[lvm-devel] master - pvmove: correcting read_ahead setting

[Zdenek Kabelac]

Gitweb:        https://sourceware.org/git/?p=lvm2.git;a=commitdiff;h=0451225c1997273ffaa36c3eb595eae7737b7dda
Commit:        0451225c1997273ffaa36c3eb595eae7737b7dda
Parent:        a53c157db0110e21aaa9971f0fec7ac19afe91dd
Author:        Zdenek Kabelac <zkabelac at redhat.com>
AuthorDate:    Tue Aug 20 12:30:25 2019 +0200
Committer:     Zdenek Kabelac <zkabelac at redhat.com>
CommitterDate: Tue Aug 20 12:37:32 2019 +0200

pvmove: correcting read_ahead setting

When pvmove is finished, we do a tricky operation since we try to
resume multiple different device that were all joined into 1 big tree.

Currently we use the infromation from existing live DM table,
where we can get list of all holders of pvmove device.
We look for these nodes (by uuid) in new metadata, and we do now a full
regular device add into dm tree structure.  All devices should be
already PRELOAD with correct table before entering suspend state,
however for correctly working readahead we need to put correct info
also into RESUME tree.  Since table are preloaded, the same table
is skip and resume, but correct read ahead is now set.
---
 WHATS_NEW                  |    1 +
 lib/activate/dev_manager.c |   47 +++++++++++++++++++++++++++++++------------
 2 files changed, 35 insertions(+), 13 deletions(-)

diff --git a/WHATS_NEW b/WHATS_NEW
index 63c3c27..91543a7 100644
--- a/WHATS_NEW
+++ b/WHATS_NEW
@@ -1,5 +1,6 @@
 Version 2.03.06 - 
 ================================
+  Correctly set read_ahead for LVs when pvmove is finished.
   Remove unsupported OPTIONS+="event_timeout" udev rule from 11-dm-lvm.rules.
   Prevent creating VGs with PVs with different logical block sizes.
   Fix metadata writes from corrupting with large physical block size.
diff --git a/lib/activate/dev_manager.c b/lib/activate/dev_manager.c
index 981f467..b218225 100644
--- a/lib/activate/dev_manager.c
+++ b/lib/activate/dev_manager.c
@@ -2034,10 +2034,20 @@ static uint16_t _get_udev_flags(struct dev_manager *dm, const struct logical_vol
 
 static int _add_lv_to_dtree(struct dev_manager *dm, struct dm_tree *dtree,
 			    const struct logical_volume *lv, int origin_only);
-
+static int _add_new_lv_to_dtree(struct dev_manager *dm, struct dm_tree *dtree,
+				const struct logical_volume *lv,
+				struct lv_activate_opts *laopts,
+				const char *layer);
+/*
+ * Check for device holders (ATM used only for removed pvmove targets)
+ * and add them into dtree structures.
+ * When 'laopts != NULL' add them as new nodes - which also corrects READ_AHEAD.
+ * Note: correct table are already explicitelly PRELOADED.
+ */
 static int _check_holder(struct dev_manager *dm, struct dm_tree *dtree,
-			 const struct logical_volume *lv, uint32_t major,
-			 const char *d_name)
+			 const struct logical_volume *lv,
+			 struct lv_activate_opts *laopts,
+			 uint32_t major, const char *d_name)
 {
 	const char *default_uuid_prefix = dm_uuid_prefix();
 	const size_t default_uuid_prefix_len = strlen(default_uuid_prefix);
@@ -2089,8 +2099,11 @@ static int _check_holder(struct dev_manager *dm, struct dm_tree *dtree,
 			log_debug_activation("Found holder %s of %s.",
 					     display_lvname(lv_det),
 					     display_lvname(lv));
-			if (!_add_lv_to_dtree(dm, dtree, lv_det, 0))
-				goto_out;
+			if (!laopts) {
+				if (!_add_lv_to_dtree(dm, dtree, lv_det, 0))
+					goto_out;
+			} else if (!_add_new_lv_to_dtree(dm, dtree, lv_det, laopts, 0))
+					goto_out;
 		}
 	}
 
@@ -2107,7 +2120,9 @@ out:
  * i.e. PVMOVE is being finished and final table is going to be resumed.
  */
 static int _add_holders_to_dtree(struct dev_manager *dm, struct dm_tree *dtree,
-				 const struct logical_volume *lv, struct dm_info *info)
+				 const struct logical_volume *lv,
+				 struct lv_activate_opts *laopts,
+				 const struct dm_info *info)
 {
 	const char *sysfs_dir = dm_sysfs_dir();
 	char sysfs_path[PATH_MAX];
@@ -2130,7 +2145,7 @@ static int _add_holders_to_dtree(struct dev_manager *dm, struct dm_tree *dtree,
 	while ((dirent = readdir(d)))
 		/* Expects minor is added to 'dm-' prefix */
 		if (!strncmp(dirent->d_name, "dm-", 3) &&
-		    !_check_holder(dm, dtree, lv, info->major, dirent->d_name))
+		    !_check_holder(dm, dtree, lv, laopts, info->major, dirent->d_name))
 			goto_out;
 
 	r = 1;
@@ -2202,7 +2217,7 @@ static int _add_dev_to_dtree(struct dev_manager *dm, struct dm_tree *dtree,
 	 */
 	if (info.exists && !lv_is_pvmove(lv) &&
 	    !strchr(lv->name, '_') && !strncmp(lv->name, "pvmove", 6))
-		if (!_add_holders_to_dtree(dm, dtree, lv, &info))
+		if (!_add_holders_to_dtree(dm, dtree, lv, NULL, &info))
 			return_0;
 
 	return 1;
@@ -3000,11 +3015,6 @@ static int _add_target_to_dtree(struct dev_manager *dm,
 						  &dm->pvmove_mirror_count);
 }
 
-static int _add_new_lv_to_dtree(struct dev_manager *dm, struct dm_tree *dtree,
-				const struct logical_volume *lv,
-				struct lv_activate_opts *laopts,
-				const char *layer);
-
 static int _add_new_external_lv_to_dtree(struct dev_manager *dm,
 					 struct dm_tree *dtree,
 					 struct logical_volume *external_lv,
@@ -3439,6 +3449,17 @@ static int _add_new_lv_to_dtree(struct dev_manager *dm, struct dm_tree *dtree,
 	    !_pool_register_callback(dm, dnode, lv))
 		return_0;
 
+	/*
+	 * Update tables for ANY PVMOVE holders for active LV where the name starts with 'pvmove',
+	 * but it's not anymore PVMOVE LV and also it's not a PVMOVE _mimage LV.
+	 * When resume happens, tables MUST be already preloaded with correct entries!
+	 * (since we can't preload different table while devices are suspended)
+	 */
+	if (!lv_is_pvmove(lv) && !strncmp(lv->name, "pvmove", 6) && !strchr(lv->name, '_') &&
+	    (dinfo = _cached_dm_info(dm->mem, dtree, lv, NULL)))
+		if (!_add_holders_to_dtree(dm, dtree, lv, laopts, dinfo))
+			return_0;
+
 	if (read_ahead == DM_READ_AHEAD_AUTO) {
 		/* we need RA at least twice a whole stripe - see the comment in md/raid0.c */
 		read_ahead = max_stripe_size * 2;