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
  32#ifdef CONFIG_NFSD_V3
  33#include "xdr3.h"
  34#endif /* CONFIG_NFSD_V3 */
  35
  36#ifdef CONFIG_NFSD_V4
  37#include "acl.h"
  38#include "idmap.h"
  39#endif /* CONFIG_NFSD_V4 */
  40
  41#include "nfsd.h"
  42#include "vfs.h"
  43
  44#define NFSDDBG_FACILITY                NFSDDBG_FILEOP
  45
  46
  47/*
  48 * This is a cache of readahead params that help us choose the proper
  49 * readahead strategy. Initially, we set all readahead parameters to 0
  50 * and let the VFS handle things.
  51 * If you increase the number of cached files very much, you'll need to
  52 * add a hash table here.
  53 */
  54struct raparms {
  55        struct raparms          *p_next;
  56        unsigned int            p_count;
  57        ino_t                   p_ino;
  58        dev_t                   p_dev;
  59        int                     p_set;
  60        struct file_ra_state    p_ra;
  61        unsigned int            p_hindex;
  62};
  63
  64struct raparm_hbucket {
  65        struct raparms          *pb_head;
  66        spinlock_t              pb_lock;
  67} ____cacheline_aligned_in_smp;
  68
  69#define RAPARM_HASH_BITS        4
  70#define RAPARM_HASH_SIZE        (1<<RAPARM_HASH_BITS)
  71#define RAPARM_HASH_MASK        (RAPARM_HASH_SIZE-1)
  72static struct raparm_hbucket    raparm_hash[RAPARM_HASH_SIZE];
  73
  74/* 
  75 * Called from nfsd_lookup and encode_dirent. Check if we have crossed 
  76 * a mount point.
  77 * Returns -EAGAIN or -ETIMEDOUT leaving *dpp and *expp unchanged,
  78 *  or nfs_ok having possibly changed *dpp and *expp
  79 */
  80int
  81nfsd_cross_mnt(struct svc_rqst *rqstp, struct dentry **dpp, 
  82                        struct svc_export **expp)
  83{
  84        struct svc_export *exp = *expp, *exp2 = NULL;
  85        struct dentry *dentry = *dpp;
  86        struct path path = {.mnt = mntget(exp->ex_path.mnt),
  87                            .dentry = dget(dentry)};
  88        int err = 0;
  89
  90        err = follow_down(&path);
  91        if (err < 0)
  92                goto out;
  93
  94        exp2 = rqst_exp_get_by_name(rqstp, &path);
  95        if (IS_ERR(exp2)) {
  96                err = PTR_ERR(exp2);
  97                /*
  98                 * We normally allow NFS clients to continue
  99                 * "underneath" a mountpoint that is not exported.
 100                 * The exception is V4ROOT, where no traversal is ever
 101                 * allowed without an explicit export of the new
 102                 * directory.
 103                 */
 104                if (err == -ENOENT && !(exp->ex_flags & NFSEXP_V4ROOT))
 105                        err = 0;
 106                path_put(&path);
 107                goto out;
 108        }
 109        if (nfsd_v4client(rqstp) ||
 110                (exp->ex_flags & NFSEXP_CROSSMOUNT) || EX_NOHIDE(exp2)) {
 111                /* successfully crossed mount point */
 112                /*
 113                 * This is subtle: path.dentry is *not* on path.mnt
 114                 * at this point.  The only reason we are safe is that
 115                 * original mnt is pinned down by exp, so we should
 116                 * put path *before* putting exp
 117                 */
 118                *dpp = path.dentry;
 119                path.dentry = dentry;
 120                *expp = exp2;
 121                exp2 = exp;
 122        }
 123        path_put(&path);
 124        exp_put(exp2);
 125out:
 126        return err;
 127}
 128
 129static void follow_to_parent(struct path *path)
 130{
 131        struct dentry *dp;
 132
 133        while (path->dentry == path->mnt->mnt_root && follow_up(path))
 134                ;
 135        dp = dget_parent(path->dentry);
 136        dput(path->dentry);
 137        path->dentry = dp;
 138}
 139
 140static int nfsd_lookup_parent(struct svc_rqst *rqstp, struct dentry *dparent, struct svc_export **exp, struct dentry **dentryp)
 141{
 142        struct svc_export *exp2;
 143        struct path path = {.mnt = mntget((*exp)->ex_path.mnt),
 144                            .dentry = dget(dparent)};
 145
 146        follow_to_parent(&path);
 147
 148        exp2 = rqst_exp_parent(rqstp, &path);
 149        if (PTR_ERR(exp2) == -ENOENT) {
 150                *dentryp = dget(dparent);
 151        } else if (IS_ERR(exp2)) {
 152                path_put(&path);
 153                return PTR_ERR(exp2);
 154        } else {
 155                *dentryp = dget(path.dentry);
 156                exp_put(*exp);
 157                *exp = exp2;
 158        }
 159        path_put(&path);
 160        return 0;
 161}
 162
 163/*
 164 * For nfsd purposes, we treat V4ROOT exports as though there was an
 165 * export at *every* directory.
 166 */
 167int nfsd_mountpoint(struct dentry *dentry, struct svc_export *exp)
 168{
 169        if (d_mountpoint(dentry))
 170                return 1;
 171        if (nfsd4_is_junction(dentry))
 172                return 1;
 173        if (!(exp->ex_flags & NFSEXP_V4ROOT))
 174                return 0;
 175        return dentry->d_inode != NULL;
 176}
 177
 178__be32
 179nfsd_lookup_dentry(struct svc_rqst *rqstp, struct svc_fh *fhp,
 180                   const char *name, unsigned int len,
 181                   struct svc_export **exp_ret, struct dentry **dentry_ret)
 182{
 183        struct svc_export       *exp;
 184        struct dentry           *dparent;
 185        struct dentry           *dentry;
 186        int                     host_err;
 187
 188        dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp), len,name);
 189
 190        dparent = fhp->fh_dentry;
 191        exp  = fhp->fh_export;
 192        exp_get(exp);
 193
 194        /* Lookup the name, but don't follow links */
 195        if (isdotent(name, len)) {
 196                if (len==1)
 197                        dentry = dget(dparent);
 198                else if (dparent != exp->ex_path.dentry)
 199                        dentry = dget_parent(dparent);
 200                else if (!EX_NOHIDE(exp) && !nfsd_v4client(rqstp))
 201                        dentry = dget(dparent); /* .. == . just like at / */
 202                else {
 203                        /* checking mountpoint crossing is very different when stepping up */
 204                        host_err = nfsd_lookup_parent(rqstp, dparent, &exp, &dentry);
 205                        if (host_err)
 206                                goto out_nfserr;
 207                }
 208        } else {
 209                fh_lock(fhp);
 210                dentry = lookup_one_len(name, dparent, len);
 211                host_err = PTR_ERR(dentry);
 212                if (IS_ERR(dentry))
 213                        goto out_nfserr;
 214                /*
 215                 * check if we have crossed a mount point ...
 216                 */
 217                if (nfsd_mountpoint(dentry, exp)) {
 218                        if ((host_err = nfsd_cross_mnt(rqstp, &dentry, &exp))) {
 219                                dput(dentry);
 220                                goto out_nfserr;
 221                        }
 222                }
 223        }
 224        *dentry_ret = dentry;
 225        *exp_ret = exp;
 226        return 0;
 227
 228out_nfserr:
 229        exp_put(exp);
 230        return nfserrno(host_err);
 231}
 232
 233/*
 234 * Look up one component of a pathname.
 235 * N.B. After this call _both_ fhp and resfh need an fh_put
 236 *
 237 * If the lookup would cross a mountpoint, and the mounted filesystem
 238 * is exported to the client with NFSEXP_NOHIDE, then the lookup is
 239 * accepted as it stands and the mounted directory is
 240 * returned. Otherwise the covered directory is returned.
 241 * NOTE: this mountpoint crossing is not supported properly by all
 242 *   clients and is explicitly disallowed for NFSv3
 243 *      NeilBrown <neilb@cse.unsw.edu.au>
 244 */
 245__be32
 246nfsd_lookup(struct svc_rqst *rqstp, struct svc_fh *fhp, const char *name,
 247                                unsigned int len, struct svc_fh *resfh)
 248{
 249        struct svc_export       *exp;
 250        struct dentry           *dentry;
 251        __be32 err;
 252
 253        err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC);
 254        if (err)
 255                return err;
 256        err = nfsd_lookup_dentry(rqstp, fhp, name, len, &exp, &dentry);
 257        if (err)
 258                return err;
 259        err = check_nfsd_access(exp, rqstp);
 260        if (err)
 261                goto out;
 262        /*
 263         * Note: we compose the file handle now, but as the
 264         * dentry may be negative, it may need to be updated.
 265         */
 266        err = fh_compose(resfh, exp, dentry, fhp);
 267        if (!err && !dentry->d_inode)
 268                err = nfserr_noent;
 269out:
 270        dput(dentry);
 271        exp_put(exp);
 272        return err;
 273}
 274
 275static int nfsd_break_lease(struct inode *inode)
 276{
 277        if (!S_ISREG(inode->i_mode))
 278                return 0;
 279        return break_lease(inode, O_WRONLY | O_NONBLOCK);
 280}
 281
 282/*
 283 * Commit metadata changes to stable storage.
 284 */
 285static int
 286commit_metadata(struct svc_fh *fhp)
 287{
 288        struct inode *inode = fhp->fh_dentry->d_inode;
 289        const struct export_operations *export_ops = inode->i_sb->s_export_op;
 290
 291        if (!EX_ISSYNC(fhp->fh_export))
 292                return 0;
 293
 294        if (export_ops->commit_metadata)
 295                return export_ops->commit_metadata(inode);
 296        return sync_inode_metadata(inode, 1);
 297}
 298
 299/*
 300 * Set various file attributes.
