Linux Audio

Check our new training course

Loading...
v4.6
   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
  42#include <linux/capability.h>
  43#include <linux/xattr.h>
  44#include <linux/posix_acl.h>
  45#include <linux/security.h>
  46#include <linux/fiemap.h>
  47#include <linux/slab.h>
 
  48
  49/*
  50 * Directories have different lock order w.r.t. mmap_sem compared to regular
  51 * files. This is due to readdir potentially triggering page faults on a user
  52 * buffer inside filldir(), and this happens with the ilock on the directory
  53 * held. For regular files, the lock order is the other way around - the
  54 * mmap_sem is taken during the page fault, and then we lock the ilock to do
  55 * block mapping. Hence we need a different class for the directory ilock so
  56 * that lockdep can tell them apart.
  57 */
  58static struct lock_class_key xfs_nondir_ilock_class;
  59static struct lock_class_key xfs_dir_ilock_class;
  60
  61static int
  62xfs_initxattrs(
  63	struct inode		*inode,
  64	const struct xattr	*xattr_array,
  65	void			*fs_info)
  66{
  67	const struct xattr	*xattr;
  68	struct xfs_inode	*ip = XFS_I(inode);
  69	int			error = 0;
  70
  71	for (xattr = xattr_array; xattr->name != NULL; xattr++) {
  72		error = xfs_attr_set(ip, xattr->name, xattr->value,
  73				      xattr->value_len, ATTR_SECURE);
  74		if (error < 0)
  75			break;
  76	}
  77	return error;
  78}
  79
  80/*
  81 * Hook in SELinux.  This is not quite correct yet, what we really need
  82 * here (as we do for default ACLs) is a mechanism by which creation of
  83 * these attrs can be journalled at inode creation time (along with the
  84 * inode, of course, such that log replay can't cause these to be lost).
  85 */
  86
  87STATIC int
  88xfs_init_security(
  89	struct inode	*inode,
  90	struct inode	*dir,
  91	const struct qstr *qstr)
  92{
  93	return security_inode_init_security(inode, dir, qstr,
  94					     &xfs_initxattrs, NULL);
  95}
  96
  97static void
  98xfs_dentry_to_name(
  99	struct xfs_name	*namep,
 
 
 
 
 
 
 
 
 
 
 100	struct dentry	*dentry,
 101	int		mode)
 102{
 103	namep->name = dentry->d_name.name;
 104	namep->len = dentry->d_name.len;
 105	namep->type = xfs_mode_to_ftype[(mode & S_IFMT) >> S_SHIFT];
 
 
 
 
 
 106}
 107
 108STATIC void
 109xfs_cleanup_inode(
 110	struct inode	*dir,
 111	struct inode	*inode,
 112	struct dentry	*dentry)
 113{
 114	struct xfs_name	teardown;
 115
 116	/* Oh, the horror.
 117	 * If we can't add the ACL or we fail in
 118	 * xfs_init_security we must back out.
 119	 * ENOSPC can hit here, among other things.
 120	 */
 121	xfs_dentry_to_name(&teardown, dentry, 0);
 122
 123	xfs_remove(XFS_I(dir), &teardown, XFS_I(inode));
 124}
 125
 126STATIC int
 127xfs_generic_create(
 128	struct inode	*dir,
 129	struct dentry	*dentry,
 130	umode_t		mode,
 131	dev_t		rdev,
 132	bool		tmpfile)	/* unnamed file */
 133{
 134	struct inode	*inode;
 135	struct xfs_inode *ip = NULL;
 136	struct posix_acl *default_acl, *acl;
 137	struct xfs_name	name;
 138	int		error;
 139
 140	/*
 141	 * Irix uses Missed'em'V split, but doesn't want to see
 142	 * the upper 5 bits of (14bit) major.
 143	 */
 144	if (S_ISCHR(mode) || S_ISBLK(mode)) {
 145		if (unlikely(!sysv_valid_dev(rdev) || MAJOR(rdev) & ~0x1ff))
 146			return -EINVAL;
 147		rdev = sysv_encode_dev(rdev);
 148	} else {
 149		rdev = 0;
 150	}
 151
 152	error = posix_acl_create(dir, &mode, &default_acl, &acl);
 153	if (error)
 154		return error;
 155
 
 
 
 
 
 156	if (!tmpfile) {
 157		xfs_dentry_to_name(&name, dentry, mode);
 158		error = xfs_create(XFS_I(dir), &name, mode, rdev, &ip);
 159	} else {
 160		error = xfs_create_tmpfile(XFS_I(dir), dentry, mode, &ip);
 161	}
 162	if (unlikely(error))
 163		goto out_free_acl;
 164
 165	inode = VFS_I(ip);
 166
 167	error = xfs_init_security(inode, dir, &dentry->d_name);
 168	if (unlikely(error))
 169		goto out_cleanup_inode;
 170
 171#ifdef CONFIG_XFS_POSIX_ACL
 172	if (default_acl) {
 173		error = xfs_set_acl(inode, default_acl, ACL_TYPE_DEFAULT);
 174		if (error)
 175			goto out_cleanup_inode;
 176	}
 177	if (acl) {
 178		error = xfs_set_acl(inode, acl, ACL_TYPE_ACCESS);
 179		if (error)
 180			goto out_cleanup_inode;
 181	}
 182#endif
 183
 
 
 184	if (tmpfile)
 185		d_tmpfile(dentry, inode);
 186	else
 187		d_instantiate(dentry, inode);
 188
 189	xfs_finish_inode_setup(ip);
 190
 191 out_free_acl:
 192	if (default_acl)
 193		posix_acl_release(default_acl);
 194	if (acl)
 195		posix_acl_release(acl);
 196	return error;
 197
 198 out_cleanup_inode:
 199	xfs_finish_inode_setup(ip);
 200	if (!tmpfile)
 201		xfs_cleanup_inode(dir, inode, dentry);
 202	iput(inode);
 203	goto out_free_acl;
 204}
 205
 206STATIC int
 207xfs_vn_mknod(
 208	struct inode	*dir,
 209	struct dentry	*dentry,
 210	umode_t		mode,
 211	dev_t		rdev)
 212{
 213	return xfs_generic_create(dir, dentry, mode, rdev, false);
 214}
 215
 216STATIC int
 217xfs_vn_create(
 218	struct inode	*dir,
 219	struct dentry	*dentry,
 220	umode_t		mode,
 221	bool		flags)
 222{
 223	return xfs_vn_mknod(dir, dentry, mode, 0);
 224}
 225
 226STATIC int
 227xfs_vn_mkdir(
 228	struct inode	*dir,
 229	struct dentry	*dentry,
 230	umode_t		mode)
 231{
 232	return xfs_vn_mknod(dir, dentry, mode|S_IFDIR, 0);
 233}
 234
 235STATIC struct dentry *
 236xfs_vn_lookup(
 237	struct inode	*dir,
 238	struct dentry	*dentry,
 239	unsigned int flags)
 240{
 241	struct xfs_inode *cip;
 242	struct xfs_name	name;
 243	int		error;
 244
 245	if (dentry->d_name.len >= MAXNAMELEN)
 246		return ERR_PTR(-ENAMETOOLONG);
 247
 248	xfs_dentry_to_name(&name, dentry, 0);
 249	error = xfs_lookup(XFS_I(dir), &name, &cip, NULL);
 250	if (unlikely(error)) {
 251		if (unlikely(error != -ENOENT))
 252			return ERR_PTR(error);
 253		d_add(dentry, NULL);
 254		return NULL;
 255	}
 256
 257	return d_splice_alias(VFS_I(cip), dentry);
 258}
 259
 260STATIC struct dentry *
 261xfs_vn_ci_lookup(
 262	struct inode	*dir,
 263	struct dentry	*dentry,
 264	unsigned int flags)
 265{
 266	struct xfs_inode *ip;
 267	struct xfs_name	xname;
 268	struct xfs_name ci_name;
 269	struct qstr	dname;
 270	int		error;
 271
 272	if (dentry->d_name.len >= MAXNAMELEN)
 273		return ERR_PTR(-ENAMETOOLONG);
 274
 275	xfs_dentry_to_name(&xname, dentry, 0);
 276	error = xfs_lookup(XFS_I(dir), &xname, &ip, &ci_name);
 277	if (unlikely(error)) {
 278		if (unlikely(error != -ENOENT))
 279			return ERR_PTR(error);
 280		/*
 281		 * call d_add(dentry, NULL) here when d_drop_negative_children
 282		 * is called in xfs_vn_mknod (ie. allow negative dentries
 283		 * with CI filesystems).
 284		 */
 285		return NULL;
 286	}
 287
 288	/* if exact match, just splice and exit */
 289	if (!ci_name.name)
 290		return d_splice_alias(VFS_I(ip), dentry);
 291
 292	/* else case-insensitive match... */
 293	dname.name = ci_name.name;
 294	dname.len = ci_name.len;
 295	dentry = d_add_ci(dentry, VFS_I(ip), &dname);
 296	kmem_free(ci_name.name);
 297	return dentry;
 298}
 299
 300STATIC int
 301xfs_vn_link(
 302	struct dentry	*old_dentry,
 303	struct inode	*dir,
 304	struct dentry	*dentry)
 305{
 306	struct inode	*inode = d_inode(old_dentry);
 307	struct xfs_name	name;
 308	int		error;
 309
 310	xfs_dentry_to_name(&name, dentry, inode->i_mode);
 
 
 311
 312	error = xfs_link(XFS_I(dir), XFS_I(inode), &name);
 313	if (unlikely(error))
 314		return error;
 315
 316	ihold(inode);
 317	d_instantiate(dentry, inode);
 318	return 0;
 319}
 320
 321STATIC int
 322xfs_vn_unlink(
 323	struct inode	*dir,
 324	struct dentry	*dentry)
 325{
 326	struct xfs_name	name;
 327	int		error;
 328
 329	xfs_dentry_to_name(&name, dentry, 0);
 330
 331	error = xfs_remove(XFS_I(dir), &name, XFS_I(d_inode(dentry)));
 332	if (error)
 333		return error;
 334
 335	/*
 336	 * With unlink, the VFS makes the dentry "negative": no inode,
 337	 * but still hashed. This is incompatible with case-insensitive
 338	 * mode, so invalidate (unhash) the dentry in CI-mode.
 339	 */
 340	if (xfs_sb_version_hasasciici(&XFS_M(dir->i_sb)->m_sb))
 341		d_invalidate(dentry);
 342	return 0;
 343}
 344
 345STATIC int
 346xfs_vn_symlink(
 347	struct inode	*dir,
 348	struct dentry	*dentry,
 349	const char	*symname)
 350{
 351	struct inode	*inode;
 352	struct xfs_inode *cip = NULL;
 353	struct xfs_name	name;
 354	int		error;
 355	umode_t		mode;
 356
 357	mode = S_IFLNK |
 358		(irix_symlink_mode ? 0777 & ~current_umask() : S_IRWXUGO);
 359	xfs_dentry_to_name(&name, dentry, mode);
 
