[linux.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)
{
	int nr_vecs;

	nr_vecs = bio_get_nr_vecs(bdev);
	return btrfs_bio_alloc(bdev, first_byte >> 9, nr_vecs, gfp_flags);
}

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 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.
	 */
	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	{
c8b97818 CM	94	int nr_vecs;
	95
	96	nr_vecs = bio_get_nr_vecs(bdev);
88f794ed	97	return btrfs_bio_alloc(bdev, first_byte >> 9, nr_vecs, gfp_flags);
c8b97818 CM	98	}
c8b97818 CM	99
d20f7043 CM	100	static int check_compressed_csum(struct inode *inode,
	101	struct compressed_bio *cb,
	102	u64 disk_start)
	103	{
	104	int ret;
	105	struct btrfs_root *root = BTRFS_I(inode)->root;
	106	struct page *page;
	107	unsigned long i;
	108	char *kaddr;
	109	u32 csum;
	110	u32 *cb_sum = &cb->sums;
	111
6cbff00f	112	if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
d20f7043 CM	113	return 0;
	114
	115	for (i = 0; i < cb->nr_pages; i++) {
	116	page = cb->compressed_pages[i];
	117	csum = ~(u32)0;
	118
	119	kaddr = kmap_atomic(page, KM_USER0);
	120	csum = btrfs_csum_data(root, kaddr, csum, PAGE_CACHE_SIZE);
	121	btrfs_csum_final(csum, (char *)&csum);
	122	kunmap_atomic(kaddr, KM_USER0);
	123
	124	if (csum != *cb_sum) {
d397712b CM	125	printk(KERN_INFO "btrfs csum failed ino %lu "
d397712b CM	126	"extent %llu csum %u "
d20f7043 CM	127	"wanted %u mirror %d\n", inode->i_ino,
	128	(unsigned long long)disk_start,
	129	csum, *cb_sum, cb->mirror_num);
	130	ret = -EIO;
	131	goto fail;
	132	}
	133	cb_sum++;
	134
	135	}
	136	ret = 0;
	137	fail:
	138	return ret;
	139	}
	140
c8b97818 CM	141	/* when we finish reading compressed pages from the disk, we
	142	* decompress them and then run the bio end_io routines on the
	143	* decompressed pages (in the inode address space).
	144	*
	145	* This allows the checksumming and other IO error handling routines
	146	* to work normally
	147	*
	148	* The compressed pages are freed here, and it must be run
	149	* in process context
	150	*/
	151	static void end_compressed_bio_read(struct bio *bio, int err)
	152	{
c8b97818 CM	153	struct compressed_bio *cb = bio->bi_private;
	154	struct inode *inode;
	155	struct page *page;
	156	unsigned long index;
	157	int ret;
	158
	159	if (err)
	160	cb->errors = 1;
	161
	162	/* if there are more bios still pending for this compressed
	163	* extent, just exit
	164	*/
	165	if (!atomic_dec_and_test(&cb->pending_bios))
	166	goto out;
	167
d20f7043 CM	168	inode = cb->inode;
	169	ret = check_compressed_csum(inode, cb, (u64)bio->bi_sector << 9);
	170	if (ret)
	171	goto csum_failed;
	172
c8b97818 CM	173	/* ok, we're the last bio for this extent, lets start
	174	* the decompression.
