Age Owner Branch data TLA Line data Source code
1 : : /*-------------------------------------------------------------------------
2 : : *
3 : : * shmem.c
4 : : * create shared memory and initialize shared memory data structures.
5 : : *
6 : : * Portions Copyright (c) 1996-2026, PostgreSQL Global Development Group
7 : : * Portions Copyright (c) 1994, Regents of the University of California
8 : : *
9 : : *
10 : : * IDENTIFICATION
11 : : * src/backend/storage/ipc/shmem.c
12 : : *
13 : : *-------------------------------------------------------------------------
14 : : */
15 : : /*
16 : : * POSTGRES processes share one or more regions of shared memory.
17 : : * The shared memory is created by a postmaster and is inherited
18 : : * by each backend via fork() (or, in some ports, via other OS-specific
19 : : * methods). The routines in this file are used for allocating and
20 : : * binding to shared memory data structures.
21 : : *
22 : : * This module provides facilities to allocate fixed-size structures in shared
23 : : * memory, for things like variables shared between all backend processes.
24 : : * Each such structure has a string name to identify it, specified when it is
25 : : * requested. shmem_hash.c provides a shared hash table implementation on top
26 : : * of that.
27 : : *
28 : : * Shared memory areas should usually not be allocated after postmaster
29 : : * startup, although we do allow small allocations later for the benefit of
30 : : * extension modules that are loaded after startup. Despite that allowance,
31 : : * extensions that need shared memory should be added in
32 : : * shared_preload_libraries, because the allowance is quite small and there is
33 : : * no guarantee that any memory is available after startup.
34 : : *
35 : : * Nowadays, there is also another way to allocate shared memory called
36 : : * Dynamic Shared Memory. See dsm.c for that facility. One big difference
37 : : * between traditional shared memory handled by shmem.c and dynamic shared
38 : : * memory is that traditional shared memory areas are mapped to the same
39 : : * address in all processes, so you can use normal pointers in shared memory
40 : : * structs. With Dynamic Shared Memory, you must use offsets or DSA pointers
41 : : * instead.
42 : : *
43 : : * Shared memory managed by shmem.c can never be freed, once allocated. Each
44 : : * hash table has its own free list, so hash buckets can be reused when an
45 : : * item is deleted.
46 : : *
47 : : * Usage
48 : : * -----
49 : : *
50 : : * To allocate shared memory, you need to register a set of callback functions
51 : : * which handle the lifecycle of the allocation. In the request_fn callback,
52 : : * call ShmemRequestStruct() with the desired name and size. When the area is
53 : : * later allocated or attached to, the global variable pointed to by the .ptr
54 : : * option is set to the shared memory location of the allocation. The init_fn
55 : : * callback can perform additional initialization.
56 : : *
57 : : * typedef struct MyShmemData {
58 : : * ...
59 : : * } MyShmemData;
60 : : *
61 : : * static MyShmemData *MyShmem;
62 : : *
63 : : * static void my_shmem_request(void *arg);
64 : : * static void my_shmem_init(void *arg);
65 : : *
66 : : * const ShmemCallbacks MyShmemCallbacks = {
67 : : * .request_fn = my_shmem_request,
68 : : * .init_fn = my_shmem_init,
69 : : * };
70 : : *
71 : : * static void
72 : : * my_shmem_request(void *arg)
73 : : * {
74 : : * ShmemRequestStruct(.name = "My shmem area",
75 : : * .size = sizeof(MyShmemData),
76 : : * .ptr = (void **) &MyShmem,
77 : : * );
78 : : * }
79 : : *
80 : : * In builtin PostgreSQL code, add the callbacks to the list in
81 : : * src/include/storage/subsystemlist.h. In an add-in module, you can register
82 : : * the callbacks by calling RegisterShmemCallbacks(&MyShmemCallbacks) in the
83 : : * extension's _PG_init() function.
84 : : *
85 : : * Lifecycle
86 : : * ---------
87 : : *
88 : : * Initializing shared memory happens in multiple phases. In the first phase,
89 : : * during postmaster startup, all the request_fn callbacks are called. Only
90 : : * after all the request_fn callbacks have been called and all the shmem areas
91 : : * have been requested by the ShmemRequestStruct() calls we know how much
92 : : * shared memory we need in total. After that, postmaster allocates global
93 : : * shared memory segment, and calls all the init_fn callbacks to initialize
94 : : * all the requested shmem areas.
95 : : *
96 : : * In standard Unix-ish environments, individual backends do not need to
97 : : * re-establish their local pointers into shared memory, because they inherit
98 : : * correct values of those variables via fork() from the postmaster. However,
99 : : * this does not work in the EXEC_BACKEND case. In ports using EXEC_BACKEND,
100 : : * backend startup also calls the shmem_request callbacks to re-establish the
101 : : * knowledge about each shared memory area, sets the pointer variables
102 : : * (*options->ptr), and calls the attach_fn callback, if any, for additional
103 : : * per-backend setup.
104 : : *
105 : : * Legacy ShmemInitStruct()/ShmemInitHash() functions
106 : : * --------------------------------------------------
107 : : *
108 : : * ShmemInitStruct()/ShmemInitHash() is another way of registering shmem
109 : : * areas. It pre-dates the ShmemRequestStruct()/ShmemRequestHash() functions,
110 : : * and should not be used in new code, but as of this writing it is still
111 : : * widely used in extensions.
112 : : *
113 : : * To allocate a shmem area with ShmemInitStruct(), you need to separately
114 : : * register the size needed for the area by calling RequestAddinShmemSpace()
115 : : * from the extension's shmem_request_hook, and allocate the area by calling
116 : : * ShmemInitStruct() from the extension's shmem_startup_hook. There are no
117 : : * init/attach callbacks. Instead, the caller of ShmemInitStruct() must check
118 : : * the return status of ShmemInitStruct() and initialize the struct if it was
119 : : * not previously initialized.
120 : : *
121 : : * Calling ShmemAlloc() directly
122 : : * -----------------------------
123 : : *
124 : : * There's a more low-level way of allocating shared memory too: you can call
125 : : * ShmemAlloc() directly. It's used to implement the higher level mechanisms,
126 : : * and should generally not be called directly.
127 : : */
128 : :
129 : : #include "postgres.h"
130 : :
131 : : #include <unistd.h>
132 : :
133 : : #include "access/slru.h"
134 : : #include "common/int.h"
135 : : #include "fmgr.h"
136 : : #include "funcapi.h"
137 : : #include "miscadmin.h"
138 : : #include "port/pg_bitutils.h"
139 : : #include "port/pg_numa.h"
140 : : #include "storage/lwlock.h"
141 : : #include "storage/pg_shmem.h"
142 : : #include "storage/shmem.h"
143 : : #include "storage/shmem_internal.h"
144 : : #include "storage/spin.h"
145 : : #include "utils/builtins.h"
146 : : #include "utils/tuplestore.h"
147 : :
148 : : typedef struct ShmemIndexEnt ShmemIndexEnt;
149 : :
150 : : /*
151 : : * Registered callbacks.
152 : : *
153 : : * During postmaster startup, we accumulate the callbacks from all subsystems
154 : : * in this list.
155 : : *
156 : : * This is in process private memory, although on Unix-like systems, we expect
157 : : * all the registrations to happen at postmaster startup time and be inherited
158 : : * by all the child processes via fork().
159 : : */
160 : : static List *registered_shmem_callbacks;
161 : :
162 : : /*
163 : : * In the shmem request phase, all the shmem areas requested with the
164 : : * ShmemRequest*() functions are accumulated in the 'pending_shmem_requests'
165 : : * list. The List, the ShmemRequest structs, and the 'options' are all
166 : : * allocated in TopMemoryContext.
167 : : */
168 : : typedef struct
169 : : {
170 : : ShmemStructOpts *options;
171 : : ShmemRequestKind kind;
172 : :
173 : : /* InitShmemIndexEntry() sets this pointer when the area is allocated */
174 : : ShmemIndexEnt *index_entry;
175 : : } ShmemRequest;
176 : :
177 : : static List *pending_shmem_requests; /* List of ShmemRequests */
178 : :
179 : : /*
180 : : * Per-process state machine, for sanity checking that we do things in the
181 : : * right order.
182 : : *
183 : : * Postmaster:
184 : : * INITIAL -> REQUESTING -> INITIALIZING -> DONE
185 : : *
186 : : * Backends in EXEC_BACKEND mode:
187 : : * INITIAL -> REQUESTING -> ATTACHING -> DONE
188 : : *
189 : : * Late request:
190 : : * DONE -> REQUESTING -> AFTER_STARTUP_ATTACH_OR_INIT -> DONE
191 : : */
192 : : enum shmem_request_state
193 : : {
194 : : /* Initial state */
195 : : SRS_INITIAL,
196 : :
197 : : /*
198 : : * When we start calling the shmem_request callbacks, we enter the
199 : : * SRS_REQUESTING phase. All ShmemRequestStruct calls happen in this
200 : : * state.