 
 360
 361	error = xfs_symlink(XFS_I(dir), &name, symname, mode, &cip);
 362	if (unlikely(error))
 363		goto out;
 364
 365	inode = VFS_I(cip);
 366
 367	error = xfs_init_security(inode, dir, &dentry->d_name);
 368	if (unlikely(error))
 369		goto out_cleanup_inode;
 370
 
 
 371	d_instantiate(dentry, inode);
 372	xfs_finish_inode_setup(cip);
 373	return 0;
 374
 375 out_cleanup_inode:
 376	xfs_finish_inode_setup(cip);
 377	xfs_cleanup_inode(dir, inode, dentry);
 378	iput(inode);
 379 out:
 380	return error;
 381}
 382
 383STATIC int
 384xfs_vn_rename(
 385	struct inode	*odir,
 386	struct dentry	*odentry,
 387	struct inode	*ndir,
 388	struct dentry	*ndentry,
 389	unsigned int	flags)
 390{
 391	struct inode	*new_inode = d_inode(ndentry);
 392	int		omode = 0;
 
 393	struct xfs_name	oname;
 394	struct xfs_name	nname;
 395
 396	if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
 397		return -EINVAL;
 398
 399	/* if we are exchanging files, we need to set i_mode of both files */
 400	if (flags & RENAME_EXCHANGE)
 401		omode = d_inode(ndentry)->i_mode;
 402
 403	xfs_dentry_to_name(&oname, odentry, omode);
 404	xfs_dentry_to_name(&nname, ndentry, d_inode(odentry)->i_mode);
 
 
 
 
 
 
 405
 406	return xfs_rename(XFS_I(odir), &oname, XFS_I(d_inode(odentry)),
 407			  XFS_I(ndir), &nname,
 408			  new_inode ? XFS_I(new_inode) : NULL, flags);
 409}
 410
 411/*
 412 * careful here - this function can get called recursively, so
 413 * we need to be very careful about how much stack we use.
 414 * uio is kmalloced for this reason...
 415 */
 416STATIC const char *
 417xfs_vn_get_link(
 418	struct dentry		*dentry,
 419	struct inode		*inode,
 420	struct delayed_call	*done)
 421{
 422	char			*link;
 423	int			error = -ENOMEM;
 424
 425	if (!dentry)
 426		return ERR_PTR(-ECHILD);
 427
 428	link = kmalloc(MAXPATHLEN+1, GFP_KERNEL);
 429	if (!link)
 430		goto out_err;
 431
 432	error = xfs_readlink(XFS_I(d_inode(dentry)), link);
 433	if (unlikely(error))
 434		goto out_kfree;
 435
 436	set_delayed_call(done, kfree_link, link);
 437	return link;
 438
 439 out_kfree:
 440	kfree(link);
 441 out_err:
 442	return ERR_PTR(error);
 443}
 444
 
 
 
 
 
 
 
 
 
 
 445STATIC int
 446xfs_vn_getattr(
 447	struct vfsmount		*mnt,
 448	struct dentry		*dentry,
 449	struct kstat		*stat)
 
 450{
 451	struct inode		*inode = d_inode(dentry);
 452	struct xfs_inode	*ip = XFS_I(inode);
 453	struct xfs_mount	*mp = ip->i_mount;
 454
 455	trace_xfs_getattr(ip);
 456
 457	if (XFS_FORCED_SHUTDOWN(mp))
 458		return -EIO;
 459
 460	stat->size = XFS_ISIZE(ip);
 461	stat->dev = inode->i_sb->s_dev;
 462	stat->mode = inode->i_mode;
 463	stat->nlink = inode->i_nlink;
 464	stat->uid = inode->i_uid;
 465	stat->gid = inode->i_gid;
 466	stat->ino = ip->i_ino;
 467	stat->atime = inode->i_atime;
 468	stat->mtime = inode->i_mtime;
 469	stat->ctime = inode->i_ctime;
 470	stat->blocks =
 471		XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks);
 472
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 473
 474	switch (inode->i_mode & S_IFMT) {
 475	case S_IFBLK:
 476	case S_IFCHR:
 477		stat->blksize = BLKDEV_IOSIZE;
 478		stat->rdev = MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff,
 479				   sysv_minor(ip->i_df.if_u2.if_rdev));
 480		break;
 481	default:
 482		if (XFS_IS_REALTIME_INODE(ip)) {
 483			/*
 484			 * If the file blocks are being allocated from a
 485			 * realtime volume, then return the inode's realtime
 486			 * extent size or the realtime volume's extent size.
 487			 */
 488			stat->blksize =
 489				xfs_get_extsz_hint(ip) << mp->m_sb.sb_blocklog;
 490		} else
 491			stat->blksize = xfs_preferred_iosize(mp);
 492		stat->rdev = 0;
 493		break;
 494	}
 495
 496	return 0;
 497}
 498
 499static void
 500xfs_setattr_mode(
 501	struct xfs_inode	*ip,
 502	struct iattr		*iattr)
 503{
 504	struct inode		*inode = VFS_I(ip);
 505	umode_t			mode = iattr->ia_mode;
 506
 507	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
 508
 509	inode->i_mode &= S_IFMT;
 510	inode->i_mode |= mode & ~S_IFMT;
 511}
 512
 513void
 514xfs_setattr_time(
 515	struct xfs_inode	*ip,
 516	struct iattr		*iattr)
 517{
 518	struct inode		*inode = VFS_I(ip);
 519
 520	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
 521
 522	if (iattr->ia_valid & ATTR_ATIME)
 523		inode->i_atime = iattr->ia_atime;
 524	if (iattr->ia_valid & ATTR_CTIME)
 525		inode->i_ctime = iattr->ia_ctime;
 526	if (iattr->ia_valid & ATTR_MTIME)
 527		inode->i_mtime = iattr->ia_mtime;
 528}
 529
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 530int
 531xfs_setattr_nonsize(
 532	struct xfs_inode	*ip,
 533	struct iattr		*iattr,
 534	int			flags)
 535{
 536	xfs_mount_t		*mp = ip->i_mount;
 537	struct inode		*inode = VFS_I(ip);
 538	int			mask = iattr->ia_valid;
 539	xfs_trans_t		*tp;
 540	int			error;
 541	kuid_t			uid = GLOBAL_ROOT_UID, iuid = GLOBAL_ROOT_UID;
 542	kgid_t			gid = GLOBAL_ROOT_GID, igid = GLOBAL_ROOT_GID;
 543	struct xfs_dquot	*udqp = NULL, *gdqp = NULL;
 544	struct xfs_dquot	*olddquot1 = NULL, *olddquot2 = NULL;
 545
 546	trace_xfs_setattr(ip);
 547
 548	/* If acls are being inherited, we already have this checked */
 549	if (!(flags & XFS_ATTR_NOACL)) {
 550		if (mp->m_flags & XFS_MOUNT_RDONLY)
 551			return -EROFS;
 552
 553		if (XFS_FORCED_SHUTDOWN(mp))
 554			return -EIO;
 555
 556		error = inode_change_ok(inode, iattr);
 557		if (error)
 558			return error;
 559	}
 560
 561	ASSERT((mask & ATTR_SIZE) == 0);
 562
 563	/*
 564	 * If disk quotas is on, we make sure that the dquots do exist on disk,
 565	 * before we start any other transactions. Trying to do this later
 566	 * is messy. We don't care to take a readlock to look at the ids
 567	 * in inode here, because we can't hold it across the trans_reserve.
 568	 * If the IDs do change before we take the ilock, we're covered
 569	 * because the i_*dquot fields will get updated anyway.
 570	 */
 571	if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) {
 572		uint	qflags = 0;
 573
 574		if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) {
 575			uid = iattr->ia_uid;
 576			qflags |= XFS_QMOPT_UQUOTA;
 577		} else {
 578			uid = inode->i_uid;
 579		}
 580		if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) {
 581			gid = iattr->ia_gid;
 582			qflags |= XFS_QMOPT_GQUOTA;
 583		}  else {
 584			gid = inode->i_gid;
 585		}
 586
 587		/*
 588		 * We take a reference when we initialize udqp and gdqp,
 589		 * so it is important that we never blindly double trip on
 590		 * the same variable. See xfs_create() for an example.
 591		 */
 592		ASSERT(udqp == NULL);
 593		ASSERT(gdqp == NULL);
 594		error = xfs_qm_vop_dqalloc(ip, xfs_kuid_to_uid(uid),
 595					   xfs_kgid_to_gid(gid),
 596					   xfs_get_projid(ip),
 597					   qflags, &udqp, &gdqp, NULL);
 598		if (error)
 599			return error;
 600	}
 601
 602	tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_NOT_SIZE);
 603	error = xfs_trans_reserve(tp, &M_RES(mp)->tr_ichange, 0, 0);
 604	if (error)
 605		goto out_trans_cancel;
 606
 607	xfs_ilock(ip, XFS_ILOCK_EXCL);
 
