xfs_iops.c

   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
   4 * All Rights Reserved.
   5 */
   6#include "xfs.h"
   7#include "xfs_fs.h"
   8#include "xfs_shared.h"
   9#include "xfs_format.h"
  10#include "xfs_log_format.h"
  11#include "xfs_trans_resv.h"
  12#include "xfs_mount.h"
  13#include "xfs_inode.h"
  14#include "xfs_acl.h"
  15#include "xfs_quota.h"
  16#include "xfs_da_format.h"
  17#include "xfs_da_btree.h"
  18#include "xfs_attr.h"
  19#include "xfs_trans.h"
  20#include "xfs_trans_space.h"
  21#include "xfs_bmap_btree.h"
  22#include "xfs_trace.h"
  23#include "xfs_icache.h"
  24#include "xfs_symlink.h"
  25#include "xfs_dir2.h"
  26#include "xfs_iomap.h"
  27#include "xfs_error.h"
  28#include "xfs_ioctl.h"
  29#include "xfs_xattr.h"
  30#include "xfs_file.h"
  31#include "xfs_bmap.h"
  32#include "xfs_zone_alloc.h"
  33
  34#include <linux/posix_acl.h>
  35#include <linux/security.h>
  36#include <linux/iversion.h>
  37#include <linux/fiemap.h>
  38
  39/*
  40 * Directories have different lock order w.r.t. mmap_lock compared to regular
  41 * files. This is due to readdir potentially triggering page faults on a user
  42 * buffer inside filldir(), and this happens with the ilock on the directory
  43 * held. For regular files, the lock order is the other way around - the
  44 * mmap_lock is taken during the page fault, and then we lock the ilock to do
  45 * block mapping. Hence we need a different class for the directory ilock so
  46 * that lockdep can tell them apart.  Directories in the metadata directory
  47 * tree get a separate class so that lockdep reports will warn us if someone
  48 * ever tries to lock regular directories after locking metadata directories.
  49 */
  50static struct lock_class_key xfs_nondir_ilock_class;
  51static struct lock_class_key xfs_dir_ilock_class;
  52
  53static int
  54xfs_initxattrs(
  55	struct inode		*inode,
  56	const struct xattr	*xattr_array,
  57	void			*fs_info)
  58{
  59	const struct xattr	*xattr;
  60	struct xfs_inode	*ip = XFS_I(inode);
  61	int			error = 0;
  62
  63	for (xattr = xattr_array; xattr->name != NULL; xattr++) {
  64		struct xfs_da_args	args = {
  65			.dp		= ip,
  66			.attr_filter	= XFS_ATTR_SECURE,
  67			.name		= xattr->name,
  68			.namelen	= strlen(xattr->name),
  69			.value		= xattr->value,
  70			.valuelen	= xattr->value_len,
  71		};
  72		error = xfs_attr_change(&args, XFS_ATTRUPDATE_UPSERT);
  73		if (error < 0)
  74			break;
  75	}
  76	return error;
  77}
  78
  79/*
  80 * Hook in SELinux.  This is not quite correct yet, what we really need
  81 * here (as we do for default ACLs) is a mechanism by which creation of
  82 * these attrs can be journalled at inode creation time (along with the
  83 * inode, of course, such that log replay can't cause these to be lost).
  84 */
  85int
  86xfs_inode_init_security(
  87	struct inode	*inode,
  88	struct inode	*dir,
  89	const struct qstr *qstr)
  90{
  91	return security_inode_init_security(inode, dir, qstr,
  92					     &xfs_initxattrs, NULL);
  93}
  94
  95static void
  96xfs_dentry_to_name(
  97	struct xfs_name	*namep,
  98	struct dentry	*dentry)
  99{
 100	namep->name = dentry->d_name.name;
 101	namep->len = dentry->d_name.len;
 102	namep->type = XFS_DIR3_FT_UNKNOWN;
 103}
 104
 105static int
 106xfs_dentry_mode_to_name(
 107	struct xfs_name	*namep,
 108	struct dentry	*dentry,
 109	int		mode)
 110{
 111	namep->name = dentry->d_name.name;
 112	namep->len = dentry->d_name.len;
 113	namep->type = xfs_mode_to_ftype(mode);
 114
 115	if (unlikely(namep->type == XFS_DIR3_FT_UNKNOWN))
 116		return -EFSCORRUPTED;
 117
 118	return 0;
 119}
 120
 121STATIC void
 122xfs_cleanup_inode(
 123	struct inode	*dir,
 124	struct inode	*inode,
 125	struct dentry	*dentry)
 126{
 127	struct xfs_name	teardown;
 128
 129	/* Oh, the horror.
 130	 * If we can't add the ACL or we fail in
 131	 * xfs_inode_init_security we must back out.
 132	 * ENOSPC can hit here, among other things.
 133	 */
 134	xfs_dentry_to_name(&teardown, dentry);
 135
 136	xfs_remove(XFS_I(dir), &teardown, XFS_I(inode));
 137}
 138
 139/*
 140 * Check to see if we are likely to need an extended attribute to be added to
 141 * the inode we are about to allocate. This allows the attribute fork to be
 142 * created during the inode allocation, reducing the number of transactions we
 143 * need to do in this fast path.
 144 *
 145 * The security checks are optimistic, but not guaranteed. The two LSMs that
 146 * require xattrs to be added here (selinux and smack) are also the only two
 147 * LSMs that add a sb->s_security structure to the superblock. Hence if security
 148 * is enabled and sb->s_security is set, we have a pretty good idea that we are
 149 * going to be asked to add a security xattr immediately after allocating the
 150 * xfs inode and instantiating the VFS inode.
 151 */
 152static inline bool
 153xfs_create_need_xattr(
 154	struct inode	*dir,
 155	struct posix_acl *default_acl,
 156	struct posix_acl *acl)
 157{
 158	if (acl)
 159		return true;
 160	if (default_acl)
 161		return true;
 162#if IS_ENABLED(CONFIG_SECURITY)
 163	if (dir->i_sb->s_security)
 164		return true;
 165#endif
 166	return false;
 167}
 168
 169
 170STATIC int
 171xfs_generic_create(
 172	struct mnt_idmap	*idmap,
 173	struct inode		*dir,
 174	struct dentry		*dentry,
 175	umode_t			mode,
 176	dev_t			rdev,
 177	struct file		*tmpfile)	/* unnamed file */
 178{
 179	struct xfs_icreate_args	args = {
 180		.idmap		= idmap,
 181		.pip		= XFS_I(dir),
 182		.rdev		= rdev,
 183		.mode		= mode,
 184	};
 185	struct inode		*inode;
 186	struct xfs_inode	*ip = NULL;
 187	struct posix_acl	*default_acl, *acl;
 188	struct xfs_name		name;
 189	int			error;
 190
 191	/*
 192	 * Irix uses Missed'em'V split, but doesn't want to see
 193	 * the upper 5 bits of (14bit) major.
