[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

[lvm-devel] [PATCH v3] Add --align parameter to pvcreate



- fixed pe_start alignment if value is so small that the whole
mda starts after initial pe_start

- I forgot that we have man lvm.conf.5 :-) fixed, also added
missing md alignment description and fixed missing backslash

No renames, this can be done easy before commit...
--

Add --align parameter to pvcreate.

Some RAID controllers suggest align volumes
to stripe or chunk size.

Currently lvm2 align LVs to underlying MD device chunk automatically,
but there is no proper way to set this parameter directly
for other RAID drivers.

Note that it is not as the same as metadata area size.

I think we should provide possibility for system administrator
to force PV data payload alignment during PV creation.
(Also it is administrator or distro installator responsibility
to set proper physical extent size later during vgcreate.)

Patch adds --align to pvcreate (default is size in KB).
Also it add pv_alignment to device section of lvm.conf,
where the default PV alignment can be specified.

pe_align() function now takes force_align parameter, which
can overwrite data alignment if set to non-zero.

Patch also fixes that after vgremove the orphan PV data align
value is not rewritten by default.

Signed-off-by: Milan Broz <mbroz redhat com>
---
 WHATS_NEW                        |    1 +
 doc/example.conf                 |   16 ++++++++---
 lib/format_text/format-text.c    |   54 ++++++++++++++++++++++++++++---------
 lib/metadata/metadata-exported.h |    1 +
 lib/metadata/metadata.c          |   50 ++++++++++++++++++++++++++++-------
 lib/metadata/metadata.h          |    2 +-
 man/lvm.conf.5.in                |   18 +++++++++++-
 man/pvcreate.8.in                |   11 +++++++
 tools/args.h                     |    1 +
 tools/commands.h                 |    5 ++-
 tools/pvcreate.c                 |   18 ++++++++++++-
 tools/vgconvert.c                |    2 +-
 12 files changed, 145 insertions(+), 34 deletions(-)

diff --git a/WHATS_NEW b/WHATS_NEW
index 29c3cd2..c9b3d8e 100644
--- a/WHATS_NEW
+++ b/WHATS_NEW
@@ -1,5 +1,6 @@
 Version 2.02.45 - 
 ===================================
+  Add --align parameter to pvcreate to allow specify payload align directly.
   Replace internal vg_check_status() implementation.
   Rename vg_read() to vg_read_internal().
 
diff --git a/doc/example.conf b/doc/example.conf
index 16cefb2..3bcfff0 100644
--- a/doc/example.conf
+++ b/doc/example.conf
@@ -86,7 +86,7 @@ devices {
     # If sysfs is mounted (2.6 kernels) restrict device scanning to 
     # the block devices it believes are valid.
     # 1 enables; 0 disables.
-    sysfs_scan = 1	
+    sysfs_scan = 1
 
     # By default, LVM2 will ignore devices used as components of
     # software RAID (md) devices by looking for md superblocks.
@@ -98,6 +98,14 @@ devices {
     # 1 enables; 0 disables.
     md_chunk_alignment = 1
 
+    # Default alignment (in KB) for PV, data payload will be aligned
+    # to multiple of this boundary.
+    # If not defined (or set to zero) alignment will be 64k or page size,
+    # if page size is bigger.
+    # Also note that specialized alignment (md_chunk_alignment) takes
+    # precedence before this setting.
+    #pv_alignment = 4096
+
     # If, while scanning the system for PVs, LVM2 encounters a device-mapper
     # device that has its I/O suspended, it waits for it to become accessible.
     # Set this to 1 to skip such devices.  This should only be needed
@@ -129,7 +137,7 @@ log {
     # There are 6 syslog-like log levels currently in use - 2 to 7 inclusive.
     # 7 is the most verbose (LOG_DEBUG).
     level = 0
-    
+
     # Format of output messages
     # Whether or not (1 or 0) to indent messages according to their severity
     indent = 1
@@ -175,7 +183,7 @@ backup {
     # Where should archived files go ?
     # Remember to back up this directory regularly!
     archive_dir = "/etc/lvm/archive"
-    
+
     # What is the minimum number of archive files you wish to keep ?
     retain_min = 10
 
