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 0; 440 if (entry->fh->size && nfs_compare_fh(entry->fh, &nfsi->fh) != 0) 441 return 0; 442 return 1; 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 and getattr code to request the 459 * use of 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 struct nfs_inode *nfsi = NFS_I(dir); 466 467 if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) && 468 !list_empty(&nfsi->open_files)) 469 set_bit(NFS_INO_ADVISE_RDPLUS, &nfsi->flags); 470 } 471 472 /* 473 * This function is mainly for use by nfs_getattr(). 474 * 475 * If this is an 'ls -l', we want to force use of readdirplus. 476 * Do this by checking if there is an active file descriptor 477 * and calling nfs_advise_use_readdirplus, then forcing a 478 * cache flush. 479 */ 480 void nfs_force_use_readdirplus(struct inode *dir) 481 { 482 struct nfs_inode *nfsi = NFS_I(dir); 483 484 if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) && 485 !list_empty(&nfsi->open_files)) { 486 set_bit(NFS_INO_ADVISE_RDPLUS, &nfsi->flags); 487 invalidate_mapping_pages(dir->i_mapping, 0, -1); 488 } 489 } 490 491 static 492 void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry) 493 { 494 struct qstr filename = QSTR_INIT(entry->name, entry->len); 495 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 496 struct dentry *dentry; 497 struct dentry *alias; 498 struct inode *dir = d_inode(parent); 499 struct inode *inode; 500 int status; 501 502 if (!(entry->fattr->valid & NFS_ATTR_FATTR_FILEID)) 503 return; 504 if (!(entry->fattr->valid & NFS_ATTR_FATTR_FSID)) 505 return; 506 if (filename.len == 0) 507 return; 508 /* Validate that the name doesn't contain any illegal '\0' */ 509 if (strnlen(filename.name, filename.len) != filename.len) 510 return; 511 /* ...or '/' */ 512 if (strnchr(filename.name, filename.len, '/')) 513 return; 514 if (filename.name[0] == '.') { 515 if (filename.len == 1) 516 return; 517 if (filename.len == 2 && filename.name[1] == '.') 518 return; 519 } 520 filename.hash = full_name_hash(parent, filename.name, filename.len); 521 522 dentry = d_lookup(parent, &filename); 523 again: 524 if (!dentry) { 525 dentry = d_alloc_parallel(parent, &filename, &wq); 526 if (IS_ERR(dentry)) 527 return; 528 } 529 if (!d_in_lookup(dentry)) { 530 /* Is there a mountpoint here? If so, just exit */ 531 if (!nfs_fsid_equal(&NFS_SB(dentry->d_sb)->fsid, 532 &entry->fattr->fsid)) 533 goto out; 534 if (nfs_same_file(dentry, entry)) { 535 if (!entry->fh->size) 536 goto out; 537 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 538 status = nfs_refresh_inode(d_inode(dentry), entry->fattr); 539 if (!status) 540 nfs_setsecurity(d_inode(dentry), entry->fattr, entry->label); 541 goto out; 542 } else { 543 d_invalidate(dentry); 544 dput(dentry); 545 dentry = NULL; 546 goto again; 547 } 548 } 549 if (!entry->fh->size) { 550 d_lookup_done(dentry); 551 goto out; 552 } 553 554 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr, entry->label); 555 alias = d_splice_alias(inode, dentry); 556 d_lookup_done(dentry); 557 if (alias) { 558 if (IS_ERR(alias)) 559 goto out; 560 dput(dentry); 561 dentry = alias; 562 } 563 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 564 out: 565 dput(dentry); 566 } 567 568 /* Perform conversion from xdr to cache array */ 569 static 570 int nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry, 571 struct page **xdr_pages, struct page *page, unsigned int buflen) 572 { 573 struct xdr_stream stream; 574 struct xdr_buf buf; 575 struct page *scratch; 576 struct nfs_cache_array *array; 577 unsigned int count = 0; 578 int status; 579 580 scratch = alloc_page(GFP_KERNEL); 581 if (scratch == NULL) 582 return -ENOMEM; 583 584 if (buflen == 0) 585 goto out_nopages; 586 587 xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen); 588 xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE); 589 590 do { 591 status = xdr_decode(desc, entry, &stream); 592 if (status != 0) { 593 if (status == -EAGAIN) 594 status = 0; 595 break; 596 } 597 598 count++; 599 600 if (desc->plus != 0) 601 nfs_prime_dcache(file_dentry(desc->file), entry); 602 603 status = nfs_readdir_add_to_array(entry, page); 604 if (status != 0) 605 break; 606 } while (!entry->eof); 607 608 out_nopages: 609 if (count == 0 || (status == -EBADCOOKIE && entry->eof != 0)) { 610 array = nfs_readdir_get_array(page); 611 if (!IS_ERR(array)) { 612 array->eof_index = array->size; 613 status = 0; 614 nfs_readdir_release_array(page); 615 } else 616 status = PTR_ERR(array); 617 } 618 619 put_page(scratch); 620 return status; 621 } 622 623 static 624 void nfs_readdir_free_pages(struct page **pages, unsigned int npages) 625 { 626 unsigned int i; 627 for (i = 0; i < npages; i++) 628 put_page(pages[i]); 629 } 630 631 /* 632 * nfs_readdir_large_page will allocate pages that must be freed with a call 633 * to nfs_readdir_free_pagearray 634 */ 635 static 636 int nfs_readdir_alloc_pages(struct page **pages, unsigned int npages) 637 { 638 unsigned int i; 639 640 for (i = 0; i < npages; i++) { 641 struct page *page = alloc_page(GFP_KERNEL); 642 if (page == NULL) 643 goto out_freepages; 644 pages[i] = page; 645 } 646 return 0; 647 648 out_freepages: 649 nfs_readdir_free_pages(pages, i); 650 return -ENOMEM; 651 } 652 653 static 654 int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t *desc, struct page *page, struct inode *inode) 655 { 656 struct page *pages[NFS_MAX_READDIR_PAGES]; 657 struct nfs_entry entry; 658 struct file *file = desc->file; 659 struct nfs_cache_array *array; 660 int status = -ENOMEM; 661 unsigned int array_size = ARRAY_SIZE(pages); 662 663 entry.prev_cookie = 0; 664 entry.cookie = desc->last_cookie; 665 entry.eof = 0; 666 entry.fh = nfs_alloc_fhandle(); 667 entry.fattr = nfs_alloc_fattr(); 668 entry.server = NFS_SERVER(inode); 669 if (entry.fh == NULL || entry.fattr == NULL) 670 goto out; 671 672 entry.label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT); 673 if (IS_ERR(entry.label)) { 674 status = PTR_ERR(entry.label); 675 goto out; 676 } 677 678 array = nfs_readdir_get_array(page); 679 if (IS_ERR(array)) { 680 status = PTR_ERR(array); 681 goto out_label_free; 682 } 683 memset(array, 0, sizeof(struct nfs_cache_array)); 684 atomic_set(&array->refcount, 1); 685 array->eof_index = -1; 686 687 status = nfs_readdir_alloc_pages(pages, array_size); 688 if (status < 0) 689 goto out_release_array; 690 do { 691 unsigned int pglen; 692 status = nfs_readdir_xdr_filler(pages, desc, &entry, file, inode); 693 694 if (status < 0) 695 break; 696 pglen = status; 697 status = nfs_readdir_page_filler(desc, &entry, pages, page, pglen); 698 if (status < 0) { 699 if (status == -ENOSPC) 700 status = 0; 701 break; 702 } 703 } while (array->eof_index < 0); 704 705 nfs_readdir_free_pages(pages, array_size); 706 out_release_array: 707 nfs_readdir_release_array(page); 708 out_label_free: 709 nfs4_label_free(entry.label); 710 out: 711 nfs_free_fattr(entry.fattr); 712 nfs_free_fhandle(entry.fh); 713 return status; 714 } 715 716 /* 717 * Now we cache directories properly, by converting xdr information 718 * to an array that can be used for lookups later. This results in 719 * fewer cache pages, since we can store more information on each page. 720 * We only need to convert from xdr once so future lookups are much simpler 721 */ 722 static 723 int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page* page) 724 { 725 struct inode *inode = file_inode(desc->file); 726 int ret; 727 728 ret = nfs_readdir_xdr_to_array(desc, page, inode); 729 if (ret < 0) 730 goto error; 731 SetPageUptodate(page); 732 733 if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) { 734 /* Should never happen */ 