 608
 609	/*
 610	 * Change file ownership.  Must be the owner or privileged.
 611	 */
 612	if (mask & (ATTR_UID|ATTR_GID)) {
 613		/*
 614		 * These IDs could have changed since we last looked at them.
 615		 * But, we're assured that if the ownership did change
 616		 * while we didn't have the inode locked, inode's dquot(s)
 617		 * would have changed also.
 618		 */
 619		iuid = inode->i_uid;
 620		igid = inode->i_gid;
 621		gid = (mask & ATTR_GID) ? iattr->ia_gid : igid;
 622		uid = (mask & ATTR_UID) ? iattr->ia_uid : iuid;
 623
 624		/*
 625		 * Do a quota reservation only if uid/gid is actually
 626		 * going to change.
 627		 */
 628		if (XFS_IS_QUOTA_RUNNING(mp) &&
 629		    ((XFS_IS_UQUOTA_ON(mp) && !uid_eq(iuid, uid)) ||
 630		     (XFS_IS_GQUOTA_ON(mp) && !gid_eq(igid, gid)))) {
 631			ASSERT(tp);
 632			error = xfs_qm_vop_chown_reserve(tp, ip, udqp, gdqp,
 633						NULL, capable(CAP_FOWNER) ?
 634						XFS_QMOPT_FORCE_RES : 0);
 635			if (error)	/* out of quota */
 636				goto out_unlock;
 637		}
 638	}
 639
 640	xfs_trans_ijoin(tp, ip, 0);
 641
 642	/*
 643	 * Change file ownership.  Must be the owner or privileged.
 644	 */
 645	if (mask & (ATTR_UID|ATTR_GID)) {
 646		/*
 647		 * CAP_FSETID overrides the following restrictions:
 648		 *
 649		 * The set-user-ID and set-group-ID bits of a file will be
 650		 * cleared upon successful return from chown()
 651		 */
 652		if ((inode->i_mode & (S_ISUID|S_ISGID)) &&
 653		    !capable(CAP_FSETID))
 654			inode->i_mode &= ~(S_ISUID|S_ISGID);
 655
 656		/*
 657		 * Change the ownerships and register quota modifications
 658		 * in the transaction.
 659		 */
 660		if (!uid_eq(iuid, uid)) {
 661			if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_UQUOTA_ON(mp)) {
 662				ASSERT(mask & ATTR_UID);
 663				ASSERT(udqp);
 664				olddquot1 = xfs_qm_vop_chown(tp, ip,
 665							&ip->i_udquot, udqp);
 666			}
 667			ip->i_d.di_uid = xfs_kuid_to_uid(uid);
 668			inode->i_uid = uid;
 669		}
 670		if (!gid_eq(igid, gid)) {
 671			if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_GQUOTA_ON(mp)) {
 672				ASSERT(xfs_sb_version_has_pquotino(&mp->m_sb) ||
 673				       !XFS_IS_PQUOTA_ON(mp));
 674				ASSERT(mask & ATTR_GID);
 675				ASSERT(gdqp);
 676				olddquot2 = xfs_qm_vop_chown(tp, ip,
 677							&ip->i_gdquot, gdqp);
 678			}
 679			ip->i_d.di_gid = xfs_kgid_to_gid(gid);
 680			inode->i_gid = gid;
 681		}
 682	}
 683
 684	if (mask & ATTR_MODE)
 685		xfs_setattr_mode(ip, iattr);
 686	if (mask & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME))
 687		xfs_setattr_time(ip, iattr);
 688
 689	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
 690
 691	XFS_STATS_INC(mp, xs_ig_attrchg);
 692
 693	if (mp->m_flags & XFS_MOUNT_WSYNC)
 694		xfs_trans_set_sync(tp);
 695	error = xfs_trans_commit(tp);
 696
 697	xfs_iunlock(ip, XFS_ILOCK_EXCL);
 698
 699	/*
 700	 * Release any dquot(s) the inode had kept before chown.
 701	 */
 702	xfs_qm_dqrele(olddquot1);
 703	xfs_qm_dqrele(olddquot2);
 704	xfs_qm_dqrele(udqp);
 705	xfs_qm_dqrele(gdqp);
 706
 707	if (error)
 708		return error;
 709
 710	/*
 711	 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode
 712	 * 	     update.  We could avoid this with linked transactions
 713	 * 	     and passing down the transaction pointer all the way
 714	 *	     to attr_set.  No previous user of the generic
 715	 * 	     Posix ACL code seems to care about this issue either.
 716	 */
 717	if ((mask & ATTR_MODE) && !(flags & XFS_ATTR_NOACL)) {
 718		error = posix_acl_chmod(inode, inode->i_mode);
 719		if (error)
 720			return error;
 721	}
 722
 723	return 0;
 724
 725out_unlock:
 726	xfs_iunlock(ip, XFS_ILOCK_EXCL);
 727out_trans_cancel:
 728	xfs_trans_cancel(tp);
 
 729	xfs_qm_dqrele(udqp);
 730	xfs_qm_dqrele(gdqp);
 731	return error;
 732}
 733
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 734/*
 735 * Truncate file.  Must have write permission and not be a directory.
 
 
 
 736 */
 737int
 738xfs_setattr_size(
 739	struct xfs_inode	*ip,
 740	struct iattr		*iattr)
 741{
 742	struct xfs_mount	*mp = ip->i_mount;
 743	struct inode		*inode = VFS_I(ip);
 744	xfs_off_t		oldsize, newsize;
 745	struct xfs_trans	*tp;
 746	int			error;
 747	uint			lock_flags = 0;
 748	bool			did_zeroing = false;
 749
 750	trace_xfs_setattr(ip);
 751
 752	if (mp->m_flags & XFS_MOUNT_RDONLY)
 753		return -EROFS;
 754
 755	if (XFS_FORCED_SHUTDOWN(mp))
 756		return -EIO;
 757
 758	error = inode_change_ok(inode, iattr);
 759	if (error)
 760		return error;
 761
 762	ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
 763	ASSERT(xfs_isilocked(ip, XFS_MMAPLOCK_EXCL));
 764	ASSERT(S_ISREG(inode->i_mode));
 765	ASSERT((iattr->ia_valid & (ATTR_UID|ATTR_GID|ATTR_ATIME|ATTR_ATIME_SET|
 766		ATTR_MTIME_SET|ATTR_KILL_PRIV|ATTR_TIMES_SET)) == 0);
 767
 768	oldsize = inode->i_size;
 769	newsize = iattr->ia_size;
 770
 771	/*
 772	 * Short circuit the truncate case for zero length files.
 773	 */
 774	if (newsize == 0 && oldsize == 0 && ip->i_d.di_nextents == 0) {
 775		if (!(iattr->ia_valid & (ATTR_CTIME|ATTR_MTIME)))
 776			return 0;
 777
 778		/*
 779		 * Use the regular setattr path to update the timestamps.
 780		 */
 781		iattr->ia_valid &= ~ATTR_SIZE;
 782		return xfs_setattr_nonsize(ip, iattr, 0);
 783	}
 784
 785	/*
 786	 * Make sure that the dquots are attached to the inode.
 787	 */
 788	error = xfs_qm_dqattach(ip, 0);
 789	if (error)
 790		return error;
 791
 792	/*
 
 
 
 
 
 793	 * File data changes must be complete before we start the transaction to
 794	 * modify the inode.  This needs to be done before joining the inode to
 795	 * the transaction because the inode cannot be unlocked once it is a
 796	 * part of the transaction.
 797	 *
 798	 * Start with zeroing any data block beyond EOF that we may expose on
 799	 * file extension.
 
 800	 */
 801	if (newsize > oldsize) {
 802		error = xfs_zero_eof(ip, newsize, oldsize, &did_zeroing);
 803		if (error)
 804			return error;
 805	}
 806
 807	/*
 808	 * We are going to log the inode size change in this transaction so
 809	 * any previous writes that are beyond the on disk EOF and the new
 810	 * EOF that have not been written out need to be written here.  If we
 811	 * do not write the data out, we expose ourselves to the null files
 812	 * problem. Note that this includes any block zeroing we did above;
 813	 * otherwise those blocks may not be zeroed after a crash.
 814	 */
 815	if (newsize > ip->i_d.di_size &&
 816	    (oldsize != ip->i_d.di_size || did_zeroing)) {
 817		error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
 818						      ip->i_d.di_size, newsize);
 819		if (error)
 820			return error;
 821	}
 822
 823	/* Now wait for all direct I/O to complete. */
 824	inode_dio_wait(inode);
 825
 826	/*
 827	 * We've already locked out new page faults, so now we can safely remove
 828	 * pages from the page cache knowing they won't get refaulted until we
 829	 * drop the XFS_MMAP_EXCL lock after the extent manipulations are
 830	 * complete. The truncate_setsize() call also cleans partial EOF page
 831	 * PTEs on extending truncates and hence ensures sub-page block size
 832	 * filesystems are correctly handled, too.
 833	 *
 834	 * We have to do all the page cache truncate work outside the
 835	 * transaction context as the "lock" order is page lock->log space
 836	 * reservation as defined by extent allocation in the writeback path.
 837	 * Hence a truncate can fail with ENOMEM from xfs_trans_reserve(), but
 838	 * having already truncated the in-memory version of the file (i.e. made
 839	 * user visible changes). There's not much we can do about this, except
 840	 * to hope that the caller sees ENOMEM and retries the truncate
 841	 * operation.
 
