Linux Audio

Check our new training course

Loading...
v4.6
 
   1/**
   2 * inode.c - NTFS kernel inode handling.
   3 *
   4 * Copyright (c) 2001-2014 Anton Altaparmakov and Tuxera Inc.
   5 *
   6 * This program/include file is free software; you can redistribute it and/or
   7 * modify it under the terms of the GNU General Public License as published
   8 * by the Free Software Foundation; either version 2 of the License, or
   9 * (at your option) any later version.
  10 *
  11 * This program/include file is distributed in the hope that it will be
  12 * useful, but WITHOUT ANY WARRANTY; without even the implied warranty
  13 * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  14 * GNU General Public License for more details.
  15 *
  16 * You should have received a copy of the GNU General Public License
  17 * along with this program (in the main directory of the Linux-NTFS
  18 * distribution in the file COPYING); if not, write to the Free Software
  19 * Foundation,Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
  20 */
  21
  22#include <linux/buffer_head.h>
  23#include <linux/fs.h>
  24#include <linux/mm.h>
  25#include <linux/mount.h>
  26#include <linux/mutex.h>
  27#include <linux/pagemap.h>
  28#include <linux/quotaops.h>
  29#include <linux/slab.h>
  30#include <linux/log2.h>
  31
  32#include "aops.h"
  33#include "attrib.h"
  34#include "bitmap.h"
  35#include "dir.h"
  36#include "debug.h"
  37#include "inode.h"
  38#include "lcnalloc.h"
  39#include "malloc.h"
  40#include "mft.h"
  41#include "time.h"
  42#include "ntfs.h"
  43
  44/**
  45 * ntfs_test_inode - compare two (possibly fake) inodes for equality
  46 * @vi:		vfs inode which to test
  47 * @na:		ntfs attribute which is being tested with
  48 *
  49 * Compare the ntfs attribute embedded in the ntfs specific part of the vfs
  50 * inode @vi for equality with the ntfs attribute @na.
  51 *
  52 * If searching for the normal file/directory inode, set @na->type to AT_UNUSED.
  53 * @na->name and @na->name_len are then ignored.
  54 *
  55 * Return 1 if the attributes match and 0 if not.
  56 *
  57 * NOTE: This function runs with the inode_hash_lock spin lock held so it is not
  58 * allowed to sleep.
  59 */
  60int ntfs_test_inode(struct inode *vi, ntfs_attr *na)
  61{
 
  62	ntfs_inode *ni;
  63
  64	if (vi->i_ino != na->mft_no)
  65		return 0;
  66	ni = NTFS_I(vi);
  67	/* If !NInoAttr(ni), @vi is a normal file or directory inode. */
  68	if (likely(!NInoAttr(ni))) {
  69		/* If not looking for a normal inode this is a mismatch. */
  70		if (unlikely(na->type != AT_UNUSED))
  71			return 0;
  72	} else {
  73		/* A fake inode describing an attribute. */
  74		if (ni->type != na->type)
  75			return 0;
  76		if (ni->name_len != na->name_len)
  77			return 0;
  78		if (na->name_len && memcmp(ni->name, na->name,
  79				na->name_len * sizeof(ntfschar)))
  80			return 0;
  81	}
  82	/* Match! */
  83	return 1;
  84}
  85
  86/**
  87 * ntfs_init_locked_inode - initialize an inode
  88 * @vi:		vfs inode to initialize
  89 * @na:		ntfs attribute which to initialize @vi to
  90 *
  91 * Initialize the vfs inode @vi with the values from the ntfs attribute @na in
  92 * order to enable ntfs_test_inode() to do its work.
  93 *
  94 * If initializing the normal file/directory inode, set @na->type to AT_UNUSED.
  95 * In that case, @na->name and @na->name_len should be set to NULL and 0,
  96 * respectively. Although that is not strictly necessary as
  97 * ntfs_read_locked_inode() will fill them in later.
  98 *
  99 * Return 0 on success and -errno on error.
 100 *
 101 * NOTE: This function runs with the inode->i_lock spin lock held so it is not
 102 * allowed to sleep. (Hence the GFP_ATOMIC allocation.)
 103 */
 104static int ntfs_init_locked_inode(struct inode *vi, ntfs_attr *na)
 105{
 
