linux/fs/nfsd/vfs.c
<<
>>
Prefs
   1/*
   2 * File operations used by nfsd. Some of these have been ripped from
   3 * other parts of the kernel because they weren't exported, others
   4 * are partial duplicates with added or changed functionality.
   5 *
   6 * Note that several functions dget() the dentry upon which they want
   7 * to act, most notably those that create directory entries. Response
   8 * dentry's are dput()'d if necessary in the release callback.
   9 * So if you notice code paths that apparently fail to dput() the
  10 * dentry, don't worry--they have been taken care of.
  11 *
  12 * Copyright (C) 1995-1999 Olaf Kirch <okir@monad.swb.de>
  13 * Zerocpy NFS support (C) 2002 Hirokazu Takahashi <taka@valinux.co.jp>
  14 */
  15
  16#include <linux/fs.h>
  17#include <linux/file.h>
  18#include <linux/splice.h>
  19#include <linux/fcntl.h>
  20#include <linux/namei.h>
  21#include <linux/delay.h>
  22#include <linux/fsnotify.h>
  23#include <linux/posix_acl_xattr.h>
  24#include <linux/xattr.h>
  25#include <linux/jhash.h>
  26#include <linux/ima.h>
  27#include <linux/slab.h>
  28#include <asm/uaccess.h>
  29#include <linux/exportfs.h>
  30#include <linux/writeback.h>
  31#include <linux/security.h>
  32
  33#ifdef CONFIG_NFSD_V3
  34#include "xdr3.h"
  35#endif /* CONFIG_NFSD_V3 */
  36
  37#ifdef CONFIG_NFSD_V4
  38#include "acl.h"
  39#include "idmap.h"
  40#endif /* CONFIG_NFSD_V4 */
  41
  42#include "nfsd.h"
  43#include "vfs.h"
  44
  45#define NFSDDBG_FACILITY                NFSDDBG_FILEOP
  46
  47
  48/*
  49 * This is a cache of readahead params that help us choose the proper
  50 * readahead strategy. Initially, we set all readahead parameters to 0
  51 * and let the VFS handle things.
  52 * If you increase the number of cached files very much, you'll need to
  53 * add a hash table here.
  54 */
  55struct raparms {
  56        struct raparms          *p_next;
  57        unsigned int            p_count;
  58        ino_t                   p_ino;
  59        dev_t                   p_dev;
  60        int                     p_set;
  61        struct file_ra_state    p_ra;
  62        unsigned int            p_hindex;
  63};
  64
  65struct raparm_hbucket {
  66        struct raparms          *pb_head;
  67        spinlock_t              pb_lock;
  68} ____cacheline_aligned_in_smp;
  69
  70#define RAPARM_HASH_BITS        4
  71#define RAPARM_HASH_SIZE        (1<<RAPARM_HASH_BITS)
  72#define RAPARM_HASH_MASK        (RAPARM_HASH_SIZE-1)
  73static struct raparm_hbucket    raparm_hash[RAPARM_HASH_SIZE];
  74
  75/* 
  76 * Called from nfsd_lookup and encode_dirent. Check if we have crossed 
  77 * a mount point.
  78 * Returns -EAGAIN or -ETIMEDOUT leaving *dpp and *expp unchanged,
  79 *  or nfs_ok having possibly changed *dpp and *expp
  80 */
  81int
  82nfsd_cross_mnt(struct svc_rqst *rqstp, struct dentry **dpp, 
  83                        struct svc_export **expp)
  84{
  85        struct svc_export *exp = *expp, *exp2 = NULL;
  86        struct dentry *dentry = *dpp;
  87        struct path path = {.mnt = mntget(exp->ex_path.mnt),
  88                            .dentry = dget(dentry)};
  89        int err = 0;
  90
  91        err = follow_down(&path);
  92        if (err < 0)
  93                goto out;
  94
  95        exp2 = rqst_exp_get_by_name(rqstp, &path);
  96        if (IS_ERR(exp2)) {
  97                err = PTR_ERR(exp2);
  98                /*
  99                 * We normally allow NFS clients to continue
 100                 * "underneath" a mountpoint that is not exported.
 101                 * The exception is V4ROOT, where no traversal is ever
 102                 * allowed without an explicit export of the new
 103                 * directory.
 104                 */
 105                if (err == -ENOENT && !(exp->ex_flags & NFSEXP_V4ROOT))
 106                        err = 0;
 107                path_put(&path);
 108                goto out;
 109        }
 110        if (nfsd_v4client(rqstp) ||
 111                (exp->ex_flags & NFSEXP_CROSSMOUNT) || EX_NOHIDE(exp2)) {
 112                /* successfully crossed mount point */
 113                /*
 114                 * This is subtle: path.dentry is *not* on path.mnt
 115                 * at this point.  The only reason we are safe is that
 116                 * original mnt is pinned down by exp, so we should
 117                 * put path *before* putting exp
 118                 */
 119                *dpp = path.dentry;
 120                path.dentry = dentry;
 121                *expp = exp2;
 122                exp2 = exp;
 123        }
 124        path_put(&path);
 125        exp_put(exp2);
 126out:
 127        return err;
 128}
 129
 130static void follow_to_parent(struct path *path)
 131{
 132        struct dentry *dp;
 133
 134        while (path->dentry == path->mnt->mnt_root && follow_up(path))
 135                ;
 136        dp = dget_parent(path->dentry);
 137        dput(path->dentry);
 138        path->dentry = dp;
 139}
 140
 141static int nfsd_lookup_parent(struct svc_rqst *rqstp, struct dentry *dparent, struct svc_export **exp, struct dentry **dentryp)
 142{
 143        struct svc_export *exp2;
 144        struct path path = {.mnt = mntget((*exp)->ex_path.mnt),
 145                            .dentry = dget(dparent)};
 146
 147        follow_to_parent(&path);
 148
 149        exp2 = rqst_exp_parent(rqstp, &path);
 150        if (PTR_ERR(exp2) == -ENOENT) {
 151                *dentryp = dget(dparent);
 152        } else if (IS_ERR(exp2)) {
 153                path_put(&path);
 154                return PTR_ERR(exp2);
 155        } else {
 156                *dentryp = dget(path.dentry);
 157                exp_put(*exp);
 158                *exp = exp2;
 159        }
 160        path_put(&path);
 161        return 0;
 162}
 163
 164/*
 165 * For nfsd purposes, we treat V4ROOT exports as though there was an
 166 * export at *every* directory.
 167 */
 168int nfsd_mountpoint(struct dentry *dentry, struct svc_export *exp)
 169{
 170        if (d_mountpoint(dentry))
 171                return 1;
 172        if (nfsd4_is_junction(dentry))
 173                return 1;
 174        if (!(exp->ex_flags & NFSEXP_V4ROOT))
 175                return 0;
 176        return dentry->d_inode != NULL;
 177}
 178
 179__be32
 180nfsd_lookup_dentry(struct svc_rqst *rqstp, struct svc_fh *fhp,
 181                   const char *name, unsigned int len,
 182                   struct svc_export **exp_ret, struct dentry **dentry_ret)
 183{
 184        struct svc_export       *exp;
 185        struct dentry           *dparent;
 186        struct dentry           *dentry;
 187        int                     host_err;
 188
 189        dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp), len,name);
 190
 191        dparent = fhp->fh_dentry;
 192        exp  = fhp->fh_export;
 193        exp_get(exp);
 194
 195        /* Lookup the name, but don't follow links */
 196        if (isdotent(name, len)) {
 197                if (len==1)
 198                        dentry = dget(dparent);
 199                else if (dparent != exp->ex_path.dentry)
 200                        dentry = dget_parent(dparent);
 201                else if (!EX_NOHIDE(exp) && !nfsd_v4client(rqstp))
 202                        dentry = dget(dparent); /* .. == . just like at / */
 203                else {
 204                        /* checking mountpoint crossing is very different when stepping up */
 205                        host_err = nfsd_lookup_parent(rqstp, dparent, &exp, &dentry);
 206                        if (host_err)
 207                                goto out_nfserr;
 208                }
 209        } else {
 210                fh_lock(fhp);
 211                dentry = lookup_one_len(name, dparent, len);
 212                host_err = PTR_ERR(dentry);
 213                if (IS_ERR(dentry))
 214                        goto out_nfserr;
 215                /*
 216                 * check if we have crossed a mount point ...
 217                 */
 218                if (nfsd_mountpoint(dentry, exp)) {
 219                        if ((host_err = nfsd_cross_mnt(rqstp, &dentry, &exp))) {
 220                                dput(dentry);
 221                                goto out_nfserr;
 222                        }
 223                }
 224        }
 225        *dentry_ret = dentry;
 226        *exp_ret = exp;
 227        return 0;
 228
 229out_nfserr:
 230        exp_put(exp);
 231        return nfserrno(host_err);
 232}
 233
 234/*
 235 * Look up one component of a pathname.
 236 * N.B. After this call _both_ fhp and resfh need an fh_put
 237 *
 238 * If the lookup would cross a mountpoint, and the mounted filesystem
 239 * is exported to the client with NFSEXP_NOHIDE, then the lookup is
 240 * accepted as it stands and the mounted directory is
 241 * returned. Otherwise the covered directory is returned.
 242 * NOTE: this mountpoint crossing is not supported properly by all
 243 *   clients and is explicitly disallowed for NFSv3
 244 *      NeilBrown <neilb@cse.unsw.edu.au>
 245 */
 246__be32
 247nfsd_lookup(struct svc_rqst *rqstp, struct svc_fh *fhp, const char *name,
 248                                unsigned int len, struct svc_fh *resfh)
 249{
 250        struct svc_export       *exp;
 251        struct dentry           *dentry;
 252        __be32 err;
 253
 254        err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC);
 255        if (err)
 256                return err;
 257        err = nfsd_lookup_dentry(rqstp, fhp, name, len, &exp, &dentry);
 258        if (err)
 259                return err;
 260        err = check_nfsd_access(exp, rqstp);
 261        if (err)
 262                goto out;
 263        /*
 264         * Note: we compose the file handle now, but as the
 265         * dentry may be negative, it may need to be updated.