735 nfs_zap_mapping(inode, inode->i_mapping); 736 } 737 unlock_page(page); 738 return 0; 739 error: 740 unlock_page(page); 741 return ret; 742 } 743 744 static 745 void cache_page_release(nfs_readdir_descriptor_t *desc) 746 { 747 nfs_readdir_clear_array(desc->page); 748 put_page(desc->page); 749 desc->page = NULL; 750 } 751 752 static 753 struct page *get_cache_page(nfs_readdir_descriptor_t *desc) 754 { 755 struct page *page; 756 757 for (;;) { 758 page = read_cache_page(desc->file->f_mapping, 759 desc->page_index, (filler_t *)nfs_readdir_filler, desc); 760 if (IS_ERR(page) || grab_page(page)) 761 break; 762 put_page(page); 763 } 764 return page; 765 } 766 767 /* 768 * Returns 0 if desc->dir_cookie was found on page desc->page_index 769 */ 770 static 771 int find_cache_page(nfs_readdir_descriptor_t *desc) 772 { 773 int res; 774 775 desc->page = get_cache_page(desc); 776 if (IS_ERR(desc->page)) 777 return PTR_ERR(desc->page); 778 779 res = nfs_readdir_search_array(desc); 780 if (res != 0) 781 cache_page_release(desc); 782 return res; 783 } 784 785 /* Search for desc->dir_cookie from the beginning of the page cache */ 786 static inline 787 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc) 788 { 789 int res; 790 791 if (desc->page_index == 0) { 792 desc->current_index = 0; 793 desc->last_cookie = 0; 794 } 795 do { 796 res = find_cache_page(desc); 797 } while (res == -EAGAIN); 798 return res; 799 } 800 801 /* 802 * Once we've found the start of the dirent within a page: fill 'er up... 803 */ 804 static 805 int nfs_do_filldir(nfs_readdir_descriptor_t *desc) 806 { 807 struct file *file = desc->file; 808 int i = 0; 809 int res = 0; 810 struct nfs_cache_array *array = NULL; 811 struct nfs_open_dir_context *ctx = file->private_data; 812 813 array = nfs_readdir_get_array(desc->page); 814 if (IS_ERR(array)) { 815 res = PTR_ERR(array); 816 goto out; 817 } 818 819 for (i = desc->cache_entry_index; i < array->size; i++) { 820 struct nfs_cache_array_entry *ent; 821 822 ent = &array->array[i]; 823 if (!dir_emit(desc->ctx, ent->string.name, ent->string.len, 824 nfs_compat_user_ino64(ent->ino), ent->d_type)) { 825 desc->eof = 1; 826 break; 827 } 828 desc->ctx->pos++; 829 if (i < (array->size-1)) 830 *desc->dir_cookie = array->array[i+1].cookie; 831 else 832 *desc->dir_cookie = array->last_cookie; 833 if (ctx->duped != 0) 834 ctx->duped = 1; 835 } 836 if (array->eof_index >= 0) 837 desc->eof = 1; 838 839 nfs_readdir_release_array(desc->page); 840 out: 841 cache_page_release(desc); 842 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n", 843 (unsigned long long)*desc->dir_cookie, res); 844 return res; 845 } 846 847 /* 848 * If we cannot find a cookie in our cache, we suspect that this is 849 * because it points to a deleted file, so we ask the server to return 850 * whatever it thinks is the next entry. We then feed this to filldir. 851 * If all goes well, we should then be able to find our way round the 852 * cache on the next call to readdir_search_pagecache(); 853 * 854 * NOTE: we cannot add the anonymous page to the pagecache because 855 * the data it contains might not be page aligned. Besides, 856 * we should already have a complete representation of the 857 * directory in the page cache by the time we get here. 858 */ 859 static inline 860 int uncached_readdir(nfs_readdir_descriptor_t *desc) 861 { 862 struct page *page = NULL; 863 int status; 864 struct inode *inode = file_inode(desc->file); 865 struct nfs_open_dir_context *ctx = desc->file->private_data; 866 867 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n", 868 (unsigned long long)*desc->dir_cookie); 869 870 page = alloc_page(GFP_HIGHUSER); 871 if (!page) { 872 status = -ENOMEM; 873 goto out; 874 } 875 876 desc->page_index = 0; 877 desc->last_cookie = *desc->dir_cookie; 878 desc->page = page; 879 ctx->duped = 0; 880 881 status = nfs_readdir_xdr_to_array(desc, page, inode); 882 if (status < 0) 883 goto out_release; 884 885 status = nfs_do_filldir(desc); 886 887 out: 888 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", 889 __func__, status); 890 return status; 891 out_release: 892 cache_page_release(desc); 893 goto out; 894 } 895 896 /* The file offset position represents the dirent entry number. A 897 last cookie cache takes care of the common case of reading the 898 whole directory. 899 */ 900 static int nfs_readdir(struct file *file, struct dir_context *ctx) 901 { 902 struct dentry *dentry = file_dentry(file); 903 struct inode *inode = d_inode(dentry); 904 nfs_readdir_descriptor_t my_desc, 905 *desc = &my_desc; 906 struct nfs_open_dir_context *dir_ctx = file->private_data; 907 int res = 0; 908 909 dfprintk(FILE, "NFS: readdir(%pD2) starting at cookie %llu\n", 910 file, (long long)ctx->pos); 911 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS); 912 913 /* 914 * ctx->pos points to the dirent entry number. 915 * *desc->dir_cookie has the cookie for the next entry. We have 916 * to either find the entry with the appropriate number or 917 * revalidate the cookie. 918 */ 919 memset(desc, 0, sizeof(*desc)); 920 921 desc->file = file; 922 desc->ctx = ctx; 923 desc->dir_cookie = &dir_ctx->dir_cookie; 924 desc->decode = NFS_PROTO(inode)->decode_dirent; 925 desc->plus = nfs_use_readdirplus(inode, ctx) ? 1 : 0; 926 927 if (ctx->pos == 0 || nfs_attribute_cache_expired(inode)) 928 res = nfs_revalidate_mapping(inode, file->f_mapping); 929 if (res < 0) 930 goto out; 931 932 do { 933 res = readdir_search_pagecache(desc); 934 935 if (res == -EBADCOOKIE) { 936 res = 0; 937 /* This means either end of directory */ 938 if (*desc->dir_cookie && desc->eof == 0) { 939 /* Or that the server has 'lost' a cookie */ 940 res = uncached_readdir(desc); 941 if (res == 0) 942 continue; 943 } 944 break; 945 } 946 if (res == -ETOOSMALL && desc->plus) { 947 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags); 948 nfs_zap_caches(inode); 949 desc->page_index = 0; 950 desc->plus = 0; 951 desc->eof = 0; 952 continue; 953 } 954 if (res < 0) 955 break; 956 957 res = nfs_do_filldir(desc); 958 if (res < 0) 959 break; 960 } while (!desc->eof); 961 out: 962 if (res > 0) 963 res = 0; 964 dfprintk(FILE, "NFS: readdir(%pD2) returns %d\n", file, res); 965 return res; 966 } 967 968 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int whence) 969 { 970 struct nfs_open_dir_context *dir_ctx = filp->private_data; 971 972 dfprintk(FILE, "NFS: llseek dir(%pD2, %lld, %d)\n", 973 filp, offset, whence); 974 975 switch (whence) { 976 case 1: 977 offset += filp->f_pos; 978 case 0: 979 if (offset >= 0) 980 break; 981 default: 982 return -EINVAL; 983 } 984 if (offset != filp->f_pos) { 985 filp->f_pos = offset; 986 dir_ctx->dir_cookie = 0; 987 dir_ctx->duped = 0; 988 } 989 return offset; 990 } 991 992 /* 993 * All directory operations under NFS are synchronous, so fsync() 994 * is a dummy operation. 995 */ 996 static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end, 997 int datasync) 998 { 999 struct inode *inode = file_inode(filp); 1000 1001 dfprintk(FILE, "NFS: fsync dir(%pD2) datasync %d\n", filp, datasync); 1002 1003 inode_lock(inode); 1004 nfs_inc_stats(inode, NFSIOS_VFSFSYNC); 1005 inode_unlock(inode); 1006 return 0; 1007 } 1008 1009 /** 1010 * nfs_force_lookup_revalidate - Mark the directory as having changed 1011 * @dir - pointer to directory inode 1012 * 1013 * This forces the revalidation code in nfs_lookup_revalidate() to do a 1014 * full lookup on all child dentries of 'dir' whenever a change occurs 1015 * on the server that might have invalidated our dcache. 