMODE_EXEC; 1628 if ((flags & O_ACCMODE) != O_WRONLY) 1629 res |= FMODE_READ; 1630 if ((flags & O_ACCMODE) != O_RDONLY) 1631 res |= FMODE_WRITE; 1632 return res; 1633 } 1634 1635 static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags, struct file *filp) 1636 { 1637 return alloc_nfs_open_context(dentry, flags_to_mode(open_flags), filp); 1638 } 1639 1640 static int do_open(struct inode *inode, struct file *filp) 1641 { 1642 nfs_fscache_open_file(inode, filp); 1643 return 0; 1644 } 1645 1646 static int nfs_finish_open(struct nfs_open_context *ctx, 1647 struct dentry *dentry, 1648 struct file *file, unsigned open_flags) 1649 { 1650 int err; 1651 1652 err = finish_open(file, dentry, do_open); 1653 if (err) 1654 goto out; 1655 if (S_ISREG(file->f_path.dentry->d_inode->i_mode)) 1656 nfs_file_set_open_context(file, ctx); 1657 else 1658 err = -EOPENSTALE; 1659 out: 1660 return err; 1661 } 1662 1663 int nfs_atomic_open(struct inode *dir, struct dentry *dentry, 1664 struct file *file, unsigned open_flags, 1665 umode_t mode) 1666 { 1667 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 1668 struct nfs_open_context *ctx; 1669 struct dentry *res; 1670 struct iattr attr = { .ia_valid = ATTR_OPEN }; 1671 struct inode *inode; 1672 unsigned int lookup_flags = 0; 1673 bool switched = false; 1674 int created = 0; 1675 int err; 1676 1677 /* Expect a negative dentry */ 1678 BUG_ON(d_inode(dentry)); 1679 1680 dfprintk(VFS, "NFS: atomic_open(%s/%lu), %pd\n", 1681 dir->i_sb->s_id, dir->i_ino, dentry); 1682 1683 err = nfs_check_flags(open_flags); 1684 if (err) 1685 return err; 1686 1687 /* NFS only supports OPEN on regular files */ 1688 if ((open_flags & O_DIRECTORY)) { 1689 if (!d_in_lookup(dentry)) { 1690 /* 1691 * Hashed negative dentry with O_DIRECTORY: dentry was 1692 * revalidated and is fine, no need to perform lookup 1693 * again 1694 */ 1695 return -ENOENT; 1696 } 1697 lookup_flags = LOOKUP_OPEN|LOOKUP_DIRECTORY; 1698 goto no_open; 1699 } 1700 1701 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) 1702 return -ENAMETOOLONG; 1703 1704 if (open_flags & O_CREAT) { 1705 struct nfs_server *server = NFS_SERVER(dir); 1706 1707 if (!(server->attr_bitmask[2] & FATTR4_WORD2_MODE_UMASK)) 1708 mode &= ~current_umask(); 1709 1710 attr.ia_valid |= ATTR_MODE; 1711 attr.ia_mode = mode; 1712 } 1713 if (open_flags & O_TRUNC) { 1714 attr.ia_valid |= ATTR_SIZE; 1715 attr.ia_size = 0; 1716 } 1717 1718 if (!(open_flags & O_CREAT) && !d_in_lookup(dentry)) { 1719 d_drop(dentry); 1720 switched = true; 1721 dentry = d_alloc_parallel(dentry->d_parent, 1722 &dentry->d_name, &wq); 1723 if (IS_ERR(dentry)) 1724 return PTR_ERR(dentry); 1725 if (unlikely(!d_in_lookup(dentry))) 1726 return finish_no_open(file, dentry); 1727 } 1728 1729 ctx = create_nfs_open_context(dentry, open_flags, file); 1730 err = PTR_ERR(ctx); 1731 if (IS_ERR(ctx)) 1732 goto out; 1733 1734 trace_nfs_atomic_open_enter(dir, ctx, open_flags); 1735 inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr, &created); 1736 if (created) 1737 file->f_mode |= FMODE_CREATED; 1738 if (IS_ERR(inode)) { 1739 err = PTR_ERR(inode); 1740 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err); 1741 put_nfs_open_context(ctx); 1742 d_drop(dentry); 1743 switch (err) { 1744 case -ENOENT: 1745 d_splice_alias(NULL, dentry); 1746 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1747 break; 1748 case -EISDIR: 1749 case -ENOTDIR: 1750 goto no_open; 1751 case -ELOOP: 1752 if (!(open_flags & O_NOFOLLOW)) 1753 goto no_open; 1754 break; 1755 /* case -EINVAL: */ 1756 default: 1757 break; 1758 } 1759 goto out; 1760 } 1761 1762 err = nfs_finish_open(ctx, ctx->dentry, file, open_flags); 1763 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err); 1764 put_nfs_open_context(ctx); 1765 out: 1766 if (unlikely(switched)) { 1767 d_lookup_done(dentry); 1768 dput(dentry); 1769 } 1770 return err; 1771 1772 no_open: 1773 res = nfs_lookup(dir, dentry, lookup_flags); 1774 if (switched) { 1775 d_lookup_done(dentry); 1776 if (!res) 1777 res = dentry; 1778 else 1779 dput(dentry); 1780 } 1781 if (IS_ERR(res)) 1782 return PTR_ERR(res); 1783 return finish_no_open(file, res); 1784 } 1785 EXPORT_SYMBOL_GPL(nfs_atomic_open); 1786 1787 static int 1788 nfs4_do_lookup_revalidate(struct inode *dir, struct dentry *dentry, 1789 unsigned int flags) 1790 { 1791 struct inode *inode; 1792 1793 if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY)) 1794 goto full_reval; 1795 if (d_mountpoint(dentry)) 1796 goto full_reval; 1797 1798 inode = d_inode(dentry); 1799 1800 /* We can't create new files in nfs_open_revalidate(), so we 1801 * optimize away revalidation of negative dentries. 