xfs_log.c

linux 内核源代码

xfs_log_reserve(xfs_mount_t	 *mp,
		int		 unit_bytes,
		int		 cnt,
		xfs_log_ticket_t *ticket,
		__uint8_t	 client,
		uint		 flags,
		uint		 t_type)
{
	xlog_t		*log = mp->m_log;
	xlog_ticket_t	*internal_ticket;
	int		retval = 0;

	ASSERT(client == XFS_TRANSACTION || client == XFS_LOG);
	ASSERT((flags & XFS_LOG_NOSLEEP) == 0);

	if (XLOG_FORCED_SHUTDOWN(log))
		return XFS_ERROR(EIO);

	XFS_STATS_INC(xs_try_logspace);

	if (*ticket != NULL) {
		ASSERT(flags & XFS_LOG_PERM_RESERV);
		internal_ticket = (xlog_ticket_t *)*ticket;
		xlog_trace_loggrant(log, internal_ticket, "xfs_log_reserve: existing ticket (permanent trans)");
		xlog_grant_push_ail(mp, internal_ticket->t_unit_res);
		retval = xlog_regrant_write_log_space(log, internal_ticket);
	} else {
		/* may sleep if need to allocate more tickets */
		internal_ticket = xlog_ticket_get(log, unit_bytes, cnt,
						  client, flags);
		internal_ticket->t_trans_type = t_type;
		*ticket = internal_ticket;
		xlog_trace_loggrant(log, internal_ticket, 
			(internal_ticket->t_flags & XLOG_TIC_PERM_RESERV) ?
			"xfs_log_reserve: create new ticket (permanent trans)" :
			"xfs_log_reserve: create new ticket");
		xlog_grant_push_ail(mp,
				    (internal_ticket->t_unit_res *
				     internal_ticket->t_cnt));
		retval = xlog_grant_log_space(log, internal_ticket);
	}

	return retval;
}	/* xfs_log_reserve */


/*
 * Mount a log filesystem
 *
 * mp		- ubiquitous xfs mount point structure
 * log_target	- buftarg of on-disk log device
 * blk_offset	- Start block # where block size is 512 bytes (BBSIZE)
 * num_bblocks	- Number of BBSIZE blocks in on-disk log
 *
 * Return error or zero.
 */
int
xfs_log_mount(xfs_mount_t	*mp,
	      xfs_buftarg_t	*log_target,
	      xfs_daddr_t	blk_offset,
	      int		num_bblks)
{
	if (!(mp->m_flags & XFS_MOUNT_NORECOVERY))
		cmn_err(CE_NOTE, "XFS mounting filesystem %s", mp->m_fsname);
	else {
		cmn_err(CE_NOTE,
			"!Mounting filesystem \"%s\" in no-recovery mode.  Filesystem will be inconsistent.",
			mp->m_fsname);
		ASSERT(mp->m_flags & XFS_MOUNT_RDONLY);
	}

	mp->m_log = xlog_alloc_log(mp, log_target, blk_offset, num_bblks);

	/*
	 * skip log recovery on a norecovery mount.  pretend it all
	 * just worked.
	 */
	if (!(mp->m_flags & XFS_MOUNT_NORECOVERY)) {
		int		error, readonly = (mp->m_flags & XFS_MOUNT_RDONLY);

		if (readonly)
			mp->m_flags &= ~XFS_MOUNT_RDONLY;

		error = xlog_recover(mp->m_log);

		if (readonly)
			mp->m_flags |= XFS_MOUNT_RDONLY;
		if (error) {
			cmn_err(CE_WARN, "XFS: log mount/recovery failed: error %d", error);
			xlog_dealloc_log(mp->m_log);
			return error;
		}
	}

	/* Normal transactions can now occur */
	mp->m_log->l_flags &= ~XLOG_ACTIVE_RECOVERY;

	/* End mounting message in xfs_log_mount_finish */
	return 0;
}	/* xfs_log_mount */

