1// SPDX-License-Identifier: GPL-2.0-or-later
   2/*
   3 * file.c - NTFS kernel file operations.  Part of the Linux-NTFS project.
   4 *
   5 * Copyright (c) 2001-2015 Anton Altaparmakov and Tuxera Inc.
   6 */
   7
   8#include <linux/backing-dev.h>
   9#include <linux/buffer_head.h>
  10#include <linux/gfp.h>
  11#include <linux/pagemap.h>
  12#include <linux/pagevec.h>
  13#include <linux/sched/signal.h>
  14#include <linux/swap.h>
  15#include <linux/uio.h>
  16#include <linux/writeback.h>
  17
  18#include <asm/page.h>
  19#include <linux/uaccess.h>
  20
  21#include "attrib.h"
  22#include "bitmap.h"
  23#include "inode.h"
  24#include "debug.h"
  25#include "lcnalloc.h"
  26#include "malloc.h"
  27#include "mft.h"
  28#include "ntfs.h"
  29
  30/**
  31 * ntfs_file_open - called when an inode is about to be opened
  32 * @vi:		inode to be opened
  33 * @filp:	file structure describing the inode
  34 *
  35 * Limit file size to the page cache limit on architectures where unsigned long
  36 * is 32-bits. This is the most we can do for now without overflowing the page
  37 * cache page index. Doing it this way means we don't run into problems because
  38 * of existing too large files. It would be better to allow the user to read
  39 * the beginning of the file but I doubt very much anyone is going to hit this
  40 * check on a 32-bit architecture, so there is no point in adding the extra
  41 * complexity required to support this.
  42 *
  43 * On 64-bit architectures, the check is hopefully optimized away by the
  44 * compiler.
  45 *
  46 * After the check passes, just call generic_file_open() to do its work.
  47 */
  48static int ntfs_file_open(struct inode *vi, struct file *filp)
  49{
  50	if (sizeof(unsigned long) < 8) {
  51		if (i_size_read(vi) > MAX_LFS_FILESIZE)
  52			return -EOVERFLOW;
  53	}
  54	return generic_file_open(vi, filp);
  55}
  56
  57#ifdef NTFS_RW
  58
  59/**
  60 * ntfs_attr_extend_initialized - extend the initialized size of an attribute
  61 * @ni:			ntfs inode of the attribute to extend
  62 * @new_init_size:	requested new initialized size in bytes
  63 *
  64 * Extend the initialized size of an attribute described by the ntfs inode @ni
  65 * to @new_init_size bytes.  This involves zeroing any non-sparse space between
  66 * the old initialized size and @new_init_size both in the page cache and on
  67 * disk (if relevant complete pages are already uptodate in the page cache then
  68 * these are simply marked dirty).
  69 *
  70 * As a side-effect, the file size (vfs inode->i_size) may be incremented as,
  71 * in the resident attribute case, it is tied to the initialized size and, in
  72 * the non-resident attribute case, it may not fall below the initialized size.
  73 *
  74 * Note that if the attribute is resident, we do not need to touch the page
  75 * cache at all.  This is because if the page cache page is not uptodate we
  76 * bring it uptodate later, when doing the write to the mft record since we
  77 * then already have the page mapped.  And if the page is uptodate, the
  78 * non-initialized region will already have been zeroed when the page was
  79 * brought uptodate and the region may in fact already have been overwritten
  80 * with new data via mmap() based writes, so we cannot just zero it.  And since
  81 * POSIX specifies that the behaviour of resizing a file whilst it is mmap()ped
  82 * is unspecified, we choose not to do zeroing and thus we do not need to touch
  83 * the page at all.  For a more detailed explanation see ntfs_truncate() in
  84 * fs/ntfs/inode.c.
  85 *
  86 * Return 0 on success and -errno on error.  In the case that an error is
  87 * encountered it is possible that the initialized size will already have been
  88 * incremented some way towards @new_init_size but it is guaranteed that if
  89 * this is the case, the necessary zeroing will also have happened and that all
  90 * metadata is self-consistent.
  91 *
  92 * Locking: i_mutex on the vfs inode corrseponsind to the ntfs inode @ni must be
  93 *	    held by the caller.
  94 */
  95static int ntfs_attr_extend_initialized(ntfs_inode *ni, const s64 new_init_size)
  96{
  97	s64 old_init_size;
  98	loff_t old_i_size;
  99	pgoff_t index, end_index;
 100	unsigned long flags;
 101	struct inode *vi = VFS_I(ni);
 102	ntfs_inode *base_ni;
 103	MFT_RECORD *m = NULL;
 104	ATTR_RECORD *a;
 105	ntfs_attr_search_ctx *ctx = NULL;
 106	struct address_space *mapping;
 107	struct page *page = NULL;
 108	u8 *kattr;
 109	int err;
 110	u32 attr_len;
 111
 112	read_lock_irqsave(&ni->size_lock, flags);
 113	old_init_size = ni->initialized_size;
 114	old_i_size = i_size_read(vi);
 115	BUG_ON(new_init_size > ni->allocated_size);
 116	read_unlock_irqrestore(&ni->size_lock, flags);
 117	ntfs_debug("Entering for i_ino 0x%lx, attribute type 0x%x, "
 118			"old_initialized_size 0x%llx, "
 119			"new_initialized_size 0x%llx, i_size 0x%llx.",
 120			vi->i_ino, (unsigned)le32_to_cpu(ni->type),
 121			(unsigned long long)old_init_size,
 122			(unsigned long long)new_init_size, old_i_size);
 123	if (!NInoAttr(ni))
 124		base_ni = ni;
 125	else
 126		base_ni = ni->ext.base_ntfs_ino;
 127	/* Use goto to reduce indentation and we need the label below anyway. */
 128	if (NInoNonResident(ni))
 129		goto do_non_resident_extend;
 130	BUG_ON(old_init_size != old_i_size);
 131	m = map_mft_record(base_ni);
 132	if (IS_ERR(m)) {
 133		err = PTR_ERR(m);
 134		m = NULL;
 135		goto err_out;
 136	}
 137	ctx = ntfs_attr_get_search_ctx(base_ni, m);
 138	if (unlikely(!ctx)) {
 139		err = -ENOMEM;
 140		goto err_out;
 141	}
 142	err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len,
 143			CASE_SENSITIVE, 0, NULL, 0, ctx);
 144	if (unlikely(err)) {
 145		if (err == -ENOENT)
 146			err = -EIO;
 147		goto err_out;
 148	}
 149	m = ctx->mrec;
 150	a = ctx->attr;
 151	BUG_ON(a->non_resident);
 152	/* The total length of the attribute value. */
 153	attr_len = le32_to_cpu(a->data.resident.value_length);
 154	BUG_ON(old_i_size != (loff_t)attr_len);
 155	/*
 156	 * Do the zeroing in the mft record and update the attribute size in
 157	 * the mft record.
 158	 */
 159	kattr = (u8*)a + le16_to_cpu(a->data.resident.value_offset);
 160	memset(kattr + attr_len, 0, new_init_size - attr_len);
 161	a->data.resident.value_length = cpu_to_le32((u32)new_init_size);
 162	/* Finally, update the sizes in the vfs and ntfs inodes. */
 163	write_lock_irqsave(&ni->size_lock, flags);
 164	i_size_write(vi, new_init_size);
 165	ni->initialized_size = new_init_size;
 166	write_unlock_irqrestore(&ni->size_lock, flags);
 167	goto done;
 168do_non_resident_extend:
 169	/*
 170	 * If the new initialized size @new_init_size exceeds the current file
 171	 * size (vfs inode->i_size), we need to extend the file size to the
 172	 * new initialized size.
 173	 */
 174	if (new_init_size > old_i_size) {
 175		m = map_mft_record(base_ni);
 176		if (IS_ERR(m)) {
 177			err = PTR_ERR(m);
 178			m = NULL;
 179			goto err_out;
 180		}
 181		ctx = ntfs_attr_get_search_ctx(base_ni, m);
 182		if (unlikely(!ctx)) {
 183			err = -ENOMEM;
 184			goto err_out;
 185		}
 186		err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len,
 187				CASE_SENSITIVE, 0, NULL, 0, ctx);
 188		if (unlikely(err)) {
 189			if (err == -ENOENT)
 190				err = -EIO;
 191			goto err_out;
 192		}
 193		m = ctx->mrec;
 194		a = ctx->attr;
 195		BUG_ON(!a->non_resident);
 196		BUG_ON(old_i_size != (loff_t)
 197				sle64_to_cpu(a->data.non_resident.data_size));
 198		a->data.non_resident.data_size = cpu_to_sle64(new_init_size);
 199		flush_dcache_mft_record_page(ctx->ntfs_ino);
 200		mark_mft_record_dirty(ctx->ntfs_ino);
 201		/* Update the file size in the vfs inode. */
 202		i_size_write(vi, new_init_size);
 203		ntfs_attr_put_search_ctx(ctx);
 204		ctx = NULL;
 205		unmap_mft_record(base_ni);
 206		m = NULL;
 207	}
 208	mapping = vi->i_mapping;
 209	index = old_init_size >> PAGE_SHIFT;
 210	end_index = (new_init_size + PAGE_SIZE - 1) >> PAGE_SHIFT;
 211	do {
 212		/*
 213		 * Read the page.  If the page is not present, this will zero
 214		 * the uninitialized regions for us.
 215		 */
 216		page = read_mapping_page(mapping, index, NULL);
 217		if (IS_ERR(page)) {
 218			err = PTR_ERR(page);
 219			goto init_err_out;
 220		}
 221		if (unlikely(PageError(page))) {
 222			put_page(page);
 223			err = -EIO;
 224			goto init_err_out;
 225		}
 226		/*
 227		 * Update the initialized size in the ntfs inode.  This is
 228		 * enough to make ntfs_writepage() work.
 229		 */
 230		write_lock_irqsave(&ni->size_lock, flags);
 231		ni->initialized_size = (s64)(index + 1) << PAGE_SHIFT;
 232		if (ni->initialized_size > new_init_size)
 233			ni->initialized_size = new_init_size;
 234		write_unlock_irqrestore(&ni->size_lock, flags);
 235		/* Set the page dirty so it gets written out. */
 236		set_page_dirty(page);
 237		put_page(page);
 238		/*
 239		 * Play nice with the vm and the rest of the system.  This is
 240		 * very much needed as we can potentially be modifying the
 241		 * initialised size from a very small value to a really huge
 242		 * value, e.g.
