summaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
Diffstat
-rw-r--r--src/include/compact_map.h15
-rw-r--r--src/include/compact_set.h19
-rw-r--r--src/mds/CDir.cc15
-rw-r--r--src/mds/CDir.h1
-rw-r--r--src/mds/CInode.cc547
-rw-r--r--src/mds/CInode.h248
-rw-r--r--src/mds/Locker.cc129
-rw-r--r--src/mds/Locker.h8
-rw-r--r--src/mds/MDCache.cc120
-rw-r--r--src/mds/MDCache.h2
-rw-r--r--src/mds/MDSDaemon.cc3
-rw-r--r--src/mds/RecoveryQueue.cc2
-rw-r--r--src/mds/Server.cc373
-rw-r--r--src/mds/StrayManager.cc40
-rw-r--r--src/mds/events/EMetaBlob.h21
-rw-r--r--src/mds/journal.cc19
-rw-r--r--src/mds/mdstypes.cc350
-rw-r--r--src/mds/mdstypes.h441
-rw-r--r--src/test/encoding/types.h4
-rw-r--r--src/tools/cephfs/DataScan.cc2
20 files changed, 1202 insertions, 1157 deletions
diff --git a/src/include/compact_map.h b/src/include/compact_map.h
index 52a7792adc8..9ccf97c1139 100644
--- a/src/include/compact_map.h
+++ b/src/include/compact_map.h
@@ -181,13 +181,18 @@ public:
size_t count (const Key& k) const {
return map ? map->count(k) : 0;
}
- void erase (iterator p) {
+ iterator erase (iterator p) {
if (map) {
assert(this == p.map);
- map->erase(p.it);
+ auto it = map->erase(p.it);
if (map->empty()) {
free_internal();
+ return iterator(this);
+ } else {
+ return iterator(this, it);
}
+ } else {
+ return iterator(this);
}
}
size_t erase (const Key& k) {
@@ -216,6 +221,12 @@ public:
alloc_internal();
return iterator(this, map->insert(val));
}
+ template <class... Args>
+ std::pair<iterator,bool> emplace ( Args&&... args ) {
+ alloc_internal();
+ auto em = map->emplace(std::forward<Args>(args)...);
+ return std::pair<iterator,bool>(iterator(this, em.first), em.second);
+ }
iterator begin() {
if (!map)
return iterator(this);
diff --git a/src/include/compact_set.h b/src/include/compact_set.h
index dd2ddd31165..e41ca95ed81 100644
--- a/src/include/compact_set.h
+++ b/src/include/compact_set.h
@@ -140,12 +140,18 @@ public:
size_t count(const T& t) const {
return set ? set->count(t) : 0;
}
- void erase (iterator p) {
+ iterator erase (iterator p) {
if (set) {
assert(this == p.set);
- set->erase(p.it);
- if (set->empty())
+ auto it = set->erase(p.it);
+ if (set->empty()) {
free_internal();
+ return iterator(this);
+ } else {
+ return iterator(this, it);
+ }
+ } else {
+ return iterator(this);
}
}
size_t erase (const T& t) {
@@ -175,6 +181,13 @@ public:
std::pair<typename Set::iterator,bool> r = set->insert(t);
return std::make_pair(iterator(this, r.first), r.second);
}
+ template <class... Args>
+ std::pair<iterator,bool> emplace ( Args&&... args ) {
+ alloc_internal();
+ auto em = set->emplace(std::forward<Args>(args)...);
+ return std::pair<iterator,bool>(iterator(this, em.first), em.second);
+ }
+
iterator begin() {
if (!set)
return iterator(this);
diff --git a/src/mds/CDir.cc b/src/mds/CDir.cc
index cb0cd07778d..9fb6a323fe8 100644
--- a/src/mds/CDir.cc
+++ b/src/mds/CDir.cc
@@ -200,14 +200,11 @@ CDir::CDir(CInode *in, frag_t fg, MDCache *mdcache, bool auth) :
num_dentries_auth_subtree_nested(0),
dir_auth(CDIR_AUTH_DEFAULT)
{
- state = STATE_INITIAL;
-
memset(&fnode, 0, sizeof(fnode));
// auth
assert(in->is_dir());
- if (auth)
- state |= STATE_AUTH;
+ if (auth) state_set(STATE_AUTH);
}
/**
@@ -825,7 +822,7 @@ void CDir::steal_dentry(CDentry *dn)
if (dn->get_linkage()->is_primary()) {
CInode *in = dn->get_linkage()->get_inode();
- inode_t *pi = in->get_projected_inode();
+ auto pi = in->get_projected_inode();
if (dn->get_linkage()->get_inode()->is_dir())
fnode.fragstat.nsubdirs++;
else
@@ -1143,7 +1140,7 @@ void CDir::merge(list<CDir*>& subs, list<MDSInternalContextBase*>& waiters, bool
void CDir::resync_accounted_fragstat()
{
fnode_t *pf = get_projected_fnode();
- inode_t *pi = inode->get_projected_inode();
+ auto pi = inode->get_projected_inode();
if (pf->accounted_fragstat.version != pi->dirstat.version) {
pf->fragstat.version = pi->dirstat.version;
@@ -1158,7 +1155,7 @@ void CDir::resync_accounted_fragstat()
void CDir::resync_accounted_rstat()
{
fnode_t *pf = get_projected_fnode();
- inode_t *pi = inode->get_projected_inode();
+ auto pi = inode->get_projected_inode();
if (pf->accounted_rstat.version != pi->rstat.version) {
pf->rstat.version = pi->rstat.version;
@@ -1178,8 +1175,8 @@ void CDir::assimilate_dirty_rstat_inodes()
if (in->is_frozen())
continue;
- inode_t *pi = in->project_inode();
- pi->version = in->pre_dirty();
+ auto &pi = in->project_inode();
+ pi.inode.version = in->pre_dirty();
inode->mdcache->project_rstat_inode_to_frag(in, this, 0, 0, NULL);
}
diff --git a/src/mds/CDir.h b/src/mds/CDir.h
index f9137491829..3e6bc0eb2d2 100644
--- a/src/mds/CDir.h
+++ b/src/mds/CDir.h
@@ -95,7 +95,6 @@ public:
// common states
static const unsigned STATE_CLEAN = 0;
- static const unsigned STATE_INITIAL = 0;
// these state bits are preserved by an import/export
// ...except if the directory is hashed, in which case none of them are!
diff --git a/src/mds/CInode.cc b/src/mds/CInode.cc
index ab68a7e054e..e01a0611393 100644
--- a/src/mds/CInode.cc
+++ b/src/mds/CInode.cc
@@ -147,14 +147,14 @@ ostream& operator<<(ostream& out, const CInode& in)
if (in.is_frozen_inode()) out << " FROZEN";
if (in.is_frozen_auth_pin()) out << " FROZEN_AUTHPIN";
- const inode_t *pi = in.get_projected_inode();
+ const CInode::mempool_inode *pi = in.get_projected_inode();
if (pi->is_truncating())
out << " truncating(" << pi->truncate_from << " to " << pi->truncate_size << ")";
if (in.inode.is_dir()) {
out << " " << in.inode.dirstat;
if (g_conf->mds_debug_scatterstat && in.is_projected()) {
- const inode_t *pi = in.get_projected_inode();
+ const CInode::mempool_inode *pi = in.get_projected_inode();
out << "->" << pi->dirstat;
}
} else {
@@ -168,7 +168,7 @@ ostream& operator<<(ostream& out, const CInode& in)
if (!(in.inode.rstat == in.inode.accounted_rstat))
out << "/" << in.inode.accounted_rstat;
if (g_conf->mds_debug_scatterstat && in.is_projected()) {
- const inode_t *pi = in.get_projected_inode();
+ const CInode::mempool_inode *pi = in.get_projected_inode();
out << "->" << pi->rstat;
if (!(pi->rstat == pi->accounted_rstat))
out << "/" << pi->accounted_rstat;
@@ -229,12 +229,12 @@ ostream& operator<<(ostream& out, const CInode& in)
}
if (!in.get_mds_caps_wanted().empty()) {
out << " mcw={";
- for (compact_map<int,int>::const_iterator p = in.get_mds_caps_wanted().begin();
- p != in.get_mds_caps_wanted().end();
- ++p) {
- if (p != in.get_mds_caps_wanted().begin())
+ bool first = true;
+ for (const auto &p : in.get_mds_caps_wanted()) {
+ if (!first)
out << ',';
- out << p->first << '=' << ccap_string(p->second);
+ out << p.first << '=' << ccap_string(p.second);
+ first = false;
}
out << '}';
}
@@ -283,7 +283,7 @@ void CInode::add_need_snapflush(CInode *snapin, snapid_t snapid, client_t client
auth_pin(this); // pin head inode...
}
- set<client_t>& clients = client_need_snapflush[snapid];
+ auto &clients = client_need_snapflush[snapid];
if (clients.empty())
snapin->auth_pin(this); // ...and pin snapped/old inode!
@@ -293,18 +293,18 @@ void CInode::add_need_snapflush(CInode *snapin, snapid_t snapid, client_t client
void CInode::remove_need_snapflush(CInode *snapin, snapid_t snapid, client_t client)
{
dout(10) << __func__ << " client." << client << " snapid " << snapid << " on " << snapin << dendl;
- compact_map<snapid_t, std::set<client_t> >::iterator p = client_need_snapflush.find(snapid);
- if (p == client_need_snapflush.end()) {
+ auto it = client_need_snapflush.find(snapid);
+ if (it == client_need_snapflush.end()) {
dout(10) << " snapid not found" << dendl;
return;
}
- if (!p->second.count(client)) {
+ size_t n = it->second.erase(client);
+ if (n == 0) {
dout(10) << " client not found" << dendl;
return;
}
- p->second.erase(client);
- if (p->second.empty()) {
- client_need_snapflush.erase(p);
+ if (it->second.empty()) {
+ client_need_snapflush.erase(it);
snapin->auth_unpin(this);
if (client_need_snapflush.empty()) {
@@ -318,16 +318,16 @@ bool CInode::split_need_snapflush(CInode *cowin, CInode *in)
{
dout(10) << __func__ << " [" << cowin->first << "," << cowin->last << "] for " << *cowin << dendl;
bool need_flush = false;
- for (compact_map<snapid_t, set<client_t> >::iterator p = client_need_snapflush.lower_bound(cowin->first);
- p != client_need_snapflush.end() && p->first < in->first; ) {
- compact_map<snapid_t, set<client_t> >::iterator q = p;
- ++p;
- assert(!q->second.empty());
- if (cowin->last >= q->first) {
+ for (auto it = client_need_snapflush.lower_bound(cowin->first);
+ it != client_need_snapflush.end() && it->first < in->first; ) {
+ assert(!it->second.empty());
+ if (cowin->last >= it->first) {
cowin->auth_pin(this);
need_flush = true;
- } else
- client_need_snapflush.erase(q);
+ ++it;
+ } else {
+ it = client_need_snapflush.erase(it);
+ }
in->auth_unpin(this);
}
return need_flush;
@@ -361,105 +361,102 @@ void CInode::clear_dirty_rstat()
}
}
-inode_t *CInode::project_inode(map<string,bufferptr> *px)
+/* Ideally this function would be subsumed by project_inode but it is also
+ * needed by CInode::project_past_snaprealm_parent so we keep it.
+ */
+sr_t &CInode::project_snaprealm(projected_inode &pi)
{
- if (projected_nodes.empty()) {
- projected_nodes.push_back(new projected_inode_t(new inode_t(inode)));
- if (px)
- *px = xattrs;
+ const sr_t *cur_srnode = get_projected_srnode();
+
+ assert(!pi.snapnode);
+ if (cur_srnode) {
+ pi.snapnode.reset(new sr_t(*cur_srnode));
} else {
- projected_nodes.push_back(new projected_inode_t(
- new inode_t(*projected_nodes.back()->inode)));
- if (px)
- *px = *get_projected_xattrs();
+ pi.snapnode.reset(new sr_t());
+ pi.snapnode->created = 0;
+ pi.snapnode->current_parent_since = get_oldest_snap();
}
+ ++num_projected_srnodes;
- projected_inode_t &pi = *projected_nodes.back();
+ dout(10) << __func__ << " " << pi.snapnode.get() << dendl;
+ return *pi.snapnode.get();
+}
- if (px) {
- pi.xattrs = px;
- ++num_projected_xattrs;
- }
+CInode::projected_inode &CInode::project_inode(bool xattr, bool snap)
+{
+ auto &pi = projected_nodes.empty() ?
+ projected_nodes.emplace_back(inode) :
+ projected_nodes.emplace_back(projected_nodes.back().inode);
if (scrub_infop && scrub_infop->last_scrub_dirty) {
- pi.inode->last_scrub_stamp = scrub_infop->last_scrub_stamp;
- pi.inode->last_scrub_version = scrub_infop->last_scrub_version;
+ pi.inode.last_scrub_stamp = scrub_infop->last_scrub_stamp;
+ pi.inode.last_scrub_version = scrub_infop->last_scrub_version;
scrub_infop->last_scrub_dirty = false;
scrub_maybe_delete_info();
}
- dout(15) << __func__ << " " << pi.inode << dendl;
- return pi.inode;
+
+ if (xattr) {
+ pi.xattrs.reset(new mempool_xattr_map(*get_projected_xattrs()));
+ ++num_projected_xattrs;
+ }
+
+ if (snap) {
+ project_snaprealm(pi);
+ }
+
+ dout(15) << __func__ << " " << pi.inode.ino << dendl;
+ return pi;
}
void CInode::pop_and_dirty_projected_inode(LogSegment *ls)
{
assert(!projected_nodes.empty());
- dout(15) << __func__ << " " << projected_nodes.front()->inode
- << " v" << projected_nodes.front()->inode->version << dendl;
+ auto &front = projected_nodes.front();
+ dout(15) << __func__ << " " << front.inode.ino
+ << " v" << front.inode.version << dendl;
int64_t old_pool = inode.layout.pool_id;
- mark_dirty(projected_nodes.front()->inode->version, ls);
- inode = *projected_nodes.front()->inode;
+ mark_dirty(front.inode.version, ls);
+ inode = front.inode;
if (inode.is_backtrace_updated())
_mark_dirty_parent(ls, old_pool != inode.layout.pool_id);
- map<string,bufferptr> *px = projected_nodes.front()->xattrs;
- if (px) {
+ if (front.xattrs) {
--num_projected_xattrs;
- xattrs = *px;
- delete px;
+ xattrs = *front.xattrs;
}
- if (projected_nodes.front()->snapnode) {
- pop_projected_snaprealm(projected_nodes.front()->snapnode);
+ auto &snapnode = front.snapnode;
+ if (snapnode) {
+ pop_projected_snaprealm(snapnode.get());
--num_projected_srnodes;
}
- delete projected_nodes.front()->inode;
- delete projected_nodes.front();
-
projected_nodes.pop_front();
}
-sr_t *CInode::project_snaprealm(snapid_t snapid)
-{
- const sr_t *cur_srnode = get_projected_srnode();
- sr_t *new_srnode;
-
- if (cur_srnode) {
- new_srnode = new sr_t(*cur_srnode);
- } else {
- new_srnode = new sr_t();
- new_srnode->created = snapid;
- new_srnode->current_parent_since = get_oldest_snap();
- }
- dout(10) << __func__ << " " << new_srnode << dendl;
- projected_nodes.back()->snapnode = new_srnode;
- ++num_projected_srnodes;
- return new_srnode;
-}
-
/* if newparent != parent, add parent to past_parents
if parent DNE, we need to find what the parent actually is and fill that in */
void CInode::project_past_snaprealm_parent(SnapRealm *newparent)
{
- sr_t *new_snap = project_snaprealm();
+ assert(!projected_nodes.empty());
+ sr_t &new_snap = project_snaprealm(projected_nodes.back());
SnapRealm *oldparent;
if (!snaprealm) {
oldparent = find_snaprealm();
- new_snap->seq = oldparent->get_newest_seq();
+ new_snap.seq = oldparent->get_newest_seq();
}
else
oldparent = snaprealm->parent;
if (newparent != oldparent) {
snapid_t oldparentseq = oldparent->get_newest_seq();
- if (oldparentseq + 1 > new_snap->current_parent_since) {
- new_snap->past_parents[oldparentseq].ino = oldparent->inode->ino();
- new_snap->past_parents[oldparentseq].first = new_snap->current_parent_since;
+ if (oldparentseq + 1 > new_snap.current_parent_since) {
+ new_snap.past_parents[oldparentseq].ino = oldparent->inode->ino();
+ new_snap.past_parents[oldparentseq].first = new_snap.current_parent_since;
}
- new_snap->current_parent_since = std::max(oldparentseq, newparent->get_last_created()) + 1;
+ new_snap.current_parent_since = std::max(oldparentseq, newparent->get_last_created()) + 1;
}
}
@@ -481,7 +478,6 @@ void CInode::pop_projected_snaprealm(sr_t *next_snaprealm)
<< " -> " << next_snaprealm->past_parents << dendl;
}
snaprealm->srnode = *next_snaprealm;
- delete next_snaprealm;
// we should be able to open these up (or have them already be open).
bool ok = snaprealm->_open_parents(NULL);
@@ -520,7 +516,7 @@ frag_t InodeStoreBase::pick_dirfrag(std::string_view dn)
bool CInode::get_dirfrags_under(frag_t fg, list<CDir*>& ls)
{
bool all = true;
- list<frag_t> fglist;
+ std::list<frag_t> fglist;
dirfragtree.get_leaves_under(fg, fglist);
for (list<frag_t>::iterator p = fglist.begin(); p != fglist.end(); ++p)
if (dirfrags.count(*p))
@@ -533,19 +529,20 @@ bool CInode::get_dirfrags_under(frag_t fg, list<CDir*>& ls)
fragtree_t tmpdft;
tmpdft.force_to_leaf(g_ceph_context, fg);
- for (compact_map<frag_t,CDir*>::iterator p = dirfrags.begin(); p != dirfrags.end(); ++p) {
- tmpdft.force_to_leaf(g_ceph_context, p->first);
- if (fg.contains(p->first) && !dirfragtree.is_leaf(p->first))
- ls.push_back(p->second);
+ for (auto &p : dirfrags) {
+ tmpdft.force_to_leaf(g_ceph_context, p.first);
+ if (fg.contains(p.first) && !dirfragtree.is_leaf(p.first))
+ ls.push_back(p.second);
}
all = true;
tmpdft.get_leaves_under(fg, fglist);
- for (list<frag_t>::iterator p = fglist.begin(); p != fglist.end(); ++p)
- if (!dirfrags.count(*p)) {
+ for (const auto &p : fglist) {
+ if (!dirfrags.count(p)) {
all = false;
break;
}
+ }
return all;
}
@@ -553,10 +550,10 @@ bool CInode::get_dirfrags_under(frag_t fg, list<CDir*>& ls)
void CInode::verify_dirfrags()
{
bool bad = false;
- for (compact_map<frag_t,CDir*>::iterator p = dirfrags.begin(); p != dirfrags.end(); ++p) {
- if (!dirfragtree.is_leaf(p->first)) {
- dout(0) << "have open dirfrag " << p->first << " but not leaf in " << dirfragtree
- << ": " << *p->second << dendl;
+ for (const auto &p : dirfrags) {
+ if (!dirfragtree.is_leaf(p.first)) {
+ dout(0) << "have open dirfrag " << p.first << " but not leaf in " << dirfragtree
+ << ": " << *p.second << dendl;
bad = true;
}
}
@@ -566,10 +563,10 @@ void CInode::verify_dirfrags()
void CInode::force_dirfrags()
{
bool bad = false;
- for (compact_map<frag_t,CDir*>::iterator p = dirfrags.begin(); p != dirfrags.end(); ++p) {
- if (!dirfragtree.is_leaf(p->first)) {
- dout(0) << "have open dirfrag " << p->first << " but not leaf in " << dirfragtree
- << ": " << *p->second << dendl;
+ for (auto &p : dirfrags) {
+ if (!dirfragtree.is_leaf(p.first)) {
+ dout(0) << "have open dirfrag " << p.first << " but not leaf in " << dirfragtree
+ << ": " << *p.second << dendl;
bad = true;
}
}
@@ -604,31 +601,28 @@ CDir *CInode::get_approx_dirfrag(frag_t fg)
return NULL;
}
-void CInode::get_dirfrags(list<CDir*>& ls)
+void CInode::get_dirfrags(std::list<CDir*>& ls)
{
// all dirfrags
- for (compact_map<frag_t,CDir*>::iterator p = dirfrags.begin();
- p != dirfrags.end();
- ++p)
- ls.push_back(p->second);
+ for (const auto &p : dirfrags) {
+ ls.push_back(p.second);
+ }
}
void CInode::get_nested_dirfrags(list<CDir*>& ls)
{
// dirfrags in same subtree
- for (compact_map<frag_t,CDir*>::iterator p = dirfrags.begin();
- p != dirfrags.end();
- ++p)
- if (!p->second->is_subtree_root())
- ls.push_back(p->second);
+ for (const auto &p : dirfrags) {
+ if (!p.second->is_subtree_root())
+ ls.push_back(p.second);
+ }
}
void CInode::get_subtree_dirfrags(list<CDir*>& ls)
{
// dirfrags that are roots of new subtrees
- for (compact_map<frag_t,CDir*>::iterator p = dirfrags.begin();
- p != dirfrags.end();
- ++p)
- if (p->second->is_subtree_root())
- ls.push_back(p->second);
+ for (const auto &p : dirfrags) {
+ if (p.second->is_subtree_root())
+ ls.push_back(p.second);
+ }
}
@@ -696,23 +690,21 @@ void CInode::close_dirfrags()
bool CInode::has_subtree_root_dirfrag(int auth)
{
- for (compact_map<frag_t,CDir*>::iterator p = dirfrags.begin();
- p != dirfrags.end();
- ++p)
- if (p->second->is_subtree_root() &&
- (auth == -1 || p->second->dir_auth.first == auth))
+ for (const auto &p : dirfrags) {
+ if (p.second->is_subtree_root() &&
+ (auth == -1 || p.second->dir_auth.first == auth))
return true;
+ }
return false;
}
bool CInode::has_subtree_or_exporting_dirfrag()
{
- for (compact_map<frag_t,CDir*>::iterator p = dirfrags.begin();
- p != dirfrags.end();
- ++p)
- if (p->second->is_subtree_root() ||
- p->second->state_test(CDir::STATE_EXPORTING))
+ for (const auto &p : dirfrags) {
+ if (p.second->is_subtree_root() ||
+ p.second->state_test(CDir::STATE_EXPORTING))
return true;
+ }
return false;
}
@@ -720,11 +712,9 @@ void CInode::get_stickydirs()
{
if (stickydir_ref == 0) {
get(PIN_STICKYDIRS);
- for (compact_map<frag_t,CDir*>::iterator p = dirfrags.begin();
- p != dirfrags.end();
- ++p) {
- p->second->state_set(CDir::STATE_STICKY);
- p->second->get(CDir::PIN_STICKY);
+ for (const auto &p : dirfrags) {
+ p.second->state_set(CDir::STATE_STICKY);
+ p.second->get(CDir::PIN_STICKY);
}
}
stickydir_ref++;
@@ -736,11 +726,9 @@ void CInode::put_stickydirs()
stickydir_ref--;
if (stickydir_ref == 0) {
put(PIN_STICKYDIRS);
- for (compact_map<frag_t,CDir*>::iterator p = dirfrags.begin();
- p != dirfrags.end();
- ++p) {
- p->second->state_clear(CDir::STATE_STICKY);
- p->second->put(CDir::PIN_STICKY);
+ for (const auto &p : dirfrags) {
+ p.second->state_clear(CDir::STATE_STICKY);
+ p.second->put(CDir::PIN_STICKY);
}
}
}
@@ -899,11 +887,11 @@ version_t CInode::pre_dirty()
assert(is_base());
pv = get_projected_version() + 1;
}
- // force update backtrace for old format inode (see inode_t::decode)
+ // force update backtrace for old format inode (see mempool_inode::decode)
if (inode.backtrace_version == 0 && !projected_nodes.empty()) {
- inode_t *pi = projected_nodes.back()->inode;
- if (pi->backtrace_version == 0)
- pi->update_backtrace(pv);
+ mempool_inode &pi = projected_nodes.back().inode;
+ if (pi.backtrace_version == 0)
+ pi.update_backtrace(pv);
}
return pv;
}
@@ -1137,12 +1125,10 @@ void CInode::build_backtrace(int64_t pool, inode_backtrace_t& bt)
in = diri;
pdn = in->get_parent_dn();
}
- for (compact_set<int64_t>::iterator i = inode.old_pools.begin();
- i != inode.old_pools.end();
- ++i) {
+ for (auto &p : inode.old_pools) {
// don't add our own pool id to old_pools to avoid looping (e.g. setlayout 0, 1, 0)
- if (*i != pool)
- bt.old_pools.insert(*i);
+ if (p != pool)
+ bt.old_pools.insert(p);
}
}
@@ -1204,20 +1190,18 @@ void CInode::store_backtrace(MDSInternalContextBase *fin, int op_prio)
// In the case where DIRTYPOOL is set, we update all old pools backtraces
// such that anyone reading them will see the new pool ID in
// inode_backtrace_t::pool and go read everything else from there.
