fs/nfs/dir.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  *  linux/fs/nfs/dir.c
   4  *
   5  *  Copyright (C) 1992  Rick Sladkey
   6  *
   7  *  nfs directory handling functions
   8  *
   9  * 10 Apr 1996  Added silly rename for unlink   --okir
  10  * 28 Sep 1996  Improved directory cache --okir
  11  * 23 Aug 1997  Claus Heine claus@momo.math.rwth-aachen.de
  12  *              Re-implemented silly rename for unlink, newly implemented
  13  *              silly rename for nfs_rename() following the suggestions
  14  *              of Olaf Kirch (okir) found in this file.
  15  *              Following Linus comments on my original hack, this version
  16  *              depends only on the dcache stuff and doesn't touch the inode
  17  *              layer (iput() and friends).
  18  *  6 Jun 1999  Cache readdir lookups in the page cache. -DaveM
  19  */
  20
  21 #include <linux/compat.h>
  22 #include <linux/module.h>
  23 #include <linux/time.h>
  24 #include <linux/errno.h>
  25 #include <linux/stat.h>
  26 #include <linux/fcntl.h>
  27 #include <linux/string.h>
  28 #include <linux/kernel.h>
  29 #include <linux/slab.h>
  30 #include <linux/mm.h>
  31 #include <linux/sunrpc/clnt.h>
  32 #include <linux/nfs_fs.h>
  33 #include <linux/nfs_mount.h>
  34 #include <linux/pagemap.h>
  35 #include <linux/pagevec.h>
  36 #include <linux/namei.h>
  37 #include <linux/mount.h>
  38 #include <linux/swap.h>
  39 #include <linux/sched.h>
  40 #include <linux/kmemleak.h>
  41 #include <linux/xattr.h>
  42 #include <linux/hash.h>
  43
  44 #include "delegation.h"
  45 #include "iostat.h"
  46 #include "internal.h"
  47 #include "fscache.h"
  48
  49 #include "nfstrace.h"
  50
  51 /* #define NFS_DEBUG_VERBOSE 1 */
  52
  53 static int nfs_opendir(struct inode *, struct file *);
  54 static int nfs_closedir(struct inode *, struct file *);
  55 static int nfs_readdir(struct file *, struct dir_context *);
  56 static int nfs_fsync_dir(struct file *, loff_t, loff_t, int);
  57 static loff_t nfs_llseek_dir(struct file *, loff_t, int);
  58 static void nfs_readdir_free_folio(struct folio *);
  59
  60 const struct file_operations nfs_dir_operations = {
  61         .llseek         = nfs_llseek_dir,
  62         .read           = generic_read_dir,
  63         .iterate_shared = nfs_readdir,
  64         .open           = nfs_opendir,
  65         .release        = nfs_closedir,
  66         .fsync          = nfs_fsync_dir,
  67 };
  68
  69 const struct address_space_operations nfs_dir_aops = {
  70         .free_folio = nfs_readdir_free_folio,
  71 };
  72
  73 #define NFS_INIT_DTSIZE PAGE_SIZE
  74
  75 static struct nfs_open_dir_context *
  76 alloc_nfs_open_dir_context(struct inode *dir)
  77 {
  78         struct nfs_inode *nfsi = NFS_I(dir);
  79         struct nfs_open_dir_context *ctx;
  80
  81         ctx = kzalloc(sizeof(*ctx), GFP_KERNEL_ACCOUNT);
  82         if (ctx != NULL) {
  83                 ctx->attr_gencount = nfsi->attr_gencount;
  84                 ctx->dtsize = NFS_INIT_DTSIZE;
  85                 spin_lock(&dir->i_lock);
  86                 if (list_empty(&nfsi->open_files) &&
  87                     (nfsi->cache_validity & NFS_INO_DATA_INVAL_DEFER))
  88                         nfs_set_cache_invalid(dir,
  89                                               NFS_INO_INVALID_DATA |
  90                                                       NFS_INO_REVAL_FORCED);
  91                 list_add_tail_rcu(&ctx->list, &nfsi->open_files);
  92                 memcpy(ctx->verf, nfsi->cookieverf, sizeof(ctx->verf));
  93                 spin_unlock(&dir->i_lock);
  94                 return ctx;
  95         }
  96         return  ERR_PTR(-ENOMEM);
  97 }
  98
  99 static void put_nfs_open_dir_context(struct inode *dir, struct nfs_open_dir_context *ctx)
 100 {
 101         spin_lock(&dir->i_lock);
 102         list_del_rcu(&ctx->list);
 103         spin_unlock(&dir->i_lock);
 104         kfree_rcu(ctx, rcu_head);
 105 }
 106
 107 /*
 108  * Open file
 109  */
 110 static int
 111 nfs_opendir(struct inode *inode, struct file *filp)
 112 {
 113         int res = 0;
 114         struct nfs_open_dir_context *ctx;
 115
 116         dfprintk(FILE, "NFS: open dir(%pD2)\n", filp);
 117
 118         nfs_inc_stats(inode, NFSIOS_VFSOPEN);
 119
 120         ctx = alloc_nfs_open_dir_context(inode);
 121         if (IS_ERR(ctx)) {
 122                 res = PTR_ERR(ctx);
 123                 goto out;
 124         }
 125         filp->private_data = ctx;
 126 out:
 127         return res;
 128 }
 129
 130 static int
 131 nfs_closedir(struct inode *inode, struct file *filp)
 132 {
 133         put_nfs_open_dir_context(file_inode(filp), filp->private_data);
 134         return 0;
 135 }
 136
 137 struct nfs_cache_array_entry {
 138         u64 cookie;
 139         u64 ino;
 140         const char *name;
 141         unsigned int name_len;
 142         unsigned char d_type;
 143 };
 144
 145 struct nfs_cache_array {
 146         u64 change_attr;
 147         u64 last_cookie;
 148         unsigned int size;
 149         unsigned char page_full : 1,
 150                       page_is_eof : 1,
 151                       cookies_are_ordered : 1;
 152         struct nfs_cache_array_entry array[];
 153 };
 154
 155 struct nfs_readdir_descriptor {
 156         struct file     *file;
 157         struct page     *page;
 158         struct dir_context *ctx;
 159         pgoff_t         page_index;
 160         pgoff_t         page_index_max;
 161         u64             dir_cookie;
 162         u64             last_cookie;
 163         loff_t          current_index;
 164
 165         __be32          verf[NFS_DIR_VERIFIER_SIZE];
 166         unsigned long   dir_verifier;
 167         unsigned long   timestamp;
 168         unsigned long   gencount;
 169         unsigned long   attr_gencount;
 170         unsigned int    cache_entry_index;
 171         unsigned int    buffer_fills;
 172         unsigned int    dtsize;
 173         bool clear_cache;
 174         bool plus;
 175         bool eob;
 176         bool eof;
 177 };
 178
 179 static void nfs_set_dtsize(struct nfs_readdir_descriptor *desc, unsigned int sz)
 180 {
 181         struct nfs_server *server = NFS_SERVER(file_inode(desc->file));
 182         unsigned int maxsize = server->dtsize;
 183
 184         if (sz > maxsize)
 185                 sz = maxsize;
 186         if (sz < NFS_MIN_FILE_IO_SIZE)
 187                 sz = NFS_MIN_FILE_IO_SIZE;
 188         desc->dtsize = sz;
 189 }
 190
 191 static void nfs_shrink_dtsize(struct nfs_readdir_descriptor *desc)
 192 {
 193         nfs_set_dtsize(desc, desc->dtsize >> 1);
 194 }
 195
 196 static void nfs_grow_dtsize(struct nfs_readdir_descriptor *desc)
 197 {
 198         nfs_set_dtsize(desc, desc->dtsize << 1);
 199 }
 200
 201 static void nfs_readdir_page_init_array(struct page *page, u64 last_cookie,
 202                                         u64 change_attr)
 203 {
 204         struct nfs_cache_array *array;
 205
 206         array = kmap_atomic(page);
 207         array->change_attr = change_attr;
 208         array->last_cookie = last_cookie;
 209         array->size = 0;
 210         array->page_full = 0;
 211         array->page_is_eof = 0;
 212         array->cookies_are_ordered = 1;
 213         kunmap_atomic(array);
 214 }
 215
 216 /*
 217  * we are freeing strings created by nfs_add_to_readdir_array()
 218  */
 219 static void nfs_readdir_clear_array(struct page *page)
 220 {
 221         struct nfs_cache_array *array;
 222         unsigned int i;
 223
 224         array = kmap_atomic(page);
 225         for (i = 0; i < array->size; i++)
 226                 kfree(array->array[i].name);
 227         array->size = 0;
 228         kunmap_atomic(array);
 229 }
 230
 231 static void nfs_readdir_free_folio(struct folio *folio)
 232 {
 233         nfs_readdir_clear_array(&folio->page);
 234 }
 235
 236 static void nfs_readdir_page_reinit_array(struct page *page, u64 last_cookie,
 237                                           u64 change_attr)
 238 {
 239         nfs_readdir_clear_array(page);
 240         nfs_readdir_page_init_array(page, last_cookie, change_attr);
 241 }
 242
 243 static struct page *
 244 nfs_readdir_page_array_alloc(u64 last_cookie, gfp_t gfp_flags)
 245 {
 246         struct page *page = alloc_page(gfp_flags);
 247         if (page)
 248                 nfs_readdir_page_init_array(page, last_cookie, 0);
 249         return page;
 250 }
 251
 252 static void nfs_readdir_page_array_free(struct page *page)
 253 {
 254         if (page) {
 255                 nfs_readdir_clear_array(page);
 256                 put_page(page);
 257         }
 258 }
 259
 260 static u64 nfs_readdir_array_index_cookie(struct nfs_cache_array *array)
 261 {
 262         return array->size == 0 ? array->last_cookie : array->array[0].cookie;
 263 }
 264
 265 static void nfs_readdir_array_set_eof(struct nfs_cache_array *array)
 266 {
 267         array->page_is_eof = 1;
 268         array->page_full = 1;
 269 }
 270
 271 static bool nfs_readdir_array_is_full(struct nfs_cache_array *array)
 272 {
 273         return array->page_full;
 274 }
 275
 276 /*
 277  * the caller is responsible for freeing qstr.name
 278  * when called by nfs_readdir_add_to_array, the strings will be freed in
 279  * nfs_clear_readdir_array()
 280  */
 281 static const char *nfs_readdir_copy_name(const char *name, unsigned int len)
 282 {
 283         const char *ret = kmemdup_nul(name, len, GFP_KERNEL);
 284
 285         /*
 286          * Avoid a kmemleak false positive. The pointer to the name is stored
 287          * in a page cache page which kmemleak does not scan.
 288          */
 289         if (ret != NULL)
 290                 kmemleak_not_leak(ret);
 291         return ret;
 292 }
 293
 294 static size_t nfs_readdir_array_maxentries(void)
 295 {
 296         return (PAGE_SIZE - sizeof(struct nfs_cache_array)) /
 297                sizeof(struct nfs_cache_array_entry);
 298 }
 299
 300 /*
 301  * Check that the next array entry lies entirely within the page bounds
 302  */
 303 static int nfs_readdir_array_can_expand(struct nfs_cache_array *array)
 304 {
 305         if (array->page_full)
 306                 return -ENOSPC;
 307         if (array->size == nfs_readdir_array_maxentries()) {
 308                 array->page_full = 1;
 309                 return -ENOSPC;
 310         }
 311         return 0;
 312 }
 313
 314 static int nfs_readdir_page_array_append(struct page *page,
 315                                          const struct nfs_entry *entry,
 316                                          u64 *cookie)
 317 {
 318         struct nfs_cache_array *array;
 319         struct nfs_cache_array_entry *cache_entry;
 320         const char *name;
 321         int ret = -ENOMEM;
 322
 323         name = nfs_readdir_copy_name(entry->name, entry->len);
 324
 325         array = kmap_atomic(page);
 326         if (!name)
 327                 goto out;
 328         ret = nfs_readdir_array_can_expand(array);
 329         if (ret) {
 330                 kfree(name);
 331                 goto out;
 332         }
 333
 334         cache_entry = &array->array[array->size];
 335         cache_entry->cookie = array->last_cookie;
 336         cache_entry->ino = entry->ino;
 337         cache_entry->d_type = entry->d_type;
 338         cache_entry->name_len = entry->len;
 339         cache_entry->name = name;
 340         array->last_cookie = entry->cookie;
 341         if (array->last_cookie <= cache_entry->cookie)
 342                 array->cookies_are_ordered = 0;
 343         array->size++;
 344         if (entry->eof != 0)
 345                 nfs_readdir_array_set_eof(array);
 346 out:
 347         *cookie = array->last_cookie;
 348         kunmap_atomic(array);
 349         return ret;
 350 }
 351
 352 #define NFS_READDIR_COOKIE_MASK (U32_MAX >> 14)
 353 /*
 354  * Hash algorithm allowing content addressible access to sequences
 355  * of directory cookies. Content is addressed by the value of the
 356  * cookie index of the first readdir entry in a page.
 357  *
 358  * We select only the first 18 bits to avoid issues with excessive
 359  * memory use for the page cache XArray. 18 bits should allow the caching
 360  * of 262144 pages of sequences of readdir entries. Since each page holds
 361  * 127 readdir entries for a typical 64-bit system, that works out to a
 362  * cache of ~ 33 million entries per directory.
