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 u64 last_cookie; 142 unsigned int size; 143 unsigned char page_full : 1, 144 page_is_eof : 1; 145 struct nfs_cache_array_entry array[]; 146 }; 147 148 typedef struct nfs_readdir_descriptor { 149 struct file *file; 150 struct page *page; 151 struct dir_context *ctx; 152 pgoff_t page_index; 153 u64 dir_cookie; 154 u64 last_cookie; 155 u64 dup_cookie; 156 loff_t current_index; 157 loff_t prev_index; 158 159 unsigned long dir_verifier; 160 unsigned long timestamp; 161 unsigned long gencount; 162 unsigned long attr_gencount; 163 unsigned int cache_entry_index; 164 signed char duped; 165 bool plus; 166 bool eof; 167 } nfs_readdir_descriptor_t; 168 169 static void nfs_readdir_array_init(struct nfs_cache_array *array) 170 { 171 memset(array, 0, sizeof(struct nfs_cache_array)); 172 } 173 174 static void nfs_readdir_page_init_array(struct page *page, u64 last_cookie) 175 { 176 struct nfs_cache_array *array; 177 178 array = kmap_atomic(page); 179 nfs_readdir_array_init(array); 180 array->last_cookie = last_cookie; 181 kunmap_atomic(array); 182 } 183 184 /* 185 * we are freeing strings created by nfs_add_to_readdir_array() 186 */ 187 static 188 void nfs_readdir_clear_array(struct page *page) 189 { 190 struct nfs_cache_array *array; 191 int i; 192 193 array = kmap_atomic(page); 194 for (i = 0; i < array->size; i++) 195 kfree(array->array[i].string.name); 196 nfs_readdir_array_init(array); 197 kunmap_atomic(array); 198 } 199 200 static void nfs_readdir_array_set_eof(struct nfs_cache_array *array) 201 { 202 array->page_is_eof = 1; 203 array->page_full = 1; 204 } 205 206 static bool nfs_readdir_array_is_full(struct nfs_cache_array *array) 207 { 208 return array->page_full; 209 } 210 211 /* 212 * the caller is responsible for freeing qstr.name 213 * when called by nfs_readdir_add_to_array, the strings will be freed in 214 * nfs_clear_readdir_array() 215 */ 216 static 217 int nfs_readdir_make_qstr(struct qstr *string, const char *name, unsigned int len) 218 { 219 string->len = len; 220 string->name = kmemdup_nul(name, len, GFP_KERNEL); 221 if (string->name == NULL) 222 return -ENOMEM; 223 /* 224 * Avoid a kmemleak false positive. The pointer to the name is stored 225 * in a page cache page which kmemleak does not scan. 226 */ 227 kmemleak_not_leak(string->name); 228 string->hash = full_name_hash(NULL, name, len); 229 return 0; 230 } 231 232 /* 233 * Check that the next array entry lies entirely within the page bounds 234 */ 235 static int nfs_readdir_array_can_expand(struct nfs_cache_array *array) 236 { 237 struct nfs_cache_array_entry *cache_entry; 238 239 if (array->page_full) 240 return -ENOSPC; 241 cache_entry = &array->array[array->size + 1]; 242 if ((char *)cache_entry - (char *)array > PAGE_SIZE) { 243 array->page_full = 1; 244 return -ENOSPC; 245 } 246 return 0; 247 } 248 249 static 250 int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page) 251 { 252 struct nfs_cache_array *array = kmap(page); 253 struct nfs_cache_array_entry *cache_entry; 254 int ret; 255 256 ret = nfs_readdir_array_can_expand(array); 257 if (ret) 258 goto out; 259 260 cache_entry = &array->array[array->size]; 261 cache_entry->cookie = entry->prev_cookie; 262 cache_entry->ino = entry->ino; 263 cache_entry->d_type = entry->d_type; 264 ret = nfs_readdir_make_qstr(&cache_entry->string, entry->name, entry->len); 265 if (ret) 266 goto out; 267 array->last_cookie = entry->cookie; 268 array->size++; 269 if (entry->eof != 0) 270 nfs_readdir_array_set_eof(array); 271 out: 272 kunmap(page); 273 return ret; 274 } 275 276 static struct page *nfs_readdir_page_get_locked(struct address_space *mapping, 277 pgoff_t index, u64 last_cookie) 278 { 279 struct page *page; 280 281 page = grab_cache_page(mapping, index); 282 if (page && !PageUptodate(page)) { 283 nfs_readdir_page_init_array(page, last_cookie); 284 if (invalidate_inode_pages2_range(mapping, index + 1, -1) < 0) 285 nfs_zap_mapping(mapping->host, mapping); 286 SetPageUptodate(page); 287 } 288 289 return page; 290 } 291 292 static u64 nfs_readdir_page_last_cookie(struct page *page) 293 { 294 struct nfs_cache_array *array; 295 u64 ret; 296 297 array = kmap_atomic(page); 298 ret = array->last_cookie; 299 kunmap_atomic(array); 300 return ret; 301 } 302 303 static bool nfs_readdir_page_needs_filling(struct page *page) 304 { 305 struct nfs_cache_array *array; 306 bool ret; 307 308 array = kmap_atomic(page); 309 ret = !nfs_readdir_array_is_full(array); 310 kunmap_atomic(array); 311 return ret; 312 } 313 314 static void nfs_readdir_page_set_eof(struct page *page) 315 { 316 struct nfs_cache_array *array; 317 318 array = kmap_atomic(page); 319 nfs_readdir_array_set_eof(array); 320 kunmap_atomic(array); 321 } 322 323 static inline 324 int is_32bit_api(void) 325 { 326 #ifdef CONFIG_COMPAT 327 return in_compat_syscall(); 328 #else 329 return (BITS_PER_LONG == 32); 330 #endif 331 } 332 333 static 334 bool nfs_readdir_use_cookie(const struct file *filp) 335 { 336 if ((filp->f_mode & FMODE_32BITHASH) || 337 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api())) 338 return false; 339 return true; 340 } 341 342 static 343 int nfs_readdir_search_for_pos(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc) 344 { 345 loff_t diff = desc->ctx->pos - desc->current_index; 346 unsigned int index; 347 348 if (diff < 0) 349 goto out_eof; 350 if (diff >= array->size) { 351 if (array->page_is_eof) 352 goto out_eof; 353 return -EAGAIN; 354 } 355 356 index = (unsigned int)diff; 357 desc->dir_cookie = array->array[index].cookie; 358 desc->cache_entry_index = index; 359 return 0; 360 out_eof: 361 desc->eof = true; 362 return -EBADCOOKIE; 363 } 364 365 static bool 366 nfs_readdir_inode_mapping_valid(struct nfs_inode *nfsi) 367 { 368 if (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA)) 369 return false; 370 smp_rmb(); 371 return !test_bit(NFS_INO_INVALIDATING, &nfsi->flags); 372 } 373 374 static 375 int nfs_readdir_search_for_cookie(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc) 376 { 377 int i; 378 loff_t new_pos; 379 int status = -EAGAIN; 380 381 for (i = 0; i < array->size; i++) { 382 if (array->array[i].cookie == desc->dir_cookie) { 383 struct nfs_inode *nfsi = NFS_I(file_inode(desc->file)); 384 385 new_pos = desc->current_index + i; 386 if (desc->attr_gencount != nfsi->attr_gencount || 387 !nfs_readdir_inode_mapping_valid(nfsi)) { 388 desc->duped = 0; 389 desc->attr_gencount = nfsi->attr_gencount; 390 } else if (new_pos < desc->prev_index) { 391 if (desc->duped > 0 392 && desc->dup_cookie == desc->dir_cookie) { 393 if (printk_ratelimit()) { 394 pr_notice("NFS: directory %pD2 contains a readdir loop." 395 "Please contact your server vendor. " 396 "The file: %.*s has duplicate cookie %llu\n", 397 desc->file, array->array[i].string.len, 398 array->array[i].string.name, desc->dir_cookie); 399 } 400 status = -ELOOP; 401 goto out; 402 } 403 desc->dup_cookie = desc->dir_cookie; 404 desc->duped = -1; 405 } 406 if (nfs_readdir_use_cookie(desc->file)) 407 desc->ctx->pos = desc->dir_cookie; 408 else 409 desc->ctx->pos = new_pos; 410 desc->prev_index = new_pos; 411 desc->cache_entry_index = i; 412 return 0; 413 } 414 } 415 if (array->page_is_eof) { 416 status = -EBADCOOKIE; 417 if (desc->dir_cookie == array->last_cookie) 418 desc->eof = true; 419 } 420 out: 421 return status; 422 } 423 424 static 425 int nfs_readdir_search_array(nfs_readdir_descriptor_t *desc) 426 { 427 struct nfs_cache_array *array; 428 int status; 429 430 array = kmap(desc->page); 431 432 if (desc->dir_cookie == 0) 433 status = nfs_readdir_search_for_pos(array, desc); 434 else 435 status = nfs_readdir_search_for_cookie(array, desc); 436 437 if (status == -EAGAIN) { 438 desc->last_cookie = array->last_cookie; 439 desc->current_index += array->size; 440 desc->page_index++; 441 } 442 kunmap(desc->page); 443 return status; 444 } 445 446 /* Fill a page with xdr information before transferring to the cache page */ 447 static 448 int nfs_readdir_xdr_filler(struct page **pages, nfs_readdir_descriptor_t *desc, 449 struct nfs_entry *entry, struct file *file, struct inode *inode) 450 { 451 struct nfs_open_dir_context *ctx = file->private_data; 452 const struct cred *cred = ctx->cred; 453 unsigned long timestamp, gencount; 454 int error; 455 456 again: 457 timestamp = jiffies; 458 gencount = nfs_inc_attr_generation_counter(); 459 desc->dir_verifier = nfs_save_change_attribute(inode); 460 error = NFS_PROTO(inode)->readdir(file_dentry(file), cred, entry->cookie, pages, 461 NFS_SERVER(inode)->dtsize, desc->plus); 462 if (error < 0) { 463 /* We requested READDIRPLUS, but the server doesn't grok it */ 464 if (error == -ENOTSUPP && desc->plus) { 465 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS; 466 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags); 467 desc->plus = false; 468 goto again; 469 } 470 goto error; 471 } 472 desc->timestamp = timestamp; 473 desc->gencount = gencount; 474 error: 475 return error; 476 } 477 478 static int xdr_decode(nfs_readdir_descriptor_t *desc, 479 struct nfs_entry *entry, struct xdr_stream *xdr) 480 { 481 struct inode *inode = file_inode(desc->file); 482 int error; 483 484 error = NFS_PROTO(inode)->decode_dirent(xdr, entry, desc->plus); 485 if (error) 486 return error; 487 entry->fattr->time_start = desc->timestamp; 488 entry->fattr->gencount = desc->gencount; 489 return 0; 490 } 491 492 /* Match file and dirent using either filehandle or fileid 493 * Note: caller is responsible for checking the fsid 494 */ 495 static 496 int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry) 497 { 498 struct inode *inode; 499 struct nfs_inode *nfsi; 500 501 if (d_really_is_negative(dentry)) 502 return 0; 503 504 inode = d_inode(dentry); 505 if (is_bad_inode(inode) || NFS_STALE(inode)) 506 return 0; 507 508 nfsi = NFS_I(inode); 509 if (entry->fattr->fileid != nfsi->fileid) 510 return 0; 511 if (entry->fh->size && nfs_compare_fh(entry->fh, &nfsi->fh) != 0) 512 return 0; 513 return 1; 514 } 515 516 static 517 bool nfs_use_readdirplus(struct inode *dir, struct dir_context *ctx) 518 { 519 if (!nfs_server_capable(dir, NFS_CAP_READDIRPLUS)) 520 return false; 521 if (test_and_clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags)) 522 return true; 523 if (ctx->pos == 0) 524 return true; 525 return false; 526 } 527 528 /* 529 * This function is called by the lookup and getattr code to request the 530 * use of readdirplus to accelerate any future lookups in the same 531 * directory. 