[linux-2.6-block.git] / fs / nilfs2 / btnode.c

// SPDX-License-Identifier: GPL-2.0+
/*
 * btnode.c - NILFS B-tree node cache
 *
 * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
 *
 * Originally written by Seiji Kihara.
 * Fully revised by Ryusuke Konishi for stabilization and simplification.
 *
 */

#include <linux/types.h>
#include <linux/buffer_head.h>
#include <linux/mm.h>
#include <linux/backing-dev.h>
#include <linux/gfp.h>
#include "nilfs.h"
#include "mdt.h"
#include "dat.h"
#include "page.h"
#include "btnode.h"

void nilfs_btnode_cache_clear(struct address_space *btnc)
{
	invalidate_mapping_pages(btnc, 0, -1);
	truncate_inode_pages(btnc, 0);
}

struct buffer_head *
nilfs_btnode_create_block(struct address_space *btnc, __u64 blocknr)
{
	struct inode *inode = NILFS_BTNC_I(btnc);
	struct buffer_head *bh;

	bh = nilfs_grab_buffer(inode, btnc, blocknr, BIT(BH_NILFS_Node));
	if (unlikely(!bh))
		return NULL;

	if (unlikely(buffer_mapped(bh) || buffer_uptodate(bh) ||
		     buffer_dirty(bh))) {
		brelse(bh);
		BUG();
	}
	memset(bh->b_data, 0, i_blocksize(inode));
	bh->b_bdev = inode->i_sb->s_bdev;
	bh->b_blocknr = blocknr;
	set_buffer_mapped(bh);
	set_buffer_uptodate(bh);

	unlock_page(bh->b_page);
	put_page(bh->b_page);
	return bh;
}

int nilfs_btnode_submit_block(struct address_space *btnc, __u64 blocknr,
			      sector_t pblocknr, int mode, int mode_flags,
			      struct buffer_head **pbh, sector_t *submit_ptr)
{
	struct buffer_head *bh;
	struct inode *inode = NILFS_BTNC_I(btnc);
	struct page *page;
	int err;

	bh = nilfs_grab_buffer(inode, btnc, blocknr, BIT(BH_NILFS_Node));
	if (unlikely(!bh))
		return -ENOMEM;

	err = -EEXIST; /* internal code */
	page = bh->b_page;

	if (buffer_uptodate(bh) || buffer_dirty(bh))
		goto found;

	if (pblocknr == 0) {
		pblocknr = blocknr;
		if (inode->i_ino != NILFS_DAT_INO) {
			struct the_nilfs *nilfs = inode->i_sb->s_fs_info;

			/* blocknr is a virtual block number */
			err = nilfs_dat_translate(nilfs->ns_dat, blocknr,
						  &pblocknr);
			if (unlikely(err)) {
				brelse(bh);
				goto out_locked;
			}
		}
	}

	if (mode_flags & REQ_RAHEAD) {
		if (pblocknr != *submit_ptr + 1 || !trylock_buffer(bh)) {
			err = -EBUSY; /* internal code */
			brelse(bh);
			goto out_locked;
		}
	} else { /* mode == READ */
		lock_buffer(bh);
	}
	if (buffer_uptodate(bh)) {
		unlock_buffer(bh);
		err = -EEXIST; /* internal code */
		goto found;
	}
	set_buffer_mapped(bh);
	bh->b_bdev = inode->i_sb->s_bdev;
	bh->b_blocknr = pblocknr; /* set block address for read */
	bh->b_end_io = end_buffer_read_sync;
	get_bh(bh);
	submit_bh(mode, mode_flags, bh);
	bh->b_blocknr = blocknr; /* set back to the given block address */
	*submit_ptr = pblocknr;
	err = 0;
found:
	*pbh = bh;

out_locked:
	unlock_page(page);
	put_page(page);
	return err;
}

/**
 * nilfs_btnode_delete - delete B-tree node buffer
 * @bh: buffer to be deleted
 *
 * nilfs_btnode_delete() invalidates the specified buffer and delete the page
 * including the buffer if the page gets unbusy.
 */
void nilfs_btnode_delete(struct buffer_head *bh)
{
	struct address_space *mapping;
	struct page *page = bh->b_page;
	pgoff_t index = page_index(page);
	int still_dirty;

	get_page(page);
	lock_page(page);
	wait_on_page_writeback(page);

	nilfs_forget_buffer(bh);
	still_dirty = PageDirty(page);
	mapping = page->mapping;
	unlock_page(page);
	put_page(page);

	if (!still_dirty && mapping)
		invalidate_inode_pages2_range(mapping, index, index);
}

