1// SPDX-License-Identifier: GPL-2.0-or-later
   2/**
   3 * aops.c - NTFS kernel address space operations and page cache handling.
 
   4 *
   5 * Copyright (c) 2001-2014 Anton Altaparmakov and Tuxera Inc.
   6 * Copyright (c) 2002 Richard Russon
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
   7 */
   8
   9#include <linux/errno.h>
  10#include <linux/fs.h>
  11#include <linux/gfp.h>
  12#include <linux/mm.h>
  13#include <linux/pagemap.h>
  14#include <linux/swap.h>
  15#include <linux/buffer_head.h>
  16#include <linux/writeback.h>
  17#include <linux/bit_spinlock.h>
  18#include <linux/bio.h>
  19
  20#include "aops.h"
  21#include "attrib.h"
  22#include "debug.h"
  23#include "inode.h"
  24#include "mft.h"
  25#include "runlist.h"
  26#include "types.h"
  27#include "ntfs.h"
  28
  29/**
  30 * ntfs_end_buffer_async_read - async io completion for reading attributes
  31 * @bh:		buffer head on which io is completed
  32 * @uptodate:	whether @bh is now uptodate or not
  33 *
  34 * Asynchronous I/O completion handler for reading pages belonging to the
  35 * attribute address space of an inode.  The inodes can either be files or
  36 * directories or they can be fake inodes describing some attribute.
  37 *
  38 * If NInoMstProtected(), perform the post read mst fixups when all IO on the
  39 * page has been completed and mark the page uptodate or set the error bit on
  40 * the page.  To determine the size of the records that need fixing up, we
  41 * cheat a little bit by setting the index_block_size in ntfs_inode to the ntfs
  42 * record size, and index_block_size_bits, to the log(base 2) of the ntfs
  43 * record size.
  44 */
  45static void ntfs_end_buffer_async_read(struct buffer_head *bh, int uptodate)
  46{
  47	unsigned long flags;
  48	struct buffer_head *first, *tmp;
  49	struct page *page;
  50	struct inode *vi;
  51	ntfs_inode *ni;
  52	int page_uptodate = 1;
  53
  54	page = bh->b_page;
  55	vi = page->mapping->host;
  56	ni = NTFS_I(vi);
  57
  58	if (likely(uptodate)) {
  59		loff_t i_size;
  60		s64 file_ofs, init_size;
  61
  62		set_buffer_uptodate(bh);
  63
  64		file_ofs = ((s64)page->index << PAGE_SHIFT) +
  65				bh_offset(bh);
  66		read_lock_irqsave(&ni->size_lock, flags);
  67		init_size = ni->initialized_size;
  68		i_size = i_size_read(vi);
  69		read_unlock_irqrestore(&ni->size_lock, flags);
  70		if (unlikely(init_size > i_size)) {
  71			/* Race with shrinking truncate. */
  72			init_size = i_size;
  73		}
  74		/* Check for the current buffer head overflowing. */
  75		if (unlikely(file_ofs + bh->b_size > init_size)) {
  76			int ofs;
  77			void *kaddr;
  78
  79			ofs = 0;
  80			if (file_ofs < init_size)
  81				ofs = init_size - file_ofs;
 
  82			kaddr = kmap_atomic(page);
  83			memset(kaddr + bh_offset(bh) + ofs, 0,
  84					bh->b_size - ofs);
  85			flush_dcache_page(page);
  86			kunmap_atomic(kaddr);
 
  87		}
  88	} else {
  89		clear_buffer_uptodate(bh);
  90		SetPageError(page);
  91		ntfs_error(ni->vol->sb, "Buffer I/O error, logical block "
  92				"0x%llx.", (unsigned long long)bh->b_blocknr);
  93	}
  94	first = page_buffers(page);
  95	spin_lock_irqsave(&first->b_uptodate_lock, flags);
 
  96	clear_buffer_async_read(bh);
  97	unlock_buffer(bh);
  98	tmp = bh;
  99	do {
 100		if (!buffer_uptodate(tmp))
 101			page_uptodate = 0;
 102		if (buffer_async_read(tmp)) {
 103			if (likely(buffer_locked(tmp)))
 104				goto still_busy;
 105			/* Async buffers must be locked. */
 106			BUG();
 107		}
 108		tmp = tmp->b_this_page;
 109	} while (tmp != bh);
 110	spin_unlock_irqrestore(&first->b_uptodate_lock, flags);
 
 111	/*
 112	 * If none of the buffers had errors then we can set the page uptodate,
 113	 * but we first have to perform the post read mst fixups, if the
 114	 * attribute is mst protected, i.e. if NInoMstProteced(ni) is true.
 115	 * Note we ignore fixup errors as those are detected when
 116	 * map_mft_record() is called which gives us per record granularity
 117	 * rather than per page granularity.
 118	 */
 119	if (!NInoMstProtected(ni)) {
 120		if (likely(page_uptodate && !PageError(page)))
 121			SetPageUptodate(page);
 122	} else {
 123		u8 *kaddr;
 124		unsigned int i, recs;
 125		u32 rec_size;
 126
 127		rec_size = ni->itype.index.block_size;
 128		recs = PAGE_SIZE / rec_size;
 129		/* Should have been verified before we got here... */
 130		BUG_ON(!recs);
 
 131		kaddr = kmap_atomic(page);
 132		for (i = 0; i < recs; i++)
 133			post_read_mst_fixup((NTFS_RECORD*)(kaddr +
 134					i * rec_size), rec_size);
 135		kunmap_atomic(kaddr);
 
 136		flush_dcache_page(page);
 137		if (likely(page_uptodate && !PageError(page)))
 138			SetPageUptodate(page);
 139	}
 140	unlock_page(page);
 141	return;
 142still_busy:
 143	spin_unlock_irqrestore(&first->b_uptodate_lock, flags);
 
 144	return;
 145}
 146
 147/**
 148 * ntfs_read_block - fill a @page of an address space with data
 149 * @page:	page cache page to fill with data
 150 *
 151 * Fill the page @page of the address space belonging to the @page->host inode.
 152 * We read each buffer asynchronously and when all buffers are read in, our io
 153 * completion handler ntfs_end_buffer_read_async(), if required, automatically
 154 * applies the mst fixups to the page before finally marking it uptodate and
 155 * unlocking it.
 156 *
 157 * We only enforce allocated_size limit because i_size is checked for in
 158 * generic_file_read().
 159 *
 160 * Return 0 on success and -errno on error.
 161 *
 162 * Contains an adapted version of fs/buffer.c::block_read_full_folio().
 163 */
 164static int ntfs_read_block(struct page *page)
 165{
 166	loff_t i_size;
 167	VCN vcn;
 168	LCN lcn;
 169	s64 init_size;
 170	struct inode *vi;
 171	ntfs_inode *ni;
 172	ntfs_volume *vol;
 173	runlist_element *rl;
 174	struct buffer_head *bh, *head, *arr[MAX_BUF_PER_PAGE];
 175	sector_t iblock, lblock, zblock;
 176	unsigned long flags;
 177	unsigned int blocksize, vcn_ofs;
 178	int i, nr;
 179	unsigned char blocksize_bits;
 180
 181	vi = page->mapping->host;
 182	ni = NTFS_I(vi);
 183	vol = ni->vol;
 184
 185	/* $MFT/$DATA must have its complete runlist in memory at all times. */
 186	BUG_ON(!ni->runlist.rl && !ni->mft_no && !NInoAttr(ni));
 187
 188	blocksize = vol->sb->s_blocksize;
 189	blocksize_bits = vol->sb->s_blocksize_bits;
 190
 191	if (!page_has_buffers(page)) {
 192		create_empty_buffers(page, blocksize, 0);
 193		if (unlikely(!page_has_buffers(page))) {
 194			unlock_page(page);
 195			return -ENOMEM;
 196		}
 197	}
 198	bh = head = page_buffers(page);
 199	BUG_ON(!bh);
 200
 201	/*
 202	 * We may be racing with truncate.  To avoid some of the problems we
 203	 * now take a snapshot of the various sizes and use those for the whole
 204	 * of the function.  In case of an extending truncate it just means we
 205	 * may leave some buffers unmapped which are now allocated.  This is
 206	 * not a problem since these buffers will just get mapped when a write
 207	 * occurs.  In case of a shrinking truncate, we will detect this later
 208	 * on due to the runlist being incomplete and if the page is being
 209	 * fully truncated, truncate will throw it away as soon as we unlock
 210	 * it so no need to worry what we do with it.
