Feature/virtual mem flags parse

docs/docs/events/builtin/extra/access_remote_vm.md

@@ -0,0 +1,34 @@
+# access_remote_vm
+
+## Intro
+access_remote_vm - gain access to the virtual memory of a separate process through the use of the procfs mem file.
+
+## Description
+This event marks access attempt of a process to the virtual memory of another process using the procfs mem file associated with that specific process (/proc/<pid>/mem).
+It is a more elaborated event than the `security_file_open` of the mem file.
+
+## Arguments
+* `remote_pid`: `int`[K] - PID of the process the memory area belongs to.
+* `start_address`: `void *`[K] - Start address of the operation.
+* `gup_flags`: `unsigned int`[K] - Flags for get_user_pages operation.
+* `vm_flags`: `unsigned long`[K] - Virtual memory flags.
+* `mapped.path`: `const char*`[K] - Path of the mapped file, or the name of memory area if no file is mapped.
+* `mapped.device_id`: `dev_t`[K,OPT] - Device ID of the mapped file.
+* `mapped.inode_number`: `unsigned long`[K,OPT] - Inode number of the mapped file.
+* `mapped.ctime`: `unsigned long`[K,OPT] - Creation time of the mapped file.
+
+## Hooks
+### get_user_pages_remote
+#### Type
+kprobe + kretprobe
+#### Purpose
+The main function that implements the access to the virtual memory area of the other process.
+
+### generic_access_phys
+#### Type
+kprobe
+#### Purpose
+A fallback function, implementing the `access` method of the `vma_operations` struct for most of the vmas. It is used to access special memory areas.
+
+## Related Events
+`security_file_open`,`security_mmap_file`,`vfs_write`,`vfs_writev`,`vfs_read`,`vfs_readv`

go.mod

@@ -6,7 +6,7 @@ require (
 	github.com/IBM/fluent-forward-go v0.2.1
 	github.com/Masterminds/sprig/v3 v3.2.3
 	github.com/aquasecurity/libbpfgo v0.5.0-libbpf-1.2
-	github.com/aquasecurity/libbpfgo/helpers v0.4.6-0.20230321190037-f591a2c5734f
+	github.com/aquasecurity/libbpfgo/helpers v0.4.6-0.20231023122045-7a58b8ec6cc3
 	github.com/aquasecurity/tracee/types v0.0.0-20230929172705-29c9b630aeed
 	github.com/containerd/containerd v1.7.0
 	github.com/docker/docker v23.0.5+incompatible
@@ -27,7 +27,7 @@ require (
 	github.com/urfave/cli/v2 v2.3.0
 	go.uber.org/goleak v1.1.12
 	go.uber.org/zap v1.24.0
-	golang.org/x/sys v0.8.0
+	golang.org/x/sys v0.13.0
 	google.golang.org/genproto v0.0.0-20230410155749-daa745c078e1
 	google.golang.org/grpc v1.56.2
 	google.golang.org/protobuf v1.30.0

