Merge pull request #15769 from dfr/freebsd-stats

Add support for 'podman stats' on FreeBSD

5 files changed, 265 insertions, 38 deletions

diff --git a/libpod/stats_common.go b/libpod/stats_common.go
new file mode 100644
index 000000000..122160bda
--- /dev/null
+++ b/libpod/stats_common.go
@@ -0,0 +1,49 @@
+//go:build linux || freebsd
+// +build linux freebsd
+
+package libpod
+
+import (
+	"fmt"
+
+	"github.com/containers/podman/v4/libpod/define"
+)
+
+// GetContainerStats gets the running stats for a given container.
+// The previousStats is used to correctly calculate cpu percentages. You
+// should pass nil if there is no previous stat for this container.
+func (c *Container) GetContainerStats(previousStats *define.ContainerStats) (*define.ContainerStats, error) {
+	stats := new(define.ContainerStats)
+	stats.ContainerID = c.ID()
+	stats.Name = c.Name()
+
+	if c.config.NoCgroups {
+		return nil, fmt.Errorf("cannot run top on container %s as it did not create a cgroup: %w", c.ID(), define.ErrNoCgroups)
+	}
+
+	if !c.batched {
+		c.lock.Lock()
+		defer c.lock.Unlock()
+		if err := c.syncContainer(); err != nil {
+			return stats, err
+		}
+	}
+
+	// returns stats with the fields' default values respective of their type
+	if c.state.State != define.ContainerStateRunning && c.state.State != define.ContainerStatePaused {
+		return stats, nil
+	}
+
+	if previousStats == nil {
+		previousStats = &define.ContainerStats{
+			// if we have no prev stats use the container start time as prev time
+			// otherwise we cannot correctly calculate the CPU percentage
+			SystemNano: uint64(c.state.StartedTime.UnixNano()),
+		}
+	}
+
+	if err := c.getPlatformContainerStats(stats, previousStats); err != nil {
+		return nil, err
+	}
+	return stats, nil
+}
diff --git a/libpod/stats_freebsd.go b/libpod/stats_freebsd.go
new file mode 100644
index 000000000..53bc3f19a
--- /dev/null
+++ b/libpod/stats_freebsd.go
@@ -0,0 +1,153 @@
+package libpod
+
+import (
+	"fmt"
+	"math"
+	"strings"
+	"time"
+
+	"github.com/containers/common/pkg/cgroups"
+	"github.com/containers/podman/v4/libpod/define"
+	"github.com/containers/podman/v4/pkg/rctl"
+	"github.com/containers/storage/pkg/system"
+	"github.com/sirupsen/logrus"
+)
+
+// getPlatformContainerStats gets the platform-specific running stats
+// for a given container.  The previousStats is used to correctly
+// calculate cpu percentages. You should pass nil if there is no
+// previous stat for this container.
+func (c *Container) getPlatformContainerStats(stats *define.ContainerStats, previousStats *define.ContainerStats) error {
+	now := uint64(time.Now().UnixNano())
+
+	jailName := c.ID()
+	if c.state.NetNS != nil {
+		jailName = c.state.NetNS.Name + "." + jailName
+	}
+	entries, err := rctl.GetRacct("jail:" + jailName)
+	if err != nil {
+		return fmt.Errorf("unable to read accounting for %s: %w", jailName, err)
+	}
+
+	// If the current total usage is less than what was previously
+	// recorded then it means the container was restarted and runs
+	// in a new jail
+	if dur, ok := entries["wallclock"]; ok {
+		if previousStats.Duration > dur*1000000000 {
+			previousStats = &define.ContainerStats{}
+		}
+	}
+
+	for key, val := range entries {
+		switch key {
+		case "cputime": // CPU time, in seconds
+			stats.CPUNano = val * 1000000000
+			stats.AvgCPU = calculateCPUPercent(stats.CPUNano, 0, now, uint64(c.state.StartedTime.UnixNano()))
+		case "datasize": // data size, in bytes
+		case "stacksize": // stack size, in bytes
+		case "coredumpsize": // core dump size, in bytes
+		case "memoryuse": // resident set size, in bytes
+			stats.MemUsage = val
+		case "memorylocked": // locked memory, in bytes
+		case "maxproc": // number of processes
+			stats.PIDs = val
+		case "openfiles": // file descriptor table size
+		case "vmemoryuse": // address space limit, in bytes
+		case "pseudoterminals": // number of PTYs
+		case "swapuse": // swap space that may be reserved or used, in bytes
+		case "nthr": // number of threads
+		case "msgqqueued": // number of queued SysV messages
+		case "msgqsize": // SysV message queue size, in bytes
+		case "nmsgq": // number of SysV message queues
+		case "nsem": // number of SysV semaphores
+		case "nsemop": // number of SysV semaphores modified in a single semop(2) call
+		case "nshm": // number of SysV shared memory segments
+		case "shmsize": // SysV shared memory size, in bytes
+		case "wallclock": // wallclock time, in seconds
+			stats.Duration = val * 1000000000
+			stats.UpTime = time.Duration(stats.Duration)
+		case "pcpu": // %CPU, in percents of a single CPU core
+			stats.CPU = float64(val)
+		case "readbps": // filesystem reads, in bytes per second
+			stats.BlockInput = val
+		case "writebps": // filesystem writes, in bytes per second
+			stats.BlockOutput = val
+		case "readiops": // filesystem reads, in operations per second
+		case "writeiops": // filesystem writes, in operations per second
+		}
+	}
+	stats.MemLimit = c.getMemLimit()
+	stats.SystemNano = now
+
+	netStats, err := getContainerNetIO(c)
+	if err != nil {
+		return err
+	}
+
+	// Handle case where the container is not in a network namespace
+	if netStats != nil {
+		stats.NetInput = netStats.TxBytes
+		stats.NetOutput = netStats.RxBytes
+	} else {
+		stats.NetInput = 0
+		stats.NetOutput = 0
+	}
+
+	return nil
+}
+
+// getMemory limit returns the memory limit for a container
+func (c *Container) getMemLimit() uint64 {
+	memLimit := uint64(math.MaxUint64)
+
+	if c.config.Spec.Linux != nil && c.config.Spec.Linux.Resources != nil &&
+		c.config.Spec.Linux.Resources.Memory != nil && c.config.Spec.Linux.Resources.Memory.Limit != nil {
+		memLimit = uint64(*c.config.Spec.Linux.Resources.Memory.Limit)
+	}
+
+	mi, err := system.ReadMemInfo()
+	if err != nil {
+		logrus.Errorf("ReadMemInfo error: %v", err)
+		return 0
+	}
+
+	//nolint:unconvert
+	physicalLimit := uint64(mi.MemTotal)
+
+	if memLimit <= 0 || memLimit > physicalLimit {
+		return physicalLimit
+	}
+
+	return memLimit
+}
+
+// calculateCPUPercent calculates the cpu usage using the latest measurement in stats.
+// previousCPU is the last value of stats.CPU.Usage.Total measured at the time previousSystem.
+//
+//	(now - previousSystem) is the time delta in nanoseconds, between the measurement in previousCPU
+//
+// and the updated value in stats.
+func calculateCPUPercent(currentCPU, previousCPU, now, previousSystem uint64) float64 {
+	var (
+		cpuPercent  = 0.0
+		cpuDelta    = float64(currentCPU - previousCPU)
+		systemDelta = float64(now - previousSystem)
+	)
+	if systemDelta > 0.0 && cpuDelta > 0.0 {
+		// gets a ratio of container cpu usage total, and multiplies that by 100 to get a percentage
+		cpuPercent = (cpuDelta / systemDelta) * 100
+	}
+	return cpuPercent
+}
+
+func calculateBlockIO(stats *cgroups.Metrics) (read uint64, write uint64) {
+	for _, blkIOEntry := range stats.Blkio.IoServiceBytesRecursive {
+		switch strings.ToLower(blkIOEntry.Op) {
+		case "read":
+			read += blkIOEntry.Value
+		case "write":
+			write += blkIOEntry.Value
+		}
+	}
+	return
+}
diff --git a/libpod/stats.go b/libpod/stats_linux.go
index c7e9e5128..ad8f33c91 100644
--- a/libpod/stats.go
+++ b/libpod/stats_linux.go
@@ -16,57 +16,33 @@ import (
 	"github.com/containers/podman/v4/libpod/define"
 )
 