 
 
 
 842	 */
 843	if (IS_DAX(inode))
 844		error = dax_truncate_page(inode, newsize, xfs_get_blocks_direct);
 845	else
 846		error = block_truncate_page(inode->i_mapping, newsize,
 847					    xfs_get_blocks);
 848	if (error)
 849		return error;
 850	truncate_setsize(inode, newsize);
 851
 852	tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_SIZE);
 853	error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 854	if (error)
 855		goto out_trans_cancel;
 856
 857	lock_flags |= XFS_ILOCK_EXCL;
 858	xfs_ilock(ip, XFS_ILOCK_EXCL);
 859	xfs_trans_ijoin(tp, ip, 0);
 860
 861	/*
 862	 * Only change the c/mtime if we are changing the size or we are
 863	 * explicitly asked to change it.  This handles the semantic difference
 864	 * between truncate() and ftruncate() as implemented in the VFS.
 865	 *
 866	 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
 867	 * special case where we need to update the times despite not having
 868	 * these flags set.  For all other operations the VFS set these flags
 869	 * explicitly if it wants a timestamp update.
 870	 */
 871	if (newsize != oldsize &&
 872	    !(iattr->ia_valid & (ATTR_CTIME | ATTR_MTIME))) {
 873		iattr->ia_ctime = iattr->ia_mtime =
 874			current_fs_time(inode->i_sb);
 875		iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME;
 876	}
 877
 878	/*
 879	 * The first thing we do is set the size to new_size permanently on
 880	 * disk.  This way we don't have to worry about anyone ever being able
 881	 * to look at the data being freed even in the face of a crash.
 882	 * What we're getting around here is the case where we free a block, it
 883	 * is allocated to another file, it is written to, and then we crash.
 884	 * If the new data gets written to the file but the log buffers
 885	 * containing the free and reallocation don't, then we'd end up with
 886	 * garbage in the blocks being freed.  As long as we make the new size
 887	 * permanent before actually freeing any blocks it doesn't matter if
 888	 * they get written to.
 889	 */
 890	ip->i_d.di_size = newsize;
 891	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
 892
 893	if (newsize <= oldsize) {
 894		error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, newsize);
 895		if (error)
 896			goto out_trans_cancel;
 897
 898		/*
 899		 * Truncated "down", so we're removing references to old data
 900		 * here - if we delay flushing for a long time, we expose
 901		 * ourselves unduly to the notorious NULL files problem.  So,
 902		 * we mark this inode and flush it when the file is closed,
 903		 * and do not wait the usual (long) time for writeout.
 904		 */
 905		xfs_iflags_set(ip, XFS_ITRUNCATED);
 906
 907		/* A truncate down always removes post-EOF blocks. */
 908		xfs_inode_clear_eofblocks_tag(ip);
 909	}
 910
 911	if (iattr->ia_valid & ATTR_MODE)
 912		xfs_setattr_mode(ip, iattr);
 913	if (iattr->ia_valid & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME))
 914		xfs_setattr_time(ip, iattr);
 915
 916	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
 917
 918	XFS_STATS_INC(mp, xs_ig_attrchg);
 919
 920	if (mp->m_flags & XFS_MOUNT_WSYNC)
 921		xfs_trans_set_sync(tp);
 922
 923	error = xfs_trans_commit(tp);
 924out_unlock:
 925	if (lock_flags)
 926		xfs_iunlock(ip, lock_flags);
 927	return error;
 928
 929out_trans_cancel:
 930	xfs_trans_cancel(tp);
 931	goto out_unlock;
 932}
 933
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 934STATIC int
 935xfs_vn_setattr(
 936	struct dentry		*dentry,
 937	struct iattr		*iattr)
 938{
 939	struct xfs_inode	*ip = XFS_I(d_inode(dentry));
 940	int			error;
 941
 942	if (iattr->ia_valid & ATTR_SIZE) {
 943		uint		iolock = XFS_IOLOCK_EXCL;
 
 944
 945		xfs_ilock(ip, iolock);
 946		error = xfs_break_layouts(d_inode(dentry), &iolock, true);
 947		if (!error) {
 948			xfs_ilock(ip, XFS_MMAPLOCK_EXCL);
 949			iolock |= XFS_MMAPLOCK_EXCL;
 950
 951			error = xfs_setattr_size(ip, iattr);
 952		}
 953		xfs_iunlock(ip, iolock);
 954	} else {
 955		error = xfs_setattr_nonsize(ip, iattr, 0);
 956	}
 957
 958	return error;
 959}
 960
 961STATIC int
 962xfs_vn_update_time(
 963	struct inode		*inode,
 964	struct timespec		*now,
 965	int			flags)
 966{
 967	struct xfs_inode	*ip = XFS_I(inode);
 968	struct xfs_mount	*mp = ip->i_mount;
 
 969	struct xfs_trans	*tp;
 970	int			error;
 971
 972	trace_xfs_update_time(ip);
 973
 974	tp = xfs_trans_alloc(mp, XFS_TRANS_FSYNC_TS);
 975	error = xfs_trans_reserve(tp, &M_RES(mp)->tr_fsyncts, 0, 0);
 976	if (error) {
 977		xfs_trans_cancel(tp);
 978		return error;
 
 
 979	}
 980
 
 
 
 
 981	xfs_ilock(ip, XFS_ILOCK_EXCL);
 982	if (flags & S_CTIME)
 983		inode->i_ctime = *now;
 984	if (flags & S_MTIME)
 985		inode->i_mtime = *now;
 986	if (flags & S_ATIME)
 987		inode->i_atime = *now;
 988
 989	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
 990	xfs_trans_log_inode(tp, ip, XFS_ILOG_TIMESTAMP);
 991	return xfs_trans_commit(tp);
 992}
 993
 994#define XFS_FIEMAP_FLAGS	(FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
 995
 996/*
 997 * Call fiemap helper to fill in user data.
 998 * Returns positive errors to xfs_getbmap.
 999 */
1000STATIC int
1001xfs_fiemap_format(
1002	void			**arg,
1003	struct getbmapx		*bmv,
1004	int			*full)
1005{
1006	int			error;
1007	struct fiemap_extent_info *fieinfo = *arg;
1008	u32			fiemap_flags = 0;
1009	u64			logical, physical, length;
1010
1011	/* Do nothing for a hole */
1012	if (bmv->bmv_block == -1LL)
1013		return 0;
1014
1015	logical = BBTOB(bmv->bmv_offset);
1016	physical = BBTOB(bmv->bmv_block);
1017	length = BBTOB(bmv->bmv_length);
1018
1019	if (bmv->bmv_oflags & BMV_OF_PREALLOC)
1020		fiemap_flags |= FIEMAP_EXTENT_UNWRITTEN;
1021	else if (bmv->bmv_oflags & BMV_OF_DELALLOC) {
1022		fiemap_flags |= (FIEMAP_EXTENT_DELALLOC |
1023				 FIEMAP_EXTENT_UNKNOWN);
1024		physical = 0;   /* no block yet */
1025	}
1026	if (bmv->bmv_oflags & BMV_OF_LAST)
1027		fiemap_flags |= FIEMAP_EXTENT_LAST;
1028
1029	error = fiemap_fill_next_extent(fieinfo, logical, physical,
1030					length, fiemap_flags);
1031	if (error > 0) {
1032		error = 0;
1033		*full = 1;	/* user array now full */
1034	}
1035
1036	return error;
1037}
1038
1039STATIC int
1040xfs_vn_fiemap(
1041	struct inode		*inode,
1042	struct fiemap_extent_info *fieinfo,
1043	u64			start,
1044	u64			length)
1045{
1046	xfs_inode_t		*ip = XFS_I(inode);
1047	struct getbmapx		bm;
1048	int			error;
1049
1050	error = fiemap_check_flags(fieinfo, XFS_FIEMAP_FLAGS);
1051	if (error)
1052		return error;
1053
1054	/* Set up bmap header for xfs internal routine */
1055	bm.bmv_offset = BTOBBT(start);
1056	/* Special case for whole file */
1057	if (length == FIEMAP_MAX_OFFSET)
1058		bm.bmv_length = -1LL;
1059	else
1060		bm.bmv_length = BTOBB(start + length) - bm.bmv_offset;
1061
1062	/* We add one because in getbmap world count includes the header */
1063	bm.bmv_count = !fieinfo->fi_extents_max ? MAXEXTNUM :
1064					fieinfo->fi_extents_max + 1;
1065	bm.bmv_count = min_t(__s32, bm.bmv_count,
1066			     (PAGE_SIZE * 16 / sizeof(struct getbmapx)));
1067	bm.bmv_iflags = BMV_IF_PREALLOC | BMV_IF_NO_HOLES;
1068	if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR)
1069		bm.bmv_iflags |= BMV_IF_ATTRFORK;
1070	if (!(fieinfo->fi_flags & FIEMAP_FLAG_SYNC))
1071		bm.bmv_iflags |= BMV_IF_DELALLOC;
1072
1073	error = xfs_getbmap(ip, &bm, xfs_fiemap_format, fieinfo);
1074	if (error)
1075		return error;
1076
1077	return 0;
1078}
1079
1080STATIC int
1081xfs_vn_tmpfile(
1082	struct inode	*dir,
1083	struct dentry	*dentry,
1084	umode_t		mode)
1085{
1086	return xfs_generic_create(dir, dentry, mode, 0, true);
1087}
1088
1089static const struct inode_operations xfs_inode_operations = {
1090	.get_acl		= xfs_get_acl,
1091	.set_acl		= xfs_set_acl,
1092	.getattr		= xfs_vn_getattr,
1093	.setattr		= xfs_vn_setattr,
1094	.setxattr		= generic_setxattr,
1095	.getxattr		= generic_getxattr,
1096	.removexattr		= generic_removexattr,
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	.rename2		= 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	.setxattr		= generic_setxattr,
1123	.getxattr		= generic_getxattr,
1124	.removexattr		= generic_removexattr,
1125	.listxattr		= xfs_vn_listxattr,
1126	.update_time		= xfs_vn_update_time,
1127	.tmpfile		= xfs_vn_tmpfile,
1128};
1129
1130static const struct inode_operations xfs_dir_ci_inode_operations = {
1131	.create			= xfs_vn_create,
1132	.lookup			= xfs_vn_ci_lookup,
1133	.link			= xfs_vn_link,
1134	.unlink			= xfs_vn_unlink,
1135	.symlink		= xfs_vn_symlink,
1136	.mkdir			= xfs_vn_mkdir,
1137	/*
1138	 * Yes, XFS uses the same method for rmdir and unlink.
1139	 *
1140	 * There are some subtile differences deeper in the code,
1141	 * but we use S_ISDIR to check for those.
1142	 */
1143	.rmdir			= xfs_vn_unlink,
1144	.mknod			= xfs_vn_mknod,
1145	.rename2		= xfs_vn_rename,
1146	.get_acl		= xfs_get_acl,
1147	.set_acl		= xfs_set_acl,
1148	.getattr		= xfs_vn_getattr,
1149	.setattr		= xfs_vn_setattr,
1150	.setxattr		= generic_setxattr,
1151	.getxattr		= generic_getxattr,
1152	.removexattr		= generic_removexattr,
1153	.listxattr		= xfs_vn_listxattr,
1154	.update_time		= xfs_vn_update_time,
1155	.tmpfile		= xfs_vn_tmpfile,
1156};
1157
1158static const struct inode_operations xfs_symlink_inode_operations = {
1159	.readlink		= generic_readlink,
1160	.get_link		= xfs_vn_get_link,
1161	.getattr		= xfs_vn_getattr,
1162	.setattr		= xfs_vn_setattr,
1163	.setxattr		= generic_setxattr,
1164	.getxattr		= generic_getxattr,
1165	.removexattr		= generic_removexattr,
 