 194	 */
 195	if (S_ISCHR(args.mode) || S_ISBLK(args.mode)) {
 196		if (unlikely(!sysv_valid_dev(args.rdev) ||
 197			     MAJOR(args.rdev) & ~0x1ff))
 198			return -EINVAL;
 199	} else {
 200		args.rdev = 0;
 201	}
 202
 203	error = posix_acl_create(dir, &args.mode, &default_acl, &acl);
 204	if (error)
 205		return error;
 206
 207	/* Verify mode is valid also for tmpfile case */
 208	error = xfs_dentry_mode_to_name(&name, dentry, args.mode);
 209	if (unlikely(error))
 210		goto out_free_acl;
 211
 212	if (!tmpfile) {
 213		if (xfs_create_need_xattr(dir, default_acl, acl))
 214			args.flags |= XFS_ICREATE_INIT_XATTRS;
 215
 216		error = xfs_create(&args, &name, &ip);
 217	} else {
 218		args.flags |= XFS_ICREATE_TMPFILE;
 219
 220		/*
 221		 * If this temporary file will not be linkable, don't bother
 222		 * creating an attr fork to receive a parent pointer.
 223		 */
 224		if (tmpfile->f_flags & O_EXCL)
 225			args.flags |= XFS_ICREATE_UNLINKABLE;
 226
 227		error = xfs_create_tmpfile(&args, &ip);
 228	}
 229	if (unlikely(error))
 230		goto out_free_acl;
 231
 232	inode = VFS_I(ip);
 233
 234	error = xfs_inode_init_security(inode, dir, &dentry->d_name);
 235	if (unlikely(error))
 236		goto out_cleanup_inode;
 237
 238	if (default_acl) {
 239		error = __xfs_set_acl(inode, default_acl, ACL_TYPE_DEFAULT);
 240		if (error)
 241			goto out_cleanup_inode;
 242	}
 243	if (acl) {
 244		error = __xfs_set_acl(inode, acl, ACL_TYPE_ACCESS);
 245		if (error)
 246			goto out_cleanup_inode;
 247	}
 248
 249	xfs_setup_iops(ip);
 250
 251	if (tmpfile) {
 252		/*
 253		 * The VFS requires that any inode fed to d_tmpfile must have
 254		 * nlink == 1 so that it can decrement the nlink in d_tmpfile.
 255		 * However, we created the temp file with nlink == 0 because
 256		 * we're not allowed to put an inode with nlink > 0 on the
 257		 * unlinked list.  Therefore we have to set nlink to 1 so that
 258		 * d_tmpfile can immediately set it back to zero.
 259		 */
 260		set_nlink(inode, 1);
 261		d_tmpfile(tmpfile, inode);
 262	} else
 263		d_instantiate(dentry, inode);
 264
 265	xfs_finish_inode_setup(ip);
 266
 267 out_free_acl:
 268	posix_acl_release(default_acl);
 269	posix_acl_release(acl);
 270	return error;
 271
 272 out_cleanup_inode:
 273	xfs_finish_inode_setup(ip);
 274	if (!tmpfile)
 275		xfs_cleanup_inode(dir, inode, dentry);
 276	xfs_irele(ip);
 277	goto out_free_acl;
 278}
 279
 280STATIC int
 281xfs_vn_mknod(
 282	struct mnt_idmap	*idmap,
 283	struct inode		*dir,
 284	struct dentry		*dentry,
 285	umode_t			mode,
 286	dev_t			rdev)
 287{
 288	return xfs_generic_create(idmap, dir, dentry, mode, rdev, NULL);
 289}
 290
 291STATIC int
 292xfs_vn_create(
 293	struct mnt_idmap	*idmap,
 294	struct inode		*dir,
 295	struct dentry		*dentry,
 296	umode_t			mode,
 297	bool			flags)
 298{
 299	return xfs_generic_create(idmap, dir, dentry, mode, 0, NULL);
 300}
 301
 302STATIC struct dentry *
 303xfs_vn_mkdir(
 304	struct mnt_idmap	*idmap,
 305	struct inode		*dir,
 306	struct dentry		*dentry,
 307	umode_t			mode)
 308{
 309	return ERR_PTR(xfs_generic_create(idmap, dir, dentry, mode | S_IFDIR, 0, NULL));
 310}
 311
 312STATIC struct dentry *
 313xfs_vn_lookup(
 314	struct inode	*dir,
 315	struct dentry	*dentry,
 316	unsigned int flags)
 317{
 318	struct inode *inode;
 319	struct xfs_inode *cip;
 320	struct xfs_name	name;
 321	int		error;
 322
 323	if (dentry->d_name.len >= MAXNAMELEN)
 324		return ERR_PTR(-ENAMETOOLONG);
 325
 326	xfs_dentry_to_name(&name, dentry);
 327	error = xfs_lookup(XFS_I(dir), &name, &cip, NULL);
 328	if (likely(!error))
 329		inode = VFS_I(cip);
 330	else if (likely(error == -ENOENT))
 331		inode = NULL;
 332	else
 333		inode = ERR_PTR(error);
 334	return d_splice_alias(inode, dentry);
 335}
 336
 337STATIC struct dentry *
 338xfs_vn_ci_lookup(
 339	struct inode	*dir,
 340	struct dentry	*dentry,
 341	unsigned int flags)
 342{
 343	struct xfs_inode *ip;
 344	struct xfs_name	xname;
 345	struct xfs_name ci_name;
 346	struct qstr	dname;
 347	int		error;
 348
 349	if (dentry->d_name.len >= MAXNAMELEN)
 350		return ERR_PTR(-ENAMETOOLONG);
 351
 352	xfs_dentry_to_name(&xname, dentry);
 353	error = xfs_lookup(XFS_I(dir), &xname, &ip, &ci_name);
 354	if (unlikely(error)) {
 355		if (unlikely(error != -ENOENT))
 356			return ERR_PTR(error);
 357		/*
 358		 * call d_add(dentry, NULL) here when d_drop_negative_children
 359		 * is called in xfs_vn_mknod (ie. allow negative dentries
 360		 * with CI filesystems).
 361		 */
 362		return NULL;
 363	}
 364
 365	/* if exact match, just splice and exit */
 366	if (!ci_name.name)
 367		return d_splice_alias(VFS_I(ip), dentry);
 368
 369	/* else case-insensitive match... */
 370	dname.name = ci_name.name;
 371	dname.len = ci_name.len;
 372	dentry = d_add_ci(dentry, VFS_I(ip), &dname);
 373	kfree(ci_name.name);
 374	return dentry;
 375}
 376
 377STATIC int
 378xfs_vn_link(
 379	struct dentry	*old_dentry,
 380	struct inode	*dir,
 381	struct dentry	*dentry)
 382{
 383	struct inode	*inode = d_inode(old_dentry);
 384	struct xfs_name	name;
 385	int		error;
 386
 387	error = xfs_dentry_mode_to_name(&name, dentry, inode->i_mode);
 388	if (unlikely(error))
 389		return error;
 390
 391	if (IS_PRIVATE(inode))
 392		return -EPERM;
 393
 394	error = xfs_link(XFS_I(dir), XFS_I(inode), &name);
 395	if (unlikely(error))
 396		return error;
 397
 398	ihold(inode);
 399	d_instantiate(dentry, inode);
 400	return 0;
 401}
 402
 403STATIC int
 404xfs_vn_unlink(
 405	struct inode	*dir,
 406	struct dentry	*dentry)
 407{
 408	struct xfs_name	name;
 409	int		error;
 410
 411	xfs_dentry_to_name(&name, dentry);
 412
 413	error = xfs_remove(XFS_I(dir), &name, XFS_I(d_inode(dentry)));
 414	if (error)
 415		return error;
 416
 417	/*
 418	 * With unlink, the VFS makes the dentry "negative": no inode,
 419	 * but still hashed. This is incompatible with case-insensitive
 420	 * mode, so invalidate (unhash) the dentry in CI-mode.