/**
 * nilfs_btnode_prepare_change_key
 *  prepare to move contents of the block for old key to one of new key.
 *  the old buffer will not be removed, but might be reused for new buffer.
 *  it might return -ENOMEM because of memory allocation errors,
 *  and might return -EIO because of disk read errors.
 */
int nilfs_btnode_prepare_change_key(struct address_space *btnc,
				    struct nilfs_btnode_chkey_ctxt *ctxt)
{
	struct buffer_head *obh, *nbh;
	struct inode *inode = NILFS_BTNC_I(btnc);
	__u64 oldkey = ctxt->oldkey, newkey = ctxt->newkey;
	int err;

	if (oldkey == newkey)
		return 0;

	obh = ctxt->bh;
	ctxt->newbh = NULL;

	if (inode->i_blkbits == PAGE_SHIFT) {
		lock_page(obh->b_page);
		/*
		 * We cannot call radix_tree_preload for the kernels older
		 * than 2.6.23, because it is not exported for modules.
		 */
retry:
		err = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
		if (err)
			goto failed_unlock;
		/* BUG_ON(oldkey != obh->b_page->index); */
		if (unlikely(oldkey != obh->b_page->index))
			NILFS_PAGE_BUG(obh->b_page,
				       "invalid oldkey %lld (newkey=%lld)",
				       (unsigned long long)oldkey,
				       (unsigned long long)newkey);

		xa_lock_irq(&btnc->i_pages);
		err = radix_tree_insert(&btnc->i_pages, newkey, obh->b_page);
		xa_unlock_irq(&btnc->i_pages);
		/*
		 * Note: page->index will not change to newkey until
		 * nilfs_btnode_commit_change_key() will be called.
		 * To protect the page in intermediate state, the page lock
		 * is held.
		 */
		radix_tree_preload_end();
		if (!err)
			return 0;
		else if (err != -EEXIST)
			goto failed_unlock;

		err = invalidate_inode_pages2_range(btnc, newkey, newkey);
		if (!err)
			goto retry;
		/* fallback to copy mode */
		unlock_page(obh->b_page);
	}

	nbh = nilfs_btnode_create_block(btnc, newkey);
	if (!nbh)
		return -ENOMEM;

	BUG_ON(nbh == obh);
	ctxt->newbh = nbh;
	return 0;

 failed_unlock:
	unlock_page(obh->b_page);
	return err;
}

/**
 * nilfs_btnode_commit_change_key
 *  commit the change_key operation prepared by prepare_change_key().
 */
void nilfs_btnode_commit_change_key(struct address_space *btnc,
				    struct nilfs_btnode_chkey_ctxt *ctxt)
{
	struct buffer_head *obh = ctxt->bh, *nbh = ctxt->newbh;
	__u64 oldkey = ctxt->oldkey, newkey = ctxt->newkey;
	struct page *opage;

	if (oldkey == newkey)
		return;

	if (nbh == NULL) {	/* blocksize == pagesize */
		opage = obh->b_page;
		if (unlikely(oldkey != opage->index))
			NILFS_PAGE_BUG(opage,
				       "invalid oldkey %lld (newkey=%lld)",
				       (unsigned long long)oldkey,
				       (unsigned long long)newkey);
		mark_buffer_dirty(obh);

		xa_lock_irq(&btnc->i_pages);
		radix_tree_delete(&btnc->i_pages, oldkey);
		radix_tree_tag_set(&btnc->i_pages, newkey,
				   PAGECACHE_TAG_DIRTY);
		xa_unlock_irq(&btnc->i_pages);

		opage->index = obh->b_blocknr = newkey;
		unlock_page(opage);
	} else {
		nilfs_copy_buffer(nbh, obh);
		mark_buffer_dirty(nbh);

		nbh->b_blocknr = newkey;
		ctxt->bh = nbh;
		nilfs_btnode_delete(obh); /* will decrement bh->b_count */
	}
}

