ZFS monitoring (#241)

Add new eBPF program to monitor ZFS filesystem.

kernel/Makefile

@@ -48,6 +48,7 @@ NETDATA_APPS= cachestat \
 	      swap \
 	      vfs \
 	      xfs \
+	      zfs \
 	      #
 
 all: $(NETDATA_APPS)

kernel/README.md

@@ -37,3 +37,4 @@ Right now we have two `eBPF` program collections:
 -  `syncfs_kern.c`         : eBPF program that monitor calls for syscall `syncfs`.
 -  `vfs_kern.c`            : eBPF program that monitor Virtual Filesystem functions.
 -  `xfs_kern.c`            : eBPF program that provides XFS monitoring.
+-  `zfs_kern.c`            : eBPF program that provides ZFS monitoring.

kernel/zfs_kern.c

@@ -0,0 +1,208 @@
+#define KBUILD_MODNAME "zfs_netdata"
+#include <linux/bpf.h>
+#include <linux/ptrace.h>
+#include <linux/genhd.h>
+
+#include "bpf_helpers.h"
+#include "netdata_ebpf.h"
+
+/************************************************************************************
+ *     
+ *                                 MAP Section
+ *     
+ ***********************************************************************************/
+
+struct bpf_map_def SEC("maps") tbl_zfs = {
+    .type = BPF_MAP_TYPE_PERCPU_ARRAY,
+    .key_size = sizeof(__u32),
+    .value_size = sizeof(__u64),
+    .max_entries = NETDATA_FS_MAX_ELEMENTS
+};
+
+struct bpf_map_def SEC("maps") tmp_zfs = {
+#if (LINUX_VERSION_CODE < KERNEL_VERSION(4,15,0)) 
+    .type = BPF_MAP_TYPE_HASH,
+#else
+    .type = BPF_MAP_TYPE_PERCPU_HASH,
+#endif    
+    .key_size = sizeof(__u32),
+    .value_size = sizeof(__u64),
+    .max_entries = 4192
+};
+
+/************************************************************************************
+ *     
+ *                                 ENTRY Section
+ *     
+ ***********************************************************************************/
+
+#if (LINUX_VERSION_CODE > KERNEL_VERSION(5,0,0))
+static int netdata_zfs_entry()
+#elif (LINUX_VERSION_CODE > KERNEL_VERSION(4,19,0)) 
+static __always_inline int netdata_zfs_entry()
+#else
+static inline int netdata_zfs_entry()
+#endif
+{
+    __u64 pid_tgid = bpf_get_current_pid_tgid();
+    __u32 pid = (__u32)(pid_tgid >> 32);
+    __u64 ts = bpf_ktime_get_ns();
+
+    bpf_map_update_elem(&tmp_zfs, &pid, &ts, BPF_ANY);
+
+    return 0;
+}
+
+SEC("kprobe/zpl_iter_read")
+int netdata_zpl_iter_read(struct pt_regs *ctx) 
+{
+    return netdata_zfs_entry();
+}
+
+SEC("kprobe/zpl_iter_write")
+int netdata_zpl_iter_write(struct pt_regs *ctx) 
+{
+    return netdata_zfs_entry();
+}
+
+SEC("kprobe/zpl_open")
+int netdata_zpl_open(struct pt_regs *ctx) 
+{
+    return netdata_zfs_entry();
+}
+
+SEC("kprobe/zpl_fsync")
+int netdata_zpl_fsync(struct pt_regs *ctx) 
+{
+    return netdata_zfs_entry();
+}
+
+/************************************************************************************
+ *     
+ *                                 END Section
+ *     
+ ***********************************************************************************/
+
+static void netdata_zfs_store_bin(__u32 bin, __u32 selection)
