aops.c

来自「Linux Kernel 2.6.9 for OMAP1710」· C语言 代码 · 共 2,031 行 · 第 1/4 页

/**
 * aops.c - NTFS kernel address space operations and page cache handling.
 *	    Part of the Linux-NTFS project.
 *
 * Copyright (c) 2001-2004 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/errno.h>
#include <linux/mm.h>
#include <linux/pagemap.h>
#include <linux/swap.h>
#include <linux/buffer_head.h>

#include "ntfs.h"

/**
 * ntfs_end_buffer_async_read - async io completion for reading attributes
 * @bh:		buffer head on which io is completed
 * @uptodate:	whether @bh is now uptodate or not
 *
 * Asynchronous I/O completion handler for reading pages belonging to the
 * attribute address space of an inode. The inodes can either be files or
 * directories or they can be fake inodes describing some attribute.
 *
 * If NInoMstProtected(), perform the post read mst fixups when all IO on the
 * page has been completed and mark the page uptodate or set the error bit on
 * the page. To determine the size of the records that need fixing up, we cheat
 * a little bit by setting the index_block_size in ntfs_inode to the ntfs
 * record size, and index_block_size_bits, to the log(base 2) of the ntfs
 * record size.
 */
static void ntfs_end_buffer_async_read(struct buffer_head *bh, int uptodate)
{
	static spinlock_t page_uptodate_lock = SPIN_LOCK_UNLOCKED;
	unsigned long flags;
	struct buffer_head *tmp;
	struct page *page;
	ntfs_inode *ni;
	int page_uptodate = 1;

	page = bh->b_page;
	ni = NTFS_I(page->mapping->host);

	if (likely(uptodate)) {
		s64 file_ofs;

		set_buffer_uptodate(bh);

		file_ofs = ((s64)page->index << PAGE_CACHE_SHIFT) +
				bh_offset(bh);
		/* Check for the current buffer head overflowing. */
		if (file_ofs + bh->b_size > ni->initialized_size) {
			char *addr;
			int ofs = 0;

			if (file_ofs < ni->initialized_size)
				ofs = ni->initialized_size - file_ofs;
			addr = kmap_atomic(page, KM_BIO_SRC_IRQ);
			memset(addr + bh_offset(bh) + ofs, 0, bh->b_size - ofs);
			flush_dcache_page(page);
			kunmap_atomic(addr, KM_BIO_SRC_IRQ);
		}
	} else {
		clear_buffer_uptodate(bh);
		ntfs_error(ni->vol->sb, "Buffer I/O error, logical block %llu.",
				(unsigned long long)bh->b_blocknr);
		SetPageError(page);
	}

	spin_lock_irqsave(&page_uptodate_lock, flags);
	clear_buffer_async_read(bh);
	unlock_buffer(bh);
	tmp = bh;
	do {
		if (!buffer_uptodate(tmp))
			page_uptodate = 0;
		if (buffer_async_read(tmp)) {
			if (likely(buffer_locked(tmp)))
				goto still_busy;
			/* Async buffers must be locked. */
			BUG();
		}
		tmp = tmp->b_this_page;
	} while (tmp != bh);
	spin_unlock_irqrestore(&page_uptodate_lock, flags);
	/*
	 * If none of the buffers had errors then we can set the page uptodate,
	 * but we first have to perform the post read mst fixups, if the
	 * attribute is mst protected, i.e. if NInoMstProteced(ni) is true.
	 */
	if (!NInoMstProtected(ni)) {
		if (likely(page_uptodate && !PageError(page)))
			SetPageUptodate(page);
	} else {
		char *addr;
		unsigned int i, recs, nr_err;
		u32 rec_size;

		rec_size = ni->itype.index.block_size;
		recs = PAGE_CACHE_SIZE / rec_size;
		addr = kmap_atomic(page, KM_BIO_SRC_IRQ);
		for (i = nr_err = 0; i < recs; i++) {
			if (likely(!post_read_mst_fixup((NTFS_RECORD*)(addr +
					i * rec_size), rec_size)))
				continue;
			nr_err++;
			ntfs_error(ni->vol->sb, "post_read_mst_fixup() failed, "
					"corrupt %s record 0x%llx. Run chkdsk.",
					ni->mft_no ? "index" : "mft",
					(unsigned long long)(((s64)page->index
					<< PAGE_CACHE_SHIFT >>
					ni->itype.index.block_size_bits) + i));
		}
		flush_dcache_page(page);
		kunmap_atomic(addr, KM_BIO_SRC_IRQ);
		if (likely(!PageError(page))) {
			if (likely(!nr_err && recs)) {
				if (likely(page_uptodate))
					SetPageUptodate(page);
			} else {
				ntfs_error(ni->vol->sb, "Setting page error, "
						"index 0x%lx.", page->index);
				SetPageError(page);
			}
		}
	}
	unlock_page(page);
	return;
still_busy:
	spin_unlock_irqrestore(&page_uptodate_lock, flags);
	return;
}