 421	 */
 422	if (xfs_has_asciici(XFS_M(dir->i_sb)))
 423		d_invalidate(dentry);
 424	return 0;
 425}
 426
 427STATIC int
 428xfs_vn_symlink(
 429	struct mnt_idmap	*idmap,
 430	struct inode		*dir,
 431	struct dentry		*dentry,
 432	const char		*symname)
 433{
 434	struct inode		*inode;
 435	struct xfs_inode	*cip = NULL;
 436	struct xfs_name		name;
 437	int			error;
 438	umode_t			mode = S_IFLNK | S_IRWXUGO;
 439
 440	error = xfs_dentry_mode_to_name(&name, dentry, mode);
 441	if (unlikely(error))
 442		goto out;
 443
 444	error = xfs_symlink(idmap, XFS_I(dir), &name, symname, mode, &cip);
 445	if (unlikely(error))
 446		goto out;
 447
 448	inode = VFS_I(cip);
 449
 450	error = xfs_inode_init_security(inode, dir, &dentry->d_name);
 451	if (unlikely(error))
 452		goto out_cleanup_inode;
 453
 454	xfs_setup_iops(cip);
 455
 456	d_instantiate(dentry, inode);
 457	xfs_finish_inode_setup(cip);
 458	return 0;
 459
 460 out_cleanup_inode:
 461	xfs_finish_inode_setup(cip);
 462	xfs_cleanup_inode(dir, inode, dentry);
 463	xfs_irele(cip);
 464 out:
 465	return error;
 466}
 467
 468STATIC int
 469xfs_vn_rename(
 470	struct mnt_idmap	*idmap,
 471	struct inode		*odir,
 472	struct dentry		*odentry,
 473	struct inode		*ndir,
 474	struct dentry		*ndentry,
 475	unsigned int		flags)
 476{
 477	struct inode	*new_inode = d_inode(ndentry);
 478	int		omode = 0;
 479	int		error;
 480	struct xfs_name	oname;
 481	struct xfs_name	nname;
 482
 483	if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
 484		return -EINVAL;
 485
 486	/* if we are exchanging files, we need to set i_mode of both files */
 487	if (flags & RENAME_EXCHANGE)
 488		omode = d_inode(ndentry)->i_mode;
 489
 490	error = xfs_dentry_mode_to_name(&oname, odentry, omode);
 491	if (omode && unlikely(error))
 492		return error;
 493
 494	error = xfs_dentry_mode_to_name(&nname, ndentry,
 495					d_inode(odentry)->i_mode);
 496	if (unlikely(error))
 497		return error;
 498
 499	return xfs_rename(idmap, XFS_I(odir), &oname,
 500			  XFS_I(d_inode(odentry)), XFS_I(ndir), &nname,
 501			  new_inode ? XFS_I(new_inode) : NULL, flags);
 502}
 503
 504/*
 505 * careful here - this function can get called recursively, so
 506 * we need to be very careful about how much stack we use.
 507 * uio is kmalloced for this reason...
 508 */
 509STATIC const char *
 510xfs_vn_get_link(
 511	struct dentry		*dentry,
 512	struct inode		*inode,
 513	struct delayed_call	*done)
 514{
 515	char			*link;
 516	int			error = -ENOMEM;
 517
 518	if (!dentry)
 519		return ERR_PTR(-ECHILD);
 520
 521	link = kmalloc(XFS_SYMLINK_MAXLEN+1, GFP_KERNEL);
 522	if (!link)
 523		goto out_err;
 524
 525	error = xfs_readlink(XFS_I(d_inode(dentry)), link);
 526	if (unlikely(error))
 527		goto out_kfree;
 528
 529	set_delayed_call(done, kfree_link, link);
 530	return link;
 531
 532 out_kfree:
 533	kfree(link);
 534 out_err:
 535	return ERR_PTR(error);
 536}
 537
 538static uint32_t
 539xfs_stat_blksize(
 540	struct xfs_inode	*ip)
 541{
 542	struct xfs_mount	*mp = ip->i_mount;
 543
 544	/*
 545	 * If the file blocks are being allocated from a realtime volume, then
 546	 * always return the realtime extent size.
 547	 */
 548	if (XFS_IS_REALTIME_INODE(ip))
 549		return XFS_FSB_TO_B(mp, xfs_get_extsz_hint(ip) ? : 1);
 550
 551	/*
 552	 * Allow large block sizes to be reported to userspace programs if the
 553	 * "largeio" mount option is used.
 554	 *
 555	 * If compatibility mode is specified, simply return the basic unit of
 556	 * caching so that we don't get inefficient read/modify/write I/O from
 557	 * user apps. Otherwise....
 558	 *
 559	 * If the underlying volume is a stripe, then return the stripe width in
 560	 * bytes as the recommended I/O size. It is not a stripe and we've set a
 561	 * default buffered I/O size, return that, otherwise return the compat
 562	 * default.
 563	 */
 564	if (xfs_has_large_iosize(mp)) {
 565		if (mp->m_swidth)
 566			return XFS_FSB_TO_B(mp, mp->m_swidth);
 567		if (xfs_has_allocsize(mp))
 568			return 1U << mp->m_allocsize_log;
 569	}
 570
 571	return max_t(uint32_t, PAGE_SIZE, mp->m_sb.sb_blocksize);
 572}
 573
 574static void
 575xfs_report_dioalign(
 576	struct xfs_inode	*ip,
 577	struct kstat		*stat)
 578{
 579	struct xfs_buftarg	*target = xfs_inode_buftarg(ip);
 580	struct block_device	*bdev = target->bt_bdev;
 581
 582	stat->result_mask |= STATX_DIOALIGN | STATX_DIO_READ_ALIGN;
 583	stat->dio_mem_align = bdev_dma_alignment(bdev) + 1;
 584
 585	/*
 586	 * For COW inodes, we can only perform out of place writes of entire
 587	 * allocation units (blocks or RT extents).
 588	 * For writes smaller than the allocation unit, we must fall back to
 589	 * buffered I/O to perform read-modify-write cycles.  At best this is
 590	 * highly inefficient; at worst it leads to page cache invalidation
 591	 * races.  Tell applications to avoid this by reporting the larger write
 592	 * alignment in dio_offset_align, and the smaller read alignment in
 593	 * dio_read_offset_align.
 594	 */
 595	stat->dio_read_offset_align = bdev_logical_block_size(bdev);
 596	if (xfs_is_cow_inode(ip))
 597		stat->dio_offset_align = xfs_inode_alloc_unitsize(ip);
 598	else
 599		stat->dio_offset_align = stat->dio_read_offset_align;
 600}
 601
 602unsigned int
 603xfs_get_atomic_write_min(
 604	struct xfs_inode	*ip)
 605{
 606	struct xfs_mount	*mp = ip->i_mount;
 607
 608	/*
 609	 * If we can complete an atomic write via atomic out of place writes,
 610	 * then advertise a minimum size of one fsblock.  Without this
 611	 * mechanism, we can only guarantee atomic writes up to a single LBA.
