xfs_log.c

linux 内核源代码

	if (threshold_block >= log->l_logBBsize) {
	    threshold_block -= log->l_logBBsize;
	    threshold_cycle += 1;
	}
	ASSIGN_ANY_LSN_HOST(threshold_lsn, threshold_cycle,
		       threshold_block);

	/* Don't pass in an lsn greater than the lsn of the last
	 * log record known to be on disk.
	 */
	if (XFS_LSN_CMP(threshold_lsn, log->l_last_sync_lsn) > 0)
	    threshold_lsn = log->l_last_sync_lsn;
    }
    GRANT_UNLOCK(log, s);

    /*
     * Get the transaction layer to kick the dirty buffers out to
     * disk asynchronously. No point in trying to do this if
     * the filesystem is shutting down.
     */
    if (threshold_lsn &&
	!XLOG_FORCED_SHUTDOWN(log))
	    xfs_trans_push_ail(mp, threshold_lsn);
}	/* xlog_grant_push_ail */


/*
 * Flush out the in-core log (iclog) to the on-disk log in an asynchronous 
 * fashion.  Previously, we should have moved the current iclog
 * ptr in the log to point to the next available iclog.  This allows further
 * write to continue while this code syncs out an iclog ready to go.
 * Before an in-core log can be written out, the data section must be scanned
 * to save away the 1st word of each BBSIZE block into the header.  We replace
 * it with the current cycle count.  Each BBSIZE block is tagged with the
 * cycle count because there in an implicit assumption that drives will
 * guarantee that entire 512 byte blocks get written at once.  In other words,
 * we can't have part of a 512 byte block written and part not written.  By
 * tagging each block, we will know which blocks are valid when recovering
 * after an unclean shutdown.
 *
 * This routine is single threaded on the iclog.  No other thread can be in
 * this routine with the same iclog.  Changing contents of iclog can there-
 * fore be done without grabbing the state machine lock.  Updating the global
 * log will require grabbing the lock though.
 *
 * The entire log manager uses a logical block numbering scheme.  Only
 * log_sync (and then only bwrite()) know about the fact that the log may
 * not start with block zero on a given device.  The log block start offset
 * is added immediately before calling bwrite().
 */

int
xlog_sync(xlog_t		*log,
	  xlog_in_core_t	*iclog)
{
	xfs_caddr_t	dptr;		/* pointer to byte sized element */
	xfs_buf_t	*bp;
	int		i, ops;
	uint		count;		/* byte count of bwrite */
	uint		count_init;	/* initial count before roundup */
	int		roundoff;       /* roundoff to BB or stripe */
	int		split = 0;	/* split write into two regions */
	int		error;
	SPLDECL(s);
	int		v2 = XFS_SB_VERSION_HASLOGV2(&log->l_mp->m_sb);

	XFS_STATS_INC(xs_log_writes);
	ASSERT(iclog->ic_refcnt == 0);

	/* Add for LR header */
	count_init = log->l_iclog_hsize + iclog->ic_offset;

	/* Round out the log write size */
	if (v2 && log->l_mp->m_sb.sb_logsunit > 1) {
		/* we have a v2 stripe unit to use */
		count = XLOG_LSUNITTOB(log, XLOG_BTOLSUNIT(log, count_init));
	} else {
		count = BBTOB(BTOBB(count_init));
	}
	roundoff = count - count_init;
	ASSERT(roundoff >= 0);
	ASSERT((v2 && log->l_mp->m_sb.sb_logsunit > 1 && 
                roundoff < log->l_mp->m_sb.sb_logsunit)
		|| 
		(log->l_mp->m_sb.sb_logsunit <= 1 && 
		 roundoff < BBTOB(1)));

	/* move grant heads by roundoff in sync */
	s = GRANT_LOCK(log);
	xlog_grant_add_space(log, roundoff);
	GRANT_UNLOCK(log, s);

	/* put cycle number in every block */
	xlog_pack_data(log, iclog, roundoff); 

	/* real byte length */
	if (v2) {
		INT_SET(iclog->ic_header.h_len, 
			ARCH_CONVERT,
			iclog->ic_offset + roundoff);
	} else {
		INT_SET(iclog->ic_header.h_len, ARCH_CONVERT, iclog->ic_offset);
	}

