linux/fs/nfs/dir.c
<<
>>
Prefs
   1/*
   2 *  linux/fs/nfs/dir.c
   3 *
   4 *  Copyright (C) 1992  Rick Sladkey
   5 *
   6 *  nfs directory handling functions
   7 *
   8 * 10 Apr 1996  Added silly rename for unlink   --okir
   9 * 28 Sep 1996  Improved directory cache --okir
  10 * 23 Aug 1997  Claus Heine claus@momo.math.rwth-aachen.de 
  11 *              Re-implemented silly rename for unlink, newly implemented
  12 *              silly rename for nfs_rename() following the suggestions
  13 *              of Olaf Kirch (okir) found in this file.
  14 *              Following Linus comments on my original hack, this version
  15 *              depends only on the dcache stuff and doesn't touch the inode
  16 *              layer (iput() and friends).
  17 *  6 Jun 1999  Cache readdir lookups in the page cache. -DaveM
  18 */
  19
  20#include <linux/time.h>
  21#include <linux/errno.h>
  22#include <linux/stat.h>
  23#include <linux/fcntl.h>
  24#include <linux/string.h>
  25#include <linux/kernel.h>
  26#include <linux/slab.h>
  27#include <linux/mm.h>
  28#include <linux/sunrpc/clnt.h>
  29#include <linux/nfs_fs.h>
  30#include <linux/nfs_mount.h>
  31#include <linux/pagemap.h>
  32#include <linux/pagevec.h>
  33#include <linux/namei.h>
  34#include <linux/mount.h>
  35#include <linux/sched.h>
  36#include <linux/kmemleak.h>
  37#include <linux/xattr.h>
  38
  39#include "delegation.h"
  40#include "iostat.h"
  41#include "internal.h"
  42#include "fscache.h"
  43
  44/* #define NFS_DEBUG_VERBOSE 1 */
  45
  46static int nfs_opendir(struct inode *, struct file *);
  47static int nfs_closedir(struct inode *, struct file *);
  48static int nfs_readdir(struct file *, void *, filldir_t);
  49static struct dentry *nfs_lookup(struct inode *, struct dentry *, struct nameidata *);
  50static int nfs_create(struct inode *, struct dentry *, int, struct nameidata *);
  51static int nfs_mkdir(struct inode *, struct dentry *, int);
  52static int nfs_rmdir(struct inode *, struct dentry *);
  53static int nfs_unlink(struct inode *, struct dentry *);
  54static int nfs_symlink(struct inode *, struct dentry *, const char *);
  55static int nfs_link(struct dentry *, struct inode *, struct dentry *);
  56static int nfs_mknod(struct inode *, struct dentry *, int, dev_t);
  57static int nfs_rename(struct inode *, struct dentry *,
  58                      struct inode *, struct dentry *);
  59static int nfs_fsync_dir(struct file *, int);
  60static loff_t nfs_llseek_dir(struct file *, loff_t, int);
  61static void nfs_readdir_clear_array(struct page*);
  62
  63const struct file_operations nfs_dir_operations = {
  64        .llseek         = nfs_llseek_dir,
  65        .read           = generic_read_dir,
  66        .readdir        = nfs_readdir,
  67        .open           = nfs_opendir,
  68        .release        = nfs_closedir,
  69        .fsync          = nfs_fsync_dir,
  70};
  71
  72const struct inode_operations nfs_dir_inode_operations = {
  73        .create         = nfs_create,
  74        .lookup         = nfs_lookup,
  75        .link           = nfs_link,
  76        .unlink         = nfs_unlink,
  77        .symlink        = nfs_symlink,
  78        .mkdir          = nfs_mkdir,
  79        .rmdir          = nfs_rmdir,
  80        .mknod          = nfs_mknod,
  81        .rename         = nfs_rename,
  82        .permission     = nfs_permission,
  83        .getattr        = nfs_getattr,
  84        .setattr        = nfs_setattr,
  85};
  86
  87const struct address_space_operations nfs_dir_aops = {
  88        .freepage = nfs_readdir_clear_array,
  89};
  90
  91#ifdef CONFIG_NFS_V3
  92const struct inode_operations nfs3_dir_inode_operations = {
  93        .create         = nfs_create,
  94        .lookup         = nfs_lookup,
  95        .link           = nfs_link,
  96        .unlink         = nfs_unlink,
  97        .symlink        = nfs_symlink,
  98        .mkdir          = nfs_mkdir,
  99        .rmdir          = nfs_rmdir,
 100        .mknod          = nfs_mknod,
 101        .rename         = nfs_rename,
 102        .permission     = nfs_permission,
 103        .getattr        = nfs_getattr,
 104        .setattr        = nfs_setattr,
 105        .listxattr      = nfs3_listxattr,
 106        .getxattr       = nfs3_getxattr,
 107        .setxattr       = nfs3_setxattr,
 108        .removexattr    = nfs3_removexattr,
 109};
 110#endif  /* CONFIG_NFS_V3 */
 111
 112#ifdef CONFIG_NFS_V4
 113
 114static struct dentry *nfs_atomic_lookup(struct inode *, struct dentry *, struct nameidata *);
 115static int nfs_open_create(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd);
 116const struct inode_operations nfs4_dir_inode_operations = {
 117        .create         = nfs_open_create,
 118        .lookup         = nfs_atomic_lookup,
 119        .link           = nfs_link,
 120        .unlink         = nfs_unlink,
 121        .symlink        = nfs_symlink,
 122        .mkdir          = nfs_mkdir,
 123        .rmdir          = nfs_rmdir,
 124        .mknod          = nfs_mknod,
 125        .rename         = nfs_rename,
 126        .permission     = nfs_permission,
 127        .getattr        = nfs_getattr,
 128        .setattr        = nfs_setattr,
 129        .getxattr       = generic_getxattr,
 130        .setxattr       = generic_setxattr,
 131        .listxattr      = generic_listxattr,
 132        .removexattr    = generic_removexattr,
 133};
 134
 135#endif /* CONFIG_NFS_V4 */
 136
 137static struct nfs_open_dir_context *alloc_nfs_open_dir_context(struct rpc_cred *cred)
 138{
 139        struct nfs_open_dir_context *ctx;
 140        ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
 141        if (ctx != NULL) {
 142                ctx->duped = 0;
 143                ctx->dir_cookie = 0;
 144                ctx->dup_cookie = 0;
 145                ctx->cred = get_rpccred(cred);
 146        } else
 147                ctx = ERR_PTR(-ENOMEM);
 148        return ctx;
 149}
 150
 151static void put_nfs_open_dir_context(struct nfs_open_dir_context *ctx)
 152{
 153        put_rpccred(ctx->cred);
 154        kfree(ctx);
 155}
 156
 157/*
 158 * Open file
 159 */
 160static int
 161nfs_opendir(struct inode *inode, struct file *filp)
 162{
 163        int res = 0;
 164        struct nfs_open_dir_context *ctx;
 165        struct rpc_cred *cred;
 166
 167        dfprintk(FILE, "NFS: open dir(%s/%s)\n",
 168                        filp->f_path.dentry->d_parent->d_name.name,
 169                        filp->f_path.dentry->d_name.name);
 170
 171        nfs_inc_stats(inode, NFSIOS_VFSOPEN);
 172
 173        cred = rpc_lookup_cred();
 174        if (IS_ERR(cred))
 175                return PTR_ERR(cred);
 176        ctx = alloc_nfs_open_dir_context(cred);
 177        if (IS_ERR(ctx)) {
 178                res = PTR_ERR(ctx);
 179                goto out;
 180        }
 181        filp->private_data = ctx;
 182        if (filp->f_path.dentry == filp->f_path.mnt->mnt_root) {
 183                /* This is a mountpoint, so d_revalidate will never
 184                 * have been called, so we need to refresh the
 185                 * inode (for close-open consistency) ourselves.
 186                 */
 187                __nfs_revalidate_inode(NFS_SERVER(inode), inode);
 188        }
 189out:
 190        put_rpccred(cred);
 191        return res;
 192}
 193
 194static int
 195nfs_closedir(struct inode *inode, struct file *filp)
 196{
 197        put_nfs_open_dir_context(filp->private_data);
 198        return 0;
 199}
 200
 201struct nfs_cache_array_entry {
 202        u64 cookie;
 203        u64 ino;
 204        struct qstr string;
 205        unsigned char d_type;
 206};
 207
 208struct nfs_cache_array {
 209        unsigned int size;
 210        int eof_index;
 211        u64 last_cookie;
 212        struct nfs_cache_array_entry array[0];
 213};
 214
 215typedef int (*decode_dirent_t)(struct xdr_stream *, struct nfs_entry *, int);
 216typedef struct {
 217        struct file     *file;
 218        struct page     *page;
 219        unsigned long   page_index;
 220        u64             *dir_cookie;
 221        u64             last_cookie;
 222        loff_t          current_index;
 223        decode_dirent_t decode;
 224
 225        unsigned long   timestamp;
 226        unsigned long   gencount;
 227        unsigned int    cache_entry_index;
 228        unsigned int    plus:1;
 229        unsigned int    eof:1;
 230} nfs_readdir_descriptor_t;
 231
 232/*
 233 * The caller is responsible for calling nfs_readdir_release_array(page)
 234 */
 235static
 236struct nfs_cache_array *nfs_readdir_get_array(struct page *page)
 237{
 238        void *ptr;
 239        if (page == NULL)
 240                return ERR_PTR(-EIO);
 241        ptr = kmap(page);
 242        if (ptr == NULL)
 243                return ERR_PTR(-ENOMEM);
 244        return ptr;
 245}
 246
 247static
 248void nfs_readdir_release_array(struct page *page)
 249{
 250        kunmap(page);
 251}
 252
 253/*
 254 * we are freeing strings created by nfs_add_to_readdir_array()
 255 */
 256static
 257void nfs_readdir_clear_array(struct page *page)
 258{
 259        struct nfs_cache_array *array;
 260        int i;
 261
 262        array = kmap_atomic(page, KM_USER0);
 263        for (i = 0; i < array->size; i++)
 264                kfree(array->array[i].string.name);
 265        kunmap_atomic(array, KM_USER0);
 266}
 267
 268/*
 269 * the caller is responsible for freeing qstr.name
 270 * when called by nfs_readdir_add_to_array, the strings will be freed in
 271 * nfs_clear_readdir_array()
 272 */
 273static
 274int nfs_readdir_make_qstr(struct qstr *string, const char *name, unsigned int len)
 275{
 276        string->len = len;
 277        string->name = kmemdup(name, len, GFP_KERNEL);
 278        if (string->name == NULL)
 279                return -ENOMEM;
 280        /*
 281         * Avoid a kmemleak false positive. The pointer to the name is stored
 282         * in a page cache page which kmemleak does not scan.