- for (compact_set<int64_t>::iterator p = inode.old_pools.begin();
- p != inode.old_pools.end();
- ++p) {
- if (*p == pool)
+ for (const auto &p : inode.old_pools) {
+ if (p == pool)
continue;
- dout(20) << __func__ << ": updating old pool " << *p << dendl;
+ dout(20) << __func__ << ": updating old pool " << p << dendl;
ObjectOperation op;
op.priority = op_prio;
op.create(false);
op.setxattr("parent", parent_bl);
- object_locator_t oloc(*p);
+ object_locator_t oloc(p);
mdcache->mds->objecter->mutate(oid, oloc, op, snapc,
ceph::real_clock::now(),
0, gather.new_sub());
@@ -1366,8 +1350,11 @@ void InodeStoreBase::decode_bare(bufferlist::iterator &bl,
{
using ceph::decode;
decode(inode, bl);
- if (is_symlink())
- decode(symlink, bl);
+ if (is_symlink()) {
+ std::string tmp;
+ decode(tmp, bl);
+ symlink = std::string_view(tmp);
+ }
decode(dirfragtree, bl);
decode(xattrs, bl);
decode(snap_blob, bl);
@@ -1492,11 +1479,9 @@ void CInode::encode_lock_state(int type, bufferlist& bl)
encode(inode.dirstat, bl); // only meaningful if i am auth.
bufferlist tmp;
__u32 n = 0;
- for (compact_map<frag_t,CDir*>::iterator p = dirfrags.begin();
- p != dirfrags.end();
- ++p) {
- frag_t fg = p->first;
- CDir *dir = p->second;
+ for (const auto &p : dirfrags) {
+ frag_t fg = p.first;
+ CDir *dir = p.second;
if (is_auth() || dir->is_auth()) {
fnode_t *pf = dir->get_projected_fnode();
dout(15) << fg << " " << *dir << dendl;
@@ -1527,11 +1512,9 @@ void CInode::encode_lock_state(int type, bufferlist& bl)
encode(inode.rstat, bl); // only meaningful if i am auth.
bufferlist tmp;
__u32 n = 0;
- for (compact_map<frag_t,CDir*>::iterator p = dirfrags.begin();
- p != dirfrags.end();
- ++p) {
- frag_t fg = p->first;
- CDir *dir = p->second;
+ for (const auto &p : dirfrags) {
+ frag_t fg = p.first;
+ CDir *dir = p.second;
if (is_auth() || dir->is_auth()) {
fnode_t *pf = dir->get_projected_fnode();
dout(10) << fg << " " << *dir << dendl;
@@ -1653,15 +1636,13 @@ void CInode::decode_lock_state(int type, bufferlist& bl)
// dft was scattered, or we may still be be waiting on the
// notify from the auth)
dirfragtree.swap(temp);
- for (compact_map<frag_t,CDir*>::iterator p = dirfrags.begin();
- p != dirfrags.end();
- ++p) {
- if (!dirfragtree.is_leaf(p->first)) {
- dout(10) << " forcing open dirfrag " << p->first << " to leaf (racing with split|merge)" << dendl;
- dirfragtree.force_to_leaf(g_ceph_context, p->first);
+ for (const auto &p : dirfrags) {
+ if (!dirfragtree.is_leaf(p.first)) {
+ dout(10) << " forcing open dirfrag " << p.first << " to leaf (racing with split|merge)" << dendl;
+ dirfragtree.force_to_leaf(g_ceph_context, p.first);
}
- if (p->second->is_auth())
- p->second->state_clear(CDir::STATE_DIRTYDFT);
+ if (p.second->is_auth())
+ p.second->state_clear(CDir::STATE_DIRTYDFT);
}
}
if (g_conf->mds_debug_frag)
@@ -1898,13 +1879,11 @@ void CInode::start_scatter(ScatterLock *lock)
{
dout(10) << __func__ << " " << *lock << " on " << *this << dendl;
assert(is_auth());
- inode_t *pi = get_projected_inode();
+ mempool_inode *pi = get_projected_inode();
- for (compact_map<frag_t,CDir*>::iterator p = dirfrags.begin();
- p != dirfrags.end();
- ++p) {
- frag_t fg = p->first;
- CDir *dir = p->second;
+ for (const auto &p : dirfrags) {
+ frag_t fg = p.first;
+ CDir *dir = p.second;
fnode_t *pf = dir->get_projected_fnode();
dout(20) << fg << " " << *dir << dendl;
@@ -1960,7 +1939,7 @@ void CInode::finish_scatter_update(ScatterLock *lock, CDir *dir,
MutationRef mut(new MutationImpl());
mut->ls = mdlog->get_current_segment();
- inode_t *pi = get_projected_inode();
+ mempool_inode *pi = get_projected_inode();
fnode_t *pf = dir->project_fnode();
const char *ename = 0;
@@ -2064,16 +2043,14 @@ void CInode::finish_scatter_gather_update(int type)
// adjust summation
assert(is_auth());
- inode_t *pi = get_projected_inode();
+ mempool_inode *pi = get_projected_inode();
bool touched_mtime = false, touched_chattr = false;
dout(20) << " orig dirstat " << pi->dirstat << dendl;
pi->dirstat.version++;
- for (compact_map<frag_t,CDir*>::iterator p = dirfrags.begin();
- p != dirfrags.end();
- ++p) {
- frag_t fg = p->first;
- CDir *dir = p->second;
+ for (const auto &p : dirfrags) {
+ frag_t fg = p.first;
+ CDir *dir = p.second;
dout(20) << fg << " " << *dir << dendl;
bool update;
@@ -2175,14 +2152,12 @@ void CInode::finish_scatter_gather_update(int type)
// adjust summation
assert(is_auth());
- inode_t *pi = get_projected_inode();
+ mempool_inode *pi = get_projected_inode();
dout(20) << " orig rstat " << pi->rstat << dendl;
pi->rstat.version++;
- for (compact_map<frag_t,CDir*>::iterator p = dirfrags.begin();
- p != dirfrags.end();
- ++p) {
- frag_t fg = p->first;
- CDir *dir = p->second;
+ for (const auto &p : dirfrags) {
+ frag_t fg = p.first;
+ CDir *dir = p.second;
dout(20) << fg << " " << *dir << dendl;
bool update;
@@ -2274,10 +2249,8 @@ void CInode::finish_scatter_gather_update_accounted(int type, MutationRef& mut,
dout(10) << __func__ << " " << type << " on " << *this << dendl;
assert(is_auth());
- for (compact_map<frag_t,CDir*>::iterator p = dirfrags.begin();
- p != dirfrags.end();
- ++p) {
- CDir *dir = p->second;
+ for (const auto &p : dirfrags) {
+ CDir *dir = p.second;
if (!dir->is_auth() || dir->get_version() == 0 || dir->is_frozen())
continue;
@@ -2332,11 +2305,11 @@ void CInode::take_dir_waiting(frag_t fg, list<MDSInternalContextBase*>& ls)
if (waiting_on_dir.empty())
return;
- compact_map<frag_t, list<MDSInternalContextBase*> >::iterator p = waiting_on_dir.find(fg);
- if (p != waiting_on_dir.end()) {
+ auto it = waiting_on_dir.find(fg);
+ if (it != waiting_on_dir.end()) {
dout(10) << __func__ << " frag " << fg << " on " << *this << dendl;
- ls.splice(ls.end(), p->second);
- waiting_on_dir.erase(p);
+ ls.splice(ls.end(), it->second);
+ waiting_on_dir.erase(it);
if (waiting_on_dir.empty())
put(PIN_DIRWAITER);
@@ -2368,10 +2341,10 @@ void CInode::take_waiting(uint64_t mask, list<MDSInternalContextBase*>& ls)
if ((mask & WAIT_DIR) && !waiting_on_dir.empty()) {
// take all dentry waiters
while (!waiting_on_dir.empty()) {
- compact_map<frag_t, list<MDSInternalContextBase*> >::iterator p = waiting_on_dir.begin();
- dout(10) << __func__ << " dirfrag " << p->first << " on " << *this << dendl;
- ls.splice(ls.end(), p->second);
- waiting_on_dir.erase(p);
+ auto it = waiting_on_dir.begin();
+ dout(10) << __func__ << " dirfrag " << it->first << " on " << *this << dendl;
+ ls.splice(ls.end(), it->second);
+ waiting_on_dir.erase(it);
}
put(PIN_DIRWAITER);
}
@@ -2524,10 +2497,8 @@ void CInode::adjust_nested_auth_pins(int a, void *by)
if (g_conf->mds_debug_auth_pins) {
// audit
int s = 0;
- for (compact_map<frag_t,CDir*>::iterator p = dirfrags.begin();
- p != dirfrags.end();
- ++p) {
- CDir *dir = p->second;
+ for (const auto &p : dirfrags) {
+ CDir *dir = p.second;
if (!dir->is_subtree_root() && dir->get_cum_auth_pins())
s++;
}
@@ -2568,14 +2539,14 @@ snapid_t CInode::get_oldest_snap()
return std::min(t, oldest_snap);
}
-old_inode_t& CInode::cow_old_inode(snapid_t follows, bool cow_head)
+CInode::mempool_old_inode& CInode::cow_old_inode(snapid_t follows, bool cow_head)
{
assert(follows >= first);
- inode_t *pi = cow_head ? get_projected_inode() : get_previous_projected_inode();
- map<string,bufferptr> *px = cow_head ? get_projected_xattrs() : get_previous_projected_xattrs();
+ mempool_inode *pi = cow_head ? get_projected_inode() : get_previous_projected_inode();
+ mempool_xattr_map *px = cow_head ? get_projected_xattrs() : get_previous_projected_xattrs();
- old_inode_t &old = old_inodes[follows];
+ mempool_old_inode &old = old_inodes[follows];
old.first = first;
old.inode = *pi;
old.xattrs = *px;
@@ -2602,15 +2573,15 @@ old_inode_t& CInode::cow_old_inode(snapid_t follows, bool cow_head)
void CInode::split_old_inode(snapid_t snap)
{
- compact_map<snapid_t, old_inode_t>::iterator p = old_inodes.lower_bound(snap);
- assert(p != old_inodes.end() && p->second.first < snap);
+ auto it = old_inodes.lower_bound(snap);
+ assert(it != old_inodes.end() && it->second.first < snap);
- old_inode_t &old = old_inodes[snap - 1];
- old = p->second;
+ mempool_old_inode &old = old_inodes[snap - 1];
+ old = it->second;
- p->second.first = snap;
- dout(10) << __func__ << " " << "[" << old.first << "," << p->first
- << "] to [" << snap << "," << p->first << "] on " << *this << dendl;
+ it->second.first = snap;
+ dout(10) << __func__ << " " << "[" << old.first << "," << it->first
+ << "] to [" << snap << "," << it->first << "] on " << *this << dendl;
}
void CInode::pre_cow_old_inode()
@@ -2642,29 +2613,27 @@ void CInode::purge_stale_snap_data(const set<snapid_t>& snaps)
{
dout(10) << __func__ << " " << snaps << dendl;
- if (old_inodes.empty())
- return;
-
- compact_map<snapid_t,old_inode_t>::iterator p = old_inodes.begin();
- while (p != old_inodes.end()) {
- set<snapid_t>::const_iterator q = snaps.lower_bound(p->second.first);
- if (q == snaps.end() || *q > p->first) {
- dout(10) << " purging old_inode [" << p->second.first << "," << p->first << "]" << dendl;
- old_inodes.erase(p++);
- } else
- ++p;
+ for (auto it = old_inodes.begin(); it != old_inodes.end(); ) {
+ const snapid_t &id = it->first;
+ const auto &s = snaps.lower_bound(it->second.first);
+ if (s == snaps.end() || *s > id) {
+ dout(10) << " purging old_inode [" << it->second.first << "," << id << "]" << dendl;
+ it = old_inodes.erase(it);
+ } else {
+ ++it;
+ }
}
}
/*
* pick/create an old_inode
*/
-old_inode_t * CInode::pick_old_inode(snapid_t snap)
+CInode::mempool_old_inode * CInode::pick_old_inode(snapid_t snap)
{
- compact_map<snapid_t, old_inode_t>::iterator p = old_inodes.lower_bound(snap); // p is first key >= to snap
- if (p != old_inodes.end() && p->second.first <= snap) {
- dout(10) << __func__ << " snap " << snap << " -> [" << p->second.first << "," << p->first << "]" << dendl;
- return &p->second;
+ auto it = old_inodes.lower_bound(snap); // p is first key >= to snap
+ if (it != old_inodes.end() && it->second.first <= snap) {
+ dout(10) << __func__ << " snap " << snap << " -> [" << it->second.first << "," << it->first << "]" << dendl;
+ return &it->second;
}
dout(10) << __func__ << " snap " << snap << " -> nothing" << dendl;
return NULL;
@@ -3042,7 +3011,7 @@ int CInode::get_xlocker_mask(client_t client) const
(linklock.gcaps_xlocker_mask(client) << linklock.get_cap_shift());
}
-int CInode::get_caps_allowed_for_client(Session *session, inode_t *file_i) const
+int CInode::get_caps_allowed_for_client(Session *session, mempool_inode *file_i) const
{
client_t client = session->get_client();
int allowed;
@@ -3120,11 +3089,9 @@ int CInode::get_caps_wanted(int *ploner, int *pother, int shift, int mask) const
//cout << " get_caps_wanted client " << it->first << " " << cap_string(it->second.wanted()) << endl;
}
if (is_auth())
- for (compact_map<int,int>::const_iterator it = mds_caps_wanted.begin();
- it != mds_caps_wanted.end();
- ++it) {
- w |= it->second;
- other |= it->second;
+ for (const auto &p : mds_caps_wanted) {
+ w |= p.second;
+ other |= p.second;
//cout << " get_caps_wanted mds " << it->first << " " << cap_string(it->second) << endl;
}
if (ploner) *ploner = (loner >> shift) & mask;
@@ -3178,10 +3145,10 @@ int CInode::encode_inodestat(bufferlist& bl, Session *session,
bool valid = true;
// pick a version!
