xfs: format log items write directly into the linear CIL buffer

Instead of setting up pointers to memory locations in iop_format which then
get copied into the CIL linear buffer after return move the copy into
the individual inode items. This avoids the need to always have a memory
block in the exact same layout that gets written into the log around, and
allow the log items to be much more flexible in their in-memory layouts.

The only caveat is that we need to properly align the data for each
iovec so that don't have structures misaligned in subsequent iovecs.

Note that all log item format routines now need to be careful to modify
the copy of the item that was placed into the CIL after calls to
xlog_copy_iovec instead of the in-memory copy.

Signed-off-by: Christoph Hellwig <[e-mail]>
Reviewed-by: Dave Chinner <[e-mail]>
Signed-off-by: Dave Chinner <[e-mail]>

fs/xfs/xfs_buf_item.c

···
 
 static inline void
 xfs_buf_item_copy_iovec(
+	struct xfs_log_vec	*lv,
 	struct xfs_log_iovec	**vecp,
 	struct xfs_buf		*bp,
 	uint			offset,
···
 	uint			nbits)
 {
 	offset += first_bit * XFS_BLF_CHUNK;
-	xlog_copy_iovec(vecp, XLOG_REG_TYPE_BCHUNK,
+	xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_BCHUNK,
 			xfs_buf_offset(bp, offset),
 			nbits * XFS_BLF_CHUNK);
 }
···
 static void
 xfs_buf_item_format_segment(
 	struct xfs_buf_log_item	*bip,
+	struct xfs_log_vec	*lv,
 	struct xfs_log_iovec	**vecp,
 	uint			offset,
 	struct xfs_buf_log_format *blfp)
 {
 	struct xfs_buf	*bp = bip->bli_buf;
 	uint		base_size;
-	uint		nvecs;
 	int		first_bit;
 	int		last_bit;
 	int		next_bit;
···
 	 */
 	base_size = xfs_buf_log_format_size(blfp);
 
-	nvecs = 0;
 	first_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size, 0);
 	if (!(bip->bli_flags & XFS_BLI_STALE) && first_bit == -1) {
 		/*
 		 * If the map is not be dirty in the transaction, mark
 		 * the size as zero and do not advance the vector pointer.
 		 */
-		goto out;
+		return;
 	}
 
-	xlog_copy_iovec(vecp, XLOG_REG_TYPE_BFORMAT, blfp, base_size);
-	nvecs = 1;
+	blfp = xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_BFORMAT, blfp, base_size);
+	blfp->blf_size = 1;
 
 	if (bip->bli_flags & XFS_BLI_STALE) {
 		/*
···
 		 */
 		trace_xfs_buf_item_format_stale(bip);
 		ASSERT(blfp->blf_flags & XFS_BLF_CANCEL);
-		goto out;
+		return;
 	}
 
 
···
 		 * same set of bits so just keep counting and scanning.
 		 */
 		if (next_bit == -1) {
-			xfs_buf_item_copy_iovec(vecp, bp, offset,
+			xfs_buf_item_copy_iovec(lv, vecp, bp, offset,
 						first_bit, nbits);
-			nvecs++;
+			blfp->blf_size++;
 			break;
 		} else if (next_bit != last_bit + 1 ||
 		           xfs_buf_item_straddle(bp, offset, next_bit, last_bit)) {
-			xfs_buf_item_copy_iovec(vecp, bp, offset,
+			xfs_buf_item_copy_iovec(lv, vecp, bp, offset,
 						first_bit, nbits);
-			nvecs++;
+			blfp->blf_size++;
 			first_bit = next_bit;
 			last_bit = next_bit;
 			nbits = 1;
···
 			nbits++;
 		}
 	}
-out:
-	blfp->blf_size = nvecs;
 }
 