 106	ntfs_inode *ni = NTFS_I(vi);
 107
 108	vi->i_ino = na->mft_no;
 109
 110	ni->type = na->type;
 111	if (na->type == AT_INDEX_ALLOCATION)
 112		NInoSetMstProtected(ni);
 113
 114	ni->name = na->name;
 115	ni->name_len = na->name_len;
 116
 117	/* If initializing a normal inode, we are done. */
 118	if (likely(na->type == AT_UNUSED)) {
 119		BUG_ON(na->name);
 120		BUG_ON(na->name_len);
 121		return 0;
 122	}
 123
 124	/* It is a fake inode. */
 125	NInoSetAttr(ni);
 126
 127	/*
 128	 * We have I30 global constant as an optimization as it is the name
 129	 * in >99.9% of named attributes! The other <0.1% incur a GFP_ATOMIC
 130	 * allocation but that is ok. And most attributes are unnamed anyway,
 131	 * thus the fraction of named attributes with name != I30 is actually
 132	 * absolutely tiny.
 133	 */
 134	if (na->name_len && na->name != I30) {
 135		unsigned int i;
 136
 137		BUG_ON(!na->name);
 138		i = na->name_len * sizeof(ntfschar);
 139		ni->name = kmalloc(i + sizeof(ntfschar), GFP_ATOMIC);
 140		if (!ni->name)
 141			return -ENOMEM;
 142		memcpy(ni->name, na->name, i);
 143		ni->name[na->name_len] = 0;
 144	}
 145	return 0;
 146}
 147
 148typedef int (*set_t)(struct inode *, void *);
 149static int ntfs_read_locked_inode(struct inode *vi);
 150static int ntfs_read_locked_attr_inode(struct inode *base_vi, struct inode *vi);
 151static int ntfs_read_locked_index_inode(struct inode *base_vi,
 152		struct inode *vi);
 153
 154/**
 155 * ntfs_iget - obtain a struct inode corresponding to a specific normal inode
 156 * @sb:		super block of mounted volume
 157 * @mft_no:	mft record number / inode number to obtain
 158 *
 159 * Obtain the struct inode corresponding to a specific normal inode (i.e. a
 160 * file or directory).
 161 *
 162 * If the inode is in the cache, it is just returned with an increased
 163 * reference count. Otherwise, a new struct inode is allocated and initialized,
 164 * and finally ntfs_read_locked_inode() is called to read in the inode and
 165 * fill in the remainder of the inode structure.
 166 *
 167 * Return the struct inode on success. Check the return value with IS_ERR() and
 168 * if true, the function failed and the error code is obtained from PTR_ERR().
 169 */
 170struct inode *ntfs_iget(struct super_block *sb, unsigned long mft_no)
 171{
 172	struct inode *vi;
 173	int err;
 174	ntfs_attr na;
 175
 176	na.mft_no = mft_no;
 177	na.type = AT_UNUSED;
 178	na.name = NULL;
 179	na.name_len = 0;
 180
 181	vi = iget5_locked(sb, mft_no, (test_t)ntfs_test_inode,
 182			(set_t)ntfs_init_locked_inode, &na);
 183	if (unlikely(!vi))
 184		return ERR_PTR(-ENOMEM);
 185
 186	err = 0;
 187
 188	/* If this is a freshly allocated inode, need to read it now. */
 189	if (vi->i_state & I_NEW) {
 190		err = ntfs_read_locked_inode(vi);
 191		unlock_new_inode(vi);
 192	}
 193	/*
 194	 * There is no point in keeping bad inodes around if the failure was
 195	 * due to ENOMEM. We want to be able to retry again later.
 196	 */
 197	if (unlikely(err == -ENOMEM)) {
 198		iput(vi);
 199		vi = ERR_PTR(err);
 200	}
 201	return vi;
 202}
 203
 204/**
 205 * ntfs_attr_iget - obtain a struct inode corresponding to an attribute
 206 * @base_vi:	vfs base inode containing the attribute
 207 * @type:	attribute type
 208 * @name:	Unicode name of the attribute (NULL if unnamed)
 209 * @name_len:	length of @name in Unicode characters (0 if unnamed)
 210 *
 211 * Obtain the (fake) struct inode corresponding to the attribute specified by
 212 * @type, @name, and @name_len, which is present in the base mft record
 213 * specified by the vfs inode @base_vi.
 214 *
 215 * If the attribute inode is in the cache, it is just returned with an
 216 * increased reference count. Otherwise, a new struct inode is allocated and
 217 * initialized, and finally ntfs_read_locked_attr_inode() is called to read the
 218 * attribute and fill in the inode structure.
 219 *
 220 * Note, for index allocation attributes, you need to use ntfs_index_iget()
 221 * instead of ntfs_attr_iget() as working with indices is a lot more complex.
 222 *
 223 * Return the struct inode of the attribute inode on success. Check the return
 224 * value with IS_ERR() and if true, the function failed and the error code is
 225 * obtained from PTR_ERR().
 226 */
 227struct inode *ntfs_attr_iget(struct inode *base_vi, ATTR_TYPE type,
 228		ntfschar *name, u32 name_len)
 229{
 230	struct inode *vi;
 231	int err;
 232	ntfs_attr na;
 233
 234	/* Make sure no one calls ntfs_attr_iget() for indices. */
 235	BUG_ON(type == AT_INDEX_ALLOCATION);
 236
 237	na.mft_no = base_vi->i_ino;
 238	na.type = type;
 239	na.name = name;
 240	na.name_len = name_len;
 241
 242	vi = iget5_locked(base_vi->i_sb, na.mft_no, (test_t)ntfs_test_inode,
 243			(set_t)ntfs_init_locked_inode, &na);
 244	if (unlikely(!vi))
 245		return ERR_PTR(-ENOMEM);
 246
 247	err = 0;
 248
 249	/* If this is a freshly allocated inode, need to read it now. */
 250	if (vi->i_state & I_NEW) {
 251		err = ntfs_read_locked_attr_inode(base_vi, vi);
 252		unlock_new_inode(vi);
 253	}
 254	/*
 255	 * There is no point in keeping bad attribute inodes around. This also
 256	 * simplifies things in that we never need to check for bad attribute
 257	 * inodes elsewhere.
 258	 */
 259	if (unlikely(err)) {
 260		iput(vi);
 261		vi = ERR_PTR(err);
 262	}
 263	return vi;
 264}
 265
 266/**
 267 * ntfs_index_iget - obtain a struct inode corresponding to an index
 268 * @base_vi:	vfs base inode containing the index related attributes
 269 * @name:	Unicode name of the index
 270 * @name_len:	length of @name in Unicode characters
 271 *
 272 * Obtain the (fake) struct inode corresponding to the index specified by @name
 273 * and @name_len, which is present in the base mft record specified by the vfs
 274 * inode @base_vi.
 275 *
 276 * If the index inode is in the cache, it is just returned with an increased
 277 * reference count.  Otherwise, a new struct inode is allocated and
 278 * initialized, and finally ntfs_read_locked_index_inode() is called to read
 279 * the index related attributes and fill in the inode structure.
 280 *
 281 * Return the struct inode of the index inode on success. Check the return
 282 * value with IS_ERR() and if true, the function failed and the error code is
 283 * obtained from PTR_ERR().
 284 */
 285struct inode *ntfs_index_iget(struct inode *base_vi, ntfschar *name,
 286		u32 name_len)
 287{
 288	struct inode *vi;
 289	int err;
 290	ntfs_attr na;
 291
 292	na.mft_no = base_vi->i_ino;
 293	na.type = AT_INDEX_ALLOCATION;
 294	na.name = name;
 295	na.name_len = name_len;
 296
 297	vi = iget5_locked(base_vi->i_sb, na.mft_no, (test_t)ntfs_test_inode,
 298			(set_t)ntfs_init_locked_inode, &na);
 299	if (unlikely(!vi))
 300		return ERR_PTR(-ENOMEM);
 301
 302	err = 0;
 303
 304	/* If this is a freshly allocated inode, need to read it now. */
 305	if (vi->i_state & I_NEW) {
 306		err = ntfs_read_locked_index_inode(base_vi, vi);
 307		unlock_new_inode(vi);
 308	}
 309	/*
 310	 * There is no point in keeping bad index inodes around.  This also
 311	 * simplifies things in that we never need to check for bad index
 312	 * inodes elsewhere.
 313	 */
 314	if (unlikely(err)) {
 315		iput(vi);
 316		vi = ERR_PTR(err);
 317	}
 318	return vi;
 319}
 320
 321struct inode *ntfs_alloc_big_inode(struct super_block *sb)
 322{
 323	ntfs_inode *ni;
 324
 325	ntfs_debug("Entering.");
 326	ni = kmem_cache_alloc(ntfs_big_inode_cache, GFP_NOFS);
 327	if (likely(ni != NULL)) {
 328		ni->state = 0;
 329		return VFS_I(ni);
 330	}
 331	ntfs_error(sb, "Allocation of NTFS big inode structure failed.");
 332	return NULL;
 333}
 334
 335static void ntfs_i_callback(struct rcu_head *head)
 336{
 337	struct inode *inode = container_of(head, struct inode, i_rcu);
 338	kmem_cache_free(ntfs_big_inode_cache, NTFS_I(inode));
 339}
 340
 341void ntfs_destroy_big_inode(struct inode *inode)
 342{
 343	ntfs_inode *ni = NTFS_I(inode);
 344
 345	ntfs_debug("Entering.");
 346	BUG_ON(ni->page);
 347	if (!atomic_dec_and_test(&ni->count))
 348		BUG();
 349	call_rcu(&inode->i_rcu, ntfs_i_callback);
 350}
 351
 352static inline ntfs_inode *ntfs_alloc_extent_inode(void)
 353{
 354	ntfs_inode *ni;
 355
 356	ntfs_debug("Entering.");
 357	ni = kmem_cache_alloc(ntfs_inode_cache, GFP_NOFS);
 358	if (likely(ni != NULL)) {
 359		ni->state = 0;
 360		return ni;
 361	}
 362	ntfs_error(NULL, "Allocation of NTFS inode structure failed.");
 363	return NULL;
 364}
 365
 366static void ntfs_destroy_extent_inode(ntfs_inode *ni)
 367{
 368	ntfs_debug("Entering.");
 369	BUG_ON(ni->page);
 370	if (!atomic_dec_and_test(&ni->count))
 371		BUG();
 372	kmem_cache_free(ntfs_inode_cache, ni);
 373}
 374
 375/*
 376 * The attribute runlist lock has separate locking rules from the
 377 * normal runlist lock, so split the two lock-classes:
 378 */
 379static struct lock_class_key attr_list_rl_lock_class;
 380
 381/**
 382 * __ntfs_init_inode - initialize ntfs specific part of an inode
 383 * @sb:		super block of mounted volume
 384 * @ni:		freshly allocated ntfs inode which to initialize
 385 *
 386 * Initialize an ntfs inode to defaults.
 387 *
 388 * NOTE: ni->mft_no, ni->state, ni->type, ni->name, and ni->name_len are left
 389 * untouched. Make sure to initialize them elsewhere.
 390 *
 391 * Return zero on success and -ENOMEM on error.
 392 */
 393void __ntfs_init_inode(struct super_block *sb, ntfs_inode *ni)
 394{
 395	ntfs_debug("Entering.");
 396	rwlock_init(&ni->size_lock);
 397	ni->initialized_size = ni->allocated_size = 0;
 398	ni->seq_no = 0;
 399	atomic_set(&ni->count, 1);
 400	ni->vol = NTFS_SB(sb);
 401	ntfs_init_runlist(&ni->runlist);
 402	mutex_init(&ni->mrec_lock);
 403	ni->page = NULL;
 404	ni->page_ofs = 0;
 405	ni->attr_list_size = 0;
 406	ni->attr_list = NULL;
 407	ntfs_init_runlist(&ni->attr_list_rl);
 408	lockdep_set_class(&ni->attr_list_rl.lock,
 409				&attr_list_rl_lock_class);
 410	ni->itype.index.block_size = 0;
 411	ni->itype.index.vcn_size = 0;
 412	ni->itype.index.collation_rule = 0;
 413	ni->itype.index.block_size_bits = 0;
 414	ni->itype.index.vcn_size_bits = 0;
 415	mutex_init(&ni->extent_lock);
 416	ni->nr_extents = 0;
 417	ni->ext.base_ntfs_ino = NULL;
 418}
 419
 420/*
 421 * Extent inodes get MFT-mapped in a nested way, while the base inode
 422 * is still mapped. Teach this nesting to the lock validator by creating
 423 * a separate class for nested inode's mrec_lock's:
 424 */
 425static struct lock_class_key extent_inode_mrec_lock_key;
 426
 427inline ntfs_inode *ntfs_new_extent_inode(struct super_block *sb,
 428		unsigned long mft_no)
 429{
 430	ntfs_inode *ni = ntfs_alloc_extent_inode();
 431
 432	ntfs_debug("Entering.");
 433	if (likely(ni != NULL)) {
 434		__ntfs_init_inode(sb, ni);
 435		lockdep_set_class(&ni->mrec_lock, &extent_inode_mrec_lock_key);
 436		ni->mft_no = mft_no;
 437		ni->type = AT_UNUSED;
 438		ni->name = NULL;
 439		ni->name_len = 0;
 440	}
 441	return ni;
 442}
 443
 444/**
 445 * ntfs_is_extended_system_file - check if a file is in the $Extend directory
 446 * @ctx:	initialized attribute search context
 447 *
 448 * Search all file name attributes in the inode described by the attribute
 449 * search context @ctx and check if any of the names are in the $Extend system
 450 * directory.
 451 *
 452 * Return values:
 453 *	   1: file is in $Extend directory
 454 *	   0: file is not in $Extend directory
 455 *    -errno: failed to determine if the file is in the $Extend directory
 456 */
 457static int ntfs_is_extended_system_file(ntfs_attr_search_ctx *ctx)
 458{
 459	int nr_links, err;
 460
 461	/* Restart search. */
 462	ntfs_attr_reinit_search_ctx(ctx);
 463
 464	/* Get number of hard links. */
 465	nr_links = le16_to_cpu(ctx->mrec->link_count);
 466
 467	/* Loop through all hard links. */
 468	while (!(err = ntfs_attr_lookup(AT_FILE_NAME, NULL, 0, 0, 0, NULL, 0,
 469			ctx))) {
 470		FILE_NAME_ATTR *file_name_attr;
 471		ATTR_RECORD *attr = ctx->attr;
 472		u8 *p, *p2;
 473
 474		nr_links--;
 475		/*
 476		 * Maximum sanity checking as we are called on an inode that
 477		 * we suspect might be corrupt.
 478		 */
 479		p = (u8*)attr + le32_to_cpu(attr->length);
 480		if (p < (u8*)ctx->mrec || (u8*)p > (u8*)ctx->mrec +
 481				le32_to_cpu(ctx->mrec->bytes_in_use)) {
 482err_corrupt_attr:
 483			ntfs_error(ctx->ntfs_ino->vol->sb, "Corrupt file name "
 484					"attribute. You should run chkdsk.");
 485			return -EIO;
 486		}
 487		if (attr->non_resident) {
 488			ntfs_error(ctx->ntfs_ino->vol->sb, "Non-resident file "
 489					"name. You should run chkdsk.");
 490			return -EIO;
 491		}
 492		if (attr->flags) {
 493			ntfs_error(ctx->ntfs_ino->vol->sb, "File name with "
 494					"invalid flags. You should run "
 495					"chkdsk.");
 496			return -EIO;
 497		}
 498		if (!(attr->data.resident.flags & RESIDENT_ATTR_IS_INDEXED)) {
 499			ntfs_error(ctx->ntfs_ino->vol->sb, "Unindexed file "
 500					"name. You should run chkdsk.");
 501			return -EIO;
 502		}
 503		file_name_attr = (FILE_NAME_ATTR*)((u8*)attr +
 504				le16_to_cpu(attr->data.resident.value_offset));
 505		p2 = (u8*)attr + le32_to_cpu(attr->data.resident.value_length);
 506		if (p2 < (u8*)attr || p2 > p)
 507			goto err_corrupt_attr;
 508		/* This attribute is ok, but is it in the $Extend directory? */
 509		if (MREF_LE(file_name_attr->parent_directory) == FILE_Extend)
 510			return 1;	/* YES, it's an extended system file. */
 511	}
 512	if (unlikely(err != -ENOENT))
 513		return err;
 514	if (unlikely(nr_links)) {
 515		ntfs_error(ctx->ntfs_ino->vol->sb, "Inode hard link count "
 516				"doesn't match number of name attributes. You "
 517				"should run chkdsk.");
 518		return -EIO;
 519	}
 520	return 0;	/* NO, it is not an extended system file. */
 521}
 522
 523/**
 524 * ntfs_read_locked_inode - read an inode from its device
 525 * @vi:		inode to read
 526 *
 527 * ntfs_read_locked_inode() is called from ntfs_iget() to read the inode
 528 * described by @vi into memory from the device.
 529 *
 530 * The only fields in @vi that we need to/can look at when the function is
 531 * called are i_sb, pointing to the mounted device's super block, and i_ino,
 532 * the number of the inode to load.
 533 *
 534 * ntfs_read_locked_inode() maps, pins and locks the mft record number i_ino
 535 * for reading and sets up the necessary @vi fields as well as initializing
 536 * the ntfs inode.
 537 *
 538 * Q: What locks are held when the function is called?
 539 * A: i_state has I_NEW set, hence the inode is locked, also
 540 *    i_count is set to 1, so it is not going to go away
 541 *    i_flags is set to 0 and we have no business touching it.  Only an ioctl()
 542 *    is allowed to write to them. We should of course be honouring them but
 543 *    we need to do that using the IS_* macros defined in include/linux/fs.h.
 544 *    In any case ntfs_read_locked_inode() has nothing to do with i_flags.
 545 *
 546 * Return 0 on success and -errno on error.  In the error case, the inode will
 547 * have had make_bad_inode() executed on it.
 548 */
 549static int ntfs_read_locked_inode(struct inode *vi)
 550{
 551	ntfs_volume *vol = NTFS_SB(vi->i_sb);
 552	ntfs_inode *ni;
 553	struct inode *bvi;
 554	MFT_RECORD *m;
 555	ATTR_RECORD *a;
 556	STANDARD_INFORMATION *si;
 557	ntfs_attr_search_ctx *ctx;
 558	int err = 0;
 559
 560	ntfs_debug("Entering for i_ino 0x%lx.", vi->i_ino);
 561
 562	/* Setup the generic vfs inode parts now. */
 563
 564	/*
 565	 * This is for checking whether an inode has changed w.r.t. a file so
 566	 * that the file can be updated if necessary (compare with f_version).
 567	 */
 568	vi->i_version = 1;
 569
 570	vi->i_uid = vol->uid;
 571	vi->i_gid = vol->gid;
 572	vi->i_mode = 0;
 573
 574	/*
 575	 * Initialize the ntfs specific part of @vi special casing
 576	 * FILE_MFT which we need to do at mount time.
 577	 */
 578	if (vi->i_ino != FILE_MFT)
 579		ntfs_init_big_inode(vi);
 580	ni = NTFS_I(vi);
 581
 582	m = map_mft_record(ni);
 583	if (IS_ERR(m)) {
 584		err = PTR_ERR(m);
 585		goto err_out;
 586	}
 587	ctx = ntfs_attr_get_search_ctx(ni, m);
 588	if (!ctx) {
 589		err = -ENOMEM;
 590		goto unm_err_out;
 591	}
 592
 593	if (!(m->flags & MFT_RECORD_IN_USE)) {
 594		ntfs_error(vi->i_sb, "Inode is not in use!");
 595		goto unm_err_out;
 596	}
 597	if (m->base_mft_record) {
 598		ntfs_error(vi->i_sb, "Inode is an extent inode!");
 599		goto unm_err_out;
 600	}
 601
 602	/* Transfer information from mft record into vfs and ntfs inodes. */
 603	vi->i_generation = ni->seq_no = le16_to_cpu(m->sequence_number);
 604
 605	/*
 606	 * FIXME: Keep in mind that link_count is two for files which have both
 607	 * a long file name and a short file name as separate entries, so if
 608	 * we are hiding short file names this will be too high. Either we need
 609	 * to account for the short file names by subtracting them or we need
 610	 * to make sure we delete files even though i_nlink is not zero which
 611	 * might be tricky due to vfs interactions. Need to think about this
 612	 * some more when implementing the unlink command.
 613	 */
 614	set_nlink(vi, le16_to_cpu(m->link_count));
 615	/*
 616	 * FIXME: Reparse points can have the directory bit set even though
 617	 * they would be S_IFLNK. Need to deal with this further below when we
 618	 * implement reparse points / symbolic links but it will do for now.
 619	 * Also if not a directory, it could be something else, rather than
 620	 * a regular file. But again, will do for now.
 621	 */
 622	/* Everyone gets all permissions. */
 623	vi->i_mode |= S_IRWXUGO;
 624	/* If read-only, no one gets write permissions. */
 625	if (IS_RDONLY(vi))
 626		vi->i_mode &= ~S_IWUGO;
 627	if (m->flags & MFT_RECORD_IS_DIRECTORY) {
 628		vi->i_mode |= S_IFDIR;
 629		/*
 630		 * Apply the directory permissions mask set in the mount
 631		 * options.
 632		 */
 633		vi->i_mode &= ~vol->dmask;
 634		/* Things break without this kludge! */
 635		if (vi->i_nlink > 1)
 636			set_nlink(vi, 1);
 637	} else {
 638		vi->i_mode |= S_IFREG;
 639		/* Apply the file permissions mask set in the mount options. */
 640		vi->i_mode &= ~vol->fmask;
 641	}
 642	/*
 643	 * Find the standard information attribute in the mft record. At this
 644	 * stage we haven't setup the attribute list stuff yet, so this could
 645	 * in fact fail if the standard information is in an extent record, but
 646	 * I don't think this actually ever happens.
 647	 */
 648	err = ntfs_attr_lookup(AT_STANDARD_INFORMATION, NULL, 0, 0, 0, NULL, 0,
 649			ctx);
 650	if (unlikely(err)) {
 651		if (err == -ENOENT) {
 652			/*
 653			 * TODO: We should be performing a hot fix here (if the
 654			 * recover mount option is set) by creating a new
 655			 * attribute.
 656			 */
 657			ntfs_error(vi->i_sb, "$STANDARD_INFORMATION attribute "
 658					"is missing.");
 659		}
 660		goto unm_err_out;
 661	}
 662	a = ctx->attr;
 663	/* Get the standard information attribute value. */
 664	si = (STANDARD_INFORMATION*)((u8*)a +
 665			le16_to_cpu(a->data.resident.value_offset));
 666
 667	/* Transfer information from the standard information into vi. */
 668	/*
 669	 * Note: The i_?times do not quite map perfectly onto the NTFS times,
 670	 * but they are close enough, and in the end it doesn't really matter
 671	 * that much...
 672	 */
 673	/*
 674	 * mtime is the last change of the data within the file. Not changed
 675	 * when only metadata is changed, e.g. a rename doesn't affect mtime.
 676	 */
 677	vi->i_mtime = ntfs2utc(si->last_data_change_time);
 678	/*
 679	 * ctime is the last change of the metadata of the file. This obviously
 680	 * always changes, when mtime is changed. ctime can be changed on its
 681	 * own, mtime is then not changed, e.g. when a file is renamed.
 682	 */
 683	vi->i_ctime = ntfs2utc(si->last_mft_change_time);
 684	/*
 685	 * Last access to the data within the file. Not changed during a rename
 686	 * for example but changed whenever the file is written to.
 687	 */
 688	vi->i_atime = ntfs2utc(si->last_access_time);
 689
 690	/* Find the attribute list attribute if present. */
 691	ntfs_attr_reinit_search_ctx(ctx);
 692	err = ntfs_attr_lookup(AT_ATTRIBUTE_LIST, NULL, 0, 0, 0, NULL, 0, ctx);
 693	if (err) {
 694		if (unlikely(err != -ENOENT)) {
 695			ntfs_error(vi->i_sb, "Failed to lookup attribute list "
 696					"attribute.");
 697			goto unm_err_out;
 698		}
 699	} else /* if (!err) */ {
 700		if (vi->i_ino == FILE_MFT)
 701			goto skip_attr_list_load;
 702		ntfs_debug("Attribute list found in inode 0x%lx.", vi->i_ino);
 703		NInoSetAttrList(ni);
 704		a = ctx->attr;
 705		if (a->flags & ATTR_COMPRESSION_MASK) {
 706			ntfs_error(vi->i_sb, "Attribute list attribute is "
 707					"compressed.");
 708			goto unm_err_out;
 709		}
 710		if (a->flags & ATTR_IS_ENCRYPTED ||
 711				a->flags & ATTR_IS_SPARSE) {
 712			if (a->non_resident) {
 713				ntfs_error(vi->i_sb, "Non-resident attribute "
 714						"list attribute is encrypted/"
 715						"sparse.");
 716				goto unm_err_out;
 717			}
 718			ntfs_warning(vi->i_sb, "Resident attribute list "
 719					"attribute in inode 0x%lx is marked "
 720					"encrypted/sparse which is not true.  "
 721					"However, Windows allows this and "
 722					"chkdsk does not detect or correct it "
 723					"so we will just ignore the invalid "
 724					"flags and pretend they are not set.",
 725					vi->i_ino);
 726		}
 727		/* Now allocate memory for the attribute list. */
 728		ni->attr_list_size = (u32)ntfs_attr_size(a);
 729		ni->attr_list = ntfs_malloc_nofs(ni->attr_list_size);
 730		if (!ni->attr_list) {
 731			ntfs_error(vi->i_sb, "Not enough memory to allocate "
 732					"buffer for attribute list.");
 733			err = -ENOMEM;
 734			goto unm_err_out;
 735		}
 736		if (a->non_resident) {
 737			NInoSetAttrListNonResident(ni);
 738			if (a->data.non_resident.lowest_vcn) {
 739				ntfs_error(vi->i_sb, "Attribute list has non "
 740						"zero lowest_vcn.");
 741				goto unm_err_out;
 742			}
 743			/*
 744			 * Setup the runlist. No need for locking as we have
 745			 * exclusive access to the inode at this time.
 746			 */
 747			ni->attr_list_rl.rl = ntfs_mapping_pairs_decompress(vol,
 748					a, NULL);
 749			if (IS_ERR(ni->attr_list_rl.rl)) {
 750				err = PTR_ERR(ni->attr_list_rl.rl);
 751				ni->attr_list_rl.rl = NULL;
 752				ntfs_error(vi->i_sb, "Mapping pairs "
 753						"decompression failed.");
 754				goto unm_err_out;
 755			}
 756			/* Now load the attribute list. */
 757			if ((err = load_attribute_list(vol, &ni->attr_list_rl,
 758					ni->attr_list, ni->attr_list_size,
 759					sle64_to_cpu(a->data.non_resident.
 760					initialized_size)))) {
 761				ntfs_error(vi->i_sb, "Failed to load "
 762						"attribute list attribute.");
 763				goto unm_err_out;
 764			}
 765		} else /* if (!a->non_resident) */ {
 766			if ((u8*)a + le16_to_cpu(a->data.resident.value_offset)
 767					+ le32_to_cpu(
 768					a->data.resident.value_length) >
 769					(u8*)ctx->mrec + vol->mft_record_size) {
 770				ntfs_error(vi->i_sb, "Corrupt attribute list "
 771						"in inode.");
 772				goto unm_err_out;
 773			}
 774			/* Now copy the attribute list. */
 775			memcpy(ni->attr_list, (u8*)a + le16_to_cpu(
 776					a->data.resident.value_offset),
 777					le32_to_cpu(
 778					a->data.resident.value_length));
 779		}
 780	}
 781skip_attr_list_load:
 782	/*
 783	 * If an attribute list is present we now have the attribute list value
 784	 * in ntfs_ino->attr_list and it is ntfs_ino->attr_list_size bytes.
 785	 */
 786	if (S_ISDIR(vi->i_mode)) {
 787		loff_t bvi_size;
 788		ntfs_inode *bni;
 789		INDEX_ROOT *ir;
 790		u8 *ir_end, *index_end;
 791
 792		/* It is a directory, find index root attribute. */
 793		ntfs_attr_reinit_search_ctx(ctx);
 794		err = ntfs_attr_lookup(AT_INDEX_ROOT, I30, 4, CASE_SENSITIVE,
 795				0, NULL, 0, ctx);
 796		if (unlikely(err)) {
 797			if (err == -ENOENT) {
 798				// FIXME: File is corrupt! Hot-fix with empty
 799				// index root attribute if recovery option is
 800				// set.
 801				ntfs_error(vi->i_sb, "$INDEX_ROOT attribute "
 802						"is missing.");
 803			}
 804			goto unm_err_out;
 805		}
 806		a = ctx->attr;
 807		/* Set up the state. */
 808		if (unlikely(a->non_resident)) {
 809			ntfs_error(vol->sb, "$INDEX_ROOT attribute is not "
 810					"resident.");
 811			goto unm_err_out;
 812		}
 813		/* Ensure the attribute name is placed before the value. */
 814		if (unlikely(a->name_length && (le16_to_cpu(a->name_offset) >=
 815				le16_to_cpu(a->data.resident.value_offset)))) {
 816			ntfs_error(vol->sb, "$INDEX_ROOT attribute name is "
 817					"placed after the attribute value.");
 818			goto unm_err_out;
 819		}
 820		/*
 821		 * Compressed/encrypted index root just means that the newly
 822		 * created files in that directory should be created compressed/
 823		 * encrypted. However index root cannot be both compressed and
 824		 * encrypted.
 825		 */
 826		if (a->flags & ATTR_COMPRESSION_MASK)
 827			NInoSetCompressed(ni);
 828		if (a->flags & ATTR_IS_ENCRYPTED) {
 829			if (a->flags & ATTR_COMPRESSION_MASK) {
 830				ntfs_error(vi->i_sb, "Found encrypted and "
 831						"compressed attribute.");
 832				goto unm_err_out;
 833			}
 834			NInoSetEncrypted(ni);
 835		}
 836		if (a->flags & ATTR_IS_SPARSE)
 837			NInoSetSparse(ni);
 838		ir = (INDEX_ROOT*)((u8*)a +
 839				le16_to_cpu(a->data.resident.value_offset));
 840		ir_end = (u8*)ir + le32_to_cpu(a->data.resident.value_length);
 841		if (ir_end > (u8*)ctx->mrec + vol->mft_record_size) {
 842			ntfs_error(vi->i_sb, "$INDEX_ROOT attribute is "
 843					"corrupt.");
 844			goto unm_err_out;
 845		}
 846		index_end = (u8*)&ir->index +
 847				le32_to_cpu(ir->index.index_length);
 848		if (index_end > ir_end) {
 849			ntfs_error(vi->i_sb, "Directory index is corrupt.");
 850			goto unm_err_out;
 851		}
 852		if (ir->type != AT_FILE_NAME) {
 853			ntfs_error(vi->i_sb, "Indexed attribute is not "
 854					"$FILE_NAME.");
 855			goto unm_err_out;
 856		}
 857		if (ir->collation_rule != COLLATION_FILE_NAME) {
 858			ntfs_error(vi->i_sb, "Index collation rule is not "
 859					"COLLATION_FILE_NAME.");
 860			goto unm_err_out;
 861		}
 862		ni->itype.index.collation_rule = ir->collation_rule;
 863		ni->itype.index.block_size = le32_to_cpu(ir->index_block_size);
 864		if (ni->itype.index.block_size &
 865				(ni->itype.index.block_size - 1)) {
 866			ntfs_error(vi->i_sb, "Index block size (%u) is not a "
 867					"power of two.",
 868					ni->itype.index.block_size);
 869			goto unm_err_out;
 870		}
 871		if (ni->itype.index.block_size > PAGE_SIZE) {
 872			ntfs_error(vi->i_sb, "Index block size (%u) > "
 873					"PAGE_SIZE (%ld) is not "
 874					"supported.  Sorry.",
 875					ni->itype.index.block_size,
 876					PAGE_SIZE);
 877			err = -EOPNOTSUPP;
 878			goto unm_err_out;
 879		}
 880		if (ni->itype.index.block_size < NTFS_BLOCK_SIZE) {
 881			ntfs_error(vi->i_sb, "Index block size (%u) < "
 882					"NTFS_BLOCK_SIZE (%i) is not "
 883					"supported.  Sorry.",
 884					ni->itype.index.block_size,
 885					NTFS_BLOCK_SIZE);
 886			err = -EOPNOTSUPP;
 887			goto unm_err_out;
 888		}
 889		ni->itype.index.block_size_bits =
 890				ffs(ni->itype.index.block_size) - 1;
 891		/* Determine the size of a vcn in the directory index. */
 892		if (vol->cluster_size <= ni->itype.index.block_size) {
 893			ni->itype.index.vcn_size = vol->cluster_size;
 894			ni->itype.index.vcn_size_bits = vol->cluster_size_bits;
 895		} else {
 896			ni->itype.index.vcn_size = vol->sector_size;
 897			ni->itype.index.vcn_size_bits = vol->sector_size_bits;
 898		}
 899
 900		/* Setup the index allocation attribute, even if not present. */
 901		NInoSetMstProtected(ni);
 902		ni->type = AT_INDEX_ALLOCATION;
 903		ni->name = I30;
 904		ni->name_len = 4;
 905
 906		if (!(ir->index.flags & LARGE_INDEX)) {
 907			/* No index allocation. */
 908			vi->i_size = ni->initialized_size =
 909					ni->allocated_size = 0;
 910			/* We are done with the mft record, so we release it. */
 911			ntfs_attr_put_search_ctx(ctx);
 912			unmap_mft_record(ni);
 913			m = NULL;
 914			ctx = NULL;
 915			goto skip_large_dir_stuff;
 916		} /* LARGE_INDEX: Index allocation present. Setup state. */
 917		NInoSetIndexAllocPresent(ni);
 918		/* Find index allocation attribute. */
 919		ntfs_attr_reinit_search_ctx(ctx);
 920		err = ntfs_attr_lookup(AT_INDEX_ALLOCATION, I30, 4,
 921				CASE_SENSITIVE, 0, NULL, 0, ctx);
 922		if (unlikely(err)) {
 923			if (err == -ENOENT)
 924				ntfs_error(vi->i_sb, "$INDEX_ALLOCATION "
 925						"attribute is not present but "
 926						"$INDEX_ROOT indicated it is.");
 927			else
 928				ntfs_error(vi->i_sb, "Failed to lookup "
 929						"$INDEX_ALLOCATION "
 930						"attribute.");
 931			goto unm_err_out;
 932		}
 933		a = ctx->attr;
 934		if (!a->non_resident) {
 935			ntfs_error(vi->i_sb, "$INDEX_ALLOCATION attribute "
 936					"is resident.");
 937			goto unm_err_out;
 938		}
 939		/*
 940		 * Ensure the attribute name is placed before the mapping pairs
 941		 * array.
 942		 */
 943		if (unlikely(a->name_length && (le16_to_cpu(a->name_offset) >=
 944				le16_to_cpu(
 945				a->data.non_resident.mapping_pairs_offset)))) {
 946			ntfs_error(vol->sb, "$INDEX_ALLOCATION attribute name "
 947					"is placed after the mapping pairs "
 948					"array.");
 949			goto unm_err_out;
 950		}
 951		if (a->flags & ATTR_IS_ENCRYPTED) {
 952			ntfs_error(vi->i_sb, "$INDEX_ALLOCATION attribute "
 953					"is encrypted.");
 954			goto unm_err_out;
 955		}
 956		if (a->flags & ATTR_IS_SPARSE) {
 957			ntfs_error(vi->i_sb, "$INDEX_ALLOCATION attribute "
 958					"is sparse.");
 959			goto unm_err_out;
 960		}
 961		if (a->flags & ATTR_COMPRESSION_MASK) {
 962			ntfs_error(vi->i_sb, "$INDEX_ALLOCATION attribute "
 963					"is compressed.");
 964			goto unm_err_out;
 965		}
 966		if (a->data.non_resident.lowest_vcn) {
 967			ntfs_error(vi->i_sb, "First extent of "
 968					"$INDEX_ALLOCATION attribute has non "
 969					"zero lowest_vcn.");
 970			goto unm_err_out;
 971		}
 972		vi->i_size = sle64_to_cpu(a->data.non_resident.data_size);
 973		ni->initialized_size = sle64_to_cpu(
 974				a->data.non_resident.initialized_size);
 975		ni->allocated_size = sle64_to_cpu(
 976				a->data.non_resident.allocated_size);
 977		/*
 978		 * We are done with the mft record, so we release it. Otherwise
 979		 * we would deadlock in ntfs_attr_iget().
 980		 */
 981		ntfs_attr_put_search_ctx(ctx);
 982		unmap_mft_record(ni);
 983		m = NULL;
 984		ctx = NULL;
 985		/* Get the index bitmap attribute inode. */
 986		bvi = ntfs_attr_iget(vi, AT_BITMAP, I30, 4);
 987		if (IS_ERR(bvi)) {
 988			ntfs_error(vi->i_sb, "Failed to get bitmap attribute.");
 989			err = PTR_ERR(bvi);
 990			goto unm_err_out;
 991		}
 992		bni = NTFS_I(bvi);
 993		if (NInoCompressed(bni) || NInoEncrypted(bni) ||
 994				NInoSparse(bni)) {
 995			ntfs_error(vi->i_sb, "$BITMAP attribute is compressed "
 996					"and/or encrypted and/or sparse.");
 997			goto iput_unm_err_out;
 998		}
 999		/* Consistency check bitmap size vs. index allocation size. */
1000		bvi_size = i_size_read(bvi);
1001		if ((bvi_size << 3) < (vi->i_size >>
1002				ni->itype.index.block_size_bits)) {
1003			ntfs_error(vi->i_sb, "Index bitmap too small (0x%llx) "
1004					"for index allocation (0x%llx).",
1005					bvi_size << 3, vi->i_size);
1006			goto iput_unm_err_out;
1007		}
1008		/* No longer need the bitmap attribute inode. */
1009		iput(bvi);
1010skip_large_dir_stuff:
1011		/* Setup the operations for this inode. */
1012		vi->i_op = &ntfs_dir_inode_ops;
1013		vi->i_fop = &ntfs_dir_ops;
1014		vi->i_mapping->a_ops = &ntfs_mst_aops;
1015	} else {
1016		/* It is a file. */
1017		ntfs_attr_reinit_search_ctx(ctx);
1018
1019		/* Setup the data attribute, even if not present. */
1020		ni->type = AT_DATA;
1021		ni->name = NULL;
1022		ni->name_len = 0;
1023
1024		/* Find first extent of the unnamed data attribute. */
1025		err = ntfs_attr_lookup(AT_DATA, NULL, 0, 0, 0, NULL, 0, ctx);
1026		if (unlikely(err)) {
1027			vi->i_size = ni->initialized_size =
1028					ni->allocated_size = 0;
1029			if (err != -ENOENT) {
1030				ntfs_error(vi->i_sb, "Failed to lookup $DATA "
1031						"attribute.");
1032				goto unm_err_out;
1033			}
1034			/*
1035			 * FILE_Secure does not have an unnamed $DATA
1036			 * attribute, so we special case it here.
1037			 */
1038			if (vi->i_ino == FILE_Secure)
1039				goto no_data_attr_special_case;
1040			/*
1041			 * Most if not all the system files in the $Extend
1042			 * system directory do not have unnamed data
1043			 * attributes so we need to check if the parent
1044			 * directory of the file is FILE_Extend and if it is
1045			 * ignore this error. To do this we need to get the
1046			 * name of this inode from the mft record as the name
1047			 * contains the back reference to the parent directory.
1048			 */
1049			if (ntfs_is_extended_system_file(ctx) > 0)
1050				goto no_data_attr_special_case;
1051			// FIXME: File is corrupt! Hot-fix with empty data
1052			// attribute if recovery option is set.
1053			ntfs_error(vi->i_sb, "$DATA attribute is missing.");
1054			goto unm_err_out;
1055		}
1056		a = ctx->attr;
1057		/* Setup the state. */
1058		if (a->flags & (ATTR_COMPRESSION_MASK | ATTR_IS_SPARSE)) {
1059			if (a->flags & ATTR_COMPRESSION_MASK) {
1060				NInoSetCompressed(ni);
1061				if (vol->cluster_size > 4096) {
1062					ntfs_error(vi->i_sb, "Found "
1063							"compressed data but "
1064							"compression is "
1065							"disabled due to "
1066							"cluster size (%i) > "
1067							"4kiB.",
1068							vol->cluster_size);
1069					goto unm_err_out;
1070				}
1071				if ((a->flags & ATTR_COMPRESSION_MASK)
1072						!= ATTR_IS_COMPRESSED) {
1073					ntfs_error(vi->i_sb, "Found unknown "
1074							"compression method "
1075							"or corrupt file.");
1076					goto unm_err_out;
1077				}
1078			}
1079			if (a->flags & ATTR_IS_SPARSE)
1080				NInoSetSparse(ni);
1081		}
1082		if (a->flags & ATTR_IS_ENCRYPTED) {
1083			if (NInoCompressed(ni)) {
1084				ntfs_error(vi->i_sb, "Found encrypted and "
1085						"compressed data.");
1086				goto unm_err_out;
1087			}
1088			NInoSetEncrypted(ni);
1089		}
1090		if (a->non_resident) {
1091			NInoSetNonResident(ni);
1092			if (NInoCompressed(ni) || NInoSparse(ni)) {
1093				if (NInoCompressed(ni) && a->data.non_resident.
1094						compression_unit != 4) {
1095					ntfs_error(vi->i_sb, "Found "
1096							"non-standard "
1097							"compression unit (%u "
1098							"instead of 4).  "
1099							"Cannot handle this.",
1100							a->data.non_resident.
1101							compression_unit);
1102					err = -EOPNOTSUPP;
1103					goto unm_err_out;
1104				}
1105				if (a->data.non_resident.compression_unit) {
1106					ni->itype.compressed.block_size = 1U <<
1107							(a->data.non_resident.
1108							compression_unit +
1109							vol->cluster_size_bits);
1110					ni->itype.compressed.block_size_bits =
1111							ffs(ni->itype.
1112							compressed.
1113							block_size) - 1;
1114					ni->itype.compressed.block_clusters =
1115							1U << a->data.
1116							non_resident.
1117							compression_unit;
1118				} else {
1119					ni->itype.compressed.block_size = 0;
1120					ni->itype.compressed.block_size_bits =
1121							0;
1122					ni->itype.compressed.block_clusters =
1123							0;
1124				}
1125				ni->itype.compressed.size = sle64_to_cpu(
1126						a->data.non_resident.
1127						compressed_size);
1128			}
1129			if (a->data.non_resident.lowest_vcn) {
1130				ntfs_error(vi->i_sb, "First extent of $DATA "
1131						"attribute has non zero "
1132						"lowest_vcn.");
1133				goto unm_err_out;
1134			}
1135			vi->i_size = sle64_to_cpu(
1136					a->data.non_resident.data_size);
1137			ni->initialized_size = sle64_to_cpu(
1138					a->data.non_resident.initialized_size);
1139			ni->allocated_size = sle64_to_cpu(
1140					a->data.non_resident.allocated_size);
1141		} else { /* Resident attribute. */
1142			vi->i_size = ni->initialized_size = le32_to_cpu(
1143					a->data.resident.value_length);
1144			ni->allocated_size = le32_to_cpu(a->length) -
1145					le16_to_cpu(
1146					a->data.resident.value_offset);
1147			if (vi->i_size > ni->allocated_size) {
1148				ntfs_error(vi->i_sb, "Resident data attribute "
1149						"is corrupt (size exceeds "
1150						"allocation).");
1151				goto unm_err_out;
1152			}
1153		}
1154no_data_attr_special_case:
1155		/* We are done with the mft record, so we release it. */
1156		ntfs_attr_put_search_ctx(ctx);
1157		unmap_mft_record(ni);
1158		m = NULL;
1159		ctx = NULL;
1160		/* Setup the operations for this inode. */
1161		vi->i_op = &ntfs_file_inode_ops;
1162		vi->i_fop = &ntfs_file_ops;
1163		vi->i_mapping->a_ops = &ntfs_normal_aops;
1164		if (NInoMstProtected(ni))
1165			vi->i_mapping->a_ops = &ntfs_mst_aops;
1166		else if (NInoCompressed(ni))
1167			vi->i_mapping->a_ops = &ntfs_compressed_aops;
1168	}
1169	/*
1170	 * The number of 512-byte blocks used on disk (for stat). This is in so
1171	 * far inaccurate as it doesn't account for any named streams or other
1172	 * special non-resident attributes, but that is how Windows works, too,
1173	 * so we are at least consistent with Windows, if not entirely
1174	 * consistent with the Linux Way. Doing it the Linux Way would cause a
1175	 * significant slowdown as it would involve iterating over all
1176	 * attributes in the mft record and adding the allocated/compressed
1177	 * sizes of all non-resident attributes present to give us the Linux
1178	 * correct size that should go into i_blocks (after division by 512).
1179	 */
1180	if (S_ISREG(vi->i_mode) && (NInoCompressed(ni) || NInoSparse(ni)))
1181		vi->i_blocks = ni->itype.compressed.size >> 9;
1182	else
1183		vi->i_blocks = ni->allocated_size >> 9;
1184	ntfs_debug("Done.");
1185	return 0;
1186iput_unm_err_out:
1187	iput(bvi);
1188unm_err_out:
1189	if (!err)
1190		err = -EIO;
1191	if (ctx)
1192		ntfs_attr_put_search_ctx(ctx);
1193	if (m)
1194		unmap_mft_record(ni);
1195err_out:
1196	ntfs_error(vol->sb, "Failed with error code %i.  Marking corrupt "
1197			"inode 0x%lx as bad.  Run chkdsk.", err, vi->i_ino);
1198	make_bad_inode(vi);
1199	if (err != -EOPNOTSUPP && err != -ENOMEM)
1200		NVolSetErrors(vol);
1201	return err;
1202}
1203
1204/**
1205 * ntfs_read_locked_attr_inode - read an attribute inode from its base inode
1206 * @base_vi:	base inode
1207 * @vi:		attribute inode to read
1208 *
1209 * ntfs_read_locked_attr_inode() is called from ntfs_attr_iget() to read the
1210 * attribute inode described by @vi into memory from the base mft record
1211 * described by @base_ni.
1212 *
1213 * ntfs_read_locked_attr_inode() maps, pins and locks the base inode for
