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, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 1999, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2013, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/obdclass/genops.c
38 * These are the only exported functions, they provide some generic
39 * infrastructure for managing object devices
42 #define DEBUG_SUBSYSTEM S_CLASS
43 #include <obd_class.h>
44 #include <lprocfs_status.h>
46 spinlock_t obd_types_lock;
48 struct kmem_cache *obd_device_cachep;
49 struct kmem_cache *obdo_cachep;
50 EXPORT_SYMBOL(obdo_cachep);
51 struct kmem_cache *import_cachep;
53 struct list_head obd_zombie_imports;
54 struct list_head obd_zombie_exports;
55 spinlock_t obd_zombie_impexp_lock;
56 static void obd_zombie_impexp_notify(void);
57 static void obd_zombie_export_add(struct obd_export *exp);
58 static void obd_zombie_import_add(struct obd_import *imp);
59 static void print_export_data(struct obd_export *exp,
60 const char *status, int locks);
62 int (*ptlrpc_put_connection_superhack)(struct ptlrpc_connection *c);
63 EXPORT_SYMBOL(ptlrpc_put_connection_superhack);
66 * support functions: we could use inter-module communication, but this
67 * is more portable to other OS's
69 static struct obd_device *obd_device_alloc(void)
71 struct obd_device *obd;
73 OBD_SLAB_ALLOC_PTR_GFP(obd, obd_device_cachep, GFP_NOFS);
75 obd->obd_magic = OBD_DEVICE_MAGIC;
80 static void obd_device_free(struct obd_device *obd)
83 LASSERTF(obd->obd_magic == OBD_DEVICE_MAGIC, "obd %p obd_magic %08x != %08x\n",
84 obd, obd->obd_magic, OBD_DEVICE_MAGIC);
85 if (obd->obd_namespace != NULL) {
86 CERROR("obd %p: namespace %p was not properly cleaned up (obd_force=%d)!\n",
87 obd, obd->obd_namespace, obd->obd_force);
90 lu_ref_fini(&obd->obd_reference);
91 OBD_SLAB_FREE_PTR(obd, obd_device_cachep);
94 struct obd_type *class_search_type(const char *name)
96 struct list_head *tmp;
97 struct obd_type *type;
99 spin_lock(&obd_types_lock);
100 list_for_each(tmp, &obd_types) {
101 type = list_entry(tmp, struct obd_type, typ_chain);
102 if (strcmp(type->typ_name, name) == 0) {
103 spin_unlock(&obd_types_lock);
107 spin_unlock(&obd_types_lock);
110 EXPORT_SYMBOL(class_search_type);
112 struct obd_type *class_get_type(const char *name)
114 struct obd_type *type = class_search_type(name);
116 #ifdef HAVE_MODULE_LOADING_SUPPORT
118 const char *modname = name;
120 if (strcmp(modname, "obdfilter") == 0)
123 if (strcmp(modname, LUSTRE_LWP_NAME) == 0)
124 modname = LUSTRE_OSP_NAME;
126 if (!strncmp(modname, LUSTRE_MDS_NAME, strlen(LUSTRE_MDS_NAME)))
127 modname = LUSTRE_MDT_NAME;
129 if (!request_module("%s", modname)) {
130 CDEBUG(D_INFO, "Loaded module '%s'\n", modname);
131 type = class_search_type(name);
133 LCONSOLE_ERROR_MSG(0x158, "Can't load module '%s'\n",
139 spin_lock(&type->obd_type_lock);
141 try_module_get(type->typ_dt_ops->o_owner);
142 spin_unlock(&type->obd_type_lock);
146 EXPORT_SYMBOL(class_get_type);
148 void class_put_type(struct obd_type *type)
151 spin_lock(&type->obd_type_lock);
153 module_put(type->typ_dt_ops->o_owner);
154 spin_unlock(&type->obd_type_lock);
156 EXPORT_SYMBOL(class_put_type);
158 #define CLASS_MAX_NAME 1024
160 int class_register_type(struct obd_ops *dt_ops, struct md_ops *md_ops,
161 bool enable_proc, struct lprocfs_seq_vars *vars,
162 const char *name, struct lu_device_type *ldt)
164 struct obd_type *type;
169 LASSERT(strnlen(name, CLASS_MAX_NAME) < CLASS_MAX_NAME);
171 if (class_search_type(name)) {
172 CDEBUG(D_IOCTL, "Type %s already registered\n", name);
177 OBD_ALLOC(type, sizeof(*type));
181 OBD_ALLOC_PTR(type->typ_dt_ops);
182 OBD_ALLOC_PTR(type->typ_md_ops);
183 OBD_ALLOC(type->typ_name, strlen(name) + 1);
185 if (type->typ_dt_ops == NULL ||
186 type->typ_md_ops == NULL ||
187 type->typ_name == NULL)
190 *(type->typ_dt_ops) = *dt_ops;
191 /* md_ops is optional */
193 *(type->typ_md_ops) = *md_ops;
194 strcpy(type->typ_name, name);
195 spin_lock_init(&type->obd_type_lock);
199 type->typ_procroot = lprocfs_seq_register(type->typ_name,
202 if (IS_ERR(type->typ_procroot)) {
203 rc = PTR_ERR(type->typ_procroot);
204 type->typ_procroot = NULL;
211 rc = lu_device_type_init(ldt);
216 spin_lock(&obd_types_lock);
217 list_add(&type->typ_chain, &obd_types);
218 spin_unlock(&obd_types_lock);
223 if (type->typ_name != NULL) {
225 if (type->typ_procroot != NULL) {
226 #ifndef HAVE_ONLY_PROCFS_SEQ
227 lprocfs_try_remove_proc_entry(type->typ_name,
230 remove_proc_subtree(type->typ_name, proc_lustre_root);
234 OBD_FREE(type->typ_name, strlen(name) + 1);
236 if (type->typ_md_ops != NULL)
237 OBD_FREE_PTR(type->typ_md_ops);
238 if (type->typ_dt_ops != NULL)
239 OBD_FREE_PTR(type->typ_dt_ops);
240 OBD_FREE(type, sizeof(*type));
243 EXPORT_SYMBOL(class_register_type);
245 int class_unregister_type(const char *name)
247 struct obd_type *type = class_search_type(name);
251 CERROR("unknown obd type\n");
255 if (type->typ_refcnt) {
256 CERROR("type %s has refcount (%d)\n", name, type->typ_refcnt);
257 /* This is a bad situation, let's make the best of it */
258 /* Remove ops, but leave the name for debugging */
259 OBD_FREE_PTR(type->typ_dt_ops);
260 OBD_FREE_PTR(type->typ_md_ops);
264 /* we do not use type->typ_procroot as for compatibility purposes
265 * other modules can share names (i.e. lod can use lov entry). so
266 * we can't reference pointer as it can get invalided when another
267 * module removes the entry */
269 if (type->typ_procroot != NULL) {
270 #ifndef HAVE_ONLY_PROCFS_SEQ
271 lprocfs_try_remove_proc_entry(type->typ_name, proc_lustre_root);
273 remove_proc_subtree(type->typ_name, proc_lustre_root);
277 if (type->typ_procsym != NULL)
278 lprocfs_remove(&type->typ_procsym);
281 lu_device_type_fini(type->typ_lu);
283 spin_lock(&obd_types_lock);
284 list_del(&type->typ_chain);
285 spin_unlock(&obd_types_lock);
286 OBD_FREE(type->typ_name, strlen(name) + 1);
287 if (type->typ_dt_ops != NULL)
288 OBD_FREE_PTR(type->typ_dt_ops);
289 if (type->typ_md_ops != NULL)
290 OBD_FREE_PTR(type->typ_md_ops);
291 OBD_FREE(type, sizeof(*type));
293 } /* class_unregister_type */
294 EXPORT_SYMBOL(class_unregister_type);
297 * Create a new obd device.
299 * Find an empty slot in ::obd_devs[], create a new obd device in it.
301 * \param[in] type_name obd device type string.
302 * \param[in] name obd device name.
