4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; if not, write to the
18 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
19 * Boston, MA 021110-1307, USA
24 * Copyright (c) 2013, 2016, Intel Corporation.
25 * Use is subject to license terms.
27 * lustre/osp/lwp_dev.c
29 * This file provides code related to the Light Weight Proxy (LWP) managing
30 * the connections established from OST to MDT, and MDT to MDT0.
32 * A LWP connection is used to send quota and FLD query requests. It's not
33 * recoverable, which means target server doesn't have an on-disk record in
34 * the last_rcvd file to remember the connection. Once LWP reconnect after
35 * server reboot, server will always regard it as a new connection.
37 * Author: <di.wang@intel.com>
38 * Author: <yawei.niu@intel.com>
40 #define DEBUG_SUBSYSTEM S_OST
42 #include <obd_class.h>
43 #include <uapi/linux/lustre_param.h>
44 #include <lustre_log.h>
45 #include <linux/kthread.h>
47 #include "osp_internal.h"
50 struct lu_device lpd_dev;
51 struct obd_device *lpd_obd; /* corresponding OBD device */
52 struct obd_uuid lpd_cluuid;/* UUID of LWP */
53 struct obd_export *lpd_exp; /* export of LWP */
54 struct ptlrpc_thread lpd_notify_thread; /* notify thread */
55 int lpd_connects; /* use count, 0 or 1 */
58 static inline struct lwp_device *lu2lwp_dev(struct lu_device *d)
60 return container_of0(d, struct lwp_device, lpd_dev);
63 static inline struct lu_device *lwp2lu_dev(struct lwp_device *d)
71 * \param[in] env environment passed by caller
72 * \param[in] lwp LWP device to be setup
73 * \param[in] nidstring remote target NID
75 * \retval 0 on success
76 * \retval negative number on error
78 static int lwp_setup(const struct lu_env *env, struct lwp_device *lwp,
81 struct lustre_cfg_bufs *bufs = NULL;
82 struct lustre_cfg *lcfg = NULL;
83 char *lwp_name = lwp->lpd_obd->obd_name;
84 char *server_uuid = NULL;
87 struct obd_import *imp;
88 int len = strlen(lwp_name) + 1;
92 thread_set_flags(&lwp->lpd_notify_thread, SVC_STOPPED);
93 init_waitqueue_head(&lwp->lpd_notify_thread.t_ctl_waitq);
99 OBD_ALLOC(server_uuid, len);
100 if (server_uuid == NULL)
101 GOTO(out, rc = -ENOMEM);
103 snprintf(server_uuid, len, "-%s-", LUSTRE_LWP_NAME);
104 ptr = cfs_strrstr(lwp_name, server_uuid);
106 CERROR("%s: failed to get server_uuid from lwp_name: rc = %d\n",
108 GOTO(out, rc = -EINVAL);
111 strncpy(server_uuid, lwp_name, ptr - lwp_name);
112 server_uuid[ptr - lwp_name] = '\0';
113 strlcat(server_uuid, "_UUID", len);
114 lustre_cfg_bufs_reset(bufs, lwp_name);
115 lustre_cfg_bufs_set_string(bufs, 1, server_uuid);
116 lustre_cfg_bufs_set_string(bufs, 2, nidstring);
117 lcfg = lustre_cfg_new(LCFG_SETUP, bufs);
119 GOTO(out, rc = -ENOMEM);
121 rc = client_obd_setup(lwp->lpd_obd, lcfg);
123 CERROR("%s: client obd setup error: rc = %d\n",
124 lwp->lpd_obd->obd_name, rc);
128 imp = lwp->lpd_obd->u.cli.cl_import;
129 rc = ptlrpc_init_import(imp);
133 ll_generate_random_uuid(uuid);
134 class_uuid_unparse(uuid, &lwp->lpd_cluuid);
138 if (server_uuid != NULL)
139 OBD_FREE(server_uuid, len);
141 lustre_cfg_free(lcfg);
143 client_obd_cleanup(lwp->lpd_obd);
149 * Disconnect the import from LWP.
