root/fs/nfs/dir.c

/* [<][>][^][v][top][bottom][index][help] */

DEFINITIONS

This source file includes following definitions.
  1. alloc_nfs_open_dir_context
  2. put_nfs_open_dir_context
  3. nfs_opendir
  4. nfs_closedir
  5. nfs_readdir_init_array
  6. nfs_readdir_clear_array
  7. nfs_readdir_make_qstr
  8. nfs_readdir_add_to_array
  9. nfs_readdir_search_for_pos
  10. nfs_readdir_inode_mapping_valid
  11. nfs_readdir_search_for_cookie
  12. nfs_readdir_search_array
  13. nfs_readdir_xdr_filler
  14. xdr_decode
  15. nfs_same_file
  16. nfs_use_readdirplus
  17. nfs_advise_use_readdirplus
  18. nfs_force_use_readdirplus
  19. nfs_prime_dcache
  20. nfs_readdir_page_filler
  21. nfs_readdir_free_pages
  22. nfs_readdir_alloc_pages
  23. nfs_readdir_xdr_to_array
  24. nfs_readdir_filler
  25. cache_page_release
  26. get_cache_page
  27. find_and_lock_cache_page
  28. readdir_search_pagecache
  29. nfs_do_filldir
  30. uncached_readdir
  31. nfs_readdir
  32. nfs_llseek_dir
  33. nfs_fsync_dir
  34. nfs_force_lookup_revalidate
  35. nfs_check_verifier
  36. nfs_is_exclusive_create
  37. nfs_lookup_verify_inode
  38. nfs_neg_need_reval
  39. nfs_lookup_revalidate_done
  40. nfs_lookup_revalidate_negative
  41. nfs_lookup_revalidate_delegated
  42. nfs_lookup_revalidate_dentry
  43. nfs_do_lookup_revalidate
  44. __nfs_lookup_revalidate
  45. nfs_lookup_revalidate
  46. nfs_weak_revalidate
  47. nfs_dentry_delete
  48. nfs_drop_nlink
  49. nfs_dentry_iput
  50. nfs_d_release
  51. nfs_lookup
  52. flags_to_mode
  53. create_nfs_open_context
  54. do_open
  55. nfs_finish_open
  56. nfs_atomic_open
  57. nfs4_do_lookup_revalidate
  58. nfs4_lookup_revalidate
  59. nfs_add_or_obtain
  60. nfs_instantiate
  61. nfs_create
  62. nfs_mknod
  63. nfs_mkdir
  64. nfs_dentry_handle_enoent
  65. nfs_rmdir
  66. nfs_safe_remove
  67. nfs_unlink
  68. nfs_symlink
  69. nfs_link
  70. nfs_rename
  71. nfs_access_free_entry
  72. nfs_access_free_list
  73. nfs_do_access_cache_scan
  74. nfs_access_cache_scan
  75. nfs_access_cache_count
  76. nfs_access_cache_enforce_limit
  77. __nfs_access_zap_cache
  78. nfs_access_zap_cache
  79. nfs_access_search_rbtree
  80. nfs_access_get_cached
  81. nfs_access_get_cached_rcu
  82. nfs_access_add_rbtree
  83. nfs_access_add_cache
  84. nfs_access_calc_mask
  85. nfs_access_set_mask
  86. nfs_do_access
  87. nfs_open_permission_mask
  88. nfs_may_open
  89. nfs_execute_ok
  90. nfs_permission

   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/module.h>
  22 #include <linux/time.h>
  23 #include <linux/errno.h>
  24 #include <linux/stat.h>
  25 #include <linux/fcntl.h>
  26 #include <linux/string.h>
  27 #include <linux/kernel.h>
  28 #include <linux/slab.h>
  29 #include <linux/mm.h>
  30 #include <linux/sunrpc/clnt.h>
  31 #include <linux/nfs_fs.h>
  32 #include <linux/nfs_mount.h>
  33 #include <linux/pagemap.h>
  34 #include <linux/pagevec.h>
  35 #include <linux/namei.h>
  36 #include <linux/mount.h>
  37 #include <linux/swap.h>
  38 #include <linux/sched.h>
  39 #include <linux/kmemleak.h>
  40 #include <linux/xattr.h>
  41 
  42 #include "delegation.h"
  43 #include "iostat.h"
  44 #include "internal.h"
  45 #include "fscache.h"
  46 
  47 #include "nfstrace.h"
  48 
  49 /* #define NFS_DEBUG_VERBOSE 1 */
  50 
  51 static int nfs_opendir(struct inode *, struct file *);
  52 static int nfs_closedir(struct inode *, struct file *);
  53 static int nfs_readdir(struct file *, struct dir_context *);
  54 static int nfs_fsync_dir(struct file *, loff_t, loff_t, int);
  55 static loff_t nfs_llseek_dir(struct file *, loff_t, int);
  56 static void nfs_readdir_clear_array(struct page*);
  57 
  58 const struct file_operations nfs_dir_operations = {
  59         .llseek         = nfs_llseek_dir,
  60         .read           = generic_read_dir,
  61         .iterate        = nfs_readdir,
  62         .open           = nfs_opendir,
  63         .release        = nfs_closedir,
  64         .fsync          = nfs_fsync_dir,
  65 };
  66 
  67 const struct address_space_operations nfs_dir_aops = {
  68         .freepage = nfs_readdir_clear_array,
  69 };
  70 
  71 static struct nfs_open_dir_context *alloc_nfs_open_dir_context(struct inode *dir, const struct cred *cred)
  72 {
  73         struct nfs_inode *nfsi = NFS_I(dir);
  74         struct nfs_open_dir_context *ctx;
  75         ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
  76         if (ctx != NULL) {
  77                 ctx->duped = 0;
  78                 ctx->attr_gencount = nfsi->attr_gencount;
  79                 ctx->dir_cookie = 0;
  80                 ctx->dup_cookie = 0;
  81                 ctx->cred = get_cred(cred);
  82                 spin_lock(&dir->i_lock);
  83                 if (list_empty(&nfsi->open_files) &&
  84                     (nfsi->cache_validity & NFS_INO_DATA_INVAL_DEFER))
  85                         nfsi->cache_validity |= NFS_INO_INVALID_DATA |
  86                                 NFS_INO_REVAL_FORCED;
  87                 list_add(&ctx->list, &nfsi->open_files);
  88                 spin_unlock(&dir->i_lock);
  89                 return ctx;
  90         }
  91         return  ERR_PTR(-ENOMEM);
  92 }
  93 
  94 static void put_nfs_open_dir_context(struct inode *dir, struct nfs_open_dir_context *ctx)
  95 {
  96         spin_lock(&dir->i_lock);
  97         list_del(&ctx->list);
  98         spin_unlock(&dir->i_lock);
  99         put_cred(ctx->cred);
 100         kfree(ctx);
 101 }
 102 
 103 /*
 104  * Open file
 105  */
 106 static int
 107 nfs_opendir(struct inode *inode, struct file *filp)
 108 {
 109         int res = 0;
 110         struct nfs_open_dir_context *ctx;
 111 
 112         dfprintk(FILE, "NFS: open dir(%pD2)\n", filp);
 113 
 114         nfs_inc_stats(inode, NFSIOS_VFSOPEN);
 115 
 116         ctx = alloc_nfs_open_dir_context(inode, current_cred());
 117         if (IS_ERR(ctx)) {
 118                 res = PTR_ERR(ctx);
 119                 goto out;
 120         }
 121         filp->private_data = ctx;
 122 out:
 123         return res;
 124 }
 125 
 126 static int
 127 nfs_closedir(struct inode *inode, struct file *filp)
 128 {
 129         put_nfs_open_dir_context(file_inode(filp), filp->private_data);
 130         return 0;
 131 }
 132 
 133 struct nfs_cache_array_entry {
 134         u64 cookie;
 135         u64 ino;
 136         struct qstr string;
 137         unsigned char d_type;
 138 };
 139 
 140 struct nfs_cache_array {
 141         int size;
 142         int eof_index;
 143         u64 last_cookie;
 144         struct nfs_cache_array_entry array[0];
 145 };
 146 
 147 typedef int (*decode_dirent_t)(struct xdr_stream *, struct nfs_entry *, bool);
 148 typedef struct {
 149         struct file     *file;
 150         struct page     *page;
 151         struct dir_context *ctx;
 152         unsigned long   page_index;
 153         u64             *dir_cookie;
 154         u64             last_cookie;
 155         loff_t          current_index;
 156         decode_dirent_t decode;
 157 
 158         unsigned long   timestamp;
 159         unsigned long   gencount;
 160         unsigned int    cache_entry_index;
 161         bool plus;
 162         bool eof;
 163 } nfs_readdir_descriptor_t;
 164 
 165 static
 166 void nfs_readdir_init_array(struct page *page)
 167 {
 168         struct nfs_cache_array *array;
 169 
 170         array = kmap_atomic(page);
 171         memset(array, 0, sizeof(struct nfs_cache_array));
 172         array->eof_index = -1;
 173         kunmap_atomic(array);
 174 }
 175 
 176 /*
 177  * we are freeing strings created by nfs_add_to_readdir_array()
 178  */
 179 static
 180 void nfs_readdir_clear_array(struct page *page)
 181 {
 182         struct nfs_cache_array *array;
 183         int i;
 184 
 185         array = kmap_atomic(page);
 186         for (i = 0; i < array->size; i++)
 187                 kfree(array->array[i].string.name);
 188         array->size = 0;
 189         kunmap_atomic(array);
 190 }
 191 
 192 /*
 193  * the caller is responsible for freeing qstr.name
 194  * when called by nfs_readdir_add_to_array, the strings will be freed in
 195  * nfs_clear_readdir_array()
 196  */
 197 static
 198 int nfs_readdir_make_qstr(struct qstr *string, const char *name, unsigned int len)
 199 {
 200         string->len = len;
 201         string->name = kmemdup(name, len, GFP_KERNEL);
 202         if (string->name == NULL)
 203                 return -ENOMEM;
 204         /*
 205          * Avoid a kmemleak false positive. The pointer to the name is stored
 206          * in a page cache page which kmemleak does not scan.
 207          */
 208         kmemleak_not_leak(string->name);
 209         string->hash = full_name_hash(NULL, name, len);
 210         return 0;
 211 }
 212 
 213 static
 214 int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page)
 215 {
 216         struct nfs_cache_array *array = kmap(page);
 217         struct nfs_cache_array_entry *cache_entry;
 218         int ret;
 219 
 220         cache_entry = &array->array[array->size];
 221 
 222         /* Check that this entry lies within the page bounds */
 223         ret = -ENOSPC;
 224         if ((char *)&cache_entry[1] - (char *)page_address(page) > PAGE_SIZE)
 225                 goto out;
 226 
 227         cache_entry->cookie = entry->prev_cookie;
 228         cache_entry->ino = entry->ino;
 229         cache_entry->d_type = entry->d_type;
 230         ret = nfs_readdir_make_qstr(&cache_entry->string, entry->name, entry->len);
 231         if (ret)
 232                 goto out;
 233         array->last_cookie = entry->cookie;
 234         array->size++;
 235         if (entry->eof != 0)
 236                 array->eof_index = array->size;
 237 out:
 238         kunmap(page);
 239         return ret;
 240 }
 241 
 242 static
 243 int nfs_readdir_search_for_pos(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
 244 {
 245         loff_t diff = desc->ctx->pos - desc->current_index;
 246         unsigned int index;
 247 
 248         if (diff < 0)
 249                 goto out_eof;
 250         if (diff >= array->size) {
 251                 if (array->eof_index >= 0)
 252                         goto out_eof;
 253                 return -EAGAIN;
 254         }
 255 
 256         index = (unsigned int)diff;
 257         *desc->dir_cookie = array->array[index].cookie;
 258         desc->cache_entry_index = index;
 259         return 0;
 260 out_eof:
 261         desc->eof = true;
 262         return -EBADCOOKIE;
 263 }
 264 
 265 static bool
 266 nfs_readdir_inode_mapping_valid(struct nfs_inode *nfsi)
 267 {
 268         if (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA))
 269                 return false;
 270         smp_rmb();
 271         return !test_bit(NFS_INO_INVALIDATING, &nfsi->flags);
 272 }
 273 
 274 static
 275 int nfs_readdir_search_for_cookie(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
 276 {
 277         int i;
 278         loff_t new_pos;
 279         int status = -EAGAIN;
 280 
 281         for (i = 0; i < array->size; i++) {
 282                 if (array->array[i].cookie == *desc->dir_cookie) {
 283                         struct nfs_inode *nfsi = NFS_I(file_inode(desc->file));
 284                         struct nfs_open_dir_context *ctx = desc->file->private_data;
 285 
 286                         new_pos = desc->current_index + i;
 287                         if (ctx->attr_gencount != nfsi->attr_gencount ||
 288                             !nfs_readdir_inode_mapping_valid(nfsi)) {
 289                                 ctx->duped = 0;
 290                                 ctx->attr_gencount = nfsi->attr_gencount;
 291                         } else if (new_pos < desc->ctx->pos) {
 292                                 if (ctx->duped > 0
 293                                     && ctx->dup_cookie == *desc->dir_cookie) {
 294                                         if (printk_ratelimit()) {
 295                                                 pr_notice("NFS: directory %pD2 contains a readdir loop."
