1 /* 2 * linux/fs/nfs/dir.c 3 * 4 * Copyright (C) 1992 Rick Sladkey 5 * 6 * nfs directory handling functions 7 * 8 * 10 Apr 1996 Added silly rename for unlink --okir 9 * 28 Sep 1996 Improved directory cache --okir 10 * 23 Aug 1997 Claus Heine claus@momo.math.rwth-aachen.de 11 * Re-implemented silly rename for unlink, newly implemented 12 * silly rename for nfs_rename() following the suggestions 13 * of Olaf Kirch (okir) found in this file. 14 * Following Linus comments on my original hack, this version 15 * depends only on the dcache stuff and doesn't touch the inode 16 * layer (iput() and friends). 17 * 6 Jun 1999 Cache readdir lookups in the page cache. -DaveM 18 */ 19 20 #include <linux/module.h> 21 #include <linux/time.h> 22 #include <linux/errno.h> 23 #include <linux/stat.h> 24 #include <linux/fcntl.h> 25 #include <linux/string.h> 26 #include <linux/kernel.h> 27 #include <linux/slab.h> 28 #include <linux/mm.h> 29 #include <linux/sunrpc/clnt.h> 30 #include <linux/nfs_fs.h> 31 #include <linux/nfs_mount.h> 32 #include <linux/pagemap.h> 33 #include <linux/pagevec.h> 34 #include <linux/namei.h> 35 #include <linux/mount.h> 36 #include <linux/swap.h> 37 #include <linux/sched.h> 38 #include <linux/kmemleak.h> 39 #include <linux/xattr.h> 40 41 #include "delegation.h" 42 #include "iostat.h" 43 #include "internal.h" 44 #include "fscache.h" 45 46 #include "nfstrace.h" 47 48 /* #define NFS_DEBUG_VERBOSE 1 */ 49 50 static int nfs_opendir(struct inode *, struct file *); 51 static int nfs_closedir(struct inode *, struct file *); 52 static int nfs_readdir(struct file *, struct dir_context *); 53 static int nfs_fsync_dir(struct file *, loff_t, loff_t, int); 54 static loff_t nfs_llseek_dir(struct file *, loff_t, int); 55 static void nfs_readdir_clear_array(struct page*); 56 57 const struct file_operations nfs_dir_operations = { 58 .llseek = nfs_llseek_dir, 59 .read = generic_read_dir, 60 .iterate = nfs_readdir, 61 .open = nfs_opendir, 62 .release = nfs_closedir, 63 .fsync = nfs_fsync_dir, 64 }; 65 66 const struct address_space_operations nfs_dir_aops = { 67 .freepage = nfs_readdir_clear_array, 68 }; 69 70 static struct nfs_open_dir_context *alloc_nfs_open_dir_context(struct inode *dir, struct rpc_cred *cred) 71 { 72 struct nfs_inode *nfsi = NFS_I(dir); 73 struct nfs_open_dir_context *ctx; 74 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL); 75 if (ctx != NULL) { 76 ctx->duped = 0; 77 ctx->attr_gencount = nfsi->attr_gencount; 78 ctx->dir_cookie = 0; 79 ctx->dup_cookie = 0; 80 ctx->cred = get_rpccred(cred); 81 spin_lock(&dir->i_lock); 82 list_add(&ctx->list, &nfsi->open_files); 83 spin_unlock(&dir->i_lock); 84 return ctx; 85 } 86 return ERR_PTR(-ENOMEM); 87 } 88 89 static void put_nfs_open_dir_context(struct inode *dir, struct nfs_open_dir_context *ctx) 90 { 91 spin_lock(&dir->i_lock); 92 list_del(&ctx->list); 93 spin_unlock(&dir->i_lock); 94 put_rpccred(ctx->cred); 95 kfree(ctx); 96 } 97 98 /* 99 * Open file 100 */ 101 static int 102 nfs_opendir(struct inode *inode, struct file *filp) 103 { 104 int res = 0; 105 struct nfs_open_dir_context *ctx; 106 struct rpc_cred *cred; 107 108 dfprintk(FILE, "NFS: open dir(%pD2)\n", filp); 109 110 nfs_inc_stats(inode, NFSIOS_VFSOPEN); 111 112 cred = rpc_lookup_cred(); 113 if (IS_ERR(cred)) 114 return PTR_ERR(cred); 115 ctx = alloc_nfs_open_dir_context(inode, cred); 116 if (IS_ERR(ctx)) { 117 res = PTR_ERR(ctx); 118 goto out; 119 } 120 filp->private_data = ctx; 121 if (filp->f_path.dentry == filp->f_path.mnt->mnt_root) { 122 /* This is a mountpoint, so d_revalidate will never 123 * have been called, so we need to refresh the 124 * inode (for close-open consistency) ourselves. 125 */ 126 __nfs_revalidate_inode(NFS_SERVER(inode), inode); 127 } 128 out: 129 put_rpccred(cred); 130 return res; 131 } 132 133 static int 134 nfs_closedir(struct inode *inode, struct file *filp) 135 { 136 put_nfs_open_dir_context(file_inode(filp), filp->private_data); 137 return 0; 138 } 139 140 struct nfs_cache_array_entry { 141 u64 cookie; 142 u64 ino; 143 struct qstr string; 144 unsigned char d_type; 145 }; 146 147 struct nfs_cache_array { 148 int size; 149 int eof_index; 150 u64 last_cookie; 151 struct nfs_cache_array_entry array[0]; 152 }; 153 154 typedef int (*decode_dirent_t)(struct xdr_stream *, struct nfs_entry *, int); 155 typedef struct { 156 struct file *file; 157 struct page *page; 158 struct dir_context *ctx; 159 unsigned long page_index; 160 u64 *dir_cookie; 161 u64 last_cookie; 162 loff_t current_index; 163 decode_dirent_t decode; 164 165 unsigned long timestamp; 166 unsigned long gencount; 167 unsigned int cache_entry_index; 168 unsigned int plus:1; 169 unsigned int eof:1; 170 } nfs_readdir_descriptor_t; 171 172 /* 173 * The caller is responsible for calling nfs_readdir_release_array(page) 174 */ 175 static 176 struct nfs_cache_array *nfs_readdir_get_array(struct page *page) 177 { 178 void *ptr; 179 if (page == NULL) 180 return ERR_PTR(-EIO); 181 ptr = kmap(page); 182 if (ptr == NULL) 183 return ERR_PTR(-ENOMEM); 184 return ptr; 185 } 186 187 static 188 void nfs_readdir_release_array(struct page *page) 189 { 190 kunmap(page); 191 } 192 193 /* 194 * we are freeing strings created by nfs_add_to_readdir_array() 195 */ 196 static 197 void nfs_readdir_clear_array(struct page *page) 198 { 199 struct nfs_cache_array *array; 200 int i; 201 202 array = kmap_atomic(page); 203 for (i = 0; i < array->size; i++) 204 kfree(array->array[i].string.name); 205 kunmap_atomic(array); 206 } 207 208 /* 209 * the caller is responsible for freeing qstr.name 210 * when called by nfs_readdir_add_to_array, the strings will be freed in 211 * nfs_clear_readdir_array() 212 */ 213 static 214 int nfs_readdir_make_qstr(struct qstr *string, const char *name, unsigned int len) 215 { 216 string->len = len; 217 string->name = kmemdup(name, len, GFP_KERNEL); 218 if (string->name == NULL) 219 return -ENOMEM; 220 /* 221 * Avoid a kmemleak false positive. The pointer to the name is stored 222 * in a page cache page which kmemleak does not scan. 223 */ 224 kmemleak_not_leak(string->name); 225 string->hash = full_name_hash(name, len); 226 return 0; 227 } 228 229 static 230 int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page) 231 { 232 struct nfs_cache_array *array = nfs_readdir_get_array(page); 233 struct nfs_cache_array_entry *cache_entry; 234 int ret; 235 236 if (IS_ERR(array)) 237 return PTR_ERR(array); 238 239 cache_entry = &array->array[array->size]; 240 241 /* Check that this entry lies within the page bounds */ 242 ret = -ENOSPC; 243 if ((char *)&cache_entry[1] - (char *)page_address(page) > PAGE_SIZE) 244 goto out; 245 246 cache_entry->cookie = entry->prev_cookie; 247 cache_entry->ino = entry->ino; 248 cache_entry->d_type = entry->d_type; 249 ret = nfs_readdir_make_qstr(&cache_entry->string, entry->name, entry->len); 250 if (ret) 251 goto out; 252 array->last_cookie = entry->cookie; 253 array->size++; 254 if (entry->eof != 0) 255 array->eof_index = array->size; 256 out: 257 nfs_readdir_release_array(page); 258 return ret; 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 = 1; 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->ctx->pos) { 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 desc->ctx->pos = new_pos; 327 desc->cache_entry_index = i; 328 return 0; 329 } 330 } 331 if (array->eof_index >= 0) { 332 status = -EBADCOOKIE; 333 if (*desc->dir_cookie == array->last_cookie) 334 desc->eof = 1; 335 } 336 out: 337 return status; 338 } 339 340 static 341 int nfs_readdir_search_array(nfs_readdir_descriptor_t *desc) 342 { 343 struct nfs_cache_array *array; 344 int status; 345 346 array = nfs_readdir_get_array(desc->page); 347 if (IS_ERR(array)) { 348 status = PTR_ERR(array); 349 goto out; 350 } 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 nfs_readdir_release_array(desc->page); 363 out: 364 return status; 365 } 366 367 /* Fill a page with xdr information before transferring to the cache page */ 368 static 369 int nfs_readdir_xdr_filler(struct page **pages, nfs_readdir_descriptor_t *desc, 370 struct nfs_entry *entry, struct file *file, struct inode *inode) 371 { 372 struct nfs_open_dir_context *ctx = file->private_data; 373 struct rpc_cred *cred = ctx->cred; 374 unsigned long timestamp, gencount; 375 int error; 376 377 again: 378 timestamp = jiffies; 379 gencount = nfs_inc_attr_generation_counter(); 380 error = NFS_PROTO(inode)->readdir(file->f_path.dentry, 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 = 0; 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 int error; 402 403 error = desc->decode(xdr, entry, desc->plus); 404 if (error) 405 return error; 406 entry->fattr->time_start = desc->timestamp; 407 entry->fattr->gencount = desc->gencount; 408 return 0; 409 } 410 411 /* Match file and dirent using either filehandle or fileid 412 * Note: caller is responsible for checking the fsid 413 */ 414 static 415 int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry) 416 { 417 struct nfs_inode *nfsi; 418 419 if (dentry->d_inode == NULL) 420 goto different; 421 422 nfsi = NFS_I(dentry->d_inode); 423 if (entry->fattr->fileid == nfsi->fileid) 424 return 1; 425 if (nfs_compare_fh(entry->fh, &nfsi->fh) == 0) 426 return 1; 427 different: 428 return 0; 429 } 430 431 static 432 bool nfs_use_readdirplus(struct inode *dir, struct dir_context *ctx) 433 { 434 if (!nfs_server_capable(dir, NFS_CAP_READDIRPLUS)) 435 return false; 436 if (test_and_clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags)) 437 return true; 438 if (ctx->pos == 0) 439 return true; 440 return false; 441 } 442 443 /* 444 * This function is called by the lookup code to request the use of 445 * readdirplus to accelerate any future lookups in the same 446 * directory. 