From 7dcf8221160474b703e649db11738802cd4747e2 Mon Sep 17 00:00:00 2001
From: Igor Fedotov <ifedotov@suse.com>
Date: Mon, 28 Jan 2019 18:17:52 +0300
Subject: [PATCH] os/bluestore: cap reads issued during deep fsck

Fixes: https://tracker.ceph.com/issues/38065
Signed-off-by: Igor Fedotov <ifedotov@suse.com>
(cherry picked from commit ced308000aef6050ca1fdbdfe3b4b63a887a1706)

Conflicts:
	src/common/options.cc
- use TYPE_UINT instead of TYPE_SIZE for new bluestore_fsck_read_bytes_cap
  option
- drop .set_flag(Option::FLAG_RUNTIME) because no .set_flag in luminous
---
 src/common/legacy_config_opts.h |  1 +
 src/common/options.cc           |  4 ++++
 src/os/bluestore/BlueStore.cc   | 24 +++++++++++++++++-------
 3 files changed, 22 insertions(+), 7 deletions(-)

diff --git a/src/common/legacy_config_opts.h b/src/common/legacy_config_opts.h
index 828697758124..5a36e6bd49bc 100644
--- a/src/common/legacy_config_opts.h
+++ b/src/common/legacy_config_opts.h
@@ -1099,6 +1099,7 @@ OPTION(bluestore_fsck_on_umount_deep, OPT_BOOL)
 OPTION(bluestore_fsck_on_mkfs, OPT_BOOL)
 OPTION(bluestore_fsck_on_mkfs_deep, OPT_BOOL)
 OPTION(bluestore_sync_submit_transaction, OPT_BOOL) // submit kv txn in queueing thread (not kv_sync_thread)
+OPTION(bluestore_fsck_read_bytes_cap, OPT_U64)
 OPTION(bluestore_throttle_bytes, OPT_U64)
 OPTION(bluestore_throttle_deferred_bytes, OPT_U64)
 OPTION(bluestore_throttle_cost_per_io_hdd, OPT_U64)
diff --git a/src/common/options.cc b/src/common/options.cc
index 1c474f3047a1..fc759287d2a3 100644
--- a/src/common/options.cc
+++ b/src/common/options.cc
@@ -3691,6 +3691,10 @@ std::vector<Option> get_global_options() {
     .set_default(false)
     .set_description("Try to submit metadata transaction to rocksdb in queuing thread context"),
 
+    Option("bluestore_fsck_read_bytes_cap", Option::TYPE_UINT, Option::LEVEL_ADVANCED)
+    .set_default(64_M)
+    .set_description("Maximum bytes read at once by deep fsck"),
+
     Option("bluestore_throttle_bytes", Option::TYPE_UINT, Option::LEVEL_ADVANCED)
     .set_default(64_M)
     .set_safe()
diff --git a/src/os/bluestore/BlueStore.cc b/src/os/bluestore/BlueStore.cc
index b957428ea8f1..dda28e0cfac6 100644
--- a/src/os/bluestore/BlueStore.cc
+++ b/src/os/bluestore/BlueStore.cc
@@ -6184,13 +6184,23 @@ int BlueStore::_fsck(bool deep, bool repair)
       }
       if (deep) {
 	bufferlist bl;
-	int r = _do_read(c.get(), o, 0, o->onode.size, bl,
-	  CEPH_OSD_OP_FLAG_FADVISE_NOCACHE);
-	if (r < 0) {
-	  ++errors;
-	  derr << "fsck error: " << oid << " error during read: "
-	       << cpp_strerror(r) << dendl;
-	}
+	uint64_t max_read_block = cct->_conf->bluestore_fsck_read_bytes_cap;
+	uint64_t offset = 0;
+	do {
+	  uint64_t l = std::min(uint64_t(o->onode.size - offset), max_read_block);
+	  int r = _do_read(c.get(), o, offset, l, bl,
+	    CEPH_OSD_OP_FLAG_FADVISE_NOCACHE);
+	  if (r < 0) {
+	    ++errors;
+	    derr << "fsck error: " << oid << std::hex
+	         << " error during read: "
+		 << " " << offset << "~" << l
+		 << " " << cpp_strerror(r) << std::dec
+		 << dendl;
+	    break;
+	  }
+	  offset += l;
+	} while (offset < o->onode.size);
       }
       // omap
       if (o->onode.has_omap()) {
-- 
2.47.3