githubmirror/redis

src/networking.c
6 023 строки238 KB
YaacovHazan
Strip CRLF from error and simple string replies (#15496)
20 июл 2026, 21:00
20 июл 2026, 21:00e1cc3dc
/*
 * Copyright (c) 2009-Present, Redis Ltd.
 * All rights reserved.
 *
 * Copyright (c) 2024-present, Valkey contributors.
 * All rights reserved.
 *
 * Licensed under your choice of (a) the Redis Source Available License 2.0
 * (RSALv2); or (b) the Server Side Public License v1 (SSPLv1); or (c) the
 * GNU Affero General Public License v3 (AGPLv3).
 *
 * Portions of this file are available under BSD3 terms; see REDISCONTRIBUTIONS for more information.
 */

#include "server.h"
#include "atomicvar.h"
#include "cluster.h"
#include "cluster_slot_stats.h"
#include "script.h"
#include "fpconv_dtoa.h"
#include "fmtargs.h"
#include "cluster_asm.h"
#include "memory_prefetch.h"
#include "connection.h"
#include <sys/socket.h>
#include <sys/uio.h>
#include <math.h>
#include <ctype.h>

static void setProtocolError(const char *errstr, client *c);
static void pauseClientsByClient(mstime_t end, int isPauseClientAll);
char *getClientSockname(client *c);
static inline int clientTypeIsSlave(client *c);
static inline int _clientHasPendingRepliesSlave(client *c);
static inline int _clientHasPendingRepliesNonSlave(client *c);
static inline int _writeToClientNonSlave(client *c, ssize_t *nwritten);
static inline int _writeToClientSlave(client *c, ssize_t *nwritten);
static pendingCommand *acquirePendingCommand(void);
static inline void reclaimPendingCommand(client *c, pendingCommand *pcmd);
static size_t getClientOutputBufferLogicalSize(client *c);

int ProcessingEventsWhileBlocked = 0; /* See processEventsWhileBlocked(). */
__thread sds thread_reusable_qb = NULL;
__thread int thread_reusable_qb_used = 0; /* Avoid multiple clients using reusable query
                                         * buffer due to nested command execution. */

/* Return the size consumed from the allocator, for the specified SDS string,
 * including internal fragmentation. This function is used in order to compute
 * the client output buffer size. */
size_t sdsZmallocSize(sds s) {
    void *sh = sdsAllocPtr(s);
    return zmalloc_size(sh);
}

/* Return the amount of memory used by the sds string at object->ptr
 * for a string object. This includes internal fragmentation. */
size_t getStringObjectSdsUsedMemory(robj *o) {
    serverAssertWithInfo(NULL,o,o->type == OBJ_STRING);
    switch(o->encoding) {
    case OBJ_ENCODING_RAW: return sdsZmallocSize(o->ptr);
    case OBJ_ENCODING_EMBSTR: return zmalloc_size(o)-sizeof(robj);
    default: return 0; /* Just integer encoding for now. */
    }
}

/* Return the length of a string object.
 * This does NOT includes internal fragmentation or sds unused space. */
size_t getStringObjectLen(robj *o) {
    serverAssertWithInfo(NULL,o,o->type == OBJ_STRING);
    switch(o->encoding) {
    case OBJ_ENCODING_RAW: return sdslen(o->ptr);
    case OBJ_ENCODING_EMBSTR: return sdslen(o->ptr);
    default: return 0; /* Just integer encoding for now. */
    }
}

/* Client.reply list dup and free methods. */
void *dupClientReplyValue(void *o) {
    clientReplyBlock *old = o;
    clientReplyBlock *buf = zmalloc(sizeof(clientReplyBlock) + old->size);
    memcpy(buf, o, sizeof(clientReplyBlock) + old->size);
    return buf;
}

void freeClientReplyValue(void *o) {
    zfree(o);
}

/* This function links the client to the global linked list of clients.
 * unlinkClient() does the opposite, among other things. */
void linkClient(client *c) {
    listAddNodeTail(server.clients,c);
    /* Note that we remember the linked list node where the client is stored,
     * this way removing the client in unlinkClient() will not require
     * a linear scan, but just a constant time operation. */
    c->client_list_node = listLast(server.clients);
    uint64_t id = htonu64(c->id);
    raxInsert(server.clients_index,(unsigned char*)&id,sizeof(id),c,NULL);
}

/* Initialize client authentication state.
 */
static void clientSetDefaultAuth(client *c) {
    /* If the default user does not require authentication, the user is
     * directly authenticated. */
    clientSetUser(c, DefaultUser);
    c->authenticated = (c->user->flags & USER_FLAG_NOPASS) &&
                       !(c->user->flags & USER_FLAG_DISABLED);
}

int authRequired(client *c) {
    /* Check if the user is authenticated. This check is skipped in case
     * the default user is flagged as "nopass" and is active. */
    uint32_t default_flags;
    atomicGet(DefaultUser->flags, default_flags);
    int auth_required = (!(default_flags & USER_FLAG_NOPASS) ||
                          (default_flags & USER_FLAG_DISABLED)) &&
                        !c->authenticated;
    return auth_required;
}

client *createClient(connection *conn) {
    client *c = zmalloc(sizeof(client));

    /* passing NULL as conn it is possible to create a non connected client.
     * This is useful since all the commands needs to be executed
     * in the context of a client. When commands are executed in other
     * contexts (for instance a Lua script) we need a non connected client. */
    if (conn) {
        connEnableTcpNoDelay(conn);
        if (server.tcpkeepalive)
            connKeepAlive(conn,server.tcpkeepalive);
        connSetReadHandler(conn, readQueryFromClient);
        connSetPrivateData(conn, c);
    }
    c->buf = zmalloc_usable(PROTO_REPLY_CHUNK_BYTES, &c->buf_usable_size);
    selectDb(c,0);
    uint64_t client_id;
    atomicGetIncr(server.next_client_id, client_id, 1);
    c->id = client_id;
    c->tid = IOTHREAD_MAIN_THREAD_ID;
    c->running_tid = IOTHREAD_MAIN_THREAD_ID;
    if (conn) server.io_threads_clients_num[c->tid]++;
#ifdef LOG_REQ_RES
    reqresReset(c, 0);
    c->resp = server.client_default_resp;
#else
    c->resp = 2;
#endif
    c->conn = conn;
    c->name = NULL;
    c->lib_name = NULL;
    c->lib_ver = NULL;
    c->bufpos = 0;
    c->buf_peak = c->buf_usable_size;
    c->buf_peak_last_reset_time = server.unixtime;
    c->buf_encoded = 0;
    c->last_header = NULL;
    c->ref_repl_buf_node = NULL;
    c->ref_block_pos = 0;
    c->io_curr_repl_node = NULL;
    c->io_curr_block_pos = 0;
    c->io_bound_repl_node = NULL;
    c->io_bound_block_pos = 0;
    c->qb_pos = 0;
    c->querybuf = NULL;
    c->querybuf_peak = 0;
    c->reqtype = 0;
    c->argc = 0;
    c->argv = NULL;
    c->argv_len = 0;
    c->all_argv_len_sum = 0;
    c->pending_cmds.head = c->pending_cmds.tail = NULL;
    c->pending_cmds.len = c->pending_cmds.ready_len = 0;
    c->current_pending_cmd = NULL;
    c->original_argc = 0;
    c->original_argv = NULL;
    c->deferred_objects = NULL;
    c->deferred_objects_num = 0;
    c->io_deferred_objects = NULL;
    c->io_deferred_objects_num = 0;
    c->io_deferred_objects_size = 0;
    c->cmd = c->lastcmd = c->realcmd = c->lookedcmd = NULL;
    c->cur_script = NULL;
    c->multibulklen = 0;
    c->bulklen = -1;
    c->sentlen = 0;
    c->flags = 0;
    c->io_flags = CLIENT_IO_READ_ENABLED | CLIENT_IO_WRITE_ENABLED;
    c->read_error = 0;
    c->slot = -1;
    c->cluster_compatibility_check_slot = -2;
    c->ctime = c->lastinteraction = server.unixtime;
    c->io_lastinteraction = 0;
    c->duration = 0;
    c->user = DefaultUser; /* Set a safe default value: clientSetDefaultAuth reads c->user. */
    clientSetDefaultAuth(c);
    c->replstate = REPL_STATE_NONE;
    c->repl_start_cmd_stream_on_ack = 0;
    c->reploff = 0;
    c->reploff_next = 0;
    c->read_reploff = 0;
    c->io_read_reploff = 0;
    c->repl_applied = 0;
    c->repl_ack_off = 0;
    c->repl_ack_time = 0;
    c->io_repl_ack_time = 0;
    c->repl_aof_off = 0;
    c->repl_last_partial_write = 0;
    c->slave_listening_port = 0;
    c->slave_addr = NULL;
    c->slave_capa = SLAVE_CAPA_NONE;
    c->slave_req = SLAVE_REQ_NONE;
    c->main_ch_client_id = 0;
    c->reply = listCreate();
    c->deferred_reply_errors = NULL;
    c->reply_bytes = c->reply_bytes_shared = c->reply_bytes_unshared = 0;
    c->obuf_soft_limit_reached_time = 0;
    listSetFreeMethod(c->reply,freeClientReplyValue);
    listSetDupMethod(c->reply,dupClientReplyValue);
    initClientBlockingState(c);
    c->woff = 0;
    c->watched_keys = listCreate();
    c->pubsub_channels = dictCreate(&objectKeyPointerValueDictType);
    c->pubsub_patterns = dictCreate(&objectKeyPointerValueDictType);
    c->pubsubshard_channels = dictCreate(&objectKeyPointerValueDictType);
    c->peerid = NULL;
    c->sockname = NULL;
    c->client_list_node = NULL;
    c->io_thread_client_list_node = NULL;
    c->io_thread_compression_clients_node = NULL;
    c->postponed_list_node = NULL;
    c->client_tracking_redirection = 0;
    c->client_tracking_prefixes = NULL;
    c->io_last_client_cron = 0;
    c->io_last_repl_cron = 0;
    c->last_memory_usage = 0;
    c->last_memory_type = CLIENT_TYPE_NORMAL;
    c->module_blocked_client = NULL;
    c->module_auth_ctx = NULL;
    c->auth_callback = NULL;
    c->auth_callback_privdata = NULL;
    c->auth_module = NULL;
    listInitNode(&c->clients_pending_write_node, c);
    listInitNode(&c->pending_ref_reply_node, c);
    c->mem_usage_bucket = NULL;
    c->mem_usage_bucket_node = NULL;
    c->net_input_bytes_curr_cmd = 0;
    c->net_output_bytes_curr_cmd = 0;
    if (conn) linkClient(c);
    initClientMultiState(c);
    c->net_input_bytes = 0;
    c->net_output_bytes = 0;
    c->commands_processed = 0;
    c->last_ts_when_counted_as_active = 0;
    c->stat_total_read_events = 0;
    c->stat_avg_pipeline_length_sum = 0;
    c->stat_avg_pipeline_length_cnt = 0;
    c->task = NULL;
    c->node_id = NULL;
    c->himport_fieldsets = NULL;
    c->compression_level = 0;
    c->compression_state = NULL;
    atomicSet(c->pending_read, 0);
    return c;
}

void installClientWriteHandler(client *c) {
    int ae_barrier = 0;
    /* For the fsync=always policy, we want that a given FD is never
     * served for reading and writing in the same event loop iteration,
     * so that in the middle of receiving the query, and serving it
     * to the client, we'll call beforeSleep() that will do the
     * actual fsync of AOF to disk. the write barrier ensures that. */
    if (server.aof_state == AOF_ON &&
        server.aof_fsync == AOF_FSYNC_ALWAYS)
    {
        ae_barrier = 1;
    }
    if (connSetWriteHandlerWithBarrier(c->conn, sendReplyToClient, ae_barrier) == C_ERR) {
        freeClientAsync(c);
    }
}

/* This function puts the client in the queue of clients that should write
 * their output buffers to the socket. Note that it does not *yet* install
 * the write handler, to start clients are put in a queue of clients that need
 * to write, so we try to do that before returning in the event loop (see the
 * handleClientsWithPendingWrites() function).
 * If we fail and there is more data to write, compared to what the socket
 * buffers can hold, then we'll really install the handler. */
void putClientInPendingWriteQueue(client *c) {
    /* Schedule the client to write the output buffers to the socket only
     * if not already done and, for slaves, if the slave can actually receive
     * writes at this stage. */
    if (!(c->flags & CLIENT_PENDING_WRITE) &&
        (c->replstate == REPL_STATE_NONE ||
         c->replstate == SLAVE_STATE_SEND_BULK_AND_STREAM ||
         (c->replstate == SLAVE_STATE_ONLINE && !c->repl_start_cmd_stream_on_ack)))
    {
        /* Here instead of installing the write handler, we just flag the
         * client and put it into a list of clients that have something
         * to write to the socket. This way before re-entering the event
         * loop, we can try to directly write to the client sockets avoiding
         * a system call. We'll only really install the write handler if
         * we'll not be able to write the whole reply at once. */
        c->flags |= CLIENT_PENDING_WRITE;
        listLinkNodeHead(server.clients_pending_write, &c->clients_pending_write_node);
    }
}

static inline int _prepareClientToWrite(client *c) {
    const uint64_t _flags = c->flags;
    /* If it's the Lua client we always return ok without installing any
     * handler since there is no socket at all. */
    if (unlikely(_flags & (CLIENT_SCRIPT|CLIENT_MODULE))) return C_OK;

    /* If CLIENT_CLOSE_ASAP flag is set, we need not write anything. */
    if (unlikely(_flags & CLIENT_CLOSE_ASAP)) return C_ERR;

    /* CLIENT REPLY OFF / SKIP handling: don't send replies.
     * CLIENT_PUSHING handling: disables the reply silencing flags. */
    if (unlikely((_flags & (CLIENT_REPLY_OFF|CLIENT_REPLY_SKIP)) &&
        !(_flags & CLIENT_PUSHING))) return C_ERR;

    /* Masters don't receive replies, unless CLIENT_MASTER_FORCE_REPLY flag
     * is set. */
    if (unlikely((_flags & CLIENT_MASTER) &&
        !(_flags & CLIENT_MASTER_FORCE_REPLY))) return C_ERR;

    if (unlikely(!c->conn)) return C_ERR; /* Fake client for AOF loading. */

    /* Schedule the client to write the output buffers to the socket, unless
     * it should already be setup to do so (it has already pending data).
     *
     * If the client runs in an IO thread, we should not put the client in the
     * pending write queue. Instead, we will install the write handler to the
     * corresponding IO thread’s event loop and let it handle the reply. */
    if (likely(c->running_tid == IOTHREAD_MAIN_THREAD_ID) && !clientHasPendingReplies(c))
        putClientInPendingWriteQueue(c);

    /* Authorize the caller to queue in the output buffer of this client. */
    return C_OK;
}

/* 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 C_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
 * (used to load AOF in memory), a master or because the setup of the write
 * handler failed, the function returns C_ERR.
 *
 * The function may return C_OK without actually installing the write
 * event handler in the following cases:
 *
 * 1) The event handler should already be installed since the output buffer
 *    already contains something.
 * 2) The client is a slave but not yet online, so we want to just accumulate
 *    writes in the buffer but not actually sending them yet.
 *
 * Typically gets called every time a reply is built, before adding more
 * data to the clients output buffers. If the function returns C_ERR no
 * data should be appended to the output buffers. */
int prepareClientToWrite(client *c) {
    return _prepareClientToWrite(c);
}

/* -----------------------------------------------------------------------------
 * Low level functions to add more data to output buffers.
 * -------------------------------------------------------------------------- */

static int tryAddPayload(client *c, char *buf, size_t *used, size_t size, uint8_t type, const void *payload, size_t len) {
    if (*used + sizeof(payloadHeader) + len > size) return 0;

    /* Start a new payload chunk */
    payloadHeader *header = (payloadHeader *)(buf + *used);
    header->payload_type = type;
    header->payload_len = len;
    memcpy((char *)header + sizeof(payloadHeader), payload, len);
    *used += sizeof(payloadHeader) + len;

    /* Track referenced reply bytes for copy avoidance. */
    if (type == BULK_STR_REF) {
        const bulkStrRef *str_ref = (const bulkStrRef *)payload;
        c->reply_bytes_shared += sdslen(str_ref->obj->ptr);
    }

    return 1;
}

/* Adds the payload to the reply linked list.
 * Note: some edits to this function need to be relayed to AddReplyFromClient. */
static void _addReplyPayloadToList(client *c, list *reply_list, const char *payload, size_t len, uint8_t payload_type) {
    listNode *ln = listLast(reply_list);
    clientReplyBlock *tail = ln ? listNodeValue(ln) : NULL;
    /* Determine if encoded buffer is required */
    int encoded = payload_type == BULK_STR_REF;

    /* Note that 'tail' may be NULL even if we have a tail node, because when
     * addReplyDeferredLen() is used, it sets a dummy node to NULL just
     * to fill it later, when the size of the bulk length is set. */

    /* Append to tail node when possible. */
    if (tail) {
        if (unlikely(tail->buf_encoded)) {
            /* Try to add to encoded buffer */
            if (tryAddPayload(c, tail->buf, &tail->used, tail->size, payload_type, (void *)payload, len)) {
                /* For BULK_STR_REF payloads, tryAddPayload updates shared reply bytes
                 * which accounts for referenced strings. */
                if (encoded) closeClientOnOutputBufferLimitReached(c, 1);
                return;
            }
        } else if (!encoded) {
            /* Both tail and new payload are non-encoded, can append directly */
            size_t avail = tail->size - tail->used;
            size_t copy = avail >= len ? len : avail;
            if (copy > 0) {
                memcpy(tail->buf + tail->used, payload, copy);
                tail->used += copy;
                payload += copy;
                len -= copy;
            }
        }
        /* else: tail is non-encoded but new payload needs encoding, can't append */
    }

    if (len) {
        /* Create a new node, make sure it is allocated to at
         * least PROTO_REPLY_CHUNK_BYTES */
        size_t usable_size;
        size_t required_size = encoded ? len + sizeof(payloadHeader) : len;
        size_t size = required_size < PROTO_REPLY_CHUNK_BYTES ? PROTO_REPLY_CHUNK_BYTES : required_size;
        tail = zmalloc_usable(size + sizeof(clientReplyBlock), &usable_size);
        /* take over the allocation's internal fragmentation */
        tail->size = usable_size - sizeof(clientReplyBlock);
        tail->used = 0;
        tail->buf_encoded = encoded;
        if (tail->buf_encoded) {
            serverAssert(tryAddPayload(c, tail->buf, &tail->used, tail->size, payload_type, (void *)payload, len));
        } else {
            tail->used = len;
            memcpy(tail->buf, payload, len);
        }
        listAddNodeTail(reply_list, tail);
        c->reply_bytes += tail->size;

        closeClientOnOutputBufferLimitReached(c, 1);
    }
}

/* The subscribe / unsubscribe command family has a push as a reply,
 * or in other words, it responds with a push (or several of them
 * depending on how many arguments it got), and has no reply. */
int cmdHasPushAsReply(struct redisCommand *cmd) {
    if (!cmd) return 0;
    return cmd->proc == subscribeCommand  || cmd->proc == unsubscribeCommand ||
           cmd->proc == psubscribeCommand || cmd->proc == punsubscribeCommand ||
           cmd->proc == ssubscribeCommand || cmd->proc == sunsubscribeCommand;
}

/* Attempts to add the reply to the static buffer in the client struct.
 * Returns the length of data that is added to the reply buffer. */
static size_t _addReplyPayloadToBuffer(client *c, const void *payload, size_t len, uint8_t payload_type) {
    /* If there already are entries in the reply list, we cannot
     * add anything more to the static buffer. */
    if (listLength(c->reply) > 0) return 0;

    size_t available = c->buf_usable_size - c->bufpos;
    size_t reply_len = min(available, len);
    if (c->buf_encoded) {
        if (!tryAddPayload(c, c->buf, &c->bufpos, c->buf_usable_size, payload_type, payload, len))
            return 0;
        reply_len = len;
    } else {
        memcpy(c->buf + c->bufpos, payload, reply_len);
        c->bufpos+=reply_len;
    }

    /* We update the buffer peak after appending the reply to the buffer */
    if (c->buf_peak < (size_t)c->bufpos) c->buf_peak = (size_t)c->bufpos;
    return reply_len;
}

/* Adds bulk string reference (i.e. pointer to object and pointer to string itself) to static buffer
 * Returns non-zero value if succeeded to add */
static size_t _addBulkStrRefToBuffer(client *c, const void *payload, size_t len) {
    size_t result;
    if (!c->buf_encoded) {
        /* If buffer is plain and not empty then can't add bulk string reference to it */
        if (c->bufpos) return 0;
        c->buf_encoded = 1; /* Set c->buf to encoded mode to allow bulk string reference to be stored in it */
        result = _addReplyPayloadToBuffer(c, payload, len, BULK_STR_REF);
        if (!result) {
            /* Failed to add bulk string reference to buffer, need to revert to plain mode. */
            c->buf_encoded = 0;
            return 0;
        }
    } else {
        result = _addReplyPayloadToBuffer(c, payload, len, BULK_STR_REF);
        if (!result) return 0;
    }

    /* Even though the bulk string is stored by reference and the underlying
     * memory is shared, we still account this shared memory towards this
     * client's output buffer usage, so we need to check the output buffer limits. */
    closeClientOnOutputBufferLimitReached(c, 1);
    return result;
}

void _addReplyToBufferOrList(client *c, const char *s, size_t len) {
    if (c->flags & CLIENT_CLOSE_AFTER_REPLY) return;

    /* Replicas should normally not cause any writes to the reply buffer. In case a rogue replica sent a command on the
     * replication link that caused a reply to be generated we'll simply disconnect it.
     * Note this is the simplest way to check a command added a response. Replication links are used to write data but
     * not for responses, so we should normally never get here on a replica client. */
    if (unlikely(clientTypeIsSlave(c))) {
        sds cmdname = c->lastcmd ? c->lastcmd->fullname : NULL;
        logInvalidUseAndFreeClientAsync(c, "Replica generated a reply to command '%s'",
                                        cmdname ? cmdname : "<unknown>");
        return;
    }

    c->net_output_bytes_curr_cmd += len;
    /* We call it here because this function may affect the reply
     * buffer offset (see function comment) */
    reqresSaveClientReplyOffset(c);

    /* If we're processing a push message into the current client (i.e. executing PUBLISH
     * to a channel which we are subscribed to, then we wanna postpone that message to be added
     * after the command's reply (specifically important during multi-exec). the exception is
     * the SUBSCRIBE command family, which (currently) have a push message instead of a proper reply.
     * The check for executing_client also avoids affecting push messages that are part of eviction.
     * Check CLIENT_PUSHING first to avoid race conditions, as it's absent in module's fake client. */
    if ((c->flags & CLIENT_PUSHING) && c == server.current_client &&
        server.executing_client && !cmdHasPushAsReply(server.executing_client->cmd))
    {
        _addReplyPayloadToList(c,server.pending_push_messages,s,len,PLAIN_REPLY);
        return;
    }

    size_t reply_len = _addReplyPayloadToBuffer(c, s, len, PLAIN_REPLY);
    if (len > reply_len)
        _addReplyPayloadToList(c, c->reply, s + reply_len, len - reply_len, PLAIN_REPLY);
}

/* Check if the client's pending_ref_reply_node is currently linked in the list.
 * A node is considered linked if it has neighbors (prev/next), or if it's the
 * only node in the list (head points to it). */
static inline int clientIsInPendingRefReplyList(client *c) {
    return listNextNode(&c->pending_ref_reply_node) != NULL ||
           listPrevNode(&c->pending_ref_reply_node) != NULL ||
           listFirst(server.clients_with_pending_ref_reply) == &c->pending_ref_reply_node;
}

/* Format "$<len>\r\n" into str_ref->prefix and set prefix_cnt. Called at write time. */
unsigned int formatBulkStrRefPrefix(bulkStrRef *str_ref) {
    if (str_ref->prefix_cnt != 0) return str_ref->prefix_cnt;
    str_ref->prefix[0] = '$';
    size_t num_len = ll2string(str_ref->prefix + 1, sizeof(str_ref->prefix) - 3, sdslen(str_ref->obj->ptr));
    str_ref->prefix[num_len + 1] = '\r';
    str_ref->prefix[num_len + 2] = '\n';
    str_ref->prefix_cnt = (unsigned int)(num_len + 3);
    return str_ref->prefix_cnt;
}

/* Increment reference to object and add pointer to object and
 * pointer to string itself to current reply buffer */
static void _addBulkStrRefToBufferOrList(client *c, robj *obj, size_t len) {
    if (c->flags & CLIENT_CLOSE_AFTER_REPLY) return;

    bulkStrRef str_ref;
    str_ref.obj = obj;
    incrRefCount(obj); /* Refcount will be decremented in write handler */
    str_ref.prefix_cnt = 0; /* filled at write time */
    str_ref.crlf[0] = '\r';
    str_ref.crlf[1] = '\n';

    /* Track output bytes: bulk string prefix + content + trailing CRLF */
    c->net_output_bytes_curr_cmd += digits10(len) + 3 + len + 2;

    /* We call it here because this function may affect the reply
     * buffer offset (see function comment) */
    reqresSaveClientReplyOffset(c);

    if (!_addBulkStrRefToBuffer(c, (void *)&str_ref, sizeof(str_ref))) {
        _addReplyPayloadToList(c, c->reply, (void *)&str_ref, sizeof(str_ref), BULK_STR_REF);
    }

    /* Track clients with pending referenced reply objects for async flushdb protection. */
    if (!clientIsInPendingRefReplyList(c)) {
        listLinkNodeTail(server.clients_with_pending_ref_reply, &c->pending_ref_reply_node);
    }
}

/* -----------------------------------------------------------------------------
 * Higher level functions to queue data on the client output buffer.
 * The following functions are the ones that commands implementations will call.
 * -------------------------------------------------------------------------- */

/* Add the object 'obj' string representation to the client output buffer. */
void addReply(client *c, robj *obj) {
    if (_prepareClientToWrite(c) != C_OK) return;

    if (sdsEncodedObject(obj)) {
        _addReplyToBufferOrList(c,obj->ptr,sdslen(obj->ptr));
    } else if (obj->encoding == OBJ_ENCODING_INT) {
        /* For integer encoded strings we just convert it into a string
         * using our optimized function, and attach the resulting string
         * to the output buffer. */
        char buf[32];
        size_t len = ll2string(buf,sizeof(buf),(long)obj->ptr);
        _addReplyToBufferOrList(c,buf,len);
    } else {
        serverPanic("Wrong obj->encoding in addReply()");
    }
}

/* Add the SDS 's' string to the client output buffer, as a side effect
 * the SDS string is freed. */
void addReplySds(client *c, sds s) {
    if (_prepareClientToWrite(c) != C_OK) {
        /* The caller expects the sds to be free'd. */
        sdsfree(s);
        return;
    }
    _addReplyToBufferOrList(c,s,sdslen(s));
    sdsfree(s);
}

/* This low level function just adds whatever protocol you send it to the
 * client buffer, trying the static buffer initially, and using the string
 * of objects if not possible.
 *
 * It is efficient because does not create an SDS object nor an Redis object
 * if not needed. The object will only be created by calling
 * _addReplyProtoToList() if we fail to extend the existing tail object
 * in the list of objects. */
void addReplyProto(client *c, const char *s, size_t len) {
    if (_prepareClientToWrite(c) != C_OK) return;
    _addReplyToBufferOrList(c,s,len);
}

/* Low level function called by the addReplyError...() functions.
 * It emits the protocol for a Redis error, in the form:
 *
 * -ERRORCODE Error Message<CR><LF>
 *
 * If the error code is already passed in the string 's', the error
 * code provided is used, otherwise the string "-ERR " for the generic
 * error code is automatically added.
 * Note that 's' must NOT end with \r\n. */
void addReplyErrorLength(client *c, const char *s, size_t len) {
    /* If the string already starts with "-..." then the error code
     * is provided by the caller. Otherwise we use "-ERR". */
    if (!len || s[0] != '-') addReplyProto(c,"-ERR ",5);
    addReplyProto(c,s,len);
    addReplyProto(c,"\r\n",2);
}

/* Do some actions after an error reply was sent (Log if needed, updates stats, etc.)
 * Possible flags:
 * * ERR_REPLY_FLAG_NO_STATS_UPDATE - indicate not to update any error stats. */
void afterErrorReply(client *c, const char *s, size_t len, int flags) {
    /* Module clients fall into two categories:
     * Calls to RM_Call, in which case the error isn't being returned to a client, so should not be counted.
     * Module thread safe context calls to RM_ReplyWithError, which will be added to a real client by the main thread later. */
    if (c->flags & CLIENT_MODULE) {
        if (!c->deferred_reply_errors) {
            c->deferred_reply_errors = listCreate();
            listSetFreeMethod(c->deferred_reply_errors, sdsfreegeneric);
        }
        listAddNodeTail(c->deferred_reply_errors, sdsnewlen(s, len));
        return;
    }

    if (!(flags & ERR_REPLY_FLAG_NO_STATS_UPDATE)) {
        /* Increment the global error counter */
        server.stat_total_error_replies++;
        /* Increment the error stats
         * If the string already starts with "-..." then the error prefix
         * is provided by the caller ( we limit the search to 32 chars). Otherwise we use "-ERR". */
        if (s[0] != '-') {
            incrementErrorCount("ERR", 3);
        } else {
            const char *spaceloc = memchr(s, ' ', len < 32 ? len : 32);
            if (spaceloc) {
                const size_t errEndPos = (size_t)(spaceloc - s);
                incrementErrorCount(s+1, errEndPos-1);
            } else {
                /* Fallback to ERR if we can't retrieve the error prefix */
                incrementErrorCount("ERR", 3);
            }
        }
    } else {
        /* stat_total_error_replies will not be updated, which means that
         * the cmd stats will not be updated as well, we still want this command
         * to be counted as failed so we update it here. We update c->realcmd in
         * case c->cmd was changed (like in GEOADD). */
        c->realcmd->failed_calls++;
    }