 211	 */
 212	iblock = (s64)page->index << (PAGE_SHIFT - blocksize_bits);
 213	read_lock_irqsave(&ni->size_lock, flags);
 214	lblock = (ni->allocated_size + blocksize - 1) >> blocksize_bits;
 215	init_size = ni->initialized_size;
 216	i_size = i_size_read(vi);
 217	read_unlock_irqrestore(&ni->size_lock, flags);
 218	if (unlikely(init_size > i_size)) {
 219		/* Race with shrinking truncate. */
 220		init_size = i_size;
 221	}
 222	zblock = (init_size + blocksize - 1) >> blocksize_bits;
 223
 224	/* Loop through all the buffers in the page. */
 225	rl = NULL;
 226	nr = i = 0;
 227	do {
 228		int err = 0;
 229
 230		if (unlikely(buffer_uptodate(bh)))
 231			continue;
 232		if (unlikely(buffer_mapped(bh))) {
 233			arr[nr++] = bh;
 234			continue;
 235		}
 236		bh->b_bdev = vol->sb->s_bdev;
 237		/* Is the block within the allowed limits? */
 238		if (iblock < lblock) {
 239			bool is_retry = false;
 240
 241			/* Convert iblock into corresponding vcn and offset. */
 242			vcn = (VCN)iblock << blocksize_bits >>
 243					vol->cluster_size_bits;
 244			vcn_ofs = ((VCN)iblock << blocksize_bits) &
 245					vol->cluster_size_mask;
 246			if (!rl) {
 247lock_retry_remap:
 248				down_read(&ni->runlist.lock);
 249				rl = ni->runlist.rl;
 250			}
 251			if (likely(rl != NULL)) {
 252				/* Seek to element containing target vcn. */
 253				while (rl->length && rl[1].vcn <= vcn)
 254					rl++;
 255				lcn = ntfs_rl_vcn_to_lcn(rl, vcn);
 256			} else
 257				lcn = LCN_RL_NOT_MAPPED;
 258			/* Successful remap. */
 259			if (lcn >= 0) {
 260				/* Setup buffer head to correct block. */
 261				bh->b_blocknr = ((lcn << vol->cluster_size_bits)
 262						+ vcn_ofs) >> blocksize_bits;
 263				set_buffer_mapped(bh);
 264				/* Only read initialized data blocks. */
 265				if (iblock < zblock) {
 266					arr[nr++] = bh;
 267					continue;
 268				}
 269				/* Fully non-initialized data block, zero it. */
 270				goto handle_zblock;
 271			}
 272			/* It is a hole, need to zero it. */
 273			if (lcn == LCN_HOLE)
 274				goto handle_hole;
 275			/* If first try and runlist unmapped, map and retry. */
 276			if (!is_retry && lcn == LCN_RL_NOT_MAPPED) {
 277				is_retry = true;
 278				/*
 279				 * Attempt to map runlist, dropping lock for
 280				 * the duration.
 281				 */
 282				up_read(&ni->runlist.lock);
 283				err = ntfs_map_runlist(ni, vcn);
 284				if (likely(!err))
 285					goto lock_retry_remap;
 286				rl = NULL;
 287			} else if (!rl)
 288				up_read(&ni->runlist.lock);
 289			/*
 290			 * If buffer is outside the runlist, treat it as a
 291			 * hole.  This can happen due to concurrent truncate
 292			 * for example.
 293			 */
 294			if (err == -ENOENT || lcn == LCN_ENOENT) {
 295				err = 0;
 296				goto handle_hole;
 297			}
 298			/* Hard error, zero out region. */
 299			if (!err)
 300				err = -EIO;
 301			bh->b_blocknr = -1;
 302			SetPageError(page);
 303			ntfs_error(vol->sb, "Failed to read from inode 0x%lx, "
 304					"attribute type 0x%x, vcn 0x%llx, "
 305					"offset 0x%x because its location on "
 306					"disk could not be determined%s "
 307					"(error code %i).", ni->mft_no,
 308					ni->type, (unsigned long long)vcn,
 309					vcn_ofs, is_retry ? " even after "
 310					"retrying" : "", err);
 311		}
 312		/*
 313		 * Either iblock was outside lblock limits or
 314		 * ntfs_rl_vcn_to_lcn() returned error.  Just zero that portion
 315		 * of the page and set the buffer uptodate.
 316		 */
 317handle_hole:
 318		bh->b_blocknr = -1UL;
 319		clear_buffer_mapped(bh);
 320handle_zblock:
 321		zero_user(page, i * blocksize, blocksize);
 322		if (likely(!err))
 323			set_buffer_uptodate(bh);
 324	} while (i++, iblock++, (bh = bh->b_this_page) != head);
 325
 326	/* Release the lock if we took it. */
 327	if (rl)
 328		up_read(&ni->runlist.lock);
 329
 330	/* Check we have at least one buffer ready for i/o. */
 331	if (nr) {
 332		struct buffer_head *tbh;
 333
 334		/* Lock the buffers. */
 335		for (i = 0; i < nr; i++) {
 336			tbh = arr[i];
 337			lock_buffer(tbh);
 338			tbh->b_end_io = ntfs_end_buffer_async_read;
 339			set_buffer_async_read(tbh);
 340		}
 341		/* Finally, start i/o on the buffers. */
 342		for (i = 0; i < nr; i++) {
 343			tbh = arr[i];
 344			if (likely(!buffer_uptodate(tbh)))
 345				submit_bh(REQ_OP_READ, tbh);
 346			else
 347				ntfs_end_buffer_async_read(tbh, 1);
 348		}
 349		return 0;
 350	}
 351	/* No i/o was scheduled on any of the buffers. */
 352	if (likely(!PageError(page)))
 353		SetPageUptodate(page);
 354	else /* Signal synchronous i/o error. */
 355		nr = -EIO;
 356	unlock_page(page);
 357	return nr;
 358}
 359
 360/**
 361 * ntfs_read_folio - fill a @folio of a @file with data from the device
 362 * @file:	open file to which the folio @folio belongs or NULL
 363 * @folio:	page cache folio to fill with data
 364 *
 365 * For non-resident attributes, ntfs_read_folio() fills the @folio of the open
 366 * file @file by calling the ntfs version of the generic block_read_full_folio()
 367 * function, ntfs_read_block(), which in turn creates and reads in the buffers
 368 * associated with the folio asynchronously.
 369 *
 370 * For resident attributes, OTOH, ntfs_read_folio() fills @folio by copying the
 371 * data from the mft record (which at this stage is most likely in memory) and
 372 * fills the remainder with zeroes. Thus, in this case, I/O is synchronous, as
 373 * even if the mft record is not cached at this point in time, we need to wait
 374 * for it to be read in before we can do the copy.
 375 *
 376 * Return 0 on success and -errno on error.
 377 */
 378static int ntfs_read_folio(struct file *file, struct folio *folio)
 379{
 380	struct page *page = &folio->page;
 381	loff_t i_size;
 382	struct inode *vi;
 383	ntfs_inode *ni, *base_ni;
 384	u8 *addr;
 385	ntfs_attr_search_ctx *ctx;
 386	MFT_RECORD *mrec;
 387	unsigned long flags;
 388	u32 attr_len;
 389	int err = 0;
 390
 391retry_readpage:
 392	BUG_ON(!PageLocked(page));
 393	vi = page->mapping->host;
 394	i_size = i_size_read(vi);
 395	/* Is the page fully outside i_size? (truncate in progress) */
 396	if (unlikely(page->index >= (i_size + PAGE_SIZE - 1) >>
 397			PAGE_SHIFT)) {
 398		zero_user(page, 0, PAGE_SIZE);
 399		ntfs_debug("Read outside i_size - truncated?");
 400		goto done;
 401	}
 402	/*
 403	 * This can potentially happen because we clear PageUptodate() during
 404	 * ntfs_writepage() of MstProtected() attributes.
 405	 */
 406	if (PageUptodate(page)) {
 407		unlock_page(page);
 408		return 0;
 409	}
 410	ni = NTFS_I(vi);
 411	/*
 412	 * Only $DATA attributes can be encrypted and only unnamed $DATA
 413	 * attributes can be compressed.  Index root can have the flags set but
 414	 * this means to create compressed/encrypted files, not that the
 415	 * attribute is compressed/encrypted.  Note we need to check for
 416	 * AT_INDEX_ALLOCATION since this is the type of both directory and
 417	 * index inodes.
 418	 */
 419	if (ni->type != AT_INDEX_ALLOCATION) {
 420		/* If attribute is encrypted, deny access, just like NT4. */
 421		if (NInoEncrypted(ni)) {
 422			BUG_ON(ni->type != AT_DATA);
 423			err = -EACCES;
 424			goto err_out;
 425		}
 426		/* Compressed data streams are handled in compress.c. */
 427		if (NInoNonResident(ni) && NInoCompressed(ni)) {
 428			BUG_ON(ni->type != AT_DATA);
 429			BUG_ON(ni->name_len);
 430			return ntfs_read_compressed_block(page);
 431		}
 432	}
 433	/* NInoNonResident() == NInoIndexAllocPresent() */
 434	if (NInoNonResident(ni)) {
 435		/* Normal, non-resident data stream. */
 436		return ntfs_read_block(page);
 437	}
 438	/*
 439	 * Attribute is resident, implying it is not compressed or encrypted.
