[net-next-2.6.git] / fs / btrfs / compression.c

/*
 * Copyright (C) 2008 Oracle.  All rights reserved.
 *
 * This program is free software; you can redistribute it and/or
 * modify it under the terms of the GNU General Public
 * License v2 as published by the Free Software Foundation.
 *
 * This program 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; if not, write to the
 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
 * Boston, MA 021110-1307, USA.
 */

#include <linux/kernel.h>
#include <linux/bio.h>
#include <linux/buffer_head.h>
#include <linux/file.h>
#include <linux/fs.h>
#include <linux/pagemap.h>
#include <linux/highmem.h>
#include <linux/time.h>
#include <linux/init.h>
#include <linux/string.h>
#include <linux/backing-dev.h>
#include <linux/mpage.h>
#include <linux/swap.h>
#include <linux/writeback.h>
#include <linux/bit_spinlock.h>
#include <linux/slab.h>
#include "compat.h"
#include "ctree.h"
#include "disk-io.h"
#include "transaction.h"
#include "btrfs_inode.h"
#include "volumes.h"
#include "ordered-data.h"
#include "compression.h"
#include "extent_io.h"
#include "extent_map.h"

struct compressed_bio {
	/* number of bios pending for this compressed extent */
	atomic_t pending_bios;

	/* the pages with the compressed data on them */
	struct page **compressed_pages;

	/* inode that owns this data */
	struct inode *inode;

	/* starting offset in the inode for our pages */
	u64 start;

	/* number of bytes in the inode we're working on */
	unsigned long len;

	/* number of bytes on disk */
	unsigned long compressed_len;

	/* number of compressed pages in the array */
	unsigned long nr_pages;

	/* IO errors */
	int errors;
	int mirror_num;

	/* for reads, this is the bio we are copying the data into */
	struct bio *orig_bio;

	/*
	 * the start of a variable length array of checksums only
	 * used by reads
	 */
	u32 sums;
};

static inline int compressed_bio_size(struct btrfs_root *root,
				      unsigned long disk_size)
{
	u16 csum_size = btrfs_super_csum_size(&root->fs_info->super_copy);
	return sizeof(struct compressed_bio) +
		((disk_size + root->sectorsize - 1) / root->sectorsize) *
		csum_size;
}

static struct bio *compressed_bio_alloc(struct block_device *bdev,
					u64 first_byte, gfp_t gfp_flags)
{
	struct bio *bio;
	int nr_vecs;

	nr_vecs = bio_get_nr_vecs(bdev);
	bio = bio_alloc(gfp_flags, nr_vecs);

	if (bio == NULL && (current->flags & PF_MEMALLOC)) {
		while (!bio && (nr_vecs /= 2))
			bio = bio_alloc(gfp_flags, nr_vecs);
	}

	if (bio) {
		bio->bi_size = 0;
		bio->bi_bdev = bdev;
		bio->bi_sector = first_byte >> 9;
	}
	return bio;
}

static int check_compressed_csum(struct inode *inode,
				 struct compressed_bio *cb,
				 u64 disk_start)
{
	int ret;
	struct btrfs_root *root = BTRFS_I(inode)->root;
	struct page *page;
	unsigned long i;
	char *kaddr;
	u32 csum;
	u32 *cb_sum = &cb->sums;

	if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
		return 0;

	for (i = 0; i < cb->nr_pages; i++) {
		page = cb->compressed_pages[i];
		csum = ~(u32)0;

		kaddr = kmap_atomic(page, KM_USER0);
		csum = btrfs_csum_data(root, kaddr, csum, PAGE_CACHE_SIZE);
		btrfs_csum_final(csum, (char *)&csum);
		kunmap_atomic(kaddr, KM_USER0);

		if (csum != *cb_sum) {
			printk(KERN_INFO "btrfs csum failed ino %lu "
			       "extent %llu csum %u "
			       "wanted %u mirror %d\n", inode->i_ino,
			       (unsigned long long)disk_start,
			       csum, *cb_sum, cb->mirror_num);
			ret = -EIO;
			goto fail;
		}
		cb_sum++;

	}
	ret = 0;
fail:
	return ret;
}

/* when we finish reading compressed pages from the disk, we
 * decompress them and then run the bio end_io routines on the
 * decompressed pages (in the inode address space).
 *
 * This allows the checksumming and other IO error handling routines
 * to work normally
 *
 * The compressed pages are freed here, and it must be run
 * in process context
 */
static void end_compressed_bio_read(struct bio *bio, int err)
{
	struct extent_io_tree *tree;
	struct compressed_bio *cb = bio->bi_private;
	struct inode *inode;
	struct page *page;
	unsigned long index;
	int ret;

	if (err)
		cb->errors = 1;

	/* if there are more bios still pending for this compressed
	 * extent, just exit
	 */
	if (!atomic_dec_and_test(&cb->pending_bios))
		goto out;

	inode = cb->inode;
	ret = check_compressed_csum(inode, cb, (u64)bio->bi_sector << 9);
	if (ret)
		goto csum_failed;

	/* ok, we're the last bio for this extent, lets start
	 * the decompression.
	 */
	tree = &BTRFS_I(inode)->io_tree;
	ret = btrfs_zlib_decompress_biovec(cb->compressed_pages,
					cb->start,
					cb->orig_bio->bi_io_vec,
					cb->orig_bio->bi_vcnt,
					cb->compressed_len);
csum_failed:
	if (ret)
		cb->errors = 1;

	/* release the compressed pages */
	index = 0;
	for (index = 0; index < cb->nr_pages; index++) {
		page = cb->compressed_pages[index];
		page->mapping = NULL;
		page_cache_release(page);
	}

	/* do io completion on the original bio */
	if (cb->errors) {
		bio_io_error(cb->orig_bio);
	} else {
		int bio_index = 0;
		struct bio_vec *bvec = cb->orig_bio->bi_io_vec;

		/*
		 * we have verified the checksum already, set page
		 * checked so the end_io handlers know about it
		 */
		while (bio_index < cb->orig_bio->bi_vcnt) {
			SetPageChecked(bvec->bv_page);
			bvec++;
			bio_index++;
		}
		bio_endio(cb->orig_bio, 0);
	}

