xnu-4570.51.1.tar.gz

[apple/xnu.git] / bsd / sys / dtrace_impl.h

diff --git a/bsd/sys/dtrace_impl.h b/bsd/sys/dtrace_impl.h
index 7f42cff5e07a306f2268445a1b7e4327f951c811..125293dbfb084b652ea16865d9b583fcde782222 100644 (file)
--- a/bsd/sys/dtrace_impl.h
+++ b/bsd/sys/dtrace_impl.h
@@ -22,6 +22,9 @@
 /*
  * Copyright 2007 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
 /*
  * Copyright 2007 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.

+ *
+ * Portions Copyright (c) 2012 by Delphix. All rights reserved.
+ * Portions Copyright (c) 2016 by Joyent, Inc.

  */
 
 #ifndef _SYS_DTRACE_IMPL_H
  */
 
 #ifndef _SYS_DTRACE_IMPL_H


@@ -46,6 +49,11 @@ extern "C" {
 
 #include <sys/dtrace.h>
 
 
 #include <sys/dtrace.h>
 

+/*
+ * DTrace Implementation Locks
+ */
+extern lck_mtx_t dtrace_procwaitfor_lock;
+

 /*
  * DTrace Implementation Constants and Typedefs
  */
 /*
  * DTrace Implementation Constants and Typedefs
  */


@@ -197,15 +205,18 @@ typedef struct dtrace_hash {
  * predicate is non-NULL, the DIF object is executed.  If the result is
  * non-zero, the action list is processed, with each action being executed
  * accordingly.  When the action list has been completely executed, processing
  * predicate is non-NULL, the DIF object is executed.  If the result is
  * non-zero, the action list is processed, with each action being executed
  * accordingly.  When the action list has been completely executed, processing

- * advances to the next ECB.  processing advances to the next ECB.  If the
- * result is non-zero; For each ECB, it first determines the The ECB
- * abstraction allows disjoint consumers to multiplex on single probes.
+ * advances to the next ECB. The ECB abstraction allows disjoint consumers
+ * to multiplex on single probes.
+ *
+ * Execution of the ECB results in consuming dte_size bytes in the buffer
+ * to record data.  During execution, dte_needed bytes must be available in
+ * the buffer.  This space is used for both recorded data and tuple data.

  */
 struct dtrace_ecb {
        dtrace_epid_t dte_epid;                 /* enabled probe ID */
        uint32_t dte_alignment;                 /* required alignment */
  */
 struct dtrace_ecb {
        dtrace_epid_t dte_epid;                 /* enabled probe ID */
        uint32_t dte_alignment;                 /* required alignment */

-       size_t dte_needed;                      /* bytes needed */
-       size_t dte_size;                        /* total size of payload */
+       size_t dte_needed;                      /* space needed for execution */
+       size_t dte_size;                        /* size of recorded payload */

        dtrace_predicate_t *dte_predicate;      /* predicate, if any */
        dtrace_action_t *dte_action;            /* actions, if any */
        dtrace_ecb_t *dte_next;                 /* next ECB on probe */
        dtrace_predicate_t *dte_predicate;      /* predicate, if any */
        dtrace_action_t *dte_action;            /* actions, if any */
        dtrace_ecb_t *dte_next;                 /* next ECB on probe */


@@ -263,27 +274,30 @@ typedef struct dtrace_aggregation {
  * the EPID, the consumer can determine the data layout.  (The data buffer
  * layout is shown schematically below.)  By assuring that one can determine
  * data layout from the EPID, the metadata stream can be separated from the
  * the EPID, the consumer can determine the data layout.  (The data buffer
  * layout is shown schematically below.)  By assuring that one can determine
  * data layout from the EPID, the metadata stream can be separated from the

