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/module.h>
  21#include <linux/time.h>
  22#include <linux/errno.h>
  23#include <linux/stat.h>
  24#include <linux/fcntl.h>
  25#include <linux/string.h>
  26#include <linux/kernel.h>
  27#include <linux/slab.h>
  28#include <linux/mm.h>
  29#include <linux/sunrpc/clnt.h>
  30#include <linux/nfs_fs.h>
  31#include <linux/nfs_mount.h>
  32#include <linux/pagemap.h>
  33#include <linux/pagevec.h>
  34#include <linux/namei.h>
  35#include <linux/mount.h>
  36#include <linux/sched.h>
  37#include <linux/kmemleak.h>
  38#include <linux/xattr.h>
  39
  40#include "delegation.h"
  41#include "iostat.h"
  42#include "internal.h"
  43#include "fscache.h"
  44
  45/* #define NFS_DEBUG_VERBOSE 1 */
  46
  47static int nfs_opendir(struct inode *, struct file *);
  48static int nfs_closedir(struct inode *, struct file *);
  49static int nfs_readdir(struct file *, void *, filldir_t);
  50static int nfs_fsync_dir(struct file *, loff_t, loff_t, int);
  51static loff_t nfs_llseek_dir(struct file *, loff_t, int);
  52static void nfs_readdir_clear_array(struct page*);
  53
  54const struct file_operations nfs_dir_operations = {
  55        .llseek         = nfs_llseek_dir,
  56        .read           = generic_read_dir,
  57        .readdir        = nfs_readdir,
  58        .open           = nfs_opendir,
  59        .release        = nfs_closedir,
  60        .fsync          = nfs_fsync_dir,
  61};
  62
  63const struct address_space_operations nfs_dir_aops = {
  64        .freepage = nfs_readdir_clear_array,
  65};
  66
  67static struct nfs_open_dir_context *alloc_nfs_open_dir_context(struct inode *dir, struct rpc_cred *cred)
  68{
  69        struct nfs_open_dir_context *ctx;
  70        ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
  71        if (ctx != NULL) {
  72                ctx->duped = 0;
  73                ctx->attr_gencount = NFS_I(dir)->attr_gencount;
  74                ctx->dir_cookie = 0;
  75                ctx->dup_cookie = 0;
  76                ctx->cred = get_rpccred(cred);
  77                return ctx;
  78        }
  79        return  ERR_PTR(-ENOMEM);
  80}
  81
  82static void put_nfs_open_dir_context(struct nfs_open_dir_context *ctx)
  83{
  84        put_rpccred(ctx->cred);
  85        kfree(ctx);
  86}
  87
  88/*
  89 * Open file
  90 */
  91static int
  92nfs_opendir(struct inode *inode, struct file *filp)
  93{
  94        int res = 0;
  95        struct nfs_open_dir_context *ctx;
  96        struct rpc_cred *cred;
  97
  98        dfprintk(FILE, "NFS: open dir(%s/%s)\n",
  99                        filp->f_path.dentry->d_parent->d_name.name,
 100                        filp->f_path.dentry->d_name.name);
 101
 102        nfs_inc_stats(inode, NFSIOS_VFSOPEN);
 103
 104        cred = rpc_lookup_cred();
 105        if (IS_ERR(cred))
 106                return PTR_ERR(cred);
 107        ctx = alloc_nfs_open_dir_context(inode, cred);
 108        if (IS_ERR(ctx)) {
 109                res = PTR_ERR(ctx);
 110                goto out;
 111        }
 112        filp->private_data = ctx;
 113        if (filp->f_path.dentry == filp->f_path.mnt->mnt_root) {
 114                /* This is a mountpoint, so d_revalidate will never
 115                 * have been called, so we need to refresh the
 116                 * inode (for close-open consistency) ourselves.
 117                 */
 118                __nfs_revalidate_inode(NFS_SERVER(inode), inode);
 119        }
 120out:
 121        put_rpccred(cred);
 122        return res;
 123}
 124
 125static int
 126nfs_closedir(struct inode *inode, struct file *filp)
 127{
 128        put_nfs_open_dir_context(filp->private_data);
 129        return 0;
 130}
 131
 132struct nfs_cache_array_entry {
 133        u64 cookie;
 134        u64 ino;
 135        struct qstr string;
 136        unsigned char d_type;
 137};
 138
 139struct nfs_cache_array {
 140        int size;
 141        int eof_index;
 142        u64 last_cookie;
 143        struct nfs_cache_array_entry array[0];
 144};
 145
 146typedef int (*decode_dirent_t)(struct xdr_stream *, struct nfs_entry *, int);
 147typedef struct {
 148        struct file     *file;
 149        struct page     *page;
 150        unsigned long   page_index;
 151        u64             *dir_cookie;
 152        u64             last_cookie;
 153        loff_t          current_index;
 154        decode_dirent_t decode;
 155
 156        unsigned long   timestamp;
 157        unsigned long   gencount;
 158        unsigned int    cache_entry_index;
 159        unsigned int    plus:1;
 160        unsigned int    eof:1;
 161} nfs_readdir_descriptor_t;
 162
 163/*
 164 * The caller is responsible for calling nfs_readdir_release_array(page)
 165 */
 166static
 167struct nfs_cache_array *nfs_readdir_get_array(struct page *page)
 168{
 169        void *ptr;
 170        if (page == NULL)
 171                return ERR_PTR(-EIO);
 172        ptr = kmap(page);
 173        if (ptr == NULL)
 174                return ERR_PTR(-ENOMEM);
 175        return ptr;
 176}
 177
 178static
 179void nfs_readdir_release_array(struct page *page)
 180{
 181        kunmap(page);
 182}
 183
 184/*
 185 * we are freeing strings created by nfs_add_to_readdir_array()
 186 */
 187static
 188void nfs_readdir_clear_array(struct page *page)
 189{
 190        struct nfs_cache_array *array;
 191        int i;
 192
 193        array = kmap_atomic(page);
 194        for (i = 0; i < array->size; i++)
 195                kfree(array->array[i].string.name);
 196        kunmap_atomic(array);
 197}
 198
 199/*
 200 * the caller is responsible for freeing qstr.name
 201 * when called by nfs_readdir_add_to_array, the strings will be freed in
 202 * nfs_clear_readdir_array()
 203 */
 204static
 205int nfs_readdir_make_qstr(struct qstr *string, const char *name, unsigned int len)
 206{
 207        string->len = len;
 208        string->name = kmemdup(name, len, GFP_KERNEL);
 209        if (string->name == NULL)
 210                return -ENOMEM;
 211        /*
 212         * Avoid a kmemleak false positive. The pointer to the name is stored
 213         * in a page cache page which kmemleak does not scan.
 214         */
 215        kmemleak_not_leak(string->name);
 216        string->hash = full_name_hash(name, len);
 217        return 0;
 218}
 219
 220static
 221int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page)
 222{
 223        struct nfs_cache_array *array = nfs_readdir_get_array(page);
 224        struct nfs_cache_array_entry *cache_entry;
 225        int ret;
 226
 227        if (IS_ERR(array))
 228                return PTR_ERR(array);
 229
 230        cache_entry = &array->array[array->size];
 231
 232        /* Check that this entry lies within the page bounds */
 233        ret = -ENOSPC;
 234        if ((char *)&cache_entry[1] - (char *)page_address(page) > PAGE_SIZE)
 235                goto out;
 236
 237        cache_entry->cookie = entry->prev_cookie;
 238        cache_entry->ino = entry->ino;
 239        cache_entry->d_type = entry->d_type;
 240        ret = nfs_readdir_make_qstr(&cache_entry->string, entry->name, entry->len);
 241        if (ret)
 242                goto out;
 243        array->last_cookie = entry->cookie;
 244        array->size++;
 245        if (entry->eof != 0)
 246                array->eof_index = array->size;
 247out:
 248        nfs_readdir_release_array(page);
 249        return ret;
 250}
 251
 252static
 253int nfs_readdir_search_for_pos(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
 254{
 255        loff_t diff = desc->file->f_pos - desc->current_index;
 256        unsigned int index;
 257
 258        if (diff < 0)
 259                goto out_eof;
 260        if (diff >= array->size) {
 261                if (array->eof_index >= 0)
 262                        goto out_eof;
 263                return -EAGAIN;
 264        }
 265
 266        index = (unsigned int)diff;
 267        *desc->dir_cookie = array->array[index].cookie;
 268        desc->cache_entry_index = index;
 269        return 0;
 270out_eof:
 271        desc->eof = 1;
 272        return -EBADCOOKIE;
 273}
 274
 275static
 276int nfs_readdir_search_for_cookie(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
 277{
 278        int i;
 279        loff_t new_pos;
 280        int status = -EAGAIN;
 281
 282        for (i = 0; i < array->size; i++) {
 283                if (array->array[i].cookie == *desc->dir_cookie) {
 284                        struct nfs_inode *nfsi = NFS_I(desc->file->f_path.dentry->d_inode);
 285                        struct nfs_open_dir_context *ctx = desc->file->private_data;
 286
 287                        new_pos = desc->current_index + i;
 288                        if (ctx->attr_gencount != nfsi->attr_gencount
 289                            || (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA))) {
 290                                ctx->duped = 0;
 291                                ctx->attr_gencount = nfsi->attr_gencount;
 292                        } else if (new_pos < desc->file->f_pos) {
 293                                if (ctx->duped > 0
 294                                    && ctx->dup_cookie == *desc->dir_cookie) {
 295                                        if (printk_ratelimit()) {
 296                                                pr_notice("NFS: directory %s/%s contains a readdir loop."
