fs/nfsd/vfs.c

   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  */
  54 struct 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
  64 struct 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)
  72 static 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  */
  80 int
  81 nfsd_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);
 125 out:
 126         return err;
 127 }
 128
 129 static 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
 140 static 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  */
 167 int 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
 179 nfsd_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
 228 out_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
 246 nfsd_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;
 269 out:
 270         dput(dentry);
 271         exp_put(exp);
 272         return err;
 273 }
 274
 275 static 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  */
 285 static int
 286 commit_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
 304 nfsd_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         int             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);
 437 out:
 438         return err;
 439
 440 out_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)
 448 static 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)
 469 static int
 470 set_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(pacl, buf, buflen);
 484         if (len < 0) {
 485                 error = len;
 486                 goto out;
 487         }
 488
 489         error = vfs_setxattr(dentry, key, buf, len, 0);
 490 out:
 491         kfree(buf);
 492         return error;
 493 }
 494
 495 __be32
 496 nfsd4_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
 529 out_release:
 530         posix_acl_release(pacl);
 531         posix_acl_release(dpacl);
 532 out_nfserr:
 533         if (host_error == -EOPNOTSUPP)
 534                 return nfserr_attrnotsupp;
 535         else
 536                 return nfserrno(host_error);
 537 }
 538
 539 static 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(buf, buflen);
 553         kfree(buf);
 554         return pacl;
 555 }
 556
 557 int
 558 nfsd4_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 #define NFSD_XATTR_JUNCTION_PREFIX XATTR_TRUSTED_PREFIX "junction."
 598 #define NFSD_XATTR_JUNCTION_TYPE NFSD_XATTR_JUNCTION_PREFIX "type"
 599 int nfsd4_is_junction(struct dentry *dentry)
 600 {
 601         struct inode *inode = dentry->d_inode;
 602
 603         if (inode == NULL)
 604                 return 0;
 605         if (inode->i_mode & S_IXUGO)
 606                 return 0;
 607         if (!(inode->i_mode & S_ISVTX))
 608                 return 0;
 609         if (vfs_getxattr(dentry, NFSD_XATTR_JUNCTION_TYPE, NULL, 0) <= 0)
 610                 return 0;
 611         return 1;
 612 }
 613 #endif /* defined(CONFIG_NFSD_V4) */
 614
 615 #ifdef CONFIG_NFSD_V3
 616 /*
 617  * Check server access rights to a file system object
 618  */
 619 struct accessmap {
 620         u32             access;
 621         int             how;
 622 };
 623 static struct accessmap nfs3_regaccess[] = {
 624     {   NFS3_ACCESS_READ,       NFSD_MAY_READ                   },
 625     {   NFS3_ACCESS_EXECUTE,    NFSD_MAY_EXEC                   },
 626     {   NFS3_ACCESS_MODIFY,     NFSD_MAY_WRITE|NFSD_MAY_TRUNC   },
 627     {   NFS3_ACCESS_EXTEND,     NFSD_MAY_WRITE                  },
 628
 629     {   0,                      0                               }
 630 };
 631
 632 static struct accessmap nfs3_diraccess[] = {
 633     {   NFS3_ACCESS_READ,       NFSD_MAY_READ                   },
 634     {   NFS3_ACCESS_LOOKUP,     NFSD_MAY_EXEC                   },
 635     {   NFS3_ACCESS_MODIFY,     NFSD_MAY_EXEC|NFSD_MAY_WRITE|NFSD_MAY_TRUNC},
 636     {   NFS3_ACCESS_EXTEND,     NFSD_MAY_EXEC|NFSD_MAY_WRITE    },
 637     {   NFS3_ACCESS_DELETE,     NFSD_MAY_REMOVE                 },
 638
 639     {   0,                      0                               }
 640 };
 641
 642 static struct accessmap nfs3_anyaccess[] = {
 643         /* Some clients - Solaris 2.6 at least, make an access call
 644          * to the server to check for access for things like /dev/null
 645          * (which really, the server doesn't care about).  So
 646          * We provide simple access checking for them, looking
 647          * mainly at mode bits, and we make sure to ignore read-only
 648          * filesystem checks
 649          */
 650     {   NFS3_ACCESS_READ,       NFSD_MAY_READ                   },
 651     {   NFS3_ACCESS_EXECUTE,    NFSD_MAY_EXEC                   },
 652     {   NFS3_ACCESS_MODIFY,     NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS    },
 653     {   NFS3_ACCESS_EXTEND,     NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS    },
 654
 655     {   0,                      0                               }
 656 };
 657
 658 __be32
 659 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
 660 {
 661         struct accessmap        *map;
 662         struct svc_export       *export;
 663         struct dentry           *dentry;
 664         u32                     query, result = 0, sresult = 0;
 665         __be32                  error;
 666
 667         error = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP);
 668         if (error)
 669                 goto out;
 670
 671         export = fhp->fh_export;
 672         dentry = fhp->fh_dentry;
 673
 674         if (S_ISREG(dentry->d_inode->i_mode))
 675                 map = nfs3_regaccess;
 676         else if (S_ISDIR(dentry->d_inode->i_mode))
 677                 map = nfs3_diraccess;
 678         else
 679                 map = nfs3_anyaccess;
 680
 681
 682         query = *access;
 683         for  (; map->access; map++) {
 684                 if (map->access & query) {
 685                         __be32 err2;
 686
 687                         sresult |= map->access;
 688
 689                         err2 = nfsd_permission(rqstp, export, dentry, map->how);
 690                         switch (err2) {
 691                         case nfs_ok:
 692                                 result |= map->access;
 693                                 break;
 694
 695                         /* the following error codes just mean the access was not allowed,
 696                          * rather than an error occurred */
 697                         case nfserr_rofs:
 698                         case nfserr_acces:
 699                         case nfserr_perm:
 700                                 /* simply don't "or" in the access bit. */
 701                                 break;
 702                         default:
 703                                 error = err2;
 704                                 goto out;
 705                         }
 706                 }
 707         }
 708         *access = result;
 709         if (supported)
 710                 *supported = sresult;
 711
 712  out:
 713         return error;
 714 }
 715 #endif /* CONFIG_NFSD_V3 */
 716
 717 static int nfsd_open_break_lease(struct inode *inode, int access)
 718 {
 719         unsigned int mode;
 720
 721         if (access & NFSD_MAY_NOT_BREAK_LEASE)
 722                 return 0;
 723         mode = (access & NFSD_MAY_WRITE) ? O_WRONLY : O_RDONLY;
 724         return break_lease(inode, mode | O_NONBLOCK);
 725 }
 726
 727 /*
 728  * Open an existing file or directory.
 729  * The access argument indicates the type of open (read/write/lock)
 730  * N.B. After this call fhp needs an fh_put
 731  */
 732 __be32
 733 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
 734                         int access, struct file **filp)
 735 {
 736         struct dentry   *dentry;
 737         struct inode    *inode;
 738         int             flags = O_RDONLY|O_LARGEFILE;
 739         __be32          err;
 740         int             host_err = 0;
 741
 742         validate_process_creds();
 743
 744         /*
 745          * If we get here, then the client has already done an "open",
 746          * and (hopefully) checked permission - so allow OWNER_OVERRIDE
 747          * in case a chmod has now revoked permission.
