]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blob - drivers/staging/lustre/lustre/ldlm/ldlm_lib.c
Merge branch 'turbostat' of git://git.kernel.org/pub/scm/linux/kernel/git/lenb/linux...
[mirror_ubuntu-jammy-kernel.git] / drivers / staging / lustre / lustre / ldlm / ldlm_lib.c
1 /*
2 * GPL HEADER START
3 *
4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5 *
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.
9 *
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).
15 *
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
19 *
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
22 * have any questions.
23 *
24 * GPL HEADER END
25 */
26 /*
27 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
29 *
30 * Copyright (c) 2010, 2015, Intel Corporation.
31 */
32 /*
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
35 */
36
37 /**
38 * This file deals with various client/target related logic including recovery.
39 *
40 * TODO: This code more logically belongs in the ptlrpc module than in ldlm and
41 * should be moved.
42 */
43
44 #define DEBUG_SUBSYSTEM S_LDLM
45
46 #include "../../include/linux/libcfs/libcfs.h"
47 #include "../include/obd.h"
48 #include "../include/obd_class.h"
49 #include "../include/lustre_dlm.h"
50 #include "../include/lustre_net.h"
51 #include "../include/lustre_sec.h"
52 #include "ldlm_internal.h"
53
54 /* @priority: If non-zero, move the selected connection to the list head.
55 * @create: If zero, only search in existing connections.
56 */
57 static int import_set_conn(struct obd_import *imp, struct obd_uuid *uuid,
58 int priority, int create)
59 {
60 struct ptlrpc_connection *ptlrpc_conn;
61 struct obd_import_conn *imp_conn = NULL, *item;
62 int rc = 0;
63
64 if (!create && !priority) {
65 CDEBUG(D_HA, "Nothing to do\n");
66 return -EINVAL;
67 }
68
69 ptlrpc_conn = ptlrpc_uuid_to_connection(uuid);
70 if (!ptlrpc_conn) {
71 CDEBUG(D_HA, "can't find connection %s\n", uuid->uuid);
72 return -ENOENT;
73 }
74
75 if (create) {
76 imp_conn = kzalloc(sizeof(*imp_conn), GFP_NOFS);
77 if (!imp_conn) {
78 rc = -ENOMEM;
79 goto out_put;
80 }
81 }
82
83 spin_lock(&imp->imp_lock);
84 list_for_each_entry(item, &imp->imp_conn_list, oic_item) {
85 if (obd_uuid_equals(uuid, &item->oic_uuid)) {
86 if (priority) {
87 list_del(&item->oic_item);
88 list_add(&item->oic_item,
89 &imp->imp_conn_list);
90 item->oic_last_attempt = 0;
91 }
92 CDEBUG(D_HA, "imp %p@%s: found existing conn %s%s\n",
93 imp, imp->imp_obd->obd_name, uuid->uuid,
94 (priority ? ", moved to head" : ""));
95 spin_unlock(&imp->imp_lock);
96 rc = 0;
97 goto out_free;
98 }
99 }
100 /* No existing import connection found for \a uuid. */
101 if (create) {
102 imp_conn->oic_conn = ptlrpc_conn;
103 imp_conn->oic_uuid = *uuid;
104 imp_conn->oic_last_attempt = 0;
105 if (priority)
106 list_add(&imp_conn->oic_item, &imp->imp_conn_list);
107 else
108 list_add_tail(&imp_conn->oic_item,
109 &imp->imp_conn_list);
110 CDEBUG(D_HA, "imp %p@%s: add connection %s at %s\n",
111 imp, imp->imp_obd->obd_name, uuid->uuid,
112 (priority ? "head" : "tail"));
113 } else {
114 spin_unlock(&imp->imp_lock);
115 rc = -ENOENT;