- inode_t *oi = &inode;
- inode_t *pi = get_projected_inode();
+ mempool_inode *oi = &inode;
+ mempool_inode *pi = get_projected_inode();
- map<string, bufferptr> *pxattrs = 0;
+ CInode::mempool_xattr_map *pxattrs = nullptr;
if (snapid != CEPH_NOSNAP) {
@@ -3190,19 +3157,20 @@ int CInode::encode_inodestat(bufferlist& bl, Session *session,
valid = false;
if (is_multiversion()) {
- compact_map<snapid_t,old_inode_t>::iterator p = old_inodes.lower_bound(snapid);
- if (p != old_inodes.end()) {
- if (p->second.first > snapid) {
- if (p != old_inodes.begin())
- --p;
+ auto it = old_inodes.lower_bound(snapid);
+ if (it != old_inodes.end()) {
+ if (it->second.first > snapid) {
+ if (it != old_inodes.begin())
+ --it;
}
- if (p->second.first <= snapid && snapid <= p->first) {
+ if (it->second.first <= snapid && snapid <= it->first) {
dout(15) << __func__ << " snapid " << snapid
- << " to old_inode [" << p->second.first << "," << p->first << "]"
- << " " << p->second.inode.rstat
+ << " to old_inode [" << it->second.first << "," << it->first << "]"
+ << " " << it->second.inode.rstat
<< dendl;
- pi = oi = &p->second.inode;
- pxattrs = &p->second.xattrs;
+ auto &p = it->second;
+ pi = oi = &p.inode;
+ pxattrs = &p.xattrs;
} else {
// snapshoted remote dentry can result this
dout(0) << __func__ << " old_inode for snapid " << snapid
@@ -3246,7 +3214,7 @@ int CInode::encode_inodestat(bufferlist& bl, Session *session,
bool plocal = versionlock.get_last_wrlock_client() == client;
bool ppolicy = policylock.is_xlocked_by_client(client) || get_loner()==client;
- inode_t *any_i = (pfile|pauth|plink|pxattr|plocal) ? pi : oi;
+ mempool_inode *any_i = (pfile|pauth|plink|pxattr|plocal) ? pi : oi;
dout(20) << " pfile " << pfile << " pauth " << pauth
<< " plink " << plink << " pxattr " << pxattr
@@ -3255,7 +3223,7 @@ int CInode::encode_inodestat(bufferlist& bl, Session *session,
<< " valid=" << valid << dendl;
// file
- inode_t *file_i = pfile ? pi:oi;
+ mempool_inode *file_i = pfile ? pi:oi;
file_layout_t layout;
if (is_dir()) {
layout = (ppolicy ? pi : oi)->layout;
@@ -3290,13 +3258,13 @@ int CInode::encode_inodestat(bufferlist& bl, Session *session,
}
// auth
- inode_t *auth_i = pauth ? pi:oi;
+ mempool_inode *auth_i = pauth ? pi:oi;
// link
- inode_t *link_i = plink ? pi:oi;
+ mempool_inode *link_i = plink ? pi:oi;
// xattr
- inode_t *xattr_i = pxattr ? pi:oi;
+ mempool_inode *xattr_i = pxattr ? pi:oi;
using ceph::encode;
// xattr
@@ -3479,7 +3447,7 @@ int CInode::encode_inodestat(bufferlist& bl, Session *session,
encode(inline_data, bl);
}
if (session->connection->has_feature(CEPH_FEATURE_MDS_QUOTA)) {
- inode_t *policy_i = ppolicy ? pi : oi;
+ mempool_inode *policy_i = ppolicy ? pi : oi;
encode(policy_i->quota, bl);
}
if (session->connection->has_feature(CEPH_FEATURE_FS_FILE_LAYOUT_V2)) {
@@ -3504,9 +3472,9 @@ void CInode::encode_cap_message(MClientCaps *m, Capability *cap)
bool plink = linklock.is_xlocked_by_client(client);
bool pxattr = xattrlock.is_xlocked_by_client(client);
- inode_t *oi = &inode;
- inode_t *pi = get_projected_inode();
- inode_t *i = (pfile|pauth|plink|pxattr) ? pi : oi;
+ mempool_inode *oi = &inode;
+ mempool_inode *pi = get_projected_inode();
+ mempool_inode *i = (pfile|pauth|plink|pxattr) ? pi : oi;
dout(20) << __func__ << " pfile " << pfile
<< " pauth " << pauth << " plink " << plink << " pxattr " << pxattr
@@ -3546,7 +3514,7 @@ void CInode::encode_cap_message(MClientCaps *m, Capability *cap)
using ceph::encode;
i = pxattr ? pi:oi;
- map<string,bufferptr> *ix = pxattr ? get_projected_xattrs() : &xattrs;
+ auto ix = pxattr ? get_projected_xattrs() : &xattrs;
if ((cap->pending() & CEPH_CAP_XATTR_SHARED) &&
i->xattr_version > cap->client_xattr_version) {
dout(10) << " including xattrs v " << i->xattr_version << dendl;
@@ -3575,7 +3543,11 @@ void CInode::_decode_base(bufferlist::iterator& p)
using ceph::decode;
decode(first, p);
decode(inode, p);
- decode(symlink, p);
+ {
+ std::string tmp;
+ decode(tmp, p);
+ symlink = std::string_view(tmp);
+ }
decode(dirfragtree, p);
decode(xattrs, p);
decode(old_inodes, p);
@@ -3711,12 +3683,10 @@ void CInode::encode_export(bufferlist& bl)
// include scatterlock info for any bounding CDirs
bufferlist bounding;
if (inode.is_dir())
- for (compact_map<frag_t,CDir*>::iterator p = dirfrags.begin();
- p != dirfrags.end();
- ++p) {
- CDir *dir = p->second;
+ for (const auto &p : dirfrags) {
+ CDir *dir = p.second;
if (dir->state_test(CDir::STATE_EXPORTBOUND)) {
- encode(p->first, bounding);
+ encode(p.first, bounding);
encode(dir->fnode.fragstat, bounding);
encode(dir->fnode.accounted_fragstat, bounding);
encode(dir->fnode.rstat, bounding);
@@ -3833,14 +3803,11 @@ void InodeStoreBase::dump(Formatter *f) const
inode.dump(f);
f->dump_string("symlink", symlink);
f->open_array_section("old_inodes");
- for (compact_map<snapid_t, old_inode_t>::const_iterator i = old_inodes.begin();
- i != old_inodes.end(); ++i) {
+ for (const auto &p : old_inodes) {
f->open_object_section("old_inode");
- {
- // The key is the last snapid, the first is in the old_inode_t
- f->dump_int("last", i->first);
- i->second.dump(f);
- }
+ // The key is the last snapid, the first is in the mempool_old_inode
+ f->dump_int("last", p.first);
+ p.second.dump(f);
f->close_section(); // old_inode
}
f->close_section(); // old_inodes
@@ -3935,7 +3902,7 @@ void CInode::validate_disk_state(CInode::validated_data *results,
bool _start(int rval) {
if (in->is_dirty()) {
MDCache *mdcache = in->mdcache;
- inode_t& inode = in->inode;
+ mempool_inode& inode = in->inode;
dout(20) << "validating a dirty CInode; results will be inconclusive"
<< dendl;
}
@@ -3978,7 +3945,7 @@ void CInode::validate_disk_state(CInode::validated_data *results,
int memory_newer;
MDCache *mdcache = in->mdcache; // For the benefit of dout
- const inode_t& inode = in->inode; // For the benefit of dout
+ const mempool_inode& inode = in->inode; // For the benefit of dout
// Ignore rval because it's the result of a FAILOK operation
// from fetch_backtrace_and_tag: the real result is in
@@ -4094,8 +4061,8 @@ next:
results->inode.ondisk_value = shadow_in->inode;
results->inode.memory_value = in->inode;
- inode_t& si = shadow_in->inode;
- inode_t& i = in->inode;
+ mempool_inode& si = shadow_in->inode;
+ mempool_inode& i = in->inode;
if (si.version > i.version) {
// uh, what?
results->inode.error_str << "On-disk inode is newer than in-memory one!";
@@ -4158,10 +4125,8 @@ next:
}
// check each dirfrag...
- for (compact_map<frag_t,CDir*>::iterator p = in->dirfrags.begin();
- p != in->dirfrags.end();
- ++p) {
- CDir *dir = p->second;
+ for (const auto &p : in->dirfrags) {
+ CDir *dir = p.second;
assert(dir->get_version() > 0);
nest_info.add(dir->fnode.accounted_rstat);
dir_info.add(dir->fnode.accounted_fragstat);
@@ -4171,10 +4136,10 @@ next:
if (dir->scrub_infop->header->get_repair()) {
results->raw_stats.repaired = true;
results->raw_stats.error_str
- << "dirfrag(" << p->first << ") has bad stats (will be fixed); ";
+ << "dirfrag(" << p.first << ") has bad stats (will be fixed); ";
} else {
results->raw_stats.error_str
- << "dirfrag(" << p->first << ") has bad stats; ";
+ << "dirfrag(" << p.first << ") has bad stats; ";
}
frags_errors++;
}
@@ -4371,14 +4336,15 @@ void CInode::dump(Formatter *f, int flags) const
if (flags & DUMP_CAPS) {
f->open_array_section("client_caps");
- for (map<client_t,Capability*>::const_iterator it = client_caps.begin();
- it != client_caps.end(); ++it) {
+ for (const auto &p : client_caps) {
+ auto &client = p.first;
+ auto &cap = p.second;
f->open_object_section("client_cap");
- f->dump_int("client_id", it->first.v);
- f->dump_string("pending", ccap_string(it->second->pending()));
- f->dump_string("issued", ccap_string(it->second->issued()));
- f->dump_string("wanted", ccap_string(it->second->wanted()));
- f->dump_int("last_sent", it->second->get_last_sent());
+ f->dump_int("client_id", client.v);
+ f->dump_string("pending", ccap_string(cap->pending()));
+ f->dump_string("issued", ccap_string(cap->issued()));
+ f->dump_string("wanted", ccap_string(cap->wanted()));
+ f->dump_int("last_sent", cap->get_last_sent());
f->close_section();
}
f->close_section();
@@ -4387,11 +4353,10 @@ void CInode::dump(Formatter *f, int flags) const
f->dump_int("want_loner", want_loner_cap.v);
f->open_array_section("mds_caps_wanted");
- for (compact_map<int,int>::const_iterator p = mds_caps_wanted.begin();
- p != mds_caps_wanted.end(); ++p) {
+ for (const auto &p : mds_caps_wanted) {
f->open_object_section("mds_cap_wanted");
- f->dump_int("rank", p->first);
- f->dump_string("cap", ccap_string(p->second));
+ f->dump_int("rank", p.first);
+ f->dump_string("cap", ccap_string(p.second));
f->close_section();
}
f->close_section();
@@ -4406,7 +4371,7 @@ void CInode::scrub_info_create() const
// break out of const-land to set up implicit initial state
CInode *me = const_cast<CInode*>(this);
- inode_t *in = me->get_projected_inode();
+ mempool_inode *in = me->get_projected_inode();
scrub_info_t *si = new scrub_info_t();
si->scrub_start_stamp = si->last_scrub_stamp = in->last_scrub_stamp;
@@ -4640,7 +4605,7 @@ mds_rank_t CInode::get_export_pin(bool inherit) const
const CDentry *pdn = in->get_projected_parent_dn();
if (!pdn)
break;
- const inode_t *pi = in->get_projected_inode();
+ const mempool_inode *pi = in->get_projected_inode();
// ignore export pin for unlinked directory
if (pi->nlink == 0)
break;
diff --git a/src/mds/CInode.h b/src/mds/CInode.h
index 41f066240b1..537d98760e7 100644
--- a/src/mds/CInode.h
+++ b/src/mds/CInode.h
@@ -75,15 +75,20 @@ extern int num_cinode_locks;
*/
class InodeStoreBase {
public:
- inode_t inode; // the inode itself
- std::string symlink; // symlink dest, if symlink
- std::map<std::string, bufferptr> xattrs;
+ typedef inode_t<mempool::mds_co::pool_allocator> mempool_inode;
+ typedef old_inode_t<mempool::mds_co::pool_allocator> mempool_old_inode;
+ typedef mempool::mds_co::compact_map<snapid_t, mempool_old_inode> mempool_old_inode_map;
+ typedef xattr_map<mempool::mds_co::pool_allocator> mempool_xattr_map; // FIXME bufferptr not in mempool
+
+ mempool_inode inode; // the inode itself
+ mempool::mds_co::string symlink; // symlink dest, if symlink
+ mempool_xattr_map xattrs;
fragtree_t dirfragtree; // dir frag tree, if any. always consistent with our dirfrag map.
- compact_map<snapid_t, old_inode_t> old_inodes; // key = last, value.first = first
- snapid_t oldest_snap;
- damage_flags_t damage_flags;
+ mempool_old_inode_map old_inodes; // key = last, value.first = first
+ snapid_t oldest_snap = CEPH_NOSNAP;
+ damage_flags_t damage_flags = 0;
- InodeStoreBase() : oldest_snap(CEPH_NOSNAP), damage_flags(0) { }
+ InodeStoreBase() {}
/* Helpers */
bool is_file() const { return inode.is_file(); }
@@ -109,6 +114,7 @@ public:
class InodeStore : public InodeStoreBase {
public:
+ // FIXME bufferlist not part of mempool
bufferlist snap_blob; // Encoded copy of SnapRealm, because we can't
// rehydrate it without full MDCache
void encode(bufferlist &bl, uint64_t features) const {
@@ -250,22 +256,22 @@ class CInode : public MDSCacheObject, public InodeStoreBase, public Counter<CIno
public:
MDCache *mdcache;
- SnapRealm *snaprealm;
- SnapRealm *containing_realm;
+ SnapRealm *snaprealm = nullptr;
+ SnapRealm *containing_realm = nullptr;
snapid_t first, last;
- compact_set<snapid_t> dirty_old_rstats;
+ mempool::mds_co::compact_set<snapid_t> dirty_old_rstats;
class scrub_stamp_info_t {
public:
/// version we started our latest scrub (whether in-progress or finished)
- version_t scrub_start_version;
+ version_t scrub_start_version = 0;
/// time we started our latest scrub (whether in-progress or finished)
utime_t scrub_start_stamp;
/// version we started our most recent finished scrub
- version_t last_scrub_version;
+ version_t last_scrub_version = 0;
/// time we started our most recent finished scrub
utime_t last_scrub_stamp;
- scrub_stamp_info_t() : scrub_start_version(0), last_scrub_version(0) {}
+ scrub_stamp_info_t() {}
void reset() {
scrub_start_version = last_scrub_version = 0;
scrub_start_stamp = last_scrub_stamp = utime_t();
@@ -274,22 +280,19 @@ class CInode : public MDSCacheObject, public InodeStoreBase, public Counter<CIno
class scrub_info_t : public scrub_stamp_info_t {
public:
- CDentry *scrub_parent;
- MDSInternalContextBase *on_finish;
+ CDentry *scrub_parent = nullptr;
+ MDSInternalContextBase *on_finish = nullptr;
- bool last_scrub_dirty; /// are our stamps dirty with respect to disk state?
- bool scrub_in_progress; /// are we currently scrubbing?
- bool children_scrubbed;
+ bool last_scrub_dirty = false; /// are our stamps dirty with respect to disk state?
+ bool scrub_in_progress = false; /// are we currently scrubbing?
+ bool children_scrubbed = false;
/// my own (temporary) stamps and versions for each dirfrag we have
- std::map<frag_t, scrub_stamp_info_t> dirfrag_stamps;
+ std::map<frag_t, scrub_stamp_info_t> dirfrag_stamps; // XXX not part of mempool
ScrubHeaderRef header;
- scrub_info_t() : scrub_stamp_info_t(),
- scrub_parent(NULL), on_finish(NULL),
- last_scrub_dirty(false), scrub_in_progress(false),
- children_scrubbed(false) {}
+ scrub_info_t() {}
};
const scrub_info_t *scrub_info() const{
@@ -381,7 +384,7 @@ public:
}
snapid_t get_oldest_snap();
- uint64_t last_journaled; // log offset for the last time i was journaled
+ uint64_t last_journaled = 0; // log offset for the last time i was journaled
//loff_t last_open_journaled; // log offset for the last journaled EOpen
utime_t last_dirstat_prop;
@@ -395,108 +398,104 @@ public:
void mark_dirty_rstat();
void clear_dirty_rstat();
- //bool hack_accessed;
+ //bool hack_accessed = false;
//utime_t hack_load_stamp;
/**
* Projection methods, used to store inode changes until they have been journaled,
* at which point they are popped.
* Usage:
- * project_inode as needed. If you're also projecting xattrs, pass
- * in an xattr map (by pointer), then edit the map.
- * If you're also projecting the snaprealm, call project_snaprealm after
- * calling project_inode, and modify the snaprealm as necessary.
+ * project_inode as needed. If you're changing xattrs or sr_t, then pass true
+ * as needed then change the xattrs/snapnode member as needed. (Dirty
+ * exception: project_past_snaprealm_parent allows you to project the
+ * snapnode after doing project_inode (i.e. you don't need to pass
+ * snap=true).
*
* Then, journal. Once journaling is done, pop_and_dirty_projected_inode.
* This function will take care of the inode itself, the xattrs, and the snaprealm.
*/
- struct projected_inode_t {
- inode_t *inode;
- std::map<std::string,bufferptr> *xattrs;
- sr_t *snapnode;
-
- projected_inode_t()
- : inode(NULL), xattrs(NULL), snapnode(NULL) {}
- projected_inode_t(inode_t *in, sr_t *sn)
- : inode(in), xattrs(NULL), snapnode(sn) {}
- projected_inode_t(inode_t *in, std::map<std::string, bufferptr> *xp = NULL, sr_t *sn = NULL)
- : inode(in), xattrs(xp), snapnode(sn) {}
+ class projected_inode {
+ public:
+ mempool_inode inode;
+ std::unique_ptr<mempool_xattr_map> xattrs;
+ std::unique_ptr<sr_t> snapnode;
+
+ projected_inode() = delete;
+ projected_inode(const mempool_inode &in) : inode(in) {}
};
- std::list<projected_inode_t*> projected_nodes; // projected values (only defined while dirty)
- int num_projected_xattrs;
- int num_projected_srnodes;
-
- inode_t *project_inode(std::map<std::string,bufferptr> *px=0);
+
+private:
+ mempool::mds_co::list<projected_inode> projected_nodes; // projected values (only defined while dirty)
+ size_t num_projected_xattrs = 0;
+ size_t num_projected_srnodes = 0;
+
+ sr_t &project_snaprealm(projected_inode &pi);
+public:
+ CInode::projected_inode &project_inode(bool xattr = false, bool snap = false);
void pop_and_dirty_projected_inode(LogSegment *ls);
- projected_inode_t *get_projected_node() {
+ projected_inode *get_projected_node() {
if (projected_nodes.empty())
return NULL;
else
- return projected_nodes.back();
+ return &projected_nodes.back();
}
version_t get_projected_version() const {
if (projected_nodes.empty())
return inode.version;
else
- return projected_nodes.back()->inode->version;
+ return projected_nodes.back().inode.version;
}
bool is_projected() const {
return !projected_nodes.empty();
}
- const inode_t *get_projected_inode() const {
+ const mempool_inode *get_projected_inode() const {
if (projected_nodes.empty())
return &inode;
else
- return projected_nodes.back()->inode;
+ return &projected_nodes.back().inode;
}
- inode_t *get_projected_inode() {
+ mempool_inode *get_projected_inode() {
if (projected_nodes.empty())
return &inode;
else
- return projected_nodes.back()->inode;
+ return &projected_nodes.back().inode;
}
- inode_t *get_previous_projected_inode() {
+ mempool_inode *get_previous_projected_inode() {
assert(!projected_nodes.empty());
- std::list<projected_inode_t*>::reverse_iterator p = projected_nodes.rbegin();
- ++p;
- if (p != projected_nodes.rend())
- return (*p)->inode;
+ auto it = projected_nodes.rbegin();
+ ++it;
+ if (it != projected_nodes.rend())
+ return &it->inode;
else
return &inode;
}
- std::map<std::string,bufferptr> *get_projected_xattrs() {
+ mempool_xattr_map *get_projected_xattrs() {
if (num_projected_xattrs > 0) {
- for (std::list<projected_inode_t*>::reverse_iterator p = projected_nodes.rbegin();
- p != projected_nodes.rend();
- ++p)
- if ((*p)->xattrs)
- return (*p)->xattrs;
+ for (auto it = projected_nodes.rbegin(); it != projected_nodes.rend(); ++it)
+ if (it->xattrs)
+ return it->xattrs.get();
}
return &xattrs;
}
- std::map<std::string,bufferptr> *get_previous_projected_xattrs() {
- std::list<projected_inode_t*>::reverse_iterator p = projected_nodes.rbegin();
- for (++p; // skip the most recent projected value
- p != projected_nodes.rend();
- ++p)
- if ((*p)->xattrs)
- return (*p)->xattrs;
+ mempool_xattr_map *get_previous_projected_xattrs() {
+ if (num_projected_xattrs > 0) {
+ for (auto it = ++projected_nodes.rbegin(); it != projected_nodes.rend(); ++it)
+ if (it->xattrs)
+ return it->xattrs.get();
+ }
return &xattrs;
}
- sr_t *project_snaprealm(snapid_t snapid=0);
const sr_t *get_projected_srnode() const {
if (num_projected_srnodes > 0) {
- for (std::list<projected_inode_t*>::const_reverse_iterator p = projected_nodes.rbegin();
- p != projected_nodes.rend();
- ++p)
- if ((*p)->snapnode)
- return (*p)->snapnode;
+ for (auto it = projected_nodes.rbegin(); it != projected_nodes.rend(); ++it)
+ if (it->snapnode)
+ return it->snapnode.get();
}
if (snaprealm)
return &snaprealm->srnode;
@@ -509,18 +508,18 @@ private:
void pop_projected_snaprealm(sr_t *next_snaprealm);
public:
- old_inode_t& cow_old_inode(snapid_t follows, bool cow_head);
+ mempool_old_inode& cow_old_inode(snapid_t follows, bool cow_head);
void split_old_inode(snapid_t snap);
- old_inode_t *pick_old_inode(snapid_t last);
+ mempool_old_inode *pick_old_inode(snapid_t last);
void pre_cow_old_inode();
bool has_snap_data(snapid_t s);
void purge_stale_snap_data(const std::set<snapid_t>& snaps);
// -- cache infrastructure --
private:
- compact_map<frag_t,CDir*> dirfrags; // cached dir fragments under this Inode
- int stickydir_ref;
- scrub_info_t *scrub_infop;
+ mempool::mds_co::compact_map<frag_t,CDir*> dirfrags; // cached dir fragments under this Inode
+ int stickydir_ref = 0;
+ scrub_info_t *scrub_infop = nullptr;
public:
bool has_dirfrags() { return !dirfrags.empty(); }
@@ -552,23 +551,24 @@ public:
protected:
// parent dentries in cache
- CDentry *parent; // primary link
- compact_set<CDentry*> remote_parents; // if hard linked
+ CDentry *parent = nullptr; // primary link
+ mempool::mds_co::compact_set<CDentry*> remote_parents; // if hard linked
- std::list<CDentry*> projected_parent; // for in-progress rename, (un)link, etc.
+ mempool::mds_co::list<CDentry*> projected_parent; // for in-progress rename, (un)link, etc.