 748          */
 749         err = fh_verify(rqstp, fhp, type, access | NFSD_MAY_OWNER_OVERRIDE);
 750         if (err)
 751                 goto out;
 752
 753         dentry = fhp->fh_dentry;
 754         inode = dentry->d_inode;
 755
 756         /* Disallow write access to files with the append-only bit set
 757          * or any access when mandatory locking enabled
 758          */
 759         err = nfserr_perm;
 760         if (IS_APPEND(inode) && (access & NFSD_MAY_WRITE))
 761                 goto out;
 762         /*
 763          * We must ignore files (but only files) which might have mandatory
 764          * locks on them because there is no way to know if the accesser has
 765          * the lock.
 766          */
 767         if (S_ISREG((inode)->i_mode) && mandatory_lock(inode))
 768                 goto out;
 769
 770         if (!inode->i_fop)
 771                 goto out;
 772
 773         host_err = nfsd_open_break_lease(inode, access);
 774         if (host_err) /* NOMEM or WOULDBLOCK */
 775                 goto out_nfserr;
 776
 777         if (access & NFSD_MAY_WRITE) {
 778                 if (access & NFSD_MAY_READ)
 779                         flags = O_RDWR|O_LARGEFILE;
 780                 else
 781                         flags = O_WRONLY|O_LARGEFILE;
 782         }
 783         *filp = dentry_open(dget(dentry), mntget(fhp->fh_export->ex_path.mnt),
 784                             flags, current_cred());
 785         if (IS_ERR(*filp))
 786                 host_err = PTR_ERR(*filp);
 787         else
 788                 host_err = ima_file_check(*filp, access);
 789 out_nfserr:
 790         err = nfserrno(host_err);
 791 out:
 792         validate_process_creds();
 793         return err;
 794 }
 795
 796 /*
 797  * Close a file.
 798  */
 799 void
 800 nfsd_close(struct file *filp)
 801 {
 802         fput(filp);
 803 }
 804
 805 /*
 806  * Obtain the readahead parameters for the file
 807  * specified by (dev, ino).
 808  */
 809
 810 static inline struct raparms *
 811 nfsd_get_raparms(dev_t dev, ino_t ino)
 812 {
 813         struct raparms  *ra, **rap, **frap = NULL;
 814         int depth = 0;
 815         unsigned int hash;
 816         struct raparm_hbucket *rab;
 817
 818         hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK;
 819         rab = &raparm_hash[hash];
 820
 821         spin_lock(&rab->pb_lock);
 822         for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) {
 823                 if (ra->p_ino == ino && ra->p_dev == dev)
 824                         goto found;
 825                 depth++;
 826                 if (ra->p_count == 0)
 827                         frap = rap;
 828         }
 829         depth = nfsdstats.ra_size;
 830         if (!frap) {
 831                 spin_unlock(&rab->pb_lock);
 832                 return NULL;
 833         }
 834         rap = frap;
 835         ra = *frap;
 836         ra->p_dev = dev;
 837         ra->p_ino = ino;
 838         ra->p_set = 0;
 839         ra->p_hindex = hash;
 840 found:
 841         if (rap != &rab->pb_head) {
 842                 *rap = ra->p_next;
 843                 ra->p_next   = rab->pb_head;
 844                 rab->pb_head = ra;
 845         }
 846         ra->p_count++;
 847         nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
 848         spin_unlock(&rab->pb_lock);
 849         return ra;
 850 }
 851
 852 /*
 853  * Grab and keep cached pages associated with a file in the svc_rqst
 854  * so that they can be passed to the network sendmsg/sendpage routines
 855  * directly. They will be released after the sending has completed.
 856  */
 857 static int
 858 nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
 859                   struct splice_desc *sd)
 860 {
 861         struct svc_rqst *rqstp = sd->u.data;
 862         struct page **pp = rqstp->rq_respages + rqstp->rq_resused;
 863         struct page *page = buf->page;
 864         size_t size;
 865
 866         size = sd->len;
 867
 868         if (rqstp->rq_res.page_len == 0) {
 869                 get_page(page);
 870                 put_page(*pp);
 871                 *pp = page;
 872                 rqstp->rq_resused++;
 873                 rqstp->rq_res.page_base = buf->offset;
 874                 rqstp->rq_res.page_len = size;
 875         } else if (page != pp[-1]) {
 876                 get_page(page);
 877                 if (*pp)
 878                         put_page(*pp);
 879                 *pp = page;
 880                 rqstp->rq_resused++;
 881                 rqstp->rq_res.page_len += size;
 882         } else
 883                 rqstp->rq_res.page_len += size;
 884
 885         return size;
 886 }
 887
 888 static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe,
 889                                     struct splice_desc *sd)
 890 {
 891         return __splice_from_pipe(pipe, sd, nfsd_splice_actor);
 892 }
 893
 894 static __be32
 895 nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
 896               loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
 897 {
 898         mm_segment_t    oldfs;
 899         __be32          err;
 900         int             host_err;
 901
 902         err = nfserr_perm;
 903
 904         if (file->f_op->splice_read && rqstp->rq_splice_ok) {
 905                 struct splice_desc sd = {
 906                         .len            = 0,
 907                         .total_len      = *count,
 908                         .pos            = offset,
 909                         .u.data         = rqstp,
 910                 };
 911
 912                 rqstp->rq_resused = 1;
 913                 host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor);
 914         } else {
 915                 oldfs = get_fs();
 916                 set_fs(KERNEL_DS);
 917                 host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
 918                 set_fs(oldfs);
 919         }
 920
 921         if (host_err >= 0) {
 922                 nfsdstats.io_read += host_err;
 923                 *count = host_err;
 924                 err = 0;
 925                 fsnotify_access(file);
 926         } else
 927                 err = nfserrno(host_err);
 928         return err;
 929 }
 930
 931 static void kill_suid(struct dentry *dentry)
 932 {
 933         struct iattr    ia;
 934         ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
 935
 936         mutex_lock(&dentry->d_inode->i_mutex);
 937         notify_change(dentry, &ia);
 938         mutex_unlock(&dentry->d_inode->i_mutex);
 939 }
 940
 941 /*
 942  * Gathered writes: If another process is currently writing to the file,
 943  * there's a high chance this is another nfsd (triggered by a bulk write
 944  * from a client's biod). Rather than syncing the file with each write
 945  * request, we sleep for 10 msec.
 946  *
 947  * I don't know if this roughly approximates C. Juszak's idea of
 948  * gathered writes, but it's a nice and simple solution (IMHO), and it
 949  * seems to work:-)
 950  *
 951  * Note: we do this only in the NFSv2 case, since v3 and higher have a
 952  * better tool (separate unstable writes and commits) for solving this
 953  * problem.
 954  */
 955 static int wait_for_concurrent_writes(struct file *file)
 956 {
 957         struct inode *inode = file->f_path.dentry->d_inode;
 958         static ino_t last_ino;
 959         static dev_t last_dev;
 960         int err = 0;
 961
 962         if (atomic_read(&inode->i_writecount) > 1
 963             || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
 964                 dprintk("nfsd: write defer %d\n", task_pid_nr(current));
 965                 msleep(10);
 966                 dprintk("nfsd: write resume %d\n", task_pid_nr(current));
 967         }
 968
 969         if (inode->i_state & I_DIRTY) {
 970                 dprintk("nfsd: write sync %d\n", task_pid_nr(current));
 971                 err = vfs_fsync(file, 0);
 972         }
 973         last_ino = inode->i_ino;
 974         last_dev = inode->i_sb->s_dev;
 975         return err;
 976 }
 977
 978 static __be32
 979 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
 980                                 loff_t offset, struct kvec *vec, int vlen,
 981                                 unsigned long *cnt, int *stablep)
 982 {
 983         struct svc_export       *exp;
 984         struct dentry           *dentry;
 985         struct inode            *inode;
 986         mm_segment_t            oldfs;
 987         __be32                  err = 0;
 988         int                     host_err;
 989         int                     stable = *stablep;
 990         int                     use_wgather;
 991
 992         dentry = file->f_path.dentry;
 993         inode = dentry->d_inode;
 994         exp   = fhp->fh_export;
 995
 996         /*
 997          * Request sync writes if
 998          *  -   the sync export option has been set, or
 999          *  -   the client requested O_SYNC behavior (NFSv3 feature).
