Linux Audio

Check our new training course

Loading...
v3.1
   1/*
   2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
   3 * All Rights Reserved.
   4 *
   5 * This program is free software; you can redistribute it and/or
   6 * modify it under the terms of the GNU General Public License as
   7 * published by the Free Software Foundation.
   8 *
   9 * This program is distributed in the hope that it would be useful,
  10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12 * GNU General Public License for more details.
  13 *
  14 * You should have received a copy of the GNU General Public License
  15 * along with this program; if not, write the Free Software Foundation,
  16 * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  17 */
  18#include "xfs.h"
  19#include "xfs_fs.h"
  20#include "xfs_acl.h"
  21#include "xfs_bit.h"
  22#include "xfs_log.h"
  23#include "xfs_inum.h"
  24#include "xfs_trans.h"
  25#include "xfs_sb.h"
  26#include "xfs_ag.h"
  27#include "xfs_alloc.h"
  28#include "xfs_quota.h"
  29#include "xfs_mount.h"
  30#include "xfs_bmap_btree.h"
  31#include "xfs_dinode.h"
  32#include "xfs_inode.h"
  33#include "xfs_bmap.h"
  34#include "xfs_rtalloc.h"
 
 
  35#include "xfs_error.h"
  36#include "xfs_itable.h"
  37#include "xfs_rw.h"
  38#include "xfs_attr.h"
  39#include "xfs_buf_item.h"
  40#include "xfs_utils.h"
  41#include "xfs_vnodeops.h"
  42#include "xfs_inode_item.h"
  43#include "xfs_trace.h"
 
 
 
 
 
 
 
  44
  45#include <linux/capability.h>
  46#include <linux/xattr.h>
  47#include <linux/namei.h>
  48#include <linux/posix_acl.h>
  49#include <linux/security.h>
  50#include <linux/fiemap.h>
  51#include <linux/slab.h>
  52
  53/*
  54 * Bring the timestamps in the XFS inode uptodate.
  55 *
  56 * Used before writing the inode to disk.
  57 */
  58void
  59xfs_synchronize_times(
  60	xfs_inode_t	*ip)
  61{
  62	struct inode	*inode = VFS_I(ip);
  63
  64	ip->i_d.di_atime.t_sec = (__int32_t)inode->i_atime.tv_sec;
  65	ip->i_d.di_atime.t_nsec = (__int32_t)inode->i_atime.tv_nsec;
  66	ip->i_d.di_ctime.t_sec = (__int32_t)inode->i_ctime.tv_sec;
  67	ip->i_d.di_ctime.t_nsec = (__int32_t)inode->i_ctime.tv_nsec;
  68	ip->i_d.di_mtime.t_sec = (__int32_t)inode->i_mtime.tv_sec;
  69	ip->i_d.di_mtime.t_nsec = (__int32_t)inode->i_mtime.tv_nsec;
  70}
  71
  72/*
  73 * If the linux inode is valid, mark it dirty, else mark the dirty state
  74 * in the XFS inode to make sure we pick it up when reclaiming the inode.
  75 */
  76void
  77xfs_mark_inode_dirty_sync(
  78	xfs_inode_t	*ip)
  79{
  80	struct inode	*inode = VFS_I(ip);
  81
  82	if (!(inode->i_state & (I_WILL_FREE|I_FREEING)))
  83		mark_inode_dirty_sync(inode);
  84	else {
  85		barrier();
  86		ip->i_update_core = 1;
  87	}
  88}
  89
  90void
  91xfs_mark_inode_dirty(
  92	xfs_inode_t	*ip)
  93{
  94	struct inode	*inode = VFS_I(ip);
 
 
  95
  96	if (!(inode->i_state & (I_WILL_FREE|I_FREEING)))
  97		mark_inode_dirty(inode);
  98	else {
  99		barrier();
 100		ip->i_update_core = 1;
 101	}
 102
 103}
 104
 105/*
 106 * Hook in SELinux.  This is not quite correct yet, what we really need
 107 * here (as we do for default ACLs) is a mechanism by which creation of
 108 * these attrs can be journalled at inode creation time (along with the
 109 * inode, of course, such that log replay can't cause these to be lost).
 110 */
 
 111STATIC int
 112xfs_init_security(
 113	struct inode	*inode,
 114	struct inode	*dir,
 115	const struct qstr *qstr)
 116{
 117	struct xfs_inode *ip = XFS_I(inode);
 118	size_t		length;
 119	void		*value;
 120	unsigned char	*name;
 121	int		error;
 122
 123	error = security_inode_init_security(inode, dir, qstr, (char **)&name,
 124					     &value, &length);
 125	if (error) {
 126		if (error == -EOPNOTSUPP)
 127			return 0;
 128		return -error;
 129	}
 130
 131	error = xfs_attr_set(ip, name, value, length, ATTR_SECURE);
 132
 133	kfree(name);
 134	kfree(value);
 135	return error;
 136}
 137
 138static void
 139xfs_dentry_to_name(
 140	struct xfs_name	*namep,
 141	struct dentry	*dentry)
 142{
 143	namep->name = dentry->d_name.name;
 144	namep->len = dentry->d_name.len;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 145}
 146
 147STATIC void
 148xfs_cleanup_inode(
 149	struct inode	*dir,
 150	struct inode	*inode,
 151	struct dentry	*dentry)
 152{
 153	struct xfs_name	teardown;
 154
 155	/* Oh, the horror.
 156	 * If we can't add the ACL or we fail in
 157	 * xfs_init_security we must back out.
 158	 * ENOSPC can hit here, among other things.
 159	 */
 160	xfs_dentry_to_name(&teardown, dentry);
 161
 162	xfs_remove(XFS_I(dir), &teardown, XFS_I(inode));
 163	iput(inode);
 164}
 165
 166STATIC int
 167xfs_vn_mknod(
 168	struct inode	*dir,
 169	struct dentry	*dentry,
 170	int		mode,
 171	dev_t		rdev)
 
 172{
 173	struct inode	*inode;
 174	struct xfs_inode *ip = NULL;
 175	struct posix_acl *default_acl = NULL;
 176	struct xfs_name	name;
 177	int		error;
 178
 179	/*
 180	 * Irix uses Missed'em'V split, but doesn't want to see
 181	 * the upper 5 bits of (14bit) major.
 182	 */
 183	if (S_ISCHR(mode) || S_ISBLK(mode)) {
 184		if (unlikely(!sysv_valid_dev(rdev) || MAJOR(rdev) & ~0x1ff))
 185			return -EINVAL;
 186		rdev = sysv_encode_dev(rdev);
 187	} else {
 188		rdev = 0;
 189	}
 190
 191	if (IS_POSIXACL(dir)) {
 192		default_acl = xfs_get_acl(dir, ACL_TYPE_DEFAULT);
 193		if (IS_ERR(default_acl))
 194			return PTR_ERR(default_acl);
 195
 196		if (!default_acl)
 197			mode &= ~current_umask();
 198	}
 
 199
 200	xfs_dentry_to_name(&name, dentry);
 201	error = xfs_create(XFS_I(dir), &name, mode, rdev, &ip);
 
 
 
 202	if (unlikely(error))
 203		goto out_free_acl;
 204
 205	inode = VFS_I(ip);
 206
 207	error = xfs_init_security(inode, dir, &dentry->d_name);
 208	if (unlikely(error))
 209		goto out_cleanup_inode;
 210
 
 211	if (default_acl) {
 212		error = -xfs_inherit_acl(inode, default_acl);
 213		default_acl = NULL;
 214		if (unlikely(error))
 215			goto out_cleanup_inode;
 216	}
 
 
 
 
 
 
 217
 
 218
 219	d_instantiate(dentry, inode);
 220	return -error;
 
 
 
 
 221
 222 out_cleanup_inode:
 223	xfs_cleanup_inode(dir, inode, dentry);
 224 out_free_acl:
 225	posix_acl_release(default_acl);
 226	return -error;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 227}
 228
 229STATIC int
 230xfs_vn_create(
 231	struct inode	*dir,
 232	struct dentry	*dentry,
 233	int		mode,
 234	struct nameidata *nd)
 235{
 236	return xfs_vn_mknod(dir, dentry, mode, 0);
 237}
 238
 239STATIC int
 240xfs_vn_mkdir(
 241	struct inode	*dir,
 242	struct dentry	*dentry,
 243	int		mode)
 244{
 245	return xfs_vn_mknod(dir, dentry, mode|S_IFDIR, 0);
 246}
 247
 248STATIC struct dentry *
 249xfs_vn_lookup(
 250	struct inode	*dir,
 251	struct dentry	*dentry,
 252	struct nameidata *nd)
 253{
 254	struct xfs_inode *cip;
 255	struct xfs_name	name;
 256	int		error;
 257
 258	if (dentry->d_name.len >= MAXNAMELEN)
 259		return ERR_PTR(-ENAMETOOLONG);
 260
 261	xfs_dentry_to_name(&name, dentry);
 262	error = xfs_lookup(XFS_I(dir), &name, &cip, NULL);
 263	if (unlikely(error)) {
 264		if (unlikely(error != ENOENT))
 265			return ERR_PTR(-error);
 266		d_add(dentry, NULL);
 267		return NULL;
 268	}
 269
 270	return d_splice_alias(VFS_I(cip), dentry);
 271}
 272
 273STATIC struct dentry *
 274xfs_vn_ci_lookup(
 275	struct inode	*dir,
 276	struct dentry	*dentry,
 277	struct nameidata *nd)
 278{
 279	struct xfs_inode *ip;
 280	struct xfs_name	xname;
 281	struct xfs_name ci_name;
 282	struct qstr	dname;
 283	int		error;
 284
 285	if (dentry->d_name.len >= MAXNAMELEN)
 286		return ERR_PTR(-ENAMETOOLONG);
 287
 288	xfs_dentry_to_name(&xname, dentry);
 289	error = xfs_lookup(XFS_I(dir), &xname, &ip, &ci_name);
 290	if (unlikely(error)) {
 291		if (unlikely(error != ENOENT))
 292			return ERR_PTR(-error);
 293		/*
 294		 * call d_add(dentry, NULL) here when d_drop_negative_children
 295		 * is called in xfs_vn_mknod (ie. allow negative dentries
 296		 * with CI filesystems).
 297		 */
 298		return NULL;
 299	}
 300
 301	/* if exact match, just splice and exit */
 302	if (!ci_name.name)
 303		return d_splice_alias(VFS_I(ip), dentry);
 304
 305	/* else case-insensitive match... */
 306	dname.name = ci_name.name;
 307	dname.len = ci_name.len;
 308	dentry = d_add_ci(dentry, VFS_I(ip), &dname);
 309	kmem_free(ci_name.name);
 310	return dentry;
 311}
 312
 313STATIC int
 314xfs_vn_link(
 315	struct dentry	*old_dentry,
 316	struct inode	*dir,
 317	struct dentry	*dentry)
 318{
 319	struct inode	*inode = old_dentry->d_inode;
 320	struct xfs_name	name;
 321	int		error;
 322
 323	xfs_dentry_to_name(&name, dentry);
 
