From 7ab8752705115055cb32019cea920527e8006a4f Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Pavol=20=C5=BD=C3=A1=C4=8Dik?= <zacik.pa@gmail.com>
Date: Tue, 17 Sep 2024 16:17:50 +0200
Subject: [PATCH 1/3] vm: Add support for dirty_(bytes|ratio) sysctl parameters

And for dirty_background_(bytes|ratio).

These parameters interact with each other; it is therefore
not sufficient to configure them via the sysctl plugin.

For more details, see
https://docs.kernel.org/admin-guide/sysctl/vm.html#dirty-bytes.

Resolves: RHEL-58820
---
 tuned/plugins/plugin_vm.py | 96 ++++++++++++++++++++++++++++++++++++++
 tuned/utils/commands.py    |  3 ++
 2 files changed, 99 insertions(+)

diff --git a/tuned/plugins/plugin_vm.py b/tuned/plugins/plugin_vm.py
index b8b7e918..a2f3ae75 100644
--- a/tuned/plugins/plugin_vm.py
+++ b/tuned/plugins/plugin_vm.py
@@ -38,11 +38,26 @@ def _get_config_options(self):
 			"transparent_hugepages" : None,
 			"transparent_hugepage" : None,
 			"transparent_hugepage.defrag" : None,
+			"dirty_bytes" : None,
+			"dirty_ratio" : None,
+			"dirty_background_bytes" : None,
+			"dirty_background_ratio" : None
 		}
 
+	@staticmethod
+	def _check_conflicting_dirty_options(instance, first, second):
+		if instance.options[first] is not None and instance.options[second] is not None:
+			log.error("Conflicting options '%s' and '%s', this may cause undefined behavior." % (first, second))
+
+	@staticmethod
+	def _proc_sys_vm_option_path(option):
+		return os.path.join("/proc/sys/vm", option)
+
 	def _instance_init(self, instance):
 		instance._has_static_tuning = True
 		instance._has_dynamic_tuning = False
+		self._check_conflicting_dirty_options(instance, "dirty_bytes", "dirty_ratio")
+		self._check_conflicting_dirty_options(instance, "dirty_background_bytes", "dirty_background_ratio")
 
 	def _instance_cleanup(self, instance):
 		pass
@@ -117,3 +132,84 @@ def _get_transparent_hugepage_defrag(self):
 			return cmd.get_active_option(cmd.read_file(sys_file))
 		else:
 			return None
+
+	def _check_twice_pagesize(self, option, int_value):
+		min_bytes = 2 * int(cmd.getconf("PAGESIZE"))
+		if int_value < min_bytes:
+			log.error("The value of '%s' must be at least twice the page size (%s)." % (option, min_bytes))
+			return False
+		return True
+
+	def _check_positive(self, option, int_value):
+		if int_value <= 0:
+			log.error("The value of '%s' must be positive." % option)
+			return False
+		return True
+
+	def _check_ratio(self, option, int_value):
+		if not 0 <= int_value <= 100:
+			log.error("The value of '%s' must be between 0 and 100." % option)
+			return False
+		return True
+
+	@command_custom("dirty_bytes")
+	def _dirty_bytes(self, enabling, value, verify, ignore_missing):
+		return self._dirty_option("dirty_bytes", "dirty_ratio", self._check_twice_pagesize, enabling, value, verify)
+
+	@command_custom("dirty_ratio")
+	def _dirty_ratio(self, enabling, value, verify, ignore_missing):
+		return self._dirty_option("dirty_ratio", "dirty_bytes", self._check_ratio, enabling, value, verify)
+
+	@command_custom("dirty_background_bytes")
+	def _dirty_background_bytes(self, enabling, value, verify, ignore_missing):
+		return self._dirty_option("dirty_background_bytes", "dirty_background_ratio", self._check_positive, enabling, value, verify)
+
+	@command_custom("dirty_background_ratio")
+	def _dirty_background_ratio(self, enabling, value, verify, ignore_missing):
+		return self._dirty_option("dirty_background_ratio", "dirty_background_bytes", self._check_ratio, enabling, value, verify)
+
+	def _dirty_option(self, option, counterpart, check_fun, enabling, value, verify):
+		option_path = self._proc_sys_vm_option_path(option)
+		counterpart_path = self._proc_sys_vm_option_path(counterpart)
+		option_key = self._storage_key(command_name=option)
+		counterpart_key = self._storage_key(command_name=counterpart)
+		if not os.path.isfile(option_path):
+			log.warning("Option '%s' is not supported on the current hardware." % option)
+		current_value = cmd.read_file(option_path).strip()
+		if verify:
+			return current_value == value
+		if enabling:
+			try:
+				int_value = int(value)
+			except ValueError:
+				log.error("The value of '%s' must be an integer." % option)
+			if not check_fun(option, int_value):
+				return None
+			if current_value == value:
+				log.info("Not setting option '%s' to '%s', already set." % (option, value))
+				return value
+			# Backup: if the option (e.g. dirty_bytes) is currently 0,
+			# its counterpart (dirty_ratio) is the active one, so we
+			# back up that one instead.
+			if int(current_value) == 0:
+				current_counterpart_value = cmd.read_file(counterpart_path).strip()
+				self._storage.set(counterpart_key, current_counterpart_value)
+			else:
+				self._storage.set(option_key, current_value)
+			log.info("Setting option '%s' to '%s'." % (option, value))
+			cmd.write_to_file(option_path, value)
+			return value
+		# Rollback is analogous to the backup: if there is no backed up
+		# value for this option, it means that its counterpart was active
+		# and we have to restore that one.
+		old_value = self._storage.get(option_key)
+		old_counterpart_value = self._storage.get(counterpart_key)
+		if old_value is not None:
+			log.info("Setting option '%s' to '%s'" % (option, old_value))
+			cmd.write_to_file(option_path, old_value)
+		elif old_counterpart_value is not None:
+			log.info("Setting option '%s' to '%s'" % (counterpart, old_counterpart_value))
+			cmd.write_to_file(counterpart_path, old_counterpart_value)
+		else:
+			log.info("Not restoring '%s', previous value is the same or unknown." % option)
+		return None
diff --git a/tuned/utils/commands.py b/tuned/utils/commands.py
index c4f7c936..10d589a9 100644
--- a/tuned/utils/commands.py
+++ b/tuned/utils/commands.py
@@ -552,3 +552,6 @@ def tr(self, text, source_chars, dest_chars):
 	# Checks if name contains only valid characters and has valid length or is empty string or None
 	def is_valid_name(self, name):
 		return not name or (all(c in consts.NAMES_ALLOWED_CHARS for c in name) and len(name) <= consts.NAMES_MAX_LENGTH)
+
+	def getconf(self, variable):
+		return check_output(["getconf", variable]).decode().strip()