	175	*/
c8b97818 CM	176	ret = btrfs_zlib_decompress_biovec(cb->compressed_pages,
	177	cb->start,
	178	cb->orig_bio->bi_io_vec,
	179	cb->orig_bio->bi_vcnt,
	180	cb->compressed_len);
d20f7043	181	csum_failed:
c8b97818 CM	182	if (ret)
	183	cb->errors = 1;
	184
	185	/* release the compressed pages */
	186	index = 0;
	187	for (index = 0; index < cb->nr_pages; index++) {
	188	page = cb->compressed_pages[index];
	189	page->mapping = NULL;
	190	page_cache_release(page);
	191	}
	192
	193	/* do io completion on the original bio */
771ed689	194	if (cb->errors) {
c8b97818	195	bio_io_error(cb->orig_bio);
d20f7043 CM	196	} else {
	197	int bio_index = 0;
	198	struct bio_vec *bvec = cb->orig_bio->bi_io_vec;
	199
	200	/*
	201	* we have verified the checksum already, set page
	202	* checked so the end_io handlers know about it
	203	*/
d397712b	204	while (bio_index < cb->orig_bio->bi_vcnt) {
d20f7043 CM	205	SetPageChecked(bvec->bv_page);
	206	bvec++;
	207	bio_index++;
	208	}
c8b97818	209	bio_endio(cb->orig_bio, 0);
d20f7043	210	}
c8b97818 CM	211
	212	/* finally free the cb struct */
	213	kfree(cb->compressed_pages);
	214	kfree(cb);
	215	out:
	216	bio_put(bio);
	217	}
	218
	219	/*
	220	* Clear the writeback bits on all of the file
	221	* pages for a compressed write
	222	*/
	223	static noinline int end_compressed_writeback(struct inode *inode, u64 start,
	224	unsigned long ram_size)
	225	{
	226	unsigned long index = start >> PAGE_CACHE_SHIFT;
	227	unsigned long end_index = (start + ram_size - 1) >> PAGE_CACHE_SHIFT;
	228	struct page *pages[16];
	229	unsigned long nr_pages = end_index - index + 1;
	230	int i;
	231	int ret;
	232
d397712b	233	while (nr_pages > 0) {
c8b97818	234	ret = find_get_pages_contig(inode->i_mapping, index,
5b050f04 CM	235	min_t(unsigned long,
5b050f04 CM	236	nr_pages, ARRAY_SIZE(pages)), pages);
c8b97818 CM	237	if (ret == 0) {
	238	nr_pages -= 1;
	239	index += 1;
	240	continue;
	241	}
	242	for (i = 0; i < ret; i++) {
	243	end_page_writeback(pages[i]);
	244	page_cache_release(pages[i]);
	245	}
	246	nr_pages -= ret;
	247	index += ret;
	248	}
	249	/* the inode may be gone now */
	250	return 0;
	251	}
	252
	253	/*
	254	* do the cleanup once all the compressed pages hit the disk.
	255	* This will clear writeback on the file pages and free the compressed
	256	* pages.
	257	*
	258	* This also calls the writeback end hooks for the file pages so that
	259	* metadata and checksums can be updated in the file.
	260	*/
	261	static void end_compressed_bio_write(struct bio *bio, int err)
	262	{
	263	struct extent_io_tree *tree;
	264	struct compressed_bio *cb = bio->bi_private;
	265	struct inode *inode;
	266	struct page *page;
	267	unsigned long index;
	268
	269	if (err)
	270	cb->errors = 1;
	271
	272	/* if there are more bios still pending for this compressed
	273	* extent, just exit
	274	*/
	275	if (!atomic_dec_and_test(&cb->pending_bios))
	276	goto out;
	277
	278	/* ok, we're the last bio for this extent, step one is to
	279	* call back into the FS and do all the end_io operations
	280	*/
	281	inode = cb->inode;
	282	tree = &BTRFS_I(inode)->io_tree;
70b99e69	283	cb->compressed_pages[0]->mapping = cb->inode->i_mapping;
c8b97818 CM	284	tree->ops->writepage_end_io_hook(cb->compressed_pages[0],
	285	cb->start,
	286	cb->start + cb->len - 1,
	287	NULL, 1);
70b99e69	288	cb->compressed_pages[0]->mapping = NULL;
c8b97818 CM	289
	290	end_compressed_writeback(inode, cb->start, cb->len);
	291	/* note, our inode could be gone now */
	292
	293	/*
	294	* release the compressed pages, these came from alloc_page and
	295	* are not attached to the inode at all
	296	*/
	297	index = 0;
	298	for (index = 0; index < cb->nr_pages; index++) {
	299	page = cb->compressed_pages[index];
	300	page->mapping = NULL;
	301	page_cache_release(page);
	302	}
	303
	304	/* finally free the cb struct */
	305	kfree(cb->compressed_pages);
	306	kfree(cb);
	307	out:
	308	bio_put(bio);
	309	}
	310
	311	/*
	312	* worker function to build and submit bios for previously compressed pages.