1214 * reading and looks up the attribute described by @vi before setting up the
1215 * necessary fields in @vi as well as initializing the ntfs inode.
1216 *
1217 * Q: What locks are held when the function is called?
1218 * A: i_state has I_NEW set, hence the inode is locked, also
1219 *    i_count is set to 1, so it is not going to go away
1220 *
1221 * Return 0 on success and -errno on error.  In the error case, the inode will
1222 * have had make_bad_inode() executed on it.
1223 *
1224 * Note this cannot be called for AT_INDEX_ALLOCATION.
1225 */
1226static int ntfs_read_locked_attr_inode(struct inode *base_vi, struct inode *vi)
1227{
1228	ntfs_volume *vol = NTFS_SB(vi->i_sb);
1229	ntfs_inode *ni, *base_ni;
1230	MFT_RECORD *m;
1231	ATTR_RECORD *a;
1232	ntfs_attr_search_ctx *ctx;
1233	int err = 0;
1234
1235	ntfs_debug("Entering for i_ino 0x%lx.", vi->i_ino);
1236
1237	ntfs_init_big_inode(vi);
1238
1239	ni	= NTFS_I(vi);
1240	base_ni = NTFS_I(base_vi);
1241
1242	/* Just mirror the values from the base inode. */
1243	vi->i_version	= base_vi->i_version;
1244	vi->i_uid	= base_vi->i_uid;
1245	vi->i_gid	= base_vi->i_gid;
1246	set_nlink(vi, base_vi->i_nlink);
1247	vi->i_mtime	= base_vi->i_mtime;
1248	vi->i_ctime	= base_vi->i_ctime;
1249	vi->i_atime	= base_vi->i_atime;
1250	vi->i_generation = ni->seq_no = base_ni->seq_no;
1251
1252	/* Set inode type to zero but preserve permissions. */
1253	vi->i_mode	= base_vi->i_mode & ~S_IFMT;
1254
1255	m = map_mft_record(base_ni);
1256	if (IS_ERR(m)) {
1257		err = PTR_ERR(m);
1258		goto err_out;
1259	}
1260	ctx = ntfs_attr_get_search_ctx(base_ni, m);
1261	if (!ctx) {
1262		err = -ENOMEM;
1263		goto unm_err_out;
1264	}
1265	/* Find the attribute. */
1266	err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len,
1267			CASE_SENSITIVE, 0, NULL, 0, ctx);
1268	if (unlikely(err))
1269		goto unm_err_out;
1270	a = ctx->attr;
1271	if (a->flags & (ATTR_COMPRESSION_MASK | ATTR_IS_SPARSE)) {
1272		if (a->flags & ATTR_COMPRESSION_MASK) {
1273			NInoSetCompressed(ni);
1274			if ((ni->type != AT_DATA) || (ni->type == AT_DATA &&
1275					ni->name_len)) {
1276				ntfs_error(vi->i_sb, "Found compressed "
1277						"non-data or named data "
1278						"attribute.  Please report "
1279						"you saw this message to "
1280						"linux-ntfs-dev@lists."
1281						"sourceforge.net");
1282				goto unm_err_out;
1283			}
1284			if (vol->cluster_size > 4096) {
1285				ntfs_error(vi->i_sb, "Found compressed "
1286						"attribute but compression is "
1287						"disabled due to cluster size "
1288						"(%i) > 4kiB.",
1289						vol->cluster_size);
1290				goto unm_err_out;
1291			}
1292			if ((a->flags & ATTR_COMPRESSION_MASK) !=
1293					ATTR_IS_COMPRESSED) {
1294				ntfs_error(vi->i_sb, "Found unknown "
1295						"compression method.");
1296				goto unm_err_out;
1297			}
1298		}
1299		/*
1300		 * The compressed/sparse flag set in an index root just means
1301		 * to compress all files.
1302		 */
1303		if (NInoMstProtected(ni) && ni->type != AT_INDEX_ROOT) {
1304			ntfs_error(vi->i_sb, "Found mst protected attribute "
1305					"but the attribute is %s.  Please "
1306					"report you saw this message to "
1307					"linux-ntfs-dev@lists.sourceforge.net",
1308					NInoCompressed(ni) ? "compressed" :
1309					"sparse");
1310			goto unm_err_out;
1311		}
1312		if (a->flags & ATTR_IS_SPARSE)
1313			NInoSetSparse(ni);
1314	}
1315	if (a->flags & ATTR_IS_ENCRYPTED) {
1316		if (NInoCompressed(ni)) {
1317			ntfs_error(vi->i_sb, "Found encrypted and compressed "
1318					"data.");
1319			goto unm_err_out;
1320		}
1321		/*
1322		 * The encryption flag set in an index root just means to
1323		 * encrypt all files.
1324		 */
1325		if (NInoMstProtected(ni) && ni->type != AT_INDEX_ROOT) {
1326			ntfs_error(vi->i_sb, "Found mst protected attribute "
1327					"but the attribute is encrypted.  "
1328					"Please report you saw this message "
1329					"to linux-ntfs-dev@lists.sourceforge."
1330					"net");
1331			goto unm_err_out;
1332		}
1333		if (ni->type != AT_DATA) {
1334			ntfs_error(vi->i_sb, "Found encrypted non-data "
1335					"attribute.");
1336			goto unm_err_out;
1337		}
1338		NInoSetEncrypted(ni);
1339	}
1340	if (!a->non_resident) {
1341		/* Ensure the attribute name is placed before the value. */
1342		if (unlikely(a->name_length && (le16_to_cpu(a->name_offset) >=
1343				le16_to_cpu(a->data.resident.value_offset)))) {
1344			ntfs_error(vol->sb, "Attribute name is placed after "
1345					"the attribute value.");
1346			goto unm_err_out;
1347		}
1348		if (NInoMstProtected(ni)) {
1349			ntfs_error(vi->i_sb, "Found mst protected attribute "
1350					"but the attribute is resident.  "
1351					"Please report you saw this message to "
1352					"linux-ntfs-dev@lists.sourceforge.net");
1353			goto unm_err_out;
1354		}
1355		vi->i_size = ni->initialized_size = le32_to_cpu(
1356				a->data.resident.value_length);
1357		ni->allocated_size = le32_to_cpu(a->length) -
1358				le16_to_cpu(a->data.resident.value_offset);
1359		if (vi->i_size > ni->allocated_size) {
1360			ntfs_error(vi->i_sb, "Resident attribute is corrupt "
1361					"(size exceeds allocation).");
1362			goto unm_err_out;
1363		}
1364	} else {
1365		NInoSetNonResident(ni);
1366		/*
1367		 * Ensure the attribute name is placed before the mapping pairs
1368		 * array.
1369		 */
1370		if (unlikely(a->name_length && (le16_to_cpu(a->name_offset) >=
1371				le16_to_cpu(
1372				a->data.non_resident.mapping_pairs_offset)))) {
1373			ntfs_error(vol->sb, "Attribute name is placed after "
1374					"the mapping pairs array.");
1375			goto unm_err_out;
1376		}
1377		if (NInoCompressed(ni) || NInoSparse(ni)) {
1378			if (NInoCompressed(ni) && a->data.non_resident.
1379					compression_unit != 4) {
1380				ntfs_error(vi->i_sb, "Found non-standard "
1381						"compression unit (%u instead "
1382						"of 4).  Cannot handle this.",
1383						a->data.non_resident.
1384						compression_unit);
1385				err = -EOPNOTSUPP;
1386				goto unm_err_out;
1387			}
1388			if (a->data.non_resident.compression_unit) {
1389				ni->itype.compressed.block_size = 1U <<
1390						(a->data.non_resident.
1391						compression_unit +
1392						vol->cluster_size_bits);
1393				ni->itype.compressed.block_size_bits =
1394						ffs(ni->itype.compressed.
1395						block_size) - 1;
1396				ni->itype.compressed.block_clusters = 1U <<
1397						a->data.non_resident.
1398						compression_unit;
1399			} else {
1400				ni->itype.compressed.block_size = 0;
1401				ni->itype.compressed.block_size_bits = 0;
1402				ni->itype.compressed.block_clusters = 0;
1403			}
1404			ni->itype.compressed.size = sle64_to_cpu(
1405					a->data.non_resident.compressed_size);
1406		}
1407		if (a->data.non_resident.lowest_vcn) {
1408			ntfs_error(vi->i_sb, "First extent of attribute has "
1409					"non-zero lowest_vcn.");
1410			goto unm_err_out;
1411		}
1412		vi->i_size = sle64_to_cpu(a->data.non_resident.data_size);
1413		ni->initialized_size = sle64_to_cpu(
1414				a->data.non_resident.initialized_size);
1415		ni->allocated_size = sle64_to_cpu(
1416				a->data.non_resident.allocated_size);
1417	}
1418	vi->i_mapping->a_ops = &ntfs_normal_aops;
1419	if (NInoMstProtected(ni))
1420		vi->i_mapping->a_ops = &ntfs_mst_aops;
1421	else if (NInoCompressed(ni))
1422		vi->i_mapping->a_ops = &ntfs_compressed_aops;
1423	if ((NInoCompressed(ni) || NInoSparse(ni)) && ni->type != AT_INDEX_ROOT)
1424		vi->i_blocks = ni->itype.compressed.size >> 9;
1425	else
1426		vi->i_blocks = ni->allocated_size >> 9;
1427	/*
1428	 * Make sure the base inode does not go away and attach it to the
1429	 * attribute inode.
1430	 */
1431	igrab(base_vi);
1432	ni->ext.base_ntfs_ino = base_ni;
1433	ni->nr_extents = -1;
1434
1435	ntfs_attr_put_search_ctx(ctx);
1436	unmap_mft_record(base_ni);
1437
1438	ntfs_debug("Done.");
1439	return 0;
1440
1441unm_err_out:
1442	if (!err)
1443		err = -EIO;
1444	if (ctx)
1445		ntfs_attr_put_search_ctx(ctx);
1446	unmap_mft_record(base_ni);
1447err_out:
1448	ntfs_error(vol->sb, "Failed with error code %i while reading attribute "
1449			"inode (mft_no 0x%lx, type 0x%x, name_len %i).  "
1450			"Marking corrupt inode and base inode 0x%lx as bad.  "
1451			"Run chkdsk.", err, vi->i_ino, ni->type, ni->name_len,
1452			base_vi->i_ino);
1453	make_bad_inode(vi);
1454	if (err != -ENOMEM)
1455		NVolSetErrors(vol);
1456	return err;
1457}
1458
1459/**
1460 * ntfs_read_locked_index_inode - read an index inode from its base inode
1461 * @base_vi:	base inode
1462 * @vi:		index inode to read
1463 *
1464 * ntfs_read_locked_index_inode() is called from ntfs_index_iget() to read the
1465 * index inode described by @vi into memory from the base mft record described
1466 * by @base_ni.
1467 *
1468 * ntfs_read_locked_index_inode() maps, pins and locks the base inode for
1469 * reading and looks up the attributes relating to the index described by @vi
1470 * before setting up the necessary fields in @vi as well as initializing the
1471 * ntfs inode.
1472 *
1473 * Note, index inodes are essentially attribute inodes (NInoAttr() is true)
1474 * with the attribute type set to AT_INDEX_ALLOCATION.  Apart from that, they
1475 * are setup like directory inodes since directories are a special case of
1476 * indices ao they need to be treated in much the same way.  Most importantly,
1477 * for small indices the index allocation attribute might not actually exist.
1478 * However, the index root attribute always exists but this does not need to
1479 * have an inode associated with it and this is why we define a new inode type
1480 * index.  Also, like for directories, we need to have an attribute inode for
1481 * the bitmap attribute corresponding to the index allocation attribute and we
1482 * can store this in the appropriate field of the inode, just like we do for
1483 * normal directory inodes.
1484 *
1485 * Q: What locks are held when the function is called?
1486 * A: i_state has I_NEW set, hence the inode is locked, also
1487 *    i_count is set to 1, so it is not going to go away
1488 *
1489 * Return 0 on success and -errno on error.  In the error case, the inode will
1490 * have had make_bad_inode() executed on it.
1491 */
1492static int ntfs_read_locked_index_inode(struct inode *base_vi, struct inode *vi)
1493{
1494	loff_t bvi_size;
1495	ntfs_volume *vol = NTFS_SB(vi->i_sb);
1496	ntfs_inode *ni, *base_ni, *bni;
1497	struct inode *bvi;
1498	MFT_RECORD *m;
1499	ATTR_RECORD *a;
1500	ntfs_attr_search_ctx *ctx;
1501	INDEX_ROOT *ir;
1502	u8 *ir_end, *index_end;
1503	int err = 0;
1504
1505	ntfs_debug("Entering for i_ino 0x%lx.", vi->i_ino);
1506	ntfs_init_big_inode(vi);
1507	ni	= NTFS_I(vi);
1508	base_ni = NTFS_I(base_vi);
1509	/* Just mirror the values from the base inode. */
1510	vi->i_version	= base_vi->i_version;
1511	vi->i_uid	= base_vi->i_uid;
1512	vi->i_gid	= base_vi->i_gid;
1513	set_nlink(vi, base_vi->i_nlink);
1514	vi->i_mtime	= base_vi->i_mtime;
1515	vi->i_ctime	= base_vi->i_ctime;
1516	vi->i_atime	= base_vi->i_atime;
1517	vi->i_generation = ni->seq_no = base_ni->seq_no;
1518	/* Set inode type to zero but preserve permissions. */
1519	vi->i_mode	= base_vi->i_mode & ~S_IFMT;
1520	/* Map the mft record for the base inode. */
1521	m = map_mft_record(base_ni);
1522	if (IS_ERR(m)) {
1523		err = PTR_ERR(m);
1524		goto err_out;
1525	}
1526	ctx = ntfs_attr_get_search_ctx(base_ni, m);
1527	if (!ctx) {
1528		err = -ENOMEM;
1529		goto unm_err_out;
1530	}
1531	/* Find the index root attribute. */
1532	err = ntfs_attr_lookup(AT_INDEX_ROOT, ni->name, ni->name_len,
1533			CASE_SENSITIVE, 0, NULL, 0, ctx);
1534	if (unlikely(err)) {
1535		if (err == -ENOENT)
1536			ntfs_error(vi->i_sb, "$INDEX_ROOT attribute is "
1537					"missing.");
1538		goto unm_err_out;
1539	}
1540	a = ctx->attr;
1541	/* Set up the state. */
1542	if (unlikely(a->non_resident)) {
1543		ntfs_error(vol->sb, "$INDEX_ROOT attribute is not resident.");
1544		goto unm_err_out;
1545	}
1546	/* Ensure the attribute name is placed before the value. */
1547	if (unlikely(a->name_length && (le16_to_cpu(a->name_offset) >=
1548			le16_to_cpu(a->data.resident.value_offset)))) {
1549		ntfs_error(vol->sb, "$INDEX_ROOT attribute name is placed "
1550				"after the attribute value.");
1551		goto unm_err_out;
1552	}
1553	/*
1554	 * Compressed/encrypted/sparse index root is not allowed, except for
1555	 * directories of course but those are not dealt with here.
1556	 */
1557	if (a->flags & (ATTR_COMPRESSION_MASK | ATTR_IS_ENCRYPTED |
1558			ATTR_IS_SPARSE)) {
1559		ntfs_error(vi->i_sb, "Found compressed/encrypted/sparse index "
1560				"root attribute.");
1561		goto unm_err_out;
1562	}
1563	ir = (INDEX_ROOT*)((u8*)a + le16_to_cpu(a->data.resident.value_offset));
1564	ir_end = (u8*)ir + le32_to_cpu(a->data.resident.value_length);
1565	if (ir_end > (u8*)ctx->mrec + vol->mft_record_size) {
1566		ntfs_error(vi->i_sb, "$INDEX_ROOT attribute is corrupt.");
1567		goto unm_err_out;
1568	}
1569	index_end = (u8*)&ir->index + le32_to_cpu(ir->index.index_length);
1570	if (index_end > ir_end) {
1571		ntfs_error(vi->i_sb, "Index is corrupt.");
1572		goto unm_err_out;
1573	}
1574	if (ir->type) {
1575		ntfs_error(vi->i_sb, "Index type is not 0 (type is 0x%x).",
1576				le32_to_cpu(ir->type));
1577		goto unm_err_out;
1578	}
1579	ni->itype.index.collation_rule = ir->collation_rule;
1580	ntfs_debug("Index collation rule is 0x%x.",
1581			le32_to_cpu(ir->collation_rule));
1582	ni->itype.index.block_size = le32_to_cpu(ir->index_block_size);
1583	if (!is_power_of_2(ni->itype.index.block_size)) {
1584		ntfs_error(vi->i_sb, "Index block size (%u) is not a power of "
1585				"two.", ni->itype.index.block_size);
1586		goto unm_err_out;
1587	}
1588	if (ni->itype.index.block_size > PAGE_SIZE) {
1589		ntfs_error(vi->i_sb, "Index block size (%u) > PAGE_SIZE "
1590				"(%ld) is not supported.  Sorry.",
1591				ni->itype.index.block_size, PAGE_SIZE);
1592		err = -EOPNOTSUPP;
1593		goto unm_err_out;
1594	}
1595	if (ni->itype.index.block_size < NTFS_BLOCK_SIZE) {
1596		ntfs_error(vi->i_sb, "Index block size (%u) < NTFS_BLOCK_SIZE "
1597				"(%i) is not supported.  Sorry.",
1598				ni->itype.index.block_size, NTFS_BLOCK_SIZE);
1599		err = -EOPNOTSUPP;
1600		goto unm_err_out;
1601	}
1602	ni->itype.index.block_size_bits = ffs(ni->itype.index.block_size) - 1;
1603	/* Determine the size of a vcn in the index. */
1604	if (vol->cluster_size <= ni->itype.index.block_size) {
1605		ni->itype.index.vcn_size = vol->cluster_size;
1606		ni->itype.index.vcn_size_bits = vol->cluster_size_bits;
1607	} else {
1608		ni->itype.index.vcn_size = vol->sector_size;
1609		ni->itype.index.vcn_size_bits = vol->sector_size_bits;
1610	}
1611	/* Check for presence of index allocation attribute. */
1612	if (!(ir->index.flags & LARGE_INDEX)) {
1613		/* No index allocation. */
1614		vi->i_size = ni->initialized_size = ni->allocated_size = 0;
1615		/* We are done with the mft record, so we release it. */
1616		ntfs_attr_put_search_ctx(ctx);
1617		unmap_mft_record(base_ni);
1618		m = NULL;
1619		ctx = NULL;
1620		goto skip_large_index_stuff;
1621	} /* LARGE_INDEX:  Index allocation present.  Setup state. */
1622	NInoSetIndexAllocPresent(ni);
1623	/* Find index allocation attribute. */
1624	ntfs_attr_reinit_search_ctx(ctx);
1625	err = ntfs_attr_lookup(AT_INDEX_ALLOCATION, ni->name, ni->name_len,
1626			CASE_SENSITIVE, 0, NULL, 0, ctx);
1627	if (unlikely(err)) {
1628		if (err == -ENOENT)
1629			ntfs_error(vi->i_sb, "$INDEX_ALLOCATION attribute is "
1630					"not present but $INDEX_ROOT "
1631					"indicated it is.");
1632		else
1633			ntfs_error(vi->i_sb, "Failed to lookup "
1634					"$INDEX_ALLOCATION attribute.");
1635		goto unm_err_out;
1636	}
1637	a = ctx->attr;
1638	if (!a->non_resident) {
1639		ntfs_error(vi->i_sb, "$INDEX_ALLOCATION attribute is "
1640				"resident.");
1641		goto unm_err_out;
1642	}
1643	/*
1644	 * Ensure the attribute name is placed before the mapping pairs array.
1645	 */
1646	if (unlikely(a->name_length && (le16_to_cpu(a->name_offset) >=
1647			le16_to_cpu(
1648			a->data.non_resident.mapping_pairs_offset)))) {
1649		ntfs_error(vol->sb, "$INDEX_ALLOCATION attribute name is "
1650				"placed after the mapping pairs array.");
1651		goto unm_err_out;
1652	}
1653	if (a->flags & ATTR_IS_ENCRYPTED) {
1654		ntfs_error(vi->i_sb, "$INDEX_ALLOCATION attribute is "
1655				"encrypted.");
1656		goto unm_err_out;
1657	}
1658	if (a->flags & ATTR_IS_SPARSE) {
1659		ntfs_error(vi->i_sb, "$INDEX_ALLOCATION attribute is sparse.");
1660		goto unm_err_out;
1661	}
1662	if (a->flags & ATTR_COMPRESSION_MASK) {
1663		ntfs_error(vi->i_sb, "$INDEX_ALLOCATION attribute is "
1664				"compressed.");
1665		goto unm_err_out;
1666	}
1667	if (a->data.non_resident.lowest_vcn) {
1668		ntfs_error(vi->i_sb, "First extent of $INDEX_ALLOCATION "
1669				"attribute has non zero lowest_vcn.");
1670		goto unm_err_out;
1671	}
1672	vi->i_size = sle64_to_cpu(a->data.non_resident.data_size);
1673	ni->initialized_size = sle64_to_cpu(
1674			a->data.non_resident.initialized_size);
1675	ni->allocated_size = sle64_to_cpu(a->data.non_resident.allocated_size);
1676	/*
1677	 * We are done with the mft record, so we release it.  Otherwise
1678	 * we would deadlock in ntfs_attr_iget().
1679	 */
1680	ntfs_attr_put_search_ctx(ctx);
1681	unmap_mft_record(base_ni);
1682	m = NULL;
1683	ctx = NULL;
1684	/* Get the index bitmap attribute inode. */
1685	bvi = ntfs_attr_iget(base_vi, AT_BITMAP, ni->name, ni->name_len);
1686	if (IS_ERR(bvi)) {
1687		ntfs_error(vi->i_sb, "Failed to get bitmap attribute.");
1688		err = PTR_ERR(bvi);
1689		goto unm_err_out;
1690	}
1691	bni = NTFS_I(bvi);
1692	if (NInoCompressed(bni) || NInoEncrypted(bni) ||
1693			NInoSparse(bni)) {
1694		ntfs_error(vi->i_sb, "$BITMAP attribute is compressed and/or "
1695				"encrypted and/or sparse.");
1696		goto iput_unm_err_out;
1697	}
1698	/* Consistency check bitmap size vs. index allocation size. */
1699	bvi_size = i_size_read(bvi);
1700	if ((bvi_size << 3) < (vi->i_size >> ni->itype.index.block_size_bits)) {
1701		ntfs_error(vi->i_sb, "Index bitmap too small (0x%llx) for "
1702				"index allocation (0x%llx).", bvi_size << 3,
1703				vi->i_size);
1704		goto iput_unm_err_out;
1705	}
1706	iput(bvi);
1707skip_large_index_stuff:
1708	/* Setup the operations for this index inode. */
1709	vi->i_mapping->a_ops = &ntfs_mst_aops;
1710	vi->i_blocks = ni->allocated_size >> 9;
1711	/*
1712	 * Make sure the base inode doesn't go away and attach it to the
1713	 * index inode.
1714	 */
1715	igrab(base_vi);
1716	ni->ext.base_ntfs_ino = base_ni;
1717	ni->nr_extents = -1;
1718
1719	ntfs_debug("Done.");
1720	return 0;
1721iput_unm_err_out:
1722	iput(bvi);
1723unm_err_out:
1724	if (!err)
1725		err = -EIO;
1726	if (ctx)
1727		ntfs_attr_put_search_ctx(ctx);
1728	if (m)
1729		unmap_mft_record(base_ni);
1730err_out:
1731	ntfs_error(vi->i_sb, "Failed with error code %i while reading index "
1732			"inode (mft_no 0x%lx, name_len %i.", err, vi->i_ino,
1733			ni->name_len);
1734	make_bad_inode(vi);
1735	if (err != -EOPNOTSUPP && err != -ENOMEM)
1736		NVolSetErrors(vol);
1737	return err;
1738}
1739
1740/*
1741 * The MFT inode has special locking, so teach the lock validator
1742 * about this by splitting off the locking rules of the MFT from
1743 * the locking rules of other inodes. The MFT inode can never be
1744 * accessed from the VFS side (or even internally), only by the
1745 * map_mft functions.
1746 */
1747static struct lock_class_key mft_ni_runlist_lock_key, mft_ni_mrec_lock_key;
1748
1749/**
1750 * ntfs_read_inode_mount - special read_inode for mount time use only
1751 * @vi:		inode to read
1752 *
1753 * Read inode FILE_MFT at mount time, only called with super_block lock
1754 * held from within the read_super() code path.
1755 *
1756 * This function exists because when it is called the page cache for $MFT/$DATA
1757 * is not initialized and hence we cannot get at the contents of mft records
1758 * by calling map_mft_record*().
1759 *
1760 * Further it needs to cope with the circular references problem, i.e. cannot
1761 * load any attributes other than $ATTRIBUTE_LIST until $DATA is loaded, because
1762 * we do not know where the other extent mft records are yet and again, because
1763 * we cannot call map_mft_record*() yet.  Obviously this applies only when an
1764 * attribute list is actually present in $MFT inode.
1765 *
1766 * We solve these problems by starting with the $DATA attribute before anything
1767 * else and iterating using ntfs_attr_lookup($DATA) over all extents.  As each
1768 * extent is found, we ntfs_mapping_pairs_decompress() including the implied
1769 * ntfs_runlists_merge().  Each step of the iteration necessarily provides
1770 * sufficient information for the next step to complete.
1771 *
1772 * This should work but there are two possible pit falls (see inline comments
1773 * below), but only time will tell if they are real pits or just smoke...
1774 */
1775int ntfs_read_inode_mount(struct inode *vi)
1776{
1777	VCN next_vcn, last_vcn, highest_vcn;
1778	s64 block;
1779	struct super_block *sb = vi->i_sb;
1780	ntfs_volume *vol = NTFS_SB(sb);
1781	struct buffer_head *bh;
1782	ntfs_inode *ni;
1783	MFT_RECORD *m = NULL;
1784	ATTR_RECORD *a;
1785	ntfs_attr_search_ctx *ctx;
1786	unsigned int i, nr_blocks;
1787	int err;
1788
1789	ntfs_debug("Entering.");
1790
1791	/* Initialize the ntfs specific part of @vi. */
1792	ntfs_init_big_inode(vi);
1793
1794	ni = NTFS_I(vi);
1795
1796	/* Setup the data attribute. It is special as it is mst protected. */
1797	NInoSetNonResident(ni);
1798	NInoSetMstProtected(ni);
1799	NInoSetSparseDisabled(ni);
1800	ni->type = AT_DATA;
1801	ni->name = NULL;
1802	ni->name_len = 0;
1803	/*
1804	 * This sets up our little cheat allowing us to reuse the async read io
1805	 * completion handler for directories.
1806	 */
1807	ni->itype.index.block_size = vol->mft_record_size;
1808	ni->itype.index.block_size_bits = vol->mft_record_size_bits;
1809
1810	/* Very important! Needed to be able to call map_mft_record*(). */
1811	vol->mft_ino = vi;
1812
1813	/* Allocate enough memory to read the first mft record. */
1814	if (vol->mft_record_size > 64 * 1024) {
1815		ntfs_error(sb, "Unsupported mft record size %i (max 64kiB).",
1816				vol->mft_record_size);
1817		goto err_out;
1818	}
1819	i = vol->mft_record_size;
1820	if (i < sb->s_blocksize)
1821		i = sb->s_blocksize;
1822	m = (MFT_RECORD*)ntfs_malloc_nofs(i);
1823	if (!m) {
1824		ntfs_error(sb, "Failed to allocate buffer for $MFT record 0.");
1825		goto err_out;
1826	}
1827
1828	/* Determine the first block of the $MFT/$DATA attribute. */
1829	block = vol->mft_lcn << vol->cluster_size_bits >>
1830			sb->s_blocksize_bits;
1831	nr_blocks = vol->mft_record_size >> sb->s_blocksize_bits;
1832	if (!nr_blocks)
1833		nr_blocks = 1;
1834
1835	/* Load $MFT/$DATA's first mft record. */
1836	for (i = 0; i < nr_blocks; i++) {
1837		bh = sb_bread(sb, block++);
1838		if (!bh) {
1839			ntfs_error(sb, "Device read failed.");
1840			goto err_out;
1841		}
1842		memcpy((char*)m + (i << sb->s_blocksize_bits), bh->b_data,
1843				sb->s_blocksize);
1844		brelse(bh);
1845	}
1846
1847	/* Apply the mst fixups. */
1848	if (post_read_mst_fixup((NTFS_RECORD*)m, vol->mft_record_size)) {
1849		/* FIXME: Try to use the $MFTMirr now. */
1850		ntfs_error(sb, "MST fixup failed. $MFT is corrupt.");
1851		goto err_out;
1852	}
1853
1854	/* Need this to sanity check attribute list references to $MFT. */
1855	vi->i_generation = ni->seq_no = le16_to_cpu(m->sequence_number);
1856
1857	/* Provides readpage() and sync_page() for map_mft_record(). */
1858	vi->i_mapping->a_ops = &ntfs_mst_aops;
1859
1860	ctx = ntfs_attr_get_search_ctx(ni, m);
1861	if (!ctx) {
1862		err = -ENOMEM;
1863		goto err_out;
1864	}
1865
1866	/* Find the attribute list attribute if present. */
1867	err = ntfs_attr_lookup(AT_ATTRIBUTE_LIST, NULL, 0, 0, 0, NULL, 0, ctx);
1868	if (err) {
1869		if (unlikely(err != -ENOENT)) {
1870			ntfs_error(sb, "Failed to lookup attribute list "
1871					"attribute. You should run chkdsk.");
1872			goto put_err_out;
1873		}
1874	} else /* if (!err) */ {
1875		ATTR_LIST_ENTRY *al_entry, *next_al_entry;
1876		u8 *al_end;
1877		static const char *es = "  Not allowed.  $MFT is corrupt.  "
1878				"You should run chkdsk.";
1879
1880		ntfs_debug("Attribute list attribute found in $MFT.");
1881		NInoSetAttrList(ni);
1882		a = ctx->attr;
1883		if (a->flags & ATTR_COMPRESSION_MASK) {
1884			ntfs_error(sb, "Attribute list attribute is "
1885					"compressed.%s", es);
1886			goto put_err_out;
1887		}
1888		if (a->flags & ATTR_IS_ENCRYPTED ||
1889				a->flags & ATTR_IS_SPARSE) {
1890			if (a->non_resident) {
1891				ntfs_error(sb, "Non-resident attribute list "
1892						"attribute is encrypted/"
1893						"sparse.%s", es);
1894				goto put_err_out;
1895			}
1896			ntfs_warning(sb, "Resident attribute list attribute "
1897					"in $MFT system file is marked "
1898					"encrypted/sparse which is not true.  "
1899					"However, Windows allows this and "
1900					"chkdsk does not detect or correct it "
1901					"so we will just ignore the invalid "
1902					"flags and pretend they are not set.");
1903		}
1904		/* Now allocate memory for the attribute list. */
1905		ni->attr_list_size = (u32)ntfs_attr_size(a);
1906		ni->attr_list = ntfs_malloc_nofs(ni->attr_list_size);
1907		if (!ni->attr_list) {
1908			ntfs_error(sb, "Not enough memory to allocate buffer "
1909					"for attribute list.");
1910			goto put_err_out;
1911		}
1912		if (a->non_resident) {
1913			NInoSetAttrListNonResident(ni);
1914			if (a->data.non_resident.lowest_vcn) {
1915				ntfs_error(sb, "Attribute list has non zero "
1916						"lowest_vcn. $MFT is corrupt. "
1917						"You should run chkdsk.");
1918				goto put_err_out;
1919			}
1920			/* Setup the runlist. */
1921			ni->attr_list_rl.rl = ntfs_mapping_pairs_decompress(vol,
1922					a, NULL);
1923			if (IS_ERR(ni->attr_list_rl.rl)) {
1924				err = PTR_ERR(ni->attr_list_rl.rl);
1925				ni->attr_list_rl.rl = NULL;
1926				ntfs_error(sb, "Mapping pairs decompression "
1927						"failed with error code %i.",
1928						-err);
1929				goto put_err_out;
1930			}
1931			/* Now load the attribute list. */
1932			if ((err = load_attribute_list(vol, &ni->attr_list_rl,
1933					ni->attr_list, ni->attr_list_size,
1934					sle64_to_cpu(a->data.
1935					non_resident.initialized_size)))) {
1936				ntfs_error(sb, "Failed to load attribute list "
1937						"attribute with error code %i.",
1938						-err);
1939				goto put_err_out;
1940			}
1941		} else /* if (!ctx.attr->non_resident) */ {
1942			if ((u8*)a + le16_to_cpu(
1943					a->data.resident.value_offset) +
1944					le32_to_cpu(
1945					a->data.resident.value_length) >
1946					(u8*)ctx->mrec + vol->mft_record_size) {
1947				ntfs_error(sb, "Corrupt attribute list "
1948						"attribute.");
1949				goto put_err_out;
1950			}
1951			/* Now copy the attribute list. */
1952			memcpy(ni->attr_list, (u8*)a + le16_to_cpu(
1953					a->data.resident.value_offset),
1954					le32_to_cpu(
1955					a->data.resident.value_length));
1956		}
1957		/* The attribute list is now setup in memory. */
1958		/*
1959		 * FIXME: I don't know if this case is actually possible.
1960		 * According to logic it is not possible but I have seen too
1961		 * many weird things in MS software to rely on logic... Thus we
1962		 * perform a manual search and make sure the first $MFT/$DATA
1963		 * extent is in the base inode. If it is not we abort with an
1964		 * error and if we ever see a report of this error we will need
1965		 * to do some magic in order to have the necessary mft record
1966		 * loaded and in the right place in the page cache. But
1967		 * hopefully logic will prevail and this never happens...
1968		 */
1969		al_entry = (ATTR_LIST_ENTRY*)ni->attr_list;
1970		al_end = (u8*)al_entry + ni->attr_list_size;
1971		for (;; al_entry = next_al_entry) {
1972			/* Out of bounds check. */
1973			if ((u8*)al_entry < ni->attr_list ||
1974					(u8*)al_entry > al_end)
1975				goto em_put_err_out;
1976			/* Catch the end of the attribute list. */
1977			if ((u8*)al_entry == al_end)
1978				goto em_put_err_out;
1979			if (!al_entry->length)
1980				goto em_put_err_out;
1981			if ((u8*)al_entry + 6 > al_end || (u8*)al_entry +
1982					le16_to_cpu(al_entry->length) > al_end)
1983				goto em_put_err_out;
1984			next_al_entry = (ATTR_LIST_ENTRY*)((u8*)al_entry +
1985					le16_to_cpu(al_entry->length));
1986			if (le32_to_cpu(al_entry->type) > le32_to_cpu(AT_DATA))
1987				goto em_put_err_out;
1988			if (AT_DATA != al_entry->type)
1989				continue;
1990			/* We want an unnamed attribute. */
1991			if (al_entry->name_length)
1992				goto em_put_err_out;
1993			/* Want the first entry, i.e. lowest_vcn == 0. */
1994			if (al_entry->lowest_vcn)
1995				goto em_put_err_out;
1996			/* First entry has to be in the base mft record. */
1997			if (MREF_LE(al_entry->mft_reference) != vi->i_ino) {
1998				/* MFT references do not match, logic fails. */
1999				ntfs_error(sb, "BUG: The first $DATA extent "
2000						"of $MFT is not in the base "
2001						"mft record. Please report "
2002						"you saw this message to "
2003						"linux-ntfs-dev@lists."
2004						"sourceforge.net");
2005				goto put_err_out;
2006			} else {
2007				/* Sequence numbers must match. */
2008				if (MSEQNO_LE(al_entry->mft_reference) !=
2009						ni->seq_no)
2010					goto em_put_err_out;
2011				/* Got it. All is ok. We can stop now. */
2012				break;
2013			}
2014		}
2015	}
2016
2017	ntfs_attr_reinit_search_ctx(ctx);
2018
2019	/* Now load all attribute extents. */
2020	a = NULL;
2021	next_vcn = last_vcn = highest_vcn = 0;
2022	while (!(err = ntfs_attr_lookup(AT_DATA, NULL, 0, 0, next_vcn, NULL, 0,
2023			ctx))) {
2024		runlist_element *nrl;
2025
2026		/* Cache the current attribute. */
2027		a = ctx->attr;
2028		/* $MFT must be non-resident. */
2029		if (!a->non_resident) {
2030			ntfs_error(sb, "$MFT must be non-resident but a "
2031					"resident extent was found. $MFT is "
2032					"corrupt. Run chkdsk.");
2033			goto put_err_out;
2034		}
2035		/* $MFT must be uncompressed and unencrypted. */
2036		if (a->flags & ATTR_COMPRESSION_MASK ||
2037				a->flags & ATTR_IS_ENCRYPTED ||
2038				a->flags & ATTR_IS_SPARSE) {
2039			ntfs_error(sb, "$MFT must be uncompressed, "
2040					"non-sparse, and unencrypted but a "
2041					"compressed/sparse/encrypted extent "
2042					"was found. $MFT is corrupt. Run "
2043					"chkdsk.");
2044			goto put_err_out;
2045		}
2046		/*
2047		 * Decompress the mapping pairs array of this extent and merge
2048		 * the result into the existing runlist. No need for locking
2049		 * as we have exclusive access to the inode at this time and we
2050		 * are a mount in progress task, too.
2051		 */
2052		nrl = ntfs_mapping_pairs_decompress(vol, a, ni->runlist.rl);
2053		if (IS_ERR(nrl)) {
2054			ntfs_error(sb, "ntfs_mapping_pairs_decompress() "
2055					"failed with error code %ld.  $MFT is "
2056					"corrupt.", PTR_ERR(nrl));
2057			goto put_err_out;
2058		}
2059		ni->runlist.rl = nrl;
2060
2061		/* Are we in the first extent? */
2062		if (!next_vcn) {
2063			if (a->data.non_resident.lowest_vcn) {
2064				ntfs_error(sb, "First extent of $DATA "
2065						"attribute has non zero "
2066						"lowest_vcn. $MFT is corrupt. "
2067						"You should run chkdsk.");
2068				goto put_err_out;
2069			}
2070			/* Get the last vcn in the $DATA attribute. */
2071			last_vcn = sle64_to_cpu(
2072					a->data.non_resident.allocated_size)
2073					>> vol->cluster_size_bits;
2074			/* Fill in the inode size. */
2075			vi->i_size = sle64_to_cpu(
2076					a->data.non_resident.data_size);
2077			ni->initialized_size = sle64_to_cpu(
2078					a->data.non_resident.initialized_size);
2079			ni->allocated_size = sle64_to_cpu(
2080					a->data.non_resident.allocated_size);
2081			/*
2082			 * Verify the number of mft records does not exceed
2083			 * 2^32 - 1.
2084			 */
2085			if ((vi->i_size >> vol->mft_record_size_bits) >=
2086					(1ULL << 32)) {
2087				ntfs_error(sb, "$MFT is too big! Aborting.");
2088				goto put_err_out;
2089			}
2090			/*
2091			 * We have got the first extent of the runlist for
2092			 * $MFT which means it is now relatively safe to call
2093			 * the normal ntfs_read_inode() function.
2094			 * Complete reading the inode, this will actually
2095			 * re-read the mft record for $MFT, this time entering
2096			 * it into the page cache with which we complete the
2097			 * kick start of the volume. It should be safe to do
2098			 * this now as the first extent of $MFT/$DATA is
2099			 * already known and we would hope that we don't need
2100			 * further extents in order to find the other
2101			 * attributes belonging to $MFT. Only time will tell if
2102			 * this is really the case. If not we will have to play
2103			 * magic at this point, possibly duplicating a lot of
2104			 * ntfs_read_inode() at this point. We will need to
2105			 * ensure we do enough of its work to be able to call
2106			 * ntfs_read_inode() on extents of $MFT/$DATA. But lets
2107			 * hope this never happens...
2108			 */
2109			ntfs_read_locked_inode(vi);
2110			if (is_bad_inode(vi)) {
2111				ntfs_error(sb, "ntfs_read_inode() of $MFT "
2112						"failed. BUG or corrupt $MFT. "
2113						"Run chkdsk and if no errors "
2114						"are found, please report you "
2115						"saw this message to "
2116						"linux-ntfs-dev@lists."
2117						"sourceforge.net");
2118				ntfs_attr_put_search_ctx(ctx);
2119				/* Revert to the safe super operations. */
2120				ntfs_free(m);
2121				return -1;
2122			}
2123			/*
2124			 * Re-initialize some specifics about $MFT's inode as
2125			 * ntfs_read_inode() will have set up the default ones.
2126			 */
2127			/* Set uid and gid to root. */
2128			vi->i_uid = GLOBAL_ROOT_UID;
2129			vi->i_gid = GLOBAL_ROOT_GID;
2130			/* Regular file. No access for anyone. */
2131			vi->i_mode = S_IFREG;
2132			/* No VFS initiated operations allowed for $MFT. */
2133			vi->i_op = &ntfs_empty_inode_ops;
2134			vi->i_fop = &ntfs_empty_file_ops;
2135		}
2136
2137		/* Get the lowest vcn for the next extent. */
2138		highest_vcn = sle64_to_cpu(a->data.non_resident.highest_vcn);
2139		next_vcn = highest_vcn + 1;
2140
2141		/* Only one extent or error, which we catch below. */
2142		if (next_vcn <= 0)
2143			break;
2144
2145		/* Avoid endless loops due to corruption. */
2146		if (next_vcn < sle64_to_cpu(
2147				a->data.non_resident.lowest_vcn)) {
2148			ntfs_error(sb, "$MFT has corrupt attribute list "
2149					"attribute. Run chkdsk.");
2150			goto put_err_out;
2151		}
2152	}
2153	if (err != -ENOENT) {
2154		ntfs_error(sb, "Failed to lookup $MFT/$DATA attribute extent. "
2155				"$MFT is corrupt. Run chkdsk.");
2156		goto put_err_out;
2157	}
2158	if (!a) {
2159		ntfs_error(sb, "$MFT/$DATA attribute not found. $MFT is "
2160				"corrupt. Run chkdsk.");
2161		goto put_err_out;
2162	}
2163	if (highest_vcn && highest_vcn != last_vcn - 1) {
2164		ntfs_error(sb, "Failed to load the complete runlist for "
2165				"$MFT/$DATA. Driver bug or corrupt $MFT. "
2166				"Run chkdsk.");
2167		ntfs_debug("highest_vcn = 0x%llx, last_vcn - 1 = 0x%llx",
2168				(unsigned long long)highest_vcn,
2169				(unsigned long long)last_vcn - 1);
2170		goto put_err_out;
2171	}
2172	ntfs_attr_put_search_ctx(ctx);
2173	ntfs_debug("Done.");
2174	ntfs_free(m);
2175
2176	/*
2177	 * Split the locking rules of the MFT inode from the
2178	 * locking rules of other inodes:
2179	 */
2180	lockdep_set_class(&ni->runlist.lock, &mft_ni_runlist_lock_key);
2181	lockdep_set_class(&ni->mrec_lock, &mft_ni_mrec_lock_key);
2182
2183	return 0;
2184
2185em_put_err_out:
2186	ntfs_error(sb, "Couldn't find first extent of $DATA attribute in "
2187			"attribute list. $MFT is corrupt. Run chkdsk.");
2188put_err_out:
2189	ntfs_attr_put_search_ctx(ctx);
2190err_out:
2191	ntfs_error(sb, "Failed. Marking inode as bad.");
2192	make_bad_inode(vi);
2193	ntfs_free(m);
2194	return -1;
2195}
2196
2197static void __ntfs_clear_inode(ntfs_inode *ni)
2198{
2199	/* Free all alocated memory. */
2200	down_write(&ni->runlist.lock);
2201	if (ni->runlist.rl) {
2202		ntfs_free(ni->runlist.rl);
2203		ni->runlist.rl = NULL;
2204	}
2205	up_write(&ni->runlist.lock);
2206
2207	if (ni->attr_list) {
2208		ntfs_free(ni->attr_list);
2209		ni->attr_list = NULL;
2210	}
2211
2212	down_write(&ni->attr_list_rl.lock);
2213	if (ni->attr_list_rl.rl) {
2214		ntfs_free(ni->attr_list_rl.rl);
2215		ni->attr_list_rl.rl = NULL;
2216	}
2217	up_write(&ni->attr_list_rl.lock);
2218
2219	if (ni->name_len && ni->name != I30) {
2220		/* Catch bugs... */
2221		BUG_ON(!ni->name);
2222		kfree(ni->name);
2223	}
2224}
2225
2226void ntfs_clear_extent_inode(ntfs_inode *ni)
2227{
2228	ntfs_debug("Entering for inode 0x%lx.", ni->mft_no);
2229
2230	BUG_ON(NInoAttr(ni));
2231	BUG_ON(ni->nr_extents != -1);
2232
2233#ifdef NTFS_RW
2234	if (NInoDirty(ni)) {
2235		if (!is_bad_inode(VFS_I(ni->ext.base_ntfs_ino)))
2236			ntfs_error(ni->vol->sb, "Clearing dirty extent inode!  "
2237					"Losing data!  This is a BUG!!!");
2238		// FIXME:  Do something!!!
2239	}
2240#endif /* NTFS_RW */
2241
2242	__ntfs_clear_inode(ni);
2243
2244	/* Bye, bye... */
2245	ntfs_destroy_extent_inode(ni);
2246}
2247
2248/**
2249 * ntfs_evict_big_inode - clean up the ntfs specific part of an inode
2250 * @vi:		vfs inode pending annihilation
2251 *
2252 * When the VFS is going to remove an inode from memory, ntfs_clear_big_inode()
2253 * is called, which deallocates all memory belonging to the NTFS specific part
2254 * of the inode and returns.
2255 *
2256 * If the MFT record is dirty, we commit it before doing anything else.
2257 */
2258void ntfs_evict_big_inode(struct inode *vi)
2259{
2260	ntfs_inode *ni = NTFS_I(vi);
2261
2262	truncate_inode_pages_final(&vi->i_data);
2263	clear_inode(vi);
2264
2265#ifdef NTFS_RW
2266	if (NInoDirty(ni)) {
2267		bool was_bad = (is_bad_inode(vi));
2268
2269		/* Committing the inode also commits all extent inodes. */
2270		ntfs_commit_inode(vi);
2271
2272		if (!was_bad && (is_bad_inode(vi) || NInoDirty(ni))) {
2273			ntfs_error(vi->i_sb, "Failed to commit dirty inode "
2274					"0x%lx.  Losing data!", vi->i_ino);
2275			// FIXME:  Do something!!!
2276		}
2277	}
2278#endif /* NTFS_RW */
2279
2280	/* No need to lock at this stage as no one else has a reference. */
2281	if (ni->nr_extents > 0) {
2282		int i;
2283
2284		for (i = 0; i < ni->nr_extents; i++)
2285			ntfs_clear_extent_inode(ni->ext.extent_ntfs_inos[i]);
2286		kfree(ni->ext.extent_ntfs_inos);
2287	}
2288
2289	__ntfs_clear_inode(ni);
2290
2291	if (NInoAttr(ni)) {
2292		/* Release the base inode if we are holding it. */
2293		if (ni->nr_extents == -1) {
2294			iput(VFS_I(ni->ext.base_ntfs_ino));
2295			ni->nr_extents = 0;
2296			ni->ext.base_ntfs_ino = NULL;
2297		}
2298	}
 