 440	 * This also means the attribute is smaller than an mft record and
 441	 * hence smaller than a page, so can simply zero out any pages with
 442	 * index above 0.  Note the attribute can actually be marked compressed
 443	 * but if it is resident the actual data is not compressed so we are
 444	 * ok to ignore the compressed flag here.
 445	 */
 446	if (unlikely(page->index > 0)) {
 447		zero_user(page, 0, PAGE_SIZE);
 448		goto done;
 449	}
 450	if (!NInoAttr(ni))
 451		base_ni = ni;
 452	else
 453		base_ni = ni->ext.base_ntfs_ino;
 454	/* Map, pin, and lock the mft record. */
 455	mrec = map_mft_record(base_ni);
 456	if (IS_ERR(mrec)) {
 457		err = PTR_ERR(mrec);
 458		goto err_out;
 459	}
 460	/*
 461	 * If a parallel write made the attribute non-resident, drop the mft
 462	 * record and retry the read_folio.
 463	 */
 464	if (unlikely(NInoNonResident(ni))) {
 465		unmap_mft_record(base_ni);
 466		goto retry_readpage;
 467	}
 468	ctx = ntfs_attr_get_search_ctx(base_ni, mrec);
 469	if (unlikely(!ctx)) {
 470		err = -ENOMEM;
 471		goto unm_err_out;
 472	}
 473	err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len,
 474			CASE_SENSITIVE, 0, NULL, 0, ctx);
 475	if (unlikely(err))
 476		goto put_unm_err_out;
 477	attr_len = le32_to_cpu(ctx->attr->data.resident.value_length);
 478	read_lock_irqsave(&ni->size_lock, flags);
 479	if (unlikely(attr_len > ni->initialized_size))
 480		attr_len = ni->initialized_size;
 481	i_size = i_size_read(vi);
 482	read_unlock_irqrestore(&ni->size_lock, flags);
 483	if (unlikely(attr_len > i_size)) {
 484		/* Race with shrinking truncate. */
 485		attr_len = i_size;
 486	}
 487	addr = kmap_atomic(page);
 488	/* Copy the data to the page. */
 489	memcpy(addr, (u8*)ctx->attr +
 490			le16_to_cpu(ctx->attr->data.resident.value_offset),
 491			attr_len);
 492	/* Zero the remainder of the page. */
 493	memset(addr + attr_len, 0, PAGE_SIZE - attr_len);
 494	flush_dcache_page(page);
 495	kunmap_atomic(addr);
 496put_unm_err_out:
 497	ntfs_attr_put_search_ctx(ctx);
 498unm_err_out:
 499	unmap_mft_record(base_ni);
 500done:
 501	SetPageUptodate(page);
 502err_out:
 503	unlock_page(page);
 504	return err;
 505}
 506
 507#ifdef NTFS_RW
 508
 509/**
 510 * ntfs_write_block - write a @page to the backing store
 511 * @page:	page cache page to write out
 512 * @wbc:	writeback control structure
 513 *
 514 * This function is for writing pages belonging to non-resident, non-mst
 515 * protected attributes to their backing store.
 516 *
 517 * For a page with buffers, map and write the dirty buffers asynchronously
 518 * under page writeback. For a page without buffers, create buffers for the
 519 * page, then proceed as above.
 520 *
 521 * If a page doesn't have buffers the page dirty state is definitive. If a page
 522 * does have buffers, the page dirty state is just a hint, and the buffer dirty
 523 * state is definitive. (A hint which has rules: dirty buffers against a clean
 524 * page is illegal. Other combinations are legal and need to be handled. In
 525 * particular a dirty page containing clean buffers for example.)
 526 *
 527 * Return 0 on success and -errno on error.
 528 *
 529 * Based on ntfs_read_block() and __block_write_full_page().
 530 */
 531static int ntfs_write_block(struct page *page, struct writeback_control *wbc)
 532{
 533	VCN vcn;
 534	LCN lcn;
 535	s64 initialized_size;
 536	loff_t i_size;
 537	sector_t block, dblock, iblock;
 538	struct inode *vi;
 539	ntfs_inode *ni;
 540	ntfs_volume *vol;
 541	runlist_element *rl;
 542	struct buffer_head *bh, *head;
 543	unsigned long flags;
 544	unsigned int blocksize, vcn_ofs;
 545	int err;
 546	bool need_end_writeback;
 547	unsigned char blocksize_bits;
 548
 549	vi = page->mapping->host;
 550	ni = NTFS_I(vi);
 551	vol = ni->vol;
 552
 553	ntfs_debug("Entering for inode 0x%lx, attribute type 0x%x, page index "
 554			"0x%lx.", ni->mft_no, ni->type, page->index);
 555
 556	BUG_ON(!NInoNonResident(ni));
 557	BUG_ON(NInoMstProtected(ni));
 558	blocksize = vol->sb->s_blocksize;
 559	blocksize_bits = vol->sb->s_blocksize_bits;
 560	if (!page_has_buffers(page)) {
 561		BUG_ON(!PageUptodate(page));
 562		create_empty_buffers(page, blocksize,
 563				(1 << BH_Uptodate) | (1 << BH_Dirty));
 564		if (unlikely(!page_has_buffers(page))) {
 565			ntfs_warning(vol->sb, "Error allocating page "
 566					"buffers.  Redirtying page so we try "
 567					"again later.");
 568			/*
 569			 * Put the page back on mapping->dirty_pages, but leave
 570			 * its buffers' dirty state as-is.
 571			 */
 572			redirty_page_for_writepage(wbc, page);
 573			unlock_page(page);
 574			return 0;
 575		}
 576	}
 577	bh = head = page_buffers(page);
 578	BUG_ON(!bh);
 579
 580	/* NOTE: Different naming scheme to ntfs_read_block()! */
 581
 582	/* The first block in the page. */
 583	block = (s64)page->index << (PAGE_SHIFT - blocksize_bits);
 584
 585	read_lock_irqsave(&ni->size_lock, flags);
 586	i_size = i_size_read(vi);
 587	initialized_size = ni->initialized_size;
 588	read_unlock_irqrestore(&ni->size_lock, flags);
 589
 590	/* The first out of bounds block for the data size. */
 591	dblock = (i_size + blocksize - 1) >> blocksize_bits;
 592
 593	/* The last (fully or partially) initialized block. */
 594	iblock = initialized_size >> blocksize_bits;
 595
 596	/*
 597	 * Be very careful.  We have no exclusion from block_dirty_folio
 598	 * here, and the (potentially unmapped) buffers may become dirty at
 599	 * any time.  If a buffer becomes dirty here after we've inspected it
 600	 * then we just miss that fact, and the page stays dirty.
 601	 *
 602	 * Buffers outside i_size may be dirtied by block_dirty_folio;
 603	 * handle that here by just cleaning them.
 604	 */
 605
 606	/*
 607	 * Loop through all the buffers in the page, mapping all the dirty
 608	 * buffers to disk addresses and handling any aliases from the
 609	 * underlying block device's mapping.
 610	 */
 611	rl = NULL;
 612	err = 0;
 613	do {
 614		bool is_retry = false;
 615
 616		if (unlikely(block >= dblock)) {
 617			/*
 618			 * Mapped buffers outside i_size will occur, because
 619			 * this page can be outside i_size when there is a
 620			 * truncate in progress. The contents of such buffers
 621			 * were zeroed by ntfs_writepage().
 622			 *
 623			 * FIXME: What about the small race window where
 624			 * ntfs_writepage() has not done any clearing because
 625			 * the page was within i_size but before we get here,
 626			 * vmtruncate() modifies i_size?
 627			 */
 628			clear_buffer_dirty(bh);
 629			set_buffer_uptodate(bh);
 630			continue;
 631		}
 632
 633		/* Clean buffers are not written out, so no need to map them. */
 634		if (!buffer_dirty(bh))
 635			continue;
 636
 637		/* Make sure we have enough initialized size. */
 638		if (unlikely((block >= iblock) &&
 639				(initialized_size < i_size))) {
 640			/*
 641			 * If this page is fully outside initialized
 642			 * size, zero out all pages between the current
 643			 * initialized size and the current page. Just
 644			 * use ntfs_read_folio() to do the zeroing
 645			 * transparently.
 646			 */
 647			if (block > iblock) {
 648				// TODO:
 649				// For each page do:
 650				// - read_cache_page()
 651				// Again for each page do:
 652				// - wait_on_page_locked()
 653				// - Check (PageUptodate(page) &&
 654				//			!PageError(page))
 655				// Update initialized size in the attribute and
 656				// in the inode.
 657				// Again, for each page do:
 658				//	block_dirty_folio();
 659				// put_page()
 660				// We don't need to wait on the writes.
 661				// Update iblock.
 662			}
 663			/*
 664			 * The current page straddles initialized size. Zero
 665			 * all non-uptodate buffers and set them uptodate (and
 666			 * dirty?). Note, there aren't any non-uptodate buffers
 667			 * if the page is uptodate.
 668			 * FIXME: For an uptodate page, the buffers may need to
 669			 * be written out because they were not initialized on
 670			 * disk before.