+{
+    __u64 *fill, data;
+    __u32 idx = selection * NETDATA_FS_MAX_BINS + bin;
+    if (idx >= NETDATA_FS_MAX_ELEMENTS)
+        return;
+
+    fill = bpf_map_lookup_elem(&tbl_zfs, &idx);
+    if (fill) {
+        libnetdata_update_u64(fill, 1);
+		return;
+    } 
+
+    data = 1;
+    bpf_map_update_elem(&tbl_zfs, &idx, &data, BPF_ANY);
+}
+
+SEC("kretprobe/zpl_iter_read")
+int netdata_ret_zpl_iter_read(struct pt_regs *ctx)
+{
+    __u64 *fill, data;
+    __u64 pid_tgid = bpf_get_current_pid_tgid();
+    __u32 bin, pid = (__u32)(pid_tgid >> 32);
+
+    fill = bpf_map_lookup_elem(&tmp_zfs, &pid);
+    if (!fill)
+        return 0;
+
+    data = bpf_ktime_get_ns() - *fill;
+    bpf_map_delete_elem(&tmp_zfs, &pid);
+
+    // Skip entries with backward time
+    if ( (s64)data < 0)
+        return 0;
+
+    // convert to microseconds
+    data /= 1000;
+    bin = libnetdata_select_idx(data, NETDATA_FS_MAX_BINS_POS);
+    netdata_zfs_store_bin(bin, NETDATA_KEY_CALLS_READ);
+
+    return 0;
+}
+
+SEC("kretprobe/zpl_iter_write")
+int netdata_ret_zpl_iter_write(struct pt_regs *ctx)
+{
+    __u64 *fill, data;
+    __u64 pid_tgid = bpf_get_current_pid_tgid();
+    __u32 bin, pid = (__u32)(pid_tgid >> 32);
+
+    fill = bpf_map_lookup_elem(&tmp_zfs, &pid);
+    if (!fill)
+        return 0;
+
+    data = bpf_ktime_get_ns() - *fill;
+    bpf_map_delete_elem(&tmp_zfs, &pid);
+
+    // Skip entries with backward time
+    if ( (s64)data < 0)
+        return 0;
+
+    // convert to microseconds
+    data /= 1000;
+    bin = libnetdata_select_idx(data, NETDATA_FS_MAX_BINS_POS);
+    netdata_zfs_store_bin(bin, NETDATA_KEY_CALLS_WRITE);
+
+    return 0;
+}
+
+SEC("kretprobe/zpl_open")
+int netdata_ret_zpl_open(struct pt_regs *ctx)
+{
+    __u64 *fill, data;
+    __u64 pid_tgid = bpf_get_current_pid_tgid();
+    __u32 bin, pid = (__u32)(pid_tgid >> 32);
+
+    fill = bpf_map_lookup_elem(&tmp_zfs, &pid);
+    if (!fill)
+        return 0;
+
+    data = bpf_ktime_get_ns() - *fill;
+    bpf_map_delete_elem(&tmp_zfs, &pid);
+
+    // Skip entries with backward time
+    if ( (s64)data < 0)
+        return 0;
+
+    // convert to microseconds
+    data /= 1000;
+    bin = libnetdata_select_idx(data, NETDATA_FS_MAX_BINS_POS);
+    netdata_zfs_store_bin(bin, NETDATA_KEY_CALLS_OPEN);
+
+    return 0;
+}
+
+SEC("kretprobe/zpl_fsync")
+int netdata_ret_zpl_fsync(struct pt_regs *ctx) 
+{
+    __u64 *fill, data;
+    __u64 pid_tgid = bpf_get_current_pid_tgid();
+    __u32 bin, pid = (__u32)(pid_tgid >> 32);
+
+    fill = bpf_map_lookup_elem(&tmp_zfs, &pid);
+    if (!fill)
+        return 0;
+
+    data = bpf_ktime_get_ns() - *fill;
+    bpf_map_delete_elem(&tmp_zfs, &pid);
+
+    // Skip entries with backward time
+    if ( (s64)data < 0)
+        return 0;
+
+    // convert to microseconds
+    data /= 1000;
+    bin = libnetdata_select_idx(data, NETDATA_FS_MAX_BINS_POS);
+    netdata_zfs_store_bin(bin, NETDATA_KEY_CALLS_SYNC);
+
+    return 0;
+}
+
+char _license[] SEC("license") = "GPL";
+