file.c 68.4 KB
Newer Older
Linus Torvalds's avatar
Linus Torvalds committed
1
/*
2
 * file.c - NTFS kernel file operations.  Part of the Linux-NTFS project.
Linus Torvalds's avatar
Linus Torvalds committed
3
 *
4
 * Copyright (c) 2001-2007 Anton Altaparmakov
Linus Torvalds's avatar
Linus Torvalds committed
5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22
 *
 * This program/include file is free software; you can redistribute it and/or
 * modify it under the terms of the GNU General Public License as published
 * by the Free Software Foundation; either version 2 of the License, or
 * (at your option) any later version.
 *
 * This program/include file is distributed in the hope that it will be
 * useful, but WITHOUT ANY WARRANTY; without even the implied warranty
 * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with this program (in the main directory of the Linux-NTFS
 * distribution in the file COPYING); if not, write to the Free Software
 * Foundation,Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
 */

#include <linux/buffer_head.h>
23 24 25 26 27 28
#include <linux/pagemap.h>
#include <linux/pagevec.h>
#include <linux/sched.h>
#include <linux/swap.h>
#include <linux/uio.h>
#include <linux/writeback.h>
Linus Torvalds's avatar
Linus Torvalds committed
29

30 31 32 33 34
#include <asm/page.h>
#include <asm/uaccess.h>

#include "attrib.h"
#include "bitmap.h"
Linus Torvalds's avatar
Linus Torvalds committed
35 36
#include "inode.h"
#include "debug.h"
37 38 39
#include "lcnalloc.h"
#include "malloc.h"
#include "mft.h"
Linus Torvalds's avatar
Linus Torvalds committed
40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62
#include "ntfs.h"

/**
 * ntfs_file_open - called when an inode is about to be opened
 * @vi:		inode to be opened
 * @filp:	file structure describing the inode
 *
 * Limit file size to the page cache limit on architectures where unsigned long
 * is 32-bits. This is the most we can do for now without overflowing the page
 * cache page index. Doing it this way means we don't run into problems because
 * of existing too large files. It would be better to allow the user to read
 * the beginning of the file but I doubt very much anyone is going to hit this
 * check on a 32-bit architecture, so there is no point in adding the extra
 * complexity required to support this.
 *
 * On 64-bit architectures, the check is hopefully optimized away by the
 * compiler.
 *
 * After the check passes, just call generic_file_open() to do its work.
 */
static int ntfs_file_open(struct inode *vi, struct file *filp)
{
	if (sizeof(unsigned long) < 8) {
63
		if (i_size_read(vi) > MAX_LFS_FILESIZE)
64
			return -EOVERFLOW;
Linus Torvalds's avatar
Linus Torvalds committed
65 66 67 68 69 70
	}
	return generic_file_open(vi, filp);
}

#ifdef NTFS_RW

71 72 73 74 75 76 77 78 79 80
/**
 * ntfs_attr_extend_initialized - extend the initialized size of an attribute
 * @ni:			ntfs inode of the attribute to extend
 * @new_init_size:	requested new initialized size in bytes
 * @cached_page:	store any allocated but unused page here
 * @lru_pvec:		lru-buffering pagevec of the caller
 *
 * Extend the initialized size of an attribute described by the ntfs inode @ni
 * to @new_init_size bytes.  This involves zeroing any non-sparse space between
 * the old initialized size and @new_init_size both in the page cache and on
81 82
 * disk (if relevant complete pages are already uptodate in the page cache then
 * these are simply marked dirty).
83 84 85 86 87 88 89 90 91 92 93 94 95 96
 *
 * As a side-effect, the file size (vfs inode->i_size) may be incremented as,
 * in the resident attribute case, it is tied to the initialized size and, in
 * the non-resident attribute case, it may not fall below the initialized size.
 *
 * Note that if the attribute is resident, we do not need to touch the page
 * cache at all.  This is because if the page cache page is not uptodate we
 * bring it uptodate later, when doing the write to the mft record since we
 * then already have the page mapped.  And if the page is uptodate, the
 * non-initialized region will already have been zeroed when the page was
 * brought uptodate and the region may in fact already have been overwritten
 * with new data via mmap() based writes, so we cannot just zero it.  And since
 * POSIX specifies that the behaviour of resizing a file whilst it is mmap()ped
 * is unspecified, we choose not to do zeroing and thus we do not need to touch
97 98
 * the page at all.  For a more detailed explanation see ntfs_truncate() in
 * fs/ntfs/inode.c.
99
 *
100
 * @cached_page and @lru_pvec are just optimizations for dealing with multiple
101 102 103 104 105 106 107 108
 * pages.
 *
 * Return 0 on success and -errno on error.  In the case that an error is
 * encountered it is possible that the initialized size will already have been
 * incremented some way towards @new_init_size but it is guaranteed that if
 * this is the case, the necessary zeroing will also have happened and that all
 * metadata is self-consistent.
 *
109
 * Locking: i_mutex on the vfs inode corrseponsind to the ntfs inode @ni must be
110
 *	    held by the caller.
111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233
 */
static int ntfs_attr_extend_initialized(ntfs_inode *ni, const s64 new_init_size,
		struct page **cached_page, struct pagevec *lru_pvec)
{
	s64 old_init_size;
	loff_t old_i_size;
	pgoff_t index, end_index;
	unsigned long flags;
	struct inode *vi = VFS_I(ni);
	ntfs_inode *base_ni;
	MFT_RECORD *m = NULL;
	ATTR_RECORD *a;
	ntfs_attr_search_ctx *ctx = NULL;
	struct address_space *mapping;
	struct page *page = NULL;
	u8 *kattr;
	int err;
	u32 attr_len;