 
 
 
 
1166	.listxattr		= xfs_vn_listxattr,
1167	.update_time		= xfs_vn_update_time,
1168};
1169
1170STATIC void
1171xfs_diflags_to_iflags(
1172	struct inode		*inode,
1173	struct xfs_inode	*ip)
1174{
1175	uint16_t		flags = ip->i_d.di_flags;
1176
1177	inode->i_flags &= ~(S_IMMUTABLE | S_APPEND | S_SYNC |
1178			    S_NOATIME | S_DAX);
1179
1180	if (flags & XFS_DIFLAG_IMMUTABLE)
1181		inode->i_flags |= S_IMMUTABLE;
1182	if (flags & XFS_DIFLAG_APPEND)
1183		inode->i_flags |= S_APPEND;
1184	if (flags & XFS_DIFLAG_SYNC)
1185		inode->i_flags |= S_SYNC;
1186	if (flags & XFS_DIFLAG_NOATIME)
1187		inode->i_flags |= S_NOATIME;
1188	if (S_ISREG(inode->i_mode) &&
1189	    ip->i_mount->m_sb.sb_blocksize == PAGE_SIZE &&
 
1190	    (ip->i_mount->m_flags & XFS_MOUNT_DAX ||
1191	     ip->i_d.di_flags2 & XFS_DIFLAG2_DAX))
1192		inode->i_flags |= S_DAX;
1193}
1194
1195/*
1196 * Initialize the Linux inode and set up the operation vectors.
1197 *
1198 * When reading existing inodes from disk this is called directly from xfs_iget,
1199 * when creating a new inode it is called from xfs_ialloc after setting up the
1200 * inode. These callers have different criteria for clearing XFS_INEW, so leave
1201 * it up to the caller to deal with unlocking the inode appropriately.
1202 */
1203void
1204xfs_setup_inode(
1205	struct xfs_inode	*ip)
1206{
1207	struct inode		*inode = &ip->i_vnode;
1208	gfp_t			gfp_mask;
1209
1210	inode->i_ino = ip->i_ino;
1211	inode->i_state = I_NEW;
1212
1213	inode_sb_list_add(inode);
1214	/* make the inode look hashed for the writeback code */
1215	hlist_add_fake(&inode->i_hash);
1216
1217	inode->i_uid    = xfs_uid_to_kuid(ip->i_d.di_uid);
1218	inode->i_gid    = xfs_gid_to_kgid(ip->i_d.di_gid);
1219
1220	switch (inode->i_mode & S_IFMT) {
1221	case S_IFBLK:
1222	case S_IFCHR:
1223		inode->i_rdev =
1224			MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff,
1225			      sysv_minor(ip->i_df.if_u2.if_rdev));
1226		break;
1227	default:
1228		inode->i_rdev = 0;
1229		break;
1230	}
1231
1232	i_size_write(inode, ip->i_d.di_size);
1233	xfs_diflags_to_iflags(inode, ip);
1234
1235	ip->d_ops = ip->i_mount->m_nondir_inode_ops;
1236	lockdep_set_class(&ip->i_lock.mr_lock, &xfs_nondir_ilock_class);
1237	switch (inode->i_mode & S_IFMT) {
1238	case S_IFREG:
1239		inode->i_op = &xfs_inode_operations;
1240		inode->i_fop = &xfs_file_operations;
1241		inode->i_mapping->a_ops = &xfs_address_space_operations;
1242		break;
1243	case S_IFDIR:
1244		lockdep_set_class(&ip->i_lock.mr_lock, &xfs_dir_ilock_class);
1245		if (xfs_sb_version_hasasciici(&XFS_M(inode->i_sb)->m_sb))
1246			inode->i_op = &xfs_dir_ci_inode_operations;
1247		else
1248			inode->i_op = &xfs_dir_inode_operations;
1249		inode->i_fop = &xfs_dir_file_operations;
1250		ip->d_ops = ip->i_mount->m_dir_inode_ops;
1251		break;
1252	case S_IFLNK:
1253		inode->i_op = &xfs_symlink_inode_operations;
1254		if (!(ip->i_df.if_flags & XFS_IFINLINE))
1255			inode->i_mapping->a_ops = &xfs_address_space_operations;
1256		break;
1257	default:
1258		inode->i_op = &xfs_inode_operations;
1259		init_special_inode(inode, inode->i_mode, inode->i_rdev);
1260		break;
1261	}
1262
1263	/*
1264	 * Ensure all page cache allocations are done from GFP_NOFS context to
1265	 * prevent direct reclaim recursion back into the filesystem and blowing
1266	 * stacks or deadlocking.
1267	 */
1268	gfp_mask = mapping_gfp_mask(inode->i_mapping);
1269	mapping_set_gfp_mask(inode->i_mapping, (gfp_mask & ~(__GFP_FS)));
1270
1271	/*
1272	 * If there is no attribute fork no ACL can exist on this inode,
1273	 * and it can't have any file capabilities attached to it either.
1274	 */
1275	if (!XFS_IFORK_Q(ip)) {
1276		inode_has_no_xattr(inode);
1277		cache_no_acl(inode);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1278	}
1279}
v4.17
   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#include <linux/iversion.h>
  50
  51/*
  52 * Directories have different lock order w.r.t. mmap_sem compared to regular
  53 * files. This is due to readdir potentially triggering page faults on a user
  54 * buffer inside filldir(), and this happens with the ilock on the directory
  55 * held. For regular files, the lock order is the other way around - the
  56 * mmap_sem is taken during the page fault, and then we lock the ilock to do
  57 * block mapping. Hence we need a different class for the directory ilock so
  58 * that lockdep can tell them apart.
  59 */
  60static struct lock_class_key xfs_nondir_ilock_class;
  61static struct lock_class_key xfs_dir_ilock_class;
  62
  63static int
  64xfs_initxattrs(
  65	struct inode		*inode,
  66	const struct xattr	*xattr_array,
  67	void			*fs_info)
  68{
  69	const struct xattr	*xattr;
  70	struct xfs_inode	*ip = XFS_I(inode);
  71	int			error = 0;
  72
  73	for (xattr = xattr_array; xattr->name != NULL; xattr++) {
  74		error = xfs_attr_set(ip, xattr->name, xattr->value,
  75				      xattr->value_len, ATTR_SECURE);
  76		if (error < 0)
  77			break;
  78	}
  79	return error;
  80}
  81
  82/*
  83 * Hook in SELinux.  This is not quite correct yet, what we really need
  84 * here (as we do for default ACLs) is a mechanism by which creation of
  85 * these attrs can be journalled at inode creation time (along with the
  86 * inode, of course, such that log replay can't cause these to be lost).
  87 */
  88
  89STATIC int
  90xfs_init_security(
  91	struct inode	*inode,
  92	struct inode	*dir,
  93	const struct qstr *qstr)
  94{
  95	return security_inode_init_security(inode, dir, qstr,
  96					     &xfs_initxattrs, NULL);
  97}
  98
  99static void
 100xfs_dentry_to_name(
 101	struct xfs_name	*namep,
 102	struct dentry	*dentry)
 103{
 104	namep->name = dentry->d_name.name;
 105	namep->len = dentry->d_name.len;
 106	namep->type = XFS_DIR3_FT_UNKNOWN;
 107}
 108
 109static int
 110xfs_dentry_mode_to_name(
 111	struct xfs_name	*namep,
 112	struct dentry	*dentry,
 113	int		mode)
 114{
 115	namep->name = dentry->d_name.name;
 116	namep->len = dentry->d_name.len;
 117	namep->type = xfs_mode_to_ftype(mode);
 118
 119	if (unlikely(namep->type == XFS_DIR3_FT_UNKNOWN))
 120		return -EFSCORRUPTED;
 121
 122	return 0;
 123}
 124
 125STATIC void
 126xfs_cleanup_inode(
 127	struct inode	*dir,
 128	struct inode	*inode,
 129	struct dentry	*dentry)
 130{
 131	struct xfs_name	teardown;
 132
 133	/* Oh, the horror.
 134	 * If we can't add the ACL or we fail in
 135	 * xfs_init_security we must back out.
 136	 * ENOSPC can hit here, among other things.
 137	 */
 138	xfs_dentry_to_name(&teardown, dentry);
 139
 140	xfs_remove(XFS_I(dir), &teardown, XFS_I(inode));
 141}
 142
 143STATIC int
 144xfs_generic_create(
 145	struct inode	*dir,
 146	struct dentry	*dentry,
 147	umode_t		mode,
 148	dev_t		rdev,
 149	bool		tmpfile)	/* unnamed file */
 150{
 151	struct inode	*inode;
 152	struct xfs_inode *ip = NULL;
 153	struct posix_acl *default_acl, *acl;
 154	struct xfs_name	name;
 155	int		error;
 156
 157	/*
 158	 * Irix uses Missed'em'V split, but doesn't want to see
 159	 * the upper 5 bits of (14bit) major.
 160	 */
 161	if (S_ISCHR(mode) || S_ISBLK(mode)) {
 162		if (unlikely(!sysv_valid_dev(rdev) || MAJOR(rdev) & ~0x1ff))
 163			return -EINVAL;
 
 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), 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(XFS_SYMLINK_MAXLEN+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	const struct path	*path,
 493	struct kstat		*stat,
 494	u32			request_mask,
 495	unsigned int		query_flags)
 496{
 497	struct inode		*inode = d_inode(path->dentry);
 498	struct xfs_inode	*ip = XFS_I(inode);
 499	struct xfs_mount	*mp = ip->i_mount;
 500
 501	trace_xfs_getattr(ip);
 502
 503	if (XFS_FORCED_SHUTDOWN(mp))
 504		return -EIO;
 505
 506	stat->size = XFS_ISIZE(ip);
 507	stat->dev = inode->i_sb->s_dev;
 508	stat->mode = inode->i_mode;
 509	stat->nlink = inode->i_nlink;
 510	stat->uid = inode->i_uid;
 511	stat->gid = inode->i_gid;
 512	stat->ino = ip->i_ino;
 513	stat->atime = inode->i_atime;
 514	stat->mtime = inode->i_mtime;
 515	stat->ctime = inode->i_ctime;
 516	stat->blocks =
 517		XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks);
 518
 519	if (ip->i_d.di_version == 3) {
 520		if (request_mask & STATX_BTIME) {
 521			stat->result_mask |= STATX_BTIME;
 522			stat->btime.tv_sec = ip->i_d.di_crtime.t_sec;
 523			stat->btime.tv_nsec = ip->i_d.di_crtime.t_nsec;
 524		}
 525	}
 526
 527	if (ip->i_d.di_flags & XFS_DIFLAG_IMMUTABLE)
 528		stat->attributes |= STATX_ATTR_IMMUTABLE;
 529	if (ip->i_d.di_flags & XFS_DIFLAG_APPEND)
 530		stat->attributes |= STATX_ATTR_APPEND;
 531	if (ip->i_d.di_flags & XFS_DIFLAG_NODUMP)
 532		stat->attributes |= STATX_ATTR_NODUMP;
 533
 534	switch (inode->i_mode & S_IFMT) {
 535	case S_IFBLK:
 536	case S_IFCHR:
 537		stat->blksize = BLKDEV_IOSIZE;
 538		stat->rdev = inode->i_rdev;
 