 363  */
 364 static pgoff_t nfs_readdir_page_cookie_hash(u64 cookie)
 365 {
 366         if (cookie == 0)
 367                 return 0;
 368         return hash_64(cookie, 18);
 369 }
 370
 371 static bool nfs_readdir_page_validate(struct page *page, u64 last_cookie,
 372                                       u64 change_attr)
 373 {
 374         struct nfs_cache_array *array = kmap_atomic(page);
 375         int ret = true;
 376
 377         if (array->change_attr != change_attr)
 378                 ret = false;
 379         if (nfs_readdir_array_index_cookie(array) != last_cookie)
 380                 ret = false;
 381         kunmap_atomic(array);
 382         return ret;
 383 }
 384
 385 static void nfs_readdir_page_unlock_and_put(struct page *page)
 386 {
 387         unlock_page(page);
 388         put_page(page);
 389 }
 390
 391 static void nfs_readdir_page_init_and_validate(struct page *page, u64 cookie,
 392                                                u64 change_attr)
 393 {
 394         if (PageUptodate(page)) {
 395                 if (nfs_readdir_page_validate(page, cookie, change_attr))
 396                         return;
 397                 nfs_readdir_clear_array(page);
 398         }
 399         nfs_readdir_page_init_array(page, cookie, change_attr);
 400         SetPageUptodate(page);
 401 }
 402
 403 static struct page *nfs_readdir_page_get_locked(struct address_space *mapping,
 404                                                 u64 cookie, u64 change_attr)
 405 {
 406         pgoff_t index = nfs_readdir_page_cookie_hash(cookie);
 407         struct page *page;
 408
 409         page = grab_cache_page(mapping, index);
 410         if (!page)
 411                 return NULL;
 412         nfs_readdir_page_init_and_validate(page, cookie, change_attr);
 413         return page;
 414 }
 415
 416 static u64 nfs_readdir_page_last_cookie(struct page *page)
 417 {
 418         struct nfs_cache_array *array;
 419         u64 ret;
 420
 421         array = kmap_atomic(page);
 422         ret = array->last_cookie;
 423         kunmap_atomic(array);
 424         return ret;
 425 }
 426
 427 static bool nfs_readdir_page_needs_filling(struct page *page)
 428 {
 429         struct nfs_cache_array *array;
 430         bool ret;
 431
 432         array = kmap_atomic(page);
 433         ret = !nfs_readdir_array_is_full(array);
 434         kunmap_atomic(array);
 435         return ret;
 436 }
 437
 438 static void nfs_readdir_page_set_eof(struct page *page)
 439 {
 440         struct nfs_cache_array *array;
 441
 442         array = kmap_atomic(page);
 443         nfs_readdir_array_set_eof(array);
 444         kunmap_atomic(array);
 445 }
 446
 447 static struct page *nfs_readdir_page_get_next(struct address_space *mapping,
 448                                               u64 cookie, u64 change_attr)
 449 {
 450         pgoff_t index = nfs_readdir_page_cookie_hash(cookie);
 451         struct page *page;
 452
 453         page = grab_cache_page_nowait(mapping, index);
 454         if (!page)
 455                 return NULL;
 456         nfs_readdir_page_init_and_validate(page, cookie, change_attr);
 457         if (nfs_readdir_page_last_cookie(page) != cookie)
 458                 nfs_readdir_page_reinit_array(page, cookie, change_attr);
 459         return page;
 460 }
 461
 462 static inline
 463 int is_32bit_api(void)
 464 {
 465 #ifdef CONFIG_COMPAT
 466         return in_compat_syscall();
 467 #else
 468         return (BITS_PER_LONG == 32);
 469 #endif
 470 }
 471
 472 static
 473 bool nfs_readdir_use_cookie(const struct file *filp)
 474 {
 475         if ((filp->f_mode & FMODE_32BITHASH) ||
 476             (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
 477                 return false;
 478         return true;
 479 }
 480
 481 static void nfs_readdir_seek_next_array(struct nfs_cache_array *array,
 482                                         struct nfs_readdir_descriptor *desc)
 483 {
 484         if (array->page_full) {
 485                 desc->last_cookie = array->last_cookie;
 486                 desc->current_index += array->size;
 487                 desc->cache_entry_index = 0;
 488                 desc->page_index++;
 489         } else
 490                 desc->last_cookie = nfs_readdir_array_index_cookie(array);
 491 }
 492
 493 static void nfs_readdir_rewind_search(struct nfs_readdir_descriptor *desc)
 494 {
 495         desc->current_index = 0;
 496         desc->last_cookie = 0;
 497         desc->page_index = 0;
 498 }
 499
 500 static int nfs_readdir_search_for_pos(struct nfs_cache_array *array,
 501                                       struct nfs_readdir_descriptor *desc)
 502 {
 503         loff_t diff = desc->ctx->pos - desc->current_index;
 504         unsigned int index;
 505
 506         if (diff < 0)
 507                 goto out_eof;
 508         if (diff >= array->size) {
 509                 if (array->page_is_eof)
 510                         goto out_eof;
 511                 nfs_readdir_seek_next_array(array, desc);
 512                 return -EAGAIN;
 513         }
 514
 515         index = (unsigned int)diff;
 516         desc->dir_cookie = array->array[index].cookie;
 517         desc->cache_entry_index = index;
 518         return 0;
 519 out_eof:
 520         desc->eof = true;
 521         return -EBADCOOKIE;
 522 }
 523
 524 static bool nfs_readdir_array_cookie_in_range(struct nfs_cache_array *array,
 525                                               u64 cookie)
 526 {
 527         if (!array->cookies_are_ordered)
 528                 return true;
 529         /* Optimisation for monotonically increasing cookies */
 530         if (cookie >= array->last_cookie)
 531                 return false;
 532         if (array->size && cookie < array->array[0].cookie)
 533                 return false;
 534         return true;
 535 }
 536
 537 static int nfs_readdir_search_for_cookie(struct nfs_cache_array *array,
 538                                          struct nfs_readdir_descriptor *desc)
 539 {
 540         unsigned int i;
 541         int status = -EAGAIN;
 542
 543         if (!nfs_readdir_array_cookie_in_range(array, desc->dir_cookie))
 544                 goto check_eof;
 545
 546         for (i = 0; i < array->size; i++) {
 547                 if (array->array[i].cookie == desc->dir_cookie) {
 548                         if (nfs_readdir_use_cookie(desc->file))
 549                                 desc->ctx->pos = desc->dir_cookie;
 550                         else
 551                                 desc->ctx->pos = desc->current_index + i;
 552                         desc->cache_entry_index = i;
 553                         return 0;
 554                 }
 555         }
 556 check_eof:
 557         if (array->page_is_eof) {
 558                 status = -EBADCOOKIE;
 559                 if (desc->dir_cookie == array->last_cookie)
 560                         desc->eof = true;
 561         } else
 562                 nfs_readdir_seek_next_array(array, desc);
 563         return status;
 564 }
 565
 566 static int nfs_readdir_search_array(struct nfs_readdir_descriptor *desc)
 567 {
 568         struct nfs_cache_array *array;
 569         int status;
 570
 571         array = kmap_atomic(desc->page);
 572
 573         if (desc->dir_cookie == 0)
 574                 status = nfs_readdir_search_for_pos(array, desc);
 575         else
 576                 status = nfs_readdir_search_for_cookie(array, desc);
 577
 578         kunmap_atomic(array);
 579         return status;
 580 }
 581
 582 /* Fill a page with xdr information before transferring to the cache page */
 583 static int nfs_readdir_xdr_filler(struct nfs_readdir_descriptor *desc,
 584                                   __be32 *verf, u64 cookie,
 585                                   struct page **pages, size_t bufsize,
 586                                   __be32 *verf_res)
 587 {
 588         struct inode *inode = file_inode(desc->file);
 589         struct nfs_readdir_arg arg = {
 590                 .dentry = file_dentry(desc->file),
 591                 .cred = desc->file->f_cred,
 592                 .verf = verf,
 593                 .cookie = cookie,
 594                 .pages = pages,
 595                 .page_len = bufsize,
 596                 .plus = desc->plus,
 597         };
 598         struct nfs_readdir_res res = {
 599                 .verf = verf_res,
 600         };
 601         unsigned long   timestamp, gencount;
 602         int             error;
 603
 604  again:
 605         timestamp = jiffies;
 606         gencount = nfs_inc_attr_generation_counter();
 607         desc->dir_verifier = nfs_save_change_attribute(inode);
 608         error = NFS_PROTO(inode)->readdir(&arg, &res);
 609         if (error < 0) {
 610                 /* We requested READDIRPLUS, but the server doesn't grok it */
 611                 if (error == -ENOTSUPP && desc->plus) {
 612                         NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
 613                         desc->plus = arg.plus = false;
 614                         goto again;
 615                 }
 616                 goto error;
 617         }
 618         desc->timestamp = timestamp;
 619         desc->gencount = gencount;
 620 error:
 621         return error;
 622 }
 623
 624 static int xdr_decode(struct nfs_readdir_descriptor *desc,
 625                       struct nfs_entry *entry, struct xdr_stream *xdr)
 626 {
 627         struct inode *inode = file_inode(desc->file);
 628         int error;
 629
 630         error = NFS_PROTO(inode)->decode_dirent(xdr, entry, desc->plus);
 631         if (error)
 632                 return error;
 633         entry->fattr->time_start = desc->timestamp;
 634         entry->fattr->gencount = desc->gencount;
 635         return 0;
 636 }
 637
 638 /* Match file and dirent using either filehandle or fileid
 639  * Note: caller is responsible for checking the fsid
 640  */
 641 static
 642 int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry)
 643 {
 644         struct inode *inode;
 645         struct nfs_inode *nfsi;
 646
 647         if (d_really_is_negative(dentry))
 648                 return 0;
 649
 650         inode = d_inode(dentry);
 651         if (is_bad_inode(inode) || NFS_STALE(inode))
 652                 return 0;
 653
 654         nfsi = NFS_I(inode);
 655         if (entry->fattr->fileid != nfsi->fileid)
 656                 return 0;
 657         if (entry->fh->size && nfs_compare_fh(entry->fh, &nfsi->fh) != 0)
 658                 return 0;
 659         return 1;
 660 }
 661
 662 #define NFS_READDIR_CACHE_USAGE_THRESHOLD (8UL)
 663
 664 static bool nfs_use_readdirplus(struct inode *dir, struct dir_context *ctx,
 665                                 unsigned int cache_hits,
 666                                 unsigned int cache_misses)
 667 {
 668         if (!nfs_server_capable(dir, NFS_CAP_READDIRPLUS))
 669                 return false;
 670         if (ctx->pos == 0 ||
 671             cache_hits + cache_misses > NFS_READDIR_CACHE_USAGE_THRESHOLD)
 672                 return true;
 673         return false;
 674 }
 675
 676 /*
 677  * This function is called by the getattr code to request the
 678  * use of readdirplus to accelerate any future lookups in the same
 679  * directory.
 680  */
 681 void nfs_readdir_record_entry_cache_hit(struct inode *dir)
 682 {
 683         struct nfs_inode *nfsi = NFS_I(dir);
 684         struct nfs_open_dir_context *ctx;
 685
 686         if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) &&
 687             S_ISDIR(dir->i_mode)) {
 688                 rcu_read_lock();
 689                 list_for_each_entry_rcu (ctx, &nfsi->open_files, list)
 690                         atomic_inc(&ctx->cache_hits);
 691                 rcu_read_unlock();
 692         }
 693 }
 694
 695 /*
 696  * This function is mainly for use by nfs_getattr().
 697  *
 698  * If this is an 'ls -l', we want to force use of readdirplus.
 699  */
 700 void nfs_readdir_record_entry_cache_miss(struct inode *dir)
 701 {
 702         struct nfs_inode *nfsi = NFS_I(dir);
 703         struct nfs_open_dir_context *ctx;
 704
 705         if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) &&
 706             S_ISDIR(dir->i_mode)) {
 707                 rcu_read_lock();
 708                 list_for_each_entry_rcu (ctx, &nfsi->open_files, list)
 709                         atomic_inc(&ctx->cache_misses);
 710                 rcu_read_unlock();
 711         }
 712 }
 713
 714 static void nfs_lookup_advise_force_readdirplus(struct inode *dir,
 715                                                 unsigned int flags)
 716 {
 717         if (nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE))
 718                 return;
 719         if (flags & (LOOKUP_EXCL | LOOKUP_PARENT | LOOKUP_REVAL))
 720                 return;
 721         nfs_readdir_record_entry_cache_miss(dir);
 722 }
 723
 724 static
 725 void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry,
 726                 unsigned long dir_verifier)
 727 {
 728         struct qstr filename = QSTR_INIT(entry->name, entry->len);
 729         DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
 730         struct dentry *dentry;
 731         struct dentry *alias;
 732         struct inode *inode;
 733         int status;
 734
 735         if (!(entry->fattr->valid & NFS_ATTR_FATTR_FILEID))
 736                 return;
 737         if (!(entry->fattr->valid & NFS_ATTR_FATTR_FSID))
 738                 return;
 739         if (filename.len == 0)
 740                 return;
 741         /* Validate that the name doesn't contain any illegal '\0' */
 742         if (strnlen(filename.name, filename.len) != filename.len)
 743                 return;
 744         /* ...or '/' */
 745         if (strnchr(filename.name, filename.len, '/'))
 746                 return;
 747         if (filename.name[0] == '.') {
 748                 if (filename.len == 1)
 749                         return;
 750                 if (filename.len == 2 && filename.name[1] == '.')
