xfs_iwalk.c | Explore in Territory

// SPDX-License-Identifier: GPL-2.0-or-later
/*
 * Copyright (C) 2019 Oracle.  All Rights Reserved.
 * Author: Darrick J. Wong <[email protected]>
 */
#include "xfs.h"
#include "xfs_fs.h"
#include "xfs_shared.h"
#include "xfs_format.h"
#include "xfs_log_format.h"
#include "xfs_trans_resv.h"
#include "xfs_mount.h"
#include "xfs_inode.h"
#include "xfs_btree.h"
#include "xfs_ialloc.h"
#include "xfs_ialloc_btree.h"
#include "xfs_iwalk.h"
#include "xfs_error.h"
#include "xfs_trace.h"
#include "xfs_icache.h"
#include "xfs_health.h"
#include "xfs_trans.h"
#include "xfs_pwork.h"
#include "xfs_ag.h"
#include "xfs_bit.h"

/*
 * Walking Inodes in the Filesystem
 * ================================
 *
 * This iterator function walks a subset of filesystem inodes in increasing
 * order from @startino until there are no more inodes.  For each allocated
 * inode it finds, it calls a walk function with the relevant inode number and
 * a pointer to caller-provided data.  The walk function can return the usual
 * negative error code to stop the iteration; 0 to continue the iteration; or
 * -ECANCELED to stop the iteration.  This return value is returned to the
 * caller.
 *
 * Internally, we allow the walk function to do anything, which means that we
 * cannot maintain the inobt cursor or our lock on the AGI buffer.  We
 * therefore cache the inobt records in kernel memory and only call the walk
 * function when our memory buffer is full.  @nr_recs is the number of records
 * that we've cached, and @sz_recs is the size of our cache.
 *
 * It is the responsibility of the walk function to ensure it accesses
 * allocated inodes, as the inobt records may be stale by the time they are
 * acted upon.
 */

struct xfs_iwalk_ag { … };

/*
 * Loop over all clusters in a chunk for a given incore inode allocation btree
 * record.  Do a readahead if there are any allocated inodes in that cluster.
 */
STATIC void
xfs_iwalk_ichunk_ra(
	struct xfs_mount		*mp,
	struct xfs_perag		*pag,
	struct xfs_inobt_rec_incore	*irec)
{ … }

/*
 * Set the bits in @irec's free mask that correspond to the inodes before
 * @agino so that we skip them.  This is how we restart an inode walk that was
 * interrupted in the middle of an inode record.
 */
STATIC void
xfs_iwalk_adjust_start(
	xfs_agino_t			agino,	/* starting inode of chunk */
	struct xfs_inobt_rec_incore	*irec)	/* btree record */
{ … }

/* Allocate memory for a walk. */
STATIC int
xfs_iwalk_alloc(
	struct xfs_iwalk_ag	*iwag)
{ … }

/* Free memory we allocated for a walk. */
STATIC void
xfs_iwalk_free(
	struct xfs_iwalk_ag	*iwag)
{ … }

/* For each inuse inode in each cached inobt record, call our function. */
STATIC int
xfs_iwalk_ag_recs(
	struct xfs_iwalk_ag	*iwag)
{ … }

/* Delete cursor and let go of AGI. */
static inline void
xfs_iwalk_del_inobt(
	struct xfs_trans	*tp,
	struct xfs_btree_cur	**curpp,
	struct xfs_buf		**agi_bpp,
	int			error)
{ … }

/*
 * Set ourselves up for walking inobt records starting from a given point in
 * the filesystem.
 *
 * If caller passed in a nonzero start inode number, load the record from the
 * inobt and make the record look like all the inodes before agino are free so
 * that we skip them, and then move the cursor to the next inobt record.  This
 * is how we support starting an iwalk in the middle of an inode chunk.
 *
 * If the caller passed in a start number of zero, move the cursor to the first
 * inobt record.
 *
 * The caller is responsible for cleaning up the cursor and buffer pointer
 * regardless of the error status.
 */
STATIC int
xfs_iwalk_ag_start(
	struct xfs_iwalk_ag	*iwag,
	xfs_agino_t		agino,
	struct xfs_btree_cur	**curpp,
	struct xfs_buf		**agi_bpp,
	int			*has_more)
{ … }