 243		 *	f = open(somefile, O_TRUNC);
 244		 *	truncate(f, 10GiB);
 245		 *	seek(f, 10GiB);
 246		 *	write(f, 1);
 247		 * And this would mean we would be marking dirty hundreds of
 248		 * thousands of pages or as in the above example more than
 249		 * two and a half million pages!
 250		 *
 251		 * TODO: For sparse pages could optimize this workload by using
 252		 * the FsMisc / MiscFs page bit as a "PageIsSparse" bit.  This
 253		 * would be set in readpage for sparse pages and here we would
 254		 * not need to mark dirty any pages which have this bit set.
 255		 * The only caveat is that we have to clear the bit everywhere
 256		 * where we allocate any clusters that lie in the page or that
 257		 * contain the page.
 258		 *
 259		 * TODO: An even greater optimization would be for us to only
 260		 * call readpage() on pages which are not in sparse regions as
 261		 * determined from the runlist.  This would greatly reduce the
 262		 * number of pages we read and make dirty in the case of sparse
 263		 * files.
 264		 */
 265		balance_dirty_pages_ratelimited(mapping);
 266		cond_resched();
 267	} while (++index < end_index);
 268	read_lock_irqsave(&ni->size_lock, flags);
 269	BUG_ON(ni->initialized_size != new_init_size);
 270	read_unlock_irqrestore(&ni->size_lock, flags);
 271	/* Now bring in sync the initialized_size in the mft record. */
 272	m = map_mft_record(base_ni);
 273	if (IS_ERR(m)) {
 274		err = PTR_ERR(m);
 275		m = NULL;
 276		goto init_err_out;
 277	}
 278	ctx = ntfs_attr_get_search_ctx(base_ni, m);
 279	if (unlikely(!ctx)) {
 280		err = -ENOMEM;
 281		goto init_err_out;
 282	}
 283	err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len,
 284			CASE_SENSITIVE, 0, NULL, 0, ctx);
 285	if (unlikely(err)) {
 286		if (err == -ENOENT)
 287			err = -EIO;
 288		goto init_err_out;
 289	}
 290	m = ctx->mrec;
 291	a = ctx->attr;
 292	BUG_ON(!a->non_resident);
 293	a->data.non_resident.initialized_size = cpu_to_sle64(new_init_size);
 294done:
 295	flush_dcache_mft_record_page(ctx->ntfs_ino);
 296	mark_mft_record_dirty(ctx->ntfs_ino);
 297	if (ctx)
 298		ntfs_attr_put_search_ctx(ctx);
 299	if (m)
 300		unmap_mft_record(base_ni);
 301	ntfs_debug("Done, initialized_size 0x%llx, i_size 0x%llx.",
 302			(unsigned long long)new_init_size, i_size_read(vi));
 303	return 0;
 304init_err_out:
 305	write_lock_irqsave(&ni->size_lock, flags);
 306	ni->initialized_size = old_init_size;
 307	write_unlock_irqrestore(&ni->size_lock, flags);
 308err_out:
 309	if (ctx)
 310		ntfs_attr_put_search_ctx(ctx);
 311	if (m)
 312		unmap_mft_record(base_ni);
 313	ntfs_debug("Failed.  Returning error code %i.", err);
 314	return err;
 315}
 316
 317static ssize_t ntfs_prepare_file_for_write(struct kiocb *iocb,
 318		struct iov_iter *from)
 319{
 320	loff_t pos;
 321	s64 end, ll;
 322	ssize_t err;
 323	unsigned long flags;
 324	struct file *file = iocb->ki_filp;
 325	struct inode *vi = file_inode(file);
 326	ntfs_inode *base_ni, *ni = NTFS_I(vi);
 327	ntfs_volume *vol = ni->vol;
 328
 329	ntfs_debug("Entering for i_ino 0x%lx, attribute type 0x%x, pos "
 330			"0x%llx, count 0x%zx.", vi->i_ino,
 331			(unsigned)le32_to_cpu(ni->type),
 332			(unsigned long long)iocb->ki_pos,
 333			iov_iter_count(from));
 334	err = generic_write_checks(iocb, from);
 335	if (unlikely(err <= 0))
 336		goto out;
 337	/*
 338	 * All checks have passed.  Before we start doing any writing we want
 339	 * to abort any totally illegal writes.
 340	 */
 341	BUG_ON(NInoMstProtected(ni));
 342	BUG_ON(ni->type != AT_DATA);
 343	/* If file is encrypted, deny access, just like NT4. */
 344	if (NInoEncrypted(ni)) {
 345		/* Only $DATA attributes can be encrypted. */
 346		/*
 347		 * Reminder for later: Encrypted files are _always_
 348		 * non-resident so that the content can always be encrypted.
 349		 */
 350		ntfs_debug("Denying write access to encrypted file.");
 351		err = -EACCES;
 352		goto out;
 353	}
 354	if (NInoCompressed(ni)) {
 355		/* Only unnamed $DATA attribute can be compressed. */
 356		BUG_ON(ni->name_len);
 357		/*
 358		 * Reminder for later: If resident, the data is not actually
 359		 * compressed.  Only on the switch to non-resident does
 360		 * compression kick in.  This is in contrast to encrypted files
 361		 * (see above).
 362		 */
 363		ntfs_error(vi->i_sb, "Writing to compressed files is not "
 364				"implemented yet.  Sorry.");
 365		err = -EOPNOTSUPP;
 366		goto out;
 367	}
 368	base_ni = ni;
 369	if (NInoAttr(ni))
 370		base_ni = ni->ext.base_ntfs_ino;
 371	err = file_remove_privs(file);
 372	if (unlikely(err))
 373		goto out;
 374	/*
 375	 * Our ->update_time method always succeeds thus file_update_time()
 376	 * cannot fail either so there is no need to check the return code.
 377	 */
 378	file_update_time(file);
 379	pos = iocb->ki_pos;
 380	/* The first byte after the last cluster being written to. */
 381	end = (pos + iov_iter_count(from) + vol->cluster_size_mask) &
 382			~(u64)vol->cluster_size_mask;
 383	/*
 384	 * If the write goes beyond the allocated size, extend the allocation
 385	 * to cover the whole of the write, rounded up to the nearest cluster.
 386	 */
 387	read_lock_irqsave(&ni->size_lock, flags);
 388	ll = ni->allocated_size;
 389	read_unlock_irqrestore(&ni->size_lock, flags);
 390	if (end > ll) {
 391		/*
 392		 * Extend the allocation without changing the data size.
 393		 *
 394		 * Note we ensure the allocation is big enough to at least
 395		 * write some data but we do not require the allocation to be
 396		 * complete, i.e. it may be partial.
 397		 */
 398		ll = ntfs_attr_extend_allocation(ni, end, -1, pos);
 399		if (likely(ll >= 0)) {
 400			BUG_ON(pos >= ll);
 401			/* If the extension was partial truncate the write. */
 402			if (end > ll) {
 403				ntfs_debug("Truncating write to inode 0x%lx, "
 404						"attribute type 0x%x, because "
 405						"the allocation was only "
 406						"partially extended.",
 407						vi->i_ino, (unsigned)
 408						le32_to_cpu(ni->type));
 409				iov_iter_truncate(from, ll - pos);
 410			}
 411		} else {
 412			err = ll;
 413			read_lock_irqsave(&ni->size_lock, flags);
 414			ll = ni->allocated_size;
 415			read_unlock_irqrestore(&ni->size_lock, flags);
 416			/* Perform a partial write if possible or fail. */
 417			if (pos < ll) {
 418				ntfs_debug("Truncating write to inode 0x%lx "
 419						"attribute type 0x%x, because "
 420						"extending the allocation "
 421						"failed (error %d).",
 422						vi->i_ino, (unsigned)
 423						le32_to_cpu(ni->type),
 424						(int)-err);
 425				iov_iter_truncate(from, ll - pos);
 426			} else {
 427				if (err != -ENOSPC)
 428					ntfs_error(vi->i_sb, "Cannot perform "
 429							"write to inode "
 430							"0x%lx, attribute "
 431							"type 0x%x, because "
 432							"extending the "
 433							"allocation failed "
 434							"(error %ld).",
 435							vi->i_ino, (unsigned)
 436							le32_to_cpu(ni->type),
 437							(long)-err);
 438				else
 439					ntfs_debug("Cannot perform write to "
 440							"inode 0x%lx, "
 441							"attribute type 0x%x, "
 442							"because there is not "
 443							"space left.",
 444							vi->i_ino, (unsigned)
 445							le32_to_cpu(ni->type));
 446				goto out;
 447			}
 448		}
 449	}
 450	/*
 451	 * If the write starts beyond the initialized size, extend it up to the
 452	 * beginning of the write and initialize all non-sparse space between
 453	 * the old initialized size and the new one.  This automatically also
 454	 * increments the vfs inode->i_size to keep it above or equal to the
 455	 * initialized_size.
 456	 */
 457	read_lock_irqsave(&ni->size_lock, flags);
 458	ll = ni->initialized_size;
 459	read_unlock_irqrestore(&ni->size_lock, flags);
 460	if (pos > ll) {
 461		/*
 462		 * Wait for ongoing direct i/o to complete before proceeding.
 463		 * New direct i/o cannot start as we hold i_mutex.
 464		 */
 465		inode_dio_wait(vi);
 466		err = ntfs_attr_extend_initialized(ni, pos);
 467		if (unlikely(err < 0))
 468			ntfs_error(vi->i_sb, "Cannot perform write to inode "
 469					"0x%lx, attribute type 0x%x, because "
 470					"extending the initialized size "
 471					"failed (error %d).", vi->i_ino,
 472					(unsigned)le32_to_cpu(ni->type),
 473					(int)-err);
 474	}
 475out:
 476	return err;
 477}
 478
 479/**
 480 * __ntfs_grab_cache_pages - obtain a number of locked pages
 481 * @mapping:	address space mapping from which to obtain page cache pages
 482 * @index:	starting index in @mapping at which to begin obtaining pages
 483 * @nr_pages:	number of page cache pages to obtain
 484 * @pages:	array of pages in which to return the obtained page cache pages
 485 * @cached_page: allocated but as yet unused page
 486 *
 487 * Obtain @nr_pages locked page cache pages from the mapping @mapping and
 488 * starting at index @index.