 751                         return;
 752         }
 753         filename.hash = full_name_hash(parent, filename.name, filename.len);
 754
 755         dentry = d_lookup(parent, &filename);
 756 again:
 757         if (!dentry) {
 758                 dentry = d_alloc_parallel(parent, &filename, &wq);
 759                 if (IS_ERR(dentry))
 760                         return;
 761         }
 762         if (!d_in_lookup(dentry)) {
 763                 /* Is there a mountpoint here? If so, just exit */
 764                 if (!nfs_fsid_equal(&NFS_SB(dentry->d_sb)->fsid,
 765                                         &entry->fattr->fsid))
 766                         goto out;
 767                 if (nfs_same_file(dentry, entry)) {
 768                         if (!entry->fh->size)
 769                                 goto out;
 770                         nfs_set_verifier(dentry, dir_verifier);
 771                         status = nfs_refresh_inode(d_inode(dentry), entry->fattr);
 772                         if (!status)
 773                                 nfs_setsecurity(d_inode(dentry), entry->fattr);
 774                         trace_nfs_readdir_lookup_revalidate(d_inode(parent),
 775                                                             dentry, 0, status);
 776                         goto out;
 777                 } else {
 778                         trace_nfs_readdir_lookup_revalidate_failed(
 779                                 d_inode(parent), dentry, 0);
 780                         d_invalidate(dentry);
 781                         dput(dentry);
 782                         dentry = NULL;
 783                         goto again;
 784                 }
 785         }
 786         if (!entry->fh->size) {
 787                 d_lookup_done(dentry);
 788                 goto out;
 789         }
 790
 791         inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr);
 792         alias = d_splice_alias(inode, dentry);
 793         d_lookup_done(dentry);
 794         if (alias) {
 795                 if (IS_ERR(alias))
 796                         goto out;
 797                 dput(dentry);
 798                 dentry = alias;
 799         }
 800         nfs_set_verifier(dentry, dir_verifier);
 801         trace_nfs_readdir_lookup(d_inode(parent), dentry, 0);
 802 out:
 803         dput(dentry);
 804 }
 805
 806 static int nfs_readdir_entry_decode(struct nfs_readdir_descriptor *desc,
 807                                     struct nfs_entry *entry,
 808                                     struct xdr_stream *stream)
 809 {
 810         int ret;
 811
 812         if (entry->fattr->label)
 813                 entry->fattr->label->len = NFS4_MAXLABELLEN;
 814         ret = xdr_decode(desc, entry, stream);
 815         if (ret || !desc->plus)
 816                 return ret;
 817         nfs_prime_dcache(file_dentry(desc->file), entry, desc->dir_verifier);
 818         return 0;
 819 }
 820
 821 /* Perform conversion from xdr to cache array */
 822 static int nfs_readdir_page_filler(struct nfs_readdir_descriptor *desc,
 823                                    struct nfs_entry *entry,
 824                                    struct page **xdr_pages, unsigned int buflen,
 825                                    struct page **arrays, size_t narrays,
 826                                    u64 change_attr)
 827 {
 828         struct address_space *mapping = desc->file->f_mapping;
 829         struct xdr_stream stream;
 830         struct xdr_buf buf;
 831         struct page *scratch, *new, *page = *arrays;
 832         u64 cookie;
 833         int status;
 834
 835         scratch = alloc_page(GFP_KERNEL);
 836         if (scratch == NULL)
 837                 return -ENOMEM;
 838
 839         xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen);
 840         xdr_set_scratch_page(&stream, scratch);
 841
 842         do {
 843                 status = nfs_readdir_entry_decode(desc, entry, &stream);
 844                 if (status != 0)
 845                         break;
 846
 847                 status = nfs_readdir_page_array_append(page, entry, &cookie);
 848                 if (status != -ENOSPC)
 849                         continue;
 850
 851                 if (page->mapping != mapping) {
 852                         if (!--narrays)
 853                                 break;
 854                         new = nfs_readdir_page_array_alloc(cookie, GFP_KERNEL);
 855                         if (!new)
 856                                 break;
 857                         arrays++;
 858                         *arrays = page = new;
 859                 } else {
 860                         new = nfs_readdir_page_get_next(mapping, cookie,
 861                                                         change_attr);
 862                         if (!new)
 863                                 break;
 864                         if (page != *arrays)
 865                                 nfs_readdir_page_unlock_and_put(page);
 866                         page = new;
 867                 }
 868                 desc->page_index_max++;
 869                 status = nfs_readdir_page_array_append(page, entry, &cookie);
 870         } while (!status && !entry->eof);
 871
 872         switch (status) {
 873         case -EBADCOOKIE:
 874                 if (!entry->eof)
 875                         break;
 876                 nfs_readdir_page_set_eof(page);
 877                 fallthrough;
 878         case -EAGAIN:
 879                 status = 0;
 880                 break;
 881         case -ENOSPC:
 882                 status = 0;
 883                 if (!desc->plus)
 884                         break;
 885                 while (!nfs_readdir_entry_decode(desc, entry, &stream))
 886                         ;
 887         }
 888
 889         if (page != *arrays)
 890                 nfs_readdir_page_unlock_and_put(page);
 891
 892         put_page(scratch);
 893         return status;
 894 }
 895
 896 static void nfs_readdir_free_pages(struct page **pages, size_t npages)
 897 {
 898         while (npages--)
 899                 put_page(pages[npages]);
 900         kfree(pages);
 901 }
 902
 903 /*
 904  * nfs_readdir_alloc_pages() will allocate pages that must be freed with a call
 905  * to nfs_readdir_free_pages()
 906  */
 907 static struct page **nfs_readdir_alloc_pages(size_t npages)
 908 {
 909         struct page **pages;
 910         size_t i;
 911
 912         pages = kmalloc_array(npages, sizeof(*pages), GFP_KERNEL);
 913         if (!pages)
 914                 return NULL;
 915         for (i = 0; i < npages; i++) {
 916                 struct page *page = alloc_page(GFP_KERNEL);
 917                 if (page == NULL)
 918                         goto out_freepages;
 919                 pages[i] = page;
 920         }
 921         return pages;
 922
 923 out_freepages:
 924         nfs_readdir_free_pages(pages, i);
 925         return NULL;
 926 }
 927
 928 static int nfs_readdir_xdr_to_array(struct nfs_readdir_descriptor *desc,
 929                                     __be32 *verf_arg, __be32 *verf_res,
 930                                     struct page **arrays, size_t narrays)
 931 {
 932         u64 change_attr;
 933         struct page **pages;
 934         struct page *page = *arrays;
 935         struct nfs_entry *entry;
 936         size_t array_size;
 937         struct inode *inode = file_inode(desc->file);
 938         unsigned int dtsize = desc->dtsize;
 939         unsigned int pglen;
 940         int status = -ENOMEM;
 941
 942         entry = kzalloc(sizeof(*entry), GFP_KERNEL);
 943         if (!entry)
 944                 return -ENOMEM;
 945         entry->cookie = nfs_readdir_page_last_cookie(page);
 946         entry->fh = nfs_alloc_fhandle();
 947         entry->fattr = nfs_alloc_fattr_with_label(NFS_SERVER(inode));
 948         entry->server = NFS_SERVER(inode);
 949         if (entry->fh == NULL || entry->fattr == NULL)
 950                 goto out;
 951
 952         array_size = (dtsize + PAGE_SIZE - 1) >> PAGE_SHIFT;
 953         pages = nfs_readdir_alloc_pages(array_size);
 954         if (!pages)
 955                 goto out;
 956
 957         change_attr = inode_peek_iversion_raw(inode);
 958         status = nfs_readdir_xdr_filler(desc, verf_arg, entry->cookie, pages,
 959                                         dtsize, verf_res);
 960         if (status < 0)
 961                 goto free_pages;
 962
 963         pglen = status;
 964         if (pglen != 0)
 965                 status = nfs_readdir_page_filler(desc, entry, pages, pglen,
 966                                                  arrays, narrays, change_attr);
 967         else
 968                 nfs_readdir_page_set_eof(page);
 969         desc->buffer_fills++;
 970
 971 free_pages:
 972         nfs_readdir_free_pages(pages, array_size);
 973 out:
 974         nfs_free_fattr(entry->fattr);
 975         nfs_free_fhandle(entry->fh);
 976         kfree(entry);
 977         return status;
 978 }
 979
 980 static void nfs_readdir_page_put(struct nfs_readdir_descriptor *desc)
 981 {
 982         put_page(desc->page);
 983         desc->page = NULL;
 984 }
 985
 986 static void
 987 nfs_readdir_page_unlock_and_put_cached(struct nfs_readdir_descriptor *desc)
 988 {
 989         unlock_page(desc->page);
 990         nfs_readdir_page_put(desc);
 991 }
 992
 993 static struct page *
 994 nfs_readdir_page_get_cached(struct nfs_readdir_descriptor *desc)
 995 {
 996         struct address_space *mapping = desc->file->f_mapping;
 997         u64 change_attr = inode_peek_iversion_raw(mapping->host);
 998         u64 cookie = desc->last_cookie;
 999         struct page *page;
1000
1001         page = nfs_readdir_page_get_locked(mapping, cookie, change_attr);
1002         if (!page)
1003                 return NULL;
1004         if (desc->clear_cache && !nfs_readdir_page_needs_filling(page))
1005                 nfs_readdir_page_reinit_array(page, cookie, change_attr);
1006         return page;
1007 }
1008
1009 /*
1010  * Returns 0 if desc->dir_cookie was found on page desc->page_index
1011  * and locks the page to prevent removal from the page cache.
1012  */
1013 static int find_and_lock_cache_page(struct nfs_readdir_descriptor *desc)
1014 {
1015         struct inode *inode = file_inode(desc->file);
1016         struct nfs_inode *nfsi = NFS_I(inode);
1017         __be32 verf[NFS_DIR_VERIFIER_SIZE];
1018         int res;
1019
1020         desc->page = nfs_readdir_page_get_cached(desc);
1021         if (!desc->page)
1022                 return -ENOMEM;
1023         if (nfs_readdir_page_needs_filling(desc->page)) {
1024                 /* Grow the dtsize if we had to go back for more pages */
1025                 if (desc->page_index == desc->page_index_max)
1026                         nfs_grow_dtsize(desc);
1027                 desc->page_index_max = desc->page_index;
1028                 trace_nfs_readdir_cache_fill(desc->file, nfsi->cookieverf,
1029                                              desc->last_cookie,
1030                                              desc->page->index, desc->dtsize);
1031                 res = nfs_readdir_xdr_to_array(desc, nfsi->cookieverf, verf,
1032                                                &desc->page, 1);
1033                 if (res < 0) {
1034                         nfs_readdir_page_unlock_and_put_cached(desc);
1035                         trace_nfs_readdir_cache_fill_done(inode, res);
1036                         if (res == -EBADCOOKIE || res == -ENOTSYNC) {
1037                                 invalidate_inode_pages2(desc->file->f_mapping);
1038                                 nfs_readdir_rewind_search(desc);
1039                                 trace_nfs_readdir_invalidate_cache_range(
1040                                         inode, 0, MAX_LFS_FILESIZE);
1041                                 return -EAGAIN;
1042                         }
1043                         return res;
1044                 }
1045                 /*
1046                  * Set the cookie verifier if the page cache was empty
1047                  */
1048                 if (desc->last_cookie == 0 &&
1049                     memcmp(nfsi->cookieverf, verf, sizeof(nfsi->cookieverf))) {
1050                         memcpy(nfsi->cookieverf, verf,
1051                                sizeof(nfsi->cookieverf));
1052                         invalidate_inode_pages2_range(desc->file->f_mapping, 1,
1053                                                       -1);
1054                         trace_nfs_readdir_invalidate_cache_range(
1055                                 inode, 1, MAX_LFS_FILESIZE);
1056                 }
1057                 desc->clear_cache = false;
1058         }
1059         res = nfs_readdir_search_array(desc);
1060         if (res == 0)
1061                 return 0;
1062         nfs_readdir_page_unlock_and_put_cached(desc);
1063         return res;
1064 }
1065
1066 /* Search for desc->dir_cookie from the beginning of the page cache */
1067 static int readdir_search_pagecache(struct nfs_readdir_descriptor *desc)
1068 {
1069         int res;
1070
1071         do {
1072                 res = find_and_lock_cache_page(desc);
1073         } while (res == -EAGAIN);
1074         return res;
1075 }
1076
1077 /*
1078  * Once we've found the start of the dirent within a page: fill 'er up...
1079  */
1080 static void nfs_do_filldir(struct nfs_readdir_descriptor *desc,
1081                            const __be32 *verf)
1082 {
1083         struct file     *file = desc->file;
1084         struct nfs_cache_array *array;
1085         unsigned int i;
1086
1087         array = kmap_local_page(desc->page);
1088         for (i = desc->cache_entry_index; i < array->size; i++) {
1089                 struct nfs_cache_array_entry *ent;
1090
1091                 ent = &array->array[i];
1092                 if (!dir_emit(desc->ctx, ent->name, ent->name_len,
1093                     nfs_compat_user_ino64(ent->ino), ent->d_type)) {
1094                         desc->eob = true;
1095                         break;
1096                 }
1097                 memcpy(desc->verf, verf, sizeof(desc->verf));
1098                 if (i == array->size - 1) {
1099                         desc->dir_cookie = array->last_cookie;
1100                         nfs_readdir_seek_next_array(array, desc);
1101                 } else {
1102                         desc->dir_cookie = array->array[i + 1].cookie;
1103                         desc->last_cookie = array->array[0].cookie;
1104                 }
1105                 if (nfs_readdir_use_cookie(file))
1106                         desc->ctx->pos = desc->dir_cookie;
1107                 else
1108                         desc->ctx->pos++;
1109         }
1110         if (array->page_is_eof)
1111                 desc->eof = !desc->eob;
1112
1113         kunmap_local(array);
1114         dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %llu\n",
1115                         (unsigned long long)desc->dir_cookie);
1116 }
1117
1118 /*
1119  * If we cannot find a cookie in our cache, we suspect that this is
1120  * because it points to a deleted file, so we ask the server to return
1121  * whatever it thinks is the next entry. We then feed this to filldir.
1122  * If all goes well, we should then be able to find our way round the
1123  * cache on the next call to readdir_search_pagecache();
1124  *
1125  * NOTE: we cannot add the anonymous page to the pagecache because
1126  *       the data it contains might not be page aligned. Besides,
1127  *       we should already have a complete representation of the
1128  *       directory in the page cache by the time we get here.
1129  */
1130 static int uncached_readdir(struct nfs_readdir_descriptor *desc)
1131 {
1132         struct page     **arrays;
1133         size_t          i, sz = 512;
1134         __be32          verf[NFS_DIR_VERIFIER_SIZE];
1135         int             status = -ENOMEM;
1136
1137         dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %llu\n",
1138                         (unsigned long long)desc->dir_cookie);
1139
1140         arrays = kcalloc(sz, sizeof(*arrays), GFP_KERNEL);
1141         if (!arrays)
1142                 goto out;
1143         arrays[0] = nfs_readdir_page_array_alloc(desc->dir_cookie, GFP_KERNEL);
1144         if (!arrays[0])
1145                 goto out;
1146
1147         desc->page_index = 0;
1148         desc->cache_entry_index = 0;
1149         desc->last_cookie = desc->dir_cookie;
1150         desc->page_index_max = 0;
1151
1152         trace_nfs_readdir_uncached(desc->file, desc->verf, desc->last_cookie,
1153                                    -1, desc->dtsize);
1154
1155         status = nfs_readdir_xdr_to_array(desc, desc->verf, verf, arrays, sz);
1156         if (status < 0) {
1157                 trace_nfs_readdir_uncached_done(file_inode(desc->file), status);
1158                 goto out_free;
1159         }
1160
1161         for (i = 0; !desc->eob && i < sz && arrays[i]; i++) {
1162                 desc->page = arrays[i];
1163                 nfs_do_filldir(desc, verf);
1164         }
1165         desc->page = NULL;
1166
1167         /*
1168          * Grow the dtsize if we have to go back for more pages,
1169          * or shrink it if we're reading too many.
1170          */
1171         if (!desc->eof) {
1172                 if (!desc->eob)
1173                         nfs_grow_dtsize(desc);
1174                 else if (desc->buffer_fills == 1 &&
1175                          i < (desc->page_index_max >> 1))
1176                         nfs_shrink_dtsize(desc);
1177         }
1178 out_free:
1179         for (i = 0; i < sz && arrays[i]; i++)
1180                 nfs_readdir_page_array_free(arrays[i]);
1181 out:
1182         if (!nfs_readdir_use_cookie(desc->file))
1183                 nfs_readdir_rewind_search(desc);
1184         desc->page_index_max = -1;
1185         kfree(arrays);
1186         dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __func__, status);
1187         return status;
1188 }
1189
1190 #define NFS_READDIR_CACHE_MISS_THRESHOLD (16UL)
1191
1192 static bool nfs_readdir_handle_cache_misses(struct inode *inode,
1193                                             struct nfs_readdir_descriptor *desc,
1194                                             unsigned int cache_misses,
1195                                             bool force_clear)
1196 {
1197         if (desc->ctx->pos == 0 || !desc->plus)
1198                 return false;
1199         if (cache_misses <= NFS_READDIR_CACHE_MISS_THRESHOLD && !force_clear)
1200                 return false;
1201         trace_nfs_readdir_force_readdirplus(inode);
1202         return true;
1203 }
1204
1205 /* The file offset position represents the dirent entry number.  A
1206    last cookie cache takes care of the common case of reading the
1207    whole directory.
1208  */
1209 static int nfs_readdir(struct file *file, struct dir_context *ctx)
1210 {
1211         struct dentry   *dentry = file_dentry(file);
1212         struct inode    *inode = d_inode(dentry);
1213         struct nfs_inode *nfsi = NFS_I(inode);
1214         struct nfs_open_dir_context *dir_ctx = file->private_data;
1215         struct nfs_readdir_descriptor *desc;
1216         unsigned int cache_hits, cache_misses;
1217         bool force_clear;
1218         int res;
1219
1220         dfprintk(FILE, "NFS: readdir(%pD2) starting at cookie %llu\n",
1221                         file, (long long)ctx->pos);
1222         nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
1223
1224         /*
1225          * ctx->pos points to the dirent entry number.
1226          * *desc->dir_cookie has the cookie for the next entry. We have
1227          * to either find the entry with the appropriate number or
1228          * revalidate the cookie.