201 : : */
202 : : SRS_REQUESTING,
203 : :
204 : : /*
205 : : * Postmaster has finished all shmem requests, and is now initializing the
206 : : * shared memory segment. init_fn callbacks are called in this state.
207 : : */
208 : : SRS_INITIALIZING,
209 : :
210 : : /*
211 : : * A postmaster child process is starting up. attach_fn callbacks are
212 : : * called in this state.
213 : : */
214 : : SRS_ATTACHING,
215 : :
216 : : /* An after-startup allocation or attachment is in progress */
217 : : SRS_AFTER_STARTUP_ATTACH_OR_INIT,
218 : :
219 : : /* Normal state after shmem initialization / attachment */
220 : : SRS_DONE,
221 : : };
222 : : static enum shmem_request_state shmem_request_state = SRS_INITIAL;
223 : :
224 : : /*
225 : : * This is the first data structure stored in the shared memory segment, at
226 : : * the offset that PGShmemHeader->content_offset points to. Allocations by
227 : : * ShmemAlloc() are carved out of the space after this.
228 : : *
229 : : * For the base pointer and the total size of the shmem segment, we rely on
230 : : * the PGShmemHeader.
231 : : */
232 : : typedef struct ShmemAllocatorData
233 : : {
234 : : Size free_offset; /* offset to first free space from ShmemBase */
235 : :
236 : : /* protects 'free_offset' */
237 : : slock_t shmem_lock;
238 : :
239 : : HASHHDR *index; /* location of ShmemIndex */
240 : : size_t index_size; /* size of shmem region holding ShmemIndex */
241 : : LWLock index_lock; /* protects ShmemIndex */
242 : : } ShmemAllocatorData;
243 : :
244 : : #define ShmemIndexLock (&ShmemAllocator->index_lock)
245 : :
246 : : static void *ShmemAllocRaw(Size size, Size alignment, Size *allocated_size);
247 : :
248 : : /* shared memory global variables */
249 : :
250 : : static PGShmemHeader *ShmemSegHdr; /* shared mem segment header */
251 : : static void *ShmemBase; /* start address of shared memory */
252 : : static void *ShmemEnd; /* end+1 address of shared memory */
253 : :
254 : : static ShmemAllocatorData *ShmemAllocator;
255 : :
256 : : /*
257 : : * ShmemIndex is a global directory of shmem areas, itself also stored in the
258 : : * shared memory.
259 : : */
260 : : static HTAB *ShmemIndex;
261 : :
262 : : /* max size of data structure string name */
263 : : #define SHMEM_INDEX_KEYSIZE (48)
264 : :
265 : : /*
266 : : * # of additional entries to reserve in the shmem index table, for
267 : : * allocations after postmaster startup. (This is not a hard limit, the hash
268 : : * table can grow larger than that if there is shared memory available)
269 : : */
270 : : #define SHMEM_INDEX_ADDITIONAL_SIZE (128)
271 : :
272 : : /* this is a hash bucket in the shmem index table */
273 : : typedef struct ShmemIndexEnt
274 : : {
275 : : char key[SHMEM_INDEX_KEYSIZE]; /* string name */
276 : : void *location; /* location in shared mem */
277 : : Size size; /* # bytes requested for the structure */
278 : : Size allocated_size; /* # bytes actually allocated */
279 : : bool initialized; /* has the init callback been run? */
280 : : } ShmemIndexEnt;
281 : :
282 : : /* To get reliable results for NUMA inquiry we need to "touch pages" once */
283 : : static bool firstNumaTouch = true;
284 : :
285 : : static void CallShmemCallbacksAfterStartup(const ShmemCallbacks *callbacks);
286 : : static void ProcessShmemRequestsAfterStartup(const ShmemCallbacks *callbacks);
287 : : static void InitShmemIndexEntry(ShmemRequest *request);
288 : : static bool AttachShmemIndexEntry(ShmemRequest *request, bool missing_ok);
289 : :
290 : : Datum pg_numa_available(PG_FUNCTION_ARGS);
291 : :
292 : : /*
293 : : * ShmemRequestStruct() --- request a named shared memory area
294 : : *
295 : : * Subsystems call this to register their shared memory needs. This is
296 : : * usually done early in postmaster startup, before the shared memory segment
297 : : * has been created, so that the size can be included in the estimate for
298 : : * total amount of shared memory needed. We set aside a small amount of
299 : : * memory for allocations that happen later, for the benefit of non-preloaded
300 : : * extensions, but that should not be relied upon.
301 : : *
302 : : * This does not yet allocate the memory, but merely registers the need for
303 : : * it. The actual allocation happens later in the postmaster startup
304 : : * sequence.
305 : : *
306 : : * This must be called from a shmem_request callback function, registered with
307 : : * RegisterShmemCallbacks(). This enforces a coding pattern that works the
308 : : * same in normal Unix systems and with EXEC_BACKEND. On Unix systems, the
309 : : * shmem_request callbacks are called once, early in postmaster startup, and
310 : : * the child processes inherit the struct descriptors and any other
311 : : * per-process state from the postmaster. In EXEC_BACKEND mode, shmem_request
312 : : * callbacks are *also* called in each backend, at backend startup, to
313 : : * re-establish the struct descriptors. By calling the same function in both
314 : : * cases, we ensure that all the shmem areas are registered the same way in
315 : : * all processes.
316 : : *
317 : : * 'options' defines the name and size of the area, and any other optional
318 : : * features. Leave unused options as zeros. The options are copied to
319 : : * longer-lived memory, so it doesn't need to live after the
320 : : * ShmemRequestStruct() call and can point to a local variable in the calling
321 : : * function. The 'name' must point to a long-lived string though, only the
322 : : * pointer to it is copied.
323 : : */
324 : : void
189 heikki.linnakangas@i 325 :CBC 82386 : ShmemRequestStructWithOpts(const ShmemStructOpts *options)
326 : : {
327 : : ShmemStructOpts *options_copy;
328 : :
329 : 82386 : options_copy = MemoryContextAlloc(TopMemoryContext,
330 : : sizeof(ShmemStructOpts));
331 : 82386 : memcpy(options_copy, options, sizeof(ShmemStructOpts));
332 : :
333 : 82386 : ShmemRequestInternal(options_copy, SHMEM_KIND_STRUCT);
334 : 82385 : }
335 : :
336 : : /*
337 : : * Internal workhorse of ShmemRequestStruct() and ShmemRequestHash().
338 : : *
339 : : * Note: Unlike in the public ShmemRequestStruct() and ShmemRequestHash()
340 : : * functions, 'options' is *not* copied. It must be allocated in
341 : : * TopMemoryContext by the caller, and will be freed after the init/attach
342 : : * callbacks have been called. This allows ShmemRequestHash() to pass a
343 : : * pointer to the extended ShmemHashOpts struct instead.