116 goto out_free;
117 }
118
119 spin_unlock(&imp->imp_lock);
120 return 0;
121 out_free:
122 kfree(imp_conn);
123 out_put:
124 ptlrpc_connection_put(ptlrpc_conn);
125 return rc;
126 }
127
128 int import_set_conn_priority(struct obd_import *imp, struct obd_uuid *uuid)
129 {
130 return import_set_conn(imp, uuid, 1, 0);
131 }
132
133 int client_import_add_conn(struct obd_import *imp, struct obd_uuid *uuid,
134 int priority)
135 {
136 return import_set_conn(imp, uuid, priority, 1);
137 }
138 EXPORT_SYMBOL(client_import_add_conn);
139
140 int client_import_del_conn(struct obd_import *imp, struct obd_uuid *uuid)
141 {
142 struct obd_import_conn *imp_conn;
143 struct obd_export *dlmexp;
144 int rc = -ENOENT;
145
146 spin_lock(&imp->imp_lock);
147 if (list_empty(&imp->imp_conn_list)) {
148 LASSERT(!imp->imp_connection);
149 goto out;
150 }
151
152 list_for_each_entry(imp_conn, &imp->imp_conn_list, oic_item) {
153 if (!obd_uuid_equals(uuid, &imp_conn->oic_uuid))
154 continue;
155 LASSERT(imp_conn->oic_conn);
156
157 if (imp_conn == imp->imp_conn_current) {
158 LASSERT(imp_conn->oic_conn == imp->imp_connection);
159
160 if (imp->imp_state != LUSTRE_IMP_CLOSED &&
161 imp->imp_state != LUSTRE_IMP_DISCON) {
162 CERROR("can't remove current connection\n");
163 rc = -EBUSY;
164 goto out;
165 }
166
167 ptlrpc_connection_put(imp->imp_connection);
168 imp->imp_connection = NULL;
169
170 dlmexp = class_conn2export(&imp->imp_dlm_handle);
171 if (dlmexp && dlmexp->exp_connection) {
172 LASSERT(dlmexp->exp_connection ==
173 imp_conn->oic_conn);
174 ptlrpc_connection_put(dlmexp->exp_connection);
175 dlmexp->exp_connection = NULL;
176 }
177 }
178
179 list_del(&imp_conn->oic_item);
180 ptlrpc_connection_put(imp_conn->oic_conn);
181 kfree(imp_conn);
182 CDEBUG(D_HA, "imp %p@%s: remove connection %s\n",
183 imp, imp->imp_obd->obd_name, uuid->uuid);
184 rc = 0;
185 break;
186 }
187 out:
188 spin_unlock(&imp->imp_lock);
189 if (rc == -ENOENT)
190 CERROR("connection %s not found\n", uuid->uuid);
191 return rc;
192 }
193 EXPORT_SYMBOL(client_import_del_conn);
194
195 /**
196 * Find conn UUID by peer NID. \a peer is a server NID. This function is used
197 * to find a conn uuid of \a imp which can reach \a peer.
198 */
199 int client_import_find_conn(struct obd_import *imp, lnet_nid_t peer,
200 struct obd_uuid *uuid)
201 {
202 struct obd_import_conn *conn;
203 int rc = -ENOENT;
204
205 spin_lock(&imp->imp_lock);
206 list_for_each_entry(conn, &imp->imp_conn_list, oic_item) {
207 /* Check if conn UUID does have this peer NID. */
208 if (class_check_uuid(&conn->oic_uuid, peer)) {
209 *uuid = conn->oic_uuid;
210 rc = 0;
211 break;
212 }
213 }
214 spin_unlock(&imp->imp_lock);
215 return rc;
216 }
217 EXPORT_SYMBOL(client_import_find_conn);
218
219 void client_destroy_import(struct obd_import *imp)
220 {
221 /* Drop security policy instance after all RPCs have finished/aborted
222 * to let all busy contexts be released. */
223 class_import_get(imp);
224 class_destroy_import(imp);
225 sptlrpc_import_sec_put(imp);
226 class_import_put(imp);
227 }
228 EXPORT_SYMBOL(client_destroy_import);
229
230 /**
231 * Check whether or not the OSC is on MDT.