/**
 * nilfs_btnode_abort_change_key
 *  abort the change_key operation prepared by prepare_change_key().
 */
void nilfs_btnode_abort_change_key(struct address_space *btnc,
				   struct nilfs_btnode_chkey_ctxt *ctxt)
{
	struct buffer_head *nbh = ctxt->newbh;
	__u64 oldkey = ctxt->oldkey, newkey = ctxt->newkey;

	if (oldkey == newkey)
		return;

	if (nbh == NULL) {	/* blocksize == pagesize */
		xa_lock_irq(&btnc->i_pages);
		radix_tree_delete(&btnc->i_pages, newkey);
		xa_unlock_irq(&btnc->i_pages);
		unlock_page(ctxt->bh->b_page);
	} else
		brelse(nbh);
}
Commit	Line	Data
ae98043f	1	// SPDX-License-Identifier: GPL-2.0+
a60be987 RK	2	/*
	3	* btnode.c - NILFS B-tree node cache
	4	*
	5	* Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
	6	*
4b420ab4 RK	7	* Originally written by Seiji Kihara.
4b420ab4 RK	8	* Fully revised by Ryusuke Konishi for stabilization and simplification.
a60be987 RK	9	*
	10	*/
	11
	12	#include <linux/types.h>
	13	#include <linux/buffer_head.h>
	14	#include <linux/mm.h>
	15	#include <linux/backing-dev.h>
5a0e3ad6	16	#include <linux/gfp.h>
a60be987 RK	17	#include "nilfs.h"
	18	#include "mdt.h"
	19	#include "dat.h"
	20	#include "page.h"
	21	#include "btnode.h"
	22
a60be987 RK	23	void nilfs_btnode_cache_clear(struct address_space *btnc)
	24	{
	25	invalidate_mapping_pages(btnc, 0, -1);
	26	truncate_inode_pages(btnc, 0);
	27	}
	28
d501d736 RK	29	struct buffer_head *
	30	nilfs_btnode_create_block(struct address_space *btnc, __u64 blocknr)
	31	{
	32	struct inode *inode = NILFS_BTNC_I(btnc);
	33	struct buffer_head *bh;
	34
4ce5c342	35	bh = nilfs_grab_buffer(inode, btnc, blocknr, BIT(BH_NILFS_Node));
d501d736 RK	36	if (unlikely(!bh))
	37	return NULL;
	38
	39	if (unlikely(buffer_mapped(bh) \|\| buffer_uptodate(bh) \|\|
	40	buffer_dirty(bh))) {
	41	brelse(bh);
	42	BUG();
	43	}
93407472	44	memset(bh->b_data, 0, i_blocksize(inode));
0ef28f9a	45	bh->b_bdev = inode->i_sb->s_bdev;
d501d736 RK	46	bh->b_blocknr = blocknr;
	47	set_buffer_mapped(bh);
	48	set_buffer_uptodate(bh);
	49
	50	unlock_page(bh->b_page);
09cbfeaf	51	put_page(bh->b_page);
d501d736 RK	52	return bh;
	53	}
	54
a60be987	55	int nilfs_btnode_submit_block(struct address_space *btnc, __u64 blocknr,
2a222ca9	56	sector_t pblocknr, int mode, int mode_flags,
26dfdd8e	57	struct buffer_head *pbh, sector_t submit_ptr)
a60be987 RK	58	{
	59	struct buffer_head *bh;
	60	struct inode *inode = NILFS_BTNC_I(btnc);
f8e6cc01	61	struct page *page;
a60be987 RK	62	int err;
a60be987 RK	63
4ce5c342	64	bh = nilfs_grab_buffer(inode, btnc, blocknr, BIT(BH_NILFS_Node));
a60be987 RK	65	if (unlikely(!bh))
	66	return -ENOMEM;
	67
	68	err = -EEXIST; /* internal code */
f8e6cc01	69	page = bh->b_page;
a60be987 RK	70
	71	if (buffer_uptodate(bh) \|\| buffer_dirty(bh))
	72	goto found;
	73
	74	if (pblocknr == 0) {
	75	pblocknr = blocknr;
	76	if (inode->i_ino != NILFS_DAT_INO) {
0ef28f9a	77	struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
a60be987 RK	78
a60be987 RK	79	/* blocknr is a virtual block number */
0ef28f9a RK	80	err = nilfs_dat_translate(nilfs->ns_dat, blocknr,