304 * \retval NULL if create fails, otherwise return the obd device
307 struct obd_device *class_newdev(const char *type_name, const char *name)
309 struct obd_device *result = NULL;
310 struct obd_device *newdev;
311 struct obd_type *type = NULL;
313 int new_obd_minor = 0;
316 if (strlen(name) >= MAX_OBD_NAME) {
317 CERROR("name/uuid must be < %u bytes long\n", MAX_OBD_NAME);
318 RETURN(ERR_PTR(-EINVAL));
321 type = class_get_type(type_name);
323 CERROR("OBD: unknown type: %s\n", type_name);
324 RETURN(ERR_PTR(-ENODEV));
327 newdev = obd_device_alloc();
329 GOTO(out_type, result = ERR_PTR(-ENOMEM));
331 LASSERT(newdev->obd_magic == OBD_DEVICE_MAGIC);
333 write_lock(&obd_dev_lock);
334 for (i = 0; i < class_devno_max(); i++) {
335 struct obd_device *obd = class_num2obd(i);
337 if (obd && (strcmp(name, obd->obd_name) == 0)) {
338 CERROR("Device %s already exists at %d, won't add\n",
341 LASSERTF(result->obd_magic == OBD_DEVICE_MAGIC,
342 "%p obd_magic %08x != %08x\n", result,
343 result->obd_magic, OBD_DEVICE_MAGIC);
344 LASSERTF(result->obd_minor == new_obd_minor,
345 "%p obd_minor %d != %d\n", result,
346 result->obd_minor, new_obd_minor);
348 obd_devs[result->obd_minor] = NULL;
349 result->obd_name[0]='\0';
351 result = ERR_PTR(-EEXIST);
354 if (!result && !obd) {
356 result->obd_minor = i;
358 result->obd_type = type;
359 strncpy(result->obd_name, name,
360 sizeof(result->obd_name) - 1);
361 obd_devs[i] = result;
364 write_unlock(&obd_dev_lock);
366 if (result == NULL && i >= class_devno_max()) {
367 CERROR("all %u OBD devices used, increase MAX_OBD_DEVICES\n",
369 GOTO(out, result = ERR_PTR(-EOVERFLOW));
375 CDEBUG(D_IOCTL, "Adding new device %s (%p)\n",
376 result->obd_name, result);
380 obd_device_free(newdev);
382 class_put_type(type);
386 void class_release_dev(struct obd_device *obd)
388 struct obd_type *obd_type = obd->obd_type;
390 LASSERTF(obd->obd_magic == OBD_DEVICE_MAGIC, "%p obd_magic %08x != %08x\n",
391 obd, obd->obd_magic, OBD_DEVICE_MAGIC);
392 LASSERTF(obd == obd_devs[obd->obd_minor], "obd %p != obd_devs[%d] %p\n",
393 obd, obd->obd_minor, obd_devs[obd->obd_minor]);
394 LASSERT(obd_type != NULL);
396 CDEBUG(D_INFO, "Release obd device %s at %d obd_type name =%s\n",
397 obd->obd_name, obd->obd_minor, obd->obd_type->typ_name);
399 write_lock(&obd_dev_lock);
400 obd_devs[obd->obd_minor] = NULL;
401 write_unlock(&obd_dev_lock);
402 obd_device_free(obd);
404 class_put_type(obd_type);
407 int class_name2dev(const char *name)
414 read_lock(&obd_dev_lock);
415 for (i = 0; i < class_devno_max(); i++) {
416 struct obd_device *obd = class_num2obd(i);
418 if (obd && strcmp(name, obd->obd_name) == 0) {
419 /* Make sure we finished attaching before we give
420 out any references */
421 LASSERT(obd->obd_magic == OBD_DEVICE_MAGIC);
422 if (obd->obd_attached) {
423 read_unlock(&obd_dev_lock);
429 read_unlock(&obd_dev_lock);
433 EXPORT_SYMBOL(class_name2dev);
435 struct obd_device *class_name2obd(const char *name)
437 int dev = class_name2dev(name);
439 if (dev < 0 || dev > class_devno_max())
441 return class_num2obd(dev);
443 EXPORT_SYMBOL(class_name2obd);
445 int class_uuid2dev(struct obd_uuid *uuid)
449 read_lock(&obd_dev_lock);
450 for (i = 0; i < class_devno_max(); i++) {
451 struct obd_device *obd = class_num2obd(i);
453 if (obd && obd_uuid_equals(uuid, &obd->obd_uuid)) {
454 LASSERT(obd->obd_magic == OBD_DEVICE_MAGIC);
455 read_unlock(&obd_dev_lock);
459 read_unlock(&obd_dev_lock);
463 EXPORT_SYMBOL(class_uuid2dev);
465 struct obd_device *class_uuid2obd(struct obd_uuid *uuid)
467 int dev = class_uuid2dev(uuid);
470 return class_num2obd(dev);
472 EXPORT_SYMBOL(class_uuid2obd);
475 * Get obd device from ::obd_devs[]
477 * \param num [in] array index
479 * \retval NULL if ::obd_devs[\a num] does not contains an obd device
480 * otherwise return the obd device there.
482 struct obd_device *class_num2obd(int num)
484 struct obd_device *obd = NULL;
486 if (num < class_devno_max()) {
491 LASSERTF(obd->obd_magic == OBD_DEVICE_MAGIC,
492 "%p obd_magic %08x != %08x\n",
493 obd, obd->obd_magic, OBD_DEVICE_MAGIC);
494 LASSERTF(obd->obd_minor == num,
495 "%p obd_minor %0d != %0d\n",
496 obd, obd->obd_minor, num);
501 EXPORT_SYMBOL(class_num2obd);
504 * Get obd devices count. Device in any
506 * \retval obd device count
508 int get_devices_count(void)
510 int index, max_index = class_devno_max(), dev_count = 0;
512 read_lock(&obd_dev_lock);
513 for (index = 0; index <= max_index; index++) {
514 struct obd_device *obd = class_num2obd(index);
518 read_unlock(&obd_dev_lock);
522 EXPORT_SYMBOL(get_devices_count);
524 void class_obd_list(void)
529 read_lock(&obd_dev_lock);
530 for (i = 0; i < class_devno_max(); i++) {
531 struct obd_device *obd = class_num2obd(i);
535 if (obd->obd_stopping)
537 else if (obd->obd_set_up)
539 else if (obd->obd_attached)
543 LCONSOLE(D_CONFIG, "%3d %s %s %s %s %d\n",
544 i, status, obd->obd_type->typ_name,
545 obd->obd_name, obd->obd_uuid.uuid,
546 atomic_read(&obd->obd_refcount));
548 read_unlock(&obd_dev_lock);
552 /* Search for a client OBD connected to tgt_uuid. If grp_uuid is
553 specified, then only the client with that uuid is returned,
554 otherwise any client connected to the tgt is returned. */
555 struct obd_device * class_find_client_obd(struct obd_uuid *tgt_uuid,
556 const char * typ_name,
557 struct obd_uuid *grp_uuid)
561 read_lock(&obd_dev_lock);
562 for (i = 0; i < class_devno_max(); i++) {
563 struct obd_device *obd = class_num2obd(i);
567 if ((strncmp(obd->obd_type->typ_name, typ_name,
568 strlen(typ_name)) == 0)) {
569 if (obd_uuid_equals(tgt_uuid,
570 &obd->u.cli.cl_target_uuid) &&
571 ((grp_uuid)? obd_uuid_equals(grp_uuid,
572 &obd->obd_uuid) : 1)) {
573 read_unlock(&obd_dev_lock);
578 read_unlock(&obd_dev_lock);
582 EXPORT_SYMBOL(class_find_client_obd);
584 /* Iterate the obd_device list looking devices have grp_uuid. Start
585 searching at *next, and if a device is found, the next index to look
586 at is saved in *next. If next is NULL, then the first matching device
587 will always be returned. */
588 struct obd_device * class_devices_in_group(struct obd_uuid *grp_uuid, int *next)
594 else if (*next >= 0 && *next < class_devno_max())
599 read_lock(&obd_dev_lock);
600 for (; i < class_devno_max(); i++) {
601 struct obd_device *obd = class_num2obd(i);
605 if (obd_uuid_equals(grp_uuid, &obd->obd_uuid)) {
608 read_unlock(&obd_dev_lock);
612 read_unlock(&obd_dev_lock);
616 EXPORT_SYMBOL(class_devices_in_group);
619 * to notify sptlrpc log for \a fsname has changed, let every relevant OBD
620 * adjust sptlrpc settings accordingly.