1016 * 1017 * The caller should be holding dir->i_lock 1018 */ 1019 void nfs_force_lookup_revalidate(struct inode *dir) 1020 { 1021 NFS_I(dir)->cache_change_attribute++; 1022 } 1023 EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate); 1024 1025 /* 1026 * A check for whether or not the parent directory has changed. 1027 * In the case it has, we assume that the dentries are untrustworthy 1028 * and may need to be looked up again. 1029 * If rcu_walk prevents us from performing a full check, return 0. 1030 */ 1031 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry, 1032 int rcu_walk) 1033 { 1034 int ret; 1035 1036 if (IS_ROOT(dentry)) 1037 return 1; 1038 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE) 1039 return 0; 1040 if (!nfs_verify_change_attribute(dir, dentry->d_time)) 1041 return 0; 1042 /* Revalidate nfsi->cache_change_attribute before we declare a match */ 1043 if (rcu_walk) 1044 ret = nfs_revalidate_inode_rcu(NFS_SERVER(dir), dir); 1045 else 1046 ret = nfs_revalidate_inode(NFS_SERVER(dir), dir); 1047 if (ret < 0) 1048 return 0; 1049 if (!nfs_verify_change_attribute(dir, dentry->d_time)) 1050 return 0; 1051 return 1; 1052 } 1053 1054 /* 1055 * Use intent information to check whether or not we're going to do 1056 * an O_EXCL create using this path component. 1057 */ 1058 static int nfs_is_exclusive_create(struct inode *dir, unsigned int flags) 1059 { 1060 if (NFS_PROTO(dir)->version == 2) 1061 return 0; 1062 return flags & LOOKUP_EXCL; 1063 } 1064 1065 /* 1066 * Inode and filehandle revalidation for lookups. 1067 * 1068 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL, 1069 * or if the intent information indicates that we're about to open this 1070 * particular file and the "nocto" mount flag is not set. 1071 * 1072 */ 1073 static 1074 int nfs_lookup_verify_inode(struct inode *inode, unsigned int flags) 1075 { 1076 struct nfs_server *server = NFS_SERVER(inode); 1077 int ret; 1078 1079 if (IS_AUTOMOUNT(inode)) 1080 return 0; 1081 /* VFS wants an on-the-wire revalidation */ 1082 if (flags & LOOKUP_REVAL) 1083 goto out_force; 1084 /* This is an open(2) */ 1085 if ((flags & LOOKUP_OPEN) && !(server->flags & NFS_MOUNT_NOCTO) && 1086 (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode))) 1087 goto out_force; 1088 out: 1089 return (inode->i_nlink == 0) ? -ENOENT : 0; 1090 out_force: 1091 if (flags & LOOKUP_RCU) 1092 return -ECHILD; 1093 ret = __nfs_revalidate_inode(server, inode); 1094 if (ret != 0) 1095 return ret; 1096 goto out; 1097 } 1098 1099 /* 1100 * We judge how long we want to trust negative 1101 * dentries by looking at the parent inode mtime. 1102 * 1103 * If parent mtime has changed, we revalidate, else we wait for a 1104 * period corresponding to the parent's attribute cache timeout value. 1105 * 1106 * If LOOKUP_RCU prevents us from performing a full check, return 1 1107 * suggesting a reval is needed. 1108 */ 1109 static inline 1110 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry, 1111 unsigned int flags) 1112 { 1113 /* Don't revalidate a negative dentry if we're creating a new file */ 1114 if (flags & LOOKUP_CREATE) 1115 return 0; 1116 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG) 1117 return 1; 1118 return !nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU); 1119 } 1120 1121 /* 1122 * This is called every time the dcache has a lookup hit, 1123 * and we should check whether we can really trust that 1124 * lookup. 1125 * 1126 * NOTE! The hit can be a negative hit too, don't assume 1127 * we have an inode! 1128 * 1129 * If the parent directory is seen to have changed, we throw out the 1130 * cached dentry and do a new lookup. 1131 */ 1132 static int nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags) 1133 { 1134 struct inode *dir; 1135 struct inode *inode; 1136 struct dentry *parent; 1137 struct nfs_fh *fhandle = NULL; 1138 struct nfs_fattr *fattr = NULL; 1139 struct nfs4_label *label = NULL; 1140 int error; 1141 1142 if (flags & LOOKUP_RCU) { 1143 parent = ACCESS_ONCE(dentry->d_parent); 1144 dir = d_inode_rcu(parent); 1145 if (!dir) 1146 return -ECHILD; 1147 } else { 1148 parent = dget_parent(dentry); 1149 dir = d_inode(parent); 1150 } 1151 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE); 1152 inode = d_inode(dentry); 1153 1154 if (!inode) { 1155 if (nfs_neg_need_reval(dir, dentry, flags)) { 1156 if (flags & LOOKUP_RCU) 1157 return -ECHILD; 1158 goto out_bad; 1159 } 1160 goto out_valid; 1161 } 1162 1163 if (is_bad_inode(inode)) { 1164 if (flags & LOOKUP_RCU) 1165 return -ECHILD; 1166 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n", 1167 __func__, dentry); 1168 goto out_bad; 1169 } 1170 1171 if (NFS_PROTO(dir)->have_delegation(inode, FMODE_READ)) 1172 goto out_set_verifier; 1173 1174 /* Force a full look up iff the parent directory has changed */ 1175 if (!nfs_is_exclusive_create(dir, flags) && 1176 nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU)) { 1177 1178 if (nfs_lookup_verify_inode(inode, flags)) { 1179 if (flags & LOOKUP_RCU) 1180 return -ECHILD; 1181 goto out_zap_parent; 1182 } 1183 nfs_advise_use_readdirplus(dir); 1184 goto out_valid; 1185 } 1186 1187 if (flags & LOOKUP_RCU) 1188 return -ECHILD; 1189 1190 if (NFS_STALE(inode)) 1191 goto out_bad; 1192 1193 error = -ENOMEM; 1194 fhandle = nfs_alloc_fhandle(); 1195 fattr = nfs_alloc_fattr(); 1196 if (fhandle == NULL || fattr == NULL) 1197 goto out_error; 1198 1199 label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT); 1200 if (IS_ERR(label)) 1201 goto out_error; 1202 1203 trace_nfs_lookup_revalidate_enter(dir, dentry, flags); 1204 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, label); 1205 trace_nfs_lookup_revalidate_exit(dir, dentry, flags, error); 1206 if (error) 1207 goto out_bad; 1208 if (nfs_compare_fh(NFS_FH(inode), fhandle)) 1209 goto out_bad; 1210 if ((error = nfs_refresh_inode(inode, fattr)) != 0) 1211 goto out_bad; 1212 1213 nfs_setsecurity(inode, fattr, label); 1214 1215 nfs_free_fattr(fattr); 1216 nfs_free_fhandle(fhandle); 1217 nfs4_label_free(label); 1218 1219 /* set a readdirplus hint that we had a cache miss */ 1220 nfs_force_use_readdirplus(dir); 1221 1222 out_set_verifier: 1223 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1224 out_valid: 1225 if (flags & LOOKUP_RCU) { 1226 if (parent != ACCESS_ONCE(dentry->d_parent)) 1227 return -ECHILD; 1228 } else 1229 dput(parent); 1230 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is valid\n", 1231 __func__, dentry); 1232 return 1; 1233 out_zap_parent: 1234 nfs_zap_caches(dir); 1235 out_bad: 1236 WARN_ON(flags & LOOKUP_RCU); 1237 nfs_free_fattr(fattr); 1238 nfs_free_fhandle(fhandle); 1239 nfs4_label_free(label); 1240 nfs_mark_for_revalidate(dir); 1241 if (inode && S_ISDIR(inode->i_mode)) { 1242 /* Purge readdir caches. */ 1243 nfs_zap_caches(inode); 1244 /* 1245 * We can't d_drop the root of a disconnected tree: 1246 * its d_hash is on the s_anon list and d_drop() would hide 1247 * it from shrink_dcache_for_unmount(), leading to busy 1248 * inodes on unmount and further oopses. 1249 */ 1250 if (IS_ROOT(dentry)) 1251 goto out_valid; 1252 } 1253 dput(parent); 1254 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is invalid\n", 1255 __func__, dentry); 1256 return 0; 1257 out_error: 1258 WARN_ON(flags & LOOKUP_RCU); 1259 nfs_free_fattr(fattr); 1260 nfs_free_fhandle(fhandle); 1261 nfs4_label_free(label); 1262 dput(parent); 1263 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) lookup returned error %d\n", 1264 __func__, dentry, error); 1265 return error; 1266 } 1267 1268 /* 1269 * A weaker form of d_revalidate for revalidating just the d_inode(dentry) 1270 * when we don't really care about the dentry name. This is called when a 1271 * pathwalk ends on a dentry that was not found via a normal lookup in the 1272 * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals). 