0ef28f9a RK	81	&pblocknr);
a60be987 RK	82	if (unlikely(err)) {
	83	brelse(bh);
	84	goto out_locked;
	85	}
	86	}
	87	}
26dfdd8e	88
2a222ca9	89	if (mode_flags & REQ_RAHEAD) {
26dfdd8e RK	90	if (pblocknr != *submit_ptr + 1 \|\| !trylock_buffer(bh)) {
	91	err = -EBUSY; /* internal code */
	92	brelse(bh);
	93	goto out_locked;
	94	}
	95	} else { /* mode == READ */
	96	lock_buffer(bh);
	97	}
a60be987 RK	98	if (buffer_uptodate(bh)) {
	99	unlock_buffer(bh);
	100	err = -EEXIST; /* internal code */
	101	goto found;
	102	}
	103	set_buffer_mapped(bh);
0ef28f9a	104	bh->b_bdev = inode->i_sb->s_bdev;
a60be987 RK	105	bh->b_blocknr = pblocknr; /* set block address for read */
	106	bh->b_end_io = end_buffer_read_sync;
	107	get_bh(bh);
2a222ca9	108	submit_bh(mode, mode_flags, bh);
a60be987	109	bh->b_blocknr = blocknr; /* set back to the given block address */
26dfdd8e	110	*submit_ptr = pblocknr;
a60be987 RK	111	err = 0;
	112	found:
	113	*pbh = bh;
	114
	115	out_locked:
f8e6cc01	116	unlock_page(page);
09cbfeaf	117	put_page(page);
a60be987 RK	118	return err;
	119	}
	120
a60be987 RK	121	/**
	122	* nilfs_btnode_delete - delete B-tree node buffer
	123	* @bh: buffer to be deleted
	124	*
	125	* nilfs_btnode_delete() invalidates the specified buffer and delete the page
	126	* including the buffer if the page gets unbusy.
	127	*/
	128	void nilfs_btnode_delete(struct buffer_head *bh)
	129	{
	130	struct address_space *mapping;
	131	struct page *page = bh->b_page;
	132	pgoff_t index = page_index(page);
	133	int still_dirty;
	134
09cbfeaf	135	get_page(page);
a60be987 RK	136	lock_page(page);
	137	wait_on_page_writeback(page);
	138
	139	nilfs_forget_buffer(bh);
	140	still_dirty = PageDirty(page);
	141	mapping = page->mapping;
	142	unlock_page(page);
09cbfeaf	143	put_page(page);
a60be987 RK	144
	145	if (!still_dirty && mapping)
	146	invalidate_inode_pages2_range(mapping, index, index);
	147	}
	148
	149	/**
	150	* nilfs_btnode_prepare_change_key
	151	* prepare to move contents of the block for old key to one of new key.
	152	* the old buffer will not be removed, but might be reused for new buffer.
	153	* it might return -ENOMEM because of memory allocation errors,
	154	* and might return -EIO because of disk read errors.
	155	*/
	156	int nilfs_btnode_prepare_change_key(struct address_space *btnc,
	157	struct nilfs_btnode_chkey_ctxt *ctxt)
	158	{
	159	struct buffer_head obh, nbh;
	160	struct inode *inode = NILFS_BTNC_I(btnc);
	161	__u64 oldkey = ctxt->oldkey, newkey = ctxt->newkey;
	162	int err;
	163
	164	if (oldkey == newkey)
	165	return 0;
	166
	167	obh = ctxt->bh;
	168	ctxt->newbh = NULL;
	169
09cbfeaf	170	if (inode->i_blkbits == PAGE_SHIFT) {
a60be987 RK	171	lock_page(obh->b_page);
	172	/*
	173	* We cannot call radix_tree_preload for the kernels older
	174	* than 2.6.23, because it is not exported for modules.
	175	*/
b1f1b8ce	176	retry:
a60be987 RK	177	err = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
	178	if (err)
	179	goto failed_unlock;
	180	/* BUG_ON(oldkey != obh->b_page->index); */
