Linux Audio

Check our new training course

Real-Time Linux with PREEMPT_RT training

Feb 18-20, 2025
Register
Loading...
v4.17
   1// SPDX-License-Identifier: GPL-2.0
   2#include <linux/ceph/ceph_debug.h>
   3
   4#include <linux/module.h>
   5#include <linux/types.h>
   6#include <linux/slab.h>
   7#include <linux/random.h>
   8#include <linux/sched.h>
   9
  10#include <linux/ceph/ceph_features.h>
  11#include <linux/ceph/mon_client.h>
  12#include <linux/ceph/libceph.h>
  13#include <linux/ceph/debugfs.h>
  14#include <linux/ceph/decode.h>
  15#include <linux/ceph/auth.h>
  16
  17/*
  18 * Interact with Ceph monitor cluster.  Handle requests for new map
  19 * versions, and periodically resend as needed.  Also implement
  20 * statfs() and umount().
  21 *
  22 * A small cluster of Ceph "monitors" are responsible for managing critical
  23 * cluster configuration and state information.  An odd number (e.g., 3, 5)
  24 * of cmon daemons use a modified version of the Paxos part-time parliament
  25 * algorithm to manage the MDS map (mds cluster membership), OSD map, and
  26 * list of clients who have mounted the file system.
  27 *
  28 * We maintain an open, active session with a monitor at all times in order to
  29 * receive timely MDSMap updates.  We periodically send a keepalive byte on the
  30 * TCP socket to ensure we detect a failure.  If the connection does break, we
  31 * randomly hunt for a new monitor.  Once the connection is reestablished, we
  32 * resend any outstanding requests.
  33 */
  34
  35static const struct ceph_connection_operations mon_con_ops;
  36
  37static int __validate_auth(struct ceph_mon_client *monc);
  38
  39/*
  40 * Decode a monmap blob (e.g., during mount).
  41 */
  42struct ceph_monmap *ceph_monmap_decode(void *p, void *end)
  43{
  44	struct ceph_monmap *m = NULL;
  45	int i, err = -EINVAL;
  46	struct ceph_fsid fsid;
  47	u32 epoch, num_mon;
 
  48	u32 len;
  49
  50	ceph_decode_32_safe(&p, end, len, bad);
  51	ceph_decode_need(&p, end, len, bad);
  52
  53	dout("monmap_decode %p %p len %d\n", p, end, (int)(end-p));
  54	p += sizeof(u16);  /* skip version */
 
  55
  56	ceph_decode_need(&p, end, sizeof(fsid) + 2*sizeof(u32), bad);
  57	ceph_decode_copy(&p, &fsid, sizeof(fsid));
  58	epoch = ceph_decode_32(&p);
  59
  60	num_mon = ceph_decode_32(&p);
  61	ceph_decode_need(&p, end, num_mon*sizeof(m->mon_inst[0]), bad);
  62
  63	if (num_mon > CEPH_MAX_MON)
  64		goto bad;
  65	m = kmalloc(sizeof(*m) + sizeof(m->mon_inst[0])*num_mon, GFP_NOFS);
  66	if (m == NULL)
  67		return ERR_PTR(-ENOMEM);
  68	m->fsid = fsid;
  69	m->epoch = epoch;
  70	m->num_mon = num_mon;
  71	ceph_decode_copy(&p, m->mon_inst, num_mon*sizeof(m->mon_inst[0]));
  72	for (i = 0; i < num_mon; i++)
  73		ceph_decode_addr(&m->mon_inst[i].addr);
  74
  75	dout("monmap_decode epoch %d, num_mon %d\n", m->epoch,
  76	     m->num_mon);
  77	for (i = 0; i < m->num_mon; i++)
  78		dout("monmap_decode  mon%d is %s\n", i,
  79		     ceph_pr_addr(&m->mon_inst[i].addr.in_addr));
  80	return m;
  81
  82bad:
  83	dout("monmap_decode failed with %d\n", err);
  84	kfree(m);
  85	return ERR_PTR(err);
  86}
  87
  88/*
  89 * return true if *addr is included in the monmap.
  90 */
  91int ceph_monmap_contains(struct ceph_monmap *m, struct ceph_entity_addr *addr)
  92{
  93	int i;
  94
  95	for (i = 0; i < m->num_mon; i++)
  96		if (memcmp(addr, &m->mon_inst[i].addr, sizeof(*addr)) == 0)
  97			return 1;
  98	return 0;
  99}
 100
 101/*
 102 * Send an auth request.
 103 */
 104static void __send_prepared_auth_request(struct ceph_mon_client *monc, int len)
 105{
 106	monc->pending_auth = 1;
 107	monc->m_auth->front.iov_len = len;
 108	monc->m_auth->hdr.front_len = cpu_to_le32(len);
 109	ceph_msg_revoke(monc->m_auth);
 110	ceph_msg_get(monc->m_auth);  /* keep our ref */
 111	ceph_con_send(&monc->con, monc->m_auth);
 112}
 113
 114/*
 115 * Close monitor session, if any.
 116 */
 117static void __close_session(struct ceph_mon_client *monc)
 118{
 119	dout("__close_session closing mon%d\n", monc->cur_mon);
 120	ceph_msg_revoke(monc->m_auth);
 121	ceph_msg_revoke_incoming(monc->m_auth_reply);
 122	ceph_msg_revoke(monc->m_subscribe);
 123	ceph_msg_revoke_incoming(monc->m_subscribe_ack);
 124	ceph_con_close(&monc->con);
 125
 126	monc->pending_auth = 0;
 127	ceph_auth_reset(monc->auth);
 128}
 129
 130/*
 131 * Pick a new monitor at random and set cur_mon.  If we are repicking
 132 * (i.e. cur_mon is already set), be sure to pick a different one.
 133 */
 134static void pick_new_mon(struct ceph_mon_client *monc)
 135{
 136	int old_mon = monc->cur_mon;
 137
 138	BUG_ON(monc->monmap->num_mon < 1);
 139
 140	if (monc->monmap->num_mon == 1) {
 141		monc->cur_mon = 0;
 142	} else {
 143		int max = monc->monmap->num_mon;
 144		int o = -1;
 145		int n;
 146
 147		if (monc->cur_mon >= 0) {
 148			if (monc->cur_mon < monc->monmap->num_mon)
 149				o = monc->cur_mon;
 150			if (o >= 0)
 151				max--;
 152		}
 153
 154		n = prandom_u32() % max;
 155		if (o >= 0 && n >= o)
 156			n++;
 157
 158		monc->cur_mon = n;
 159	}
 160
 161	dout("%s mon%d -> mon%d out of %d mons\n", __func__, old_mon,
 162	     monc->cur_mon, monc->monmap->num_mon);
 163}
 164
 165/*
 166 * Open a session with a new monitor.
 167 */
 168static void __open_session(struct ceph_mon_client *monc)
 169{
 
 170	int ret;
 171
 172	pick_new_mon(monc);
 173
 174	monc->hunting = true;
 175	if (monc->had_a_connection) {
 176		monc->hunt_mult *= CEPH_MONC_HUNT_BACKOFF;
 177		if (monc->hunt_mult > CEPH_MONC_HUNT_MAX_MULT)
 178			monc->hunt_mult = CEPH_MONC_HUNT_MAX_MULT;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 179	}
 180
 181	monc->sub_renew_after = jiffies; /* i.e., expired */
 182	monc->sub_renew_sent = 0;
 183
 184	dout("%s opening mon%d\n", __func__, monc->cur_mon);
 185	ceph_con_open(&monc->con, CEPH_ENTITY_TYPE_MON, monc->cur_mon,
 186		      &monc->monmap->mon_inst[monc->cur_mon].addr);
 187
 188	/*
 189	 * send an initial keepalive to ensure our timestamp is valid
 190	 * by the time we are in an OPENED state
 191	 */
 192	ceph_con_keepalive(&monc->con);
 193
 194	/* initiate authentication handshake */
 195	ret = ceph_auth_build_hello(monc->auth,
 196				    monc->m_auth->front.iov_base,
 197				    monc->m_auth->front_alloc_len);
 198	BUG_ON(ret <= 0);
 199	__send_prepared_auth_request(monc, ret);
 200}
 201
 202static void reopen_session(struct ceph_mon_client *monc)
 203{
 204	if (!monc->hunting)
 205		pr_info("mon%d %s session lost, hunting for new mon\n",
 206		    monc->cur_mon, ceph_pr_addr(&monc->con.peer_addr.in_addr));
 207
 208	__close_session(monc);
 209	__open_session(monc);
 210}
 211
 212static void un_backoff(struct ceph_mon_client *monc)
 213{
 214	monc->hunt_mult /= 2; /* reduce by 50% */
 215	if (monc->hunt_mult < 1)
 216		monc->hunt_mult = 1;
 217	dout("%s hunt_mult now %d\n", __func__, monc->hunt_mult);
 218}
 219
 220/*
 221 * Reschedule delayed work timer.
 222 */
 223static void __schedule_delayed(struct ceph_mon_client *monc)
 224{
 225	unsigned long delay;
 226
 227	if (monc->hunting)
 228		delay = CEPH_MONC_HUNT_INTERVAL * monc->hunt_mult;
 229	else
 230		delay = CEPH_MONC_PING_INTERVAL;
 231
 232	dout("__schedule_delayed after %lu\n", delay);
 233	mod_delayed_work(system_wq, &monc->delayed_work,
 234			 round_jiffies_relative(delay));
 235}
 236
 237const char *ceph_sub_str[] = {
 238	[CEPH_SUB_MONMAP] = "monmap",
 239	[CEPH_SUB_OSDMAP] = "osdmap",
 240	[CEPH_SUB_FSMAP]  = "fsmap.user",
 241	[CEPH_SUB_MDSMAP] = "mdsmap",
 242};
 243
 244/*
 245 * Send subscribe request for one or more maps, according to
 246 * monc->subs.
 247 */
 248static void __send_subscribe(struct ceph_mon_client *monc)
 249{
 250	struct ceph_msg *msg = monc->m_subscribe;
 251	void *p = msg->front.iov_base;
 252	void *const end = p + msg->front_alloc_len;
 253	int num = 0;
 254	int i;
 255
 256	dout("%s sent %lu\n", __func__, monc->sub_renew_sent);
 257
 258	BUG_ON(monc->cur_mon < 0);
 259
 260	if (!monc->sub_renew_sent)
 261		monc->sub_renew_sent = jiffies | 1; /* never 0 */
 262
 263	msg->hdr.version = cpu_to_le16(2);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 264
 265	for (i = 0; i < ARRAY_SIZE(monc->subs); i++) {
 266		if (monc->subs[i].want)
 267			num++;
 268	}
 269	BUG_ON(num < 1); /* monmap sub is always there */
 270	ceph_encode_32(&p, num);
 271	for (i = 0; i < ARRAY_SIZE(monc->subs); i++) {
 272		char buf[32];
 273		int len;
 274
 275		if (!monc->subs[i].want)
 276			continue;
 277
 278		len = sprintf(buf, "%s", ceph_sub_str[i]);
 279		if (i == CEPH_SUB_MDSMAP &&
 280		    monc->fs_cluster_id != CEPH_FS_CLUSTER_ID_NONE)
 281			len += sprintf(buf + len, ".%d", monc->fs_cluster_id);
 282
 283		dout("%s %s start %llu flags 0x%x\n", __func__, buf,
 284		     le64_to_cpu(monc->subs[i].item.start),
 285		     monc->subs[i].item.flags);
 286		ceph_encode_string(&p, end, buf, len);
 287		memcpy(p, &monc->subs[i].item, sizeof(monc->subs[i].item));
 288		p += sizeof(monc->subs[i].item);
 289	}
 290
 291	BUG_ON(p > end);
 292	msg->front.iov_len = p - msg->front.iov_base;
 293	msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
 294	ceph_msg_revoke(msg);
 295	ceph_con_send(&monc->con, ceph_msg_get(msg));
 296}
 297
 298static void handle_subscribe_ack(struct ceph_mon_client *monc,
 299				 struct ceph_msg *msg)
 300{
 301	unsigned int seconds;
 302	struct ceph_mon_subscribe_ack *h = msg->front.iov_base;
 303
 304	if (msg->front.iov_len < sizeof(*h))
 305		goto bad;
 306	seconds = le32_to_cpu(h->duration);
 307
 308	mutex_lock(&monc->mutex);
 309	if (monc->sub_renew_sent) {
 310		/*
 311		 * This is only needed for legacy (infernalis or older)
 312		 * MONs -- see delayed_work().
 313		 */
 314		monc->sub_renew_after = monc->sub_renew_sent +
 315					    (seconds >> 1) * HZ - 1;
 316		dout("%s sent %lu duration %d renew after %lu\n", __func__,
 317		     monc->sub_renew_sent, seconds, monc->sub_renew_after);
 318		monc->sub_renew_sent = 0;
 319	} else {
 320		dout("%s sent %lu renew after %lu, ignoring\n", __func__,
 321		     monc->sub_renew_sent, monc->sub_renew_after);
 322	}
 