 
 324
 325	error = xfs_link(XFS_I(dir), XFS_I(inode), &name);
 326	if (unlikely(error))
 327		return -error;
 328
 329	ihold(inode);
 330	d_instantiate(dentry, inode);
 331	return 0;
 332}
 333
 334STATIC int
 335xfs_vn_unlink(
 336	struct inode	*dir,
 337	struct dentry	*dentry)
 338{
 339	struct xfs_name	name;
 340	int		error;
 341
 342	xfs_dentry_to_name(&name, dentry);
 343
 344	error = -xfs_remove(XFS_I(dir), &name, XFS_I(dentry->d_inode));
 345	if (error)
 346		return error;
 347
 348	/*
 349	 * With unlink, the VFS makes the dentry "negative": no inode,
 350	 * but still hashed. This is incompatible with case-insensitive
 351	 * mode, so invalidate (unhash) the dentry in CI-mode.
 352	 */
 353	if (xfs_sb_version_hasasciici(&XFS_M(dir->i_sb)->m_sb))
 354		d_invalidate(dentry);
 355	return 0;
 356}
 357
 358STATIC int
 359xfs_vn_symlink(
 360	struct inode	*dir,
 361	struct dentry	*dentry,
 362	const char	*symname)
 363{
 364	struct inode	*inode;
 365	struct xfs_inode *cip = NULL;
 366	struct xfs_name	name;
 367	int		error;
 368	mode_t		mode;
 369
 370	mode = S_IFLNK |
 371		(irix_symlink_mode ? 0777 & ~current_umask() : S_IRWXUGO);
 372	xfs_dentry_to_name(&name, dentry);
 
 
 373
 374	error = xfs_symlink(XFS_I(dir), &name, symname, mode, &cip);
 375	if (unlikely(error))
 376		goto out;
 377
 378	inode = VFS_I(cip);
 379
 380	error = xfs_init_security(inode, dir, &dentry->d_name);
 381	if (unlikely(error))
 382		goto out_cleanup_inode;
 383
 
 
 384	d_instantiate(dentry, inode);
 
 385	return 0;
 386
 387 out_cleanup_inode:
 
 388	xfs_cleanup_inode(dir, inode, dentry);
 
 389 out:
 390	return -error;
 391}
 392
 393STATIC int
 394xfs_vn_rename(
 395	struct inode	*odir,
 396	struct dentry	*odentry,
 397	struct inode	*ndir,
 398	struct dentry	*ndentry)
 
 399{
 400	struct inode	*new_inode = ndentry->d_inode;
 
 
 401	struct xfs_name	oname;
 402	struct xfs_name	nname;
 403
 404	xfs_dentry_to_name(&oname, odentry);
 405	xfs_dentry_to_name(&nname, ndentry);
 
 
 
 
 406
 407	return -xfs_rename(XFS_I(odir), &oname, XFS_I(odentry->d_inode),
 408			   XFS_I(ndir), &nname, new_inode ?
 409			   			XFS_I(new_inode) : NULL);
 
 
 
 
 
 
 
 
 
 410}
 411
 412/*
 413 * careful here - this function can get called recursively, so
 414 * we need to be very careful about how much stack we use.
 415 * uio is kmalloced for this reason...
 416 */
 417STATIC void *
 418xfs_vn_follow_link(
 419	struct dentry		*dentry,
 420	struct nameidata	*nd)
 
 421{
 422	char			*link;
 423	int			error = -ENOMEM;
 424
 
 
 
 425	link = kmalloc(MAXPATHLEN+1, GFP_KERNEL);
 426	if (!link)
 427		goto out_err;
 428
 429	error = -xfs_readlink(XFS_I(dentry->d_inode), link);
 430	if (unlikely(error))
 431		goto out_kfree;
 432
 433	nd_set_link(nd, link);
 434	return NULL;
 435
 436 out_kfree:
 437	kfree(link);
 438 out_err:
 439	nd_set_link(nd, ERR_PTR(error));
 440	return NULL;
 441}
 442
 443STATIC void
 444xfs_vn_put_link(
 445	struct dentry	*dentry,
 446	struct nameidata *nd,
 447	void		*p)
 448{
 449	char		*s = nd_get_link(nd);
 450
 451	if (!IS_ERR(s))
 452		kfree(s);
 453}
 454
 455STATIC int
 456xfs_vn_getattr(
 457	struct vfsmount		*mnt,
 458	struct dentry		*dentry,
 459	struct kstat		*stat)
 460{
 461	struct inode		*inode = dentry->d_inode;
 462	struct xfs_inode	*ip = XFS_I(inode);
 463	struct xfs_mount	*mp = ip->i_mount;
 464
 465	trace_xfs_getattr(ip);
 466
 467	if (XFS_FORCED_SHUTDOWN(mp))
 468		return XFS_ERROR(EIO);
 469
 470	stat->size = XFS_ISIZE(ip);
 471	stat->dev = inode->i_sb->s_dev;
 472	stat->mode = ip->i_d.di_mode;
 473	stat->nlink = ip->i_d.di_nlink;
 474	stat->uid = ip->i_d.di_uid;
 475	stat->gid = ip->i_d.di_gid;
 476	stat->ino = ip->i_ino;
 477	stat->atime = inode->i_atime;
 478	stat->mtime = inode->i_mtime;
 479	stat->ctime = inode->i_ctime;
 480	stat->blocks =
 481		XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks);
 482
 483
 484	switch (inode->i_mode & S_IFMT) {
 485	case S_IFBLK:
 486	case S_IFCHR:
 487		stat->blksize = BLKDEV_IOSIZE;
 488		stat->rdev = MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff,
 489				   sysv_minor(ip->i_df.if_u2.if_rdev));
 490		break;
 491	default:
 492		if (XFS_IS_REALTIME_INODE(ip)) {
 493			/*
 494			 * If the file blocks are being allocated from a
 495			 * realtime volume, then return the inode's realtime
 496			 * extent size or the realtime volume's extent size.
 497			 */
 498			stat->blksize =
 499				xfs_get_extsz_hint(ip) << mp->m_sb.sb_blocklog;
 500		} else
 501			stat->blksize = xfs_preferred_iosize(mp);
 502		stat->rdev = 0;
 503		break;
 504	}
 505
 506	return 0;
 507}
 508
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 509int
 510xfs_setattr_nonsize(
 511	struct xfs_inode	*ip,
 512	struct iattr		*iattr,
 513	int			flags)
 514{
 515	xfs_mount_t		*mp = ip->i_mount;
 516	struct inode		*inode = VFS_I(ip);
 517	int			mask = iattr->ia_valid;
 518	xfs_trans_t		*tp;
 519	int			error;
 520	uid_t			uid = 0, iuid = 0;
 521	gid_t			gid = 0, igid = 0;
 522	struct xfs_dquot	*udqp = NULL, *gdqp = NULL;
 523	struct xfs_dquot	*olddquot1 = NULL, *olddquot2 = NULL;
 524
 525	trace_xfs_setattr(ip);
 526
 527	if (mp->m_flags & XFS_MOUNT_RDONLY)
 528		return XFS_ERROR(EROFS);
 529
 530	if (XFS_FORCED_SHUTDOWN(mp))
 531		return XFS_ERROR(EIO);
 532
 533	error = -inode_change_ok(inode, iattr);
 534	if (error)
 535		return XFS_ERROR(error);
 536
 537	ASSERT((mask & ATTR_SIZE) == 0);
 538
 539	/*
 540	 * If disk quotas is on, we make sure that the dquots do exist on disk,
 541	 * before we start any other transactions. Trying to do this later
 542	 * is messy. We don't care to take a readlock to look at the ids
 543	 * in inode here, because we can't hold it across the trans_reserve.
 544	 * If the IDs do change before we take the ilock, we're covered
 545	 * because the i_*dquot fields will get updated anyway.
 546	 */
 547	if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) {
 548		uint	qflags = 0;
 549
 550		if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) {
 551			uid = iattr->ia_uid;
 552			qflags |= XFS_QMOPT_UQUOTA;
 553		} else {
 554			uid = ip->i_d.di_uid;
 555		}
 556		if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) {
 557			gid = iattr->ia_gid;
 558			qflags |= XFS_QMOPT_GQUOTA;
 559		}  else {
 560			gid = ip->i_d.di_gid;
 561		}
 562
 563		/*
 564		 * We take a reference when we initialize udqp and gdqp,
 565		 * so it is important that we never blindly double trip on
 566		 * the same variable. See xfs_create() for an example.
 567		 */
 568		ASSERT(udqp == NULL);
 569		ASSERT(gdqp == NULL);
 570		error = xfs_qm_vop_dqalloc(ip, uid, gid, xfs_get_projid(ip),
 571					 qflags, &udqp, &gdqp);
 
 
 572		if (error)
 573			return error;
 574	}
 575
 576	tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_NOT_SIZE);
 577	error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES(mp), 0, 0, 0);
 578	if (error)
 579		goto out_dqrele;
 580
 581	xfs_ilock(ip, XFS_ILOCK_EXCL);
 
 582
 583	/*
 584	 * Change file ownership.  Must be the owner or privileged.
 585	 */
 586	if (mask & (ATTR_UID|ATTR_GID)) {
 587		/*
 588		 * These IDs could have changed since we last looked at them.
 589		 * But, we're assured that if the ownership did change
 590		 * while we didn't have the inode locked, inode's dquot(s)
 591		 * would have changed also.
 592		 */
 593		iuid = ip->i_d.di_uid;
 594		igid = ip->i_d.di_gid;
 595		gid = (mask & ATTR_GID) ? iattr->ia_gid : igid;
 596		uid = (mask & ATTR_UID) ? iattr->ia_uid : iuid;
 597
 598		/*
 599		 * Do a quota reservation only if uid/gid is actually
 600		 * going to change.
 601		 */
 602		if (XFS_IS_QUOTA_RUNNING(mp) &&
 603		    ((XFS_IS_UQUOTA_ON(mp) && iuid != uid) ||
 604		     (XFS_IS_GQUOTA_ON(mp) && igid != gid))) {
 605			ASSERT(tp);
 606			error = xfs_qm_vop_chown_reserve(tp, ip, udqp, gdqp,
 607						capable(CAP_FOWNER) ?
 608						XFS_QMOPT_FORCE_RES : 0);
 609			if (error)	/* out of quota */
 610				goto out_trans_cancel;
 611		}
 612	}
 613
 614	xfs_trans_ijoin(tp, ip);
 615
 616	/*
 617	 * Change file ownership.  Must be the owner or privileged.
 618	 */
 619	if (mask & (ATTR_UID|ATTR_GID)) {
 620		/*
 621		 * CAP_FSETID overrides the following restrictions:
 622		 *
 623		 * The set-user-ID and set-group-ID bits of a file will be
 624		 * cleared upon successful return from chown()
 625		 */
 626		if ((ip->i_d.di_mode & (S_ISUID|S_ISGID)) &&
 627		    !capable(CAP_FSETID))
 628			ip->i_d.di_mode &= ~(S_ISUID|S_ISGID);
 629
 630		/*
 631		 * Change the ownerships and register quota modifications
 632		 * in the transaction.
 633		 */
 634		if (iuid != uid) {
 635			if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_UQUOTA_ON(mp)) {
 636				ASSERT(mask & ATTR_UID);
 637				ASSERT(udqp);
 638				olddquot1 = xfs_qm_vop_chown(tp, ip,
 639							&ip->i_udquot, udqp);
 640			}
 641			ip->i_d.di_uid = uid;
 642			inode->i_uid = uid;
 643		}
 644		if (igid != gid) {
 645			if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_GQUOTA_ON(mp)) {
 646				ASSERT(!XFS_IS_PQUOTA_ON(mp));
 
