scoutfs/kmod/src/client.c

/*
 * Copyright (C) 2017 Versity Software, Inc.  All rights reserved.
 *
 * This program is free software; you can redistribute it and/or
 * modify it under the terms of the GNU General Public
 * License v2 as published by the Free Software Foundation.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * General Public License for more details.
 */
#include <linux/kernel.h>
#include <linux/fs.h>
#include <linux/slab.h>
#include <asm/ioctls.h>
#include <linux/net.h>
#include <linux/inet.h>
#include <linux/in.h>
#include <net/sock.h>
#include <net/tcp.h>
#include <linux/sort.h>
#include <asm/barrier.h>

#include "format.h"
#include "counters.h"
#include "inode.h"
#include "btree.h"
#include "manifest.h"
#include "seg.h"
#include "compact.h"
#include "scoutfs_trace.h"
#include "msg.h"
#include "server.h"
#include "client.h"
#include "sock.h"
#include "endian_swap.h"

/*
 * Client callers block sending requests to the server.  Senders connect
 * and send down the socket in their blocked context under a mutex.
 * Once a socket is connected recv work is fired up.  Destroying a
 * socket shuts down the socket and cancels the work.
 *
 * Clients are responsible for resending their requests after
 * reconnecting to a new socket.  These new socket connections might be
 * connecting to the same server.  The message sending and processing
 * paths are responsible for dealing with duplicate requests.
 */

#define SIN_FMT		"%pIS:%u"
#define SIN_ARG(sin)	sin, be16_to_cpu((sin)->sin_port)

/*
 * Have a pretty aggressive keepalive timeout of around 10 seconds.  The
 * TCP keepalives are being processed out of task context so they should
 * be responsive even when mounts are under load.  We also derive the
 * connect timeout from this.
 */
#define KEEPCNT			3
#define KEEPIDLE		7
#define KEEPINTVL		1
#define KEEP_TIMEO_SECS		(KEEPIDLE + (KEEPCNT * KEEPINTVL))
#define CONNECT_TIMEO_SECS	KEEP_TIMEO_SECS
#define CONNECT_TIMEO_MSECS	(KEEP_TIMEO_SECS * MSEC_PER_SEC)

struct client_info {
	struct super_block *sb;

	/* spinlock protects quick critical sections between send,recv,umount */
	spinlock_t recv_lock;
	struct rb_root sender_root;

	/* the sock mutex serializes connecting and sending */
	struct mutex send_mutex;
	bool recv_shutdown;
	u64 next_id;
	u64 sock_gen;
	struct socket *sock;
	struct sockaddr_in peername;
	struct sockaddr_in sockname;

	/* blocked senders sit on a waitq that's woken for resends */
	wait_queue_head_t waitq;

	struct workqueue_struct *recv_wq;
	struct work_struct recv_work;
};

struct waiting_sender {
	struct rb_node node;
	struct task_struct *task;

	u64 id;
	void *rx;
	size_t rx_size;
	int result;
};

static struct waiting_sender *walk_sender_tree(struct client_info *client,
					       u64 id,
					       struct waiting_sender *ins)
{
	struct rb_node **node = &client->sender_root.rb_node;
	struct waiting_sender *found = NULL;
	struct waiting_sender *sender;
	struct rb_node *parent = NULL;

	assert_spin_locked(&client->recv_lock);

	while (*node) {
		parent = *node;
		sender = container_of(*node, struct waiting_sender, node);

		if (id < sender->id) {
			node = &(*node)->rb_left;
		} else if (id > sender->id) {
			node = &(*node)->rb_right;
		} else {
			found = sender;
			break;
		}
	}

	if (ins) {
		/* ids are never reused and assigned under lock */
		BUG_ON(found);
		rb_link_node(&ins->node, parent, node);
		rb_insert_color(&ins->node, &client->sender_root);
		found = ins;
	}

	return found;
}

/*
 * This work is queued once the socket is created.  It blocks trying to
 * receive replies to sent messages.  If the sender is still around it
 * receives the reply data into their buffer.  If the sender has left
 * then it silently drops the reply.
 *
 * This exits once someone shuts down the socket.  If this sees a fatal
 * error it shuts down the socket which causes senders to reconnect.
 */
static void scoutfs_client_recv_func(struct work_struct *work)
{
	struct client_info *client = container_of(work, struct client_info,
						  recv_work);
	struct waiting_sender *sender;
	struct scoutfs_net_header nh;
	void *rx_alloc = NULL;
	int result = 0;
	u16 data_len;
	void *rx;
	int ret;

	for (;;) {
		/* receive the header */
		ret = scoutfs_sock_recvmsg(client->sock, &nh, sizeof(nh));
		if (ret)
			break;

		data_len = le16_to_cpu(nh.data_len);

		trace_scoutfs_client_recv_reply(client->sb,
						&client->sockname,
						&client->peername, &nh);

