kernel-ark/fs/jfs/jfs_logmgr.h
Dave Kleikamp cbc3d65ebc JFS: Improve sync barrier processing
Under heavy load, hot metadata pages are often locked by non-committed
transactions, making them difficult to flush to disk.  This prevents
the sync point from advancing past a transaction that had modified the
page.

There is a point during the sync barrier processing where all
outstanding transactions have been committed to disk, but no new
transaction have been allowed to proceed.  This is the best time
to write the metadata.

Signed-off-by: Dave Kleikamp <shaggy@austin.ibm.com>
2005-07-27 09:17:57 -05:00

516 lines
15 KiB
C

/*
* Copyright (C) International Business Machines Corp., 2000-2004
* Portions Copyright (C) Christoph Hellwig, 2001-2002
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation; either version 2 of the License, or
* (at your option) any later version.
*
* 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 02111-1307 USA
*/
#ifndef _H_JFS_LOGMGR
#define _H_JFS_LOGMGR
#include "jfs_filsys.h"
#include "jfs_lock.h"
/*
* log manager configuration parameters
*/
/* log page size */
#define LOGPSIZE 4096
#define L2LOGPSIZE 12
#define LOGPAGES 16 /* Log pages per mounted file system */
/*
* log logical volume
*
* a log is used to make the commit operation on journalled
* files within the same logical volume group atomic.
* a log is implemented with a logical volume.
* there is one log per logical volume group.
*
* block 0 of the log logical volume is not used (ipl etc).
* block 1 contains a log "superblock" and is used by logFormat(),
* lmLogInit(), lmLogShutdown(), and logRedo() to record status
* of the log but is not otherwise used during normal processing.
* blocks 2 - (N-1) are used to contain log records.
*
* when a volume group is varied-on-line, logRedo() must have
* been executed before the file systems (logical volumes) in
* the volume group can be mounted.
*/
/*
* log superblock (block 1 of logical volume)
*/
#define LOGSUPER_B 1
#define LOGSTART_B 2
#define LOGMAGIC 0x87654321
#define LOGVERSION 1
#define MAX_ACTIVE 128 /* Max active file systems sharing log */
struct logsuper {
__le32 magic; /* 4: log lv identifier */
__le32 version; /* 4: version number */
__le32 serial; /* 4: log open/mount counter */
__le32 size; /* 4: size in number of LOGPSIZE blocks */
__le32 bsize; /* 4: logical block size in byte */
__le32 l2bsize; /* 4: log2 of bsize */
__le32 flag; /* 4: option */
__le32 state; /* 4: state - see below */
__le32 end; /* 4: addr of last log record set by logredo */
char uuid[16]; /* 16: 128-bit journal uuid */
char label[16]; /* 16: journal label */
struct {
char uuid[16];
} active[MAX_ACTIVE]; /* 2048: active file systems list */
};
#define NULL_UUID "\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0"
/* log flag: commit option (see jfs_filsys.h) */
/* log state */
#define LOGMOUNT 0 /* log mounted by lmLogInit() */
#define LOGREDONE 1 /* log shutdown by lmLogShutdown().
* log redo completed by logredo().
*/
#define LOGWRAP 2 /* log wrapped */
#define LOGREADERR 3 /* log read error detected in logredo() */
/*
* log logical page
*
* (this comment should be rewritten !)
* the header and trailer structures (h,t) will normally have
* the same page and eor value.
* An exception to this occurs when a complete page write is not
* accomplished on a power failure. Since the hardware may "split write"
* sectors in the page, any out of order sequence may occur during powerfail
* and needs to be recognized during log replay. The xor value is
* an "exclusive or" of all log words in the page up to eor. This
* 32 bit eor is stored with the top 16 bits in the header and the
* bottom 16 bits in the trailer. logredo can easily recognize pages
* that were not completed by reconstructing this eor and checking
* the log page.
*
* Previous versions of the operating system did not allow split
* writes and detected partially written records in logredo by
* ordering the updates to the header, trailer, and the move of data
* into the logdata area. The order: (1) data is moved (2) header
* is updated (3) trailer is updated. In logredo, when the header
* differed from the trailer, the header and trailer were reconciled
* as follows: if h.page != t.page they were set to the smaller of
* the two and h.eor and t.eor set to 8 (i.e. empty page). if (only)
* h.eor != t.eor they were set to the smaller of their two values.