 266         */
 267        err = fh_compose(resfh, exp, dentry, fhp);
 268        if (!err && !dentry->d_inode)
 269                err = nfserr_noent;
 270out:
 271        dput(dentry);
 272        exp_put(exp);
 273        return err;
 274}
 275
 276static int nfsd_break_lease(struct inode *inode)
 277{
 278        if (!S_ISREG(inode->i_mode))
 279                return 0;
 280        return break_lease(inode, O_WRONLY | O_NONBLOCK);
 281}
 282
 283/*
 284 * Commit metadata changes to stable storage.
 285 */
 286static int
 287commit_metadata(struct svc_fh *fhp)
 288{
 289        struct inode *inode = fhp->fh_dentry->d_inode;
 290        const struct export_operations *export_ops = inode->i_sb->s_export_op;
 291
 292        if (!EX_ISSYNC(fhp->fh_export))
 293                return 0;
 294
 295        if (export_ops->commit_metadata)
 296                return export_ops->commit_metadata(inode);
 297        return sync_inode_metadata(inode, 1);
 298}
 299
 300/*
 301 * Set various file attributes.
 302 * N.B. After this call fhp needs an fh_put
 303 */
 304__be32
 305nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
 306             int check_guard, time_t guardtime)
 307{
 308        struct dentry   *dentry;
 309        struct inode    *inode;
 310        int             accmode = NFSD_MAY_SATTR;
 311        umode_t         ftype = 0;
 312        __be32          err;
 313        int             host_err;
 314        int             size_change = 0;
 315
 316        if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE))
 317                accmode |= NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE;
 318        if (iap->ia_valid & ATTR_SIZE)
 319                ftype = S_IFREG;
 320
 321        /* Get inode */
 322        err = fh_verify(rqstp, fhp, ftype, accmode);
 323        if (err)
 324                goto out;
 325
 326        dentry = fhp->fh_dentry;
 327        inode = dentry->d_inode;
 328
 329        /* Ignore any mode updates on symlinks */
 330        if (S_ISLNK(inode->i_mode))
 331                iap->ia_valid &= ~ATTR_MODE;
 332
 333        if (!iap->ia_valid)
 334                goto out;
 335
 336        /*
 337         * NFSv2 does not differentiate between "set-[ac]time-to-now"
 338         * which only requires access, and "set-[ac]time-to-X" which
 339         * requires ownership.
 340         * So if it looks like it might be "set both to the same time which
 341         * is close to now", and if inode_change_ok fails, then we
 342         * convert to "set to now" instead of "set to explicit time"
 343         *
 344         * We only call inode_change_ok as the last test as technically
 345         * it is not an interface that we should be using.  It is only
 346         * valid if the filesystem does not define it's own i_op->setattr.
 347         */
 348#define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
 349#define MAX_TOUCH_TIME_ERROR (30*60)
 350        if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET &&
 351            iap->ia_mtime.tv_sec == iap->ia_atime.tv_sec) {
 352                /*
 353                 * Looks probable.
 354                 *
 355                 * Now just make sure time is in the right ballpark.
 356                 * Solaris, at least, doesn't seem to care what the time
 357                 * request is.  We require it be within 30 minutes of now.
 358                 */
 359                time_t delta = iap->ia_atime.tv_sec - get_seconds();
 360                if (delta < 0)
 361                        delta = -delta;
 362                if (delta < MAX_TOUCH_TIME_ERROR &&
 363                    inode_change_ok(inode, iap) != 0) {
 364                        /*
 365                         * Turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME.
 366                         * This will cause notify_change to set these times
 367                         * to "now"
 368                         */
 369                        iap->ia_valid &= ~BOTH_TIME_SET;
 370                }
 371        }
 372            
 373        /*
 374         * The size case is special.
 375         * It changes the file as well as the attributes.
 376         */
 377        if (iap->ia_valid & ATTR_SIZE) {
 378                if (iap->ia_size < inode->i_size) {
 379                        err = nfsd_permission(rqstp, fhp->fh_export, dentry,
 380                                        NFSD_MAY_TRUNC|NFSD_MAY_OWNER_OVERRIDE);
 381                        if (err)
 382                                goto out;
 383                }
 384
 385                host_err = get_write_access(inode);
 386                if (host_err)
 387                        goto out_nfserr;
 388
 389                size_change = 1;
 390                host_err = locks_verify_truncate(inode, NULL, iap->ia_size);
 391                if (host_err) {
 392                        put_write_access(inode);
 393                        goto out_nfserr;
 394                }
 395        }
 396
 397        /* sanitize the mode change */
 398        if (iap->ia_valid & ATTR_MODE) {
 399                iap->ia_mode &= S_IALLUGO;
 400                iap->ia_mode |= (inode->i_mode & ~S_IALLUGO);
 401        }
 402
 403        /* Revoke setuid/setgid on chown */
 404        if (!S_ISDIR(inode->i_mode) &&
 405            (((iap->ia_valid & ATTR_UID) && !uid_eq(iap->ia_uid, inode->i_uid)) ||
 406             ((iap->ia_valid & ATTR_GID) && !gid_eq(iap->ia_gid, inode->i_gid)))) {
 407                iap->ia_valid |= ATTR_KILL_PRIV;
 408                if (iap->ia_valid & ATTR_MODE) {
 409                        /* we're setting mode too, just clear the s*id bits */
 410                        iap->ia_mode &= ~S_ISUID;
 411                        if (iap->ia_mode & S_IXGRP)
 412                                iap->ia_mode &= ~S_ISGID;
 413                } else {
 414                        /* set ATTR_KILL_* bits and let VFS handle it */
 415                        iap->ia_valid |= (ATTR_KILL_SUID | ATTR_KILL_SGID);
 416                }
 417        }
 418
 419        /* Change the attributes. */
 420
 421        iap->ia_valid |= ATTR_CTIME;
 422
 423        err = nfserr_notsync;
 424        if (!check_guard || guardtime == inode->i_ctime.tv_sec) {
 425                host_err = nfsd_break_lease(inode);
 426                if (host_err)
 427                        goto out_nfserr;
 428                fh_lock(fhp);
 429
 430                host_err = notify_change(dentry, iap);
 431                err = nfserrno(host_err);
 432                fh_unlock(fhp);
 433        }
 434        if (size_change)
 435                put_write_access(inode);
 436        if (!err)
 437                commit_metadata(fhp);
 438out:
 439        return err;
 440
 441out_nfserr:
 442        err = nfserrno(host_err);
 443        goto out;
 444}
 445
 446#if defined(CONFIG_NFSD_V2_ACL) || \
 447    defined(CONFIG_NFSD_V3_ACL) || \
 448    defined(CONFIG_NFSD_V4)
 449static ssize_t nfsd_getxattr(struct dentry *dentry, char *key, void **buf)
 450{
 451        ssize_t buflen;
 452        ssize_t ret;
 453
 454        buflen = vfs_getxattr(dentry, key, NULL, 0);
 455        if (buflen <= 0)
 456                return buflen;
 457
 458        *buf = kmalloc(buflen, GFP_KERNEL);
 459        if (!*buf)
 460                return -ENOMEM;
 461
 462        ret = vfs_getxattr(dentry, key, *buf, buflen);
 463        if (ret < 0)
 464                kfree(*buf);
 465        return ret;
 466}
 467#endif
 468
 469#if defined(CONFIG_NFSD_V4)
 470static int
 471set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key)
 472{
 473        int len;
 474        size_t buflen;
 475        char *buf = NULL;
 476        int error = 0;
 477
 478        buflen = posix_acl_xattr_size(pacl->a_count);
 479        buf = kmalloc(buflen, GFP_KERNEL);
 480        error = -ENOMEM;
 481        if (buf == NULL)
 482                goto out;
 483
 484        len = posix_acl_to_xattr(&init_user_ns, pacl, buf, buflen);
 485        if (len < 0) {
 486                error = len;
 487                goto out;
 488        }
 489
 490        error = vfs_setxattr(dentry, key, buf, len, 0);
 491out:
 492        kfree(buf);
 493        return error;
 494}
 495
 496__be32
 497nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp,
 498    struct nfs4_acl *acl)
 499{
 500        __be32 error;
 501        int host_error;
 502        struct dentry *dentry;
 503        struct inode *inode;
 504        struct posix_acl *pacl = NULL, *dpacl = NULL;
 505        unsigned int flags = 0;
 506
 507        /* Get inode */
 508        error = fh_verify(rqstp, fhp, 0, NFSD_MAY_SATTR);
 509        if (error)
 510                return error;
 511
 512        dentry = fhp->fh_dentry;
 513        inode = dentry->d_inode;
 514        if (S_ISDIR(inode->i_mode))
 515                flags = NFS4_ACL_DIR;
 516
 517        host_error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags);
 518        if (host_error == -EINVAL) {
 519                return nfserr_attrnotsupp;
 520        } else if (host_error < 0)
 521                goto out_nfserr;
 522
 523        host_error = set_nfsv4_acl_one(dentry, pacl, POSIX_ACL_XATTR_ACCESS);
 524        if (host_error < 0)
 525                goto out_release;
 526
 527        if (S_ISDIR(inode->i_mode))
 528                host_error = set_nfsv4_acl_one(dentry, dpacl, POSIX_ACL_XATTR_DEFAULT);
 529
 530out_release:
 531        posix_acl_release(pacl);
 532        posix_acl_release(dpacl);
 533out_nfserr:
 534        if (host_error == -EOPNOTSUPP)
 535                return nfserr_attrnotsupp;
 536        else
 537                return nfserrno(host_error);
 538}
 539
 540static struct posix_acl *
 541_get_posix_acl(struct dentry *dentry, char *key)
 542{
 543        void *buf = NULL;
 544        struct posix_acl *pacl = NULL;
 545        int buflen;
 546
 547        buflen = nfsd_getxattr(dentry, key, &buf);
 548        if (!buflen)
 549                buflen = -ENODATA;
 550        if (buflen <= 0)
 551                return ERR_PTR(buflen);
 552
 553        pacl = posix_acl_from_xattr(&init_user_ns, buf, buflen);
 554        kfree(buf);
 555        return pacl;
 556}
 557
 558int
 559nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl)
 560{
 561        struct inode *inode = dentry->d_inode;
 562        int error = 0;
 563        struct posix_acl *pacl = NULL, *dpacl = NULL;
 564        unsigned int flags = 0;
 565
 566        pacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_ACCESS);
 567        if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA)
 568                pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL);
 569        if (IS_ERR(pacl)) {
 570                error = PTR_ERR(pacl);
 571                pacl = NULL;
 572                goto out;
 573        }
 574
 575        if (S_ISDIR(inode->i_mode)) {
 576                dpacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_DEFAULT);
 577                if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA)
 578                        dpacl = NULL;
 579                else if (IS_ERR(dpacl)) {
 580                        error = PTR_ERR(dpacl);
 581                        dpacl = NULL;
 582                        goto out;
 583                }
 584                flags = NFS4_ACL_DIR;
 585        }
 586
 587        *acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags);
 588        if (IS_ERR(*acl)) {
 589                error = PTR_ERR(*acl);
 590                *acl = NULL;
 591        }
 592 out:
 593        posix_acl_release(pacl);
 594        posix_acl_release(dpacl);
 595        return error;
 596}
 597
 598/*
 599 * NFS junction information is stored in an extended attribute.