From 155f2323a41b866217ee76b52b5648588d09f583 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Pavol=20=C5=BD=C3=A1=C4=8Dik?= <zacik.pa@gmail.com>
Date: Tue, 17 Sep 2024 16:24:47 +0200
Subject: [PATCH 2/3] Adjust profiles to set dirty_(bytes|ratio) via the vm
 plugin

---
 profiles/accelerator-performance/tuned.conf   |  7 ++---
 profiles/latency-performance/tuned.conf       |  7 ++---
 profiles/mssql/tuned.conf                     |  4 +--
 profiles/oracle/tuned.conf                    |  4 +--
 profiles/postgresql/tuned.conf                | 12 ++++-----
 profiles/sap-hana/tuned.conf                  |  4 +--
 profiles/spectrumscale-ece/tuned.conf         |  6 +++--
 profiles/spindown-disk/tuned.conf             |  4 ++-
 profiles/throughput-performance/tuned.conf    | 27 ++++++++++---------
 profiles/virtual-guest/tuned.conf             |  5 ++--
 profiles/virtual-host/tuned.conf              |  4 +--
 .../post-vars/tuned.conf                      |  4 +--
 .../post/tuned.conf                           |  4 +--
 .../post2/tuned.conf                          |  4 +--
 14 files changed, 51 insertions(+), 45 deletions(-)

diff --git a/profiles/accelerator-performance/tuned.conf b/profiles/accelerator-performance/tuned.conf
index a0f3da45..01f1a780 100644
--- a/profiles/accelerator-performance/tuned.conf
+++ b/profiles/accelerator-performance/tuned.conf
@@ -17,7 +17,7 @@ platform_profile=performance
 [disk]
 readahead=>4096
 
-[sysctl]
+[vm]
 # If a workload mostly uses anonymous memory and it hits this limit, the entire
 # working set is buffered for I/O, and any more write buffering would require
 # swapping, so it's time to throttle writes until I/O can catch up.  Workloads
@@ -25,12 +25,13 @@ readahead=>4096
 #
 # The generator of dirty data starts writeback at this percentage (system default
 # is 20%)
-vm.dirty_ratio = 40
+dirty_ratio = 40
 