232 * In the config log,
233 * osc on MDT
234 * setup 0:{fsname}-OSTxxxx-osc[-MDTxxxx] 1:lustre-OST0000_UUID 2:NID
235 * osc on client
236 * setup 0:{fsname}-OSTxxxx-osc 1:lustre-OST0000_UUID 2:NID
237 *
238 **/
239 static int osc_on_mdt(char *obdname)
240 {
241 char *ptr;
242
243 ptr = strrchr(obdname, '-');
244 if (ptr == NULL)
245 return 0;
246
247 if (strncmp(ptr + 1, "MDT", 3) == 0)
248 return 1;
249
250 return 0;
251 }
252
253 /* Configure an RPC client OBD device.
254 *
255 * lcfg parameters:
256 * 1 - client UUID
257 * 2 - server UUID
258 * 3 - inactive-on-startup
259 */
260 int client_obd_setup(struct obd_device *obddev, struct lustre_cfg *lcfg)
261 {
262 struct client_obd *cli = &obddev->u.cli;
263 struct obd_import *imp;
264 struct obd_uuid server_uuid;
265 int rq_portal, rp_portal, connect_op;
266 char *name = obddev->obd_type->typ_name;
267 ldlm_ns_type_t ns_type = LDLM_NS_TYPE_UNKNOWN;
268 int rc;
269
270 /* In a more perfect world, we would hang a ptlrpc_client off of
271 * obd_type and just use the values from there. */
272 if (!strcmp(name, LUSTRE_OSC_NAME)) {
273 rq_portal = OST_REQUEST_PORTAL;
274 rp_portal = OSC_REPLY_PORTAL;
275 connect_op = OST_CONNECT;
276 cli->cl_sp_me = LUSTRE_SP_CLI;
277 cli->cl_sp_to = LUSTRE_SP_OST;
278 ns_type = LDLM_NS_TYPE_OSC;
279 } else if (!strcmp(name, LUSTRE_MDC_NAME) ||
280 !strcmp(name, LUSTRE_LWP_NAME)) {
281 rq_portal = MDS_REQUEST_PORTAL;
282 rp_portal = MDC_REPLY_PORTAL;
283 connect_op = MDS_CONNECT;
284 cli->cl_sp_me = LUSTRE_SP_CLI;
285 cli->cl_sp_to = LUSTRE_SP_MDT;
286 ns_type = LDLM_NS_TYPE_MDC;
287 } else if (!strcmp(name, LUSTRE_OSP_NAME)) {
288 if (strstr(lustre_cfg_buf(lcfg, 1), "OST") == NULL) {
289 /* OSP_on_MDT for other MDTs */
290 connect_op = MDS_CONNECT;
291 cli->cl_sp_to = LUSTRE_SP_MDT;
292 ns_type = LDLM_NS_TYPE_MDC;
293 rq_portal = OUT_PORTAL;
294 } else {
295 /* OSP on MDT for OST */
296 connect_op = OST_CONNECT;
297 cli->cl_sp_to = LUSTRE_SP_OST;
298 ns_type = LDLM_NS_TYPE_OSC;
299 rq_portal = OST_REQUEST_PORTAL;
300 }
301 rp_portal = OSC_REPLY_PORTAL;
302 cli->cl_sp_me = LUSTRE_SP_CLI;
303 } else if (!strcmp(name, LUSTRE_MGC_NAME)) {
304 rq_portal = MGS_REQUEST_PORTAL;
305 rp_portal = MGC_REPLY_PORTAL;
306 connect_op = MGS_CONNECT;
307 cli->cl_sp_me = LUSTRE_SP_MGC;
308 cli->cl_sp_to = LUSTRE_SP_MGS;
309 cli->cl_flvr_mgc.sf_rpc = SPTLRPC_FLVR_INVALID;
310 ns_type = LDLM_NS_TYPE_MGC;
311 } else {
312 CERROR("unknown client OBD type \"%s\", can't setup\n",
313 name);
314 return -EINVAL;
315 }
316
317 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
318 CERROR("requires a TARGET UUID\n");