 296                                                                 "Please contact your server vendor.  "
 297                                                                 "The file: %.*s has duplicate cookie %llu\n",
 298                                                                 desc->file, array->array[i].string.len,
 299                                                                 array->array[i].string.name, *desc->dir_cookie);
 300                                         }
 301                                         status = -ELOOP;
 302                                         goto out;
 303                                 }
 304                                 ctx->dup_cookie = *desc->dir_cookie;
 305                                 ctx->duped = -1;
 306                         }
 307                         desc->ctx->pos = new_pos;
 308                         desc->cache_entry_index = i;
 309                         return 0;
 310                 }
 311         }
 312         if (array->eof_index >= 0) {
 313                 status = -EBADCOOKIE;
 314                 if (*desc->dir_cookie == array->last_cookie)
 315                         desc->eof = true;
 316         }
 317 out:
 318         return status;
 319 }
 320 
 321 static
 322 int nfs_readdir_search_array(nfs_readdir_descriptor_t *desc)
 323 {
 324         struct nfs_cache_array *array;
 325         int status;
 326 
 327         array = kmap(desc->page);
 328 
 329         if (*desc->dir_cookie == 0)
 330                 status = nfs_readdir_search_for_pos(array, desc);
 331         else
 332                 status = nfs_readdir_search_for_cookie(array, desc);
 333 
 334         if (status == -EAGAIN) {
 335                 desc->last_cookie = array->last_cookie;
 336                 desc->current_index += array->size;
 337                 desc->page_index++;
 338         }
 339         kunmap(desc->page);
 340         return status;
 341 }
 342 
 343 /* Fill a page with xdr information before transferring to the cache page */
 344 static
 345 int nfs_readdir_xdr_filler(struct page **pages, nfs_readdir_descriptor_t *desc,
 346                         struct nfs_entry *entry, struct file *file, struct inode *inode)
 347 {
 348         struct nfs_open_dir_context *ctx = file->private_data;
 349         const struct cred *cred = ctx->cred;
 350         unsigned long   timestamp, gencount;
 351         int             error;
 352 
 353  again:
 354         timestamp = jiffies;
 355         gencount = nfs_inc_attr_generation_counter();
 356         error = NFS_PROTO(inode)->readdir(file_dentry(file), cred, entry->cookie, pages,
 357                                           NFS_SERVER(inode)->dtsize, desc->plus);
 358         if (error < 0) {
 359                 /* We requested READDIRPLUS, but the server doesn't grok it */
 360                 if (error == -ENOTSUPP && desc->plus) {
 361                         NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
 362                         clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
 363                         desc->plus = false;
 364                         goto again;
 365                 }
 366                 goto error;
 367         }
 368         desc->timestamp = timestamp;
 369         desc->gencount = gencount;
 370 error:
 371         return error;
 372 }
 373 
 374 static int xdr_decode(nfs_readdir_descriptor_t *desc,
 375                       struct nfs_entry *entry, struct xdr_stream *xdr)
 376 {
 377         int error;
 378 
 379         error = desc->decode(xdr, entry, desc->plus);
 380         if (error)
 381                 return error;
 382         entry->fattr->time_start = desc->timestamp;
 383         entry->fattr->gencount = desc->gencount;
 384         return 0;
 385 }
 386 
 387 /* Match file and dirent using either filehandle or fileid
 388  * Note: caller is responsible for checking the fsid
 389  */
 390 static
 391 int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry)
 392 {
 393         struct inode *inode;
 394         struct nfs_inode *nfsi;
 395 
 396         if (d_really_is_negative(dentry))
 397                 return 0;
 398 
 399         inode = d_inode(dentry);
 400         if (is_bad_inode(inode) || NFS_STALE(inode))
 401                 return 0;
 402 
 403         nfsi = NFS_I(inode);
 404         if (entry->fattr->fileid != nfsi->fileid)
 405                 return 0;
 406         if (entry->fh->size && nfs_compare_fh(entry->fh, &nfsi->fh) != 0)
 407                 return 0;
 408         return 1;
 409 }
 410 
 411 static
 412 bool nfs_use_readdirplus(struct inode *dir, struct dir_context *ctx)
 413 {
 414         if (!nfs_server_capable(dir, NFS_CAP_READDIRPLUS))
 415                 return false;
 416         if (test_and_clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags))
 417                 return true;
 418         if (ctx->pos == 0)
 419                 return true;
 420         return false;
 421 }
 422 
 423 /*
 424  * This function is called by the lookup and getattr code to request the
 425  * use of readdirplus to accelerate any future lookups in the same
 426  * directory.
 427  */
 428 void nfs_advise_use_readdirplus(struct inode *dir)
 429 {
 430         struct nfs_inode *nfsi = NFS_I(dir);
 431 
 432         if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) &&
 433             !list_empty(&nfsi->open_files))
 434                 set_bit(NFS_INO_ADVISE_RDPLUS, &nfsi->flags);
 435 }
 436 
 437 /*
 438  * This function is mainly for use by nfs_getattr().
 439  *
 440  * If this is an 'ls -l', we want to force use of readdirplus.
 441  * Do this by checking if there is an active file descriptor
 442  * and calling nfs_advise_use_readdirplus, then forcing a
 443  * cache flush.
 444  */
 445 void nfs_force_use_readdirplus(struct inode *dir)
 446 {
 447         struct nfs_inode *nfsi = NFS_I(dir);
 448 
 449         if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) &&
 450             !list_empty(&nfsi->open_files)) {
 451                 set_bit(NFS_INO_ADVISE_RDPLUS, &nfsi->flags);
 452                 invalidate_mapping_pages(dir->i_mapping, 0, -1);
 453         }
 454 }
 455 
 456 static
 457 void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry)
 458 {
 459         struct qstr filename = QSTR_INIT(entry->name, entry->len);
 460         DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
 461         struct dentry *dentry;
 462         struct dentry *alias;
 463         struct inode *dir = d_inode(parent);
 464         struct inode *inode;
 465         int status;
 466 
 467         if (!(entry->fattr->valid & NFS_ATTR_FATTR_FILEID))
 468                 return;
 469         if (!(entry->fattr->valid & NFS_ATTR_FATTR_FSID))
 470                 return;
 471         if (filename.len == 0)
 472                 return;
 473         /* Validate that the name doesn't contain any illegal '\0' */
 474         if (strnlen(filename.name, filename.len) != filename.len)
 475                 return;
 476         /* ...or '/' */
 477         if (strnchr(filename.name, filename.len, '/'))
 478                 return;
 479         if (filename.name[0] == '.') {
 480                 if (filename.len == 1)
 481                         return;
 482                 if (filename.len == 2 && filename.name[1] == '.')
 483                         return;
 484         }
 485         filename.hash = full_name_hash(parent, filename.name, filename.len);
 486 
 487         dentry = d_lookup(parent, &filename);
 488 again:
 489         if (!dentry) {
 490                 dentry = d_alloc_parallel(parent, &filename, &wq);
 491                 if (IS_ERR(dentry))
 492                         return;
 493         }
 494         if (!d_in_lookup(dentry)) {
 495                 /* Is there a mountpoint here? If so, just exit */
 496                 if (!nfs_fsid_equal(&NFS_SB(dentry->d_sb)->fsid,
 497                                         &entry->fattr->fsid))
 498                         goto out;
 499                 if (nfs_same_file(dentry, entry)) {
 500                         if (!entry->fh->size)
 501                                 goto out;
 502                         nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
 503                         status = nfs_refresh_inode(d_inode(dentry), entry->fattr);
 504                         if (!status)
 505                                 nfs_setsecurity(d_inode(dentry), entry->fattr, entry->label);
 506                         goto out;
 507                 } else {
 508                         d_invalidate(dentry);
 509                         dput(dentry);
 510                         dentry = NULL;
 511                         goto again;
 512                 }
 513         }
 514         if (!entry->fh->size) {
 515                 d_lookup_done(dentry);
 516                 goto out;
 517         }
 518 
 519         inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr, entry->label);
 520         alias = d_splice_alias(inode, dentry);
 521         d_lookup_done(dentry);
 522         if (alias) {
 523                 if (IS_ERR(alias))
 524                         goto out;
 525                 dput(dentry);
 526                 dentry = alias;
 527         }
 528         nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
 529 out:
 530         dput(dentry);
 531 }
 532 
 533 /* Perform conversion from xdr to cache array */
 534 static
 535 int nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry,
 536                                 struct page **xdr_pages, struct page *page, unsigned int buflen)
 537 {
 538         struct xdr_stream stream;
 539         struct xdr_buf buf;
 540         struct page *scratch;
 541         struct nfs_cache_array *array;
 542         unsigned int count = 0;
 543         int status;
 544 
 545         scratch = alloc_page(GFP_KERNEL);
 546         if (scratch == NULL)
 547                 return -ENOMEM;
 548 
 549         if (buflen == 0)
 550                 goto out_nopages;
 551 
 552         xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen);
 553         xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
 554 
 555         do {
 556                 status = xdr_decode(desc, entry, &stream);
 557                 if (status != 0) {
 558                         if (status == -EAGAIN)
 559                                 status = 0;
 560                         break;
 561                 }
 562 
 563                 count++;
 564 
 565                 if (desc->plus)
 566                         nfs_prime_dcache(file_dentry(desc->file), entry);
 567 
 568                 status = nfs_readdir_add_to_array(entry, page);
 569                 if (status != 0)
 570                         break;
 571         } while (!entry->eof);
 572 
 573 out_nopages:
 574         if (count == 0 || (status == -EBADCOOKIE && entry->eof != 0)) {
 575                 array = kmap(page);
 576                 array->eof_index = array->size;
 577                 status = 0;
 578                 kunmap(page);
 579         }
 580 
 581         put_page(scratch);
 582         return status;
 583 }
 584 
 585 static
 586 void nfs_readdir_free_pages(struct page **pages, unsigned int npages)
 587 {
 588         unsigned int i;
 589         for (i = 0; i < npages; i++)
 590                 put_page(pages[i]);
 591 }
 592 
 593 /*
 594  * nfs_readdir_alloc_pages() will allocate pages that must be freed with a call
 595  * to nfs_readdir_free_pages()
 596  */
 597 static
 598 int nfs_readdir_alloc_pages(struct page **pages, unsigned int npages)
 599 {
 600         unsigned int i;
 601 
 602         for (i = 0; i < npages; i++) {
 603                 struct page *page = alloc_page(GFP_KERNEL);
 604                 if (page == NULL)
 605                         goto out_freepages;
 606                 pages[i] = page;
 607         }
 608         return 0;
 609 
 610 out_freepages:
 611         nfs_readdir_free_pages(pages, i);
 612         return -ENOMEM;
 613 }
 614 
 615 static
 616 int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t *desc, struct page *page, struct inode *inode)
 617 {
 618         struct page *pages[NFS_MAX_READDIR_PAGES];
 619         struct nfs_entry entry;
 620         struct file     *file = desc->file;
 621         struct nfs_cache_array *array;
 622         int status = -ENOMEM;
 623         unsigned int array_size = ARRAY_SIZE(pages);
 624 
 625         nfs_readdir_init_array(page);
 626 
 627         entry.prev_cookie = 0;
 628         entry.cookie = desc->last_cookie;
 629         entry.eof = 0;
 630         entry.fh = nfs_alloc_fhandle();
 631         entry.fattr = nfs_alloc_fattr();
 632         entry.server = NFS_SERVER(inode);
 633         if (entry.fh == NULL || entry.fattr == NULL)
 634                 goto out;
 635 
 636         entry.label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT);
 637         if (IS_ERR(entry.label)) {
 638                 status = PTR_ERR(entry.label);
 639                 goto out;
 640         }
 641 
 642         array = kmap(page);
 643 
 644         status = nfs_readdir_alloc_pages(pages, array_size);
 645         if (status < 0)
 646                 goto out_release_array;
 647         do {
 648                 unsigned int pglen;
 649                 status = nfs_readdir_xdr_filler(pages, desc, &entry, file, inode);
 650 
 651                 if (status < 0)
 652                         break;
 653                 pglen = status;
 654                 status = nfs_readdir_page_filler(desc, &entry, pages, page, pglen);
 655                 if (status < 0) {
 656                         if (status == -ENOSPC)
 657                                 status = 0;
 658                         break;
 659                 }
 660         } while (array->eof_index < 0);
 661 
 662         nfs_readdir_free_pages(pages, array_size);
 663 out_release_array:
 664         kunmap(page);
 665         nfs4_label_free(entry.label);
 666 out:
 667         nfs_free_fattr(entry.fattr);
 668         nfs_free_fhandle(entry.fh);
 669         return status;
 670 }
 671 
 672 /*
 673  * Now we cache directories properly, by converting xdr information
 674  * to an array that can be used for lookups later.  This results in
 675  * fewer cache pages, since we can store more information on each page.