 301 * N.B. After this call fhp needs an fh_put
 302 */
 303__be32
 304nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
 305             int check_guard, time_t guardtime)
 306{
 307        struct dentry   *dentry;
 308        struct inode    *inode;
 309        int             accmode = NFSD_MAY_SATTR;
 310        umode_t         ftype = 0;
 311        __be32          err;
 312        int             host_err;
 313        int             size_change = 0;
 314
 315        if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE))
 316                accmode |= NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE;
 317        if (iap->ia_valid & ATTR_SIZE)
 318                ftype = S_IFREG;
 319
 320        /* Get inode */
 321        err = fh_verify(rqstp, fhp, ftype, accmode);
 322        if (err)
 323                goto out;
 324
 325        dentry = fhp->fh_dentry;
 326        inode = dentry->d_inode;
 327
 328        /* Ignore any mode updates on symlinks */
 329        if (S_ISLNK(inode->i_mode))
 330                iap->ia_valid &= ~ATTR_MODE;
 331
 332        if (!iap->ia_valid)
 333                goto out;
 334
 335        /*
 336         * NFSv2 does not differentiate between "set-[ac]time-to-now"
 337         * which only requires access, and "set-[ac]time-to-X" which
 338         * requires ownership.
 339         * So if it looks like it might be "set both to the same time which
 340         * is close to now", and if inode_change_ok fails, then we
 341         * convert to "set to now" instead of "set to explicit time"
 342         *
 343         * We only call inode_change_ok as the last test as technically
 344         * it is not an interface that we should be using.  It is only
 345         * valid if the filesystem does not define it's own i_op->setattr.
 346         */
 347#define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
 348#define MAX_TOUCH_TIME_ERROR (30*60)
 349        if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET &&
 350            iap->ia_mtime.tv_sec == iap->ia_atime.tv_sec) {
 351                /*
 352                 * Looks probable.
 353                 *
 354                 * Now just make sure time is in the right ballpark.
 355                 * Solaris, at least, doesn't seem to care what the time
 356                 * request is.  We require it be within 30 minutes of now.
 357                 */
 358                time_t delta = iap->ia_atime.tv_sec - get_seconds();
 359                if (delta < 0)
 360                        delta = -delta;
 361                if (delta < MAX_TOUCH_TIME_ERROR &&
 362                    inode_change_ok(inode, iap) != 0) {
 363                        /*
 364                         * Turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME.
 365                         * This will cause notify_change to set these times
 366                         * to "now"
 367                         */
 368                        iap->ia_valid &= ~BOTH_TIME_SET;
 369                }
 370        }
 371            
 372        /*
 373         * The size case is special.
 374         * It changes the file as well as the attributes.
 375         */
 376        if (iap->ia_valid & ATTR_SIZE) {
 377                if (iap->ia_size < inode->i_size) {
 378                        err = nfsd_permission(rqstp, fhp->fh_export, dentry,
 379                                        NFSD_MAY_TRUNC|NFSD_MAY_OWNER_OVERRIDE);
 380                        if (err)
 381                                goto out;
 382                }
 383
 384                host_err = get_write_access(inode);
 385                if (host_err)
 386                        goto out_nfserr;
 387
 388                size_change = 1;
 389                host_err = locks_verify_truncate(inode, NULL, iap->ia_size);
 390                if (host_err) {
 391                        put_write_access(inode);
 392                        goto out_nfserr;
 393                }
 394        }
 395
 396        /* sanitize the mode change */
 397        if (iap->ia_valid & ATTR_MODE) {
 398                iap->ia_mode &= S_IALLUGO;
 399                iap->ia_mode |= (inode->i_mode & ~S_IALLUGO);
 400        }
 401
 402        /* Revoke setuid/setgid on chown */
 403        if (!S_ISDIR(inode->i_mode) &&
 404            (((iap->ia_valid & ATTR_UID) && iap->ia_uid != inode->i_uid) ||
 405             ((iap->ia_valid & ATTR_GID) && iap->ia_gid != inode->i_gid))) {
 406                iap->ia_valid |= ATTR_KILL_PRIV;
 407                if (iap->ia_valid & ATTR_MODE) {
 408                        /* we're setting mode too, just clear the s*id bits */
 409                        iap->ia_mode &= ~S_ISUID;
 410                        if (iap->ia_mode & S_IXGRP)
 411                                iap->ia_mode &= ~S_ISGID;
 412                } else {
 413                        /* set ATTR_KILL_* bits and let VFS handle it */
 414                        iap->ia_valid |= (ATTR_KILL_SUID | ATTR_KILL_SGID);
 415                }
 416        }
 417
 418        /* Change the attributes. */
 419
 420        iap->ia_valid |= ATTR_CTIME;
 421
 422        err = nfserr_notsync;
 423        if (!check_guard || guardtime == inode->i_ctime.tv_sec) {
 424                host_err = nfsd_break_lease(inode);
 425                if (host_err)
 426                        goto out_nfserr;
 427                fh_lock(fhp);
 428
 429                host_err = notify_change(dentry, iap);
 430                err = nfserrno(host_err);
 431                fh_unlock(fhp);
 432        }
 433        if (size_change)
 434                put_write_access(inode);
 435        if (!err)
 436                commit_metadata(fhp);
 437out:
 438        return err;
 439
 440out_nfserr:
 441        err = nfserrno(host_err);
 442        goto out;
 443}
 444
 445#if defined(CONFIG_NFSD_V2_ACL) || \
 446    defined(CONFIG_NFSD_V3_ACL) || \
 447    defined(CONFIG_NFSD_V4)
 448static ssize_t nfsd_getxattr(struct dentry *dentry, char *key, void **buf)
 449{
 450        ssize_t buflen;
 451        ssize_t ret;
 452
 453        buflen = vfs_getxattr(dentry, key, NULL, 0);
 454        if (buflen <= 0)
 455                return buflen;
 456
 457        *buf = kmalloc(buflen, GFP_KERNEL);
 458        if (!*buf)
 459                return -ENOMEM;
 460
 461        ret = vfs_getxattr(dentry, key, *buf, buflen);
 462        if (ret < 0)
 463                kfree(*buf);
 464        return ret;
 465}
 466#endif
 467
 468#if defined(CONFIG_NFSD_V4)
 469static int
 470set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key)
 471{
 472        int len;
 473        size_t buflen;
 474        char *buf = NULL;
 475        int error = 0;
 476
 477        buflen = posix_acl_xattr_size(pacl->a_count);
 478        buf = kmalloc(buflen, GFP_KERNEL);
 479        error = -ENOMEM;
 480        if (buf == NULL)
 481                goto out;
 482
 483        len = posix_acl_to_xattr(&init_user_ns, pacl, buf, buflen);
 484        if (len < 0) {
 485                error = len;
 486                goto out;
 487        }
 488
 489        error = vfs_setxattr(dentry, key, buf, len, 0);
 490out:
 491        kfree(buf);
 492        return error;
 493}
 494
 495__be32
 496nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp,
 497    struct nfs4_acl *acl)
 498{
 499        __be32 error;
 500        int host_error;
 501        struct dentry *dentry;
 502        struct inode *inode;
 503        struct posix_acl *pacl = NULL, *dpacl = NULL;
 504        unsigned int flags = 0;
 505
 506        /* Get inode */
 507        error = fh_verify(rqstp, fhp, 0, NFSD_MAY_SATTR);
 508        if (error)
 509                return error;
 510
 511        dentry = fhp->fh_dentry;
 512        inode = dentry->d_inode;
 513        if (S_ISDIR(inode->i_mode))
 514                flags = NFS4_ACL_DIR;
 515
 516        host_error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags);
 517        if (host_error == -EINVAL) {
 518                return nfserr_attrnotsupp;
 519        } else if (host_error < 0)
 520                goto out_nfserr;
 521
 522        host_error = set_nfsv4_acl_one(dentry, pacl, POSIX_ACL_XATTR_ACCESS);
 523        if (host_error < 0)
 524                goto out_release;
 525
 526        if (S_ISDIR(inode->i_mode))
 527                host_error = set_nfsv4_acl_one(dentry, dpacl, POSIX_ACL_XATTR_DEFAULT);
 528
 529out_release:
 530        posix_acl_release(pacl);
 531        posix_acl_release(dpacl);
 532out_nfserr:
 533        if (host_error == -EOPNOTSUPP)
 534                return nfserr_attrnotsupp;
 535        else
 536                return nfserrno(host_error);
 537}
 538
 539static struct posix_acl *
 540_get_posix_acl(struct dentry *dentry, char *key)
 541{
 542        void *buf = NULL;
 543        struct posix_acl *pacl = NULL;
 544        int buflen;
 545
 546        buflen = nfsd_getxattr(dentry, key, &buf);
 547        if (!buflen)
 548                buflen = -ENODATA;
 549        if (buflen <= 0)
 550                return ERR_PTR(buflen);
 551
 552        pacl = posix_acl_from_xattr(&init_user_ns, buf, buflen);
 553        kfree(buf);
 554        return pacl;
 555}
 556
 557int
 558nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl)
 559{
 560        struct inode *inode = dentry->d_inode;
 561        int error = 0;
 562        struct posix_acl *pacl = NULL, *dpacl = NULL;
 563        unsigned int flags = 0;
 564
 565        pacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_ACCESS);
 566        if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA)
 567                pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL);
 568        if (IS_ERR(pacl)) {
 569                error = PTR_ERR(pacl);
 570                pacl = NULL;
 571                goto out;
 572        }
 573
 574        if (S_ISDIR(inode->i_mode)) {
 575                dpacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_DEFAULT);
 576                if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA)
 577                        dpacl = NULL;
 578                else if (IS_ERR(dpacl)) {
 579                        error = PTR_ERR(dpacl);
 580                        dpacl = NULL;
 581                        goto out;
 582                }
 583                flags = NFS4_ACL_DIR;
 584        }
 585
 586        *acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags);
 587        if (IS_ERR(*acl)) {
 588                error = PTR_ERR(*acl);
 589                *acl = NULL;
 590        }
 591 out:
 592        posix_acl_release(pacl);
 593        posix_acl_release(dpacl);
 594        return error;
 595}
 596
 597/*
 598 * NFS junction information is stored in an extended attribute.