		/* see if we have a waiting sender */
		spin_lock(&client->recv_lock);
		sender = walk_sender_tree(client, le64_to_cpu(nh.id), NULL);
		spin_unlock(&client->recv_lock);

		if (sender) {
			if (sender->rx_size < data_len) {
				/* protocol mismatch is fatal */
				rx = NULL;
				result = -EIO;
			} else {
				rx = sender->rx;
				result = 0;
			}
		} else {
			rx = NULL;
		}

		if (!rx) {
			kfree(rx_alloc);
			rx_alloc = kmalloc(data_len, GFP_NOFS);
			if (!rx_alloc) {
				ret = -ENOMEM;
				break;
			}
			rx = rx_alloc;
		}

		/* recv failure can be server crashing, not fatal */
		ret = scoutfs_sock_recvmsg(client->sock, rx, data_len);
		if (ret) {
			break;
		}

		if (sender) {
			/* lock to keep sender around until after we wake */
			spin_lock(&client->recv_lock);
			sender->result = result;
			smp_mb(); /* store result before waking */
			wake_up_process(sender->task);
			spin_unlock(&client->recv_lock);
		}
	}

	/* make senders reconnect if we see an rx error */
	if (ret) {
		/* XXX would need to break out send */
		kernel_sock_shutdown(client->sock, SHUT_RDWR);
		client->recv_shutdown = true;
	}

	kfree(rx_alloc);
}


/*
 * Spin discovering the address of the server and trying to connect to
 * it until either we connect or we're interrupted by a signal.
 *
 * A single mount coming up starts both the server and the client.  The
 * server takes a few IOs and network messages to get going and communicate
 * its address.  We want to aggressively retry getting the address so that
 * these mounts can be quick.  But we back off to avoid storms waiting for
 * recovery after an existing server explodes.
 */
static int client_connect(struct client_info *client)
{
	struct super_block *sb = client->sb;
	struct scoutfs_super_block super;
	struct sockaddr_in *sin;
	struct socket *sock = NULL;
	struct timeval tv;
	unsigned int msecs = MSEC_PER_SEC / 10;
	int addrlen;
	int optval;
	int ret;

	BUG_ON(!mutex_is_locked(&client->send_mutex));

	for(;;) {
		if (sock) {
			sock_release(sock);
			sock = NULL;
		}

		ret = scoutfs_read_supers(sb, &super);
		if (ret)
			continue;

		if (super.server_addr.addr == cpu_to_le32(INADDR_ANY)) {
			msleep_interruptible(msecs);
			if (msecs < CONNECT_TIMEO_MSECS)
				msecs = max(msecs + MSEC_PER_SEC,
					    CONNECT_TIMEO_MSECS);
			continue;
		}

		if (signal_pending(current)) {
			ret = -ERESTARTSYS;
			break;
		}

		sin = &client->peername;
		sin->sin_family = AF_INET;
		sin->sin_addr.s_addr = le32_to_be32(super.server_addr.addr);
		sin->sin_port = le16_to_be16(super.server_addr.port);

		ret = sock_create_kern(AF_INET, SOCK_STREAM, IPPROTO_TCP,
				       &sock);
		if (ret)
			continue;

		optval = 1;
		ret = kernel_setsockopt(sock, SOL_TCP, TCP_NODELAY,
					(char *)&optval, sizeof(optval));
		if (ret)
			continue;

		/* start with a connect timeout */
		tv.tv_sec = CONNECT_TIMEO_SECS;
		tv.tv_usec = 0;
		ret = kernel_setsockopt(sock, SOL_SOCKET, SO_SNDTIMEO,
					(char *)&tv, sizeof(tv));
		if (ret)
			continue;

		client->sock = sock;

		ret = kernel_connect(sock, (struct sockaddr *)sin,
				     sizeof(struct sockaddr_in), 0);
		if (ret)
			continue;

