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