 
 
 323	mutex_unlock(&monc->mutex);
 324	return;
 325bad:
 326	pr_err("got corrupt subscribe-ack msg\n");
 327	ceph_msg_dump(msg);
 328}
 329
 330/*
 331 * Register interest in a map
 332 *
 333 * @sub: one of CEPH_SUB_*
 334 * @epoch: X for "every map since X", or 0 for "just the latest"
 335 */
 336static bool __ceph_monc_want_map(struct ceph_mon_client *monc, int sub,
 337				 u32 epoch, bool continuous)
 338{
 339	__le64 start = cpu_to_le64(epoch);
 340	u8 flags = !continuous ? CEPH_SUBSCRIBE_ONETIME : 0;
 341
 342	dout("%s %s epoch %u continuous %d\n", __func__, ceph_sub_str[sub],
 343	     epoch, continuous);
 344
 345	if (monc->subs[sub].want &&
 346	    monc->subs[sub].item.start == start &&
 347	    monc->subs[sub].item.flags == flags)
 348		return false;
 349
 350	monc->subs[sub].item.start = start;
 351	monc->subs[sub].item.flags = flags;
 352	monc->subs[sub].want = true;
 353
 354	return true;
 355}
 356
 357bool ceph_monc_want_map(struct ceph_mon_client *monc, int sub, u32 epoch,
 358			bool continuous)
 359{
 360	bool need_request;
 361
 362	mutex_lock(&monc->mutex);
 363	need_request = __ceph_monc_want_map(monc, sub, epoch, continuous);
 364	mutex_unlock(&monc->mutex);
 365
 366	return need_request;
 367}
 368EXPORT_SYMBOL(ceph_monc_want_map);
 369
 370/*
 371 * Keep track of which maps we have
 372 *
 373 * @sub: one of CEPH_SUB_*
 374 */
 375static void __ceph_monc_got_map(struct ceph_mon_client *monc, int sub,
 376				u32 epoch)
 377{
 378	dout("%s %s epoch %u\n", __func__, ceph_sub_str[sub], epoch);
 379
 380	if (monc->subs[sub].want) {
 381		if (monc->subs[sub].item.flags & CEPH_SUBSCRIBE_ONETIME)
 382			monc->subs[sub].want = false;
 383		else
 384			monc->subs[sub].item.start = cpu_to_le64(epoch + 1);
 385	}
 386
 387	monc->subs[sub].have = epoch;
 388}
 389
 390void ceph_monc_got_map(struct ceph_mon_client *monc, int sub, u32 epoch)
 391{
 392	mutex_lock(&monc->mutex);
 393	__ceph_monc_got_map(monc, sub, epoch);
 394	mutex_unlock(&monc->mutex);
 
 395}
 396EXPORT_SYMBOL(ceph_monc_got_map);
 397
 398void ceph_monc_renew_subs(struct ceph_mon_client *monc)
 399{
 400	mutex_lock(&monc->mutex);
 401	__send_subscribe(monc);
 
 402	mutex_unlock(&monc->mutex);
 
 403}
 404EXPORT_SYMBOL(ceph_monc_renew_subs);
 405
 406/*
 407 * Wait for an osdmap with a given epoch.
 408 *
 409 * @epoch: epoch to wait for
 410 * @timeout: in jiffies, 0 means "wait forever"
 411 */
 412int ceph_monc_wait_osdmap(struct ceph_mon_client *monc, u32 epoch,
 413			  unsigned long timeout)
 414{
 415	unsigned long started = jiffies;
 416	long ret;
 417
 418	mutex_lock(&monc->mutex);
 419	while (monc->subs[CEPH_SUB_OSDMAP].have < epoch) {
 420		mutex_unlock(&monc->mutex);
 421
 422		if (timeout && time_after_eq(jiffies, started + timeout))
 423			return -ETIMEDOUT;
 424
 425		ret = wait_event_interruptible_timeout(monc->client->auth_wq,
 426				     monc->subs[CEPH_SUB_OSDMAP].have >= epoch,
 427				     ceph_timeout_jiffies(timeout));
 428		if (ret < 0)
 429			return ret;
 430
 431		mutex_lock(&monc->mutex);
 432	}
 433
 434	mutex_unlock(&monc->mutex);
 435	return 0;
 436}
 437EXPORT_SYMBOL(ceph_monc_wait_osdmap);
 438
 439/*
 440 * Open a session with a random monitor.  Request monmap and osdmap,
 441 * which are waited upon in __ceph_open_session().
 442 */
 443int ceph_monc_open_session(struct ceph_mon_client *monc)
 444{
 445	mutex_lock(&monc->mutex);
 446	__ceph_monc_want_map(monc, CEPH_SUB_MONMAP, 0, true);
 447	__ceph_monc_want_map(monc, CEPH_SUB_OSDMAP, 0, false);
 448	__open_session(monc);
 449	__schedule_delayed(monc);
 450	mutex_unlock(&monc->mutex);
 451	return 0;
 452}
 453EXPORT_SYMBOL(ceph_monc_open_session);
 454
 
 
 
 
 455static void ceph_monc_handle_map(struct ceph_mon_client *monc,
 456				 struct ceph_msg *msg)
 457{
 458	struct ceph_client *client = monc->client;
 459	struct ceph_monmap *monmap = NULL, *old = monc->monmap;
 460	void *p, *end;
 461
 462	mutex_lock(&monc->mutex);
 463
 464	dout("handle_monmap\n");
 465	p = msg->front.iov_base;
 466	end = p + msg->front.iov_len;
 467
 468	monmap = ceph_monmap_decode(p, end);
 469	if (IS_ERR(monmap)) {
 470		pr_err("problem decoding monmap, %d\n",
 471		       (int)PTR_ERR(monmap));
 472		goto out;
 473	}
 474
 475	if (ceph_check_fsid(monc->client, &monmap->fsid) < 0) {
 476		kfree(monmap);
 477		goto out;
 478	}
 479
 480	client->monc.monmap = monmap;
 481	kfree(old);
 482
 483	__ceph_monc_got_map(monc, CEPH_SUB_MONMAP, monc->monmap->epoch);
 484	client->have_fsid = true;
 485
 
 
 
 
 
 
 
 486out:
 487	mutex_unlock(&monc->mutex);
 
 488	wake_up_all(&client->auth_wq);
 489}
 490
 491/*
 492 * generic requests (currently statfs, mon_get_version)
 493 */
 494DEFINE_RB_FUNCS(generic_request, struct ceph_mon_generic_request, tid, node)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 495
 496static void release_generic_request(struct kref *kref)
 497{
 498	struct ceph_mon_generic_request *req =
 499		container_of(kref, struct ceph_mon_generic_request, kref);
 500
 501	dout("%s greq %p request %p reply %p\n", __func__, req, req->request,
 502	     req->reply);
 503	WARN_ON(!RB_EMPTY_NODE(&req->node));
 504
 505	if (req->reply)
 506		ceph_msg_put(req->reply);
 507	if (req->request)
 508		ceph_msg_put(req->request);
 509
 510	kfree(req);
 511}
 512
 513static void put_generic_request(struct ceph_mon_generic_request *req)
 514{
 515	if (req)
 516		kref_put(&req->kref, release_generic_request);
 517}
 518
 519static void get_generic_request(struct ceph_mon_generic_request *req)
 520{
 521	kref_get(&req->kref);
 522}
 523
 524static struct ceph_mon_generic_request *
 525alloc_generic_request(struct ceph_mon_client *monc, gfp_t gfp)
 526{
 527	struct ceph_mon_generic_request *req;
 528
 529	req = kzalloc(sizeof(*req), gfp);
 530	if (!req)
 531		return NULL;
 532
 533	req->monc = monc;
 534	kref_init(&req->kref);
 535	RB_CLEAR_NODE(&req->node);
 536	init_completion(&req->completion);
 537
 538	dout("%s greq %p\n", __func__, req);
 539	return req;
 540}
 541
 542static void register_generic_request(struct ceph_mon_generic_request *req)
 543{
 544	struct ceph_mon_client *monc = req->monc;
 545
 546	WARN_ON(req->tid);
 547
 548	get_generic_request(req);
 549	req->tid = ++monc->last_tid;
 550	insert_generic_request(&monc->generic_request_tree, req);
 551}
 552
 553static void send_generic_request(struct ceph_mon_client *monc,
 554				 struct ceph_mon_generic_request *req)
 555{
 556	WARN_ON(!req->tid);
 557
 558	dout("%s greq %p tid %llu\n", __func__, req, req->tid);
 559	req->request->hdr.tid = cpu_to_le64(req->tid);
 560	ceph_con_send(&monc->con, ceph_msg_get(req->request));
 561}
 562
 563static void __finish_generic_request(struct ceph_mon_generic_request *req)
 564{
 565	struct ceph_mon_client *monc = req->monc;
 566
 567	dout("%s greq %p tid %llu\n", __func__, req, req->tid);
 568	erase_generic_request(&monc->generic_request_tree, req);
 569
 570	ceph_msg_revoke(req->request);
 571	ceph_msg_revoke_incoming(req->reply);
 572}
 573
 574static void finish_generic_request(struct ceph_mon_generic_request *req)
 575{
 576	__finish_generic_request(req);
 577	put_generic_request(req);
 578}
 579
 580static void complete_generic_request(struct ceph_mon_generic_request *req)
 581{
 582	if (req->complete_cb)
 583		req->complete_cb(req);
 584	else
 585		complete_all(&req->completion);
 586	put_generic_request(req);
 587}
 588
 589static void cancel_generic_request(struct ceph_mon_generic_request *req)
 590{
 591	struct ceph_mon_client *monc = req->monc;
 592	struct ceph_mon_generic_request *lookup_req;
 593
 594	dout("%s greq %p tid %llu\n", __func__, req, req->tid);
 595
 596	mutex_lock(&monc->mutex);
 597	lookup_req = lookup_generic_request(&monc->generic_request_tree,
 598					    req->tid);
 599	if (lookup_req) {
 600		WARN_ON(lookup_req != req);
 601		finish_generic_request(req);
 602	}
 603
 604	mutex_unlock(&monc->mutex);
 605}
 606
 607static int wait_generic_request(struct ceph_mon_generic_request *req)
 608{
 609	int ret;
 610
 611	dout("%s greq %p tid %llu\n", __func__, req, req->tid);
 612	ret = wait_for_completion_interruptible(&req->completion);
 613	if (ret)
 614		cancel_generic_request(req);
 615	else
 616		ret = req->result; /* completed */
 617
 618	return ret;
 619}
 620
 621static struct ceph_msg *get_generic_reply(struct ceph_connection *con,
 622					 struct ceph_msg_header *hdr,
 623					 int *skip)
 624{
 625	struct ceph_mon_client *monc = con->private;
 626	struct ceph_mon_generic_request *req;
 627	u64 tid = le64_to_cpu(hdr->tid);
 628	struct ceph_msg *m;
 629
 630	mutex_lock(&monc->mutex);
 631	req = lookup_generic_request(&monc->generic_request_tree, tid);
 632	if (!req) {
 633		dout("get_generic_reply %lld dne\n", tid);
 634		*skip = 1;
 635		m = NULL;
 636	} else {
 637		dout("get_generic_reply %lld got %p\n", tid, req->reply);
 638		*skip = 0;
 639		m = ceph_msg_get(req->reply);
 640		/*
 641		 * we don't need to track the connection reading into
 642		 * this reply because we only have one open connection
 643		 * at a time, ever.
 644		 */
 645	}
 646	mutex_unlock(&monc->mutex);
 647	return m;
 648}
 649
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 650/*
 651 * statfs
 652 */
 653static void handle_statfs_reply(struct ceph_mon_client *monc,
 654				struct ceph_msg *msg)
 655{
 656	struct ceph_mon_generic_request *req;
 657	struct ceph_mon_statfs_reply *reply = msg->front.iov_base;
 658	u64 tid = le64_to_cpu(msg->hdr.tid);
 659
 660	dout("%s msg %p tid %llu\n", __func__, msg, tid);
 661
 662	if (msg->front.iov_len != sizeof(*reply))
 663		goto bad;
 
