Revision d7831a0bdf06b9f722b947bb0c205ff7d77cebd8 authored by Richard Kennedy on 30 June 2009, 18:41:35 UTC, committed by Linus Torvalds on 01 July 2009, 01:56:01 UTC
balance_dirty_pages can overreact and move all of the dirty pages to writeback unnecessarily. balance_dirty_pages makes its decision to throttle based on the number of dirty plus writeback pages that are over the calculated limit,so it will continue to move pages even when there are plenty of pages in writeback and less than the threshold still dirty. This allows it to overshoot its limits and move all the dirty pages to writeback while waiting for the drives to catch up and empty the writeback list. A simple fio test easily demonstrates this problem. fio --name=f1 --directory=/disk1 --size=2G -rw=write --name=f2 --directory=/disk2 --size=1G --rw=write --startdelay=10 This is the simplest fix I could find, but I'm not entirely sure that it alone will be enough for all cases. But it certainly is an improvement on my desktop machine writing to 2 disks. Do we need something more for machines with large arrays where bdi_threshold * number_of_drives is greater than the dirty_ratio ? Signed-off-by: Richard Kennedy <richard@rsk.demon.co.uk> Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Jens Axboe <jens.axboe@oracle.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
1 parent df279ca
bitmap.h
/*
* bitmap.h: Copyright (C) Peter T. Breuer (ptb@ot.uc3m.es) 2003
*
* additions: Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
*/
#ifndef BITMAP_H
#define BITMAP_H 1
#define BITMAP_MAJOR_LO 3
/* version 4 insists the bitmap is in little-endian order
* with version 3, it is host-endian which is non-portable
*/
#define BITMAP_MAJOR_HI 4
#define BITMAP_MAJOR_HOSTENDIAN 3
#define BITMAP_MINOR 39
/*
* in-memory bitmap:
*
* Use 16 bit block counters to track pending writes to each "chunk".
* The 2 high order bits are special-purpose, the first is a flag indicating
* whether a resync is needed. The second is a flag indicating whether a
* resync is active.
* This means that the counter is actually 14 bits:
*
* +--------+--------+------------------------------------------------+
* | resync | resync | counter |
* | needed | active | |
* | (0-1) | (0-1) | (0-16383) |
* +--------+--------+------------------------------------------------+
*
* The "resync needed" bit is set when:
* a '1' bit is read from storage at startup.
* a write request fails on some drives
* a resync is aborted on a chunk with 'resync active' set
* It is cleared (and resync-active set) when a resync starts across all drives
* of the chunk.
*
*
* The "resync active" bit is set when:
* a resync is started on all drives, and resync_needed is set.
* resync_needed will be cleared (as long as resync_active wasn't already set).
* It is cleared when a resync completes.
*
* The counter counts pending write requests, plus the on-disk bit.
* When the counter is '1' and the resync bits are clear, the on-disk
* bit can be cleared aswell, thus setting the counter to 0.
* When we set a bit, or in the counter (to start a write), if the fields is
* 0, we first set the disk bit and set the counter to 1.
*
* If the counter is 0, the on-disk bit is clear and the stipe is clean
* Anything that dirties the stipe pushes the counter to 2 (at least)
* and sets the on-disk bit (lazily).
* If a periodic sweep find the counter at 2, it is decremented to 1.
* If the sweep find the counter at 1, the on-disk bit is cleared and the
* counter goes to zero.