 297                                                                "Please contact your server vendor.  "
 298                                                                "The file: %s has duplicate cookie %llu\n",
 299                                                                desc->file->f_dentry->d_parent->d_name.name,
 300                                                                desc->file->f_dentry->d_name.name,
 301                                                                array->array[i].string.name,
 302                                                                *desc->dir_cookie);
 303                                        }
 304                                        status = -ELOOP;
 305                                        goto out;
 306                                }
 307                                ctx->dup_cookie = *desc->dir_cookie;
 308                                ctx->duped = -1;
 309                        }
 310                        desc->file->f_pos = new_pos;
 311                        desc->cache_entry_index = i;
 312                        return 0;
 313                }
 314        }
 315        if (array->eof_index >= 0) {
 316                status = -EBADCOOKIE;
 317                if (*desc->dir_cookie == array->last_cookie)
 318                        desc->eof = 1;
 319        }
 320out:
 321        return status;
 322}
 323
 324static
 325int nfs_readdir_search_array(nfs_readdir_descriptor_t *desc)
 326{
 327        struct nfs_cache_array *array;
 328        int status;
 329
 330        array = nfs_readdir_get_array(desc->page);
 331        if (IS_ERR(array)) {
 332                status = PTR_ERR(array);
 333                goto out;
 334        }
 335
 336        if (*desc->dir_cookie == 0)
 337                status = nfs_readdir_search_for_pos(array, desc);
 338        else
 339                status = nfs_readdir_search_for_cookie(array, desc);
 340
 341        if (status == -EAGAIN) {
 342                desc->last_cookie = array->last_cookie;
 343                desc->current_index += array->size;
 344                desc->page_index++;
 345        }
 346        nfs_readdir_release_array(desc->page);
 347out:
 348        return status;
 349}
 350
 351/* Fill a page with xdr information before transferring to the cache page */
 352static
 353int nfs_readdir_xdr_filler(struct page **pages, nfs_readdir_descriptor_t *desc,
 354                        struct nfs_entry *entry, struct file *file, struct inode *inode)
 355{
 356        struct nfs_open_dir_context *ctx = file->private_data;
 357        struct rpc_cred *cred = ctx->cred;
 358        unsigned long   timestamp, gencount;
 359        int             error;
 360
 361 again:
 362        timestamp = jiffies;
 363        gencount = nfs_inc_attr_generation_counter();
 364        error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, entry->cookie, pages,
 365                                          NFS_SERVER(inode)->dtsize, desc->plus);
 366        if (error < 0) {
 367                /* We requested READDIRPLUS, but the server doesn't grok it */
 368                if (error == -ENOTSUPP && desc->plus) {
 369                        NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
 370                        clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
 371                        desc->plus = 0;
 372                        goto again;
 373                }
 374                goto error;
 375        }
 376        desc->timestamp = timestamp;
 377        desc->gencount = gencount;
 378error:
 379        return error;
 380}
 381
 382static int xdr_decode(nfs_readdir_descriptor_t *desc,
 383                      struct nfs_entry *entry, struct xdr_stream *xdr)
 384{
 385        int error;
 386
 387        error = desc->decode(xdr, entry, desc->plus);
 388        if (error)
 389                return error;
 390        entry->fattr->time_start = desc->timestamp;
 391        entry->fattr->gencount = desc->gencount;
 392        return 0;
 393}
 394
 395static
 396int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry)
 397{
 398        if (dentry->d_inode == NULL)
 399                goto different;
 400        if (nfs_compare_fh(entry->fh, NFS_FH(dentry->d_inode)) != 0)
 401                goto different;
 402        return 1;
 403different:
 404        return 0;
 405}
 406
 407static
 408bool nfs_use_readdirplus(struct inode *dir, struct file *filp)
 409{
 410        if (!nfs_server_capable(dir, NFS_CAP_READDIRPLUS))
 411                return false;
 412        if (test_and_clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags))
 413                return true;
 414        if (filp->f_pos == 0)
 415                return true;
 416        return false;
 417}
 418
 419/*
 420 * This function is called by the lookup code to request the use of
 421 * readdirplus to accelerate any future lookups in the same
 422 * directory.
 423 */
 424static
 425void nfs_advise_use_readdirplus(struct inode *dir)
 426{
 427        set_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags);
 428}
 429
 430static
 431void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry)
 432{
 433        struct qstr filename = QSTR_INIT(entry->name, entry->len);
 434        struct dentry *dentry;
 435        struct dentry *alias;
 436        struct inode *dir = parent->d_inode;
 437        struct inode *inode;
 438
 439        if (filename.name[0] == '.') {
 440                if (filename.len == 1)
 441                        return;
 442                if (filename.len == 2 && filename.name[1] == '.')
 443                        return;
 444        }
 445        filename.hash = full_name_hash(filename.name, filename.len);
 446
 447        dentry = d_lookup(parent, &filename);
 448        if (dentry != NULL) {
 449                if (nfs_same_file(dentry, entry)) {
 450                        nfs_refresh_inode(dentry->d_inode, entry->fattr);
 451                        goto out;
 452                } else {
 453                        if (d_invalidate(dentry) != 0)
 454                                goto out;
 455                        dput(dentry);
 456                }
 457        }
 458
 459        dentry = d_alloc(parent, &filename);
 460        if (dentry == NULL)
 461                return;
 462
 463        inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr);
 464        if (IS_ERR(inode))
 465                goto out;
 466
 467        alias = d_materialise_unique(dentry, inode);
 468        if (IS_ERR(alias))
 469                goto out;
 470        else if (alias) {
 471                nfs_set_verifier(alias, nfs_save_change_attribute(dir));
 472                dput(alias);
 473        } else
 474                nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
 475
 476out:
 477        dput(dentry);
 478}
 479
 480/* Perform conversion from xdr to cache array */
 481static
 482int nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry,
 483                                struct page **xdr_pages, struct page *page, unsigned int buflen)
 484{
 485        struct xdr_stream stream;
 486        struct xdr_buf buf;
 487        struct page *scratch;
 488        struct nfs_cache_array *array;
 489        unsigned int count = 0;
 490        int status;
 491
 492        scratch = alloc_page(GFP_KERNEL);
 493        if (scratch == NULL)
 494                return -ENOMEM;
 495
 496        xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen);
 497        xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
 498
 499        do {
 500                status = xdr_decode(desc, entry, &stream);
 501                if (status != 0) {
 502                        if (status == -EAGAIN)
 503                                status = 0;
 504                        break;
 505                }
 506
 507                count++;
 508
 509                if (desc->plus != 0)
 510                        nfs_prime_dcache(desc->file->f_path.dentry, entry);
 511
 512                status = nfs_readdir_add_to_array(entry, page);
 513                if (status != 0)
 514                        break;
 515        } while (!entry->eof);
 516
 517        if (count == 0 || (status == -EBADCOOKIE && entry->eof != 0)) {
 518                array = nfs_readdir_get_array(page);
 519                if (!IS_ERR(array)) {
 520                        array->eof_index = array->size;
 521                        status = 0;
 522                        nfs_readdir_release_array(page);
 523                } else
 524                        status = PTR_ERR(array);
 525        }
 526
 527        put_page(scratch);
 528        return status;
 529}
 530
 531static
 532void nfs_readdir_free_pagearray(struct page **pages, unsigned int npages)
 533{
 534        unsigned int i;
 535        for (i = 0; i < npages; i++)
 536                put_page(pages[i]);
 537}
 538
 539static
 540void nfs_readdir_free_large_page(void *ptr, struct page **pages,
 541                unsigned int npages)
 542{
 543        nfs_readdir_free_pagearray(pages, npages);
 544}
 545
 546/*
 547 * nfs_readdir_large_page will allocate pages that must be freed with a call
 548 * to nfs_readdir_free_large_page
 549 */
 550static
 551int nfs_readdir_large_page(struct page **pages, unsigned int npages)
 552{
 553        unsigned int i;
 554
 555        for (i = 0; i < npages; i++) {
 556                struct page *page = alloc_page(GFP_KERNEL);
 557                if (page == NULL)
 558                        goto out_freepages;
 559                pages[i] = page;
 560        }
 561        return 0;
 562
 563out_freepages:
 564        nfs_readdir_free_pagearray(pages, i);
 565        return -ENOMEM;
 566}
 567
 568static
 569int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t *desc, struct page *page, struct inode *inode)
 570{
 571        struct page *pages[NFS_MAX_READDIR_PAGES];
 572        void *pages_ptr = NULL;
 573        struct nfs_entry entry;
 574        struct file     *file = desc->file;
 575        struct nfs_cache_array *array;
 576        int status = -ENOMEM;
 577        unsigned int array_size = ARRAY_SIZE(pages);
 578
 579        entry.prev_cookie = 0;
 580        entry.cookie = desc->last_cookie;
 581        entry.eof = 0;
 582        entry.fh = nfs_alloc_fhandle();
 583        entry.fattr = nfs_alloc_fattr();
 584        entry.server = NFS_SERVER(inode);
 585        if (entry.fh == NULL || entry.fattr == NULL)
 586                goto out;
 587
 588        array = nfs_readdir_get_array(page);
 589        if (IS_ERR(array)) {
 590                status = PTR_ERR(array);
 591                goto out;
 592        }
 593        memset(array, 0, sizeof(struct nfs_cache_array));
 594        array->eof_index = -1;
 595
 596        status = nfs_readdir_large_page(pages, array_size);
 597        if (status < 0)
 598                goto out_release_array;
 599        do {
 600                unsigned int pglen;
 601                status = nfs_readdir_xdr_filler(pages, desc, &entry, file, inode);
 602
 603                if (status < 0)
 604                        break;
 605                pglen = status;
 606                status = nfs_readdir_page_filler(desc, &entry, pages, page, pglen);
 607                if (status < 0) {
 608                        if (status == -ENOSPC)
 609                                status = 0;
 610                        break;
 611                }
 612        } while (array->eof_index < 0);
 613
 614        nfs_readdir_free_large_page(pages_ptr, pages, array_size);
 615out_release_array:
 616        nfs_readdir_release_array(page);
 617out:
 618        nfs_free_fattr(entry.fattr);
 619        nfs_free_fhandle(entry.fh);
 620        return status;
 621}
 622
 623/*
 624 * Now we cache directories properly, by converting xdr information
 625 * to an array that can be used for lookups later.  This results in
 626 * fewer cache pages, since we can store more information on each page.