532 */ 533 void nfs_advise_use_readdirplus(struct inode *dir) 534 { 535 struct nfs_inode *nfsi = NFS_I(dir); 536 537 if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) && 538 !list_empty(&nfsi->open_files)) 539 set_bit(NFS_INO_ADVISE_RDPLUS, &nfsi->flags); 540 } 541 542 /* 543 * This function is mainly for use by nfs_getattr(). 544 * 545 * If this is an 'ls -l', we want to force use of readdirplus. 546 * Do this by checking if there is an active file descriptor 547 * and calling nfs_advise_use_readdirplus, then forcing a 548 * cache flush. 549 */ 550 void nfs_force_use_readdirplus(struct inode *dir) 551 { 552 struct nfs_inode *nfsi = NFS_I(dir); 553 554 if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) && 555 !list_empty(&nfsi->open_files)) { 556 set_bit(NFS_INO_ADVISE_RDPLUS, &nfsi->flags); 557 invalidate_mapping_pages(dir->i_mapping, 558 nfsi->page_index + 1, -1); 559 } 560 } 561 562 static 563 void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry, 564 unsigned long dir_verifier) 565 { 566 struct qstr filename = QSTR_INIT(entry->name, entry->len); 567 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 568 struct dentry *dentry; 569 struct dentry *alias; 570 struct inode *inode; 571 int status; 572 573 if (!(entry->fattr->valid & NFS_ATTR_FATTR_FILEID)) 574 return; 575 if (!(entry->fattr->valid & NFS_ATTR_FATTR_FSID)) 576 return; 577 if (filename.len == 0) 578 return; 579 /* Validate that the name doesn't contain any illegal '\0' */ 580 if (strnlen(filename.name, filename.len) != filename.len) 581 return; 582 /* ...or '/' */ 583 if (strnchr(filename.name, filename.len, '/')) 584 return; 585 if (filename.name[0] == '.') { 586 if (filename.len == 1) 587 return; 588 if (filename.len == 2 && filename.name[1] == '.') 589 return; 590 } 591 filename.hash = full_name_hash(parent, filename.name, filename.len); 592 593 dentry = d_lookup(parent, &filename); 594 again: 595 if (!dentry) { 596 dentry = d_alloc_parallel(parent, &filename, &wq); 597 if (IS_ERR(dentry)) 598 return; 599 } 600 if (!d_in_lookup(dentry)) { 601 /* Is there a mountpoint here? If so, just exit */ 602 if (!nfs_fsid_equal(&NFS_SB(dentry->d_sb)->fsid, 603 &entry->fattr->fsid)) 604 goto out; 605 if (nfs_same_file(dentry, entry)) { 606 if (!entry->fh->size) 607 goto out; 608 nfs_set_verifier(dentry, dir_verifier); 609 status = nfs_refresh_inode(d_inode(dentry), entry->fattr); 610 if (!status) 611 nfs_setsecurity(d_inode(dentry), entry->fattr, entry->label); 612 goto out; 613 } else { 614 d_invalidate(dentry); 615 dput(dentry); 616 dentry = NULL; 617 goto again; 618 } 619 } 620 if (!entry->fh->size) { 621 d_lookup_done(dentry); 622 goto out; 623 } 624 625 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr, entry->label); 626 alias = d_splice_alias(inode, dentry); 627 d_lookup_done(dentry); 628 if (alias) { 629 if (IS_ERR(alias)) 630 goto out; 631 dput(dentry); 632 dentry = alias; 633 } 634 nfs_set_verifier(dentry, dir_verifier); 635 out: 636 dput(dentry); 637 } 638 639 /* Perform conversion from xdr to cache array */ 640 static 641 int nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry, 642 struct page **xdr_pages, struct page *page, unsigned int buflen) 643 { 644 struct xdr_stream stream; 645 struct xdr_buf buf; 646 struct page *scratch; 647 int status; 648 649 scratch = alloc_page(GFP_KERNEL); 650 if (scratch == NULL) 651 return -ENOMEM; 652 653 xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen); 654 xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE); 655 656 do { 657 if (entry->label) 658 entry->label->len = NFS4_MAXLABELLEN; 659 660 status = xdr_decode(desc, entry, &stream); 661 if (status != 0) 662 break; 663 664 if (desc->plus) 665 nfs_prime_dcache(file_dentry(desc->file), entry, 666 desc->dir_verifier); 667 668 status = nfs_readdir_add_to_array(entry, page); 669 } while (!status && !entry->eof); 670 671 switch (status) { 672 case -EBADCOOKIE: 673 if (entry->eof) { 674 nfs_readdir_page_set_eof(page); 675 status = 0; 676 } 677 break; 678 case -ENOSPC: 679 case -EAGAIN: 680 status = 0; 681 break; 682 } 683 684 put_page(scratch); 685 return status; 686 } 687 688 static 689 void nfs_readdir_free_pages(struct page **pages, unsigned int npages) 690 { 691 unsigned int i; 692 for (i = 0; i < npages; i++) 693 put_page(pages[i]); 694 } 695 696 /* 697 * nfs_readdir_alloc_pages() will allocate pages that must be freed with a call 698 * to nfs_readdir_free_pages() 699 */ 700 static 701 int nfs_readdir_alloc_pages(struct page **pages, unsigned int npages) 702 { 703 unsigned int i; 704 705 for (i = 0; i < npages; i++) { 706 struct page *page = alloc_page(GFP_KERNEL); 707 if (page == NULL) 708 goto out_freepages; 709 pages[i] = page; 710 } 711 return 0; 712 713 out_freepages: 714 nfs_readdir_free_pages(pages, i); 715 return -ENOMEM; 716 } 717 718 static 719 int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t *desc, struct page *page, struct inode *inode) 720 { 721 struct page *pages[NFS_MAX_READDIR_PAGES]; 722 struct nfs_entry entry; 723 struct file *file = desc->file; 724 struct nfs_cache_array *array; 725 int status = -ENOMEM; 726 unsigned int array_size = ARRAY_SIZE(pages); 727 728 entry.prev_cookie = 0; 729 entry.cookie = nfs_readdir_page_last_cookie(page); 730 entry.eof = 0; 731 entry.fh = nfs_alloc_fhandle(); 732 entry.fattr = nfs_alloc_fattr(); 733 entry.server = NFS_SERVER(inode); 734 if (entry.fh == NULL || entry.fattr == NULL) 735 goto out; 736 737 entry.label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT); 738 if (IS_ERR(entry.label)) { 739 status = PTR_ERR(entry.label); 740 goto out; 741 } 742 743 array = kmap(page); 744 745 status = nfs_readdir_alloc_pages(pages, array_size); 746 if (status < 0) 747 goto out_release_array; 748 do { 749 unsigned int pglen; 750 status = nfs_readdir_xdr_filler(pages, desc, &entry, file, inode); 751 752 if (status < 0) 753 break; 754 755 pglen = status; 756 if (pglen == 0) { 757 nfs_readdir_page_set_eof(page); 758 break; 759 } 760 761 status = nfs_readdir_page_filler(desc, &entry, pages, page, pglen); 762 } while (!status && !nfs_readdir_array_is_full(array)); 763 764 nfs_readdir_free_pages(pages, array_size); 765 out_release_array: 766 kunmap(page); 767 nfs4_label_free(entry.label); 768 out: 769 nfs_free_fattr(entry.fattr); 770 nfs_free_fhandle(entry.fh); 771 return status; 772 } 773 774 static void nfs_readdir_page_put(struct nfs_readdir_descriptor *desc) 775 { 776 put_page(desc->page); 777 desc->page = NULL; 778 } 779 780 static void 781 nfs_readdir_page_unlock_and_put_cached(struct nfs_readdir_descriptor *desc) 782 { 783 unlock_page(desc->page); 784 nfs_readdir_page_put(desc); 785 } 786 787 static struct page * 788 nfs_readdir_page_get_cached(struct nfs_readdir_descriptor *desc) 789 { 790 return nfs_readdir_page_get_locked(desc->file->f_mapping, 791 desc->page_index, 792 desc->last_cookie); 793 } 794 795 /* 796 * Returns 0 if desc->dir_cookie was found on page desc->page_index 797 * and locks the page to prevent removal from the page cache. 798 */ 799 static 800 int find_and_lock_cache_page(nfs_readdir_descriptor_t *desc) 801 { 802 struct inode *inode = file_inode(desc->file); 803 struct nfs_inode *nfsi = NFS_I(inode); 804 int res; 805 806 desc->page = nfs_readdir_page_get_cached(desc); 807 if (!desc->page) 808 return -ENOMEM; 809 if (nfs_readdir_page_needs_filling(desc->page)) { 810 res = nfs_readdir_xdr_to_array(desc, desc->page, inode); 811 if (res < 0) 812 goto error; 813 } 814 res = nfs_readdir_search_array(desc); 815 if (res == 0) { 816 nfsi->page_index = desc->page_index; 817 return 0; 818 } 819 error: 820 nfs_readdir_page_unlock_and_put_cached(desc); 821 return res; 822 } 823 824 /* Search for desc->dir_cookie from the beginning of the page cache */ 825 static inline 826 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc) 827 { 828 int res; 829 830 if (desc->page_index == 0) { 831 desc->current_index = 0; 832 desc->prev_index = 0; 833 desc->last_cookie = 0; 834 } 835 do { 836 res = find_and_lock_cache_page(desc); 837 } while (res == -EAGAIN); 838 return res; 839 } 840 841 /* 842 * Once we've found the start of the dirent within a page: fill 'er up... 843 */ 844 static 845 int nfs_do_filldir(nfs_readdir_descriptor_t *desc) 846 { 847 struct file *file = desc->file; 848 int i = 0; 849 int res = 0; 850 struct nfs_cache_array *array = NULL; 851 852 array = kmap(desc->page); 853 for (i = desc->cache_entry_index; i < array->size; i++) { 854 struct nfs_cache_array_entry *ent; 855 856 ent = &array->array[i]; 857 if (!dir_emit(desc->ctx, ent->string.name, ent->string.len, 858 nfs_compat_user_ino64(ent->ino), ent->d_type)) { 859 desc->eof = true; 860 break; 861 } 862 if (i < (array->size-1)) 863 desc->dir_cookie = array->array[i+1].cookie; 864 else 865 desc->dir_cookie = array->last_cookie; 866 if (nfs_readdir_use_cookie(file)) 867 desc->ctx->pos = desc->dir_cookie; 868 else 869 desc->ctx->pos++; 870 if (desc->duped != 0) 871 desc->duped = 1; 872 } 873 if (array->page_is_eof) 874 desc->eof = true; 875 876 kunmap(desc->page); 877 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n", 878 (unsigned long long)desc->dir_cookie, res); 879 return res; 880 } 881 882 /* 883 * If we cannot find a cookie in our cache, we suspect that this is 884 * because it points to a deleted file, so we ask the server to return 885 * whatever it thinks is the next entry. We then feed this to filldir. 