344 : : */
345 : : void
346 : 101706 : ShmemRequestInternal(ShmemStructOpts *options, ShmemRequestKind kind)
347 : : {
348 : : MemoryContext oldcontext;
349 : : ShmemRequest *request;
350 : :
351 : : /* Check that we're in the right state */
12 352 [ - + ]: 101706 : if (shmem_request_state != SRS_REQUESTING)
12 heikki.linnakangas@i 353 [ # # ]:UBC 0 : elog(ERROR, "ShmemRequestStruct can only be called from a shmem_request callback");
354 : :
355 : : /* Check the options */
189 heikki.linnakangas@i 356 [ - + ]:CBC 101706 : if (options->name == NULL)
189 heikki.linnakangas@i 357 [ # # ]:UBC 0 : elog(ERROR, "shared memory request is missing 'name' option");
358 : :
12 heikki.linnakangas@i 359 [ + + ]:CBC 101706 : if (options->size == SHMEM_ATTACH_UNKNOWN_SIZE)
360 : : {
361 [ + + ]: 4 : if (ShmemIndex == NULL)
189 362 [ + - ]: 1 : elog(ERROR, "SHMEM_ATTACH_UNKNOWN_SIZE cannot be used during startup");
363 : : }
12 364 [ - + ]: 101702 : else if (options->size <= 0)
12 heikki.linnakangas@i 365 [ # # ]:UBC 0 : elog(ERROR, "invalid size %zd for shared memory request for \"%s\"",
366 : : options->size, options->name);
367 : :
189 heikki.linnakangas@i 368 [ + + - + ]:CBC 101705 : if (options->alignment != 0 && pg_nextpower2_size_t(options->alignment) != options->alignment)
189 heikki.linnakangas@i 369 [ # # ]:UBC 0 : elog(ERROR, "invalid alignment %zu for shared memory request for \"%s\"",
370 : : options->alignment, options->name);
371 : :
372 : : /* Check that it's not already registered in this process */
189 heikki.linnakangas@i 373 [ + + + + :CBC 4169494 : foreach_ptr(ShmemRequest, existing, pending_shmem_requests)
+ + ]
374 : : {
375 [ - + ]: 3966084 : if (strcmp(existing->options->name, options->name) == 0)
189 heikki.linnakangas@i 376 [ # # ]:UBC 0 : ereport(ERROR,
377 : : (errmsg("shared memory struct \"%s\" is already registered",
378 : : options->name)));
379 : : }
380 : :
381 : : /* Request looks valid, remember it */
45 heikki.linnakangas@i 382 :CBC 101705 : oldcontext = MemoryContextSwitchTo(TopMemoryContext);
56 michael@paquier.xyz 383 : 101705 : request = palloc_object(ShmemRequest);
189 heikki.linnakangas@i 384 : 101705 : request->options = options;
385 : 101705 : request->kind = kind;
45 386 : 101705 : request->index_entry = NULL;
189 387 : 101705 : pending_shmem_requests = lappend(pending_shmem_requests, request);
45 388 : 101705 : MemoryContextSwitchTo(oldcontext);
189 389 : 101705 : }
390 : :
391 : : /*
392 : : * ShmemGetRequestedSize() --- estimate the total size of all registered shared
393 : : * memory structures.
394 : : *
395 : : * This is called at postmaster startup, before the shared memory segment has
396 : : * been created.
397 : : */
398 : : size_t
399 : 2391 : ShmemGetRequestedSize(void)
400 : : {
401 : : size_t size;
402 : :
403 : : /* memory needed for the ShmemIndex */
404 : 2391 : size = hash_estimate_size(list_length(pending_shmem_requests) + SHMEM_INDEX_ADDITIONAL_SIZE,
405 : : sizeof(ShmemIndexEnt));
406 : 2391 : size = CACHELINEALIGN(size);
407 : :
408 : : /* memory needed for all the requested areas */
409 [ + - + + : 193693 : foreach_ptr(ShmemRequest, request, pending_shmem_requests)
+ + ]
410 : : {
411 : 188911 : size_t alignment = request->options->alignment;
412 : :
413 : : /* pad the start address for alignment like ShmemAllocRaw() does */
414 [ + + ]: 188911 : if (alignment < PG_CACHE_LINE_SIZE)
415 : 181738 : alignment = PG_CACHE_LINE_SIZE;
416 : 188911 : size = TYPEALIGN(alignment, size);
417 : :
418 : 188911 : size = add_size(size, request->options->size);
419 : : }
420 : :
421 : 2391 : return size;
422 : : }
423 : :
424 : : /*
425 : : * ShmemInitRequested() --- allocate and initialize requested shared memory
426 : : * structures.
427 : : *
428 : : * This is called once at postmaster startup, after the shared memory segment
429 : : * has been created.
430 : : */
431 : : void
432 : 1283 : ShmemInitRequested(void)
433 : : {
434 : : /* should be called only by the postmaster or a standalone backend */
435 [ - + ]: 1283 : Assert(!IsUnderPostmaster);
436 [ - + ]: 1283 : Assert(shmem_request_state == SRS_INITIALIZING);
437 : :
438 : : /*
439 : : * Initialize the ShmemIndex entries and perform basic initialization of
440 : : * all the requested memory areas. There are no concurrent processes yet,
441 : : * so no need for locking.
442 : : */
443 [ + - + + : 103933 : foreach_ptr(ShmemRequest, request, pending_shmem_requests)
+ + ]
444 : : {
445 : 101367 : InitShmemIndexEntry(request);
446 : : }
447 : :
448 : : /*
449 : : * Call the subsystem-specific init callbacks to finish initialization of
450 : : * all the areas.
451 : : */
452 [ + - + + : 59044 : foreach_ptr(const ShmemCallbacks, callbacks, registered_shmem_callbacks)
+ + ]
453 : : {
454 [ + + ]: 56478 : if (callbacks->init_fn)
455 : 52625 : callbacks->init_fn(callbacks->opaque_arg);
456 : : }
457 : :
458 : : /* Now we can mark all the areas as initialized and free the requests */
45 459 [ + - + + : 103933 : foreach_ptr(ShmemRequest, request, pending_shmem_requests)
+ + ]
460 : : {
461 : 101367 : request->index_entry->initialized = true;
462 : 101367 : pfree(request->options);
463 : : }
464 : 1283 : list_free_deep(pending_shmem_requests);
465 : 1283 : pending_shmem_requests = NIL;
466 : :
189 467 : 1283 : shmem_request_state = SRS_DONE;
468 : 1283 : }
469 : :
470 : : /*
471 : : * Re-establish process private state related to shmem areas.
472 : : *
473 : : * This is called at backend startup in EXEC_BACKEND mode, in every backend.
474 : : */
475 : : #ifdef EXEC_BACKEND
476 : : void
477 : : ShmemAttachRequested(void)
478 : : {
479 : : ListCell *lc;
480 : :
481 : : /* Must be initializing a (non-standalone) backend */
482 : : Assert(IsUnderPostmaster);
483 : : Assert(ShmemAllocator->index != NULL);
484 : : Assert(shmem_request_state == SRS_REQUESTING);
485 : : shmem_request_state = SRS_ATTACHING;
486 : :
487 : : LWLockAcquire(ShmemIndexLock, LW_SHARED);
488 : :
489 : : /*
490 : : * Attach to all the requested memory areas.
491 : : */
492 : : foreach_ptr(ShmemRequest, request, pending_shmem_requests)
493 : : {
494 : : AttachShmemIndexEntry(request, false);
495 : : pfree(request->options);
496 : : }
497 : : list_free_deep(pending_shmem_requests);
498 : : pending_shmem_requests = NIL;
499 : :
500 : : /* Call attach callbacks */
501 : : foreach(lc, registered_shmem_callbacks)
502 : : {
503 : : const ShmemCallbacks *callbacks = (const ShmemCallbacks *) lfirst(lc);
504 : :
505 : : if (callbacks->attach_fn)
506 : : callbacks->attach_fn(callbacks->opaque_arg);
507 : : }
508 : :
509 : : LWLockRelease(ShmemIndexLock);
510 : :
511 : : shmem_request_state = SRS_DONE;
512 : : }
513 : : #endif
514 : :
515 : : /*
516 : : * Insert requested shmem area into the shared memory index and initialize it.
517 : : *
518 : : * Note that this only does performs basic initialization depending on
519 : : * ShmemRequestKind, like setting the global pointer variable to the area for
520 : : * SHMEM_KIND_STRUCT or setting up the backend-private HTAB control struct.
521 : : * This does *not* call the subsystem-specific init callbacks. That's done
522 : : * later after all the shmem areas have been initialized or attached to.
523 : : */
524 : : static void
525 : 101376 : InitShmemIndexEntry(ShmemRequest *request)
526 : : {
527 : 101376 : const char *name = request->options->name;
528 : : ShmemIndexEnt *index_entry;
529 : : bool found;
530 : : size_t allocated_size;
531 : : void *structPtr;
532 : :
533 : : /* Size must be known at this point. */
24 534 [ - + ]: 101376 : Assert(request->options->size != SHMEM_ATTACH_UNKNOWN_SIZE);
535 : :
536 : : /* look it up in the shmem index */
537 : : index_entry = (ShmemIndexEnt *)
189 538 : 101376 : hash_search(ShmemIndex, name, HASH_ENTER_NULL, &found);
539 [ - + ]: 101376 : if (found)
189 heikki.linnakangas@i 540 [ # # ]:UBC 0 : elog(ERROR, "shared memory struct \"%s\" is already initialized", name);
189 heikki.linnakangas@i 541 [ - + ]:CBC 101376 : if (!index_entry)
542 : : {
543 : : /* tried to add it to the hash table, but there was no space */
189 heikki.linnakangas@i 544 [ # # ]:UBC 0 : ereport(ERROR,
545 : : (errcode(ERRCODE_OUT_OF_MEMORY),
546 : : errmsg("could not create ShmemIndex entry for data structure \"%s\"",
547 : : name)));
548 : : }
549 : :
550 : : /*
551 : : * We inserted the entry to the shared memory index. Allocate requested
552 : : * amount of shared memory for it, and initialize the index entry.