-// GetContainerStats gets the running stats for a given container.
-// The previousStats is used to correctly calculate cpu percentages. You
-// should pass nil if there is no previous stat for this container.
-func (c *Container) GetContainerStats(previousStats *define.ContainerStats) (*define.ContainerStats, error) {
-	stats := new(define.ContainerStats)
-	stats.ContainerID = c.ID()
-	stats.Name = c.Name()
-
+// getPlatformContainerStats gets the platform-specific running stats
+// for a given container.  The previousStats is used to correctly
+// calculate cpu percentages. You should pass nil if there is no
+// previous stat for this container.
+func (c *Container) getPlatformContainerStats(stats *define.ContainerStats, previousStats *define.ContainerStats) error {
 	if c.config.NoCgroups {
-		return nil, fmt.Errorf("cannot run top on container %s as it did not create a cgroup: %w", c.ID(), define.ErrNoCgroups)
-	}
-
-	if !c.batched {
-		c.lock.Lock()
-		defer c.lock.Unlock()
-		if err := c.syncContainer(); err != nil {
-			return stats, err
-		}
-	}
-
-	// returns stats with the fields' default values respective of their type
-	if c.state.State != define.ContainerStateRunning && c.state.State != define.ContainerStatePaused {
-		return stats, nil
-	}
-
-	if previousStats == nil {
-		previousStats = &define.ContainerStats{
-			// if we have no prev stats use the container start time as prev time
-			// otherwise we cannot correctly calculate the CPU percentage
-			SystemNano: uint64(c.state.StartedTime.UnixNano()),
-		}
+		return fmt.Errorf("cannot run top on container %s as it did not create a cgroup: %w", c.ID(), define.ErrNoCgroups)
 	}
 