1802 */ 1803 if (inode == NULL) 1804 goto full_reval; 1805 1806 if (nfs_verifier_is_delegated(dentry)) 1807 return nfs_lookup_revalidate_delegated(dir, dentry, inode); 1808 1809 /* NFS only supports OPEN on regular files */ 1810 if (!S_ISREG(inode->i_mode)) 1811 goto full_reval; 1812 1813 /* We cannot do exclusive creation on a positive dentry */ 1814 if (flags & (LOOKUP_EXCL | LOOKUP_REVAL)) 1815 goto reval_dentry; 1816 1817 /* Check if the directory changed */ 1818 if (!nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU)) 1819 goto reval_dentry; 1820 1821 /* Let f_op->open() actually open (and revalidate) the file */ 1822 return 1; 1823 reval_dentry: 1824 if (flags & LOOKUP_RCU) 1825 return -ECHILD; 1826 return nfs_lookup_revalidate_dentry(dir, dentry, inode); 1827 1828 full_reval: 1829 return nfs_do_lookup_revalidate(dir, dentry, flags); 1830 } 1831 1832 static int nfs4_lookup_revalidate(struct dentry *dentry, unsigned int flags) 1833 { 1834 return __nfs_lookup_revalidate(dentry, flags, 1835 nfs4_do_lookup_revalidate); 1836 } 1837 1838 #endif /* CONFIG_NFSV4 */ 1839 1840 struct dentry * 1841 nfs_add_or_obtain(struct dentry *dentry, struct nfs_fh *fhandle, 1842 struct nfs_fattr *fattr, 1843 struct nfs4_label *label) 1844 { 1845 struct dentry *parent = dget_parent(dentry); 1846 struct inode *dir = d_inode(parent); 1847 struct inode *inode; 1848 struct dentry *d; 1849 int error; 1850 1851 d_drop(dentry); 1852 1853 if (fhandle->size == 0) { 1854 error = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr, NULL); 1855 if (error) 1856 goto out_error; 1857 } 1858 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1859 if (!(fattr->valid & NFS_ATTR_FATTR)) { 1860 struct nfs_server *server = NFS_SB(dentry->d_sb); 1861 error = server->nfs_client->rpc_ops->getattr(server, fhandle, 1862 fattr, NULL, NULL); 1863 if (error < 0) 1864 goto out_error; 1865 } 1866 inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label); 1867 d = d_splice_alias(inode, dentry); 1868 out: 1869 dput(parent); 1870 return d; 1871 out_error: 1872 nfs_mark_for_revalidate(dir); 1873 d = ERR_PTR(error); 1874 goto out; 1875 } 1876 EXPORT_SYMBOL_GPL(nfs_add_or_obtain); 1877 1878 /* 1879 * Code common to create, mkdir, and mknod. 1880 */ 1881 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle, 1882 struct nfs_fattr *fattr, 1883 struct nfs4_label *label) 1884 { 1885 struct dentry *d; 1886 1887 d = nfs_add_or_obtain(dentry, fhandle, fattr, label); 1888 if (IS_ERR(d)) 1889 return PTR_ERR(d); 1890 1891 /* Callers don't care */ 1892 dput(d); 1893 return 0; 1894 } 1895 EXPORT_SYMBOL_GPL(nfs_instantiate); 1896 1897 /* 1898 * Following a failed create operation, we drop the dentry rather 1899 * than retain a negative dentry. This avoids a problem in the event 1900 * that the operation succeeded on the server, but an error in the 1901 * reply path made it appear to have failed. 1902 */ 1903 int nfs_create(struct inode *dir, struct dentry *dentry, 1904 umode_t mode, bool excl) 1905 { 1906 struct iattr attr; 1907 int open_flags = excl ? O_CREAT | O_EXCL : O_CREAT; 1908 int error; 1909 1910 dfprintk(VFS, "NFS: create(%s/%lu), %pd\n", 1911 dir->i_sb->s_id, dir->i_ino, dentry); 1912 1913 attr.ia_mode = mode; 1914 attr.ia_valid = ATTR_MODE; 1915 1916 trace_nfs_create_enter(dir, dentry, open_flags); 1917 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags); 1918 trace_nfs_create_exit(dir, dentry, open_flags, error); 1919 if (error != 0) 1920 goto out_err; 1921 return 0; 1922 out_err: 1923 d_drop(dentry); 1924 return error; 1925 } 1926 EXPORT_SYMBOL_GPL(nfs_create); 1927 1928 /* 1929 * See comments for nfs_proc_create regarding failed operations. 1930 */ 1931 int 1932 nfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t rdev) 1933 { 1934 struct iattr attr; 1935 int status; 1936 1937 dfprintk(VFS, "NFS: mknod(%s/%lu), %pd\n", 1938 dir->i_sb->s_id, dir->i_ino, dentry); 1939 1940 attr.ia_mode = mode; 1941 attr.ia_valid = ATTR_MODE; 1942 1943 trace_nfs_mknod_enter(dir, dentry); 1944 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev); 1945 trace_nfs_mknod_exit(dir, dentry, status); 1946 if (status != 0) 1947 goto out_err; 1948 return 0; 1949 out_err: 1950 d_drop(dentry); 1951 return status; 1952 } 1953 EXPORT_SYMBOL_GPL(nfs_mknod); 1954 1955 /* 1956 * See comments for nfs_proc_create regarding failed operations. 