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_shared = 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 atomic_t refcount; 149 int size; 150 int eof_index; 151 u64 last_cookie; 152 struct nfs_cache_array_entry array[0]; 153 }; 154 155 typedef int (*decode_dirent_t)(struct xdr_stream *, struct nfs_entry *, int); 156 typedef struct { 157 struct file *file; 158 struct page *page; 159 struct dir_context *ctx; 160 unsigned long page_index; 161 u64 *dir_cookie; 162 u64 last_cookie; 163 loff_t current_index; 164 decode_dirent_t decode; 165 166 unsigned long timestamp; 167 unsigned long gencount; 168 unsigned int cache_entry_index; 169 unsigned int plus:1; 170 unsigned int eof:1; 171 } nfs_readdir_descriptor_t; 172 173 /* 174 * The caller is responsible for calling nfs_readdir_release_array(page) 175 */ 176 static 177 struct nfs_cache_array *nfs_readdir_get_array(struct page *page) 178 { 179 void *ptr; 180 if (page == NULL) 181 return ERR_PTR(-EIO); 182 ptr = kmap(page); 183 if (ptr == NULL) 184 return ERR_PTR(-ENOMEM); 185 return ptr; 186 } 187 188 static 189 void nfs_readdir_release_array(struct page *page) 190 { 191 kunmap(page); 192 } 193 194 /* 195 * we are freeing strings created by nfs_add_to_readdir_array() 196 */ 197 static 198 void nfs_readdir_clear_array(struct page *page) 199 { 200 struct nfs_cache_array *array; 201 int i; 202 203 array = kmap_atomic(page); 204 if (atomic_dec_and_test(&array->refcount)) 205 for (i = 0; i < array->size; i++) 206 kfree(array->array[i].string.name); 207 kunmap_atomic(array); 208 } 209 210 static bool grab_page(struct page *page) 211 { 212 struct nfs_cache_array *array = kmap_atomic(page); 213 bool res = atomic_inc_not_zero(&array->refcount); 214 kunmap_atomic(array); 215 return res; 216 } 217 218 /* 219 * the caller is responsible for freeing qstr.name 220 * when called by nfs_readdir_add_to_array, the strings will be freed in 221 * nfs_clear_readdir_array() 222 */ 223 static 224 int nfs_readdir_make_qstr(struct qstr *string, const char *name, unsigned int len) 225 { 226 string->len = len; 227 string->name = kmemdup(name, len, GFP_KERNEL); 228 if (string->name == NULL) 229 return -ENOMEM; 230 /* 231 * Avoid a kmemleak false positive. The pointer to the name is stored 232 * in a page cache page which kmemleak does not scan. 233 */ 234 kmemleak_not_leak(string->name); 235 string->hash = full_name_hash(NULL, name, len); 236 return 0; 237 } 238 239 static 240 int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page) 241 { 242 struct nfs_cache_array *array = nfs_readdir_get_array(page); 243 struct nfs_cache_array_entry *cache_entry; 244 int ret; 245 246 if (IS_ERR(array)) 247 return PTR_ERR(array); 248 249 cache_entry = &array->array[array->size]; 250 251 /* Check that this entry lies within the page bounds */ 252 ret = -ENOSPC; 253 if ((char *)&cache_entry[1] - (char *)page_address(page) > PAGE_SIZE) 254 goto out; 255 256 cache_entry->cookie = entry->prev_cookie; 257 cache_entry->ino = entry->ino; 258 cache_entry->d_type = entry->d_type; 259 ret = nfs_readdir_make_qstr(&cache_entry->string, entry->name, entry->len); 260 if (ret) 261 goto out; 262 array->last_cookie = entry->cookie; 263 array->size++; 264 if (entry->eof != 0) 265 array->eof_index = array->size; 266 out: 267 nfs_readdir_release_array(page); 268 return ret; 269 } 270 271 static 272 int nfs_readdir_search_for_pos(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc) 273 { 274 loff_t diff = desc->ctx->pos - desc->current_index; 275 unsigned int index; 276 277 if (diff < 0) 278 goto out_eof; 279 if (diff >= array->size) { 280 if (array->eof_index >= 0) 281 goto out_eof; 282 return -EAGAIN; 283 } 284 285 index = (unsigned int)diff; 286 *desc->dir_cookie = array->array[index].cookie; 287 desc->cache_entry_index = index; 288 return 0; 289 out_eof: 290 desc->eof = 1; 291 return -EBADCOOKIE; 292 } 293 294 static bool 295 nfs_readdir_inode_mapping_valid(struct nfs_inode *nfsi) 296 { 297 if (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA)) 298 return false; 299 smp_rmb(); 300 return !test_bit(NFS_INO_INVALIDATING, &nfsi->flags); 301 } 302 303 static 304 int nfs_readdir_search_for_cookie(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc) 305 { 306 int i; 307 loff_t new_pos; 308 int status = -EAGAIN; 309 310 for (i = 0; i < array->size; i++) { 311 if (array->array[i].cookie == *desc->dir_cookie) { 312 struct nfs_inode *nfsi = NFS_I(file_inode(desc->file)); 313 struct nfs_open_dir_context *ctx = desc->file->private_data; 314 315 new_pos = desc->current_index + i; 316 if (ctx->attr_gencount != nfsi->attr_gencount || 317 !nfs_readdir_inode_mapping_valid(nfsi)) { 318 ctx->duped = 0; 319 ctx->attr_gencount = nfsi->attr_gencount; 320 } else if (new_pos < desc->ctx->pos) { 321 if (ctx->duped > 0 322 && ctx->dup_cookie == *desc->dir_cookie) { 323 if (printk_ratelimit()) { 324 pr_notice("NFS: directory %pD2 contains a readdir loop." 325 "Please contact your server vendor. " 326 "The file: %.*s has duplicate cookie %llu\n", 327 desc->file, array->array[i].string.len, 328 array->array[i].string.name, *desc->dir_cookie); 329 } 330 status = -ELOOP; 331 goto out; 332 } 333 ctx->dup_cookie = *desc->dir_cookie; 334 ctx->duped = -1; 335 } 336 desc->ctx->pos = new_pos; 337 desc->cache_entry_index = i; 338 return 0; 339 } 340 } 341 if (array->eof_index >= 0) { 342 status = -EBADCOOKIE; 343 if (*desc->dir_cookie == array->last_cookie) 344 desc->eof = 1; 345 } 346 out: 347 return status; 348 } 349 350 static 351 int nfs_readdir_search_array(nfs_readdir_descriptor_t *desc) 352 { 353 struct nfs_cache_array *array; 354 int status; 355 356 array = nfs_readdir_get_array(desc->page); 357 if (IS_ERR(array)) { 358 status = PTR_ERR(array); 359 goto out; 360 } 361 362 if (*desc->dir_cookie == 0) 363 status = nfs_readdir_search_for_pos(array, desc); 364 else 365 status = nfs_readdir_search_for_cookie(array, desc); 366 367 if (status == -EAGAIN) { 368 desc->last_cookie = array->last_cookie; 369 desc->current_index += array->size; 370 desc->page_index++; 371 } 372 nfs_readdir_release_array(desc->page); 373 out: 374 return status; 375 } 376 377 /* Fill a page with xdr information before transferring to the cache page */ 378 static 379 int nfs_readdir_xdr_filler(struct page **pages, nfs_readdir_descriptor_t *desc, 380 struct nfs_entry *entry, struct file *file, struct inode *inode) 381 { 382 struct nfs_open_dir_context *ctx = file->private_data; 383 struct rpc_cred *cred = ctx->cred; 384 unsigned long timestamp, gencount; 385 int error; 386 387 again: 388 timestamp = jiffies; 389 gencount = nfs_inc_attr_generation_counter(); 390 error = NFS_PROTO(inode)->readdir(file_dentry(file), cred, entry->cookie, pages, 391 NFS_SERVER(inode)->dtsize, desc->plus); 392 if (error < 0) { 393 /* We requested READDIRPLUS, but the server doesn't grok it */ 394 if (error == -ENOTSUPP && desc->plus) { 395 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS; 396 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags); 397 desc->plus = 0; 398 goto again; 399 } 400 goto error; 401 } 402 desc->timestamp = timestamp; 403 desc->gencount = gencount; 404 error: 405 return error; 406 } 407 408 static int xdr_decode(nfs_readdir_descriptor_t *desc, 409 struct nfs_entry *entry, struct xdr_stream *xdr) 410 { 411 int error; 412 413 error = desc->decode(xdr, entry, desc->plus); 414 if (error) 415 return error; 416 entry->fattr->time_start = desc->timestamp; 417 entry->fattr->gencount = desc->gencount; 418 return 0; 419 } 420 421 /* Match file and dirent using either filehandle or fileid 422 * Note: caller is responsible for checking the fsid 423 */ 424 static 425 int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry) 426 { 427 struct inode *inode; 428 struct nfs_inode *nfsi; 429 430 if (d_really_is_negative(dentry)) 431 return 0; 432 433 inode = d_inode(dentry); 434 if (is_bad_inode(inode) || NFS_STALE(inode)) 435 return 0; 436 437 nfsi = NFS_I(inode); 438 if (entry->fattr->fileid == nfsi->fileid) 439 return 1; 440 if (nfs_compare_fh(entry->fh, &nfsi->fh) == 0) 441 return 1; 442 return 0; 443 } 444 445 static 446 bool nfs_use_readdirplus(struct inode *dir, struct dir_context *ctx) 447 { 448 if (!nfs_server_capable(dir, NFS_CAP_READDIRPLUS)) 449 return false; 450 if (test_and_clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags)) 451 return true; 452 if (ctx->pos == 0) 453 return true; 454 return false; 455 } 456 457 /* 458 * This function is called by the lookup code to request the use of 459 * readdirplus to accelerate any future lookups in the same 460 * directory. 461 */ 462 static 463 void nfs_advise_use_readdirplus(struct inode *dir) 464 { 465 set_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags); 466 } 467 468 /* 469 * This function is mainly for use by nfs_getattr(). 