 664
 665	mutex_lock(&monc->mutex);
 666	req = lookup_generic_request(&monc->generic_request_tree, tid);
 667	if (!req) {
 668		mutex_unlock(&monc->mutex);
 669		return;
 
 670	}
 671
 672	req->result = 0;
 673	*req->u.st = reply->st; /* struct */
 674	__finish_generic_request(req);
 675	mutex_unlock(&monc->mutex);
 676
 677	complete_generic_request(req);
 
 
 678	return;
 679
 680bad:
 681	pr_err("corrupt statfs reply, tid %llu\n", tid);
 682	ceph_msg_dump(msg);
 683}
 684
 685/*
 686 * Do a synchronous statfs().
 687 */
 688int ceph_monc_do_statfs(struct ceph_mon_client *monc, u64 data_pool,
 689			struct ceph_statfs *buf)
 690{
 691	struct ceph_mon_generic_request *req;
 692	struct ceph_mon_statfs *h;
 693	int ret = -ENOMEM;
 694
 695	req = alloc_generic_request(monc, GFP_NOFS);
 696	if (!req)
 697		goto out;
 
 
 
 
 
 698
 
 699	req->request = ceph_msg_new(CEPH_MSG_STATFS, sizeof(*h), GFP_NOFS,
 700				    true);
 701	if (!req->request)
 702		goto out;
 703
 704	req->reply = ceph_msg_new(CEPH_MSG_STATFS_REPLY, 64, GFP_NOFS, true);
 705	if (!req->reply)
 706		goto out;
 707
 708	req->u.st = buf;
 709	req->request->hdr.version = cpu_to_le16(2);
 710
 711	mutex_lock(&monc->mutex);
 712	register_generic_request(req);
 713	/* fill out request */
 714	h = req->request->front.iov_base;
 715	h->monhdr.have_version = 0;
 716	h->monhdr.session_mon = cpu_to_le16(-1);
 717	h->monhdr.session_mon_tid = 0;
 718	h->fsid = monc->monmap->fsid;
 719	h->contains_data_pool = (data_pool != CEPH_NOPOOL);
 720	h->data_pool = cpu_to_le64(data_pool);
 721	send_generic_request(monc, req);
 722	mutex_unlock(&monc->mutex);
 723
 724	ret = wait_generic_request(req);
 
 725out:
 726	put_generic_request(req);
 727	return ret;
 728}
 729EXPORT_SYMBOL(ceph_monc_do_statfs);
 730
 731static void handle_get_version_reply(struct ceph_mon_client *monc,
 732				     struct ceph_msg *msg)
 733{
 734	struct ceph_mon_generic_request *req;
 735	u64 tid = le64_to_cpu(msg->hdr.tid);
 736	void *p = msg->front.iov_base;
 737	void *end = p + msg->front_alloc_len;
 738	u64 handle;
 739
 740	dout("%s msg %p tid %llu\n", __func__, msg, tid);
 741
 742	ceph_decode_need(&p, end, 2*sizeof(u64), bad);
 743	handle = ceph_decode_64(&p);
 744	if (tid != 0 && tid != handle)
 745		goto bad;
 746
 747	mutex_lock(&monc->mutex);
 748	req = lookup_generic_request(&monc->generic_request_tree, handle);
 749	if (!req) {
 750		mutex_unlock(&monc->mutex);
 751		return;
 752	}
 753
 754	req->result = 0;
 755	req->u.newest = ceph_decode_64(&p);
 756	__finish_generic_request(req);
 757	mutex_unlock(&monc->mutex);
 758
 759	complete_generic_request(req);
 760	return;
 761
 762bad:
 763	pr_err("corrupt mon_get_version reply, tid %llu\n", tid);
 764	ceph_msg_dump(msg);
 765}
 766
 767static struct ceph_mon_generic_request *
 768__ceph_monc_get_version(struct ceph_mon_client *monc, const char *what,
 769			ceph_monc_callback_t cb, u64 private_data)
 770{
 771	struct ceph_mon_generic_request *req;
 772
 773	req = alloc_generic_request(monc, GFP_NOIO);
 774	if (!req)
 775		goto err_put_req;
 776
 777	req->request = ceph_msg_new(CEPH_MSG_MON_GET_VERSION,
 778				    sizeof(u64) + sizeof(u32) + strlen(what),
 779				    GFP_NOIO, true);
 780	if (!req->request)
 781		goto err_put_req;
 782
 783	req->reply = ceph_msg_new(CEPH_MSG_MON_GET_VERSION_REPLY, 32, GFP_NOIO,
 784				  true);
 785	if (!req->reply)
 786		goto err_put_req;
 787
 788	req->complete_cb = cb;
 789	req->private_data = private_data;
 790
 791	mutex_lock(&monc->mutex);
 792	register_generic_request(req);
 793	{
 794		void *p = req->request->front.iov_base;
 795		void *const end = p + req->request->front_alloc_len;
 796
 797		ceph_encode_64(&p, req->tid); /* handle */
 798		ceph_encode_string(&p, end, what, strlen(what));
 799		WARN_ON(p != end);
 800	}
 801	send_generic_request(monc, req);
 802	mutex_unlock(&monc->mutex);
 803
 804	return req;
 805
 806err_put_req:
 807	put_generic_request(req);
 808	return ERR_PTR(-ENOMEM);
 809}
 810
 811/*
 812 * Send MMonGetVersion and wait for the reply.
 813 *
 814 * @what: one of "mdsmap", "osdmap" or "monmap"
 815 */
 816int ceph_monc_get_version(struct ceph_mon_client *monc, const char *what,
 817			  u64 *newest)
 818{
 819	struct ceph_mon_generic_request *req;
 820	int ret;
 821
 822	req = __ceph_monc_get_version(monc, what, NULL, 0);
 823	if (IS_ERR(req))
 824		return PTR_ERR(req);
 825
 826	ret = wait_generic_request(req);
 827	if (!ret)
 828		*newest = req->u.newest;
 829
 830	put_generic_request(req);
 831	return ret;
 832}
 833EXPORT_SYMBOL(ceph_monc_get_version);
 834
 835/*
 836 * Send MMonGetVersion,
 837 *
 838 * @what: one of "mdsmap", "osdmap" or "monmap"
 839 */
 840int ceph_monc_get_version_async(struct ceph_mon_client *monc, const char *what,
 841				ceph_monc_callback_t cb, u64 private_data)
 842{
 843	struct ceph_mon_generic_request *req;
 844
 845	req = __ceph_monc_get_version(monc, what, cb, private_data);
 846	if (IS_ERR(req))
 847		return PTR_ERR(req);
 848
 849	put_generic_request(req);
 850	return 0;
 851}
 852EXPORT_SYMBOL(ceph_monc_get_version_async);
 853
 854static void handle_command_ack(struct ceph_mon_client *monc,
 855			       struct ceph_msg *msg)
 856{
 857	struct ceph_mon_generic_request *req;
 858	void *p = msg->front.iov_base;
 859	void *const end = p + msg->front_alloc_len;
 860	u64 tid = le64_to_cpu(msg->hdr.tid);
 861
 862	dout("%s msg %p tid %llu\n", __func__, msg, tid);
 863
 864	ceph_decode_need(&p, end, sizeof(struct ceph_mon_request_header) +
 865							    sizeof(u32), bad);
 866	p += sizeof(struct ceph_mon_request_header);
 867
 868	mutex_lock(&monc->mutex);
 869	req = lookup_generic_request(&monc->generic_request_tree, tid);
 870	if (!req) {
 871		mutex_unlock(&monc->mutex);
 872		return;
 
 
 
 
 
 
 
 873	}
 874
 875	req->result = ceph_decode_32(&p);
 876	__finish_generic_request(req);
 877	mutex_unlock(&monc->mutex);
 878
 879	complete_generic_request(req);
 
 
 880	return;
 881
 882bad:
 883	pr_err("corrupt mon_command ack, tid %llu\n", tid);
 884	ceph_msg_dump(msg);
 885}
 886
 887int ceph_monc_blacklist_add(struct ceph_mon_client *monc,
 888			    struct ceph_entity_addr *client_addr)
 
 
 
 
 889{
 890	struct ceph_mon_generic_request *req;
 891	struct ceph_mon_command *h;
 892	int ret = -ENOMEM;
 893	int len;
 894
 895	req = alloc_generic_request(monc, GFP_NOIO);
 896	if (!req)
 897		goto out;
 898
 899	req->request = ceph_msg_new(CEPH_MSG_MON_COMMAND, 256, GFP_NOIO, true);
 
 
 
 
 
 
 