 # Start background writeback (via writeback threads) at this percentage (system
 # default is 10%)
-vm.dirty_background_ratio = 10
+dirty_background_ratio = 10
 
+[sysctl]
 # PID allocation wrap value.  When the kernel's next PID value
 # reaches this value, it wraps back to a minimum PID value.
 # PIDs of value pid_max or larger are not allocated.
diff --git a/profiles/latency-performance/tuned.conf b/profiles/latency-performance/tuned.conf
index c7806023..1e044f54 100644
--- a/profiles/latency-performance/tuned.conf
+++ b/profiles/latency-performance/tuned.conf
@@ -14,7 +14,7 @@ min_perf_pct=100
 [acpi]
 platform_profile=performance
 
-[sysctl]
+[vm]
 # If a workload mostly uses anonymous memory and it hits this limit, the entire
 # working set is buffered for I/O, and any more write buffering would require
 # swapping, so it's time to throttle writes until I/O can catch up.  Workloads
@@ -22,12 +22,13 @@ platform_profile=performance
 #
 # The generator of dirty data starts writeback at this percentage (system default
 # is 20%)
-vm.dirty_ratio=10
+dirty_ratio=10
 
 # Start background writeback (via writeback threads) at this percentage (system
 # default is 10%)
-vm.dirty_background_ratio=3
+dirty_background_ratio=3
 
+[sysctl]
 # The swappiness parameter controls the tendency of the kernel to move
 # processes out of physical memory and onto the swap disk.
 # 0 tells the kernel to avoid swapping processes out of physical memory
diff --git a/profiles/mssql/tuned.conf b/profiles/mssql/tuned.conf
index a1c97863..cf3e0f94 100644
--- a/profiles/mssql/tuned.conf
+++ b/profiles/mssql/tuned.conf
@@ -12,11 +12,11 @@ force_latency=5
 [vm]
 # For multi-instance SQL deployments use 'madvise' instead of 'always'
 transparent_hugepages=always
+dirty_background_ratio=3
+dirty_ratio=80
 
 [sysctl]
 vm.swappiness=1
-vm.dirty_background_ratio=3
-vm.dirty_ratio=80
 vm.dirty_expire_centisecs=500
 vm.dirty_writeback_centisecs=100
 vm.max_map_count=1600000
diff --git a/profiles/oracle/tuned.conf b/profiles/oracle/tuned.conf
index f84cb2f9..a1673ff7 100644
--- a/profiles/oracle/tuned.conf
+++ b/profiles/oracle/tuned.conf
@@ -8,8 +8,6 @@ include=throughput-performance
 
 [sysctl]
 vm.swappiness = 10
-vm.dirty_background_ratio = 3
-vm.dirty_ratio = 40
 vm.dirty_expire_centisecs = 500
 vm.dirty_writeback_centisecs = 100
 kernel.shmmax = 4398046511104
@@ -27,4 +25,6 @@ kernel.panic_on_oops = 1
 kernel.numa_balancing = 0
 
 [vm]
+dirty_background_ratio = 3
+dirty_ratio = 40
 transparent_hugepages=never
diff --git a/profiles/postgresql/tuned.conf b/profiles/postgresql/tuned.conf
index 88da8e44..8165fd8b 100644
--- a/profiles/postgresql/tuned.conf
+++ b/profiles/postgresql/tuned.conf
@@ -16,22 +16,20 @@ force_latency=1
 [vm]
 transparent_hugepages=never
 
-[sysctl]
-
 # The dirty_background_ratio and dirty_ratio controls percentage of memory
 # that file system cache have to fill with dirty data before kernel will
 # will start to flush data to disks. The default values are 10% and 20%
 # accordingly. On a systems with a big amount of memory this values can
 # be tens of gigabytes and produce IO spikes when PostgreSQL server writes
-# checkpoints.
+# checkpoints. The values can alternatively be set in absolute number of bytes
+# via dirty_background_bytes and dirty_bytes.
 #
 # Keep this values reasonable small - about size of RAID controller write-back
 # cache size (typcal 512MB - 2GB).
-vm.dirty_background_ratio = 0
-vm.dirty_ratio = 0
-vm.dirty_background_bytes = 67108864
-vm.dirty_bytes = 536870912
+dirty_background_bytes = 67108864
+dirty_bytes = 536870912
 
+[sysctl]
 # The swappiness parameter controls the tendency of the kernel to move
 # processes out of physical memory and onto the swap disk.
 # 0 tells the kernel to avoid swapping processes out of physical memory
diff --git a/profiles/sap-hana/tuned.conf b/profiles/sap-hana/tuned.conf
index aeecf53c..e369ddf7 100644
--- a/profiles/sap-hana/tuned.conf
+++ b/profiles/sap-hana/tuned.conf
@@ -13,10 +13,10 @@ min_perf_pct=100
 