1229          */
1230         nfs_revalidate_mapping(inode, file->f_mapping);
1231
1232         res = -ENOMEM;
1233         desc = kzalloc(sizeof(*desc), GFP_KERNEL);
1234         if (!desc)
1235                 goto out;
1236         desc->file = file;
1237         desc->ctx = ctx;
1238         desc->page_index_max = -1;
1239
1240         spin_lock(&file->f_lock);
1241         desc->dir_cookie = dir_ctx->dir_cookie;
1242         desc->page_index = dir_ctx->page_index;
1243         desc->last_cookie = dir_ctx->last_cookie;
1244         desc->attr_gencount = dir_ctx->attr_gencount;
1245         desc->eof = dir_ctx->eof;
1246         nfs_set_dtsize(desc, dir_ctx->dtsize);
1247         memcpy(desc->verf, dir_ctx->verf, sizeof(desc->verf));
1248         cache_hits = atomic_xchg(&dir_ctx->cache_hits, 0);
1249         cache_misses = atomic_xchg(&dir_ctx->cache_misses, 0);
1250         force_clear = dir_ctx->force_clear;
1251         spin_unlock(&file->f_lock);
1252
1253         if (desc->eof) {
1254                 res = 0;
1255                 goto out_free;
1256         }
1257
1258         desc->plus = nfs_use_readdirplus(inode, ctx, cache_hits, cache_misses);
1259         force_clear = nfs_readdir_handle_cache_misses(inode, desc, cache_misses,
1260                                                       force_clear);
1261         desc->clear_cache = force_clear;
1262
1263         do {
1264                 res = readdir_search_pagecache(desc);
1265
1266                 if (res == -EBADCOOKIE) {
1267                         res = 0;
1268                         /* This means either end of directory */
1269                         if (desc->dir_cookie && !desc->eof) {
1270                                 /* Or that the server has 'lost' a cookie */
1271                                 res = uncached_readdir(desc);
1272                                 if (res == 0)
1273                                         continue;
1274                                 if (res == -EBADCOOKIE || res == -ENOTSYNC)
1275                                         res = 0;
1276                         }
1277                         break;
1278                 }
1279                 if (res == -ETOOSMALL && desc->plus) {
1280                         nfs_zap_caches(inode);
1281                         desc->plus = false;
1282                         desc->eof = false;
1283                         continue;
1284                 }
1285                 if (res < 0)
1286                         break;
1287
1288                 nfs_do_filldir(desc, nfsi->cookieverf);
1289                 nfs_readdir_page_unlock_and_put_cached(desc);
1290                 if (desc->page_index == desc->page_index_max)
1291                         desc->clear_cache = force_clear;
1292         } while (!desc->eob && !desc->eof);
1293
1294         spin_lock(&file->f_lock);
1295         dir_ctx->dir_cookie = desc->dir_cookie;
1296         dir_ctx->last_cookie = desc->last_cookie;
1297         dir_ctx->attr_gencount = desc->attr_gencount;
1298         dir_ctx->page_index = desc->page_index;
1299         dir_ctx->force_clear = force_clear;
1300         dir_ctx->eof = desc->eof;
1301         dir_ctx->dtsize = desc->dtsize;
1302         memcpy(dir_ctx->verf, desc->verf, sizeof(dir_ctx->verf));
1303         spin_unlock(&file->f_lock);
1304 out_free:
1305         kfree(desc);
1306
1307 out:
1308         dfprintk(FILE, "NFS: readdir(%pD2) returns %d\n", file, res);
1309         return res;
1310 }
1311
1312 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int whence)
1313 {
1314         struct nfs_open_dir_context *dir_ctx = filp->private_data;
1315
1316         dfprintk(FILE, "NFS: llseek dir(%pD2, %lld, %d)\n",
1317                         filp, offset, whence);
1318
1319         switch (whence) {
1320         default:
1321                 return -EINVAL;
1322         case SEEK_SET:
1323                 if (offset < 0)
1324                         return -EINVAL;
1325                 spin_lock(&filp->f_lock);
1326                 break;
1327         case SEEK_CUR:
1328                 if (offset == 0)
1329                         return filp->f_pos;
1330                 spin_lock(&filp->f_lock);
1331                 offset += filp->f_pos;
1332                 if (offset < 0) {
1333                         spin_unlock(&filp->f_lock);
1334                         return -EINVAL;
1335                 }
1336         }
1337         if (offset != filp->f_pos) {
1338                 filp->f_pos = offset;
1339                 dir_ctx->page_index = 0;
1340                 if (!nfs_readdir_use_cookie(filp)) {
1341                         dir_ctx->dir_cookie = 0;
1342                         dir_ctx->last_cookie = 0;
1343                 } else {
1344                         dir_ctx->dir_cookie = offset;
1345                         dir_ctx->last_cookie = offset;
1346                 }
1347                 dir_ctx->eof = false;
1348         }
1349         spin_unlock(&filp->f_lock);
1350         return offset;
1351 }
1352
1353 /*
1354  * All directory operations under NFS are synchronous, so fsync()
1355  * is a dummy operation.
1356  */
1357 static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end,
1358                          int datasync)
1359 {
1360         dfprintk(FILE, "NFS: fsync dir(%pD2) datasync %d\n", filp, datasync);
1361
1362         nfs_inc_stats(file_inode(filp), NFSIOS_VFSFSYNC);
1363         return 0;
1364 }
1365
1366 /**
1367  * nfs_force_lookup_revalidate - Mark the directory as having changed
1368  * @dir: pointer to directory inode
1369  *
1370  * This forces the revalidation code in nfs_lookup_revalidate() to do a
1371  * full lookup on all child dentries of 'dir' whenever a change occurs
1372  * on the server that might have invalidated our dcache.
1373  *
1374  * Note that we reserve bit '0' as a tag to let us know when a dentry
1375  * was revalidated while holding a delegation on its inode.
1376  *
1377  * The caller should be holding dir->i_lock
1378  */
1379 void nfs_force_lookup_revalidate(struct inode *dir)
1380 {
1381         NFS_I(dir)->cache_change_attribute += 2;
1382 }
1383 EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate);
1384
1385 /**
1386  * nfs_verify_change_attribute - Detects NFS remote directory changes
1387  * @dir: pointer to parent directory inode
1388  * @verf: previously saved change attribute
1389  *
1390  * Return "false" if the verifiers doesn't match the change attribute.
1391  * This would usually indicate that the directory contents have changed on
1392  * the server, and that any dentries need revalidating.
1393  */
1394 static bool nfs_verify_change_attribute(struct inode *dir, unsigned long verf)
1395 {
1396         return (verf & ~1UL) == nfs_save_change_attribute(dir);
1397 }
1398
1399 static void nfs_set_verifier_delegated(unsigned long *verf)
1400 {
1401         *verf |= 1UL;
1402 }
1403
1404 #if IS_ENABLED(CONFIG_NFS_V4)
1405 static void nfs_unset_verifier_delegated(unsigned long *verf)
1406 {
1407         *verf &= ~1UL;
1408 }
1409 #endif /* IS_ENABLED(CONFIG_NFS_V4) */
1410
1411 static bool nfs_test_verifier_delegated(unsigned long verf)
1412 {
1413         return verf & 1;
1414 }
1415
1416 static bool nfs_verifier_is_delegated(struct dentry *dentry)
1417 {
1418         return nfs_test_verifier_delegated(dentry->d_time);
1419 }
1420
1421 static void nfs_set_verifier_locked(struct dentry *dentry, unsigned long verf)
1422 {
1423         struct inode *inode = d_inode(dentry);
1424         struct inode *dir = d_inode(dentry->d_parent);
1425
1426         if (!nfs_verify_change_attribute(dir, verf))
1427                 return;
1428         if (inode && NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
1429                 nfs_set_verifier_delegated(&verf);
1430         dentry->d_time = verf;
1431 }
1432
1433 /**
1434  * nfs_set_verifier - save a parent directory verifier in the dentry
1435  * @dentry: pointer to dentry
1436  * @verf: verifier to save
1437  *
1438  * Saves the parent directory verifier in @dentry. If the inode has
1439  * a delegation, we also tag the dentry as having been revalidated
1440  * while holding a delegation so that we know we don't have to
1441  * look it up again after a directory change.
1442  */
1443 void nfs_set_verifier(struct dentry *dentry, unsigned long verf)
1444 {
1445
1446         spin_lock(&dentry->d_lock);
1447         nfs_set_verifier_locked(dentry, verf);
1448         spin_unlock(&dentry->d_lock);
1449 }
1450 EXPORT_SYMBOL_GPL(nfs_set_verifier);
1451
1452 #if IS_ENABLED(CONFIG_NFS_V4)
1453 /**
1454  * nfs_clear_verifier_delegated - clear the dir verifier delegation tag
1455  * @inode: pointer to inode
1456  *
1457  * Iterates through the dentries in the inode alias list and clears
1458  * the tag used to indicate that the dentry has been revalidated
1459  * while holding a delegation.
1460  * This function is intended for use when the delegation is being
1461  * returned or revoked.
1462  */
1463 void nfs_clear_verifier_delegated(struct inode *inode)
1464 {
1465         struct dentry *alias;
1466
1467         if (!inode)
1468                 return;
1469         spin_lock(&inode->i_lock);
1470         hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
1471                 spin_lock(&alias->d_lock);
1472                 nfs_unset_verifier_delegated(&alias->d_time);
1473                 spin_unlock(&alias->d_lock);
1474         }
1475         spin_unlock(&inode->i_lock);
1476 }
1477 EXPORT_SYMBOL_GPL(nfs_clear_verifier_delegated);
1478 #endif /* IS_ENABLED(CONFIG_NFS_V4) */
1479
1480 static int nfs_dentry_verify_change(struct inode *dir, struct dentry *dentry)
1481 {
1482         if (nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE) &&
1483             d_really_is_negative(dentry))
1484                 return dentry->d_time == inode_peek_iversion_raw(dir);
1485         return nfs_verify_change_attribute(dir, dentry->d_time);
1486 }
1487
1488 /*
1489  * A check for whether or not the parent directory has changed.
1490  * In the case it has, we assume that the dentries are untrustworthy
1491  * and may need to be looked up again.
1492  * If rcu_walk prevents us from performing a full check, return 0.
1493  */
1494 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry,
1495                               int rcu_walk)
1496 {
1497         if (IS_ROOT(dentry))
1498                 return 1;
1499         if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
1500                 return 0;
1501         if (!nfs_dentry_verify_change(dir, dentry))
1502                 return 0;
1503         /* Revalidate nfsi->cache_change_attribute before we declare a match */
1504         if (nfs_mapping_need_revalidate_inode(dir)) {
1505                 if (rcu_walk)
1506                         return 0;
1507                 if (__nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
1508                         return 0;
1509         }
1510         if (!nfs_dentry_verify_change(dir, dentry))
1511                 return 0;
1512         return 1;
1513 }
1514
1515 /*
1516  * Use intent information to check whether or not we're going to do
1517  * an O_EXCL create using this path component.
1518  */
1519 static int nfs_is_exclusive_create(struct inode *dir, unsigned int flags)
1520 {
1521         if (NFS_PROTO(dir)->version == 2)
1522                 return 0;
1523         return flags & LOOKUP_EXCL;
1524 }
1525
1526 /*
1527  * Inode and filehandle revalidation for lookups.
1528  *
1529  * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
1530  * or if the intent information indicates that we're about to open this
1531  * particular file and the "nocto" mount flag is not set.
1532  *
1533  */
1534 static
1535 int nfs_lookup_verify_inode(struct inode *inode, unsigned int flags)
1536 {
1537         struct nfs_server *server = NFS_SERVER(inode);
1538         int ret;
1539
1540         if (IS_AUTOMOUNT(inode))
1541                 return 0;
1542
1543         if (flags & LOOKUP_OPEN) {
1544                 switch (inode->i_mode & S_IFMT) {
1545                 case S_IFREG:
1546                         /* A NFSv4 OPEN will revalidate later */
1547                         if (server->caps & NFS_CAP_ATOMIC_OPEN)
1548                                 goto out;
1549                         fallthrough;
1550                 case S_IFDIR:
1551                         if (server->flags & NFS_MOUNT_NOCTO)
1552                                 break;
1553                         /* NFS close-to-open cache consistency validation */
1554                         goto out_force;
1555                 }
1556         }
1557
1558         /* VFS wants an on-the-wire revalidation */
1559         if (flags & LOOKUP_REVAL)
1560                 goto out_force;
1561 out:
1562         if (inode->i_nlink > 0 ||
1563             (inode->i_nlink == 0 &&
1564              test_bit(NFS_INO_PRESERVE_UNLINKED, &NFS_I(inode)->flags)))
1565                 return 0;
1566         else
1567                 return -ESTALE;
1568 out_force:
1569         if (flags & LOOKUP_RCU)
1570                 return -ECHILD;
1571         ret = __nfs_revalidate_inode(server, inode);
1572         if (ret != 0)
1573                 return ret;
1574         goto out;
1575 }
1576
1577 static void nfs_mark_dir_for_revalidate(struct inode *inode)
1578 {
1579         spin_lock(&inode->i_lock);
1580         nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE);
1581         spin_unlock(&inode->i_lock);
1582 }
1583
1584 /*
1585  * We judge how long we want to trust negative
1586  * dentries by looking at the parent inode mtime.
1587  *
1588  * If parent mtime has changed, we revalidate, else we wait for a
1589  * period corresponding to the parent's attribute cache timeout value.
1590  *
1591  * If LOOKUP_RCU prevents us from performing a full check, return 1
1592  * suggesting a reval is needed.
1593  *
1594  * Note that when creating a new file, or looking up a rename target,
1595  * then it shouldn't be necessary to revalidate a negative dentry.
1596  */
1597 static inline
1598 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
1599                        unsigned int flags)
1600 {
1601         if (flags & (LOOKUP_CREATE | LOOKUP_RENAME_TARGET))
1602                 return 0;
1603         if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
1604                 return 1;
1605         /* Case insensitive server? Revalidate negative dentries */
1606         if (nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE))
1607                 return 1;
1608         return !nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU);
1609 }
1610
1611 static int
1612 nfs_lookup_revalidate_done(struct inode *dir, struct dentry *dentry,
1613                            struct inode *inode, int error)
1614 {
1615         switch (error) {
1616         case 1:
1617                 break;
1618         case 0:
1619                 /*
1620                  * We can't d_drop the root of a disconnected tree:
1621                  * its d_hash is on the s_anon list and d_drop() would hide
1622                  * it from shrink_dcache_for_unmount(), leading to busy
1623                  * inodes on unmount and further oopses.