1957 */ 1958 int nfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) 1959 { 1960 struct iattr attr; 1961 int error; 1962 1963 dfprintk(VFS, "NFS: mkdir(%s/%lu), %pd\n", 1964 dir->i_sb->s_id, dir->i_ino, dentry); 1965 1966 attr.ia_valid = ATTR_MODE; 1967 attr.ia_mode = mode | S_IFDIR; 1968 1969 trace_nfs_mkdir_enter(dir, dentry); 1970 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr); 1971 trace_nfs_mkdir_exit(dir, dentry, error); 1972 if (error != 0) 1973 goto out_err; 1974 return 0; 1975 out_err: 1976 d_drop(dentry); 1977 return error; 1978 } 1979 EXPORT_SYMBOL_GPL(nfs_mkdir); 1980 1981 static void nfs_dentry_handle_enoent(struct dentry *dentry) 1982 { 1983 if (simple_positive(dentry)) 1984 d_delete(dentry); 1985 } 1986 1987 int nfs_rmdir(struct inode *dir, struct dentry *dentry) 1988 { 1989 int error; 1990 1991 dfprintk(VFS, "NFS: rmdir(%s/%lu), %pd\n", 1992 dir->i_sb->s_id, dir->i_ino, dentry); 1993 1994 trace_nfs_rmdir_enter(dir, dentry); 1995 if (d_really_is_positive(dentry)) { 1996 down_write(&NFS_I(d_inode(dentry))->rmdir_sem); 1997 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name); 1998 /* Ensure the VFS deletes this inode */ 1999 switch (error) { 2000 case 0: 2001 clear_nlink(d_inode(dentry)); 2002 break; 2003 case -ENOENT: 2004 nfs_dentry_handle_enoent(dentry); 2005 } 2006 up_write(&NFS_I(d_inode(dentry))->rmdir_sem); 2007 } else 2008 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name); 2009 trace_nfs_rmdir_exit(dir, dentry, error); 2010 2011 return error; 2012 } 2013 EXPORT_SYMBOL_GPL(nfs_rmdir); 2014 2015 /* 2016 * Remove a file after making sure there are no pending writes, 2017 * and after checking that the file has only one user. 2018 * 2019 * We invalidate the attribute cache and free the inode prior to the operation 2020 * to avoid possible races if the server reuses the inode. 2021 */ 2022 static int nfs_safe_remove(struct dentry *dentry) 2023 { 2024 struct inode *dir = d_inode(dentry->d_parent); 2025 struct inode *inode = d_inode(dentry); 2026 int error = -EBUSY; 2027 2028 dfprintk(VFS, "NFS: safe_remove(%pd2)\n", dentry); 2029 2030 /* If the dentry was sillyrenamed, we simply call d_delete() */ 2031 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { 2032 error = 0; 2033 goto out; 2034 } 2035 2036 trace_nfs_remove_enter(dir, dentry); 2037 if (inode != NULL) { 2038 error = NFS_PROTO(dir)->remove(dir, dentry); 2039 if (error == 0) 2040 nfs_drop_nlink(inode); 2041 } else 2042 error = NFS_PROTO(dir)->remove(dir, dentry); 2043 if (error == -ENOENT) 2044 nfs_dentry_handle_enoent(dentry); 2045 trace_nfs_remove_exit(dir, dentry, error); 2046 out: 2047 return error; 2048 } 2049 2050 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode 2051 * belongs to an active ".nfs..." file and we return -EBUSY. 2052 * 2053 * If sillyrename() returns 0, we do nothing, otherwise we unlink. 2054 */ 2055 int nfs_unlink(struct inode *dir, struct dentry *dentry) 2056 { 2057 int error; 2058 int need_rehash = 0; 2059 2060 dfprintk(VFS, "NFS: unlink(%s/%lu, %pd)\n", dir->i_sb->s_id, 2061 dir->i_ino, dentry); 2062 2063 trace_nfs_unlink_enter(dir, dentry); 2064 spin_lock(&dentry->d_lock); 2065 if (d_count(dentry) > 1) { 2066 spin_unlock(&dentry->d_lock); 2067 /* Start asynchronous writeout of the inode */ 2068 write_inode_now(d_inode(dentry), 0); 2069 error = nfs_sillyrename(dir, dentry); 2070 goto out; 2071 } 2072 if (!d_unhashed(dentry)) { 2073 __d_drop(dentry); 2074 need_rehash = 1; 2075 } 2076 spin_unlock(&dentry->d_lock); 2077 error = nfs_safe_remove(dentry); 2078 if (!error || error == -ENOENT) { 2079 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 2080 } else if (need_rehash) 2081 d_rehash(dentry); 2082 out: 2083 trace_nfs_unlink_exit(dir, dentry, error); 2084 return error; 2085 } 2086 EXPORT_SYMBOL_GPL(nfs_unlink); 2087 2088 /* 2089 * To create a symbolic link, most file systems instantiate a new inode, 2090 * add a page to it containing the path, then write it out to the disk 2091 * using prepare_write/commit_write. 