From dbdc717ac6743074c3a55fc5c380638c91d24afd Mon Sep 17 00:00:00 2001 From: Nathan Bossart Date: Wed, 26 Nov 2025 15:12:25 -0600 Subject: [PATCH] Teach DSM registry to retry entry initialization if needed. If DSM registry entry initialization fails, backends could try to use an uninitialized DSM segment, DSA, or dshash table (since the entry is still added to the registry). To fix, restructure the code so that the registry retries initialization as needed. This commit also modifies pg_get_dsm_registry_allocations() to leave out partially-initialized entries, as they shouldn't have any allocated memory. DSM registry entry initialization shouldn't fail often in practice, but retrying was deemed better than leaving entries in a permanently failed state (as was done by commit 1165a933aa, which has since been reverted). Suggested-by: Robert Haas Reviewed-by: Robert Haas Discussion: https://postgr.es/m/E1vJHUk-006I7r-37%40gemulon.postgresql.org Backpatch-through: 17 --- src/backend/storage/ipc/dsm_registry.c | 123 ++++++++++++++++--------- 1 file changed, 79 insertions(+), 44 deletions(-) diff --git a/src/backend/storage/ipc/dsm_registry.c b/src/backend/storage/ipc/dsm_registry.c index a926b9c3f32..ef6533b1100 100644 --- a/src/backend/storage/ipc/dsm_registry.c +++ b/src/backend/storage/ipc/dsm_registry.c @@ -155,9 +155,10 @@ init_dsm_registry(void) { /* Initialize dynamic shared hash table for registry. */ dsm_registry_dsa = dsa_create(LWTRANCHE_DSM_REGISTRY_DSA); + dsm_registry_table = dshash_create(dsm_registry_dsa, &dsh_params, NULL); + dsa_pin(dsm_registry_dsa); dsa_pin_mapping(dsm_registry_dsa); - dsm_registry_table = dshash_create(dsm_registry_dsa, &dsh_params, NULL); /* Store handles in shared memory for other backends to use. */ DSMRegistryCtx->dsah = dsa_get_handle(dsm_registry_dsa); @@ -188,6 +189,8 @@ GetNamedDSMSegment(const char *name, size_t size, DSMRegistryEntry *entry; MemoryContext oldcontext; void *ret; + NamedDSMState *state; + dsm_segment *seg; Assert(found); @@ -210,36 +213,36 @@ GetNamedDSMSegment(const char *name, size_t size, init_dsm_registry(); entry = dshash_find_or_insert(dsm_registry_table, name, found); + state = &entry->dsm; if (!(*found)) { - NamedDSMState *state = &entry->dsm; - dsm_segment *seg; - entry->type = DSMR_ENTRY_TYPE_DSM; - - /* Initialize the segment. */ - seg = dsm_create(size, 0); - - dsm_pin_segment(seg); - dsm_pin_mapping(seg); - state->handle = dsm_segment_handle(seg); + state->handle = DSM_HANDLE_INVALID; state->size = size; - ret = dsm_segment_address(seg); - - if (init_callback) - (*init_callback) (ret); } else if (entry->type != DSMR_ENTRY_TYPE_DSM) ereport(ERROR, (errmsg("requested DSM segment does not match type of existing entry"))); - else if (entry->dsm.size != size) + else if (state->size != size) ereport(ERROR, (errmsg("requested DSM segment size does not match size of existing segment"))); + + if (state->handle == DSM_HANDLE_INVALID) + { + *found = false; + + /* Initialize the segment. */ + seg = dsm_create(size, 0); + + if (init_callback) + (*init_callback) (dsm_segment_address(seg)); + + dsm_pin_segment(seg); + dsm_pin_mapping(seg); + state->handle = dsm_segment_handle(seg); + } else { - NamedDSMState *state = &entry->dsm; - dsm_segment *seg; - /* If the existing segment is not already attached, attach it now. */ seg = dsm_find_mapping(state->handle); if (seg == NULL) @@ -250,10 +253,9 @@ GetNamedDSMSegment(const char *name, size_t size, dsm_pin_mapping(seg); } - - ret = dsm_segment_address(seg); } + ret = dsm_segment_address(seg); dshash_release_lock(dsm_registry_table, entry); MemoryContextSwitchTo(oldcontext); @@ -274,6 +276,7 @@ GetNamedDSA(const char *name, bool *found) DSMRegistryEntry *entry; MemoryContext oldcontext; dsa_area *ret; + NamedDSAState *state; Assert(found); @@ -292,14 +295,28 @@ GetNamedDSA(const char *name, bool *found) init_dsm_registry(); entry = dshash_find_or_insert(dsm_registry_table, name, found); + state = &entry->dsa; if (!