1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Copyright (c) 2001-2003 Cluster File Systems, Inc.
6 * This file is part of Lustre, http://www.lustre.org.
8 * Lustre is free software; you can redistribute it and/or
9 * modify it under the terms of version 2 of the GNU General Public
10 * License as published by the Free Software Foundation.
12 * Lustre is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
17 * You should have received a copy of the GNU General Public License
18 * along with Lustre; if not, write to the Free Software
19 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
21 * These are the only exported functions, they provide some generic
22 * infrastructure for managing object devices
25 #define DEBUG_SUBSYSTEM S_CLASS
27 #include <linux/kmod.h> /* for request_module() */
28 #include <linux/module.h>
29 #include <linux/obd_class.h>
30 #include <linux/random.h>
31 #include <linux/slab.h>
32 #include <linux/pagemap.h>
34 #include <liblustre.h>
35 #include <linux/obd_class.h>
36 #include <linux/obd.h>
38 #include <linux/lprocfs_status.h>
40 extern struct list_head obd_types;
41 static spinlock_t obd_types_lock = SPIN_LOCK_UNLOCKED;
42 kmem_cache_t *obdo_cachep = NULL;
43 kmem_cache_t *import_cachep = NULL;
45 int (*ptlrpc_put_connection_superhack)(struct ptlrpc_connection *c);
46 void (*ptlrpc_abort_inflight_superhack)(struct obd_import *imp);
49 * support functions: we could use inter-module communication, but this
50 * is more portable to other OS's
52 static struct obd_type *class_search_type(char *name)
54 struct list_head *tmp;
55 struct obd_type *type;
57 spin_lock(&obd_types_lock);
58 list_for_each(tmp, &obd_types) {
59 type = list_entry(tmp, struct obd_type, typ_chain);
60 if (strlen(type->typ_name) == strlen(name) &&
61 strcmp(type->typ_name, name) == 0) {
62 spin_unlock(&obd_types_lock);
66 spin_unlock(&obd_types_lock);
70 struct obd_type *class_get_type(char *name)
72 struct obd_type *type = class_search_type(name);
76 if (!request_module(name)) {
77 CDEBUG(D_INFO, "Loaded module '%s'\n", name);
78 type = class_search_type(name);
80 CDEBUG(D_INFO, "Can't load module '%s'\n", name);
84 try_module_get(type->typ_ops->o_owner);
88 void class_put_type(struct obd_type *type)
91 module_put(type->typ_ops->o_owner);
94 int class_register_type(struct obd_ops *ops, struct lprocfs_vars *vars,
97 struct obd_type *type;
101 LASSERT(strnlen(name, 1024) < 1024); /* sanity check */
103 if (class_search_type(name)) {
104 CDEBUG(D_IOCTL, "Type %s already registered\n", name);
109 OBD_ALLOC(type, sizeof(*type));
113 OBD_ALLOC(type->typ_ops, sizeof(*type->typ_ops));
114 OBD_ALLOC(type->typ_name, strlen(name) + 1);
115 if (type->typ_ops == NULL || type->typ_name == NULL)
118 *(type->typ_ops) = *ops;
119 strcpy(type->typ_name, name);
122 type->typ_procroot = lprocfs_register(type->typ_name, proc_lustre_root,
125 if (IS_ERR(type->typ_procroot)) {
126 rc = PTR_ERR(type->typ_procroot);
127 type->typ_procroot = NULL;
131 spin_lock(&obd_types_lock);
132 list_add(&type->typ_chain, &obd_types);
133 spin_unlock(&obd_types_lock);
138 if (type->typ_name != NULL)
139 OBD_FREE(type->typ_name, strlen(name) + 1);
140 if (type->typ_ops != NULL)
141 OBD_FREE (type->typ_ops, sizeof (*type->typ_ops));
142 OBD_FREE(type, sizeof(*type));
146 int class_unregister_type(char *name)
148 struct obd_type *type = class_search_type(name);
152 CERROR("unknown obd type\n");
