Hi all, This is a follow-up of the following thread where I have touched the topic of corrupted 2PC files being completely ignored by recovery: https://www.postgresql.org/message-id/20180709012955.GD1467%40paquier.xyz I have posted a patch on this thread, but after more reviews I have noticed that it does not close all the holes.
What happens is that when a two-phase state file cannot be read by recovery when the file is on disk, which is what ReadTwoPhaseFile does, then its data is simply skipped. It is perfectly possible that an on-disk 2PC is missing at crash recovery if it has been already processed, so if trying to open the file and seeing an ENOENT, then the file can be properly skipped. However, if you look at this API, it skips *all* errors instead of the ones that matter. For example, if a file needs to be processed and is cannot be opened because of permission issues, then it is simply skipped. If its size, its CRC or its magic number is incorrect, the same thing happens. Skipping unconditionally files this way can result in data loss. I think that we really need to harden things, by making ReadTwoPhaseFile() fail hard is it finds something unexpected, which is in this case anything except trying to open a file which fails on ENOENT, and that this stuff should be back-patched. Thoughts? -- Michael
From c53e3f6b604ceffcf75484975331e1b141bb6139 Mon Sep 17 00:00:00 2001 From: Michael Paquier <mich...@paquier.xyz> Date: Mon, 9 Jul 2018 13:59:46 +0900 Subject: [PATCH] Fail hard when facing corrupted two-phase state files at recovery When a corrupted file is found by WAL replay, be it for crash recovery or archive recovery, then the file is simply skipped and a WARNING is logged to the user. Facing an on-disk WAL file which is corrupted is as likely to happen as its pair recorded in dedicated WAL records, but WAL records are already able to fail hard if there is a CRC mismatch at record level. The situation got better since 978b2f6 (9.6) which has added two-phase state information directly in WAL instead of using on-disk files, so the risk is limited to two-phase transactions which live across at least one checkpoint. Reported-by: Michael Paquier Author: Michael Paquier Discussion: https://postgr.es/m/XXX --- src/backend/access/transam/twophase.c | 131 +++++++++++--------------- 1 file changed, 53 insertions(+), 78 deletions(-) diff --git a/src/backend/access/transam/twophase.c b/src/backend/access/transam/twophase.c index e8d4e37fe3..3a048f9a4b 100644 --- a/src/backend/access/transam/twophase.c +++ b/src/backend/access/transam/twophase.c @@ -1206,10 +1206,11 @@ RegisterTwoPhaseRecord(TwoPhaseRmgrId rmid, uint16 info, * Read and validate the state file for xid. * * If it looks OK (has a valid magic number and CRC), return the palloc'd - * contents of the file. Otherwise return NULL. + * contents of the file. If missing_ok is true, do not issue an ERROR and + * return NULL. */ static char * -ReadTwoPhaseFile(TransactionId xid, bool give_warnings) +ReadTwoPhaseFile(TransactionId xid, bool missing_ok) { char path[MAXPGPATH]; char *buf; @@ -1225,12 +1226,12 @@ ReadTwoPhaseFile(TransactionId xid, bool give_warnings) fd = OpenTransientFile(path, O_RDONLY | PG_BINARY); if (fd < 0) { - if (give_warnings) - ereport(WARNING, - (errcode_for_file_access(), - errmsg("could not open two-phase state file \"%s\": %m", - path))); - return NULL; + if (missing_ok && errno == ENOENT) + return NULL; + ereport(ERROR, + (errcode_for_file_access(), + errmsg("could not open two-phase state file \"%s\": %m", + path))); } /* @@ -1240,36 +1241,26 @@ ReadTwoPhaseFile(TransactionId xid, bool give_warnings) * even on a valid file. */ if (fstat(fd, &stat)) - { - int save_errno = errno; - - CloseTransientFile(fd); - if (give_warnings) - { - errno = save_errno; - ereport(WARNING, - (errcode_for_file_access(), - errmsg("could not stat two-phase state file \"%s\": %m", - path))); - } - return NULL; - } + ereport(ERROR, + (errcode_for_file_access(), + errmsg("could not stat two-phase state file \"%s\": %m", + path))); if (stat.st_size < (MAXALIGN(sizeof(TwoPhaseFileHeader)) + MAXALIGN(sizeof(TwoPhaseRecordOnDisk)) + sizeof(pg_crc32c)) || stat.st_size > MaxAllocSize) - { - CloseTransientFile(fd); - return NULL; - } + ereport(ERROR, + (errcode(ERRCODE_DATA_CORRUPTED), + errmsg("incorrect size of two-phase state file \"%s\": %zu bytes", + path, stat.st_size))); crc_offset = stat.st_size - sizeof(pg_crc32c); if (crc_offset != MAXALIGN(crc_offset)) - { - CloseTransientFile(fd); - return NULL; - } + ereport(ERROR, + (errcode(ERRCODE_DATA_CORRUPTED), + errmsg("incorrect alignment of CRC offset for two-phase state file \"%s\"", + path))); /* * OK, slurp in the file. @@ -1278,32 +1269,26 @@ ReadTwoPhaseFile(TransactionId xid, bool give_warnings) pgstat_report_wait_start(WAIT_EVENT_TWOPHASE_FILE_READ); if (read(fd, buf, stat.st_size) != stat.st_size) - { - int save_errno = errno; - - pgstat_report_wait_end(); - CloseTransientFile(fd); - if (give_warnings) - { - errno = save_errno; - ereport(WARNING, - (errcode_for_file_access(), - errmsg("could not read two-phase state file \"%s\": %m", - path))); - } - pfree(buf); - return NULL; - } + ereport(ERROR, + (errcode_for_file_access(), + errmsg("could not read two-phase state file \"%s\": %m", + path))); pgstat_report_wait_end(); CloseTransientFile(fd); hdr = (TwoPhaseFileHeader *) buf; - if (hdr->magic != TWOPHASE_MAGIC || hdr->total_len != stat.st_size) - { - pfree(buf); - return NULL; - } + if (hdr->magic != TWOPHASE_MAGIC) + ereport(ERROR, + (errcode(ERRCODE_DATA_CORRUPTED), + errmsg("invalid magic number stored in two-phase state file \"%s\"", + path))); + + if (hdr->total_len != stat.st_size) + ereport(ERROR, + (errcode(ERRCODE_DATA_CORRUPTED), + errmsg("invalid size stored in two-phase state file \"%s\"", + path))); INIT_CRC32C(calc_crc); COMP_CRC32C(calc_crc, buf, crc_offset); @@ -1312,10 +1297,10 @@ ReadTwoPhaseFile(TransactionId xid, bool give_warnings) file_crc = *((pg_crc32c *) (buf + crc_offset)); if (!EQ_CRC32C(calc_crc, file_crc)) - { - pfree(buf); - return NULL; - } + ereport(ERROR, + (errcode(ERRCODE_DATA_CORRUPTED), + errmsg("calculated CRC checksum does not match value stored in two^phase state file \"%s\"", + path))); return buf; } @@ -1424,7 +1409,7 @@ StandbyTransactionIdIsPrepared(TransactionId xid) return false; /* nothing to do */ /* Read and validate file */ - buf = ReadTwoPhaseFile(xid, false); + buf = ReadTwoPhaseFile(xid, true); if (buf == NULL) return false; @@ -1472,7 +1457,7 @@ FinishPreparedTransaction(const char *gid, bool isCommit) * to disk if for some reason they have lived for a long time. */ if (gxact->ondisk) - buf = ReadTwoPhaseFile(xid, true); + buf = ReadTwoPhaseFile(xid, false); else XlogReadTwoPhaseData(gxact->prepare_start_lsn, &buf, NULL); @@ -1868,6 +1853,10 @@ restoreTwoPhaseData(void) * write a WAL entry, and so there might be no evidence in WAL of those * subxact XIDs. * + * On corrupted two-phase files, fail immediately. Keeping around broken + * entries and let replay continue causes harm on the system, and a new + * backup should be rolled in. + * * Our other responsibility is to determine and return the oldest valid XID * among the prepared xacts (if none, return ShmemVariableCache->nextXid). * This is needed to synchronize pg_subtrans startup properly. @@ -2158,15 +2147,7 @@ ProcessTwoPhaseBuffer(TransactionId xid, if (fromdisk) { /* Read and validate file */ - buf = ReadTwoPhaseFile(xid, true); - if (buf == NULL) - { - ereport(WARNING, - (errmsg("removing corrupt two-phase state file for transaction %u", - xid))); - RemoveTwoPhaseFile(xid, true); - return NULL; - } + buf = ReadTwoPhaseFile(xid, false); } else { @@ -2179,21 +2160,15 @@ ProcessTwoPhaseBuffer(TransactionId xid, if (!TransactionIdEquals(hdr->xid, xid)) { if (fromdisk) - { - ereport(WARNING, - (errmsg("removing corrupt two-phase state file for transaction %u", + ereport(ERROR, + (errcode(ERRCODE_DATA_CORRUPTED), + errmsg("corrupted two-phase state file for \"%u\"", xid))); - RemoveTwoPhaseFile(xid, true); - } else - { - ereport(WARNING, - (errmsg("removing corrupt two-phase state from memory for transaction %u", + ereport(ERROR, + (errcode(ERRCODE_DATA_CORRUPTED), + errmsg("corrupted two-phase state in memory for \"%u\"", xid))); - PrepareRedoRemove(xid, true); - } - pfree(buf); - return NULL; } /* -- 2.18.0
signature.asc
Description: PGP signature