 539		break;
 540	default:
 541		if (XFS_IS_REALTIME_INODE(ip)) {
 542			/*
 543			 * If the file blocks are being allocated from a
 544			 * realtime volume, then return the inode's realtime
 545			 * extent size or the realtime volume's extent size.
 546			 */
 547			stat->blksize =
 548				xfs_get_extsz_hint(ip) << mp->m_sb.sb_blocklog;
 549		} else
 550			stat->blksize = xfs_preferred_iosize(mp);
 551		stat->rdev = 0;
 552		break;
 553	}
 554
 555	return 0;
 556}
 557
 558static void
 559xfs_setattr_mode(
 560	struct xfs_inode	*ip,
 561	struct iattr		*iattr)
 562{
 563	struct inode		*inode = VFS_I(ip);
 564	umode_t			mode = iattr->ia_mode;
 565
 566	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
 567
 568	inode->i_mode &= S_IFMT;
 569	inode->i_mode |= mode & ~S_IFMT;
 570}
 571
 572void
 573xfs_setattr_time(
 574	struct xfs_inode	*ip,
 575	struct iattr		*iattr)
 576{
 577	struct inode		*inode = VFS_I(ip);
 578
 579	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
 580
 581	if (iattr->ia_valid & ATTR_ATIME)
 582		inode->i_atime = iattr->ia_atime;
 583	if (iattr->ia_valid & ATTR_CTIME)
 584		inode->i_ctime = iattr->ia_ctime;
 585	if (iattr->ia_valid & ATTR_MTIME)
 586		inode->i_mtime = iattr->ia_mtime;
 587}
 588
 589static int
 590xfs_vn_change_ok(
 591	struct dentry	*dentry,
 592	struct iattr	*iattr)
 593{
 594	struct xfs_mount	*mp = XFS_I(d_inode(dentry))->i_mount;
 595
 596	if (mp->m_flags & XFS_MOUNT_RDONLY)
 597		return -EROFS;
 598
 599	if (XFS_FORCED_SHUTDOWN(mp))
 600		return -EIO;
 601
 602	return setattr_prepare(dentry, iattr);
 603}
 604
 605/*
 606 * Set non-size attributes of an inode.
 607 *
 608 * Caution: The caller of this function is responsible for calling
 609 * setattr_prepare() or otherwise verifying the change is fine.
 610 */
 611int
 612xfs_setattr_nonsize(
 613	struct xfs_inode	*ip,
 614	struct iattr		*iattr,
 615	int			flags)
 616{
 617	xfs_mount_t		*mp = ip->i_mount;
 618	struct inode		*inode = VFS_I(ip);
 619	int			mask = iattr->ia_valid;
 620	xfs_trans_t		*tp;
 621	int			error;
 622	kuid_t			uid = GLOBAL_ROOT_UID, iuid = GLOBAL_ROOT_UID;
 623	kgid_t			gid = GLOBAL_ROOT_GID, igid = GLOBAL_ROOT_GID;
 624	struct xfs_dquot	*udqp = NULL, *gdqp = NULL;
 625	struct xfs_dquot	*olddquot1 = NULL, *olddquot2 = NULL;
 626
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 627	ASSERT((mask & ATTR_SIZE) == 0);
 628
 629	/*
 630	 * If disk quotas is on, we make sure that the dquots do exist on disk,
 631	 * before we start any other transactions. Trying to do this later
 632	 * is messy. We don't care to take a readlock to look at the ids
 633	 * in inode here, because we can't hold it across the trans_reserve.
 634	 * If the IDs do change before we take the ilock, we're covered
 635	 * because the i_*dquot fields will get updated anyway.
 636	 */
 637	if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) {
 638		uint	qflags = 0;
 639
 640		if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) {
 641			uid = iattr->ia_uid;
 642			qflags |= XFS_QMOPT_UQUOTA;
 643		} else {
 644			uid = inode->i_uid;
 645		}
 646		if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) {
 647			gid = iattr->ia_gid;
 648			qflags |= XFS_QMOPT_GQUOTA;
 649		}  else {
 650			gid = inode->i_gid;
 651		}
 652
 653		/*
 654		 * We take a reference when we initialize udqp and gdqp,
 655		 * so it is important that we never blindly double trip on
 656		 * the same variable. See xfs_create() for an example.
 657		 */
 658		ASSERT(udqp == NULL);
 659		ASSERT(gdqp == NULL);
 660		error = xfs_qm_vop_dqalloc(ip, xfs_kuid_to_uid(uid),
 661					   xfs_kgid_to_gid(gid),
 662					   xfs_get_projid(ip),
 663					   qflags, &udqp, &gdqp, NULL);
 664		if (error)
 665			return error;
 666	}
 667
 668	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ichange, 0, 0, 0, &tp);
 
 669	if (error)
 670		goto out_dqrele;
 671
 672	xfs_ilock(ip, XFS_ILOCK_EXCL);
 673	xfs_trans_ijoin(tp, ip, 0);
 674
 675	/*
 676	 * Change file ownership.  Must be the owner or privileged.
 677	 */
 678	if (mask & (ATTR_UID|ATTR_GID)) {
 679		/*
 680		 * These IDs could have changed since we last looked at them.
 681		 * But, we're assured that if the ownership did change
 682		 * while we didn't have the inode locked, inode's dquot(s)
 683		 * would have changed also.
 684		 */
 685		iuid = inode->i_uid;
 686		igid = inode->i_gid;
 687		gid = (mask & ATTR_GID) ? iattr->ia_gid : igid;
 688		uid = (mask & ATTR_UID) ? iattr->ia_uid : iuid;
 689
 690		/*
 691		 * Do a quota reservation only if uid/gid is actually
 692		 * going to change.
 693		 */
 694		if (XFS_IS_QUOTA_RUNNING(mp) &&
 695		    ((XFS_IS_UQUOTA_ON(mp) && !uid_eq(iuid, uid)) ||
 696		     (XFS_IS_GQUOTA_ON(mp) && !gid_eq(igid, gid)))) {
 697			ASSERT(tp);
 698			error = xfs_qm_vop_chown_reserve(tp, ip, udqp, gdqp,
 699						NULL, capable(CAP_FOWNER) ?
 700						XFS_QMOPT_FORCE_RES : 0);
 701			if (error)	/* out of quota */
 702				goto out_cancel;
 703		}
 704	}
 705
 
 
 706	/*
 707	 * Change file ownership.  Must be the owner or privileged.
 708	 */
 709	if (mask & (ATTR_UID|ATTR_GID)) {
 710		/*
 711		 * CAP_FSETID overrides the following restrictions:
 712		 *
 713		 * The set-user-ID and set-group-ID bits of a file will be
 714		 * cleared upon successful return from chown()
 715		 */
 716		if ((inode->i_mode & (S_ISUID|S_ISGID)) &&
 717		    !capable(CAP_FSETID))
 718			inode->i_mode &= ~(S_ISUID|S_ISGID);
 719
 720		/*
 721		 * Change the ownerships and register quota modifications
 722		 * in the transaction.
 723		 */
 724		if (!uid_eq(iuid, uid)) {
 725			if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_UQUOTA_ON(mp)) {
 726				ASSERT(mask & ATTR_UID);
 727				ASSERT(udqp);
 728				olddquot1 = xfs_qm_vop_chown(tp, ip,
 729							&ip->i_udquot, udqp);
 730			}
 731			ip->i_d.di_uid = xfs_kuid_to_uid(uid);
 732			inode->i_uid = uid;
 733		}
 734		if (!gid_eq(igid, gid)) {
 735			if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_GQUOTA_ON(mp)) {
 736				ASSERT(xfs_sb_version_has_pquotino(&mp->m_sb) ||
 737				       !XFS_IS_PQUOTA_ON(mp));
 738				ASSERT(mask & ATTR_GID);
 739				ASSERT(gdqp);
 740				olddquot2 = xfs_qm_vop_chown(tp, ip,
 741							&ip->i_gdquot, gdqp);
 742			}
 743			ip->i_d.di_gid = xfs_kgid_to_gid(gid);
 744			inode->i_gid = gid;
 745		}
 746	}
 747
 748	if (mask & ATTR_MODE)
 749		xfs_setattr_mode(ip, iattr);
 750	if (mask & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME))
 751		xfs_setattr_time(ip, iattr);
 752
 753	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
 754
 755	XFS_STATS_INC(mp, xs_ig_attrchg);
 756
 757	if (mp->m_flags & XFS_MOUNT_WSYNC)
 758		xfs_trans_set_sync(tp);
 759	error = xfs_trans_commit(tp);
 760
 761	xfs_iunlock(ip, XFS_ILOCK_EXCL);
 762
 763	/*
 764	 * Release any dquot(s) the inode had kept before chown.
 765	 */
 766	xfs_qm_dqrele(olddquot1);
 767	xfs_qm_dqrele(olddquot2);
 768	xfs_qm_dqrele(udqp);
 769	xfs_qm_dqrele(gdqp);
 770
 771	if (error)
 772		return error;
 773
 774	/*
 775	 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode
 776	 * 	     update.  We could avoid this with linked transactions
 777	 * 	     and passing down the transaction pointer all the way
 778	 *	     to attr_set.  No previous user of the generic
 779	 * 	     Posix ACL code seems to care about this issue either.
 780	 */
 781	if ((mask & ATTR_MODE) && !(flags & XFS_ATTR_NOACL)) {
 782		error = posix_acl_chmod(inode, inode->i_mode);
 783		if (error)
 784			return error;
 785	}
 786
 787	return 0;
 788
 789out_cancel:
 