156 if (type->typ_refcnt) {
157 CERROR("type %s has refcount (%d)\n", name, type->typ_refcnt);
158 /* This is a bad situation, let's make the best of it */
159 /* Remove ops, but leave the name for debugging */
160 OBD_FREE(type->typ_ops, sizeof(*type->typ_ops));
164 if (type->typ_procroot) {
165 lprocfs_remove(type->typ_procroot);
166 type->typ_procroot = NULL;
169 spin_lock(&obd_types_lock);
170 list_del(&type->typ_chain);
171 spin_unlock(&obd_types_lock);
172 OBD_FREE(type->typ_name, strlen(name) + 1);
173 if (type->typ_ops != NULL)
174 OBD_FREE(type->typ_ops, sizeof(*type->typ_ops));
175 OBD_FREE(type, sizeof(*type));
177 } /* class_unregister_type */
179 struct obd_device *class_newdev(int *dev)
181 struct obd_device *result = NULL;
184 for (i = 0 ; i < MAX_OBD_DEVICES ; i++) {
185 struct obd_device *obd = &obd_dev[i];
186 if (!obd->obd_type) {
196 int class_name2dev(char *name)
203 for (i = 0; i < MAX_OBD_DEVICES; i++) {
204 struct obd_device *obd = &obd_dev[i];
205 if (obd->obd_name && strcmp(name, obd->obd_name) == 0)
212 struct obd_device *class_name2obd(char *name)
214 int dev = class_name2dev(name);
217 return &obd_dev[dev];
220 int class_uuid2dev(struct obd_uuid *uuid)
224 for (i = 0; i < MAX_OBD_DEVICES; i++) {
225 struct obd_device *obd = &obd_dev[i];
226 if (obd_uuid_equals(uuid, &obd->obd_uuid))
233 struct obd_device *class_uuid2obd(struct obd_uuid *uuid)
235 int dev = class_uuid2dev(uuid);
238 return &obd_dev[dev];
241 /* Search for a client OBD connected to tgt_uuid. If grp_uuid is
242 specified, then only the client with that uuid is returned,
243 otherwise any client connected to the tgt is returned. */
244 struct obd_device * class_find_client_obd(struct obd_uuid *tgt_uuid,
246 struct obd_uuid *grp_uuid)
250 for (i = 0; i < MAX_OBD_DEVICES; i++) {
251 struct obd_device *obd = &obd_dev[i];
252 if (obd->obd_type == NULL)
254 if ((strncmp(obd->obd_type->typ_name, typ_name,
255 strlen(typ_name)) == 0)) {
256 struct client_obd *cli = &obd->u.cli;
257 struct obd_import *imp = cli->cl_import;
258 if (obd_uuid_equals(tgt_uuid, &imp->imp_target_uuid) &&
259 ((grp_uuid)? obd_uuid_equals(grp_uuid,
260 &obd->obd_uuid) : 1)) {
269 /* Iterate the obd_device list looking devices have grp_uuid. Start
270 searching at *next, and if a device is found, the next index to look
271 it is saved in *next. If next is NULL, then the first matching device
272 will always be returned. */
273 struct obd_device * class_devices_in_group(struct obd_uuid *grp_uuid, int *next)
278 else if (*next >= 0 && *next < MAX_OBD_DEVICES)
283 for (; i < MAX_OBD_DEVICES; i++) {
284 struct obd_device *obd = &obd_dev[i];
285 if (obd->obd_type == NULL)
287 if (obd_uuid_equals(grp_uuid, &obd->obd_uuid)) {
298 void obd_cleanup_caches(void)
303 rc = kmem_cache_destroy(obdo_cachep);
305 CERROR("Cannot destory ll_obdo_cache\n");
309 rc = kmem_cache_destroy(import_cachep);
311 CERROR("Cannot destory ll_import_cache\n");
312 import_cachep = NULL;
317 int obd_init_caches(void)
320 LASSERT(obdo_cachep == NULL);
321 obdo_cachep = kmem_cache_create("ll_obdo_cache", sizeof(struct obdo),
326 LASSERT(import_cachep == NULL);
327 import_cachep = kmem_cache_create("ll_import_cache",
328 sizeof(struct obd_import),
335 obd_cleanup_caches();
340 /* map connection to client */
341 struct obd_export *class_conn2export(struct lustre_handle *conn)
343 struct obd_export *export;
