compress.c

winNT技术操作系统,国外开放的原代码和LIUX一样

/**
 * compress.c - NTFS kernel compressed attributes handling.
 *		Part of the Linux-NTFS project.
 *
 * Copyright (c) 2001-2003 Anton Altaparmakov
 * Copyright (c) 2002 Richard Russon
 *
 * 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/fs.h>
#include <linux/buffer_head.h>

#include "ntfs.h"

/**
 * ntfs_compression_constants - enum of constants used in the compression code
 */
typedef enum {
	/* Token types and access mask. */
	NTFS_SYMBOL_TOKEN	=	0,
	NTFS_PHRASE_TOKEN	=	1,
	NTFS_TOKEN_MASK		=	1,

	/* Compression sub-block constants. */
	NTFS_SB_SIZE_MASK	=	0x0fff,
	NTFS_SB_SIZE		=	0x1000,
	NTFS_SB_IS_COMPRESSED	=	0x8000,

	/*
	 * The maximum compression block size is by definition 16 * the cluster
	 * size, with the maximum supported cluster size being 4kiB. Thus the
	 * maximum compression buffer size is 64kiB, so we use this when
	 * initializing the compression buffer.
	 */
	NTFS_MAX_CB_SIZE	= 64 * 1024,
} ntfs_compression_constants;

/**
 * ntfs_compression_buffer - one buffer for the decompression engine
 */
static u8 *ntfs_compression_buffer = NULL;

/**
 * ntfs_cb_lock - spinlock which protects ntfs_compression_buffer
 */
static spinlock_t ntfs_cb_lock = SPIN_LOCK_UNLOCKED;

/**
 * allocate_compression_buffers - allocate the decompression buffers
 *
 * Caller has to hold the ntfs_lock semaphore.
 *
 * Return 0 on success or -ENOMEM if the allocations failed.
 */
int allocate_compression_buffers(void)
{
	BUG_ON(ntfs_compression_buffer);

	ntfs_compression_buffer = vmalloc(NTFS_MAX_CB_SIZE);
	if (!ntfs_compression_buffer)
		return -ENOMEM;
	return 0;
}

/**
 * free_compression_buffers - free the decompression buffers
 *
 * Caller has to hold the ntfs_lock semaphore.
 */
void free_compression_buffers(void)
{
	BUG_ON(!ntfs_compression_buffer);
	vfree(ntfs_compression_buffer);
	ntfs_compression_buffer = NULL;
}

/**
 * zero_partial_compressed_page - zero out of bounds compressed page region
 */
static void zero_partial_compressed_page(ntfs_inode *ni, struct page *page)
{
	u8 *kp = page_address(page);
	unsigned int kp_ofs;

	ntfs_debug("Zeroing page region outside initialized size.");
	if (((s64)page->index << PAGE_CACHE_SHIFT) >= ni->initialized_size) {
		/*
		 * FIXME: Using clear_page() will become wrong when we get
		 * PAGE_CACHE_SIZE != PAGE_SIZE but for now there is no problem.
		 */
		clear_page(kp);
		return;
	}
	kp_ofs = ni->initialized_size & ~PAGE_CACHE_MASK;
	memset(kp + kp_ofs, 0, PAGE_CACHE_SIZE - kp_ofs);
	return;
}

/**
 * handle_bounds_compressed_page - test for&handle out of bounds compressed page
 */
static inline void handle_bounds_compressed_page(ntfs_inode *ni,
		struct page *page)
{
	if ((page->index >= (ni->initialized_size >> PAGE_CACHE_SHIFT)) &&
			(ni->initialized_size < VFS_I(ni)->i_size))
		zero_partial_compressed_page(ni, page);
	return;
}