1000          *  -   The file system doesn't support fsync().
1001          * When NFSv2 gathered writes have been configured for this volume,
1002          * flushing the data to disk is handled separately below.
1003          */
1004         use_wgather = (rqstp->rq_vers == 2) && EX_WGATHER(exp);
1005
1006         if (!file->f_op->fsync) {/* COMMIT3 cannot work */
1007                stable = 2;
1008                *stablep = 2; /* FILE_SYNC */
1009         }
1010
1011         if (!EX_ISSYNC(exp))
1012                 stable = 0;
1013         if (stable && !use_wgather) {
1014                 spin_lock(&file->f_lock);
1015                 file->f_flags |= O_SYNC;
1016                 spin_unlock(&file->f_lock);
1017         }
1018
1019         /* Write the data. */
1020         oldfs = get_fs(); set_fs(KERNEL_DS);
1021         host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset);
1022         set_fs(oldfs);
1023         if (host_err < 0)
1024                 goto out_nfserr;
1025         *cnt = host_err;
1026         nfsdstats.io_write += host_err;
1027         fsnotify_modify(file);
1028
1029         /* clear setuid/setgid flag after write */
1030         if (inode->i_mode & (S_ISUID | S_ISGID))
1031                 kill_suid(dentry);
1032
1033         if (stable && use_wgather)
1034                 host_err = wait_for_concurrent_writes(file);
1035
1036 out_nfserr:
1037         dprintk("nfsd: write complete host_err=%d\n", host_err);
1038         if (host_err >= 0)
1039                 err = 0;
1040         else
1041                 err = nfserrno(host_err);
1042         return err;
1043 }
1044
1045 /*
1046  * Read data from a file. count must contain the requested read count
1047  * on entry. On return, *count contains the number of bytes actually read.
1048  * N.B. After this call fhp needs an fh_put
1049  */
1050 __be32 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp,
1051         loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
1052 {
1053         struct file *file;
1054         struct inode *inode;
1055         struct raparms  *ra;
1056         __be32 err;
1057
1058         err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
1059         if (err)
1060                 return err;
1061
1062         inode = file->f_path.dentry->d_inode;
1063
1064         /* Get readahead parameters */
1065         ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
1066
1067         if (ra && ra->p_set)
1068                 file->f_ra = ra->p_ra;
1069
1070         err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1071
1072         /* Write back readahead params */
1073         if (ra) {
1074                 struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex];
1075                 spin_lock(&rab->pb_lock);
1076                 ra->p_ra = file->f_ra;
1077                 ra->p_set = 1;
1078                 ra->p_count--;
1079                 spin_unlock(&rab->pb_lock);
1080         }
1081
1082         nfsd_close(file);
1083         return err;
1084 }
1085
1086 /* As above, but use the provided file descriptor. */
1087 __be32
1088 nfsd_read_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1089                 loff_t offset, struct kvec *vec, int vlen,
1090                 unsigned long *count)
1091 {
1092         __be32          err;
1093
1094         if (file) {
1095                 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1096                                 NFSD_MAY_READ|NFSD_MAY_OWNER_OVERRIDE);
1097                 if (err)
1098                         goto out;
1099                 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1100         } else /* Note file may still be NULL in NFSv4 special stateid case: */
1101                 err = nfsd_read(rqstp, fhp, offset, vec, vlen, count);
1102 out:
1103         return err;
1104 }
1105
1106 /*
1107  * Write data to a file.
1108  * The stable flag requests synchronous writes.
1109  * N.B. After this call fhp needs an fh_put
1110  */
1111 __be32
1112 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1113                 loff_t offset, struct kvec *vec, int vlen, unsigned long *cnt,
1114                 int *stablep)
1115 {
1116         __be32                  err = 0;
1117
1118         if (file) {
1119                 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1120                                 NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE);
1121                 if (err)
1122                         goto out;
1123                 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1124                                 stablep);
1125         } else {
1126                 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file);
1127                 if (err)
1128                         goto out;
1129
1130                 if (cnt)
1131                         err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1132                                              cnt, stablep);
1133                 nfsd_close(file);
1134         }
1135 out:
1136         return err;
1137 }
1138
1139 #ifdef CONFIG_NFSD_V3
1140 /*
1141  * Commit all pending writes to stable storage.
1142  *
1143  * Note: we only guarantee that data that lies within the range specified
1144  * by the 'offset' and 'count' parameters will be synced.
1145  *
1146  * Unfortunately we cannot lock the file to make sure we return full WCC
1147  * data to the client, as locking happens lower down in the filesystem.
1148  */
1149 __be32
1150 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1151                loff_t offset, unsigned long count)
1152 {
1153         struct file     *file;
1154         loff_t          end = LLONG_MAX;
1155         __be32          err = nfserr_inval;
1156
1157         if (offset < 0)
1158                 goto out;
1159         if (count != 0) {
1160                 end = offset + (loff_t)count - 1;
1161                 if (end < offset)
1162                         goto out;
1163         }
1164
1165         err = nfsd_open(rqstp, fhp, S_IFREG,
1166                         NFSD_MAY_WRITE|NFSD_MAY_NOT_BREAK_LEASE, &file);
1167         if (err)
1168                 goto out;
1169         if (EX_ISSYNC(fhp->fh_export)) {
1170                 int err2 = vfs_fsync_range(file, offset, end, 0);
1171
1172                 if (err2 != -EINVAL)
1173                         err = nfserrno(err2);
1174                 else
1175                         err = nfserr_notsupp;
1176         }
1177
1178         nfsd_close(file);
1179 out:
1180         return err;
1181 }
1182 #endif /* CONFIG_NFSD_V3 */
1183
1184 static __be32
1185 nfsd_create_setattr(struct svc_rqst *rqstp, struct svc_fh *resfhp,
1186                         struct iattr *iap)
1187 {
1188         /*
1189          * Mode has already been set earlier in create:
1190          */
1191         iap->ia_valid &= ~ATTR_MODE;
1192         /*
1193          * Setting uid/gid works only for root.  Irix appears to
1194          * send along the gid on create when it tries to implement
1195          * setgid directories via NFS:
1196          */
1197         if (current_fsuid() != 0)
1198                 iap->ia_valid &= ~(ATTR_UID|ATTR_GID);
1199         if (iap->ia_valid)
1200                 return nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1201         return 0;
1202 }
1203
1204 /* HPUX client sometimes creates a file in mode 000, and sets size to 0.
1205  * setting size to 0 may fail for some specific file systems by the permission
1206  * checking which requires WRITE permission but the mode is 000.
1207  * we ignore the resizing(to 0) on the just new created file, since the size is
1208  * 0 after file created.