347 CDEBUG(D_CACHE, "looking for null handle\n");
351 if (conn->cookie == -1) { /* this means assign a new connection */
352 CDEBUG(D_CACHE, "want a new connection\n");
356 CDEBUG(D_IOCTL, "looking for export cookie "LPX64"\n", conn->cookie);
357 export = class_handle2object(conn->cookie);
361 struct obd_device *class_exp2obd(struct obd_export *exp)
368 struct obd_device *class_conn2obd(struct lustre_handle *conn)
370 struct obd_export *export;
371 export = class_conn2export(conn);
373 struct obd_device *obd = export->exp_obd;
374 class_export_put(export);
380 struct obd_import *class_exp2cliimp(struct obd_export *exp)
382 struct obd_device *obd = exp->exp_obd;
385 return obd->u.cli.cl_import;
388 struct obd_import *class_conn2cliimp(struct lustre_handle *conn)
390 struct obd_device *obd = class_conn2obd(conn);
393 return obd->u.cli.cl_import;
396 /* Export management functions */
397 static void export_handle_addref(void *export)
399 class_export_get(export);
402 void __class_export_put(struct obd_export *exp)
404 if (atomic_dec_and_test(&exp->exp_refcount)) {
405 struct obd_device *obd = exp->exp_obd;
406 CDEBUG(D_IOCTL, "destroying export %p/%s\n", exp,
407 exp->exp_client_uuid.uuid);
409 LASSERT(obd != NULL);
411 /* "Local" exports (lctl, LOV->{mdc,osc}) have no connection. */
412 if (exp->exp_connection)
413 ptlrpc_put_connection_superhack(exp->exp_connection);
415 LASSERT(list_empty(&exp->exp_outstanding_replies));
416 LASSERT(list_empty(&exp->exp_handle.h_link));
417 obd_destroy_export(exp);
419 OBD_FREE(exp, sizeof(*exp));
420 if (obd->obd_set_up) {
421 atomic_dec(&obd->obd_refcount);
422 wake_up(&obd->obd_refcount_waitq);
427 /* Creates a new export, adds it to the hash table, and returns a
428 * pointer to it. The refcount is 2: one for the hash reference, and
429 * one for the pointer returned by this function. */
430 struct obd_export *class_new_export(struct obd_device *obd)
432 struct obd_export *export;
434 OBD_ALLOC(export, sizeof(*export));
436 CERROR("no memory! (minor %d)\n", obd->obd_minor);
440 export->exp_conn_cnt = 0;
441 atomic_set(&export->exp_refcount, 2);
442 export->exp_obd = obd;
443 INIT_LIST_HEAD(&export->exp_outstanding_replies);
444 /* XXX this should be in LDLM init */
445 INIT_LIST_HEAD(&export->exp_ldlm_data.led_held_locks);
447 INIT_LIST_HEAD(&export->exp_handle.h_link);
448 class_handle_hash(&export->exp_handle, export_handle_addref);
449 spin_lock_init(&export->exp_lock);
451 spin_lock(&obd->obd_dev_lock);
452 LASSERT(!obd->obd_stopping); /* shouldn't happen, but might race */
453 atomic_inc(&obd->obd_refcount);
454 list_add(&export->exp_obd_chain, &export->exp_obd->obd_exports);
455 export->exp_obd->obd_num_exports++;
456 spin_unlock(&obd->obd_dev_lock);
457 obd_init_export(export);
461 void class_unlink_export(struct obd_export *exp)
463 class_handle_unhash(&exp->exp_handle);
465 spin_lock(&exp->exp_obd->obd_dev_lock);
466 list_del_init(&exp->exp_obd_chain);
467 exp->exp_obd->obd_num_exports--;
468 spin_unlock(&exp->exp_obd->obd_dev_lock);
470 class_export_put(exp);
473 /* Import management functions */
474 static void import_handle_addref(void *import)
476 class_import_get(import);
479 struct obd_import *class_import_get(struct obd_import *import)
481 atomic_inc(&import->imp_refcount);
482 CDEBUG(D_IOCTL, "import %p refcount=%d\n", import,
483 atomic_read(&import->imp_refcount));
487 void class_import_put(struct obd_import *import)