@@ -193,7 +201,7 @@ shell {
 
 # Miscellaneous global LVM2 settings
 global {
-    
+
     # The file creation mask for any files and directories created.
     # Interpreted as octal if the first digit is zero.
     umask = 077
diff --git a/lib/format_text/format-text.c b/lib/format_text/format-text.c
index 428b5e2..a4ad44e 100644
--- a/lib/format_text/format-text.c
+++ b/lib/format_text/format-text.c
@@ -1182,7 +1182,7 @@ static int _mda_setup(const struct format_type *fmt,
 	if (!pvmetadatacopies)
 		return 1;
 
-	alignment = pe_align(pv) << SECTOR_SHIFT;
+	alignment = pe_align(pv, 0) << SECTOR_SHIFT;
 	disk_size = pv->size << SECTOR_SHIFT;
 	pe_start <<= SECTOR_SHIFT;
 	pe_end <<= SECTOR_SHIFT;
@@ -1294,6 +1294,37 @@ static int _mda_setup(const struct format_type *fmt,
 	return 1;
 }
 
+
+static void _pv_align_mda(struct physical_volume *pv,
+			  uint64_t mda_start_bytes,
+			  uint64_t mda_end_bytes)
+{
+	uint64_t pe_start_bytes = pv->pe_start << SECTOR_SHIFT;
+	uint64_t adjustment;
+
+	/*
+	 * Fix the offset if mda is in the beginning of device
+	 * but pe_start is too small
+	 */
+	if ((pe_start_bytes < 65536) &&
+	    (pe_start_bytes < mda_start_bytes)) {
+		pv->pe_start = mda_end_bytes >> SECTOR_SHIFT;
+		pe_start_bytes = pv->pe_start << SECTOR_SHIFT;
+	}
+
+	/*
+	 * Align pe_start that there is enough space for mda
+	 */
+	if ((mda_start_bytes <= pe_start_bytes) &&
+	    (mda_end_bytes > pe_start_bytes)) {
+		pv->pe_start = mda_end_bytes >> SECTOR_SHIFT;
+
+		adjustment = pv->pe_start % pe_align(pv, 0);
+		if (adjustment)
+			pv->pe_start += (pe_align(pv, 0) - adjustment);
+	}
+}
+
 /* Only for orphans */
 /* Set label_sector to -1 if rewriting existing label into same sector */
 static int _text_pv_write(const struct format_type *fmt, struct physical_volume *pv,
@@ -1305,7 +1336,6 @@ static int _text_pv_write(const struct format_type *fmt, struct physical_volume
 	struct metadata_area *mda;
 	char buf[MDA_HEADER_SIZE] __attribute((aligned(8)));
 	struct mda_header *mdah = (struct mda_header *) buf;
-	uint64_t adjustment;
 
 	/* FIXME Test mode don't update cache? */
 
@@ -1347,21 +1377,19 @@ static int _text_pv_write(const struct format_type *fmt, struct physical_volume
 	else
 		dm_list_init(&info->das);
 
