[linux.git] / fs / afs / vl_probe.c

// SPDX-License-Identifier: GPL-2.0-or-later
/* AFS vlserver probing
 *
 * Copyright (C) 2018 Red Hat, Inc. All Rights Reserved.
 * Written by David Howells ([email protected])
 */

#include <linux/sched.h>
#include <linux/slab.h>
#include "afs_fs.h"
#include "internal.h"
#include "protocol_yfs.h"


/*
 * Handle the completion of a set of probes.
 */
static void afs_finished_vl_probe(struct afs_vlserver *server)
{
	if (!(server->probe.flags & AFS_VLSERVER_PROBE_RESPONDED)) {
		server->rtt = UINT_MAX;
		clear_bit(AFS_VLSERVER_FL_RESPONDING, &server->flags);
	}

	clear_bit_unlock(AFS_VLSERVER_FL_PROBING, &server->flags);
	wake_up_bit(&server->flags, AFS_VLSERVER_FL_PROBING);
}

/*
 * Handle the completion of a probe RPC call.
 */
static void afs_done_one_vl_probe(struct afs_vlserver *server, bool wake_up)
{
	if (atomic_dec_and_test(&server->probe_outstanding)) {
		afs_finished_vl_probe(server);
		wake_up = true;
	}

	if (wake_up)
		wake_up_all(&server->probe_wq);
}

/*
 * Process the result of probing a vlserver.  This is called after successful
 * or failed delivery of an VL.GetCapabilities operation.
 */
void afs_vlserver_probe_result(struct afs_call *call)
{
	struct afs_addr_list *alist = call->alist;
	struct afs_vlserver *server = call->vlserver;
	unsigned int server_index = call->server_index;
	unsigned int rtt_us = 0;
	unsigned int index = call->addr_ix;
	bool have_result = false;
	int ret = call->error;

	_enter("%s,%u,%u,%d,%d", server->name, server_index, index, ret, call->abort_code);

	spin_lock(&server->probe_lock);

	switch (ret) {
	case 0:
		server->probe.error = 0;
		goto responded;
	case -ECONNABORTED:
		if (!(server->probe.flags & AFS_VLSERVER_PROBE_RESPONDED)) {
			server->probe.abort_code = call->abort_code;
			server->probe.error = ret;
		}
		goto responded;
	case -ENOMEM:
	case -ENONET:
	case -EKEYEXPIRED:
	case -EKEYREVOKED:
	case -EKEYREJECTED:
		server->probe.flags |= AFS_VLSERVER_PROBE_LOCAL_FAILURE;
		if (server->probe.error == 0)
			server->probe.error = ret;
		trace_afs_io_error(call->debug_id, ret, afs_io_error_vl_probe_fail);
		goto out;
	case -ECONNRESET: /* Responded, but call expired. */
	case -ERFKILL:
	case -EADDRNOTAVAIL:
	case -ENETUNREACH:
	case -EHOSTUNREACH:
	case -EHOSTDOWN:
	case -ECONNREFUSED:
	case -ETIMEDOUT:
	case -ETIME:
	default:
		clear_bit(index, &alist->responded);
		set_bit(index, &alist->failed);
		if (!(server->probe.flags & AFS_VLSERVER_PROBE_RESPONDED) &&
		    (server->probe.error == 0 ||
		     server->probe.error == -ETIMEDOUT ||
		     server->probe.error == -ETIME))
			server->probe.error = ret;
		trace_afs_io_error(call->debug_id, ret, afs_io_error_vl_probe_fail);
		goto out;
	}

responded:
	set_bit(index, &alist->responded);
	clear_bit(index, &alist->failed);