 283         */
 284        kmemleak_not_leak(string->name);
 285        string->hash = full_name_hash(name, len);
 286        return 0;
 287}
 288
 289static
 290int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page)
 291{
 292        struct nfs_cache_array *array = nfs_readdir_get_array(page);
 293        struct nfs_cache_array_entry *cache_entry;
 294        int ret;
 295
 296        if (IS_ERR(array))
 297                return PTR_ERR(array);
 298
 299        cache_entry = &array->array[array->size];
 300
 301        /* Check that this entry lies within the page bounds */
 302        ret = -ENOSPC;
 303        if ((char *)&cache_entry[1] - (char *)page_address(page) > PAGE_SIZE)
 304                goto out;
 305
 306        cache_entry->cookie = entry->prev_cookie;
 307        cache_entry->ino = entry->ino;
 308        cache_entry->d_type = entry->d_type;
 309        ret = nfs_readdir_make_qstr(&cache_entry->string, entry->name, entry->len);
 310        if (ret)
 311                goto out;
 312        array->last_cookie = entry->cookie;
 313        array->size++;
 314        if (entry->eof != 0)
 315                array->eof_index = array->size;
 316out:
 317        nfs_readdir_release_array(page);
 318        return ret;
 319}
 320
 321static
 322int nfs_readdir_search_for_pos(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
 323{
 324        loff_t diff = desc->file->f_pos - desc->current_index;
 325        unsigned int index;
 326        struct nfs_open_dir_context *ctx = desc->file->private_data;
 327
 328        if (diff < 0)
 329                goto out_eof;
 330        if (diff >= array->size) {
 331                if (array->eof_index >= 0)
 332                        goto out_eof;
 333                return -EAGAIN;
 334        }
 335
 336        index = (unsigned int)diff;
 337        *desc->dir_cookie = array->array[index].cookie;
 338        desc->cache_entry_index = index;
 339        ctx->duped = 0;
 340        return 0;
 341out_eof:
 342        desc->eof = 1;
 343        return -EBADCOOKIE;
 344}
 345
 346static
 347int nfs_readdir_search_for_cookie(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
 348{
 349        int i;
 350        loff_t new_pos;
 351        int status = -EAGAIN;
 352        struct nfs_open_dir_context *ctx = desc->file->private_data;
 353
 354        for (i = 0; i < array->size; i++) {
 355                if (array->array[i].cookie == *desc->dir_cookie) {
 356                        new_pos = desc->current_index + i;
 357                        if (new_pos < desc->file->f_pos) {
 358                                ctx->dup_cookie = *desc->dir_cookie;
 359                                ctx->duped = 1;
 360                        }
 361                        desc->file->f_pos = new_pos;
 362                        desc->cache_entry_index = i;
 363                        return 0;
 364                }
 365        }
 366        if (array->eof_index >= 0) {
 367                status = -EBADCOOKIE;
 368                if (*desc->dir_cookie == array->last_cookie)
 369                        desc->eof = 1;
 370        }
 371        return status;
 372}
 373
 374static
 375int nfs_readdir_search_array(nfs_readdir_descriptor_t *desc)
 376{
 377        struct nfs_cache_array *array;
 378        int status;
 379
 380        array = nfs_readdir_get_array(desc->page);
 381        if (IS_ERR(array)) {
 382                status = PTR_ERR(array);
 383                goto out;
 384        }
 385
 386        if (*desc->dir_cookie == 0)
 387                status = nfs_readdir_search_for_pos(array, desc);
 388        else
 389                status = nfs_readdir_search_for_cookie(array, desc);
 390
 391        if (status == -EAGAIN) {
 392                desc->last_cookie = array->last_cookie;
 393                desc->current_index += array->size;
 394                desc->page_index++;
 395        }
 396        nfs_readdir_release_array(desc->page);
 397out:
 398        return status;
 399}
 400
 401/* Fill a page with xdr information before transferring to the cache page */
 402static
 403int nfs_readdir_xdr_filler(struct page **pages, nfs_readdir_descriptor_t *desc,
 404                        struct nfs_entry *entry, struct file *file, struct inode *inode)
 405{
 406        struct nfs_open_dir_context *ctx = file->private_data;
 407        struct rpc_cred *cred = ctx->cred;
 408        unsigned long   timestamp, gencount;
 409        int             error;
 410
 411 again:
 412        timestamp = jiffies;
 413        gencount = nfs_inc_attr_generation_counter();
 414        error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, entry->cookie, pages,
 415                                          NFS_SERVER(inode)->dtsize, desc->plus);
 416        if (error < 0) {
 417                /* We requested READDIRPLUS, but the server doesn't grok it */
 418                if (error == -ENOTSUPP && desc->plus) {
 419                        NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
 420                        clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
 421                        desc->plus = 0;
 422                        goto again;
 423                }
 424                goto error;
 425        }
 426        desc->timestamp = timestamp;
 427        desc->gencount = gencount;
 428error:
 429        return error;
 430}
 431
 432static int xdr_decode(nfs_readdir_descriptor_t *desc,
 433                      struct nfs_entry *entry, struct xdr_stream *xdr)
 434{
 435        int error;
 436
 437        error = desc->decode(xdr, entry, desc->plus);
 438        if (error)
 439                return error;
 440        entry->fattr->time_start = desc->timestamp;
 441        entry->fattr->gencount = desc->gencount;
 442        return 0;
 443}
 444
 445static
 446int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry)
 447{
 448        if (dentry->d_inode == NULL)
 449                goto different;
 450        if (nfs_compare_fh(entry->fh, NFS_FH(dentry->d_inode)) != 0)
 451                goto different;
 452        return 1;
 453different:
 454        return 0;
 455}
 456
 457static
 458void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry)
 459{
 460        struct qstr filename = {
 461                .len = entry->len,
 462                .name = entry->name,
 463        };
 464        struct dentry *dentry;
 465        struct dentry *alias;
 466        struct inode *dir = parent->d_inode;
 467        struct inode *inode;
 468
 469        if (filename.name[0] == '.') {
 470                if (filename.len == 1)
 471                        return;
 472                if (filename.len == 2 && filename.name[1] == '.')
 473                        return;
 474        }
 475        filename.hash = full_name_hash(filename.name, filename.len);
 476
 477        dentry = d_lookup(parent, &filename);
 478        if (dentry != NULL) {
 479                if (nfs_same_file(dentry, entry)) {
 480                        nfs_refresh_inode(dentry->d_inode, entry->fattr);
 481                        goto out;
 482                } else {
 483                        d_drop(dentry);
 484                        dput(dentry);
 485                }
 486        }
 487
 488        dentry = d_alloc(parent, &filename);
 489        if (dentry == NULL)
 490                return;
 491
 492        inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr);
 493        if (IS_ERR(inode))
 494                goto out;
 495
 496        alias = d_materialise_unique(dentry, inode);
 497        if (IS_ERR(alias))
 498                goto out;
 499        else if (alias) {
 500                nfs_set_verifier(alias, nfs_save_change_attribute(dir));
 501                dput(alias);
 502        } else
 503                nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
 504
 505out:
 506        dput(dentry);
 507}
 508
 509/* Perform conversion from xdr to cache array */
 510static
 511int nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry,
 512                                struct page **xdr_pages, struct page *page, unsigned int buflen)
 513{
 514        struct xdr_stream stream;
 515        struct xdr_buf buf;
 516        struct page *scratch;
 517        struct nfs_cache_array *array;
 518        unsigned int count = 0;
 519        int status;
 520
 521        scratch = alloc_page(GFP_KERNEL);
 522        if (scratch == NULL)
 523                return -ENOMEM;
 524
 525        xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen);
 526        xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
 527
 528        do {
 529                status = xdr_decode(desc, entry, &stream);
 530                if (status != 0) {
 531                        if (status == -EAGAIN)
 532                                status = 0;
 533                        break;
 534                }
 535
 536                count++;
 537
 538                if (desc->plus != 0)
 539                        nfs_prime_dcache(desc->file->f_path.dentry, entry);
 540
 541                status = nfs_readdir_add_to_array(entry, page);
 542                if (status != 0)
 543                        break;
 544        } while (!entry->eof);
 545
 546        if (count == 0 || (status == -EBADCOOKIE && entry->eof != 0)) {
 547                array = nfs_readdir_get_array(page);
 548                if (!IS_ERR(array)) {
 549                        array->eof_index = array->size;
 550                        status = 0;
 551                        nfs_readdir_release_array(page);
 552                } else
 553                        status = PTR_ERR(array);
 554        }
 555
 556        put_page(scratch);
 557        return status;
 558}
 559
 560static
 561void nfs_readdir_free_pagearray(struct page **pages, unsigned int npages)
 562{
 563        unsigned int i;
 564        for (i = 0; i < npages; i++)
 565                put_page(pages[i]);
 566}
 567
 568static
 569void nfs_readdir_free_large_page(void *ptr, struct page **pages,
 570                unsigned int npages)
 571{
 572        nfs_readdir_free_pagearray(pages, npages);
 573}
 574
 575/*
 576 * nfs_readdir_large_page will allocate pages that must be freed with a call
 577 * to nfs_readdir_free_large_page
 578 */
 579static
 580int nfs_readdir_large_page(struct page **pages, unsigned int npages)
 581{
 582        unsigned int i;
 583
 584        for (i = 0; i < npages; i++) {
 585                struct page *page = alloc_page(GFP_KERNEL);
 586                if (page == NULL)
 587                        goto out_freepages;
 588                pages[i] = page;
 589        }
 590        return 0;
 591
 592out_freepages:
 593        nfs_readdir_free_pagearray(pages, i);
 594        return -ENOMEM;
 595}
 596
 597static
 598int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t *desc, struct page *page, struct inode *inode)
 599{
 600        struct page *pages[NFS_MAX_READDIR_PAGES];
 601        void *pages_ptr = NULL;
 602        struct nfs_entry entry;
 603        struct file     *file = desc->file;
 604        struct nfs_cache_array *array;
 605        int status = -ENOMEM;
 606        unsigned int array_size = ARRAY_SIZE(pages);
 607
 608        entry.prev_cookie = 0;
 609        entry.cookie = desc->last_cookie;
 610        entry.eof = 0;
 611        entry.fh = nfs_alloc_fhandle();
 612        entry.fattr = nfs_alloc_fattr();
 613        entry.server = NFS_SERVER(inode);
 614        if (entry.fh == NULL || entry.fattr == NULL)
 615                goto out;
 616
 617        array = nfs_readdir_get_array(page);
 618        if (IS_ERR(array)) {
 619                status = PTR_ERR(array);
 620                goto out;
 621        }
 622        memset(array, 0, sizeof(struct nfs_cache_array));
 623        array->eof_index = -1;
 624
 625        status = nfs_readdir_large_page(pages, array_size);
 626        if (status < 0)
 627                goto out_release_array;
 628        do {
 629                unsigned int pglen;
 630                status = nfs_readdir_xdr_filler(pages, desc, &entry, file, inode);
 631
 632                if (status < 0)
 633                        break;
 634                pglen = status;
 635                status = nfs_readdir_page_filler(desc, &entry, pages, page, pglen);
 636                if (status < 0) {
 637                        if (status == -ENOSPC)
 638                                status = 0;
 639                        break;
 640                }
 641        } while (array->eof_index < 0);
 642
 643        nfs_readdir_free_large_page(pages_ptr, pages, array_size);
 644out_release_array:
 645        nfs_readdir_release_array(page);
 646out:
 647        nfs_free_fattr(entry.fattr);
 648        nfs_free_fhandle(entry.fh);
 649        return status;
 650}
 651
 652/*
 653 * Now we cache directories properly, by converting xdr information
 654 * to an array that can be used for lookups later.  This results in
 655 * fewer cache pages, since we can store more information on each page.