- * data stream -- simplifying the data stream enormously.
- *
- *      base of data buffer --->  +------+--------------------+------+
- *                                | EPID | data               | EPID |
- *                                +------+--------+------+----+------+
- *                                | data          | EPID | data      |
- *                                +---------------+------+-----------+
- *                                | data, cont.                      |
- *                                +------+--------------------+------+
- *                                | EPID | data               |      |
- *                                +------+--------------------+      |
- *                                |                ||                |
- *                                |                ||                |
- *                                |                \/                |
- *                                :                                  :
- *                                .                                  .
- *                                .                                  .
- *                                .                                  .
- *                                :                                  :
- *                                |                                  |
- *     limit of data buffer --->  +----------------------------------+
+ * data stream -- simplifying the data stream enormously.  The ECB always
+ * proceeds the recorded data as part of the dtrace_rechdr_t structure that
+ * includes the EPID and a high-resolution timestamp used for output ordering
+ * consistency.
+ *
+ *      base of data buffer --->  +--------+--------------------+--------+
+ *                                | rechdr | data               | rechdr |
+ *                                +--------+------+--------+----+--------+
+ *                                | data          | rechdr | data        |
+ *                                +---------------+--------+-------------+
+ *                                | data, cont.                          |
+ *                                +--------+--------------------+--------+
+ *                                | rechdr | data               |        |
+ *                                +--------+--------------------+        |
+ *                                |                ||                    |
+ *                                |                ||                    |
+ *                                |                \/                    |
+ *                                :                                      :
+ *                                .                                      .
+ *                                .                                      .
+ *                                .                                      .
+ *                                :                                      :
+ *                                |                                      |
+ *     limit of data buffer --->  +--------------------------------------+

  *
  * When evaluating an ECB, dtrace_probe() determines if the ECB's needs of the
  * principal buffer (both scratch and payload) exceed the available space.  If
  *
  * When evaluating an ECB, dtrace_probe() determines if the ECB's needs of the
  * principal buffer (both scratch and payload) exceed the available space.  If


@@ -408,6 +422,8 @@ typedef struct dtrace_aggregation {
 
 typedef struct dtrace_buffer {
        uint64_t dtb_offset;                    /* current offset in buffer */
 
 typedef struct dtrace_buffer {
        uint64_t dtb_offset;                    /* current offset in buffer */

+       uint64_t dtb_cur_limit;                 /* current limit before signaling/dropping */
+       uint64_t dtb_limit;                     /* limit before signaling */

        uint64_t dtb_size;                      /* size of buffer */
        uint32_t dtb_flags;                     /* flags */
        uint32_t dtb_drops;                     /* number of drops */
        uint64_t dtb_size;                      /* size of buffer */
        uint32_t dtb_flags;                     /* flags */
        uint32_t dtb_drops;                     /* number of drops */


@@ -421,6 +437,9 @@ typedef struct dtrace_buffer {
 #ifndef _LP64
        uint64_t dtb_pad1;
 #endif
 #ifndef _LP64
        uint64_t dtb_pad1;
 #endif

+       uint64_t dtb_switched;                  /* time of last switch */
+       uint64_t dtb_interval;                  /* observed switch interval */
+       uint64_t dtb_pad2[4];                   /* pad to avoid false sharing */

 } dtrace_buffer_t;
 
 /*
 } dtrace_buffer_t;
 
 /*


@@ -892,6 +911,7 @@ typedef struct dtrace_vstate {
 #define        DTRACE_MSTATE_WALLTIMESTAMP     0x00000100
 #define        DTRACE_MSTATE_USTACKDEPTH       0x00000200
 #define        DTRACE_MSTATE_UCALLER           0x00000400
 #define        DTRACE_MSTATE_WALLTIMESTAMP     0x00000100
 #define        DTRACE_MSTATE_USTACKDEPTH       0x00000200
 #define        DTRACE_MSTATE_UCALLER           0x00000400

+#define        DTRACE_MSTATE_MACHTIMESTAMP     0x00000800

 
 typedef struct dtrace_mstate {
        uintptr_t dtms_scratch_base;            /* base of scratch space */
 
 typedef struct dtrace_mstate {
        uintptr_t dtms_scratch_base;            /* base of scratch space */