 627 * We only need to convert from xdr once so future lookups are much simpler
 628 */
 629static
 630int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page* page)
 631{
 632        struct inode    *inode = desc->file->f_path.dentry->d_inode;
 633        int ret;
 634
 635        ret = nfs_readdir_xdr_to_array(desc, page, inode);
 636        if (ret < 0)
 637                goto error;
 638        SetPageUptodate(page);
 639
 640        if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) {
 641                /* Should never happen */
 642                nfs_zap_mapping(inode, inode->i_mapping);
 643        }
 644        unlock_page(page);
 645        return 0;
 646 error:
 647        unlock_page(page);
 648        return ret;
 649}
 650
 651static
 652void cache_page_release(nfs_readdir_descriptor_t *desc)
 653{
 654        if (!desc->page->mapping)
 655                nfs_readdir_clear_array(desc->page);
 656        page_cache_release(desc->page);
 657        desc->page = NULL;
 658}
 659
 660static
 661struct page *get_cache_page(nfs_readdir_descriptor_t *desc)
 662{
 663        return read_cache_page(desc->file->f_path.dentry->d_inode->i_mapping,
 664                        desc->page_index, (filler_t *)nfs_readdir_filler, desc);
 665}
 666
 667/*
 668 * Returns 0 if desc->dir_cookie was found on page desc->page_index
 669 */
 670static
 671int find_cache_page(nfs_readdir_descriptor_t *desc)
 672{
 673        int res;
 674
 675        desc->page = get_cache_page(desc);
 676        if (IS_ERR(desc->page))
 677                return PTR_ERR(desc->page);
 678
 679        res = nfs_readdir_search_array(desc);
 680        if (res != 0)
 681                cache_page_release(desc);
 682        return res;
 683}
 684
 685/* Search for desc->dir_cookie from the beginning of the page cache */
 686static inline
 687int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
 688{
 689        int res;
 690
 691        if (desc->page_index == 0) {
 692                desc->current_index = 0;
 693                desc->last_cookie = 0;
 694        }
 695        do {
 696                res = find_cache_page(desc);
 697        } while (res == -EAGAIN);
 698        return res;
 699}
 700
 701/*
 702 * Once we've found the start of the dirent within a page: fill 'er up...
 703 */
 704static 
 705int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent,
 706                   filldir_t filldir)
 707{
 708        struct file     *file = desc->file;
 709        int i = 0;
 710        int res = 0;
 711        struct nfs_cache_array *array = NULL;
 712        struct nfs_open_dir_context *ctx = file->private_data;
 713
 714        array = nfs_readdir_get_array(desc->page);
 715        if (IS_ERR(array)) {
 716                res = PTR_ERR(array);
 717                goto out;
 718        }
 719
 720        for (i = desc->cache_entry_index; i < array->size; i++) {
 721                struct nfs_cache_array_entry *ent;
 722
 723                ent = &array->array[i];
 724                if (filldir(dirent, ent->string.name, ent->string.len,
 725                    file->f_pos, nfs_compat_user_ino64(ent->ino),
 726                    ent->d_type) < 0) {
 727                        desc->eof = 1;
 728                        break;
 729                }
 730                file->f_pos++;
 731                if (i < (array->size-1))
 732                        *desc->dir_cookie = array->array[i+1].cookie;
 733                else
 734                        *desc->dir_cookie = array->last_cookie;
 735                if (ctx->duped != 0)
 736                        ctx->duped = 1;
 737        }
 738        if (array->eof_index >= 0)
 739                desc->eof = 1;
 740
 741        nfs_readdir_release_array(desc->page);
 742out:
 743        cache_page_release(desc);
 744        dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
 745                        (unsigned long long)*desc->dir_cookie, res);
 746        return res;
 747}
 748
 749/*
 750 * If we cannot find a cookie in our cache, we suspect that this is
 751 * because it points to a deleted file, so we ask the server to return
 752 * whatever it thinks is the next entry. We then feed this to filldir.
 753 * If all goes well, we should then be able to find our way round the
 754 * cache on the next call to readdir_search_pagecache();
 755 *
 756 * NOTE: we cannot add the anonymous page to the pagecache because
 757 *       the data it contains might not be page aligned. Besides,
 758 *       we should already have a complete representation of the
 759 *       directory in the page cache by the time we get here.
 760 */
 761static inline
 762int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent,
 763                     filldir_t filldir)
 764{
 765        struct page     *page = NULL;
 766        int             status;
 767        struct inode *inode = desc->file->f_path.dentry->d_inode;
 768        struct nfs_open_dir_context *ctx = desc->file->private_data;
 769
 770        dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n",
 771                        (unsigned long long)*desc->dir_cookie);
 772
 773        page = alloc_page(GFP_HIGHUSER);
 774        if (!page) {
 775                status = -ENOMEM;
 776                goto out;
 777        }
 778
 779        desc->page_index = 0;
 780        desc->last_cookie = *desc->dir_cookie;
 781        desc->page = page;
 782        ctx->duped = 0;
 783
 784        status = nfs_readdir_xdr_to_array(desc, page, inode);
 785        if (status < 0)
 786                goto out_release;
 787
 788        status = nfs_do_filldir(desc, dirent, filldir);
 789
 790 out:
 791        dfprintk(DIRCACHE, "NFS: %s: returns %d\n",
 792                        __func__, status);
 793        return status;
 794 out_release:
 795        cache_page_release(desc);
 796        goto out;
 797}
 798
 799/* The file offset position represents the dirent entry number.  A
 800   last cookie cache takes care of the common case of reading the
 801   whole directory.
 802 */
 803static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
 804{
 805        struct dentry   *dentry = filp->f_path.dentry;
 806        struct inode    *inode = dentry->d_inode;
 807        nfs_readdir_descriptor_t my_desc,
 808                        *desc = &my_desc;
 809        struct nfs_open_dir_context *dir_ctx = filp->private_data;
 810        int res;
 811
 812        dfprintk(FILE, "NFS: readdir(%s/%s) starting at cookie %llu\n",
 813                        dentry->d_parent->d_name.name, dentry->d_name.name,
 814                        (long long)filp->f_pos);
 815        nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
 816
 817        /*
 818         * filp->f_pos points to the dirent entry number.
 819         * *desc->dir_cookie has the cookie for the next entry. We have
 820         * to either find the entry with the appropriate number or
 821         * revalidate the cookie.