 656 * We only need to convert from xdr once so future lookups are much simpler
 657 */
 658static
 659int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page* page)
 660{
 661        struct inode    *inode = desc->file->f_path.dentry->d_inode;
 662        int ret;
 663
 664        ret = nfs_readdir_xdr_to_array(desc, page, inode);
 665        if (ret < 0)
 666                goto error;
 667        SetPageUptodate(page);
 668
 669        if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) {
 670                /* Should never happen */
 671                nfs_zap_mapping(inode, inode->i_mapping);
 672        }
 673        unlock_page(page);
 674        return 0;
 675 error:
 676        unlock_page(page);
 677        return ret;
 678}
 679
 680static
 681void cache_page_release(nfs_readdir_descriptor_t *desc)
 682{
 683        if (!desc->page->mapping)
 684                nfs_readdir_clear_array(desc->page);
 685        page_cache_release(desc->page);
 686        desc->page = NULL;
 687}
 688
 689static
 690struct page *get_cache_page(nfs_readdir_descriptor_t *desc)
 691{
 692        return read_cache_page(desc->file->f_path.dentry->d_inode->i_mapping,
 693                        desc->page_index, (filler_t *)nfs_readdir_filler, desc);
 694}
 695
 696/*
 697 * Returns 0 if desc->dir_cookie was found on page desc->page_index
 698 */
 699static
 700int find_cache_page(nfs_readdir_descriptor_t *desc)
 701{
 702        int res;
 703
 704        desc->page = get_cache_page(desc);
 705        if (IS_ERR(desc->page))
 706                return PTR_ERR(desc->page);
 707
 708        res = nfs_readdir_search_array(desc);
 709        if (res != 0)
 710                cache_page_release(desc);
 711        return res;
 712}
 713
 714/* Search for desc->dir_cookie from the beginning of the page cache */
 715static inline
 716int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
 717{
 718        int res;
 719
 720        if (desc->page_index == 0) {
 721                desc->current_index = 0;
 722                desc->last_cookie = 0;
 723        }
 724        do {
 725                res = find_cache_page(desc);
 726        } while (res == -EAGAIN);
 727        return res;
 728}
 729
 730/*
 731 * Once we've found the start of the dirent within a page: fill 'er up...
 732 */
 733static 
 734int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent,
 735                   filldir_t filldir)
 736{
 737        struct file     *file = desc->file;
 738        int i = 0;
 739        int res = 0;
 740        struct nfs_cache_array *array = NULL;
 741        struct nfs_open_dir_context *ctx = file->private_data;
 742
 743        if (ctx->duped != 0 && ctx->dup_cookie == *desc->dir_cookie) {
 744                if (printk_ratelimit()) {
 745                        pr_notice("NFS: directory %s/%s contains a readdir loop.  "
 746                                "Please contact your server vendor.  "
 747                                "Offending cookie: %llu\n",
 748                                file->f_dentry->d_parent->d_name.name,
 749                                file->f_dentry->d_name.name,
 750                                *desc->dir_cookie);
 751                }
 752                res = -ELOOP;
 753                goto out;
 754        }
 755
 756        array = nfs_readdir_get_array(desc->page);
 757        if (IS_ERR(array)) {
 758                res = PTR_ERR(array);
 759                goto out;
 760        }
 761
 762        for (i = desc->cache_entry_index; i < array->size; i++) {
 763                struct nfs_cache_array_entry *ent;
 764
 765                ent = &array->array[i];
 766                if (filldir(dirent, ent->string.name, ent->string.len,
 767                    file->f_pos, nfs_compat_user_ino64(ent->ino),
 768                    ent->d_type) < 0) {
 769                        desc->eof = 1;
 770                        break;
 771                }
 772                file->f_pos++;
 773                if (i < (array->size-1))
 774                        *desc->dir_cookie = array->array[i+1].cookie;
 775                else
 776                        *desc->dir_cookie = array->last_cookie;
 777        }
 778        if (array->eof_index >= 0)
 779                desc->eof = 1;
 780
 781        nfs_readdir_release_array(desc->page);
 782out:
 783        cache_page_release(desc);
 784        dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
 785                        (unsigned long long)*desc->dir_cookie, res);
 786        return res;
 787}
 788
 789/*
 790 * If we cannot find a cookie in our cache, we suspect that this is
 791 * because it points to a deleted file, so we ask the server to return
 792 * whatever it thinks is the next entry. We then feed this to filldir.
 793 * If all goes well, we should then be able to find our way round the
 794 * cache on the next call to readdir_search_pagecache();
 795 *
 796 * NOTE: we cannot add the anonymous page to the pagecache because
 797 *       the data it contains might not be page aligned. Besides,
 798 *       we should already have a complete representation of the
 799 *       directory in the page cache by the time we get here.
 800 */
 801static inline
 802int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent,
 803                     filldir_t filldir)
 804{
 805        struct page     *page = NULL;
 806        int             status;
 807        struct inode *inode = desc->file->f_path.dentry->d_inode;
 808
 809        dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n",
 810                        (unsigned long long)*desc->dir_cookie);
 811
 812        page = alloc_page(GFP_HIGHUSER);
 813        if (!page) {
 814                status = -ENOMEM;
 815                goto out;
 816        }
 817
 818        desc->page_index = 0;
 819        desc->last_cookie = *desc->dir_cookie;
 820        desc->page = page;
 821
 822        status = nfs_readdir_xdr_to_array(desc, page, inode);
 823        if (status < 0)
 824                goto out_release;
 825
 826        status = nfs_do_filldir(desc, dirent, filldir);
 827
 828 out:
 829        dfprintk(DIRCACHE, "NFS: %s: returns %d\n",
 830                        __func__, status);
 831        return status;
 832 out_release:
 833        cache_page_release(desc);
 834        goto out;
 835}
 836
 837/* The file offset position represents the dirent entry number.  A
 838   last cookie cache takes care of the common case of reading the
 839   whole directory.
 840 */
 841static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
 842{
 843        struct dentry   *dentry = filp->f_path.dentry;
 844        struct inode    *inode = dentry->d_inode;
 845        nfs_readdir_descriptor_t my_desc,
 846                        *desc = &my_desc;
 847        struct nfs_open_dir_context *dir_ctx = filp->private_data;
 848        int res;
 849
 850        dfprintk(FILE, "NFS: readdir(%s/%s) starting at cookie %llu\n",
 851                        dentry->d_parent->d_name.name, dentry->d_name.name,
 852                        (long long)filp->f_pos);
 853        nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
 854
 855        /*
 856         * filp->f_pos points to the dirent entry number.
 857         * *desc->dir_cookie has the cookie for the next entry. We have
 858         * to either find the entry with the appropriate number or
 859         * revalidate the cookie.