 647				ASSERT(mask & ATTR_GID);
 648				ASSERT(gdqp);
 649				olddquot2 = xfs_qm_vop_chown(tp, ip,
 650							&ip->i_gdquot, gdqp);
 651			}
 652			ip->i_d.di_gid = gid;
 653			inode->i_gid = gid;
 654		}
 655	}
 656
 657	/*
 658	 * Change file access modes.
 659	 */
 660	if (mask & ATTR_MODE) {
 661		umode_t mode = iattr->ia_mode;
 662
 663		if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID))
 664			mode &= ~S_ISGID;
 665
 666		ip->i_d.di_mode &= S_IFMT;
 667		ip->i_d.di_mode |= mode & ~S_IFMT;
 668
 669		inode->i_mode &= S_IFMT;
 670		inode->i_mode |= mode & ~S_IFMT;
 671	}
 672
 673	/*
 674	 * Change file access or modified times.
 675	 */
 676	if (mask & ATTR_ATIME) {
 677		inode->i_atime = iattr->ia_atime;
 678		ip->i_d.di_atime.t_sec = iattr->ia_atime.tv_sec;
 679		ip->i_d.di_atime.t_nsec = iattr->ia_atime.tv_nsec;
 680		ip->i_update_core = 1;
 681	}
 682	if (mask & ATTR_CTIME) {
 683		inode->i_ctime = iattr->ia_ctime;
 684		ip->i_d.di_ctime.t_sec = iattr->ia_ctime.tv_sec;
 685		ip->i_d.di_ctime.t_nsec = iattr->ia_ctime.tv_nsec;
 686		ip->i_update_core = 1;
 687	}
 688	if (mask & ATTR_MTIME) {
 689		inode->i_mtime = iattr->ia_mtime;
 690		ip->i_d.di_mtime.t_sec = iattr->ia_mtime.tv_sec;
 691		ip->i_d.di_mtime.t_nsec = iattr->ia_mtime.tv_nsec;
 692		ip->i_update_core = 1;
 693	}
 694
 695	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
 696
 697	XFS_STATS_INC(xs_ig_attrchg);
 698
 699	if (mp->m_flags & XFS_MOUNT_WSYNC)
 700		xfs_trans_set_sync(tp);
 701	error = xfs_trans_commit(tp, 0);
 702
 703	xfs_iunlock(ip, XFS_ILOCK_EXCL);
 704
 705	/*
 706	 * Release any dquot(s) the inode had kept before chown.
 707	 */
 708	xfs_qm_dqrele(olddquot1);
 709	xfs_qm_dqrele(olddquot2);
 710	xfs_qm_dqrele(udqp);
 711	xfs_qm_dqrele(gdqp);
 712
 713	if (error)
 714		return XFS_ERROR(error);
 715
 716	/*
 717	 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode
 718	 * 	     update.  We could avoid this with linked transactions
 719	 * 	     and passing down the transaction pointer all the way
 720	 *	     to attr_set.  No previous user of the generic
 721	 * 	     Posix ACL code seems to care about this issue either.
 722	 */
 723	if ((mask & ATTR_MODE) && !(flags & XFS_ATTR_NOACL)) {
 724		error = -xfs_acl_chmod(inode);
 725		if (error)
 726			return XFS_ERROR(error);
 727	}
 728
 729	return 0;
 730
 731out_trans_cancel:
 732	xfs_trans_cancel(tp, 0);
 733	xfs_iunlock(ip, XFS_ILOCK_EXCL);
 734out_dqrele:
 735	xfs_qm_dqrele(udqp);
 736	xfs_qm_dqrele(gdqp);
 737	return error;
 738}
 739
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 740/*
 741 * Truncate file.  Must have write permission and not be a directory.
 
 
 
 742 */
 743int
 744xfs_setattr_size(
 745	struct xfs_inode	*ip,
 746	struct iattr		*iattr,
 747	int			flags)
 748{
 749	struct xfs_mount	*mp = ip->i_mount;
 750	struct inode		*inode = VFS_I(ip);
 751	int			mask = iattr->ia_valid;
 752	struct xfs_trans	*tp;
 753	int			error;
 754	uint			lock_flags;
 755	uint			commit_flags = 0;
 756
 757	trace_xfs_setattr(ip);
 
 
 
 
 758
 759	if (mp->m_flags & XFS_MOUNT_RDONLY)
 760		return XFS_ERROR(EROFS);
 761
 762	if (XFS_FORCED_SHUTDOWN(mp))
 763		return XFS_ERROR(EIO);
 764
 765	error = -inode_change_ok(inode, iattr);
 766	if (error)
 767		return XFS_ERROR(error);
 768
 769	ASSERT(S_ISREG(ip->i_d.di_mode));
 770	ASSERT((mask & (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_ATIME|ATTR_ATIME_SET|
 771			ATTR_MTIME_SET|ATTR_KILL_SUID|ATTR_KILL_SGID|
 772			ATTR_KILL_PRIV|ATTR_TIMES_SET)) == 0);
 773
 774	lock_flags = XFS_ILOCK_EXCL;
 775	if (!(flags & XFS_ATTR_NOLOCK))
 776		lock_flags |= XFS_IOLOCK_EXCL;
 777	xfs_ilock(ip, lock_flags);
 778
 779	/*
 780	 * Short circuit the truncate case for zero length files.
 781	 */
 782	if (iattr->ia_size == 0 &&
 783	    ip->i_size == 0 && ip->i_d.di_nextents == 0) {
 784		if (!(mask & (ATTR_CTIME|ATTR_MTIME)))
 785			goto out_unlock;
 786
 787		/*
 788		 * Use the regular setattr path to update the timestamps.
 789		 */
 790		xfs_iunlock(ip, lock_flags);
 791		iattr->ia_valid &= ~ATTR_SIZE;
 792		return xfs_setattr_nonsize(ip, iattr, 0);
 793	}
 794
 795	/*
 796	 * Make sure that the dquots are attached to the inode.
 797	 */
 798	error = xfs_qm_dqattach_locked(ip, 0);
 799	if (error)
 800		goto out_unlock;
 801
 802	/*
 803	 * Now we can make the changes.  Before we join the inode to the
 804	 * transaction, take care of the part of the truncation that must be
 805	 * done without the inode lock.  This needs to be done before joining
 806	 * the inode to the transaction, because the inode cannot be unlocked
 807	 * once it is a part of the transaction.
 808	 */
 809	if (iattr->ia_size > ip->i_size) {
 810		/*
 811		 * Do the first part of growing a file: zero any data in the
 812		 * last block that is beyond the old EOF.  We need to do this
 813		 * before the inode is joined to the transaction to modify
 814		 * i_size.
 815		 */
 816		error = xfs_zero_eof(ip, iattr->ia_size, ip->i_size);
 817		if (error)
 818			goto out_unlock;
 
 
 
 
 
 
 
 819	}
 820	xfs_iunlock(ip, XFS_ILOCK_EXCL);
 821	lock_flags &= ~XFS_ILOCK_EXCL;
 
 822
 823	/*
 824	 * We are going to log the inode size change in this transaction so
 825	 * any previous writes that are beyond the on disk EOF and the new
 826	 * EOF that have not been written out need to be written here.  If we
 827	 * do not write the data out, we expose ourselves to the null files
 828	 * problem.
 829	 *
 830	 * Only flush from the on disk size to the smaller of the in memory
 831	 * file size or the new size as that's the range we really care about
 832	 * here and prevents waiting for other data not within the range we
 833	 * care about here.
 834	 */
 835	if (ip->i_size != ip->i_d.di_size && iattr->ia_size > ip->i_d.di_size) {
 836		error = xfs_flush_pages(ip, ip->i_d.di_size, iattr->ia_size,
 837					XBF_ASYNC, FI_NONE);
 
 838		if (error)
 839			goto out_unlock;
 840	}
 841
 842	/*
 843	 * Wait for all I/O to complete.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 844	 */
 845	xfs_ioend_wait(ip);
 846
 847	error = -block_truncate_page(inode->i_mapping, iattr->ia_size,
 848				     xfs_get_blocks);
 849	if (error)
 850		goto out_unlock;
 851
 852	tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_SIZE);
 853	error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
 854				 XFS_TRANS_PERM_LOG_RES,
 855				 XFS_ITRUNCATE_LOG_COUNT);
 856	if (error)
 857		goto out_trans_cancel;
 858
 859	truncate_setsize(inode, iattr->ia_size);
 860
 861	commit_flags = XFS_TRANS_RELEASE_LOG_RES;
 862	lock_flags |= XFS_ILOCK_EXCL;
 863
 864	xfs_ilock(ip, XFS_ILOCK_EXCL);
 865
 866	xfs_trans_ijoin(tp, ip);
 867
 868	/*
 869	 * Only change the c/mtime if we are changing the size or we are
 870	 * explicitly asked to change it.  This handles the semantic difference
 871	 * between truncate() and ftruncate() as implemented in the VFS.
 872	 *
 873	 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
 874	 * special case where we need to update the times despite not having
 875	 * these flags set.  For all other operations the VFS set these flags
 876	 * explicitly if it wants a timestamp update.
 877	 */
 878	if (iattr->ia_size != ip->i_size &&
 879	    (!(mask & (ATTR_CTIME | ATTR_MTIME)))) {
 880		iattr->ia_ctime = iattr->ia_mtime =
 881			current_fs_time(inode->i_sb);
 882		mask |= ATTR_CTIME | ATTR_MTIME;
 883	}
 884
 885	if (iattr->ia_size > ip->i_size) {
 886		ip->i_d.di_size = iattr->ia_size;
 887		ip->i_size = iattr->ia_size;
 888	} else if (iattr->ia_size <= ip->i_size ||
 889		   (iattr->ia_size == 0 && ip->i_d.di_nextents)) {
 890		error = xfs_itruncate_data(&tp, ip, iattr->ia_size);
 
 
 
 
 
 
 
 
 
 
 
 891		if (error)
 892			goto out_trans_abort;
 893
 894		/*
 895		 * Truncated "down", so we're removing references to old data
 896		 * here - if we delay flushing for a long time, we expose
 897		 * ourselves unduly to the notorious NULL files problem.  So,
 898		 * we mark this inode and flush it when the file is closed,
 899		 * and do not wait the usual (long) time for writeout.
 900		 */
 901		xfs_iflags_set(ip, XFS_ITRUNCATED);
 902	}
 903
 904	if (mask & ATTR_CTIME) {
 905		inode->i_ctime = iattr->ia_ctime;
 906		ip->i_d.di_ctime.t_sec = iattr->ia_ctime.tv_sec;
 907		ip->i_d.di_ctime.t_nsec = iattr->ia_ctime.tv_nsec;
 908		ip->i_update_core = 1;
 909	}
 910	if (mask & ATTR_MTIME) {
 911		inode->i_mtime = iattr->ia_mtime;
 912		ip->i_d.di_mtime.t_sec = iattr->ia_mtime.tv_sec;
 913		ip->i_d.di_mtime.t_nsec = iattr->ia_mtime.tv_nsec;
 914		ip->i_update_core = 1;
 915	}
 916
 
 
 
 
 
 917	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
 918
 919	XFS_STATS_INC(xs_ig_attrchg);
 920
 921	if (mp->m_flags & XFS_MOUNT_WSYNC)
 922		xfs_trans_set_sync(tp);
 923
 924	error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
 925out_unlock:
 926	if (lock_flags)
 927		xfs_iunlock(ip, lock_flags);
 928	return error;
 929
 930out_trans_abort:
 931	commit_flags |= XFS_TRANS_ABORT;
 932out_trans_cancel:
 933	xfs_trans_cancel(tp, commit_flags);
 934	goto out_unlock;
 935}
 936
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 937STATIC int
 938xfs_vn_setattr(
 939	struct dentry	*dentry,
 940	struct iattr	*iattr)
 941{
 942	if (iattr->ia_valid & ATTR_SIZE)
 943		return -xfs_setattr_size(XFS_I(dentry->d_inode), iattr, 0);
 944	return -xfs_setattr_nonsize(XFS_I(dentry->d_inode), iattr, 0);
 945}
 946
 947#define XFS_FIEMAP_FLAGS	(FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 948
 949/*
 950 * Call fiemap helper to fill in user data.
 951 * Returns positive errors to xfs_getbmap.
 952 */
 953STATIC int
 954xfs_fiemap_format(
 955	void			**arg,
 956	struct getbmapx		*bmv,
 957	int			*full)
 958{
 
 
 