447 */ 448 static 449 void nfs_advise_use_readdirplus(struct inode *dir) 450 { 451 set_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags); 452 } 453 454 /* 455 * This function is mainly for use by nfs_getattr(). 456 * 457 * If this is an 'ls -l', we want to force use of readdirplus. 458 * Do this by checking if there is an active file descriptor 459 * and calling nfs_advise_use_readdirplus, then forcing a 460 * cache flush. 461 */ 462 void nfs_force_use_readdirplus(struct inode *dir) 463 { 464 if (!list_empty(&NFS_I(dir)->open_files)) { 465 nfs_advise_use_readdirplus(dir); 466 nfs_zap_mapping(dir, dir->i_mapping); 467 } 468 } 469 470 static 471 void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry) 472 { 473 struct qstr filename = QSTR_INIT(entry->name, entry->len); 474 struct dentry *dentry; 475 struct dentry *alias; 476 struct inode *dir = parent->d_inode; 477 struct inode *inode; 478 int status; 479 480 if (!(entry->fattr->valid & NFS_ATTR_FATTR_FILEID)) 481 return; 482 if (!(entry->fattr->valid & NFS_ATTR_FATTR_FSID)) 483 return; 484 if (filename.name[0] == '.') { 485 if (filename.len == 1) 486 return; 487 if (filename.len == 2 && filename.name[1] == '.') 488 return; 489 } 490 filename.hash = full_name_hash(filename.name, filename.len); 491 492 dentry = d_lookup(parent, &filename); 493 if (dentry != NULL) { 494 /* Is there a mountpoint here? If so, just exit */ 495 if (!nfs_fsid_equal(&NFS_SB(dentry->d_sb)->fsid, 496 &entry->fattr->fsid)) 497 goto out; 498 if (nfs_same_file(dentry, entry)) { 499 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 500 status = nfs_refresh_inode(dentry->d_inode, entry->fattr); 501 if (!status) 502 nfs_setsecurity(dentry->d_inode, entry->fattr, entry->label); 503 goto out; 504 } else { 505 d_invalidate(dentry); 506 dput(dentry); 507 } 508 } 509 510 dentry = d_alloc(parent, &filename); 511 if (dentry == NULL) 512 return; 513 514 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr, entry->label); 515 if (IS_ERR(inode)) 516 goto out; 517 518 alias = d_splice_alias(inode, dentry); 519 if (IS_ERR(alias)) 520 goto out; 521 else if (alias) { 522 nfs_set_verifier(alias, nfs_save_change_attribute(dir)); 523 dput(alias); 524 } else 525 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 526 527 out: 528 dput(dentry); 529 } 530 531 /* Perform conversion from xdr to cache array */ 532 static 533 int nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry, 534 struct page **xdr_pages, struct page *page, unsigned int buflen) 535 { 536 struct xdr_stream stream; 537 struct xdr_buf buf; 538 struct page *scratch; 539 struct nfs_cache_array *array; 540 unsigned int count = 0; 541 int status; 542 543 scratch = alloc_page(GFP_KERNEL); 544 if (scratch == NULL) 545 return -ENOMEM; 546 547 if (buflen == 0) 548 goto out_nopages; 549 550 xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen); 551 xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE); 552 553 do { 554 status = xdr_decode(desc, entry, &stream); 555 if (status != 0) { 556 if (status == -EAGAIN) 557 status = 0; 558 break; 559 } 560 561 count++; 562 563 if (desc->plus != 0) 564 nfs_prime_dcache(desc->file->f_path.dentry, entry); 565 566 status = nfs_readdir_add_to_array(entry, page); 567 if (status != 0) 568 break; 569 } while (!entry->eof); 570 571 out_nopages: 572 if (count == 0 || (status == -EBADCOOKIE && entry->eof != 0)) { 573 array = nfs_readdir_get_array(page); 574 if (!IS_ERR(array)) { 575 array->eof_index = array->size; 576 status = 0; 577 nfs_readdir_release_array(page); 578 } else 579 status = PTR_ERR(array); 580 } 581 582 put_page(scratch); 583 return status; 584 } 585 586 static 587 void nfs_readdir_free_pagearray(struct page **pages, unsigned int npages) 588 { 589 unsigned int i; 590 for (i = 0; i < npages; i++) 591 put_page(pages[i]); 592 } 593 594 static 595 void nfs_readdir_free_large_page(void *ptr, struct page **pages, 596 unsigned int npages) 597 { 598 nfs_readdir_free_pagearray(pages, npages); 599 } 600 601 /* 602 * nfs_readdir_large_page will allocate pages that must be freed with a call 603 * to nfs_readdir_free_large_page 604 */ 605 static 606 int nfs_readdir_large_page(struct page **pages, unsigned int npages) 607 { 608 unsigned int i; 609 610 for (i = 0; i < npages; i++) { 611 struct page *page = alloc_page(GFP_KERNEL); 612 if (page == NULL) 613 goto out_freepages; 614 pages[i] = page; 615 } 616 return 0; 617 618 out_freepages: 619 nfs_readdir_free_pagearray(pages, i); 620 return -ENOMEM; 621 } 622 623 static 624 int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t *desc, struct page *page, struct inode *inode) 625 { 626 struct page *pages[NFS_MAX_READDIR_PAGES]; 627 void *pages_ptr = NULL; 628 struct nfs_entry entry; 629 struct file *file = desc->file; 630 struct nfs_cache_array *array; 631 int status = -ENOMEM; 632 unsigned int array_size = ARRAY_SIZE(pages); 633 634 entry.prev_cookie = 0; 635 entry.cookie = desc->last_cookie; 636 entry.eof = 0; 637 entry.fh = nfs_alloc_fhandle(); 638 entry.fattr = nfs_alloc_fattr(); 639 entry.server = NFS_SERVER(inode); 640 if (entry.fh == NULL || entry.fattr == NULL) 641 goto out; 642 643 entry.label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT); 644 if (IS_ERR(entry.label)) { 645 status = PTR_ERR(entry.label); 646 goto out; 647 } 648 649 array = nfs_readdir_get_array(page); 650 if (IS_ERR(array)) { 651 status = PTR_ERR(array); 652 goto out_label_free; 653 } 654 memset(array, 0, sizeof(struct nfs_cache_array)); 655 array->eof_index = -1; 656 657 status = nfs_readdir_large_page(pages, array_size); 658 if (status < 0) 659 goto out_release_array; 660 do { 661 unsigned int pglen; 662 status = nfs_readdir_xdr_filler(pages, desc, &entry, file, inode); 663 664 if (status < 0) 665 break; 666 pglen = status; 667 status = nfs_readdir_page_filler(desc, &entry, pages, page, pglen); 668 if (status < 0) { 669 if (status == -ENOSPC) 670 status = 0; 671 break; 672 } 673 } while (array->eof_index < 0); 674 675 nfs_readdir_free_large_page(pages_ptr, pages, array_size); 676 out_release_array: 677 nfs_readdir_release_array(page); 678 out_label_free: 679 nfs4_label_free(entry.label); 680 out: 681 nfs_free_fattr(entry.fattr); 682 nfs_free_fhandle(entry.fh); 683 return status; 684 } 685 686 /* 687 * Now we cache directories properly, by converting xdr information 688 * to an array that can be used for lookups later. This results in 689 * fewer cache pages, since we can store more information on each page. 690 * We only need to convert from xdr once so future lookups are much simpler 691 */ 692 static 693 int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page* page) 694 { 695 struct inode *inode = file_inode(desc->file); 696 int ret; 697 698 ret = nfs_readdir_xdr_to_array(desc, page, inode); 699 if (ret < 0) 700 goto error; 701 SetPageUptodate(page); 702 703 if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) { 704 /* Should never happen */ 705 nfs_zap_mapping(inode, inode->i_mapping); 706 } 707 unlock_page(page); 708 return 0; 709 error: 710 unlock_page(page); 711 return ret; 712 } 713 714 static 715 void cache_page_release(nfs_readdir_descriptor_t *desc) 716 { 717 if (!desc->page->mapping) 718 nfs_readdir_clear_array(desc->page); 719 page_cache_release(desc->page); 720 desc->page = NULL; 721 } 722 723 static 724 struct page *get_cache_page(nfs_readdir_descriptor_t *desc) 725 { 726 return read_cache_page(file_inode(desc->file)->i_mapping, 727 desc->page_index, (filler_t *)nfs_readdir_filler, desc); 728 } 729 730 /* 731 * Returns 0 if desc->dir_cookie was found on page desc->page_index 732 */ 733 static 734 int find_cache_page(nfs_readdir_descriptor_t *desc) 735 { 736 int res; 737 738 desc->page = get_cache_page(desc); 739 if (IS_ERR(desc->page)) 740 return PTR_ERR(desc->page); 741 