	if (call->service_id == YFS_VL_SERVICE) {
		server->probe.flags |= AFS_VLSERVER_PROBE_IS_YFS;
		set_bit(AFS_VLSERVER_FL_IS_YFS, &server->flags);
		alist->addrs[index].srx_service = call->service_id;
	} else {
		server->probe.flags |= AFS_VLSERVER_PROBE_NOT_YFS;
		if (!(server->probe.flags & AFS_VLSERVER_PROBE_IS_YFS)) {
			clear_bit(AFS_VLSERVER_FL_IS_YFS, &server->flags);
			alist->addrs[index].srx_service = call->service_id;
		}
	}

	if (rxrpc_kernel_get_srtt(call->net->socket, call->rxcall, &rtt_us) &&
	    rtt_us < server->probe.rtt) {
		server->probe.rtt = rtt_us;
		server->rtt = rtt_us;
		alist->preferred = index;
	}

	smp_wmb(); /* Set rtt before responded. */
	server->probe.flags |= AFS_VLSERVER_PROBE_RESPONDED;
	set_bit(AFS_VLSERVER_FL_PROBED, &server->flags);
	set_bit(AFS_VLSERVER_FL_RESPONDING, &server->flags);
	have_result = true;
out:
	spin_unlock(&server->probe_lock);

	_debug("probe [%u][%u] %pISpc rtt=%u ret=%d",
	       server_index, index, &alist->addrs[index].transport, rtt_us, ret);

	afs_done_one_vl_probe(server, have_result);
}

/*
 * Probe all of a vlserver's addresses to find out the best route and to
 * query its capabilities.
 */
static bool afs_do_probe_vlserver(struct afs_net *net,
				  struct afs_vlserver *server,
				  struct key *key,
				  unsigned int server_index,
				  struct afs_error *_e)
{
	struct afs_addr_cursor ac = {
		.index = 0,
	};
	struct afs_call *call;
	bool in_progress = false;

	_enter("%s", server->name);

	read_lock(&server->lock);
	ac.alist = rcu_dereference_protected(server->addresses,
					     lockdep_is_held(&server->lock));
	read_unlock(&server->lock);

	atomic_set(&server->probe_outstanding, ac.alist->nr_addrs);
	memset(&server->probe, 0, sizeof(server->probe));
	server->probe.rtt = UINT_MAX;

	for (ac.index = 0; ac.index < ac.alist->nr_addrs; ac.index++) {
		call = afs_vl_get_capabilities(net, &ac, key, server,
					       server_index);
		if (!IS_ERR(call)) {
			afs_put_call(call);
			in_progress = true;
		} else {
			afs_prioritise_error(_e, PTR_ERR(call), ac.abort_code);
			afs_done_one_vl_probe(server, false);
		}
	}

	return in_progress;
}

/*
 * Send off probes to all unprobed servers.
 */
int afs_send_vl_probes(struct afs_net *net, struct key *key,
		       struct afs_vlserver_list *vllist)
{
	struct afs_vlserver *server;
	struct afs_error e;
	bool in_progress = false;
	int i;

	e.error = 0;
	e.responded = false;
	for (i = 0; i < vllist->nr_servers; i++) {
		server = vllist->servers[i].server;
		if (test_bit(AFS_VLSERVER_FL_PROBED, &server->flags))
			continue;

		if (!test_and_set_bit_lock(AFS_VLSERVER_FL_PROBING, &server->flags) &&
		    afs_do_probe_vlserver(net, server, key, i, &e))
			in_progress = true;
	}

	return in_progress ? 0 : e.error;
}

/*
 * Wait for the first as-yet untried server to respond.
 */
int afs_wait_for_vl_probes(struct afs_vlserver_list *vllist,
			   unsigned long untried)
{
	struct wait_queue_entry *waits;
	struct afs_vlserver *server;
	unsigned int rtt = UINT_MAX, rtt_s;
	bool have_responders = false;
	int pref = -1, i;

	_enter("%u,%lx", vllist->nr_servers, untried);