 959	int			error;
 960	struct fiemap_extent_info *fieinfo = *arg;
 961	u32			fiemap_flags = 0;
 962	u64			logical, physical, length;
 963
 964	/* Do nothing for a hole */
 965	if (bmv->bmv_block == -1LL)
 966		return 0;
 967
 968	logical = BBTOB(bmv->bmv_offset);
 969	physical = BBTOB(bmv->bmv_block);
 970	length = BBTOB(bmv->bmv_length);
 971
 972	if (bmv->bmv_oflags & BMV_OF_PREALLOC)
 973		fiemap_flags |= FIEMAP_EXTENT_UNWRITTEN;
 974	else if (bmv->bmv_oflags & BMV_OF_DELALLOC) {
 975		fiemap_flags |= FIEMAP_EXTENT_DELALLOC;
 976		physical = 0;   /* no block yet */
 977	}
 978	if (bmv->bmv_oflags & BMV_OF_LAST)
 979		fiemap_flags |= FIEMAP_EXTENT_LAST;
 980
 981	error = fiemap_fill_next_extent(fieinfo, logical, physical,
 982					length, fiemap_flags);
 983	if (error > 0) {
 984		error = 0;
 985		*full = 1;	/* user array now full */
 986	}
 987
 988	return -error;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 989}
 990
 991STATIC int
 992xfs_vn_fiemap(
 993	struct inode		*inode,
 994	struct fiemap_extent_info *fieinfo,
 995	u64			start,
 996	u64			length)
 997{
 998	xfs_inode_t		*ip = XFS_I(inode);
 999	struct getbmapx		bm;
1000	int			error;
1001
1002	error = fiemap_check_flags(fieinfo, XFS_FIEMAP_FLAGS);
1003	if (error)
1004		return error;
1005
1006	/* Set up bmap header for xfs internal routine */
1007	bm.bmv_offset = BTOBB(start);
1008	/* Special case for whole file */
1009	if (length == FIEMAP_MAX_OFFSET)
1010		bm.bmv_length = -1LL;
1011	else
1012		bm.bmv_length = BTOBB(length);
1013
1014	/* We add one because in getbmap world count includes the header */
1015	bm.bmv_count = !fieinfo->fi_extents_max ? MAXEXTNUM :
1016					fieinfo->fi_extents_max + 1;
1017	bm.bmv_count = min_t(__s32, bm.bmv_count,
1018			     (PAGE_SIZE * 16 / sizeof(struct getbmapx)));
1019	bm.bmv_iflags = BMV_IF_PREALLOC | BMV_IF_NO_HOLES;
1020	if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR)
1021		bm.bmv_iflags |= BMV_IF_ATTRFORK;
1022	if (!(fieinfo->fi_flags & FIEMAP_FLAG_SYNC))
1023		bm.bmv_iflags |= BMV_IF_DELALLOC;
1024
1025	error = xfs_getbmap(ip, &bm, xfs_fiemap_format, fieinfo);
1026	if (error)
1027		return -error;
1028
1029	return 0;
 
 
 
 
 
 
1030}
1031
1032static const struct inode_operations xfs_inode_operations = {
1033	.get_acl		= xfs_get_acl,
 
1034	.getattr		= xfs_vn_getattr,
1035	.setattr		= xfs_vn_setattr,
1036	.setxattr		= generic_setxattr,
1037	.getxattr		= generic_getxattr,
1038	.removexattr		= generic_removexattr,
1039	.listxattr		= xfs_vn_listxattr,
1040	.fiemap			= xfs_vn_fiemap,
 
1041};
1042
1043static const struct inode_operations xfs_dir_inode_operations = {
1044	.create			= xfs_vn_create,
1045	.lookup			= xfs_vn_lookup,
1046	.link			= xfs_vn_link,
1047	.unlink			= xfs_vn_unlink,
1048	.symlink		= xfs_vn_symlink,
1049	.mkdir			= xfs_vn_mkdir,
1050	/*
1051	 * Yes, XFS uses the same method for rmdir and unlink.
1052	 *
1053	 * There are some subtile differences deeper in the code,
1054	 * but we use S_ISDIR to check for those.
1055	 */
1056	.rmdir			= xfs_vn_unlink,
1057	.mknod			= xfs_vn_mknod,
1058	.rename			= xfs_vn_rename,
1059	.get_acl		= xfs_get_acl,
 
1060	.getattr		= xfs_vn_getattr,
1061	.setattr		= xfs_vn_setattr,
1062	.setxattr		= generic_setxattr,
1063	.getxattr		= generic_getxattr,
1064	.removexattr		= generic_removexattr,
1065	.listxattr		= xfs_vn_listxattr,
 
 
1066};
1067
1068static const struct inode_operations xfs_dir_ci_inode_operations = {
1069	.create			= xfs_vn_create,
1070	.lookup			= xfs_vn_ci_lookup,
1071	.link			= xfs_vn_link,
1072	.unlink			= xfs_vn_unlink,
1073	.symlink		= xfs_vn_symlink,
1074	.mkdir			= xfs_vn_mkdir,
1075	/*
1076	 * Yes, XFS uses the same method for rmdir and unlink.
1077	 *
1078	 * There are some subtile differences deeper in the code,
1079	 * but we use S_ISDIR to check for those.
1080	 */
1081	.rmdir			= xfs_vn_unlink,
1082	.mknod			= xfs_vn_mknod,
1083	.rename			= xfs_vn_rename,
1084	.get_acl		= xfs_get_acl,
 
1085	.getattr		= xfs_vn_getattr,
1086	.setattr		= xfs_vn_setattr,
1087	.setxattr		= generic_setxattr,
1088	.getxattr		= generic_getxattr,
1089	.removexattr		= generic_removexattr,
1090	.listxattr		= xfs_vn_listxattr,
 
 
1091};
1092
1093static const struct inode_operations xfs_symlink_inode_operations = {
1094	.readlink		= generic_readlink,
1095	.follow_link		= xfs_vn_follow_link,
1096	.put_link		= xfs_vn_put_link,
1097	.get_acl		= xfs_get_acl,
1098	.getattr		= xfs_vn_getattr,
1099	.setattr		= xfs_vn_setattr,
1100	.setxattr		= generic_setxattr,
1101	.getxattr		= generic_getxattr,
1102	.removexattr		= generic_removexattr,
1103	.listxattr		= xfs_vn_listxattr,
 
 
 
 
 
 
 
 
 
1104};
1105
1106STATIC void
1107xfs_diflags_to_iflags(
1108	struct inode		*inode,
1109	struct xfs_inode	*ip)
1110{
1111	if (ip->i_d.di_flags & XFS_DIFLAG_IMMUTABLE)
 
 
 
 
 
1112		inode->i_flags |= S_IMMUTABLE;
1113	else
1114		inode->i_flags &= ~S_IMMUTABLE;
1115	if (ip->i_d.di_flags & XFS_DIFLAG_APPEND)
1116		inode->i_flags |= S_APPEND;
1117	else
1118		inode->i_flags &= ~S_APPEND;
1119	if (ip->i_d.di_flags & XFS_DIFLAG_SYNC)
1120		inode->i_flags |= S_SYNC;
1121	else
1122		inode->i_flags &= ~S_SYNC;
1123	if (ip->i_d.di_flags & XFS_DIFLAG_NOATIME)
1124		inode->i_flags |= S_NOATIME;
1125	else
1126		inode->i_flags &= ~S_NOATIME;
 
 
 
 
1127}
1128
1129/*
1130 * Initialize the Linux inode, set up the operation vectors and
1131 * unlock the inode.
1132 *
1133 * When reading existing inodes from disk this is called directly
1134 * from xfs_iget, when creating a new inode it is called from
1135 * xfs_ialloc after setting up the inode.
1136 *
1137 * We are always called with an uninitialised linux inode here.
1138 * We need to initialise the necessary fields and take a reference
1139 * on it.
1140 */
1141void
1142xfs_setup_inode(
1143	struct xfs_inode	*ip)
1144{
1145	struct inode		*inode = &ip->i_vnode;
 
1146
1147	inode->i_ino = ip->i_ino;
1148	inode->i_state = I_NEW;
1149
1150	inode_sb_list_add(inode);
1151	/* make the inode look hashed for the writeback code */
1152	hlist_add_fake(&inode->i_hash);
1153
1154	inode->i_mode	= ip->i_d.di_mode;
1155	inode->i_nlink	= ip->i_d.di_nlink;
1156	inode->i_uid	= ip->i_d.di_uid;
1157	inode->i_gid	= ip->i_d.di_gid;
1158
1159	switch (inode->i_mode & S_IFMT) {
1160	case S_IFBLK:
1161	case S_IFCHR:
1162		inode->i_rdev =
1163			MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff,
1164			      sysv_minor(ip->i_df.if_u2.if_rdev));
1165		break;
1166	default:
1167		inode->i_rdev = 0;
1168		break;
1169	}
1170
1171	inode->i_generation = ip->i_d.di_gen;
1172	i_size_write(inode, ip->i_d.di_size);
1173	inode->i_atime.tv_sec	= ip->i_d.di_atime.t_sec;
1174	inode->i_atime.tv_nsec	= ip->i_d.di_atime.t_nsec;
1175	inode->i_mtime.tv_sec	= ip->i_d.di_mtime.t_sec;
1176	inode->i_mtime.tv_nsec	= ip->i_d.di_mtime.t_nsec;
1177	inode->i_ctime.tv_sec	= ip->i_d.di_ctime.t_sec;
1178	inode->i_ctime.tv_nsec	= ip->i_d.di_ctime.t_nsec;
1179	xfs_diflags_to_iflags(inode, ip);
1180
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1181	switch (inode->i_mode & S_IFMT) {
1182	case S_IFREG:
1183		inode->i_op = &xfs_inode_operations;
1184		inode->i_fop = &xfs_file_operations;
1185		inode->i_mapping->a_ops = &xfs_address_space_operations;
1186		break;
1187	case S_IFDIR:
1188		if (xfs_sb_version_hasasciici(&XFS_M(inode->i_sb)->m_sb))
1189			inode->i_op = &xfs_dir_ci_inode_operations;
1190		else
1191			inode->i_op = &xfs_dir_inode_operations;
1192		inode->i_fop = &xfs_dir_file_operations;
1193		break;
1194	case S_IFLNK:
1195		inode->i_op = &xfs_symlink_inode_operations;
1196		if (!(ip->i_df.if_flags & XFS_IFINLINE))
1197			inode->i_mapping->a_ops = &xfs_address_space_operations;
 
1198		break;
1199	default:
1200		inode->i_op = &xfs_inode_operations;
1201		init_special_inode(inode, inode->i_mode, inode->i_rdev);
1202		break;
1203	}
1204
1205	/*
1206	 * If there is no attribute fork no ACL can exist on this inode,
1207	 * and it can't have any file capabilities attached to it either.
1208	 */
1209	if (!XFS_IFORK_Q(ip)) {
1210		inode_has_no_xattr(inode);
1211		cache_no_acl(inode);
1212	}
1213
1214	xfs_iflags_clear(ip, XFS_INEW);
1215	barrier();
1216
1217	unlock_new_inode(inode);
1218}
v4.10.11
   1/*
   2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
   3 * All Rights Reserved.
   4 *
   5 * This program is free software; you can redistribute it and/or
   6 * modify it under the terms of the GNU General Public License as
   7 * published by the Free Software Foundation.
   8 *
   9 * This program is distributed in the hope that it would be useful,
  10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12 * GNU General Public License for more details.
  13 *
  14 * You should have received a copy of the GNU General Public License
  15 * along with this program; if not, write the Free Software Foundation,
  16 * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  17 */
  18#include "xfs.h"
  19#include "xfs_fs.h"
  20#include "xfs_shared.h"
  21#include "xfs_format.h"
  22#include "xfs_log_format.h"
  23#include "xfs_trans_resv.h"
 
 
 
 
 