151 * \param[in] d LWP device to be disconnected
153 * \retval 0 on success
154 * \retval negative number on error
156 static int lwp_disconnect(struct lwp_device *d)
158 struct obd_import *imp;
161 imp = d->lpd_obd->u.cli.cl_import;
164 * Mark import deactivated now, so we don't try to reconnect if any
165 * of the cleanup RPCs fails (e.g. ldlm cancel, etc). We don't
166 * fully deactivate the import because that would cause all requests
169 LASSERT(imp != NULL);
170 spin_lock(&imp->imp_lock);
171 imp->imp_deactive = 1;
172 spin_unlock(&imp->imp_lock);
174 ptlrpc_deactivate_import(imp);
177 * Some non-replayable imports (MDS's OSCs) are pinged, so just
178 * delete it regardless. (It's safe to delete an import that was
181 ptlrpc_pinger_del_import(imp);
182 rc = ptlrpc_disconnect_import(imp, 0);
184 CWARN("%s: can't disconnect: rc = %d\n",
185 d->lpd_obd->obd_name, rc);
187 ptlrpc_invalidate_import(imp);
193 * Implementation of lu_device_operations::ldo_process_config.
195 * Process a Lustre configuration request.
197 * \param[in] env environment passed by caller
198 * \param[in] dev device to be processed
199 * \param[in] lcfg lustre_cfg, LCFG_PRE_CLEANUP or LCFG_CLEANUP
201 * \retval 0 on success
202 * \retval negative number on error
204 static int lwp_process_config(const struct lu_env *env,
205 struct lu_device *dev, struct lustre_cfg *lcfg)
207 struct lwp_device *d = lu2lwp_dev(dev);
211 switch (lcfg->lcfg_command) {
212 case LCFG_PRE_CLEANUP:
214 rc = lwp_disconnect(d);
220 CERROR("%s: unknown command %u\n",
221 (char *)lustre_cfg_string(lcfg, 0), lcfg->lcfg_command);
229 static const struct lu_device_operations lwp_lu_ops = {
230 .ldo_process_config = lwp_process_config,
234 * Initialize LWP device.
236 * \param[in] env environment passed by caller
237 * \param[in] lwp device to be initialized
238 * \param[in] ldt not used
239 * \param[in] cfg lustre_cfg contains remote target uuid
241 * \retval 0 on success
242 * \retval -ENODEV if the device name cannot be found
243 * \retval negative numbers on other errors
245 static int lwp_init0(const struct lu_env *env, struct lwp_device *lwp,
246 struct lu_device_type *ldt, struct lustre_cfg *cfg)
251 lwp->lpd_obd = class_name2obd(lustre_cfg_string(cfg, 0));
252 if (lwp->lpd_obd == NULL) {
253 CERROR("Cannot find obd with name %s\n",
254 lustre_cfg_string(cfg, 0));
258 lwp->lpd_dev.ld_ops = &lwp_lu_ops;
259 lwp->lpd_obd->obd_lu_dev = &lwp->lpd_dev;
261 rc = ptlrpcd_addref();
263 CERROR("%s: ptlrpcd addref error: rc =%d\n",
264 lwp->lpd_obd->obd_name, rc);
268 rc = lwp_setup(env, lwp, lustre_cfg_string(cfg, 1));
270 CERROR("%s: setup lwp failed. %d\n",
271 lwp->lpd_obd->obd_name, rc);
276 if (lprocfs_obd_setup(lwp->lpd_obd) == 0) {
277 sptlrpc_lprocfs_cliobd_attach(lwp->lpd_obd);
278 ptlrpc_lprocfs_register_obd(lwp->lpd_obd);
285 * Implementation of lu_device_type_operations::ldto_device_free.
289 * \param[in] env environment passed by caller
290 * \param[in] lu device to be freed
292 * \retval NULL to indicate that this is the bottom device
293 * of the stack and there are no more devices
294 * below this one to be cleaned up.
296 static struct lu_device *lwp_device_free(const struct lu_env *env,
297 struct lu_device *lu)
299 struct lwp_device *m = lu2lwp_dev(lu);
302 if (atomic_read(&lu->ld_ref) && lu->ld_site) {
303 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
304 lu_site_print(env, lu->ld_site, &msgdata, lu_cdebug_printer);
306 lu_device_fini(&m->lpd_dev);
312 * Implementation of lu_device_type_operations::ldto_device_alloc.
314 * Allocate a LWP device.
