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 md_ops *md_ops,
95 struct lprocfs_vars *vars, char *name)
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);
116 OBD_ALLOC(type->typ_md_ops, sizeof(*type->typ_md_ops));
117 if (type->typ_ops == NULL || type->typ_name == NULL ||
118 (md_ops && type->typ_md_ops == NULL))
121 *(type->typ_ops) = *ops;
123 *(type->typ_md_ops) = *md_ops;
125 type->typ_md_ops = NULL;
126 strcpy(type->typ_name, name);
129 type->typ_procroot = lprocfs_register(type->typ_name, proc_lustre_root,
132 if (IS_ERR(type->typ_procroot)) {
133 rc = PTR_ERR(type->typ_procroot);
134 type->typ_procroot = NULL;
138 spin_lock(&obd_types_lock);
139 list_add(&type->typ_chain, &obd_types);
140 spin_unlock(&obd_types_lock);
145 if (type->typ_name != NULL)
146 OBD_FREE(type->typ_name, strlen(name) + 1);
147 if (type->typ_ops != NULL)
148 OBD_FREE (type->typ_ops, sizeof (*type->typ_ops));
149 if (type->typ_md_ops != NULL)
150 OBD_FREE (type->typ_md_ops, sizeof (*type->typ_md_ops));
151 OBD_FREE(type, sizeof(*type));
155 int class_unregister_type(char *name)
157 struct obd_type *type = class_search_type(name);
161 CERROR("unknown obd type\n");
165 if (type->typ_refcnt) {
166 CERROR("type %s has refcount (%d)\n", name, type->typ_refcnt);
167 /* This is a bad situation, let's make the best of it */
168 /* Remove ops, but leave the name for debugging */
169 OBD_FREE(type->typ_ops, sizeof(*type->typ_ops));
173 if (type->typ_procroot) {
174 lprocfs_remove(type->typ_procroot);
175 type->typ_procroot = NULL;
178 spin_lock(&obd_types_lock);
179 list_del(&type->typ_chain);
180 spin_unlock(&obd_types_lock);
181 OBD_FREE(type->typ_name, strlen(name) + 1);
182 if (type->typ_ops != NULL)
183 OBD_FREE(type->typ_ops, sizeof(*type->typ_ops));
184 if (type->typ_md_ops != NULL)
185 OBD_FREE (type->typ_md_ops, sizeof (*type->typ_md_ops));
186 OBD_FREE(type, sizeof(*type));
188 } /* class_unregister_type */
190 struct obd_device *class_newdev(int *dev)
192 struct obd_device *result = NULL;
195 for (i = 0 ; i < MAX_OBD_DEVICES ; i++) {
196 struct obd_device *obd = &obd_dev[i];
197 if (!obd->obd_type) {
207 int class_name2dev(char *name)
214 for (i = 0; i < MAX_OBD_DEVICES; i++) {
215 struct obd_device *obd = &obd_dev[i];
216 if (obd->obd_name && strcmp(name, obd->obd_name) == 0)
223 struct obd_device *class_name2obd(char *name)
225 int dev = class_name2dev(name);
228 return &obd_dev[dev];
231 int class_uuid2dev(struct obd_uuid *uuid)
235 for (i = 0; i < MAX_OBD_DEVICES; i++) {
236 struct obd_device *obd = &obd_dev[i];
237 if (obd_uuid_equals(uuid, &obd->obd_uuid))
244 struct obd_device *class_uuid2obd(struct obd_uuid *uuid)
246 int dev = class_uuid2dev(uuid);
249 return &obd_dev[dev];
252 /* Search for a client OBD connected to tgt_uuid. If grp_uuid is
253 specified, then only the client with that uuid is returned,
254 otherwise any client connected to the tgt is returned.