	read_lock_irqsave(&ni->size_lock, flags);
	old_init_size = ni->initialized_size;
	old_i_size = i_size_read(vi);
	BUG_ON(new_init_size > ni->allocated_size);
	read_unlock_irqrestore(&ni->size_lock, flags);
	ntfs_debug("Entering for i_ino 0x%lx, attribute type 0x%x, "
			"old_initialized_size 0x%llx, "
			"new_initialized_size 0x%llx, i_size 0x%llx.",
			vi->i_ino, (unsigned)le32_to_cpu(ni->type),
			(unsigned long long)old_init_size,
			(unsigned long long)new_init_size, old_i_size);
	if (!NInoAttr(ni))
		base_ni = ni;
	else
		base_ni = ni->ext.base_ntfs_ino;
	/* Use goto to reduce indentation and we need the label below anyway. */
	if (NInoNonResident(ni))
		goto do_non_resident_extend;
	BUG_ON(old_init_size != old_i_size);
	m = map_mft_record(base_ni);
	if (IS_ERR(m)) {
		err = PTR_ERR(m);
		m = NULL;
		goto err_out;
	}
	ctx = ntfs_attr_get_search_ctx(base_ni, m);
	if (unlikely(!ctx)) {
		err = -ENOMEM;
		goto err_out;
	}
	err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len,
			CASE_SENSITIVE, 0, NULL, 0, ctx);
	if (unlikely(err)) {
		if (err == -ENOENT)
			err = -EIO;
		goto err_out;
	}
	m = ctx->mrec;
	a = ctx->attr;
	BUG_ON(a->non_resident);
	/* The total length of the attribute value. */
	attr_len = le32_to_cpu(a->data.resident.value_length);
	BUG_ON(old_i_size != (loff_t)attr_len);
	/*
	 * Do the zeroing in the mft record and update the attribute size in
	 * the mft record.
	 */
	kattr = (u8*)a + le16_to_cpu(a->data.resident.value_offset);
	memset(kattr + attr_len, 0, new_init_size - attr_len);
	a->data.resident.value_length = cpu_to_le32((u32)new_init_size);
	/* Finally, update the sizes in the vfs and ntfs inodes. */
	write_lock_irqsave(&ni->size_lock, flags);
	i_size_write(vi, new_init_size);
	ni->initialized_size = new_init_size;
	write_unlock_irqrestore(&ni->size_lock, flags);
	goto done;
do_non_resident_extend:
	/*
	 * If the new initialized size @new_init_size exceeds the current file
	 * size (vfs inode->i_size), we need to extend the file size to the
	 * new initialized size.
	 */
	if (new_init_size > old_i_size) {
		m = map_mft_record(base_ni);
		if (IS_ERR(m)) {
			err = PTR_ERR(m);
			m = NULL;
			goto err_out;
		}
		ctx = ntfs_attr_get_search_ctx(base_ni, m);
		if (unlikely(!ctx)) {
			err = -ENOMEM;
			goto err_out;
		}
		err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len,
				CASE_SENSITIVE, 0, NULL, 0, ctx);
		if (unlikely(err)) {
			if (err == -ENOENT)
				err = -EIO;
			goto err_out;
		}
		m = ctx->mrec;
		a = ctx->attr;
		BUG_ON(!a->non_resident);
		BUG_ON(old_i_size != (loff_t)
				sle64_to_cpu(a->data.non_resident.data_size));
		a->data.non_resident.data_size = cpu_to_sle64(new_init_size);
		flush_dcache_mft_record_page(ctx->ntfs_ino);
		mark_mft_record_dirty(ctx->ntfs_ino);
		/* Update the file size in the vfs inode. */
		i_size_write(vi, new_init_size);
		ntfs_attr_put_search_ctx(ctx);
		ctx = NULL;
		unmap_mft_record(base_ni);
		m = NULL;
	}
	mapping = vi->i_mapping;
	index = old_init_size >> PAGE_CACHE_SHIFT;
	end_index = (new_init_size + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
	do {
		/*
		 * Read the page.  If the page is not present, this will zero
		 * the uninitialized regions for us.
		 */
234
		page = read_mapping_page(mapping, index, NULL);
235 236 237 238
		if (IS_ERR(page)) {
			err = PTR_ERR(page);
			goto init_err_out;
		}
239
		if (unlikely(PageError(page))) {
240 241 242 243 244 245 246 247 248
			page_cache_release(page);
			err = -EIO;
			goto init_err_out;
		}
		/*
		 * Update the initialized size in the ntfs inode.  This is
		 * enough to make ntfs_writepage() work.
		 */
		write_lock_irqsave(&ni->size_lock, flags);
249
		ni->initialized_size = (s64)(index + 1) << PAGE_CACHE_SHIFT;
250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363
		if (ni->initialized_size > new_init_size)
			ni->initialized_size = new_init_size;
		write_unlock_irqrestore(&ni->size_lock, flags);
		/* Set the page dirty so it gets written out. */
		set_page_dirty(page);
		page_cache_release(page);
		/*
		 * Play nice with the vm and the rest of the system.  This is
		 * very much needed as we can potentially be modifying the
		 * initialised size from a very small value to a really huge
		 * value, e.g.
		 *	f = open(somefile, O_TRUNC);
		 *	truncate(f, 10GiB);
		 *	seek(f, 10GiB);
		 *	write(f, 1);
		 * And this would mean we would be marking dirty hundreds of
		 * thousands of pages or as in the above example more than
		 * two and a half million pages!
		 *
		 * TODO: For sparse pages could optimize this workload by using
		 * the FsMisc / MiscFs page bit as a "PageIsSparse" bit.  This
		 * would be set in readpage for sparse pages and here we would
		 * not need to mark dirty any pages which have this bit set.
		 * The only caveat is that we have to clear the bit everywhere
		 * where we allocate any clusters that lie in the page or that
		 * contain the page.
		 *
		 * TODO: An even greater optimization would be for us to only
		 * call readpage() on pages which are not in sparse regions as
		 * determined from the runlist.  This would greatly reduce the
		 * number of pages we read and make dirty in the case of sparse
		 * files.
		 */
		balance_dirty_pages_ratelimited(mapping);
		cond_resched();
	} while (++index < end_index);
	read_lock_irqsave(&ni->size_lock, flags);
	BUG_ON(ni->initialized_size != new_init_size);
	read_unlock_irqrestore(&ni->size_lock, flags);
	/* Now bring in sync the initialized_size in the mft record. */
	m = map_mft_record(base_ni);
	if (IS_ERR(m)) {
		err = PTR_ERR(m);
		m = NULL;
		goto init_err_out;
	}
	ctx = ntfs_attr_get_search_ctx(base_ni, m);
	if (unlikely(!ctx)) {
		err = -ENOMEM;
		goto init_err_out;
	}
	err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len,
			CASE_SENSITIVE, 0, NULL, 0, ctx);
	if (unlikely(err)) {
		if (err == -ENOENT)
			err = -EIO;
		goto init_err_out;
	}
	m = ctx->mrec;
	a = ctx->attr;
	BUG_ON(!a->non_resident);
	a->data.non_resident.initialized_size = cpu_to_sle64(new_init_size);
done:
	flush_dcache_mft_record_page(ctx->ntfs_ino);
	mark_mft_record_dirty(ctx->ntfs_ino);
	if (ctx)
		ntfs_attr_put_search_ctx(ctx);
	if (m)
		unmap_mft_record(base_ni);
	ntfs_debug("Done, initialized_size 0x%llx, i_size 0x%llx.",
			(unsigned long long)new_init_size, i_size_read(vi));
	return 0;
init_err_out:
	write_lock_irqsave(&ni->size_lock, flags);
	ni->initialized_size = old_init_size;
	write_unlock_irqrestore(&ni->size_lock, flags);
err_out:
	if (ctx)
		ntfs_attr_put_search_ctx(ctx);
	if (m)
		unmap_mft_record(base_ni);
	ntfs_debug("Failed.  Returning error code %i.", err);
	return err;
}

/**
 * ntfs_fault_in_pages_readable -
 *
 * Fault a number of userspace pages into pagetables.
 *
 * Unlike include/linux/pagemap.h::fault_in_pages_readable(), this one copes
 * with more than two userspace pages as well as handling the single page case
 * elegantly.
 *
 * If you find this difficult to understand, then think of the while loop being
 * the following code, except that we do without the integer variable ret:
 *
 *	do {
 *		ret = __get_user(c, uaddr);
 *		uaddr += PAGE_SIZE;
 *	} while (!ret && uaddr < end);
 *
 * Note, the final __get_user() may well run out-of-bounds of the user buffer,
 * but _not_ out-of-bounds of the page the user buffer belongs to, and since
 * this is only a read and not a write, and since it is still in the same page,
 * it should not matter and this makes the code much simpler.
 */
static inline void ntfs_fault_in_pages_readable(const char __user *uaddr,
		int bytes)
{
	const char __user *end;
	volatile char c;

	/* Set @end to the first byte outside the last page we care about. */
364
	end = (const char __user*)PAGE_ALIGN((unsigned long)uaddr + bytes);
365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473

	while (!__get_user(c, uaddr) && (uaddr += PAGE_SIZE, uaddr < end))
		;
}

/**
 * ntfs_fault_in_pages_readable_iovec -
 *
 * Same as ntfs_fault_in_pages_readable() but operates on an array of iovecs.
 */
static inline void ntfs_fault_in_pages_readable_iovec(const struct iovec *iov,
		size_t iov_ofs, int bytes)
{
	do {
		const char __user *buf;
		unsigned len;

		buf = iov->iov_base + iov_ofs;
		len = iov->iov_len - iov_ofs;
		if (len > bytes)
			len = bytes;
		ntfs_fault_in_pages_readable(buf, len);
		bytes -= len;
		iov++;
		iov_ofs = 0;
	} while (bytes);
}

