dynamorio/core/os_shared.h

/* **********************************************************
 * Copyright (c) 2010-2026 Google, Inc.  All rights reserved.
 * Copyright (c) 2003-2010 VMware, Inc.  All rights reserved.
 * **********************************************************/

/*
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions are met:
 *
 * * Redistributions of source code must retain the above copyright notice,
 *   this list of conditions and the following disclaimer.
 *
 * * Redistributions in binary form must reproduce the above copyright notice,
 *   this list of conditions and the following disclaimer in the documentation
 *   and/or other materials provided with the distribution.
 *
 * * Neither the name of VMware, Inc. nor the names of its contributors may be
 *   used to endorse or promote products derived from this software without
 *   specific prior written permission.
 *
 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
 * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
 * ARE DISCLAIMED. IN NO EVENT SHALL VMWARE, INC. OR CONTRIBUTORS BE LIABLE
 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
 * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
 * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH
 * DAMAGE.
 */

/* Copyright (c) 2003-2007 Determina Corp. */

/*
 * os_shared.h - shared declarations for os facilities from unix/win32
 */

#ifndef OS_SHARED_H
#define OS_SHARED_H

#include "os_api.h"

/* Keep in sync with the ptrace_attach.c OS_SRCS guard in core/CMakeLists.txt. */
#if defined(LINUX) && !defined(ANDROID) && (defined(AARCH64) || defined(X86))
#    define PTRACE_TAKEOVER_SUPPORTED 1
#endif

enum { VM_ALLOCATION_BOUNDARY = 64 * 1024 }; /* 64KB allocation size for Windows */

struct _local_state_t; /* in arch_exports.h */

/* in os.c */
void
d_r_os_init(void);
/* called on detach and on process exit in debug builds */
void
os_slow_exit(void);
/* called on detach and on process exit in both debug and release builds */
void
os_fast_exit(void);

void
os_tls_init(void);
/* Frees local_state.  If the calling thread is exiting (i.e.,
 * !other_thread) then also frees kernel resources for the calling
 * thread; if other_thread then that may not be possible.
 */
void
os_tls_exit(struct _local_state_t *local_state, bool other_thread);
/* os_data is passed through from dynamo_thread_init */
void
os_thread_init(dcontext_t *dcontext, void *os_data);
/* Called right before giving up thread_initexit_lock, after other components
 * like synch have initialized.
 */
void
os_thread_init_finalize(dcontext_t *dcontext, void *os_data);
void
os_thread_exit(dcontext_t *dcontext, bool other_thread);

/* must only be called for the executing thread */
void
os_thread_under_dynamo(dcontext_t *dcontext);
/* must only be called for the executing thread */
void
os_thread_not_under_dynamo(dcontext_t *dcontext, bool restore_sigblocked);
void
os_process_under_dynamorio_initiate(dcontext_t *dcontext);
void
os_process_under_dynamorio_complete(dcontext_t *dcontext);
void
os_process_not_under_dynamorio(dcontext_t *dcontext);

bool
os_take_over_all_unknown_threads(dcontext_t *dcontext);

#ifdef PTRACE_TAKEOVER_SUPPORTED
bool
os_unmask_suspend_signal_via_ptrace(thread_id_t skip_tid);
#endif

bool
detach_do_not_translate(thread_record_t *tr);
void
detach_finalize_translation(thread_record_t *tr, priv_mcontext_t *mc);
void
detach_finalize_cleanup(void);

void
os_heap_init(void);
void
os_heap_exit(void);

/* os provided heap routines */
/* caller is required to handle thread synchronization and to update dynamo vm areas.
 * size must be PAGE_SIZE-aligned.
 * returns NULL if fails to allocate memory!
 * N.B.: only heap.c should call these routines, as it contains the logic to
 * handle memory allocation failure!
 *
 * error_code is set to 0 on success
 * on failure it is set to one of the other enum values below or an OS specific status
 * code used only for reporting
 */
enum {
    HEAP_ERROR_SUCCESS = 0,
    /* os_heap_reserve_in_region() only, couldn't find a place to reserve within region */
    HEAP_ERROR_CANT_RESERVE_IN_REGION = 1,
    /* os_heap_reserve() only, Linux only, mmap failed to reserve at preferred address */
    HEAP_ERROR_NOT_AT_PREFERRED = 2,
};
typedef uint heap_error_code_t;

/* flag values for os_raw_mem_alloc */
enum {
#ifdef WINDOWS
    /* default is reserve+commit */
    RAW_ALLOC_RESERVE_ONLY = 0x0001,
    RAW_ALLOC_COMMIT_ONLY = 0x0002,
#endif
#ifdef UNIX
    RAW_ALLOC_32BIT = 0x0004,
#endif
};

/* For dr_raw_mem_alloc, try to allocate memory at preferred address. */
void *
os_raw_mem_alloc(void *preferred, size_t size, uint prot, uint flags,
                 heap_error_code_t *error_code);
/* For dr_raw_mem_free, free memory allocated from os_raw_mem_alloc */
bool
os_raw_mem_free(void *p, size_t size, uint flags, heap_error_code_t *error_code);

/* Reserve size bytes of virtual address space in one piece without committing swap
 * space for it.  If preferred is non-NULL then memory will be reserved at that address
 * only (if size bytes are unavailable at preferred then the allocation will fail).
 * The executable flag is for platforms that divide executable from data memory.
 */
void *
os_heap_reserve(void *preferred, size_t size, heap_error_code_t *error_code,
                bool executable);
/* Reserve size bytes of virtual address space in one piece entirely within the
 * address range specified without committing swap space for it.
 * The executable flag is for platforms that divide executable from data memory.
 */
void *
os_heap_reserve_in_region(void *start, void *end, size_t size,
                          heap_error_code_t *error_code, bool executable);
/* commit previously reserved pages, returns false when out of memory */
bool
os_heap_commit(void *p, size_t size, uint prot, heap_error_code_t *error_code);
/* decommit previously committed page, so it is reserved for future reuse */
void
os_heap_decommit(void *p, size_t size, heap_error_code_t *error_code);
/* frees size bytes starting at address p (note - on windows the entire allocation
 * containing p is freed and size is ignored) */
void
os_heap_free(void *p, size_t size, heap_error_code_t *error_code);