886 * If all goes well, we should then be able to find our way round the 887 * cache on the next call to readdir_search_pagecache(); 888 * 889 * NOTE: we cannot add the anonymous page to the pagecache because 890 * the data it contains might not be page aligned. Besides, 891 * we should already have a complete representation of the 892 * directory in the page cache by the time we get here. 893 */ 894 static inline 895 int uncached_readdir(nfs_readdir_descriptor_t *desc) 896 { 897 struct page *page = NULL; 898 int status; 899 struct inode *inode = file_inode(desc->file); 900 901 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n", 902 (unsigned long long)desc->dir_cookie); 903 904 page = alloc_page(GFP_HIGHUSER); 905 if (!page) { 906 status = -ENOMEM; 907 goto out; 908 } 909 910 desc->page_index = 0; 911 desc->last_cookie = desc->dir_cookie; 912 desc->page = page; 913 desc->duped = 0; 914 915 nfs_readdir_page_init_array(page, desc->dir_cookie); 916 status = nfs_readdir_xdr_to_array(desc, page, inode); 917 if (status < 0) 918 goto out_release; 919 920 status = nfs_do_filldir(desc); 921 922 out_release: 923 nfs_readdir_clear_array(desc->page); 924 nfs_readdir_page_put(desc); 925 out: 926 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", 927 __func__, status); 928 return status; 929 } 930 931 /* The file offset position represents the dirent entry number. A 932 last cookie cache takes care of the common case of reading the 933 whole directory. 934 */ 935 static int nfs_readdir(struct file *file, struct dir_context *ctx) 936 { 937 struct dentry *dentry = file_dentry(file); 938 struct inode *inode = d_inode(dentry); 939 struct nfs_open_dir_context *dir_ctx = file->private_data; 940 nfs_readdir_descriptor_t my_desc = { 941 .file = file, 942 .ctx = ctx, 943 .plus = nfs_use_readdirplus(inode, ctx), 944 }, 945 *desc = &my_desc; 946 int res = 0; 947 948 dfprintk(FILE, "NFS: readdir(%pD2) starting at cookie %llu\n", 949 file, (long long)ctx->pos); 950 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS); 951 952 /* 953 * ctx->pos points to the dirent entry number. 954 * *desc->dir_cookie has the cookie for the next entry. We have 955 * to either find the entry with the appropriate number or 956 * revalidate the cookie. 957 */ 958 if (ctx->pos == 0 || nfs_attribute_cache_expired(inode)) 959 res = nfs_revalidate_mapping(inode, file->f_mapping); 960 if (res < 0) 961 goto out; 962 963 spin_lock(&file->f_lock); 964 desc->dir_cookie = dir_ctx->dir_cookie; 965 desc->dup_cookie = dir_ctx->dup_cookie; 966 desc->duped = dir_ctx->duped; 967 desc->attr_gencount = dir_ctx->attr_gencount; 968 spin_unlock(&file->f_lock); 969 970 do { 971 res = readdir_search_pagecache(desc); 972 973 if (res == -EBADCOOKIE) { 974 res = 0; 975 /* This means either end of directory */ 976 if (desc->dir_cookie && !desc->eof) { 977 /* Or that the server has 'lost' a cookie */ 978 res = uncached_readdir(desc); 979 if (res == 0) 980 continue; 981 } 982 break; 983 } 984 if (res == -ETOOSMALL && desc->plus) { 985 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags); 986 nfs_zap_caches(inode); 987 desc->page_index = 0; 988 desc->plus = false; 989 desc->eof = false; 990 continue; 991 } 992 if (res < 0) 993 break; 994 995 res = nfs_do_filldir(desc); 996 nfs_readdir_page_unlock_and_put_cached(desc); 997 if (res < 0) 998 break; 999 } while (!desc->eof); 1000 1001 spin_lock(&file->f_lock); 1002 dir_ctx->dir_cookie = desc->dir_cookie; 1003 dir_ctx->dup_cookie = desc->dup_cookie; 1004 dir_ctx->duped = desc->duped; 1005 dir_ctx->attr_gencount = desc->attr_gencount; 1006 spin_unlock(&file->f_lock); 1007 1008 out: 1009 if (res > 0) 1010 res = 0; 1011 dfprintk(FILE, "NFS: readdir(%pD2) returns %d\n", file, res); 1012 return res; 1013 } 1014 1015 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int whence) 1016 { 1017 struct nfs_open_dir_context *dir_ctx = filp->private_data; 1018 1019 dfprintk(FILE, "NFS: llseek dir(%pD2, %lld, %d)\n", 1020 filp, offset, whence); 1021 1022 switch (whence) { 1023 default: 1024 return -EINVAL; 1025 case SEEK_SET: 1026 if (offset < 0) 1027 return -EINVAL; 1028 spin_lock(&filp->f_lock); 1029 break; 1030 case SEEK_CUR: 1031 if (offset == 0) 1032 return filp->f_pos; 1033 spin_lock(&filp->f_lock); 1034 offset += filp->f_pos; 1035 if (offset < 0) { 1036 spin_unlock(&filp->f_lock); 1037 return -EINVAL; 1038 } 1039 } 1040 if (offset != filp->f_pos) { 1041 filp->f_pos = offset; 1042 if (nfs_readdir_use_cookie(filp)) 1043 dir_ctx->dir_cookie = offset; 1044 else 1045 dir_ctx->dir_cookie = 0; 1046 dir_ctx->duped = 0; 1047 } 1048 spin_unlock(&filp->f_lock); 1049 return offset; 1050 } 1051 1052 /* 1053 * All directory operations under NFS are synchronous, so fsync() 1054 * is a dummy operation. 1055 */ 1056 static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end, 1057 int datasync) 1058 { 1059 dfprintk(FILE, "NFS: fsync dir(%pD2) datasync %d\n", filp, datasync); 1060 1061 nfs_inc_stats(file_inode(filp), NFSIOS_VFSFSYNC); 1062 return 0; 1063 } 1064 1065 /** 1066 * nfs_force_lookup_revalidate - Mark the directory as having changed 1067 * @dir: pointer to directory inode 1068 * 1069 * This forces the revalidation code in nfs_lookup_revalidate() to do a 1070 * full lookup on all child dentries of 'dir' whenever a change occurs 1071 * on the server that might have invalidated our dcache. 1072 * 1073 * Note that we reserve bit '0' as a tag to let us know when a dentry 1074 * was revalidated while holding a delegation on its inode. 1075 * 1076 * The caller should be holding dir->i_lock 1077 */ 1078 void nfs_force_lookup_revalidate(struct inode *dir) 1079 { 1080 NFS_I(dir)->cache_change_attribute += 2; 1081 } 1082 EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate); 1083 1084 /** 1085 * nfs_verify_change_attribute - Detects NFS remote directory changes 1086 * @dir: pointer to parent directory inode 1087 * @verf: previously saved change attribute 1088 * 1089 * Return "false" if the verifiers doesn't match the change attribute. 1090 * This would usually indicate that the directory contents have changed on 1091 * the server, and that any dentries need revalidating. 1092 */ 1093 static bool nfs_verify_change_attribute(struct inode *dir, unsigned long verf) 1094 { 1095 return (verf & ~1UL) == nfs_save_change_attribute(dir); 1096 } 1097 1098 static void nfs_set_verifier_delegated(unsigned long *verf) 1099 { 1100 *verf |= 1UL; 1101 } 1102 1103 #if IS_ENABLED(CONFIG_NFS_V4) 1104 static void nfs_unset_verifier_delegated(unsigned long *verf) 1105 { 1106 *verf &= ~1UL; 1107 } 1108 #endif /* IS_ENABLED(CONFIG_NFS_V4) */ 1109 1110 static bool nfs_test_verifier_delegated(unsigned long verf) 1111 { 1112 return verf & 1; 1113 } 1114 1115 static bool nfs_verifier_is_delegated(struct dentry *dentry) 1116 { 1117 return nfs_test_verifier_delegated(dentry->d_time); 1118 } 1119 1120 static void nfs_set_verifier_locked(struct dentry *dentry, unsigned long verf) 1121 { 1122 struct inode *inode = d_inode(dentry); 1123 1124 if (!nfs_verifier_is_delegated(dentry) && 1125 !nfs_verify_change_attribute(d_inode(dentry->d_parent), verf)) 1126 goto out; 1127 if (inode && NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) 1128 nfs_set_verifier_delegated(&verf); 1129 out: 1130 dentry->d_time = verf; 1131 } 1132 1133 /** 1134 * nfs_set_verifier - save a parent directory verifier in the dentry 1135 * @dentry: pointer to dentry 1136 * @verf: verifier to save 1137 * 1138 * Saves the parent directory verifier in @dentry. If the inode has 1139 * a delegation, we also tag the dentry as having been revalidated 1140 * while holding a delegation so that we know we don't have to 1141 * look it up again after a directory change. 1142 */ 1143 void nfs_set_verifier(struct dentry *dentry, unsigned long verf) 1144 { 1145 1146 spin_lock(&dentry->d_lock); 1147 nfs_set_verifier_locked(dentry, verf); 1148 spin_unlock(&dentry->d_lock); 1149 } 1150 EXPORT_SYMBOL_GPL(nfs_set_verifier); 1151 1152 #if IS_ENABLED(CONFIG_NFS_V4) 1153 /** 1154 * nfs_clear_verifier_delegated - clear the dir verifier delegation tag 1155 * @inode: pointer to inode 1156 * 1157 * Iterates through the dentries in the inode alias list and clears 1158 * the tag used to indicate that the dentry has been revalidated 1159 * while holding a delegation. 1160 * This function is intended for use when the delegation is being 1161 * returned or revoked. 1162 */ 1163 void nfs_clear_verifier_delegated(struct inode *inode) 1164 { 1165 struct dentry *alias; 1166 1167 if (!inode) 1168 return; 1169 spin_lock(&inode->i_lock); 1170 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) { 1171 spin_lock(&alias->d_lock); 1172 nfs_unset_verifier_delegated(&alias->d_time); 1173 spin_unlock(&alias->d_lock); 1174 } 1175 spin_unlock(&inode->i_lock); 1176 } 1177 EXPORT_SYMBOL_GPL(nfs_clear_verifier_delegated); 1178 #endif /* IS_ENABLED(CONFIG_NFS_V4) */ 1179 1180 /* 1181 * A check for whether or not the parent directory has changed. 1182 * In the case it has, we assume that the dentries are untrustworthy 1183 * and may need to be looked up again. 1184 * If rcu_walk prevents us from performing a full check, return 0. 1185 */ 1186 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry, 1187 int rcu_walk) 1188 { 1189 if (IS_ROOT(dentry)) 1190 return 1; 1191 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE) 1192 return 0; 1193 if (!nfs_verify_change_attribute(dir, dentry->d_time)) 1194 return 0; 1195 /* Revalidate nfsi->cache_change_attribute before we declare a match */ 1196 if (nfs_mapping_need_revalidate_inode(dir)) { 1197 if (rcu_walk) 1198 return 0; 1199 if (__nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0) 1200 return 0; 1201 } 1202 if (!nfs_verify_change_attribute(dir, dentry->d_time)) 1203 return 0; 1204 return 1; 1205 } 1206 1207 /* 1208 * Use intent information to check whether or not we're going to do 1209 * an O_EXCL create using this path component. 