/**
 * __ntfs_grab_cache_pages - obtain a number of locked pages
 * @mapping:	address space mapping from which to obtain page cache pages
 * @index:	starting index in @mapping at which to begin obtaining pages
 * @nr_pages:	number of page cache pages to obtain
 * @pages:	array of pages in which to return the obtained page cache pages
 * @cached_page: allocated but as yet unused page
 * @lru_pvec:	lru-buffering pagevec of caller
 *
 * Obtain @nr_pages locked page cache pages from the mapping @maping and
 * starting at index @index.
 *
 * If a page is newly created, increment its refcount and add it to the
 * caller's lru-buffering pagevec @lru_pvec.
 *
 * This is the same as mm/filemap.c::__grab_cache_page(), except that @nr_pages
 * are obtained at once instead of just one page and that 0 is returned on
 * success and -errno on error.
 *
 * Note, the page locks are obtained in ascending page index order.
 */
static inline int __ntfs_grab_cache_pages(struct address_space *mapping,
		pgoff_t index, const unsigned nr_pages, struct page **pages,
		struct page **cached_page, struct pagevec *lru_pvec)
{
	int err, nr;

	BUG_ON(!nr_pages);
	err = nr = 0;
	do {
		pages[nr] = find_lock_page(mapping, index);
		if (!pages[nr]) {
			if (!*cached_page) {
				*cached_page = page_cache_alloc(mapping);
				if (unlikely(!*cached_page)) {
					err = -ENOMEM;
					goto err_out;
				}
			}
			err = add_to_page_cache(*cached_page, mapping, index,
					GFP_KERNEL);
			if (unlikely(err)) {
				if (err == -EEXIST)
					continue;
				goto err_out;
			}
			pages[nr] = *cached_page;
			page_cache_get(*cached_page);
			if (unlikely(!pagevec_add(lru_pvec, *cached_page)))
				__pagevec_lru_add(lru_pvec);
			*cached_page = NULL;
		}
		index++;
		nr++;
	} while (nr < nr_pages);
out:
	return err;
err_out:
	while (nr > 0) {
		unlock_page(pages[--nr]);
		page_cache_release(pages[nr]);
	}
	goto out;
}

static inline int ntfs_submit_bh_for_read(struct buffer_head *bh)
{
	lock_buffer(bh);
	get_bh(bh);
	bh->b_end_io = end_buffer_read_sync;
	return submit_bh(READ, bh);
}

/**
 * ntfs_prepare_pages_for_non_resident_write - prepare pages for receiving data
 * @pages:	array of destination pages
 * @nr_pages:	number of pages in @pages
 * @pos:	byte position in file at which the write begins
 * @bytes:	number of bytes to be written
 *
 * This is called for non-resident attributes from ntfs_file_buffered_write()
474
 * with i_mutex held on the inode (@pages[0]->mapping->host).  There are
475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510
 * @nr_pages pages in @pages which are locked but not kmap()ped.  The source
 * data has not yet been copied into the @pages.
 * 
 * Need to fill any holes with actual clusters, allocate buffers if necessary,
 * ensure all the buffers are mapped, and bring uptodate any buffers that are
 * only partially being written to.
 *
 * If @nr_pages is greater than one, we are guaranteed that the cluster size is
 * greater than PAGE_CACHE_SIZE, that all pages in @pages are entirely inside
 * the same cluster and that they are the entirety of that cluster, and that
 * the cluster is sparse, i.e. we need to allocate a cluster to fill the hole.
 *
 * i_size is not to be modified yet.
 *
 * Return 0 on success or -errno on error.
 */
static int ntfs_prepare_pages_for_non_resident_write(struct page **pages,
		unsigned nr_pages, s64 pos, size_t bytes)
{
	VCN vcn, highest_vcn = 0, cpos, cend, bh_cpos, bh_cend;
	LCN lcn;
	s64 bh_pos, vcn_len, end, initialized_size;
	sector_t lcn_block;
	struct page *page;
	struct inode *vi;
	ntfs_inode *ni, *base_ni = NULL;
	ntfs_volume *vol;
	runlist_element *rl, *rl2;
	struct buffer_head *bh, *head, *wait[2], **wait_bh = wait;
	ntfs_attr_search_ctx *ctx = NULL;
	MFT_RECORD *m = NULL;
	ATTR_RECORD *a = NULL;
	unsigned long flags;
	u32 attr_rec_len = 0;
	unsigned blocksize, u;
	int err, mp_size;
511
	bool rl_write_locked, was_hole, is_retry;
512 513 514 515 516 517 518 519 520 521 522 523 524 525 526
	unsigned char blocksize_bits;
	struct {
		u8 runlist_merged:1;
		u8 mft_attr_mapped:1;
		u8 mp_rebuilt:1;
		u8 attr_switched:1;
	} status = { 0, 0, 0, 0 };

	BUG_ON(!nr_pages);
	BUG_ON(!pages);
	BUG_ON(!*pages);
	vi = pages[0]->mapping->host;
	ni = NTFS_I(vi);
	vol = ni->vol;
	ntfs_debug("Entering for inode 0x%lx, attribute type 0x%x, start page "
527
			"index 0x%lx, nr_pages 0x%x, pos 0x%llx, bytes 0x%zx.",
528 529
			vi->i_ino, ni->type, pages[0]->index, nr_pages,
			(long long)pos, bytes);
530 531
	blocksize = vol->sb->s_blocksize;
	blocksize_bits = vol->sb->s_blocksize_bits;
532 533
	u = 0;
	do {
534 535
		page = pages[u];
		BUG_ON(!page);
536 537 538 539 540 541 542 543 544 545
		/*
		 * create_empty_buffers() will create uptodate/dirty buffers if
		 * the page is uptodate/dirty.
		 */
		if (!page_has_buffers(page)) {
			create_empty_buffers(page, blocksize, 0);
			if (unlikely(!page_has_buffers(page)))
				return -ENOMEM;
		}
	} while (++u < nr_pages);
546
	rl_write_locked = false;
547 548 549 550 551
	rl = NULL;
	err = 0;
	vcn = lcn = -1;
	vcn_len = 0;
	lcn_block = -1;
552
	was_hole = false;
553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609
	cpos = pos >> vol->cluster_size_bits;
	end = pos + bytes;
	cend = (end + vol->cluster_size - 1) >> vol->cluster_size_bits;
	/*
	 * Loop over each page and for each page over each buffer.  Use goto to
	 * reduce indentation.
	 */
	u = 0;
do_next_page:
	page = pages[u];
	bh_pos = (s64)page->index << PAGE_CACHE_SHIFT;
	bh = head = page_buffers(page);
	do {
		VCN cdelta;
		s64 bh_end;
		unsigned bh_cofs;