 600 */
 601#define NFSD_JUNCTION_XATTR_NAME        XATTR_TRUSTED_PREFIX "junction.nfs"
 602
 603/**
 604 * nfsd4_is_junction - Test if an object could be an NFS junction
 605 *
 606 * @dentry: object to test
 607 *
 608 * Returns 1 if "dentry" appears to contain NFS junction information.
 609 * Otherwise 0 is returned.
 610 */
 611int nfsd4_is_junction(struct dentry *dentry)
 612{
 613        struct inode *inode = dentry->d_inode;
 614
 615        if (inode == NULL)
 616                return 0;
 617        if (inode->i_mode & S_IXUGO)
 618                return 0;
 619        if (!(inode->i_mode & S_ISVTX))
 620                return 0;
 621        if (vfs_getxattr(dentry, NFSD_JUNCTION_XATTR_NAME, NULL, 0) <= 0)
 622                return 0;
 623        return 1;
 624}
 625#ifdef CONFIG_NFSD_V4_SECURITY_LABEL
 626__be32 nfsd4_set_nfs4_label(struct svc_rqst *rqstp, struct svc_fh *fhp,
 627                struct xdr_netobj *label)
 628{
 629        __be32 error;
 630        int host_error;
 631        struct dentry *dentry;
 632
 633        error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, NFSD_MAY_SATTR);
 634        if (error)
 635                return error;
 636
 637        dentry = fhp->fh_dentry;
 638
 639        mutex_lock(&dentry->d_inode->i_mutex);
 640        host_error = security_inode_setsecctx(dentry, label->data, label->len);
 641        mutex_unlock(&dentry->d_inode->i_mutex);
 642        return nfserrno(host_error);
 643}
 644#else
 645__be32 nfsd4_set_nfs4_label(struct svc_rqst *rqstp, struct svc_fh *fhp,
 646                struct xdr_netobj *label)
 647{
 648        return nfserr_notsupp;
 649}
 650#endif
 651
 652#endif /* defined(CONFIG_NFSD_V4) */
 653
 654#ifdef CONFIG_NFSD_V3
 655/*
 656 * Check server access rights to a file system object
 657 */
 658struct accessmap {
 659        u32             access;
 660        int             how;
 661};
 662static struct accessmap nfs3_regaccess[] = {
 663    {   NFS3_ACCESS_READ,       NFSD_MAY_READ                   },
 664    {   NFS3_ACCESS_EXECUTE,    NFSD_MAY_EXEC                   },
 665    {   NFS3_ACCESS_MODIFY,     NFSD_MAY_WRITE|NFSD_MAY_TRUNC   },
 666    {   NFS3_ACCESS_EXTEND,     NFSD_MAY_WRITE                  },
 667
 668    {   0,                      0                               }
 669};
 670
 671static struct accessmap nfs3_diraccess[] = {
 672    {   NFS3_ACCESS_READ,       NFSD_MAY_READ                   },
 673    {   NFS3_ACCESS_LOOKUP,     NFSD_MAY_EXEC                   },
 674    {   NFS3_ACCESS_MODIFY,     NFSD_MAY_EXEC|NFSD_MAY_WRITE|NFSD_MAY_TRUNC},
 675    {   NFS3_ACCESS_EXTEND,     NFSD_MAY_EXEC|NFSD_MAY_WRITE    },
 676    {   NFS3_ACCESS_DELETE,     NFSD_MAY_REMOVE                 },
 677
 678    {   0,                      0                               }
 679};
 680
 681static struct accessmap nfs3_anyaccess[] = {
 682        /* Some clients - Solaris 2.6 at least, make an access call
 683         * to the server to check for access for things like /dev/null
 684         * (which really, the server doesn't care about).  So
 685         * We provide simple access checking for them, looking
 686         * mainly at mode bits, and we make sure to ignore read-only
 687         * filesystem checks
 688         */
 689    {   NFS3_ACCESS_READ,       NFSD_MAY_READ                   },
 690    {   NFS3_ACCESS_EXECUTE,    NFSD_MAY_EXEC                   },
 691    {   NFS3_ACCESS_MODIFY,     NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS    },
 692    {   NFS3_ACCESS_EXTEND,     NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS    },
 693
 694    {   0,                      0                               }
 695};
 696
 697__be32
 698nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
 699{
 700        struct accessmap        *map;
 701        struct svc_export       *export;
 702        struct dentry           *dentry;
 703        u32                     query, result = 0, sresult = 0;
 704        __be32                  error;
 705
 706        error = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP);
 707        if (error)
 708                goto out;
 709
 710        export = fhp->fh_export;
 711        dentry = fhp->fh_dentry;
 712
 713        if (S_ISREG(dentry->d_inode->i_mode))
 714                map = nfs3_regaccess;
 715        else if (S_ISDIR(dentry->d_inode->i_mode))
 716                map = nfs3_diraccess;
 717        else
 718                map = nfs3_anyaccess;
 719
 720
 721        query = *access;
 722        for  (; map->access; map++) {
 723                if (map->access & query) {
 724                        __be32 err2;
 725
 726                        sresult |= map->access;
 727
 728                        err2 = nfsd_permission(rqstp, export, dentry, map->how);
 729                        switch (err2) {
 730                        case nfs_ok:
 731                                result |= map->access;
 732                                break;
 733                                
 734                        /* the following error codes just mean the access was not allowed,
 735                         * rather than an error occurred */
 736                        case nfserr_rofs:
 737                        case nfserr_acces:
 738                        case nfserr_perm:
 739                                /* simply don't "or" in the access bit. */
 740                                break;
 741                        default:
 742                                error = err2;
 743                                goto out;
 744                        }
 745                }
 746        }
 747        *access = result;
 748        if (supported)
 749                *supported = sresult;
 750
 751 out:
 752        return error;
 753}
 754#endif /* CONFIG_NFSD_V3 */
 755
 756static int nfsd_open_break_lease(struct inode *inode, int access)
 757{
 758        unsigned int mode;
 759
 760        if (access & NFSD_MAY_NOT_BREAK_LEASE)
 761                return 0;
 762        mode = (access & NFSD_MAY_WRITE) ? O_WRONLY : O_RDONLY;
 763        return break_lease(inode, mode | O_NONBLOCK);
 764}
 765
 766/*
 767 * Open an existing file or directory.
 768 * The may_flags argument indicates the type of open (read/write/lock)
 769 * and additional flags.
 770 * N.B. After this call fhp needs an fh_put
 771 */
 772__be32
 773nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, umode_t type,
 774                        int may_flags, struct file **filp)
 775{
 776        struct path     path;
 777        struct inode    *inode;
 778        int             flags = O_RDONLY|O_LARGEFILE;
 779        __be32          err;
 780        int             host_err = 0;
 781
 782        validate_process_creds();
 783
 784        /*
 785         * If we get here, then the client has already done an "open",
 786         * and (hopefully) checked permission - so allow OWNER_OVERRIDE
 787         * in case a chmod has now revoked permission.
 788         *
 789         * Arguably we should also allow the owner override for
 790         * directories, but we never have and it doesn't seem to have
 791         * caused anyone a problem.  If we were to change this, note
 792         * also that our filldir callbacks would need a variant of
 793         * lookup_one_len that doesn't check permissions.
 794         */
 795        if (type == S_IFREG)
 796                may_flags |= NFSD_MAY_OWNER_OVERRIDE;
 797        err = fh_verify(rqstp, fhp, type, may_flags);
 798        if (err)
 799                goto out;
 800
 801        path.mnt = fhp->fh_export->ex_path.mnt;
 802        path.dentry = fhp->fh_dentry;
 803        inode = path.dentry->d_inode;
 804
 805        /* Disallow write access to files with the append-only bit set
 806         * or any access when mandatory locking enabled
 807         */
 808        err = nfserr_perm;
 809        if (IS_APPEND(inode) && (may_flags & NFSD_MAY_WRITE))
 810                goto out;
 811        /*
 812         * We must ignore files (but only files) which might have mandatory
 813         * locks on them because there is no way to know if the accesser has
 814         * the lock.