+	/*
+	 * Keep PV data alignment if already set
+	 */
+	if (pv->pe_start < pe_align(pv, 0))
+		pv->pe_start = pe_align(pv, 0);
+
 	/* Set pe_start to first aligned sector after any metadata
 	 * areas that begin before pe_start */
-	pv->pe_start = pe_align(pv);
 	dm_list_iterate_items(mda, &info->mdas) {
 		mdac = (struct mda_context *) mda->metadata_locn;
-		if (pv->dev == mdac->area.dev &&
-		    (mdac->area.start <= (pv->pe_start << SECTOR_SHIFT)) &&
-		    (mdac->area.start + mdac->area.size >
-		     (pv->pe_start << SECTOR_SHIFT))) {
-			pv->pe_start = (mdac->area.start + mdac->area.size)
-			    >> SECTOR_SHIFT;
-			adjustment = pv->pe_start % pe_align(pv);
-			if (adjustment)
-				pv->pe_start += (pe_align(pv) - adjustment);
-		}
+		if (pv->dev == mdac->area.dev)
+			_pv_align_mda(pv, mdac->area.start,
+				      mdac->area.start + mdac->area.size);
 	}
 	if (!add_da
 	    (NULL, &info->das, pv->pe_start << SECTOR_SHIFT, UINT64_C(0)))
diff --git a/lib/metadata/metadata-exported.h b/lib/metadata/metadata-exported.h
index e785383..b25e30d 100644
--- a/lib/metadata/metadata-exported.h
+++ b/lib/metadata/metadata-exported.h
@@ -407,6 +407,7 @@ pv_t *pv_create(const struct cmd_context *cmd,
 		      struct device *dev,
 		      struct id *id,
 		      uint64_t size,
+		      unsigned long req_pe_align,
 		      uint64_t pe_start,
 		      uint32_t existing_extent_count,
 		      uint32_t existing_extent_size,
diff --git a/lib/metadata/metadata.c b/lib/metadata/metadata.c
index 0130bd3..cf7cb70 100644
--- a/lib/metadata/metadata.c
+++ b/lib/metadata/metadata.c
@@ -46,6 +46,7 @@ static struct physical_volume *_pv_read(struct cmd_context *cmd,
 static struct physical_volume *_pv_create(const struct format_type *fmt,
 				  struct device *dev,
 				  struct id *id, uint64_t size,
+				  unsigned long align,
 				  uint64_t pe_start,
 				  uint32_t existing_extent_count,
 				  uint32_t existing_extent_size,
@@ -65,19 +66,33 @@ static struct pv_list *_find_pv_in_vg(const struct volume_group *vg,
 static struct physical_volume *_find_pv_in_vg_by_uuid(const struct volume_group *vg,
 						      const struct id *id);
 
-unsigned long pe_align(struct physical_volume *pv)
+unsigned long pe_align(struct physical_volume *pv, unsigned long force_pe_align)
 {
-	if (pv->pe_align)
+	if (pv->pe_align && !force_pe_align)
 		goto out;
 
-	pv->pe_align = MAX(65536UL, lvm_getpagesize()) >> SECTOR_SHIFT;
+	if (force_pe_align)
+		pv->pe_align = force_pe_align;
+	else 
+		/*
+		 * Config value is in KB
+		 */
+		pv->pe_align = (find_config_tree_int(pv->fmt->cmd,
+				"devices/pv_alignment", 0) * 1024) >> SECTOR_SHIFT;
 
 	/*
-	 * Align to chunk size of underlying md device if present
+	 * Use old default if not specified
 	 */
+	if (!pv->pe_align)
+		pv->pe_align = MAX(65536UL, lvm_getpagesize()) >> SECTOR_SHIFT;
+
+
 	if (!pv->dev)
 		goto out;
 
+	/*
+	 * Align to chunk size of underlying md device if present
+	 */
 	if (find_config_tree_bool(pv->fmt->cmd, "devices/md_chunk_alignment",
 				  DEFAULT_MD_CHUNK_ALIGNMENT))
 		pv->pe_align = MAX(pv->pe_align,
@@ -148,8 +163,8 @@ int add_pv_to_vg(struct volume_group *vg, const char *pv_name,
 
 	/* FIXME Do proper rounding-up alignment? */
 	/* Reserved space for label; this holds 0 for PVs created by LVM1 */
-	if (pv->pe_start < pe_align(pv))
-		pv->pe_start = pe_align(pv);
+	if (pv->pe_start < pe_align(pv, 0))
+		pv->pe_start = pe_align(pv, 0);
 