*/
struct logpage {
struct { /* header */
__le32 page; /* 4: log sequence page number */
__le16 rsrvd; /* 2: */
__le16 eor; /* 2: end-of-log offset of lasrt record write */
} h;
__le32 data[LOGPSIZE / 4 - 4]; /* log record area */
struct { /* trailer */
__le32 page; /* 4: normally the same as h.page */
__le16 rsrvd; /* 2: */
__le16 eor; /* 2: normally the same as h.eor */
} t;
};
#define LOGPHDRSIZE 8 /* log page header size */
#define LOGPTLRSIZE 8 /* log page trailer size */
/*
* log record
*
* (this comment should be rewritten !)
* jfs uses only "after" log records (only a single writer is allowed
* in a page, pages are written to temporary paging space if
* if they must be written to disk before commit, and i/o is
* scheduled for modified pages to their home location after
* the log records containing the after values and the commit
* record is written to the log on disk, undo discards the copy
* in main-memory.)
*
* a log record consists of a data area of variable length followed by
* a descriptor of fixed size LOGRDSIZE bytes.
* the data area is rounded up to an integral number of 4-bytes and
* must be no longer than LOGPSIZE.
* the descriptor is of size of multiple of 4-bytes and aligned on a
* 4-byte boundary.
* records are packed one after the other in the data area of log pages.
* (sometimes a DUMMY record is inserted so that at least one record ends
* on every page or the longest record is placed on at most two pages).
* the field eor in page header/trailer points to the byte following
* the last record on a page.
*/
/* log record types */
#define LOG_COMMIT 0x8000
#define LOG_SYNCPT 0x4000
#define LOG_MOUNT 0x2000
#define LOG_REDOPAGE 0x0800
#define LOG_NOREDOPAGE 0x0080
#define LOG_NOREDOINOEXT 0x0040
#define LOG_UPDATEMAP 0x0008
#define LOG_NOREDOFILE 0x0001
/* REDOPAGE/NOREDOPAGE log record data type */
#define LOG_INODE 0x0001
#define LOG_XTREE 0x0002
#define LOG_DTREE 0x0004
#define LOG_BTROOT 0x0010
#define LOG_EA 0x0020
#define LOG_ACL 0x0040
#define LOG_DATA 0x0080
#define LOG_NEW 0x0100
#define LOG_EXTEND 0x0200
#define LOG_RELOCATE 0x0400
#define LOG_DIR_XTREE 0x0800 /* Xtree is in directory inode */
/* UPDATEMAP log record descriptor type */
#define LOG_ALLOCXADLIST 0x0080
#define LOG_ALLOCPXDLIST 0x0040
#define LOG_ALLOCXAD 0x0020
#define LOG_ALLOCPXD 0x0010
#define LOG_FREEXADLIST 0x0008
#define LOG_FREEPXDLIST 0x0004
#define LOG_FREEXAD 0x0002
#define LOG_FREEPXD 0x0001
struct lrd {
/*
* type independent area
*/
__le32 logtid; /* 4: log transaction identifier */
__le32 backchain; /* 4: ptr to prev record of same transaction */
__le16 type; /* 2: record type */
__le16 length; /* 2: length of data in record (in byte) */
__le32 aggregate; /* 4: file system lv/aggregate */
/* (16) */
/*
* type dependent area (20)
*/
union {
/*
* COMMIT: commit
*
* transaction commit: no type-dependent information;
*/
/*
* REDOPAGE: after-image
*
* apply after-image;
*
* N.B. REDOPAGE, NOREDOPAGE, and UPDATEMAP must be same format;
*/
struct {
__le32 fileset; /* 4: fileset number */
__le32 inode; /* 4: inode number */
__le16 type; /* 2: REDOPAGE record type */
__le16 l2linesize; /* 2: log2 of line size */
pxd_t pxd; /* 8: on-disk page pxd */
} redopage; /* (20) */
/*
* NOREDOPAGE: the page is freed
*
* do not apply after-image records which precede this record
* in the log with the same page block number to this page.
