X-Git-Url: https://git.saurik.com/redis.git/blobdiff_plain/4a7893ca9ce334f2a144faa96ef02113bef4b2b2..97e7f8aec3f464967bbe1b38ccfb357141134d09:/src/networking.c diff --git a/src/networking.c b/src/networking.c index f37ecac1..634e2107 100644 --- a/src/networking.c +++ b/src/networking.c @@ -11,29 +11,28 @@ int listMatchObjects(void *a, void *b) { } redisClient *createClient(int fd) { - redisClient *c; - - /* Make sure to allocate a multiple of the page size to prevent wasting - * memory. A page size of 4096 is assumed here. We need to compensate - * for the zmalloc overhead of sizeof(size_t) bytes. */ - size_t size = 8192-sizeof(size_t); - redisAssert(size > sizeof(redisClient)); - c = zmalloc(size); - c->buflen = size-sizeof(redisClient); + redisClient *c = zmalloc(sizeof(redisClient)); c->bufpos = 0; anetNonBlock(NULL,fd); anetTcpNoDelay(NULL,fd); if (!c) return NULL; + if (aeCreateFileEvent(server.el,fd,AE_READABLE, + readQueryFromClient, c) == AE_ERR) + { + close(fd); + zfree(c); + return NULL; + } + selectDb(c,0); c->fd = fd; c->querybuf = sdsempty(); + c->reqtype = 0; c->argc = 0; c->argv = NULL; + c->multibulklen = 0; c->bulklen = -1; - c->multibulk = 0; - c->mbargc = 0; - c->mbargv = NULL; c->sentlen = 0; c->flags = 0; c->lastinteraction = time(NULL); @@ -51,17 +50,17 @@ redisClient *createClient(int fd) { c->pubsub_patterns = listCreate(); listSetFreeMethod(c->pubsub_patterns,decrRefCount); listSetMatchMethod(c->pubsub_patterns,listMatchObjects); - if (aeCreateFileEvent(server.el, c->fd, AE_READABLE, - readQueryFromClient, c) == AE_ERR) { - freeClient(c); - return NULL; - } listAddNodeTail(server.clients,c); initClientMultiState(c); return c; } -int _ensureFileEvent(redisClient *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) { + /* When CLOSE_AFTER_REPLY is set, no more replies may be added! */ + redisAssert(!(c->flags & REDIS_CLOSE_AFTER_REPLY)); + if (c->fd <= 0) return REDIS_ERR; if (c->bufpos == 0 && listLength(c->reply) == 0 && (c->replstate == REDIS_REPL_NONE || @@ -71,90 +70,187 @@ int _ensureFileEvent(redisClient *c) { return REDIS_OK; } -void _addReplyObjectToList(redisClient *c, robj *obj) { - redisAssert(obj->type == REDIS_STRING && - obj->encoding == REDIS_ENCODING_RAW); - listAddNodeTail(c->reply,obj); +/* Create a duplicate of the last object in the reply list when + * it is not exclusively owned by the reply list. */ +robj *dupLastObjectIfNeeded(list *reply) { + robj *new, *cur; + listNode *ln; + redisAssert(listLength(reply) > 0); + ln = listLast(reply); + cur = listNodeValue(ln); + if (cur->refcount > 1) { + new = dupStringObject(cur); + decrRefCount(cur); + listNodeValue(ln) = new; + } + return listNodeValue(ln); } -void _ensureBufferInReplyList(redisClient *c) { - sds buffer = sdsnewlen(NULL,REDIS_REPLY_CHUNK_SIZE); - sdsupdatelen(buffer); /* sdsnewlen expects non-empty string */ - listAddNodeTail(c->reply,createObject(REDIS_REPLY_NODE,buffer)); +int _addReplyToBuffer(redisClient *c, char *s, size_t len) { + size_t available = sizeof(c->buf)-c->bufpos; + + /* If there already are entries in the reply list, we cannot + * add anything more to the static buffer. */ + if (listLength(c->reply) > 0) return REDIS_ERR; + + /* Check that the buffer has enough space available for this string. */ + if (len > available) return REDIS_ERR; + + memcpy(c->buf+c->bufpos,s,len); + c->bufpos+=len; + return