    /* Sometimes it could be normal that a slave replies to a master with
     * an error and this function gets called. Actually the error will never
     * be sent because addReply*() against master clients has no effect...
     * A notable example is:
     *
     *    EVAL 'redis.call("incr",KEYS[1]); redis.call("nonexisting")' 1 x
     *
     * Where the master must propagate the first change even if the second
     * will produce an error. However it is useful to log such events since
     * they are rare and may hint at errors in a script or a bug in Redis. */
    int ctype = getClientType(c);
    if (ctype == CLIENT_TYPE_MASTER || ctype == CLIENT_TYPE_SLAVE || c->id == CLIENT_ID_AOF) {
        char *to, *from;

        if (c->id == CLIENT_ID_AOF) {
            to = "AOF-loading-client";
            from = "server";
        } else if (ctype == CLIENT_TYPE_MASTER) {
            if (c->flags & CLIENT_ASM_IMPORTING) {
                to = "source";
                from = "destination";
            } else {
                to = "master";
                from = "replica";
            }
        } else {
            to = "replica";
            from = "master";
        }

        if (len > 4096) len = 4096;
        sds cmdname = c->lastcmd ? c->lastcmd->fullname : NULL;
        serverLog(LL_WARNING,"== CRITICAL == This %s is sending an error "
                             "to its %s: '%.*s' after processing the command "
                             "'%s'", from, to, (int)len, s, cmdname ? cmdname : "<unknown>");
        if (ctype == CLIENT_TYPE_MASTER && server.repl_backlog &&
            !(c->flags & CLIENT_ASM_IMPORTING) && server.repl_backlog->histlen > 0)
        {
            showLatestBacklog();
        }
        server.stat_unexpected_error_replies++;

        /* Based off the propagation error behavior, check if we need to panic here. There
         * are currently two checked cases:
         * * If this command was from our master and we are not a writable replica.
         * * We are reading from an AOF file. */
        int panic_in_replicas = (ctype == CLIENT_TYPE_MASTER && server.repl_slave_ro)
            && (server.propagation_error_behavior == PROPAGATION_ERR_BEHAVIOR_PANIC ||
            server.propagation_error_behavior == PROPAGATION_ERR_BEHAVIOR_PANIC_ON_REPLICAS);
        int panic_in_aof = c->id == CLIENT_ID_AOF 
            && server.propagation_error_behavior == PROPAGATION_ERR_BEHAVIOR_PANIC;
        if (panic_in_replicas || panic_in_aof) {
            serverPanic("This %s panicked sending an error to its %s"
                " after processing the command '%s'",
                from, to, cmdname ? cmdname : "<unknown>");
        }
    }
}

/* The 'err' object is expected to start with -ERRORCODE and end with \r\n.
 * Unlike addReplyErrorSds and others alike which rely on addReplyErrorLength. */
void addReplyErrorObject(client *c, robj *err) {
    addReply(c, err);
    afterErrorReply(c, err->ptr, sdslen(err->ptr)-2, 0); /* Ignore trailing \r\n */
}

/* Sends either a reply or an error reply by checking the first char.
 * If the first char is '-' the reply is considered an error.
 * In any case the given reply is sent, if the reply is also recognize
 * as an error we also perform some post reply operations such as
 * logging and stats update. */
void addReplyOrErrorObject(client *c, robj *reply) {
    serverAssert(sdsEncodedObject(reply));
    sds rep = reply->ptr;
    if (sdslen(rep) > 1 && rep[0] == '-') {
        addReplyErrorObject(c, reply);
    } else {
        addReply(c, reply);
    }
}

/* See addReplyErrorLength for expectations from the input string. */
void addReplyError(client *c, const char *err) {
    addReplyErrorLength(c,err,strlen(err));
    afterErrorReply(c,err,strlen(err),0);
}

/* Add error reply to the given client.
 * Supported flags:
 * * ERR_REPLY_FLAG_NO_STATS_UPDATE - indicate not to perform any error stats updates */
void addReplyErrorSdsEx(client *c, sds err, int flags) {
    addReplyErrorLength(c,err,sdslen(err));
    afterErrorReply(c,err,sdslen(err),flags);
    sdsfree(err);
}

/* See addReplyErrorLength for expectations from the input string.
 * The string is safe to contain \r and \n anywhere. */
void addReplyErrorSdsExSafe(client *c, sds err, int flags) {
    err = sdsmapchars(err, "\r\n", "  ",  2);
    addReplyErrorSdsEx(c, err, flags);
}

/* See addReplyErrorLength for expectations from the input string. */
/* As a side effect the SDS string is freed. */
void addReplyErrorSds(client *c, sds err) {
    addReplyErrorSdsEx(c, err, 0);
}

/* See addReplyErrorLength for expectations from the input string. */
/* As a side effect the SDS string is freed. */
void addReplyErrorSdsSafe(client *c, sds err) {
    err = sdsmapchars(err, "\r\n", "  ",  2);
    addReplyErrorSdsEx(c, err, 0);
}

/* Internal function used by addReplyErrorFormat, addReplyErrorFormatEx and RM_ReplyWithErrorFormat.
 * Refer to afterErrorReply for more information about the flags. */
void addReplyErrorFormatInternal(client *c, int flags, const char *fmt, va_list ap) {
    va_list cpy;
    va_copy(cpy,ap);
    sds s = sdscatvprintf(sdsempty(),fmt,cpy);
    va_end(cpy);
    /* Trim any newlines at the end (ones will be added by addReplyErrorLength) */
    s = sdstrim(s, "\r\n");
    /* Make sure there are no newlines in the middle of the string, otherwise
     * invalid protocol is emitted. */
    s = sdsmapchars(s, "\r\n", "  ",  2);
    addReplyErrorLength(c,s,sdslen(s));
    afterErrorReply(c,s,sdslen(s),flags);
    sdsfree(s);
}

void addReplyErrorFormatEx(client *c, int flags, const char *fmt, ...) {
    va_list ap;
    va_start(ap,fmt);
    addReplyErrorFormatInternal(c, flags, fmt, ap);
    va_end(ap);
}

/* See addReplyErrorLength for expectations from the formatted string.
 * The formatted string is safe to contain \r and \n anywhere. */
void addReplyErrorFormat(client *c, const char *fmt, ...) {
    va_list ap;
    va_start(ap,fmt);
    addReplyErrorFormatInternal(c, 0, fmt, ap);
    va_end(ap);
}

void addReplyErrorArity(client *c) {
    addReplyErrorFormat(c, "wrong number of arguments for '%s' command",
                        c->cmd->fullname);
}

void addReplyErrorExpireTime(client *c) {
    addReplyErrorFormat(c, "invalid expire time in '%s' command",
                        c->cmd->fullname);
}

void addReplyStatusLength(client *c, const char *s, size_t len) {
    addReplyProto(c,"+",1);
    addReplyProto(c,s,len);
    addReplyProto(c,"\r\n",2);
}

void addReplyStatus(client *c, const char *status) {
    addReplyStatusLength(c,status,strlen(status));
}

/* Like addReplyStatus() but the string is safe to contain \r and \n anywhere. */
void addReplyStatusSafe(client *c, const char *status) {
    sds s = sdsmapchars(sdsnew(status), "\r\n", "  ",  2);
    addReplyStatusLength(c,s,sdslen(s));
    sdsfree(s);
}

void addReplyStatusFormat(client *c, const char *fmt, ...) {
    va_list ap;
    va_start(ap,fmt);
    sds s = sdscatvprintf(sdsempty(),fmt,ap);
    va_end(ap);
    /* Trim any newlines at the end (ones will be added by addReplyStatusLength) */
    s = sdstrim(s, "\r\n");
    /* Make sure there are no newlines in the middle of the string, otherwise
     * invalid protocol is emitted. */
    s = sdsmapchars(s, "\r\n", "  ",  2);
    addReplyStatusLength(c,s,sdslen(s));
    sdsfree(s);
}

/* Sometimes we are forced to create a new reply node, and we can't append to
 * the previous one, when that happens, we wanna try to trim the unused space
 * at the end of the last reply node which we won't use anymore. */
void trimReplyUnusedTailSpace(client *c) {
    listNode *ln = listLast(c->reply);
    clientReplyBlock *tail = ln? listNodeValue(ln): NULL;

    /* Note that 'tail' may be NULL even if we have a tail node, because when
     * addReplyDeferredLen() is used */
    if (!tail) return;

    /* We only try to trim the space is relatively high (more than a 1/4 of the
     * allocation), otherwise there's a high chance realloc will NOP.
     * Also, to avoid large memmove which happens as part of realloc, we only do
     * that if the used part is small.  */
    if (tail->size - tail->used > tail->size / 4 &&
        tail->used < PROTO_REPLY_CHUNK_BYTES && !tail->buf_encoded)
    {
        size_t usable_size;
        size_t old_size = tail->size;
        tail = zrealloc_usable(tail, tail->used + sizeof(clientReplyBlock), &usable_size, NULL);
        /* take over the allocation's internal fragmentation (at least for
         * memory usage tracking) */
        tail->size = usable_size - sizeof(clientReplyBlock);
        c->reply_bytes = c->reply_bytes + tail->size - old_size;
        listNodeValue(ln) = tail;
    }
}

/* 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 *addReplyDeferredLen(client *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 setDeferredAggregateLen() will be called. */
    if (_prepareClientToWrite(c) != C_OK) return NULL;

    /* Replicas should normally not cause any writes to the reply buffer. In case a rogue replica sent a command on the
     * replication link that caused a reply to be generated we'll simply disconnect it.
     * Note this is the simplest way to check a command added a response. Replication links are used to write data but
     * not for responses, so we should normally never get here on a replica client. */
    if (unlikely(clientTypeIsSlave(c))) {
        sds cmdname = c->lastcmd ? c->lastcmd->fullname : NULL;
        logInvalidUseAndFreeClientAsync(c, "Replica generated a reply to command '%s'",
                                        cmdname ? cmdname : "<unknown>");
        return NULL;
    }

    /* We call it here because this function conceptually affects the reply
     * buffer offset (see function comment) */
    reqresSaveClientReplyOffset(c);

    trimReplyUnusedTailSpace(c);
    listAddNodeTail(c->reply,NULL); /* NULL is our placeholder. */
    return listLast(c->reply);
}

void setDeferredReply(client *c, void *node, const char *s, size_t length) {
    listNode *ln = (listNode*)node;
    clientReplyBlock *next, *prev;

    /* Abort when *node is NULL: when the client should not accept writes
     * we return NULL in addReplyDeferredLen() */
    if (node == NULL) return;
    serverAssert(!listNodeValue(ln));

    /* Normally we fill this dummy NULL node, added by addReplyDeferredLen(),
     * with a new buffer structure containing the protocol needed to specify
     * the length of the array following. However sometimes there might be room
     * in the previous/next node so we can instead remove this NULL node, and
     * suffix/prefix our data in the node immediately before/after it, in order
     * to save a write(2) syscall later. Conditions needed to do it:
     *
     * - The prev node is non-NULL and has space in it or
     * - The next node is non-NULL,
     * - It has enough room already allocated
     * - And not too large (avoid large memmove) */
    if (ln->prev != NULL && (prev = listNodeValue(ln->prev)) &&
        prev->used < prev->size && !prev->buf_encoded)
    {
        size_t len_to_copy = prev->size - prev->used;
        if (len_to_copy > length)
            len_to_copy = length;
        memcpy(prev->buf + prev->used, s, len_to_copy);
        c->net_output_bytes_curr_cmd += len_to_copy;
        prev->used += len_to_copy;
        length -= len_to_copy;
        if (length == 0) {
            listDelNode(c->reply, ln);
            return;
        }
        s += len_to_copy;
    }

    if (ln->next != NULL && (next = listNodeValue(ln->next)) &&
        next->size - next->used >= length &&
        next->used < PROTO_REPLY_CHUNK_BYTES * 4 &&
        !next->buf_encoded)
    {
        memmove(next->buf + length, next->buf, next->used);
        memcpy(next->buf, s, length);
        c->net_output_bytes_curr_cmd += length;
        next->used += length;
        listDelNode(c->reply,ln);
    } else {
        /* Create a new node */
        size_t usable_size;
        clientReplyBlock *buf = zmalloc_usable(length + sizeof(clientReplyBlock), &usable_size);
        /* Take over the allocation's internal fragmentation */
        buf->size = usable_size - sizeof(clientReplyBlock);
        buf->used = length;
        buf->buf_encoded = 0;
        memcpy(buf->buf, s, length);
        c->net_output_bytes_curr_cmd += length;
        listNodeValue(ln) = buf;
        c->reply_bytes += buf->size;

        closeClientOnOutputBufferLimitReached(c, 1);
    }
}

/* Populate the length object and try gluing it to the next chunk. */
void setDeferredAggregateLen(client *c, void *node, long length, char prefix) {
    serverAssert(length >= 0);

    /* Abort when *node is NULL: when the client should not accept writes
     * we return NULL in addReplyDeferredLen() */
    if (node == NULL) return;

    /* Things like *2\r\n, %3\r\n or ~4\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. */
    const size_t hdr_len = OBJ_SHARED_HDR_STRLEN(length);
    const int opt_hdr = length < OBJ_SHARED_BULKHDR_LEN;
    if (prefix == '*' && opt_hdr) {
        setDeferredReply(c, node, shared.mbulkhdr[length]->ptr, hdr_len);
        return;
    }
    if (prefix == '%' && opt_hdr) {
        setDeferredReply(c, node, shared.maphdr[length]->ptr, hdr_len);
        return;
    }
    if (prefix == '~' && opt_hdr) {
        setDeferredReply(c, node, shared.sethdr[length]->ptr, hdr_len);
        return;
    }

    char lenstr[128];
    size_t lenstr_len = snprintf(lenstr, sizeof(lenstr), "%c%ld\r\n", prefix, length);
    setDeferredReply(c, node, lenstr, lenstr_len);
}

void setDeferredArrayLen(client *c, void *node, long length) {
    setDeferredAggregateLen(c,node,length,'*');
}

void setDeferredMapLen(client *c, void *node, long length) {
    int prefix = c->resp == 2 ? '*' : '%';
    if (c->resp == 2) length *= 2;
    setDeferredAggregateLen(c,node,length,prefix);
}

void setDeferredSetLen(client *c, void *node, long length) {
    int prefix = c->resp == 2 ? '*' : '~';
    setDeferredAggregateLen(c,node,length,prefix);
}

void setDeferredAttributeLen(client *c, void *node, long length) {
    serverAssert(c->resp >= 3);
    setDeferredAggregateLen(c,node,length,'|');
}

void setDeferredPushLen(client *c, void *node, long length) {
    serverAssert(c->resp >= 3);
    setDeferredAggregateLen(c,node,length,'>');
}

/* Add a double as a bulk reply */
void addReplyDouble(client *c, double d) {
    if (c->resp == 3) {
        char dbuf[MAX_D2STRING_CHARS+3];
        dbuf[0] = ',';
        const int dlen = d2string(dbuf+1,sizeof(dbuf)-1,d);
        dbuf[dlen+1] = '\r';
        dbuf[dlen+2] = '\n';
        dbuf[dlen+3] = '\0';
        addReplyProto(c,dbuf,dlen+3);
    } else {
        char dbuf[MAX_LONG_DOUBLE_CHARS+32];
        /* In order to prepend the string length before the formatted number,
         * but still avoid an extra memcpy of the whole number, we reserve space
         * for maximum header `$0000\r\n`, print double, add the resp header in
         * front of it, and then send the buffer with the right `start` offset. */
        const int dlen = d2string(dbuf+7,sizeof(dbuf)-7,d);
        int digits = digits10(dlen);
        int start = 4 - digits;
        serverAssert(start >= 0);
        dbuf[start] = '$';

        /* Convert `dlen` to string, putting it's digits after '$' and before the
            * formatted double string. */
        for(int i = digits, val = dlen; val && i > 0 ; --i, val /= 10) {
            dbuf[start + i] = "0123456789"[val % 10];
        }
        dbuf[5] = '\r';
        dbuf[6] = '\n';
        dbuf[dlen+7] = '\r';
        dbuf[dlen+8] = '\n';
        dbuf[dlen+9] = '\0';
        addReplyProto(c,dbuf+start,dlen+9-start);
    }
}

void addReplyBigNum(client *c, const char* num, size_t len) {
    if (c->resp == 2) {
        addReplyBulkCBuffer(c, num, len);
    } else {
        addReplyProto(c,"(",1);
        addReplyProto(c,num,len);
        addReplyProto(c,"\r\n",2);
    }
}

/* Add a long double as a bulk reply, but uses a human readable formatting
 * of the double instead of exposing the crude behavior of doubles to the
 * dear user. */
void addReplyHumanLongDouble(client *c, long double d) {
    if (c->resp == 2) {
        robj *o = createStringObjectFromLongDouble(d,1);
        addReplyBulk(c,o);
        decrRefCount(o);
    } else {
        char buf[MAX_LONG_DOUBLE_CHARS];
        int len = ld2string(buf,sizeof(buf),d,LD_STR_HUMAN);
        addReplyProto(c,",",1);
        addReplyProto(c,buf,len);
        addReplyProto(c,"\r\n",2);
    }
}

static inline void _addReplyLongLongSharedHdr(client *c, long long ll, char prefix,
                                              robj *shared_hdr[OBJ_SHARED_BULKHDR_LEN])
{
    char buf[128];
    int len;
    const int opt_hdr = ll < OBJ_SHARED_BULKHDR_LEN && ll >= 0;

    if (opt_hdr) {
        _addReplyToBufferOrList(c, shared_hdr[ll]->ptr, OBJ_SHARED_HDR_STRLEN(ll));
        return;
    }

    buf[0] = prefix;
    len = ll2string(buf + 1, sizeof(buf) - 1, ll);
    buf[len + 1] = '\r';
    buf[len + 2] = '\n';
    _addReplyToBufferOrList(c, buf, len + 3);
}

static inline void _addReplyLongLongBulk(client *c, long long ll) {
    _addReplyLongLongSharedHdr(c, ll, '$', shared.bulkhdr);
}

static inline void _addReplyLongLongMBulk(client *c, long long ll) {
    _addReplyLongLongSharedHdr(c, ll, '*', shared.mbulkhdr);
}

/* Add a long long as integer reply or bulk len / multi bulk count.
 * Basically this is used to output <prefix><long long><crlf>. */
static void _addReplyLongLongWithPrefix(client *c, long long ll, char prefix) {
    char buf[128];
    int len;

    /* Things like $3\r\n or *2\r\n are emitted very often by the protocol
     * so we have a few shared objects to use if the integer is small
     * like it is most of the times. */
    const int opt_hdr = ll < OBJ_SHARED_BULKHDR_LEN && ll >= 0;
    const size_t hdr_len = OBJ_SHARED_HDR_STRLEN(ll);
    if (prefix == '*' && opt_hdr) {
        _addReplyToBufferOrList(c, shared.mbulkhdr[ll]->ptr, hdr_len);
        return;
    } else if (prefix == '$' && opt_hdr) {
        _addReplyToBufferOrList(c, shared.bulkhdr[ll]->ptr, hdr_len);
        return;
    } else if (prefix == '%' && opt_hdr) {
        _addReplyToBufferOrList(c, shared.maphdr[ll]->ptr, hdr_len);
        return;
    } else if (prefix == '~' && opt_hdr) {
        _addReplyToBufferOrList(c, shared.sethdr[ll]->ptr, hdr_len);
        return;
    }

    buf[0] = prefix;
    len = ll2string(buf + 1, sizeof(buf) - 1, ll);
    buf[len + 1] = '\r';
    buf[len + 2] = '\n';
    _addReplyToBufferOrList(c, buf, len + 3);
}

void addReplyLongLong(client *c, long long ll) {
    if (ll == 0)
        addReply(c,shared.czero);
    else if (ll == 1)
        addReply(c, shared.cone);
    else {
        if (_prepareClientToWrite(c) != C_OK) return;
        _addReplyLongLongWithPrefix(c, ll, ':');
    }
}

void addReplyLongLongFromStr(client *c, robj *str) {
    addReplyProto(c,":",1);
    addReply(c,str);
    addReplyProto(c,"\r\n",2);
}

/* Reply with unsigned 64-bit value. Uses integer reply when value fits in
 * signed long long, otherwise big number (RESP3) or bulk string (RESP2). */
void addReplyUnsignedLongLong(client *c, uint64_t v) {
    if (v <= (uint64_t)LLONG_MAX) {
        addReplyLongLong(c, (long long)v);
    } else {
        char buf[LONG_STR_SIZE];
        int len = ull2string(buf, sizeof(buf), v);
        addReplyBigNum(c, buf, len);
    }
}

void addReplyAggregateLen(client *c, long length, int prefix) {
    serverAssert(length >= 0);
    if (_prepareClientToWrite(c) != C_OK) return;
    _addReplyLongLongWithPrefix(c, length, prefix);
}

void addReplyArrayLen(client *c, long length) {
    serverAssert(length >= 0);
    if (_prepareClientToWrite(c) != C_OK) return;
    _addReplyLongLongMBulk(c, length);
}

void addReplyMapLen(client *c, long length) {
    int prefix = c->resp == 2 ? '*' : '%';
    if (c->resp == 2) length *= 2;
    addReplyAggregateLen(c,length,prefix);
}

void addReplySetLen(client *c, long length) {
    int prefix = c->resp == 2 ? '*' : '~';
    addReplyAggregateLen(c,length,prefix);
}

void addReplyAttributeLen(client *c, long length) {
    serverAssert(c->resp >= 3);
    addReplyAggregateLen(c,length,'|');
}

void addReplyPushLen(client *c, long length) {
    serverAssert(c->resp >= 3);
    serverAssertWithInfo(c, NULL, c->flags & CLIENT_PUSHING);
    addReplyAggregateLen(c,length,'>');
}

void addReplyNull(client *c) {
    if (c->resp == 2) {
        addReplyProto(c,"$-1\r\n",5);
    } else {
        addReplyProto(c,"_\r\n",3);
    }
}

void addReplyBool(client *c, int b) {
    if (c->resp == 2) {
        addReply(c, b ? shared.cone : shared.czero);
    } else {
        addReplyProto(c, b ? "#t\r\n" : "#f\r\n",4);
    }
}

/* A null array is a concept that no longer exists in RESP3. However
 * RESP2 had it, so API-wise we have this call, that will emit the correct
 * RESP2 protocol, however for RESP3 the reply will always be just the
 * Null type "_\r\n". */
void addReplyNullArray(client *c) {
    if (c->resp == 2) {
        addReplyProto(c,"*-1\r\n",5);
    } else {
        addReplyProto(c,"_\r\n",3);
    }
}

/* Create the length prefix of a bulk reply, example: $2234 */
void addReplyBulkLen(client *c, robj *obj) {
    size_t len = stringObjectLen(obj);
    if (_prepareClientToWrite(c) != C_OK) return;
    _addReplyLongLongBulk(c, len);
}

/* Check if copy avoidance is preferred for this client and object.
 * Copy avoidance allows I/O threads to directly reference obj->ptr
 * instead of copying data to reply buffers. */
static int isCopyAvoidPreferred(client *c, robj *obj, size_t len) {
    /* Don't use copy avoidance for fake clients. */
    if (!c->conn || !server.reply_copy_avoidance_enabled) return 0;

    int type = getClientType(c);
    if (type != CLIENT_TYPE_NORMAL) return 0;

    /* Don't use copy avoidance for push messages. Push messages need to be deferred
     * to server.pending_push_messages when CLIENT_PUSHING is set. */
    if (c->flags & CLIENT_PUSHING) return 0;

    if (obj->encoding != OBJ_ENCODING_RAW || obj->refcount >= OBJ_FIRST_SPECIAL_REFCOUNT) return 0;

    /* Copy avoidance is preferred for any string size starting certain number of I/O threads  */
    if (server.io_threads_num >= COPY_AVOID_MIN_IO_THREADS) return 1;

    /* Main thread only. No I/O threads */
    if (server.io_threads_num == 1) {
        /* Copy avoidance is preferred starting certain string size */
        return len >= COPY_AVOID_MIN_STRING_SIZE;
    }

    /* Main thread + I/O threads */
    return len >= COPY_AVOID_MIN_STRING_SIZE_THREADED;
}

/* Try to avoid whole bulk string copy to a reply buffer
 * If copy avoidance allowed then only pointer to object and string will be copied to the buffer */
static int tryAvoidBulkStrCopyToReply(client *c, robj *obj, size_t len) {
    if (!isCopyAvoidPreferred(c, obj, len)) return C_ERR;
    _addBulkStrRefToBufferOrList(c, obj, len);
    return C_OK;
}

/* Add a Redis Object as a bulk reply.
 * If avoid_copy is non-zero, attempt to use copy avoidance optimization. */
void addReplyBulkWithFlag(client *c, robj *obj, int avoid_copy) {
    if (_prepareClientToWrite(c) != C_OK) return;

    if (sdsEncodedObject(obj)) {
        const size_t len = sdslen(obj->ptr);
        if (avoid_copy && tryAvoidBulkStrCopyToReply(c, obj, len) == C_OK)
            return;
        _addReplyLongLongBulk(c, len);
        _addReplyToBufferOrList(c,obj->ptr,len);
        _addReplyToBufferOrList(c,"\r\n",2);
    } else if (obj->encoding == OBJ_ENCODING_INT) {
        /* For integer encoded strings we just convert it into a string
         * using our optimized function, and attach the resulting string
         * to the output buffer. */
        char buf[34];
        size_t len = ll2string(buf,sizeof(buf),(long)obj->ptr);
        buf[len] = '\r';
        buf[len+1] = '\n';
        _addReplyLongLongBulk(c, len);
        _addReplyToBufferOrList(c,buf,len+2);
    } else {
        serverPanic("Wrong obj->encoding in addReply()");
    }
}

/* Add a Redis Object as a bulk reply */
void addReplyBulk(client *c, robj *obj) {
    addReplyBulkWithFlag(c, obj, 1);
}

/* Add a C buffer as bulk reply */
void addReplyBulkCBuffer(client *c, const void *p, size_t len) {
    if (_prepareClientToWrite(c) != C_OK) return;
    _addReplyLongLongBulk(c, len);
    _addReplyToBufferOrList(c, p, len);
    _addReplyToBufferOrList(c, "\r\n", 2);
}

/* Add sds to reply (takes ownership of sds and frees it) */
void addReplyBulkSds(client *c, sds s) {
    if (_prepareClientToWrite(c) != C_OK) {
        sdsfree(s);
        return;
    }
    _addReplyLongLongWithPrefix(c, sdslen(s), '$');
    _addReplyToBufferOrList(c, s, sdslen(s));
    sdsfree(s);
    _addReplyToBufferOrList(c, "\r\n", 2);
}

/* Set sds to a deferred reply (for symmetry with addReplyBulkSds it also frees the sds) */
void setDeferredReplyBulkSds(client *c, void *node, sds s) {
    sds reply = sdscatprintf(sdsempty(), "$%d\r\n%s\r\n", (unsigned)sdslen(s), s);
    setDeferredReply(c, node, reply, sdslen(reply));
    sdsfree(reply);
    sdsfree(s);
}

/* Add a C null term string as bulk reply */
void addReplyBulkCString(client *c, const char *s) {
    if (s == NULL) {
        addReplyNull(c);
    } else {
        addReplyBulkCBuffer(c,s,strlen(s));
    }
}

/* Add a long long as a bulk reply */
void addReplyBulkLongLong(client *c, long long ll) {
    char buf[64];
    int len;

    len = ll2string(buf,64,ll);
    addReplyBulkCBuffer(c,buf,len);
}

/* Reply with a verbatim type having the specified extension.
 *
 * The 'ext' is the "extension" of the file, actually just a three
 * character type that describes the format of the verbatim string.
 * For instance "txt" means it should be interpreted as a text only
 * file by the receiver, "md " as markdown, and so forth. Only the
 * three first characters of the extension are used, and if the
 * provided one is shorter than that, the remaining is filled with
 * spaces. */
void addReplyVerbatim(client *c, const char *s, size_t len, const char *ext) {
    if (c->resp == 2) {
        addReplyBulkCBuffer(c,s,len);
    } else {
        char buf[32];
        size_t preflen = snprintf(buf,sizeof(buf),"=%zu\r\nxxx:",len+4);
        char *p = buf+preflen-4;
        for (int i = 0; i < 3; i++) {
            if (*ext == '\0') {
                p[i] = ' ';
            } else {
                p[i] = *ext++;
            }
        }
        addReplyProto(c,buf,preflen);
        addReplyProto(c,s,len);
        addReplyProto(c,"\r\n",2);
    }
}

/* This function is similar to the addReplyHelp function but adds the
 * ability to pass in two arrays of strings. Some commands have
 * some additional subcommands based on the specific feature implementation
 * Redis is compiled with (currently just clustering). This function allows
 * to pass is the common subcommands in `help` and any implementation
 * specific subcommands in `extended_help`.
 */
void addExtendedReplyHelp(client *c, const char **help, const char **extended_help) {
    sds cmd = sdsnew((char*) c->argv[0]->ptr);
    void *blenp = addReplyDeferredLen(c);
    int blen = 0;
    int idx = 0;

    sdstoupper(cmd);
    addReplyStatusFormat(c,
        "%s <subcommand> [<arg> [value] [opt] ...]. Subcommands are:",cmd);
    sdsfree(cmd);

    while (help[blen]) addReplyStatus(c,help[blen++]);
    if (extended_help) {
        while (extended_help[idx]) addReplyStatus(c,extended_help[idx++]);
    }
    blen += idx;

    addReplyStatus(c,"HELP");
    addReplyStatus(c,"    Print this help.");

    blen += 1;  /* Account for the header. */
    blen += 2;  /* Account for the footer. */
    setDeferredArrayLen(c,blenp,blen);
}

/* Add an array of C strings as status replies with a heading.
 * This function is typically invoked by commands that support
 * subcommands in response to the 'help' subcommand. The help array
 * is terminated by NULL sentinel. */
void addReplyHelp(client *c, const char **help) {
    addExtendedReplyHelp(c, help, NULL);
}

/* Add a suggestive error reply.
 * This function is typically invoked by from commands that support
 * subcommands in response to an unknown subcommand or argument error. */
void addReplySubcommandSyntaxError(client *c) {
    sds cmd = sdsnew((char*) c->argv[0]->ptr);
    sdstoupper(cmd);
    addReplyErrorFormat(c,
        "unknown subcommand or wrong number of arguments for '%.128s'. Try %s HELP.",
        (char*)c->argv[1]->ptr,cmd);
    sdsfree(cmd);
}