*
* Also, we'll hijack the "map" pointer itself and use it as two 16 bit block
* counters as a fallback when "page" memory cannot be allocated:
*
* Normal case (page memory allocated):
*
* page pointer (32-bit)
*
* [ ] ------+
* |
* +-------> [ ][ ]..[ ] (4096 byte page == 2048 counters)
* c1 c2 c2048
*
* Hijacked case (page memory allocation failed):
*
* hijacked page pointer (32-bit)
*
* [ ][ ] (no page memory allocated)
* counter #1 (16-bit) counter #2 (16-bit)
*
*/
#ifdef __KERNEL__
#define PAGE_BITS (PAGE_SIZE << 3)
#define PAGE_BIT_SHIFT (PAGE_SHIFT + 3)
typedef __u16 bitmap_counter_t;
#define COUNTER_BITS 16
#define COUNTER_BIT_SHIFT 4
#define COUNTER_BYTE_RATIO (COUNTER_BITS / 8)
#define COUNTER_BYTE_SHIFT (COUNTER_BIT_SHIFT - 3)
#define NEEDED_MASK ((bitmap_counter_t) (1 << (COUNTER_BITS - 1)))
#define RESYNC_MASK ((bitmap_counter_t) (1 << (COUNTER_BITS - 2)))
#define COUNTER_MAX ((bitmap_counter_t) RESYNC_MASK - 1)
#define NEEDED(x) (((bitmap_counter_t) x) & NEEDED_MASK)
#define RESYNC(x) (((bitmap_counter_t) x) & RESYNC_MASK)
#define COUNTER(x) (((bitmap_counter_t) x) & COUNTER_MAX)
/* how many counters per page? */
#define PAGE_COUNTER_RATIO (PAGE_BITS / COUNTER_BITS)
/* same, except a shift value for more efficient bitops */
#define PAGE_COUNTER_SHIFT (PAGE_BIT_SHIFT - COUNTER_BIT_SHIFT)
/* same, except a mask value for more efficient bitops */
#define PAGE_COUNTER_MASK (PAGE_COUNTER_RATIO - 1)
#define BITMAP_BLOCK_SIZE 512
#define BITMAP_BLOCK_SHIFT 9
/* how many blocks per chunk? (this is variable) */
#define CHUNK_BLOCK_RATIO(bitmap) ((bitmap)->chunksize >> BITMAP_BLOCK_SHIFT)
#define CHUNK_BLOCK_SHIFT(bitmap) ((bitmap)->chunkshift - BITMAP_BLOCK_SHIFT)
#define CHUNK_BLOCK_MASK(bitmap) (CHUNK_BLOCK_RATIO(bitmap) - 1)
/* when hijacked, the counters and bits represent even larger "chunks" */
/* there will be 1024 chunks represented by each counter in the page pointers */
#define PAGEPTR_BLOCK_RATIO(bitmap) \
(CHUNK_BLOCK_RATIO(bitmap) << PAGE_COUNTER_SHIFT >> 1)
#define PAGEPTR_BLOCK_SHIFT(bitmap) \
(CHUNK_BLOCK_SHIFT(bitmap) + PAGE_COUNTER_SHIFT - 1)
#define PAGEPTR_BLOCK_MASK(bitmap) (PAGEPTR_BLOCK_RATIO(bitmap) - 1)
/*
* on-disk bitmap:
*
* Use one bit per "chunk" (block set). We do the disk I/O on the bitmap
* file a page at a time. There's a superblock at the start of the file.
*/
/* map chunks (bits) to file pages - offset by the size of the superblock */
#define CHUNK_BIT_OFFSET(chunk) ((chunk) + (sizeof(bitmap_super_t) << 3))
#endif
/*
* bitmap structures:
*/
#define BITMAP_MAGIC 0x6d746962
/* use these for bitmap->flags and bitmap->sb->state bit-fields */
enum bitmap_state {
BITMAP_STALE = 0x002, /* the bitmap file is out of date or had -EIO */
BITMAP_WRITE_ERROR = 0x004, /* A write error has occurred */
BITMAP_HOSTENDIAN = 0x8000,
};
/* the superblock at the front of the bitmap file -- little endian */
typedef struct bitmap_super_s {
__le32 magic; /* 0 BITMAP_MAGIC */
__le32 version; /* 4 the bitmap major for now, could change... */
__u8 uuid[16]; /* 8 128 bit uuid - must match md device uuid */
__le64 events; /* 24 event counter for the bitmap (1)*/
__le64 events_cleared;/*32 event counter when last bit cleared (2) */
__le64 sync_size; /* 40 the size of the md device's sync range(3) */
__le32 state; /* 48 bitmap state information */
__le32 chunksize; /* 52 the bitmap chunk size in bytes */
__le32 daemon_sleep; /* 56 seconds between disk flushes */
__le32 write_behind; /* 60 number of outstanding write-behind writes */
__u8 pad[256 - 64]; /* set to zero */
} bitmap_super_t;
/* notes:
* (1) This event counter is updated before the eventcounter in the md superblock
* When a bitmap is loaded, it is only accepted if this event counter is equal
* to, or one greater than, the event counter in the superblock.