 900	if (!req->request)
 901		goto out;
 902
 903	req->reply = ceph_msg_new(CEPH_MSG_MON_COMMAND_ACK, 512, GFP_NOIO,
 904				  true);
 905	if (!req->reply)
 906		goto out;
 907
 908	mutex_lock(&monc->mutex);
 909	register_generic_request(req);
 910	h = req->request->front.iov_base;
 911	h->monhdr.have_version = 0;
 912	h->monhdr.session_mon = cpu_to_le16(-1);
 913	h->monhdr.session_mon_tid = 0;
 914	h->fsid = monc->monmap->fsid;
 915	h->num_strs = cpu_to_le32(1);
 916	len = sprintf(h->str, "{ \"prefix\": \"osd blacklist\", \
 917		                 \"blacklistop\": \"add\", \
 918				 \"addr\": \"%pISpc/%u\" }",
 919		      &client_addr->in_addr, le32_to_cpu(client_addr->nonce));
 920	h->str_len = cpu_to_le32(len);
 921	send_generic_request(monc, req);
 922	mutex_unlock(&monc->mutex);
 923
 924	ret = wait_generic_request(req);
 925out:
 926	put_generic_request(req);
 927	return ret;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 928}
 929EXPORT_SYMBOL(ceph_monc_blacklist_add);
 930
 931/*
 932 * Resend pending generic requests.
 933 */
 934static void __resend_generic_request(struct ceph_mon_client *monc)
 935{
 936	struct ceph_mon_generic_request *req;
 937	struct rb_node *p;
 938
 939	for (p = rb_first(&monc->generic_request_tree); p; p = rb_next(p)) {
 940		req = rb_entry(p, struct ceph_mon_generic_request, node);
 941		ceph_msg_revoke(req->request);
 942		ceph_msg_revoke_incoming(req->reply);
 943		ceph_con_send(&monc->con, ceph_msg_get(req->request));
 944	}
 945}
 946
 947/*
 948 * Delayed work.  If we haven't mounted yet, retry.  Otherwise,
 949 * renew/retry subscription as needed (in case it is timing out, or we
 950 * got an ENOMEM).  And keep the monitor connection alive.
 951 */
 952static void delayed_work(struct work_struct *work)
 953{
 954	struct ceph_mon_client *monc =
 955		container_of(work, struct ceph_mon_client, delayed_work.work);
 956
 957	dout("monc delayed_work\n");
 958	mutex_lock(&monc->mutex);
 959	if (monc->hunting) {
 960		dout("%s continuing hunt\n", __func__);
 961		reopen_session(monc);
 962	} else {
 963		int is_auth = ceph_auth_is_authenticated(monc->auth);
 964		if (ceph_con_keepalive_expired(&monc->con,
 965					       CEPH_MONC_PING_TIMEOUT)) {
 966			dout("monc keepalive timeout\n");
 967			is_auth = 0;
 968			reopen_session(monc);
 969		}
 970
 971		if (!monc->hunting) {
 972			ceph_con_keepalive(&monc->con);
 973			__validate_auth(monc);
 974			un_backoff(monc);
 975		}
 976
 977		if (is_auth &&
 978		    !(monc->con.peer_features & CEPH_FEATURE_MON_STATEFUL_SUB)) {
 979			unsigned long now = jiffies;
 980
 981			dout("%s renew subs? now %lu renew after %lu\n",
 982			     __func__, now, monc->sub_renew_after);
 983			if (time_after_eq(now, monc->sub_renew_after))
 984				__send_subscribe(monc);
 985		}
 986	}
 987	__schedule_delayed(monc);
 988	mutex_unlock(&monc->mutex);
 989}
 990
 991/*
 992 * On startup, we build a temporary monmap populated with the IPs
 993 * provided by mount(2).
 994 */
 995static int build_initial_monmap(struct ceph_mon_client *monc)
 996{
 997	struct ceph_options *opt = monc->client->options;
 998	struct ceph_entity_addr *mon_addr = opt->mon_addr;
 999	int num_mon = opt->num_mon;
1000	int i;
1001
1002	/* build initial monmap */
1003	monc->monmap = kzalloc(sizeof(*monc->monmap) +
1004			       num_mon*sizeof(monc->monmap->mon_inst[0]),
1005			       GFP_KERNEL);
1006	if (!monc->monmap)
1007		return -ENOMEM;
1008	for (i = 0; i < num_mon; i++) {
1009		monc->monmap->mon_inst[i].addr = mon_addr[i];
1010		monc->monmap->mon_inst[i].addr.nonce = 0;
1011		monc->monmap->mon_inst[i].name.type =
1012			CEPH_ENTITY_TYPE_MON;
1013		monc->monmap->mon_inst[i].name.num = cpu_to_le64(i);
1014	}
1015	monc->monmap->num_mon = num_mon;
 
1016	return 0;
1017}
1018
1019int ceph_monc_init(struct ceph_mon_client *monc, struct ceph_client *cl)
1020{
1021	int err = 0;
1022
1023	dout("init\n");
1024	memset(monc, 0, sizeof(*monc));
1025	monc->client = cl;
1026	monc->monmap = NULL;
1027	mutex_init(&monc->mutex);
1028
1029	err = build_initial_monmap(monc);
1030	if (err)
1031		goto out;
1032
1033	/* connection */
 
 
 
 
 
 
 
1034	/* authentication */
1035	monc->auth = ceph_auth_init(cl->options->name,
1036				    cl->options->key);
1037	if (IS_ERR(monc->auth)) {
1038		err = PTR_ERR(monc->auth);
1039		goto out_monmap;
1040	}
1041	monc->auth->want_keys =
1042		CEPH_ENTITY_TYPE_AUTH | CEPH_ENTITY_TYPE_MON |
1043		CEPH_ENTITY_TYPE_OSD | CEPH_ENTITY_TYPE_MDS;
1044
1045	/* msgs */
1046	err = -ENOMEM;
1047	monc->m_subscribe_ack = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE_ACK,
1048				     sizeof(struct ceph_mon_subscribe_ack),
1049				     GFP_KERNEL, true);
1050	if (!monc->m_subscribe_ack)
1051		goto out_auth;
1052
1053	monc->m_subscribe = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE, 128,
1054					 GFP_KERNEL, true);
1055	if (!monc->m_subscribe)
1056		goto out_subscribe_ack;
1057
1058	monc->m_auth_reply = ceph_msg_new(CEPH_MSG_AUTH_REPLY, 4096,
1059					  GFP_KERNEL, true);
1060	if (!monc->m_auth_reply)
1061		goto out_subscribe;
1062
1063	monc->m_auth = ceph_msg_new(CEPH_MSG_AUTH, 4096, GFP_KERNEL, true);
1064	monc->pending_auth = 0;
1065	if (!monc->m_auth)
1066		goto out_auth_reply;
1067
1068	ceph_con_init(&monc->con, monc, &mon_con_ops,
1069		      &monc->client->msgr);
1070
1071	monc->cur_mon = -1;
1072	monc->had_a_connection = false;
1073	monc->hunt_mult = 1;
 
1074
1075	INIT_DELAYED_WORK(&monc->delayed_work, delayed_work);
1076	monc->generic_request_tree = RB_ROOT;
 
1077	monc->last_tid = 0;
1078
1079	monc->fs_cluster_id = CEPH_FS_CLUSTER_ID_NONE;
1080
 
1081	return 0;
1082
1083out_auth_reply:
1084	ceph_msg_put(monc->m_auth_reply);
1085out_subscribe:
1086	ceph_msg_put(monc->m_subscribe);
1087out_subscribe_ack:
1088	ceph_msg_put(monc->m_subscribe_ack);
1089out_auth:
1090	ceph_auth_destroy(monc->auth);
 
 
1091out_monmap:
1092	kfree(monc->monmap);
1093out:
1094	return err;
1095}
1096EXPORT_SYMBOL(ceph_monc_init);
1097
1098void ceph_monc_stop(struct ceph_mon_client *monc)
1099{
1100	dout("stop\n");
1101	cancel_delayed_work_sync(&monc->delayed_work);
1102
1103	mutex_lock(&monc->mutex);
1104	__close_session(monc);
1105	monc->cur_mon = -1;
 
 
 
 
1106	mutex_unlock(&monc->mutex);
1107
1108	/*
1109	 * flush msgr queue before we destroy ourselves to ensure that:
1110	 *  - any work that references our embedded con is finished.
1111	 *  - any osd_client or other work that may reference an authorizer
1112	 *    finishes before we shut down the auth subsystem.
1113	 */
1114	ceph_msgr_flush();
1115
1116	ceph_auth_destroy(monc->auth);
1117
1118	WARN_ON(!RB_EMPTY_ROOT(&monc->generic_request_tree));
1119
1120	ceph_msg_put(monc->m_auth);
1121	ceph_msg_put(monc->m_auth_reply);
1122	ceph_msg_put(monc->m_subscribe);
1123	ceph_msg_put(monc->m_subscribe_ack);
1124
1125	kfree(monc->monmap);
1126}
1127EXPORT_SYMBOL(ceph_monc_stop);
1128
1129static void finish_hunting(struct ceph_mon_client *monc)
1130{
1131	if (monc->hunting) {
1132		dout("%s found mon%d\n", __func__, monc->cur_mon);
1133		monc->hunting = false;
1134		monc->had_a_connection = true;
1135		un_backoff(monc);
1136		__schedule_delayed(monc);
1137	}
1138}
1139
1140static void handle_auth_reply(struct ceph_mon_client *monc,
1141			      struct ceph_msg *msg)
1142{
1143	int ret;
1144	int was_auth = 0;
1145
1146	mutex_lock(&monc->mutex);
1147	was_auth = ceph_auth_is_authenticated(monc->auth);
 
1148	monc->pending_auth = 0;
1149	ret = ceph_handle_auth_reply(monc->auth, msg->front.iov_base,
1150				     msg->front.iov_len,
1151				     monc->m_auth->front.iov_base,
1152				     monc->m_auth->front_alloc_len);
1153	if (ret > 0) {
1154		__send_prepared_auth_request(monc, ret);
1155		goto out;
1156	}
1157
1158	finish_hunting(monc);
1159
1160	if (ret < 0) {
1161		monc->client->auth_err = ret;
1162	} else if (!was_auth && ceph_auth_is_authenticated(monc->auth)) {
 
 
 