		/* Clear buffer_new on all buffers to reinitialise state. */
		if (buffer_new(bh))
			clear_buffer_new(bh);
		bh_end = bh_pos + blocksize;
		bh_cpos = bh_pos >> vol->cluster_size_bits;
		bh_cofs = bh_pos & vol->cluster_size_mask;
		if (buffer_mapped(bh)) {
			/*
			 * The buffer is already mapped.  If it is uptodate,
			 * ignore it.
			 */
			if (buffer_uptodate(bh))
				continue;
			/*
			 * The buffer is not uptodate.  If the page is uptodate
			 * set the buffer uptodate and otherwise ignore it.
			 */
			if (PageUptodate(page)) {
				set_buffer_uptodate(bh);
				continue;
			}
			/*
			 * Neither the page nor the buffer are uptodate.  If
			 * the buffer is only partially being written to, we
			 * need to read it in before the write, i.e. now.
			 */
			if ((bh_pos < pos && bh_end > pos) ||
					(bh_pos < end && bh_end > end)) {
				/*
				 * If the buffer is fully or partially within
				 * the initialized size, do an actual read.
				 * Otherwise, simply zero the buffer.
				 */
				read_lock_irqsave(&ni->size_lock, flags);
				initialized_size = ni->initialized_size;
				read_unlock_irqrestore(&ni->size_lock, flags);
				if (bh_pos < initialized_size) {
					ntfs_submit_bh_for_read(bh);
					*wait_bh++ = bh;
				} else {
610 611
					zero_user(page, bh_offset(bh),
							blocksize);
612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666
					set_buffer_uptodate(bh);
				}
			}
			continue;
		}
		/* Unmapped buffer.  Need to map it. */
		bh->b_bdev = vol->sb->s_bdev;
		/*
		 * If the current buffer is in the same clusters as the map
		 * cache, there is no need to check the runlist again.  The
		 * map cache is made up of @vcn, which is the first cached file
		 * cluster, @vcn_len which is the number of cached file
		 * clusters, @lcn is the device cluster corresponding to @vcn,
		 * and @lcn_block is the block number corresponding to @lcn.
		 */
		cdelta = bh_cpos - vcn;
		if (likely(!cdelta || (cdelta > 0 && cdelta < vcn_len))) {
map_buffer_cached:
			BUG_ON(lcn < 0);
			bh->b_blocknr = lcn_block +
					(cdelta << (vol->cluster_size_bits -
					blocksize_bits)) +
					(bh_cofs >> blocksize_bits);
			set_buffer_mapped(bh);
			/*
			 * If the page is uptodate so is the buffer.  If the
			 * buffer is fully outside the write, we ignore it if
			 * it was already allocated and we mark it dirty so it
			 * gets written out if we allocated it.  On the other
			 * hand, if we allocated the buffer but we are not
			 * marking it dirty we set buffer_new so we can do
			 * error recovery.
			 */
			if (PageUptodate(page)) {
				if (!buffer_uptodate(bh))
					set_buffer_uptodate(bh);
				if (unlikely(was_hole)) {
					/* We allocated the buffer. */
					unmap_underlying_metadata(bh->b_bdev,
							bh->b_blocknr);
					if (bh_end <= pos || bh_pos >= end)
						mark_buffer_dirty(bh);
					else
						set_buffer_new(bh);
				}
				continue;
			}
			/* Page is _not_ uptodate. */
			if (likely(!was_hole)) {
				/*
				 * Buffer was already allocated.  If it is not
				 * uptodate and is only partially being written
				 * to, we need to read it in before the write,
				 * i.e. now.
				 */
667 668 669 670
				if (!buffer_uptodate(bh) && bh_pos < end &&
						bh_end > pos &&
						(bh_pos < pos ||
						bh_end > end)) {
671 672 673 674 675 676 677 678 679 680 681 682 683 684 685
					/*
					 * If the buffer is fully or partially
					 * within the initialized size, do an
					 * actual read.  Otherwise, simply zero
					 * the buffer.
					 */
					read_lock_irqsave(&ni->size_lock,
							flags);
					initialized_size = ni->initialized_size;
					read_unlock_irqrestore(&ni->size_lock,
							flags);
					if (bh_pos < initialized_size) {
						ntfs_submit_bh_for_read(bh);
						*wait_bh++ = bh;
					} else {
686 687
						zero_user(page, bh_offset(bh),
								blocksize);
688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704
						set_buffer_uptodate(bh);
					}
				}
				continue;
			}
			/* We allocated the buffer. */
			unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr);
			/*
			 * If the buffer is fully outside the write, zero it,
			 * set it uptodate, and mark it dirty so it gets
			 * written out.  If it is partially being written to,
			 * zero region surrounding the write but leave it to
			 * commit write to do anything else.  Finally, if the
			 * buffer is fully being overwritten, do nothing.
			 */
			if (bh_end <= pos || bh_pos >= end) {
				if (!buffer_uptodate(bh)) {
705 706
					zero_user(page, bh_offset(bh),
							blocksize);
707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744
					set_buffer_uptodate(bh);
				}
				mark_buffer_dirty(bh);
				continue;
			}
			set_buffer_new(bh);
			if (!buffer_uptodate(bh) &&
					(bh_pos < pos || bh_end > end)) {
				u8 *kaddr;
				unsigned pofs;
					