622 int class_notify_sptlrpc_conf(const char *fsname, int namelen)
624 struct obd_device *obd;
628 LASSERT(namelen > 0);
630 read_lock(&obd_dev_lock);
631 for (i = 0; i < class_devno_max(); i++) {
632 obd = class_num2obd(i);
634 if (obd == NULL || obd->obd_set_up == 0 || obd->obd_stopping)
637 /* only notify mdc, osc, mdt, ost */
638 type = obd->obd_type->typ_name;
639 if (strcmp(type, LUSTRE_MDC_NAME) != 0 &&
640 strcmp(type, LUSTRE_OSC_NAME) != 0 &&
641 strcmp(type, LUSTRE_MDT_NAME) != 0 &&
642 strcmp(type, LUSTRE_OST_NAME) != 0)
645 if (strncmp(obd->obd_name, fsname, namelen))
648 class_incref(obd, __FUNCTION__, obd);
649 read_unlock(&obd_dev_lock);
650 rc2 = obd_set_info_async(NULL, obd->obd_self_export,
651 sizeof(KEY_SPTLRPC_CONF),
652 KEY_SPTLRPC_CONF, 0, NULL, NULL);
654 class_decref(obd, __FUNCTION__, obd);
655 read_lock(&obd_dev_lock);
657 read_unlock(&obd_dev_lock);
660 EXPORT_SYMBOL(class_notify_sptlrpc_conf);
662 void obd_cleanup_caches(void)
665 if (obd_device_cachep) {
666 kmem_cache_destroy(obd_device_cachep);
667 obd_device_cachep = NULL;
670 kmem_cache_destroy(obdo_cachep);
674 kmem_cache_destroy(import_cachep);
675 import_cachep = NULL;
678 kmem_cache_destroy(capa_cachep);
684 int obd_init_caches(void)
689 LASSERT(obd_device_cachep == NULL);
690 obd_device_cachep = kmem_cache_create("ll_obd_dev_cache",
691 sizeof(struct obd_device),
693 if (!obd_device_cachep)
694 GOTO(out, rc = -ENOMEM);
696 LASSERT(obdo_cachep == NULL);
697 obdo_cachep = kmem_cache_create("ll_obdo_cache", sizeof(struct obdo),
700 GOTO(out, rc = -ENOMEM);
702 LASSERT(import_cachep == NULL);
703 import_cachep = kmem_cache_create("ll_import_cache",
704 sizeof(struct obd_import),
707 GOTO(out, rc = -ENOMEM);
709 LASSERT(capa_cachep == NULL);
710 capa_cachep = kmem_cache_create("capa_cache", sizeof(struct obd_capa),
713 GOTO(out, rc = -ENOMEM);
717 obd_cleanup_caches();
721 /* map connection to client */
722 struct obd_export *class_conn2export(struct lustre_handle *conn)
724 struct obd_export *export;
728 CDEBUG(D_CACHE, "looking for null handle\n");
732 if (conn->cookie == -1) { /* this means assign a new connection */
733 CDEBUG(D_CACHE, "want a new connection\n");
737 CDEBUG(D_INFO, "looking for export cookie "LPX64"\n", conn->cookie);
738 export = class_handle2object(conn->cookie, NULL);
741 EXPORT_SYMBOL(class_conn2export);
743 struct obd_device *class_exp2obd(struct obd_export *exp)
749 EXPORT_SYMBOL(class_exp2obd);
751 struct obd_device *class_conn2obd(struct lustre_handle *conn)
753 struct obd_export *export;
754 export = class_conn2export(conn);
756 struct obd_device *obd = export->exp_obd;
757 class_export_put(export);
762 EXPORT_SYMBOL(class_conn2obd);
764 struct obd_import *class_exp2cliimp(struct obd_export *exp)
766 struct obd_device *obd = exp->exp_obd;
769 return obd->u.cli.cl_import;
771 EXPORT_SYMBOL(class_exp2cliimp);
773 struct obd_import *class_conn2cliimp(struct lustre_handle *conn)
775 struct obd_device *obd = class_conn2obd(conn);
778 return obd->u.cli.cl_import;
780 EXPORT_SYMBOL(class_conn2cliimp);
782 /* Export management functions */
783 static void class_export_destroy(struct obd_export *exp)
785 struct obd_device *obd = exp->exp_obd;
788 LASSERT_ATOMIC_ZERO(&exp->exp_refcount);
789 LASSERT(obd != NULL);
791 CDEBUG(D_IOCTL, "destroying export %p/%s for %s\n", exp,
792 exp->exp_client_uuid.uuid, obd->obd_name);
794 /* "Local" exports (lctl, LOV->{mdc,osc}) have no connection. */
795 if (exp->exp_connection)
796 ptlrpc_put_connection_superhack(exp->exp_connection);
798 LASSERT(list_empty(&exp->exp_outstanding_replies));
799 LASSERT(list_empty(&exp->exp_uncommitted_replies));
800 LASSERT(list_empty(&exp->exp_req_replay_queue));
801 LASSERT(list_empty(&exp->exp_hp_rpcs));
802 obd_destroy_export(exp);
803 class_decref(obd, "export", exp);
805 OBD_FREE_RCU(exp, sizeof(*exp), &exp->exp_handle);
809 static void export_handle_addref(void *export)
811 class_export_get(export);
814 static struct portals_handle_ops export_handle_ops = {
815 .hop_addref = export_handle_addref,
819 struct obd_export *class_export_get(struct obd_export *exp)
821 atomic_inc(&exp->exp_refcount);
822 CDEBUG(D_INFO, "GETting export %p : new refcount %d\n", exp,
823 atomic_read(&exp->exp_refcount));
826 EXPORT_SYMBOL(class_export_get);
828 void class_export_put(struct obd_export *exp)
830 LASSERT(exp != NULL);
831 LASSERT_ATOMIC_GT_LT(&exp->exp_refcount, 0, LI_POISON);
832 CDEBUG(D_INFO, "PUTting export %p : new refcount %d\n", exp,
833 atomic_read(&exp->exp_refcount) - 1);
835 if (atomic_dec_and_test(&exp->exp_refcount)) {
836 LASSERT(!list_empty(&exp->exp_obd_chain));
837 CDEBUG(D_IOCTL, "final put %p/%s\n",
838 exp, exp->exp_client_uuid.uuid);
840 /* release nid stat refererence */
841 lprocfs_exp_cleanup(exp);
843 obd_zombie_export_add(exp);
846 EXPORT_SYMBOL(class_export_put);
848 /* Creates a new export, adds it to the hash table, and returns a
849 * pointer to it. The refcount is 2: one for the hash reference, and
850 * one for the pointer returned by this function. */
851 struct obd_export *class_new_export(struct obd_device *obd,
852 struct obd_uuid *cluuid)
854 struct obd_export *export;
855 cfs_hash_t *hash = NULL;
859 OBD_ALLOC_PTR(export);
861 return ERR_PTR(-ENOMEM);
863 export->exp_conn_cnt = 0;
864 export->exp_lock_hash = NULL;
865 export->exp_flock_hash = NULL;
866 atomic_set(&export->exp_refcount, 2);
867 atomic_set(&export->exp_rpc_count, 0);
868 atomic_set(&export->exp_cb_count, 0);
869 atomic_set(&export->exp_locks_count, 0);
870 #if LUSTRE_TRACKS_LOCK_EXP_REFS
871 INIT_LIST_HEAD(&export->exp_locks_list);
872 spin_lock_init(&export->exp_locks_list_guard);
874 atomic_set(&export->exp_replay_count, 0);
875 export->exp_obd = obd;
876 INIT_LIST_HEAD(&export->exp_outstanding_replies);
877 spin_lock_init(&export->exp_uncommitted_replies_lock);
878 INIT_LIST_HEAD(&export->exp_uncommitted_replies);
879 INIT_LIST_HEAD(&export->exp_req_replay_queue);
880 INIT_LIST_HEAD(&export->exp_handle.h_link);
881 INIT_LIST_HEAD(&export->exp_hp_rpcs);
882 INIT_LIST_HEAD(&export->exp_reg_rpcs);
883 class_handle_hash(&export->exp_handle, &export_handle_ops);
884 export->exp_last_request_time = cfs_time_current_sec();
885 spin_lock_init(&export->exp_lock);
886 spin_lock_init(&export->exp_rpc_lock);
887 INIT_HLIST_NODE(&export->exp_uuid_hash);
888 INIT_HLIST_NODE(&export->exp_nid_hash);
889 spin_lock_init(&export->exp_bl_list_lock);
890 INIT_LIST_HEAD(&export->exp_bl_list);