319 return -EINVAL;
320 }
321
322 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > 37) {
323 CERROR("client UUID must be less than 38 characters\n");
324 return -EINVAL;
325 }
326
327 if (LUSTRE_CFG_BUFLEN(lcfg, 2) < 1) {
328 CERROR("setup requires a SERVER UUID\n");
329 return -EINVAL;
330 }
331
332 if (LUSTRE_CFG_BUFLEN(lcfg, 2) > 37) {
333 CERROR("target UUID must be less than 38 characters\n");
334 return -EINVAL;
335 }
336
337 init_rwsem(&cli->cl_sem);
338 cli->cl_conn_count = 0;
339 memcpy(server_uuid.uuid, lustre_cfg_buf(lcfg, 2),
340 min_t(unsigned int, LUSTRE_CFG_BUFLEN(lcfg, 2),
341 sizeof(server_uuid)));
342
343 cli->cl_dirty = 0;
344 cli->cl_avail_grant = 0;
345 /* FIXME: Should limit this for the sum of all cl_dirty_max. */
346 cli->cl_dirty_max = OSC_MAX_DIRTY_DEFAULT * 1024 * 1024;
347 if (cli->cl_dirty_max >> PAGE_CACHE_SHIFT > totalram_pages / 8)
348 cli->cl_dirty_max = totalram_pages << (PAGE_CACHE_SHIFT - 3);
349 INIT_LIST_HEAD(&cli->cl_cache_waiters);
350 INIT_LIST_HEAD(&cli->cl_loi_ready_list);
351 INIT_LIST_HEAD(&cli->cl_loi_hp_ready_list);
352 INIT_LIST_HEAD(&cli->cl_loi_write_list);
353 INIT_LIST_HEAD(&cli->cl_loi_read_list);
354 client_obd_list_lock_init(&cli->cl_loi_list_lock);
355 atomic_set(&cli->cl_pending_w_pages, 0);
356 atomic_set(&cli->cl_pending_r_pages, 0);
357 cli->cl_r_in_flight = 0;
358 cli->cl_w_in_flight = 0;
359
360 spin_lock_init(&cli->cl_read_rpc_hist.oh_lock);
361 spin_lock_init(&cli->cl_write_rpc_hist.oh_lock);
362 spin_lock_init(&cli->cl_read_page_hist.oh_lock);
363 spin_lock_init(&cli->cl_write_page_hist.oh_lock);
364 spin_lock_init(&cli->cl_read_offset_hist.oh_lock);
365 spin_lock_init(&cli->cl_write_offset_hist.oh_lock);
366
367 /* lru for osc. */
368 INIT_LIST_HEAD(&cli->cl_lru_osc);
369 atomic_set(&cli->cl_lru_shrinkers, 0);
370 atomic_set(&cli->cl_lru_busy, 0);
371 atomic_set(&cli->cl_lru_in_list, 0);
372 INIT_LIST_HEAD(&cli->cl_lru_list);
373 client_obd_list_lock_init(&cli->cl_lru_list_lock);
374
375 init_waitqueue_head(&cli->cl_destroy_waitq);
376 atomic_set(&cli->cl_destroy_in_flight, 0);
377 /* Turn on checksumming by default. */
378 cli->cl_checksum = 1;
379 /*
380 * The supported checksum types will be worked out at connect time
381 * Set cl_chksum* to CRC32 for now to avoid returning screwed info
382 * through procfs.
383 */
384 cli->cl_cksum_type = cli->cl_supp_cksum_types = OBD_CKSUM_CRC32;
385 atomic_set(&cli->cl_resends, OSC_DEFAULT_RESENDS);
386
387 /* This value may be reduced at connect time in
388 * ptlrpc_connect_interpret() . We initialize it to only
389 * 1MB until we know what the performance looks like.
