mds: check multiversion inode when purging stray

we can't purge a multiversion directory inode when it is still referenced
by some snapshots. For regular files that are still referenced by some
snapshots, we need to keep the backtrace object.

Signed-off-by: Yan, Zheng <zyan@redhat.com>

diff --git a/src/mds/MDCache.cc b/src/mds/MDCache.cc
index 3e2203e021ce8e196cf2b251303063e2d81fe2a9..2a9d5f0a1575086564ae21a80a92a60a67cf88cd 100644 (file)
--- a/src/mds/MDCache.cc
+++ b/src/mds/MDCache.cc
@@ -8944,19 +8944,19 @@ void MDCache::eval_stray(CDentry *dn, bool delay)
 
   // purge?
   if (in->inode.nlink == 0) {
-    if (in->is_dir()) {
       // past snaprealm parents imply snapped dentry remote links.
       // only important for directories.  normal file data snaps are handled
       // by the object store.
+    if (in->snaprealm && in->snaprealm->has_past_parents()) {
+      if (!in->snaprealm->have_past_parents_open() &&
+         !in->snaprealm->open_parents(new C_MDC_EvalStray(this, dn)))
+       return;
+      in->snaprealm->prune_past_parents();
+    }
+    if (in->is_dir()) {
       if (in->snaprealm && in->snaprealm->has_past_parents()) {
-       if (!in->snaprealm->have_past_parents_open() &&
-           !in->snaprealm->open_parents(new C_MDC_EvalStray(this, dn)))
-         return;
-       in->snaprealm->prune_past_parents();
-       if (in->snaprealm->has_past_parents()) {
-         dout(20) << "  has past parents " << in->snaprealm->srnode.past_parents << dendl;
-         return;  // not until some snaps are deleted.
-       }
+       dout(20) << "  directory has past parents " << in->snaprealm->srnode.past_parents << dendl;
+       return;  // not until some snaps are deleted.
       }
       if (in->has_dirfrags()) {
        list<CDir*> ls;
@@ -8996,6 +8996,11 @@ void MDCache::eval_stray(CDentry *dn, bool delay)
        num_strays_delayed++;
        logger->set(l_mdc_num_strays_delayed, num_strays_delayed);
       }
+    } else if (in->snaprealm && in->snaprealm->has_past_parents()) {
+      assert(!in->is_dir());
+      dout(20) << " file has past parents " << in->snaprealm->srnode.past_parents << dendl;
+      if (in->is_file() && in->get_projected_inode()->size > 0)
+       truncate_stray(dn); // truncate head objects
     } else {
       if (in->is_dir())
        in->close_dirfrags();
@@ -9048,15 +9053,68 @@ void MDCache::fetch_backtrace(inodeno_t ino, int64_t pool, bufferlist& bl, Conte
 
 class C_IO_MDC_PurgeStrayPurged : public MDCacheIOContext {
   CDentry *dn;
+  bool only_head;
 public:
-  C_IO_MDC_PurgeStrayPurged(MDCache *c, CDentry *d) : 
-    MDCacheIOContext(c), dn(d) { }
+  C_IO_MDC_PurgeStrayPurged(MDCache *c, CDentry *d, bool oh) :
+    MDCacheIOContext(c), dn(d), only_head(oh) { }
   void finish(int r) {
     assert(r == 0 || r == -ENOENT);
-    mdcache->_purge_stray_purged(dn, r);
+    mdcache->_purge_stray_purged(dn, only_head);
   }
 };
 