1210 */ 1211 static int nfs_is_exclusive_create(struct inode *dir, unsigned int flags) 1212 { 1213 if (NFS_PROTO(dir)->version == 2) 1214 return 0; 1215 return flags & LOOKUP_EXCL; 1216 } 1217 1218 /* 1219 * Inode and filehandle revalidation for lookups. 1220 * 1221 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL, 1222 * or if the intent information indicates that we're about to open this 1223 * particular file and the "nocto" mount flag is not set. 1224 * 1225 */ 1226 static 1227 int nfs_lookup_verify_inode(struct inode *inode, unsigned int flags) 1228 { 1229 struct nfs_server *server = NFS_SERVER(inode); 1230 int ret; 1231 1232 if (IS_AUTOMOUNT(inode)) 1233 return 0; 1234 1235 if (flags & LOOKUP_OPEN) { 1236 switch (inode->i_mode & S_IFMT) { 1237 case S_IFREG: 1238 /* A NFSv4 OPEN will revalidate later */ 1239 if (server->caps & NFS_CAP_ATOMIC_OPEN) 1240 goto out; 1241 fallthrough; 1242 case S_IFDIR: 1243 if (server->flags & NFS_MOUNT_NOCTO) 1244 break; 1245 /* NFS close-to-open cache consistency validation */ 1246 goto out_force; 1247 } 1248 } 1249 1250 /* VFS wants an on-the-wire revalidation */ 1251 if (flags & LOOKUP_REVAL) 1252 goto out_force; 1253 out: 1254 return (inode->i_nlink == 0) ? -ESTALE : 0; 1255 out_force: 1256 if (flags & LOOKUP_RCU) 1257 return -ECHILD; 1258 ret = __nfs_revalidate_inode(server, inode); 1259 if (ret != 0) 1260 return ret; 1261 goto out; 1262 } 1263 1264 /* 1265 * We judge how long we want to trust negative 1266 * dentries by looking at the parent inode mtime. 1267 * 1268 * If parent mtime has changed, we revalidate, else we wait for a 1269 * period corresponding to the parent's attribute cache timeout value. 1270 * 1271 * If LOOKUP_RCU prevents us from performing a full check, return 1 1272 * suggesting a reval is needed. 1273 * 1274 * Note that when creating a new file, or looking up a rename target, 1275 * then it shouldn't be necessary to revalidate a negative dentry. 1276 */ 1277 static inline 1278 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry, 1279 unsigned int flags) 1280 { 1281 if (flags & (LOOKUP_CREATE | LOOKUP_RENAME_TARGET)) 1282 return 0; 1283 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG) 1284 return 1; 1285 return !nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU); 1286 } 1287 1288 static int 1289 nfs_lookup_revalidate_done(struct inode *dir, struct dentry *dentry, 1290 struct inode *inode, int error) 1291 { 1292 switch (error) { 1293 case 1: 1294 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is valid\n", 1295 __func__, dentry); 1296 return 1; 1297 case 0: 1298 nfs_mark_for_revalidate(dir); 1299 if (inode && S_ISDIR(inode->i_mode)) { 1300 /* Purge readdir caches. */ 1301 nfs_zap_caches(inode); 1302 /* 1303 * We can't d_drop the root of a disconnected tree: 1304 * its d_hash is on the s_anon list and d_drop() would hide 1305 * it from shrink_dcache_for_unmount(), leading to busy 1306 * inodes on unmount and further oopses. 1307 */ 1308 if (IS_ROOT(dentry)) 1309 return 1; 1310 } 1311 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is invalid\n", 1312 __func__, dentry); 1313 return 0; 1314 } 1315 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) lookup returned error %d\n", 1316 __func__, dentry, error); 1317 return error; 1318 } 1319 1320 static int 1321 nfs_lookup_revalidate_negative(struct inode *dir, struct dentry *dentry, 1322 unsigned int flags) 1323 { 1324 int ret = 1; 1325 if (nfs_neg_need_reval(dir, dentry, flags)) { 1326 if (flags & LOOKUP_RCU) 1327 return -ECHILD; 1328 ret = 0; 1329 } 1330 return nfs_lookup_revalidate_done(dir, dentry, NULL, ret); 1331 } 1332 1333 static int 1334 nfs_lookup_revalidate_delegated(struct inode *dir, struct dentry *dentry, 1335 struct inode *inode) 1336 { 1337 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1338 return nfs_lookup_revalidate_done(dir, dentry, inode, 1); 1339 } 1340 1341 static int 1342 nfs_lookup_revalidate_dentry(struct inode *dir, struct dentry *dentry, 1343 struct inode *inode) 1344 { 1345 struct nfs_fh *fhandle; 1346 struct nfs_fattr *fattr; 1347 struct nfs4_label *label; 1348 unsigned long dir_verifier; 1349 int ret; 1350 1351 ret = -ENOMEM; 1352 fhandle = nfs_alloc_fhandle(); 1353 fattr = nfs_alloc_fattr(); 1354 label = nfs4_label_alloc(NFS_SERVER(inode), GFP_KERNEL); 1355 if (fhandle == NULL || fattr == NULL || IS_ERR(label)) 1356 goto out; 1357 1358 dir_verifier = nfs_save_change_attribute(dir); 1359 ret = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr, label); 1360 if (ret < 0) { 1361 switch (ret) { 1362 case -ESTALE: 1363 case -ENOENT: 1364 ret = 0; 1365 break; 1366 case -ETIMEDOUT: 1367 if (NFS_SERVER(inode)->flags & NFS_MOUNT_SOFTREVAL) 1368 ret = 1; 1369 } 1370 goto out; 1371 } 1372 ret = 0; 1373 if (nfs_compare_fh(NFS_FH(inode), fhandle)) 1374 goto out; 1375 if (nfs_refresh_inode(inode, fattr) < 0) 1376 goto out; 1377 1378 nfs_setsecurity(inode, fattr, label); 1379 nfs_set_verifier(dentry, dir_verifier); 1380 1381 /* set a readdirplus hint that we had a cache miss */ 1382 nfs_force_use_readdirplus(dir); 1383 ret = 1; 1384 out: 1385 nfs_free_fattr(fattr); 1386 nfs_free_fhandle(fhandle); 1387 nfs4_label_free(label); 1388 return nfs_lookup_revalidate_done(dir, dentry, inode, ret); 1389 } 1390 1391 /* 1392 * This is called every time the dcache has a lookup hit, 1393 * and we should check whether we can really trust that 1394 * lookup. 1395 * 1396 * NOTE! The hit can be a negative hit too, don't assume 1397 * we have an inode! 1398 * 1399 * If the parent directory is seen to have changed, we throw out the 1400 * cached dentry and do a new lookup. 1401 */ 1402 static int 1403 nfs_do_lookup_revalidate(struct inode *dir, struct dentry *dentry, 1404 unsigned int flags) 1405 { 1406 struct inode *inode; 1407 int error; 1408 1409 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE); 1410 inode = d_inode(dentry); 1411 1412 if (!inode) 1413 return nfs_lookup_revalidate_negative(dir, dentry, flags); 1414 1415 if (is_bad_inode(inode)) { 1416 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n", 1417 __func__, dentry); 1418 goto out_bad; 1419 } 1420 1421 if (nfs_verifier_is_delegated(dentry)) 1422 return nfs_lookup_revalidate_delegated(dir, dentry, inode); 1423 1424 /* Force a full look up iff the parent directory has changed */ 1425 if (!(flags & (LOOKUP_EXCL | LOOKUP_REVAL)) && 1426 nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU)) { 1427 error = nfs_lookup_verify_inode(inode, flags); 1428 if (error) { 1429 if (error == -ESTALE) 1430 nfs_zap_caches(dir); 1431 goto out_bad; 1432 } 1433 nfs_advise_use_readdirplus(dir); 1434 goto out_valid; 1435 } 1436 1437 if (flags & LOOKUP_RCU) 1438 return -ECHILD; 1439 1440 if (NFS_STALE(inode)) 1441 goto out_bad; 1442 1443 trace_nfs_lookup_revalidate_enter(dir, dentry, flags); 1444 error = nfs_lookup_revalidate_dentry(dir, dentry, inode); 1445 trace_nfs_lookup_revalidate_exit(dir, dentry, flags, error); 1446 return error; 1447 out_valid: 1448 return nfs_lookup_revalidate_done(dir, dentry, inode, 1); 1449 out_bad: 1450 if (flags & LOOKUP_RCU) 1451 return -ECHILD; 1452 return nfs_lookup_revalidate_done(dir, dentry, inode, 0); 1453 } 1454 1455 static int 1456 __nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags, 1457 int (*reval)(struct inode *, struct dentry *, unsigned int)) 1458 { 1459 struct dentry *parent; 1460 struct inode *dir; 1461 int ret; 1462 1463 if (flags & LOOKUP_RCU) { 1464 parent = READ_ONCE(dentry->d_parent); 1465 dir = d_inode_rcu(parent); 1466 if (!dir) 1467 return -ECHILD; 1468 ret = reval(dir, dentry, flags); 1469 if (parent != READ_ONCE(dentry->d_parent)) 1470 return -ECHILD; 1471 } else { 1472 parent = dget_parent(dentry); 1473 ret = reval(d_inode(parent), dentry, flags); 1474 dput(parent); 1475 } 1476 return ret; 1477 } 1478 1479 static int nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags) 1480 { 1481 return __nfs_lookup_revalidate(dentry, flags, nfs_do_lookup_revalidate); 1482 } 1483 1484 /* 1485 * A weaker form of d_revalidate for revalidating just the d_inode(dentry) 1486 * when we don't really care about the dentry name. This is called when a 1487 * pathwalk ends on a dentry that was not found via a normal lookup in the 1488 * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals). 1489 * 1490 * In this situation, we just want to verify that the inode itself is OK 1491 * since the dentry might have changed on the server. 1492 */ 1493 static int nfs_weak_revalidate(struct dentry *dentry, unsigned int flags) 1494 { 1495 struct inode *inode = d_inode(dentry); 1496 int error = 0; 1497 1498 /* 1499 * I believe we can only get a negative dentry here in the case of a 1500 * procfs-style symlink. Just assume it's correct for now, but we may 1501 * eventually need to do something more here. 1502 */ 1503 if (!inode) { 1504 dfprintk(LOOKUPCACHE, "%s: %pd2 has negative inode\n", 1505 __func__, dentry); 1506 return 1; 1507 } 1508 1509 if (is_bad_inode(inode)) { 1510 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n", 1511 __func__, dentry); 1512 return 0; 1513 } 1514 1515 error = nfs_lookup_verify_inode(inode, flags); 1516 dfprintk(LOOKUPCACHE, "NFS: %s: inode %lu is %s\n", 1517 __func__, inode->i_ino, error ? "invalid" : "valid"); 1518 return !error; 1519 } 1520 1521 /* 1522 * This is called from dput() when d_count is going to 0. 1523 */ 1524 static int nfs_dentry_delete(const struct dentry *dentry) 1525 { 1526 dfprintk(VFS, "NFS: dentry_delete(%pd2, %x)\n", 1527 dentry, dentry->d_flags); 1528 1529 /* Unhash any dentry with a stale inode */ 1530 if (d_really_is_positive(dentry) && NFS_STALE(d_inode(dentry))) 1531 return 1; 1532 1533 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { 1534 /* Unhash it, so that ->d_iput() would be called */ 1535 return 1; 1536 } 1537 if (!