- mds_authority_t inode_auth;
+ mds_authority_t inode_auth = CDIR_AUTH_DEFAULT;
// -- distributed state --
protected:
- // file capabilities
- std::map<client_t, Capability*> client_caps; // client -> caps
- compact_map<int32_t, int32_t> mds_caps_wanted; // [auth] mds -> caps wanted
- int replica_caps_wanted; // [replica] what i've requested from auth
+ // file capabilities FIXME Capability not part of mempool
+ using cap_map = mempool::mds_co::map<client_t, Capability*>;
+ cap_map client_caps; // client -> caps
+ mempool::mds_co::compact_map<int32_t, int32_t> mds_caps_wanted; // [auth] mds -> caps wanted
+ int replica_caps_wanted = 0; // [replica] what i've requested from auth
public:
- compact_map<int, std::set<client_t> > client_snap_caps; // [auth] [snap] dirty metadata we still need from the head
- compact_map<snapid_t, std::set<client_t> > client_need_snapflush;
+ mempool::mds_co::compact_map<int, mempool::mds_co::set<client_t> > client_snap_caps; // [auth] [snap] dirty metadata we still need from the head
+ mempool::mds_co::compact_map<snapid_t, mempool::mds_co::set<client_t> > client_need_snapflush;
void add_need_snapflush(CInode *snapin, snapid_t snapid, client_t client);
void remove_need_snapflush(CInode *snapin, snapid_t snapid, client_t client);
@@ -576,8 +576,8 @@ public:
protected:
- ceph_lock_state_t *fcntl_locks;
- ceph_lock_state_t *flock_locks;
+ ceph_lock_state_t *fcntl_locks = nullptr;
+ ceph_lock_state_t *flock_locks = nullptr;
ceph_lock_state_t *get_fcntl_lock_state() {
if (!fcntl_locks)
@@ -644,7 +644,7 @@ public:
elist<CInode*>::item& item_recover_queue_front = item_dirty_dirfrag_nest;
public:
- int auth_pin_freeze_allowance;
+ int auth_pin_freeze_allowance = 0;
inode_load_vec_t pop;
@@ -658,25 +658,17 @@ public:
friend class CInodeExport;
// ---------------------------
+ CInode() = delete;
CInode(MDCache *c, bool auth=true, snapid_t f=2, snapid_t l=CEPH_NOSNAP) :
mdcache(c),
- snaprealm(0), containing_realm(0),
first(f), last(l),
- last_journaled(0), //last_open_journaled(0),
- //hack_accessed(true),
- num_projected_xattrs(0),
- num_projected_srnodes(0),
- stickydir_ref(0),
- scrub_infop(NULL),
- parent(0),
- inode_auth(CDIR_AUTH_DEFAULT),
- replica_caps_wanted(0),
- fcntl_locks(0), flock_locks(0),
- item_dirty(this), item_caps(this), item_open_file(this), item_dirty_parent(this),
+ item_dirty(this),
+ item_caps(this),
+ item_open_file(this),
+ item_dirty_parent(this),
item_dirty_dirfrag_dir(this),
item_dirty_dirfrag_nest(this),
item_dirty_dirfrag_dirfragtree(this),
- auth_pin_freeze_allowance(0),
pop(ceph_clock_now()),
versionlock(this, &versionlock_type),
authlock(this, &authlock_type),
@@ -687,10 +679,8 @@ public:
snaplock(this, &snaplock_type),
nestlock(this, &nestlock_type),
flocklock(this, &flocklock_type),
- policylock(this, &policylock_type),
- loner_cap(-1), want_loner_cap(-1)
+ policylock(this, &policylock_type)
{
- state = 0;
if (auth) state_set(STATE_AUTH);
}
~CInode() override {
@@ -730,7 +720,7 @@ public:
vinodeno_t vino() const { return vinodeno_t(inode.ino, last); }
int d_type() const { return IFTODT(inode.mode); }
- inode_t& get_inode() { return inode; }
+ mempool_inode& get_inode() { return inode; }
CDentry* get_parent_dn() { return parent; }
const CDentry* get_parent_dn() const { return parent; }
CDentry* get_projected_parent_dn() { return !projected_parent.empty() ? projected_parent.back() : parent; }
@@ -831,7 +821,7 @@ public:
// -- waiting --
protected:
- compact_map<frag_t, std::list<MDSInternalContextBase*> > waiting_on_dir;
+ mempool::mds_co::compact_map<frag_t, std::list<MDSInternalContextBase*> > waiting_on_dir;
public:
void add_dir_waiter(frag_t fg, MDSInternalContextBase *c);
void take_dir_waiting(frag_t fg, std::list<MDSInternalContextBase*>& ls);
@@ -884,6 +874,7 @@ public:
static LockType flocklock_type;
static LockType policylock_type;
+ // FIXME not part of mempool
LocalLock versionlock;
SimpleLock authlock;
SimpleLock linklock;
@@ -935,7 +926,7 @@ public:
// -- caps -- (new)
// client caps
- client_t loner_cap, want_loner_cap;
+ client_t loner_cap = -1, want_loner_cap = -1;
client_t get_loner() const { return loner_cap; }
client_t get_wanted_loner() const { return want_loner_cap; }
@@ -960,33 +951,31 @@ public:
int count_nonstale_caps() {
int n = 0;
- for (std::map<client_t,Capability*>::iterator it = client_caps.begin();
- it != client_caps.end();
- ++it)
- if (!it->second->is_stale())
+ for (const auto &p : client_caps) {
+ if (!p.second->is_stale())
n++;
+ }
return n;
}
bool multiple_nonstale_caps() {
int n = 0;
- for (std::map<client_t,Capability*>::iterator it = client_caps.begin();
- it != client_caps.end();
- ++it)
- if (!it->second->is_stale()) {
+ for (const auto &p : client_caps) {
+ if (!p.second->is_stale()) {
if (n)
return true;
n++;
}
+ }
return false;
}
bool is_any_caps() { return !client_caps.empty(); }
bool is_any_nonstale_caps() { return count_nonstale_caps(); }
- const compact_map<int32_t,int32_t>& get_mds_caps_wanted() const { return mds_caps_wanted; }
- compact_map<int32_t,int32_t>& get_mds_caps_wanted() { return mds_caps_wanted; }
+ const mempool::mds_co::compact_map<int32_t,int32_t>& get_mds_caps_wanted() const { return mds_caps_wanted; }
+ mempool::mds_co::compact_map<int32_t,int32_t>& get_mds_caps_wanted() { return mds_caps_wanted; }
- const std::map<client_t,Capability*>& get_client_caps() const { return client_caps; }
+ const cap_map& get_client_caps() const { return client_caps; }
Capability *get_client_cap(client_t client) {
auto client_caps_entry = client_caps.find(client);
if (client_caps_entry != client_caps.end())
@@ -1016,7 +1005,7 @@ public:
int get_caps_allowed_by_type(int type) const;
int get_caps_careful() const;
int get_xlocker_mask(client_t client) const;
- int get_caps_allowed_for_client(Session *s, inode_t *file_i) const;
+ int get_caps_allowed_for_client(Session *s, mempool_inode *file_i) const;
// caps issued, wanted
int get_caps_issued(int *ploner = 0, int *pother = 0, int *pxlocker = 0,
@@ -1139,8 +1128,8 @@ public:
std::stringstream error_str;
};
- bool performed_validation;
- bool passed_validation;
+ bool performed_validation = false;
+ bool passed_validation = false;
struct raw_stats_t {
frag_info_t dirstat;
@@ -1148,11 +1137,10 @@ public:
};
member_status<inode_backtrace_t> backtrace;
- member_status<inode_t> inode;
+ member_status<mempool_inode> inode; // XXX should not be in mempool; wait for pmr
member_status<raw_stats_t> raw_stats;
- validated_data() : performed_validation(false),
- passed_validation(false) {}
+ validated_data() {}
void dump(Formatter *f) const;
diff --git a/src/mds/Locker.cc b/src/mds/Locker.cc
index 3d9fa6c3a56..7daa84bb199 100644
--- a/src/mds/Locker.cc
+++ b/src/mds/Locker.cc
@@ -1857,7 +1857,7 @@ void Locker::file_update_finish(CInode *in, MutationRef& mut, bool share_max, bo
dout(10) << " client_snap_caps " << in->client_snap_caps << dendl;
// check for snap writeback completion
bool gather = false;
- compact_map<int,set<client_t> >::iterator p = in->client_snap_caps.begin();
+ auto p = in->client_snap_caps.begin();
while (p != in->client_snap_caps.end()) {
auto q = p->second.find(client);
if (q != p->second.end()) {
@@ -2294,7 +2294,7 @@ public:
}
};
-uint64_t Locker::calc_new_max_size(inode_t *pi, uint64_t size)
+uint64_t Locker::calc_new_max_size(CInode::mempool_inode *pi, uint64_t size)
{
uint64_t new_max = (size + 1) << 1;
uint64_t max_inc = g_conf->mds_client_writeable_range_max_inc_objs;
@@ -2306,10 +2306,10 @@ uint64_t Locker::calc_new_max_size(inode_t *pi, uint64_t size)
}
void Locker::calc_new_client_ranges(CInode *in, uint64_t size,
- map<client_t,client_writeable_range_t> *new_ranges,
+ CInode::mempool_inode::client_range_map *new_ranges,
bool *max_increased)
{
- inode_t *latest = in->get_projected_inode();
+ auto latest = in->get_projected_inode();
uint64_t ms;
if(latest->has_layout()) {
ms = calc_new_max_size(latest, size);
@@ -2348,8 +2348,8 @@ bool Locker::check_inode_max_size(CInode *in, bool force_wrlock,
assert(in->is_auth());
assert(in->is_file());
- inode_t *latest = in->get_projected_inode();
- map<client_t, client_writeable_range_t> new_ranges;
+ CInode::mempool_inode *latest = in->get_projected_inode();
+ CInode::mempool_inode::client_range_map new_ranges;
uint64_t size = latest->size;
bool update_size = new_size > 0;
bool update_max = false;
@@ -2409,20 +2409,20 @@ bool Locker::check_inode_max_size(CInode *in, bool force_wrlock,
MutationRef mut(new MutationImpl());
mut->ls = mds->mdlog->get_current_segment();
- inode_t *pi = in->project_inode();
- pi->version = in->pre_dirty();
+ auto &pi = in->project_inode();
+ pi.inode.version = in->pre_dirty();
if (update_max) {
- dout(10) << "check_inode_max_size client_ranges " << pi->client_ranges << " -> " << new_ranges << dendl;
- pi->client_ranges = new_ranges;
+ dout(10) << "check_inode_max_size client_ranges " << pi.inode.client_ranges << " -> " << new_ranges << dendl;
+ pi.inode.client_ranges = new_ranges;
}
if (update_size) {
- dout(10) << "check_inode_max_size size " << pi->size << " -> " << new_size << dendl;
- pi->size = new_size;
- pi->rstat.rbytes = new_size;
- dout(10) << "check_inode_max_size mtime " << pi->mtime << " -> " << new_mtime << dendl;
- pi->mtime = new_mtime;
+ dout(10) << "check_inode_max_size size " << pi.inode.size << " -> " << new_size << dendl;
+ pi.inode.size = new_size;
+ pi.inode.rstat.rbytes = new_size;
+ dout(10) << "check_inode_max_size mtime " << pi.inode.mtime << " -> " << new_mtime << dendl;
+ pi.inode.mtime = new_mtime;
}
// use EOpen if the file is still open; otherwise, use EUpdate.
@@ -2615,7 +2615,7 @@ void Locker::_do_null_snapflush(CInode *head_in, client_t client, snapid_t last)
for (auto p = head_in->client_need_snapflush.begin();
p != head_in->client_need_snapflush.end() && p->first < last; ) {
snapid_t snapid = p->first;
- set<client_t>& clients = p->second;
+ auto &clients = p->second;
++p; // be careful, q loop below depends on this
if (clients.count(client)) {
@@ -3120,53 +3120,55 @@ void Locker::_do_snap_update(CInode *in, snapid_t snap, int dirty, snapid_t foll
// normal metadata updates that we can apply to the head as well.
// update xattrs?
- bool xattrs = false;
- map<string,bufferptr> *px = 0;
- if ((dirty & CEPH_CAP_XATTR_EXCL) &&
- m->xattrbl.length() &&
- m->head.xattr_version > in->get_projected_inode()->xattr_version)
- xattrs = true;
-
- old_inode_t *oi = 0;
+ CInode::mempool_xattr_map *px = nullptr;
+ bool xattrs = (dirty & CEPH_CAP_XATTR_EXCL) &&
+ m->xattrbl.length() &&
+ m->head.xattr_version > in->get_projected_inode()->xattr_version;
+
+ CInode::mempool_old_inode *oi = 0;
if (in->is_multiversion()) {
oi = in->pick_old_inode(snap);
}
- inode_t *pi;
+ CInode::mempool_inode *i;
if (oi) {
dout(10) << " writing into old inode" << dendl;
- pi = in->project_inode();
- pi->version = in->pre_dirty();
+ auto &pi = in->project_inode();
+ pi.inode.version = in->pre_dirty();
if (snap > oi->first)
in->split_old_inode(snap);
- pi = &oi->inode;
+ i = &oi->inode;
if (xattrs)
px = &oi->xattrs;
} else {
+ auto &pi = in->project_inode(xattrs);
+ pi.inode.version = in->pre_dirty();
+ i = &pi.inode;
if (xattrs)
- px = new map<string,bufferptr>;
- pi = in->project_inode(px);
- pi->version = in->pre_dirty();
+ px = pi.xattrs.get();
}
- _update_cap_fields(in, dirty, m, pi);
+ _update_cap_fields(in, dirty, m, i);
// xattr
- if (px) {
- dout(7) << " xattrs v" << pi->xattr_version << " -> " << m->head.xattr_version
+ if (xattrs) {
+ dout(7) << " xattrs v" << i->xattr_version << " -> " << m->head.xattr_version
<< " len " << m->xattrbl.length() << dendl;
- pi->xattr_version = m->head.xattr_version;
+ i->xattr_version = m->head.xattr_version;
bufferlist::iterator p = m->xattrbl.begin();
decode(*px, p);
}
- if (pi->client_ranges.count(client)) {
- if (in->last == snap) {
- dout(10) << " removing client_range entirely" << dendl;
- pi->client_ranges.erase(client);
- } else {
- dout(10) << " client_range now follows " << snap << dendl;
- pi->client_ranges[client].follows = snap;
+ {
+ auto it = i->client_ranges.find(client);
+ if (it != i->client_ranges.end()) {
+ if (in->last == snap) {
+ dout(10) << " removing client_range entirely" << dendl;
+ i->client_ranges.erase(it);
+ } else {
+ dout(10) << " client_range now follows " << snap << dendl;
+ it->second.follows = snap;
+ }
}
}
@@ -3183,7 +3185,7 @@ void Locker::_do_snap_update(CInode *in, snapid_t snap, int dirty, snapid_t foll
client, ack));
}
-void Locker::_update_cap_fields(CInode *in, int dirty, MClientCaps *m, inode_t *pi)
+void Locker::_update_cap_fields(CInode *in, int dirty, MClientCaps *m, CInode::mempool_inode *pi)
{
if (dirty == 0)
return;
@@ -3291,7 +3293,7 @@ bool Locker::_do_cap_update(CInode *in, Capability *cap,
<< " on " << *in << dendl;
assert(in->is_auth());
client_t client = m->get_source().num();
- inode_t *latest = in->get_projected_inode();
+ CInode::mempool_inode *latest = in->get_projected_inode();
// increase or zero max_size?
uint64_t size = m->get_size();
@@ -3395,30 +3397,28 @@ bool Locker::_do_cap_update(CInode *in, Capability *cap,
EUpdate *le = new EUpdate(mds->mdlog, "cap update");
mds->mdlog->start_entry(le);
- // xattrs update?
- map<string,bufferptr> *px = 0;
- if ((dirty & CEPH_CAP_XATTR_EXCL) &&
- m->xattrbl.length() &&
- m->head.xattr_version > in->get_projected_inode()->xattr_version)
- px = new map<string,bufferptr>;
+ bool xattr = (dirty & CEPH_CAP_XATTR_EXCL) &&
+ m->xattrbl.length() &&
+ m->head.xattr_version > in->get_projected_inode()->xattr_version;
- inode_t *pi = in->project_inode(px);
- pi->version = in->pre_dirty();
+ auto &pi = in->project_inode(xattr);
+ pi.inode.version = in->pre_dirty();
MutationRef mut(new MutationImpl());
mut->ls = mds->mdlog->get_current_segment();
- _update_cap_fields(in, dirty, m, pi);
+ _update_cap_fields(in, dirty, m, &pi.inode);
if (change_max) {
dout(7) << " max_size " << old_max << " -> " << new_max
<< " for " << *in << dendl;
if (new_max) {
- pi->client_ranges[client].range.first = 0;
- pi->client_ranges[client].range.last = new_max;
- pi->client_ranges[client].follows = in->first - 1;
+ auto &cr = pi.inode.client_ranges[client];
+ cr.range.first = 0;
+ cr.range.last = new_max;
+ cr.follows = in->first - 1;
} else
- pi->client_ranges.erase(client);
+ pi.inode.client_ranges.erase(client);
}
if (change_max || (dirty & (CEPH_CAP_FILE_EXCL|CEPH_CAP_FILE_WR)))
@@ -3428,13 +3428,12 @@ bool Locker::_do_cap_update(CInode *in, Capability *cap,
if (dirty & CEPH_CAP_AUTH_EXCL)
wrlock_force(&in->authlock, mut);
- // xattr
- if (px) {
- dout(7) << " xattrs v" << pi->xattr_version << " -> " << m->head.xattr_version << dendl;
- pi->xattr_version = m->head.xattr_version;
+ // xattrs update?
+ if (xattr) {
+ dout(7) << " xattrs v" << pi.inode.xattr_version << " -> " << m->head.xattr_version << dendl;
+ pi.inode.xattr_version = m->head.xattr_version;
bufferlist::iterator p = m->xattrbl.begin();
- decode(*px, p);
-
+ decode(*pi.xattrs, p);
wrlock_force(&in->xattrlock, mut);
}
@@ -4435,8 +4434,8 @@ void Locker::scatter_writebehind(ScatterLock *lock)
in->pre_cow_old_inode(); // avoid cow mayhem
- inode_t *pi = in->project_inode();
- pi->version = in->pre_dirty();
+ auto &pi = in->project_inode();
+ pi.inode.version = in->pre_dirty();
in->finish_scatter_gather_update(lock->get_type());
lock->start_flush();
diff --git a/src/mds/Locker.h b/src/mds/Locker.h
index 1c0d3888e08..315f2a207b4 100644
--- a/src/mds/Locker.h
+++ b/src/mds/Locker.h
@@ -28,7 +28,6 @@ using std::set;
class MDSRank;
class Session;
-class CInode;
class CDentry;
struct SnapRealm;
@@ -42,6 +41,7 @@ class SimpleLock;
class ScatterLock;
class LocalLock;
+#include "CInode.h"
#include "SimpleLock.h"
#include "Mutation.h"
@@ -193,7 +193,7 @@ protected:
bool _need_flush_mdlog(CInode *in, int wanted_caps);
void adjust_cap_wanted(Capability *cap, int wanted, int issue_seq);
void handle_client_caps(class MClientCaps *m);
- void _update_cap_fields(CInode *in, int dirty, MClientCaps *m, inode_t *pi);
+ void _update_cap_fields(CInode *in, int dirty, MClientCaps *m, CInode::mempool_inode *pi);
void _do_snap_update(CInode *in, snapid_t snap, int dirty, snapid_t follows, client_t client, MClientCaps *m, MClientCaps *ack);
void _do_null_snapflush(CInode *head_in, client_t client, snapid_t last=CEPH_NOSNAP);
bool _do_cap_update(CInode *in, Capability *cap, int dirty, snapid_t follows, MClientCaps *m,
@@ -259,10 +259,10 @@ protected:
void file_update_finish(CInode *in, MutationRef& mut, bool share_max, bool issue_client_cap,
client_t client, MClientCaps *ack);
private:
- uint64_t calc_new_max_size(inode_t *pi, uint64_t size);
+ uint64_t calc_new_max_size(CInode::mempool_inode *pi, uint64_t size);
public:
void calc_new_client_ranges(CInode *in, uint64_t size,
- map<client_t, client_writeable_range_t>* new_ranges,
+ CInode::mempool_inode::client_range_map* new_ranges,
bool *max_increased);
bool check_inode_max_size(CInode *in, bool force_wrlock=false,
uint64_t newmax=0, uint64_t newsize=0,
diff --git a/src/mds/MDCache.cc b/src/mds/MDCache.cc
index 038ce8ee1e8..de600c84f02 100644
--- a/src/mds/MDCache.cc
+++ b/src/mds/MDCache.cc
@@ -1513,15 +1513,14 @@ CInode *MDCache::cow_inode(CInode *in, snapid_t last)
assert(head_in);
if (head_in->split_need_snapflush(oldin, in)) {
oldin->client_snap_caps = in->client_snap_caps;
- for (compact_map<int,set<client_t> >::iterator p = in->client_snap_caps.begin();
- p != in->client_snap_caps.end();
- ++p) {
- SimpleLock *lock = oldin->get_lock(p->first);
+ for (const auto &p : in->client_snap_caps) {
+ SimpleLock *lock = oldin->get_lock(p.first);
assert(lock);
- for (auto q = p->second.begin(); q != p->second.end(); ++q) {
+ for (const auto &q : p.second) {
oldin->auth_pin(lock);
lock->set_state(LOCK_SNAP_SYNC); // gathering
lock->get_wrlock(true);
+ (void)q; /* unused */
}
}
}
@@ -1729,7 +1728,7 @@ void MDCache::project_rstat_inode_to_frag(CInode *cur, CDir *parent, snapid_t fi
int linkunlink, SnapRealm *prealm)
{
CDentry *parentdn = cur->get_projected_parent_dn();
- inode_t *curi = cur->get_projected_inode();
+ CInode::mempool_inode *curi = cur->get_projected_inode();
if (cur->first > first)
first = cur->first;
@@ -1772,23 +1771,21 @@ void MDCache::project_rstat_inode_to_frag(CInode *cur, CDir *parent, snapid_t fi
}
if (g_conf->mds_snap_rstat) {
- for (compact_set<snapid_t>::iterator p = cur->dirty_old_rstats.begin();
- p != cur->dirty_old_rstats.end();
- ++p) {
- old_inode_t& old = cur->old_inodes[*p];
+ for (const auto &p : cur->dirty_old_rstats) {
+ auto &old = cur->old_inodes[p];
snapid_t ofirst = std::max(old.first, floor);
- set<snapid_t>::const_iterator q = snaps.lower_bound(ofirst);
- if (q == snaps.end() || *q > *p)
+ auto it = snaps.lower_bound(ofirst);
+ if (it == snaps.end() || *it > p)
continue;
- if (*p >= floor)
- _project_rstat_inode_to_frag(old.inode, ofirst, *p, parent, 0, false);
+ if (p >= floor)
+ _project_rstat_inode_to_frag(old.inode, ofirst, p, parent, 0, false);
}
}
cur->dirty_old_rstats.clear();
}
-void MDCache::_project_rstat_inode_to_frag(inode_t& inode, snapid_t ofirst, snapid_t last,
+void MDCache::_project_rstat_inode_to_frag(CInode::mempool_inode& inode, snapid_t ofirst, snapid_t last,
CDir *parent, int linkunlink, bool update_inode)
{
dout(10) << "_project_rstat_inode_to_frag [" << ofirst << "," << last << "]" << dendl;
@@ -1919,13 +1916,13 @@ void MDCache::project_rstat_frag_to_inode(nest_info_t& rstat, nest_info_t& accou
dout(20) << " delta " << delta << dendl;
while (last >= ofirst) {
- inode_t *pi;
+ CInode::mempool_inode *pi;
snapid_t first;
if (last == pin->last) {
pi = pin->get_projected_inode();
first = std::max(ofirst, pin->first);
if (first > pin->first) {
- old_inode_t& old = pin->cow_old_inode(first-1, cow_head);
+ auto &old = pin->cow_old_inode(first-1, cow_head);
dout(20) << " cloned old_inode rstat is " << old.inode.rstat << dendl;
}
} else {
@@ -1935,23 +1932,23 @@ void MDCache::project_rstat_frag_to_inode(nest_info_t& rstat, nest_info_t& accou
} else {
// our life is easier here because old_inodes is not sparse
// (although it may not begin at snapid 1)
- compact_map<snapid_t,old_inode_t>::iterator p = pin->old_inodes.lower_bound(last);
- if (p == pin->old_inodes.end()) {
+ auto it = pin->old_inodes.lower_bound(last);
+ if (it == pin->old_inodes.end()) {
dout(10) << " no old_inode <= " << last << ", done." << dendl;
break;
}
- first = p->second.first;
+ first = it->second.first;
if (first > last) {
- dout(10) << " oldest old_inode is [" << first << "," << p->first << "], done." << dendl;
+ dout(10) << " oldest old_inode is [" << first << "," << it->first << "], done." << dendl;
//assert(p == pin->old_inodes.begin());
break;
}
- if (p->first > last) {
- dout(10) << " splitting right old_inode [" << first << "," << p->first << "] to ["
- << (last+1) << "," << p->first << "]" << dendl;
- pin->old_inodes[last] = p->second;
- p->second.first = last+1;
- pin->dirty_old_rstats.insert(p->first);
+ if (it->first > last) {
+ dout(10) << " splitting right old_inode [" << first << "," << it->first << "] to ["
+ << (last+1) << "," << it->first << "]" << dendl;
+ pin->old_inodes[last] = it->second;
+ it->second.first = last+1;
+ pin->dirty_old_rstats.insert(it->first);
}
}
if (first < ofirst) {
@@ -1977,7 +1974,7 @@ void MDCache::broadcast_quota_to_client(CInode *in)
if (!in->is_auth() || in->is_frozen())
return;
- inode_t *i = in->get_projected_inode();
+ auto i = in->get_projected_inode();
if (!i->quota.is_enable())
return;
@@ -2249,32 +2246,32 @@ void MDCache::predirty_journal_parents(MutationRef mut, EMetaBlob *blob,
pin->pre_cow_old_inode(); // avoid cow mayhem!