 
 790	xfs_trans_cancel(tp);
 791out_dqrele:
 792	xfs_qm_dqrele(udqp);
 793	xfs_qm_dqrele(gdqp);
 794	return error;
 795}
 796
 797int
 798xfs_vn_setattr_nonsize(
 799	struct dentry		*dentry,
 800	struct iattr		*iattr)
 801{
 802	struct xfs_inode	*ip = XFS_I(d_inode(dentry));
 803	int error;
 804
 805	trace_xfs_setattr(ip);
 806
 807	error = xfs_vn_change_ok(dentry, iattr);
 808	if (error)
 809		return error;
 810	return xfs_setattr_nonsize(ip, iattr, 0);
 811}
 812
 813/*
 814 * Truncate file.  Must have write permission and not be a directory.
 815 *
 816 * Caution: The caller of this function is responsible for calling
 817 * setattr_prepare() or otherwise verifying the change is fine.
 818 */
 819STATIC int
 820xfs_setattr_size(
 821	struct xfs_inode	*ip,
 822	struct iattr		*iattr)
 823{
 824	struct xfs_mount	*mp = ip->i_mount;
 825	struct inode		*inode = VFS_I(ip);
 826	xfs_off_t		oldsize, newsize;
 827	struct xfs_trans	*tp;
 828	int			error;
 829	uint			lock_flags = 0;
 830	bool			did_zeroing = false;
 831
 
 
 
 
 
 
 
 
 
 
 
 
 832	ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
 833	ASSERT(xfs_isilocked(ip, XFS_MMAPLOCK_EXCL));
 834	ASSERT(S_ISREG(inode->i_mode));
 835	ASSERT((iattr->ia_valid & (ATTR_UID|ATTR_GID|ATTR_ATIME|ATTR_ATIME_SET|
 836		ATTR_MTIME_SET|ATTR_KILL_PRIV|ATTR_TIMES_SET)) == 0);
 837
 838	oldsize = inode->i_size;
 839	newsize = iattr->ia_size;
 840
 841	/*
 842	 * Short circuit the truncate case for zero length files.
 843	 */
 844	if (newsize == 0 && oldsize == 0 && ip->i_d.di_nextents == 0) {
 845		if (!(iattr->ia_valid & (ATTR_CTIME|ATTR_MTIME)))
 846			return 0;
 847
 848		/*
 849		 * Use the regular setattr path to update the timestamps.
 850		 */
 851		iattr->ia_valid &= ~ATTR_SIZE;
 852		return xfs_setattr_nonsize(ip, iattr, 0);
 853	}
 854
 855	/*
 856	 * Make sure that the dquots are attached to the inode.
 857	 */
 858	error = xfs_qm_dqattach(ip, 0);
 859	if (error)
 860		return error;
 861
 862	/*
 863	 * Wait for all direct I/O to complete.
 864	 */
 865	inode_dio_wait(inode);
 866
 867	/*
 868	 * File data changes must be complete before we start the transaction to
 869	 * modify the inode.  This needs to be done before joining the inode to
 870	 * the transaction because the inode cannot be unlocked once it is a
 871	 * part of the transaction.
 872	 *
 873	 * Start with zeroing any data beyond EOF that we may expose on file
 874	 * extension, or zeroing out the rest of the block on a downward
 875	 * truncate.
 876	 */
 877	if (newsize > oldsize) {
 878		trace_xfs_zero_eof(ip, oldsize, newsize - oldsize);
 879		error = iomap_zero_range(inode, oldsize, newsize - oldsize,
 880				&did_zeroing, &xfs_iomap_ops);
 881	} else {
 882		error = iomap_truncate_page(inode, newsize, &did_zeroing,
 883				&xfs_iomap_ops);
 
 
 
 
 
 
 
 
 
 
 
 
 
 884	}
 885
 886	if (error)
 887		return error;
 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	 * And we update in-core i_size and truncate page cache beyond newsize
 907	 * before writeback the [di_size, newsize] range, so we're guaranteed
 908	 * not to write stale data past the new EOF on truncate down.
 909	 */
 
 
 
 
 
 
 
 910	truncate_setsize(inode, newsize);
 911
 912	/*
 913	 * We are going to log the inode size change in this transaction so
 914	 * any previous writes that are beyond the on disk EOF and the new
 915	 * EOF that have not been written out need to be written here.  If we
 916	 * do not write the data out, we expose ourselves to the null files
 917	 * problem. Note that this includes any block zeroing we did above;
 918	 * otherwise those blocks may not be zeroed after a crash.
 919	 */
 920	if (did_zeroing ||
 921	    (newsize > ip->i_d.di_size && oldsize != ip->i_d.di_size)) {
 922		error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
 923						ip->i_d.di_size, newsize - 1);
 924		if (error)
 925			return error;
 926	}
 927
 928	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
 929	if (error)
 930		return error;
 931
 932	lock_flags |= XFS_ILOCK_EXCL;
 933	xfs_ilock(ip, XFS_ILOCK_EXCL);
 934	xfs_trans_ijoin(tp, ip, 0);
 935
 936	/*
 937	 * Only change the c/mtime if we are changing the size or we are
 938	 * explicitly asked to change it.  This handles the semantic difference
 939	 * between truncate() and ftruncate() as implemented in the VFS.
 940	 *
 941	 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
 942	 * special case where we need to update the times despite not having
 943	 * these flags set.  For all other operations the VFS set these flags
 944	 * explicitly if it wants a timestamp update.
 945	 */
 946	if (newsize != oldsize &&
 947	    !(iattr->ia_valid & (ATTR_CTIME | ATTR_MTIME))) {
 948		iattr->ia_ctime = iattr->ia_mtime =
 949			current_time(inode);
 950		iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME;
 951	}
 952
 953	/*
 954	 * The first thing we do is set the size to new_size permanently on
 955	 * disk.  This way we don't have to worry about anyone ever being able
 956	 * to look at the data being freed even in the face of a crash.
 957	 * What we're getting around here is the case where we free a block, it
 958	 * is allocated to another file, it is written to, and then we crash.
 959	 * If the new data gets written to the file but the log buffers
 960	 * containing the free and reallocation don't, then we'd end up with
 961	 * garbage in the blocks being freed.  As long as we make the new size
 962	 * permanent before actually freeing any blocks it doesn't matter if
 963	 * they get written to.
 964	 */
 965	ip->i_d.di_size = newsize;
 966	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
 967
 968	if (newsize <= oldsize) {
 969		error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, newsize);
 970		if (error)
 971			goto out_trans_cancel;
 972
 973		/*
 974		 * Truncated "down", so we're removing references to old data
 975		 * here - if we delay flushing for a long time, we expose
 976		 * ourselves unduly to the notorious NULL files problem.  So,
 977		 * we mark this inode and flush it when the file is closed,
 978		 * and do not wait the usual (long) time for writeout.
 979		 */
 980		xfs_iflags_set(ip, XFS_ITRUNCATED);
 981
 982		/* A truncate down always removes post-EOF blocks. */
 983		xfs_inode_clear_eofblocks_tag(ip);
 984	}
 985
 986	if (iattr->ia_valid & ATTR_MODE)
 987		xfs_setattr_mode(ip, iattr);
 988	if (iattr->ia_valid & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME))
 989		xfs_setattr_time(ip, iattr);
 990
 991	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
 992
 993	XFS_STATS_INC(mp, xs_ig_attrchg);
 994
 995	if (mp->m_flags & XFS_MOUNT_WSYNC)
 996		xfs_trans_set_sync(tp);
 997
 998	error = xfs_trans_commit(tp);
 999out_unlock:
1000	if (lock_flags)
1001		xfs_iunlock(ip, lock_flags);
1002	return error;
1003
1004out_trans_cancel:
1005	xfs_trans_cancel(tp);
1006	goto out_unlock;
1007}
1008
1009int
1010xfs_vn_setattr_size(
1011	struct dentry		*dentry,
1012	struct iattr		*iattr)
1013{
1014	struct xfs_inode	*ip = XFS_I(d_inode(dentry));
1015	int error;
1016
1017	trace_xfs_setattr(ip);
1018
1019	error = xfs_vn_change_ok(dentry, iattr);
1020	if (error)
1021		return error;
1022	return xfs_setattr_size(ip, iattr);
1023}
1024
1025STATIC int
1026xfs_vn_setattr(
1027	struct dentry		*dentry,
1028	struct iattr		*iattr)
1029{
 
1030	int			error;
1031
1032	if (iattr->ia_valid & ATTR_SIZE) {
1033		struct xfs_inode	*ip = XFS_I(d_inode(dentry));
1034		uint			iolock = XFS_IOLOCK_EXCL;
1035
1036		error = xfs_break_layouts(d_inode(dentry), &iolock);
1037		if (error)
1038			return error;
 
 
1039
1040		xfs_ilock(ip, XFS_MMAPLOCK_EXCL);
1041		error = xfs_vn_setattr_size(dentry, iattr);
1042		xfs_iunlock(ip, XFS_MMAPLOCK_EXCL);
1043	} else {
1044		error = xfs_vn_setattr_nonsize(dentry, iattr);
1045	}
1046
1047	return error;
1048}
1049
1050STATIC int
1051xfs_vn_update_time(
1052	struct inode		*inode,
1053	struct timespec		*now,
1054	int			flags)
1055{
1056	struct xfs_inode	*ip = XFS_I(inode);
1057	struct xfs_mount	*mp = ip->i_mount;
1058	int			log_flags = XFS_ILOG_TIMESTAMP;
1059	struct xfs_trans	*tp;
1060	int			error;
1061
1062	trace_xfs_update_time(ip);
1063
1064	if (inode->i_sb->s_flags & SB_LAZYTIME) {
1065		if (!((flags & S_VERSION) &&
1066		      inode_maybe_inc_iversion(inode, false)))
1067			return generic_update_time(inode, now, flags);
1068
1069		/* Capture the iversion update that just occurred */
1070		log_flags |= XFS_ILOG_CORE;
1071	}
1072
1073	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp);
1074	if (error)
1075		return error;
1076
1077	xfs_ilock(ip, XFS_ILOCK_EXCL);
1078	if (flags & S_CTIME)
1079		inode->i_ctime = *now;
1080	if (flags & S_MTIME)
1081		inode->i_mtime = *now;
1082	if (flags & S_ATIME)
1083		inode->i_atime = *now;
1084
1085	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1086	xfs_trans_log_inode(tp, ip, log_flags);
1087	return xfs_trans_commit(tp);
1088}
1089
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1090STATIC int
1091xfs_vn_fiemap(
1092	struct inode		*inode,
1093	struct fiemap_extent_info *fieinfo,
1094	u64			start,
1095	u64			length)
1096{
 