(dentry->d_sb->s_flags & SB_ACTIVE)) { 1538 /* Unhash it, so that ancestors of killed async unlink 1539 * files will be cleaned up during umount */ 1540 return 1; 1541 } 1542 return 0; 1543 1544 } 1545 1546 /* Ensure that we revalidate inode->i_nlink */ 1547 static void nfs_drop_nlink(struct inode *inode) 1548 { 1549 spin_lock(&inode->i_lock); 1550 /* drop the inode if we're reasonably sure this is the last link */ 1551 if (inode->i_nlink > 0) 1552 drop_nlink(inode); 1553 NFS_I(inode)->attr_gencount = nfs_inc_attr_generation_counter(); 1554 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_CHANGE 1555 | NFS_INO_INVALID_CTIME 1556 | NFS_INO_INVALID_OTHER 1557 | NFS_INO_REVAL_FORCED; 1558 spin_unlock(&inode->i_lock); 1559 } 1560 1561 /* 1562 * Called when the dentry loses inode. 1563 * We use it to clean up silly-renamed files. 1564 */ 1565 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode) 1566 { 1567 if (S_ISDIR(inode->i_mode)) 1568 /* drop any readdir cache as it could easily be old */ 1569 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA; 1570 1571 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { 1572 nfs_complete_unlink(dentry, inode); 1573 nfs_drop_nlink(inode); 1574 } 1575 iput(inode); 1576 } 1577 1578 static void nfs_d_release(struct dentry *dentry) 1579 { 1580 /* free cached devname value, if it survived that far */ 1581 if (unlikely(dentry->d_fsdata)) { 1582 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) 1583 WARN_ON(1); 1584 else 1585 kfree(dentry->d_fsdata); 1586 } 1587 } 1588 1589 const struct dentry_operations nfs_dentry_operations = { 1590 .d_revalidate = nfs_lookup_revalidate, 1591 .d_weak_revalidate = nfs_weak_revalidate, 1592 .d_delete = nfs_dentry_delete, 1593 .d_iput = nfs_dentry_iput, 1594 .d_automount = nfs_d_automount, 1595 .d_release = nfs_d_release, 1596 }; 1597 EXPORT_SYMBOL_GPL(nfs_dentry_operations); 1598 1599 struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags) 1600 { 1601 struct dentry *res; 1602 struct inode *inode = NULL; 1603 struct nfs_fh *fhandle = NULL; 1604 struct nfs_fattr *fattr = NULL; 1605 struct nfs4_label *label = NULL; 1606 unsigned long dir_verifier; 1607 int error; 1608 1609 dfprintk(VFS, "NFS: lookup(%pd2)\n", dentry); 1610 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP); 1611 1612 if (unlikely(dentry->d_name.len > NFS_SERVER(dir)->namelen)) 1613 return ERR_PTR(-ENAMETOOLONG); 1614 1615 /* 1616 * If we're doing an exclusive create, optimize away the lookup 1617 * but don't hash the dentry. 1618 */ 1619 if (nfs_is_exclusive_create(dir, flags) || flags & LOOKUP_RENAME_TARGET) 1620 return NULL; 1621 1622 res = ERR_PTR(-ENOMEM); 1623 fhandle = nfs_alloc_fhandle(); 1624 fattr = nfs_alloc_fattr(); 1625 if (fhandle == NULL || fattr == NULL) 1626 goto out; 1627 1628 label = nfs4_label_alloc(NFS_SERVER(dir), GFP_NOWAIT); 1629 if (IS_ERR(label)) 1630 goto out; 1631 1632 dir_verifier = nfs_save_change_attribute(dir); 1633 trace_nfs_lookup_enter(dir, dentry, flags); 1634 error = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr, label); 1635 if (error == -ENOENT) 1636 goto no_entry; 1637 if (error < 0) { 1638 res = ERR_PTR(error); 1639 goto out_label; 1640 } 1641 inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label); 1642 res = ERR_CAST(inode); 1643 if (IS_ERR(res)) 1644 goto out_label; 1645 1646 /* Notify readdir to use READDIRPLUS */ 1647 nfs_force_use_readdirplus(dir); 1648 1649 no_entry: 1650 res = d_splice_alias(inode, dentry); 1651 if (res != NULL) { 1652 if (IS_ERR(res)) 1653 goto out_label; 1654 dentry = res; 1655 } 1656 nfs_set_verifier(dentry, dir_verifier); 1657 out_label: 1658 trace_nfs_lookup_exit(dir, dentry, flags, error); 1659 nfs4_label_free(label); 1660 out: 1661 nfs_free_fattr(fattr); 1662 nfs_free_fhandle(fhandle); 1663 return res; 1664 } 1665 EXPORT_SYMBOL_GPL(nfs_lookup); 1666 1667 #if IS_ENABLED(CONFIG_NFS_V4) 1668 static int nfs4_lookup_revalidate(struct dentry *, unsigned int); 1669 1670 const struct dentry_operations nfs4_dentry_operations = { 1671 .d_revalidate = nfs4_lookup_revalidate, 1672 .d_weak_revalidate = nfs_weak_revalidate, 1673 .d_delete = nfs_dentry_delete, 1674 .d_iput = nfs_dentry_iput, 1675 .d_automount = nfs_d_automount, 1676 .d_release = nfs_d_release, 1677 }; 1678 EXPORT_SYMBOL_GPL(nfs4_dentry_operations); 1679 1680 static fmode_t flags_to_mode(int flags) 1681 { 1682 fmode_t res = (__force fmode_t)flags & FMODE_EXEC; 1683 if ((flags & O_ACCMODE) != O_WRONLY) 1684 res |= FMODE_READ; 1685 if ((flags & O_ACCMODE) != O_RDONLY) 1686 res |= FMODE_WRITE; 1687 return res; 1688 } 1689 1690 static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags, struct file *filp) 1691 { 1692 return alloc_nfs_open_context(dentry, flags_to_mode(open_flags), filp); 1693 } 1694 1695 static int do_open(struct inode *inode, struct file *filp) 1696 { 1697 nfs_fscache_open_file(inode, filp); 1698 return 0; 1699 } 1700 1701 static int nfs_finish_open(struct nfs_open_context *ctx, 1702 struct dentry *dentry, 1703 struct file *file, unsigned open_flags) 1704 { 1705 int err; 1706 1707 err = finish_open(file, dentry, do_open); 1708 if (err) 1709 goto out; 1710 if (S_ISREG(file->f_path.dentry->d_inode->i_mode)) 1711 nfs_file_set_open_context(file, ctx); 1712 else 1713 err = -EOPENSTALE; 1714 out: 1715 return err; 1716 } 1717 1718 int nfs_atomic_open(struct inode *dir, struct dentry *dentry, 1719 struct file *file, unsigned open_flags, 1720 umode_t mode) 1721 { 1722 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 1723 struct nfs_open_context *ctx; 1724 struct dentry *res; 1725 struct iattr attr = { .ia_valid = ATTR_OPEN }; 1726 struct inode *inode; 1727 unsigned int lookup_flags = 0; 1728 bool switched = false; 1729 int created = 0; 1730 int err; 1731 1732 /* Expect a negative dentry */ 1733 BUG_ON(d_inode(dentry)); 1734 1735 dfprintk(VFS, "NFS: atomic_open(%s/%lu), %pd\n", 1736 dir->i_sb->s_id, dir->i_ino, dentry); 1737 1738 err = nfs_check_flags(open_flags); 1739 if (err) 1740 return err; 1741 1742 /* NFS only supports OPEN on regular files */ 1743 if ((open_flags & O_DIRECTORY)) { 1744 if (!d_in_lookup(dentry)) { 1745 /* 1746 * Hashed negative dentry with O_DIRECTORY: dentry was 1747 * revalidated and is fine, no need to perform lookup 1748 * again 1749 */ 1750 return -ENOENT; 1751 } 1752 lookup_flags = LOOKUP_OPEN|LOOKUP_DIRECTORY; 1753 goto no_open; 1754 } 1755 1756 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) 1757 return -ENAMETOOLONG; 1758 1759 if (open_flags & O_CREAT) { 1760 struct nfs_server *server = NFS_SERVER(dir); 1761 1762 if (!(server->attr_bitmask[2] & FATTR4_WORD2_MODE_UMASK)) 1763 mode &= ~current_umask(); 1764 1765 attr.ia_valid |= ATTR_MODE; 1766 attr.ia_mode = mode; 1767 } 1768 if (open_flags & O_TRUNC) { 1769 attr.ia_valid |= ATTR_SIZE; 1770 attr.ia_size = 0; 1771 } 1772 1773 if (!(open_flags & O_CREAT) && !d_in_lookup(dentry)) { 1774 d_drop(dentry); 1775 switched = true; 1776 dentry = d_alloc_parallel(dentry->d_parent, 1777 &dentry->d_name, &wq); 1778 if (IS_ERR(dentry)) 1779 return PTR_ERR(dentry); 1780 if (unlikely(!d_in_lookup(dentry))) 1781 return finish_no_open(file, dentry); 1782 } 1783 1784 ctx = create_nfs_open_context(dentry, open_flags, file); 1785 err = PTR_ERR(ctx); 1786 if (IS_ERR(ctx)) 1787 goto out; 1788 1789 trace_nfs_atomic_open_enter(dir, ctx, open_flags); 1790 inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr, &created); 1791 if (created) 1792 file->f_mode |= FMODE_CREATED; 1793 if (IS_ERR(inode)) { 1794 err = PTR_ERR(inode); 1795 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err); 1796 put_nfs_open_context(ctx); 1797 d_drop(dentry); 1798 switch (err) { 1799 case -ENOENT: 1800 d_splice_alias(NULL, dentry); 1801 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1802 break; 1803 case -EISDIR: 1804 case -ENOTDIR: 1805 goto no_open; 1806 case -ELOOP: 1807 if (!(open_flags & O_NOFOLLOW)) 1808 goto no_open; 1809 break; 1810 /* case -EINVAL: */ 1811 default: 1812 break; 1813 } 1814 goto out; 1815 } 1816 1817 err = nfs_finish_open(ctx, ctx->dentry, file, open_flags); 1818 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err); 1819 put_nfs_open_context(ctx); 1820 out: 1821 if (unlikely(switched)) { 1822 d_lookup_done(dentry); 1823 dput(dentry); 1824 } 1825 return err; 1826 1827 no_open: 1828 res = nfs_lookup(dir, dentry, lookup_flags); 1829 if (switched) { 1830 d_lookup_done(dentry); 1831 if (!res) 1832 res = dentry; 1833 else 1834 dput(dentry); 1835 } 1836 if (IS_ERR(res)) 1837 return PTR_ERR(res); 1838 return finish_no_open(file, res); 1839 } 1840 EXPORT_SYMBOL_GPL(nfs_atomic_open); 1841 1842 static int 1843 nfs4_do_lookup_revalidate(struct inode *dir, struct dentry *dentry, 1844 unsigned int flags) 1845 { 1846 struct inode *inode; 1847 1848 if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY)) 1849 goto full_reval; 1850 if (d_mountpoint(dentry)) 1851 goto full_reval; 1852 1853 inode = d_inode(dentry); 1854 1855 /* We can't create new files in nfs_open_revalidate(), so we 1856 * optimize away revalidation of negative dentries. 