 /*
···
 STATIC void
 xfs_buf_item_format(
 	struct xfs_log_item	*lip,
-	struct xfs_log_iovec	*vecp)
+	struct xfs_log_vec	*lv)
 {
 	struct xfs_buf_log_item	*bip = BUF_ITEM(lip);
 	struct xfs_buf		*bp = bip->bli_buf;
+	struct xfs_log_iovec	*vecp = NULL;
 	uint			offset = 0;
 	int			i;
 
···
 	}
 
 	for (i = 0; i < bip->bli_format_count; i++) {
-		xfs_buf_item_format_segment(bip, &vecp, offset,
+		xfs_buf_item_format_segment(bip, lv, &vecp, offset,
 					    &bip->bli_formats[i]);
 		offset += bp->b_maps[i].bm_len;
 	}

fs/xfs/xfs_dquot_item.c

···
 STATIC void
 xfs_qm_dquot_logitem_format(
 	struct xfs_log_item	*lip,
-	struct xfs_log_iovec	*vecp)
+	struct xfs_log_vec	*lv)
 {
 	struct xfs_dq_logitem	*qlip = DQUOT_ITEM(lip);
-
-	xlog_copy_iovec(&vecp, XLOG_REG_TYPE_QFORMAT,
-			&qlip->qli_format,
-			sizeof(struct xfs_dq_logformat));
-	xlog_copy_iovec(&vecp, XLOG_REG_TYPE_DQUOT,
-			&qlip->qli_dquot->q_core,
-			sizeof(struct xfs_disk_dquot));
+	struct xfs_log_iovec	*vecp = NULL;
 
 	qlip->qli_format.qlf_size = 2;
+
+	xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_QFORMAT,
+			&qlip->qli_format,
+			sizeof(struct xfs_dq_logformat));
+	xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_DQUOT,
+			&qlip->qli_dquot->q_core,
+			sizeof(struct xfs_disk_dquot));
 }
 
 /*
···
 STATIC void
 xfs_qm_qoff_logitem_format(
 	struct xfs_log_item	*lip,
-	struct xfs_log_iovec	*vecp)
+	struct xfs_log_vec	*lv)
 {
 	struct xfs_qoff_logitem	*qflip = QOFF_ITEM(lip);
+	struct xfs_log_iovec	*vecp = NULL;
 
 	ASSERT(qflip->qql_format.qf_type == XFS_LI_QUOTAOFF);
+	qflip->qql_format.qf_size = 1;
 
-	xlog_copy_iovec(&vecp, XLOG_REG_TYPE_QUOTAOFF,
+	xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_QUOTAOFF,
 			&qflip->qql_format,
 			sizeof(struct xfs_qoff_logitem));
-
-	qflip->qql_format.qf_size = 1;
 }
 
 /*

fs/xfs/xfs_extfree_item.c

···
 STATIC void
 xfs_efi_item_format(
 	struct xfs_log_item	*lip,
-	struct xfs_log_iovec	*vecp)
+	struct xfs_log_vec	*lv)
 {
 	struct xfs_efi_log_item	*efip = EFI_ITEM(lip);
+	struct xfs_log_iovec	*vecp = NULL;
 
 	ASSERT(atomic_read(&efip->efi_next_extent) ==
 				efip->efi_format.efi_nextents);
···
 	efip->efi_format.efi_type = XFS_LI_EFI;
 	efip->efi_format.efi_size = 1;
 
-	xlog_copy_iovec(&vecp, XLOG_REG_TYPE_EFI_FORMAT,
+	xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_EFI_FORMAT,
 			&efip->efi_format,
 			xfs_efi_item_sizeof(efip));
 }
···
 STATIC void
 xfs_efd_item_format(
 	struct xfs_log_item	*lip,
-	struct xfs_log_iovec	*vecp)
+	struct xfs_log_vec	*lv)
 {
 	struct xfs_efd_log_item	*efdp = EFD_ITEM(lip);
+	struct xfs_log_iovec	*vecp = NULL;
 