*
* N.B. REDOPAGE, NOREDOPAGE, and UPDATEMAP must be same format;
*/
struct {
__le32 fileset; /* 4: fileset number */
__le32 inode; /* 4: inode number */
__le16 type; /* 2: NOREDOPAGE record type */
__le16 rsrvd; /* 2: reserved */
pxd_t pxd; /* 8: on-disk page pxd */
} noredopage; /* (20) */
/*
* UPDATEMAP: update block allocation map
*
* either in-line PXD,
* or out-of-line XADLIST;
*
* N.B. REDOPAGE, NOREDOPAGE, and UPDATEMAP must be same format;
*/
struct {
__le32 fileset; /* 4: fileset number */
__le32 inode; /* 4: inode number */
__le16 type; /* 2: UPDATEMAP record type */
__le16 nxd; /* 2: number of extents */
pxd_t pxd; /* 8: pxd */
} updatemap; /* (20) */
/*
* NOREDOINOEXT: the inode extent is freed
*
* do not apply after-image records which precede this
* record in the log with the any of the 4 page block
* numbers in this inode extent.
*
* NOTE: The fileset and pxd fields MUST remain in
* the same fields in the REDOPAGE record format.
*
*/
struct {
__le32 fileset; /* 4: fileset number */
__le32 iagnum; /* 4: IAG number */
__le32 inoext_idx; /* 4: inode extent index */
pxd_t pxd; /* 8: on-disk page pxd */
} noredoinoext; /* (20) */
/*
* SYNCPT: log sync point
*
* replay log upto syncpt address specified;
*/
struct {
__le32 sync; /* 4: syncpt address (0 = here) */
} syncpt;
/*
* MOUNT: file system mount
*
* file system mount: no type-dependent information;
*/
/*
* ? FREEXTENT: free specified extent(s)
*
* free specified extent(s) from block allocation map
* N.B.: nextents should be length of data/sizeof(xad_t)
*/
struct {
__le32 type; /* 4: FREEXTENT record type */
__le32 nextent; /* 4: number of extents */
/* data: PXD or XAD list */
} freextent;
/*
* ? NOREDOFILE: this file is freed
*
* do not apply records which precede this record in the log
* with the same inode number.
*
* NOREDILE must be the first to be written at commit
* (last to be read in logredo()) - it prevents
* replay of preceding updates of all preceding generations
* of the inumber esp. the on-disk inode itself,
* but does NOT prevent
* replay of the
*/
struct {
__le32 fileset; /* 4: fileset number */
__le32 inode; /* 4: inode number */
} noredofile;
/*
* ? NEWPAGE:
*
* metadata type dependent
*/
struct {
__le32 fileset; /* 4: fileset number */
__le32 inode; /* 4: inode number */
__le32 type; /* 4: NEWPAGE record type */
pxd_t pxd; /* 8: on-disk page pxd */
} newpage;
/*
* ? DUMMY: filler
*
* no type-dependent information
*/
} log;
}; /* (36) */
#define LOGRDSIZE (sizeof(struct lrd))
/*
* line vector descriptor
*/
struct lvd {
__le16 offset;
__le16 length;
};
/*
* log logical volume
*/
struct jfs_log {
struct list_head sb_list;/* This is used to sync metadata
* before writing syncpt.