	313	* The corresponding pages in the inode should be marked for writeback
	314	* and the compressed pages should have a reference on them for dropping
	315	* when the IO is complete.
	316	*
	317	* This also checksums the file bytes and gets things ready for
	318	* the end io hooks.
	319	*/
	320	int btrfs_submit_compressed_write(struct inode *inode, u64 start,
	321	unsigned long len, u64 disk_start,
	322	unsigned long compressed_len,
	323	struct page **compressed_pages,
	324	unsigned long nr_pages)
	325	{
	326	struct bio *bio = NULL;
	327	struct btrfs_root *root = BTRFS_I(inode)->root;
	328	struct compressed_bio *cb;
	329	unsigned long bytes_left;
	330	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
	331	int page_index = 0;
	332	struct page *page;
	333	u64 first_byte = disk_start;
	334	struct block_device *bdev;
	335	int ret;
	336
	337	WARN_ON(start & ((u64)PAGE_CACHE_SIZE - 1));
d20f7043	338	cb = kmalloc(compressed_bio_size(root, compressed_len), GFP_NOFS);
c8b97818 CM	339	atomic_set(&cb->pending_bios, 0);
	340	cb->errors = 0;
	341	cb->inode = inode;
	342	cb->start = start;
	343	cb->len = len;
d20f7043	344	cb->mirror_num = 0;
c8b97818 CM	345	cb->compressed_pages = compressed_pages;
	346	cb->compressed_len = compressed_len;
	347	cb->orig_bio = NULL;
	348	cb->nr_pages = nr_pages;
	349
	350	bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
	351
c8b97818 CM	352	bio = compressed_bio_alloc(bdev, first_byte, GFP_NOFS);
	353	bio->bi_private = cb;
	354	bio->bi_end_io = end_compressed_bio_write;
	355	atomic_inc(&cb->pending_bios);
	356
	357	/* create and submit bios for the compressed pages */
	358	bytes_left = compressed_len;
cfbc246e	359	for (page_index = 0; page_index < cb->nr_pages; page_index++) {
c8b97818 CM	360	page = compressed_pages[page_index];
	361	page->mapping = inode->i_mapping;
	362	if (bio->bi_size)
	363	ret = io_tree->ops->merge_bio_hook(page, 0,
	364	PAGE_CACHE_SIZE,
	365	bio, 0);
	366	else
	367	ret = 0;
	368
70b99e69	369	page->mapping = NULL;
c8b97818 CM	370	if (ret \|\| bio_add_page(bio, page, PAGE_CACHE_SIZE, 0) <
	371	PAGE_CACHE_SIZE) {
	372	bio_get(bio);
	373
af09abfe CM	374	/*
	375	* inc the count before we submit the bio so
	376	* we know the end IO handler won't happen before
	377	* we inc the count. Otherwise, the cb might get
	378	* freed before we're done setting it up
	379	*/
	380	atomic_inc(&cb->pending_bios);
c8b97818 CM	381	ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
	382	BUG_ON(ret);
	383
d20f7043 CM	384	ret = btrfs_csum_one_bio(root, inode, bio, start, 1);
	385	BUG_ON(ret);
	386
c8b97818 CM	387	ret = btrfs_map_bio(root, WRITE, bio, 0, 1);
	388	BUG_ON(ret);