1163		dout("authenticated, starting session\n");
1164
1165		monc->client->msgr.inst.name.type = CEPH_ENTITY_TYPE_CLIENT;
1166		monc->client->msgr.inst.name.num =
1167					cpu_to_le64(monc->auth->global_id);
1168
1169		__send_subscribe(monc);
1170		__resend_generic_request(monc);
1171
1172		pr_info("mon%d %s session established\n", monc->cur_mon,
1173			ceph_pr_addr(&monc->con.peer_addr.in_addr));
1174	}
1175
1176out:
1177	mutex_unlock(&monc->mutex);
1178	if (monc->client->auth_err < 0)
1179		wake_up_all(&monc->client->auth_wq);
1180}
1181
1182static int __validate_auth(struct ceph_mon_client *monc)
1183{
1184	int ret;
1185
1186	if (monc->pending_auth)
1187		return 0;
1188
1189	ret = ceph_build_auth(monc->auth, monc->m_auth->front.iov_base,
1190			      monc->m_auth->front_alloc_len);
1191	if (ret <= 0)
1192		return ret; /* either an error, or no need to authenticate */
1193	__send_prepared_auth_request(monc, ret);
1194	return 0;
1195}
1196
1197int ceph_monc_validate_auth(struct ceph_mon_client *monc)
1198{
1199	int ret;
1200
1201	mutex_lock(&monc->mutex);
1202	ret = __validate_auth(monc);
1203	mutex_unlock(&monc->mutex);
1204	return ret;
1205}
1206EXPORT_SYMBOL(ceph_monc_validate_auth);
1207
1208/*
1209 * handle incoming message
1210 */
1211static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
1212{
1213	struct ceph_mon_client *monc = con->private;
1214	int type = le16_to_cpu(msg->hdr.type);
1215
1216	if (!monc)
1217		return;
1218
1219	switch (type) {
1220	case CEPH_MSG_AUTH_REPLY:
1221		handle_auth_reply(monc, msg);
1222		break;
1223
1224	case CEPH_MSG_MON_SUBSCRIBE_ACK:
1225		handle_subscribe_ack(monc, msg);
1226		break;
1227
1228	case CEPH_MSG_STATFS_REPLY:
1229		handle_statfs_reply(monc, msg);
1230		break;
1231
1232	case CEPH_MSG_MON_GET_VERSION_REPLY:
1233		handle_get_version_reply(monc, msg);
1234		break;
1235
1236	case CEPH_MSG_MON_COMMAND_ACK:
1237		handle_command_ack(monc, msg);
1238		break;
1239
1240	case CEPH_MSG_MON_MAP:
1241		ceph_monc_handle_map(monc, msg);
1242		break;
1243
1244	case CEPH_MSG_OSD_MAP:
1245		ceph_osdc_handle_map(&monc->client->osdc, msg);
1246		break;
1247
1248	default:
1249		/* can the chained handler handle it? */
1250		if (monc->client->extra_mon_dispatch &&
1251		    monc->client->extra_mon_dispatch(monc->client, msg) == 0)
1252			break;
1253			
1254		pr_err("received unknown message type %d %s\n", type,
1255		       ceph_msg_type_name(type));
1256	}
1257	ceph_msg_put(msg);
1258}
1259
1260/*
1261 * Allocate memory for incoming message
1262 */
1263static struct ceph_msg *mon_alloc_msg(struct ceph_connection *con,
1264				      struct ceph_msg_header *hdr,
1265				      int *skip)
1266{
1267	struct ceph_mon_client *monc = con->private;
1268	int type = le16_to_cpu(hdr->type);
1269	int front_len = le32_to_cpu(hdr->front_len);
1270	struct ceph_msg *m = NULL;
1271
1272	*skip = 0;
1273
1274	switch (type) {
1275	case CEPH_MSG_MON_SUBSCRIBE_ACK:
1276		m = ceph_msg_get(monc->m_subscribe_ack);
1277		break;
 
1278	case CEPH_MSG_STATFS_REPLY:
1279	case CEPH_MSG_MON_COMMAND_ACK:
1280		return get_generic_reply(con, hdr, skip);
1281	case CEPH_MSG_AUTH_REPLY:
1282		m = ceph_msg_get(monc->m_auth_reply);
1283		break;
1284	case CEPH_MSG_MON_GET_VERSION_REPLY:
1285		if (le64_to_cpu(hdr->tid) != 0)
1286			return get_generic_reply(con, hdr, skip);
1287
1288		/*
1289		 * Older OSDs don't set reply tid even if the orignal
1290		 * request had a non-zero tid.  Work around this weirdness
1291		 * by allocating a new message.
1292		 */
1293		/* fall through */
1294	case CEPH_MSG_MON_MAP:
1295	case CEPH_MSG_MDS_MAP:
1296	case CEPH_MSG_OSD_MAP:
1297	case CEPH_MSG_FS_MAP_USER:
1298		m = ceph_msg_new(type, front_len, GFP_NOFS, false);
1299		if (!m)
1300			return NULL;	/* ENOMEM--return skip == 0 */
1301		break;
1302	}
1303
1304	if (!m) {
1305		pr_info("alloc_msg unknown type %d\n", type);
1306		*skip = 1;
1307	} else if (front_len > m->front_alloc_len) {
1308		pr_warn("mon_alloc_msg front %d > prealloc %d (%u#%llu)\n",
1309			front_len, m->front_alloc_len,
1310			(unsigned int)con->peer_name.type,
1311			le64_to_cpu(con->peer_name.num));
1312		ceph_msg_put(m);
1313		m = ceph_msg_new(type, front_len, GFP_NOFS, false);
1314	}
1315
1316	return m;
1317}
1318
1319/*
1320 * If the monitor connection resets, pick a new monitor and resubmit
1321 * any pending requests.
1322 */
1323static void mon_fault(struct ceph_connection *con)
1324{
1325	struct ceph_mon_client *monc = con->private;
1326
 
 
 
 
1327	mutex_lock(&monc->mutex);
1328	dout("%s mon%d\n", __func__, monc->cur_mon);
1329	if (monc->cur_mon >= 0) {
1330		if (!monc->hunting) {
1331			dout("%s hunting for new mon\n", __func__);
1332			reopen_session(monc);
1333			__schedule_delayed(monc);
1334		} else {
1335			dout("%s already hunting\n", __func__);
1336		}
1337	}
1338	mutex_unlock(&monc->mutex);
1339}
1340
1341/*
1342 * We can ignore refcounting on the connection struct, as all references
1343 * will come from the messenger workqueue, which is drained prior to
1344 * mon_client destruction.
1345 */
1346static struct ceph_connection *con_get(struct ceph_connection *con)
1347{
1348	return con;
1349}
1350
1351static void con_put(struct ceph_connection *con)
1352{
 
 
 
 
 
 
 
 
 
1353}
1354
1355static const struct ceph_connection_operations mon_con_ops = {
1356	.get = con_get,
1357	.put = con_put,
1358	.dispatch = dispatch,
1359	.fault = mon_fault,
1360	.alloc_msg = mon_alloc_msg,
1361};
v3.5.6
 
   1#include <linux/ceph/ceph_debug.h>
   2
   3#include <linux/module.h>
   4#include <linux/types.h>
   5#include <linux/slab.h>
   6#include <linux/random.h>
   7#include <linux/sched.h>
   8
 
   9#include <linux/ceph/mon_client.h>
  10#include <linux/ceph/libceph.h>
  11#include <linux/ceph/debugfs.h>
  12#include <linux/ceph/decode.h>
  13#include <linux/ceph/auth.h>
  14
  15/*
  16 * Interact with Ceph monitor cluster.  Handle requests for new map
  17 * versions, and periodically resend as needed.  Also implement
  18 * statfs() and umount().
  19 *
  20 * A small cluster of Ceph "monitors" are responsible for managing critical
  21 * cluster configuration and state information.  An odd number (e.g., 3, 5)
  22 * of cmon daemons use a modified version of the Paxos part-time parliament
  23 * algorithm to manage the MDS map (mds cluster membership), OSD map, and
  24 * list of clients who have mounted the file system.
  25 *
  26 * We maintain an open, active session with a monitor at all times in order to
  27 * receive timely MDSMap updates.  We periodically send a keepalive byte on the
  28 * TCP socket to ensure we detect a failure.  If the connection does break, we
  29 * randomly hunt for a new monitor.  Once the connection is reestablished, we
  30 * resend any outstanding requests.
  31 */
  32
  33static const struct ceph_connection_operations mon_con_ops;
  34
  35static int __validate_auth(struct ceph_mon_client *monc);
  36
  37/*
  38 * Decode a monmap blob (e.g., during mount).
  39 */
  40struct ceph_monmap *ceph_monmap_decode(void *p, void *end)
  41{
  42	struct ceph_monmap *m = NULL;
  43	int i, err = -EINVAL;
  44	struct ceph_fsid fsid;
  45	u32 epoch, num_mon;
  46	u16 version;
  47	u32 len;
  48
  49	ceph_decode_32_safe(&p, end, len, bad);
  50	ceph_decode_need(&p, end, len, bad);
  51
  52	dout("monmap_decode %p %p len %d\n", p, end, (int)(end-p));
  53
  54	ceph_decode_16_safe(&p, end, version, bad);
  55
  56	ceph_decode_need(&p, end, sizeof(fsid) + 2*sizeof(u32), bad);
  57	ceph_decode_copy(&p, &fsid, sizeof(fsid));
  58	epoch = ceph_decode_32(&p);
  59
  60	num_mon = ceph_decode_32(&p);
  61	ceph_decode_need(&p, end, num_mon*sizeof(m->mon_inst[0]), bad);
  62
  63	if (num_mon >= CEPH_MAX_MON)
  64		goto bad;
  65	m = kmalloc(sizeof(*m) + sizeof(m->mon_inst[0])*num_mon, GFP_NOFS);
  66	if (m == NULL)
  67		return ERR_PTR(-ENOMEM);
  68	m->fsid = fsid;
  69	m->epoch = epoch;
  70	m->num_mon = num_mon;
  71	ceph_decode_copy(&p, m->mon_inst, num_mon*sizeof(m->mon_inst[0]));
  72	for (i = 0; i < num_mon; i++)
  73		ceph_decode_addr(&m->mon_inst[i].addr);
  74
  75	dout("monmap_decode epoch %d, num_mon %d\n", m->epoch,
  76	     m->num_mon);
  77	for (i = 0; i < m->num_mon; i++)
  78		dout("monmap_decode  mon%d is %s\n", i,
  79		     ceph_pr_addr(&m->mon_inst[i].addr.in_addr));
  80	return m;
  81
  82bad:
  83	dout("monmap_decode failed with %d\n", err);
  84	kfree(m);
  85	return ERR_PTR(err);
  86}
  87
  88/*
  89 * return true if *addr is included in the monmap.
  90 */
  91int ceph_monmap_contains(struct ceph_monmap *m, struct ceph_entity_addr *addr)
  92{
  93	int i;
  94
  95	for (i = 0; i < m->num_mon; i++)
  96		if (memcmp(addr, &m->mon_inst[i].addr, sizeof(*addr)) == 0)
  97			return 1;
  98	return 0;
  99}
 100
 101/*
 102 * Send an auth request.
 103 */
 104static void __send_prepared_auth_request(struct ceph_mon_client *monc, int len)
 105{
 106	monc->pending_auth = 1;
 107	monc->m_auth->front.iov_len = len;
 108	monc->m_auth->hdr.front_len = cpu_to_le32(len);
 109	ceph_con_revoke(monc->con, monc->m_auth);
 110	ceph_msg_get(monc->m_auth);  /* keep our ref */
 111	ceph_con_send(monc->con, monc->m_auth);
 112}
 113
 114/*
 115 * Close monitor session, if any.
 116 */
 117static void __close_session(struct ceph_mon_client *monc)
 118{
 119	dout("__close_session closing mon%d\n", monc->cur_mon);
 120	ceph_con_revoke(monc->con, monc->m_auth);
 121	ceph_con_close(monc->con);
 122	monc->cur_mon = -1;
 
 
 