  24#include "xfs_mount.h"
  25#include "xfs_da_format.h"
 
  26#include "xfs_inode.h"
  27#include "xfs_bmap.h"
  28#include "xfs_bmap_util.h"
  29#include "xfs_acl.h"
  30#include "xfs_quota.h"
  31#include "xfs_error.h"
 
 
  32#include "xfs_attr.h"
  33#include "xfs_trans.h"
 
 
 
  34#include "xfs_trace.h"
  35#include "xfs_icache.h"
  36#include "xfs_symlink.h"
  37#include "xfs_da_btree.h"
  38#include "xfs_dir2.h"
  39#include "xfs_trans_space.h"
  40#include "xfs_pnfs.h"
  41#include "xfs_iomap.h"
  42
  43#include <linux/capability.h>
  44#include <linux/xattr.h>
 
  45#include <linux/posix_acl.h>
  46#include <linux/security.h>
  47#include <linux/iomap.h>
  48#include <linux/slab.h>
  49
  50/*
  51 * Directories have different lock order w.r.t. mmap_sem compared to regular
  52 * files. This is due to readdir potentially triggering page faults on a user
  53 * buffer inside filldir(), and this happens with the ilock on the directory
  54 * held. For regular files, the lock order is the other way around - the
  55 * mmap_sem is taken during the page fault, and then we lock the ilock to do
  56 * block mapping. Hence we need a different class for the directory ilock so
  57 * that lockdep can tell them apart.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  58 */
  59static struct lock_class_key xfs_nondir_ilock_class;
  60static struct lock_class_key xfs_dir_ilock_class;
 
 
 
  61
  62static int
  63xfs_initxattrs(
  64	struct inode		*inode,
  65	const struct xattr	*xattr_array,
  66	void			*fs_info)
 
 
 
 
 
 
  67{
  68	const struct xattr	*xattr;
  69	struct xfs_inode	*ip = XFS_I(inode);
  70	int			error = 0;
  71
  72	for (xattr = xattr_array; xattr->name != NULL; xattr++) {
  73		error = xfs_attr_set(ip, xattr->name, xattr->value,
  74				      xattr->value_len, ATTR_SECURE);
  75		if (error < 0)
  76			break;
  77	}
  78	return error;
  79}
  80
  81/*
  82 * Hook in SELinux.  This is not quite correct yet, what we really need
  83 * here (as we do for default ACLs) is a mechanism by which creation of
  84 * these attrs can be journalled at inode creation time (along with the
  85 * inode, of course, such that log replay can't cause these to be lost).
  86 */
  87
  88STATIC int
  89xfs_init_security(
  90	struct inode	*inode,
  91	struct inode	*dir,
  92	const struct qstr *qstr)
  93{
  94	return security_inode_init_security(inode, dir, qstr,
  95					     &xfs_initxattrs, NULL);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  96}
  97
  98static void
  99xfs_dentry_to_name(
 100	struct xfs_name	*namep,
 101	struct dentry	*dentry)
 102{
 103	namep->name = dentry->d_name.name;
 104	namep->len = dentry->d_name.len;
 105	namep->type = XFS_DIR3_FT_UNKNOWN;
 106}
 107
 108static int
 109xfs_dentry_mode_to_name(
 110	struct xfs_name	*namep,
 111	struct dentry	*dentry,
 112	int		mode)
 113{
 114	namep->name = dentry->d_name.name;
 115	namep->len = dentry->d_name.len;
 116	namep->type = xfs_mode_to_ftype(mode);
 117
 118	if (unlikely(namep->type == XFS_DIR3_FT_UNKNOWN))
 119		return -EFSCORRUPTED;
 120
 121	return 0;
 122}
 123
 124STATIC void
 125xfs_cleanup_inode(
 126	struct inode	*dir,
 127	struct inode	*inode,
 128	struct dentry	*dentry)
 129{
 130	struct xfs_name	teardown;
 131
 132	/* Oh, the horror.
 133	 * If we can't add the ACL or we fail in
 134	 * xfs_init_security we must back out.
 135	 * ENOSPC can hit here, among other things.
 136	 */
 137	xfs_dentry_to_name(&teardown, dentry);
 138
 139	xfs_remove(XFS_I(dir), &teardown, XFS_I(inode));
 
 140}
 141
 142STATIC int
 143xfs_generic_create(
 144	struct inode	*dir,
 145	struct dentry	*dentry,
 146	umode_t		mode,
 147	dev_t		rdev,
 148	bool		tmpfile)	/* unnamed file */
 149{
 150	struct inode	*inode;
 151	struct xfs_inode *ip = NULL;
 152	struct posix_acl *default_acl, *acl;
 153	struct xfs_name	name;
 154	int		error;
 155
 156	/*
 157	 * Irix uses Missed'em'V split, but doesn't want to see
 158	 * the upper 5 bits of (14bit) major.
 159	 */
 160	if (S_ISCHR(mode) || S_ISBLK(mode)) {
 161		if (unlikely(!sysv_valid_dev(rdev) || MAJOR(rdev) & ~0x1ff))
 162			return -EINVAL;
 163		rdev = sysv_encode_dev(rdev);
 164	} else {
 165		rdev = 0;
 166	}
 167
 168	error = posix_acl_create(dir, &mode, &default_acl, &acl);
 169	if (error)
 170		return error;
 
 171
 172	/* Verify mode is valid also for tmpfile case */
 173	error = xfs_dentry_mode_to_name(&name, dentry, mode);
 174	if (unlikely(error))
 175		goto out_free_acl;
 176
 177	if (!tmpfile) {
 178		error = xfs_create(XFS_I(dir), &name, mode, rdev, &ip);
 179	} else {
 180		error = xfs_create_tmpfile(XFS_I(dir), dentry, mode, &ip);
 181	}
 182	if (unlikely(error))
 183		goto out_free_acl;
 184
 185	inode = VFS_I(ip);
 186
 187	error = xfs_init_security(inode, dir, &dentry->d_name);
 188	if (unlikely(error))
 189		goto out_cleanup_inode;
 190
 191#ifdef CONFIG_XFS_POSIX_ACL
 192	if (default_acl) {
 193		error = xfs_set_acl(inode, default_acl, ACL_TYPE_DEFAULT);
 194		if (error)
 
 195			goto out_cleanup_inode;
 196	}
 197	if (acl) {
 198		error = xfs_set_acl(inode, acl, ACL_TYPE_ACCESS);
 199		if (error)
 200			goto out_cleanup_inode;
 201	}
 202#endif
 203
 204	xfs_setup_iops(ip);
 205
 206	if (tmpfile)
 207		d_tmpfile(dentry, inode);
 208	else
 209		d_instantiate(dentry, inode);
 210
 211	xfs_finish_inode_setup(ip);
 212
 
 
 213 out_free_acl:
 214	if (default_acl)
 215		posix_acl_release(default_acl);
 216	if (acl)
 217		posix_acl_release(acl);
 218	return error;
 219
 220 out_cleanup_inode:
 221	xfs_finish_inode_setup(ip);
 222	if (!tmpfile)
 223		xfs_cleanup_inode(dir, inode, dentry);
 224	iput(inode);
 225	goto out_free_acl;
 226}
 227
 228STATIC int
 229xfs_vn_mknod(
 230	struct inode	*dir,
 231	struct dentry	*dentry,
 232	umode_t		mode,
 233	dev_t		rdev)
 234{
 235	return xfs_generic_create(dir, dentry, mode, rdev, false);
 236}
 237
 238STATIC int
 239xfs_vn_create(
 240	struct inode	*dir,
 241	struct dentry	*dentry,
 242	umode_t		mode,
 243	bool		flags)
 244{
 245	return xfs_vn_mknod(dir, dentry, mode, 0);
 246}
 247
 248STATIC int
 249xfs_vn_mkdir(
 250	struct inode	*dir,
 251	struct dentry	*dentry,
 252	umode_t		mode)
 253{
 254	return xfs_vn_mknod(dir, dentry, mode|S_IFDIR, 0);
 255}
 256
 257STATIC struct dentry *
 258xfs_vn_lookup(
 259	struct inode	*dir,
 260	struct dentry	*dentry,
 261	unsigned int flags)
 262{
 263	struct xfs_inode *cip;
 264	struct xfs_name	name;
 265	int		error;
 266
 267	if (dentry->d_name.len >= MAXNAMELEN)
 268		return ERR_PTR(-ENAMETOOLONG);
 269
 270	xfs_dentry_to_name(&name, dentry);
 271	error = xfs_lookup(XFS_I(dir), &name, &cip, NULL);
 272	if (unlikely(error)) {
 273		if (unlikely(error != -ENOENT))
 274			return ERR_PTR(error);
 275		d_add(dentry, NULL);
 276		return NULL;
 277	}
 278
 279	return d_splice_alias(VFS_I(cip), dentry);
 280}
 281
 282STATIC struct dentry *
 283xfs_vn_ci_lookup(
 284	struct inode	*dir,
 285	struct dentry	*dentry,
 286	unsigned int flags)
 287{
 288	struct xfs_inode *ip;
 289	struct xfs_name	xname;
 290	struct xfs_name ci_name;
 291	struct qstr	dname;
 292	int		error;
 293
 294	if (dentry->d_name.len >= MAXNAMELEN)
 295		return ERR_PTR(-ENAMETOOLONG);
 296
 297	xfs_dentry_to_name(&xname, dentry);
 298	error = xfs_lookup(XFS_I(dir), &xname, &ip, &ci_name);
 299	if (unlikely(error)) {
 300		if (unlikely(error != -ENOENT))
 301			return ERR_PTR(error);
 302		/*
 303		 * call d_add(dentry, NULL) here when d_drop_negative_children
 304		 * is called in xfs_vn_mknod (ie. allow negative dentries
 305		 * with CI filesystems).
 306		 */
 307		return NULL;
 308	}
 309
 310	/* if exact match, just splice and exit */
 311	if (!ci_name.name)
 312		return d_splice_alias(VFS_I(ip), dentry);
 313
 314	/* else case-insensitive match... */
 315	dname.name = ci_name.name;
 316	dname.len = ci_name.len;
 317	dentry = d_add_ci(dentry, VFS_I(ip), &dname);
 318	kmem_free(ci_name.name);
 319	return dentry;
 320}
 321
 322STATIC int
 323xfs_vn_link(
 324	struct dentry	*old_dentry,
 325	struct inode	*dir,
 326	struct dentry	*dentry)
 327{
 328	struct inode	*inode = d_inode(old_dentry);
 329	struct xfs_name	name;
 330	int		error;
 331
 332	error = xfs_dentry_mode_to_name(&name, dentry, inode->i_mode);
 333	if (unlikely(error))
 334		return error;
 335
 336	error = xfs_link(XFS_I(dir), XFS_I(inode), &name);
 337	if (unlikely(error))
 338		return error;
 339
 340	ihold(inode);
 341	d_instantiate(dentry, inode);
 342	return 0;
 343}
 344
 345STATIC int
 346xfs_vn_unlink(
 347	struct inode	*dir,
 348	struct dentry	*dentry)
 349{
 350	struct xfs_name	name;
 351	int		error;
 352
 353	xfs_dentry_to_name(&name, dentry);
 354
 355	error = xfs_remove(XFS_I(dir), &name, XFS_I(d_inode(dentry)));
 356	if (error)
 357		return error;
 358
 359	/*
 360	 * With unlink, the VFS makes the dentry "negative": no inode,
 361	 * but still hashed. This is incompatible with case-insensitive
 362	 * mode, so invalidate (unhash) the dentry in CI-mode.
 363	 */
 364	if (xfs_sb_version_hasasciici(&XFS_M(dir->i_sb)->m_sb))
 365		d_invalidate(dentry);
 366	return 0;
 367}
 368
 369STATIC int
 370xfs_vn_symlink(
 371	struct inode	*dir,
 372	struct dentry	*dentry,
 373	const char	*symname)
 374{
 375	struct inode	*inode;
 376	struct xfs_inode *cip = NULL;
 377	struct xfs_name	name;
 378	int		error;
 379	umode_t		mode;
 380
 381	mode = S_IFLNK |
 382		(irix_symlink_mode ? 0777 & ~current_umask() : S_IRWXUGO);
 383	error = xfs_dentry_mode_to_name(&name, dentry, mode);
 384	if (unlikely(error))
 385		goto out;
 386
 387	error = xfs_symlink(XFS_I(dir), &name, symname, mode, &cip);
 388	if (unlikely(error))
 389		goto out;
 390
 391	inode = VFS_I(cip);
 392
 393	error = xfs_init_security(inode, dir, &dentry->d_name);
 394	if (unlikely(error))
 395		goto out_cleanup_inode;
 396
 397	xfs_setup_iops(cip);
 398
 399	d_instantiate(dentry, inode);
 400	xfs_finish_inode_setup(cip);
 401	return 0;
 402
 403 out_cleanup_inode:
 404	xfs_finish_inode_setup(cip);
 405	xfs_cleanup_inode(dir, inode, dentry);
 406	iput(inode);
 407 out:
 408	return error;
 409}
 410
 411STATIC int
 412xfs_vn_rename(
 413	struct inode	*odir,
 414	struct dentry	*odentry,
 415	struct inode	*ndir,
 416	struct dentry	*ndentry,
 417	unsigned int	flags)
 418{
 419	struct inode	*new_inode = d_inode(ndentry);
 420	int		omode = 0;
 421	int		error;
 422	struct xfs_name	oname;
 423	struct xfs_name	nname;
 424
 425	if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
 426		return -EINVAL;
 427
 428	/* if we are exchanging files, we need to set i_mode of both files */
 429	if (flags & RENAME_EXCHANGE)
 430		omode = d_inode(ndentry)->i_mode;
 431
 432	error = xfs_dentry_mode_to_name(&oname, odentry, omode);
 433	if (omode && unlikely(error))
 434		return error;
 435
 436	error = xfs_dentry_mode_to_name(&nname, ndentry,
 437					d_inode(odentry)->i_mode);
 438	if (unlikely(error))
 439		return error;
 440
 441	return xfs_rename(XFS_I(odir), &oname, XFS_I(d_inode(odentry)),
 442			  XFS_I(ndir), &nname,
 443			  new_inode ? XFS_I(new_inode) : NULL, flags);
 444}
 445
 446/*
 447 * careful here - this function can get called recursively, so
 448 * we need to be very careful about how much stack we use.
 449 * uio is kmalloced for this reason...
 450 */
 451STATIC const char *
 452xfs_vn_get_link(
 453	struct dentry		*dentry,
 454	struct inode		*inode,
 455	struct delayed_call	*done)
 456{
 457	char			*link;
 458	int			error = -ENOMEM;
 459
 460	if (!dentry)
 461		return ERR_PTR(-ECHILD);
 462
 463	link = kmalloc(MAXPATHLEN+1, GFP_KERNEL);
 464	if (!link)
 465		goto out_err;
 466
 467	error = xfs_readlink(XFS_I(d_inode(dentry)), link);
 468	if (unlikely(error))
 469		goto out_kfree;
 470
 471	set_delayed_call(done, kfree_link, link);
 472	return link;
 473
 474 out_kfree:
 475	kfree(link);
 476 out_err:
 477	return ERR_PTR(error);
 