+void MDCache::truncate_stray(CDentry *dn)
+{
+  CDentry::linkage_t *dnl = dn->get_projected_linkage();
+  CInode *in = dnl->get_inode();
+  dout(10) << "truncate_stray " << *dn << " " << *in << dendl;
+  assert(!dn->is_replicated());
+
+  dn->state_set(CDentry::STATE_PURGING);
+  dn->get(CDentry::PIN_PURGING);
+  in->state_set(CInode::STATE_PURGING);
+
+  if (dn->item_stray.is_on_list())
+    dn->item_stray.remove_myself();
+
+  C_GatherBuilder gather(
+    g_ceph_context,
+    new C_OnFinisher(new C_IO_MDC_PurgeStrayPurged(this, dn, true),
+                    &mds->finisher));
+
+  SnapRealm *realm = in->find_snaprealm();
+  assert(realm);
+  dout(10) << " realm " << *realm << dendl;
+  const SnapContext *snapc = &realm->get_snap_context();
+
+  uint64_t period = (uint64_t)in->inode.layout.fl_object_size *
+                   (uint64_t)in->inode.layout.fl_stripe_count;
+  uint64_t to = in->inode.get_max_size();
+  to = MAX(in->inode.size, to);
+  // when truncating a file, the filer does not delete stripe objects that are
+  // truncated to zero. so we need to purge stripe objects up to the max size
+  // the file has ever been.
+  to = MAX(in->inode.max_size_ever, to);
+  if (period && to > period) {
+    uint64_t num = (to - 1) / period;
+    dout(10) << "purge_stray 0~" << to << " objects 0~" << num
+      << " snapc " << snapc << " on " << *in << dendl;
+    mds->filer->purge_range(in->ino(), &in->inode.layout, *snapc,
+                           1, num, ceph_clock_now(g_ceph_context),
+                           0, gather.new_sub());
+  }
+
+  // keep backtrace object
+  if (period && to > 0) {
+    mds->filer->zero(in->ino(), &in->inode.layout, *snapc,
+                    0, period, ceph_clock_now(g_ceph_context),
+                    0, true, NULL, gather.new_sub());
+  }
+
+  assert(gather.has_subs());
+  gather.activate();
+}
+
 void MDCache::purge_stray(CDentry *dn)
 {
   CDentry::linkage_t *dnl = dn->get_projected_linkage();
@@ -9087,7 +9145,8 @@ void MDCache::purge_stray(CDentry *dn)
   SnapContext nullsnapc;
   C_GatherBuilder gather(
     g_ceph_context,
-    new C_OnFinisher(new C_IO_MDC_PurgeStrayPurged(this, dn), &mds->finisher));
+    new C_OnFinisher(new C_IO_MDC_PurgeStrayPurged(this, dn, false),
+                    &mds->finisher));
 
   if (in->is_dir()) {
     object_locator_t oloc(mds->mdsmap->get_metadata_pool());
@@ -9184,13 +9243,13 @@ public:
   }
 };
 
-void MDCache::_purge_stray_purged(CDentry *dn, int r)
+void MDCache::_purge_stray_purged(CDentry *dn, bool only_head)
 {
-  assert (r == 0 || r == -ENOENT);
   CInode *in = dn->get_projected_linkage()->get_inode();
   dout(10) << "_purge_stray_purged " << *dn << " " << *in << dendl;
 
-  if (in->get_num_ref() == (int)in->is_dirty() &&
+  if (!only_head &&
+      in->get_num_ref() == (int)in->is_dirty() &&
       dn->get_num_ref() == (int)dn->is_dirty() + !!in->get_num_ref() + 1/*PIN_PURGING*/) {
     // kill dentry.
     version_t pdv = dn->pre_dirty();
@@ -9228,6 +9287,7 @@ void MDCache::_purge_stray_purged(CDentry *dn, int r)
     
     inode_t *pi = in->project_inode();
     pi->size = 0;
+    pi->max_size_ever = 0;
     pi->client_ranges.clear();
     pi->truncate_size = 0;
     pi->truncate_from = 0;
@@ -9287,8 +9347,6 @@ void MDCache::_purge_stray_logged_truncate(CDentry *dn, LogSegment *ls)
   eval_stray(dn);
 }
 
-
-
 void MDCache::reintegrate_stray(CDentry *straydn, CDentry *rdn)
 {
   dout(10) << "reintegrate_stray " << *straydn << " into " << *rdn << dendl;

diff --git a/src/mds/MDCache.h b/src/mds/MDCache.h
index b1459cb97b94e5fd3be744fb29dfe72f5a4a6a74..c28cc9d397dd6a667bf0b64873396d2008b16db3 100644 (file)
--- a/src/mds/MDCache.h
+++ b/src/mds/MDCache.h
@@ -912,8 +912,9 @@ public:
 
 protected:
   void scan_stray_dir(dirfrag_t next=dirfrag_t());
+  void truncate_stray(CDentry *dn);
   void purge_stray(CDentry *dn);
-  void _purge_stray_purged(CDentry *dn, int r=0);
+  void _purge_stray_purged(CDentry *dn, bool only_head);
   void _purge_stray_logged(CDentry *dn, version_t pdv, LogSegment *ls);
   void _purge_stray_logged_truncate(CDentry *dn, LogSegment *ls);
   friend struct C_MDC_RetryScanStray;