316 * \param[in] env environment passed by caller
317 * \param[in] ldt device type whose name is LUSTRE_LWP_NAME
318 * \param[in] lcfg lustre_cfg contains remote target UUID
320 * \retval pointer of allocated LWP device on success
321 * \retval ERR_PTR(errno) on error
323 static struct lu_device *lwp_device_alloc(const struct lu_env *env,
324 struct lu_device_type *ldt,
325 struct lustre_cfg *lcfg)
327 struct lwp_device *lwp;
328 struct lu_device *ludev;
332 ludev = ERR_PTR(-ENOMEM);
336 ludev = lwp2lu_dev(lwp);
337 lu_device_init(&lwp->lpd_dev, ldt);
338 rc = lwp_init0(env, lwp, ldt, lcfg);
340 lwp_device_free(env, ludev);
349 * Implementation of lu_device_type_operations::ltdo_device_fini.
351 * Finalize LWP device.
353 * \param[in] env environment passed by caller
354 * \param[in] ludev device to be finalized
356 * \retval NULL on success
358 static struct lu_device *lwp_device_fini(const struct lu_env *env,
359 struct lu_device *ludev)
361 struct lwp_device *m = lu2lwp_dev(ludev);
362 struct ptlrpc_thread *thread = &m->lpd_notify_thread;
363 struct l_wait_info lwi = { 0 };
367 if (!thread_is_stopped(thread))
368 l_wait_event(thread->t_ctl_waitq, thread_is_stopped(thread),
371 if (m->lpd_exp != NULL)
372 class_disconnect(m->lpd_exp);
375 ptlrpc_lprocfs_unregister_obd(m->lpd_obd);
376 lprocfs_obd_cleanup(m->lpd_obd);
378 rc = client_obd_cleanup(m->lpd_obd);
379 LASSERTF(rc == 0, "error %d\n", rc);
386 static struct lu_device_type_operations lwp_device_type_ops = {
387 .ldto_device_alloc = lwp_device_alloc,
388 .ldto_device_free = lwp_device_free,
389 .ldto_device_fini = lwp_device_fini
392 struct lu_device_type lwp_device_type = {
393 .ldt_tags = LU_DEVICE_DT,
394 .ldt_name = LUSTRE_LWP_NAME,
395 .ldt_ops = &lwp_device_type_ops,
396 .ldt_ctx_tags = LCT_MD_THREAD
399 static int lwp_notify_main(void *args)
401 struct obd_export *exp = (struct obd_export *)args;
402 struct lwp_device *lwp;
403 struct ptlrpc_thread *thread;
405 LASSERT(exp != NULL);
406 class_export_get(exp);
408 lwp = lu2lwp_dev(exp->exp_obd->obd_lu_dev);
409 thread = &lwp->lpd_notify_thread;
411 thread_set_flags(thread, SVC_RUNNING);
412 wake_up(&thread->t_ctl_waitq);
414 lustre_notify_lwp_list(exp);
416 class_export_put(exp);
417 thread_set_flags(thread, SVC_STOPPED);
418 wake_up(&thread->t_ctl_waitq);
423 * Some notify callbacks may cause deadlock in failover
424 * scenario, so we have to start thread to run callbacks
425 * asynchronously. See LU-6273.
427 static void lwp_notify_users(struct obd_export *exp)
429 struct lwp_device *lwp;
430 struct ptlrpc_thread *thread;
431 struct task_struct *task;
432 struct l_wait_info lwi = { 0 };
433 char name[MTI_NAME_MAXLEN];
435 LASSERT(exp != NULL);
436 lwp = lu2lwp_dev(exp->exp_obd->obd_lu_dev);
437 thread = &lwp->lpd_notify_thread;
439 snprintf(name, MTI_NAME_MAXLEN, "lwp_notify_%s",
440 exp->exp_obd->obd_name);
442 /* Notify happens only on LWP setup, so there shouldn't
443 * be notify thread running */
444 if (!thread_is_stopped(thread)) {
445 CERROR("LWP notify thread: %s wasn't stopped\n", name);
449 task = kthread_run(lwp_notify_main, exp, name);
451 thread_set_flags(thread, SVC_STOPPED);
452 CERROR("Failed to start LWP notify thread:%s. %lu\n",
453 name, PTR_ERR(task));
456 l_wait_event(thread->t_ctl_waitq,
457 thread_is_running(thread) || thread_is_stopped(thread),
462 * Implementation of OBD device operations obd_ops::o_connect.