 815         */
 816        if (S_ISREG((inode)->i_mode) && mandatory_lock(inode))
 817                goto out;
 818
 819        if (!inode->i_fop)
 820                goto out;
 821
 822        host_err = nfsd_open_break_lease(inode, may_flags);
 823        if (host_err) /* NOMEM or WOULDBLOCK */
 824                goto out_nfserr;
 825
 826        if (may_flags & NFSD_MAY_WRITE) {
 827                if (may_flags & NFSD_MAY_READ)
 828                        flags = O_RDWR|O_LARGEFILE;
 829                else
 830                        flags = O_WRONLY|O_LARGEFILE;
 831        }
 832        *filp = dentry_open(&path, flags, current_cred());
 833        if (IS_ERR(*filp)) {
 834                host_err = PTR_ERR(*filp);
 835                *filp = NULL;
 836        } else {
 837                host_err = ima_file_check(*filp, may_flags);
 838
 839                if (may_flags & NFSD_MAY_64BIT_COOKIE)
 840                        (*filp)->f_mode |= FMODE_64BITHASH;
 841                else
 842                        (*filp)->f_mode |= FMODE_32BITHASH;
 843        }
 844
 845out_nfserr:
 846        err = nfserrno(host_err);
 847out:
 848        validate_process_creds();
 849        return err;
 850}
 851
 852/*
 853 * Close a file.
 854 */
 855void
 856nfsd_close(struct file *filp)
 857{
 858        fput(filp);
 859}
 860
 861/*
 862 * Obtain the readahead parameters for the file
 863 * specified by (dev, ino).
 864 */
 865
 866static inline struct raparms *
 867nfsd_get_raparms(dev_t dev, ino_t ino)
 868{
 869        struct raparms  *ra, **rap, **frap = NULL;
 870        int depth = 0;
 871        unsigned int hash;
 872        struct raparm_hbucket *rab;
 873
 874        hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK;
 875        rab = &raparm_hash[hash];
 876
 877        spin_lock(&rab->pb_lock);
 878        for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) {
 879                if (ra->p_ino == ino && ra->p_dev == dev)
 880                        goto found;
 881                depth++;
 882                if (ra->p_count == 0)
 883                        frap = rap;
 884        }
 885        depth = nfsdstats.ra_size;
 886        if (!frap) {    
 887                spin_unlock(&rab->pb_lock);
 888                return NULL;
 889        }
 890        rap = frap;
 891        ra = *frap;
 892        ra->p_dev = dev;
 893        ra->p_ino = ino;
 894        ra->p_set = 0;
 895        ra->p_hindex = hash;
 896found:
 897        if (rap != &rab->pb_head) {
 898                *rap = ra->p_next;
 899                ra->p_next   = rab->pb_head;
 900                rab->pb_head = ra;
 901        }
 902        ra->p_count++;
 903        nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
 904        spin_unlock(&rab->pb_lock);
 905        return ra;
 906}
 907
 908/*
 909 * Grab and keep cached pages associated with a file in the svc_rqst
 910 * so that they can be passed to the network sendmsg/sendpage routines
 911 * directly. They will be released after the sending has completed.
 912 */
 913static int
 914nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
 915                  struct splice_desc *sd)
 916{
 917        struct svc_rqst *rqstp = sd->u.data;
 918        struct page **pp = rqstp->rq_next_page;
 919        struct page *page = buf->page;
 920        size_t size;
 921
 922        size = sd->len;
 923
 924        if (rqstp->rq_res.page_len == 0) {
 925                get_page(page);
 926                put_page(*rqstp->rq_next_page);
 927                *(rqstp->rq_next_page++) = page;
 928                rqstp->rq_res.page_base = buf->offset;
 929                rqstp->rq_res.page_len = size;
 930        } else if (page != pp[-1]) {
 931                get_page(page);
 932                if (*rqstp->rq_next_page)
 933                        put_page(*rqstp->rq_next_page);
 934                *(rqstp->rq_next_page++) = page;
 935                rqstp->rq_res.page_len += size;
 936        } else
 937                rqstp->rq_res.page_len += size;
 938
 939        return size;
 940}
 941
 942static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe,
 943                                    struct splice_desc *sd)
 944{
 945        return __splice_from_pipe(pipe, sd, nfsd_splice_actor);
 946}
 947
 948static __be32
 949nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
 950              loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
 951{
 952        mm_segment_t    oldfs;
 953        __be32          err;
 954        int             host_err;
 955
 956        err = nfserr_perm;
 957
 958        if (file->f_op->splice_read && rqstp->rq_splice_ok) {
 959                struct splice_desc sd = {
 960                        .len            = 0,
 961                        .total_len      = *count,
 962                        .pos            = offset,
 963                        .u.data         = rqstp,
 964                };
 965
 966                rqstp->rq_next_page = rqstp->rq_respages + 1;
 967                host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor);
 968        } else {
 969                oldfs = get_fs();
 970                set_fs(KERNEL_DS);
 971                host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
 972                set_fs(oldfs);
 973        }
 974
 975        if (host_err >= 0) {
 976                nfsdstats.io_read += host_err;
 977                *count = host_err;
 978                err = 0;
 979                fsnotify_access(file);
 980        } else 
 981                err = nfserrno(host_err);
 982        return err;
 983}
 984
 985static void kill_suid(struct dentry *dentry)
 986{
 987        struct iattr    ia;
 988        ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
 989
 990        mutex_lock(&dentry->d_inode->i_mutex);
 991        notify_change(dentry, &ia);
 992        mutex_unlock(&dentry->d_inode->i_mutex);
 993}
 994
 995/*
 996 * Gathered writes: If another process is currently writing to the file,
 997 * there's a high chance this is another nfsd (triggered by a bulk write
 998 * from a client's biod). Rather than syncing the file with each write
 999 * request, we sleep for 10 msec.
1000 *
1001 * I don't know if this roughly approximates C. Juszak's idea of
1002 * gathered writes, but it's a nice and simple solution (IMHO), and it
1003 * seems to work:-)
1004 *
1005 * Note: we do this only in the NFSv2 case, since v3 and higher have a
1006 * better tool (separate unstable writes and commits) for solving this
1007 * problem.
1008 */
1009static int wait_for_concurrent_writes(struct file *file)
1010{
1011        struct inode *inode = file_inode(file);
1012        static ino_t last_ino;
1013        static dev_t last_dev;
1014        int err = 0;
1015
1016        if (atomic_read(&inode->i_writecount) > 1
1017            || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
1018                dprintk("nfsd: write defer %d\n", task_pid_nr(current));
1019                msleep(10);
1020                dprintk("nfsd: write resume %d\n", task_pid_nr(current));
1021        }
1022
1023        if (inode->i_state & I_DIRTY) {
1024                dprintk("nfsd: write sync %d\n", task_pid_nr(current));
1025                err = vfs_fsync(file, 0);
1026        }
1027        last_ino = inode->i_ino;
1028        last_dev = inode->i_sb->s_dev;
1029        return err;
1030}
1031
1032static __be32
1033nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1034                                loff_t offset, struct kvec *vec, int vlen,
1035                                unsigned long *cnt, int *stablep)
1036{
1037        struct svc_export       *exp;
1038        struct dentry           *dentry;
1039        struct inode            *inode;
1040        mm_segment_t            oldfs;
1041        __be32                  err = 0;
1042        int                     host_err;
1043        int                     stable = *stablep;
1044        int                     use_wgather;
1045        loff_t                  pos = offset;
1046
1047        dentry = file->f_path.dentry;
1048        inode = dentry->d_inode;
1049        exp   = fhp->fh_export;
1050
1051        use_wgather = (rqstp->rq_vers == 2) && EX_WGATHER(exp);
1052
1053        if (!EX_ISSYNC(exp))
1054                stable = 0;
1055
1056        /* Write the data. */
1057        oldfs = get_fs(); set_fs(KERNEL_DS);
1058        host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &pos);
1059        set_fs(oldfs);
1060        if (host_err < 0)
1061                goto out_nfserr;
1062        *cnt = host_err;
1063        nfsdstats.io_write += host_err;
1064        fsnotify_modify(file);
1065
1066        /* clear setuid/setgid flag after write */
1067        if (inode->i_mode & (S_ISUID | S_ISGID))
1068                kill_suid(dentry);
1069
1070        if (stable) {
1071                if (use_wgather)
1072                        host_err = wait_for_concurrent_writes(file);
1073                else
1074                        host_err = vfs_fsync_range(file, offset, offset+*cnt, 0);
1075        }
1076
1077out_nfserr:
1078        dprintk("nfsd: write complete host_err=%d\n", host_err);
1079        if (host_err >= 0)
1080                err = 0;
1081        else
1082                err = nfserrno(host_err);
1083        return err;
1084}
1085
1086/*
1087 * Read data from a file. count must contain the requested read count
1088 * on entry. On return, *count contains the number of bytes actually read.
1089 * N.B. After this call fhp needs an fh_put
1090 */
1091__be32 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp,
1092        loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
1093{
1094        struct file *file;
1095        struct inode *inode;
1096        struct raparms  *ra;
1097        __be32 err;
1098
1099        err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
1100        if (err)
1101                return err;
1102
1103        inode = file_inode(file);
1104
1105        /* Get readahead parameters */
1106        ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
1107
1108        if (ra && ra->p_set)
1109                file->f_ra = ra->p_ra;
1110
1111        err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1112
1113        /* Write back readahead params */
1114        if (ra) {
1115                struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex];
1116                spin_lock(&rab->pb_lock);
1117                ra->p_ra = file->f_ra;
1118                ra->p_set = 1;
1119                ra->p_count--;
1120                spin_unlock(&rab->pb_lock);
1121        }
1122
1123        nfsd_close(file);
1124        return err;
1125}
1126
1127/* As above, but use the provided file descriptor. */
1128__be32
1129nfsd_read_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1130                loff_t offset, struct kvec *vec, int vlen,
1131                unsigned long *count)
1132{
1133        __be32          err;
1134
1135        if (file) {
1136                err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1137                                NFSD_MAY_READ|NFSD_MAY_OWNER_OVERRIDE);
1138                if (err)
1139                        goto out;
1140                err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1141        } else /* Note file may still be NULL in NFSv4 special stateid case: */
1142                err = nfsd_read(rqstp, fhp, offset, vec, vlen, count);
1143out:
1144        return err;
1145}
1146
1147/*
1148 * Write data to a file.