go.sum

@@ -69,6 +69,8 @@ github.com/aquasecurity/libbpfgo v0.5.0-libbpf-1.2 h1:Yywi9wC3GPDOgR8wr6P9geY2qv
 github.com/aquasecurity/libbpfgo v0.5.0-libbpf-1.2/go.mod h1:0rEApF1YBHGuZ4C8OYI9q5oDBVpgqtRqYATePl9mCDk=
 github.com/aquasecurity/libbpfgo/helpers v0.4.6-0.20230321190037-f591a2c5734f h1:l127H3NqJBmw+XMt+haBOeZIrBppuw7TJz26cWMI9kY=
 github.com/aquasecurity/libbpfgo/helpers v0.4.6-0.20230321190037-f591a2c5734f/go.mod h1:j/TQLmsZpOIdF3CnJODzYngG4yu1YoDCoRMELxkQSSA=
+github.com/aquasecurity/libbpfgo/helpers v0.4.6-0.20231023122045-7a58b8ec6cc3 h1:Znl2aCgGDhJmsJTd5FG753cZpzyBZD6OU4CBE+yqTxs=
+github.com/aquasecurity/libbpfgo/helpers v0.4.6-0.20231023122045-7a58b8ec6cc3/go.mod h1:RKGRrng1b+79fklwa81IP98tt4S3ioCM/x6TFcnxbDU=
 github.com/aquasecurity/tracee/types v0.0.0-20230929172705-29c9b630aeed h1:d8ehyRJFP5qS0zfTss65UC2SoF7WX+/4+UcTL/73x9M=
 github.com/aquasecurity/tracee/types v0.0.0-20230929172705-29c9b630aeed/go.mod h1:sorAC3uGSNB/b8A9PjG1MSbENl81dR7azTSj+HclE+A=
 github.com/arbovm/levenshtein v0.0.0-20160628152529-48b4e1c0c4d0 h1:jfIu9sQUG6Ig+0+Ap1h4unLjW6YQJpKZVmUzxsD4E/Q=
@@ -650,6 +652,8 @@ golang.org/x/sys v0.2.0/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
 golang.org/x/sys v0.3.0/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
 golang.org/x/sys v0.8.0 h1:EBmGv8NaZBZTWvrbjNoL6HVt+IVy3QDQpJs7VRIw3tU=
 golang.org/x/sys v0.8.0/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
+golang.org/x/sys v0.13.0 h1:Af8nKPmuFypiUBjVoU9V20FiaFXOcuZI21p0ycVYYGE=
+golang.org/x/sys v0.13.0/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
 golang.org/x/term v0.0.0-20201126162022-7de9c90e9dd1/go.mod h1:bj7SfCRtBDWHUb9snDiAeCFNEtKQo2Wmx5Cou7ajbmo=
 golang.org/x/term v0.0.0-20210927222741-03fcf44c2211/go.mod h1:jbD1KX2456YbFQfuXm/mYQcufACuNUgVhRMnK/tPxf8=
 golang.org/x/term v0.2.0/go.mod h1:TVmDHMZPmdnySmBfhjOoOdhjzdE1h4u1VwSiw2l1Nuc=

pkg/ebpf/c/common/memory.h

@@ -4,6 +4,7 @@
 #include <vmlinux.h>
 
 #include <common/common.h>
+#include <common/ksymbols.h>
 
 // PROTOTYPES
 
@@ -12,7 +13,9 @@ statfunc unsigned long get_arg_start_from_mm(struct mm_struct *);
 statfunc unsigned long get_arg_end_from_mm(struct mm_struct *);
 statfunc unsigned long get_env_start_from_mm(struct mm_struct *);
 statfunc unsigned long get_env_end_from_mm(struct mm_struct *);
+statfunc struct task_struct *get_owner_task_from_mm(struct mm_struct *);
 statfunc unsigned long get_vma_flags(struct vm_area_struct *);
+statfunc int get_vma_location(struct vm_area_struct *, size_t, char *);
 
 // FUNCTIONS
 
@@ -41,11 +44,71 @@ statfunc unsigned long get_env_end_from_mm(struct mm_struct *mm)
     return BPF_CORE_READ(mm, env_end);
 }
 
+statfunc struct task_struct *get_owner_task_from_mm(struct mm_struct *mm)
+{
+    return BPF_CORE_READ(mm, owner);
+}
+
+statfunc struct file *get_mapped_file_from_vma(struct vm_area_struct *vma)
+{
+    return BPF_CORE_READ(vma, vm_file);
+}
+
 statfunc unsigned long get_vma_flags(struct vm_area_struct *vma)
 {
     return BPF_CORE_READ(vma, vm_flags);
 }
 