	389
	390	bio_put(bio);
	391
	392	bio = compressed_bio_alloc(bdev, first_byte, GFP_NOFS);
c8b97818 CM	393	bio->bi_private = cb;
	394	bio->bi_end_io = end_compressed_bio_write;
	395	bio_add_page(bio, page, PAGE_CACHE_SIZE, 0);
	396	}
cfbc246e CM	397	if (bytes_left < PAGE_CACHE_SIZE) {
	398	printk("bytes left %lu compress len %lu nr %lu\n",
	399	bytes_left, cb->compressed_len, cb->nr_pages);
	400	}
c8b97818 CM	401	bytes_left -= PAGE_CACHE_SIZE;
c8b97818 CM	402	first_byte += PAGE_CACHE_SIZE;
771ed689	403	cond_resched();
c8b97818 CM	404	}
	405	bio_get(bio);
	406
	407	ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
	408	BUG_ON(ret);
	409
d20f7043 CM	410	ret = btrfs_csum_one_bio(root, inode, bio, start, 1);
	411	BUG_ON(ret);
	412
c8b97818 CM	413	ret = btrfs_map_bio(root, WRITE, bio, 0, 1);
	414	BUG_ON(ret);
	415
	416	bio_put(bio);
	417	return 0;
	418	}
	419
771ed689 CM	420	static noinline int add_ra_bio_pages(struct inode *inode,
	421	u64 compressed_end,
	422	struct compressed_bio *cb)
	423	{
	424	unsigned long end_index;
	425	unsigned long page_index;
	426	u64 last_offset;
	427	u64 isize = i_size_read(inode);
	428	int ret;
	429	struct page *page;
	430	unsigned long nr_pages = 0;
	431	struct extent_map *em;
	432	struct address_space *mapping = inode->i_mapping;
771ed689 CM	433	struct extent_map_tree *em_tree;
	434	struct extent_io_tree *tree;
	435	u64 end;
	436	int misses = 0;
	437
	438	page = cb->orig_bio->bi_io_vec[cb->orig_bio->bi_vcnt - 1].bv_page;
	439	last_offset = (page_offset(page) + PAGE_CACHE_SIZE);
	440	em_tree = &BTRFS_I(inode)->extent_tree;
	441	tree = &BTRFS_I(inode)->io_tree;
	442
	443	if (isize == 0)
	444	return 0;
	445
	446	end_index = (i_size_read(inode) - 1) >> PAGE_CACHE_SHIFT;
	447
d397712b	448	while (last_offset < compressed_end) {
771ed689 CM	449	page_index = last_offset >> PAGE_CACHE_SHIFT;
	450
	451	if (page_index > end_index)
	452	break;
	453
	454	rcu_read_lock();
	455	page = radix_tree_lookup(&mapping->page_tree, page_index);
	456	rcu_read_unlock();
	457	if (page) {
	458	misses++;
	459	if (misses > 4)
	460	break;
	461	goto next;
	462	}
	463
28ecb609 NP	464	page = __page_cache_alloc(mapping_gfp_mask(mapping) &
28ecb609 NP	465	~__GFP_FS);
771ed689 CM	466	if (!page)
	467	break;
	468
28ecb609 NP	469	if (add_to_page_cache_lru(page, mapping, page_index,
28ecb609 NP	470	GFP_NOFS)) {
771ed689 CM	471	page_cache_release(page);
	472	goto next;
	473	}
	474
771ed689 CM	475	end = last_offset + PAGE_CACHE_SIZE - 1;
	476	/*
	477	* at this point, we have a locked page in the page cache
	478	* for these bytes in the file. But, we have to make
	479	* sure they map to this compressed extent on disk.