@@ -902,6 +922,7 @@ typedef struct dtrace_mstate {
        dtrace_epid_t dtms_epid;                /* current EPID */
        uint64_t dtms_timestamp;                /* cached timestamp */
        hrtime_t dtms_walltimestamp;            /* cached wall timestamp */
        dtrace_epid_t dtms_epid;                /* current EPID */
        uint64_t dtms_timestamp;                /* cached timestamp */
        hrtime_t dtms_walltimestamp;            /* cached wall timestamp */

+       uint64_t dtms_machtimestamp;            /* cached mach absolute timestamp */

        int dtms_stackdepth;                    /* cached stackdepth */
        int dtms_ustackdepth;                   /* cached ustackdepth */
        struct dtrace_probe *dtms_probe;        /* current probe */
        int dtms_stackdepth;                    /* cached stackdepth */
        int dtms_ustackdepth;                   /* cached ustackdepth */
        struct dtrace_probe *dtms_probe;        /* current probe */


@@ -910,6 +931,7 @@ typedef struct dtrace_mstate {
        int dtms_ipl;                           /* cached interrupt pri lev */
        int dtms_fltoffs;                       /* faulting DIFO offset */
        uintptr_t dtms_strtok;                  /* saved strtok() pointer */
        int dtms_ipl;                           /* cached interrupt pri lev */
        int dtms_fltoffs;                       /* faulting DIFO offset */
        uintptr_t dtms_strtok;                  /* saved strtok() pointer */

+       uintptr_t dtms_strtok_limit;            /* upper bound of strtok ptr */

        uint32_t dtms_access;                   /* memory access rights */
        dtrace_difo_t *dtms_difo;               /* current dif object */
 } dtrace_mstate_t;
        uint32_t dtms_access;                   /* memory access rights */
        dtrace_difo_t *dtms_difo;               /* current dif object */
 } dtrace_mstate_t;


@@ -937,6 +959,7 @@ typedef struct dtrace_mstate {
  * directed acyclic graph.  The activity transition diagram is as follows:
  *
  *
  * directed acyclic graph.  The activity transition diagram is as follows:
  *
  *

+ *

  * +----------+                   +--------+                   +--------+
  * | INACTIVE |------------------>| WARMUP |------------------>| ACTIVE |
  * +----------+   dtrace_go(),    +--------+   dtrace_go(),    +--------+
  * +----------+                   +--------+                   +--------+
  * | INACTIVE |------------------>| WARMUP |------------------>| ACTIVE |
  * +----------+   dtrace_go(),    +--------+   dtrace_go(),    +--------+


@@ -978,9 +1001,9 @@ typedef enum dtrace_activity {
        DTRACE_ACTIVITY_KILLED                  /* killed */
 } dtrace_activity_t;
 
        DTRACE_ACTIVITY_KILLED                  /* killed */
 } dtrace_activity_t;
 

-#if defined(__APPLE__)
+

 /*
 /*

- * DTrace dof modes
+ * APPLE NOTE:  DTrace dof modes implementation

  *
  * DTrace has four "dof modes". They are:
  *
  *
  * DTrace has four "dof modes". They are:
  *


@@ -1047,7 +1070,6 @@ typedef enum dtrace_activity {
 #define DTRACE_KERNEL_SYMBOLS_FROM_USERSPACE           2
 #define DTRACE_KERNEL_SYMBOLS_ALWAYS_FROM_KERNEL       3
        
 #define DTRACE_KERNEL_SYMBOLS_FROM_USERSPACE           2
 #define DTRACE_KERNEL_SYMBOLS_ALWAYS_FROM_KERNEL       3
        

-#endif /* __APPLE__ */

 
 /*
  * DTrace Helper Implementation
 
 /*
  * DTrace Helper Implementation


@@ -1109,6 +1131,7 @@ typedef struct dtrace_helpers {
 #define        DTRACE_HELPTRACE_DONE   (-2)
 #define        DTRACE_HELPTRACE_ERR    (-3)
 