 676  * We only need to convert from xdr once so future lookups are much simpler
 677  */
 678 static
 679 int nfs_readdir_filler(void *data, struct page* page)
 680 {
 681         nfs_readdir_descriptor_t *desc = data;
 682         struct inode    *inode = file_inode(desc->file);
 683         int ret;
 684 
 685         ret = nfs_readdir_xdr_to_array(desc, page, inode);
 686         if (ret < 0)
 687                 goto error;
 688         SetPageUptodate(page);
 689 
 690         if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) {
 691                 /* Should never happen */
 692                 nfs_zap_mapping(inode, inode->i_mapping);
 693         }
 694         unlock_page(page);
 695         return 0;
 696  error:
 697         nfs_readdir_clear_array(page);
 698         unlock_page(page);
 699         return ret;
 700 }
 701 
 702 static
 703 void cache_page_release(nfs_readdir_descriptor_t *desc)
 704 {
 705         put_page(desc->page);
 706         desc->page = NULL;
 707 }
 708 
 709 static
 710 struct page *get_cache_page(nfs_readdir_descriptor_t *desc)
 711 {
 712         return read_cache_page(desc->file->f_mapping, desc->page_index,
 713                         nfs_readdir_filler, desc);
 714 }
 715 
 716 /*
 717  * Returns 0 if desc->dir_cookie was found on page desc->page_index
 718  * and locks the page to prevent removal from the page cache.
 719  */
 720 static
 721 int find_and_lock_cache_page(nfs_readdir_descriptor_t *desc)
 722 {
 723         int res;
 724 
 725         desc->page = get_cache_page(desc);
 726         if (IS_ERR(desc->page))
 727                 return PTR_ERR(desc->page);
 728         res = lock_page_killable(desc->page);
 729         if (res != 0)
 730                 goto error;
 731         res = -EAGAIN;
 732         if (desc->page->mapping != NULL) {
 733                 res = nfs_readdir_search_array(desc);
 734                 if (res == 0)
 735                         return 0;
 736         }
 737         unlock_page(desc->page);
 738 error:
 739         cache_page_release(desc);
 740         return res;
 741 }
 742 
 743 /* Search for desc->dir_cookie from the beginning of the page cache */
 744 static inline
 745 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
 746 {
 747         int res;
 748 
 749         if (desc->page_index == 0) {
 750                 desc->current_index = 0;
 751                 desc->last_cookie = 0;
 752         }
 753         do {
 754                 res = find_and_lock_cache_page(desc);
 755         } while (res == -EAGAIN);
 756         return res;
 757 }
 758 
 759 /*
 760  * Once we've found the start of the dirent within a page: fill 'er up...
 761  */
 762 static 
 763 int nfs_do_filldir(nfs_readdir_descriptor_t *desc)
 764 {
 765         struct file     *file = desc->file;
 766         int i = 0;
 767         int res = 0;
 768         struct nfs_cache_array *array = NULL;
 769         struct nfs_open_dir_context *ctx = file->private_data;
 770 
 771         array = kmap(desc->page);
 772         for (i = desc->cache_entry_index; i < array->size; i++) {
 773                 struct nfs_cache_array_entry *ent;
 774 
 775                 ent = &array->array[i];
 776                 if (!dir_emit(desc->ctx, ent->string.name, ent->string.len,
 777                     nfs_compat_user_ino64(ent->ino), ent->d_type)) {
 778                         desc->eof = true;
 779                         break;
 780                 }
 781                 desc->ctx->pos++;
 782                 if (i < (array->size-1))
 783                         *desc->dir_cookie = array->array[i+1].cookie;
 784                 else
 785                         *desc->dir_cookie = array->last_cookie;
 786                 if (ctx->duped != 0)
 787                         ctx->duped = 1;
 788         }
 789         if (array->eof_index >= 0)
 790                 desc->eof = true;
 791 
 792         kunmap(desc->page);
 793         dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
 794                         (unsigned long long)*desc->dir_cookie, res);
 795         return res;
 796 }
 797 
 798 /*
 799  * If we cannot find a cookie in our cache, we suspect that this is
 800  * because it points to a deleted file, so we ask the server to return
 801  * whatever it thinks is the next entry. We then feed this to filldir.
 802  * If all goes well, we should then be able to find our way round the
 803  * cache on the next call to readdir_search_pagecache();
 804  *
 805  * NOTE: we cannot add the anonymous page to the pagecache because
 806  *       the data it contains might not be page aligned. Besides,
 807  *       we should already have a complete representation of the
 808  *       directory in the page cache by the time we get here.
 809  */
 810 static inline
 811 int uncached_readdir(nfs_readdir_descriptor_t *desc)
 812 {
 813         struct page     *page = NULL;
 814         int             status;
 815         struct inode *inode = file_inode(desc->file);
 816         struct nfs_open_dir_context *ctx = desc->file->private_data;
 817 
 818         dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n",
 819                         (unsigned long long)*desc->dir_cookie);
 820 
 821         page = alloc_page(GFP_HIGHUSER);
 822         if (!page) {
 823                 status = -ENOMEM;
 824                 goto out;
 825         }
 826 
 827         desc->page_index = 0;
 828         desc->last_cookie = *desc->dir_cookie;
 829         desc->page = page;
 830         ctx->duped = 0;
 831 
 832         status = nfs_readdir_xdr_to_array(desc, page, inode);
 833         if (status < 0)
 834                 goto out_release;
 835 
 836         status = nfs_do_filldir(desc);
 837 
 838  out_release:
 839         nfs_readdir_clear_array(desc->page);
 840         cache_page_release(desc);
 841  out:
 842         dfprintk(DIRCACHE, "NFS: %s: returns %d\n",
 843                         __func__, status);
 844         return status;
 845 }
 846 
 847 /* The file offset position represents the dirent entry number.  A
 848    last cookie cache takes care of the common case of reading the
 849    whole directory.
 850  */
 851 static int nfs_readdir(struct file *file, struct dir_context *ctx)
 852 {
 853         struct dentry   *dentry = file_dentry(file);
 854         struct inode    *inode = d_inode(dentry);
 855         nfs_readdir_descriptor_t my_desc,
 856                         *desc = &my_desc;
 857         struct nfs_open_dir_context *dir_ctx = file->private_data;
 858         int res = 0;
 859 
 860         dfprintk(FILE, "NFS: readdir(%pD2) starting at cookie %llu\n",
 861                         file, (long long)ctx->pos);
 862         nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
 863 
 864         /*
 865          * ctx->pos points to the dirent entry number.
 866          * *desc->dir_cookie has the cookie for the next entry. We have
 867          * to either find the entry with the appropriate number or
 868          * revalidate the cookie.
 869          */
 870         memset(desc, 0, sizeof(*desc));
 871 
 872         desc->file = file;
 873         desc->ctx = ctx;
 874         desc->dir_cookie = &dir_ctx->dir_cookie;
 875         desc->decode = NFS_PROTO(inode)->decode_dirent;
 876         desc->plus = nfs_use_readdirplus(inode, ctx);
 877 
 878         if (ctx->pos == 0 || nfs_attribute_cache_expired(inode))
 879                 res = nfs_revalidate_mapping(inode, file->f_mapping);
 880         if (res < 0)
 881                 goto out;
 882 
 883         do {
 884                 res = readdir_search_pagecache(desc);
 885 
 886                 if (res == -EBADCOOKIE) {
 887                         res = 0;
 888                         /* This means either end of directory */
 889                         if (*desc->dir_cookie && !desc->eof) {
 890                                 /* Or that the server has 'lost' a cookie */
 891                                 res = uncached_readdir(desc);
 892                                 if (res == 0)
 893                                         continue;
 894                         }
 895                         break;
 896                 }
 897                 if (res == -ETOOSMALL && desc->plus) {
 898                         clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
 899                         nfs_zap_caches(inode);
 900                         desc->page_index = 0;
 901                         desc->plus = false;
 902                         desc->eof = false;
 903                         continue;
 904                 }
 905                 if (res < 0)
 906                         break;
 907 
 908                 res = nfs_do_filldir(desc);
 909                 unlock_page(desc->page);
 910                 cache_page_release(desc);
 911                 if (res < 0)
 912                         break;
 913         } while (!desc->eof);
 914 out:
 915         if (res > 0)
 916                 res = 0;
 917         dfprintk(FILE, "NFS: readdir(%pD2) returns %d\n", file, res);
 918         return res;
 919 }
 920 
 921 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int whence)
 922 {
 923         struct inode *inode = file_inode(filp);
 924         struct nfs_open_dir_context *dir_ctx = filp->private_data;
 925 
 926         dfprintk(FILE, "NFS: llseek dir(%pD2, %lld, %d)\n",
 927                         filp, offset, whence);
 928 
 929         switch (whence) {
 930         default:
 931                 return -EINVAL;
 932         case SEEK_SET:
 933                 if (offset < 0)
 934                         return -EINVAL;
 935                 inode_lock(inode);
 936                 break;
 937         case SEEK_CUR:
 938                 if (offset == 0)
 939                         return filp->f_pos;
 940                 inode_lock(inode);
 941                 offset += filp->f_pos;
 942                 if (offset < 0) {
 943                         inode_unlock(inode);
 944                         return -EINVAL;
 945                 }
 946         }
 947         if (offset != filp->f_pos) {
 948                 filp->f_pos = offset;
 949                 dir_ctx->dir_cookie = 0;
 950                 dir_ctx->duped = 0;
 951         }
 952         inode_unlock(inode);
 953         return offset;
 954 }
 955 
 956 /*
 957  * All directory operations under NFS are synchronous, so fsync()
 958  * is a dummy operation.
 959  */
 960 static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end,
 961                          int datasync)
 962 {
 963         struct inode *inode = file_inode(filp);
 964 
 965         dfprintk(FILE, "NFS: fsync dir(%pD2) datasync %d\n", filp, datasync);
 966 
 967         inode_lock(inode);
 968         nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
 969         inode_unlock(inode);
 970         return 0;
 971 }
 972 
 973 /**
 974  * nfs_force_lookup_revalidate - Mark the directory as having changed
 975  * @dir: pointer to directory inode
 976  *
 977  * This forces the revalidation code in nfs_lookup_revalidate() to do a
 978  * full lookup on all child dentries of 'dir' whenever a change occurs
 979  * on the server that might have invalidated our dcache.