 671			 */
 672			if (!PageUptodate(page)) {
 673				// TODO:
 674				// Zero any non-uptodate buffers up to i_size.
 675				// Set them uptodate and dirty.
 676			}
 677			// TODO:
 678			// Update initialized size in the attribute and in the
 679			// inode (up to i_size).
 680			// Update iblock.
 681			// FIXME: This is inefficient. Try to batch the two
 682			// size changes to happen in one go.
 683			ntfs_error(vol->sb, "Writing beyond initialized size "
 684					"is not supported yet. Sorry.");
 685			err = -EOPNOTSUPP;
 686			break;
 687			// Do NOT set_buffer_new() BUT DO clear buffer range
 688			// outside write request range.
 689			// set_buffer_uptodate() on complete buffers as well as
 690			// set_buffer_dirty().
 691		}
 692
 693		/* No need to map buffers that are already mapped. */
 694		if (buffer_mapped(bh))
 695			continue;
 696
 697		/* Unmapped, dirty buffer. Need to map it. */
 698		bh->b_bdev = vol->sb->s_bdev;
 699
 700		/* Convert block into corresponding vcn and offset. */
 701		vcn = (VCN)block << blocksize_bits;
 702		vcn_ofs = vcn & vol->cluster_size_mask;
 703		vcn >>= vol->cluster_size_bits;
 704		if (!rl) {
 705lock_retry_remap:
 706			down_read(&ni->runlist.lock);
 707			rl = ni->runlist.rl;
 708		}
 709		if (likely(rl != NULL)) {
 710			/* Seek to element containing target vcn. */
 711			while (rl->length && rl[1].vcn <= vcn)
 712				rl++;
 713			lcn = ntfs_rl_vcn_to_lcn(rl, vcn);
 714		} else
 715			lcn = LCN_RL_NOT_MAPPED;
 716		/* Successful remap. */
 717		if (lcn >= 0) {
 718			/* Setup buffer head to point to correct block. */
 719			bh->b_blocknr = ((lcn << vol->cluster_size_bits) +
 720					vcn_ofs) >> blocksize_bits;
 721			set_buffer_mapped(bh);
 722			continue;
 723		}
 724		/* It is a hole, need to instantiate it. */
 725		if (lcn == LCN_HOLE) {
 726			u8 *kaddr;
 727			unsigned long *bpos, *bend;
 728
 729			/* Check if the buffer is zero. */
 730			kaddr = kmap_atomic(page);
 731			bpos = (unsigned long *)(kaddr + bh_offset(bh));
 732			bend = (unsigned long *)((u8*)bpos + blocksize);
 733			do {
 734				if (unlikely(*bpos))
 735					break;
 736			} while (likely(++bpos < bend));
 737			kunmap_atomic(kaddr);
 738			if (bpos == bend) {
 739				/*
 740				 * Buffer is zero and sparse, no need to write
 741				 * it.
 742				 */
 743				bh->b_blocknr = -1;
 744				clear_buffer_dirty(bh);
 745				continue;
 746			}
 747			// TODO: Instantiate the hole.
 748			// clear_buffer_new(bh);
 749			// clean_bdev_bh_alias(bh);
 750			ntfs_error(vol->sb, "Writing into sparse regions is "
 751					"not supported yet. Sorry.");
 752			err = -EOPNOTSUPP;
 753			break;
 754		}
 755		/* If first try and runlist unmapped, map and retry. */
 756		if (!is_retry && lcn == LCN_RL_NOT_MAPPED) {
 757			is_retry = true;
 758			/*
 759			 * Attempt to map runlist, dropping lock for
 760			 * the duration.
 761			 */
 762			up_read(&ni->runlist.lock);
 763			err = ntfs_map_runlist(ni, vcn);
 764			if (likely(!err))
 765				goto lock_retry_remap;
 766			rl = NULL;
 767		} else if (!rl)
 768			up_read(&ni->runlist.lock);
 769		/*
 770		 * If buffer is outside the runlist, truncate has cut it out
 771		 * of the runlist.  Just clean and clear the buffer and set it
 772		 * uptodate so it can get discarded by the VM.
 773		 */
 774		if (err == -ENOENT || lcn == LCN_ENOENT) {
 775			bh->b_blocknr = -1;
 776			clear_buffer_dirty(bh);
 777			zero_user(page, bh_offset(bh), blocksize);
 778			set_buffer_uptodate(bh);
 779			err = 0;
 780			continue;
 781		}
 782		/* Failed to map the buffer, even after retrying. */
 783		if (!err)
 784			err = -EIO;
 785		bh->b_blocknr = -1;
 786		ntfs_error(vol->sb, "Failed to write to inode 0x%lx, "
 787				"attribute type 0x%x, vcn 0x%llx, offset 0x%x "
 788				"because its location on disk could not be "
 789				"determined%s (error code %i).", ni->mft_no,
 790				ni->type, (unsigned long long)vcn,
 791				vcn_ofs, is_retry ? " even after "
 792				"retrying" : "", err);
 793		break;
 794	} while (block++, (bh = bh->b_this_page) != head);
 795
 796	/* Release the lock if we took it. */
 797	if (rl)
 798		up_read(&ni->runlist.lock);
 799
 800	/* For the error case, need to reset bh to the beginning. */
 801	bh = head;
 802
 803	/* Just an optimization, so ->read_folio() is not called later. */
 804	if (unlikely(!PageUptodate(page))) {
 805		int uptodate = 1;
 806		do {
 807			if (!buffer_uptodate(bh)) {
 808				uptodate = 0;
 809				bh = head;
 810				break;
 811			}
 812		} while ((bh = bh->b_this_page) != head);
 813		if (uptodate)
 814			SetPageUptodate(page);
 815	}
 816
 817	/* Setup all mapped, dirty buffers for async write i/o. */
 818	do {
 819		if (buffer_mapped(bh) && buffer_dirty(bh)) {
 820			lock_buffer(bh);
 821			if (test_clear_buffer_dirty(bh)) {
 822				BUG_ON(!buffer_uptodate(bh));
 823				mark_buffer_async_write(bh);
 824			} else
 825				unlock_buffer(bh);
 826		} else if (unlikely(err)) {
 827			/*
 828			 * For the error case. The buffer may have been set
 829			 * dirty during attachment to a dirty page.
 830			 */
 831			if (err != -ENOMEM)
 832				clear_buffer_dirty(bh);
 833		}
 834	} while ((bh = bh->b_this_page) != head);
 835
 836	if (unlikely(err)) {
 837		// TODO: Remove the -EOPNOTSUPP check later on...
 838		if (unlikely(err == -EOPNOTSUPP))
 839			err = 0;
 840		else if (err == -ENOMEM) {
 841			ntfs_warning(vol->sb, "Error allocating memory. "
 842					"Redirtying page so we try again "
 843					"later.");
 844			/*
 845			 * Put the page back on mapping->dirty_pages, but
 846			 * leave its buffer's dirty state as-is.
 847			 */
 848			redirty_page_for_writepage(wbc, page);
 849			err = 0;
 850		} else
 851			SetPageError(page);
 852	}
 853
 854	BUG_ON(PageWriteback(page));
 855	set_page_writeback(page);	/* Keeps try_to_free_buffers() away. */
 856
 857	/* Submit the prepared buffers for i/o. */
 858	need_end_writeback = true;
 859	do {
 860		struct buffer_head *next = bh->b_this_page;
 861		if (buffer_async_write(bh)) {
 862			submit_bh(REQ_OP_WRITE, bh);
 863			need_end_writeback = false;
 864		}
 865		bh = next;
 866	} while (bh != head);
 867	unlock_page(page);
 868
 869	/* If no i/o was started, need to end_page_writeback(). */
 870	if (unlikely(need_end_writeback))
 871		end_page_writeback(page);
 872
 873	ntfs_debug("Done.");
 874	return err;
 875}
 876
 877/**
 878 * ntfs_write_mst_block - write a @page to the backing store
 879 * @page:	page cache page to write out
 880 * @wbc:	writeback control structure
 881 *
 882 * This function is for writing pages belonging to non-resident, mst protected
 883 * attributes to their backing store.  The only supported attributes are index
 884 * allocation and $MFT/$DATA.  Both directory inodes and index inodes are
 885 * supported for the index allocation case.
 886 *
 887 * The page must remain locked for the duration of the write because we apply
 888 * the mst fixups, write, and then undo the fixups, so if we were to unlock the
 889 * page before undoing the fixups, any other user of the page will see the
 890 * page contents as corrupt.
 891 *
 892 * We clear the page uptodate flag for the duration of the function to ensure
 893 * exclusion for the $MFT/$DATA case against someone mapping an mft record we
 894 * are about to apply the mst fixups to.
 895 *
 896 * Return 0 on success and -errno on error.
 897 *
 898 * Based on ntfs_write_block(), ntfs_mft_writepage(), and
 899 * write_mft_record_nolock().