		/* but use a keepalive timeout instead of send timeout */
		tv.tv_sec = 0;
		tv.tv_usec = 0;
		ret = kernel_setsockopt(sock, SOL_SOCKET, SO_SNDTIMEO,
					(char *)&tv, sizeof(tv));
		if (ret)
			continue;

		optval = KEEPCNT;
		ret = kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT,
					(char *)&optval, sizeof(optval));
		if (ret)
			continue;

		optval = KEEPIDLE;
		ret = kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE,
					(char *)&optval, sizeof(optval));
		if (ret)
			continue;

		optval = KEEPINTVL;
		ret = kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL,
					(char *)&optval, sizeof(optval));
		if (ret)
			continue;

		optval = 1;
		ret = kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
					(char *)&optval, sizeof(optval));
		if (ret)
			continue;

		addrlen = sizeof(struct sockaddr_in);
		ret = kernel_getsockname(sock,
					 (struct sockaddr *)&client->sockname,
					 &addrlen);
		if (ret)
			continue;

		scoutfs_info(sb, "client connected "SIN_FMT" -> "SIN_FMT,
			     SIN_ARG(&client->sockname),
			     SIN_ARG(&client->peername));

		client->sock_gen++;
		client->recv_shutdown = false;
		queue_work(client->recv_wq, &client->recv_work);
		wake_up(&client->waitq);
		ret = 0;
		break;
	}

	if (ret && sock)
		sock_release(sock);

	return ret;
}

/* either a sender or unmount is destroying the socket */
static void shutdown_sock_sync(struct client_info *client)
{
	struct super_block *sb = client->sb;
	struct socket *sock = client->sock;

	if (sock) {
		kernel_sock_shutdown(sock, SHUT_RDWR);
		cancel_work_sync(&client->recv_work);
		sock_release(sock);
		client->sock = NULL;

		scoutfs_info(sb, "client disconnected "SIN_FMT" -> "SIN_FMT,
			     SIN_ARG(&client->sockname),
			     SIN_ARG(&client->peername));
	}
}

/*
 * Senders sleep waiting for a reply to come down the connection out
 * which they just sent a request.  They need to wake up when the recv
 * work has given them a reply or when it's given up and the sender
 * needs to reconnect and resend.
 *
 * This is a condition for wait_event.  The barrier orders the task
 * state store before loading the sender and client fields.
 */
static int sender_should_wake(struct client_info *client,
			      struct waiting_sender *sender)
{
	smp_mb();
	return sender->result != -EINPROGRESS || client->recv_shutdown;
}

/*
 * Block sending a request and then waiting for the reply.  All senders
 * are responsible for connecting sockets and sending their requests.
 * recv work blocks receiving from the socket and waking senders if
 * they're reply has been copied to their buffer.  If the socket sees an
 * error the recv work will shutdown and wake us to reconnect.
 */
static int client_request(struct client_info *client, int type, void *data,
			  unsigned data_len, void *rx, size_t rx_size)
{
	struct waiting_sender sender;
	struct scoutfs_net_header nh;
	struct kvec kv[2];
	unsigned kv_len;
	u64 sent_to_gen = ~0ULL;
	int ret = 0;

	if (WARN_ON_ONCE(!data && data_len))
		return -EINVAL;

	spin_lock(&client->recv_lock);

	sender.task = current;
	sender.id = client->next_id++;
	sender.rx = rx;
	sender.rx_size = rx_size;
	sender.result = -EINPROGRESS;

	nh.id = cpu_to_le64(sender.id);
	nh.data_len = cpu_to_le16(data_len);
	nh.type = type;
	nh.status = SCOUTFS_NET_STATUS_REQUEST;

	walk_sender_tree(client, sender.id, &sender);

	spin_unlock(&client->recv_lock);

	mutex_lock(&client->send_mutex);

	while (sender.result == -EINPROGRESS) {

		if (!client->sock) {
			ret = client_connect(client);
			if (ret < 0)
				break;
		}

