src/firehose/firehose_inline_internal.h

   1 /*
   2  * Copyright (c) 2015 Apple Inc. All rights reserved.
   3  *
   4  * @APPLE_APACHE_LICENSE_HEADER_START@
   5  *
   6  * Licensed under the Apache License, Version 2.0 (the "License");
   7  * you may not use this file except in compliance with the License.
   8  * You may obtain a copy of the License at
   9  *
  10  *     http://www.apache.org/licenses/LICENSE-2.0
  11  *
  12  * Unless required by applicable law or agreed to in writing, software
  13  * distributed under the License is distributed on an "AS IS" BASIS,
  14  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  15  * See the License for the specific language governing permissions and
  16  * limitations under the License.
  17  *
  18  * @APPLE_APACHE_LICENSE_HEADER_END@
  19  */
  20
  21 #ifndef __FIREHOSE_INLINE_INTERNAL__
  22 #define __FIREHOSE_INLINE_INTERNAL__
  23
  24 #define firehose_atomic_maxv2o(p, f, v, o, m) \
  25                 os_atomic_rmw_loop2o(p, f, *(o), (v), m, { \
  26                         if (*(o) >= (v)) os_atomic_rmw_loop_give_up(break); \
  27                 })
  28
  29 #define firehose_atomic_max2o(p, f, v, m)   ({ \
  30                 typeof((p)->f) _old; \
  31                 firehose_atomic_maxv2o(p, f, v, &_old, m); \
  32         })
  33
  34 #ifndef KERNEL
  35 // caller must test for non zero first
  36 OS_ALWAYS_INLINE
  37 static inline uint16_t
  38 firehose_bitmap_first_set(uint64_t bitmap)
  39 {
  40         dispatch_assert(bitmap != 0);
  41         // this builtin returns 0 if bitmap is 0, or (first bit set + 1)
  42         return (uint16_t)__builtin_ffsll((long long)bitmap) - 1;
  43 }
  44 #endif
  45
  46 #pragma mark -
  47 #pragma mark Mach Misc.
  48 #ifndef KERNEL
  49
  50 OS_ALWAYS_INLINE
  51 static inline mach_port_t
  52 firehose_mach_port_allocate(uint32_t flags, void *ctx)
  53 {
  54         mach_port_t port = MACH_PORT_NULL;
  55         mach_port_options_t opts = {
  56                 .flags = flags,
  57         };
  58         kern_return_t kr = mach_port_construct(mach_task_self(), &opts,
  59                         (mach_port_context_t)ctx, &port);
  60         if (unlikely(kr)) {
  61                 DISPATCH_VERIFY_MIG(kr);
  62                 DISPATCH_CLIENT_CRASH(kr, "Unable to allocate mach port");
  63         }
  64         return port;
  65 }
  66
  67 OS_ALWAYS_INLINE
  68 static inline kern_return_t
  69 firehose_mach_port_recv_dispose(mach_port_t port, void *ctx)
  70 {
  71         kern_return_t kr;
  72         kr = mach_port_destruct(mach_task_self(), port, 0,
  73                         (mach_port_context_t)ctx);
  74         DISPATCH_VERIFY_MIG(kr);
  75         return kr;
  76 }
  77
  78 OS_ALWAYS_INLINE
  79 static inline void
  80 firehose_mach_port_send_release(mach_port_t port)
  81 {
  82         kern_return_t kr = mach_port_deallocate(mach_task_self(), port);
  83         DISPATCH_VERIFY_MIG(kr);
  84         dispatch_assume_zero(kr);
  85 }
  86
  87 OS_ALWAYS_INLINE
  88 static inline void
  89 firehose_mach_port_guard(mach_port_t port, bool strict, void *ctx)
  90 {
  91         kern_return_t kr = mach_port_guard(mach_task_self(), port,
  92                         (mach_port_context_t)ctx, strict);
  93         DISPATCH_VERIFY_MIG(kr);
  94         dispatch_assume_zero(kr);
  95 }
  96
  97 OS_ALWAYS_INLINE
  98 static inline void
  99 firehose_mig_server(dispatch_mig_callback_t demux, size_t maxmsgsz,
 100                 mach_msg_header_t *hdr)
 101 {
 102         mig_reply_error_t *msg_reply = (mig_reply_error_t *)alloca(maxmsgsz);
 103         kern_return_t rc = KERN_SUCCESS;
 104         bool expects_reply = false;
 105
 106         if (MACH_MSGH_BITS_REMOTE(hdr->msgh_bits) == MACH_MSG_TYPE_MOVE_SEND_ONCE) {
 107                 expects_reply = true;
 108         }
 109
 110         if (!fastpath(demux(hdr, &msg_reply->Head))) {
 111                 rc = MIG_BAD_ID;
 112         } else if (msg_reply->Head.msgh_bits & MACH_MSGH_BITS_COMPLEX) {
 113                 rc = KERN_SUCCESS;
 114         } else {
 115                 // if MACH_MSGH_BITS_COMPLEX is _not_ set, then msg_reply->RetCode