1149 * The stable flag requests synchronous writes.
1150 * N.B. After this call fhp needs an fh_put
1151 */
1152__be32
1153nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1154                loff_t offset, struct kvec *vec, int vlen, unsigned long *cnt,
1155                int *stablep)
1156{
1157        __be32                  err = 0;
1158
1159        if (file) {
1160                err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1161                                NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE);
1162                if (err)
1163                        goto out;
1164                err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1165                                stablep);
1166        } else {
1167                err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file);
1168                if (err)
1169                        goto out;
1170
1171                if (cnt)
1172                        err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1173                                             cnt, stablep);
1174                nfsd_close(file);
1175        }
1176out:
1177        return err;
1178}
1179
1180#ifdef CONFIG_NFSD_V3
1181/*
1182 * Commit all pending writes to stable storage.
1183 *
1184 * Note: we only guarantee that data that lies within the range specified
1185 * by the 'offset' and 'count' parameters will be synced.
1186 *
1187 * Unfortunately we cannot lock the file to make sure we return full WCC
1188 * data to the client, as locking happens lower down in the filesystem.
1189 */
1190__be32
1191nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1192               loff_t offset, unsigned long count)
1193{
1194        struct file     *file;
1195        loff_t          end = LLONG_MAX;
1196        __be32          err = nfserr_inval;
1197
1198        if (offset < 0)
1199                goto out;
1200        if (count != 0) {
1201                end = offset + (loff_t)count - 1;
1202                if (end < offset)
1203                        goto out;
1204        }
1205
1206        err = nfsd_open(rqstp, fhp, S_IFREG,
1207                        NFSD_MAY_WRITE|NFSD_MAY_NOT_BREAK_LEASE, &file);
1208        if (err)
1209                goto out;
1210        if (EX_ISSYNC(fhp->fh_export)) {
1211                int err2 = vfs_fsync_range(file, offset, end, 0);
1212
1213                if (err2 != -EINVAL)
1214                        err = nfserrno(err2);
1215                else
1216                        err = nfserr_notsupp;
1217        }
1218
1219        nfsd_close(file);
1220out:
1221        return err;
1222}
1223#endif /* CONFIG_NFSD_V3 */
1224
1225static __be32
1226nfsd_create_setattr(struct svc_rqst *rqstp, struct svc_fh *resfhp,
1227                        struct iattr *iap)
1228{
1229        /*
1230         * Mode has already been set earlier in create:
1231         */
1232        iap->ia_valid &= ~ATTR_MODE;
1233        /*
1234         * Setting uid/gid works only for root.  Irix appears to
1235         * send along the gid on create when it tries to implement
1236         * setgid directories via NFS:
1237         */
1238        if (!uid_eq(current_fsuid(), GLOBAL_ROOT_UID))
1239                iap->ia_valid &= ~(ATTR_UID|ATTR_GID);
1240        if (iap->ia_valid)
1241                return nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1242        return 0;
1243}
1244
1245/* HPUX client sometimes creates a file in mode 000, and sets size to 0.
1246 * setting size to 0 may fail for some specific file systems by the permission
1247 * checking which requires WRITE permission but the mode is 000.
1248 * we ignore the resizing(to 0) on the just new created file, since the size is
1249 * 0 after file created.
1250 *
1251 * call this only after vfs_create() is called.
1252 * */
1253static void
1254nfsd_check_ignore_resizing(struct iattr *iap)
1255{
1256        if ((iap->ia_valid & ATTR_SIZE) && (iap->ia_size == 0))
1257                iap->ia_valid &= ~ATTR_SIZE;
1258}
1259
1260/*
1261 * Create a file (regular, directory, device, fifo); UNIX sockets 
1262 * not yet implemented.
1263 * If the response fh has been verified, the parent directory should
1264 * already be locked. Note that the parent directory is left locked.
1265 *
1266 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1267 */
1268__be32
1269nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1270                char *fname, int flen, struct iattr *iap,
1271                int type, dev_t rdev, struct svc_fh *resfhp)
1272{
1273        struct dentry   *dentry, *dchild = NULL;
1274        struct inode    *dirp;
1275        __be32          err;
1276        __be32          err2;
1277        int             host_err;
1278
1279        err = nfserr_perm;
1280        if (!flen)
1281                goto out;
1282        err = nfserr_exist;
1283        if (isdotent(fname, flen))
1284                goto out;
1285
1286        err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1287        if (err)
1288                goto out;
1289
1290        dentry = fhp->fh_dentry;
1291        dirp = dentry->d_inode;
1292
1293        err = nfserr_notdir;
1294        if (!dirp->i_op->lookup)
1295                goto out;
1296        /*
1297         * Check whether the response file handle has been verified yet.
1298         * If it has, the parent directory should already be locked.
1299         */
1300        if (!resfhp->fh_dentry) {
1301                host_err = fh_want_write(fhp);
1302                if (host_err)
1303                        goto out_nfserr;
1304
1305                /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1306                fh_lock_nested(fhp, I_MUTEX_PARENT);
1307                dchild = lookup_one_len(fname, dentry, flen);
1308                host_err = PTR_ERR(dchild);
1309                if (IS_ERR(dchild))
1310                        goto out_nfserr;
1311                err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1312                if (err)
1313                        goto out;
1314        } else {
1315                /* called from nfsd_proc_create */
1316                dchild = dget(resfhp->fh_dentry);
1317                if (!fhp->fh_locked) {
1318                        /* not actually possible */
1319                        printk(KERN_ERR
1320                                "nfsd_create: parent %s/%s not locked!\n",
1321                                dentry->d_parent->d_name.name,
1322                                dentry->d_name.name);
1323                        err = nfserr_io;
1324                        goto out;
1325                }
1326        }
1327        /*
1328         * Make sure the child dentry is still negative ...
1329         */
1330        err = nfserr_exist;
1331        if (dchild->d_inode) {
1332                dprintk("nfsd_create: dentry %s/%s not negative!\n",
1333                        dentry->d_name.name, dchild->d_name.name);
1334                goto out; 
1335        }
1336
1337        if (!(iap->ia_valid & ATTR_MODE))
1338                iap->ia_mode = 0;
1339        iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1340
1341        err = nfserr_inval;
1342        if (!S_ISREG(type) && !S_ISDIR(type) && !special_file(type)) {
1343                printk(KERN_WARNING "nfsd: bad file type %o in nfsd_create\n",
1344                       type);
1345                goto out;
1346        }
1347
1348        /*
1349         * Get the dir op function pointer.
1350         */
1351        err = 0;
1352        host_err = 0;
1353        switch (type) {
1354        case S_IFREG:
1355                host_err = vfs_create(dirp, dchild, iap->ia_mode, true);
1356                if (!host_err)
1357                        nfsd_check_ignore_resizing(iap);
1358                break;
1359        case S_IFDIR:
1360                host_err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1361                break;
1362        case S_IFCHR:
1363        case S_IFBLK:
1364        case S_IFIFO:
1365        case S_IFSOCK:
1366                host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1367                break;
1368        }
1369        if (host_err < 0)
1370                goto out_nfserr;
1371
1372        err = nfsd_create_setattr(rqstp, resfhp, iap);
1373
1374        /*
1375         * nfsd_setattr already committed the child.  Transactional filesystems
1376         * had a chance to commit changes for both parent and child
1377         * simultaneously making the following commit_metadata a noop.
1378         */
1379        err2 = nfserrno(commit_metadata(fhp));
1380        if (err2)
1381                err = err2;
1382        /*
1383         * Update the file handle to get the new inode info.
1384         */
1385        if (!err)
1386                err = fh_update(resfhp);
1387out:
1388        if (dchild && !IS_ERR(dchild))
1389                dput(dchild);
1390        return err;
1391
1392out_nfserr:
1393        err = nfserrno(host_err);
1394        goto out;
1395}
1396
1397#ifdef CONFIG_NFSD_V3
1398
1399static inline int nfsd_create_is_exclusive(int createmode)
1400{
1401        return createmode == NFS3_CREATE_EXCLUSIVE
1402               || createmode == NFS4_CREATE_EXCLUSIVE4_1;
1403}
1404
1405/*
1406 * NFSv3 and NFSv4 version of nfsd_create
1407 */
1408__be32
1409do_nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1410                char *fname, int flen, struct iattr *iap,
1411                struct svc_fh *resfhp, int createmode, u32 *verifier,
1412                bool *truncp, bool *created)
1413{
1414        struct dentry   *dentry, *dchild = NULL;
1415        struct inode    *dirp;
1416        __be32          err;
1417        int             host_err;
1418        __u32           v_mtime=0, v_atime=0;
1419
1420        err = nfserr_perm;
1421        if (!flen)
1422                goto out;
1423        err = nfserr_exist;
1424        if (isdotent(fname, flen))
1425                goto out;
1426        if (!(iap->ia_valid & ATTR_MODE))
1427                iap->ia_mode = 0;
1428        err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC);
1429        if (err)
1430                goto out;
1431
1432        dentry = fhp->fh_dentry;
1433        dirp = dentry->d_inode;
1434
1435        /* Get all the sanity checks out of the way before
1436         * we lock the parent. */
1437        err = nfserr_notdir;
1438        if (!dirp->i_op->lookup)
1439                goto out;
1440
1441        host_err = fh_want_write(fhp);
1442        if (host_err)
1443                goto out_nfserr;
1444
1445        fh_lock_nested(fhp, I_MUTEX_PARENT);
1446
1447        /*
1448         * Compose the response file handle.