	/* finally free the cb struct */
	kfree(cb->compressed_pages);
	kfree(cb);
out:
	bio_put(bio);
}

/*
 * Clear the writeback bits on all of the file
 * pages for a compressed write
 */
static noinline int end_compressed_writeback(struct inode *inode, u64 start,
					     unsigned long ram_size)
{
	unsigned long index = start >> PAGE_CACHE_SHIFT;
	unsigned long end_index = (start + ram_size - 1) >> PAGE_CACHE_SHIFT;
	struct page *pages[16];
	unsigned long nr_pages = end_index - index + 1;
	int i;
	int ret;

	while (nr_pages > 0) {
		ret = find_get_pages_contig(inode->i_mapping, index,
				     min_t(unsigned long,
				     nr_pages, ARRAY_SIZE(pages)), pages);
		if (ret == 0) {
			nr_pages -= 1;
			index += 1;
			continue;
		}
		for (i = 0; i < ret; i++) {
			end_page_writeback(pages[i]);
			page_cache_release(pages[i]);
		}
		nr_pages -= ret;
		index += ret;
	}
	/* the inode may be gone now */
	return 0;
}

/*
 * do the cleanup once all the compressed pages hit the disk.
 * This will clear writeback on the file pages and free the compressed
 * pages.
 *
 * This also calls the writeback end hooks for the file pages so that
 * metadata and checksums can be updated in the file.
 */
static void end_compressed_bio_write(struct bio *bio, int err)
{
	struct extent_io_tree *tree;
	struct compressed_bio *cb = bio->bi_private;
	struct inode *inode;
	struct page *page;
	unsigned long index;

	if (err)
		cb->errors = 1;

	/* if there are more bios still pending for this compressed
	 * extent, just exit
	 */
	if (!atomic_dec_and_test(&cb->pending_bios))
		goto out;

	/* ok, we're the last bio for this extent, step one is to
	 * call back into the FS and do all the end_io operations
	 */
	inode = cb->inode;
	tree = &BTRFS_I(inode)->io_tree;
	cb->compressed_pages[0]->mapping = cb->inode->i_mapping;
	tree->ops->writepage_end_io_hook(cb->compressed_pages[0],
					 cb->start,
					 cb->start + cb->len - 1,
					 NULL, 1);
	cb->compressed_pages[0]->mapping = NULL;

	end_compressed_writeback(inode, cb->start, cb->len);
	/* note, our inode could be gone now */

	/*
	 * release the compressed pages, these came from alloc_page and
	 * are not attached to the inode at all
	 */
	index = 0;
	for (index = 0; index < cb->nr_pages; index++) {
		page = cb->compressed_pages[index];
		page->mapping = NULL;
		page_cache_release(page);
	}

	/* finally free the cb struct */
	kfree(cb->compressed_pages);
	kfree(cb);
out:
	bio_put(bio);
}

/*
 * worker function to build and submit bios for previously compressed pages.
 * The corresponding pages in the inode should be marked for writeback
 * and the compressed pages should have a reference on them for dropping
 * when the IO is complete.
 *
 * This also checksums the file bytes and gets things ready for
 * the end io hooks.
 */
int btrfs_submit_compressed_write(struct inode *inode, u64 start,
				 unsigned long len, u64 disk_start,
				 unsigned long compressed_len,
				 struct page **compressed_pages,
				 unsigned long nr_pages)
{
	struct bio *bio = NULL;
	struct btrfs_root *root = BTRFS_I(inode)->root;
	struct compressed_bio *cb;
	unsigned long bytes_left;
	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
	int page_index = 0;
	struct page *page;
	u64 first_byte = disk_start;
	struct block_device *bdev;
	int ret;

	WARN_ON(start & ((u64)PAGE_CACHE_SIZE - 1));
	cb = kmalloc(compressed_bio_size(root, compressed_len), GFP_NOFS);
	atomic_set(&cb->pending_bios, 0);
	cb->errors = 0;
	cb->inode = inode;
	cb->start = start;
	cb->len = len;
	cb->mirror_num = 0;
	cb->compressed_pages = compressed_pages;
	cb->compressed_len = compressed_len;
	cb->orig_bio = NULL;
	cb->nr_pages = nr_pages;

	bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;

	bio = compressed_bio_alloc(bdev, first_byte, GFP_NOFS);
	bio->bi_private = cb;
	bio->bi_end_io = end_compressed_bio_write;
	atomic_inc(&cb->pending_bios);

	/* create and submit bios for the compressed pages */
	bytes_left = compressed_len;
	for (page_index = 0; page_index < cb->nr_pages; page_index++) {
		page = compressed_pages[page_index];
		page->mapping = inode->i_mapping;
		if (bio->bi_size)
			ret = io_tree->ops->merge_bio_hook(page, 0,
							   PAGE_CACHE_SIZE,
							   bio, 0);
		else
			ret = 0;

		page->mapping = NULL;
		if (ret || bio_add_page(bio, page, PAGE_CACHE_SIZE, 0) <
		    PAGE_CACHE_SIZE) {
			bio_get(bio);

			/*
			 * inc the count before we submit the bio so
			 * we know the end IO handler won't happen before
			 * we inc the count.  Otherwise, the cb might get
			 * freed before we're done setting it up
			 */
			atomic_inc(&cb->pending_bios);
			ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
			BUG_ON(ret);

			ret = btrfs_csum_one_bio(root, inode, bio, start, 1);
			BUG_ON(ret);

			ret = btrfs_map_bio(root, WRITE, bio, 0, 1);
			BUG_ON(ret);

			bio_put(bio);

			bio = compressed_bio_alloc(bdev, first_byte, GFP_NOFS);
			bio->bi_private = cb;
			bio->bi_end_io = end_compressed_bio_write;
			bio_add_page(bio, page, PAGE_CACHE_SIZE, 0);
		}
		if (bytes_left < PAGE_CACHE_SIZE) {
			printk("bytes left %lu compress len %lu nr %lu\n",
			       bytes_left, cb->compressed_len, cb->nr_pages);
		}
		bytes_left -= PAGE_CACHE_SIZE;
		first_byte += PAGE_CACHE_SIZE;
		cond_resched();
	}
	bio_get(bio);

	ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
	BUG_ON(ret);

	ret = btrfs_csum_one_bio(root, inode, bio, start, 1);
	BUG_ON(ret);

	ret = btrfs_map_bio(root, WRITE, bio, 0, 1);
	BUG_ON(ret);

	bio_put(bio);
	return 0;
}

static noinline int add_ra_bio_pages(struct inode *inode,
				     u64 compressed_end,
				     struct compressed_bio *cb)
{
	unsigned long end_index;
	unsigned long page_index;
	u64 last_offset;
	u64 isize = i_size_read(inode);
	int ret;
	struct page *page;
	unsigned long nr_pages = 0;
	struct extent_map *em;
	struct address_space *mapping = inode->i_mapping;
	struct extent_map_tree *em_tree;
	struct extent_io_tree *tree;
	u64 end;
	int misses = 0;

	page = cb->orig_bio->bi_io_vec[cb->orig_bio->bi_vcnt - 1].bv_page;
	last_offset = (page_offset(page) + PAGE_CACHE_SIZE);
	em_tree = &BTRFS_I(inode)->extent_tree;
	tree = &BTRFS_I(inode)->io_tree;

	if (isize == 0)
		return 0;

	end_index = (i_size_read(inode) - 1) >> PAGE_CACHE_SHIFT;

	while (last_offset < compressed_end) {
		page_index = last_offset >> PAGE_CACHE_SHIFT;

		if (page_index > end_index)
			break;

		rcu_read_lock();
		page = radix_tree_lookup(&mapping->page_tree, page_index);
		rcu_read_unlock();
		if (page) {
			misses++;
			if (misses > 4)
				break;
			goto next;
		}

		page = __page_cache_alloc(mapping_gfp_mask(mapping) &
								~__GFP_FS);
		if (!page)
			break;

		if (add_to_page_cache_lru(page, mapping, page_index,
								GFP_NOFS)) {
			page_cache_release(page);
			goto next;
		}

		end = last_offset + PAGE_CACHE_SIZE - 1;
		/*
		 * at this point, we have a locked page in the page cache
		 * for these bytes in the file.  But, we have to make
		 * sure they map to this compressed extent on disk.
		 */
		set_page_extent_mapped(page);
		lock_extent(tree, last_offset, end, GFP_NOFS);
		read_lock(&em_tree->lock);
		em = lookup_extent_mapping(em_tree, last_offset,
					   PAGE_CACHE_SIZE);
		read_unlock(&em_tree->lock);

		if (!em || last_offset < em->start ||
		    (last_offset + PAGE_CACHE_SIZE > extent_map_end(em)) ||
		    (em->block_start >> 9) != cb->orig_bio->bi_sector) {
			free_extent_map(em);
			unlock_extent(tree, last_offset, end, GFP_NOFS);
			unlock_page(page);
			page_cache_release(page);
			break;
		}
		free_extent_map(em);

		if (page->index == end_index) {
			char *userpage;
			size_t zero_offset = isize & (PAGE_CACHE_SIZE - 1);

			if (zero_offset) {
				int zeros;
				zeros = PAGE_CACHE_SIZE - zero_offset;
				userpage = kmap_atomic(page, KM_USER0);
				memset(userpage + zero_offset, 0, zeros);
				flush_dcache_page(page);
				kunmap_atomic(userpage, KM_USER0);
			}
		}

		ret = bio_add_page(cb->orig_bio, page,
				   PAGE_CACHE_SIZE, 0);

		if (ret == PAGE_CACHE_SIZE) {
			nr_pages++;
			page_cache_release(page);
		} else {
			unlock_extent(tree, last_offset, end, GFP_NOFS);
			unlock_page(page);
			page_cache_release(page);
			break;
		}
next:
		last_offset += PAGE_CACHE_SIZE;
	}
	return 0;
}

/*
 * for a compressed read, the bio we get passed has all the inode pages
 * in it.  We don't actually do IO on those pages but allocate new ones
 * to hold the compressed pages on disk.
 *
 * bio->bi_sector points to the compressed extent on disk
 * bio->bi_io_vec points to all of the inode pages
 * bio->bi_vcnt is a count of pages
 *
 * After the compressed pages are read, we copy the bytes into the
 * bio we were passed and then call the bio end_io calls
 */
int btrfs_submit_compressed_read(struct inode *inode, struct bio *bio,
				 int mirror_num, unsigned long bio_flags)
{
	struct extent_io_tree *tree;
	struct extent_map_tree *em_tree;
	struct compressed_bio *cb;
	struct btrfs_root *root = BTRFS_I(inode)->root;
	unsigned long uncompressed_len = bio->bi_vcnt * PAGE_CACHE_SIZE;
	unsigned long compressed_len;
	unsigned long nr_pages;
	unsigned long page_index;
	struct page *page;
	struct block_device *bdev;
	struct bio *comp_bio;
	u64 cur_disk_byte = (u64)bio->bi_sector << 9;
	u64 em_len;
	u64 em_start;
	struct extent_map *em;
	int ret;
	u32 *sums;

	tree = &BTRFS_I(inode)->io_tree;
	em_tree = &BTRFS_I(inode)->extent_tree;

	/* we need the actual starting offset of this extent in the file */
	read_lock(&em_tree->lock);
	em = lookup_extent_mapping(em_tree,
				   page_offset(bio->bi_io_vec->bv_page),
				   PAGE_CACHE_SIZE);
	read_unlock(&em_tree->lock);

	compressed_len = em->block_len;
	cb = kmalloc(compressed_bio_size(root, compressed_len), GFP_NOFS);
	atomic_set(&cb->pending_bios, 0);
	cb->errors = 0;
	cb->inode = inode;
	cb->mirror_num = mirror_num;
	sums = &cb->sums;

	cb->start = em->orig_start;
	em_len = em->len;
	em_start = em->start;

	free_extent_map(em);
	em = NULL;

	cb->len = uncompressed_len;
	cb->compressed_len = compressed_len;
	cb->orig_bio = bio;

	nr_pages = (compressed_len + PAGE_CACHE_SIZE - 1) /
				 PAGE_CACHE_SIZE;
	cb->compressed_pages = kmalloc(sizeof(struct page *) * nr_pages,
				       GFP_NOFS);
	bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;

	for (page_index = 0; page_index < nr_pages; page_index++) {
		cb->compressed_pages[page_index] = alloc_page(GFP_NOFS |
							      __GFP_HIGHMEM);
	}
	cb->nr_pages = nr_pages;

	add_ra_bio_pages(inode, em_start + em_len, cb);