255 If tgt_uuid is NULL, the lov with grp_uuid is returned. */
256 struct obd_device * class_find_client_obd(struct obd_uuid *tgt_uuid,
258 struct obd_uuid *grp_uuid)
262 for (i = 0; i < MAX_OBD_DEVICES; i++) {
263 struct obd_device *obd = &obd_dev[i];
264 if (obd->obd_type == NULL)
266 if ((strncmp(obd->obd_type->typ_name, typ_name,
267 strlen(typ_name)) == 0)) {
268 struct client_obd *cli = &obd->u.cli;
269 struct obd_import *imp = cli->cl_import;
270 if (tgt_uuid == NULL) {
272 if (obd_uuid_equals(grp_uuid, &obd->obd_uuid))
276 if (obd_uuid_equals(tgt_uuid, &imp->imp_target_uuid) &&
277 ((grp_uuid)? obd_uuid_equals(grp_uuid,
278 &obd->obd_uuid) : 1)) {
287 /* Iterate the obd_device list looking devices have grp_uuid. Start
288 searching at *next, and if a device is found, the next index to look
289 it is saved in *next. If next is NULL, then the first matching device
290 will always be returned. */
291 struct obd_device * class_devices_in_group(struct obd_uuid *grp_uuid, int *next)
296 else if (*next >= 0 && *next < MAX_OBD_DEVICES)
301 for (; i < MAX_OBD_DEVICES; i++) {
302 struct obd_device *obd = &obd_dev[i];
303 if (obd->obd_type == NULL)
305 if (obd_uuid_equals(grp_uuid, &obd->obd_uuid)) {
316 void obd_cleanup_caches(void)
321 rc = kmem_cache_destroy(obdo_cachep);
323 CERROR("Cannot destory ll_obdo_cache\n");
327 rc = kmem_cache_destroy(import_cachep);
329 CERROR("Cannot destory ll_import_cache\n");
330 import_cachep = NULL;
335 int obd_init_caches(void)
338 LASSERT(obdo_cachep == NULL);
339 obdo_cachep = kmem_cache_create("ll_obdo_cache", sizeof(struct obdo),
344 LASSERT(import_cachep == NULL);
345 import_cachep = kmem_cache_create("ll_import_cache",
346 sizeof(struct obd_import),
353 obd_cleanup_caches();
358 /* map connection to client */
359 struct obd_export *class_conn2export(struct lustre_handle *conn)
361 struct obd_export *export;
365 CDEBUG(D_CACHE, "looking for null handle\n");
369 if (conn->cookie == -1) { /* this means assign a new connection */
370 CDEBUG(D_CACHE, "want a new connection\n");
374 CDEBUG(D_IOCTL, "looking for export cookie "LPX64"\n", conn->cookie);
375 export = class_handle2object(conn->cookie);
379 struct obd_device *class_exp2obd(struct obd_export *exp)
386 struct obd_device *class_conn2obd(struct lustre_handle *conn)
388 struct obd_export *export;
389 export = class_conn2export(conn);
391 struct obd_device *obd = export->exp_obd;
392 class_export_put(export);
398 struct obd_import *class_exp2cliimp(struct obd_export *exp)
400 struct obd_device *obd = exp->exp_obd;
403 return obd->u.cli.cl_import;
406 struct obd_import *class_conn2cliimp(struct lustre_handle *conn)
408 struct obd_device *obd = class_conn2obd(conn);
411 return obd->u.cli.cl_import;
414 /* Export management functions */
415 static void export_handle_addref(void *export)
417 class_export_get(export);
420 void __class_export_put(struct obd_export *exp)
422 if (atomic_dec_and_test(&exp->exp_refcount)) {
423 struct obd_device *obd = exp->exp_obd;
424 CDEBUG(D_IOCTL, "destroying export %p/%s\n", exp,
425 exp->exp_client_uuid.uuid);
427 LASSERT(obd != NULL);
429 /* "Local" exports (lctl, LOV->{mdc,osc}) have no connection. */
430 if (exp->exp_connection)
431 ptlrpc_put_connection_superhack(exp->exp_connection);
433 LASSERT(list_empty(&exp->exp_outstanding_replies));
434 LASSERT(list_empty(&exp->exp_handle.h_link));
435 obd_destroy_export(exp);
437 OBD_FREE(exp, sizeof(*exp));
438 if (obd->obd_set_up) {
439 atomic_dec(&obd->obd_refcount);
440 wake_up(&obd->obd_refcount_waitq);
442 CERROR("removing export %p from obd %s (%p) -- OBD "
443 "not set up (refcount = %d)\n", exp,
445 atomic_read(&obd->obd_refcount));
450 /* Creates a new export, adds it to the hash table, and returns a
451 * pointer to it. The refcount is 2: one for the hash reference, and
452 * one for the pointer returned by this function. */
453 struct obd_export *class_new_export(struct obd_device *obd)
455 struct obd_export *export;
457 OBD_ALLOC(export, sizeof(*export));
459 CERROR("no memory! (minor %d)\n", obd->obd_minor);
463 export->exp_conn_cnt = 0;