 	cgroupPath, err := c.cGroupPath()
 	if err != nil {
-		return nil, err
+		return err
 	}
 	cgroup, err := cgroups.Load(cgroupPath)
 	if err != nil {
-		return stats, fmt.Errorf("unable to load cgroup at %s: %w", cgroupPath, err)
+		return fmt.Errorf("unable to load cgroup at %s: %w", cgroupPath, err)
 	}
 
 	// Ubuntu does not have swap memory in cgroups because swap is often not enabled.
 	cgroupStats, err := cgroup.Stat()
 	if err != nil {
-		return stats, fmt.Errorf("unable to obtain cgroup stats: %w", err)
+		return fmt.Errorf("unable to obtain cgroup stats: %w", err)
 	}
 	conState := c.state.State
 	netStats, err := getContainerNetIO(c)
 	if err != nil {
-		return nil, err
+		return err
 	}
 
 	// If the current total usage in the cgroup is less than what was previously
@@ -103,7 +79,7 @@ func (c *Container) GetContainerStats(previousStats *define.ContainerStats) (*de
 		stats.NetOutput = 0
 	}
 
-	return stats, nil
+	return nil
 }
 
 // getMemory limit returns the memory limit for a container
@@ -133,7 +109,9 @@ func (c *Container) getMemLimit() uint64 {
 
 // calculateCPUPercent calculates the cpu usage using the latest measurement in stats.
 // previousCPU is the last value of stats.CPU.Usage.Total measured at the time previousSystem.
-//  (now - previousSystem) is the time delta in nanoseconds, between the measurement in previousCPU
+//
+//	(now - previousSystem) is the time delta in nanoseconds, between the measurement in previousCPU
+//
 // and the updated value in stats.
 func calculateCPUPercent(stats *runccgroup.Stats, previousCPU, now, previousSystem uint64) float64 {
 	var (
diff --git a/libpod/stats_unsupported.go b/libpod/stats_unsupported.go
index b23333c2e..3094e2eaa 100644
--- a/libpod/stats_unsupported.go
+++ b/libpod/stats_unsupported.go
@@ -1,5 +1,5 @@
-//go:build !linux
-// +build !linux
+//go:build !linux && !freebsd
+// +build !linux,!freebsd
 
 package libpod
 
diff --git a/pkg/rctl/rctl.go b/pkg/rctl/rctl.go
new file mode 100644
index 000000000..135cc60cb
--- /dev/null
+++ b/pkg/rctl/rctl.go
@@ -0,0 +1,47 @@
+//go:build freebsd
+// +build freebsd
+
+package rctl
+
+// #include <sys/rctl.h>
+import "C"
+
+import (
+	"bytes"
+	"fmt"
+	"strconv"
+	"strings"
+	"syscall"
+	"unsafe"
+
+	"github.com/sirupsen/logrus"
+)
+
+func GetRacct(filter string) (map[string]uint64, error) {
+	bp, err := syscall.ByteSliceFromString(filter)
+	if err != nil {
+		return nil, err
+	}
+	var buf [1024]byte
+	_, _, errno := syscall.Syscall6(syscall.SYS_RCTL_GET_RACCT,
+		uintptr(unsafe.Pointer(&bp[0])),
+		uintptr(len(bp)),
+		uintptr(unsafe.Pointer(&buf[0])),
+		uintptr(len(buf)), 0, 0)
+	if errno != 0 {
+		return nil, fmt.Errorf("error calling rctl_get_racct with filter %s: %v", errno)
+	}
+	len := bytes.IndexByte(buf[:], byte(0))
+	entries := strings.Split(string(buf[:len]), ",")
+	res := make(map[string]uint64)
+	for _, entry := range entries {
+		kv := strings.SplitN(entry, "=", 2)
+		key := kv[0]
+		val, err := strconv.ParseUint(kv[1], 10, 0)
+		if err != nil {
+			logrus.Warnf("unexpected rctl entry, ignoring: %s", entry)
+		}
+		res[key] = val
+	}
+	return res, nil
+}