470 * 471 * If this is an 'ls -l', we want to force use of readdirplus. 472 * Do this by checking if there is an active file descriptor 473 * and calling nfs_advise_use_readdirplus, then forcing a 474 * cache flush. 475 */ 476 void nfs_force_use_readdirplus(struct inode *dir) 477 { 478 if (!list_empty(&NFS_I(dir)->open_files)) { 479 nfs_advise_use_readdirplus(dir); 480 nfs_zap_mapping(dir, dir->i_mapping); 481 } 482 } 483 484 static 485 void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry) 486 { 487 struct qstr filename = QSTR_INIT(entry->name, entry->len); 488 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 489 struct dentry *dentry; 490 struct dentry *alias; 491 struct inode *dir = d_inode(parent); 492 struct inode *inode; 493 int status; 494 495 if (!(entry->fattr->valid & NFS_ATTR_FATTR_FILEID)) 496 return; 497 if (!(entry->fattr->valid & NFS_ATTR_FATTR_FSID)) 498 return; 499 if (filename.len == 0) 500 return; 501 /* Validate that the name doesn't contain any illegal '\0' */ 502 if (strnlen(filename.name, filename.len) != filename.len) 503 return; 504 /* ...or '/' */ 505 if (strnchr(filename.name, filename.len, '/')) 506 return; 507 if (filename.name[0] == '.') { 508 if (filename.len == 1) 509 return; 510 if (filename.len == 2 && filename.name[1] == '.') 511 return; 512 } 513 filename.hash = full_name_hash(parent, filename.name, filename.len); 514 515 dentry = d_lookup(parent, &filename); 516 again: 517 if (!dentry) { 518 dentry = d_alloc_parallel(parent, &filename, &wq); 519 if (IS_ERR(dentry)) 520 return; 521 } 522 if (!d_in_lookup(dentry)) { 523 /* Is there a mountpoint here? If so, just exit */ 524 if (!nfs_fsid_equal(&NFS_SB(dentry->d_sb)->fsid, 525 &entry->fattr->fsid)) 526 goto out; 527 if (nfs_same_file(dentry, entry)) { 528 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 529 status = nfs_refresh_inode(d_inode(dentry), entry->fattr); 530 if (!status) 531 nfs_setsecurity(d_inode(dentry), entry->fattr, entry->label); 532 goto out; 533 } else { 534 d_invalidate(dentry); 535 dput(dentry); 536 dentry = NULL; 537 goto again; 538 } 539 } 540 541 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr, entry->label); 542 alias = d_splice_alias(inode, dentry); 543 d_lookup_done(dentry); 544 if (alias) { 545 if (IS_ERR(alias)) 546 goto out; 547 dput(dentry); 548 dentry = alias; 549 } 550 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 551 out: 552 dput(dentry); 553 } 554 555 /* Perform conversion from xdr to cache array */ 556 static 557 int nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry, 558 struct page **xdr_pages, struct page *page, unsigned int buflen) 559 { 560 struct xdr_stream stream; 561 struct xdr_buf buf; 562 struct page *scratch; 563 struct nfs_cache_array *array; 564 unsigned int count = 0; 565 int status; 566 567 scratch = alloc_page(GFP_KERNEL); 568 if (scratch == NULL) 569 return -ENOMEM; 570 571 if (buflen == 0) 572 goto out_nopages; 573 574 xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen); 575 xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE); 576 577 do { 578 status = xdr_decode(desc, entry, &stream); 579 if (status != 0) { 580 if (status == -EAGAIN) 581 status = 0; 582 break; 583 } 584 585 count++; 586 587 if (desc->plus != 0) 588 nfs_prime_dcache(file_dentry(desc->file), entry); 589 590 status = nfs_readdir_add_to_array(entry, page); 591 if (status != 0) 592 break; 593 } while (!entry->eof); 594 595 out_nopages: 596 if (count == 0 || (status == -EBADCOOKIE && entry->eof != 0)) { 597 array = nfs_readdir_get_array(page); 598 if (!IS_ERR(array)) { 599 array->eof_index = array->size; 600 status = 0; 601 nfs_readdir_release_array(page); 602 } else 603 status = PTR_ERR(array); 604 } 605 606 put_page(scratch); 607 return status; 608 } 609 610 static 611 void nfs_readdir_free_pages(struct page **pages, unsigned int npages) 612 { 613 unsigned int i; 614 for (i = 0; i < npages; i++) 615 put_page(pages[i]); 616 } 617 618 /* 619 * nfs_readdir_large_page will allocate pages that must be freed with a call 620 * to nfs_readdir_free_pagearray 621 */ 622 static 623 int nfs_readdir_alloc_pages(struct page **pages, unsigned int npages) 624 { 625 unsigned int i; 626 627 for (i = 0; i < npages; i++) { 628 struct page *page = alloc_page(GFP_KERNEL); 629 if (page == NULL) 630 goto out_freepages; 631 pages[i] = page; 632 } 633 return 0; 634 635 out_freepages: 636 nfs_readdir_free_pages(pages, i); 637 return -ENOMEM; 638 } 639 640 static 641 int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t *desc, struct page *page, struct inode *inode) 642 { 643 struct page *pages[NFS_MAX_READDIR_PAGES]; 644 struct nfs_entry entry; 645 struct file *file = desc->file; 646 struct nfs_cache_array *array; 647 int status = -ENOMEM; 648 unsigned int array_size = ARRAY_SIZE(pages); 649 650 entry.prev_cookie = 0; 651 entry.cookie = desc->last_cookie; 652 entry.eof = 0; 653 entry.fh = nfs_alloc_fhandle(); 654 entry.fattr = nfs_alloc_fattr(); 655 entry.server = NFS_SERVER(inode); 656 if (entry.fh == NULL || entry.fattr == NULL) 657 goto out; 658 659 entry.label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT); 660 if (IS_ERR(entry.label)) { 661 status = PTR_ERR(entry.label); 662 goto out; 663 } 664 665 array = nfs_readdir_get_array(page); 666 if (IS_ERR(array)) { 667 status = PTR_ERR(array); 668 goto out_label_free; 669 } 670 memset(array, 0, sizeof(struct nfs_cache_array)); 671 atomic_set(&array->refcount, 1); 672 array->eof_index = -1; 673 674 status = nfs_readdir_alloc_pages(pages, array_size); 675 if (status < 0) 676 goto out_release_array; 677 do { 678 unsigned int pglen; 679 status = nfs_readdir_xdr_filler(pages, desc, &entry, file, inode); 680 681 if (status < 0) 682 break; 683 pglen = status; 684 status = nfs_readdir_page_filler(desc, &entry, pages, page, pglen); 685 if (status < 0) { 686 if (status == -ENOSPC) 687 status = 0; 688 break; 689 } 690 } while (array->eof_index < 0); 691 692 nfs_readdir_free_pages(pages, array_size); 693 out_release_array: 694 nfs_readdir_release_array(page); 695 out_label_free: 696 nfs4_label_free(entry.label); 697 out: 698 nfs_free_fattr(entry.fattr); 699 nfs_free_fhandle(entry.fh); 700 return status; 701 } 702 703 /* 704 * Now we cache directories properly, by converting xdr information 705 * to an array that can be used for lookups later. This results in 706 * fewer cache pages, since we can store more information on each page. 707 * We only need to convert from xdr once so future lookups are much simpler 708 */ 709 static 710 int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page* page) 711 { 712 struct inode *inode = file_inode(desc->file); 713 int ret; 714 715 ret = nfs_readdir_xdr_to_array(desc, page, inode); 716 if (ret < 0) 717 goto error; 718 SetPageUptodate(page); 719 720 if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) { 721 /* Should never happen */ 722 nfs_zap_mapping(inode, inode->i_mapping); 723 } 724 unlock_page(page); 725 return 0; 726 error: 727 unlock_page(page); 728 return ret; 729 } 730 731 static 732 void cache_page_release(nfs_readdir_descriptor_t *desc) 733 { 734 nfs_readdir_clear_array(desc->page); 735 put_page(desc->page); 736 desc->page = NULL; 737 } 738 739 static 740 struct page *get_cache_page(nfs_readdir_descriptor_t *desc) 741 { 742 struct page *page; 743 744 for (;;) { 745 page = read_cache_page(desc->file->f_mapping, 746 desc->page_index, (filler_t *)nfs_readdir_filler, desc); 747 if (IS_ERR(page) || grab_page(page)) 748 break; 749 put_page(page); 750 } 751 return page; 752 } 753 754 /* 755 * Returns 0 if desc->dir_cookie was found on page desc->page_index 756 */ 757 static 758 int find_cache_page(nfs_readdir_descriptor_t *desc) 759 { 760 int res; 761 762 desc->page = get_cache_page(desc); 763 if (IS_ERR(desc->page)) 764 return PTR_ERR(desc->page); 765 766 res = nfs_readdir_search_array(desc); 767 if (res != 0) 768 cache_page_release(desc); 769 return res; 770 } 771 772 /* Search for desc->dir_cookie from the beginning of the page cache */ 773 static inline 774 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc) 775 { 776 int res; 777 778 if (desc->page_index == 0) { 779 desc->current_index = 0; 780 desc->last_cookie = 0; 781 } 782 do { 783 res = find_cache_page(desc); 784 } while (res == -EAGAIN); 785 return res; 786 } 787 788 /* 789 * Once we've found the start of the dirent within a page: fill 'er up... 790 */ 791 static 792 int nfs_do_filldir(nfs_readdir_descriptor_t *desc) 793 { 794 struct file *file = desc->file; 795 int i = 0; 796 int res = 0; 797 struct nfs_cache_array *array = NULL; 798 struct nfs_open_dir_context *ctx = file->private_data; 799 800 array = nfs_readdir_get_array(desc->page); 801 if (IS_ERR(array)) { 802 res = PTR_ERR(array); 803 goto out; 804 } 805 806 for (i = desc->cache_entry_index; i < array->size; i++) { 807 struct nfs_cache_array_entry *ent; 808 809 ent = &array->array[i]; 810 if (!dir_emit(desc->ctx, ent->string.name, ent->string.len, 811 nfs_compat_user_ino64(ent->ino), ent->d_type)) { 812 desc->eof = 1; 813 break; 814 } 815 desc->ctx->pos++; 816 if (i < (array->size-1)) 817 *desc->dir_cookie = array->array[i+1].cookie; 818 else 819 *desc->dir_cookie = array->last_cookie; 820 if (ctx->duped != 0) 821 ctx->duped = 1; 822 } 823 if (array->eof_index >= 0) 824 desc->eof = 1; 825 826 nfs_readdir_release_array(desc->page); 827 out: 828 cache_page_release(desc); 829 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n", 830 (unsigned long long)*desc->dir_cookie, res); 831 return res; 832 } 833 834 /* 835 * If we cannot find a cookie in our cache, we suspect that this is 836 * because it points to a deleted file, so we ask the server to return 837 * whatever it thinks is the next entry. We then feed this to filldir. 