 478}
 479
 480STATIC const char *
 481xfs_vn_get_link_inline(
 482	struct dentry		*dentry,
 483	struct inode		*inode,
 484	struct delayed_call	*done)
 485{
 486	ASSERT(XFS_I(inode)->i_df.if_flags & XFS_IFINLINE);
 487	return XFS_I(inode)->i_df.if_u1.if_data;
 
 
 488}
 489
 490STATIC int
 491xfs_vn_getattr(
 492	struct vfsmount		*mnt,
 493	struct dentry		*dentry,
 494	struct kstat		*stat)
 495{
 496	struct inode		*inode = d_inode(dentry);
 497	struct xfs_inode	*ip = XFS_I(inode);
 498	struct xfs_mount	*mp = ip->i_mount;
 499
 500	trace_xfs_getattr(ip);
 501
 502	if (XFS_FORCED_SHUTDOWN(mp))
 503		return -EIO;
 504
 505	stat->size = XFS_ISIZE(ip);
 506	stat->dev = inode->i_sb->s_dev;
 507	stat->mode = inode->i_mode;
 508	stat->nlink = inode->i_nlink;
 509	stat->uid = inode->i_uid;
 510	stat->gid = inode->i_gid;
 511	stat->ino = ip->i_ino;
 512	stat->atime = inode->i_atime;
 513	stat->mtime = inode->i_mtime;
 514	stat->ctime = inode->i_ctime;
 515	stat->blocks =
 516		XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks);
 517
 518
 519	switch (inode->i_mode & S_IFMT) {
 520	case S_IFBLK:
 521	case S_IFCHR:
 522		stat->blksize = BLKDEV_IOSIZE;
 523		stat->rdev = MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff,
 524				   sysv_minor(ip->i_df.if_u2.if_rdev));
 525		break;
 526	default:
 527		if (XFS_IS_REALTIME_INODE(ip)) {
 528			/*
 529			 * If the file blocks are being allocated from a
 530			 * realtime volume, then return the inode's realtime
 531			 * extent size or the realtime volume's extent size.
 532			 */
 533			stat->blksize =
 534				xfs_get_extsz_hint(ip) << mp->m_sb.sb_blocklog;
 535		} else
 536			stat->blksize = xfs_preferred_iosize(mp);
 537		stat->rdev = 0;
 538		break;
 539	}
 540
 541	return 0;
 542}
 543
 544static void
 545xfs_setattr_mode(
 546	struct xfs_inode	*ip,
 547	struct iattr		*iattr)
 548{
 549	struct inode		*inode = VFS_I(ip);
 550	umode_t			mode = iattr->ia_mode;
 551
 552	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
 553
 554	inode->i_mode &= S_IFMT;
 555	inode->i_mode |= mode & ~S_IFMT;
 556}
 557
 558void
 559xfs_setattr_time(
 560	struct xfs_inode	*ip,
 561	struct iattr		*iattr)
 562{
 563	struct inode		*inode = VFS_I(ip);
 564
 565	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
 566
 567	if (iattr->ia_valid & ATTR_ATIME)
 568		inode->i_atime = iattr->ia_atime;
 569	if (iattr->ia_valid & ATTR_CTIME)
 570		inode->i_ctime = iattr->ia_ctime;
 571	if (iattr->ia_valid & ATTR_MTIME)
 572		inode->i_mtime = iattr->ia_mtime;
 573}
 574
 575static int
 576xfs_vn_change_ok(
 577	struct dentry	*dentry,
 578	struct iattr	*iattr)
 579{
 580	struct xfs_mount	*mp = XFS_I(d_inode(dentry))->i_mount;
 581
 582	if (mp->m_flags & XFS_MOUNT_RDONLY)
 583		return -EROFS;
 584
 585	if (XFS_FORCED_SHUTDOWN(mp))
 586		return -EIO;
 587
 588	return setattr_prepare(dentry, iattr);
 589}
 590
 591/*
 592 * Set non-size attributes of an inode.
 593 *
 594 * Caution: The caller of this function is responsible for calling
 595 * setattr_prepare() or otherwise verifying the change is fine.
 596 */
 597int
 598xfs_setattr_nonsize(
 599	struct xfs_inode	*ip,
 600	struct iattr		*iattr,
 601	int			flags)
 602{
 603	xfs_mount_t		*mp = ip->i_mount;
 604	struct inode		*inode = VFS_I(ip);
 605	int			mask = iattr->ia_valid;
 606	xfs_trans_t		*tp;
 607	int			error;
 608	kuid_t			uid = GLOBAL_ROOT_UID, iuid = GLOBAL_ROOT_UID;
 609	kgid_t			gid = GLOBAL_ROOT_GID, igid = GLOBAL_ROOT_GID;
 610	struct xfs_dquot	*udqp = NULL, *gdqp = NULL;
 611	struct xfs_dquot	*olddquot1 = NULL, *olddquot2 = NULL;
 612
 
 
 
 
 
 
 
 
 
 
 
 
 613	ASSERT((mask & ATTR_SIZE) == 0);
 614
 615	/*
 616	 * If disk quotas is on, we make sure that the dquots do exist on disk,
 617	 * before we start any other transactions. Trying to do this later
 618	 * is messy. We don't care to take a readlock to look at the ids
 619	 * in inode here, because we can't hold it across the trans_reserve.
 620	 * If the IDs do change before we take the ilock, we're covered
 621	 * because the i_*dquot fields will get updated anyway.
 622	 */
 623	if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) {
 624		uint	qflags = 0;
 625
 626		if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) {
 627			uid = iattr->ia_uid;
 628			qflags |= XFS_QMOPT_UQUOTA;
 629		} else {
 630			uid = inode->i_uid;
 631		}
 632		if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) {
 633			gid = iattr->ia_gid;
 634			qflags |= XFS_QMOPT_GQUOTA;
 635		}  else {
 636			gid = inode->i_gid;
 637		}
 638
 639		/*
 640		 * We take a reference when we initialize udqp and gdqp,
 641		 * so it is important that we never blindly double trip on
 642		 * the same variable. See xfs_create() for an example.
 643		 */
 644		ASSERT(udqp == NULL);
 645		ASSERT(gdqp == NULL);
 646		error = xfs_qm_vop_dqalloc(ip, xfs_kuid_to_uid(uid),
 647					   xfs_kgid_to_gid(gid),
 648					   xfs_get_projid(ip),
 649					   qflags, &udqp, &gdqp, NULL);
 650		if (error)
 651			return error;
 652	}
 653
 654	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ichange, 0, 0, 0, &tp);
 
 655	if (error)
 656		goto out_dqrele;
 657
 658	xfs_ilock(ip, XFS_ILOCK_EXCL);
 659	xfs_trans_ijoin(tp, ip, 0);
 660
 661	/*
 662	 * Change file ownership.  Must be the owner or privileged.
 663	 */
 664	if (mask & (ATTR_UID|ATTR_GID)) {
 665		/*
 666		 * These IDs could have changed since we last looked at them.
 667		 * But, we're assured that if the ownership did change
 668		 * while we didn't have the inode locked, inode's dquot(s)
 669		 * would have changed also.
 670		 */
 671		iuid = inode->i_uid;
 672		igid = inode->i_gid;
 673		gid = (mask & ATTR_GID) ? iattr->ia_gid : igid;
 674		uid = (mask & ATTR_UID) ? iattr->ia_uid : iuid;
 675
 676		/*
 677		 * Do a quota reservation only if uid/gid is actually
 678		 * going to change.
 679		 */
 680		if (XFS_IS_QUOTA_RUNNING(mp) &&
 681		    ((XFS_IS_UQUOTA_ON(mp) && !uid_eq(iuid, uid)) ||
 682		     (XFS_IS_GQUOTA_ON(mp) && !gid_eq(igid, gid)))) {
 683			ASSERT(tp);
 684			error = xfs_qm_vop_chown_reserve(tp, ip, udqp, gdqp,
 685						NULL, capable(CAP_FOWNER) ?
 686						XFS_QMOPT_FORCE_RES : 0);
 687			if (error)	/* out of quota */
 688				goto out_cancel;
 689		}
 690	}
 691
 