 900 */
 901static int ntfs_write_mst_block(struct page *page,
 902		struct writeback_control *wbc)
 903{
 904	sector_t block, dblock, rec_block;
 905	struct inode *vi = page->mapping->host;
 906	ntfs_inode *ni = NTFS_I(vi);
 907	ntfs_volume *vol = ni->vol;
 908	u8 *kaddr;
 909	unsigned int rec_size = ni->itype.index.block_size;
 910	ntfs_inode *locked_nis[PAGE_SIZE / NTFS_BLOCK_SIZE];
 911	struct buffer_head *bh, *head, *tbh, *rec_start_bh;
 912	struct buffer_head *bhs[MAX_BUF_PER_PAGE];
 913	runlist_element *rl;
 914	int i, nr_locked_nis, nr_recs, nr_bhs, max_bhs, bhs_per_rec, err, err2;
 915	unsigned bh_size, rec_size_bits;
 916	bool sync, is_mft, page_is_dirty, rec_is_dirty;
 917	unsigned char bh_size_bits;
 918
 919	if (WARN_ON(rec_size < NTFS_BLOCK_SIZE))
 920		return -EINVAL;
 921
 922	ntfs_debug("Entering for inode 0x%lx, attribute type 0x%x, page index "
 923			"0x%lx.", vi->i_ino, ni->type, page->index);
 924	BUG_ON(!NInoNonResident(ni));
 925	BUG_ON(!NInoMstProtected(ni));
 926	is_mft = (S_ISREG(vi->i_mode) && !vi->i_ino);
 927	/*
 928	 * NOTE: ntfs_write_mst_block() would be called for $MFTMirr if a page
 929	 * in its page cache were to be marked dirty.  However this should
 930	 * never happen with the current driver and considering we do not
 931	 * handle this case here we do want to BUG(), at least for now.
 932	 */
 933	BUG_ON(!(is_mft || S_ISDIR(vi->i_mode) ||
 934			(NInoAttr(ni) && ni->type == AT_INDEX_ALLOCATION)));
 935	bh_size = vol->sb->s_blocksize;
 936	bh_size_bits = vol->sb->s_blocksize_bits;
 937	max_bhs = PAGE_SIZE / bh_size;
 938	BUG_ON(!max_bhs);
 939	BUG_ON(max_bhs > MAX_BUF_PER_PAGE);
 940
 941	/* Were we called for sync purposes? */
 942	sync = (wbc->sync_mode == WB_SYNC_ALL);
 943
 944	/* Make sure we have mapped buffers. */
 945	bh = head = page_buffers(page);
 946	BUG_ON(!bh);
 947
 948	rec_size_bits = ni->itype.index.block_size_bits;
 949	BUG_ON(!(PAGE_SIZE >> rec_size_bits));
 950	bhs_per_rec = rec_size >> bh_size_bits;
 951	BUG_ON(!bhs_per_rec);
 952
 953	/* The first block in the page. */
 954	rec_block = block = (sector_t)page->index <<
 955			(PAGE_SHIFT - bh_size_bits);
 956
 957	/* The first out of bounds block for the data size. */
 958	dblock = (i_size_read(vi) + bh_size - 1) >> bh_size_bits;
 959
 960	rl = NULL;
 961	err = err2 = nr_bhs = nr_recs = nr_locked_nis = 0;
 962	page_is_dirty = rec_is_dirty = false;
 963	rec_start_bh = NULL;
 964	do {
 965		bool is_retry = false;
 966
 967		if (likely(block < rec_block)) {
 968			if (unlikely(block >= dblock)) {
 969				clear_buffer_dirty(bh);
 970				set_buffer_uptodate(bh);
 971				continue;
 972			}
 973			/*
 974			 * This block is not the first one in the record.  We
 975			 * ignore the buffer's dirty state because we could
 976			 * have raced with a parallel mark_ntfs_record_dirty().
 977			 */
 978			if (!rec_is_dirty)
 979				continue;
 980			if (unlikely(err2)) {
 981				if (err2 != -ENOMEM)
 982					clear_buffer_dirty(bh);
 983				continue;
 984			}
 985		} else /* if (block == rec_block) */ {
 986			BUG_ON(block > rec_block);
 987			/* This block is the first one in the record. */
 988			rec_block += bhs_per_rec;
 989			err2 = 0;
 990			if (unlikely(block >= dblock)) {
 991				clear_buffer_dirty(bh);
 992				continue;
 993			}
 994			if (!buffer_dirty(bh)) {
 995				/* Clean records are not written out. */
 996				rec_is_dirty = false;
 997				continue;
 998			}
 999			rec_is_dirty = true;
1000			rec_start_bh = bh;
1001		}
1002		/* Need to map the buffer if it is not mapped already. */
1003		if (unlikely(!buffer_mapped(bh))) {
1004			VCN vcn;
1005			LCN lcn;
1006			unsigned int vcn_ofs;
1007
1008			bh->b_bdev = vol->sb->s_bdev;
1009			/* Obtain the vcn and offset of the current block. */
1010			vcn = (VCN)block << bh_size_bits;
1011			vcn_ofs = vcn & vol->cluster_size_mask;
1012			vcn >>= vol->cluster_size_bits;
1013			if (!rl) {
1014lock_retry_remap:
1015				down_read(&ni->runlist.lock);
1016				rl = ni->runlist.rl;
1017			}
1018			if (likely(rl != NULL)) {
1019				/* Seek to element containing target vcn. */
1020				while (rl->length && rl[1].vcn <= vcn)
1021					rl++;
1022				lcn = ntfs_rl_vcn_to_lcn(rl, vcn);
1023			} else
1024				lcn = LCN_RL_NOT_MAPPED;
1025			/* Successful remap. */
1026			if (likely(lcn >= 0)) {
1027				/* Setup buffer head to correct block. */
1028				bh->b_blocknr = ((lcn <<
1029						vol->cluster_size_bits) +
1030						vcn_ofs) >> bh_size_bits;
1031				set_buffer_mapped(bh);
1032			} else {
1033				/*
1034				 * Remap failed.  Retry to map the runlist once
1035				 * unless we are working on $MFT which always
1036				 * has the whole of its runlist in memory.
1037				 */
1038				if (!is_mft && !is_retry &&
1039						lcn == LCN_RL_NOT_MAPPED) {
1040					is_retry = true;
1041					/*
1042					 * Attempt to map runlist, dropping
1043					 * lock for the duration.
1044					 */
1045					up_read(&ni->runlist.lock);
1046					err2 = ntfs_map_runlist(ni, vcn);
1047					if (likely(!err2))
1048						goto lock_retry_remap;
1049					if (err2 == -ENOMEM)
1050						page_is_dirty = true;
1051					lcn = err2;
1052				} else {
1053					err2 = -EIO;
1054					if (!rl)
1055						up_read(&ni->runlist.lock);
1056				}
1057				/* Hard error.  Abort writing this record. */
1058				if (!err || err == -ENOMEM)
1059					err = err2;
1060				bh->b_blocknr = -1;
1061				ntfs_error(vol->sb, "Cannot write ntfs record "
1062						"0x%llx (inode 0x%lx, "
1063						"attribute type 0x%x) because "
1064						"its location on disk could "
1065						"not be determined (error "
1066						"code %lli).",
1067						(long long)block <<
1068						bh_size_bits >>
1069						vol->mft_record_size_bits,
1070						ni->mft_no, ni->type,
1071						(long long)lcn);
1072				/*
1073				 * If this is not the first buffer, remove the
1074				 * buffers in this record from the list of
1075				 * buffers to write and clear their dirty bit
1076				 * if not error -ENOMEM.
1077				 */
1078				if (rec_start_bh != bh) {
1079					while (bhs[--nr_bhs] != rec_start_bh)
1080						;
1081					if (err2 != -ENOMEM) {
1082						do {
1083							clear_buffer_dirty(
1084								rec_start_bh);
1085						} while ((rec_start_bh =
1086								rec_start_bh->
1087								b_this_page) !=
1088								bh);
1089					}
1090				}
1091				continue;
1092			}
1093		}
1094		BUG_ON(!buffer_uptodate(bh));
1095		BUG_ON(nr_bhs >= max_bhs);
1096		bhs[nr_bhs++] = bh;
1097	} while (block++, (bh = bh->b_this_page) != head);
1098	if (unlikely(rl))
1099		up_read(&ni->runlist.lock);
1100	/* If there were no dirty buffers, we are done. */
1101	if (!nr_bhs)
1102		goto done;
1103	/* Map the page so we can access its contents. */
1104	kaddr = kmap(page);
1105	/* Clear the page uptodate flag whilst the mst fixups are applied. */
1106	BUG_ON(!PageUptodate(page));
1107	ClearPageUptodate(page);
1108	for (i = 0; i < nr_bhs; i++) {
1109		unsigned int ofs;
1110
1111		/* Skip buffers which are not at the beginning of records. */
1112		if (i % bhs_per_rec)
1113			continue;
1114		tbh = bhs[i];
1115		ofs = bh_offset(tbh);
1116		if (is_mft) {
1117			ntfs_inode *tni;
1118			unsigned long mft_no;
1119
1120			/* Get the mft record number. */
1121			mft_no = (((s64)page->index << PAGE_SHIFT) + ofs)
1122					>> rec_size_bits;
1123			/* Check whether to write this mft record. */
1124			tni = NULL;
1125			if (!ntfs_may_write_mft_record(vol, mft_no,
1126					(MFT_RECORD*)(kaddr + ofs), &tni)) {
1127				/*
1128				 * The record should not be written.  This
1129				 * means we need to redirty the page before
1130				 * returning.