390 * In the future this should likely be increased. LU-1431 */
391 cli->cl_max_pages_per_rpc = min_t(int, PTLRPC_MAX_BRW_PAGES,
392 LNET_MTU >> PAGE_CACHE_SHIFT);
393
394 if (!strcmp(name, LUSTRE_MDC_NAME)) {
395 cli->cl_max_rpcs_in_flight = MDC_MAX_RIF_DEFAULT;
396 } else if (totalram_pages >> (20 - PAGE_CACHE_SHIFT) <= 128 /* MB */) {
397 cli->cl_max_rpcs_in_flight = 2;
398 } else if (totalram_pages >> (20 - PAGE_CACHE_SHIFT) <= 256 /* MB */) {
399 cli->cl_max_rpcs_in_flight = 3;
400 } else if (totalram_pages >> (20 - PAGE_CACHE_SHIFT) <= 512 /* MB */) {
401 cli->cl_max_rpcs_in_flight = 4;
402 } else {
403 if (osc_on_mdt(obddev->obd_name))
404 cli->cl_max_rpcs_in_flight = MDS_OSC_MAX_RIF_DEFAULT;
405 else
406 cli->cl_max_rpcs_in_flight = OSC_MAX_RIF_DEFAULT;
407 }
408 rc = ldlm_get_ref();
409 if (rc) {
410 CERROR("ldlm_get_ref failed: %d\n", rc);
411 goto err;
412 }
413
414 ptlrpc_init_client(rq_portal, rp_portal, name,
415 &obddev->obd_ldlm_client);
416
417 imp = class_new_import(obddev);
418 if (imp == NULL) {
419 rc = -ENOENT;
420 goto err_ldlm;
421 }
422 imp->imp_client = &obddev->obd_ldlm_client;
423 imp->imp_connect_op = connect_op;
424 memcpy(cli->cl_target_uuid.uuid, lustre_cfg_buf(lcfg, 1),
425 LUSTRE_CFG_BUFLEN(lcfg, 1));
426 class_import_put(imp);
427
428 rc = client_import_add_conn(imp, &server_uuid, 1);
429 if (rc) {
430 CERROR("can't add initial connection\n");
431 goto err_import;
432 }
433
434 cli->cl_import = imp;
435 /* cli->cl_max_mds_{easize,cookiesize} updated by mdc_init_ea_size() */
436 cli->cl_max_mds_easize = sizeof(struct lov_mds_md_v3);
437 cli->cl_max_mds_cookiesize = sizeof(struct llog_cookie);
438
439 if (LUSTRE_CFG_BUFLEN(lcfg, 3) > 0) {
440 if (!strcmp(lustre_cfg_string(lcfg, 3), "inactive")) {
441 CDEBUG(D_HA, "marking %s %s->%s as inactive\n",
442 name, obddev->obd_name,
443 cli->cl_target_uuid.uuid);
444 spin_lock(&imp->imp_lock);
445 imp->imp_deactive = 1;
446 spin_unlock(&imp->imp_lock);
447 }
448 }
449
450 obddev->obd_namespace = ldlm_namespace_new(obddev, obddev->obd_name,
451 LDLM_NAMESPACE_CLIENT,
452 LDLM_NAMESPACE_GREEDY,
453 ns_type);
454 if (obddev->obd_namespace == NULL) {
455 CERROR("Unable to create client namespace - %s\n",
456 obddev->obd_name);
457 rc = -ENOMEM;
458 goto err_import;
459 }
460
461 cli->cl_qchk_stat = CL_NOT_QUOTACHECKED;
462
463 return rc;
464
465 err_import:
466 class_destroy_import(imp);
467 err_ldlm:
468 ldlm_put_ref();
469 err:
470 return rc;
471
472 }
473 EXPORT_SYMBOL(client_obd_setup);
474
475 int client_obd_cleanup(struct obd_device *obddev)
476 {
477 ldlm_namespace_free_post(obddev->obd_namespace);
478 obddev->obd_namespace = NULL;
479
480 LASSERT(obddev->u.cli.cl_import == NULL);
481
482 ldlm_put_ref();
483 return 0;
484 }
485 EXPORT_SYMBOL(client_obd_cleanup);
486
487 /* ->o_connect() method for client side (OSC and MDC and MGC) */
488 int client_connect_import(const struct lu_env *env,
489 struct obd_export **exp,
490 struct obd_device *obd, struct obd_uuid *cluuid,
491 struct obd_connect_data *data, void *localdata)
492 {
493 struct client_obd *cli = &obd->u.cli;
494 struct obd_import *imp = cli->cl_import;
495 struct obd_connect_data *ocd;
496 struct lustre_handle conn = { 0 };
497 int rc;
498
499 *exp = NULL;
500 down_write(&cli->cl_sem);
501 if (cli->cl_conn_count > 0) {