838 * If all goes well, we should then be able to find our way round the 839 * cache on the next call to readdir_search_pagecache(); 840 * 841 * NOTE: we cannot add the anonymous page to the pagecache because 842 * the data it contains might not be page aligned. Besides, 843 * we should already have a complete representation of the 844 * directory in the page cache by the time we get here. 845 */ 846 static inline 847 int uncached_readdir(nfs_readdir_descriptor_t *desc) 848 { 849 struct page *page = NULL; 850 int status; 851 struct inode *inode = file_inode(desc->file); 852 struct nfs_open_dir_context *ctx = desc->file->private_data; 853 854 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n", 855 (unsigned long long)*desc->dir_cookie); 856 857 page = alloc_page(GFP_HIGHUSER); 858 if (!page) { 859 status = -ENOMEM; 860 goto out; 861 } 862 863 desc->page_index = 0; 864 desc->last_cookie = *desc->dir_cookie; 865 desc->page = page; 866 ctx->duped = 0; 867 868 status = nfs_readdir_xdr_to_array(desc, page, inode); 869 if (status < 0) 870 goto out_release; 871 872 status = nfs_do_filldir(desc); 873 874 out: 875 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", 876 __func__, status); 877 return status; 878 out_release: 879 cache_page_release(desc); 880 goto out; 881 } 882 883 static bool nfs_dir_mapping_need_revalidate(struct inode *dir) 884 { 885 struct nfs_inode *nfsi = NFS_I(dir); 886 887 if (nfs_attribute_cache_expired(dir)) 888 return true; 889 if (nfsi->cache_validity & NFS_INO_INVALID_DATA) 890 return true; 891 return false; 892 } 893 894 /* The file offset position represents the dirent entry number. A 895 last cookie cache takes care of the common case of reading the 896 whole directory. 897 */ 898 static int nfs_readdir(struct file *file, struct dir_context *ctx) 899 { 900 struct dentry *dentry = file_dentry(file); 901 struct inode *inode = d_inode(dentry); 902 nfs_readdir_descriptor_t my_desc, 903 *desc = &my_desc; 904 struct nfs_open_dir_context *dir_ctx = file->private_data; 905 int res = 0; 906 907 dfprintk(FILE, "NFS: readdir(%pD2) starting at cookie %llu\n", 908 file, (long long)ctx->pos); 909 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS); 910 911 /* 912 * ctx->pos points to the dirent entry number. 913 * *desc->dir_cookie has the cookie for the next entry. We have 914 * to either find the entry with the appropriate number or 915 * revalidate the cookie. 916 */ 917 memset(desc, 0, sizeof(*desc)); 918 919 desc->file = file; 920 desc->ctx = ctx; 921 desc->dir_cookie = &dir_ctx->dir_cookie; 922 desc->decode = NFS_PROTO(inode)->decode_dirent; 923 desc->plus = nfs_use_readdirplus(inode, ctx) ? 1 : 0; 924 925 if (ctx->pos == 0 || nfs_dir_mapping_need_revalidate(inode)) 926 res = nfs_revalidate_mapping(inode, file->f_mapping); 927 if (res < 0) 928 goto out; 929 930 do { 931 res = readdir_search_pagecache(desc); 932 933 if (res == -EBADCOOKIE) { 934 res = 0; 935 /* This means either end of directory */ 936 if (*desc->dir_cookie && desc->eof == 0) { 937 /* Or that the server has 'lost' a cookie */ 938 res = uncached_readdir(desc); 939 if (res == 0) 940 continue; 941 } 942 break; 943 } 944 if (res == -ETOOSMALL && desc->plus) { 945 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags); 946 nfs_zap_caches(inode); 947 desc->page_index = 0; 948 desc->plus = 0; 949 desc->eof = 0; 950 continue; 951 } 952 if (res < 0) 953 break; 954 955 res = nfs_do_filldir(desc); 956 if (res < 0) 957 break; 958 } while (!desc->eof); 959 out: 960 if (res > 0) 961 res = 0; 962 dfprintk(FILE, "NFS: readdir(%pD2) returns %d\n", file, res); 963 return res; 964 } 965 966 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int whence) 967 { 968 struct nfs_open_dir_context *dir_ctx = filp->private_data; 969 970 dfprintk(FILE, "NFS: llseek dir(%pD2, %lld, %d)\n", 971 filp, offset, whence); 972 973 switch (whence) { 974 case 1: 975 offset += filp->f_pos; 976 case 0: 977 if (offset >= 0) 978 break; 979 default: 980 return -EINVAL; 981 } 982 if (offset != filp->f_pos) { 983 filp->f_pos = offset; 984 dir_ctx->dir_cookie = 0; 985 dir_ctx->duped = 0; 986 } 987 return offset; 988 } 989 990 /* 991 * All directory operations under NFS are synchronous, so fsync() 992 * is a dummy operation. 993 */ 994 static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end, 995 int datasync) 996 { 997 struct inode *inode = file_inode(filp); 998 999 dfprintk(FILE, "NFS: fsync dir(%pD2) datasync %d\n", filp, datasync); 1000 1001 inode_lock(inode); 1002 nfs_inc_stats(inode, NFSIOS_VFSFSYNC); 1003 inode_unlock(inode); 1004 return 0; 1005 } 1006 1007 /** 1008 * nfs_force_lookup_revalidate - Mark the directory as having changed 1009 * @dir - pointer to directory inode 1010 * 1011 * This forces the revalidation code in nfs_lookup_revalidate() to do a 1012 * full lookup on all child dentries of 'dir' whenever a change occurs 1013 * on the server that might have invalidated our dcache. 1014 * 1015 * The caller should be holding dir->i_lock 1016 */ 1017 void nfs_force_lookup_revalidate(struct inode *dir) 1018 { 1019 NFS_I(dir)->cache_change_attribute++; 1020 } 1021 EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate); 1022 1023 /* 1024 * A check for whether or not the parent directory has changed. 1025 * In the case it has, we assume that the dentries are untrustworthy 1026 * and may need to be looked up again. 1027 * If rcu_walk prevents us from performing a full check, return 0. 1028 */ 1029 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry, 1030 int rcu_walk) 1031 { 1032 int ret; 1033 1034 if (IS_ROOT(dentry)) 1035 return 1; 1036 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE) 1037 return 0; 1038 if (!nfs_verify_change_attribute(dir, dentry->d_time)) 1039 return 0; 1040 /* Revalidate nfsi->cache_change_attribute before we declare a match */ 1041 if (rcu_walk) 1042 ret = nfs_revalidate_inode_rcu(NFS_SERVER(dir), dir); 1043 else 1044 ret = nfs_revalidate_inode(NFS_SERVER(dir), dir); 1045 if (ret < 0) 1046 return 0; 1047 if (!nfs_verify_change_attribute(dir, dentry->d_time)) 1048 return 0; 1049 return 1; 1050 } 1051 1052 /* 1053 * Use intent information to check whether or not we're going to do 1054 * an O_EXCL create using this path component. 1055 */ 1056 static int nfs_is_exclusive_create(struct inode *dir, unsigned int flags) 1057 { 1058 if (NFS_PROTO(dir)->version == 2) 1059 return 0; 1060 return flags & LOOKUP_EXCL; 1061 } 1062 1063 /* 1064 * Inode and filehandle revalidation for lookups. 1065 * 1066 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL, 1067 * or if the intent information indicates that we're about to open this 1068 * particular file and the "nocto" mount flag is not set. 1069 * 1070 */ 1071 static 1072 int nfs_lookup_verify_inode(struct inode *inode, unsigned int flags) 1073 { 1074 struct nfs_server *server = NFS_SERVER(inode); 1075 int ret; 1076 1077 if (IS_AUTOMOUNT(inode)) 1078 return 0; 1079 /* VFS wants an on-the-wire revalidation */ 1080 if (flags & LOOKUP_REVAL) 1081 goto out_force; 1082 /* This is an open(2) */ 1083 if ((flags & LOOKUP_OPEN) && !(server->flags & NFS_MOUNT_NOCTO) && 1084 (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode))) 1085 goto out_force; 1086 out: 1087 return (inode->i_nlink == 0) ? -ENOENT : 0; 1088 out_force: 1089 if (flags & LOOKUP_RCU) 1090 return -ECHILD; 1091 ret = __nfs_revalidate_inode(server, inode); 1092 if (ret != 0) 1093 return ret; 1094 goto out; 1095 } 1096 1097 /* 1098 * We judge how long we want to trust negative 1099 * dentries by looking at the parent inode mtime. 