 	ASSERT(efdp->efd_next_extent == efdp->efd_format.efd_nextents);
 
 	efdp->efd_format.efd_type = XFS_LI_EFD;
 	efdp->efd_format.efd_size = 1;
 
-	xlog_copy_iovec(&vecp, XLOG_REG_TYPE_EFD_FORMAT,
+	xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_EFD_FORMAT,
 			&efdp->efd_format,
 			xfs_efd_item_sizeof(efdp));
 }

fs/xfs/xfs_icreate_item.c

···
 STATIC void
 xfs_icreate_item_format(
 	struct xfs_log_item	*lip,
-	struct xfs_log_iovec	*vecp)
+	struct xfs_log_vec	*lv)
 {
 	struct xfs_icreate_item	*icp = ICR_ITEM(lip);
+	struct xfs_log_iovec	*vecp = NULL;
 
-	xlog_copy_iovec(&vecp, XLOG_REG_TYPE_ICREATE,
+	xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_ICREATE,
 			&icp->ic_format,
 			sizeof(struct xfs_icreate_log));
 }

fs/xfs/xfs_inode_item.c

···
 STATIC int
 xfs_inode_item_format_extents(
 	struct xfs_inode	*ip,
+	struct xfs_log_vec	*lv,
 	struct xfs_log_iovec	**vecp,
 	int			whichfork,
 	int			type)
···
 		ip->i_itemp->ili_aextents_buf = ext_buffer;
 
 	len = xfs_iextents_copy(ip, ext_buffer, whichfork);
-	xlog_copy_iovec(vecp, type, ext_buffer, len);
+	xlog_copy_iovec(lv, vecp, type, ext_buffer, len);
 	return len;
 }
 