/**
 * ntfs_read_block - fill a @page of an address space with data
 * @page:	page cache page to fill with data
 *
 * Fill the page @page of the address space belonging to the @page->host inode.
 * We read each buffer asynchronously and when all buffers are read in, our io
 * completion handler ntfs_end_buffer_read_async(), if required, automatically
 * applies the mst fixups to the page before finally marking it uptodate and
 * unlocking it.
 *
 * We only enforce allocated_size limit because i_size is checked for in
 * generic_file_read().
 *
 * Return 0 on success and -errno on error.
 *
 * Contains an adapted version of fs/buffer.c::block_read_full_page().
 */
static int ntfs_read_block(struct page *page)
{
	VCN vcn;
	LCN lcn;
	ntfs_inode *ni;
	ntfs_volume *vol;
	runlist_element *rl;
	struct buffer_head *bh, *head, *arr[MAX_BUF_PER_PAGE];
	sector_t iblock, lblock, zblock;
	unsigned int blocksize, vcn_ofs;
	int i, nr;
	unsigned char blocksize_bits;

	ni = NTFS_I(page->mapping->host);
	vol = ni->vol;

	blocksize_bits = VFS_I(ni)->i_blkbits;
	blocksize = 1 << blocksize_bits;

	if (!page_has_buffers(page))
		create_empty_buffers(page, blocksize, 0);
	bh = head = page_buffers(page);
	if (unlikely(!bh)) {
		unlock_page(page);
		return -ENOMEM;
	}

	iblock = (s64)page->index << (PAGE_CACHE_SHIFT - blocksize_bits);
	lblock = (ni->allocated_size + blocksize - 1) >> blocksize_bits;
	zblock = (ni->initialized_size + blocksize - 1) >> blocksize_bits;

#ifdef DEBUG
	if (unlikely(!ni->runlist.rl && !ni->mft_no && !NInoAttr(ni)))
		panic("NTFS: $MFT/$DATA runlist has been unmapped! This is a "
				"very serious bug! Cannot continue...");
#endif

	/* Loop through all the buffers in the page. */
	rl = NULL;
	nr = i = 0;
	do {
		u8 *kaddr;

		if (unlikely(buffer_uptodate(bh)))
			continue;
		if (unlikely(buffer_mapped(bh))) {
			arr[nr++] = bh;
			continue;
		}
		bh->b_bdev = vol->sb->s_bdev;
		/* Is the block within the allowed limits? */
		if (iblock < lblock) {
			BOOL is_retry = FALSE;

			/* Convert iblock into corresponding vcn and offset. */
			vcn = (VCN)iblock << blocksize_bits >>
					vol->cluster_size_bits;
			vcn_ofs = ((VCN)iblock << blocksize_bits) &
					vol->cluster_size_mask;
			if (!rl) {
lock_retry_remap:
				down_read(&ni->runlist.lock);
				rl = ni->runlist.rl;
			}
			if (likely(rl != NULL)) {
				/* Seek to element containing target vcn. */
				while (rl->length && rl[1].vcn <= vcn)
					rl++;
				lcn = ntfs_vcn_to_lcn(rl, vcn);
			} else
				lcn = (LCN)LCN_RL_NOT_MAPPED;
			/* Successful remap. */
			if (lcn >= 0) {
				/* Setup buffer head to correct block. */
				bh->b_blocknr = ((lcn << vol->cluster_size_bits)
						+ vcn_ofs) >> blocksize_bits;
				set_buffer_mapped(bh);
				/* Only read initialized data blocks. */
				if (iblock < zblock) {
					arr[nr++] = bh;
					continue;
				}
				/* Fully non-initialized data block, zero it. */
				goto handle_zblock;
			}
			/* It is a hole, need to zero it. */
			if (lcn == LCN_HOLE)
				goto handle_hole;
			/* If first try and runlist unmapped, map and retry. */
			if (!is_retry && lcn == LCN_RL_NOT_MAPPED) {
				is_retry = TRUE;
				/*
				 * Attempt to map runlist, dropping lock for
				 * the duration.
				 */
				up_read(&ni->runlist.lock);
				if (!ntfs_map_runlist(ni, vcn))
					goto lock_retry_remap;
				rl = NULL;
			}
			/* Hard error, zero out region. */
			SetPageError(page);
			ntfs_error(vol->sb, "ntfs_vcn_to_lcn(vcn = 0x%llx) "
					"failed with error code 0x%llx%s.",
					(unsigned long long)vcn,
					(unsigned long long)-lcn,
					is_retry ? " even after retrying" : "");
			// FIXME: Depending on vol->on_errors, do something.
		}
		/*
		 * Either iblock was outside lblock limits or ntfs_vcn_to_lcn()
		 * returned error. Just zero that portion of the page and set
		 * the buffer uptodate.
		 */
handle_hole:
		bh->b_blocknr = -1UL;
		clear_buffer_mapped(bh);
handle_zblock:
		kaddr = kmap_atomic(page, KM_USER0);
		memset(kaddr + i * blocksize, 0, blocksize);
		flush_dcache_page(page);
		kunmap_atomic(kaddr, KM_USER0);
		set_buffer_uptodate(bh);
	} while (i++, iblock++, (bh = bh->b_this_page) != head);

