X-Git-Url: https://git.saurik.com/redis.git/blobdiff_plain/ac1f6fd07a15ac7a14f27b8a5f4b5fb3dc6cf8ac..749aac72ad980bf08b04655634590b80e2fd014c:/src/networking.c?ds=inline diff --git a/src/networking.c b/src/networking.c index 3979ab62..3bc084f7 100644 --- a/src/networking.c +++ b/src/networking.c @@ -1,6 +1,16 @@ #include "redis.h" #include +static void setProtocolError(redisClient *c, int pos); + +/* To evaluate the output buffer size of a client we need to get size of + * allocated objects, however we can't used zmalloc_size() directly on sds + * strings because of the trick they use to work (the header is before the + * returned pointer), so we use this helper function. */ +size_t zmalloc_size_sds(sds s) { + return zmalloc_size(s-sizeof(struct sdshdr)); +} + void *dupClientReplyValue(void *o) { incrRefCount((robj*)o); return o; @@ -12,7 +22,6 @@ int listMatchObjects(void *a, void *b) { redisClient *createClient(int fd) { redisClient *c = zmalloc(sizeof(redisClient)); - c->bufpos = 0; /* passing -1 as fd it is possible to create a non connected client. * This is useful since all the Redis commands needs to be executed @@ -32,19 +41,24 @@ redisClient *createClient(int fd) { selectDb(c,0); c->fd = fd; + c->bufpos = 0; c->querybuf = sdsempty(); + c->querybuf_peak = 0; c->reqtype = 0; c->argc = 0; c->argv = NULL; - c->cmd = NULL; + c->cmd = c->lastcmd = NULL; c->multibulklen = 0; c->bulklen = -1; c->sentlen = 0; c->flags = 0; - c->lastinteraction = time(NULL); + c->ctime = c->lastinteraction = server.unixtime; c->authenticated = 0; c->replstate = REDIS_REPL_NONE; + c->slave_listening_port = 0; c->reply = listCreate(); + c->reply_bytes = 0; + c->obuf_soft_limit_reached_time = 0; listSetFreeMethod(c->reply,decrRefCount); listSetDupMethod(c->reply,dupClientReplyValue); c->bpop.keys = NULL; @@ -63,11 +77,23 @@ redisClient *createClient(int fd) { return c; } -/* Set the event loop to listen for write events on the client's socket. - * Typically gets called every time a reply is built. */ -int _installWriteEvent(redisClient *c) { +/* This function is called every time we are going to transmit new data + * to the client. The behavior is the following: + * + * If the client should receive new data (normal clients will) the function + * returns REDIS_OK, and make sure to install the write handler in our event + * loop so that when the socket is writable new data gets written. + * + * If the client should not receive new data, because it is a fake client + * or a slave, or because the setup of the write handler failed, the function + * returns REDIS_ERR. + * + * Typically gets called every time a reply is built, before adding more + * data to the clients output buffers. If the function returns REDIS_ERR no + * data should be appended to the output buffers. */ +int prepareClientToWrite(redisClient *c) { if (c->flags & REDIS_LUA_CLIENT) return REDIS_OK; - if (c->fd <= 0) return REDIS_ERR; + if (c->fd <= 0) return REDIS_ERR; /* Fake client */ if (c->bufpos == 0 && listLength(c->reply) == 0 && (c->replstate == REDIS_REPL_NONE || c->replstate == REDIS_REPL_ONLINE) && @@ -121,6 +147,7 @@ void _addReplyObjectToList(redisClient *c, robj *o) { if (listLength(c->reply) == 0) { incrRefCount(o); listAddNodeTail(c->reply,o); + c->reply_bytes += zmalloc_size_sds(o->ptr); } else { tail = listNodeValue(listLast(c->reply)); @@ -128,13 +155,17 @@ void _addReplyObjectToList(redisClient *c, robj *o) { if (tail->ptr != NULL && sdslen(tail->ptr)+sdslen(o->ptr) <= REDIS_REPLY_CHUNK_BYTES) { + c->reply_bytes -= zmalloc_size_sds(tail->ptr); tail = dupLastObjectIfNeeded(c->reply); tail->ptr = sdscatlen(tail->ptr,o->ptr,sdslen(o->ptr)); + c->reply_bytes += zmalloc_size_sds(tail->ptr); } else { incrRefCount(o); listAddNodeTail(c->reply,o); + c->reply_bytes += zmalloc_size_sds(o->ptr); } } + asyncCloseClientOnOutputBufferLimitReached(c); } /* This method takes responsibility over the sds. When it is no longer @@ -149,6 +180,7 @@ void _addReplySdsToList(redisClient *c, sds s) { if (listLength(c->reply) == 0) { listAddNodeTail(c->reply,createObject(REDIS_STRING,s)); + c->reply_bytes += zmalloc_size_sds(s); } else { tail = listNodeValue(listLast(c->reply)); @@ -156,13 +188,17 @@ void _addReplySdsToList(redisClient *c, sds s) { if (tail->ptr != NULL && sdslen(tail->ptr)+sdslen(s) <= REDIS_REPLY_CHUNK_BYTES) { + c->reply_bytes -= zmalloc_size_sds(tail->ptr); tail = dupLastObjectIfNeeded(c->reply); tail->ptr = sdscatlen(tail->ptr,s,sdslen(s)); + c->reply_bytes += zmalloc_size_sds(tail->ptr); sdsfree(s); } else { listAddNodeTail(c->reply,createObject(REDIS_STRING,s)); + c->reply_bytes += zmalloc_size_sds(s); } } + asyncCloseClientOnOutputBufferLimitReached(c); } void _addReplyStringToList(redisClient *c, char *s, size_t len) { @@ -171,7 +207,10 @@ void _addReplyStringToList(redisClient *c, char *s, size_t len) { if (c->flags & REDIS_CLOSE_AFTER_REPLY) return; if (listLength(c->reply) == 0) { - listAddNodeTail(c->reply,createStringObject(s,len)); + robj *o = createStringObject(s,len); + + listAddNodeTail(c->reply,o); + c->reply_bytes += zmalloc_size_sds(o->ptr); } else { tail = listNodeValue(listLast(c->reply)); @@ -179,12 +218,18 @@ void _addReplyStringToList(redisClient *c, char *s, size_t len) { if (tail->ptr != NULL && sdslen(tail->ptr)+len <= REDIS_REPLY_CHUNK_BYTES) { + c->reply_bytes -= zmalloc_size_sds(tail->ptr); tail = dupLastObjectIfNeeded(c->reply); tail->ptr = sdscatlen(tail->ptr,s,len); + c->reply_bytes += zmalloc_size_sds(tail->ptr); } else { - listAddNodeTail(c->reply,createStringObject(s,len)); + robj *o = createStringObject(s,len); + + listAddNodeTail(c->reply,o); + c->reply_bytes += zmalloc_size_sds(o->ptr); } } + asyncCloseClientOnOutputBufferLimitReached(c); } /* ----------------------------------------------------------------------------- @@ -193,7 +238,7 @@ void _addReplyStringToList(redisClient *c, char *s, size_t len) { * -------------------------------------------------------------------------- */ void addReply(redisClient *c, robj *obj) { - if (_installWriteEvent(c) != REDIS_OK) return; + if (prepareClientToWrite(c) != REDIS_OK) return; /* This is an important place where we can avoid copy-on-write * when there is a saving child running, avoiding touching the @@ -205,19 +250,31 @@ void addReply(redisClient *c, robj *obj) { if (obj->encoding == REDIS_ENCODING_RAW) { if (_addReplyToBuffer(c,obj->ptr,sdslen(obj->ptr)) != REDIS_OK) _addReplyObjectToList(c,obj); - } else { - /* FIXME: convert the long into string and use _addReplyToBuffer() - * instead of calling getDecodedObject. As this place in the - * code is too performance critical. */ + } else if (obj->encoding == REDIS_ENCODING_INT) { + /* Optimization: if there is room in the static buffer for 32 bytes + * (more than the max chars a 64 bit integer can take as string) we + * avoid decoding the object and go for the lower level approach. */ + if (listLength(c->reply) == 0 && (sizeof(c->buf) - c->bufpos) >= 32) { + char buf[32]; + int len; + + len = ll2string(buf,sizeof(buf),(long)obj->ptr); + if (_addReplyToBuffer(c,buf,len) == REDIS_OK) + return; + /* else... continue with the normal code path, but should never + * happen actually since we verified there is room. */ + } obj = getDecodedObject(obj); if (_addReplyToBuffer(c,obj->ptr,sdslen(obj->ptr)) != REDIS_OK) _addReplyObjectToList(c,obj); decrRefCount(obj); + } else { + redisPanic("Wrong obj->encoding in addReply()"); } } void addReplySds(redisClient *c, sds s) { - if (_installWriteEvent(c) != REDIS_OK) { + if (prepareClientToWrite(c) != REDIS_OK) { /* The caller expects the sds to be free'd. */ sdsfree(s); return; @@ -231,19 +288,19 @@ void addReplySds(redisClient *c, sds s) { } void addReplyString(redisClient *c, char *s, size_t len) { - if (_installWriteEvent(c) != REDIS_OK) return; + if (prepareClientToWrite(c) != REDIS_OK) return; if (_addReplyToBuffer(c,s,len) != REDIS_OK) _addReplyStringToList(c,s,len); } -void _addReplyError(redisClient *c, char *s, size_t len) { +void addReplyErrorLength(redisClient *c, char *s, size_t len) { addReplyString(c,"-ERR ",5); addReplyString(c,s,len); addReplyString(c,"\r\n",2); } void addReplyError(redisClient *c, char *err) { - _addReplyError(c,err,strlen(err)); + addReplyErrorLength(c,err,strlen(err)); } void addReplyErrorFormat(redisClient *c, const char *fmt, ...) { @@ -258,18 +315,18 @@ void addReplyErrorFormat(redisClient *c, const char *fmt, ...) { for (j = 0; j < l; j++) { if (s[j] == '\r' || s[j] == '\n') s[j] = ' '; } - _addReplyError(c,s,sdslen(s)); + addReplyErrorLength(c,s,sdslen(s)); sdsfree(s); } -void _addReplyStatus(redisClient *c, char *s, size_t len) { +void addReplyStatusLength(redisClient *c, char *s, size_t len) { addReplyString(c,"+",1); addReplyString(c,s,len); addReplyString(c,"\r\n",2); } void addReplyStatus(redisClient *c, char *status) { - _addReplyStatus(c,status,strlen(status)); + addReplyStatusLength(c,status,strlen(status)); } void addReplyStatusFormat(redisClient *c, const char *fmt, ...) { @@ -277,7 +334,7 @@ void addReplyStatusFormat(redisClient *c, const char *fmt, ...) { va_start(ap,fmt); sds s = sdscatvprintf(sdsempty(),fmt,ap); va_end(ap); - _addReplyStatus(c,s,sdslen(s)); + addReplyStatusLength(c,s,sdslen(s)); sdsfree(s); } @@ -287,7 +344,7 @@ void *addDeferredMultiBulkLength(redisClient *c) { /* Note that we install the write event here even if the object is not * ready to be sent, since we are sure that before returning to the * event loop setDeferredMultiBulkLength() will be called. */ - if (_installWriteEvent(c) != REDIS_OK) return NULL; + if (prepareClientToWrite(c) != REDIS_OK) return NULL; listAddNodeTail(c->reply,createObject(REDIS_STRING,NULL)); return listLast(c->reply); } @@ -302,15 +359,20 @@ void setDeferredMultiBulkLength(redisClient *c, void *node, long length) { len = listNodeValue(ln); len->ptr = sdscatprintf(sdsempty(),"*%ld\r\n",length); + c->reply_bytes += zmalloc_size_sds(len->ptr); if (ln->next != NULL) { next = listNodeValue(ln->next); /* Only glue when the next node is non-NULL (an sds in this case) */ if (next->ptr != NULL) { + c->reply_bytes -= zmalloc_size_sds(len->ptr); + c->reply_bytes -= zmalloc_size_sds(next->ptr); len->ptr = sdscatlen(len->ptr,next->ptr,sdslen(next->ptr)); + c->reply_bytes += zmalloc_size_sds(len->ptr); listDelNode(c->reply,ln->next); } } + asyncCloseClientOnOutputBufferLimitReached(c); } /* Add a duble as a bulk reply */ @@ -324,9 +386,21 @@ void addReplyDouble(redisClient *c, double d) { /* Add a long long as integer reply or bulk len / multi bulk count. * Basically this is used to output . */ -void _addReplyLongLong(redisClient *c, long long ll, char prefix) { +void addReplyLongLongWithPrefix(redisClient *c, long long ll, char prefix) { char buf[128]; int len; + + /* Things like $3\r\n or *2\r\n are emitted very often by the protocol + * so we have a few shared objects to use if the integer is small + * like it is most of the times. */ + if (prefix == '*' && ll < REDIS_SHARED_BULKHDR_LEN) { + addReply(c,shared.mbulkhdr[ll]); + return; + } else if (prefix == '$' && ll < REDIS_SHARED_BULKHDR_LEN) { + addReply(c,shared.bulkhdr[ll]); + return; + } + buf[0] = prefix; len = ll2string(buf+1,sizeof(buf)-1,ll); buf[len+1] = '\r'; @@ -340,11 +414,11 @@ void addReplyLongLong(redisClient *c, long long ll) { else if (ll == 1) addReply(c,shared.cone); else - _addReplyLongLong(c,ll,':'); + addReplyLongLongWithPrefix(c,ll,':'); } void addReplyMultiBulkLen(redisClient *c, long length) { - _addReplyLongLong(c,length,'*'); + addReplyLongLongWithPrefix(c,length,'*'); } /* Create the length prefix of a bulk reply, example: $2234 */ @@ -366,7 +440,7 @@ void addReplyBulkLen(redisClient *c, robj *obj) { len++; } } - _addReplyLongLong(c,len,'$'); + addReplyLongLongWithPrefix(c,len,'$'); } /* Add a Redis Object as a bulk reply */ @@ -378,7 +452,7 @@ void addReplyBulk(redisClient *c, robj *obj) { /* Add a C buffer as bulk reply */ void addReplyBulkCBuffer(redisClient *c, void *p, size_t len) { - _addReplyLongLong(c,len,'$'); + addReplyLongLongWithPrefix(c,len,'$'); addReplyString(c,p,len); addReply(c,shared.crlf); } @@ -401,6 +475,17 @@ void addReplyBulkLongLong(redisClient *c, long long ll) { addReplyBulkCBuffer(c,buf,len); } +/* Copy 'src' client output buffers into 'dst' client output buffers. + * The function takes care of freeing the old output buffers of the + * destination client. */ +void copyClientOutputBuffer(redisClient *dst, redisClient *src) { + listRelease(dst->reply); + dst->reply = listDup(src->reply); + memcpy(dst->buf,src->buf,src->bufpos); + dst->bufpos = src->bufpos; + dst->reply_bytes = src->reply_bytes; +} + static void acceptCommonHandler(int fd) { redisClient *c; if ((c = createClient(fd)) == NULL) { @@ -412,13 +497,14 @@ static void acceptCommonHandler(int fd) { * connection. Note that we create the client instead to check before * for this condition, since now the socket is already set in nonblocking * mode and we can send an error for free using the Kernel I/O */ - if (server.maxclients && listLength(server.clients) > server.maxclients) { + if (listLength(server.clients) > server.maxclients) { char *err = "-ERR max number of clients reached\r\n"; /* That's a best effort error message, don't check write