1624                  */
1625                 if (inode && IS_ROOT(dentry))
1626                         error = 1;
1627                 break;
1628         }
1629         trace_nfs_lookup_revalidate_exit(dir, dentry, 0, error);
1630         return error;
1631 }
1632
1633 static int
1634 nfs_lookup_revalidate_negative(struct inode *dir, struct dentry *dentry,
1635                                unsigned int flags)
1636 {
1637         int ret = 1;
1638         if (nfs_neg_need_reval(dir, dentry, flags)) {
1639                 if (flags & LOOKUP_RCU)
1640                         return -ECHILD;
1641                 ret = 0;
1642         }
1643         return nfs_lookup_revalidate_done(dir, dentry, NULL, ret);
1644 }
1645
1646 static int
1647 nfs_lookup_revalidate_delegated(struct inode *dir, struct dentry *dentry,
1648                                 struct inode *inode)
1649 {
1650         nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1651         return nfs_lookup_revalidate_done(dir, dentry, inode, 1);
1652 }
1653
1654 static int nfs_lookup_revalidate_dentry(struct inode *dir,
1655                                         struct dentry *dentry,
1656                                         struct inode *inode, unsigned int flags)
1657 {
1658         struct nfs_fh *fhandle;
1659         struct nfs_fattr *fattr;
1660         unsigned long dir_verifier;
1661         int ret;
1662
1663         trace_nfs_lookup_revalidate_enter(dir, dentry, flags);
1664
1665         ret = -ENOMEM;
1666         fhandle = nfs_alloc_fhandle();
1667         fattr = nfs_alloc_fattr_with_label(NFS_SERVER(inode));
1668         if (fhandle == NULL || fattr == NULL)
1669                 goto out;
1670
1671         dir_verifier = nfs_save_change_attribute(dir);
1672         ret = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr);
1673         if (ret < 0) {
1674                 switch (ret) {
1675                 case -ESTALE:
1676                 case -ENOENT:
1677                         ret = 0;
1678                         break;
1679                 case -ETIMEDOUT:
1680                         if (NFS_SERVER(inode)->flags & NFS_MOUNT_SOFTREVAL)
1681                                 ret = 1;
1682                 }
1683                 goto out;
1684         }
1685
1686         /* Request help from readdirplus */
1687         nfs_lookup_advise_force_readdirplus(dir, flags);
1688
1689         ret = 0;
1690         if (nfs_compare_fh(NFS_FH(inode), fhandle))
1691                 goto out;
1692         if (nfs_refresh_inode(inode, fattr) < 0)
1693                 goto out;
1694
1695         nfs_setsecurity(inode, fattr);
1696         nfs_set_verifier(dentry, dir_verifier);
1697
1698         ret = 1;
1699 out:
1700         nfs_free_fattr(fattr);
1701         nfs_free_fhandle(fhandle);
1702
1703         /*
1704          * If the lookup failed despite the dentry change attribute being
1705          * a match, then we should revalidate the directory cache.
1706          */
1707         if (!ret && nfs_dentry_verify_change(dir, dentry))
1708                 nfs_mark_dir_for_revalidate(dir);
1709         return nfs_lookup_revalidate_done(dir, dentry, inode, ret);
1710 }
1711
1712 /*
1713  * This is called every time the dcache has a lookup hit,
1714  * and we should check whether we can really trust that
1715  * lookup.
1716  *
1717  * NOTE! The hit can be a negative hit too, don't assume
1718  * we have an inode!
1719  *
1720  * If the parent directory is seen to have changed, we throw out the
1721  * cached dentry and do a new lookup.
1722  */
1723 static int
1724 nfs_do_lookup_revalidate(struct inode *dir, struct dentry *dentry,
1725                          unsigned int flags)
1726 {
1727         struct inode *inode;
1728         int error;
1729
1730         nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
1731         inode = d_inode(dentry);
1732
1733         if (!inode)
1734                 return nfs_lookup_revalidate_negative(dir, dentry, flags);
1735
1736         if (is_bad_inode(inode)) {
1737                 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1738                                 __func__, dentry);
1739                 goto out_bad;
1740         }
1741
1742         if ((flags & LOOKUP_RENAME_TARGET) && d_count(dentry) < 2 &&
1743             nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE))
1744                 goto out_bad;
1745
1746         if (nfs_verifier_is_delegated(dentry))
1747                 return nfs_lookup_revalidate_delegated(dir, dentry, inode);
1748
1749         /* Force a full look up iff the parent directory has changed */
1750         if (!(flags & (LOOKUP_EXCL | LOOKUP_REVAL)) &&
1751             nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU)) {
1752                 error = nfs_lookup_verify_inode(inode, flags);
1753                 if (error) {
1754                         if (error == -ESTALE)
1755                                 nfs_mark_dir_for_revalidate(dir);
1756                         goto out_bad;
1757                 }
1758                 goto out_valid;
1759         }
1760
1761         if (flags & LOOKUP_RCU)
1762                 return -ECHILD;
1763
1764         if (NFS_STALE(inode))
1765                 goto out_bad;
1766
1767         return nfs_lookup_revalidate_dentry(dir, dentry, inode, flags);
1768 out_valid:
1769         return nfs_lookup_revalidate_done(dir, dentry, inode, 1);
1770 out_bad:
1771         if (flags & LOOKUP_RCU)
1772                 return -ECHILD;
1773         return nfs_lookup_revalidate_done(dir, dentry, inode, 0);
1774 }
1775
1776 static int
1777 __nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags,
1778                         int (*reval)(struct inode *, struct dentry *, unsigned int))
1779 {
1780         struct dentry *parent;
1781         struct inode *dir;
1782         int ret;
1783
1784         if (flags & LOOKUP_RCU) {
1785                 if (dentry->d_fsdata == NFS_FSDATA_BLOCKED)
1786                         return -ECHILD;
1787                 parent = READ_ONCE(dentry->d_parent);
1788                 dir = d_inode_rcu(parent);
1789                 if (!dir)
1790                         return -ECHILD;
1791                 ret = reval(dir, dentry, flags);
1792                 if (parent != READ_ONCE(dentry->d_parent))
1793                         return -ECHILD;
1794         } else {
1795                 /* Wait for unlink to complete */
1796                 wait_var_event(&dentry->d_fsdata,
1797                                dentry->d_fsdata != NFS_FSDATA_BLOCKED);
1798                 parent = dget_parent(dentry);
1799                 ret = reval(d_inode(parent), dentry, flags);
1800                 dput(parent);
1801         }
1802         return ret;
1803 }
1804
1805 static int nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1806 {
1807         return __nfs_lookup_revalidate(dentry, flags, nfs_do_lookup_revalidate);
1808 }
1809
1810 /*
1811  * A weaker form of d_revalidate for revalidating just the d_inode(dentry)
1812  * when we don't really care about the dentry name. This is called when a
1813  * pathwalk ends on a dentry that was not found via a normal lookup in the
1814  * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals).
1815  *
1816  * In this situation, we just want to verify that the inode itself is OK
1817  * since the dentry might have changed on the server.
1818  */
1819 static int nfs_weak_revalidate(struct dentry *dentry, unsigned int flags)
1820 {
1821         struct inode *inode = d_inode(dentry);
1822         int error = 0;
1823
1824         /*
1825          * I believe we can only get a negative dentry here in the case of a
1826          * procfs-style symlink. Just assume it's correct for now, but we may
1827          * eventually need to do something more here.
1828          */
1829         if (!inode) {
1830                 dfprintk(LOOKUPCACHE, "%s: %pd2 has negative inode\n",
1831                                 __func__, dentry);
1832                 return 1;
1833         }
1834
1835         if (is_bad_inode(inode)) {
1836                 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1837                                 __func__, dentry);
1838                 return 0;
1839         }
1840
1841         error = nfs_lookup_verify_inode(inode, flags);
1842         dfprintk(LOOKUPCACHE, "NFS: %s: inode %lu is %s\n",
1843                         __func__, inode->i_ino, error ? "invalid" : "valid");
1844         return !error;
1845 }
1846
1847 /*
1848  * This is called from dput() when d_count is going to 0.
1849  */
1850 static int nfs_dentry_delete(const struct dentry *dentry)
1851 {
1852         dfprintk(VFS, "NFS: dentry_delete(%pd2, %x)\n",
1853                 dentry, dentry->d_flags);
1854
1855         /* Unhash any dentry with a stale inode */
1856         if (d_really_is_positive(dentry) && NFS_STALE(d_inode(dentry)))
1857                 return 1;
1858
1859         if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1860                 /* Unhash it, so that ->d_iput() would be called */
1861                 return 1;
1862         }
1863         if (!(dentry->d_sb->s_flags & SB_ACTIVE)) {
1864                 /* Unhash it, so that ancestors of killed async unlink
1865                  * files will be cleaned up during umount */
1866                 return 1;
1867         }
1868         return 0;
1869
1870 }
1871
1872 /* Ensure that we revalidate inode->i_nlink */
1873 static void nfs_drop_nlink(struct inode *inode)
1874 {
1875         spin_lock(&inode->i_lock);
1876         /* drop the inode if we're reasonably sure this is the last link */
1877         if (inode->i_nlink > 0)
1878                 drop_nlink(inode);
1879         NFS_I(inode)->attr_gencount = nfs_inc_attr_generation_counter();
1880         nfs_set_cache_invalid(
1881                 inode, NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_CTIME |
1882                                NFS_INO_INVALID_NLINK);
1883         spin_unlock(&inode->i_lock);
1884 }
1885
1886 /*
1887  * Called when the dentry loses inode.
1888  * We use it to clean up silly-renamed files.
1889  */
1890 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
1891 {
1892         if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1893                 nfs_complete_unlink(dentry, inode);
1894                 nfs_drop_nlink(inode);
1895         }
1896         iput(inode);
1897 }
1898
1899 static void nfs_d_release(struct dentry *dentry)
1900 {
1901         /* free cached devname value, if it survived that far */
1902         if (unlikely(dentry->d_fsdata)) {
1903                 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1904                         WARN_ON(1);
1905                 else
1906                         kfree(dentry->d_fsdata);
1907         }
1908 }
1909
1910 const struct dentry_operations nfs_dentry_operations = {
1911         .d_revalidate   = nfs_lookup_revalidate,
1912         .d_weak_revalidate      = nfs_weak_revalidate,
1913         .d_delete       = nfs_dentry_delete,
1914         .d_iput         = nfs_dentry_iput,
1915         .d_automount    = nfs_d_automount,
1916         .d_release      = nfs_d_release,
1917 };
1918 EXPORT_SYMBOL_GPL(nfs_dentry_operations);
1919
1920 struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
1921 {
1922         struct dentry *res;
1923         struct inode *inode = NULL;
1924         struct nfs_fh *fhandle = NULL;
1925         struct nfs_fattr *fattr = NULL;
1926         unsigned long dir_verifier;
1927         int error;
1928
1929         dfprintk(VFS, "NFS: lookup(%pd2)\n", dentry);
1930         nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
1931
1932         if (unlikely(dentry->d_name.len > NFS_SERVER(dir)->namelen))
1933                 return ERR_PTR(-ENAMETOOLONG);
1934
1935         /*
1936          * If we're doing an exclusive create, optimize away the lookup
1937          * but don't hash the dentry.
1938          */
1939         if (nfs_is_exclusive_create(dir, flags) || flags & LOOKUP_RENAME_TARGET)
1940                 return NULL;
1941
1942         res = ERR_PTR(-ENOMEM);
1943         fhandle = nfs_alloc_fhandle();
1944         fattr = nfs_alloc_fattr_with_label(NFS_SERVER(dir));
1945         if (fhandle == NULL || fattr == NULL)
1946                 goto out;
1947
1948         dir_verifier = nfs_save_change_attribute(dir);
1949         trace_nfs_lookup_enter(dir, dentry, flags);
1950         error = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr);
1951         if (error == -ENOENT) {
1952                 if (nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE))
1953                         dir_verifier = inode_peek_iversion_raw(dir);
1954                 goto no_entry;
1955         }
1956         if (error < 0) {
1957                 res = ERR_PTR(error);
1958                 goto out;
1959         }
1960         inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1961         res = ERR_CAST(inode);
1962         if (IS_ERR(res))
1963                 goto out;
1964
1965         /* Notify readdir to use READDIRPLUS */
1966         nfs_lookup_advise_force_readdirplus(dir, flags);
1967
1968 no_entry:
1969         res = d_splice_alias(inode, dentry);
1970         if (res != NULL) {
1971                 if (IS_ERR(res))
1972                         goto out;
1973                 dentry = res;
1974         }
1975         nfs_set_verifier(dentry, dir_verifier);
1976 out:
1977         trace_nfs_lookup_exit(dir, dentry, flags, PTR_ERR_OR_ZERO(res));
1978         nfs_free_fattr(fattr);
1979         nfs_free_fhandle(fhandle);
1980         return res;
1981 }
1982 EXPORT_SYMBOL_GPL(nfs_lookup);
1983
1984 void nfs_d_prune_case_insensitive_aliases(struct inode *inode)
1985 {
1986         /* Case insensitive server? Revalidate dentries */
1987         if (inode && nfs_server_capable(inode, NFS_CAP_CASE_INSENSITIVE))
1988                 d_prune_aliases(inode);
1989 }
1990 EXPORT_SYMBOL_GPL(nfs_d_prune_case_insensitive_aliases);
1991
1992 #if IS_ENABLED(CONFIG_NFS_V4)
1993 static int nfs4_lookup_revalidate(struct dentry *, unsigned int);
1994
1995 const struct dentry_operations nfs4_dentry_operations = {
1996         .d_revalidate   = nfs4_lookup_revalidate,
1997         .d_weak_revalidate      = nfs_weak_revalidate,
1998         .d_delete       = nfs_dentry_delete,
1999         .d_iput         = nfs_dentry_iput,
2000         .d_automount    = nfs_d_automount,
2001         .d_release      = nfs_d_release,
2002 };
2003 EXPORT_SYMBOL_GPL(nfs4_dentry_operations);
2004
2005 static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags, struct file *filp)
2006 {
2007         return alloc_nfs_open_context(dentry, flags_to_mode(open_flags), filp);
2008 }
2009
2010 static int do_open(struct inode *inode, struct file *filp)
2011 {
2012         nfs_fscache_open_file(inode, filp);
2013         return 0;
2014 }
2015
2016 static int nfs_finish_open(struct nfs_open_context *ctx,
2017                            struct dentry *dentry,
2018                            struct file *file, unsigned open_flags)
2019 {
2020         int err;
2021
2022         err = finish_open(file, dentry, do_open);
2023         if (err)
2024                 goto out;
2025         if (S_ISREG(file->f_path.dentry->d_inode->i_mode))
2026                 nfs_file_set_open_context(file, ctx);
2027         else
2028                 err = -EOPENSTALE;
2029 out:
2030         return err;
2031 }
2032
2033 int nfs_atomic_open(struct inode *dir, struct dentry *dentry,
2034                     struct file *file, unsigned open_flags,
2035                     umode_t mode)
2036 {
2037         DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
2038         struct nfs_open_context *ctx;
2039         struct dentry *res;
2040         struct iattr attr = { .ia_valid = ATTR_OPEN };
2041         struct inode *inode;
2042         unsigned int lookup_flags = 0;
2043         unsigned long dir_verifier;
2044         bool switched = false;
2045         int created = 0;
2046         int err;
2047
2048         /* Expect a negative dentry */
2049         BUG_ON(d_inode(dentry));
2050
2051         dfprintk(VFS, "NFS: atomic_open(%s/%lu), %pd\n",
2052                         dir->i_sb->s_id, dir->i_ino, dentry);
2053
2054         err = nfs_check_flags(open_flags);
2055         if (err)