 860         */
 861        memset(desc, 0, sizeof(*desc));
 862
 863        desc->file = filp;
 864        desc->dir_cookie = &dir_ctx->dir_cookie;
 865        desc->decode = NFS_PROTO(inode)->decode_dirent;
 866        desc->plus = NFS_USE_READDIRPLUS(inode);
 867
 868        nfs_block_sillyrename(dentry);
 869        res = nfs_revalidate_mapping(inode, filp->f_mapping);
 870        if (res < 0)
 871                goto out;
 872
 873        do {
 874                res = readdir_search_pagecache(desc);
 875
 876                if (res == -EBADCOOKIE) {
 877                        res = 0;
 878                        /* This means either end of directory */
 879                        if (*desc->dir_cookie && desc->eof == 0) {
 880                                /* Or that the server has 'lost' a cookie */
 881                                res = uncached_readdir(desc, dirent, filldir);
 882                                if (res == 0)
 883                                        continue;
 884                        }
 885                        break;
 886                }
 887                if (res == -ETOOSMALL && desc->plus) {
 888                        clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
 889                        nfs_zap_caches(inode);
 890                        desc->page_index = 0;
 891                        desc->plus = 0;
 892                        desc->eof = 0;
 893                        continue;
 894                }
 895                if (res < 0)
 896                        break;
 897
 898                res = nfs_do_filldir(desc, dirent, filldir);
 899                if (res < 0)
 900                        break;
 901        } while (!desc->eof);
 902out:
 903        nfs_unblock_sillyrename(dentry);
 904        if (res > 0)
 905                res = 0;
 906        dfprintk(FILE, "NFS: readdir(%s/%s) returns %d\n",
 907                        dentry->d_parent->d_name.name, dentry->d_name.name,
 908                        res);
 909        return res;
 910}
 911
 912static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int origin)
 913{
 914        struct dentry *dentry = filp->f_path.dentry;
 915        struct inode *inode = dentry->d_inode;
 916        struct nfs_open_dir_context *dir_ctx = filp->private_data;
 917
 918        dfprintk(FILE, "NFS: llseek dir(%s/%s, %lld, %d)\n",
 919                        dentry->d_parent->d_name.name,
 920                        dentry->d_name.name,
 921                        offset, origin);
 922
 923        mutex_lock(&inode->i_mutex);
 924        switch (origin) {
 925                case 1:
 926                        offset += filp->f_pos;
 927                case 0:
 928                        if (offset >= 0)
 929                                break;
 930                default:
 931                        offset = -EINVAL;
 932                        goto out;
 933        }
 934        if (offset != filp->f_pos) {
 935                filp->f_pos = offset;
 936                dir_ctx->dir_cookie = 0;
 937                dir_ctx->duped = 0;
 938        }
 939out:
 940        mutex_unlock(&inode->i_mutex);
 941        return offset;
 942}
 943
 944/*
 945 * All directory operations under NFS are synchronous, so fsync()
 946 * is a dummy operation.
 947 */
 948static int nfs_fsync_dir(struct file *filp, int datasync)
 949{
 950        struct dentry *dentry = filp->f_path.dentry;
 951
 952        dfprintk(FILE, "NFS: fsync dir(%s/%s) datasync %d\n",
 953                        dentry->d_parent->d_name.name, dentry->d_name.name,
 954                        datasync);
 955
 956        nfs_inc_stats(dentry->d_inode, NFSIOS_VFSFSYNC);
 957        return 0;
 958}
 959
 960/**
 961 * nfs_force_lookup_revalidate - Mark the directory as having changed
 962 * @dir - pointer to directory inode
 963 *
 964 * This forces the revalidation code in nfs_lookup_revalidate() to do a
 965 * full lookup on all child dentries of 'dir' whenever a change occurs
 966 * on the server that might have invalidated our dcache.
 967 *
 968 * The caller should be holding dir->i_lock
 969 */
 970void nfs_force_lookup_revalidate(struct inode *dir)
 971{
 972        NFS_I(dir)->cache_change_attribute++;
 973}
 974
 975/*
 976 * A check for whether or not the parent directory has changed.
 977 * In the case it has, we assume that the dentries are untrustworthy
 978 * and may need to be looked up again.
 979 */
 980static int nfs_check_verifier(struct inode *dir, struct dentry *dentry)
 981{
 982        if (IS_ROOT(dentry))
 983                return 1;
 984        if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
 985                return 0;
 986        if (!nfs_verify_change_attribute(dir, dentry->d_time))
 987                return 0;
 988        /* Revalidate nfsi->cache_change_attribute before we declare a match */
 989        if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
 990                return 0;
 991        if (!nfs_verify_change_attribute(dir, dentry->d_time))
 992                return 0;
 993        return 1;
 994}
 995
 996/*
 997 * Return the intent data that applies to this particular path component
 998 *
 999 * Note that the current set of intents only apply to the very last
1000 * component of the path.
1001 * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT.
1002 */
1003static inline unsigned int nfs_lookup_check_intent(struct nameidata *nd,
1004                                                unsigned int mask)
1005{
1006        if (nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))
1007                return 0;
1008        return nd->flags & mask;
1009}
1010
1011/*
1012 * Use intent information to check whether or not we're going to do
1013 * an O_EXCL create using this path component.
1014 */
1015static int nfs_is_exclusive_create(struct inode *dir, struct nameidata *nd)
1016{
1017        if (NFS_PROTO(dir)->version == 2)
1018                return 0;
1019        return nd && nfs_lookup_check_intent(nd, LOOKUP_EXCL);
1020}
1021
1022/*
1023 * Inode and filehandle revalidation for lookups.
1024 *
1025 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
1026 * or if the intent information indicates that we're about to open this
1027 * particular file and the "nocto" mount flag is not set.
1028 *
1029 */
1030static inline
1031int nfs_lookup_verify_inode(struct inode *inode, struct nameidata *nd)
1032{
1033        struct nfs_server *server = NFS_SERVER(inode);
1034
1035        if (IS_AUTOMOUNT(inode))
1036                return 0;
1037        if (nd != NULL) {
1038                /* VFS wants an on-the-wire revalidation */
1039                if (nd->flags & LOOKUP_REVAL)
1040                        goto out_force;
1041                /* This is an open(2) */
1042                if (nfs_lookup_check_intent(nd, LOOKUP_OPEN) != 0 &&
1043                                !(server->flags & NFS_MOUNT_NOCTO) &&
1044                                (S_ISREG(inode->i_mode) ||
1045                                 S_ISDIR(inode->i_mode)))
1046                        goto out_force;
1047                return 0;
1048        }
1049        return nfs_revalidate_inode(server, inode);
1050out_force:
1051        return __nfs_revalidate_inode(server, inode);
1052}
1053
1054/*
1055 * We judge how long we want to trust negative
1056 * dentries by looking at the parent inode mtime.
1057 *
1058 * If parent mtime has changed, we revalidate, else we wait for a
1059 * period corresponding to the parent's attribute cache timeout value.
1060 */
1061static inline
1062int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
1063                       struct nameidata *nd)
1064{
1065        /* Don't revalidate a negative dentry if we're creating a new file */
1066        if (nd != NULL && nfs_lookup_check_intent(nd, LOOKUP_CREATE) != 0)
1067                return 0;
1068        if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
1069                return 1;
1070        return !nfs_check_verifier(dir, dentry);
1071}
1072
1073/*
1074 * This is called every time the dcache has a lookup hit,
1075 * and we should check whether we can really trust that
1076 * lookup.
1077 *
1078 * NOTE! The hit can be a negative hit too, don't assume
1079 * we have an inode!
1080 *
1081 * If the parent directory is seen to have changed, we throw out the
1082 * cached dentry and do a new lookup.
1083 */
1084static int nfs_lookup_revalidate(struct dentry *dentry, struct nameidata *nd)
1085{
1086        struct inode *dir;
1087        struct inode *inode;
1088        struct dentry *parent;
1089        struct nfs_fh *fhandle = NULL;
1090        struct nfs_fattr *fattr = NULL;
1091        int error;
1092
1093        if (nd->flags & LOOKUP_RCU)
1094                return -ECHILD;
1095
1096        parent = dget_parent(dentry);
1097        dir = parent->d_inode;
1098        nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
1099        inode = dentry->d_inode;
1100
1101        if (!inode) {
1102                if (nfs_neg_need_reval(dir, dentry, nd))
1103                        goto out_bad;
1104                goto out_valid;
1105        }
1106
1107        if (is_bad_inode(inode)) {
1108                dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n",
1109                                __func__, dentry->d_parent->d_name.name,
1110                                dentry->d_name.name);
1111                goto out_bad;
1112        }
1113
1114        if (nfs_have_delegation(inode, FMODE_READ))
1115                goto out_set_verifier;
1116
1117        /* Force a full look up iff the parent directory has changed */
1118        if (!nfs_is_exclusive_create(dir, nd) && nfs_check_verifier(dir, dentry)) {
1119                if (nfs_lookup_verify_inode(inode, nd))
1120                        goto out_zap_parent;
1121                goto out_valid;
1122        }
1123
1124        if (NFS_STALE(inode))
1125                goto out_bad;
1126
1127        error = -ENOMEM;
1128        fhandle = nfs_alloc_fhandle();
1129        fattr = nfs_alloc_fattr();
1130        if (fhandle == NULL || fattr == NULL)
1131                goto out_error;
1132
1133        error = NFS_PROTO(dir)->lookup(NFS_SERVER(dir)->client, dir, &dentry->d_name, fhandle, fattr);
1134        if (error)
1135                goto out_bad;
1136        if (nfs_compare_fh(NFS_FH(inode), fhandle))
1137                goto out_bad;
1138        if ((error = nfs_refresh_inode(inode, fattr)) != 0)
1139                goto out_bad;
1140
1141        nfs_free_fattr(fattr);
1142        nfs_free_fhandle(fhandle);
1143out_set_verifier:
1144        nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1145 out_valid:
1146        dput(parent);
1147        dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is valid\n",
1148                        __func__, dentry->d_parent->d_name.name,
1149                        dentry->d_name.name);
1150        return 1;
1151out_zap_parent:
1152        nfs_zap_caches(dir);
1153 out_bad:
1154        nfs_mark_for_revalidate(dir);
1155        if (inode && S_ISDIR(inode->i_mode)) {
1156                /* Purge readdir caches. */
1157                nfs_zap_caches(inode);
1158                /* If we have submounts, don't unhash ! */
1159                if (have_submounts(dentry))
1160                        goto out_valid;
1161                if (dentry->d_flags & DCACHE_DISCONNECTED)
1162                        goto out_valid;
1163                shrink_dcache_parent(dentry);
1164        }
1165        d_drop(dentry);
1166        nfs_free_fattr(fattr);
1167        nfs_free_fhandle(fhandle);
1168        dput(parent);
1169        dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is invalid\n",
1170                        __func__, dentry->d_parent->d_name.name,
1171                        dentry->d_name.name);
1172        return 0;
1173out_error:
1174        nfs_free_fattr(fattr);
1175        nfs_free_fhandle(fhandle);
1176        dput(parent);
1177        dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) lookup returned error %d\n",
1178                        __func__, dentry->d_parent->d_name.name,
1179                        dentry->d_name.name, error);
1180        return error;
1181}
1182
1183/*
1184 * This is called from dput() when d_count is going to 0.