1131				 */
1132				page_is_dirty = true;
1133				/*
1134				 * Remove the buffers in this mft record from
1135				 * the list of buffers to write.
1136				 */
1137				do {
1138					bhs[i] = NULL;
1139				} while (++i % bhs_per_rec);
1140				continue;
1141			}
1142			/*
1143			 * The record should be written.  If a locked ntfs
1144			 * inode was returned, add it to the array of locked
1145			 * ntfs inodes.
1146			 */
1147			if (tni)
1148				locked_nis[nr_locked_nis++] = tni;
1149		}
1150		/* Apply the mst protection fixups. */
1151		err2 = pre_write_mst_fixup((NTFS_RECORD*)(kaddr + ofs),
1152				rec_size);
1153		if (unlikely(err2)) {
1154			if (!err || err == -ENOMEM)
1155				err = -EIO;
1156			ntfs_error(vol->sb, "Failed to apply mst fixups "
1157					"(inode 0x%lx, attribute type 0x%x, "
1158					"page index 0x%lx, page offset 0x%x)!"
1159					"  Unmount and run chkdsk.", vi->i_ino,
1160					ni->type, page->index, ofs);
1161			/*
1162			 * Mark all the buffers in this record clean as we do
1163			 * not want to write corrupt data to disk.
1164			 */
1165			do {
1166				clear_buffer_dirty(bhs[i]);
1167				bhs[i] = NULL;
1168			} while (++i % bhs_per_rec);
1169			continue;
1170		}
1171		nr_recs++;
1172	}
1173	/* If no records are to be written out, we are done. */
1174	if (!nr_recs)
1175		goto unm_done;
1176	flush_dcache_page(page);
1177	/* Lock buffers and start synchronous write i/o on them. */
1178	for (i = 0; i < nr_bhs; i++) {
1179		tbh = bhs[i];
1180		if (!tbh)
1181			continue;
1182		if (!trylock_buffer(tbh))
1183			BUG();
1184		/* The buffer dirty state is now irrelevant, just clean it. */
1185		clear_buffer_dirty(tbh);
1186		BUG_ON(!buffer_uptodate(tbh));
1187		BUG_ON(!buffer_mapped(tbh));
1188		get_bh(tbh);
1189		tbh->b_end_io = end_buffer_write_sync;
1190		submit_bh(REQ_OP_WRITE, tbh);
1191	}
1192	/* Synchronize the mft mirror now if not @sync. */
1193	if (is_mft && !sync)
1194		goto do_mirror;
1195do_wait:
1196	/* Wait on i/o completion of buffers. */
1197	for (i = 0; i < nr_bhs; i++) {
1198		tbh = bhs[i];
1199		if (!tbh)
1200			continue;
1201		wait_on_buffer(tbh);
1202		if (unlikely(!buffer_uptodate(tbh))) {
1203			ntfs_error(vol->sb, "I/O error while writing ntfs "
1204					"record buffer (inode 0x%lx, "
1205					"attribute type 0x%x, page index "
1206					"0x%lx, page offset 0x%lx)!  Unmount "
1207					"and run chkdsk.", vi->i_ino, ni->type,
1208					page->index, bh_offset(tbh));
1209			if (!err || err == -ENOMEM)
1210				err = -EIO;
1211			/*
1212			 * Set the buffer uptodate so the page and buffer
1213			 * states do not become out of sync.
1214			 */
1215			set_buffer_uptodate(tbh);
1216		}
1217	}
1218	/* If @sync, now synchronize the mft mirror. */
1219	if (is_mft && sync) {
1220do_mirror:
1221		for (i = 0; i < nr_bhs; i++) {
1222			unsigned long mft_no;
1223			unsigned int ofs;
1224
1225			/*
1226			 * Skip buffers which are not at the beginning of
1227			 * records.
1228			 */
1229			if (i % bhs_per_rec)
1230				continue;
1231			tbh = bhs[i];
1232			/* Skip removed buffers (and hence records). */
1233			if (!tbh)
1234				continue;
1235			ofs = bh_offset(tbh);
1236			/* Get the mft record number. */
1237			mft_no = (((s64)page->index << PAGE_SHIFT) + ofs)
1238					>> rec_size_bits;
1239			if (mft_no < vol->mftmirr_size)
1240				ntfs_sync_mft_mirror(vol, mft_no,
1241						(MFT_RECORD*)(kaddr + ofs),
1242						sync);
1243		}
1244		if (!sync)
1245			goto do_wait;
1246	}
1247	/* Remove the mst protection fixups again. */
1248	for (i = 0; i < nr_bhs; i++) {
1249		if (!(i % bhs_per_rec)) {
1250			tbh = bhs[i];
1251			if (!tbh)
1252				continue;
1253			post_write_mst_fixup((NTFS_RECORD*)(kaddr +
1254					bh_offset(tbh)));
1255		}
1256	}
1257	flush_dcache_page(page);
1258unm_done:
1259	/* Unlock any locked inodes. */
1260	while (nr_locked_nis-- > 0) {
1261		ntfs_inode *tni, *base_tni;
1262		
1263		tni = locked_nis[nr_locked_nis];
1264		/* Get the base inode. */
1265		mutex_lock(&tni->extent_lock);
1266		if (tni->nr_extents >= 0)
1267			base_tni = tni;
1268		else {
1269			base_tni = tni->ext.base_ntfs_ino;
1270			BUG_ON(!base_tni);
1271		}
1272		mutex_unlock(&tni->extent_lock);
1273		ntfs_debug("Unlocking %s inode 0x%lx.",
1274				tni == base_tni ? "base" : "extent",
1275				tni->mft_no);
1276		mutex_unlock(&tni->mrec_lock);
1277		atomic_dec(&tni->count);
1278		iput(VFS_I(base_tni));
1279	}
1280	SetPageUptodate(page);
1281	kunmap(page);
1282done:
1283	if (unlikely(err && err != -ENOMEM)) {
1284		/*
1285		 * Set page error if there is only one ntfs record in the page.
1286		 * Otherwise we would loose per-record granularity.
1287		 */
1288		if (ni->itype.index.block_size == PAGE_SIZE)
1289			SetPageError(page);
1290		NVolSetErrors(vol);
1291	}
1292	if (page_is_dirty) {
1293		ntfs_debug("Page still contains one or more dirty ntfs "
1294				"records.  Redirtying the page starting at "
1295				"record 0x%lx.", page->index <<
1296				(PAGE_SHIFT - rec_size_bits));
1297		redirty_page_for_writepage(wbc, page);
1298		unlock_page(page);
1299	} else {
1300		/*
1301		 * Keep the VM happy.  This must be done otherwise the
1302		 * radix-tree tag PAGECACHE_TAG_DIRTY remains set even though
1303		 * the page is clean.
1304		 */
1305		BUG_ON(PageWriteback(page));
1306		set_page_writeback(page);
1307		unlock_page(page);
1308		end_page_writeback(page);
1309	}
1310	if (likely(!err))
1311		ntfs_debug("Done.");
1312	return err;
1313}
1314
1315/**
1316 * ntfs_writepage - write a @page to the backing store
1317 * @page:	page cache page to write out
1318 * @wbc:	writeback control structure
1319 *
1320 * This is called from the VM when it wants to have a dirty ntfs page cache
1321 * page cleaned.  The VM has already locked the page and marked it clean.
1322 *
1323 * For non-resident attributes, ntfs_writepage() writes the @page by calling
1324 * the ntfs version of the generic block_write_full_page() function,
1325 * ntfs_write_block(), which in turn if necessary creates and writes the
1326 * buffers associated with the page asynchronously.
1327 *
1328 * For resident attributes, OTOH, ntfs_writepage() writes the @page by copying
1329 * the data to the mft record (which at this stage is most likely in memory).
1330 * The mft record is then marked dirty and written out asynchronously via the
1331 * vfs inode dirty code path for the inode the mft record belongs to or via the
1332 * vm page dirty code path for the page the mft record is in.
1333 *
1334 * Based on ntfs_read_folio() and fs/buffer.c::block_write_full_page().
1335 *
1336 * Return 0 on success and -errno on error.