2056                 return err;
2057
2058         /* NFS only supports OPEN on regular files */
2059         if ((open_flags & O_DIRECTORY)) {
2060                 if (!d_in_lookup(dentry)) {
2061                         /*
2062                          * Hashed negative dentry with O_DIRECTORY: dentry was
2063                          * revalidated and is fine, no need to perform lookup
2064                          * again
2065                          */
2066                         return -ENOENT;
2067                 }
2068                 lookup_flags = LOOKUP_OPEN|LOOKUP_DIRECTORY;
2069                 goto no_open;
2070         }
2071
2072         if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
2073                 return -ENAMETOOLONG;
2074
2075         if (open_flags & O_CREAT) {
2076                 struct nfs_server *server = NFS_SERVER(dir);
2077
2078                 if (!(server->attr_bitmask[2] & FATTR4_WORD2_MODE_UMASK))
2079                         mode &= ~current_umask();
2080
2081                 attr.ia_valid |= ATTR_MODE;
2082                 attr.ia_mode = mode;
2083         }
2084         if (open_flags & O_TRUNC) {
2085                 attr.ia_valid |= ATTR_SIZE;
2086                 attr.ia_size = 0;
2087         }
2088
2089         if (!(open_flags & O_CREAT) && !d_in_lookup(dentry)) {
2090                 d_drop(dentry);
2091                 switched = true;
2092                 dentry = d_alloc_parallel(dentry->d_parent,
2093                                           &dentry->d_name, &wq);
2094                 if (IS_ERR(dentry))
2095                         return PTR_ERR(dentry);
2096                 if (unlikely(!d_in_lookup(dentry)))
2097                         return finish_no_open(file, dentry);
2098         }
2099
2100         ctx = create_nfs_open_context(dentry, open_flags, file);
2101         err = PTR_ERR(ctx);
2102         if (IS_ERR(ctx))
2103                 goto out;
2104
2105         trace_nfs_atomic_open_enter(dir, ctx, open_flags);
2106         inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr, &created);
2107         if (created)
2108                 file->f_mode |= FMODE_CREATED;
2109         if (IS_ERR(inode)) {
2110                 err = PTR_ERR(inode);
2111                 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
2112                 put_nfs_open_context(ctx);
2113                 d_drop(dentry);
2114                 switch (err) {
2115                 case -ENOENT:
2116                         d_splice_alias(NULL, dentry);
2117                         if (nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE))
2118                                 dir_verifier = inode_peek_iversion_raw(dir);
2119                         else
2120                                 dir_verifier = nfs_save_change_attribute(dir);
2121                         nfs_set_verifier(dentry, dir_verifier);
2122                         break;
2123                 case -EISDIR:
2124                 case -ENOTDIR:
2125                         goto no_open;
2126                 case -ELOOP:
2127                         if (!(open_flags & O_NOFOLLOW))
2128                                 goto no_open;
2129                         break;
2130                         /* case -EINVAL: */
2131                 default:
2132                         break;
2133                 }
2134                 goto out;
2135         }
2136         file->f_mode |= FMODE_CAN_ODIRECT;
2137
2138         err = nfs_finish_open(ctx, ctx->dentry, file, open_flags);
2139         trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
2140         put_nfs_open_context(ctx);
2141 out:
2142         if (unlikely(switched)) {
2143                 d_lookup_done(dentry);
2144                 dput(dentry);
2145         }
2146         return err;
2147
2148 no_open:
2149         res = nfs_lookup(dir, dentry, lookup_flags);
2150         if (!res) {
2151                 inode = d_inode(dentry);
2152                 if ((lookup_flags & LOOKUP_DIRECTORY) && inode &&
2153                     !(S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)))
2154                         res = ERR_PTR(-ENOTDIR);
2155                 else if (inode && S_ISREG(inode->i_mode))
2156                         res = ERR_PTR(-EOPENSTALE);
2157         } else if (!IS_ERR(res)) {
2158                 inode = d_inode(res);
2159                 if ((lookup_flags & LOOKUP_DIRECTORY) && inode &&
2160                     !(S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))) {
2161                         dput(res);
2162                         res = ERR_PTR(-ENOTDIR);
2163                 } else if (inode && S_ISREG(inode->i_mode)) {
2164                         dput(res);
2165                         res = ERR_PTR(-EOPENSTALE);
2166                 }
2167         }
2168         if (switched) {
2169                 d_lookup_done(dentry);
2170                 if (!res)
2171                         res = dentry;
2172                 else
2173                         dput(dentry);
2174         }
2175         if (IS_ERR(res))
2176                 return PTR_ERR(res);
2177         return finish_no_open(file, res);
2178 }
2179 EXPORT_SYMBOL_GPL(nfs_atomic_open);
2180
2181 static int
2182 nfs4_do_lookup_revalidate(struct inode *dir, struct dentry *dentry,
2183                           unsigned int flags)
2184 {
2185         struct inode *inode;
2186
2187         if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY))
2188                 goto full_reval;
2189         if (d_mountpoint(dentry))
2190                 goto full_reval;
2191
2192         inode = d_inode(dentry);
2193
2194         /* We can't create new files in nfs_open_revalidate(), so we
2195          * optimize away revalidation of negative dentries.
2196          */
2197         if (inode == NULL)
2198                 goto full_reval;
2199
2200         if (nfs_verifier_is_delegated(dentry))
2201                 return nfs_lookup_revalidate_delegated(dir, dentry, inode);
2202
2203         /* NFS only supports OPEN on regular files */
2204         if (!S_ISREG(inode->i_mode))
2205                 goto full_reval;
2206
2207         /* We cannot do exclusive creation on a positive dentry */
2208         if (flags & (LOOKUP_EXCL | LOOKUP_REVAL))
2209                 goto reval_dentry;
2210
2211         /* Check if the directory changed */
2212         if (!nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU))
2213                 goto reval_dentry;
2214
2215         /* Let f_op->open() actually open (and revalidate) the file */
2216         return 1;
2217 reval_dentry:
2218         if (flags & LOOKUP_RCU)
2219                 return -ECHILD;
2220         return nfs_lookup_revalidate_dentry(dir, dentry, inode, flags);
2221
2222 full_reval:
2223         return nfs_do_lookup_revalidate(dir, dentry, flags);
2224 }
2225
2226 static int nfs4_lookup_revalidate(struct dentry *dentry, unsigned int flags)
2227 {
2228         return __nfs_lookup_revalidate(dentry, flags,
2229                         nfs4_do_lookup_revalidate);
2230 }
2231
2232 #endif /* CONFIG_NFSV4 */
2233
2234 struct dentry *
2235 nfs_add_or_obtain(struct dentry *dentry, struct nfs_fh *fhandle,
2236                                 struct nfs_fattr *fattr)
2237 {
2238         struct dentry *parent = dget_parent(dentry);
2239         struct inode *dir = d_inode(parent);
2240         struct inode *inode;
2241         struct dentry *d;
2242         int error;
2243
2244         d_drop(dentry);
2245
2246         if (fhandle->size == 0) {
2247                 error = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr);
2248                 if (error)
2249                         goto out_error;
2250         }
2251         nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2252         if (!(fattr->valid & NFS_ATTR_FATTR)) {
2253                 struct nfs_server *server = NFS_SB(dentry->d_sb);
2254                 error = server->nfs_client->rpc_ops->getattr(server, fhandle,
2255                                 fattr, NULL);
2256                 if (error < 0)
2257                         goto out_error;
2258         }
2259         inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
2260         d = d_splice_alias(inode, dentry);
2261 out:
2262         dput(parent);
2263         return d;
2264 out_error:
2265         d = ERR_PTR(error);
2266         goto out;
2267 }
2268 EXPORT_SYMBOL_GPL(nfs_add_or_obtain);
2269
2270 /*
2271  * Code common to create, mkdir, and mknod.
2272  */
2273 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
2274                                 struct nfs_fattr *fattr)
2275 {
2276         struct dentry *d;
2277
2278         d = nfs_add_or_obtain(dentry, fhandle, fattr);
2279         if (IS_ERR(d))
2280                 return PTR_ERR(d);
2281
2282         /* Callers don't care */
2283         dput(d);
2284         return 0;
2285 }
2286 EXPORT_SYMBOL_GPL(nfs_instantiate);
2287
2288 /*
2289  * Following a failed create operation, we drop the dentry rather
2290  * than retain a negative dentry. This avoids a problem in the event
2291  * that the operation succeeded on the server, but an error in the
2292  * reply path made it appear to have failed.
2293  */
2294 int nfs_create(struct user_namespace *mnt_userns, struct inode *dir,
2295                struct dentry *dentry, umode_t mode, bool excl)
2296 {
2297         struct iattr attr;
2298         int open_flags = excl ? O_CREAT | O_EXCL : O_CREAT;
2299         int error;
2300
2301         dfprintk(VFS, "NFS: create(%s/%lu), %pd\n",
2302                         dir->i_sb->s_id, dir->i_ino, dentry);
2303
2304         attr.ia_mode = mode;
2305         attr.ia_valid = ATTR_MODE;
2306
2307         trace_nfs_create_enter(dir, dentry, open_flags);
2308         error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags);
2309         trace_nfs_create_exit(dir, dentry, open_flags, error);
2310         if (error != 0)
2311                 goto out_err;
2312         return 0;
2313 out_err:
2314         d_drop(dentry);
2315         return error;
2316 }
2317 EXPORT_SYMBOL_GPL(nfs_create);
2318
2319 /*
2320  * See comments for nfs_proc_create regarding failed operations.
2321  */
2322 int
2323 nfs_mknod(struct user_namespace *mnt_userns, struct inode *dir,
2324           struct dentry *dentry, umode_t mode, dev_t rdev)
2325 {
2326         struct iattr attr;
2327         int status;
2328
2329         dfprintk(VFS, "NFS: mknod(%s/%lu), %pd\n",
2330                         dir->i_sb->s_id, dir->i_ino, dentry);
2331
2332         attr.ia_mode = mode;
2333         attr.ia_valid = ATTR_MODE;
2334
2335         trace_nfs_mknod_enter(dir, dentry);
2336         status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
2337         trace_nfs_mknod_exit(dir, dentry, status);
2338         if (status != 0)
2339                 goto out_err;
2340         return 0;
2341 out_err:
2342         d_drop(dentry);
2343         return status;
2344 }
2345 EXPORT_SYMBOL_GPL(nfs_mknod);
2346
2347 /*
2348  * See comments for nfs_proc_create regarding failed operations.
2349  */
2350 int nfs_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
2351               struct dentry *dentry, umode_t mode)
2352 {
2353         struct iattr attr;
2354         int error;
2355
2356         dfprintk(VFS, "NFS: mkdir(%s/%lu), %pd\n",
2357                         dir->i_sb->s_id, dir->i_ino, dentry);
2358
2359         attr.ia_valid = ATTR_MODE;
2360         attr.ia_mode = mode | S_IFDIR;
2361
2362         trace_nfs_mkdir_enter(dir, dentry);
2363         error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
2364         trace_nfs_mkdir_exit(dir, dentry, error);
2365         if (error != 0)
2366                 goto out_err;
2367         return 0;
2368 out_err:
2369         d_drop(dentry);
2370         return error;
2371 }
2372 EXPORT_SYMBOL_GPL(nfs_mkdir);
2373
2374 static void nfs_dentry_handle_enoent(struct dentry *dentry)
2375 {
2376         if (simple_positive(dentry))
2377                 d_delete(dentry);
2378 }
2379
2380 static void nfs_dentry_remove_handle_error(struct inode *dir,
2381                                            struct dentry *dentry, int error)
2382 {
2383         switch (error) {
2384         case -ENOENT:
2385                 d_delete(dentry);
2386                 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2387                 break;
2388         case 0:
2389                 nfs_d_prune_case_insensitive_aliases(d_inode(dentry));
2390                 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2391         }
2392 }
2393
2394 int nfs_rmdir(struct inode *dir, struct dentry *dentry)
2395 {
2396         int error;
2397
2398         dfprintk(VFS, "NFS: rmdir(%s/%lu), %pd\n",
2399                         dir->i_sb->s_id, dir->i_ino, dentry);
2400
2401         trace_nfs_rmdir_enter(dir, dentry);
2402         if (d_really_is_positive(dentry)) {
2403                 down_write(&NFS_I(d_inode(dentry))->rmdir_sem);
2404                 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
2405                 /* Ensure the VFS deletes this inode */
2406                 switch (error) {
2407                 case 0:
2408                         clear_nlink(d_inode(dentry));
2409                         break;
2410                 case -ENOENT:
2411                         nfs_dentry_handle_enoent(dentry);
2412                 }
2413                 up_write(&NFS_I(d_inode(dentry))->rmdir_sem);
2414         } else
2415                 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
2416         nfs_dentry_remove_handle_error(dir, dentry, error);
2417         trace_nfs_rmdir_exit(dir, dentry, error);
2418
2419         return error;
2420 }
2421 EXPORT_SYMBOL_GPL(nfs_rmdir);
2422
2423 /*
2424  * Remove a file after making sure there are no pending writes,
2425  * and after checking that the file has only one user.
2426  *
2427  * We invalidate the attribute cache and free the inode prior to the operation
2428  * to avoid possible races if the server reuses the inode.
2429  */
2430 static int nfs_safe_remove(struct dentry *dentry)
2431 {
2432         struct inode *dir = d_inode(dentry->d_parent);
2433         struct inode *inode = d_inode(dentry);
2434         int error = -EBUSY;
2435
2436         dfprintk(VFS, "NFS: safe_remove(%pd2)\n", dentry);
2437
2438         /* If the dentry was sillyrenamed, we simply call d_delete() */
2439         if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
2440                 error = 0;
2441                 goto out;
2442         }
2443
2444         trace_nfs_remove_enter(dir, dentry);
2445         if (inode != NULL) {
2446                 error = NFS_PROTO(dir)->remove(dir, dentry);
2447                 if (error == 0)
2448                         nfs_drop_nlink(inode);
2449         } else
2450                 error = NFS_PROTO(dir)->remove(dir, dentry);
2451         if (error == -ENOENT)
2452                 nfs_dentry_handle_enoent(dentry);
2453         trace_nfs_remove_exit(dir, dentry, error);
2454 out:
2455         return error;
2456 }
2457
2458 /*  We do silly rename. In case sillyrename() returns -EBUSY, the inode
2459  *  belongs to an active ".nfs..." file and we return -EBUSY.
2460  *
2461  *  If sillyrename() returns 0, we do nothing, otherwise we unlink.
2462  */
2463 int nfs_unlink(struct inode *dir, struct dentry *dentry)
2464 {
2465         int error;
2466
2467         dfprintk(VFS, "NFS: unlink(%s/%lu, %pd)\n", dir->i_sb->s_id,
2468                 dir->i_ino, dentry);
2469
2470         trace_nfs_unlink_enter(dir, dentry);
2471         spin_lock(&dentry->d_lock);
2472         if (d_count(dentry) > 1 && !test_bit(NFS_INO_PRESERVE_UNLINKED,
2473                                              &NFS_I(d_inode(dentry))->flags)) {
2474                 spin_unlock(&dentry->d_lock);
2475                 /* Start asynchronous writeout of the inode */
2476                 write_inode_now(d_inode(dentry), 0);
2477                 error = nfs_sillyrename(dir, dentry);
2478                 goto out;
2479         }
2480         /* We must prevent any concurrent open until the unlink
2481          * completes.  ->d_revalidate will wait for ->d_fsdata
2482          * to clear.  We set it here to ensure no lookup succeeds until
2483          * the unlink is complete on the server.