	/* Only wait for servers that have a probe outstanding. */
	for (i = 0; i < vllist->nr_servers; i++) {
		if (test_bit(i, &untried)) {
			server = vllist->servers[i].server;
			if (!test_bit(AFS_VLSERVER_FL_PROBING, &server->flags))
				__clear_bit(i, &untried);
			if (server->probe.flags & AFS_VLSERVER_PROBE_RESPONDED)
				have_responders = true;
		}
	}
	if (have_responders || !untried)
		return 0;

	waits = kmalloc(array_size(vllist->nr_servers, sizeof(*waits)), GFP_KERNEL);
	if (!waits)
		return -ENOMEM;

	for (i = 0; i < vllist->nr_servers; i++) {
		if (test_bit(i, &untried)) {
			server = vllist->servers[i].server;
			init_waitqueue_entry(&waits[i], current);
			add_wait_queue(&server->probe_wq, &waits[i]);
		}
	}

	for (;;) {
		bool still_probing = false;

		set_current_state(TASK_INTERRUPTIBLE);
		for (i = 0; i < vllist->nr_servers; i++) {
			if (test_bit(i, &untried)) {
				server = vllist->servers[i].server;
				if (server->probe.flags & AFS_VLSERVER_PROBE_RESPONDED)
					goto stop;
				if (test_bit(AFS_VLSERVER_FL_PROBING, &server->flags))
					still_probing = true;
			}
		}

		if (!still_probing || signal_pending(current))
			goto stop;
		schedule();
	}

stop:
	set_current_state(TASK_RUNNING);

	for (i = 0; i < vllist->nr_servers; i++) {
		if (test_bit(i, &untried)) {
			server = vllist->servers[i].server;
			rtt_s = READ_ONCE(server->rtt);
			if (test_bit(AFS_VLSERVER_FL_RESPONDING, &server->flags) &&
			    rtt_s < rtt) {
				pref = i;
				rtt = rtt_s;
			}

			remove_wait_queue(&server->probe_wq, &waits[i]);
		}
	}

	kfree(waits);

	if (pref == -1 && signal_pending(current))
		return -ERESTARTSYS;

	if (pref >= 0)
		vllist->preferred = pref;