(*found)) { - NamedDSAState *state = &entry->dsa; - entry->type = DSMR_ENTRY_TYPE_DSA; + state->handle = DSA_HANDLE_INVALID; + state->tranche = -1; + } + else if (entry->type != DSMR_ENTRY_TYPE_DSA) + ereport(ERROR, + (errmsg("requested DSA does not match type of existing entry"))); + + if (state->tranche == -1) + { + *found = false; /* Initialize the LWLock tranche for the DSA. */ state->tranche = LWLockNewTrancheId(name); + } + + if (state->handle == DSA_HANDLE_INVALID) + { + *found = false; /* Initialize the DSA. */ ret = dsa_create(state->tranche); @@ -309,17 +326,11 @@ GetNamedDSA(const char *name, bool *found) /* Store handle for other backends to use. */ state->handle = dsa_get_handle(ret); } - else if (entry->type != DSMR_ENTRY_TYPE_DSA) + else if (dsa_is_attached(state->handle)) ereport(ERROR, - (errmsg("requested DSA does not match type of existing entry"))); + (errmsg("requested DSA already attached to current process"))); else { - NamedDSAState *state = &entry->dsa; - - if (dsa_is_attached(state->handle)) - ereport(ERROR, - (errmsg("requested DSA already attached to current process"))); - /* Attach to existing DSA. */ ret = dsa_attach(state->handle); dsa_pin_mapping(ret); @@ -346,6 +357,7 @@ GetNamedDSHash(const char *name, const dshash_parameters *params, bool *found) DSMRegistryEntry *entry; MemoryContext oldcontext; dshash_table *ret; + NamedDSHState *dsh_state; Assert(params); Assert(found); @@ -365,45 +377,57 @@ GetNamedDSHash(const char *name, const dshash_parameters *params, bool *found) init_dsm_registry(); entry = dshash_find_or_insert(dsm_registry_table, name, found); + dsh_state = &entry->dsh; if (!(*found)) { - NamedDSHState *dsh_state = &entry->dsh; - dshash_parameters params_copy; - dsa_area *dsa; - entry->type = DSMR_ENTRY_TYPE_DSH; + dsh_state->dsa_handle = DSA_HANDLE_INVALID; + dsh_state->dsh_handle = DSHASH_HANDLE_INVALID; + dsh_state->tranche = -1; + } + else if (entry->type != DSMR_ENTRY_TYPE_DSH) + ereport(ERROR, + (errmsg("requested DSHash does not match type of existing entry"))); + + if (dsh_state->tranche == -1) + { + *found = false; /* Initialize the LWLock tranche for the hash table. */ dsh_state->tranche = LWLockNewTrancheId(name); + } + + if (dsh_state->dsa_handle == DSA_HANDLE_INVALID) + { + dshash_parameters params_copy; + dsa_area *dsa; + + *found = false; /* Initialize the DSA for the hash table. */ dsa = dsa_create(dsh_state->tranche); - dsa_pin(dsa); - dsa_pin_mapping(dsa); /* Initialize the dshash table. */ memcpy(¶ms_copy, params, sizeof(dshash_parameters)); params_copy.tranche_id = dsh_state->tranche; ret = dshash_create(dsa, ¶ms_copy, NULL); + dsa_pin(dsa); + dsa_pin_mapping(dsa); + /* Store handles for other backends to use. */ dsh_state->dsa_handle = dsa_get_handle(dsa); dsh_state->dsh_handle = dshash_get_hash_table_handle(ret); } - else if (entry->type != DSMR_ENTRY_TYPE_DSH) + else if (dsa_is_attached(dsh_state->dsa_handle)) ereport(ERROR, - (errmsg("requested DSHash does not match type of existing entry"))); + (errmsg("requested DSHash already attached to current process"))); else { - NamedDSHState *dsh_state = &entry->dsh; dsa_area *dsa; /* XXX: Should we verify params matches what table was created with? */ - if (dsa_is_attached(dsh_state->dsa_handle)) - ereport(ERROR, - (errmsg("requested DSHash already attached to current process"))); - /* Attach to existing DSA for the hash table. */ dsa = dsa_attach(dsh_state->dsa_handle); dsa_pin_mapping(dsa); @@ -439,6 +463,17 @@ pg_get_dsm_registry_allocations(PG_FUNCTION_ARGS) Datum vals[3]; bool nulls[3] = {0}; + /* Do not show partially-initialized entries. */ + if (entry->type == DSMR_ENTRY_TYPE_DSM && + entry->dsm.handle == DSM_HANDLE_INVALID) + continue; + if (entry->type == DSMR_ENTRY_TYPE_DSA && + entry->dsa.handle == DSA_HANDLE_INVALID) + continue; + if (entry->type == DSMR_ENTRY_TYPE_DSH && + entry->dsh.dsa_handle == DSA_HANDLE_INVALID) + continue; + vals[0] = CStringGetTextDatum(entry->name); vals[1] = CStringGetTextDatum(DSMREntryTypeNames[entry->type]);