1449         */
1450        dchild = lookup_one_len(fname, dentry, flen);
1451        host_err = PTR_ERR(dchild);
1452        if (IS_ERR(dchild))
1453                goto out_nfserr;
1454
1455        /* If file doesn't exist, check for permissions to create one */
1456        if (!dchild->d_inode) {
1457                err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1458                if (err)
1459                        goto out;
1460        }
1461
1462        err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1463        if (err)
1464                goto out;
1465
1466        if (nfsd_create_is_exclusive(createmode)) {
1467                /* solaris7 gets confused (bugid 4218508) if these have
1468                 * the high bit set, so just clear the high bits. If this is
1469                 * ever changed to use different attrs for storing the
1470                 * verifier, then do_open_lookup() will also need to be fixed
1471                 * accordingly.
1472                 */
1473                v_mtime = verifier[0]&0x7fffffff;
1474                v_atime = verifier[1]&0x7fffffff;
1475        }
1476        
1477        if (dchild->d_inode) {
1478                err = 0;
1479
1480                switch (createmode) {
1481                case NFS3_CREATE_UNCHECKED:
1482                        if (! S_ISREG(dchild->d_inode->i_mode))
1483                                goto out;
1484                        else if (truncp) {
1485                                /* in nfsv4, we need to treat this case a little
1486                                 * differently.  we don't want to truncate the
1487                                 * file now; this would be wrong if the OPEN
1488                                 * fails for some other reason.  furthermore,
1489                                 * if the size is nonzero, we should ignore it
1490                                 * according to spec!
1491                                 */
1492                                *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1493                        }
1494                        else {
1495                                iap->ia_valid &= ATTR_SIZE;
1496                                goto set_attr;
1497                        }
1498                        break;
1499                case NFS3_CREATE_EXCLUSIVE:
1500                        if (   dchild->d_inode->i_mtime.tv_sec == v_mtime
1501                            && dchild->d_inode->i_atime.tv_sec == v_atime
1502                            && dchild->d_inode->i_size  == 0 ) {
1503                                if (created)
1504                                        *created = 1;
1505                                break;
1506                        }
1507                case NFS4_CREATE_EXCLUSIVE4_1:
1508                        if (   dchild->d_inode->i_mtime.tv_sec == v_mtime
1509                            && dchild->d_inode->i_atime.tv_sec == v_atime
1510                            && dchild->d_inode->i_size  == 0 ) {
1511                                if (created)
1512                                        *created = 1;
1513                                goto set_attr;
1514                        }
1515                         /* fallthru */
1516                case NFS3_CREATE_GUARDED:
1517                        err = nfserr_exist;
1518                }
1519                fh_drop_write(fhp);
1520                goto out;
1521        }
1522
1523        host_err = vfs_create(dirp, dchild, iap->ia_mode, true);
1524        if (host_err < 0) {
1525                fh_drop_write(fhp);
1526                goto out_nfserr;
1527        }
1528        if (created)
1529                *created = 1;
1530
1531        nfsd_check_ignore_resizing(iap);
1532
1533        if (nfsd_create_is_exclusive(createmode)) {
1534                /* Cram the verifier into atime/mtime */
1535                iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1536                        | ATTR_MTIME_SET|ATTR_ATIME_SET;
1537                /* XXX someone who knows this better please fix it for nsec */ 
1538                iap->ia_mtime.tv_sec = v_mtime;
1539                iap->ia_atime.tv_sec = v_atime;
1540                iap->ia_mtime.tv_nsec = 0;
1541                iap->ia_atime.tv_nsec = 0;
1542        }
1543
1544 set_attr:
1545        err = nfsd_create_setattr(rqstp, resfhp, iap);
1546
1547        /*
1548         * nfsd_setattr already committed the child (and possibly also the parent).
1549         */
1550        if (!err)
1551                err = nfserrno(commit_metadata(fhp));
1552
1553        /*
1554         * Update the filehandle to get the new inode info.
1555         */
1556        if (!err)
1557                err = fh_update(resfhp);
1558
1559 out:
1560        fh_unlock(fhp);
1561        if (dchild && !IS_ERR(dchild))
1562                dput(dchild);
1563        fh_drop_write(fhp);
1564        return err;
1565 
1566 out_nfserr:
1567        err = nfserrno(host_err);
1568        goto out;
1569}
1570#endif /* CONFIG_NFSD_V3 */
1571
1572/*
1573 * Read a symlink. On entry, *lenp must contain the maximum path length that
1574 * fits into the buffer. On return, it contains the true length.
1575 * N.B. After this call fhp needs an fh_put
1576 */
1577__be32
1578nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1579{
1580        struct inode    *inode;
1581        mm_segment_t    oldfs;
1582        __be32          err;
1583        int             host_err;
1584        struct path path;
1585
1586        err = fh_verify(rqstp, fhp, S_IFLNK, NFSD_MAY_NOP);
1587        if (err)
1588                goto out;
1589
1590        path.mnt = fhp->fh_export->ex_path.mnt;
1591        path.dentry = fhp->fh_dentry;
1592        inode = path.dentry->d_inode;
1593
1594        err = nfserr_inval;
1595        if (!inode->i_op->readlink)
1596                goto out;
1597
1598        touch_atime(&path);
1599        /* N.B. Why does this call need a get_fs()??
1600         * Remove the set_fs and watch the fireworks:-) --okir
1601         */
1602
1603        oldfs = get_fs(); set_fs(KERNEL_DS);
1604        host_err = inode->i_op->readlink(path.dentry, (char __user *)buf, *lenp);
1605        set_fs(oldfs);
1606
1607        if (host_err < 0)
1608                goto out_nfserr;
1609        *lenp = host_err;
1610        err = 0;
1611out:
1612        return err;
1613
1614out_nfserr:
1615        err = nfserrno(host_err);
1616        goto out;
1617}
1618
1619/*
1620 * Create a symlink and look up its inode
1621 * N.B. After this call _both_ fhp and resfhp need an fh_put
1622 */
1623__be32
1624nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1625                                char *fname, int flen,
1626                                char *path,  int plen,
1627                                struct svc_fh *resfhp,
1628                                struct iattr *iap)
1629{
1630        struct dentry   *dentry, *dnew;
1631        __be32          err, cerr;
1632        int             host_err;
1633
1634        err = nfserr_noent;
1635        if (!flen || !plen)
1636                goto out;
1637        err = nfserr_exist;
1638        if (isdotent(fname, flen))
1639                goto out;
1640
1641        err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1642        if (err)
1643                goto out;
1644
1645        host_err = fh_want_write(fhp);
1646        if (host_err)
1647                goto out_nfserr;
1648
1649        fh_lock(fhp);
1650        dentry = fhp->fh_dentry;
1651        dnew = lookup_one_len(fname, dentry, flen);
1652        host_err = PTR_ERR(dnew);
1653        if (IS_ERR(dnew))
1654                goto out_nfserr;
1655
1656        if (unlikely(path[plen] != 0)) {
1657                char *path_alloced = kmalloc(plen+1, GFP_KERNEL);
1658                if (path_alloced == NULL)
1659                        host_err = -ENOMEM;
1660                else {
1661                        strncpy(path_alloced, path, plen);
1662                        path_alloced[plen] = 0;
1663                        host_err = vfs_symlink(dentry->d_inode, dnew, path_alloced);
1664                        kfree(path_alloced);
1665                }
1666        } else
1667                host_err = vfs_symlink(dentry->d_inode, dnew, path);
1668        err = nfserrno(host_err);
1669        if (!err)
1670                err = nfserrno(commit_metadata(fhp));
1671        fh_unlock(fhp);
1672
1673        fh_drop_write(fhp);
1674
1675        cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1676        dput(dnew);
1677        if (err==0) err = cerr;
1678out:
1679        return err;
1680
1681out_nfserr:
1682        err = nfserrno(host_err);
1683        goto out;
1684}
1685
1686/*
1687 * Create a hardlink
1688 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1689 */
1690__be32
1691nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1692                                char *name, int len, struct svc_fh *tfhp)
1693{
1694        struct dentry   *ddir, *dnew, *dold;
1695        struct inode    *dirp;
1696        __be32          err;
1697        int             host_err;
1698
1699        err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_CREATE);
1700        if (err)
1701                goto out;
1702        err = fh_verify(rqstp, tfhp, 0, NFSD_MAY_NOP);
1703        if (err)
1704                goto out;
1705        err = nfserr_isdir;
1706        if (S_ISDIR(tfhp->fh_dentry->d_inode->i_mode))
1707                goto out;
1708        err = nfserr_perm;
1709        if (!len)
1710                goto out;
1711        err = nfserr_exist;
1712        if (isdotent(name, len))
1713                goto out;
1714
1715        host_err = fh_want_write(tfhp);
1716        if (host_err) {
1717                err = nfserrno(host_err);
1718                goto out;
1719        }
1720
1721        fh_lock_nested(ffhp, I_MUTEX_PARENT);
1722        ddir = ffhp->fh_dentry;
1723        dirp = ddir->d_inode;
1724
1725        dnew = lookup_one_len(name, ddir, len);
1726        host_err = PTR_ERR(dnew);
1727        if (IS_ERR(dnew))
1728                goto out_nfserr;
1729
1730        dold = tfhp->fh_dentry;
1731
1732        err = nfserr_noent;
1733        if (!dold->d_inode)
1734                goto out_dput;
1735        host_err = nfsd_break_lease(dold->d_inode);
1736        if (host_err) {
1737                err = nfserrno(host_err);