1100 * 1101 * If parent mtime has changed, we revalidate, else we wait for a 1102 * period corresponding to the parent's attribute cache timeout value. 1103 * 1104 * If LOOKUP_RCU prevents us from performing a full check, return 1 1105 * suggesting a reval is needed. 1106 */ 1107 static inline 1108 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry, 1109 unsigned int flags) 1110 { 1111 /* Don't revalidate a negative dentry if we're creating a new file */ 1112 if (flags & LOOKUP_CREATE) 1113 return 0; 1114 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG) 1115 return 1; 1116 return !nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU); 1117 } 1118 1119 /* 1120 * This is called every time the dcache has a lookup hit, 1121 * and we should check whether we can really trust that 1122 * lookup. 1123 * 1124 * NOTE! The hit can be a negative hit too, don't assume 1125 * we have an inode! 1126 * 1127 * If the parent directory is seen to have changed, we throw out the 1128 * cached dentry and do a new lookup. 1129 */ 1130 static int nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags) 1131 { 1132 struct inode *dir; 1133 struct inode *inode; 1134 struct dentry *parent; 1135 struct nfs_fh *fhandle = NULL; 1136 struct nfs_fattr *fattr = NULL; 1137 struct nfs4_label *label = NULL; 1138 int error; 1139 1140 if (flags & LOOKUP_RCU) { 1141 parent = ACCESS_ONCE(dentry->d_parent); 1142 dir = d_inode_rcu(parent); 1143 if (!dir) 1144 return -ECHILD; 1145 } else { 1146 parent = dget_parent(dentry); 1147 dir = d_inode(parent); 1148 } 1149 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE); 1150 inode = d_inode(dentry); 1151 1152 if (!inode) { 1153 if (nfs_neg_need_reval(dir, dentry, flags)) { 1154 if (flags & LOOKUP_RCU) 1155 return -ECHILD; 1156 goto out_bad; 1157 } 1158 goto out_valid_noent; 1159 } 1160 1161 if (is_bad_inode(inode)) { 1162 if (flags & LOOKUP_RCU) 1163 return -ECHILD; 1164 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n", 1165 __func__, dentry); 1166 goto out_bad; 1167 } 1168 1169 if (NFS_PROTO(dir)->have_delegation(inode, FMODE_READ)) 1170 goto out_set_verifier; 1171 1172 /* Force a full look up iff the parent directory has changed */ 1173 if (!nfs_is_exclusive_create(dir, flags) && 1174 nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU)) { 1175 1176 if (nfs_lookup_verify_inode(inode, flags)) { 1177 if (flags & LOOKUP_RCU) 1178 return -ECHILD; 1179 goto out_zap_parent; 1180 } 1181 goto out_valid; 1182 } 1183 1184 if (flags & LOOKUP_RCU) 1185 return -ECHILD; 1186 1187 if (NFS_STALE(inode)) 1188 goto out_bad; 1189 1190 error = -ENOMEM; 1191 fhandle = nfs_alloc_fhandle(); 1192 fattr = nfs_alloc_fattr(); 1193 if (fhandle == NULL || fattr == NULL) 1194 goto out_error; 1195 1196 label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT); 1197 if (IS_ERR(label)) 1198 goto out_error; 1199 1200 trace_nfs_lookup_revalidate_enter(dir, dentry, flags); 1201 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, label); 1202 trace_nfs_lookup_revalidate_exit(dir, dentry, flags, error); 1203 if (error) 1204 goto out_bad; 1205 if (nfs_compare_fh(NFS_FH(inode), fhandle)) 1206 goto out_bad; 1207 if ((error = nfs_refresh_inode(inode, fattr)) != 0) 1208 goto out_bad; 1209 1210 nfs_setsecurity(inode, fattr, label); 1211 1212 nfs_free_fattr(fattr); 1213 nfs_free_fhandle(fhandle); 1214 nfs4_label_free(label); 1215 1216 out_set_verifier: 1217 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1218 out_valid: 1219 /* Success: notify readdir to use READDIRPLUS */ 1220 nfs_advise_use_readdirplus(dir); 1221 out_valid_noent: 1222 if (flags & LOOKUP_RCU) { 1223 if (parent != ACCESS_ONCE(dentry->d_parent)) 1224 return -ECHILD; 1225 } else 1226 dput(parent); 1227 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is valid\n", 1228 __func__, dentry); 1229 return 1; 1230 out_zap_parent: 1231 nfs_zap_caches(dir); 1232 out_bad: 1233 WARN_ON(flags & LOOKUP_RCU); 1234 nfs_free_fattr(fattr); 1235 nfs_free_fhandle(fhandle); 1236 nfs4_label_free(label); 1237 nfs_mark_for_revalidate(dir); 1238 if (inode && S_ISDIR(inode->i_mode)) { 1239 /* Purge readdir caches. */ 1240 nfs_zap_caches(inode); 1241 /* 1242 * We can't d_drop the root of a disconnected tree: 1243 * its d_hash is on the s_anon list and d_drop() would hide 1244 * it from shrink_dcache_for_unmount(), leading to busy 1245 * inodes on unmount and further oopses. 1246 */ 1247 if (IS_ROOT(dentry)) 1248 goto out_valid; 1249 } 1250 dput(parent); 1251 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is invalid\n", 1252 __func__, dentry); 1253 return 0; 1254 out_error: 1255 WARN_ON(flags & LOOKUP_RCU); 1256 nfs_free_fattr(fattr); 1257 nfs_free_fhandle(fhandle); 1258 nfs4_label_free(label); 1259 dput(parent); 1260 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) lookup returned error %d\n", 1261 __func__, dentry, error); 1262 return error; 1263 } 1264 1265 /* 1266 * A weaker form of d_revalidate for revalidating just the d_inode(dentry) 1267 * when we don't really care about the dentry name. This is called when a 1268 * pathwalk ends on a dentry that was not found via a normal lookup in the 1269 * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals). 1270 * 1271 * In this situation, we just want to verify that the inode itself is OK 1272 * since the dentry might have changed on the server. 1273 */ 1274 static int nfs_weak_revalidate(struct dentry *dentry, unsigned int flags) 1275 { 1276 int error; 1277 struct inode *inode = d_inode(dentry); 1278 1279 /* 1280 * I believe we can only get a negative dentry here in the case of a 1281 * procfs-style symlink. Just assume it's correct for now, but we may 1282 * eventually need to do something more here. 1283 */ 1284 if (!inode) { 1285 dfprintk(LOOKUPCACHE, "%s: %pd2 has negative inode\n", 1286 __func__, dentry); 1287 return 1; 1288 } 1289 1290 if (is_bad_inode(inode)) { 1291 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n", 1292 __func__, dentry); 1293 return 0; 1294 } 1295 1296 error = nfs_revalidate_inode(NFS_SERVER(inode), inode); 1297 dfprintk(LOOKUPCACHE, "NFS: %s: inode %lu is %s\n", 1298 __func__, inode->i_ino, error ? "invalid" : "valid"); 1299 return !error; 1300 } 1301 1302 /* 1303 * This is called from dput() when d_count is going to 0. 1304 */ 1305 static int nfs_dentry_delete(const struct dentry *dentry) 1306 { 1307 dfprintk(VFS, "NFS: dentry_delete(%pd2, %x)\n", 1308 dentry, dentry->d_flags); 1309 1310 /* Unhash any dentry with a stale inode */ 1311 if (d_really_is_positive(dentry) && NFS_STALE(d_inode(dentry))) 1312 return 1; 1313 1314 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { 1315 /* Unhash it, so that ->d_iput() would be called */ 1316 return 1; 1317 } 1318 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) { 1319 /* Unhash it, so that ancestors of killed async unlink 1320 * files will be cleaned up during umount */ 1321 return 1; 1322 } 1323 return 0; 1324 1325 } 1326 1327 /* Ensure that we revalidate inode->i_nlink */ 1328 static void nfs_drop_nlink(struct inode *inode) 1329 { 1330 spin_lock(&inode->i_lock); 1331 /* drop the inode if we're reasonably sure this is the last link */ 1332 if (inode->i_nlink == 1) 1333 clear_nlink(inode); 1334 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATTR; 1335 spin_unlock(&inode->i_lock); 1336 } 1337 1338 /* 1339 * Called when the dentry loses inode. 1340 * We use it to clean up silly-renamed files. 1341 */ 1342 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode) 1343 { 1344 if (S_ISDIR(inode->i_mode)) 1345 /* drop any readdir cache as it could easily be old */ 1346 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA; 1347 1348 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { 1349 nfs_complete_unlink(dentry, inode); 1350 nfs_drop_nlink(inode); 1351 } 1352 iput(inode); 1353 } 1354 1355 static void nfs_d_release(struct dentry *dentry) 1356 { 1357 /* free cached devname value, if it survived that far */ 1358 if (unlikely(dentry->d_fsdata)) { 1359 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) 1360 WARN_ON(1); 1361 else 1362 kfree(dentry->d_fsdata); 1363 } 1364 } 1365 1366 const struct dentry_operations nfs_dentry_operations = { 1367 .d_revalidate = nfs_lookup_revalidate, 1368 .d_weak_revalidate = nfs_weak_revalidate, 1369 .d_delete = nfs_dentry_delete, 1370 .d_iput = nfs_dentry_iput, 1371 .d_automount = nfs_d_automount, 1372 .d_release = nfs_d_release, 1373 }; 1374 EXPORT_SYMBOL_GPL(nfs_dentry_operations); 1375 1376 struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags) 1377 { 1378 struct dentry *res; 1379 struct inode *inode = NULL; 1380 struct nfs_fh *fhandle = NULL; 1381 struct nfs_fattr *fattr = NULL; 1382 struct nfs4_label *label = NULL; 1383 int error; 1384 1385 dfprintk(VFS, "NFS: lookup(%pd2)\n", dentry); 1386 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP); 1387 1388 if (unlikely(dentry->d_name.len > NFS_SERVER(dir)->namelen)) 1389 return ERR_PTR(-ENAMETOOLONG); 1390 1391 /* 1392 * If we're doing an exclusive create, optimize away the lookup 1393 * but don't hash the dentry. 