		if (sent_to_gen != client->sock_gen) {
			kv[0].iov_base = &nh;
			kv[0].iov_len = sizeof(nh);
			kv[1].iov_base = data;
			kv[1].iov_len = data_len;
			kv_len = data ? 2 : 1;

			trace_scoutfs_client_send_request(client->sb,
							  &client->sockname,
							  &client->peername,
							  &nh);

			ret = scoutfs_sock_sendmsg(client->sock, kv, kv_len);
			if (ret) {
				shutdown_sock_sync(client);
				continue;
			}

			sent_to_gen = client->sock_gen;
		}

		/* XXX would need to protect erase during rx if interruptible */
		mutex_unlock(&client->send_mutex);

		wait_event(client->waitq, sender_should_wake(client, &sender));

		mutex_lock(&client->send_mutex);

		/* finish tearing down the socket if recv shutdown */
		if (client->sock && client->recv_shutdown) {
			shutdown_sock_sync(client);
			continue;
		}
	}

	mutex_unlock(&client->send_mutex);

	/* safe to remove, we only finish after canceling recv or we're woke */
	spin_lock(&client->recv_lock);
	rb_erase(&sender.node, &client->sender_root);
	spin_unlock(&client->recv_lock);

	if (ret == 0)
		ret = sender.result;

	return ret;
}

int scoutfs_client_alloc_inodes(struct super_block *sb)
{
	struct client_info *client = SCOUTFS_SB(sb)->client_info;
	struct scoutfs_net_inode_alloc ial;
	u64 ino = 0;
	u64 nr = 0;
	int ret;

	ret = client_request(client, SCOUTFS_NET_ALLOC_INODES, NULL, 0,
			     &ial, sizeof(ial));
	if (ret == 0) {
		ino = le64_to_cpu(ial.ino);
		nr = le64_to_cpu(ial.nr);

		/* catch wrapping */
		if (ino + nr < ino)
			ret = -EINVAL;
	}

	if (ret < 0)
		scoutfs_inode_fill_pool(sb, 0, 0);
	else
		scoutfs_inode_fill_pool(sb, ino, nr);

	return ret;
}

int scoutfs_client_alloc_segno(struct super_block *sb, u64 *segno)
{
	struct client_info *client = SCOUTFS_SB(sb)->client_info;
	__le64 lesegno;
	int ret;

	ret = client_request(client, SCOUTFS_NET_ALLOC_SEGNO, NULL, 0,
			     &lesegno, sizeof(lesegno));
	if (ret == 0) {
		if (lesegno == 0)
			ret = -ENOSPC;
		else
			*segno = le64_to_cpu(lesegno);
	}

	return ret;
}

int scoutfs_client_record_segment(struct super_block *sb,
				  struct scoutfs_segment *seg, u8 level)
{
	struct client_info *client = SCOUTFS_SB(sb)->client_info;
	struct scoutfs_net_manifest_entry *net_ment;
	struct scoutfs_manifest_entry ment;
	int ret;

	scoutfs_seg_init_ment(&ment, level, seg);
	net_ment = scoutfs_alloc_net_ment(&ment);
	if (net_ment) {
		ret = client_request(client, SCOUTFS_NET_RECORD_SEGMENT,
				     net_ment, scoutfs_net_ment_bytes(net_ment),
				     NULL, 0);
		kfree(net_ment);
	} else {
		ret = -ENOMEM;
	}

	return ret;
}

static int sort_cmp_u64s(const void *A, const void *B)
{
	const u64 *a = A;
	const u64 *b = B;

	return *a < *b ? -1  : *a > *b ? 1 : 0;
}

static void sort_swap_u64s(void *A, void *B, int size)
{
	u64 *a = A;
	u64 *b = B;

	swap(*a, *b);
}

/*
 * Returns a 0-terminated allocated array of segnos, the caller is
 * responsible for freeing it.
 *
 * This double alloc is silly.  But the caller does have an easier time
 * with native u64s.  We'll probably clean this up.
 */
u64 *scoutfs_client_bulk_alloc(struct super_block *sb)
{
	struct client_info *client = SCOUTFS_SB(sb)->client_info;
	struct scoutfs_net_segnos *ns = NULL;
	u64 *segnos = NULL;
	size_t size;
	unsigned nr;
	u64 prev;
	int ret;
	int i;