 980  *
 981  * The caller should be holding dir->i_lock
 982  */
 983 void nfs_force_lookup_revalidate(struct inode *dir)
 984 {
 985         NFS_I(dir)->cache_change_attribute++;
 986 }
 987 EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate);
 988 
 989 /*
 990  * A check for whether or not the parent directory has changed.
 991  * In the case it has, we assume that the dentries are untrustworthy
 992  * and may need to be looked up again.
 993  * If rcu_walk prevents us from performing a full check, return 0.
 994  */
 995 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry,
 996                               int rcu_walk)
 997 {
 998         if (IS_ROOT(dentry))
 999                 return 1;
1000         if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
1001                 return 0;
1002         if (!nfs_verify_change_attribute(dir, dentry->d_time))
1003                 return 0;
1004         /* Revalidate nfsi->cache_change_attribute before we declare a match */
1005         if (nfs_mapping_need_revalidate_inode(dir)) {
1006                 if (rcu_walk)
1007                         return 0;
1008                 if (__nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
1009                         return 0;
1010         }
1011         if (!nfs_verify_change_attribute(dir, dentry->d_time))
1012                 return 0;
1013         return 1;
1014 }
1015 
1016 /*
1017  * Use intent information to check whether or not we're going to do
1018  * an O_EXCL create using this path component.
1019  */
1020 static int nfs_is_exclusive_create(struct inode *dir, unsigned int flags)
1021 {
1022         if (NFS_PROTO(dir)->version == 2)
1023                 return 0;
1024         return flags & LOOKUP_EXCL;
1025 }
1026 
1027 /*
1028  * Inode and filehandle revalidation for lookups.
1029  *
1030  * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
1031  * or if the intent information indicates that we're about to open this
1032  * particular file and the "nocto" mount flag is not set.
1033  *
1034  */
1035 static
1036 int nfs_lookup_verify_inode(struct inode *inode, unsigned int flags)
1037 {
1038         struct nfs_server *server = NFS_SERVER(inode);
1039         int ret;
1040 
1041         if (IS_AUTOMOUNT(inode))
1042                 return 0;
1043 
1044         if (flags & LOOKUP_OPEN) {
1045                 switch (inode->i_mode & S_IFMT) {
1046                 case S_IFREG:
1047                         /* A NFSv4 OPEN will revalidate later */
1048                         if (server->caps & NFS_CAP_ATOMIC_OPEN)
1049                                 goto out;
1050                         /* Fallthrough */
1051                 case S_IFDIR:
1052                         if (server->flags & NFS_MOUNT_NOCTO)
1053                                 break;
1054                         /* NFS close-to-open cache consistency validation */
1055                         goto out_force;
1056                 }
1057         }
1058 
1059         /* VFS wants an on-the-wire revalidation */
1060         if (flags & LOOKUP_REVAL)
1061                 goto out_force;
1062 out:
1063         return (inode->i_nlink == 0) ? -ESTALE : 0;
1064 out_force:
1065         if (flags & LOOKUP_RCU)
1066                 return -ECHILD;
1067         ret = __nfs_revalidate_inode(server, inode);
1068         if (ret != 0)
1069                 return ret;
1070         goto out;
1071 }
1072 
1073 /*
1074  * We judge how long we want to trust negative
1075  * dentries by looking at the parent inode mtime.
1076  *
1077  * If parent mtime has changed, we revalidate, else we wait for a
1078  * period corresponding to the parent's attribute cache timeout value.
1079  *
1080  * If LOOKUP_RCU prevents us from performing a full check, return 1
1081  * suggesting a reval is needed.
1082  *
1083  * Note that when creating a new file, or looking up a rename target,
1084  * then it shouldn't be necessary to revalidate a negative dentry.
1085  */
1086 static inline
1087 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
1088                        unsigned int flags)
1089 {
1090         if (flags & (LOOKUP_CREATE | LOOKUP_RENAME_TARGET))
1091                 return 0;
1092         if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
1093                 return 1;
1094         return !nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU);
1095 }
1096 
1097 static int
1098 nfs_lookup_revalidate_done(struct inode *dir, struct dentry *dentry,
1099                            struct inode *inode, int error)
1100 {
1101         switch (error) {
1102         case 1:
1103                 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is valid\n",
1104                         __func__, dentry);
1105                 return 1;
1106         case 0:
1107                 nfs_mark_for_revalidate(dir);
1108                 if (inode && S_ISDIR(inode->i_mode)) {
1109                         /* Purge readdir caches. */
1110                         nfs_zap_caches(inode);
1111                         /*
1112                          * We can't d_drop the root of a disconnected tree:
1113                          * its d_hash is on the s_anon list and d_drop() would hide
1114                          * it from shrink_dcache_for_unmount(), leading to busy
1115                          * inodes on unmount and further oopses.
1116                          */
1117                         if (IS_ROOT(dentry))
1118                                 return 1;
1119                 }
1120                 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is invalid\n",
1121                                 __func__, dentry);
1122                 return 0;
1123         }
1124         dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) lookup returned error %d\n",
1125                                 __func__, dentry, error);
1126         return error;
1127 }
1128 
1129 static int
1130 nfs_lookup_revalidate_negative(struct inode *dir, struct dentry *dentry,
1131                                unsigned int flags)
1132 {
1133         int ret = 1;
1134         if (nfs_neg_need_reval(dir, dentry, flags)) {
1135                 if (flags & LOOKUP_RCU)
1136                         return -ECHILD;
1137                 ret = 0;
1138         }
1139         return nfs_lookup_revalidate_done(dir, dentry, NULL, ret);
1140 }
1141 
1142 static int
1143 nfs_lookup_revalidate_delegated(struct inode *dir, struct dentry *dentry,
1144                                 struct inode *inode)
1145 {
1146         nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1147         return nfs_lookup_revalidate_done(dir, dentry, inode, 1);
1148 }
1149 
1150 static int
1151 nfs_lookup_revalidate_dentry(struct inode *dir, struct dentry *dentry,
1152                              struct inode *inode)
1153 {
1154         struct nfs_fh *fhandle;
1155         struct nfs_fattr *fattr;
1156         struct nfs4_label *label;
1157         int ret;
1158 
1159         ret = -ENOMEM;
1160         fhandle = nfs_alloc_fhandle();
1161         fattr = nfs_alloc_fattr();
1162         label = nfs4_label_alloc(NFS_SERVER(inode), GFP_KERNEL);
1163         if (fhandle == NULL || fattr == NULL || IS_ERR(label))
1164                 goto out;
1165 
1166         ret = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, label);
1167         if (ret < 0) {
1168                 if (ret == -ESTALE || ret == -ENOENT)
1169                         ret = 0;
1170                 goto out;
1171         }
1172         ret = 0;
1173         if (nfs_compare_fh(NFS_FH(inode), fhandle))
1174                 goto out;
1175         if (nfs_refresh_inode(inode, fattr) < 0)
1176                 goto out;
1177 
1178         nfs_setsecurity(inode, fattr, label);
1179         nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1180 
1181         /* set a readdirplus hint that we had a cache miss */
1182         nfs_force_use_readdirplus(dir);
1183         ret = 1;
1184 out:
1185         nfs_free_fattr(fattr);
1186         nfs_free_fhandle(fhandle);
1187         nfs4_label_free(label);
1188         return nfs_lookup_revalidate_done(dir, dentry, inode, ret);
1189 }
1190 
1191 /*
1192  * This is called every time the dcache has a lookup hit,
1193  * and we should check whether we can really trust that
1194  * lookup.
1195  *
1196  * NOTE! The hit can be a negative hit too, don't assume
1197  * we have an inode!
1198  *
1199  * If the parent directory is seen to have changed, we throw out the
1200  * cached dentry and do a new lookup.
1201  */
1202 static int
1203 nfs_do_lookup_revalidate(struct inode *dir, struct dentry *dentry,
1204                          unsigned int flags)
1205 {
1206         struct inode *inode;
1207         int error;
1208 
1209         nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
1210         inode = d_inode(dentry);
1211 
1212         if (!inode)
1213                 return nfs_lookup_revalidate_negative(dir, dentry, flags);
1214 
1215         if (is_bad_inode(inode)) {
1216                 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1217                                 __func__, dentry);
1218                 goto out_bad;
1219         }
1220 
1221         if (NFS_PROTO(dir)->have_delegation(inode, FMODE_READ))
1222                 return nfs_lookup_revalidate_delegated(dir, dentry, inode);
1223 
1224         /* Force a full look up iff the parent directory has changed */
1225         if (!(flags & (LOOKUP_EXCL | LOOKUP_REVAL)) &&
1226             nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU)) {
1227                 error = nfs_lookup_verify_inode(inode, flags);
1228                 if (error) {
1229                         if (error == -ESTALE)
1230                                 nfs_zap_caches(dir);
1231                         goto out_bad;
1232                 }
1233                 nfs_advise_use_readdirplus(dir);
1234                 goto out_valid;
1235         }
1236 
1237         if (flags & LOOKUP_RCU)
1238                 return -ECHILD;
1239 
1240         if (NFS_STALE(inode))
1241                 goto out_bad;
1242 
1243         trace_nfs_lookup_revalidate_enter(dir, dentry, flags);
1244         error = nfs_lookup_revalidate_dentry(dir, dentry, inode);
1245         trace_nfs_lookup_revalidate_exit(dir, dentry, flags, error);
1246         return error;
1247 out_valid:
1248         return nfs_lookup_revalidate_done(dir, dentry, inode, 1);
1249 out_bad:
1250         if (flags & LOOKUP_RCU)
1251                 return -ECHILD;
1252         return nfs_lookup_revalidate_done(dir, dentry, inode, 0);
1253 }
1254 
1255 static int
1256 __nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags,
1257                         int (*reval)(struct inode *, struct dentry *, unsigned int))
1258 {
1259         struct dentry *parent;
1260         struct inode *dir;
1261         int ret;
1262 
1263         if (flags & LOOKUP_RCU) {
1264                 parent = READ_ONCE(dentry->d_parent);
1265                 dir = d_inode_rcu(parent);
1266                 if (!dir)
1267                         return -ECHILD;
1268                 ret = reval(dir, dentry, flags);
1269                 if (parent != READ_ONCE(dentry->d_parent))
1270                         return -ECHILD;
1271         } else {
1272                 parent = dget_parent(dentry);
1273                 ret = reval(d_inode(parent), dentry, flags);
1274                 dput(parent);
1275         }
1276         return ret;
1277 }
1278 
1279 static int nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1280 {
1281         return __nfs_lookup_revalidate(dentry, flags, nfs_do_lookup_revalidate);
1282 }
1283 
1284 /*
1285  * A weaker form of d_revalidate for revalidating just the d_inode(dentry)
1286  * when we don't really care about the dentry name. This is called when a
1287  * pathwalk ends on a dentry that was not found via a normal lookup in the
1288  * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals).
1289  *
1290  * In this situation, we just want to verify that the inode itself is OK
1291  * since the dentry might have changed on the server.
1292  */
1293 static int nfs_weak_revalidate(struct dentry *dentry, unsigned int flags)
1294 {
1295         struct inode *inode = d_inode(dentry);
1296         int error = 0;
1297 
1298         /*
1299          * I believe we can only get a negative dentry here in the case of a
1300          * procfs-style symlink. Just assume it's correct for now, but we may
1301          * eventually need to do something more here.
1302          */
1303         if (!inode) {
1304                 dfprintk(LOOKUPCACHE, "%s: %pd2 has negative inode\n",
1305                                 __func__, dentry);
1306                 return 1;
1307         }
1308 
1309         if (is_bad_inode(inode)) {
1310                 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1311                                 __func__, dentry);
1312                 return 0;
1313         }
1314 
1315         error = nfs_lookup_verify_inode(inode, flags);
1316         dfprintk(LOOKUPCACHE, "NFS: %s: inode %lu is %s\n",
1317                         __func__, inode->i_ino, error ? "invalid" : "valid");
1318         return !error;
1319 }
1320 
1321 /*
1322  * This is called from dput() when d_count is going to 0.