/* Append 'src' client output buffers into 'dst' client output buffers.
 * This function clears the output buffers of 'src' */
void AddReplyFromClient(client *dst, client *src) {
    /* If the source client contains a partial response due to client output
     * buffer limits, propagate that to the dest rather than copy a partial
     * reply. We don't wanna run the risk of copying partial response in case
     * for some reason the output limits don't reach the same decision (maybe
     * they changed) */
    if (src->flags & CLIENT_CLOSE_ASAP) {
        sds client = catClientInfoString(sdsempty(),dst);
        freeClientAsync(dst);
        serverLog(LL_WARNING,"Client %s scheduled to be closed ASAP for overcoming of output buffer limits.", client);
        sdsfree(client);
        return;
    }

    /* First add the static buffer (either into the static buffer or reply list) */
    addReplyProto(dst,src->buf, src->bufpos);

    /* We need to check with _prepareClientToWrite again (after addReplyProto)
     * since addReplyProto may have changed something (like CLIENT_CLOSE_ASAP) */
    if (_prepareClientToWrite(dst) != C_OK)
        return;

    /* We're bypassing _addReplyProtoToList, so we need to add the pre/post
     * checks in it. */
    if (dst->flags & CLIENT_CLOSE_AFTER_REPLY) return;

    /* Concatenate the reply list into the dest */
    if (listLength(src->reply))
        listJoin(dst->reply,src->reply);
    serverAssert(src->reply_bytes_shared == 0); /* It is non-normal client, never has references. */
    dst->reply_bytes += src->reply_bytes;
    src->reply_bytes = 0;
    src->bufpos = 0;

    if (src->deferred_reply_errors) {
        deferredAfterErrorReply(dst, src->deferred_reply_errors);
        listRelease(src->deferred_reply_errors);
        src->deferred_reply_errors = NULL;
    }

    /* Check output buffer limits */
    closeClientOnOutputBufferLimitReached(dst, 1);
}

/* Append the listed errors to the server error statistics. the input
 * list is not modified and remains the responsibility of the caller. */
void deferredAfterErrorReply(client *c, list *errors) {
    listIter li;
    listNode *ln;
    listRewind(errors,&li);
    while((ln = listNext(&li))) {
        sds err = ln->value;
        afterErrorReply(c, err, sdslen(err), 0);
    }
}

/* Logically copy 'src' replica client buffers info to 'dst' replica.
 * Basically increase referenced buffer block node reference count. */
void copyReplicaOutputBuffer(client *dst, client *src) {
    serverAssert(src->bufpos == 0 && listLength(src->reply) == 0); 
    serverAssert(src->running_tid == IOTHREAD_MAIN_THREAD_ID &&
                 dst->running_tid == IOTHREAD_MAIN_THREAD_ID);
    if (src->ref_repl_buf_node == NULL) return;
    dst->ref_repl_buf_node = src->ref_repl_buf_node;
    dst->ref_block_pos = src->ref_block_pos;
    ((replBufBlock *)listNodeValue(dst->ref_repl_buf_node))->refcount++;
}

static inline int _clientHasPendingRepliesNonSlave(client *c) {
    return c->bufpos || listLength(c->reply);
}

static inline int _clientHasPendingRepliesSlave(client *c) {
    /* Replicas use global shared replication buffer instead of
     * private output buffer. */
    serverAssert(c->bufpos == 0 && listLength(c->reply) == 0);
    if (c->ref_repl_buf_node == NULL) return 0;

    /* If the last replication buffer block content is totally sent,
     * we have nothing to send. */
    if (c->running_tid == IOTHREAD_MAIN_THREAD_ID) {
        listNode *ln = listLast(server.repl_buffer_blocks);
        replBufBlock *tail = listNodeValue(ln);
        if (ln == c->ref_repl_buf_node &&
            c->ref_block_pos == tail->used) return 0;
    } else {
        if (c->io_bound_repl_node == c->io_curr_repl_node &&
            c->io_bound_block_pos == c->io_curr_block_pos) return 0;
    }
    return 1;
}

/* Return true if the specified client has pending reply buffers to write to
 * the socket. */
int clientHasPendingReplies(client *c) {
    if (unlikely(clientTypeIsSlave(c))) {
        return _clientHasPendingRepliesSlave(c);
    }
    return _clientHasPendingRepliesNonSlave(c);
}

void clientAcceptHandler(connection *conn) {
    client *c = connGetPrivateData(conn);

    if (connGetState(conn) != CONN_STATE_CONNECTED) {
        serverLog(LL_WARNING,
                  "Error accepting a client connection: %s (addr=%s laddr=%s)",
                  connGetLastError(conn), getClientPeerId(c), getClientSockname(c));
        freeClientAsync(c);
        return;
    }

    /* If the server is running in protected mode (the default) and there
     * is no password set, nor a specific interface is bound, we don't accept
     * requests from non loopback interfaces. Instead we try to explain the
     * user what to do to fix it if needed. */
    if (server.protected_mode &&
        DefaultUser->flags & USER_FLAG_NOPASS)
    {
        if (connIsLocal(conn) != 1) {
            char *err =
                "-DENIED Redis is running in protected mode because protected "
                "mode is enabled and no password is set for the default user. "
                "In this mode connections are only accepted from the loopback interface. "
                "If you want to connect from external computers to Redis you "
                "may adopt one of the following solutions: "
                "1) Just disable protected mode sending the command "
                "'CONFIG SET protected-mode no' from the loopback interface "
                "by connecting to Redis from the same host the server is "
                "running, however MAKE SURE Redis is not publicly accessible "
                "from internet if you do so. Use CONFIG REWRITE to make this "
                "change permanent. "
                "2) Alternatively you can just disable the protected mode by "
                "editing the Redis configuration file, and setting the protected "
                "mode option to 'no', and then restarting the server. "
                "3) If you started the server manually just for testing, restart "
                "it with the '--protected-mode no' option. "
                "4) Set up an authentication password for the default user. "
                "NOTE: You only need to do one of the above things in order for "
                "the server to start accepting connections from the outside.\r\n";
            if (connWrite(c->conn,err,strlen(err)) == -1) {
                /* Nothing to do, Just to avoid the warning... */
            }
            server.stat_rejected_conn++;
            freeClientAsync(c);
            return;
        }
    }

    /* Auto-authenticate from cert_user field if set */
    sds username = connGetPeerUsername(conn);
    if (username != NULL) {
        user *u = ACLGetUserByName(username, sdslen(username));
        if (u && !(u->flags & USER_FLAG_DISABLED)) {
            c->authenticated = 1;
            clientSetUser(c, u);
            moduleNotifyUserChanged(c);
            serverLog(LL_VERBOSE, "TLS: Auto-authenticated client as %s",
                      server.hide_user_data_from_log ? "*redacted*" : u->name);
        } else {
            addACLLogEntry(c, ACL_INVALID_TLS_CERT_AUTH, ACL_LOG_CTX_TOPLEVEL, 0, username, NULL);
        }
        sdsfree(username);
    }

    server.stat_numconnections++;
    moduleFireServerEvent(REDISMODULE_EVENT_CLIENT_CHANGE,
                          REDISMODULE_SUBEVENT_CLIENT_CHANGE_CONNECTED,
                          c);

    /* Assign the client to an IO thread */
    if (server.io_threads_num > 1) assignClientToIOThread(c);
}

void acceptCommonHandler(connection *conn, int flags, char *ip) {
    client *c;
    UNUSED(ip);

    if (connGetState(conn) != CONN_STATE_ACCEPTING) {
        char addr[NET_ADDR_STR_LEN] = {0};
        char laddr[NET_ADDR_STR_LEN] = {0};
        connFormatAddr(conn, addr, sizeof(addr), 1);
        connFormatAddr(conn, laddr, sizeof(laddr), 0);
        serverLog(LL_VERBOSE,
                  "Accepted client connection in error state: %s (addr=%s laddr=%s)",
                  connGetLastError(conn), addr, laddr);
        connClose(conn);
        return;
    }

    /* Limit the number of connections we take at the same time.
     *
     * Admission control will happen before a client is created and connAccept()
     * called, because we don't want to even start transport-level negotiation
     * if rejected. */
    if (listLength(server.clients) + getClusterConnectionsCount()
        >= server.maxclients)
    {
        char *err;
        if (server.cluster_enabled)
            err = "-ERR max number of clients + cluster "
                  "connections reached\r\n";
        else
            err = "-ERR max number of clients reached\r\n";

        /* That's a best effort error message, don't check write errors.
         * Note that for TLS connections, no handshake was done yet so nothing
         * is written and the connection will just drop. */
        if (connWrite(conn,err,strlen(err)) == -1) {
            /* Nothing to do, Just to avoid the warning... */
        }
        server.stat_rejected_conn++;
        connClose(conn);
        return;
    }

    /* Create connection and client */
    if ((c = createClient(conn)) == NULL) {
        char addr[NET_ADDR_STR_LEN] = {0};
        char laddr[NET_ADDR_STR_LEN] = {0};
        connFormatAddr(conn, addr, sizeof(addr), 1);
        connFormatAddr(conn, laddr, sizeof(laddr), 0);
        serverLog(LL_WARNING,
                  "Error registering fd event for the new client connection: %s (addr=%s laddr=%s)",
                  connGetLastError(conn), addr, laddr);
        connClose(conn); /* May be already closed, just ignore errors */
        return;
    }

    /* Last chance to keep flags */
    c->flags |= flags;

    /* Initiate accept.
     *
     * Note that connAccept() is free to do two things here:
     * 1. Call clientAcceptHandler() immediately;
     * 2. Schedule a future call to clientAcceptHandler().
     *
     * Because of that, we must do nothing else afterwards.
     */
    if (connAccept(conn, clientAcceptHandler) == C_ERR) {
        if (connGetState(conn) == CONN_STATE_ERROR)
            serverLog(LL_WARNING,
                      "Error accepting a client connection: %s (addr=%s laddr=%s)",
                      connGetLastError(conn), getClientPeerId(c), getClientSockname(c));
        freeClient(connGetPrivateData(conn));
        return;
    }
}

static void freeDeferredObject(client *c, int type, void *ptr) {
    if (type == DEFERRED_OBJECT_TYPE_PENDING_COMMAND) {
        freePendingCommand(c, ptr);
    } else if (type == DEFERRED_OBJECT_TYPE_ROBJ) {
        decrRefCount(ptr);
    } else {
        serverPanic("Unknown deferred object type: %d", type);
    }
}

/* Attempt to defer freeing the object to the IO thread. We usually call this since
 * we know the object is allocated in the IO thread, to avoid memory arena contention,
 * and also reducing the load of the main thread. */
void tryDeferFreeClientObject(client *c, int type, void *ptr) {
    if (!c || c->tid == IOTHREAD_MAIN_THREAD_ID) {
        freeDeferredObject(c, type, ptr);
        return;
    }

    /* Put the object in the deferred objects array. */
    if (c->deferred_objects && c->deferred_objects_num < CLIENT_MAX_DEFERRED_OBJECTS) {
        c->deferred_objects[c->deferred_objects_num].type = type;
        c->deferred_objects[c->deferred_objects_num].ptr = ptr;
        c->deferred_objects_num++;
    } else {
        freeDeferredObject(c, type, ptr);
    }
}

/* Free the objects in the deferred_pending_cmds array. If free_array is true
 * then free the array itself as well. */
void freeClientDeferredObjects(client *c, int free_array) {
    for (int j = 0; j < c->deferred_objects_num; j++) {
        deferredObject *obj = &c->deferred_objects[j];
        freeDeferredObject(c, obj->type, obj->ptr);
    }
    c->deferred_objects_num = 0;

    if (free_array) {
        zfree(c->deferred_objects);
        c->deferred_objects = NULL;
    }
}

/* Queue an robj to be freed by the main thread when client returns from IO thread.
 * This is used in IO thread write path to avoid refcount race conditions. */
#define IO_DEFERRED_OBJECTS_INIT_SIZE 8
void ioDeferFreeRobj(client *c, robj *obj) {
    if (c->io_deferred_objects_num >= c->io_deferred_objects_size) {
        int new_size = !c->io_deferred_objects_size ?
            IO_DEFERRED_OBJECTS_INIT_SIZE : c->io_deferred_objects_size * 2;
        c->io_deferred_objects = zrealloc(c->io_deferred_objects, new_size * sizeof(robj *));
        c->io_deferred_objects_size = new_size;
    }
    c->io_deferred_objects[c->io_deferred_objects_num++] = obj;
}

/* Free all objects queued by IO thread for deferred freeing.
 * Called by main thread when client returns from IO thread.
 * If free_array is true then free the array itself as well. */
void freeClientIODeferredObjects(client *c, int free_array) {
    if (!c->conn) return;

    for (int i = 0; i < c->io_deferred_objects_num; i++) {
        robj *obj = c->io_deferred_objects[i];
        decrRefCount(obj);
    }

    if (!free_array) {
        /* If the utilization rate is less than 1/4, reduce the size to 1/2 to avoid thrashing */
        if (c->io_deferred_objects_size > IO_DEFERRED_OBJECTS_INIT_SIZE &&
            c->io_deferred_objects_num * 4 < c->io_deferred_objects_size)
        {
            int new_size = c->io_deferred_objects_size / 2;
            c->io_deferred_objects = zrealloc(c->io_deferred_objects, new_size * sizeof(robj *));
            c->io_deferred_objects_size = new_size;
        }
        c->io_deferred_objects_num = 0;
    } else {
        zfree(c->io_deferred_objects);
        c->io_deferred_objects = NULL;
        c->io_deferred_objects_num = 0;
        c->io_deferred_objects_size = 0;
    }
}

void freeClientOriginalArgv(client *c) {
    /* We didn't rewrite this client */
    if (!c->original_argv) return;

    for (int j = 0; j < c->original_argc; j++)
        decrRefCount(c->original_argv[j]);
    zfree(c->original_argv);
    c->original_argv = NULL;
    c->original_argc = 0;
}

static inline void freeClientArgvInternal(client *c, int free_argv) {
    int j;
    for (j = 0; j < c->argc; j++)
        decrRefCount(c->argv[j]);
    c->argc = 0;
    c->cmd = NULL;
    c->lookedcmd = NULL;
    if (free_argv) {
        c->argv_len = 0;
        zfree(c->argv);
        c->argv = NULL;
    }
}

void freeClientArgv(client *c) {
    freeClientArgvInternal(c, 1);
}

static inline void freeClientPendingCommands(client *c, int num_pcmds_to_free) {
    /* (-1) means free all pending commands */
    if (num_pcmds_to_free == -1)
        num_pcmds_to_free = c->pending_cmds.len;

    while (num_pcmds_to_free--) {
        pendingCommand *pcmd = popPendingCommandFromHead(&c->pending_cmds);
        serverAssert(pcmd);
        reclaimPendingCommand(c, pcmd);
    }
}

/* 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) {
    listIter li;
    listNode *ln;
    listRewind(server.slaves,&li);
    while((ln = listNext(&li))) {
        freeClient((client*)ln->value);
    }
}

/* Check if there is any other slave waiting dumping RDB finished expect me.
 * This function is useful to judge current dumping RDB can be used for full
 * synchronization or not. */
int anyOtherSlaveWaitRdb(client *except_me) {
    listIter li;
    listNode *ln;

    listRewind(server.slaves, &li);
    while((ln = listNext(&li))) {
        client *slave = ln->value;
        if (slave != except_me &&
            slave->replstate == SLAVE_STATE_WAIT_BGSAVE_END)
        {
            return 1;
        }
    }
    return 0;
}

/* Remove the specified client from global lists where the client could
 * be referenced, not including the Pub/Sub channels.
 * This is used by freeClient() and replicationCacheMaster(). */
void unlinkClient(client *c) {
    listNode *ln;

    /* If this is marked as current client unset it. */
    if (c->conn && server.current_client == c) server.current_client = NULL;

    /* Certain operations must be done only if the client has an active connection.
     * If the client was already unlinked or if it's a "fake client" the
     * conn is already set to NULL. */
    if (c->conn) {
        /* Remove from the list of active clients. */
        if (c->client_list_node) {
            uint64_t id = htonu64(c->id);
            raxRemove(server.clients_index,(unsigned char*)&id,sizeof(id),NULL);
            listDelNode(server.clients,c->client_list_node);
            c->client_list_node = NULL;
        }

        /* Check if this is a replica waiting for diskless replication (rdb pipe),
         * in which case it needs to be cleaned from that list */
        if (c->flags & CLIENT_SLAVE &&
            c->replstate == SLAVE_STATE_WAIT_BGSAVE_END &&
            server.rdb_pipe_conns)
        {
            int i;
            for (i=0; i < server.rdb_pipe_numconns; i++) {
                if (server.rdb_pipe_conns[i] == c->conn) {
                    rdbPipeWriteHandlerConnRemoved(c->conn);
                    server.rdb_pipe_conns[i] = NULL;
                    break;
                }
            }
        }
        /* Only use shutdown when the fork is active and we are the parent. */
        if (server.child_type) {
            /* connShutdown() may access TLS state. If this is a rdbchannel
             * client, bgsave fork is writing to the connection and TLS state in
             * the main process is stale. SSL_shutdown() involves a handshake,
             * and it may block the caller when used with stale TLS state.*/
            if (c->flags & CLIENT_REPL_RDB_CHANNEL)
                shutdown(c->conn->fd, SHUT_RDWR);
            else
                connShutdown(c->conn);
        }
        connClose(c->conn);
        c->conn = NULL;
    }

    /* Remove from the list of pending writes if needed. */
    if (c->flags & CLIENT_PENDING_WRITE) {
        serverAssert(&c->clients_pending_write_node.next != NULL || 
                     &c->clients_pending_write_node.prev != NULL);
        listUnlinkNode(server.clients_pending_write, &c->clients_pending_write_node);
        c->flags &= ~CLIENT_PENDING_WRITE;
    }

    /* When client was just unblocked because of a blocking operation,
     * remove it from the list of unblocked clients. */
    if (c->flags & CLIENT_UNBLOCKED) {
        ln = listSearchKey(server.unblocked_clients,c);
        serverAssert(ln != NULL);
        listDelNode(server.unblocked_clients,ln);
        c->flags &= ~CLIENT_UNBLOCKED;
    }

    freeClientPendingCommands(c, -1);
    c->argv_len = 0;
    c->argv = NULL;
    c->argc = 0;
    c->cmd = NULL;

    /* Clear the tracking status. */
    if (c->flags & CLIENT_TRACKING) disableTracking(c);
}

/* Remove client from the list of clients with pending referenced replies.
 * This is called when the client has finished sending all pending replies,
 * or when the client is being freed.
 *
 * If 'force' is true, the client is removed unconditionally.
 * This should only be used when we are certain that the replies no longer
 * contain any referenced robj. */
void tryUnlinkClientFromPendingRefReply(client *c, int force) {
    if (clientIsInPendingRefReplyList(c) && (force || !clientHasPendingReplies(c))) {
        listUnlinkNode(server.clients_with_pending_ref_reply, &c->pending_ref_reply_node);
    }
}

/* Count bytes in an encoded buffer where the client holds the last remaining
 * reference to the underlying string object (refcount == 1), meaning the key
 * has been deleted from the keyspace and only this client buffer keeps the
 * memory alive.
 *
 * Note: when multiple clients share a reference to the same object,
 * the object's refcount stays above 1 even after the key is deleted. In that
 * case none of those clients will be counted here, so the shared memory is
 * under-reported until all but one client has consumed its copy. */
static size_t computeUnsharedReplyBytes(char *buf, size_t bufpos) {
    size_t total = 0;
    char *ptr = buf;
    while (ptr < buf + bufpos) {
        payloadHeader *header = (payloadHeader *)ptr;
        ptr += sizeof(payloadHeader);
        if (header->payload_type == BULK_STR_REF) {
            bulkStrRef *str_ref = (bulkStrRef *)ptr;
            if (str_ref->obj != NULL && str_ref->obj->refcount == 1)
                total += sdslen(str_ref->obj->ptr);
        }
        ptr += header->payload_len;
    }
    return total;
}

/* Update the client's unshared reply memory (solely owned). */
void updateClientUnsharedReplyBytes(client *c) {
    c->reply_bytes_unshared = 0;

    /* No shared memory means no unshared memory either. */
    if (c->reply_bytes_shared == 0) return;

    /* Scan the static output buffer. */
    if (c->buf_encoded)
        c->reply_bytes_unshared += computeUnsharedReplyBytes(c->buf, c->bufpos);

    /* Scan each block in the reply list. */
    listIter reply_li;
    listNode *reply_ln;
    listRewind(c->reply, &reply_li);
    while ((reply_ln = listNext(&reply_li))) {
        clientReplyBlock *block = listNodeValue(reply_ln);
        if (block == NULL) continue; /* deferred-length placeholder */
        if (block->buf_encoded)
            c->reply_bytes_unshared += computeUnsharedReplyBytes(block->buf, block->used);
    }
}

/* Compute shared reply memory: total shared reply bytes and the unshared subset where the key
 * has been deleted and the client buffer is the sole holder. */
void getClientsSharedMemoryUsage(size_t *shared_mem, size_t *unshared_mem) {
    listNode *ln;
    listIter li;
    listRewind(server.clients_with_pending_ref_reply, &li);
    while ((ln = listNext(&li))) {
        client *c = listNodeValue(ln);

        /* Total shared reply bytes (logical size, shared with keyspace). */
        *shared_mem += c->reply_bytes_shared;

        /* Unshared reply bytes: the client is the sole owner because the key was deleted. */
        updateClientUnsharedReplyBytes(c);
        *unshared_mem += c->reply_bytes_unshared;
    }
}

/* Clear the client state to resemble a newly connected client. */
void clearClientConnectionState(client *c) {
    listNode *ln;

    /* MONITOR clients are also marked with CLIENT_SLAVE, we need to
     * distinguish between the two.
     */
    if (c->flags & CLIENT_MONITOR) {
        ln = listSearchKey(server.monitors,c);
        serverAssert(ln != NULL);
        listDelNode(server.monitors,ln);

        c->flags &= ~(CLIENT_MONITOR|CLIENT_SLAVE);
    }

    serverAssert(!(c->flags &(CLIENT_SLAVE|CLIENT_MASTER)));

    if (c->flags & CLIENT_TRACKING) disableTracking(c);
    selectDb(c,0);
#ifdef LOG_REQ_RES
    c->resp = server.client_default_resp;
#else
    c->resp = 2;
#endif

    clientSetDefaultAuth(c);
    moduleNotifyUserChanged(c);
    discardTransaction(c);
    himportFieldsetsFree(c);

    pubsubUnsubscribeAllChannels(c,0);
    pubsubUnsubscribeShardAllChannels(c, 0);
    pubsubUnsubscribeAllPatterns(c,0);
    unmarkClientAsPubSub(c);

    if (c->name) {
        decrRefCount(c->name);
        c->name = NULL;
    }

    /* Note: lib_name and lib_ver are not reset since they still
     * represent the client library behind the connection. */
    
    /* Selectively clear state flags not covered above */
    c->flags &= ~(CLIENT_ASKING|CLIENT_READONLY|CLIENT_REPLY_OFF|
                  CLIENT_REPLY_SKIP_NEXT|CLIENT_NO_TOUCH|CLIENT_NO_EVICT);
}

void deauthenticateAndCloseClient(client *c) {
    disableTracking(c);
    c->user = DefaultUser;
    c->authenticated = 0;
    /* We will write replies to this client later, so we can't
     * close it directly even if async. */
    if (c == server.current_client) {
        c->flags |= CLIENT_CLOSE_AFTER_COMMAND;
    } else {
        freeClientAsync(c);
    }
}

/* Resets the reusable query buffer used by the given client.
 * If any data remained in the buffer, the client will take ownership of the buffer
 * and a new empty buffer will be allocated for the reusable buffer. */
static void resetReusableQueryBuf(client *c) {
    serverAssert(c->io_flags & CLIENT_IO_REUSABLE_QUERYBUFFER);
    if (c->querybuf != thread_reusable_qb || sdslen(c->querybuf) > c->qb_pos) {
        /* If querybuf has been reallocated or there is still data left,
         * let the client take ownership of the reusable buffer. */
        thread_reusable_qb = NULL;
    } else {
        /* It is safe to dereference and reuse the reusable query buffer. */
        c->querybuf = NULL;
        c->qb_pos = 0;
        sdsclear(thread_reusable_qb);
    } 

    /* Mark that the client is no longer using the reusable query buffer
     * and indicate that it is no longer used by any client. */
    c->io_flags &= ~CLIENT_IO_REUSABLE_QUERYBUFFER;
    thread_reusable_qb_used = 0;
}

/* Release references to string objects inside an encoded buffer.
 * If running in IO thread, defer the free to main thread via io_deferred_objects. */
static void releaseBufReferences(client *c, char *buf, size_t bufpos) {
    int in_io_thread = (c && c->running_tid != IOTHREAD_MAIN_THREAD_ID);
    char *ptr = buf;
    while (ptr < buf + bufpos) {
        payloadHeader *header = (payloadHeader *)ptr;
        ptr += sizeof(payloadHeader);

        if (header->payload_type == BULK_STR_REF) {
            bulkStrRef *str_ref = (bulkStrRef *)ptr;
            /* Only release if not already released. */
            if (str_ref->obj != NULL) {
                c->reply_bytes_shared -= sdslen(str_ref->obj->ptr);
                if (in_io_thread)
                    ioDeferFreeRobj(c, str_ref->obj);
                else
                    decrRefCount(str_ref->obj);
                str_ref->obj = NULL;
            }
        } else {
            serverAssert(header->payload_type == PLAIN_REPLY);
        }

        ptr += header->payload_len;
    }
}

/* Release all references to string objects in all encoded buffers */
static void releaseAllBufReferences(client *c) {
    if (c->buf_encoded) {
        releaseBufReferences(c, c->buf, c->bufpos);
    }

    listIter iter;
    listNode *next;
    listRewind(c->reply, &iter);
    while ((next = listNext(&iter))) {
        clientReplyBlock *o = (clientReplyBlock *)listNodeValue(next);
        if (o && o->buf_encoded) {
            releaseBufReferences(c, o->buf, o->used);
        }
    }
}

void freeClient(client *c) {
    listNode *ln;

    /* If a client is protected, yet we need to free it right now, make sure
     * to at least use asynchronous freeing. */
    if (c->flags & CLIENT_PROTECTED) {
        freeClientAsync(c);
        return;
    }

    /* If the client is running in io thread, we can't free it directly. */
    if (c->running_tid != IOTHREAD_MAIN_THREAD_ID) {
        fetchClientFromIOThread(c);
    }

    /* We need to unbind connection of client from io thread event loop first. */
    if (c->tid != IOTHREAD_MAIN_THREAD_ID) {
        keepClientInMainThread(c);
    }

    /* Update the number of clients in the IO thread. */
    if (c->conn) server.io_threads_clients_num[c->tid]--;

    /* For connected clients, call the disconnection event of modules hooks. */
    if (c->conn) {
        moduleFireServerEvent(REDISMODULE_EVENT_CLIENT_CHANGE,
                              REDISMODULE_SUBEVENT_CLIENT_CHANGE_DISCONNECTED,
                              c);
    }

    asmCallbackOnFreeClient(c);

    /* Notify module system that this client auth status changed. */
    moduleNotifyUserChanged(c);

    /* Free the RedisModuleBlockedClient held onto for reprocessing if not already freed. */
    zfree(c->module_blocked_client);

    /* If this client was scheduled for async freeing we need to remove it
     * from the queue. Note that we need to do this here, because later
     * we may call replicationCacheMaster() and the client should already
     * be removed from the list of clients to free. */
    if (c->flags & CLIENT_CLOSE_ASAP) {
        ln = listSearchKey(server.clients_to_close,c);
        serverAssert(ln != NULL);
        listDelNode(server.clients_to_close,ln);
    }

    /* Disable compression if present */
    clientDestroyCompressionState(c);

    /* If it is our master that's being disconnected we should make sure
     * to cache the state to try a partial resynchronization later.
     *
     * Note that before doing this we make sure that the client is not in
     * some unexpected state, by checking its flags. */
    if (server.master && c->flags & CLIENT_MASTER) {
        serverLog(LL_NOTICE,"Connection with master lost.");
        if (!(c->flags & (CLIENT_PROTOCOL_ERROR|CLIENT_BLOCKED))) {
            c->flags &= ~(CLIENT_CLOSE_ASAP|CLIENT_CLOSE_AFTER_REPLY);
            c->io_flags &= ~CLIENT_IO_CLOSE_ASAP;
            replicationCacheMaster(c);
            return;
        }
    }

    /* Log link disconnection with slave */
    if (clientTypeIsSlave(c)) {
        const char *type = c->flags & CLIENT_REPL_RDB_CHANNEL ? " (rdbchannel)" : "";
        serverLog(LL_NOTICE,"Connection with replica%s %s lost.", type,
            replicationGetSlaveName(c));
    }

    /* Free the query buffer */
    if (c->io_flags & CLIENT_IO_REUSABLE_QUERYBUFFER)
        resetReusableQueryBuf(c);
    sdsfree(c->querybuf);
    c->querybuf = NULL;

    /* Deallocate structures used to block on blocking ops. */
    /* If there is any in-flight command, we don't record their duration. */
    c->duration = 0;
    if (c->flags & CLIENT_BLOCKED) unblockClient(c, 1);
    dictRelease(c->bstate.keys);

    /* UNWATCH all the keys */
    unwatchAllKeys(c);
    listRelease(c->watched_keys);

    /* Unsubscribe from all the pubsub channels */
    pubsubUnsubscribeAllChannels(c,0);
    pubsubUnsubscribeShardAllChannels(c, 0);
    pubsubUnsubscribeAllPatterns(c,0);
    unmarkClientAsPubSub(c);
    dictRelease(c->pubsub_channels);
    dictRelease(c->pubsub_patterns);
    dictRelease(c->pubsubshard_channels);