1209  *
1210  * call this only after vfs_create() is called.
1211  * */
1212 static void
1213 nfsd_check_ignore_resizing(struct iattr *iap)
1214 {
1215         if ((iap->ia_valid & ATTR_SIZE) && (iap->ia_size == 0))
1216                 iap->ia_valid &= ~ATTR_SIZE;
1217 }
1218
1219 /*
1220  * Create a file (regular, directory, device, fifo); UNIX sockets
1221  * not yet implemented.
1222  * If the response fh has been verified, the parent directory should
1223  * already be locked. Note that the parent directory is left locked.
1224  *
1225  * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1226  */
1227 __be32
1228 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1229                 char *fname, int flen, struct iattr *iap,
1230                 int type, dev_t rdev, struct svc_fh *resfhp)
1231 {
1232         struct dentry   *dentry, *dchild = NULL;
1233         struct inode    *dirp;
1234         __be32          err;
1235         __be32          err2;
1236         int             host_err;
1237
1238         err = nfserr_perm;
1239         if (!flen)
1240                 goto out;
1241         err = nfserr_exist;
1242         if (isdotent(fname, flen))
1243                 goto out;
1244
1245         err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1246         if (err)
1247                 goto out;
1248
1249         dentry = fhp->fh_dentry;
1250         dirp = dentry->d_inode;
1251
1252         err = nfserr_notdir;
1253         if (!dirp->i_op->lookup)
1254                 goto out;
1255         /*
1256          * Check whether the response file handle has been verified yet.
1257          * If it has, the parent directory should already be locked.
1258          */
1259         if (!resfhp->fh_dentry) {
1260                 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1261                 fh_lock_nested(fhp, I_MUTEX_PARENT);
1262                 dchild = lookup_one_len(fname, dentry, flen);
1263                 host_err = PTR_ERR(dchild);
1264                 if (IS_ERR(dchild))
1265                         goto out_nfserr;
1266                 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1267                 if (err)
1268                         goto out;
1269         } else {
1270                 /* called from nfsd_proc_create */
1271                 dchild = dget(resfhp->fh_dentry);
1272                 if (!fhp->fh_locked) {
1273                         /* not actually possible */
1274                         printk(KERN_ERR
1275                                 "nfsd_create: parent %s/%s not locked!\n",
1276                                 dentry->d_parent->d_name.name,
1277                                 dentry->d_name.name);
1278                         err = nfserr_io;
1279                         goto out;
1280                 }
1281         }
1282         /*
1283          * Make sure the child dentry is still negative ...
1284          */
1285         err = nfserr_exist;
1286         if (dchild->d_inode) {
1287                 dprintk("nfsd_create: dentry %s/%s not negative!\n",
1288                         dentry->d_name.name, dchild->d_name.name);
1289                 goto out;
1290         }
1291
1292         if (!(iap->ia_valid & ATTR_MODE))
1293                 iap->ia_mode = 0;
1294         iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1295
1296         err = nfserr_inval;
1297         if (!S_ISREG(type) && !S_ISDIR(type) && !special_file(type)) {
1298                 printk(KERN_WARNING "nfsd: bad file type %o in nfsd_create\n",
1299                        type);
1300                 goto out;
1301         }
1302
1303         host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1304         if (host_err)
1305                 goto out_nfserr;
1306
1307         /*
1308          * Get the dir op function pointer.
1309          */
1310         err = 0;
1311         switch (type) {
1312         case S_IFREG:
1313                 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1314                 if (!host_err)
1315                         nfsd_check_ignore_resizing(iap);
1316                 break;
1317         case S_IFDIR:
1318                 host_err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1319                 break;
1320         case S_IFCHR:
1321         case S_IFBLK:
1322         case S_IFIFO:
1323         case S_IFSOCK:
1324                 host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1325                 break;
1326         }
1327         if (host_err < 0) {
1328                 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1329                 goto out_nfserr;
1330         }
1331
1332         err = nfsd_create_setattr(rqstp, resfhp, iap);
1333
1334         /*
1335          * nfsd_setattr already committed the child.  Transactional filesystems
1336          * had a chance to commit changes for both parent and child
1337          * simultaneously making the following commit_metadata a noop.
1338          */
1339         err2 = nfserrno(commit_metadata(fhp));
1340         if (err2)
1341                 err = err2;
1342         mnt_drop_write(fhp->fh_export->ex_path.mnt);
1343         /*
1344          * Update the file handle to get the new inode info.
1345          */
1346         if (!err)
1347                 err = fh_update(resfhp);
1348 out:
1349         if (dchild && !IS_ERR(dchild))
1350                 dput(dchild);
1351         return err;
1352
1353 out_nfserr:
1354         err = nfserrno(host_err);
1355         goto out;
1356 }
1357
1358 #ifdef CONFIG_NFSD_V3
1359
1360 static inline int nfsd_create_is_exclusive(int createmode)
1361 {
1362         return createmode == NFS3_CREATE_EXCLUSIVE
1363                || createmode == NFS4_CREATE_EXCLUSIVE4_1;
1364 }
1365
1366 /*
1367  * NFSv3 and NFSv4 version of nfsd_create
1368  */
1369 __be32
1370 do_nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1371                 char *fname, int flen, struct iattr *iap,
1372                 struct svc_fh *resfhp, int createmode, u32 *verifier,
1373                 bool *truncp, bool *created)
1374 {
1375         struct dentry   *dentry, *dchild = NULL;
1376         struct inode    *dirp;
1377         __be32          err;
1378         int             host_err;
1379         __u32           v_mtime=0, v_atime=0;
1380
1381         err = nfserr_perm;
1382         if (!flen)
1383                 goto out;
1384         err = nfserr_exist;
1385         if (isdotent(fname, flen))
1386                 goto out;
1387         if (!(iap->ia_valid & ATTR_MODE))
1388                 iap->ia_mode = 0;
1389         err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC);
1390         if (err)
1391                 goto out;
1392
1393         dentry = fhp->fh_dentry;
1394         dirp = dentry->d_inode;
1395
1396         /* Get all the sanity checks out of the way before
1397          * we lock the parent. */
1398         err = nfserr_notdir;
1399         if (!dirp->i_op->lookup)
1400                 goto out;
1401         fh_lock_nested(fhp, I_MUTEX_PARENT);
1402
1403         /*
1404          * Compose the response file handle.
1405          */
1406         dchild = lookup_one_len(fname, dentry, flen);
1407         host_err = PTR_ERR(dchild);
1408         if (IS_ERR(dchild))
1409                 goto out_nfserr;
1410
1411         /* If file doesn't exist, check for permissions to create one */
1412         if (!dchild->d_inode) {
1413                 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1414                 if (err)
1415                         goto out;
1416         }
1417
1418         err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1419         if (err)
1420                 goto out;
1421
1422         if (nfsd_create_is_exclusive(createmode)) {
1423                 /* solaris7 gets confused (bugid 4218508) if these have
1424                  * the high bit set, so just clear the high bits. If this is
1425                  * ever changed to use different attrs for storing the
1426                  * verifier, then do_open_lookup() will also need to be fixed
1427                  * accordingly.