 822         */
 823        memset(desc, 0, sizeof(*desc));
 824
 825        desc->file = filp;
 826        desc->dir_cookie = &dir_ctx->dir_cookie;
 827        desc->decode = NFS_PROTO(inode)->decode_dirent;
 828        desc->plus = nfs_use_readdirplus(inode, filp) ? 1 : 0;
 829
 830        nfs_block_sillyrename(dentry);
 831        res = nfs_revalidate_mapping(inode, filp->f_mapping);
 832        if (res < 0)
 833                goto out;
 834
 835        do {
 836                res = readdir_search_pagecache(desc);
 837
 838                if (res == -EBADCOOKIE) {
 839                        res = 0;
 840                        /* This means either end of directory */
 841                        if (*desc->dir_cookie && desc->eof == 0) {
 842                                /* Or that the server has 'lost' a cookie */
 843                                res = uncached_readdir(desc, dirent, filldir);
 844                                if (res == 0)
 845                                        continue;
 846                        }
 847                        break;
 848                }
 849                if (res == -ETOOSMALL && desc->plus) {
 850                        clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
 851                        nfs_zap_caches(inode);
 852                        desc->page_index = 0;
 853                        desc->plus = 0;
 854                        desc->eof = 0;
 855                        continue;
 856                }
 857                if (res < 0)
 858                        break;
 859
 860                res = nfs_do_filldir(desc, dirent, filldir);
 861                if (res < 0)
 862                        break;
 863        } while (!desc->eof);
 864out:
 865        nfs_unblock_sillyrename(dentry);
 866        if (res > 0)
 867                res = 0;
 868        dfprintk(FILE, "NFS: readdir(%s/%s) returns %d\n",
 869                        dentry->d_parent->d_name.name, dentry->d_name.name,
 870                        res);
 871        return res;
 872}
 873
 874static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int origin)
 875{
 876        struct dentry *dentry = filp->f_path.dentry;
 877        struct inode *inode = dentry->d_inode;
 878        struct nfs_open_dir_context *dir_ctx = filp->private_data;
 879
 880        dfprintk(FILE, "NFS: llseek dir(%s/%s, %lld, %d)\n",
 881                        dentry->d_parent->d_name.name,
 882                        dentry->d_name.name,
 883                        offset, origin);
 884
 885        mutex_lock(&inode->i_mutex);
 886        switch (origin) {
 887                case 1:
 888                        offset += filp->f_pos;
 889                case 0:
 890                        if (offset >= 0)
 891                                break;
 892                default:
 893                        offset = -EINVAL;
 894                        goto out;
 895        }
 896        if (offset != filp->f_pos) {
 897                filp->f_pos = offset;
 898                dir_ctx->dir_cookie = 0;
 899                dir_ctx->duped = 0;
 900        }
 901out:
 902        mutex_unlock(&inode->i_mutex);
 903        return offset;
 904}
 905
 906/*
 907 * All directory operations under NFS are synchronous, so fsync()
 908 * is a dummy operation.
 909 */
 910static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end,
 911                         int datasync)
 912{
 913        struct dentry *dentry = filp->f_path.dentry;
 914        struct inode *inode = dentry->d_inode;
 915
 916        dfprintk(FILE, "NFS: fsync dir(%s/%s) datasync %d\n",
 917                        dentry->d_parent->d_name.name, dentry->d_name.name,
 918                        datasync);
 919
 920        mutex_lock(&inode->i_mutex);
 921        nfs_inc_stats(dentry->d_inode, NFSIOS_VFSFSYNC);
 922        mutex_unlock(&inode->i_mutex);
 923        return 0;
 924}
 925
 926/**
 927 * nfs_force_lookup_revalidate - Mark the directory as having changed
 928 * @dir - pointer to directory inode
 929 *
 930 * This forces the revalidation code in nfs_lookup_revalidate() to do a
 931 * full lookup on all child dentries of 'dir' whenever a change occurs
 932 * on the server that might have invalidated our dcache.
 933 *
 934 * The caller should be holding dir->i_lock
 935 */
 936void nfs_force_lookup_revalidate(struct inode *dir)
 937{
 938        NFS_I(dir)->cache_change_attribute++;
 939}
 940EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate);
 941
 942/*
 943 * A check for whether or not the parent directory has changed.
 944 * In the case it has, we assume that the dentries are untrustworthy
 945 * and may need to be looked up again.
 946 */
 947static int nfs_check_verifier(struct inode *dir, struct dentry *dentry)
 948{
 949        if (IS_ROOT(dentry))
 950                return 1;
 951        if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
 952                return 0;
 953        if (!nfs_verify_change_attribute(dir, dentry->d_time))
 954                return 0;
 955        /* Revalidate nfsi->cache_change_attribute before we declare a match */
 956        if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
 957                return 0;
 958        if (!nfs_verify_change_attribute(dir, dentry->d_time))
 959                return 0;
 960        return 1;
 961}
 962
 963/*
 964 * Use intent information to check whether or not we're going to do
 965 * an O_EXCL create using this path component.
 966 */
 967static int nfs_is_exclusive_create(struct inode *dir, unsigned int flags)
 968{
 969        if (NFS_PROTO(dir)->version == 2)
 970                return 0;
 971        return flags & LOOKUP_EXCL;
 972}
 973
 974/*
 975 * Inode and filehandle revalidation for lookups.
 976 *
 977 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
 978 * or if the intent information indicates that we're about to open this
 979 * particular file and the "nocto" mount flag is not set.
 980 *
 981 */
 982static inline
 983int nfs_lookup_verify_inode(struct inode *inode, unsigned int flags)
 984{
 985        struct nfs_server *server = NFS_SERVER(inode);
 986
 987        if (IS_AUTOMOUNT(inode))
 988                return 0;
 989        /* VFS wants an on-the-wire revalidation */
 990        if (flags & LOOKUP_REVAL)
 991                goto out_force;
 992        /* This is an open(2) */
 993        if ((flags & LOOKUP_OPEN) && !(server->flags & NFS_MOUNT_NOCTO) &&
 994            (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)))
 995                goto out_force;
 996        return 0;
 997out_force:
 998        return __nfs_revalidate_inode(server, inode);
 999}
1000
1001/*
1002 * We judge how long we want to trust negative
1003 * dentries by looking at the parent inode mtime.
1004 *
1005 * If parent mtime has changed, we revalidate, else we wait for a
1006 * period corresponding to the parent's attribute cache timeout value.
1007 */
1008static inline
1009int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
1010                       unsigned int flags)
1011{
1012        /* Don't revalidate a negative dentry if we're creating a new file */
1013        if (flags & LOOKUP_CREATE)
1014                return 0;
1015        if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
1016                return 1;
1017        return !nfs_check_verifier(dir, dentry);
1018}
1019
1020/*
1021 * This is called every time the dcache has a lookup hit,
1022 * and we should check whether we can really trust that
1023 * lookup.
1024 *
1025 * NOTE! The hit can be a negative hit too, don't assume
1026 * we have an inode!
1027 *
1028 * If the parent directory is seen to have changed, we throw out the
1029 * cached dentry and do a new lookup.
1030 */
1031static int nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1032{
1033        struct inode *dir;
1034        struct inode *inode;
1035        struct dentry *parent;
1036        struct nfs_fh *fhandle = NULL;
1037        struct nfs_fattr *fattr = NULL;
1038        int error;
1039
1040        if (flags & LOOKUP_RCU)
1041                return -ECHILD;
1042
1043        parent = dget_parent(dentry);
1044        dir = parent->d_inode;
1045        nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
1046        inode = dentry->d_inode;
1047
1048        if (!inode) {
1049                if (nfs_neg_need_reval(dir, dentry, flags))
1050                        goto out_bad;
1051                goto out_valid_noent;
1052        }
1053
1054        if (is_bad_inode(inode)) {
1055                dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n",
1056                                __func__, dentry->d_parent->d_name.name,
1057                                dentry->d_name.name);
1058                goto out_bad;
1059        }
1060
1061        if (NFS_PROTO(dir)->have_delegation(inode, FMODE_READ))
1062                goto out_set_verifier;
1063
1064        /* Force a full look up iff the parent directory has changed */
1065        if (!nfs_is_exclusive_create(dir, flags) && nfs_check_verifier(dir, dentry)) {
1066                if (nfs_lookup_verify_inode(inode, flags))
1067                        goto out_zap_parent;
1068                goto out_valid;
1069        }
1070
1071        if (NFS_STALE(inode))
1072                goto out_bad;
1073
1074        error = -ENOMEM;
1075        fhandle = nfs_alloc_fhandle();
1076        fattr = nfs_alloc_fattr();
1077        if (fhandle == NULL || fattr == NULL)
1078                goto out_error;
1079
1080        error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1081        if (error)
1082                goto out_bad;
1083        if (nfs_compare_fh(NFS_FH(inode), fhandle))
1084                goto out_bad;
1085        if ((error = nfs_refresh_inode(inode, fattr)) != 0)
1086                goto out_bad;
1087
1088        nfs_free_fattr(fattr);
1089        nfs_free_fhandle(fhandle);
1090out_set_verifier:
1091        nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1092 out_valid:
1093        /* Success: notify readdir to use READDIRPLUS */
1094        nfs_advise_use_readdirplus(dir);
1095 out_valid_noent:
1096        dput(parent);
1097        dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is valid\n",
1098                        __func__, dentry->d_parent->d_name.name,
1099                        dentry->d_name.name);
1100        return 1;
1101out_zap_parent:
1102        nfs_zap_caches(dir);
1103 out_bad:
1104        nfs_free_fattr(fattr);
1105        nfs_free_fhandle(fhandle);
1106        nfs_mark_for_revalidate(dir);
1107        if (inode && S_ISDIR(inode->i_mode)) {
1108                /* Purge readdir caches. */
1109                nfs_zap_caches(inode);
1110                /* If we have submounts, don't unhash ! */
1111                if (have_submounts(dentry))
1112                        goto out_valid;
1113                if (dentry->d_flags & DCACHE_DISCONNECTED)
1114                        goto out_valid;
1115                shrink_dcache_parent(dentry);
1116        }
1117        d_drop(dentry);
1118        dput(parent);
1119        dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is invalid\n",
1120                        __func__, dentry->d_parent->d_name.name,
1121                        dentry->d_name.name);
1122        return 0;
1123out_error:
1124        nfs_free_fattr(fattr);
1125        nfs_free_fhandle(fhandle);
1126        dput(parent);
1127        dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) lookup returned error %d\n",
1128                        __func__, dentry->d_parent->d_name.name,
1129                        dentry->d_name.name, error);
1130        return error;
1131}
1132
1133/*
1134 * This is called from dput() when d_count is going to 0.