	/* Release the lock if we took it. */
	if (rl)
		up_read(&ni->runlist.lock);

	/* Check we have at least one buffer ready for i/o. */
	if (nr) {
		struct buffer_head *tbh;

		/* Lock the buffers. */
		for (i = 0; i < nr; i++) {
			tbh = arr[i];
			lock_buffer(tbh);
			tbh->b_end_io = ntfs_end_buffer_async_read;
			set_buffer_async_read(tbh);
		}
		/* Finally, start i/o on the buffers. */
		for (i = 0; i < nr; i++) {
			tbh = arr[i];
			if (likely(!buffer_uptodate(tbh)))
				submit_bh(READ, tbh);
			else
				ntfs_end_buffer_async_read(tbh, 1);
		}
		return 0;
	}
	/* No i/o was scheduled on any of the buffers. */
	if (likely(!PageError(page)))
		SetPageUptodate(page);
	else /* Signal synchronous i/o error. */
		nr = -EIO;
	unlock_page(page);
	return nr;
}

/**
 * ntfs_readpage - fill a @page of a @file with data from the device
 * @file:	open file to which the page @page belongs or NULL
 * @page:	page cache page to fill with data
 *
 * For non-resident attributes, ntfs_readpage() fills the @page of the open
 * file @file by calling the ntfs version of the generic block_read_full_page()
 * function, ntfs_read_block(), which in turn creates and reads in the buffers
 * associated with the page asynchronously.
 *
 * For resident attributes, OTOH, ntfs_readpage() fills @page by copying the
 * data from the mft record (which at this stage is most likely in memory) and
 * fills the remainder with zeroes. Thus, in this case, I/O is synchronous, as
 * even if the mft record is not cached at this point in time, we need to wait
 * for it to be read in before we can do the copy.
 *
 * Return 0 on success and -errno on error.
 *
 * WARNING: Do not make this function static! It is used by mft.c!
 */
int ntfs_readpage(struct file *file, struct page *page)
{
	s64 attr_pos;
	ntfs_inode *ni, *base_ni;
	u8 *kaddr;
	ntfs_attr_search_ctx *ctx;
	MFT_RECORD *mrec;
	u32 attr_len;
	int err = 0;

	BUG_ON(!PageLocked(page));

	/*
	 * This can potentially happen because we clear PageUptodate() during
	 * ntfs_writepage() of MstProtected() attributes.
	 */
	if (PageUptodate(page)) {
		unlock_page(page);
		return 0;
	}

	ni = NTFS_I(page->mapping->host);

	/* NInoNonResident() == NInoIndexAllocPresent() */
	if (NInoNonResident(ni)) {
		/*
		 * Only unnamed $DATA attributes can be compressed or
		 * encrypted.
		 */
		if (ni->type == AT_DATA && !ni->name_len) {
			/* If file is encrypted, deny access, just like NT4. */
			if (NInoEncrypted(ni)) {
				err = -EACCES;
				goto err_out;
			}
			/* Compressed data streams are handled in compress.c. */
			if (NInoCompressed(ni))
				return ntfs_read_compressed_block(page);
		}
		/* Normal data stream. */
		return ntfs_read_block(page);
	}
	/* Attribute is resident, implying it is not compressed or encrypted. */
	if (!NInoAttr(ni))
		base_ni = ni;
	else
		base_ni = ni->ext.base_ntfs_ino;