	480	*/
	481	set_page_extent_mapped(page);
	482	lock_extent(tree, last_offset, end, GFP_NOFS);
890871be	483	read_lock(&em_tree->lock);
771ed689 CM	484	em = lookup_extent_mapping(em_tree, last_offset,
771ed689 CM	485	PAGE_CACHE_SIZE);
890871be	486	read_unlock(&em_tree->lock);
771ed689 CM	487
	488	if (!em \|\| last_offset < em->start \|\|
	489	(last_offset + PAGE_CACHE_SIZE > extent_map_end(em)) \|\|
	490	(em->block_start >> 9) != cb->orig_bio->bi_sector) {
	491	free_extent_map(em);
	492	unlock_extent(tree, last_offset, end, GFP_NOFS);
	493	unlock_page(page);
	494	page_cache_release(page);
	495	break;
	496	}
	497	free_extent_map(em);
	498
	499	if (page->index == end_index) {
	500	char *userpage;
	501	size_t zero_offset = isize & (PAGE_CACHE_SIZE - 1);
	502
	503	if (zero_offset) {
	504	int zeros;
	505	zeros = PAGE_CACHE_SIZE - zero_offset;
	506	userpage = kmap_atomic(page, KM_USER0);
	507	memset(userpage + zero_offset, 0, zeros);
	508	flush_dcache_page(page);
	509	kunmap_atomic(userpage, KM_USER0);
	510	}
	511	}
	512
	513	ret = bio_add_page(cb->orig_bio, page,
	514	PAGE_CACHE_SIZE, 0);
	515
	516	if (ret == PAGE_CACHE_SIZE) {
	517	nr_pages++;
	518	page_cache_release(page);
	519	} else {
	520	unlock_extent(tree, last_offset, end, GFP_NOFS);
	521	unlock_page(page);
	522	page_cache_release(page);
	523	break;
	524	}
	525	next:
	526	last_offset += PAGE_CACHE_SIZE;
	527	}
771ed689 CM	528	return 0;
	529	}
	530
c8b97818 CM	531	/*
	532	* for a compressed read, the bio we get passed has all the inode pages
	533	* in it. We don't actually do IO on those pages but allocate new ones
	534	* to hold the compressed pages on disk.
	535	*
	536	* bio->bi_sector points to the compressed extent on disk
	537	* bio->bi_io_vec points to all of the inode pages
	538	* bio->bi_vcnt is a count of pages
	539	*
	540	* After the compressed pages are read, we copy the bytes into the
	541	* bio we were passed and then call the bio end_io calls
	542	*/
	543	int btrfs_submit_compressed_read(struct inode inode, struct bio bio,
	544	int mirror_num, unsigned long bio_flags)
	545	{
	546	struct extent_io_tree *tree;
	547	struct extent_map_tree *em_tree;
	548	struct compressed_bio *cb;
	549	struct btrfs_root *root = BTRFS_I(inode)->root;
	550	unsigned long uncompressed_len = bio->bi_vcnt * PAGE_CACHE_SIZE;
	551	unsigned long compressed_len;
	552	unsigned long nr_pages;
	553	unsigned long page_index;
	554	struct page *page;
	555	struct block_device *bdev;
	556	struct bio *comp_bio;
	557	u64 cur_disk_byte = (u64)bio->bi_sector << 9;
e04ca626 CM	558	u64 em_len;
e04ca626 CM	559	u64 em_start;
c8b97818 CM	560	struct extent_map *em;
c8b97818 CM	561	int ret;
d20f7043	562	u32 *sums;
c8b97818 CM	563
	564	tree = &BTRFS_I(inode)->io_tree;
	565	em_tree = &BTRFS_I(inode)->extent_tree;
	566
	567	/* we need the actual starting offset of this extent in the file */
890871be	568	read_lock(&em_tree->lock);
c8b97818 CM	569	em = lookup_extent_mapping(em_tree,
	570	page_offset(bio->bi_io_vec->bv_page),
	571	PAGE_CACHE_SIZE);
890871be	572	read_unlock(&em_tree->lock);
c8b97818	573
d20f7043 CM	574	compressed_len = em->block_len;
d20f7043 CM	575	cb = kmalloc(compressed_bio_size(root, compressed_len), GFP_NOFS);
c8b97818 CM	576	atomic_set(&cb->pending_bios, 0);
	577	cb->errors = 0;