+// The function uses the "special_mapping_vmops" and "legacy_special_mapping_vmops" symbols, so
+// they should be requested by the event using this function.
+statfunc bool is_special_mapping(struct vm_area_struct *vma)
+{
+    char special_mapping_vmops_symbol[22] = "special_mapping_vmops";
+    void *special_mapping_vmops_addr = get_symbol_addr(special_mapping_vmops_symbol);
+    char legacy_special_mapping_vmops_symbol[29] = "legacy_special_mapping_vmops";
+    void *legacy_special_mapping_vmops_addr = get_symbol_addr(legacy_special_mapping_vmops_symbol);
+    const struct vm_operations_struct *vm_ops = BPF_CORE_READ(vma, vm_ops);
+    if (vm_ops != NULL &&
+        (vm_ops == special_mapping_vmops_addr || vm_ops == legacy_special_mapping_vmops_addr)) {
+        return true;
+    }
+    return false;
+}
+
+// Return the name of the location related to the given vma. For example [stack], [heap] and [vdso].
+// This function doesn't resolve file mapped vmas, so if the vma is a memory mapped file its
+// details should be collected in a different way.
+// The function uses the "special_mapping_vmops" and "legacy_special_mapping_vmops" symbols, so
+// they should be requested by the event using this function.
+statfunc int get_vma_location(struct vm_area_struct *vma, size_t max_len, char *location)
+{
+    // Avoid buffer overflow and satisfy the verifier.
+    if (max_len < 7) {
+        return 0;
+    }
+    struct mm_struct *vm_mm = BPF_CORE_READ(vma, vm_mm);
+    if (vm_mm) {
+        unsigned long start_stack = BPF_CORE_READ(vm_mm, start_stack);
+        unsigned long start_brk = BPF_CORE_READ(vm_mm, start_brk);
+        unsigned long brk = BPF_CORE_READ(vm_mm, brk);
+        u64 vm_start = BPF_CORE_READ(vma, vm_start);
+        u64 vm_end = BPF_CORE_READ(vma, vm_end);
+
+        if (vm_start <= start_stack && vm_end >= start_stack) {
+            __builtin_memcpy(location, "[stack]", 7);
+        } else if (vm_start <= brk && vm_end >= start_brk) {
+            __builtin_memcpy(location, "[heap]", 6);
+        } else if (is_special_mapping(vma)) {
+            struct vm_special_mapping *special_mapping = BPF_CORE_READ(vma, vm_private_data);
+            const char *name = BPF_CORE_READ(special_mapping, name);
+            if (name) {
+                bpf_probe_read_str(location, max_len, name);
+            }
+        }
+    }
+    return 0;
+}
+
 statfunc struct mount *real_mount(struct vfsmount *mnt)
 {
     return container_of(mnt, struct mount, mnt);

pkg/ebpf/c/tracee.bpf.c

@@ -4861,6 +4861,111 @@ int BPF_KPROBE(trace_ret_exec_binprm2)
     return events_perf_submit(&p, PROCESS_EXECUTION_FAILED, ret);
 }
 