1394 */ 1395 if (nfs_is_exclusive_create(dir, flags)) 1396 return NULL; 1397 1398 res = ERR_PTR(-ENOMEM); 1399 fhandle = nfs_alloc_fhandle(); 1400 fattr = nfs_alloc_fattr(); 1401 if (fhandle == NULL || fattr == NULL) 1402 goto out; 1403 1404 label = nfs4_label_alloc(NFS_SERVER(dir), GFP_NOWAIT); 1405 if (IS_ERR(label)) 1406 goto out; 1407 1408 trace_nfs_lookup_enter(dir, dentry, flags); 1409 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, label); 1410 if (error == -ENOENT) 1411 goto no_entry; 1412 if (error < 0) { 1413 res = ERR_PTR(error); 1414 goto out_label; 1415 } 1416 inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label); 1417 res = ERR_CAST(inode); 1418 if (IS_ERR(res)) 1419 goto out_label; 1420 1421 /* Success: notify readdir to use READDIRPLUS */ 1422 nfs_advise_use_readdirplus(dir); 1423 1424 no_entry: 1425 res = d_splice_alias(inode, dentry); 1426 if (res != NULL) { 1427 if (IS_ERR(res)) 1428 goto out_label; 1429 dentry = res; 1430 } 1431 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1432 out_label: 1433 trace_nfs_lookup_exit(dir, dentry, flags, error); 1434 nfs4_label_free(label); 1435 out: 1436 nfs_free_fattr(fattr); 1437 nfs_free_fhandle(fhandle); 1438 return res; 1439 } 1440 EXPORT_SYMBOL_GPL(nfs_lookup); 1441 1442 #if IS_ENABLED(CONFIG_NFS_V4) 1443 static int nfs4_lookup_revalidate(struct dentry *, unsigned int); 1444 1445 const struct dentry_operations nfs4_dentry_operations = { 1446 .d_revalidate = nfs4_lookup_revalidate, 1447 .d_delete = nfs_dentry_delete, 1448 .d_iput = nfs_dentry_iput, 1449 .d_automount = nfs_d_automount, 1450 .d_release = nfs_d_release, 1451 }; 1452 EXPORT_SYMBOL_GPL(nfs4_dentry_operations); 1453 1454 static fmode_t flags_to_mode(int flags) 1455 { 1456 fmode_t res = (__force fmode_t)flags & FMODE_EXEC; 1457 if ((flags & O_ACCMODE) != O_WRONLY) 1458 res |= FMODE_READ; 1459 if ((flags & O_ACCMODE) != O_RDONLY) 1460 res |= FMODE_WRITE; 1461 return res; 1462 } 1463 1464 static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags) 1465 { 1466 return alloc_nfs_open_context(dentry, flags_to_mode(open_flags)); 1467 } 1468 1469 static int do_open(struct inode *inode, struct file *filp) 1470 { 1471 nfs_fscache_open_file(inode, filp); 1472 return 0; 1473 } 1474 1475 static int nfs_finish_open(struct nfs_open_context *ctx, 1476 struct dentry *dentry, 1477 struct file *file, unsigned open_flags, 1478 int *opened) 1479 { 1480 int err; 1481 1482 err = finish_open(file, dentry, do_open, opened); 1483 if (err) 1484 goto out; 1485 nfs_file_set_open_context(file, ctx); 1486 1487 out: 1488 return err; 1489 } 1490 1491 int nfs_atomic_open(struct inode *dir, struct dentry *dentry, 1492 struct file *file, unsigned open_flags, 1493 umode_t mode, int *opened) 1494 { 1495 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 1496 struct nfs_open_context *ctx; 1497 struct dentry *res; 1498 struct iattr attr = { .ia_valid = ATTR_OPEN }; 1499 struct inode *inode; 1500 unsigned int lookup_flags = 0; 1501 bool switched = false; 1502 int err; 1503 1504 /* Expect a negative dentry */ 1505 BUG_ON(d_inode(dentry)); 1506 1507 dfprintk(VFS, "NFS: atomic_open(%s/%lu), %pd\n", 1508 dir->i_sb->s_id, dir->i_ino, dentry); 1509 1510 err = nfs_check_flags(open_flags); 1511 if (err) 1512 return err; 1513 1514 /* NFS only supports OPEN on regular files */ 1515 if ((open_flags & O_DIRECTORY)) { 1516 if (!d_in_lookup(dentry)) { 1517 /* 1518 * Hashed negative dentry with O_DIRECTORY: dentry was 1519 * revalidated and is fine, no need to perform lookup 1520 * again 1521 */ 1522 return -ENOENT; 1523 } 1524 lookup_flags = LOOKUP_OPEN|LOOKUP_DIRECTORY; 1525 goto no_open; 1526 } 1527 1528 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) 1529 return -ENAMETOOLONG; 1530 1531 if (open_flags & O_CREAT) { 1532 attr.ia_valid |= ATTR_MODE; 1533 attr.ia_mode = mode & ~current_umask(); 1534 } 1535 if (open_flags & O_TRUNC) { 1536 attr.ia_valid |= ATTR_SIZE; 1537 attr.ia_size = 0; 1538 } 1539 1540 if (!(open_flags & O_CREAT) && !d_in_lookup(dentry)) { 1541 d_drop(dentry); 1542 switched = true; 1543 dentry = d_alloc_parallel(dentry->d_parent, 1544 &dentry->d_name, &wq); 1545 if (IS_ERR(dentry)) 1546 return PTR_ERR(dentry); 1547 if (unlikely(!d_in_lookup(dentry))) 1548 return finish_no_open(file, dentry); 1549 } 1550 1551 ctx = create_nfs_open_context(dentry, open_flags); 1552 err = PTR_ERR(ctx); 1553 if (IS_ERR(ctx)) 1554 goto out; 1555 1556 trace_nfs_atomic_open_enter(dir, ctx, open_flags); 1557 inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr, opened); 1558 if (IS_ERR(inode)) { 1559 err = PTR_ERR(inode); 1560 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err); 1561 put_nfs_open_context(ctx); 1562 d_drop(dentry); 1563 switch (err) { 1564 case -ENOENT: 1565 d_add(dentry, NULL); 1566 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1567 break; 1568 case -EISDIR: 1569 case -ENOTDIR: 1570 goto no_open; 1571 case -ELOOP: 1572 if (!(open_flags & O_NOFOLLOW)) 1573 goto no_open; 1574 break; 1575 /* case -EINVAL: */ 1576 default: 1577 break; 1578 } 1579 goto out; 1580 } 1581 1582 err = nfs_finish_open(ctx, ctx->dentry, file, open_flags, opened); 1583 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err); 1584 put_nfs_open_context(ctx); 1585 out: 1586 if (unlikely(switched)) { 1587 d_lookup_done(dentry); 1588 dput(dentry); 1589 } 1590 return err; 1591 1592 no_open: 1593 res = nfs_lookup(dir, dentry, lookup_flags); 1594 if (switched) { 1595 d_lookup_done(dentry); 1596 if (!res) 1597 res = dentry; 1598 else 1599 dput(dentry); 1600 } 1601 if (IS_ERR(res)) 1602 return PTR_ERR(res); 1603 return finish_no_open(file, res); 1604 } 1605 EXPORT_SYMBOL_GPL(nfs_atomic_open); 1606 1607 static int nfs4_lookup_revalidate(struct dentry *dentry, unsigned int flags) 1608 { 1609 struct inode *inode; 1610 int ret = 0; 1611 1612 if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY)) 1613 goto no_open; 1614 if (d_mountpoint(dentry)) 1615 goto no_open; 1616 if (NFS_SB(dentry->d_sb)->caps & NFS_CAP_ATOMIC_OPEN_V1) 1617 goto no_open; 1618 1619 inode = d_inode(dentry); 1620 1621 /* We can't create new files in nfs_open_revalidate(), so we 1622 * optimize away revalidation of negative dentries. 1623 */ 1624 if (inode == NULL) { 1625 struct dentry *parent; 1626 struct inode *dir; 1627 1628 if (flags & LOOKUP_RCU) { 1629 parent = ACCESS_ONCE(dentry->d_parent); 1630 dir = d_inode_rcu(parent); 1631 if (!dir) 1632 return -ECHILD; 1633 } else { 1634 parent = dget_parent(dentry); 1635 dir = d_inode(parent); 1636 } 1637 if (!nfs_neg_need_reval(dir, dentry, flags)) 1638 ret = 1; 1639 else if (flags & LOOKUP_RCU) 1640 ret = -ECHILD; 1641 if (!(flags & LOOKUP_RCU)) 1642 dput(parent); 1643 else if (parent != ACCESS_ONCE(dentry->d_parent)) 1644 return -ECHILD; 1645 goto out; 1646 } 1647 1648 /* NFS only supports OPEN on regular files */ 1649 if (!S_ISREG(inode->i_mode)) 1650 goto no_open; 1651 /* We cannot do exclusive creation on a positive dentry */ 1652 if (flags & LOOKUP_EXCL) 1653 goto no_open; 1654 1655 /* Let f_op->open() actually open (and revalidate) the file */ 1656 ret = 1; 1657 1658 out: 1659 return ret; 1660 1661 no_open: 1662 return nfs_lookup_revalidate(dentry, flags); 1663 } 1664 1665 #endif /* CONFIG_NFSV4 */ 1666 1667 /* 1668 * Code common to create, mkdir, and mknod. 1669 */ 1670 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle, 1671 struct nfs_fattr *fattr, 1672 struct nfs4_label *label) 1673 { 1674 struct dentry *parent = dget_parent(dentry); 1675 struct inode *dir = d_inode(parent); 1676 struct inode *inode; 1677 int error = -EACCES; 1678 1679 d_drop(dentry); 1680 1681 /* We may have been initialized further down */ 1682 if (d_really_is_positive(dentry)) 1683 goto out; 1684 if (fhandle->size == 0) { 1685 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, NULL); 1686 if (error) 1687 goto out_error; 1688 } 1689 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1690 if (!