    /* Free data structures. */
    releaseAllBufReferences(c); /* Release all references to string objects in encoded buffers before freeing */
    listRelease(c->reply);
    zfree(c->buf);
    freeReplicaReferencedReplBuffer(c);
    freeClientOriginalArgv(c);
    freeClientDeferredObjects(c, 1);
    freeClientIODeferredObjects(c, 1);
    tryUnlinkClientFromPendingRefReply(c, 1);
    if (c->deferred_reply_errors)
        listRelease(c->deferred_reply_errors);
#ifdef LOG_REQ_RES
    reqresReset(c, 1);
#endif

    /* Remove the contribution that this client gave to our incrementally
     * computed memory usage. The cached master must be included too: its conn
     * was already detached by replicationCacheMaster(), so gating on c->conn
     * alone would skip the subtraction and leak its size from
     * mem_clients_normal on every discarded partial resync. */
    if (c->conn || c == server.cached_master)
        server.stat_clients_type_memory[c->last_memory_type] -=
            c->last_memory_usage;

    /* Unlink the client: this will close the socket, remove the I/O
     * handlers, and remove references of the client from different
     * places where active clients may be referenced.
     * This will also clean all remaining pending commands in the client,
     * as they are no longer valid.
     */
    unlinkClient(c);

    freeClientMultiState(c);
    serverAssert(c->pending_cmds.len == 0);

    /* Master/slave cleanup Case 1:
     * we lost the connection with a slave. */
    if (c->flags & CLIENT_SLAVE) {
        /* If there is no any other slave waiting dumping RDB finished, the
         * current child process need not continue to dump RDB, then we kill it.
         * So child process won't use more memory, and we also can fork a new
         * child process asap to dump rdb for next full synchronization or bgsave.
         * But we also need to check if users enable 'save' RDB, if enable, we
         * should not remove directly since that means RDB is important for users
         * to keep data safe and we may delay configured 'save' for full sync. */
        if (server.saveparamslen == 0 &&
            c->replstate == SLAVE_STATE_WAIT_BGSAVE_END &&
            server.child_type == CHILD_TYPE_RDB &&
            server.rdb_child_type == RDB_CHILD_TYPE_DISK &&
            anyOtherSlaveWaitRdb(c) == 0)
        {
            killRDBChild();
        }
        if (c->replstate == SLAVE_STATE_SEND_BULK) {
            if (c->repldbfd != -1) close(c->repldbfd);
            if (c->replpreamble) sdsfree(c->replpreamble);
        }
        list *l = (c->flags & CLIENT_MONITOR) ? server.monitors : server.slaves;
        ln = listSearchKey(l,c);
        serverAssert(ln != NULL);
        listDelNode(l,ln);
        /* We need to remember the time when we started to have zero
         * attached slaves, as after some time we'll free the replication
         * backlog. */
        if (clientTypeIsSlave(c) && listLength(server.slaves) == 0)
            server.repl_no_slaves_since = server.unixtime;
        refreshGoodSlavesCount();
        /* Fire the replica change modules event. */
        if (c->replstate == SLAVE_STATE_ONLINE)
            moduleFireServerEvent(REDISMODULE_EVENT_REPLICA_CHANGE,
                                  REDISMODULE_SUBEVENT_REPLICA_CHANGE_OFFLINE,
                                  NULL);
    }

    /* Master/slave cleanup Case 2:
     * we lost the connection with the master. */
    if (c->flags & CLIENT_MASTER) replicationHandleMasterDisconnection();

    /* Remove client from memory usage buckets */
    if (c->mem_usage_bucket) {
        c->mem_usage_bucket->mem_usage_sum -= c->last_memory_usage;
        listDelNode(c->mem_usage_bucket->clients, c->mem_usage_bucket_node);
    }

    /* Release other dynamically allocated client structure fields,
     * and finally release the client structure itself. */
    himportFieldsetsFree(c);
    if (c->name) decrRefCount(c->name);
    if (c->lib_name) decrRefCount(c->lib_name);
    if (c->lib_ver) decrRefCount(c->lib_ver);
    serverAssert(c->all_argv_len_sum == 0);
    sdsfree(c->peerid);
    sdsfree(c->sockname);
    sdsfree(c->slave_addr);
    sdsfree(c->node_id);
    zfree(c);
}

/* Schedule a client to free it at a safe time in the beforeSleep() 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(client *c) {
    if (c->running_tid != IOTHREAD_MAIN_THREAD_ID) {
        int main_thread = pthread_equal(pthread_self(), server.main_thread_id);
        /* Make sure the main thread can access IO thread data safely. */
        if (main_thread) pauseIOThread(c->tid);
        if (!(c->io_flags & CLIENT_IO_CLOSE_ASAP)) {
            c->io_flags |= CLIENT_IO_CLOSE_ASAP;
            enqueuePendingClientsToMainThread(c, 1);
        }
        if (main_thread) resumeIOThread(c->tid);
        return;
    }

    if (c->flags & CLIENT_CLOSE_ASAP || c->flags & CLIENT_SCRIPT) return;
    c->flags |= CLIENT_CLOSE_ASAP;
    /* Replicas that was marked as CLIENT_CLOSE_ASAP should not keep the
     * replication backlog from been trimmed. */
    if (c->flags & CLIENT_SLAVE) freeReplicaReferencedReplBuffer(c);
    listAddNodeTail(server.clients_to_close,c);
}

/* Log errors for invalid use and free the client in async way.
 * We will add additional information about the client to the message. */
void logInvalidUseAndFreeClientAsync(client *c, const char *fmt, ...) {
    va_list ap;
    va_start(ap, fmt);
    sds info = sdscatvprintf(sdsempty(), fmt, ap);
    va_end(ap);

    sds client = catClientInfoString(sdsempty(), c);
    serverLog(LL_WARNING, "%s, disconnecting it: %s", info, client);

    sdsfree(info);
    sdsfree(client);
    freeClientAsync(c);
}

/* Perform processing of the client before moving on to processing the next client
 * this is useful for performing operations that affect the global state but can't
 * wait until we're done with all clients. In other words can't wait until beforeSleep()
 * return C_ERR in case client is no longer valid after call.
 * The input client argument: c, may be NULL in case the previous client was
 * freed before the call. */
int beforeNextClient(client *c) {
    /* Notice, this code is also called from 'processUnblockedClients'.
     * But in case of a module blocked client (see RM_Call 'K' flag) we do not reach this code path.
     * So whenever we change the code here we need to consider if we need this change on module
     * blocked client as well */

    /* Skip the client processing if we're in an IO thread, in that case we'll perform
       this operation later (this function is called again) in the fan-in stage of the threading mechanism */
    if (c && c->running_tid != IOTHREAD_MAIN_THREAD_ID)
        return C_OK;
    /* Handle async frees */
    /* Note: this doesn't make the server.clients_to_close list redundant because of
     * cases where we want an async free of a client other than myself. For example
     * in ACL modifications we disconnect clients authenticated to non-existent
     * users (see ACL LOAD). */
    if (c && (c->flags & CLIENT_CLOSE_ASAP)) {
        freeClient(c);
        return C_ERR;
    }
    return C_OK;
}

/* Free the clients marked as CLOSE_ASAP, return the number of clients
 * freed. */
int freeClientsInAsyncFreeQueue(void) {
    int freed = 0;
    listIter li;
    listNode *ln;

    listRewind(server.clients_to_close,&li);
    while ((ln = listNext(&li)) != NULL) {
        client *c = listNodeValue(ln);

        if (c->flags & CLIENT_PROTECTED) continue;

        c->flags &= ~CLIENT_CLOSE_ASAP;
        freeClient(c);
        listDelNode(server.clients_to_close,ln);
        freed++;
    }
    return freed;
}

/* Return a client by ID, or NULL if the client ID is not in the set
 * of registered clients. Note that "fake clients", created with -1 as FD,
 * are not registered clients. */
client *lookupClientByID(uint64_t id) {
    id = htonu64(id);
    void *c = NULL;
    raxFind(server.clients_index,(unsigned char*)&id,sizeof(id),&c);
    return c;
}

/* This struct is used by writevToClient to prepare iovec array for submitting to connWritev */
typedef struct ReplyIOV {
    struct iovec *iov;      /* Array of iovec structures for writev() */
    int iovmax;             /* Maximum number of iovec entries allocated */
    int iovcnt;             /* Current number of iovec entries in use */
    size_t iov_bytes_len;   /* Total bytes across all iovec entries */
} ReplyIOV;

/* Check if the reply IOV has reached its limit yet. */
static int replyIOVReachLimit(ReplyIOV *reply_iov) {
    return reply_iov->iovcnt >= reply_iov->iovmax || reply_iov->iov_bytes_len >= NET_MAX_WRITES_PER_EVENT;
}

/* Helper function to process encoded buffer and build iov array. */
static void processEncodedBufferForWrite(ReplyIOV *reply_iov, char *start_ptr, char *end_ptr, size_t offset) {
    char *ptr = start_ptr;
    while (ptr < end_ptr && !replyIOVReachLimit(reply_iov)) {
        payloadHeader *head = (payloadHeader *)ptr;

        if (head->payload_type == PLAIN_REPLY) {
            /* Plain data - add directly */
            reply_iov->iov[reply_iov->iovcnt].iov_base = ptr + sizeof(payloadHeader) + offset;
            reply_iov->iov[reply_iov->iovcnt].iov_len = head->payload_len - offset;
            reply_iov->iov_bytes_len += reply_iov->iov[reply_iov->iovcnt++].iov_len;
        } else {
            /* BULK_STR_REF - expand to prefix + string + crlf (format prefix at write time) */
            bulkStrRef *str_ref = (bulkStrRef *)(ptr + sizeof(payloadHeader));
            size_t prefix_len = formatBulkStrRefPrefix(str_ref);
            size_t str_len = sdslen(str_ref->obj->ptr);

            /* Add prefix */
            if (offset < prefix_len) {
                if (replyIOVReachLimit(reply_iov)) return;
                reply_iov->iov[reply_iov->iovcnt].iov_base = str_ref->prefix + offset;
                reply_iov->iov[reply_iov->iovcnt].iov_len = prefix_len - offset;
                reply_iov->iov_bytes_len += reply_iov->iov[reply_iov->iovcnt++].iov_len;
                offset = 0;
            } else {
                offset -= prefix_len;
            }

            /* Add string data */
            if (offset < str_len) {
                if (replyIOVReachLimit(reply_iov)) return;
                reply_iov->iov[reply_iov->iovcnt].iov_base = (char *)str_ref->obj->ptr + offset;
                reply_iov->iov[reply_iov->iovcnt].iov_len = str_len - offset;
                reply_iov->iov_bytes_len += reply_iov->iov[(reply_iov->iovcnt)++].iov_len;
                offset = 0;
            } else {
                offset -= str_len;
            }

            /* Add crlf */
            if (offset < 2) {
                if (replyIOVReachLimit(reply_iov)) return;
                reply_iov->iov[reply_iov->iovcnt].iov_base = str_ref->crlf + offset;
                reply_iov->iov[reply_iov->iovcnt].iov_len = 2 - offset;
                reply_iov->iov_bytes_len += reply_iov->iov[reply_iov->iovcnt++].iov_len;
            }
        }

        offset = 0;
        ptr += sizeof(payloadHeader) + head->payload_len;
    }
}

/* Process sent data in the encoded buffer.
 * Returns pointer to the current payload header being processed, or NULL if all data is processed.
 * If running in IO thread, defer the free to main thread via io_deferred_objects. */
static payloadHeader *processSentDataInEncodedBuffer(client *c, char *start_ptr, char *end_ptr,
                                                     size_t *sentlen, ssize_t *remaining)
{
    int in_io_thread = (c && c->running_tid != IOTHREAD_MAIN_THREAD_ID);
    char *ptr = start_ptr;
    while (ptr < end_ptr && *remaining > 0) {
        payloadHeader *head = (payloadHeader *)ptr;

        if (head->payload_type == PLAIN_REPLY) {
            if (*remaining < (ssize_t)(head->payload_len - *sentlen)) {
                *sentlen += *remaining;
                *remaining = 0;
                return head;
            }
            *remaining -= (head->payload_len - *sentlen);
            *sentlen = 0;
        } else {
            /* BULK_STR_REF - release object references */
            bulkStrRef *str_ref = (bulkStrRef *)(ptr + sizeof(payloadHeader));
            formatBulkStrRefPrefix(str_ref); /* ensure prefix_cnt is set for writen_len */

            size_t writen_len = str_ref->prefix_cnt + sdslen(str_ref->obj->ptr) + 2;
            if (*remaining < (ssize_t)(writen_len - *sentlen)) {
                *sentlen += *remaining;
                *remaining = 0;
                return head;
            }
            *remaining -= (writen_len - *sentlen);
            c->reply_bytes_shared -= sdslen(str_ref->obj->ptr);
            if (in_io_thread) {
                ioDeferFreeRobj(c, str_ref->obj);
            } else {
                decrRefCount(str_ref->obj);
            }
            str_ref->obj = NULL; /* Mark as released to prevent double free */
            *sentlen = 0;
        }

        ptr += sizeof(payloadHeader) + head->payload_len;
    }

    return (ptr == end_ptr) ? NULL : (payloadHeader *)ptr;
}

/* This function should be called from _writeToClient when the reply list is not empty,
 * it gathers the scattered buffers from reply list and sends them away with connWritev.
 * If we write successfully, it returns C_OK, otherwise, C_ERR is returned,
 * and 'nwritten' is an output parameter, it means how many bytes server write
 * to client. */
static int _writevToClient(client *c, ssize_t *nwritten) {
    int iovmax = min(IOV_MAX, c->conn->iovcnt);
    struct iovec iov[iovmax];
    ReplyIOV reply_iov = {iov, iovmax};

    /* Add c->buf to iov array */
    if (c->bufpos > 0) {
        if (likely(!c->buf_encoded)) {
            /* Non-encoded buffer - add directly */
            iov[reply_iov.iovcnt].iov_base = c->buf + c->sentlen;
            iov[reply_iov.iovcnt].iov_len = c->bufpos - c->sentlen;
            reply_iov.iov_bytes_len += iov[reply_iov.iovcnt++].iov_len;
        } else {
            /* Encoded buffer */
            char *start_ptr = c->last_header ? (char *)c->last_header : c->buf;
            serverAssert(start_ptr >= c->buf && start_ptr < (c->buf + c->bufpos));
            processEncodedBufferForWrite(&reply_iov, start_ptr, c->buf + c->bufpos, c->sentlen);
        }
    }

    /* Add c->reply list nodes to iov array */
    if (!replyIOVReachLimit(&reply_iov)) {
        /* The first node of reply list might be incomplete from the last call,
         * thus it needs to be calibrated to get the actual data address and length. */
        size_t offset = c->bufpos > 0 ? 0 : c->sentlen;
        payloadHeader *last_header = c->bufpos > 0 ? NULL : c->last_header;
        listIter iter;
        listNode *next;
        listRewind(c->reply, &iter);
        while ((next = listNext(&iter)) && !replyIOVReachLimit(&reply_iov)) {
            clientReplyBlock *o = listNodeValue(next);
            if (o->used == 0) { /* empty node, just release it and skip. */
                c->reply_bytes -= o->size;
                listDelNode(c->reply, next);
                offset = 0;
                last_header = NULL;
                continue;
            }

            if (!o->buf_encoded) {
                serverAssert(!last_header);
                /* Non-encoded reply block - add directly */
                iov[reply_iov.iovcnt].iov_base = o->buf + offset;
                iov[reply_iov.iovcnt].iov_len = o->used - offset;
                reply_iov.iov_bytes_len += iov[reply_iov.iovcnt++].iov_len;
                offset = 0;
            } else {
                /* Encoded reply block */
                char *start_ptr = last_header ? (char *)last_header : o->buf;
                processEncodedBufferForWrite(&reply_iov, start_ptr, o->buf + o->used, offset);
                offset = 0;
                last_header = NULL;
            }
        }
    }

    if (reply_iov.iovcnt == 0) return C_OK;
    *nwritten = connWritev(c->conn, iov, reply_iov.iovcnt);
    if (*nwritten <= 0) return C_ERR;

    /* Locate the new node which has leftover data and
     * release all nodes in front of it. */
    ssize_t remaining = *nwritten;
    if (c->bufpos > 0) {
        if (likely(!c->buf_encoded)) {
            int buf_len = c->bufpos - c->sentlen;
            c->sentlen += remaining;
            /* If the buffer was sent, set bufpos to zero to continue with
            * the remainder of the reply. */
            if (remaining >= buf_len) {
                c->bufpos = 0;
                c->sentlen = 0;
            }
            remaining -= buf_len;
        } else {
            /* For encoded buffers */
            char *start_ptr = c->last_header ? (char *)c->last_header : c->buf;
            c->last_header = processSentDataInEncodedBuffer(c, start_ptr, c->buf + c->bufpos, &c->sentlen, &remaining);
            if (!c->last_header) { /* reach end */
                c->bufpos = 0;
                c->buf_encoded = 0;
                c->sentlen = 0;
            }
        }
    }

    /* Process c->reply list nodes */
    listIter iter;
    listNode *next;
    listRewind(c->reply, &iter);
    while (remaining > 0) {
        next = listNext(&iter);
        clientReplyBlock *o = listNodeValue(next);

        if (!o->buf_encoded) {
            if (remaining < (ssize_t)(o->used - c->sentlen)) {
                c->sentlen += remaining;
                break;
            }
            remaining -= (ssize_t)(o->used - c->sentlen);
            c->reply_bytes -= o->size;
            listDelNode(c->reply, next);
            c->sentlen = 0;
        } else {
            /* Encoded reply block */
            char *start_ptr = c->last_header ? (char *)c->last_header : o->buf;
            c->last_header = processSentDataInEncodedBuffer(c, start_ptr, o->buf + o->used, &c->sentlen, &remaining);
            if (!c->last_header) { /* reach end */
                /* Block fully consumed, remove it */
                c->reply_bytes -= o->size;
                listDelNode(c->reply, next);
                c->sentlen = 0;
            } else {
                /* Partial write, c->sentlen and o->last_header already updated, stop processing */
                break;
            }
        }
    }

    return C_OK;
}

/* This function does actual writing output buffers for non slave client types,
 * it is called by writeToClient.
 * If we write successfully, it returns C_OK, otherwise, C_ERR is returned,
 * and 'nwritten' is an output parameter, it means how many bytes server write
 * to client. */
static inline int _writeToClientNonSlave(client *c, ssize_t *nwritten) {
    *nwritten = 0;
    /* When the reply list is not empty, it's better to use writev to save us some
     * system calls and TCP packets. */
    if (listLength(c->reply) > 0) {
        int ret = _writevToClient(c, nwritten);
        if (ret != C_OK) return ret;

        /* If there are no longer objects in the list, we expect
         * the count of reply bytes to be exactly zero. */
        if (listLength(c->reply) == 0)
            serverAssert(c->reply_bytes == 0 && c->reply_bytes_shared == 0);
    } else if (c->bufpos > 0) {
        /* For encoded buffers, we need to use writev to handle bulk string references */
        if (c->buf_encoded) {
            int ret = _writevToClient(c, nwritten);
            return ret;
        }

        *nwritten = connWrite(c->conn, c->buf + c->sentlen, c->bufpos - c->sentlen);
        if (*nwritten <= 0) return C_ERR;
        c->sentlen += *nwritten;

        /* If the buffer was sent, set bufpos to zero to continue with
         * the remainder of the reply. */
        if (c->sentlen == c->bufpos) {
            c->bufpos = 0;
            c->sentlen = 0;
        }
    }
    return C_OK;
}

static inline int _writeToClientSlaveInIOThread(client *c, ssize_t *nwritten) {
    replBufBlock *o = listNodeValue(c->io_curr_repl_node);
    /* The IO thread must not send data beyond the bound position. */
    size_t pos = c->io_curr_repl_node == c->io_bound_repl_node ?
                 c->io_bound_block_pos : o->used;
    if (pos > c->io_curr_block_pos) {
        /* Compression is applied at the client level (see client_comp.c): when
         * compression is active `consumed` is how many uncompressed bytes were
         * taken from the repl buffer while `sock_written` is how many bytes were
         * actually written to the socket. Without compression they are equal. */
        int network_written = 0;
        int consumed = clientConnWrite(c, o->buf+c->io_curr_block_pos,
                                       pos-c->io_curr_block_pos, &network_written);

        if (consumed <= 0) return C_ERR;

        /* *nwritten counts bytes actually put on the socket (compressed when
         * compression is active), while `consumed` is how many bytes were read
         * from the repl buffer node. */
        *nwritten += network_written;
        c->io_curr_block_pos += consumed;
    }
    /* If we fully sent the object and there are more nodes to send, go to the next one. */
    if (c->io_curr_block_pos == pos && c->io_curr_repl_node != c->io_bound_repl_node) {
        c->io_curr_repl_node = listNextNode(c->io_curr_repl_node);
        c->io_curr_block_pos = 0;
    }

    /* If the replica has drained all currently available replication data (it
     * "caught up"), force flush any data still buffered in its compressor so it
     * reaches the replica now instead of waiting for the periodic
     * repl-compression-max-latency flush. Under sustained load the replica won't
     * be caught up between writes, so the compressor keeps batching data into
     * the current frame. No-op when compression is inactive or nothing is
     * buffered. */
    if (c->io_curr_repl_node == c->io_bound_repl_node &&
        c->io_curr_block_pos == c->io_bound_block_pos)
    {
        int flushed = clientFlushCompressedData(c);
        if (flushed < 0) return C_ERR;
        *nwritten += flushed;
    }
    return C_OK;
}

/* This function does actual writing output buffers for slave client types,
 * it is called by writeToClient.
 * If we write successfully, it returns C_OK, otherwise, C_ERR is returned,
 * and 'nwritten' is an output parameter, it means how many bytes server write
 * to client. */
static inline int _writeToClientSlave(client *c, ssize_t *nwritten) {
    *nwritten = 0;
    serverAssert(c->bufpos == 0 && listLength(c->reply) == 0);

    if (c->running_tid != IOTHREAD_MAIN_THREAD_ID) {
        return _writeToClientSlaveInIOThread(c, nwritten);
    }

    replBufBlock *o = listNodeValue(c->ref_repl_buf_node);
    serverAssert(o->used >= c->ref_block_pos);
    /* Send current block if it is not fully sent. */
    if (o->used > c->ref_block_pos) {
        *nwritten = connWrite(c->conn, o->buf+c->ref_block_pos,
                                o->used-c->ref_block_pos);
        if (*nwritten <= 0) return C_ERR;
        c->ref_block_pos += *nwritten;
    }

    /* If we fully sent the object on head, go to the next one. */
    listNode *next = listNextNode(c->ref_repl_buf_node);
    if (next && c->ref_block_pos == o->used) {
        o->refcount--;
        ((replBufBlock *)(listNodeValue(next)))->refcount++;
        c->ref_repl_buf_node = next;
        c->ref_block_pos = 0;
        incrementalTrimReplicationBacklog(REPL_BACKLOG_TRIM_BLOCKS_PER_CALL);
    }
    return C_OK;
}

/* Write data in output buffers to client. Return C_OK if the client
 * is still valid after the call, C_ERR if it was freed because of some
 * error.  If handler_installed is set, it will attempt to clear the
 * write event.
 *
 * This function is called by threads, but always with handler_installed
 * set to 0. So when handler_installed is set to 0 the function must be
 * thread safe. */
int writeToClient(client *c, int handler_installed) {
    if (!(c->io_flags & CLIENT_IO_WRITE_ENABLED)) return C_OK;
    /* Update the number of writes of io threads on server */
    atomicIncr(IOThreads[c->running_tid].io_writes_processed, 1);

    ssize_t nwritten = 0, totwritten = 0;
    const int is_slave = clientTypeIsSlave(c);

    if (unlikely(is_slave)) {
        /* We send as much as possible if the client is
         * a slave (otherwise, on high-speed traffic, the
         * replication buffer will grow indefinitely) */
        while(_clientHasPendingRepliesSlave(c)) {
            int ret = _writeToClientSlave(c, &nwritten);
            if (ret == C_ERR) break;
            totwritten += nwritten;
        }
        atomicIncr(server.stat_net_repl_output_bytes, totwritten);
    } else {
        /* If we reach this block and client is marked with CLIENT_SLAVE flag
         * it's because it's a MONITOR/slot-migration client, which are marked
         * as replicas, but exposed as normal clients */
        const int is_normal_client = !(c->flags & CLIENT_SLAVE);
        while (_clientHasPendingRepliesNonSlave(c)) {
            int ret = _writeToClientNonSlave(c, &nwritten);
            if (ret == C_ERR) break;
            totwritten += nwritten;
            /* Note that we avoid to send more than NET_MAX_WRITES_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 interface).
             *
             * However if we are over the maxmemory limit we ignore that and
             * just deliver as much data as it is possible to deliver.
             *
             * Moreover, we also send as much as possible if the client is
             * a slave (covered above) or a monitor (covered here).
             * (otherwise, on high-speed traffic, the
             * output buffer will grow indefinitely) */
            if (totwritten > NET_MAX_WRITES_PER_EVENT &&
                (server.maxmemory == 0 ||
                zmalloc_used_memory() < server.maxmemory) &&
                is_normal_client) break;
        }
        atomicIncr(server.stat_net_output_bytes, totwritten);
    }
    c->net_output_bytes += totwritten;

    if (nwritten == -1) {
        if (connGetState(c->conn) != CONN_STATE_CONNECTED) {
            serverLog(LL_VERBOSE,
                "Error writing to client: %s", connGetLastError(c->conn));
            freeClientAsync(c);
            return C_ERR;
        }
    }
    if (totwritten > 0) {
        /* For clients representing masters we don't count sending data
         * as an interaction, since we always send REPLCONF ACK commands
         * that take some time to just fill the socket output buffer.
         * We just rely on data / pings received for timeout detection. */
        if (!(c->flags & CLIENT_MASTER)) c->lastinteraction = server.unixtime;
    }
    if (!clientHasPendingReplies(c)) {
        c->sentlen = 0;
        /* Note that writeToClient() is called in a threaded way, but
         * aeDeleteFileEvent() is not thread safe: however writeToClient()
         * is always called with handler_installed set to 0 from threads
         * so we are fine. */
        if (handler_installed) {
            /* IO Thread also can do that now. */
            connSetWriteHandler(c->conn, NULL);
        }

        /* Close connection after entire reply has been sent. */
        if (c->flags & CLIENT_CLOSE_AFTER_REPLY) {
            freeClientAsync(c);
            return C_ERR;
        }

        /* Remove client from pending referenced reply clients list. */
        if (c->running_tid == IOTHREAD_MAIN_THREAD_ID)
            tryUnlinkClientFromPendingRefReply(c, 1);

        /* If replica client has sent all the replication data it knows about
         * we send it to main thread so it can pick up new repl data ASAP.
         * Note, that we keep it in IO thread in case we have a pending ACK read. */
        if (c->flags & CLIENT_SLAVE && c->running_tid != IOTHREAD_MAIN_THREAD_ID) {
            if (!replicaFromIOThreadHasPendingRead(c))
                enqueuePendingClientsToMainThread(c, 0);
        }
    }
    /* Update client's memory usage after writing.
     * Since this isn't thread safe we do this conditionally. */
    if (c->running_tid == IOTHREAD_MAIN_THREAD_ID) {
        updateClientMemUsageAndBucket(c);
    }
    return C_OK;
}

/* Write event handler. Just send data to the client. */
void sendReplyToClient(connection *conn) {
    client *c = connGetPrivateData(conn);
    writeToClient(c,1);
}

/* This function is called just before entering the event loop, in the hope
 * we can just write the replies to the client output buffer without any
 * need to use a syscall in order to install the writable event handler,
 * get it called, and so forth. */
int handleClientsWithPendingWrites(void) {
    listIter li;
    listNode *ln;
    int processed = listLength(server.clients_pending_write);

    listRewind(server.clients_pending_write,&li);
    while((ln = listNext(&li))) {
        client *c = listNodeValue(ln);

        /* We handle IO thread replicas in putReplicasInPendingClientsToIOThreads */
        if (c->flags & CLIENT_SLAVE && c->tid != IOTHREAD_MAIN_THREAD_ID)
            continue;

        c->flags &= ~CLIENT_PENDING_WRITE;
        listUnlinkNode(server.clients_pending_write,ln);

        /* If a client is protected, don't do anything,
         * that may trigger write error or recreate handler. */
        if (c->flags & CLIENT_PROTECTED) continue;

        /* Don't write to clients that are going to be closed anyway. */
        if (c->flags & CLIENT_CLOSE_ASAP) continue;

        /* Let IO thread handle the client if possible. */
        if (server.io_threads_num > 1 &&
            !(c->flags & CLIENT_CLOSE_AFTER_REPLY) &&
            c->tid == IOTHREAD_MAIN_THREAD_ID &&
            !isClientMustHandledByMainThread(c))
        {
            assignClientToIOThread(c);
            continue;
        }

        /* Try to write buffers to the client socket. */
        if (writeToClient(c,0) == C_ERR) continue;

        /* If after the synchronous writes above we still have data to
         * output to the client, we need to install the writable handler. */
        if (clientHasPendingReplies(c)) {
            installClientWriteHandler(c);
        }
    }
    /* Record the flushed offset so flushSlavesOutputBuffersIfNeeded() only
     * fires when a single event loop accumulates more than the threshold. */
    server.repl_last_flush_offset = server.master_repl_offset;
    return processed;
}

/* Prepare the client for the parsing of the next command. */
void resetClientQbufState(client *c) {
    c->reqtype = 0;
    c->multibulklen = 0;
    c->bulklen = -1;
}

static inline void resetClientInternal(client *c, int num_pcmds_to_free) {
    redisCommandProc *prevcmd = c->cmd ? c->cmd->proc : NULL;

    /* We may get here with no pending commands but with an argv that needs freeing.
     * An example is in the case of modules (RM_Call) */
    if (c->current_pending_cmd) {
        freeClientPendingCommands(c, num_pcmds_to_free);
        if (c->pending_cmds.len == 0)
            serverAssert(c->all_argv_len_sum == 0);
        c->current_pending_cmd = NULL;
    } else if (c->argv) {
        freeClientArgvInternal(c, 1 /* free_argv */);
        /* If we're dealing with a client that doesn't create pendingCommand structs (e.g.: a Lua client),
         * clear the all_argv_len_sum counter so we don't get to freeing the client with it non-zero. */
        c->all_argv_len_sum = 0;
    }

    c->argc = 0;
    c->cmd = NULL;
    c->argv_len = 0;
    c->argv = NULL;
    c->cur_script = NULL;
    c->slot = -1;
    c->cluster_compatibility_check_slot = -2;
    if (c->flags & CLIENT_EXECUTING_COMMAND)
        c->flags &= ~CLIENT_EXECUTING_COMMAND;