742 res = nfs_readdir_search_array(desc); 743 if (res != 0) 744 cache_page_release(desc); 745 return res; 746 } 747 748 /* Search for desc->dir_cookie from the beginning of the page cache */ 749 static inline 750 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc) 751 { 752 int res; 753 754 if (desc->page_index == 0) { 755 desc->current_index = 0; 756 desc->last_cookie = 0; 757 } 758 do { 759 res = find_cache_page(desc); 760 } while (res == -EAGAIN); 761 return res; 762 } 763 764 /* 765 * Once we've found the start of the dirent within a page: fill 'er up... 766 */ 767 static 768 int nfs_do_filldir(nfs_readdir_descriptor_t *desc) 769 { 770 struct file *file = desc->file; 771 int i = 0; 772 int res = 0; 773 struct nfs_cache_array *array = NULL; 774 struct nfs_open_dir_context *ctx = file->private_data; 775 776 array = nfs_readdir_get_array(desc->page); 777 if (IS_ERR(array)) { 778 res = PTR_ERR(array); 779 goto out; 780 } 781 782 for (i = desc->cache_entry_index; i < array->size; i++) { 783 struct nfs_cache_array_entry *ent; 784 785 ent = &array->array[i]; 786 if (!dir_emit(desc->ctx, ent->string.name, ent->string.len, 787 nfs_compat_user_ino64(ent->ino), ent->d_type)) { 788 desc->eof = 1; 789 break; 790 } 791 desc->ctx->pos++; 792 if (i < (array->size-1)) 793 *desc->dir_cookie = array->array[i+1].cookie; 794 else 795 *desc->dir_cookie = array->last_cookie; 796 if (ctx->duped != 0) 797 ctx->duped = 1; 798 } 799 if (array->eof_index >= 0) 800 desc->eof = 1; 801 802 nfs_readdir_release_array(desc->page); 803 out: 804 cache_page_release(desc); 805 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n", 806 (unsigned long long)*desc->dir_cookie, res); 807 return res; 808 } 809 810 /* 811 * If we cannot find a cookie in our cache, we suspect that this is 812 * because it points to a deleted file, so we ask the server to return 813 * whatever it thinks is the next entry. We then feed this to filldir. 814 * If all goes well, we should then be able to find our way round the 815 * cache on the next call to readdir_search_pagecache(); 816 * 817 * NOTE: we cannot add the anonymous page to the pagecache because 818 * the data it contains might not be page aligned. Besides, 819 * we should already have a complete representation of the 820 * directory in the page cache by the time we get here. 821 */ 822 static inline 823 int uncached_readdir(nfs_readdir_descriptor_t *desc) 824 { 825 struct page *page = NULL; 826 int status; 827 struct inode *inode = file_inode(desc->file); 828 struct nfs_open_dir_context *ctx = desc->file->private_data; 829 830 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n", 831 (unsigned long long)*desc->dir_cookie); 832 833 page = alloc_page(GFP_HIGHUSER); 834 if (!page) { 835 status = -ENOMEM; 836 goto out; 837 } 838 839 desc->page_index = 0; 840 desc->last_cookie = *desc->dir_cookie; 841 desc->page = page; 842 ctx->duped = 0; 843 844 status = nfs_readdir_xdr_to_array(desc, page, inode); 845 if (status < 0) 846 goto out_release; 847 848 status = nfs_do_filldir(desc); 849 850 out: 851 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", 852 __func__, status); 853 return status; 854 out_release: 855 cache_page_release(desc); 856 goto out; 857 } 858 859 static bool nfs_dir_mapping_need_revalidate(struct inode *dir) 860 { 861 struct nfs_inode *nfsi = NFS_I(dir); 862 863 if (nfs_attribute_cache_expired(dir)) 864 return true; 865 if (nfsi->cache_validity & NFS_INO_INVALID_DATA) 866 return true; 867 return false; 868 } 869 870 /* The file offset position represents the dirent entry number. A 871 last cookie cache takes care of the common case of reading the 872 whole directory. 873 */ 874 static int nfs_readdir(struct file *file, struct dir_context *ctx) 875 { 876 struct dentry *dentry = file->f_path.dentry; 877 struct inode *inode = dentry->d_inode; 878 nfs_readdir_descriptor_t my_desc, 879 *desc = &my_desc; 880 struct nfs_open_dir_context *dir_ctx = file->private_data; 881 int res = 0; 882 883 dfprintk(FILE, "NFS: readdir(%pD2) starting at cookie %llu\n", 884 file, (long long)ctx->pos); 885 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS); 886 887 /* 888 * ctx->pos points to the dirent entry number. 889 * *desc->dir_cookie has the cookie for the next entry. We have 890 * to either find the entry with the appropriate number or 891 * revalidate the cookie. 892 */ 893 memset(desc, 0, sizeof(*desc)); 894 895 desc->file = file; 896 desc->ctx = ctx; 897 desc->dir_cookie = &dir_ctx->dir_cookie; 898 desc->decode = NFS_PROTO(inode)->decode_dirent; 899 desc->plus = nfs_use_readdirplus(inode, ctx) ? 1 : 0; 900 901 nfs_block_sillyrename(dentry); 902 if (ctx->pos == 0 || nfs_dir_mapping_need_revalidate(inode)) 903 res = nfs_revalidate_mapping(inode, file->f_mapping); 904 if (res < 0) 905 goto out; 906 907 do { 908 res = readdir_search_pagecache(desc); 909 910 if (res == -EBADCOOKIE) { 911 res = 0; 912 /* This means either end of directory */ 913 if (*desc->dir_cookie && desc->eof == 0) { 914 /* Or that the server has 'lost' a cookie */ 915 res = uncached_readdir(desc); 916 if (res == 0) 917 continue; 918 } 919 break; 920 } 921 if (res == -ETOOSMALL && desc->plus) { 922 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags); 923 nfs_zap_caches(inode); 924 desc->page_index = 0; 925 desc->plus = 0; 926 desc->eof = 0; 927 continue; 928 } 929 if (res < 0) 930 break; 931 932 res = nfs_do_filldir(desc); 933 if (res < 0) 934 break; 935 } while (!desc->eof); 936 out: 937 nfs_unblock_sillyrename(dentry); 938 if (res > 0) 939 res = 0; 940 dfprintk(FILE, "NFS: readdir(%pD2) returns %d\n", file, res); 941 return res; 942 } 943 944 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int whence) 945 { 946 struct inode *inode = file_inode(filp); 947 struct nfs_open_dir_context *dir_ctx = filp->private_data; 948 949 dfprintk(FILE, "NFS: llseek dir(%pD2, %lld, %d)\n", 950 filp, offset, whence); 951 952 mutex_lock(&inode->i_mutex); 953 switch (whence) { 954 case 1: 955 offset += filp->f_pos; 956 case 0: 957 if (offset >= 0) 958 break; 959 default: 960 offset = -EINVAL; 961 goto out; 962 } 963 if (offset != filp->f_pos) { 964 filp->f_pos = offset; 965 dir_ctx->dir_cookie = 0; 966 dir_ctx->duped = 0; 967 } 968 out: 969 mutex_unlock(&inode->i_mutex); 970 return offset; 971 } 972 973 /* 974 * All directory operations under NFS are synchronous, so fsync() 975 * is a dummy operation. 976 */ 977 static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end, 978 int datasync) 979 { 980 struct inode *inode = file_inode(filp); 981 982 dfprintk(FILE, "NFS: fsync dir(%pD2) datasync %d\n", filp, datasync); 983 984 mutex_lock(&inode->i_mutex); 985 nfs_inc_stats(inode, NFSIOS_VFSFSYNC); 986 mutex_unlock(&inode->i_mutex); 987 return 0; 988 } 989 990 /** 991 * nfs_force_lookup_revalidate - Mark the directory as having changed 992 * @dir - pointer to directory inode 993 * 994 * This forces the revalidation code in nfs_lookup_revalidate() to do a 995 * full lookup on all child dentries of 'dir' whenever a change occurs 996 * on the server that might have invalidated our dcache. 997 * 998 * The caller should be holding dir->i_lock 999 */ 1000 void nfs_force_lookup_revalidate(struct inode *dir) 1001 { 1002 NFS_I(dir)->cache_change_attribute++; 1003 } 1004 EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate); 1005 1006 /* 1007 * A check for whether or not the parent directory has changed. 1008 * In the case it has, we assume that the dentries are untrustworthy 1009 * and may need to be looked up again. 1010 * If rcu_walk prevents us from performing a full check, return 0. 1011 */ 1012 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry, 1013 int rcu_walk) 1014 { 1015 int ret; 1016 1017 if (IS_ROOT(dentry)) 1018 return 1; 1019 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE) 1020 return 0; 1021 if (!nfs_verify_change_attribute(dir, dentry->d_time)) 1022 return 0; 1023 /* Revalidate nfsi->cache_change_attribute before we declare a match */ 1024 if (rcu_walk) 1025 ret = nfs_revalidate_inode_rcu(NFS_SERVER(dir), dir); 1026 else 1027 ret = nfs_revalidate_inode(NFS_SERVER(dir), dir); 1028 if (ret < 0) 1029 return 0; 1030 if (!nfs_verify_change_attribute(dir, dentry->d_time)) 1031 return 0; 1032 return 1; 1033 } 1034 1035 /* 1036 * Use intent information to check whether or not we're going to do 1037 * an O_EXCL create using this path component. 1038 */ 1039 static int nfs_is_exclusive_create(struct inode *dir, unsigned int flags) 1040 { 1041 if (NFS_PROTO(dir)->version == 2) 1042 return 0; 1043 return flags & LOOKUP_EXCL; 1044 } 1045 1046 /* 1047 * Inode and filehandle revalidation for lookups. 