	/* include any pages we added in add_ra-bio_pages */
	uncompressed_len = bio->bi_vcnt * PAGE_CACHE_SIZE;
	cb->len = uncompressed_len;

	comp_bio = compressed_bio_alloc(bdev, cur_disk_byte, GFP_NOFS);
	comp_bio->bi_private = cb;
	comp_bio->bi_end_io = end_compressed_bio_read;
	atomic_inc(&cb->pending_bios);

	for (page_index = 0; page_index < nr_pages; page_index++) {
		page = cb->compressed_pages[page_index];
		page->mapping = inode->i_mapping;
		page->index = em_start >> PAGE_CACHE_SHIFT;

		if (comp_bio->bi_size)
			ret = tree->ops->merge_bio_hook(page, 0,
							PAGE_CACHE_SIZE,
							comp_bio, 0);
		else
			ret = 0;

		page->mapping = NULL;
		if (ret || bio_add_page(comp_bio, page, PAGE_CACHE_SIZE, 0) <
		    PAGE_CACHE_SIZE) {
			bio_get(comp_bio);

			ret = btrfs_bio_wq_end_io(root->fs_info, comp_bio, 0);
			BUG_ON(ret);

			/*
			 * inc the count before we submit the bio so
			 * we know the end IO handler won't happen before
			 * we inc the count.  Otherwise, the cb might get
			 * freed before we're done setting it up
			 */
			atomic_inc(&cb->pending_bios);

			if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
				btrfs_lookup_bio_sums(root, inode, comp_bio,
						      sums);
			}
			sums += (comp_bio->bi_size + root->sectorsize - 1) /
				root->sectorsize;

			ret = btrfs_map_bio(root, READ, comp_bio,
					    mirror_num, 0);
			BUG_ON(ret);

			bio_put(comp_bio);

			comp_bio = compressed_bio_alloc(bdev, cur_disk_byte,
							GFP_NOFS);
			comp_bio->bi_private = cb;
			comp_bio->bi_end_io = end_compressed_bio_read;

			bio_add_page(comp_bio, page, PAGE_CACHE_SIZE, 0);
		}
		cur_disk_byte += PAGE_CACHE_SIZE;
	}
	bio_get(comp_bio);

	ret = btrfs_bio_wq_end_io(root->fs_info, comp_bio, 0);
	BUG_ON(ret);

	if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM))
		btrfs_lookup_bio_sums(root, inode, comp_bio, sums);

	ret = btrfs_map_bio(root, READ, comp_bio, mirror_num, 0);
	BUG_ON(ret);