    /* Make sure the duration has been recorded to some command. */
    serverAssert(c->duration == 0);
#ifdef LOG_REQ_RES
    reqresReset(c, 1);
#endif

    if (c->deferred_reply_errors)
        listRelease(c->deferred_reply_errors);
    c->deferred_reply_errors = NULL;

    /* We clear the ASKING flag as well if we are not inside a MULTI, and
     * if what we just executed is not the ASKING command itself. */
    if (c->flags & CLIENT_ASKING && !(c->flags & CLIENT_MULTI) &&
        prevcmd != askingCommand)
    {
        c->flags &= ~CLIENT_ASKING;
    }

    /* We do the same for the CACHING command as well. It also affects
     * the next command or transaction executed, in a way very similar
     * to ASKING. */
    if (c->flags & CLIENT_TRACKING_CACHING && !(c->flags & CLIENT_MULTI) &&
        prevcmd != clientCommand)
    {
        c->flags &= ~CLIENT_TRACKING_CACHING;
    }

    /* Remove the CLIENT_REPLY_SKIP flag if any so that the reply
     * to the next command will be sent, but set the flag if the command
     * we just processed was "CLIENT REPLY SKIP". */
    if (c->flags & CLIENT_REPLY_SKIP)
        c->flags &= ~CLIENT_REPLY_SKIP;

    if (c->flags & CLIENT_REPLY_SKIP_NEXT) {
        c->flags |= CLIENT_REPLY_SKIP;
        c->flags &= ~CLIENT_REPLY_SKIP_NEXT;
    }

    c->net_input_bytes_curr_cmd = 0;
    c->net_output_bytes_curr_cmd = 0;
}

/* resetClient prepare the client to process the next command */
void resetClient(client *c, int num_pcmds_to_free) {
    resetClientInternal(c, num_pcmds_to_free);
}

/* This function is used when we want to re-enter the event loop but there
 * is the risk that the client we are dealing with will be freed in some
 * way. This happens for instance in:
 *
 * * DEBUG RELOAD and similar.
 * * When a Lua script is in -BUSY state.
 *
 * So the function will protect the client by doing two things:
 *
 * 1) It removes the file events. This way it is not possible that an
 *    error is signaled on the socket, freeing the client.
 * 2) Moreover it makes sure that if the client is freed in a different code
 *    path, it is not really released, but only marked for later release. */
void protectClient(client *c) {
    c->flags |= CLIENT_PROTECTED;
    if (c->conn && c->tid == IOTHREAD_MAIN_THREAD_ID) {
        connSetReadHandler(c->conn,NULL);
        connSetWriteHandler(c->conn,NULL);
    }
}

/* This will undo the client protection done by protectClient() */
void unprotectClient(client *c) {
    if (c->flags & CLIENT_PROTECTED) {
        c->flags &= ~CLIENT_PROTECTED;
        if (c->conn) {
            if (c->tid == IOTHREAD_MAIN_THREAD_ID)
                connSetReadHandler(c->conn,readQueryFromClient);
            if (clientHasPendingReplies(c)) putClientInPendingWriteQueue(c);
        }
    }
}

/* Like processMultibulkBuffer(), but for the inline protocol instead of RESP,
 * this function consumes the client query buffer and creates a command ready
 * to be executed inside the client structure. Returns C_OK if the command
 * is ready to be executed, or C_ERR if there is still protocol to read to
 * have a well formed command. The function also returns C_ERR when there is
 * a protocol error: in such a case the client structure is setup to reply
 * with the error and close the connection. */
int processInlineBuffer(client *c, pendingCommand *pcmd) {
    char *newline;
    int argc, j, linefeed_chars = 1;
    sds *argv, aux;
    size_t querylen;

    /* Search for end of line */
    newline = strchr(c->querybuf+c->qb_pos,'\n');

    /* Nothing to do without a \r\n */
    if (newline == NULL) {
        if (sdslen(c->querybuf)-c->qb_pos > PROTO_INLINE_MAX_SIZE) {
            pcmd->read_error = CLIENT_READ_TOO_BIG_INLINE_REQUEST;
        }
        return C_ERR;
    }

    /* Handle the \r\n case. */
    if (newline != c->querybuf+c->qb_pos && *(newline-1) == '\r')
        newline--, linefeed_chars++;

    /* Split the input buffer up to the \r\n */
    querylen = newline-(c->querybuf+c->qb_pos);
    aux = sdsnewlen(c->querybuf+c->qb_pos,querylen);
    argv = sdssplitargs(aux,&argc);
    sdsfree(aux);
    if (argv == NULL) {
        pcmd->read_error = CLIENT_READ_UNBALANCED_QUOTES;
        return C_ERR;
    }

    /* Newline from slaves can be used to refresh the last ACK time.
     * This is useful for a slave to ping back while loading a big
     * RDB file. */
    if (querylen == 0 && clientTypeIsSlave(c)) {
        if (c->running_tid == IOTHREAD_MAIN_THREAD_ID)
            c->repl_ack_time = server.unixtime;
        else
            /* If this is a replica client running in an IO thread we cache the
             * last ack time in a different member variable in order to avoid
             * contention with main thread. f.e see refreshGoodSlavesCount()
             * Note c->repl_ack_time will still be updated in
             * updateClientDataFromIOThread with the value of c->io_repl_ack_time
             * when the client moves from IO to main thread. */
            c->io_repl_ack_time = server.unixtime;
    }

    /* Masters should never send us inline protocol to run actual
     * commands. If this happens, it is likely due to a bug in Redis where
     * we got some desynchronization in the protocol, for example
     * because of a PSYNC gone bad.
     *
     * However there is an exception: masters may send us just a newline
     * to keep the connection active. */
    if (querylen != 0 && c->flags & CLIENT_MASTER) {
        sdsfreesplitres(argv,argc);
        pcmd->read_error = CLIENT_READ_MASTER_USING_INLINE_PROTOCAL;
        return C_ERR;
    }

    /* Move querybuffer position to the next query in the buffer. */
    c->qb_pos += querylen+linefeed_chars;

    /* Setup argv array on client structure */
    if (argc) {
        /* Create new argv if space is insufficient. */
        if (argc > pcmd->argv_len) {
            zfree(pcmd->argv);
            pcmd->argv = zmalloc(sizeof(robj*)*argc);
            pcmd->argv_len = argc;
            pcmd->argv_len_sum = 0;
        }
    }

    /* Create redis objects for all arguments. */
    for (pcmd->argc = 0, j = 0; j < argc; j++) {
        /* Trim greedy allocation from sdssplitargs(). */
        argv[j] = sdsRemoveFreeSpace(argv[j], 0);
        pcmd->argv[pcmd->argc] = createObject(OBJ_STRING,argv[j]);
        pcmd->argc++;
        pcmd->argv_len_sum += sdslen(argv[j]);
        c->all_argv_len_sum += sdslen(argv[j]);
    }
    zfree(argv);

    /* Per-slot network bytes-in calculation.
     *
     * We calculate and store the current command's ingress bytes under
     * c->net_input_bytes_curr_cmd, for which its per-slot aggregation is deferred
     * until c->slot is parsed later within processCommand().
     *
     * Calculation: For inline buffer, every whitespace is of length 1,
     * with the exception of the trailing '\r\n' being length 2.
     *
     * For example;
     * Command) SET key value
     * Inline) SET key value\r\n
     */
    pcmd->input_bytes = (pcmd->argv_len_sum + (pcmd->argc - 1) + 2);

    return C_OK;
}

/* Helper function. Record protocol error details in server log,
 * and set the client as CLIENT_CLOSE_AFTER_REPLY and
 * CLIENT_PROTOCOL_ERROR. */
#define PROTO_DUMP_LEN 128
static void setProtocolError(const char *errstr, client *c) {
    if (server.verbosity <= LL_VERBOSE || c->flags & CLIENT_MASTER) {
        sds client = catClientInfoString(sdsempty(),c);

        /* Sample some protocol to given an idea about what was inside. */
        char buf[256];
        if (server.hide_user_data_from_log) {
            snprintf(buf,sizeof(buf),"Query buffer during protocol error: '*redacted*'");  
        } else if (sdslen(c->querybuf)-c->qb_pos < PROTO_DUMP_LEN) {
            snprintf(buf,sizeof(buf),"Query buffer during protocol error: '%s'", c->querybuf+c->qb_pos);  
        } else {
            snprintf(buf,sizeof(buf),"Query buffer during protocol error: '%.*s' (... more %zu bytes ...) '%.*s'", PROTO_DUMP_LEN/2, c->querybuf+c->qb_pos, sdslen(c->querybuf)-c->qb_pos-PROTO_DUMP_LEN, PROTO_DUMP_LEN/2, c->querybuf+sdslen(c->querybuf)-PROTO_DUMP_LEN/2);  
        }

        /* Remove non printable chars. */  
        if (!server.hide_user_data_from_log) {
            char *p = buf;
            while (*p != '\0') {
                if (!isprint(*p)) *p = '.';
                p++;
            }
        }

        /* Log all the client and protocol info. */
        int loglevel = (c->flags & CLIENT_MASTER) ? LL_WARNING :
                                                    LL_VERBOSE;
        serverLog(loglevel,
            "Protocol error (%s) from client: %s. %s", errstr, client, buf);
        sdsfree(client);
    }
    c->flags |= (CLIENT_CLOSE_AFTER_REPLY|CLIENT_PROTOCOL_ERROR);
}

/* Process the query buffer for client 'c', setting up the client argument
 * vector for command execution. Returns C_OK if after running the function
 * the client has a well-formed ready to be processed command, otherwise
 * C_ERR if there is still to read more buffer to get the full command.
 * The function also returns C_ERR when there is a protocol error: in such a
 * case the client structure is setup to reply with the error and close
 * the connection.
 *
 * This function is called if processInputBuffer() detects that the next
 * command is in RESP format, so the first byte in the command is found
 * to be '*'. Otherwise for inline commands processInlineBuffer() is called. */
static int processMultibulkBuffer(client *c, pendingCommand *pcmd) {
    char *newline = NULL;
    int ok;
    long long ll;
    size_t querybuf_len = sdslen(c->querybuf); /* Cache sdslen */

    if (c->multibulklen == 0) {
        /* The pending command should have been reset */
        serverAssertWithInfo(c,NULL,pcmd->argc == 0);

        /* Multi bulk length cannot be read without a \r\n */
        newline = strchr(c->querybuf+c->qb_pos,'\r');
        if (newline == NULL) {
            if (querybuf_len-c->qb_pos > PROTO_INLINE_MAX_SIZE) {
                pcmd->read_error = CLIENT_READ_TOO_BIG_MBULK_COUNT_STRING;
            }
            return C_ERR;
        }

        /* Buffer should also contain \n */
        if (newline-(c->querybuf+c->qb_pos) > (ssize_t)(querybuf_len-c->qb_pos-2))
            return C_ERR;

        /* We know for sure there is a whole line since newline != NULL,
         * so go ahead and find out the multi bulk length. */
        serverAssertWithInfo(c,NULL,c->querybuf[c->qb_pos] == '*');
        size_t multibulklen_slen = newline - (c->querybuf + 1 + c->qb_pos);
        ok = string2ll(c->querybuf+1+c->qb_pos,newline-(c->querybuf+1+c->qb_pos),&ll);
        if (!ok || ll > INT_MAX) {
            pcmd->read_error = CLIENT_READ_INVALID_MULTIBUCK_LENGTH;
            return C_ERR;
        } else if (ll > 10 && authRequired(c)) {
            pcmd->read_error = CLIENT_READ_UNAUTH_MBUCK_COUNT;
            return C_ERR;
        }

        c->qb_pos = (newline-c->querybuf)+2;

        if (ll <= 0) return C_OK;

        c->multibulklen = ll;
        c->bulklen = -1;

        /* Setup argv array on pending command structure.
         * Reallocate argv array when the requested size is greater than current size. */
        if (c->multibulklen > pcmd->argv_len) {
            zfree(pcmd->argv);
            pcmd->argv_len = min(c->multibulklen, 1024);
            pcmd->argv = zmalloc(sizeof(robj*)*(pcmd->argv_len));
            pcmd->argv_len_sum = 0;
        }

        /* Per-slot network bytes-in calculation.
         *
         * We calculate and store the current command's ingress bytes under
         * c->net_input_bytes_curr_cmd, for which its per-slot aggregation is deferred
         * until c->slot is parsed later within processCommand().
         *
         * Calculation: For multi bulk buffer, we accumulate four factors, namely;
         *
         * 1) multibulklen_slen + 3
         *    Cumulative string length (and not the value of) of multibulklen,
         *    including the first "*" byte and last "\r\n" 2 bytes from RESP.
         * 2) bulklen_slen + 3
         *    Cumulative string length (and not the value of) of bulklen,
         *    including +3 from RESP first "$" byte and last "\r\n" 2 bytes per argument count.
         * 3) c->argv_len_sum
         *    Cumulative string length of all argument vectors.
         * 4) c->argc * 2
         *    Cumulative string length of the arguments' white-spaces, for which there exists a total of
         *    "\r\n" 2 bytes per argument.
         *
         * For example;
         * Command) SET key value
         * RESP) *3\r\n$3\r\nSET\r\n$3\r\nkey\r\n$5\r\nvalue\r\n
         *
         * 1) String length of "*3\r\n" is 4, obtained from (multibulklen_slen + 3).
         * 2) String length of "$3\r\n" "$3\r\n" "$5\r\n" is 12, obtained from (bulklen_slen + 3).
         * 3) String length of "SET" "key" "value" is 11, obtained from (c->argv_len_sum).
         * 4) String length of the 3 arguments' white-spaces "\r\n" is 6, obtained from (c->argc * 2).
         *
         * The 1st component is calculated within the below line.
         * */
        pcmd->input_bytes += (multibulklen_slen + 3);
    }

    serverAssertWithInfo(c,NULL,c->multibulklen > 0);
    while(c->multibulklen) {
        /* Read bulk length if unknown */
        if (c->bulklen == -1) {
            newline = memchr(c->querybuf+c->qb_pos,'\r',sdslen(c->querybuf) - c->qb_pos);
            if (newline == NULL) {
                if (querybuf_len-c->qb_pos > PROTO_INLINE_MAX_SIZE) {
                    pcmd->read_error = CLIENT_READ_TOO_BIG_BUCK_COUNT_STRING;
                    return C_ERR;
                }
                break;
            }

            /* Buffer should also contain \n */
            if (newline-(c->querybuf+c->qb_pos) > (ssize_t)(querybuf_len-c->qb_pos-2))
                break;

            if (c->querybuf[c->qb_pos] != '$') {
                pcmd->read_error = CLIENT_READ_EXPECTED_DOLLAR;
                return C_ERR;
            }

            size_t bulklen_slen = newline - (c->querybuf + c->qb_pos + 1);
            ok = string2ll(c->querybuf+c->qb_pos+1,newline-(c->querybuf+c->qb_pos+1),&ll);
            if (!ok || ll < 0 ||
                (!(c->flags & CLIENT_MASTER) && ll > server.proto_max_bulk_len)) {
                pcmd->read_error = CLIENT_READ_INVALID_BUCK_LENGTH;
                return C_ERR;
            } else if (ll > 16384 && authRequired(c)) {
                pcmd->read_error = CLIENT_READ_UNAUTH_BUCK_LENGTH;
                return C_ERR;
            }

            c->qb_pos = newline-c->querybuf+2;
            if (!(c->flags & CLIENT_MASTER) && ll >= PROTO_MBULK_BIG_ARG) {
                /* When the client is not a master client (because master
                 * client's querybuf can only be trimmed after data applied
                 * and sent to replicas).
                 *
                 * 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 optimize object creation
                 * avoiding a large copy of data.
                 *
                 * But only when the data we have not parsed is less than
                 * or equal to ll+2. If the data length is greater than
                 * ll+2, trimming querybuf is just a waste of time, because
                 * at this time the querybuf contains not only our bulk. */
                if (querybuf_len-c->qb_pos <= (size_t)ll+2) {
                    sdsrange(c->querybuf,c->qb_pos,-1);
                    querybuf_len = sdslen(c->querybuf);
                    c->qb_pos = 0;
                    /* Hint the sds library about the amount of bytes this string is
                     * going to contain. */
                    c->querybuf = sdsMakeRoomForNonGreedy(c->querybuf,ll+2-querybuf_len);
                    /* We later set the peak to the used portion of the buffer, but here we over
                     * allocated because we know what we need, make sure it'll not be shrunk before used. */
                    if (c->querybuf_peak < (size_t)ll + 2) c->querybuf_peak = ll + 2;
                    querybuf_len = sdslen(c->querybuf); /* Update cached length */
                }
            }
            c->bulklen = ll;
            /* Per-slot network bytes-in calculation, 2nd component. */
            pcmd->input_bytes += (bulklen_slen + 3);
        } else {
            serverAssert(pcmd->flags & PENDING_CMD_FLAG_INCOMPLETE);
        }

        /* Read bulk argument */
        if (querybuf_len-c->qb_pos < (size_t)(c->bulklen+2)) {
            break;
        } else {
            /* Check if we have space in argv, grow if needed */
            if (pcmd->argc >= pcmd->argv_len) {
                pcmd->argv_len = min(pcmd->argv_len < INT_MAX/2 ? (pcmd->argv_len)*2 : INT_MAX, pcmd->argc+c->multibulklen);
                pcmd->argv = zrealloc(pcmd->argv, sizeof(robj*)*(pcmd->argv_len));
            }

            /* Optimization: if a non-master client's buffer contains JUST our bulk element
             * instead of creating a new object by *copying* the sds we
             * just use the current sds string. */
            if (!(c->flags & CLIENT_MASTER) &&
                c->qb_pos == 0 &&
                c->bulklen >= PROTO_MBULK_BIG_ARG &&
                querybuf_len == (size_t)(c->bulklen+2))
            {
                (pcmd->argv)[(pcmd->argc)++] = createObject(OBJ_STRING,c->querybuf);
                pcmd->argv_len_sum += c->bulklen;
                c->all_argv_len_sum += c->bulklen;
                sdsIncrLen(c->querybuf,-2); /* remove CRLF */
                /* Assume that if we saw a fat argument we'll see another one likely...
                 * But only if that fat argument is not too big compared to the memory limit. */
                if (!server.maxmemory || (size_t)c->bulklen < server.maxmemory / 32) {
                    c->querybuf = sdsnewlen(SDS_NOINIT,c->bulklen+2);
                } else {
                    c->querybuf = sdsnewlen(SDS_NOINIT, PROTO_IOBUF_LEN);
                }
                sdsclear(c->querybuf);
                querybuf_len = sdslen(c->querybuf); /* Update cached length */
            } else {
                (pcmd->argv)[(pcmd->argc)++] =
                    createStringObject(c->querybuf+c->qb_pos,c->bulklen);
                pcmd->argv_len_sum += c->bulklen;
                c->all_argv_len_sum += c->bulklen;
                c->qb_pos += c->bulklen+2;
            }
            c->bulklen = -1;
            c->multibulklen--;
        }
    }

    /* We're done when c->multibulk == 0 */
    if (c->multibulklen == 0) {
        /* Per-slot network bytes-in calculation, 3rd and 4th components. */
        pcmd->input_bytes += (pcmd->argv_len_sum + (pcmd->argc * 2));
        pcmd->flags &= ~PENDING_CMD_FLAG_INCOMPLETE;
        return C_OK;
    }

    /* Still not ready to process the command */
    pcmd->flags |= PENDING_CMD_FLAG_INCOMPLETE;
    return C_OK;
}

/* Prepare the client for executing the next command:
 *
 * 1. Append the response, if necessary.
 * 2. Reset the client.
 * 3. Update the all_argv_len_sum counter and advance the pending_cmd cyclic buffer.
 * 4. Update the cluster slot stats, if necessary.
 */
void prepareForNextCommand(client *c, int update_slot_stats) {
    reqresAppendResponse(c);
    if (update_slot_stats) {
        /* We should do this before reset client. */
        clusterSlotStatsAddNetworkBytesInForUserClient(c);
    }
    resetClientInternal(c, 1);
}

/* Perform necessary tasks after a command was executed:
 *
 * 1. The client is reset unless there are reasons to avoid doing it.
 * 2. In the case of master clients, the replication offset is updated.
 * 3. Propagate commands we got from our master to replicas down the line. */
void commandProcessed(client *c) {
    /* If client is blocked(including paused), just return avoid reset and replicate.
     *
     * 1. Don't reset the client structure for blocked clients, so that the reply
     *    callback will still be able to access the client argv and argc fields.
     *    The client will be reset in unblockClient().
     * 2. Don't update replication offset or propagate commands to replicas,
     *    since we have not applied the command. */
    if (c->flags & CLIENT_BLOCKED) return;

    prepareForNextCommand(c, 1);

    long long prev_offset = c->reploff;
    if (c->flags & CLIENT_MASTER && !(c->flags & CLIENT_MULTI)) {
        /* Update the applied replication offset of our master. */
        serverAssert(c->reploff_next > 0);
        c->reploff = c->reploff_next;
    }

    /* If the client is a master we need to compute the difference
     * between the applied offset before and after processing the buffer,
     * to understand how much of the replication stream was actually
     * applied to the master state: this quantity, and its corresponding
     * part of the replication stream, will be propagated to the
     * sub-replicas and to the replication backlog. */
    if (c->flags & CLIENT_MASTER) {
        long long applied = c->reploff - prev_offset;
        if (applied) {
            replicationFeedStreamFromMasterStream(c->querybuf+c->repl_applied,applied);
            c->repl_applied += applied;

            /* Update the atomic slot migration task's applied bytes. */
            if (c->flags & CLIENT_ASM_IMPORTING)
                asmImportIncrAppliedBytes(c->task, applied);
        }
    }
}

/* This function calls processCommand(), but also performs a few sub tasks
 * for the client that are useful in that context:
 *
 * 1. It sets the current client to the client 'c'.
 * 2. calls commandProcessed() if the command was handled.
 *
 * The function returns C_ERR in case the client was freed as a side effect
 * of processing the command, otherwise C_OK is returned. */
int processCommandAndResetClient(client *c) {
    int deadclient = 0;
    client *old_client = server.current_client;
    server.current_client = c;
    if (processCommand(c) == C_OK) {
        commandProcessed(c);
        /* Update the client's memory to include output buffer growth following the
         * processed command. */
        if (c->conn) updateClientMemUsageAndBucket(c);
    }

    if (server.current_client == NULL) deadclient = 1;
    /*
     * Restore the old client, this is needed because when a script
     * times out, we will get into this code from processEventsWhileBlocked.
     * Which will cause to set the server.current_client. If not restored
     * we will return 1 to our caller which will falsely indicate the client
     * is dead and will stop reading from its buffer.
     */
    server.current_client = old_client;
    /* performEvictions may flush slave output buffers. This may
     * result in a slave, that may be the active client, to be
     * freed. */
    return deadclient ? C_ERR : C_OK;
}


/* This function will execute any fully parsed commands pending on
 * the client. Returns C_ERR if the client is no longer valid after executing
 * the command, and C_OK for all other cases. */
int processPendingCommandAndInputBuffer(client *c) {
    /* Notice, this code is also called from 'processUnblockedClients'.
     * But in case of a module blocked client (see RM_Call 'K' flag) we do not reach this code path.
     * So whenever we change the code here we need to consider if we need this change on module
     * blocked client as well */
    if (c->flags & CLIENT_PENDING_COMMAND) {
        c->flags &= ~CLIENT_PENDING_COMMAND;
        if (processCommandAndResetClient(c) == C_ERR) {
            return C_ERR;
        }
    }

    /* Now process client if it has more data in it's buffer.
     *
     * Note: when a master client steps into this function,
     * it can always satisfy this condition, because its querybuf
     * contains data not applied. */
    if ((c->querybuf && sdslen(c->querybuf) > 0) || c->pending_cmds.ready_len > 0) {
        return processInputBuffer(c);
    }
    return C_OK;
}

void handleClientReadError(client *c) {
    switch (c->read_error) {
        case CLIENT_READ_TOO_BIG_INLINE_REQUEST:
            addReplyError(c,"Protocol error: too big inline request");
            setProtocolError("too big inline request",c);
            break;
        case CLIENT_READ_UNBALANCED_QUOTES:
            addReplyError(c,"Protocol error: unbalanced quotes in request");
            setProtocolError("unbalanced quotes in request",c);
            break;
        case CLIENT_READ_MASTER_USING_INLINE_PROTOCAL:
            serverLog(LL_WARNING,"WARNING: Receiving inline protocol from master, master stream corruption? Closing the master connection and discarding the cached master.");
            setProtocolError("Master using the inline protocol. Desync?",c);
            break;
        case CLIENT_READ_TOO_BIG_MBULK_COUNT_STRING:
            addReplyError(c,"Protocol error: too big mbulk count string");
            setProtocolError("too big mbulk count string",c);
            break;
        case CLIENT_READ_TOO_BIG_BUCK_COUNT_STRING:
            addReplyError(c, "Protocol error: too big bulk count string");
            setProtocolError("too big bulk count string",c);
            break;
        case CLIENT_READ_EXPECTED_DOLLAR:
            addReplyErrorFormat(c,
                "Protocol error: expected '$', got '%c'",
                c->querybuf[c->qb_pos]);
            setProtocolError("expected $ but got something else",c);
            break;
        case CLIENT_READ_INVALID_BUCK_LENGTH:
            addReplyError(c,"Protocol error: invalid bulk length");
            setProtocolError("invalid bulk length",c);
            break;
        case CLIENT_READ_UNAUTH_BUCK_LENGTH:
            addReplyError(c, "Protocol error: unauthenticated bulk length");
            setProtocolError("unauth bulk length", c);
            break;
        case CLIENT_READ_INVALID_MULTIBUCK_LENGTH:
            addReplyError(c,"Protocol error: invalid multibulk length");
            setProtocolError("invalid mbulk count",c);
            break;
        case CLIENT_READ_UNAUTH_MBUCK_COUNT:
            addReplyError(c, "Protocol error: unauthenticated multibulk length");
            setProtocolError("unauth mbulk count", c);
            break;
        case CLIENT_READ_CONN_DISCONNECTED:
            serverLog(LL_VERBOSE, "Reading from client: %s",connGetLastError(c->conn));
            break;
        case CLIENT_READ_CONN_CLOSED:
            if (server.verbosity <= LL_VERBOSE) {
                sds info = catClientInfoString(sdsempty(), c);
                serverLog(LL_VERBOSE, "Client closed connection %s", info);
                sdsfree(info);
            }
            break;
        case CLIENT_READ_REACHED_MAX_QUERYBUF: {
            sds ci = catClientInfoString(sdsempty(),c), bytes = sdsempty();
            bytes = sdscatrepr(bytes,c->querybuf,64);
            serverLog(LL_WARNING,"Closing client that reached max query buffer length: %s (qbuf initial bytes: %s)", ci, bytes);
            sdsfree(ci);
            sdsfree(bytes);
            break;
        }
        default:
            serverPanic("Unknown client read error: %d", c->read_error);
            break;
    }
}


/* Helper function to check if a read error is fatal (should stop processing) */
int isClientReadErrorFatal(client *c) {
    return c->read_error != 0 &&
           c->read_error != CLIENT_READ_COMMAND_NOT_FOUND &&
           c->read_error != CLIENT_READ_BAD_ARITY &&
           c->read_error != CLIENT_READ_CROSS_SLOT;
}

/* This function is called every time, in the client structure 'c', there is
 * more query buffer to process, because we read more data from the socket
 * or because a client was blocked and later reactivated, so there could be
 * pending query buffer, already representing a full command, to process.
 * return C_ERR in case the client was freed during the processing */
int processInputBuffer(client *c) {
    atomicIncr(server.stat_total_client_process_input_buff_events, 1);

    /* Keep active-client window updates on main-thread paths only (here and
     * in IO-thread handoff processing) to avoid races with serverCron()
     * maintenance of the circular slots. */
    if (c->running_tid == IOTHREAD_MAIN_THREAD_ID)
        statsUpdateActiveClients(c);

    /* We limit the lookahead for unauthenticated connections to 1.
     * This is both to reduce memory overhead, and to prevent errors: AUTH can
     * affect the handling of succeeding commands. Parsing of "large"
     * unauthenticated multibulk commands is rejected, which would cause those
     * commands to incorrectly return an error to the client. */
    const int lookahead = authRequired(c) ? 1 : server.lookahead;

    /* Keep processing while there is something in the input buffer */
    while ((c->querybuf && c->qb_pos < sdslen(c->querybuf)) ||
           c->pending_cmds.ready_len > 0)
    {
        /* Immediately abort if the client is in the middle of something. */
        if (c->flags & CLIENT_BLOCKED || c->flags & CLIENT_UNBLOCKED) break;

        /* Don't process more buffers from clients that have already pending
         * commands to execute in c->argv. */
        if (c->flags & CLIENT_PENDING_COMMAND) break;

        /* Don't process input from the master while there is a busy script
         * condition on the slave. We want just to accumulate the replication
         * stream (instead of replying -BUSY like we do with other clients) and
         * later resume the processing. */
        if (c->flags & CLIENT_MASTER && isInsideYieldingLongCommand()) break;

        /* CLIENT_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).
         *
         * The same applies for clients we want to terminate ASAP. */
        if (c->flags & (CLIENT_CLOSE_AFTER_REPLY|CLIENT_CLOSE_ASAP)) break;

        /* Determine if we need to parse more commands from the query buffer.
         * Only parse when there are no ready commands waiting to be processed. */
        const int parse_more = !c->pending_cmds.ready_len;
        int pending_cmd_before_reading = c->pending_cmds.ready_len;