1048 * 1049 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL, 1050 * or if the intent information indicates that we're about to open this 1051 * particular file and the "nocto" mount flag is not set. 1052 * 1053 */ 1054 static 1055 int nfs_lookup_verify_inode(struct inode *inode, unsigned int flags) 1056 { 1057 struct nfs_server *server = NFS_SERVER(inode); 1058 int ret; 1059 1060 if (IS_AUTOMOUNT(inode)) 1061 return 0; 1062 /* VFS wants an on-the-wire revalidation */ 1063 if (flags & LOOKUP_REVAL) 1064 goto out_force; 1065 /* This is an open(2) */ 1066 if ((flags & LOOKUP_OPEN) && !(server->flags & NFS_MOUNT_NOCTO) && 1067 (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode))) 1068 goto out_force; 1069 out: 1070 return (inode->i_nlink == 0) ? -ENOENT : 0; 1071 out_force: 1072 if (flags & LOOKUP_RCU) 1073 return -ECHILD; 1074 ret = __nfs_revalidate_inode(server, inode); 1075 if (ret != 0) 1076 return ret; 1077 goto out; 1078 } 1079 1080 /* 1081 * We judge how long we want to trust negative 1082 * dentries by looking at the parent inode mtime. 1083 * 1084 * If parent mtime has changed, we revalidate, else we wait for a 1085 * period corresponding to the parent's attribute cache timeout value. 1086 * 1087 * If LOOKUP_RCU prevents us from performing a full check, return 1 1088 * suggesting a reval is needed. 1089 */ 1090 static inline 1091 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry, 1092 unsigned int flags) 1093 { 1094 /* Don't revalidate a negative dentry if we're creating a new file */ 1095 if (flags & LOOKUP_CREATE) 1096 return 0; 1097 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG) 1098 return 1; 1099 return !nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU); 1100 } 1101 1102 /* 1103 * This is called every time the dcache has a lookup hit, 1104 * and we should check whether we can really trust that 1105 * lookup. 1106 * 1107 * NOTE! The hit can be a negative hit too, don't assume 1108 * we have an inode! 1109 * 1110 * If the parent directory is seen to have changed, we throw out the 1111 * cached dentry and do a new lookup. 1112 */ 1113 static int nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags) 1114 { 1115 struct inode *dir; 1116 struct inode *inode; 1117 struct dentry *parent; 1118 struct nfs_fh *fhandle = NULL; 1119 struct nfs_fattr *fattr = NULL; 1120 struct nfs4_label *label = NULL; 1121 int error; 1122 1123 if (flags & LOOKUP_RCU) { 1124 parent = ACCESS_ONCE(dentry->d_parent); 1125 dir = ACCESS_ONCE(parent->d_inode); 1126 if (!dir) 1127 return -ECHILD; 1128 } else { 1129 parent = dget_parent(dentry); 1130 dir = parent->d_inode; 1131 } 1132 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE); 1133 inode = dentry->d_inode; 1134 1135 if (!inode) { 1136 if (nfs_neg_need_reval(dir, dentry, flags)) { 1137 if (flags & LOOKUP_RCU) 1138 return -ECHILD; 1139 goto out_bad; 1140 } 1141 goto out_valid_noent; 1142 } 1143 1144 if (is_bad_inode(inode)) { 1145 if (flags & LOOKUP_RCU) 1146 return -ECHILD; 1147 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n", 1148 __func__, dentry); 1149 goto out_bad; 1150 } 1151 1152 if (NFS_PROTO(dir)->have_delegation(inode, FMODE_READ)) 1153 goto out_set_verifier; 1154 1155 /* Force a full look up iff the parent directory has changed */ 1156 if (!nfs_is_exclusive_create(dir, flags) && 1157 nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU)) { 1158 1159 if (nfs_lookup_verify_inode(inode, flags)) { 1160 if (flags & LOOKUP_RCU) 1161 return -ECHILD; 1162 goto out_zap_parent; 1163 } 1164 goto out_valid; 1165 } 1166 1167 if (flags & LOOKUP_RCU) 1168 return -ECHILD; 1169 1170 if (NFS_STALE(inode)) 1171 goto out_bad; 1172 1173 error = -ENOMEM; 1174 fhandle = nfs_alloc_fhandle(); 1175 fattr = nfs_alloc_fattr(); 1176 if (fhandle == NULL || fattr == NULL) 1177 goto out_error; 1178 1179 label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT); 1180 if (IS_ERR(label)) 1181 goto out_error; 1182 1183 trace_nfs_lookup_revalidate_enter(dir, dentry, flags); 1184 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, label); 1185 trace_nfs_lookup_revalidate_exit(dir, dentry, flags, error); 1186 if (error) 1187 goto out_bad; 1188 if (nfs_compare_fh(NFS_FH(inode), fhandle)) 1189 goto out_bad; 1190 if ((error = nfs_refresh_inode(inode, fattr)) != 0) 1191 goto out_bad; 1192 1193 nfs_setsecurity(inode, fattr, label); 1194 1195 nfs_free_fattr(fattr); 1196 nfs_free_fhandle(fhandle); 1197 nfs4_label_free(label); 1198 1199 out_set_verifier: 1200 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1201 out_valid: 1202 /* Success: notify readdir to use READDIRPLUS */ 1203 nfs_advise_use_readdirplus(dir); 1204 out_valid_noent: 1205 if (flags & LOOKUP_RCU) { 1206 if (parent != ACCESS_ONCE(dentry->d_parent)) 1207 return -ECHILD; 1208 } else 1209 dput(parent); 1210 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is valid\n", 1211 __func__, dentry); 1212 return 1; 1213 out_zap_parent: 1214 nfs_zap_caches(dir); 1215 out_bad: 1216 WARN_ON(flags & LOOKUP_RCU); 1217 nfs_free_fattr(fattr); 1218 nfs_free_fhandle(fhandle); 1219 nfs4_label_free(label); 1220 nfs_mark_for_revalidate(dir); 1221 if (inode && S_ISDIR(inode->i_mode)) { 1222 /* Purge readdir caches. */ 1223 nfs_zap_caches(inode); 1224 /* 1225 * We can't d_drop the root of a disconnected tree: 1226 * its d_hash is on the s_anon list and d_drop() would hide 1227 * it from shrink_dcache_for_unmount(), leading to busy 1228 * inodes on unmount and further oopses. 1229 */ 1230 if (IS_ROOT(dentry)) 1231 goto out_valid; 1232 } 1233 dput(parent); 1234 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is invalid\n", 1235 __func__, dentry); 1236 return 0; 1237 out_error: 1238 WARN_ON(flags & LOOKUP_RCU); 1239 nfs_free_fattr(fattr); 1240 nfs_free_fhandle(fhandle); 1241 nfs4_label_free(label); 1242 dput(parent); 1243 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) lookup returned error %d\n", 1244 __func__, dentry, error); 1245 return error; 1246 } 1247 1248 /* 1249 * A weaker form of d_revalidate for revalidating just the dentry->d_inode 1250 * when we don't really care about the dentry name. This is called when a 1251 * pathwalk ends on a dentry that was not found via a normal lookup in the 1252 * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals). 1253 * 1254 * In this situation, we just want to verify that the inode itself is OK 1255 * since the dentry might have changed on the server. 1256 */ 1257 static int nfs_weak_revalidate(struct dentry *dentry, unsigned int flags) 1258 { 1259 int error; 1260 struct inode *inode = dentry->d_inode; 1261 1262 /* 1263 * I believe we can only get a negative dentry here in the case of a 1264 * procfs-style symlink. Just assume it's correct for now, but we may 1265 * eventually need to do something more here. 1266 */ 1267 if (!inode) { 1268 dfprintk(LOOKUPCACHE, "%s: %pd2 has negative inode\n", 1269 __func__, dentry); 1270 return 1; 1271 } 1272 1273 if (is_bad_inode(inode)) { 1274 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n", 1275 __func__, dentry); 1276 return 0; 1277 } 1278 1279 error = nfs_revalidate_inode(NFS_SERVER(inode), inode); 1280 dfprintk(LOOKUPCACHE, "NFS: %s: inode %lu is %s\n", 1281 __func__, inode->i_ino, error ? "invalid" : "valid"); 1282 return !error; 1283 } 1284 1285 /* 1286 * This is called from dput() when d_count is going to 0. 1287 */ 1288 static int nfs_dentry_delete(const struct dentry *dentry) 1289 { 1290 dfprintk(VFS, "NFS: dentry_delete(%pd2, %x)\n", 1291 dentry, dentry->d_flags); 1292 1293 /* Unhash any dentry with a stale inode */ 1294 if (dentry->d_inode != NULL && NFS_STALE(dentry->d_inode)) 1295 return 1; 1296 1297 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { 1298 /* Unhash it, so that ->d_iput() would be called */ 1299 return 1; 1300 } 1301 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) { 1302 /* Unhash it, so that ancestors of killed async unlink 1303 * files will be cleaned up during umount */ 1304 return 1; 1305 } 1306 return 0; 1307 1308 } 1309 1310 /* Ensure that we revalidate inode->i_nlink */ 1311 static void nfs_drop_nlink(struct inode *inode) 1312 { 1313 spin_lock(&inode->i_lock); 1314 /* drop the inode if we're reasonably sure this is the last link */ 1315 if (inode->i_nlink == 1) 1316 clear_nlink(inode); 1317 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATTR; 1318 spin_unlock(&inode->i_lock); 1319 } 1320 1321 /* 1322 * Called when the dentry loses inode. 1323 * We use it to clean up silly-renamed files. 