/*
 * Finish the recovery of the file system.  This is separate from
 * the xfs_log_mount() call, because it depends on the code in
 * xfs_mountfs() to read in the root and real-time bitmap inodes
 * between calling xfs_log_mount() and here.
 *
 * mp		- ubiquitous xfs mount point structure
 */
int
xfs_log_mount_finish(xfs_mount_t *mp, int mfsi_flags)
{
	int	error;

	if (!(mp->m_flags & XFS_MOUNT_NORECOVERY))
		error = xlog_recover_finish(mp->m_log, mfsi_flags);
	else {
		error = 0;
		ASSERT(mp->m_flags & XFS_MOUNT_RDONLY);
	}

	return error;
}

/*
 * Unmount processing for the log.
 */
int
xfs_log_unmount(xfs_mount_t *mp)
{
	int		error;

	error = xfs_log_unmount_write(mp);
	xfs_log_unmount_dealloc(mp);
	return error;
}

/*
 * Final log writes as part of unmount.
 *
 * Mark the filesystem clean as unmount happens.  Note that during relocation
 * this routine needs to be executed as part of source-bag while the
 * deallocation must not be done until source-end.
 */

/*
 * Unmount record used to have a string "Unmount filesystem--" in the
 * data section where the "Un" was really a magic number (XLOG_UNMOUNT_TYPE).
 * We just write the magic number now since that particular field isn't
 * currently architecture converted and "nUmount" is a bit foo.
 * As far as I know, there weren't any dependencies on the old behaviour.
 */

int
xfs_log_unmount_write(xfs_mount_t *mp)
{
	xlog_t		 *log = mp->m_log;
	xlog_in_core_t	 *iclog;
#ifdef DEBUG
	xlog_in_core_t	 *first_iclog;
#endif
	xfs_log_iovec_t  reg[1];
	xfs_log_ticket_t tic = NULL;
	xfs_lsn_t	 lsn;
	int		 error;
	SPLDECL(s);

	/* the data section must be 32 bit size aligned */
	struct {
	    __uint16_t magic;
	    __uint16_t pad1;
	    __uint32_t pad2; /* may as well make it 64 bits */
	} magic = { XLOG_UNMOUNT_TYPE, 0, 0 };

	/*
	 * Don't write out unmount record on read-only mounts.
	 * Or, if we are doing a forced umount (typically because of IO errors).
	 */
	if (mp->m_flags & XFS_MOUNT_RDONLY)
		return 0;

	xfs_log_force(mp, 0, XFS_LOG_FORCE|XFS_LOG_SYNC);

#ifdef DEBUG
	first_iclog = iclog = log->l_iclog;
	do {
		if (!(iclog->ic_state & XLOG_STATE_IOERROR)) {
			ASSERT(iclog->ic_state & XLOG_STATE_ACTIVE);
			ASSERT(iclog->ic_offset == 0);
		}
		iclog = iclog->ic_next;
	} while (iclog != first_iclog);
#endif
	if (! (XLOG_FORCED_SHUTDOWN(log))) {
		reg[0].i_addr = (void*)&magic;
		reg[0].i_len  = sizeof(magic);
		XLOG_VEC_SET_TYPE(&reg[0], XLOG_REG_TYPE_UNMOUNT);

		error = xfs_log_reserve(mp, 600, 1, &tic,
					XFS_LOG, 0, XLOG_UNMOUNT_REC_TYPE);
		if (!error) {
			/* remove inited flag */
			((xlog_ticket_t *)tic)->t_flags = 0;
			error = xlog_write(mp, reg, 1, tic, &lsn,
					   NULL, XLOG_UNMOUNT_TRANS);
			/*
			 * At this point, we're umounting anyway,
			 * so there's no point in transitioning log state
			 * to IOERROR. Just continue...
			 */
		}

		if (error) {
			xfs_fs_cmn_err(CE_ALERT, mp,
				"xfs_log_unmount: unmount record failed");
		}