2092 * 2093 * Unfortunately the NFS client can't create the in-core inode first 2094 * because it needs a file handle to create an in-core inode (see 2095 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the 2096 * symlink request has completed on the server. 2097 * 2098 * So instead we allocate a raw page, copy the symname into it, then do 2099 * the SYMLINK request with the page as the buffer. If it succeeds, we 2100 * now have a new file handle and can instantiate an in-core NFS inode 2101 * and move the raw page into its mapping. 2102 */ 2103 int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname) 2104 { 2105 struct page *page; 2106 char *kaddr; 2107 struct iattr attr; 2108 unsigned int pathlen = strlen(symname); 2109 int error; 2110 2111 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s)\n", dir->i_sb->s_id, 2112 dir->i_ino, dentry, symname); 2113 2114 if (pathlen > PAGE_SIZE) 2115 return -ENAMETOOLONG; 2116 2117 attr.ia_mode = S_IFLNK | S_IRWXUGO; 2118 attr.ia_valid = ATTR_MODE; 2119 2120 page = alloc_page(GFP_USER); 2121 if (!page) 2122 return -ENOMEM; 2123 2124 kaddr = page_address(page); 2125 memcpy(kaddr, symname, pathlen); 2126 if (pathlen < PAGE_SIZE) 2127 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen); 2128 2129 trace_nfs_symlink_enter(dir, dentry); 2130 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr); 2131 trace_nfs_symlink_exit(dir, dentry, error); 2132 if (error != 0) { 2133 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s) error %d\n", 2134 dir->i_sb->s_id, dir->i_ino, 2135 dentry, symname, error); 2136 d_drop(dentry); 2137 __free_page(page); 2138 return error; 2139 } 2140 2141 /* 2142 * No big deal if we can't add this page to the page cache here. 2143 * READLINK will get the missing page from the server if needed. 2144 */ 2145 if (!add_to_page_cache_lru(page, d_inode(dentry)->i_mapping, 0, 2146 GFP_KERNEL)) { 2147 SetPageUptodate(page); 2148 unlock_page(page); 2149 /* 2150 * add_to_page_cache_lru() grabs an extra page refcount. 2151 * Drop it here to avoid leaking this page later. 2152 */ 2153 put_page(page); 2154 } else 2155 __free_page(page); 2156 2157 return 0; 2158 } 2159 EXPORT_SYMBOL_GPL(nfs_symlink); 2160 2161 int 2162 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) 2163 { 2164 struct inode *inode = d_inode(old_dentry); 2165 int error; 2166 2167 dfprintk(VFS, "NFS: link(%pd2 -> %pd2)\n", 2168 old_dentry, dentry); 2169 2170 trace_nfs_link_enter(inode, dir, dentry); 2171 d_drop(dentry); 2172 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name); 2173 if (error == 0) { 2174 ihold(inode); 2175 d_add(dentry, inode); 2176 } 2177 trace_nfs_link_exit(inode, dir, dentry, error); 2178 return error; 2179 } 2180 EXPORT_SYMBOL_GPL(nfs_link); 2181 2182 /* 2183 * RENAME 2184 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a 2185 * different file handle for the same inode after a rename (e.g. when 2186 * moving to a different directory). A fail-safe method to do so would 2187 * be to look up old_dir/old_name, create a link to new_dir/new_name and 2188 * rename the old file using the sillyrename stuff. This way, the original 2189 * file in old_dir will go away when the last process iput()s the inode. 2190 * 2191 * FIXED. 2192 * 2193 * It actually works quite well. One needs to have the possibility for 2194 * at least one ".nfs..." file in each directory the file ever gets 2195 * moved or linked to which happens automagically with the new 2196 * implementation that only depends on the dcache stuff instead of 2197 * using the inode layer 2198 * 2199 * Unfortunately, things are a little more complicated than indicated 2200 * above. For a cross-directory move, we want to make sure we can get 2201 * rid of the old inode after the operation. This means there must be 2202 * no pending writes (if it's a file), and the use count must be 1. 2203 * If these conditions are met, we can drop the dentries before doing 2204 * the rename. 2205 */ 2206 int nfs_rename(struct inode *old_dir, struct dentry *old_dentry, 2207 struct inode *new_dir, struct dentry *new_dentry, 2208 unsigned int flags) 2209 { 2210 struct inode *old_inode = d_inode(old_dentry); 2211 struct inode *new_inode = d_inode(new_dentry); 2212 struct dentry *dentry = NULL, *rehash = NULL; 2213 struct rpc_task *task; 2214 int error = -EBUSY; 2215 2216 if (flags) 2217 return -EINVAL; 2218 2219 dfprintk(VFS, "NFS: rename(%pd2 -> %pd2, ct=%d)\n", 2220 old_dentry, new_dentry, 2221 d_count(new_dentry)); 2222 2223 trace_nfs_rename_enter(old_dir, old_dentry, new_dir, new_dentry); 2224 /* 2225 * For non-directories, check whether the target is busy and if so, 2226 * make a copy of the dentry and then do a silly-rename. If the 2227 * silly-rename succeeds, the copied dentry is hashed and becomes 2228 * the new target. 