 612	 *
 613	 * If out of place writes are not available, we can guarantee an atomic
 614	 * write of exactly one single fsblock if the bdev will make that
 615	 * guarantee for us.
 616	 */
 617	if (xfs_inode_can_hw_atomic_write(ip) ||
 618	    xfs_inode_can_sw_atomic_write(ip))
 619		return mp->m_sb.sb_blocksize;
 620
 621	return 0;
 622}
 623
 624unsigned int
 625xfs_get_atomic_write_max(
 626	struct xfs_inode	*ip)
 627{
 628	struct xfs_mount	*mp = ip->i_mount;
 629
 630	/*
 631	 * If out of place writes are not available, we can guarantee an atomic
 632	 * write of exactly one single fsblock if the bdev will make that
 633	 * guarantee for us.
 634	 */
 635	if (!xfs_inode_can_sw_atomic_write(ip)) {
 636		if (xfs_inode_can_hw_atomic_write(ip))
 637			return mp->m_sb.sb_blocksize;
 638		return 0;
 639	}
 640
 641	/*
 642	 * If we can complete an atomic write via atomic out of place writes,
 643	 * then advertise a maximum size of whatever we can complete through
 644	 * that means.  Hardware support is reported via max_opt, not here.
 645	 */
 646	if (XFS_IS_REALTIME_INODE(ip))
 647		return XFS_FSB_TO_B(mp, mp->m_groups[XG_TYPE_RTG].awu_max);
 648	return XFS_FSB_TO_B(mp, mp->m_groups[XG_TYPE_AG].awu_max);
 649}
 650
 651unsigned int
 652xfs_get_atomic_write_max_opt(
 653	struct xfs_inode	*ip)
 654{
 655	unsigned int		awu_max = xfs_get_atomic_write_max(ip);
 656
 657	/* if the max is 1x block, then just keep behaviour that opt is 0 */
 658	if (awu_max <= ip->i_mount->m_sb.sb_blocksize)
 659		return 0;
 660
 661	/*
 662	 * Advertise the maximum size of an atomic write that we can tell the
 663	 * block device to perform for us.  In general the bdev limit will be
 664	 * less than our out of place write limit, but we don't want to exceed
 665	 * the awu_max.
 666	 */
 667	return min(awu_max, xfs_inode_buftarg(ip)->bt_awu_max);
 668}
 669
 670static void
 671xfs_report_atomic_write(
 672	struct xfs_inode	*ip,
 673	struct kstat		*stat)
 674{
 675	generic_fill_statx_atomic_writes(stat,
 676			xfs_get_atomic_write_min(ip),
 677			xfs_get_atomic_write_max(ip),
 678			xfs_get_atomic_write_max_opt(ip));
 679}
 680
 681STATIC int
 682xfs_vn_getattr(
 683	struct mnt_idmap	*idmap,
 684	const struct path	*path,
 685	struct kstat		*stat,
 686	u32			request_mask,
 687	unsigned int		query_flags)
 688{
 689	struct inode		*inode = d_inode(path->dentry);
 690	struct xfs_inode	*ip = XFS_I(inode);
 691	struct xfs_mount	*mp = ip->i_mount;
 692	vfsuid_t		vfsuid = i_uid_into_vfsuid(idmap, inode);
 693	vfsgid_t		vfsgid = i_gid_into_vfsgid(idmap, inode);
 694
 695	trace_xfs_getattr(ip);
 696
 697	if (xfs_is_shutdown(mp))
 698		return -EIO;
 699
 700	stat->size = XFS_ISIZE(ip);
 701	stat->dev = inode->i_sb->s_dev;
 702	stat->mode = inode->i_mode;
 703	stat->nlink = inode->i_nlink;
 704	stat->uid = vfsuid_into_kuid(vfsuid);
 705	stat->gid = vfsgid_into_kgid(vfsgid);
 706	stat->ino = ip->i_ino;
 707	stat->atime = inode_get_atime(inode);
 708
 709	fill_mg_cmtime(stat, request_mask, inode);
 710
 711	stat->blocks = XFS_FSB_TO_BB(mp, ip->i_nblocks + ip->i_delayed_blks);
 712
 713	if (xfs_has_v3inodes(mp)) {
 714		if (request_mask & STATX_BTIME) {
 715			stat->result_mask |= STATX_BTIME;
 716			stat->btime = ip->i_crtime;
 717		}
 718	}
 719
 720	/*
 721	 * Note: If you add another clause to set an attribute flag, please
 722	 * update attributes_mask below.
 723	 */
 724	if (ip->i_diflags & XFS_DIFLAG_IMMUTABLE)
 725		stat->attributes |= STATX_ATTR_IMMUTABLE;
 726	if (ip->i_diflags & XFS_DIFLAG_APPEND)
 727		stat->attributes |= STATX_ATTR_APPEND;
 728	if (ip->i_diflags & XFS_DIFLAG_NODUMP)
 729		stat->attributes |= STATX_ATTR_NODUMP;
 730
 731	stat->attributes_mask |= (STATX_ATTR_IMMUTABLE |
 732				  STATX_ATTR_APPEND |
 733				  STATX_ATTR_NODUMP);
 734
 735	switch (inode->i_mode & S_IFMT) {
 736	case S_IFBLK:
 737	case S_IFCHR:
 738		stat->blksize = BLKDEV_IOSIZE;
 739		stat->rdev = inode->i_rdev;
 740		break;
 741	case S_IFREG:
 742		if (request_mask & (STATX_DIOALIGN | STATX_DIO_READ_ALIGN))
 743			xfs_report_dioalign(ip, stat);
 744		if (request_mask & STATX_WRITE_ATOMIC)
 745			xfs_report_atomic_write(ip, stat);
 746		fallthrough;
 747	default:
 748		stat->blksize = xfs_stat_blksize(ip);
 749		stat->rdev = 0;
 750		break;
 751	}
 752
 753	return 0;
 754}
 755
 756static int
 757xfs_vn_change_ok(
 758	struct mnt_idmap	*idmap,
 759	struct dentry		*dentry,
 760	struct iattr		*iattr)
 761{
 762	struct xfs_mount	*mp = XFS_I(d_inode(dentry))->i_mount;
 763
 764	if (xfs_is_readonly(mp))
 765		return -EROFS;
 766
 767	if (xfs_is_shutdown(mp))
 768		return -EIO;
 769
 770	return setattr_prepare(idmap, dentry, iattr);
 771}
 772
 773/*
 774 * Set non-size attributes of an inode.
 775 *
 776 * Caution: The caller of this function is responsible for calling
 777 * setattr_prepare() or otherwise verifying the change is fine.