553 : : */
189 heikki.linnakangas@i 554 :CBC 101376 : structPtr = ShmemAllocRaw(request->options->size,
555 : 101376 : request->options->alignment,
556 : : &allocated_size);
557 [ + + ]: 101376 : if (structPtr == NULL)
558 : : {
559 : : /* out of memory; remove the failed ShmemIndex entry */
560 : 1 : hash_search(ShmemIndex, name, HASH_REMOVE, NULL);
561 [ + - ]: 1 : ereport(ERROR,
562 : : (errcode(ERRCODE_OUT_OF_MEMORY),
563 : : errmsg("not enough shared memory for data structure"
564 : : " \"%s\" (%zd bytes requested)",
565 : : name, request->options->size)));
566 : : }
567 : 101375 : index_entry->size = request->options->size;
568 : 101375 : index_entry->allocated_size = allocated_size;
569 : 101375 : index_entry->location = structPtr;
570 : :
571 : : /*
572 : : * The area is considered fully initialized only after the subsystem's
573 : : * init callback has been called. For now, perform only basic
574 : : * initialization based on the kind of shmem area it is.
575 : : */
45 576 : 101375 : index_entry->initialized = false;
189 577 [ + + + - ]: 101375 : switch (request->kind)
578 : : {
579 : 82115 : case SHMEM_KIND_STRUCT:
580 [ + - ]: 82115 : if (request->options->ptr)
581 : 82115 : *(request->options->ptr) = index_entry->location;
582 : 82115 : break;
583 : 10275 : case SHMEM_KIND_HASH:
584 : 10275 : shmem_hash_init(structPtr, request->options);
585 : 10275 : break;
586 : 8985 : case SHMEM_KIND_SLRU:
587 : 8985 : shmem_slru_init(structPtr, request->options);
588 : 8985 : break;
589 : : }
590 : :
591 : : /* return the pointer to the entry to the caller */
45 592 : 101375 : request->index_entry = index_entry;
189 593 : 101375 : }
594 : :
595 : : /*
596 : : * Look up a named shmem area in the shared memory index and attach to it.
597 : : *
598 : : * Note that this only performs the basic attachment actions depending on
599 : : * ShmemRequestKind, like setting the global pointer variable to the area for
600 : : * SHMEM_KIND_STRUCT or setting up the backend-private HTAB control struct.
601 : : * This does *not* call the subsystem-specific attach callbacks. That's done
602 : : * later after all the shmem areas have been initialized or attached to.
603 : : */
604 : : static bool
605 : 11 : AttachShmemIndexEntry(ShmemRequest *request, bool missing_ok)
606 : : {
607 : 11 : const char *name = request->options->name;
608 : : ShmemIndexEnt *index_entry;
609 : :
610 : : /* Look it up in the shmem index */
611 : : index_entry = (ShmemIndexEnt *)
612 : 11 : hash_search(ShmemIndex, name, HASH_FIND, NULL);
613 [ - + ]: 11 : if (!index_entry)
614 : : {
189 heikki.linnakangas@i 615 [ # # ]:UBC 0 : if (!missing_ok)
616 [ # # ]: 0 : ereport(ERROR,
617 : : (errmsg("could not find ShmemIndex entry for data structure \"%s\"",
618 : : request->options->name)));
619 : 0 : return false;
620 : : }
621 : :
622 : : /*
623 : : * If it was previously allocated but not fully initialized, error out.
624 : : * There is currently no way of retrying or cleaning up an uninitialized
625 : : * entry, it just lingers until the server is shut down. But this can
626 : : * only happen when allocating areas after postmaster startup, and it's
627 : : * unlikely that you could successfully retry anyway. The most likely
628 : : * reason for failed initialization is that you are out of shared memory
629 : : * and retrying won't help with that.
630 : : */
45 heikki.linnakangas@i 631 [ - + ]:CBC 11 : if (!index_entry->initialized)
45 heikki.linnakangas@i 632 [ # # ]:UBC 0 : ereport(ERROR,
633 : : (errmsg("cannot attach to shared memory struct \"%s\" because it was not fully initialized",
634 : : request->options->name)));
635 : :
636 : : /* Check that the size in the index matches the request */
189 heikki.linnakangas@i 637 [ + + ]:CBC 11 : if (index_entry->size != request->options->size &&
638 [ + + ]: 3 : request->options->size != SHMEM_ATTACH_UNKNOWN_SIZE)
639 : : {
640 [ + - ]: 1 : ereport(ERROR,
641 : : (errmsg("shared memory struct \"%s\" was created with"
642 : : " different size: existing %zu, requested %zd",
643 : : name, index_entry->size, request->options->size)));
644 : : }
645 : :
646 : : /*
647 : : * Re-establish the caller's pointer variable, or do other actions to
648 : : * attach depending on the kind of shmem area it is.
649 : : */
650 [ + - - - ]: 10 : switch (request->kind)
651 : : {
652 : 10 : case SHMEM_KIND_STRUCT:
653 [ + - ]: 10 : if (request->options->ptr)
654 : 10 : *(request->options->ptr) = index_entry->location;
655 : 10 : break;
189 heikki.linnakangas@i 656 :UBC 0 : case SHMEM_KIND_HASH:
657 : 0 : shmem_hash_attach(index_entry->location, request->options);
658 : 0 : break;
659 : 0 : case SHMEM_KIND_SLRU:
660 : 0 : shmem_slru_attach(index_entry->location, request->options);
661 : 0 : break;
662 : : }
663 : :
45 heikki.linnakangas@i 664 :CBC 10 : request->index_entry = index_entry;
665 : :
189 666 : 10 : return true;
667 : : }
668 : :
669 : : /*
670 : : * InitShmemAllocator() --- set up basic pointers to shared memory.
671 : : *
672 : : * Called at postmaster or stand-alone backend startup, to initialize the
673 : : * allocator's data structure in the shared memory segment. In EXEC_BACKEND,
674 : : * this is also called at backend startup, to set up pointers to the
675 : : * already-initialized data structure.
676 : : */
677 : : void
255 678 : 1283 : InitShmemAllocator(PGShmemHeader *seghdr)
679 : : {
680 : : Size offset;
681 : : int64 hash_nelems;
682 : : HASHCTL info;
683 : : int hash_flags;
684 : :
685 : : #ifndef EXEC_BACKEND
200 686 [ - + ]: 1283 : Assert(!IsUnderPostmaster);
687 : : #endif
255 688 [ - + ]: 1283 : Assert(seghdr != NULL);
689 : :
189 690 [ - + ]: 1283 : if (IsUnderPostmaster)
691 : : {
189 heikki.linnakangas@i 692 [ # # ]:UBC 0 : Assert(shmem_request_state == SRS_INITIAL);
693 : : }
694 : : else
695 : : {
189 heikki.linnakangas@i 696 [ - + ]:CBC 1283 : Assert(shmem_request_state == SRS_REQUESTING);
697 : 1283 : shmem_request_state = SRS_INITIALIZING;
698 : : }
699 : :
700 : : /*
701 : : * We assume the pointer and offset are MAXALIGN. Not a hard requirement,
702 : : * but it's true today and keeps the math below simpler.
703 : : */
255 704 [ - + ]: 1283 : Assert(seghdr == (void *) MAXALIGN(seghdr));
705 [ - + ]: 1283 : Assert(seghdr->content_offset == MAXALIGN(seghdr->content_offset));
706 : :
707 : : /*
708 : : * Allocations after this point should go through ShmemAlloc, which
709 : : * expects to allocate everything on cache line boundaries. Make sure the
710 : : * first allocation begins on a cache line boundary.
711 : : */
200 712 : 1283 : offset = CACHELINEALIGN(seghdr->content_offset + sizeof(ShmemAllocatorData));
713 [ - + ]: 1283 : if (offset > seghdr->totalsize)
200 heikki.linnakangas@i 714 [ # # ]:UBC 0 : ereport(ERROR,
715 : : (errcode(ERRCODE_OUT_OF_MEMORY),
716 : : errmsg("out of shared memory (%zu bytes requested)",
717 : : offset)));
718 : :
719 : : /*
720 : : * In postmaster or stand-alone backend, initialize the shared memory
721 : : * allocator so that we can allocate shared memory for ShmemIndex using
722 : : * ShmemAlloc(). In a regular backend just set up the pointers required
723 : : * by ShmemAlloc().