errors */ if (write(c->fd,err,strlen(err)) == -1) { /* Nothing to do, Just to avoid the warning... */ } + server.stat_rejected_conn++; freeClient(c); return; } @@ -465,9 +551,22 @@ static void freeClientArgv(redisClient *c) { c->cmd = NULL; } +/* Close all the slaves connections. This is useful in chained replication + * when we resync with our own master and want to force all our slaves to + * resync with us as well. */ +void disconnectSlaves(void) { + while (listLength(server.slaves)) { + listNode *ln = listFirst(server.slaves); + freeClient((redisClient*)ln->value); + } +} + void freeClient(redisClient *c) { listNode *ln; + /* If this is marked as current client unset it */ + if (server.current_client == c) server.current_client = NULL; + /* Note that if the client we are freeing is blocked into a blocking * call, we have to set querybuf to NULL *before* to call * unblockClientWaitingData() to avoid processInputBuffer() will get @@ -518,33 +617,55 @@ void freeClient(redisClient *c) { /* Case 2: we lost the connection with the master. */ if (c->flags & REDIS_MASTER) { server.master = NULL; - server.replstate = REDIS_REPL_CONNECT; - server.repl_down_since = time(NULL); - /* Since we lost the connection with the master, we should also - * close the connection with all our slaves if we have any, so - * when we'll resync with the master the other slaves will sync again - * with us as well. Note that also when the slave is not connected - * to the master it will keep refusing connections by other slaves. + server.repl_state = REDIS_REPL_CONNECT; + server.repl_down_since = server.unixtime; + /* We lost connection with our master, force our slaves to resync + * with us as well to load the new data set. * - * We do this only if server.masterhost != NULL. If it is NULL this - * means the user called SLAVEOF NO ONE and we are freeing our - * link with the master, so no need to close link with slaves. */ - if (server.masterhost != NULL) { - while (listLength(server.slaves)) { - ln = listFirst(server.slaves); - freeClient((redisClient*)ln->value); - } - } + * If server.masterhost is NULL the user called SLAVEOF NO ONE so + * slave resync is not needed. */ + if (server.masterhost != NULL) disconnectSlaves(); } + + /* If this client was scheduled for async freeing we need to remove it + * from the queue. */ + if (c->flags & REDIS_CLOSE_ASAP) { + ln = listSearchKey(server.clients_to_close,c); + redisAssert(ln != NULL); + listDelNode(server.clients_to_close,ln); + } + /* Release memory */ zfree(c->argv); freeClientMultiState(c); zfree(c); } +/* Schedule a client to free it at a safe time in the serverCron() function. + * This function is useful when we need to terminate a client but we are in + * a context where calling freeClient() is not possible, because the client + * should be valid for the continuation of the flow of the program. */ +void freeClientAsync(redisClient *c) { + if (c->flags & REDIS_CLOSE_ASAP) return; + c->flags |= REDIS_CLOSE_ASAP; + listAddNodeTail(server.clients_to_close,c); +} + +void freeClientsInAsyncFreeQueue(void) { + while (listLength(server.clients_to_close)) { + listNode *ln = listFirst(server.clients_to_close); + redisClient *c = listNodeValue(ln); + + c->flags &= ~REDIS_CLOSE_ASAP; + freeClient(c); + listDelNode(server.clients_to_close,ln); + } +} + void sendReplyToClient(aeEventLoop *el, int fd, void *privdata, int mask) { redisClient *c = privdata; int nwritten = 0, totwritten = 0, objlen; + size_t objmem; robj *o; REDIS_NOTUSED(el); REDIS_NOTUSED(mask); @@ -570,6 +691,7 @@ void sendReplyToClient(aeEventLoop *el, int fd, void *privdata, int mask) { } else { o = listNodeValue(listFirst(c->reply)); objlen = sdslen(o->ptr); + objmem = zmalloc_size_sds(o->ptr); if (objlen == 0) { listDelNode(c->reply,listFirst(c->reply)); @@ -590,14 +712,20 @@ void sendReplyToClient(aeEventLoop *el, int fd, void *privdata, int mask) { if (c->sentlen == objlen) { listDelNode(c->reply,listFirst(c->reply)); c->sentlen = 0; + c->reply_bytes -= objmem; } } - /* Note that we avoid to send more thank REDIS_MAX_WRITE_PER_EVENT + /* Note that we avoid to send more than REDIS_MAX_WRITE_PER_EVENT * bytes, in a single threaded server it's a good idea to serve * other clients as well, even if a very large request comes from * super fast link that is always able to accept data (in real world - * scenario think about 'KEYS *' against the loopback interfae) */ - if (totwritten > REDIS_MAX_WRITE_PER_EVENT) break; + * scenario think about 'KEYS *' against the loopback interface). + * + * However if we are over the maxmemory limit we ignore that and + * just deliver as much data as it is possible to deliver. */ + if (totwritten > REDIS_MAX_WRITE_PER_EVENT && + (server.maxmemory == 0 || + zmalloc_used_memory() < server.maxmemory)) break; } if (nwritten == -1) { if (errno == EAGAIN) { @@ -609,7 +737,7 @@ void sendReplyToClient(aeEventLoop *el, int fd, void *privdata, int mask) { return; } } - if (totwritten > 0) c->lastinteraction = time(NULL); + if (totwritten > 0) c->lastinteraction = server.unixtime; if (c->bufpos == 0 && listLength(c->reply) == 0) { c->sentlen = 0; aeDeleteFileEvent(server.el,c->fd,AE_WRITABLE); @@ -625,34 +753,8 @@ void resetClient(redisClient *c) { c->reqtype = 0; c->multibulklen = 0; c->bulklen = -1; -} - -void closeTimedoutClients(void) { - redisClient *c; - listNode *ln; - time_t now = time(NULL); - listIter li; - - listRewind(server.clients,&li); - while ((ln = listNext(&li)) != NULL) { - c = listNodeValue(ln); - if (server.maxidletime && - !(c->flags & REDIS_SLAVE) && /* no timeout for slaves */ - !(c->flags & REDIS_MASTER) && /* no timeout for masters */ - !(c->flags & REDIS_BLOCKED) && /* no timeout for BLPOP */ - dictSize(c->pubsub_channels) == 0 && /* no timeout for pubsub */ - listLength(c->pubsub_patterns) == 0 && - (now - c->lastinteraction > server.maxidletime)) - { - redisLog(REDIS_VERBOSE,"Closing idle client"); - freeClient(c); - } else if (c->flags & REDIS_BLOCKED) { - if (c->bpop.timeout != 0 && c->bpop.timeout < now) { - addReply(c,shared.nullmultibulk); - unblockClientWaitingData(c); - } - } - } + /* We clear the ASKING flag as well if we are not inside a MULTI. */ + if (!(c->flags & REDIS_MULTI)) c->flags &= (~REDIS_ASKING); } int processInlineBuffer(redisClient *c) { @@ -662,8 +764,13 @@ int processInlineBuffer(redisClient *c) { size_t querylen; /* Nothing to do without a \r\n */ - if (newline == NULL) + if (newline == NULL) { + if (sdslen(c->querybuf) > REDIS_INLINE_MAX_SIZE) { + addReplyError(c,"Protocol error: too big inline request"); + setProtocolError(c,0); + } return REDIS_ERR; + } /* Split the input buffer up to the \r\n */ querylen = newline-(c->querybuf); @@ -692,6 +799,12 @@ int processInlineBuffer(redisClient *c) { /* Helper function. Trims query buffer to make the function that processes * multi bulk requests idempotent. */ static void setProtocolError(redisClient *c, int pos) { + if (server.verbosity >= REDIS_VERBOSE) { + sds client = getClientInfoString(c); + redisLog(REDIS_VERBOSE, + "Protocol error from client: %s", client); + sdsfree(client); + } c->flags |= REDIS_CLOSE_AFTER_REPLY; c->querybuf = sdsrange(c->querybuf,pos,-1); } @@ -703,12 +816,17 @@ int processMultibulkBuffer(redisClient *c) { if (c->multibulklen == 0) { /* The client should have been reset */ - redisAssert(c->argc == 0); + redisAssertWithInfo(c,NULL,c->argc == 0); /* Multi bulk length cannot be read without a \r\n */ newline = strchr(c->querybuf,'\r'); - if (newline == NULL) + if (newline == NULL) { + if (sdslen(c->querybuf) > REDIS_INLINE_MAX_SIZE) { + addReplyError(c,"Protocol error: too big mbulk count string"); + setProtocolError(c,0); + } return REDIS_ERR; + } /* Buffer should also contain \n */ if (newline-(c->querybuf) > ((signed)sdslen(c->querybuf)-2)) @@ -716,7 +834,7 @@ int processMultibulkBuffer(redisClient *c) { /* We know for sure there is a whole line since newline != NULL, * so go ahead and find out the multi bulk length. */ - redisAssert(c->querybuf[0] == '*'); + redisAssertWithInfo(c,NULL,c->querybuf[0] == '*'); ok = string2ll(c->querybuf+1,newline-(c->querybuf+1),&ll); if (!ok || ll > 1024*1024) { addReplyError(c,"Protocol error: invalid multibulk length"); @@ -737,13 +855,18 @@ int processMultibulkBuffer(redisClient *c) { c->argv = zmalloc(sizeof(robj*)*c->multibulklen); } - redisAssert(c->multibulklen > 0); + redisAssertWithInfo(c,NULL,c->multibulklen > 0); while(c->multibulklen) { /* Read bulk length if unknown */ if (c->bulklen == -1) { newline = strchr(c->querybuf+pos,'\r'); - if (newline == NULL) + if (newline == NULL) { + if (sdslen(c->querybuf) > REDIS_INLINE_MAX_SIZE) { + addReplyError(c,"Protocol error: too big bulk count string"); + setProtocolError(c,0); + } break; + } /* Buffer should also contain \n */ if (newline-(c->querybuf) > ((signed)sdslen(c->querybuf)-2)) @@ -765,6 +888,17 @@ int processMultibulkBuffer(redisClient *c) { } pos += newline-(c->querybuf+pos)+2; + if (ll >= REDIS_MBULK_BIG_ARG) { + /* If we are going to read a large object from network + * try to make it likely that it will start at c->querybuf + * boundary so that we can optimized object creation + * avoiding a large copy of data. */ + c->querybuf = sdsrange(c->querybuf,pos,-1); + pos = 0; + /* Hint the sds library about the amount of bytes this string is + * going to contain. */ + c->querybuf = sdsMakeRoomFor(c->querybuf,ll+2); + } c->bulklen = ll; } @@ -773,20 +907,37 @@ int processMultibulkBuffer(redisClient *c) { /* Not enough data (+2 == trailing \r\n) */ break; } else { - c->argv[c->argc++] = createStringObject(c->querybuf+pos,c->bulklen); - pos += c->bulklen+2; + /* Optimization: if the buffer contanins JUST our bulk element + * instead of creating a new object by *copying* the sds we + * just use the current sds string. */ + if (pos == 0 && + c->bulklen >= REDIS_MBULK_BIG_ARG && + (signed) sdslen(c->querybuf) == c->bulklen+2) + { + c->argv[c->argc++] = createObject(REDIS_STRING,c->querybuf); + sdsIncrLen(c->querybuf,-2); /* remove CRLF */ + c->querybuf = sdsempty(); + /* Assume that if we saw a fat argument we'll see another one + * likely... */ + c->querybuf = sdsMakeRoomFor(c->querybuf,c->bulklen+2); + pos = 0; + } else { + c->argv[c->argc++] = + createStringObject(c->querybuf+pos,c->bulklen); + pos += c->bulklen+2; + } c->bulklen = -1; c->multibulklen--; } } /* Trim to pos */ - c->querybuf = sdsrange(c->querybuf,pos,-1); + if (pos) c->querybuf = sdsrange(c->querybuf,pos,-1); /* We're done when c->multibulk == 0 */ - if (c->multibulklen == 0) { - return REDIS_OK; - } + if (c->multibulklen == 0) return REDIS_OK; + + /* Still not read to process the command */ return REDIS_ERR; } @@ -831,12 +982,31 @@ void processInputBuffer(redisClient *c) { void readQueryFromClient(aeEventLoop *el, int fd, void *privdata, int mask) { redisClient *c = (redisClient*) privdata; - char buf[REDIS_IOBUF_LEN]; - int nread; + int nread, readlen; + size_t qblen; REDIS_NOTUSED(el); REDIS_NOTUSED(mask); - nread = read(fd, buf, REDIS_IOBUF_LEN); + server.current_client = c; + readlen = REDIS_IOBUF_LEN; + /* If this is a multi bulk request, and we are processing a bulk reply + * that is large enough, try to maximize the probabilty that the query + * buffer contains excatly the SDS string representing the object, even + * at the risk of requring more read(2) calls. This way the function + * processMultiBulkBuffer() can avoid copying buffers to create the + * Redis Object representing the argument. */ + if (c->reqtype == REDIS_REQ_MULTIBULK && c->multibulklen && c->bulklen != -1 + && c->bulklen >= REDIS_MBULK_BIG_ARG) + { + int remaining = (unsigned)(c->bulklen+2)-sdslen(c->querybuf); + + if (remaining < readlen) readlen = remaining; + } + + qblen = sdslen(c->querybuf); + if (c->querybuf_peak < qblen) c->querybuf_peak = qblen; + c->querybuf = sdsMakeRoomFor(c->querybuf, readlen); + nread = read(fd, c->querybuf+qblen, readlen); if (nread == -1) { if (errno == EAGAIN) { nread = 0; @@ -851,12 +1021,24 @@ void readQueryFromClient(aeEventLoop *el, int fd, void *privdata, int mask) { return; } if (nread) { - c->querybuf = sdscatlen(c->querybuf,buf,nread); - c->lastinteraction = time(NULL); + sdsIncrLen(c->querybuf,nread); + c->lastinteraction = server.unixtime; } else { + server.current_client = NULL; + return; + } + if (sdslen(c->querybuf) > server.client_max_querybuf_len) { + sds ci = getClientInfoString(c), bytes = sdsempty(); + + bytes = sdscatrepr(bytes,c->querybuf,64); + redisLog(REDIS_WARNING,"Closing client that reached max query buffer length: %s (qbuf initial bytes: %s)", ci, bytes); + sdsfree(ci); + sdsfree(bytes); + freeClient(c); return; } processInputBuffer(c); + server.current_client = NULL; } void getClientsMaxBuffers(unsigned long *longest_output_list, @@ -877,46 +1059,80 @@ void getClientsMaxBuffers(unsigned long *longest_output_list, *biggest_input_buffer = bib; } -void clientCommand(redisClient *c) { +/* Turn a Redis client into an sds string representing its state. */ +sds getClientInfoString(redisClient *client) { + char ip[32], flags[16], events[3], *p; + int port; + int emask; + + anetPeerToString(client->fd,ip,&port); + p = flags; + if (client->flags & REDIS_SLAVE) { + if (client->flags & REDIS_MONITOR) + *p++ = 'O'; + else + *p++ = 'S'; + } + if (client->flags & REDIS_MASTER) *p++ = 'M'; + if (client->flags & REDIS_MULTI) *p++ = 'x'; + if (client->flags & REDIS_BLOCKED) *p++ = 'b'; + if (client->flags & REDIS_DIRTY_CAS) *p++ = 'd'; + if (client->flags & REDIS_CLOSE_AFTER_REPLY) *p++ = 'c'; + if (client->flags & REDIS_UNBLOCKED) *p++ = 'u'; + if (client->flags & REDIS_CLOSE_ASAP) *p++ = 'A'; + if (p == flags) *p++ = 'N'; + *p++ = '\0'; + + emask = client->fd == -1 ? 