1428                  */
1429                 v_mtime = verifier[0]&0x7fffffff;
1430                 v_atime = verifier[1]&0x7fffffff;
1431         }
1432
1433         host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1434         if (host_err)
1435                 goto out_nfserr;
1436         if (dchild->d_inode) {
1437                 err = 0;
1438
1439                 switch (createmode) {
1440                 case NFS3_CREATE_UNCHECKED:
1441                         if (! S_ISREG(dchild->d_inode->i_mode))
1442                                 err = nfserr_exist;
1443                         else if (truncp) {
1444                                 /* in nfsv4, we need to treat this case a little
1445                                  * differently.  we don't want to truncate the
1446                                  * file now; this would be wrong if the OPEN
1447                                  * fails for some other reason.  furthermore,
1448                                  * if the size is nonzero, we should ignore it
1449                                  * according to spec!
1450                                  */
1451                                 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1452                         }
1453                         else {
1454                                 iap->ia_valid &= ATTR_SIZE;
1455                                 goto set_attr;
1456                         }
1457                         break;
1458                 case NFS3_CREATE_EXCLUSIVE:
1459                         if (   dchild->d_inode->i_mtime.tv_sec == v_mtime
1460                             && dchild->d_inode->i_atime.tv_sec == v_atime
1461                             && dchild->d_inode->i_size  == 0 )
1462                                 break;
1463                 case NFS4_CREATE_EXCLUSIVE4_1:
1464                         if (   dchild->d_inode->i_mtime.tv_sec == v_mtime
1465                             && dchild->d_inode->i_atime.tv_sec == v_atime
1466                             && dchild->d_inode->i_size  == 0 )
1467                                 goto set_attr;
1468                          /* fallthru */
1469                 case NFS3_CREATE_GUARDED:
1470                         err = nfserr_exist;
1471                 }
1472                 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1473                 goto out;
1474         }
1475
1476         host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1477         if (host_err < 0) {
1478                 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1479                 goto out_nfserr;
1480         }
1481         if (created)
1482                 *created = 1;
1483
1484         nfsd_check_ignore_resizing(iap);
1485
1486         if (nfsd_create_is_exclusive(createmode)) {
1487                 /* Cram the verifier into atime/mtime */
1488                 iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1489                         | ATTR_MTIME_SET|ATTR_ATIME_SET;
1490                 /* XXX someone who knows this better please fix it for nsec */
1491                 iap->ia_mtime.tv_sec = v_mtime;
1492                 iap->ia_atime.tv_sec = v_atime;
1493                 iap->ia_mtime.tv_nsec = 0;
1494                 iap->ia_atime.tv_nsec = 0;
1495         }
1496
1497  set_attr:
1498         err = nfsd_create_setattr(rqstp, resfhp, iap);
1499
1500         /*
1501          * nfsd_setattr already committed the child (and possibly also the parent).
1502          */
1503         if (!err)
1504                 err = nfserrno(commit_metadata(fhp));
1505
1506         mnt_drop_write(fhp->fh_export->ex_path.mnt);
1507         /*
1508          * Update the filehandle to get the new inode info.
1509          */
1510         if (!err)
1511                 err = fh_update(resfhp);
1512
1513  out:
1514         fh_unlock(fhp);
1515         if (dchild && !IS_ERR(dchild))
1516                 dput(dchild);
1517         return err;
1518
1519  out_nfserr:
1520         err = nfserrno(host_err);
1521         goto out;
1522 }
1523 #endif /* CONFIG_NFSD_V3 */
1524
1525 /*
1526  * Read a symlink. On entry, *lenp must contain the maximum path length that
1527  * fits into the buffer. On return, it contains the true length.
1528  * N.B. After this call fhp needs an fh_put
1529  */
1530 __be32
1531 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1532 {
1533         struct dentry   *dentry;
1534         struct inode    *inode;
1535         mm_segment_t    oldfs;
1536         __be32          err;
1537         int             host_err;
1538
1539         err = fh_verify(rqstp, fhp, S_IFLNK, NFSD_MAY_NOP);
1540         if (err)
1541                 goto out;
1542
1543         dentry = fhp->fh_dentry;
1544         inode = dentry->d_inode;
1545
1546         err = nfserr_inval;
1547         if (!inode->i_op->readlink)
1548                 goto out;
1549
1550         touch_atime(fhp->fh_export->ex_path.mnt, dentry);
1551         /* N.B. Why does this call need a get_fs()??
1552          * Remove the set_fs and watch the fireworks:-) --okir
1553          */
1554
1555         oldfs = get_fs(); set_fs(KERNEL_DS);
1556         host_err = inode->i_op->readlink(dentry, buf, *lenp);
1557         set_fs(oldfs);
1558
1559         if (host_err < 0)
1560                 goto out_nfserr;
1561         *lenp = host_err;
1562         err = 0;
1563 out:
1564         return err;
1565
1566 out_nfserr:
1567         err = nfserrno(host_err);
1568         goto out;
1569 }
1570
1571 /*
1572  * Create a symlink and look up its inode
1573  * N.B. After this call _both_ fhp and resfhp need an fh_put
1574  */
1575 __be32
1576 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1577                                 char *fname, int flen,
1578                                 char *path,  int plen,
1579                                 struct svc_fh *resfhp,
1580                                 struct iattr *iap)
1581 {
1582         struct dentry   *dentry, *dnew;
1583         __be32          err, cerr;
1584         int             host_err;
1585
1586         err = nfserr_noent;
1587         if (!flen || !plen)
1588                 goto out;
1589         err = nfserr_exist;
1590         if (isdotent(fname, flen))
1591                 goto out;
1592
1593         err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1594         if (err)
1595                 goto out;
1596         fh_lock(fhp);
1597         dentry = fhp->fh_dentry;
1598         dnew = lookup_one_len(fname, dentry, flen);
1599         host_err = PTR_ERR(dnew);
1600         if (IS_ERR(dnew))
1601                 goto out_nfserr;
1602
1603         host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1604         if (host_err)
1605                 goto out_nfserr;
1606
1607         if (unlikely(path[plen] != 0)) {
1608                 char *path_alloced = kmalloc(plen+1, GFP_KERNEL);
1609                 if (path_alloced == NULL)
1610                         host_err = -ENOMEM;
1611                 else {
1612                         strncpy(path_alloced, path, plen);
1613                         path_alloced[plen] = 0;
1614                         host_err = vfs_symlink(dentry->d_inode, dnew, path_alloced);
1615                         kfree(path_alloced);
1616                 }
1617         } else
1618                 host_err = vfs_symlink(dentry->d_inode, dnew, path);
1619         err = nfserrno(host_err);
1620         if (!err)
1621                 err = nfserrno(commit_metadata(fhp));
1622         fh_unlock(fhp);
1623
1624         mnt_drop_write(fhp->fh_export->ex_path.mnt);
1625
1626         cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1627         dput(dnew);
1628         if (err==0) err = cerr;
1629 out:
1630         return err;
1631
1632 out_nfserr:
1633         err = nfserrno(host_err);
1634         goto out;
1635 }
1636
1637 /*
1638  * Create a hardlink
1639  * N.B. After this call _both_ ffhp and tfhp need an fh_put
1640  */
1641 __be32
1642 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1643                                 char *name, int len, struct svc_fh *tfhp)
1644 {