2484          */
2485         error = -ETXTBSY;
2486         if (WARN_ON(dentry->d_flags & DCACHE_NFSFS_RENAMED) ||
2487             WARN_ON(dentry->d_fsdata == NFS_FSDATA_BLOCKED))
2488                 goto out;
2489         if (dentry->d_fsdata)
2490                 /* old devname */
2491                 kfree(dentry->d_fsdata);
2492         dentry->d_fsdata = NFS_FSDATA_BLOCKED;
2493
2494         spin_unlock(&dentry->d_lock);
2495         error = nfs_safe_remove(dentry);
2496         nfs_dentry_remove_handle_error(dir, dentry, error);
2497         dentry->d_fsdata = NULL;
2498         wake_up_var(&dentry->d_fsdata);
2499 out:
2500         trace_nfs_unlink_exit(dir, dentry, error);
2501         return error;
2502 }
2503 EXPORT_SYMBOL_GPL(nfs_unlink);
2504
2505 /*
2506  * To create a symbolic link, most file systems instantiate a new inode,
2507  * add a page to it containing the path, then write it out to the disk
2508  * using prepare_write/commit_write.
2509  *
2510  * Unfortunately the NFS client can't create the in-core inode first
2511  * because it needs a file handle to create an in-core inode (see
2512  * fs/nfs/inode.c:nfs_fhget).  We only have a file handle *after* the
2513  * symlink request has completed on the server.
2514  *
2515  * So instead we allocate a raw page, copy the symname into it, then do
2516  * the SYMLINK request with the page as the buffer.  If it succeeds, we
2517  * now have a new file handle and can instantiate an in-core NFS inode
2518  * and move the raw page into its mapping.
2519  */
2520 int nfs_symlink(struct user_namespace *mnt_userns, struct inode *dir,
2521                 struct dentry *dentry, const char *symname)
2522 {
2523         struct page *page;
2524         char *kaddr;
2525         struct iattr attr;
2526         unsigned int pathlen = strlen(symname);
2527         int error;
2528
2529         dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s)\n", dir->i_sb->s_id,
2530                 dir->i_ino, dentry, symname);
2531
2532         if (pathlen > PAGE_SIZE)
2533                 return -ENAMETOOLONG;
2534
2535         attr.ia_mode = S_IFLNK | S_IRWXUGO;
2536         attr.ia_valid = ATTR_MODE;
2537
2538         page = alloc_page(GFP_USER);
2539         if (!page)
2540                 return -ENOMEM;
2541
2542         kaddr = page_address(page);
2543         memcpy(kaddr, symname, pathlen);
2544         if (pathlen < PAGE_SIZE)
2545                 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
2546
2547         trace_nfs_symlink_enter(dir, dentry);
2548         error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
2549         trace_nfs_symlink_exit(dir, dentry, error);
2550         if (error != 0) {
2551                 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s) error %d\n",
2552                         dir->i_sb->s_id, dir->i_ino,
2553                         dentry, symname, error);
2554                 d_drop(dentry);
2555                 __free_page(page);
2556                 return error;
2557         }
2558
2559         nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2560
2561         /*
2562          * No big deal if we can't add this page to the page cache here.
2563          * READLINK will get the missing page from the server if needed.
2564          */
2565         if (!add_to_page_cache_lru(page, d_inode(dentry)->i_mapping, 0,
2566                                                         GFP_KERNEL)) {
2567                 SetPageUptodate(page);
2568                 unlock_page(page);
2569                 /*
2570                  * add_to_page_cache_lru() grabs an extra page refcount.
2571                  * Drop it here to avoid leaking this page later.
2572                  */
2573                 put_page(page);
2574         } else
2575                 __free_page(page);
2576
2577         return 0;
2578 }
2579 EXPORT_SYMBOL_GPL(nfs_symlink);
2580
2581 int
2582 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2583 {
2584         struct inode *inode = d_inode(old_dentry);
2585         int error;
2586
2587         dfprintk(VFS, "NFS: link(%pd2 -> %pd2)\n",
2588                 old_dentry, dentry);
2589
2590         trace_nfs_link_enter(inode, dir, dentry);
2591         d_drop(dentry);
2592         if (S_ISREG(inode->i_mode))
2593                 nfs_sync_inode(inode);
2594         error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
2595         if (error == 0) {
2596                 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2597                 ihold(inode);
2598                 d_add(dentry, inode);
2599         }
2600         trace_nfs_link_exit(inode, dir, dentry, error);
2601         return error;
2602 }
2603 EXPORT_SYMBOL_GPL(nfs_link);
2604
2605 static void
2606 nfs_unblock_rename(struct rpc_task *task, struct nfs_renamedata *data)
2607 {
2608         struct dentry *new_dentry = data->new_dentry;
2609
2610         new_dentry->d_fsdata = NULL;
2611         wake_up_var(&new_dentry->d_fsdata);
2612 }
2613
2614 /*
2615  * RENAME
2616  * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
2617  * different file handle for the same inode after a rename (e.g. when
2618  * moving to a different directory). A fail-safe method to do so would
2619  * be to look up old_dir/old_name, create a link to new_dir/new_name and
2620  * rename the old file using the sillyrename stuff. This way, the original
2621  * file in old_dir will go away when the last process iput()s the inode.
2622  *
2623  * FIXED.
2624  *
2625  * It actually works quite well. One needs to have the possibility for
2626  * at least one ".nfs..." file in each directory the file ever gets
2627  * moved or linked to which happens automagically with the new
2628  * implementation that only depends on the dcache stuff instead of
2629  * using the inode layer
2630  *
2631  * Unfortunately, things are a little more complicated than indicated
2632  * above. For a cross-directory move, we want to make sure we can get
2633  * rid of the old inode after the operation.  This means there must be
2634  * no pending writes (if it's a file), and the use count must be 1.
2635  * If these conditions are met, we can drop the dentries before doing
2636  * the rename.
2637  */
2638 int nfs_rename(struct user_namespace *mnt_userns, struct inode *old_dir,
2639                struct dentry *old_dentry, struct inode *new_dir,
2640                struct dentry *new_dentry, unsigned int flags)
2641 {
2642         struct inode *old_inode = d_inode(old_dentry);
2643         struct inode *new_inode = d_inode(new_dentry);
2644         struct dentry *dentry = NULL;
2645         struct rpc_task *task;
2646         bool must_unblock = false;
2647         int error = -EBUSY;
2648
2649         if (flags)
2650                 return -EINVAL;
2651
2652         dfprintk(VFS, "NFS: rename(%pd2 -> %pd2, ct=%d)\n",
2653                  old_dentry, new_dentry,
2654                  d_count(new_dentry));
2655
2656         trace_nfs_rename_enter(old_dir, old_dentry, new_dir, new_dentry);
2657         /*
2658          * For non-directories, check whether the target is busy and if so,
2659          * make a copy of the dentry and then do a silly-rename. If the
2660          * silly-rename succeeds, the copied dentry is hashed and becomes
2661          * the new target.
2662          */
2663         if (new_inode && !S_ISDIR(new_inode->i_mode)) {
2664                 /* We must prevent any concurrent open until the unlink
2665                  * completes.  ->d_revalidate will wait for ->d_fsdata
2666                  * to clear.  We set it here to ensure no lookup succeeds until
2667                  * the unlink is complete on the server.
2668                  */
2669                 error = -ETXTBSY;
2670                 if (WARN_ON(new_dentry->d_flags & DCACHE_NFSFS_RENAMED) ||
2671                     WARN_ON(new_dentry->d_fsdata == NFS_FSDATA_BLOCKED))
2672                         goto out;
2673                 if (new_dentry->d_fsdata) {
2674                         /* old devname */
2675                         kfree(new_dentry->d_fsdata);
2676                         new_dentry->d_fsdata = NULL;
2677                 }
2678
2679                 spin_lock(&new_dentry->d_lock);
2680                 if (d_count(new_dentry) > 2) {
2681                         int err;
2682
2683                         spin_unlock(&new_dentry->d_lock);
2684
2685                         /* copy the target dentry's name */
2686                         dentry = d_alloc(new_dentry->d_parent,
2687                                          &new_dentry->d_name);
2688                         if (!dentry)
2689                                 goto out;
2690
2691                         /* silly-rename the existing target ... */
2692                         err = nfs_sillyrename(new_dir, new_dentry);
2693                         if (err)
2694                                 goto out;
2695
2696                         new_dentry = dentry;
2697                         new_inode = NULL;
2698                 } else {
2699                         new_dentry->d_fsdata = NFS_FSDATA_BLOCKED;
2700                         must_unblock = true;
2701                         spin_unlock(&new_dentry->d_lock);
2702                 }
2703
2704         }
2705
2706         if (S_ISREG(old_inode->i_mode))
2707                 nfs_sync_inode(old_inode);
2708         task = nfs_async_rename(old_dir, new_dir, old_dentry, new_dentry,
2709                                 must_unblock ? nfs_unblock_rename : NULL);
2710         if (IS_ERR(task)) {
2711                 error = PTR_ERR(task);
2712                 goto out;
2713         }
2714
2715         error = rpc_wait_for_completion_task(task);
2716         if (error != 0) {
2717                 ((struct nfs_renamedata *)task->tk_calldata)->cancelled = 1;
2718                 /* Paired with the atomic_dec_and_test() barrier in rpc_do_put_task() */
2719                 smp_wmb();
2720         } else
2721                 error = task->tk_status;
2722         rpc_put_task(task);
2723         /* Ensure the inode attributes are revalidated */
2724         if (error == 0) {
2725                 spin_lock(&old_inode->i_lock);
2726                 NFS_I(old_inode)->attr_gencount = nfs_inc_attr_generation_counter();
2727                 nfs_set_cache_invalid(old_inode, NFS_INO_INVALID_CHANGE |
2728                                                          NFS_INO_INVALID_CTIME |
2729                                                          NFS_INO_REVAL_FORCED);
2730                 spin_unlock(&old_inode->i_lock);
2731         }
2732 out:
2733         trace_nfs_rename_exit(old_dir, old_dentry,
2734                         new_dir, new_dentry, error);
2735         if (!error) {
2736                 if (new_inode != NULL)
2737                         nfs_drop_nlink(new_inode);
2738                 /*
2739                  * The d_move() should be here instead of in an async RPC completion
2740                  * handler because we need the proper locks to move the dentry.  If
2741                  * we're interrupted by a signal, the async RPC completion handler
2742                  * should mark the directories for revalidation.
2743                  */
2744                 d_move(old_dentry, new_dentry);
2745                 nfs_set_verifier(old_dentry,
2746                                         nfs_save_change_attribute(new_dir));
2747         } else if (error == -ENOENT)
2748                 nfs_dentry_handle_enoent(old_dentry);
2749
2750         /* new dentry created? */
2751         if (dentry)
2752                 dput(dentry);
2753         return error;
2754 }
2755 EXPORT_SYMBOL_GPL(nfs_rename);
2756
2757 static DEFINE_SPINLOCK(nfs_access_lru_lock);
2758 static LIST_HEAD(nfs_access_lru_list);
2759 static atomic_long_t nfs_access_nr_entries;
2760
2761 static unsigned long nfs_access_max_cachesize = 4*1024*1024;
2762 module_param(nfs_access_max_cachesize, ulong, 0644);
2763 MODULE_PARM_DESC(nfs_access_max_cachesize, "NFS access maximum total cache length");
2764
2765 static void nfs_access_free_entry(struct nfs_access_entry *entry)
2766 {
2767         put_group_info(entry->group_info);
2768         kfree_rcu(entry, rcu_head);
2769         smp_mb__before_atomic();
2770         atomic_long_dec(&nfs_access_nr_entries);
2771         smp_mb__after_atomic();
2772 }
2773
2774 static void nfs_access_free_list(struct list_head *head)
2775 {
2776         struct nfs_access_entry *cache;
2777
2778         while (!list_empty(head)) {
2779                 cache = list_entry(head->next, struct nfs_access_entry, lru);
2780                 list_del(&cache->lru);
2781                 nfs_access_free_entry(cache);
2782         }
2783 }
2784
2785 static unsigned long
2786 nfs_do_access_cache_scan(unsigned int nr_to_scan)
2787 {
2788         LIST_HEAD(head);
2789         struct nfs_inode *nfsi, *next;
2790         struct nfs_access_entry *cache;
2791         long freed = 0;
2792
2793         spin_lock(&nfs_access_lru_lock);
2794         list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) {
2795                 struct inode *inode;
2796
2797                 if (nr_to_scan-- == 0)
2798                         break;
2799                 inode = &nfsi->vfs_inode;
2800                 spin_lock(&inode->i_lock);
2801                 if (list_empty(&nfsi->access_cache_entry_lru))
2802                         goto remove_lru_entry;
2803                 cache = list_entry(nfsi->access_cache_entry_lru.next,
2804                                 struct nfs_access_entry, lru);
2805                 list_move(&cache->lru, &head);
2806                 rb_erase(&cache->rb_node, &nfsi->access_cache);
2807                 freed++;
2808                 if (!list_empty(&nfsi->access_cache_entry_lru))
2809                         list_move_tail(&nfsi->access_cache_inode_lru,
2810                                         &nfs_access_lru_list);
2811                 else {
2812 remove_lru_entry:
2813                         list_del_init(&nfsi->access_cache_inode_lru);
2814                         smp_mb__before_atomic();
2815                         clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
2816                         smp_mb__after_atomic();
2817                 }
2818                 spin_unlock(&inode->i_lock);
2819         }
2820         spin_unlock(&nfs_access_lru_lock);
2821         nfs_access_free_list(&head);
2822         return freed;
2823 }
2824
2825 unsigned long
2826 nfs_access_cache_scan(struct shrinker *shrink, struct shrink_control *sc)
2827 {
2828         int nr_to_scan = sc->nr_to_scan;
2829         gfp_t gfp_mask = sc->gfp_mask;
2830
2831         if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL)
2832                 return SHRINK_STOP;
2833         return nfs_do_access_cache_scan(nr_to_scan);
2834 }
2835
2836
2837 unsigned long
2838 nfs_access_cache_count(struct shrinker *shrink, struct shrink_control *sc)
2839 {
2840         return vfs_pressure_ratio(atomic_long_read(&nfs_access_nr_entries));
2841 }
2842
2843 static void
2844 nfs_access_cache_enforce_limit(void)
2845 {
2846         long nr_entries = atomic_long_read(&nfs_access_nr_entries);
2847         unsigned long diff;
2848         unsigned int nr_to_scan;
2849
2850         if (nr_entries < 0 || nr_entries <= nfs_access_max_cachesize)
2851                 return;
2852         nr_to_scan = 100;
2853         diff = nr_entries - nfs_access_max_cachesize;
2854         if (diff < nr_to_scan)
2855                 nr_to_scan = diff;
2856         nfs_do_access_cache_scan(nr_to_scan);
2857 }
2858
2859 static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head)
2860 {
2861         struct rb_root *root_node = &nfsi->access_cache;
2862         struct rb_node *n;
2863         struct nfs_access_entry *entry;
2864
2865         /* Unhook entries from the cache */
2866         while ((n = rb_first(root_node)) != NULL) {
2867                 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2868                 rb_erase(n, root_node);
2869                 list_move(&entry->lru, head);
2870         }
2871         nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
2872 }
2873
2874 void nfs_access_zap_cache(struct inode *inode)
2875 {
2876         LIST_HEAD(head);
2877
2878         if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0)
2879                 return;
2880         /* Remove from global LRU init */
2881         spin_lock(&nfs_access_lru_lock);
2882         if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2883                 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
2884
2885         spin_lock(&inode->i_lock);
2886         __nfs_access_zap_cache(NFS_I(inode), &head);
2887         spin_unlock(&inode->i_lock);
2888         spin_unlock(&nfs_access_lru_lock);
2889         nfs_access_free_list(&head);
2890 }
2891 EXPORT_SYMBOL_GPL(nfs_access_zap_cache);
2892
2893 static int access_cmp(const struct cred *a, const struct nfs_access_entry *b)
2894 {
2895         struct group_info *ga, *gb;
2896         int g;
2897
2898         if (uid_lt(a->fsuid, b->fsuid))
2899                 return -1;
2900         if (uid_gt(a->fsuid, b->fsuid))
2901                 return 1;
2902
2903         if (gid_lt(a->fsgid, b->fsgid))
2904                 return -1;
2905         if (gid_gt(a->fsgid, b->fsgid))
2906                 return 1;
2907
2908         ga = a->group_info;
2909         gb = b->group_info;
2910         if (ga == gb)
2911                 return 0;
2912         if (ga == NULL)
2913                 return -1;
2914         if (gb == NULL)
2915                 return 1;
2916         if (ga->ngroups < gb->ngroups)
2917                 return -1;
2918         if (ga->ngroups > gb->ngroups)
2919                 return 1;
2920
2921         for (g = 0; g < ga->ngroups; g++) {
2922                 if (gid_lt(ga->gid[g], gb->gid[g]))
2923                         return -1;
2924                 if (gid_gt(ga->gid[g], gb->gid[g]))
2925                         return 1;
2926         }
2927         return 0;
2928 }
2929
2930 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, const struct cred *cred)
2931 {
2932         struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
2933
2934         while (n != NULL) {
2935                 struct nfs_access_entry *entry =
2936                         rb_entry(n, struct nfs_access_entry, rb_node);
2937                 int cmp = access_cmp(cred, entry);
2938
2939                 if (cmp < 0)
2940                         n = n->rb_left;
2941                 else if (cmp > 0)
2942                         n = n->rb_right;
2943                 else
2944                         return entry;
2945         }
2946         return NULL;
2947 }
2948
2949 static int nfs_access_get_cached_locked(struct inode *inode, const struct cred *cred, u32 *mask, bool may_block)
2950 {
2951         struct nfs_inode *nfsi = NFS_I(inode);
2952         struct nfs_access_entry *cache;
2953         bool retry = true;
2954         int err;
2955
2956         spin_lock(&inode->i_lock);
2957         for(;;) {
2958                 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2959                         goto out_zap;
2960                 cache = nfs_access_search_rbtree(inode, cred);
2961                 err = -ENOENT;
2962                 if (cache == NULL)
2963                         goto out;
2964                 /* Found an entry, is our attribute cache valid? */
2965                 if (!nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS))
2966                         break;
2967                 if (!retry)
2968                         break;
2969                 err = -ECHILD;
2970                 if (!may_block)
2971                         goto out;
2972                 spin_unlock(&inode->i_lock);
2973                 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
2974                 if (err)
2975                         return err;
2976                 spin_lock(&inode->i_lock);
2977                 retry = false;
2978         }
2979         *mask = cache->mask;
2980         list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
2981         err = 0;
2982 out:
2983         spin_unlock(&inode->i_lock);
2984         return err;
2985 out_zap:
2986         spin_unlock(&inode->i_lock);
2987         nfs_access_zap_cache(inode);
2988         return -ENOENT;
2989 }
2990
2991 static int nfs_access_get_cached_rcu(struct inode *inode, const struct cred *cred, u32 *mask)
2992 {
2993         /* Only check the most recently returned cache entry,
2994          * but do it without locking.