892 export->exp_sp_peer = LUSTRE_SP_ANY;
893 export->exp_flvr.sf_rpc = SPTLRPC_FLVR_INVALID;
894 export->exp_client_uuid = *cluuid;
895 obd_init_export(export);
897 spin_lock(&obd->obd_dev_lock);
898 /* shouldn't happen, but might race */
899 if (obd->obd_stopping)
900 GOTO(exit_unlock, rc = -ENODEV);
902 hash = cfs_hash_getref(obd->obd_uuid_hash);
904 GOTO(exit_unlock, rc = -ENODEV);
905 spin_unlock(&obd->obd_dev_lock);
907 if (!obd_uuid_equals(cluuid, &obd->obd_uuid)) {
908 rc = cfs_hash_add_unique(hash, cluuid, &export->exp_uuid_hash);
910 LCONSOLE_WARN("%s: denying duplicate export for %s, %d\n",
911 obd->obd_name, cluuid->uuid, rc);
912 GOTO(exit_err, rc = -EALREADY);
916 spin_lock(&obd->obd_dev_lock);
917 if (obd->obd_stopping) {
918 cfs_hash_del(hash, cluuid, &export->exp_uuid_hash);
919 GOTO(exit_unlock, rc = -ENODEV);
922 class_incref(obd, "export", export);
923 list_add(&export->exp_obd_chain, &export->exp_obd->obd_exports);
924 list_add_tail(&export->exp_obd_chain_timed,
925 &export->exp_obd->obd_exports_timed);
926 export->exp_obd->obd_num_exports++;
927 spin_unlock(&obd->obd_dev_lock);
928 cfs_hash_putref(hash);
932 spin_unlock(&obd->obd_dev_lock);
935 cfs_hash_putref(hash);
936 class_handle_unhash(&export->exp_handle);
937 LASSERT(hlist_unhashed(&export->exp_uuid_hash));
938 obd_destroy_export(export);
939 OBD_FREE_PTR(export);
942 EXPORT_SYMBOL(class_new_export);
944 void class_unlink_export(struct obd_export *exp)
946 class_handle_unhash(&exp->exp_handle);
948 spin_lock(&exp->exp_obd->obd_dev_lock);
949 /* delete an uuid-export hashitem from hashtables */
950 if (!hlist_unhashed(&exp->exp_uuid_hash))
951 cfs_hash_del(exp->exp_obd->obd_uuid_hash,
952 &exp->exp_client_uuid,
953 &exp->exp_uuid_hash);
955 list_move(&exp->exp_obd_chain, &exp->exp_obd->obd_unlinked_exports);
956 list_del_init(&exp->exp_obd_chain_timed);
957 exp->exp_obd->obd_num_exports--;
958 spin_unlock(&exp->exp_obd->obd_dev_lock);
959 class_export_put(exp);
961 EXPORT_SYMBOL(class_unlink_export);
963 /* Import management functions */
964 void class_import_destroy(struct obd_import *imp)
968 CDEBUG(D_IOCTL, "destroying import %p for %s\n", imp,
969 imp->imp_obd->obd_name);
971 LASSERT_ATOMIC_ZERO(&imp->imp_refcount);
973 ptlrpc_put_connection_superhack(imp->imp_connection);
975 while (!list_empty(&imp->imp_conn_list)) {
976 struct obd_import_conn *imp_conn;
978 imp_conn = list_entry(imp->imp_conn_list.next,
979 struct obd_import_conn, oic_item);
980 list_del_init(&imp_conn->oic_item);
981 ptlrpc_put_connection_superhack(imp_conn->oic_conn);
982 OBD_FREE(imp_conn, sizeof(*imp_conn));
985 LASSERT(imp->imp_sec == NULL);
986 class_decref(imp->imp_obd, "import", imp);
987 OBD_FREE_RCU(imp, sizeof(*imp), &imp->imp_handle);
991 static void import_handle_addref(void *import)
993 class_import_get(import);
996 static struct portals_handle_ops import_handle_ops = {
997 .hop_addref = import_handle_addref,
1001 struct obd_import *class_import_get(struct obd_import *import)
1003 atomic_inc(&import->imp_refcount);
1004 CDEBUG(D_INFO, "import %p refcount=%d obd=%s\n", import,
1005 atomic_read(&import->imp_refcount),
1006 import->imp_obd->obd_name);
1009 EXPORT_SYMBOL(class_import_get);
1011 void class_import_put(struct obd_import *imp)
1015 LASSERT(list_empty(&imp->imp_zombie_chain));
1016 LASSERT_ATOMIC_GT_LT(&imp->imp_refcount, 0, LI_POISON);
1018 CDEBUG(D_INFO, "import %p refcount=%d obd=%s\n", imp,
1019 atomic_read(&imp->imp_refcount) - 1,
1020 imp->imp_obd->obd_name);
1022 if (atomic_dec_and_test(&imp->imp_refcount)) {
1023 CDEBUG(D_INFO, "final put import %p\n", imp);
1024 obd_zombie_import_add(imp);
1027 /* catch possible import put race */
1028 LASSERT_ATOMIC_GE_LT(&imp->imp_refcount, 0, LI_POISON);
1031 EXPORT_SYMBOL(class_import_put);
1033 static void init_imp_at(struct imp_at *at) {
1035 at_init(&at->iat_net_latency, 0, 0);
1036 for (i = 0; i < IMP_AT_MAX_PORTALS; i++) {
1037 /* max service estimates are tracked on the server side, so
1038 don't use the AT history here, just use the last reported
1039 val. (But keep hist for proc histogram, worst_ever) */
1040 at_init(&at->iat_service_estimate[i], INITIAL_CONNECT_TIMEOUT,
1045 struct obd_import *class_new_import(struct obd_device *obd)
1047 struct obd_import *imp;
1049 OBD_ALLOC(imp, sizeof(*imp));
1053 INIT_LIST_HEAD(&imp->imp_pinger_chain);
1054 INIT_LIST_HEAD(&imp->imp_zombie_chain);
1055 INIT_LIST_HEAD(&imp->imp_replay_list);
1056 INIT_LIST_HEAD(&imp->imp_sending_list);
1057 INIT_LIST_HEAD(&imp->imp_delayed_list);
1058 INIT_LIST_HEAD(&imp->imp_committed_list);
1059 imp->imp_replay_cursor = &imp->imp_committed_list;
1060 spin_lock_init(&imp->imp_lock);
1061 imp->imp_last_success_conn = 0;
1062 imp->imp_state = LUSTRE_IMP_NEW;
1063 imp->imp_obd = class_incref(obd, "import", imp);
1064 mutex_init(&imp->imp_sec_mutex);
1065 init_waitqueue_head(&imp->imp_recovery_waitq);
1067 atomic_set(&imp->imp_refcount, 2);
1068 atomic_set(&imp->imp_unregistering, 0);
1069 atomic_set(&imp->imp_inflight, 0);
1070 atomic_set(&imp->imp_replay_inflight, 0);
1071 atomic_set(&imp->imp_inval_count, 0);
1072 INIT_LIST_HEAD(&imp->imp_conn_list);
1073 INIT_LIST_HEAD(&imp->imp_handle.h_link);
1074 class_handle_hash(&imp->imp_handle, &import_handle_ops);
1075 init_imp_at(&imp->imp_at);
1077 /* the default magic is V2, will be used in connect RPC, and
1078 * then adjusted according to the flags in request/reply. */
1079 imp->imp_msg_magic = LUSTRE_MSG_MAGIC_V2;
1083 EXPORT_SYMBOL(class_new_import);
1085 void class_destroy_import(struct obd_import *import)
1087 LASSERT(import != NULL);
1088 LASSERT(import != LP_POISON);
1090 class_handle_unhash(&import->imp_handle);
1092 spin_lock(&import->imp_lock);
1093 import->imp_generation++;
1094 spin_unlock(&import->imp_lock);
1095 class_import_put(import);
1097 EXPORT_SYMBOL(class_destroy_import);
1099 #if LUSTRE_TRACKS_LOCK_EXP_REFS
1101 void __class_export_add_lock_ref(struct obd_export *exp, struct ldlm_lock *lock)
1103 spin_lock(&exp->exp_locks_list_guard);
1105 LASSERT(lock->l_exp_refs_nr >= 0);
1107 if (lock->l_exp_refs_target != NULL &&
1108 lock->l_exp_refs_target != exp) {
1109 LCONSOLE_WARN("setting export %p for lock %p which already has export %p\n",
1110 exp, lock, lock->l_exp_refs_target);
1112 if ((lock->l_exp_refs_nr ++) == 0) {
1113 list_add(&lock->l_exp_refs_link, &exp->exp_locks_list);
1114 lock->l_exp_refs_target = exp;
1116 CDEBUG(D_INFO, "lock = %p, export = %p, refs = %u\n",