/**
 * ntfs_decompress - decompress a compression block into an array of pages
 * @dest_pages:		destination array of pages
 * @dest_index:		current index into @dest_pages (IN/OUT)
 * @dest_ofs:		current offset within @dest_pages[@dest_index] (IN/OUT)
 * @dest_max_index:	maximum index into @dest_pages (IN)
 * @dest_max_ofs:	maximum offset within @dest_pages[@dest_max_index] (IN)
 * @xpage:		the target page (-1 if none) (IN)
 * @xpage_done:		set to 1 if xpage was completed successfully (IN/OUT)
 * @cb_start:		compression block to decompress (IN)
 * @cb_size:		size of compression block @cb_start in bytes (IN)
 *
 * The caller must have disabled preemption. ntfs_decompress() reenables it when
 * the critical section is finished.
 *
 * This decompresses the compression block @cb_start into the array of
 * destination pages @dest_pages starting at index @dest_index into @dest_pages
 * and at offset @dest_pos into the page @dest_pages[@dest_index].
 *
 * When the page @dest_pages[@xpage] is completed, @xpage_done is set to 1.
 * If xpage is -1 or @xpage has not been completed, @xpage_done is not modified.
 *
 * @cb_start is a pointer to the compression block which needs decompressing
 * and @cb_size is the size of @cb_start in bytes (8-64kiB).
 *
 * Return 0 if success or -EOVERFLOW on error in the compressed stream.
 * @xpage_done indicates whether the target page (@dest_pages[@xpage]) was
 * completed during the decompression of the compression block (@cb_start).
 *
 * Warning: This function *REQUIRES* PAGE_CACHE_SIZE >= 4096 or it will blow up
 * unpredicatbly! You have been warned!
 *
 * Note to hackers: This function may not sleep until it has finished accessing
 * the compression block @cb_start as it is a per-CPU buffer.
 */
static int ntfs_decompress(struct page *dest_pages[], int *dest_index,
		int *dest_ofs, const int dest_max_index, const int dest_max_ofs,
		const int xpage, char *xpage_done, u8 *const cb_start,
		const u32 cb_size)
{
	/*
	 * Pointers into the compressed data, i.e. the compression block (cb),
	 * and the therein contained sub-blocks (sb).
	 */
	u8 *cb_end = cb_start + cb_size; /* End of cb. */
	u8 *cb = cb_start;	/* Current position in cb. */
	u8 *cb_sb_start = cb;	/* Beginning of the current sb in the cb. */
	u8 *cb_sb_end;		/* End of current sb / beginning of next sb. */

	/* Variables for uncompressed data / destination. */
	struct page *dp;	/* Current destination page being worked on. */
	u8 *dp_addr;		/* Current pointer into dp. */
	u8 *dp_sb_start;	/* Start of current sub-block in dp. */
	u8 *dp_sb_end;		/* End of current sb in dp (dp_sb_start +
				   NTFS_SB_SIZE). */
	u16 do_sb_start;	/* @dest_ofs when starting this sub-block. */
	u16 do_sb_end;		/* @dest_ofs of end of this sb (do_sb_start +
				   NTFS_SB_SIZE). */

	/* Variables for tag and token parsing. */
	u8 tag;			/* Current tag. */
	int token;		/* Loop counter for the eight tokens in tag. */

	/* Need this because we can't sleep, so need two stages. */
	int completed_pages[dest_max_index - *dest_index + 1];
	int nr_completed_pages = 0;

	/* Default error code. */
	int err = -EOVERFLOW;

	ntfs_debug("Entering, cb_size = 0x%x.", cb_size);
do_next_sb:
	ntfs_debug("Beginning sub-block at offset = 0x%x in the cb.",
			cb - cb_start);

	/* Have we reached the end of the compression block? */
	if (cb == cb_end || !le16_to_cpup((u16*)cb)) {
		int i;

		ntfs_debug("Completed. Returning success (0).");
		err = 0;
return_error:
		/* We can sleep from now on, so we drop lock. */
		spin_unlock(&ntfs_cb_lock);
		/* Second stage: finalize completed pages. */
		if (nr_completed_pages > 0) {
			struct page *page = dest_pages[completed_pages[0]];
			ntfs_inode *ni = NTFS_I(page->mapping->host);

			for (i = 0; i < nr_completed_pages; i++) {
				int di = completed_pages[i];

				dp = dest_pages[di];
				/*
				 * If we are outside the initialized size, zero
				 * the out of bounds page range.
				 */
				handle_bounds_compressed_page(ni, dp);
				flush_dcache_page(dp);
				kunmap(dp);
				SetPageUptodate(dp);
				unlock_page(dp);
				if (di == xpage)
					*xpage_done = 1;
				else
					page_cache_release(dp);
				dest_pages[di] = NULL;
			}
		}
		return err;
	}

	/* Setup offsets for the current sub-block destination. */
	do_sb_start = *dest_ofs;
	do_sb_end = do_sb_start + NTFS_SB_SIZE;