 #define        DTRACE_HELPTRACE_DONE   (-2)
 #define        DTRACE_HELPTRACE_ERR    (-3)
 

+

 typedef struct dtrace_helptrace {
        dtrace_helper_action_t  *dtht_helper;   /* helper action */
        int dtht_where;                         /* where in helper action */
 typedef struct dtrace_helptrace {
        dtrace_helper_action_t  *dtht_helper;   /* helper action */
        int dtht_where;                         /* where in helper action */


@@ -1202,9 +1225,8 @@ struct dtrace_state {
        dtrace_optval_t dts_options[DTRACEOPT_MAX]; /* options */
        dtrace_cred_t dts_cred;                 /* credentials */
        size_t dts_nretained;                   /* number of retained enabs */
        dtrace_optval_t dts_options[DTRACEOPT_MAX]; /* options */
        dtrace_cred_t dts_cred;                 /* credentials */
        size_t dts_nretained;                   /* number of retained enabs */

-#if defined(__APPLE__)

        uint64_t dts_arg_error_illval;
        uint64_t dts_arg_error_illval;

-#endif /* __APPLE__ */
+       uint32_t dts_buf_over_limit;            /* number of bufs over dtb_limit */

 };
 
 struct dtrace_provider {
 };
 
 struct dtrace_provider {


@@ -1215,13 +1237,15 @@ struct dtrace_provider {
        void *dtpv_arg;                         /* provider argument */
        uint_t dtpv_defunct;                    /* boolean: defunct provider */
        struct dtrace_provider *dtpv_next;      /* next provider */
        void *dtpv_arg;                         /* provider argument */
        uint_t dtpv_defunct;                    /* boolean: defunct provider */
        struct dtrace_provider *dtpv_next;      /* next provider */

+       uint64_t dtpv_probe_count;              /* number of associated probes */
+       uint64_t dtpv_ecb_count;                /* number of associated enabled ECBs */

 };
 
 struct dtrace_meta {
        dtrace_mops_t dtm_mops;                 /* meta provider operations */
        char *dtm_name;                         /* meta provider name */
        void *dtm_arg;                          /* meta provider user arg */
 };
 
 struct dtrace_meta {
        dtrace_mops_t dtm_mops;                 /* meta provider operations */
        char *dtm_name;                         /* meta provider name */
        void *dtm_arg;                          /* meta provider user arg */

-       uint64_t dtm_count;                     /* no. of associated provs. */
+       uint64_t dtm_count;                     /* number of associated providers */

 };
 
 /*
 };
 
 /*


@@ -1286,6 +1310,18 @@ typedef struct dtrace_errhash {
 
 #endif /* DTRACE_ERRDEBUG */
 
 
 #endif /* DTRACE_ERRDEBUG */
 

+/**
+ * DTrace Matching pre-conditions
+ *
+ * Used when matching new probes to discard matching of enablings that
+ * doesn't match the condition tested by dmc_func
+ */
+typedef struct dtrace_match_cond {
+       int (*dmc_func)(dtrace_probedesc_t*, void*);
+       void *dmc_data;
+} dtrace_match_cond_t;
+
+

 /*
  * DTrace Toxic Ranges
  *
 /*
  * DTrace Toxic Ranges
  *


@@ -1307,76 +1343,76 @@ typedef struct dtrace_toxrange {
        uintptr_t       dtt_limit;              /* limit of toxic range */
 } dtrace_toxrange_t;
 
        uintptr_t       dtt_limit;              /* limit of toxic range */
 } dtrace_toxrange_t;
 