1738                goto out_dput;
1739        }
1740        host_err = vfs_link(dold, dirp, dnew);
1741        if (!host_err) {
1742                err = nfserrno(commit_metadata(ffhp));
1743                if (!err)
1744                        err = nfserrno(commit_metadata(tfhp));
1745        } else {
1746                if (host_err == -EXDEV && rqstp->rq_vers == 2)
1747                        err = nfserr_acces;
1748                else
1749                        err = nfserrno(host_err);
1750        }
1751out_dput:
1752        dput(dnew);
1753out_unlock:
1754        fh_unlock(ffhp);
1755        fh_drop_write(tfhp);
1756out:
1757        return err;
1758
1759out_nfserr:
1760        err = nfserrno(host_err);
1761        goto out_unlock;
1762}
1763
1764/*
1765 * Rename a file
1766 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1767 */
1768__be32
1769nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1770                            struct svc_fh *tfhp, char *tname, int tlen)
1771{
1772        struct dentry   *fdentry, *tdentry, *odentry, *ndentry, *trap;
1773        struct inode    *fdir, *tdir;
1774        __be32          err;
1775        int             host_err;
1776
1777        err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_REMOVE);
1778        if (err)
1779                goto out;
1780        err = fh_verify(rqstp, tfhp, S_IFDIR, NFSD_MAY_CREATE);
1781        if (err)
1782                goto out;
1783
1784        fdentry = ffhp->fh_dentry;
1785        fdir = fdentry->d_inode;
1786
1787        tdentry = tfhp->fh_dentry;
1788        tdir = tdentry->d_inode;
1789
1790        err = nfserr_perm;
1791        if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1792                goto out;
1793
1794        host_err = fh_want_write(ffhp);
1795        if (host_err) {
1796                err = nfserrno(host_err);
1797                goto out;
1798        }
1799
1800        /* cannot use fh_lock as we need deadlock protective ordering
1801         * so do it by hand */
1802        trap = lock_rename(tdentry, fdentry);
1803        ffhp->fh_locked = tfhp->fh_locked = 1;
1804        fill_pre_wcc(ffhp);
1805        fill_pre_wcc(tfhp);
1806
1807        odentry = lookup_one_len(fname, fdentry, flen);
1808        host_err = PTR_ERR(odentry);
1809        if (IS_ERR(odentry))
1810                goto out_nfserr;
1811
1812        host_err = -ENOENT;
1813        if (!odentry->d_inode)
1814                goto out_dput_old;
1815        host_err = -EINVAL;
1816        if (odentry == trap)
1817                goto out_dput_old;
1818
1819        ndentry = lookup_one_len(tname, tdentry, tlen);
1820        host_err = PTR_ERR(ndentry);
1821        if (IS_ERR(ndentry))
1822                goto out_dput_old;
1823        host_err = -ENOTEMPTY;
1824        if (ndentry == trap)
1825                goto out_dput_new;
1826
1827        host_err = -EXDEV;
1828        if (ffhp->fh_export->ex_path.mnt != tfhp->fh_export->ex_path.mnt)
1829                goto out_dput_new;
1830        if (ffhp->fh_export->ex_path.dentry != tfhp->fh_export->ex_path.dentry)
1831                goto out_dput_new;
1832
1833        host_err = nfsd_break_lease(odentry->d_inode);
1834        if (host_err)
1835                goto out_dput_new;
1836        if (ndentry->d_inode) {
1837                host_err = nfsd_break_lease(ndentry->d_inode);
1838                if (host_err)
1839                        goto out_dput_new;
1840        }
1841        host_err = vfs_rename(fdir, odentry, tdir, ndentry);
1842        if (!host_err) {
1843                host_err = commit_metadata(tfhp);
1844                if (!host_err)
1845                        host_err = commit_metadata(ffhp);
1846        }
1847 out_dput_new:
1848        dput(ndentry);
1849 out_dput_old:
1850        dput(odentry);
1851 out_nfserr:
1852        err = nfserrno(host_err);
1853
1854        /* we cannot reply on fh_unlock on the two filehandles,
1855         * as that would do the wrong thing if the two directories
1856         * were the same, so again we do it by hand
1857         */
1858        fill_post_wcc(ffhp);
1859        fill_post_wcc(tfhp);
1860        unlock_rename(tdentry, fdentry);
1861        ffhp->fh_locked = tfhp->fh_locked = 0;
1862        fh_drop_write(ffhp);
1863
1864out:
1865        return err;
1866}
1867
1868/*
1869 * Unlink a file or directory
1870 * N.B. After this call fhp needs an fh_put
1871 */
1872__be32
1873nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1874                                char *fname, int flen)
1875{
1876        struct dentry   *dentry, *rdentry;
1877        struct inode    *dirp;
1878        __be32          err;
1879        int             host_err;
1880
1881        err = nfserr_acces;
1882        if (!flen || isdotent(fname, flen))
1883                goto out;
1884        err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_REMOVE);
1885        if (err)
1886                goto out;
1887
1888        host_err = fh_want_write(fhp);
1889        if (host_err)
1890                goto out_nfserr;
1891
1892        fh_lock_nested(fhp, I_MUTEX_PARENT);
1893        dentry = fhp->fh_dentry;
1894        dirp = dentry->d_inode;
1895
1896        rdentry = lookup_one_len(fname, dentry, flen);
1897        host_err = PTR_ERR(rdentry);
1898        if (IS_ERR(rdentry))
1899                goto out_nfserr;
1900
1901        if (!rdentry->d_inode) {
1902                dput(rdentry);
1903                err = nfserr_noent;
1904                goto out;
1905        }
1906
1907        if (!type)
1908                type = rdentry->d_inode->i_mode & S_IFMT;
1909
1910        host_err = nfsd_break_lease(rdentry->d_inode);
1911        if (host_err)
1912                goto out_put;
1913        if (type != S_IFDIR)
1914                host_err = vfs_unlink(dirp, rdentry);
1915        else
1916                host_err = vfs_rmdir(dirp, rdentry);
1917        if (!host_err)
1918                host_err = commit_metadata(fhp);
1919out_put:
1920        dput(rdentry);
1921
1922out_nfserr:
1923        err = nfserrno(host_err);
1924out:
1925        return err;
1926}
1927
1928/*
1929 * We do this buffering because we must not call back into the file
1930 * system's ->lookup() method from the filldir callback. That may well
1931 * deadlock a number of file systems.
1932 *
1933 * This is based heavily on the implementation of same in XFS.
1934 */
1935struct buffered_dirent {
1936        u64             ino;
1937        loff_t          offset;
1938        int             namlen;
1939        unsigned int    d_type;
1940        char            name[];
1941};
1942
1943struct readdir_data {
1944        struct dir_context ctx;
1945        char            *dirent;
1946        size_t          used;
1947        int             full;
1948};
1949
1950static int nfsd_buffered_filldir(void *__buf, const char *name, int namlen,
1951                                 loff_t offset, u64 ino, unsigned int d_type)
1952{
1953        struct readdir_data *buf = __buf;
1954        struct buffered_dirent *de = (void *)(buf->dirent + buf->used);
1955        unsigned int reclen;
1956
1957        reclen = ALIGN(sizeof(struct buffered_dirent) + namlen, sizeof(u64));
1958        if (buf->used + reclen > PAGE_SIZE) {
1959                buf->full = 1;
1960                return -EINVAL;
1961        }
1962
1963        de->namlen = namlen;
1964        de->offset = offset;
1965        de->ino = ino;
1966        de->d_type = d_type;
1967        memcpy(de->name, name, namlen);
1968        buf->used += reclen;
1969
1970        return 0;
1971}
1972
1973static __be32 nfsd_buffered_readdir(struct file *file, filldir_t func,
1974                                    struct readdir_cd *cdp, loff_t *offsetp)
1975{
1976        struct buffered_dirent *de;
1977        int host_err;
1978        int size;
1979        loff_t offset;
1980        struct readdir_data buf = {
1981                .ctx.actor = nfsd_buffered_filldir,
1982                .dirent = (void *)__get_free_page(GFP_KERNEL)
1983        };
1984
1985        if (!buf.dirent)
1986                return nfserrno(-ENOMEM);
1987
1988        offset = *offsetp;
1989
1990        while (1) {
1991                struct inode *dir_inode = file_inode(file);
1992                unsigned int reclen;
1993
1994                cdp->err = nfserr_eof; /* will be cleared on successful read */
1995                buf.used = 0;
1996                buf.full = 0;
1997
1998                host_err = iterate_dir(file, &buf.ctx);
1999                if (buf.full)
2000                        host_err = 0;
2001
2002                if (host_err < 0)
2003                        break;
2004
2005                size = buf.used;
2006
2007                if (!size)
2008                        break;
2009
2010                /*
2011                 * Various filldir functions may end up calling back into
2012                 * lookup_one_len() and the file system's ->lookup() method.
2013                 * These expect i_mutex to be held, as it would within readdir.
2014                 */
2015                host_err = mutex_lock_killable(&dir_inode->i_mutex);
2016                if (host_err)
2017                        break;
2018
2019                de = (struct buffered_dirent *)buf.dirent;
2020                while (size > 0) {
2021                        offset = de->offset;
2022
2023                        if (func(cdp, de->name, de->namlen, de->offset,
2024                                 de->ino, de->d_type))
2025                                break;
2026
2027                        if (cdp->err != nfs_ok)
2028                                break;
2029
2030                        reclen = ALIGN(sizeof(*de) + de->namlen,
2031                                       sizeof(u64));
2032                        size -= reclen;
2033                        de = (struct buffered_dirent *)((char *)de + reclen);
2034                }
2035                mutex_unlock(&dir_inode->i_mutex);
2036                if (size > 0) /* We bailed out early */
2037                        break;
2038
2039                offset = vfs_llseek(file, 0, SEEK_CUR);
2040        }
2041
2042        free_page((unsigned long)(buf.dirent));
2043
2044        if (host_err)
2045                return nfserrno(host_err);
2046
2047        *offsetp = offset;
2048        return cdp->err;
2049}
2050
2051/*
2052 * Read entries from a directory.
2053 * The  NFSv3/4 verifier we ignore for now.