(fattr->valid & NFS_ATTR_FATTR)) { 1691 struct nfs_server *server = NFS_SB(dentry->d_sb); 1692 error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr, NULL); 1693 if (error < 0) 1694 goto out_error; 1695 } 1696 inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label); 1697 error = PTR_ERR(inode); 1698 if (IS_ERR(inode)) 1699 goto out_error; 1700 d_add(dentry, inode); 1701 out: 1702 dput(parent); 1703 return 0; 1704 out_error: 1705 nfs_mark_for_revalidate(dir); 1706 dput(parent); 1707 return error; 1708 } 1709 EXPORT_SYMBOL_GPL(nfs_instantiate); 1710 1711 /* 1712 * Following a failed create operation, we drop the dentry rather 1713 * than retain a negative dentry. This avoids a problem in the event 1714 * that the operation succeeded on the server, but an error in the 1715 * reply path made it appear to have failed. 1716 */ 1717 int nfs_create(struct inode *dir, struct dentry *dentry, 1718 umode_t mode, bool excl) 1719 { 1720 struct iattr attr; 1721 int open_flags = excl ? O_CREAT | O_EXCL : O_CREAT; 1722 int error; 1723 1724 dfprintk(VFS, "NFS: create(%s/%lu), %pd\n", 1725 dir->i_sb->s_id, dir->i_ino, dentry); 1726 1727 attr.ia_mode = mode; 1728 attr.ia_valid = ATTR_MODE; 1729 1730 trace_nfs_create_enter(dir, dentry, open_flags); 1731 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags); 1732 trace_nfs_create_exit(dir, dentry, open_flags, error); 1733 if (error != 0) 1734 goto out_err; 1735 return 0; 1736 out_err: 1737 d_drop(dentry); 1738 return error; 1739 } 1740 EXPORT_SYMBOL_GPL(nfs_create); 1741 1742 /* 1743 * See comments for nfs_proc_create regarding failed operations. 1744 */ 1745 int 1746 nfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t rdev) 1747 { 1748 struct iattr attr; 1749 int status; 1750 1751 dfprintk(VFS, "NFS: mknod(%s/%lu), %pd\n", 1752 dir->i_sb->s_id, dir->i_ino, dentry); 1753 1754 attr.ia_mode = mode; 1755 attr.ia_valid = ATTR_MODE; 1756 1757 trace_nfs_mknod_enter(dir, dentry); 1758 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev); 1759 trace_nfs_mknod_exit(dir, dentry, status); 1760 if (status != 0) 1761 goto out_err; 1762 return 0; 1763 out_err: 1764 d_drop(dentry); 1765 return status; 1766 } 1767 EXPORT_SYMBOL_GPL(nfs_mknod); 1768 1769 /* 1770 * See comments for nfs_proc_create regarding failed operations. 1771 */ 1772 int nfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) 1773 { 1774 struct iattr attr; 1775 int error; 1776 1777 dfprintk(VFS, "NFS: mkdir(%s/%lu), %pd\n", 1778 dir->i_sb->s_id, dir->i_ino, dentry); 1779 1780 attr.ia_valid = ATTR_MODE; 1781 attr.ia_mode = mode | S_IFDIR; 1782 1783 trace_nfs_mkdir_enter(dir, dentry); 1784 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr); 1785 trace_nfs_mkdir_exit(dir, dentry, error); 1786 if (error != 0) 1787 goto out_err; 1788 return 0; 1789 out_err: 1790 d_drop(dentry); 1791 return error; 1792 } 1793 EXPORT_SYMBOL_GPL(nfs_mkdir); 1794 1795 static void nfs_dentry_handle_enoent(struct dentry *dentry) 1796 { 1797 if (simple_positive(dentry)) 1798 d_delete(dentry); 1799 } 1800 1801 int nfs_rmdir(struct inode *dir, struct dentry *dentry) 1802 { 1803 int error; 1804 1805 dfprintk(VFS, "NFS: rmdir(%s/%lu), %pd\n", 1806 dir->i_sb->s_id, dir->i_ino, dentry); 1807 1808 trace_nfs_rmdir_enter(dir, dentry); 1809 if (d_really_is_positive(dentry)) { 1810 down_write(&NFS_I(d_inode(dentry))->rmdir_sem); 1811 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name); 1812 /* Ensure the VFS deletes this inode */ 1813 switch (error) { 1814 case 0: 1815 clear_nlink(d_inode(dentry)); 1816 break; 1817 case -ENOENT: 1818 nfs_dentry_handle_enoent(dentry); 1819 } 1820 up_write(&NFS_I(d_inode(dentry))->rmdir_sem); 1821 } else 1822 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name); 1823 trace_nfs_rmdir_exit(dir, dentry, error); 1824 1825 return error; 1826 } 1827 EXPORT_SYMBOL_GPL(nfs_rmdir); 1828 1829 /* 1830 * Remove a file after making sure there are no pending writes, 1831 * and after checking that the file has only one user. 1832 * 1833 * We invalidate the attribute cache and free the inode prior to the operation 1834 * to avoid possible races if the server reuses the inode. 1835 */ 1836 static int nfs_safe_remove(struct dentry *dentry) 1837 { 1838 struct inode *dir = d_inode(dentry->d_parent); 1839 struct inode *inode = d_inode(dentry); 1840 int error = -EBUSY; 1841 1842 dfprintk(VFS, "NFS: safe_remove(%pd2)\n", dentry); 1843 1844 /* If the dentry was sillyrenamed, we simply call d_delete() */ 1845 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { 1846 error = 0; 1847 goto out; 1848 } 1849 1850 trace_nfs_remove_enter(dir, dentry); 1851 if (inode != NULL) { 1852 NFS_PROTO(inode)->return_delegation(inode); 1853 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name); 1854 if (error == 0) 1855 nfs_drop_nlink(inode); 1856 } else 1857 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name); 1858 if (error == -ENOENT) 1859 nfs_dentry_handle_enoent(dentry); 1860 trace_nfs_remove_exit(dir, dentry, error); 1861 out: 1862 return error; 1863 } 1864 1865 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode 1866 * belongs to an active ".nfs..." file and we return -EBUSY. 1867 * 1868 * If sillyrename() returns 0, we do nothing, otherwise we unlink. 1869 */ 1870 int nfs_unlink(struct inode *dir, struct dentry *dentry) 1871 { 1872 int error; 1873 int need_rehash = 0; 1874 1875 dfprintk(VFS, "NFS: unlink(%s/%lu, %pd)\n", dir->i_sb->s_id, 1876 dir->i_ino, dentry); 1877 1878 trace_nfs_unlink_enter(dir, dentry); 1879 spin_lock(&dentry->d_lock); 1880 if (d_count(dentry) > 1) { 1881 spin_unlock(&dentry->d_lock); 1882 /* Start asynchronous writeout of the inode */ 1883 write_inode_now(d_inode(dentry), 0); 1884 error = nfs_sillyrename(dir, dentry); 1885 goto out; 1886 } 1887 if (!d_unhashed(dentry)) { 1888 __d_drop(dentry); 1889 need_rehash = 1; 1890 } 1891 spin_unlock(&dentry->d_lock); 1892 error = nfs_safe_remove(dentry); 1893 if (!error || error == -ENOENT) { 1894 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1895 } else if (need_rehash) 1896 d_rehash(dentry); 1897 out: 1898 trace_nfs_unlink_exit(dir, dentry, error); 1899 return error; 1900 } 1901 EXPORT_SYMBOL_GPL(nfs_unlink); 1902 1903 /* 1904 * To create a symbolic link, most file systems instantiate a new inode, 1905 * add a page to it containing the path, then write it out to the disk 1906 * using prepare_write/commit_write. 1907 * 1908 * Unfortunately the NFS client can't create the in-core inode first 1909 * because it needs a file handle to create an in-core inode (see 1910 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the 1911 * symlink request has completed on the server. 1912 * 1913 * So instead we allocate a raw page, copy the symname into it, then do 1914 * the SYMLINK request with the page as the buffer. If it succeeds, we 1915 * now have a new file handle and can instantiate an in-core NFS inode 1916 * and move the raw page into its mapping. 1917 */ 1918 int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname) 1919 { 1920 struct page *page; 1921 char *kaddr; 1922 struct iattr attr; 1923 unsigned int pathlen = strlen(symname); 1924 int error; 1925 1926 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s)\n", dir->i_sb->s_id, 1927 dir->i_ino, dentry, symname); 1928 1929 if (pathlen > PAGE_SIZE) 1930 return -ENAMETOOLONG; 1931 1932 attr.ia_mode = S_IFLNK | S_IRWXUGO; 1933 attr.ia_valid = ATTR_MODE; 1934 1935 page = alloc_page(GFP_USER); 1936 if (!page) 1937 return -ENOMEM; 1938 1939 kaddr = page_address(page); 1940 memcpy(kaddr, symname, pathlen); 1941 if (pathlen < PAGE_SIZE) 1942 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen); 1943 1944 trace_nfs_symlink_enter(dir, dentry); 1945 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr); 1946 trace_nfs_symlink_exit(dir, dentry, error); 1947 if (error != 0) { 1948 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s) error %d\n", 1949 dir->i_sb->s_id, dir->i_ino, 1950 dentry, symname, error); 1951 d_drop(dentry); 1952 __free_page(page); 1953 return error; 1954 } 1955 1956 /* 1957 * No big deal if we can't add this page to the page cache here. 1958 * READLINK will get the missing page from the server if needed. 1959 */ 1960 if (!add_to_page_cache_lru(page, d_inode(dentry)->i_mapping, 0, 1961 GFP_KERNEL)) { 1962 SetPageUptodate(page); 1963 unlock_page(page); 1964 /* 1965 * add_to_page_cache_lru() grabs an extra page refcount. 1966 * Drop it here to avoid leaking this page later. 1967 */ 1968 put_page(page); 1969 } else 1970 __free_page(page); 1971 1972 return 0; 1973 } 1974 EXPORT_SYMBOL_GPL(nfs_symlink); 1975 1976 int 1977 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) 1978 { 1979 struct inode *inode = d_inode(old_dentry); 1980 int error; 1981 1982 dfprintk(VFS, "NFS: link(%pd2 -> %pd2)\n", 1983 old_dentry, dentry); 1984 1985 trace_nfs_link_enter(inode, dir, dentry); 1986 NFS_PROTO(inode)->return_delegation(inode); 1987 1988 d_drop(dentry); 1989 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name); 1990 if (error == 0) { 1991 ihold(inode); 1992 d_add(dentry, inode); 1993 } 1994 trace_nfs_link_exit(inode, dir, dentry, error); 1995 return error; 1996 } 1997 EXPORT_SYMBOL_GPL(nfs_link); 1998 1999 /* 2000 * RENAME 2001 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a 2002 * different file handle for the same inode after a rename (e.g. when 2003 * moving to a different directory). A fail-safe method to do so would 2004 * be to look up old_dir/old_name, create a link to new_dir/new_name and 2005 * rename the old file using the sillyrename stuff. This way, the original 2006 * file in old_dir will go away when the last process iput()s the inode. 