        /* Parse up to lookahead commands only if we don't have enough ready commands */
        while (parse_more && c->pending_cmds.ready_len < lookahead &&
               c->querybuf && c->qb_pos < sdslen(c->querybuf))
        {
            /* Determine request type when unknown. */
            if (!c->reqtype) {
                if (c->querybuf[c->qb_pos] == '*') {
                    c->reqtype = PROTO_REQ_MULTIBULK;
                } else {
                    c->reqtype = PROTO_REQ_INLINE;
                }
            }

            pendingCommand *pcmd = NULL;
            if (c->reqtype == PROTO_REQ_INLINE) {
                pcmd = acquirePendingCommand();
                if (processInlineBuffer(c, pcmd) == C_ERR && !pcmd->read_error) {
                    /* If it fails but there are no errors, it means that it might just be
                     * that the desired content cannot be parsed. At this point, we exit and wait for the next time. */
                    freePendingCommand(c, pcmd);
                    break;
                }
            } else if (c->reqtype == PROTO_REQ_MULTIBULK) {
                int incomplete = (c->pending_cmds.len != c->pending_cmds.ready_len);
                if (unlikely(incomplete)) {
                    pcmd = popPendingCommandFromTail(&c->pending_cmds);
                } else {
                    pcmd = acquirePendingCommand();
                }

                if (processMultibulkBuffer(c, pcmd) == C_ERR && !pcmd->read_error) {
                    /* If it fails but there are no errors, it means that it might just be
                     * that the desired content cannot be parsed. At this point, we exit and wait for the next time. */
                    freePendingCommand(c, pcmd);
                    break;
                }
            } else {
                serverPanic("Unknown request type");
            }

            addPendingCommand(&c->pending_cmds, pcmd);
            if (unlikely(pcmd->read_error || (pcmd->flags & PENDING_CMD_FLAG_INCOMPLETE)))
                break;

            if (c->running_tid == IOTHREAD_MAIN_THREAD_ID)
                pcmd->reploff = c->read_reploff - sdslen(c->querybuf) + c->qb_pos;
            else
                pcmd->reploff = c->io_read_reploff - sdslen(c->querybuf) + c->qb_pos;

            preprocessCommand(c, pcmd);
            pcmd->flags |= PENDING_CMD_FLAG_PREPROCESSED;
            resetClientQbufState(c);
        }

        if (c->pending_cmds.ready_len != pending_cmd_before_reading) {
            int newly_parsed_cmds = c->pending_cmds.ready_len - pending_cmd_before_reading;
            atomicIncr(server.stat_avg_pipeline_length_sum, newly_parsed_cmds);
            atomicIncr(server.stat_avg_pipeline_length_cnt, 1);

            c->stat_avg_pipeline_length_sum += newly_parsed_cmds;
            c->stat_avg_pipeline_length_cnt++;
        }

        /* Try to consume the next ready command from the pending command list. */
        if (!c->pending_cmds.ready_len)
            break;
        pendingCommand *curcmd = c->pending_cmds.head;

        /* We populate the old client fields so we don't have to modify all existing logic to work with pendingCommands */
        c->argc = curcmd->argc;
        c->argv = curcmd->argv;
        c->argv_len = curcmd->argv_len;
        c->net_input_bytes_curr_cmd += curcmd->input_bytes;
        c->reploff_next = curcmd->reploff;
        c->slot = curcmd->slot;
        c->lookedcmd = curcmd->cmd;
        c->read_error = curcmd->read_error;
        c->current_pending_cmd = curcmd;

        /* Prefetch the command only when more commands have been parsed and we
         * are in the main thread. If running in an IO thread, prefetch will be
         * deferred until the client is processed by the main thread. Skip prefetch
         * if there are too few commands to avoid meaningless prefetching. */
        if (parse_more && c->running_tid == IOTHREAD_MAIN_THREAD_ID &&
            c->pending_cmds.ready_len > 1)
        {
            /* Prefetch the commands. */
            resetCommandsBatch();
            addCommandToBatch(c);
            prefetchCommands();
        }

        /* Check if the client has a fatal read error that requires stopping processing. */
        if (isClientReadErrorFatal(c)) {
            if (c->running_tid != IOTHREAD_MAIN_THREAD_ID) {
                enqueuePendingClientsToMainThread(c, 0);
            }
            break;
        }

        /* Multibulk processing could see a <= 0 length. */
        if (!c->argc) {
            /* A naked newline can be sent from masters as a keep-alive, or from slaves to refresh
             * the last ACK time. In that case there's no command to actually execute. */
            prepareForNextCommand(c, 0);
        } else {
            /* If we are in the context of an I/O thread, we can't really
             * execute the command here. All we can do is to flag the client
             * as one that needs to process the command. */
            if (c->running_tid != IOTHREAD_MAIN_THREAD_ID) {
                c->io_flags |= CLIENT_IO_PENDING_COMMAND;
                enqueuePendingClientsToMainThread(c, 0);
                break;
            }

            /* We are finally ready to execute the command. */
            if (processCommandAndResetClient(c) == C_ERR) {
                /* If the client is no longer valid, we avoid exiting this
                 * loop and trimming the client buffer later. So we return
                 * ASAP in that case. */
                return C_ERR;
            }
        }
    }

    if (c->flags & CLIENT_MASTER) {
        /* If the client is a master, trim the querybuf to repl_applied,
         * since master client is very special, its querybuf not only
         * used to parse command, but also proxy to sub-replicas.
         *
         * Here are some scenarios we cannot trim to qb_pos:
         * 1. we don't receive complete command from master
         * 2. master client blocked cause of client pause
         * 3. io threads operate read, master client flagged with CLIENT_PENDING_COMMAND
         *
         * In these scenarios, qb_pos points to the part of the current command
         * or the beginning of next command, and the current command is not applied yet,
         * so the repl_applied is not equal to qb_pos. */
        if (c->repl_applied) {
            sdsrange(c->querybuf,c->repl_applied,-1);
            serverAssert(c->qb_pos >= (size_t)c->repl_applied);
            c->qb_pos -= c->repl_applied;
            c->repl_applied = 0;
        }
    } else if (c->qb_pos) {
        /* Trim to pos */
        sdsrange(c->querybuf,c->qb_pos,-1);
        c->qb_pos = 0;
    }

    /* Update client memory usage after processing the query buffer, this is
     * important in case the query buffer is big and wasn't drained during
     * the above loop (because of partially sent big commands). */
    if (c->running_tid == IOTHREAD_MAIN_THREAD_ID)
        updateClientMemUsageAndBucket(c);

    return C_OK;
}

void readQueryFromClient(connection *conn) {
    client *c = connGetPrivateData(conn);
    int nread, big_arg = 0;
    size_t qblen, readlen;

    /* We have to read compressed data but compression for this client is
     * currently disabled. This could happened f.e when client was just fetched
     * to main thread. */
    int pending_compression_read = c->compression_state &&
                                   !(c->io_flags & CLIENT_IO_COMPRESSION_ENABLED);

    if (!(c->io_flags & CLIENT_IO_READ_ENABLED) || pending_compression_read) {
        atomicSetWithSync(c->pending_read, 1);
        return;
    } else if (server.io_threads_num > 1) {
        atomicSetWithSync(c->pending_read, 0);
    }

    c->read_error = 0;

    c->stat_total_read_events++;

    /* Update the number of reads of io threads on server */
    atomicIncr(IOThreads[c->running_tid].io_reads_processed, 1);

    readlen = PROTO_IOBUF_LEN;
    /* If this is a multi bulk request, and we are processing a bulk reply
     * that is large enough, try to maximize the probability that the query
     * buffer contains exactly the SDS string representing the object, even
     * at the risk of requiring more read(2) calls. This way the function
     * processMultiBulkBuffer() can avoid copying buffers to create the
     * Redis Object representing the argument. */
    if (c->reqtype == PROTO_REQ_MULTIBULK && c->multibulklen && c->bulklen != -1
        && c->bulklen >= PROTO_MBULK_BIG_ARG)
    {
        /* For big argv, the client always uses its private query buffer.
         * Using the reusable query buffer would eventually expand it beyond 32k,
         * causing the client to take ownership of the reusable query buffer. */
        if (!c->querybuf) c->querybuf = sdsempty();

        ssize_t remaining = (size_t)(c->bulklen+2)-(sdslen(c->querybuf)-c->qb_pos);
        big_arg = 1;

        /* Note that the 'remaining' variable may be zero in some edge case,
         * for example once we resume a blocked client after CLIENT PAUSE. */
        if (remaining > 0) readlen = remaining;

        /* Master client needs expand the readlen when meet BIG_ARG(see #9100),
         * but doesn't need align to the next arg, we can read more data. */
        if (c->flags & CLIENT_MASTER && readlen < PROTO_IOBUF_LEN)
            readlen = PROTO_IOBUF_LEN;
    } else if (c->querybuf == NULL) {
        if (unlikely(thread_reusable_qb_used)) {
            /* The reusable query buffer is already used by another client,
             * switch to using the client's private query buffer. This only
             * occurs when commands are executed nested via processEventsWhileBlocked(). */
            c->querybuf = sdsnewlen(NULL, PROTO_IOBUF_LEN);
            sdsclear(c->querybuf);
        } else {
            /* Create the reusable query buffer if it doesn't exist. */
            if (!thread_reusable_qb) {
                thread_reusable_qb = sdsnewlen(NULL, PROTO_IOBUF_LEN);
                sdsclear(thread_reusable_qb);
            }

            /* Assign the reusable query buffer to the client and mark it as in use. */
            serverAssert(sdslen(thread_reusable_qb) == 0);
            c->querybuf = thread_reusable_qb;
            c->io_flags |= CLIENT_IO_REUSABLE_QUERYBUFFER;
            thread_reusable_qb_used = 1;
        }
    }

    qblen = sdslen(c->querybuf);
    if (!(c->flags & CLIENT_MASTER) && // master client's querybuf can grow greedy.
        (big_arg || sdsalloc(c->querybuf) < PROTO_IOBUF_LEN)) {
        /* When reading a BIG_ARG we won't be reading more than that one arg
         * into the query buffer, so we don't need to pre-allocate more than we
         * need, so using the non-greedy growing. For an initial allocation of
         * the query buffer, we also don't wanna use the greedy growth, in order
         * to avoid collision with the RESIZE_THRESHOLD mechanism. */
        c->querybuf = sdsMakeRoomForNonGreedy(c->querybuf, readlen);
        /* We later set the peak to the used portion of the buffer, but here we over
         * allocated because we know what we need, make sure it'll not be shrunk before used. */
        if (c->querybuf_peak < qblen + readlen) c->querybuf_peak = qblen + readlen;
    } else {
        c->querybuf = sdsMakeRoomFor(c->querybuf, readlen);

        /* Read as much as possible from the socket to save read(2) system calls. */
        readlen = sdsavail(c->querybuf);
    }
    /* Compression is applied at the client level (see client_comp.c): when
     * decompression is active `nread` is the number of decompressed bytes placed
     * in the query buffer while `network_read` is the number of bytes actually
     * read from the socket. Without compression they are equal. */
    int network_read = 0;
    nread = clientConnRead(c, c->querybuf+qblen, readlen, &network_read);
    if (nread == -1) {
        if (connGetState(conn) == CONN_STATE_CONNECTED) {
            goto done;
        } else {
            c->read_error = CLIENT_READ_CONN_DISCONNECTED;
            freeClientAsync(c);
            goto done;
        }
    } else if (nread == 0) {
        c->read_error = CLIENT_READ_CONN_CLOSED;
        freeClientAsync(c);
        goto done;
    }

    sdsIncrLen(c->querybuf,nread);
    qblen = sdslen(c->querybuf);
    if (c->querybuf_peak < qblen) c->querybuf_peak = qblen;

    if (!(c->flags & CLIENT_MASTER) || c->running_tid == IOTHREAD_MAIN_THREAD_ID)
        c->lastinteraction = server.unixtime;
    else
        /* Avoid contention with genRedisInfoString as it can access master
         * client's data. If this is a master running in IO thread the value of
         * c->lastinteraction will be updated during processClientsFromIOThread */
        c->io_lastinteraction = server.unixtime;

    if (c->flags & CLIENT_MASTER) {
        if (c->running_tid == IOTHREAD_MAIN_THREAD_ID) {
            c->read_reploff += nread;
        } else {
            /* Same comment as for c->io_lastinteraction */
            c->io_read_reploff += nread;
        }
        atomicIncr(server.stat_net_repl_input_bytes, network_read);
    } else {
        atomicIncr(server.stat_net_input_bytes, network_read);
    }
    c->net_input_bytes += network_read;

    if (!(c->flags & CLIENT_MASTER) &&
        /* The commands cached in the MULTI/EXEC queue have not been executed yet,
         * so they are also considered a part of the query buffer in a broader sense.
         *
         * For unauthenticated clients, the query buffer cannot exceed 1MB at most. */
        (c->mstate.argv_len_sums + sdslen(c->querybuf) > server.client_max_querybuf_len ||
         (c->mstate.argv_len_sums + sdslen(c->querybuf) > 1024*1024 && authRequired(c))))
    {
        c->read_error = CLIENT_READ_REACHED_MAX_QUERYBUF;
        freeClientAsync(c);
        atomicIncr(server.stat_client_qbuf_limit_disconnections, 1);
        goto done;
    }

    /* There is more data in the client input buffer, continue parsing it
     * and check if there is a full command to execute. */
    if (processInputBuffer(c) == C_ERR)
         c = NULL;

done:
    if (c && isClientReadErrorFatal(c)) {
        if (c->running_tid == IOTHREAD_MAIN_THREAD_ID) {
            handleClientReadError(c);
        }
    }

    if (c && (c->io_flags & CLIENT_IO_REUSABLE_QUERYBUFFER)) {
        serverAssert(c->qb_pos == 0); /* Ensure the client's query buffer is trimmed in processInputBuffer */
        resetReusableQueryBuf(c);
    }
    beforeNextClient(c);
}

/* A Redis "Address String" is a colon separated ip:port pair.
 * For IPv4 it's in the form x.y.z.k:port, example: "127.0.0.1:1234".
 * For IPv6 addresses we use [] around the IP part, like in "[::1]:1234".
 * For Unix sockets we use path:0, like in "/tmp/redis:0".
 *
 * An Address String always fits inside a buffer of NET_ADDR_STR_LEN bytes,
 * including the null term.
 *
 * On failure the function still populates 'addr' with the "?:0" string in case
 * you want to relax error checking or need to display something anyway (see
 * anetFdToString implementation for more info). */
void genClientAddrString(client *client, char *addr,
                         size_t addr_len, int remote) {
    if (client->flags & CLIENT_UNIX_SOCKET) {
        /* Unix socket client. */
        snprintf(addr,addr_len,"%s:0",server.unixsocket);
    } else {
        /* TCP client. */
        connFormatAddr(client->conn,addr,addr_len,remote);
    }
}

/* This function returns the client peer id, by creating and caching it
 * if client->peerid is NULL, otherwise returning the cached value.
 * The Peer ID never changes during the life of the client, however it
 * is expensive to compute. */
char *getClientPeerId(client *c) {
    char peerid[NET_ADDR_STR_LEN] = {0};

    if (c->peerid == NULL) {
        genClientAddrString(c,peerid,sizeof(peerid),1);
        c->peerid = sdsnew(peerid);
    }
    return c->peerid;
}

/* This function returns the client bound socket name, by creating and caching
 * it if client->sockname is NULL, otherwise returning the cached value.
 * The Socket Name never changes during the life of the client, however it
 * is expensive to compute. */
char *getClientSockname(client *c) {
    char sockname[NET_ADDR_STR_LEN] = {0};

    if (c->sockname == NULL) {
        genClientAddrString(c,sockname,sizeof(sockname),0);
        c->sockname = sdsnew(sockname);
    }
    return c->sockname;
}

static inline int isCrashing(void) {
    int crashing;
    atomicGet(server.crashing, crashing);
    return crashing;
}

/* Concatenate a string representing the state of a client in a human
 * readable format, into the sds string 's'. */
sds catClientInfoString(sds s, client *client) {
    char flags[17], events[3], conninfo[CONN_INFO_LEN], *p;

    /* Pause IO thread to access data of the client safely. */
    int paused = 0;
    if (client->running_tid != IOTHREAD_MAIN_THREAD_ID &&
        pthread_equal(server.main_thread_id, pthread_self()) &&
        !isCrashing())
    {
        paused = 1;
        pauseIOThread(client->running_tid);
    }

    p = flags;
    if (client->flags & CLIENT_SLAVE) {
        if (client->flags & CLIENT_MONITOR)
            *p++ = 'O';
        else if (client->flags & CLIENT_ASM_MIGRATING)
            *p++ = 'g';
        else
            *p++ = 'S';
    }
    if (client->flags & CLIENT_MASTER) {
        if (client->flags & CLIENT_ASM_IMPORTING)
            *p++ = 'o';
        else
            *p++ = 'M';
    }
    if (client->flags & CLIENT_PUBSUB) *p++ = 'P';
    if (client->flags & CLIENT_MULTI) *p++ = 'x';
    if (client->flags & CLIENT_BLOCKED) *p++ = 'b';
    if (client->flags & CLIENT_TRACKING) *p++ = 't';
    if (client->flags & CLIENT_TRACKING_BROKEN_REDIR) *p++ = 'R';
    if (client->flags & CLIENT_TRACKING_BCAST) *p++ = 'B';
    if (client->flags & CLIENT_DIRTY_CAS) *p++ = 'd';
    if (client->flags & CLIENT_CLOSE_AFTER_REPLY) *p++ = 'c';
    if (client->flags & CLIENT_UNBLOCKED) *p++ = 'u';
    if (client->flags & CLIENT_CLOSE_ASAP) *p++ = 'A';
    if (client->flags & CLIENT_UNIX_SOCKET) *p++ = 'U';
    if (client->flags & CLIENT_READONLY) *p++ = 'r';
    if (client->flags & CLIENT_NO_EVICT) *p++ = 'e';
    if (client->flags & CLIENT_NO_TOUCH) *p++ = 'T';
    if (client->flags & CLIENT_REPL_RDB_CHANNEL) *p++ = 'C';
    if (client->flags & CLIENT_INTERNAL) *p++ = 'I';
    if (p == flags) *p++ = 'N';
    *p++ = '\0';

    p = events;
    if (client->conn) {
        if (connHasReadHandler(client->conn)) *p++ = 'r';
        if (connHasWriteHandler(client->conn)) *p++ = 'w';
    }
    *p = '\0';

    /* Refresh the cached unshared reply bytes before computing memory stats below. */
    updateClientUnsharedReplyBytes(client);

    /* Compute the total memory consumed by this client. */
    size_t obufmem = getClientOutputBufferLogicalSize(client);
    size_t total_mem = getClientMemoryUsage(client);

    size_t used_blocks_of_repl_buf = 0;
    if (client->ref_repl_buf_node) {
        replBufBlock *last = listNodeValue(listLast(server.repl_buffer_blocks));
        replBufBlock *cur = listNodeValue(client->ref_repl_buf_node);
        used_blocks_of_repl_buf = last->id - cur->id + 1;
    }

    sds ret = sdscatfmt(s, FMTARGS(
        "id=%U", (unsigned long long) client->id,
        " addr=%s", getClientPeerId(client),
        " laddr=%s", getClientSockname(client),
        " %s", connGetInfo(client->conn, conninfo, sizeof(conninfo)),
        " name=%s", client->name ? (char*)client->name->ptr : "",
        " age=%I", (long long)(commandTimeSnapshot() / 1000 - client->ctime),
        " idle=%I", (long long)(server.unixtime - client->lastinteraction),
        " flags=%s", flags,
        " db=%i", client->db->id,
        " sub=%i", (int) dictSize(client->pubsub_channels),
        " psub=%i", (int) dictSize(client->pubsub_patterns),
        " ssub=%i", (int) dictSize(client->pubsubshard_channels),
        " multi=%i", (client->flags & CLIENT_MULTI) ? client->mstate.count : -1,
        " watch=%i", (int) listLength(client->watched_keys),
        " qbuf=%U", client->querybuf ? (unsigned long long) sdslen(client->querybuf) : 0,
        " qbuf-free=%U", client->querybuf ? (unsigned long long) sdsavail(client->querybuf) : 0,
        " argv-mem=%U", (unsigned long long) client->all_argv_len_sum,
        " multi-mem=%U", (unsigned long long) client->mstate.argv_len_sums,
        " rbs=%U", (unsigned long long) client->buf_usable_size,
        " rbp=%U", (unsigned long long) client->buf_peak,
        " obl=%U", (unsigned long long) client->bufpos,
        " oll=%U", (unsigned long long) listLength(client->reply) + used_blocks_of_repl_buf,
        " omem=%U", (unsigned long long) obufmem, /* logical output buffer memory (includes shared memory; excludes client->buf so static clients show 0) */
        " omem-shared=%U", (unsigned long long) client->reply_bytes_shared, /* shared memory (not solely owned by this client) */
        " omem-unshared=%U", (unsigned long long) client->reply_bytes_unshared, /* unshared memory (solely owned by this client) */
        " tot-mem=%U", (unsigned long long) total_mem, /* actual memory usage (includes unshared memory, excludes shared memory) */
        " events=%s", events,
        " cmd=%s", client->lastcmd ? client->lastcmd->fullname : "NULL",
        " user=%s", client->user ? client->user->name : "(superuser)",
        " redir=%I", (client->flags & CLIENT_TRACKING) ? (long long) client->client_tracking_redirection : -1,
        " resp=%i", client->resp,
        " lib-name=%s", client->lib_name ? (char*)client->lib_name->ptr : "",
        " lib-ver=%s", client->lib_ver ? (char*)client->lib_ver->ptr : "",
        " io-thread=%i", client->tid,
        " tot-net-in=%U", client->net_input_bytes,
        " tot-net-out=%U", client->net_output_bytes,
        " tot-cmds=%U", client->commands_processed,
        " read-events=%U", (unsigned long long)client->stat_total_read_events,
        " avg-pipeline-len-sum=%U", (unsigned long long)client->stat_avg_pipeline_length_sum,
        " avg-pipeline-len-cnt=%U", (unsigned long long)client->stat_avg_pipeline_length_cnt));

    if (paused) resumeIOThread(client->running_tid);
    return ret;
}

sds getAllClientsInfoString(int type) {
    listNode *ln;
    listIter li;
    client *client;
    sds o = sdsnewlen(SDS_NOINIT,200*listLength(server.clients));
    sdsclear(o);

    /* Pause all IO threads to access data of clients safely, and pausing the
     * specific IO thread will not repeatedly execute in catClientInfoString. */
    int allpaused = 0;
    if (server.io_threads_num > 1 && !isCrashing() &&
        pthread_equal(server.main_thread_id, pthread_self()))
    {
        allpaused = 1;
        pauseAllIOThreads();
    }

    listRewind(server.clients,&li);
    while ((ln = listNext(&li)) != NULL) {
        client = listNodeValue(ln);
        if (type != -1 && getClientType(client) != type) continue;
        o = catClientInfoString(o,client);
        o = sdscatlen(o,"\n",1);
    }

    if (allpaused) resumeAllIOThreads();
    return o;
}

/* Check validity of an attribute that's gonna be shown in CLIENT LIST. */
int validateClientAttr(const char *val) {
    /* Check if the charset is ok. We need to do this otherwise
     * CLIENT LIST format will break. You should always be able to
     * split by space to get the different fields. */
    while (*val) {
        if (*val < '!' || *val > '~') { /* ASCII is assumed. */
            return C_ERR;
        }
        val++;
    }
    return C_OK;
}

/* Returns C_OK if the name is valid. Returns C_ERR & sets `err` (when provided) otherwise. */
int validateClientName(robj *name, const char **err) {
    const char *err_msg = "Client names cannot contain spaces, newlines or special characters.";
    int len = (name != NULL) ? sdslen(name->ptr) : 0;
    /* We allow setting the client name to an empty string. */
    if (len == 0)
        return C_OK;
    if (validateClientAttr(name->ptr) == C_ERR) {
        if (err) *err = err_msg;
        return C_ERR;
    }
    return C_OK;
}

/* Returns C_OK if the name has been set or C_ERR if the name is invalid. */
int clientSetName(client *c, robj *name, const char **err) {
    if (validateClientName(name, err) == C_ERR) {
        return C_ERR;
    }
    int len = (name != NULL) ? sdslen(name->ptr) : 0;
    /* Setting the client name to an empty string actually removes
     * the current name. */
    if (len == 0) {
        if (c->name) decrRefCount(c->name);
        c->name = NULL;
        return C_OK;
    }
    if (c->name) decrRefCount(c->name);
    c->name = name;
    incrRefCount(name);
    return C_OK;
}

/* This function implements CLIENT SETNAME, including replying to the
 * user with an error if the charset is wrong (in that case C_ERR is
 * returned). If the function succeeded C_OK is returned, and it's up
 * to the caller to send a reply if needed.
 *
 * Setting an empty string as name has the effect of unsetting the
 * currently set name: the client will remain unnamed.
 *
 * This function is also used to implement the HELLO SETNAME option. */
int clientSetNameOrReply(client *c, robj *name) {
    const char *err = NULL;
    int result = clientSetName(c, name, &err);
    if (result == C_ERR) {
        addReplyError(c, err);
    }
    return result;
}

/* Set client or connection related info */
void clientSetinfoCommand(client *c) {
    sds attr = c->argv[2]->ptr;
    robj *valob = c->argv[3];
    sds val = valob->ptr;
    robj **destvar = NULL;
    if (!strcasecmp(attr,"lib-name")) {
        destvar = &c->lib_name;
    } else if (!strcasecmp(attr,"lib-ver")) {
        destvar = &c->lib_ver;
    } else {
        addReplyErrorFormat(c,"Unrecognized option '%s'", attr);
        return;
    }

    if (validateClientAttr(val)==C_ERR) {
        addReplyErrorFormat(c,
            "%s cannot contain spaces, newlines or special characters.", attr);
        return;
    }
    if (*destvar) decrRefCount(*destvar);
    if (sdslen(val)) {
        *destvar = valob;
        incrRefCount(valob);
    } else
        *destvar = NULL;
    addReply(c,shared.ok);
}

/* Reset the client state to resemble a newly connected client.
 */
void resetCommand(client *c) {
    /* MONITOR clients are also marked with CLIENT_SLAVE, we need to
     * distinguish between the two.
     */
    uint64_t flags = c->flags;
    if (flags & CLIENT_MONITOR) flags &= ~(CLIENT_MONITOR|CLIENT_SLAVE);

    if (flags & (CLIENT_SLAVE|CLIENT_MASTER|CLIENT_MODULE)) {
        addReplyError(c,"can only reset normal client connections");
        return;
    }

    clearClientConnectionState(c);
    addReplyStatus(c,"RESET");
}

/* Disconnect the current client */
void quitCommand(client *c) {
    addReply(c,shared.ok);
    c->flags |= CLIENT_CLOSE_AFTER_REPLY;
}

void clientCommand(client *c) {
    listNode *ln;
    listIter li;

    if (c->argc == 2 && !strcasecmp(c->argv[1]->ptr,"help")) {
        const char *help[] = {
"CACHING (YES|NO)",
"    Enable/disable tracking of the keys for next command in OPTIN/OPTOUT modes.",
"GETREDIR",
"    Return the client ID we are redirecting to when tracking is enabled.",
"GETNAME",
"    Return the name of the current connection.",
"ID",
"    Return the ID of the current connection.",
"INFO",
"    Return information about the current client connection.",
"KILL <ip:port>",
"    Kill connection made from <ip:port>.",
"KILL <option> <value> [<option> <value> [...]]",
"    Kill connections. Options are:",
"    * ADDR (<ip:port>|<unixsocket>:0)",
"      Kill connections made from the specified address",
"    * LADDR (<ip:port>|<unixsocket>:0)",
"      Kill connections made to specified local address",
"    * TYPE (NORMAL|MASTER|REPLICA|PUBSUB)",
"      Kill connections by type.",
"    * USER <username>",
"      Kill connections authenticated by <username>.",
"    * SKIPME (YES|NO)",
"      Skip killing current connection (default: yes).",
"    * ID <client-id>",
"      Kill connections by client id.",
"    * MAXAGE <maxage>",
"      Kill connections older than the specified age.",
"LIST [options ...]",
"    Return information about client connections. Options:",
"    * TYPE (NORMAL|MASTER|REPLICA|PUBSUB)",
"      Return clients of specified type.",
"UNPAUSE",
"    Stop the current client pause, resuming traffic.",
"PAUSE <timeout> [WRITE|ALL]",
"    Suspend all, or just write, clients for <timeout> milliseconds.",
"REPLY (ON|OFF|SKIP)",
"    Control the replies sent to the current connection.",
"SETNAME <name>",
"    Assign the name <name> to the current connection.",
"SETINFO <option> <value>",
"    Set client meta attr. Options are:",
"    * LIB-NAME: the client lib name.",
"    * LIB-VER: the client lib version.",
"UNBLOCK <clientid> [TIMEOUT|ERROR]",
"    Unblock the specified blocked client.",
"TRACKING (ON|OFF) [REDIRECT <id>] [BCAST] [PREFIX <prefix> [...]]",
"         [OPTIN] [OPTOUT] [NOLOOP]",
"    Control server assisted client side caching.",
"TRACKINGINFO",
"    Report tracking status for the current connection.",
"NO-EVICT (ON|OFF)",
"    Protect current client connection from eviction.",
"NO-TOUCH (ON|OFF)",
"    Will not touch LRU/LFU stats when this mode is on.",
NULL
        };
        addReplyHelp(c, help);
    } else if (!strcasecmp(c->argv[1]->ptr,"id") && c->argc == 2) {
        /* CLIENT ID */
        addReplyLongLong(c,c->id);
    } else if (!strcasecmp(c->argv[1]->ptr,"info") && c->argc == 2) {
        /* CLIENT INFO */
        sds o = catClientInfoString(sdsempty(), c);
        o = sdscatlen(o,"\n",1);
        addReplyVerbatim(c,o,sdslen(o),"txt");
        sdsfree(o);
    } else if (!strcasecmp(c->argv[1]->ptr,"list")) {
        /* CLIENT LIST */
        int type = -1;
        sds o = NULL;
        if (c->argc == 4 && !strcasecmp(c->argv[2]->ptr,"type")) {
            type = getClientTypeByName(c->argv[3]->ptr);
            if (type == -1) {
                addReplyErrorFormat(c,"Unknown client type '%s'",
                    (char*) c->argv[3]->ptr);
                return;
            }
        } else if (c->argc > 3 && !strcasecmp(c->argv[2]->ptr,"id")) {
            int j;
            o = sdsempty();
            for (j = 3; j < c->argc; j++) {
                long long cid;
                if (getLongLongFromObjectOrReply(c, c->argv[j], &cid,
                            "Invalid client ID")) {
                    sdsfree(o);
                    return;
                }
                client *cl = lookupClientByID(cid);
                if (cl) {
                    o = catClientInfoString(o, cl);
                    o = sdscatlen(o, "\n", 1);
                }
            }
        } else if (c->argc != 2) {
            addReplyErrorObject(c,shared.syntaxerr);
            return;
        }

        if (!o)
            o = getAllClientsInfoString(type);
        addReplyVerbatim(c,o,sdslen(o),"txt");
        sdsfree(o);
    } else if (!strcasecmp(c->argv[1]->ptr,"reply") && c->argc == 3) {
        /* CLIENT REPLY ON|OFF|SKIP */
        if (!strcasecmp(c->argv[2]->ptr,"on")) {
            c->flags &= ~(CLIENT_REPLY_SKIP|CLIENT_REPLY_OFF);
            addReply(c,shared.ok);
        } else if (!strcasecmp(c->argv[2]->ptr,"off")) {
            c->flags |= CLIENT_REPLY_OFF;
        } else if (!strcasecmp(c->argv[2]->ptr,"skip")) {
            if (!(c->flags & CLIENT_REPLY_OFF))
                c->flags |= CLIENT_REPLY_SKIP_NEXT;
        } else {
            addReplyErrorObject(c,shared.syntaxerr);
            return;
        }
    } else if (!strcasecmp(c->argv[1]->ptr,"no-evict") && c->argc == 3) {
        /* CLIENT NO-EVICT ON|OFF */
        if (!strcasecmp(c->argv[2]->ptr,"on")) {
            c->flags |= CLIENT_NO_EVICT;
            removeClientFromMemUsageBucket(c, 0);
            addReply(c,shared.ok);
        } else if (!strcasecmp(c->argv[2]->ptr,"off")) {
            c->flags &= ~CLIENT_NO_EVICT;
            updateClientMemUsageAndBucket(c);
            addReply(c,shared.ok);
        } else {
            addReplyErrorObject(c,shared.syntaxerr);
            return;
        }
    } else if (!strcasecmp(c->argv[1]->ptr,"kill")) {
        /* CLIENT KILL <ip:port>
         * CLIENT KILL <option> [value] ... <option> [value] */
        char *addr = NULL;
        char *laddr = NULL;
        user *user = NULL;
        int type = -1;
        uint64_t id = 0;
        long long max_age = 0;
        int skipme = 1;
        int killed = 0, close_this_client = 0;

        if (c->argc == 3) {
            /* Old style syntax: CLIENT KILL <addr> */
            addr = c->argv[2]->ptr;
            skipme = 0; /* With the old form, you can kill yourself. */
        } else if (c->argc > 3) {
            int i = 2; /* Next option index. */