+// Submit "access_remote_vm" event for the given vma and info.
+statfunc int submit_access_remote_vm(program_data_t *p,
+                                     struct vm_area_struct *vma,
+                                     void *address,
+                                     unsigned int gup_flags)
+{
+    unsigned long vm_flags = get_vma_flags(vma);
+    struct mm_struct *mm = BPF_CORE_READ(vma, vm_mm);
+    file_info_t file_info = {};
+
+    char location[64] = {0};
+
+    // Get mapped memory information
+    struct file *file = get_mapped_file_from_vma(vma);
+    file_info = get_file_info(file);
+    // If there is no file that is mapped to the vma, try to get memory location name instead.
+    if (!file_info.pathname_p) {
+        get_vma_location(vma, sizeof(location), location);
+    }
+
+    struct task_struct *task = get_owner_task_from_mm(mm);
+    u32 pid = get_task_host_tgid(task);
+
+    save_to_submit_buf(&p->event->args_buf, &pid, sizeof(u32), 0);
+    save_to_submit_buf(&p->event->args_buf, &address, sizeof(void *), 1);
+    save_to_submit_buf(&p->event->args_buf, &gup_flags, sizeof(unsigned int), 2);
+    save_to_submit_buf(&p->event->args_buf, &vm_flags, sizeof(unsigned long), 3);
+    if (file_info.pathname_p != NULL) {
+        save_str_to_buf(&p->event->args_buf, file_info.pathname_p, 4);
+        save_to_submit_buf(&p->event->args_buf, &file_info.id.device, sizeof(dev_t), 5);
+        save_to_submit_buf(&p->event->args_buf, &file_info.id.inode, sizeof(unsigned long), 6);
+        save_to_submit_buf(&p->event->args_buf, &file_info.id.ctime, sizeof(u64), 7);
+    } else {
+        save_str_to_buf(&p->event->args_buf, location, 4);
+    }
+
+    return events_perf_submit(p, ACCESS_REMOTE_VM, 0);
+}
+
+SEC("kprobe/generic_access_phys")
+int BPF_KPROBE(trace_generic_access_phys)
+{
+    program_data_t p = {};
+    if (!init_program_data(&p, ctx))
+        return 0;
+
+    if (!should_trace(&p))
+        return 0;
+
+    struct vm_area_struct *vma = (struct vm_area_struct *) PT_REGS_PARM1(ctx);
+    unsigned long addr = (unsigned long) PT_REGS_PARM2(ctx);
+    int write = (int) PT_REGS_PARM5(ctx);
+    return submit_access_remote_vm(&p, vma, (void *) addr, write);
+}
+
+SEC("kprobe/get_user_pages_remote")
+TRACE_ENT_FUNC(get_user_pages_remote, ACCESS_REMOTE_VM);
+
+SEC("kretprobe/get_user_pages_remote")
+int BPF_KPROBE(trace_ret_get_user_pages_remote)
+{
+    args_t saved_args;
+    if (load_args(&saved_args, ACCESS_REMOTE_VM) != 0) {
+        // missed entry or not traced
+        return 0;
+    }
+    del_args(ACCESS_REMOTE_VM);
+
+    program_data_t p = {};
+    if (!init_program_data(&p, ctx))
+        return 0;
+
+    if (!should_trace(&p))
+        return 0;
+
+    int ret_val = PT_REGS_RC(ctx);
+    // if the page cannot be pinned the return code would be <= 0.
+    // This could mean that the page is a memory mapped page, for example.
+    // In this case, the access to that page has to be done through vma->vm_ops->access which we
+    // trace for most cases (all normal cases) using generic_access_phys.
+    // So, we let the flow continue so it would be traced there.
+    if (ret_val <= 0)
+        return 0;
+
+    unsigned long address = (unsigned long) saved_args.args[1];
+    unsigned int gup_flags = (unsigned int) saved_args.args[3];
+    struct vm_area_struct **vmas = (struct vm_area_struct **) saved_args.args[5];
+    struct vm_area_struct *vma = NULL;
+    unsigned long vm_flags = 0;
+    file_info_t file_info = {};
+    // Without a vma we can't get enough information for the event.
+    // Moreover, this will filter out the calls for it from the execve flow.
+    if (!vmas)
+        return 0;
+
+    // In the flow of interest, via the access_remote_vm function, the logic iterate over the
+    // pages 1 by 1. This means that 1 and only 1 vma would be resolved here if the function
+    // succeed.
+    bpf_core_read(&vma, sizeof(void *), vmas);
+    if (!vma)
+        return 0;
+
+    return submit_access_remote_vm(&p, vma, (void *) address, gup_flags);
+}
+
 // clang-format off
 
 // Network Packets (works from ~5.2 and beyond)

pkg/ebpf/c/tracee.h

@@ -61,5 +61,7 @@ statfunc int capture_file_write(struct pt_regs *, u32, bool);
 statfunc bool filter_file_read_capture(program_data_t *, struct file *, io_data_t, off_t);
 statfunc int capture_file_read(struct pt_regs *, u32, bool);
 statfunc struct pipe_buffer *get_last_write_pipe_buffer(struct pipe_inode_info *);
+statfunc int
+submit_access_remote_vm(program_data_t *, struct vm_area_struct *, void *, unsigned int);
 