502 rc = -EALREADY;
503 goto out_sem;
504 }
505
506 rc = class_connect(&conn, obd, cluuid);
507 if (rc)
508 goto out_sem;
509
510 cli->cl_conn_count++;
511 *exp = class_conn2export(&conn);
512
513 LASSERT(obd->obd_namespace);
514
515 imp->imp_dlm_handle = conn;
516 rc = ptlrpc_init_import(imp);
517 if (rc != 0)
518 goto out_ldlm;
519
520 ocd = &imp->imp_connect_data;
521 if (data) {
522 *ocd = *data;
523 imp->imp_connect_flags_orig = data->ocd_connect_flags;
524 }
525
526 rc = ptlrpc_connect_import(imp);
527 if (rc != 0) {
528 LASSERT(imp->imp_state == LUSTRE_IMP_DISCON);
529 goto out_ldlm;
530 }
531 LASSERT(*exp != NULL && (*exp)->exp_connection);
532
533 if (data) {
534 LASSERTF((ocd->ocd_connect_flags & data->ocd_connect_flags) ==
535 ocd->ocd_connect_flags, "old %#llx, new %#llx\n",
536 data->ocd_connect_flags, ocd->ocd_connect_flags);
537 data->ocd_connect_flags = ocd->ocd_connect_flags;
538 }
539
540 ptlrpc_pinger_add_import(imp);
541
542 if (rc) {
543 out_ldlm:
544 cli->cl_conn_count--;
545 class_disconnect(*exp);
546 *exp = NULL;
547 }
548 out_sem:
549 up_write(&cli->cl_sem);
550
551 return rc;
552 }
553 EXPORT_SYMBOL(client_connect_import);
554
555 int client_disconnect_export(struct obd_export *exp)
556 {
557 struct obd_device *obd = class_exp2obd(exp);
558 struct client_obd *cli;
559 struct obd_import *imp;
560 int rc = 0, err;
561
562 if (!obd) {
563 CERROR("invalid export for disconnect: exp %p cookie %#llx\n",
564 exp, exp ? exp->exp_handle.h_cookie : -1);
565 return -EINVAL;
566 }
567
568 cli = &obd->u.cli;
569 imp = cli->cl_import;
570
571 down_write(&cli->cl_sem);
572 CDEBUG(D_INFO, "disconnect %s - %d\n", obd->obd_name,
573 cli->cl_conn_count);
574
575 if (!cli->cl_conn_count) {
576 CERROR("disconnecting disconnected device (%s)\n",
577 obd->obd_name);
578 rc = -EINVAL;
579 goto out_disconnect;
580 }
581
582 cli->cl_conn_count--;
583 if (cli->cl_conn_count) {
584 rc = 0;
585 goto out_disconnect;
586 }
587
588 /* Mark import deactivated now, so we don't try to reconnect if any
589 * of the cleanup RPCs fails (e.g. LDLM cancel, etc). We don't
590 * fully deactivate the import, or that would drop all requests. */
591 spin_lock(&imp->imp_lock);
592 imp->imp_deactive = 1;
593 spin_unlock(&imp->imp_lock);
594
595 /* Some non-replayable imports (MDS's OSCs) are pinged, so just
596 * delete it regardless. (It's safe to delete an import that was
597 * never added.) */
598 (void)ptlrpc_pinger_del_import(imp);
599
600 if (obd->obd_namespace != NULL) {
601 /* obd_force == local only */
602 ldlm_cli_cancel_unused(obd->obd_namespace, NULL,
603 obd->obd_force ? LCF_LOCAL : 0, NULL);
604 ldlm_namespace_free_prior(obd->obd_namespace, imp,
605 obd->obd_force);
606 }
607
608 /* There's no need to hold sem while disconnecting an import,
609 * and it may actually cause deadlock in GSS. */
610 up_write(&cli->cl_sem);
611 rc = ptlrpc_disconnect_import(imp, 0);
612 down_write(&cli->cl_sem);
613
614 ptlrpc_invalidate_import(imp);
615
616 out_disconnect:
617 /* Use server style - class_disconnect should be always called for
618 * o_disconnect. */
619 err = class_disconnect(exp);
620 if (!rc && err)
621 rc = err;
622
623 up_write(&cli->cl_sem);
624
625 return rc;
626 }
627 EXPORT_SYMBOL(client_disconnect_export);
628
629 /**
630 * Packs current SLV and Limit into \a req.