0 : aeGetFileEvents(server.el,client->fd); + p = events; + if (emask & AE_READABLE) *p++ = 'r'; + if (emask & AE_WRITABLE) *p++ = 'w'; + *p = '\0'; + return sdscatprintf(sdsempty(), + "addr=%s:%d fd=%d age=%ld idle=%ld flags=%s db=%d sub=%d psub=%d multi=%d qbuf=%lu qbuf-free=%lu obl=%lu oll=%lu omem=%lu events=%s cmd=%s", + ip,port,client->fd, + (long)(server.unixtime - client->ctime), + (long)(server.unixtime - client->lastinteraction), + flags, + client->db->id, + (int) dictSize(client->pubsub_channels), + (int) listLength(client->pubsub_patterns), + (client->flags & REDIS_MULTI) ? client->mstate.count : -1, + (unsigned long) sdslen(client->querybuf), + (unsigned long) sdsavail(client->querybuf), + (unsigned long) client->bufpos, + (unsigned long) listLength(client->reply), + getClientOutputBufferMemoryUsage(client), + events, + client->lastcmd ? client->lastcmd->name : "NULL"); +} + +sds getAllClientsInfoString(void) { listNode *ln; listIter li; redisClient *client; + sds o = sdsempty(); - if (!strcasecmp(c->argv[1]->ptr,"list") && c->argc == 2) { - sds o = sdsempty(); - time_t now = time(NULL); + listRewind(server.clients,&li); + while ((ln = listNext(&li)) != NULL) { + sds cs; - listRewind(server.clients,&li); - while ((ln = listNext(&li)) != NULL) { - char ip[32], flags[16], *p; - int port; + client = listNodeValue(ln); + cs = getClientInfoString(client); + o = sdscatsds(o,cs); + sdsfree(cs); + o = sdscatlen(o,"\n",1); + } + return o; +} - client = listNodeValue(ln); - if (anetPeerToString(client->fd,ip,&port) == -1) continue; - p = flags; - if (client->flags & REDIS_SLAVE) { - if (client->flags & REDIS_MONITOR) - *p++ = 'O'; - else - *p++ = 'S'; - } - if (client->flags & REDIS_MASTER) *p++ = 'M'; - if (p == flags) *p++ = 'N'; - if (client->flags & REDIS_MULTI) *p++ = 'x'; - if (client->flags & REDIS_BLOCKED) *p++ = 'b'; - if (client->flags & REDIS_DIRTY_CAS) *p++ = 'd'; - if (client->flags & REDIS_CLOSE_AFTER_REPLY) *p++ = 'c'; - if (client->flags & REDIS_UNBLOCKED) *p++ = 'u'; - *p++ = '\0'; - o = sdscatprintf(o, - "addr=%s:%d fd=%d idle=%ld flags=%s db=%d sub=%d psub=%d\n", - ip,port,client->fd, - (long)(now - client->lastinteraction), - flags, - client->db->id, - (int) dictSize(client->pubsub_channels), - (int) listLength(client->pubsub_patterns)); - } +void clientCommand(redisClient *c) { + listNode *ln; + listIter li; + redisClient *client; + + if (!strcasecmp(c->argv[1]->ptr,"list") && c->argc == 2) { + sds o = getAllClientsInfoString(); addReplyBulkCBuffer(c,o,sdslen(o)); sdsfree(o); } else if (!strcasecmp(c->argv[1]->ptr,"kill") && c->argc == 3) { @@ -970,7 +1186,7 @@ void rewriteClientCommandVector(redisClient *c, int argc, ...) { c->argv = argv; c->argc = argc; c->cmd = lookupCommand(c->argv[0]->ptr); - redisAssert(c->cmd != NULL); + redisAssertWithInfo(c,NULL,c->cmd != NULL); va_end(ap); } @@ -979,7 +1195,7 @@ void rewriteClientCommandVector(redisClient *c, int argc, ...) { void rewriteClientCommandArgument(redisClient *c, int i, robj *newval) { robj *oldval; - redisAssert(i < c->argc); + redisAssertWithInfo(c,NULL,i < c->argc); oldval = c->argv[i]; c->argv[i] = newval; incrRefCount(newval); @@ -988,6 +1204,136 @@ void rewriteClientCommandArgument(redisClient *c, int i, robj *newval) { /* If this is the command name make sure to fix c->cmd. */ if (i == 0) { c->cmd = lookupCommand(c->argv[0]->ptr); - redisAssert(c->cmd != NULL); + redisAssertWithInfo(c,NULL,c->cmd != NULL); + } +} + +/* This function returns the number of bytes that Redis is virtually + * using to store the reply still not read by the client. + * It is "virtual" since the reply output list may contain objects that + * are shared and are not really using additional memory. + * + * The function returns the total sum of the length of all the objects + * stored in the output list, plus the memory used to allocate every + * list node. The static reply buffer is not taken into account since it + * is allocated