		s = LOG_LOCK(log);
		iclog = log->l_iclog;
		iclog->ic_refcnt++;
		LOG_UNLOCK(log, s);
		xlog_state_want_sync(log, iclog);
		(void) xlog_state_release_iclog(log, iclog);

		s = LOG_LOCK(log);
		if (!(iclog->ic_state == XLOG_STATE_ACTIVE ||
		      iclog->ic_state == XLOG_STATE_DIRTY)) {
			if (!XLOG_FORCED_SHUTDOWN(log)) {
				sv_wait(&iclog->ic_forcesema, PMEM,
					&log->l_icloglock, s);
			} else {
				LOG_UNLOCK(log, s);
			}
		} else {
			LOG_UNLOCK(log, s);
		}
		if (tic) {
			xlog_trace_loggrant(log, tic, "unmount rec");
			xlog_ungrant_log_space(log, tic);
			xlog_state_put_ticket(log, tic);
		}
	} else {
		/*
		 * We're already in forced_shutdown mode, couldn't
		 * even attempt to write out the unmount transaction.
		 *
		 * Go through the motions of sync'ing and releasing
		 * the iclog, even though no I/O will actually happen,
		 * we need to wait for other log I/Os that may already
		 * be in progress.  Do this as a separate section of
		 * code so we'll know if we ever get stuck here that
		 * we're in this odd situation of trying to unmount
		 * a file system that went into forced_shutdown as
		 * the result of an unmount..
		 */
		s = LOG_LOCK(log);
		iclog = log->l_iclog;
		iclog->ic_refcnt++;
		LOG_UNLOCK(log, s);

		xlog_state_want_sync(log, iclog);
		(void) xlog_state_release_iclog(log, iclog);

		s = LOG_LOCK(log);

		if ( ! (   iclog->ic_state == XLOG_STATE_ACTIVE
			|| iclog->ic_state == XLOG_STATE_DIRTY
			|| iclog->ic_state == XLOG_STATE_IOERROR) ) {

				sv_wait(&iclog->ic_forcesema, PMEM,
					&log->l_icloglock, s);
		} else {
			LOG_UNLOCK(log, s);
		}
	}

	return 0;
}	/* xfs_log_unmount_write */

/*
 * Deallocate log structures for unmount/relocation.
 */
void
xfs_log_unmount_dealloc(xfs_mount_t *mp)
{
	xlog_dealloc_log(mp->m_log);
}

/*
 * Write region vectors to log.  The write happens using the space reservation
 * of the ticket (tic).  It is not a requirement that all writes for a given
 * transaction occur with one call to xfs_log_write().
 */
int
xfs_log_write(xfs_mount_t *	mp,
	      xfs_log_iovec_t	reg[],
	      int		nentries,
	      xfs_log_ticket_t	tic,
	      xfs_lsn_t		*start_lsn)
{
	int	error;
	xlog_t *log = mp->m_log;

	if (XLOG_FORCED_SHUTDOWN(log))
		return XFS_ERROR(EIO);

	if ((error = xlog_write(mp, reg, nentries, tic, start_lsn, NULL, 0))) {
		xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR);
	}
	return error;
}	/* xfs_log_write */


void
xfs_log_move_tail(xfs_mount_t	*mp,
		  xfs_lsn_t	tail_lsn)
{
	xlog_ticket_t	*tic;
	xlog_t		*log = mp->m_log;
	int		need_bytes, free_bytes, cycle, bytes;
	SPLDECL(s);

	if (XLOG_FORCED_SHUTDOWN(log))
		return;
	ASSERT(!XFS_FORCED_SHUTDOWN(mp));

	if (tail_lsn == 0) {
		/* needed since sync_lsn is 64 bits */
		s = LOG_LOCK(log);
		tail_lsn = log->l_last_sync_lsn;
		LOG_UNLOCK(log, s);
	}

	s = GRANT_LOCK(log);