	_leave(" = 0 [%u]", pref);
	return 0;
}
Commit	Line	Data
b4d0d230	1	// SPDX-License-Identifier: GPL-2.0-or-later
3bf0fb6f DH	2	/* AFS vlserver probing
	3	*
	4	* Copyright (C) 2018 Red Hat, Inc. All Rights Reserved.
	5	* Written by David Howells ([email protected])
3bf0fb6f DH	6	*/
	7
	8	#include <linux/sched.h>
	9	#include <linux/slab.h>
	10	#include "afs_fs.h"
	11	#include "internal.h"
	12	#include "protocol_yfs.h"
	13
b95b3094 DH	14
	15	/*
	16	* Handle the completion of a set of probes.
	17	*/
	18	static void afs_finished_vl_probe(struct afs_vlserver *server)
3bf0fb6f	19	{
b95b3094 DH	20	if (!(server->probe.flags & AFS_VLSERVER_PROBE_RESPONDED)) {
	21	server->rtt = UINT_MAX;
	22	clear_bit(AFS_VLSERVER_FL_RESPONDING, &server->flags);
	23	}
3bf0fb6f	24
3bf0fb6f DH	25	clear_bit_unlock(AFS_VLSERVER_FL_PROBING, &server->flags);
3bf0fb6f DH	26	wake_up_bit(&server->flags, AFS_VLSERVER_FL_PROBING);
b95b3094 DH	27	}
	28
	29	/*
	30	* Handle the completion of a probe RPC call.
	31	*/
	32	static void afs_done_one_vl_probe(struct afs_vlserver *server, bool wake_up)
	33	{
	34	if (atomic_dec_and_test(&server->probe_outstanding)) {
	35	afs_finished_vl_probe(server);
	36	wake_up = true;
	37	}
	38
	39	if (wake_up)
	40	wake_up_all(&server->probe_wq);
3bf0fb6f DH	41	}
	42
	43	/*
	44	* Process the result of probing a vlserver. This is called after successful
	45	* or failed delivery of an VL.GetCapabilities operation.
	46	*/
	47	void afs_vlserver_probe_result(struct afs_call *call)
	48	{
	49	struct afs_addr_list *alist = call->alist;
ffba718e DH	50	struct afs_vlserver *server = call->vlserver;
ffba718e DH	51	unsigned int server_index = call->server_index;
c410bf01	52	unsigned int rtt_us = 0;
3bf0fb6f	53	unsigned int index = call->addr_ix;
3bf0fb6f	54	bool have_result = false;
3bf0fb6f DH	55	int ret = call->error;
	56
	57	_enter("%s,%u,%u,%d,%d", server->name, server_index, index, ret, call->abort_code);
	58
	59	spin_lock(&server->probe_lock);
	60
	61	switch (ret) {
	62	case 0:
	63	server->probe.error = 0;
	64	goto responded;
	65	case -ECONNABORTED:
fb72cd3d	66	if (!(server->probe.flags & AFS_VLSERVER_PROBE_RESPONDED)) {
3bf0fb6f DH	67	server->probe.abort_code = call->abort_code;
	68	server->probe.error = ret;
	69	}
	70	goto responded;
	71	case -ENOMEM:
	72	case -ENONET:
b95b3094 DH	73	case -EKEYEXPIRED:
	74	case -EKEYREVOKED:
	75	case -EKEYREJECTED:
fb72cd3d	76	server->probe.flags \|= AFS_VLSERVER_PROBE_LOCAL_FAILURE;
b95b3094 DH	77	if (server->probe.error == 0)
	78	server->probe.error = ret;
	79	trace_afs_io_error(call->debug_id, ret, afs_io_error_vl_probe_fail);
3bf0fb6f DH	80	goto out;
3bf0fb6f DH	81	case -ECONNRESET: /* Responded, but call expired. */
4584ae96 DH	82	case -ERFKILL:
4584ae96 DH	83	case -EADDRNOTAVAIL:
3bf0fb6f DH	84	case -ENETUNREACH:
3bf0fb6f DH	85	case -EHOSTUNREACH:
4584ae96	86	case -EHOSTDOWN:
3bf0fb6f DH	87	case -ECONNREFUSED:
	88	case -ETIMEDOUT:
	89	case -ETIME:
	90	default:
	91	clear_bit(index, &alist->responded);
	92	set_bit(index, &alist->failed);
fb72cd3d	93	if (!(server->probe.flags & AFS_VLSERVER_PROBE_RESPONDED) &&
3bf0fb6f DH	94	(server->probe.error == 0 \|\|