	578	cb->inode = inode;
d20f7043 CM	579	cb->mirror_num = mirror_num;
d20f7043 CM	580	sums = &cb->sums;
c8b97818	581
ff5b7ee3	582	cb->start = em->orig_start;
e04ca626 CM	583	em_len = em->len;
e04ca626 CM	584	em_start = em->start;
d20f7043	585
c8b97818	586	free_extent_map(em);
e04ca626	587	em = NULL;
c8b97818 CM	588
	589	cb->len = uncompressed_len;
	590	cb->compressed_len = compressed_len;
	591	cb->orig_bio = bio;
	592
	593	nr_pages = (compressed_len + PAGE_CACHE_SIZE - 1) /
	594	PAGE_CACHE_SIZE;
	595	cb->compressed_pages = kmalloc(sizeof(struct page ) nr_pages,
	596	GFP_NOFS);
	597	bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
	598
	599	for (page_index = 0; page_index < nr_pages; page_index++) {
	600	cb->compressed_pages[page_index] = alloc_page(GFP_NOFS \|
	601	__GFP_HIGHMEM);
	602	}
	603	cb->nr_pages = nr_pages;
	604
e04ca626	605	add_ra_bio_pages(inode, em_start + em_len, cb);
771ed689	606
771ed689 CM	607	/* include any pages we added in add_ra-bio_pages */
	608	uncompressed_len = bio->bi_vcnt * PAGE_CACHE_SIZE;
	609	cb->len = uncompressed_len;
	610
c8b97818 CM	611	comp_bio = compressed_bio_alloc(bdev, cur_disk_byte, GFP_NOFS);
	612	comp_bio->bi_private = cb;
	613	comp_bio->bi_end_io = end_compressed_bio_read;
	614	atomic_inc(&cb->pending_bios);
	615
	616	for (page_index = 0; page_index < nr_pages; page_index++) {
	617	page = cb->compressed_pages[page_index];
	618	page->mapping = inode->i_mapping;
d20f7043 CM	619	page->index = em_start >> PAGE_CACHE_SHIFT;
d20f7043 CM	620
c8b97818 CM	621	if (comp_bio->bi_size)
	622	ret = tree->ops->merge_bio_hook(page, 0,
	623	PAGE_CACHE_SIZE,
	624	comp_bio, 0);
	625	else
	626	ret = 0;
	627
70b99e69	628	page->mapping = NULL;
c8b97818 CM	629	if (ret \|\| bio_add_page(comp_bio, page, PAGE_CACHE_SIZE, 0) <
	630	PAGE_CACHE_SIZE) {
	631	bio_get(comp_bio);
	632
	633	ret = btrfs_bio_wq_end_io(root->fs_info, comp_bio, 0);
	634	BUG_ON(ret);
	635
af09abfe CM	636	/*
	637	* inc the count before we submit the bio so
	638	* we know the end IO handler won't happen before
	639	* we inc the count. Otherwise, the cb might get
	640	* freed before we're done setting it up
	641	*/
	642	atomic_inc(&cb->pending_bios);
	643
6cbff00f	644	if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
d20f7043 CM	645	btrfs_lookup_bio_sums(root, inode, comp_bio,
	646	sums);
	647	}
	648	sums += (comp_bio->bi_size + root->sectorsize - 1) /
	649	root->sectorsize;
	650
	651	ret = btrfs_map_bio(root, READ, comp_bio,
	652	mirror_num, 0);
c8b97818 CM	653	BUG_ON(ret);
	654
	655	bio_put(comp_bio);
	656
	657	comp_bio = compressed_bio_alloc(bdev, cur_disk_byte,
	658	GFP_NOFS);
771ed689 CM	659	comp_bio->bi_private = cb;
	660	comp_bio->bi_end_io = end_compressed_bio_read;
	661
	662	bio_add_page(comp_bio, page, PAGE_CACHE_SIZE, 0);
c8b97818 CM	663	}
	664	cur_disk_byte += PAGE_CACHE_SIZE;
	665	}
	666	bio_get(comp_bio);
	667
	668	ret = btrfs_bio_wq_end_io(root->fs_info, comp_bio, 0);
	669	BUG_ON(ret);
	670
6cbff00f	671	if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM))
d20f7043	672	btrfs_lookup_bio_sums(root, inode, comp_bio, sums);
d20f7043 CM	673
d20f7043 CM	674	ret = btrfs_map_bio(root, READ, comp_bio, mirror_num, 0);
c8b97818 CM	675	BUG_ON(ret);
	676
	677	bio_put(comp_bio);
	678	return 0;
	679	}