/* prognosticate whether systemwide memory pressure based on
 * last_error_code and systemwide omens
 *
 * Note it doesn't answer if the hog is current process, nor whether
 * it is our own fault in current process.
 */
bool
os_heap_systemwide_overcommit(heap_error_code_t last_error_code);

bool
os_heap_get_commit_limit(size_t *commit_used, size_t *commit_limit);

thread_id_t
d_r_get_thread_id(void);
process_id_t
get_process_id(void);
void
os_thread_yield(void);
void
os_thread_sleep(uint64 milliseconds);
bool
os_thread_suspend(thread_record_t *tr);
bool
os_thread_resume(thread_record_t *tr);
bool
os_thread_terminate(thread_record_t *tr);

bool
is_thread_currently_native(thread_record_t *tr);

/* If state beyond that in our priv_mcontext_t is needed, os-specific routines
 * must be used.  These only deal with priv_mcontext_t state.
 */
bool
thread_get_mcontext(thread_record_t *tr, priv_mcontext_t *mc);

#ifdef LINUX
bool
thread_get_nudged_mcontext(thread_record_t *tr, priv_mcontext_t *mc);
#endif

bool
thread_set_mcontext(thread_record_t *tr, priv_mcontext_t *mc);

/* Takes an os-specific context. Does not return. */
void
thread_set_self_context(void *cxt, bool is_detach_external);
/* Only sets the priv_mcontext_t state.  Does not return. */
void
thread_set_self_mcontext(priv_mcontext_t *mc, bool is_detach_external);

/* Assumes target thread is suspended */
bool
os_thread_take_over_suspended_native(dcontext_t *dcontext);

/* Initializes the executing thread by calling dynamo_thread_init(). */
dcontext_t *
os_thread_take_over_secondary(priv_mcontext_t *mc);

/* Readies a known but currently-native thread for takeover.
 * Returns whether the thread is known.
 */
bool
os_thread_re_take_over(void);

dcontext_t *
get_thread_private_dcontext(void);
void
set_thread_private_dcontext(dcontext_t *dcontext);

/* converts a local_state_t offset to a segment offset */
ushort
os_tls_offset(ushort tls_offs);

ushort
os_local_state_offset(ushort seg_offs);

struct _local_state_t;          /* in arch_exports.h */
struct _local_state_extended_t; /* in arch_exports.h */
struct _local_state_t *
get_local_state(void);
struct _local_state_extended_t *
get_local_state_extended(void);

/* Returns POINTER_MAX on failure.
 * On X86, we assume that cs, ss, ds, and es are flat.
 * On ARM, there is no segment. We use it to get TLS base instead.
 */
byte *
get_segment_base(uint seg);

byte *
get_app_segment_base(uint seg);

/* Allocates num_slots tls slots aligned with alignment align */
bool
os_tls_calloc(DR_PARAM_OUT uint *offset, uint num_slots, uint alignment);

bool
os_tls_cfree(uint offset, uint num_slots);

bool
os_should_swap_state(void);

bool
os_using_app_state(dcontext_t *dcontext);

void
os_swap_context(dcontext_t *dcontext, bool to_app, dr_state_flags_t flags);

bool
pre_system_call(dcontext_t *dcontext);
void
post_system_call(dcontext_t *dcontext);
int
os_normalized_sysnum(int num_raw, instr_t *gateway, dcontext_t *dcontext_live);

char *
get_application_pid(void);
char *
get_application_name(void);

int
num_app_args(void);
int
get_app_args(DR_PARAM_OUT dr_app_arg_t *args_array, int args_count);
const char *
get_application_short_name(void);
char *
get_computer_name(void); /* implemented on win32 only, in eventlog.c */

app_pc
get_application_base(void);

app_pc
get_application_end(void);

int
get_num_processors(void);

/* Terminate types - the best choice here is TERMINATE_PROCESS, no cleanup*/
typedef enum {
    TERMINATE_PROCESS = 0x1,
    TERMINATE_THREAD = 0x2, /* terminate thread (and process if last) */
    /* Use of TERMINATE_THREAD mode is dangerous,
       and can result in the following problems (see MSDN):
       * If the target thread owns a critical section,
       the critical section will not be released.
       * If the target thread is allocating memory from the heap,
       the heap lock will not be released.
       * If the target thread is executing certain kernel32 calls when it is terminated,
       the kernel32 state for the thread's process could be inconsistent.
       * If the target thread is manipulating the global state of a shared DLL,
       the state of the DLL could be destroyed, affecting other users of the DLL.
    */
    TERMINATE_CLEANUP = 0x4 /* cleanup our state before issuing terminal syscall */
} terminate_flags_t;

void
os_terminate(dcontext_t *dcontext, terminate_flags_t flags);
void
os_terminate_with_code(dcontext_t *dcontext, terminate_flags_t flags, int exit_code);

typedef enum {
    ILLEGAL_INSTRUCTION_EXCEPTION,
    UNREADABLE_MEMORY_EXECUTION_EXCEPTION,
    IN_PAGE_ERROR_EXCEPTION,
    GUARD_PAGE_EXCEPTION,
    SINGLE_STEP_EXCEPTION,
} dr_exception_type_t;

void
os_forge_exception(app_pc exception_address, dr_exception_type_t type);