1323  */
1324 static int nfs_dentry_delete(const struct dentry *dentry)
1325 {
1326         dfprintk(VFS, "NFS: dentry_delete(%pd2, %x)\n",
1327                 dentry, dentry->d_flags);
1328 
1329         /* Unhash any dentry with a stale inode */
1330         if (d_really_is_positive(dentry) && NFS_STALE(d_inode(dentry)))
1331                 return 1;
1332 
1333         if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1334                 /* Unhash it, so that ->d_iput() would be called */
1335                 return 1;
1336         }
1337         if (!(dentry->d_sb->s_flags & SB_ACTIVE)) {
1338                 /* Unhash it, so that ancestors of killed async unlink
1339                  * files will be cleaned up during umount */
1340                 return 1;
1341         }
1342         return 0;
1343 
1344 }
1345 
1346 /* Ensure that we revalidate inode->i_nlink */
1347 static void nfs_drop_nlink(struct inode *inode)
1348 {
1349         spin_lock(&inode->i_lock);
1350         /* drop the inode if we're reasonably sure this is the last link */
1351         if (inode->i_nlink > 0)
1352                 drop_nlink(inode);
1353         NFS_I(inode)->attr_gencount = nfs_inc_attr_generation_counter();
1354         NFS_I(inode)->cache_validity |= NFS_INO_INVALID_CHANGE
1355                 | NFS_INO_INVALID_CTIME
1356                 | NFS_INO_INVALID_OTHER
1357                 | NFS_INO_REVAL_FORCED;
1358         spin_unlock(&inode->i_lock);
1359 }
1360 
1361 /*
1362  * Called when the dentry loses inode.
1363  * We use it to clean up silly-renamed files.
1364  */
1365 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
1366 {
1367         if (S_ISDIR(inode->i_mode))
1368                 /* drop any readdir cache as it could easily be old */
1369                 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
1370 
1371         if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1372                 nfs_complete_unlink(dentry, inode);
1373                 nfs_drop_nlink(inode);
1374         }
1375         iput(inode);
1376 }
1377 
1378 static void nfs_d_release(struct dentry *dentry)
1379 {
1380         /* free cached devname value, if it survived that far */
1381         if (unlikely(dentry->d_fsdata)) {
1382                 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1383                         WARN_ON(1);
1384                 else
1385                         kfree(dentry->d_fsdata);
1386         }
1387 }
1388 
1389 const struct dentry_operations nfs_dentry_operations = {
1390         .d_revalidate   = nfs_lookup_revalidate,
1391         .d_weak_revalidate      = nfs_weak_revalidate,
1392         .d_delete       = nfs_dentry_delete,
1393         .d_iput         = nfs_dentry_iput,
1394         .d_automount    = nfs_d_automount,
1395         .d_release      = nfs_d_release,
1396 };
1397 EXPORT_SYMBOL_GPL(nfs_dentry_operations);
1398 
1399 struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
1400 {
1401         struct dentry *res;
1402         struct inode *inode = NULL;
1403         struct nfs_fh *fhandle = NULL;
1404         struct nfs_fattr *fattr = NULL;
1405         struct nfs4_label *label = NULL;
1406         int error;
1407 
1408         dfprintk(VFS, "NFS: lookup(%pd2)\n", dentry);
1409         nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
1410 
1411         if (unlikely(dentry->d_name.len > NFS_SERVER(dir)->namelen))
1412                 return ERR_PTR(-ENAMETOOLONG);
1413 
1414         /*
1415          * If we're doing an exclusive create, optimize away the lookup
1416          * but don't hash the dentry.
1417          */
1418         if (nfs_is_exclusive_create(dir, flags) || flags & LOOKUP_RENAME_TARGET)
1419                 return NULL;
1420 
1421         res = ERR_PTR(-ENOMEM);
1422         fhandle = nfs_alloc_fhandle();
1423         fattr = nfs_alloc_fattr();
1424         if (fhandle == NULL || fattr == NULL)
1425                 goto out;
1426 
1427         label = nfs4_label_alloc(NFS_SERVER(dir), GFP_NOWAIT);
1428         if (IS_ERR(label))
1429                 goto out;
1430 
1431         trace_nfs_lookup_enter(dir, dentry, flags);
1432         error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, label);
1433         if (error == -ENOENT)
1434                 goto no_entry;
1435         if (error < 0) {
1436                 res = ERR_PTR(error);
1437                 goto out_label;
1438         }
1439         inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label);
1440         res = ERR_CAST(inode);
1441         if (IS_ERR(res))
1442                 goto out_label;
1443 
1444         /* Notify readdir to use READDIRPLUS */
1445         nfs_force_use_readdirplus(dir);
1446 
1447 no_entry:
1448         res = d_splice_alias(inode, dentry);
1449         if (res != NULL) {
1450                 if (IS_ERR(res))
1451                         goto out_label;
1452                 dentry = res;
1453         }
1454         nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1455 out_label:
1456         trace_nfs_lookup_exit(dir, dentry, flags, error);
1457         nfs4_label_free(label);
1458 out:
1459         nfs_free_fattr(fattr);
1460         nfs_free_fhandle(fhandle);
1461         return res;
1462 }
1463 EXPORT_SYMBOL_GPL(nfs_lookup);
1464 
1465 #if IS_ENABLED(CONFIG_NFS_V4)
1466 static int nfs4_lookup_revalidate(struct dentry *, unsigned int);
1467 
1468 const struct dentry_operations nfs4_dentry_operations = {
1469         .d_revalidate   = nfs4_lookup_revalidate,
1470         .d_weak_revalidate      = nfs_weak_revalidate,
1471         .d_delete       = nfs_dentry_delete,
1472         .d_iput         = nfs_dentry_iput,
1473         .d_automount    = nfs_d_automount,
1474         .d_release      = nfs_d_release,
1475 };
1476 EXPORT_SYMBOL_GPL(nfs4_dentry_operations);
1477 
1478 static fmode_t flags_to_mode(int flags)
1479 {
1480         fmode_t res = (__force fmode_t)flags & FMODE_EXEC;
1481         if ((flags & O_ACCMODE) != O_WRONLY)
1482                 res |= FMODE_READ;
1483         if ((flags & O_ACCMODE) != O_RDONLY)
1484                 res |= FMODE_WRITE;
1485         return res;
1486 }
1487 
1488 static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags, struct file *filp)
1489 {
1490         return alloc_nfs_open_context(dentry, flags_to_mode(open_flags), filp);
1491 }
1492 
1493 static int do_open(struct inode *inode, struct file *filp)
1494 {
1495         nfs_fscache_open_file(inode, filp);
1496         return 0;
1497 }
1498 
1499 static int nfs_finish_open(struct nfs_open_context *ctx,
1500                            struct dentry *dentry,
1501                            struct file *file, unsigned open_flags)
1502 {
1503         int err;
1504 
1505         err = finish_open(file, dentry, do_open);
1506         if (err)
1507                 goto out;
1508         if (S_ISREG(file->f_path.dentry->d_inode->i_mode))
1509                 nfs_file_set_open_context(file, ctx);
1510         else
1511                 err = -EOPENSTALE;
1512 out:
1513         return err;
1514 }
1515 
1516 int nfs_atomic_open(struct inode *dir, struct dentry *dentry,
1517                     struct file *file, unsigned open_flags,
1518                     umode_t mode)
1519 {
1520         DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1521         struct nfs_open_context *ctx;
1522         struct dentry *res;
1523         struct iattr attr = { .ia_valid = ATTR_OPEN };
1524         struct inode *inode;
1525         unsigned int lookup_flags = 0;
1526         bool switched = false;
1527         int created = 0;
1528         int err;
1529 
1530         /* Expect a negative dentry */
1531         BUG_ON(d_inode(dentry));
1532 
1533         dfprintk(VFS, "NFS: atomic_open(%s/%lu), %pd\n",
1534                         dir->i_sb->s_id, dir->i_ino, dentry);
1535 
1536         err = nfs_check_flags(open_flags);
1537         if (err)
1538                 return err;
1539 
1540         /* NFS only supports OPEN on regular files */
1541         if ((open_flags & O_DIRECTORY)) {
1542                 if (!d_in_lookup(dentry)) {
1543                         /*
1544                          * Hashed negative dentry with O_DIRECTORY: dentry was
1545                          * revalidated and is fine, no need to perform lookup
1546                          * again
1547                          */
1548                         return -ENOENT;
1549                 }
1550                 lookup_flags = LOOKUP_OPEN|LOOKUP_DIRECTORY;
1551                 goto no_open;
1552         }
1553 
1554         if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1555                 return -ENAMETOOLONG;
1556 
1557         if (open_flags & O_CREAT) {
1558                 struct nfs_server *server = NFS_SERVER(dir);
1559 
1560                 if (!(server->attr_bitmask[2] & FATTR4_WORD2_MODE_UMASK))
1561                         mode &= ~current_umask();
1562 
1563                 attr.ia_valid |= ATTR_MODE;
1564                 attr.ia_mode = mode;
1565         }
1566         if (open_flags & O_TRUNC) {
1567                 attr.ia_valid |= ATTR_SIZE;
1568                 attr.ia_size = 0;
1569         }
1570 
1571         if (!(open_flags & O_CREAT) && !d_in_lookup(dentry)) {
1572                 d_drop(dentry);
1573                 switched = true;
1574                 dentry = d_alloc_parallel(dentry->d_parent,
1575                                           &dentry->d_name, &wq);
1576                 if (IS_ERR(dentry))
1577                         return PTR_ERR(dentry);
1578                 if (unlikely(!d_in_lookup(dentry)))
1579                         return finish_no_open(file, dentry);
1580         }
1581 
1582         ctx = create_nfs_open_context(dentry, open_flags, file);
1583         err = PTR_ERR(ctx);
1584         if (IS_ERR(ctx))
1585                 goto out;
1586 
1587         trace_nfs_atomic_open_enter(dir, ctx, open_flags);
1588         inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr, &created);
1589         if (created)
1590                 file->f_mode |= FMODE_CREATED;
1591         if (IS_ERR(inode)) {
1592                 err = PTR_ERR(inode);
1593                 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
1594                 put_nfs_open_context(ctx);
1595                 d_drop(dentry);
1596                 switch (err) {
1597                 case -ENOENT:
1598                         d_splice_alias(NULL, dentry);
1599                         nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1600                         break;
1601                 case -EISDIR:
1602                 case -ENOTDIR:
1603                         goto no_open;
1604                 case -ELOOP:
1605                         if (!(open_flags & O_NOFOLLOW))
1606                                 goto no_open;
1607                         break;
1608                         /* case -EINVAL: */
1609                 default:
1610                         break;
1611                 }
1612                 goto out;
1613         }
1614 
1615         err = nfs_finish_open(ctx, ctx->dentry, file, open_flags);
1616         trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
1617         put_nfs_open_context(ctx);
1618 out:
1619         if (unlikely(switched)) {
1620                 d_lookup_done(dentry);
1621                 dput(dentry);
1622         }
1623         return err;
1624 
1625 no_open:
1626         res = nfs_lookup(dir, dentry, lookup_flags);
1627         if (switched) {
1628                 d_lookup_done(dentry);
1629                 if (!res)
1630                         res = dentry;
1631                 else
1632                         dput(dentry);
1633         }
1634         if (IS_ERR(res))
1635                 return PTR_ERR(res);
1636         return finish_no_open(file, res);
1637 }
1638 EXPORT_SYMBOL_GPL(nfs_atomic_open);
1639 
1640 static int
1641 nfs4_do_lookup_revalidate(struct inode *dir, struct dentry *dentry,
1642                           unsigned int flags)
1643 {
1644         struct inode *inode;
1645 
1646         if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY))
1647                 goto full_reval;
1648         if (d_mountpoint(dentry))
1649                 goto full_reval;
1650 
1651         inode = d_inode(dentry);
1652 
1653         /* We can't create new files in nfs_open_revalidate(), so we
1654          * optimize away revalidation of negative dentries.