1273 * 1274 * In this situation, we just want to verify that the inode itself is OK 1275 * since the dentry might have changed on the server. 1276 */ 1277 static int nfs_weak_revalidate(struct dentry *dentry, unsigned int flags) 1278 { 1279 int error; 1280 struct inode *inode = d_inode(dentry); 1281 1282 /* 1283 * I believe we can only get a negative dentry here in the case of a 1284 * procfs-style symlink. Just assume it's correct for now, but we may 1285 * eventually need to do something more here. 1286 */ 1287 if (!inode) { 1288 dfprintk(LOOKUPCACHE, "%s: %pd2 has negative inode\n", 1289 __func__, dentry); 1290 return 1; 1291 } 1292 1293 if (is_bad_inode(inode)) { 1294 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n", 1295 __func__, dentry); 1296 return 0; 1297 } 1298 1299 error = nfs_revalidate_inode(NFS_SERVER(inode), inode); 1300 dfprintk(LOOKUPCACHE, "NFS: %s: inode %lu is %s\n", 1301 __func__, inode->i_ino, error ? "invalid" : "valid"); 1302 return !error; 1303 } 1304 1305 /* 1306 * This is called from dput() when d_count is going to 0. 1307 */ 1308 static int nfs_dentry_delete(const struct dentry *dentry) 1309 { 1310 dfprintk(VFS, "NFS: dentry_delete(%pd2, %x)\n", 1311 dentry, dentry->d_flags); 1312 1313 /* Unhash any dentry with a stale inode */ 1314 if (d_really_is_positive(dentry) && NFS_STALE(d_inode(dentry))) 1315 return 1; 1316 1317 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { 1318 /* Unhash it, so that ->d_iput() would be called */ 1319 return 1; 1320 } 1321 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) { 1322 /* Unhash it, so that ancestors of killed async unlink 1323 * files will be cleaned up during umount */ 1324 return 1; 1325 } 1326 return 0; 1327 1328 } 1329 1330 /* Ensure that we revalidate inode->i_nlink */ 1331 static void nfs_drop_nlink(struct inode *inode) 1332 { 1333 spin_lock(&inode->i_lock); 1334 /* drop the inode if we're reasonably sure this is the last link */ 1335 if (inode->i_nlink == 1) 1336 clear_nlink(inode); 1337 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATTR; 1338 spin_unlock(&inode->i_lock); 1339 } 1340 1341 /* 1342 * Called when the dentry loses inode. 1343 * We use it to clean up silly-renamed files. 1344 */ 1345 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode) 1346 { 1347 if (S_ISDIR(inode->i_mode)) 1348 /* drop any readdir cache as it could easily be old */ 1349 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA; 1350 1351 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { 1352 nfs_complete_unlink(dentry, inode); 1353 nfs_drop_nlink(inode); 1354 } 1355 iput(inode); 1356 } 1357 1358 static void nfs_d_release(struct dentry *dentry) 1359 { 1360 /* free cached devname value, if it survived that far */ 1361 if (unlikely(dentry->d_fsdata)) { 1362 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) 1363 WARN_ON(1); 1364 else 1365 kfree(dentry->d_fsdata); 1366 } 1367 } 1368 1369 const struct dentry_operations nfs_dentry_operations = { 1370 .d_revalidate = nfs_lookup_revalidate, 1371 .d_weak_revalidate = nfs_weak_revalidate, 1372 .d_delete = nfs_dentry_delete, 1373 .d_iput = nfs_dentry_iput, 1374 .d_automount = nfs_d_automount, 1375 .d_release = nfs_d_release, 1376 }; 1377 EXPORT_SYMBOL_GPL(nfs_dentry_operations); 1378 1379 struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags) 1380 { 1381 struct dentry *res; 1382 struct inode *inode = NULL; 1383 struct nfs_fh *fhandle = NULL; 1384 struct nfs_fattr *fattr = NULL; 1385 struct nfs4_label *label = NULL; 1386 int error; 1387 1388 dfprintk(VFS, "NFS: lookup(%pd2)\n", dentry); 1389 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP); 1390 1391 if (unlikely(dentry->d_name.len > NFS_SERVER(dir)->namelen)) 1392 return ERR_PTR(-ENAMETOOLONG); 1393 1394 /* 1395 * If we're doing an exclusive create, optimize away the lookup 1396 * but don't hash the dentry. 1397 */ 1398 if (nfs_is_exclusive_create(dir, flags)) 1399 return NULL; 1400 1401 res = ERR_PTR(-ENOMEM); 1402 fhandle = nfs_alloc_fhandle(); 1403 fattr = nfs_alloc_fattr(); 1404 if (fhandle == NULL || fattr == NULL) 1405 goto out; 1406 1407 label = nfs4_label_alloc(NFS_SERVER(dir), GFP_NOWAIT); 1408 if (IS_ERR(label)) 1409 goto out; 1410 1411 trace_nfs_lookup_enter(dir, dentry, flags); 1412 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, label); 1413 if (error == -ENOENT) 1414 goto no_entry; 1415 if (error < 0) { 1416 res = ERR_PTR(error); 1417 goto out_label; 1418 } 1419 inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label); 1420 res = ERR_CAST(inode); 1421 if (IS_ERR(res)) 1422 goto out_label; 1423 1424 /* Notify readdir to use READDIRPLUS */ 1425 nfs_force_use_readdirplus(dir); 1426 1427 no_entry: 1428 res = d_splice_alias(inode, dentry); 1429 if (res != NULL) { 1430 if (IS_ERR(res)) 1431 goto out_label; 1432 dentry = res; 1433 } 1434 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1435 out_label: 1436 trace_nfs_lookup_exit(dir, dentry, flags, error); 1437 nfs4_label_free(label); 1438 out: 1439 nfs_free_fattr(fattr); 1440 nfs_free_fhandle(fhandle); 1441 return res; 1442 } 1443 EXPORT_SYMBOL_GPL(nfs_lookup); 1444 1445 #if IS_ENABLED(CONFIG_NFS_V4) 1446 static int nfs4_lookup_revalidate(struct dentry *, unsigned int); 1447 1448 const struct dentry_operations nfs4_dentry_operations = { 1449 .d_revalidate = nfs4_lookup_revalidate, 1450 .d_delete = nfs_dentry_delete, 1451 .d_iput = nfs_dentry_iput, 1452 .d_automount = nfs_d_automount, 1453 .d_release = nfs_d_release, 1454 }; 1455 EXPORT_SYMBOL_GPL(nfs4_dentry_operations); 1456 1457 static fmode_t flags_to_mode(int flags) 1458 { 1459 fmode_t res = (__force fmode_t)flags & FMODE_EXEC; 1460 if ((flags & O_ACCMODE) != O_WRONLY) 1461 res |= FMODE_READ; 1462 if ((flags & O_ACCMODE) != O_RDONLY) 1463 res |= FMODE_WRITE; 1464 return res; 1465 } 1466 1467 static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags, struct file *filp) 1468 { 1469 return alloc_nfs_open_context(dentry, flags_to_mode(open_flags), filp); 1470 } 1471 1472 static int do_open(struct inode *inode, struct file *filp) 1473 { 1474 nfs_fscache_open_file(inode, filp); 1475 return 0; 1476 } 1477 1478 static int nfs_finish_open(struct nfs_open_context *ctx, 1479 struct dentry *dentry, 1480 struct file *file, unsigned open_flags, 1481 int *opened) 1482 { 1483 int err; 1484 1485 err = finish_open(file, dentry, do_open, opened); 1486 if (err) 1487 goto out; 1488 nfs_file_set_open_context(file, ctx); 1489 1490 out: 1491 return err; 1492 } 1493 1494 int nfs_atomic_open(struct inode *dir, struct dentry *dentry, 1495 struct file *file, unsigned open_flags, 1496 umode_t mode, int *opened) 1497 { 1498 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 1499 struct nfs_open_context *ctx; 1500 struct dentry *res; 1501 struct iattr attr = { .ia_valid = ATTR_OPEN }; 1502 struct inode *inode; 1503 unsigned int lookup_flags = 0; 1504 bool switched = false; 1505 int err; 1506 1507 /* Expect a negative dentry */ 1508 BUG_ON(d_inode(dentry)); 1509 1510 dfprintk(VFS, "NFS: atomic_open(%s/%lu), %pd\n", 1511 dir->i_sb->s_id, dir->i_ino, dentry); 1512 1513 err = nfs_check_flags(open_flags); 1514 if (err) 1515 return err; 1516 1517 /* NFS only supports OPEN on regular files */ 1518 if ((open_flags & O_DIRECTORY)) { 1519 if (!d_in_lookup(dentry)) { 1520 /* 1521 * Hashed negative dentry with O_DIRECTORY: dentry was 1522 * revalidated and is fine, no need to perform lookup 1523 * again 1524 */ 1525 return -ENOENT; 1526 } 1527 lookup_flags = LOOKUP_OPEN|LOOKUP_DIRECTORY; 1528 goto no_open; 1529 } 1530 1531 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) 1532 return -ENAMETOOLONG; 1533 1534 if (open_flags & O_CREAT) { 1535 attr.ia_valid |= ATTR_MODE; 1536 attr.ia_mode = mode & ~current_umask(); 1537 } 1538 if (open_flags & O_TRUNC) { 1539 attr.ia_valid |= ATTR_SIZE; 1540 attr.ia_size = 0; 1541 } 1542 1543 if (!