1857 */ 1858 if (inode == NULL) 1859 goto full_reval; 1860 1861 if (nfs_verifier_is_delegated(dentry)) 1862 return nfs_lookup_revalidate_delegated(dir, dentry, inode); 1863 1864 /* NFS only supports OPEN on regular files */ 1865 if (!S_ISREG(inode->i_mode)) 1866 goto full_reval; 1867 1868 /* We cannot do exclusive creation on a positive dentry */ 1869 if (flags & (LOOKUP_EXCL | LOOKUP_REVAL)) 1870 goto reval_dentry; 1871 1872 /* Check if the directory changed */ 1873 if (!nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU)) 1874 goto reval_dentry; 1875 1876 /* Let f_op->open() actually open (and revalidate) the file */ 1877 return 1; 1878 reval_dentry: 1879 if (flags & LOOKUP_RCU) 1880 return -ECHILD; 1881 return nfs_lookup_revalidate_dentry(dir, dentry, inode); 1882 1883 full_reval: 1884 return nfs_do_lookup_revalidate(dir, dentry, flags); 1885 } 1886 1887 static int nfs4_lookup_revalidate(struct dentry *dentry, unsigned int flags) 1888 { 1889 return __nfs_lookup_revalidate(dentry, flags, 1890 nfs4_do_lookup_revalidate); 1891 } 1892 1893 #endif /* CONFIG_NFSV4 */ 1894 1895 struct dentry * 1896 nfs_add_or_obtain(struct dentry *dentry, struct nfs_fh *fhandle, 1897 struct nfs_fattr *fattr, 1898 struct nfs4_label *label) 1899 { 1900 struct dentry *parent = dget_parent(dentry); 1901 struct inode *dir = d_inode(parent); 1902 struct inode *inode; 1903 struct dentry *d; 1904 int error; 1905 1906 d_drop(dentry); 1907 1908 if (fhandle->size == 0) { 1909 error = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr, NULL); 1910 if (error) 1911 goto out_error; 1912 } 1913 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1914 if (!(fattr->valid & NFS_ATTR_FATTR)) { 1915 struct nfs_server *server = NFS_SB(dentry->d_sb); 1916 error = server->nfs_client->rpc_ops->getattr(server, fhandle, 1917 fattr, NULL, NULL); 1918 if (error < 0) 1919 goto out_error; 1920 } 1921 inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label); 1922 d = d_splice_alias(inode, dentry); 1923 out: 1924 dput(parent); 1925 return d; 1926 out_error: 1927 nfs_mark_for_revalidate(dir); 1928 d = ERR_PTR(error); 1929 goto out; 1930 } 1931 EXPORT_SYMBOL_GPL(nfs_add_or_obtain); 1932 1933 /* 1934 * Code common to create, mkdir, and mknod. 1935 */ 1936 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle, 1937 struct nfs_fattr *fattr, 1938 struct nfs4_label *label) 1939 { 1940 struct dentry *d; 1941 1942 d = nfs_add_or_obtain(dentry, fhandle, fattr, label); 1943 if (IS_ERR(d)) 1944 return PTR_ERR(d); 1945 1946 /* Callers don't care */ 1947 dput(d); 1948 return 0; 1949 } 1950 EXPORT_SYMBOL_GPL(nfs_instantiate); 1951 1952 /* 1953 * Following a failed create operation, we drop the dentry rather 1954 * than retain a negative dentry. This avoids a problem in the event 1955 * that the operation succeeded on the server, but an error in the 1956 * reply path made it appear to have failed. 1957 */ 1958 int nfs_create(struct inode *dir, struct dentry *dentry, 1959 umode_t mode, bool excl) 1960 { 1961 struct iattr attr; 1962 int open_flags = excl ? O_CREAT | O_EXCL : O_CREAT; 1963 int error; 1964 1965 dfprintk(VFS, "NFS: create(%s/%lu), %pd\n", 1966 dir->i_sb->s_id, dir->i_ino, dentry); 1967 1968 attr.ia_mode = mode; 1969 attr.ia_valid = ATTR_MODE; 1970 1971 trace_nfs_create_enter(dir, dentry, open_flags); 1972 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags); 1973 trace_nfs_create_exit(dir, dentry, open_flags, error); 1974 if (error != 0) 1975 goto out_err; 1976 return 0; 1977 out_err: 1978 d_drop(dentry); 1979 return error; 1980 } 1981 EXPORT_SYMBOL_GPL(nfs_create); 1982 1983 /* 1984 * See comments for nfs_proc_create regarding failed operations. 1985 */ 1986 int 1987 nfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t rdev) 1988 { 1989 struct iattr attr; 1990 int status; 1991 1992 dfprintk(VFS, "NFS: mknod(%s/%lu), %pd\n", 1993 dir->i_sb->s_id, dir->i_ino, dentry); 1994 1995 attr.ia_mode = mode; 1996 attr.ia_valid = ATTR_MODE; 1997 1998 trace_nfs_mknod_enter(dir, dentry); 1999 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev); 2000 trace_nfs_mknod_exit(dir, dentry, status); 2001 if (status != 0) 2002 goto out_err; 2003 return 0; 2004 out_err: 2005 d_drop(dentry); 2006 return status; 2007 } 2008 EXPORT_SYMBOL_GPL(nfs_mknod); 2009 2010 /* 2011 * See comments for nfs_proc_create regarding failed operations. 2012 */ 2013 int nfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) 2014 { 2015 struct iattr attr; 2016 int error; 2017 2018 dfprintk(VFS, "NFS: mkdir(%s/%lu), %pd\n", 2019 dir->i_sb->s_id, dir->i_ino, dentry); 2020 2021 attr.ia_valid = ATTR_MODE; 2022 attr.ia_mode = mode | S_IFDIR; 2023 2024 trace_nfs_mkdir_enter(dir, dentry); 2025 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr); 2026 trace_nfs_mkdir_exit(dir, dentry, error); 2027 if (error != 0) 2028 goto out_err; 2029 return 0; 2030 out_err: 2031 d_drop(dentry); 2032 return error; 2033 } 2034 EXPORT_SYMBOL_GPL(nfs_mkdir); 2035 2036 static void nfs_dentry_handle_enoent(struct dentry *dentry) 2037 { 2038 if (simple_positive(dentry)) 2039 d_delete(dentry); 2040 } 2041 2042 int nfs_rmdir(struct inode *dir, struct dentry *dentry) 2043 { 2044 int error; 2045 2046 dfprintk(VFS, "NFS: rmdir(%s/%lu), %pd\n", 2047 dir->i_sb->s_id, dir->i_ino, dentry); 2048 2049 trace_nfs_rmdir_enter(dir, dentry); 2050 if (d_really_is_positive(dentry)) { 2051 down_write(&NFS_I(d_inode(dentry))->rmdir_sem); 2052 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name); 2053 /* Ensure the VFS deletes this inode */ 2054 switch (error) { 2055 case 0: 2056 clear_nlink(d_inode(dentry)); 2057 break; 2058 case -ENOENT: 2059 nfs_dentry_handle_enoent(dentry); 2060 } 2061 up_write(&NFS_I(d_inode(dentry))->rmdir_sem); 2062 } else 2063 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name); 2064 trace_nfs_rmdir_exit(dir, dentry, error); 2065 2066 return error; 2067 } 2068 EXPORT_SYMBOL_GPL(nfs_rmdir); 2069 2070 /* 2071 * Remove a file after making sure there are no pending writes, 2072 * and after checking that the file has only one user. 2073 * 2074 * We invalidate the attribute cache and free the inode prior to the operation 2075 * to avoid possible races if the server reuses the inode. 2076 */ 2077 static int nfs_safe_remove(struct dentry *dentry) 2078 { 2079 struct inode *dir = d_inode(dentry->d_parent); 2080 struct inode *inode = d_inode(dentry); 2081 int error = -EBUSY; 2082 2083 dfprintk(VFS, "NFS: safe_remove(%pd2)\n", dentry); 2084 2085 /* If the dentry was sillyrenamed, we simply call d_delete() */ 2086 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { 2087 error = 0; 2088 goto out; 2089 } 2090 2091 trace_nfs_remove_enter(dir, dentry); 2092 if (inode != NULL) { 2093 error = NFS_PROTO(dir)->remove(dir, dentry); 2094 if (error == 0) 2095 nfs_drop_nlink(inode); 2096 } else 2097 error = NFS_PROTO(dir)->remove(dir, dentry); 2098 if (error == -ENOENT) 2099 nfs_dentry_handle_enoent(dentry); 2100 trace_nfs_remove_exit(dir, dentry, error); 2101 out: 2102 return error; 2103 } 2104 2105 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode 2106 * belongs to an active ".nfs..." file and we return -EBUSY. 2107 * 2108 * If sillyrename() returns 0, we do nothing, otherwise we unlink. 2109 */ 2110 int nfs_unlink(struct inode *dir, struct dentry *dentry) 2111 { 2112 int error; 2113 int need_rehash = 0; 2114 2115 dfprintk(VFS, "NFS: unlink(%s/%lu, %pd)\n", dir->i_sb->s_id, 2116 dir->i_ino, dentry); 2117 2118 trace_nfs_unlink_enter(dir, dentry); 2119 spin_lock(&dentry->d_lock); 2120 if (d_count(dentry) > 1) { 2121 spin_unlock(&dentry->d_lock); 2122 /* Start asynchronous writeout of the inode */ 2123 write_inode_now(d_inode(dentry), 0); 2124 error = nfs_sillyrename(dir, dentry); 2125 goto out; 2126 } 2127 if (!d_unhashed(dentry)) { 2128 __d_drop(dentry); 2129 need_rehash = 1; 2130 } 2131 spin_unlock(&dentry->d_lock); 2132 error = nfs_safe_remove(dentry); 2133 if (!error || error == -ENOENT) { 2134 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 2135 } else if (need_rehash) 2136 d_rehash(dentry); 2137 out: 2138 trace_nfs_unlink_exit(dir, dentry, error); 2139 return error; 2140 } 2141 EXPORT_SYMBOL_GPL(nfs_unlink); 2142 2143 /* 2144 * To create a symbolic link, most file systems instantiate a new inode, 2145 * add a page to it containing the path, then write it out to the disk 2146 * using prepare_write/commit_write. 2147 * 2148 * Unfortunately the NFS client can't create the in-core inode first 2149 * because it needs a file handle to create an in-core inode (see 2150 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the 2151 * symlink request has completed on the server. 2152 * 2153 * So instead we allocate a raw page, copy the symname into it, then do 2154 * the SYMLINK request with the page as the buffer. If it succeeds, we 2155 * now have a new file handle and can instantiate an in-core NFS inode 2156 * and move the raw page into its mapping. 2157 */ 2158 int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname) 2159 { 2160 struct page *page; 2161 char *kaddr; 2162 struct iattr attr; 2163 unsigned int pathlen = strlen(symname); 2164 int error; 2165 2166 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s)\n", dir->i_sb->s_id, 2167 dir->i_ino, dentry, symname); 2168 2169 if (pathlen > PAGE_SIZE) 2170 return -ENAMETOOLONG; 2171 2172 attr.ia_mode = S_IFLNK | S_IRWXUGO; 2173 attr.ia_valid = ATTR_MODE; 2174 2175 page = alloc_page(GFP_USER); 2176 if (!page) 2177 return -ENOMEM; 2178 2179 kaddr = page_address(page); 2180 memcpy(kaddr, symname, pathlen); 2181 if (pathlen < PAGE_SIZE) 2182 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen); 2183 2184 trace_nfs_symlink_enter(dir, dentry); 2185 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr); 2186 trace_nfs_symlink_exit(dir, dentry, error); 2187 if (error != 0) { 2188 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s) error %d\n", 2189 dir->i_sb->s_id, dir->i_ino, 2190 dentry, symname, error); 2191 d_drop(dentry); 2192 __free_page(page); 2193 return error; 2194 } 2195 2196 /* 2197 * No big deal if we can't add this page to the page cache here. 2198 * READLINK will get the missing page from the server if needed. 2199 */ 2200 if (!add_to_page_cache_lru(page, d_inode(dentry)->i_mapping, 0, 2201 GFP_KERNEL)) { 2202 SetPageUptodate(page); 2203 unlock_page(page); 2204 /* 2205 * add_to_page_cache_lru() grabs an extra page refcount. 2206 * Drop it here to avoid leaking this page later. 