/* events for dumpcore_mask */
/* NOTE with DUMPCORE_DEADLOCK and DUMPCORE_ASSERT you will get 2 dumps for
 * rank order violations in debug builds */
enum {
    DUMPCORE_INTERNAL_EXCEPTION = 0x0001,
    DUMPCORE_SECURITY_VIOLATION = 0x0002,
    DUMPCORE_DEADLOCK = 0x0004,
    DUMPCORE_ASSERTION = 0x0008,
    DUMPCORE_FATAL_USAGE_ERROR = 0x0010,
    DUMPCORE_CLIENT_EXCEPTION = 0x0020,
    DUMPCORE_TIMEOUT = 0x0040,
    DUMPCORE_CURIOSITY = 0x0080,
#ifdef HOT_PATCHING_INTERFACE /* Part of fix for 5357 & 5988. */
    /* Errors and exceptions in hot patches are treated the same.  Case 5696. */
    DUMPCORE_HOTP_FAILURE = 0x0100,
#endif
    DUMPCORE_OUT_OF_MEM = 0x0200,
    DUMPCORE_OUT_OF_MEM_SILENT = 0x0400, /* not on by default in DEBUG */
#ifdef UNIX
    DUMPCORE_INCLUDE_STACKDUMP = 0x0800,
    DUMPCORE_WAIT_FOR_DEBUGGER = 0x1000, /* not on by default in DEBUG */
#endif
#ifdef HOT_PATCHING_INTERFACE          /* Part of fix for 5988. */
    DUMPCORE_HOTP_DETECTION = 0x2000,  /* not on by default in DEBUG */
    DUMPCORE_HOTP_PROTECTION = 0x4000, /* not on by default in DEBUG */
#endif
    DUMPCORE_DR_ABORT = 0x8000,
    /* all exception cases here are off by default since we expect them to
     * usually be the app's fault (or normal behavior for the app) */
    DUMPCORE_FORGE_ILLEGAL_INST = 0x10000,
    DUMPCORE_FORGE_UNREAD_EXEC = 0x20000, /* not including -throw_exception */
    /* Note this is ALL app exceptions (including those the app may expect and
     * handle without issue) except those created via RaiseException (note
     * dr forged exceptions use the eqv. of RaiseException).  Would be nice to
     * have a flag for just unhandled app exceptions but that's harder to
     * implement. */
    DUMPCORE_APP_EXCEPTION = 0x40000,
    DUMPCORE_TRY_EXCEPT = 0x80000, /* even when we do have a handler */
    DUMPCORE_UNSUPPORTED_APP = 0x100000,
    DUMPCORE_STACK_OVERFLOW = 0x200000, /* modifies DUMPCORE_INTERNAL_EXCEPTION */

#ifdef UNIX
    DUMPCORE_OPTION_PAUSE = DUMPCORE_WAIT_FOR_DEBUGGER | DUMPCORE_INTERNAL_EXCEPTION |
        DUMPCORE_SECURITY_VIOLATION | DUMPCORE_DEADLOCK | DUMPCORE_ASSERTION |
        DUMPCORE_FATAL_USAGE_ERROR | DUMPCORE_CLIENT_EXCEPTION |
        DUMPCORE_UNSUPPORTED_APP | DUMPCORE_TIMEOUT | DUMPCORE_CURIOSITY |
        DUMPCORE_DR_ABORT | DUMPCORE_OUT_OF_MEM | DUMPCORE_OUT_OF_MEM_SILENT,
#endif
};
void
os_dump_core(const char *msg);

int
os_timeout(int time_in_milliseconds);

void
os_syslog(syslog_event_type_t priority, uint message_id, uint substitutions_num,
          va_list args);

/* Note that this is NOT identical to module_handle_t: on Linux this
 * is a pointer to a loader data structure and NOT the base address
 * (xref PR 366195).
 * XXX: we're duplicating these types above as dr_auxlib*
 */
typedef void *shlib_handle_t;
typedef void (*shlib_routine_ptr_t)(void);

shlib_handle_t
load_shared_library(const char *name, bool reachable);

shlib_routine_ptr_t
lookup_library_routine(shlib_handle_t lib, const char *name);
void
unload_shared_library(shlib_handle_t lib);
void
shared_library_error(char *buf, int maxlen);
/* addr is any pointer known to lie within the library
 * for linux, one of addr or name is needed; for windows, neither is needed.
 */
bool
shared_library_bounds(DR_PARAM_IN shlib_handle_t lib, DR_PARAM_IN byte *addr,
                      DR_PARAM_IN const char *name, DR_PARAM_OUT byte **start,
                      DR_PARAM_OUT byte **end);
char *
get_dynamorio_library_path(void);

#define MEMPROT_NONE DR_MEMPROT_NONE
#define MEMPROT_READ DR_MEMPROT_READ
#define MEMPROT_WRITE DR_MEMPROT_WRITE
#define MEMPROT_EXEC DR_MEMPROT_EXEC
#ifdef WINDOWS
#    define MEMPROT_GUARD DR_MEMPROT_GUARD
#else
#    define MEMPROT_VDSO DR_MEMPROT_VDSO
#endif
/* i#1861: avoid merging Android regions w/ different custom comments */
#define MEMPROT_HAS_COMMENT DR_MEMPROT_GUARD /* Android-only */
#define MEMPROT_META_FLAGS (MEMPROT_VDSO | MEMPROT_HAS_COMMENT)

/* Ignore any PAGE_SIZE provided by the tool chain and define a new
 * version for DynamoRIO's internal use. Since PAGE_SIZE looks like
 * it might be a constant, in new code it would be better to use an
 * explicit function call.
 */
#undef PAGE_SIZE
#define PAGE_SIZE os_page_size()

/* Convenience macro to align to the start of a page of memory.
 * It uses a function call so be careful where performance is critical.
 */
#define PAGE_START(x) (((ptr_uint_t)(x)) & ~(os_page_size() - 1))
#define PAGE_START64(x) (((uint64)(x)) & ~((uint64)os_page_size() - 1))

size_t
os_page_size(void);
#ifdef UNIX
/* This also tries to set other auxv values. */
void
os_page_size_init(const char **env, bool env_followed_by_auxv);
size_t
os_minsigstksz(void);
#endif
bool
get_memory_info(const byte *pc, byte **base_pc, size_t *size, uint *prot);
bool
query_memory_ex(const byte *pc, DR_PARAM_OUT dr_mem_info_t *info);
/* We provide this b/c getting the bounds is expensive on Windows (i#1462) */
bool
query_memory_cur_base(const byte *pc, DR_PARAM_OUT dr_mem_info_t *info);
#ifdef UNIX
bool
get_memory_info_from_os(const byte *pc, byte **base_pc, size_t *size, uint *prot);
bool
query_memory_ex_from_os(const byte *pc, DR_PARAM_OUT dr_mem_info_t *info);
void
os_check_new_app_module(dcontext_t *dcontext, app_pc pc);
#endif

bool
get_stack_bounds(dcontext_t *dcontext, byte **base, byte **top);