1645         struct dentry   *ddir, *dnew, *dold;
1646         struct inode    *dirp;
1647         __be32          err;
1648         int             host_err;
1649
1650         err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_CREATE);
1651         if (err)
1652                 goto out;
1653         err = fh_verify(rqstp, tfhp, 0, NFSD_MAY_NOP);
1654         if (err)
1655                 goto out;
1656         err = nfserr_isdir;
1657         if (S_ISDIR(tfhp->fh_dentry->d_inode->i_mode))
1658                 goto out;
1659         err = nfserr_perm;
1660         if (!len)
1661                 goto out;
1662         err = nfserr_exist;
1663         if (isdotent(name, len))
1664                 goto out;
1665
1666         fh_lock_nested(ffhp, I_MUTEX_PARENT);
1667         ddir = ffhp->fh_dentry;
1668         dirp = ddir->d_inode;
1669
1670         dnew = lookup_one_len(name, ddir, len);
1671         host_err = PTR_ERR(dnew);
1672         if (IS_ERR(dnew))
1673                 goto out_nfserr;
1674
1675         dold = tfhp->fh_dentry;
1676
1677         host_err = mnt_want_write(tfhp->fh_export->ex_path.mnt);
1678         if (host_err) {
1679                 err = nfserrno(host_err);
1680                 goto out_dput;
1681         }
1682         err = nfserr_noent;
1683         if (!dold->d_inode)
1684                 goto out_drop_write;
1685         host_err = nfsd_break_lease(dold->d_inode);
1686         if (host_err) {
1687                 err = nfserrno(host_err);
1688                 goto out_drop_write;
1689         }
1690         host_err = vfs_link(dold, dirp, dnew);
1691         if (!host_err) {
1692                 err = nfserrno(commit_metadata(ffhp));
1693                 if (!err)
1694                         err = nfserrno(commit_metadata(tfhp));
1695         } else {
1696                 if (host_err == -EXDEV && rqstp->rq_vers == 2)
1697                         err = nfserr_acces;
1698                 else
1699                         err = nfserrno(host_err);
1700         }
1701 out_drop_write:
1702         mnt_drop_write(tfhp->fh_export->ex_path.mnt);
1703 out_dput:
1704         dput(dnew);
1705 out_unlock:
1706         fh_unlock(ffhp);
1707 out:
1708         return err;
1709
1710 out_nfserr:
1711         err = nfserrno(host_err);
1712         goto out_unlock;
1713 }
1714
1715 /*
1716  * Rename a file
1717  * N.B. After this call _both_ ffhp and tfhp need an fh_put
1718  */
1719 __be32
1720 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1721                             struct svc_fh *tfhp, char *tname, int tlen)
1722 {
1723         struct dentry   *fdentry, *tdentry, *odentry, *ndentry, *trap;
1724         struct inode    *fdir, *tdir;
1725         __be32          err;
1726         int             host_err;
1727
1728         err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_REMOVE);
1729         if (err)
1730                 goto out;
1731         err = fh_verify(rqstp, tfhp, S_IFDIR, NFSD_MAY_CREATE);
1732         if (err)
1733                 goto out;
1734
1735         fdentry = ffhp->fh_dentry;
1736         fdir = fdentry->d_inode;
1737
1738         tdentry = tfhp->fh_dentry;
1739         tdir = tdentry->d_inode;
1740
1741         err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev;
1742         if (ffhp->fh_export != tfhp->fh_export)
1743                 goto out;
1744
1745         err = nfserr_perm;
1746         if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1747                 goto out;
1748
1749         /* cannot use fh_lock as we need deadlock protective ordering
1750          * so do it by hand */
1751         trap = lock_rename(tdentry, fdentry);
1752         ffhp->fh_locked = tfhp->fh_locked = 1;
1753         fill_pre_wcc(ffhp);
1754         fill_pre_wcc(tfhp);
1755
1756         odentry = lookup_one_len(fname, fdentry, flen);
1757         host_err = PTR_ERR(odentry);
1758         if (IS_ERR(odentry))
1759                 goto out_nfserr;
1760
1761         host_err = -ENOENT;
1762         if (!odentry->d_inode)
1763                 goto out_dput_old;
1764         host_err = -EINVAL;
1765         if (odentry == trap)
1766                 goto out_dput_old;
1767
1768         ndentry = lookup_one_len(tname, tdentry, tlen);
1769         host_err = PTR_ERR(ndentry);
1770         if (IS_ERR(ndentry))
1771                 goto out_dput_old;
1772         host_err = -ENOTEMPTY;
1773         if (ndentry == trap)
1774                 goto out_dput_new;
1775
1776         host_err = -EXDEV;
1777         if (ffhp->fh_export->ex_path.mnt != tfhp->fh_export->ex_path.mnt)
1778                 goto out_dput_new;
1779         host_err = mnt_want_write(ffhp->fh_export->ex_path.mnt);
1780         if (host_err)
1781                 goto out_dput_new;
1782
1783         host_err = nfsd_break_lease(odentry->d_inode);
1784         if (host_err)
1785                 goto out_drop_write;
1786         if (ndentry->d_inode) {
1787                 host_err = nfsd_break_lease(ndentry->d_inode);
1788                 if (host_err)
1789                         goto out_drop_write;
1790         }
1791         host_err = vfs_rename(fdir, odentry, tdir, ndentry);
1792         if (!host_err) {
1793                 host_err = commit_metadata(tfhp);
1794                 if (!host_err)
1795                         host_err = commit_metadata(ffhp);
1796         }
1797 out_drop_write:
1798         mnt_drop_write(ffhp->fh_export->ex_path.mnt);
1799  out_dput_new:
1800         dput(ndentry);
1801  out_dput_old:
1802         dput(odentry);
1803  out_nfserr:
1804         err = nfserrno(host_err);
1805
1806         /* we cannot reply on fh_unlock on the two filehandles,
1807          * as that would do the wrong thing if the two directories
1808          * were the same, so again we do it by hand
1809          */
1810         fill_post_wcc(ffhp);
1811         fill_post_wcc(tfhp);
1812         unlock_rename(tdentry, fdentry);
1813         ffhp->fh_locked = tfhp->fh_locked = 0;
1814
1815 out:
1816         return err;
1817 }
1818
1819 /*
1820  * Unlink a file or directory
1821  * N.B. After this call fhp needs an fh_put
1822  */
1823 __be32
1824 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1825                                 char *fname, int flen)
1826 {
1827         struct dentry   *dentry, *rdentry;
1828         struct inode    *dirp;
1829         __be32          err;
1830         int             host_err;
1831
1832         err = nfserr_acces;
1833         if (!flen || isdotent(fname, flen))
1834                 goto out;
1835         err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_REMOVE);
1836         if (err)
1837                 goto out;
1838
1839         fh_lock_nested(fhp, I_MUTEX_PARENT);
1840         dentry = fhp->fh_dentry;
1841         dirp = dentry->d_inode;
1842
1843         rdentry = lookup_one_len(fname, dentry, flen);
1844         host_err = PTR_ERR(rdentry);
1845         if (IS_ERR(rdentry))
1846                 goto out_nfserr;
1847
1848         if (!rdentry->d_inode) {
1849                 dput(rdentry);
1850                 err = nfserr_noent;
1851                 goto out;
1852         }
1853
1854         if (!type)
1855                 type = rdentry->d_inode->i_mode & S_IFMT;
1856
1857         host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1858         if (host_err)
1859                 goto out_put;
1860
1861         host_err = nfsd_break_lease(rdentry->d_inode);
1862         if (host_err)
1863                 goto out_drop_write;
1864         if (type != S_IFDIR)
1865                 host_err = vfs_unlink(dirp, rdentry);
1866         else
1867                 host_err = vfs_rmdir(dirp, rdentry);
1868         if (!host_err)
1869                 host_err = commit_metadata(fhp);
1870 out_drop_write:
1871         mnt_drop_write(fhp->fh_export->ex_path.mnt);
1872 out_put:
1873         dput(rdentry);
1874
1875 out_nfserr:
1876         err = nfserrno(host_err);
1877 out:
1878         return err;
1879 }
1880
1881 /*
1882  * We do this buffering because we must not call back into the file
1883  * system's ->lookup() method from the filldir callback. That may well
1884  * deadlock a number of file systems.