 #endif

pkg/ebpf/c/types.h

@@ -122,6 +122,7 @@ enum event_id_e
     HIDDEN_KERNEL_MODULE_SEEKER,
     MODULE_LOAD,
     MODULE_FREE,
+    ACCESS_REMOTE_VM,
     MAX_EVENT_ID,
 };
 

pkg/ebpf/c/vmlinux.h

@@ -283,10 +283,24 @@ typedef struct {
 struct signal_struct {
     atomic_t live;
 };
+typedef unsigned long vm_flags_t;
+
+struct vm_special_mapping {
+    const char *name;
+};
 
 struct vm_area_struct {
-    long unsigned int vm_flags;
+    union {
+        struct {
+            unsigned long vm_start;
+            unsigned long vm_end;
+        };
+    };
+    struct mm_struct *vm_mm;
+    const vm_flags_t vm_flags;
     struct file *vm_file;
+    void *vm_private_data;
+    const struct vm_operations_struct *vm_ops;
 };
 
 typedef unsigned int __kernel_gid32_t;
@@ -642,9 +656,13 @@ struct mm_struct {
     struct {
         long unsigned int arg_start;
         long unsigned int arg_end;
+        unsigned long start_brk;
+        unsigned long brk;
+        unsigned long start_stack;
         long unsigned int env_start;
         long unsigned int env_end;
     };
+    struct task_struct *owner;
 };
 
 struct vfsmount {

pkg/ebpf/probes/probe_group.go

@@ -215,6 +215,9 @@ func NewDefaultProbeGroup(module *bpf.Module, netEnabled bool) (*ProbeGroup, err
 		SignalSchedProcessFork:     NewTraceProbe(RawTracepoint, "sched:sched_process_fork", "sched_process_fork_signal"),
 		SignalSchedProcessExec:     NewTraceProbe(RawTracepoint, "sched:sched_process_exec", "sched_process_exec_signal"),
 		SignalSchedProcessExit:     NewTraceProbe(RawTracepoint, "sched:sched_process_exit", "sched_process_exit_signal"),
+		GetUserPagesRemote:         NewTraceProbe(KProbe, "get_user_pages_remote", "trace_get_user_pages_remote"),
+		GetUserPagesRemoteRet:      NewTraceProbe(KretProbe, "get_user_pages_remote", "trace_ret_get_user_pages_remote"),
+		GenericAccessPhys:          NewTraceProbe(KProbe, "generic_access_phys", "trace_generic_access_phys"),
 	}
 
 	if !netEnabled {

pkg/ebpf/probes/probes.go

@@ -132,4 +132,7 @@ const (
 	SignalSchedProcessFork
 	SignalSchedProcessExec
 	SignalSchedProcessExit
+	GetUserPagesRemote
+	GetUserPagesRemoteRet
+	GenericAccessPhys
 )

pkg/events/core.go

@@ -104,6 +104,7 @@ const (
 	HiddenKernelModuleSeeker
 	ModuleLoad
 	ModuleFree
+	AccessRemoteVm
 	MaxCommonID
 )
 
@@ -11259,6 +11260,33 @@ var CoreEvents = map[ID]Definition{
 			{Type: "bool", Name: "process_group_exit"},
 		},
 	},
+	AccessRemoteVm: {
+		id:      AccessRemoteVm,
+		id32Bit: Sys32Undefined,
+		name:    "access_remote_vm",
+		dependencies: Dependencies{
+			probes: []Probe{
+				{handle: probes.GetUserPagesRemote, required: true},
+				{handle: probes.GetUserPagesRemoteRet, required: true},
+				{handle: probes.GenericAccessPhys, required: true},
+			},
+			kSymbols: []KSymbol{
+				{symbol: "special_mapping_vmops", required: false},
+				{symbol: "legacy_special_mapping_vmops", required: false},
+			},
+		},
+		sets: []string{},
+		params: []trace.ArgMeta{
+			{Type: "int", Name: "remote_pid"},
+			{Type: "void *", Name: "start_address"},
+			{Type: "unsigned int", Name: "gup_flags"},
+			{Type: "unsigned long", Name: "vm_flags"},
+			{Type: "const char*", Name: "mapped.path"},
+			{Type: "dev_t", Name: "mapped.device_id"},
+			{Type: "unsigned long", Name: "mapped.inode_number"},
+			{Type: "unsigned long", Name: "mapped.ctime"},
+		},
+	},
 	//
 	// Begin of Network Protocol Event Types
 	//