	/* Also an invalid lsn.  1 implies that we aren't passing in a valid
	 * tail_lsn.
	 */
	if (tail_lsn != 1) {
		log->l_tail_lsn = tail_lsn;
	}

	if ((tic = log->l_write_headq)) {
#ifdef DEBUG
		if (log->l_flags & XLOG_ACTIVE_RECOVERY)
			panic("Recovery problem");
#endif
		cycle = log->l_grant_write_cycle;
		bytes = log->l_grant_write_bytes;
		free_bytes = xlog_space_left(log, cycle, bytes);
		do {
			ASSERT(tic->t_flags & XLOG_TIC_PERM_RESERV);

			if (free_bytes < tic->t_unit_res && tail_lsn != 1)
				break;
			tail_lsn = 0;
			free_bytes -= tic->t_unit_res;
			sv_signal(&tic->t_sema);
			tic = tic->t_next;
		} while (tic != log->l_write_headq);
	}
	if ((tic = log->l_reserve_headq)) {
#ifdef DEBUG
		if (log->l_flags & XLOG_ACTIVE_RECOVERY)
			panic("Recovery problem");
#endif
		cycle = log->l_grant_reserve_cycle;
		bytes = log->l_grant_reserve_bytes;
		free_bytes = xlog_space_left(log, cycle, bytes);
		do {
			if (tic->t_flags & XLOG_TIC_PERM_RESERV)
				need_bytes = tic->t_unit_res*tic->t_cnt;
			else
				need_bytes = tic->t_unit_res;
			if (free_bytes < need_bytes && tail_lsn != 1)
				break;
			tail_lsn = 0;
			free_bytes -= need_bytes;
			sv_signal(&tic->t_sema);
			tic = tic->t_next;
		} while (tic != log->l_reserve_headq);
	}
	GRANT_UNLOCK(log, s);
}	/* xfs_log_move_tail */

/*
 * Determine if we have a transaction that has gone to disk
 * that needs to be covered. Log activity needs to be idle (no AIL and
 * nothing in the iclogs). And, we need to be in the right state indicating
 * something has gone out.
 */
int
xfs_log_need_covered(xfs_mount_t *mp)
{
	SPLDECL(s);
	int		needed = 0, gen;
	xlog_t		*log = mp->m_log;

	if (!xfs_fs_writable(mp))
		return 0;

	s = LOG_LOCK(log);
	if (((log->l_covered_state == XLOG_STATE_COVER_NEED) ||
		(log->l_covered_state == XLOG_STATE_COVER_NEED2))
			&& !xfs_trans_first_ail(mp, &gen)
			&& xlog_iclogs_empty(log)) {
		if (log->l_covered_state == XLOG_STATE_COVER_NEED)
			log->l_covered_state = XLOG_STATE_COVER_DONE;
		else {
			ASSERT(log->l_covered_state == XLOG_STATE_COVER_NEED2);
			log->l_covered_state = XLOG_STATE_COVER_DONE2;
		}
		needed = 1;
	}
	LOG_UNLOCK(log, s);
	return needed;
}

/******************************************************************************
 *
 *	local routines
 *
 ******************************************************************************
 */

/* xfs_trans_tail_ail returns 0 when there is nothing in the list.
 * The log manager must keep track of the last LR which was committed
 * to disk.  The lsn of this LR will become the new tail_lsn whenever
 * xfs_trans_tail_ail returns 0.  If we don't do this, we run into
 * the situation where stuff could be written into the log but nothing
 * was ever in the AIL when asked.  Eventually, we panic since the
 * tail hits the head.
 *
 * We may be holding the log iclog lock upon entering this routine.
 */
xfs_lsn_t
xlog_assign_tail_lsn(xfs_mount_t *mp)
{
	xfs_lsn_t tail_lsn;
	SPLDECL(s);
	xlog_t	  *log = mp->m_log;