snapshot: refactor snapshot-merge deptree and device removal
PATCH 1/1 snapshot: refactor snapshot-merge deptree and device removal
Add a merging snapshot using the "error" target rather than avoid adding it entirely. This allows proper cleanup of the -cow device without having to rename the -cow to use the origin's name as a prefix.
Moved the preloading of the origin LV, after a merge, from lv_remove_single() to vg_remove_single(). Having vg_remove_single() preload the origin free up the -cow device so that it can be removed via deactivate_lv(). This is reliable because the associated snapshot LV, that is to be removed when a snapshot-merge completes, is always added no to the deptree (and kernel -- via "error" target).
Now when the snapshot LV is removed both the -cow and -real devices get removed using uuid rather than device name. This paves the way for us to switch over to info-by-uuid queries.
--- lib/activate/dev_manager.c | 33 ++++++++++----------------------- lib/metadata/lv_manip.c | 32 ++++++++++---------------------- lib/metadata/snapshot_manip.c | 37 ++++++++++++++++++++++++++++++++----- tools/lvconvert.c | 5 ----- 4 files changed, 52 insertions(+), 55 deletions(-)
lib/activate/dev_manager.c +10 -23
@@ -1035,10 +1035,6 @@ static int _add_snapshot_target_to_dtree(struct dev_manager *dm,return 0 ; } -/* cow is to be merged so skip adding it */ -if (lv_is_merging_cow (lv)) -return 1 ; -if (!(origin_dlid =build_dlid (dm, snap_seg->origin->lvid.s,"real" ))) return_0;
@@ -1047,7 +1043,13 @@ static int _add_snapshot_target_to_dtree(struct dev_manager *dm, size = (uint64_t ) snap_seg->len * snap_seg->origin->vg->extent_size; -if (!dm_tree_node_add_snapshot_target (dnode, size, origin_dlid, cow_dlid,1 , snap_seg->chunk_size)) +if (lv_is_merging_cow (lv)) { +/* cow is to be merged so load the error target */ +if (!dm_tree_node_add_error_target (dnode, size)) + return_0; + } +else if (!dm_tree_node_add_snapshot_target (dnode, size, origin_dlid, + cow_dlid,1 , snap_seg->chunk_size)) return_0;return 1 ;
@@ -1166,7 +1168,7 @@ static int _add_new_lv_to_dtree(struct dev_manager *dm, struct dm_tree *dtree,struct lv_layer *lvlayer;struct dm_tree_node *dnode;const struct dm_info *dinfo; -char *name, *dlid, *lv_name; +char *name, *dlid;uint32_t max_stripe_size =UINT32_C (0 );uint32_t read_ahead = lv->read_ahead;uint32_t read_ahead_flags =UINT32_C (0 );
@@ -1192,22 +1194,7 @@ static int _add_new_lv_to_dtree(struct dev_manager *dm, struct dm_tree *dtree, } } - lv_name = lv->name; -if (lv_is_cow (lv) &&lv_is_merging_cow (lv)) { -if (layer) { -/* -* use origin's name as basis for snapshot-merge device names; -* this allows _clean_tree to automatically cleanup "-cow" -* when the origin is resumed (after merge completes) -*/ - lv_name =origin_from_cow (lv)->name; - }else { -/* top-level snapshot device is not needed during merge */ -return 1 ; - } - } - -if (!(name =build_dm_name (dm->mem, lv->vg->name, lv_name, layer))) +if (!(name =build_dm_name (dm->mem, lv->vg->name, lv->name, layer))) return_0;if (!(dlid =build_dlid (dm, lv->lvid.s, layer)))
@@ -1220,7 +1207,7 @@ static int _add_new_lv_to_dtree(struct dev_manager *dm, struct dm_tree *dtree,if (!(lvlayer =dm_pool_alloc (dm->mem,sizeof (*lvlayer)))) {log_error ("_add_new_lv_to_dtree: pool alloc failed for %s %s." , - lv_name, layer); + lv->name, layer);return 0 ; }
lib/metadata/lv_manip.c +10 -22
@@ -2064,7 +2064,6 @@ int lv_remove_single(struct cmd_context *cmd, struct logical_volume *lv,struct volume_group *vg;struct lvinfo info;struct logical_volume *origin =NULL ; -int was_merging =0 , preload_origin =0 ; vg = lv->vg;
@@ -2117,24 +2116,18 @@ int lv_remove_single(struct cmd_context *cmd, struct logical_volume *lv,if (!archive (vg))return 0 ; -/* FIXME Snapshot commit out of sequence if it fails after here? */ -if (!deactivate_lv (cmd, lv)) { -log_error ("Unable to deactivate logical volume \"%s\"" , - lv->name); -return 0 ; - } -if (lv_is_cow (lv)) { origin =origin_from_cow (lv); - was_merging =lv_is_merging_origin (origin);log_verbose ("Removing snapshot %s" , lv->name); +/* vg_remove_snapshot() will preload origin if it was merging */ if (!vg_remove_snapshot (lv)) return_0; -if (was_merging &&lv_is_origin (origin)) { -/* snapshot(s) still exist. preload new origin prior to suspend. */ -/* FIXME Seek a simpler way of dealing with this within the tree. */ - preload_origin =1 ; - } + } + +if (!deactivate_lv (cmd, lv)) { +log_error ("Unable to deactivate logical volume \"%s\"" , + lv->name); +return 0 ; }log_verbose ("Releasing logical volume \"%s\"" , lv->name);