724 : : */
200 heikki.linnakangas@i 725 :CBC 1283 : ShmemAllocator = (ShmemAllocatorData *) ((char *) seghdr + seghdr->content_offset);
726 [ + - ]: 1283 : if (!IsUnderPostmaster)
727 : : {
728 : 1283 : SpinLockInit(&ShmemAllocator->shmem_lock);
729 : 1283 : ShmemAllocator->free_offset = offset;
730 : 1283 : LWLockInitialize(&ShmemAllocator->index_lock, LWTRANCHE_SHMEM_INDEX);
731 : : }
732 : :
685 peter@eisentraut.org 733 : 1283 : ShmemSegHdr = seghdr;
734 : 1283 : ShmemBase = seghdr;
735 : 1283 : ShmemEnd = (char *) ShmemBase + seghdr->totalsize;
736 : :
737 : : /*
738 : : * Create (or attach to) the shared memory index of shmem areas.
739 : : *
740 : : * This is the same initialization as ShmemInitHash() does, but we cannot
741 : : * use ShmemInitHash() here because it relies on ShmemIndex being already
742 : : * initialized.
743 : : */
189 heikki.linnakangas@i 744 : 1283 : hash_nelems = list_length(pending_shmem_requests) + SHMEM_INDEX_ADDITIONAL_SIZE;
745 : :
200 746 : 1283 : info.keysize = SHMEM_INDEX_KEYSIZE;
747 : 1283 : info.entrysize = sizeof(ShmemIndexEnt);
191 748 : 1283 : hash_flags = HASH_ELEM | HASH_STRINGS | HASH_FIXED_SIZE;
749 : :
200 750 [ + - ]: 1283 : if (!IsUnderPostmaster)
751 : : {
189 752 : 1283 : ShmemAllocator->index_size = hash_estimate_size(hash_nelems, info.entrysize);
191 753 : 1283 : ShmemAllocator->index = (HASHHDR *) ShmemAlloc(ShmemAllocator->index_size);
754 : : }
755 : 2566 : ShmemIndex = shmem_hash_create(ShmemAllocator->index,
756 : 1283 : ShmemAllocator->index_size,
757 : : IsUnderPostmaster,
758 : : "ShmemIndex", hash_nelems,
759 : : &info, hash_flags);
200 760 [ - + ]: 1283 : Assert(ShmemIndex != NULL);
761 : :
762 : : /*
763 : : * Add an entry for ShmemIndex itself into ShmemIndex, so that it's
764 : : * visible in the pg_shmem_allocations view
765 : : */
194 766 [ + - ]: 1283 : if (!IsUnderPostmaster)
767 : : {
768 : : bool found;
769 : : ShmemIndexEnt *result = (ShmemIndexEnt *)
770 : 1283 : hash_search(ShmemIndex, "ShmemIndex", HASH_ENTER, &found);
771 : :
772 [ - + ]: 1283 : Assert(!found);
191 773 : 1283 : result->size = ShmemAllocator->index_size;
774 : 1283 : result->allocated_size = ShmemAllocator->index_size;
194 775 : 1283 : result->location = ShmemAllocator->index;
45 776 : 1283 : result->initialized = true;
777 : : }
11052 scrappy@hub.org 778 : 1283 : }
779 : :
780 : : /*
781 : : * Reset state on postmaster crash restart.
782 : : */
783 : : void
189 heikki.linnakangas@i 784 : 6 : ResetShmemAllocator(void)
785 : : {
786 [ - + ]: 6 : Assert(!IsUnderPostmaster);
12 787 : 6 : ShmemAllocator = NULL;
788 : 6 : ShmemIndex = NULL;
189 789 : 6 : shmem_request_state = SRS_INITIAL;
790 : :
791 : 6 : pending_shmem_requests = NIL;
792 : :
793 : : /*
794 : : * Note that we don't clear the registered callbacks. We will need to
795 : : * call them again as we restart
796 : : */
797 : 6 : }
798 : :
799 : : /*
800 : : * ShmemAlloc -- allocate max-aligned chunk from shared memory
801 : : *
802 : : * Throws error if request cannot be satisfied.
803 : : *
804 : : * Assumes ShmemSegHdr is initialized.
805 : : */
806 : : void *
7302 tgl@sss.pgh.pa.us 807 : 1283 : ShmemAlloc(Size size)
808 : : {
809 : : void *newSpace;
810 : : Size allocated_size;
811 : :
189 heikki.linnakangas@i 812 : 1283 : newSpace = ShmemAllocRaw(size, 0, &allocated_size);
3693 tgl@sss.pgh.pa.us 813 [ - + ]: 1283 : if (!newSpace)
3693 tgl@sss.pgh.pa.us 814 [ # # ]:UBC 0 : ereport(ERROR,
815 : : (errcode(ERRCODE_OUT_OF_MEMORY),
816 : : errmsg("out of shared memory (%zu bytes requested)",
817 : : size)));
3693 tgl@sss.pgh.pa.us 818 :CBC 1283 : return newSpace;
819 : : }
820 : :
821 : : /*
822 : : * ShmemAllocNoError -- allocate max-aligned chunk from shared memory
823 : : *
824 : : * As ShmemAlloc, but returns NULL if out of space, rather than erroring.
825 : : */
826 : : void *
3693 tgl@sss.pgh.pa.us 827 :UBC 0 : ShmemAllocNoError(Size size)
828 : : {
829 : : Size allocated_size;
830 : :
189 heikki.linnakangas@i 831 : 0 : return ShmemAllocRaw(size, 0, &allocated_size);
832 : : }
833 : :
834 : : /*
835 : : * ShmemAllocRaw -- allocate align chunk and return allocated size
836 : : *
837 : : * Also sets *allocated_size to the number of bytes allocated, which will
838 : : * be equal to the number requested plus any padding we choose to add.
839 : : *
840 : : * Returns NULL in case space can not be allocated.
841 : : */
842 : : static void *
189 heikki.linnakangas@i 843 :CBC 102659 : ShmemAllocRaw(Size size, Size alignment, Size *allocated_size)
844 : : {
845 : : Size rawStart;
846 : : Size newStart;
847 : : Size newFree;
848 : : void *newSpace;
849 : :
850 : : /*
851 : : * Ensure all space is adequately aligned. We used to only MAXALIGN this
852 : : * space but experience has proved that on modern systems that is not good
853 : : * enough. Many parts of the system are very sensitive to critical data
854 : : * structures getting split across cache line boundaries. To avoid that,
855 : : * attempt to align the beginning of the allocation to a cache line
856 : : * boundary. The calling code will still need to be careful about how it
857 : : * uses the allocated space - e.g. by padding each element in an array of
858 : : * structures out to a power-of-two size - but without this, even that
859 : : * won't be sufficient.
860 : : */
861 [ + + ]: 102659 : if (alignment < PG_CACHE_LINE_SIZE)
862 : 98810 : alignment = PG_CACHE_LINE_SIZE;
863 : :
4014 rhaas@postgresql.org 864 [ - + ]: 102659 : Assert(ShmemSegHdr != NULL);
865 : :
200 heikki.linnakangas@i 866 : 102659 : SpinLockAcquire(&ShmemAllocator->shmem_lock);
867 : :
189 868 : 102659 : rawStart = ShmemAllocator->free_offset;
869 : 102659 : newStart = TYPEALIGN(alignment, rawStart);
870 : :
871 : : /*
872 : : * newFree = newStart + size, which is the start of the remaining space
873 : : * after the allocation. If it exceeds the shmem segment size, we don't
874 : : * have enough space available.
875 : : */
24 876 [ + - ]: 102659 : if (!pg_add_size_overflow(newStart, size, &newFree) &&
877 [ + + ]: 102659 : newFree <= ShmemSegHdr->totalsize)
878 : : {
683 peter@eisentraut.org 879 : 102658 : newSpace = (char *) ShmemBase + newStart;
255 heikki.linnakangas@i 880 : 102658 : ShmemAllocator->free_offset = newFree;
881 : : }
882 : : else
10627 bruce@momjian.us 883 : 1 : newSpace = NULL;
884 : :
200 heikki.linnakangas@i 885 : 102659 : SpinLockRelease(&ShmemAllocator->shmem_lock);
886 : :
887 : : /* note this assert is okay with newSpace == NULL */
189 888 [ - + ]: 102659 : Assert(newSpace == (void *) TYPEALIGN(alignment, newSpace));
889 : :
890 : 102659 : *allocated_size = newFree - rawStart;
10268 bruce@momjian.us 891 : 102659 : return newSpace;
892 : : }
893 : :
894 : : /*
895 : : * ShmemAddrIsValid -- test if an address refers to shared memory
896 : : *
897 : : * Returns true if the pointer points within the shared memory segment.
898 : : */
899 : : bool
5726 heikki.linnakangas@i 900 : 1703 : ShmemAddrIsValid(const void *addr)
901 : : {
6553 tgl@sss.pgh.pa.us 902 [ + - + - ]: 1703 : return (addr >= ShmemBase) && (addr < ShmemEnd);
903 : : }
904 : :
905 : : /*
906 : : * Register callbacks that define a shared memory area (or multiple areas).