	bio_put(comp_bio);
	return 0;
}
Commit	Line	Data
c8b97818 CM	1	/*
	2	* Copyright (C) 2008 Oracle. All rights reserved.
	3	*
	4	* This program is free software; you can redistribute it and/or
	5	* modify it under the terms of the GNU General Public
	6	* License v2 as published by the Free Software Foundation.
	7	*
	8	* This program is distributed in the hope that it will be useful,
	9	* but WITHOUT ANY WARRANTY; without even the implied warranty of
	10	* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
	11	* General Public License for more details.
	12	*
	13	* You should have received a copy of the GNU General Public
	14	* License along with this program; if not, write to the
	15	* Free Software Foundation, Inc., 59 Temple Place - Suite 330,
	16	* Boston, MA 021110-1307, USA.
	17	*/
	18
	19	#include <linux/kernel.h>
	20	#include <linux/bio.h>
	21	#include <linux/buffer_head.h>
	22	#include <linux/file.h>
	23	#include <linux/fs.h>
	24	#include <linux/pagemap.h>
	25	#include <linux/highmem.h>
	26	#include <linux/time.h>
	27	#include <linux/init.h>
	28	#include <linux/string.h>
c8b97818 CM	29	#include <linux/backing-dev.h>
	30	#include <linux/mpage.h>
	31	#include <linux/swap.h>
	32	#include <linux/writeback.h>
	33	#include <linux/bit_spinlock.h>
5a0e3ad6	34	#include <linux/slab.h>
4b4e25f2	35	#include "compat.h"
c8b97818 CM	36	#include "ctree.h"
	37	#include "disk-io.h"
	38	#include "transaction.h"
	39	#include "btrfs_inode.h"
	40	#include "volumes.h"
	41	#include "ordered-data.h"
c8b97818 CM	42	#include "compression.h"
	43	#include "extent_io.h"
	44	#include "extent_map.h"
	45
	46	struct compressed_bio {
	47	/* number of bios pending for this compressed extent */
	48	atomic_t pending_bios;
	49
	50	/* the pages with the compressed data on them */
	51	struct page **compressed_pages;
	52
	53	/* inode that owns this data */
	54	struct inode *inode;
	55
	56	/* starting offset in the inode for our pages */
	57	u64 start;
	58
	59	/* number of bytes in the inode we're working on */
	60	unsigned long len;
	61
	62	/* number of bytes on disk */
	63	unsigned long compressed_len;
	64
	65	/* number of compressed pages in the array */
	66	unsigned long nr_pages;
	67
	68	/* IO errors */
	69	int errors;
d20f7043	70	int mirror_num;
c8b97818 CM	71
	72	/* for reads, this is the bio we are copying the data into */
	73	struct bio *orig_bio;
d20f7043 CM	74
	75	/*
	76	* the start of a variable length array of checksums only
	77	* used by reads
	78	*/
	79	u32 sums;
c8b97818 CM	80	};
c8b97818 CM	81
d20f7043 CM	82	static inline int compressed_bio_size(struct btrfs_root *root,
	83	unsigned long disk_size)
	84	{
	85	u16 csum_size = btrfs_super_csum_size(&root->fs_info->super_copy);
	86	return sizeof(struct compressed_bio) +
	87	((disk_size + root->sectorsize - 1) / root->sectorsize) *
	88	csum_size;
	89	}
	90
c8b97818 CM	91	static struct bio compressed_bio_alloc(struct block_device bdev,
	92	u64 first_byte, gfp_t gfp_flags)
	93	{
	94	struct bio *bio;
	95	int nr_vecs;
	96
	97	nr_vecs = bio_get_nr_vecs(bdev);
	98	bio = bio_alloc(gfp_flags, nr_vecs);
	99
	100	if (bio == NULL && (current->flags & PF_MEMALLOC)) {
	101	while (!bio && (nr_vecs /= 2))
	102	bio = bio_alloc(gfp_flags, nr_vecs);
	103	}
	104
	105	if (bio) {
	106	bio->bi_size = 0;
	107	bio->bi_bdev = bdev;
	108	bio->bi_sector = first_byte >> 9;
	109	}
	110	return bio;
	111	}
	112
d20f7043 CM	113	static int check_compressed_csum(struct inode *inode,
	114	struct compressed_bio *cb,
	115	u64 disk_start)
	116	{
	117	int ret;
	118	struct btrfs_root *root = BTRFS_I(inode)->root;
	119	struct page *page;
	120	unsigned long i;
	121	char *kaddr;
	122	u32 csum;
	123	u32 *cb_sum = &cb->sums;
	124
6cbff00f	125	if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
d20f7043 CM	126	return 0;
	127
	128	for (i = 0; i < cb->nr_pages; i++) {
	129	page = cb->compressed_pages[i];
	130	csum = ~(u32)0;
	131
	132	kaddr = kmap_atomic(page, KM_USER0);
	133	csum = btrfs_csum_data(root, kaddr, csum, PAGE_CACHE_SIZE);
	134	btrfs_csum_final(csum, (char *)&csum);
	135	kunmap_atomic(kaddr, KM_USER0);
	136
	137	if (csum != *cb_sum) {
d397712b CM	138	printk(KERN_INFO "btrfs csum failed ino %lu "
d397712b CM	139	"extent %llu csum %u "
d20f7043 CM	140	"wanted %u mirror %d\n", inode->i_ino,
	141	(unsigned long long)disk_start,
	142	csum, *cb_sum, cb->mirror_num);
	143	ret = -EIO;
	144	goto fail;
	145	}
	146	cb_sum++;
	147
	148	}
	149	ret = 0;
	150	fail:
	151	return ret;
	152	}
	153
c8b97818 CM	154	/* when we finish reading compressed pages from the disk, we
	155	* decompress them and then run the bio end_io routines on the
	156	* decompressed pages (in the inode address space).
	157	*
	158	* This allows the checksumming and other IO error handling routines
	159	* to work normally
	160	*
	161	* The compressed pages are freed here, and it must be run
	162	* in process context
	163	*/
	164	static void end_compressed_bio_read(struct bio *bio, int err)
	165	{
	166	struct extent_io_tree *tree;
	167	struct compressed_bio *cb = bio->bi_private;
	168	struct inode *inode;
	169	struct page *page;
	170	unsigned long index;
	171	int ret;
	172
	173	if (err)
	174	cb->errors = 1;
	175
	176	/* if there are more bios still pending for this compressed
	177	* extent, just exit
	178	*/
	179	if (!atomic_dec_and_test(&cb->pending_bios))
	180	goto out;
	181
d20f7043 CM	182	inode = cb->inode;