1337 */
1338static int ntfs_writepage(struct page *page, struct writeback_control *wbc)
1339{
1340	loff_t i_size;
1341	struct inode *vi = page->mapping->host;
1342	ntfs_inode *base_ni = NULL, *ni = NTFS_I(vi);
1343	char *addr;
1344	ntfs_attr_search_ctx *ctx = NULL;
1345	MFT_RECORD *m = NULL;
1346	u32 attr_len;
1347	int err;
1348
1349retry_writepage:
1350	BUG_ON(!PageLocked(page));
1351	i_size = i_size_read(vi);
1352	/* Is the page fully outside i_size? (truncate in progress) */
1353	if (unlikely(page->index >= (i_size + PAGE_SIZE - 1) >>
1354			PAGE_SHIFT)) {
1355		struct folio *folio = page_folio(page);
1356		/*
1357		 * The page may have dirty, unmapped buffers.  Make them
1358		 * freeable here, so the page does not leak.
1359		 */
1360		block_invalidate_folio(folio, 0, folio_size(folio));
1361		folio_unlock(folio);
1362		ntfs_debug("Write outside i_size - truncated?");
1363		return 0;
1364	}
1365	/*
1366	 * Only $DATA attributes can be encrypted and only unnamed $DATA
1367	 * attributes can be compressed.  Index root can have the flags set but
1368	 * this means to create compressed/encrypted files, not that the
1369	 * attribute is compressed/encrypted.  Note we need to check for
1370	 * AT_INDEX_ALLOCATION since this is the type of both directory and
1371	 * index inodes.
1372	 */
1373	if (ni->type != AT_INDEX_ALLOCATION) {
1374		/* If file is encrypted, deny access, just like NT4. */
1375		if (NInoEncrypted(ni)) {
1376			unlock_page(page);
1377			BUG_ON(ni->type != AT_DATA);
1378			ntfs_debug("Denying write access to encrypted file.");
1379			return -EACCES;
1380		}
1381		/* Compressed data streams are handled in compress.c. */
1382		if (NInoNonResident(ni) && NInoCompressed(ni)) {
1383			BUG_ON(ni->type != AT_DATA);
1384			BUG_ON(ni->name_len);
1385			// TODO: Implement and replace this with
1386			// return ntfs_write_compressed_block(page);
1387			unlock_page(page);
1388			ntfs_error(vi->i_sb, "Writing to compressed files is "
1389					"not supported yet.  Sorry.");
1390			return -EOPNOTSUPP;
1391		}
1392		// TODO: Implement and remove this check.
1393		if (NInoNonResident(ni) && NInoSparse(ni)) {
1394			unlock_page(page);
1395			ntfs_error(vi->i_sb, "Writing to sparse files is not "
1396					"supported yet.  Sorry.");
1397			return -EOPNOTSUPP;
1398		}
1399	}
1400	/* NInoNonResident() == NInoIndexAllocPresent() */
1401	if (NInoNonResident(ni)) {
1402		/* We have to zero every time due to mmap-at-end-of-file. */
1403		if (page->index >= (i_size >> PAGE_SHIFT)) {
1404			/* The page straddles i_size. */
1405			unsigned int ofs = i_size & ~PAGE_MASK;
1406			zero_user_segment(page, ofs, PAGE_SIZE);
1407		}
1408		/* Handle mst protected attributes. */
1409		if (NInoMstProtected(ni))
1410			return ntfs_write_mst_block(page, wbc);
1411		/* Normal, non-resident data stream. */
1412		return ntfs_write_block(page, wbc);
1413	}
1414	/*
1415	 * Attribute is resident, implying it is not compressed, encrypted, or
1416	 * mst protected.  This also means the attribute is smaller than an mft
1417	 * record and hence smaller than a page, so can simply return error on
1418	 * any pages with index above 0.  Note the attribute can actually be
1419	 * marked compressed but if it is resident the actual data is not
1420	 * compressed so we are ok to ignore the compressed flag here.
1421	 */
1422	BUG_ON(page_has_buffers(page));
1423	BUG_ON(!PageUptodate(page));
1424	if (unlikely(page->index > 0)) {
1425		ntfs_error(vi->i_sb, "BUG()! page->index (0x%lx) > 0.  "
1426				"Aborting write.", page->index);
1427		BUG_ON(PageWriteback(page));
1428		set_page_writeback(page);
1429		unlock_page(page);
1430		end_page_writeback(page);
1431		return -EIO;
1432	}
1433	if (!NInoAttr(ni))
1434		base_ni = ni;
1435	else
1436		base_ni = ni->ext.base_ntfs_ino;
1437	/* Map, pin, and lock the mft record. */
1438	m = map_mft_record(base_ni);
1439	if (IS_ERR(m)) {
1440		err = PTR_ERR(m);
1441		m = NULL;
1442		ctx = NULL;
1443		goto err_out;
1444	}
1445	/*
1446	 * If a parallel write made the attribute non-resident, drop the mft
1447	 * record and retry the writepage.
1448	 */
1449	if (unlikely(NInoNonResident(ni))) {
1450		unmap_mft_record(base_ni);
1451		goto retry_writepage;
1452	}
1453	ctx = ntfs_attr_get_search_ctx(base_ni, m);
1454	if (unlikely(!ctx)) {
1455		err = -ENOMEM;
1456		goto err_out;
1457	}
1458	err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len,
1459			CASE_SENSITIVE, 0, NULL, 0, ctx);
1460	if (unlikely(err))
1461		goto err_out;
1462	/*
1463	 * Keep the VM happy.  This must be done otherwise the radix-tree tag
1464	 * PAGECACHE_TAG_DIRTY remains set even though the page is clean.
1465	 */
1466	BUG_ON(PageWriteback(page));
1467	set_page_writeback(page);
1468	unlock_page(page);
1469	attr_len = le32_to_cpu(ctx->attr->data.resident.value_length);
1470	i_size = i_size_read(vi);
1471	if (unlikely(attr_len > i_size)) {
1472		/* Race with shrinking truncate or a failed truncate. */
1473		attr_len = i_size;
1474		/*
1475		 * If the truncate failed, fix it up now.  If a concurrent
1476		 * truncate, we do its job, so it does not have to do anything.
1477		 */
1478		err = ntfs_resident_attr_value_resize(ctx->mrec, ctx->attr,
1479				attr_len);
1480		/* Shrinking cannot fail. */
1481		BUG_ON(err);
1482	}
1483	addr = kmap_atomic(page);
1484	/* Copy the data from the page to the mft record. */
1485	memcpy((u8*)ctx->attr +
1486			le16_to_cpu(ctx->attr->data.resident.value_offset),
1487			addr, attr_len);
1488	/* Zero out of bounds area in the page cache page. */
1489	memset(addr + attr_len, 0, PAGE_SIZE - attr_len);
1490	kunmap_atomic(addr);
1491	flush_dcache_page(page);
1492	flush_dcache_mft_record_page(ctx->ntfs_ino);
1493	/* We are done with the page. */
1494	end_page_writeback(page);
1495	/* Finally, mark the mft record dirty, so it gets written back. */
1496	mark_mft_record_dirty(ctx->ntfs_ino);
1497	ntfs_attr_put_search_ctx(ctx);
1498	unmap_mft_record(base_ni);
1499	return 0;
1500err_out:
1501	if (err == -ENOMEM) {
1502		ntfs_warning(vi->i_sb, "Error allocating memory. Redirtying "
1503				"page so we try again later.");
1504		/*
1505		 * Put the page back on mapping->dirty_pages, but leave its
1506		 * buffers' dirty state as-is.
1507		 */
1508		redirty_page_for_writepage(wbc, page);
1509		err = 0;
1510	} else {
1511		ntfs_error(vi->i_sb, "Resident attribute write failed with "
1512				"error %i.", err);
1513		SetPageError(page);
1514		NVolSetErrors(ni->vol);
1515	}
1516	unlock_page(page);
1517	if (ctx)
1518		ntfs_attr_put_search_ctx(ctx);
1519	if (m)
1520		unmap_mft_record(base_ni);
1521	return err;
1522}
1523
1524#endif	/* NTFS_RW */
1525
1526/**
1527 * ntfs_bmap - map logical file block to physical device block
1528 * @mapping:	address space mapping to which the block to be mapped belongs
1529 * @block:	logical block to map to its physical device block
1530 *
1531 * For regular, non-resident files (i.e. not compressed and not encrypted), map
1532 * the logical @block belonging to the file described by the address space
1533 * mapping @mapping to its physical device block.
1534 *
1535 * The size of the block is equal to the @s_blocksize field of the super block
1536 * of the mounted file system which is guaranteed to be smaller than or equal
1537 * to the cluster size thus the block is guaranteed to fit entirely inside the
1538 * cluster which means we do not need to care how many contiguous bytes are
1539 * available after the beginning of the block.
1540 *
1541 * Return the physical device block if the mapping succeeded or 0 if the block
1542 * is sparse or there was an error.
1543 *
1544 * Note: This is a problem if someone tries to run bmap() on $Boot system file
1545 * as that really is in block zero but there is nothing we can do.  bmap() is
1546 * just broken in that respect (just like it cannot distinguish sparse from
1547 * not available or error).