-extern uint64_t dtrace_getarg(int, int);
+extern uint64_t dtrace_getarg(int, int, dtrace_mstate_t*, dtrace_vstate_t*);

 extern int dtrace_getipl(void);
 extern uintptr_t dtrace_caller(int);
 extern uint32_t dtrace_cas32(uint32_t *, uint32_t, uint32_t);
 extern void *dtrace_casptr(void *, void *, void *);
 extern int dtrace_getipl(void);
 extern uintptr_t dtrace_caller(int);
 extern uint32_t dtrace_cas32(uint32_t *, uint32_t, uint32_t);
 extern void *dtrace_casptr(void *, void *, void *);

-#if !defined(__APPLE__)
-extern void dtrace_copyin(uintptr_t, uintptr_t, size_t, volatile uint16_t *);
-extern void dtrace_copyinstr(uintptr_t, uintptr_t, size_t, volatile uint16_t *);
-extern void dtrace_copyout(uintptr_t, uintptr_t, size_t, volatile uint16_t *);
-extern void dtrace_copyoutstr(uintptr_t, uintptr_t, size_t,
-    volatile uint16_t *);
-#else

 extern void dtrace_copyin(user_addr_t, uintptr_t, size_t, volatile uint16_t *);
 extern void dtrace_copyinstr(user_addr_t, uintptr_t, size_t, volatile uint16_t *);
 extern void dtrace_copyout(uintptr_t, user_addr_t, size_t, volatile uint16_t *);
 extern void dtrace_copyoutstr(uintptr_t, user_addr_t, size_t, volatile uint16_t *);
 extern void dtrace_copyin(user_addr_t, uintptr_t, size_t, volatile uint16_t *);
 extern void dtrace_copyinstr(user_addr_t, uintptr_t, size_t, volatile uint16_t *);
 extern void dtrace_copyout(uintptr_t, user_addr_t, size_t, volatile uint16_t *);
 extern void dtrace_copyoutstr(uintptr_t, user_addr_t, size_t, volatile uint16_t *);

-#endif /* __APPLE__ */

 extern void dtrace_getpcstack(pc_t *, int, int, uint32_t *);
 extern void dtrace_getpcstack(pc_t *, int, int, uint32_t *);

-#if !defined(__APPLE__)
-extern ulong_t dtrace_getreg(struct regs *, uint_t);
-#else
+extern uint64_t dtrace_load64(uintptr_t);
+extern int dtrace_canload(uint64_t, size_t, dtrace_mstate_t*, dtrace_vstate_t*);
+

 extern uint64_t dtrace_getreg(struct regs *, uint_t);
 extern uint64_t dtrace_getreg(struct regs *, uint_t);

-#endif /* __APPLE__ */

 extern int dtrace_getstackdepth(int);
 extern void dtrace_getupcstack(uint64_t *, int);
 extern void dtrace_getufpstack(uint64_t *, uint64_t *, int);
 extern int dtrace_getustackdepth(void);
 extern uintptr_t dtrace_fulword(void *);
 extern int dtrace_getstackdepth(int);
 extern void dtrace_getupcstack(uint64_t *, int);
 extern void dtrace_getufpstack(uint64_t *, uint64_t *, int);
 extern int dtrace_getustackdepth(void);
 extern uintptr_t dtrace_fulword(void *);

-#if !defined(__APPLE__)
-extern uint8_t dtrace_fuword8(void *);
-extern uint16_t dtrace_fuword16(void *);
-extern uint32_t dtrace_fuword32(void *);
-extern uint64_t dtrace_fuword64(void *);
-extern void dtrace_probe_error(dtrace_state_t *, dtrace_epid_t, int, int,
-    int, uintptr_t);
-#else

 extern uint8_t dtrace_fuword8(user_addr_t);
 extern uint16_t dtrace_fuword16(user_addr_t);
 extern uint32_t dtrace_fuword32(user_addr_t);
 extern uint64_t dtrace_fuword64(user_addr_t);
 extern uint8_t dtrace_fuword8(user_addr_t);
 extern uint16_t dtrace_fuword16(user_addr_t);
 extern uint32_t dtrace_fuword32(user_addr_t);
 extern uint64_t dtrace_fuword64(user_addr_t);