	/* put ops count in correct order */
	ops = iclog->ic_header.h_num_logops;
	INT_SET(iclog->ic_header.h_num_logops, ARCH_CONVERT, ops);

	bp = iclog->ic_bp;
	ASSERT(XFS_BUF_FSPRIVATE2(bp, unsigned long) == (unsigned long)1);
	XFS_BUF_SET_FSPRIVATE2(bp, (unsigned long)2);
	XFS_BUF_SET_ADDR(bp, BLOCK_LSN(INT_GET(iclog->ic_header.h_lsn, ARCH_CONVERT)));

	XFS_STATS_ADD(xs_log_blocks, BTOBB(count));

	/* Do we need to split this write into 2 parts? */
	if (XFS_BUF_ADDR(bp) + BTOBB(count) > log->l_logBBsize) {
		split = count - (BBTOB(log->l_logBBsize - XFS_BUF_ADDR(bp)));
		count = BBTOB(log->l_logBBsize - XFS_BUF_ADDR(bp));
		iclog->ic_bwritecnt = 2;	/* split into 2 writes */
	} else {
		iclog->ic_bwritecnt = 1;
	}
	XFS_BUF_SET_COUNT(bp, count);
	XFS_BUF_SET_FSPRIVATE(bp, iclog);	/* save for later */
	XFS_BUF_ZEROFLAGS(bp);
	XFS_BUF_BUSY(bp);
	XFS_BUF_ASYNC(bp);
	/*
	 * Do an ordered write for the log block.
	 * Its unnecessary to flush the first split block in the log wrap case.
	 */
	if (!split && (log->l_mp->m_flags & XFS_MOUNT_BARRIER))
		XFS_BUF_ORDERED(bp);

	ASSERT(XFS_BUF_ADDR(bp) <= log->l_logBBsize-1);
	ASSERT(XFS_BUF_ADDR(bp) + BTOBB(count) <= log->l_logBBsize);

	xlog_verify_iclog(log, iclog, count, B_TRUE);

	/* account for log which doesn't start at block #0 */
	XFS_BUF_SET_ADDR(bp, XFS_BUF_ADDR(bp) + log->l_logBBstart);
	/*
	 * Don't call xfs_bwrite here. We do log-syncs even when the filesystem
	 * is shutting down.
	 */
	XFS_BUF_WRITE(bp);

	if ((error = XFS_bwrite(bp))) {
		xfs_ioerror_alert("xlog_sync", log->l_mp, bp,
				  XFS_BUF_ADDR(bp));
		return error;
	}
	if (split) {
		bp = iclog->ic_log->l_xbuf;
		ASSERT(XFS_BUF_FSPRIVATE2(bp, unsigned long) ==
							(unsigned long)1);
		XFS_BUF_SET_FSPRIVATE2(bp, (unsigned long)2);
		XFS_BUF_SET_ADDR(bp, 0);	     /* logical 0 */
		XFS_BUF_SET_PTR(bp, (xfs_caddr_t)((__psint_t)&(iclog->ic_header)+
					    (__psint_t)count), split);
		XFS_BUF_SET_FSPRIVATE(bp, iclog);
		XFS_BUF_ZEROFLAGS(bp);
		XFS_BUF_BUSY(bp);
		XFS_BUF_ASYNC(bp);
		if (log->l_mp->m_flags & XFS_MOUNT_BARRIER)
			XFS_BUF_ORDERED(bp);
		dptr = XFS_BUF_PTR(bp);
		/*
		 * Bump the cycle numbers at the start of each block
		 * since this part of the buffer is at the start of
		 * a new cycle.  Watch out for the header magic number
		 * case, though.
		 */
		for (i=0; i<split; i += BBSIZE) {
			INT_MOD(*(uint *)dptr, ARCH_CONVERT, +1);
			if (INT_GET(*(uint *)dptr, ARCH_CONVERT) == XLOG_HEADER_MAGIC_NUM)
				INT_MOD(*(uint *)dptr, ARCH_CONVERT, +1);
			dptr += BBSIZE;
		}

		ASSERT(XFS_BUF_ADDR(bp) <= log->l_logBBsize-1);
		ASSERT(XFS_BUF_ADDR(bp) + BTOBB(count) <= log->l_logBBsize);

		/* account for internal log which doesn't start at block #0 */
		XFS_BUF_SET_ADDR(bp, XFS_BUF_ADDR(bp) + log->l_logBBstart);
		XFS_BUF_WRITE(bp);
		if ((error = XFS_bwrite(bp))) {
			xfs_ioerror_alert("xlog_sync (split)", log->l_mp,
					  bp, XFS_BUF_ADDR(bp));
			return error;
		}
	}
	return 0;
}	/* xlog_sync */