2229 */ 2230 if (new_inode && !S_ISDIR(new_inode->i_mode)) { 2231 /* 2232 * To prevent any new references to the target during the 2233 * rename, we unhash the dentry in advance. 2234 */ 2235 if (!d_unhashed(new_dentry)) { 2236 d_drop(new_dentry); 2237 rehash = new_dentry; 2238 } 2239 2240 if (d_count(new_dentry) > 2) { 2241 int err; 2242 2243 /* copy the target dentry's name */ 2244 dentry = d_alloc(new_dentry->d_parent, 2245 &new_dentry->d_name); 2246 if (!dentry) 2247 goto out; 2248 2249 /* silly-rename the existing target ... */ 2250 err = nfs_sillyrename(new_dir, new_dentry); 2251 if (err) 2252 goto out; 2253 2254 new_dentry = dentry; 2255 rehash = NULL; 2256 new_inode = NULL; 2257 } 2258 } 2259 2260 task = nfs_async_rename(old_dir, new_dir, old_dentry, new_dentry, NULL); 2261 if (IS_ERR(task)) { 2262 error = PTR_ERR(task); 2263 goto out; 2264 } 2265 2266 error = rpc_wait_for_completion_task(task); 2267 if (error != 0) { 2268 ((struct nfs_renamedata *)task->tk_calldata)->cancelled = 1; 2269 /* Paired with the atomic_dec_and_test() barrier in rpc_do_put_task() */ 2270 smp_wmb(); 2271 } else 2272 error = task->tk_status; 2273 rpc_put_task(task); 2274 /* Ensure the inode attributes are revalidated */ 2275 if (error == 0) { 2276 spin_lock(&old_inode->i_lock); 2277 NFS_I(old_inode)->attr_gencount = nfs_inc_attr_generation_counter(); 2278 NFS_I(old_inode)->cache_validity |= NFS_INO_INVALID_CHANGE 2279 | NFS_INO_INVALID_CTIME 2280 | NFS_INO_REVAL_FORCED; 2281 spin_unlock(&old_inode->i_lock); 2282 } 2283 out: 2284 if (rehash) 2285 d_rehash(rehash); 2286 trace_nfs_rename_exit(old_dir, old_dentry, 2287 new_dir, new_dentry, error); 2288 if (!error) { 2289 if (new_inode != NULL) 2290 nfs_drop_nlink(new_inode); 2291 /* 2292 * The d_move() should be here instead of in an async RPC completion 2293 * handler because we need the proper locks to move the dentry. If 2294 * we're interrupted by a signal, the async RPC completion handler 2295 * should mark the directories for revalidation. 2296 */ 2297 d_move(old_dentry, new_dentry); 2298 nfs_set_verifier(old_dentry, 2299 nfs_save_change_attribute(new_dir)); 2300 } else if (error == -ENOENT) 2301 nfs_dentry_handle_enoent(old_dentry); 2302 2303 /* new dentry created? */ 2304 if (dentry) 2305 dput(dentry); 2306 return error; 2307 } 2308 EXPORT_SYMBOL_GPL(nfs_rename); 2309 2310 static DEFINE_SPINLOCK(nfs_access_lru_lock); 2311 static LIST_HEAD(nfs_access_lru_list); 2312 static atomic_long_t nfs_access_nr_entries; 2313 2314 static unsigned long nfs_access_max_cachesize = 4*1024*1024; 2315 module_param(nfs_access_max_cachesize, ulong, 0644); 2316 MODULE_PARM_DESC(nfs_access_max_cachesize, "NFS access maximum total cache length"); 2317 2318 static void nfs_access_free_entry(struct nfs_access_entry *entry) 2319 { 2320 put_cred(entry->cred); 2321 kfree_rcu(entry, rcu_head); 2322 smp_mb__before_atomic(); 2323 atomic_long_dec(&nfs_access_nr_entries); 2324 smp_mb__after_atomic(); 2325 } 2326 2327 static void nfs_access_free_list(struct list_head *head) 2328 { 2329 struct nfs_access_entry *cache; 2330 2331 while (!list_empty(head)) { 2332 cache = list_entry(head->next, struct nfs_access_entry, lru); 2333 list_del(&cache->lru); 2334 nfs_access_free_entry(cache); 2335 } 2336 } 2337 2338 static unsigned long 2339 nfs_do_access_cache_scan(unsigned int nr_to_scan) 2340 { 2341 LIST_HEAD(head); 2342 struct nfs_inode *nfsi, *next; 2343 struct nfs_access_entry *cache; 2344 long freed = 0; 2345 2346 spin_lock(&nfs_access_lru_lock); 2347 list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) { 2348 struct inode *inode; 2349 2350 if (nr_to_scan-- == 0) 2351 break; 2352 inode = &nfsi->vfs_inode; 2353 spin_lock(&inode->i_lock); 2354 if (list_empty(&nfsi->access_cache_entry_lru)) 2355 goto remove_lru_entry; 2356 cache = list_entry(nfsi->access_cache_entry_lru.next, 2357 struct nfs_access_entry, lru); 2358 list_move(&cache->lru, &head); 2359 rb_erase(&cache->rb_node, &nfsi->access_cache); 2360 