 
 692	/*
 693	 * Change file ownership.  Must be the owner or privileged.
 694	 */
 695	if (mask & (ATTR_UID|ATTR_GID)) {
 696		/*
 697		 * CAP_FSETID overrides the following restrictions:
 698		 *
 699		 * The set-user-ID and set-group-ID bits of a file will be
 700		 * cleared upon successful return from chown()
 701		 */
 702		if ((inode->i_mode & (S_ISUID|S_ISGID)) &&
 703		    !capable(CAP_FSETID))
 704			inode->i_mode &= ~(S_ISUID|S_ISGID);
 705
 706		/*
 707		 * Change the ownerships and register quota modifications
 708		 * in the transaction.
 709		 */
 710		if (!uid_eq(iuid, uid)) {
 711			if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_UQUOTA_ON(mp)) {
 712				ASSERT(mask & ATTR_UID);
 713				ASSERT(udqp);
 714				olddquot1 = xfs_qm_vop_chown(tp, ip,
 715							&ip->i_udquot, udqp);
 716			}
 717			ip->i_d.di_uid = xfs_kuid_to_uid(uid);
 718			inode->i_uid = uid;
 719		}
 720		if (!gid_eq(igid, gid)) {
 721			if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_GQUOTA_ON(mp)) {
 722				ASSERT(xfs_sb_version_has_pquotino(&mp->m_sb) ||
 723				       !XFS_IS_PQUOTA_ON(mp));
 724				ASSERT(mask & ATTR_GID);
 725				ASSERT(gdqp);
 726				olddquot2 = xfs_qm_vop_chown(tp, ip,
 727							&ip->i_gdquot, gdqp);
 728			}
 729			ip->i_d.di_gid = xfs_kgid_to_gid(gid);
 730			inode->i_gid = gid;
 731		}
 732	}
 733
 734	if (mask & ATTR_MODE)
 735		xfs_setattr_mode(ip, iattr);
 736	if (mask & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME))
 737		xfs_setattr_time(ip, iattr);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 738
 739	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
 740
 741	XFS_STATS_INC(mp, xs_ig_attrchg);
 742
 743	if (mp->m_flags & XFS_MOUNT_WSYNC)
 744		xfs_trans_set_sync(tp);
 745	error = xfs_trans_commit(tp);
 746
 747	xfs_iunlock(ip, XFS_ILOCK_EXCL);
 748
 749	/*
 750	 * Release any dquot(s) the inode had kept before chown.
 751	 */
 752	xfs_qm_dqrele(olddquot1);
 753	xfs_qm_dqrele(olddquot2);
 754	xfs_qm_dqrele(udqp);
 755	xfs_qm_dqrele(gdqp);
 756
 757	if (error)
 758		return error;
 759
 760	/*
 761	 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode
 762	 * 	     update.  We could avoid this with linked transactions
 763	 * 	     and passing down the transaction pointer all the way
 764	 *	     to attr_set.  No previous user of the generic
 765	 * 	     Posix ACL code seems to care about this issue either.
 766	 */
 767	if ((mask & ATTR_MODE) && !(flags & XFS_ATTR_NOACL)) {
 768		error = posix_acl_chmod(inode, inode->i_mode);
 769		if (error)
 770			return error;
 771	}
 772
 773	return 0;
 774
 775out_cancel:
 776	xfs_trans_cancel(tp);
 
 777out_dqrele:
 778	xfs_qm_dqrele(udqp);
 779	xfs_qm_dqrele(gdqp);
 780	return error;
 781}
 782
 783int
 784xfs_vn_setattr_nonsize(
 785	struct dentry		*dentry,
 786	struct iattr		*iattr)
 787{
 788	struct xfs_inode	*ip = XFS_I(d_inode(dentry));
 789	int error;
 790
 791	trace_xfs_setattr(ip);
 792
 793	error = xfs_vn_change_ok(dentry, iattr);
 794	if (error)
 795		return error;
 796	return xfs_setattr_nonsize(ip, iattr, 0);
 797}
 798
 799/*
 800 * Truncate file.  Must have write permission and not be a directory.
 801 *
 802 * Caution: The caller of this function is responsible for calling
 803 * setattr_prepare() or otherwise verifying the change is fine.
 804 */
 805int
 806xfs_setattr_size(
 807	struct xfs_inode	*ip,
 808	struct iattr		*iattr)
 
 809{
 810	struct xfs_mount	*mp = ip->i_mount;
 811	struct inode		*inode = VFS_I(ip);
 812	xfs_off_t		oldsize, newsize;
 813	struct xfs_trans	*tp;
 814	int			error;
 815	uint			lock_flags = 0;
 816	bool			did_zeroing = false;
 817
 818	ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
 819	ASSERT(xfs_isilocked(ip, XFS_MMAPLOCK_EXCL));
 820	ASSERT(S_ISREG(inode->i_mode));
 821	ASSERT((iattr->ia_valid & (ATTR_UID|ATTR_GID|ATTR_ATIME|ATTR_ATIME_SET|
 822		ATTR_MTIME_SET|ATTR_KILL_PRIV|ATTR_TIMES_SET)) == 0);
 823
 824	oldsize = inode->i_size;
 825	newsize = iattr->ia_size;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 826
 827	/*
 828	 * Short circuit the truncate case for zero length files.
 829	 */
 830	if (newsize == 0 && oldsize == 0 && ip->i_d.di_nextents == 0) {
 831		if (!(iattr->ia_valid & (ATTR_CTIME|ATTR_MTIME)))
 832			return 0;
 
 833
 834		/*
 835		 * Use the regular setattr path to update the timestamps.
 836		 */
 
 837		iattr->ia_valid &= ~ATTR_SIZE;
 838		return xfs_setattr_nonsize(ip, iattr, 0);
 839	}
 840
 841	/*
 842	 * Make sure that the dquots are attached to the inode.
 843	 */
 844	error = xfs_qm_dqattach(ip, 0);
 845	if (error)
 846		return error;
 847
 848	/*
 849	 * Wait for all direct I/O to complete.
 
 
 
 
 850	 */
 851	inode_dio_wait(inode);
 852
 853	/*
 854	 * File data changes must be complete before we start the transaction to
 855	 * modify the inode.  This needs to be done before joining the inode to
 856	 * the transaction because the inode cannot be unlocked once it is a
 857	 * part of the transaction.
 858	 *
 859	 * Start with zeroing any data beyond EOF that we may expose on file
 860	 * extension, or zeroing out the rest of the block on a downward
 861	 * truncate.
 862	 */
 863	if (newsize > oldsize) {
 864		error = xfs_zero_eof(ip, newsize, oldsize, &did_zeroing);
 865	} else {
 866		error = iomap_truncate_page(inode, newsize, &did_zeroing,
 867				&xfs_iomap_ops);
 868	}
 869
 870	if (error)
 871		return error;
 872
 873	/*
 874	 * We are going to log the inode size change in this transaction so
 875	 * any previous writes that are beyond the on disk EOF and the new
 876	 * EOF that have not been written out need to be written here.  If we
 877	 * do not write the data out, we expose ourselves to the null files
 878	 * problem. Note that this includes any block zeroing we did above;
 879	 * otherwise those blocks may not be zeroed after a crash.
 
 
 
 
 880	 */
 881	if (did_zeroing ||
 882	    (newsize > ip->i_d.di_size && oldsize != ip->i_d.di_size)) {
 883		error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
 884						      ip->i_d.di_size, newsize);
 885		if (error)
 886			return error;
 887	}
 888
 889	/*
 890	 * We've already locked out new page faults, so now we can safely remove
 891	 * pages from the page cache knowing they won't get refaulted until we
 892	 * drop the XFS_MMAP_EXCL lock after the extent manipulations are
 893	 * complete. The truncate_setsize() call also cleans partial EOF page
 894	 * PTEs on extending truncates and hence ensures sub-page block size
 895	 * filesystems are correctly handled, too.
 896	 *
 897	 * We have to do all the page cache truncate work outside the
 898	 * transaction context as the "lock" order is page lock->log space
 899	 * reservation as defined by extent allocation in the writeback path.
 900	 * Hence a truncate can fail with ENOMEM from xfs_trans_alloc(), but
 901	 * having already truncated the in-memory version of the file (i.e. made
 902	 * user visible changes). There's not much we can do about this, except
 903	 * to hope that the caller sees ENOMEM and retries the truncate
 904	 * operation.
 905	 */
 906	truncate_setsize(inode, newsize);
 
 
 
 
 
 907
 908	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
 
 
 
 909	if (error)
 910		return error;
 
 
 911
 
 912	lock_flags |= XFS_ILOCK_EXCL;
 
 913	xfs_ilock(ip, XFS_ILOCK_EXCL);
 914	xfs_trans_ijoin(tp, ip, 0);
 
 915
 916	/*
 917	 * Only change the c/mtime if we are changing the size or we are
 918	 * explicitly asked to change it.  This handles the semantic difference
 919	 * between truncate() and ftruncate() as implemented in the VFS.
 920	 *
 921	 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
 922	 * special case where we need to update the times despite not having
 923	 * these flags set.  For all other operations the VFS set these flags
 924	 * explicitly if it wants a timestamp update.
 925	 */
 926	if (newsize != oldsize &&
 927	    !(iattr->ia_valid & (ATTR_CTIME | ATTR_MTIME))) {
 928		iattr->ia_ctime = iattr->ia_mtime =
 929			current_time(inode);
 930		iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME;
 931	}
 932
 933	/*
 934	 * The first thing we do is set the size to new_size permanently on
 935	 * disk.  This way we don't have to worry about anyone ever being able
 936	 * to look at the data being freed even in the face of a crash.
 937	 * What we're getting around here is the case where we free a block, it
 938	 * is allocated to another file, it is written to, and then we crash.
 939	 * If the new data gets written to the file but the log buffers
 940	 * containing the free and reallocation don't, then we'd end up with
 941	 * garbage in the blocks being freed.  As long as we make the new size
 942	 * permanent before actually freeing any blocks it doesn't matter if
 943	 * they get written to.
 944	 */
 945	ip->i_d.di_size = newsize;
 946	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
 947
 948	if (newsize <= oldsize) {
 949		error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, newsize);
 950		if (error)
 951			goto out_trans_cancel;
 952
 953		/*
 954		 * Truncated "down", so we're removing references to old data
 955		 * here - if we delay flushing for a long time, we expose
 956		 * ourselves unduly to the notorious NULL files problem.  So,
 957		 * we mark this inode and flush it when the file is closed,
 958		 * and do not wait the usual (long) time for writeout.
 959		 */
 960		xfs_iflags_set(ip, XFS_ITRUNCATED);
 
 961
 962		/* A truncate down always removes post-EOF blocks. */
 963		xfs_inode_clear_eofblocks_tag(ip);
 
 
 
 
 
 
 
 
 
 964	}
 965
 966	if (iattr->ia_valid & ATTR_MODE)
 967		xfs_setattr_mode(ip, iattr);
 968	if (iattr->ia_valid & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME))
 969		xfs_setattr_time(ip, iattr);
 970
 971	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
 972
 973	XFS_STATS_INC(mp, xs_ig_attrchg);
 974
 975	if (mp->m_flags & XFS_MOUNT_WSYNC)
 976		xfs_trans_set_sync(tp);
 977
 978	error = xfs_trans_commit(tp);
 979out_unlock:
 980	if (lock_flags)
 981		xfs_iunlock(ip, lock_flags);
 982	return error;
 983
 
 
 984out_trans_cancel:
 985	xfs_trans_cancel(tp);
 986	goto out_unlock;
 987}
 988
 989int
 990xfs_vn_setattr_size(
 991	struct dentry		*dentry,
 992	struct iattr		*iattr)
 993{
 994	struct xfs_inode	*ip = XFS_I(d_inode(dentry));
 995	int error;
 996
 997	trace_xfs_setattr(ip);
 998
 999	error = xfs_vn_change_ok(dentry, iattr);
1000	if (error)
1001		return error;
1002	return xfs_setattr_size(ip, iattr);
1003}
1004
1005STATIC int
1006xfs_vn_setattr(
1007	struct dentry		*dentry,
1008	struct iattr		*iattr)
1009{
1010	int			error;
 