*/
struct list_head journal_list; /* Global list */
struct block_device *bdev; /* 4: log lv pointer */
int serial; /* 4: log mount serial number */
s64 base; /* @8: log extent address (inline log ) */
int size; /* 4: log size in log page (in page) */
int l2bsize; /* 4: log2 of bsize */
long flag; /* 4: flag */
struct lbuf *lbuf_free; /* 4: free lbufs */
wait_queue_head_t free_wait; /* 4: */
/* log write */
int logtid; /* 4: log tid */
int page; /* 4: page number of eol page */
int eor; /* 4: eor of last record in eol page */
struct lbuf *bp; /* 4: current log page buffer */
struct semaphore loglock; /* 4: log write serialization lock */
/* syncpt */
int nextsync; /* 4: bytes to write before next syncpt */
int active; /* 4: */
wait_queue_head_t syncwait; /* 4: */
/* commit */
uint cflag; /* 4: */
struct list_head cqueue; /* FIFO commit queue */
struct tblock *flush_tblk; /* tblk we're waiting on for flush */
int gcrtc; /* 4: GC_READY transaction count */
struct tblock *gclrt; /* 4: latest GC_READY transaction */
spinlock_t gclock; /* 4: group commit lock */
int logsize; /* 4: log data area size in byte */
int lsn; /* 4: end-of-log */
int clsn; /* 4: clsn */
int syncpt; /* 4: addr of last syncpt record */
int sync; /* 4: addr from last logsync() */
struct list_head synclist; /* 8: logsynclist anchor */
spinlock_t synclock; /* 4: synclist lock */
struct lbuf *wqueue; /* 4: log pageout queue */
int count; /* 4: count */
char uuid[16]; /* 16: 128-bit uuid of log device */
int no_integrity; /* 3: flag to disable journaling to disk */
};
/*
* Log flag
*/
#define log_INLINELOG 1
#define log_SYNCBARRIER 2
#define log_QUIESCE 3
#define log_FLUSH 4
/*
* group commit flag
*/
/* jfs_log */
#define logGC_PAGEOUT 0x00000001
/* tblock/lbuf */
#define tblkGC_QUEUE 0x0001
#define tblkGC_READY 0x0002
#define tblkGC_COMMIT 0x0004
#define tblkGC_COMMITTED 0x0008
#define tblkGC_EOP 0x0010
#define tblkGC_FREE 0x0020
#define tblkGC_LEADER 0x0040
#define tblkGC_ERROR 0x0080
#define tblkGC_LAZY 0x0100 // D230860
#define tblkGC_UNLOCKED 0x0200 // D230860
/*
* log cache buffer header
*/
struct lbuf {
struct jfs_log *l_log; /* 4: log associated with buffer */
/*
* data buffer base area
*/
uint l_flag; /* 4: pageout control flags */
struct lbuf *l_wqnext; /* 4: write queue link */
struct lbuf *l_freelist; /* 4: freelistlink */
int l_pn; /* 4: log page number */
int l_eor; /* 4: log record eor */
int l_ceor; /* 4: committed log record eor */
s64 l_blkno; /* 8: log page block number */
caddr_t l_ldata; /* 4: data page */
struct page *l_page; /* The page itself */
uint l_offset; /* Offset of l_ldata within the page */
wait_queue_head_t l_ioevent; /* 4: i/o done event */
};
/* Reuse l_freelist for redrive list */
#define l_redrive_next l_freelist
/*
* logsynclist block
*
* common logsyncblk prefix for jbuf_t and tblock
*/
struct logsyncblk {
u16 xflag; /* flags */
u16 flag; /* only meaninful in tblock */
lid_t lid; /* lock id */
s32 lsn; /* log sequence number */
struct list_head synclist; /* log sync list link */
};
/*
* logsynclist serialization (per log)
*/
#define LOGSYNC_LOCK_INIT(log) spin_lock_init(&(log)->synclock)
#define LOGSYNC_LOCK(log, flags) spin_lock_irqsave(&(log)->synclock, flags)
#define LOGSYNC_UNLOCK(log, flags) \
spin_unlock_irqrestore(&(log)->synclock, flags)
/* compute the difference in bytes of lsn from sync point */
#define logdiff(diff, lsn, log)\
{\
diff = (lsn) - (log)->syncpt;\
if (diff < 0)\
diff += (log)->logsize;\
}
extern int lmLogOpen(struct super_block *sb);
extern int lmLogClose(struct super_block *sb);
extern int lmLogShutdown(struct jfs_log * log);
extern int lmLogInit(struct jfs_log * log);
extern int lmLogFormat(struct jfs_log *log, s64 logAddress, int logSize);
extern int lmGroupCommit(struct jfs_log *, struct tblock *);
extern int jfsIOWait(void *);
extern void jfs_flush_journal(struct jfs_log * log, int wait);
extern void jfs_syncpt(struct jfs_log *log, int hard_sync);
#endif /* _H_JFS_LOGMGR */