REDIS_OK; } -void _addReplyStringToBuffer(redisClient *c, char *s, size_t len) { - size_t available = 0; - redisAssert(len < REDIS_REPLY_CHUNK_THRESHOLD); - if (listLength(c->reply) > 0) { - robj *o = listNodeValue(listLast(c->reply)); +void _addReplyObjectToList(redisClient *c, robj *o) { + robj *tail; + if (listLength(c->reply) == 0) { + incrRefCount(o); + listAddNodeTail(c->reply,o); + } else { + tail = listNodeValue(listLast(c->reply)); - /* Make sure to append to a reply node with enough bytes available. */ - if (o->type == REDIS_REPLY_NODE) available = sdsavail(o->ptr); - if (o->type != REDIS_REPLY_NODE || len > available) { - _ensureBufferInReplyList(c); - _addReplyStringToBuffer(c,s,len); + /* Append to this object when possible. */ + if (tail->ptr != NULL && + sdslen(tail->ptr)+sdslen(o->ptr) <= REDIS_REPLY_CHUNK_BYTES) + { + tail = dupLastObjectIfNeeded(c->reply); + tail->ptr = sdscatlen(tail->ptr,o->ptr,sdslen(o->ptr)); } else { - o->ptr = sdscatlen(o->ptr,s,len); + incrRefCount(o); + listAddNodeTail(c->reply,o); } + } +} + +/* This method takes responsibility over the sds. When it is no longer + * needed it will be free'd, otherwise it ends up in a robj. */ +void _addReplySdsToList(redisClient *c, sds s) { + robj *tail; + if (listLength(c->reply) == 0) { + listAddNodeTail(c->reply,createObject(REDIS_STRING,s)); } else { - available = c->buflen-c->bufpos; - if (len > available) { - _ensureBufferInReplyList(c); - _addReplyStringToBuffer(c,s,len); + tail = listNodeValue(listLast(c->reply)); + + /* Append to this object when possible. */ + if (tail->ptr != NULL && + sdslen(tail->ptr)+sdslen(s) <= REDIS_REPLY_CHUNK_BYTES) + { + tail = dupLastObjectIfNeeded(c->reply); + tail->ptr = sdscatlen(tail->ptr,s,sdslen(s)); + sdsfree(s); } else { - memcpy(c->buf+c->bufpos,s,len); - c->bufpos += len; + listAddNodeTail(c->reply,createObject(REDIS_STRING,s)); } } } -void addReply(redisClient *c, robj *obj) { - if (_ensureFileEvent(c) != REDIS_OK) return; - if (server.vm_enabled && obj->storage != REDIS_VM_MEMORY) { - /* Returns a new object with refcount 1 */ - obj = dupStringObject(obj); +void _addReplyStringToList(redisClient *c, char *s, size_t len) { + robj *tail; + if (listLength(c->reply) == 0) { + listAddNodeTail(c->reply,createStringObject(s,len)); } else { - /* This increments the refcount. */ - obj = getDecodedObject(obj); + tail = listNodeValue(listLast(c->reply)); + + /* Append to this object when possible. */ + if (tail->ptr != NULL && + sdslen(tail->ptr)+len <= REDIS_REPLY_CHUNK_BYTES) + { + tail = dupLastObjectIfNeeded(c->reply); + tail->ptr = sdscatlen(tail->ptr,s,len); + } else { + listAddNodeTail(c->reply,createStringObject(s,len)); + } } +} - if (sdslen(obj->ptr) < REDIS_REPLY_CHUNK_THRESHOLD) { - _addReplyStringToBuffer(c,obj->ptr,sdslen(obj->ptr)); - decrRefCount(obj); +void addReply(redisClient *c, robj *obj) { + if (_installWriteEvent(c) != REDIS_OK) return; + redisAssert(!server.vm_enabled || obj->storage == REDIS_VM_MEMORY); + + /* This is an important place where we can avoid copy-on-write + * when there is a saving child running, avoiding touching the + * refcount field of the object if it's not needed. + * + * If the encoding is RAW and there is room in the static buffer + * we'll be able to send the object to the client without + * messing with its page. */ + if (obj->encoding == REDIS_ENCODING_RAW) { + if (_addReplyToBuffer(c,obj->ptr,sdslen(obj->ptr)) != REDIS_OK) + _addReplyObjectToList(c,obj); } else { - _addReplyObjectToList(c,obj); + obj = getDecodedObject(obj); + if (_addReplyToBuffer(c,obj->ptr,sdslen(obj->ptr)) != REDIS_OK) + _addReplyObjectToList(c,obj); + decrRefCount(obj); } } void addReplySds(redisClient *c, sds s) { - if (_ensureFileEvent(c) != REDIS_OK) { + if (_installWriteEvent(c) != REDIS_OK) { /* The caller expects the sds to be free'd. */ sdsfree(s); return; } - if (sdslen(s) < REDIS_REPLY_CHUNK_THRESHOLD) { - _addReplyStringToBuffer(c,s,sdslen(s)); + if (_addReplyToBuffer(c,s,sdslen(s)) == REDIS_OK) { sdsfree(s); } else { - _addReplyObjectToList(c,createObject(REDIS_STRING,s)); + /* This method free's the sds when it is no longer needed. */ + _addReplySdsToList(c,s); } } void addReplyString(redisClient *c, char *s, size_t len) { - if (_ensureFileEvent(c) != REDIS_OK) return; - if (len < REDIS_REPLY_CHUNK_THRESHOLD) { - _addReplyStringToBuffer(c,s,len); - } else { - _addReplyObjectToList(c,createStringObject(s,len)); - } + if (_installWriteEvent(c) != REDIS_OK) return; + if (_addReplyToBuffer(c,s,len) != REDIS_OK) + _addReplyStringToList(c,s,len); +} + +void _addReplyError(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)); +} + +void addReplyErrorFormat(redisClient *c, const char *fmt, ...) { + va_list ap; + va_start(ap,fmt); + sds s = sdscatvprintf(sdsempty(),fmt,ap); + va_end(ap); + _addReplyError(c,s,sdslen(s)); + sdsfree(s); +} + +void _addReplyStatus(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)); +} + +void addReplyStatusFormat(redisClient *c, const char *fmt, ...) { + va_list ap; + va_start(ap,fmt); + sds s = sdscatvprintf(sdsempty(),fmt,ap); + va_end(ap); + _addReplyStatus(c,s,sdslen(s)); + sdsfree(s); } /* Adds an empty object to the reply list that will contain the multi bulk * length, which is not known when this function is called. */ void *addDeferredMultiBulkLength(redisClient *c) { - if (_ensureFileEvent(c) != REDIS_OK) return NULL; - _addReplyObjectToList(c,createObject(REDIS_STRING,NULL)); + /* 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; + listAddNodeTail(c->reply,createObject(REDIS_STRING,NULL)); return listLast(c->reply); } @@ -170,8 +266,9 @@ void setDeferredMultiBulkLength(redisClient *c, void *node, long length) { len->ptr = sdscatprintf(sdsempty(),"*%ld\r\n",length); if (ln->next != NULL) { next = listNodeValue(ln->next); - /* Only glue when the next node is a reply chunk. */ - if (next->type == REDIS_REPLY_NODE) { + + /* Only glue when the next node is non-NULL (an sds in this case) */ + if (next->ptr != NULL) { len->ptr = sdscatlen(len->ptr,next->ptr,sdslen(next->ptr)); listDelNode(c->reply,ln->next); } @@ -242,23 +339,11 @@ void addReplyBulkCString(redisClient *c, char *s) { } } -void acceptHandler(aeEventLoop *el, int fd, void *privdata, int mask) { - int cport, cfd; - char cip[128]; +static void acceptCommonHandler(int fd) { redisClient *c; - REDIS_NOTUSED(el); - REDIS_NOTUSED(mask); - REDIS_NOTUSED(privdata); - - cfd = anetAccept(server.neterr, fd, cip, &cport); - if (cfd == AE_ERR) { - redisLog(REDIS_VERBOSE,"Accepting client connection: %s", server.neterr); - return; - } - redisLog(REDIS_VERBOSE,"Accepted %s:%d", cip, cport); - if ((c = createClient(cfd)) == NULL) { + if ((c = createClient(fd)) == NULL) { redisLog(REDIS_WARNING,"Error