	183	ret = check_compressed_csum(inode, cb, (u64)bio->bi_sector << 9);
	184	if (ret)
	185	goto csum_failed;
	186
c8b97818 CM	187	/* ok, we're the last bio for this extent, lets start
	188	* the decompression.
	189	*/
c8b97818 CM	190	tree = &BTRFS_I(inode)->io_tree;
	191	ret = btrfs_zlib_decompress_biovec(cb->compressed_pages,
	192	cb->start,
	193	cb->orig_bio->bi_io_vec,
	194	cb->orig_bio->bi_vcnt,
	195	cb->compressed_len);
d20f7043	196	csum_failed:
c8b97818 CM	197	if (ret)
	198	cb->errors = 1;
	199
	200	/* release the compressed pages */
	201	index = 0;
	202	for (index = 0; index < cb->nr_pages; index++) {
	203	page = cb->compressed_pages[index];
	204	page->mapping = NULL;
	205	page_cache_release(page);
	206	}
	207
	208	/* do io completion on the original bio */
771ed689	209	if (cb->errors) {
c8b97818	210	bio_io_error(cb->orig_bio);
d20f7043 CM	211	} else {
	212	int bio_index = 0;
	213	struct bio_vec *bvec = cb->orig_bio->bi_io_vec;
	214
	215	/*
	216	* we have verified the checksum already, set page
	217	* checked so the end_io handlers know about it
	218	*/
d397712b	219	while (bio_index < cb->orig_bio->bi_vcnt) {
d20f7043 CM	220	SetPageChecked(bvec->bv_page);
	221	bvec++;
	222	bio_index++;
	223	}
c8b97818	224	bio_endio(cb->orig_bio, 0);
d20f7043	225	}
c8b97818 CM	226
	227	/* finally free the cb struct */
	228	kfree(cb->compressed_pages);
	229	kfree(cb);
	230	out:
	231	bio_put(bio);
	232	}
	233
	234	/*
	235	* Clear the writeback bits on all of the file
	236	* pages for a compressed write
	237	*/
	238	static noinline int end_compressed_writeback(struct inode *inode, u64 start,
	239	unsigned long ram_size)
	240	{
	241	unsigned long index = start >> PAGE_CACHE_SHIFT;
	242	unsigned long end_index = (start + ram_size - 1) >> PAGE_CACHE_SHIFT;
	243	struct page *pages[16];
	244	unsigned long nr_pages = end_index - index + 1;
	245	int i;
	246	int ret;
	247
d397712b	248	while (nr_pages > 0) {
c8b97818	249	ret = find_get_pages_contig(inode->i_mapping, index,
5b050f04 CM	250	min_t(unsigned long,
5b050f04 CM	251	nr_pages, ARRAY_SIZE(pages)), pages);
c8b97818 CM	252	if (ret == 0) {
	253	nr_pages -= 1;
	254	index += 1;
	255	continue;
	256	}
	257	for (i = 0; i < ret; i++) {
	258	end_page_writeback(pages[i]);
	259	page_cache_release(pages[i]);
	260	}
	261	nr_pages -= ret;
	262	index += ret;
	263	}
	264	/* the inode may be gone now */
	265	return 0;
	266	}
	267
	268	/*
	269	* do the cleanup once all the compressed pages hit the disk.
	270	* This will clear writeback on the file pages and free the compressed
	271	* pages.
	272	*
	273	* This also calls the writeback end hooks for the file pages so that
	274	* metadata and checksums can be updated in the file.
	275	*/
	276	static void end_compressed_bio_write(struct bio *bio, int err)
	277	{
	278	struct extent_io_tree *tree;
	279	struct compressed_bio *cb = bio->bi_private;
	280	struct inode *inode;
	281	struct page *page;
	282	unsigned long index;
	283
	284	if (err)
	285	cb->errors = 1;
	286
	287	/* if there are more bios still pending for this compressed
	288	* extent, just exit
	289	*/
	290	if (!atomic_dec_and_test(&cb->pending_bios))
	291	goto out;
	292
	293	/* ok, we're the last bio for this extent, step one is to
	294	* call back into the FS and do all the end_io operations
	295	*/
	296	inode = cb->inode;
	297	tree = &BTRFS_I(inode)->io_tree;
70b99e69	298	cb->compressed_pages[0]->mapping = cb->inode->i_mapping;
c8b97818 CM	299	tree->ops->writepage_end_io_hook(cb->compressed_pages[0],
	300	cb->start,
	301	cb->start + cb->len - 1,
	302	NULL, 1);
70b99e69	303	cb->compressed_pages[0]->mapping = NULL;
c8b97818 CM	304
	305	end_compressed_writeback(inode, cb->start, cb->len);
	306	/* note, our inode could be gone now */
	307
	308	/*
	309	* release the compressed pages, these came from alloc_page and
	310	* are not attached to the inode at all
	311	*/
	312	index = 0;
	313	for (index = 0; index < cb->nr_pages; index++) {
	314	page = cb->compressed_pages[index];
	315	page->mapping = NULL;
	316	page_cache_release(page);
	317	}
	318
	319	/* finally free the cb struct */
	320	kfree(cb->compressed_pages);
	321	kfree(cb);
	322	out:
	323	bio_put(bio);
	324	}
	325
	326	/*
	327	* worker function to build and submit bios for previously compressed pages.
	328	* The corresponding pages in the inode should be marked for writeback
	329	* and the compressed pages should have a reference on them for dropping
	330	* when the IO is complete.
	331	*
	332	* This also checksums the file bytes and gets things ready for
	333	* the end io hooks.
	334	*/
	335	int btrfs_submit_compressed_write(struct inode *inode, u64 start,
	336	unsigned long len, u64 disk_start,
	337	unsigned long compressed_len,
	338	struct page **compressed_pages,
	339	unsigned long nr_pages)
	340	{
	341	struct bio *bio = NULL;
	342	struct btrfs_root *root = BTRFS_I(inode)->root;
	343	struct compressed_bio *cb;
	344	unsigned long bytes_left;
	345	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
	346	int page_index = 0;
	347	struct page *page;
	348	u64 first_byte = disk_start;
	349	struct block_device *bdev;
	350	int ret;
	351
	352	WARN_ON(start & ((u64)PAGE_CACHE_SIZE - 1));
d20f7043	353	cb = kmalloc(compressed_bio_size(root, compressed_len), GFP_NOFS);
c8b97818 CM	354	atomic_set(&cb->pending_bios, 0);
	355	cb->errors = 0;
	356	cb->inode = inode;
	357	cb->start = start;
	358	cb->len = len;