···
 STATIC void
 xfs_inode_item_format_data_fork(
 	struct xfs_inode_log_item *iip,
-	struct xfs_log_iovec	**vecp,
-	int			*nvecs)
+	struct xfs_inode_log_format *ilf,
+	struct xfs_log_vec	*lv,
+	struct xfs_log_iovec	**vecp)
 {
 	struct xfs_inode	*ip = iip->ili_inode;
 	size_t			data_bytes;
···
 				 * extents, so just point to the
 				 * real extents array.
 				 */
-				xlog_copy_iovec(vecp, XLOG_REG_TYPE_IEXT,
+				xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_IEXT,
 						ip->i_df.if_u1.if_extents,
 						ip->i_df.if_bytes);
-				iip->ili_format.ilf_dsize = ip->i_df.if_bytes;
+				ilf->ilf_dsize = ip->i_df.if_bytes;
 			} else
 #endif
 			{
-				iip->ili_format.ilf_dsize =
-					xfs_inode_item_format_extents(ip, vecp,
+				ilf->ilf_dsize =
+					xfs_inode_item_format_extents(ip, lv, vecp,
 						XFS_DATA_FORK, XLOG_REG_TYPE_IEXT);
 				ASSERT(iip->ili_format.ilf_dsize <= ip->i_df.if_bytes);
 			}
-			(*nvecs)++;
+			ilf->ilf_size++;
 		} else {
 			iip->ili_fields &= ~XFS_ILOG_DEXT;
 		}
···
 		if ((iip->ili_fields & XFS_ILOG_DBROOT) &&
 		    ip->i_df.if_broot_bytes > 0) {
 			ASSERT(ip->i_df.if_broot != NULL);
-			xlog_copy_iovec(vecp, XLOG_REG_TYPE_IBROOT,
+			xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_IBROOT,
 					ip->i_df.if_broot,
 					ip->i_df.if_broot_bytes);
-			(*nvecs)++;
-			iip->ili_format.ilf_dsize = ip->i_df.if_broot_bytes;
+			ilf->ilf_dsize = ip->i_df.if_broot_bytes;
+			ilf->ilf_size++;
 		} else {
 			ASSERT(!(iip->ili_fields &
 				 XFS_ILOG_DBROOT));
···
 			       ip->i_df.if_real_bytes == data_bytes);
 			ASSERT(ip->i_df.if_u1.if_data != NULL);
 			ASSERT(ip->i_d.di_size > 0);
-			xlog_copy_iovec(vecp, XLOG_REG_TYPE_ILOCAL,
+			xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_ILOCAL,
 					ip->i_df.if_u1.if_data, data_bytes);
-			(*nvecs)++;
-			iip->ili_format.ilf_dsize = (unsigned)data_bytes;
+			ilf->ilf_dsize = (unsigned)data_bytes;
+			ilf->ilf_size++;
 		} else {
 			iip->ili_fields &= ~XFS_ILOG_DDATA;
 		}
···
 		iip->ili_fields &=
 			~(XFS_ILOG_DDATA | XFS_ILOG_DBROOT |
 			  XFS_ILOG_DEXT | XFS_ILOG_UUID);
-		if (iip->ili_fields & XFS_ILOG_DEV) {
-			iip->ili_format.ilf_u.ilfu_rdev =
-				ip->i_df.if_u2.if_rdev;
-		}
+		if (iip->ili_fields & XFS_ILOG_DEV)
+			ilf->ilf_u.ilfu_rdev = ip->i_df.if_u2.if_rdev;
 		break;
 	case XFS_DINODE_FMT_UUID:
 		iip->ili_fields &=
 			~(XFS_ILOG_DDATA | XFS_ILOG_DBROOT |
 			  XFS_ILOG_DEXT | XFS_ILOG_DEV);
-		if (iip->ili_fields & XFS_ILOG_UUID) {
-			iip->ili_format.ilf_u.ilfu_uuid =
-				ip->i_df.if_u2.if_uuid;
-		}
+		if (iip->ili_fields & XFS_ILOG_UUID)
+			ilf->ilf_u.ilfu_uuid = ip->i_df.if_u2.if_uuid;
 		break;
 	default:
 		ASSERT(0);
···
 STATIC void
 xfs_inode_item_format_attr_fork(
 	struct xfs_inode_log_item *iip,
-	struct xfs_log_iovec	**vecp,
-	int			*nvecs)
+	struct xfs_inode_log_format *ilf,
+	struct xfs_log_vec	*lv,
+	struct xfs_log_iovec	**vecp)
 {
 	struct xfs_inode	*ip = iip->ili_inode;
 	size_t			data_bytes;
···
 			 * There are not delayed allocation extents
 			 * for attributes, so just point at the array.
 			 */
-			xlog_copy_iovec(vecp, XLOG_REG_TYPE_IATTR_EXT,
+			xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_IATTR_EXT,
 					ip->i_afp->if_u1.if_extents,
 					ip->i_afp->if_bytes);
-			iip->ili_format.ilf_asize = ip->i_afp->if_bytes;
+			ilf->ilf_asize = ip->i_afp->if_bytes;
 #else
 			ASSERT(iip->ili_aextents_buf == NULL);
-			iip->ili_format.ilf_asize =
-				xfs_inode_item_format_extents(ip, vecp,
+			ilf->ilf_asize =
+				xfs_inode_item_format_extents(ip, lv, vecp,
 					XFS_ATTR_FORK, XLOG_REG_TYPE_IATTR_EXT);
 #endif
-			(*nvecs)++;
+			ilf->ilf_size++;
 		} else {
 			iip->ili_fields &= ~XFS_ILOG_AEXT;
 		}
···
 		    ip->i_afp->if_broot_bytes > 0) {
 			ASSERT(ip->i_afp->if_broot != NULL);
 