1185 */
1186static int nfs_dentry_delete(const struct dentry *dentry)
1187{
1188        dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n",
1189                dentry->d_parent->d_name.name, dentry->d_name.name,
1190                dentry->d_flags);
1191
1192        /* Unhash any dentry with a stale inode */
1193        if (dentry->d_inode != NULL && NFS_STALE(dentry->d_inode))
1194                return 1;
1195
1196        if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1197                /* Unhash it, so that ->d_iput() would be called */
1198                return 1;
1199        }
1200        if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
1201                /* Unhash it, so that ancestors of killed async unlink
1202                 * files will be cleaned up during umount */
1203                return 1;
1204        }
1205        return 0;
1206
1207}
1208
1209static void nfs_drop_nlink(struct inode *inode)
1210{
1211        spin_lock(&inode->i_lock);
1212        if (inode->i_nlink > 0)
1213                drop_nlink(inode);
1214        spin_unlock(&inode->i_lock);
1215}
1216
1217/*
1218 * Called when the dentry loses inode.
1219 * We use it to clean up silly-renamed files.
1220 */
1221static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
1222{
1223        if (S_ISDIR(inode->i_mode))
1224                /* drop any readdir cache as it could easily be old */
1225                NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
1226
1227        if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1228                drop_nlink(inode);
1229                nfs_complete_unlink(dentry, inode);
1230        }
1231        iput(inode);
1232}
1233
1234static void nfs_d_release(struct dentry *dentry)
1235{
1236        /* free cached devname value, if it survived that far */
1237        if (unlikely(dentry->d_fsdata)) {
1238                if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1239                        WARN_ON(1);
1240                else
1241                        kfree(dentry->d_fsdata);
1242        }
1243}
1244
1245const struct dentry_operations nfs_dentry_operations = {
1246        .d_revalidate   = nfs_lookup_revalidate,
1247        .d_delete       = nfs_dentry_delete,
1248        .d_iput         = nfs_dentry_iput,
1249        .d_automount    = nfs_d_automount,
1250        .d_release      = nfs_d_release,
1251};
1252
1253static struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
1254{
1255        struct dentry *res;
1256        struct dentry *parent;
1257        struct inode *inode = NULL;
1258        struct nfs_fh *fhandle = NULL;
1259        struct nfs_fattr *fattr = NULL;
1260        int error;
1261
1262        dfprintk(VFS, "NFS: lookup(%s/%s)\n",
1263                dentry->d_parent->d_name.name, dentry->d_name.name);
1264        nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
1265
1266        res = ERR_PTR(-ENAMETOOLONG);
1267        if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1268                goto out;
1269
1270        /*
1271         * If we're doing an exclusive create, optimize away the lookup
1272         * but don't hash the dentry.
1273         */
1274        if (nfs_is_exclusive_create(dir, nd)) {
1275                d_instantiate(dentry, NULL);
1276                res = NULL;
1277                goto out;
1278        }
1279
1280        res = ERR_PTR(-ENOMEM);
1281        fhandle = nfs_alloc_fhandle();
1282        fattr = nfs_alloc_fattr();
1283        if (fhandle == NULL || fattr == NULL)
1284                goto out;
1285
1286        parent = dentry->d_parent;
1287        /* Protect against concurrent sillydeletes */
1288        nfs_block_sillyrename(parent);
1289        error = NFS_PROTO(dir)->lookup(NFS_SERVER(dir)->client, dir, &dentry->d_name, fhandle, fattr);
1290        if (error == -ENOENT)
1291                goto no_entry;
1292        if (error < 0) {
1293                res = ERR_PTR(error);
1294                goto out_unblock_sillyrename;
1295        }
1296        inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1297        res = ERR_CAST(inode);
1298        if (IS_ERR(res))
1299                goto out_unblock_sillyrename;
1300
1301no_entry:
1302        res = d_materialise_unique(dentry, inode);
1303        if (res != NULL) {
1304                if (IS_ERR(res))
1305                        goto out_unblock_sillyrename;
1306                dentry = res;
1307        }
1308        nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1309out_unblock_sillyrename:
1310        nfs_unblock_sillyrename(parent);
1311out:
1312        nfs_free_fattr(fattr);
1313        nfs_free_fhandle(fhandle);
1314        return res;
1315}
1316
1317#ifdef CONFIG_NFS_V4
1318static int nfs_open_revalidate(struct dentry *, struct nameidata *);
1319
1320const struct dentry_operations nfs4_dentry_operations = {
1321        .d_revalidate   = nfs_open_revalidate,
1322        .d_delete       = nfs_dentry_delete,
1323        .d_iput         = nfs_dentry_iput,
1324        .d_automount    = nfs_d_automount,
1325        .d_release      = nfs_d_release,
1326};
1327
1328/*
1329 * Use intent information to determine whether we need to substitute
1330 * the NFSv4-style stateful OPEN for the LOOKUP call
1331 */
1332static int is_atomic_open(struct nameidata *nd)
1333{
1334        if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_OPEN) == 0)
1335                return 0;
1336        /* NFS does not (yet) have a stateful open for directories */
1337        if (nd->flags & LOOKUP_DIRECTORY)
1338                return 0;
1339        /* Are we trying to write to a read only partition? */
1340        if (__mnt_is_readonly(nd->path.mnt) &&
1341            (nd->intent.open.flags & (O_CREAT|O_TRUNC|FMODE_WRITE)))
1342                return 0;
1343        return 1;
1344}
1345
1346static struct nfs_open_context *nameidata_to_nfs_open_context(struct dentry *dentry, struct nameidata *nd)
1347{
1348        struct path path = {
1349                .mnt = nd->path.mnt,
1350                .dentry = dentry,
1351        };
1352        struct nfs_open_context *ctx;
1353        struct rpc_cred *cred;
1354        fmode_t fmode = nd->intent.open.flags & (FMODE_READ | FMODE_WRITE | FMODE_EXEC);
1355
1356        cred = rpc_lookup_cred();
1357        if (IS_ERR(cred))
1358                return ERR_CAST(cred);
1359        ctx = alloc_nfs_open_context(&path, cred, fmode);
1360        put_rpccred(cred);
1361        if (ctx == NULL)
1362                return ERR_PTR(-ENOMEM);
1363        return ctx;
1364}
1365
1366static int do_open(struct inode *inode, struct file *filp)
1367{
1368        nfs_fscache_set_inode_cookie(inode, filp);
1369        return 0;
1370}
1371
1372static int nfs_intent_set_file(struct nameidata *nd, struct nfs_open_context *ctx)
1373{
1374        struct file *filp;
1375        int ret = 0;
1376
1377        /* If the open_intent is for execute, we have an extra check to make */
1378        if (ctx->mode & FMODE_EXEC) {
1379                ret = nfs_may_open(ctx->path.dentry->d_inode,
1380                                ctx->cred,
1381                                nd->intent.open.flags);
1382                if (ret < 0)
1383                        goto out;
1384        }
1385        filp = lookup_instantiate_filp(nd, ctx->path.dentry, do_open);
1386        if (IS_ERR(filp))
1387                ret = PTR_ERR(filp);
1388        else
1389                nfs_file_set_open_context(filp, ctx);
1390out:
1391        put_nfs_open_context(ctx);
1392        return ret;
1393}
1394
1395static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
1396{
1397        struct nfs_open_context *ctx;
1398        struct iattr attr;
1399        struct dentry *res = NULL;
1400        struct inode *inode;
1401        int open_flags;
1402        int err;
1403
1404        dfprintk(VFS, "NFS: atomic_lookup(%s/%ld), %s\n",
1405                        dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1406
1407        /* Check that we are indeed trying to open this file */
1408        if (!is_atomic_open(nd))
1409                goto no_open;
1410
1411        if (dentry->d_name.len > NFS_SERVER(dir)->namelen) {
1412                res = ERR_PTR(-ENAMETOOLONG);
1413                goto out;
1414        }
1415
1416        /* Let vfs_create() deal with O_EXCL. Instantiate, but don't hash
1417         * the dentry. */
1418        if (nd->flags & LOOKUP_EXCL) {
1419                d_instantiate(dentry, NULL);
1420                goto out;
1421        }
1422
1423        ctx = nameidata_to_nfs_open_context(dentry, nd);
1424        res = ERR_CAST(ctx);
1425        if (IS_ERR(ctx))
1426                goto out;
1427
1428        open_flags = nd->intent.open.flags;
1429        if (nd->flags & LOOKUP_CREATE) {
1430                attr.ia_mode = nd->intent.open.create_mode;
1431                attr.ia_valid = ATTR_MODE;
1432                attr.ia_mode &= ~current_umask();
1433        } else {
1434                open_flags &= ~(O_EXCL | O_CREAT);
1435                attr.ia_valid = 0;
1436        }
1437
1438        /* Open the file on the server */
1439        nfs_block_sillyrename(dentry->d_parent);
1440        inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr);
1441        if (IS_ERR(inode)) {
1442                nfs_unblock_sillyrename(dentry->d_parent);
1443                put_nfs_open_context(ctx);
1444                switch (PTR_ERR(inode)) {
1445                        /* Make a negative dentry */
1446                        case -ENOENT:
1447                                d_add(dentry, NULL);
1448                                res = NULL;
1449                                goto out;
1450                        /* This turned out not to be a regular file */
1451                        case -ENOTDIR:
1452                                goto no_open;
1453                        case -ELOOP:
1454                                if (!(nd->intent.open.flags & O_NOFOLLOW))
1455                                        goto no_open;
1456                        /* case -EISDIR: */
1457                        /* case -EINVAL: */
1458                        default:
1459                                res = ERR_CAST(inode);
1460                                goto out;
1461                }
1462        }
1463        res = d_add_unique(dentry, inode);
1464        nfs_unblock_sillyrename(dentry->d_parent);
1465        if (res != NULL) {
1466                dput(ctx->path.dentry);
1467                ctx->path.dentry = dget(res);
1468                dentry = res;
1469        }
1470        err = nfs_intent_set_file(nd, ctx);
1471        if (err < 0) {
1472                if (res != NULL)
1473                        dput(res);
1474                return ERR_PTR(err);
1475        }
1476out:
1477        nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1478        return res;
1479no_open:
1480        return nfs_lookup(dir, dentry, nd);
1481}
1482
1483static int nfs_open_revalidate(struct dentry *dentry, struct nameidata *nd)
1484{
1485        struct dentry *parent = NULL;
1486        struct inode *inode;
1487        struct inode *dir;
1488        struct nfs_open_context *ctx;
1489        int openflags, ret = 0;
1490
1491        if (nd->flags & LOOKUP_RCU)
1492                return -ECHILD;
1493
1494        inode = dentry->d_inode;
1495        if (!is_atomic_open(nd) || d_mountpoint(dentry))
1496                goto no_open;
1497
1498        parent = dget_parent(dentry);
1499        dir = parent->d_inode;
1500
1501        /* We can't create new files in nfs_open_revalidate(), so we
1502         * optimize away revalidation of negative dentries.