 
 
2299	return;
2300}
2301
2302/**
2303 * ntfs_show_options - show mount options in /proc/mounts
2304 * @sf:		seq_file in which to write our mount options
2305 * @root:	root of the mounted tree whose mount options to display
2306 *
2307 * Called by the VFS once for each mounted ntfs volume when someone reads
2308 * /proc/mounts in order to display the NTFS specific mount options of each
2309 * mount. The mount options of fs specified by @root are written to the seq file
2310 * @sf and success is returned.
2311 */
2312int ntfs_show_options(struct seq_file *sf, struct dentry *root)
2313{
2314	ntfs_volume *vol = NTFS_SB(root->d_sb);
2315	int i;
2316
2317	seq_printf(sf, ",uid=%i", from_kuid_munged(&init_user_ns, vol->uid));
2318	seq_printf(sf, ",gid=%i", from_kgid_munged(&init_user_ns, vol->gid));
2319	if (vol->fmask == vol->dmask)
2320		seq_printf(sf, ",umask=0%o", vol->fmask);
2321	else {
2322		seq_printf(sf, ",fmask=0%o", vol->fmask);
2323		seq_printf(sf, ",dmask=0%o", vol->dmask);
2324	}
2325	seq_printf(sf, ",nls=%s", vol->nls_map->charset);
2326	if (NVolCaseSensitive(vol))
2327		seq_printf(sf, ",case_sensitive");
2328	if (NVolShowSystemFiles(vol))
2329		seq_printf(sf, ",show_sys_files");
2330	if (!NVolSparseEnabled(vol))
2331		seq_printf(sf, ",disable_sparse");
2332	for (i = 0; on_errors_arr[i].val; i++) {
2333		if (on_errors_arr[i].val & vol->on_errors)
2334			seq_printf(sf, ",errors=%s", on_errors_arr[i].str);
2335	}
2336	seq_printf(sf, ",mft_zone_multiplier=%i", vol->mft_zone_multiplier);
2337	return 0;
2338}
2339
2340#ifdef NTFS_RW
2341
2342static const char *es = "  Leaving inconsistent metadata.  Unmount and run "
2343		"chkdsk.";
2344
2345/**
2346 * ntfs_truncate - called when the i_size of an ntfs inode is changed
2347 * @vi:		inode for which the i_size was changed
2348 *
2349 * We only support i_size changes for normal files at present, i.e. not
2350 * compressed and not encrypted.  This is enforced in ntfs_setattr(), see
2351 * below.
2352 *
2353 * The kernel guarantees that @vi is a regular file (S_ISREG() is true) and
2354 * that the change is allowed.
2355 *
2356 * This implies for us that @vi is a file inode rather than a directory, index,
2357 * or attribute inode as well as that @vi is a base inode.
2358 *
2359 * Returns 0 on success or -errno on error.
2360 *
2361 * Called with ->i_mutex held.
2362 */
2363int ntfs_truncate(struct inode *vi)
2364{
2365	s64 new_size, old_size, nr_freed, new_alloc_size, old_alloc_size;
2366	VCN highest_vcn;
2367	unsigned long flags;
2368	ntfs_inode *base_ni, *ni = NTFS_I(vi);
2369	ntfs_volume *vol = ni->vol;
2370	ntfs_attr_search_ctx *ctx;
2371	MFT_RECORD *m;
2372	ATTR_RECORD *a;
2373	const char *te = "  Leaving file length out of sync with i_size.";
2374	int err, mp_size, size_change, alloc_change;
2375	u32 attr_len;
2376
2377	ntfs_debug("Entering for inode 0x%lx.", vi->i_ino);
2378	BUG_ON(NInoAttr(ni));
2379	BUG_ON(S_ISDIR(vi->i_mode));
2380	BUG_ON(NInoMstProtected(ni));
2381	BUG_ON(ni->nr_extents < 0);
2382retry_truncate:
2383	/*
2384	 * Lock the runlist for writing and map the mft record to ensure it is
2385	 * safe to mess with the attribute runlist and sizes.
2386	 */
2387	down_write(&ni->runlist.lock);
2388	if (!NInoAttr(ni))
2389		base_ni = ni;
2390	else
2391		base_ni = ni->ext.base_ntfs_ino;
2392	m = map_mft_record(base_ni);
2393	if (IS_ERR(m)) {
2394		err = PTR_ERR(m);
2395		ntfs_error(vi->i_sb, "Failed to map mft record for inode 0x%lx "
2396				"(error code %d).%s", vi->i_ino, err, te);
2397		ctx = NULL;
2398		m = NULL;
2399		goto old_bad_out;
2400	}
2401	ctx = ntfs_attr_get_search_ctx(base_ni, m);
2402	if (unlikely(!ctx)) {
2403		ntfs_error(vi->i_sb, "Failed to allocate a search context for "
2404				"inode 0x%lx (not enough memory).%s",
2405				vi->i_ino, te);
2406		err = -ENOMEM;
2407		goto old_bad_out;
2408	}
2409	err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len,
2410			CASE_SENSITIVE, 0, NULL, 0, ctx);
2411	if (unlikely(err)) {
2412		if (err == -ENOENT) {
2413			ntfs_error(vi->i_sb, "Open attribute is missing from "
2414					"mft record.  Inode 0x%lx is corrupt.  "
2415					"Run chkdsk.%s", vi->i_ino, te);
2416			err = -EIO;
2417		} else
2418			ntfs_error(vi->i_sb, "Failed to lookup attribute in "
2419					"inode 0x%lx (error code %d).%s",
2420					vi->i_ino, err, te);
2421		goto old_bad_out;
2422	}
2423	m = ctx->mrec;
2424	a = ctx->attr;
2425	/*
2426	 * The i_size of the vfs inode is the new size for the attribute value.
2427	 */
2428	new_size = i_size_read(vi);
2429	/* The current size of the attribute value is the old size. */
2430	old_size = ntfs_attr_size(a);
2431	/* Calculate the new allocated size. */
2432	if (NInoNonResident(ni))
2433		new_alloc_size = (new_size + vol->cluster_size - 1) &
2434				~(s64)vol->cluster_size_mask;
2435	else
2436		new_alloc_size = (new_size + 7) & ~7;
2437	/* The current allocated size is the old allocated size. */
2438	read_lock_irqsave(&ni->size_lock, flags);
2439	old_alloc_size = ni->allocated_size;
2440	read_unlock_irqrestore(&ni->size_lock, flags);
2441	/*
2442	 * The change in the file size.  This will be 0 if no change, >0 if the
2443	 * size is growing, and <0 if the size is shrinking.
2444	 */
2445	size_change = -1;
2446	if (new_size - old_size >= 0) {
2447		size_change = 1;
2448		if (new_size == old_size)
2449			size_change = 0;
2450	}
2451	/* As above for the allocated size. */
2452	alloc_change = -1;
2453	if (new_alloc_size - old_alloc_size >= 0) {
2454		alloc_change = 1;
2455		if (new_alloc_size == old_alloc_size)
2456			alloc_change = 0;
2457	}
2458	/*
2459	 * If neither the size nor the allocation are being changed there is
2460	 * nothing to do.
2461	 */
2462	if (!size_change && !alloc_change)
2463		goto unm_done;
2464	/* If the size is changing, check if new size is allowed in $AttrDef. */
2465	if (size_change) {
2466		err = ntfs_attr_size_bounds_check(vol, ni->type, new_size);
2467		if (unlikely(err)) {
2468			if (err == -ERANGE) {
2469				ntfs_error(vol->sb, "Truncate would cause the "
2470						"inode 0x%lx to %simum size "
2471						"for its attribute type "
2472						"(0x%x).  Aborting truncate.",
2473						vi->i_ino,
2474						new_size > old_size ? "exceed "
2475						"the max" : "go under the min",
2476						le32_to_cpu(ni->type));
2477				err = -EFBIG;
2478			} else {
2479				ntfs_error(vol->sb, "Inode 0x%lx has unknown "
2480						"attribute type 0x%x.  "
2481						"Aborting truncate.",
2482						vi->i_ino,
2483						le32_to_cpu(ni->type));
2484				err = -EIO;
2485			}
2486			/* Reset the vfs inode size to the old size. */
2487			i_size_write(vi, old_size);
2488			goto err_out;
2489		}
2490	}
2491	if (NInoCompressed(ni) || NInoEncrypted(ni)) {
2492		ntfs_warning(vi->i_sb, "Changes in inode size are not "
2493				"supported yet for %s files, ignoring.",
2494				NInoCompressed(ni) ? "compressed" :
2495				"encrypted");
2496		err = -EOPNOTSUPP;
2497		goto bad_out;
2498	}
2499	if (a->non_resident)
2500		goto do_non_resident_truncate;
2501	BUG_ON(NInoNonResident(ni));
2502	/* Resize the attribute record to best fit the new attribute size. */
2503	if (new_size < vol->mft_record_size &&
2504			!ntfs_resident_attr_value_resize(m, a, new_size)) {
2505		/* The resize succeeded! */
2506		flush_dcache_mft_record_page(ctx->ntfs_ino);
2507		mark_mft_record_dirty(ctx->ntfs_ino);
2508		write_lock_irqsave(&ni->size_lock, flags);
2509		/* Update the sizes in the ntfs inode and all is done. */
2510		ni->allocated_size = le32_to_cpu(a->length) -
2511				le16_to_cpu(a->data.resident.value_offset);
2512		/*
2513		 * Note ntfs_resident_attr_value_resize() has already done any
2514		 * necessary data clearing in the attribute record.  When the
2515		 * file is being shrunk vmtruncate() will already have cleared
2516		 * the top part of the last partial page, i.e. since this is
2517		 * the resident case this is the page with index 0.  However,
2518		 * when the file is being expanded, the page cache page data
2519		 * between the old data_size, i.e. old_size, and the new_size
2520		 * has not been zeroed.  Fortunately, we do not need to zero it
2521		 * either since on one hand it will either already be zero due
2522		 * to both readpage and writepage clearing partial page data
2523		 * beyond i_size in which case there is nothing to do or in the
2524		 * case of the file being mmap()ped at the same time, POSIX
2525		 * specifies that the behaviour is unspecified thus we do not
2526		 * have to do anything.  This means that in our implementation
2527		 * in the rare case that the file is mmap()ped and a write
2528		 * occurred into the mmap()ped region just beyond the file size
2529		 * and writepage has not yet been called to write out the page
2530		 * (which would clear the area beyond the file size) and we now
2531		 * extend the file size to incorporate this dirty region
2532		 * outside the file size, a write of the page would result in
2533		 * this data being written to disk instead of being cleared.
2534		 * Given both POSIX and the Linux mmap(2) man page specify that
2535		 * this corner case is undefined, we choose to leave it like
2536		 * that as this is much simpler for us as we cannot lock the
2537		 * relevant page now since we are holding too many ntfs locks
2538		 * which would result in a lock reversal deadlock.
2539		 */
2540		ni->initialized_size = new_size;
2541		write_unlock_irqrestore(&ni->size_lock, flags);
2542		goto unm_done;
2543	}
2544	/* If the above resize failed, this must be an attribute extension. */
2545	BUG_ON(size_change < 0);
2546	/*
2547	 * We have to drop all the locks so we can call
2548	 * ntfs_attr_make_non_resident().  This could be optimised by try-
2549	 * locking the first page cache page and only if that fails dropping
2550	 * the locks, locking the page, and redoing all the locking and
2551	 * lookups.  While this would be a huge optimisation, it is not worth
2552	 * it as this is definitely a slow code path as it only ever can happen
2553	 * once for any given file.
2554	 */
2555	ntfs_attr_put_search_ctx(ctx);
2556	unmap_mft_record(base_ni);
2557	up_write(&ni->runlist.lock);
2558	/*
2559	 * Not enough space in the mft record, try to make the attribute
2560	 * non-resident and if successful restart the truncation process.
2561	 */
2562	err = ntfs_attr_make_non_resident(ni, old_size);
2563	if (likely(!err))
2564		goto retry_truncate;
2565	/*
2566	 * Could not make non-resident.  If this is due to this not being
2567	 * permitted for this attribute type or there not being enough space,
2568	 * try to make other attributes non-resident.  Otherwise fail.
2569	 */
2570	if (unlikely(err != -EPERM && err != -ENOSPC)) {
2571		ntfs_error(vol->sb, "Cannot truncate inode 0x%lx, attribute "
2572				"type 0x%x, because the conversion from "
2573				"resident to non-resident attribute failed "
2574				"with error code %i.", vi->i_ino,
2575				(unsigned)le32_to_cpu(ni->type), err);
2576		if (err != -ENOMEM)
2577			err = -EIO;
2578		goto conv_err_out;
2579	}
2580	/* TODO: Not implemented from here, abort. */
2581	if (err == -ENOSPC)
2582		ntfs_error(vol->sb, "Not enough space in the mft record/on "
2583				"disk for the non-resident attribute value.  "
2584				"This case is not implemented yet.");
2585	else /* if (err == -EPERM) */
2586		ntfs_error(vol->sb, "This attribute type may not be "
2587				"non-resident.  This case is not implemented "
2588				"yet.");
2589	err = -EOPNOTSUPP;
2590	goto conv_err_out;
2591#if 0
2592	// TODO: Attempt to make other attributes non-resident.
2593	if (!err)
2594		goto do_resident_extend;
2595	/*
2596	 * Both the attribute list attribute and the standard information
2597	 * attribute must remain in the base inode.  Thus, if this is one of
2598	 * these attributes, we have to try to move other attributes out into
2599	 * extent mft records instead.
2600	 */
2601	if (ni->type == AT_ATTRIBUTE_LIST ||
2602			ni->type == AT_STANDARD_INFORMATION) {
2603		// TODO: Attempt to move other attributes into extent mft
2604		// records.
2605		err = -EOPNOTSUPP;
2606		if (!err)
2607			goto do_resident_extend;
2608		goto err_out;
2609	}
2610	// TODO: Attempt to move this attribute to an extent mft record, but
2611	// only if it is not already the only attribute in an mft record in
2612	// which case there would be nothing to gain.
2613	err = -EOPNOTSUPP;
2614	if (!err)
2615		goto do_resident_extend;
2616	/* There is nothing we can do to make enough space. )-: */
2617	goto err_out;
2618#endif
2619do_non_resident_truncate:
2620	BUG_ON(!NInoNonResident(ni));
2621	if (alloc_change < 0) {
2622		highest_vcn = sle64_to_cpu(a->data.non_resident.highest_vcn);
2623		if (highest_vcn > 0 &&
2624				old_alloc_size >> vol->cluster_size_bits >
2625				highest_vcn + 1) {
2626			/*
2627			 * This attribute has multiple extents.  Not yet
2628			 * supported.
2629			 */
2630			ntfs_error(vol->sb, "Cannot truncate inode 0x%lx, "
2631					"attribute type 0x%x, because the "
2632					"attribute is highly fragmented (it "
2633					"consists of multiple extents) and "
2634					"this case is not implemented yet.",
2635					vi->i_ino,
2636					(unsigned)le32_to_cpu(ni->type));
2637			err = -EOPNOTSUPP;
2638			goto bad_out;
2639		}
2640	}
2641	/*
2642	 * If the size is shrinking, need to reduce the initialized_size and
2643	 * the data_size before reducing the allocation.
2644	 */
2645	if (size_change < 0) {
2646		/*
2647		 * Make the valid size smaller (i_size is already up-to-date).
2648		 */
2649		write_lock_irqsave(&ni->size_lock, flags);
2650		if (new_size < ni->initialized_size) {
2651			ni->initialized_size = new_size;
2652			a->data.non_resident.initialized_size =
2653					cpu_to_sle64(new_size);
2654		}
2655		a->data.non_resident.data_size = cpu_to_sle64(new_size);
2656		write_unlock_irqrestore(&ni->size_lock, flags);
2657		flush_dcache_mft_record_page(ctx->ntfs_ino);
2658		mark_mft_record_dirty(ctx->ntfs_ino);
2659		/* If the allocated size is not changing, we are done. */
2660		if (!alloc_change)
2661			goto unm_done;
2662		/*
2663		 * If the size is shrinking it makes no sense for the
2664		 * allocation to be growing.
2665		 */
2666		BUG_ON(alloc_change > 0);
2667	} else /* if (size_change >= 0) */ {
2668		/*
2669		 * The file size is growing or staying the same but the
2670		 * allocation can be shrinking, growing or staying the same.
2671		 */
2672		if (alloc_change > 0) {
2673			/*
2674			 * We need to extend the allocation and possibly update
2675			 * the data size.  If we are updating the data size,
2676			 * since we are not touching the initialized_size we do
2677			 * not need to worry about the actual data on disk.
2678			 * And as far as the page cache is concerned, there
2679			 * will be no pages beyond the old data size and any
2680			 * partial region in the last page between the old and
2681			 * new data size (or the end of the page if the new
2682			 * data size is outside the page) does not need to be
2683			 * modified as explained above for the resident
2684			 * attribute truncate case.  To do this, we simply drop
2685			 * the locks we hold and leave all the work to our
2686			 * friendly helper ntfs_attr_extend_allocation().
2687			 */
2688			ntfs_attr_put_search_ctx(ctx);
2689			unmap_mft_record(base_ni);
2690			up_write(&ni->runlist.lock);
2691			err = ntfs_attr_extend_allocation(ni, new_size,
2692					size_change > 0 ? new_size : -1, -1);
2693			/*
2694			 * ntfs_attr_extend_allocation() will have done error
2695			 * output already.
2696			 */
2697			goto done;
2698		}
2699		if (!alloc_change)
2700			goto alloc_done;
2701	}
2702	/* alloc_change < 0 */
2703	/* Free the clusters. */
2704	nr_freed = ntfs_cluster_free(ni, new_alloc_size >>
2705			vol->cluster_size_bits, -1, ctx);
2706	m = ctx->mrec;
2707	a = ctx->attr;
2708	if (unlikely(nr_freed < 0)) {
2709		ntfs_error(vol->sb, "Failed to release cluster(s) (error code "
2710				"%lli).  Unmount and run chkdsk to recover "
2711				"the lost cluster(s).", (long long)nr_freed);
2712		NVolSetErrors(vol);
2713		nr_freed = 0;
2714	}
2715	/* Truncate the runlist. */
2716	err = ntfs_rl_truncate_nolock(vol, &ni->runlist,
2717			new_alloc_size >> vol->cluster_size_bits);
2718	/*
2719	 * If the runlist truncation failed and/or the search context is no
2720	 * longer valid, we cannot resize the attribute record or build the
2721	 * mapping pairs array thus we mark the inode bad so that no access to
2722	 * the freed clusters can happen.
2723	 */
2724	if (unlikely(err || IS_ERR(m))) {
2725		ntfs_error(vol->sb, "Failed to %s (error code %li).%s",
2726				IS_ERR(m) ?
2727				"restore attribute search context" :
2728				"truncate attribute runlist",
2729				IS_ERR(m) ? PTR_ERR(m) : err, es);
2730		err = -EIO;
2731		goto bad_out;
2732	}
2733	/* Get the size for the shrunk mapping pairs array for the runlist. */
2734	mp_size = ntfs_get_size_for_mapping_pairs(vol, ni->runlist.rl, 0, -1);
2735	if (unlikely(mp_size <= 0)) {
2736		ntfs_error(vol->sb, "Cannot shrink allocation of inode 0x%lx, "
2737				"attribute type 0x%x, because determining the "
2738				"size for the mapping pairs failed with error "
2739				"code %i.%s", vi->i_ino,
2740				(unsigned)le32_to_cpu(ni->type), mp_size, es);
2741		err = -EIO;
2742		goto bad_out;
2743	}
2744	/*
2745	 * Shrink the attribute record for the new mapping pairs array.  Note,
2746	 * this cannot fail since we are making the attribute smaller thus by
2747	 * definition there is enough space to do so.
2748	 */
2749	attr_len = le32_to_cpu(a->length);
2750	err = ntfs_attr_record_resize(m, a, mp_size +
2751			le16_to_cpu(a->data.non_resident.mapping_pairs_offset));
2752	BUG_ON(err);
2753	/*
2754	 * Generate the mapping pairs array directly into the attribute record.
2755	 */
2756	err = ntfs_mapping_pairs_build(vol, (u8*)a +
2757			le16_to_cpu(a->data.non_resident.mapping_pairs_offset),
2758			mp_size, ni->runlist.rl, 0, -1, NULL);
2759	if (unlikely(err)) {
2760		ntfs_error(vol->sb, "Cannot shrink allocation of inode 0x%lx, "
2761				"attribute type 0x%x, because building the "
2762				"mapping pairs failed with error code %i.%s",
2763				vi->i_ino, (unsigned)le32_to_cpu(ni->type),
2764				err, es);
2765		err = -EIO;
2766		goto bad_out;
2767	}
2768	/* Update the allocated/compressed size as well as the highest vcn. */
2769	a->data.non_resident.highest_vcn = cpu_to_sle64((new_alloc_size >>
2770			vol->cluster_size_bits) - 1);
2771	write_lock_irqsave(&ni->size_lock, flags);
2772	ni->allocated_size = new_alloc_size;
2773	a->data.non_resident.allocated_size = cpu_to_sle64(new_alloc_size);
2774	if (NInoSparse(ni) || NInoCompressed(ni)) {
2775		if (nr_freed) {
2776			ni->itype.compressed.size -= nr_freed <<
2777					vol->cluster_size_bits;
2778			BUG_ON(ni->itype.compressed.size < 0);
2779			a->data.non_resident.compressed_size = cpu_to_sle64(
2780					ni->itype.compressed.size);
2781			vi->i_blocks = ni->itype.compressed.size >> 9;
2782		}
2783	} else
2784		vi->i_blocks = new_alloc_size >> 9;
2785	write_unlock_irqrestore(&ni->size_lock, flags);
2786	/*
2787	 * We have shrunk the allocation.  If this is a shrinking truncate we
2788	 * have already dealt with the initialized_size and the data_size above
2789	 * and we are done.  If the truncate is only changing the allocation
2790	 * and not the data_size, we are also done.  If this is an extending
2791	 * truncate, need to extend the data_size now which is ensured by the
2792	 * fact that @size_change is positive.
2793	 */
2794alloc_done:
2795	/*
2796	 * If the size is growing, need to update it now.  If it is shrinking,
2797	 * we have already updated it above (before the allocation change).
2798	 */
2799	if (size_change > 0)
2800		a->data.non_resident.data_size = cpu_to_sle64(new_size);
2801	/* Ensure the modified mft record is written out. */
2802	flush_dcache_mft_record_page(ctx->ntfs_ino);
2803	mark_mft_record_dirty(ctx->ntfs_ino);
2804unm_done:
2805	ntfs_attr_put_search_ctx(ctx);
2806	unmap_mft_record(base_ni);
2807	up_write(&ni->runlist.lock);
2808done:
2809	/* Update the mtime and ctime on the base inode. */
2810	/* normally ->truncate shouldn't update ctime or mtime,
2811	 * but ntfs did before so it got a copy & paste version
2812	 * of file_update_time.  one day someone should fix this
2813	 * for real.
2814	 */
2815	if (!IS_NOCMTIME(VFS_I(base_ni)) && !IS_RDONLY(VFS_I(base_ni))) {
2816		struct timespec now = current_fs_time(VFS_I(base_ni)->i_sb);
2817		int sync_it = 0;
2818
2819		if (!timespec_equal(&VFS_I(base_ni)->i_mtime, &now) ||
2820		    !timespec_equal(&VFS_I(base_ni)->i_ctime, &now))
2821			sync_it = 1;
2822		VFS_I(base_ni)->i_mtime = now;
2823		VFS_I(base_ni)->i_ctime = now;
2824
2825		if (sync_it)
2826			mark_inode_dirty_sync(VFS_I(base_ni));
2827	}
2828
2829	if (likely(!err)) {
2830		NInoClearTruncateFailed(ni);
2831		ntfs_debug("Done.");
2832	}
2833	return err;
2834old_bad_out:
2835	old_size = -1;
2836bad_out:
2837	if (err != -ENOMEM && err != -EOPNOTSUPP)
2838		NVolSetErrors(vol);
2839	if (err != -EOPNOTSUPP)
2840		NInoSetTruncateFailed(ni);
2841	else if (old_size >= 0)
2842		i_size_write(vi, old_size);
2843err_out:
2844	if (ctx)
2845		ntfs_attr_put_search_ctx(ctx);
2846	if (m)
2847		unmap_mft_record(base_ni);
2848	up_write(&ni->runlist.lock);
2849out:
2850	ntfs_debug("Failed.  Returning error code %i.", err);
2851	return err;
2852conv_err_out:
2853	if (err != -ENOMEM && err != -EOPNOTSUPP)
2854		NVolSetErrors(vol);
2855	if (err != -EOPNOTSUPP)
2856		NInoSetTruncateFailed(ni);
2857	else
2858		i_size_write(vi, old_size);
2859	goto out;
2860}
2861
2862/**
2863 * ntfs_truncate_vfs - wrapper for ntfs_truncate() that has no return value
2864 * @vi:		inode for which the i_size was changed
2865 *
2866 * Wrapper for ntfs_truncate() that has no return value.
2867 *
2868 * See ntfs_truncate() description above for details.
2869 */
2870#ifdef NTFS_RW
2871void ntfs_truncate_vfs(struct inode *vi) {
2872	ntfs_truncate(vi);
2873}
2874#endif
2875
2876/**
2877 * ntfs_setattr - called from notify_change() when an attribute is being changed
2878 * @dentry:	dentry whose attributes to change
2879 * @attr:	structure describing the attributes and the changes
2880 *
2881 * We have to trap VFS attempts to truncate the file described by @dentry as
2882 * soon as possible, because we do not implement changes in i_size yet.  So we
2883 * abort all i_size changes here.
2884 *
2885 * We also abort all changes of user, group, and mode as we do not implement
2886 * the NTFS ACLs yet.
2887 *
2888 * Called with ->i_mutex held.
2889 */
2890int ntfs_setattr(struct dentry *dentry, struct iattr *attr)
2891{
2892	struct inode *vi = d_inode(dentry);
2893	int err;
2894	unsigned int ia_valid = attr->ia_valid;
2895
2896	err = inode_change_ok(vi, attr);
2897	if (err)
2898		goto out;
2899	/* We do not support NTFS ACLs yet. */
2900	if (ia_valid & (ATTR_UID | ATTR_GID | ATTR_MODE)) {
2901		ntfs_warning(vi->i_sb, "Changes in user/group/mode are not "
2902				"supported yet, ignoring.");
2903		err = -EOPNOTSUPP;
2904		goto out;
2905	}
2906	if (ia_valid & ATTR_SIZE) {
2907		if (attr->ia_size != i_size_read(vi)) {
2908			ntfs_inode *ni = NTFS_I(vi);
2909			/*
2910			 * FIXME: For now we do not support resizing of
2911			 * compressed or encrypted files yet.
2912			 */
2913			if (NInoCompressed(ni) || NInoEncrypted(ni)) {
2914				ntfs_warning(vi->i_sb, "Changes in inode size "
2915						"are not supported yet for "
2916						"%s files, ignoring.",
2917						NInoCompressed(ni) ?
2918						"compressed" : "encrypted");
2919				err = -EOPNOTSUPP;
2920			} else {
2921				truncate_setsize(vi, attr->ia_size);
2922				ntfs_truncate_vfs(vi);
2923			}
2924			if (err || ia_valid == ATTR_SIZE)
2925				goto out;
2926		} else {
2927			/*
2928			 * We skipped the truncate but must still update
2929			 * timestamps.
2930			 */
2931			ia_valid |= ATTR_MTIME | ATTR_CTIME;
2932		}
2933	}
2934	if (ia_valid & ATTR_ATIME)
2935		vi->i_atime = timespec_trunc(attr->ia_atime,
2936				vi->i_sb->s_time_gran);
2937	if (ia_valid & ATTR_MTIME)
2938		vi->i_mtime = timespec_trunc(attr->ia_mtime,
2939				vi->i_sb->s_time_gran);
2940	if (ia_valid & ATTR_CTIME)
2941		vi->i_ctime = timespec_trunc(attr->ia_ctime,
2942				vi->i_sb->s_time_gran);
2943	mark_inode_dirty(vi);
2944out:
2945	return err;
2946}
2947
2948/**
2949 * ntfs_write_inode - write out a dirty inode
2950 * @vi:		inode to write out
2951 * @sync:	if true, write out synchronously
2952 *
2953 * Write out a dirty inode to disk including any extent inodes if present.
2954 *
2955 * If @sync is true, commit the inode to disk and wait for io completion.  This
2956 * is done using write_mft_record().
2957 *
2958 * If @sync is false, just schedule the write to happen but do not wait for i/o
2959 * completion.  In 2.6 kernels, scheduling usually happens just by virtue of
2960 * marking the page (and in this case mft record) dirty but we do not implement
2961 * this yet as write_mft_record() largely ignores the @sync parameter and
2962 * always performs synchronous writes.
2963 *
2964 * Return 0 on success and -errno on error.
2965 */
2966int __ntfs_write_inode(struct inode *vi, int sync)
2967{
2968	sle64 nt;
2969	ntfs_inode *ni = NTFS_I(vi);
2970	ntfs_attr_search_ctx *ctx;
2971	MFT_RECORD *m;
2972	STANDARD_INFORMATION *si;
2973	int err = 0;
2974	bool modified = false;
2975
2976	ntfs_debug("Entering for %sinode 0x%lx.", NInoAttr(ni) ? "attr " : "",
2977			vi->i_ino);
2978	/*
2979	 * Dirty attribute inodes are written via their real inodes so just
2980	 * clean them here.  Access time updates are taken care off when the
2981	 * real inode is written.
2982	 */
2983	if (NInoAttr(ni)) {
2984		NInoClearDirty(ni);
2985		ntfs_debug("Done.");
2986		return 0;
2987	}
2988	/* Map, pin, and lock the mft record belonging to the inode. */
2989	m = map_mft_record(ni);
2990	if (IS_ERR(m)) {
2991		err = PTR_ERR(m);
2992		goto err_out;
2993	}
2994	/* Update the access times in the standard information attribute. */
2995	ctx = ntfs_attr_get_search_ctx(ni, m);
2996	if (unlikely(!ctx)) {
2997		err = -ENOMEM;
2998		goto unm_err_out;
2999	}
3000	err = ntfs_attr_lookup(AT_STANDARD_INFORMATION, NULL, 0,
3001			CASE_SENSITIVE, 0, NULL, 0, ctx);
3002	if (unlikely(err)) {
3003		ntfs_attr_put_search_ctx(ctx);
3004		goto unm_err_out;
3005	}
3006	si = (STANDARD_INFORMATION*)((u8*)ctx->attr +
3007			le16_to_cpu(ctx->attr->data.resident.value_offset));
3008	/* Update the access times if they have changed. */
3009	nt = utc2ntfs(vi->i_mtime);
3010	if (si->last_data_change_time != nt) {
3011		ntfs_debug("Updating mtime for inode 0x%lx: old = 0x%llx, "
3012				"new = 0x%llx", vi->i_ino, (long long)
3013				sle64_to_cpu(si->last_data_change_time),
3014				(long long)sle64_to_cpu(nt));
3015		si->last_data_change_time = nt;
3016		modified = true;
3017	}
3018	nt = utc2ntfs(vi->i_ctime);
3019	if (si->last_mft_change_time != nt) {
3020		ntfs_debug("Updating ctime for inode 0x%lx: old = 0x%llx, "
3021				"new = 0x%llx", vi->i_ino, (long long)
3022				sle64_to_cpu(si->last_mft_change_time),
3023				(long long)sle64_to_cpu(nt));
3024		si->last_mft_change_time = nt;
3025		modified = true;
3026	}
3027	nt = utc2ntfs(vi->i_atime);
3028	if (si->last_access_time != nt) {
3029		ntfs_debug("Updating atime for inode 0x%lx: old = 0x%llx, "
3030				"new = 0x%llx", vi->i_ino,
3031				(long long)sle64_to_cpu(si->last_access_time),
3032				(long long)sle64_to_cpu(nt));
3033		si->last_access_time = nt;
3034		modified = true;
3035	}
3036	/*
3037	 * If we just modified the standard information attribute we need to
3038	 * mark the mft record it is in dirty.  We do this manually so that
3039	 * mark_inode_dirty() is not called which would redirty the inode and
3040	 * hence result in an infinite loop of trying to write the inode.
3041	 * There is no need to mark the base inode nor the base mft record
3042	 * dirty, since we are going to write this mft record below in any case
3043	 * and the base mft record may actually not have been modified so it
3044	 * might not need to be written out.
3045	 * NOTE: It is not a problem when the inode for $MFT itself is being
3046	 * written out as mark_ntfs_record_dirty() will only set I_DIRTY_PAGES
3047	 * on the $MFT inode and hence ntfs_write_inode() will not be
3048	 * re-invoked because of it which in turn is ok since the dirtied mft
3049	 * record will be cleaned and written out to disk below, i.e. before
3050	 * this function returns.
3051	 */
3052	if (modified) {
3053		flush_dcache_mft_record_page(ctx->ntfs_ino);
3054		if (!NInoTestSetDirty(ctx->ntfs_ino))
3055			mark_ntfs_record_dirty(ctx->ntfs_ino->page,
3056					ctx->ntfs_ino->page_ofs);
3057	}
3058	ntfs_attr_put_search_ctx(ctx);
3059	/* Now the access times are updated, write the base mft record. */
3060	if (NInoDirty(ni))
3061		err = write_mft_record(ni, m, sync);
3062	/* Write all attached extent mft records. */
3063	mutex_lock(&ni->extent_lock);
3064	if (ni->nr_extents > 0) {
3065		ntfs_inode **extent_nis = ni->ext.extent_ntfs_inos;
3066		int i;
3067
3068		ntfs_debug("Writing %i extent inodes.", ni->nr_extents);
3069		for (i = 0; i < ni->nr_extents; i++) {
3070			ntfs_inode *tni = extent_nis[i];
3071
3072			if (NInoDirty(tni)) {
3073				MFT_RECORD *tm = map_mft_record(tni);
3074				int ret;
3075
3076				if (IS_ERR(tm)) {
3077					if (!err || err == -ENOMEM)
3078						err = PTR_ERR(tm);
3079					continue;
3080				}
3081				ret = write_mft_record(tni, tm, sync);
3082				unmap_mft_record(tni);
3083				if (unlikely(ret)) {
3084					if (!err || err == -ENOMEM)
3085						err = ret;
3086				}
3087			}
3088		}
3089	}
3090	mutex_unlock(&ni->extent_lock);
3091	unmap_mft_record(ni);
3092	if (unlikely(err))
3093		goto err_out;
3094	ntfs_debug("Done.");
3095	return 0;
3096unm_err_out:
3097	unmap_mft_record(ni);
3098err_out:
3099	if (err == -ENOMEM) {
3100		ntfs_warning(vi->i_sb, "Not enough memory to write inode.  "
3101				"Marking the inode dirty again, so the VFS "
3102				"retries later.");
3103		mark_inode_dirty(vi);
3104	} else {
3105		ntfs_error(vi->i_sb, "Failed (error %i):  Run chkdsk.", -err);
3106		NVolSetErrors(ni->vol);
3107	}
3108	return err;
3109}
3110
3111#endif /* NTFS_RW */
v5.9
   1// SPDX-License-Identifier: GPL-2.0-or-later
   2/**
   3 * inode.c - NTFS kernel inode handling.
   4 *
   5 * Copyright (c) 2001-2014 Anton Altaparmakov and Tuxera Inc.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
   6 */
   7
   8#include <linux/buffer_head.h>
   9#include <linux/fs.h>
  10#include <linux/mm.h>
  11#include <linux/mount.h>
  12#include <linux/mutex.h>
  13#include <linux/pagemap.h>
  14#include <linux/quotaops.h>
  15#include <linux/slab.h>
  16#include <linux/log2.h>
  17
  18#include "aops.h"
  19#include "attrib.h"
  20#include "bitmap.h"
  21#include "dir.h"
  22#include "debug.h"
  23#include "inode.h"
  24#include "lcnalloc.h"
  25#include "malloc.h"
  26#include "mft.h"
  27#include "time.h"
  28#include "ntfs.h"
  29
  30/**
  31 * ntfs_test_inode - compare two (possibly fake) inodes for equality
  32 * @vi:		vfs inode which to test
  33 * @data:	data which is being tested with
  34 *
  35 * Compare the ntfs attribute embedded in the ntfs specific part of the vfs
  36 * inode @vi for equality with the ntfs attribute @data.
  37 *
  38 * If searching for the normal file/directory inode, set @na->type to AT_UNUSED.
  39 * @na->name and @na->name_len are then ignored.
  40 *
  41 * Return 1 if the attributes match and 0 if not.
  42 *
  43 * NOTE: This function runs with the inode_hash_lock spin lock held so it is not
  44 * allowed to sleep.
  45 */
  46int ntfs_test_inode(struct inode *vi, void *data)
  47{
  48	ntfs_attr *na = (ntfs_attr *)data;
  49	ntfs_inode *ni;
  50
  51	if (vi->i_ino != na->mft_no)
  52		return 0;
  53	ni = NTFS_I(vi);
  54	/* If !NInoAttr(ni), @vi is a normal file or directory inode. */
  55	if (likely(!NInoAttr(ni))) {
  56		/* If not looking for a normal inode this is a mismatch. */
  57		if (unlikely(na->type != AT_UNUSED))
  58			return 0;
  59	} else {
  60		/* A fake inode describing an attribute. */
  61		if (ni->type != na->type)
  62			return 0;
  63		if (ni->name_len != na->name_len)
  64			return 0;
  65		if (na->name_len && memcmp(ni->name, na->name,
  66				na->name_len * sizeof(ntfschar)))
  67			return 0;
  68	}
  69	/* Match! */
  70	return 1;
  71}
  72
  73/**
  74 * ntfs_init_locked_inode - initialize an inode
  75 * @vi:		vfs inode to initialize
  76 * @data:	data which to initialize @vi to
  77 *
  78 * Initialize the vfs inode @vi with the values from the ntfs attribute @data in
  79 * order to enable ntfs_test_inode() to do its work.
  80 *
  81 * If initializing the normal file/directory inode, set @na->type to AT_UNUSED.
  82 * In that case, @na->name and @na->name_len should be set to NULL and 0,
  83 * respectively. Although that is not strictly necessary as
  84 * ntfs_read_locked_inode() will fill them in later.
  85 *
  86 * Return 0 on success and -errno on error.
  87 *
  88 * NOTE: This function runs with the inode->i_lock spin lock held so it is not
  89 * allowed to sleep. (Hence the GFP_ATOMIC allocation.)
  90 */
  91static int ntfs_init_locked_inode(struct inode *vi, void *data)
  92{
  93	ntfs_attr *na = (ntfs_attr *)data;
  94	ntfs_inode *ni = NTFS_I(vi);
  95
  96	vi->i_ino = na->mft_no;
  97
  98	ni->type = na->type;
  99	if (na->type == AT_INDEX_ALLOCATION)
 100		NInoSetMstProtected(ni);
 101
 102	ni->name = na->name;
 103	ni->name_len = na->name_len;
 104
 105	/* If initializing a normal inode, we are done. */
 106	if (likely(na->type == AT_UNUSED)) {
 107		BUG_ON(na->name);
 108		BUG_ON(na->name_len);
 109		return 0;
 110	}
 111
 112	/* It is a fake inode. */
 113	NInoSetAttr(ni);
 114
 115	/*
 116	 * We have I30 global constant as an optimization as it is the name
 117	 * in >99.9% of named attributes! The other <0.1% incur a GFP_ATOMIC
 118	 * allocation but that is ok. And most attributes are unnamed anyway,
 119	 * thus the fraction of named attributes with name != I30 is actually
 120	 * absolutely tiny.
 121	 */
 122	if (na->name_len && na->name != I30) {
 123		unsigned int i;
 124
 125		BUG_ON(!na->name);
 126		i = na->name_len * sizeof(ntfschar);
 127		ni->name = kmalloc(i + sizeof(ntfschar), GFP_ATOMIC);
 128		if (!ni->name)
 129			return -ENOMEM;
 130		memcpy(ni->name, na->name, i);
 131		ni->name[na->name_len] = 0;
 132	}
 133	return 0;
 134}
 135
 