+extern int dtrace_proc_waitfor(dtrace_procdesc_t*);

 extern void dtrace_probe_error(dtrace_state_t *, dtrace_epid_t, int, int,
     int, uint64_t);
 extern void dtrace_probe_error(dtrace_state_t *, dtrace_epid_t, int, int,
     int, uint64_t);

-#endif /* __APPLE__ */

 extern int dtrace_assfail(const char *, const char *, int);
 extern int dtrace_attached(void);
 extern hrtime_t dtrace_gethrestime(void);
 extern int dtrace_assfail(const char *, const char *, int);
 extern int dtrace_attached(void);
 extern hrtime_t dtrace_gethrestime(void);

+extern void dtrace_isa_init(void);

 
 

-#ifdef __sparc
-extern void dtrace_flush_windows(void);
-extern void dtrace_flush_user_windows(void);
-extern uint_t dtrace_getotherwin(void);
-extern uint_t dtrace_getfprs(void);
-#else

 extern void dtrace_copy(uintptr_t, uintptr_t, size_t);
 extern void dtrace_copystr(uintptr_t, uintptr_t, size_t, volatile uint16_t *);
 extern void dtrace_copy(uintptr_t, uintptr_t, size_t);
 extern void dtrace_copystr(uintptr_t, uintptr_t, size_t, volatile uint16_t *);

-#endif
+
+/*
+ * DTrace state handling
+ */
+extern minor_t dtrace_state_reserve(void);
+extern dtrace_state_t* dtrace_state_allocate(minor_t minor);
+extern dtrace_state_t* dtrace_state_get(minor_t minor);
+extern void dtrace_state_free(minor_t minor);
+
+/*
+ * DTrace restriction checks
+ */
+extern void dtrace_restriction_policy_load(void);
+extern boolean_t dtrace_is_restricted(void);
+extern boolean_t dtrace_are_restrictions_relaxed(void);
+extern boolean_t dtrace_fbt_probes_restricted(void);
+extern boolean_t dtrace_sdt_probes_restricted(void);
+extern boolean_t dtrace_can_attach_to_proc(proc_t);

 
 /*
  * DTrace Assertions
  *
 
 /*
  * DTrace Assertions
  *

- * DTrace calls ASSERT from probe context.  To assure that a failed ASSERT
- * does not induce a markedly more catastrophic failure (e.g., one from which
- * a dump cannot be gleaned), DTrace must define its own ASSERT to be one that
- * may safely be called from probe context.  This header file must thus be
- * included by any DTrace component that calls ASSERT from probe context, and
- * _only_ by those components.  (The only exception to this is kernel
- * debugging infrastructure at user-level that doesn't depend on calling
- * ASSERT.)
+ * DTrace calls ASSERT and VERIFY from probe context.  To assure that a failed
+ * ASSERT or VERIFYdoes not induce a markedly more catastrophic failure (e.g.,
+ * one from which a dump cannot be gleaned), DTrace must define its own ASSERT
+ * and VERIFY macros to be ones that may safely be called from probe context.
+ * This header file must thus be included by any DTrace component that calls
+ * ASSERT and/or VERIFY from probe context, and _only_ by those components.
+ * (The only exception to this is kernel debugging infrastructure at user-level
+ * that doesn't depend on calling ASSERT.)

  */
 #undef ASSERT
  */
 #undef ASSERT

+#undef VERIFY
+
+#define        VERIFY(EX)      ((void)((EX) || \
+                       dtrace_assfail(#EX, __FILE__, __LINE__)))
+

 #if DEBUG
 #define        ASSERT(EX)      ((void)((EX) || \
                        dtrace_assfail(#EX, __FILE__, __LINE__)))
 #if DEBUG
 #define        ASSERT(EX)      ((void)((EX) || \
                        dtrace_assfail(#EX, __FILE__, __LINE__)))