/* Does a safe_read of *src_ptr into dst_var, returning true for success.  We
 * assert that the size of dst and src match.  The other advantage over plain
 * safe_read is that the caller doesn't need to pass sizeof(dst), which is
 * useful for repeated small memory accesses.
 */
#define SAFE_READ_VAL(dst_var, src_ptr)           \
    (ASSERT(sizeof(dst_var) == sizeof(*src_ptr)), \
     d_r_safe_read(src_ptr, sizeof(dst_var), &dst_var))

bool
is_readable_without_exception(const byte *pc, size_t size);
bool
is_readable_without_exception_query_os(byte *pc, size_t size);
bool
is_readable_without_exception_query_os_noblock(byte *pc, size_t size);
bool
d_r_safe_read(const void *base, size_t size, void *out_buf);
bool
safe_read_ex(const void *base, size_t size, void *out_buf, size_t *bytes_read);
bool
safe_write_ex(void *base, size_t size, const void *in_buf, size_t *bytes_written);
bool
is_user_address(byte *pc);
/* returns osprot flags preserving all native protection flags except
 * for RWX, which are replaced according to memprot */
uint
osprot_replace_memprot(uint old_osprot, uint memprot);

/* returns false if out of memory */
bool
set_protection(byte *pc, size_t size, uint prot);
/* Change protections on memory region starting at pc of length size
 * (padded to page boundaries).  This method is meant to be used on DR memory
 * as part of protect from app and is safe with respect to stats and changing
 * the protection of the data segment. */
/* returns false if out of memory */
bool
change_protection(byte *pc, size_t size, bool writable);
#ifdef WINDOWS
/* makes pc:pc+size (page_padded) writable preserving other flags
 * returns false if out of memory
 */
bool
make_hookable(byte *pc, size_t size, bool *changed_prot);
/* if changed_prot makes pc:pc+size (page padded) unwritable preserving
 * other flags */
void
make_unhookable(byte *pc, size_t size, bool changed_prot);
#endif
/* requires that pc is page aligned and size is multiple of the page size
 * and marks that memory writable, preserves other flags,
 * returns false if out of memory
 */
bool
make_writable(byte *pc, size_t size);
/* requires that pc is page aligned and size is multiple of the page size
 * and marks that memory NOT writable, preserves other flags */
void
make_unwritable(byte *pc, size_t size);
/* like make_writable but adds COW (note: only usable if allocated COW) */
bool
make_copy_on_writable(byte *pc, size_t size);

/***************************************************************************
 * SELF_PROTECTION
 */

/* Values for protect_mask to specify what is write-protected from malicious or
 * inadvertent modification by the application.
 * DATA_CXTSW and GLOBAL are done on each context switch
 * the rest are on-demand:
 * DATASEGMENT, DATA_FREQ, and GENCODE only on the rare occasions when we write to them
 * CACHE only when emitting or {,un}linking
 * LOCAL only on path in DR that needs to write to local
 */
enum {
    /* Options specifying protection of our DR dll data sections: */
    /* .data == variables written only at init or exit time or rarely in between */
    SELFPROT_DATA_RARE = 0x001,
    /* .fspdata == frequently written enough that we separate from .data.
     * XXX case 8073: currently these are unprotected on every cxt switch
     */
    SELFPROT_DATA_FREQ = 0x002,
    /* .cspdata == so frequently written that to protect them requires unprotecting
     * every context switch.
     */
    SELFPROT_DATA_CXTSW = 0x004,

    /* if GLOBAL && !DCONTEXT, entire dcontext is unprotected, rest of
     *   global allocs are protected;
     * if GLOBAL && DCONTEXT, cache-written fields of dcontext are unprotected,
     *   rest are protected;
     * if !GLOBAL, DCONTEXT should not be used
     */
    SELFPROT_GLOBAL = 0x008,
    SELFPROT_DCONTEXT = 0x010, /* means we split out unprotected_context_t --
                                * no actual protection unless SELFPROT_GLOBAL */
    SELFPROT_LOCAL = 0x020,
    SELFPROT_CACHE = 0x040, /* TODO: thread-safe NYI when doing all units */
    SELFPROT_STACK = 0x080, /* essentially always on with clean-dstack d_r_dispatch()
                             * design, leaving as a bit in case we do more later */
    /* protect our generated thread-shared and thread-private code */
    SELFPROT_GENCODE = 0x100,
    /* XXX: TEB page on Win32
     * Other global structs, like thread-local callbacks on Win32?
     * PEB page?
     */
    /* options that require action on every context switch
     * XXX: global heap used to be much rarer before shared
     * fragments, only containing "important" data, which is why we
     * un-protected on every context switch.  We should re-think that
     * now that most things are shared.
     */
    SELFPROT_ON_CXT_SWITCH = (SELFPROT_DATA_CXTSW |
                              SELFPROT_GLOBAL
                              /* XXX case 8073: this is only temporary until
                               * we finish implementing .fspdata unprots */
                              | SELFPROT_DATA_FREQ),
    SELFPROT_ANY_DATA_SECTION =
        (SELFPROT_DATA_RARE | SELFPROT_DATA_FREQ | SELFPROT_DATA_CXTSW),
};

/* Values to refer to individual data sections.  The order is not
 * important here.
 */
enum {
    DATASEC_NEVER_PROT = 0,
    DATASEC_RARELY_PROT,
    DATASEC_FREQ_PROT,
    DATASEC_CXTSW_PROT,
    DATASEC_NUM,
};

/* in dynamo.c */
extern const uint DATASEC_SELFPROT[];
extern const char *const DATASEC_NAMES[];
/* see dynamo.c for why this is not an array */
extern const uint datasec_writable_neverprot;
extern uint datasec_writable_rareprot;
extern uint datasec_writable_freqprot;
extern uint datasec_writable_cxtswprot;

/* this is a uint not a bool */
#define DATASEC_WRITABLE(which)                                             \
    ((which) == DATASEC_RARELY_PROT                                         \
         ? datasec_writable_rareprot                                        \
         : ((which) == DATASEC_CXTSW_PROT                                   \
                ? datasec_writable_cxtswprot                                \
                : ((which) == DATASEC_FREQ_PROT ? datasec_writable_freqprot \
                                                : datasec_writable_neverprot)))