1135 */
1136static int nfs_dentry_delete(const struct dentry *dentry)
1137{
1138        dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n",
1139                dentry->d_parent->d_name.name, dentry->d_name.name,
1140                dentry->d_flags);
1141
1142        /* Unhash any dentry with a stale inode */
1143        if (dentry->d_inode != NULL && NFS_STALE(dentry->d_inode))
1144                return 1;
1145
1146        if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1147                /* Unhash it, so that ->d_iput() would be called */
1148                return 1;
1149        }
1150        if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
1151                /* Unhash it, so that ancestors of killed async unlink
1152                 * files will be cleaned up during umount */
1153                return 1;
1154        }
1155        return 0;
1156
1157}
1158
1159/* Ensure that we revalidate inode->i_nlink */
1160static void nfs_drop_nlink(struct inode *inode)
1161{
1162        spin_lock(&inode->i_lock);
1163        /* drop the inode if we're reasonably sure this is the last link */
1164        if (inode->i_nlink == 1)
1165                clear_nlink(inode);
1166        NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATTR;
1167        spin_unlock(&inode->i_lock);
1168}
1169
1170/*
1171 * Called when the dentry loses inode.
1172 * We use it to clean up silly-renamed files.
1173 */
1174static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
1175{
1176        if (S_ISDIR(inode->i_mode))
1177                /* drop any readdir cache as it could easily be old */
1178                NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
1179
1180        if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1181                nfs_complete_unlink(dentry, inode);
1182                nfs_drop_nlink(inode);
1183        }
1184        iput(inode);
1185}
1186
1187static void nfs_d_release(struct dentry *dentry)
1188{
1189        /* free cached devname value, if it survived that far */
1190        if (unlikely(dentry->d_fsdata)) {
1191                if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1192                        WARN_ON(1);
1193                else
1194                        kfree(dentry->d_fsdata);
1195        }
1196}
1197
1198const struct dentry_operations nfs_dentry_operations = {
1199        .d_revalidate   = nfs_lookup_revalidate,
1200        .d_delete       = nfs_dentry_delete,
1201        .d_iput         = nfs_dentry_iput,
1202        .d_automount    = nfs_d_automount,
1203        .d_release      = nfs_d_release,
1204};
1205EXPORT_SYMBOL_GPL(nfs_dentry_operations);
1206
1207struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
1208{
1209        struct dentry *res;
1210        struct dentry *parent;
1211        struct inode *inode = NULL;
1212        struct nfs_fh *fhandle = NULL;
1213        struct nfs_fattr *fattr = NULL;
1214        int error;
1215
1216        dfprintk(VFS, "NFS: lookup(%s/%s)\n",
1217                dentry->d_parent->d_name.name, dentry->d_name.name);
1218        nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
1219
1220        res = ERR_PTR(-ENAMETOOLONG);
1221        if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1222                goto out;
1223
1224        /*
1225         * If we're doing an exclusive create, optimize away the lookup
1226         * but don't hash the dentry.
1227         */
1228        if (nfs_is_exclusive_create(dir, flags)) {
1229                d_instantiate(dentry, NULL);
1230                res = NULL;
1231                goto out;
1232        }
1233
1234        res = ERR_PTR(-ENOMEM);
1235        fhandle = nfs_alloc_fhandle();
1236        fattr = nfs_alloc_fattr();
1237        if (fhandle == NULL || fattr == NULL)
1238                goto out;
1239
1240        parent = dentry->d_parent;
1241        /* Protect against concurrent sillydeletes */
1242        nfs_block_sillyrename(parent);
1243        error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1244        if (error == -ENOENT)
1245                goto no_entry;
1246        if (error < 0) {
1247                res = ERR_PTR(error);
1248                goto out_unblock_sillyrename;
1249        }
1250        inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1251        res = ERR_CAST(inode);
1252        if (IS_ERR(res))
1253                goto out_unblock_sillyrename;
1254
1255        /* Success: notify readdir to use READDIRPLUS */
1256        nfs_advise_use_readdirplus(dir);
1257
1258no_entry:
1259        res = d_materialise_unique(dentry, inode);
1260        if (res != NULL) {
1261                if (IS_ERR(res))
1262                        goto out_unblock_sillyrename;
1263                dentry = res;
1264        }
1265        nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1266out_unblock_sillyrename:
1267        nfs_unblock_sillyrename(parent);
1268out:
1269        nfs_free_fattr(fattr);
1270        nfs_free_fhandle(fhandle);
1271        return res;
1272}
1273EXPORT_SYMBOL_GPL(nfs_lookup);
1274
1275#if IS_ENABLED(CONFIG_NFS_V4)
1276static int nfs4_lookup_revalidate(struct dentry *, unsigned int);
1277
1278const struct dentry_operations nfs4_dentry_operations = {
1279        .d_revalidate   = nfs4_lookup_revalidate,
1280        .d_delete       = nfs_dentry_delete,
1281        .d_iput         = nfs_dentry_iput,
1282        .d_automount    = nfs_d_automount,
1283        .d_release      = nfs_d_release,
1284};
1285EXPORT_SYMBOL_GPL(nfs4_dentry_operations);
1286
1287static fmode_t flags_to_mode(int flags)
1288{
1289        fmode_t res = (__force fmode_t)flags & FMODE_EXEC;
1290        if ((flags & O_ACCMODE) != O_WRONLY)
1291                res |= FMODE_READ;
1292        if ((flags & O_ACCMODE) != O_RDONLY)
1293                res |= FMODE_WRITE;
1294        return res;
1295}
1296
1297static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags)
1298{
1299        return alloc_nfs_open_context(dentry, flags_to_mode(open_flags));
1300}
1301
1302static int do_open(struct inode *inode, struct file *filp)
1303{
1304        nfs_fscache_set_inode_cookie(inode, filp);
1305        return 0;
1306}
1307
1308static int nfs_finish_open(struct nfs_open_context *ctx,
1309                           struct dentry *dentry,
1310                           struct file *file, unsigned open_flags,
1311                           int *opened)
1312{
1313        int err;
1314
1315        if (ctx->dentry != dentry) {
1316                dput(ctx->dentry);
1317                ctx->dentry = dget(dentry);
1318        }
1319
1320        /* If the open_intent is for execute, we have an extra check to make */
1321        if (ctx->mode & FMODE_EXEC) {
1322                err = nfs_may_open(dentry->d_inode, ctx->cred, open_flags);
1323                if (err < 0)
1324                        goto out;
1325        }
1326
1327        err = finish_open(file, dentry, do_open, opened);
1328        if (err)
1329                goto out;
1330        nfs_file_set_open_context(file, ctx);
1331
1332out:
1333        put_nfs_open_context(ctx);
1334        return err;
1335}
1336
1337int nfs_atomic_open(struct inode *dir, struct dentry *dentry,
1338                    struct file *file, unsigned open_flags,
1339                    umode_t mode, int *opened)
1340{
1341        struct nfs_open_context *ctx;
1342        struct dentry *res;
1343        struct iattr attr = { .ia_valid = ATTR_OPEN };
1344        struct inode *inode;
1345        int err;
1346
1347        /* Expect a negative dentry */
1348        BUG_ON(dentry->d_inode);
1349
1350        dfprintk(VFS, "NFS: atomic_open(%s/%ld), %s\n",
1351                        dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1352
1353        /* NFS only supports OPEN on regular files */
1354        if ((open_flags & O_DIRECTORY)) {
1355                if (!d_unhashed(dentry)) {
1356                        /*
1357                         * Hashed negative dentry with O_DIRECTORY: dentry was
1358                         * revalidated and is fine, no need to perform lookup
1359                         * again
1360                         */
1361                        return -ENOENT;
1362                }
1363                goto no_open;
1364        }
1365
1366        if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1367                return -ENAMETOOLONG;
1368
1369        if (open_flags & O_CREAT) {
1370                attr.ia_valid |= ATTR_MODE;
1371                attr.ia_mode = mode & ~current_umask();
1372        }
1373        if (open_flags & O_TRUNC) {
1374                attr.ia_valid |= ATTR_SIZE;
1375                attr.ia_size = 0;
1376        }
1377
1378        ctx = create_nfs_open_context(dentry, open_flags);
1379        err = PTR_ERR(ctx);
1380        if (IS_ERR(ctx))
1381                goto out;
1382
1383        nfs_block_sillyrename(dentry->d_parent);
1384        inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr);
1385        d_drop(dentry);
1386        if (IS_ERR(inode)) {
1387                nfs_unblock_sillyrename(dentry->d_parent);
1388                put_nfs_open_context(ctx);
1389                err = PTR_ERR(inode);
1390                switch (err) {
1391                case -ENOENT:
1392                        d_add(dentry, NULL);
1393                        break;
1394                case -EISDIR:
1395                case -ENOTDIR:
1396                        goto no_open;
1397                case -ELOOP:
1398                        if (!(open_flags & O_NOFOLLOW))
1399                                goto no_open;
1400                        break;
1401                        /* case -EINVAL: */
1402                default:
1403                        break;
1404                }
1405                goto out;
1406        }
1407        res = d_add_unique(dentry, inode);
1408        if (res != NULL)
1409                dentry = res;
1410
1411        nfs_unblock_sillyrename(dentry->d_parent);
1412        nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1413
1414        err = nfs_finish_open(ctx, dentry, file, open_flags, opened);
1415
1416        dput(res);
1417out:
1418        return err;
1419
1420no_open:
1421        res = nfs_lookup(dir, dentry, 0);
1422        err = PTR_ERR(res);
1423        if (IS_ERR(res))
1424                goto out;
1425
1426        return finish_no_open(file, res);
1427}
1428EXPORT_SYMBOL_GPL(nfs_atomic_open);
1429
1430static int nfs4_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1431{
1432        struct dentry *parent = NULL;
1433        struct inode *inode;
1434        struct inode *dir;
1435        int ret = 0;
1436
1437        if (flags & LOOKUP_RCU)
1438                return -ECHILD;
1439
1440        if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY))
1441                goto no_open;
1442        if (d_mountpoint(dentry))
1443                goto no_open;
1444
1445        inode = dentry->d_inode;
1446        parent = dget_parent(dentry);
1447        dir = parent->d_inode;
1448
1449        /* We can't create new files in nfs_open_revalidate(), so we
1450         * optimize away revalidation of negative dentries.