anyway. + * + * Note: this function is very fast so can be called as many time as + * the caller wishes. The main usage of this function currently is + * enforcing the client output length limits. */ +unsigned long getClientOutputBufferMemoryUsage(redisClient *c) { + unsigned long list_item_size = sizeof(listNode)+sizeof(robj); + + return c->reply_bytes + (list_item_size*listLength(c->reply)); +} + +/* Get the class of a client, used in order to envorce limits to different + * classes of clients. + * + * The function will return one of the following: + * REDIS_CLIENT_LIMIT_CLASS_NORMAL -> Normal client + * REDIS_CLIENT_LIMIT_CLASS_SLAVE -> Slave or client executing MONITOR command + * REDIS_CLIENT_LIMIT_CLASS_PUBSUB -> Client subscribed to Pub/Sub channels + */ +int getClientLimitClass(redisClient *c) { + if (c->flags & REDIS_SLAVE) return REDIS_CLIENT_LIMIT_CLASS_SLAVE; + if (dictSize(c->pubsub_channels) || listLength(c->pubsub_patterns)) + return REDIS_CLIENT_LIMIT_CLASS_PUBSUB; + return REDIS_CLIENT_LIMIT_CLASS_NORMAL; +} + +int getClientLimitClassByName(char *name) { + if (!strcasecmp(name,"normal")) return REDIS_CLIENT_LIMIT_CLASS_NORMAL; + else if (!strcasecmp(name,"slave")) return REDIS_CLIENT_LIMIT_CLASS_SLAVE; + else if (!strcasecmp(name,"pubsub")) return REDIS_CLIENT_LIMIT_CLASS_PUBSUB; + else return -1; +} + +char *getClientLimitClassName(int class) { + switch(class) { + case REDIS_CLIENT_LIMIT_CLASS_NORMAL: return "normal"; + case REDIS_CLIENT_LIMIT_CLASS_SLAVE: return "slave"; + case REDIS_CLIENT_LIMIT_CLASS_PUBSUB: return "pubsub"; + default: return NULL; + } +} + +/* The function checks if the client reached output buffer soft or hard + * limit, and also update the state needed to check the soft limit as + * a side effect. + * + * Return value: non-zero if the client reached the soft or the hard limit. + * Otherwise zero is returned. */ +int checkClientOutputBufferLimits(redisClient *c) { + int soft = 0, hard = 0, class; + unsigned long used_mem = getClientOutputBufferMemoryUsage(c); + + class = getClientLimitClass(c); + if (server.client_obuf_limits[class].hard_limit_bytes && + used_mem >= server.client_obuf_limits[class].hard_limit_bytes) + hard = 1; + if (server.client_obuf_limits[class].soft_limit_bytes && + used_mem >= server.client_obuf_limits[class].soft_limit_bytes) + soft = 1; + + /* We need to check if the soft limit is reached continuously for the + * specified amount of seconds. */ + if (soft) { + if (c->obuf_soft_limit_reached_time == 0) { + c->obuf_soft_limit_reached_time = server.unixtime; + soft = 0; /* First time we see the soft limit reached */ + } else { + time_t elapsed = server.unixtime - c->obuf_soft_limit_reached_time; + + if (elapsed <= + server.client_obuf_limits[class].soft_limit_seconds) { + soft = 0; /* The client still did not reached the max number of + seconds for the soft limit to be considered + reached. */ + } + } + } else { + c->obuf_soft_limit_reached_time = 0; + } + return soft || hard; +} + +/* Asynchronously close a client if soft or hard limit is reached on the + * output buffer size. The caller can check if the client will be closed + * checking if the client REDIS_CLOSE_ASAP flag is set. + * + * Note: we need to close the client asynchronously because this function is + * called from contexts where the client can't be freed safely, i.e. from the + * lower level functions pushing data inside the client output buffers. */ +void asyncCloseClientOnOutputBufferLimitReached(redisClient *c) { + redisAssert(c->reply_bytes < ULONG_MAX-(1024*64)); + if (c->reply_bytes == 0 || c->flags & REDIS_CLOSE_ASAP) return; + if (checkClientOutputBufferLimits(c)) { + sds client = getClientInfoString(c); + + freeClientAsync(c); + redisLog(REDIS_WARNING,"Client %s scheduled to be closed ASAP for overcoming of output buffer limits.", client); + sdsfree(client); + } +} + +/* Helper function used by freeMemoryIfNeeded() in order to flush slaves + * output buffers without returning control to the event loop. */ +void flushSlavesOutputBuffers(void) { + listIter li; + listNode *ln; + + listRewind(server.slaves,&li); + while((ln = listNext(&li))) { + redisClient *slave = listNodeValue(ln); + int events; + + events = aeGetFileEvents(server.el,slave->fd); + if (events & AE_WRITABLE && + slave->replstate == REDIS_REPL_ONLINE && + listLength(slave->reply)) + { + sendReplyToClient(server.el,slave->fd,slave,0); + } } }