-			xlog_copy_iovec(vecp, XLOG_REG_TYPE_IATTR_BROOT,
+			xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_IATTR_BROOT,
 					ip->i_afp->if_broot,
 					ip->i_afp->if_broot_bytes);
-			(*nvecs)++;
-			iip->ili_format.ilf_asize = ip->i_afp->if_broot_bytes;
+			ilf->ilf_asize = ip->i_afp->if_broot_bytes;
+			ilf->ilf_size++;
 		} else {
 			iip->ili_fields &= ~XFS_ILOG_ABROOT;
 		}
···
 			ASSERT(ip->i_afp->if_real_bytes == 0 ||
 			       ip->i_afp->if_real_bytes == data_bytes);
 			ASSERT(ip->i_afp->if_u1.if_data != NULL);
-			xlog_copy_iovec(vecp, XLOG_REG_TYPE_IATTR_LOCAL,
+			xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_IATTR_LOCAL,
 					ip->i_afp->if_u1.if_data,
 					data_bytes);
-			(*nvecs)++;
-			iip->ili_format.ilf_asize = (unsigned)data_bytes;
+			ilf->ilf_asize = (unsigned)data_bytes;
+			ilf->ilf_size++;
 		} else {
 			iip->ili_fields &= ~XFS_ILOG_ADATA;
 		}
···
 STATIC void
 xfs_inode_item_format(
 	struct xfs_log_item	*lip,
-	struct xfs_log_iovec	*vecp)
+	struct xfs_log_vec	*lv)
 {
 	struct xfs_inode_log_item *iip = INODE_ITEM(lip);
 	struct xfs_inode	*ip = iip->ili_inode;
-	uint			nvecs;
+	struct xfs_inode_log_format *ilf;
+	struct xfs_log_iovec	*vecp = NULL;
 
-	xlog_copy_iovec(&vecp, XLOG_REG_TYPE_IFORMAT,
+	ilf = xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_IFORMAT,
 			&iip->ili_format,
 			sizeof(struct xfs_inode_log_format));
-	nvecs = 1;
-
-	xlog_copy_iovec(&vecp, XLOG_REG_TYPE_ICORE,
-			&ip->i_d,
-			xfs_icdinode_size(ip->i_d.di_version));
-	nvecs++;
+	ilf->ilf_size = 1;
 
 	if (ip->i_d.di_version == 1)
 		xfs_inode_item_format_v1_inode(ip);
+	xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_ICORE,
+			&ip->i_d,
+			xfs_icdinode_size(ip->i_d.di_version));
+	ilf->ilf_size++;
 
-	xfs_inode_item_format_data_fork(iip, &vecp, &nvecs);
+	xfs_inode_item_format_data_fork(iip, ilf, lv, &vecp);
 	if (XFS_IFORK_Q(ip)) {
-		xfs_inode_item_format_attr_fork(iip, &vecp, &nvecs);
+		xfs_inode_item_format_attr_fork(iip, ilf, lv, &vecp);
 	} else {
 		iip->ili_fields &=
 			~(XFS_ILOG_ADATA | XFS_ILOG_ABROOT | XFS_ILOG_AEXT);
···
 	 */
 	iip->ili_format.ilf_fields = XFS_ILOG_CORE |
 		(iip->ili_fields & ~XFS_ILOG_TIMESTAMP);
-	iip->ili_format.ilf_size = nvecs;
 }
 
 /*

fs/xfs/xfs_log.h

···
 #define XFS_LOG_VEC_ORDERED	(-1)
 
 static inline void *
-xlog_copy_iovec(struct xfs_log_iovec **vecp, uint type, void *data, int len)
+xlog_prepare_iovec(struct xfs_log_vec *lv, struct xfs_log_iovec **vecp,
+		uint type)
 {
 	struct xfs_log_iovec *vec = *vecp;
 
-	vec->i_type = type;
-	vec->i_addr = data;
-	vec->i_len = len;
+	if (vec) {
+		ASSERT(vec - lv->lv_iovecp < lv->lv_niovecs);
+		vec++;
+	} else {
+		vec = &lv->lv_iovecp[0];
+	}
 