            /* New style syntax: parse options. */
            while(i < c->argc) {
                int moreargs = c->argc > i+1;

                if (!strcasecmp(c->argv[i]->ptr,"id") && moreargs) {
                    long tmp;

                    if (getRangeLongFromObjectOrReply(c, c->argv[i+1], 1, LONG_MAX, &tmp,
                                                      "client-id should be greater than 0") != C_OK)
                        return;
                    id = tmp;
                } else if (!strcasecmp(c->argv[i]->ptr,"maxage") && moreargs) {
                    long long tmp;

                    if (getLongLongFromObjectOrReply(c, c->argv[i+1], &tmp,
                                                     "maxage is not an integer or out of range") != C_OK)
                        return;
                    if (tmp <= 0) {
                        addReplyError(c, "maxage should be greater than 0");
                        return;
                    }

                    max_age = tmp;
                } else if (!strcasecmp(c->argv[i]->ptr,"type") && moreargs) {
                    type = getClientTypeByName(c->argv[i+1]->ptr);
                    if (type == -1) {
                        addReplyErrorFormat(c,"Unknown client type '%s'",
                            (char*) c->argv[i+1]->ptr);
                        return;
                    }
                } else if (!strcasecmp(c->argv[i]->ptr,"addr") && moreargs) {
                    addr = c->argv[i+1]->ptr;
                } else if (!strcasecmp(c->argv[i]->ptr,"laddr") && moreargs) {
                    laddr = c->argv[i+1]->ptr;
                } else if (!strcasecmp(c->argv[i]->ptr,"user") && moreargs) {
                    user = ACLGetUserByName(c->argv[i+1]->ptr,
                                            sdslen(c->argv[i+1]->ptr));
                    if (user == NULL) {
                        addReplyErrorFormat(c,"No such user '%s'",
                            (char*) c->argv[i+1]->ptr);
                        return;
                    }
                } else if (!strcasecmp(c->argv[i]->ptr,"skipme") && moreargs) {
                    if (!strcasecmp(c->argv[i+1]->ptr,"yes")) {
                        skipme = 1;
                    } else if (!strcasecmp(c->argv[i+1]->ptr,"no")) {
                        skipme = 0;
                    } else {
                        addReplyErrorObject(c,shared.syntaxerr);
                        return;
                    }
                } else {
                    addReplyErrorObject(c,shared.syntaxerr);
                    return;
                }
                i += 2;
            }
        } else {
            addReplyErrorObject(c,shared.syntaxerr);
            return;
        }

        /* Iterate clients killing all the matching clients. */
        listRewind(server.clients,&li);
        while ((ln = listNext(&li)) != NULL) {
            client *client = listNodeValue(ln);
            if (addr && strcmp(getClientPeerId(client),addr) != 0) continue;
            if (laddr && strcmp(getClientSockname(client),laddr) != 0) continue;
            if (type != -1 && getClientType(client) != type) continue;
            if (id != 0 && client->id != id) continue;
            if (user && client->user != user) continue;
            if (c == client && skipme) continue;
            if (max_age != 0 && (long long)(commandTimeSnapshot() / 1000 - client->ctime) < max_age) continue;

            /* Kill it. */
            if (c == client) {
                close_this_client = 1;
            } else {
                freeClient(client);
            }
            killed++;
        }

        /* Reply according to old/new format. */
        if (c->argc == 3) {
            if (killed == 0)
                addReplyError(c,"No such client");
            else
                addReply(c,shared.ok);
        } else {
            addReplyLongLong(c,killed);
        }

        /* If this client has to be closed, flag it as CLOSE_AFTER_REPLY
         * only after we queued the reply to its output buffers. */
        if (close_this_client) c->flags |= CLIENT_CLOSE_AFTER_REPLY;
    } else if (!strcasecmp(c->argv[1]->ptr,"unblock") && (c->argc == 3 ||
                                                          c->argc == 4))
    {
        /* CLIENT UNBLOCK <id> [timeout|error] */
        long long id;
        int unblock_error = 0;

        if (c->argc == 4) {
            if (!strcasecmp(c->argv[3]->ptr,"timeout")) {
                unblock_error = 0;
            } else if (!strcasecmp(c->argv[3]->ptr,"error")) {
                unblock_error = 1;
            } else {
                addReplyError(c,
                    "CLIENT UNBLOCK reason should be TIMEOUT or ERROR");
                return;
            }
        }
        if (getLongLongFromObjectOrReply(c,c->argv[2],&id,NULL)
            != C_OK) return;
        struct client *target = lookupClientByID(id);
        /* Note that we never try to unblock a client blocked on a module command,
         * or a client blocked by CLIENT PAUSE or some other blocking type which
         * doesn't have a timeout callback (even in the case of UNBLOCK ERROR).
         * The reason is that we assume that if a command doesn't expect to be timedout,
         * it also doesn't expect to be unblocked by CLIENT UNBLOCK */
        if (target && target->flags & CLIENT_BLOCKED && blockedClientMayTimeout(target)) {
            if (unblock_error)
                unblockClientOnError(target,
                    "-UNBLOCKED client unblocked via CLIENT UNBLOCK");
            else
                unblockClientOnTimeout(target);

            addReply(c,shared.cone);
        } else {
            addReply(c,shared.czero);
        }
    } else if (!strcasecmp(c->argv[1]->ptr,"setname") && c->argc == 3) {
        /* CLIENT SETNAME */
        if (clientSetNameOrReply(c,c->argv[2]) == C_OK)
            addReply(c,shared.ok);
    } else if (!strcasecmp(c->argv[1]->ptr,"getname") && c->argc == 2) {
        /* CLIENT GETNAME */
        if (c->name)
            addReplyBulk(c,c->name);
        else
            addReplyNull(c);
    } else if (!strcasecmp(c->argv[1]->ptr,"unpause") && c->argc == 2) {
        /* CLIENT UNPAUSE */
        unpauseActions(PAUSE_BY_CLIENT_COMMAND);
        addReply(c,shared.ok);
    } else if (!strcasecmp(c->argv[1]->ptr,"pause") && (c->argc == 3 ||
                                                        c->argc == 4))
    {
        /* CLIENT PAUSE TIMEOUT [WRITE|ALL] */
        mstime_t end;
        int isPauseClientAll = 1;
        if (c->argc == 4) {
            if (!strcasecmp(c->argv[3]->ptr,"write")) {
                isPauseClientAll = 0;
            } else if (strcasecmp(c->argv[3]->ptr,"all")) {
                addReplyError(c,
                    "CLIENT PAUSE mode must be WRITE or ALL");  
                return;       
            }
        }

        if (getTimeoutFromObjectOrReply(c,c->argv[2],&end,
            UNIT_MILLISECONDS) != C_OK) return;
        pauseClientsByClient(end, isPauseClientAll);
        addReply(c,shared.ok);
    } else if (!strcasecmp(c->argv[1]->ptr,"tracking") && c->argc >= 3) {
        /* CLIENT TRACKING (on|off) [REDIRECT <id>] [BCAST] [PREFIX first]
         *                          [PREFIX second] [OPTIN] [OPTOUT] [NOLOOP]... */
        long long redir = 0;
        uint64_t options = 0;
        robj **prefix = NULL;
        size_t numprefix = 0;

        /* Parse the options. */
        for (int j = 3; j < c->argc; j++) {
            int moreargs = (c->argc-1) - j;

            if (!strcasecmp(c->argv[j]->ptr,"redirect") && moreargs) {
                j++;
                if (redir != 0) {
                    addReplyError(c,"A client can only redirect to a single "
                                    "other client");
                    zfree(prefix);
                    return;
                }

                if (getLongLongFromObjectOrReply(c,c->argv[j],&redir,NULL) !=
                    C_OK)
                {
                    zfree(prefix);
                    return;
                }
                /* We will require the client with the specified ID to exist
                 * right now, even if it is possible that it gets disconnected
                 * later. Still a valid sanity check. */
                if (lookupClientByID(redir) == NULL) {
                    addReplyError(c,"The client ID you want redirect to "
                                    "does not exist");
                    zfree(prefix);
                    return;
                }
            } else if (!strcasecmp(c->argv[j]->ptr,"bcast")) {
                options |= CLIENT_TRACKING_BCAST;
            } else if (!strcasecmp(c->argv[j]->ptr,"optin")) {
                options |= CLIENT_TRACKING_OPTIN;
            } else if (!strcasecmp(c->argv[j]->ptr,"optout")) {
                options |= CLIENT_TRACKING_OPTOUT;
            } else if (!strcasecmp(c->argv[j]->ptr,"noloop")) {
                options |= CLIENT_TRACKING_NOLOOP;
            } else if (!strcasecmp(c->argv[j]->ptr,"prefix") && moreargs) {
                j++;
                prefix = zrealloc(prefix,sizeof(robj*)*(numprefix+1));
                prefix[numprefix++] = c->argv[j];
            } else {
                zfree(prefix);
                addReplyErrorObject(c,shared.syntaxerr);
                return;
            }
        }

        /* Options are ok: enable or disable the tracking for this client. */
        if (!strcasecmp(c->argv[2]->ptr,"on")) {
            /* Before enabling tracking, make sure options are compatible
             * among each other and with the current state of the client. */
            if (!(options & CLIENT_TRACKING_BCAST) && numprefix) {
                addReplyError(c,
                    "PREFIX option requires BCAST mode to be enabled");
                zfree(prefix);
                return;
            }

            if (c->flags & CLIENT_TRACKING) {
                int oldbcast = !!(c->flags & CLIENT_TRACKING_BCAST);
                int newbcast = !!(options & CLIENT_TRACKING_BCAST);
                if (oldbcast != newbcast) {
                    addReplyError(c,
                    "You can't switch BCAST mode on/off before disabling "
                    "tracking for this client, and then re-enabling it with "
                    "a different mode.");
                    zfree(prefix);
                    return;
                }
            }

            if (options & CLIENT_TRACKING_BCAST &&
                options & (CLIENT_TRACKING_OPTIN|CLIENT_TRACKING_OPTOUT))
            {
                addReplyError(c,
                "OPTIN and OPTOUT are not compatible with BCAST");
                zfree(prefix);
                return;
            }

            if (options & CLIENT_TRACKING_OPTIN && options & CLIENT_TRACKING_OPTOUT)
            {
                addReplyError(c,
                "You can't specify both OPTIN mode and OPTOUT mode");
                zfree(prefix);
                return;
            }

            if ((options & CLIENT_TRACKING_OPTIN && c->flags & CLIENT_TRACKING_OPTOUT) ||
                (options & CLIENT_TRACKING_OPTOUT && c->flags & CLIENT_TRACKING_OPTIN))
            {
                addReplyError(c,
                "You can't switch OPTIN/OPTOUT mode before disabling "
                "tracking for this client, and then re-enabling it with "
                "a different mode.");
                zfree(prefix);
                return;
            }

            if (options & CLIENT_TRACKING_BCAST) {
                if (!checkPrefixCollisionsOrReply(c,prefix,numprefix)) {
                    zfree(prefix);
                    return;
                }
            }

            enableTracking(c,redir,options,prefix,numprefix);
        } else if (!strcasecmp(c->argv[2]->ptr,"off")) {
            disableTracking(c);
        } else {
            zfree(prefix);
            addReplyErrorObject(c,shared.syntaxerr);
            return;
        }
        zfree(prefix);
        addReply(c,shared.ok);
    } else if (!strcasecmp(c->argv[1]->ptr,"caching") && c->argc >= 3) {
        if (!(c->flags & CLIENT_TRACKING)) {
            addReplyError(c,"CLIENT CACHING can be called only when the "
                            "client is in tracking mode with OPTIN or "
                            "OPTOUT mode enabled");
            return;
        }

        char *opt = c->argv[2]->ptr;
        if (!strcasecmp(opt,"yes")) {
            if (c->flags & CLIENT_TRACKING_OPTIN) {
                c->flags |= CLIENT_TRACKING_CACHING;
            } else {
                addReplyError(c,"CLIENT CACHING YES is only valid when tracking is enabled in OPTIN mode.");
                return;
            }
        } else if (!strcasecmp(opt,"no")) {
            if (c->flags & CLIENT_TRACKING_OPTOUT) {
                c->flags |= CLIENT_TRACKING_CACHING;
            } else {
                addReplyError(c,"CLIENT CACHING NO is only valid when tracking is enabled in OPTOUT mode.");
                return;
            }
        } else {
            addReplyErrorObject(c,shared.syntaxerr);
            return;
        }

        /* Common reply for when we succeeded. */
        addReply(c,shared.ok);
    } else if (!strcasecmp(c->argv[1]->ptr,"getredir") && c->argc == 2) {
        /* CLIENT GETREDIR */
        if (c->flags & CLIENT_TRACKING) {
            addReplyLongLong(c,c->client_tracking_redirection);
        } else {
            addReplyLongLong(c,-1);
        }
    } else if (!strcasecmp(c->argv[1]->ptr,"trackinginfo") && c->argc == 2) {
        addReplyMapLen(c,3);

        /* Flags */
        addReplyBulkCString(c,"flags");
        void *arraylen_ptr = addReplyDeferredLen(c);
        int numflags = 0;
        addReplyBulkCString(c,c->flags & CLIENT_TRACKING ? "on" : "off");
        numflags++;
        if (c->flags & CLIENT_TRACKING_BCAST) {
            addReplyBulkCString(c,"bcast");
            numflags++;
        }
        if (c->flags & CLIENT_TRACKING_OPTIN) {
            addReplyBulkCString(c,"optin");
            numflags++;
            if (c->flags & CLIENT_TRACKING_CACHING) {
                addReplyBulkCString(c,"caching-yes");
                numflags++;        
            }
        }
        if (c->flags & CLIENT_TRACKING_OPTOUT) {
            addReplyBulkCString(c,"optout");
            numflags++;
            if (c->flags & CLIENT_TRACKING_CACHING) {
                addReplyBulkCString(c,"caching-no");
                numflags++;        
            }
        }
        if (c->flags & CLIENT_TRACKING_NOLOOP) {
            addReplyBulkCString(c,"noloop");
            numflags++;
        }
        if (c->flags & CLIENT_TRACKING_BROKEN_REDIR) {
            addReplyBulkCString(c,"broken_redirect");
            numflags++;
        }
        setDeferredSetLen(c,arraylen_ptr,numflags);

        /* Redirect */
        addReplyBulkCString(c,"redirect");
        if (c->flags & CLIENT_TRACKING) {
            addReplyLongLong(c,c->client_tracking_redirection);
        } else {
            addReplyLongLong(c,-1);
        }

        /* Prefixes */
        addReplyBulkCString(c,"prefixes");
        if (c->client_tracking_prefixes) {
            addReplyArrayLen(c,raxSize(c->client_tracking_prefixes));
            raxIterator ri;
            raxStart(&ri,c->client_tracking_prefixes);
            raxSeek(&ri,"^",NULL,0);
            while(raxNext(&ri)) {
                addReplyBulkCBuffer(c,ri.key,ri.key_len);
            }
            raxStop(&ri);
        } else {
            addReplyArrayLen(c,0);
        }
    } else if (!strcasecmp(c->argv[1]->ptr, "no-touch")) {
        /* CLIENT NO-TOUCH ON|OFF */
        if (!strcasecmp(c->argv[2]->ptr,"on")) {
            c->flags |= CLIENT_NO_TOUCH;
            addReply(c,shared.ok);
        } else if (!strcasecmp(c->argv[2]->ptr,"off")) {
            c->flags &= ~CLIENT_NO_TOUCH;
            addReply(c,shared.ok);
        } else {
            addReplyErrorObject(c,shared.syntaxerr);
        }
    } else {
        addReplySubcommandSyntaxError(c);
    }
}

/* HELLO [<protocol-version> [AUTH <user> <password>] [SETNAME <name>] ] */
void helloCommand(client *c) {
    long long ver = 0;
    int next_arg = 1;

    if (c->argc >= 2) {
        if (getLongLongFromObjectOrReply(c, c->argv[next_arg++], &ver,
            "Protocol version is not an integer or out of range") != C_OK) {
            return;
        }

        if (ver < 2 || ver > 3) {
            addReplyError(c,"-NOPROTO unsupported protocol version");
            return;
        }
    }

    robj *username = NULL;
    robj *password = NULL;
    robj *clientname = NULL;
    for (int j = next_arg; j < c->argc; j++) {
        int moreargs = (c->argc-1) - j;
        const char *opt = c->argv[j]->ptr;
        if (!strcasecmp(opt,"AUTH") && moreargs >= 2) {
            redactClientCommandArgument(c, j+1);
            redactClientCommandArgument(c, j+2);
            username = c->argv[j+1];
            password = c->argv[j+2];
            j += 2;
        } else if (!strcasecmp(opt,"SETNAME") && moreargs) {
            clientname = c->argv[j+1];
            const char *err = NULL;
            if (validateClientName(clientname, &err) == C_ERR) {
                addReplyError(c, err);
                return;
            }
            j++;
        } else {
            addReplyErrorFormat(c,"Syntax error in HELLO option '%s'",opt);
            return;
        }
    }

    if (username && password) {
        robj *err = NULL;
        int auth_result = ACLAuthenticateUser(c, username, password, &err);
        if (auth_result == AUTH_ERR) {
            addAuthErrReply(c, err);
        }
        if (err) decrRefCount(err);
        /* In case of auth errors, return early since we already replied with an ERR.
         * In case of blocking module auth, we reply to the client/setname later upon unblocking. */
        if (auth_result == AUTH_ERR || auth_result == AUTH_BLOCKED) {
            return;
        }
    }

    /* At this point we need to be authenticated to continue. */
    if (!c->authenticated) {
        addReplyError(c,"-NOAUTH HELLO must be called with the client already "
                        "authenticated, otherwise the HELLO <proto> AUTH <user> <pass> "
                        "option can be used to authenticate the client and "
                        "select the RESP protocol version at the same time");
        return;
    }

    /* Now that we're authenticated, set the client name. */
    if (clientname) clientSetName(c, clientname, NULL);

    /* Let's switch to the specified RESP mode. */
    if (ver) c->resp = ver;
    addReplyMapLen(c,6 + !server.sentinel_mode);

    ADD_REPLY_BULK_CBUFFER_STRING_CONSTANT(c,"server");
    ADD_REPLY_BULK_CBUFFER_STRING_CONSTANT(c,"redis");

    ADD_REPLY_BULK_CBUFFER_STRING_CONSTANT(c,"version");
    ADD_REPLY_BULK_CBUFFER_STRING_CONSTANT(c,REDIS_VERSION);

    ADD_REPLY_BULK_CBUFFER_STRING_CONSTANT(c,"proto");

    addReplyLongLong(c,c->resp);

    ADD_REPLY_BULK_CBUFFER_STRING_CONSTANT(c,"id");
    addReplyLongLong(c,c->id);

    ADD_REPLY_BULK_CBUFFER_STRING_CONSTANT(c,"mode");
    if (server.sentinel_mode) addReplyBulkCString(c,"sentinel");
    else if (server.cluster_enabled) addReplyBulkCString(c,"cluster");
    else addReplyBulkCString(c,"standalone");

    if (!server.sentinel_mode) {
        ADD_REPLY_BULK_CBUFFER_STRING_CONSTANT(c,"role");
        addReplyBulkCString(c,server.masterhost ? "replica" : "master");
    }

    ADD_REPLY_BULK_CBUFFER_STRING_CONSTANT(c,"modules");
    addReplyLoadedModules(c);
}

/* This callback is bound to POST and "Host:" command names. Those are not
 * really commands, but are used in security attacks in order to talk to
 * Redis instances via HTTP, with a technique called "cross protocol scripting"
 * which exploits the fact that services like Redis will discard invalid
 * HTTP headers and will process what follows.
 *
 * As a protection against this attack, Redis will terminate the connection
 * when a POST or "Host:" header is seen, and will log the event from
 * time to time (to avoid creating a DOS as a result of too many logs). */
void securityWarningCommand(client *c) {
    static time_t logged_time = 0;
    time_t now = time(NULL);

    if (llabs(now-logged_time) > 60) {
        char ip[NET_IP_STR_LEN];
        int port;
        if (connAddrPeerName(c->conn, ip, sizeof(ip), &port) == -1) {
            serverLog(LL_WARNING,"Possible SECURITY ATTACK detected. It looks like somebody is sending POST or Host: commands to Redis. This is likely due to an attacker attempting to use Cross Protocol Scripting to compromise your Redis instance. Connection aborted.");
        } else {
            serverLog(LL_WARNING,"Possible SECURITY ATTACK detected. It looks like somebody is sending POST or Host: commands to Redis. This is likely due to an attacker attempting to use Cross Protocol Scripting to compromise your Redis instance. Connection from %s:%d aborted.", ip, port);
        }
        logged_time = now;
    }
    freeClientAsync(c);
}

/* Keep track of the original command arguments so that we can generate
 * an accurate slowlog entry after the command has been executed. */
static void retainOriginalCommandVector(client *c) {
    /* We already rewrote this command, so don't rewrite it again */
    if (c->original_argv) return;
    c->original_argc = c->argc;
    c->original_argv = zmalloc(sizeof(robj*)*(c->argc));
    for (int j = 0; j < c->argc; j++) {
        c->original_argv[j] = c->argv[j];
        incrRefCount(c->argv[j]);
    }
}

/* Redact a given argument to prevent it from being shown
 * in the slowlog. This information is stored in the
 * original_argv array. */
void redactClientCommandArgument(client *c, int argc) {
    retainOriginalCommandVector(c);
    if (c->original_argv[argc] == shared.redacted) {
        /* This argument has already been redacted */
        return;
    }
    decrRefCount(c->original_argv[argc]);
    c->original_argv[argc] = shared.redacted;
}

/* Rewrite the command vector of the client. All the new objects ref count
 * is incremented. The old command vector is freed, and the old objects
 * ref count is decremented. */
void rewriteClientCommandVector(client *c, int argc, ...) {
    va_list ap;
    int j;
    robj **argv; /* The new argument vector */

    argv = zmalloc(sizeof(robj*)*argc);
    va_start(ap,argc);
    for (j = 0; j < argc; j++) {
        robj *a;

        a = va_arg(ap, robj*);
        argv[j] = a;
        incrRefCount(a);
    }
    replaceClientCommandVector(c, argc, argv);
    va_end(ap);
}

/* Completely replace the client command vector with the provided one. */
void replaceClientCommandVector(client *c, int argc, robj **argv) {
    int j;
    retainOriginalCommandVector(c);

    /* We don't need to just fix the client argv, we also need to fix the pending command (same argv),
     * But sometimes we reach here not from a real client, but from a Lua 'scriptRunCtx'. This flow bypasses the
     * pending-command system entirely and uses c->argv directly. In this case there's no pending commands
     * to update, so we skip that code. */
    pendingCommand *pcmd = NULL;
    int is_mstate = 0;
    if (c->mstate.executing_cmd < 0) {
        is_mstate = 0;
        if (c->pending_cmds.ready_len > 0) {
            pcmd = c->pending_cmds.head;
            serverAssert(!(pcmd->flags & PENDING_CMD_FLAG_INCOMPLETE));
        }
    } else {
        is_mstate = 1;
        serverAssert(c->mstate.executing_cmd < c->mstate.count);
        pcmd = c->mstate.commands[c->mstate.executing_cmd];
    }

    if (pcmd) {
        serverAssert(pcmd->argv == c->argv);
        pcmd->argv = argv;
        pcmd->argc = argc;
        pcmd->argv_len = argc;
    }
    freeClientArgv(c);
    c->argv = argv;
    c->argc = c->argv_len = argc;

    if (!is_mstate) {  /* multi-state does not track all_argv_len_sum, see code in queueMultiCommand */
        size_t new_argv_len_sum = 0;
        for (j = 0; j < c->argc; j++)
            if (c->argv[j])
                new_argv_len_sum += getStringObjectLen(c->argv[j]);

        if (!pcmd) {
            c->all_argv_len_sum = new_argv_len_sum;
        } else {
            c->all_argv_len_sum -= pcmd->argv_len_sum;
            pcmd->argv_len_sum = new_argv_len_sum;
            c->all_argv_len_sum += pcmd->argv_len_sum;
        }
    }
    c->cmd = lookupCommandOrOriginal(c->argv,c->argc);
    if (pcmd)
        pcmd->cmd = c->cmd;
    serverAssertWithInfo(c,NULL,c->cmd != NULL);
}

/* Rewrite a single item in the command vector.
 * The new val ref count is incremented, and the old decremented.
 *
 * It is possible to specify an argument over the current size of the
 * argument vector: in this case the array of objects gets reallocated
 * and c->argc set to the max value. However it's up to the caller to
 *
 * 1. Make sure there are no "holes" and all the arguments are set.
 * 2. If the original argument vector was longer than the one we
 *    want to end with, it's up to the caller to set c->argc and
 *    free the no longer used objects on c->argv.
 * 3. To remove argument at i'th index, pass NULL as new value
 */
void rewriteClientCommandArgument(client *c, int i, robj *newval) {
    robj *oldval;
    retainOriginalCommandVector(c);

    /* We don't need to just fix the client argv, we also need to fix the pending command (same argv),
     * But sometimes we reach here not from a real client, but from a Lua 'scriptRunCtx'. This flow bypasses the
     * pending-command system entirely and uses c->argv directly. In this case there's no pending commands
     * to update, so we skip that code. */
    pendingCommand *pcmd = c->pending_cmds.head ? c->pending_cmds.head: NULL;
    int update_pcmd = pcmd && pcmd->argv == c->argv;

    /* We need to handle both extending beyond argc (just update it and
     * initialize the new element) or beyond argv_len (realloc is needed).
     */
    if (i >= c->argc) {
        if (i >= c->argv_len) {
            c->argv = zrealloc(c->argv,sizeof(robj*)*(i+1));
            c->argv_len = i+1;
        }
        c->argc = i+1;
        c->argv[i] = NULL;
    }
    oldval = c->argv[i];
    if (oldval) c->all_argv_len_sum -= getStringObjectLen(oldval);

    if (newval) {
        c->argv[i] = newval;
        incrRefCount(newval);
        c->all_argv_len_sum += getStringObjectLen(newval);
    } else {
        /* move the remaining arguments one step left */
        for (int j = i+1; j < c->argc; j++) {
            c->argv[j-1] = c->argv[j];
        }
        c->argv[--c->argc] = NULL;
    }
    if (oldval) decrRefCount(oldval);

    if (update_pcmd) {
        pcmd->argv = c->argv;
        pcmd->argc = c->argc;
        pcmd->argv_len = c->argv_len;
        if (oldval) pcmd->argv_len_sum -= getStringObjectLen(oldval);
        if (newval) pcmd->argv_len_sum += getStringObjectLen(newval);
    }

    /* If this is the command name make sure to fix c->cmd. */
    if (i == 0) {
        c->cmd = lookupCommandOrOriginal(c->argv,c->argc);
        serverAssertWithInfo(c,NULL,c->cmd != NULL);
        if (update_pcmd)
            pcmd->cmd = c->cmd;
    }
}