freed++; 2361 if (!list_empty(&nfsi->access_cache_entry_lru)) 2362 list_move_tail(&nfsi->access_cache_inode_lru, 2363 &nfs_access_lru_list); 2364 else { 2365 remove_lru_entry: 2366 list_del_init(&nfsi->access_cache_inode_lru); 2367 smp_mb__before_atomic(); 2368 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags); 2369 smp_mb__after_atomic(); 2370 } 2371 spin_unlock(&inode->i_lock); 2372 } 2373 spin_unlock(&nfs_access_lru_lock); 2374 nfs_access_free_list(&head); 2375 return freed; 2376 } 2377 2378 unsigned long 2379 nfs_access_cache_scan(struct shrinker *shrink, struct shrink_control *sc) 2380 { 2381 int nr_to_scan = sc->nr_to_scan; 2382 gfp_t gfp_mask = sc->gfp_mask; 2383 2384 if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL) 2385 return SHRINK_STOP; 2386 return nfs_do_access_cache_scan(nr_to_scan); 2387 } 2388 2389 2390 unsigned long 2391 nfs_access_cache_count(struct shrinker *shrink, struct shrink_control *sc) 2392 { 2393 return vfs_pressure_ratio(atomic_long_read(&nfs_access_nr_entries)); 2394 } 2395 2396 static void 2397 nfs_access_cache_enforce_limit(void) 2398 { 2399 long nr_entries = atomic_long_read(&nfs_access_nr_entries); 2400 unsigned long diff; 2401 unsigned int nr_to_scan; 2402 2403 if (nr_entries < 0 || nr_entries <= nfs_access_max_cachesize) 2404 return; 2405 nr_to_scan = 100; 2406 diff = nr_entries - nfs_access_max_cachesize; 2407 if (diff < nr_to_scan) 2408 nr_to_scan = diff; 2409 nfs_do_access_cache_scan(nr_to_scan); 2410 } 2411 2412 static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head) 2413 { 2414 struct rb_root *root_node = &nfsi->access_cache; 2415 struct rb_node *n; 2416 struct nfs_access_entry *entry; 2417 2418 /* Unhook entries from the cache */ 2419 while ((n = rb_first(root_node)) != NULL) { 2420 entry = rb_entry(n, struct nfs_access_entry, rb_node); 2421 rb_erase(n, root_node); 2422 list_move(&entry->lru, head); 2423 } 2424 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS; 2425 } 2426 2427 void nfs_access_zap_cache(struct inode *inode) 2428 { 2429 LIST_HEAD(head); 2430 2431 if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0) 2432 return; 2433 /* Remove from global LRU init */ 2434 spin_lock(&nfs_access_lru_lock); 2435 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) 2436 list_del_init(&NFS_I(inode)->access_cache_inode_lru); 2437 2438 spin_lock(&inode->i_lock); 2439 __nfs_access_zap_cache(NFS_I(inode), &head); 2440 spin_unlock(&inode->i_lock); 2441 spin_unlock(&nfs_access_lru_lock); 2442 nfs_access_free_list(&head); 2443 } 2444 EXPORT_SYMBOL_GPL(nfs_access_zap_cache); 2445 2446 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, const struct cred *cred) 2447 { 2448 struct rb_node *n = NFS_I(inode)->access_cache.rb_node; 2449 2450 while (n != NULL) { 2451 struct nfs_access_entry *entry = 2452 rb_entry(n, struct nfs_access_entry, rb_node); 2453 int cmp = cred_fscmp(cred, entry->cred); 2454 2455 if (cmp < 0) 2456 n = n->rb_left; 2457 else if (cmp > 0) 2458 n = n->rb_right; 2459 else 2460 return entry; 2461 } 2462 return NULL; 2463 } 2464 2465 static int nfs_access_get_cached_locked(struct inode *inode, const struct cred *cred, struct nfs_access_entry *res, bool may_block) 2466 { 2467 struct nfs_inode *nfsi = NFS_I(inode); 2468 struct nfs_access_entry *cache; 2469 bool retry = true; 2470 int err; 2471 2472 spin_lock(&inode->i_lock); 2473 for(;;) { 2474 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS) 2475 goto out_zap; 2476 cache = nfs_access_search_rbtree(inode, cred); 2477 err = -ENOENT; 2478 if (cache == NULL) 2479 goto out; 2480 /* Found an entry, is our attribute cache valid? */ 2481 if (!nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS)) 2482 break; 2483 if (!retry) 2484 break; 2485 err = -ECHILD; 2486 if (!may_block) 2487 goto out; 2488 spin_unlock(&inode->i_lock); 2489 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode); 2490 if (err) 2491 return err; 2492 spin_lock(&inode->i_lock); 2493 retry = false; 2494 } 2495 res->cred = cache->cred; 2496 res->mask = cache->mask; 2497 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru); 2498 err = 0; 2499 out: 2500 spin_unlock(&inode->i_lock); 2501 return err; 2502 out_zap: 2503 spin_unlock(&inode->i_lock); 2504 nfs_access_zap_cache(inode); 2505 return -ENOENT; 2506 } 2507 2508 static int nfs_access_get_cached_rcu(struct inode *inode, const struct cred *cred, struct nfs_access_entry *res) 2509 { 2510 /* Only check the most recently returned cache entry, 2511 * but do it without locking. 