1655          */
1656         if (inode == NULL)
1657                 goto full_reval;
1658 
1659         if (NFS_PROTO(dir)->have_delegation(inode, FMODE_READ))
1660                 return nfs_lookup_revalidate_delegated(dir, dentry, inode);
1661 
1662         /* NFS only supports OPEN on regular files */
1663         if (!S_ISREG(inode->i_mode))
1664                 goto full_reval;
1665 
1666         /* We cannot do exclusive creation on a positive dentry */
1667         if (flags & (LOOKUP_EXCL | LOOKUP_REVAL))
1668                 goto reval_dentry;
1669 
1670         /* Check if the directory changed */
1671         if (!nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU))
1672                 goto reval_dentry;
1673 
1674         /* Let f_op->open() actually open (and revalidate) the file */
1675         return 1;
1676 reval_dentry:
1677         if (flags & LOOKUP_RCU)
1678                 return -ECHILD;
1679         return nfs_lookup_revalidate_dentry(dir, dentry, inode);
1680 
1681 full_reval:
1682         return nfs_do_lookup_revalidate(dir, dentry, flags);
1683 }
1684 
1685 static int nfs4_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1686 {
1687         return __nfs_lookup_revalidate(dentry, flags,
1688                         nfs4_do_lookup_revalidate);
1689 }
1690 
1691 #endif /* CONFIG_NFSV4 */
1692 
1693 struct dentry *
1694 nfs_add_or_obtain(struct dentry *dentry, struct nfs_fh *fhandle,
1695                                 struct nfs_fattr *fattr,
1696                                 struct nfs4_label *label)
1697 {
1698         struct dentry *parent = dget_parent(dentry);
1699         struct inode *dir = d_inode(parent);
1700         struct inode *inode;
1701         struct dentry *d;
1702         int error;
1703 
1704         d_drop(dentry);
1705 
1706         if (fhandle->size == 0) {
1707                 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, NULL);
1708                 if (error)
1709                         goto out_error;
1710         }
1711         nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1712         if (!(fattr->valid & NFS_ATTR_FATTR)) {
1713                 struct nfs_server *server = NFS_SB(dentry->d_sb);
1714                 error = server->nfs_client->rpc_ops->getattr(server, fhandle,
1715                                 fattr, NULL, NULL);
1716                 if (error < 0)
1717                         goto out_error;
1718         }
1719         inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label);
1720         d = d_splice_alias(inode, dentry);
1721 out:
1722         dput(parent);
1723         return d;
1724 out_error:
1725         nfs_mark_for_revalidate(dir);
1726         d = ERR_PTR(error);
1727         goto out;
1728 }
1729 EXPORT_SYMBOL_GPL(nfs_add_or_obtain);
1730 
1731 /*
1732  * Code common to create, mkdir, and mknod.
1733  */
1734 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1735                                 struct nfs_fattr *fattr,
1736                                 struct nfs4_label *label)
1737 {
1738         struct dentry *d;
1739 
1740         d = nfs_add_or_obtain(dentry, fhandle, fattr, label);
1741         if (IS_ERR(d))
1742                 return PTR_ERR(d);
1743 
1744         /* Callers don't care */
1745         dput(d);
1746         return 0;
1747 }
1748 EXPORT_SYMBOL_GPL(nfs_instantiate);
1749 
1750 /*
1751  * Following a failed create operation, we drop the dentry rather
1752  * than retain a negative dentry. This avoids a problem in the event
1753  * that the operation succeeded on the server, but an error in the
1754  * reply path made it appear to have failed.
1755  */
1756 int nfs_create(struct inode *dir, struct dentry *dentry,
1757                 umode_t mode, bool excl)
1758 {
1759         struct iattr attr;
1760         int open_flags = excl ? O_CREAT | O_EXCL : O_CREAT;
1761         int error;
1762 
1763         dfprintk(VFS, "NFS: create(%s/%lu), %pd\n",
1764                         dir->i_sb->s_id, dir->i_ino, dentry);
1765 
1766         attr.ia_mode = mode;
1767         attr.ia_valid = ATTR_MODE;
1768 
1769         trace_nfs_create_enter(dir, dentry, open_flags);
1770         error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags);
1771         trace_nfs_create_exit(dir, dentry, open_flags, error);
1772         if (error != 0)
1773                 goto out_err;
1774         return 0;
1775 out_err:
1776         d_drop(dentry);
1777         return error;
1778 }
1779 EXPORT_SYMBOL_GPL(nfs_create);
1780 
1781 /*
1782  * See comments for nfs_proc_create regarding failed operations.
1783  */
1784 int
1785 nfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t rdev)
1786 {
1787         struct iattr attr;
1788         int status;
1789 
1790         dfprintk(VFS, "NFS: mknod(%s/%lu), %pd\n",
1791                         dir->i_sb->s_id, dir->i_ino, dentry);
1792 
1793         attr.ia_mode = mode;
1794         attr.ia_valid = ATTR_MODE;
1795 
1796         trace_nfs_mknod_enter(dir, dentry);
1797         status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1798         trace_nfs_mknod_exit(dir, dentry, status);
1799         if (status != 0)
1800                 goto out_err;
1801         return 0;
1802 out_err:
1803         d_drop(dentry);
1804         return status;
1805 }
1806 EXPORT_SYMBOL_GPL(nfs_mknod);
1807 
1808 /*
1809  * See comments for nfs_proc_create regarding failed operations.
1810  */
1811 int nfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1812 {
1813         struct iattr attr;
1814         int error;
1815 
1816         dfprintk(VFS, "NFS: mkdir(%s/%lu), %pd\n",
1817                         dir->i_sb->s_id, dir->i_ino, dentry);
1818 
1819         attr.ia_valid = ATTR_MODE;
1820         attr.ia_mode = mode | S_IFDIR;
1821 
1822         trace_nfs_mkdir_enter(dir, dentry);
1823         error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1824         trace_nfs_mkdir_exit(dir, dentry, error);
1825         if (error != 0)
1826                 goto out_err;
1827         return 0;
1828 out_err:
1829         d_drop(dentry);
1830         return error;
1831 }
1832 EXPORT_SYMBOL_GPL(nfs_mkdir);
1833 
1834 static void nfs_dentry_handle_enoent(struct dentry *dentry)
1835 {
1836         if (simple_positive(dentry))
1837                 d_delete(dentry);
1838 }
1839 
1840 int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1841 {
1842         int error;
1843 
1844         dfprintk(VFS, "NFS: rmdir(%s/%lu), %pd\n",
1845                         dir->i_sb->s_id, dir->i_ino, dentry);
1846 
1847         trace_nfs_rmdir_enter(dir, dentry);
1848         if (d_really_is_positive(dentry)) {
1849                 down_write(&NFS_I(d_inode(dentry))->rmdir_sem);
1850                 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1851                 /* Ensure the VFS deletes this inode */
1852                 switch (error) {
1853                 case 0:
1854                         clear_nlink(d_inode(dentry));
1855                         break;
1856                 case -ENOENT:
1857                         nfs_dentry_handle_enoent(dentry);
1858                 }
1859                 up_write(&NFS_I(d_inode(dentry))->rmdir_sem);
1860         } else
1861                 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1862         trace_nfs_rmdir_exit(dir, dentry, error);
1863 
1864         return error;
1865 }
1866 EXPORT_SYMBOL_GPL(nfs_rmdir);
1867 
1868 /*
1869  * Remove a file after making sure there are no pending writes,
1870  * and after checking that the file has only one user. 
1871  *
1872  * We invalidate the attribute cache and free the inode prior to the operation
1873  * to avoid possible races if the server reuses the inode.
1874  */
1875 static int nfs_safe_remove(struct dentry *dentry)
1876 {
1877         struct inode *dir = d_inode(dentry->d_parent);
1878         struct inode *inode = d_inode(dentry);
1879         int error = -EBUSY;
1880                 
1881         dfprintk(VFS, "NFS: safe_remove(%pd2)\n", dentry);
1882 
1883         /* If the dentry was sillyrenamed, we simply call d_delete() */
1884         if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1885                 error = 0;
1886                 goto out;
1887         }
1888 
1889         trace_nfs_remove_enter(dir, dentry);
1890         if (inode != NULL) {
1891                 error = NFS_PROTO(dir)->remove(dir, dentry);
1892                 if (error == 0)
1893                         nfs_drop_nlink(inode);
1894         } else
1895                 error = NFS_PROTO(dir)->remove(dir, dentry);
1896         if (error == -ENOENT)
1897                 nfs_dentry_handle_enoent(dentry);
1898         trace_nfs_remove_exit(dir, dentry, error);
1899 out:
1900         return error;
1901 }
1902 
1903 /*  We do silly rename. In case sillyrename() returns -EBUSY, the inode
1904  *  belongs to an active ".nfs..." file and we return -EBUSY.
1905  *
1906  *  If sillyrename() returns 0, we do nothing, otherwise we unlink.
1907  */
1908 int nfs_unlink(struct inode *dir, struct dentry *dentry)
1909 {
1910         int error;
1911         int need_rehash = 0;
1912 
1913         dfprintk(VFS, "NFS: unlink(%s/%lu, %pd)\n", dir->i_sb->s_id,
1914                 dir->i_ino, dentry);
1915 
1916         trace_nfs_unlink_enter(dir, dentry);
1917         spin_lock(&dentry->d_lock);
1918         if (d_count(dentry) > 1) {
1919                 spin_unlock(&dentry->d_lock);
1920                 /* Start asynchronous writeout of the inode */
1921                 write_inode_now(d_inode(dentry), 0);
1922                 error = nfs_sillyrename(dir, dentry);
1923                 goto out;
1924         }
1925         if (!d_unhashed(dentry)) {
1926                 __d_drop(dentry);
1927                 need_rehash = 1;
1928         }
1929         spin_unlock(&dentry->d_lock);
1930         error = nfs_safe_remove(dentry);
1931         if (!error || error == -ENOENT) {
1932                 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1933         } else if (need_rehash)
1934                 d_rehash(dentry);
1935 out:
1936         trace_nfs_unlink_exit(dir, dentry, error);
1937         return error;
1938 }
1939 EXPORT_SYMBOL_GPL(nfs_unlink);
1940 
1941 /*
1942  * To create a symbolic link, most file systems instantiate a new inode,
1943  * add a page to it containing the path, then write it out to the disk
1944  * using prepare_write/commit_write.
1945  *
1946  * Unfortunately the NFS client can't create the in-core inode first
1947  * because it needs a file handle to create an in-core inode (see
1948  * fs/nfs/inode.c:nfs_fhget).  We only have a file handle *after* the
1949  * symlink request has completed on the server.
1950  *
1951  * So instead we allocate a raw page, copy the symname into it, then do
1952  * the SYMLINK request with the page as the buffer.  If it succeeds, we
1953  * now have a new file handle and can instantiate an in-core NFS inode
1954  * and move the raw page into its mapping.
1955  */
1956 int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1957 {
1958         struct page *page;
1959         char *kaddr;
1960         struct iattr attr;
1961         unsigned int pathlen = strlen(symname);
1962         int error;
1963 
1964         dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s)\n", dir->i_sb->s_id,
1965                 dir->i_ino, dentry, symname);
1966 
1967         if (pathlen > PAGE_SIZE)
1968                 return -ENAMETOOLONG;
1969 
1970         attr.ia_mode = S_IFLNK | S_IRWXUGO;
1971         attr.ia_valid = ATTR_MODE;
1972 
1973         page = alloc_page(GFP_USER);
1974         if (!page)
1975                 return -ENOMEM;
1976 
1977         kaddr = page_address(page);
1978         memcpy(kaddr, symname, pathlen);
1979         if (pathlen < PAGE_SIZE)
1980                 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
1981 
1982         trace_nfs_symlink_enter(dir, dentry);
1983         error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
1984         trace_nfs_symlink_exit(dir, dentry, error);
1985         if (error != 0) {
1986                 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s) error %d\n",
1987                         dir->i_sb->s_id, dir->i_ino,
1988                         dentry, symname, error);
1989                 d_drop(dentry);
1990                 __free_page(page);
1991                 return error;
1992         }
1993 
1994         /*
1995          * No big deal if we can't add this page to the page cache here.
1996          * READLINK will get the missing page from the server if needed.
