mds: add vxattr to block quiesce on an inode

Signed-off-by: Patrick Donnelly <pdonnell@redhat.com>
(cherry picked from commit e2529fc74808b102b1ca85a3ee6644160da32e60)

diff --git a/src/include/cephfs/types.h b/src/include/cephfs/types.h
index 5ab5c229a063b5819a2f4473697d82582706cd76..108878794f755074ceb73f9ec678ce21fa90e06c 100644 (file)
--- a/src/include/cephfs/types.h
+++ b/src/include/cephfs/types.h
@@ -395,6 +395,7 @@ struct inode_t {
   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>>>;
 
   static const uint8_t F_EPHEMERAL_DISTRIBUTED_PIN = (1<<0);
+  static const uint8_t F_QUIESCE_BLOCK             = (1<<1);
 
   inode_t()
   {
@@ -500,6 +501,12 @@ struct inode_t {
   bool get_ephemeral_distributed_pin() const {
     return get_flag(F_EPHEMERAL_DISTRIBUTED_PIN);
   }
+  void set_quiesce_block(bool v) {
+    set_flag(v, F_QUIESCE_BLOCK);
+  }
+  bool get_quiesce_block() const {
+    return get_flag(F_QUIESCE_BLOCK);
+  }
 
   void encode(ceph::buffer::list &bl, uint64_t features) const;
   void decode(ceph::buffer::list::const_iterator& bl);
@@ -828,6 +835,7 @@ void inode_t<Allocator>::dump(ceph::Formatter *f) const
   f->dump_int("export_pin", export_pin);
   f->dump_int("export_ephemeral_random_pin", export_ephemeral_random_pin);
   f->dump_bool("export_ephemeral_distributed_pin", get_ephemeral_distributed_pin());
+  f->dump_bool("quiesce_block", get_quiesce_block());
 
   f->open_array_section("client_ranges");
   for (const auto &p : client_ranges) {

diff --git a/src/mds/CInode.cc b/src/mds/CInode.cc
index 8225e4fbb5a7fa6572405fb6a9b8030525e965b9..ed9a74f4435b456a6cd6478c6023270c7b8021f4 100644 (file)
--- a/src/mds/CInode.cc
+++ b/src/mds/CInode.cc
@@ -310,6 +310,9 @@ ostream& operator<<(ostream& out, const CInode& in)
   if (in.state_test(CInode::STATE_RANDEPHEMERALPIN)) {
     out << " randepin";
   }
+  if (in.get_inode()->get_quiesce_block()) {
+    out << " qblock";
+  }
 
   out << " " << &in;
   out << "]";

diff --git a/src/mds/MDCache.cc b/src/mds/MDCache.cc
index 8a3dd21d57be561c7fe2584acfad858b7f084fa4..a6df735106922661f3d83a5393f77d360d9d5efe 100644 (file)
--- a/src/mds/MDCache.cc
+++ b/src/mds/MDCache.cc
@@ -13614,6 +13614,17 @@ void MDCache::dispatch_quiesce_inode(const MDRequestRef& mdr)
     return;
   }
 
+  if (in->get_projected_inode()->get_quiesce_block()) {
+    dout(10) << __func__ << " quiesce is blocked for this inode; dropping locks!" << dendl;
+    mdr->mark_event("quiesce blocked");
+    mds->locker->drop_locks(mdr.get());
+    /* keep authpins! */
+    qs.inc_inodes_blocked();
+    mdr->internal_op_finish->complete(0);
+    mdr->internal_op_finish = nullptr;
+    return;
+  }
+
   if (in->is_dir()) {
     for (auto& dir : in->get_dirfrags()) {
       if (!dir->is_auth() && !splitauth) {

diff --git a/src/mds/MDCache.h b/src/mds/MDCache.h
index ab34d238eafb2cd8b317cead3f5f9ef293ead516..92a6be6d2a08b4361c70eed69da4b305d6ae66e7 100644 (file)
--- a/src/mds/MDCache.h
+++ b/src/mds/MDCache.h
@@ -540,12 +540,18 @@ public:
     uint64_t inc_heartbeat_count() {
       return ++heartbeat_count;
     }
+    void inc_inodes_blocked() {
+      inodes_blocked++;
+    }
     uint64_t get_inodes() const {
       return inodes;
     }
     uint64_t get_inodes_quiesced() const {
       return inodes_quiesced;
     }
+    uint64_t get_inodes_blocked() const {
+      return inodes_blocked;
+    }
     void add_failed(const MDRequestRef& mdr, int rc) {
       failed[mdr] = rc;
     }
@@ -559,6 +565,7 @@ public:
     void dump(Formatter* f) const {
       f->dump_unsigned("inodes", inodes);
       f->dump_unsigned("inodes_quiesced", inodes_quiesced);
+      f->dump_unsigned("inodes_blocked", inodes_blocked);
       f->open_array_section("failed");
       for (auto& [mdr, rc] : failed) {
         f->open_object_section("failure");
@@ -572,6 +579,7 @@ private:
     uint64_t heartbeat_count = 0;
     uint64_t inodes = 0;
     uint64_t inodes_quiesced = 0;
+    uint64_t inodes_blocked = 0;
     std::map<MDRequestRef, int> failed;
   };
   class C_MDS_QuiescePath : public MDSInternalContext {