 778 */
 779static int
 780xfs_setattr_nonsize(
 781	struct mnt_idmap	*idmap,
 782	struct dentry		*dentry,
 783	struct xfs_inode	*ip,
 784	struct iattr		*iattr)
 785{
 786	xfs_mount_t		*mp = ip->i_mount;
 787	struct inode		*inode = VFS_I(ip);
 788	int			mask = iattr->ia_valid;
 789	xfs_trans_t		*tp;
 790	int			error;
 791	kuid_t			uid = GLOBAL_ROOT_UID;
 792	kgid_t			gid = GLOBAL_ROOT_GID;
 793	struct xfs_dquot	*udqp = NULL, *gdqp = NULL;
 794	struct xfs_dquot	*old_udqp = NULL, *old_gdqp = NULL;
 795
 796	ASSERT((mask & ATTR_SIZE) == 0);
 797
 798	/*
 799	 * If disk quotas is on, we make sure that the dquots do exist on disk,
 800	 * before we start any other transactions. Trying to do this later
 801	 * is messy. We don't care to take a readlock to look at the ids
 802	 * in inode here, because we can't hold it across the trans_reserve.
 803	 * If the IDs do change before we take the ilock, we're covered
 804	 * because the i_*dquot fields will get updated anyway.
 805	 */
 806	if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) {
 807		uint	qflags = 0;
 808
 809		if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) {
 810			uid = from_vfsuid(idmap, i_user_ns(inode),
 811					  iattr->ia_vfsuid);
 812			qflags |= XFS_QMOPT_UQUOTA;
 813		} else {
 814			uid = inode->i_uid;
 815		}
 816		if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) {
 817			gid = from_vfsgid(idmap, i_user_ns(inode),
 818					  iattr->ia_vfsgid);
 819			qflags |= XFS_QMOPT_GQUOTA;
 820		}  else {
 821			gid = inode->i_gid;
 822		}
 823
 824		/*
 825		 * We take a reference when we initialize udqp and gdqp,
 826		 * so it is important that we never blindly double trip on
 827		 * the same variable. See xfs_create() for an example.
 828		 */
 829		ASSERT(udqp == NULL);
 830		ASSERT(gdqp == NULL);
 831		error = xfs_qm_vop_dqalloc(ip, uid, gid, ip->i_projid,
 832					   qflags, &udqp, &gdqp, NULL);
 833		if (error)
 834			return error;
 835	}
 836
 837	error = xfs_trans_alloc_ichange(ip, udqp, gdqp, NULL,
 838			has_capability_noaudit(current, CAP_FOWNER), &tp);
 839	if (error)
 840		goto out_dqrele;
 841
 842	/*
 843	 * Register quota modifications in the transaction.  Must be the owner
 844	 * or privileged.  These IDs could have changed since we last looked at
 845	 * them.  But, we're assured that if the ownership did change while we
 846	 * didn't have the inode locked, inode's dquot(s) would have changed
 847	 * also.
 848	 */
 849	if (XFS_IS_UQUOTA_ON(mp) &&
 850	    i_uid_needs_update(idmap, iattr, inode)) {
 851		ASSERT(udqp);
 852		old_udqp = xfs_qm_vop_chown(tp, ip, &ip->i_udquot, udqp);
 853	}
 854	if (XFS_IS_GQUOTA_ON(mp) &&
 855	    i_gid_needs_update(idmap, iattr, inode)) {
 856		ASSERT(xfs_has_pquotino(mp) || !XFS_IS_PQUOTA_ON(mp));
 857		ASSERT(gdqp);
 858		old_gdqp = xfs_qm_vop_chown(tp, ip, &ip->i_gdquot, gdqp);
 859	}
 860
 861	setattr_copy(idmap, inode, iattr);
 862	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
 863
 864	XFS_STATS_INC(mp, xs_ig_attrchg);
 865
 866	if (xfs_has_wsync(mp))
 867		xfs_trans_set_sync(tp);
 868	error = xfs_trans_commit(tp);
 869
 870	/*
 871	 * Release any dquot(s) the inode had kept before chown.
 872	 */
 873	xfs_qm_dqrele(old_udqp);
 874	xfs_qm_dqrele(old_gdqp);
 875	xfs_qm_dqrele(udqp);
 876	xfs_qm_dqrele(gdqp);
 877
 878	if (error)
 879		return error;
 880
 881	/*
 882	 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode
 883	 * 	     update.  We could avoid this with linked transactions
 884	 * 	     and passing down the transaction pointer all the way
 885	 *	     to attr_set.  No previous user of the generic
 886	 * 	     Posix ACL code seems to care about this issue either.
 887	 */
 888	if (mask & ATTR_MODE) {
 889		error = posix_acl_chmod(idmap, dentry, inode->i_mode);
 890		if (error)
 891			return error;
 892	}
 893
 894	return 0;
 895
 896out_dqrele:
 897	xfs_qm_dqrele(udqp);
 898	xfs_qm_dqrele(gdqp);
 899	return error;
 900}
 901
 902/*
 903 * Truncate file.  Must have write permission and not be a directory.
 904 *
 905 * Caution: The caller of this function is responsible for calling
 906 * setattr_prepare() or otherwise verifying the change is fine.
 907 */
 908STATIC int
 909xfs_setattr_size(
 910	struct mnt_idmap	*idmap,
 911	struct dentry		*dentry,
 912	struct xfs_inode	*ip,
 913	struct iattr		*iattr)
 914{
 915	struct xfs_mount	*mp = ip->i_mount;
 916	struct inode		*inode = VFS_I(ip);
 917	xfs_off_t		oldsize, newsize;
 918	struct xfs_trans	*tp;
 919	int			error;
 920	uint			lock_flags = 0;
 921	uint			resblks = 0;
 922	bool			did_zeroing = false;
 923	struct xfs_zone_alloc_ctx ac = { };
 924
 925	xfs_assert_ilocked(ip, XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL);
 926	ASSERT(S_ISREG(inode->i_mode));
 927	ASSERT((iattr->ia_valid & (ATTR_UID|ATTR_GID|ATTR_ATIME|ATTR_ATIME_SET|
 928		ATTR_MTIME_SET|ATTR_TIMES_SET)) == 0);
 929
 930	oldsize = inode->i_size;
 931	newsize = iattr->ia_size;
 932
 933	/*
 934	 * Short circuit the truncate case for zero length files.
 935	 */
 936	if (newsize == 0 && oldsize == 0 && ip->i_df.if_nextents == 0) {
 937		if (!(iattr->ia_valid & (ATTR_CTIME|ATTR_MTIME)))
 938			return 0;
 939
 940		/*
 941		 * Use the regular setattr path to update the timestamps.
 942		 */
 943		iattr->ia_valid &= ~ATTR_SIZE;
 944		return xfs_setattr_nonsize(idmap, dentry, ip, iattr);
 945	}
 946
 947	/*
 948	 * Make sure that the dquots are attached to the inode.
 949	 */
 950	error = xfs_qm_dqattach(ip);
 951	if (error)
 952		return error;
 953
 954	/*
 955	 * Wait for all direct I/O to complete.
 956	 */
 957	inode_dio_wait(inode);
 958
 959	/*
 960	 * Normally xfs_zoned_space_reserve is supposed to be called outside the
 961	 * IOLOCK.  For truncate we can't do that since ->setattr is called with
 962	 * it already held by the VFS.  So for now chicken out and try to
 963	 * allocate space under it.