- inode_t *pi = pin->project_inode();
- pi->version = pin->pre_dirty();
+ auto &pi = pin->project_inode();
+ pi.inode.version = pin->pre_dirty();
// dirstat
if (do_parent_mtime || linkunlink) {
dout(20) << "predirty_journal_parents add_delta " << pf->fragstat << dendl;
dout(20) << "predirty_journal_parents - " << pf->accounted_fragstat << dendl;
bool touched_mtime = false, touched_chattr = false;
- pi->dirstat.add_delta(pf->fragstat, pf->accounted_fragstat, &touched_mtime, &touched_chattr);
+ pi.inode.dirstat.add_delta(pf->fragstat, pf->accounted_fragstat, &touched_mtime, &touched_chattr);
pf->accounted_fragstat = pf->fragstat;
if (touched_mtime)
- pi->mtime = pi->ctime = pi->dirstat.mtime;
+ pi.inode.mtime = pi.inode.ctime = pi.inode.dirstat.mtime;
if (touched_chattr)
- pi->change_attr = pi->dirstat.change_attr;
- dout(20) << "predirty_journal_parents gives " << pi->dirstat << " on " << *pin << dendl;
+ pi.inode.change_attr = pi.inode.dirstat.change_attr;
+ dout(20) << "predirty_journal_parents gives " << pi.inode.dirstat << " on " << *pin << dendl;
if (parent->get_frag() == frag_t()) { // i.e., we are the only frag
- if (pi->dirstat.size() < 0)
+ if (pi.inode.dirstat.size() < 0)
assert(!"negative dirstat size" == g_conf->mds_verify_scatter);
- if (pi->dirstat.size() != pf->fragstat.size()) {
+ if (pi.inode.dirstat.size() != pf->fragstat.size()) {
mds->clog->error() << "unmatched fragstat size on single dirfrag "
- << parent->dirfrag() << ", inode has " << pi->dirstat
+ << parent->dirfrag() << ", inode has " << pi.inode.dirstat
<< ", dirfrag has " << pf->fragstat;
// trust the dirfrag for now
- pi->dirstat = pf->fragstat;
+ pi.inode.dirstat = pf->fragstat;
assert(!"unmatched fragstat size" == g_conf->mds_verify_scatter);
}
@@ -2315,13 +2312,13 @@ void MDCache::predirty_journal_parents(MutationRef mut, EMetaBlob *blob,
pf->accounted_rstat = pf->rstat;
if (parent->get_frag() == frag_t()) { // i.e., we are the only frag
- if (pi->rstat.rbytes != pf->rstat.rbytes) {
+ if (pi.inode.rstat.rbytes != pf->rstat.rbytes) {
mds->clog->error() << "unmatched rstat rbytes on single dirfrag "
- << parent->dirfrag() << ", inode has " << pi->rstat
+ << parent->dirfrag() << ", inode has " << pi.inode.rstat
<< ", dirfrag has " << pf->rstat;
// trust the dirfrag for now
- pi->rstat = pf->rstat;
+ pi.inode.rstat = pf->rstat;
assert(!"unmatched rstat rbytes" == g_conf->mds_verify_scatter);
}
@@ -4201,10 +4198,9 @@ void MDCache::rejoin_walk(CDir *dir, MMDSCacheRejoin *rejoin)
if (dn->last != CEPH_NOSNAP) {
if (in && !in->remote_parents.empty()) {
// unlink any stale remote snap dentry.
- for (compact_set<CDentry*>::iterator q = in->remote_parents.begin();
- q != in->remote_parents.end(); ) {
- CDentry *remote_dn = *q;
- ++q;
+ for (auto it2 = in->remote_parents.begin(); it2 != in->remote_parents.end(); ) {
+ CDentry *remote_dn = *it2;
+ ++it2;
assert(remote_dn->last != CEPH_NOSNAP);
remote_dn->unlink_remote(remote_dn->get_linkage());
}
@@ -6177,7 +6173,7 @@ void MDCache::queue_file_recover(CInode *in)
s.erase(*s.rbegin());
dout(10) << " snaps in [" << in->first << "," << in->last << "] are " << s << dendl;
if (s.size() > 1) {
- inode_t *pi = in->project_inode();
+ CInode::mempool_inode pi = in->project_inode();
pi->version = in->pre_dirty();
auto mut(std::make_shared<MutationImpl>());
@@ -6302,7 +6298,7 @@ public:
void MDCache::truncate_inode(CInode *in, LogSegment *ls)
{
- inode_t *pi = in->get_projected_inode();
+ auto pi = in->get_projected_inode();
dout(10) << "truncate_inode "
<< pi->truncate_from << " -> " << pi->truncate_size
<< " on " << *in
@@ -6336,7 +6332,7 @@ struct C_IO_MDC_TruncateFinish : public MDCacheIOContext {
void MDCache::_truncate_inode(CInode *in, LogSegment *ls)
{
- inode_t *pi = &in->inode;
+ auto pi = &in->inode;
dout(10) << "_truncate_inode "
<< pi->truncate_from << " -> " << pi->truncate_size
<< " on " << *in << dendl;
@@ -6385,10 +6381,10 @@ void MDCache::truncate_inode_finish(CInode *in, LogSegment *ls)
ls->truncating_inodes.erase(p);
// update
- inode_t *pi = in->project_inode();
- pi->version = in->pre_dirty();
- pi->truncate_from = 0;
- pi->truncate_pending--;
+ auto &pi = in->project_inode();
+ pi.inode.version = in->pre_dirty();
+ pi.inode.truncate_from = 0;
+ pi.inode.truncate_pending--;
MutationRef mut(new MutationImpl());
mut->ls = mds->mdlog->get_current_segment();
@@ -9372,17 +9368,18 @@ void MDCache::snaprealm_create(MDRequestRef& mdr, CInode *in)
le->metablob.add_table_transaction(TABLE_SNAP, mdr->more()->stid);
- inode_t *pi = in->project_inode();
- pi->version = in->pre_dirty();
- pi->rstat.rsnaprealms++;
+ auto &pi = in->project_inode(false, true);
+ pi.inode.version = in->pre_dirty();
+ pi.inode.rstat.rsnaprealms++;
bufferlist::iterator p = mdr->more()->snapidbl.begin();
snapid_t seq;
decode(seq, p);
- sr_t *newsnap = in->project_snaprealm(seq);
- newsnap->seq = seq;
- newsnap->last_created = seq;
+ auto &newsnap = *pi.snapnode;
+ newsnap.created = seq;
+ newsnap.seq = seq;
+ newsnap.last_created = seq;
predirty_journal_parents(mut, &le->metablob, in, 0, PREDIRTY_PRIMARY);
journal_cow_inode(mut, &le->metablob, in);
@@ -11351,8 +11348,8 @@ void MDCache::dispatch_fragment_dir(MDRequestRef& mdr)
// dft lock
if (diri->is_auth()) {
// journal dirfragtree
- inode_t *pi = diri->project_inode();
- pi->version = diri->pre_dirty();
+ auto &pi = diri->project_inode();
+ pi.inode.version = diri->pre_dirty();
journal_dirty_inode(mdr.get(), &le->metablob, diri);
} else {
mds->locker->mark_updated_scatterlock(&diri->dirfragtreelock);
@@ -11715,7 +11712,8 @@ void MDCache::rollback_uncommitted_fragments()
}
if (diri_auth) {
- diri->project_inode()->version = diri->pre_dirty();
+ auto &pi = diri->project_inode();
+ pi.inode.version = diri->pre_dirty();
diri->pop_and_dirty_projected_inode(ls); // hacky
le->metablob.add_primary_dentry(diri->get_projected_parent_dn(), diri, true);
} else {
diff --git a/src/mds/MDCache.h b/src/mds/MDCache.h
index 7107fb1a358..bc551a15c68 100644
--- a/src/mds/MDCache.h
+++ b/src/mds/MDCache.h
@@ -397,7 +397,7 @@ public:
void project_rstat_inode_to_frag(CInode *cur, CDir *parent, snapid_t first,
int linkunlink, SnapRealm *prealm);
- void _project_rstat_inode_to_frag(inode_t& inode, snapid_t ofirst, snapid_t last,
+ void _project_rstat_inode_to_frag(CInode::mempool_inode & inode, snapid_t ofirst, snapid_t last,
CDir *parent, int linkunlink, bool update_inode);
void project_rstat_frag_to_inode(nest_info_t& rstat, nest_info_t& accounted_rstat,
snapid_t ofirst, snapid_t last,
diff --git a/src/mds/MDSDaemon.cc b/src/mds/MDSDaemon.cc
index ca0dc8cfed8..bbded0e8af6 100644
--- a/src/mds/MDSDaemon.cc
+++ b/src/mds/MDSDaemon.cc
@@ -437,7 +437,8 @@ int MDSDaemon::init()
dout(10) << sizeof(MDSCacheObject) << "\tMDSCacheObject" << dendl;
dout(10) << sizeof(CInode) << "\tCInode" << dendl;
dout(10) << sizeof(elist<void*>::item) << "\t elist<>::item *7=" << 7*sizeof(elist<void*>::item) << dendl;
- dout(10) << sizeof(inode_t) << "\t inode_t " << dendl;
+ dout(10) << sizeof(CInode::mempool_inode) << "\t inode " << dendl;
+ dout(10) << sizeof(CInode::mempool_old_inode) << "\t old_inode " << dendl;
dout(10) << sizeof(nest_info_t) << "\t nest_info_t " << dendl;
dout(10) << sizeof(frag_info_t) << "\t frag_info_t " << dendl;
dout(10) << sizeof(SimpleLock) << "\t SimpleLock *5=" << 5*sizeof(SimpleLock) << dendl;
diff --git a/src/mds/RecoveryQueue.cc b/src/mds/RecoveryQueue.cc
index d205c6f278b..2e6ff30f2b8 100644
--- a/src/mds/RecoveryQueue.cc
+++ b/src/mds/RecoveryQueue.cc
@@ -87,7 +87,7 @@ void RecoveryQueue::advance()
void RecoveryQueue::_start(CInode *in)
{
- inode_t *pi = in->get_projected_inode();
+ auto pi = in->get_projected_inode();
// blech
if (pi->client_ranges.size() && !pi->get_max_size()) {
diff --git a/src/mds/Server.cc b/src/mds/Server.cc
index d923fecd93a..7ada87c9da3 100644
--- a/src/mds/Server.cc
+++ b/src/mds/Server.cc
@@ -2665,11 +2665,13 @@ CInode* Server::prepare_new_inode(MDRequestRef& mdr, CDir *dir, inodeno_t useino
bufferlist::iterator p = req->get_data().begin();
// xattrs on new inode?
- map<string,bufferptr> xattrs;
+ CInode::mempool_xattr_map xattrs;
decode(xattrs, p);
- for (map<string,bufferptr>::iterator p = xattrs.begin(); p != xattrs.end(); ++p) {
- dout(10) << "prepare_new_inode setting xattr " << p->first << dendl;
- in->xattrs[p->first] = p->second;
+ for (const auto &p : xattrs) {
+ dout(10) << "prepare_new_inode setting xattr " << p.first << dendl;
+ auto em = in->xattrs.emplace(std::piecewise_construct, std::forward_as_tuple(p.first), std::forward_as_tuple(p.second));
+ if (!em.second)
+ em.first->second = p.second;
}
}
@@ -3373,7 +3375,7 @@ void Server::handle_client_open(MDRequestRef& mdr)
return;
// wait for pending truncate?
- const inode_t *pi = cur->get_projected_inode();
+ const auto pi = cur->get_projected_inode();
if (pi->is_truncating()) {
dout(10) << " waiting for pending truncate from " << pi->truncate_from
<< " to " << pi->truncate_size << " to complete on " << *cur << dendl;
@@ -4147,9 +4149,9 @@ void Server::handle_client_setattr(MDRequestRef& mdr)
return;
// trunc from bigger -> smaller?
- inode_t *pi = cur->get_projected_inode();
+ auto pip = cur->get_projected_inode();
- uint64_t old_size = std::max<uint64_t>(pi->size, req->head.args.setattr.old_size);
+ uint64_t old_size = std::max<uint64_t>(pip->size, req->head.args.setattr.old_size);
// ENOSPC on growing file while full, but allow shrinks
if (is_full && req->head.args.setattr.size > old_size) {
@@ -4161,9 +4163,9 @@ void Server::handle_client_setattr(MDRequestRef& mdr)
bool truncating_smaller = false;
if (mask & CEPH_SETATTR_SIZE) {
truncating_smaller = req->head.args.setattr.size < old_size;
- if (truncating_smaller && pi->is_truncating()) {
- dout(10) << " waiting for pending truncate from " << pi->truncate_from
- << " to " << pi->truncate_size << " to complete on " << *cur << dendl;
+ if (truncating_smaller && pip->is_truncating()) {
+ dout(10) << " waiting for pending truncate from " << pip->truncate_from
+ << " to " << pip->truncate_size << " to complete on " << *cur << dendl;
mds->locker->drop_locks(mdr.get());
mdr->drop_local_auth_pins();
cur->add_waiter(CInode::WAIT_TRUNC, new C_MDS_RetryRequest(mdcache, mdr));
@@ -4178,53 +4180,53 @@ void Server::handle_client_setattr(MDRequestRef& mdr)
EUpdate *le = new EUpdate(mdlog, "setattr");
mdlog->start_entry(le);
- pi = cur->project_inode();
+ auto &pi = cur->project_inode();
if (mask & CEPH_SETATTR_UID)
- pi->uid = req->head.args.setattr.uid;
+ pi.inode.uid = req->head.args.setattr.uid;
if (mask & CEPH_SETATTR_GID)
- pi->gid = req->head.args.setattr.gid;
+ pi.inode.gid = req->head.args.setattr.gid;
if (mask & CEPH_SETATTR_MODE)
- pi->mode = (pi->mode & ~07777) | (req->head.args.setattr.mode & 07777);
+ pi.inode.mode = (pi.inode.mode & ~07777) | (req->head.args.setattr.mode & 07777);
else if ((mask & (CEPH_SETATTR_UID|CEPH_SETATTR_GID|CEPH_SETATTR_KILL_SGUID)) &&
- S_ISREG(pi->mode) &&
- (pi->mode & (S_IXUSR|S_IXGRP|S_IXOTH))) {
- pi->mode &= ~(S_ISUID|S_ISGID);
+ S_ISREG(pi.inode.mode) &&
+ (pi.inode.mode & (S_IXUSR|S_IXGRP|S_IXOTH))) {
+ pi.inode.mode &= ~(S_ISUID|S_ISGID);
}
if (mask & CEPH_SETATTR_MTIME)
- pi->mtime = req->head.args.setattr.mtime;
+ pi.inode.mtime = req->head.args.setattr.mtime;
if (mask & CEPH_SETATTR_ATIME)
- pi->atime = req->head.args.setattr.atime;
+ pi.inode.atime = req->head.args.setattr.atime;
if (mask & CEPH_SETATTR_BTIME)
- pi->btime = req->head.args.setattr.btime;
+ pi.inode.btime = req->head.args.setattr.btime;
if (mask & (CEPH_SETATTR_ATIME | CEPH_SETATTR_MTIME | CEPH_SETATTR_BTIME))
- pi->time_warp_seq++; // maybe not a timewarp, but still a serialization point.
+ pi.inode.time_warp_seq++; // maybe not a timewarp, but still a serialization point.
if (mask & CEPH_SETATTR_SIZE) {
if (truncating_smaller) {
- pi->truncate(old_size, req->head.args.setattr.size);
+ pi.inode.truncate(old_size, req->head.args.setattr.size);
le->metablob.add_truncate_start(cur->ino());
} else {
- pi->size = req->head.args.setattr.size;
- pi->rstat.rbytes = pi->size;
+ pi.inode.size = req->head.args.setattr.size;
+ pi.inode.rstat.rbytes = pi.inode.size;
}
- pi->mtime = mdr->get_op_stamp();
+ pi.inode.mtime = mdr->get_op_stamp();
// adjust client's max_size?