 116                 // is present
 117                 rc = msg_reply->RetCode;
 118         }
 119
 120         if (slowpath(rc == KERN_SUCCESS && expects_reply)) {
 121                 // if crashing here, some handler returned KERN_SUCCESS
 122                 // hoping for firehose_mig_server to perform the mach_msg()
 123                 // call to reply, and it doesn't know how to do that
 124                 DISPATCH_INTERNAL_CRASH(msg_reply->Head.msgh_id,
 125                                 "firehose_mig_server doesn't handle replies");
 126         }
 127         if (slowpath(rc != KERN_SUCCESS && rc != MIG_NO_REPLY)) {
 128                 // destroy the request - but not the reply port
 129                 hdr->msgh_remote_port = 0;
 130                 mach_msg_destroy(hdr);
 131         }
 132 }
 133
 134 #endif // !KERNEL
 135 #pragma mark -
 136 #pragma mark firehose buffer
 137
 138 OS_ALWAYS_INLINE
 139 static inline firehose_chunk_t
 140 firehose_buffer_chunk_for_address(void *addr)
 141 {
 142         uintptr_t chunk_addr = (uintptr_t)addr & ~(FIREHOSE_CHUNK_SIZE - 1);
 143         return (firehose_chunk_t)chunk_addr;
 144 }
 145
 146 OS_ALWAYS_INLINE
 147 static inline uint16_t
 148 firehose_buffer_chunk_to_ref(firehose_buffer_t fb, firehose_chunk_t fbc)
 149 {
 150         return (uint16_t)(fbc - fb->fb_chunks);
 151 }
 152
 153 OS_ALWAYS_INLINE
 154 static inline firehose_chunk_t
 155 firehose_buffer_ref_to_chunk(firehose_buffer_t fb, uint16_t ref)
 156 {
 157         return fb->fb_chunks + ref;
 158 }
 159
 160 #ifndef FIREHOSE_SERVER
 161 #if DISPATCH_PURE_C
 162
 163 OS_ALWAYS_INLINE
 164 static inline uint8_t
 165 firehose_buffer_qos_bits_propagate(void)
 166 {
 167 #ifndef KERNEL
 168         pthread_priority_t pp = _dispatch_priority_propagate();
 169
 170         pp &= _PTHREAD_PRIORITY_QOS_CLASS_MASK;
 171         return (uint8_t)(pp >> _PTHREAD_PRIORITY_QOS_CLASS_SHIFT);
 172 #else
 173         return 0;
 174 #endif
 175 }
 176
 177 OS_ALWAYS_INLINE
 178 static inline void
 179 firehose_buffer_stream_flush(firehose_buffer_t fb, firehose_stream_t stream)
 180 {
 181         firehose_buffer_stream_t fbs = &fb->fb_header.fbh_stream[stream];
 182         firehose_stream_state_u old_state, new_state;
 183         firehose_chunk_t fc;
 184         uint64_t stamp = UINT64_MAX; // will cause the reservation to fail
 185         uint16_t ref;
 186         long result;
 187
 188         old_state.fss_atomic_state =
 189                         os_atomic_load2o(fbs, fbs_state.fss_atomic_state, relaxed);
 190         ref = old_state.fss_current;
 191         if (!ref || ref == FIREHOSE_STREAM_STATE_PRISTINE) {
 192                 // there is no installed page, nothing to flush, go away
 193 #ifndef KERNEL
 194                 firehose_buffer_force_connect(fb);
 195 #endif
 196                 return;
 197         }
 198
 199         fc = firehose_buffer_ref_to_chunk(fb, old_state.fss_current);
 200         result = firehose_chunk_tracepoint_try_reserve(fc, stamp, stream,
 201                         firehose_buffer_qos_bits_propagate(), 1, 0, NULL);
 202         if (likely(result < 0)) {
 203                 firehose_buffer_ring_enqueue(fb, old_state.fss_current);
 204         }
 205         if (unlikely(result > 0)) {
 206                 // because we pass a silly stamp that requires a flush
 207                 DISPATCH_INTERNAL_CRASH(result, "Allocation should always fail");
 208         }
 209
 210         // as a best effort try to uninstall the page we just flushed
 211         // but failing is okay, let's not contend stupidly for something
 212         // allocators know how to handle in the first place
 213         new_state = old_state;
 214         new_state.fss_current = 0;
 215         (void)os_atomic_cmpxchg2o(fbs, fbs_state.fss_atomic_state,
 216                         old_state.fss_atomic_state, new_state.fss_atomic_state, relaxed);
 217 }
 218
 219 /**
 220  * @function firehose_buffer_tracepoint_reserve
 221  *
 222  * @abstract
 223  * Reserves space in the firehose buffer for the tracepoint with specified
 224  * characteristics.