907 : : *
908 : : * The system will call the callbacks at different stages of postmaster or
909 : : * backend startup, to allocate and initialize the area.
910 : : *
911 : : * This is normally called early during postmaster startup, but if the
912 : : * SHMEM_CALLBACKS_ALLOW_AFTER_STARTUP is set, this can also be used after
913 : : * startup, although after startup there's no guarantee that there's enough
914 : : * shared memory available. When called after startup, this immediately calls
915 : : * the right callbacks depending on whether another backend had already
916 : : * initialized the area.
917 : : *
918 : : * Note: In EXEC_BACKEND mode, this needs to be called in every backend
919 : : * process. That's needed because we cannot pass down the callback function
920 : : * pointers from the postmaster process, because different processes may have
921 : : * loaded libraries to different addresses.
922 : : */
923 : : void
189 heikki.linnakangas@i 924 : 56983 : RegisterShmemCallbacks(const ShmemCallbacks *callbacks)
925 : : {
47 926 [ + + ]: 56983 : if (shmem_request_state == SRS_DONE)
927 : : {
928 : : /*
929 : : * After-startup initialization or attachment. Call the appropriate
930 : : * callbacks immediately.
931 : : *
932 : : * This is not allowed from the postmaster, because the postmaster
933 : : * cannot acquire locks.
934 : : */
189 935 [ - + ]: 22 : if ((callbacks->flags & SHMEM_CALLBACKS_ALLOW_AFTER_STARTUP) == 0)
189 heikki.linnakangas@i 936 [ # # ]:UBC 0 : elog(ERROR, "cannot request shared memory at this time");
47 heikki.linnakangas@i 937 [ + + - + ]:CBC 22 : Assert(IsUnderPostmaster || !IsPostmasterEnvironment);
938 : :
189 939 : 22 : CallShmemCallbacksAfterStartup(callbacks);
940 : : }
47 941 [ + - ]: 56961 : else if (shmem_request_state == SRS_INITIAL)
942 : : {
943 : : /* Remember the callbacks for later */
189 944 : 56961 : registered_shmem_callbacks = lappend(registered_shmem_callbacks,
945 : : unconstify(ShmemCallbacks *, callbacks));
946 : : }
947 : : else
47 heikki.linnakangas@i 948 [ # # ]:UBC 0 : elog(ERROR, "cannot request shared memory at this time");
189 heikki.linnakangas@i 949 :CBC 56978 : }
950 : :
951 : : /*
952 : : * Register a shmem area (or multiple areas) after startup.
953 : : */
954 : : static void
955 : 22 : CallShmemCallbacksAfterStartup(const ShmemCallbacks *callbacks)
956 : : {
957 [ - + ]: 22 : Assert(shmem_request_state == SRS_DONE);
958 [ - + ]: 22 : Assert(pending_shmem_requests == NIL);
959 : :
45 960 [ + + ]: 22 : PG_TRY();
961 : : {
962 : 22 : shmem_request_state = SRS_REQUESTING;
963 : :
964 : : /*
965 : : * Call the request callback first. The callback makes
966 : : * ShmemRequest*() calls for each shmem area, adding them to
967 : : * pending_shmem_requests.
968 : : */
969 [ + - ]: 22 : if (callbacks->request_fn)
970 : 22 : callbacks->request_fn(callbacks->opaque_arg);
971 : :
972 : : /* Process all the requests */
973 : 22 : shmem_request_state = SRS_AFTER_STARTUP_ATTACH_OR_INIT;
974 [ + - ]: 22 : if (pending_shmem_requests != NIL)
975 : 22 : ProcessShmemRequestsAfterStartup(callbacks);
976 : : }
977 : 5 : PG_FINALLY();
978 : : {
979 [ + - + + : 66 : foreach_ptr(ShmemRequest, request, pending_shmem_requests)
+ + ]
980 : 22 : pfree(request->options);
981 : 22 : list_free_deep(pending_shmem_requests);
982 : 22 : pending_shmem_requests = NIL;
983 : :
189 984 : 22 : shmem_request_state = SRS_DONE;
985 : : }
45 986 [ + + ]: 22 : PG_END_TRY();
987 : 17 : }
988 : :
989 : : static void
990 : 22 : ProcessShmemRequestsAfterStartup(const ShmemCallbacks *callbacks)
991 : : {
992 : : bool found_any;
993 : : bool notfound_any;
994 : :
995 : : /* There should be some requests to process */
996 [ - + ]: 22 : Assert(pending_shmem_requests != NIL);
997 : :
998 : : /* Caller manages the global state variable */
999 [ - + ]: 22 : Assert(shmem_request_state == SRS_AFTER_STARTUP_ATTACH_OR_INIT);
1000 : :
1001 : : /*
1002 : : * Hold ShmemIndexLock while we allocate all the shmem entries and run all
1003 : : * the initializers.
1004 : : */
200 1005 : 22 : LWLockAcquire(ShmemIndexLock, LW_EXCLUSIVE);
1006 : :
1007 : : /*
1008 : : * Check if the requested shared memory areas have already been
1009 : : * initialized. We assume all the areas requested by the request callback
1010 : : * to form a coherent unit such that they're all already initialized or
1011 : : * none. Otherwise it would be ambiguous which callback, init or attach,
1012 : : * to callback afterwards.
1013 : : */
189 1014 : 22 : found_any = notfound_any = false;
1015 [ + - + + : 62 : foreach_ptr(ShmemRequest, request, pending_shmem_requests)
+ + ]
1016 : : {
1017 : : ShmemIndexEnt *index_entry;
1018 : :
1019 : : index_entry = (ShmemIndexEnt *)
45 1020 : 22 : hash_search(ShmemIndex, request->options->name, HASH_FIND, NULL);
1021 [ + + ]: 22 : if (index_entry)
1022 : : {
1023 : : /*
1024 : : * Check for a half-initialized area. (See also similar check in
1025 : : * AttachShmemIndexEntry())
1026 : : */
1027 [ + + ]: 12 : if (!index_entry->initialized)
1028 [ + - ]: 1 : ereport(ERROR,
1029 : : (errmsg("cannot attach to shared memory struct \"%s\" because it was not fully initialized",
1030 : : request->options->name)));
189 1031 : 11 : found_any = true;
1032 : : }
1033 : : else
1034 : : {
24 1035 [ + + ]: 10 : if (request->options->size == SHMEM_ATTACH_UNKNOWN_SIZE)
1036 [ + - ]: 1 : ereport(ERROR,
1037 : : (errmsg("cannot attach to shared memory struct \"%s\" because it does not exist",
1038 : : request->options->name),
1039 : : errdetail("SHMEM_ATTACH_UNKNOWN_SIZE can only be used to attach to an existing shared memory structure.")));
189 1040 : 9 : notfound_any = true;
1041 : : }
1042 : : }
1043 [ + + - + ]: 20 : if (found_any && notfound_any)
97 heikki.linnakangas@i 1044 [ # # ]:UBC 0 : elog(ERROR, "some of the requested shmem areas have already been initialized");
1045 : :
1046 : : /*
1047 : : * Allocate or attach all the shmem areas requested by the request_fn
1048 : : * callback.
1049 : : */
189 heikki.linnakangas@i 1050 [ + - + + :CBC 56 : foreach_ptr(ShmemRequest, request, pending_shmem_requests)
+ + ]
1051 : : {
1052 [ + + ]: 20 : if (found_any)
1053 : 11 : AttachShmemIndexEntry(request, false);
1054 : : else
1055 : 9 : InitShmemIndexEntry(request);
1056 : : }
1057 : :
1058 : : /* Finish by calling the appropriate subsystem-specific callback */
1059 [ + + ]: 18 : if (found_any)
1060 : : {
1061 [ + + ]: 10 : if (callbacks->attach_fn)
1062 : 6 : callbacks->attach_fn(callbacks->opaque_arg);
1063 : : }
1064 : : else
1065 : : {
1066 [ + + ]: 8 : if (callbacks->init_fn)
1067 : 6 : callbacks->init_fn(callbacks->opaque_arg);
1068 : : }
1069 : :
45 1070 [ + - + + : 51 : foreach_ptr(ShmemRequest, request, pending_shmem_requests)
+ + ]
1071 : : {
1072 : 17 : request->index_entry->initialized = true;
1073 : : }
1074 : :
7586 tgl@sss.pgh.pa.us 1075 : 17 : LWLockRelease(ShmemIndexLock);
189 heikki.linnakangas@i 1076 : 17 : }
1077 : :
1078 : : /*
1079 : : * Call all shmem request callbacks.