491 CDEBUG(D_IOCTL, "import %p refcount=%d\n", import,
492 atomic_read(&import->imp_refcount) - 1);
494 LASSERT(atomic_read(&import->imp_refcount) > 0);
495 LASSERT(atomic_read(&import->imp_refcount) < 0x5a5a5a);
496 if (!atomic_dec_and_test(&import->imp_refcount)) {
501 CDEBUG(D_IOCTL, "destroying import %p\n", import);
503 ptlrpc_put_connection_superhack(import->imp_connection);
505 LASSERT(list_empty(&import->imp_handle.h_link));
506 OBD_FREE(import, sizeof(*import));
510 struct obd_import *class_new_import(void)
512 struct obd_import *imp;
514 OBD_ALLOC(imp, sizeof(*imp));
518 INIT_LIST_HEAD(&imp->imp_replay_list);
519 INIT_LIST_HEAD(&imp->imp_sending_list);
520 INIT_LIST_HEAD(&imp->imp_delayed_list);
521 spin_lock_init(&imp->imp_lock);
522 imp->imp_conn_cnt = 0;
523 imp->imp_max_transno = 0;
524 imp->imp_peer_committed_transno = 0;
525 imp->imp_state = LUSTRE_IMP_NEW;
526 init_waitqueue_head(&imp->imp_recovery_waitq);
528 atomic_set(&imp->imp_refcount, 2);
529 atomic_set(&imp->imp_replay_inflight, 0);
530 INIT_LIST_HEAD(&imp->imp_handle.h_link);
531 class_handle_hash(&imp->imp_handle, import_handle_addref);
536 void class_destroy_import(struct obd_import *import)
538 LASSERT(import != NULL);
539 LASSERT((unsigned long)import != 0x5a5a5a5a);
541 class_handle_unhash(&import->imp_handle);
543 /* Abort any inflight DLM requests and NULL out their (about to be
545 /* Invalidate all requests on import, would be better to call
546 ptlrpc_set_import_active(imp, 0); */
547 import->imp_generation++;
548 ptlrpc_abort_inflight_superhack(import);
550 class_import_put(import);
553 /* A connection defines an export context in which preallocation can
554 be managed. This releases the export pointer reference, and returns
555 the export handle, so the export refcount is 1 when this function
557 int class_connect(struct lustre_handle *conn, struct obd_device *obd,
558 struct obd_uuid *cluuid)
560 struct obd_export *export;
561 LASSERT(conn != NULL);
562 LASSERT(obd != NULL);
563 LASSERT(cluuid != NULL);
566 export = class_new_export(obd);
570 conn->cookie = export->exp_handle.h_cookie;
571 memcpy(&export->exp_client_uuid, cluuid,
572 sizeof(export->exp_client_uuid));
573 class_export_put(export);
575 CDEBUG(D_IOCTL, "connect: client %s, cookie "LPX64"\n",
576 cluuid->uuid, conn->cookie);
580 /* This function removes two references from the export: one for the
581 * hash entry and one for the export pointer passed in. The export
582 * pointer passed to this function is destroyed should not be used
584 int class_disconnect(struct obd_export *export, int flags)
588 if (export == NULL) {
590 CDEBUG(D_IOCTL, "attempting to free NULL export %p\n", export);
594 /* XXX this shouldn't have to be here, but double-disconnect will crash
595 * otherwise, and sometimes double-disconnect happens. abort_recovery,
597 if (list_empty(&export->exp_handle.h_link))
600 CDEBUG(D_IOCTL, "disconnect: cookie "LPX64"\n",
601 export->exp_handle.h_cookie);
603 class_unlink_export(export);
604 class_export_put(export);
608 void class_disconnect_exports(struct obd_device *obd, int flags)
611 struct list_head *tmp, *n, work_list;
612 struct lustre_handle fake_conn;
613 struct obd_export *fake_exp, *exp;
616 /* Move all of the exports from obd_exports to a work list, en masse. */
617 spin_lock(&obd->obd_dev_lock);
618 list_add(&work_list, &obd->obd_exports);
619 list_del_init(&obd->obd_exports);