 123	monc->pending_auth = 0;
 124	ceph_auth_reset(monc->auth);
 125}
 126
 127/*
 128 * Open a session with a (new) monitor.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 129 */
 130static int __open_session(struct ceph_mon_client *monc)
 131{
 132	char r;
 133	int ret;
 134
 135	if (monc->cur_mon < 0) {
 136		get_random_bytes(&r, 1);
 137		monc->cur_mon = r % monc->monmap->num_mon;
 138		dout("open_session num=%d r=%d -> mon%d\n",
 139		     monc->monmap->num_mon, r, monc->cur_mon);
 140		monc->sub_sent = 0;
 141		monc->sub_renew_after = jiffies;  /* i.e., expired */
 142		monc->want_next_osdmap = !!monc->want_next_osdmap;
 143
 144		dout("open_session mon%d opening\n", monc->cur_mon);
 145		monc->con->peer_name.type = CEPH_ENTITY_TYPE_MON;
 146		monc->con->peer_name.num = cpu_to_le64(monc->cur_mon);
 147		ceph_con_open(monc->con,
 148			      &monc->monmap->mon_inst[monc->cur_mon].addr);
 149
 150		/* initiatiate authentication handshake */
 151		ret = ceph_auth_build_hello(monc->auth,
 152					    monc->m_auth->front.iov_base,
 153					    monc->m_auth->front_max);
 154		__send_prepared_auth_request(monc, ret);
 155	} else {
 156		dout("open_session mon%d already open\n", monc->cur_mon);
 157	}
 158	return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 159}
 160
 161static bool __sub_expired(struct ceph_mon_client *monc)
 162{
 163	return time_after_eq(jiffies, monc->sub_renew_after);
 
 
 
 164}
 165
 166/*
 167 * Reschedule delayed work timer.
 168 */
 169static void __schedule_delayed(struct ceph_mon_client *monc)
 170{
 171	unsigned int delay;
 172
 173	if (monc->cur_mon < 0 || __sub_expired(monc))
 174		delay = 10 * HZ;
 175	else
 176		delay = 20 * HZ;
 177	dout("__schedule_delayed after %u\n", delay);
 178	schedule_delayed_work(&monc->delayed_work, delay);
 
 
 179}
 180
 
 
 
 
 
 
 
 181/*
 182 * Send subscribe request for mdsmap and/or osdmap.
 
 183 */
 184static void __send_subscribe(struct ceph_mon_client *monc)
 185{
 186	dout("__send_subscribe sub_sent=%u exp=%u want_osd=%d\n",
 187	     (unsigned int)monc->sub_sent, __sub_expired(monc),
 188	     monc->want_next_osdmap);
 189	if ((__sub_expired(monc) && !monc->sub_sent) ||
 190	    monc->want_next_osdmap == 1) {
 191		struct ceph_msg *msg = monc->m_subscribe;
 192		struct ceph_mon_subscribe_item *i;
 193		void *p, *end;
 194		int num;
 195
 196		p = msg->front.iov_base;
 197		end = p + msg->front_max;
 198
 199		num = 1 + !!monc->want_next_osdmap + !!monc->want_mdsmap;
 200		ceph_encode_32(&p, num);
 201
 202		if (monc->want_next_osdmap) {
 203			dout("__send_subscribe to 'osdmap' %u\n",
 204			     (unsigned int)monc->have_osdmap);
 205			ceph_encode_string(&p, end, "osdmap", 6);
 206			i = p;
 207			i->have = cpu_to_le64(monc->have_osdmap);
 208			i->onetime = 1;
 209			p += sizeof(*i);
 210			monc->want_next_osdmap = 2;  /* requested */
 211		}
 212		if (monc->want_mdsmap) {
 213			dout("__send_subscribe to 'mdsmap' %u+\n",
 214			     (unsigned int)monc->have_mdsmap);
 215			ceph_encode_string(&p, end, "mdsmap", 6);
 216			i = p;
 217			i->have = cpu_to_le64(monc->have_mdsmap);
 218			i->onetime = 0;
 219			p += sizeof(*i);
 220		}
 221		ceph_encode_string(&p, end, "monmap", 6);
 222		i = p;
 223		i->have = 0;
 224		i->onetime = 0;
 225		p += sizeof(*i);
 226
 227		msg->front.iov_len = p - msg->front.iov_base;
 228		msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
 229		ceph_con_revoke(monc->con, msg);
 230		ceph_con_send(monc->con, ceph_msg_get(msg));
 231
 232		monc->sub_sent = jiffies | 1;  /* never 0 */
 233	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 234}
 235
 236static void handle_subscribe_ack(struct ceph_mon_client *monc,
 237				 struct ceph_msg *msg)
 238{
 239	unsigned int seconds;
 240	struct ceph_mon_subscribe_ack *h = msg->front.iov_base;
 241
 242	if (msg->front.iov_len < sizeof(*h))
 243		goto bad;
 244	seconds = le32_to_cpu(h->duration);
 245
 246	mutex_lock(&monc->mutex);
 247	if (monc->hunting) {
 248		pr_info("mon%d %s session established\n",
 249			monc->cur_mon,
 250			ceph_pr_addr(&monc->con->peer_addr.in_addr));
 251		monc->hunting = false;
 
 
 
 
 
 
 
 
 252	}
 253	dout("handle_subscribe_ack after %d seconds\n", seconds);
 254	monc->sub_renew_after = monc->sub_sent + (seconds >> 1)*HZ - 1;
 255	monc->sub_sent = 0;
 256	mutex_unlock(&monc->mutex);
 257	return;
 258bad:
 259	pr_err("got corrupt subscribe-ack msg\n");
 260	ceph_msg_dump(msg);
 261}
 262
 263/*
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 264 * Keep track of which maps we have
 
 
 265 */
 266int ceph_monc_got_mdsmap(struct ceph_mon_client *monc, u32 got)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 267{
 268	mutex_lock(&monc->mutex);
 269	monc->have_mdsmap = got;
 270	mutex_unlock(&monc->mutex);
 271	return 0;
 272}
 273EXPORT_SYMBOL(ceph_monc_got_mdsmap);
 274
 275int ceph_monc_got_osdmap(struct ceph_mon_client *monc, u32 got)
 276{
 277	mutex_lock(&monc->mutex);
 278	monc->have_osdmap = got;
 279	monc->want_next_osdmap = 0;
 280	mutex_unlock(&monc->mutex);
 281	return 0;
 282}
 
 283
 284/*
 285 * Register interest in the next osdmap
 
 
 
 286 */
 287void ceph_monc_request_next_osdmap(struct ceph_mon_client *monc)
 
 288{
 289	dout("request_next_osdmap have %u\n", monc->have_osdmap);
 
 
 290	mutex_lock(&monc->mutex);
 291	if (!monc->want_next_osdmap)
 292		monc->want_next_osdmap = 1;
 293	if (monc->want_next_osdmap < 2)
 294		__send_subscribe(monc);
 
 
 
 
 
 
 
 
 
 
 
 295	mutex_unlock(&monc->mutex);
 
 296}
 
 297
 298/*
 299 *
 
 300 */
 301int ceph_monc_open_session(struct ceph_mon_client *monc)
 302{
 303	mutex_lock(&monc->mutex);
 
 
 304	__open_session(monc);
 305	__schedule_delayed(monc);
 306	mutex_unlock(&monc->mutex);
 307	return 0;
 308}
 309EXPORT_SYMBOL(ceph_monc_open_session);
 310
 311/*
 312 * The monitor responds with mount ack indicate mount success.  The
 313 * included client ticket allows the client to talk to MDSs and OSDs.
 314 */
 315static void ceph_monc_handle_map(struct ceph_mon_client *monc,
 316				 struct ceph_msg *msg)
 317{
 318	struct ceph_client *client = monc->client;
 319	struct ceph_monmap *monmap = NULL, *old = monc->monmap;
 320	void *p, *end;
 321
 322	mutex_lock(&monc->mutex);
 323
 324	dout("handle_monmap\n");
 325	p = msg->front.iov_base;
 326	end = p + msg->front.iov_len;
 327
 328	monmap = ceph_monmap_decode(p, end);
 329	if (IS_ERR(monmap)) {
 330		pr_err("problem decoding monmap, %d\n",
 331		       (int)PTR_ERR(monmap));
 332		goto out;
 333	}
 334
 335	if (ceph_check_fsid(monc->client, &monmap->fsid) < 0) {
 336		kfree(monmap);
 337		goto out;
 338	}
 339
 340	client->monc.monmap = monmap;
 341	kfree(old);
 342
 343	if (!client->have_fsid) {
 344		client->have_fsid = true;
 345		mutex_unlock(&monc->mutex);
 346		/*
 347		 * do debugfs initialization without mutex to avoid
 348		 * creating a locking dependency
 349		 */
 350		ceph_debugfs_client_init(client);
 351		goto out_unlocked;
 352	}
 353out:
 354	mutex_unlock(&monc->mutex);
 355out_unlocked:
 356	wake_up_all(&client->auth_wq);
 357}
 358
 359/*
 360 * generic requests (e.g., statfs, poolop)
 361 */
 362static struct ceph_mon_generic_request *__lookup_generic_req(
 363	struct ceph_mon_client *monc, u64 tid)
 364{
 365	struct ceph_mon_generic_request *req;
 366	struct rb_node *n = monc->generic_request_tree.rb_node;
 367
 368	while (n) {
 369		req = rb_entry(n, struct ceph_mon_generic_request, node);
 370		if (tid < req->tid)
 371			n = n->rb_left;
 372		else if (tid > req->tid)
 373			n = n->rb_right;
 374		else
 375			return req;
 376	}
 377	return NULL;
 378}
 379
 380static void __insert_generic_request(struct ceph_mon_client *monc,
 381			    struct ceph_mon_generic_request *new)
 382{
 383	struct rb_node **p = &monc->generic_request_tree.rb_node;
 384	struct rb_node *parent = NULL;
 385	struct ceph_mon_generic_request *req = NULL;
 386
 387	while (*p) {
 388		parent = *p;
 389		req = rb_entry(parent, struct ceph_mon_generic_request, node);
 390		if (new->tid < req->tid)
 391			p = &(*p)->rb_left;
 392		else if (new->tid > req->tid)
 393			p = &(*p)->rb_right;
 394		else
 395			BUG();
 396	}
 397
 398	rb_link_node(&new->node, parent, p);
 399	rb_insert_color(&new->node, &monc->generic_request_tree);
 400}
 401
 402static void release_generic_request(struct kref *kref)
 403{
 404	struct ceph_mon_generic_request *req =
 405		container_of(kref, struct ceph_mon_generic_request, kref);
 406
 
 
 
 
 407	if (req->reply)
 408		ceph_msg_put(req->reply);
 409	if (req->request)
 410		ceph_msg_put(req->request);
 411
 412	kfree(req);
 413}
 414
 415static void put_generic_request(struct ceph_mon_generic_request *req)
 416{
 417	kref_put(&req->kref, release_generic_request);
 
 418}
 419
 420static void get_generic_request(struct ceph_mon_generic_request *req)
 421{
 422	kref_get(&req->kref);
 423}
 424
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 425static struct ceph_msg *get_generic_reply(struct ceph_connection *con,
 426					 struct ceph_msg_header *hdr,
 427					 int *skip)
 428{
 429	struct ceph_mon_client *monc = con->private;
 430	struct ceph_mon_generic_request *req;
 431	u64 tid = le64_to_cpu(hdr->tid);
 432	struct ceph_msg *m;
 433
 434	mutex_lock(&monc->mutex);
 435	req = __lookup_generic_req(monc, tid);
 436	if (!req) {
 437		dout("get_generic_reply %lld dne\n", tid);
 438		*skip = 1;
 439		m = NULL;
 440	} else {
 441		dout("get_generic_reply %lld got %p\n", tid, req->reply);
 