* (2) This event counter is updated when the other one is *if*and*only*if* the
* array is not degraded. As bits are not cleared when the array is degraded,
* this represents the last time that any bits were cleared.
* If a device is being added that has an event count with this value or
* higher, it is accepted as conforming to the bitmap.
* (3)This is the number of sectors represented by the bitmap, and is the range that
* resync happens across. For raid1 and raid5/6 it is the size of individual
* devices. For raid10 it is the size of the array.
*/
#ifdef __KERNEL__
/* the in-memory bitmap is represented by bitmap_pages */
struct bitmap_page {
/*
* map points to the actual memory page
*/
char *map;
/*
* in emergencies (when map cannot be alloced), hijack the map
* pointer and use it as two counters itself
*/
unsigned int hijacked:1;
/*
* count of dirty bits on the page
*/
unsigned int count:31;
};
/* keep track of bitmap file pages that have pending writes on them */
struct page_list {
struct list_head list;
struct page *page;
};
/* the main bitmap structure - one per mddev */
struct bitmap {
struct bitmap_page *bp;
unsigned long pages; /* total number of pages in the bitmap */
unsigned long missing_pages; /* number of pages not yet allocated */
mddev_t *mddev; /* the md device that the bitmap is for */
int counter_bits; /* how many bits per block counter */
/* bitmap chunksize -- how much data does each bit represent? */
unsigned long chunksize;
unsigned long chunkshift; /* chunksize = 2^chunkshift (for bitops) */
unsigned long chunks; /* total number of data chunks for the array */
/* We hold a count on the chunk currently being synced, and drop
* it when the last block is started. If the resync is aborted
* midway, we need to be able to drop that count, so we remember
* the counted chunk..
*/
unsigned long syncchunk;
__u64 events_cleared;
int need_sync;
/* bitmap spinlock */
spinlock_t lock;
long offset; /* offset from superblock if file is NULL */
struct file *file; /* backing disk file */
struct page *sb_page; /* cached copy of the bitmap file superblock */
struct page **filemap; /* list of cache pages for the file */
unsigned long *filemap_attr; /* attributes associated w/ filemap pages */
unsigned long file_pages; /* number of pages in the file */
int last_page_size; /* bytes in the last page */
unsigned long flags;
int allclean;
unsigned long max_write_behind; /* write-behind mode */
atomic_t behind_writes;
/*
* the bitmap daemon - periodically wakes up and sweeps the bitmap
* file, cleaning up bits and flushing out pages to disk as necessary
*/
unsigned long daemon_lastrun; /* jiffies of last run */
unsigned long daemon_sleep; /* how many seconds between updates? */
unsigned long last_end_sync; /* when we lasted called end_sync to
* update bitmap with resync progress */
atomic_t pending_writes; /* pending writes to the bitmap file */
wait_queue_head_t write_wait;
wait_queue_head_t overflow_wait;
};
/* the bitmap API */
/* these are used only by md/bitmap */
int bitmap_create(mddev_t *mddev);
void bitmap_flush(mddev_t *mddev);
void bitmap_destroy(mddev_t *mddev);
void bitmap_print_sb(struct bitmap *bitmap);
void bitmap_update_sb(struct bitmap *bitmap);
int bitmap_setallbits(struct bitmap *bitmap);
void bitmap_write_all(struct bitmap *bitmap);
void bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e);
/* these are exported */
int bitmap_startwrite(struct bitmap *bitmap, sector_t offset,
unsigned long sectors, int behind);
void bitmap_endwrite(struct bitmap *bitmap, sector_t offset,
unsigned long sectors, int success, int behind);
int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, int *blocks, int degraded);
void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, int *blocks, int aborted);
void bitmap_close_sync(struct bitmap *bitmap);
void bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector);
void bitmap_unplug(struct bitmap *bitmap);
void bitmap_daemon_work(struct bitmap *bitmap);
#endif
#endif
![swh spinner](/static/img/swh-spinner.gif)
Computing file changes ...