2207 */ 2208 put_page(page); 2209 } else 2210 __free_page(page); 2211 2212 return 0; 2213 } 2214 EXPORT_SYMBOL_GPL(nfs_symlink); 2215 2216 int 2217 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) 2218 { 2219 struct inode *inode = d_inode(old_dentry); 2220 int error; 2221 2222 dfprintk(VFS, "NFS: link(%pd2 -> %pd2)\n", 2223 old_dentry, dentry); 2224 2225 trace_nfs_link_enter(inode, dir, dentry); 2226 d_drop(dentry); 2227 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name); 2228 if (error == 0) { 2229 ihold(inode); 2230 d_add(dentry, inode); 2231 } 2232 trace_nfs_link_exit(inode, dir, dentry, error); 2233 return error; 2234 } 2235 EXPORT_SYMBOL_GPL(nfs_link); 2236 2237 /* 2238 * RENAME 2239 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a 2240 * different file handle for the same inode after a rename (e.g. when 2241 * moving to a different directory). A fail-safe method to do so would 2242 * be to look up old_dir/old_name, create a link to new_dir/new_name and 2243 * rename the old file using the sillyrename stuff. This way, the original 2244 * file in old_dir will go away when the last process iput()s the inode. 2245 * 2246 * FIXED. 2247 * 2248 * It actually works quite well. One needs to have the possibility for 2249 * at least one ".nfs..." file in each directory the file ever gets 2250 * moved or linked to which happens automagically with the new 2251 * implementation that only depends on the dcache stuff instead of 2252 * using the inode layer 2253 * 2254 * Unfortunately, things are a little more complicated than indicated 2255 * above. For a cross-directory move, we want to make sure we can get 2256 * rid of the old inode after the operation. This means there must be 2257 * no pending writes (if it's a file), and the use count must be 1. 2258 * If these conditions are met, we can drop the dentries before doing 2259 * the rename. 2260 */ 2261 int nfs_rename(struct inode *old_dir, struct dentry *old_dentry, 2262 struct inode *new_dir, struct dentry *new_dentry, 2263 unsigned int flags) 2264 { 2265 struct inode *old_inode = d_inode(old_dentry); 2266 struct inode *new_inode = d_inode(new_dentry); 2267 struct dentry *dentry = NULL, *rehash = NULL; 2268 struct rpc_task *task; 2269 int error = -EBUSY; 2270 2271 if (flags) 2272 return -EINVAL; 2273 2274 dfprintk(VFS, "NFS: rename(%pd2 -> %pd2, ct=%d)\n", 2275 old_dentry, new_dentry, 2276 d_count(new_dentry)); 2277 2278 trace_nfs_rename_enter(old_dir, old_dentry, new_dir, new_dentry); 2279 /* 2280 * For non-directories, check whether the target is busy and if so, 2281 * make a copy of the dentry and then do a silly-rename. If the 2282 * silly-rename succeeds, the copied dentry is hashed and becomes 2283 * the new target. 2284 */ 2285 if (new_inode && !S_ISDIR(new_inode->i_mode)) { 2286 /* 2287 * To prevent any new references to the target during the 2288 * rename, we unhash the dentry in advance. 2289 */ 2290 if (!d_unhashed(new_dentry)) { 2291 d_drop(new_dentry); 2292 rehash = new_dentry; 2293 } 2294 2295 if (d_count(new_dentry) > 2) { 2296 int err; 2297 2298 /* copy the target dentry's name */ 2299 dentry = d_alloc(new_dentry->d_parent, 2300 &new_dentry->d_name); 2301 if (!dentry) 2302 goto out; 2303 2304 /* silly-rename the existing target ... */ 2305 err = nfs_sillyrename(new_dir, new_dentry); 2306 if (err) 2307 goto out; 2308 2309 new_dentry = dentry; 2310 rehash = NULL; 2311 new_inode = NULL; 2312 } 2313 } 2314 2315 task = nfs_async_rename(old_dir, new_dir, old_dentry, new_dentry, NULL); 2316 if (IS_ERR(task)) { 2317 error = PTR_ERR(task); 2318 goto out; 2319 } 2320 2321 error = rpc_wait_for_completion_task(task); 2322 if (error != 0) { 2323 ((struct nfs_renamedata *)task->tk_calldata)->cancelled = 1; 2324 /* Paired with the atomic_dec_and_test() barrier in rpc_do_put_task() */ 2325 smp_wmb(); 2326 } else 2327 error = task->tk_status; 2328 rpc_put_task(task); 2329 /* Ensure the inode attributes are revalidated */ 2330 if (error == 0) { 2331 spin_lock(&old_inode->i_lock); 2332 NFS_I(old_inode)->attr_gencount = nfs_inc_attr_generation_counter(); 2333 NFS_I(old_inode)->cache_validity |= NFS_INO_INVALID_CHANGE 2334 | NFS_INO_INVALID_CTIME 2335 | NFS_INO_REVAL_FORCED; 2336 spin_unlock(&old_inode->i_lock); 2337 } 2338 out: 2339 if (rehash) 2340 d_rehash(rehash); 2341 trace_nfs_rename_exit(old_dir, old_dentry, 2342 new_dir, new_dentry, error); 2343 if (!error) { 2344 if (new_inode != NULL) 2345 nfs_drop_nlink(new_inode); 2346 /* 2347 * The d_move() should be here instead of in an async RPC completion 2348 * handler because we need the proper locks to move the dentry. If 2349 * we're interrupted by a signal, the async RPC completion handler 2350 * should mark the directories for revalidation. 2351 */ 2352 d_move(old_dentry, new_dentry); 2353 nfs_set_verifier(old_dentry, 2354 nfs_save_change_attribute(new_dir)); 2355 } else if (error == -ENOENT) 2356 nfs_dentry_handle_enoent(old_dentry); 2357 2358 /* new dentry created? */ 2359 if (dentry) 2360 dput(dentry); 2361 return error; 2362 } 2363 EXPORT_SYMBOL_GPL(nfs_rename); 2364 2365 static DEFINE_SPINLOCK(nfs_access_lru_lock); 2366 static LIST_HEAD(nfs_access_lru_list); 2367 static atomic_long_t nfs_access_nr_entries; 2368 2369 static unsigned long nfs_access_max_cachesize = 4*1024*1024; 2370 module_param(nfs_access_max_cachesize, ulong, 0644); 2371 MODULE_PARM_DESC(nfs_access_max_cachesize, "NFS access maximum total cache length"); 2372 2373 static void nfs_access_free_entry(struct nfs_access_entry *entry) 2374 { 2375 put_cred(entry->cred); 2376 kfree_rcu(entry, rcu_head); 2377 smp_mb__before_atomic(); 2378 atomic_long_dec(&nfs_access_nr_entries); 2379 smp_mb__after_atomic(); 2380 } 2381 2382 static void nfs_access_free_list(struct list_head *head) 2383 { 2384 struct nfs_access_entry *cache; 2385 2386 while (!list_empty(head)) { 2387 cache = list_entry(head->next, struct nfs_access_entry, lru); 2388 list_del(&cache->lru); 2389 nfs_access_free_entry(cache); 2390 } 2391 } 2392 2393 static unsigned long 2394 nfs_do_access_cache_scan(unsigned int nr_to_scan) 2395 { 2396 LIST_HEAD(head); 2397 struct nfs_inode *nfsi, *next; 2398 struct nfs_access_entry *cache; 2399 long freed = 0; 2400 2401 spin_lock(&nfs_access_lru_lock); 2402 list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) { 2403 struct inode *inode; 2404 2405 if (nr_to_scan-- == 0) 2406 break; 2407 inode = &nfsi->vfs_inode; 2408 spin_lock(&inode->i_lock); 2409 if (list_empty(&nfsi->access_cache_entry_lru)) 2410 goto remove_lru_entry; 2411 cache = list_entry(nfsi->access_cache_entry_lru.next, 2412 struct nfs_access_entry, lru); 2413 list_move(&cache->lru, &head); 2414 rb_erase(&cache->rb_node, &nfsi->access_cache); 2415 freed++; 2416 if (!list_empty(&nfsi->access_cache_entry_lru)) 2417 list_move_tail(&nfsi->access_cache_inode_lru, 2418 &nfs_access_lru_list); 2419 else { 2420 remove_lru_entry: 2421 list_del_init(&nfsi->access_cache_inode_lru); 2422 smp_mb__before_atomic(); 2423 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags); 2424 smp_mb__after_atomic(); 2425 } 2426 spin_unlock(&inode->i_lock); 2427 } 2428 spin_unlock(&nfs_access_lru_lock); 2429 nfs_access_free_list(&head); 2430 return freed; 2431 } 2432 2433 unsigned long 2434 nfs_access_cache_scan(struct shrinker *shrink, struct shrink_control *sc) 2435 { 2436 int nr_to_scan = sc->nr_to_scan; 2437 gfp_t gfp_mask = sc->gfp_mask; 2438 2439 if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL) 2440 return SHRINK_STOP; 2441 return nfs_do_access_cache_scan(nr_to_scan); 2442 } 2443 2444 2445 unsigned long 2446 nfs_access_cache_count(struct shrinker *shrink, struct shrink_control *sc) 2447 { 2448 return vfs_pressure_ratio(atomic_long_read(&nfs_access_nr_entries)); 2449 } 2450 2451 static void 2452 nfs_access_cache_enforce_limit(void) 2453 { 2454 long nr_entries = atomic_long_read(&nfs_access_nr_entries); 2455 unsigned long diff; 2456 unsigned int nr_to_scan; 2457 2458 if (nr_entries < 0 || nr_entries <= nfs_access_max_cachesize) 2459 return; 2460 nr_to_scan = 100; 2461 diff = nr_entries - nfs_access_max_cachesize; 2462 if (diff < nr_to_scan) 2463 nr_to_scan = diff; 2464 nfs_do_access_cache_scan(nr_to_scan); 2465 } 2466 2467 static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head) 2468 { 2469 struct rb_root *root_node = &nfsi->access_cache; 2470 struct rb_node *n; 2471 struct nfs_access_entry *entry; 2472 2473 /* Unhook entries from the cache */ 2474 while ((n = rb_first(root_node)) != NULL) { 2475 entry = rb_entry(n, struct nfs_access_entry, rb_node); 2476 rb_erase(n, root_node); 2477 list_move(&entry->lru, head); 2478 } 2479 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS; 2480 } 2481 2482 void nfs_access_zap_cache(struct inode *inode) 2483 { 2484 LIST_HEAD(head); 2485 2486 if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0) 2487 return; 2488 /* Remove from global LRU init */ 2489 spin_lock(&nfs_access_lru_lock); 2490 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) 2491 list_del_init(&NFS_I(inode)->access_cache_inode_lru); 2492 2493 spin_lock(&inode->i_lock); 2494 __nfs_access_zap_cache(NFS_I(inode), &head); 2495 spin_unlock(&inode->i_lock); 2496 spin_unlock(&nfs_access_lru_lock); 2497 nfs_access_free_list(&head); 2498 } 2499 EXPORT_SYMBOL_GPL(nfs_access_zap_cache); 2500 2501 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, const struct cred *cred) 2502 { 2503 struct rb_node *n = NFS_I(inode)->access_cache.rb_node; 2504 2505 while (n != NULL) { 2506 struct nfs_access_entry *entry = 2507 rb_entry(n, struct nfs_access_entry, rb_node); 2508 int cmp = cred_fscmp(cred, entry->cred); 2509 2510 if (cmp < 0) 2511 n = n->rb_left; 2512 else if (cmp > 0) 2513 n = n->rb_right; 2514 else 2515 return entry; 2516 } 2517 return NULL; 2518 } 2519 2520 static int nfs_access_get_cached_locked(struct inode *inode, const struct cred *cred, struct nfs_access_entry *res, bool may_block) 2521 { 2522 struct nfs_inode *nfsi = NFS_I(inode); 2523 struct nfs_access_entry *cache; 2524 bool retry = true; 2525 int err; 2526 2527 spin_lock(&inode->i_lock); 2528 for(;;) { 2529 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS) 2530 goto out_zap; 2531 cache = nfs_access_search_rbtree(inode, cred); 2532 err = -ENOENT; 2533 if (cache == NULL) 2534 goto out; 2535 /* Found an entry, is our attribute cache valid? */ 2536 if (!nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS)) 2537 break; 2538 if (!retry) 2539 break; 2540 err = -ECHILD; 2541 if (!may_block) 2542 goto out; 2543 spin_unlock(&inode->i_lock); 2544 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode); 2545 if (err) 2546 return err; 2547 spin_lock(&inode->i_lock); 2548 retry = false; 2549 } 2550 res->cred = cache->cred; 2551 res->mask = cache->mask; 2552 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru); 2553 err = 0; 2554 out: 2555 spin_unlock(&inode->i_lock); 2556 return err; 2557 out_zap: 2558 spin_unlock(&inode->i_lock); 2559 nfs_access_zap_cache(inode); 2560 return -ENOENT; 2561 } 2562 2563 static int nfs_access_get_cached_rcu(struct inode *inode, const struct cred *cred, struct nfs_access_entry *res) 2564 { 2565 /* Only check the most recently returned cache entry, 2566 * but do it without locking. 