 225  *
 226  * @discussion
 227  * This returns a slot, with the length of the tracepoint already set, so
 228  * that in case of a crash, we maximize our chance to be able to skip the
 229  * tracepoint in case of a partial write.
 230  *
 231  * Once the tracepoint has been written, firehose_buffer_tracepoint_flush()
 232  * must be called.
 233  *
 234  * @param fb
 235  * The buffer to allocate from.
 236  *
 237  * @param stream
 238  * The buffer stream to use.
 239  *
 240  * @param pubsize
 241  * The size of the public data for this tracepoint, cannot be 0, doesn't
 242  * take the size of the tracepoint header into account.
 243  *
 244  * @param privsize
 245  * The size of the private data for this tracepoint, can be 0.
 246  *
 247  * @param privptr
 248  * The pointer to the private buffer, can be NULL
 249  *
 250  * @result
 251  * The pointer to the tracepoint.
 252  */
 253 OS_ALWAYS_INLINE
 254 static inline firehose_tracepoint_t
 255 firehose_buffer_tracepoint_reserve(firehose_buffer_t fb, uint64_t stamp,
 256                 firehose_stream_t stream, uint16_t pubsize,
 257                 uint16_t privsize, uint8_t **privptr)
 258 {
 259         firehose_buffer_stream_t fbs = &fb->fb_header.fbh_stream[stream];
 260         firehose_stream_state_u old_state, new_state;
 261         firehose_chunk_t fc;
 262 #if KERNEL
 263         bool failable = false;
 264 #endif
 265         bool success;
 266         long result;
 267         uint16_t ref;
 268
 269         // cannot use os_atomic_rmw_loop2o, _page_try_reserve does a store
 270         old_state.fss_atomic_state =
 271                         os_atomic_load2o(fbs, fbs_state.fss_atomic_state, relaxed);
 272         for (;;) {
 273                 new_state = old_state;
 274
 275                 ref = old_state.fss_current;
 276                 if (likely(ref && ref != FIREHOSE_STREAM_STATE_PRISTINE)) {
 277                         fc = firehose_buffer_ref_to_chunk(fb, ref);
 278                         result = firehose_chunk_tracepoint_try_reserve(fc, stamp, stream,
 279                                         firehose_buffer_qos_bits_propagate(),
 280                                         pubsize, privsize, privptr);
 281                         if (likely(result > 0)) {
 282                                 uint64_t thread;
 283 #ifdef KERNEL
 284                                 thread = thread_tid(current_thread());
 285 #else
 286                                 thread = _pthread_threadid_self_np_direct();
 287 #endif
 288                                 return firehose_chunk_tracepoint_begin(fc,
 289                                                 stamp, pubsize, thread, result);
 290                         }
 291                         if (likely(result < 0)) {
 292                                 firehose_buffer_ring_enqueue(fb, old_state.fss_current);
 293                         }
 294                         new_state.fss_current = 0;
 295                 }
 296 #if KERNEL
 297                 if (failable) {
 298                         return NULL;
 299                 }
 300 #endif
 301
 302                 if (unlikely(old_state.fss_allocator)) {
 303                         _dispatch_gate_wait(&fbs->fbs_state.fss_gate,
 304                                         DLOCK_LOCK_DATA_CONTENTION);
 305                         old_state.fss_atomic_state =
 306                                         os_atomic_load2o(fbs, fbs_state.fss_atomic_state, relaxed);
 307 #if KERNEL
 308                         failable = true;
 309 #endif
 310                         continue;
 311                 }
 312
 313                 // if the thread doing the allocation is a low priority one
 314                 // we may starve high priority ones.