1080 : : */
1081 : : void
1082 : 1287 : ShmemCallRequestCallbacks(void)
1083 : : {
1084 : : ListCell *lc;
1085 : :
1086 [ - + ]: 1287 : Assert(shmem_request_state == SRS_INITIAL);
1087 : 1287 : shmem_request_state = SRS_REQUESTING;
1088 : :
1089 [ + - + + : 57941 : foreach(lc, registered_shmem_callbacks)
+ + ]
1090 : : {
1091 : 56655 : const ShmemCallbacks *callbacks = (const ShmemCallbacks *) lfirst(lc);
1092 : :
1093 [ + - ]: 56655 : if (callbacks->request_fn)
1094 : 56655 : callbacks->request_fn(callbacks->opaque_arg);
1095 : : }
11052 scrappy@hub.org 1096 : 1286 : }
1097 : :
1098 : : /*
1099 : : * ShmemInitStruct -- Create/attach to a structure in shared memory.
1100 : : *
1101 : : * This is called during initialization to find or allocate
1102 : : * a data structure in shared memory. If no other process
1103 : : * has created the structure, this routine allocates space
1104 : : * for it. If it exists already, a pointer to the existing
1105 : : * structure is returned.
1106 : : *
1107 : : * Returns: pointer to the object. *foundPtr is set true if the object was
1108 : : * already in the shmem index (hence, already initialized).
1109 : : *
1110 : : * Note: This is a legacy interface, kept for backwards compatibility with
1111 : : * extensions. Use ShmemRequestStruct() in new code!
1112 : : */
1113 : : void *
189 heikki.linnakangas@i 1114 :UBC 0 : ShmemInitStruct(const char *name, Size size, bool *foundPtr)
1115 : : {
1116 : 0 : void *ptr = NULL;
1117 : 0 : ShmemStructOpts options = {
1118 : : .name = name,
1119 : : .size = size,
1120 : : .ptr = &ptr,
1121 : : };
1122 : 0 : ShmemRequest request = {&options, SHMEM_KIND_STRUCT};
1123 : :
1124 [ # # # # : 0 : Assert(shmem_request_state == SRS_DONE ||
# # ]
1125 : : shmem_request_state == SRS_INITIALIZING ||
1126 : : shmem_request_state == SRS_REQUESTING);
1127 : :
1128 : 0 : LWLockAcquire(ShmemIndexLock, LW_EXCLUSIVE);
1129 : :
1130 : : /* Look up the existing entry if any */
18 1131 : 0 : *foundPtr = AttachShmemIndexEntry(&request, true);
1132 : :
1133 : : /* Initialize it if not found */
189 1134 [ # # ]: 0 : if (!*foundPtr)
1135 : : {
1136 : 0 : InitShmemIndexEntry(&request);
1137 : : /* no additional initialization needed */
45 1138 : 0 : request.index_entry->initialized = true;
1139 : : }
1140 : :
189 1141 : 0 : LWLockRelease(ShmemIndexLock);
1142 : :
1143 [ # # ]: 0 : Assert(ptr != NULL);
1144 : 0 : return ptr;
1145 : : }
1146 : :
1147 : : /* SQL SRF showing allocated shared memory */
1148 : : Datum
2468 rhaas@postgresql.org 1149 :CBC 4 : pg_get_shmem_allocations(PG_FUNCTION_ARGS)
1150 : : {
1151 : : #define PG_GET_SHMEM_SIZES_COLS 4
1152 : 4 : ReturnSetInfo *rsinfo = (ReturnSetInfo *) fcinfo->resultinfo;
1153 : : HASH_SEQ_STATUS hstat;
1154 : : ShmemIndexEnt *ent;
2342 tgl@sss.pgh.pa.us 1155 : 4 : Size named_allocated = 0;
1156 : : Datum values[PG_GET_SHMEM_SIZES_COLS];
1157 : : bool nulls[PG_GET_SHMEM_SIZES_COLS];
1158 : :
1455 michael@paquier.xyz 1159 : 4 : InitMaterializedSRF(fcinfo, 0);
1160 : :
2468 rhaas@postgresql.org 1161 : 4 : LWLockAcquire(ShmemIndexLock, LW_SHARED);
1162 : :
1163 : 4 : hash_seq_init(&hstat, ShmemIndex);
1164 : :
1165 : : /* output all allocated entries */
1166 : 4 : memset(nulls, 0, sizeof(nulls));
1167 [ + + ]: 326 : while ((ent = (ShmemIndexEnt *) hash_seq_search(&hstat)) != NULL)
1168 : : {
1169 : 322 : values[0] = CStringGetTextDatum(ent->key);
1170 : 322 : values[1] = Int64GetDatum((char *) ent->location - (char *) ShmemSegHdr);
1171 : 322 : values[2] = Int64GetDatum(ent->size);
1172 : 322 : values[3] = Int64GetDatum(ent->allocated_size);
1173 : 322 : named_allocated += ent->allocated_size;
1174 : :
1680 michael@paquier.xyz 1175 : 322 : tuplestore_putvalues(rsinfo->setResult, rsinfo->setDesc,
1176 : : values, nulls);
1177 : : }
1178 : :
1179 : : /* output shared memory allocated but not counted via the shmem index */
2468 rhaas@postgresql.org 1180 : 4 : values[0] = CStringGetTextDatum("<anonymous>");
1181 : 4 : nulls[1] = true;
255 heikki.linnakangas@i 1182 : 4 : values[2] = Int64GetDatum(ShmemAllocator->free_offset - named_allocated);
2468 rhaas@postgresql.org 1183 : 4 : values[3] = values[2];
1680 michael@paquier.xyz 1184 : 4 : tuplestore_putvalues(rsinfo->setResult, rsinfo->setDesc, values, nulls);
1185 : :
1186 : : /* output as-of-yet unused shared memory */
2468 rhaas@postgresql.org 1187 : 4 : nulls[0] = true;
255 heikki.linnakangas@i 1188 : 4 : values[1] = Int64GetDatum(ShmemAllocator->free_offset);
2468 rhaas@postgresql.org 1189 : 4 : nulls[1] = false;
255 heikki.linnakangas@i 1190 : 4 : values[2] = Int64GetDatum(ShmemSegHdr->totalsize - ShmemAllocator->free_offset);
2468 rhaas@postgresql.org 1191 : 4 : values[3] = values[2];
1680 michael@paquier.xyz 1192 : 4 : tuplestore_putvalues(rsinfo->setResult, rsinfo->setDesc, values, nulls);
1193 : :
2468 rhaas@postgresql.org 1194 : 4 : LWLockRelease(ShmemIndexLock);
1195 : :
1196 : 4 : return (Datum) 0;
1197 : : }
1198 : :
1199 : : /*
1200 : : * SQL SRF showing NUMA memory nodes for allocated shared memory
1201 : : *
1202 : : * Compared to pg_get_shmem_allocations(), this function does not return
1203 : : * information about shared anonymous allocations and unused shared memory.
1204 : : */
1205 : : Datum
553 tomas.vondra@postgre 1206 : 4 : pg_get_shmem_allocations_numa(PG_FUNCTION_ARGS)
1207 : : {
1208 : : #define PG_GET_SHMEM_NUMA_SIZES_COLS 3
1209 : 4 : ReturnSetInfo *rsinfo = (ReturnSetInfo *) fcinfo->resultinfo;
1210 : : HASH_SEQ_STATUS hstat;
1211 : : ShmemIndexEnt *ent;
1212 : : Datum values[PG_GET_SHMEM_NUMA_SIZES_COLS];
1213 : : bool nulls[PG_GET_SHMEM_NUMA_SIZES_COLS];
1214 : : Size os_page_size;
1215 : : void **page_ptrs;
1216 : : int *pages_status;
1217 : : uint64 shm_total_page_count,
1218 : : shm_ent_page_count,
1219 : : max_nodes;
1220 : : Size *nodes;
1221 : :
1222 [ + - ]: 4 : if (pg_numa_init() == -1)
20 michael@paquier.xyz 1223 [ + - ]:GNC 4 : ereport(ERROR,
1224 : : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1225 : : errmsg("libnuma initialization failed or NUMA is not supported on this platform")));
1226 : :
553 tomas.vondra@postgre 1227 :UBC 0 : InitMaterializedSRF(fcinfo, 0);
1228 : :
1229 : 0 : max_nodes = pg_numa_get_max_node();
259 1230 : 0 : nodes = palloc_array(Size, max_nodes + 2);
1231 : :
1232 : : /*
1233 : : * Shared memory allocations can vary in size and may not align with OS
1234 : : * memory page boundaries, while NUMA queries work on pages.