1548 */
1549static sector_t ntfs_bmap(struct address_space *mapping, sector_t block)
1550{
1551	s64 ofs, size;
1552	loff_t i_size;
1553	LCN lcn;
1554	unsigned long blocksize, flags;
1555	ntfs_inode *ni = NTFS_I(mapping->host);
1556	ntfs_volume *vol = ni->vol;
1557	unsigned delta;
1558	unsigned char blocksize_bits, cluster_size_shift;
1559
1560	ntfs_debug("Entering for mft_no 0x%lx, logical block 0x%llx.",
1561			ni->mft_no, (unsigned long long)block);
1562	if (ni->type != AT_DATA || !NInoNonResident(ni) || NInoEncrypted(ni)) {
1563		ntfs_error(vol->sb, "BMAP does not make sense for %s "
1564				"attributes, returning 0.",
1565				(ni->type != AT_DATA) ? "non-data" :
1566				(!NInoNonResident(ni) ? "resident" :
1567				"encrypted"));
1568		return 0;
1569	}
1570	/* None of these can happen. */
1571	BUG_ON(NInoCompressed(ni));
1572	BUG_ON(NInoMstProtected(ni));
1573	blocksize = vol->sb->s_blocksize;
1574	blocksize_bits = vol->sb->s_blocksize_bits;
1575	ofs = (s64)block << blocksize_bits;
1576	read_lock_irqsave(&ni->size_lock, flags);
1577	size = ni->initialized_size;
1578	i_size = i_size_read(VFS_I(ni));
1579	read_unlock_irqrestore(&ni->size_lock, flags);
1580	/*
1581	 * If the offset is outside the initialized size or the block straddles
1582	 * the initialized size then pretend it is a hole unless the
1583	 * initialized size equals the file size.
1584	 */
1585	if (unlikely(ofs >= size || (ofs + blocksize > size && size < i_size)))
1586		goto hole;
1587	cluster_size_shift = vol->cluster_size_bits;
1588	down_read(&ni->runlist.lock);
1589	lcn = ntfs_attr_vcn_to_lcn_nolock(ni, ofs >> cluster_size_shift, false);
1590	up_read(&ni->runlist.lock);
1591	if (unlikely(lcn < LCN_HOLE)) {
1592		/*
1593		 * Step down to an integer to avoid gcc doing a long long
1594		 * comparision in the switch when we know @lcn is between
1595		 * LCN_HOLE and LCN_EIO (i.e. -1 to -5).
1596		 *
1597		 * Otherwise older gcc (at least on some architectures) will
1598		 * try to use __cmpdi2() which is of course not available in
1599		 * the kernel.
1600		 */
1601		switch ((int)lcn) {
1602		case LCN_ENOENT:
1603			/*
1604			 * If the offset is out of bounds then pretend it is a
1605			 * hole.
1606			 */
1607			goto hole;
1608		case LCN_ENOMEM:
1609			ntfs_error(vol->sb, "Not enough memory to complete "
1610					"mapping for inode 0x%lx.  "
1611					"Returning 0.", ni->mft_no);
1612			break;
1613		default:
1614			ntfs_error(vol->sb, "Failed to complete mapping for "
1615					"inode 0x%lx.  Run chkdsk.  "
1616					"Returning 0.", ni->mft_no);
1617			break;
1618		}
1619		return 0;
1620	}
1621	if (lcn < 0) {
1622		/* It is a hole. */
1623hole:
1624		ntfs_debug("Done (returning hole).");
1625		return 0;
1626	}
1627	/*
1628	 * The block is really allocated and fullfils all our criteria.
1629	 * Convert the cluster to units of block size and return the result.
1630	 */
1631	delta = ofs & vol->cluster_size_mask;
1632	if (unlikely(sizeof(block) < sizeof(lcn))) {
1633		block = lcn = ((lcn << cluster_size_shift) + delta) >>
1634				blocksize_bits;
1635		/* If the block number was truncated return 0. */
1636		if (unlikely(block != lcn)) {
1637			ntfs_error(vol->sb, "Physical block 0x%llx is too "
1638					"large to be returned, returning 0.",
1639					(long long)lcn);
1640			return 0;
1641		}
1642	} else
1643		block = ((lcn << cluster_size_shift) + delta) >>
1644				blocksize_bits;
1645	ntfs_debug("Done (returning block 0x%llx).", (unsigned long long)lcn);
1646	return block;
1647}
1648
1649/**
1650 * ntfs_normal_aops - address space operations for normal inodes and attributes
1651 *
1652 * Note these are not used for compressed or mst protected inodes and
1653 * attributes.
1654 */
1655const struct address_space_operations ntfs_normal_aops = {
1656	.read_folio	= ntfs_read_folio,
1657#ifdef NTFS_RW
1658	.writepage	= ntfs_writepage,
1659	.dirty_folio	= block_dirty_folio,
1660#endif /* NTFS_RW */
1661	.bmap		= ntfs_bmap,
1662	.migrate_folio	= buffer_migrate_folio,
1663	.is_partially_uptodate = block_is_partially_uptodate,
1664	.error_remove_page = generic_error_remove_page,
1665};
1666
1667/**
1668 * ntfs_compressed_aops - address space operations for compressed inodes
1669 */
1670const struct address_space_operations ntfs_compressed_aops = {
1671	.read_folio	= ntfs_read_folio,
1672#ifdef NTFS_RW
1673	.writepage	= ntfs_writepage,
1674	.dirty_folio	= block_dirty_folio,
1675#endif /* NTFS_RW */
1676	.migrate_folio	= buffer_migrate_folio,
1677	.is_partially_uptodate = block_is_partially_uptodate,
 
1678	.error_remove_page = generic_error_remove_page,
1679};
1680
1681/**
1682 * ntfs_mst_aops - general address space operations for mst protecteed inodes
1683 *		   and attributes
1684 */
1685const struct address_space_operations ntfs_mst_aops = {
1686	.read_folio	= ntfs_read_folio,	/* Fill page with data. */
1687#ifdef NTFS_RW
1688	.writepage	= ntfs_writepage,	/* Write dirty page to disk. */
1689	.dirty_folio	= filemap_dirty_folio,
 
 
1690#endif /* NTFS_RW */
1691	.migrate_folio	= buffer_migrate_folio,
1692	.is_partially_uptodate	= block_is_partially_uptodate,
 
1693	.error_remove_page = generic_error_remove_page,
1694};
1695
1696#ifdef NTFS_RW
1697
1698/**
1699 * mark_ntfs_record_dirty - mark an ntfs record dirty
1700 * @page:	page containing the ntfs record to mark dirty
1701 * @ofs:	byte offset within @page at which the ntfs record begins
1702 *
1703 * Set the buffers and the page in which the ntfs record is located dirty.
1704 *
1705 * The latter also marks the vfs inode the ntfs record belongs to dirty
1706 * (I_DIRTY_PAGES only).
1707 *
1708 * If the page does not have buffers, we create them and set them uptodate.
1709 * The page may not be locked which is why we need to handle the buffers under
1710 * the mapping->private_lock.  Once the buffers are marked dirty we no longer
1711 * need the lock since try_to_free_buffers() does not free dirty buffers.
1712 */
1713void mark_ntfs_record_dirty(struct page *page, const unsigned int ofs) {
1714	struct address_space *mapping = page->mapping;
1715	ntfs_inode *ni = NTFS_I(mapping->host);
1716	struct buffer_head *bh, *head, *buffers_to_free = NULL;
1717	unsigned int end, bh_size, bh_ofs;
1718
1719	BUG_ON(!PageUptodate(page));
1720	end = ofs + ni->itype.index.block_size;
1721	bh_size = VFS_I(ni)->i_sb->s_blocksize;
1722	spin_lock(&mapping->private_lock);
1723	if (unlikely(!page_has_buffers(page))) {
1724		spin_unlock(&mapping->private_lock);
1725		bh = head = alloc_page_buffers(page, bh_size, true);
1726		spin_lock(&mapping->private_lock);
1727		if (likely(!page_has_buffers(page))) {
1728			struct buffer_head *tail;
1729
1730			do {
1731				set_buffer_uptodate(bh);
1732				tail = bh;
1733				bh = bh->b_this_page;
1734			} while (bh);
1735			tail->b_this_page = head;
1736			attach_page_private(page, head);
1737		} else
1738			buffers_to_free = bh;
1739	}
1740	bh = head = page_buffers(page);
1741	BUG_ON(!bh);
1742	do {
1743		bh_ofs = bh_offset(bh);
1744		if (bh_ofs + bh_size <= ofs)
1745			continue;
1746		if (unlikely(bh_ofs >= end))
1747			break;
1748		set_buffer_dirty(bh);
1749	} while ((bh = bh->b_this_page) != head);
1750	spin_unlock(&mapping->private_lock);
1751	filemap_dirty_folio(mapping, page_folio(page));
1752	if (unlikely(buffers_to_free)) {
1753		do {
1754			bh = buffers_to_free->b_this_page;
1755			free_buffer_head(buffers_to_free);
1756			buffers_to_free = bh;
1757		} while (buffers_to_free);
1758	}
1759}
1760
1761#endif /* NTFS_RW */