1451         */
1452        if (inode == NULL) {
1453                if (!nfs_neg_need_reval(dir, dentry, flags))
1454                        ret = 1;
1455                goto out;
1456        }
1457
1458        /* NFS only supports OPEN on regular files */
1459        if (!S_ISREG(inode->i_mode))
1460                goto no_open_dput;
1461        /* We cannot do exclusive creation on a positive dentry */
1462        if (flags & LOOKUP_EXCL)
1463                goto no_open_dput;
1464
1465        /* Let f_op->open() actually open (and revalidate) the file */
1466        ret = 1;
1467
1468out:
1469        dput(parent);
1470        return ret;
1471
1472no_open_dput:
1473        dput(parent);
1474no_open:
1475        return nfs_lookup_revalidate(dentry, flags);
1476}
1477
1478#endif /* CONFIG_NFSV4 */
1479
1480/*
1481 * Code common to create, mkdir, and mknod.
1482 */
1483int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1484                                struct nfs_fattr *fattr)
1485{
1486        struct dentry *parent = dget_parent(dentry);
1487        struct inode *dir = parent->d_inode;
1488        struct inode *inode;
1489        int error = -EACCES;
1490
1491        d_drop(dentry);
1492
1493        /* We may have been initialized further down */
1494        if (dentry->d_inode)
1495                goto out;
1496        if (fhandle->size == 0) {
1497                error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1498                if (error)
1499                        goto out_error;
1500        }
1501        nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1502        if (!(fattr->valid & NFS_ATTR_FATTR)) {
1503                struct nfs_server *server = NFS_SB(dentry->d_sb);
1504                error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr);
1505                if (error < 0)
1506                        goto out_error;
1507        }
1508        inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1509        error = PTR_ERR(inode);
1510        if (IS_ERR(inode))
1511                goto out_error;
1512        d_add(dentry, inode);
1513out:
1514        dput(parent);
1515        return 0;
1516out_error:
1517        nfs_mark_for_revalidate(dir);
1518        dput(parent);
1519        return error;
1520}
1521EXPORT_SYMBOL_GPL(nfs_instantiate);
1522
1523/*
1524 * Following a failed create operation, we drop the dentry rather
1525 * than retain a negative dentry. This avoids a problem in the event
1526 * that the operation succeeded on the server, but an error in the
1527 * reply path made it appear to have failed.
1528 */
1529int nfs_create(struct inode *dir, struct dentry *dentry,
1530                umode_t mode, bool excl)
1531{
1532        struct iattr attr;
1533        int open_flags = excl ? O_CREAT | O_EXCL : O_CREAT;
1534        int error;
1535
1536        dfprintk(VFS, "NFS: create(%s/%ld), %s\n",
1537                        dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1538
1539        attr.ia_mode = mode;
1540        attr.ia_valid = ATTR_MODE;
1541
1542        error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags);
1543        if (error != 0)
1544                goto out_err;
1545        return 0;
1546out_err:
1547        d_drop(dentry);
1548        return error;
1549}
1550EXPORT_SYMBOL_GPL(nfs_create);
1551
1552/*
1553 * See comments for nfs_proc_create regarding failed operations.
1554 */
1555int
1556nfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t rdev)
1557{
1558        struct iattr attr;
1559        int status;
1560
1561        dfprintk(VFS, "NFS: mknod(%s/%ld), %s\n",
1562                        dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1563
1564        if (!new_valid_dev(rdev))
1565                return -EINVAL;
1566
1567        attr.ia_mode = mode;
1568        attr.ia_valid = ATTR_MODE;
1569
1570        status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1571        if (status != 0)
1572                goto out_err;
1573        return 0;
1574out_err:
1575        d_drop(dentry);
1576        return status;
1577}
1578EXPORT_SYMBOL_GPL(nfs_mknod);
1579
1580/*
1581 * See comments for nfs_proc_create regarding failed operations.
1582 */
1583int nfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1584{
1585        struct iattr attr;
1586        int error;
1587
1588        dfprintk(VFS, "NFS: mkdir(%s/%ld), %s\n",
1589                        dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1590
1591        attr.ia_valid = ATTR_MODE;
1592        attr.ia_mode = mode | S_IFDIR;
1593
1594        error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1595        if (error != 0)
1596                goto out_err;
1597        return 0;
1598out_err:
1599        d_drop(dentry);
1600        return error;
1601}
1602EXPORT_SYMBOL_GPL(nfs_mkdir);
1603
1604static void nfs_dentry_handle_enoent(struct dentry *dentry)
1605{
1606        if (dentry->d_inode != NULL && !d_unhashed(dentry))
1607                d_delete(dentry);
1608}
1609
1610int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1611{
1612        int error;
1613
1614        dfprintk(VFS, "NFS: rmdir(%s/%ld), %s\n",
1615                        dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1616
1617        error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1618        /* Ensure the VFS deletes this inode */
1619        if (error == 0 && dentry->d_inode != NULL)
1620                clear_nlink(dentry->d_inode);
1621        else if (error == -ENOENT)
1622                nfs_dentry_handle_enoent(dentry);
1623
1624        return error;
1625}
1626EXPORT_SYMBOL_GPL(nfs_rmdir);
1627
1628/*
1629 * Remove a file after making sure there are no pending writes,
1630 * and after checking that the file has only one user. 
1631 *
1632 * We invalidate the attribute cache and free the inode prior to the operation
1633 * to avoid possible races if the server reuses the inode.
1634 */
1635static int nfs_safe_remove(struct dentry *dentry)
1636{
1637        struct inode *dir = dentry->d_parent->d_inode;
1638        struct inode *inode = dentry->d_inode;
1639        int error = -EBUSY;
1640                
1641        dfprintk(VFS, "NFS: safe_remove(%s/%s)\n",
1642                dentry->d_parent->d_name.name, dentry->d_name.name);
1643
1644        /* If the dentry was sillyrenamed, we simply call d_delete() */
1645        if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1646                error = 0;
1647                goto out;
1648        }
1649
1650        if (inode != NULL) {
1651                NFS_PROTO(inode)->return_delegation(inode);
1652                error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1653                if (error == 0)
1654                        nfs_drop_nlink(inode);
1655        } else
1656                error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1657        if (error == -ENOENT)
1658                nfs_dentry_handle_enoent(dentry);
1659out:
1660        return error;
1661}
1662
1663/*  We do silly rename. In case sillyrename() returns -EBUSY, the inode
1664 *  belongs to an active ".nfs..." file and we return -EBUSY.
1665 *
1666 *  If sillyrename() returns 0, we do nothing, otherwise we unlink.
1667 */
1668int nfs_unlink(struct inode *dir, struct dentry *dentry)
1669{
1670        int error;
1671        int need_rehash = 0;
1672
1673        dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id,
1674                dir->i_ino, dentry->d_name.name);
1675
1676        spin_lock(&dentry->d_lock);
1677        if (dentry->d_count > 1) {
1678                spin_unlock(&dentry->d_lock);
1679                /* Start asynchronous writeout of the inode */
1680                write_inode_now(dentry->d_inode, 0);
1681                error = nfs_sillyrename(dir, dentry);
1682                return error;
1683        }
1684        if (!d_unhashed(dentry)) {
1685                __d_drop(dentry);
1686                need_rehash = 1;
1687        }
1688        spin_unlock(&dentry->d_lock);
1689        error = nfs_safe_remove(dentry);
1690        if (!error || error == -ENOENT) {
1691                nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1692        } else if (need_rehash)
1693                d_rehash(dentry);
1694        return error;
1695}
1696EXPORT_SYMBOL_GPL(nfs_unlink);
1697
1698/*
1699 * To create a symbolic link, most file systems instantiate a new inode,
1700 * add a page to it containing the path, then write it out to the disk
1701 * using prepare_write/commit_write.