464 atomic_set(&export->exp_refcount, 2);
465 atomic_set(&export->exp_rpc_count, 0);
466 export->exp_obd = obd;
467 INIT_LIST_HEAD(&export->exp_outstanding_replies);
468 /* XXX this should be in LDLM init */
469 INIT_LIST_HEAD(&export->exp_ldlm_data.led_held_locks);
471 INIT_LIST_HEAD(&export->exp_handle.h_link);
472 class_handle_hash(&export->exp_handle, export_handle_addref);
473 spin_lock_init(&export->exp_lock);
475 spin_lock(&obd->obd_dev_lock);
476 LASSERT(!obd->obd_stopping); /* shouldn't happen, but might race */
477 atomic_inc(&obd->obd_refcount);
478 list_add(&export->exp_obd_chain, &export->exp_obd->obd_exports);
479 export->exp_obd->obd_num_exports++;
480 spin_unlock(&obd->obd_dev_lock);
481 obd_init_export(export);
485 void class_unlink_export(struct obd_export *exp)
487 class_handle_unhash(&exp->exp_handle);
489 spin_lock(&exp->exp_obd->obd_dev_lock);
490 list_del_init(&exp->exp_obd_chain);
491 exp->exp_obd->obd_num_exports--;
492 spin_unlock(&exp->exp_obd->obd_dev_lock);
494 class_export_put(exp);
497 /* Import management functions */
498 static void import_handle_addref(void *import)
500 class_import_get(import);
503 struct obd_import *class_import_get(struct obd_import *import)
505 atomic_inc(&import->imp_refcount);
506 CDEBUG(D_IOCTL, "import %p refcount=%d\n", import,
507 atomic_read(&import->imp_refcount));
511 void class_import_put(struct obd_import *import)
515 CDEBUG(D_IOCTL, "import %p refcount=%d\n", import,
516 atomic_read(&import->imp_refcount) - 1);
518 LASSERT(atomic_read(&import->imp_refcount) > 0);
519 LASSERT(atomic_read(&import->imp_refcount) < 0x5a5a5a);
520 if (!atomic_dec_and_test(&import->imp_refcount)) {
525 CDEBUG(D_IOCTL, "destroying import %p\n", import);
527 ptlrpc_put_connection_superhack(import->imp_connection);
529 LASSERT(list_empty(&import->imp_handle.h_link));
530 OBD_FREE(import, sizeof(*import));
534 struct obd_import *class_new_import(void)
536 struct obd_import *imp;
538 OBD_ALLOC(imp, sizeof(*imp));
542 INIT_LIST_HEAD(&imp->imp_replay_list);
543 INIT_LIST_HEAD(&imp->imp_sending_list);
544 INIT_LIST_HEAD(&imp->imp_delayed_list);
545 spin_lock_init(&imp->imp_lock);
546 imp->imp_conn_cnt = 0;
547 imp->imp_max_transno = 0;
548 imp->imp_peer_committed_transno = 0;
549 imp->imp_state = LUSTRE_IMP_NEW;
550 init_waitqueue_head(&imp->imp_recovery_waitq);
552 atomic_set(&imp->imp_refcount, 2);
553 atomic_set(&imp->imp_inflight, 0);
554 atomic_set(&imp->imp_replay_inflight, 0);
555 INIT_LIST_HEAD(&imp->imp_handle.h_link);
556 class_handle_hash(&imp->imp_handle, import_handle_addref);
561 void class_destroy_import(struct obd_import *import)
563 LASSERT(import != NULL);
564 LASSERT(import != LP_POISON);
566 class_handle_unhash(&import->imp_handle);
568 /* Abort any inflight DLM requests and NULL out their (about to be
570 /* Invalidate all requests on import, would be better to call
571 ptlrpc_set_import_active(imp, 0); */
572 import->imp_generation++;
573 ptlrpc_abort_inflight_superhack(import);
575 class_import_put(import);
578 /* A connection defines an export context in which preallocation can
579 be managed. This releases the export pointer reference, and returns
580 the export handle, so the export refcount is 1 when this function
582 int class_connect(struct lustre_handle *conn, struct obd_device *obd,
583 struct obd_uuid *cluuid)
585 struct obd_export *export;
586 LASSERT(conn != NULL);
587 LASSERT(obd != NULL);
588 LASSERT(cluuid != NULL);
591 export = class_new_export(obd);
595 conn->cookie = export->exp_handle.h_cookie;
596 memcpy(&export->exp_client_uuid, cluuid,
597 sizeof(export->exp_client_uuid));
598 class_export_put(export);
600 CDEBUG(D_IOCTL, "connect: client %s, cookie "LPX64"\n",
601 cluuid->uuid, conn->cookie);
605 /* This function removes two references from the export: one for the
606 * hash entry and one for the export pointer passed in. The export
607 * pointer passed to this function is destroyed should not be used
609 int class_disconnect(struct obd_export *export, int flags)
613 if (export == NULL) {
615 CDEBUG(D_IOCTL, "attempting to free NULL export %p\n", export);
