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

// SPDX-License-Identifier: GPL-2.0+
/*
 * 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"


/**
 * nilfs_init_btnc_inode - initialize B-tree node cache inode
 * @btnc_inode: inode to be initialized
 *
 * nilfs_init_btnc_inode() sets up an inode for B-tree node cache.
 */
void nilfs_init_btnc_inode(struct inode *btnc_inode)
{
	struct nilfs_inode_info *ii = NILFS_I(btnc_inode);

	btnc_inode->i_mode = S_IFREG;
	ii->i_flags = 0;
	memset(&ii->i_bmap_data, 0, sizeof(struct nilfs_bmap));
	mapping_set_gfp_mask(btnc_inode->i_mapping, GFP_NOFS);
}

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 = btnc->host;
	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, blk_opf_t opf,
			      struct buffer_head **pbh, sector_t *submit_ptr)
{
	struct buffer_head *bh;
	struct inode *inode = btnc->host;
	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 (opf & REQ_RAHEAD) {
		if (pblocknr != *submit_ptr + 1 || !trylock_buffer(bh)) {
			err = -EBUSY; /* internal code */
			brelse(bh);
			goto out_locked;
		}
	} else { /* opf == REQ_OP_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(opf, 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 = btnc->host;
	__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) {
		struct page *opage = obh->b_page;
		lock_page(opage);
retry:
		/* BUG_ON(oldkey != obh->b_folio->index); */
		if (unlikely(oldkey != opage->index))
			NILFS_PAGE_BUG(opage,
				       "invalid oldkey %lld (newkey=%lld)",
				       (unsigned long long)oldkey,
				       (unsigned long long)newkey);

		xa_lock_irq(&btnc->i_pages);
		err = __xa_insert(&btnc->i_pages, newkey, opage, GFP_NOFS);
		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.
		 */
		if (!err)
			return 0;
		else if (err != -EBUSY)
			goto failed_unlock;