 	/*
 	 * pe_count must always be calculated by pv_setup
@@ -759,6 +774,7 @@ int vg_split_mdas(struct cmd_context *cmd __attribute((unused)),
  * @dev: PV device to initialize
  * @id: PV UUID to use for initialization
  * @size: size of the PV in sectors
+ * @req_pe_align: requested alignment of payload
  * @pe_start: physical extent start
  * @existing_extent_count
  * @existing_extent_size
@@ -776,13 +792,14 @@ int vg_split_mdas(struct cmd_context *cmd __attribute((unused)),
 pv_t *pv_create(const struct cmd_context *cmd,
 		struct device *dev,
 		struct id *id, uint64_t size,
+		unsigned long req_pe_align,
 		uint64_t pe_start,
 		uint32_t existing_extent_count,
 		uint32_t existing_extent_size,
 		int pvmetadatacopies,
 		uint64_t pvmetadatasize, struct dm_list *mdas)
 {
-	return _pv_create(cmd->fmt, dev, id, size, pe_start,
+	return _pv_create(cmd->fmt, dev, id, size, req_pe_align, pe_start,
 			  existing_extent_count,
 			  existing_extent_size,
 			  pvmetadatacopies,
@@ -826,6 +843,7 @@ static struct physical_volume *_alloc_pv(struct dm_pool *mem, struct device *dev
 static struct physical_volume *_pv_create(const struct format_type *fmt,
 				  struct device *dev,
 				  struct id *id, uint64_t size,
+				  unsigned long req_pe_align,
 				  uint64_t pe_start,
 				  uint32_t existing_extent_count,
 				  uint32_t existing_extent_size,
@@ -860,15 +878,25 @@ static struct physical_volume *_pv_create(const struct format_type *fmt,
 		pv->size = size;
 	}
 
-	if (pv->size < PV_MIN_SIZE) {
-		log_error("%s: Size must exceed minimum of %ld sectors.",
-			  pv_dev_name(pv), PV_MIN_SIZE);
+	if (pv->size < (req_pe_align + PV_MIN_SIZE)) {
+		log_error("%s: Size %smust exceed minimum of %ld sectors.",
+			  pv_dev_name(pv),
+			  req_pe_align ? "after data alignment " : "",
+			  PV_MIN_SIZE);
 		goto bad;
 	}
 
 	pv->fmt = fmt;
 	pv->vg_name = fmt->orphan_vg_name;
 
+	/*
+	 * Force data alignment if specified
+	 */
+	if (req_pe_align && pe_align(pv, req_pe_align) != req_pe_align)
+		log_warn("%s: Overriding data aligmnent to %lu sectors"
+			 " (requested %lu sectors)", pv_dev_name(pv),
+			 pe_align(pv, 0), req_pe_align);
+
 	if (!fmt->ops->pv_setup(fmt, pe_start, existing_extent_count,
 				existing_extent_size,
 				pvmetadatacopies, pvmetadatasize, mdas,
@@ -877,6 +905,8 @@ static struct physical_volume *_pv_create(const struct format_type *fmt,
 			  "failed.", pv_dev_name(pv));
 		goto bad;
 	}
+
+
 	return pv;
 
       bad:
diff --git a/lib/metadata/metadata.h b/lib/metadata/metadata.h
index c425fbf..c982b78 100644
--- a/lib/metadata/metadata.h
+++ b/lib/metadata/metadata.h
@@ -263,7 +263,7 @@ struct format_handler {
 /*
  * Utility functions
  */
-unsigned long pe_align(struct physical_volume *pv);
+unsigned long pe_align(struct physical_volume *pv, unsigned long force_pe_align);
 int vg_validate(struct volume_group *vg);
 