1997          */
1998         if (!add_to_page_cache_lru(page, d_inode(dentry)->i_mapping, 0,
1999                                                         GFP_KERNEL)) {
2000                 SetPageUptodate(page);
2001                 unlock_page(page);
2002                 /*
2003                  * add_to_page_cache_lru() grabs an extra page refcount.
2004                  * Drop it here to avoid leaking this page later.
2005                  */
2006                 put_page(page);
2007         } else
2008                 __free_page(page);
2009 
2010         return 0;
2011 }
2012 EXPORT_SYMBOL_GPL(nfs_symlink);
2013 
2014 int
2015 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2016 {
2017         struct inode *inode = d_inode(old_dentry);
2018         int error;
2019 
2020         dfprintk(VFS, "NFS: link(%pd2 -> %pd2)\n",
2021                 old_dentry, dentry);
2022 
2023         trace_nfs_link_enter(inode, dir, dentry);
2024         d_drop(dentry);
2025         error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
2026         if (error == 0) {
2027                 ihold(inode);
2028                 d_add(dentry, inode);
2029         }
2030         trace_nfs_link_exit(inode, dir, dentry, error);
2031         return error;
2032 }
2033 EXPORT_SYMBOL_GPL(nfs_link);
2034 
2035 /*
2036  * RENAME
2037  * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
2038  * different file handle for the same inode after a rename (e.g. when
2039  * moving to a different directory). A fail-safe method to do so would
2040  * be to look up old_dir/old_name, create a link to new_dir/new_name and
2041  * rename the old file using the sillyrename stuff. This way, the original
2042  * file in old_dir will go away when the last process iput()s the inode.
2043  *
2044  * FIXED.
2045  * 
2046  * It actually works quite well. One needs to have the possibility for
2047  * at least one ".nfs..." file in each directory the file ever gets
2048  * moved or linked to which happens automagically with the new
2049  * implementation that only depends on the dcache stuff instead of
2050  * using the inode layer
2051  *
2052  * Unfortunately, things are a little more complicated than indicated
2053  * above. For a cross-directory move, we want to make sure we can get
2054  * rid of the old inode after the operation.  This means there must be
2055  * no pending writes (if it's a file), and the use count must be 1.
2056  * If these conditions are met, we can drop the dentries before doing
2057  * the rename.
2058  */
2059 int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2060                struct inode *new_dir, struct dentry *new_dentry,
2061                unsigned int flags)
2062 {
2063         struct inode *old_inode = d_inode(old_dentry);
2064         struct inode *new_inode = d_inode(new_dentry);
2065         struct dentry *dentry = NULL, *rehash = NULL;
2066         struct rpc_task *task;
2067         int error = -EBUSY;
2068 
2069         if (flags)
2070                 return -EINVAL;
2071 
2072         dfprintk(VFS, "NFS: rename(%pd2 -> %pd2, ct=%d)\n",
2073                  old_dentry, new_dentry,
2074                  d_count(new_dentry));
2075 
2076         trace_nfs_rename_enter(old_dir, old_dentry, new_dir, new_dentry);
2077         /*
2078          * For non-directories, check whether the target is busy and if so,
2079          * make a copy of the dentry and then do a silly-rename. If the
2080          * silly-rename succeeds, the copied dentry is hashed and becomes
2081          * the new target.
2082          */
2083         if (new_inode && !S_ISDIR(new_inode->i_mode)) {
2084                 /*
2085                  * To prevent any new references to the target during the
2086                  * rename, we unhash the dentry in advance.
2087                  */
2088                 if (!d_unhashed(new_dentry)) {
2089                         d_drop(new_dentry);
2090                         rehash = new_dentry;
2091                 }
2092 
2093                 if (d_count(new_dentry) > 2) {
2094                         int err;
2095 
2096                         /* copy the target dentry's name */
2097                         dentry = d_alloc(new_dentry->d_parent,
2098                                          &new_dentry->d_name);
2099                         if (!dentry)
2100                                 goto out;
2101 
2102                         /* silly-rename the existing target ... */
2103                         err = nfs_sillyrename(new_dir, new_dentry);
2104                         if (err)
2105                                 goto out;
2106 
2107                         new_dentry = dentry;
2108                         rehash = NULL;
2109                         new_inode = NULL;
2110                 }
2111         }
2112 
2113         task = nfs_async_rename(old_dir, new_dir, old_dentry, new_dentry, NULL);
2114         if (IS_ERR(task)) {
2115                 error = PTR_ERR(task);
2116                 goto out;
2117         }
2118 
2119         error = rpc_wait_for_completion_task(task);
2120         if (error != 0) {
2121                 ((struct nfs_renamedata *)task->tk_calldata)->cancelled = 1;
2122                 /* Paired with the atomic_dec_and_test() barrier in rpc_do_put_task() */
2123                 smp_wmb();
2124         } else
2125                 error = task->tk_status;
2126         rpc_put_task(task);
2127         /* Ensure the inode attributes are revalidated */
2128         if (error == 0) {
2129                 spin_lock(&old_inode->i_lock);
2130                 NFS_I(old_inode)->attr_gencount = nfs_inc_attr_generation_counter();
2131                 NFS_I(old_inode)->cache_validity |= NFS_INO_INVALID_CHANGE
2132                         | NFS_INO_INVALID_CTIME
2133                         | NFS_INO_REVAL_FORCED;
2134                 spin_unlock(&old_inode->i_lock);
2135         }
2136 out:
2137         if (rehash)
2138                 d_rehash(rehash);
2139         trace_nfs_rename_exit(old_dir, old_dentry,
2140                         new_dir, new_dentry, error);
2141         if (!error) {
2142                 if (new_inode != NULL)
2143                         nfs_drop_nlink(new_inode);
2144                 /*
2145                  * The d_move() should be here instead of in an async RPC completion
2146                  * handler because we need the proper locks to move the dentry.  If
2147                  * we're interrupted by a signal, the async RPC completion handler
2148                  * should mark the directories for revalidation.
2149                  */
2150                 d_move(old_dentry, new_dentry);
2151                 nfs_set_verifier(old_dentry,
2152                                         nfs_save_change_attribute(new_dir));
2153         } else if (error == -ENOENT)
2154                 nfs_dentry_handle_enoent(old_dentry);
2155 
2156         /* new dentry created? */
2157         if (dentry)
2158                 dput(dentry);
2159         return error;
2160 }
2161 EXPORT_SYMBOL_GPL(nfs_rename);
2162 
2163 static DEFINE_SPINLOCK(nfs_access_lru_lock);
2164 static LIST_HEAD(nfs_access_lru_list);
2165 static atomic_long_t nfs_access_nr_entries;
2166 
2167 static unsigned long nfs_access_max_cachesize = ULONG_MAX;
2168 module_param(nfs_access_max_cachesize, ulong, 0644);
2169 MODULE_PARM_DESC(nfs_access_max_cachesize, "NFS access maximum total cache length");
2170 
2171 static void nfs_access_free_entry(struct nfs_access_entry *entry)
2172 {
2173         put_cred(entry->cred);
2174         kfree_rcu(entry, rcu_head);
2175         smp_mb__before_atomic();
2176         atomic_long_dec(&nfs_access_nr_entries);
2177         smp_mb__after_atomic();
2178 }
2179 
2180 static void nfs_access_free_list(struct list_head *head)
2181 {
2182         struct nfs_access_entry *cache;
2183 
2184         while (!list_empty(head)) {
2185                 cache = list_entry(head->next, struct nfs_access_entry, lru);
2186                 list_del(&cache->lru);
2187                 nfs_access_free_entry(cache);
2188         }
2189 }
2190 
2191 static unsigned long
2192 nfs_do_access_cache_scan(unsigned int nr_to_scan)
2193 {
2194         LIST_HEAD(head);
2195         struct nfs_inode *nfsi, *next;
2196         struct nfs_access_entry *cache;
2197         long freed = 0;
2198 
2199         spin_lock(&nfs_access_lru_lock);
2200         list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) {
2201                 struct inode *inode;
2202 
2203                 if (nr_to_scan-- == 0)
2204                         break;
2205                 inode = &nfsi->vfs_inode;
2206                 spin_lock(&inode->i_lock);
2207                 if (list_empty(&nfsi->access_cache_entry_lru))
2208                         goto remove_lru_entry;
2209                 cache = list_entry(nfsi->access_cache_entry_lru.next,
2210                                 struct nfs_access_entry, lru);
2211                 list_move(&cache->lru, &head);
2212                 rb_erase(&cache->rb_node, &nfsi->access_cache);
2213                 freed++;
2214                 if (!list_empty(&nfsi->access_cache_entry_lru))
2215                         list_move_tail(&nfsi->access_cache_inode_lru,
2216                                         &nfs_access_lru_list);
2217                 else {
2218 remove_lru_entry:
2219                         list_del_init(&nfsi->access_cache_inode_lru);
2220                         smp_mb__before_atomic();
2221                         clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
2222                         smp_mb__after_atomic();
2223                 }
2224                 spin_unlock(&inode->i_lock);
2225         }
2226         spin_unlock(&nfs_access_lru_lock);
2227         nfs_access_free_list(&head);
2228         return freed;
2229 }
2230 
2231 unsigned long
2232 nfs_access_cache_scan(struct shrinker *shrink, struct shrink_control *sc)
2233 {
2234         int nr_to_scan = sc->nr_to_scan;
2235         gfp_t gfp_mask = sc->gfp_mask;
2236 
2237         if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL)
2238                 return SHRINK_STOP;
2239         return nfs_do_access_cache_scan(nr_to_scan);
2240 }
2241 
2242 
2243 unsigned long
2244 nfs_access_cache_count(struct shrinker *shrink, struct shrink_control *sc)
2245 {
2246         return vfs_pressure_ratio(atomic_long_read(&nfs_access_nr_entries));
2247 }
2248 
2249 static void
2250 nfs_access_cache_enforce_limit(void)
2251 {
2252         long nr_entries = atomic_long_read(&nfs_access_nr_entries);
2253         unsigned long diff;
2254         unsigned int nr_to_scan;
2255 
2256         if (nr_entries < 0 || nr_entries <= nfs_access_max_cachesize)
2257                 return;
2258         nr_to_scan = 100;
2259         diff = nr_entries - nfs_access_max_cachesize;
2260         if (diff < nr_to_scan)
2261                 nr_to_scan = diff;
2262         nfs_do_access_cache_scan(nr_to_scan);
2263 }
2264 
2265 static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head)
2266 {
2267         struct rb_root *root_node = &nfsi->access_cache;
2268         struct rb_node *n;
2269         struct nfs_access_entry *entry;
2270 
2271         /* Unhook entries from the cache */
2272         while ((n = rb_first(root_node)) != NULL) {
2273                 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2274                 rb_erase(n, root_node);
2275                 list_move(&entry->lru, head);
2276         }
2277         nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
2278 }
2279 
2280 void nfs_access_zap_cache(struct inode *inode)
2281 {
2282         LIST_HEAD(head);
2283 
2284         if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0)
2285                 return;
2286         /* Remove from global LRU init */
2287         spin_lock(&nfs_access_lru_lock);
2288         if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2289                 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
2290 
2291         spin_lock(&inode->i_lock);
2292         __nfs_access_zap_cache(NFS_I(inode), &head);
2293         spin_unlock(&inode->i_lock);
2294         spin_unlock(&nfs_access_lru_lock);
2295         nfs_access_free_list(&head);
2296 }
2297 EXPORT_SYMBOL_GPL(nfs_access_zap_cache);
2298 
2299 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, const struct cred *cred)
2300 {
2301         struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
2302 
2303         while (n != NULL) {
2304                 struct nfs_access_entry *entry =
2305                         rb_entry(n, struct nfs_access_entry, rb_node);
2306                 int cmp = cred_fscmp(cred, entry->cred);
2307 
2308                 if (cmp < 0)
2309                         n = n->rb_left;
2310                 else if (cmp > 0)
2311                         n = n->rb_right;
2312                 else
2313                         return entry;
2314         }
2315         return NULL;
2316 }
2317 
2318 static int nfs_access_get_cached(struct inode *inode, const struct cred *cred, struct nfs_access_entry *res, bool may_block)
2319 {
2320         struct nfs_inode *nfsi = NFS_I(inode);
2321         struct nfs_access_entry *cache;
2322         bool retry = true;
2323         int err;
2324 
2325         spin_lock(&inode->i_lock);
2326         for(;;) {
2327                 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2328                         goto out_zap;
2329                 cache = nfs_access_search_rbtree(inode, cred);
2330                 err = -ENOENT;
2331                 if (cache == NULL)
2332                         goto out;
2333                 /* Found an entry, is our attribute cache valid? */
2334                 if (!nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS))
2335                         break;
2336                 err = -ECHILD;
2337                 if (!may_block)
2338                         goto out;
2339                 if (!retry)
2340                         goto out_zap;
2341                 spin_unlock(&inode->i_lock);
2342                 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
2343                 if (err)
2344                         return err;
2345                 spin_lock(&inode->i_lock);
2346                 retry = false;
2347         }
2348         res->cred = cache->cred;
2349         res->mask = cache->mask;
2350         list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
2351         err = 0;
2352 out:
2353         spin_unlock(&inode->i_lock);
2354         return err;
2355 out_zap:
2356         spin_unlock(&inode->i_lock);
2357         nfs_access_zap_cache(inode);
2358         return -ENOENT;
2359 }
2360 
2361 static int nfs_access_get_cached_rcu(struct inode *inode, const struct cred *cred, struct nfs_access_entry *res)
2362 {
2363         /* Only check the most recently returned cache entry,
2364          * but do it without locking.