				kaddr = kmap_atomic(page, KM_USER0);
				if (bh_pos < pos) {
					pofs = bh_pos & ~PAGE_CACHE_MASK;
					memset(kaddr + pofs, 0, pos - bh_pos);
				}
				if (bh_end > end) {
					pofs = end & ~PAGE_CACHE_MASK;
					memset(kaddr + pofs, 0, bh_end - end);
				}
				kunmap_atomic(kaddr, KM_USER0);
				flush_dcache_page(page);
			}
			continue;
		}
		/*
		 * Slow path: this is the first buffer in the cluster.  If it
		 * is outside allocated size and is not uptodate, zero it and
		 * set it uptodate.
		 */
		read_lock_irqsave(&ni->size_lock, flags);
		initialized_size = ni->allocated_size;
		read_unlock_irqrestore(&ni->size_lock, flags);
		if (bh_pos > initialized_size) {
			if (PageUptodate(page)) {
				if (!buffer_uptodate(bh))
					set_buffer_uptodate(bh);
			} else if (!buffer_uptodate(bh)) {
745
				zero_user(page, bh_offset(bh), blocksize);
746 747 748 749
				set_buffer_uptodate(bh);
			}
			continue;
		}
750
		is_retry = false;
751 752 753 754 755 756 757 758 759 760 761 762 763 764 765
		if (!rl) {
			down_read(&ni->runlist.lock);
retry_remap:
			rl = ni->runlist.rl;
		}
		if (likely(rl != NULL)) {
			/* Seek to element containing target cluster. */
			while (rl->length && rl[1].vcn <= bh_cpos)
				rl++;
			lcn = ntfs_rl_vcn_to_lcn(rl, bh_cpos);
			if (likely(lcn >= 0)) {
				/*
				 * Successful remap, setup the map cache and
				 * use that to deal with the buffer.
				 */
766
				was_hole = false;
767 768 769 770
				vcn = bh_cpos;
				vcn_len = rl[1].vcn - vcn;
				lcn_block = lcn << (vol->cluster_size_bits -
						blocksize_bits);
771
				cdelta = 0;
772
				/*
773 774 775 776 777
				 * If the number of remaining clusters touched
				 * by the write is smaller or equal to the
				 * number of cached clusters, unlock the
				 * runlist as the map cache will be used from
				 * now on.
778 779 780 781
				 */
				if (likely(vcn + vcn_len >= cend)) {
					if (rl_write_locked) {
						up_write(&ni->runlist.lock);
782
						rl_write_locked = false;
783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807
					} else
						up_read(&ni->runlist.lock);
					rl = NULL;
				}
				goto map_buffer_cached;
			}
		} else
			lcn = LCN_RL_NOT_MAPPED;
		/*
		 * If it is not a hole and not out of bounds, the runlist is
		 * probably unmapped so try to map it now.
		 */
		if (unlikely(lcn != LCN_HOLE && lcn != LCN_ENOENT)) {
			if (likely(!is_retry && lcn == LCN_RL_NOT_MAPPED)) {
				/* Attempt to map runlist. */
				if (!rl_write_locked) {
					/*
					 * We need the runlist locked for
					 * writing, so if it is locked for
					 * reading relock it now and retry in
					 * case it changed whilst we dropped
					 * the lock.
					 */
					up_read(&ni->runlist.lock);
					down_write(&ni->runlist.lock);
808
					rl_write_locked = true;
809 810 811 812 813
					goto retry_remap;
				}
				err = ntfs_map_runlist_nolock(ni, bh_cpos,
						NULL);
				if (likely(!err)) {
814
					is_retry = true;
815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868
					goto retry_remap;
				}
				/*
				 * If @vcn is out of bounds, pretend @lcn is
				 * LCN_ENOENT.  As long as the buffer is out
				 * of bounds this will work fine.
				 */
				if (err == -ENOENT) {
					lcn = LCN_ENOENT;
					err = 0;
					goto rl_not_mapped_enoent;
				}
			} else
				err = -EIO;
			/* Failed to map the buffer, even after retrying. */
			bh->b_blocknr = -1;
			ntfs_error(vol->sb, "Failed to write to inode 0x%lx, "
					"attribute type 0x%x, vcn 0x%llx, "
					"vcn offset 0x%x, because its "
					"location on disk could not be "
					"determined%s (error code %i).",
					ni->mft_no, ni->type,
					(unsigned long long)bh_cpos,
					(unsigned)bh_pos &
					vol->cluster_size_mask,
					is_retry ? " even after retrying" : "",
					err);
			break;
		}
rl_not_mapped_enoent:
		/*
		 * The buffer is in a hole or out of bounds.  We need to fill
		 * the hole, unless the buffer is in a cluster which is not
		 * touched by the write, in which case we just leave the buffer
		 * unmapped.  This can only happen when the cluster size is
		 * less than the page cache size.
		 */
		if (unlikely(vol->cluster_size < PAGE_CACHE_SIZE)) {
			bh_cend = (bh_end + vol->cluster_size - 1) >>
					vol->cluster_size_bits;
			if ((bh_cend <= cpos || bh_cpos >= cend)) {
				bh->b_blocknr = -1;
				/*
				 * If the buffer is uptodate we skip it.  If it
				 * is not but the page is uptodate, we can set
				 * the buffer uptodate.  If the page is not
				 * uptodate, we can clear the buffer and set it
				 * uptodate.  Whether this is worthwhile is
				 * debatable and this could be removed.
				 */
				if (PageUptodate(page)) {
					if (!buffer_uptodate(bh))
						set_buffer_uptodate(bh);
				} else if (!buffer_uptodate(bh)) {
869 870
					zero_user(page, bh_offset(bh),
						blocksize);
871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889
					set_buffer_uptodate(bh);
				}
				continue;
			}
		}
		/*
		 * Out of bounds buffer is invalid if it was not really out of
		 * bounds.
		 */
		BUG_ON(lcn != LCN_HOLE);
		/*
		 * We need the runlist locked for writing, so if it is locked
		 * for reading relock it now and retry in case it changed
		 * whilst we dropped the lock.
		 */
		BUG_ON(!rl);
		if (!rl_write_locked) {
			up_read(&ni->runlist.lock);
			down_write(&ni->runlist.lock);
890
			rl_write_locked = true;
891 892 893 894 895 896 897 898 899 900 901 902 903
			goto retry_remap;
		}
		/* Find the previous last allocated cluster. */
		BUG_ON(rl->lcn != LCN_HOLE);
		lcn = -1;
		rl2 = rl;
		while (--rl2 >= ni->runlist.rl) {
			if (rl2->lcn >= 0) {
				lcn = rl2->lcn + rl2->length;
				break;
			}
		}
		rl2 = ntfs_cluster_alloc(vol, bh_cpos, 1, lcn, DATA_ZONE,
904
				false);
905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928
		if (IS_ERR(rl2)) {
			err = PTR_ERR(rl2);
			ntfs_debug("Failed to allocate cluster, error code %i.",
					err);
			break;
		}
		lcn = rl2->lcn;
		rl = ntfs_runlists_merge(ni->runlist.rl, rl2);
		if (IS_ERR(rl)) {
			err = PTR_ERR(rl);
			if (err != -ENOMEM)
				err = -EIO;
			if (ntfs_cluster_free_from_rl(vol, rl2)) {
				ntfs_error(vol->sb, "Failed to release "
						"allocated cluster in error "
						"code path.  Run chkdsk to "
						"recover the lost cluster.");
				NVolSetErrors(vol);
			}
			ntfs_free(rl2);
			break;
		}
		ni->runlist.rl = rl;
		status.runlist_merged = 1;
929 930
		ntfs_debug("Allocated cluster, lcn 0x%llx.",
				(unsigned long long)lcn);
931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079
		/* Map and lock the mft record and get the attribute record. */
		if (!NInoAttr(ni))
			base_ni = ni;
		else
			base_ni = ni->ext.base_ntfs_ino;
		m = map_mft_record(base_ni);
		if (IS_ERR(m)) {
			err = PTR_ERR(m);
			break;
		}
		ctx = ntfs_attr_get_search_ctx(base_ni, m);
		if (unlikely(!ctx)) {
			err = -ENOMEM;
			unmap_mft_record(base_ni);
			break;
		}
		status.mft_attr_mapped = 1;
		err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len,
				CASE_SENSITIVE, bh_cpos, NULL, 0, ctx);
		if (unlikely(err)) {
			if (err == -ENOENT)
				err = -EIO;
			break;
		}
		m = ctx->mrec;
		a = ctx->attr;
		/*
		 * Find the runlist element with which the attribute extent
		 * starts.  Note, we cannot use the _attr_ version because we
		 * have mapped the mft record.  That is ok because we know the
		 * runlist fragment must be mapped already to have ever gotten
		 * here, so we can just use the _rl_ version.
		 */
		vcn = sle64_to_cpu(a->data.non_resident.lowest_vcn);
		rl2 = ntfs_rl_find_vcn_nolock(rl, vcn);
		BUG_ON(!rl2);
		BUG_ON(!rl2->length);
		BUG_ON(rl2->lcn < LCN_HOLE);
		highest_vcn = sle64_to_cpu(a->data.non_resident.highest_vcn);
		/*
		 * If @highest_vcn is zero, calculate the real highest_vcn
		 * (which can really be zero).
		 */
		if (!highest_vcn)
			highest_vcn = (sle64_to_cpu(
					a->data.non_resident.allocated_size) >>
					vol->cluster_size_bits) - 1;
		/*
		 * Determine the size of the mapping pairs array for the new
		 * extent, i.e. the old extent with the hole filled.
		 */
		mp_size = ntfs_get_size_for_mapping_pairs(vol, rl2, vcn,
				highest_vcn);
		if (unlikely(mp_size <= 0)) {
			if (!(err = mp_size))
				err = -EIO;
			ntfs_debug("Failed to get size for mapping pairs "
					"array, error code %i.", err);
			break;
		}
		/*
		 * Resize the attribute record to fit the new mapping pairs
		 * array.
		 */
		attr_rec_len = le32_to_cpu(a->length);
		err = ntfs_attr_record_resize(m, a, mp_size + le16_to_cpu(
				a->data.non_resident.mapping_pairs_offset));
		if (unlikely(err)) {
			BUG_ON(err != -ENOSPC);
			// TODO: Deal with this by using the current attribute
			// and fill it with as much of the mapping pairs
			// array as possible.  Then loop over each attribute
			// extent rewriting the mapping pairs arrays as we go
			// along and if when we reach the end we have not
			// enough space, try to resize the last attribute
			// extent and if even that fails, add a new attribute
			// extent.
			// We could also try to resize at each step in the hope
			// that we will not need to rewrite every single extent.
			// Note, we may need to decompress some extents to fill
			// the runlist as we are walking the extents...
			ntfs_error(vol->sb, "Not enough space in the mft "
					"record for the extended attribute "
					"record.  This case is not "
					"implemented yet.");
			err = -EOPNOTSUPP;
			break ;
		}
		status.mp_rebuilt = 1;
		/*
		 * Generate the mapping pairs array directly into the attribute
		 * record.
		 */
		err = ntfs_mapping_pairs_build(vol, (u8*)a + le16_to_cpu(
				a->data.non_resident.mapping_pairs_offset),
				mp_size, rl2, vcn, highest_vcn, NULL);
		if (unlikely(err)) {
			ntfs_error(vol->sb, "Cannot fill hole in inode 0x%lx, "
					"attribute type 0x%x, because building "
					"the mapping pairs failed with error "
					"code %i.", vi->i_ino,
					(unsigned)le32_to_cpu(ni->type), err);
			err = -EIO;
			break;
		}
		/* Update the highest_vcn but only if it was not set. */
		if (unlikely(!a->data.non_resident.highest_vcn))
			a->data.non_resident.highest_vcn =
					cpu_to_sle64(highest_vcn);
		/*
		 * If the attribute is sparse/compressed, update the compressed
		 * size in the ntfs_inode structure and the attribute record.
		 */
		if (likely(NInoSparse(ni) || NInoCompressed(ni))) {
			/*
			 * If we are not in the first attribute extent, switch
			 * to it, but first ensure the changes will make it to
			 * disk later.
			 */
			if (a->data.non_resident.lowest_vcn) {
				flush_dcache_mft_record_page(ctx->ntfs_ino);
				mark_mft_record_dirty(ctx->ntfs_ino);
				ntfs_attr_reinit_search_ctx(ctx);
				err = ntfs_attr_lookup(ni->type, ni->name,
						ni->name_len, CASE_SENSITIVE,
						0, NULL, 0, ctx);
				if (unlikely(err)) {
					status.attr_switched = 1;
					break;
				}
				/* @m is not used any more so do not set it. */
				a = ctx->attr;
			}
			write_lock_irqsave(&ni->size_lock, flags);
			ni->itype.compressed.size += vol->cluster_size;
			a->data.non_resident.compressed_size =
					cpu_to_sle64(ni->itype.compressed.size);
			write_unlock_irqrestore(&ni->size_lock, flags);
		}
		/* Ensure the changes make it to disk. */
		flush_dcache_mft_record_page(ctx->ntfs_ino);
		mark_mft_record_dirty(ctx->ntfs_ino);
		ntfs_attr_put_search_ctx(ctx);
		unmap_mft_record(base_ni);
		/* Successfully filled the hole. */
		status.runlist_merged = 0;
		status.mft_attr_mapped = 0;
		status.mp_rebuilt = 0;
		/* Setup the map cache and use that to deal with the buffer. */
1080
		was_hole = true;
1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091
		vcn = bh_cpos;
		vcn_len = 1;
		lcn_block = lcn << (vol->cluster_size_bits - blocksize_bits);
		cdelta = 0;
		/*
		 * If the number of remaining clusters in the @pages is smaller
		 * or equal to the number of cached clusters, unlock the
		 * runlist as the map cache will be used from now on.
		 */
		if (likely(vcn + vcn_len >= cend)) {
			up_write(&ni->runlist.lock);
1092
			rl_write_locked = false;
1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103
			rl = NULL;
		}
		goto map_buffer_cached;
	} while (bh_pos += blocksize, (bh = bh->b_this_page) != head);
	/* If there are no errors, do the next page. */
	if (likely(!err && ++u < nr_pages))
		goto do_next_page;
	/* If there are no errors, release the runlist lock if we took it. */
	if (likely(!err)) {
		if (unlikely(rl_write_locked)) {
			up_write(&ni->runlist.lock);
1104
			rl_write_locked = false;
1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128
		} else if (unlikely(rl))
			up_read(&ni->runlist.lock);
		rl = NULL;
	}
	/* If we issued read requests, let them complete. */
	read_lock_irqsave(&ni->size_lock, flags);
	initialized_size = ni->initialized_size;
	read_unlock_irqrestore(&ni->size_lock, flags);
	while (wait_bh > wait) {
		bh = *--wait_bh;
		wait_on_buffer(bh);
		if (likely(buffer_uptodate(bh))) {
			page = bh->b_page;
			bh_pos = ((s64)page->index << PAGE_CACHE_SHIFT) +
					bh_offset(bh);
			/*
			 * If the buffer overflows the initialized size, need
			 * to zero the overflowing region.
			 */
			if (unlikely(bh_pos + blocksize > initialized_size)) {
				int ofs = 0;