 
1097	int			error;
1098
1099	xfs_ilock(XFS_I(inode), XFS_IOLOCK_SHARED);
1100	if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
1101		fieinfo->fi_flags &= ~FIEMAP_FLAG_XATTR;
1102		error = iomap_fiemap(inode, fieinfo, start, length,
1103				&xfs_xattr_iomap_ops);
1104	} else {
1105		error = iomap_fiemap(inode, fieinfo, start, length,
1106				&xfs_iomap_ops);
1107	}
1108	xfs_iunlock(XFS_I(inode), XFS_IOLOCK_SHARED);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1109
1110	return error;
1111}
1112
1113STATIC int
1114xfs_vn_tmpfile(
1115	struct inode	*dir,
1116	struct dentry	*dentry,
1117	umode_t		mode)
1118{
1119	return xfs_generic_create(dir, dentry, mode, 0, true);
1120}
1121
1122static const struct inode_operations xfs_inode_operations = {
1123	.get_acl		= xfs_get_acl,
1124	.set_acl		= xfs_set_acl,
1125	.getattr		= xfs_vn_getattr,
1126	.setattr		= xfs_vn_setattr,
 
 
 
1127	.listxattr		= xfs_vn_listxattr,
1128	.fiemap			= xfs_vn_fiemap,
1129	.update_time		= xfs_vn_update_time,
1130};
1131
1132static const struct inode_operations xfs_dir_inode_operations = {
1133	.create			= xfs_vn_create,
1134	.lookup			= xfs_vn_lookup,
1135	.link			= xfs_vn_link,
1136	.unlink			= xfs_vn_unlink,
1137	.symlink		= xfs_vn_symlink,
1138	.mkdir			= xfs_vn_mkdir,
1139	/*
1140	 * Yes, XFS uses the same method for rmdir and unlink.
1141	 *
1142	 * There are some subtile differences deeper in the code,
1143	 * but we use S_ISDIR to check for those.
1144	 */
1145	.rmdir			= xfs_vn_unlink,
1146	.mknod			= xfs_vn_mknod,
1147	.rename			= xfs_vn_rename,
1148	.get_acl		= xfs_get_acl,
1149	.set_acl		= xfs_set_acl,
1150	.getattr		= xfs_vn_getattr,
1151	.setattr		= xfs_vn_setattr,
 
 
 
1152	.listxattr		= xfs_vn_listxattr,
1153	.update_time		= xfs_vn_update_time,
1154	.tmpfile		= xfs_vn_tmpfile,
1155};
1156
1157static const struct inode_operations xfs_dir_ci_inode_operations = {
1158	.create			= xfs_vn_create,
1159	.lookup			= xfs_vn_ci_lookup,
1160	.link			= xfs_vn_link,
1161	.unlink			= xfs_vn_unlink,
1162	.symlink		= xfs_vn_symlink,
1163	.mkdir			= xfs_vn_mkdir,
1164	/*
1165	 * Yes, XFS uses the same method for rmdir and unlink.
1166	 *
1167	 * There are some subtile differences deeper in the code,
1168	 * but we use S_ISDIR to check for those.
1169	 */
1170	.rmdir			= xfs_vn_unlink,
1171	.mknod			= xfs_vn_mknod,
1172	.rename			= xfs_vn_rename,
1173	.get_acl		= xfs_get_acl,
1174	.set_acl		= xfs_set_acl,
1175	.getattr		= xfs_vn_getattr,
1176	.setattr		= xfs_vn_setattr,
 
 
 
1177	.listxattr		= xfs_vn_listxattr,
1178	.update_time		= xfs_vn_update_time,
1179	.tmpfile		= xfs_vn_tmpfile,
1180};
1181
1182static const struct inode_operations xfs_symlink_inode_operations = {
 
1183	.get_link		= xfs_vn_get_link,
1184	.getattr		= xfs_vn_getattr,
1185	.setattr		= xfs_vn_setattr,
1186	.listxattr		= xfs_vn_listxattr,
1187	.update_time		= xfs_vn_update_time,
1188};
1189
1190static const struct inode_operations xfs_inline_symlink_inode_operations = {
1191	.get_link		= xfs_vn_get_link_inline,
1192	.getattr		= xfs_vn_getattr,
1193	.setattr		= xfs_vn_setattr,
1194	.listxattr		= xfs_vn_listxattr,
1195	.update_time		= xfs_vn_update_time,
1196};
1197
1198STATIC void
1199xfs_diflags_to_iflags(
1200	struct inode		*inode,
1201	struct xfs_inode	*ip)
1202{
1203	uint16_t		flags = ip->i_d.di_flags;
1204
1205	inode->i_flags &= ~(S_IMMUTABLE | S_APPEND | S_SYNC |
1206			    S_NOATIME | S_DAX);
1207
1208	if (flags & XFS_DIFLAG_IMMUTABLE)
1209		inode->i_flags |= S_IMMUTABLE;
1210	if (flags & XFS_DIFLAG_APPEND)
1211		inode->i_flags |= S_APPEND;
1212	if (flags & XFS_DIFLAG_SYNC)
1213		inode->i_flags |= S_SYNC;
1214	if (flags & XFS_DIFLAG_NOATIME)
1215		inode->i_flags |= S_NOATIME;
1216	if (S_ISREG(inode->i_mode) &&
1217	    ip->i_mount->m_sb.sb_blocksize == PAGE_SIZE &&
1218	    !xfs_is_reflink_inode(ip) &&
1219	    (ip->i_mount->m_flags & XFS_MOUNT_DAX ||
1220	     ip->i_d.di_flags2 & XFS_DIFLAG2_DAX))
1221		inode->i_flags |= S_DAX;
1222}
1223
1224/*
1225 * Initialize the Linux inode.
1226 *
1227 * When reading existing inodes from disk this is called directly from xfs_iget,
1228 * when creating a new inode it is called from xfs_ialloc after setting up the
1229 * inode. These callers have different criteria for clearing XFS_INEW, so leave
1230 * it up to the caller to deal with unlocking the inode appropriately.
1231 */
1232void
1233xfs_setup_inode(
1234	struct xfs_inode	*ip)
1235{
1236	struct inode		*inode = &ip->i_vnode;
1237	gfp_t			gfp_mask;
1238
1239	inode->i_ino = ip->i_ino;
1240	inode->i_state = I_NEW;
1241
1242	inode_sb_list_add(inode);
1243	/* make the inode look hashed for the writeback code */
1244	hlist_add_fake(&inode->i_hash);
1245
1246	inode->i_uid    = xfs_uid_to_kuid(ip->i_d.di_uid);
1247	inode->i_gid    = xfs_gid_to_kgid(ip->i_d.di_gid);
1248
 
 
 
 
 
 
 
 
 
 
 
 
1249	i_size_write(inode, ip->i_d.di_size);
1250	xfs_diflags_to_iflags(inode, ip);
1251
1252	if (S_ISDIR(inode->i_mode)) {
 
 
 
 
 
 
 
 
1253		lockdep_set_class(&ip->i_lock.mr_lock, &xfs_dir_ilock_class);
 
 
 
 
 
1254		ip->d_ops = ip->i_mount->m_dir_inode_ops;
1255	} else {
1256		ip->d_ops = ip->i_mount->m_nondir_inode_ops;
1257		lockdep_set_class(&ip->i_lock.mr_lock, &xfs_nondir_ilock_class);
 
 
 
 
 
 
 
1258	}
1259
1260	/*
1261	 * Ensure all page cache allocations are done from GFP_NOFS context to
1262	 * prevent direct reclaim recursion back into the filesystem and blowing
1263	 * stacks or deadlocking.
1264	 */
1265	gfp_mask = mapping_gfp_mask(inode->i_mapping);
1266	mapping_set_gfp_mask(inode->i_mapping, (gfp_mask & ~(__GFP_FS)));
1267
1268	/*
1269	 * If there is no attribute fork no ACL can exist on this inode,
1270	 * and it can't have any file capabilities attached to it either.
1271	 */
1272	if (!XFS_IFORK_Q(ip)) {
1273		inode_has_no_xattr(inode);
1274		cache_no_acl(inode);
1275	}
1276}
1277
1278void
1279xfs_setup_iops(
1280	struct xfs_inode	*ip)
1281{
1282	struct inode		*inode = &ip->i_vnode;
1283
1284	switch (inode->i_mode & S_IFMT) {
1285	case S_IFREG:
1286		inode->i_op = &xfs_inode_operations;
1287		inode->i_fop = &xfs_file_operations;
1288		if (IS_DAX(inode))
1289			inode->i_mapping->a_ops = &xfs_dax_aops;
1290		else
1291			inode->i_mapping->a_ops = &xfs_address_space_operations;
1292		break;
1293	case S_IFDIR:
1294		if (xfs_sb_version_hasasciici(&XFS_M(inode->i_sb)->m_sb))
1295			inode->i_op = &xfs_dir_ci_inode_operations;
1296		else
1297			inode->i_op = &xfs_dir_inode_operations;
1298		inode->i_fop = &xfs_dir_file_operations;
1299		break;
1300	case S_IFLNK:
1301		if (ip->i_df.if_flags & XFS_IFINLINE)
1302			inode->i_op = &xfs_inline_symlink_inode_operations;
1303		else
1304			inode->i_op = &xfs_symlink_inode_operations;
1305		break;
1306	default:
1307		inode->i_op = &xfs_inode_operations;
1308		init_special_inode(inode, inode->i_mode, inode->i_rdev);
1309		break;
1310	}
1311}