d20f7043	359	cb->mirror_num = 0;
c8b97818 CM	360	cb->compressed_pages = compressed_pages;
	361	cb->compressed_len = compressed_len;
	362	cb->orig_bio = NULL;
	363	cb->nr_pages = nr_pages;
	364
	365	bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
	366
c8b97818 CM	367	bio = compressed_bio_alloc(bdev, first_byte, GFP_NOFS);
	368	bio->bi_private = cb;
	369	bio->bi_end_io = end_compressed_bio_write;
	370	atomic_inc(&cb->pending_bios);
	371
	372	/* create and submit bios for the compressed pages */
	373	bytes_left = compressed_len;
cfbc246e	374	for (page_index = 0; page_index < cb->nr_pages; page_index++) {
c8b97818 CM	375	page = compressed_pages[page_index];
	376	page->mapping = inode->i_mapping;
	377	if (bio->bi_size)
	378	ret = io_tree->ops->merge_bio_hook(page, 0,
	379	PAGE_CACHE_SIZE,
	380	bio, 0);
	381	else
	382	ret = 0;
	383
70b99e69	384	page->mapping = NULL;
c8b97818 CM	385	if (ret \|\| bio_add_page(bio, page, PAGE_CACHE_SIZE, 0) <
	386	PAGE_CACHE_SIZE) {
	387	bio_get(bio);
	388
af09abfe CM	389	/*
	390	* inc the count before we submit the bio so
	391	* we know the end IO handler won't happen before
	392	* we inc the count. Otherwise, the cb might get
	393	* freed before we're done setting it up
	394	*/
	395	atomic_inc(&cb->pending_bios);
c8b97818 CM	396	ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
	397	BUG_ON(ret);
	398
d20f7043 CM	399	ret = btrfs_csum_one_bio(root, inode, bio, start, 1);
	400	BUG_ON(ret);
	401
c8b97818 CM	402	ret = btrfs_map_bio(root, WRITE, bio, 0, 1);
	403	BUG_ON(ret);
	404
	405	bio_put(bio);
	406
	407	bio = compressed_bio_alloc(bdev, first_byte, GFP_NOFS);
c8b97818 CM	408	bio->bi_private = cb;
	409	bio->bi_end_io = end_compressed_bio_write;
	410	bio_add_page(bio, page, PAGE_CACHE_SIZE, 0);
	411	}
cfbc246e CM	412	if (bytes_left < PAGE_CACHE_SIZE) {
	413	printk("bytes left %lu compress len %lu nr %lu\n",
	414	bytes_left, cb->compressed_len, cb->nr_pages);
	415	}
c8b97818 CM	416	bytes_left -= PAGE_CACHE_SIZE;
c8b97818 CM	417	first_byte += PAGE_CACHE_SIZE;
771ed689	418	cond_resched();
c8b97818 CM	419	}
	420	bio_get(bio);
	421
	422	ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
	423	BUG_ON(ret);
	424
d20f7043 CM	425	ret = btrfs_csum_one_bio(root, inode, bio, start, 1);
	426	BUG_ON(ret);
	427
c8b97818 CM	428	ret = btrfs_map_bio(root, WRITE, bio, 0, 1);
	429	BUG_ON(ret);
	430
	431	bio_put(bio);
	432	return 0;
	433	}
	434
771ed689 CM	435	static noinline int add_ra_bio_pages(struct inode *inode,
	436	u64 compressed_end,
	437	struct compressed_bio *cb)
	438	{
	439	unsigned long end_index;
	440	unsigned long page_index;
	441	u64 last_offset;
	442	u64 isize = i_size_read(inode);
	443	int ret;
	444	struct page *page;
	445	unsigned long nr_pages = 0;
	446	struct extent_map *em;
	447	struct address_space *mapping = inode->i_mapping;
771ed689 CM	448	struct extent_map_tree *em_tree;
	449	struct extent_io_tree *tree;
	450	u64 end;
	451	int misses = 0;
	452
	453	page = cb->orig_bio->bi_io_vec[cb->orig_bio->bi_vcnt - 1].bv_page;
	454	last_offset = (page_offset(page) + PAGE_CACHE_SIZE);
	455	em_tree = &BTRFS_I(inode)->extent_tree;
	456	tree = &BTRFS_I(inode)->io_tree;
	457
	458	if (isize == 0)
	459	return 0;
	460
	461	end_index = (i_size_read(inode) - 1) >> PAGE_CACHE_SHIFT;
	462
d397712b	463	while (last_offset < compressed_end) {
771ed689 CM	464	page_index = last_offset >> PAGE_CACHE_SHIFT;
	465
	466	if (page_index > end_index)
	467	break;
	468
	469	rcu_read_lock();
	470	page = radix_tree_lookup(&mapping->page_tree, page_index);
	471	rcu_read_unlock();
	472	if (page) {
	473	misses++;
	474	if (misses > 4)
	475	break;
	476	goto next;
	477	}
	478
28ecb609 NP	479	page = __page_cache_alloc(mapping_gfp_mask(mapping) &
28ecb609 NP	480	~__GFP_FS);
771ed689 CM	481	if (!page)
	482	break;
	483
28ecb609 NP	484	if (add_to_page_cache_lru(page, mapping, page_index,
28ecb609 NP	485	GFP_NOFS)) {
771ed689 CM	486	page_cache_release(page);
	487	goto next;
	488	}
	489
771ed689 CM	490	end = last_offset + PAGE_CACHE_SIZE - 1;
	491	/*
	492	* at this point, we have a locked page in the page cache
	493	* for these bytes in the file. But, we have to make
	494	* sure they map to this compressed extent on disk.
	495	*/
	496	set_page_extent_mapped(page);
	497	lock_extent(tree, last_offset, end, GFP_NOFS);
890871be	498	read_lock(&em_tree->lock);
771ed689 CM	499	em = lookup_extent_mapping(em_tree, last_offset,
771ed689 CM	500	PAGE_CACHE_SIZE);
890871be	501	read_unlock(&em_tree->lock);
771ed689 CM	502
	503	if (!em \|\| last_offset < em->start \|\|
	504	(last_offset + PAGE_CACHE_SIZE > extent_map_end(em)) \|\|
	505	(em->block_start >> 9) != cb->orig_bio->bi_sector) {
	506	free_extent_map(em);
	507	unlock_extent(tree, last_offset, end, GFP_NOFS);
	508	unlock_page(page);
	509	page_cache_release(page);
	510	break;
	511	}
	512	free_extent_map(em);
	513
	514	if (page->index == end_index) {
	515	char *userpage;
	516	size_t zero_offset = isize & (PAGE_CACHE_SIZE - 1);
	517
	518	if (zero_offset) {
	519	int zeros;
	520	zeros = PAGE_CACHE_SIZE - zero_offset;
	521	userpage = kmap_atomic(page, KM_USER0);
	522	memset(userpage + zero_offset, 0, zeros);
	523	flush_dcache_page(page);
	524	kunmap_atomic(userpage, KM_USER0);
	525	}
	526	}
	527
	528	ret = bio_add_page(cb->orig_bio, page,
	529	PAGE_CACHE_SIZE, 0);
	530
	531	if (ret == PAGE_CACHE_SIZE) {
	532	nr_pages++;