(open_flags & O_CREAT) && !d_in_lookup(dentry)) { 1544 d_drop(dentry); 1545 switched = true; 1546 dentry = d_alloc_parallel(dentry->d_parent, 1547 &dentry->d_name, &wq); 1548 if (IS_ERR(dentry)) 1549 return PTR_ERR(dentry); 1550 if (unlikely(!d_in_lookup(dentry))) 1551 return finish_no_open(file, dentry); 1552 } 1553 1554 ctx = create_nfs_open_context(dentry, open_flags, file); 1555 err = PTR_ERR(ctx); 1556 if (IS_ERR(ctx)) 1557 goto out; 1558 1559 trace_nfs_atomic_open_enter(dir, ctx, open_flags); 1560 inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr, opened); 1561 if (IS_ERR(inode)) { 1562 err = PTR_ERR(inode); 1563 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err); 1564 put_nfs_open_context(ctx); 1565 d_drop(dentry); 1566 switch (err) { 1567 case -ENOENT: 1568 d_add(dentry, NULL); 1569 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1570 break; 1571 case -EISDIR: 1572 case -ENOTDIR: 1573 goto no_open; 1574 case -ELOOP: 1575 if (!(open_flags & O_NOFOLLOW)) 1576 goto no_open; 1577 break; 1578 /* case -EINVAL: */ 1579 default: 1580 break; 1581 } 1582 goto out; 1583 } 1584 1585 err = nfs_finish_open(ctx, ctx->dentry, file, open_flags, opened); 1586 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err); 1587 put_nfs_open_context(ctx); 1588 out: 1589 if (unlikely(switched)) { 1590 d_lookup_done(dentry); 1591 dput(dentry); 1592 } 1593 return err; 1594 1595 no_open: 1596 res = nfs_lookup(dir, dentry, lookup_flags); 1597 if (switched) { 1598 d_lookup_done(dentry); 1599 if (!res) 1600 res = dentry; 1601 else 1602 dput(dentry); 1603 } 1604 if (IS_ERR(res)) 1605 return PTR_ERR(res); 1606 return finish_no_open(file, res); 1607 } 1608 EXPORT_SYMBOL_GPL(nfs_atomic_open); 1609 1610 static int nfs4_lookup_revalidate(struct dentry *dentry, unsigned int flags) 1611 { 1612 struct inode *inode; 1613 int ret = 0; 1614 1615 if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY)) 1616 goto no_open; 1617 if (d_mountpoint(dentry)) 1618 goto no_open; 1619 if (NFS_SB(dentry->d_sb)->caps & NFS_CAP_ATOMIC_OPEN_V1) 1620 goto no_open; 1621 1622 inode = d_inode(dentry); 1623 1624 /* We can't create new files in nfs_open_revalidate(), so we 1625 * optimize away revalidation of negative dentries. 1626 */ 1627 if (inode == NULL) { 1628 struct dentry *parent; 1629 struct inode *dir; 1630 1631 if (flags & LOOKUP_RCU) { 1632 parent = ACCESS_ONCE(dentry->d_parent); 1633 dir = d_inode_rcu(parent); 1634 if (!dir) 1635 return -ECHILD; 1636 } else { 1637 parent = dget_parent(dentry); 1638 dir = d_inode(parent); 1639 } 1640 if (!nfs_neg_need_reval(dir, dentry, flags)) 1641 ret = 1; 1642 else if (flags & LOOKUP_RCU) 1643 ret = -ECHILD; 1644 if (!(flags & LOOKUP_RCU)) 1645 dput(parent); 1646 else if (parent != ACCESS_ONCE(dentry->d_parent)) 1647 return -ECHILD; 1648 goto out; 1649 } 1650 1651 /* NFS only supports OPEN on regular files */ 1652 if (!S_ISREG(inode->i_mode)) 1653 goto no_open; 1654 /* We cannot do exclusive creation on a positive dentry */ 1655 if (flags & LOOKUP_EXCL) 1656 goto no_open; 1657 1658 /* Let f_op->open() actually open (and revalidate) the file */ 1659 ret = 1; 1660 1661 out: 1662 return ret; 1663 1664 no_open: 1665 return nfs_lookup_revalidate(dentry, flags); 1666 } 1667 1668 #endif /* CONFIG_NFSV4 */ 1669 1670 /* 1671 * Code common to create, mkdir, and mknod. 1672 */ 1673 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle, 1674 struct nfs_fattr *fattr, 1675 struct nfs4_label *label) 1676 { 1677 struct dentry *parent = dget_parent(dentry); 1678 struct inode *dir = d_inode(parent); 1679 struct inode *inode; 1680 int error = -EACCES; 1681 1682 d_drop(dentry); 1683 1684 /* We may have been initialized further down */ 1685 if (d_really_is_positive(dentry)) 1686 goto out; 1687 if (fhandle->size == 0) { 1688 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, NULL); 1689 if (error) 1690 goto out_error; 1691 } 1692 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1693 if (!(fattr->valid & NFS_ATTR_FATTR)) { 1694 struct nfs_server *server = NFS_SB(dentry->d_sb); 1695 error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr, NULL); 1696 if (error < 0) 1697 goto out_error; 1698 } 1699 inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label); 1700 error = PTR_ERR(inode); 1701 if (IS_ERR(inode)) 1702 goto out_error; 1703 d_add(dentry, inode); 1704 out: 1705 dput(parent); 1706 return 0; 1707 out_error: 1708 nfs_mark_for_revalidate(dir); 1709 dput(parent); 1710 return error; 1711 } 1712 EXPORT_SYMBOL_GPL(nfs_instantiate); 1713 1714 /* 1715 * Following a failed create operation, we drop the dentry rather 1716 * than retain a negative dentry. This avoids a problem in the event 1717 * that the operation succeeded on the server, but an error in the 1718 * reply path made it appear to have failed. 1719 */ 1720 int nfs_create(struct inode *dir, struct dentry *dentry, 1721 umode_t mode, bool excl) 1722 { 1723 struct iattr attr; 1724 int open_flags = excl ? O_CREAT | O_EXCL : O_CREAT; 1725 int error; 1726 1727 dfprintk(VFS, "NFS: create(%s/%lu), %pd\n", 1728 dir->i_sb->s_id, dir->i_ino, dentry); 1729 1730 attr.ia_mode = mode; 1731 attr.ia_valid = ATTR_MODE; 1732 1733 trace_nfs_create_enter(dir, dentry, open_flags); 1734 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags); 1735 trace_nfs_create_exit(dir, dentry, open_flags, error); 1736 if (error != 0) 1737 goto out_err; 1738 return 0; 1739 out_err: 1740 d_drop(dentry); 1741 return error; 1742 } 1743 EXPORT_SYMBOL_GPL(nfs_create); 1744 1745 /* 1746 * See comments for nfs_proc_create regarding failed operations. 1747 */ 1748 int 1749 nfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t rdev) 1750 { 1751 struct iattr attr; 1752 int status; 1753 1754 dfprintk(VFS, "NFS: mknod(%s/%lu), %pd\n", 1755 dir->i_sb->s_id, dir->i_ino, dentry); 1756 1757 attr.ia_mode = mode; 1758 attr.ia_valid = ATTR_MODE; 1759 1760 trace_nfs_mknod_enter(dir, dentry); 1761 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev); 1762 trace_nfs_mknod_exit(dir, dentry, status); 1763 if (status != 0) 1764 goto out_err; 1765 return 0; 1766 out_err: 1767 d_drop(dentry); 1768 return status; 1769 } 1770 EXPORT_SYMBOL_GPL(nfs_mknod); 1771 1772 /* 1773 * See comments for nfs_proc_create regarding failed operations. 1774 */ 1775 int nfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) 1776 { 1777 struct iattr attr; 1778 int error; 1779 1780 dfprintk(VFS, "NFS: mkdir(%s/%lu), %pd\n", 1781 dir->i_sb->s_id, dir->i_ino, dentry); 1782 1783 attr.ia_valid = ATTR_MODE; 1784 attr.ia_mode = mode | S_IFDIR; 1785 1786 trace_nfs_mkdir_enter(dir, dentry); 1787 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr); 1788 trace_nfs_mkdir_exit(dir, dentry, error); 1789 if (error != 0) 1790 goto out_err; 1791 return 0; 1792 out_err: 1793 d_drop(dentry); 1794 return error; 1795 } 1796 EXPORT_SYMBOL_GPL(nfs_mkdir); 1797 1798 static void nfs_dentry_handle_enoent(struct dentry *dentry) 1799 { 1800 if (simple_positive(dentry)) 1801 d_delete(dentry); 1802 } 1803 1804 int nfs_rmdir(struct inode *dir, struct dentry *dentry) 1805 { 1806 int error; 1807 1808 dfprintk(VFS, "NFS: rmdir(%s/%lu), %pd\n", 1809 dir->i_sb->s_id, dir->i_ino, dentry); 1810 1811 trace_nfs_rmdir_enter(dir, dentry); 1812 if (d_really_is_positive(dentry)) { 1813 down_write(&NFS_I(d_inode(dentry))->rmdir_sem); 1814 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name); 1815 /* Ensure the VFS deletes this inode */ 1816 switch (error) { 1817 case 0: 1818 clear_nlink(d_inode(dentry)); 1819 break; 1820 case -ENOENT: 1821 nfs_dentry_handle_enoent(dentry); 1822 } 1823 up_write(&NFS_I(d_inode(dentry))->rmdir_sem); 1824 } else 1825 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name); 1826 trace_nfs_rmdir_exit(dir, dentry, error); 1827 1828 return error; 1829 } 1830 EXPORT_SYMBOL_GPL(nfs_rmdir); 1831 1832 /* 1833 * Remove a file after making sure there are no pending writes, 1834 * and after checking that the file has only one user. 1835 * 1836 * We invalidate the attribute cache and free the inode prior to the operation 1837 * to avoid possible races if the server reuses the inode. 