2054 */
2055__be32
2056nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp, 
2057             struct readdir_cd *cdp, filldir_t func)
2058{
2059        __be32          err;
2060        struct file     *file;
2061        loff_t          offset = *offsetp;
2062        int             may_flags = NFSD_MAY_READ;
2063
2064        /* NFSv2 only supports 32 bit cookies */
2065        if (rqstp->rq_vers > 2)
2066                may_flags |= NFSD_MAY_64BIT_COOKIE;
2067
2068        err = nfsd_open(rqstp, fhp, S_IFDIR, may_flags, &file);
2069        if (err)
2070                goto out;
2071
2072        offset = vfs_llseek(file, offset, SEEK_SET);
2073        if (offset < 0) {
2074                err = nfserrno((int)offset);
2075                goto out_close;
2076        }
2077
2078        err = nfsd_buffered_readdir(file, func, cdp, offsetp);
2079
2080        if (err == nfserr_eof || err == nfserr_toosmall)
2081                err = nfs_ok; /* can still be found in ->err */
2082out_close:
2083        nfsd_close(file);
2084out:
2085        return err;
2086}
2087
2088/*
2089 * Get file system stats
2090 * N.B. After this call fhp needs an fh_put
2091 */
2092__be32
2093nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat, int access)
2094{
2095        __be32 err;
2096
2097        err = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP | access);
2098        if (!err) {
2099                struct path path = {
2100                        .mnt    = fhp->fh_export->ex_path.mnt,
2101                        .dentry = fhp->fh_dentry,
2102                };
2103                if (vfs_statfs(&path, stat))
2104                        err = nfserr_io;
2105        }
2106        return err;
2107}
2108
2109static int exp_rdonly(struct svc_rqst *rqstp, struct svc_export *exp)
2110{
2111        return nfsexp_flags(rqstp, exp) & NFSEXP_READONLY;
2112}
2113
2114/*
2115 * Check for a user's access permissions to this inode.
2116 */
2117__be32
2118nfsd_permission(struct svc_rqst *rqstp, struct svc_export *exp,
2119                                        struct dentry *dentry, int acc)
2120{
2121        struct inode    *inode = dentry->d_inode;
2122        int             err;
2123
2124        if ((acc & NFSD_MAY_MASK) == NFSD_MAY_NOP)
2125                return 0;
2126#if 0
2127        dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
2128                acc,
2129                (acc & NFSD_MAY_READ)?  " read"  : "",
2130                (acc & NFSD_MAY_WRITE)? " write" : "",
2131                (acc & NFSD_MAY_EXEC)?  " exec"  : "",
2132                (acc & NFSD_MAY_SATTR)? " sattr" : "",
2133                (acc & NFSD_MAY_TRUNC)? " trunc" : "",
2134                (acc & NFSD_MAY_LOCK)?  " lock"  : "",
2135                (acc & NFSD_MAY_OWNER_OVERRIDE)? " owneroverride" : "",
2136                inode->i_mode,
2137                IS_IMMUTABLE(inode)?    " immut" : "",
2138                IS_APPEND(inode)?       " append" : "",
2139                __mnt_is_readonly(exp->ex_path.mnt)?    " ro" : "");
2140        dprintk("      owner %d/%d user %d/%d\n",
2141                inode->i_uid, inode->i_gid, current_fsuid(), current_fsgid());
2142#endif
2143
2144        /* Normally we reject any write/sattr etc access on a read-only file
2145         * system.  But if it is IRIX doing check on write-access for a 
2146         * device special file, we ignore rofs.
2147         */
2148        if (!(acc & NFSD_MAY_LOCAL_ACCESS))
2149                if (acc & (NFSD_MAY_WRITE | NFSD_MAY_SATTR | NFSD_MAY_TRUNC)) {
2150                        if (exp_rdonly(rqstp, exp) ||
2151                            __mnt_is_readonly(exp->ex_path.mnt))
2152                                return nfserr_rofs;
2153                        if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode))
2154                                return nfserr_perm;
2155                }
2156        if ((acc & NFSD_MAY_TRUNC) && IS_APPEND(inode))
2157                return nfserr_perm;
2158
2159        if (acc & NFSD_MAY_LOCK) {
2160                /* If we cannot rely on authentication in NLM requests,
2161                 * just allow locks, otherwise require read permission, or
2162                 * ownership
2163                 */
2164                if (exp->ex_flags & NFSEXP_NOAUTHNLM)
2165                        return 0;
2166                else
2167                        acc = NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE;
2168        }
2169        /*
2170         * The file owner always gets access permission for accesses that
2171         * would normally be checked at open time. This is to make
2172         * file access work even when the client has done a fchmod(fd, 0).
2173         *
2174         * However, `cp foo bar' should fail nevertheless when bar is
2175         * readonly. A sensible way to do this might be to reject all
2176         * attempts to truncate a read-only file, because a creat() call
2177         * always implies file truncation.
2178         * ... but this isn't really fair.  A process may reasonably call
2179         * ftruncate on an open file descriptor on a file with perm 000.
2180         * We must trust the client to do permission checking - using "ACCESS"
2181         * with NFSv3.
2182         */
2183        if ((acc & NFSD_MAY_OWNER_OVERRIDE) &&
2184            uid_eq(inode->i_uid, current_fsuid()))
2185                return 0;
2186
2187        /* This assumes  NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */
2188        err = inode_permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC));
2189
2190        /* Allow read access to binaries even when mode 111 */
2191        if (err == -EACCES && S_ISREG(inode->i_mode) &&
2192             (acc == (NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE) ||
2193              acc == (NFSD_MAY_READ | NFSD_MAY_READ_IF_EXEC)))
2194                err = inode_permission(inode, MAY_EXEC);
2195
2196        return err? nfserrno(err) : 0;
2197}
2198
2199void
2200nfsd_racache_shutdown(void)
2201{
2202        struct raparms *raparm, *last_raparm;
2203        unsigned int i;
2204
2205        dprintk("nfsd: freeing readahead buffers.\n");
2206
2207        for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2208                raparm = raparm_hash[i].pb_head;
2209                while(raparm) {
2210                        last_raparm = raparm;
2211                        raparm = raparm->p_next;
2212                        kfree(last_raparm);
2213                }
2214                raparm_hash[i].pb_head = NULL;
2215        }
2216}
2217/*
2218 * Initialize readahead param cache
2219 */
2220int
2221nfsd_racache_init(int cache_size)
2222{
2223        int     i;
2224        int     j = 0;
2225        int     nperbucket;
2226        struct raparms **raparm = NULL;
2227
2228
2229        if (raparm_hash[0].pb_head)
2230                return 0;
2231        nperbucket = DIV_ROUND_UP(cache_size, RAPARM_HASH_SIZE);
2232        if (nperbucket < 2)
2233                nperbucket = 2;
2234        cache_size = nperbucket * RAPARM_HASH_SIZE;
2235
2236        dprintk("nfsd: allocating %d readahead buffers.\n", cache_size);
2237
2238        for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2239                spin_lock_init(&raparm_hash[i].pb_lock);
2240
2241                raparm = &raparm_hash[i].pb_head;
2242                for (j = 0; j < nperbucket; j++) {
2243                        *raparm = kzalloc(sizeof(struct raparms), GFP_KERNEL);
2244                        if (!*raparm)
2245                                goto out_nomem;
2246                        raparm = &(*raparm)->p_next;
2247                }
2248                *raparm = NULL;
2249        }
2250
2251        nfsdstats.ra_size = cache_size;
2252        return 0;
2253
2254out_nomem:
2255        dprintk("nfsd: kmalloc failed, freeing readahead buffers\n");
2256        nfsd_racache_shutdown();
2257        return -ENOMEM;
2258}
2259
2260#if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
2261struct posix_acl *
2262nfsd_get_posix_acl(struct svc_fh *fhp, int type)
2263{
2264        struct inode *inode = fhp->fh_dentry->d_inode;
2265        char *name;
2266        void *value = NULL;
2267        ssize_t size;
2268        struct posix_acl *acl;
2269
2270        if (!IS_POSIXACL(inode))
2271                return ERR_PTR(-EOPNOTSUPP);
2272
2273        switch (type) {
2274        case ACL_TYPE_ACCESS:
2275                name = POSIX_ACL_XATTR_ACCESS;
2276                break;
2277        case ACL_TYPE_DEFAULT:
2278                name = POSIX_ACL_XATTR_DEFAULT;
2279                break;
2280        default:
2281                return ERR_PTR(-EOPNOTSUPP);
2282        }
2283
2284        size = nfsd_getxattr(fhp->fh_dentry, name, &value);
2285        if (size < 0)
2286                return ERR_PTR(size);
2287
2288        acl = posix_acl_from_xattr(&init_user_ns, value, size);
2289        kfree(value);
2290        return acl;
2291}
2292
2293int
2294nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl)
2295{
2296        struct inode *inode = fhp->fh_dentry->d_inode;
2297        char *name;
2298        void *value = NULL;
2299        size_t size;
2300        int error;
2301
2302        if (!IS_POSIXACL(inode) ||
2303            !inode->i_op->setxattr || !inode->i_op->removexattr)
2304                return -EOPNOTSUPP;
2305        switch(type) {
2306                case ACL_TYPE_ACCESS:
2307                        name = POSIX_ACL_XATTR_ACCESS;
2308                        break;
2309                case ACL_TYPE_DEFAULT:
2310                        name = POSIX_ACL_XATTR_DEFAULT;
2311                        break;
2312                default:
2313                        return -EOPNOTSUPP;
2314        }
2315
2316        if (acl && acl->a_count) {
2317                size = posix_acl_xattr_size(acl->a_count);
2318                value = kmalloc(size, GFP_KERNEL);
2319                if (!value)
2320                        return -ENOMEM;
2321                error = posix_acl_to_xattr(&init_user_ns, acl, value, size);
2322                if (error < 0)
2323                        goto getout;
2324                size = error;
2325        } else
2326                size = 0;
2327
2328        error = fh_want_write(fhp);
2329        if (error)
2330                goto getout;
2331        if (size)
2332                error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0);
2333        else {
2334                if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT)
2335                        error = 0;
2336                else {
2337                        error = vfs_removexattr(fhp->fh_dentry, name);
2338                        if (error == -ENODATA)
2339                                error = 0;
2340                }
2341        }
2342        fh_drop_write(fhp);
2343
2344getout:
2345        kfree(value);
2346        return error;
2347}
2348#endif  /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
2349
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.