464 * Create export for LWP, and connect to target server.
466 * \param[in] env the environment passed by caller
467 * \param[out] exp export for the connection to be established
468 * \param[in] obd OBD device to perform the connect on
469 * \param[in] cluuid UUID of the OBD device
470 * \param[in] data connect data containing compatibility flags
471 * \param[in] localdata not used
473 * \retval 0 on success
474 * \retval negative number on error
476 static int lwp_obd_connect(const struct lu_env *env, struct obd_export **exp,
477 struct obd_device *obd, struct obd_uuid *cluuid,
478 struct obd_connect_data *data, void *localdata)
480 struct lwp_device *lwp = lu2lwp_dev(obd->obd_lu_dev);
481 struct client_obd *cli = &lwp->lpd_obd->u.cli;
482 struct obd_import *imp = cli->cl_import;
483 struct obd_connect_data *ocd;
484 struct lustre_handle conn;
489 CDEBUG(D_CONFIG, "connect #%d\n", lwp->lpd_connects);
492 down_write(&cli->cl_sem);
493 rc = class_connect(&conn, obd, cluuid);
497 *exp = class_conn2export(&conn);
501 LASSERT(lwp->lpd_connects == 1);
503 imp->imp_dlm_handle = conn;
504 rc = ptlrpc_init_import(imp);
508 LASSERT(data != NULL);
509 ocd = &imp->imp_connect_data;
512 LASSERT(ocd->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT);
514 ocd->ocd_version = LUSTRE_VERSION_CODE;
515 imp->imp_connect_flags_orig = ocd->ocd_connect_flags;
516 imp->imp_connect_flags2_orig = ocd->ocd_connect_flags2;
518 rc = ptlrpc_connect_import(imp);
520 CERROR("%s: can't connect obd: rc = %d\n", obd->obd_name, rc);
524 ptlrpc_pinger_add_import(imp);
526 GOTO(out_dis, rc = 0);
530 class_disconnect(*exp);
536 up_write(&cli->cl_sem);
539 lwp_notify_users(*exp);
545 * Implementation of OBD device operations obd_ops::o_disconnect.
547 * Release export for the LWP. Only disconnect the underlying layers
548 * on the final disconnect.
550 * \param[in] exp the export to perform disconnect on
552 * \retval 0 on success
553 * \retval negative number on error
555 static int lwp_obd_disconnect(struct obd_export *exp)
557 struct obd_device *obd = exp->exp_obd;
558 struct lwp_device *lwp = lu2lwp_dev(obd->obd_lu_dev);
562 LASSERT(lwp->lpd_connects == 1);
565 rc = class_disconnect(exp);
567 CERROR("%s: class disconnect error: rc = %d\n",
574 * Handle import events for the LWP device.
576 * \param[in] obd OBD device associated with the import
577 * \param[in] imp the import which event happened on
578 * \param[in] event event type
580 * \retval 0 on success
581 * \retval negative number on error
583 static int lwp_import_event(struct obd_device *obd, struct obd_import *imp,
584 enum obd_import_event event)
587 case IMP_EVENT_DISCON:
588 case IMP_EVENT_INACTIVE:
589 case IMP_EVENT_ACTIVE:
591 case IMP_EVENT_INVALIDATE:
592 if (obd->obd_namespace == NULL)
594 ldlm_namespace_cleanup(obd->obd_namespace, LDLM_FL_LOCAL_ONLY);
599 CERROR("%s: unsupported import event: %#x\n",
600 obd->obd_name, event);
605 static int lwp_set_info_async(const struct lu_env *env,
606 struct obd_export *exp,
607 u32 keylen, void *key,
608 u32 vallen, void *val,
609 struct ptlrpc_request_set *set)
613 if (KEY_IS(KEY_SPTLRPC_CONF)) {
614 sptlrpc_conf_client_adapt(exp->exp_obd);
618 CERROR("Unknown key %s\n", (char *)key);
622 struct obd_ops lwp_obd_device_ops = {
623 .o_owner = THIS_MODULE,
624 .o_add_conn = client_import_add_conn,
625 .o_del_conn = client_import_del_conn,
626 .o_connect = lwp_obd_connect,
627 .o_disconnect = lwp_obd_disconnect,
628 .o_import_event = lwp_import_event,
629 .o_set_info_async = lwp_set_info_async,