 489 *
 490 * If a page is newly created, add it to lru list
 491 *
 492 * Note, the page locks are obtained in ascending page index order.
 493 */
 494static inline int __ntfs_grab_cache_pages(struct address_space *mapping,
 495		pgoff_t index, const unsigned nr_pages, struct page **pages,
 496		struct page **cached_page)
 497{
 498	int err, nr;
 499
 500	BUG_ON(!nr_pages);
 501	err = nr = 0;
 502	do {
 503		pages[nr] = find_get_page_flags(mapping, index, FGP_LOCK |
 504				FGP_ACCESSED);
 505		if (!pages[nr]) {
 506			if (!*cached_page) {
 507				*cached_page = page_cache_alloc(mapping);
 508				if (unlikely(!*cached_page)) {
 509					err = -ENOMEM;
 510					goto err_out;
 511				}
 512			}
 513			err = add_to_page_cache_lru(*cached_page, mapping,
 514				   index,
 515				   mapping_gfp_constraint(mapping, GFP_KERNEL));
 516			if (unlikely(err)) {
 517				if (err == -EEXIST)
 518					continue;
 519				goto err_out;
 520			}
 521			pages[nr] = *cached_page;
 522			*cached_page = NULL;
 523		}
 524		index++;
 525		nr++;
 526	} while (nr < nr_pages);
 527out:
 528	return err;
 529err_out:
 530	while (nr > 0) {
 531		unlock_page(pages[--nr]);
 532		put_page(pages[nr]);
 533	}
 534	goto out;
 535}
 536
 537static inline int ntfs_submit_bh_for_read(struct buffer_head *bh)
 538{
 539	lock_buffer(bh);
 540	get_bh(bh);
 541	bh->b_end_io = end_buffer_read_sync;
 542	return submit_bh(REQ_OP_READ, 0, bh);
 543}
 544
 545/**
 546 * ntfs_prepare_pages_for_non_resident_write - prepare pages for receiving data
 547 * @pages:	array of destination pages
 548 * @nr_pages:	number of pages in @pages
 549 * @pos:	byte position in file at which the write begins
 550 * @bytes:	number of bytes to be written
 551 *
 552 * This is called for non-resident attributes from ntfs_file_buffered_write()
 553 * with i_mutex held on the inode (@pages[0]->mapping->host).  There are
 554 * @nr_pages pages in @pages which are locked but not kmap()ped.  The source
 555 * data has not yet been copied into the @pages.
 556 * 
 557 * Need to fill any holes with actual clusters, allocate buffers if necessary,
 558 * ensure all the buffers are mapped, and bring uptodate any buffers that are
 559 * only partially being written to.
 560 *
 561 * If @nr_pages is greater than one, we are guaranteed that the cluster size is
 562 * greater than PAGE_SIZE, that all pages in @pages are entirely inside
 563 * the same cluster and that they are the entirety of that cluster, and that
 564 * the cluster is sparse, i.e. we need to allocate a cluster to fill the hole.
 565 *
 566 * i_size is not to be modified yet.
 567 *
 568 * Return 0 on success or -errno on error.
 569 */
 570static int ntfs_prepare_pages_for_non_resident_write(struct page **pages,
 571		unsigned nr_pages, s64 pos, size_t bytes)
 572{
 573	VCN vcn, highest_vcn = 0, cpos, cend, bh_cpos, bh_cend;
 574	LCN lcn;
 575	s64 bh_pos, vcn_len, end, initialized_size;
 576	sector_t lcn_block;
 577	struct page *page;
 578	struct inode *vi;
 579	ntfs_inode *ni, *base_ni = NULL;
 580	ntfs_volume *vol;
 581	runlist_element *rl, *rl2;
 582	struct buffer_head *bh, *head, *wait[2], **wait_bh = wait;
 583	ntfs_attr_search_ctx *ctx = NULL;
 584	MFT_RECORD *m = NULL;
 585	ATTR_RECORD *a = NULL;
 586	unsigned long flags;
 587	u32 attr_rec_len = 0;
 588	unsigned blocksize, u;
 589	int err, mp_size;
 590	bool rl_write_locked, was_hole, is_retry;
 591	unsigned char blocksize_bits;
 592	struct {
 593		u8 runlist_merged:1;
 594		u8 mft_attr_mapped:1;
 595		u8 mp_rebuilt:1;
 596		u8 attr_switched:1;
 597	} status = { 0, 0, 0, 0 };
 598
 599	BUG_ON(!nr_pages);
 600	BUG_ON(!pages);
 601	BUG_ON(!*pages);
 602	vi = pages[0]->mapping->host;
 603	ni = NTFS_I(vi);
 604	vol = ni->vol;
 605	ntfs_debug("Entering for inode 0x%lx, attribute type 0x%x, start page "
 606			"index 0x%lx, nr_pages 0x%x, pos 0x%llx, bytes 0x%zx.",
 607			vi->i_ino, ni->type, pages[0]->index, nr_pages,
 608			(long long)pos, bytes);
 609	blocksize = vol->sb->s_blocksize;
 610	blocksize_bits = vol->sb->s_blocksize_bits;
 611	u = 0;
 612	do {
 613		page = pages[u];
 614		BUG_ON(!page);
 615		/*
 616		 * create_empty_buffers() will create uptodate/dirty buffers if
 617		 * the page is uptodate/dirty.
 618		 */
 619		if (!page_has_buffers(page)) {
 620			create_empty_buffers(page, blocksize, 0);
 621			if (unlikely(!page_has_buffers(page)))
 622				return -ENOMEM;
 623		}
 624	} while (++u < nr_pages);
 625	rl_write_locked = false;
 626	rl = NULL;
 627	err = 0;
 628	vcn = lcn = -1;
 629	vcn_len = 0;
 630	lcn_block = -1;
 631	was_hole = false;
 632	cpos = pos >> vol->cluster_size_bits;
 633	end = pos + bytes;
 634	cend = (end + vol->cluster_size - 1) >> vol->cluster_size_bits;
 635	/*
 636	 * Loop over each page and for each page over each buffer.  Use goto to
 637	 * reduce indentation.
 638	 */
 639	u = 0;
 640do_next_page:
 641	page = pages[u];
 642	bh_pos = (s64)page->index << PAGE_SHIFT;
 643	bh = head = page_buffers(page);
 644	do {
 645		VCN cdelta;
 646		s64 bh_end;
 647		unsigned bh_cofs;
 648
 649		/* Clear buffer_new on all buffers to reinitialise state. */
 650		if (buffer_new(bh))
 651			clear_buffer_new(bh);
 652		bh_end = bh_pos + blocksize;
 653		bh_cpos = bh_pos >> vol->cluster_size_bits;
 654		bh_cofs = bh_pos & vol->cluster_size_mask;
 655		if (buffer_mapped(bh)) {
 656			/*
 657			 * The buffer is already mapped.  If it is uptodate,
 658			 * ignore it.
 659			 */
 660			if (buffer_uptodate(bh))
 661				continue;
 662			/*
 663			 * The buffer is not uptodate.  If the page is uptodate
 664			 * set the buffer uptodate and otherwise ignore it.
 665			 */
 666			if (PageUptodate(page)) {
 667				set_buffer_uptodate(bh);
 668				continue;
 669			}
 670			/*
 671			 * Neither the page nor the buffer are uptodate.  If
 672			 * the buffer is only partially being written to, we
 673			 * need to read it in before the write, i.e. now.
 674			 */
 675			if ((bh_pos < pos && bh_end > pos) ||
 676					(bh_pos < end && bh_end > end)) {
 677				/*
 678				 * If the buffer is fully or partially within
 679				 * the initialized size, do an actual read.
 680				 * Otherwise, simply zero the buffer.
 681				 */
 682				read_lock_irqsave(&ni->size_lock, flags);
 683				initialized_size = ni->initialized_size;
 684				read_unlock_irqrestore(&ni->size_lock, flags);
 685				if (bh_pos < initialized_size) {
 686					ntfs_submit_bh_for_read(bh);
 687					*wait_bh++ = bh;
 688				} else {
 689					zero_user(page, bh_offset(bh),
 690							blocksize);
 691					set_buffer_uptodate(bh);
 692				}
 693			}
 694			continue;
 695		}
 696		/* Unmapped buffer.  Need to map it. */
 697		bh->b_bdev = vol->sb->s_bdev;
 698		/*
 699		 * If the current buffer is in the same clusters as the map
 700		 * cache, there is no need to check the runlist again.  The
 701		 * map cache is made up of @vcn, which is the first cached file
 702		 * cluster, @vcn_len which is the number of cached file
 703		 * clusters, @lcn is the device cluster corresponding to @vcn,
 704		 * and @lcn_block is the block number corresponding to @lcn.
 705		 */
 706		cdelta = bh_cpos - vcn;
 707		if (likely(!cdelta || (cdelta > 0 && cdelta < vcn_len))) {
 708map_buffer_cached:
 709			BUG_ON(lcn < 0);
 710			bh->b_blocknr = lcn_block +
 711					(cdelta << (vol->cluster_size_bits -
 712					blocksize_bits)) +
 713					(bh_cofs >> blocksize_bits);
 714			set_buffer_mapped(bh);
 715			/*
 716			 * If the page is uptodate so is the buffer.  If the
 717			 * buffer is fully outside the write, we ignore it if
 718			 * it was already allocated and we mark it dirty so it
 719			 * gets written out if we allocated it.  On the other
 720			 * hand, if we allocated the buffer but we are not
 721			 * marking it dirty we set buffer_new so we can do
 722			 * error recovery.
 723			 */
 724			if (PageUptodate(page)) {
 725				if (!buffer_uptodate(bh))
 726					set_buffer_uptodate(bh);
 727				if (unlikely(was_hole)) {
 728					/* We allocated the buffer. */
 729					clean_bdev_bh_alias(bh);
 730					if (bh_end <= pos || bh_pos >= end)
 731						mark_buffer_dirty(bh);
 732					else
 733						set_buffer_new(bh);
 734				}
 735				continue;
 736			}
 737			/* Page is _not_ uptodate. */
 738			if (likely(!was_hole)) {
 739				/*
 740				 * Buffer was already allocated.  If it is not
 741				 * uptodate and is only partially being written
 742				 * to, we need to read it in before the write,
 743				 * i.e. now.
 744				 */
 745				if (!buffer_uptodate(bh) && bh_pos < end &&
 746						bh_end > pos &&
 747						(bh_pos < pos ||
 748						bh_end > end)) {
 749					/*
 750					 * If the buffer is fully or partially
 751					 * within the initialized size, do an
 752					 * actual read.  Otherwise, simply zero
 753					 * the buffer.
 754					 */
 755					read_lock_irqsave(&ni->size_lock,
 756							flags);
 757					initialized_size = ni->initialized_size;
 758					read_unlock_irqrestore(&ni->size_lock,
 759							flags);
 760					if (bh_pos < initialized_size) {
 761						ntfs_submit_bh_for_read(bh);
 762						*wait_bh++ = bh;
 763					} else {
 764						zero_user(page, bh_offset(bh),
 765								blocksize);
 766						set_buffer_uptodate(bh);
 767					}
 768				}
 769				continue;
 770			}
 771			/* We allocated the buffer. */
 772			clean_bdev_bh_alias(bh);
 773			/*
 774			 * If the buffer is fully outside the write, zero it,
 775			 * set it uptodate, and mark it dirty so it gets
 776			 * written out.  If it is partially being written to,
 777			 * zero region surrounding the write but leave it to
 778			 * commit write to do anything else.  Finally, if the
 779			 * buffer is fully being overwritten, do nothing.
 780			 */
 781			if (bh_end <= pos || bh_pos >= end) {
 782				if (!buffer_uptodate(bh)) {
 783					zero_user(page, bh_offset(bh),
 784							blocksize);
 785					set_buffer_uptodate(bh);
 786				}
 787				mark_buffer_dirty(bh);
 788				continue;
 789			}
 790			set_buffer_new(bh);
 791			if (!buffer_uptodate(bh) &&
 792					(bh_pos < pos || bh_end > end)) {
 793				u8 *kaddr;
 794				unsigned pofs;
 795					
 796				kaddr = kmap_atomic(page);
 797				if (bh_pos < pos) {
 798					pofs = bh_pos & ~PAGE_MASK;
 799					memset(kaddr + pofs, 0, pos - bh_pos);
 800				}
 801				if (bh_end > end) {
 802					pofs = end & ~PAGE_MASK;
 803					memset(kaddr + pofs, 0, bh_end - end);
 804				}
 805				kunmap_atomic(kaddr);
 806				flush_dcache_page(page);
 807			}
 808			continue;
 809		}
 810		/*
 811		 * Slow path: this is the first buffer in the cluster.  If it
 812		 * is outside allocated size and is not uptodate, zero it and
 813		 * set it uptodate.
 814		 */
 815		read_lock_irqsave(&ni->size_lock, flags);
 816		initialized_size = ni->allocated_size;
 817		read_unlock_irqrestore(&ni->size_lock, flags);
 818		if (bh_pos > initialized_size) {
 819			if (PageUptodate(page)) {
 820				if (!buffer_uptodate(bh))
 821					set_buffer_uptodate(bh);
 822			} else if (!buffer_uptodate(bh)) {
 823				zero_user(page, bh_offset(bh), blocksize);
 824				set_buffer_uptodate(bh);
 825			}
 826			continue;
 827		}
 828		is_retry = false;
 829		if (!rl) {
 830			down_read(&ni->runlist.lock);
 831retry_remap:
 832			rl = ni->runlist.rl;
 833		}
 834		if (likely(rl != NULL)) {
 835			/* Seek to element containing target cluster. */
 836			while (rl->length && rl[1].vcn <= bh_cpos)
 837				rl++;
 838			lcn = ntfs_rl_vcn_to_lcn(rl, bh_cpos);
 839			if (likely(lcn >= 0)) {
 840				/*
 841				 * Successful remap, setup the map cache and
 842				 * use that to deal with the buffer.
 843				 */
 844				was_hole = false;
 845				vcn = bh_cpos;
 846				vcn_len = rl[1].vcn - vcn;
 847				lcn_block = lcn << (vol->cluster_size_bits -
 848						blocksize_bits);
 849				cdelta = 0;
 850				/*
 851				 * If the number of remaining clusters touched
 852				 * by the write is smaller or equal to the
 853				 * number of cached clusters, unlock the
 854				 * runlist as the map cache will be used from
 855				 * now on.
 856				 */
 857				if (likely(vcn + vcn_len >= cend)) {
 858					if (rl_write_locked) {
 859						up_write(&ni->runlist.lock);
 860						rl_write_locked = false;
 861					} else
 862						up_read(&ni->runlist.lock);
 863					rl = NULL;
 864				}
 865				goto map_buffer_cached;
 866			}
 867		} else
 868			lcn = LCN_RL_NOT_MAPPED;
 869		/*
 870		 * If it is not a hole and not out of bounds, the runlist is
 871		 * probably unmapped so try to map it now.
 872		 */
 873		if (unlikely(lcn != LCN_HOLE && lcn != LCN_ENOENT)) {
 874			if (likely(!is_retry && lcn == LCN_RL_NOT_MAPPED)) {
 875				/* Attempt to map runlist. */
 876				if (!rl_write_locked) {
 877					/*
 878					 * We need the runlist locked for
 879					 * writing, so if it is locked for
 880					 * reading relock it now and retry in
 881					 * case it changed whilst we dropped
 882					 * the lock.
 883					 */
 884					up_read(&ni->runlist.lock);
 885					down_write(&ni->runlist.lock);
 886					rl_write_locked = true;
 887					goto retry_remap;
 888				}
 889				err = ntfs_map_runlist_nolock(ni, bh_cpos,
 890						NULL);
 891				if (likely(!err)) {
 892					is_retry = true;
 893					goto retry_remap;
 894				}
 895				/*
 896				 * If @vcn is out of bounds, pretend @lcn is
 897				 * LCN_ENOENT.  As long as the buffer is out
 898				 * of bounds this will work fine.
 899				 */
 900				if (err == -ENOENT) {
 901					lcn = LCN_ENOENT;
 902					err = 0;
 903					goto rl_not_mapped_enoent;
 904				}
 905			} else
 906				err = -EIO;
 907			/* Failed to map the buffer, even after retrying. */
 908			bh->b_blocknr = -1;
 909			ntfs_error(vol->sb, "Failed to write to inode 0x%lx, "
 910					"attribute type 0x%x, vcn 0x%llx, "
 911					"vcn offset 0x%x, because its "
 912					"location on disk could not be "
 913					"determined%s (error code %i).",
 914					ni->mft_no, ni->type,
 915					(unsigned long long)bh_cpos,
 916					(unsigned)bh_pos &
 917					vol->cluster_size_mask,
 918					is_retry ? " even after retrying" : "",
 919					err);
 920			break;
 921		}
 922rl_not_mapped_enoent:
 923		/*
 924		 * The buffer is in a hole or out of bounds.  We need to fill
 925		 * the hole, unless the buffer is in a cluster which is not
 926		 * touched by the write, in which case we just leave the buffer
 927		 * unmapped.  This can only happen when the cluster size is
 928		 * less than the page cache size.
 929		 */
 930		if (unlikely(vol->cluster_size < PAGE_SIZE)) {
 931			bh_cend = (bh_end + vol->cluster_size - 1) >>
 932					vol->cluster_size_bits;
 933			if ((bh_cend <= cpos || bh_cpos >= cend)) {
 934				bh->b_blocknr = -1;
 935				/*
 936				 * If the buffer is uptodate we skip it.  If it
 937				 * is not but the page is uptodate, we can set
 938				 * the buffer uptodate.  If the page is not
 939				 * uptodate, we can clear the buffer and set it
 940				 * uptodate.  Whether this is worthwhile is
 941				 * debatable and this could be removed.
 942				 */
 943				if (PageUptodate(page)) {
 944					if (!buffer_uptodate(bh))
 945						set_buffer_uptodate(bh);
 946				} else if (!buffer_uptodate(bh)) {
 947					zero_user(page, bh_offset(bh),
 948						blocksize);
 949					set_buffer_uptodate(bh);
 950				}
 951				continue;
 952			}
 953		}
 954		/*
 955		 * Out of bounds buffer is invalid if it was not really out of
 956		 * bounds.
 957		 */
 958		BUG_ON(lcn != LCN_HOLE);
 959		/*
 960		 * We need the runlist locked for writing, so if it is locked
 961		 * for reading relock it now and retry in case it changed
 962		 * whilst we dropped the lock.
 963		 */
 964		BUG_ON(!rl);
 965		if (!rl_write_locked) {
 966			up_read(&ni->runlist.lock);
 967			down_write(&ni->runlist.lock);
 968			rl_write_locked = true;
 969			goto retry_remap;
 970		}
 971		/* Find the previous last allocated cluster. */
 972		BUG_ON(rl->lcn != LCN_HOLE);
 973		lcn = -1;
 974		rl2 = rl;
 975		while (--rl2 >= ni->runlist.rl) {
 976			if (rl2->lcn >= 0) {
 977				lcn = rl2->lcn + rl2->length;
 978				break;
 979			}
 980		}
 981		rl2 = ntfs_cluster_alloc(vol, bh_cpos, 1, lcn, DATA_ZONE,
 982				false);
 983		if (IS_ERR(rl2)) {
 984			err = PTR_ERR(rl2);
 985			ntfs_debug("Failed to allocate cluster, error code %i.",
 986					err);
 987			break;
 988		}
 989		lcn = rl2->lcn;
 990		rl = ntfs_runlists_merge(ni->runlist.rl, rl2);
 991		if (IS_ERR(rl)) {
 992			err = PTR_ERR(rl);
 993			if (err != -ENOMEM)
 994				err = -EIO;
 995			if (ntfs_cluster_free_from_rl(vol, rl2)) {
 996				ntfs_error(vol->sb, "Failed to release "
 997						"allocated cluster in error "
 998						"code path.  Run chkdsk to "
 999						"recover the lost cluster.");
1000				NVolSetErrors(vol);
1001			}
1002			ntfs_free(rl2);
1003			break;
1004		}
1005		ni->runlist.rl = rl;
1006		status.runlist_merged = 1;
1007		ntfs_debug("Allocated cluster, lcn 0x%llx.",
1008				(unsigned long long)lcn);
1009		/* Map and lock the mft record and get the attribute record. */
1010		if (!NInoAttr(ni))
1011			base_ni = ni;
1012		else
1013			base_ni = ni->ext.base_ntfs_ino;
1014		m = map_mft_record(base_ni);
1015		if (IS_ERR(m)) {
1016			err = PTR_ERR(m);
1017			break;
1018		}
1019		ctx = ntfs_attr_get_search_ctx(base_ni, m);
1020		if (unlikely(!ctx)) {
1021			err = -ENOMEM;
1022			unmap_mft_record(base_ni);
1023			break;
1024		}
1025		status.mft_attr_mapped = 1;
1026		err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len,
1027				CASE_SENSITIVE, bh_cpos, NULL, 0, ctx);
1028		if (unlikely(err)) {
1029			if (err == -ENOENT)
1030				err = -EIO;
1031			break;
1032		}
1033		m = ctx->mrec;
1034		a = ctx->attr;
1035		/*
1036		 * Find the runlist element with which the attribute extent
1037		 * starts.  Note, we cannot use the _attr_ version because we
1038		 * have mapped the mft record.  That is ok because we know the
1039		 * runlist fragment must be mapped already to have ever gotten
1040		 * here, so we can just use the _rl_ version.
1041		 */
1042		vcn = sle64_to_cpu(a->data.non_resident.lowest_vcn);
1043		rl2 = ntfs_rl_find_vcn_nolock(rl, vcn);
1044		BUG_ON(!rl2);
1045		BUG_ON(!rl2->length);
1046		BUG_ON(rl2->lcn < LCN_HOLE);
1047		highest_vcn = sle64_to_cpu(a->data.non_resident.highest_vcn);
1048		/*
1049		 * If @highest_vcn is zero, calculate the real highest_vcn
1050		 * (which can really be zero).
1051		 */
1052		if (!highest_vcn)
1053			highest_vcn = (sle64_to_cpu(
1054					a->data.non_resident.allocated_size) >>
1055					vol->cluster_size_bits) - 1;
1056		/*
1057		 * Determine the size of the mapping pairs array for the new
1058		 * extent, i.e. the old extent with the hole filled.
1059		 */
1060		mp_size = ntfs_get_size_for_mapping_pairs(vol, rl2, vcn,
1061				highest_vcn);
1062		if (unlikely(mp_size <= 0)) {
1063			if (!(err = mp_size))
1064				err = -EIO;
1065			ntfs_debug("Failed to get size for mapping pairs "
1066					"array, error code %i.", err);
1067			break;
1068		}
1069		/*
1070		 * Resize the attribute record to fit the new mapping pairs
1071		 * array.
1072		 */
1073		attr_rec_len = le32_to_cpu(a->length);
1074		err = ntfs_attr_record_resize(m, a, mp_size + le16_to_cpu(
1075				a->data.non_resident.mapping_pairs_offset));
1076		if (unlikely(err)) {
1077			BUG_ON(err != -ENOSPC);
1078			// TODO: Deal with this by using the current attribute
1079			// and fill it with as much of the mapping pairs
1080			// array as possible.  Then loop over each attribute
1081			// extent rewriting the mapping pairs arrays as we go
1082			// along and if when we reach the end we have not
1083			// enough space, try to resize the last attribute
1084			// extent and if even that fails, add a new attribute
1085			// extent.
1086			// We could also try to resize at each step in the hope
1087			// that we will not need to rewrite every single extent.
1088			// Note, we may need to decompress some extents to fill
1089			// the runlist as we are walking the extents...
1090			ntfs_error(vol->sb, "Not enough space in the mft "
1091					"record for the extended attribute "
1092					"record.  This case is not "
1093					"implemented yet.");
1094			err = -EOPNOTSUPP;
1095			break ;
1096		}
1097		status.mp_rebuilt = 1;
1098		/*
1099		 * Generate the mapping pairs array directly into the attribute
1100		 * record.
1101		 */
1102		err = ntfs_mapping_pairs_build(vol, (u8*)a + le16_to_cpu(
1103				a->data.non_resident.mapping_pairs_offset),
1104				mp_size, rl2, vcn, highest_vcn, NULL);
1105		if (unlikely(err)) {
1106			ntfs_error(vol->sb, "Cannot fill hole in inode 0x%lx, "
1107					"attribute type 0x%x, because building "
1108					"the mapping pairs failed with error "
1109					"code %i.", vi->i_ino,
1110					(unsigned)le32_to_cpu(ni->type), err);
1111			err = -EIO;
1112			break;
1113		}
1114		/* Update the highest_vcn but only if it was not set. */
1115		if (unlikely(!a->data.non_resident.highest_vcn))
1116			a->data.non_resident.highest_vcn =
1117					cpu_to_sle64(highest_vcn);
1118		/*
1119		 * If the attribute is sparse/compressed, update the compressed
1120		 * size in the ntfs_inode structure and the attribute record.
1121		 */
1122		if (likely(NInoSparse(ni) || NInoCompressed(ni))) {
1123			/*
1124			 * If we are not in the first attribute extent, switch
1125			 * to it, but first ensure the changes will make it to
1126			 * disk later.
1127			 */
1128			if (a->data.non_resident.lowest_vcn) {
1129				flush_dcache_mft_record_page(ctx->ntfs_ino);
1130				mark_mft_record_dirty(ctx->ntfs_ino);
1131				ntfs_attr_reinit_search_ctx(ctx);
1132				err = ntfs_attr_lookup(ni->type, ni->name,
1133						ni->name_len, CASE_SENSITIVE,
1134						0, NULL, 0, ctx);
1135				if (unlikely(err)) {
1136					status.attr_switched = 1;
1137					break;
1138				}
1139				/* @m is not used any more so do not set it. */
1140				a = ctx->attr;
1141			}
1142			write_lock_irqsave(&ni->size_lock, flags);
1143			ni->itype.compressed.size += vol->cluster_size;
1144			a->data.non_resident.compressed_size =
1145					cpu_to_sle64(ni->itype.compressed.size);
1146			write_unlock_irqrestore(&ni->size_lock, flags);
1147		}
1148		/* Ensure the changes make it to disk. */
1149		flush_dcache_mft_record_page(ctx->ntfs_ino);
1150		mark_mft_record_dirty(ctx->ntfs_ino);
1151		ntfs_attr_put_search_ctx(ctx);
1152		unmap_mft_record(base_ni);
1153		/* Successfully filled the hole. */
1154		status.runlist_merged = 0;
1155		status.mft_attr_mapped = 0;
1156		status.mp_rebuilt = 0;
1157		/* Setup the map cache and use that to deal with the buffer. */
1158		was_hole = true;
1159		vcn = bh_cpos;
1160		vcn_len = 1;
1161		lcn_block = lcn << (vol->cluster_size_bits - blocksize_bits);
1162		cdelta = 0;
1163		/*
1164		 * If the number of remaining clusters in the @pages is smaller
1165		 * or equal to the number of cached clusters, unlock the
1166		 * runlist as the map cache will be used from now on.
1167		 */
1168		if (likely(vcn + vcn_len >= cend)) {
1169			up_write(&ni->runlist.lock);
1170			rl_write_locked = false;
1171			rl = NULL;
1172		}
1173		goto map_buffer_cached;
1174	} while (bh_pos += blocksize, (bh = bh->b_this_page) != head);
1175	/* If there are no errors, do the next page. */
1176	if (likely(!err && ++u < nr_pages))
1177		goto do_next_page;
1178	/* If there are no errors, release the runlist lock if we took it. */
1179	if (likely(!err)) {
1180		if (unlikely(rl_write_locked)) {
1181			up_write(&ni->runlist.lock);
1182			rl_write_locked = false;
1183		} else if (unlikely(rl))
1184			up_read(&ni->runlist.lock);
1185		rl = NULL;
1186	}
1187	/* If we issued read requests, let them complete. */
1188	read_lock_irqsave(&ni->size_lock, flags);
1189	initialized_size = ni->initialized_size;
1190	read_unlock_irqrestore(&ni->size_lock, flags);
1191	while (wait_bh > wait) {
1192		bh = *--wait_bh;
1193		wait_on_buffer(bh);
1194		if (likely(buffer_uptodate(bh))) {
1195			page = bh->b_page;
1196			bh_pos = ((s64)page->index << PAGE_SHIFT) +
1197					bh_offset(bh);
1198			/*
1199			 * If the buffer overflows the initialized size, need
1200			 * to zero the overflowing region.
1201			 */
1202			if (unlikely(bh_pos + blocksize > initialized_size)) {
1203				int ofs = 0;
1204
1205				if (likely(bh_pos < initialized_size))
1206					ofs = initialized_size - bh_pos;
1207				zero_user_segment(page, bh_offset(bh) + ofs,
1208						blocksize);
1209			}
1210		} else /* if (unlikely(!buffer_uptodate(bh))) */
1211			err = -EIO;
1212	}
1213	if (likely(!err)) {
1214		/* Clear buffer_new on all buffers. */
1215		u = 0;
1216		do {
1217			bh = head = page_buffers(pages[u]);
1218			do {
1219				if (buffer_new(bh))
1220					clear_buffer_new(bh);
1221			} while ((bh = bh->b_this_page) != head);
1222		} while (++u < nr_pages);
1223		ntfs_debug("Done.");
1224		return err;
1225	}
1226	if (status.attr_switched) {
1227		/* Get back to the attribute extent we modified. */
1228		ntfs_attr_reinit_search_ctx(ctx);
1229		if (ntfs_attr_lookup(ni->type, ni->name, ni->name_len,
1230				CASE_SENSITIVE, bh_cpos, NULL, 0, ctx)) {
1231			ntfs_error(vol->sb, "Failed to find required "
1232					"attribute extent of attribute in "
1233					"error code path.  Run chkdsk to "
1234					"recover.");
1235			write_lock_irqsave(&ni->size_lock, flags);
1236			ni->itype.compressed.size += vol->cluster_size;
1237			write_unlock_irqrestore(&ni->size_lock, flags);
1238			flush_dcache_mft_record_page(ctx->ntfs_ino);
1239			mark_mft_record_dirty(ctx->ntfs_ino);
1240			/*
1241			 * The only thing that is now wrong is the compressed
1242			 * size of the base attribute extent which chkdsk
1243			 * should be able to fix.
1244			 */
1245			NVolSetErrors(vol);
1246		} else {
1247			m = ctx->mrec;
1248			a = ctx->attr;
1249			status.attr_switched = 0;
1250		}
1251	}
1252	/*
1253	 * If the runlist has been modified, need to restore it by punching a
1254	 * hole into it and we then need to deallocate the on-disk cluster as
1255	 * well.  Note, we only modify the runlist if we are able to generate a
1256	 * new mapping pairs array, i.e. only when the mapped attribute extent
1257	 * is not switched.
1258	 */
1259	if (status.runlist_merged && !status.attr_switched) {
1260		BUG_ON(!rl_write_locked);
1261		/* Make the file cluster we allocated sparse in the runlist. */
1262		if (ntfs_rl_punch_nolock(vol, &ni->runlist, bh_cpos, 1)) {
1263			ntfs_error(vol->sb, "Failed to punch hole into "
1264					"attribute runlist in error code "
1265					"path.  Run chkdsk to recover the "
1266					"lost cluster.");
1267			NVolSetErrors(vol);
1268		} else /* if (success) */ {
1269			status.runlist_merged = 0;
1270			/*
1271			 * Deallocate the on-disk cluster we allocated but only
1272			 * if we succeeded in punching its vcn out of the
1273			 * runlist.
1274			 */
1275			down_write(&vol->lcnbmp_lock);
1276			if (ntfs_bitmap_clear_bit(vol->lcnbmp_ino, lcn)) {
1277				ntfs_error(vol->sb, "Failed to release "
1278						"allocated cluster in error "
1279						"code path.  Run chkdsk to "
1280						"recover the lost cluster.");
1281				NVolSetErrors(vol);
1282			}
1283			up_write(&vol->lcnbmp_lock);
1284		}
1285	}
1286	/*
1287	 * Resize the attribute record to its old size and rebuild the mapping
1288	 * pairs array.  Note, we only can do this if the runlist has been
1289	 * restored to its old state which also implies that the mapped
1290	 * attribute extent is not switched.
1291	 */
1292	if (status.mp_rebuilt && !status.runlist_merged) {
1293		if (ntfs_attr_record_resize(m, a, attr_rec_len)) {
1294			ntfs_error(vol->sb, "Failed to restore attribute "
1295					"record in error code path.  Run "
1296					"chkdsk to recover.");
1297			NVolSetErrors(vol);
1298		} else /* if (success) */ {
1299			if (ntfs_mapping_pairs_build(vol, (u8*)a +
1300					le16_to_cpu(a->data.non_resident.
1301					mapping_pairs_offset), attr_rec_len -
1302					le16_to_cpu(a->data.non_resident.
1303					mapping_pairs_offset), ni->runlist.rl,
1304					vcn, highest_vcn, NULL)) {
1305				ntfs_error(vol->sb, "Failed to restore "
1306						"mapping pairs array in error "
1307						"code path.  Run chkdsk to "
1308						"recover.");
1309				NVolSetErrors(vol);
1310			}
1311			flush_dcache_mft_record_page(ctx->ntfs_ino);
1312			mark_mft_record_dirty(ctx->ntfs_ino);
1313		}
1314	}
1315	/* Release the mft record and the attribute. */
1316	if (status.mft_attr_mapped) {
1317		ntfs_attr_put_search_ctx(ctx);
1318		unmap_mft_record(base_ni);
1319	}
1320	/* Release the runlist lock. */
1321	if (rl_write_locked)
1322		up_write(&ni->runlist.lock);
1323	else if (rl)
1324		up_read(&ni->runlist.lock);
1325	/*
1326	 * Zero out any newly allocated blocks to avoid exposing stale data.
1327	 * If BH_New is set, we know that the block was newly allocated above
1328	 * and that it has not been fully zeroed and marked dirty yet.
1329	 */
1330	nr_pages = u;
1331	u = 0;
1332	end = bh_cpos << vol->cluster_size_bits;
1333	do {
1334		page = pages[u];
1335		bh = head = page_buffers(page);
1336		do {
1337			if (u == nr_pages &&
1338					((s64)page->index << PAGE_SHIFT) +
1339					bh_offset(bh) >= end)
1340				break;
1341			if (!buffer_new(bh))
1342				continue;
1343			clear_buffer_new(bh);
1344			if (!buffer_uptodate(bh)) {
1345				if (PageUptodate(page))
1346					set_buffer_uptodate(bh);
1347				else {
1348					zero_user(page, bh_offset(bh),
1349							blocksize);
1350					set_buffer_uptodate(bh);
1351				}
1352			}
1353			mark_buffer_dirty(bh);
1354		} while ((bh = bh->b_this_page) != head);
1355	} while (++u <= nr_pages);
1356	ntfs_error(vol->sb, "Failed.  Returning error code %i.", err);
1357	return err;
1358}
1359
1360static inline void ntfs_flush_dcache_pages(struct page **pages,
1361		unsigned nr_pages)
1362{
1363	BUG_ON(!nr_pages);
1364	/*
1365	 * Warning: Do not do the decrement at the same time as the call to
1366	 * flush_dcache_page() because it is a NULL macro on i386 and hence the
1367	 * decrement never happens so the loop never terminates.
1368	 */
1369	do {
1370		--nr_pages;
1371		flush_dcache_page(pages[nr_pages]);
1372	} while (nr_pages > 0);
1373}
1374
1375/**
1376 * ntfs_commit_pages_after_non_resident_write - commit the received data
1377 * @pages:	array of destination pages
1378 * @nr_pages:	number of pages in @pages
1379 * @pos:	byte position in file at which the write begins
1380 * @bytes:	number of bytes to be written
1381 *
1382 * See description of ntfs_commit_pages_after_write(), below.
1383 */
1384static inline int ntfs_commit_pages_after_non_resident_write(
1385		struct page **pages, const unsigned nr_pages,
1386		s64 pos, size_t bytes)
1387{
1388	s64 end, initialized_size;
1389	struct inode *vi;
1390	ntfs_inode *ni, *base_ni;
1391	struct buffer_head *bh, *head;
1392	ntfs_attr_search_ctx *ctx;
1393	MFT_RECORD *m;
1394	ATTR_RECORD *a;
1395	unsigned long flags;
1396	unsigned blocksize, u;
1397	int err;
1398
1399	vi = pages[0]->mapping->host;
1400	ni = NTFS_I(vi);
1401	blocksize = vi->i_sb->s_blocksize;
1402	end = pos + bytes;
1403	u = 0;
1404	do {
1405		s64 bh_pos;
1406		struct page *page;
1407		bool partial;
1408
1409		page = pages[u];
1410		bh_pos = (s64)page->index << PAGE_SHIFT;
1411		bh = head = page_buffers(page);
1412		partial = false;
1413		do {
1414			s64 bh_end;
1415
1416			bh_end = bh_pos + blocksize;
1417			if (bh_end <= pos || bh_pos >= end) {
1418				if (!buffer_uptodate(bh))
1419					partial = true;
1420			} else {
1421				set_buffer_uptodate(bh);
1422				mark_buffer_dirty(bh);
1423			}
1424		} while (bh_pos += blocksize, (bh = bh->b_this_page) != head);
1425		/*
1426		 * If all buffers are now uptodate but the page is not, set the
1427		 * page uptodate.
1428		 */
1429		if (!partial && !PageUptodate(page))
1430			SetPageUptodate(page);
1431	} while (++u < nr_pages);
1432	/*
1433	 * Finally, if we do not need to update initialized_size or i_size we
1434	 * are finished.
1435	 */
1436	read_lock_irqsave(&ni->size_lock, flags);
1437	initialized_size = ni->initialized_size;
1438	read_unlock_irqrestore(&ni->size_lock, flags);
1439	if (end <= initialized_size) {
1440		ntfs_debug("Done.");
1441		return 0;
1442	}
1443	/*
1444	 * Update initialized_size/i_size as appropriate, both in the inode and
1445	 * the mft record.
1446	 */
1447	if (!NInoAttr(ni))
1448		base_ni = ni;
1449	else
1450		base_ni = ni->ext.base_ntfs_ino;
1451	/* Map, pin, and lock the mft record. */
1452	m = map_mft_record(base_ni);
1453	if (IS_ERR(m)) {
1454		err = PTR_ERR(m);
1455		m = NULL;
1456		ctx = NULL;
1457		goto err_out;
1458	}
1459	BUG_ON(!NInoNonResident(ni));
1460	ctx = ntfs_attr_get_search_ctx(base_ni, m);
1461	if (unlikely(!ctx)) {
1462		err = -ENOMEM;
1463		goto err_out;
1464	}
1465	err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len,
1466			CASE_SENSITIVE, 0, NULL, 0, ctx);
1467	if (unlikely(err)) {
1468		if (err == -ENOENT)
1469			err = -EIO;
1470		goto err_out;
1471	}
1472	a = ctx->attr;
1473	BUG_ON(!a->non_resident);
1474	write_lock_irqsave(&ni->size_lock, flags);
1475	BUG_ON(end > ni->allocated_size);
1476	ni->initialized_size = end;
1477	a->data.non_resident.initialized_size = cpu_to_sle64(end);
1478	if (end > i_size_read(vi)) {
1479		i_size_write(vi, end);
1480		a->data.non_resident.data_size =
1481				a->data.non_resident.initialized_size;
1482	}
1483	write_unlock_irqrestore(&ni->size_lock, flags);
1484	/* Mark the mft record dirty, so it gets written back. */
1485	flush_dcache_mft_record_page(ctx->ntfs_ino);
1486	mark_mft_record_dirty(ctx->ntfs_ino);
1487	ntfs_attr_put_search_ctx(ctx);
1488	unmap_mft_record(base_ni);
1489	ntfs_debug("Done.");
1490	return 0;
1491err_out:
1492	if (ctx)
1493		ntfs_attr_put_search_ctx(ctx);
1494	if (m)
1495		unmap_mft_record(base_ni);
1496	ntfs_error(vi->i_sb, "Failed to update initialized_size/i_size (error "
1497			"code %i).", err);
1498	if (err != -ENOMEM)
1499		NVolSetErrors(ni->vol);
1500	return err;
1501}
1502
1503/**
1504 * ntfs_commit_pages_after_write - commit the received data
1505 * @pages:	array of destination pages
1506 * @nr_pages:	number of pages in @pages
1507 * @pos:	byte position in file at which the write begins
1508 * @bytes:	number of bytes to be written
1509 *
1510 * This is called from ntfs_file_buffered_write() with i_mutex held on the inode
1511 * (@pages[0]->mapping->host).  There are @nr_pages pages in @pages which are
1512 * locked but not kmap()ped.  The source data has already been copied into the
1513 * @page.  ntfs_prepare_pages_for_non_resident_write() has been called before
1514 * the data was copied (for non-resident attributes only) and it returned
1515 * success.
1516 *
1517 * Need to set uptodate and mark dirty all buffers within the boundary of the
1518 * write.  If all buffers in a page are uptodate we set the page uptodate, too.
1519 *
1520 * Setting the buffers dirty ensures that they get written out later when
1521 * ntfs_writepage() is invoked by the VM.
1522 *
1523 * Finally, we need to update i_size and initialized_size as appropriate both
1524 * in the inode and the mft record.
1525 *
1526 * This is modelled after fs/buffer.c::generic_commit_write(), which marks
1527 * buffers uptodate and dirty, sets the page uptodate if all buffers in the
1528 * page are uptodate, and updates i_size if the end of io is beyond i_size.  In
1529 * that case, it also marks the inode dirty.
1530 *
1531 * If things have gone as outlined in
1532 * ntfs_prepare_pages_for_non_resident_write(), we do not need to do any page
1533 * content modifications here for non-resident attributes.  For resident
1534 * attributes we need to do the uptodate bringing here which we combine with
1535 * the copying into the mft record which means we save one atomic kmap.
1536 *
1537 * Return 0 on success or -errno on error.
1538 */
1539static int ntfs_commit_pages_after_write(struct page **pages,
1540		const unsigned nr_pages, s64 pos, size_t bytes)
1541{
1542	s64 end, initialized_size;
1543	loff_t i_size;
1544	struct inode *vi;
1545	ntfs_inode *ni, *base_ni;
1546	struct page *page;
1547	ntfs_attr_search_ctx *ctx;
1548	MFT_RECORD *m;
1549	ATTR_RECORD *a;
1550	char *kattr, *kaddr;
1551	unsigned long flags;
1552	u32 attr_len;
1553	int err;
1554
1555	BUG_ON(!nr_pages);
1556	BUG_ON(!pages);
1557	page = pages[0];
1558	BUG_ON(!page);
1559	vi = page->mapping->host;
1560	ni = NTFS_I(vi);
1561	ntfs_debug("Entering for inode 0x%lx, attribute type 0x%x, start page "
1562			"index 0x%lx, nr_pages 0x%x, pos 0x%llx, bytes 0x%zx.",
1563			vi->i_ino, ni->type, page->index, nr_pages,
1564			(long long)pos, bytes);
1565	if (NInoNonResident(ni))
1566		return ntfs_commit_pages_after_non_resident_write(pages,
1567				nr_pages, pos, bytes);
1568	BUG_ON(nr_pages > 1);
1569	/*
1570	 * Attribute is resident, implying it is not compressed, encrypted, or
1571	 * sparse.
1572	 */
1573	if (!NInoAttr(ni))
1574		base_ni = ni;
1575	else
1576		base_ni = ni->ext.base_ntfs_ino;
1577	BUG_ON(NInoNonResident(ni));
1578	/* Map, pin, and lock the mft record. */
1579	m = map_mft_record(base_ni);
1580	if (IS_ERR(m)) {
1581		err = PTR_ERR(m);
1582		m = NULL;
1583		ctx = NULL;
1584		goto err_out;
1585	}
1586	ctx = ntfs_attr_get_search_ctx(base_ni, m);
1587	if (unlikely(!ctx)) {
1588		err = -ENOMEM;
1589		goto err_out;
1590	}
1591	err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len,
1592			CASE_SENSITIVE, 0, NULL, 0, ctx);
1593	if (unlikely(err)) {
1594		if (err == -ENOENT)
1595			err = -EIO;
1596		goto err_out;
1597	}
1598	a = ctx->attr;
1599	BUG_ON(a->non_resident);
1600	/* The total length of the attribute value. */
1601	attr_len = le32_to_cpu(a->data.resident.value_length);
1602	i_size = i_size_read(vi);
1603	BUG_ON(attr_len != i_size);
1604	BUG_ON(pos > attr_len);
1605	end = pos + bytes;
1606	BUG_ON(end > le32_to_cpu(a->length) -
1607			le16_to_cpu(a->data.resident.value_offset));
1608	kattr = (u8*)a + le16_to_cpu(a->data.resident.value_offset);
1609	kaddr = kmap_atomic(page);
1610	/* Copy the received data from the page to the mft record. */
1611	memcpy(kattr + pos, kaddr + pos, bytes);
1612	/* Update the attribute length if necessary. */
1613	if (end > attr_len) {
1614		attr_len = end;
1615		a->data.resident.value_length = cpu_to_le32(attr_len);
1616	}
1617	/*
1618	 * If the page is not uptodate, bring the out of bounds area(s)
1619	 * uptodate by copying data from the mft record to the page.
1620	 */
1621	if (!PageUptodate(page)) {
1622		if (pos > 0)
1623			memcpy(kaddr, kattr, pos);
1624		if (end < attr_len)
1625			memcpy(kaddr + end, kattr + end, attr_len - end);
1626		/* Zero the region outside the end of the attribute value. */
1627		memset(kaddr + attr_len, 0, PAGE_SIZE - attr_len);
1628		flush_dcache_page(page);
1629		SetPageUptodate(page);
1630	}
1631	kunmap_atomic(kaddr);
1632	/* Update initialized_size/i_size if necessary. */
1633	read_lock_irqsave(&ni->size_lock, flags);
1634	initialized_size = ni->initialized_size;
1635	BUG_ON(end > ni->allocated_size);
1636	read_unlock_irqrestore(&ni->size_lock, flags);
1637	BUG_ON(initialized_size != i_size);
1638	if (end > initialized_size) {
1639		write_lock_irqsave(&ni->size_lock, flags);
1640		ni->initialized_size = end;
1641		i_size_write(vi, end);
1642		write_unlock_irqrestore(&ni->size_lock, flags);
1643	}
1644	/* Mark the mft record dirty, so it gets written back. */
1645	flush_dcache_mft_record_page(ctx->ntfs_ino);
1646	mark_mft_record_dirty(ctx->ntfs_ino);
1647	ntfs_attr_put_search_ctx(ctx);
1648	unmap_mft_record(base_ni);
1649	ntfs_debug("Done.");
1650	return 0;
1651err_out:
1652	if (err == -ENOMEM) {
1653		ntfs_warning(vi->i_sb, "Error allocating memory required to "
1654				"commit the write.");
1655		if (PageUptodate(page)) {
1656			ntfs_warning(vi->i_sb, "Page is uptodate, setting "
1657					"dirty so the write will be retried "
1658					"later on by the VM.");
1659			/*
1660			 * Put the page on mapping->dirty_pages, but leave its
1661			 * buffers' dirty state as-is.
1662			 */
1663			__set_page_dirty_nobuffers(page);
1664			err = 0;
1665		} else
1666			ntfs_error(vi->i_sb, "Page is not uptodate.  Written "
1667					"data has been lost.");
1668	} else {
1669		ntfs_error(vi->i_sb, "Resident attribute commit write failed "
1670				"with error %i.", err);
1671		NVolSetErrors(ni->vol);
1672	}
1673	if (ctx)
1674		ntfs_attr_put_search_ctx(ctx);
1675	if (m)
1676		unmap_mft_record(base_ni);
1677	return err;
1678}
1679
1680/*
1681 * Copy as much as we can into the pages and return the number of bytes which
1682 * were successfully copied.  If a fault is encountered then clear the pages
1683 * out to (ofs + bytes) and return the number of bytes which were copied.
1684 */
1685static size_t ntfs_copy_from_user_iter(struct page **pages, unsigned nr_pages,
1686		unsigned ofs, struct iov_iter *i, size_t bytes)
1687{
1688	struct page **last_page = pages + nr_pages;
1689	size_t total = 0;
1690	struct iov_iter data = *i;
1691	unsigned len, copied;
1692
1693	do {
1694		len = PAGE_SIZE - ofs;
1695		if (len > bytes)
1696			len = bytes;
1697		copied = iov_iter_copy_from_user_atomic(*pages, &data, ofs,
1698				len);
1699		total += copied;
1700		bytes -= copied;
1701		if (!bytes)
1702			break;
1703		iov_iter_advance(&data, copied);
1704		if (copied < len)
1705			goto err;
1706		ofs = 0;
1707	} while (++pages < last_page);
1708out:
1709	return total;
1710err:
1711	/* Zero the rest of the target like __copy_from_user(). */
1712	len = PAGE_SIZE - copied;
1713	do {
1714		if (len > bytes)
1715			len = bytes;
1716		zero_user(*pages, copied, len);
1717		bytes -= len;
1718		copied = 0;
1719		len = PAGE_SIZE;
1720	} while (++pages < last_page);
1721	goto out;
1722}
1723
1724/**
1725 * ntfs_perform_write - perform buffered write to a file
1726 * @file:	file to write to
1727 * @i:		iov_iter with data to write
1728 * @pos:	byte offset in file at which to begin writing to
1729 */
1730static ssize_t ntfs_perform_write(struct file *file, struct iov_iter *i,
1731		loff_t pos)
1732{
1733	struct address_space *mapping = file->f_mapping;
1734	struct inode *vi = mapping->host;
1735	ntfs_inode *ni = NTFS_I(vi);
1736	ntfs_volume *vol = ni->vol;
1737	struct page *pages[NTFS_MAX_PAGES_PER_CLUSTER];
1738	struct page *cached_page = NULL;
1739	VCN last_vcn;
1740	LCN lcn;
1741	size_t bytes;
1742	ssize_t status, written = 0;
1743	unsigned nr_pages;
1744
1745	ntfs_debug("Entering for i_ino 0x%lx, attribute type 0x%x, pos "
1746			"0x%llx, count 0x%lx.", vi->i_ino,
1747			(unsigned)le32_to_cpu(ni->type),
1748			(unsigned long long)pos,
1749			(unsigned long)iov_iter_count(i));
1750	/*
1751	 * If a previous ntfs_truncate() failed, repeat it and abort if it
1752	 * fails again.
1753	 */
1754	if (unlikely(NInoTruncateFailed(ni))) {
1755		int err;
1756
1757		inode_dio_wait(vi);
1758		err = ntfs_truncate(vi);
1759		if (err || NInoTruncateFailed(ni)) {
1760			if (!err)
1761				err = -EIO;
1762			ntfs_error(vol->sb, "Cannot perform write to inode "
1763					"0x%lx, attribute type 0x%x, because "
1764					"ntfs_truncate() failed (error code "
1765					"%i).", vi->i_ino,
1766					(unsigned)le32_to_cpu(ni->type), err);
1767			return err;
1768		}
1769	}
1770	/*
1771	 * Determine the number of pages per cluster for non-resident
1772	 * attributes.
1773	 */
1774	nr_pages = 1;
1775	if (vol->cluster_size > PAGE_SIZE && NInoNonResident(ni))
1776		nr_pages = vol->cluster_size >> PAGE_SHIFT;
1777	last_vcn = -1;
1778	do {
1779		VCN vcn;
1780		pgoff_t idx, start_idx;
1781		unsigned ofs, do_pages, u;
1782		size_t copied;
1783
1784		start_idx = idx = pos >> PAGE_SHIFT;
1785		ofs = pos & ~PAGE_MASK;
1786		bytes = PAGE_SIZE - ofs;
1787		do_pages = 1;
1788		if (nr_pages > 1) {
1789			vcn = pos >> vol->cluster_size_bits;
1790			if (vcn != last_vcn) {
1791				last_vcn = vcn;
1792				/*
1793				 * Get the lcn of the vcn the write is in.  If
1794				 * it is a hole, need to lock down all pages in
1795				 * the cluster.
1796				 */
1797				down_read(&ni->runlist.lock);
1798				lcn = ntfs_attr_vcn_to_lcn_nolock(ni, pos >>
1799						vol->cluster_size_bits, false);
1800				up_read(&ni->runlist.lock);
1801				if (unlikely(lcn < LCN_HOLE)) {
1802					if (lcn == LCN_ENOMEM)
1803						status = -ENOMEM;
1804					else {
1805						status = -EIO;
1806						ntfs_error(vol->sb, "Cannot "
1807							"perform write to "
1808							"inode 0x%lx, "
1809							"attribute type 0x%x, "
1810							"because the attribute "
1811							"is corrupt.",
1812							vi->i_ino, (unsigned)
1813							le32_to_cpu(ni->type));
1814					}
1815					break;
1816				}
1817				if (lcn == LCN_HOLE) {
1818					start_idx = (pos & ~(s64)
1819							vol->cluster_size_mask)
1820							>> PAGE_SHIFT;
1821					bytes = vol->cluster_size - (pos &
1822							vol->cluster_size_mask);
1823					do_pages = nr_pages;
1824				}
1825			}
1826		}
1827		if (bytes > iov_iter_count(i))
1828			bytes = iov_iter_count(i);
1829again:
1830		/*
1831		 * Bring in the user page(s) that we will copy from _first_.
1832		 * Otherwise there is a nasty deadlock on copying from the same
1833		 * page(s) as we are writing to, without it/them being marked
1834		 * up-to-date.  Note, at present there is nothing to stop the
1835		 * pages being swapped out between us bringing them into memory
1836		 * and doing the actual copying.
1837		 */
1838		if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
1839			status = -EFAULT;
1840			break;
1841		}
1842		/* Get and lock @do_pages starting at index @start_idx. */
1843		status = __ntfs_grab_cache_pages(mapping, start_idx, do_pages,
1844				pages, &cached_page);
1845		if (unlikely(status))
1846			break;
1847		/*
1848		 * For non-resident attributes, we need to fill any holes with
1849		 * actual clusters and ensure all bufferes are mapped.  We also
1850		 * need to bring uptodate any buffers that are only partially
1851		 * being written to.
1852		 */
1853		if (NInoNonResident(ni)) {
1854			status = ntfs_prepare_pages_for_non_resident_write(
1855					pages, do_pages, pos, bytes);
1856			if (unlikely(status)) {
1857				do {
1858					unlock_page(pages[--do_pages]);
1859					put_page(pages[do_pages]);
1860				} while (do_pages);
1861				break;
1862			}
1863		}
1864		u = (pos >> PAGE_SHIFT) - pages[0]->index;
1865		copied = ntfs_copy_from_user_iter(pages + u, do_pages - u, ofs,
1866					i, bytes);
1867		ntfs_flush_dcache_pages(pages + u, do_pages - u);
1868		status = 0;
1869		if (likely(copied == bytes)) {
1870			status = ntfs_commit_pages_after_write(pages, do_pages,
1871					pos, bytes);
1872			if (!status)
1873				status = bytes;
1874		}
1875		do {
1876			unlock_page(pages[--do_pages]);
1877			put_page(pages[do_pages]);
1878		} while (do_pages);
1879		if (unlikely(status < 0))
 
1880			break;
1881		copied = status;
1882		cond_resched();
1883		if (unlikely(!copied)) {
1884			size_t sc;
1885
1886			/*
1887			 * We failed to copy anything.  Fall back to single
1888			 * segment length write.
1889			 *
1890			 * This is needed to avoid possible livelock in the
1891			 * case that all segments in the iov cannot be copied
1892			 * at once without a pagefault.
1893			 */
1894			sc = iov_iter_single_seg_count(i);
1895			if (bytes > sc)
1896				bytes = sc;
1897			goto again;
1898		}
1899		iov_iter_advance(i, copied);
1900		pos += copied;
1901		written += copied;
1902		balance_dirty_pages_ratelimited(mapping);
1903		if (fatal_signal_pending(current)) {
1904			status = -EINTR;
1905			break;
1906		}
1907	} while (iov_iter_count(i));
1908	if (cached_page)
1909		put_page(cached_page);
1910	ntfs_debug("Done.  Returning %s (written 0x%lx, status %li).",
1911			written ? "written" : "status", (unsigned long)written,
1912			(long)status);
1913	return written ? written : status;
1914}
1915
1916/**
1917 * ntfs_file_write_iter - simple wrapper for ntfs_file_write_iter_nolock()
1918 * @iocb:	IO state structure
1919 * @from:	iov_iter with data to write
1920 *
1921 * Basically the same as generic_file_write_iter() except that it ends up
1922 * up calling ntfs_perform_write() instead of generic_perform_write() and that
1923 * O_DIRECT is not implemented.
1924 */
1925static ssize_t ntfs_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
1926{
1927	struct file *file = iocb->ki_filp;
1928	struct inode *vi = file_inode(file);
1929	ssize_t written = 0;
1930	ssize_t err;
1931
1932	inode_lock(vi);
1933	/* We can write back this queue in page reclaim. */
1934	current->backing_dev_info = inode_to_bdi(vi);
1935	err = ntfs_prepare_file_for_write(iocb, from);
1936	if (iov_iter_count(from) && !err)
1937		written = ntfs_perform_write(file, from, iocb->ki_pos);
1938	current->backing_dev_info = NULL;
1939	inode_unlock(vi);
1940	iocb->ki_pos += written;
1941	if (likely(written > 0))
1942		written = generic_write_sync(iocb, written);
1943	return written ? written : err;
1944}
1945
1946/**
1947 * ntfs_file_fsync - sync a file to disk
1948 * @filp:	file to be synced
1949 * @datasync:	if non-zero only flush user data and not metadata
1950 *
1951 * Data integrity sync of a file to disk.  Used for fsync, fdatasync, and msync
1952 * system calls.  This function is inspired by fs/buffer.c::file_fsync().
1953 *
1954 * If @datasync is false, write the mft record and all associated extent mft
1955 * records as well as the $DATA attribute and then sync the block device.
1956 *
1957 * If @datasync is true and the attribute is non-resident, we skip the writing
1958 * of the mft record and all associated extent mft records (this might still
1959 * happen due to the write_inode_now() call).
1960 *
1961 * Also, if @datasync is true, we do not wait on the inode to be written out
1962 * but we always wait on the page cache pages to be written out.
1963 *
1964 * Locking: Caller must hold i_mutex on the inode.
1965 *
1966 * TODO: We should probably also write all attribute/index inodes associated
1967 * with this inode but since we have no simple way of getting to them we ignore
1968 * this problem for now.
1969 */
1970static int ntfs_file_fsync(struct file *filp, loff_t start, loff_t end,
1971			   int datasync)
1972{
1973	struct inode *vi = filp->f_mapping->host;
1974	int err, ret = 0;
1975
1976	ntfs_debug("Entering for inode 0x%lx.", vi->i_ino);
1977
1978	err = file_write_and_wait_range(filp, start, end);
1979	if (err)
1980		return err;
1981	inode_lock(vi);
1982
1983	BUG_ON(S_ISDIR(vi->i_mode));
1984	if (!datasync || !NInoNonResident(NTFS_I(vi)))
1985		ret = __ntfs_write_inode(vi, 1);
1986	write_inode_now(vi, !datasync);
1987	/*
1988	 * NOTE: If we were to use mapping->private_list (see ext2 and
1989	 * fs/buffer.c) for dirty blocks then we could optimize the below to be
1990	 * sync_mapping_buffers(vi->i_mapping).
1991	 */
1992	err = sync_blockdev(vi->i_sb->s_bdev);
1993	if (unlikely(err && !ret))
1994		ret = err;
1995	if (likely(!ret))
1996		ntfs_debug("Done.");
1997	else
1998		ntfs_warning(vi->i_sb, "Failed to f%ssync inode 0x%lx.  Error "
1999				"%u.", datasync ? "data" : "", vi->i_ino, -ret);
2000	inode_unlock(vi);
2001	return ret;
2002}
2003
2004#endif /* NTFS_RW */
2005
2006const struct file_operations ntfs_file_ops = {
2007	.llseek		= generic_file_llseek,
2008	.read_iter	= generic_file_read_iter,
2009#ifdef NTFS_RW
2010	.write_iter	= ntfs_file_write_iter,
2011	.fsync		= ntfs_file_fsync,
2012#endif /* NTFS_RW */
2013	.mmap		= generic_file_mmap,
2014	.open		= ntfs_file_open,
2015	.splice_read	= generic_file_splice_read,
2016};
2017
2018const struct inode_operations ntfs_file_inode_ops = {
2019#ifdef NTFS_RW
2020	.setattr	= ntfs_setattr,
2021#endif /* NTFS_RW */
2022};
2023
2024const struct file_operations ntfs_empty_file_ops = {};
2025
2026const struct inode_operations ntfs_empty_inode_ops = {};