1702 *
1703 * Unfortunately the NFS client can't create the in-core inode first
1704 * because it needs a file handle to create an in-core inode (see
1705 * fs/nfs/inode.c:nfs_fhget).  We only have a file handle *after* the
1706 * symlink request has completed on the server.
1707 *
1708 * So instead we allocate a raw page, copy the symname into it, then do
1709 * the SYMLINK request with the page as the buffer.  If it succeeds, we
1710 * now have a new file handle and can instantiate an in-core NFS inode
1711 * and move the raw page into its mapping.
1712 */
1713int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1714{
1715        struct pagevec lru_pvec;
1716        struct page *page;
1717        char *kaddr;
1718        struct iattr attr;
1719        unsigned int pathlen = strlen(symname);
1720        int error;
1721
1722        dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id,
1723                dir->i_ino, dentry->d_name.name, symname);
1724
1725        if (pathlen > PAGE_SIZE)
1726                return -ENAMETOOLONG;
1727
1728        attr.ia_mode = S_IFLNK | S_IRWXUGO;
1729        attr.ia_valid = ATTR_MODE;
1730
1731        page = alloc_page(GFP_HIGHUSER);
1732        if (!page)
1733                return -ENOMEM;
1734
1735        kaddr = kmap_atomic(page);
1736        memcpy(kaddr, symname, pathlen);
1737        if (pathlen < PAGE_SIZE)
1738                memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
1739        kunmap_atomic(kaddr);
1740
1741        error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
1742        if (error != 0) {
1743                dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s) error %d\n",
1744                        dir->i_sb->s_id, dir->i_ino,
1745                        dentry->d_name.name, symname, error);
1746                d_drop(dentry);
1747                __free_page(page);
1748                return error;
1749        }
1750
1751        /*
1752         * No big deal if we can't add this page to the page cache here.
1753         * READLINK will get the missing page from the server if needed.
1754         */
1755        pagevec_init(&lru_pvec, 0);
1756        if (!add_to_page_cache(page, dentry->d_inode->i_mapping, 0,
1757                                                        GFP_KERNEL)) {
1758                pagevec_add(&lru_pvec, page);
1759                pagevec_lru_add_file(&lru_pvec);
1760                SetPageUptodate(page);
1761                unlock_page(page);
1762        } else
1763                __free_page(page);
1764
1765        return 0;
1766}
1767EXPORT_SYMBOL_GPL(nfs_symlink);
1768
1769int
1770nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1771{
1772        struct inode *inode = old_dentry->d_inode;
1773        int error;
1774
1775        dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n",
1776                old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1777                dentry->d_parent->d_name.name, dentry->d_name.name);
1778
1779        NFS_PROTO(inode)->return_delegation(inode);
1780
1781        d_drop(dentry);
1782        error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
1783        if (error == 0) {
1784                ihold(inode);
1785                d_add(dentry, inode);
1786        }
1787        return error;
1788}
1789EXPORT_SYMBOL_GPL(nfs_link);
1790
1791/*
1792 * RENAME
1793 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1794 * different file handle for the same inode after a rename (e.g. when
1795 * moving to a different directory). A fail-safe method to do so would
1796 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1797 * rename the old file using the sillyrename stuff. This way, the original
1798 * file in old_dir will go away when the last process iput()s the inode.
1799 *
1800 * FIXED.
1801 * 
1802 * It actually works quite well. One needs to have the possibility for
1803 * at least one ".nfs..." file in each directory the file ever gets
1804 * moved or linked to which happens automagically with the new
1805 * implementation that only depends on the dcache stuff instead of
1806 * using the inode layer
1807 *
1808 * Unfortunately, things are a little more complicated than indicated
1809 * above. For a cross-directory move, we want to make sure we can get
1810 * rid of the old inode after the operation.  This means there must be
1811 * no pending writes (if it's a file), and the use count must be 1.
1812 * If these conditions are met, we can drop the dentries before doing
1813 * the rename.
1814 */
1815int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1816                      struct inode *new_dir, struct dentry *new_dentry)
1817{
1818        struct inode *old_inode = old_dentry->d_inode;
1819        struct inode *new_inode = new_dentry->d_inode;
1820        struct dentry *dentry = NULL, *rehash = NULL;
1821        int error = -EBUSY;
1822
1823        dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1824                 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1825                 new_dentry->d_parent->d_name.name, new_dentry->d_name.name,
1826                 new_dentry->d_count);
1827
1828        /*
1829         * For non-directories, check whether the target is busy and if so,
1830         * make a copy of the dentry and then do a silly-rename. If the
1831         * silly-rename succeeds, the copied dentry is hashed and becomes
1832         * the new target.
1833         */
1834        if (new_inode && !S_ISDIR(new_inode->i_mode)) {
1835                /*
1836                 * To prevent any new references to the target during the
1837                 * rename, we unhash the dentry in advance.
1838                 */
1839                if (!d_unhashed(new_dentry)) {
1840                        d_drop(new_dentry);
1841                        rehash = new_dentry;
1842                }
1843
1844                if (new_dentry->d_count > 2) {
1845                        int err;
1846
1847                        /* copy the target dentry's name */
1848                        dentry = d_alloc(new_dentry->d_parent,
1849                                         &new_dentry->d_name);
1850                        if (!dentry)
1851                                goto out;
1852
1853                        /* silly-rename the existing target ... */
1854                        err = nfs_sillyrename(new_dir, new_dentry);
1855                        if (err)
1856                                goto out;
1857
1858                        new_dentry = dentry;
1859                        rehash = NULL;
1860                        new_inode = NULL;
1861                }
1862        }
1863
1864        NFS_PROTO(old_inode)->return_delegation(old_inode);
1865        if (new_inode != NULL)
1866                NFS_PROTO(new_inode)->return_delegation(new_inode);
1867
1868        error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name,
1869                                           new_dir, &new_dentry->d_name);
1870        nfs_mark_for_revalidate(old_inode);
1871out:
1872        if (rehash)
1873                d_rehash(rehash);
1874        if (!error) {
1875                if (new_inode != NULL)
1876                        nfs_drop_nlink(new_inode);
1877                d_move(old_dentry, new_dentry);
1878                nfs_set_verifier(new_dentry,
1879                                        nfs_save_change_attribute(new_dir));
1880        } else if (error == -ENOENT)
1881                nfs_dentry_handle_enoent(old_dentry);
1882
1883        /* new dentry created? */
1884        if (dentry)
1885                dput(dentry);
1886        return error;
1887}
1888EXPORT_SYMBOL_GPL(nfs_rename);
1889
1890static DEFINE_SPINLOCK(nfs_access_lru_lock);
1891static LIST_HEAD(nfs_access_lru_list);
1892static atomic_long_t nfs_access_nr_entries;
1893
1894static void nfs_access_free_entry(struct nfs_access_entry *entry)
1895{
1896        put_rpccred(entry->cred);
1897        kfree(entry);
1898        smp_mb__before_atomic_dec();
1899        atomic_long_dec(&nfs_access_nr_entries);
1900        smp_mb__after_atomic_dec();
1901}
1902
1903static void nfs_access_free_list(struct list_head *head)
1904{
1905        struct nfs_access_entry *cache;
1906
1907        while (!list_empty(head)) {
1908                cache = list_entry(head->next, struct nfs_access_entry, lru);
1909                list_del(&cache->lru);
1910                nfs_access_free_entry(cache);
1911        }
1912}
1913
1914int nfs_access_cache_shrinker(struct shrinker *shrink,
1915                              struct shrink_control *sc)
1916{
1917        LIST_HEAD(head);
1918        struct nfs_inode *nfsi, *next;
1919        struct nfs_access_entry *cache;
1920        int nr_to_scan = sc->nr_to_scan;
1921        gfp_t gfp_mask = sc->gfp_mask;
1922
1923        if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL)
1924                return (nr_to_scan == 0) ? 0 : -1;
1925
1926        spin_lock(&nfs_access_lru_lock);
1927        list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) {
1928                struct inode *inode;
1929
1930                if (nr_to_scan-- == 0)
1931                        break;
1932                inode = &nfsi->vfs_inode;
1933                spin_lock(&inode->i_lock);
1934                if (list_empty(&nfsi->access_cache_entry_lru))
1935                        goto remove_lru_entry;
1936                cache = list_entry(nfsi->access_cache_entry_lru.next,
1937                                struct nfs_access_entry, lru);
1938                list_move(&cache->lru, &head);
1939                rb_erase(&cache->rb_node, &nfsi->access_cache);
1940                if (!list_empty(&nfsi->access_cache_entry_lru))
1941                        list_move_tail(&nfsi->access_cache_inode_lru,
1942                                        &nfs_access_lru_list);
1943                else {
1944remove_lru_entry:
1945                        list_del_init(&nfsi->access_cache_inode_lru);
1946                        smp_mb__before_clear_bit();
1947                        clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
1948                        smp_mb__after_clear_bit();
1949                }
1950                spin_unlock(&inode->i_lock);
1951        }
1952        spin_unlock(&nfs_access_lru_lock);
1953        nfs_access_free_list(&head);
1954        return (atomic_long_read(&nfs_access_nr_entries) / 100) * sysctl_vfs_cache_pressure;