-	*vecp = vec + 1;
+	vec->i_type = type;
+	vec->i_addr = lv->lv_buf + lv->lv_buf_len;
+
+	ASSERT(IS_ALIGNED((unsigned long)vec->i_addr, sizeof(uint64_t)));
+
+	*vecp = vec;
 	return vec->i_addr;
+}
+
+static inline void
+xlog_finish_iovec(struct xfs_log_vec *lv, struct xfs_log_iovec *vec, int len)
+{
+	/*
+	 * We need to make sure the next buffer is naturally aligned for the
+	 * biggest basic data type we put into it.  We already accounted for
+	 * this when sizing the buffer.
+	 */
+	lv->lv_buf_len += round_up(len, sizeof(uint64_t));
+	vec->i_len = len;
+}
+
+static inline void *
+xlog_copy_iovec(struct xfs_log_vec *lv, struct xfs_log_iovec **vecp,
+		uint type, void *data, int len)
+{
+	void *buf;
+
+	buf = xlog_prepare_iovec(lv, vecp, type);
+	memcpy(buf, data, len);
+	xlog_finish_iovec(lv, *vecp, len);
+	return buf;
 }
 
 /*

fs/xfs/xfs_log_cil.c

···
 								log->l_curr_block);
 }
 
-STATIC int
-xlog_cil_lv_item_format(
-	struct xfs_log_item	*lip,
-	struct xfs_log_vec	*lv)
-{
-	int	index;
-	char	*ptr;
-
-	/* format new vectors into array */
-	lip->li_ops->iop_format(lip, lv->lv_iovecp);
-
-	/* copy data into existing array */
-	ptr = lv->lv_buf;
-	for (index = 0; index < lv->lv_niovecs; index++) {
-		struct xfs_log_iovec *vec = &lv->lv_iovecp[index];
-
-		memcpy(ptr, vec->i_addr, vec->i_len);
-		vec->i_addr = ptr;
-		ptr += vec->i_len;
-	}
-
-	/*
-	 * some size calculations for log vectors over-estimate, so the caller
-	 * doesn't know the amount of space actually used by the item. Return
-	 * the byte count to the caller so they can check and store it
-	 * appropriately.
-	 */
-	return ptr - lv->lv_buf;
-}
-
 /*
  * Prepare the log item for insertion into the CIL. Calculate the difference in
  * log space and vectors it will consume, and if it is a new item pin it as
···
 			nbytes = 0;
 		}
 
+		/*
+		 * We 64-bit align the length of each iovec so that the start
+		 * of the next one is naturally aligned.  We'll need to
+		 * account for that slack space here.
+		 */
+		nbytes += niovecs * sizeof(uint64_t);
+
 		/* grab the old item if it exists for reservation accounting */
 		old_lv = lip->li_lv;
 
···
 		lv->lv_niovecs = niovecs;
 
 		/* The allocated data region lies beyond the iovec region */
+		lv->lv_buf_len = 0;
 		lv->lv_buf = (char *)lv + buf_size - nbytes;
-
-		lv->lv_buf_len = xlog_cil_lv_item_format(lip, lv);
+		lip->li_ops->iop_format(lip, lv);
 insert:
 		ASSERT(lv->lv_buf_len <= nbytes);
 		xfs_cil_prepare_item(log, lv, old_lv, diff_len, diff_iovecs);

fs/xfs/xfs_trans.h

···
 
 struct xfs_item_ops {
 	void (*iop_size)(xfs_log_item_t *, int *, int *);
-	void (*iop_format)(xfs_log_item_t *, struct xfs_log_iovec *);
+	void (*iop_format)(xfs_log_item_t *, struct xfs_log_vec *);
 	void (*iop_pin)(xfs_log_item_t *);
 	void (*iop_unpin)(xfs_log_item_t *, int remove);
 	uint (*iop_push)(struct xfs_log_item *, struct list_head *);