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

	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);
		__xa_erase(&btnc->i_pages, oldkey);
		__xa_set_mark(&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_erase_irq(&btnc->i_pages, newkey);
		unlock_page(ctxt->bh->b_page);
	} else
		brelse(nbh);
}
Commit	Line	Data
ae98043f	1	// SPDX-License-Identifier: GPL-2.0+
a60be987	2	/*
94ee1d91	3	* NILFS B-tree node cache
a60be987 RK	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
e897be17 RK	23
	24	/**
	25	* nilfs_init_btnc_inode - initialize B-tree node cache inode
	26	* @btnc_inode: inode to be initialized
	27	*
	28	* nilfs_init_btnc_inode() sets up an inode for B-tree node cache.
	29	*/
	30	void nilfs_init_btnc_inode(struct inode *btnc_inode)
	31	{
	32	struct nilfs_inode_info *ii = NILFS_I(btnc_inode);
	33
	34	btnc_inode->i_mode = S_IFREG;
	35	ii->i_flags = 0;
	36	memset(&ii->i_bmap_data, 0, sizeof(struct nilfs_bmap));
	37	mapping_set_gfp_mask(btnc_inode->i_mapping, GFP_NOFS);
	38	}
	39
a60be987 RK	40	void nilfs_btnode_cache_clear(struct address_space *btnc)
	41	{
	42	invalidate_mapping_pages(btnc, 0, -1);
	43	truncate_inode_pages(btnc, 0);
	44	}
	45
d501d736 RK	46	struct buffer_head *
	47	nilfs_btnode_create_block(struct address_space *btnc, __u64 blocknr)
	48	{
e897be17	49	struct inode *inode = btnc->host;
d501d736 RK	50	struct buffer_head *bh;
d501d736 RK	51
4ce5c342	52	bh = nilfs_grab_buffer(inode, btnc, blocknr, BIT(BH_NILFS_Node));
d501d736 RK	53	if (unlikely(!bh))
	54	return NULL;
	55
	56	if (unlikely(buffer_mapped(bh) \|\| buffer_uptodate(bh) \|\|
	57	buffer_dirty(bh))) {
	58	brelse(bh);
	59	BUG();
	60	}
93407472	61	memset(bh->b_data, 0, i_blocksize(inode));
0ef28f9a	62	bh->b_bdev = inode->i_sb->s_bdev;
d501d736 RK	63	bh->b_blocknr = blocknr;
	64	set_buffer_mapped(bh);
	65	set_buffer_uptodate(bh);
	66
	67	unlock_page(bh->b_page);
09cbfeaf	68	put_page(bh->b_page);
d501d736 RK	69	return bh;
	70	}
	71
a60be987	72	int nilfs_btnode_submit_block(struct address_space *btnc, __u64 blocknr,
ed451259	73	sector_t pblocknr, blk_opf_t opf,
26dfdd8e	74	struct buffer_head *pbh, sector_t submit_ptr)
a60be987 RK	75	{
a60be987 RK	76	struct buffer_head *bh;
e897be17	77	struct inode *inode = btnc->host;
f8e6cc01	78	struct page *page;
a60be987 RK	79	int err;
a60be987 RK	80
4ce5c342	81	bh = nilfs_grab_buffer(inode, btnc, blocknr, BIT(BH_NILFS_Node));
a60be987 RK	82	if (unlikely(!bh))
	83	return -ENOMEM;
	84
	85	err = -EEXIST; /* internal code */
f8e6cc01	86	page = bh->b_page;
a60be987 RK	87
	88	if (buffer_uptodate(bh) \|\| buffer_dirty(bh))
	89	goto found;
	90
	91	if (pblocknr == 0) {
	92	pblocknr = blocknr;
	93	if (inode->i_ino != NILFS_DAT_INO) {
0ef28f9a	94	struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
a60be987 RK	95
a60be987 RK	96	/* blocknr is a virtual block number */
0ef28f9a RK	97	err = nilfs_dat_translate(nilfs->ns_dat, blocknr,
0ef28f9a RK	98	&pblocknr);
a60be987 RK	99	if (unlikely(err)) {
	100	brelse(bh);
	101	goto out_locked;
	102	}
	103	}
	104	}
26dfdd8e	105
ed451259	106	if (opf & REQ_RAHEAD) {
26dfdd8e RK	107	if (pblocknr != *submit_ptr + 1 \|\| !trylock_buffer(bh)) {
	108	err = -EBUSY; /* internal code */
	109	brelse(bh);
	110	goto out_locked;
	111	}
ed451259	112	} else { /* opf == REQ_OP_READ */
26dfdd8e RK	113	lock_buffer(bh);
26dfdd8e RK	114	}
a60be987 RK	115	if (buffer_uptodate(bh)) {
	116	unlock_buffer(bh);
	117	err = -EEXIST; /* internal code */
	118	goto found;
	119	}
	120	set_buffer_mapped(bh);
0ef28f9a	121	bh->b_bdev = inode->i_sb->s_bdev;
a60be987 RK	122	bh->b_blocknr = pblocknr; /* set block address for read */
	123	bh->b_end_io = end_buffer_read_sync;
	124	get_bh(bh);
ed451259	125	submit_bh(opf, bh);
a60be987	126	bh->b_blocknr = blocknr; /* set back to the given block address */
26dfdd8e	127	*submit_ptr = pblocknr;
a60be987 RK	128	err = 0;
	129	found:
	130	*pbh = bh;
	131
	132	out_locked:
f8e6cc01	133	unlock_page(page);
09cbfeaf	134	put_page(page);
a60be987 RK	135	return err;
	136	}
	137
a60be987 RK	138	/**
	139	* nilfs_btnode_delete - delete B-tree node buffer
	140	* @bh: buffer to be deleted
	141	*
	142	* nilfs_btnode_delete() invalidates the specified buffer and delete the page
	143	* including the buffer if the page gets unbusy.
	144	*/
	145	void nilfs_btnode_delete(struct buffer_head *bh)
	146	{
	147	struct address_space *mapping;
	148	struct page *page = bh->b_page;