/* these must be plain literals since we need these in pragmas/attributes */
#define NEVER_PROTECTED_SECTION ".nspdata"
#define RARELY_PROTECTED_SECTION ".data"
#define FREQ_PROTECTED_SECTION ".fspdata"
#define CXTSW_PROTECTED_SECTION ".cspdata"

/* note that asserting !protected is safe, but asserting protection
 * is racy as another thread could be in an unprot window.
 * see also check_should_be_protected().
 */
#define DATASEC_PROTECTED(which) (DATASEC_WRITABLE(which) == 0)

/* User MUST supply an init value, or else cl will leave in .bss (will show up
 * at end of .data), which is why we hardcode the = here!
 * We use varargs to allow commas in the init if a struct.
 */
#define DECLARE_FREQPROT_VAR(var, ...)                        \
    START_DATA_SECTION(FREQ_PROTECTED_SECTION, "w")           \
    var VAR_IN_SECTION(FREQ_PROTECTED_SECTION) = __VA_ARGS__; \
    END_DATA_SECTION()

#define DECLARE_CXTSWPROT_VAR(var, ...)                        \
    START_DATA_SECTION(CXTSW_PROTECTED_SECTION, "w")           \
    var VAR_IN_SECTION(CXTSW_PROTECTED_SECTION) = __VA_ARGS__; \
    END_DATA_SECTION()

#define DECLARE_NEVERPROT_VAR(var, ...)                        \
    START_DATA_SECTION(NEVER_PROTECTED_SECTION, "w")           \
    var VAR_IN_SECTION(NEVER_PROTECTED_SECTION) = __VA_ARGS__; \
    END_DATA_SECTION()

#define SELF_PROTECT_ON_CXT_SWITCH                                   \
    (TESTANY(SELFPROT_ON_CXT_SWITCH, DYNAMO_OPTION(protect_mask)) || \
     INTERNAL_OPTION(single_privileged_thread))

/* macros to put mask check outside of function, for efficiency */
#define SELF_PROTECT_LOCAL(dc, w)                                 \
    do {                                                          \
        if (TESTANY(SELFPROT_LOCAL, dynamo_options.protect_mask)) \
            protect_local_heap(dc, w);                            \
    } while (0);

#define SELF_PROTECT_GLOBAL(w)                                     \
    do {                                                           \
        if (TESTANY(SELFPROT_GLOBAL, dynamo_options.protect_mask)) \
            protect_global_heap(w);                                \
    } while (0);

#define ASSERT_LOCAL_HEAP_PROTECTED(dcontext)                       \
    ASSERT(!TESTANY(SELFPROT_LOCAL, dynamo_options.protect_mask) || \
           local_heap_protected(dcontext))
#define ASSERT_LOCAL_HEAP_UNPROTECTED(dcontext)                     \
    ASSERT(!TESTANY(SELFPROT_LOCAL, dynamo_options.protect_mask) || \
           !local_heap_protected(dcontext))

#define SELF_PROTECT_DATASEC(which)                                        \
    do {                                                                   \
        if (TESTANY(DATASEC_SELFPROT[which], DYNAMO_OPTION(protect_mask))) \
            protect_data_section(which, READONLY);                         \
    } while (0);
#define SELF_UNPROTECT_DATASEC(which)                                      \
    do {                                                                   \
        if (TESTANY(DATASEC_SELFPROT[which], DYNAMO_OPTION(protect_mask))) \
            protect_data_section(which, WRITABLE);                         \
    } while (0);

/***************************************************************************/

#ifdef DEBUG
void
mem_stats_snapshot(void);
#endif

app_pc
get_dynamorio_dll_start(void);
app_pc
get_dynamorio_dll_preferred_base(void);

bool
is_in_dynamo_dll(app_pc pc);
/* Returns the number of separate regions added to the dynamo vm areas list. */
int
find_dynamo_library_vm_areas(void);
/* Returns the number of executable regions found in the address space. */
int
find_executable_vm_areas(void);

/* all_memory_areas is !HAVE_MEMINFO-only: nop elsewhere */
void
all_memory_areas_lock(void);
void
all_memory_areas_unlock(void);
/* pass -1 if type is unchanged */
void
update_all_memory_areas(app_pc start, app_pc end, uint prot, int type);
bool
remove_from_all_memory_areas(app_pc start, app_pc end);

/* file operations */
/* defaults to read only access, if write is not set ignores others */
#define OS_OPEN_READ 0x001
#define OS_OPEN_WRITE 0x002
#define OS_OPEN_WRITE_ONLY 0x004 /* for linux pipes: ignores _APPEND + _NEW flags */
#define OS_OPEN_APPEND 0x008     /* if not set, the file is truncated */
#define OS_OPEN_REQUIRE_NEW 0x010
#define OS_EXECUTE 0x020            /* only used on win32, currently */
#define OS_SHARE_DELETE 0x040       /* only used on win32, currently */
#define OS_OPEN_FORCE_OWNER 0x080   /* only used on win32, currently */
#define OS_OPEN_ALLOW_LARGE 0x100   /* only used on linux32, currently */
#define OS_OPEN_CLOSE_ON_FORK 0x200 /* only used on linux */
#define OS_OPEN_RESERVED 0x10000000 /* used for fd_table on linux */
/* always use OS_OPEN_REQUIRE_NEW when asking for OS_OPEN_WRITE, in
 * order to avoid hard link or symbolic link attacks if the file is in
 * a world writable locations and the process may have high
 * privileges.
 */
file_t
os_open(const char *fname, int os_open_flags);
file_t
os_open_protected(const char *fname, int os_open_flags);
file_t
os_open_directory(const char *fname, int os_open_flags);
bool
os_file_exists(const char *fname, bool is_dir);
bool
os_get_file_size(const char *file, uint64 *size); /* NYI on Linux */
bool
os_get_file_size_by_handle(file_t fd, uint64 *size);
bool
os_get_current_dir(char *buf, size_t bufsz);

typedef enum {
    CREATE_DIR_ALLOW_EXISTING = 0x0,
    /* should always use CREATE_DIR_REQUIRE_NEW, see reference in
     * OS_OPEN_REQUIRE_NEW - to avoid symlink attacks (although only
     * an issue when files we create in these directories have
     * predictible names - case 9138)
     */
    CREATE_DIR_REQUIRE_NEW = 0x1,
    CREATE_DIR_FORCE_OWNER = 0x2,
} create_directory_flags_t;

bool
os_create_dir(const char *fname, create_directory_flags_t create_dir_flags);
bool
os_delete_dir(const char *fname);
void
os_close(file_t f);
void
os_close_protected(file_t f);
/* returns number of bytes written, negative if failure */
ssize_t
os_write(file_t f, const void *buf, size_t count);
/* returns number of bytes read, negative if failure */
ssize_t
os_read(file_t f, void *buf, size_t count);
void
os_flush(file_t f);