	181	if (unlikely(oldkey != obh->b_page->index))
	182	NILFS_PAGE_BUG(obh->b_page,
	183	"invalid oldkey %lld (newkey=%lld)",
	184	(unsigned long long)oldkey,
	185	(unsigned long long)newkey);
	186
b93b0163 MW	187	xa_lock_irq(&btnc->i_pages);
	188	err = radix_tree_insert(&btnc->i_pages, newkey, obh->b_page);
	189	xa_unlock_irq(&btnc->i_pages);
a60be987 RK	190	/*
	191	* Note: page->index will not change to newkey until
	192	* nilfs_btnode_commit_change_key() will be called.
	193	* To protect the page in intermediate state, the page lock
	194	* is held.
	195	*/
	196	radix_tree_preload_end();
	197	if (!err)
	198	return 0;
	199	else if (err != -EEXIST)
	200	goto failed_unlock;
	201
	202	err = invalidate_inode_pages2_range(btnc, newkey, newkey);
	203	if (!err)
	204	goto retry;
	205	/* fallback to copy mode */
	206	unlock_page(obh->b_page);
	207	}
	208
45f4910b RK	209	nbh = nilfs_btnode_create_block(btnc, newkey);
	210	if (!nbh)
	211	return -ENOMEM;
	212
	213	BUG_ON(nbh == obh);
	214	ctxt->newbh = nbh;
	215	return 0;
a60be987 RK	216
	217	failed_unlock:
	218	unlock_page(obh->b_page);
	219	return err;
	220	}
	221
	222	/**
	223	* nilfs_btnode_commit_change_key
	224	* commit the change_key operation prepared by prepare_change_key().
	225	*/
	226	void nilfs_btnode_commit_change_key(struct address_space *btnc,
	227	struct nilfs_btnode_chkey_ctxt *ctxt)
	228	{
	229	struct buffer_head obh = ctxt->bh, nbh = ctxt->newbh;
	230	__u64 oldkey = ctxt->oldkey, newkey = ctxt->newkey;
	231	struct page *opage;
	232
	233	if (oldkey == newkey)
	234	return;
	235
	236	if (nbh == NULL) { /* blocksize == pagesize */
	237	opage = obh->b_page;
	238	if (unlikely(oldkey != opage->index))
	239	NILFS_PAGE_BUG(opage,
	240	"invalid oldkey %lld (newkey=%lld)",
	241	(unsigned long long)oldkey,
	242	(unsigned long long)newkey);
5fc7b141	243	mark_buffer_dirty(obh);
a60be987	244
b93b0163 MW	245	xa_lock_irq(&btnc->i_pages);
	246	radix_tree_delete(&btnc->i_pages, oldkey);
	247	radix_tree_tag_set(&btnc->i_pages, newkey,
a60be987	248	PAGECACHE_TAG_DIRTY);
b93b0163	249	xa_unlock_irq(&btnc->i_pages);
a60be987 RK	250
	251	opage->index = obh->b_blocknr = newkey;
	252	unlock_page(opage);
	253	} else {
	254	nilfs_copy_buffer(nbh, obh);
5fc7b141	255	mark_buffer_dirty(nbh);
a60be987 RK	256
	257	nbh->b_blocknr = newkey;
	258	ctxt->bh = nbh;
	259	nilfs_btnode_delete(obh); /* will decrement bh->b_count */
	260	}
	261	}
	262
	263	/**
	264	* nilfs_btnode_abort_change_key
	265	* abort the change_key operation prepared by prepare_change_key().
	266	*/
	267	void nilfs_btnode_abort_change_key(struct address_space *btnc,
	268	struct nilfs_btnode_chkey_ctxt *ctxt)
	269	{
	270	struct buffer_head *nbh = ctxt->newbh;
	271	__u64 oldkey = ctxt->oldkey, newkey = ctxt->newkey;
	272
	273	if (oldkey == newkey)
	274	return;
	275
	276	if (nbh == NULL) { /* blocksize == pagesize */
b93b0163 MW	277	xa_lock_irq(&btnc->i_pages);
	278	radix_tree_delete(&btnc->i_pages, newkey);
	279	xa_unlock_irq(&btnc->i_pages);
a60be987 RK	280	unlock_page(ctxt->bh->b_page);
	281	} else
	282	brelse(nbh);
	283	}