 442		m = ceph_msg_get(req->reply);
 443		/*
 444		 * we don't need to track the connection reading into
 445		 * this reply because we only have one open connection
 446		 * at a time, ever.
 447		 */
 448	}
 449	mutex_unlock(&monc->mutex);
 450	return m;
 451}
 452
 453static int do_generic_request(struct ceph_mon_client *monc,
 454			      struct ceph_mon_generic_request *req)
 455{
 456	int err;
 457
 458	/* register request */
 459	mutex_lock(&monc->mutex);
 460	req->tid = ++monc->last_tid;
 461	req->request->hdr.tid = cpu_to_le64(req->tid);
 462	__insert_generic_request(monc, req);
 463	monc->num_generic_requests++;
 464	ceph_con_send(monc->con, ceph_msg_get(req->request));
 465	mutex_unlock(&monc->mutex);
 466
 467	err = wait_for_completion_interruptible(&req->completion);
 468
 469	mutex_lock(&monc->mutex);
 470	rb_erase(&req->node, &monc->generic_request_tree);
 471	monc->num_generic_requests--;
 472	mutex_unlock(&monc->mutex);
 473
 474	if (!err)
 475		err = req->result;
 476	return err;
 477}
 478
 479/*
 480 * statfs
 481 */
 482static void handle_statfs_reply(struct ceph_mon_client *monc,
 483				struct ceph_msg *msg)
 484{
 485	struct ceph_mon_generic_request *req;
 486	struct ceph_mon_statfs_reply *reply = msg->front.iov_base;
 487	u64 tid = le64_to_cpu(msg->hdr.tid);
 488
 
 
 489	if (msg->front.iov_len != sizeof(*reply))
 490		goto bad;
 491	dout("handle_statfs_reply %p tid %llu\n", msg, tid);
 492
 493	mutex_lock(&monc->mutex);
 494	req = __lookup_generic_req(monc, tid);
 495	if (req) {
 496		*(struct ceph_statfs *)req->buf = reply->st;
 497		req->result = 0;
 498		get_generic_request(req);
 499	}
 
 
 
 
 500	mutex_unlock(&monc->mutex);
 501	if (req) {
 502		complete_all(&req->completion);
 503		put_generic_request(req);
 504	}
 505	return;
 506
 507bad:
 508	pr_err("corrupt generic reply, tid %llu\n", tid);
 509	ceph_msg_dump(msg);
 510}
 511
 512/*
 513 * Do a synchronous statfs().
 514 */
 515int ceph_monc_do_statfs(struct ceph_mon_client *monc, struct ceph_statfs *buf)
 
 516{
 517	struct ceph_mon_generic_request *req;
 518	struct ceph_mon_statfs *h;
 519	int err;
 520
 521	req = kzalloc(sizeof(*req), GFP_NOFS);
 522	if (!req)
 523		return -ENOMEM;
 524
 525	kref_init(&req->kref);
 526	req->buf = buf;
 527	req->buf_len = sizeof(*buf);
 528	init_completion(&req->completion);
 529
 530	err = -ENOMEM;
 531	req->request = ceph_msg_new(CEPH_MSG_STATFS, sizeof(*h), GFP_NOFS,
 532				    true);
 533	if (!req->request)
 534		goto out;
 535	req->reply = ceph_msg_new(CEPH_MSG_STATFS_REPLY, 1024, GFP_NOFS,
 536				  true);
 537	if (!req->reply)
 538		goto out;
 539
 
 
 
 
 
 540	/* fill out request */
 541	h = req->request->front.iov_base;
 542	h->monhdr.have_version = 0;
 543	h->monhdr.session_mon = cpu_to_le16(-1);
 544	h->monhdr.session_mon_tid = 0;
 545	h->fsid = monc->monmap->fsid;
 
 
 
 
 546
 547	err = do_generic_request(monc, req);
 548
 549out:
 550	kref_put(&req->kref, release_generic_request);
 551	return err;
 552}
 553EXPORT_SYMBOL(ceph_monc_do_statfs);
 554
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 555/*
 556 * pool ops
 
 
 557 */
 558static int get_poolop_reply_buf(const char *src, size_t src_len,
 559				char *dst, size_t dst_len)
 560{
 561	u32 buf_len;
 
 562
 563	if (src_len != sizeof(u32) + dst_len)
 564		return -EINVAL;
 
 
 
 
 
 
 
 
 
 
 565
 566	buf_len = le32_to_cpu(*(u32 *)src);
 567	if (buf_len != dst_len)
 568		return -EINVAL;
 
 
 
 
 
 
 569
 570	memcpy(dst, src + sizeof(u32), dst_len);
 
 
 
 
 571	return 0;
 572}
 
 573
 574static void handle_poolop_reply(struct ceph_mon_client *monc,
 575				struct ceph_msg *msg)
 576{
 577	struct ceph_mon_generic_request *req;
 578	struct ceph_mon_poolop_reply *reply = msg->front.iov_base;
 
 579	u64 tid = le64_to_cpu(msg->hdr.tid);
 580
 581	if (msg->front.iov_len < sizeof(*reply))
 582		goto bad;
 583	dout("handle_poolop_reply %p tid %llu\n", msg, tid);
 
 
 584
 585	mutex_lock(&monc->mutex);
 586	req = __lookup_generic_req(monc, tid);
 587	if (req) {
 588		if (req->buf_len &&
 589		    get_poolop_reply_buf(msg->front.iov_base + sizeof(*reply),
 590				     msg->front.iov_len - sizeof(*reply),
 591				     req->buf, req->buf_len) < 0) {
 592			mutex_unlock(&monc->mutex);
 593			goto bad;
 594		}
 595		req->result = le32_to_cpu(reply->reply_code);
 596		get_generic_request(req);
 597	}
 
 
 
 598	mutex_unlock(&monc->mutex);
 599	if (req) {
 600		complete(&req->completion);
 601		put_generic_request(req);
 602	}
 603	return;
 604
 605bad:
 606	pr_err("corrupt generic reply, tid %llu\n", tid);
 607	ceph_msg_dump(msg);
 608}
 609
 610/*
 611 * Do a synchronous pool op.
 612 */
 613int ceph_monc_do_poolop(struct ceph_mon_client *monc, u32 op,
 614			u32 pool, u64 snapid,
 615			char *buf, int len)
 616{
 617	struct ceph_mon_generic_request *req;
 618	struct ceph_mon_poolop *h;
 619	int err;
 
 620
 621	req = kzalloc(sizeof(*req), GFP_NOFS);
 622	if (!req)
 623		return -ENOMEM;
 624
 625	kref_init(&req->kref);
 626	req->buf = buf;
 627	req->buf_len = len;
 628	init_completion(&req->completion);
 629
 630	err = -ENOMEM;
 631	req->request = ceph_msg_new(CEPH_MSG_POOLOP, sizeof(*h), GFP_NOFS,
 632				    true);
 633	if (!req->request)
 634		goto out;
 635	req->reply = ceph_msg_new(CEPH_MSG_POOLOP_REPLY, 1024, GFP_NOFS,
 
 636				  true);
 637	if (!req->reply)
 638		goto out;
 639
 640	/* fill out request */
 641	req->request->hdr.version = cpu_to_le16(2);
 642	h = req->request->front.iov_base;
 643	h->monhdr.have_version = 0;
 644	h->monhdr.session_mon = cpu_to_le16(-1);
 645	h->monhdr.session_mon_tid = 0;
 646	h->fsid = monc->monmap->fsid;
 647	h->pool = cpu_to_le32(pool);
 648	h->op = cpu_to_le32(op);
 649	h->auid = 0;
 650	h->snapid = cpu_to_le64(snapid);
 651	h->name_len = 0;
 652
 653	err = do_generic_request(monc, req);
 
 654
 
 655out:
 656	kref_put(&req->kref, release_generic_request);
 657	return err;
 658}
 659
 660int ceph_monc_create_snapid(struct ceph_mon_client *monc,
 661			    u32 pool, u64 *snapid)
 662{
 663	return ceph_monc_do_poolop(monc,  POOL_OP_CREATE_UNMANAGED_SNAP,
 664				   pool, 0, (char *)snapid, sizeof(*snapid));
 665
 666}
 667EXPORT_SYMBOL(ceph_monc_create_snapid);
 668
 669int ceph_monc_delete_snapid(struct ceph_mon_client *monc,
 670			    u32 pool, u64 snapid)
 671{
 672	return ceph_monc_do_poolop(monc,  POOL_OP_CREATE_UNMANAGED_SNAP,
 673				   pool, snapid, 0, 0);
 674
 675}
 
 676
 677/*
 678 * Resend pending generic requests.
 679 */
 680static void __resend_generic_request(struct ceph_mon_client *monc)
 681{
 682	struct ceph_mon_generic_request *req;
 683	struct rb_node *p;
 684
 685	for (p = rb_first(&monc->generic_request_tree); p; p = rb_next(p)) {
 686		req = rb_entry(p, struct ceph_mon_generic_request, node);
 687		ceph_con_revoke(monc->con, req->request);
 688		ceph_con_send(monc->con, ceph_msg_get(req->request));
 
 689	}
 690}
 691
 692/*
 693 * Delayed work.  If we haven't mounted yet, retry.  Otherwise,
 694 * renew/retry subscription as needed (in case it is timing out, or we
 695 * got an ENOMEM).  And keep the monitor connection alive.
 696 */
 697static void delayed_work(struct work_struct *work)
 698{
 699	struct ceph_mon_client *monc =
 700		container_of(work, struct ceph_mon_client, delayed_work.work);
 701
 702	dout("monc delayed_work\n");
 703	mutex_lock(&monc->mutex);
 704	if (monc->hunting) {
 705		__close_session(monc);
 706		__open_session(monc);  /* continue hunting */
 707	} else {
 708		ceph_con_keepalive(monc->con);
 
 
 
 
 
 
 709
 710		__validate_auth(monc);
 
 
 
 
 711
 712		if (monc->auth->ops->is_authenticated(monc->auth))
 713			__send_subscribe(monc);
 
 
 
 
 
 
 