/* For use with os_file_seek(), specifies the origin at which to apply the offset
 * NOTE - keep in synch with DR_SEEK_* in insturment.h and SEEK_* from Linux headers */
#define OS_SEEK_SET 0 /* start of file */
#define OS_SEEK_CUR 1 /* current file position */
#define OS_SEEK_END 2 /* end of file */
/* seek current file position to offset bytes from origin, return true if successful */
bool
os_seek(file_t f, int64 offset, int origin);
/* return the current file position, -1 on failure */
int64
os_tell(file_t f);

bool
os_delete_file(const char *file_name);
bool
os_delete_mapped_file(const char *filename);
bool
os_rename_file(const char *orig_name, const char *new_name, bool replace);
/* These routines do not update dynamo_areas; use the non-os_-prefixed versions */
/* File mapping on Linux uses a 32-bit offset, with mmap2 considering
 * it a multiple of the page size in order to have a larger reach; on
 * Windows we have a 64-bit offset.  We go with the widest here, a
 * 64-bit offset, to avoid limiting Windows.
 */
/* When fixed==true, DR tries to allocate memory from the address specified
 * by addr.
 * In Linux, it has the same semantic as mmap system call with MAP_FIXED flags,
 * If the memory region specified by addr and size overlaps pages of
 * any existing mapping(s), then the overlapped part of the existing mapping(s)
 * will be discarded. If the specified address cannot be used, it will fail
 * and returns NULL.
 * XXX: in Windows, fixed argument is currently ignored (PR 214077 / case 9642),
 * and handling it is covered by PR 214097.
 */
byte *
os_map_file(file_t f, size_t *size DR_PARAM_INOUT, uint64 offs, app_pc addr, uint prot,
            map_flags_t map_flags);
bool
os_unmap_file(byte *map, size_t size);
file_t
os_create_memory_file(const char *name, size_t size);
void
os_delete_memory_file(const char *name, file_t fd);
/* unlike set_protection, os_set_protection does not update
 * the allmem info in Linux. */
bool
os_set_protection(byte *pc, size_t length, uint prot /*MEMPROT_*/);

bool
os_current_user_directory(char *directory_prefix /* INOUT */, uint directory_size,
                          bool create);
bool
os_validate_user_owned(file_t file_or_directory_handle);

bool
os_get_disk_free_space(/*IN*/ file_t file_handle,
                       /*OUT*/ uint64 *AvailableQuotaBytes /*OPTIONAL*/,
                       /*OUT*/ uint64 *TotalQuotaBytes /*OPTIONAL*/,
                       /*OUT*/ uint64 *TotalVolumeBytes /*OPTIONAL*/);
file_t
redirect_open(const char *fname, int flags, int mode);
void
redirect_close(file_t f);

#ifdef PROFILE_RDTSC
extern uint kilo_hertz;
#endif

/* Despite the name, this enum is used for all types of critical events:
 * asserts and crashes for all builds, and security violations for
 * PROGRAM_SHEPHERDING builds.
 * When a security_violation is being reported, enum value must be negative!
 */
typedef enum {
#ifdef PROGRAM_SHEPHERDING
    STACK_EXECUTION_VIOLATION = -1,
    HEAP_EXECUTION_VIOLATION = -2,
    RETURN_TARGET_VIOLATION = -3,
    RETURN_DIRECT_RCT_VIOLATION = -4, /* NYI DIRECT_CALL_CHECK */
    INDIRECT_CALL_RCT_VIOLATION = -5,
    INDIRECT_JUMP_RCT_VIOLATION = -6,
#    ifdef HOT_PATCHING_INTERFACE
    HOT_PATCH_DETECTOR_VIOLATION = -7,
    HOT_PATCH_PROTECTOR_VIOLATION = -8,
    /* Errors and exceptions in hot patches are treated the same.  Case 5696. */
    HOT_PATCH_FAILURE = -9,
#    endif
    /* internal */
    ATTACK_SIMULATION_VIOLATION = -10,
    ATTACK_SIM_NUDGE_VIOLATION = -11,
#endif
    /* not really PROGRAM_SHEPHERDING */
    ASLR_TARGET_VIOLATION = -12,
#ifdef GBOP
    GBOP_SOURCE_VIOLATION = -13,
#endif /* GBOP */
#ifdef PROCESS_CONTROL
    PROCESS_CONTROL_VIOLATION = -14, /* case 8594 */
#endif
    APC_THREAD_SHELLCODE_VIOLATION = -15, /* note still presented externally as .B */
    /* Not a valid violation; used for initializing security_violation_t vars. */
    INVALID_VIOLATION = 0,
#ifdef PROGRAM_SHEPHERDING
    /* Add new violation types above this line as consecutive negative
     * numbers, and update get_security_violation_name() for the
     * appropriate letter obfuscation */
    ALLOWING_OK = 1,
    ALLOWING_BAD = 2,
#endif
    NO_VIOLATION_BAD_INTERNAL_STATE = 3,
    NO_VIOLATION_OK_INTERNAL_STATE = 4
} security_violation_t;

/*  in diagnost.c */
void
report_diagnostics(const char *message, const char *name,
                   security_violation_t violation_type);
void
append_diagnostics(file_t diagnostics_file, const char *message, const char *name,
                   security_violation_t violation_type);
void
diagnost_exit(void);
bool
check_for_unsupported_modules(void);

#ifdef RETURN_AFTER_CALL
typedef enum {
    INITIAL_STACK_EMPTY = 0,
    INITIAL_STACK_BOTTOM_REACHED = 1,
    INITIAL_STACK_BOTTOM_NOT_REACHED = 2
} initial_call_stack_status_t;

initial_call_stack_status_t
at_initial_stack_bottom(dcontext_t *dcontext, app_pc target_pc);
bool
at_known_exception(dcontext_t *dcontext, app_pc target_pc, app_pc source_fragment);
#endif