 964	 *
 965	 * To avoid deadlocks this means we can't block waiting for space, which
 966	 * can lead to spurious -ENOSPC if there are no directly available
 967	 * blocks.  We mitigate this a bit by allowing zeroing to dip into the
 968	 * reserved pool, but eventually the VFS calling convention needs to
 969	 * change.
 970	 */
 971	if (xfs_is_zoned_inode(ip)) {
 972		error = xfs_zoned_space_reserve(mp, 1,
 973				XFS_ZR_NOWAIT | XFS_ZR_RESERVED, &ac);
 974		if (error) {
 975			if (error == -EAGAIN)
 976				return -ENOSPC;
 977			return error;
 978		}
 979	}
 980
 981	/*
 982	 * File data changes must be complete before we start the transaction to
 983	 * modify the inode.  This needs to be done before joining the inode to
 984	 * the transaction because the inode cannot be unlocked once it is a
 985	 * part of the transaction.
 986	 *
 987	 * Start with zeroing any data beyond EOF that we may expose on file
 988	 * extension, or zeroing out the rest of the block on a downward
 989	 * truncate.
 990	 */
 991	if (newsize > oldsize) {
 992		trace_xfs_zero_eof(ip, oldsize, newsize - oldsize);
 993		error = xfs_zero_range(ip, oldsize, newsize - oldsize,
 994				&ac, &did_zeroing);
 995	} else {
 996		error = xfs_truncate_page(ip, newsize, &ac, &did_zeroing);
 997	}
 998
 999	if (xfs_is_zoned_inode(ip))
1000		xfs_zoned_space_unreserve(mp, &ac);
1001
1002	if (error)
1003		return error;
1004
1005	/*
1006	 * We've already locked out new page faults, so now we can safely remove
1007	 * pages from the page cache knowing they won't get refaulted until we
1008	 * drop the XFS_MMAP_EXCL lock after the extent manipulations are
1009	 * complete. The truncate_setsize() call also cleans partial EOF page
1010	 * PTEs on extending truncates and hence ensures sub-page block size
1011	 * filesystems are correctly handled, too.
1012	 *
1013	 * We have to do all the page cache truncate work outside the
1014	 * transaction context as the "lock" order is page lock->log space
1015	 * reservation as defined by extent allocation in the writeback path.
1016	 * Hence a truncate can fail with ENOMEM from xfs_trans_alloc(), but
1017	 * having already truncated the in-memory version of the file (i.e. made
1018	 * user visible changes). There's not much we can do about this, except
1019	 * to hope that the caller sees ENOMEM and retries the truncate
1020	 * operation.
1021	 *
1022	 * And we update in-core i_size and truncate page cache beyond newsize
1023	 * before writeback the [i_disk_size, newsize] range, so we're
1024	 * guaranteed not to write stale data past the new EOF on truncate down.
1025	 */
1026	truncate_setsize(inode, newsize);
1027
1028	/*
1029	 * We are going to log the inode size change in this transaction so
1030	 * any previous writes that are beyond the on disk EOF and the new
1031	 * EOF that have not been written out need to be written here.  If we
1032	 * do not write the data out, we expose ourselves to the null files
1033	 * problem. Note that this includes any block zeroing we did above;
1034	 * otherwise those blocks may not be zeroed after a crash.
1035	 */
1036	if (did_zeroing ||
1037	    (newsize > ip->i_disk_size && oldsize != ip->i_disk_size)) {
1038		error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
1039						ip->i_disk_size, newsize - 1);
1040		if (error)
1041			return error;
1042	}
1043
1044	/*
1045	 * For realtime inode with more than one block rtextsize, we need the
1046	 * block reservation for bmap btree block allocations/splits that can
1047	 * happen since it could split the tail written extent and convert the
1048	 * right beyond EOF one to unwritten.
1049	 */
1050	if (xfs_inode_has_bigrtalloc(ip))
1051		resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
1052
1053	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, resblks,
1054				0, 0, &tp);
1055	if (error)
1056		return error;
1057
1058	lock_flags |= XFS_ILOCK_EXCL;
1059	xfs_ilock(ip, XFS_ILOCK_EXCL);
1060	xfs_trans_ijoin(tp, ip, 0);
1061
1062	/*
1063	 * Only change the c/mtime if we are changing the size or we are
1064	 * explicitly asked to change it.  This handles the semantic difference
1065	 * between truncate() and ftruncate() as implemented in the VFS.
1066	 *
1067	 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
1068	 * special case where we need to update the times despite not having
1069	 * these flags set.  For all other operations the VFS set these flags
1070	 * explicitly if it wants a timestamp update.
1071	 */
1072	if (newsize != oldsize &&
1073	    !(iattr->ia_valid & (ATTR_CTIME | ATTR_MTIME))) {
1074		iattr->ia_ctime = iattr->ia_mtime =
1075			current_time(inode);
1076		iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME;
1077	}
1078
1079	/*
1080	 * The first thing we do is set the size to new_size permanently on
1081	 * disk.  This way we don't have to worry about anyone ever being able
1082	 * to look at the data being freed even in the face of a crash.
1083	 * What we're getting around here is the case where we free a block, it
1084	 * is allocated to another file, it is written to, and then we crash.
1085	 * If the new data gets written to the file but the log buffers
1086	 * containing the free and reallocation don't, then we'd end up with
1087	 * garbage in the blocks being freed.  As long as we make the new size
1088	 * permanent before actually freeing any blocks it doesn't matter if
1089	 * they get written to.
1090	 */
1091	ip->i_disk_size = newsize;
1092	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1093
1094	if (newsize <= oldsize) {
1095		error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, newsize);
1096		if (error)
1097			goto out_trans_cancel;
1098
1099		/*
1100		 * Truncated "down", so we're removing references to old data
1101		 * here - if we delay flushing for a long time, we expose
1102		 * ourselves unduly to the notorious NULL files problem.  So,
1103		 * we mark this inode and flush it when the file is closed,
1104		 * and do not wait the usual (long) time for writeout.