2995          */
2996         struct nfs_inode *nfsi = NFS_I(inode);
2997         struct nfs_access_entry *cache;
2998         int err = -ECHILD;
2999         struct list_head *lh;
3000
3001         rcu_read_lock();
3002         if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
3003                 goto out;
3004         lh = rcu_dereference(list_tail_rcu(&nfsi->access_cache_entry_lru));
3005         cache = list_entry(lh, struct nfs_access_entry, lru);
3006         if (lh == &nfsi->access_cache_entry_lru ||
3007             access_cmp(cred, cache) != 0)
3008                 cache = NULL;
3009         if (cache == NULL)
3010                 goto out;
3011         if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS))
3012                 goto out;
3013         *mask = cache->mask;
3014         err = 0;
3015 out:
3016         rcu_read_unlock();
3017         return err;
3018 }
3019
3020 int nfs_access_get_cached(struct inode *inode, const struct cred *cred,
3021                           u32 *mask, bool may_block)
3022 {
3023         int status;
3024
3025         status = nfs_access_get_cached_rcu(inode, cred, mask);
3026         if (status != 0)
3027                 status = nfs_access_get_cached_locked(inode, cred, mask,
3028                     may_block);
3029
3030         return status;
3031 }
3032 EXPORT_SYMBOL_GPL(nfs_access_get_cached);
3033
3034 static void nfs_access_add_rbtree(struct inode *inode,
3035                                   struct nfs_access_entry *set,
3036                                   const struct cred *cred)
3037 {
3038         struct nfs_inode *nfsi = NFS_I(inode);
3039         struct rb_root *root_node = &nfsi->access_cache;
3040         struct rb_node **p = &root_node->rb_node;
3041         struct rb_node *parent = NULL;
3042         struct nfs_access_entry *entry;
3043         int cmp;
3044
3045         spin_lock(&inode->i_lock);
3046         while (*p != NULL) {
3047                 parent = *p;
3048                 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
3049                 cmp = access_cmp(cred, entry);
3050
3051                 if (cmp < 0)
3052                         p = &parent->rb_left;
3053                 else if (cmp > 0)
3054                         p = &parent->rb_right;
3055                 else
3056                         goto found;
3057         }
3058         rb_link_node(&set->rb_node, parent, p);
3059         rb_insert_color(&set->rb_node, root_node);
3060         list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
3061         spin_unlock(&inode->i_lock);
3062         return;
3063 found:
3064         rb_replace_node(parent, &set->rb_node, root_node);
3065         list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
3066         list_del(&entry->lru);
3067         spin_unlock(&inode->i_lock);
3068         nfs_access_free_entry(entry);
3069 }
3070
3071 void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set,
3072                           const struct cred *cred)
3073 {
3074         struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
3075         if (cache == NULL)
3076                 return;
3077         RB_CLEAR_NODE(&cache->rb_node);
3078         cache->fsuid = cred->fsuid;
3079         cache->fsgid = cred->fsgid;
3080         cache->group_info = get_group_info(cred->group_info);
3081         cache->mask = set->mask;
3082
3083         /* The above field assignments must be visible
3084          * before this item appears on the lru.  We cannot easily
3085          * use rcu_assign_pointer, so just force the memory barrier.
3086          */
3087         smp_wmb();
3088         nfs_access_add_rbtree(inode, cache, cred);
3089
3090         /* Update accounting */
3091         smp_mb__before_atomic();
3092         atomic_long_inc(&nfs_access_nr_entries);
3093         smp_mb__after_atomic();
3094
3095         /* Add inode to global LRU list */
3096         if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
3097                 spin_lock(&nfs_access_lru_lock);
3098                 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
3099                         list_add_tail(&NFS_I(inode)->access_cache_inode_lru,
3100                                         &nfs_access_lru_list);
3101                 spin_unlock(&nfs_access_lru_lock);
3102         }
3103         nfs_access_cache_enforce_limit();
3104 }
3105 EXPORT_SYMBOL_GPL(nfs_access_add_cache);
3106
3107 #define NFS_MAY_READ (NFS_ACCESS_READ)
3108 #define NFS_MAY_WRITE (NFS_ACCESS_MODIFY | \
3109                 NFS_ACCESS_EXTEND | \
3110                 NFS_ACCESS_DELETE)
3111 #define NFS_FILE_MAY_WRITE (NFS_ACCESS_MODIFY | \
3112                 NFS_ACCESS_EXTEND)
3113 #define NFS_DIR_MAY_WRITE NFS_MAY_WRITE
3114 #define NFS_MAY_LOOKUP (NFS_ACCESS_LOOKUP)
3115 #define NFS_MAY_EXECUTE (NFS_ACCESS_EXECUTE)
3116 static int
3117 nfs_access_calc_mask(u32 access_result, umode_t umode)
3118 {
3119         int mask = 0;
3120
3121         if (access_result & NFS_MAY_READ)
3122                 mask |= MAY_READ;
3123         if (S_ISDIR(umode)) {
3124                 if ((access_result & NFS_DIR_MAY_WRITE) == NFS_DIR_MAY_WRITE)
3125                         mask |= MAY_WRITE;
3126                 if ((access_result & NFS_MAY_LOOKUP) == NFS_MAY_LOOKUP)
3127                         mask |= MAY_EXEC;
3128         } else if (S_ISREG(umode)) {
3129                 if ((access_result & NFS_FILE_MAY_WRITE) == NFS_FILE_MAY_WRITE)
3130                         mask |= MAY_WRITE;
3131                 if ((access_result & NFS_MAY_EXECUTE) == NFS_MAY_EXECUTE)
3132                         mask |= MAY_EXEC;
3133         } else if (access_result & NFS_MAY_WRITE)
3134                         mask |= MAY_WRITE;
3135         return mask;
3136 }
3137
3138 void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result)
3139 {
3140         entry->mask = access_result;
3141 }
3142 EXPORT_SYMBOL_GPL(nfs_access_set_mask);
3143
3144 static int nfs_do_access(struct inode *inode, const struct cred *cred, int mask)
3145 {
3146         struct nfs_access_entry cache;
3147         bool may_block = (mask & MAY_NOT_BLOCK) == 0;
3148         int cache_mask = -1;
3149         int status;
3150
3151         trace_nfs_access_enter(inode);
3152
3153         status = nfs_access_get_cached(inode, cred, &cache.mask, may_block);
3154         if (status == 0)
3155                 goto out_cached;
3156
3157         status = -ECHILD;
3158         if (!may_block)
3159                 goto out;
3160
3161         /*
3162          * Determine which access bits we want to ask for...
3163          */
3164         cache.mask = NFS_ACCESS_READ | NFS_ACCESS_MODIFY | NFS_ACCESS_EXTEND |
3165                      nfs_access_xattr_mask(NFS_SERVER(inode));
3166         if (S_ISDIR(inode->i_mode))
3167                 cache.mask |= NFS_ACCESS_DELETE | NFS_ACCESS_LOOKUP;
3168         else
3169                 cache.mask |= NFS_ACCESS_EXECUTE;
3170         status = NFS_PROTO(inode)->access(inode, &cache, cred);
3171         if (status != 0) {
3172                 if (status == -ESTALE) {
3173                         if (!S_ISDIR(inode->i_mode))
3174                                 nfs_set_inode_stale(inode);
3175                         else
3176                                 nfs_zap_caches(inode);
3177                 }
3178                 goto out;
3179         }
3180         nfs_access_add_cache(inode, &cache, cred);
3181 out_cached:
3182         cache_mask = nfs_access_calc_mask(cache.mask, inode->i_mode);
3183         if ((mask & ~cache_mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) != 0)
3184                 status = -EACCES;
3185 out:
3186         trace_nfs_access_exit(inode, mask, cache_mask, status);
3187         return status;
3188 }
3189
3190 static int nfs_open_permission_mask(int openflags)
3191 {
3192         int mask = 0;
3193
3194         if (openflags & __FMODE_EXEC) {
3195                 /* ONLY check exec rights */
3196                 mask = MAY_EXEC;
3197         } else {
3198                 if ((openflags & O_ACCMODE) != O_WRONLY)
3199                         mask |= MAY_READ;
3200                 if ((openflags & O_ACCMODE) != O_RDONLY)
3201                         mask |= MAY_WRITE;
3202         }
3203
3204         return mask;
3205 }
3206
3207 int nfs_may_open(struct inode *inode, const struct cred *cred, int openflags)
3208 {
3209         return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
3210 }
3211 EXPORT_SYMBOL_GPL(nfs_may_open);
3212
3213 static int nfs_execute_ok(struct inode *inode, int mask)
3214 {
3215         struct nfs_server *server = NFS_SERVER(inode);
3216         int ret = 0;
3217
3218         if (S_ISDIR(inode->i_mode))
3219                 return 0;
3220         if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_MODE)) {
3221                 if (mask & MAY_NOT_BLOCK)
3222                         return -ECHILD;
3223                 ret = __nfs_revalidate_inode(server, inode);
3224         }
3225         if (ret == 0 && !execute_ok(inode))
3226                 ret = -EACCES;
3227         return ret;
3228 }
3229
3230 int nfs_permission(struct user_namespace *mnt_userns,
3231                    struct inode *inode,
3232                    int mask)
3233 {
3234         const struct cred *cred = current_cred();
3235         int res = 0;
3236
3237         nfs_inc_stats(inode, NFSIOS_VFSACCESS);
3238
3239         if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
3240                 goto out;
3241         /* Is this sys_access() ? */
3242         if (mask & (MAY_ACCESS | MAY_CHDIR))
3243                 goto force_lookup;
3244
3245         switch (inode->i_mode & S_IFMT) {
3246                 case S_IFLNK:
3247                         goto out;
3248                 case S_IFREG:
3249                         if ((mask & MAY_OPEN) &&
3250                            nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN))
3251                                 return 0;
3252                         break;
3253                 case S_IFDIR:
3254                         /*
3255                          * Optimize away all write operations, since the server
3256                          * will check permissions when we perform the op.
3257                          */
3258                         if ((mask & MAY_WRITE) && !(mask & MAY_READ))
3259                                 goto out;
3260         }
3261
3262 force_lookup:
3263         if (!NFS_PROTO(inode)->access)
3264                 goto out_notsup;
3265
3266         res = nfs_do_access(inode, cred, mask);
3267 out:
3268         if (!res && (mask & MAY_EXEC))
3269                 res = nfs_execute_ok(inode, mask);
3270
3271         dfprintk(VFS, "NFS: permission(%s/%lu), mask=0x%x, res=%d\n",
3272                 inode->i_sb->s_id, inode->i_ino, mask, res);
3273         return res;
3274 out_notsup:
3275         if (mask & MAY_NOT_BLOCK)
3276                 return -ECHILD;
3277
3278         res = nfs_revalidate_inode(inode, NFS_INO_INVALID_MODE |
3279                                                   NFS_INO_INVALID_OTHER);
3280         if (res == 0)
3281                 res = generic_permission(&init_user_ns, inode, mask);
3282         goto out;
3283 }
3284 EXPORT_SYMBOL_GPL(nfs_permission);