	/* Check that we are still within allowed boundaries. */
	if (*dest_index == dest_max_index && do_sb_end > dest_max_ofs)
		goto return_overflow;

	/* Does the minimum size of a compressed sb overflow valid range? */
	if (cb + 6 > cb_end)
		goto return_overflow;

	/* Setup the current sub-block source pointers and validate range. */
	cb_sb_start = cb;
	cb_sb_end = cb_sb_start + (le16_to_cpup((u16*)cb) & NTFS_SB_SIZE_MASK)
			+ 3;
	if (cb_sb_end > cb_end)
		goto return_overflow;

	/* Get the current destination page. */
	dp = dest_pages[*dest_index];
	if (!dp) {
		/* No page present. Skip decompression of this sub-block. */
		cb = cb_sb_end;

		/* Advance destination position to next sub-block. */
		*dest_ofs = (*dest_ofs + NTFS_SB_SIZE) & ~PAGE_CACHE_MASK;
		if (!*dest_ofs && (++*dest_index > dest_max_index))
			goto return_overflow;
		goto do_next_sb;
	}

	/* We have a valid destination page. Setup the destination pointers. */
	dp_addr = (u8*)page_address(dp) + do_sb_start;

	/* Now, we are ready to process the current sub-block (sb). */
	if (!(le16_to_cpup((u16*)cb) & NTFS_SB_IS_COMPRESSED)) {
		ntfs_debug("Found uncompressed sub-block.");
		/* This sb is not compressed, just copy it into destination. */

		/* Advance source position to first data byte. */
		cb += 2;

		/* An uncompressed sb must be full size. */
		if (cb_sb_end - cb != NTFS_SB_SIZE)
			goto return_overflow;

		/* Copy the block and advance the source position. */
		memcpy(dp_addr, cb, NTFS_SB_SIZE);
		cb += NTFS_SB_SIZE;

		/* Advance destination position to next sub-block. */
		*dest_ofs += NTFS_SB_SIZE;
		if (!(*dest_ofs &= ~PAGE_CACHE_MASK)) {
finalize_page:
			/*
			 * First stage: add current page index to array of
			 * completed pages.
			 */
			completed_pages[nr_completed_pages++] = *dest_index;
			if (++*dest_index > dest_max_index)
				goto return_overflow;
		}
		goto do_next_sb;
	}
	ntfs_debug("Found compressed sub-block.");
	/* This sb is compressed, decompress it into destination. */

	/* Setup destination pointers. */
	dp_sb_start = dp_addr;
	dp_sb_end = dp_sb_start + NTFS_SB_SIZE;

	/* Forward to the first tag in the sub-block. */
	cb += 2;
do_next_tag:
	if (cb == cb_sb_end) {
		/* Check if the decompressed sub-block was not full-length. */
		if (dp_addr < dp_sb_end) {
			int nr_bytes = do_sb_end - *dest_ofs;

			ntfs_debug("Filling incomplete sub-block with "
					"zeroes.");
			/* Zero remainder and update destination position. */
			memset(dp_addr, 0, nr_bytes);
			*dest_ofs += nr_bytes;
		}
		/* We have finished the current sub-block. */
		if (!(*dest_ofs &= ~PAGE_CACHE_MASK))
			goto finalize_page;
		goto do_next_sb;
	}

	/* Check we are still in range. */
	if (cb > cb_sb_end || dp_addr > dp_sb_end)
		goto return_overflow;

	/* Get the next tag and advance to first token. */
	tag = *cb++;

	/* Parse the eight tokens described by the tag. */
	for (token = 0; token < 8; token++, tag >>= 1) {
		u16 lg, pt, length, max_non_overlap;
		register u16 i;
		u8 *dp_back_addr;

		/* Check if we are done / still in range. */
		if (cb >= cb_sb_end || dp_addr > dp_sb_end)
			break;

		/* Determine token type and parse appropriately.*/
		if ((tag & NTFS_TOKEN_MASK) == NTFS_SYMBOL_TOKEN) {
			/*
			 * We have a symbol token, copy the symbol across, and
			 * advance the source and destination positions.
			 */
			*dp_addr++ = *cb++;
			++*dest_ofs;