1105		 */
1106		xfs_iflags_set(ip, XFS_ITRUNCATED);
1107
1108		/* A truncate down always removes post-EOF blocks. */
1109		xfs_inode_clear_eofblocks_tag(ip);
1110	}
1111
1112	ASSERT(!(iattr->ia_valid & (ATTR_UID | ATTR_GID)));
1113	setattr_copy(idmap, inode, iattr);
1114	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1115
1116	XFS_STATS_INC(mp, xs_ig_attrchg);
1117
1118	if (xfs_has_wsync(mp))
1119		xfs_trans_set_sync(tp);
1120
1121	error = xfs_trans_commit(tp);
1122out_unlock:
1123	if (lock_flags)
1124		xfs_iunlock(ip, lock_flags);
1125	return error;
1126
1127out_trans_cancel:
1128	xfs_trans_cancel(tp);
1129	goto out_unlock;
1130}
1131
1132int
1133xfs_vn_setattr_size(
1134	struct mnt_idmap	*idmap,
1135	struct dentry		*dentry,
1136	struct iattr		*iattr)
1137{
1138	struct xfs_inode	*ip = XFS_I(d_inode(dentry));
1139	int error;
1140
1141	trace_xfs_setattr(ip);
1142
1143	error = xfs_vn_change_ok(idmap, dentry, iattr);
1144	if (error)
1145		return error;
1146	return xfs_setattr_size(idmap, dentry, ip, iattr);
1147}
1148
1149STATIC int
1150xfs_vn_setattr(
1151	struct mnt_idmap	*idmap,
1152	struct dentry		*dentry,
1153	struct iattr		*iattr)
1154{
1155	struct inode		*inode = d_inode(dentry);
1156	struct xfs_inode	*ip = XFS_I(inode);
1157	int			error;
1158
1159	if (iattr->ia_valid & ATTR_SIZE) {
1160		uint			iolock;
1161
1162		xfs_ilock(ip, XFS_MMAPLOCK_EXCL);
1163		iolock = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL;
1164
1165		error = xfs_break_layouts(inode, &iolock, BREAK_UNMAP);
1166		if (error) {
1167			xfs_iunlock(ip, XFS_MMAPLOCK_EXCL);
1168			return error;
1169		}
1170
1171		error = xfs_vn_setattr_size(idmap, dentry, iattr);
1172		xfs_iunlock(ip, XFS_MMAPLOCK_EXCL);
1173	} else {
1174		trace_xfs_setattr(ip);
1175
1176		error = xfs_vn_change_ok(idmap, dentry, iattr);
1177		if (!error)
1178			error = xfs_setattr_nonsize(idmap, dentry, ip, iattr);
1179	}
1180
1181	return error;
1182}
1183
1184STATIC int
1185xfs_vn_update_time(
1186	struct inode		*inode,
1187	int			flags)
1188{
1189	struct xfs_inode	*ip = XFS_I(inode);
1190	struct xfs_mount	*mp = ip->i_mount;
1191	int			log_flags = XFS_ILOG_TIMESTAMP;
1192	struct xfs_trans	*tp;
1193	int			error;
1194	struct timespec64	now;
1195
1196	trace_xfs_update_time(ip);
1197
1198	if (inode->i_sb->s_flags & SB_LAZYTIME) {
1199		if (!((flags & S_VERSION) &&
1200		      inode_maybe_inc_iversion(inode, false))) {
1201			generic_update_time(inode, flags);
1202			return 0;
1203		}
1204
1205		/* Capture the iversion update that just occurred */
1206		log_flags |= XFS_ILOG_CORE;
1207	}
1208
1209	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp);
1210	if (error)
1211		return error;
1212
1213	xfs_ilock(ip, XFS_ILOCK_EXCL);
1214	if (flags & (S_CTIME|S_MTIME))
1215		now = inode_set_ctime_current(inode);
1216	else
1217		now = current_time(inode);
1218
1219	if (flags & S_MTIME)
1220		inode_set_mtime_to_ts(inode, now);
1221	if (flags & S_ATIME)
1222		inode_set_atime_to_ts(inode, now);
1223
1224	xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1225	xfs_trans_log_inode(tp, ip, log_flags);
1226	return xfs_trans_commit(tp);
1227}
1228
1229STATIC int
1230xfs_vn_fiemap(
1231	struct inode		*inode,
1232	struct fiemap_extent_info *fieinfo,
1233	u64			start,
1234	u64			length)
1235{
1236	int			error;
1237
1238	xfs_ilock(XFS_I(inode), XFS_IOLOCK_SHARED);
1239	if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
1240		fieinfo->fi_flags &= ~FIEMAP_FLAG_XATTR;
1241		error = iomap_fiemap(inode, fieinfo, start, length,
1242				&xfs_xattr_iomap_ops);
1243	} else {
1244		error = iomap_fiemap(inode, fieinfo, start, length,
1245				&xfs_read_iomap_ops);
1246	}
1247	xfs_iunlock(XFS_I(inode), XFS_IOLOCK_SHARED);
1248
1249	return error;
1250}
1251
1252STATIC int
1253xfs_vn_tmpfile(
1254	struct mnt_idmap	*idmap,
1255	struct inode		*dir,
1256	struct file		*file,
1257	umode_t			mode)
1258{
1259	int err = xfs_generic_create(idmap, dir, file->f_path.dentry, mode, 0, file);
1260
1261	return finish_open_simple(file, err);
1262}
1263
1264static const struct inode_operations xfs_inode_operations = {
1265	.get_inode_acl		= xfs_get_acl,
1266	.set_acl		= xfs_set_acl,
1267	.getattr		= xfs_vn_getattr,
1268	.setattr		= xfs_vn_setattr,
1269	.listxattr		= xfs_vn_listxattr,
1270	.fiemap			= xfs_vn_fiemap,
1271	.update_time		= xfs_vn_update_time,
1272	.fileattr_get		= xfs_fileattr_get,
1273	.fileattr_set		= xfs_fileattr_set,
1274};
1275
1276static const struct inode_operations xfs_dir_inode_operations = {
1277	.create			= xfs_vn_create,
1278	.lookup			= xfs_vn_lookup,
1279	.link			= xfs_vn_link,
1280	.unlink			= xfs_vn_unlink,
1281	.symlink		= xfs_vn_symlink,
1282	.mkdir			= xfs_vn_mkdir,
1283	/*
1284	 * Yes, XFS uses the same method for rmdir and unlink.
1285	 *
1286	 * There are some subtile differences deeper in the code,
1287	 * but we use S_ISDIR to check for those.
1288	 */
1289	.rmdir			= xfs_vn_unlink,
1290	.mknod			= xfs_vn_mknod,
1291	.rename			= xfs_vn_rename,
1292	.get_inode_acl		= xfs_get_acl,
1293	.set_acl		= xfs_set_acl,
1294	.getattr		= xfs_vn_getattr,
1295	.setattr		= xfs_vn_setattr,
1296	.listxattr		= xfs_vn_listxattr,
1297	.update_time		= xfs_vn_update_time,
1298	.tmpfile		= xfs_vn_tmpfile,
1299	.fileattr_get		= xfs_fileattr_get,
1300	.fileattr_set		= xfs_fileattr_set,
1301};
1302
1303static const struct inode_operations xfs_dir_ci_inode_operations = {
1304	.create			= xfs_vn_create,
1305	.lookup			= xfs_vn_ci_lookup,
1306	.link			= xfs_vn_link,
1307	.unlink			= xfs_vn_unlink,
1308	.symlink		= xfs_vn_symlink,
1309	.mkdir			= xfs_vn_mkdir,
1310	/*
1311	 * Yes, XFS uses the same method for rmdir and unlink.
1312	 *
1313	 * There are some subtile differences deeper in the code,
1314	 * but we use S_ISDIR to check for those.