1503         */
1504        if (inode == NULL) {
1505                if (!nfs_neg_need_reval(dir, dentry, nd))
1506                        ret = 1;
1507                goto out;
1508        }
1509
1510        /* NFS only supports OPEN on regular files */
1511        if (!S_ISREG(inode->i_mode))
1512                goto no_open_dput;
1513        openflags = nd->intent.open.flags;
1514        /* We cannot do exclusive creation on a positive dentry */
1515        if ((openflags & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
1516                goto no_open_dput;
1517        /* We can't create new files, or truncate existing ones here */
1518        openflags &= ~(O_CREAT|O_EXCL|O_TRUNC);
1519
1520        ctx = nameidata_to_nfs_open_context(dentry, nd);
1521        ret = PTR_ERR(ctx);
1522        if (IS_ERR(ctx))
1523                goto out;
1524        /*
1525         * Note: we're not holding inode->i_mutex and so may be racing with
1526         * operations that change the directory. We therefore save the
1527         * change attribute *before* we do the RPC call.
1528         */
1529        inode = NFS_PROTO(dir)->open_context(dir, ctx, openflags, NULL);
1530        if (IS_ERR(inode)) {
1531                ret = PTR_ERR(inode);
1532                switch (ret) {
1533                case -EPERM:
1534                case -EACCES:
1535                case -EDQUOT:
1536                case -ENOSPC:
1537                case -EROFS:
1538                        goto out_put_ctx;
1539                default:
1540                        goto out_drop;
1541                }
1542        }
1543        iput(inode);
1544        if (inode != dentry->d_inode)
1545                goto out_drop;
1546
1547        nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1548        ret = nfs_intent_set_file(nd, ctx);
1549        if (ret >= 0)
1550                ret = 1;
1551out:
1552        dput(parent);
1553        return ret;
1554out_drop:
1555        d_drop(dentry);
1556        ret = 0;
1557out_put_ctx:
1558        put_nfs_open_context(ctx);
1559        goto out;
1560
1561no_open_dput:
1562        dput(parent);
1563no_open:
1564        return nfs_lookup_revalidate(dentry, nd);
1565}
1566
1567static int nfs_open_create(struct inode *dir, struct dentry *dentry, int mode,
1568                struct nameidata *nd)
1569{
1570        struct nfs_open_context *ctx = NULL;
1571        struct iattr attr;
1572        int error;
1573        int open_flags = 0;
1574
1575        dfprintk(VFS, "NFS: create(%s/%ld), %s\n",
1576                        dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1577
1578        attr.ia_mode = mode;
1579        attr.ia_valid = ATTR_MODE;
1580
1581        if ((nd->flags & LOOKUP_CREATE) != 0) {
1582                open_flags = nd->intent.open.flags;
1583
1584                ctx = nameidata_to_nfs_open_context(dentry, nd);
1585                error = PTR_ERR(ctx);
1586                if (IS_ERR(ctx))
1587                        goto out_err_drop;
1588        }
1589
1590        error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, ctx);
1591        if (error != 0)
1592                goto out_put_ctx;
1593        if (ctx != NULL) {
1594                error = nfs_intent_set_file(nd, ctx);
1595                if (error < 0)
1596                        goto out_err;
1597        }
1598        return 0;
1599out_put_ctx:
1600        if (ctx != NULL)
1601                put_nfs_open_context(ctx);
1602out_err_drop:
1603        d_drop(dentry);
1604out_err:
1605        return error;
1606}
1607
1608#endif /* CONFIG_NFSV4 */
1609
1610/*
1611 * Code common to create, mkdir, and mknod.
1612 */
1613int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1614                                struct nfs_fattr *fattr)
1615{
1616        struct dentry *parent = dget_parent(dentry);
1617        struct inode *dir = parent->d_inode;
1618        struct inode *inode;
1619        int error = -EACCES;
1620
1621        d_drop(dentry);
1622
1623        /* We may have been initialized further down */
1624        if (dentry->d_inode)
1625                goto out;
1626        if (fhandle->size == 0) {
1627                error = NFS_PROTO(dir)->lookup(NFS_SERVER(dir)->client, dir, &dentry->d_name, fhandle, fattr);
1628                if (error)
1629                        goto out_error;
1630        }
1631        nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1632        if (!(fattr->valid & NFS_ATTR_FATTR)) {
1633                struct nfs_server *server = NFS_SB(dentry->d_sb);
1634                error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr);
1635                if (error < 0)
1636                        goto out_error;
1637        }
1638        inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1639        error = PTR_ERR(inode);
1640        if (IS_ERR(inode))
1641                goto out_error;
1642        d_add(dentry, inode);
1643out:
1644        dput(parent);
1645        return 0;
1646out_error:
1647        nfs_mark_for_revalidate(dir);
1648        dput(parent);
1649        return error;
1650}
1651
1652/*
1653 * Following a failed create operation, we drop the dentry rather
1654 * than retain a negative dentry. This avoids a problem in the event
1655 * that the operation succeeded on the server, but an error in the
1656 * reply path made it appear to have failed.
1657 */
1658static int nfs_create(struct inode *dir, struct dentry *dentry, int mode,
1659                struct nameidata *nd)
1660{
1661        struct iattr attr;
1662        int error;
1663        int open_flags = 0;
1664
1665        dfprintk(VFS, "NFS: create(%s/%ld), %s\n",
1666                        dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1667
1668        attr.ia_mode = mode;
1669        attr.ia_valid = ATTR_MODE;
1670
1671        if ((nd->flags & LOOKUP_CREATE) != 0)
1672                open_flags = nd->intent.open.flags;
1673
1674        error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, NULL);
1675        if (error != 0)
1676                goto out_err;
1677        return 0;
1678out_err:
1679        d_drop(dentry);
1680        return error;
1681}
1682
1683/*
1684 * See comments for nfs_proc_create regarding failed operations.
1685 */
1686static int
1687nfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t rdev)
1688{
1689        struct iattr attr;
1690        int status;
1691
1692        dfprintk(VFS, "NFS: mknod(%s/%ld), %s\n",
1693                        dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1694
1695        if (!new_valid_dev(rdev))
1696                return -EINVAL;
1697
1698        attr.ia_mode = mode;
1699        attr.ia_valid = ATTR_MODE;
1700
1701        status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1702        if (status != 0)
1703                goto out_err;
1704        return 0;
1705out_err:
1706        d_drop(dentry);
1707        return status;
1708}
1709
1710/*
1711 * See comments for nfs_proc_create regarding failed operations.
1712 */
1713static int nfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1714{
1715        struct iattr attr;
1716        int error;
1717
1718        dfprintk(VFS, "NFS: mkdir(%s/%ld), %s\n",
1719                        dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1720
1721        attr.ia_valid = ATTR_MODE;
1722        attr.ia_mode = mode | S_IFDIR;
1723
1724        error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1725        if (error != 0)
1726                goto out_err;
1727        return 0;
1728out_err:
1729        d_drop(dentry);
1730        return error;
1731}
1732
1733static void nfs_dentry_handle_enoent(struct dentry *dentry)
1734{
1735        if (dentry->d_inode != NULL && !d_unhashed(dentry))
1736                d_delete(dentry);
1737}
1738
1739static int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1740{
1741        int error;
1742
1743        dfprintk(VFS, "NFS: rmdir(%s/%ld), %s\n",
1744                        dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1745
1746        error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1747        /* Ensure the VFS deletes this inode */
1748        if (error == 0 && dentry->d_inode != NULL)
1749                clear_nlink(dentry->d_inode);
1750        else if (error == -ENOENT)
1751                nfs_dentry_handle_enoent(dentry);
1752
1753        return error;
1754}
1755
1756/*
1757 * Remove a file after making sure there are no pending writes,
1758 * and after checking that the file has only one user. 