	533	page_cache_release(page);
	534	} else {
	535	unlock_extent(tree, last_offset, end, GFP_NOFS);
	536	unlock_page(page);
	537	page_cache_release(page);
	538	break;
	539	}
	540	next:
	541	last_offset += PAGE_CACHE_SIZE;
	542	}
771ed689 CM	543	return 0;
	544	}
	545
c8b97818 CM	546	/*
	547	* for a compressed read, the bio we get passed has all the inode pages
	548	* in it. We don't actually do IO on those pages but allocate new ones
	549	* to hold the compressed pages on disk.
	550	*
	551	* bio->bi_sector points to the compressed extent on disk
	552	* bio->bi_io_vec points to all of the inode pages
	553	* bio->bi_vcnt is a count of pages
	554	*
	555	* After the compressed pages are read, we copy the bytes into the
	556	* bio we were passed and then call the bio end_io calls
	557	*/
	558	int btrfs_submit_compressed_read(struct inode inode, struct bio bio,
	559	int mirror_num, unsigned long bio_flags)
	560	{
	561	struct extent_io_tree *tree;
	562	struct extent_map_tree *em_tree;
	563	struct compressed_bio *cb;
	564	struct btrfs_root *root = BTRFS_I(inode)->root;
	565	unsigned long uncompressed_len = bio->bi_vcnt * PAGE_CACHE_SIZE;
	566	unsigned long compressed_len;
	567	unsigned long nr_pages;
	568	unsigned long page_index;
	569	struct page *page;
	570	struct block_device *bdev;
	571	struct bio *comp_bio;
	572	u64 cur_disk_byte = (u64)bio->bi_sector << 9;
e04ca626 CM	573	u64 em_len;
e04ca626 CM	574	u64 em_start;
c8b97818 CM	575	struct extent_map *em;
c8b97818 CM	576	int ret;
d20f7043	577	u32 *sums;
c8b97818 CM	578
	579	tree = &BTRFS_I(inode)->io_tree;
	580	em_tree = &BTRFS_I(inode)->extent_tree;
	581
	582	/* we need the actual starting offset of this extent in the file */
890871be	583	read_lock(&em_tree->lock);
c8b97818 CM	584	em = lookup_extent_mapping(em_tree,
	585	page_offset(bio->bi_io_vec->bv_page),
	586	PAGE_CACHE_SIZE);
890871be	587	read_unlock(&em_tree->lock);
c8b97818	588
d20f7043 CM	589	compressed_len = em->block_len;
d20f7043 CM	590	cb = kmalloc(compressed_bio_size(root, compressed_len), GFP_NOFS);
c8b97818 CM	591	atomic_set(&cb->pending_bios, 0);
	592	cb->errors = 0;
	593	cb->inode = inode;
d20f7043 CM	594	cb->mirror_num = mirror_num;
d20f7043 CM	595	sums = &cb->sums;
c8b97818	596
ff5b7ee3	597	cb->start = em->orig_start;
e04ca626 CM	598	em_len = em->len;
e04ca626 CM	599	em_start = em->start;
d20f7043	600
c8b97818	601	free_extent_map(em);
e04ca626	602	em = NULL;
c8b97818 CM	603
	604	cb->len = uncompressed_len;
	605	cb->compressed_len = compressed_len;
	606	cb->orig_bio = bio;
	607
	608	nr_pages = (compressed_len + PAGE_CACHE_SIZE - 1) /
	609	PAGE_CACHE_SIZE;
	610	cb->compressed_pages = kmalloc(sizeof(struct page ) nr_pages,
	611	GFP_NOFS);
	612	bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
	613
	614	for (page_index = 0; page_index < nr_pages; page_index++) {
	615	cb->compressed_pages[page_index] = alloc_page(GFP_NOFS \|
	616	__GFP_HIGHMEM);
	617	}
	618	cb->nr_pages = nr_pages;
	619
e04ca626	620	add_ra_bio_pages(inode, em_start + em_len, cb);
771ed689	621
771ed689 CM	622	/* include any pages we added in add_ra-bio_pages */
	623	uncompressed_len = bio->bi_vcnt * PAGE_CACHE_SIZE;
	624	cb->len = uncompressed_len;
	625
c8b97818 CM	626	comp_bio = compressed_bio_alloc(bdev, cur_disk_byte, GFP_NOFS);
	627	comp_bio->bi_private = cb;
	628	comp_bio->bi_end_io = end_compressed_bio_read;
	629	atomic_inc(&cb->pending_bios);
	630
	631	for (page_index = 0; page_index < nr_pages; page_index++) {
	632	page = cb->compressed_pages[page_index];
	633	page->mapping = inode->i_mapping;
d20f7043 CM	634	page->index = em_start >> PAGE_CACHE_SHIFT;
d20f7043 CM	635
c8b97818 CM	636	if (comp_bio->bi_size)
	637	ret = tree->ops->merge_bio_hook(page, 0,
	638	PAGE_CACHE_SIZE,
	639	comp_bio, 0);
	640	else
	641	ret = 0;
	642
70b99e69	643	page->mapping = NULL;
c8b97818 CM	644	if (ret \|\| bio_add_page(comp_bio, page, PAGE_CACHE_SIZE, 0) <
	645	PAGE_CACHE_SIZE) {
	646	bio_get(comp_bio);
	647
	648	ret = btrfs_bio_wq_end_io(root->fs_info, comp_bio, 0);
	649	BUG_ON(ret);
	650
af09abfe CM	651	/*
	652	* inc the count before we submit the bio so
	653	* we know the end IO handler won't happen before
	654	* we inc the count. Otherwise, the cb might get
	655	* freed before we're done setting it up
	656	*/
	657	atomic_inc(&cb->pending_bios);
	658
6cbff00f	659	if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
d20f7043 CM	660	btrfs_lookup_bio_sums(root, inode, comp_bio,
	661	sums);
	662	}
	663	sums += (comp_bio->bi_size + root->sectorsize - 1) /
	664	root->sectorsize;
	665
	666	ret = btrfs_map_bio(root, READ, comp_bio,
	667	mirror_num, 0);
c8b97818 CM	668	BUG_ON(ret);
	669
	670	bio_put(comp_bio);
	671
	672	comp_bio = compressed_bio_alloc(bdev, cur_disk_byte,
	673	GFP_NOFS);
771ed689 CM	674	comp_bio->bi_private = cb;
	675	comp_bio->bi_end_io = end_compressed_bio_read;
	676
	677	bio_add_page(comp_bio, page, PAGE_CACHE_SIZE, 0);
c8b97818 CM	678	}
	679	cur_disk_byte += PAGE_CACHE_SIZE;
	680	}
	681	bio_get(comp_bio);
	682
	683	ret = btrfs_bio_wq_end_io(root->fs_info, comp_bio, 0);
	684	BUG_ON(ret);
	685
6cbff00f	686	if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM))
d20f7043	687	btrfs_lookup_bio_sums(root, inode, comp_bio, sums);
d20f7043 CM	688
d20f7043 CM	689	ret = btrfs_map_bio(root, READ, comp_bio, mirror_num, 0);
c8b97818 CM	690	BUG_ON(ret);
	691
	692	bio_put(comp_bio);
	693	return 0;
	694	}