1315	 */
1316	.rmdir			= xfs_vn_unlink,
1317	.mknod			= xfs_vn_mknod,
1318	.rename			= xfs_vn_rename,
1319	.get_inode_acl		= xfs_get_acl,
1320	.set_acl		= xfs_set_acl,
1321	.getattr		= xfs_vn_getattr,
1322	.setattr		= xfs_vn_setattr,
1323	.listxattr		= xfs_vn_listxattr,
1324	.update_time		= xfs_vn_update_time,
1325	.tmpfile		= xfs_vn_tmpfile,
1326	.fileattr_get		= xfs_fileattr_get,
1327	.fileattr_set		= xfs_fileattr_set,
1328};
1329
1330static const struct inode_operations xfs_symlink_inode_operations = {
1331	.get_link		= xfs_vn_get_link,
1332	.getattr		= xfs_vn_getattr,
1333	.setattr		= xfs_vn_setattr,
1334	.listxattr		= xfs_vn_listxattr,
1335	.update_time		= xfs_vn_update_time,
1336	.fileattr_get		= xfs_fileattr_get,
1337	.fileattr_set		= xfs_fileattr_set,
1338};
1339
1340/* Figure out if this file actually supports DAX. */
1341static bool
1342xfs_inode_supports_dax(
1343	struct xfs_inode	*ip)
1344{
1345	struct xfs_mount	*mp = ip->i_mount;
1346
1347	/* Only supported on regular files. */
1348	if (!S_ISREG(VFS_I(ip)->i_mode))
1349		return false;
1350
1351	/* Block size must match page size */
1352	if (mp->m_sb.sb_blocksize != PAGE_SIZE)
1353		return false;
1354
1355	/* Device has to support DAX too. */
1356	return xfs_inode_buftarg(ip)->bt_daxdev != NULL;
1357}
1358
1359static bool
1360xfs_inode_should_enable_dax(
1361	struct xfs_inode *ip)
1362{
1363	if (!IS_ENABLED(CONFIG_FS_DAX))
1364		return false;
1365	if (xfs_has_dax_never(ip->i_mount))
1366		return false;
1367	if (!xfs_inode_supports_dax(ip))
1368		return false;
1369	if (xfs_has_dax_always(ip->i_mount))
1370		return true;
1371	if (ip->i_diflags2 & XFS_DIFLAG2_DAX)
1372		return true;
1373	return false;
1374}
1375
1376void
1377xfs_diflags_to_iflags(
1378	struct xfs_inode	*ip,
1379	bool init)
1380{
1381	struct inode            *inode = VFS_I(ip);
1382	unsigned int            xflags = xfs_ip2xflags(ip);
1383	unsigned int            flags = 0;
1384
1385	ASSERT(!(IS_DAX(inode) && init));
1386
1387	if (xflags & FS_XFLAG_IMMUTABLE)
1388		flags |= S_IMMUTABLE;
1389	if (xflags & FS_XFLAG_APPEND)
1390		flags |= S_APPEND;
1391	if (xflags & FS_XFLAG_SYNC)
1392		flags |= S_SYNC;
1393	if (xflags & FS_XFLAG_NOATIME)
1394		flags |= S_NOATIME;
1395	if (init && xfs_inode_should_enable_dax(ip))
1396		flags |= S_DAX;
1397
1398	/*
1399	 * S_DAX can only be set during inode initialization and is never set by
1400	 * the VFS, so we cannot mask off S_DAX in i_flags.
1401	 */
1402	inode->i_flags &= ~(S_IMMUTABLE | S_APPEND | S_SYNC | S_NOATIME);
1403	inode->i_flags |= flags;
1404}
1405
1406/*
1407 * Initialize the Linux inode.
1408 *
1409 * When reading existing inodes from disk this is called directly from xfs_iget,
1410 * when creating a new inode it is called from xfs_init_new_inode after setting
1411 * up the inode. These callers have different criteria for clearing XFS_INEW, so
1412 * leave it up to the caller to deal with unlocking the inode appropriately.
1413 */
1414void
1415xfs_setup_inode(
1416	struct xfs_inode	*ip)
1417{
1418	struct inode		*inode = &ip->i_vnode;
1419	gfp_t			gfp_mask;
1420	bool			is_meta = xfs_is_internal_inode(ip);
1421
1422	inode->i_ino = ip->i_ino;
1423	inode_state_set_raw(inode, I_NEW);
1424
1425	inode_sb_list_add(inode);
1426	/* make the inode look hashed for the writeback code */
1427	inode_fake_hash(inode);
1428
1429	i_size_write(inode, ip->i_disk_size);
1430	xfs_diflags_to_iflags(ip, true);
1431
1432	/*
1433	 * Mark our metadata files as private so that LSMs and the ACL code
1434	 * don't try to add their own metadata or reason about these files,
1435	 * and users cannot ever obtain file handles to them.
1436	 */
1437	if (is_meta) {
1438		inode->i_flags |= S_PRIVATE;
1439		inode->i_opflags &= ~IOP_XATTR;
1440	}
1441
1442	if (S_ISDIR(inode->i_mode)) {
1443		/*
1444		 * We set the i_rwsem class here to avoid potential races with
1445		 * lockdep_annotate_inode_mutex_key() reinitialising the lock
1446		 * after a filehandle lookup has already found the inode in
1447		 * cache before it has been unlocked via unlock_new_inode().
1448		 */
1449		lockdep_set_class(&inode->i_rwsem,
1450				  &inode->i_sb->s_type->i_mutex_dir_key);
1451		lockdep_set_class(&ip->i_lock, &xfs_dir_ilock_class);
1452	} else {
1453		lockdep_set_class(&ip->i_lock, &xfs_nondir_ilock_class);
1454	}
1455
1456	/*
1457	 * Ensure all page cache allocations are done from GFP_NOFS context to
1458	 * prevent direct reclaim recursion back into the filesystem and blowing
1459	 * stacks or deadlocking.
1460	 */
1461	gfp_mask = mapping_gfp_mask(inode->i_mapping);
1462	mapping_set_gfp_mask(inode->i_mapping, (gfp_mask & ~(__GFP_FS)));
1463
1464	/*
1465	 * For real-time inodes update the stable write flags to that of the RT
1466	 * device instead of the data device.
1467	 */
1468	if (S_ISREG(inode->i_mode) && XFS_IS_REALTIME_INODE(ip))
1469		xfs_update_stable_writes(ip);
1470
1471	/*
1472	 * If there is no attribute fork no ACL can exist on this inode,
1473	 * and it can't have any file capabilities attached to it either.
1474	 */
1475	if (!xfs_inode_has_attr_fork(ip)) {
1476		inode_has_no_xattr(inode);
1477		cache_no_acl(inode);
1478	}
1479}
1480
1481void
1482xfs_setup_iops(
1483	struct xfs_inode	*ip)
1484{
1485	struct inode		*inode = &ip->i_vnode;
1486
1487	switch (inode->i_mode & S_IFMT) {
1488	case S_IFREG:
1489		inode->i_op = &xfs_inode_operations;
1490		inode->i_fop = &xfs_file_operations;
1491		if (IS_DAX(inode))
1492			inode->i_mapping->a_ops = &xfs_dax_aops;
1493		else
1494			inode->i_mapping->a_ops = &xfs_address_space_operations;
1495		break;
1496	case S_IFDIR:
1497		if (xfs_has_asciici(XFS_M(inode->i_sb)))
1498			inode->i_op = &xfs_dir_ci_inode_operations;
1499		else
1500			inode->i_op = &xfs_dir_inode_operations;
1501		inode->i_fop = &xfs_dir_file_operations;
1502		break;
1503	case S_IFLNK:
1504		inode->i_op = &xfs_symlink_inode_operations;
1505		break;
1506	default:
1507		inode->i_op = &xfs_inode_operations;
1508		init_special_inode(inode, inode->i_mode, inode->i_rdev);
1509		break;
1510	}
1511}