1117 lock, exp, lock->l_exp_refs_nr);
1118 spin_unlock(&exp->exp_locks_list_guard);
1120 EXPORT_SYMBOL(__class_export_add_lock_ref);
1122 void __class_export_del_lock_ref(struct obd_export *exp, struct ldlm_lock *lock)
1124 spin_lock(&exp->exp_locks_list_guard);
1125 LASSERT(lock->l_exp_refs_nr > 0);
1126 if (lock->l_exp_refs_target != exp) {
1127 LCONSOLE_WARN("lock %p, "
1128 "mismatching export pointers: %p, %p\n",
1129 lock, lock->l_exp_refs_target, exp);
1131 if (-- lock->l_exp_refs_nr == 0) {
1132 list_del_init(&lock->l_exp_refs_link);
1133 lock->l_exp_refs_target = NULL;
1135 CDEBUG(D_INFO, "lock = %p, export = %p, refs = %u\n",
1136 lock, exp, lock->l_exp_refs_nr);
1137 spin_unlock(&exp->exp_locks_list_guard);
1139 EXPORT_SYMBOL(__class_export_del_lock_ref);
1142 /* A connection defines an export context in which preallocation can
1143 be managed. This releases the export pointer reference, and returns
1144 the export handle, so the export refcount is 1 when this function
1146 int class_connect(struct lustre_handle *conn, struct obd_device *obd,
1147 struct obd_uuid *cluuid)
1149 struct obd_export *export;
1150 LASSERT(conn != NULL);
1151 LASSERT(obd != NULL);
1152 LASSERT(cluuid != NULL);
1155 export = class_new_export(obd, cluuid);
1157 RETURN(PTR_ERR(export));
1159 conn->cookie = export->exp_handle.h_cookie;
1160 class_export_put(export);
1162 CDEBUG(D_IOCTL, "connect: client %s, cookie "LPX64"\n",
1163 cluuid->uuid, conn->cookie);
1166 EXPORT_SYMBOL(class_connect);
1168 /* if export is involved in recovery then clean up related things */
1169 void class_export_recovery_cleanup(struct obd_export *exp)
1171 struct obd_device *obd = exp->exp_obd;
1173 spin_lock(&obd->obd_recovery_task_lock);
1174 if (obd->obd_recovering) {
1175 if (exp->exp_in_recovery) {
1176 spin_lock(&exp->exp_lock);
1177 exp->exp_in_recovery = 0;
1178 spin_unlock(&exp->exp_lock);
1179 LASSERT_ATOMIC_POS(&obd->obd_connected_clients);
1180 atomic_dec(&obd->obd_connected_clients);
1183 /* if called during recovery then should update
1184 * obd_stale_clients counter,
1185 * lightweight exports are not counted */
1186 if ((exp_connect_flags(exp) & OBD_CONNECT_LIGHTWEIGHT) == 0)
1187 exp->exp_obd->obd_stale_clients++;
1189 spin_unlock(&obd->obd_recovery_task_lock);
1191 spin_lock(&exp->exp_lock);
1192 /** Cleanup req replay fields */
1193 if (exp->exp_req_replay_needed) {
1194 exp->exp_req_replay_needed = 0;
1196 LASSERT(atomic_read(&obd->obd_req_replay_clients));
1197 atomic_dec(&obd->obd_req_replay_clients);
1200 /** Cleanup lock replay data */
1201 if (exp->exp_lock_replay_needed) {
1202 exp->exp_lock_replay_needed = 0;
1204 LASSERT(atomic_read(&obd->obd_lock_replay_clients));
1205 atomic_dec(&obd->obd_lock_replay_clients);
1207 spin_unlock(&exp->exp_lock);
1210 /* This function removes 1-3 references from the export:
1211 * 1 - for export pointer passed
1212 * and if disconnect really need
1213 * 2 - removing from hash
1214 * 3 - in client_unlink_export
1215 * The export pointer passed to this function can destroyed */
1216 int class_disconnect(struct obd_export *export)
1218 int already_disconnected;
1221 if (export == NULL) {
1222 CWARN("attempting to free NULL export %p\n", export);
1226 spin_lock(&export->exp_lock);
1227 already_disconnected = export->exp_disconnected;
1228 export->exp_disconnected = 1;
1229 spin_unlock(&export->exp_lock);
1231 /* class_cleanup(), abort_recovery(), and class_fail_export()
1232 * all end up in here, and if any of them race we shouldn't
1233 * call extra class_export_puts(). */
1234 if (already_disconnected) {
1235 LASSERT(hlist_unhashed(&export->exp_nid_hash));
1236 GOTO(no_disconn, already_disconnected);
1239 CDEBUG(D_IOCTL, "disconnect: cookie "LPX64"\n",
1240 export->exp_handle.h_cookie);
1242 if (!hlist_unhashed(&export->exp_nid_hash))
1243 cfs_hash_del(export->exp_obd->obd_nid_hash,
1244 &export->exp_connection->c_peer.nid,
1245 &export->exp_nid_hash);
1247 class_export_recovery_cleanup(export);
1248 class_unlink_export(export);
1250 class_export_put(export);
1253 EXPORT_SYMBOL(class_disconnect);
1255 /* Return non-zero for a fully connected export */
1256 int class_connected_export(struct obd_export *exp)
1261 spin_lock(&exp->exp_lock);
1262 connected = (exp->exp_conn_cnt > 0) && !exp->exp_failed;
1263 spin_unlock(&exp->exp_lock);
1267 EXPORT_SYMBOL(class_connected_export);
1269 static void class_disconnect_export_list(struct list_head *list,
1270 enum obd_option flags)
1273 struct obd_export *exp;
1276 /* It's possible that an export may disconnect itself, but
1277 * nothing else will be added to this list. */
1278 while (!list_empty(list)) {
1279 exp = list_entry(list->next, struct obd_export,
1281 /* need for safe call CDEBUG after obd_disconnect */
1282 class_export_get(exp);
1284 spin_lock(&exp->exp_lock);
1285 exp->exp_flags = flags;
1286 spin_unlock(&exp->exp_lock);
1288 if (obd_uuid_equals(&exp->exp_client_uuid,
1289 &exp->exp_obd->obd_uuid)) {
1291 "exp %p export uuid == obd uuid, don't discon\n",
1293 /* Need to delete this now so we don't end up pointing
1294 * to work_list later when this export is cleaned up. */
1295 list_del_init(&exp->exp_obd_chain);
1296 class_export_put(exp);
1300 class_export_get(exp);
1301 CDEBUG(D_HA, "%s: disconnecting export at %s (%p), "
1302 "last request at "CFS_TIME_T"\n",
1303 exp->exp_obd->obd_name, obd_export_nid2str(exp),
1304 exp, exp->exp_last_request_time);
1305 /* release one export reference anyway */
1306 rc = obd_disconnect(exp);
1308 CDEBUG(D_HA, "disconnected export at %s (%p): rc %d\n",
1309 obd_export_nid2str(exp), exp, rc);
1310 class_export_put(exp);
1315 void class_disconnect_exports(struct obd_device *obd)
1317 struct list_head work_list;
1320 /* Move all of the exports from obd_exports to a work list, en masse. */
1321 INIT_LIST_HEAD(&work_list);
1322 spin_lock(&obd->obd_dev_lock);
1323 list_splice_init(&obd->obd_exports, &work_list);
1324 list_splice_init(&obd->obd_delayed_exports, &work_list);
1325 spin_unlock(&obd->obd_dev_lock);
1327 if (!list_empty(&work_list)) {
1328 CDEBUG(D_HA, "OBD device %d (%p) has exports, "
1329 "disconnecting them\n", obd->obd_minor, obd);
1330 class_disconnect_export_list(&work_list,
1331 exp_flags_from_obd(obd));
1333 CDEBUG(D_HA, "OBD device %d (%p) has no exports\n",
1334 obd->obd_minor, obd);
1337 EXPORT_SYMBOL(class_disconnect_exports);
1339 /* Remove exports that have not completed recovery.