2007 * 2008 * FIXED. 2009 * 2010 * It actually works quite well. One needs to have the possibility for 2011 * at least one ".nfs..." file in each directory the file ever gets 2012 * moved or linked to which happens automagically with the new 2013 * implementation that only depends on the dcache stuff instead of 2014 * using the inode layer 2015 * 2016 * Unfortunately, things are a little more complicated than indicated 2017 * above. For a cross-directory move, we want to make sure we can get 2018 * rid of the old inode after the operation. This means there must be 2019 * no pending writes (if it's a file), and the use count must be 1. 2020 * If these conditions are met, we can drop the dentries before doing 2021 * the rename. 2022 */ 2023 int nfs_rename(struct inode *old_dir, struct dentry *old_dentry, 2024 struct inode *new_dir, struct dentry *new_dentry) 2025 { 2026 struct inode *old_inode = d_inode(old_dentry); 2027 struct inode *new_inode = d_inode(new_dentry); 2028 struct dentry *dentry = NULL, *rehash = NULL; 2029 struct rpc_task *task; 2030 int error = -EBUSY; 2031 2032 dfprintk(VFS, "NFS: rename(%pd2 -> %pd2, ct=%d)\n", 2033 old_dentry, new_dentry, 2034 d_count(new_dentry)); 2035 2036 trace_nfs_rename_enter(old_dir, old_dentry, new_dir, new_dentry); 2037 /* 2038 * For non-directories, check whether the target is busy and if so, 2039 * make a copy of the dentry and then do a silly-rename. If the 2040 * silly-rename succeeds, the copied dentry is hashed and becomes 2041 * the new target. 2042 */ 2043 if (new_inode && !S_ISDIR(new_inode->i_mode)) { 2044 /* 2045 * To prevent any new references to the target during the 2046 * rename, we unhash the dentry in advance. 2047 */ 2048 if (!d_unhashed(new_dentry)) { 2049 d_drop(new_dentry); 2050 rehash = new_dentry; 2051 } 2052 2053 if (d_count(new_dentry) > 2) { 2054 int err; 2055 2056 /* copy the target dentry's name */ 2057 dentry = d_alloc(new_dentry->d_parent, 2058 &new_dentry->d_name); 2059 if (!dentry) 2060 goto out; 2061 2062 /* silly-rename the existing target ... */ 2063 err = nfs_sillyrename(new_dir, new_dentry); 2064 if (err) 2065 goto out; 2066 2067 new_dentry = dentry; 2068 rehash = NULL; 2069 new_inode = NULL; 2070 } 2071 } 2072 2073 NFS_PROTO(old_inode)->return_delegation(old_inode); 2074 if (new_inode != NULL) 2075 NFS_PROTO(new_inode)->return_delegation(new_inode); 2076 2077 task = nfs_async_rename(old_dir, new_dir, old_dentry, new_dentry, NULL); 2078 if (IS_ERR(task)) { 2079 error = PTR_ERR(task); 2080 goto out; 2081 } 2082 2083 error = rpc_wait_for_completion_task(task); 2084 if (error == 0) 2085 error = task->tk_status; 2086 rpc_put_task(task); 2087 nfs_mark_for_revalidate(old_inode); 2088 out: 2089 if (rehash) 2090 d_rehash(rehash); 2091 trace_nfs_rename_exit(old_dir, old_dentry, 2092 new_dir, new_dentry, error); 2093 if (!error) { 2094 if (new_inode != NULL) 2095 nfs_drop_nlink(new_inode); 2096 d_move(old_dentry, new_dentry); 2097 nfs_set_verifier(new_dentry, 2098 nfs_save_change_attribute(new_dir)); 2099 } else if (error == -ENOENT) 2100 nfs_dentry_handle_enoent(old_dentry); 2101 2102 /* new dentry created? */ 2103 if (dentry) 2104 dput(dentry); 2105 return error; 2106 } 2107 EXPORT_SYMBOL_GPL(nfs_rename); 2108 2109 static DEFINE_SPINLOCK(nfs_access_lru_lock); 2110 static LIST_HEAD(nfs_access_lru_list); 2111 static atomic_long_t nfs_access_nr_entries; 2112 2113 static unsigned long nfs_access_max_cachesize = ULONG_MAX; 2114 module_param(nfs_access_max_cachesize, ulong, 0644); 2115 MODULE_PARM_DESC(nfs_access_max_cachesize, "NFS access maximum total cache length"); 2116 2117 static void nfs_access_free_entry(struct nfs_access_entry *entry) 2118 { 2119 put_rpccred(entry->cred); 2120 kfree_rcu(entry, rcu_head); 2121 smp_mb__before_atomic(); 2122 atomic_long_dec(&nfs_access_nr_entries); 2123 smp_mb__after_atomic(); 2124 } 2125 2126 static void nfs_access_free_list(struct list_head *head) 2127 { 2128 struct nfs_access_entry *cache; 2129 2130 while (!list_empty(head)) { 2131 cache = list_entry(head->next, struct nfs_access_entry, lru); 2132 list_del(&cache->lru); 2133 nfs_access_free_entry(cache); 2134 } 2135 } 2136 2137 static unsigned long 2138 nfs_do_access_cache_scan(unsigned int nr_to_scan) 2139 { 2140 LIST_HEAD(head); 2141 struct nfs_inode *nfsi, *next; 2142 struct nfs_access_entry *cache; 2143 long freed = 0; 2144 2145 spin_lock(&nfs_access_lru_lock); 2146 list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) { 2147 struct inode *inode; 2148 2149 if (nr_to_scan-- == 0) 2150 break; 2151 inode = &nfsi->vfs_inode; 2152 spin_lock(&inode->i_lock); 2153 if (list_empty(&nfsi->access_cache_entry_lru)) 2154 goto remove_lru_entry; 2155 cache = list_entry(nfsi->access_cache_entry_lru.next, 2156 struct nfs_access_entry, lru); 2157 list_move(&cache->lru, &head); 2158 rb_erase(&cache->rb_node, &nfsi->access_cache); 2159 freed++; 2160 if (!list_empty(&nfsi->access_cache_entry_lru)) 2161 list_move_tail(&nfsi->access_cache_inode_lru, 2162 &nfs_access_lru_list); 2163 else { 2164 remove_lru_entry: 2165 list_del_init(&nfsi->access_cache_inode_lru); 2166 smp_mb__before_atomic(); 2167 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags); 2168 smp_mb__after_atomic(); 2169 } 2170 spin_unlock(&inode->i_lock); 2171 } 2172 spin_unlock(&nfs_access_lru_lock); 2173 nfs_access_free_list(&head); 2174 return freed; 2175 } 2176 2177 unsigned long 2178 nfs_access_cache_scan(struct shrinker *shrink, struct shrink_control *sc) 2179 { 2180 int nr_to_scan = sc->nr_to_scan; 2181 gfp_t gfp_mask = sc->gfp_mask; 2182 2183 if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL) 2184 return SHRINK_STOP; 2185 return nfs_do_access_cache_scan(nr_to_scan); 2186 } 2187 2188 2189 unsigned long 2190 nfs_access_cache_count(struct shrinker *shrink, struct shrink_control *sc) 2191 { 2192 return vfs_pressure_ratio(atomic_long_read(&nfs_access_nr_entries)); 2193 } 2194 2195 static void 2196 nfs_access_cache_enforce_limit(void) 2197 { 2198 long nr_entries = atomic_long_read(&nfs_access_nr_entries); 2199 unsigned long diff; 2200 unsigned int nr_to_scan; 2201 2202 if (nr_entries < 0 || nr_entries <= nfs_access_max_cachesize) 2203 return; 2204 nr_to_scan = 100; 2205 diff = nr_entries - nfs_access_max_cachesize; 2206 if (diff < nr_to_scan) 2207 nr_to_scan = diff; 2208 nfs_do_access_cache_scan(nr_to_scan); 2209 } 2210 2211 static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head) 2212 { 2213 struct rb_root *root_node = &nfsi->access_cache; 2214 struct rb_node *n; 2215 struct nfs_access_entry *entry; 2216 2217 /* Unhook entries from the cache */ 2218 while ((n = rb_first(root_node)) != NULL) { 2219 entry = rb_entry(n, struct nfs_access_entry, rb_node); 2220 rb_erase(n, root_node); 2221 list_move(&entry->lru, head); 2222 } 2223 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS; 2224 } 2225 2226 void nfs_access_zap_cache(struct inode *inode) 2227 { 2228 LIST_HEAD(head); 2229 2230 if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0) 2231 return; 2232 /* Remove from global LRU init */ 2233 spin_lock(&nfs_access_lru_lock); 2234 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) 2235 list_del_init(&NFS_I(inode)->access_cache_inode_lru); 2236 2237 spin_lock(&inode->i_lock); 2238 __nfs_access_zap_cache(NFS_I(inode), &head); 2239 spin_unlock(&inode->i_lock); 2240 spin_unlock(&nfs_access_lru_lock); 2241 nfs_access_free_list(&head); 2242 } 2243 EXPORT_SYMBOL_GPL(nfs_access_zap_cache); 2244 2245 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred) 2246 { 2247 struct rb_node *n = NFS_I(inode)->access_cache.rb_node; 2248 struct nfs_access_entry *entry; 2249 2250 while (n != NULL) { 2251 entry = rb_entry(n, struct nfs_access_entry, rb_node); 2252 2253 if (cred < entry->cred) 2254 n = n->rb_left; 2255 else if (cred > entry->cred) 2256 n = n->rb_right; 2257 else 2258 return entry; 2259 } 2260 return NULL; 2261 } 2262 2263 static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res, bool may_block) 2264 { 2265 struct nfs_inode *nfsi = NFS_I(inode); 2266 struct nfs_access_entry *cache; 2267 bool retry = true; 2268 int err; 2269 2270 spin_lock(&inode->i_lock); 2271 for(;;) { 2272 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS) 2273 goto out_zap; 2274 cache = nfs_access_search_rbtree(inode, cred); 2275 err = -ENOENT; 2276 if (cache == NULL) 2277 goto out; 2278 /* Found an entry, is our attribute cache valid? */ 2279 if (!nfs_attribute_cache_expired(inode) && 2280 !