	149	pgoff_t index = page_index(page);
	150	int still_dirty;
	151
09cbfeaf	152	get_page(page);
a60be987 RK	153	lock_page(page);
	154	wait_on_page_writeback(page);
	155
	156	nilfs_forget_buffer(bh);
	157	still_dirty = PageDirty(page);
	158	mapping = page->mapping;
	159	unlock_page(page);
09cbfeaf	160	put_page(page);
a60be987 RK	161
	162	if (!still_dirty && mapping)
	163	invalidate_inode_pages2_range(mapping, index, index);
	164	}
	165
	166	/**
	167	* nilfs_btnode_prepare_change_key
	168	* prepare to move contents of the block for old key to one of new key.
	169	* the old buffer will not be removed, but might be reused for new buffer.
	170	* it might return -ENOMEM because of memory allocation errors,
	171	* and might return -EIO because of disk read errors.
	172	*/
	173	int nilfs_btnode_prepare_change_key(struct address_space *btnc,
	174	struct nilfs_btnode_chkey_ctxt *ctxt)
	175	{
	176	struct buffer_head obh, nbh;
e897be17	177	struct inode *inode = btnc->host;
a60be987 RK	178	__u64 oldkey = ctxt->oldkey, newkey = ctxt->newkey;
	179	int err;
	180
	181	if (oldkey == newkey)
	182	return 0;
	183
	184	obh = ctxt->bh;
	185	ctxt->newbh = NULL;
	186
09cbfeaf	187	if (inode->i_blkbits == PAGE_SHIFT) {
f611ff63 MW	188	struct page *opage = obh->b_page;
f611ff63 MW	189	lock_page(opage);
b1f1b8ce	190	retry:
6ad4cd7f	191	/* BUG_ON(oldkey != obh->b_folio->index); */
f611ff63 MW	192	if (unlikely(oldkey != opage->index))
f611ff63 MW	193	NILFS_PAGE_BUG(opage,
a60be987 RK	194	"invalid oldkey %lld (newkey=%lld)",
	195	(unsigned long long)oldkey,
	196	(unsigned long long)newkey);
	197
b93b0163	198	xa_lock_irq(&btnc->i_pages);
f611ff63	199	err = __xa_insert(&btnc->i_pages, newkey, opage, GFP_NOFS);
b93b0163	200	xa_unlock_irq(&btnc->i_pages);
a60be987 RK	201	/*
	202	* Note: page->index will not change to newkey until
	203	* nilfs_btnode_commit_change_key() will be called.
	204	* To protect the page in intermediate state, the page lock
	205	* is held.
	206	*/
a60be987 RK	207	if (!err)
a60be987 RK	208	return 0;
fd9dc93e	209	else if (err != -EBUSY)
a60be987 RK	210	goto failed_unlock;
	211
	212	err = invalidate_inode_pages2_range(btnc, newkey, newkey);
	213	if (!err)
	214	goto retry;
	215	/* fallback to copy mode */
f611ff63	216	unlock_page(opage);
a60be987 RK	217	}
a60be987 RK	218
45f4910b RK	219	nbh = nilfs_btnode_create_block(btnc, newkey);
	220	if (!nbh)
	221	return -ENOMEM;
	222
	223	BUG_ON(nbh == obh);
	224	ctxt->newbh = nbh;
	225	return 0;
a60be987 RK	226
	227	failed_unlock:
	228	unlock_page(obh->b_page);
	229	return err;
	230	}
	231
	232	/**
	233	* nilfs_btnode_commit_change_key
	234	* commit the change_key operation prepared by prepare_change_key().
	235	*/
	236	void nilfs_btnode_commit_change_key(struct address_space *btnc,
	237	struct nilfs_btnode_chkey_ctxt *ctxt)
	238	{
	239	struct buffer_head obh = ctxt->bh, nbh = ctxt->newbh;
	240	__u64 oldkey = ctxt->oldkey, newkey = ctxt->newkey;
	241	struct page *opage;
	242
	243	if (oldkey == newkey)
	244	return;
	245
	246	if (nbh == NULL) { /* blocksize == pagesize */
	247	opage = obh->b_page;
	248	if (unlikely(oldkey != opage->index))
	249	NILFS_PAGE_BUG(opage,
	250	"invalid oldkey %lld (newkey=%lld)",
	251	(unsigned long long)oldkey,
	252	(unsigned long long)newkey);
5fc7b141	253	mark_buffer_dirty(obh);
a60be987	254
b93b0163	255	xa_lock_irq(&btnc->i_pages);
f611ff63 MW	256	__xa_erase(&btnc->i_pages, oldkey);
f611ff63 MW	257	__xa_set_mark(&btnc->i_pages, newkey, PAGECACHE_TAG_DIRTY);
b93b0163	258	xa_unlock_irq(&btnc->i_pages);
a60be987 RK	259
	260	opage->index = obh->b_blocknr = newkey;
	261	unlock_page(opage);
	262	} else {
	263	nilfs_copy_buffer(nbh, obh);
5fc7b141	264	mark_buffer_dirty(nbh);
a60be987 RK	265
	266	nbh->b_blocknr = newkey;
	267	ctxt->bh = nbh;
	268	nilfs_btnode_delete(obh); /* will decrement bh->b_count */
	269	}
	270	}
	271
	272	/**
	273	* nilfs_btnode_abort_change_key
	274	* abort the change_key operation prepared by prepare_change_key().
	275	*/
	276	void nilfs_btnode_abort_change_key(struct address_space *btnc,
	277	struct nilfs_btnode_chkey_ctxt *ctxt)
	278	{
	279	struct buffer_head *nbh = ctxt->newbh;
	280	__u64 oldkey = ctxt->oldkey, newkey = ctxt->newkey;
	281
	282	if (oldkey == newkey)
	283	return;
	284
	285	if (nbh == NULL) { /* blocksize == pagesize */
fe2b5114	286	xa_erase_irq(&btnc->i_pages, newkey);
a60be987 RK	287	unlock_page(ctxt->bh->b_page);
	288	} else
	289	brelse(nbh);
	290	}