				if (likely(bh_pos < initialized_size))
					ofs = initialized_size - bh_pos;
1129 1130
				zero_user_segment(page, bh_offset(bh) + ofs,
						blocksize);
1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269
			}
		} else /* if (unlikely(!buffer_uptodate(bh))) */
			err = -EIO;
	}
	if (likely(!err)) {
		/* Clear buffer_new on all buffers. */
		u = 0;
		do {
			bh = head = page_buffers(pages[u]);
			do {
				if (buffer_new(bh))
					clear_buffer_new(bh);
			} while ((bh = bh->b_this_page) != head);
		} while (++u < nr_pages);
		ntfs_debug("Done.");
		return err;
	}
	if (status.attr_switched) {
		/* Get back to the attribute extent we modified. */
		ntfs_attr_reinit_search_ctx(ctx);
		if (ntfs_attr_lookup(ni->type, ni->name, ni->name_len,
				CASE_SENSITIVE, bh_cpos, NULL, 0, ctx)) {
			ntfs_error(vol->sb, "Failed to find required "
					"attribute extent of attribute in "
					"error code path.  Run chkdsk to "
					"recover.");
			write_lock_irqsave(&ni->size_lock, flags);
			ni->itype.compressed.size += vol->cluster_size;
			write_unlock_irqrestore(&ni->size_lock, flags);
			flush_dcache_mft_record_page(ctx->ntfs_ino);
			mark_mft_record_dirty(ctx->ntfs_ino);
			/*
			 * The only thing that is now wrong is the compressed
			 * size of the base attribute extent which chkdsk
			 * should be able to fix.
			 */
			NVolSetErrors(vol);
		} else {
			m = ctx->mrec;
			a = ctx->attr;
			status.attr_switched = 0;
		}
	}
	/*
	 * If the runlist has been modified, need to restore it by punching a
	 * hole into it and we then need to deallocate the on-disk cluster as
	 * well.  Note, we only modify the runlist if we are able to generate a
	 * new mapping pairs array, i.e. only when the mapped attribute extent
	 * is not switched.
	 */
	if (status.runlist_merged && !status.attr_switched) {
		BUG_ON(!rl_write_locked);
		/* Make the file cluster we allocated sparse in the runlist. */
		if (ntfs_rl_punch_nolock(vol, &ni->runlist, bh_cpos, 1)) {
			ntfs_error(vol->sb, "Failed to punch hole into "
					"attribute runlist in error code "
					"path.  Run chkdsk to recover the "
					"lost cluster.");
			NVolSetErrors(vol);
		} else /* if (success) */ {
			status.runlist_merged = 0;
			/*
			 * Deallocate the on-disk cluster we allocated but only
			 * if we succeeded in punching its vcn out of the
			 * runlist.
			 */
			down_write(&vol->lcnbmp_lock);
			if (ntfs_bitmap_clear_bit(vol->lcnbmp_ino, lcn)) {
				ntfs_error(vol->sb, "Failed to release "
						"allocated cluster in error "
						"code path.  Run chkdsk to "
						"recover the lost cluster.");
				NVolSetErrors(vol);
			}
			up_write(&vol->lcnbmp_lock);
		}
	}
	/*
	 * Resize the attribute record to its old size and rebuild the mapping
	 * pairs array.  Note, we only can do this if the runlist has been
	 * restored to its old state which also implies that the mapped
	 * attribute extent is not switched.
	 */
	if (status.mp_rebuilt && !status.runlist_merged) {
		if (ntfs_attr_record_resize(m, a, attr_rec_len)) {
			ntfs_error(vol->sb, "Failed to restore attribute "
					"record in error code path.  Run "
					"chkdsk to recover.");
			NVolSetErrors(vol);
		} else /* if (success) */ {
			if (ntfs_mapping_pairs_build(vol, (u8*)a +
					le16_to_cpu(a->data.non_resident.
					mapping_pairs_offset), attr_rec_len -
					le16_to_cpu(a->data.non_resident.
					mapping_pairs_offset), ni->runlist.rl,
					vcn, highest_vcn, NULL)) {
				ntfs_error(vol->sb, "Failed to restore "
						"mapping pairs array in error "
						"code path.  Run chkdsk to "
						"recover.");
				NVolSetErrors(vol);
			}
			flush_dcache_mft_record_page(ctx->ntfs_ino);
			mark_mft_record_dirty(ctx->ntfs_ino);
		}
	}
	/* Release the mft record and the attribute. */
	if (status.mft_attr_mapped) {
		ntfs_attr_put_search_ctx(ctx);
		unmap_mft_record(base_ni);
	}
	/* Release the runlist lock. */
	if (rl_write_locked)
		up_write(&ni->runlist.lock);
	else if (rl)
		up_read(&ni->runlist.lock);
	/*
	 * Zero out any newly allocated blocks to avoid exposing stale data.
	 * If BH_New is set, we know that the block was newly allocated above
	 * and that it has not been fully zeroed and marked dirty yet.
	 */
	nr_pages = u;
	u = 0;
	end = bh_cpos << vol->cluster_size_bits;
	do {
		page = pages[u];
		bh = head = page_buffers(page);
		do {
			if (u == nr_pages &&
					((s64)page->index << PAGE_CACHE_SHIFT) +
					bh_offset(bh) >= end)
				break;
			if (!buffer_new(bh))
				continue;
			clear_buffer_new(bh);
			if (!buffer_uptodate(bh)) {
				if (PageUptodate(page))
					set_buffer_uptodate(bh);
				else {
1270 1271
					zero_user(page, bh_offset(bh),
							blocksize);
1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291
					set_buffer_uptodate(bh);
				}
			}
			mark_buffer_dirty(bh);
		} while ((bh = bh->b_this_page) != head);
	} while (++u <= nr_pages);
	ntfs_error(vol->sb, "Failed.  Returning error code %i.", err);
	return err;
}