 int pv_write_orphan(struct cmd_context *cmd, struct physical_volume *pv);
diff --git a/man/lvm.conf.5.in b/man/lvm.conf.5.in
index fdd728a..0d7f40f 100644
--- a/man/lvm.conf.5.in
+++ b/man/lvm.conf.5.in
@@ -124,14 +124,28 @@ you'll need \fBtypes = ["device-mapper", 16]\fP.  But if you do this,
 be careful to avoid recursion within LVM2.  The figure for number 
 of partitions is not currently used in LVM2 - and might never be.
 .IP
-\fBsysfs_scan\fP (em If set to 1 and your kernel supports sysfs and 
+\fBsysfs_scan\fP \(em If set to 1 and your kernel supports sysfs and 
 it is mounted, sysfs will be used as a quick way of filtering out
 block devices that are not present.
 .IP
-\fBmd_component_detection\fP (em If set to 1, LVM2 will ignore devices
+\fBmd_component_detection\fP \(em If set to 1, LVM2 will ignore devices
 used as components of software RAID (md) devices by looking for md
 superblocks. This doesn't always work satisfactorily e.g. if a device 
 has been reused without wiping the md superblocks first.
+.IP
+\fBmd_chunk_alignment\fP \(em If set to 1, and a PV is placed directly
+upon an md device, LVM2 will align its data blocks with the the chunk_size
+exposed in sysfs.
+.IP
+\fBpv_alignment\fP \(em Default alignment (in KB) for PV, data payload will
+be aligned to multiple of this boundary. If not defined (or set to zero)
+alignment will be 64k or page size, if page size is bigger.
+Also note that specialized alignment (md_chunk_alignment) takes precedence
+before this setting.
+.sp
+To print the current data payload offset for the Physical Volume use
+\fBpvs -o +pe_start\fP command.  The reported value is always multiple
+of requested alignment value.
 .TP
 \fBlog\fP \(em Default log settings
 .IP
diff --git a/man/pvcreate.8.in b/man/pvcreate.8.in
index 7ecda56..57e46e7 100644
--- a/man/pvcreate.8.in
+++ b/man/pvcreate.8.in
@@ -13,6 +13,7 @@ pvcreate \- initialize a disk or partition for use by LVM
 .RB [ \-M | \-\-metadatatype type ]
 .RB [ \-\-metadatacopies #copies ]
 .RB [ \-\-metadatasize size ]
+.RB [ \-\-align size ]
 .RB [ \-\-restorefile file ]
 .RB [ \-\-setphysicalvolumesize size ]
 .RB [ \-u | \-\-uuid uuid ]
@@ -89,6 +90,16 @@ to see where the metadata areas are placed.
 The approximate amount of space to be set aside for each metadata area.
 (The size you specify may get rounded.)
 .TP
+.BR \-\-align " size"
+Force align start of the payload to specified boundary (to align
+with underlying RAID device stripe or chunk).
+Note that you should use properly sized physical extent later
+in vgcreate to correctly align all Logical Volumes start too.
+.sp
+To print the current data payload offset for the Physical Volume use
+\fBpvs -o +pe_start\fP command.  The reported value is always multiple
+of requested alignment value.
+.TP
 .BR \-\-metadatacopies " copies"
 The number of metadata areas to set aside on each PV.  Currently
 this can be 0, 1 or 2.  
diff --git a/tools/args.h b/tools/args.h
index 8f026fc..776fc0c 100644
--- a/tools/args.h
+++ b/tools/args.h
@@ -56,6 +56,7 @@ arg(ignoremonitoring_ARG, '\0', "ignoremonitoring", NULL, 0)
 arg(nameprefixes_ARG, '\0', "nameprefixes", NULL, 0)
 arg(unquoted_ARG, '\0', "unquoted", NULL, 0)
 arg(rows_ARG, '\0', "rows", NULL, 0)
+arg(physicalvolumealign_ARG, '\0', "align", size_kb_arg, 0)
 