diff --git a/src/mds/Server.cc b/src/mds/Server.cc
index 3bc4f9196b6df9d911a4d2f9ab6ca8752281cea8..04330cd308d1cd60ac70b8079240e0ca9f8ff4da 100644 (file)
--- a/src/mds/Server.cc
+++ b/src/mds/Server.cc
@@ -6175,6 +6175,57 @@ void Server::handle_set_vxattr(const MDRequestRef& mdr, CInode *cur)
 
     client_t exclude_ct = mdr->get_client();
     mdcache->broadcast_quota_to_client(cur, exclude_ct, true);
+  } else if (name == "ceph.quiesce.block"sv) {
+    bool val;
+    try {
+      val = boost::lexical_cast<bool>(value);
+    } catch (boost::bad_lexical_cast const&) {
+      dout(10) << "bad vxattr value, unable to parse bool for " << name << dendl;
+      respond_to_request(mdr, -CEPHFS_EINVAL);
+      return;
+    }
+
+    /* Verify it's not already marked with lighter weight
+     * rdlock.
+     */
+    if (!mdr->more()->rdonly_checks) {
+      if (!(mdr->locking_state & MutationImpl::ALL_LOCKED)) {
+        lov.add_rdlock(&cur->policylock);
+        if (!mds->locker->acquire_locks(mdr, lov))
+          return;
+        mdr->locking_state |= MutationImpl::ALL_LOCKED;
+      }
+      bool is_blocked = cur->get_projected_inode()->get_quiesce_block();
+      if (is_blocked == val) {
+        dout(20) << "already F_QUIESCE_BLOCK set" << dendl;
+        respond_to_request(mdr, 0);
+        return;
+      }
+      mdr->more()->rdonly_checks = true;
+    }
+
+    if ((mdr->locking_state & MutationImpl::ALL_LOCKED) && !mdr->is_xlocked(&cur->policylock)) {
+      /* drop the rdlock and acquire xlocks */
+      dout(20) << "dropping rdlocks" << dendl;
+      mds->locker->drop_locks(mdr.get());
+      if (!xlock_policylock(mdr, cur, false, true))
+        return;
+    }
+
+    /* repeat rdonly checks in case changed between rdlock -> xlock */
+    bool is_blocked = cur->get_projected_inode()->get_quiesce_block();
+    if (is_blocked == val) {
+      dout(20) << "already F_QUIESCE_BLOCK set" << dendl;
+      respond_to_request(mdr, 0);
+      return;
+    }
+
+    auto pi = cur->project_inode(mdr);
+    pi.inode->set_quiesce_block(val);
+    dout(20) << (val ? "setting" : "unsetting") << " F_QUIESCE_BLOCK on ino: " << cur->ino() << dendl;
+
+    mdr->no_early_reply = true;
+    pip = pi.inode.get();
   } else if (name == "ceph.dir.subvolume"sv) {
     if (!cur->is_dir()) {
       respond_to_request(mdr, -CEPHFS_EINVAL);
@@ -6879,6 +6930,8 @@ void Server::handle_client_getvxattr(const MDRequestRef& mdr)
     } else {
       r = -CEPHFS_ENODATA; // no such attribute
     }
+  } else if (xattr_name == "ceph.quiesce.block"sv) {
+    *css << cur->get_projected_inode()->get_quiesce_block();
   } else if (xattr_name.substr(0, 12) == "ceph.dir.pin"sv) {
     if (xattr_name == "ceph.dir.pin"sv) {
       *css << cur->get_projected_inode()->export_pin;

diff --git a/src/mds/Server.h b/src/mds/Server.h
index 63d05ea32abbba2990102961af33e8fea2908a2d..152ab1b701992a52801b1fbf4b227c9478e96926 100644 (file)
--- a/src/mds/Server.h
+++ b/src/mds/Server.h
@@ -17,6 +17,8 @@
 
 #include <string_view>
 
+using namespace std::literals::string_view_literals;
+
 #include <common/DecayCounter.h>
 
 #include "include/common_fwd.h"
@@ -441,6 +443,7 @@ private:
     return xattr_name.rfind("ceph.dir.layout", 0) == 0 ||
            xattr_name.rfind("ceph.file.layout", 0) == 0 ||
            xattr_name.rfind("ceph.quota", 0) == 0 ||
+           xattr_name == "ceph.quiesce.block"sv ||
            xattr_name == "ceph.dir.subvolume" ||
            xattr_name == "ceph.dir.pin" ||
            xattr_name == "ceph.dir.pin.random" ||