 315                 // so disable preemption before we become an allocator
 316                 // the reenabling of the preemption is in
 317                 // firehose_buffer_stream_chunk_install
 318                 __firehose_critical_region_enter();
 319 #if KERNEL
 320                 new_state.fss_allocator = (uint32_t)cpu_number();
 321 #else
 322                 new_state.fss_allocator = _dispatch_lock_value_for_self();
 323 #endif
 324                 success = os_atomic_cmpxchgv2o(fbs, fbs_state.fss_atomic_state,
 325                                 old_state.fss_atomic_state, new_state.fss_atomic_state,
 326                                 &old_state.fss_atomic_state, relaxed);
 327                 if (likely(success)) {
 328                         break;
 329                 }
 330                 __firehose_critical_region_leave();
 331         }
 332
 333         struct firehose_tracepoint_query_s ask = {
 334                 .pubsize = pubsize,
 335                 .privsize = privsize,
 336                 .stream = stream,
 337                 .for_io = (firehose_stream_uses_io_bank & (1UL << stream)) != 0,
 338 #ifndef KERNEL
 339                 .quarantined = fb->fb_header.fbh_quarantined,
 340 #endif
 341                 .stamp = stamp,
 342         };
 343         return firehose_buffer_tracepoint_reserve_slow(fb, &ask, privptr);
 344 }
 345
 346 /**
 347  * @function firehose_buffer_tracepoint_flush
 348  *
 349  * @abstract
 350  * Flushes a firehose tracepoint, and sends the chunk to the daemon when full
 351  * and this was the last tracepoint writer for this chunk.
 352  *
 353  * @param fb
 354  * The buffer the tracepoint belongs to.
 355  *
 356  * @param ft
 357  * The tracepoint to flush.
 358  *
 359  * @param ftid
 360  * The firehose tracepoint ID for that tracepoint.
 361  * It is written last, preventing compiler reordering, so that its absence
 362  * on crash recovery means the tracepoint is partial.
 363  */
 364 OS_ALWAYS_INLINE
 365 static inline void
 366 firehose_buffer_tracepoint_flush(firehose_buffer_t fb,
 367                 firehose_tracepoint_t ft, firehose_tracepoint_id_u ftid)
 368 {
 369         firehose_chunk_t fc = firehose_buffer_chunk_for_address(ft);
 370
 371         // Needed for process death handling (tracepoint-flush):
 372         // We want to make sure the observers
 373         // will see memory effects in program (asm) order.
 374         // 1. write all the data to the tracepoint
 375         // 2. write the tracepoint ID, so that seeing it means the tracepoint
 376         //    is valid
 377         if (firehose_chunk_tracepoint_end(fc, ft, ftid)) {
 378                 firehose_buffer_ring_enqueue(fb, firehose_buffer_chunk_to_ref(fb, fc));
 379         }
 380 }
 381
 382 #ifndef KERNEL
 383 OS_ALWAYS_INLINE
 384 static inline void
 385 firehose_buffer_clear_bank_flags(firehose_buffer_t fb, unsigned long bits)
 386 {
 387         firehose_buffer_bank_t fbb = &fb->fb_header.fbh_bank;
 388         unsigned long orig_flags;
 389
 390         orig_flags = os_atomic_and_orig2o(fbb, fbb_flags, ~bits, relaxed);
 391         if (orig_flags != (orig_flags & ~bits)) {
 392                 firehose_buffer_update_limits(fb);
 393         }
 394 }
 395
 396 OS_ALWAYS_INLINE
 397 static inline void
 398 firehose_buffer_set_bank_flags(firehose_buffer_t fb, unsigned long bits)
 399 {
 400         firehose_buffer_bank_t fbb = &fb->fb_header.fbh_bank;
 401         unsigned long orig_flags;
 402
 403         orig_flags = os_atomic_or_orig2o(fbb, fbb_flags, bits, relaxed);
 404         if (orig_flags != (orig_flags | bits)) {
 405                 firehose_buffer_update_limits(fb);
 406         }
 407 }
 408 #endif // !KERNEL
 409
 410 #endif // !defined(FIREHOSE_SERVER)
 411
 412 #endif // DISPATCH_PURE_C
 413
 414 #endif // __FIREHOSE_INLINE_INTERNAL__