/*
 * The inobt record cache is full, so preserve the inobt cursor state and
 * run callbacks on the cached inobt records.  When we're done, restore the
 * cursor state to wherever the cursor would have been had the cache not been
 * full (and therefore we could've just incremented the cursor) if *@has_more
 * is true.  On exit, *@has_more will indicate whether or not the caller should
 * try for more inode records.
 */
STATIC int
xfs_iwalk_run_callbacks(
	struct xfs_iwalk_ag		*iwag,
	struct xfs_btree_cur		**curpp,
	struct xfs_buf			**agi_bpp,
	int				*has_more)
{ … }

/* Walk all inodes in a single AG, from @iwag->startino to the end of the AG. */
STATIC int
xfs_iwalk_ag(
	struct xfs_iwalk_ag		*iwag)
{ … }

/*
 * We experimentally determined that the reduction in ioctl call overhead
 * diminishes when userspace asks for more than 2048 inodes, so we'll cap
 * prefetch at this point.
 */
#define IWALK_MAX_INODE_PREFETCH …

/*
 * Given the number of inodes to prefetch, set the number of inobt records that
 * we cache in memory, which controls the number of inodes we try to read
 * ahead.  Set the maximum if @inodes == 0.
 */
static inline unsigned int
xfs_iwalk_prefetch(
	unsigned int		inodes)
{ … }

/*
 * Walk all inodes in the filesystem starting from @startino.  The @iwalk_fn
 * will be called for each allocated inode, being passed the inode's number and
 * @data.  @max_prefetch controls how many inobt records' worth of inodes we
 * try to readahead.
 */
int
xfs_iwalk(
	struct xfs_mount	*mp,
	struct xfs_trans	*tp,
	xfs_ino_t		startino,
	unsigned int		flags,
	xfs_iwalk_fn		iwalk_fn,
	unsigned int		inode_records,
	void			*data)
{ … }

/* Run per-thread iwalk work. */
static int
xfs_iwalk_ag_work(
	struct xfs_mount	*mp,
	struct xfs_pwork	*pwork)
{ … }

/*
 * Walk all the inodes in the filesystem using multiple threads to process each
 * AG.
 */
int
xfs_iwalk_threaded(
	struct xfs_mount	*mp,
	xfs_ino_t		startino,
	unsigned int		flags,
	xfs_iwalk_fn		iwalk_fn,
	unsigned int		inode_records,
	bool			polled,
	void			*data)
{ … }

/*
 * Allow callers to cache up to a page's worth of inobt records.  This reflects
 * the existing inumbers prefetching behavior.  Since the inobt walk does not
 * itself do anything with the inobt records, we can set a fairly high limit
 * here.
 */
#define MAX_INOBT_WALK_PREFETCH …

/*
 * Given the number of records that the user wanted, set the number of inobt
 * records that we buffer in memory.  Set the maximum if @inobt_records == 0.
 */
static inline unsigned int
xfs_inobt_walk_prefetch(
	unsigned int		inobt_records)
{ … }

/*
 * Walk all inode btree records in the filesystem starting from @startino.  The
 * @inobt_walk_fn will be called for each btree record, being passed the incore
 * record and @data.  @max_prefetch controls how many inobt records we try to
 * cache ahead of time.
 */
int
xfs_inobt_walk(
	struct xfs_mount	*mp,
	struct xfs_trans	*tp,
	xfs_ino_t		startino,
	unsigned int		flags,
	xfs_inobt_walk_fn	inobt_walk_fn,
	unsigned int		inobt_records,
	void			*data)
{ … }
linux/fs/xfs/xfs_iwalk.c