 [vm]
 transparent_hugepages=never
+dirty_ratio = 40
+dirty_background_ratio = 10
 
 [sysctl]
 kernel.sem = 32000 1024000000 500 32000
 kernel.numa_balancing = 0
-vm.dirty_ratio = 40
-vm.dirty_background_ratio = 10
 vm.swappiness = 10
diff --git a/profiles/spectrumscale-ece/tuned.conf b/profiles/spectrumscale-ece/tuned.conf
index c111e31e..3b3bebab 100644
--- a/profiles/spectrumscale-ece/tuned.conf
+++ b/profiles/spectrumscale-ece/tuned.conf
@@ -11,10 +11,12 @@ governor=performance
 energy_perf_bias=performance
 min_perf_pct=100
 
+[vm]
+dirty_ratio = 40
+dirty_background_ratio = 10
+
 [sysctl]
 kernel.numa_balancing = 1
-vm.dirty_ratio = 40
-vm.dirty_background_ratio = 10
 vm.swappiness=10
 net.ipv4.tcp_window_scaling = 1
 net.ipv4.tcp_timestamps = 1
diff --git a/profiles/spindown-disk/tuned.conf b/profiles/spindown-disk/tuned.conf
index 6f126846..5252bb73 100644
--- a/profiles/spindown-disk/tuned.conf
+++ b/profiles/spindown-disk/tuned.conf
@@ -26,10 +26,12 @@ spindown=6
 [scsi_host]
 alpm=medium_power
 
+[vm]
+dirty_ratio=60
+
 [sysctl]
 vm.dirty_writeback_centisecs=6000
 vm.dirty_expire_centisecs=9000
-vm.dirty_ratio=60
 vm.laptop_mode=5
 vm.swappiness=30
 
diff --git a/profiles/throughput-performance/tuned.conf b/profiles/throughput-performance/tuned.conf
index b5e266d7..b55989c5 100644
--- a/profiles/throughput-performance/tuned.conf
+++ b/profiles/throughput-performance/tuned.conf
@@ -17,6 +17,20 @@ energy_performance_preference=performance
 [acpi]
 platform_profile=performance
 
+[vm]
+# If a workload mostly uses anonymous memory and it hits this limit, the entire
+# working set is buffered for I/O, and any more write buffering would require
+# swapping, so it's time to throttle writes until I/O can catch up.  Workloads
+# that mostly use file mappings may be able to use even higher values.
+#
+# The generator of dirty data starts writeback at this percentage (system default
+# is 20%)
+dirty_ratio = 40
+
+# Start background writeback (via writeback threads) at this percentage (system
+# default is 10%)
+dirty_background_ratio = 10
+
 # Marvell ThunderX
 [vm.thunderx]
 type=vm
@@ -31,19 +45,6 @@ transparent_hugepages=never
 readahead=>4096
 
 [sysctl]
-# If a workload mostly uses anonymous memory and it hits this limit, the entire
-# working set is buffered for I/O, and any more write buffering would require
-# swapping, so it's time to throttle writes until I/O can catch up.  Workloads
-# that mostly use file mappings may be able to use even higher values.
-#
-# The generator of dirty data starts writeback at this percentage (system default
-# is 20%)
-vm.dirty_ratio = 40
-
-# Start background writeback (via writeback threads) at this percentage (system
-# default is 10%)
-vm.dirty_background_ratio = 10
-
 # PID allocation wrap value.  When the kernel's next PID value
 # reaches this value, it wraps back to a minimum PID value.
 # PIDs of value pid_max or larger are not allocated.
diff --git a/profiles/virtual-guest/tuned.conf b/profiles/virtual-guest/tuned.conf
index 28b85c97..bc4c0399 100644
--- a/profiles/virtual-guest/tuned.conf
+++ b/profiles/virtual-guest/tuned.conf
@@ -6,7 +6,7 @@
 summary=Optimize for running inside a virtual guest
 include=throughput-performance
 
-[sysctl]
+[vm]
 # If a workload mostly uses anonymous memory and it hits this limit, the entire
 # working set is buffered for I/O, and any more write buffering would require
 # swapping, so it's time to throttle writes until I/O can catch up.  Workloads
@@ -14,8 +14,9 @@ include=throughput-performance
 #
 # The generator of dirty data starts writeback at this percentage (system default
 # is 20%)
-vm.dirty_ratio = 30
+dirty_ratio = 30
 