1235 : : *
1236 : : * To correctly map each allocation to NUMA nodes, we need to: 1.
1237 : : * Determine the OS memory page size. 2. Align each allocation's start/end
1238 : : * addresses to page boundaries. 3. Query NUMA node information for all
1239 : : * pages spanning the allocation.
1240 : : */
551 1241 : 0 : os_page_size = pg_get_shmem_pagesize();
1242 : :
1243 : : /*
1244 : : * Allocate memory for page pointers and status based on total shared
1245 : : * memory size. This simplified approach allocates enough space for all
1246 : : * pages in shared memory rather than calculating the exact requirements
1247 : : * for each segment.
1248 : : *
1249 : : * Add 1, because we don't know how exactly the segments align to OS
1250 : : * pages, so the allocation might use one more memory page. In practice
1251 : : * this is not very likely, and moreover we have more entries, each of
1252 : : * them using only fraction of the total pages.
1253 : : */
553 1254 : 0 : shm_total_page_count = (ShmemSegHdr->totalsize / os_page_size) + 1;
306 michael@paquier.xyz 1255 : 0 : page_ptrs = palloc0_array(void *, shm_total_page_count);
1256 : 0 : pages_status = palloc_array(int, shm_total_page_count);
1257 : :
553 tomas.vondra@postgre 1258 [ # # ]: 0 : if (firstNumaTouch)
1259 [ # # ]: 0 : elog(DEBUG1, "NUMA: page-faulting shared memory segments for proper NUMA readouts");
1260 : :
1261 : 0 : LWLockAcquire(ShmemIndexLock, LW_SHARED);
1262 : :
1263 : 0 : hash_seq_init(&hstat, ShmemIndex);
1264 : :
1265 : : /* output all allocated entries */
1266 [ # # ]: 0 : while ((ent = (ShmemIndexEnt *) hash_seq_search(&hstat)) != NULL)
1267 : : {
1268 : : char *startptr,
1269 : : *endptr;
1270 : : Size total_len;
1271 : :
1272 : : /*
1273 : : * Calculate the range of OS pages used by this segment. The segment
1274 : : * may start / end half-way through a page, we want to count these
1275 : : * pages too. So we align the start/end pointers down/up, and then
1276 : : * calculate the number of pages from that.
1277 : : */
1278 : 0 : startptr = (char *) TYPEALIGN_DOWN(os_page_size, ent->location);
1279 : 0 : endptr = (char *) TYPEALIGN(os_page_size,
1280 : : (char *) ent->location + ent->allocated_size);
1281 : 0 : total_len = (endptr - startptr);
1282 : :
1283 : 0 : shm_ent_page_count = total_len / os_page_size;
1284 : :
1285 : : /*
1286 : : * If we ever get 0xff (-1) back from kernel inquiry, then we probably
1287 : : * have a bug in mapping buffers to OS pages.
1288 : : */
1289 : 0 : memset(pages_status, 0xff, sizeof(int) * shm_ent_page_count);
1290 : :
1291 : : /*
1292 : : * Setup page_ptrs[] with pointers to all OS pages for this segment,
1293 : : * and get the NUMA status using pg_numa_query_pages.
1294 : : *
1295 : : * In order to get reliable results we also need to touch memory
1296 : : * pages, so that inquiry about NUMA memory node doesn't return -2
1297 : : * (ENOENT, which indicates unmapped/unallocated pages).
1298 : : */
93 peter@eisentraut.org 1299 [ # # ]:UNC 0 : for (uint64 i = 0; i < shm_ent_page_count; i++)
1300 : : {
553 tomas.vondra@postgre 1301 :UBC 0 : page_ptrs[i] = startptr + (i * os_page_size);
1302 : :
1303 : 0 : if (firstNumaTouch)
1304 : : pg_numa_touch_mem_if_required(page_ptrs[i]);
1305 : :
1306 [ # # ]: 0 : CHECK_FOR_INTERRUPTS();
1307 : : }
1308 : :
1309 [ # # ]: 0 : if (pg_numa_query_pages(0, shm_ent_page_count, page_ptrs, pages_status) == -1)
1310 [ # # ]: 0 : elog(ERROR, "failed NUMA pages inquiry status: %m");
1311 : :
1312 : : /* Count number of NUMA nodes used for this shared memory entry */
259 1313 : 0 : memset(nodes, 0, sizeof(Size) * (max_nodes + 2));
1314 : :
93 peter@eisentraut.org 1315 [ # # ]:UNC 0 : for (uint64 i = 0; i < shm_ent_page_count; i++)
1316 : : {
553 tomas.vondra@postgre 1317 :UBC 0 : int s = pages_status[i];
1318 : :
1319 : : /* Ensure we are adding only valid index to the array */
259 1320 [ # # # # ]: 0 : if (s >= 0 && s <= max_nodes)
1321 : : {
1322 : : /* valid NUMA node */
1323 : 0 : nodes[s]++;
1324 : 0 : continue;
1325 : : }
1326 [ # # ]: 0 : else if (s == -2)
1327 : : {
1328 : : /* -2 means ENOENT (e.g. page was moved to swap) */
1329 : 0 : nodes[max_nodes + 1]++;
1330 : 0 : continue;
1331 : : }
1332 : :
1333 [ # # ]: 0 : elog(ERROR, "invalid NUMA node id outside of allowed range "
1334 : : "[0, " UINT64_FORMAT "]: %d", max_nodes, s);
1335 : : }
1336 : :
1337 : : /* no NULLs for regular nodes */
1338 : 0 : memset(nulls, 0, sizeof(nulls));
1339 : :
1340 : : /*
1341 : : * Add one entry for each NUMA node, including those without allocated
1342 : : * memory for this segment.
1343 : : */
93 peter@eisentraut.org 1344 [ # # ]:UNC 0 : for (uint64 i = 0; i <= max_nodes; i++)
1345 : : {
553 tomas.vondra@postgre 1346 :UBC 0 : values[0] = CStringGetTextDatum(ent->key);
430 peter@eisentraut.org 1347 : 0 : values[1] = Int32GetDatum(i);
553 tomas.vondra@postgre 1348 : 0 : values[2] = Int64GetDatum(nodes[i] * os_page_size);
1349 : :
1350 : 0 : tuplestore_putvalues(rsinfo->setResult, rsinfo->setDesc,
1351 : : values, nulls);
1352 : : }
1353 : :
1354 : : /* The last entry is used for pages without a NUMA node. */
259 1355 : 0 : nulls[1] = true;
1356 : 0 : values[0] = CStringGetTextDatum(ent->key);
1357 : 0 : values[2] = Int64GetDatum(nodes[max_nodes + 1] * os_page_size);
1358 : :
1359 : 0 : tuplestore_putvalues(rsinfo->setResult, rsinfo->setDesc,
1360 : : values, nulls);
1361 : : }
1362 : :
553 1363 : 0 : LWLockRelease(ShmemIndexLock);
1364 : 0 : firstNumaTouch = false;
1365 : :
1366 : 0 : return (Datum) 0;
1367 : : }
1368 : :
1369 : : /*
1370 : : * Determine the memory page size used for the shared memory segment.
1371 : : *
1372 : : * If the shared segment was allocated using huge pages, returns the size of
1373 : : * a huge page. Otherwise returns the size of regular memory page.
1374 : : *
1375 : : * This should be used only after shared memory has been initialized.
1376 : : */
1377 : : Size
551 tomas.vondra@postgre 1378 :CBC 2 : pg_get_shmem_pagesize(void)
1379 : : {
1380 : : Size os_page_size;
1381 : : #ifdef WIN32
1382 : : SYSTEM_INFO sysinfo;
1383 : :
1384 : : GetSystemInfo(&sysinfo);
1385 : : os_page_size = sysinfo.dwPageSize;
1386 : : #else
1387 : 2 : os_page_size = sysconf(_SC_PAGESIZE);
1388 : : #endif
1389 : :
1390 [ - + ]: 2 : Assert(huge_pages_status != HUGE_PAGES_UNKNOWN);
1391 : :
1392 [ - + ]: 2 : if (huge_pages_status == HUGE_PAGES_ON)
551 tomas.vondra@postgre 1393 :UBC 0 : GetHugePageSize(&os_page_size, NULL);
1394 : :
551 tomas.vondra@postgre 1395 :CBC 2 : return os_page_size;
1396 : : }
1397 : :
1398 : : Datum
1399 : 5 : pg_numa_available(PG_FUNCTION_ARGS)
1400 : : {
1401 : 5 : PG_RETURN_BOOL(pg_numa_init() != -1);
1402 : : }
|