 
 
1011
1012	if (iattr->ia_valid & ATTR_SIZE) {
1013		struct xfs_inode	*ip = XFS_I(d_inode(dentry));
1014		uint			iolock = XFS_IOLOCK_EXCL;
1015
1016		error = xfs_break_layouts(d_inode(dentry), &iolock);
1017		if (error)
1018			return error;
1019
1020		xfs_ilock(ip, XFS_MMAPLOCK_EXCL);
1021		error = xfs_vn_setattr_size(dentry, iattr);
1022		xfs_iunlock(ip, XFS_MMAPLOCK_EXCL);
1023	} else {
1024		error = xfs_vn_setattr_nonsize(dentry, iattr);
1025	}
1026
1027	return error;
1028}
1029
 
 
 
 
1030STATIC int
1031xfs_vn_update_time(
1032	struct inode		*inode,
1033	struct timespec		*now,
1034	int			flags)
1035{
1036	struct xfs_inode	*ip = XFS_I(inode);
1037	struct xfs_mount	*mp = ip->i_mount;
1038	struct xfs_trans	*tp;
1039	int			error;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1040
1041	trace_xfs_update_time(ip);
1042
1043	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp);
1044	if (error)
1045		return error;
1046
1047	xfs_ilock(ip, XFS_ILOCK_EXCL);
1048	if (flags & S_CTIME)
1049		inode->i_ctime = *now;
1050	if (flags & S_MTIME)
1051		inode->i_mtime = *now;
1052	if (flags & S_ATIME)
1053		inode->i_atime = *now;
1054
1055	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1056	xfs_trans_log_inode(tp, ip, XFS_ILOG_TIMESTAMP);
1057	return xfs_trans_commit(tp);
1058}
1059
1060STATIC int
1061xfs_vn_fiemap(
1062	struct inode		*inode,
1063	struct fiemap_extent_info *fieinfo,
1064	u64			start,
1065	u64			length)
1066{
 
 
1067	int			error;
1068
1069	xfs_ilock(XFS_I(inode), XFS_IOLOCK_SHARED);
1070	if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
1071		fieinfo->fi_flags &= ~FIEMAP_FLAG_XATTR;
1072		error = iomap_fiemap(inode, fieinfo, start, length,
1073				&xfs_xattr_iomap_ops);
1074	} else {
1075		error = iomap_fiemap(inode, fieinfo, start, length,
1076				&xfs_iomap_ops);
1077	}
1078	xfs_iunlock(XFS_I(inode), XFS_IOLOCK_SHARED);
 
 
 
 
 
 
 
 
 
 
 
 
1079
1080	return error;
1081}
 
1082
1083STATIC int
1084xfs_vn_tmpfile(
1085	struct inode	*dir,
1086	struct dentry	*dentry,
1087	umode_t		mode)
1088{
1089	return xfs_generic_create(dir, dentry, mode, 0, true);
1090}
1091
1092static const struct inode_operations xfs_inode_operations = {
1093	.get_acl		= xfs_get_acl,
1094	.set_acl		= xfs_set_acl,
1095	.getattr		= xfs_vn_getattr,
1096	.setattr		= xfs_vn_setattr,
 
 
 
1097	.listxattr		= xfs_vn_listxattr,
1098	.fiemap			= xfs_vn_fiemap,
1099	.update_time		= xfs_vn_update_time,
1100};
1101
1102static const struct inode_operations xfs_dir_inode_operations = {
1103	.create			= xfs_vn_create,
1104	.lookup			= xfs_vn_lookup,
1105	.link			= xfs_vn_link,
1106	.unlink			= xfs_vn_unlink,
1107	.symlink		= xfs_vn_symlink,
1108	.mkdir			= xfs_vn_mkdir,
1109	/*
1110	 * Yes, XFS uses the same method for rmdir and unlink.
1111	 *
1112	 * There are some subtile differences deeper in the code,
1113	 * but we use S_ISDIR to check for those.
1114	 */
1115	.rmdir			= xfs_vn_unlink,
1116	.mknod			= xfs_vn_mknod,
1117	.rename			= xfs_vn_rename,
1118	.get_acl		= xfs_get_acl,
1119	.set_acl		= xfs_set_acl,
1120	.getattr		= xfs_vn_getattr,
1121	.setattr		= xfs_vn_setattr,
 
 
 
1122	.listxattr		= xfs_vn_listxattr,
1123	.update_time		= xfs_vn_update_time,
1124	.tmpfile		= xfs_vn_tmpfile,
1125};
1126
1127static const struct inode_operations xfs_dir_ci_inode_operations = {
1128	.create			= xfs_vn_create,
1129	.lookup			= xfs_vn_ci_lookup,
1130	.link			= xfs_vn_link,
1131	.unlink			= xfs_vn_unlink,
1132	.symlink		= xfs_vn_symlink,
1133	.mkdir			= xfs_vn_mkdir,
1134	/*
1135	 * Yes, XFS uses the same method for rmdir and unlink.
1136	 *
1137	 * There are some subtile differences deeper in the code,
1138	 * but we use S_ISDIR to check for those.
1139	 */
1140	.rmdir			= xfs_vn_unlink,
1141	.mknod			= xfs_vn_mknod,
1142	.rename			= xfs_vn_rename,
1143	.get_acl		= xfs_get_acl,
1144	.set_acl		= xfs_set_acl,
1145	.getattr		= xfs_vn_getattr,
1146	.setattr		= xfs_vn_setattr,
 
 
 
1147	.listxattr		= xfs_vn_listxattr,
1148	.update_time		= xfs_vn_update_time,
1149	.tmpfile		= xfs_vn_tmpfile,
1150};
1151
1152static const struct inode_operations xfs_symlink_inode_operations = {
1153	.get_link		= xfs_vn_get_link,
 
 
 
1154	.getattr		= xfs_vn_getattr,
1155	.setattr		= xfs_vn_setattr,
 
 
 
1156	.listxattr		= xfs_vn_listxattr,
1157	.update_time		= xfs_vn_update_time,
1158};
1159
1160static const struct inode_operations xfs_inline_symlink_inode_operations = {
1161	.get_link		= xfs_vn_get_link_inline,
1162	.getattr		= xfs_vn_getattr,
1163	.setattr		= xfs_vn_setattr,
1164	.listxattr		= xfs_vn_listxattr,
1165	.update_time		= xfs_vn_update_time,
1166};
1167
1168STATIC void
1169xfs_diflags_to_iflags(
1170	struct inode		*inode,
1171	struct xfs_inode	*ip)
1172{
1173	uint16_t		flags = ip->i_d.di_flags;
1174
1175	inode->i_flags &= ~(S_IMMUTABLE | S_APPEND | S_SYNC |
1176			    S_NOATIME | S_DAX);
1177
1178	if (flags & XFS_DIFLAG_IMMUTABLE)
1179		inode->i_flags |= S_IMMUTABLE;
1180	if (flags & XFS_DIFLAG_APPEND)
 
 
1181		inode->i_flags |= S_APPEND;
1182	if (flags & XFS_DIFLAG_SYNC)
 
 
1183		inode->i_flags |= S_SYNC;
1184	if (flags & XFS_DIFLAG_NOATIME)
 
 
1185		inode->i_flags |= S_NOATIME;
1186	if (S_ISREG(inode->i_mode) &&
1187	    ip->i_mount->m_sb.sb_blocksize == PAGE_SIZE &&
1188	    !xfs_is_reflink_inode(ip) &&
1189	    (ip->i_mount->m_flags & XFS_MOUNT_DAX ||
1190	     ip->i_d.di_flags2 & XFS_DIFLAG2_DAX))
1191		inode->i_flags |= S_DAX;
1192}
1193
1194/*
1195 * Initialize the Linux inode.
 
1196 *
1197 * When reading existing inodes from disk this is called directly from xfs_iget,
1198 * when creating a new inode it is called from xfs_ialloc after setting up the
1199 * inode. These callers have different criteria for clearing XFS_INEW, so leave
1200 * it up to the caller to deal with unlocking the inode appropriately.
 
 
 
1201 */
1202void
1203xfs_setup_inode(
1204	struct xfs_inode	*ip)
1205{
1206	struct inode		*inode = &ip->i_vnode;
1207	gfp_t			gfp_mask;
1208
1209	inode->i_ino = ip->i_ino;
1210	inode->i_state = I_NEW;
1211
1212	inode_sb_list_add(inode);
1213	/* make the inode look hashed for the writeback code */
1214	hlist_add_fake(&inode->i_hash);
1215
1216	inode->i_uid    = xfs_uid_to_kuid(ip->i_d.di_uid);
1217	inode->i_gid    = xfs_gid_to_kgid(ip->i_d.di_gid);
 
 
1218
1219	switch (inode->i_mode & S_IFMT) {
1220	case S_IFBLK:
1221	case S_IFCHR:
1222		inode->i_rdev =
1223			MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff,
1224			      sysv_minor(ip->i_df.if_u2.if_rdev));
1225		break;
1226	default:
1227		inode->i_rdev = 0;
1228		break;
1229	}
1230
 
1231	i_size_write(inode, ip->i_d.di_size);
 
 
 
 
 
 
1232	xfs_diflags_to_iflags(inode, ip);
1233
1234	if (S_ISDIR(inode->i_mode)) {
1235		lockdep_set_class(&ip->i_lock.mr_lock, &xfs_dir_ilock_class);
1236		ip->d_ops = ip->i_mount->m_dir_inode_ops;
1237	} else {
1238		ip->d_ops = ip->i_mount->m_nondir_inode_ops;
1239		lockdep_set_class(&ip->i_lock.mr_lock, &xfs_nondir_ilock_class);
1240	}
1241
1242	/*
1243	 * Ensure all page cache allocations are done from GFP_NOFS context to
1244	 * prevent direct reclaim recursion back into the filesystem and blowing
1245	 * stacks or deadlocking.
1246	 */
1247	gfp_mask = mapping_gfp_mask(inode->i_mapping);
1248	mapping_set_gfp_mask(inode->i_mapping, (gfp_mask & ~(__GFP_FS)));
1249
1250	/*
1251	 * If there is no attribute fork no ACL can exist on this inode,
1252	 * and it can't have any file capabilities attached to it either.
1253	 */
1254	if (!XFS_IFORK_Q(ip)) {
1255		inode_has_no_xattr(inode);
1256		cache_no_acl(inode);
1257	}
1258}
1259
1260void
1261xfs_setup_iops(
1262	struct xfs_inode	*ip)
1263{
1264	struct inode		*inode = &ip->i_vnode;
1265
1266	switch (inode->i_mode & S_IFMT) {
1267	case S_IFREG:
1268		inode->i_op = &xfs_inode_operations;
1269		inode->i_fop = &xfs_file_operations;
1270		inode->i_mapping->a_ops = &xfs_address_space_operations;
1271		break;
1272	case S_IFDIR:
1273		if (xfs_sb_version_hasasciici(&XFS_M(inode->i_sb)->m_sb))
1274			inode->i_op = &xfs_dir_ci_inode_operations;
1275		else
1276			inode->i_op = &xfs_dir_inode_operations;
1277		inode->i_fop = &xfs_dir_file_operations;
1278		break;
1279	case S_IFLNK:
1280		if (ip->i_df.if_flags & XFS_IFINLINE)
1281			inode->i_op = &xfs_inline_symlink_inode_operations;
1282		else
1283			inode->i_op = &xfs_symlink_inode_operations;
1284		break;
1285	default:
1286		inode->i_op = &xfs_inode_operations;
1287		init_special_inode(inode, inode->i_mode, inode->i_rdev);
1288		break;
1289	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1290}