1324 */ 1325 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode) 1326 { 1327 if (S_ISDIR(inode->i_mode)) 1328 /* drop any readdir cache as it could easily be old */ 1329 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA; 1330 1331 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { 1332 nfs_complete_unlink(dentry, inode); 1333 nfs_drop_nlink(inode); 1334 } 1335 iput(inode); 1336 } 1337 1338 static void nfs_d_release(struct dentry *dentry) 1339 { 1340 /* free cached devname value, if it survived that far */ 1341 if (unlikely(dentry->d_fsdata)) { 1342 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) 1343 WARN_ON(1); 1344 else 1345 kfree(dentry->d_fsdata); 1346 } 1347 } 1348 1349 const struct dentry_operations nfs_dentry_operations = { 1350 .d_revalidate = nfs_lookup_revalidate, 1351 .d_weak_revalidate = nfs_weak_revalidate, 1352 .d_delete = nfs_dentry_delete, 1353 .d_iput = nfs_dentry_iput, 1354 .d_automount = nfs_d_automount, 1355 .d_release = nfs_d_release, 1356 }; 1357 EXPORT_SYMBOL_GPL(nfs_dentry_operations); 1358 1359 struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags) 1360 { 1361 struct dentry *res; 1362 struct dentry *parent; 1363 struct inode *inode = NULL; 1364 struct nfs_fh *fhandle = NULL; 1365 struct nfs_fattr *fattr = NULL; 1366 struct nfs4_label *label = NULL; 1367 int error; 1368 1369 dfprintk(VFS, "NFS: lookup(%pd2)\n", dentry); 1370 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP); 1371 1372 res = ERR_PTR(-ENAMETOOLONG); 1373 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) 1374 goto out; 1375 1376 /* 1377 * If we're doing an exclusive create, optimize away the lookup 1378 * but don't hash the dentry. 1379 */ 1380 if (nfs_is_exclusive_create(dir, flags)) { 1381 d_instantiate(dentry, NULL); 1382 res = NULL; 1383 goto out; 1384 } 1385 1386 res = ERR_PTR(-ENOMEM); 1387 fhandle = nfs_alloc_fhandle(); 1388 fattr = nfs_alloc_fattr(); 1389 if (fhandle == NULL || fattr == NULL) 1390 goto out; 1391 1392 label = nfs4_label_alloc(NFS_SERVER(dir), GFP_NOWAIT); 1393 if (IS_ERR(label)) 1394 goto out; 1395 1396 parent = dentry->d_parent; 1397 /* Protect against concurrent sillydeletes */ 1398 trace_nfs_lookup_enter(dir, dentry, flags); 1399 nfs_block_sillyrename(parent); 1400 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, label); 1401 if (error == -ENOENT) 1402 goto no_entry; 1403 if (error < 0) { 1404 res = ERR_PTR(error); 1405 goto out_unblock_sillyrename; 1406 } 1407 inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label); 1408 res = ERR_CAST(inode); 1409 if (IS_ERR(res)) 1410 goto out_unblock_sillyrename; 1411 1412 /* Success: notify readdir to use READDIRPLUS */ 1413 nfs_advise_use_readdirplus(dir); 1414 1415 no_entry: 1416 res = d_splice_alias(inode, dentry); 1417 if (res != NULL) { 1418 if (IS_ERR(res)) 1419 goto out_unblock_sillyrename; 1420 dentry = res; 1421 } 1422 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1423 out_unblock_sillyrename: 1424 nfs_unblock_sillyrename(parent); 1425 trace_nfs_lookup_exit(dir, dentry, flags, error); 1426 nfs4_label_free(label); 1427 out: 1428 nfs_free_fattr(fattr); 1429 nfs_free_fhandle(fhandle); 1430 return res; 1431 } 1432 EXPORT_SYMBOL_GPL(nfs_lookup); 1433 1434 #if IS_ENABLED(CONFIG_NFS_V4) 1435 static int nfs4_lookup_revalidate(struct dentry *, unsigned int); 1436 1437 const struct dentry_operations nfs4_dentry_operations = { 1438 .d_revalidate = nfs4_lookup_revalidate, 1439 .d_delete = nfs_dentry_delete, 1440 .d_iput = nfs_dentry_iput, 1441 .d_automount = nfs_d_automount, 1442 .d_release = nfs_d_release, 1443 }; 1444 EXPORT_SYMBOL_GPL(nfs4_dentry_operations); 1445 1446 static fmode_t flags_to_mode(int flags) 1447 { 1448 fmode_t res = (__force fmode_t)flags & FMODE_EXEC; 1449 if ((flags & O_ACCMODE) != O_WRONLY) 1450 res |= FMODE_READ; 1451 if ((flags & O_ACCMODE) != O_RDONLY) 1452 res |= FMODE_WRITE; 1453 return res; 1454 } 1455 1456 static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags) 1457 { 1458 return alloc_nfs_open_context(dentry, flags_to_mode(open_flags)); 1459 } 1460 1461 static int do_open(struct inode *inode, struct file *filp) 1462 { 1463 nfs_fscache_open_file(inode, filp); 1464 return 0; 1465 } 1466 1467 static int nfs_finish_open(struct nfs_open_context *ctx, 1468 struct dentry *dentry, 1469 struct file *file, unsigned open_flags, 1470 int *opened) 1471 { 1472 int err; 1473 1474 if ((open_flags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL)) 1475 *opened |= FILE_CREATED; 1476 1477 err = finish_open(file, dentry, do_open, opened); 1478 if (err) 1479 goto out; 1480 nfs_file_set_open_context(file, ctx); 1481 1482 out: 1483 return err; 1484 } 1485 1486 int nfs_atomic_open(struct inode *dir, struct dentry *dentry, 1487 struct file *file, unsigned open_flags, 1488 umode_t mode, int *opened) 1489 { 1490 struct nfs_open_context *ctx; 1491 struct dentry *res; 1492 struct iattr attr = { .ia_valid = ATTR_OPEN }; 1493 struct inode *inode; 1494 unsigned int lookup_flags = 0; 1495 int err; 1496 1497 /* Expect a negative dentry */ 1498 BUG_ON(dentry->d_inode); 1499 1500 dfprintk(VFS, "NFS: atomic_open(%s/%lu), %pd\n", 1501 dir->i_sb->s_id, dir->i_ino, dentry); 1502 1503 err = nfs_check_flags(open_flags); 1504 if (err) 1505 return err; 1506 1507 /* NFS only supports OPEN on regular files */ 1508 if ((open_flags & O_DIRECTORY)) { 1509 if (!d_unhashed(dentry)) { 1510 /* 1511 * Hashed negative dentry with O_DIRECTORY: dentry was 1512 * revalidated and is fine, no need to perform lookup 1513 * again 1514 */ 1515 return -ENOENT; 1516 } 1517 lookup_flags = LOOKUP_OPEN|LOOKUP_DIRECTORY; 1518 goto no_open; 1519 } 1520 1521 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) 1522 return -ENAMETOOLONG; 1523 1524 if (open_flags & O_CREAT) { 1525 attr.ia_valid |= ATTR_MODE; 1526 attr.ia_mode = mode & ~current_umask(); 1527 } 1528 if (open_flags & O_TRUNC) { 1529 attr.ia_valid |= ATTR_SIZE; 1530 attr.ia_size = 0; 1531 } 1532 1533 ctx = create_nfs_open_context(dentry, open_flags); 1534 err = PTR_ERR(ctx); 1535 if (IS_ERR(ctx)) 1536 goto out; 1537 1538 trace_nfs_atomic_open_enter(dir, ctx, open_flags); 1539 nfs_block_sillyrename(dentry->d_parent); 1540 inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr, opened); 1541 nfs_unblock_sillyrename(dentry->d_parent); 1542 if (IS_ERR(inode)) { 1543 err = PTR_ERR(inode); 1544 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err); 1545 put_nfs_open_context(ctx); 1546 switch (err) { 1547 case -ENOENT: 1548 d_drop(dentry); 1549 d_add(dentry, NULL); 1550 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1551 break; 1552 case -EISDIR: 1553 case -ENOTDIR: 1554 goto no_open; 1555 case -ELOOP: 1556 if (!(open_flags & O_NOFOLLOW)) 1557 goto no_open; 1558 break; 1559 /* case -EINVAL: */ 1560 default: 1561 break; 1562 } 1563 goto out; 1564 } 1565 1566 err = nfs_finish_open(ctx, ctx->dentry, file, open_flags, opened); 1567 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err); 1568 put_nfs_open_context(ctx); 1569 out: 1570 return err; 1571 1572 no_open: 1573 res = nfs_lookup(dir, dentry, lookup_flags); 1574 err = PTR_ERR(res); 1575 if (IS_ERR(res)) 1576 goto out; 1577 1578 return finish_no_open(file, res); 1579 } 1580 EXPORT_SYMBOL_GPL(nfs_atomic_open); 1581 1582 static int nfs4_lookup_revalidate(struct dentry *dentry, unsigned int flags) 1583 { 1584 struct inode *inode; 1585 int ret = 0; 1586 1587 if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY)) 1588 goto no_open; 1589 if (d_mountpoint(dentry)) 1590 goto no_open; 1591 if (NFS_SB(dentry->d_sb)->caps & NFS_CAP_ATOMIC_OPEN_V1) 1592 goto no_open; 1593 1594 inode = dentry->d_inode; 1595 1596 /* We can't create new files in nfs_open_revalidate(), so we 1597 * optimize away revalidation of negative dentries. 1598 */ 1599 if (inode == NULL) { 1600 struct dentry *parent; 1601 struct inode *dir; 1602 1603 if (flags & LOOKUP_RCU) { 1604 parent = ACCESS_ONCE(dentry->d_parent); 1605 dir = ACCESS_ONCE(parent->d_inode); 1606 if (!dir) 1607 return -ECHILD; 1608 } else { 1609 parent = dget_parent(dentry); 1610 dir = parent->d_inode; 1611 } 1612 if (!nfs_neg_need_reval(dir, dentry, flags)) 1613 ret = 1; 1614 else if (flags & LOOKUP_RCU) 1615 ret = -ECHILD; 1616 if (!(flags & LOOKUP_RCU)) 1617 dput(parent); 1618 else if (parent != ACCESS_ONCE(dentry->d_parent)) 1619 return -ECHILD; 1620 goto out; 1621 } 1622 1623 /* NFS only supports OPEN on regular files */ 1624 if (!S_ISREG(inode->i_mode)) 1625 goto no_open; 1626 /* We cannot do exclusive creation on a positive dentry */ 1627 if (flags & LOOKUP_EXCL) 1628 goto no_open; 1629 1630 /* Let f_op->open() actually open (and revalidate) the file */ 1631 ret = 1; 1632 1633 out: 1634 return ret; 1635 1636 no_open: 1637 return nfs_lookup_revalidate(dentry, flags); 1638 } 1639 1640 #endif /* CONFIG_NFSV4 */ 1641 1642 /* 1643 * Code common to create, mkdir, and mknod. 