 136static int ntfs_read_locked_inode(struct inode *vi);
 137static int ntfs_read_locked_attr_inode(struct inode *base_vi, struct inode *vi);
 138static int ntfs_read_locked_index_inode(struct inode *base_vi,
 139		struct inode *vi);
 140
 141/**
 142 * ntfs_iget - obtain a struct inode corresponding to a specific normal inode
 143 * @sb:		super block of mounted volume
 144 * @mft_no:	mft record number / inode number to obtain
 145 *
 146 * Obtain the struct inode corresponding to a specific normal inode (i.e. a
 147 * file or directory).
 148 *
 149 * If the inode is in the cache, it is just returned with an increased
 150 * reference count. Otherwise, a new struct inode is allocated and initialized,
 151 * and finally ntfs_read_locked_inode() is called to read in the inode and
 152 * fill in the remainder of the inode structure.
 153 *
 154 * Return the struct inode on success. Check the return value with IS_ERR() and
 155 * if true, the function failed and the error code is obtained from PTR_ERR().
 156 */
 157struct inode *ntfs_iget(struct super_block *sb, unsigned long mft_no)
 158{
 159	struct inode *vi;
 160	int err;
 161	ntfs_attr na;
 162
 163	na.mft_no = mft_no;
 164	na.type = AT_UNUSED;
 165	na.name = NULL;
 166	na.name_len = 0;
 167
 168	vi = iget5_locked(sb, mft_no, ntfs_test_inode,
 169			ntfs_init_locked_inode, &na);
 170	if (unlikely(!vi))
 171		return ERR_PTR(-ENOMEM);
 172
 173	err = 0;
 174
 175	/* If this is a freshly allocated inode, need to read it now. */
 176	if (vi->i_state & I_NEW) {
 177		err = ntfs_read_locked_inode(vi);
 178		unlock_new_inode(vi);
 179	}
 180	/*
 181	 * There is no point in keeping bad inodes around if the failure was
 182	 * due to ENOMEM. We want to be able to retry again later.
 183	 */
 184	if (unlikely(err == -ENOMEM)) {
 185		iput(vi);
 186		vi = ERR_PTR(err);
 187	}
 188	return vi;
 189}
 190
 191/**
 192 * ntfs_attr_iget - obtain a struct inode corresponding to an attribute
 193 * @base_vi:	vfs base inode containing the attribute
 194 * @type:	attribute type
 195 * @name:	Unicode name of the attribute (NULL if unnamed)
 196 * @name_len:	length of @name in Unicode characters (0 if unnamed)
 197 *
 198 * Obtain the (fake) struct inode corresponding to the attribute specified by
 199 * @type, @name, and @name_len, which is present in the base mft record
 200 * specified by the vfs inode @base_vi.
 201 *
 202 * If the attribute inode is in the cache, it is just returned with an
 203 * increased reference count. Otherwise, a new struct inode is allocated and
 204 * initialized, and finally ntfs_read_locked_attr_inode() is called to read the
 205 * attribute and fill in the inode structure.
 206 *
 207 * Note, for index allocation attributes, you need to use ntfs_index_iget()
 208 * instead of ntfs_attr_iget() as working with indices is a lot more complex.
 209 *
 210 * Return the struct inode of the attribute inode on success. Check the return
 211 * value with IS_ERR() and if true, the function failed and the error code is
 212 * obtained from PTR_ERR().
 213 */
 214struct inode *ntfs_attr_iget(struct inode *base_vi, ATTR_TYPE type,
 215		ntfschar *name, u32 name_len)
 216{
 217	struct inode *vi;
 218	int err;
 219	ntfs_attr na;
 220
 221	/* Make sure no one calls ntfs_attr_iget() for indices. */
 222	BUG_ON(type == AT_INDEX_ALLOCATION);
 223
 224	na.mft_no = base_vi->i_ino;
 225	na.type = type;
 226	na.name = name;
 227	na.name_len = name_len;
 228
 229	vi = iget5_locked(base_vi->i_sb, na.mft_no, ntfs_test_inode,
 230			ntfs_init_locked_inode, &na);
 231	if (unlikely(!vi))
 232		return ERR_PTR(-ENOMEM);
 233
 234	err = 0;
 235
 236	/* If this is a freshly allocated inode, need to read it now. */
 237	if (vi->i_state & I_NEW) {
 238		err = ntfs_read_locked_attr_inode(base_vi, vi);
 239		unlock_new_inode(vi);
 240	}
 241	/*
 242	 * There is no point in keeping bad attribute inodes around. This also
 243	 * simplifies things in that we never need to check for bad attribute
 244	 * inodes elsewhere.
 245	 */
 246	if (unlikely(err)) {
 247		iput(vi);
 248		vi = ERR_PTR(err);
 249	}
 250	return vi;
 251}
 252
 253/**
 254 * ntfs_index_iget - obtain a struct inode corresponding to an index
 255 * @base_vi:	vfs base inode containing the index related attributes
 256 * @name:	Unicode name of the index
 257 * @name_len:	length of @name in Unicode characters
 258 *
 259 * Obtain the (fake) struct inode corresponding to the index specified by @name
 260 * and @name_len, which is present in the base mft record specified by the vfs
 261 * inode @base_vi.
 262 *
 263 * If the index inode is in the cache, it is just returned with an increased
 264 * reference count.  Otherwise, a new struct inode is allocated and
 265 * initialized, and finally ntfs_read_locked_index_inode() is called to read
 266 * the index related attributes and fill in the inode structure.
 267 *
 268 * Return the struct inode of the index inode on success. Check the return
 269 * value with IS_ERR() and if true, the function failed and the error code is
 270 * obtained from PTR_ERR().
 271 */
 272struct inode *ntfs_index_iget(struct inode *base_vi, ntfschar *name,
 273		u32 name_len)
 274{
 275	struct inode *vi;
 276	int err;
 277	ntfs_attr na;
 278
 279	na.mft_no = base_vi->i_ino;
 280	na.type = AT_INDEX_ALLOCATION;
 281	na.name = name;
 282	na.name_len = name_len;
 283
 284	vi = iget5_locked(base_vi->i_sb, na.mft_no, ntfs_test_inode,
 285			ntfs_init_locked_inode, &na);
 286	if (unlikely(!vi))
 287		return ERR_PTR(-ENOMEM);
 288
 289	err = 0;
 290
 291	/* If this is a freshly allocated inode, need to read it now. */
 292	if (vi->i_state & I_NEW) {
 293		err = ntfs_read_locked_index_inode(base_vi, vi);
 294		unlock_new_inode(vi);
 295	}
 296	/*
 297	 * There is no point in keeping bad index inodes around.  This also
 298	 * simplifies things in that we never need to check for bad index
 299	 * inodes elsewhere.
 300	 */
 301	if (unlikely(err)) {
 302		iput(vi);
 303		vi = ERR_PTR(err);
 304	}
 305	return vi;
 306}
 307
 308struct inode *ntfs_alloc_big_inode(struct super_block *sb)
 309{
 310	ntfs_inode *ni;
 311
 312	ntfs_debug("Entering.");
 313	ni = kmem_cache_alloc(ntfs_big_inode_cache, GFP_NOFS);
 314	if (likely(ni != NULL)) {
 315		ni->state = 0;
 316		return VFS_I(ni);
 317	}
 318	ntfs_error(sb, "Allocation of NTFS big inode structure failed.");
 319	return NULL;
 320}
 321
 322void ntfs_free_big_inode(struct inode *inode)
 323{
 
 324	kmem_cache_free(ntfs_big_inode_cache, NTFS_I(inode));
 325}
 326
 
 
 
 
 
 
 
 
 
 
 
 327static inline ntfs_inode *ntfs_alloc_extent_inode(void)
 328{
 329	ntfs_inode *ni;
 330
 331	ntfs_debug("Entering.");
 332	ni = kmem_cache_alloc(ntfs_inode_cache, GFP_NOFS);
 333	if (likely(ni != NULL)) {
 334		ni->state = 0;
 335		return ni;
 336	}
 337	ntfs_error(NULL, "Allocation of NTFS inode structure failed.");
 338	return NULL;
 339}
 340
 341static void ntfs_destroy_extent_inode(ntfs_inode *ni)
 342{
 343	ntfs_debug("Entering.");
 344	BUG_ON(ni->page);
 345	if (!atomic_dec_and_test(&ni->count))
 346		BUG();
 347	kmem_cache_free(ntfs_inode_cache, ni);
 348}
 349
 350/*
 351 * The attribute runlist lock has separate locking rules from the
 352 * normal runlist lock, so split the two lock-classes:
 353 */
 354static struct lock_class_key attr_list_rl_lock_class;
 355
 356/**
 357 * __ntfs_init_inode - initialize ntfs specific part of an inode
 358 * @sb:		super block of mounted volume
 359 * @ni:		freshly allocated ntfs inode which to initialize
 360 *
 361 * Initialize an ntfs inode to defaults.
 362 *
 363 * NOTE: ni->mft_no, ni->state, ni->type, ni->name, and ni->name_len are left
 364 * untouched. Make sure to initialize them elsewhere.
 365 *
 366 * Return zero on success and -ENOMEM on error.
 367 */
 368void __ntfs_init_inode(struct super_block *sb, ntfs_inode *ni)
 369{
 370	ntfs_debug("Entering.");
 371	rwlock_init(&ni->size_lock);
 372	ni->initialized_size = ni->allocated_size = 0;
 373	ni->seq_no = 0;
 374	atomic_set(&ni->count, 1);
 375	ni->vol = NTFS_SB(sb);
 376	ntfs_init_runlist(&ni->runlist);
 377	mutex_init(&ni->mrec_lock);
 378	ni->page = NULL;
 379	ni->page_ofs = 0;
 380	ni->attr_list_size = 0;
 381	ni->attr_list = NULL;
 382	ntfs_init_runlist(&ni->attr_list_rl);
 383	lockdep_set_class(&ni->attr_list_rl.lock,
 384				&attr_list_rl_lock_class);
 385	ni->itype.index.block_size = 0;
 386	ni->itype.index.vcn_size = 0;
 387	ni->itype.index.collation_rule = 0;
 388	ni->itype.index.block_size_bits = 0;
 389	ni->itype.index.vcn_size_bits = 0;
 390	mutex_init(&ni->extent_lock);
 391	ni->nr_extents = 0;
 392	ni->ext.base_ntfs_ino = NULL;
 393}
 394
 395/*
 396 * Extent inodes get MFT-mapped in a nested way, while the base inode
 397 * is still mapped. Teach this nesting to the lock validator by creating
 398 * a separate class for nested inode's mrec_lock's:
 399 */
 400static struct lock_class_key extent_inode_mrec_lock_key;
 401
 402inline ntfs_inode *ntfs_new_extent_inode(struct super_block *sb,
 403		unsigned long mft_no)
 404{
 405	ntfs_inode *ni = ntfs_alloc_extent_inode();
 406
 407	ntfs_debug("Entering.");
 408	if (likely(ni != NULL)) {
 409		__ntfs_init_inode(sb, ni);
 410		lockdep_set_class(&ni->mrec_lock, &extent_inode_mrec_lock_key);
 411		ni->mft_no = mft_no;
 412		ni->type = AT_UNUSED;
 413		ni->name = NULL;
 414		ni->name_len = 0;
 415	}
 416	return ni;
 417}
 418
 419/**
 420 * ntfs_is_extended_system_file - check if a file is in the $Extend directory
 421 * @ctx:	initialized attribute search context
 422 *
 423 * Search all file name attributes in the inode described by the attribute
 424 * search context @ctx and check if any of the names are in the $Extend system
 425 * directory.
 426 *
 427 * Return values:
 428 *	   1: file is in $Extend directory
 429 *	   0: file is not in $Extend directory
 430 *    -errno: failed to determine if the file is in the $Extend directory
 431 */
 432static int ntfs_is_extended_system_file(ntfs_attr_search_ctx *ctx)
 433{
 434	int nr_links, err;
 435
 436	/* Restart search. */
 437	ntfs_attr_reinit_search_ctx(ctx);
 438
 439	/* Get number of hard links. */
 440	nr_links = le16_to_cpu(ctx->mrec->link_count);
 441
 442	/* Loop through all hard links. */
 443	while (!(err = ntfs_attr_lookup(AT_FILE_NAME, NULL, 0, 0, 0, NULL, 0,
 444			ctx))) {
 445		FILE_NAME_ATTR *file_name_attr;
 446		ATTR_RECORD *attr = ctx->attr;
 447		u8 *p, *p2;
 448
 449		nr_links--;
 450		/*
 451		 * Maximum sanity checking as we are called on an inode that
 452		 * we suspect might be corrupt.
 453		 */
 454		p = (u8*)attr + le32_to_cpu(attr->length);
 455		if (p < (u8*)ctx->mrec || (u8*)p > (u8*)ctx->mrec +
 456				le32_to_cpu(ctx->mrec->bytes_in_use)) {
 457err_corrupt_attr:
 458			ntfs_error(ctx->ntfs_ino->vol->sb, "Corrupt file name "
 459					"attribute. You should run chkdsk.");
 460			return -EIO;
 461		}
 462		if (attr->non_resident) {
 463			ntfs_error(ctx->ntfs_ino->vol->sb, "Non-resident file "
 464					"name. You should run chkdsk.");
 465			return -EIO;
 466		}
 467		if (attr->flags) {
 468			ntfs_error(ctx->ntfs_ino->vol->sb, "File name with "
 469					"invalid flags. You should run "
 470					"chkdsk.");
 471			return -EIO;
 472		}
 473		if (!(attr->data.resident.flags & RESIDENT_ATTR_IS_INDEXED)) {
 474			ntfs_error(ctx->ntfs_ino->vol->sb, "Unindexed file "
 475					"name. You should run chkdsk.");
 476			return -EIO;
 477		}
 478		file_name_attr = (FILE_NAME_ATTR*)((u8*)attr +
 479				le16_to_cpu(attr->data.resident.value_offset));
 480		p2 = (u8*)attr + le32_to_cpu(attr->data.resident.value_length);
 481		if (p2 < (u8*)attr || p2 > p)
 482			goto err_corrupt_attr;
 483		/* This attribute is ok, but is it in the $Extend directory? */
 484		if (MREF_LE(file_name_attr->parent_directory) == FILE_Extend)
 485			return 1;	/* YES, it's an extended system file. */
 486	}
 487	if (unlikely(err != -ENOENT))
 488		return err;
 489	if (unlikely(nr_links)) {
 490		ntfs_error(ctx->ntfs_ino->vol->sb, "Inode hard link count "
 491				"doesn't match number of name attributes. You "
 492				"should run chkdsk.");
 493		return -EIO;
 494	}
 495	return 0;	/* NO, it is not an extended system file. */
 496}
 497
 498/**
 499 * ntfs_read_locked_inode - read an inode from its device
 500 * @vi:		inode to read
 501 *
 502 * ntfs_read_locked_inode() is called from ntfs_iget() to read the inode
 503 * described by @vi into memory from the device.
 504 *
 505 * The only fields in @vi that we need to/can look at when the function is
 506 * called are i_sb, pointing to the mounted device's super block, and i_ino,
 507 * the number of the inode to load.
 508 *
 509 * ntfs_read_locked_inode() maps, pins and locks the mft record number i_ino
 510 * for reading and sets up the necessary @vi fields as well as initializing
 511 * the ntfs inode.
 512 *
 513 * Q: What locks are held when the function is called?
 514 * A: i_state has I_NEW set, hence the inode is locked, also
 515 *    i_count is set to 1, so it is not going to go away
 516 *    i_flags is set to 0 and we have no business touching it.  Only an ioctl()
 517 *    is allowed to write to them. We should of course be honouring them but
 518 *    we need to do that using the IS_* macros defined in include/linux/fs.h.
 519 *    In any case ntfs_read_locked_inode() has nothing to do with i_flags.
 520 *
 521 * Return 0 on success and -errno on error.  In the error case, the inode will
 522 * have had make_bad_inode() executed on it.
 523 */
 524static int ntfs_read_locked_inode(struct inode *vi)
 525{
 526	ntfs_volume *vol = NTFS_SB(vi->i_sb);
 527	ntfs_inode *ni;
 528	struct inode *bvi;
 529	MFT_RECORD *m;
 530	ATTR_RECORD *a;
 531	STANDARD_INFORMATION *si;
 532	ntfs_attr_search_ctx *ctx;
 533	int err = 0;
 534
 535	ntfs_debug("Entering for i_ino 0x%lx.", vi->i_ino);
 536
 537	/* Setup the generic vfs inode parts now. */
 
 
 
 
 
 
 