1759 *
1760 * We invalidate the attribute cache and free the inode prior to the operation
1761 * to avoid possible races if the server reuses the inode.
1762 */
1763static int nfs_safe_remove(struct dentry *dentry)
1764{
1765        struct inode *dir = dentry->d_parent->d_inode;
1766        struct inode *inode = dentry->d_inode;
1767        int error = -EBUSY;
1768                
1769        dfprintk(VFS, "NFS: safe_remove(%s/%s)\n",
1770                dentry->d_parent->d_name.name, dentry->d_name.name);
1771
1772        /* If the dentry was sillyrenamed, we simply call d_delete() */
1773        if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1774                error = 0;
1775                goto out;
1776        }
1777
1778        if (inode != NULL) {
1779                nfs_inode_return_delegation(inode);
1780                error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1781                /* The VFS may want to delete this inode */
1782                if (error == 0)
1783                        nfs_drop_nlink(inode);
1784                nfs_mark_for_revalidate(inode);
1785        } else
1786                error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1787        if (error == -ENOENT)
1788                nfs_dentry_handle_enoent(dentry);
1789out:
1790        return error;
1791}
1792
1793/*  We do silly rename. In case sillyrename() returns -EBUSY, the inode
1794 *  belongs to an active ".nfs..." file and we return -EBUSY.
1795 *
1796 *  If sillyrename() returns 0, we do nothing, otherwise we unlink.
1797 */
1798static int nfs_unlink(struct inode *dir, struct dentry *dentry)
1799{
1800        int error;
1801        int need_rehash = 0;
1802
1803        dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id,
1804                dir->i_ino, dentry->d_name.name);
1805
1806        spin_lock(&dentry->d_lock);
1807        if (dentry->d_count > 1) {
1808                spin_unlock(&dentry->d_lock);
1809                /* Start asynchronous writeout of the inode */
1810                write_inode_now(dentry->d_inode, 0);
1811                error = nfs_sillyrename(dir, dentry);
1812                return error;
1813        }
1814        if (!d_unhashed(dentry)) {
1815                __d_drop(dentry);
1816                need_rehash = 1;
1817        }
1818        spin_unlock(&dentry->d_lock);
1819        error = nfs_safe_remove(dentry);
1820        if (!error || error == -ENOENT) {
1821                nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1822        } else if (need_rehash)
1823                d_rehash(dentry);
1824        return error;
1825}
1826
1827/*
1828 * To create a symbolic link, most file systems instantiate a new inode,
1829 * add a page to it containing the path, then write it out to the disk
1830 * using prepare_write/commit_write.
1831 *
1832 * Unfortunately the NFS client can't create the in-core inode first
1833 * because it needs a file handle to create an in-core inode (see
1834 * fs/nfs/inode.c:nfs_fhget).  We only have a file handle *after* the
1835 * symlink request has completed on the server.
1836 *
1837 * So instead we allocate a raw page, copy the symname into it, then do
1838 * the SYMLINK request with the page as the buffer.  If it succeeds, we
1839 * now have a new file handle and can instantiate an in-core NFS inode
1840 * and move the raw page into its mapping.
1841 */
1842static int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1843{
1844        struct pagevec lru_pvec;
1845        struct page *page;
1846        char *kaddr;
1847        struct iattr attr;
1848        unsigned int pathlen = strlen(symname);
1849        int error;
1850
1851        dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id,
1852                dir->i_ino, dentry->d_name.name, symname);
1853
1854        if (pathlen > PAGE_SIZE)
1855                return -ENAMETOOLONG;
1856
1857        attr.ia_mode = S_IFLNK | S_IRWXUGO;
1858        attr.ia_valid = ATTR_MODE;
1859
1860        page = alloc_page(GFP_HIGHUSER);
1861        if (!page)
1862                return -ENOMEM;
1863
1864        kaddr = kmap_atomic(page, KM_USER0);
1865        memcpy(kaddr, symname, pathlen);
1866        if (pathlen < PAGE_SIZE)
1867                memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
1868        kunmap_atomic(kaddr, KM_USER0);
1869
1870        error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
1871        if (error != 0) {
1872                dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s) error %d\n",
1873                        dir->i_sb->s_id, dir->i_ino,
1874                        dentry->d_name.name, symname, error);
1875                d_drop(dentry);
1876                __free_page(page);
1877                return error;
1878        }
1879
1880        /*
1881         * No big deal if we can't add this page to the page cache here.
1882         * READLINK will get the missing page from the server if needed.
1883         */
1884        pagevec_init(&lru_pvec, 0);
1885        if (!add_to_page_cache(page, dentry->d_inode->i_mapping, 0,
1886                                                        GFP_KERNEL)) {
1887                pagevec_add(&lru_pvec, page);
1888                pagevec_lru_add_file(&lru_pvec);
1889                SetPageUptodate(page);
1890                unlock_page(page);
1891        } else
1892                __free_page(page);
1893
1894        return 0;
1895}
1896
1897static int 
1898nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1899{
1900        struct inode *inode = old_dentry->d_inode;
1901        int error;
1902
1903        dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n",
1904                old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1905                dentry->d_parent->d_name.name, dentry->d_name.name);
1906
1907        nfs_inode_return_delegation(inode);
1908
1909        d_drop(dentry);
1910        error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
1911        if (error == 0) {
1912                ihold(inode);
1913                d_add(dentry, inode);
1914        }
1915        return error;
1916}
1917
1918/*
1919 * RENAME
1920 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1921 * different file handle for the same inode after a rename (e.g. when
1922 * moving to a different directory). A fail-safe method to do so would
1923 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1924 * rename the old file using the sillyrename stuff. This way, the original
1925 * file in old_dir will go away when the last process iput()s the inode.
1926 *
1927 * FIXED.
1928 * 
1929 * It actually works quite well. One needs to have the possibility for
1930 * at least one ".nfs..." file in each directory the file ever gets
1931 * moved or linked to which happens automagically with the new
1932 * implementation that only depends on the dcache stuff instead of
1933 * using the inode layer
1934 *
1935 * Unfortunately, things are a little more complicated than indicated
1936 * above. For a cross-directory move, we want to make sure we can get
1937 * rid of the old inode after the operation.  This means there must be
1938 * no pending writes (if it's a file), and the use count must be 1.
1939 * If these conditions are met, we can drop the dentries before doing
1940 * the rename.
1941 */
1942static int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1943                      struct inode *new_dir, struct dentry *new_dentry)
1944{
1945        struct inode *old_inode = old_dentry->d_inode;
1946        struct inode *new_inode = new_dentry->d_inode;
1947        struct dentry *dentry = NULL, *rehash = NULL;
1948        int error = -EBUSY;
1949
1950        dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1951                 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1952                 new_dentry->d_parent->d_name.name, new_dentry->d_name.name,
1953                 new_dentry->d_count);
1954
1955        /*
1956         * For non-directories, check whether the target is busy and if so,
1957         * make a copy of the dentry and then do a silly-rename. If the
1958         * silly-rename succeeds, the copied dentry is hashed and becomes
1959         * the new target.
1960         */
1961        if (new_inode && !S_ISDIR(new_inode->i_mode)) {
1962                /*
1963                 * To prevent any new references to the target during the
1964                 * rename, we unhash the dentry in advance.