1341 void class_disconnect_stale_exports(struct obd_device *obd,
1342 int (*test_export)(struct obd_export *))
1344 struct list_head work_list;
1345 struct obd_export *exp, *n;
1349 INIT_LIST_HEAD(&work_list);
1350 spin_lock(&obd->obd_dev_lock);
1351 list_for_each_entry_safe(exp, n, &obd->obd_exports,
1353 /* don't count self-export as client */
1354 if (obd_uuid_equals(&exp->exp_client_uuid,
1355 &exp->exp_obd->obd_uuid))
1358 /* don't evict clients which have no slot in last_rcvd
1359 * (e.g. lightweight connection) */
1360 if (exp->exp_target_data.ted_lr_idx == -1)
1363 spin_lock(&exp->exp_lock);
1364 if (exp->exp_failed || test_export(exp)) {
1365 spin_unlock(&exp->exp_lock);
1368 exp->exp_failed = 1;
1369 spin_unlock(&exp->exp_lock);
1371 list_move(&exp->exp_obd_chain, &work_list);
1373 CDEBUG(D_HA, "%s: disconnect stale client %s@%s\n",
1374 obd->obd_name, exp->exp_client_uuid.uuid,
1375 exp->exp_connection == NULL ? "<unknown>" :
1376 libcfs_nid2str(exp->exp_connection->c_peer.nid));
1377 print_export_data(exp, "EVICTING", 0);
1379 spin_unlock(&obd->obd_dev_lock);
1382 LCONSOLE_WARN("%s: disconnecting %d stale clients\n",
1383 obd->obd_name, evicted);
1385 class_disconnect_export_list(&work_list, exp_flags_from_obd(obd) |
1386 OBD_OPT_ABORT_RECOV);
1389 EXPORT_SYMBOL(class_disconnect_stale_exports);
1391 void class_fail_export(struct obd_export *exp)
1393 int rc, already_failed;
1395 spin_lock(&exp->exp_lock);
1396 already_failed = exp->exp_failed;
1397 exp->exp_failed = 1;
1398 spin_unlock(&exp->exp_lock);
1400 if (already_failed) {
1401 CDEBUG(D_HA, "disconnecting dead export %p/%s; skipping\n",
1402 exp, exp->exp_client_uuid.uuid);
1406 CDEBUG(D_HA, "disconnecting export %p/%s\n",
1407 exp, exp->exp_client_uuid.uuid);
1409 if (obd_dump_on_timeout)
1410 libcfs_debug_dumplog();
1412 /* need for safe call CDEBUG after obd_disconnect */
1413 class_export_get(exp);
1415 /* Most callers into obd_disconnect are removing their own reference
1416 * (request, for example) in addition to the one from the hash table.
1417 * We don't have such a reference here, so make one. */
1418 class_export_get(exp);
1419 rc = obd_disconnect(exp);
1421 CERROR("disconnecting export %p failed: %d\n", exp, rc);
1423 CDEBUG(D_HA, "disconnected export %p/%s\n",
1424 exp, exp->exp_client_uuid.uuid);
1425 class_export_put(exp);
1427 EXPORT_SYMBOL(class_fail_export);
1429 char *obd_export_nid2str(struct obd_export *exp)
1431 if (exp->exp_connection != NULL)
1432 return libcfs_nid2str(exp->exp_connection->c_peer.nid);
1436 EXPORT_SYMBOL(obd_export_nid2str);
1438 int obd_export_evict_by_nid(struct obd_device *obd, const char *nid)
1440 cfs_hash_t *nid_hash;
1441 struct obd_export *doomed_exp = NULL;
1442 int exports_evicted = 0;
1444 lnet_nid_t nid_key = libcfs_str2nid((char *)nid);
1446 spin_lock(&obd->obd_dev_lock);
1447 /* umount has run already, so evict thread should leave
1448 * its task to umount thread now */
1449 if (obd->obd_stopping) {
1450 spin_unlock(&obd->obd_dev_lock);
1451 return exports_evicted;
1453 nid_hash = obd->obd_nid_hash;
1454 cfs_hash_getref(nid_hash);
1455 spin_unlock(&obd->obd_dev_lock);
1458 doomed_exp = cfs_hash_lookup(nid_hash, &nid_key);
1459 if (doomed_exp == NULL)
1462 LASSERTF(doomed_exp->exp_connection->c_peer.nid == nid_key,
1463 "nid %s found, wanted nid %s, requested nid %s\n",
1464 obd_export_nid2str(doomed_exp),
1465 libcfs_nid2str(nid_key), nid);
1466 LASSERTF(doomed_exp != obd->obd_self_export,
1467 "self-export is hashed by NID?\n");
1469 LCONSOLE_WARN("%s: evicting %s (at %s) by administrative "
1470 "request\n", obd->obd_name,
1471 obd_uuid2str(&doomed_exp->exp_client_uuid),
1472 obd_export_nid2str(doomed_exp));
1473 class_fail_export(doomed_exp);
1474 class_export_put(doomed_exp);
1477 cfs_hash_putref(nid_hash);
1479 if (!exports_evicted)
1480 CDEBUG(D_HA,"%s: can't disconnect NID '%s': no exports found\n",
1481 obd->obd_name, nid);
1482 return exports_evicted;
1484 EXPORT_SYMBOL(obd_export_evict_by_nid);
1486 int obd_export_evict_by_uuid(struct obd_device *obd, const char *uuid)
1488 cfs_hash_t *uuid_hash;
1489 struct obd_export *doomed_exp = NULL;
1490 struct obd_uuid doomed_uuid;
1491 int exports_evicted = 0;
1493 spin_lock(&obd->obd_dev_lock);
1494 if (obd->obd_stopping) {
1495 spin_unlock(&obd->obd_dev_lock);
1496 return exports_evicted;
1498 uuid_hash = obd->obd_uuid_hash;
1499 cfs_hash_getref(uuid_hash);
1500 spin_unlock(&obd->obd_dev_lock);
1502 obd_str2uuid(&doomed_uuid, uuid);
1503 if (obd_uuid_equals(&doomed_uuid, &obd->obd_uuid)) {
1504 CERROR("%s: can't evict myself\n", obd->obd_name);
1505 cfs_hash_putref(uuid_hash);
1506 return exports_evicted;
1509 doomed_exp = cfs_hash_lookup(uuid_hash, &doomed_uuid);
1511 if (doomed_exp == NULL) {
1512 CERROR("%s: can't disconnect %s: no exports found\n",
1513 obd->obd_name, uuid);
1515 CWARN("%s: evicting %s at adminstrative request\n",
1516 obd->obd_name, doomed_exp->exp_client_uuid.uuid);
1517 class_fail_export(doomed_exp);
1518 class_export_put(doomed_exp);
1521 cfs_hash_putref(uuid_hash);
1523 return exports_evicted;
1525 EXPORT_SYMBOL(obd_export_evict_by_uuid);
1527 #if LUSTRE_TRACKS_LOCK_EXP_REFS
1528 void (*class_export_dump_hook)(struct obd_export*) = NULL;
1529 EXPORT_SYMBOL(class_export_dump_hook);
1532 static void print_export_data(struct obd_export *exp, const char *status,
1535 struct ptlrpc_reply_state *rs;
1536 struct ptlrpc_reply_state *first_reply = NULL;
1539 spin_lock(&exp->exp_lock);