/*
 * Deallocate a log structure
 */
void
xlog_dealloc_log(xlog_t *log)
{
	xlog_in_core_t	*iclog, *next_iclog;
	xlog_ticket_t	*tic, *next_tic;
	int		i;


	iclog = log->l_iclog;
	for (i=0; i<log->l_iclog_bufs; i++) {
		sv_destroy(&iclog->ic_forcesema);
		sv_destroy(&iclog->ic_writesema);
		xfs_buf_free(iclog->ic_bp);
#ifdef XFS_LOG_TRACE
		if (iclog->ic_trace != NULL) {
			ktrace_free(iclog->ic_trace);
		}
#endif
		next_iclog = iclog->ic_next;
		kmem_free(iclog, sizeof(xlog_in_core_t));
		iclog = next_iclog;
	}
	freesema(&log->l_flushsema);
	spinlock_destroy(&log->l_icloglock);
	spinlock_destroy(&log->l_grant_lock);

	/* XXXsup take a look at this again. */
	if ((log->l_ticket_cnt != log->l_ticket_tcnt)  &&
	    !XLOG_FORCED_SHUTDOWN(log)) {
		xfs_fs_cmn_err(CE_WARN, log->l_mp,
			"xlog_dealloc_log: (cnt: %d, total: %d)",
			log->l_ticket_cnt, log->l_ticket_tcnt);
		/* ASSERT(log->l_ticket_cnt == log->l_ticket_tcnt); */

	} else {
		tic = log->l_unmount_free;
		while (tic) {
			next_tic = tic->t_next;
			kmem_free(tic, NBPP);
			tic = next_tic;
		}
	}
	xfs_buf_free(log->l_xbuf);
#ifdef XFS_LOG_TRACE
	if (log->l_trace != NULL) {
		ktrace_free(log->l_trace);
	}
	if (log->l_grant_trace != NULL) {
		ktrace_free(log->l_grant_trace);
	}
#endif
	log->l_mp->m_log = NULL;
	kmem_free(log, sizeof(xlog_t));
}	/* xlog_dealloc_log */

/*
 * Update counters atomically now that memcpy is done.
 */
/* ARGSUSED */
static inline void
xlog_state_finish_copy(xlog_t		*log,
		       xlog_in_core_t	*iclog,
		       int		record_cnt,
		       int		copy_bytes)
{
	SPLDECL(s);

	s = LOG_LOCK(log);

	iclog->ic_header.h_num_logops += record_cnt;
	iclog->ic_offset += copy_bytes;

	LOG_UNLOCK(log, s);
}	/* xlog_state_finish_copy */




/*
 * print out info relating to regions written which consume
 * the reservation
 */
STATIC void
xlog_print_tic_res(xfs_mount_t *mp, xlog_ticket_t *ticket)
{
	uint i;
	uint ophdr_spc = ticket->t_res_num_ophdrs * (uint)sizeof(xlog_op_header_t);

	/* match with XLOG_REG_TYPE_* in xfs_log.h */
	static char *res_type_str[XLOG_REG_TYPE_MAX] = {
	    "bformat",
	    "bchunk",
	    "efi_format",
	    "efd_format",
	    "iformat",
	    "icore",
	    "iext",
	    "ibroot",
	    "ilocal",
	    "iattr_ext",
	    "iattr_broot",
	    "iattr_local",
	    "qformat",
	    "dquot",
	    "quotaoff",
	    "LR header",
	    "unmount",
	    "commit",
	    "trans header"
	};
	static char *trans_type_str[XFS_TRANS_TYPE_MAX] = {
	    "SETATTR_NOT_SIZE",
	    "SETATTR_SIZE",
	    "INACTIVE",
	    "CREATE",
	    "CREATE_TRUNC",
	    "TRUNCATE_FILE",
	    "REMOVE",
	    "LINK",
	    "RENAME",
	    "MKDIR",
	    "RMDIR",
	    "SYMLINK",
	    "SET_DMATTRS",
	    "GROWFS",
	    "STRAT_WRITE",
	    "DIOSTRAT",
	    "WRITE_SYNC",
	    "WRITEID",
	    "ADDAFORK",
	    "ATTRINVAL",
	    "ATRUNCATE",
	    "ATTR_SET",
	    "ATTR_RM",
	    "ATTR_FLAG",
	    "CLEAR_AGI_BUCKET",
	    "QM_SBCHANGE",
	    "DUMMY1",
	    "DUMMY2",
	    "QM_QUOTAOFF",
	    "QM_DQALLOC",
	    "QM_SETQLIM",
	    "QM_DQCLUSTER",
	    "QM_QINOCREATE",
	    "QM_QUOTAOFF_END",
	    "SB_UNIT",
	    "FSYNC_TS",
	    "GROWFSRT_ALLOC",
	    "GROWFSRT_ZERO",
	    "GROWFSRT_FREE",
	    "SWAPEXT"
	};