 538	vi->i_uid = vol->uid;
 539	vi->i_gid = vol->gid;
 540	vi->i_mode = 0;
 541
 542	/*
 543	 * Initialize the ntfs specific part of @vi special casing
 544	 * FILE_MFT which we need to do at mount time.
 545	 */
 546	if (vi->i_ino != FILE_MFT)
 547		ntfs_init_big_inode(vi);
 548	ni = NTFS_I(vi);
 549
 550	m = map_mft_record(ni);
 551	if (IS_ERR(m)) {
 552		err = PTR_ERR(m);
 553		goto err_out;
 554	}
 555	ctx = ntfs_attr_get_search_ctx(ni, m);
 556	if (!ctx) {
 557		err = -ENOMEM;
 558		goto unm_err_out;
 559	}
 560
 561	if (!(m->flags & MFT_RECORD_IN_USE)) {
 562		ntfs_error(vi->i_sb, "Inode is not in use!");
 563		goto unm_err_out;
 564	}
 565	if (m->base_mft_record) {
 566		ntfs_error(vi->i_sb, "Inode is an extent inode!");
 567		goto unm_err_out;
 568	}
 569
 570	/* Transfer information from mft record into vfs and ntfs inodes. */
 571	vi->i_generation = ni->seq_no = le16_to_cpu(m->sequence_number);
 572
 573	/*
 574	 * FIXME: Keep in mind that link_count is two for files which have both
 575	 * a long file name and a short file name as separate entries, so if
 576	 * we are hiding short file names this will be too high. Either we need
 577	 * to account for the short file names by subtracting them or we need
 578	 * to make sure we delete files even though i_nlink is not zero which
 579	 * might be tricky due to vfs interactions. Need to think about this
 580	 * some more when implementing the unlink command.
 581	 */
 582	set_nlink(vi, le16_to_cpu(m->link_count));
 583	/*
 584	 * FIXME: Reparse points can have the directory bit set even though
 585	 * they would be S_IFLNK. Need to deal with this further below when we
 586	 * implement reparse points / symbolic links but it will do for now.
 587	 * Also if not a directory, it could be something else, rather than
 588	 * a regular file. But again, will do for now.
 589	 */
 590	/* Everyone gets all permissions. */
 591	vi->i_mode |= S_IRWXUGO;
 592	/* If read-only, no one gets write permissions. */
 593	if (IS_RDONLY(vi))
 594		vi->i_mode &= ~S_IWUGO;
 595	if (m->flags & MFT_RECORD_IS_DIRECTORY) {
 596		vi->i_mode |= S_IFDIR;
 597		/*
 598		 * Apply the directory permissions mask set in the mount
 599		 * options.
 600		 */
 601		vi->i_mode &= ~vol->dmask;
 602		/* Things break without this kludge! */
 603		if (vi->i_nlink > 1)
 604			set_nlink(vi, 1);
 605	} else {
 606		vi->i_mode |= S_IFREG;
 607		/* Apply the file permissions mask set in the mount options. */
 608		vi->i_mode &= ~vol->fmask;
 609	}
 610	/*
 611	 * Find the standard information attribute in the mft record. At this
 612	 * stage we haven't setup the attribute list stuff yet, so this could
 613	 * in fact fail if the standard information is in an extent record, but
 614	 * I don't think this actually ever happens.
 615	 */
 616	err = ntfs_attr_lookup(AT_STANDARD_INFORMATION, NULL, 0, 0, 0, NULL, 0,
 617			ctx);
 618	if (unlikely(err)) {
 619		if (err == -ENOENT) {
 620			/*
 621			 * TODO: We should be performing a hot fix here (if the
 622			 * recover mount option is set) by creating a new
 623			 * attribute.
 624			 */
 625			ntfs_error(vi->i_sb, "$STANDARD_INFORMATION attribute "
 626					"is missing.");
 627		}
 628		goto unm_err_out;
 629	}
 630	a = ctx->attr;
 631	/* Get the standard information attribute value. */
 632	si = (STANDARD_INFORMATION*)((u8*)a +
 633			le16_to_cpu(a->data.resident.value_offset));
 634
 635	/* Transfer information from the standard information into vi. */
 636	/*
 637	 * Note: The i_?times do not quite map perfectly onto the NTFS times,
 638	 * but they are close enough, and in the end it doesn't really matter
 639	 * that much...
 640	 */
 641	/*
 642	 * mtime is the last change of the data within the file. Not changed
 643	 * when only metadata is changed, e.g. a rename doesn't affect mtime.
 644	 */
 645	vi->i_mtime = ntfs2utc(si->last_data_change_time);
 646	/*
 647	 * ctime is the last change of the metadata of the file. This obviously
 648	 * always changes, when mtime is changed. ctime can be changed on its
 649	 * own, mtime is then not changed, e.g. when a file is renamed.
 650	 */
 651	vi->i_ctime = ntfs2utc(si->last_mft_change_time);
 652	/*
 653	 * Last access to the data within the file. Not changed during a rename
 654	 * for example but changed whenever the file is written to.
 655	 */
 656	vi->i_atime = ntfs2utc(si->last_access_time);
 657
 658	/* Find the attribute list attribute if present. */
 659	ntfs_attr_reinit_search_ctx(ctx);
 660	err = ntfs_attr_lookup(AT_ATTRIBUTE_LIST, NULL, 0, 0, 0, NULL, 0, ctx);
 661	if (err) {
 662		if (unlikely(err != -ENOENT)) {
 663			ntfs_error(vi->i_sb, "Failed to lookup attribute list "
 664					"attribute.");
 665			goto unm_err_out;
 666		}
 667	} else /* if (!err) */ {
 668		if (vi->i_ino == FILE_MFT)
 669			goto skip_attr_list_load;
 670		ntfs_debug("Attribute list found in inode 0x%lx.", vi->i_ino);
 671		NInoSetAttrList(ni);
 672		a = ctx->attr;
 673		if (a->flags & ATTR_COMPRESSION_MASK) {
 674			ntfs_error(vi->i_sb, "Attribute list attribute is "
 675					"compressed.");
 676			goto unm_err_out;
 677		}
 678		if (a->flags & ATTR_IS_ENCRYPTED ||
 679				a->flags & ATTR_IS_SPARSE) {
 680			if (a->non_resident) {
 681				ntfs_error(vi->i_sb, "Non-resident attribute "
 682						"list attribute is encrypted/"
 683						"sparse.");
 684				goto unm_err_out;
 685			}
 686			ntfs_warning(vi->i_sb, "Resident attribute list "
 687					"attribute in inode 0x%lx is marked "
 688					"encrypted/sparse which is not true.  "
 689					"However, Windows allows this and "
 690					"chkdsk does not detect or correct it "
 691					"so we will just ignore the invalid "
 692					"flags and pretend they are not set.",
 693					vi->i_ino);
 694		}
 695		/* Now allocate memory for the attribute list. */
 696		ni->attr_list_size = (u32)ntfs_attr_size(a);
 697		ni->attr_list = ntfs_malloc_nofs(ni->attr_list_size);
 698		if (!ni->attr_list) {
 699			ntfs_error(vi->i_sb, "Not enough memory to allocate "
 700					"buffer for attribute list.");
 701			err = -ENOMEM;
 702			goto unm_err_out;
 703		}
 704		if (a->non_resident) {
 705			NInoSetAttrListNonResident(ni);
 706			if (a->data.non_resident.lowest_vcn) {
 707				ntfs_error(vi->i_sb, "Attribute list has non "
 708						"zero lowest_vcn.");
 709				goto unm_err_out;
 710			}
 711			/*
 712			 * Setup the runlist. No need for locking as we have
 713			 * exclusive access to the inode at this time.
 714			 */
 715			ni->attr_list_rl.rl = ntfs_mapping_pairs_decompress(vol,
 716					a, NULL);
 717			if (IS_ERR(ni->attr_list_rl.rl)) {
 718				err = PTR_ERR(ni->attr_list_rl.rl);
 719				ni->attr_list_rl.rl = NULL;
 720				ntfs_error(vi->i_sb, "Mapping pairs "
 721						"decompression failed.");
 722				goto unm_err_out;
 723			}
 724			/* Now load the attribute list. */
 725			if ((err = load_attribute_list(vol, &ni->attr_list_rl,
 726					ni->attr_list, ni->attr_list_size,
 727					sle64_to_cpu(a->data.non_resident.
 728					initialized_size)))) {
 729				ntfs_error(vi->i_sb, "Failed to load "
 730						"attribute list attribute.");
 731				goto unm_err_out;
 732			}
 733		} else /* if (!a->non_resident) */ {
 734			if ((u8*)a + le16_to_cpu(a->data.resident.value_offset)
 735					+ le32_to_cpu(
 736					a->data.resident.value_length) >
 737					(u8*)ctx->mrec + vol->mft_record_size) {
 738				ntfs_error(vi->i_sb, "Corrupt attribute list "
 739						"in inode.");
 740				goto unm_err_out;
 741			}
 742			/* Now copy the attribute list. */
 743			memcpy(ni->attr_list, (u8*)a + le16_to_cpu(
 744					a->data.resident.value_offset),
 745					le32_to_cpu(
 746					a->data.resident.value_length));
 747		}
 748	}
 749skip_attr_list_load:
 750	/*
 751	 * If an attribute list is present we now have the attribute list value
 752	 * in ntfs_ino->attr_list and it is ntfs_ino->attr_list_size bytes.
 753	 */
 754	if (S_ISDIR(vi->i_mode)) {
 755		loff_t bvi_size;
 756		ntfs_inode *bni;
 757		INDEX_ROOT *ir;
 758		u8 *ir_end, *index_end;
 759
 760		/* It is a directory, find index root attribute. */
 761		ntfs_attr_reinit_search_ctx(ctx);
 762		err = ntfs_attr_lookup(AT_INDEX_ROOT, I30, 4, CASE_SENSITIVE,
 763				0, NULL, 0, ctx);
 764		if (unlikely(err)) {
 765			if (err == -ENOENT) {
 766				// FIXME: File is corrupt! Hot-fix with empty
 767				// index root attribute if recovery option is
 768				// set.
 769				ntfs_error(vi->i_sb, "$INDEX_ROOT attribute "
 770						"is missing.");
 771			}
 772			goto unm_err_out;
 773		}
 774		a = ctx->attr;
 775		/* Set up the state. */
 776		if (unlikely(a->non_resident)) {
 777			ntfs_error(vol->sb, "$INDEX_ROOT attribute is not "
 778					"resident.");
 779			goto unm_err_out;
 780		}
 781		/* Ensure the attribute name is placed before the value. */
 782		if (unlikely(a->name_length && (le16_to_cpu(a->name_offset) >=
 783				le16_to_cpu(a->data.resident.value_offset)))) {
 784			ntfs_error(vol->sb, "$INDEX_ROOT attribute name is "
 785					"placed after the attribute value.");
 786			goto unm_err_out;
 787		}
 788		/*
 789		 * Compressed/encrypted index root just means that the newly
 790		 * created files in that directory should be created compressed/
 791		 * encrypted. However index root cannot be both compressed and
 792		 * encrypted.
 793		 */
 794		if (a->flags & ATTR_COMPRESSION_MASK)
 795			NInoSetCompressed(ni);
 796		if (a->flags & ATTR_IS_ENCRYPTED) {
 797			if (a->flags & ATTR_COMPRESSION_MASK) {
 798				ntfs_error(vi->i_sb, "Found encrypted and "
 799						"compressed attribute.");
 800				goto unm_err_out;
 801			}
 802			NInoSetEncrypted(ni);
 803		}
 804		if (a->flags & ATTR_IS_SPARSE)
 805			NInoSetSparse(ni);
 806		ir = (INDEX_ROOT*)((u8*)a +
 807				le16_to_cpu(a->data.resident.value_offset));
 808		ir_end = (u8*)ir + le32_to_cpu(a->data.resident.value_length);
 809		if (ir_end > (u8*)ctx->mrec + vol->mft_record_size) {
 810			ntfs_error(vi->i_sb, "$INDEX_ROOT attribute is "
 811					"corrupt.");
 812			goto unm_err_out;
 813		}
 814		index_end = (u8*)&ir->index +
 815				le32_to_cpu(ir->index.index_length);
 816		if (index_end > ir_end) {
 817			ntfs_error(vi->i_sb, "Directory index is corrupt.");
 818			goto unm_err_out;
 819		}
 820		if (ir->type != AT_FILE_NAME) {
 821			ntfs_error(vi->i_sb, "Indexed attribute is not "
 822					"$FILE_NAME.");
 823			goto unm_err_out;
 824		}
 825		if (ir->collation_rule != COLLATION_FILE_NAME) {
 826			ntfs_error(vi->i_sb, "Index collation rule is not "
 827					"COLLATION_FILE_NAME.");
 828			goto unm_err_out;
 829		}
 830		ni->itype.index.collation_rule = ir->collation_rule;
 831		ni->itype.index.block_size = le32_to_cpu(ir->index_block_size);
 832		if (ni->itype.index.block_size &
 833				(ni->itype.index.block_size - 1)) {
 834			ntfs_error(vi->i_sb, "Index block size (%u) is not a "
 835					"power of two.",
 836					ni->itype.index.block_size);
 837			goto unm_err_out;
 838		}
 839		if (ni->itype.index.block_size > PAGE_SIZE) {
 840			ntfs_error(vi->i_sb, "Index block size (%u) > "
 841					"PAGE_SIZE (%ld) is not "
 842					"supported.  Sorry.",
 843					ni->itype.index.block_size,
 844					PAGE_SIZE);
 845			err = -EOPNOTSUPP;
 846			goto unm_err_out;
 847		}
 848		if (ni->itype.index.block_size < NTFS_BLOCK_SIZE) {
 849			ntfs_error(vi->i_sb, "Index block size (%u) < "
 850					"NTFS_BLOCK_SIZE (%i) is not "
 851					"supported.  Sorry.",
 852					ni->itype.index.block_size,
 853					NTFS_BLOCK_SIZE);
 854			err = -EOPNOTSUPP;
 855			goto unm_err_out;
 856		}
 857		ni->itype.index.block_size_bits =
 858				ffs(ni->itype.index.block_size) - 1;
 859		/* Determine the size of a vcn in the directory index. */
 860		if (vol->cluster_size <= ni->itype.index.block_size) {
 861			ni->itype.index.vcn_size = vol->cluster_size;
 862			ni->itype.index.vcn_size_bits = vol->cluster_size_bits;
 863		} else {
 864			ni->itype.index.vcn_size = vol->sector_size;
 865			ni->itype.index.vcn_size_bits = vol->sector_size_bits;
 866		}
 867
 868		/* Setup the index allocation attribute, even if not present. */
 869		NInoSetMstProtected(ni);
 870		ni->type = AT_INDEX_ALLOCATION;
 871		ni->name = I30;
 872		ni->name_len = 4;
 873
 874		if (!(ir->index.flags & LARGE_INDEX)) {
 875			/* No index allocation. */
 876			vi->i_size = ni->initialized_size =
 877					ni->allocated_size = 0;
 878			/* We are done with the mft record, so we release it. */
 879			ntfs_attr_put_search_ctx(ctx);
 880			unmap_mft_record(ni);
 881			m = NULL;
 882			ctx = NULL;
 883			goto skip_large_dir_stuff;
 884		} /* LARGE_INDEX: Index allocation present. Setup state. */
 885		NInoSetIndexAllocPresent(ni);
 886		/* Find index allocation attribute. */
 887		ntfs_attr_reinit_search_ctx(ctx);
 888		err = ntfs_attr_lookup(AT_INDEX_ALLOCATION, I30, 4,
 889				CASE_SENSITIVE, 0, NULL, 0, ctx);
 890		if (unlikely(err)) {
 891			if (err == -ENOENT)
 892				ntfs_error(vi->i_sb, "$INDEX_ALLOCATION "
 893						"attribute is not present but "
 894						"$INDEX_ROOT indicated it is.");
 895			else
 896				ntfs_error(vi->i_sb, "Failed to lookup "
 897						"$INDEX_ALLOCATION "
 898						"attribute.");
 899			goto unm_err_out;
 900		}
 901		a = ctx->attr;
 902		if (!a->non_resident) {
 903			ntfs_error(vi->i_sb, "$INDEX_ALLOCATION attribute "
 904					"is resident.");
 905			goto unm_err_out;
 906		}
 907		/*
 908		 * Ensure the attribute name is placed before the mapping pairs
 909		 * array.
 910		 */
 911		if (unlikely(a->name_length && (le16_to_cpu(a->name_offset) >=
 912				le16_to_cpu(
 913				a->data.non_resident.mapping_pairs_offset)))) {
 914			ntfs_error(vol->sb, "$INDEX_ALLOCATION attribute name "
 915					"is placed after the mapping pairs "
 916					"array.");
 917			goto unm_err_out;
 918		}
 919		if (a->flags & ATTR_IS_ENCRYPTED) {
 920			ntfs_error(vi->i_sb, "$INDEX_ALLOCATION attribute "
 921					"is encrypted.");
 922			goto unm_err_out;
 923		}
 924		if (a->flags & ATTR_IS_SPARSE) {
 925			ntfs_error(vi->i_sb, "$INDEX_ALLOCATION attribute "
 926					"is sparse.");
 927			goto unm_err_out;
 928		}
 929		if (a->flags & ATTR_COMPRESSION_MASK) {
 930			ntfs_error(vi->i_sb, "$INDEX_ALLOCATION attribute "
 931					"is compressed.");
 932			goto unm_err_out;
 933		}
 934		if (a->data.non_resident.lowest_vcn) {
 935			ntfs_error(vi->i_sb, "First extent of "
 936					"$INDEX_ALLOCATION attribute has non "
 937					"zero lowest_vcn.");
 938			goto unm_err_out;
 939		}
 940		vi->i_size = sle64_to_cpu(a->data.non_resident.data_size);
 941		ni->initialized_size = sle64_to_cpu(
 942				a->data.non_resident.initialized_size);
 943		ni->allocated_size = sle64_to_cpu(
 944				a->data.non_resident.allocated_size);
 945		/*
 946		 * We are done with the mft record, so we release it. Otherwise
 947		 * we would deadlock in ntfs_attr_iget().
 948		 */
 949		ntfs_attr_put_search_ctx(ctx);
 950		unmap_mft_record(ni);
 951		m = NULL;
 952		ctx = NULL;
 953		/* Get the index bitmap attribute inode. */
 954		bvi = ntfs_attr_iget(vi, AT_BITMAP, I30, 4);
 955		if (IS_ERR(bvi)) {
 956			ntfs_error(vi->i_sb, "Failed to get bitmap attribute.");
 957			err = PTR_ERR(bvi);
 958			goto unm_err_out;
 959		}
 960		bni = NTFS_I(bvi);
 961		if (NInoCompressed(bni) || NInoEncrypted(bni) ||
 962				NInoSparse(bni)) {
 963			ntfs_error(vi->i_sb, "$BITMAP attribute is compressed "
 964					"and/or encrypted and/or sparse.");
 965			goto iput_unm_err_out;
 966		}
 967		/* Consistency check bitmap size vs. index allocation size. */
 968		bvi_size = i_size_read(bvi);
 969		if ((bvi_size << 3) < (vi->i_size >>
 970				ni->itype.index.block_size_bits)) {
 971			ntfs_error(vi->i_sb, "Index bitmap too small (0x%llx) "
 972					"for index allocation (0x%llx).",
 973					bvi_size << 3, vi->i_size);
 974			goto iput_unm_err_out;
 975		}
 976		/* No longer need the bitmap attribute inode. */
 977		iput(bvi);
 978skip_large_dir_stuff:
 979		/* Setup the operations for this inode. */
 980		vi->i_op = &ntfs_dir_inode_ops;
 981		vi->i_fop = &ntfs_dir_ops;
 982		vi->i_mapping->a_ops = &ntfs_mst_aops;
 983	} else {
 984		/* It is a file. */
 985		ntfs_attr_reinit_search_ctx(ctx);
 986
 987		/* Setup the data attribute, even if not present. */
 988		ni->type = AT_DATA;
 989		ni->name = NULL;
 990		ni->name_len = 0;
 991
 992		/* Find first extent of the unnamed data attribute. */
 993		err = ntfs_attr_lookup(AT_DATA, NULL, 0, 0, 0, NULL, 0, ctx);
 994		if (unlikely(err)) {
 995			vi->i_size = ni->initialized_size =
 996					ni->allocated_size = 0;
 997			if (err != -ENOENT) {
 998				ntfs_error(vi->i_sb, "Failed to lookup $DATA "
 999						"attribute.");
1000				goto unm_err_out;
1001			}
1002			/*
1003			 * FILE_Secure does not have an unnamed $DATA
1004			 * attribute, so we special case it here.
1005			 */
1006			if (vi->i_ino == FILE_Secure)
1007				goto no_data_attr_special_case;
1008			/*
1009			 * Most if not all the system files in the $Extend
1010			 * system directory do not have unnamed data
1011			 * attributes so we need to check if the parent
1012			 * directory of the file is FILE_Extend and if it is
1013			 * ignore this error. To do this we need to get the
1014			 * name of this inode from the mft record as the name
1015			 * contains the back reference to the parent directory.
1016			 */
1017			if (ntfs_is_extended_system_file(ctx) > 0)
1018				goto no_data_attr_special_case;
1019			// FIXME: File is corrupt! Hot-fix with empty data
1020			// attribute if recovery option is set.
1021			ntfs_error(vi->i_sb, "$DATA attribute is missing.");
1022			goto unm_err_out;
1023		}
1024		a = ctx->attr;
1025		/* Setup the state. */
1026		if (a->flags & (ATTR_COMPRESSION_MASK | ATTR_IS_SPARSE)) {
1027			if (a->flags & ATTR_COMPRESSION_MASK) {
1028				NInoSetCompressed(ni);
1029				if (vol->cluster_size > 4096) {
1030					ntfs_error(vi->i_sb, "Found "
1031							"compressed data but "
1032							"compression is "
1033							"disabled due to "
1034							"cluster size (%i) > "
1035							"4kiB.",
1036							vol->cluster_size);
1037					goto unm_err_out;
1038				}
1039				if ((a->flags & ATTR_COMPRESSION_MASK)
1040						!= ATTR_IS_COMPRESSED) {
1041					ntfs_error(vi->i_sb, "Found unknown "
1042							"compression method "
1043							"or corrupt file.");
1044					goto unm_err_out;
1045				}
1046			}
1047			if (a->flags & ATTR_IS_SPARSE)
1048				NInoSetSparse(ni);
1049		}
1050		if (a->flags & ATTR_IS_ENCRYPTED) {
1051			if (NInoCompressed(ni)) {
1052				ntfs_error(vi->i_sb, "Found encrypted and "
1053						"compressed data.");
1054				goto unm_err_out;
1055			}
1056			NInoSetEncrypted(ni);
1057		}
1058		if (a->non_resident) {
1059			NInoSetNonResident(ni);
1060			if (NInoCompressed(ni) || NInoSparse(ni)) {
1061				if (NInoCompressed(ni) && a->data.non_resident.
1062						compression_unit != 4) {
1063					ntfs_error(vi->i_sb, "Found "
1064							"non-standard "
1065							"compression unit (%u "
1066							"instead of 4).  "
1067							"Cannot handle this.",
1068							a->data.non_resident.
1069							compression_unit);
1070					err = -EOPNOTSUPP;
1071					goto unm_err_out;
1072				}
1073				if (a->data.non_resident.compression_unit) {
1074					ni->itype.compressed.block_size = 1U <<
1075							(a->data.non_resident.
1076							compression_unit +
1077							vol->cluster_size_bits);
1078					ni->itype.compressed.block_size_bits =
1079							ffs(ni->itype.
1080							compressed.
1081							block_size) - 1;
1082					ni->itype.compressed.block_clusters =
1083							1U << a->data.
1084							non_resident.
1085							compression_unit;
1086				} else {
1087					ni->itype.compressed.block_size = 0;
1088					ni->itype.compressed.block_size_bits =
1089							0;
1090					ni->itype.compressed.block_clusters =
1091							0;
1092				}
1093				ni->itype.compressed.size = sle64_to_cpu(
1094						a->data.non_resident.
1095						compressed_size);
1096			}
1097			if (a->data.non_resident.lowest_vcn) {
1098				ntfs_error(vi->i_sb, "First extent of $DATA "
1099						"attribute has non zero "
1100						"lowest_vcn.");
1101				goto unm_err_out;
1102			}
1103			vi->i_size = sle64_to_cpu(
1104					a->data.non_resident.data_size);
1105			ni->initialized_size = sle64_to_cpu(
1106					a->data.non_resident.initialized_size);
1107			ni->allocated_size = sle64_to_cpu(
1108					a->data.non_resident.allocated_size);
1109		} else { /* Resident attribute. */
1110			vi->i_size = ni->initialized_size = le32_to_cpu(
1111					a->data.resident.value_length);
1112			ni->allocated_size = le32_to_cpu(a->length) -
1113					le16_to_cpu(
1114					a->data.resident.value_offset);
1115			if (vi->i_size > ni->allocated_size) {
1116				ntfs_error(vi->i_sb, "Resident data attribute "
1117						"is corrupt (size exceeds "
1118						"allocation).");
1119				goto unm_err_out;
1120			}
1121		}
1122no_data_attr_special_case:
1123		/* We are done with the mft record, so we release it. */
1124		ntfs_attr_put_search_ctx(ctx);
1125		unmap_mft_record(ni);
1126		m = NULL;
1127		ctx = NULL;
1128		/* Setup the operations for this inode. */
1129		vi->i_op = &ntfs_file_inode_ops;
1130		vi->i_fop = &ntfs_file_ops;
1131		vi->i_mapping->a_ops = &ntfs_normal_aops;
1132		if (NInoMstProtected(ni))
1133			vi->i_mapping->a_ops = &ntfs_mst_aops;
1134		else if (NInoCompressed(ni))
1135			vi->i_mapping->a_ops = &ntfs_compressed_aops;
1136	}
1137	/*
1138	 * The number of 512-byte blocks used on disk (for stat). This is in so
1139	 * far inaccurate as it doesn't account for any named streams or other
1140	 * special non-resident attributes, but that is how Windows works, too,
1141	 * so we are at least consistent with Windows, if not entirely
1142	 * consistent with the Linux Way. Doing it the Linux Way would cause a
1143	 * significant slowdown as it would involve iterating over all
1144	 * attributes in the mft record and adding the allocated/compressed
1145	 * sizes of all non-resident attributes present to give us the Linux
1146	 * correct size that should go into i_blocks (after division by 512).
1147	 */
1148	if (S_ISREG(vi->i_mode) && (NInoCompressed(ni) || NInoSparse(ni)))
1149		vi->i_blocks = ni->itype.compressed.size >> 9;
1150	else
1151		vi->i_blocks = ni->allocated_size >> 9;
1152	ntfs_debug("Done.");
1153	return 0;
1154iput_unm_err_out:
1155	iput(bvi);
1156unm_err_out:
1157	if (!err)
1158		err = -EIO;
1159	if (ctx)
1160		ntfs_attr_put_search_ctx(ctx);
1161	if (m)
1162		unmap_mft_record(ni);
1163err_out:
1164	ntfs_error(vol->sb, "Failed with error code %i.  Marking corrupt "
1165			"inode 0x%lx as bad.  Run chkdsk.", err, vi->i_ino);
1166	make_bad_inode(vi);
1167	if (err != -EOPNOTSUPP && err != -ENOMEM)
1168		NVolSetErrors(vol);
1169	return err;
1170}
1171
1172/**
1173 * ntfs_read_locked_attr_inode - read an attribute inode from its base inode
1174 * @base_vi:	base inode
1175 * @vi:		attribute inode to read
1176 *
1177 * ntfs_read_locked_attr_inode() is called from ntfs_attr_iget() to read the
1178 * attribute inode described by @vi into memory from the base mft record
1179 * described by @base_ni.
1180 *
1181 * ntfs_read_locked_attr_inode() maps, pins and locks the base inode for
1182 * reading and looks up the attribute described by @vi before setting up the
1183 * necessary fields in @vi as well as initializing the ntfs inode.
1184 *
1185 * Q: What locks are held when the function is called?
1186 * A: i_state has I_NEW set, hence the inode is locked, also
1187 *    i_count is set to 1, so it is not going to go away
1188 *
1189 * Return 0 on success and -errno on error.  In the error case, the inode will
1190 * have had make_bad_inode() executed on it.
1191 *
1192 * Note this cannot be called for AT_INDEX_ALLOCATION.
1193 */
1194static int ntfs_read_locked_attr_inode(struct inode *base_vi, struct inode *vi)
1195{
1196	ntfs_volume *vol = NTFS_SB(vi->i_sb);
1197	ntfs_inode *ni, *base_ni;
1198	MFT_RECORD *m;
1199	ATTR_RECORD *a;
1200	ntfs_attr_search_ctx *ctx;
1201	int err = 0;
1202
1203	ntfs_debug("Entering for i_ino 0x%lx.", vi->i_ino);
1204
1205	ntfs_init_big_inode(vi);
1206
1207	ni	= NTFS_I(vi);
1208	base_ni = NTFS_I(base_vi);
1209
1210	/* Just mirror the values from the base inode. */
 
1211	vi->i_uid	= base_vi->i_uid;
1212	vi->i_gid	= base_vi->i_gid;
1213	set_nlink(vi, base_vi->i_nlink);
1214	vi->i_mtime	= base_vi->i_mtime;
1215	vi->i_ctime	= base_vi->i_ctime;
1216	vi->i_atime	= base_vi->i_atime;
1217	vi->i_generation = ni->seq_no = base_ni->seq_no;
1218
1219	/* Set inode type to zero but preserve permissions. */
1220	vi->i_mode	= base_vi->i_mode & ~S_IFMT;
1221
1222	m = map_mft_record(base_ni);
1223	if (IS_ERR(m)) {
1224		err = PTR_ERR(m);
1225		goto err_out;
1226	}
1227	ctx = ntfs_attr_get_search_ctx(base_ni, m);
1228	if (!ctx) {
1229		err = -ENOMEM;
1230		goto unm_err_out;
1231	}
1232	/* Find the attribute. */
1233	err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len,
1234			CASE_SENSITIVE, 0, NULL, 0, ctx);
1235	if (unlikely(err))
1236		goto unm_err_out;
1237	a = ctx->attr;
1238	if (a->flags & (ATTR_COMPRESSION_MASK | ATTR_IS_SPARSE)) {
1239		if (a->flags & ATTR_COMPRESSION_MASK) {
1240			NInoSetCompressed(ni);
1241			if ((ni->type != AT_DATA) || (ni->type == AT_DATA &&
1242					ni->name_len)) {
1243				ntfs_error(vi->i_sb, "Found compressed "
1244						"non-data or named data "
1245						"attribute.  Please report "
1246						"you saw this message to "
1247						"linux-ntfs-dev@lists."
1248						"sourceforge.net");
1249				goto unm_err_out;
1250			}
1251			if (vol->cluster_size > 4096) {
1252				ntfs_error(vi->i_sb, "Found compressed "
1253						"attribute but compression is "
1254						"disabled due to cluster size "
1255						"(%i) > 4kiB.",
1256						vol->cluster_size);
1257				goto unm_err_out;
1258			}
1259			if ((a->flags & ATTR_COMPRESSION_MASK) !=
1260					ATTR_IS_COMPRESSED) {
1261				ntfs_error(vi->i_sb, "Found unknown "
1262						"compression method.");
1263				goto unm_err_out;
1264			}
1265		}
1266		/*
1267		 * The compressed/sparse flag set in an index root just means
1268		 * to compress all files.
1269		 */
1270		if (NInoMstProtected(ni) && ni->type != AT_INDEX_ROOT) {
1271			ntfs_error(vi->i_sb, "Found mst protected attribute "
1272					"but the attribute is %s.  Please "
1273					"report you saw this message to "
1274					"linux-ntfs-dev@lists.sourceforge.net",
1275					NInoCompressed(ni) ? "compressed" :
1276					"sparse");
1277			goto unm_err_out;
1278		}
1279		if (a->flags & ATTR_IS_SPARSE)
1280			NInoSetSparse(ni);
1281	}
1282	if (a->flags & ATTR_IS_ENCRYPTED) {
1283		if (NInoCompressed(ni)) {
1284			ntfs_error(vi->i_sb, "Found encrypted and compressed "
1285					"data.");
1286			goto unm_err_out;
1287		}
1288		/*
1289		 * The encryption flag set in an index root just means to
1290		 * encrypt all files.
1291		 */
1292		if (NInoMstProtected(ni) && ni->type != AT_INDEX_ROOT) {
1293			ntfs_error(vi->i_sb, "Found mst protected attribute "
1294					"but the attribute is encrypted.  "
1295					"Please report you saw this message "
1296					"to linux-ntfs-dev@lists.sourceforge."
1297					"net");
1298			goto unm_err_out;
1299		}
1300		if (ni->type != AT_DATA) {
1301			ntfs_error(vi->i_sb, "Found encrypted non-data "
1302					"attribute.");
1303			goto unm_err_out;
1304		}
1305		NInoSetEncrypted(ni);
1306	}
1307	if (!a->non_resident) {
1308		/* Ensure the attribute name is placed before the value. */
1309		if (unlikely(a->name_length && (le16_to_cpu(a->name_offset) >=
1310				le16_to_cpu(a->data.resident.value_offset)))) {
1311			ntfs_error(vol->sb, "Attribute name is placed after "
1312					"the attribute value.");
1313			goto unm_err_out;
1314		}
1315		if (NInoMstProtected(ni)) {
1316			ntfs_error(vi->i_sb, "Found mst protected attribute "
1317					"but the attribute is resident.  "
1318					"Please report you saw this message to "
1319					"linux-ntfs-dev@lists.sourceforge.net");
1320			goto unm_err_out;
1321		}
1322		vi->i_size = ni->initialized_size = le32_to_cpu(
1323				a->data.resident.value_length);
1324		ni->allocated_size = le32_to_cpu(a->length) -
1325				le16_to_cpu(a->data.resident.value_offset);
1326		if (vi->i_size > ni->allocated_size) {
1327			ntfs_error(vi->i_sb, "Resident attribute is corrupt "
1328					"(size exceeds allocation).");
1329			goto unm_err_out;
1330		}
1331	} else {
1332		NInoSetNonResident(ni);
1333		/*
1334		 * Ensure the attribute name is placed before the mapping pairs
1335		 * array.
1336		 */
1337		if (unlikely(a->name_length && (le16_to_cpu(a->name_offset) >=
1338				le16_to_cpu(
1339				a->data.non_resident.mapping_pairs_offset)))) {
1340			ntfs_error(vol->sb, "Attribute name is placed after "
1341					"the mapping pairs array.");
1342			goto unm_err_out;
1343		}
1344		if (NInoCompressed(ni) || NInoSparse(ni)) {
1345			if (NInoCompressed(ni) && a->data.non_resident.
1346					compression_unit != 4) {
1347				ntfs_error(vi->i_sb, "Found non-standard "
1348						"compression unit (%u instead "
1349						"of 4).  Cannot handle this.",
1350						a->data.non_resident.
1351						compression_unit);
1352				err = -EOPNOTSUPP;
1353				goto unm_err_out;
1354			}
1355			if (a->data.non_resident.compression_unit) {
1356				ni->itype.compressed.block_size = 1U <<
1357						(a->data.non_resident.
1358						compression_unit +
1359						vol->cluster_size_bits);
1360				ni->itype.compressed.block_size_bits =
1361						ffs(ni->itype.compressed.
1362						block_size) - 1;
1363				ni->itype.compressed.block_clusters = 1U <<
1364						a->data.non_resident.
1365						compression_unit;
1366			} else {
1367				ni->itype.compressed.block_size = 0;
1368				ni->itype.compressed.block_size_bits = 0;
1369				ni->itype.compressed.block_clusters = 0;
1370			}
1371			ni->itype.compressed.size = sle64_to_cpu(
1372					a->data.non_resident.compressed_size);
1373		}
1374		if (a->data.non_resident.lowest_vcn) {
1375			ntfs_error(vi->i_sb, "First extent of attribute has "
1376					"non-zero lowest_vcn.");
1377			goto unm_err_out;
1378		}
1379		vi->i_size = sle64_to_cpu(a->data.non_resident.data_size);
1380		ni->initialized_size = sle64_to_cpu(
1381				a->data.non_resident.initialized_size);
1382		ni->allocated_size = sle64_to_cpu(
1383				a->data.non_resident.allocated_size);
1384	}
1385	vi->i_mapping->a_ops = &ntfs_normal_aops;
1386	if (NInoMstProtected(ni))
1387		vi->i_mapping->a_ops = &ntfs_mst_aops;
1388	else if (NInoCompressed(ni))
1389		vi->i_mapping->a_ops = &ntfs_compressed_aops;
1390	if ((NInoCompressed(ni) || NInoSparse(ni)) && ni->type != AT_INDEX_ROOT)
1391		vi->i_blocks = ni->itype.compressed.size >> 9;
1392	else
1393		vi->i_blocks = ni->allocated_size >> 9;
1394	/*
1395	 * Make sure the base inode does not go away and attach it to the
1396	 * attribute inode.
1397	 */
1398	igrab(base_vi);
1399	ni->ext.base_ntfs_ino = base_ni;
1400	ni->nr_extents = -1;
1401
1402	ntfs_attr_put_search_ctx(ctx);
1403	unmap_mft_record(base_ni);
1404
1405	ntfs_debug("Done.");
1406	return 0;
1407
1408unm_err_out:
1409	if (!err)
1410		err = -EIO;
1411	if (ctx)
1412		ntfs_attr_put_search_ctx(ctx);
1413	unmap_mft_record(base_ni);
1414err_out:
1415	ntfs_error(vol->sb, "Failed with error code %i while reading attribute "
1416			"inode (mft_no 0x%lx, type 0x%x, name_len %i).  "
1417			"Marking corrupt inode and base inode 0x%lx as bad.  "
1418			"Run chkdsk.", err, vi->i_ino, ni->type, ni->name_len,
1419			base_vi->i_ino);
1420	make_bad_inode(vi);
1421	if (err != -ENOMEM)
1422		NVolSetErrors(vol);
1423	return err;
1424}
1425
1426/**
1427 * ntfs_read_locked_index_inode - read an index inode from its base inode
1428 * @base_vi:	base inode
1429 * @vi:		index inode to read
1430 *
1431 * ntfs_read_locked_index_inode() is called from ntfs_index_iget() to read the
1432 * index inode described by @vi into memory from the base mft record described
1433 * by @base_ni.
1434 *
1435 * ntfs_read_locked_index_inode() maps, pins and locks the base inode for
1436 * reading and looks up the attributes relating to the index described by @vi
1437 * before setting up the necessary fields in @vi as well as initializing the
1438 * ntfs inode.
1439 *
1440 * Note, index inodes are essentially attribute inodes (NInoAttr() is true)
1441 * with the attribute type set to AT_INDEX_ALLOCATION.  Apart from that, they
1442 * are setup like directory inodes since directories are a special case of
1443 * indices ao they need to be treated in much the same way.  Most importantly,
1444 * for small indices the index allocation attribute might not actually exist.
1445 * However, the index root attribute always exists but this does not need to
1446 * have an inode associated with it and this is why we define a new inode type
1447 * index.  Also, like for directories, we need to have an attribute inode for
1448 * the bitmap attribute corresponding to the index allocation attribute and we
1449 * can store this in the appropriate field of the inode, just like we do for
1450 * normal directory inodes.
1451 *
1452 * Q: What locks are held when the function is called?
1453 * A: i_state has I_NEW set, hence the inode is locked, also
1454 *    i_count is set to 1, so it is not going to go away
1455 *
1456 * Return 0 on success and -errno on error.  In the error case, the inode will
1457 * have had make_bad_inode() executed on it.
1458 */
1459static int ntfs_read_locked_index_inode(struct inode *base_vi, struct inode *vi)
1460{
1461	loff_t bvi_size;
1462	ntfs_volume *vol = NTFS_SB(vi->i_sb);
1463	ntfs_inode *ni, *base_ni, *bni;
1464	struct inode *bvi;
1465	MFT_RECORD *m;
1466	ATTR_RECORD *a;
1467	ntfs_attr_search_ctx *ctx;
1468	INDEX_ROOT *ir;
1469	u8 *ir_end, *index_end;
1470	int err = 0;
1471
1472	ntfs_debug("Entering for i_ino 0x%lx.", vi->i_ino);
1473	ntfs_init_big_inode(vi);
1474	ni	= NTFS_I(vi);
1475	base_ni = NTFS_I(base_vi);
1476	/* Just mirror the values from the base inode. */
 