1644 */ 1645 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle, 1646 struct nfs_fattr *fattr, 1647 struct nfs4_label *label) 1648 { 1649 struct dentry *parent = dget_parent(dentry); 1650 struct inode *dir = parent->d_inode; 1651 struct inode *inode; 1652 int error = -EACCES; 1653 1654 d_drop(dentry); 1655 1656 /* We may have been initialized further down */ 1657 if (dentry->d_inode) 1658 goto out; 1659 if (fhandle->size == 0) { 1660 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, NULL); 1661 if (error) 1662 goto out_error; 1663 } 1664 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1665 if (!(fattr->valid & NFS_ATTR_FATTR)) { 1666 struct nfs_server *server = NFS_SB(dentry->d_sb); 1667 error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr, NULL); 1668 if (error < 0) 1669 goto out_error; 1670 } 1671 inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label); 1672 error = PTR_ERR(inode); 1673 if (IS_ERR(inode)) 1674 goto out_error; 1675 d_add(dentry, inode); 1676 out: 1677 dput(parent); 1678 return 0; 1679 out_error: 1680 nfs_mark_for_revalidate(dir); 1681 dput(parent); 1682 return error; 1683 } 1684 EXPORT_SYMBOL_GPL(nfs_instantiate); 1685 1686 /* 1687 * Following a failed create operation, we drop the dentry rather 1688 * than retain a negative dentry. This avoids a problem in the event 1689 * that the operation succeeded on the server, but an error in the 1690 * reply path made it appear to have failed. 1691 */ 1692 int nfs_create(struct inode *dir, struct dentry *dentry, 1693 umode_t mode, bool excl) 1694 { 1695 struct iattr attr; 1696 int open_flags = excl ? O_CREAT | O_EXCL : O_CREAT; 1697 int error; 1698 1699 dfprintk(VFS, "NFS: create(%s/%lu), %pd\n", 1700 dir->i_sb->s_id, dir->i_ino, dentry); 1701 1702 attr.ia_mode = mode; 1703 attr.ia_valid = ATTR_MODE; 1704 1705 trace_nfs_create_enter(dir, dentry, open_flags); 1706 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags); 1707 trace_nfs_create_exit(dir, dentry, open_flags, error); 1708 if (error != 0) 1709 goto out_err; 1710 return 0; 1711 out_err: 1712 d_drop(dentry); 1713 return error; 1714 } 1715 EXPORT_SYMBOL_GPL(nfs_create); 1716 1717 /* 1718 * See comments for nfs_proc_create regarding failed operations. 1719 */ 1720 int 1721 nfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t rdev) 1722 { 1723 struct iattr attr; 1724 int status; 1725 1726 dfprintk(VFS, "NFS: mknod(%s/%lu), %pd\n", 1727 dir->i_sb->s_id, dir->i_ino, dentry); 1728 1729 if (!new_valid_dev(rdev)) 1730 return -EINVAL; 1731 1732 attr.ia_mode = mode; 1733 attr.ia_valid = ATTR_MODE; 1734 1735 trace_nfs_mknod_enter(dir, dentry); 1736 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev); 1737 trace_nfs_mknod_exit(dir, dentry, status); 1738 if (status != 0) 1739 goto out_err; 1740 return 0; 1741 out_err: 1742 d_drop(dentry); 1743 return status; 1744 } 1745 EXPORT_SYMBOL_GPL(nfs_mknod); 1746 1747 /* 1748 * See comments for nfs_proc_create regarding failed operations. 1749 */ 1750 int nfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) 1751 { 1752 struct iattr attr; 1753 int error; 1754 1755 dfprintk(VFS, "NFS: mkdir(%s/%lu), %pd\n", 1756 dir->i_sb->s_id, dir->i_ino, dentry); 1757 1758 attr.ia_valid = ATTR_MODE; 1759 attr.ia_mode = mode | S_IFDIR; 1760 1761 trace_nfs_mkdir_enter(dir, dentry); 1762 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr); 1763 trace_nfs_mkdir_exit(dir, dentry, error); 1764 if (error != 0) 1765 goto out_err; 1766 return 0; 1767 out_err: 1768 d_drop(dentry); 1769 return error; 1770 } 1771 EXPORT_SYMBOL_GPL(nfs_mkdir); 1772 1773 static void nfs_dentry_handle_enoent(struct dentry *dentry) 1774 { 1775 if (dentry->d_inode != NULL && !d_unhashed(dentry)) 1776 d_delete(dentry); 1777 } 1778 1779 int nfs_rmdir(struct inode *dir, struct dentry *dentry) 1780 { 1781 int error; 1782 1783 dfprintk(VFS, "NFS: rmdir(%s/%lu), %pd\n", 1784 dir->i_sb->s_id, dir->i_ino, dentry); 1785 1786 trace_nfs_rmdir_enter(dir, dentry); 1787 if (dentry->d_inode) { 1788 nfs_wait_on_sillyrename(dentry); 1789 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name); 1790 /* Ensure the VFS deletes this inode */ 1791 switch (error) { 1792 case 0: 1793 clear_nlink(dentry->d_inode); 1794 break; 1795 case -ENOENT: 1796 nfs_dentry_handle_enoent(dentry); 1797 } 1798 } else 1799 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name); 1800 trace_nfs_rmdir_exit(dir, dentry, error); 1801 1802 return error; 1803 } 1804 EXPORT_SYMBOL_GPL(nfs_rmdir); 1805 1806 /* 1807 * Remove a file after making sure there are no pending writes, 1808 * and after checking that the file has only one user. 1809 * 1810 * We invalidate the attribute cache and free the inode prior to the operation 1811 * to avoid possible races if the server reuses the inode. 1812 */ 1813 static int nfs_safe_remove(struct dentry *dentry) 1814 { 1815 struct inode *dir = dentry->d_parent->d_inode; 1816 struct inode *inode = dentry->d_inode; 1817 int error = -EBUSY; 1818 1819 dfprintk(VFS, "NFS: safe_remove(%pd2)\n", dentry); 1820 1821 /* If the dentry was sillyrenamed, we simply call d_delete() */ 1822 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { 1823 error = 0; 1824 goto out; 1825 } 1826 1827 trace_nfs_remove_enter(dir, dentry); 1828 if (inode != NULL) { 1829 NFS_PROTO(inode)->return_delegation(inode); 1830 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name); 1831 if (error == 0) 1832 nfs_drop_nlink(inode); 1833 } else 1834 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name); 1835 if (error == -ENOENT) 1836 nfs_dentry_handle_enoent(dentry); 1837 trace_nfs_remove_exit(dir, dentry, error); 1838 out: 1839 return error; 1840 } 1841 1842 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode 1843 * belongs to an active ".nfs..." file and we return -EBUSY. 1844 * 1845 * If sillyrename() returns 0, we do nothing, otherwise we unlink. 1846 */ 1847 int nfs_unlink(struct inode *dir, struct dentry *dentry) 1848 { 1849 int error; 1850 int need_rehash = 0; 1851 1852 dfprintk(VFS, "NFS: unlink(%s/%lu, %pd)\n", dir->i_sb->s_id, 1853 dir->i_ino, dentry); 1854 1855 trace_nfs_unlink_enter(dir, dentry); 1856 spin_lock(&dentry->d_lock); 1857 if (d_count(dentry) > 1) { 1858 spin_unlock(&dentry->d_lock); 1859 /* Start asynchronous writeout of the inode */ 1860 write_inode_now(dentry->d_inode, 0); 1861 error = nfs_sillyrename(dir, dentry); 1862 goto out; 1863 } 1864 if (!d_unhashed(dentry)) { 1865 __d_drop(dentry); 1866 need_rehash = 1; 1867 } 1868 spin_unlock(&dentry->d_lock); 1869 error = nfs_safe_remove(dentry); 1870 if (!error || error == -ENOENT) { 1871 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1872 } else if (need_rehash) 1873 d_rehash(dentry); 1874 out: 1875 trace_nfs_unlink_exit(dir, dentry, error); 1876 return error; 1877 } 1878 EXPORT_SYMBOL_GPL(nfs_unlink); 1879 1880 /* 1881 * To create a symbolic link, most file systems instantiate a new inode, 1882 * add a page to it containing the path, then write it out to the disk 1883 * using prepare_write/commit_write. 1884 * 1885 * Unfortunately the NFS client can't create the in-core inode first 1886 * because it needs a file handle to create an in-core inode (see 1887 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the 1888 * symlink request has completed on the server. 1889 * 1890 * So instead we allocate a raw page, copy the symname into it, then do 1891 * the SYMLINK request with the page as the buffer. If it succeeds, we 1892 * now have a new file handle and can instantiate an in-core NFS inode 1893 * and move the raw page into its mapping. 1894 */ 1895 int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname) 1896 { 1897 struct page *page; 1898 char *kaddr; 1899 struct iattr attr; 1900 unsigned int pathlen = strlen(symname); 1901 int error; 1902 1903 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s)\n", dir->i_sb->s_id, 1904 dir->i_ino, dentry, symname); 1905 1906 if (pathlen > PAGE_SIZE) 1907 return -ENAMETOOLONG; 1908 1909 attr.ia_mode = S_IFLNK | S_IRWXUGO; 1910 attr.ia_valid = ATTR_MODE; 1911 1912 page = alloc_page(GFP_HIGHUSER); 1913 if (!page) 1914 return -ENOMEM; 1915 1916 kaddr = kmap_atomic(page); 1917 memcpy(kaddr, symname, pathlen); 1918 if (pathlen < PAGE_SIZE) 1919 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen); 1920 kunmap_atomic(kaddr); 1921 1922 trace_nfs_symlink_enter(dir, dentry); 1923 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr); 1924 trace_nfs_symlink_exit(dir, dentry, error); 1925 if (error != 0) { 1926 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s) error %d\n", 1927 dir->i_sb->s_id, dir->i_ino, 1928 dentry, symname, error); 1929 d_drop(dentry); 1930 __free_page(page); 1931 return error; 1932 } 1933 1934 /* 1935 * No big deal if we can't add this page to the page cache here. 1936 * READLINK will get the missing page from the server if needed. 1937 */ 1938 if (!add_to_page_cache_lru(page, dentry->d_inode->i_mapping, 0, 1939 GFP_KERNEL)) { 1940 SetPageUptodate(page); 1941 unlock_page(page); 1942 /* 1943 * add_to_page_cache_lru() grabs an extra page refcount. 