- map<client_t,client_writeable_range_t> new_ranges;
+ CInode::mempool_inode::client_range_map new_ranges;
bool max_increased = false;
- mds->locker->calc_new_client_ranges(cur, pi->size, &new_ranges, &max_increased);
- if (pi->client_ranges != new_ranges) {
- dout(10) << " client_ranges " << pi->client_ranges << " -> " << new_ranges << dendl;
- pi->client_ranges = new_ranges;
+ mds->locker->calc_new_client_ranges(cur, pi.inode.size, &new_ranges, &max_increased);
+ if (pi.inode.client_ranges != new_ranges) {
+ dout(10) << " client_ranges " << pi.inode.client_ranges << " -> " << new_ranges << dendl;
+ pi.inode.client_ranges = new_ranges;
changed_ranges = true;
}
}
- pi->version = cur->pre_dirty();
- pi->ctime = mdr->get_op_stamp();
- pi->change_attr++;
+ pi.inode.version = cur->pre_dirty();
+ pi.inode.ctime = mdr->get_op_stamp();
+ pi.inode.change_attr++;
// log + wait
le->metablob.add_client_req(req->get_reqid(), req->get_oldest_client_tid());
@@ -4257,22 +4259,22 @@ void Server::do_open_truncate(MDRequestRef& mdr, int cmode)
mdlog->start_entry(le);
// prepare
- inode_t *pi = in->project_inode();
- pi->version = in->pre_dirty();
- pi->mtime = pi->ctime = mdr->get_op_stamp();
- pi->change_attr++;
+ auto &pi = in->project_inode();
+ pi.inode.version = in->pre_dirty();
+ pi.inode.mtime = pi.inode.ctime = mdr->get_op_stamp();
+ pi.inode.change_attr++;
- uint64_t old_size = std::max<uint64_t>(pi->size, mdr->client_request->head.args.open.old_size);
+ uint64_t old_size = std::max<uint64_t>(pi.inode.size, mdr->client_request->head.args.open.old_size);
if (old_size > 0) {
- pi->truncate(old_size, 0);
+ pi.inode.truncate(old_size, 0);
le->metablob.add_truncate_start(in->ino());
}
bool changed_ranges = false;
if (cmode & CEPH_FILE_MODE_WR) {
- pi->client_ranges[client].range.first = 0;
- pi->client_ranges[client].range.last = pi->get_layout_size_increment();
- pi->client_ranges[client].follows = realm->get_newest_seq();
+ pi.inode.client_ranges[client].range.first = 0;
+ pi.inode.client_ranges[client].range.last = pi.inode.get_layout_size_increment();
+ pi.inode.client_ranges[client].follows = realm->get_newest_seq();
changed_ranges = true;
}
@@ -4372,13 +4374,13 @@ void Server::handle_client_setlayout(MDRequestRef& mdr)
return;
// project update
- inode_t *pi = cur->project_inode();
- pi->layout = layout;
+ auto &pi = cur->project_inode();
+ pi.inode.layout = layout;
// add the old pool to the inode
- pi->add_old_pool(old_layout.pool_id);
- pi->version = cur->pre_dirty();
- pi->ctime = mdr->get_op_stamp();
- pi->change_attr++;
+ pi.inode.add_old_pool(old_layout.pool_id);
+ pi.inode.version = cur->pre_dirty();
+ pi.inode.ctime = mdr->get_op_stamp();
+ pi.inode.change_attr++;
// log + wait
mdr->ls = mdlog->get_current_segment();
@@ -4414,7 +4416,7 @@ void Server::handle_client_setdirlayout(MDRequestRef& mdr)
return;
// validate layout
- const inode_t *old_pi = cur->get_projected_inode();
+ const auto old_pi = cur->get_projected_inode();
file_layout_t layout;
if (old_pi->has_layout())
layout = old_pi->layout;
@@ -4461,9 +4463,9 @@ void Server::handle_client_setdirlayout(MDRequestRef& mdr)
if (!check_access(mdr, cur, access))
return;
- inode_t *pi = cur->project_inode();
- pi->layout = layout;
- pi->version = cur->pre_dirty();
+ auto &pi = cur->project_inode();
+ pi.inode.layout = layout;
+ pi.inode.version = cur->pre_dirty();
// log + wait
mdr->ls = mdlog->get_current_segment();
@@ -4674,7 +4676,7 @@ void Server::handle_set_vxattr(MDRequestRef& mdr, CInode *cur,
<< " bytes on " << *cur
<< dendl;
- inode_t *pi = NULL;
+ CInode::mempool_inode *pip = nullptr;
string rest;
if (!check_access(mdr, cur, MAY_SET_VXATTR)) {
@@ -4704,9 +4706,10 @@ void Server::handle_set_vxattr(MDRequestRef& mdr, CInode *cur,
if (!mds->locker->acquire_locks(mdr, rdlocks, wrlocks, xlocks))
return;
- pi = cur->project_inode();
- pi->layout = layout;
+ auto &pi = cur->project_inode();
+ pi.inode.layout = layout;
mdr->no_early_reply = true;
+ pip = &pi.inode;
} else if (name.compare(0, 16, "ceph.file.layout") == 0) {
if (!cur->is_file()) {
respond_to_request(mdr, -EINVAL);
@@ -4726,11 +4729,12 @@ void Server::handle_set_vxattr(MDRequestRef& mdr, CInode *cur,
if (!mds->locker->acquire_locks(mdr, rdlocks, wrlocks, xlocks))
return;
- pi = cur->project_inode();
- int64_t old_pool = pi->layout.pool_id;
- pi->add_old_pool(old_pool);
- pi->layout = layout;
- pi->ctime = mdr->get_op_stamp();
+ auto &pi = cur->project_inode();
+ int64_t old_pool = pi.inode.layout.pool_id;
+ pi.inode.add_old_pool(old_pool);
+ pi.inode.layout = layout;
+ pi.inode.ctime = mdr->get_op_stamp();
+ pip = &pi.inode;
} else if (name.compare(0, 10, "ceph.quota") == 0) {
if (!cur->is_dir() || cur->is_root()) {
respond_to_request(mdr, -EINVAL);
@@ -4755,15 +4759,18 @@ void Server::handle_set_vxattr(MDRequestRef& mdr, CInode *cur,
if (!mds->locker->acquire_locks(mdr, rdlocks, wrlocks, xlocks))
return;
- pi = cur->project_inode();
- pi->quota = quota;
+ auto &pi = cur->project_inode(false, new_realm);
+ pi.inode.quota = quota;
if (new_realm) {
SnapRealm *realm = cur->find_snaprealm();
- sr_t *newsnap = cur->project_snaprealm(realm->get_newest_seq());
- newsnap->seq = realm->get_newest_seq();
+ auto seq = realm->get_newest_seq();
+ auto &newsnap = *pi.snapnode;
+ newsnap.created = seq;
+ newsnap.seq = seq;
}
mdr->no_early_reply = true;
+ pip = &pi.inode;
} else if (name.find("ceph.dir.pin") == 0) {
if (!cur->is_dir() || cur->is_root()) {
respond_to_request(mdr, -EINVAL);
@@ -4784,19 +4791,20 @@ void Server::handle_set_vxattr(MDRequestRef& mdr, CInode *cur,
if (!mds->locker->acquire_locks(mdr, rdlocks, wrlocks, xlocks))
return;
- pi = cur->project_inode();
+ auto &pi = cur->project_inode();
cur->set_export_pin(rank);
+ pip = &pi.inode;
} else {
dout(10) << " unknown vxattr " << name << dendl;
respond_to_request(mdr, -EINVAL);
return;
}
- pi->change_attr++;
- pi->ctime = mdr->get_op_stamp();
- pi->version = cur->pre_dirty();
+ pip->change_attr++;
+ pip->ctime = mdr->get_op_stamp();
+ pip->version = cur->pre_dirty();
if (cur->is_file())
- pi->update_backtrace();
+ pip->update_backtrace();
// log + wait
mdr->ls = mdlog->get_current_segment();
@@ -4842,9 +4850,9 @@ void Server::handle_remove_vxattr(MDRequestRef& mdr, CInode *cur,
if (!mds->locker->acquire_locks(mdr, rdlocks, wrlocks, xlocks))
return;
- inode_t *pi = cur->project_inode();
- pi->clear_layout();
- pi->version = cur->pre_dirty();
+ auto &pi = cur->project_inode();
+ pi.inode.clear_layout();
+ pi.inode.version = cur->pre_dirty();
// log + wait
mdr->ls = mdlog->get_current_segment();
@@ -4925,7 +4933,7 @@ void Server::handle_client_setxattr(MDRequestRef& mdr)
if (!check_access(mdr, cur, MAY_WRITE))
return;
- map<string, bufferptr> *pxattrs = cur->get_projected_xattrs();
+ auto pxattrs = cur->get_projected_xattrs();
size_t len = req->get_data().length();
size_t inc = len + name.length();
@@ -4945,12 +4953,12 @@ void Server::handle_client_setxattr(MDRequestRef& mdr)
return;
}
- if ((flags & CEPH_XATTR_CREATE) && pxattrs->count(name)) {
+ if ((flags & CEPH_XATTR_CREATE) && pxattrs->count(mempool::mds_co::string(name))) {
dout(10) << "setxattr '" << name << "' XATTR_CREATE and EEXIST on " << *cur << dendl;
respond_to_request(mdr, -EEXIST);
return;
}
- if ((flags & CEPH_XATTR_REPLACE) && !pxattrs->count(name)) {
+ if ((flags & CEPH_XATTR_REPLACE) && !pxattrs->count(mempool::mds_co::string(name))) {
dout(10) << "setxattr '" << name << "' XATTR_REPLACE and ENODATA on " << *cur << dendl;
respond_to_request(mdr, -ENODATA);
return;
@@ -4959,17 +4967,21 @@ void Server::handle_client_setxattr(MDRequestRef& mdr)
dout(10) << "setxattr '" << name << "' len " << len << " on " << *cur << dendl;
// project update
- map<string,bufferptr> *px = new map<string,bufferptr>;
- inode_t *pi = cur->project_inode(px);
- pi->version = cur->pre_dirty();
- pi->ctime = mdr->get_op_stamp();
- pi->change_attr++;
- pi->xattr_version++;
- px->erase(name);
- if (!(flags & CEPH_XATTR_REMOVE)) {
- (*px)[name] = buffer::create(len);
+ auto &pi = cur->project_inode(true);
+ pi.inode.version = cur->pre_dirty();
+ pi.inode.ctime = mdr->get_op_stamp();
+ pi.inode.change_attr++;
+ pi.inode.xattr_version++;
+ auto &px = *pi.xattrs;
+ if ((flags & CEPH_XATTR_REMOVE)) {
+ px.erase(mempool::mds_co::string(name));
+ } else {
+ bufferptr b = buffer::create(len);
if (len)
- req->get_data().copy(0, len, (*px)[name].c_str());
+ req->get_data().copy(0, len, b.c_str());
+ auto em = px.emplace(std::piecewise_construct, std::forward_as_tuple(mempool::mds_co::string(name)), std::forward_as_tuple(b));
+ if (!em.second)
+ em.first->second = b;
}
// log + wait
@@ -4986,8 +4998,8 @@ void Server::handle_client_setxattr(MDRequestRef& mdr)
void Server::handle_client_removexattr(MDRequestRef& mdr)
{
MClientRequest *req = mdr->client_request;
- string name(req->get_path2());
- set<SimpleLock*> rdlocks, wrlocks, xlocks;
+ std::string name(req->get_path2());
+ std::set<SimpleLock*> rdlocks, wrlocks, xlocks;
file_layout_t *dir_layout = NULL;
CInode *cur;
if (name == "ceph.dir.layout")
@@ -5011,8 +5023,8 @@ void Server::handle_client_removexattr(MDRequestRef& mdr)
if (!mds->locker->acquire_locks(mdr, rdlocks, wrlocks, xlocks))
return;
- map<string, bufferptr> *pxattrs = cur->get_projected_xattrs();
- if (pxattrs->count(name) == 0) {
+ auto pxattrs = cur->get_projected_xattrs();
+ if (pxattrs->count(mempool::mds_co::string(name)) == 0) {
dout(10) << "removexattr '" << name << "' and ENODATA on " << *cur << dendl;
respond_to_request(mdr, -ENODATA);
return;
@@ -5021,13 +5033,13 @@ void Server::handle_client_removexattr(MDRequestRef& mdr)
dout(10) << "removexattr '" << name << "' on " << *cur << dendl;
// project update
- map<string,bufferptr> *px = new map<string,bufferptr>;
- inode_t *pi = cur->project_inode(px);
- pi->version = cur->pre_dirty();
- pi->ctime = mdr->get_op_stamp();
- pi->change_attr++;
- pi->xattr_version++;
- px->erase(name);
+ auto &pi = cur->project_inode(true);
+ auto &px = *pi.xattrs;
+ pi.inode.version = cur->pre_dirty();
+ pi.inode.ctime = mdr->get_op_stamp();
+ pi.inode.change_attr++;
+ pi.inode.xattr_version++;
+ px.erase(mempool::mds_co::string(name));
// log + wait
mdr->ls = mdlog->get_current_segment();
@@ -5416,11 +5428,11 @@ void Server::_link_local(MDRequestRef& mdr, CDentry *dn, CInode *targeti)
version_t tipv = targeti->pre_dirty();
// project inode update
- inode_t *pi = targeti->project_inode();
- pi->nlink++;
- pi->ctime = mdr->get_op_stamp();
- pi->change_attr++;
- pi->version = tipv;
+ auto &pi = targeti->project_inode();
+ pi.inode.nlink++;
+ pi.inode.ctime = mdr->get_op_stamp();
+ pi.inode.change_attr++;
+ pi.inode.version = tipv;
// log + wait
EUpdate *le = new EUpdate(mdlog, "link_local");
@@ -5648,16 +5660,16 @@ void Server::handle_slave_link_prep(MDRequestRef& mdr)
ESlaveUpdate::OP_PREPARE, ESlaveUpdate::LINK);
mdlog->start_entry(le);
- inode_t *pi = dnl->get_inode()->project_inode();
+ auto &pi = dnl->get_inode()->project_inode();
// update journaled target inode
bool inc;
if (mdr->slave_request->get_op() == MMDSSlaveRequest::OP_LINKPREP) {
inc = true;
- pi->nlink++;
+ pi.inode.nlink++;
} else {
inc = false;
- pi->nlink--;
+ pi.inode.nlink--;
}
link_rollback rollback;
@@ -5671,10 +5683,10 @@ void Server::handle_slave_link_prep(MDRequestRef& mdr)
encode(rollback, le->rollback);
mdr->more()->rollback_bl = le->rollback;
- pi->ctime = mdr->get_op_stamp();
- pi->version = targeti->pre_dirty();
+ pi.inode.ctime = mdr->get_op_stamp();
+ pi.inode.version = targeti->pre_dirty();
- dout(10) << " projected inode " << pi << " v " << pi->version << dendl;
+ dout(10) << " projected inode " << pi.inode.ino << " v " << pi.inode.version << dendl;
// commit case
mdcache->predirty_journal_parents(mdr, &le->commit, dnl->get_inode(), 0, PREDIRTY_SHALLOW|PREDIRTY_PRIMARY);
@@ -5793,8 +5805,8 @@ void Server::do_link_rollback(bufferlist &rbl, mds_rank_t master, MDRequestRef&
dout(10) << " target is " << *in << dendl;
assert(!in->is_projected()); // live slave request hold versionlock xlock.
- inode_t *pi = in->project_inode();
- pi->version = in->pre_dirty();
+ auto &pi = in->project_inode();
+ pi.inode.version = in->pre_dirty();
mut->add_projected_inode(in);
// parent dir rctime
@@ -5802,20 +5814,20 @@ void Server::do_link_rollback(bufferlist &rbl, mds_rank_t master, MDRequestRef&
fnode_t *pf = parent->project_fnode();
mut->add_projected_fnode(parent);
pf->version = parent->pre_dirty();
- if (pf->fragstat.mtime == pi->ctime) {
+ if (pf->fragstat.mtime == pi.inode.ctime) {
pf->fragstat.mtime = rollback.old_dir_mtime;
- if (pf->rstat.rctime == pi->ctime)
+ if (pf->rstat.rctime == pi.inode.ctime)
pf->rstat.rctime = rollback.old_dir_rctime;
mut->add_updated_lock(&parent->get_inode()->filelock);
mut->add_updated_lock(&parent->get_inode()->nestlock);
}
// inode
- pi->ctime = rollback.old_ctime;
+ pi.inode.ctime = rollback.old_ctime;
if (rollback.was_inc)
- pi->nlink--;
+ pi.inode.nlink--;
else
- pi->nlink++;
+ pi.inode.nlink++;
// journal it
ESlaveUpdate *le = new ESlaveUpdate(mdlog, "slave_link_rollback", rollback.reqid, master,
@@ -6076,14 +6088,18 @@ void Server::_unlink_local(MDRequestRef& mdr, CDentry *dn, CDentry *straydn, sna
// the unlinked dentry
dn->pre_dirty();
- inode_t *pi = in->project_inode();
- dn->make_path_string(pi->stray_prior_path, true);
+ auto &pi = in->project_inode();
+ {
+ std::string t;
+ dn->make_path_string(t, true);
+ pi.inode.stray_prior_path = std::move(t);
+ }
mdr->add_projected_inode(in); // do this _after_ my dn->pre_dirty().. we apply that one manually.
- pi->version = in->pre_dirty();
- pi->ctime = mdr->get_op_stamp();
- pi->change_attr++;
- pi->nlink--;
- if (pi->nlink == 0)
+ pi.inode.version = in->pre_dirty();
+ pi.inode.ctime = mdr->get_op_stamp();
+ pi.inode.change_attr++;
+ pi.inode.nlink--;
+ if (pi.inode.nlink == 0)
in->state_set(CInode::STATE_ORPHAN);
if (dnl->is_primary()) {
@@ -6096,7 +6112,7 @@ void Server::_unlink_local(MDRequestRef& mdr, CDentry *dn, CDentry *straydn, sna
if (in->snaprealm || follows + 1 > in->get_oldest_snap())
in->project_past_snaprealm_parent(straydn->get_dir()->inode->find_snaprealm());
- pi->update_backtrace();
+ pi.inode.update_backtrace();
le->metablob.add_primary_dentry(straydn, in, true, true);
} else {
// remote link. update remote inode.
@@ -7193,8 +7209,8 @@ void Server::_rename_prepare(MDRequestRef& mdr,
}
// prepare
- inode_t *pi = 0; // renamed inode
- inode_t *tpi = 0; // target/overwritten inode
+ CInode::mempool_inode *spi = 0; // renamed inode
+ CInode::mempool_inode *tpi = 0; // target/overwritten inode
// target inode
if (!linkmerge) {
@@ -7202,16 +7218,18 @@ void Server::_rename_prepare(MDRequestRef& mdr,
assert(straydn); // moving to straydn.
// link--, and move.
if (destdn->is_auth()) {
- tpi = oldin->project_inode(); //project_snaprealm
- tpi->version = straydn->pre_dirty(tpi->version);
- tpi->update_backtrace();
+ auto &pi= oldin->project_inode(); //project_snaprealm
+ pi.inode.version = straydn->pre_dirty(pi.inode.version);
+ pi.inode.update_backtrace();
+ tpi = &pi.inode;
}
straydn->push_projected_linkage(oldin);
} else if (destdnl->is_remote()) {
// nlink-- targeti
if (oldin->is_auth()) {
- tpi = oldin->project_inode();
- tpi->version = oldin->pre_dirty();
+ auto &pi = oldin->project_inode();
+ pi.inode.version = oldin->pre_dirty();
+ tpi = &pi.inode;
}
}
}
@@ -7225,14 +7243,16 @@ void Server::_rename_prepare(MDRequestRef& mdr,
destdn->push_projected_linkage(srcdnl->get_remote_ino(), srcdnl->get_remote_d_type());
// srci
if (srci->is_auth()) {
- pi = srci->project_inode();
- pi->version = srci->pre_dirty();
+ auto &pi = srci->project_inode();
+ pi.inode.version = srci->pre_dirty();
+ spi = &pi.inode;
}
} else {
dout(10) << " will merge remote onto primary link" << dendl;
if (destdn->is_auth()) {
- pi = oldin->project_inode();
- pi->version = mdr->more()->pvmap[destdn] = destdn->pre_dirty(oldin->inode.version);
+ auto &pi = oldin->project_inode();
+ pi.inode.version = mdr->more()->pvmap[destdn] = destdn->pre_dirty(oldin->inode.version);
+ spi = &pi.inode;
}
}
} else { // primary
@@ -7256,10 +7276,11 @@ void Server::_rename_prepare(MDRequestRef& mdr,
dout(10) << " noting renamed dir open frags " << metablob->renamed_dir_frags << dendl;
}
}
- pi = srci->project_inode(); // project snaprealm if srcdnl->is_primary
+ auto &pi = srci->project_inode(); // project snaprealm if srcdnl->is_primary
// & srcdnl->snaprealm
- pi->version = mdr->more()->pvmap[destdn] = destdn->pre_dirty(oldpv);
- pi->update_backtrace();
+ pi.inode.version = mdr->more()->pvmap[destdn] = destdn->pre_dirty(oldpv);
+ pi.inode.update_backtrace();
+ spi = &pi.inode;
}
destdn->push_projected_linkage(srci);
}
@@ -7270,16 +7291,20 @@ void Server::_rename_prepare(MDRequestRef& mdr,
srcdn->push_projected_linkage(); // push null linkage
if (!silent) {
- if (pi) {
- pi->ctime = mdr->get_op_stamp();
- pi->change_attr++;
+ if (spi) {
+ spi->ctime = mdr->get_op_stamp();
+ spi->change_attr++;
if (linkmerge)
- pi->nlink--;
+ spi->nlink--;
}
if (tpi) {
tpi->ctime = mdr->get_op_stamp();
tpi->change_attr++;
- destdn->make_path_string(tpi->stray_prior_path, true);
+ {
+ std::string t;
+ destdn->make_path_string(t, true);
+ tpi->stray_prior_path = std::move(t);
+ }
tpi->nlink--;
if (tpi->nlink == 0)
oldin->state_set(CInode::STATE_ORPHAN);
@@ -8163,22 +8188,23 @@ void Server::do_rename_rollback(bufferlist &rbl, mds_rank_t master, MDRequestRef
rollback.orig_src.remote_d_type);
}
- inode_t *pi = 0;
+ CInode::mempool_inode *pip = 0;
if (in) {
if (in->authority().first == whoami) {
- pi = in->project_inode();
+ auto &pi = in->project_inode();
mut->add_projected_inode(in);
- pi->version = in->pre_dirty();
+ pi.inode.version = in->pre_dirty();
+ pip = &pi.inode;
} else
- pi = in->get_projected_inode();
- if (pi->ctime == rollback.ctime)
- pi->ctime = rollback.orig_src.old_ctime;
+ pip = in->get_projected_inode();
+ if (pip->ctime == rollback.ctime)
+ pip->ctime = rollback.orig_src.old_ctime;
}
if (srcdn && srcdn->authority().first == whoami) {
nest_info_t blah;
_rollback_repair_dir(mut, srcdir, rollback.orig_src, rollback.ctime,
- in ? in->is_dir() : false, 1, pi ? pi->accounted_rstat : blah);
+ in ? in->is_dir() : false, 1, pip ? pip->accounted_rstat : blah);
}
// repair dest
@@ -8200,11 +8226,12 @@ void Server::do_rename_rollback(bufferlist &rbl, mds_rank_t master, MDRequestRef
straydn->push_projected_linkage();
if (target) {
- inode_t *ti = NULL;
+ CInode::mempool_inode *ti = NULL;
if (target->authority().first == whoami) {
- ti = target->project_inode();
+ auto &pi = target->project_inode();
mut->add_projected_inode(target);
- ti->version = target->pre_dirty();
+ pi.inode.version = target->pre_dirty();
+ ti = &pi.inode;
} else
ti = target->get_projected_inode();
if (ti->ctime == rollback.ctime)
@@ -8661,15 +8688,18 @@ void Server::handle_client_mksnap(MDRequestRef& mdr)
info.name = snapname;
info.stamp = mdr->get_op_stamp();
- inode_t *pi = diri->project_inode();
- pi->ctime = info.stamp;
- pi->version = diri->pre_dirty();
+ auto &pi = diri->project_inode(false, true);
+ pi.inode.ctime = info.stamp;
+ pi.inode.version = diri->pre_dirty();
// project the snaprealm
- sr_t *newsnap = diri->project_snaprealm(snapid);
- newsnap->snaps[snapid] = info;
- newsnap->seq = snapid;
- newsnap->last_created = snapid;
+ auto &newsnap = *pi.snapnode;
+ newsnap.created = snapid;
+ auto em = newsnap.snaps.emplace(std::piecewise_construct, std::forward_as_tuple(snapid), std::forward_as_tuple(info));
+ if (!em.second)
+ em.first->second = info;
+ newsnap.seq = snapid;
+ newsnap.last_created = snapid;
// journal the inode changes
mdr->ls = mdlog->get_current_segment();
@@ -8789,19 +8819,19 @@ void Server::handle_client_rmsnap(MDRequestRef& mdr)
dout(10) << " stid is " << stid << ", seq is " << seq << dendl;
// journal
- inode_t *pi = diri->project_inode();
- pi->version = diri->pre_dirty();
- pi->ctime = mdr->get_op_stamp();
+ auto &pi = diri->project_inode(false, true);
+ pi.inode.version = diri->pre_dirty();
+ pi.inode.ctime = mdr->get_op_stamp();
mdr->ls = mdlog->get_current_segment();
EUpdate *le = new EUpdate(mdlog, "rmsnap");
mdlog->start_entry(le);
// project the snaprealm
- sr_t *newnode = diri->project_snaprealm();
- newnode->snaps.erase(snapid);
- newnode->seq = seq;
- newnode->last_destroyed = seq;
+ auto &newnode = *pi.snapnode;
+ newnode.snaps.erase(snapid);
+ newnode.seq = seq;
+ newnode.last_destroyed = seq;
le->metablob.add_client_req(req->get_reqid(), req->get_oldest_client_tid());
le->metablob.add_table_transaction(TABLE_SNAP, stid);
@@ -8929,14 +8959,15 @@ void Server::handle_client_renamesnap(MDRequestRef& mdr)
dout(10) << " stid is " << stid << dendl;
// journal
- inode_t *pi = diri->project_inode();
- pi->ctime = mdr->get_op_stamp();
- pi->version = diri->pre_dirty();
+ auto &pi = diri->project_inode(false, true);
+ pi.inode.ctime = mdr->get_op_stamp();
+ pi.inode.version = diri->pre_dirty();
// project the snaprealm
- sr_t *newsnap = diri->project_snaprealm();
- assert(newsnap->snaps.count(snapid));
- newsnap->snaps[snapid].name = dstname;
+ auto &newsnap = *pi.snapnode;
+ auto it = newsnap.snaps.find(snapid);
+ assert(it != newsnap.snaps.end());
+ it->second.name = dstname;
// journal the inode changes
mdr->ls = mdlog->get_current_segment();
diff --git a/src/mds/StrayManager.cc b/src/mds/StrayManager.cc
index e1cd070c18c..b7f94fa5ef0 100644
--- a/src/mds/StrayManager.cc
+++ b/src/mds/StrayManager.cc
@@ -125,11 +125,13 @@ void StrayManager::purge(CDentry *dn)
to = std::max(in->inode.max_size_ever, to);
}
- inode_t *pi = in->get_projected_inode();
+ auto pi = in->get_projected_inode();
item.size = to;
item.layout = pi->layout;
- item.old_pools = pi->old_pools;
+ item.old_pools.clear();
+ for (const auto &p : pi->old_pools)
+ item.old_pools.insert(p);
item.snapc = *snapc;
}
@@ -175,13 +177,13 @@ void StrayManager::_purge_stray_purged(
EUpdate *le = new EUpdate(mds->mdlog, "purge_stray truncate");
mds->mdlog->start_entry(le);
- 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;
- pi->version = in->pre_dirty();
+ auto &pi = in->project_inode();
+ pi.inode.size = 0;
+ pi.inode.max_size_ever = 0;
+ pi.inode.client_ranges.clear();
+ pi.inode.truncate_size = 0;
+ pi.inode.truncate_from = 0;
+ pi.inode.version = in->pre_dirty();
le->metablob.add_dir_context(dn->dir);
le->metablob.add_primary_dentry(dn, in, true);
@@ -463,10 +465,9 @@ bool StrayManager::_eval_stray(CDentry *dn, bool delay)
if (!in->remote_parents.empty()) {
// unlink any stale remote snap dentry.