1965                 */
1966                if (!d_unhashed(new_dentry)) {
1967                        d_drop(new_dentry);
1968                        rehash = new_dentry;
1969                }
1970
1971                if (new_dentry->d_count > 2) {
1972                        int err;
1973
1974                        /* copy the target dentry's name */
1975                        dentry = d_alloc(new_dentry->d_parent,
1976                                         &new_dentry->d_name);
1977                        if (!dentry)
1978                                goto out;
1979
1980                        /* silly-rename the existing target ... */
1981                        err = nfs_sillyrename(new_dir, new_dentry);
1982                        if (err)
1983                                goto out;
1984
1985                        new_dentry = dentry;
1986                        rehash = NULL;
1987                        new_inode = NULL;
1988                }
1989        }
1990
1991        nfs_inode_return_delegation(old_inode);
1992        if (new_inode != NULL)
1993                nfs_inode_return_delegation(new_inode);
1994
1995        error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name,
1996                                           new_dir, &new_dentry->d_name);
1997        nfs_mark_for_revalidate(old_inode);
1998out:
1999        if (rehash)
2000                d_rehash(rehash);
2001        if (!error) {
2002                if (new_inode != NULL)
2003                        nfs_drop_nlink(new_inode);
2004                d_move(old_dentry, new_dentry);
2005                nfs_set_verifier(new_dentry,
2006                                        nfs_save_change_attribute(new_dir));
2007        } else if (error == -ENOENT)
2008                nfs_dentry_handle_enoent(old_dentry);
2009
2010        /* new dentry created? */
2011        if (dentry)
2012                dput(dentry);
2013        return error;
2014}
2015
2016static DEFINE_SPINLOCK(nfs_access_lru_lock);
2017static LIST_HEAD(nfs_access_lru_list);
2018static atomic_long_t nfs_access_nr_entries;
2019
2020static void nfs_access_free_entry(struct nfs_access_entry *entry)
2021{
2022        put_rpccred(entry->cred);
2023        kfree(entry);
2024        smp_mb__before_atomic_dec();
2025        atomic_long_dec(&nfs_access_nr_entries);
2026        smp_mb__after_atomic_dec();
2027}
2028
2029static void nfs_access_free_list(struct list_head *head)
2030{
2031        struct nfs_access_entry *cache;
2032
2033        while (!list_empty(head)) {
2034                cache = list_entry(head->next, struct nfs_access_entry, lru);
2035                list_del(&cache->lru);
2036                nfs_access_free_entry(cache);
2037        }
2038}
2039
2040int nfs_access_cache_shrinker(struct shrinker *shrink,
2041                              struct shrink_control *sc)
2042{
2043        LIST_HEAD(head);
2044        struct nfs_inode *nfsi, *next;
2045        struct nfs_access_entry *cache;
2046        int nr_to_scan = sc->nr_to_scan;
2047        gfp_t gfp_mask = sc->gfp_mask;
2048
2049        if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL)
2050                return (nr_to_scan == 0) ? 0 : -1;
2051
2052        spin_lock(&nfs_access_lru_lock);
2053        list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) {
2054                struct inode *inode;
2055
2056                if (nr_to_scan-- == 0)
2057                        break;
2058                inode = &nfsi->vfs_inode;
2059                spin_lock(&inode->i_lock);
2060                if (list_empty(&nfsi->access_cache_entry_lru))
2061                        goto remove_lru_entry;
2062                cache = list_entry(nfsi->access_cache_entry_lru.next,
2063                                struct nfs_access_entry, lru);
2064                list_move(&cache->lru, &head);
2065                rb_erase(&cache->rb_node, &nfsi->access_cache);
2066                if (!list_empty(&nfsi->access_cache_entry_lru))
2067                        list_move_tail(&nfsi->access_cache_inode_lru,
2068                                        &nfs_access_lru_list);
2069                else {
2070remove_lru_entry:
2071                        list_del_init(&nfsi->access_cache_inode_lru);
2072                        smp_mb__before_clear_bit();
2073                        clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
2074                        smp_mb__after_clear_bit();
2075                }
2076                spin_unlock(&inode->i_lock);
2077        }
2078        spin_unlock(&nfs_access_lru_lock);
2079        nfs_access_free_list(&head);
2080        return (atomic_long_read(&nfs_access_nr_entries) / 100) * sysctl_vfs_cache_pressure;
2081}
2082
2083static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head)
2084{
2085        struct rb_root *root_node = &nfsi->access_cache;
2086        struct rb_node *n;
2087        struct nfs_access_entry *entry;
2088
2089        /* Unhook entries from the cache */
2090        while ((n = rb_first(root_node)) != NULL) {
2091                entry = rb_entry(n, struct nfs_access_entry, rb_node);
2092                rb_erase(n, root_node);
2093                list_move(&entry->lru, head);
2094        }
2095        nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
2096}
2097
2098void nfs_access_zap_cache(struct inode *inode)
2099{
2100        LIST_HEAD(head);
2101
2102        if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0)
2103                return;
2104        /* Remove from global LRU init */
2105        spin_lock(&nfs_access_lru_lock);
2106        if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2107                list_del_init(&NFS_I(inode)->access_cache_inode_lru);
2108
2109        spin_lock(&inode->i_lock);
2110        __nfs_access_zap_cache(NFS_I(inode), &head);
2111        spin_unlock(&inode->i_lock);
2112        spin_unlock(&nfs_access_lru_lock);
2113        nfs_access_free_list(&head);
2114}
2115
2116static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred)
2117{
2118        struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
2119        struct nfs_access_entry *entry;
2120
2121        while (n != NULL) {
2122                entry = rb_entry(n, struct nfs_access_entry, rb_node);
2123
2124                if (cred < entry->cred)
2125                        n = n->rb_left;
2126                else if (cred > entry->cred)
2127                        n = n->rb_right;
2128                else
2129                        return entry;
2130        }
2131        return NULL;
2132}
2133
2134static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
2135{
2136        struct nfs_inode *nfsi = NFS_I(inode);
2137        struct nfs_access_entry *cache;
2138        int err = -ENOENT;
2139
2140        spin_lock(&inode->i_lock);
2141        if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2142                goto out_zap;
2143        cache = nfs_access_search_rbtree(inode, cred);
2144        if (cache == NULL)
2145                goto out;
2146        if (!nfs_have_delegated_attributes(inode) &&
2147            !time_in_range_open(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo))
2148                goto out_stale;
2149        res->jiffies = cache->jiffies;
2150        res->cred = cache->cred;
2151        res->mask = cache->mask;
2152        list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
2153        err = 0;
2154out:
2155        spin_unlock(&inode->i_lock);
2156        return err;
2157out_stale:
2158        rb_erase(&cache->rb_node, &nfsi->access_cache);
2159        list_del(&cache->lru);
2160        spin_unlock(&inode->i_lock);
2161        nfs_access_free_entry(cache);
2162        return -ENOENT;
2163out_zap:
2164        spin_unlock(&inode->i_lock);
2165        nfs_access_zap_cache(inode);
2166        return -ENOENT;
2167}
2168
2169static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
2170{
2171        struct nfs_inode *nfsi = NFS_I(inode);
2172        struct rb_root *root_node = &nfsi->access_cache;
2173        struct rb_node **p = &root_node->rb_node;
2174        struct rb_node *parent = NULL;
2175        struct nfs_access_entry *entry;
2176
2177        spin_lock(&inode->i_lock);
2178        while (*p != NULL) {
2179                parent = *p;
2180                entry = rb_entry(parent, struct nfs_access_entry, rb_node);
2181
2182                if (set->cred < entry->cred)
2183                        p = &parent->rb_left;
2184                else if (set->cred > entry->cred)
2185                        p = &parent->rb_right;
2186                else
2187                        goto found;
2188        }
2189        rb_link_node(&set->rb_node, parent, p);
2190        rb_insert_color(&set->rb_node, root_node);
2191        list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2192        spin_unlock(&inode->i_lock);
2193        return;
2194found:
2195        rb_replace_node(parent, &set->rb_node, root_node);
2196        list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2197        list_del(&entry->lru);
2198        spin_unlock(&inode->i_lock);
2199        nfs_access_free_entry(entry);
2200}
2201
2202static void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
2203{
2204        struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
2205        if (cache == NULL)
2206                return;
2207        RB_CLEAR_NODE(&cache->rb_node);
2208        cache->jiffies = set->jiffies;
2209        cache->cred = get_rpccred(set->cred);
2210        cache->mask = set->mask;
2211
2212        nfs_access_add_rbtree(inode, cache);
2213
2214        /* Update accounting */
2215        smp_mb__before_atomic_inc();
2216        atomic_long_inc(&nfs_access_nr_entries);
2217        smp_mb__after_atomic_inc();
2218
2219        /* Add inode to global LRU list */
2220        if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
2221                spin_lock(&nfs_access_lru_lock);
2222                if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2223                        list_add_tail(&NFS_I(inode)->access_cache_inode_lru,
2224                                        &nfs_access_lru_list);
2225                spin_unlock(&nfs_access_lru_lock);
2226        }
2227}
2228
2229static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
2230{
2231        struct nfs_access_entry cache;
2232        int status;
2233
2234        status = nfs_access_get_cached(inode, cred, &cache);
2235        if (status == 0)
2236                goto out;
2237
2238        /* Be clever: ask server to check for all possible rights */
2239        cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ;
2240        cache.cred = cred;
2241        cache.jiffies = jiffies;
2242        status = NFS_PROTO(inode)->access(inode, &cache);
2243        if (status != 0) {
2244                if (status == -ESTALE) {
2245                        nfs_zap_caches(inode);
2246                        if (!S_ISDIR(inode->i_mode))
2247                                set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
2248                }
2249                return status;
2250        }
2251        nfs_access_add_cache(inode, &cache);
2252out:
2253        if ((mask & ~cache.mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2254                return 0;
2255        return -EACCES;
2256}
2257
2258static int nfs_open_permission_mask(int openflags)
2259{
2260        int mask = 0;
2261
2262        if (openflags & FMODE_READ)
2263                mask |= MAY_READ;
2264        if (openflags & FMODE_WRITE)
2265                mask |= MAY_WRITE;
2266        if (openflags & FMODE_EXEC)
2267                mask |= MAY_EXEC;
2268        return mask;
2269}
2270
2271int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags)
2272{
2273        return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
2274}
2275
2276int nfs_permission(struct inode *inode, int mask, unsigned int flags)
2277{
2278        struct rpc_cred *cred;
2279        int res = 0;
2280
2281        if (flags & IPERM_FLAG_RCU)
2282                return -ECHILD;
2283
2284        nfs_inc_stats(inode, NFSIOS_VFSACCESS);
2285
2286        if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2287                goto out;
2288        /* Is this sys_access() ? */
2289        if (mask & (MAY_ACCESS | MAY_CHDIR))
2290                goto force_lookup;
2291
2292        switch (inode->i_mode & S_IFMT) {
2293                case S_IFLNK:
2294                        goto out;
2295                case S_IFREG:
2296                        /* NFSv4 has atomic_open... */
2297                        if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN)
2298                                        && (mask & MAY_OPEN)
2299                                        && !(mask & MAY_EXEC))
2300                                goto out;
2301                        break;
2302                case S_IFDIR:
2303                        /*
2304                         * Optimize away all write operations, since the server
2305                         * will check permissions when we perform the op.
2306                         */
2307                        if ((mask & MAY_WRITE) && !(mask & MAY_READ))
2308                                goto out;
2309        }
2310
2311force_lookup:
2312        if (!NFS_PROTO(inode)->access)
2313                goto out_notsup;
2314
2315        cred = rpc_lookup_cred();
2316        if (!IS_ERR(cred)) {
2317                res = nfs_do_access(inode, cred, mask);
2318                put_rpccred(cred);
2319        } else
2320                res = PTR_ERR(cred);
2321out:
2322        if (!res && (mask & MAY_EXEC) && !execute_ok(inode))
2323                res = -EACCES;
2324
2325        dfprintk(VFS, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n",
2326                inode->i_sb->s_id, inode->i_ino, mask, res);
2327        return res;
2328out_notsup:
2329        res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
2330        if (res == 0)
2331                res = generic_permission(inode, mask, flags, NULL);
2332        goto out;
2333}
2334
2335/*
2336 * Local variables:
2337 *  version-control: t
2338 *  kept-new-versions: 5
2339 * End:
2340 */
2341