+[sysctl]
 # Filesystem I/O is usually much more efficient than swapping, so try to keep
 # swapping low.  It's usually safe to go even lower than this on systems with
 # server-grade storage.
diff --git a/profiles/virtual-host/tuned.conf b/profiles/virtual-host/tuned.conf
index 5301d9ff..ba58cbb7 100644
--- a/profiles/virtual-host/tuned.conf
+++ b/profiles/virtual-host/tuned.conf
@@ -6,10 +6,10 @@
 summary=Optimize for running KVM guests
 include=throughput-performance
 
-[sysctl]
+[vm]
 # Start background writeback (via writeback threads) at this percentage (system
 # default is 10%)
-vm.dirty_background_ratio = 5
+dirty_background_ratio = 5
 
 [cpu]
 # Setting C3 state sleep mode/power savings
diff --git a/tests/beakerlib/bz1798183-RFE-support-post-loaded-profile/post-vars/tuned.conf b/tests/beakerlib/bz1798183-RFE-support-post-loaded-profile/post-vars/tuned.conf
index 0bef6122..a93de044 100644
--- a/tests/beakerlib/bz1798183-RFE-support-post-loaded-profile/post-vars/tuned.conf
+++ b/tests/beakerlib/bz1798183-RFE-support-post-loaded-profile/post-vars/tuned.conf
@@ -1,5 +1,5 @@
 [main]
 summary=Post-loaded profile that uses variables from the regular active profile
 
-[sysctl]
-vm.dirty_ratio=${foo}
+[vm]
+dirty_ratio=${foo}
diff --git a/tests/beakerlib/bz1798183-RFE-support-post-loaded-profile/post/tuned.conf b/tests/beakerlib/bz1798183-RFE-support-post-loaded-profile/post/tuned.conf
index 7844ca82..de0629f7 100644
--- a/tests/beakerlib/bz1798183-RFE-support-post-loaded-profile/post/tuned.conf
+++ b/tests/beakerlib/bz1798183-RFE-support-post-loaded-profile/post/tuned.conf
@@ -1,5 +1,5 @@
 [main]
 summary=Post-loaded profile
 
-[sysctl]
-vm.dirty_ratio=8
+[vm]
+dirty_ratio=8
diff --git a/tests/beakerlib/bz1798183-RFE-support-post-loaded-profile/post2/tuned.conf b/tests/beakerlib/bz1798183-RFE-support-post-loaded-profile/post2/tuned.conf
index 994db42d..c8fedd9e 100644
--- a/tests/beakerlib/bz1798183-RFE-support-post-loaded-profile/post2/tuned.conf
+++ b/tests/beakerlib/bz1798183-RFE-support-post-loaded-profile/post2/tuned.conf
@@ -1,5 +1,5 @@
 [main]
 summary=Second version of the post-loaded profile
 
-[sysctl]
-vm.dirty_ratio=7
+[vm]
+dirty_ratio=7

From 52768f78e4fcc9d657ce981a4790360444f089ec Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Pavol=20=C5=BD=C3=A1=C4=8Dik?= <zacik.pa@gmail.com>
Date: Wed, 18 Sep 2024 08:26:36 +0200
Subject: [PATCH 3/3] Document new vm plugin options

---
 tuned/plugins/plugin_vm.py | 12 +++++++++---
 1 file changed, 9 insertions(+), 3 deletions(-)

diff --git a/tuned/plugins/plugin_vm.py b/tuned/plugins/plugin_vm.py
index a2f3ae75..a671e171 100644
--- a/tuned/plugins/plugin_vm.py
+++ b/tuned/plugins/plugin_vm.py
@@ -13,9 +13,15 @@
 
 class VMPlugin(base.Plugin):
 	"""
-	Enables or disables transparent huge pages depending on value of the
-	[option]`transparent_hugepages` option. The option can have one of three
-	possible values `always`, `madvise` and `never`.
+	Tunes selected sysctl options in `/proc/sys/vm`, currently
+	[option]`dirty_ratio`, [option]`dirty_background_ratio`,
+	[option]`dirty_bytes`, and [option]`dirty_background_bytes`.
+	See https://docs.kernel.org/admin-guide/sysctl/vm.html for detailed
+	documentation of these options.
+
+	Additionaly enables or disables transparent huge pages depending on
+	the value of the [option]`transparent_hugepages` option. The option
+	can have one of three possible values: `always`, `madvise` and `never`.
 
 	.Disable transparent hugepages
 	====