1477	vi->i_uid	= base_vi->i_uid;
1478	vi->i_gid	= base_vi->i_gid;
1479	set_nlink(vi, base_vi->i_nlink);
1480	vi->i_mtime	= base_vi->i_mtime;
1481	vi->i_ctime	= base_vi->i_ctime;
1482	vi->i_atime	= base_vi->i_atime;
1483	vi->i_generation = ni->seq_no = base_ni->seq_no;
1484	/* Set inode type to zero but preserve permissions. */
1485	vi->i_mode	= base_vi->i_mode & ~S_IFMT;
1486	/* Map the mft record for the base inode. */
1487	m = map_mft_record(base_ni);
1488	if (IS_ERR(m)) {
1489		err = PTR_ERR(m);
1490		goto err_out;
1491	}
1492	ctx = ntfs_attr_get_search_ctx(base_ni, m);
1493	if (!ctx) {
1494		err = -ENOMEM;
1495		goto unm_err_out;
1496	}
1497	/* Find the index root attribute. */
1498	err = ntfs_attr_lookup(AT_INDEX_ROOT, ni->name, ni->name_len,
1499			CASE_SENSITIVE, 0, NULL, 0, ctx);
1500	if (unlikely(err)) {
1501		if (err == -ENOENT)
1502			ntfs_error(vi->i_sb, "$INDEX_ROOT attribute is "
1503					"missing.");
1504		goto unm_err_out;
1505	}
1506	a = ctx->attr;
1507	/* Set up the state. */
1508	if (unlikely(a->non_resident)) {
1509		ntfs_error(vol->sb, "$INDEX_ROOT attribute is not resident.");
1510		goto unm_err_out;
1511	}
1512	/* Ensure the attribute name is placed before the value. */
1513	if (unlikely(a->name_length && (le16_to_cpu(a->name_offset) >=
1514			le16_to_cpu(a->data.resident.value_offset)))) {
1515		ntfs_error(vol->sb, "$INDEX_ROOT attribute name is placed "
1516				"after the attribute value.");
1517		goto unm_err_out;
1518	}
1519	/*
1520	 * Compressed/encrypted/sparse index root is not allowed, except for
1521	 * directories of course but those are not dealt with here.
1522	 */
1523	if (a->flags & (ATTR_COMPRESSION_MASK | ATTR_IS_ENCRYPTED |
1524			ATTR_IS_SPARSE)) {
1525		ntfs_error(vi->i_sb, "Found compressed/encrypted/sparse index "
1526				"root attribute.");
1527		goto unm_err_out;
1528	}
1529	ir = (INDEX_ROOT*)((u8*)a + le16_to_cpu(a->data.resident.value_offset));
1530	ir_end = (u8*)ir + le32_to_cpu(a->data.resident.value_length);
1531	if (ir_end > (u8*)ctx->mrec + vol->mft_record_size) {
1532		ntfs_error(vi->i_sb, "$INDEX_ROOT attribute is corrupt.");
1533		goto unm_err_out;
1534	}
1535	index_end = (u8*)&ir->index + le32_to_cpu(ir->index.index_length);
1536	if (index_end > ir_end) {
1537		ntfs_error(vi->i_sb, "Index is corrupt.");
1538		goto unm_err_out;
1539	}
1540	if (ir->type) {
1541		ntfs_error(vi->i_sb, "Index type is not 0 (type is 0x%x).",
1542				le32_to_cpu(ir->type));
1543		goto unm_err_out;
1544	}
1545	ni->itype.index.collation_rule = ir->collation_rule;
1546	ntfs_debug("Index collation rule is 0x%x.",
1547			le32_to_cpu(ir->collation_rule));
1548	ni->itype.index.block_size = le32_to_cpu(ir->index_block_size);
1549	if (!is_power_of_2(ni->itype.index.block_size)) {
1550		ntfs_error(vi->i_sb, "Index block size (%u) is not a power of "
1551				"two.", ni->itype.index.block_size);
1552		goto unm_err_out;
1553	}
1554	if (ni->itype.index.block_size > PAGE_SIZE) {
1555		ntfs_error(vi->i_sb, "Index block size (%u) > PAGE_SIZE "
1556				"(%ld) is not supported.  Sorry.",
1557				ni->itype.index.block_size, PAGE_SIZE);
1558		err = -EOPNOTSUPP;
1559		goto unm_err_out;
1560	}
1561	if (ni->itype.index.block_size < NTFS_BLOCK_SIZE) {
1562		ntfs_error(vi->i_sb, "Index block size (%u) < NTFS_BLOCK_SIZE "
1563				"(%i) is not supported.  Sorry.",
1564				ni->itype.index.block_size, NTFS_BLOCK_SIZE);
1565		err = -EOPNOTSUPP;
1566		goto unm_err_out;
1567	}
1568	ni->itype.index.block_size_bits = ffs(ni->itype.index.block_size) - 1;
1569	/* Determine the size of a vcn in the index. */
1570	if (vol->cluster_size <= ni->itype.index.block_size) {
1571		ni->itype.index.vcn_size = vol->cluster_size;
1572		ni->itype.index.vcn_size_bits = vol->cluster_size_bits;
1573	} else {
1574		ni->itype.index.vcn_size = vol->sector_size;
1575		ni->itype.index.vcn_size_bits = vol->sector_size_bits;
1576	}
1577	/* Check for presence of index allocation attribute. */
1578	if (!(ir->index.flags & LARGE_INDEX)) {
1579		/* No index allocation. */
1580		vi->i_size = ni->initialized_size = ni->allocated_size = 0;
1581		/* We are done with the mft record, so we release it. */
1582		ntfs_attr_put_search_ctx(ctx);
1583		unmap_mft_record(base_ni);
1584		m = NULL;
1585		ctx = NULL;
1586		goto skip_large_index_stuff;
1587	} /* LARGE_INDEX:  Index allocation present.  Setup state. */
1588	NInoSetIndexAllocPresent(ni);
1589	/* Find index allocation attribute. */
1590	ntfs_attr_reinit_search_ctx(ctx);
1591	err = ntfs_attr_lookup(AT_INDEX_ALLOCATION, ni->name, ni->name_len,
1592			CASE_SENSITIVE, 0, NULL, 0, ctx);
1593	if (unlikely(err)) {
1594		if (err == -ENOENT)
1595			ntfs_error(vi->i_sb, "$INDEX_ALLOCATION attribute is "
1596					"not present but $INDEX_ROOT "
1597					"indicated it is.");
1598		else
1599			ntfs_error(vi->i_sb, "Failed to lookup "
1600					"$INDEX_ALLOCATION attribute.");
1601		goto unm_err_out;
1602	}
1603	a = ctx->attr;
1604	if (!a->non_resident) {
1605		ntfs_error(vi->i_sb, "$INDEX_ALLOCATION attribute is "
1606				"resident.");
1607		goto unm_err_out;
1608	}
1609	/*
1610	 * Ensure the attribute name is placed before the mapping pairs array.
1611	 */
1612	if (unlikely(a->name_length && (le16_to_cpu(a->name_offset) >=
1613			le16_to_cpu(
1614			a->data.non_resident.mapping_pairs_offset)))) {
1615		ntfs_error(vol->sb, "$INDEX_ALLOCATION attribute name is "
1616				"placed after the mapping pairs array.");
1617		goto unm_err_out;
1618	}
1619	if (a->flags & ATTR_IS_ENCRYPTED) {
1620		ntfs_error(vi->i_sb, "$INDEX_ALLOCATION attribute is "
1621				"encrypted.");
1622		goto unm_err_out;
1623	}
1624	if (a->flags & ATTR_IS_SPARSE) {
1625		ntfs_error(vi->i_sb, "$INDEX_ALLOCATION attribute is sparse.");
1626		goto unm_err_out;
1627	}
1628	if (a->flags & ATTR_COMPRESSION_MASK) {
1629		ntfs_error(vi->i_sb, "$INDEX_ALLOCATION attribute is "
1630				"compressed.");
1631		goto unm_err_out;
1632	}
1633	if (a->data.non_resident.lowest_vcn) {
1634		ntfs_error(vi->i_sb, "First extent of $INDEX_ALLOCATION "
1635				"attribute has non zero lowest_vcn.");
1636		goto unm_err_out;
1637	}
1638	vi->i_size = sle64_to_cpu(a->data.non_resident.data_size);
1639	ni->initialized_size = sle64_to_cpu(
1640			a->data.non_resident.initialized_size);
1641	ni->allocated_size = sle64_to_cpu(a->data.non_resident.allocated_size);
1642	/*
1643	 * We are done with the mft record, so we release it.  Otherwise
1644	 * we would deadlock in ntfs_attr_iget().
1645	 */
1646	ntfs_attr_put_search_ctx(ctx);
1647	unmap_mft_record(base_ni);
1648	m = NULL;
1649	ctx = NULL;
1650	/* Get the index bitmap attribute inode. */
1651	bvi = ntfs_attr_iget(base_vi, AT_BITMAP, ni->name, ni->name_len);
1652	if (IS_ERR(bvi)) {
1653		ntfs_error(vi->i_sb, "Failed to get bitmap attribute.");
1654		err = PTR_ERR(bvi);
1655		goto unm_err_out;
1656	}
1657	bni = NTFS_I(bvi);
1658	if (NInoCompressed(bni) || NInoEncrypted(bni) ||
1659			NInoSparse(bni)) {
1660		ntfs_error(vi->i_sb, "$BITMAP attribute is compressed and/or "
1661				"encrypted and/or sparse.");
1662		goto iput_unm_err_out;
1663	}
1664	/* Consistency check bitmap size vs. index allocation size. */
1665	bvi_size = i_size_read(bvi);
1666	if ((bvi_size << 3) < (vi->i_size >> ni->itype.index.block_size_bits)) {
1667		ntfs_error(vi->i_sb, "Index bitmap too small (0x%llx) for "
1668				"index allocation (0x%llx).", bvi_size << 3,
1669				vi->i_size);
1670		goto iput_unm_err_out;
1671	}
1672	iput(bvi);
1673skip_large_index_stuff:
1674	/* Setup the operations for this index inode. */
1675	vi->i_mapping->a_ops = &ntfs_mst_aops;
1676	vi->i_blocks = ni->allocated_size >> 9;
1677	/*
1678	 * Make sure the base inode doesn't go away and attach it to the
1679	 * index inode.
1680	 */
1681	igrab(base_vi);
1682	ni->ext.base_ntfs_ino = base_ni;
1683	ni->nr_extents = -1;
1684
1685	ntfs_debug("Done.");
1686	return 0;
1687iput_unm_err_out:
1688	iput(bvi);
1689unm_err_out:
1690	if (!err)
1691		err = -EIO;
1692	if (ctx)
1693		ntfs_attr_put_search_ctx(ctx);
1694	if (m)
1695		unmap_mft_record(base_ni);
1696err_out:
1697	ntfs_error(vi->i_sb, "Failed with error code %i while reading index "
1698			"inode (mft_no 0x%lx, name_len %i.", err, vi->i_ino,
1699			ni->name_len);
1700	make_bad_inode(vi);
1701	if (err != -EOPNOTSUPP && err != -ENOMEM)
1702		NVolSetErrors(vol);
1703	return err;
1704}
1705
1706/*
1707 * The MFT inode has special locking, so teach the lock validator
1708 * about this by splitting off the locking rules of the MFT from
1709 * the locking rules of other inodes. The MFT inode can never be
1710 * accessed from the VFS side (or even internally), only by the
1711 * map_mft functions.
1712 */
1713static struct lock_class_key mft_ni_runlist_lock_key, mft_ni_mrec_lock_key;
1714
1715/**
1716 * ntfs_read_inode_mount - special read_inode for mount time use only
1717 * @vi:		inode to read
1718 *
1719 * Read inode FILE_MFT at mount time, only called with super_block lock
1720 * held from within the read_super() code path.
1721 *
1722 * This function exists because when it is called the page cache for $MFT/$DATA
1723 * is not initialized and hence we cannot get at the contents of mft records
1724 * by calling map_mft_record*().
1725 *
1726 * Further it needs to cope with the circular references problem, i.e. cannot
1727 * load any attributes other than $ATTRIBUTE_LIST until $DATA is loaded, because
1728 * we do not know where the other extent mft records are yet and again, because
1729 * we cannot call map_mft_record*() yet.  Obviously this applies only when an
1730 * attribute list is actually present in $MFT inode.
1731 *
1732 * We solve these problems by starting with the $DATA attribute before anything
1733 * else and iterating using ntfs_attr_lookup($DATA) over all extents.  As each
1734 * extent is found, we ntfs_mapping_pairs_decompress() including the implied
1735 * ntfs_runlists_merge().  Each step of the iteration necessarily provides
1736 * sufficient information for the next step to complete.
1737 *
1738 * This should work but there are two possible pit falls (see inline comments
1739 * below), but only time will tell if they are real pits or just smoke...
1740 */
1741int ntfs_read_inode_mount(struct inode *vi)
1742{
1743	VCN next_vcn, last_vcn, highest_vcn;
1744	s64 block;
1745	struct super_block *sb = vi->i_sb;
1746	ntfs_volume *vol = NTFS_SB(sb);
1747	struct buffer_head *bh;
1748	ntfs_inode *ni;
1749	MFT_RECORD *m = NULL;
1750	ATTR_RECORD *a;
1751	ntfs_attr_search_ctx *ctx;
1752	unsigned int i, nr_blocks;
1753	int err;
1754
1755	ntfs_debug("Entering.");
1756
1757	/* Initialize the ntfs specific part of @vi. */
1758	ntfs_init_big_inode(vi);
1759
1760	ni = NTFS_I(vi);
1761
1762	/* Setup the data attribute. It is special as it is mst protected. */
1763	NInoSetNonResident(ni);
1764	NInoSetMstProtected(ni);
1765	NInoSetSparseDisabled(ni);
1766	ni->type = AT_DATA;
1767	ni->name = NULL;
1768	ni->name_len = 0;
1769	/*
1770	 * This sets up our little cheat allowing us to reuse the async read io
1771	 * completion handler for directories.
1772	 */
1773	ni->itype.index.block_size = vol->mft_record_size;
1774	ni->itype.index.block_size_bits = vol->mft_record_size_bits;
1775
1776	/* Very important! Needed to be able to call map_mft_record*(). */
1777	vol->mft_ino = vi;
1778
1779	/* Allocate enough memory to read the first mft record. */
1780	if (vol->mft_record_size > 64 * 1024) {
1781		ntfs_error(sb, "Unsupported mft record size %i (max 64kiB).",
1782				vol->mft_record_size);
1783		goto err_out;
1784	}
1785	i = vol->mft_record_size;
1786	if (i < sb->s_blocksize)
1787		i = sb->s_blocksize;
1788	m = (MFT_RECORD*)ntfs_malloc_nofs(i);
1789	if (!m) {
1790		ntfs_error(sb, "Failed to allocate buffer for $MFT record 0.");
1791		goto err_out;
1792	}
1793
1794	/* Determine the first block of the $MFT/$DATA attribute. */
1795	block = vol->mft_lcn << vol->cluster_size_bits >>
1796			sb->s_blocksize_bits;
1797	nr_blocks = vol->mft_record_size >> sb->s_blocksize_bits;
1798	if (!nr_blocks)
1799		nr_blocks = 1;
1800
1801	/* Load $MFT/$DATA's first mft record. */
1802	for (i = 0; i < nr_blocks; i++) {
1803		bh = sb_bread(sb, block++);
1804		if (!bh) {
1805			ntfs_error(sb, "Device read failed.");
1806			goto err_out;
1807		}
1808		memcpy((char*)m + (i << sb->s_blocksize_bits), bh->b_data,
1809				sb->s_blocksize);
1810		brelse(bh);
1811	}
1812
1813	/* Apply the mst fixups. */
1814	if (post_read_mst_fixup((NTFS_RECORD*)m, vol->mft_record_size)) {
1815		/* FIXME: Try to use the $MFTMirr now. */
1816		ntfs_error(sb, "MST fixup failed. $MFT is corrupt.");
1817		goto err_out;
1818	}
1819
1820	/* Need this to sanity check attribute list references to $MFT. */
1821	vi->i_generation = ni->seq_no = le16_to_cpu(m->sequence_number);
1822
1823	/* Provides readpage() for map_mft_record(). */
1824	vi->i_mapping->a_ops = &ntfs_mst_aops;
1825
1826	ctx = ntfs_attr_get_search_ctx(ni, m);
1827	if (!ctx) {
1828		err = -ENOMEM;
1829		goto err_out;
1830	}
1831
1832	/* Find the attribute list attribute if present. */
1833	err = ntfs_attr_lookup(AT_ATTRIBUTE_LIST, NULL, 0, 0, 0, NULL, 0, ctx);
1834	if (err) {
1835		if (unlikely(err != -ENOENT)) {
1836			ntfs_error(sb, "Failed to lookup attribute list "
1837					"attribute. You should run chkdsk.");
1838			goto put_err_out;
1839		}
1840	} else /* if (!err) */ {
1841		ATTR_LIST_ENTRY *al_entry, *next_al_entry;
1842		u8 *al_end;
1843		static const char *es = "  Not allowed.  $MFT is corrupt.  "
1844				"You should run chkdsk.";
1845
1846		ntfs_debug("Attribute list attribute found in $MFT.");
1847		NInoSetAttrList(ni);
1848		a = ctx->attr;
1849		if (a->flags & ATTR_COMPRESSION_MASK) {
1850			ntfs_error(sb, "Attribute list attribute is "
1851					"compressed.%s", es);
1852			goto put_err_out;
1853		}
1854		if (a->flags & ATTR_IS_ENCRYPTED ||
1855				a->flags & ATTR_IS_SPARSE) {
1856			if (a->non_resident) {
1857				ntfs_error(sb, "Non-resident attribute list "
1858						"attribute is encrypted/"
1859						"sparse.%s", es);
1860				goto put_err_out;
1861			}
1862			ntfs_warning(sb, "Resident attribute list attribute "
1863					"in $MFT system file is marked "
1864					"encrypted/sparse which is not true.  "
1865					"However, Windows allows this and "
1866					"chkdsk does not detect or correct it "
1867					"so we will just ignore the invalid "
1868					"flags and pretend they are not set.");
1869		}
1870		/* Now allocate memory for the attribute list. */
1871		ni->attr_list_size = (u32)ntfs_attr_size(a);
1872		ni->attr_list = ntfs_malloc_nofs(ni->attr_list_size);
1873		if (!ni->attr_list) {
1874			ntfs_error(sb, "Not enough memory to allocate buffer "
1875					"for attribute list.");
1876			goto put_err_out;
1877		}
1878		if (a->non_resident) {
1879			NInoSetAttrListNonResident(ni);
1880			if (a->data.non_resident.lowest_vcn) {
1881				ntfs_error(sb, "Attribute list has non zero "
1882						"lowest_vcn. $MFT is corrupt. "
1883						"You should run chkdsk.");
1884				goto put_err_out;
1885			}
1886			/* Setup the runlist. */
1887			ni->attr_list_rl.rl = ntfs_mapping_pairs_decompress(vol,
1888					a, NULL);
1889			if (IS_ERR(ni->attr_list_rl.rl)) {
1890				err = PTR_ERR(ni->attr_list_rl.rl);
1891				ni->attr_list_rl.rl = NULL;
1892				ntfs_error(sb, "Mapping pairs decompression "
1893						"failed with error code %i.",
1894						-err);
1895				goto put_err_out;
1896			}
1897			/* Now load the attribute list. */
1898			if ((err = load_attribute_list(vol, &ni->attr_list_rl,
1899					ni->attr_list, ni->attr_list_size,
1900					sle64_to_cpu(a->data.
1901					non_resident.initialized_size)))) {
1902				ntfs_error(sb, "Failed to load attribute list "
1903						"attribute with error code %i.",
1904						-err);
1905				goto put_err_out;
1906			}
1907		} else /* if (!ctx.attr->non_resident) */ {
1908			if ((u8*)a + le16_to_cpu(
1909					a->data.resident.value_offset) +
1910					le32_to_cpu(
1911					a->data.resident.value_length) >
1912					(u8*)ctx->mrec + vol->mft_record_size) {
1913				ntfs_error(sb, "Corrupt attribute list "
1914						"attribute.");
1915				goto put_err_out;
1916			}
1917			/* Now copy the attribute list. */
1918			memcpy(ni->attr_list, (u8*)a + le16_to_cpu(
1919					a->data.resident.value_offset),
1920					le32_to_cpu(
1921					a->data.resident.value_length));
1922		}
1923		/* The attribute list is now setup in memory. */
1924		/*
1925		 * FIXME: I don't know if this case is actually possible.
1926		 * According to logic it is not possible but I have seen too
1927		 * many weird things in MS software to rely on logic... Thus we
1928		 * perform a manual search and make sure the first $MFT/$DATA
1929		 * extent is in the base inode. If it is not we abort with an
1930		 * error and if we ever see a report of this error we will need
1931		 * to do some magic in order to have the necessary mft record
1932		 * loaded and in the right place in the page cache. But
1933		 * hopefully logic will prevail and this never happens...
1934		 */
1935		al_entry = (ATTR_LIST_ENTRY*)ni->attr_list;
1936		al_end = (u8*)al_entry + ni->attr_list_size;
1937		for (;; al_entry = next_al_entry) {
1938			/* Out of bounds check. */
1939			if ((u8*)al_entry < ni->attr_list ||
1940					(u8*)al_entry > al_end)
1941				goto em_put_err_out;
1942			/* Catch the end of the attribute list. */
1943			if ((u8*)al_entry == al_end)
1944				goto em_put_err_out;
1945			if (!al_entry->length)
1946				goto em_put_err_out;
1947			if ((u8*)al_entry + 6 > al_end || (u8*)al_entry +
1948					le16_to_cpu(al_entry->length) > al_end)
1949				goto em_put_err_out;
1950			next_al_entry = (ATTR_LIST_ENTRY*)((u8*)al_entry +
1951					le16_to_cpu(al_entry->length));
1952			if (le32_to_cpu(al_entry->type) > le32_to_cpu(AT_DATA))
1953				goto em_put_err_out;
1954			if (AT_DATA != al_entry->type)
1955				continue;
1956			/* We want an unnamed attribute. */
1957			if (al_entry->name_length)
1958				goto em_put_err_out;
1959			/* Want the first entry, i.e. lowest_vcn == 0. */
1960			if (al_entry->lowest_vcn)
1961				goto em_put_err_out;
1962			/* First entry has to be in the base mft record. */
1963			if (MREF_LE(al_entry->mft_reference) != vi->i_ino) {
1964				/* MFT references do not match, logic fails. */
1965				ntfs_error(sb, "BUG: The first $DATA extent "
1966						"of $MFT is not in the base "
1967						"mft record. Please report "
1968						"you saw this message to "
1969						"linux-ntfs-dev@lists."
1970						"sourceforge.net");
1971				goto put_err_out;
1972			} else {
1973				/* Sequence numbers must match. */
1974				if (MSEQNO_LE(al_entry->mft_reference) !=
1975						ni->seq_no)
1976					goto em_put_err_out;
1977				/* Got it. All is ok. We can stop now. */
1978				break;
1979			}
1980		}
1981	}
1982
1983	ntfs_attr_reinit_search_ctx(ctx);
1984
1985	/* Now load all attribute extents. */
1986	a = NULL;
1987	next_vcn = last_vcn = highest_vcn = 0;
1988	while (!(err = ntfs_attr_lookup(AT_DATA, NULL, 0, 0, next_vcn, NULL, 0,
1989			ctx))) {
1990		runlist_element *nrl;
1991
1992		/* Cache the current attribute. */
1993		a = ctx->attr;
1994		/* $MFT must be non-resident. */
1995		if (!a->non_resident) {
1996			ntfs_error(sb, "$MFT must be non-resident but a "
1997					"resident extent was found. $MFT is "
1998					"corrupt. Run chkdsk.");
1999			goto put_err_out;
2000		}
2001		/* $MFT must be uncompressed and unencrypted. */
2002		if (a->flags & ATTR_COMPRESSION_MASK ||
2003				a->flags & ATTR_IS_ENCRYPTED ||
2004				a->flags & ATTR_IS_SPARSE) {
2005			ntfs_error(sb, "$MFT must be uncompressed, "
2006					"non-sparse, and unencrypted but a "
2007					"compressed/sparse/encrypted extent "
2008					"was found. $MFT is corrupt. Run "
2009					"chkdsk.");
2010			goto put_err_out;
2011		}
2012		/*
2013		 * Decompress the mapping pairs array of this extent and merge
2014		 * the result into the existing runlist. No need for locking
2015		 * as we have exclusive access to the inode at this time and we
2016		 * are a mount in progress task, too.
2017		 */
2018		nrl = ntfs_mapping_pairs_decompress(vol, a, ni->runlist.rl);
2019		if (IS_ERR(nrl)) {
2020			ntfs_error(sb, "ntfs_mapping_pairs_decompress() "
2021					"failed with error code %ld.  $MFT is "
2022					"corrupt.", PTR_ERR(nrl));
2023			goto put_err_out;
2024		}
2025		ni->runlist.rl = nrl;
2026
2027		/* Are we in the first extent? */
2028		if (!next_vcn) {
2029			if (a->data.non_resident.lowest_vcn) {
2030				ntfs_error(sb, "First extent of $DATA "
2031						"attribute has non zero "
2032						"lowest_vcn. $MFT is corrupt. "
2033						"You should run chkdsk.");
2034				goto put_err_out;
2035			}
2036			/* Get the last vcn in the $DATA attribute. */
2037			last_vcn = sle64_to_cpu(
2038					a->data.non_resident.allocated_size)
2039					>> vol->cluster_size_bits;
2040			/* Fill in the inode size. */
2041			vi->i_size = sle64_to_cpu(
2042					a->data.non_resident.data_size);
2043			ni->initialized_size = sle64_to_cpu(
2044					a->data.non_resident.initialized_size);
2045			ni->allocated_size = sle64_to_cpu(
2046					a->data.non_resident.allocated_size);
2047			/*
2048			 * Verify the number of mft records does not exceed
2049			 * 2^32 - 1.
2050			 */
2051			if ((vi->i_size >> vol->mft_record_size_bits) >=
2052					(1ULL << 32)) {
2053				ntfs_error(sb, "$MFT is too big! Aborting.");
2054				goto put_err_out;
2055			}
2056			/*
2057			 * We have got the first extent of the runlist for
2058			 * $MFT which means it is now relatively safe to call
2059			 * the normal ntfs_read_inode() function.
2060			 * Complete reading the inode, this will actually
2061			 * re-read the mft record for $MFT, this time entering
2062			 * it into the page cache with which we complete the
2063			 * kick start of the volume. It should be safe to do
2064			 * this now as the first extent of $MFT/$DATA is
2065			 * already known and we would hope that we don't need
2066			 * further extents in order to find the other
2067			 * attributes belonging to $MFT. Only time will tell if
2068			 * this is really the case. If not we will have to play
2069			 * magic at this point, possibly duplicating a lot of
2070			 * ntfs_read_inode() at this point. We will need to
2071			 * ensure we do enough of its work to be able to call
2072			 * ntfs_read_inode() on extents of $MFT/$DATA. But lets
2073			 * hope this never happens...
2074			 */
2075			ntfs_read_locked_inode(vi);
2076			if (is_bad_inode(vi)) {
2077				ntfs_error(sb, "ntfs_read_inode() of $MFT "
2078						"failed. BUG or corrupt $MFT. "
2079						"Run chkdsk and if no errors "
2080						"are found, please report you "
2081						"saw this message to "
2082						"linux-ntfs-dev@lists."
2083						"sourceforge.net");
2084				ntfs_attr_put_search_ctx(ctx);
2085				/* Revert to the safe super operations. */
2086				ntfs_free(m);
2087				return -1;
2088			}
2089			/*
2090			 * Re-initialize some specifics about $MFT's inode as
2091			 * ntfs_read_inode() will have set up the default ones.
2092			 */
2093			/* Set uid and gid to root. */
2094			vi->i_uid = GLOBAL_ROOT_UID;
2095			vi->i_gid = GLOBAL_ROOT_GID;
2096			/* Regular file. No access for anyone. */
2097			vi->i_mode = S_IFREG;
2098			/* No VFS initiated operations allowed for $MFT. */
2099			vi->i_op = &ntfs_empty_inode_ops;
2100			vi->i_fop = &ntfs_empty_file_ops;
2101		}
2102
2103		/* Get the lowest vcn for the next extent. */
2104		highest_vcn = sle64_to_cpu(a->data.non_resident.highest_vcn);
2105		next_vcn = highest_vcn + 1;
2106
2107		/* Only one extent or error, which we catch below. */
2108		if (next_vcn <= 0)
2109			break;
2110
2111		/* Avoid endless loops due to corruption. */
2112		if (next_vcn < sle64_to_cpu(
2113				a->data.non_resident.lowest_vcn)) {
2114			ntfs_error(sb, "$MFT has corrupt attribute list "
2115					"attribute. Run chkdsk.");
2116			goto put_err_out;
2117		}
2118	}
2119	if (err != -ENOENT) {
2120		ntfs_error(sb, "Failed to lookup $MFT/$DATA attribute extent. "
2121				"$MFT is corrupt. Run chkdsk.");
2122		goto put_err_out;
2123	}
2124	if (!a) {
2125		ntfs_error(sb, "$MFT/$DATA attribute not found. $MFT is "
2126				"corrupt. Run chkdsk.");
2127		goto put_err_out;
2128	}
2129	if (highest_vcn && highest_vcn != last_vcn - 1) {
2130		ntfs_error(sb, "Failed to load the complete runlist for "
2131				"$MFT/$DATA. Driver bug or corrupt $MFT. "
2132				"Run chkdsk.");
2133		ntfs_debug("highest_vcn = 0x%llx, last_vcn - 1 = 0x%llx",
2134				(unsigned long long)highest_vcn,
2135				(unsigned long long)last_vcn - 1);
2136		goto put_err_out;
2137	}
2138	ntfs_attr_put_search_ctx(ctx);
2139	ntfs_debug("Done.");
2140	ntfs_free(m);
2141
2142	/*
2143	 * Split the locking rules of the MFT inode from the
2144	 * locking rules of other inodes:
2145	 */
2146	lockdep_set_class(&ni->runlist.lock, &mft_ni_runlist_lock_key);
2147	lockdep_set_class(&ni->mrec_lock, &mft_ni_mrec_lock_key);
2148
2149	return 0;
2150
2151em_put_err_out:
2152	ntfs_error(sb, "Couldn't find first extent of $DATA attribute in "
2153			"attribute list. $MFT is corrupt. Run chkdsk.");
2154put_err_out:
2155	ntfs_attr_put_search_ctx(ctx);
2156err_out:
2157	ntfs_error(sb, "Failed. Marking inode as bad.");
2158	make_bad_inode(vi);
2159	ntfs_free(m);
2160	return -1;
2161}
2162
2163static void __ntfs_clear_inode(ntfs_inode *ni)
2164{
2165	/* Free all alocated memory. */
2166	down_write(&ni->runlist.lock);
2167	if (ni->runlist.rl) {
2168		ntfs_free(ni->runlist.rl);
2169		ni->runlist.rl = NULL;
2170	}
2171	up_write(&ni->runlist.lock);
2172
2173	if (ni->attr_list) {
2174		ntfs_free(ni->attr_list);
2175		ni->attr_list = NULL;
2176	}
2177
2178	down_write(&ni->attr_list_rl.lock);
2179	if (ni->attr_list_rl.rl) {
2180		ntfs_free(ni->attr_list_rl.rl);
2181		ni->attr_list_rl.rl = NULL;
2182	}
2183	up_write(&ni->attr_list_rl.lock);
2184
2185	if (ni->name_len && ni->name != I30) {
2186		/* Catch bugs... */
2187		BUG_ON(!ni->name);
2188		kfree(ni->name);
2189	}
2190}
2191
2192void ntfs_clear_extent_inode(ntfs_inode *ni)
2193{
2194	ntfs_debug("Entering for inode 0x%lx.", ni->mft_no);
2195
2196	BUG_ON(NInoAttr(ni));
2197	BUG_ON(ni->nr_extents != -1);
2198
2199#ifdef NTFS_RW
2200	if (NInoDirty(ni)) {
2201		if (!is_bad_inode(VFS_I(ni->ext.base_ntfs_ino)))
2202			ntfs_error(ni->vol->sb, "Clearing dirty extent inode!  "
2203					"Losing data!  This is a BUG!!!");
2204		// FIXME:  Do something!!!
2205	}
2206#endif /* NTFS_RW */
2207
2208	__ntfs_clear_inode(ni);
2209
2210	/* Bye, bye... */
2211	ntfs_destroy_extent_inode(ni);
2212}
2213
2214/**
2215 * ntfs_evict_big_inode - clean up the ntfs specific part of an inode
2216 * @vi:		vfs inode pending annihilation
2217 *
2218 * When the VFS is going to remove an inode from memory, ntfs_clear_big_inode()
2219 * is called, which deallocates all memory belonging to the NTFS specific part
2220 * of the inode and returns.
2221 *
2222 * If the MFT record is dirty, we commit it before doing anything else.
2223 */
2224void ntfs_evict_big_inode(struct inode *vi)
2225{
2226	ntfs_inode *ni = NTFS_I(vi);
2227
2228	truncate_inode_pages_final(&vi->i_data);
2229	clear_inode(vi);
2230
2231#ifdef NTFS_RW
2232	if (NInoDirty(ni)) {
2233		bool was_bad = (is_bad_inode(vi));
2234
2235		/* Committing the inode also commits all extent inodes. */
2236		ntfs_commit_inode(vi);
2237
2238		if (!was_bad && (is_bad_inode(vi) || NInoDirty(ni))) {
2239			ntfs_error(vi->i_sb, "Failed to commit dirty inode "
2240					"0x%lx.  Losing data!", vi->i_ino);
2241			// FIXME:  Do something!!!
2242		}
2243	}
2244#endif /* NTFS_RW */
2245
2246	/* No need to lock at this stage as no one else has a reference. */
2247	if (ni->nr_extents > 0) {
2248		int i;
2249
2250		for (i = 0; i < ni->nr_extents; i++)
2251			ntfs_clear_extent_inode(ni->ext.extent_ntfs_inos[i]);
2252		kfree(ni->ext.extent_ntfs_inos);
2253	}
2254
2255	__ntfs_clear_inode(ni);
2256
2257	if (NInoAttr(ni)) {
2258		/* Release the base inode if we are holding it. */
2259		if (ni->nr_extents == -1) {
2260			iput(VFS_I(ni->ext.base_ntfs_ino));
2261			ni->nr_extents = 0;
2262			ni->ext.base_ntfs_ino = NULL;
2263		}
2264	}
2265	BUG_ON(ni->page);
2266	if (!atomic_dec_and_test(&ni->count))
2267		BUG();
2268	return;
2269}
2270
2271/**
2272 * ntfs_show_options - show mount options in /proc/mounts
2273 * @sf:		seq_file in which to write our mount options
2274 * @root:	root of the mounted tree whose mount options to display
2275 *
2276 * Called by the VFS once for each mounted ntfs volume when someone reads
2277 * /proc/mounts in order to display the NTFS specific mount options of each
2278 * mount. The mount options of fs specified by @root are written to the seq file
2279 * @sf and success is returned.
2280 */
2281int ntfs_show_options(struct seq_file *sf, struct dentry *root)
2282{
2283	ntfs_volume *vol = NTFS_SB(root->d_sb);
2284	int i;
2285
2286	seq_printf(sf, ",uid=%i", from_kuid_munged(&init_user_ns, vol->uid));
2287	seq_printf(sf, ",gid=%i", from_kgid_munged(&init_user_ns, vol->gid));
2288	if (vol->fmask == vol->dmask)
2289		seq_printf(sf, ",umask=0%o", vol->fmask);
2290	else {
2291		seq_printf(sf, ",fmask=0%o", vol->fmask);
2292		seq_printf(sf, ",dmask=0%o", vol->dmask);
2293	}
2294	seq_printf(sf, ",nls=%s", vol->nls_map->charset);
2295	if (NVolCaseSensitive(vol))
2296		seq_printf(sf, ",case_sensitive");
2297	if (NVolShowSystemFiles(vol))
2298		seq_printf(sf, ",show_sys_files");
2299	if (!NVolSparseEnabled(vol))
2300		seq_printf(sf, ",disable_sparse");
2301	for (i = 0; on_errors_arr[i].val; i++) {
2302		if (on_errors_arr[i].val & vol->on_errors)
2303			seq_printf(sf, ",errors=%s", on_errors_arr[i].str);
2304	}
2305	seq_printf(sf, ",mft_zone_multiplier=%i", vol->mft_zone_multiplier);
2306	return 0;
2307}
2308
2309#ifdef NTFS_RW
2310
2311static const char *es = "  Leaving inconsistent metadata.  Unmount and run "
2312		"chkdsk.";
2313
2314/**
2315 * ntfs_truncate - called when the i_size of an ntfs inode is changed
2316 * @vi:		inode for which the i_size was changed
2317 *
2318 * We only support i_size changes for normal files at present, i.e. not
2319 * compressed and not encrypted.  This is enforced in ntfs_setattr(), see
2320 * below.
2321 *
2322 * The kernel guarantees that @vi is a regular file (S_ISREG() is true) and
2323 * that the change is allowed.
2324 *
2325 * This implies for us that @vi is a file inode rather than a directory, index,
2326 * or attribute inode as well as that @vi is a base inode.
2327 *
2328 * Returns 0 on success or -errno on error.
2329 *
2330 * Called with ->i_mutex held.
2331 */
2332int ntfs_truncate(struct inode *vi)
2333{
2334	s64 new_size, old_size, nr_freed, new_alloc_size, old_alloc_size;
2335	VCN highest_vcn;
2336	unsigned long flags;
2337	ntfs_inode *base_ni, *ni = NTFS_I(vi);
2338	ntfs_volume *vol = ni->vol;
2339	ntfs_attr_search_ctx *ctx;
2340	MFT_RECORD *m;
2341	ATTR_RECORD *a;
2342	const char *te = "  Leaving file length out of sync with i_size.";
2343	int err, mp_size, size_change, alloc_change;
2344	u32 attr_len;
2345
2346	ntfs_debug("Entering for inode 0x%lx.", vi->i_ino);
2347	BUG_ON(NInoAttr(ni));
2348	BUG_ON(S_ISDIR(vi->i_mode));
2349	BUG_ON(NInoMstProtected(ni));
2350	BUG_ON(ni->nr_extents < 0);
2351retry_truncate:
2352	/*
2353	 * Lock the runlist for writing and map the mft record to ensure it is
2354	 * safe to mess with the attribute runlist and sizes.
2355	 */
2356	down_write(&ni->runlist.lock);
2357	if (!NInoAttr(ni))
2358		base_ni = ni;
2359	else
2360		base_ni = ni->ext.base_ntfs_ino;
2361	m = map_mft_record(base_ni);
2362	if (IS_ERR(m)) {
2363		err = PTR_ERR(m);
2364		ntfs_error(vi->i_sb, "Failed to map mft record for inode 0x%lx "
2365				"(error code %d).%s", vi->i_ino, err, te);
2366		ctx = NULL;
2367		m = NULL;
2368		goto old_bad_out;
2369	}
2370	ctx = ntfs_attr_get_search_ctx(base_ni, m);
2371	if (unlikely(!ctx)) {
2372		ntfs_error(vi->i_sb, "Failed to allocate a search context for "
2373				"inode 0x%lx (not enough memory).%s",
2374				vi->i_ino, te);
2375		err = -ENOMEM;
2376		goto old_bad_out;
2377	}
2378	err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len,
2379			CASE_SENSITIVE, 0, NULL, 0, ctx);
2380	if (unlikely(err)) {
2381		if (err == -ENOENT) {
2382			ntfs_error(vi->i_sb, "Open attribute is missing from "
2383					"mft record.  Inode 0x%lx is corrupt.  "
2384					"Run chkdsk.%s", vi->i_ino, te);
2385			err = -EIO;
2386		} else
2387			ntfs_error(vi->i_sb, "Failed to lookup attribute in "
2388					"inode 0x%lx (error code %d).%s",
2389					vi->i_ino, err, te);
2390		goto old_bad_out;
2391	}
2392	m = ctx->mrec;
2393	a = ctx->attr;
2394	/*
2395	 * The i_size of the vfs inode is the new size for the attribute value.
2396	 */
2397	new_size = i_size_read(vi);
2398	/* The current size of the attribute value is the old size. */
2399	old_size = ntfs_attr_size(a);
2400	/* Calculate the new allocated size. */
2401	if (NInoNonResident(ni))
2402		new_alloc_size = (new_size + vol->cluster_size - 1) &
2403				~(s64)vol->cluster_size_mask;
2404	else
2405		new_alloc_size = (new_size + 7) & ~7;
2406	/* The current allocated size is the old allocated size. */
2407	read_lock_irqsave(&ni->size_lock, flags);
2408	old_alloc_size = ni->allocated_size;
2409	read_unlock_irqrestore(&ni->size_lock, flags);
2410	/*
2411	 * The change in the file size.  This will be 0 if no change, >0 if the
2412	 * size is growing, and <0 if the size is shrinking.
2413	 */
2414	size_change = -1;
2415	if (new_size - old_size >= 0) {
2416		size_change = 1;
2417		if (new_size == old_size)
2418			size_change = 0;
2419	}
2420	/* As above for the allocated size. */
2421	alloc_change = -1;
2422	if (new_alloc_size - old_alloc_size >= 0) {
2423		alloc_change = 1;
2424		if (new_alloc_size == old_alloc_size)
2425			alloc_change = 0;
2426	}
2427	/*
2428	 * If neither the size nor the allocation are being changed there is
2429	 * nothing to do.
2430	 */
2431	if (!size_change && !alloc_change)
2432		goto unm_done;
2433	/* If the size is changing, check if new size is allowed in $AttrDef. */
2434	if (size_change) {
2435		err = ntfs_attr_size_bounds_check(vol, ni->type, new_size);
2436		if (unlikely(err)) {
2437			if (err == -ERANGE) {
2438				ntfs_error(vol->sb, "Truncate would cause the "
2439						"inode 0x%lx to %simum size "
2440						"for its attribute type "
2441						"(0x%x).  Aborting truncate.",
2442						vi->i_ino,
2443						new_size > old_size ? "exceed "
2444						"the max" : "go under the min",
2445						le32_to_cpu(ni->type));
2446				err = -EFBIG;
2447			} else {
2448				ntfs_error(vol->sb, "Inode 0x%lx has unknown "
2449						"attribute type 0x%x.  "
2450						"Aborting truncate.",
2451						vi->i_ino,
2452						le32_to_cpu(ni->type));
2453				err = -EIO;
2454			}
2455			/* Reset the vfs inode size to the old size. */
2456			i_size_write(vi, old_size);
2457			goto err_out;
2458		}
2459	}
2460	if (NInoCompressed(ni) || NInoEncrypted(ni)) {
2461		ntfs_warning(vi->i_sb, "Changes in inode size are not "
2462				"supported yet for %s files, ignoring.",
2463				NInoCompressed(ni) ? "compressed" :
2464				"encrypted");
2465		err = -EOPNOTSUPP;
2466		goto bad_out;
2467	}
2468	if (a->non_resident)
2469		goto do_non_resident_truncate;
2470	BUG_ON(NInoNonResident(ni));
2471	/* Resize the attribute record to best fit the new attribute size. */
2472	if (new_size < vol->mft_record_size &&
2473			!ntfs_resident_attr_value_resize(m, a, new_size)) {
2474		/* The resize succeeded! */
2475		flush_dcache_mft_record_page(ctx->ntfs_ino);
2476		mark_mft_record_dirty(ctx->ntfs_ino);
2477		write_lock_irqsave(&ni->size_lock, flags);
2478		/* Update the sizes in the ntfs inode and all is done. */
2479		ni->allocated_size = le32_to_cpu(a->length) -
2480				le16_to_cpu(a->data.resident.value_offset);
2481		/*
2482		 * Note ntfs_resident_attr_value_resize() has already done any
2483		 * necessary data clearing in the attribute record.  When the
2484		 * file is being shrunk vmtruncate() will already have cleared
2485		 * the top part of the last partial page, i.e. since this is
2486		 * the resident case this is the page with index 0.  However,
2487		 * when the file is being expanded, the page cache page data
2488		 * between the old data_size, i.e. old_size, and the new_size
2489		 * has not been zeroed.  Fortunately, we do not need to zero it
2490		 * either since on one hand it will either already be zero due
2491		 * to both readpage and writepage clearing partial page data
2492		 * beyond i_size in which case there is nothing to do or in the
2493		 * case of the file being mmap()ped at the same time, POSIX
2494		 * specifies that the behaviour is unspecified thus we do not
2495		 * have to do anything.  This means that in our implementation
2496		 * in the rare case that the file is mmap()ped and a write
2497		 * occurred into the mmap()ped region just beyond the file size
2498		 * and writepage has not yet been called to write out the page
2499		 * (which would clear the area beyond the file size) and we now
2500		 * extend the file size to incorporate this dirty region
2501		 * outside the file size, a write of the page would result in
2502		 * this data being written to disk instead of being cleared.
2503		 * Given both POSIX and the Linux mmap(2) man page specify that
2504		 * this corner case is undefined, we choose to leave it like
2505		 * that as this is much simpler for us as we cannot lock the
2506		 * relevant page now since we are holding too many ntfs locks
2507		 * which would result in a lock reversal deadlock.
2508		 */
2509		ni->initialized_size = new_size;
2510		write_unlock_irqrestore(&ni->size_lock, flags);
2511		goto unm_done;
2512	}
2513	/* If the above resize failed, this must be an attribute extension. */
2514	BUG_ON(size_change < 0);
2515	/*
2516	 * We have to drop all the locks so we can call
2517	 * ntfs_attr_make_non_resident().  This could be optimised by try-
2518	 * locking the first page cache page and only if that fails dropping
2519	 * the locks, locking the page, and redoing all the locking and
2520	 * lookups.  While this would be a huge optimisation, it is not worth
2521	 * it as this is definitely a slow code path as it only ever can happen
2522	 * once for any given file.
2523	 */
2524	ntfs_attr_put_search_ctx(ctx);
2525	unmap_mft_record(base_ni);
2526	up_write(&ni->runlist.lock);
2527	/*
2528	 * Not enough space in the mft record, try to make the attribute
2529	 * non-resident and if successful restart the truncation process.
2530	 */
2531	err = ntfs_attr_make_non_resident(ni, old_size);
2532	if (likely(!err))
2533		goto retry_truncate;
2534	/*
2535	 * Could not make non-resident.  If this is due to this not being
2536	 * permitted for this attribute type or there not being enough space,
2537	 * try to make other attributes non-resident.  Otherwise fail.
2538	 */
2539	if (unlikely(err != -EPERM && err != -ENOSPC)) {
2540		ntfs_error(vol->sb, "Cannot truncate inode 0x%lx, attribute "
2541				"type 0x%x, because the conversion from "
2542				"resident to non-resident attribute failed "
2543				"with error code %i.", vi->i_ino,
2544				(unsigned)le32_to_cpu(ni->type), err);
2545		if (err != -ENOMEM)
2546			err = -EIO;
2547		goto conv_err_out;
2548	}
2549	/* TODO: Not implemented from here, abort. */
2550	if (err == -ENOSPC)
2551		ntfs_error(vol->sb, "Not enough space in the mft record/on "
2552				"disk for the non-resident attribute value.  "
2553				"This case is not implemented yet.");
2554	else /* if (err == -EPERM) */
2555		ntfs_error(vol->sb, "This attribute type may not be "
2556				"non-resident.  This case is not implemented "
2557				"yet.");
2558	err = -EOPNOTSUPP;
2559	goto conv_err_out;
2560#if 0
2561	// TODO: Attempt to make other attributes non-resident.
2562	if (!err)
2563		goto do_resident_extend;
2564	/*
2565	 * Both the attribute list attribute and the standard information
2566	 * attribute must remain in the base inode.  Thus, if this is one of
2567	 * these attributes, we have to try to move other attributes out into
2568	 * extent mft records instead.
2569	 */
2570	if (ni->type == AT_ATTRIBUTE_LIST ||
2571			ni->type == AT_STANDARD_INFORMATION) {
2572		// TODO: Attempt to move other attributes into extent mft
2573		// records.
2574		err = -EOPNOTSUPP;
2575		if (!err)
2576			goto do_resident_extend;
2577		goto err_out;
2578	}
2579	// TODO: Attempt to move this attribute to an extent mft record, but
2580	// only if it is not already the only attribute in an mft record in
2581	// which case there would be nothing to gain.
2582	err = -EOPNOTSUPP;
2583	if (!err)
2584		goto do_resident_extend;
2585	/* There is nothing we can do to make enough space. )-: */
2586	goto err_out;
2587#endif
2588do_non_resident_truncate:
2589	BUG_ON(!NInoNonResident(ni));
2590	if (alloc_change < 0) {
2591		highest_vcn = sle64_to_cpu(a->data.non_resident.highest_vcn);
2592		if (highest_vcn > 0 &&
2593				old_alloc_size >> vol->cluster_size_bits >
2594				highest_vcn + 1) {
2595			/*
2596			 * This attribute has multiple extents.  Not yet
2597			 * supported.
2598			 */
2599			ntfs_error(vol->sb, "Cannot truncate inode 0x%lx, "
2600					"attribute type 0x%x, because the "
2601					"attribute is highly fragmented (it "
2602					"consists of multiple extents) and "
2603					"this case is not implemented yet.",
2604					vi->i_ino,
2605					(unsigned)le32_to_cpu(ni->type));
2606			err = -EOPNOTSUPP;
2607			goto bad_out;
2608		}
2609	}
2610	/*
2611	 * If the size is shrinking, need to reduce the initialized_size and
2612	 * the data_size before reducing the allocation.
2613	 */
2614	if (size_change < 0) {
2615		/*
2616		 * Make the valid size smaller (i_size is already up-to-date).
2617		 */
2618		write_lock_irqsave(&ni->size_lock, flags);
2619		if (new_size < ni->initialized_size) {
2620			ni->initialized_size = new_size;
2621			a->data.non_resident.initialized_size =
2622					cpu_to_sle64(new_size);
2623		}
2624		a->data.non_resident.data_size = cpu_to_sle64(new_size);
2625		write_unlock_irqrestore(&ni->size_lock, flags);
2626		flush_dcache_mft_record_page(ctx->ntfs_ino);
2627		mark_mft_record_dirty(ctx->ntfs_ino);
2628		/* If the allocated size is not changing, we are done. */
2629		if (!alloc_change)
2630			goto unm_done;
2631		/*
2632		 * If the size is shrinking it makes no sense for the
2633		 * allocation to be growing.
2634		 */
2635		BUG_ON(alloc_change > 0);
2636	} else /* if (size_change >= 0) */ {
2637		/*
2638		 * The file size is growing or staying the same but the
2639		 * allocation can be shrinking, growing or staying the same.
2640		 */
2641		if (alloc_change > 0) {
2642			/*
2643			 * We need to extend the allocation and possibly update
2644			 * the data size.  If we are updating the data size,
2645			 * since we are not touching the initialized_size we do
2646			 * not need to worry about the actual data on disk.
2647			 * And as far as the page cache is concerned, there
2648			 * will be no pages beyond the old data size and any
2649			 * partial region in the last page between the old and
2650			 * new data size (or the end of the page if the new
2651			 * data size is outside the page) does not need to be
2652			 * modified as explained above for the resident
2653			 * attribute truncate case.  To do this, we simply drop
2654			 * the locks we hold and leave all the work to our
2655			 * friendly helper ntfs_attr_extend_allocation().
2656			 */
2657			ntfs_attr_put_search_ctx(ctx);
2658			unmap_mft_record(base_ni);
2659			up_write(&ni->runlist.lock);
2660			err = ntfs_attr_extend_allocation(ni, new_size,
2661					size_change > 0 ? new_size : -1, -1);
2662			/*
2663			 * ntfs_attr_extend_allocation() will have done error
2664			 * output already.
2665			 */
2666			goto done;
2667		}
2668		if (!alloc_change)
2669			goto alloc_done;
2670	}
2671	/* alloc_change < 0 */
2672	/* Free the clusters. */
2673	nr_freed = ntfs_cluster_free(ni, new_alloc_size >>
2674			vol->cluster_size_bits, -1, ctx);
2675	m = ctx->mrec;
2676	a = ctx->attr;
2677	if (unlikely(nr_freed < 0)) {
2678		ntfs_error(vol->sb, "Failed to release cluster(s) (error code "
2679				"%lli).  Unmount and run chkdsk to recover "
2680				"the lost cluster(s).", (long long)nr_freed);
2681		NVolSetErrors(vol);
2682		nr_freed = 0;
2683	}
2684	/* Truncate the runlist. */
2685	err = ntfs_rl_truncate_nolock(vol, &ni->runlist,
2686			new_alloc_size >> vol->cluster_size_bits);
2687	/*
2688	 * If the runlist truncation failed and/or the search context is no
2689	 * longer valid, we cannot resize the attribute record or build the
2690	 * mapping pairs array thus we mark the inode bad so that no access to
2691	 * the freed clusters can happen.
2692	 */
2693	if (unlikely(err || IS_ERR(m))) {
2694		ntfs_error(vol->sb, "Failed to %s (error code %li).%s",
2695				IS_ERR(m) ?
2696				"restore attribute search context" :
2697				"truncate attribute runlist",
2698				IS_ERR(m) ? PTR_ERR(m) : err, es);
2699		err = -EIO;
2700		goto bad_out;
2701	}
2702	/* Get the size for the shrunk mapping pairs array for the runlist. */
2703	mp_size = ntfs_get_size_for_mapping_pairs(vol, ni->runlist.rl, 0, -1);
2704	if (unlikely(mp_size <= 0)) {
2705		ntfs_error(vol->sb, "Cannot shrink allocation of inode 0x%lx, "
2706				"attribute type 0x%x, because determining the "
2707				"size for the mapping pairs failed with error "
2708				"code %i.%s", vi->i_ino,
2709				(unsigned)le32_to_cpu(ni->type), mp_size, es);
2710		err = -EIO;
2711		goto bad_out;
2712	}
2713	/*
2714	 * Shrink the attribute record for the new mapping pairs array.  Note,
2715	 * this cannot fail since we are making the attribute smaller thus by
2716	 * definition there is enough space to do so.
2717	 */
2718	attr_len = le32_to_cpu(a->length);
2719	err = ntfs_attr_record_resize(m, a, mp_size +
2720			le16_to_cpu(a->data.non_resident.mapping_pairs_offset));
2721	BUG_ON(err);
2722	/*
2723	 * Generate the mapping pairs array directly into the attribute record.
2724	 */
2725	err = ntfs_mapping_pairs_build(vol, (u8*)a +
2726			le16_to_cpu(a->data.non_resident.mapping_pairs_offset),
2727			mp_size, ni->runlist.rl, 0, -1, NULL);
2728	if (unlikely(err)) {
2729		ntfs_error(vol->sb, "Cannot shrink allocation of inode 0x%lx, "
2730				"attribute type 0x%x, because building the "
2731				"mapping pairs failed with error code %i.%s",
2732				vi->i_ino, (unsigned)le32_to_cpu(ni->type),
2733				err, es);
2734		err = -EIO;
2735		goto bad_out;
2736	}
2737	/* Update the allocated/compressed size as well as the highest vcn. */
2738	a->data.non_resident.highest_vcn = cpu_to_sle64((new_alloc_size >>
2739			vol->cluster_size_bits) - 1);
2740	write_lock_irqsave(&ni->size_lock, flags);
2741	ni->allocated_size = new_alloc_size;
2742	a->data.non_resident.allocated_size = cpu_to_sle64(new_alloc_size);
2743	if (NInoSparse(ni) || NInoCompressed(ni)) {
2744		if (nr_freed) {
2745			ni->itype.compressed.size -= nr_freed <<
2746					vol->cluster_size_bits;
2747			BUG_ON(ni->itype.compressed.size < 0);
2748			a->data.non_resident.compressed_size = cpu_to_sle64(
2749					ni->itype.compressed.size);
2750			vi->i_blocks = ni->itype.compressed.size >> 9;
2751		}
2752	} else
2753		vi->i_blocks = new_alloc_size >> 9;
2754	write_unlock_irqrestore(&ni->size_lock, flags);
2755	/*
2756	 * We have shrunk the allocation.  If this is a shrinking truncate we
2757	 * have already dealt with the initialized_size and the data_size above
2758	 * and we are done.  If the truncate is only changing the allocation
2759	 * and not the data_size, we are also done.  If this is an extending
2760	 * truncate, need to extend the data_size now which is ensured by the
2761	 * fact that @size_change is positive.
2762	 */
2763alloc_done:
2764	/*
2765	 * If the size is growing, need to update it now.  If it is shrinking,
2766	 * we have already updated it above (before the allocation change).
2767	 */
2768	if (size_change > 0)
2769		a->data.non_resident.data_size = cpu_to_sle64(new_size);
2770	/* Ensure the modified mft record is written out. */
2771	flush_dcache_mft_record_page(ctx->ntfs_ino);
2772	mark_mft_record_dirty(ctx->ntfs_ino);
2773unm_done:
2774	ntfs_attr_put_search_ctx(ctx);
2775	unmap_mft_record(base_ni);
2776	up_write(&ni->runlist.lock);
2777done:
2778	/* Update the mtime and ctime on the base inode. */
2779	/* normally ->truncate shouldn't update ctime or mtime,
2780	 * but ntfs did before so it got a copy & paste version
2781	 * of file_update_time.  one day someone should fix this
2782	 * for real.
2783	 */
2784	if (!IS_NOCMTIME(VFS_I(base_ni)) && !IS_RDONLY(VFS_I(base_ni))) {
2785		struct timespec64 now = current_time(VFS_I(base_ni));
2786		int sync_it = 0;
2787
2788		if (!timespec64_equal(&VFS_I(base_ni)->i_mtime, &now) ||
2789		    !timespec64_equal(&VFS_I(base_ni)->i_ctime, &now))
2790			sync_it = 1;
2791		VFS_I(base_ni)->i_mtime = now;
2792		VFS_I(base_ni)->i_ctime = now;
2793
2794		if (sync_it)
2795			mark_inode_dirty_sync(VFS_I(base_ni));
2796	}
2797
2798	if (likely(!err)) {
2799		NInoClearTruncateFailed(ni);
2800		ntfs_debug("Done.");
2801	}
2802	return err;
2803old_bad_out:
2804	old_size = -1;
2805bad_out:
2806	if (err != -ENOMEM && err != -EOPNOTSUPP)
2807		NVolSetErrors(vol);
2808	if (err != -EOPNOTSUPP)
2809		NInoSetTruncateFailed(ni);
2810	else if (old_size >= 0)
2811		i_size_write(vi, old_size);
2812err_out:
2813	if (ctx)
2814		ntfs_attr_put_search_ctx(ctx);
2815	if (m)
2816		unmap_mft_record(base_ni);
2817	up_write(&ni->runlist.lock);
2818out:
2819	ntfs_debug("Failed.  Returning error code %i.", err);
2820	return err;
2821conv_err_out:
2822	if (err != -ENOMEM && err != -EOPNOTSUPP)
2823		NVolSetErrors(vol);
2824	if (err != -EOPNOTSUPP)
2825		NInoSetTruncateFailed(ni);
2826	else
2827		i_size_write(vi, old_size);
2828	goto out;
2829}
2830
2831/**
2832 * ntfs_truncate_vfs - wrapper for ntfs_truncate() that has no return value
2833 * @vi:		inode for which the i_size was changed
2834 *
2835 * Wrapper for ntfs_truncate() that has no return value.
2836 *
2837 * See ntfs_truncate() description above for details.
2838 */
2839#ifdef NTFS_RW
2840void ntfs_truncate_vfs(struct inode *vi) {
2841	ntfs_truncate(vi);
2842}
2843#endif
2844
2845/**
2846 * ntfs_setattr - called from notify_change() when an attribute is being changed
2847 * @dentry:	dentry whose attributes to change
2848 * @attr:	structure describing the attributes and the changes
2849 *
2850 * We have to trap VFS attempts to truncate the file described by @dentry as
2851 * soon as possible, because we do not implement changes in i_size yet.  So we
2852 * abort all i_size changes here.
2853 *
2854 * We also abort all changes of user, group, and mode as we do not implement
2855 * the NTFS ACLs yet.
2856 *
2857 * Called with ->i_mutex held.
2858 */
2859int ntfs_setattr(struct dentry *dentry, struct iattr *attr)
2860{
2861	struct inode *vi = d_inode(dentry);
2862	int err;
2863	unsigned int ia_valid = attr->ia_valid;
2864
2865	err = setattr_prepare(dentry, attr);
2866	if (err)
2867		goto out;
2868	/* We do not support NTFS ACLs yet. */
2869	if (ia_valid & (ATTR_UID | ATTR_GID | ATTR_MODE)) {
2870		ntfs_warning(vi->i_sb, "Changes in user/group/mode are not "
2871				"supported yet, ignoring.");
2872		err = -EOPNOTSUPP;
2873		goto out;
2874	}
2875	if (ia_valid & ATTR_SIZE) {
2876		if (attr->ia_size != i_size_read(vi)) {
2877			ntfs_inode *ni = NTFS_I(vi);
2878			/*
2879			 * FIXME: For now we do not support resizing of
2880			 * compressed or encrypted files yet.
2881			 */
2882			if (NInoCompressed(ni) || NInoEncrypted(ni)) {
2883				ntfs_warning(vi->i_sb, "Changes in inode size "
2884						"are not supported yet for "
2885						"%s files, ignoring.",
2886						NInoCompressed(ni) ?
2887						"compressed" : "encrypted");
2888				err = -EOPNOTSUPP;
2889			} else {
2890				truncate_setsize(vi, attr->ia_size);
2891				ntfs_truncate_vfs(vi);
2892			}
2893			if (err || ia_valid == ATTR_SIZE)
2894				goto out;
2895		} else {
2896			/*
2897			 * We skipped the truncate but must still update
2898			 * timestamps.
2899			 */
2900			ia_valid |= ATTR_MTIME | ATTR_CTIME;
2901		}
2902	}
2903	if (ia_valid & ATTR_ATIME)
2904		vi->i_atime = attr->ia_atime;
 