/* contended path of mutex operations */
bool
ksynch_var_initialized(KSYNCH_TYPE *var);
/* If mc != NULL we mark this thread safe to suspend and transfer with a valid
 * mcontext (THREAD_SYNCH_VALID_MCONTEXT). Note that means that all fields of
 * \p mc must be valid (e.g. PC, control, integer, MMX fields).
 */
void
mutex_wait_contended_lock(mutex_t *lock, priv_mcontext_t *mc);
void
mutex_notify_released_lock(mutex_t *lock);
void
mutex_free_contended_event(mutex_t *lock);
/* contended path of rwlock operations */
void
rwlock_wait_contended_writer(read_write_lock_t *rwlock);
void
rwlock_notify_writer(read_write_lock_t *rwlock);
void
rwlock_wait_contended_reader(read_write_lock_t *rwlock);
void
rwlock_notify_readers(read_write_lock_t *rwlock);

/* The event interface */

/* An event object. */
#ifdef WINDOWS
typedef HANDLE event_t;
#else
typedef struct linux_event_t *event_t;
#endif

event_t
create_event(void);
/* A broadcast event wakes all waiting threads when signaled. */
event_t
create_broadcast_event(void);
void
destroy_event(event_t e);
/* For a broadcast event, wakes all threads; o/w wakes at most one thread. */
void
signal_event(event_t e);
void
reset_event(event_t e);
/* 0 means to wait forever.  Returns false on timeout, true on event firing. */
bool
wait_for_event(event_t e, int timeout_ms);

/* get current timer frequency in KHz */
/* NOTE: Keep in mind that with voltage scaling in power saving mode
 * this may not be a reliable way to obtain time information */
timestamp_t
get_timer_frequency(void);

/* DR has standardized on UTC time which counts from Jan 1, 1601.
 * That's the Windows standard.  But Linux uses the Epoch of Jan 1, 1970.
 */
#define UTC_TO_EPOCH_SECONDS 11644473600

/* Returns the number of seconds since Jan 1, 1601 (this is
 * the current UTC time).
 */
uint
query_time_seconds(void);

/* Returns the number of milliseconds since Jan 1, 1601 (this is
 * the current UTC time).
 */
uint64
query_time_millis(void);

/* microseconds since 1601 */
uint64
query_time_micros(void);

/* gives a good but not necessarily crypto-strength random seed */
uint
os_random_seed(void);

/* module map/unmap processing relevant to the RCT policies */
/* cf. rct_analyze_module_at_violation() which can be called lazily */
void
rct_process_module_mmap(app_pc module_base, size_t module_size, bool add,
                        bool already_relocated);

/* module boundary and code section analysis for RCT policies */
/* XXX: should be better abstracted by calling a routine that
 * enumerates code sections, while keeping the general driver in rct.c
 */
bool
rct_analyze_module_at_violation(dcontext_t *dcontext, app_pc target_pc);
bool
aslr_is_possible_attack(app_pc target);
app_pc
aslr_possible_preferred_address(app_pc target_addr);

#ifdef WINDOWS
/* d_r_dispatch places next pc in asynch_target and clears it after syscall handling
 * completes, so a zero value means shared syscall was used and the next pc
 * is in the esi slot. If asynch_target == BACK_TO_NATIVE_AFTER_SYSCALL then the
 * thread is native at an intercepted syscall and the real post syscall target is in
 * native_exec_postsyscall. */
#    define POST_SYSCALL_PC(dc)                                 \
        ((dc)->asynch_target == NULL                            \
         ? ASSERT(DYNAMO_OPTION(shared_syscalls)),              \
         (app_pc)get_mcontext((dc))->xsi                        \
         : ((dc)->asynch_target == BACK_TO_NATIVE_AFTER_SYSCALL \
                ? (dc)->native_exec_postsyscall                 \
                : (dc)->asynch_target))
#else
/* on linux asynch target always holds the post syscall pc */
#    define POST_SYSCALL_PC(dc) ((dc)->asynch_target)
#endif

/* This has been exposed from win32 module so that hotp_only_gateway can
 * return the right return code.
 */
typedef enum {
    AFTER_INTERCEPT_LET_GO,
    AFTER_INTERCEPT_LET_GO_ALT_DYN, /* alternate direct execution target,
                                     * usable only with
                                     * AFTER_INTERCEPT_DYNAMIC_DECISION,
                                     * not by itself */
    AFTER_INTERCEPT_TAKE_OVER,
    AFTER_INTERCEPT_DYNAMIC_DECISION, /* handler returns one of prev values */

    /* handler is expected to restore original instructions */
    AFTER_INTERCEPT_TAKE_OVER_SINGLE_SHOT, /* static only with alternative target */
} after_intercept_action_t;

/* Argument structure for intercept function; contains app state at the point
 * our intercept routine will take over.  Fix for case 7597.
 *
 * -- CAUTION -- the number, order and size of fields in this structure are
 * assumed in emit_intercept_code().  Changing anything here is likely to cause
 * DR hooks & hotp_only to fail.
 */
typedef struct {
    void *callee_arg;   /* Argument passed to the intercept routine. */
    app_pc start_pc;    /* optimization: could use mc.pc instead */
    priv_mcontext_t mc; /* note: 8-byte aligned */
} app_state_at_intercept_t;

/* note that only points intercepted with DYNAMIC_DECISION (currently only
 * [un]load_dll) care about the return value */
typedef after_intercept_action_t
intercept_function_t(app_state_at_intercept_t *args);

/* opcodes and encoding constants that we sometimes directly use */
#ifdef X86
/* XXX: if we add more opcodes maybe should be exported by ir/ */
enum {
    JMP_REL32_OPCODE = 0xe9,
    JMP_REL32_SIZE = 5, /* size in bytes of 32-bit rel jmp */
    CALL_REL32_OPCODE = 0xe8,
    JMP_ABS_IND64_OPCODE = 0xff,
    JMP_ABS_IND64_SIZE = 6, /* size in bytes of a 64-bit abs ind jmp */
    JMP_ABS_MEM_IND64_MODRM = 0x25,
};
#elif defined(AARCHXX)
enum {
    /* XXX i#1551, i#1569: this is for A32 for now to get things compiling */
    JMP_REL32_OPCODE = 0xec000000,
    JMP_REL32_SIZE = 4,
    CALL_REL32_OPCODE = 0xed000000,
};
#elif defined(RISCV64)
enum {
    /* XXX i#3544: Fix proper values. Those are for compilation only. */
    JMP_REL32_OPCODE = 0xec000000,
    JMP_REL32_SIZE = 4,
    CALL_REL32_OPCODE = 0xed000000,
};
#else
#    error only X86 and ARM supported
#endif /* X86 */