			/* Continue with the next token. */
			continue;
		}

		/*
		 * We have a phrase token. Make sure it is not the first tag in
		 * the sb as this is illegal and would confuse the code below.
		 */
		if (dp_addr == dp_sb_start)
			goto return_overflow;

		/*
		 * Determine the number of bytes to go back (p) and the number
		 * of bytes to copy (l). We use an optimized algorithm in which
		 * we first calculate log2(current destination position in sb),
		 * which allows determination of l and p in O(1) rather than
		 * O(n). We just need an arch-optimized log2() function now.
		 */
		lg = 0;
		for (i = *dest_ofs - do_sb_start - 1; i >= 0x10; i >>= 1)
			lg++;

		/* Get the phrase token into i. */
		pt = le16_to_cpup((u16*)cb);

		/*
		 * Calculate starting position of the byte sequence in
		 * the destination using the fact that p = (pt >> (12 - lg)) + 1
		 * and make sure we don't go too far back.
		 */
		dp_back_addr = dp_addr - (pt >> (12 - lg)) - 1;
		if (dp_back_addr < dp_sb_start)
			goto return_overflow;

		/* Now calculate the length of the byte sequence. */
		length = (pt & (0xfff >> lg)) + 3;

		/* Advance destination position and verify it is in range. */
		*dest_ofs += length;
		if (*dest_ofs > do_sb_end)
			goto return_overflow;

		/* The number of non-overlapping bytes. */
		max_non_overlap = dp_addr - dp_back_addr;

		if (length <= max_non_overlap) {
			/* The byte sequence doesn't overlap, just copy it. */
			memcpy(dp_addr, dp_back_addr, length);

			/* Advance destination pointer. */
			dp_addr += length;
		} else {
			/*
			 * The byte sequence does overlap, copy non-overlapping
			 * part and then do a slow byte by byte copy for the
			 * overlapping part. Also, advance the destination
			 * pointer.
			 */
			memcpy(dp_addr, dp_back_addr, max_non_overlap);
			dp_addr += max_non_overlap;
			dp_back_addr += max_non_overlap;
			length -= max_non_overlap;
			while (length--)
				*dp_addr++ = *dp_back_addr++;
		}

		/* Advance source position and continue with the next token. */
		cb += 2;
	}

	/* No tokens left in the current tag. Continue with the next tag. */
	goto do_next_tag;

return_overflow:
	ntfs_error(NULL, "Failed. Returning -EOVERFLOW.\n");
	goto return_error;
}

/**
 * ntfs_read_compressed_block - read a compressed block into the page cache
 * @page:	locked page in the compression block(s) we need to read
 *
 * When we are called the page has already been verified to be locked and the
 * attribute is known to be non-resident, not encrypted, but compressed.
 *
 * 1. Determine which compression block(s) @page is in.
 * 2. Get hold of all pages corresponding to this/these compression block(s).
 * 3. Read the (first) compression block.
 * 4. Decompress it into the corresponding pages.
 * 5. Throw the compressed data away and proceed to 3. for the next compression
 *    block or return success if no more compression blocks left.
 *
 * Warning: We have to be careful what we do about existing pages. They might
 * have been written to so that we would lose data if we were to just overwrite
 * them with the out-of-date uncompressed data.
 *
 * FIXME: For PAGE_CACHE_SIZE > cb_size we are not doing the Right Thing(TM) at
 * the end of the file I think. We need to detect this case and zero the out
 * of bounds remainder of the page in question and mark it as handled. At the
 * moment we would just return -EIO on such a page. This bug will only become
 * apparent if pages are above 8kiB and the NTFS volume only uses 512 byte
 * clusters so is probably not going to be seen by anyone. Still this should
 * be fixed. (AIA)
 *
 * FIXME: Again for PAGE_CACHE_SIZE > cb_size we are screwing up both in
 * handling sparse and compressed cbs. (AIA)
 *
 * FIXME: At the moment we don't do any zeroing out in the case that
 * initialized_size is less than data_size. This should be safe because of the
 * nature of the compression algorithm used. Just in case we check and output
 * an error message in read inode if the two sizes are not equal for a
 * compressed file. (AIA)
 */
int ntfs_read_compressed_block(struct page *page)
{
	struct address_space *mapping = page->mapping;
	ntfs_inode *ni = NTFS_I(mapping->host);
	ntfs_volume *vol = ni->vol;
	struct super_block *sb = vol->sb;