 599 */
 600#define NFSD_JUNCTION_XATTR_NAME        XATTR_TRUSTED_PREFIX "junction.nfs"
 601
 602/**
 603 * nfsd4_is_junction - Test if an object could be an NFS junction
 604 *
 605 * @dentry: object to test
 606 *
 607 * Returns 1 if "dentry" appears to contain NFS junction information.
 608 * Otherwise 0 is returned.
 609 */
 610int nfsd4_is_junction(struct dentry *dentry)
 611{
 612        struct inode *inode = dentry->d_inode;
 613
 614        if (inode == NULL)
 615                return 0;
 616        if (inode->i_mode & S_IXUGO)
 617                return 0;
 618        if (!(inode->i_mode & S_ISVTX))
 619                return 0;
 620        if (vfs_getxattr(dentry, NFSD_JUNCTION_XATTR_NAME, NULL, 0) <= 0)
 621                return 0;
 622        return 1;
 623}
 624#endif /* defined(CONFIG_NFSD_V4) */
 625
 626#ifdef CONFIG_NFSD_V3
 627/*
 628 * Check server access rights to a file system object
 629 */
 630struct accessmap {
 631        u32             access;
 632        int             how;
 633};
 634static struct accessmap nfs3_regaccess[] = {
 635    {   NFS3_ACCESS_READ,       NFSD_MAY_READ                   },
 636    {   NFS3_ACCESS_EXECUTE,    NFSD_MAY_EXEC                   },
 637    {   NFS3_ACCESS_MODIFY,     NFSD_MAY_WRITE|NFSD_MAY_TRUNC   },
 638    {   NFS3_ACCESS_EXTEND,     NFSD_MAY_WRITE                  },
 639
 640    {   0,                      0                               }
 641};
 642
 643static struct accessmap nfs3_diraccess[] = {
 644    {   NFS3_ACCESS_READ,       NFSD_MAY_READ                   },
 645    {   NFS3_ACCESS_LOOKUP,     NFSD_MAY_EXEC                   },
 646    {   NFS3_ACCESS_MODIFY,     NFSD_MAY_EXEC|NFSD_MAY_WRITE|NFSD_MAY_TRUNC},
 647    {   NFS3_ACCESS_EXTEND,     NFSD_MAY_EXEC|NFSD_MAY_WRITE    },
 648    {   NFS3_ACCESS_DELETE,     NFSD_MAY_REMOVE                 },
 649
 650    {   0,                      0                               }
 651};
 652
 653static struct accessmap nfs3_anyaccess[] = {
 654        /* Some clients - Solaris 2.6 at least, make an access call
 655         * to the server to check for access for things like /dev/null
 656         * (which really, the server doesn't care about).  So
 657         * We provide simple access checking for them, looking
 658         * mainly at mode bits, and we make sure to ignore read-only
 659         * filesystem checks
 660         */
 661    {   NFS3_ACCESS_READ,       NFSD_MAY_READ                   },
 662    {   NFS3_ACCESS_EXECUTE,    NFSD_MAY_EXEC                   },
 663    {   NFS3_ACCESS_MODIFY,     NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS    },
 664    {   NFS3_ACCESS_EXTEND,     NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS    },
 665
 666    {   0,                      0                               }
 667};
 668
 669__be32
 670nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
 671{
 672        struct accessmap        *map;
 673        struct svc_export       *export;
 674        struct dentry           *dentry;
 675        u32                     query, result = 0, sresult = 0;
 676        __be32                  error;
 677
 678        error = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP);
 679        if (error)
 680                goto out;
 681
 682        export = fhp->fh_export;
 683        dentry = fhp->fh_dentry;
 684
 685        if (S_ISREG(dentry->d_inode->i_mode))
 686                map = nfs3_regaccess;
 687        else if (S_ISDIR(dentry->d_inode->i_mode))
 688                map = nfs3_diraccess;
 689        else
 690                map = nfs3_anyaccess;
 691
 692
 693        query = *access;
 694        for  (; map->access; map++) {
 695                if (map->access & query) {
 696                        __be32 err2;
 697
 698                        sresult |= map->access;
 699
 700                        err2 = nfsd_permission(rqstp, export, dentry, map->how);
 701                        switch (err2) {
 702                        case nfs_ok:
 703                                result |= map->access;
 704                                break;
 705                                
 706                        /* the following error codes just mean the access was not allowed,
 707                         * rather than an error occurred */
 708                        case nfserr_rofs:
 709                        case nfserr_acces:
 710                        case nfserr_perm:
 711                                /* simply don't "or" in the access bit. */
 712                                break;
 713                        default:
 714                                error = err2;
 715                                goto out;
 716                        }
 717                }
 718        }
 719        *access = result;
 720        if (supported)
 721                *supported = sresult;
 722
 723 out:
 724        return error;
 725}
 726#endif /* CONFIG_NFSD_V3 */
 727
 728static int nfsd_open_break_lease(struct inode *inode, int access)
 729{
 730        unsigned int mode;
 731
 732        if (access & NFSD_MAY_NOT_BREAK_LEASE)
 733                return 0;
 734        mode = (access & NFSD_MAY_WRITE) ? O_WRONLY : O_RDONLY;
 735        return break_lease(inode, mode | O_NONBLOCK);
 736}
 737
 738/*
 739 * Open an existing file or directory.
 740 * The may_flags argument indicates the type of open (read/write/lock)
 741 * and additional flags.
 742 * N.B. After this call fhp needs an fh_put
 743 */
 744__be32
 745nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, umode_t type,
 746                        int may_flags, struct file **filp)
 747{
 748        struct path     path;
 749        struct inode    *inode;
 750        int             flags = O_RDONLY|O_LARGEFILE;
 751        __be32          err;
 752        int             host_err = 0;
 753
 754        validate_process_creds();
 755
 756        /*
 757         * If we get here, then the client has already done an "open",
 758         * and (hopefully) checked permission - so allow OWNER_OVERRIDE
 759         * in case a chmod has now revoked permission.
 760         *
 761         * Arguably we should also allow the owner override for
 762         * directories, but we never have and it doesn't seem to have
 763         * caused anyone a problem.  If we were to change this, note
 764         * also that our filldir callbacks would need a variant of
 765         * lookup_one_len that doesn't check permissions.
 766         */
 767        if (type == S_IFREG)
 768                may_flags |= NFSD_MAY_OWNER_OVERRIDE;
 769        err = fh_verify(rqstp, fhp, type, may_flags);
 770        if (err)
 771                goto out;
 772
 773        path.mnt = fhp->fh_export->ex_path.mnt;
 774        path.dentry = fhp->fh_dentry;
 775        inode = path.dentry->d_inode;
 776
 777        /* Disallow write access to files with the append-only bit set
 778         * or any access when mandatory locking enabled
 779         */
 780        err = nfserr_perm;
 781        if (IS_APPEND(inode) && (may_flags & NFSD_MAY_WRITE))
 782                goto out;
 783        /*
 784         * We must ignore files (but only files) which might have mandatory
 785         * locks on them because there is no way to know if the accesser has
 786         * the lock.
 787         */
 788        if (S_ISREG((inode)->i_mode) && mandatory_lock(inode))
 789                goto out;
 790
 791        if (!inode->i_fop)
 792                goto out;
 793
 794        host_err = nfsd_open_break_lease(inode, may_flags);
 795        if (host_err) /* NOMEM or WOULDBLOCK */
 796                goto out_nfserr;
 797
 798        if (may_flags & NFSD_MAY_WRITE) {
 799                if (may_flags & NFSD_MAY_READ)
 800                        flags = O_RDWR|O_LARGEFILE;
 801                else
 802                        flags = O_WRONLY|O_LARGEFILE;
 803        }
 804        *filp = dentry_open(&path, flags, current_cred());
 805        if (IS_ERR(*filp))
 806                host_err = PTR_ERR(*filp);
 807        else {
 808                host_err = ima_file_check(*filp, may_flags);
 809
 810                if (may_flags & NFSD_MAY_64BIT_COOKIE)
 811                        (*filp)->f_mode |= FMODE_64BITHASH;
 812                else
 813                        (*filp)->f_mode |= FMODE_32BITHASH;
 814        }
 815
 816out_nfserr:
 817        err = nfserrno(host_err);
 818out:
 819        validate_process_creds();
 820        return err;
 821}
 822
 823/*
 824 * Close a file.
 825 */
 826void
 827nfsd_close(struct file *filp)
 828{
 829        fput(filp);
 830}
 831
 832/*
 833 * Obtain the readahead parameters for the file
 834 * specified by (dev, ino).
 835 */
 836
 837static inline struct raparms *
 838nfsd_get_raparms(dev_t dev, ino_t ino)
 839{
 840        struct raparms  *ra, **rap, **frap = NULL;
 841        int depth = 0;
 842        unsigned int hash;
 843        struct raparm_hbucket *rab;
 844
 845        hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK;
 846        rab = &raparm_hash[hash];
 847
 848        spin_lock(&rab->pb_lock);
 849        for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) {
 850                if (ra->p_ino == ino && ra->p_dev == dev)
 851                        goto found;
 852                depth++;
 853                if (ra->p_count == 0)
 854                        frap = rap;
 855        }
 856        depth = nfsdstats.ra_size;
 857        if (!frap) {    
 858                spin_unlock(&rab->pb_lock);
 859                return NULL;
 860        }
 861        rap = frap;
 862        ra = *frap;
 863        ra->p_dev = dev;
 864        ra->p_ino = ino;
 865        ra->p_set = 0;
 866        ra->p_hindex = hash;
 867found:
 868        if (rap != &rab->pb_head) {
 869                *rap = ra->p_next;
 870                ra->p_next   = rab->pb_head;
 871                rab->pb_head = ra;
 872        }
 873        ra->p_count++;
 874        nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
 875        spin_unlock(&rab->pb_lock);
 876        return ra;
 877}
 878
 879/*
 880 * Grab and keep cached pages associated with a file in the svc_rqst
 881 * so that they can be passed to the network sendmsg/sendpage routines
 882 * directly. They will be released after the sending has completed.
 883 */
 884static int
 885nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
 886                  struct splice_desc *sd)
 887{
 888        struct svc_rqst *rqstp = sd->u.data;
 889        struct page **pp = rqstp->rq_respages + rqstp->rq_resused;
 890        struct page *page = buf->page;
 891        size_t size;
 892
 893        size = sd->len;
 894
 895        if (rqstp->rq_res.page_len == 0) {
 896                get_page(page);
 897                put_page(*pp);
 898                *pp = page;
 899                rqstp->rq_resused++;
 900                rqstp->rq_res.page_base = buf->offset;
 901                rqstp->rq_res.page_len = size;
 902        } else if (page != pp[-1]) {
 903                get_page(page);
 904                if (*pp)
 905                        put_page(*pp);
 906                *pp = page;
 907                rqstp->rq_resused++;
 908                rqstp->rq_res.page_len += size;
 909        } else
 910                rqstp->rq_res.page_len += size;
 911
 912        return size;
 913}
 914
 915static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe,
 916                                    struct splice_desc *sd)
 917{
 918        return __splice_from_pipe(pipe, sd, nfsd_splice_actor);
 919}
 920
 921static __be32
 922nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
 923              loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
 924{
 925        mm_segment_t    oldfs;
 926        __be32          err;
 927        int             host_err;
 928
 929        err = nfserr_perm;
 930
 931        if (file->f_op->splice_read && rqstp->rq_splice_ok) {
 932                struct splice_desc sd = {
 933                        .len            = 0,
 934                        .total_len      = *count,
 935                        .pos            = offset,
 936                        .u.data         = rqstp,
 937                };
 938
 939                rqstp->rq_resused = 1;
 940                host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor);
 941        } else {
 942                oldfs = get_fs();
 943                set_fs(KERNEL_DS);
 944                host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
 945                set_fs(oldfs);
 946        }
 947
 948        if (host_err >= 0) {
 949                nfsdstats.io_read += host_err;
 950                *count = host_err;
 951                err = 0;
 952                fsnotify_access(file);
 953        } else 
 954                err = nfserrno(host_err);
 955        return err;
 956}
 957
 958static void kill_suid(struct dentry *dentry)
 959{
 960        struct iattr    ia;
 961        ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
 962
 963        mutex_lock(&dentry->d_inode->i_mutex);
 964        notify_change(dentry, &ia);
 965        mutex_unlock(&dentry->d_inode->i_mutex);
 966}
 967
 968/*
 969 * Gathered writes: If another process is currently writing to the file,
 970 * there's a high chance this is another nfsd (triggered by a bulk write
 971 * from a client's biod). Rather than syncing the file with each write
 972 * request, we sleep for 10 msec.
 973 *
 974 * I don't know if this roughly approximates C. Juszak's idea of
 975 * gathered writes, but it's a nice and simple solution (IMHO), and it
 976 * seems to work:-)
 977 *
 978 * Note: we do this only in the NFSv2 case, since v3 and higher have a
 979 * better tool (separate unstable writes and commits) for solving this
 980 * problem.
 981 */
 982static int wait_for_concurrent_writes(struct file *file)
 983{
 984        struct inode *inode = file->f_path.dentry->d_inode;
 985        static ino_t last_ino;
 986        static dev_t last_dev;
 987        int err = 0;
 988
 989        if (atomic_read(&inode->i_writecount) > 1
 990            || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
 991                dprintk("nfsd: write defer %d\n", task_pid_nr(current));
 992                msleep(10);
 993                dprintk("nfsd: write resume %d\n", task_pid_nr(current));
 994        }
 995
 996        if (inode->i_state & I_DIRTY) {
 997                dprintk("nfsd: write sync %d\n", task_pid_nr(current));
 998                err = vfs_fsync(file, 0);
 999        }
1000        last_ino = inode->i_ino;
1001        last_dev = inode->i_sb->s_dev;
1002        return err;
1003}
1004
1005static __be32
1006nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1007                                loff_t offset, struct kvec *vec, int vlen,
1008                                unsigned long *cnt, int *stablep)
1009{
1010        struct svc_export       *exp;
1011        struct dentry           *dentry;
1012        struct inode            *inode;
1013        mm_segment_t            oldfs;
1014        __be32                  err = 0;
1015        int                     host_err;
1016        int                     stable = *stablep;
1017        int                     use_wgather;
1018
1019        dentry = file->f_path.dentry;
1020        inode = dentry->d_inode;
1021        exp   = fhp->fh_export;
1022
1023        /*
1024         * Request sync writes if
1025         *  -   the sync export option has been set, or
1026         *  -   the client requested O_SYNC behavior (NFSv3 feature).
1027         *  -   The file system doesn't support fsync().
1028         * When NFSv2 gathered writes have been configured for this volume,
1029         * flushing the data to disk is handled separately below.
1030         */
1031        use_wgather = (rqstp->rq_vers == 2) && EX_WGATHER(exp);
1032
1033        if (!file->f_op->fsync) {/* COMMIT3 cannot work */
1034               stable = 2;
1035               *stablep = 2; /* FILE_SYNC */
1036        }
1037
1038        if (!EX_ISSYNC(exp))
1039                stable = 0;
1040        if (stable && !use_wgather) {
1041                spin_lock(&file->f_lock);
1042                file->f_flags |= O_SYNC;
1043                spin_unlock(&file->f_lock);
1044        }
1045
1046        /* Write the data. */
1047        oldfs = get_fs(); set_fs(KERNEL_DS);
1048        host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset);
1049        set_fs(oldfs);
1050        if (host_err < 0)
1051                goto out_nfserr;
1052        *cnt = host_err;
1053        nfsdstats.io_write += host_err;
1054        fsnotify_modify(file);
1055
1056        /* clear setuid/setgid flag after write */
1057        if (inode->i_mode & (S_ISUID | S_ISGID))
1058                kill_suid(dentry);
1059
1060        if (stable && use_wgather)
1061                host_err = wait_for_concurrent_writes(file);
1062
1063out_nfserr:
1064        dprintk("nfsd: write complete host_err=%d\n", host_err);
1065        if (host_err >= 0)
1066                err = 0;
1067        else
1068                err = nfserrno(host_err);
1069        return err;
1070}
1071
1072/*
1073 * Read data from a file. count must contain the requested read count
1074 * on entry. On return, *count contains the number of bytes actually read.
1075 * N.B. After this call fhp needs an fh_put
1076 */
1077__be32 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp,
1078        loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
1079{
1080        struct file *file;
1081        struct inode *inode;
1082        struct raparms  *ra;
1083        __be32 err;
1084
1085        err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
1086        if (err)
1087                return err;
1088
1089        inode = file->f_path.dentry->d_inode;
1090
1091        /* Get readahead parameters */
1092        ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
1093
1094        if (ra && ra->p_set)
1095                file->f_ra = ra->p_ra;
1096
1097        err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1098
1099        /* Write back readahead params */
1100        if (ra) {
1101                struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex];
1102                spin_lock(&rab->pb_lock);
1103                ra->p_ra = file->f_ra;
1104                ra->p_set = 1;
1105                ra->p_count--;
1106                spin_unlock(&rab->pb_lock);
1107        }
1108
1109        nfsd_close(file);
1110        return err;
1111}
1112
1113/* As above, but use the provided file descriptor. */
1114__be32
1115nfsd_read_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1116                loff_t offset, struct kvec *vec, int vlen,
1117                unsigned long *count)
1118{
1119        __be32          err;
1120
1121        if (file) {
1122                err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1123                                NFSD_MAY_READ|NFSD_MAY_OWNER_OVERRIDE);
1124                if (err)
1125                        goto out;
1126                err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1127        } else /* Note file may still be NULL in NFSv4 special stateid case: */
1128                err = nfsd_read(rqstp, fhp, offset, vec, vlen, count);
1129out:
1130        return err;
1131}
1132
1133/*
1134 * Write data to a file.
1135 * The stable flag requests synchronous writes.
1136 * N.B. After this call fhp needs an fh_put
1137 */
1138__be32
1139nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1140                loff_t offset, struct kvec *vec, int vlen, unsigned long *cnt,
1141                int *stablep)
1142{
1143        __be32                  err = 0;
1144
1145        if (file) {
1146                err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1147                                NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE);
1148                if (err)
1149                        goto out;
1150                err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1151                                stablep);
1152        } else {
1153                err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file);
1154                if (err)
1155                        goto out;
1156
1157                if (cnt)
1158                        err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1159                                             cnt, stablep);
1160                nfsd_close(file);
1161        }
1162out:
1163        return err;
1164}
1165
1166#ifdef CONFIG_NFSD_V3
1167/*
1168 * Commit all pending writes to stable storage.
1169 *
1170 * Note: we only guarantee that data that lies within the range specified
1171 * by the 'offset' and 'count' parameters will be synced.
1172 *
1173 * Unfortunately we cannot lock the file to make sure we return full WCC
1174 * data to the client, as locking happens lower down in the filesystem.
1175 */
1176__be32
1177nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1178               loff_t offset, unsigned long count)
1179{
1180        struct file     *file;
1181        loff_t          end = LLONG_MAX;
1182        __be32          err = nfserr_inval;
1183
1184        if (offset < 0)
1185                goto out;
1186        if (count != 0) {
1187                end = offset + (loff_t)count - 1;
1188                if (end < offset)
1189                        goto out;
1190        }
1191
1192        err = nfsd_open(rqstp, fhp, S_IFREG,
1193                        NFSD_MAY_WRITE|NFSD_MAY_NOT_BREAK_LEASE, &file);
1194        if (err)
1195                goto out;
1196        if (EX_ISSYNC(fhp->fh_export)) {
1197                int err2 = vfs_fsync_range(file, offset, end, 0);
1198
1199                if (err2 != -EINVAL)
1200                        err = nfserrno(err2);
1201                else
1202                        err = nfserr_notsupp;
1203        }
1204
1205        nfsd_close(file);
1206out:
1207        return err;
1208}
1209#endif /* CONFIG_NFSD_V3 */
1210
1211static __be32
1212nfsd_create_setattr(struct svc_rqst *rqstp, struct svc_fh *resfhp,
1213                        struct iattr *iap)
1214{
1215        /*
1216         * Mode has already been set earlier in create:
1217         */
1218        iap->ia_valid &= ~ATTR_MODE;
1219        /*
1220         * Setting uid/gid works only for root.  Irix appears to
1221         * send along the gid on create when it tries to implement
1222         * setgid directories via NFS:
1223         */
1224        if (current_fsuid() != 0)
1225                iap->ia_valid &= ~(ATTR_UID|ATTR_GID);
1226        if (iap->ia_valid)
1227                return nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1228        return 0;
1229}
1230
1231/* HPUX client sometimes creates a file in mode 000, and sets size to 0.
1232 * setting size to 0 may fail for some specific file systems by the permission
1233 * checking which requires WRITE permission but the mode is 000.
1234 * we ignore the resizing(to 0) on the just new created file, since the size is
1235 * 0 after file created.
1236 *
1237 * call this only after vfs_create() is called.
1238 * */
1239static void
1240nfsd_check_ignore_resizing(struct iattr *iap)
1241{
1242        if ((iap->ia_valid & ATTR_SIZE) && (iap->ia_size == 0))
1243                iap->ia_valid &= ~ATTR_SIZE;
1244}
1245
1246/*
1247 * Create a file (regular, directory, device, fifo); UNIX sockets 
1248 * not yet implemented.
1249 * If the response fh has been verified, the parent directory should
1250 * already be locked. Note that the parent directory is left locked.
1251 *
1252 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1253 */
1254__be32
1255nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1256                char *fname, int flen, struct iattr *iap,
1257                int type, dev_t rdev, struct svc_fh *resfhp)
1258{
1259        struct dentry   *dentry, *dchild = NULL;
1260        struct inode    *dirp;
1261        __be32          err;
1262        __be32          err2;
1263        int             host_err;
1264
1265        err = nfserr_perm;
1266        if (!flen)
1267                goto out;
1268        err = nfserr_exist;
1269        if (isdotent(fname, flen))
1270                goto out;
1271
1272        err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1273        if (err)
1274                goto out;
1275
1276        dentry = fhp->fh_dentry;
1277        dirp = dentry->d_inode;
1278
1279        err = nfserr_notdir;
1280        if (!dirp->i_op->lookup)
1281                goto out;
1282        /*
1283         * Check whether the response file handle has been verified yet.
1284         * If it has, the parent directory should already be locked.
1285         */
1286        if (!resfhp->fh_dentry) {
1287                host_err = fh_want_write(fhp);
1288                if (host_err)
1289                        goto out_nfserr;
1290
1291                /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1292                fh_lock_nested(fhp, I_MUTEX_PARENT);
1293                dchild = lookup_one_len(fname, dentry, flen);
1294                host_err = PTR_ERR(dchild);
1295                if (IS_ERR(dchild))
1296                        goto out_nfserr;
1297                err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1298                if (err)
1299                        goto out;
1300        } else {
1301                /* called from nfsd_proc_create */
1302                dchild = dget(resfhp->fh_dentry);
1303                if (!fhp->fh_locked) {
1304                        /* not actually possible */
1305                        printk(KERN_ERR
1306                                "nfsd_create: parent %s/%s not locked!\n",
1307                                dentry->d_parent->d_name.name,
1308                                dentry->d_name.name);
1309                        err = nfserr_io;
1310                        goto out;
1311                }
1312        }
1313        /*
1314         * Make sure the child dentry is still negative ...
1315         */
1316        err = nfserr_exist;
1317        if (dchild->d_inode) {
1318                dprintk("nfsd_create: dentry %s/%s not negative!\n",
1319                        dentry->d_name.name, dchild->d_name.name);
1320                goto out; 
1321        }
1322
1323        if (!(iap->ia_valid & ATTR_MODE))
1324                iap->ia_mode = 0;
1325        iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1326
1327        err = nfserr_inval;
1328        if (!S_ISREG(type) && !S_ISDIR(type) && !special_file(type)) {
1329                printk(KERN_WARNING "nfsd: bad file type %o in nfsd_create\n",
1330                       type);
1331                goto out;
1332        }
1333
1334        /*
1335         * Get the dir op function pointer.
1336         */
1337        err = 0;
1338        host_err = 0;
1339        switch (type) {
1340        case S_IFREG:
1341                host_err = vfs_create(dirp, dchild, iap->ia_mode, true);
1342                if (!host_err)
1343                        nfsd_check_ignore_resizing(iap);
1344                break;
1345        case S_IFDIR:
1346                host_err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1347                break;
1348        case S_IFCHR:
1349        case S_IFBLK:
1350        case S_IFIFO:
1351        case S_IFSOCK:
1352                host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1353                break;
1354        }
1355        if (host_err < 0)
1356                goto out_nfserr;
1357
1358        err = nfsd_create_setattr(rqstp, resfhp, iap);
1359
1360        /*
1361         * nfsd_setattr already committed the child.  Transactional filesystems
1362         * had a chance to commit changes for both parent and child
1363         * simultaneously making the following commit_metadata a noop.
1364         */
1365        err2 = nfserrno(commit_metadata(fhp));
1366        if (err2)
1367                err = err2;
1368        /*
1369         * Update the file handle to get the new inode info.
1370         */
1371        if (!err)
1372                err = fh_update(resfhp);
1373out:
1374        if (dchild && !IS_ERR(dchild))
1375                dput(dchild);
1376        return err;
1377
1378out_nfserr:
1379        err = nfserrno(host_err);
1380        goto out;
1381}
1382
1383#ifdef CONFIG_NFSD_V3
1384
1385static inline int nfsd_create_is_exclusive(int createmode)
1386{
1387        return createmode == NFS3_CREATE_EXCLUSIVE
1388               || createmode == NFS4_CREATE_EXCLUSIVE4_1;
1389}
1390
1391/*
1392 * NFSv3 and NFSv4 version of nfsd_create
1393 */
1394__be32
1395do_nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1396                char *fname, int flen, struct iattr *iap,
1397                struct svc_fh *resfhp, int createmode, u32 *verifier,
1398                bool *truncp, bool *created)
1399{
1400        struct dentry   *dentry, *dchild = NULL;
1401        struct inode    *dirp;
1402        __be32          err;
1403        int             host_err;
1404        __u32           v_mtime=0, v_atime=0;
1405
1406        err = nfserr_perm;
1407        if (!flen)
1408                goto out;
1409        err = nfserr_exist;
1410        if (isdotent(fname, flen))
1411                goto out;
1412        if (!(iap->ia_valid & ATTR_MODE))
1413                iap->ia_mode = 0;
1414        err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC);
1415        if (err)
1416                goto out;
1417
1418        dentry = fhp->fh_dentry;
1419        dirp = dentry->d_inode;
1420
1421        /* Get all the sanity checks out of the way before
1422         * we lock the parent. */
1423        err = nfserr_notdir;
1424        if (!dirp->i_op->lookup)
1425                goto out;
1426
1427        host_err = fh_want_write(fhp);
1428        if (host_err)
1429                goto out_nfserr;
1430
1431        fh_lock_nested(fhp, I_MUTEX_PARENT);
1432
1433        /*
1434         * Compose the response file handle.
1435         */
1436        dchild = lookup_one_len(fname, dentry, flen);
1437        host_err = PTR_ERR(dchild);
1438        if (IS_ERR(dchild))
1439                goto out_nfserr;
1440
1441        /* If file doesn't exist, check for permissions to create one */
1442        if (!dchild->d_inode) {
1443                err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1444                if (err)
1445                        goto out;
1446        }
1447
1448        err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1449        if (err)
1450                goto out;
1451
1452        if (nfsd_create_is_exclusive(createmode)) {
1453                /* solaris7 gets confused (bugid 4218508) if these have
1454                 * the high bit set, so just clear the high bits. If this is
1455                 * ever changed to use different attrs for storing the
1456                 * verifier, then do_open_lookup() will also need to be fixed
1457                 * accordingly.
1458                 */
1459                v_mtime = verifier[0]&0x7fffffff;
1460                v_atime = verifier[1]&0x7fffffff;
1461        }
1462        
1463        if (dchild->d_inode) {
1464                err = 0;
1465
1466                switch (createmode) {
1467                case NFS3_CREATE_UNCHECKED:
1468                        if (! S_ISREG(dchild->d_inode->i_mode))
1469                                goto out;
1470                        else if (truncp) {
1471                                /* in nfsv4, we need to treat this case a little
1472                                 * differently.  we don't want to truncate the
1473                                 * file now; this would be wrong if the OPEN
1474                                 * fails for some other reason.  furthermore,
1475                                 * if the size is nonzero, we should ignore it
1476                                 * according to spec!
1477                                 */
1478                                *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1479                        }
1480                        else {
1481                                iap->ia_valid &= ATTR_SIZE;
1482                                goto set_attr;
1483                        }
1484                        break;
1485                case NFS3_CREATE_EXCLUSIVE:
1486                        if (   dchild->d_inode->i_mtime.tv_sec == v_mtime
1487                            && dchild->d_inode->i_atime.tv_sec == v_atime
1488                            && dchild->d_inode->i_size  == 0 ) {
1489                                if (created)
1490                                        *created = 1;
1491                                break;
1492                        }
1493                case NFS4_CREATE_EXCLUSIVE4_1:
1494                        if (   dchild->d_inode->i_mtime.tv_sec == v_mtime
1495                            && dchild->d_inode->i_atime.tv_sec == v_atime
1496                            && dchild->d_inode->i_size  == 0 ) {
1497                                if (created)
1498                                        *created = 1;
1499                                goto set_attr;
1500                        }
1501                         /* fallthru */
1502                case NFS3_CREATE_GUARDED:
1503                        err = nfserr_exist;
1504                }
1505                fh_drop_write(fhp);
1506                goto out;
1507        }
1508
1509        host_err = vfs_create(dirp, dchild, iap->ia_mode, true);
1510        if (host_err < 0) {
1511                fh_drop_write(fhp);
1512                goto out_nfserr;
1513        }
1514        if (created)
1515                *created = 1;
1516
1517        nfsd_check_ignore_resizing(iap);
1518
1519        if (nfsd_create_is_exclusive(createmode)) {
1520                /* Cram the verifier into atime/mtime */
1521                iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1522                        | ATTR_MTIME_SET|ATTR_ATIME_SET;
1523                /* XXX someone who knows this better please fix it for nsec */ 
1524                iap->ia_mtime.tv_sec = v_mtime;
1525                iap->ia_atime.tv_sec = v_atime;
1526                iap->ia_mtime.tv_nsec = 0;
1527                iap->ia_atime.tv_nsec = 0;
1528        }
1529
1530 set_attr:
1531        err = nfsd_create_setattr(rqstp, resfhp, iap);
1532
1533        /*
1534         * nfsd_setattr already committed the child (and possibly also the parent).
1535         */
1536        if (!err)
1537                err = nfserrno(commit_metadata(fhp));
1538
1539        /*
1540         * Update the filehandle to get the new inode info.
1541         */
1542        if (!err)
1543                err = fh_update(resfhp);
1544
1545 out:
1546        fh_unlock(fhp);
1547        if (dchild && !IS_ERR(dchild))
1548                dput(dchild);
1549        fh_drop_write(fhp);
1550        return err;
1551 
1552 out_nfserr:
1553        err = nfserrno(host_err);
1554        goto out;
1555}
1556#endif /* CONFIG_NFSD_V3 */
1557
1558/*
1559 * Read a symlink. On entry, *lenp must contain the maximum path length that
1560 * fits into the buffer. On return, it contains the true length.
1561 * N.B. After this call fhp needs an fh_put
1562 */
1563__be32
1564nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1565{
1566        struct inode    *inode;
1567        mm_segment_t    oldfs;
1568        __be32          err;
1569        int             host_err;
1570        struct path path;
1571
1572        err = fh_verify(rqstp, fhp, S_IFLNK, NFSD_MAY_NOP);
1573        if (err)
1574                goto out;
1575
1576        path.mnt = fhp->fh_export->ex_path.mnt;
1577        path.dentry = fhp->fh_dentry;
1578        inode = path.dentry->d_inode;
1579
1580        err = nfserr_inval;
1581        if (!inode->i_op->readlink)
1582                goto out;
1583
1584        touch_atime(&path);
1585        /* N.B. Why does this call need a get_fs()??
1586         * Remove the set_fs and watch the fireworks:-) --okir
1587         */
1588
1589        oldfs = get_fs(); set_fs(KERNEL_DS);
1590        host_err = inode->i_op->readlink(path.dentry, (char __user *)buf, *lenp);
1591        set_fs(oldfs);
1592
1593        if (host_err < 0)
1594                goto out_nfserr;
1595        *lenp = host_err;
1596        err = 0;
1597out:
1598        return err;
1599
1600out_nfserr:
1601        err = nfserrno(host_err);
1602        goto out;
1603}
1604
1605/*
1606 * Create a symlink and look up its inode
1607 * N.B. After this call _both_ fhp and resfhp need an fh_put
1608 */
1609__be32
1610nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1611                                char *fname, int flen,
1612                                char *path,  int plen,
1613                                struct svc_fh *resfhp,
1614                                struct iattr *iap)
1615{
1616        struct dentry   *dentry, *dnew;
1617        __be32          err, cerr;
1618        int             host_err;
1619
1620        err = nfserr_noent;
1621        if (!flen || !plen)
1622                goto out;
1623        err = nfserr_exist;
1624        if (isdotent(fname, flen))
1625                goto out;
1626
1627        err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1628        if (err)
1629                goto out;
1630
1631        host_err = fh_want_write(fhp);
1632        if (host_err)
1633                goto out_nfserr;
1634
1635        fh_lock(fhp);
1636        dentry = fhp->fh_dentry;
1637        dnew = lookup_one_len(fname, dentry, flen);
1638        host_err = PTR_ERR(dnew);
1639        if (IS_ERR(dnew))
1640                goto out_nfserr;
1641
1642        if (unlikely(path[plen] != 0)) {
1643                char *path_alloced = kmalloc(plen+1, GFP_KERNEL);
1644                if (path_alloced == NULL)
1645                        host_err = -ENOMEM;
1646                else {
1647                        strncpy(path_alloced, path, plen);
1648                        path_alloced[plen] = 0;
1649                        host_err = vfs_symlink(dentry->d_inode, dnew, path_alloced);
1650                        kfree(path_alloced);
1651                }
1652        } else
1653                host_err = vfs_symlink(dentry->d_inode, dnew, path);
1654        err = nfserrno(host_err);
1655        if (!err)
1656                err = nfserrno(commit_metadata(fhp));
1657        fh_unlock(fhp);
1658
1659        fh_drop_write(fhp);
1660
1661        cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1662        dput(dnew);
1663        if (err==0) err = cerr;
1664out:
1665        return err;
1666
1667out_nfserr:
1668        err = nfserrno(host_err);
1669        goto out;
1670}
1671
1672/*
1673 * Create a hardlink
1674 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1675 */
1676__be32
1677nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1678                                char *name, int len, struct svc_fh *tfhp)
1679{
1680        struct dentry   *ddir, *dnew, *dold;
1681        struct inode    *dirp;
1682        __be32          err;
1683        int             host_err;
1684
1685        err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_CREATE);
1686        if (err)
1687                goto out;
1688        err = fh_verify(rqstp, tfhp, 0, NFSD_MAY_NOP);
1689        if (err)
1690                goto out;
1691        err = nfserr_isdir;
1692        if (S_ISDIR(tfhp->fh_dentry->d_inode->i_mode))
1693                goto out;
1694        err = nfserr_perm;
1695        if (!len)
1696                goto out;
1697        err = nfserr_exist;
1698        if (isdotent(name, len))
1699                goto out;
1700
1701        host_err = fh_want_write(tfhp);
1702        if (host_err) {
1703                err = nfserrno(host_err);
1704                goto out;
1705        }
1706
1707        fh_lock_nested(ffhp, I_MUTEX_PARENT);
1708        ddir = ffhp->fh_dentry;
1709        dirp = ddir->d_inode;
1710
1711        dnew = lookup_one_len(name, ddir, len);
1712        host_err = PTR_ERR(dnew);
1713        if (IS_ERR(dnew))
1714                goto out_nfserr;
1715
1716        dold = tfhp->fh_dentry;
1717
1718        err = nfserr_noent;
1719        if (!dold->d_inode)
1720                goto out_dput;
1721        host_err = nfsd_break_lease(dold->d_inode);
1722        if (host_err) {
1723                err = nfserrno(host_err);
1724                goto out_dput;
1725        }
1726        host_err = vfs_link(dold, dirp, dnew);
1727        if (!host_err) {
1728                err = nfserrno(commit_metadata(ffhp));
1729                if (!err)
1730                        err = nfserrno(commit_metadata(tfhp));
1731        } else {
1732                if (host_err == -EXDEV && rqstp->rq_vers == 2)
1733                        err = nfserr_acces;
1734                else
1735                        err = nfserrno(host_err);
1736        }
1737out_dput:
1738        dput(dnew);
1739out_unlock:
1740        fh_unlock(ffhp);
1741        fh_drop_write(tfhp);
1742out:
1743        return err;
1744
1745out_nfserr:
1746        err = nfserrno(host_err);
1747        goto out_unlock;
1748}
1749
1750/*
1751 * Rename a file
1752 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1753 */
1754__be32
1755nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1756                            struct svc_fh *tfhp, char *tname, int tlen)
1757{
1758        struct dentry   *fdentry, *tdentry, *odentry, *ndentry, *trap;
1759        struct inode    *fdir, *tdir;
1760        __be32          err;
1761        int             host_err;
1762
1763        err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_REMOVE);
1764        if (err)
1765                goto out;
1766        err = fh_verify(rqstp, tfhp, S_IFDIR, NFSD_MAY_CREATE);
1767        if (err)
1768                goto out;
1769
1770        fdentry = ffhp->fh_dentry;
1771        fdir = fdentry->d_inode;
1772
1773        tdentry = tfhp->fh_dentry;
1774        tdir = tdentry->d_inode;
1775
1776        err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev;
1777        if (ffhp->fh_export != tfhp->fh_export)
1778                goto out;
1779
1780        err = nfserr_perm;
1781        if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1782                goto out;
1783
1784        host_err = fh_want_write(ffhp);
1785        if (host_err) {
1786                err = nfserrno(host_err);
1787                goto out;
1788        }
1789
1790        /* cannot use fh_lock as we need deadlock protective ordering
1791         * so do it by hand */
1792        trap = lock_rename(tdentry, fdentry);
1793        ffhp->fh_locked = tfhp->fh_locked = 1;
1794        fill_pre_wcc(ffhp);
1795        fill_pre_wcc(tfhp);
1796
1797        odentry = lookup_one_len(fname, fdentry, flen);
1798        host_err = PTR_ERR(odentry);
1799        if (IS_ERR(odentry))
1800                goto out_nfserr;
1801
1802        host_err = -ENOENT;
1803        if (!odentry->d_inode)
1804                goto out_dput_old;
1805        host_err = -EINVAL;
1806        if (odentry == trap)
1807                goto out_dput_old;
1808
1809        ndentry = lookup_one_len(tname, tdentry, tlen);
1810        host_err = PTR_ERR(ndentry);
1811        if (IS_ERR(ndentry))
1812                goto out_dput_old;
1813        host_err = -ENOTEMPTY;
1814        if (ndentry == trap)
1815                goto out_dput_new;
1816
1817        host_err = -EXDEV;
1818        if (ffhp->fh_export->ex_path.mnt != tfhp->fh_export->ex_path.mnt)
1819                goto out_dput_new;
1820
1821        host_err = nfsd_break_lease(odentry->d_inode);
1822        if (host_err)
1823                goto out_dput_new;
1824        if (ndentry->d_inode) {
1825                host_err = nfsd_break_lease(ndentry->d_inode);
1826                if (host_err)
1827                        goto out_dput_new;
1828        }
1829        host_err = vfs_rename(fdir, odentry, tdir, ndentry);
1830        if (!host_err) {
1831                host_err = commit_metadata(tfhp);
1832                if (!host_err)
1833                        host_err = commit_metadata(ffhp);
1834        }
1835 out_dput_new:
1836        dput(ndentry);
1837 out_dput_old:
1838        dput(odentry);
1839 out_nfserr:
1840        err = nfserrno(host_err);
1841
1842        /* we cannot reply on fh_unlock on the two filehandles,
1843         * as that would do the wrong thing if the two directories
1844         * were the same, so again we do it by hand
1845         */
1846        fill_post_wcc(ffhp);
1847        fill_post_wcc(tfhp);
1848        unlock_rename(tdentry, fdentry);
1849        ffhp->fh_locked = tfhp->fh_locked = 0;
1850        fh_drop_write(ffhp);
1851
1852out:
1853        return err;
1854}
1855
1856/*
1857 * Unlink a file or directory
1858 * N.B. After this call fhp needs an fh_put
1859 */
1860__be32
1861nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1862                                char *fname, int flen)
1863{
1864        struct dentry   *dentry, *rdentry;
1865        struct inode    *dirp;
1866        __be32          err;
1867        int             host_err;
1868
1869        err = nfserr_acces;
1870        if (!flen || isdotent(fname, flen))
1871                goto out;
1872        err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_REMOVE);
1873        if (err)
1874                goto out;
1875
1876        host_err = fh_want_write(fhp);
1877        if (host_err)
1878                goto out_nfserr;
1879
1880        fh_lock_nested(fhp, I_MUTEX_PARENT);
1881        dentry = fhp->fh_dentry;
1882        dirp = dentry->d_inode;
1883
1884        rdentry = lookup_one_len(fname, dentry, flen);
1885        host_err = PTR_ERR(rdentry);
1886        if (IS_ERR(rdentry))
1887                goto out_nfserr;
1888
1889        if (!rdentry->d_inode) {
1890                dput(rdentry);
1891                err = nfserr_noent;
1892                goto out;
1893        }
1894
1895        if (!type)
1896                type = rdentry->d_inode->i_mode & S_IFMT;
1897
1898        host_err = nfsd_break_lease(rdentry->d_inode);
1899        if (host_err)
1900                goto out_put;
1901        if (type != S_IFDIR)
1902                host_err = vfs_unlink(dirp, rdentry);
1903        else
1904                host_err = vfs_rmdir(dirp, rdentry);
1905        if (!host_err)
1906                host_err = commit_metadata(fhp);
1907out_put:
1908        dput(rdentry);
1909
1910out_nfserr:
1911        err = nfserrno(host_err);
1912out:
1913        return err;
1914}
1915
1916/*
1917 * We do this buffering because we must not call back into the file
1918 * system's ->lookup() method from the filldir callback. That may well
1919 * deadlock a number of file systems.
1920 *
1921 * This is based heavily on the implementation of same in XFS.
1922 */
1923struct buffered_dirent {
1924        u64             ino;
1925        loff_t          offset;
1926        int             namlen;
1927        unsigned int    d_type;
1928        char            name[];
1929};
1930
1931struct readdir_data {
1932        char            *dirent;
1933        size_t          used;
1934        int             full;
1935};
1936
1937static int nfsd_buffered_filldir(void *__buf, const char *name, int namlen,
1938                                 loff_t offset, u64 ino, unsigned int d_type)
1939{
1940        struct readdir_data *buf = __buf;
1941        struct buffered_dirent *de = (void *)(buf->dirent + buf->used);
1942        unsigned int reclen;
1943
1944        reclen = ALIGN(sizeof(struct buffered_dirent) + namlen, sizeof(u64));
1945        if (buf->used + reclen > PAGE_SIZE) {
1946                buf->full = 1;
1947                return -EINVAL;
1948        }
1949
1950        de->namlen = namlen;
1951        de->offset = offset;
1952        de->ino = ino;
1953        de->d_type = d_type;
1954        memcpy(de->name, name, namlen);
1955        buf->used += reclen;
1956
1957        return 0;
1958}
1959
1960static __be32 nfsd_buffered_readdir(struct file *file, filldir_t func,
1961                                    struct readdir_cd *cdp, loff_t *offsetp)
1962{
1963        struct readdir_data buf;
1964        struct buffered_dirent *de;
1965        int host_err;
1966        int size;
1967        loff_t offset;
1968
1969        buf.dirent = (void *)__get_free_page(GFP_KERNEL);
1970        if (!buf.dirent)
1971                return nfserrno(-ENOMEM);
1972
1973        offset = *offsetp;
1974
1975        while (1) {
1976                struct inode *dir_inode = file->f_path.dentry->d_inode;
1977                unsigned int reclen;
1978
1979                cdp->err = nfserr_eof; /* will be cleared on successful read */
1980                buf.used = 0;
1981                buf.full = 0;
1982
1983                host_err = vfs_readdir(file, nfsd_buffered_filldir, &buf);
1984                if (buf.full)
1985                        host_err = 0;
1986
1987                if (host_err < 0)
1988                        break;
1989
1990                size = buf.used;
1991
1992                if (!size)
1993                        break;
1994
1995                /*
1996                 * Various filldir functions may end up calling back into
1997                 * lookup_one_len() and the file system's ->lookup() method.
1998                 * These expect i_mutex to be held, as it would within readdir.
1999                 */
2000                host_err = mutex_lock_killable(&dir_inode->i_mutex);
2001                if (host_err)
2002                        break;
2003
2004                de = (struct buffered_dirent *)buf.dirent;
2005                while (size > 0) {
2006                        offset = de->offset;
2007
2008                        if (func(cdp, de->name, de->namlen, de->offset,
2009                                 de->ino, de->d_type))
2010                                break;
2011
2012                        if (cdp->err != nfs_ok)
2013                                break;
2014
2015                        reclen = ALIGN(sizeof(*de) + de->namlen,
2016                                       sizeof(u64));
2017                        size -= reclen;
2018                        de = (struct buffered_dirent *)((char *)de + reclen);
2019                }
2020                mutex_unlock(&dir_inode->i_mutex);
2021                if (size > 0) /* We bailed out early */
2022                        break;
2023
2024                offset = vfs_llseek(file, 0, SEEK_CUR);
2025        }
2026
2027        free_page((unsigned long)(buf.dirent));
2028
2029        if (host_err)
2030                return nfserrno(host_err);
2031
2032        *offsetp = offset;
2033        return cdp->err;
2034}
2035
2036/*
2037 * Read entries from a directory.
2038 * The  NFSv3/4 verifier we ignore for now.
2039 */
2040__be32
2041nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp, 
2042             struct readdir_cd *cdp, filldir_t func)
2043{
2044        __be32          err;
2045        struct file     *file;
2046        loff_t          offset = *offsetp;
2047        int             may_flags = NFSD_MAY_READ;
2048
2049        /* NFSv2 only supports 32 bit cookies */
2050        if (rqstp->rq_vers > 2)
2051                may_flags |= NFSD_MAY_64BIT_COOKIE;
2052
2053        err = nfsd_open(rqstp, fhp, S_IFDIR, may_flags, &file);
2054        if (err)
2055                goto out;
2056
2057        offset = vfs_llseek(file, offset, SEEK_SET);
2058        if (offset < 0) {
2059                err = nfserrno((int)offset);
2060                goto out_close;
2061        }
2062
2063        err = nfsd_buffered_readdir(file, func, cdp, offsetp);
2064
2065        if (err == nfserr_eof || err == nfserr_toosmall)
2066                err = nfs_ok; /* can still be found in ->err */
2067out_close:
2068        nfsd_close(file);
2069out:
2070        return err;
2071}
2072
2073/*
2074 * Get file system stats
2075 * N.B. After this call fhp needs an fh_put
2076 */
2077__be32
2078nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat, int access)
2079{
2080        __be32 err;
2081
2082        err = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP | access);
2083        if (!err) {
2084                struct path path = {
2085                        .mnt    = fhp->fh_export->ex_path.mnt,
2086                        .dentry = fhp->fh_dentry,
2087                };
2088                if (vfs_statfs(&path, stat))
2089                        err = nfserr_io;
2090        }
2091        return err;
2092}
2093
2094static int exp_rdonly(struct svc_rqst *rqstp, struct svc_export *exp)
2095{
2096        return nfsexp_flags(rqstp, exp) & NFSEXP_READONLY;
2097}
2098
2099/*
2100 * Check for a user's access permissions to this inode.
2101 */
2102__be32
2103nfsd_permission(struct svc_rqst *rqstp, struct svc_export *exp,
2104                                        struct dentry *dentry, int acc)
2105{
2106        struct inode    *inode = dentry->d_inode;
2107        int             err;
2108
2109        if ((acc & NFSD_MAY_MASK) == NFSD_MAY_NOP)
2110                return 0;
2111#if 0
2112        dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
2113                acc,
2114                (acc & NFSD_MAY_READ)?  " read"  : "",
2115                (acc & NFSD_MAY_WRITE)? " write" : "",
2116                (acc & NFSD_MAY_EXEC)?  " exec"  : "",
2117                (acc & NFSD_MAY_SATTR)? " sattr" : "",
2118                (acc & NFSD_MAY_TRUNC)? " trunc" : "",
2119                (acc & NFSD_MAY_LOCK)?  " lock"  : "",
2120                (acc & NFSD_MAY_OWNER_OVERRIDE)? " owneroverride" : "",
2121                inode->i_mode,
2122                IS_IMMUTABLE(inode)?    " immut" : "",
2123                IS_APPEND(inode)?       " append" : "",
2124                __mnt_is_readonly(exp->ex_path.mnt)?    " ro" : "");
2125        dprintk("      owner %d/%d user %d/%d\n",
2126                inode->i_uid, inode->i_gid, current_fsuid(), current_fsgid());
2127#endif
2128
2129        /* Normally we reject any write/sattr etc access on a read-only file
2130         * system.  But if it is IRIX doing check on write-access for a 
2131         * device special file, we ignore rofs.
2132         */
2133        if (!(acc & NFSD_MAY_LOCAL_ACCESS))
2134                if (acc & (NFSD_MAY_WRITE | NFSD_MAY_SATTR | NFSD_MAY_TRUNC)) {
2135                        if (exp_rdonly(rqstp, exp) ||
2136                            __mnt_is_readonly(exp->ex_path.mnt))
2137                                return nfserr_rofs;
2138                        if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode))
2139                                return nfserr_perm;
2140                }
2141        if ((acc & NFSD_MAY_TRUNC) && IS_APPEND(inode))
2142                return nfserr_perm;
2143
2144        if (acc & NFSD_MAY_LOCK) {
2145                /* If we cannot rely on authentication in NLM requests,
2146                 * just allow locks, otherwise require read permission, or
2147                 * ownership
2148                 */
2149                if (exp->ex_flags & NFSEXP_NOAUTHNLM)
2150                        return 0;
2151                else
2152                        acc = NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE;
2153        }
2154        /*
2155         * The file owner always gets access permission for accesses that
2156         * would normally be checked at open time. This is to make
2157         * file access work even when the client has done a fchmod(fd, 0).
2158         *
2159         * However, `cp foo bar' should fail nevertheless when bar is
2160         * readonly. A sensible way to do this might be to reject all
2161         * attempts to truncate a read-only file, because a creat() call
2162         * always implies file truncation.
2163         * ... but this isn't really fair.  A process may reasonably call
2164         * ftruncate on an open file descriptor on a file with perm 000.
2165         * We must trust the client to do permission checking - using "ACCESS"
2166         * with NFSv3.
2167         */
2168        if ((acc & NFSD_MAY_OWNER_OVERRIDE) &&
2169            inode->i_uid == current_fsuid())
2170                return 0;
2171
2172        /* This assumes  NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */
2173        err = inode_permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC));
2174
2175        /* Allow read access to binaries even when mode 111 */
2176        if (err == -EACCES && S_ISREG(inode->i_mode) &&
2177             (acc == (NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE) ||
2178              acc == (NFSD_MAY_READ | NFSD_MAY_READ_IF_EXEC)))
2179                err = inode_permission(inode, MAY_EXEC);
2180
2181        return err? nfserrno(err) : 0;
2182}
2183
2184void
2185nfsd_racache_shutdown(void)
2186{
2187        struct raparms *raparm, *last_raparm;
2188        unsigned int i;
2189
2190        dprintk("nfsd: freeing readahead buffers.\n");
2191
2192        for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2193                raparm = raparm_hash[i].pb_head;
2194                while(raparm) {
2195                        last_raparm = raparm;
2196                        raparm = raparm->p_next;
2197                        kfree(last_raparm);
2198                }
2199                raparm_hash[i].pb_head = NULL;
2200        }
2201}
2202/*
2203 * Initialize readahead param cache
2204 */
2205int
2206nfsd_racache_init(int cache_size)
2207{
2208        int     i;
2209        int     j = 0;
2210        int     nperbucket;
2211        struct raparms **raparm = NULL;
2212
2213
2214        if (raparm_hash[0].pb_head)
2215                return 0;
2216        nperbucket = DIV_ROUND_UP(cache_size, RAPARM_HASH_SIZE);
2217        if (nperbucket < 2)
2218                nperbucket = 2;
2219        cache_size = nperbucket * RAPARM_HASH_SIZE;
2220
2221        dprintk("nfsd: allocating %d readahead buffers.\n", cache_size);
2222
2223        for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2224                spin_lock_init(&raparm_hash[i].pb_lock);
2225
2226                raparm = &raparm_hash[i].pb_head;
2227                for (j = 0; j < nperbucket; j++) {
2228                        *raparm = kzalloc(sizeof(struct raparms), GFP_KERNEL);
2229                        if (!*raparm)
2230                                goto out_nomem;
2231                        raparm = &(*raparm)->p_next;
2232                }
2233                *raparm = NULL;
2234        }
2235
2236        nfsdstats.ra_size = cache_size;
2237        return 0;
2238
2239out_nomem:
2240        dprintk("nfsd: kmalloc failed, freeing readahead buffers\n");
2241        nfsd_racache_shutdown();
2242        return -ENOMEM;
2243}
2244
2245#if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
2246struct posix_acl *
2247nfsd_get_posix_acl(struct svc_fh *fhp, int type)
2248{
2249        struct inode *inode = fhp->fh_dentry->d_inode;
2250        char *name;
2251        void *value = NULL;
2252        ssize_t size;
2253        struct posix_acl *acl;
2254
2255        if (!IS_POSIXACL(inode))
2256                return ERR_PTR(-EOPNOTSUPP);
2257
2258        switch (type) {
2259        case ACL_TYPE_ACCESS:
2260                name = POSIX_ACL_XATTR_ACCESS;
2261                break;
2262        case ACL_TYPE_DEFAULT:
2263                name = POSIX_ACL_XATTR_DEFAULT;
2264                break;
2265        default:
2266                return ERR_PTR(-EOPNOTSUPP);
2267        }
2268
2269        size = nfsd_getxattr(fhp->fh_dentry, name, &value);
2270        if (size < 0)
2271                return ERR_PTR(size);
2272
2273        acl = posix_acl_from_xattr(&init_user_ns, value, size);
2274        kfree(value);
2275        return acl;
2276}
2277
2278int
2279nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl)
2280{
2281        struct inode *inode = fhp->fh_dentry->d_inode;
2282        char *name;
2283        void *value = NULL;
2284        size_t size;
2285        int error;
2286
2287        if (!IS_POSIXACL(inode) ||
2288            !inode->i_op->setxattr || !inode->i_op->removexattr)
2289                return -EOPNOTSUPP;
2290        switch(type) {
2291                case ACL_TYPE_ACCESS:
2292                        name = POSIX_ACL_XATTR_ACCESS;
2293                        break;
2294                case ACL_TYPE_DEFAULT:
2295                        name = POSIX_ACL_XATTR_DEFAULT;
2296                        break;
2297                default:
2298                        return -EOPNOTSUPP;
2299        }
2300
2301        if (acl && acl->a_count) {
2302                size = posix_acl_xattr_size(acl->a_count);
2303                value = kmalloc(size, GFP_KERNEL);
2304                if (!value)
2305                        return -ENOMEM;
2306                error = posix_acl_to_xattr(&init_user_ns, acl, value, size);
2307                if (error < 0)
2308                        goto getout;
2309                size = error;
2310        } else
2311                size = 0;
2312
2313        error = fh_want_write(fhp);
2314        if (error)
2315                goto getout;
2316        if (size)
2317                error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0);
2318        else {
2319                if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT)
2320                        error = 0;
2321                else {
2322                        error = vfs_removexattr(fhp->fh_dentry, name);
2323                        if (error == -ENODATA)
2324                                error = 0;
2325                }
2326        }
2327        fh_drop_write(fhp);
2328
2329getout:
2330        kfree(value);
2331        return error;
2332}
2333#endif  /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
2334
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.