/*
 * Copy as much as we can into the pages and return the number of bytes which
 * were sucessfully copied.  If a fault is encountered then clear the pages
 * out to (ofs + bytes) and return the number of bytes which were copied.
 */
static inline size_t ntfs_copy_from_user(struct page **pages,
		unsigned nr_pages, unsigned ofs, const char __user *buf,
		size_t bytes)
{
	struct page **last_page = pages + nr_pages;
1292
	char *addr;
1293 1294 1295 1296 1297 1298 1299 1300
	size_t total = 0;
	unsigned len;
	int left;

	do {
		len = PAGE_CACHE_SIZE - ofs;
		if (len > bytes)
			len = bytes;
1301 1302 1303
		addr = kmap_atomic(*pages, KM_USER0);
		left = __copy_from_user_inatomic(addr + ofs, buf, len);
		kunmap_atomic(addr, KM_USER0);
1304 1305
		if (unlikely(left)) {
			/* Do it the slow way. */
1306 1307
			addr = kmap(*pages);
			left = __copy_from_user(addr + ofs, buf, len);
1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330
			kunmap(*pages);
			if (unlikely(left))
				goto err_out;
		}
		total += len;
		bytes -= len;
		if (!bytes)
			break;
		buf += len;
		ofs = 0;
	} while (++pages < last_page);
out:
	return total;
err_out:
	total += len - left;
	/* Zero the rest of the target like __copy_from_user(). */
	while (++pages < last_page) {
		bytes -= len;
		if (!bytes)
			break;
		len = PAGE_CACHE_SIZE;
		if (len > bytes)
			len = bytes;
1331
		zero_user(*pages, 0, len);
1332 1333 1334 1335
	}
	goto out;
}

1336
static size_t __ntfs_copy_from_user_iovec_inatomic(char *vaddr,
1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393
		const struct iovec *iov, size_t iov_ofs, size_t bytes)
{
	size_t total = 0;

	while (1) {
		const char __user *buf = iov->iov_base + iov_ofs;
		unsigned len;
		size_t left;

		len = iov->iov_len - iov_ofs;
		if (len > bytes)
			len = bytes;
		left = __copy_from_user_inatomic(vaddr, buf, len);
		total += len;
		bytes -= len;
		vaddr += len;
		if (unlikely(left)) {
			total -= left;
			break;
		}
		if (!bytes)
			break;
		iov++;
		iov_ofs = 0;
	}
	return total;
}

static inline void ntfs_set_next_iovec(const struct iovec **iovp,
		size_t *iov_ofsp, size_t bytes)
{
	const struct iovec *iov = *iovp;
	size_t iov_ofs = *iov_ofsp;

	while (bytes) {
		unsigned len;

		len = iov->iov_len - iov_ofs;
		if (len > bytes)
			len = bytes;
		bytes -= len;
		iov_ofs += len;
		if (iov->iov_len == iov_ofs) {
			iov++;
			iov_ofs = 0;
		}
	}
	*iovp = iov;
	*iov_ofsp = iov_ofs;
}

/*
 * This has the same side-effects and return value as ntfs_copy_from_user().
 * The difference is that on a fault we need to memset the remainder of the
 * pages (out to offset + bytes), to emulate ntfs_copy_from_user()'s
 * single-segment behaviour.
 *
1394 1395 1396 1397 1398 1399 1400
 * We call the same helper (__ntfs_copy_from_user_iovec_inatomic()) both
 * when atomic and when not atomic.  This is ok because
 * __ntfs_copy_from_user_iovec_inatomic() calls __copy_from_user_inatomic()
 * and it is ok to call this when non-atomic.
 * Infact, the only difference between __copy_from_user_inatomic() and
 * __copy_from_user() is that the latter calls might_sleep() and the former
 * should not zero the tail of the buffer on error.  And on many
1401 1402 1403 1404 1405 1406 1407 1408
 * architectures __copy_from_user_inatomic() is just defined to
 * __copy_from_user() so it makes no difference at all on those architectures.
 */
static inline size_t ntfs_copy_from_user_iovec(struct page **pages,
		unsigned nr_pages, unsigned ofs, const struct iovec **iov,
		size_t *iov_ofs, size_t bytes)
{
	struct page **last_page = pages + nr_pages;
1409
	char *addr;
1410 1411 1412 1413 1414 1415
	size_t copied, len, total = 0;

	do {
		len = PAGE_CACHE_SIZE - ofs;
		if (len > bytes)
			len = bytes;
1416 1417
		addr = kmap_atomic(*pages, KM_USER0);
		copied = __ntfs_copy_from_user_iovec_inatomic(addr + ofs,
1418
				*iov, *iov_ofs, len);
1419
		kunmap_atomic(addr, KM_USER0);
1420 1421
		if (unlikely(copied != len)) {
			/* Do it the slow way. */
1422 1423
			addr = kmap(*pages);
			copied = __ntfs_copy_from_user_iovec_inatomic(addr + ofs,
1424
					*iov, *iov_ofs, len);
1425 1426 1427
			/*
			 * Zero the rest of the target like __copy_from_user().
			 */
1428
			memset(addr + ofs + copied, 0, len - copied);
1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451
			kunmap(*pages);
			if (unlikely(copied != len))
				goto err_out;
		}
		total += len;
		bytes -= len;
		if (!bytes)
			break;
		ntfs_set_next_iovec(iov, iov_ofs, len);
		ofs = 0;
	} while (++pages < last_page);
out:
	return total;
err_out:
	total += copied;
	/* Zero the rest of the target like __copy_from_user(). */
	while (++pages < last_page) {
		bytes -= len;
		if (!bytes)
			break;
		len = PAGE_CACHE_SIZE;
		if (len > bytes)
			len = bytes;
1452
		zero_user(*pages, 0, len);
1453 1454 1455 1456 1457 1458 1459 1460
	}
	goto out;
}

static inline void ntfs_flush_dcache_pages(struct page **pages,
		unsigned nr_pages)
{
	BUG_ON(!nr_pages);
1461 1462 1463 1464 1465
	/*
	 * Warning: Do not do the decrement at the same time as the call to
	 * flush_dcache_page() because it is a NULL macro on i386 and hence the
	 * decrement never happens so the loop never terminates.
	 */
1466
	do {
1467
		--nr_pages;
1468
		flush_dcache_page(pages[nr_pages]);
1469
	} while (nr_pages > 0);
1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497
}