#ifdef WINDOWS
/* used for option early_inject_location */
/* use 0 to supply an arbitrary address via -early_inject_address */
/* NOTE - these enum values are passed between processes, don't change/reuse
 * any existing values. */
/* NtMapViewOfSection doesn't work as a place to LoadDll (ldr can't handle the
 * re-entrancy), but might be a good place to takeover if we remote mapped in
 * and needed ntdll initialized (first MapView is for kernel32.dll fom what
 * I've seen, though beware the LdrLists aren't consistent at this point). */
/* As currently implemented KiUserException location doesn't work, but we might
 * be able to find a better exception location (maybe point the image's import
 * section to invalid memory instead? TODO try). */
/* see early_inject_init() in os.c for more notes on best location for each
 * os version */
/* NOTE - be carefull changing this enum as it's shared across proccesses.
 * Also values <= LdrDefault are assumed to need address computation while
 * those > are assumed to not need it. */
#    define INJECT_LOCATION_IS_LDR(loc) (loc <= INJECT_LOCATION_LdrDefault)
#    define INJECT_LOCATION_IS_LDR_NON_DEFAULT(loc) (loc < INJECT_LOCATION_LdrDefault)
enum {
    INJECT_LOCATION_Invalid = -100,           /* invalid sentinel */
    INJECT_LOCATION_LdrpLoadDll = 0,          /* Good for XP, 2k3 */
    INJECT_LOCATION_LdrpLoadImportModule = 1, /* Good for 2k */
    INJECT_LOCATION_LdrCustom = 2,            /* Specify custom address */
    INJECT_LOCATION_LdrLoadDll = 3,           /* Good for 2k3 */
    INJECT_LOCATION_LdrDefault = 4,           /* Pick best location based on OS */
    /* For NT the best location is LdrpLoadImportModule, but we can't find that
     * automatically on NT, so LdrDefault for NT uses -early_inject_address if
     * specified or else disables early injection (xref 7806). */
    /* Beyond this point not expected to need address determination */
    /* Earliest injection via remote map.
     * On Vista+ this is treated as LdrInitializeThunk as there is no init APC.
     */
    INJECT_LOCATION_KiUserApc = 5,
    INJECT_LOCATION_KiUserException = 6, /* No good, Ldr init issues */
    /* Clients that depend on private libraries have trouble running
     * at the early and earliest injection points. At the image
     * entry point all the app libraries are loaded and hence it is suitable
     * for those clients which are using private libraries those depends
     * on some app libraries being initialized
     */
    INJECT_LOCATION_ImageEntry = 7,
    /* Similar in lateness to ImageEntry, but is more robust in that it does not
     * rely on reaching the image entry, which not all apps do (e.g., .NET).
     * This is equivalent to RtlUserThreadStart.
     */
    INJECT_LOCATION_ThreadStart = 8,
    INJECT_LOCATION_MAX = INJECT_LOCATION_ThreadStart,
};
#endif

void
take_over_primary_thread(void);

#ifdef HOT_PATCHING_INTERFACE
void *
get_drmarker_hotp_policy_status_table(void);
void
set_drmarker_hotp_policy_status_table(void *new_table);

/* Used for -hotp_only; can be exposed if needed elsewhere later on. */
byte *
hook_text(byte *hook_code_buf, const app_pc image_addr, intercept_function_t hook_func,
          const void *callee_arg, const after_intercept_action_t action_after,
          const bool abort_if_hooked, const bool ignore_cti, byte **app_code_copy_p,
          byte **alt_exit_tgt_p);
void
unhook_text(byte *hook_code_buf, app_pc image_addr);
void
insert_jmp_at_tramp_entry(dcontext_t *dcontext, byte *trampoline, byte *target);
#endif /* HOT_PATCHING_INTERFACE */

/* checks for compatibility OS specific options, returns true if
 * modified the value of any options to make them compatible
 */
bool
os_check_option_compatibility(void);

/* Introduced as part of PR 250294 - 64-bit hook reachability. */
#define LANDING_PAD_AREA_SIZE 64 * 1024
#define MAX_HOOK_DISPLACED_LENGTH (JMP_LONG_LENGTH + MAX_INSTR_LENGTH)
#ifdef X64
/* 8 bytes for the 64-bit abs addr, 6 for abs ind jmp to the trampoline and 5
 * for return jmp back to the instruction after the hook.  Plus displaced instr(s).
 */
#    define LANDING_PAD_SIZE (19 + MAX_HOOK_DISPLACED_LENGTH)
#else
/* 5 bytes each for the two relative jumps (one to the trampoline and the
 * other back to instruction after hook.  Plus displaced instr(s).
 */
#    define LANDING_PAD_SIZE (10 + MAX_HOOK_DISPLACED_LENGTH)
#endif
byte *
alloc_landing_pad(app_pc addr_to_hook);

bool
trim_landing_pad(byte *lpad_start, size_t space_used);

void
landing_pads_to_executable_areas(bool add);

/* in loader_shared.c */
app_pc
load_private_library(const char *filename, bool reachable);
bool
unload_private_library(app_pc modbase);
/* searches in standard paths instead of requiring abs path */
app_pc
locate_and_load_private_library(const char *name, bool reachable);
void
loader_init_prologue(void);
void
loader_init_epilogue(dcontext_t *dcontext);
void
loader_exit(void);
void
loader_thread_init(dcontext_t *dcontext);
void
loader_thread_exit(dcontext_t *dcontext);
void
loader_make_exit_calls(dcontext_t *dcontext);
bool
in_private_library(app_pc pc);
void
loader_allow_unsafe_static_behavior(void);

#endif /* OS_SHARED_H */