- for (compact_set<CDentry*>::iterator p = in->remote_parents.begin();
- p != in->remote_parents.end(); ) {
- CDentry *remote_dn = *p;
- ++p;
+ for (auto it = in->remote_parents.begin(); it != in->remote_parents.end(); ) {
+ CDentry *remote_dn = *it;
+ ++it;
assert(remote_dn->last != CEPH_NOSNAP);
remote_dn->unlink_remote(remote_dn->get_linkage());
}
@@ -606,18 +607,17 @@ void StrayManager::_eval_stray_remote(CDentry *stray_dn, CDentry *remote_dn)
/* If no remote_dn hinted, pick one arbitrarily */
if (remote_dn == NULL) {
if (!stray_in->remote_parents.empty()) {
- for (compact_set<CDentry*>::iterator p = stray_in->remote_parents.begin();
- p != stray_in->remote_parents.end();
- ++p)
- if ((*p)->last == CEPH_NOSNAP && !(*p)->is_projected()) {
- if ((*p)->is_auth()) {
- remote_dn = *p;
+ for (const auto &dn : stray_in->remote_parents) {
+ if (dn->last == CEPH_NOSNAP && !dn->is_projected()) {
+ if (dn->is_auth()) {
+ remote_dn = dn;
if (remote_dn->dir->can_auth_pin())
break;
} else if (!remote_dn) {
- remote_dn = *p;
+ remote_dn = dn;
}
}
+ }
}
if (!remote_dn) {
dout(20) << __func__ << ": not reintegrating (no remote parents in cache)" << dendl;
diff --git a/src/mds/events/EMetaBlob.h b/src/mds/events/EMetaBlob.h
index 55db2db31ec..c38c2ccb528 100644
--- a/src/mds/events/EMetaBlob.h
+++ b/src/mds/events/EMetaBlob.h
@@ -65,27 +65,26 @@ public:
static const int STATE_DIRTYPARENT = (1<<1);
static const int STATE_DIRTYPOOL = (1<<2);
static const int STATE_NEED_SNAPFLUSH = (1<<3);
- typedef compact_map<snapid_t, old_inode_t> old_inodes_t;
std::string dn; // dentry
snapid_t dnfirst, dnlast;
version_t dnv{0};
- inode_t inode; // if it's not
+ CInode::mempool_inode inode; // if it's not XXX should not be part of mempool; wait for std::pmr to simplify
fragtree_t dirfragtree;
- map<string,bufferptr> xattrs;
- string symlink;
+ CInode::mempool_xattr_map xattrs;
+ std::string symlink;
snapid_t oldest_snap;
bufferlist snapbl;
__u8 state{0};
- old_inodes_t old_inodes;
+ CInode::mempool_old_inode_map old_inodes; // XXX should not be part of mempool; wait for std::pmr to simplify
fullbit(const fullbit& o);
const fullbit& operator=(const fullbit& o);
fullbit(std::string_view d, snapid_t df, snapid_t dl,
- version_t v, const inode_t& i, const fragtree_t &dft,
- const map<string,bufferptr> &xa, const string& sym,
+ version_t v, const CInode::mempool_inode& i, const fragtree_t &dft,
+ const CInode::mempool_xattr_map &xa, std::string_view sym,
snapid_t os, const bufferlist &sbl, __u8 st,
- const old_inodes_t *oi = NULL) :
+ const CInode::mempool_old_inode_map *oi = NULL) :
dn(d), dnfirst(df), dnlast(dl), dnv(v), inode(i), xattrs(xa),
oldest_snap(os), state(st)
{
@@ -446,7 +445,7 @@ private:
in->last_journaled = event_seq;
//cout << "journaling " << in->inode.ino << " at " << my_offset << std::endl;
- const inode_t *pi = in->get_projected_inode();
+ const auto pi = in->get_projected_inode();
if ((state & fullbit::STATE_DIRTY) && pi->is_backtrace_updated())
state |= fullbit::STATE_DIRTYPARENT;
@@ -494,8 +493,8 @@ private:
add_primary_dentry(dn, 0, dirty, dirty_parent, dirty_pool);
}
- void add_root(bool dirty, CInode *in, const inode_t *pi=0, fragtree_t *pdft=0, bufferlist *psnapbl=0,
- map<string,bufferptr> *px=0) {
+ void add_root(bool dirty, CInode *in, const CInode::mempool_inode *pi=0, fragtree_t *pdft=0, bufferlist *psnapbl=0,
+ CInode::mempool_xattr_map *px=0) {
in->last_journaled = event_seq;
//cout << "journaling " << in->inode.ino << " at " << my_offset << std::endl;
diff --git a/src/mds/journal.cc b/src/mds/journal.cc
index a0feaed3bec..aaefd373151 100644
--- a/src/mds/journal.cc
+++ b/src/mds/journal.cc
@@ -501,10 +501,9 @@ void EMetaBlob::fullbit::dump(Formatter *f) const
inode.dump(f);
f->close_section(); // inode
f->open_object_section("xattrs");
- for (map<string, bufferptr>::const_iterator iter = xattrs.begin();
- iter != xattrs.end(); ++iter) {
- string s(iter->second.c_str(), iter->second.length());
- f->dump_string(iter->first.c_str(), s);
+ for (const auto &p : xattrs) {
+ std::string s(p.second.c_str(), p.second.length());
+ f->dump_string(p.first.c_str(), s);
}
f->close_section(); // xattrs
if (inode.is_symlink()) {
@@ -523,12 +522,10 @@ void EMetaBlob::fullbit::dump(Formatter *f) const
f->dump_string("state", state_string());
if (!old_inodes.empty()) {
f->open_array_section("old inodes");
- for (old_inodes_t::const_iterator iter = old_inodes.begin();
- iter != old_inodes.end();
- ++iter) {
+ for (const auto &p : old_inodes) {
f->open_object_section("inode");
- f->dump_int("snapid", iter->first);
- iter->second.dump(f);
+ f->dump_int("snapid", p.first);
+ p.second.dump(f);
f->close_section(); // inode
}
f->close_section(); // old inodes
@@ -537,9 +534,9 @@ void EMetaBlob::fullbit::dump(Formatter *f) const
void EMetaBlob::fullbit::generate_test_instances(list<EMetaBlob::fullbit*>& ls)
{
- inode_t inode;
+ CInode::mempool_inode inode;
fragtree_t fragtree;
- map<string,bufferptr> empty_xattrs;
+ CInode::mempool_xattr_map empty_xattrs;
bufferlist empty_snapbl;
fullbit *sample = new fullbit("/testdn", 0, 0, 0,
inode, fragtree, empty_xattrs, "", 0, empty_snapbl,
diff --git a/src/mds/mdstypes.cc b/src/mds/mdstypes.cc
index 8658f1b6851..a641f2812bf 100644
--- a/src/mds/mdstypes.cc
+++ b/src/mds/mdstypes.cc
@@ -241,338 +241,6 @@ void inline_data_t::decode(bufferlist::iterator &p)
free_data();
}
-/*
- * inode_t
- */
-void inode_t::encode(bufferlist &bl, uint64_t features) const
-{
- ENCODE_START(15, 6, bl);
-
- encode(ino, bl);
- encode(rdev, bl);
- encode(ctime, bl);
-
- encode(mode, bl);
- encode(uid, bl);
- encode(gid, bl);
-
- encode(nlink, bl);
- {
- // removed field
- bool anchored = 0;
- encode(anchored, bl);
- }
-
- encode(dir_layout, bl);
- encode(layout, bl, features);
- encode(size, bl);
- encode(truncate_seq, bl);
- encode(truncate_size, bl);
- encode(truncate_from, bl);
- encode(truncate_pending, bl);
- encode(mtime, bl);
- encode(atime, bl);
- encode(time_warp_seq, bl);
- encode(client_ranges, bl);
-
- encode(dirstat, bl);
- encode(rstat, bl);
- encode(accounted_rstat, bl);
-
- encode(version, bl);
- encode(file_data_version, bl);
- encode(xattr_version, bl);
- encode(backtrace_version, bl);
- encode(old_pools, bl);
- encode(max_size_ever, bl);
- encode(inline_data, bl);
- encode(quota, bl);
-
- encode(stray_prior_path, bl);
-
- encode(last_scrub_version, bl);
- encode(last_scrub_stamp, bl);
-
- encode(btime, bl);
- encode(change_attr, bl);
-
- encode(export_pin, bl);
-
- ENCODE_FINISH(bl);
-}
-
-void inode_t::decode(bufferlist::iterator &p)
-{
- DECODE_START_LEGACY_COMPAT_LEN(15, 6, 6, p);
-
- decode(ino, p);
- decode(rdev, p);
- decode(ctime, p);
-
- decode(mode, p);
- decode(uid, p);
- decode(gid, p);
-
- decode(nlink, p);
- {
- bool anchored;
- decode(anchored, p);
- }
-
- if (struct_v >= 4)
- decode(dir_layout, p);
- else
- memset(&dir_layout, 0, sizeof(dir_layout));
- decode(layout, p);
- decode(size, p);
- decode(truncate_seq, p);
- decode(truncate_size, p);
- decode(truncate_from, p);
- if (struct_v >= 5)
- decode(truncate_pending, p);
- else
- truncate_pending = 0;
- decode(mtime, p);
- decode(atime, p);
- decode(time_warp_seq, p);
- if (struct_v >= 3) {
- decode(client_ranges, p);
- } else {
- map<client_t, client_writeable_range_t::byte_range_t> m;
- decode(m, p);
- for (map<client_t, client_writeable_range_t::byte_range_t>::iterator
- q = m.begin(); q != m.end(); ++q)
- client_ranges[q->first].range = q->second;
- }
-
- decode(dirstat, p);
- decode(rstat, p);
- decode(accounted_rstat, p);
-
- decode(version, p);
- decode(file_data_version, p);
- decode(xattr_version, p);
- if (struct_v >= 2)
- decode(backtrace_version, p);
- if (struct_v >= 7)
- decode(old_pools, p);
- if (struct_v >= 8)
- decode(max_size_ever, p);
- if (struct_v >= 9) {
- decode(inline_data, p);
- } else {
- inline_data.version = CEPH_INLINE_NONE;
- }
- if (struct_v < 10)
- backtrace_version = 0; // force update backtrace
- if (struct_v >= 11)
- decode(quota, p);
-
- if (struct_v >= 12) {
- decode(stray_prior_path, p);
- }
-
- if (struct_v >= 13) {
- decode(last_scrub_version, p);
- decode(last_scrub_stamp, p);
- }
- if (struct_v >= 14) {
- decode(btime, p);
- decode(change_attr, p);
- } else {
- btime = utime_t();
- change_attr = 0;
- }
-
- if (struct_v >= 15) {
- decode(export_pin, p);
- } else {
- export_pin = MDS_RANK_NONE;
- }
-
- DECODE_FINISH(p);
-}
-
-void inode_t::dump(Formatter *f) const
-{
- f->dump_unsigned("ino", ino);
- f->dump_unsigned("rdev", rdev);
- f->dump_stream("ctime") << ctime;
- f->dump_stream("btime") << btime;
- f->dump_unsigned("mode", mode);
- f->dump_unsigned("uid", uid);
- f->dump_unsigned("gid", gid);
- f->dump_unsigned("nlink", nlink);
-
- f->open_object_section("dir_layout");
- ::dump(dir_layout, f);
- f->close_section();
-
- f->dump_object("layout", layout);
-
- f->open_array_section("old_pools");
- for (compact_set<int64_t>::const_iterator i = old_pools.begin();
- i != old_pools.end();
- ++i)
- f->dump_int("pool", *i);
- f->close_section();
-
- f->dump_unsigned("size", size);
- f->dump_unsigned("truncate_seq", truncate_seq);
- f->dump_unsigned("truncate_size", truncate_size);
- f->dump_unsigned("truncate_from", truncate_from);
- f->dump_unsigned("truncate_pending", truncate_pending);
- f->dump_stream("mtime") << mtime;
- f->dump_stream("atime") << atime;
- f->dump_unsigned("time_warp_seq", time_warp_seq);
- f->dump_unsigned("change_attr", change_attr);
- f->dump_int("export_pin", export_pin);
-
- f->open_array_section("client_ranges");
- for (map<client_t,client_writeable_range_t>::const_iterator p = client_ranges.begin(); p != client_ranges.end(); ++p) {
- f->open_object_section("client");
- f->dump_unsigned("client", p->first.v);
- p->second.dump(f);
- f->close_section();
- }
- f->close_section();
-
- f->open_object_section("dirstat");
- dirstat.dump(f);
- f->close_section();
-
- f->open_object_section("rstat");
- rstat.dump(f);
- f->close_section();
-
- f->open_object_section("accounted_rstat");
- accounted_rstat.dump(f);
- f->close_section();
-
- f->dump_unsigned("version", version);
- f->dump_unsigned("file_data_version", file_data_version);
- f->dump_unsigned("xattr_version", xattr_version);
- f->dump_unsigned("backtrace_version", backtrace_version);
-
- f->dump_string("stray_prior_path", stray_prior_path);
-}
-
-void inode_t::generate_test_instances(list<inode_t*>& ls)
-{
- ls.push_back(new inode_t);
- ls.push_back(new inode_t);
- ls.back()->ino = 1;
- // i am lazy.
-}
-
-int inode_t::compare(const inode_t &other, bool *divergent) const
-{
- assert(ino == other.ino);
- *divergent = false;
- if (version == other.version) {
- if (rdev != other.rdev ||
- ctime != other.ctime ||
- btime != other.btime ||
- mode != other.mode ||
- uid != other.uid ||
- gid != other.gid ||
- nlink != other.nlink ||
- memcmp(&dir_layout, &other.dir_layout, sizeof(dir_layout)) ||
- layout != other.layout ||
- old_pools != other.old_pools ||
- size != other.size ||
- max_size_ever != other.max_size_ever ||
- truncate_seq != other.truncate_seq ||
- truncate_size != other.truncate_size ||
- truncate_from != other.truncate_from ||
- truncate_pending != other.truncate_pending ||
- change_attr != other.change_attr ||
- mtime != other.mtime ||
- atime != other.atime ||
- time_warp_seq != other.time_warp_seq ||
- inline_data != other.inline_data ||
- client_ranges != other.client_ranges ||
- !(dirstat == other.dirstat) ||
- !(rstat == other.rstat) ||
- !(accounted_rstat == other.accounted_rstat) ||
- file_data_version != other.file_data_version ||
- xattr_version != other.xattr_version ||
- backtrace_version != other.backtrace_version) {
- *divergent = true;
- }
- return 0;
- } else if (version > other.version) {
- *divergent = !older_is_consistent(other);
- return 1;
- } else {
- assert(version < other.version);
- *divergent = !other.older_is_consistent(*this);
- return -1;
- }
-}
-
-bool inode_t::older_is_consistent(const inode_t &other) const
-{
- if (max_size_ever < other.max_size_ever ||
- truncate_seq < other.truncate_seq ||
- time_warp_seq < other.time_warp_seq ||
- inline_data.version < other.inline_data.version ||
- dirstat.version < other.dirstat.version ||
- rstat.version < other.rstat.version ||
- accounted_rstat.version < other.accounted_rstat.version ||
- file_data_version < other.file_data_version ||
- xattr_version < other.xattr_version ||
- backtrace_version < other.backtrace_version) {
- return false;
- }
- return true;
-}
-
-/*
- * old_inode_t
- */
-void old_inode_t::encode(bufferlist& bl, uint64_t features) const
-{
- ENCODE_START(2, 2, bl);
- encode(first, bl);
- encode(inode, bl, features);
- encode(xattrs, bl);
- ENCODE_FINISH(bl);
-}
-
-void old_inode_t::decode(bufferlist::iterator& bl)
-{
- DECODE_START_LEGACY_COMPAT_LEN(2, 2, 2, bl);
- decode(first, bl);
- decode(inode, bl);
- decode(xattrs, bl);
- DECODE_FINISH(bl);
-}
-
-void old_inode_t::dump(Formatter *f) const
-{
- f->dump_unsigned("first", first);
- inode.dump(f);
- f->open_object_section("xattrs");
- for (map<string,bufferptr>::const_iterator p = xattrs.begin(); p != xattrs.end(); ++p) {
- string v(p->second.c_str(), p->second.length());
- f->dump_string(p->first.c_str(), v);
- }
- f->close_section();
-}
-
-void old_inode_t::generate_test_instances(list<old_inode_t*>& ls)
-{
- ls.push_back(new old_inode_t);
- ls.push_back(new old_inode_t);
- ls.back()->first = 2;
- list<inode_t*> ils;
- inode_t::generate_test_instances(ils);
- ls.back()->inode = *ils.back();
- ls.back()->xattrs["user.foo"] = buffer::copy("asdf", 4);
- ls.back()->xattrs["user.unprintable"] = buffer::copy("\000\001\002", 3);
-}
-
/*
* fnode_t
@@ -926,25 +594,27 @@ void mds_table_pending_t::generate_test_instances(list<mds_table_pending_t*>& ls
void inode_load_vec_t::encode(bufferlist &bl) const
{
ENCODE_START(2, 2, bl);
- for (int i=0; i<NUM; i++)
- encode(vec[i], bl);
+ for (const auto &i : vec) {
+ encode(i, bl);
+ }
ENCODE_FINISH(bl);
}
void inode_load_vec_t::decode(const utime_t &t, bufferlist::iterator &p)
{
DECODE_START_LEGACY_COMPAT_LEN(2, 2, 2, p);
- for (int i=0; i<NUM; i++)
- decode(vec[i], t, p);
+ for (auto &i : vec) {
+ decode(i, t, p);
+ }
DECODE_FINISH(p);
}
void inode_load_vec_t::dump(Formatter *f)
{
f->open_array_section("Decay Counters");
- for (vector<DecayCounter>::const_iterator i = vec.begin(); i != vec.end(); ++i) {
+ for (const auto &i : vec) {
f->open_object_section("Decay Counter");
- i->dump(f);
+ i.dump(f);
f->close_section();
}
f->close_section();
@@ -963,9 +633,9 @@ void inode_load_vec_t::generate_test_instances(list<inode_load_vec_t*>& ls)
void dirfrag_load_vec_t::dump(Formatter *f) const
{
f->open_array_section("Decay Counters");
- for (vector<DecayCounter>::const_iterator i = vec.begin(); i != vec.end(); ++i) {
+ for (const auto &i : vec) {
f->open_object_section("Decay Counter");
- i->dump(f);
+ i.dump(f);
f->close_section();
}
f->close_section();
diff --git a/src/mds/mdstypes.h b/src/mds/mdstypes.h
index 380cea18f42..7ddf2d1f13c 100644
--- a/src/mds/mdstypes.h
+++ b/src/mds/mdstypes.h
@@ -380,8 +380,8 @@ inline std::ostream& operator<<(std::ostream &out, const vinodeno_t &vino) {
*/
struct client_writeable_range_t {
struct byte_range_t {
- uint64_t first, last; // interval client can write to
- byte_range_t() : first(0), last(0) {}
+ uint64_t first = 0, last = 0; // interval client can write to
+ byte_range_t() {}
};
byte_range_t range;
@@ -392,7 +392,7 @@ struct client_writeable_range_t {
void encode(bufferlist &bl) const;
void decode(bufferlist::iterator& bl);
void dump(Formatter *f) const;
- static void generate_test_instances(list<client_writeable_range_t*>& ls);
+ static void generate_test_instances(std::list<client_writeable_range_t*>& ls);
};
inline void decode(client_writeable_range_t::byte_range_t& range, bufferlist::iterator& bl) {
@@ -462,6 +462,7 @@ typedef uint32_t damage_flags_t;
/*
* inode_t
*/
+template<template<typename> class Allocator = std::allocator>
struct inode_t {
/**
* ***************
@@ -487,7 +488,7 @@ struct inode_t {
// file (data access)
ceph_dir_layout dir_layout; // [dir only]
file_layout_t layout;
- compact_set <int64_t> old_pools;
+ compact_set<int64_t, std::less<int64_t>, Allocator<int64_t>> old_pools;
uint64_t size = 0; // on directory, # dentries
uint64_t max_size_ever = 0; // max size the file has ever been
uint32_t truncate_seq = 0;
@@ -496,12 +497,13 @@ struct inode_t {
utime_t mtime; // file data modify time.