2567 */ 2568 struct nfs_inode *nfsi = NFS_I(inode); 2569 struct nfs_access_entry *cache; 2570 int err = -ECHILD; 2571 struct list_head *lh; 2572 2573 rcu_read_lock(); 2574 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS) 2575 goto out; 2576 lh = rcu_dereference(list_tail_rcu(&nfsi->access_cache_entry_lru)); 2577 cache = list_entry(lh, struct nfs_access_entry, lru); 2578 if (lh == &nfsi->access_cache_entry_lru || 2579 cred_fscmp(cred, cache->cred) != 0) 2580 cache = NULL; 2581 if (cache == NULL) 2582 goto out; 2583 if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS)) 2584 goto out; 2585 res->cred = cache->cred; 2586 res->mask = cache->mask; 2587 err = 0; 2588 out: 2589 rcu_read_unlock(); 2590 return err; 2591 } 2592 2593 int nfs_access_get_cached(struct inode *inode, const struct cred *cred, struct 2594 nfs_access_entry *res, bool may_block) 2595 { 2596 int status; 2597 2598 status = nfs_access_get_cached_rcu(inode, cred, res); 2599 if (status != 0) 2600 status = nfs_access_get_cached_locked(inode, cred, res, 2601 may_block); 2602 2603 return status; 2604 } 2605 EXPORT_SYMBOL_GPL(nfs_access_get_cached); 2606 2607 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set) 2608 { 2609 struct nfs_inode *nfsi = NFS_I(inode); 2610 struct rb_root *root_node = &nfsi->access_cache; 2611 struct rb_node **p = &root_node->rb_node; 2612 struct rb_node *parent = NULL; 2613 struct nfs_access_entry *entry; 2614 int cmp; 2615 2616 spin_lock(&inode->i_lock); 2617 while (*p != NULL) { 2618 parent = *p; 2619 entry = rb_entry(parent, struct nfs_access_entry, rb_node); 2620 cmp = cred_fscmp(set->cred, entry->cred); 2621 2622 if (cmp < 0) 2623 p = &parent->rb_left; 2624 else if (cmp > 0) 2625 p = &parent->rb_right; 2626 else 2627 goto found; 2628 } 2629 rb_link_node(&set->rb_node, parent, p); 2630 rb_insert_color(&set->rb_node, root_node); 2631 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); 2632 spin_unlock(&inode->i_lock); 2633 return; 2634 found: 2635 rb_replace_node(parent, &set->rb_node, root_node); 2636 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); 2637 list_del(&entry->lru); 2638 spin_unlock(&inode->i_lock); 2639 nfs_access_free_entry(entry); 2640 } 2641 2642 void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set) 2643 { 2644 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL); 2645 if (cache == NULL) 2646 return; 2647 RB_CLEAR_NODE(&cache->rb_node); 2648 cache->cred = get_cred(set->cred); 2649 cache->mask = set->mask; 2650 2651 /* The above field assignments must be visible 2652 * before this item appears on the lru. We cannot easily 2653 * use rcu_assign_pointer, so just force the memory barrier. 2654 */ 2655 smp_wmb(); 2656 nfs_access_add_rbtree(inode, cache); 2657 2658 /* Update accounting */ 2659 smp_mb__before_atomic(); 2660 atomic_long_inc(&nfs_access_nr_entries); 2661 smp_mb__after_atomic(); 2662 2663 /* Add inode to global LRU list */ 2664 if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) { 2665 spin_lock(&nfs_access_lru_lock); 2666 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) 2667 list_add_tail(&NFS_I(inode)->access_cache_inode_lru, 2668 &nfs_access_lru_list); 2669 spin_unlock(&nfs_access_lru_lock); 2670 } 2671 nfs_access_cache_enforce_limit(); 2672 } 2673 EXPORT_SYMBOL_GPL(nfs_access_add_cache); 2674 2675 #define NFS_MAY_READ (NFS_ACCESS_READ) 2676 #define NFS_MAY_WRITE (NFS_ACCESS_MODIFY | \ 2677 NFS_ACCESS_EXTEND | \ 2678 NFS_ACCESS_DELETE) 2679 #define NFS_FILE_MAY_WRITE (NFS_ACCESS_MODIFY | \ 2680 NFS_ACCESS_EXTEND) 2681 #define NFS_DIR_MAY_WRITE NFS_MAY_WRITE 2682 #define NFS_MAY_LOOKUP (NFS_ACCESS_LOOKUP) 2683 #define NFS_MAY_EXECUTE (NFS_ACCESS_EXECUTE) 2684 static int 2685 nfs_access_calc_mask(u32 access_result, umode_t umode) 2686 { 2687 int mask = 0; 2688 2689 if (access_result & NFS_MAY_READ) 2690 mask |= MAY_READ; 2691 if (S_ISDIR(umode)) { 2692 if ((access_result & NFS_DIR_MAY_WRITE) == NFS_DIR_MAY_WRITE) 2693 mask |= MAY_WRITE; 2694 if ((access_result & NFS_MAY_LOOKUP) == NFS_MAY_LOOKUP) 2695 mask |= MAY_EXEC; 2696 } else if (S_ISREG(umode)) { 2697 if ((access_result & NFS_FILE_MAY_WRITE) == NFS_FILE_MAY_WRITE) 2698 mask |= MAY_WRITE; 2699 if ((access_result & NFS_MAY_EXECUTE) == NFS_MAY_EXECUTE) 2700 mask |= MAY_EXEC; 2701 } else if (access_result & NFS_MAY_WRITE) 2702 mask |= MAY_WRITE; 2703 return mask; 2704 } 2705 2706 void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result) 2707 { 2708 entry->mask = access_result; 2709 } 2710 EXPORT_SYMBOL_GPL(nfs_access_set_mask); 2711 2712 static int nfs_do_access(struct inode *inode, const struct cred *cred, int mask) 2713 { 2714 struct nfs_access_entry cache; 2715 bool may_block = (mask & MAY_NOT_BLOCK) == 0; 2716 int cache_mask = -1; 2717 int status; 2718 2719 trace_nfs_access_enter(inode); 2720 2721 status = nfs_access_get_cached(inode, cred, &cache, may_block); 2722 if (status == 0) 2723 goto out_cached; 2724 2725 status = -ECHILD; 2726 if (!may_block) 2727 goto out; 2728 2729 /* 2730 * Determine which access bits we want to ask for... 2731 */ 2732 cache.mask = NFS_ACCESS_READ | NFS_ACCESS_MODIFY | NFS_ACCESS_EXTEND; 2733 if (nfs_server_capable(inode, NFS_CAP_XATTR)) { 2734 cache.mask |= NFS_ACCESS_XAREAD | NFS_ACCESS_XAWRITE | 2735 NFS_ACCESS_XALIST; 2736 } 2737 if (S_ISDIR(inode->i_mode)) 2738 cache.mask |= NFS_ACCESS_DELETE | NFS_ACCESS_LOOKUP; 2739 else 2740 cache.mask |= NFS_ACCESS_EXECUTE; 2741 cache.cred = cred; 2742 status = NFS_PROTO(inode)->access(inode, &cache); 2743 if (status != 0) { 2744 if (status == -ESTALE) { 2745 if (!S_ISDIR(inode->i_mode)) 2746 nfs_set_inode_stale(inode); 2747 else 2748 nfs_zap_caches(inode); 2749 } 2750 goto out; 2751 } 2752 nfs_access_add_cache(inode, &cache); 2753 out_cached: 2754 cache_mask = nfs_access_calc_mask(cache.mask, inode->i_mode); 2755 if ((mask & ~cache_mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) != 0) 2756 status = -EACCES; 2757 out: 2758 trace_nfs_access_exit(inode, mask, cache_mask, status); 2759 return status; 2760 } 2761 2762 static int nfs_open_permission_mask(int openflags) 2763 { 2764 int mask = 0; 2765 2766 if (openflags & __FMODE_EXEC) { 2767 /* ONLY check exec rights */ 2768 mask = MAY_EXEC; 2769 } else { 2770 if ((openflags & O_ACCMODE) != O_WRONLY) 2771 mask |= MAY_READ; 2772 if ((openflags & O_ACCMODE) != O_RDONLY) 2773 mask |= MAY_WRITE; 2774 } 2775 2776 return mask; 2777 } 2778 2779 int nfs_may_open(struct inode *inode, const struct cred *cred, int openflags) 2780 { 2781 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags)); 2782 } 2783 EXPORT_SYMBOL_GPL(nfs_may_open); 2784 2785 static int nfs_execute_ok(struct inode *inode, int mask) 2786 { 2787 struct nfs_server *server = NFS_SERVER(inode); 2788 int ret = 0; 2789 2790 if (S_ISDIR(inode->i_mode)) 2791 return 0; 2792 if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_OTHER)) { 2793 if (mask & MAY_NOT_BLOCK) 2794 return -ECHILD; 2795 ret = __nfs_revalidate_inode(server, inode); 2796 } 2797 if (ret == 0 && !execute_ok(inode)) 2798 ret = -EACCES; 2799 return ret; 2800 } 2801 2802 int nfs_permission(struct inode *inode, int mask) 2803 { 2804 const struct cred *cred = current_cred(); 2805 int res = 0; 2806 2807 nfs_inc_stats(inode, NFSIOS_VFSACCESS); 2808 2809 if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) 2810 goto out; 2811 /* Is this sys_access() ? */ 2812 if (mask & (MAY_ACCESS | MAY_CHDIR)) 2813 goto force_lookup; 2814 2815 switch (inode->i_mode & S_IFMT) { 2816 case S_IFLNK: 2817 goto out; 2818 case S_IFREG: 2819 if ((mask & MAY_OPEN) && 2820 nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN)) 2821 return 0; 2822 break; 2823 case S_IFDIR: 2824 /* 2825 * Optimize away all write operations, since the server 2826 * will check permissions when we perform the op. 2827 */ 2828 if ((mask & MAY_WRITE) && !(mask & MAY_READ)) 2829 goto out; 2830 } 2831 2832 force_lookup: 2833 if (!NFS_PROTO(inode)->access) 2834 goto out_notsup; 2835 2836 res = nfs_do_access(inode, cred, mask); 2837 out: 2838 if (!res && (mask & MAY_EXEC)) 2839 res = nfs_execute_ok(inode, mask); 2840 2841 dfprintk(VFS, "NFS: permission(%s/%lu), mask=0x%x, res=%d\n", 2842 inode->i_sb->s_id, inode->i_ino, mask, res); 2843 return res; 2844 out_notsup: 2845 if (mask & MAY_NOT_BLOCK) 2846 return -ECHILD; 2847 2848 res = nfs_revalidate_inode(NFS_SERVER(inode), inode); 2849 if (res == 0) 2850 res = generic_permission(inode, mask); 2851 goto out; 2852 } 2853 EXPORT_SYMBOL_GPL(nfs_permission); 2854 2855 /* 2856 * Local variables: 2857 * version-control: t 2858 * kept-new-versions: 5 2859 * End: 2860 */ 2861