1955}
1956
1957static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head)
1958{
1959        struct rb_root *root_node = &nfsi->access_cache;
1960        struct rb_node *n;
1961        struct nfs_access_entry *entry;
1962
1963        /* Unhook entries from the cache */
1964        while ((n = rb_first(root_node)) != NULL) {
1965                entry = rb_entry(n, struct nfs_access_entry, rb_node);
1966                rb_erase(n, root_node);
1967                list_move(&entry->lru, head);
1968        }
1969        nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
1970}
1971
1972void nfs_access_zap_cache(struct inode *inode)
1973{
1974        LIST_HEAD(head);
1975
1976        if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0)
1977                return;
1978        /* Remove from global LRU init */
1979        spin_lock(&nfs_access_lru_lock);
1980        if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
1981                list_del_init(&NFS_I(inode)->access_cache_inode_lru);
1982
1983        spin_lock(&inode->i_lock);
1984        __nfs_access_zap_cache(NFS_I(inode), &head);
1985        spin_unlock(&inode->i_lock);
1986        spin_unlock(&nfs_access_lru_lock);
1987        nfs_access_free_list(&head);
1988}
1989EXPORT_SYMBOL_GPL(nfs_access_zap_cache);
1990
1991static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred)
1992{
1993        struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
1994        struct nfs_access_entry *entry;
1995
1996        while (n != NULL) {
1997                entry = rb_entry(n, struct nfs_access_entry, rb_node);
1998
1999                if (cred < entry->cred)
2000                        n = n->rb_left;
2001                else if (cred > entry->cred)
2002                        n = n->rb_right;
2003                else
2004                        return entry;
2005        }
2006        return NULL;
2007}
2008
2009static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
2010{
2011        struct nfs_inode *nfsi = NFS_I(inode);
2012        struct nfs_access_entry *cache;
2013        int err = -ENOENT;
2014
2015        spin_lock(&inode->i_lock);
2016        if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2017                goto out_zap;
2018        cache = nfs_access_search_rbtree(inode, cred);
2019        if (cache == NULL)
2020                goto out;
2021        if (!nfs_have_delegated_attributes(inode) &&
2022            !time_in_range_open(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo))
2023                goto out_stale;
2024        res->jiffies = cache->jiffies;
2025        res->cred = cache->cred;
2026        res->mask = cache->mask;
2027        list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
2028        err = 0;
2029out:
2030        spin_unlock(&inode->i_lock);
2031        return err;
2032out_stale:
2033        rb_erase(&cache->rb_node, &nfsi->access_cache);
2034        list_del(&cache->lru);
2035        spin_unlock(&inode->i_lock);
2036        nfs_access_free_entry(cache);
2037        return -ENOENT;
2038out_zap:
2039        spin_unlock(&inode->i_lock);
2040        nfs_access_zap_cache(inode);
2041        return -ENOENT;
2042}
2043
2044static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
2045{
2046        struct nfs_inode *nfsi = NFS_I(inode);
2047        struct rb_root *root_node = &nfsi->access_cache;
2048        struct rb_node **p = &root_node->rb_node;
2049        struct rb_node *parent = NULL;
2050        struct nfs_access_entry *entry;
2051
2052        spin_lock(&inode->i_lock);
2053        while (*p != NULL) {
2054                parent = *p;
2055                entry = rb_entry(parent, struct nfs_access_entry, rb_node);
2056
2057                if (set->cred < entry->cred)
2058                        p = &parent->rb_left;
2059                else if (set->cred > entry->cred)
2060                        p = &parent->rb_right;
2061                else
2062                        goto found;
2063        }
2064        rb_link_node(&set->rb_node, parent, p);
2065        rb_insert_color(&set->rb_node, root_node);
2066        list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2067        spin_unlock(&inode->i_lock);
2068        return;
2069found:
2070        rb_replace_node(parent, &set->rb_node, root_node);
2071        list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2072        list_del(&entry->lru);
2073        spin_unlock(&inode->i_lock);
2074        nfs_access_free_entry(entry);
2075}
2076
2077void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
2078{
2079        struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
2080        if (cache == NULL)
2081                return;
2082        RB_CLEAR_NODE(&cache->rb_node);
2083        cache->jiffies = set->jiffies;
2084        cache->cred = get_rpccred(set->cred);
2085        cache->mask = set->mask;
2086
2087        nfs_access_add_rbtree(inode, cache);
2088
2089        /* Update accounting */
2090        smp_mb__before_atomic_inc();
2091        atomic_long_inc(&nfs_access_nr_entries);
2092        smp_mb__after_atomic_inc();
2093
2094        /* Add inode to global LRU list */
2095        if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
2096                spin_lock(&nfs_access_lru_lock);
2097                if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2098                        list_add_tail(&NFS_I(inode)->access_cache_inode_lru,
2099                                        &nfs_access_lru_list);
2100                spin_unlock(&nfs_access_lru_lock);
2101        }
2102}
2103EXPORT_SYMBOL_GPL(nfs_access_add_cache);
2104
2105void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result)
2106{
2107        entry->mask = 0;
2108        if (access_result & NFS4_ACCESS_READ)
2109                entry->mask |= MAY_READ;
2110        if (access_result &
2111            (NFS4_ACCESS_MODIFY | NFS4_ACCESS_EXTEND | NFS4_ACCESS_DELETE))
2112                entry->mask |= MAY_WRITE;
2113        if (access_result & (NFS4_ACCESS_LOOKUP|NFS4_ACCESS_EXECUTE))
2114                entry->mask |= MAY_EXEC;
2115}
2116EXPORT_SYMBOL_GPL(nfs_access_set_mask);
2117
2118static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
2119{
2120        struct nfs_access_entry cache;
2121        int status;
2122
2123        status = nfs_access_get_cached(inode, cred, &cache);
2124        if (status == 0)
2125                goto out;
2126
2127        /* Be clever: ask server to check for all possible rights */
2128        cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ;
2129        cache.cred = cred;
2130        cache.jiffies = jiffies;
2131        status = NFS_PROTO(inode)->access(inode, &cache);
2132        if (status != 0) {
2133                if (status == -ESTALE) {
2134                        nfs_zap_caches(inode);
2135                        if (!S_ISDIR(inode->i_mode))
2136                                set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
2137                }
2138                return status;
2139        }
2140        nfs_access_add_cache(inode, &cache);
2141out:
2142        if ((mask & ~cache.mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2143                return 0;
2144        return -EACCES;
2145}
2146
2147static int nfs_open_permission_mask(int openflags)
2148{
2149        int mask = 0;
2150
2151        if (openflags & __FMODE_EXEC) {
2152                /* ONLY check exec rights */
2153                mask = MAY_EXEC;
2154        } else {
2155                if ((openflags & O_ACCMODE) != O_WRONLY)
2156                        mask |= MAY_READ;
2157                if ((openflags & O_ACCMODE) != O_RDONLY)
2158                        mask |= MAY_WRITE;
2159        }
2160
2161        return mask;
2162}
2163
2164int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags)
2165{
2166        return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
2167}
2168EXPORT_SYMBOL_GPL(nfs_may_open);
2169
2170int nfs_permission(struct inode *inode, int mask)
2171{
2172        struct rpc_cred *cred;
2173        int res = 0;
2174
2175        if (mask & MAY_NOT_BLOCK)
2176                return -ECHILD;
2177
2178        nfs_inc_stats(inode, NFSIOS_VFSACCESS);
2179
2180        if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2181                goto out;
2182        /* Is this sys_access() ? */
2183        if (mask & (MAY_ACCESS | MAY_CHDIR))
2184                goto force_lookup;
2185
2186        switch (inode->i_mode & S_IFMT) {
2187                case S_IFLNK:
2188                        goto out;
2189                case S_IFREG:
2190                        /* NFSv4 has atomic_open... */
2191                        if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN)
2192                                        && (mask & MAY_OPEN)
2193                                        && !(mask & MAY_EXEC))
2194                                goto out;
2195                        break;
2196                case S_IFDIR:
2197                        /*
2198                         * Optimize away all write operations, since the server
2199                         * will check permissions when we perform the op.
2200                         */
2201                        if ((mask & MAY_WRITE) && !(mask & MAY_READ))
2202                                goto out;
2203        }
2204
2205force_lookup:
2206        if (!NFS_PROTO(inode)->access)
2207                goto out_notsup;
2208
2209        cred = rpc_lookup_cred();
2210        if (!IS_ERR(cred)) {
2211                res = nfs_do_access(inode, cred, mask);
2212                put_rpccred(cred);
2213        } else
2214                res = PTR_ERR(cred);
2215out:
2216        if (!res && (mask & MAY_EXEC) && !execute_ok(inode))
2217                res = -EACCES;
2218
2219        dfprintk(VFS, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n",
2220                inode->i_sb->s_id, inode->i_ino, mask, res);
2221        return res;
2222out_notsup:
2223        res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
2224        if (res == 0)
2225                res = generic_permission(inode, mask);
2226        goto out;
2227}
2228EXPORT_SYMBOL_GPL(nfs_permission);
2229
2230/*
2231 * Local variables:
2232 *  version-control: t
2233 *  kept-new-versions: 5
2234 * End:
2235 */
2236
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.