 714	}
 715	__schedule_delayed(monc);
 716	mutex_unlock(&monc->mutex);
 717}
 718
 719/*
 720 * On startup, we build a temporary monmap populated with the IPs
 721 * provided by mount(2).
 722 */
 723static int build_initial_monmap(struct ceph_mon_client *monc)
 724{
 725	struct ceph_options *opt = monc->client->options;
 726	struct ceph_entity_addr *mon_addr = opt->mon_addr;
 727	int num_mon = opt->num_mon;
 728	int i;
 729
 730	/* build initial monmap */
 731	monc->monmap = kzalloc(sizeof(*monc->monmap) +
 732			       num_mon*sizeof(monc->monmap->mon_inst[0]),
 733			       GFP_KERNEL);
 734	if (!monc->monmap)
 735		return -ENOMEM;
 736	for (i = 0; i < num_mon; i++) {
 737		monc->monmap->mon_inst[i].addr = mon_addr[i];
 738		monc->monmap->mon_inst[i].addr.nonce = 0;
 739		monc->monmap->mon_inst[i].name.type =
 740			CEPH_ENTITY_TYPE_MON;
 741		monc->monmap->mon_inst[i].name.num = cpu_to_le64(i);
 742	}
 743	monc->monmap->num_mon = num_mon;
 744	monc->have_fsid = false;
 745	return 0;
 746}
 747
 748int ceph_monc_init(struct ceph_mon_client *monc, struct ceph_client *cl)
 749{
 750	int err = 0;
 751
 752	dout("init\n");
 753	memset(monc, 0, sizeof(*monc));
 754	monc->client = cl;
 755	monc->monmap = NULL;
 756	mutex_init(&monc->mutex);
 757
 758	err = build_initial_monmap(monc);
 759	if (err)
 760		goto out;
 761
 762	/* connection */
 763	monc->con = kmalloc(sizeof(*monc->con), GFP_KERNEL);
 764	if (!monc->con)
 765		goto out_monmap;
 766	ceph_con_init(monc->client->msgr, monc->con);
 767	monc->con->private = monc;
 768	monc->con->ops = &mon_con_ops;
 769
 770	/* authentication */
 771	monc->auth = ceph_auth_init(cl->options->name,
 772				    cl->options->key);
 773	if (IS_ERR(monc->auth)) {
 774		err = PTR_ERR(monc->auth);
 775		goto out_con;
 776	}
 777	monc->auth->want_keys =
 778		CEPH_ENTITY_TYPE_AUTH | CEPH_ENTITY_TYPE_MON |
 779		CEPH_ENTITY_TYPE_OSD | CEPH_ENTITY_TYPE_MDS;
 780
 781	/* msgs */
 782	err = -ENOMEM;
 783	monc->m_subscribe_ack = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE_ACK,
 784				     sizeof(struct ceph_mon_subscribe_ack),
 785				     GFP_NOFS, true);
 786	if (!monc->m_subscribe_ack)
 787		goto out_auth;
 788
 789	monc->m_subscribe = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE, 96, GFP_NOFS,
 790					 true);
 791	if (!monc->m_subscribe)
 792		goto out_subscribe_ack;
 793
 794	monc->m_auth_reply = ceph_msg_new(CEPH_MSG_AUTH_REPLY, 4096, GFP_NOFS,
 795					  true);
 796	if (!monc->m_auth_reply)
 797		goto out_subscribe;
 798
 799	monc->m_auth = ceph_msg_new(CEPH_MSG_AUTH, 4096, GFP_NOFS, true);
 800	monc->pending_auth = 0;
 801	if (!monc->m_auth)
 802		goto out_auth_reply;
 803
 
 
 
 804	monc->cur_mon = -1;
 805	monc->hunting = true;
 806	monc->sub_renew_after = jiffies;
 807	monc->sub_sent = 0;
 808
 809	INIT_DELAYED_WORK(&monc->delayed_work, delayed_work);
 810	monc->generic_request_tree = RB_ROOT;
 811	monc->num_generic_requests = 0;
 812	monc->last_tid = 0;
 813
 814	monc->have_mdsmap = 0;
 815	monc->have_osdmap = 0;
 816	monc->want_next_osdmap = 1;
 817	return 0;
 818
 819out_auth_reply:
 820	ceph_msg_put(monc->m_auth_reply);
 821out_subscribe:
 822	ceph_msg_put(monc->m_subscribe);
 823out_subscribe_ack:
 824	ceph_msg_put(monc->m_subscribe_ack);
 825out_auth:
 826	ceph_auth_destroy(monc->auth);
 827out_con:
 828	monc->con->ops->put(monc->con);
 829out_monmap:
 830	kfree(monc->monmap);
 831out:
 832	return err;
 833}
 834EXPORT_SYMBOL(ceph_monc_init);
 835
 836void ceph_monc_stop(struct ceph_mon_client *monc)
 837{
 838	dout("stop\n");
 839	cancel_delayed_work_sync(&monc->delayed_work);
 840
 841	mutex_lock(&monc->mutex);
 842	__close_session(monc);
 843
 844	monc->con->private = NULL;
 845	monc->con->ops->put(monc->con);
 846	monc->con = NULL;
 847
 848	mutex_unlock(&monc->mutex);
 849
 850	/*
 851	 * flush msgr queue before we destroy ourselves to ensure that:
 852	 *  - any work that references our embedded con is finished.
 853	 *  - any osd_client or other work that may reference an authorizer
 854	 *    finishes before we shut down the auth subsystem.
 855	 */
 856	ceph_msgr_flush();
 857
 858	ceph_auth_destroy(monc->auth);
 859
 
 
 860	ceph_msg_put(monc->m_auth);
 861	ceph_msg_put(monc->m_auth_reply);
 862	ceph_msg_put(monc->m_subscribe);
 863	ceph_msg_put(monc->m_subscribe_ack);
 864
 865	kfree(monc->monmap);
 866}
 867EXPORT_SYMBOL(ceph_monc_stop);
 868
 
 
 
 
 
 
 
 
 
 
 
 869static void handle_auth_reply(struct ceph_mon_client *monc,
 870			      struct ceph_msg *msg)
 871{
 872	int ret;
 873	int was_auth = 0;
 874
 875	mutex_lock(&monc->mutex);
 876	if (monc->auth->ops)
 877		was_auth = monc->auth->ops->is_authenticated(monc->auth);
 878	monc->pending_auth = 0;
 879	ret = ceph_handle_auth_reply(monc->auth, msg->front.iov_base,
 880				     msg->front.iov_len,
 881				     monc->m_auth->front.iov_base,
 882				     monc->m_auth->front_max);
 
 
 
 
 
 
 
 883	if (ret < 0) {
 884		monc->client->auth_err = ret;
 885		wake_up_all(&monc->client->auth_wq);
 886	} else if (ret > 0) {
 887		__send_prepared_auth_request(monc, ret);
 888	} else if (!was_auth && monc->auth->ops->is_authenticated(monc->auth)) {
 889		dout("authenticated, starting session\n");
 890
 891		monc->client->msgr->inst.name.type = CEPH_ENTITY_TYPE_CLIENT;
 892		monc->client->msgr->inst.name.num =
 893					cpu_to_le64(monc->auth->global_id);
 894
 895		__send_subscribe(monc);
 896		__resend_generic_request(monc);
 
 
 
 897	}
 
 
 898	mutex_unlock(&monc->mutex);
 
 
 899}
 900
 901static int __validate_auth(struct ceph_mon_client *monc)
 902{
 903	int ret;
 904
 905	if (monc->pending_auth)
 906		return 0;
 907
 908	ret = ceph_build_auth(monc->auth, monc->m_auth->front.iov_base,
 909			      monc->m_auth->front_max);
 910	if (ret <= 0)
 911		return ret; /* either an error, or no need to authenticate */
 912	__send_prepared_auth_request(monc, ret);
 913	return 0;
 914}
 915
 916int ceph_monc_validate_auth(struct ceph_mon_client *monc)
 917{
 918	int ret;
 919
 920	mutex_lock(&monc->mutex);
 921	ret = __validate_auth(monc);
 922	mutex_unlock(&monc->mutex);
 923	return ret;
 924}
 925EXPORT_SYMBOL(ceph_monc_validate_auth);
 926
 927/*
 928 * handle incoming message
 929 */
 930static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
 931{
 932	struct ceph_mon_client *monc = con->private;
 933	int type = le16_to_cpu(msg->hdr.type);
 934
 935	if (!monc)
 936		return;
 937
 938	switch (type) {
 939	case CEPH_MSG_AUTH_REPLY:
 940		handle_auth_reply(monc, msg);
 941		break;
 942
 943	case CEPH_MSG_MON_SUBSCRIBE_ACK:
 944		handle_subscribe_ack(monc, msg);
 945		break;
 946
 947	case CEPH_MSG_STATFS_REPLY:
 948		handle_statfs_reply(monc, msg);
 949		break;
 950
 951	case CEPH_MSG_POOLOP_REPLY:
 952		handle_poolop_reply(monc, msg);
 
 
 
 
 953		break;
 954
 955	case CEPH_MSG_MON_MAP:
 956		ceph_monc_handle_map(monc, msg);
 957		break;
 958
 959	case CEPH_MSG_OSD_MAP:
 960		ceph_osdc_handle_map(&monc->client->osdc, msg);
 961		break;
 962
 963	default:
 964		/* can the chained handler handle it? */
 965		if (monc->client->extra_mon_dispatch &&
 966		    monc->client->extra_mon_dispatch(monc->client, msg) == 0)
 967			break;
 968			
 969		pr_err("received unknown message type %d %s\n", type,
 970		       ceph_msg_type_name(type));
 971	}
 972	ceph_msg_put(msg);
 973}
 974
 975/*
 976 * Allocate memory for incoming message
 977 */
 978static struct ceph_msg *mon_alloc_msg(struct ceph_connection *con,
 979				      struct ceph_msg_header *hdr,
 980				      int *skip)
 981{
 982	struct ceph_mon_client *monc = con->private;
 983	int type = le16_to_cpu(hdr->type);
 984	int front_len = le32_to_cpu(hdr->front_len);
 985	struct ceph_msg *m = NULL;
 986
 987	*skip = 0;
 988
 989	switch (type) {
 990	case CEPH_MSG_MON_SUBSCRIBE_ACK:
 991		m = ceph_msg_get(monc->m_subscribe_ack);
 992		break;
 993	case CEPH_MSG_POOLOP_REPLY:
 994	case CEPH_MSG_STATFS_REPLY:
 
 995		return get_generic_reply(con, hdr, skip);
 996	case CEPH_MSG_AUTH_REPLY:
 997		m = ceph_msg_get(monc->m_auth_reply);
 998		break;
 
 
 
 
 
 
 
 
 
 
 999	case CEPH_MSG_MON_MAP:
1000	case CEPH_MSG_MDS_MAP:
1001	case CEPH_MSG_OSD_MAP:
 
1002		m = ceph_msg_new(type, front_len, GFP_NOFS, false);
 
 
1003		break;
1004	}
1005
1006	if (!m) {
1007		pr_info("alloc_msg unknown type %d\n", type);
1008		*skip = 1;
 
 
 
 
 
 
 
1009	}
 
1010	return m;
1011}
1012
1013/*
1014 * If the monitor connection resets, pick a new monitor and resubmit
1015 * any pending requests.
1016 */
1017static void mon_fault(struct ceph_connection *con)
1018{
1019	struct ceph_mon_client *monc = con->private;
1020
1021	if (!monc)
1022		return;
1023
1024	dout("mon_fault\n");
1025	mutex_lock(&monc->mutex);
1026	if (!con->private)
1027		goto out;
 
 
 
 
 
 
 
 
 
 
1028
1029	if (!monc->hunting)
1030		pr_info("mon%d %s session lost, "
1031			"hunting for new mon\n", monc->cur_mon,
1032			ceph_pr_addr(&monc->con->peer_addr.in_addr));
 
 
 
 
 
1033
1034	__close_session(monc);
1035	if (!monc->hunting) {
1036		/* start hunting */
1037		monc->hunting = true;
1038		__open_session(monc);
1039	} else {
1040		/* already hunting, let's wait a bit */
1041		__schedule_delayed(monc);
1042	}
1043out:
1044	mutex_unlock(&monc->mutex);
1045}
1046
1047static const struct ceph_connection_operations mon_con_ops = {
1048	.get = ceph_con_get,
1049	.put = ceph_con_put,
1050	.dispatch = dispatch,
1051	.fault = mon_fault,
1052	.alloc_msg = mon_alloc_msg,
1053};