	xfs_fs_cmn_err(CE_WARN, mp,
			"xfs_log_write: reservation summary:\n"
			"  trans type  = %s (%u)\n"
			"  unit res    = %d bytes\n"
			"  current res = %d bytes\n"
			"  total reg   = %u bytes (o/flow = %u bytes)\n"
			"  ophdrs      = %u (ophdr space = %u bytes)\n"
			"  ophdr + reg = %u bytes\n"
			"  num regions = %u\n",
			((ticket->t_trans_type <= 0 ||
			  ticket->t_trans_type > XFS_TRANS_TYPE_MAX) ?
			  "bad-trans-type" : trans_type_str[ticket->t_trans_type-1]),
			ticket->t_trans_type,
			ticket->t_unit_res,
			ticket->t_curr_res,
			ticket->t_res_arr_sum, ticket->t_res_o_flow,
			ticket->t_res_num_ophdrs, ophdr_spc,
			ticket->t_res_arr_sum + 
			ticket->t_res_o_flow + ophdr_spc,
			ticket->t_res_num);

	for (i = 0; i < ticket->t_res_num; i++) {
		uint r_type = ticket->t_res_arr[i].r_type; 
		cmn_err(CE_WARN,
			    "region[%u]: %s - %u bytes\n",
			    i, 
			    ((r_type <= 0 || r_type > XLOG_REG_TYPE_MAX) ?
			    "bad-rtype" : res_type_str[r_type-1]),
			    ticket->t_res_arr[i].r_len);
	}
}

/*
 * Write some region out to in-core log
 *
 * This will be called when writing externally provided regions or when
 * writing out a commit record for a given transaction.
 *
 * General algorithm:
 *	1. Find total length of this write.  This may include adding to the
 *		lengths passed in.
 *	2. Check whether we violate the tickets reservation.
 *	3. While writing to this iclog
 *	    A. Reserve as much space in this iclog as can get
 *	    B. If this is first write, save away start lsn
 *	    C. While writing this region:
 *		1. If first write of transaction, write start record
 *		2. Write log operation header (header per region)
 *		3. Find out if we can fit entire region into this iclog
 *		4. Potentially, verify destination memcpy ptr
 *		5. Memcpy (partial) region
 *		6. If partial copy, release iclog; otherwise, continue
 *			copying more regions into current iclog
 *	4. Mark want sync bit (in simulation mode)
 *	5. Release iclog for potential flush to on-disk log.
 *
 * ERRORS:
 * 1.	Panic if reservation is overrun.  This should never happen since
 *	reservation amounts are generated internal to the filesystem.
 * NOTES:
 * 1. Tickets are single threaded data structures.
 * 2. The XLOG_END_TRANS & XLOG_CONTINUE_TRANS flags are passed down to the
 *	syncing routine.  When a single log_write region needs to span
 *	multiple in-core logs, the XLOG_CONTINUE_TRANS bit should be set
 *	on all log operation writes which don't contain the end of the
 *	region.  The XLOG_END_TRANS bit is used for the in-core log
 *	operation which contains the end of the continued log_write region.
 * 3. When xlog_state_get_iclog_space() grabs the rest of the current iclog,
 *	we don't really know exactly how much space will be used.  As a result,
 *	we don't update ic_offset until the end when we know exactly how many
 *	bytes have been written out.
 */
int
xlog_write(xfs_mount_t *	mp,
	   xfs_log_iovec_t	reg[],
	   int			nentries,
	   xfs_log_ticket_t	tic,
	   xfs_lsn_t		*start_lsn,
	   xlog_in_core_t	**commit_iclog,
	   uint			flags)
{
    xlog_t	     *log = mp->m_log;
    xlog_ticket_t    *ticket = (xlog_ticket_t *)tic;
    xlog_in_core_t   *iclog = NULL;  /* ptr to current in-core log */
    xlog_op_header_t *logop_head;    /* ptr to log operation header */
    __psint_t	     ptr;	     /* copy address into data region */
    int		     len;	     /* # xlog_write() bytes 2 still copy */
    int		     index;	     /* region index currently copying */
    int		     log_offset;     /* offset (from 0) into data region */
    int		     start_rec_copy; /* # bytes to copy for start record */