allocating resoures for the client"); - close(cfd); /* May be already closed, just ingore errors */ + close(fd); /* May be already closed, just ingore errors */ return; } /* If maxclient directive is set and this is one client more... close the @@ -278,15 +363,43 @@ void acceptHandler(aeEventLoop *el, int fd, void *privdata, int mask) { server.stat_numconnections++; } +void acceptTcpHandler(aeEventLoop *el, int fd, void *privdata, int mask) { + int cport, cfd; + char cip[128]; + REDIS_NOTUSED(el); + REDIS_NOTUSED(mask); + REDIS_NOTUSED(privdata); + + cfd = anetTcpAccept(server.neterr, fd, cip, &cport); + if (cfd == AE_ERR) { + redisLog(REDIS_VERBOSE,"Accepting client connection: %s", server.neterr); + return; + } + redisLog(REDIS_VERBOSE,"Accepted %s:%d", cip, cport); + acceptCommonHandler(cfd); +} + +void acceptUnixHandler(aeEventLoop *el, int fd, void *privdata, int mask) { + int cfd; + REDIS_NOTUSED(el); + REDIS_NOTUSED(mask); + REDIS_NOTUSED(privdata); + + cfd = anetUnixAccept(server.neterr, fd); + if (cfd == AE_ERR) { + redisLog(REDIS_VERBOSE,"Accepting client connection: %s", server.neterr); + return; + } + redisLog(REDIS_VERBOSE,"Accepted connection to %s", server.unixsocket); + acceptCommonHandler(cfd); +} + + static void freeClientArgv(redisClient *c) { int j; - for (j = 0; j < c->argc; j++) decrRefCount(c->argv[j]); - for (j = 0; j < c->mbargc; j++) - decrRefCount(c->mbargv[j]); c->argc = 0; - c->mbargc = 0; } void freeClient(redisClient *c) { @@ -354,6 +467,7 @@ void freeClient(redisClient *c) { /* Case 2: we lost the connection with the master. */ if (c->flags & REDIS_MASTER) { server.master = NULL; + /* FIXME */ server.replstate = REDIS_REPL_CONNECT; /* Since we lost the connection with the master, we should also * close the connection with all our slaves if we have any, so @@ -367,7 +481,6 @@ void freeClient(redisClient *c) { } /* Release memory */ zfree(c->argv); - zfree(c->mbargv); freeClientMultiState(c); zfree(c); } @@ -452,6 +565,9 @@ void sendReplyToClient(aeEventLoop *el, int fd, void *privdata, int mask) { if (listLength(c->reply) == 0) { c->sentlen = 0; aeDeleteFileEvent(server.el,c->fd,AE_WRITABLE); + + /* Close connection after entire reply has been sent. */ + if (c->flags & REDIS_CLOSE_AFTER_REPLY) freeClient(c); } } @@ -536,8 +652,9 @@ void sendReplyToClientWritev(aeEventLoop *el, int fd, void *privdata, int mask) /* resetClient prepare the client to process the next command */ void resetClient(redisClient *c) { freeClientArgv(c); + c->reqtype = 0; + c->multibulklen = 0; c->bulklen = -1; - c->multibulk = 0; } void closeTimedoutClients(void) { @@ -568,84 +685,172 @@ void closeTimedoutClients(void) { } } -void processInputBuffer(redisClient *c) { -again: - /* Before to process the input buffer, make sure the client is not - * waitig for a blocking operation such as BLPOP. Note that the first - * iteration the client is never blocked, otherwise the processInputBuffer - * would not be called at all, but after the execution of the first commands - * in the input buffer the client may be blocked, and the "goto again" - * will try to reiterate. The following line will make it return asap. */ - if (c->flags & REDIS_BLOCKED || c->flags & REDIS_IO_WAIT) return; - if (c->bulklen == -1) { - /* Read the first line of the query */ - char *p = strchr(c->querybuf,'\n'); - size_t querylen; - - if (p) { - sds query, *argv; - int argc, j; - - query = c->querybuf; - c->querybuf = sdsempty(); - querylen = 1+(p-(query)); - if (sdslen(query) > querylen) { - /* leave data after the first line of the query in the buffer */ - c->querybuf = sdscatlen(c->querybuf,query+querylen,sdslen(query)-querylen); - } - *p = '\0'; /* remove "\n" */ - if (*(p-1) == '\r') *(p-1) = '\0'; /* and "\r" if any */ - sdsupdatelen(query); - - /* Now we can split the query in arguments */ - argv = sdssplitlen(query,sdslen(query)," ",1,&argc); - sdsfree(query); - - if (c->argv) zfree(c->argv); - c->argv = zmalloc(sizeof(robj*)*argc); - - for (j = 0; j < argc; j++) { - if (sdslen(argv[j])) { - c->argv[c->argc] = createObject(REDIS_STRING,argv[j]); - c->argc++; - } else { - sdsfree(argv[j]); +int processInlineBuffer(redisClient *c) { + char *newline = strstr(c->querybuf,"\r\n"); + int argc, j; + sds *argv; + size_t querylen; + + /* Nothing to do without a \r\n */ + if (newline == NULL) + return REDIS_ERR; + + /* Split the input buffer up to the \r\n */ + querylen = newline-(c->querybuf); + argv = sdssplitlen(c->querybuf,querylen," ",1,&argc); + + /* Leave data after the first line of the query in the buffer */ + c->querybuf = sdsrange(c->querybuf,querylen+2,-1); + + /* Setup argv array on client structure */ + if (c->argv) zfree(c->argv); + c->argv = zmalloc(sizeof(robj*)*argc); + + /* Create redis objects for all arguments. */ + for (c->argc = 0, j = 0; j < argc; j++) { + if (sdslen(argv[j])) { + c->argv[c->argc] = createObject(REDIS_STRING,argv[j]); + c->argc++; + } else { + sdsfree(argv[j]); + } + } + zfree(argv); + return REDIS_OK; +} + +/* Helper function. Trims query buffer to make the function that processes + * multi bulk requests idempotent. */ +static void setProtocolError(redisClient *c, int pos) { + c->flags |= REDIS_CLOSE_AFTER_REPLY; + c->querybuf = sdsrange(c->querybuf,pos,-1); +} + +int processMultibulkBuffer(redisClient *c) { + char *newline = NULL; + char *eptr; + int pos = 0, tolerr; + long bulklen; + + if (c->multibulklen == 0) { + /* The client should have been reset */ + redisAssert(c->argc == 0); + + /* Multi bulk length cannot be read without a \r\n */ + newline = strstr(c->querybuf,"\r\n"); + if (newline == NULL) + return REDIS_ERR; + + /* 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] == '*'); + c->multibulklen = strtol(c->querybuf+1,&eptr,10); + pos = (newline-c->querybuf)+2; + if (c->multibulklen <= 0) { + c->querybuf = sdsrange(c->querybuf,pos,-1); + return REDIS_OK; + } else if (c->multibulklen > 1024*1024) { + addReplyError(c,"Protocol error: invalid multibulk length"); + setProtocolError(c,pos); + return REDIS_ERR; + } + + /* Setup argv array on client structure */ + if (c->argv) zfree(c->argv); + c->argv = zmalloc(sizeof(robj*)*c->multibulklen); + + /* Search new newline */ + newline = strstr(c->querybuf+pos,"\r\n"); + } + + redisAssert(c->multibulklen > 0); + while(c->multibulklen) { + /* Read bulk length if unknown */ + if (c->bulklen == -1) { + newline = strstr(c->querybuf+pos,"\r\n"); + if (newline != NULL) { + if (c->querybuf[pos] != '$') { + addReplyErrorFormat(c, + "Protocol error: expected '$', got '%c'", + c->querybuf[pos]); + setProtocolError(c,pos); + return REDIS_ERR; + } + + bulklen = strtol(c->querybuf+pos+1,&eptr,10); + tolerr = (eptr[0] != '\r'); + if (tolerr || bulklen == LONG_MIN || bulklen == LONG_MAX || + bulklen < 0 || bulklen > 1024*1024*1024) + { + addReplyError(c,"Protocol error: invalid bulk length"); + setProtocolError(c,pos); + return REDIS_ERR; } + pos += eptr-(c->querybuf+pos)+2; + c->bulklen = bulklen; + } else { + /* No newline in current buffer, so wait for more data */ + break; } - zfree(argv); - if (c->argc) { - /* Execute the command. If the client is still valid - * after processCommand() return and there is something - * on the query buffer try to process the next command. */ - if (processCommand(c) && sdslen(c->querybuf)) goto again; + } + + /* Read bulk argument */ + if (sdslen(c->querybuf)-pos < (unsigned)(c->bulklen+2)) { + /* Not enough data (+2 == trailing \r\n) */ + break; + } 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); + + /* We're done when c->multibulk == 0 */ + if (c->multibulklen == 0) { + return REDIS_OK; + } + return REDIS_ERR; +} + +void processInputBuffer(redisClient *c) { + /* Keep processing while there is something in the input buffer */ + while(sdslen(c->querybuf)) { + /* Immediately abort if the client is in the middle of something. */ + if (c->flags & REDIS_BLOCKED || c->flags & REDIS_IO_WAIT) return; + + /* REDIS_CLOSE_AFTER_REPLY closes the connection once the reply is + * written to the client. Make sure to not let the reply grow after + * this flag has been set (i.e. don't process more commands). */ + if (c->flags & REDIS_CLOSE_AFTER_REPLY) return; + + /* Determine request type when unknown. */ + if (!c->reqtype) { + if (c->querybuf[0] == '*') { + c->reqtype = REDIS_REQ_MULTIBULK; } else { - /* Nothing to process, argc == 0. Just process the query - * buffer if it's not empty or return to the caller */ - if (sdslen(c->querybuf)) goto again; + c->reqtype = REDIS_REQ_INLINE; } - return; - } else if (sdslen(c->querybuf) >= REDIS_REQUEST_MAX_SIZE) { - redisLog(REDIS_VERBOSE, "Client protocol error"); - freeClient(c); - return; } - } else { - /* Bulk read handling. Note that if we are at this point - the client already sent a command terminated with a newline, - we are reading the bulk data that is actually the last - argument of the command. */ - int qbl = sdslen(c->querybuf); - - if (c->bulklen <= qbl) { - /* Copy everything but the final CRLF as final argument */ - c->argv[c->argc] = createStringObject(c->querybuf,c->bulklen-2); - c->argc++; - c->querybuf = sdsrange(c->querybuf,c->bulklen,-1); - /* Process the command. If the client is still valid after - * the processing and there is more data in the buffer - * try to parse it. */ - if (processCommand(c) && sdslen(c->querybuf)) goto again; - return; + + if (c->reqtype == REDIS_REQ_INLINE) { + if (processInlineBuffer(c) != REDIS_OK) break; + } else if (c->reqtype == REDIS_REQ_MULTIBULK) { + if (processMultibulkBuffer(c) != REDIS_OK) break; + } else { + redisPanic("Unknown request type"); + } + + /* Multibulk processing could see a <= 0 length. */ + if (c->argc == 0) { + resetClient(c); + } else { + /* Only reset the client when the command was executed. */ + if (processCommand(c) == REDIS_OK) + resetClient(c); } } } @@ -672,7 +877,7 @@ void readQueryFromClient(aeEventLoop *el, int fd, void *privdata, int mask) { return; } if (nread) { - c->querybuf = sdscatlen(c->querybuf, buf, nread); + c->querybuf = sdscatlen(c->querybuf,buf,nread); c->lastinteraction = time(NULL); } else { return;