1540 list_for_each_entry(rs, &exp->exp_outstanding_replies,
1546 spin_unlock(&exp->exp_lock);
1548 CDEBUG(D_HA, "%s: %s %p %s %s %d (%d %d %d) %d %d %d %d: %p %s "LPU64"\n",
1549 exp->exp_obd->obd_name, status, exp, exp->exp_client_uuid.uuid,
1550 obd_export_nid2str(exp), atomic_read(&exp->exp_refcount),
1551 atomic_read(&exp->exp_rpc_count),
1552 atomic_read(&exp->exp_cb_count),
1553 atomic_read(&exp->exp_locks_count),
1554 exp->exp_disconnected, exp->exp_delayed, exp->exp_failed,
1555 nreplies, first_reply, nreplies > 3 ? "..." : "",
1556 exp->exp_last_committed);
1557 #if LUSTRE_TRACKS_LOCK_EXP_REFS
1558 if (locks && class_export_dump_hook != NULL)
1559 class_export_dump_hook(exp);
1563 void dump_exports(struct obd_device *obd, int locks)
1565 struct obd_export *exp;
1567 spin_lock(&obd->obd_dev_lock);
1568 list_for_each_entry(exp, &obd->obd_exports, exp_obd_chain)
1569 print_export_data(exp, "ACTIVE", locks);
1570 list_for_each_entry(exp, &obd->obd_unlinked_exports, exp_obd_chain)
1571 print_export_data(exp, "UNLINKED", locks);
1572 list_for_each_entry(exp, &obd->obd_delayed_exports, exp_obd_chain)
1573 print_export_data(exp, "DELAYED", locks);
1574 spin_unlock(&obd->obd_dev_lock);
1575 spin_lock(&obd_zombie_impexp_lock);
1576 list_for_each_entry(exp, &obd_zombie_exports, exp_obd_chain)
1577 print_export_data(exp, "ZOMBIE", locks);
1578 spin_unlock(&obd_zombie_impexp_lock);
1580 EXPORT_SYMBOL(dump_exports);
1582 void obd_exports_barrier(struct obd_device *obd)
1585 LASSERT(list_empty(&obd->obd_exports));
1586 spin_lock(&obd->obd_dev_lock);
1587 while (!list_empty(&obd->obd_unlinked_exports)) {
1588 spin_unlock(&obd->obd_dev_lock);
1589 schedule_timeout_and_set_state(TASK_UNINTERRUPTIBLE,
1590 cfs_time_seconds(waited));
1591 if (waited > 5 && IS_PO2(waited)) {
1592 LCONSOLE_WARN("%s is waiting for obd_unlinked_exports "
1593 "more than %d seconds. "
1594 "The obd refcount = %d. Is it stuck?\n",
1595 obd->obd_name, waited,
1596 atomic_read(&obd->obd_refcount));
1597 dump_exports(obd, 1);
1600 spin_lock(&obd->obd_dev_lock);
1602 spin_unlock(&obd->obd_dev_lock);
1604 EXPORT_SYMBOL(obd_exports_barrier);
1606 /* Total amount of zombies to be destroyed */
1607 static int zombies_count = 0;
1610 * kill zombie imports and exports
1612 void obd_zombie_impexp_cull(void)
1614 struct obd_import *import;
1615 struct obd_export *export;
1619 spin_lock(&obd_zombie_impexp_lock);
1622 if (!list_empty(&obd_zombie_imports)) {
1623 import = list_entry(obd_zombie_imports.next,
1626 list_del_init(&import->imp_zombie_chain);
1630 if (!list_empty(&obd_zombie_exports)) {
1631 export = list_entry(obd_zombie_exports.next,
1634 list_del_init(&export->exp_obd_chain);
1637 spin_unlock(&obd_zombie_impexp_lock);
1639 if (import != NULL) {
1640 class_import_destroy(import);
1641 spin_lock(&obd_zombie_impexp_lock);
1643 spin_unlock(&obd_zombie_impexp_lock);
1646 if (export != NULL) {
1647 class_export_destroy(export);
1648 spin_lock(&obd_zombie_impexp_lock);
1650 spin_unlock(&obd_zombie_impexp_lock);
1654 } while (import != NULL || export != NULL);
1658 static struct completion obd_zombie_start;
1659 static struct completion obd_zombie_stop;
1660 static unsigned long obd_zombie_flags;
1661 static wait_queue_head_t obd_zombie_waitq;
1662 static pid_t obd_zombie_pid;
1665 OBD_ZOMBIE_STOP = 0x0001,
1669 * check for work for kill zombie import/export thread.
1671 static int obd_zombie_impexp_check(void *arg)
1675 spin_lock(&obd_zombie_impexp_lock);
1676 rc = (zombies_count == 0) &&
1677 !test_bit(OBD_ZOMBIE_STOP, &obd_zombie_flags);
1678 spin_unlock(&obd_zombie_impexp_lock);
1684 * Add export to the obd_zombe thread and notify it.
1686 static void obd_zombie_export_add(struct obd_export *exp) {
1687 spin_lock(&exp->exp_obd->obd_dev_lock);
1688 LASSERT(!list_empty(&exp->exp_obd_chain));
1689 list_del_init(&exp->exp_obd_chain);
1690 spin_unlock(&exp->exp_obd->obd_dev_lock);
1691 spin_lock(&obd_zombie_impexp_lock);
1693 list_add(&exp->exp_obd_chain, &obd_zombie_exports);
1694 spin_unlock(&obd_zombie_impexp_lock);
1696 obd_zombie_impexp_notify();
1700 * Add import to the obd_zombe thread and notify it.
1702 static void obd_zombie_import_add(struct obd_import *imp) {
1703 LASSERT(imp->imp_sec == NULL);
1704 LASSERT(imp->imp_rq_pool == NULL);
1705 spin_lock(&obd_zombie_impexp_lock);
1706 LASSERT(list_empty(&imp->imp_zombie_chain));
1708 list_add(&imp->imp_zombie_chain, &obd_zombie_imports);
1709 spin_unlock(&obd_zombie_impexp_lock);
1711 obd_zombie_impexp_notify();
1715 * notify import/export destroy thread about new zombie.
1717 static void obd_zombie_impexp_notify(void)
1720 * Make sure obd_zomebie_impexp_thread get this notification.
1721 * It is possible this signal only get by obd_zombie_barrier, and
1722 * barrier gulps this notification and sleeps away and hangs ensues
1724 wake_up_all(&obd_zombie_waitq);
1728 * check whether obd_zombie is idle
1730 static int obd_zombie_is_idle(void)
1734 LASSERT(!test_bit(OBD_ZOMBIE_STOP, &obd_zombie_flags));
1735 spin_lock(&obd_zombie_impexp_lock);
1736 rc = (zombies_count == 0);
1737 spin_unlock(&obd_zombie_impexp_lock);
1742 * wait when obd_zombie import/export queues become empty
1744 void obd_zombie_barrier(void)
1746 struct l_wait_info lwi = { 0 };
1748 if (obd_zombie_pid == current_pid())
1749 /* don't wait for myself */
1751 l_wait_event(obd_zombie_waitq, obd_zombie_is_idle(), &lwi);
1753 EXPORT_SYMBOL(obd_zombie_barrier);
1757 * destroy zombie export/import thread.