631 */
632 int target_pack_pool_reply(struct ptlrpc_request *req)
633 {
634 struct obd_device *obd;
635
636 /* Check that we still have all structures alive as this may
637 * be some late RPC at shutdown time. */
638 if (unlikely(!req->rq_export || !req->rq_export->exp_obd ||
639 !exp_connect_lru_resize(req->rq_export))) {
640 lustre_msg_set_slv(req->rq_repmsg, 0);
641 lustre_msg_set_limit(req->rq_repmsg, 0);
642 return 0;
643 }
644
645 /* OBD is alive here as export is alive, which we checked above. */
646 obd = req->rq_export->exp_obd;
647
648 read_lock(&obd->obd_pool_lock);
649 lustre_msg_set_slv(req->rq_repmsg, obd->obd_pool_slv);
650 lustre_msg_set_limit(req->rq_repmsg, obd->obd_pool_limit);
651 read_unlock(&obd->obd_pool_lock);
652
653 return 0;
654 }
655 EXPORT_SYMBOL(target_pack_pool_reply);
656
657 static int
658 target_send_reply_msg(struct ptlrpc_request *req, int rc, int fail_id)
659 {
660 if (OBD_FAIL_CHECK_ORSET(fail_id & ~OBD_FAIL_ONCE, OBD_FAIL_ONCE)) {
661 DEBUG_REQ(D_ERROR, req, "dropping reply");
662 return -ECOMM;
663 }
664
665 if (unlikely(rc)) {
666 DEBUG_REQ(D_NET, req, "processing error (%d)", rc);
667 req->rq_status = rc;
668 return ptlrpc_send_error(req, 1);
669 }
670
671 DEBUG_REQ(D_NET, req, "sending reply");
672 return ptlrpc_send_reply(req, PTLRPC_REPLY_MAYBE_DIFFICULT);
673 }
674
675 void target_send_reply(struct ptlrpc_request *req, int rc, int fail_id)
676 {
677 struct ptlrpc_service_part *svcpt;
678 int netrc;
679 struct ptlrpc_reply_state *rs;
680 struct obd_export *exp;
681
682 if (req->rq_no_reply)
683 return;
684
685 svcpt = req->rq_rqbd->rqbd_svcpt;
686 rs = req->rq_reply_state;
687 if (rs == NULL || !rs->rs_difficult) {
688 /* no notifiers */
689 target_send_reply_msg(req, rc, fail_id);
690 return;
691 }
692
693 /* must be an export if locks saved */
694 LASSERT(req->rq_export != NULL);
695 /* req/reply consistent */
696 LASSERT(rs->rs_svcpt == svcpt);
697
698 /* "fresh" reply */
699 LASSERT(!rs->rs_scheduled);
700 LASSERT(!rs->rs_scheduled_ever);
701 LASSERT(!rs->rs_handled);
702 LASSERT(!rs->rs_on_net);
703 LASSERT(rs->rs_export == NULL);
704 LASSERT(list_empty(&rs->rs_obd_list));
705 LASSERT(list_empty(&rs->rs_exp_list));
706
707 exp = class_export_get(req->rq_export);
708
709 /* disable reply scheduling while I'm setting up */
710 rs->rs_scheduled = 1;
711 rs->rs_on_net = 1;
712 rs->rs_xid = req->rq_xid;
713 rs->rs_transno = req->rq_transno;
714 rs->rs_export = exp;
715 rs->rs_opc = lustre_msg_get_opc(req->rq_reqmsg);
716
717 spin_lock(&exp->exp_uncommitted_replies_lock);
718 CDEBUG(D_NET, "rs transno = %llu, last committed = %llu\n",
719 rs->rs_transno, exp->exp_last_committed);
720 if (rs->rs_transno > exp->exp_last_committed) {
721 /* not committed already */
722 list_add_tail(&rs->rs_obd_list,
723 &exp->exp_uncommitted_replies);
724 }
725 spin_unlock(&exp->exp_uncommitted_replies_lock);
726