619 /* XXX this shouldn't have to be here, but double-disconnect will crash
620 * otherwise, and sometimes double-disconnect happens. abort_recovery,
622 if (list_empty(&export->exp_handle.h_link))
625 CDEBUG(D_IOCTL, "disconnect: cookie "LPX64"\n",
626 export->exp_handle.h_cookie);
628 class_unlink_export(export);
629 class_export_put(export);
633 void class_disconnect_exports(struct obd_device *obd, int flags)
636 struct list_head *tmp, *n, work_list;
637 struct lustre_handle fake_conn;
638 struct obd_export *fake_exp, *exp;
641 /* Move all of the exports from obd_exports to a work list, en masse. */
642 spin_lock(&obd->obd_dev_lock);
643 list_add(&work_list, &obd->obd_exports);
644 list_del_init(&obd->obd_exports);
645 spin_unlock(&obd->obd_dev_lock);
647 CDEBUG(D_HA, "OBD device %d (%p) has exports, "
648 "disconnecting them\n", obd->obd_minor, obd);
649 list_for_each_safe(tmp, n, &work_list) {
650 exp = list_entry(tmp, struct obd_export, exp_obd_chain);
651 class_export_get(exp);
653 if (obd_uuid_equals(&exp->exp_client_uuid,
654 &exp->exp_obd->obd_uuid)) {
656 "exp %p export uuid == obd uuid, don't discon\n",
658 /* Need to delete this now so we don't end up pointing
659 * to work_list later when this export is cleaned up. */
660 list_del_init(&exp->exp_obd_chain);
661 class_export_put(exp);
665 fake_conn.cookie = exp->exp_handle.h_cookie;
666 fake_exp = class_conn2export(&fake_conn);
668 class_export_put(exp);
671 rc = obd_disconnect(fake_exp, flags);
672 class_export_put(exp);
674 CDEBUG(D_HA, "disconnecting export %p failed: %d\n",
677 CDEBUG(D_HA, "export %p disconnected\n", exp);
683 int oig_init(struct obd_io_group **oig_out)
685 struct obd_io_group *oig;
688 OBD_ALLOC(oig, sizeof(*oig));
692 spin_lock_init(&oig->oig_lock);
694 oig->oig_pending = 0;
695 atomic_set(&oig->oig_refcount, 1);
696 init_waitqueue_head(&oig->oig_waitq);
697 INIT_LIST_HEAD(&oig->oig_occ_list);
703 static inline void oig_grab(struct obd_io_group *oig)
705 atomic_inc(&oig->oig_refcount);
707 void oig_release(struct obd_io_group *oig)
709 if (atomic_dec_and_test(&oig->oig_refcount))
710 OBD_FREE(oig, sizeof(*oig));
713 void oig_add_one(struct obd_io_group *oig,
714 struct oig_callback_context *occ)
717 CDEBUG(D_CACHE, "oig %p ready to roll\n", oig);
718 spin_lock_irqsave(&oig->oig_lock, flags);
721 list_add_tail(&occ->occ_oig_item, &oig->oig_occ_list);
722 spin_unlock_irqrestore(&oig->oig_lock, flags);
726 void oig_complete_one(struct obd_io_group *oig,
727 struct oig_callback_context *occ, int rc)
730 wait_queue_head_t *wake = NULL;
733 spin_lock_irqsave(&oig->oig_lock, flags);
736 list_del_init(&occ->occ_oig_item);
738 old_rc = oig->oig_rc;
739 if (oig->oig_rc == 0 && rc != 0)
742 if (--oig->oig_pending <= 0)
743 wake = &oig->oig_waitq;
745 spin_unlock_irqrestore(&oig->oig_lock, flags);
747 CDEBUG(D_CACHE, "oig %p completed, rc %d -> %d via %d, %d now "
748 "pending (racey)\n", oig, old_rc, oig->oig_rc, rc,
755 static int oig_done(struct obd_io_group *oig)
759 spin_lock_irqsave(&oig->oig_lock, flags);
760 if (oig->oig_pending <= 0)
762 spin_unlock_irqrestore(&oig->oig_lock, flags);
766 static void interrupted_oig(void *data)
768 struct obd_io_group *oig = data;
769 struct list_head *pos;
770 struct oig_callback_context *occ;
773 spin_lock_irqsave(&oig->oig_lock, flags);
774 list_for_each(pos, &oig->oig_occ_list) {
775 occ = list_entry(pos, struct oig_callback_context,
777 occ->occ_interrupted(occ);
779 spin_unlock_irqrestore(&oig->oig_lock, flags);
782 int oig_wait(struct obd_io_group *oig)
784 struct l_wait_info lwi = LWI_INTR(interrupted_oig, oig);
787 CDEBUG(D_CACHE, "waiting for oig %p\n", oig);
790 rc = l_wait_event(oig->oig_waitq, oig_done(oig), &lwi);
791 LASSERTF(rc == 0 || rc == -EINTR, "rc: %d\n", rc);
792 /* we can't continue until the oig has emptied and stopped
793 * referencing state that the caller will free upon return */
795 lwi = (struct l_wait_info){ 0, };
796 } while (rc == -EINTR);
798 LASSERTF(oig->oig_pending == 0,
799 "exiting oig_wait(oig = %p) with %d pending\n", oig,
802 CDEBUG(D_CACHE, "done waiting on oig %p rc %d\n", oig, oig->oig_rc);