@@ -2144,20 +2137,15 @@ int lv_remove_single(struct cmd_context *cmd, struct logical_volume *lv, }/* store it on disks */ -if (!vg_write (vg)) - return_0; - -if (!preload_origin && !vg_commit (vg)) +if (!vg_write (vg) || !vg_commit (vg)) return_0; -/* If no snapshots left or if we stopped merging, reload */ -if (origin && (!lv_is_origin (origin) || was_merging)) { +/* If no snapshots left, reload without -real. */ +if (origin && (!lv_is_origin (origin))) {if (!suspend_lv (cmd, origin)) {log_error ("Failed to refresh %s without snapshot." , origin->name);return 0 ; } -if (preload_origin && !vg_commit (vg)) - return_0;if (!resume_lv (cmd, origin)) {log_error ("Failed to resume %s." , origin->name);return 0 ;
lib/metadata/snapshot_manip.c +32 -5
@@ -15,6 +15,7 @@#include "lib.h" #include "metadata.h" +#include "locking.h" #include "toolcontext.h" #include "lv_alloc.h"
@@ -58,9 +59,9 @@ int lv_is_merging_origin(const struct logical_volume *origin)struct lv_segment *find_merging_cow (const struct logical_volume *origin) { -/* FIXME: eliminate this wrapper and just use find_cow()? -* - find_merging_cow() adds to code clarity in caller -*/ +if (!lv_is_merging_origin (origin)) +return NULL ; +return find_cow (origin); }
@@ -170,12 +171,22 @@ int vg_add_snapshot(struct logical_volume *origin,int vg_remove_snapshot (struct logical_volume *cow) { +int preload_origin =0 ;struct logical_volume *origin =origin_from_cow (cow);dm_list_del (&cow->snapshot->origin_list); origin->origin_count--; -if (find_merging_cow (origin) ==find_cow (cow)) -clear_snapshot_merge (origin_from_cow (cow)); +if (find_merging_cow (origin) ==find_cow (cow)) { +clear_snapshot_merge (origin); +/* +* preload origin to: +* - allow proper release of -cow +* - avoid allocations with other devices suspended +* when transitioning from "snapshot-merge" to +* "snapshot-origin after a merge completes. +*/ + preload_origin =1 ; + }if (!lv_remove (cow->snapshot->lv)) { log_error("Failed to remove internal snapshot LV %s",
@@ -186,5 +197,21 @@ int vg_remove_snapshot(struct logical_volume *cow) cow->snapshot =NULL ;lv_set_visible (cow); +if (preload_origin) { +if (!vg_write (origin->vg)) + return_0; +if (!suspend_lv (origin->vg->cmd, origin)) { +log_error ("Failed to refresh %s without snapshot." , + origin->name); +return 0 ; + } +if (!vg_commit (origin->vg)) + return_0; +if (!resume_lv (origin->vg->cmd, origin)) { +log_error ("Failed to resume %s." , origin->name); +return 0 ; + } + } +return 1 ; }
tools/lvconvert.c +0 -5
@@ -1210,11 +1210,6 @@ static int lvconvert_merge(struct cmd_context *cmd, goto out; } -if (!deactivate_lv (cmd, lv)) { -log_warn ("WARNING: Unable to deactivate merging snapshot %s" , lv->name); -/* merge is running regardless of this deactivation failure */ - } - lp->need_polling =1 ; lp->lv_to_poll = origin;
Quoted 12 lines · Mike Snitzer
Add a merging snapshot using the "error" target rather than avoid adding it entirely. This allows proper cleanup of the -cow device without having to rename the -cow to use the origin's name as a prefix.
Moved the preloading of the origin LV, after a merge, from
lv_remove_single()tovg_remove_single(). Havingvg_remove_single()preload the origin free up the -cow device so that it can be removed viadeactivate_lv(). This is reliable because the associated snapshot LV, that is to be removed when a snapshot-merge completes, is always added no to the deptree (and kernel -- via "error" target).
Quoted 3 lines
Now when the snapshot LV is removed both the -cow and -real devices get removed using uuid rather than device name. This paves the way for us to switch over to info-by-uuid queries.
I tested this patch with and without Milan's info-by-uuid patches here:
http://mazyland.englab.brq.redhat.com/test-shared/
Mike