	size = offsetof(struct scoutfs_net_segnos,
			segnos[SCOUTFS_BULK_ALLOC_COUNT]);
	ns = kmalloc(size, GFP_NOFS);
	if (!ns) {
		ret = -ENOMEM;
		goto out;
	}

	ret = client_request(client, SCOUTFS_NET_BULK_ALLOC, NULL, 0, ns, size);
	if (ret)
		goto out;

	nr = le16_to_cpu(ns->nr);
	if (nr == 0) {
		ret = -ENOSPC;
		goto out;
	}

	if (nr > SCOUTFS_BULK_ALLOC_COUNT) {
		ret = -EINVAL;
		goto out;
	}

	segnos = kmalloc_array(nr + 1, sizeof(*segnos), GFP_NOFS);
	if (segnos == NULL) {
		ret = -ENOMEM;
		goto out;
	}

	for (i = 0; i < nr; i++)
		segnos[i] = le64_to_cpu(ns->segnos[i]);
	segnos[nr] = 0;

	/* sort segnos for the caller so they can merge easily */
	sort(segnos, nr, sizeof(segnos[0]), sort_cmp_u64s, sort_swap_u64s);

	/* make sure they're all non-zero and unique */
	prev = 0;
	for (i = 0; i < nr; i++) {
		if (segnos[i] == prev) {
			ret = -EINVAL;
			goto out;
		}
		prev = segnos[i];
	}

	ret = 0;
out:
	kfree(ns);
	if (ret) {
		kfree(segnos);
		segnos = ERR_PTR(ret);
	}

	return segnos;
}

int scoutfs_client_advance_seq(struct super_block *sb, u64 *seq)
{
	struct client_info *client = SCOUTFS_SB(sb)->client_info;
	__le64 before = cpu_to_le64p(seq);
	__le64 after;
	int ret;

	ret = client_request(client, SCOUTFS_NET_ADVANCE_SEQ,
			     &before, sizeof(before), &after, sizeof(after));
	if (ret == 0)
		*seq = le64_to_cpu(after);

	return ret;
}

int scoutfs_client_get_last_seq(struct super_block *sb, u64 *seq)
{
	struct client_info *client = SCOUTFS_SB(sb)->client_info;
	__le64 last_seq;
	int ret;

	ret = client_request(client, SCOUTFS_NET_GET_LAST_SEQ,
			     NULL, 0, &last_seq, sizeof(last_seq));
	if (ret == 0)
		*seq = le64_to_cpu(last_seq);

	return ret;
}

int scoutfs_client_get_manifest_root(struct super_block *sb,
				     struct scoutfs_btree_root *root)
{
	struct client_info *client = SCOUTFS_SB(sb)->client_info;

	return client_request(client, SCOUTFS_NET_GET_MANIFEST_ROOT,
			      NULL, 0, root, sizeof(struct scoutfs_btree_root));
}

int scoutfs_client_setup(struct super_block *sb)
{
	struct scoutfs_sb_info *sbi = SCOUTFS_SB(sb);
	struct client_info *client;

	client = kzalloc(sizeof(struct client_info), GFP_KERNEL);
	if (!client)
		return -ENOMEM;

	client->sb = sb;
	spin_lock_init(&client->recv_lock);
	client->sender_root = RB_ROOT;
	mutex_init(&client->send_mutex);
	init_waitqueue_head(&client->waitq);
	INIT_WORK(&client->recv_work, scoutfs_client_recv_func);

	client->recv_wq = alloc_workqueue("scoutfs_client_recv", WQ_UNBOUND, 1);
	if (!client->recv_wq) {
		kfree(client);
		return -ENOMEM;
	}

	sbi->client_info = client;
	return 0;
}

/*
 * There must be no more callers to the client send functions by the
 * time we get here.  We just need to free the socket if it's
 * still sitting around.
 */
void scoutfs_client_destroy(struct super_block *sb)
{
	struct scoutfs_sb_info *sbi = SCOUTFS_SB(sb);
	struct client_info *client = SCOUTFS_SB(sb)->client_info;

	if (client) {
		shutdown_sock_sync(client);

		cancel_work_sync(&client->recv_work);
		destroy_workqueue(client->recv_wq);

		kfree(client);
		sbi->client_info = NULL;
	}
}