	95	server->probe.error == -ETIMEDOUT \|\|
	96	server->probe.error == -ETIME))
	97	server->probe.error = ret;
b95b3094	98	trace_afs_io_error(call->debug_id, ret, afs_io_error_vl_probe_fail);
3bf0fb6f DH	99	goto out;
	100	}
	101
	102	responded:
	103	set_bit(index, &alist->responded);
	104	clear_bit(index, &alist->failed);
	105
	106	if (call->service_id == YFS_VL_SERVICE) {
fb72cd3d	107	server->probe.flags \|= AFS_VLSERVER_PROBE_IS_YFS;
3bf0fb6f DH	108	set_bit(AFS_VLSERVER_FL_IS_YFS, &server->flags);
	109	alist->addrs[index].srx_service = call->service_id;
	110	} else {
fb72cd3d DH	111	server->probe.flags \|= AFS_VLSERVER_PROBE_NOT_YFS;
fb72cd3d DH	112	if (!(server->probe.flags & AFS_VLSERVER_PROBE_IS_YFS)) {
3bf0fb6f DH	113	clear_bit(AFS_VLSERVER_FL_IS_YFS, &server->flags);
	114	alist->addrs[index].srx_service = call->service_id;
	115	}
	116	}
	117
1d4adfaf DH	118	if (rxrpc_kernel_get_srtt(call->net->socket, call->rxcall, &rtt_us) &&
1d4adfaf DH	119	rtt_us < server->probe.rtt) {
c410bf01	120	server->probe.rtt = rtt_us;
b95b3094	121	server->rtt = rtt_us;
3bf0fb6f	122	alist->preferred = index;
3bf0fb6f DH	123	}
	124
	125	smp_wmb(); /* Set rtt before responded. */
fb72cd3d	126	server->probe.flags \|= AFS_VLSERVER_PROBE_RESPONDED;
3bf0fb6f	127	set_bit(AFS_VLSERVER_FL_PROBED, &server->flags);
b95b3094 DH	128	set_bit(AFS_VLSERVER_FL_RESPONDING, &server->flags);
b95b3094 DH	129	have_result = true;
3bf0fb6f DH	130	out:
	131	spin_unlock(&server->probe_lock);
	132
	133	_debug("probe [%u][%u] %pISpc rtt=%u ret=%d",
c410bf01	134	server_index, index, &alist->addrs[index].transport, rtt_us, ret);
3bf0fb6f	135
b95b3094	136	afs_done_one_vl_probe(server, have_result);
3bf0fb6f DH	137	}
	138
	139	/*
	140	* Probe all of a vlserver's addresses to find out the best route and to
	141	* query its capabilities.
	142	*/
4584ae96 DH	143	static bool afs_do_probe_vlserver(struct afs_net *net,
	144	struct afs_vlserver *server,
	145	struct key *key,
	146	unsigned int server_index,
	147	struct afs_error *_e)
3bf0fb6f DH	148	{
	149	struct afs_addr_cursor ac = {
	150	.index = 0,
	151	};
0b9bf381	152	struct afs_call *call;
4584ae96	153	bool in_progress = false;
3bf0fb6f DH	154
	155	_enter("%s", server->name);
	156
	157	read_lock(&server->lock);
	158	ac.alist = rcu_dereference_protected(server->addresses,
	159	lockdep_is_held(&server->lock));
	160	read_unlock(&server->lock);
	161
	162	atomic_set(&server->probe_outstanding, ac.alist->nr_addrs);
	163	memset(&server->probe, 0, sizeof(server->probe));
	164	server->probe.rtt = UINT_MAX;
	165
	166	for (ac.index = 0; ac.index < ac.alist->nr_addrs; ac.index++) {
0b9bf381 DH	167	call = afs_vl_get_capabilities(net, &ac, key, server,
	168	server_index);
	169	if (!IS_ERR(call)) {
	170	afs_put_call(call);
4584ae96	171	in_progress = true;
0b9bf381 DH	172	} else {
0b9bf381 DH	173	afs_prioritise_error(_e, PTR_ERR(call), ac.abort_code);
b95b3094	174	afs_done_one_vl_probe(server, false);
0b9bf381	175	}
3bf0fb6f DH	176	}
3bf0fb6f DH	177
4584ae96	178	return in_progress;
3bf0fb6f DH	179	}
	180
	181	/*
	182	* Send off probes to all unprobed servers.
	183	*/
	184	int afs_send_vl_probes(struct afs_net net, struct key key,