1759 static int obd_zombie_impexp_thread(void *unused)
1761 unshare_fs_struct();
1762 complete(&obd_zombie_start);
1764 obd_zombie_pid = current_pid();
1766 while (!test_bit(OBD_ZOMBIE_STOP, &obd_zombie_flags)) {
1767 struct l_wait_info lwi = { 0 };
1769 l_wait_event(obd_zombie_waitq,
1770 !obd_zombie_impexp_check(NULL), &lwi);
1771 obd_zombie_impexp_cull();
1774 * Notify obd_zombie_barrier callers that queues
1777 wake_up(&obd_zombie_waitq);
1780 complete(&obd_zombie_stop);
1787 * start destroy zombie import/export thread
1789 int obd_zombie_impexp_init(void)
1791 struct task_struct *task;
1793 INIT_LIST_HEAD(&obd_zombie_imports);
1795 INIT_LIST_HEAD(&obd_zombie_exports);
1796 spin_lock_init(&obd_zombie_impexp_lock);
1797 init_completion(&obd_zombie_start);
1798 init_completion(&obd_zombie_stop);
1799 init_waitqueue_head(&obd_zombie_waitq);
1802 task = kthread_run(obd_zombie_impexp_thread, NULL, "obd_zombid");
1804 RETURN(PTR_ERR(task));
1806 wait_for_completion(&obd_zombie_start);
1810 * stop destroy zombie import/export thread
1812 void obd_zombie_impexp_stop(void)
1814 set_bit(OBD_ZOMBIE_STOP, &obd_zombie_flags);
1815 obd_zombie_impexp_notify();
1816 wait_for_completion(&obd_zombie_stop);
1819 /***** Kernel-userspace comm helpers *******/
1821 /* Get length of entire message, including header */
1822 int kuc_len(int payload_len)
1824 return sizeof(struct kuc_hdr) + payload_len;
1826 EXPORT_SYMBOL(kuc_len);
1828 /* Get a pointer to kuc header, given a ptr to the payload
1829 * @param p Pointer to payload area
1830 * @returns Pointer to kuc header
1832 struct kuc_hdr * kuc_ptr(void *p)
1834 struct kuc_hdr *lh = ((struct kuc_hdr *)p) - 1;
1835 LASSERT(lh->kuc_magic == KUC_MAGIC);
1838 EXPORT_SYMBOL(kuc_ptr);
1840 /* Test if payload is part of kuc message
1841 * @param p Pointer to payload area
1844 int kuc_ispayload(void *p)
1846 struct kuc_hdr *kh = ((struct kuc_hdr *)p) - 1;
1848 if (kh->kuc_magic == KUC_MAGIC)
1853 EXPORT_SYMBOL(kuc_ispayload);
1855 /* Alloc space for a message, and fill in header
1856 * @return Pointer to payload area
1858 void *kuc_alloc(int payload_len, int transport, int type)
1861 int len = kuc_len(payload_len);
1865 return ERR_PTR(-ENOMEM);
1867 lh->kuc_magic = KUC_MAGIC;
1868 lh->kuc_transport = transport;
1869 lh->kuc_msgtype = type;
1870 lh->kuc_msglen = len;
1872 return (void *)(lh + 1);
1874 EXPORT_SYMBOL(kuc_alloc);
1876 /* Takes pointer to payload area */
1877 inline void kuc_free(void *p, int payload_len)
1879 struct kuc_hdr *lh = kuc_ptr(p);
1880 OBD_FREE(lh, kuc_len(payload_len));
1882 EXPORT_SYMBOL(kuc_free);
1884 struct obd_request_slot_waiter {
1885 struct list_head orsw_entry;
1886 wait_queue_head_t orsw_waitq;
1890 static bool obd_request_slot_avail(struct client_obd *cli,
1891 struct obd_request_slot_waiter *orsw)
1895 client_obd_list_lock(&cli->cl_loi_list_lock);
1896 avail = !!list_empty(&orsw->orsw_entry);
1897 client_obd_list_unlock(&cli->cl_loi_list_lock);
1903 * For network flow control, the RPC sponsor needs to acquire a credit
1904 * before sending the RPC. The credits count for a connection is defined
1905 * by the "cl_max_rpcs_in_flight". If all the credits are occpuied, then
1906 * the subsequent RPC sponsors need to wait until others released their
1907 * credits, or the administrator increased the "cl_max_rpcs_in_flight".
1909 int obd_get_request_slot(struct client_obd *cli)
1911 struct obd_request_slot_waiter orsw;
1912 struct l_wait_info lwi;
1915 client_obd_list_lock(&cli->cl_loi_list_lock);
1916 if (cli->cl_r_in_flight < cli->cl_max_rpcs_in_flight) {
1917 cli->cl_r_in_flight++;
1918 client_obd_list_unlock(&cli->cl_loi_list_lock);
1922 init_waitqueue_head(&orsw.orsw_waitq);
1923 list_add_tail(&orsw.orsw_entry, &cli->cl_loi_read_list);
1924 orsw.orsw_signaled = false;
1925 client_obd_list_unlock(&cli->cl_loi_list_lock);
1927 lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1928 rc = l_wait_event(orsw.orsw_waitq,
1929 obd_request_slot_avail(cli, &orsw) ||
1933 /* Here, we must take the lock to avoid the on-stack 'orsw' to be
1934 * freed but other (such as obd_put_request_slot) is using it. */
1935 client_obd_list_lock(&cli->cl_loi_list_lock);
1937 if (!orsw.orsw_signaled) {
1938 if (list_empty(&orsw.orsw_entry))
1939 cli->cl_r_in_flight--;
1941 list_del(&orsw.orsw_entry);
1945 if (orsw.orsw_signaled) {
1946 LASSERT(list_empty(&orsw.orsw_entry));
1950 client_obd_list_unlock(&cli->cl_loi_list_lock);
1954 EXPORT_SYMBOL(obd_get_request_slot);
1956 void obd_put_request_slot(struct client_obd *cli)
1958 struct obd_request_slot_waiter *orsw;
1960 client_obd_list_lock(&cli->cl_loi_list_lock);
1961 cli->cl_r_in_flight--;
1963 /* If there is free slot, wakeup the first waiter. */
1964 if (!list_empty(&cli->cl_loi_read_list) &&
1965 likely(cli->cl_r_in_flight < cli->cl_max_rpcs_in_flight)) {
1966 orsw = list_entry(cli->cl_loi_read_list.next,
1967 struct obd_request_slot_waiter, orsw_entry);
1968 list_del_init(&orsw->orsw_entry);
1969 cli->cl_r_in_flight++;
1970 wake_up(&orsw->orsw_waitq);
1972 client_obd_list_unlock(&cli->cl_loi_list_lock);
1974 EXPORT_SYMBOL(obd_put_request_slot);
1976 __u32 obd_get_max_rpcs_in_flight(struct client_obd *cli)
1978 return cli->cl_max_rpcs_in_flight;
1980 EXPORT_SYMBOL(obd_get_max_rpcs_in_flight);
1982 int obd_set_max_rpcs_in_flight(struct client_obd *cli, __u32 max)
1984 struct obd_request_slot_waiter *orsw;
1989 if (max > OBD_MAX_RIF_MAX || max < 1)
1992 client_obd_list_lock(&cli->cl_loi_list_lock);
1993 old = cli->cl_max_rpcs_in_flight;
1994 cli->cl_max_rpcs_in_flight = max;
1997 /* We increase the max_rpcs_in_flight, then wakeup some waiters. */
1998 for (i = 0; i < diff; i++) {
1999 if (list_empty(&cli->cl_loi_read_list))
2002 orsw = list_entry(cli->cl_loi_read_list.next,
2003 struct obd_request_slot_waiter, orsw_entry);
2004 list_del_init(&orsw->orsw_entry);
2005 cli->cl_r_in_flight++;
2006 wake_up(&orsw->orsw_waitq);
2008 client_obd_list_unlock(&cli->cl_loi_list_lock);
2012 EXPORT_SYMBOL(obd_set_max_rpcs_in_flight);