1944 * Drop it here to avoid leaking this page later. 1945 */ 1946 page_cache_release(page); 1947 } else 1948 __free_page(page); 1949 1950 return 0; 1951 } 1952 EXPORT_SYMBOL_GPL(nfs_symlink); 1953 1954 int 1955 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) 1956 { 1957 struct inode *inode = old_dentry->d_inode; 1958 int error; 1959 1960 dfprintk(VFS, "NFS: link(%pd2 -> %pd2)\n", 1961 old_dentry, dentry); 1962 1963 trace_nfs_link_enter(inode, dir, dentry); 1964 NFS_PROTO(inode)->return_delegation(inode); 1965 1966 d_drop(dentry); 1967 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name); 1968 if (error == 0) { 1969 ihold(inode); 1970 d_add(dentry, inode); 1971 } 1972 trace_nfs_link_exit(inode, dir, dentry, error); 1973 return error; 1974 } 1975 EXPORT_SYMBOL_GPL(nfs_link); 1976 1977 /* 1978 * RENAME 1979 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a 1980 * different file handle for the same inode after a rename (e.g. when 1981 * moving to a different directory). A fail-safe method to do so would 1982 * be to look up old_dir/old_name, create a link to new_dir/new_name and 1983 * rename the old file using the sillyrename stuff. This way, the original 1984 * file in old_dir will go away when the last process iput()s the inode. 1985 * 1986 * FIXED. 1987 * 1988 * It actually works quite well. One needs to have the possibility for 1989 * at least one ".nfs..." file in each directory the file ever gets 1990 * moved or linked to which happens automagically with the new 1991 * implementation that only depends on the dcache stuff instead of 1992 * using the inode layer 1993 * 1994 * Unfortunately, things are a little more complicated than indicated 1995 * above. For a cross-directory move, we want to make sure we can get 1996 * rid of the old inode after the operation. This means there must be 1997 * no pending writes (if it's a file), and the use count must be 1. 1998 * If these conditions are met, we can drop the dentries before doing 1999 * the rename. 2000 */ 2001 int nfs_rename(struct inode *old_dir, struct dentry *old_dentry, 2002 struct inode *new_dir, struct dentry *new_dentry) 2003 { 2004 struct inode *old_inode = old_dentry->d_inode; 2005 struct inode *new_inode = new_dentry->d_inode; 2006 struct dentry *dentry = NULL, *rehash = NULL; 2007 struct rpc_task *task; 2008 int error = -EBUSY; 2009 2010 dfprintk(VFS, "NFS: rename(%pd2 -> %pd2, ct=%d)\n", 2011 old_dentry, new_dentry, 2012 d_count(new_dentry)); 2013 2014 trace_nfs_rename_enter(old_dir, old_dentry, new_dir, new_dentry); 2015 /* 2016 * For non-directories, check whether the target is busy and if so, 2017 * make a copy of the dentry and then do a silly-rename. If the 2018 * silly-rename succeeds, the copied dentry is hashed and becomes 2019 * the new target. 2020 */ 2021 if (new_inode && !S_ISDIR(new_inode->i_mode)) { 2022 /* 2023 * To prevent any new references to the target during the 2024 * rename, we unhash the dentry in advance. 2025 */ 2026 if (!d_unhashed(new_dentry)) { 2027 d_drop(new_dentry); 2028 rehash = new_dentry; 2029 } 2030 2031 if (d_count(new_dentry) > 2) { 2032 int err; 2033 2034 /* copy the target dentry's name */ 2035 dentry = d_alloc(new_dentry->d_parent, 2036 &new_dentry->d_name); 2037 if (!dentry) 2038 goto out; 2039 2040 /* silly-rename the existing target ... */ 2041 err = nfs_sillyrename(new_dir, new_dentry); 2042 if (err) 2043 goto out; 2044 2045 new_dentry = dentry; 2046 rehash = NULL; 2047 new_inode = NULL; 2048 } 2049 } 2050 2051 NFS_PROTO(old_inode)->return_delegation(old_inode); 2052 if (new_inode != NULL) 2053 NFS_PROTO(new_inode)->return_delegation(new_inode); 2054 2055 task = nfs_async_rename(old_dir, new_dir, old_dentry, new_dentry, NULL); 2056 if (IS_ERR(task)) { 2057 error = PTR_ERR(task); 2058 goto out; 2059 } 2060 2061 error = rpc_wait_for_completion_task(task); 2062 if (error == 0) 2063 error = task->tk_status; 2064 rpc_put_task(task); 2065 nfs_mark_for_revalidate(old_inode); 2066 out: 2067 if (rehash) 2068 d_rehash(rehash); 2069 trace_nfs_rename_exit(old_dir, old_dentry, 2070 new_dir, new_dentry, error); 2071 if (!error) { 2072 if (new_inode != NULL) 2073 nfs_drop_nlink(new_inode); 2074 d_move(old_dentry, new_dentry); 2075 nfs_set_verifier(new_dentry, 2076 nfs_save_change_attribute(new_dir)); 2077 } else if (error == -ENOENT) 2078 nfs_dentry_handle_enoent(old_dentry); 2079 2080 /* new dentry created? */ 2081 if (dentry) 2082 dput(dentry); 2083 return error; 2084 } 2085 EXPORT_SYMBOL_GPL(nfs_rename); 2086 2087 static DEFINE_SPINLOCK(nfs_access_lru_lock); 2088 static LIST_HEAD(nfs_access_lru_list); 2089 static atomic_long_t nfs_access_nr_entries; 2090 2091 static unsigned long nfs_access_max_cachesize = ULONG_MAX; 2092 module_param(nfs_access_max_cachesize, ulong, 0644); 2093 MODULE_PARM_DESC(nfs_access_max_cachesize, "NFS access maximum total cache length"); 2094 2095 static void nfs_access_free_entry(struct nfs_access_entry *entry) 2096 { 2097 put_rpccred(entry->cred); 2098 kfree_rcu(entry, rcu_head); 2099 smp_mb__before_atomic(); 2100 atomic_long_dec(&nfs_access_nr_entries); 2101 smp_mb__after_atomic(); 2102 } 2103 2104 static void nfs_access_free_list(struct list_head *head) 2105 { 2106 struct nfs_access_entry *cache; 2107 2108 while (!list_empty(head)) { 2109 cache = list_entry(head->next, struct nfs_access_entry, lru); 2110 list_del(&cache->lru); 2111 nfs_access_free_entry(cache); 2112 } 2113 } 2114 2115 static unsigned long 2116 nfs_do_access_cache_scan(unsigned int nr_to_scan) 2117 { 2118 LIST_HEAD(head); 2119 struct nfs_inode *nfsi, *next; 2120 struct nfs_access_entry *cache; 2121 long freed = 0; 2122 2123 spin_lock(&nfs_access_lru_lock); 2124 list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) { 2125 struct inode *inode; 2126 2127 if (nr_to_scan-- == 0) 2128 break; 2129 inode = &nfsi->vfs_inode; 2130 spin_lock(&inode->i_lock); 2131 if (list_empty(&nfsi->access_cache_entry_lru)) 2132 goto remove_lru_entry; 2133 cache = list_entry(nfsi->access_cache_entry_lru.next, 2134 struct nfs_access_entry, lru); 2135 list_move(&cache->lru, &head); 2136 rb_erase(&cache->rb_node, &nfsi->access_cache); 2137 freed++; 2138 if (!list_empty(&nfsi->access_cache_entry_lru)) 2139 list_move_tail(&nfsi->access_cache_inode_lru, 2140 &nfs_access_lru_list); 2141 else { 2142 remove_lru_entry: 2143 list_del_init(&nfsi->access_cache_inode_lru); 2144 smp_mb__before_atomic(); 2145 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags); 2146 smp_mb__after_atomic(); 2147 } 2148 spin_unlock(&inode->i_lock); 2149 } 2150 spin_unlock(&nfs_access_lru_lock); 2151 nfs_access_free_list(&head); 2152 return freed; 2153 } 2154 2155 unsigned long 2156 nfs_access_cache_scan(struct shrinker *shrink, struct shrink_control *sc) 2157 { 2158 int nr_to_scan = sc->nr_to_scan; 2159 gfp_t gfp_mask = sc->gfp_mask; 2160 2161 if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL) 2162 return SHRINK_STOP; 2163 return nfs_do_access_cache_scan(nr_to_scan); 2164 } 2165 2166 2167 unsigned long 2168 nfs_access_cache_count(struct shrinker *shrink, struct shrink_control *sc) 2169 { 2170 return vfs_pressure_ratio(atomic_long_read(&nfs_access_nr_entries)); 2171 } 2172 2173 static void 2174 nfs_access_cache_enforce_limit(void) 2175 { 2176 long nr_entries = atomic_long_read(&nfs_access_nr_entries); 2177 unsigned long diff; 2178 unsigned int nr_to_scan; 2179 2180 if (nr_entries < 0 || nr_entries <= nfs_access_max_cachesize) 2181 return; 2182 nr_to_scan = 100; 2183 diff = nr_entries - nfs_access_max_cachesize; 2184 if (diff < nr_to_scan) 2185 nr_to_scan = diff; 2186 nfs_do_access_cache_scan(nr_to_scan); 2187 } 2188 2189 static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head) 2190 { 2191 struct rb_root *root_node = &nfsi->access_cache; 2192 struct rb_node *n; 2193 struct nfs_access_entry *entry; 2194 2195 /* Unhook entries from the cache */ 2196 while ((n = rb_first(root_node)) != NULL) { 2197 entry = rb_entry(n, struct nfs_access_entry, rb_node); 2198 rb_erase(n, root_node); 2199 list_move(&entry->lru, head); 2200 } 2201 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS; 2202 } 2203 2204 void nfs_access_zap_cache(struct inode *inode) 2205 { 2206 LIST_HEAD(head); 2207 2208 if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0) 2209 return; 2210 /* Remove from global LRU init */ 2211 spin_lock(&nfs_access_lru_lock); 2212 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) 2213 list_del_init(&NFS_I(inode)->access_cache_inode_lru); 2214 2215 spin_lock(&inode->i_lock); 2216 __nfs_access_zap_cache(NFS_I(inode), &head); 2217 spin_unlock(&inode->i_lock); 2218 spin_unlock(&nfs_access_lru_lock); 2219 nfs_access_free_list(&head); 2220 } 2221 EXPORT_SYMBOL_GPL(nfs_access_zap_cache); 