	185	struct afs_vlserver_list *vllist)
	186	{
	187	struct afs_vlserver *server;
4584ae96 DH	188	struct afs_error e;
	189	bool in_progress = false;
	190	int i;
3bf0fb6f	191
4584ae96 DH	192	e.error = 0;
4584ae96 DH	193	e.responded = false;
3bf0fb6f DH	194	for (i = 0; i < vllist->nr_servers; i++) {
	195	server = vllist->servers[i].server;
	196	if (test_bit(AFS_VLSERVER_FL_PROBED, &server->flags))
	197	continue;
	198
4584ae96 DH	199	if (!test_and_set_bit_lock(AFS_VLSERVER_FL_PROBING, &server->flags) &&
	200	afs_do_probe_vlserver(net, server, key, i, &e))
	201	in_progress = true;
3bf0fb6f DH	202	}
3bf0fb6f DH	203
4584ae96	204	return in_progress ? 0 : e.error;
3bf0fb6f DH	205	}
	206
	207	/*
	208	* Wait for the first as-yet untried server to respond.
	209	*/
	210	int afs_wait_for_vl_probes(struct afs_vlserver_list *vllist,
	211	unsigned long untried)
	212	{
	213	struct wait_queue_entry *waits;
	214	struct afs_vlserver *server;
b95b3094	215	unsigned int rtt = UINT_MAX, rtt_s;
3bf0fb6f DH	216	bool have_responders = false;
	217	int pref = -1, i;
	218
	219	_enter("%u,%lx", vllist->nr_servers, untried);
	220
	221	/* Only wait for servers that have a probe outstanding. */
	222	for (i = 0; i < vllist->nr_servers; i++) {
	223	if (test_bit(i, &untried)) {
	224	server = vllist->servers[i].server;
	225	if (!test_bit(AFS_VLSERVER_FL_PROBING, &server->flags))
	226	__clear_bit(i, &untried);
fb72cd3d	227	if (server->probe.flags & AFS_VLSERVER_PROBE_RESPONDED)
3bf0fb6f DH	228	have_responders = true;
	229	}
	230	}
	231	if (have_responders \|\| !untried)
	232	return 0;
	233
	234	waits = kmalloc(array_size(vllist->nr_servers, sizeof(*waits)), GFP_KERNEL);
	235	if (!waits)
	236	return -ENOMEM;
	237
	238	for (i = 0; i < vllist->nr_servers; i++) {
	239	if (test_bit(i, &untried)) {
	240	server = vllist->servers[i].server;
	241	init_waitqueue_entry(&waits[i], current);
	242	add_wait_queue(&server->probe_wq, &waits[i]);
	243	}
	244	}
	245
	246	for (;;) {
	247	bool still_probing = false;
	248
	249	set_current_state(TASK_INTERRUPTIBLE);
	250	for (i = 0; i < vllist->nr_servers; i++) {
	251	if (test_bit(i, &untried)) {
	252	server = vllist->servers[i].server;
fb72cd3d	253	if (server->probe.flags & AFS_VLSERVER_PROBE_RESPONDED)
3bf0fb6f DH	254	goto stop;
	255	if (test_bit(AFS_VLSERVER_FL_PROBING, &server->flags))
	256	still_probing = true;
	257	}
	258	}
	259
08d405c8	260	if (!still_probing \|\| signal_pending(current))
3bf0fb6f DH	261	goto stop;
	262	schedule();
	263	}
	264
	265	stop:
	266	set_current_state(TASK_RUNNING);
	267
	268	for (i = 0; i < vllist->nr_servers; i++) {
	269	if (test_bit(i, &untried)) {
	270	server = vllist->servers[i].server;
b95b3094 DH	271	rtt_s = READ_ONCE(server->rtt);
	272	if (test_bit(AFS_VLSERVER_FL_RESPONDING, &server->flags) &&
	273	rtt_s < rtt) {
3bf0fb6f	274	pref = i;
b95b3094	275	rtt = rtt_s;
3bf0fb6f DH	276	}
	277
	278	remove_wait_queue(&server->probe_wq, &waits[i]);
	279	}
	280	}
	281
	282	kfree(waits);
	283
	284	if (pref == -1 && signal_pending(current))
	285	return -ERESTARTSYS;
	286
	287	if (pref >= 0)
	288	vllist->preferred = pref;
	289
	290	_leave(" = 0 [%u]", pref);
	291	return 0;
	292	}