2365          */
2366         struct nfs_inode *nfsi = NFS_I(inode);
2367         struct nfs_access_entry *cache;
2368         int err = -ECHILD;
2369         struct list_head *lh;
2370 
2371         rcu_read_lock();
2372         if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2373                 goto out;
2374         lh = rcu_dereference(nfsi->access_cache_entry_lru.prev);
2375         cache = list_entry(lh, struct nfs_access_entry, lru);
2376         if (lh == &nfsi->access_cache_entry_lru ||
2377             cred != cache->cred)
2378                 cache = NULL;
2379         if (cache == NULL)
2380                 goto out;
2381         if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS))
2382                 goto out;
2383         res->cred = cache->cred;
2384         res->mask = cache->mask;
2385         err = 0;
2386 out:
2387         rcu_read_unlock();
2388         return err;
2389 }
2390 
2391 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
2392 {
2393         struct nfs_inode *nfsi = NFS_I(inode);
2394         struct rb_root *root_node = &nfsi->access_cache;
2395         struct rb_node **p = &root_node->rb_node;
2396         struct rb_node *parent = NULL;
2397         struct nfs_access_entry *entry;
2398         int cmp;
2399 
2400         spin_lock(&inode->i_lock);
2401         while (*p != NULL) {
2402                 parent = *p;
2403                 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
2404                 cmp = cred_fscmp(set->cred, entry->cred);
2405 
2406                 if (cmp < 0)
2407                         p = &parent->rb_left;
2408                 else if (cmp > 0)
2409                         p = &parent->rb_right;
2410                 else
2411                         goto found;
2412         }
2413         rb_link_node(&set->rb_node, parent, p);
2414         rb_insert_color(&set->rb_node, root_node);
2415         list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2416         spin_unlock(&inode->i_lock);
2417         return;
2418 found:
2419         rb_replace_node(parent, &set->rb_node, root_node);
2420         list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2421         list_del(&entry->lru);
2422         spin_unlock(&inode->i_lock);
2423         nfs_access_free_entry(entry);
2424 }
2425 
2426 void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
2427 {
2428         struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
2429         if (cache == NULL)
2430                 return;
2431         RB_CLEAR_NODE(&cache->rb_node);
2432         cache->cred = get_cred(set->cred);
2433         cache->mask = set->mask;
2434 
2435         /* The above field assignments must be visible
2436          * before this item appears on the lru.  We cannot easily
2437          * use rcu_assign_pointer, so just force the memory barrier.
2438          */
2439         smp_wmb();
2440         nfs_access_add_rbtree(inode, cache);
2441 
2442         /* Update accounting */
2443         smp_mb__before_atomic();
2444         atomic_long_inc(&nfs_access_nr_entries);
2445         smp_mb__after_atomic();
2446 
2447         /* Add inode to global LRU list */
2448         if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
2449                 spin_lock(&nfs_access_lru_lock);
2450                 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2451                         list_add_tail(&NFS_I(inode)->access_cache_inode_lru,
2452                                         &nfs_access_lru_list);
2453                 spin_unlock(&nfs_access_lru_lock);
2454         }
2455         nfs_access_cache_enforce_limit();
2456 }
2457 EXPORT_SYMBOL_GPL(nfs_access_add_cache);
2458 
2459 #define NFS_MAY_READ (NFS_ACCESS_READ)
2460 #define NFS_MAY_WRITE (NFS_ACCESS_MODIFY | \
2461                 NFS_ACCESS_EXTEND | \
2462                 NFS_ACCESS_DELETE)
2463 #define NFS_FILE_MAY_WRITE (NFS_ACCESS_MODIFY | \
2464                 NFS_ACCESS_EXTEND)
2465 #define NFS_DIR_MAY_WRITE NFS_MAY_WRITE
2466 #define NFS_MAY_LOOKUP (NFS_ACCESS_LOOKUP)
2467 #define NFS_MAY_EXECUTE (NFS_ACCESS_EXECUTE)
2468 static int
2469 nfs_access_calc_mask(u32 access_result, umode_t umode)
2470 {
2471         int mask = 0;
2472 
2473         if (access_result & NFS_MAY_READ)
2474                 mask |= MAY_READ;
2475         if (S_ISDIR(umode)) {
2476                 if ((access_result & NFS_DIR_MAY_WRITE) == NFS_DIR_MAY_WRITE)
2477                         mask |= MAY_WRITE;
2478                 if ((access_result & NFS_MAY_LOOKUP) == NFS_MAY_LOOKUP)
2479                         mask |= MAY_EXEC;
2480         } else if (S_ISREG(umode)) {
2481                 if ((access_result & NFS_FILE_MAY_WRITE) == NFS_FILE_MAY_WRITE)
2482                         mask |= MAY_WRITE;
2483                 if ((access_result & NFS_MAY_EXECUTE) == NFS_MAY_EXECUTE)
2484                         mask |= MAY_EXEC;
2485         } else if (access_result & NFS_MAY_WRITE)
2486                         mask |= MAY_WRITE;
2487         return mask;
2488 }
2489 
2490 void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result)
2491 {
2492         entry->mask = access_result;
2493 }
2494 EXPORT_SYMBOL_GPL(nfs_access_set_mask);
2495 
2496 static int nfs_do_access(struct inode *inode, const struct cred *cred, int mask)
2497 {
2498         struct nfs_access_entry cache;
2499         bool may_block = (mask & MAY_NOT_BLOCK) == 0;
2500         int cache_mask;
2501         int status;
2502 
2503         trace_nfs_access_enter(inode);
2504 
2505         status = nfs_access_get_cached_rcu(inode, cred, &cache);
2506         if (status != 0)
2507                 status = nfs_access_get_cached(inode, cred, &cache, may_block);
2508         if (status == 0)
2509                 goto out_cached;
2510 
2511         status = -ECHILD;
2512         if (!may_block)
2513                 goto out;
2514 
2515         /*
2516          * Determine which access bits we want to ask for...
2517          */
2518         cache.mask = NFS_ACCESS_READ | NFS_ACCESS_MODIFY | NFS_ACCESS_EXTEND;
2519         if (S_ISDIR(inode->i_mode))
2520                 cache.mask |= NFS_ACCESS_DELETE | NFS_ACCESS_LOOKUP;
2521         else
2522                 cache.mask |= NFS_ACCESS_EXECUTE;
2523         cache.cred = cred;
2524         status = NFS_PROTO(inode)->access(inode, &cache);
2525         if (status != 0) {
2526                 if (status == -ESTALE) {
2527                         nfs_zap_caches(inode);
2528                         if (!S_ISDIR(inode->i_mode))
2529                                 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
2530                 }
2531                 goto out;
2532         }
2533         nfs_access_add_cache(inode, &cache);
2534 out_cached:
2535         cache_mask = nfs_access_calc_mask(cache.mask, inode->i_mode);
2536         if ((mask & ~cache_mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) != 0)
2537                 status = -EACCES;
2538 out:
2539         trace_nfs_access_exit(inode, status);
2540         return status;
2541 }
2542 
2543 static int nfs_open_permission_mask(int openflags)
2544 {
2545         int mask = 0;
2546 
2547         if (openflags & __FMODE_EXEC) {
2548                 /* ONLY check exec rights */
2549                 mask = MAY_EXEC;
2550         } else {
2551                 if ((openflags & O_ACCMODE) != O_WRONLY)
2552                         mask |= MAY_READ;
2553                 if ((openflags & O_ACCMODE) != O_RDONLY)
2554                         mask |= MAY_WRITE;
2555         }
2556 
2557         return mask;
2558 }
2559 
2560 int nfs_may_open(struct inode *inode, const struct cred *cred, int openflags)
2561 {
2562         return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
2563 }
2564 EXPORT_SYMBOL_GPL(nfs_may_open);
2565 
2566 static int nfs_execute_ok(struct inode *inode, int mask)
2567 {
2568         struct nfs_server *server = NFS_SERVER(inode);
2569         int ret = 0;
2570 
2571         if (S_ISDIR(inode->i_mode))
2572                 return 0;
2573         if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_OTHER)) {
2574                 if (mask & MAY_NOT_BLOCK)
2575                         return -ECHILD;
2576                 ret = __nfs_revalidate_inode(server, inode);
2577         }
2578         if (ret == 0 && !execute_ok(inode))
2579                 ret = -EACCES;
2580         return ret;
2581 }
2582 
2583 int nfs_permission(struct inode *inode, int mask)
2584 {
2585         const struct cred *cred = current_cred();
2586         int res = 0;
2587 
2588         nfs_inc_stats(inode, NFSIOS_VFSACCESS);
2589 
2590         if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2591                 goto out;
2592         /* Is this sys_access() ? */
2593         if (mask & (MAY_ACCESS | MAY_CHDIR))
2594                 goto force_lookup;
2595 
2596         switch (inode->i_mode & S_IFMT) {
2597                 case S_IFLNK:
2598                         goto out;
2599                 case S_IFREG:
2600                         if ((mask & MAY_OPEN) &&
2601                            nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN))
2602                                 return 0;
2603                         break;
2604                 case S_IFDIR:
2605                         /*
2606                          * Optimize away all write operations, since the server
2607                          * will check permissions when we perform the op.
2608                          */
2609                         if ((mask & MAY_WRITE) && !(mask & MAY_READ))
2610                                 goto out;
2611         }
2612 
2613 force_lookup:
2614         if (!NFS_PROTO(inode)->access)
2615                 goto out_notsup;
2616 
2617         /* Always try fast lookups first */
2618         rcu_read_lock();
2619         res = nfs_do_access(inode, cred, mask|MAY_NOT_BLOCK);
2620         rcu_read_unlock();
2621         if (res == -ECHILD && !(mask & MAY_NOT_BLOCK)) {
2622                 /* Fast lookup failed, try the slow way */
2623                 res = nfs_do_access(inode, cred, mask);
2624         }
2625 out:
2626         if (!res && (mask & MAY_EXEC))
2627                 res = nfs_execute_ok(inode, mask);
2628 
2629         dfprintk(VFS, "NFS: permission(%s/%lu), mask=0x%x, res=%d\n",
2630                 inode->i_sb->s_id, inode->i_ino, mask, res);
2631         return res;
2632 out_notsup:
2633         if (mask & MAY_NOT_BLOCK)
2634                 return -ECHILD;
2635 
2636         res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
2637         if (res == 0)
2638                 res = generic_permission(inode, mask);
2639         goto out;
2640 }
2641 EXPORT_SYMBOL_GPL(nfs_permission);
2642 
2643 /*
2644  * Local variables:
2645  *  version-control: t
2646  *  kept-new-versions: 5
2647  * End:
2648  */

/* [<][>][^][v][top][bottom][index][help] */