2512 */ 2513 struct nfs_inode *nfsi = NFS_I(inode); 2514 struct nfs_access_entry *cache; 2515 int err = -ECHILD; 2516 struct list_head *lh; 2517 2518 rcu_read_lock(); 2519 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS) 2520 goto out; 2521 lh = rcu_dereference(list_tail_rcu(&nfsi->access_cache_entry_lru)); 2522 cache = list_entry(lh, struct nfs_access_entry, lru); 2523 if (lh == &nfsi->access_cache_entry_lru || 2524 cred_fscmp(cred, cache->cred) != 0) 2525 cache = NULL; 2526 if (cache == NULL) 2527 goto out; 2528 if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS)) 2529 goto out; 2530 res->cred = cache->cred; 2531 res->mask = cache->mask; 2532 err = 0; 2533 out: 2534 rcu_read_unlock(); 2535 return err; 2536 } 2537 2538 int nfs_access_get_cached(struct inode *inode, const struct cred *cred, struct 2539 nfs_access_entry *res, bool may_block) 2540 { 2541 int status; 2542 2543 status = nfs_access_get_cached_rcu(inode, cred, res); 2544 if (status != 0) 2545 status = nfs_access_get_cached_locked(inode, cred, res, 2546 may_block); 2547 2548 return status; 2549 } 2550 EXPORT_SYMBOL_GPL(nfs_access_get_cached); 2551 2552 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set) 2553 { 2554 struct nfs_inode *nfsi = NFS_I(inode); 2555 struct rb_root *root_node = &nfsi->access_cache; 2556 struct rb_node **p = &root_node->rb_node; 2557 struct rb_node *parent = NULL; 2558 struct nfs_access_entry *entry; 2559 int cmp; 2560 2561 spin_lock(&inode->i_lock); 2562 while (*p != NULL) { 2563 parent = *p; 2564 entry = rb_entry(parent, struct nfs_access_entry, rb_node); 2565 cmp = cred_fscmp(set->cred, entry->cred); 2566 2567 if (cmp < 0) 2568 p = &parent->rb_left; 2569 else if (cmp > 0) 2570 p = &parent->rb_right; 2571 else 2572 goto found; 2573 } 2574 rb_link_node(&set->rb_node, parent, p); 2575 rb_insert_color(&set->rb_node, root_node); 2576 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); 2577 spin_unlock(&inode->i_lock); 2578 return; 2579 found: 2580 rb_replace_node(parent, &set->rb_node, root_node); 2581 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); 2582 list_del(&entry->lru); 2583 spin_unlock(&inode->i_lock); 2584 nfs_access_free_entry(entry); 2585 } 2586 2587 void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set) 2588 { 2589 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL); 2590 if (cache == NULL) 2591 return; 2592 RB_CLEAR_NODE(&cache->rb_node); 2593 cache->cred = get_cred(set->cred); 2594 cache->mask = set->mask; 2595 2596 /* The above field assignments must be visible 2597 * before this item appears on the lru. We cannot easily 2598 * use rcu_assign_pointer, so just force the memory barrier. 2599 */ 2600 smp_wmb(); 2601 nfs_access_add_rbtree(inode, cache); 2602 2603 /* Update accounting */ 2604 smp_mb__before_atomic(); 2605 atomic_long_inc(&nfs_access_nr_entries); 2606 smp_mb__after_atomic(); 2607 2608 /* Add inode to global LRU list */ 2609 if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) { 2610 spin_lock(&nfs_access_lru_lock); 2611 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) 2612 list_add_tail(&NFS_I(inode)->access_cache_inode_lru, 2613 &nfs_access_lru_list); 2614 spin_unlock(&nfs_access_lru_lock); 2615 } 2616 nfs_access_cache_enforce_limit(); 2617 } 2618 EXPORT_SYMBOL_GPL(nfs_access_add_cache); 2619 2620 #define NFS_MAY_READ (NFS_ACCESS_READ) 2621 #define NFS_MAY_WRITE (NFS_ACCESS_MODIFY | \ 2622 NFS_ACCESS_EXTEND | \ 2623 NFS_ACCESS_DELETE) 2624 #define NFS_FILE_MAY_WRITE (NFS_ACCESS_MODIFY | \ 2625 NFS_ACCESS_EXTEND) 2626 #define NFS_DIR_MAY_WRITE NFS_MAY_WRITE 2627 #define NFS_MAY_LOOKUP (NFS_ACCESS_LOOKUP) 2628 #define NFS_MAY_EXECUTE (NFS_ACCESS_EXECUTE) 2629 static int 2630 nfs_access_calc_mask(u32 access_result, umode_t umode) 2631 { 2632 int mask = 0; 2633 2634 if (access_result & NFS_MAY_READ) 2635 mask |= MAY_READ; 2636 if (S_ISDIR(umode)) { 2637 if ((access_result & NFS_DIR_MAY_WRITE) == NFS_DIR_MAY_WRITE) 2638 mask |= MAY_WRITE; 2639 if ((access_result & NFS_MAY_LOOKUP) == NFS_MAY_LOOKUP) 2640 mask |= MAY_EXEC; 2641 } else if (S_ISREG(umode)) { 2642 if ((access_result & NFS_FILE_MAY_WRITE) == NFS_FILE_MAY_WRITE) 2643 mask |= MAY_WRITE; 2644 if ((access_result & NFS_MAY_EXECUTE) == NFS_MAY_EXECUTE) 