2905	if (ia_valid & ATTR_MTIME)
2906		vi->i_mtime = attr->ia_mtime;
 
2907	if (ia_valid & ATTR_CTIME)
2908		vi->i_ctime = attr->ia_ctime;
 
2909	mark_inode_dirty(vi);
2910out:
2911	return err;
2912}
2913
2914/**
2915 * ntfs_write_inode - write out a dirty inode
2916 * @vi:		inode to write out
2917 * @sync:	if true, write out synchronously
2918 *
2919 * Write out a dirty inode to disk including any extent inodes if present.
2920 *
2921 * If @sync is true, commit the inode to disk and wait for io completion.  This
2922 * is done using write_mft_record().
2923 *
2924 * If @sync is false, just schedule the write to happen but do not wait for i/o
2925 * completion.  In 2.6 kernels, scheduling usually happens just by virtue of
2926 * marking the page (and in this case mft record) dirty but we do not implement
2927 * this yet as write_mft_record() largely ignores the @sync parameter and
2928 * always performs synchronous writes.
2929 *
2930 * Return 0 on success and -errno on error.
2931 */
2932int __ntfs_write_inode(struct inode *vi, int sync)
2933{
2934	sle64 nt;
2935	ntfs_inode *ni = NTFS_I(vi);
2936	ntfs_attr_search_ctx *ctx;
2937	MFT_RECORD *m;
2938	STANDARD_INFORMATION *si;
2939	int err = 0;
2940	bool modified = false;
2941
2942	ntfs_debug("Entering for %sinode 0x%lx.", NInoAttr(ni) ? "attr " : "",
2943			vi->i_ino);
2944	/*
2945	 * Dirty attribute inodes are written via their real inodes so just
2946	 * clean them here.  Access time updates are taken care off when the
2947	 * real inode is written.
2948	 */
2949	if (NInoAttr(ni)) {
2950		NInoClearDirty(ni);
2951		ntfs_debug("Done.");
2952		return 0;
2953	}
2954	/* Map, pin, and lock the mft record belonging to the inode. */
2955	m = map_mft_record(ni);
2956	if (IS_ERR(m)) {
2957		err = PTR_ERR(m);
2958		goto err_out;
2959	}
2960	/* Update the access times in the standard information attribute. */
2961	ctx = ntfs_attr_get_search_ctx(ni, m);
2962	if (unlikely(!ctx)) {
2963		err = -ENOMEM;
2964		goto unm_err_out;
2965	}
2966	err = ntfs_attr_lookup(AT_STANDARD_INFORMATION, NULL, 0,
2967			CASE_SENSITIVE, 0, NULL, 0, ctx);
2968	if (unlikely(err)) {
2969		ntfs_attr_put_search_ctx(ctx);
2970		goto unm_err_out;
2971	}
2972	si = (STANDARD_INFORMATION*)((u8*)ctx->attr +
2973			le16_to_cpu(ctx->attr->data.resident.value_offset));
2974	/* Update the access times if they have changed. */
2975	nt = utc2ntfs(vi->i_mtime);
2976	if (si->last_data_change_time != nt) {
2977		ntfs_debug("Updating mtime for inode 0x%lx: old = 0x%llx, "
2978				"new = 0x%llx", vi->i_ino, (long long)
2979				sle64_to_cpu(si->last_data_change_time),
2980				(long long)sle64_to_cpu(nt));
2981		si->last_data_change_time = nt;
2982		modified = true;
2983	}
2984	nt = utc2ntfs(vi->i_ctime);
2985	if (si->last_mft_change_time != nt) {
2986		ntfs_debug("Updating ctime for inode 0x%lx: old = 0x%llx, "
2987				"new = 0x%llx", vi->i_ino, (long long)
2988				sle64_to_cpu(si->last_mft_change_time),
2989				(long long)sle64_to_cpu(nt));
2990		si->last_mft_change_time = nt;
2991		modified = true;
2992	}
2993	nt = utc2ntfs(vi->i_atime);
2994	if (si->last_access_time != nt) {
2995		ntfs_debug("Updating atime for inode 0x%lx: old = 0x%llx, "
2996				"new = 0x%llx", vi->i_ino,
2997				(long long)sle64_to_cpu(si->last_access_time),
2998				(long long)sle64_to_cpu(nt));
2999		si->last_access_time = nt;
3000		modified = true;
3001	}
3002	/*
3003	 * If we just modified the standard information attribute we need to
3004	 * mark the mft record it is in dirty.  We do this manually so that
3005	 * mark_inode_dirty() is not called which would redirty the inode and
3006	 * hence result in an infinite loop of trying to write the inode.
3007	 * There is no need to mark the base inode nor the base mft record
3008	 * dirty, since we are going to write this mft record below in any case
3009	 * and the base mft record may actually not have been modified so it
3010	 * might not need to be written out.
3011	 * NOTE: It is not a problem when the inode for $MFT itself is being
3012	 * written out as mark_ntfs_record_dirty() will only set I_DIRTY_PAGES
3013	 * on the $MFT inode and hence ntfs_write_inode() will not be
3014	 * re-invoked because of it which in turn is ok since the dirtied mft
3015	 * record will be cleaned and written out to disk below, i.e. before
3016	 * this function returns.
3017	 */
3018	if (modified) {
3019		flush_dcache_mft_record_page(ctx->ntfs_ino);
3020		if (!NInoTestSetDirty(ctx->ntfs_ino))
3021			mark_ntfs_record_dirty(ctx->ntfs_ino->page,
3022					ctx->ntfs_ino->page_ofs);
3023	}
3024	ntfs_attr_put_search_ctx(ctx);
3025	/* Now the access times are updated, write the base mft record. */
3026	if (NInoDirty(ni))
3027		err = write_mft_record(ni, m, sync);
3028	/* Write all attached extent mft records. */
3029	mutex_lock(&ni->extent_lock);
3030	if (ni->nr_extents > 0) {
3031		ntfs_inode **extent_nis = ni->ext.extent_ntfs_inos;
3032		int i;
3033
3034		ntfs_debug("Writing %i extent inodes.", ni->nr_extents);
3035		for (i = 0; i < ni->nr_extents; i++) {
3036			ntfs_inode *tni = extent_nis[i];
3037
3038			if (NInoDirty(tni)) {
3039				MFT_RECORD *tm = map_mft_record(tni);
3040				int ret;
3041
3042				if (IS_ERR(tm)) {
3043					if (!err || err == -ENOMEM)
3044						err = PTR_ERR(tm);
3045					continue;
3046				}
3047				ret = write_mft_record(tni, tm, sync);
3048				unmap_mft_record(tni);
3049				if (unlikely(ret)) {
3050					if (!err || err == -ENOMEM)
3051						err = ret;
3052				}
3053			}
3054		}
3055	}
3056	mutex_unlock(&ni->extent_lock);
3057	unmap_mft_record(ni);
3058	if (unlikely(err))
3059		goto err_out;
3060	ntfs_debug("Done.");
3061	return 0;
3062unm_err_out:
3063	unmap_mft_record(ni);
3064err_out:
3065	if (err == -ENOMEM) {
3066		ntfs_warning(vi->i_sb, "Not enough memory to write inode.  "
3067				"Marking the inode dirty again, so the VFS "
3068				"retries later.");
3069		mark_inode_dirty(vi);
3070	} else {
3071		ntfs_error(vi->i_sb, "Failed (error %i):  Run chkdsk.", -err);
3072		NVolSetErrors(ni->vol);
3073	}
3074	return err;
3075}
3076
3077#endif /* NTFS_RW */