 /* Allow some variations */
 arg(resizable_ARG, '\0', "resizable", yes_no_arg, 0)
diff --git a/tools/commands.h b/tools/commands.h
index 58c6156..4007639 100644
--- a/tools/commands.h
+++ b/tools/commands.h
@@ -462,6 +462,7 @@ xx(pvcreate,
    "\t[-M|--metadatatype 1|2]" "\n"
    "\t[--metadatacopies #copies]" "\n"
    "\t[--metadatasize MetadataSize[kKmMgGtTpPeE]]" "\n"
+   "\t[--align Size[kKmM]]" "\n"
    "\t[--setphysicalvolumesize PhysicalVolumeSize[kKmMgGtTpPeE]" "\n"
    "\t[-t|--test] " "\n"
    "\t[-u|--uuid uuid] " "\n"
@@ -472,8 +473,8 @@ xx(pvcreate,
    "\tPhysicalVolume [PhysicalVolume...]\n",
 
    force_ARG, test_ARG, labelsector_ARG, metadatatype_ARG, metadatacopies_ARG,
-   metadatasize_ARG, physicalvolumesize_ARG, restorefile_ARG, uuidstr_ARG,
-   yes_ARG, zero_ARG)
+   metadatasize_ARG, physicalvolumealign_ARG, physicalvolumesize_ARG,
+   restorefile_ARG, uuidstr_ARG, yes_ARG, zero_ARG)
 
 xx(pvdata,
    "Display the on-disk metadata for physical volume(s)",
diff --git a/tools/pvcreate.c b/tools/pvcreate.c
index db1a0e2..9a12c33 100644
--- a/tools/pvcreate.c
+++ b/tools/pvcreate.c
@@ -19,6 +19,7 @@
 struct pvcreate_params {
 	int zero;
 	uint64_t size;
+	uint64_t align;
 	int pvmetadatacopies;
 	uint64_t pvmetadatasize;
 	int64_t labelsector;
@@ -177,7 +178,7 @@ static int pvcreate_single(struct cmd_context *cmd, const char *pv_name,
 	}
 
 	dm_list_init(&mdas);
-	if (!(pv = pv_create(cmd, dev, pp->idp, pp->size, pp->pe_start,
+	if (!(pv = pv_create(cmd, dev, pp->idp, pp->size, pp->align, pp->pe_start,
 			     pp->extent_count, pp->extent_size,
 			     pp->pvmetadatacopies,
 			     pp->pvmetadatasize,&mdas))) {
@@ -329,6 +330,21 @@ static int pvcreate_validate_params(struct cmd_context *cmd,
 	}
 	pp->size = arg_uint64_value(cmd, physicalvolumesize_ARG, UINT64_C(0));
 
+	if (arg_sign_value(cmd, physicalvolumealign_ARG, 0) == SIGN_MINUS) {
+		log_error("Physical volume alignment may not be negative");
+		return 0;
+	}
+	pp->align = arg_uint64_value(cmd, physicalvolumealign_ARG, UINT64_C(0));
+
+	if (pp->align > ULONG_MAX) {
+		log_error("Physical volume alignment is too big.");
+		return 0;
+	}
+	if (pp->align && pp->pe_start && (pp->pe_start % pp->align)) {
+		log_error("Incompatible alignment value is ignored.");
+		pp->align = 0;
+	}
+
 	if (arg_sign_value(cmd, metadatasize_ARG, 0) == SIGN_MINUS) {
 		log_error("Metadata size may not be negative");
 		return 0;
diff --git a/tools/vgconvert.c b/tools/vgconvert.c
index dc9d023..209d597 100644
--- a/tools/vgconvert.c
+++ b/tools/vgconvert.c
@@ -133,7 +133,7 @@ static int vgconvert_single(struct cmd_context *cmd, const char *vg_name,
 
 		dm_list_init(&mdas);
 		if (!(pv = pv_create(cmd, pv_dev(existing_pv),
-				     &existing_pv->id, size,
+				     &existing_pv->id, size, 0,
 				     pe_start, pv_pe_count(existing_pv),
 				     pv_pe_size(existing_pv), pvmetadatacopies,
 				     pvmetadatasize, &mdas))) {



[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]