2222 2223 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred) 2224 { 2225 struct rb_node *n = NFS_I(inode)->access_cache.rb_node; 2226 struct nfs_access_entry *entry; 2227 2228 while (n != NULL) { 2229 entry = rb_entry(n, struct nfs_access_entry, rb_node); 2230 2231 if (cred < entry->cred) 2232 n = n->rb_left; 2233 else if (cred > entry->cred) 2234 n = n->rb_right; 2235 else 2236 return entry; 2237 } 2238 return NULL; 2239 } 2240 2241 static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res) 2242 { 2243 struct nfs_inode *nfsi = NFS_I(inode); 2244 struct nfs_access_entry *cache; 2245 int err = -ENOENT; 2246 2247 spin_lock(&inode->i_lock); 2248 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS) 2249 goto out_zap; 2250 cache = nfs_access_search_rbtree(inode, cred); 2251 if (cache == NULL) 2252 goto out; 2253 if (!nfs_have_delegated_attributes(inode) && 2254 !time_in_range_open(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo)) 2255 goto out_stale; 2256 res->jiffies = cache->jiffies; 2257 res->cred = cache->cred; 2258 res->mask = cache->mask; 2259 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru); 2260 err = 0; 2261 out: 2262 spin_unlock(&inode->i_lock); 2263 return err; 2264 out_stale: 2265 rb_erase(&cache->rb_node, &nfsi->access_cache); 2266 list_del(&cache->lru); 2267 spin_unlock(&inode->i_lock); 2268 nfs_access_free_entry(cache); 2269 return -ENOENT; 2270 out_zap: 2271 spin_unlock(&inode->i_lock); 2272 nfs_access_zap_cache(inode); 2273 return -ENOENT; 2274 } 2275 2276 static int nfs_access_get_cached_rcu(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res) 2277 { 2278 /* Only check the most recently returned cache entry, 2279 * but do it without locking. 2280 */ 2281 struct nfs_inode *nfsi = NFS_I(inode); 2282 struct nfs_access_entry *cache; 2283 int err = -ECHILD; 2284 struct list_head *lh; 2285 2286 rcu_read_lock(); 2287 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS) 2288 goto out; 2289 lh = rcu_dereference(nfsi->access_cache_entry_lru.prev); 2290 cache = list_entry(lh, struct nfs_access_entry, lru); 2291 if (lh == &nfsi->access_cache_entry_lru || 2292 cred != cache->cred) 2293 cache = NULL; 2294 if (cache == NULL) 2295 goto out; 2296 if (!nfs_have_delegated_attributes(inode) && 2297 !time_in_range_open(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo)) 2298 goto out; 2299 res->jiffies = cache->jiffies; 2300 res->cred = cache->cred; 2301 res->mask = cache->mask; 2302 err = 0; 2303 out: 2304 rcu_read_unlock(); 2305 return err; 2306 } 2307 2308 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set) 2309 { 2310 struct nfs_inode *nfsi = NFS_I(inode); 2311 struct rb_root *root_node = &nfsi->access_cache; 2312 struct rb_node **p = &root_node->rb_node; 2313 struct rb_node *parent = NULL; 2314 struct nfs_access_entry *entry; 2315 2316 spin_lock(&inode->i_lock); 2317 while (*p != NULL) { 2318 parent = *p; 2319 entry = rb_entry(parent, struct nfs_access_entry, rb_node); 2320 2321 if (set->cred < entry->cred) 2322 p = &parent->rb_left; 2323 else if (set->cred > entry->cred) 2324 p = &parent->rb_right; 2325 else 2326 goto found; 2327 } 2328 rb_link_node(&set->rb_node, parent, p); 2329 rb_insert_color(&set->rb_node, root_node); 2330 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); 2331 spin_unlock(&inode->i_lock); 2332 return; 2333 found: 2334 rb_replace_node(parent, &set->rb_node, root_node); 2335 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); 2336 list_del(&entry->lru); 2337 spin_unlock(&inode->i_lock); 2338 nfs_access_free_entry(entry); 2339 } 2340 2341 void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set) 2342 { 2343 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL); 2344 if (cache == NULL) 2345 return; 2346 RB_CLEAR_NODE(&cache->rb_node); 2347 cache->jiffies = set->jiffies; 2348 cache->cred = get_rpccred(set->cred); 2349 cache->mask = set->mask; 2350 2351 /* The above field assignments must be visible 2352 * before this item appears on the lru. We cannot easily 2353 * use rcu_assign_pointer, so just force the memory barrier. 2354 */ 2355 smp_wmb(); 2356 nfs_access_add_rbtree(inode, cache); 2357 2358 /* Update accounting */ 2359 smp_mb__before_atomic(); 2360 atomic_long_inc(&nfs_access_nr_entries); 2361 smp_mb__after_atomic(); 2362 2363 /* Add inode to global LRU list */ 2364 if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) { 2365 spin_lock(&nfs_access_lru_lock); 2366 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) 2367 list_add_tail(&NFS_I(inode)->access_cache_inode_lru, 2368 &nfs_access_lru_list); 2369 spin_unlock(&nfs_access_lru_lock); 2370 } 2371 nfs_access_cache_enforce_limit(); 2372 } 2373 EXPORT_SYMBOL_GPL(nfs_access_add_cache); 2374 2375 void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result) 2376 { 2377 entry->mask = 0; 2378 if (access_result & NFS4_ACCESS_READ) 2379 entry->mask |= MAY_READ; 2380 if (access_result & 2381 (NFS4_ACCESS_MODIFY | NFS4_ACCESS_EXTEND | NFS4_ACCESS_DELETE)) 2382 entry->mask |= MAY_WRITE; 2383 if (access_result & (NFS4_ACCESS_LOOKUP|NFS4_ACCESS_EXECUTE)) 2384 entry->mask |= MAY_EXEC; 2385 } 2386 EXPORT_SYMBOL_GPL(nfs_access_set_mask); 2387 2388 static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask) 2389 { 2390 struct nfs_access_entry cache; 2391 int status; 2392 2393 trace_nfs_access_enter(inode); 2394 2395 status = nfs_access_get_cached_rcu(inode, cred, &cache); 2396 if (status != 0) 2397 status = nfs_access_get_cached(inode, cred, &cache); 2398 if (status == 0) 2399 goto out_cached; 2400 2401 status = -ECHILD; 2402 if (mask & MAY_NOT_BLOCK) 2403 goto out; 2404 2405 /* Be clever: ask server to check for all possible rights */ 2406 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ; 2407 cache.cred = cred; 2408 cache.jiffies = jiffies; 2409 status = NFS_PROTO(inode)->access(inode, &cache); 2410 if (status != 0) { 2411 if (status == -ESTALE) { 2412 nfs_zap_caches(inode); 2413 if (!S_ISDIR(inode->i_mode)) 2414 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags); 2415 } 2416 goto out; 2417 } 2418 nfs_access_add_cache(inode, &cache); 2419 out_cached: 2420 if ((mask & ~cache.mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) != 0) 2421 status = -EACCES; 2422 out: 2423 trace_nfs_access_exit(inode, status); 2424 return status; 2425 } 2426 2427 static int nfs_open_permission_mask(int openflags) 2428 { 2429 int mask = 0; 2430 2431 if (openflags & __FMODE_EXEC) { 2432 /* ONLY check exec rights */ 2433 mask = MAY_EXEC; 2434 } else { 2435 if ((openflags & O_ACCMODE) != O_WRONLY) 2436 mask |= MAY_READ; 2437 if ((openflags & O_ACCMODE) != O_RDONLY) 2438 mask |= MAY_WRITE; 2439 } 2440 2441 return mask; 2442 } 2443 2444 int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags) 2445 { 2446 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags)); 2447 } 2448 EXPORT_SYMBOL_GPL(nfs_may_open); 2449 2450 int nfs_permission(struct inode *inode, int mask) 2451 { 2452 struct rpc_cred *cred; 2453 int res = 0; 2454 2455 nfs_inc_stats(inode, NFSIOS_VFSACCESS); 2456 2457 if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) 2458 goto out; 2459 /* Is this sys_access() ? */ 2460 if (mask & (MAY_ACCESS | MAY_CHDIR)) 2461 goto force_lookup; 2462 2463 switch (inode->i_mode & S_IFMT) { 2464 case S_IFLNK: 2465 goto out; 2466 case S_IFREG: 2467 break; 2468 case S_IFDIR: 2469 /* 2470 * Optimize away all write operations, since the server 2471 * will check permissions when we perform the op. 2472 */ 2473 if ((mask & MAY_WRITE) && !(mask & MAY_READ)) 2474 goto out; 2475 } 2476 2477 force_lookup: 2478 if (!NFS_PROTO(inode)->access) 2479 goto out_notsup; 2480 2481 /* Always try fast lookups first */ 2482 rcu_read_lock(); 2483 cred = rpc_lookup_cred_nonblock(); 2484 if (!IS_ERR(cred)) 2485 res = nfs_do_access(inode, cred, mask|MAY_NOT_BLOCK); 2486 else 2487 res = PTR_ERR(cred); 2488 rcu_read_unlock(); 2489 if (res == -ECHILD && !(mask & MAY_NOT_BLOCK)) { 2490 /* Fast lookup failed, try the slow way */ 2491 cred = rpc_lookup_cred(); 2492 if (!IS_ERR(cred)) { 2493 res = nfs_do_access(inode, cred, mask); 2494 put_rpccred(cred); 2495 } else 2496 res = PTR_ERR(cred); 2497 } 2498 out: 2499 if (!res && (mask & MAY_EXEC) && !execute_ok(inode)) 2500 res = -EACCES; 2501 2502 dfprintk(VFS, "NFS: permission(%s/%lu), mask=0x%x, res=%d\n", 2503 inode->i_sb->s_id, inode->i_ino, mask, res); 2504 return res; 2505 out_notsup: 2506 if (mask & MAY_NOT_BLOCK) 2507 return -ECHILD; 2508 2509 res = nfs_revalidate_inode(NFS_SERVER(inode), inode); 2510 if (res == 0) 2511 res = generic_permission(inode, mask); 2512 goto out; 2513 } 2514 EXPORT_SYMBOL_GPL(nfs_permission); 2515 2516 /* 2517 * Local variables: 2518 * version-control: t 2519 * kept-new-versions: 5 2520 * End: 2521 */ 2522