1885  *
1886  * This is based heavily on the implementation of same in XFS.
1887  */
1888 struct buffered_dirent {
1889         u64             ino;
1890         loff_t          offset;
1891         int             namlen;
1892         unsigned int    d_type;
1893         char            name[];
1894 };
1895
1896 struct readdir_data {
1897         char            *dirent;
1898         size_t          used;
1899         int             full;
1900 };
1901
1902 static int nfsd_buffered_filldir(void *__buf, const char *name, int namlen,
1903                                  loff_t offset, u64 ino, unsigned int d_type)
1904 {
1905         struct readdir_data *buf = __buf;
1906         struct buffered_dirent *de = (void *)(buf->dirent + buf->used);
1907         unsigned int reclen;
1908
1909         reclen = ALIGN(sizeof(struct buffered_dirent) + namlen, sizeof(u64));
1910         if (buf->used + reclen > PAGE_SIZE) {
1911                 buf->full = 1;
1912                 return -EINVAL;
1913         }
1914
1915         de->namlen = namlen;
1916         de->offset = offset;
1917         de->ino = ino;
1918         de->d_type = d_type;
1919         memcpy(de->name, name, namlen);
1920         buf->used += reclen;
1921
1922         return 0;
1923 }
1924
1925 static __be32 nfsd_buffered_readdir(struct file *file, filldir_t func,
1926                                     struct readdir_cd *cdp, loff_t *offsetp)
1927 {
1928         struct readdir_data buf;
1929         struct buffered_dirent *de;
1930         int host_err;
1931         int size;
1932         loff_t offset;
1933
1934         buf.dirent = (void *)__get_free_page(GFP_KERNEL);
1935         if (!buf.dirent)
1936                 return nfserrno(-ENOMEM);
1937
1938         offset = *offsetp;
1939
1940         while (1) {
1941                 struct inode *dir_inode = file->f_path.dentry->d_inode;
1942                 unsigned int reclen;
1943
1944                 cdp->err = nfserr_eof; /* will be cleared on successful read */
1945                 buf.used = 0;
1946                 buf.full = 0;
1947
1948                 host_err = vfs_readdir(file, nfsd_buffered_filldir, &buf);
1949                 if (buf.full)
1950                         host_err = 0;
1951
1952                 if (host_err < 0)
1953                         break;
1954
1955                 size = buf.used;
1956
1957                 if (!size)
1958                         break;
1959
1960                 /*
1961                  * Various filldir functions may end up calling back into
1962                  * lookup_one_len() and the file system's ->lookup() method.
1963                  * These expect i_mutex to be held, as it would within readdir.
1964                  */
1965                 host_err = mutex_lock_killable(&dir_inode->i_mutex);
1966                 if (host_err)
1967                         break;
1968
1969                 de = (struct buffered_dirent *)buf.dirent;
1970                 while (size > 0) {
1971                         offset = de->offset;
1972
1973                         if (func(cdp, de->name, de->namlen, de->offset,
1974                                  de->ino, de->d_type))
1975                                 break;
1976
1977                         if (cdp->err != nfs_ok)
1978                                 break;
1979
1980                         reclen = ALIGN(sizeof(*de) + de->namlen,
1981                                        sizeof(u64));
1982                         size -= reclen;
1983                         de = (struct buffered_dirent *)((char *)de + reclen);
1984                 }
1985                 mutex_unlock(&dir_inode->i_mutex);
1986                 if (size > 0) /* We bailed out early */
1987                         break;
1988
1989                 offset = vfs_llseek(file, 0, SEEK_CUR);
1990         }
1991
1992         free_page((unsigned long)(buf.dirent));
1993
1994         if (host_err)
1995                 return nfserrno(host_err);
1996
1997         *offsetp = offset;
1998         return cdp->err;
1999 }
2000
2001 /*
2002  * Read entries from a directory.
2003  * The  NFSv3/4 verifier we ignore for now.
2004  */
2005 __be32
2006 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
2007              struct readdir_cd *cdp, filldir_t func)
2008 {
2009         __be32          err;
2010         struct file     *file;
2011         loff_t          offset = *offsetp;
2012
2013         err = nfsd_open(rqstp, fhp, S_IFDIR, NFSD_MAY_READ, &file);
2014         if (err)
2015                 goto out;
2016
2017         offset = vfs_llseek(file, offset, 0);
2018         if (offset < 0) {
2019                 err = nfserrno((int)offset);
2020                 goto out_close;
2021         }
2022
2023         err = nfsd_buffered_readdir(file, func, cdp, offsetp);
2024
2025         if (err == nfserr_eof || err == nfserr_toosmall)
2026                 err = nfs_ok; /* can still be found in ->err */
2027 out_close:
2028         nfsd_close(file);
2029 out:
2030         return err;
2031 }
2032
2033 /*
2034  * Get file system stats
2035  * N.B. After this call fhp needs an fh_put
2036  */
2037 __be32
2038 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat, int access)
2039 {
2040         __be32 err;
2041
2042         err = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP | access);
2043         if (!err) {
2044                 struct path path = {
2045                         .mnt    = fhp->fh_export->ex_path.mnt,
2046                         .dentry = fhp->fh_dentry,
2047                 };
2048                 if (vfs_statfs(&path, stat))
2049                         err = nfserr_io;
2050         }
2051         return err;
2052 }
2053
2054 static int exp_rdonly(struct svc_rqst *rqstp, struct svc_export *exp)
2055 {
2056         return nfsexp_flags(rqstp, exp) & NFSEXP_READONLY;
2057 }
2058
2059 /*
2060  * Check for a user's access permissions to this inode.
2061  */
2062 __be32
2063 nfsd_permission(struct svc_rqst *rqstp, struct svc_export *exp,
2064                                         struct dentry *dentry, int acc)
2065 {
2066         struct inode    *inode = dentry->d_inode;
2067         int             err;
2068
2069         if ((acc & NFSD_MAY_MASK) == NFSD_MAY_NOP)
2070                 return 0;
2071 #if 0
2072         dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
2073                 acc,
2074                 (acc & NFSD_MAY_READ)?  " read"  : "",
2075                 (acc & NFSD_MAY_WRITE)? " write" : "",
2076                 (acc & NFSD_MAY_EXEC)?  " exec"  : "",
2077                 (acc & NFSD_MAY_SATTR)? " sattr" : "",
2078                 (acc & NFSD_MAY_TRUNC)? " trunc" : "",
2079                 (acc & NFSD_MAY_LOCK)?  " lock"  : "",
2080                 (acc & NFSD_MAY_OWNER_OVERRIDE)? " owneroverride" : "",
2081                 inode->i_mode,
2082                 IS_IMMUTABLE(inode)?    " immut" : "",
2083                 IS_APPEND(inode)?       " append" : "",
2084                 __mnt_is_readonly(exp->ex_path.mnt)?    " ro" : "");
2085         dprintk("      owner %d/%d user %d/%d\n",
2086                 inode->i_uid, inode->i_gid, current_fsuid(), current_fsgid());
2087 #endif
2088
2089         /* Normally we reject any write/sattr etc access on a read-only file
2090          * system.  But if it is IRIX doing check on write-access for a
2091          * device special file, we ignore rofs.