(nfsi->cache_validity & NFS_INO_INVALID_ATTR)) 2281 break; 2282 err = -ECHILD; 2283 if (!may_block) 2284 goto out; 2285 if (!retry) 2286 goto out_zap; 2287 spin_unlock(&inode->i_lock); 2288 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode); 2289 if (err) 2290 return err; 2291 spin_lock(&inode->i_lock); 2292 retry = false; 2293 } 2294 res->jiffies = cache->jiffies; 2295 res->cred = cache->cred; 2296 res->mask = cache->mask; 2297 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru); 2298 err = 0; 2299 out: 2300 spin_unlock(&inode->i_lock); 2301 return err; 2302 out_zap: 2303 spin_unlock(&inode->i_lock); 2304 nfs_access_zap_cache(inode); 2305 return -ENOENT; 2306 } 2307 2308 static int nfs_access_get_cached_rcu(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res) 2309 { 2310 /* Only check the most recently returned cache entry, 2311 * but do it without locking. 2312 */ 2313 struct nfs_inode *nfsi = NFS_I(inode); 2314 struct nfs_access_entry *cache; 2315 int err = -ECHILD; 2316 struct list_head *lh; 2317 2318 rcu_read_lock(); 2319 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS) 2320 goto out; 2321 lh = rcu_dereference(nfsi->access_cache_entry_lru.prev); 2322 cache = list_entry(lh, struct nfs_access_entry, lru); 2323 if (lh == &nfsi->access_cache_entry_lru || 2324 cred != cache->cred) 2325 cache = NULL; 2326 if (cache == NULL) 2327 goto out; 2328 err = nfs_revalidate_inode_rcu(NFS_SERVER(inode), inode); 2329 if (err) 2330 goto out; 2331 res->jiffies = cache->jiffies; 2332 res->cred = cache->cred; 2333 res->mask = cache->mask; 2334 out: 2335 rcu_read_unlock(); 2336 return err; 2337 } 2338 2339 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set) 2340 { 2341 struct nfs_inode *nfsi = NFS_I(inode); 2342 struct rb_root *root_node = &nfsi->access_cache; 2343 struct rb_node **p = &root_node->rb_node; 2344 struct rb_node *parent = NULL; 2345 struct nfs_access_entry *entry; 2346 2347 spin_lock(&inode->i_lock); 2348 while (*p != NULL) { 2349 parent = *p; 2350 entry = rb_entry(parent, struct nfs_access_entry, rb_node); 2351 2352 if (set->cred < entry->cred) 2353 p = &parent->rb_left; 2354 else if (set->cred > entry->cred) 2355 p = &parent->rb_right; 2356 else 2357 goto found; 2358 } 2359 rb_link_node(&set->rb_node, parent, p); 2360 rb_insert_color(&set->rb_node, root_node); 2361 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); 2362 spin_unlock(&inode->i_lock); 2363 return; 2364 found: 2365 rb_replace_node(parent, &set->rb_node, root_node); 2366 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); 2367 list_del(&entry->lru); 2368 spin_unlock(&inode->i_lock); 2369 nfs_access_free_entry(entry); 2370 } 2371 2372 void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set) 2373 { 2374 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL); 2375 if (cache == NULL) 2376 return; 2377 RB_CLEAR_NODE(&cache->rb_node); 2378 cache->jiffies = set->jiffies; 2379 cache->cred = get_rpccred(set->cred); 2380 cache->mask = set->mask; 2381 2382 /* The above field assignments must be visible 2383 * before this item appears on the lru. We cannot easily 2384 * use rcu_assign_pointer, so just force the memory barrier. 2385 */ 2386 smp_wmb(); 2387 nfs_access_add_rbtree(inode, cache); 2388 2389 /* Update accounting */ 2390 smp_mb__before_atomic(); 2391 atomic_long_inc(&nfs_access_nr_entries); 2392 smp_mb__after_atomic(); 2393 2394 /* Add inode to global LRU list */ 2395 if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) { 2396 spin_lock(&nfs_access_lru_lock); 2397 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) 2398 list_add_tail(&NFS_I(inode)->access_cache_inode_lru, 2399 &nfs_access_lru_list); 2400 spin_unlock(&nfs_access_lru_lock); 2401 } 2402 nfs_access_cache_enforce_limit(); 2403 } 2404 EXPORT_SYMBOL_GPL(nfs_access_add_cache); 2405 2406 void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result) 2407 { 2408 entry->mask = 0; 2409 if (access_result & NFS4_ACCESS_READ) 2410 entry->mask |= MAY_READ; 2411 if (access_result & 2412 (NFS4_ACCESS_MODIFY | NFS4_ACCESS_EXTEND | NFS4_ACCESS_DELETE)) 2413 entry->mask |= MAY_WRITE; 2414 if (access_result & (NFS4_ACCESS_LOOKUP|NFS4_ACCESS_EXECUTE)) 2415 entry->mask |= MAY_EXEC; 2416 } 2417 EXPORT_SYMBOL_GPL(nfs_access_set_mask); 2418 2419 static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask) 2420 { 2421 struct nfs_access_entry cache; 2422 bool may_block = (mask & MAY_NOT_BLOCK) == 0; 2423 int status; 2424 2425 trace_nfs_access_enter(inode); 2426 2427 status = nfs_access_get_cached_rcu(inode, cred, &cache); 2428 if (status != 0) 2429 status = nfs_access_get_cached(inode, cred, &cache, may_block); 2430 if (status == 0) 2431 goto out_cached; 2432 2433 status = -ECHILD; 2434 if (!may_block) 2435 goto out; 2436 2437 /* Be clever: ask server to check for all possible rights */ 2438 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ; 2439 cache.cred = cred; 2440 cache.jiffies = jiffies; 2441 status = NFS_PROTO(inode)->access(inode, &cache); 2442 if (status != 0) { 2443 if (status == -ESTALE) { 2444 nfs_zap_caches(inode); 2445 if (!S_ISDIR(inode->i_mode)) 2446 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags); 2447 } 2448 goto out; 2449 } 2450 nfs_access_add_cache(inode, &cache); 2451 out_cached: 2452 if ((mask & ~cache.mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) != 0) 2453 status = -EACCES; 2454 out: 2455 trace_nfs_access_exit(inode, status); 2456 return status; 2457 } 2458 2459 static int nfs_open_permission_mask(int openflags) 2460 { 2461 int mask = 0; 2462 2463 if (openflags & __FMODE_EXEC) { 2464 /* ONLY check exec rights */ 2465 mask = MAY_EXEC; 2466 } else { 2467 if ((openflags & O_ACCMODE) != O_WRONLY) 2468 mask |= MAY_READ; 2469 if ((openflags & O_ACCMODE) != O_RDONLY) 2470 mask |= MAY_WRITE; 2471 } 2472 2473 return mask; 2474 } 2475 2476 int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags) 2477 { 2478 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags)); 2479 } 2480 EXPORT_SYMBOL_GPL(nfs_may_open); 2481 2482 static int nfs_execute_ok(struct inode *inode, int mask) 2483 { 2484 struct nfs_server *server = NFS_SERVER(inode); 2485 int ret; 2486 2487 if (mask & MAY_NOT_BLOCK) 2488 ret = nfs_revalidate_inode_rcu(server, inode); 2489 else 2490 ret = nfs_revalidate_inode(server, inode); 2491 if (ret == 0 && !execute_ok(inode)) 2492 ret = -EACCES; 2493 return ret; 2494 } 2495 2496 int nfs_permission(struct inode *inode, int mask) 2497 { 2498 struct rpc_cred *cred; 2499 int res = 0; 2500 2501 nfs_inc_stats(inode, NFSIOS_VFSACCESS); 2502 2503 if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) 2504 goto out; 2505 /* Is this sys_access() ? */ 2506 if (mask & (MAY_ACCESS | MAY_CHDIR)) 2507 goto force_lookup; 2508 2509 switch (inode->i_mode & S_IFMT) { 2510 case S_IFLNK: 2511 goto out; 2512 case S_IFREG: 2513 if ((mask & MAY_OPEN) && 2514 nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN)) 2515 return 0; 2516 break; 2517 case S_IFDIR: 2518 /* 2519 * Optimize away all write operations, since the server 2520 * will check permissions when we perform the op. 2521 */ 2522 if ((mask & MAY_WRITE) && !(mask & MAY_READ)) 2523 goto out; 2524 } 2525 2526 force_lookup: 2527 if (!NFS_PROTO(inode)->access) 2528 goto out_notsup; 2529 2530 /* Always try fast lookups first */ 2531 rcu_read_lock(); 2532 cred = rpc_lookup_cred_nonblock(); 2533 if (!IS_ERR(cred)) 2534 res = nfs_do_access(inode, cred, mask|MAY_NOT_BLOCK); 2535 else 2536 res = PTR_ERR(cred); 2537 rcu_read_unlock(); 2538 if (res == -ECHILD && !(mask & MAY_NOT_BLOCK)) { 2539 /* Fast lookup failed, try the slow way */ 2540 cred = rpc_lookup_cred(); 2541 if (!IS_ERR(cred)) { 2542 res = nfs_do_access(inode, cred, mask); 2543 put_rpccred(cred); 2544 } else 2545 res = PTR_ERR(cred); 2546 } 2547 out: 2548 if (!res && (mask & MAY_EXEC)) 2549 res = nfs_execute_ok(inode, mask); 2550 2551 dfprintk(VFS, "NFS: permission(%s/%lu), mask=0x%x, res=%d\n", 2552 inode->i_sb->s_id, inode->i_ino, mask, res); 2553 return res; 2554 out_notsup: 2555 if (mask & MAY_NOT_BLOCK) 2556 return -ECHILD; 2557 2558 res = nfs_revalidate_inode(NFS_SERVER(inode), inode); 2559 if (res == 0) 2560 res = generic_permission(inode, mask); 2561 goto out; 2562 } 2563 EXPORT_SYMBOL_GPL(nfs_permission); 2564 2565 /* 2566 * Local variables: 2567 * version-control: t 2568 * kept-new-versions: 5 2569 * End: 2570 */ 2571