/**
 * ntfs_commit_pages_after_non_resident_write - commit the received data
 * @pages:	array of destination pages
 * @nr_pages:	number of pages in @pages
 * @pos:	byte position in file at which the write begins
 * @bytes:	number of bytes to be written
 *
 * See description of ntfs_commit_pages_after_write(), below.
 */
static inline int ntfs_commit_pages_after_non_resident_write(
		struct page **pages, const unsigned nr_pages,
		s64 pos, size_t bytes)
{
	s64 end, initialized_size;
	struct inode *vi;
	ntfs_inode *ni, *base_ni;
	struct buffer_head *bh, *head;
	ntfs_attr_search_ctx *ctx;
	MFT_RECORD *m;
	ATTR_RECORD *a;
	unsigned long flags;
	unsigned blocksize, u;
	int err;

	vi = pages[0]->mapping->host;
	ni = NTFS_I(vi);
1498
	blocksize = vi->i_sb->s_blocksize;
1499 1500 1501 1502 1503
	end = pos + bytes;
	u = 0;
	do {
		s64 bh_pos;
		struct page *page;
1504
		bool partial;
1505 1506 1507 1508

		page = pages[u];
		bh_pos = (s64)page->index << PAGE_CACHE_SHIFT;
		bh = head = page_buffers(page);
1509
		partial = false;
1510 1511 1512 1513 1514 1515
		do {
			s64 bh_end;

			bh_end = bh_pos + blocksize;
			if (bh_end <= pos || bh_pos >= end) {
				if (!buffer_uptodate(bh))
1516
					partial = true;
1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594
			} else {
				set_buffer_uptodate(bh);
				mark_buffer_dirty(bh);
			}
		} while (bh_pos += blocksize, (bh = bh->b_this_page) != head);
		/*
		 * If all buffers are now uptodate but the page is not, set the
		 * page uptodate.
		 */
		if (!partial && !PageUptodate(page))
			SetPageUptodate(page);
	} while (++u < nr_pages);
	/*
	 * Finally, if we do not need to update initialized_size or i_size we
	 * are finished.
	 */
	read_lock_irqsave(&ni->size_lock, flags);
	initialized_size = ni->initialized_size;
	read_unlock_irqrestore(&ni->size_lock, flags);
	if (end <= initialized_size) {
		ntfs_debug("Done.");
		return 0;
	}
	/*
	 * Update initialized_size/i_size as appropriate, both in the inode and
	 * the mft record.
	 */
	if (!NInoAttr(ni))
		base_ni = ni;
	else
		base_ni = ni->ext.base_ntfs_ino;
	/* Map, pin, and lock the mft record. */
	m = map_mft_record(base_ni);
	if (IS_ERR(m)) {
		err = PTR_ERR(m);
		m = NULL;
		ctx = NULL;
		goto err_out;
	}
	BUG_ON(!NInoNonResident(ni));
	ctx = ntfs_attr_get_search_ctx(base_ni, m);
	if (unlikely(!ctx)) {
		err = -ENOMEM;
		goto err_out;
	}
	err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len,
			CASE_SENSITIVE, 0, NULL, 0, ctx);
	if (unlikely(err)) {
		if (err == -ENOENT)
			err = -EIO;
		goto err_out;
	}
	a = ctx->attr;
	BUG_ON(!a->non_resident);
	write_lock_irqsave(&ni->size_lock, flags);
	BUG_ON(end > ni->allocated_size);
	ni->initialized_size = end;
	a->data.non_resident.initialized_size = cpu_to_sle64(end);
	if (end > i_size_read(vi)) {
		i_size_write(vi, end);
		a->data.non_resident.data_size =
				a->data.non_resident.initialized_size;
	}
	write_unlock_irqrestore(&ni->size_lock, flags);
	/* Mark the mft record dirty, so it gets written back. */
	flush_dcache_mft_record_page(ctx->ntfs_ino);
	mark_mft_record_dirty(ctx->ntfs_ino);
	ntfs_attr_put_search_ctx(ctx);
	unmap_mft_record(base_ni);
	ntfs_debug("Done.");
	return 0;
err_out:
	if (ctx)
		ntfs_attr_put_search_ctx(ctx);
	if (m)
		unmap_mft_record(base_ni);
	ntfs_error(vi->i_sb, "Failed to update initialized_size/i_size (error "
			"code %i).", err);
1595
	if (err != -ENOMEM)
1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606
		NVolSetErrors(ni->vol);
	return err;
}

/**
 * ntfs_commit_pages_after_write - commit the received data
 * @pages:	array of destination pages
 * @nr_pages:	number of pages in @pages
 * @pos:	byte position in file at which the write begins
 * @bytes:	number of bytes to be written
 *
1607
 * This is called from ntfs_file_buffered_write() with i_mutex held on the inode
1608 1609 1610 1611 1612 1613 1614 1615 1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658
 * (@pages[0]->mapping->host).  There are @nr_pages pages in @pages which are
 * locked but not kmap()ped.  The source data has already been copied into the
 * @page.  ntfs_prepare_pages_for_non_resident_write() has been called before
 * the data was copied (for non-resident attributes only) and it returned
 * success.
 *
 * Need to set uptodate and mark dirty all buffers within the boundary of the
 * write.  If all buffers in a page are uptodate we set the page uptodate, too.
 *
 * Setting the buffers dirty ensures that they get written out later when
 * ntfs_writepage() is invoked by the VM.
 *
 * Finally, we need to update i_size and initialized_size as appropriate both
 * in the inode and the mft record.
 *
 * This is modelled after fs/buffer.c::generic_commit_write(), which marks
 * buffers uptodate and dirty, sets the page uptodate if all buffers in the
 * page are uptodate, and updates i_size if the end of io is beyond i_size.  In
 * that case, it also marks the inode dirty.
 *
 * If things have gone as outlined in
 * ntfs_prepare_pages_for_non_resident_write(), we do not need to do any page
 * content modifications here for non-resident attributes.  For resident
 * attributes we need to do the uptodate bringing here which we combine with
 * the copying into the mft record which means we save one atomic kmap.
 *
 * Return 0 on success or -errno on error.
 */
static int ntfs_commit_pages_after_write(struct page **pages,
		const unsigned nr_pages, s64 pos, size_t bytes)
{
	s64 end, initialized_size;
	loff_t i_size;
	struct inode *vi;
	ntfs_inode *ni, *base_ni;
	struct page *page;
	ntfs_attr_search_ctx *ctx;
	MFT_RECORD *m;
	ATTR_RECORD *a;
	char *kattr, *kaddr;
	unsigned long flags;
	u32 attr_len;
	int err;

	BUG_ON(!nr_pages);
	BUG_ON(!pages);
	page = pages[0];
	BUG_ON(!page);
	vi = page->mapping->host;
	ni = NTFS_I(vi);
	ntfs_debug("Entering for inode 0x%lx, attribute type 0x%x, start page "
1659
			"index 0x%lx, nr_pages 0x%x, pos 0x%llx, bytes 0x%zx.",
1660 1661 1662 1663 1664 1665 1666 1667 1668 1669 1670 1671 1672 1673 1674 1675 1676 1677 1678 1679 1680 1681 1682 1683 1684 1685 1686 1687 1688 1689 1690 1691 1692 1693 1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 1706 1707 1708 1709 1710 1711 1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726 1727 1728 1729 1730 1731 1732 1733 1734 1735 1736 1737 1738 1739 1740 1741 1742 1743 1744 1745 1746 1747 1748 1749 1750 1751 1752 1753 1754 1755