620 spin_unlock(&obd->obd_dev_lock);
622 CDEBUG(D_HA, "OBD device %d (%p) has exports, "
623 "disconnecting them\n", obd->obd_minor, obd);
624 list_for_each_safe(tmp, n, &work_list) {
625 exp = list_entry(tmp, struct obd_export, exp_obd_chain);
626 class_export_get(exp);
628 if (obd_uuid_equals(&exp->exp_client_uuid,
629 &exp->exp_obd->obd_uuid)) {
631 "exp %p export uuid == obd uuid, don't discon\n",
633 /* Need to delete this now so we don't end up pointing
634 * to work_list later when this export is cleaned up. */
635 list_del_init(&exp->exp_obd_chain);
636 class_export_put(exp);
640 fake_conn.cookie = exp->exp_handle.h_cookie;
641 fake_exp = class_conn2export(&fake_conn);
643 class_export_put(exp);
646 rc = obd_disconnect(fake_exp, flags);
647 class_export_put(exp);
649 CDEBUG(D_HA, "disconnecting export %p failed: %d\n",
652 CDEBUG(D_HA, "export %p disconnected\n", exp);
658 int oig_init(struct obd_io_group **oig_out)
660 struct obd_io_group *oig;
663 OBD_ALLOC(oig, sizeof(*oig));
667 spin_lock_init(&oig->oig_lock);
669 oig->oig_pending = 0;
670 atomic_set(&oig->oig_refcount, 1);
671 init_waitqueue_head(&oig->oig_waitq);
672 INIT_LIST_HEAD(&oig->oig_occ_list);
678 static inline void oig_grab(struct obd_io_group *oig)
680 atomic_inc(&oig->oig_refcount);
682 void oig_release(struct obd_io_group *oig)
684 if (atomic_dec_and_test(&oig->oig_refcount))
685 OBD_FREE(oig, sizeof(*oig));
688 void oig_add_one(struct obd_io_group *oig,
689 struct oig_callback_context *occ)
692 CDEBUG(D_CACHE, "oig %p ready to roll\n", oig);
693 spin_lock_irqsave(&oig->oig_lock, flags);
696 list_add_tail(&occ->occ_oig_item, &oig->oig_occ_list);
697 spin_unlock_irqrestore(&oig->oig_lock, flags);
701 void oig_complete_one(struct obd_io_group *oig,
702 struct oig_callback_context *occ, int rc)
705 wait_queue_head_t *wake = NULL;
708 spin_lock_irqsave(&oig->oig_lock, flags);
711 list_del_init(&occ->occ_oig_item);
713 old_rc = oig->oig_rc;
714 if (oig->oig_rc == 0 && rc != 0)
717 if (--oig->oig_pending <= 0)
718 wake = &oig->oig_waitq;
720 spin_unlock_irqrestore(&oig->oig_lock, flags);
722 CDEBUG(D_CACHE, "oig %p completed, rc %d -> %d via %d, %d now "
723 "pending (racey)\n", oig, old_rc, oig->oig_rc, rc,
730 static int oig_done(struct obd_io_group *oig)
734 spin_lock_irqsave(&oig->oig_lock, flags);
735 if (oig->oig_pending <= 0)
737 spin_unlock_irqrestore(&oig->oig_lock, flags);
741 static void interrupted_oig(void *data)
743 struct obd_io_group *oig = data;
744 struct list_head *pos;
745 struct oig_callback_context *occ;
748 spin_lock_irqsave(&oig->oig_lock, flags);
749 list_for_each(pos, &oig->oig_occ_list) {
750 occ = list_entry(pos, struct oig_callback_context,
752 occ->occ_interrupted(occ);
754 spin_unlock_irqrestore(&oig->oig_lock, flags);
757 int oig_wait(struct obd_io_group *oig)
759 struct l_wait_info lwi = LWI_INTR(interrupted_oig, oig);
762 CDEBUG(D_CACHE, "waiting for oig %p\n", oig);
765 rc = l_wait_event(oig->oig_waitq, oig_done(oig), &lwi);
766 LASSERTF(rc == 0 || rc == -EINTR, "rc: %d\n", rc);
767 /* we can't continue until the oig has emptied and stopped
768 * referencing state that the caller will free upon return */
770 lwi = (struct l_wait_info){ 0, };
771 } while (rc == -EINTR);
773 LASSERTF(oig->oig_pending == 0,
774 "exiting oig_wait(oig = %p) with %d pending\n", oig,
777 CDEBUG(D_CACHE, "done waiting on oig %p rc %d\n", oig, oig->oig_rc);