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