2092          */
2093         if (!(acc & NFSD_MAY_LOCAL_ACCESS))
2094                 if (acc & (NFSD_MAY_WRITE | NFSD_MAY_SATTR | NFSD_MAY_TRUNC)) {
2095                         if (exp_rdonly(rqstp, exp) ||
2096                             __mnt_is_readonly(exp->ex_path.mnt))
2097                                 return nfserr_rofs;
2098                         if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode))
2099                                 return nfserr_perm;
2100                 }
2101         if ((acc & NFSD_MAY_TRUNC) && IS_APPEND(inode))
2102                 return nfserr_perm;
2103
2104         if (acc & NFSD_MAY_LOCK) {
2105                 /* If we cannot rely on authentication in NLM requests,
2106                  * just allow locks, otherwise require read permission, or
2107                  * ownership
2108                  */
2109                 if (exp->ex_flags & NFSEXP_NOAUTHNLM)
2110                         return 0;
2111                 else
2112                         acc = NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE;
2113         }
2114         /*
2115          * The file owner always gets access permission for accesses that
2116          * would normally be checked at open time. This is to make
2117          * file access work even when the client has done a fchmod(fd, 0).
2118          *
2119          * However, `cp foo bar' should fail nevertheless when bar is
2120          * readonly. A sensible way to do this might be to reject all
2121          * attempts to truncate a read-only file, because a creat() call
2122          * always implies file truncation.
2123          * ... but this isn't really fair.  A process may reasonably call
2124          * ftruncate on an open file descriptor on a file with perm 000.
2125          * We must trust the client to do permission checking - using "ACCESS"
2126          * with NFSv3.
2127          */
2128         if ((acc & NFSD_MAY_OWNER_OVERRIDE) &&
2129             inode->i_uid == current_fsuid())
2130                 return 0;
2131
2132         /* This assumes  NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */
2133         err = inode_permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC));
2134
2135         /* Allow read access to binaries even when mode 111 */
2136         if (err == -EACCES && S_ISREG(inode->i_mode) &&
2137              (acc == (NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE) ||
2138               acc == (NFSD_MAY_READ | NFSD_MAY_READ_IF_EXEC)))
2139                 err = inode_permission(inode, MAY_EXEC);
2140
2141         return err? nfserrno(err) : 0;
2142 }
2143
2144 void
2145 nfsd_racache_shutdown(void)
2146 {
2147         struct raparms *raparm, *last_raparm;
2148         unsigned int i;
2149
2150         dprintk("nfsd: freeing readahead buffers.\n");
2151
2152         for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2153                 raparm = raparm_hash[i].pb_head;
2154                 while(raparm) {
2155                         last_raparm = raparm;
2156                         raparm = raparm->p_next;
2157                         kfree(last_raparm);
2158                 }
2159                 raparm_hash[i].pb_head = NULL;
2160         }
2161 }
2162 /*
2163  * Initialize readahead param cache
2164  */
2165 int
2166 nfsd_racache_init(int cache_size)
2167 {
2168         int     i;
2169         int     j = 0;
2170         int     nperbucket;
2171         struct raparms **raparm = NULL;
2172
2173
2174         if (raparm_hash[0].pb_head)
2175                 return 0;
2176         nperbucket = DIV_ROUND_UP(cache_size, RAPARM_HASH_SIZE);
2177         if (nperbucket < 2)
2178                 nperbucket = 2;
2179         cache_size = nperbucket * RAPARM_HASH_SIZE;
2180
2181         dprintk("nfsd: allocating %d readahead buffers.\n", cache_size);
2182
2183         for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2184                 spin_lock_init(&raparm_hash[i].pb_lock);
2185
2186                 raparm = &raparm_hash[i].pb_head;
2187                 for (j = 0; j < nperbucket; j++) {
2188                         *raparm = kzalloc(sizeof(struct raparms), GFP_KERNEL);
2189                         if (!*raparm)
2190                                 goto out_nomem;
2191                         raparm = &(*raparm)->p_next;
2192                 }
2193                 *raparm = NULL;
2194         }
2195
2196         nfsdstats.ra_size = cache_size;
2197         return 0;
2198
2199 out_nomem:
2200         dprintk("nfsd: kmalloc failed, freeing readahead buffers\n");
2201         nfsd_racache_shutdown();
2202         return -ENOMEM;
2203 }
2204
2205 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
2206 struct posix_acl *
2207 nfsd_get_posix_acl(struct svc_fh *fhp, int type)
2208 {
2209         struct inode *inode = fhp->fh_dentry->d_inode;
2210         char *name;
2211         void *value = NULL;
2212         ssize_t size;
2213         struct posix_acl *acl;
2214
2215         if (!IS_POSIXACL(inode))
2216                 return ERR_PTR(-EOPNOTSUPP);
2217
2218         switch (type) {
2219         case ACL_TYPE_ACCESS:
2220                 name = POSIX_ACL_XATTR_ACCESS;
2221                 break;
2222         case ACL_TYPE_DEFAULT:
2223                 name = POSIX_ACL_XATTR_DEFAULT;
2224                 break;
2225         default:
2226                 return ERR_PTR(-EOPNOTSUPP);
2227         }
2228
2229         size = nfsd_getxattr(fhp->fh_dentry, name, &value);
2230         if (size < 0)
2231                 return ERR_PTR(size);
2232
2233         acl = posix_acl_from_xattr(value, size);
2234         kfree(value);
2235         return acl;
2236 }
2237
2238 int
2239 nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl)
2240 {
2241         struct inode *inode = fhp->fh_dentry->d_inode;
2242         char *name;
2243         void *value = NULL;
2244         size_t size;
2245         int error;
2246
2247         if (!IS_POSIXACL(inode) ||
2248             !inode->i_op->setxattr || !inode->i_op->removexattr)
2249                 return -EOPNOTSUPP;
2250         switch(type) {
2251                 case ACL_TYPE_ACCESS:
2252                         name = POSIX_ACL_XATTR_ACCESS;
2253                         break;
2254                 case ACL_TYPE_DEFAULT:
2255                         name = POSIX_ACL_XATTR_DEFAULT;
2256                         break;
2257                 default:
2258                         return -EOPNOTSUPP;
2259         }
2260
2261         if (acl && acl->a_count) {
2262                 size = posix_acl_xattr_size(acl->a_count);
2263                 value = kmalloc(size, GFP_KERNEL);
2264                 if (!value)
2265                         return -ENOMEM;
2266                 error = posix_acl_to_xattr(acl, value, size);
2267                 if (error < 0)
2268                         goto getout;
2269                 size = error;
2270         } else
2271                 size = 0;
2272
2273         error = mnt_want_write(fhp->fh_export->ex_path.mnt);
2274         if (error)
2275                 goto getout;
2276         if (size)
2277                 error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0);
2278         else {
2279                 if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT)
2280                         error = 0;
2281                 else {
2282                         error = vfs_removexattr(fhp->fh_dentry, name);
2283                         if (error == -ENODATA)
2284                                 error = 0;
2285                 }
2286         }
2287         mnt_drop_write(fhp->fh_export->ex_path.mnt);
2288
2289 getout:
2290         kfree(value);
2291         return error;
2292 }
2293 #endif  /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */