1/*
   2 * file.c - NTFS kernel file operations.  Part of the Linux-NTFS project.
   3 *
   4 * Copyright (c) 2001-2015 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/backing-dev.h>
  23#include <linux/buffer_head.h>
  24#include <linux/gfp.h>
  25#include <linux/pagemap.h>
  26#include <linux/pagevec.h>
  27#include <linux/sched.h>
  28#include <linux/swap.h>
  29#include <linux/uio.h>
  30#include <linux/writeback.h>
  31
  32#include <asm/page.h>
  33#include <asm/uaccess.h>
  34
  35#include "attrib.h"
  36#include "bitmap.h"
  37#include "inode.h"
  38#include "debug.h"
  39#include "lcnalloc.h"
  40#include "malloc.h"
  41#include "mft.h"
  42#include "ntfs.h"
  43
  44/**
  45 * ntfs_file_open - called when an inode is about to be opened
  46 * @vi:		inode to be opened
  47 * @filp:	file structure describing the inode
  48 *
  49 * Limit file size to the page cache limit on architectures where unsigned long
  50 * is 32-bits. This is the most we can do for now without overflowing the page
  51 * cache page index. Doing it this way means we don't run into problems because
  52 * of existing too large files. It would be better to allow the user to read
  53 * the beginning of the file but I doubt very much anyone is going to hit this
  54 * check on a 32-bit architecture, so there is no point in adding the extra
  55 * complexity required to support this.
  56 *
  57 * On 64-bit architectures, the check is hopefully optimized away by the
  58 * compiler.
  59 *
  60 * After the check passes, just call generic_file_open() to do its work.
  61 */
  62static int ntfs_file_open(struct inode *vi, struct file *filp)
  63{
  64	if (sizeof(unsigned long) < 8) {
  65		if (i_size_read(vi) > MAX_LFS_FILESIZE)
  66			return -EOVERFLOW;
  67	}
  68	return generic_file_open(vi, filp);
  69}
  70
  71#ifdef NTFS_RW
  72
  73/**
  74 * ntfs_attr_extend_initialized - extend the initialized size of an attribute
  75 * @ni:			ntfs inode of the attribute to extend
  76 * @new_init_size:	requested new initialized size in bytes
  77 *
  78 * Extend the initialized size of an attribute described by the ntfs inode @ni
  79 * to @new_init_size bytes.  This involves zeroing any non-sparse space between
  80 * the old initialized size and @new_init_size both in the page cache and on
  81 * disk (if relevant complete pages are already uptodate in the page cache then
  82 * these are simply marked dirty).
  83 *
  84 * As a side-effect, the file size (vfs inode->i_size) may be incremented as,
  85 * in the resident attribute case, it is tied to the initialized size and, in
  86 * the non-resident attribute case, it may not fall below the initialized size.
  87 *
  88 * Note that if the attribute is resident, we do not need to touch the page
  89 * cache at all.  This is because if the page cache page is not uptodate we
  90 * bring it uptodate later, when doing the write to the mft record since we
  91 * then already have the page mapped.  And if the page is uptodate, the
  92 * non-initialized region will already have been zeroed when the page was
  93 * brought uptodate and the region may in fact already have been overwritten
  94 * with new data via mmap() based writes, so we cannot just zero it.  And since
  95 * POSIX specifies that the behaviour of resizing a file whilst it is mmap()ped
  96 * is unspecified, we choose not to do zeroing and thus we do not need to touch
  97 * the page at all.  For a more detailed explanation see ntfs_truncate() in
  98 * fs/ntfs/inode.c.
  99 *
 100 * Return 0 on success and -errno on error.  In the case that an error is
 101 * encountered it is possible that the initialized size will already have been
 102 * incremented some way towards @new_init_size but it is guaranteed that if
 103 * this is the case, the necessary zeroing will also have happened and that all
 104 * metadata is self-consistent.
 105 *
 106 * Locking: i_mutex on the vfs inode corrseponsind to the ntfs inode @ni must be
 107 *	    held by the caller.
 108 */
 109static int ntfs_attr_extend_initialized(ntfs_inode *ni, const s64 new_init_size)
 110{
 111	s64 old_init_size;
 112	loff_t old_i_size;
 113	pgoff_t index, end_index;
 114	unsigned long flags;
 115	struct inode *vi = VFS_I(ni);
 116	ntfs_inode *base_ni;
 117	MFT_RECORD *m = NULL;
 118	ATTR_RECORD *a;
 119	ntfs_attr_search_ctx *ctx = NULL;
 120	struct address_space *mapping;
 121	struct page *page = NULL;
 122	u8 *kattr;
 123	int err;
 124	u32 attr_len;
 125
 126	read_lock_irqsave(&ni->size_lock, flags);
 127	old_init_size = ni->initialized_size;
 128	old_i_size = i_size_read(vi);
 129	BUG_ON(new_init_size > ni->allocated_size);
 130	read_unlock_irqrestore(&ni->size_lock, flags);
 131	ntfs_debug("Entering for i_ino 0x%lx, attribute type 0x%x, "
 132			"old_initialized_size 0x%llx, "
 133			"new_initialized_size 0x%llx, i_size 0x%llx.",
 134			vi->i_ino, (unsigned)le32_to_cpu(ni->type),
 135			(unsigned long long)old_init_size,
 136			(unsigned long long)new_init_size, old_i_size);
 137	if (!NInoAttr(ni))
 138		base_ni = ni;
 139	else
 140		base_ni = ni->ext.base_ntfs_ino;
 141	/* Use goto to reduce indentation and we need the label below anyway. */
 142	if (NInoNonResident(ni))
 143		goto do_non_resident_extend;
 144	BUG_ON(old_init_size != old_i_size);
 145	m = map_mft_record(base_ni);
 146	if (IS_ERR(m)) {
 147		err = PTR_ERR(m);
 148		m = NULL;
 149		goto err_out;
 150	}
 151	ctx = ntfs_attr_get_search_ctx(base_ni, m);
 152	if (unlikely(!ctx)) {
 153		err = -ENOMEM;
 154		goto err_out;
 155	}
 156	err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len,
 157			CASE_SENSITIVE, 0, NULL, 0, ctx);
 158	if (unlikely(err)) {
 159		if (err == -ENOENT)
 160			err = -EIO;
 161		goto err_out;
 162	}
 163	m = ctx->mrec;
 164	a = ctx->attr;
 165	BUG_ON(a->non_resident);
 166	/* The total length of the attribute value. */
 167	attr_len = le32_to_cpu(a->data.resident.value_length);
 168	BUG_ON(old_i_size != (loff_t)attr_len);
 169	/*
 170	 * Do the zeroing in the mft record and update the attribute size in
 171	 * the mft record.
 172	 */
 173	kattr = (u8*)a + le16_to_cpu(a->data.resident.value_offset);
 174	memset(kattr + attr_len, 0, new_init_size - attr_len);
 175	a->data.resident.value_length = cpu_to_le32((u32)new_init_size);
 176	/* Finally, update the sizes in the vfs and ntfs inodes. */
 177	write_lock_irqsave(&ni->size_lock, flags);
 178	i_size_write(vi, new_init_size);
 179	ni->initialized_size = new_init_size;
 180	write_unlock_irqrestore(&ni->size_lock, flags);
 181	goto done;
 182do_non_resident_extend:
 183	/*
 184	 * If the new initialized size @new_init_size exceeds the current file
 185	 * size (vfs inode->i_size), we need to extend the file size to the
 186	 * new initialized size.
 187	 */
 188	if (new_init_size > old_i_size) {
 189		m = map_mft_record(base_ni);
 190		if (IS_ERR(m)) {
 191			err = PTR_ERR(m);
 192			m = NULL;
 193			goto err_out;
 194		}
 195		ctx = ntfs_attr_get_search_ctx(base_ni, m);
 196		if (unlikely(!ctx)) {
 197			err = -ENOMEM;
 198			goto err_out;
 199		}
 200		err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len,
 201				CASE_SENSITIVE, 0, NULL, 0, ctx);
 202		if (unlikely(err)) {
 203			if (err == -ENOENT)
 204				err = -EIO;
 205			goto err_out;
 206		}
 207		m = ctx->mrec;
 208		a = ctx->attr;
 209		BUG_ON(!a->non_resident);
 210		BUG_ON(old_i_size != (loff_t)
 211				sle64_to_cpu(a->data.non_resident.data_size));
 212		a->data.non_resident.data_size = cpu_to_sle64(new_init_size);
 213		flush_dcache_mft_record_page(ctx->ntfs_ino);
 214		mark_mft_record_dirty(ctx->ntfs_ino);
 215		/* Update the file size in the vfs inode. */
 216		i_size_write(vi, new_init_size);
 217		ntfs_attr_put_search_ctx(ctx);
 218		ctx = NULL;
 219		unmap_mft_record(base_ni);
 220		m = NULL;
 221	}
 222	mapping = vi->i_mapping;
 223	index = old_init_size >> PAGE_SHIFT;
 224	end_index = (new_init_size + PAGE_SIZE - 1) >> PAGE_SHIFT;
 225	do {
 226		/*
 227		 * Read the page.  If the page is not present, this will zero
 228		 * the uninitialized regions for us.
 229		 */
 230		page = read_mapping_page(mapping, index, NULL);
 231		if (IS_ERR(page)) {
 232			err = PTR_ERR(page);
 233			goto init_err_out;
 234		}
 235		if (unlikely(PageError(page))) {
 236			put_page(page);
 237			err = -EIO;
 238			goto init_err_out;
 239		}
 240		/*
 241		 * Update the initialized size in the ntfs inode.  This is
 242		 * enough to make ntfs_writepage() work.
 243		 */
 244		write_lock_irqsave(&ni->size_lock, flags);
 245		ni->initialized_size = (s64)(index + 1) << PAGE_SHIFT;
 246		if (ni->initialized_size > new_init_size)
 247			ni->initialized_size = new_init_size;
 248		write_unlock_irqrestore(&ni->size_lock, flags);
 249		/* Set the page dirty so it gets written out. */
 250		set_page_dirty(page);
 251		put_page(page);
 252		/*
 253		 * Play nice with the vm and the rest of the system.  This is
 254		 * very much needed as we can potentially be modifying the
 255		 * initialised size from a very small value to a really huge
 256		 * value, e.g.
 257		 *	f = open(somefile, O_TRUNC);
 258		 *	truncate(f, 10GiB);
 259		 *	seek(f, 10GiB);
 260		 *	write(f, 1);
 261		 * And this would mean we would be marking dirty hundreds of
 262		 * thousands of pages or as in the above example more than
 263		 * two and a half million pages!
 264		 *
 265		 * TODO: For sparse pages could optimize this workload by using
 266		 * the FsMisc / MiscFs page bit as a "PageIsSparse" bit.  This
 267		 * would be set in readpage for sparse pages and here we would
 268		 * not need to mark dirty any pages which have this bit set.
 269		 * The only caveat is that we have to clear the bit everywhere
 270		 * where we allocate any clusters that lie in the page or that
 271		 * contain the page.
 272		 *
 273		 * TODO: An even greater optimization would be for us to only
 274		 * call readpage() on pages which are not in sparse regions as
 275		 * determined from the runlist.  This would greatly reduce the
 276		 * number of pages we read and make dirty in the case of sparse
 277		 * files.
 278		 */
 279		balance_dirty_pages_ratelimited(mapping);
 280		cond_resched();
 281	} while (++index < end_index);
 282	read_lock_irqsave(&ni->size_lock, flags);
 283	BUG_ON(ni->initialized_size != new_init_size);
 284	read_unlock_irqrestore(&ni->size_lock, flags);
 285	/* Now bring in sync the initialized_size in the mft record. */
 286	m = map_mft_record(base_ni);
 287	if (IS_ERR(m)) {
 288		err = PTR_ERR(m);
 289		m = NULL;
 290		goto init_err_out;
 291	}
 292	ctx = ntfs_attr_get_search_ctx(base_ni, m);
 293	if (unlikely(!ctx)) {
 294		err = -ENOMEM;
 295		goto init_err_out;
 296	}
 297	err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len,
 298			CASE_SENSITIVE, 0, NULL, 0, ctx);
 299	if (unlikely(err)) {
 300		if (err == -ENOENT)
 301			err = -EIO;
 302		goto init_err_out;
 303	}
 304	m = ctx->mrec;
 305	a = ctx->attr;
 306	BUG_ON(!a->non_resident);
 307	a->data.non_resident.initialized_size = cpu_to_sle64(new_init_size);
 308done:
 309	flush_dcache_mft_record_page(ctx->ntfs_ino);
 310	mark_mft_record_dirty(ctx->ntfs_ino);
 311	if (ctx)
 312		ntfs_attr_put_search_ctx(ctx);
 313	if (m)
 314		unmap_mft_record(base_ni);
 315	ntfs_debug("Done, initialized_size 0x%llx, i_size 0x%llx.",
 316			(unsigned long long)new_init_size, i_size_read(vi));
 317	return 0;
 318init_err_out:
 319	write_lock_irqsave(&ni->size_lock, flags);
 320	ni->initialized_size = old_init_size;
 321	write_unlock_irqrestore(&ni->size_lock, flags);
 322err_out:
 323	if (ctx)
 324		ntfs_attr_put_search_ctx(ctx);
 325	if (m)
 326		unmap_mft_record(base_ni);
 327	ntfs_debug("Failed.  Returning error code %i.", err);
 328	return err;
 329}
 330
 331static ssize_t ntfs_prepare_file_for_write(struct kiocb *iocb,
 332		struct iov_iter *from)
 333{
 334	loff_t pos;
 335	s64 end, ll;
 336	ssize_t err;
 337	unsigned long flags;
 338	struct file *file = iocb->ki_filp;
 339	struct inode *vi = file_inode(file);
 340	ntfs_inode *base_ni, *ni = NTFS_I(vi);
 341	ntfs_volume *vol = ni->vol;
 342
 343	ntfs_debug("Entering for i_ino 0x%lx, attribute type 0x%x, pos "
 344			"0x%llx, count 0x%zx.", vi->i_ino,
 345			(unsigned)le32_to_cpu(ni->type),
 346			(unsigned long long)iocb->ki_pos,
 347			iov_iter_count(from));
 348	err = generic_write_checks(iocb, from);
 349	if (unlikely(err <= 0))
 350		goto out;
 351	/*
 352	 * All checks have passed.  Before we start doing any writing we want
 353	 * to abort any totally illegal writes.
 354	 */
 355	BUG_ON(NInoMstProtected(ni));
 356	BUG_ON(ni->type != AT_DATA);
 357	/* If file is encrypted, deny access, just like NT4. */
 358	if (NInoEncrypted(ni)) {
 359		/* Only $DATA attributes can be encrypted. */
 360		/*
 361		 * Reminder for later: Encrypted files are _always_
 362		 * non-resident so that the content can always be encrypted.
 363		 */
 364		ntfs_debug("Denying write access to encrypted file.");
 365		err = -EACCES;
 366		goto out;
 367	}
 368	if (NInoCompressed(ni)) {
 369		/* Only unnamed $DATA attribute can be compressed. */
 370		BUG_ON(ni->name_len);
 371		/*
 372		 * Reminder for later: If resident, the data is not actually
 373		 * compressed.  Only on the switch to non-resident does
 374		 * compression kick in.  This is in contrast to encrypted files
 375		 * (see above).
 376		 */
 377		ntfs_error(vi->i_sb, "Writing to compressed files is not "
 378				"implemented yet.  Sorry.");
 379		err = -EOPNOTSUPP;
 380		goto out;
 381	}
 382	base_ni = ni;
 383	if (NInoAttr(ni))
 384		base_ni = ni->ext.base_ntfs_ino;
 385	err = file_remove_privs(file);
 386	if (unlikely(err))
 387		goto out;
 388	/*
 389	 * Our ->update_time method always succeeds thus file_update_time()
 390	 * cannot fail either so there is no need to check the return code.
 391	 */
 392	file_update_time(file);
 393	pos = iocb->ki_pos;
 394	/* The first byte after the last cluster being written to. */
 395	end = (pos + iov_iter_count(from) + vol->cluster_size_mask) &
 396			~(u64)vol->cluster_size_mask;
 397	/*
 398	 * If the write goes beyond the allocated size, extend the allocation
 399	 * to cover the whole of the write, rounded up to the nearest cluster.
 400	 */
 401	read_lock_irqsave(&ni->size_lock, flags);
 402	ll = ni->allocated_size;
 403	read_unlock_irqrestore(&ni->size_lock, flags);
 404	if (end > ll) {
 405		/*
 406		 * Extend the allocation without changing the data size.
 407		 *
 408		 * Note we ensure the allocation is big enough to at least
 409		 * write some data but we do not require the allocation to be
 410		 * complete, i.e. it may be partial.
 411		 */
 412		ll = ntfs_attr_extend_allocation(ni, end, -1, pos);
 413		if (likely(ll >= 0)) {
 414			BUG_ON(pos >= ll);
 415			/* If the extension was partial truncate the write. */
 416			if (end > ll) {
 417				ntfs_debug("Truncating write to inode 0x%lx, "
 418						"attribute type 0x%x, because "
 419						"the allocation was only "
 420						"partially extended.",
 421						vi->i_ino, (unsigned)
 422						le32_to_cpu(ni->type));
 423				iov_iter_truncate(from, ll - pos);
 424			}
 425		} else {
 426			err = ll;
 427			read_lock_irqsave(&ni->size_lock, flags);
 428			ll = ni->allocated_size;
 429			read_unlock_irqrestore(&ni->size_lock, flags);
 430			/* Perform a partial write if possible or fail. */
 431			if (pos < ll) {
 432				ntfs_debug("Truncating write to inode 0x%lx "
 433						"attribute type 0x%x, because "
 434						"extending the allocation "
 435						"failed (error %d).",
 436						vi->i_ino, (unsigned)
 437						le32_to_cpu(ni->type),
 438						(int)-err);
 439				iov_iter_truncate(from, ll - pos);
 440			} else {
 441				if (err != -ENOSPC)
 442					ntfs_error(vi->i_sb, "Cannot perform "
 443							"write to inode "
 444							"0x%lx, attribute "
 445							"type 0x%x, because "
 446							"extending the "
 447							"allocation failed "
 448							"(error %ld).",
 449							vi->i_ino, (unsigned)
 450							le32_to_cpu(ni->type),
 451							(long)-err);
 452				else
 453					ntfs_debug("Cannot perform write to "
 454							"inode 0x%lx, "
 455							"attribute type 0x%x, "
 456							"because there is not "
 457							"space left.",
 458							vi->i_ino, (unsigned)
 459							le32_to_cpu(ni->type));
 460				goto out;
 461			}
 462		}
 463	}
 464	/*
 465	 * If the write starts beyond the initialized size, extend it up to the
 466	 * beginning of the write and initialize all non-sparse space between
 467	 * the old initialized size and the new one.  This automatically also
 468	 * increments the vfs inode->i_size to keep it above or equal to the
 469	 * initialized_size.
 470	 */
 471	read_lock_irqsave(&ni->size_lock, flags);
 472	ll = ni->initialized_size;
 473	read_unlock_irqrestore(&ni->size_lock, flags);
 474	if (pos > ll) {
 475		/*
 476		 * Wait for ongoing direct i/o to complete before proceeding.
 477		 * New direct i/o cannot start as we hold i_mutex.
 478		 */
 479		inode_dio_wait(vi);
 480		err = ntfs_attr_extend_initialized(ni, pos);
 481		if (unlikely(err < 0))
 482			ntfs_error(vi->i_sb, "Cannot perform write to inode "
 483					"0x%lx, attribute type 0x%x, because "
 484					"extending the initialized size "
 485					"failed (error %d).", vi->i_ino,
 486					(unsigned)le32_to_cpu(ni->type),
 487					(int)-err);
 488	}
 489out:
 490	return err;
 491}
 492
 493/**
 494 * __ntfs_grab_cache_pages - obtain a number of locked pages
 495 * @mapping:	address space mapping from which to obtain page cache pages
 496 * @index:	starting index in @mapping at which to begin obtaining pages
 497 * @nr_pages:	number of page cache pages to obtain
 498 * @pages:	array of pages in which to return the obtained page cache pages
 499 * @cached_page: allocated but as yet unused page
 500 *
 501 * Obtain @nr_pages locked page cache pages from the mapping @mapping and
 502 * starting at index @index.
 503 *
 504 * If a page is newly created, add it to lru list
 505 *
 506 * Note, the page locks are obtained in ascending page index order.
 507 */
 508static inline int __ntfs_grab_cache_pages(struct address_space *mapping,
 509		pgoff_t index, const unsigned nr_pages, struct page **pages,
 510		struct page **cached_page)
 511{
 512	int err, nr;
 513
 514	BUG_ON(!nr_pages);
 515	err = nr = 0;
 516	do {
 517		pages[nr] = find_get_page_flags(mapping, index, FGP_LOCK |
 518				FGP_ACCESSED);
 519		if (!pages[nr]) {
 520			if (!*cached_page) {
 521				*cached_page = page_cache_alloc(mapping);
 522				if (unlikely(!*cached_page)) {
 523					err = -ENOMEM;
 524					goto err_out;
 525				}
 526			}
 527			err = add_to_page_cache_lru(*cached_page, mapping,
 528				   index,
 529				   mapping_gfp_constraint(mapping, GFP_KERNEL));
 530			if (unlikely(err)) {
 531				if (err == -EEXIST)
 532					continue;
 533				goto err_out;
 534			}
 535			pages[nr] = *cached_page;
 536			*cached_page = NULL;
 537		}
 538		index++;
 539		nr++;
 540	} while (nr < nr_pages);
 541out:
 542	return err;
 543err_out:
 544	while (nr > 0) {
 545		unlock_page(pages[--nr]);
 546		put_page(pages[nr]);
 547	}
 548	goto out;
 549}
 550
 551static inline int ntfs_submit_bh_for_read(struct buffer_head *bh)
 552{
 553	lock_buffer(bh);
 554	get_bh(bh);
 555	bh->b_end_io = end_buffer_read_sync;
 556	return submit_bh(READ, bh);
 557}
 558
 559/**
 560 * ntfs_prepare_pages_for_non_resident_write - prepare pages for receiving data
 561 * @pages:	array of destination pages
 562 * @nr_pages:	number of pages in @pages
 563 * @pos:	byte position in file at which the write begins
 564 * @bytes:	number of bytes to be written
 565 *
 566 * This is called for non-resident attributes from ntfs_file_buffered_write()
 567 * with i_mutex held on the inode (@pages[0]->mapping->host).  There are
 568 * @nr_pages pages in @pages which are locked but not kmap()ped.  The source
 569 * data has not yet been copied into the @pages.
 570 * 
 571 * Need to fill any holes with actual clusters, allocate buffers if necessary,
 572 * ensure all the buffers are mapped, and bring uptodate any buffers that are
 573 * only partially being written to.
 574 *
 575 * If @nr_pages is greater than one, we are guaranteed that the cluster size is
 576 * greater than PAGE_SIZE, that all pages in @pages are entirely inside
 577 * the same cluster and that they are the entirety of that cluster, and that
 578 * the cluster is sparse, i.e. we need to allocate a cluster to fill the hole.
 579 *
 580 * i_size is not to be modified yet.
 581 *
 582 * Return 0 on success or -errno on error.
 583 */
 584static int ntfs_prepare_pages_for_non_resident_write(struct page **pages,
 585		unsigned nr_pages, s64 pos, size_t bytes)
 586{
 587	VCN vcn, highest_vcn = 0, cpos, cend, bh_cpos, bh_cend;
 588	LCN lcn;
 589	s64 bh_pos, vcn_len, end, initialized_size;
 590	sector_t lcn_block;
 591	struct page *page;
 592	struct inode *vi;
 593	ntfs_inode *ni, *base_ni = NULL;
 594	ntfs_volume *vol;
 595	runlist_element *rl, *rl2;
 596	struct buffer_head *bh, *head, *wait[2], **wait_bh = wait;
 597	ntfs_attr_search_ctx *ctx = NULL;
 598	MFT_RECORD *m = NULL;
 599	ATTR_RECORD *a = NULL;
 600	unsigned long flags;
 601	u32 attr_rec_len = 0;
 602	unsigned blocksize, u;
 603	int err, mp_size;
 604	bool rl_write_locked, was_hole, is_retry;
 605	unsigned char blocksize_bits;
 606	struct {
 607		u8 runlist_merged:1;
 608		u8 mft_attr_mapped:1;
 609		u8 mp_rebuilt:1;
 610		u8 attr_switched:1;
 611	} status = { 0, 0, 0, 0 };
 612
 613	BUG_ON(!nr_pages);
 614	BUG_ON(!pages);
 615	BUG_ON(!*pages);
 616	vi = pages[0]->mapping->host;
 617	ni = NTFS_I(vi);
 618	vol = ni->vol;
 619	ntfs_debug("Entering for inode 0x%lx, attribute type 0x%x, start page "
 620			"index 0x%lx, nr_pages 0x%x, pos 0x%llx, bytes 0x%zx.",
 621			vi->i_ino, ni->type, pages[0]->index, nr_pages,
 622			(long long)pos, bytes);
 623	blocksize = vol->sb->s_blocksize;
 624	blocksize_bits = vol->sb->s_blocksize_bits;
 625	u = 0;
 626	do {
 627		page = pages[u];
 628		BUG_ON(!page);
 629		/*
 630		 * create_empty_buffers() will create uptodate/dirty buffers if
 631		 * the page is uptodate/dirty.
 632		 */
 633		if (!page_has_buffers(page)) {
 634			create_empty_buffers(page, blocksize, 0);
 635			if (unlikely(!page_has_buffers(page)))
 636				return -ENOMEM;
 637		}
 638	} while (++u < nr_pages);
 639	rl_write_locked = false;
 640	rl = NULL;
 641	err = 0;
 642	vcn = lcn = -1;
 643	vcn_len = 0;
 644	lcn_block = -1;
 645	was_hole = false;
 646	cpos = pos >> vol->cluster_size_bits;
 647	end = pos + bytes;
 648	cend = (end + vol->cluster_size - 1) >> vol->cluster_size_bits;
 649	/*
 650	 * Loop over each page and for each page over each buffer.  Use goto to
 651	 * reduce indentation.
 652	 */
 653	u = 0;
 654do_next_page:
 655	page = pages[u];
 656	bh_pos = (s64)page->index << PAGE_SHIFT;
 657	bh = head = page_buffers(page);
 658	do {
 659		VCN cdelta;
 660		s64 bh_end;
 661		unsigned bh_cofs;
 662
 663		/* Clear buffer_new on all buffers to reinitialise state. */
 664		if (buffer_new(bh))
 665			clear_buffer_new(bh);
 666		bh_end = bh_pos + blocksize;
 667		bh_cpos = bh_pos >> vol->cluster_size_bits;
 668		bh_cofs = bh_pos & vol->cluster_size_mask;
 669		if (buffer_mapped(bh)) {
 670			/*
 671			 * The buffer is already mapped.  If it is uptodate,
 672			 * ignore it.
 673			 */
 674			if (buffer_uptodate(bh))
 675				continue;
 676			/*
 677			 * The buffer is not uptodate.  If the page is uptodate
 678			 * set the buffer uptodate and otherwise ignore it.
 679			 */
 680			if (PageUptodate(page)) {
 681				set_buffer_uptodate(bh);
 682				continue;
 683			}
 684			/*
 685			 * Neither the page nor the buffer are uptodate.  If
 686			 * the buffer is only partially being written to, we
 687			 * need to read it in before the write, i.e. now.
 688			 */
 689			if ((bh_pos < pos && bh_end > pos) ||
 690					(bh_pos < end && bh_end > end)) {
 691				/*
 692				 * If the buffer is fully or partially within
 693				 * the initialized size, do an actual read.
 694				 * Otherwise, simply zero the buffer.
 695				 */
 696				read_lock_irqsave(&ni->size_lock, flags);
 697				initialized_size = ni->initialized_size;
 698				read_unlock_irqrestore(&ni->size_lock, flags);
 699				if (bh_pos < initialized_size) {
 700					ntfs_submit_bh_for_read(bh);
 701					*wait_bh++ = bh;
 702				} else {
 703					zero_user(page, bh_offset(bh),
 704							blocksize);
 705					set_buffer_uptodate(bh);
 706				}
 707			}
 708			continue;
 709		}
 710		/* Unmapped buffer.  Need to map it. */
 711		bh->b_bdev = vol->sb->s_bdev;
 712		/*
 713		 * If the current buffer is in the same clusters as the map
 714		 * cache, there is no need to check the runlist again.  The
 715		 * map cache is made up of @vcn, which is the first cached file
 716		 * cluster, @vcn_len which is the number of cached file
 717		 * clusters, @lcn is the device cluster corresponding to @vcn,
 718		 * and @lcn_block is the block number corresponding to @lcn.
 719		 */
 720		cdelta = bh_cpos - vcn;
 721		if (likely(!cdelta || (cdelta > 0 && cdelta < vcn_len))) {
 722map_buffer_cached:
 723			BUG_ON(lcn < 0);
 724			bh->b_blocknr = lcn_block +
 725					(cdelta << (vol->cluster_size_bits -
 726					blocksize_bits)) +
 727					(bh_cofs >> blocksize_bits);
 728			set_buffer_mapped(bh);
 729			/*
 730			 * If the page is uptodate so is the buffer.  If the
 731			 * buffer is fully outside the write, we ignore it if
 732			 * it was already allocated and we mark it dirty so it
 733			 * gets written out if we allocated it.  On the other
 734			 * hand, if we allocated the buffer but we are not
 735			 * marking it dirty we set buffer_new so we can do
 736			 * error recovery.
 737			 */
 738			if (PageUptodate(page)) {
 739				if (!buffer_uptodate(bh))
 740					set_buffer_uptodate(bh);
 741				if (unlikely(was_hole)) {
 742					/* We allocated the buffer. */
 743					unmap_underlying_metadata(bh->b_bdev,
 744							bh->b_blocknr);
 745					if (bh_end <= pos || bh_pos >= end)
 746						mark_buffer_dirty(bh);
 747					else
 748						set_buffer_new(bh);
 749				}
 750				continue;
 751			}
 752			/* Page is _not_ uptodate. */
 753			if (likely(!was_hole)) {
 754				/*
 755				 * Buffer was already allocated.  If it is not
 756				 * uptodate and is only partially being written
 757				 * to, we need to read it in before the write,
 758				 * i.e. now.
 759				 */
 760				if (!buffer_uptodate(bh) && bh_pos < end &&
 761						bh_end > pos &&
 762						(bh_pos < pos ||
 763						bh_end > end)) {
 764					/*
 765					 * If the buffer is fully or partially
 766					 * within the initialized size, do an
 767					 * actual read.  Otherwise, simply zero
 768					 * the buffer.
 769					 */
 770					read_lock_irqsave(&ni->size_lock,
 771							flags);
 772					initialized_size = ni->initialized_size;
 773					read_unlock_irqrestore(&ni->size_lock,
 774							flags);
 775					if (bh_pos < initialized_size) {
 776						ntfs_submit_bh_for_read(bh);
 777						*wait_bh++ = bh;
 778					} else {
 779						zero_user(page, bh_offset(bh),
 780								blocksize);
 781						set_buffer_uptodate(bh);
 782					}
 783				}
 784				continue;
 785			}
 786			/* We allocated the buffer. */
 787			unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr);
 788			/*
 789			 * If the buffer is fully outside the write, zero it,
 790			 * set it uptodate, and mark it dirty so it gets
 791			 * written out.  If it is partially being written to,
 792			 * zero region surrounding the write but leave it to
 793			 * commit write to do anything else.  Finally, if the
 794			 * buffer is fully being overwritten, do nothing.
 795			 */
 796			if (bh_end <= pos || bh_pos >= end) {
 797				if (!buffer_uptodate(bh)) {
 798					zero_user(page, bh_offset(bh),
 799							blocksize);
 800					set_buffer_uptodate(bh);
 801				}
 802				mark_buffer_dirty(bh);
 803				continue;
 804			}
 805			set_buffer_new(bh);
 806			if (!buffer_uptodate(bh) &&
 807					(bh_pos < pos || bh_end > end)) {
 808				u8 *kaddr;
 809				unsigned pofs;
 810					
 811				kaddr = kmap_atomic(page);
 812				if (bh_pos < pos) {
 813					pofs = bh_pos & ~PAGE_MASK;
 814					memset(kaddr + pofs, 0, pos - bh_pos);
 815				}
 816				if (bh_end > end) {
 817					pofs = end & ~PAGE_MASK;
 818					memset(kaddr + pofs, 0, bh_end - end);
 819				}
 820				kunmap_atomic(kaddr);
 821				flush_dcache_page(page);
 822			}
 823			continue;
 824		}
 825		/*
 826		 * Slow path: this is the first buffer in the cluster.  If it
 827		 * is outside allocated size and is not uptodate, zero it and
 828		 * set it uptodate.
 829		 */
 830		read_lock_irqsave(&ni->size_lock, flags);
 831		initialized_size = ni->allocated_size;
 832		read_unlock_irqrestore(&ni->size_lock, flags);
 833		if (bh_pos > initialized_size) {
 834			if (PageUptodate(page)) {
 835				if (!buffer_uptodate(bh))
 836					set_buffer_uptodate(bh);
 837			} else if (!buffer_uptodate(bh)) {
 838				zero_user(page, bh_offset(bh), blocksize);
 839				set_buffer_uptodate(bh);
 840			}
 841			continue;
 842		}
 843		is_retry = false;
 844		if (!rl) {
 845			down_read(&ni->runlist.lock);
 846retry_remap:
 847			rl = ni->runlist.rl;
 848		}
 849		if (likely(rl != NULL)) {
 850			/* Seek to element containing target cluster. */
 851			while (rl->length && rl[1].vcn <= bh_cpos)
 852				rl++;
 853			lcn = ntfs_rl_vcn_to_lcn(rl, bh_cpos);
 854			if (likely(lcn >= 0)) {
 855				/*
 856				 * Successful remap, setup the map cache and
 857				 * use that to deal with the buffer.
 858				 */
 859				was_hole = false;
 860				vcn = bh_cpos;
 861				vcn_len = rl[1].vcn - vcn;
 862				lcn_block = lcn << (vol->cluster_size_bits -
 863						blocksize_bits);
 864				cdelta = 0;
 865				/*
 866				 * If the number of remaining clusters touched
 867				 * by the write is smaller or equal to the
 868				 * number of cached clusters, unlock the
 869				 * runlist as the map cache will be used from
 870				 * now on.
 871				 */
 872				if (likely(vcn + vcn_len >= cend)) {
 873					if (rl_write_locked) {
 874						up_write(&ni->runlist.lock);
 875						rl_write_locked = false;
 876					} else
 877						up_read(&ni->runlist.lock);
 878					rl = NULL;
 879				}
 880				goto map_buffer_cached;
 881			}
 882		} else
 883			lcn = LCN_RL_NOT_MAPPED;
 884		/*
 885		 * If it is not a hole and not out of bounds, the runlist is
 886		 * probably unmapped so try to map it now.
 887		 */
 888		if (unlikely(lcn != LCN_HOLE && lcn != LCN_ENOENT)) {
 889			if (likely(!is_retry && lcn == LCN_RL_NOT_MAPPED)) {
 890				/* Attempt to map runlist. */
 891				if (!rl_write_locked) {
 892					/*
 893					 * We need the runlist locked for
 894					 * writing, so if it is locked for
 895					 * reading relock it now and retry in
 896					 * case it changed whilst we dropped
 897					 * the lock.
 898					 */
 899					up_read(&ni->runlist.lock);
 900					down_write(&ni->runlist.lock);
 901					rl_write_locked = true;
 902					goto retry_remap;
 903				}
 904				err = ntfs_map_runlist_nolock(ni, bh_cpos,
 905						NULL);
 906				if (likely(!err)) {
 907					is_retry = true;
 908					goto retry_remap;
 909				}
 910				/*
 911				 * If @vcn is out of bounds, pretend @lcn is
 912				 * LCN_ENOENT.  As long as the buffer is out
 913				 * of bounds this will work fine.
 914				 */
 915				if (err == -ENOENT) {
 916					lcn = LCN_ENOENT;
 917					err = 0;
 918					goto rl_not_mapped_enoent;
 919				}
 920			} else
 921				err = -EIO;
 922			/* Failed to map the buffer, even after retrying. */
 923			bh->b_blocknr = -1;
 924			ntfs_error(vol->sb, "Failed to write to inode 0x%lx, "
 925					"attribute type 0x%x, vcn 0x%llx, "
 926					"vcn offset 0x%x, because its "
 927					"location on disk could not be "
 928					"determined%s (error code %i).",
 929					ni->mft_no, ni->type,
 930					(unsigned long long)bh_cpos,
 931					(unsigned)bh_pos &
 932					vol->cluster_size_mask,
 933					is_retry ? " even after retrying" : "",
 934					err);
 935			break;
 936		}
 937rl_not_mapped_enoent:
 938		/*
 939		 * The buffer is in a hole or out of bounds.  We need to fill
 940		 * the hole, unless the buffer is in a cluster which is not
 941		 * touched by the write, in which case we just leave the buffer
 942		 * unmapped.  This can only happen when the cluster size is
 943		 * less than the page cache size.
 944		 */
 945		if (unlikely(vol->cluster_size < PAGE_SIZE)) {
 946			bh_cend = (bh_end + vol->cluster_size - 1) >>
 947					vol->cluster_size_bits;
 948			if ((bh_cend <= cpos || bh_cpos >= cend)) {
 949				bh->b_blocknr = -1;
 950				/*
 951				 * If the buffer is uptodate we skip it.  If it
 952				 * is not but the page is uptodate, we can set
 953				 * the buffer uptodate.  If the page is not
 954				 * uptodate, we can clear the buffer and set it
 955				 * uptodate.  Whether this is worthwhile is
 956				 * debatable and this could be removed.
 957				 */
 958				if (PageUptodate(page)) {
 959					if (!buffer_uptodate(bh))
 960						set_buffer_uptodate(bh);
 961				} else if (!buffer_uptodate(bh)) {
 962					zero_user(page, bh_offset(bh),
 963						blocksize);
 964					set_buffer_uptodate(bh);
 965				}
 966				continue;
 967			}
 968		}
 969		/*
 970		 * Out of bounds buffer is invalid if it was not really out of
 971		 * bounds.
 972		 */
 973		BUG_ON(lcn != LCN_HOLE);
 974		/*
 975		 * We need the runlist locked for writing, so if it is locked
 976		 * for reading relock it now and retry in case it changed
 977		 * whilst we dropped the lock.
 978		 */
 979		BUG_ON(!rl);
 980		if (!rl_write_locked) {
 981			up_read(&ni->runlist.lock);
 982			down_write(&ni->runlist.lock);
 983			rl_write_locked = true;
 984			goto retry_remap;
 985		}
 986		/* Find the previous last allocated cluster. */
 987		BUG_ON(rl->lcn != LCN_HOLE);
 988		lcn = -1;
 989		rl2 = rl;
 990		while (--rl2 >= ni->runlist.rl) {
 991			if (rl2->lcn >= 0) {
 992				lcn = rl2->lcn + rl2->length;
 993				break;
 994			}
 995		}
 996		rl2 = ntfs_cluster_alloc(vol, bh_cpos, 1, lcn, DATA_ZONE,
 997				false);
 998		if (IS_ERR(rl2)) {
 999			err = PTR_ERR(rl2);
1000			ntfs_debug("Failed to allocate cluster, error code %i.",
1001					err);
1002			break;
1003		}
1004		lcn = rl2->lcn;
1005		rl = ntfs_runlists_merge(ni->runlist.rl, rl2);
1006		if (IS_ERR(rl)) {
1007			err = PTR_ERR(rl);
1008			if (err != -ENOMEM)
1009				err = -EIO;
1010			if (ntfs_cluster_free_from_rl(vol, rl2)) {
1011				ntfs_error(vol->sb, "Failed to release "
1012						"allocated cluster in error "
1013						"code path.  Run chkdsk to "
1014						"recover the lost cluster.");
1015				NVolSetErrors(vol);
1016			}
1017			ntfs_free(rl2);
1018			break;
1019		}
1020		ni->runlist.rl = rl;
1021		status.runlist_merged = 1;
1022		ntfs_debug("Allocated cluster, lcn 0x%llx.",
1023				(unsigned long long)lcn);
1024		/* Map and lock the mft record and get the attribute record. */
1025		if (!NInoAttr(ni))
1026			base_ni = ni;
1027		else
1028			base_ni = ni->ext.base_ntfs_ino;
1029		m = map_mft_record(base_ni);
1030		if (IS_ERR(m)) {
1031			err = PTR_ERR(m);
1032			break;
1033		}
1034		ctx = ntfs_attr_get_search_ctx(base_ni, m);
1035		if (unlikely(!ctx)) {
1036			err = -ENOMEM;
1037			unmap_mft_record(base_ni);
1038			break;
1039		}
1040		status.mft_attr_mapped = 1;
1041		err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len,
1042				CASE_SENSITIVE, bh_cpos, NULL, 0, ctx);
1043		if (unlikely(err)) {
1044			if (err == -ENOENT)
1045				err = -EIO;
1046			break;
1047		}
1048		m = ctx->mrec;
1049		a = ctx->attr;
1050		/*
1051		 * Find the runlist element with which the attribute extent
1052		 * starts.  Note, we cannot use the _attr_ version because we
1053		 * have mapped the mft record.  That is ok because we know the
1054		 * runlist fragment must be mapped already to have ever gotten
1055		 * here, so we can just use the _rl_ version.
1056		 */
1057		vcn = sle64_to_cpu(a->data.non_resident.lowest_vcn);
1058		rl2 = ntfs_rl_find_vcn_nolock(rl, vcn);
1059		BUG_ON(!rl2);
1060		BUG_ON(!rl2->length);
1061		BUG_ON(rl2->lcn < LCN_HOLE);
1062		highest_vcn = sle64_to_cpu(a->data.non_resident.highest_vcn);
1063		/*
1064		 * If @highest_vcn is zero, calculate the real highest_vcn
1065		 * (which can really be zero).
1066		 */
1067		if (!highest_vcn)
1068			highest_vcn = (sle64_to_cpu(
1069					a->data.non_resident.allocated_size) >>
1070					vol->cluster_size_bits) - 1;
1071		/*
1072		 * Determine the size of the mapping pairs array for the new
1073		 * extent, i.e. the old extent with the hole filled.
1074		 */
1075		mp_size = ntfs_get_size_for_mapping_pairs(vol, rl2, vcn,
1076				highest_vcn);
1077		if (unlikely(mp_size <= 0)) {
1078			if (!(err = mp_size))
1079				err = -EIO;
1080			ntfs_debug("Failed to get size for mapping pairs "
1081					"array, error code %i.", err);
1082			break;
1083		}
1084		/*
1085		 * Resize the attribute record to fit the new mapping pairs
1086		 * array.
1087		 */
1088		attr_rec_len = le32_to_cpu(a->length);
1089		err = ntfs_attr_record_resize(m, a, mp_size + le16_to_cpu(
1090				a->data.non_resident.mapping_pairs_offset));
1091		if (unlikely(err)) {
1092			BUG_ON(err != -ENOSPC);
1093			// TODO: Deal with this by using the current attribute
1094			// and fill it with as much of the mapping pairs
1095			// array as possible.  Then loop over each attribute
1096			// extent rewriting the mapping pairs arrays as we go
1097			// along and if when we reach the end we have not
1098			// enough space, try to resize the last attribute
1099			// extent and if even that fails, add a new attribute
1100			// extent.
1101			// We could also try to resize at each step in the hope
1102			// that we will not need to rewrite every single extent.
1103			// Note, we may need to decompress some extents to fill
1104			// the runlist as we are walking the extents...
1105			ntfs_error(vol->sb, "Not enough space in the mft "
1106					"record for the extended attribute "
1107					"record.  This case is not "
1108					"implemented yet.");
1109			err = -EOPNOTSUPP;
1110			break ;
1111		}
1112		status.mp_rebuilt = 1;
1113		/*
1114		 * Generate the mapping pairs array directly into the attribute
1115		 * record.
1116		 */
1117		err = ntfs_mapping_pairs_build(vol, (u8*)a + le16_to_cpu(
1118				a->data.non_resident.mapping_pairs_offset),
1119				mp_size, rl2, vcn, highest_vcn, NULL);
1120		if (unlikely(err)) {
1121			ntfs_error(vol->sb, "Cannot fill hole in inode 0x%lx, "
1122					"attribute type 0x%x, because building "
1123					"the mapping pairs failed with error "
1124					"code %i.", vi->i_ino,
1125					(unsigned)le32_to_cpu(ni->type), err);
1126			err = -EIO;
1127			break;
1128		}
1129		/* Update the highest_vcn but only if it was not set. */
1130		if (unlikely(!a->data.non_resident.highest_vcn))
1131			a->data.non_resident.highest_vcn =
1132					cpu_to_sle64(highest_vcn);
1133		/*
1134		 * If the attribute is sparse/compressed, update the compressed
1135		 * size in the ntfs_inode structure and the attribute record.
1136		 */
1137		if (likely(NInoSparse(ni) || NInoCompressed(ni))) {
1138			/*
1139			 * If we are not in the first attribute extent, switch
1140			 * to it, but first ensure the changes will make it to
1141			 * disk later.
1142			 */
1143			if (a->data.non_resident.lowest_vcn) {
1144				flush_dcache_mft_record_page(ctx->ntfs_ino);
1145				mark_mft_record_dirty(ctx->ntfs_ino);
1146				ntfs_attr_reinit_search_ctx(ctx);
1147				err = ntfs_attr_lookup(ni->type, ni->name,
1148						ni->name_len, CASE_SENSITIVE,
1149						0, NULL, 0, ctx);
1150				if (unlikely(err)) {
1151					status.attr_switched = 1;
1152					break;
1153				}
1154				/* @m is not used any more so do not set it. */
1155				a = ctx->attr;
1156			}
1157			write_lock_irqsave(&ni->size_lock, flags);
1158			ni->itype.compressed.size += vol->cluster_size;
1159			a->data.non_resident.compressed_size =
1160					cpu_to_sle64(ni->itype.compressed.size);
1161			write_unlock_irqrestore(&ni->size_lock, flags);
1162		}
1163		/* Ensure the changes make it to disk. */
1164		flush_dcache_mft_record_page(ctx->ntfs_ino);
1165		mark_mft_record_dirty(ctx->ntfs_ino);
1166		ntfs_attr_put_search_ctx(ctx);
1167		unmap_mft_record(base_ni);
1168		/* Successfully filled the hole. */
1169		status.runlist_merged = 0;
1170		status.mft_attr_mapped = 0;
1171		status.mp_rebuilt = 0;
1172		/* Setup the map cache and use that to deal with the buffer. */
1173		was_hole = true;
1174		vcn = bh_cpos;
1175		vcn_len = 1;
1176		lcn_block = lcn << (vol->cluster_size_bits - blocksize_bits);
1177		cdelta = 0;
1178		/*
1179		 * If the number of remaining clusters in the @pages is smaller
1180		 * or equal to the number of cached clusters, unlock the
1181		 * runlist as the map cache will be used from now on.
1182		 */
1183		if (likely(vcn + vcn_len >= cend)) {
1184			up_write(&ni->runlist.lock);
1185			rl_write_locked = false;
1186			rl = NULL;
1187		}
1188		goto map_buffer_cached;
1189	} while (bh_pos += blocksize, (bh = bh->b_this_page) != head);
1190	/* If there are no errors, do the next page. */
1191	if (likely(!err && ++u < nr_pages))
1192		goto do_next_page;
1193	/* If there are no errors, release the runlist lock if we took it. */
1194	if (likely(!err)) {
1195		if (unlikely(rl_write_locked)) {
1196			up_write(&ni->runlist.lock);
1197			rl_write_locked = false;
1198		} else if (unlikely(rl))
1199			up_read(&ni->runlist.lock);
1200		rl = NULL;
1201	}
1202	/* If we issued read requests, let them complete. */
1203	read_lock_irqsave(&ni->size_lock, flags);
1204	initialized_size = ni->initialized_size;
1205	read_unlock_irqrestore(&ni->size_lock, flags);
1206	while (wait_bh > wait) {
1207		bh = *--wait_bh;
1208		wait_on_buffer(bh);
1209		if (likely(buffer_uptodate(bh))) {
1210			page = bh->b_page;
1211			bh_pos = ((s64)page->index << PAGE_SHIFT) +
1212					bh_offset(bh);
1213			/*
1214			 * If the buffer overflows the initialized size, need
1215			 * to zero the overflowing region.
1216			 */
1217			if (unlikely(bh_pos + blocksize > initialized_size)) {
1218				int ofs = 0;
1219
1220				if (likely(bh_pos < initialized_size))
1221					ofs = initialized_size - bh_pos;
1222				zero_user_segment(page, bh_offset(bh) + ofs,
1223						blocksize);
1224			}
1225		} else /* if (unlikely(!buffer_uptodate(bh))) */
1226			err = -EIO;
1227	}
1228	if (likely(!err)) {
1229		/* Clear buffer_new on all buffers. */
1230		u = 0;
1231		do {
1232			bh = head = page_buffers(pages[u]);
1233			do {
1234				if (buffer_new(bh))
1235					clear_buffer_new(bh);
1236			} while ((bh = bh->b_this_page) != head);
1237		} while (++u < nr_pages);
1238		ntfs_debug("Done.");
1239		return err;
1240	}
1241	if (status.attr_switched) {
1242		/* Get back to the attribute extent we modified. */
1243		ntfs_attr_reinit_search_ctx(ctx);
1244		if (ntfs_attr_lookup(ni->type, ni->name, ni->name_len,
1245				CASE_SENSITIVE, bh_cpos, NULL, 0, ctx)) {
1246			ntfs_error(vol->sb, "Failed to find required "
1247					"attribute extent of attribute in "
1248					"error code path.  Run chkdsk to "
1249					"recover.");
1250			write_lock_irqsave(&ni->size_lock, flags);
1251			ni->itype.compressed.size += vol->cluster_size;
1252			write_unlock_irqrestore(&ni->size_lock, flags);
1253			flush_dcache_mft_record_page(ctx->ntfs_ino);
1254			mark_mft_record_dirty(ctx->ntfs_ino);
1255			/*
1256			 * The only thing that is now wrong is the compressed
1257			 * size of the base attribute extent which chkdsk
1258			 * should be able to fix.
1259			 */
1260			NVolSetErrors(vol);
1261		} else {
1262			m = ctx->mrec;
1263			a = ctx->attr;
1264			status.attr_switched = 0;
1265		}
1266	}
1267	/*
1268	 * If the runlist has been modified, need to restore it by punching a
1269	 * hole into it and we then need to deallocate the on-disk cluster as
1270	 * well.  Note, we only modify the runlist if we are able to generate a
1271	 * new mapping pairs array, i.e. only when the mapped attribute extent
1272	 * is not switched.
1273	 */
1274	if (status.runlist_merged && !status.attr_switched) {
1275		BUG_ON(!rl_write_locked);
1276		/* Make the file cluster we allocated sparse in the runlist. */
1277		if (ntfs_rl_punch_nolock(vol, &ni->runlist, bh_cpos, 1)) {
1278			ntfs_error(vol->sb, "Failed to punch hole into "
1279					"attribute runlist in error code "
1280					"path.  Run chkdsk to recover the "
1281					"lost cluster.");
1282			NVolSetErrors(vol);
1283		} else /* if (success) */ {
1284			status.runlist_merged = 0;
1285			/*
1286			 * Deallocate the on-disk cluster we allocated but only
1287			 * if we succeeded in punching its vcn out of the
1288			 * runlist.
1289			 */
1290			down_write(&vol->lcnbmp_lock);
1291			if (ntfs_bitmap_clear_bit(vol->lcnbmp_ino, lcn)) {
1292				ntfs_error(vol->sb, "Failed to release "
1293						"allocated cluster in error "
1294						"code path.  Run chkdsk to "
1295						"recover the lost cluster.");
1296				NVolSetErrors(vol);
1297			}
1298			up_write(&vol->lcnbmp_lock);
1299		}
1300	}
1301	/*
1302	 * Resize the attribute record to its old size and rebuild the mapping
1303	 * pairs array.  Note, we only can do this if the runlist has been
1304	 * restored to its old state which also implies that the mapped
1305	 * attribute extent is not switched.
1306	 */
1307	if (status.mp_rebuilt && !status.runlist_merged) {
1308		if (ntfs_attr_record_resize(m, a, attr_rec_len)) {
1309			ntfs_error(vol->sb, "Failed to restore attribute "
1310					"record in error code path.  Run "
1311					"chkdsk to recover.");
1312			NVolSetErrors(vol);
1313		} else /* if (success) */ {
1314			if (ntfs_mapping_pairs_build(vol, (u8*)a +
1315					le16_to_cpu(a->data.non_resident.
1316					mapping_pairs_offset), attr_rec_len -
1317					le16_to_cpu(a->data.non_resident.
1318					mapping_pairs_offset), ni->runlist.rl,
1319					vcn, highest_vcn, NULL)) {
1320				ntfs_error(vol->sb, "Failed to restore "
1321						"mapping pairs array in error "
1322						"code path.  Run chkdsk to "
1323						"recover.");
1324				NVolSetErrors(vol);
1325			}
1326			flush_dcache_mft_record_page(ctx->ntfs_ino);
1327			mark_mft_record_dirty(ctx->ntfs_ino);
1328		}
1329	}
1330	/* Release the mft record and the attribute. */
1331	if (status.mft_attr_mapped) {
1332		ntfs_attr_put_search_ctx(ctx);
1333		unmap_mft_record(base_ni);
1334	}
1335	/* Release the runlist lock. */
1336	if (rl_write_locked)
1337		up_write(&ni->runlist.lock);
1338	else if (rl)
1339		up_read(&ni->runlist.lock);
1340	/*
1341	 * Zero out any newly allocated blocks to avoid exposing stale data.
1342	 * If BH_New is set, we know that the block was newly allocated above
1343	 * and that it has not been fully zeroed and marked dirty yet.
1344	 */
1345	nr_pages = u;
1346	u = 0;
1347	end = bh_cpos << vol->cluster_size_bits;
1348	do {
1349		page = pages[u];
1350		bh = head = page_buffers(page);
1351		do {
1352			if (u == nr_pages &&
1353					((s64)page->index << PAGE_SHIFT) +
1354					bh_offset(bh) >= end)
1355				break;
1356			if (!buffer_new(bh))
1357				continue;
1358			clear_buffer_new(bh);
1359			if (!buffer_uptodate(bh)) {
1360				if (PageUptodate(page))
1361					set_buffer_uptodate(bh);
1362				else {
1363					zero_user(page, bh_offset(bh),
1364							blocksize);
1365					set_buffer_uptodate(bh);
1366				}
1367			}
1368			mark_buffer_dirty(bh);
1369		} while ((bh = bh->b_this_page) != head);
1370	} while (++u <= nr_pages);
1371	ntfs_error(vol->sb, "Failed.  Returning error code %i.", err);
1372	return err;
1373}
1374
1375static inline void ntfs_flush_dcache_pages(struct page **pages,
1376		unsigned nr_pages)
1377{
1378	BUG_ON(!nr_pages);
1379	/*
1380	 * Warning: Do not do the decrement at the same time as the call to
1381	 * flush_dcache_page() because it is a NULL macro on i386 and hence the
1382	 * decrement never happens so the loop never terminates.
1383	 */
1384	do {
1385		--nr_pages;
1386		flush_dcache_page(pages[nr_pages]);
1387	} while (nr_pages > 0);
1388}
1389
1390/**
1391 * ntfs_commit_pages_after_non_resident_write - commit the received data
1392 * @pages:	array of destination pages
1393 * @nr_pages:	number of pages in @pages
1394 * @pos:	byte position in file at which the write begins
1395 * @bytes:	number of bytes to be written
1396 *
1397 * See description of ntfs_commit_pages_after_write(), below.
1398 */
1399static inline int ntfs_commit_pages_after_non_resident_write(
1400		struct page **pages, const unsigned nr_pages,
1401		s64 pos, size_t bytes)
1402{
1403	s64 end, initialized_size;
1404	struct inode *vi;
1405	ntfs_inode *ni, *base_ni;
1406	struct buffer_head *bh, *head;
1407	ntfs_attr_search_ctx *ctx;
1408	MFT_RECORD *m;
1409	ATTR_RECORD *a;
1410	unsigned long flags;
1411	unsigned blocksize, u;
1412	int err;
1413
1414	vi = pages[0]->mapping->host;
1415	ni = NTFS_I(vi);
1416	blocksize = vi->i_sb->s_blocksize;
1417	end = pos + bytes;
1418	u = 0;
1419	do {
1420		s64 bh_pos;
1421		struct page *page;
1422		bool partial;
1423
1424		page = pages[u];
1425		bh_pos = (s64)page->index << PAGE_SHIFT;
1426		bh = head = page_buffers(page);
1427		partial = false;
1428		do {
1429			s64 bh_end;
1430
1431			bh_end = bh_pos + blocksize;
1432			if (bh_end <= pos || bh_pos >= end) {
1433				if (!buffer_uptodate(bh))
1434					partial = true;
1435			} else {
1436				set_buffer_uptodate(bh);
1437				mark_buffer_dirty(bh);
1438			}
1439		} while (bh_pos += blocksize, (bh = bh->b_this_page) != head);
1440		/*
1441		 * If all buffers are now uptodate but the page is not, set the
1442		 * page uptodate.
1443		 */
1444		if (!partial && !PageUptodate(page))
1445			SetPageUptodate(page);
1446	} while (++u < nr_pages);
1447	/*
1448	 * Finally, if we do not need to update initialized_size or i_size we
1449	 * are finished.
1450	 */
1451	read_lock_irqsave(&ni->size_lock, flags);
1452	initialized_size = ni->initialized_size;
1453	read_unlock_irqrestore(&ni->size_lock, flags);
1454	if (end <= initialized_size) {
1455		ntfs_debug("Done.");
1456		return 0;
1457	}
1458	/*
1459	 * Update initialized_size/i_size as appropriate, both in the inode and
1460	 * the mft record.
1461	 */
1462	if (!NInoAttr(ni))
1463		base_ni = ni;
1464	else
1465		base_ni = ni->ext.base_ntfs_ino;
1466	/* Map, pin, and lock the mft record. */
1467	m = map_mft_record(base_ni);
1468	if (IS_ERR(m)) {
1469		err = PTR_ERR(m);
1470		m = NULL;
1471		ctx = NULL;
1472		goto err_out;
1473	}
1474	BUG_ON(!NInoNonResident(ni));
1475	ctx = ntfs_attr_get_search_ctx(base_ni, m);
1476	if (unlikely(!ctx)) {
1477		err = -ENOMEM;
1478		goto err_out;
1479	}
1480	err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len,
1481			CASE_SENSITIVE, 0, NULL, 0, ctx);
1482	if (unlikely(err)) {
1483		if (err == -ENOENT)
1484			err = -EIO;
1485		goto err_out;
1486	}
1487	a = ctx->attr;
1488	BUG_ON(!a->non_resident);
1489	write_lock_irqsave(&ni->size_lock, flags);
1490	BUG_ON(end > ni->allocated_size);
1491	ni->initialized_size = end;
1492	a->data.non_resident.initialized_size = cpu_to_sle64(end);
1493	if (end > i_size_read(vi)) {
1494		i_size_write(vi, end);
1495		a->data.non_resident.data_size =
1496				a->data.non_resident.initialized_size;
1497	}
1498	write_unlock_irqrestore(&ni->size_lock, flags);
1499	/* Mark the mft record dirty, so it gets written back. */
1500	flush_dcache_mft_record_page(ctx->ntfs_ino);
1501	mark_mft_record_dirty(ctx->ntfs_ino);
1502	ntfs_attr_put_search_ctx(ctx);
1503	unmap_mft_record(base_ni);
1504	ntfs_debug("Done.");
1505	return 0;
1506err_out:
1507	if (ctx)
1508		ntfs_attr_put_search_ctx(ctx);
1509	if (m)
1510		unmap_mft_record(base_ni);
1511	ntfs_error(vi->i_sb, "Failed to update initialized_size/i_size (error "
1512			"code %i).", err);
1513	if (err != -ENOMEM)
1514		NVolSetErrors(ni->vol);
1515	return err;
1516}
1517
1518/**
1519 * ntfs_commit_pages_after_write - commit the received data
1520 * @pages:	array of destination pages
1521 * @nr_pages:	number of pages in @pages
1522 * @pos:	byte position in file at which the write begins
1523 * @bytes:	number of bytes to be written
1524 *
1525 * This is called from ntfs_file_buffered_write() with i_mutex held on the inode
1526 * (@pages[0]->mapping->host).  There are @nr_pages pages in @pages which are
1527 * locked but not kmap()ped.  The source data has already been copied into the
1528 * @page.  ntfs_prepare_pages_for_non_resident_write() has been called before
1529 * the data was copied (for non-resident attributes only) and it returned
1530 * success.
1531 *
1532 * Need to set uptodate and mark dirty all buffers within the boundary of the
1533 * write.  If all buffers in a page are uptodate we set the page uptodate, too.
1534 *
1535 * Setting the buffers dirty ensures that they get written out later when
1536 * ntfs_writepage() is invoked by the VM.
1537 *
1538 * Finally, we need to update i_size and initialized_size as appropriate both
1539 * in the inode and the mft record.
1540 *
1541 * This is modelled after fs/buffer.c::generic_commit_write(), which marks
1542 * buffers uptodate and dirty, sets the page uptodate if all buffers in the
1543 * page are uptodate, and updates i_size if the end of io is beyond i_size.  In
1544 * that case, it also marks the inode dirty.
1545 *
1546 * If things have gone as outlined in
1547 * ntfs_prepare_pages_for_non_resident_write(), we do not need to do any page
1548 * content modifications here for non-resident attributes.  For resident
1549 * attributes we need to do the uptodate bringing here which we combine with
1550 * the copying into the mft record which means we save one atomic kmap.
1551 *
1552 * Return 0 on success or -errno on error.
1553 */
1554static int ntfs_commit_pages_after_write(struct page **pages,
1555		const unsigned nr_pages, s64 pos, size_t bytes)
1556{
1557	s64 end, initialized_size;
1558	loff_t i_size;
1559	struct inode *vi;
1560	ntfs_inode *ni, *base_ni;
1561	struct page *page;
1562	ntfs_attr_search_ctx *ctx;
1563	MFT_RECORD *m;
1564	ATTR_RECORD *a;
1565	char *kattr, *kaddr;
1566	unsigned long flags;
1567	u32 attr_len;
1568	int err;
1569
1570	BUG_ON(!nr_pages);
1571	BUG_ON(!pages);
1572	page = pages[0];
1573	BUG_ON(!page);
1574	vi = page->mapping->host;
1575	ni = NTFS_I(vi);
1576	ntfs_debug("Entering for inode 0x%lx, attribute type 0x%x, start page "
1577			"index 0x%lx, nr_pages 0x%x, pos 0x%llx, bytes 0x%zx.",
1578			vi->i_ino, ni->type, page->index, nr_pages,
1579			(long long)pos, bytes);
1580	if (NInoNonResident(ni))
1581		return ntfs_commit_pages_after_non_resident_write(pages,
1582				nr_pages, pos, bytes);
1583	BUG_ON(nr_pages > 1);
1584	/*
1585	 * Attribute is resident, implying it is not compressed, encrypted, or
1586	 * sparse.
1587	 */
1588	if (!NInoAttr(ni))
1589		base_ni = ni;
1590	else
1591		base_ni = ni->ext.base_ntfs_ino;
1592	BUG_ON(NInoNonResident(ni));
1593	/* Map, pin, and lock the mft record. */
1594	m = map_mft_record(base_ni);
1595	if (IS_ERR(m)) {
1596		err = PTR_ERR(m);
1597		m = NULL;
1598		ctx = NULL;
1599		goto err_out;
1600	}
1601	ctx = ntfs_attr_get_search_ctx(base_ni, m);
1602	if (unlikely(!ctx)) {
1603		err = -ENOMEM;
1604		goto err_out;
1605	}
1606	err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len,
1607			CASE_SENSITIVE, 0, NULL, 0, ctx);
1608	if (unlikely(err)) {
1609		if (err == -ENOENT)
1610			err = -EIO;
1611		goto err_out;
1612	}
1613	a = ctx->attr;
1614	BUG_ON(a->non_resident);
1615	/* The total length of the attribute value. */
1616	attr_len = le32_to_cpu(a->data.resident.value_length);
1617	i_size = i_size_read(vi);
1618	BUG_ON(attr_len != i_size);
1619	BUG_ON(pos > attr_len);
1620	end = pos + bytes;
1621	BUG_ON(end > le32_to_cpu(a->length) -
1622			le16_to_cpu(a->data.resident.value_offset));
1623	kattr = (u8*)a + le16_to_cpu(a->data.resident.value_offset);
1624	kaddr = kmap_atomic(page);
1625	/* Copy the received data from the page to the mft record. */
1626	memcpy(kattr + pos, kaddr + pos, bytes);
1627	/* Update the attribute length if necessary. */
1628	if (end > attr_len) {
1629		attr_len = end;
1630		a->data.resident.value_length = cpu_to_le32(attr_len);
1631	}
1632	/*
1633	 * If the page is not uptodate, bring the out of bounds area(s)
1634	 * uptodate by copying data from the mft record to the page.
1635	 */
1636	if (!PageUptodate(page)) {
1637		if (pos > 0)
1638			memcpy(kaddr, kattr, pos);
1639		if (end < attr_len)
1640			memcpy(kaddr + end, kattr + end, attr_len - end);
1641		/* Zero the region outside the end of the attribute value. */
1642		memset(kaddr + attr_len, 0, PAGE_SIZE - attr_len);
1643		flush_dcache_page(page);
1644		SetPageUptodate(page);
1645	}
1646	kunmap_atomic(kaddr);
1647	/* Update initialized_size/i_size if necessary. */
1648	read_lock_irqsave(&ni->size_lock, flags);
1649	initialized_size = ni->initialized_size;
1650	BUG_ON(end > ni->allocated_size);
1651	read_unlock_irqrestore(&ni->size_lock, flags);
1652	BUG_ON(initialized_size != i_size);
1653	if (end > initialized_size) {
1654		write_lock_irqsave(&ni->size_lock, flags);
1655		ni->initialized_size = end;
1656		i_size_write(vi, end);
1657		write_unlock_irqrestore(&ni->size_lock, flags);
1658	}
1659	/* Mark the mft record dirty, so it gets written back. */
1660	flush_dcache_mft_record_page(ctx->ntfs_ino);
1661	mark_mft_record_dirty(ctx->ntfs_ino);
1662	ntfs_attr_put_search_ctx(ctx);
1663	unmap_mft_record(base_ni);
1664	ntfs_debug("Done.");
1665	return 0;
1666err_out:
1667	if (err == -ENOMEM) {
1668		ntfs_warning(vi->i_sb, "Error allocating memory required to "
1669				"commit the write.");
1670		if (PageUptodate(page)) {
1671			ntfs_warning(vi->i_sb, "Page is uptodate, setting "
1672					"dirty so the write will be retried "
1673					"later on by the VM.");
1674			/*
1675			 * Put the page on mapping->dirty_pages, but leave its
1676			 * buffers' dirty state as-is.
1677			 */
1678			__set_page_dirty_nobuffers(page);
1679			err = 0;
1680		} else
1681			ntfs_error(vi->i_sb, "Page is not uptodate.  Written "
1682					"data has been lost.");
1683	} else {
1684		ntfs_error(vi->i_sb, "Resident attribute commit write failed "
1685				"with error %i.", err);
1686		NVolSetErrors(ni->vol);
1687	}
1688	if (ctx)
1689		ntfs_attr_put_search_ctx(ctx);
1690	if (m)
1691		unmap_mft_record(base_ni);
1692	return err;
1693}
1694
1695/*
1696 * Copy as much as we can into the pages and return the number of bytes which
1697 * were successfully copied.  If a fault is encountered then clear the pages
1698 * out to (ofs + bytes) and return the number of bytes which were copied.
1699 */
1700static size_t ntfs_copy_from_user_iter(struct page **pages, unsigned nr_pages,
1701		unsigned ofs, struct iov_iter *i, size_t bytes)
1702{
1703	struct page **last_page = pages + nr_pages;
1704	size_t total = 0;
1705	struct iov_iter data = *i;
1706	unsigned len, copied;
1707
1708	do {
1709		len = PAGE_SIZE - ofs;
1710		if (len > bytes)
1711			len = bytes;
1712		copied = iov_iter_copy_from_user_atomic(*pages, &data, ofs,
1713				len);
1714		total += copied;
1715		bytes -= copied;
1716		if (!bytes)
1717			break;
1718		iov_iter_advance(&data, copied);
1719		if (copied < len)
1720			goto err;
1721		ofs = 0;
1722	} while (++pages < last_page);
1723out:
1724	return total;
1725err:
1726	/* Zero the rest of the target like __copy_from_user(). */
1727	len = PAGE_SIZE - copied;
1728	do {
1729		if (len > bytes)
1730			len = bytes;
1731		zero_user(*pages, copied, len);
1732		bytes -= len;
1733		copied = 0;
1734		len = PAGE_SIZE;
1735	} while (++pages < last_page);
1736	goto out;
1737}
1738
1739/**
1740 * ntfs_perform_write - perform buffered write to a file
1741 * @file:	file to write to
1742 * @i:		iov_iter with data to write
1743 * @pos:	byte offset in file at which to begin writing to
1744 */
1745static ssize_t ntfs_perform_write(struct file *file, struct iov_iter *i,
1746		loff_t pos)
1747{
1748	struct address_space *mapping = file->f_mapping;
1749	struct inode *vi = mapping->host;
1750	ntfs_inode *ni = NTFS_I(vi);
1751	ntfs_volume *vol = ni->vol;
1752	struct page *pages[NTFS_MAX_PAGES_PER_CLUSTER];
1753	struct page *cached_page = NULL;
1754	VCN last_vcn;
1755	LCN lcn;
1756	size_t bytes;
1757	ssize_t status, written = 0;
1758	unsigned nr_pages;
1759
1760	ntfs_debug("Entering for i_ino 0x%lx, attribute type 0x%x, pos "
1761			"0x%llx, count 0x%lx.", vi->i_ino,
1762			(unsigned)le32_to_cpu(ni->type),
1763			(unsigned long long)pos,
1764			(unsigned long)iov_iter_count(i));
1765	/*
1766	 * If a previous ntfs_truncate() failed, repeat it and abort if it
1767	 * fails again.
1768	 */
1769	if (unlikely(NInoTruncateFailed(ni))) {
1770		int err;
1771
1772		inode_dio_wait(vi);
1773		err = ntfs_truncate(vi);
1774		if (err || NInoTruncateFailed(ni)) {
1775			if (!err)
1776				err = -EIO;
1777			ntfs_error(vol->sb, "Cannot perform write to inode "
1778					"0x%lx, attribute type 0x%x, because "
1779					"ntfs_truncate() failed (error code "
1780					"%i).", vi->i_ino,
1781					(unsigned)le32_to_cpu(ni->type), err);
1782			return err;
1783		}
1784	}
1785	/*
1786	 * Determine the number of pages per cluster for non-resident
1787	 * attributes.
1788	 */
1789	nr_pages = 1;
1790	if (vol->cluster_size > PAGE_SIZE && NInoNonResident(ni))
1791		nr_pages = vol->cluster_size >> PAGE_SHIFT;
1792	last_vcn = -1;
1793	do {
1794		VCN vcn;
1795		pgoff_t idx, start_idx;
1796		unsigned ofs, do_pages, u;
1797		size_t copied;
1798
1799		start_idx = idx = pos >> PAGE_SHIFT;
1800		ofs = pos & ~PAGE_MASK;
1801		bytes = PAGE_SIZE - ofs;
1802		do_pages = 1;
1803		if (nr_pages > 1) {
1804			vcn = pos >> vol->cluster_size_bits;
1805			if (vcn != last_vcn) {
1806				last_vcn = vcn;
1807				/*
1808				 * Get the lcn of the vcn the write is in.  If
1809				 * it is a hole, need to lock down all pages in
1810				 * the cluster.
1811				 */
1812				down_read(&ni->runlist.lock);
1813				lcn = ntfs_attr_vcn_to_lcn_nolock(ni, pos >>
1814						vol->cluster_size_bits, false);
1815				up_read(&ni->runlist.lock);
1816				if (unlikely(lcn < LCN_HOLE)) {
1817					if (lcn == LCN_ENOMEM)
1818						status = -ENOMEM;
1819					else {
1820						status = -EIO;
1821						ntfs_error(vol->sb, "Cannot "
1822							"perform write to "
1823							"inode 0x%lx, "
1824							"attribute type 0x%x, "
1825							"because the attribute "
1826							"is corrupt.",
1827							vi->i_ino, (unsigned)
1828							le32_to_cpu(ni->type));
1829					}
1830					break;
1831				}
1832				if (lcn == LCN_HOLE) {
1833					start_idx = (pos & ~(s64)
1834							vol->cluster_size_mask)
1835							>> PAGE_SHIFT;
1836					bytes = vol->cluster_size - (pos &
1837							vol->cluster_size_mask);
1838					do_pages = nr_pages;
1839				}
1840			}
1841		}
1842		if (bytes > iov_iter_count(i))
1843			bytes = iov_iter_count(i);
1844again:
1845		/*
1846		 * Bring in the user page(s) that we will copy from _first_.
1847		 * Otherwise there is a nasty deadlock on copying from the same
1848		 * page(s) as we are writing to, without it/them being marked
1849		 * up-to-date.  Note, at present there is nothing to stop the
1850		 * pages being swapped out between us bringing them into memory
1851		 * and doing the actual copying.
1852		 */
1853		if (unlikely(iov_iter_fault_in_multipages_readable(i, bytes))) {
1854			status = -EFAULT;
1855			break;
1856		}
1857		/* Get and lock @do_pages starting at index @start_idx. */
1858		status = __ntfs_grab_cache_pages(mapping, start_idx, do_pages,
1859				pages, &cached_page);
1860		if (unlikely(status))
1861			break;
1862		/*
1863		 * For non-resident attributes, we need to fill any holes with
1864		 * actual clusters and ensure all bufferes are mapped.  We also
1865		 * need to bring uptodate any buffers that are only partially
1866		 * being written to.
1867		 */
1868		if (NInoNonResident(ni)) {
1869			status = ntfs_prepare_pages_for_non_resident_write(
1870					pages, do_pages, pos, bytes);
1871			if (unlikely(status)) {
1872				do {
1873					unlock_page(pages[--do_pages]);
1874					put_page(pages[do_pages]);
1875				} while (do_pages);
1876				break;
1877			}
1878		}
1879		u = (pos >> PAGE_SHIFT) - pages[0]->index;
1880		copied = ntfs_copy_from_user_iter(pages + u, do_pages - u, ofs,
1881					i, bytes);
1882		ntfs_flush_dcache_pages(pages + u, do_pages - u);
1883		status = 0;
1884		if (likely(copied == bytes)) {
1885			status = ntfs_commit_pages_after_write(pages, do_pages,
1886					pos, bytes);
1887			if (!status)
1888				status = bytes;
1889		}
1890		do {
1891			unlock_page(pages[--do_pages]);
1892			put_page(pages[do_pages]);
1893		} while (do_pages);
1894		if (unlikely(status < 0))
 
1895			break;
1896		copied = status;
1897		cond_resched();
1898		if (unlikely(!copied)) {
1899			size_t sc;
1900
1901			/*
1902			 * We failed to copy anything.  Fall back to single
1903			 * segment length write.
1904			 *
1905			 * This is needed to avoid possible livelock in the
1906			 * case that all segments in the iov cannot be copied
1907			 * at once without a pagefault.
1908			 */
1909			sc = iov_iter_single_seg_count(i);
1910			if (bytes > sc)
1911				bytes = sc;
1912			goto again;
1913		}
1914		iov_iter_advance(i, copied);
1915		pos += copied;
1916		written += copied;
1917		balance_dirty_pages_ratelimited(mapping);
1918		if (fatal_signal_pending(current)) {
1919			status = -EINTR;
1920			break;
1921		}
1922	} while (iov_iter_count(i));
1923	if (cached_page)
1924		put_page(cached_page);
1925	ntfs_debug("Done.  Returning %s (written 0x%lx, status %li).",
1926			written ? "written" : "status", (unsigned long)written,
1927			(long)status);
1928	return written ? written : status;
1929}
1930
1931/**
1932 * ntfs_file_write_iter - simple wrapper for ntfs_file_write_iter_nolock()
1933 * @iocb:	IO state structure
1934 * @from:	iov_iter with data to write
1935 *
1936 * Basically the same as generic_file_write_iter() except that it ends up
1937 * up calling ntfs_perform_write() instead of generic_perform_write() and that
1938 * O_DIRECT is not implemented.
1939 */
1940static ssize_t ntfs_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
1941{
1942	struct file *file = iocb->ki_filp;
1943	struct inode *vi = file_inode(file);
1944	ssize_t written = 0;
1945	ssize_t err;
1946
1947	inode_lock(vi);
1948	/* We can write back this queue in page reclaim. */
1949	current->backing_dev_info = inode_to_bdi(vi);
1950	err = ntfs_prepare_file_for_write(iocb, from);
1951	if (iov_iter_count(from) && !err)
1952		written = ntfs_perform_write(file, from, iocb->ki_pos);
1953	current->backing_dev_info = NULL;
1954	inode_unlock(vi);
1955	if (likely(written > 0)) {
1956		err = generic_write_sync(file, iocb->ki_pos, written);
1957		if (err < 0)
1958			written = 0;
1959	}
1960	iocb->ki_pos += written;
 
 
1961	return written ? written : err;
1962}
1963
1964/**
1965 * ntfs_file_fsync - sync a file to disk
1966 * @filp:	file to be synced
1967 * @datasync:	if non-zero only flush user data and not metadata
1968 *
1969 * Data integrity sync of a file to disk.  Used for fsync, fdatasync, and msync
1970 * system calls.  This function is inspired by fs/buffer.c::file_fsync().
1971 *
1972 * If @datasync is false, write the mft record and all associated extent mft
1973 * records as well as the $DATA attribute and then sync the block device.
1974 *
1975 * If @datasync is true and the attribute is non-resident, we skip the writing
1976 * of the mft record and all associated extent mft records (this might still
1977 * happen due to the write_inode_now() call).
1978 *
1979 * Also, if @datasync is true, we do not wait on the inode to be written out
1980 * but we always wait on the page cache pages to be written out.
1981 *
1982 * Locking: Caller must hold i_mutex on the inode.
1983 *
1984 * TODO: We should probably also write all attribute/index inodes associated
1985 * with this inode but since we have no simple way of getting to them we ignore
1986 * this problem for now.
1987 */
1988static int ntfs_file_fsync(struct file *filp, loff_t start, loff_t end,
1989			   int datasync)
1990{
1991	struct inode *vi = filp->f_mapping->host;
1992	int err, ret = 0;
1993
1994	ntfs_debug("Entering for inode 0x%lx.", vi->i_ino);
1995
1996	err = filemap_write_and_wait_range(vi->i_mapping, start, end);
1997	if (err)
1998		return err;
1999	inode_lock(vi);
2000
2001	BUG_ON(S_ISDIR(vi->i_mode));
2002	if (!datasync || !NInoNonResident(NTFS_I(vi)))
2003		ret = __ntfs_write_inode(vi, 1);
2004	write_inode_now(vi, !datasync);
2005	/*
2006	 * NOTE: If we were to use mapping->private_list (see ext2 and
2007	 * fs/buffer.c) for dirty blocks then we could optimize the below to be
2008	 * sync_mapping_buffers(vi->i_mapping).
2009	 */
2010	err = sync_blockdev(vi->i_sb->s_bdev);
2011	if (unlikely(err && !ret))
2012		ret = err;
2013	if (likely(!ret))
2014		ntfs_debug("Done.");
2015	else
2016		ntfs_warning(vi->i_sb, "Failed to f%ssync inode 0x%lx.  Error "
2017				"%u.", datasync ? "data" : "", vi->i_ino, -ret);
2018	inode_unlock(vi);
2019	return ret;
2020}
2021
2022#endif /* NTFS_RW */
2023
2024const struct file_operations ntfs_file_ops = {
2025	.llseek		= generic_file_llseek,
2026	.read_iter	= generic_file_read_iter,
2027#ifdef NTFS_RW
2028	.write_iter	= ntfs_file_write_iter,
2029	.fsync		= ntfs_file_fsync,
2030#endif /* NTFS_RW */
2031	.mmap		= generic_file_mmap,
2032	.open		= ntfs_file_open,
2033	.splice_read	= generic_file_splice_read,
2034};
2035
2036const struct inode_operations ntfs_file_inode_ops = {
2037#ifdef NTFS_RW
2038	.setattr	= ntfs_setattr,
2039#endif /* NTFS_RW */
2040};
2041
2042const struct file_operations ntfs_empty_file_ops = {};
2043
2044const struct inode_operations ntfs_empty_inode_ops = {};