727 spin_lock(&exp->exp_lock);
728 list_add_tail(&rs->rs_exp_list, &exp->exp_outstanding_replies);
729 spin_unlock(&exp->exp_lock);
730
731 netrc = target_send_reply_msg(req, rc, fail_id);
732
733 spin_lock(&svcpt->scp_rep_lock);
734
735 atomic_inc(&svcpt->scp_nreps_difficult);
736
737 if (netrc != 0) {
738 /* error sending: reply is off the net. Also we need +1
739 * reply ref until ptlrpc_handle_rs() is done
740 * with the reply state (if the send was successful, there
741 * would have been +1 ref for the net, which
742 * reply_out_callback leaves alone) */
743 rs->rs_on_net = 0;
744 ptlrpc_rs_addref(rs);
745 }
746
747 spin_lock(&rs->rs_lock);
748 if (rs->rs_transno <= exp->exp_last_committed ||
749 (!rs->rs_on_net && !rs->rs_no_ack) ||
750 list_empty(&rs->rs_exp_list) || /* completed already */
751 list_empty(&rs->rs_obd_list)) {
752 CDEBUG(D_HA, "Schedule reply immediately\n");
753 ptlrpc_dispatch_difficult_reply(rs);
754 } else {
755 list_add(&rs->rs_list, &svcpt->scp_rep_active);
756 rs->rs_scheduled = 0; /* allow notifier to schedule */
757 }
758 spin_unlock(&rs->rs_lock);
759 spin_unlock(&svcpt->scp_rep_lock);
760 }
761 EXPORT_SYMBOL(target_send_reply);
762
763 ldlm_mode_t lck_compat_array[] = {
764 [LCK_EX] = LCK_COMPAT_EX,
765 [LCK_PW] = LCK_COMPAT_PW,
766 [LCK_PR] = LCK_COMPAT_PR,
767 [LCK_CW] = LCK_COMPAT_CW,
768 [LCK_CR] = LCK_COMPAT_CR,
769 [LCK_NL] = LCK_COMPAT_NL,
770 [LCK_GROUP] = LCK_COMPAT_GROUP,
771 [LCK_COS] = LCK_COMPAT_COS,
772 };
773
774 /**
775 * Rather arbitrary mapping from LDLM error codes to errno values. This should
776 * not escape to the user level.
777 */
778 int ldlm_error2errno(ldlm_error_t error)
779 {
780 int result;
781
782 switch (error) {
783 case ELDLM_OK:
784 result = 0;
785 break;
786 case ELDLM_LOCK_CHANGED:
787 result = -ESTALE;
788 break;
789 case ELDLM_LOCK_ABORTED:
790 result = -ENAVAIL;
791 break;
792 case ELDLM_LOCK_REPLACED:
793 result = -ESRCH;
794 break;
795 case ELDLM_NO_LOCK_DATA:
796 result = -ENOENT;
797 break;
798 case ELDLM_NAMESPACE_EXISTS:
799 result = -EEXIST;
800 break;
801 case ELDLM_BAD_NAMESPACE:
802 result = -EBADF;
803 break;
804 default:
805 if (((int)error) < 0) /* cast to signed type */
806 result = error; /* as ldlm_error_t can be unsigned */
807 else {
808 CERROR("Invalid DLM result code: %d\n", error);
809 result = -EPROTO;
810 }
811 }
812 return result;
813 }
814 EXPORT_SYMBOL(ldlm_error2errno);
815
816 #if LUSTRE_TRACKS_LOCK_EXP_REFS
817 void ldlm_dump_export_locks(struct obd_export *exp)
818 {
819 spin_lock(&exp->exp_locks_list_guard);
820 if (!list_empty(&exp->exp_locks_list)) {
821 struct ldlm_lock *lock;
822
823 CERROR("dumping locks for export %p,ignore if the unmount doesn't hang\n",
824 exp);
825 list_for_each_entry(lock, &exp->exp_locks_list,
826 l_exp_refs_link)
827 LDLM_ERROR(lock, "lock:");
828 }
829 spin_unlock(&exp->exp_locks_list_guard);
830 }
831 #endif