1838 */ 1839 static int nfs_safe_remove(struct dentry *dentry) 1840 { 1841 struct inode *dir = d_inode(dentry->d_parent); 1842 struct inode *inode = d_inode(dentry); 1843 int error = -EBUSY; 1844 1845 dfprintk(VFS, "NFS: safe_remove(%pd2)\n", dentry); 1846 1847 /* If the dentry was sillyrenamed, we simply call d_delete() */ 1848 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { 1849 error = 0; 1850 goto out; 1851 } 1852 1853 trace_nfs_remove_enter(dir, dentry); 1854 if (inode != NULL) { 1855 NFS_PROTO(inode)->return_delegation(inode); 1856 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name); 1857 if (error == 0) 1858 nfs_drop_nlink(inode); 1859 } else 1860 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name); 1861 if (error == -ENOENT) 1862 nfs_dentry_handle_enoent(dentry); 1863 trace_nfs_remove_exit(dir, dentry, error); 1864 out: 1865 return error; 1866 } 1867 1868 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode 1869 * belongs to an active ".nfs..." file and we return -EBUSY. 1870 * 1871 * If sillyrename() returns 0, we do nothing, otherwise we unlink. 1872 */ 1873 int nfs_unlink(struct inode *dir, struct dentry *dentry) 1874 { 1875 int error; 1876 int need_rehash = 0; 1877 1878 dfprintk(VFS, "NFS: unlink(%s/%lu, %pd)\n", dir->i_sb->s_id, 1879 dir->i_ino, dentry); 1880 1881 trace_nfs_unlink_enter(dir, dentry); 1882 spin_lock(&dentry->d_lock); 1883 if (d_count(dentry) > 1) { 1884 spin_unlock(&dentry->d_lock); 1885 /* Start asynchronous writeout of the inode */ 1886 write_inode_now(d_inode(dentry), 0); 1887 error = nfs_sillyrename(dir, dentry); 1888 goto out; 1889 } 1890 if (!d_unhashed(dentry)) { 1891 __d_drop(dentry); 1892 need_rehash = 1; 1893 } 1894 spin_unlock(&dentry->d_lock); 1895 error = nfs_safe_remove(dentry); 1896 if (!error || error == -ENOENT) { 1897 nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); 1898 } else if (need_rehash) 1899 d_rehash(dentry); 1900 out: 1901 trace_nfs_unlink_exit(dir, dentry, error); 1902 return error; 1903 } 1904 EXPORT_SYMBOL_GPL(nfs_unlink); 1905 1906 /* 1907 * To create a symbolic link, most file systems instantiate a new inode, 1908 * add a page to it containing the path, then write it out to the disk 1909 * using prepare_write/commit_write. 1910 * 1911 * Unfortunately the NFS client can't create the in-core inode first 1912 * because it needs a file handle to create an in-core inode (see 1913 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the 1914 * symlink request has completed on the server. 1915 * 1916 * So instead we allocate a raw page, copy the symname into it, then do 1917 * the SYMLINK request with the page as the buffer. If it succeeds, we 1918 * now have a new file handle and can instantiate an in-core NFS inode 1919 * and move the raw page into its mapping. 1920 */ 1921 int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname) 1922 { 1923 struct page *page; 1924 char *kaddr; 1925 struct iattr attr; 1926 unsigned int pathlen = strlen(symname); 1927 int error; 1928 1929 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s)\n", dir->i_sb->s_id, 1930 dir->i_ino, dentry, symname); 1931 1932 if (pathlen > PAGE_SIZE) 1933 return -ENAMETOOLONG; 1934 1935 attr.ia_mode = S_IFLNK | S_IRWXUGO; 1936 attr.ia_valid = ATTR_MODE; 1937 1938 page = alloc_page(GFP_USER); 1939 if (!page) 1940 return -ENOMEM; 1941 1942 kaddr = page_address(page); 1943 memcpy(kaddr, symname, pathlen); 1944 if (pathlen < PAGE_SIZE) 1945 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen); 1946 1947 trace_nfs_symlink_enter(dir, dentry); 1948 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr); 1949 trace_nfs_symlink_exit(dir, dentry, error); 1950 if (error != 0) { 1951 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s) error %d\n", 1952 dir->i_sb->s_id, dir->i_ino, 1953 dentry, symname, error); 1954 d_drop(dentry); 1955 __free_page(page); 1956 return error; 1957 } 1958 1959 /* 1960 * No big deal if we can't add this page to the page cache here. 1961 * READLINK will get the missing page from the server if needed. 1962 */ 1963 if (!add_to_page_cache_lru(page, d_inode(dentry)->i_mapping, 0, 1964 GFP_KERNEL)) { 1965 SetPageUptodate(page); 1966 unlock_page(page); 1967 /* 1968 * add_to_page_cache_lru() grabs an extra page refcount. 1969 * Drop it here to avoid leaking this page later. 1970 */ 1971 put_page(page); 1972 } else 1973 __free_page(page); 1974 1975 return 0; 1976 } 1977 EXPORT_SYMBOL_GPL(nfs_symlink); 1978 1979 int 1980 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) 1981 { 1982 struct inode *inode = d_inode(old_dentry); 1983 int error; 1984 1985 dfprintk(VFS, "NFS: link(%pd2 -> %pd2)\n", 1986 old_dentry, dentry); 1987 1988 trace_nfs_link_enter(inode, dir, dentry); 1989 NFS_PROTO(inode)->return_delegation(inode); 1990 1991 d_drop(dentry); 1992 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name); 1993 if (error == 0) { 1994 ihold(inode); 1995 d_add(dentry, inode); 1996 } 1997 trace_nfs_link_exit(inode, dir, dentry, error); 1998 return error; 1999 } 2000 EXPORT_SYMBOL_GPL(nfs_link); 2001 2002 /* 2003 * RENAME 2004 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a 2005 * different file handle for the same inode after a rename (e.g. when 2006 * moving to a different directory). A fail-safe method to do so would 2007 * be to look up old_dir/old_name, create a link to new_dir/new_name and 2008 * rename the old file using the sillyrename stuff. This way, the original 2009 * file in old_dir will go away when the last process iput()s the inode. 2010 * 2011 * FIXED. 2012 * 2013 * It actually works quite well. One needs to have the possibility for 2014 * at least one ".nfs..." file in each directory the file ever gets 2015 * moved or linked to which happens automagically with the new 2016 * implementation that only depends on the dcache stuff instead of 2017 * using the inode layer 2018 * 2019 * Unfortunately, things are a little more complicated than indicated 2020 * above. For a cross-directory move, we want to make sure we can get 2021 * rid of the old inode after the operation. This means there must be 2022 * no pending writes (if it's a file), and the use count must be 1. 2023 * If these conditions are met, we can drop the dentries before doing 2024 * the rename. 2025 */ 2026 int nfs_rename(struct inode *old_dir, struct dentry *old_dentry, 2027 struct inode *new_dir, struct dentry *new_dentry, 2028 unsigned int flags) 2029 { 2030 struct inode *old_inode = d_inode(old_dentry); 2031 struct inode *new_inode = d_inode(new_dentry); 2032 struct dentry *dentry = NULL, *rehash = NULL; 2033 struct rpc_task *task; 2034 int error = -EBUSY; 2035 2036 if (flags) 2037 return -EINVAL; 2038 2039 dfprintk(VFS, "NFS: rename(%pd2 -> %pd2, ct=%d)\n", 2040 old_dentry, new_dentry, 2041 d_count(new_dentry)); 2042 2043 trace_nfs_rename_enter(old_dir, old_dentry, new_dir, new_dentry); 2044 /* 2045 * For non-directories, check whether the target is busy and if so, 2046 * make a copy of the dentry and then do a silly-rename. If the 2047 * silly-rename succeeds, the copied dentry is hashed and becomes 2048 * the new target. 2049 */ 2050 if (new_inode && !S_ISDIR(new_inode->i_mode)) { 2051 /* 2052 * To prevent any new references to the target during the 2053 * rename, we unhash the dentry in advance. 2054 */ 2055 if (!d_unhashed(new_dentry)) { 2056 d_drop(new_dentry); 2057 rehash = new_dentry; 2058 } 2059 2060 if (d_count(new_dentry) > 2) { 2061 int err; 2062 2063 /* copy the target dentry's name */ 2064 dentry = d_alloc(new_dentry->d_parent, 2065 &new_dentry->d_name); 2066 if (!dentry) 2067 goto out; 2068 2069 /* silly-rename the existing target ... */ 2070 err = nfs_sillyrename(new_dir, new_dentry); 2071 if (err) 2072 goto out; 2073 2074 new_dentry = dentry; 2075 rehash = NULL; 2076 new_inode = NULL; 2077 } 2078 } 2079 2080 NFS_PROTO(old_inode)->return_delegation(old_inode); 2081 if (new_inode != NULL) 2082 NFS_PROTO(new_inode)->return_delegation(new_inode); 2083 2084 task = nfs_async_rename(old_dir, new_dir, old_dentry, new_dentry, NULL); 2085 if (IS_ERR(task)) { 2086 error = PTR_ERR(task); 2087 goto out; 2088 } 2089 2090 error = rpc_wait_for_completion_task(task); 2091 if (error == 0) 2092 error = task->tk_status; 2093 rpc_put_task(task); 2094 nfs_mark_for_revalidate(old_inode); 2095 out: 2096 if (rehash) 2097 d_rehash(rehash); 2098 trace_nfs_rename_exit(old_dir, old_dentry, 2099 new_dir, new_dentry, error); 2100 if (!error) { 2101 if (new_inode != NULL) 2102 nfs_drop_nlink(new_inode); 2103 d_move(old_dentry, new_dentry); 2104 nfs_set_verifier(new_dentry, 2105 nfs_save_change_attribute(new_dir)); 2106 } else if (error == -ENOENT) 2107 nfs_dentry_handle_enoent(old_dentry); 2108 2109 /* new dentry created? */ 2110 if (dentry) 2111 dput(dentry); 2112 return error; 2113 } 2114 EXPORT_SYMBOL_GPL(nfs_rename); 2115 2116 static DEFINE_SPINLOCK(nfs_access_lru_lock); 2117 static LIST_HEAD(nfs_access_lru_list); 2118 static atomic_long_t nfs_access_nr_entries; 2119 2120 static unsigned long nfs_access_max_cachesize = ULONG_MAX; 2121 module_param(nfs_access_max_cachesize, ulong, 0644); 2122 MODULE_PARM_DESC(nfs_access_max_cachesize, "NFS access maximum total cache length"); 2123 2124 static void nfs_access_free_entry(struct nfs_access_entry *entry) 2125 { 2126 put_rpccred(entry->cred); 2127 kfree_rcu(entry, rcu_head); 2128 smp_mb__before_atomic(); 2129 atomic_long_dec(&nfs_access_nr_entries); 2130 smp_mb__after_atomic(); 2131 } 2132 2133 static void nfs_access_free_list(struct list_head *head) 2134 { 2135 struct nfs_access_entry *cache; 2136 2137 while (!list_empty(head)) { 2138 cache = list_entry(head->next, struct nfs_access_entry, lru); 2139 list_del(&cache->lru); 2140 nfs_access_free_entry(cache); 2141 } 2142 } 2143 2144 static unsigned long 2145 nfs_do_access_cache_scan(unsigned int nr_to_scan) 2146 { 2147 LIST_HEAD(head); 2148 struct nfs_inode *nfsi, *next; 2149 struct nfs_access_entry *cache; 2150 long freed = 0; 2151 2152 spin_lock(&nfs_access_lru_lock); 2153 list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) { 2154 struct inode *inode; 2155 2156 if (nr_to_scan-- == 0) 2157 break; 2158 inode = &nfsi->vfs_inode; 2159 spin_lock(&inode->i_lock); 2160 if (list_empty(&nfsi->access_cache_entry_lru)) 2161 goto remove_lru_entry; 2162 cache = list_entry(nfsi->access_cache_entry_lru.next, 2163 struct nfs_access_entry, lru); 2164 list_move(&cache->lru, &head); 2165 rb_erase(&cache->rb_node, &nfsi->access_cache); 2166 freed++; 2167 if (!list_empty(&nfsi->access_cache_entry_lru)) 2168 list_move_tail(&nfsi->access_cache_inode_lru, 2169 &nfs_access_lru_list); 2170 else { 2171 remove_lru_entry: 2172 list_del_init(&nfsi->access_cache_inode_lru); 2173 smp_mb__before_atomic(); 2174 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags); 2175 smp_mb__after_atomic(); 2176 } 2177 spin_unlock(&inode->i_lock); 2178 } 2179 spin_unlock(&nfs_access_lru_lock); 2180 nfs_access_free_list(&head); 2181 return freed; 2182 } 2183 2184 unsigned long 2185 nfs_access_cache_scan(struct shrinker *shrink, struct shrink_control *sc) 2186 { 2187 int nr_to_scan = sc->nr_to_scan; 2188 gfp_t gfp_mask = sc->gfp_mask; 2189 2190 if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL) 2191 return SHRINK_STOP; 2192 return nfs_do_access_cache_scan(nr_to_scan); 2193 } 2194 2195 2196 unsigned long 2197 nfs_access_cache_count(struct shrinker *shrink, struct shrink_control *sc) 2198 { 2199 return vfs_pressure_ratio(atomic_long_read(&nfs_access_nr_entries)); 2200 } 2201 2202 static void 2203 nfs_access_cache_enforce_limit(void) 2204 { 2205 long nr_entries = atomic_long_read(&nfs_access_nr_entries); 2206 unsigned long diff; 2207 unsigned int nr_to_scan; 2208 2209 if (nr_entries < 0 || nr_entries <= nfs_access_max_cachesize) 2210 return; 2211 nr_to_scan = 100; 2212 diff = nr_entries - nfs_access_max_cachesize; 2213 if (diff < nr_to_scan) 2214 nr_to_scan = diff; 2215 nfs_do_access_cache_scan(nr_to_scan); 2216 } 2217 2218 static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head) 2219 { 2220 struct rb_root *root_node = &nfsi->access_cache; 2221 struct rb_node *n; 2222 struct nfs_access_entry *entry; 2223 2224 /* Unhook entries from the cache */ 2225 while ((n = rb_first(root_node)) != NULL) { 2226 entry = rb_entry(n, struct nfs_access_entry, rb_node); 2227 rb_erase(n, root_node); 2228 list_move(&entry->lru, head); 2229 } 2230 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS; 2231 } 2232 2233 void nfs_access_zap_cache(struct inode *inode) 2234 { 2235 LIST_HEAD(head); 2236 2237 if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0) 2238 return; 2239 /* Remove from global LRU init */ 2240 spin_lock(&nfs_access_lru_lock); 2241 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) 2242 list_del_init(&NFS_I(inode)->access_cache_inode_lru); 2243 2244 spin_lock(&inode->i_lock); 2245 __nfs_access_zap_cache(NFS_I(inode), &head); 2246 spin_unlock(&inode->i_lock); 2247 spin_unlock(&nfs_access_lru_lock); 2248 nfs_access_free_list(&head); 2249 } 2250 EXPORT_SYMBOL_GPL(nfs_access_zap_cache); 2251 2252 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred) 2253 { 2254 struct rb_node *n = NFS_I(inode)->access_cache.rb_node; 2255 struct nfs_access_entry *entry; 2256 2257 while (n != NULL) { 2258 entry = rb_entry(n, struct nfs_access_entry, rb_node); 2259 2260 if (cred < entry->cred) 2261 n = n->rb_left; 2262 else if (cred > entry->cred) 2263 n = n->rb_right; 2264 else 2265 return entry; 2266 } 2267 return NULL; 2268 } 2269 2270 static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res, bool may_block) 2271 { 2272 struct nfs_inode *nfsi = NFS_I(inode); 2273 struct nfs_access_entry *cache; 2274 bool retry = true; 2275 int err; 2276 2277 spin_lock(&inode->i_lock); 2278 for(;;) { 2279 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS) 2280 goto out_zap; 2281 cache = nfs_access_search_rbtree(inode, cred); 2282 err = -ENOENT; 2283 if (cache == NULL) 2284 goto out; 2285 /* Found an entry, is our attribute cache valid? */ 2286 if (!nfs_attribute_cache_expired(inode) && 2287 !(nfsi->cache_validity & NFS_INO_INVALID_ATTR)) 2288 break; 2289 err = -ECHILD; 2290 if (!may_block) 2291 goto out; 2292 if (!retry) 2293 goto out_zap; 2294 spin_unlock(&inode->i_lock); 2295 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode); 2296 if (err) 2297 return err; 2298 spin_lock(&inode->i_lock); 2299 retry = false; 2300 } 2301 res->jiffies = cache->jiffies; 2302 res->cred = cache->cred; 2303 res->mask = cache->mask; 2304 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru); 2305 err = 0; 2306 out: 2307 spin_unlock(&inode->i_lock); 2308 return err; 2309 out_zap: 2310 spin_unlock(&inode->i_lock); 2311 nfs_access_zap_cache(inode); 2312 return -ENOENT; 2313 } 2314 2315 static int nfs_access_get_cached_rcu(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res) 2316 { 2317 /* Only check the most recently returned cache entry, 2318 * but do it without locking. 2319 */ 2320 struct nfs_inode *nfsi = NFS_I(inode); 2321 struct nfs_access_entry *cache; 2322 int err = -ECHILD; 2323 struct list_head *lh; 2324 2325 rcu_read_lock(); 2326 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS) 2327 goto out; 2328 lh = rcu_dereference(nfsi->access_cache_entry_lru.prev); 2329 cache = list_entry(lh, struct nfs_access_entry, lru); 2330 if (lh == &nfsi->access_cache_entry_lru || 2331 cred != cache->cred) 2332 cache = NULL; 2333 if (cache == NULL) 2334 goto out; 2335 err = nfs_revalidate_inode_rcu(NFS_SERVER(inode), inode); 2336 if (err) 2337 goto out; 2338 res->jiffies = cache->jiffies; 2339 res->cred = cache->cred; 2340 res->mask = cache->mask; 2341 out: 2342 rcu_read_unlock(); 2343 return err; 2344 } 2345 2346 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set) 2347 { 2348 struct nfs_inode *nfsi = NFS_I(inode); 2349 struct rb_root *root_node = &nfsi->access_cache; 2350 struct rb_node **p = &root_node->rb_node; 2351 struct rb_node *parent = NULL; 2352 struct nfs_access_entry *entry; 2353 2354 spin_lock(&inode->i_lock); 2355 while (*p != NULL) { 2356 parent = *p; 2357 entry = rb_entry(parent, struct nfs_access_entry, rb_node); 2358 2359 if (set->cred < entry->cred) 2360 p = &parent->rb_left; 2361 else if (set->cred > entry->cred) 2362 p = &parent->rb_right; 2363 else 2364 goto found; 2365 } 2366 rb_link_node(&set->rb_node, parent, p); 2367 rb_insert_color(&set->rb_node, root_node); 2368 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); 2369 spin_unlock(&inode->i_lock); 2370 return; 2371 found: 2372 rb_replace_node(parent, &set->rb_node, root_node); 2373 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); 2374 list_del(&entry->lru); 2375 spin_unlock(&inode->i_lock); 2376 nfs_access_free_entry(entry); 2377 } 2378 2379 void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set) 2380 { 2381 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL); 2382 if (cache == NULL) 2383 return; 2384 RB_CLEAR_NODE(&cache->rb_node); 2385 cache->jiffies = set->jiffies; 2386 cache->cred = get_rpccred(set->cred); 2387 cache->mask = set->mask; 2388 2389 /* The above field assignments must be visible 2390 * before this item appears on the lru. We cannot easily 2391 * use rcu_assign_pointer, so just force the memory barrier. 2392 */ 2393 smp_wmb(); 2394 nfs_access_add_rbtree(inode, cache); 2395 2396 /* Update accounting */ 2397 smp_mb__before_atomic(); 2398 atomic_long_inc(&nfs_access_nr_entries); 2399 smp_mb__after_atomic(); 2400 2401 /* Add inode to global LRU list */ 2402 if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) { 2403 spin_lock(&nfs_access_lru_lock); 2404 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) 2405 list_add_tail(&NFS_I(inode)->access_cache_inode_lru, 2406 &nfs_access_lru_list); 2407 spin_unlock(&nfs_access_lru_lock); 2408 } 2409 nfs_access_cache_enforce_limit(); 2410 } 2411 EXPORT_SYMBOL_GPL(nfs_access_add_cache); 2412 2413 void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result) 2414 { 2415 entry->mask = 0; 2416 if (access_result & NFS4_ACCESS_READ) 2417 entry->mask |= MAY_READ; 2418 if (access_result & 2419 (NFS4_ACCESS_MODIFY | NFS4_ACCESS_EXTEND | NFS4_ACCESS_DELETE)) 2420 entry->mask |= MAY_WRITE; 2421 if (access_result & (NFS4_ACCESS_LOOKUP|NFS4_ACCESS_EXECUTE)) 2422 entry->mask |= MAY_EXEC; 2423 } 2424 EXPORT_SYMBOL_GPL(nfs_access_set_mask); 2425 2426 static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask) 2427 { 2428 struct nfs_access_entry cache; 2429 bool may_block = (mask & MAY_NOT_BLOCK) == 0; 2430 int status; 2431 2432 trace_nfs_access_enter(inode); 2433 2434 status = nfs_access_get_cached_rcu(inode, cred, &cache); 2435 if (status != 0) 2436 status = nfs_access_get_cached(inode, cred, &cache, may_block); 2437 if (status == 0) 2438 goto out_cached; 2439 2440 status = -ECHILD; 2441 if (!may_block) 2442 goto out; 2443 2444 /* Be clever: ask server to check for all possible rights */ 2445 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ; 2446 cache.cred = cred; 2447 cache.jiffies = jiffies; 2448 status = NFS_PROTO(inode)->access(inode, &cache); 2449 if (status != 0) { 2450 if (status == -ESTALE) { 2451 nfs_zap_caches(inode); 2452 if (!S_ISDIR(inode->i_mode)) 2453 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags); 2454 } 2455 goto out; 2456 } 2457 nfs_access_add_cache(inode, &cache); 2458 out_cached: 2459 if ((mask & ~cache.mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) != 0) 2460 status = -EACCES; 2461 out: 2462 trace_nfs_access_exit(inode, status); 2463 return status; 2464 } 2465 2466 static int nfs_open_permission_mask(int openflags) 2467 { 2468 int mask = 0; 2469 2470 if (openflags & __FMODE_EXEC) { 2471 /* ONLY check exec rights */ 2472 mask = MAY_EXEC; 2473 } else { 2474 if ((openflags & O_ACCMODE) != O_WRONLY) 2475 mask |= MAY_READ; 2476 if ((openflags & O_ACCMODE) != O_RDONLY) 2477 mask |= MAY_WRITE; 2478 } 2479 2480 return mask; 2481 } 2482 2483 int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags) 2484 { 2485 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags)); 2486 } 2487 EXPORT_SYMBOL_GPL(nfs_may_open); 2488 2489 static int nfs_execute_ok(struct inode *inode, int mask) 2490 { 2491 struct nfs_server *server = NFS_SERVER(inode); 2492 int ret; 2493 2494 if (mask & MAY_NOT_BLOCK) 2495 ret = nfs_revalidate_inode_rcu(server, inode); 2496 else 2497 ret = nfs_revalidate_inode(server, inode); 2498 if (ret == 0 && !execute_ok(inode)) 2499 ret = -EACCES; 2500 return ret; 2501 } 2502 2503 int nfs_permission(struct inode *inode, int mask) 2504 { 2505 struct rpc_cred *cred; 2506 int res = 0; 2507 2508 nfs_inc_stats(inode, NFSIOS_VFSACCESS); 2509 2510 if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) 2511 goto out; 2512 /* Is this sys_access() ? */ 2513 if (mask & (MAY_ACCESS | MAY_CHDIR)) 2514 goto force_lookup; 2515 2516 switch (inode->i_mode & S_IFMT) { 2517 case S_IFLNK: 2518 goto out; 2519 case S_IFREG: 2520 if ((mask & MAY_OPEN) && 2521 nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN)) 2522 return 0; 2523 break; 2524 case S_IFDIR: 2525 /* 2526 * Optimize away all write operations, since the server 2527 * will check permissions when we perform the op. 2528 */ 2529 if ((mask & MAY_WRITE) && !(mask & MAY_READ)) 2530 goto out; 2531 } 2532 2533 force_lookup: 2534 if (!NFS_PROTO(inode)->access) 2535 goto out_notsup; 2536 2537 /* Always try fast lookups first */ 2538 rcu_read_lock(); 2539 cred = rpc_lookup_cred_nonblock(); 2540 if (!IS_ERR(cred)) 2541 res = nfs_do_access(inode, cred, mask|MAY_NOT_BLOCK); 2542 else 2543 res = PTR_ERR(cred); 2544 rcu_read_unlock(); 2545 if (res == -ECHILD && !(mask & MAY_NOT_BLOCK)) { 2546 /* Fast lookup failed, try the slow way */ 2547 cred = rpc_lookup_cred(); 2548 if (!IS_ERR(cred)) { 2549 res = nfs_do_access(inode, cred, mask); 2550 put_rpccred(cred); 2551 } else 2552 res = PTR_ERR(cred); 2553 } 2554 out: 2555 if (!res && (mask & MAY_EXEC)) 2556 res = nfs_execute_ok(inode, mask); 2557 2558 dfprintk(VFS, "NFS: permission(%s/%lu), mask=0x%x, res=%d\n", 2559 inode->i_sb->s_id, inode->i_ino, mask, res); 2560 return res; 2561 out_notsup: 2562 if (mask & MAY_NOT_BLOCK) 2563 return -ECHILD; 2564 2565 res = nfs_revalidate_inode(NFS_SERVER(inode), inode); 2566 if (res == 0) 2567 res = generic_permission(inode, mask); 2568 goto out; 2569 } 2570 EXPORT_SYMBOL_GPL(nfs_permission); 2571 2572 /* 2573 * Local variables: 2574 * version-control: t 2575 * kept-new-versions: 5 2576 * End: 2577 */ 2578