	/* Map, pin, and lock the mft record. */
	mrec = map_mft_record(base_ni);
	if (IS_ERR(mrec)) {
		err = PTR_ERR(mrec);
		goto err_out;
	}
	ctx = ntfs_attr_get_search_ctx(base_ni, mrec);
	if (unlikely(!ctx)) {
		err = -ENOMEM;
		goto unm_err_out;
	}
	err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len,
			CASE_SENSITIVE, 0, NULL, 0, ctx);
	if (unlikely(err))
		goto put_unm_err_out;

	/* Starting position of the page within the attribute value. */
	attr_pos = page->index << PAGE_CACHE_SHIFT;

	/* The total length of the attribute value. */
	attr_len = le32_to_cpu(ctx->attr->data.resident.value_length);

	kaddr = kmap_atomic(page, KM_USER0);
	/* Copy over in bounds data, zeroing the remainder of the page. */
	if (attr_pos < attr_len) {
		u32 bytes = attr_len - attr_pos;
		if (bytes > PAGE_CACHE_SIZE)
			bytes = PAGE_CACHE_SIZE;
		else if (bytes < PAGE_CACHE_SIZE)
			memset(kaddr + bytes, 0, PAGE_CACHE_SIZE - bytes);
		/* Copy the data to the page. */
		memcpy(kaddr, attr_pos + (char*)ctx->attr +
				le16_to_cpu(
				ctx->attr->data.resident.value_offset), bytes);
	} else
		memset(kaddr, 0, PAGE_CACHE_SIZE);
	flush_dcache_page(page);
	kunmap_atomic(kaddr, KM_USER0);

	SetPageUptodate(page);
put_unm_err_out:
	ntfs_attr_put_search_ctx(ctx);
unm_err_out:
	unmap_mft_record(base_ni);
err_out:
	unlock_page(page);
	return err;
}

#ifdef NTFS_RW

/**
 * ntfs_write_block - write a @page to the backing store
 * @wbc:	writeback control structure
 * @page:	page cache page to write out
 *
 * This function is for writing pages belonging to non-resident, non-mst
 * protected attributes to their backing store.
 *
 * For a page with buffers, map and write the dirty buffers asynchronously
 * under page writeback. For a page without buffers, create buffers for the
 * page, then proceed as above.
 *
 * If a page doesn't have buffers the page dirty state is definitive. If a page
 * does have buffers, the page dirty state is just a hint, and the buffer dirty
 * state is definitive. (A hint which has rules: dirty buffers against a clean
 * page is illegal. Other combinations are legal and need to be handled. In
 * particular a dirty page containing clean buffers for example.)
 *
 * Return 0 on success and -errno on error.
 *
 * Based on ntfs_read_block() and __block_write_full_page().
 */
static int ntfs_write_block(struct writeback_control *wbc, struct page *page)
{
	VCN vcn;
	LCN lcn;
	sector_t block, dblock, iblock;
	struct inode *vi;
	ntfs_inode *ni;
	ntfs_volume *vol;
	runlist_element *rl;
	struct buffer_head *bh, *head;
	unsigned int blocksize, vcn_ofs;
	int err;
	BOOL need_end_writeback;
	unsigned char blocksize_bits;

	vi = page->mapping->host;
	ni = NTFS_I(vi);
	vol = ni->vol;

	ntfs_debug("Entering for inode 0x%lx, attribute type 0x%x, page index "
			"0x%lx.", vi->i_ino, ni->type, page->index);

	BUG_ON(!NInoNonResident(ni));
	BUG_ON(NInoMstProtected(ni));

	blocksize_bits = vi->i_blkbits;
	blocksize = 1 << blocksize_bits;

	if (!page_has_buffers(page)) {
		BUG_ON(!PageUptodate(page));
		create_empty_buffers(page, blocksize,
				(1 << BH_Uptodate) | (1 << BH_Dirty));
	}
	bh = head = page_buffers(page);
	if (unlikely(!bh)) {
		ntfs_warning(vol->sb, "Error allocating page buffers. "
				"Redirtying page so we try again later.");
		/*
		 * Put the page back on mapping->dirty_pages, but leave its
		 * buffer's dirty state as-is.
		 */
		redirty_page_for_writepage(wbc, page);