utime_t atime; // file data access time.
uint32_t time_warp_seq = 0; // count of (potential) mtime/atime timewarps (i.e., utimes())
- inline_data_t inline_data;
+ inline_data_t inline_data; // FIXME check
// change attribute
uint64_t change_attr = 0;
- std::map<client_t,client_writeable_range_t> client_ranges; // client(s) can write to these ranges
+ using client_range_map = std::map<client_t,client_writeable_range_t,std::less<client_t>,Allocator<std::pair<const client_t,client_writeable_range_t>>>;
+ client_range_map client_ranges; // client(s) can write to these ranges
// dirfrag, recursive accountin
frag_info_t dirstat; // protected by my filelock
@@ -524,7 +526,7 @@ struct inode_t {
snapid_t oldest_snap;
- string stray_prior_path; //stores path before unlink
+ std::basic_string<char,std::char_traits<char>,Allocator<char>> stray_prior_path; //stores path before unlink
inode_t()
{
@@ -610,7 +612,7 @@ struct inode_t {
void encode(bufferlist &bl, uint64_t features) const;
void decode(bufferlist::iterator& bl);
void dump(Formatter *f) const;
- static void generate_test_instances(list<inode_t*>& ls);
+ static void generate_test_instances(std::list<inode_t*>& ls);
/**
* Compare this inode_t with another that represent *the same inode*
* at different points in time.
@@ -627,23 +629,392 @@ struct inode_t {
private:
bool older_is_consistent(const inode_t &other) const;
};
-WRITE_CLASS_ENCODER_FEATURES(inode_t)
+// These methods may be moved back to mdstypes.cc when we have pmr
+template<template<typename> class Allocator>
+void inode_t<Allocator>::encode(bufferlist &bl, uint64_t features) const
+{
+ ENCODE_START(15, 6, bl);
+
+ encode(ino, bl);
+ encode(rdev, bl);
+ encode(ctime, bl);
+
+ encode(mode, bl);
+ encode(uid, bl);
+ encode(gid, bl);
+
+ encode(nlink, bl);
+ {
+ // removed field
+ bool anchored = 0;
+ encode(anchored, bl);
+ }
+
+ encode(dir_layout, bl);
+ encode(layout, bl, features);
+ encode(size, bl);
+ encode(truncate_seq, bl);
+ encode(truncate_size, bl);
+ encode(truncate_from, bl);
+ encode(truncate_pending, bl);
+ encode(mtime, bl);
+ encode(atime, bl);
+ encode(time_warp_seq, bl);
+ encode(client_ranges, bl);
+
+ encode(dirstat, bl);
+ encode(rstat, bl);
+ encode(accounted_rstat, bl);
+
+ encode(version, bl);
+ encode(file_data_version, bl);
+ encode(xattr_version, bl);
+ encode(backtrace_version, bl);
+ encode(old_pools, bl);
+ encode(max_size_ever, bl);
+ encode(inline_data, bl);
+ encode(quota, bl);
+
+ encode(stray_prior_path, bl);
+
+ encode(last_scrub_version, bl);
+ encode(last_scrub_stamp, bl);
+
+ encode(btime, bl);
+ encode(change_attr, bl);
+
+ encode(export_pin, bl);
+
+ ENCODE_FINISH(bl);
+}
+
+template<template<typename> class Allocator>
+void inode_t<Allocator>::decode(bufferlist::iterator &p)
+{
+ DECODE_START_LEGACY_COMPAT_LEN(15, 6, 6, p);
+
+ decode(ino, p);
+ decode(rdev, p);
+ decode(ctime, p);
+
+ decode(mode, p);
+ decode(uid, p);
+ decode(gid, p);
+
+ decode(nlink, p);
+ {
+ bool anchored;
+ decode(anchored, p);
+ }
+
+ if (struct_v >= 4)
+ decode(dir_layout, p);
+ else
+ memset(&dir_layout, 0, sizeof(dir_layout));
+ decode(layout, p);
+ decode(size, p);
+ decode(truncate_seq, p);
+ decode(truncate_size, p);
+ decode(truncate_from, p);
+ if (struct_v >= 5)
+ decode(truncate_pending, p);
+ else
+ truncate_pending = 0;
+ decode(mtime, p);
+ decode(atime, p);
+ decode(time_warp_seq, p);
+ if (struct_v >= 3) {
+ decode(client_ranges, p);
+ } else {
+ map<client_t, client_writeable_range_t::byte_range_t> m;
+ decode(m, p);
+ for (map<client_t, client_writeable_range_t::byte_range_t>::iterator
+ q = m.begin(); q != m.end(); ++q)
+ client_ranges[q->first].range = q->second;
+ }
+
+ decode(dirstat, p);
+ decode(rstat, p);
+ decode(accounted_rstat, p);
+
+ decode(version, p);
+ decode(file_data_version, p);
+ decode(xattr_version, p);
+ if (struct_v >= 2)
+ decode(backtrace_version, p);
+ if (struct_v >= 7)
+ decode(old_pools, p);
+ if (struct_v >= 8)
+ decode(max_size_ever, p);
+ if (struct_v >= 9) {
+ decode(inline_data, p);
+ } else {
+ inline_data.version = CEPH_INLINE_NONE;
+ }
+ if (struct_v < 10)
+ backtrace_version = 0; // force update backtrace
+ if (struct_v >= 11)
+ decode(quota, p);
+
+ if (struct_v >= 12) {
+ std::string tmp;
+ decode(tmp, p);
+ stray_prior_path = std::string_view(tmp);
+ }
+
+ if (struct_v >= 13) {
+ decode(last_scrub_version, p);
+ decode(last_scrub_stamp, p);
+ }
+ if (struct_v >= 14) {
+ decode(btime, p);
+ decode(change_attr, p);
+ } else {
+ btime = utime_t();
+ change_attr = 0;
+ }
+
+ if (struct_v >= 15) {
+ decode(export_pin, p);
+ } else {
+ export_pin = MDS_RANK_NONE;
+ }
+
+ DECODE_FINISH(p);
+}
+
+template<template<typename> class Allocator>
+void inode_t<Allocator>::dump(Formatter *f) const
+{
+ f->dump_unsigned("ino", ino);
+ f->dump_unsigned("rdev", rdev);
+ f->dump_stream("ctime") << ctime;
+ f->dump_stream("btime") << btime;
+ f->dump_unsigned("mode", mode);
+ f->dump_unsigned("uid", uid);
+ f->dump_unsigned("gid", gid);
+ f->dump_unsigned("nlink", nlink);
+
+ f->open_object_section("dir_layout");
+ ::dump(dir_layout, f);
+ f->close_section();
+
+ f->dump_object("layout", layout);
+
+ f->open_array_section("old_pools");
+ for (const auto &p : old_pools) {
+ f->dump_int("pool", p);
+ }
+ f->close_section();
+
+ f->dump_unsigned("size", size);
+ f->dump_unsigned("truncate_seq", truncate_seq);
+ f->dump_unsigned("truncate_size", truncate_size);
+ f->dump_unsigned("truncate_from", truncate_from);
+ f->dump_unsigned("truncate_pending", truncate_pending);
+ f->dump_stream("mtime") << mtime;
+ f->dump_stream("atime") << atime;
+ f->dump_unsigned("time_warp_seq", time_warp_seq);
+ f->dump_unsigned("change_attr", change_attr);
+ f->dump_int("export_pin", export_pin);
+
+ f->open_array_section("client_ranges");
+ for (const auto &p : client_ranges) {
+ f->open_object_section("client");
+ f->dump_unsigned("client", p.first.v);
+ p.second.dump(f);
+ f->close_section();
+ }
+ f->close_section();
+
+ f->open_object_section("dirstat");
+ dirstat.dump(f);
+ f->close_section();
+
+ f->open_object_section("rstat");
+ rstat.dump(f);
+ f->close_section();
+
+ f->open_object_section("accounted_rstat");
+ accounted_rstat.dump(f);
+ f->close_section();
+
+ f->dump_unsigned("version", version);
+ f->dump_unsigned("file_data_version", file_data_version);
+ f->dump_unsigned("xattr_version", xattr_version);
+ f->dump_unsigned("backtrace_version", backtrace_version);
+
+ f->dump_string("stray_prior_path", stray_prior_path);
+}
+
+template<template<typename> class Allocator>
+void inode_t<Allocator>::generate_test_instances(list<inode_t*>& ls)
+{
+ ls.push_back(new inode_t<Allocator>);
+ ls.push_back(new inode_t<Allocator>);
+ ls.back()->ino = 1;
+ // i am lazy.
+}
+
+template<template<typename> class Allocator>
+int inode_t<Allocator>::compare(const inode_t<Allocator> &other, bool *divergent) const
+{
+ assert(ino == other.ino);
+ *divergent = false;
+ if (version == other.version) {
+ if (rdev != other.rdev ||
+ ctime != other.ctime ||
+ btime != other.btime ||
+ mode != other.mode ||
+ uid != other.uid ||
+ gid != other.gid ||
+ nlink != other.nlink ||
+ memcmp(&dir_layout, &other.dir_layout, sizeof(dir_layout)) ||
+ layout != other.layout ||
+ old_pools != other.old_pools ||
+ size != other.size ||
+ max_size_ever != other.max_size_ever ||
+ truncate_seq != other.truncate_seq ||
+ truncate_size != other.truncate_size ||
+ truncate_from != other.truncate_from ||
+ truncate_pending != other.truncate_pending ||
+ change_attr != other.change_attr ||
+ mtime != other.mtime ||
+ atime != other.atime ||
+ time_warp_seq != other.time_warp_seq ||
+ inline_data != other.inline_data ||
+ client_ranges != other.client_ranges ||
+ !(dirstat == other.dirstat) ||
+ !(rstat == other.rstat) ||
+ !(accounted_rstat == other.accounted_rstat) ||
+ file_data_version != other.file_data_version ||
+ xattr_version != other.xattr_version ||
+ backtrace_version != other.backtrace_version) {
+ *divergent = true;
+ }
+ return 0;
+ } else if (version > other.version) {
+ *divergent = !older_is_consistent(other);
+ return 1;
+ } else {
+ assert(version < other.version);
+ *divergent = !other.older_is_consistent(*this);
+ return -1;
+ }
+}
+
+template<template<typename> class Allocator>
+bool inode_t<Allocator>::older_is_consistent(const inode_t<Allocator> &other) const
+{
+ if (max_size_ever < other.max_size_ever ||
+ truncate_seq < other.truncate_seq ||
+ time_warp_seq < other.time_warp_seq ||
+ inline_data.version < other.inline_data.version ||
+ dirstat.version < other.dirstat.version ||
+ rstat.version < other.rstat.version ||
+ accounted_rstat.version < other.accounted_rstat.version ||
+ file_data_version < other.file_data_version ||
+ xattr_version < other.xattr_version ||
+ backtrace_version < other.backtrace_version) {
+ return false;
+ }
+ return true;
+}
+
+template<template<typename> class Allocator>
+inline void encode(const inode_t<Allocator> &c, ::ceph::bufferlist &bl, uint64_t features)
+{
+ ENCODE_DUMP_PRE();
+ c.encode(bl, features);
+ ENCODE_DUMP_POST(cl);
+}
+template<template<typename> class Allocator>
+inline void decode(inode_t<Allocator> &c, ::ceph::bufferlist::iterator &p)
+{
+ c.decode(p);
+}
+
+template<template<typename> class Allocator>
+using alloc_string = std::basic_string<char,std::char_traits<char>,Allocator<char>>;
+
+template<template<typename> class Allocator>
+using xattr_map = compact_map<alloc_string<Allocator>, bufferptr, std::less<alloc_string<Allocator>>, Allocator<std::pair<const alloc_string<Allocator>, bufferptr>>>; // FIXME bufferptr not in mempool
/*
* old_inode_t
*/
+template<template<typename> class Allocator = std::allocator>
struct old_inode_t {
snapid_t first;
- inode_t inode;
- std::map<string,bufferptr> xattrs;
+ inode_t<Allocator> inode;
+ xattr_map<Allocator> xattrs;
void encode(bufferlist &bl, uint64_t features) const;
void decode(bufferlist::iterator& bl);
void dump(Formatter *f) const;
- static void generate_test_instances(list<old_inode_t*>& ls);
+ static void generate_test_instances(std::list<old_inode_t*>& ls);
};
-WRITE_CLASS_ENCODER_FEATURES(old_inode_t)
+
+// These methods may be moved back to mdstypes.cc when we have pmr
+template<template<typename> class Allocator>
+void old_inode_t<Allocator>::encode(bufferlist& bl, uint64_t features) const
+{
+ ENCODE_START(2, 2, bl);
+ encode(first, bl);
+ encode(inode, bl, features);
+ encode(xattrs, bl);
+ ENCODE_FINISH(bl);
+}
+
+template<template<typename> class Allocator>
+void old_inode_t<Allocator>::decode(bufferlist::iterator& bl)
+{
+ DECODE_START_LEGACY_COMPAT_LEN(2, 2, 2, bl);
+ decode(first, bl);
+ decode(inode, bl);
+ decode(xattrs, bl);
+ DECODE_FINISH(bl);
+}
+
+template<template<typename> class Allocator>
+void old_inode_t<Allocator>::dump(Formatter *f) const
+{
+ f->dump_unsigned("first", first);
+ inode.dump(f);
+ f->open_object_section("xattrs");
+ for (const auto &p : xattrs) {
+ std::string v(p.second.c_str(), p.second.length());
+ f->dump_string(p.first.c_str(), v);
+ }
+ f->close_section();
+}
+
+template<template<typename> class Allocator>
+void old_inode_t<Allocator>::generate_test_instances(std::list<old_inode_t<Allocator>*>& ls)
+{
+ ls.push_back(new old_inode_t<Allocator>);
+ ls.push_back(new old_inode_t<Allocator>);
+ ls.back()->first = 2;
+ std::list<inode_t<Allocator>*> ils;
+ inode_t<Allocator>::generate_test_instances(ils);
+ ls.back()->inode = *ils.back();
+ ls.back()->xattrs["user.foo"] = buffer::copy("asdf", 4);
+ ls.back()->xattrs["user.unprintable"] = buffer::copy("\000\001\002", 3);
+}
+
+template<template<typename> class Allocator>
+inline void encode(const old_inode_t<Allocator> &c, ::ceph::bufferlist &bl, uint64_t features)
+{
+ ENCODE_DUMP_PRE();
+ c.encode(bl, features);
+ ENCODE_DUMP_POST(cl);
+}
+template<template<typename> class Allocator>
+inline void decode(old_inode_t<Allocator> &c, ::ceph::bufferlist::iterator &p)
+{
+ c.decode(p);
+}
/*
@@ -1037,15 +1408,13 @@ namespace std {
class inode_load_vec_t {
static const int NUM = 2;
- std::vector < DecayCounter > vec;
+ std::array<DecayCounter, NUM> vec;
public:
explicit inode_load_vec_t(const utime_t &now)
- : vec(NUM, DecayCounter(now))
+ : vec{DecayCounter(now), DecayCounter(now)}
{}
// for dencoder infrastructure
- inode_load_vec_t() :
- vec(NUM, DecayCounter())
- {}
+ inode_load_vec_t() {}
DecayCounter &get(int t) {
assert(t < NUM);
return vec[t];
@@ -1074,24 +1443,30 @@ inline void decode(inode_load_vec_t & c, bufferlist::iterator &p) {
class dirfrag_load_vec_t {
public:
static const int NUM = 5;
- std::vector < DecayCounter > vec;
+ std::array<DecayCounter, NUM> vec;
explicit dirfrag_load_vec_t(const utime_t &now)
- : vec(NUM, DecayCounter(now))
- { }
- // for dencoder infrastructure
- dirfrag_load_vec_t()
- : vec(NUM, DecayCounter())
+ : vec{
+ DecayCounter(now),
+ DecayCounter(now),
+ DecayCounter(now),
+ DecayCounter(now),
+ DecayCounter(now)
+ }
{}
+ // for dencoder infrastructure
+ dirfrag_load_vec_t() {}
void encode(bufferlist &bl) const {
ENCODE_START(2, 2, bl);
- for (int i=0; i<NUM; i++)
- encode(vec[i], bl);
+ for (const auto &i : vec) {
+ encode(i, bl);
+ }
ENCODE_FINISH(bl);
}
void decode(const utime_t &t, bufferlist::iterator &p) {
DECODE_START_LEGACY_COMPAT_LEN(2, 2, 2, p);
- for (int i=0; i<NUM; i++)
- decode(vec[i], t, p);
+ for (auto &i : vec) {
+ decode(i, t, p);
+ }
DECODE_FINISH(p);
}
// for dencoder infrastructure
@@ -1108,12 +1483,14 @@ public:
return vec[t];
}
void adjust(utime_t now, const DecayRate& rate, double d) {
- for (int i=0; i<NUM; i++)
- vec[i].adjust(now, rate, d);
+ for (auto &i : vec) {
+ i.adjust(now, rate, d);
+ }
}
void zero(utime_t now) {
- for (int i=0; i<NUM; i++)
- vec[i].reset(now);
+ for (auto &i : vec) {
+ i.reset(now);
+ }
}
double meta_load(utime_t now, const DecayRate& rate) {
return
diff --git a/src/test/encoding/types.h b/src/test/encoding/types.h
index 9748345ffc6..d725b199f0a 100644
--- a/src/test/encoding/types.h
+++ b/src/test/encoding/types.h
@@ -188,8 +188,8 @@ TYPE(sr_t)
TYPE(frag_info_t)
TYPE(nest_info_t)
TYPE(client_writeable_range_t)
-TYPE_FEATUREFUL(inode_t)
-TYPE_FEATUREFUL(old_inode_t)
+TYPE_FEATUREFUL(inode_t<std::allocator>)
+TYPE_FEATUREFUL(old_inode_t<std::allocator>)
TYPE(fnode_t)
TYPE(old_rstat_t)
TYPE_FEATUREFUL(session_info_t)
diff --git a/src/tools/cephfs/DataScan.cc b/src/tools/cephfs/DataScan.cc
index 0d3e3ace724..f8bf524097a 100644
--- a/src/tools/cephfs/DataScan.cc
+++ b/src/tools/cephfs/DataScan.cc
@@ -909,7 +909,7 @@ int DataScan::scan_links()
int nlink;
bool is_dir;
link_info_t() : version(0), nlink(0), is_dir(false) {}
- link_info_t(inodeno_t di, frag_t df, const string& n, const inode_t i) :
+ link_info_t(inodeno_t di, frag_t df, const string& n, const CInode::mempool_inode& i) :
dirino(di), frag(df), name(n),
version(i.version), nlink(i.nlink), is_dir(S_IFDIR & i.mode) {}
dirfrag_t dirfrag() const {
generated by cgit v1.2.3 (git 2.39.1) at 2025-01-19 11:43:23 +0100