2645 mask |= MAY_EXEC; 2646 } else if (access_result & NFS_MAY_WRITE) 2647 mask |= MAY_WRITE; 2648 return mask; 2649 } 2650 2651 void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result) 2652 { 2653 entry->mask = access_result; 2654 } 2655 EXPORT_SYMBOL_GPL(nfs_access_set_mask); 2656 2657 static int nfs_do_access(struct inode *inode, const struct cred *cred, int mask) 2658 { 2659 struct nfs_access_entry cache; 2660 bool may_block = (mask & MAY_NOT_BLOCK) == 0; 2661 int cache_mask = -1; 2662 int status; 2663 2664 trace_nfs_access_enter(inode); 2665 2666 status = nfs_access_get_cached(inode, cred, &cache, may_block); 2667 if (status == 0) 2668 goto out_cached; 2669 2670 status = -ECHILD; 2671 if (!may_block) 2672 goto out; 2673 2674 /* 2675 * Determine which access bits we want to ask for... 2676 */ 2677 cache.mask = NFS_ACCESS_READ | NFS_ACCESS_MODIFY | NFS_ACCESS_EXTEND; 2678 if (nfs_server_capable(inode, NFS_CAP_XATTR)) { 2679 cache.mask |= NFS_ACCESS_XAREAD | NFS_ACCESS_XAWRITE | 2680 NFS_ACCESS_XALIST; 2681 } 2682 if (S_ISDIR(inode->i_mode)) 2683 cache.mask |= NFS_ACCESS_DELETE | NFS_ACCESS_LOOKUP; 2684 else 2685 cache.mask |= NFS_ACCESS_EXECUTE; 2686 cache.cred = cred; 2687 status = NFS_PROTO(inode)->access(inode, &cache); 2688 if (status != 0) { 2689 if (status == -ESTALE) { 2690 if (!S_ISDIR(inode->i_mode)) 2691 nfs_set_inode_stale(inode); 2692 else 2693 nfs_zap_caches(inode); 2694 } 2695 goto out; 2696 } 2697 nfs_access_add_cache(inode, &cache); 2698 out_cached: 2699 cache_mask = nfs_access_calc_mask(cache.mask, inode->i_mode); 2700 if ((mask & ~cache_mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) != 0) 2701 status = -EACCES; 2702 out: 2703 trace_nfs_access_exit(inode, mask, cache_mask, status); 2704 return status; 2705 } 2706 2707 static int nfs_open_permission_mask(int openflags) 2708 { 2709 int mask = 0; 2710 2711 if (openflags & __FMODE_EXEC) { 2712 /* ONLY check exec rights */ 2713 mask = MAY_EXEC; 2714 } else { 2715 if ((openflags & O_ACCMODE) != O_WRONLY) 2716 mask |= MAY_READ; 2717 if ((openflags & O_ACCMODE) != O_RDONLY) 2718 mask |= MAY_WRITE; 2719 } 2720 2721 return mask; 2722 } 2723 2724 int nfs_may_open(struct inode *inode, const struct cred *cred, int openflags) 2725 { 2726 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags)); 2727 } 2728 EXPORT_SYMBOL_GPL(nfs_may_open); 2729 2730 static int nfs_execute_ok(struct inode *inode, int mask) 2731 { 2732 struct nfs_server *server = NFS_SERVER(inode); 2733 int ret = 0; 2734 2735 if (S_ISDIR(inode->i_mode)) 2736 return 0; 2737 if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_OTHER)) { 2738 if (mask & MAY_NOT_BLOCK) 2739 return -ECHILD; 2740 ret = __nfs_revalidate_inode(server, inode); 2741 } 2742 if (ret == 0 && !execute_ok(inode)) 2743 ret = -EACCES; 2744 return ret; 2745 } 2746 2747 int nfs_permission(struct inode *inode, int mask) 2748 { 2749 const struct cred *cred = current_cred(); 2750 int res = 0; 2751 2752 nfs_inc_stats(inode, NFSIOS_VFSACCESS); 2753 2754 if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) 2755 goto out; 2756 /* Is this sys_access() ? */ 2757 if (mask & (MAY_ACCESS | MAY_CHDIR)) 2758 goto force_lookup; 2759 2760 switch (inode->i_mode & S_IFMT) { 2761 case S_IFLNK: 2762 goto out; 2763 case S_IFREG: 2764 if ((mask & MAY_OPEN) && 2765 nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN)) 2766 return 0; 2767 break; 2768 case S_IFDIR: 2769 /* 2770 * Optimize away all write operations, since the server 2771 * will check permissions when we perform the op. 2772 */ 2773 if ((mask & MAY_WRITE) && !(mask & MAY_READ)) 2774 goto out; 2775 } 2776 2777 force_lookup: 2778 if (!NFS_PROTO(inode)->access) 2779 goto out_notsup; 2780 2781 res = nfs_do_access(inode, cred, mask); 2782 out: 2783 if (!res && (mask & MAY_EXEC)) 2784 res = nfs_execute_ok(inode, mask); 2785 2786 dfprintk(VFS, "NFS: permission(%s/%lu), mask=0x%x, res=%d\n", 2787 inode->i_sb->s_id, inode->i_ino, mask, res); 2788 return res; 2789 out_notsup: 2790 if (mask & MAY_NOT_BLOCK) 2791 return -ECHILD; 2792 2793 res = nfs_revalidate_inode(NFS_SERVER(inode), inode); 2794 if (res == 0) 2795 res = generic_permission(inode, mask); 2796 goto out; 2797 } 2798 EXPORT_SYMBOL_GPL(nfs_permission); 2799 2800 /* 2801 * Local variables: 2802 * version-control: t 2803 * kept-new-versions: 5 2804 * End: 2805 */ 2806