/* This function returns the number of bytes that Redis is
 * using to store the reply still not read by the client.
 * It does NOT include any referenced bytes (neither shared nor unshared).
 *
 * 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. */
static size_t getClientOutputBufferAllocSize(client *c) {
    if (unlikely(clientTypeIsSlave(c))) {
        size_t repl_buf_size = 0;
        size_t repl_node_num = 0;
        size_t repl_node_size = sizeof(listNode) + sizeof(replBufBlock);
        if (c->ref_repl_buf_node) {
            replBufBlock *last = listNodeValue(listLast(server.repl_buffer_blocks));
            replBufBlock *cur = listNodeValue(c->ref_repl_buf_node);
            repl_buf_size = last->repl_offset + last->size - cur->repl_offset;
            repl_node_num = last->id - cur->id + 1;
        }
        return repl_buf_size + (repl_node_size*repl_node_num);
    } else { 
        size_t list_item_size = sizeof(listNode) + sizeof(clientReplyBlock);
        return c->reply_bytes + (list_item_size*listLength(c->reply));
    }
}

/* Returns the logical output buffer size for limit enforcement.
 * This includes all shared memory (shared with the keyspace), ensuring that
 * a client requesting huge amounts of data via copy-avoidance is still
 * subject to output buffer limits. */
static size_t getClientOutputBufferLogicalSize(client *c) {
    size_t mem = getClientOutputBufferAllocSize(c);
    if (!clientTypeIsSlave(c))
        mem += c->reply_bytes_shared;
    return mem;
}

/* Returns the actual memory used to store the reply not yet read by the client.
 * This includes unshared memory (solely owned by this client), which would be
 * freed when the client disconnects. */
size_t getClientOutputBufferMemoryUsage(client *c) {
    size_t mem = getClientOutputBufferAllocSize(c);
    mem += c->reply_bytes_unshared;
    return mem;
}

size_t getNormalClientPendingReplyBytes(client *c) {
    serverAssert(!clientTypeIsSlave(c));
    if (listLength(c->reply) == 0) return c->bufpos + c->reply_bytes_shared;

    clientReplyBlock *block = listNodeValue(listLast(c->reply));
    return (c->reply_bytes + c->reply_bytes_shared - block->size + block->used) + c->bufpos;
}

/* Returns the total client's memory usage. */
size_t getClientMemoryUsage(client *c) {
    size_t mem = getClientOutputBufferMemoryUsage(c);

    mem += c->querybuf ? sdsZmallocSize(c->querybuf) : 0;
    mem += zmalloc_size(c);
    mem += c->buf_usable_size;
    /* For efficiency (less work keeping track of the argv memory), it doesn't include the used memory
     * i.e. unused sds space and internal fragmentation, just the string length. but this is enough to
     * spot problematic clients. */
    mem += c->all_argv_len_sum + sizeof(robj*)*c->argc;
    mem += multiStateMemOverhead(c);

    /* Add memory overhead of this client's HIMPORT fieldset bindings. */
    mem += himportFieldsetsMemOverhead(c);

    /* Add memory overhead of pubsub channels and patterns. Note: this is just the overhead of the robj pointers
     * to the strings themselves because they aren't stored per client. */
    mem += pubsubMemOverhead(c);

    /* Add memory overhead of the tracking prefixes, this is an underestimation so we don't need to traverse the entire rax */
    if (c->client_tracking_prefixes)
        mem += c->client_tracking_prefixes->numnodes * (sizeof(raxNode) * sizeof(raxNode*));

    return mem;
}

/* Get the class of a client, used in order to enforce limits to different
 * classes of clients.
 *
 * The function will return one of the following:
 * CLIENT_TYPE_NORMAL -> Normal client, including MONITOR
 * CLIENT_TYPE_SLAVE  -> Slave
 * CLIENT_TYPE_PUBSUB -> Client subscribed to Pub/Sub channels
 * CLIENT_TYPE_MASTER -> The client representing our replication master.
 */
int getClientType(client *c) {
    if (c->flags & CLIENT_MASTER) return CLIENT_TYPE_MASTER;
    /* Even though MONITOR clients are marked as replicas, we
     * want the expose them as normal clients. */
    if ((c->flags & CLIENT_SLAVE) && !(c->flags & CLIENT_MONITOR))
        return CLIENT_TYPE_SLAVE;
    if (c->flags & CLIENT_PUBSUB) return CLIENT_TYPE_PUBSUB;
    return CLIENT_TYPE_NORMAL;
}

static inline int clientTypeIsSlave(client *c) {
    /* Even though MONITOR clients and ASM destination RDB/main channels are
     * marked as replicas, we want to expose them as normal clients. */
    if (unlikely((c->flags & CLIENT_SLAVE) &&
        !(c->flags & (CLIENT_MONITOR | CLIENT_ASM_MIGRATING))))
    {
        return 1;
    }
    return 0;
}

int getClientTypeByName(char *name) {
    if (!strcasecmp(name,"normal")) return CLIENT_TYPE_NORMAL;
    else if (!strcasecmp(name,"slave")) return CLIENT_TYPE_SLAVE;
    else if (!strcasecmp(name,"replica")) return CLIENT_TYPE_SLAVE;
    else if (!strcasecmp(name,"pubsub")) return CLIENT_TYPE_PUBSUB;
    else if (!strcasecmp(name,"master")) return CLIENT_TYPE_MASTER;
    else return -1;
}

char *getClientTypeName(int class) {
    switch(class) {
    case CLIENT_TYPE_NORMAL: return "normal";
    case CLIENT_TYPE_SLAVE:  return "slave";
    case CLIENT_TYPE_PUBSUB: return "pubsub";
    case CLIENT_TYPE_MASTER: return "master";
    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(client *c) {
    int soft = 0, hard = 0, class;
    unsigned long used_mem = getClientOutputBufferLogicalSize(c);

    /* For unauthenticated clients the output buffer is limited to prevent
     * them from abusing it by not reading the replies */
    if (used_mem > 1024 && authRequired(c))
        return 1;

    class = getClientType(c);
    /* For the purpose of output buffer limiting, masters are handled
     * like normal clients. */
    if (class == CLIENT_TYPE_MASTER) class = CLIENT_TYPE_NORMAL;

    /* Note that it doesn't make sense to set the replica clients output buffer
     * limit lower than the repl-backlog-size config (partial sync will succeed
     * and then replica will get disconnected).
     * Such a configuration is ignored (the size of repl-backlog-size will be used).
     * This doesn't have memory consumption implications since the replica client
     * will share the backlog buffers memory. */
    size_t hard_limit_bytes = server.client_obuf_limits[class].hard_limit_bytes;
    if (class == CLIENT_TYPE_SLAVE && hard_limit_bytes &&
        (long long)hard_limit_bytes < server.repl_backlog_size)
        hard_limit_bytes = server.repl_backlog_size;
    if (server.client_obuf_limits[class].hard_limit_bytes &&
        used_mem >= 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 CLIENT_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.
 * When `async` is set to 0, we close the client immediately, this is
 * useful when called from cron.
 *
 * Returns 1 if client was (flagged) closed. */
int closeClientOnOutputBufferLimitReached(client *c, int async) {
    if (!c->conn) return 0; /* It is unsafe to free fake clients. */
    serverAssert(c->reply_bytes < SIZE_MAX-(1024*64)); /* actual memory only, logical memory may exceed SIZE_MAX */
    /* Note that c->reply_bytes is irrelevant for replica clients
     * (they use the global repl buffers). */
    if ((c->reply_bytes == 0 && c->reply_bytes_shared == 0 && !clientTypeIsSlave(c)) ||
        c->flags & CLIENT_CLOSE_ASAP) return 0;
    if (checkClientOutputBufferLimits(c)) {
        sds client = catClientInfoString(sdsempty(),c);

        if (async) {
            freeClientAsync(c);
            serverLog(LL_WARNING,
                      "Client %s scheduled to be closed ASAP for overcoming of output buffer limits.",
                      client);
        } else {
            freeClient(c);
            serverLog(LL_WARNING,
                      "Client %s closed for overcoming of output buffer limits.",
                      client);
        }
        sdsfree(client);
        server.stat_client_outbuf_limit_disconnections++;
        return  1;
    }
    return 0;
}

/* Helper function used by performEvictions() in order to flush slaves
 * output buffers without returning control to the event loop.
 * This is also called by SHUTDOWN for a best-effort attempt to send
 * slaves the latest writes. */
void flushSlavesOutputBuffers(void) {
    listIter li;
    listNode *ln;

    listRewind(server.slaves,&li);
    while((ln = listNext(&li))) {
        client *slave = listNodeValue(ln);

        /* Fetch the replica clients that are currently running in IO thread. */
        if (slave->running_tid != IOTHREAD_MAIN_THREAD_ID) {
            fetchClientFromIOThread(slave);
            /* If the slave doesn't have any pending replies nothing to do
             * anyways. */
            if (!clientHasPendingReplies(slave)) continue;
            putClientInPendingWriteQueue(slave);
        }

        int can_receive_writes = connHasWriteHandler(slave->conn) ||
                                 (slave->flags & CLIENT_PENDING_WRITE);

        /* We don't want to send the pending data to the replica in a few
         * cases:
         *
         * 1. For some reason there is neither the write handler installed
         *    nor the client is flagged as to have pending writes: for some
         *    reason this replica may not be set to receive data. This is
         *    just for the sake of defensive programming.
         *
         * 2. The put_online_on_ack flag is true. To know why we don't want
         *    to send data to the replica in this case, please grep for the
         *    flag for this flag.
         *
         * 3. Obviously if the slave is not ONLINE.
         */
        if ((slave->replstate == SLAVE_STATE_ONLINE || slave->replstate == SLAVE_STATE_SEND_BULK_AND_STREAM) &&
            !(slave->flags & CLIENT_CLOSE_ASAP) &&
            can_receive_writes &&
            !slave->repl_start_cmd_stream_on_ack &&
            clientHasPendingReplies(slave))
        {
            writeToClient(slave,0);
        }
    }
    /* Record the flushed offset for flushSlavesOutputBuffersIfNeeded(). */
    server.repl_last_flush_offset = server.master_repl_offset;
}

/* Send pending data to replicas once ~1MB has built up, instead of waiting
 * for beforeSleep, so a busy loop keeps feeding replicas as it goes rather
 * than in one burst at the end.
 *
 * Normally we write to replicas only once per event loop, at beforeSleep.
 * If one loop produces 10MB, the replica socket stays empty during the
 * loop, then at beforeSleep we try to send all 10MB at once. The socket
 * buffer only fits ~4MB, so 4MB is sent and 6MB waits, with a write handler
 * installed for the rest. But that handler runs about once per loop and only
 * sends another ~4MB, while every loop adds 10MB more, so the buffer keeps
 * growing. It eventually hits the replica client-output-buffer-limit, the
 * master drops the replica, and it has to full sync. Sending every ~1MB
 * during the loop feeds the replica as data is produced, so the buffer stays
 * small and the replica keeps up. */
#define REPL_FLUSH_THRESHOLD (1024*1024)
void flushSlavesOutputBuffersIfNeeded(void) {
    /* Only with io threads off. When on, replicas are written from a separate 
     * thread in parallel, so they keep up on their own even under load. */
    if (server.io_threads_num <= 1 && listLength(server.slaves) &&
        server.master_repl_offset - server.repl_last_flush_offset > REPL_FLUSH_THRESHOLD)
    {
        flushSlavesOutputBuffers();
    }
}

/* Compute current paused actions and its end time, aggregated for
 * all pause purposes. */
void updatePausedActions(void) {
    uint32_t prev_paused_actions = server.paused_actions;
    server.paused_actions = 0;

    for (int i = 0; i < NUM_PAUSE_PURPOSES; i++) {
        pause_event *p = &(server.client_pause_per_purpose[i]);
        if (p->end > server.mstime)
            server.paused_actions |= p->paused_actions;
        else {
            p->paused_actions = 0;
            p->end = 0;
        }
    }

    /* If the pause type is less restrictive than before, we unblock all clients
     * so they are reprocessed (may get re-paused). */
    uint32_t mask_cli = (PAUSE_ACTION_CLIENT_WRITE|PAUSE_ACTION_CLIENT_ALL);
    if ((server.paused_actions & mask_cli) < (prev_paused_actions & mask_cli)) {
        unblockPostponedClients();
    }
}

/* Unblock all paused clients (ones that where blocked by BLOCKED_POSTPONE (possibly in processCommand).
 * This means they'll get re-processed in beforeSleep, and may get paused again if needed. */
void unblockPostponedClients(void) {
    listNode *ln;
    listIter li;
    listRewind(server.postponed_clients, &li);
    while ((ln = listNext(&li)) != NULL) {
        client *c = listNodeValue(ln);
        unblockClient(c, 1);
    }
}

/* Set pause-client end-time and restricted action. If already paused, then:
 * 1. Keep higher end-time value between configured and the new one
 * 2. Keep most restrictive action between configured and the new one */
static void pauseClientsByClient(mstime_t endTime, int isPauseClientAll) {
    uint32_t actions;
    pause_event *p = &server.client_pause_per_purpose[PAUSE_BY_CLIENT_COMMAND];

    if (isPauseClientAll)
        actions = PAUSE_ACTIONS_CLIENT_ALL_SET;
    else {
        actions = PAUSE_ACTIONS_CLIENT_WRITE_SET;
        /* If currently configured most restrictive client pause, then keep it */
        if (p->paused_actions & PAUSE_ACTION_CLIENT_ALL)
            actions = PAUSE_ACTIONS_CLIENT_ALL_SET;
    }

    /* Cancel all ASM tasks when starting client pause */
    clusterAsmCancel(NULL, "client pause requested");

    pauseActions(PAUSE_BY_CLIENT_COMMAND, endTime, actions);
}

/* Pause actions up to the specified unixtime (in ms) for a given type of
 * commands.
 *
 * A main use case of this function is to allow pausing replication traffic
 * so that a failover without data loss to occur. Replicas will continue to receive
 * traffic to facilitate this functionality.
 * 
 * This function is also internally used by Redis Cluster for the manual
 * failover procedure implemented by CLUSTER FAILOVER.
 *
 * The function always succeed, even if there is already a pause in progress.
 * The new paused_actions of a given 'purpose' will override the old ones and
 * end time will be updated if new end time is bigger than currently configured */
void pauseActions(pause_purpose purpose, mstime_t end, uint32_t actions) {
    /* Manage pause type and end time per pause purpose. */
    server.client_pause_per_purpose[purpose].paused_actions = actions;

    /* If currently configured end time bigger than new one, then keep it */
    if (server.client_pause_per_purpose[purpose].end < end)
        server.client_pause_per_purpose[purpose].end = end;

    updatePausedActions();

    /* We allow write commands that were queued
     * up before and after to execute. We need
     * to track this state so that we don't assert
     * in propagateNow(). */
    if (server.in_exec) {
        server.client_pause_in_transaction = 1;
    }

    /* Assert that there is no import task in progress when we are pausing.
     * otherwise we break the promise that no writes are performed, maybe
     * causing data lost during a failover. */
    if (isPausedActions(PAUSE_ACTION_CLIENT_ALL) ||
        isPausedActions(PAUSE_ACTION_CLIENT_WRITE))
        serverAssert(!asmImportInProgress());
}

/* Unpause actions and queue them for reprocessing. */
void unpauseActions(pause_purpose purpose) {
    server.client_pause_per_purpose[purpose].end = 0;
    server.client_pause_per_purpose[purpose].paused_actions = 0;
    updatePausedActions();
}

/* Returns bitmask of paused actions */
uint32_t isPausedActions(uint32_t actions_bitmask) {
    return (server.paused_actions & actions_bitmask);
}

/* Returns bitmask of paused actions */
uint32_t isPausedActionsWithUpdate(uint32_t actions_bitmask) {
    if (!(server.paused_actions & actions_bitmask)) return 0;
    updatePausedActions();
    return (server.paused_actions & actions_bitmask);
}

/* This function is called by Redis in order to process a few events from
 * time to time while blocked into some not interruptible operation.
 * This allows to reply to clients with the -LOADING error while loading the
 * data set at startup or after a full resynchronization with the master
 * and so forth.
 *
 * It calls the event loop in order to process a few events. Specifically we
 * try to call the event loop 4 times as long as we receive acknowledge that
 * some event was processed, in order to go forward with the accept, read,
 * write, close sequence needed to serve a client.
 *
 * The function returns the total number of events processed. */
void processEventsWhileBlocked(void) {
    int iterations = 4; /* See the function top-comment. */

    /* Update our cached time since it is used to create and update the last
     * interaction time with clients and for other important things. */
    updateCachedTime(0);

    /* For the few commands that are allowed during busy scripts, we rather
     * provide a fresher time than the one from when the script started (they
     * still won't get it from the call due to execution_nesting. For commands
     * during loading this doesn't matter. */
    mstime_t prev_cmd_time_snapshot = server.cmd_time_snapshot;
    server.cmd_time_snapshot = server.mstime;

    /* Note: when we are processing events while blocked (for instance during
     * busy Lua scripts), we set a global flag. When such flag is set, we
     * avoid handling the read part of clients using threaded I/O.
     * See https://github.com/redis/redis/issues/6988 for more info.
     * Note that there could be cases of nested calls to this function,
     * specifically on a busy script during async_loading rdb, and scripts
     * that came from AOF. */
    ProcessingEventsWhileBlocked++;
    while (iterations--) {
        long long startval = server.events_processed_while_blocked;
        long long ae_events = aeProcessEvents(server.el,
            AE_FILE_EVENTS|AE_DONT_WAIT|
            AE_CALL_BEFORE_SLEEP|AE_CALL_AFTER_SLEEP);
        /* Note that server.events_processed_while_blocked will also get
         * incremented by callbacks called by the event loop handlers. */
        server.events_processed_while_blocked += ae_events;
        long long events = server.events_processed_while_blocked - startval;
        if (!events) break;
    }

    whileBlockedCron();

    ProcessingEventsWhileBlocked--;
    serverAssert(ProcessingEventsWhileBlocked >= 0);

    server.cmd_time_snapshot = prev_cmd_time_snapshot;
}

/* Returns the actual client eviction limit based on current configuration or
 * 0 if no limit. */
size_t getClientEvictionLimit(void) {
    size_t maxmemory_clients_actual = SIZE_MAX;

    /* Handle percentage of maxmemory*/
    if (server.maxmemory_clients < 0 && server.maxmemory > 0) {
        unsigned long long maxmemory_clients_bytes = (unsigned long long)((double)server.maxmemory * -(double) server.maxmemory_clients / 100);
        if (maxmemory_clients_bytes <= SIZE_MAX)
            maxmemory_clients_actual = maxmemory_clients_bytes;
    }
    else if (server.maxmemory_clients > 0)
        maxmemory_clients_actual = server.maxmemory_clients;
    else
        return 0;

    /* Don't allow a too small maxmemory-clients to avoid cases where we can't communicate
     * at all with the server because of bad configuration */
    if (maxmemory_clients_actual < 1024*128)
        maxmemory_clients_actual = 1024*128;

    return maxmemory_clients_actual;
}

void evictClients(void) {
    if (!server.client_mem_usage_buckets)
        return;
    /* Start eviction from topmost bucket (largest clients) */
    int curr_bucket = CLIENT_MEM_USAGE_BUCKETS-1;
    listIter bucket_iter;
    listRewind(server.client_mem_usage_buckets[curr_bucket].clients, &bucket_iter);
    size_t client_eviction_limit = getClientEvictionLimit();
    if (client_eviction_limit == 0)
        return;
    while (server.stat_clients_type_memory[CLIENT_TYPE_NORMAL] +
           server.stat_clients_type_memory[CLIENT_TYPE_PUBSUB] >= client_eviction_limit) {
        listNode *ln = listNext(&bucket_iter);
        if (ln) {
            client *c = ln->value;
            size_t last_memory = c->last_memory_usage;
            int tid = c->running_tid;
            if (tid != IOTHREAD_MAIN_THREAD_ID) {
                pauseIOThread(tid);
                /* We need to update the client memory usage and bucket if the client
                 * is running in IO thread. This is because the client memory usage
                 * and bucket are updated 'only' in the main thread, such as processing
                 * command and clientsCron, it may delay updating, to avoid incorrectly
                 * evicting clients, we update again before evicting, if the memory
                 * used by the client does not decrease or memory usage bucket is not
                 * changed, then we will evict it, otherwise, not evict it. */
                updateClientMemUsageAndBucket(c);
            }
            if (c->last_memory_usage >= last_memory ||
                c->mem_usage_bucket == &server.client_mem_usage_buckets[curr_bucket])
            {
                sds ci = catClientInfoString(sdsempty(),c);
                serverLog(LL_NOTICE, "Evicting client: %s", ci);
                freeClient(c);
                sdsfree(ci);
                server.stat_evictedclients++;
            }
            if (tid != IOTHREAD_MAIN_THREAD_ID) {
                resumeIOThread(tid);
                /* The 'next' of 'bucket_iter' may be changed after updating client memory
                 * usage and freeing client, so let reset 'bucket_iter'. */
                listRewind(server.client_mem_usage_buckets[curr_bucket].clients, &bucket_iter);
            }
        } else {
            curr_bucket--;
            if (curr_bucket < 0) {
                serverLog(LL_WARNING, "Over client maxmemory after evicting all evictable clients");
                break;
            }
            listRewind(server.client_mem_usage_buckets[curr_bucket].clients, &bucket_iter);
        }
    }
}

/* Acquire a pending command from the shared pool or allocate a new one.
 * Uses the shared pool when available (only when IO threads are inactive),
 * otherwise allocates a new pending command structure. */
static pendingCommand *acquirePendingCommand(void) {
    /* Ensure pool is empty when IO threads are active to avoid race conditions */
    serverAssert(server.io_threads_active == 0 || server.cmd_pool.size == 0);

    pendingCommand *pcmd = NULL;
    if (server.cmd_pool.size > 0) {
        /* Shared pool is available. */
        pcmd = server.cmd_pool.pool[--server.cmd_pool.size];
        server.cmd_pool.pool[server.cmd_pool.size] = NULL;

        /* Track minimum pool size for utilization calculation */
        if (server.cmd_pool.size < server.cmd_pool.min_size)
            server.cmd_pool.min_size = server.cmd_pool.size;
    } else {
        /* Shared pool is empty, allocate new pending command. */
        pcmd = zmalloc(sizeof(pendingCommand));
        initPendingCommand(pcmd);
    }
    return pcmd;
}

/* Try to expand the pending command pool capacity.
 * Returns 1 if expansion succeeded or wasn't needed, 0 if expansion failed. */
static int tryExpandPendingCommandPool(void) {
    /* Check if expansion is needed */
    if (server.cmd_pool.size < server.cmd_pool.capacity) {
        return 1; /* No expansion needed */
    }
    
    /* Check if we can expand further */
    if (server.cmd_pool.capacity >= PENDING_COMMAND_POOL_MAX_SIZE) {
        return 0; /* Already at maximum capacity */
    }
    
    /* Expand the pending command pool capacity by doubling it, up to the maximum size */
    int new_capacity = server.cmd_pool.capacity * 2;
    if (new_capacity > PENDING_COMMAND_POOL_MAX_SIZE)
        new_capacity = PENDING_COMMAND_POOL_MAX_SIZE;

    server.cmd_pool.pool = zrealloc(server.cmd_pool.pool, sizeof(pendingCommand*) * new_capacity);
    server.cmd_pool.capacity = new_capacity;
    return 1; /* Expansion succeeded */
}

/* Reclaim a pending command by adding it to the shared pool for reuse or freeing it.
 * The shared pool is only used when IO threads are inactive to avoid race conditions
 * between multiple clients. Additionally, pool reuse provides minimal benefit in
 * multi-threaded scenarios, so we only use it in single-threaded mode. */
static inline void reclaimPendingCommand(client *c, pendingCommand *pcmd) {
    if (!server.io_threads_active) {
        /* Try to add to shared pool for reuse if argv isn't too large */
        if (likely(pcmd->argv_len < 64)) {
            /* Check if pool needs expansion before attempting to add */
            if (!tryExpandPendingCommandPool()) {
                /* Pool is at maximum capacity, can't expand further */
                goto free_command;
            }

            /* Clean up command resources before adding to pool */
            for (int j = 0; j < pcmd->argc; j++)
                decrRefCount(pcmd->argv[j]);

            getKeysFreeResult(&pcmd->keys_result);

            if (c) {
                serverAssert(c->all_argv_len_sum >= pcmd->argv_len_sum); /* assert this doesn't try to go negative */
                c->all_argv_len_sum -= pcmd->argv_len_sum;
                pcmd->argv_len_sum = 0;
            }

            /* Reset the pending command while preserving the argv array for shared pool reuse */
            robj **argv = pcmd->argv;
            int argv_len = pcmd->argv_len;
            memset(pcmd, 0, sizeof(pendingCommand));
            pcmd->argv = argv;
            pcmd->argv_len = argv_len;
            pcmd->slot = INVALID_CLUSTER_SLOT;

            server.cmd_pool.pool[server.cmd_pool.size++] = pcmd;
            return; /* Successfully added to shared pool for reuse */
        }
    } else {
        /* IO threads are active, handle thread-specific cleanup */
        if (c && c->tid != IOTHREAD_MAIN_THREAD_ID) {
            /* Partial cleanup for IO thread commands to avoid race issues.
             * To avoid robj that may already be referenced elsewhere, we should
             * decrease the reference count to release our reference to it. */
            for (int j = 0; j < pcmd->argc; j++) {
                robj *o = pcmd->argv[j];
                if (o && o->refcount > 1) {
                    decrRefCount(o);
                    pcmd->argv[j] = NULL;
                }
            }

            serverAssert(c->all_argv_len_sum >= pcmd->argv_len_sum); /* assert this doesn't try to go negative */
            c->all_argv_len_sum -= pcmd->argv_len_sum;
            pcmd->argv_len_sum = 0;

            tryDeferFreeClientObject(c, DEFERRED_OBJECT_TYPE_PENDING_COMMAND, pcmd);
            return;
        }
    }

free_command:
    /* Shared pool is full or command argv is too large, free this pending command */
    freePendingCommand(c, pcmd);
}

void initPendingCommand(pendingCommand *pcmd) {
    memset(pcmd, 0, sizeof(pendingCommand));
    pcmd->slot = INVALID_CLUSTER_SLOT;
}

void freePendingCommand(client *c, pendingCommand *pcmd) {
    if (!pcmd)
        return;

    getKeysFreeResult(&pcmd->keys_result);

    if (pcmd->argv) {
        for (int j = 0; j < pcmd->argc; j++) {
            robj *o = pcmd->argv[j];
            if (!o) continue; /* argv[j] may be NULL when called from reclaimPendingCommand */
            decrRefCount(o);
        }

        zfree(pcmd->argv);

        /* c may be NULL when called from reclaimPendingCommand */
        if (c) {
            serverAssert(c->all_argv_len_sum >= pcmd->argv_len_sum); /* assert this doesn't try to go negative */
            c->all_argv_len_sum -= pcmd->argv_len_sum;
        }
    }

    zfree(pcmd);
}

/* Add a command to the tail of the pending command list. */
void addPendingCommand(pendingCommandList *queue, pendingCommand *cmd) {
    cmd->next = NULL;
    cmd->prev = queue->tail;

    if (queue->tail) {
        queue->tail->next = cmd;
    } else {
        /* Queue was empty */
        queue->head = cmd;
    }

    queue->tail = cmd;
    queue->len++;
    if (!(cmd->flags & PENDING_CMD_FLAG_INCOMPLETE)) queue->ready_len++;
}

/* Remove and return the first pending command from the list.
 * Returns NULL if the list is empty. */
pendingCommand *popPendingCommandFromHead(pendingCommandList *list) {
    pendingCommand *cmd = list->head;
    if (!cmd) return NULL;  /* List is empty */

    list->head = cmd->next;
    if (list->head) {
        list->head->prev = NULL;
    } else {
        /* Queue was empty */
        list->tail = NULL;
    }

    cmd->next = cmd->prev = NULL;
    list->len--;
    if (!(cmd->flags & PENDING_CMD_FLAG_INCOMPLETE)) list->ready_len--;
    return cmd;
}

/* Remove and return the last pending command from the list.
 * Returns NULL if the list is empty. */
pendingCommand *popPendingCommandFromTail(pendingCommandList *list) {
    pendingCommand *cmd = list->tail;
    if (!cmd) return NULL;  /* List is empty */

    list->tail = cmd->prev;
    if (list->tail) {
        list->tail->next = NULL;
    } else {
        /* Queue became empty */
        list->head = NULL;
    }

    cmd->next = cmd->prev = NULL;
    list->len--;
    if (!(cmd->flags & PENDING_CMD_FLAG_INCOMPLETE)) list->ready_len--;
    return cmd;
}

/* Get cached key result for current pending command */
getKeysResult *getClientCachedKeyResult(client *c) {
    pendingCommand *pcmd = c->current_pending_cmd;
    if (pcmd) {
        /* Preprocess the command if needed */
        if (!(pcmd->flags & PENDING_CMD_FLAG_PREPROCESSED)) {
            preprocessCommand(c, pcmd);
            pcmd->flags |= PENDING_CMD_FLAG_PREPROCESSED;
        }

        /* Return cached result if available */
        if (pcmd->flags & PENDING_CMD_KEYS_RESULT_VALID)
            return &c->current_pending_cmd->keys_result;
    }
    return NULL;
}

void shrinkPendingCommandPool(void) {
    /* Don't shrink if pool is too small. */
    if (server.cmd_pool.capacity <= PENDING_COMMAND_POOL_SIZE) return;

    /* Free commands until we have half the current size, but not below minimum. */
    int target_size = max(server.cmd_pool.size / 2, PENDING_COMMAND_POOL_SIZE);

    while (server.cmd_pool.size > target_size) {
        pendingCommand *cmd = server.cmd_pool.pool[--server.cmd_pool.size];
        if (cmd) {
            freePendingCommand(NULL, cmd);
            server.cmd_pool.pool[server.cmd_pool.size] = NULL;
        }
    }

    int old_capacity = server.cmd_pool.capacity;
    server.cmd_pool.capacity = target_size;
    server.cmd_pool.pool = zrealloc(server.cmd_pool.pool, sizeof(pendingCommand*) * target_size);
    serverLog(LL_DEBUG, "Shrunk pending command pool: capacity %d->%d", old_capacity, server.cmd_pool.capacity);
}