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) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2012, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
37 #define DEBUG_SUBSYSTEM S_MDC
39 # include <linux/module.h>
41 #include "../include/lustre_intent.h"
42 #include "../include/obd.h"
43 #include "../include/obd_class.h"
44 #include "../include/lustre_dlm.h"
45 #include "../include/lustre_fid.h" /* fid_res_name_eq() */
46 #include "../include/lustre_mdc.h"
47 #include "../include/lustre_net.h"
48 #include "../include/lustre_req_layout.h"
49 #include "mdc_internal.h"
51 struct mdc_getattr_args {
52 struct obd_export *ga_exp;
53 struct md_enqueue_info *ga_minfo;
54 struct ldlm_enqueue_info *ga_einfo;
57 int it_disposition(struct lookup_intent *it, int flag)
59 return it->d.lustre.it_disposition & flag;
61 EXPORT_SYMBOL(it_disposition);
63 void it_set_disposition(struct lookup_intent *it, int flag)
65 it->d.lustre.it_disposition |= flag;
67 EXPORT_SYMBOL(it_set_disposition);
69 void it_clear_disposition(struct lookup_intent *it, int flag)
71 it->d.lustre.it_disposition &= ~flag;
73 EXPORT_SYMBOL(it_clear_disposition);
75 int it_open_error(int phase, struct lookup_intent *it)
77 if (it_disposition(it, DISP_OPEN_LEASE)) {
78 if (phase >= DISP_OPEN_LEASE)
79 return it->d.lustre.it_status;
83 if (it_disposition(it, DISP_OPEN_OPEN)) {
84 if (phase >= DISP_OPEN_OPEN)
85 return it->d.lustre.it_status;
90 if (it_disposition(it, DISP_OPEN_CREATE)) {
91 if (phase >= DISP_OPEN_CREATE)
92 return it->d.lustre.it_status;
97 if (it_disposition(it, DISP_LOOKUP_EXECD)) {
98 if (phase >= DISP_LOOKUP_EXECD)
99 return it->d.lustre.it_status;
104 if (it_disposition(it, DISP_IT_EXECD)) {
105 if (phase >= DISP_IT_EXECD)
106 return it->d.lustre.it_status;
110 CERROR("it disp: %X, status: %d\n", it->d.lustre.it_disposition,
111 it->d.lustre.it_status);
115 EXPORT_SYMBOL(it_open_error);
117 /* this must be called on a lockh that is known to have a referenced lock */
118 int mdc_set_lock_data(struct obd_export *exp, __u64 *lockh, void *data,
121 struct ldlm_lock *lock;
122 struct inode *new_inode = data;
130 lock = ldlm_handle2lock((struct lustre_handle *)lockh);
132 LASSERT(lock != NULL);
133 lock_res_and_lock(lock);
134 if (lock->l_resource->lr_lvb_inode &&
135 lock->l_resource->lr_lvb_inode != data) {
136 struct inode *old_inode = lock->l_resource->lr_lvb_inode;
138 LASSERTF(old_inode->i_state & I_FREEING,
139 "Found existing inode %p/%lu/%u state %lu in lock: setting data to %p/%lu/%u\n",
140 old_inode, old_inode->i_ino, old_inode->i_generation,
141 old_inode->i_state, new_inode, new_inode->i_ino,
142 new_inode->i_generation);
144 lock->l_resource->lr_lvb_inode = new_inode;
146 *bits = lock->l_policy_data.l_inodebits.bits;
148 unlock_res_and_lock(lock);
154 ldlm_mode_t mdc_lock_match(struct obd_export *exp, __u64 flags,
155 const struct lu_fid *fid, ldlm_type_t type,
156 ldlm_policy_data_t *policy, ldlm_mode_t mode,
157 struct lustre_handle *lockh)
159 struct ldlm_res_id res_id;
162 fid_build_reg_res_name(fid, &res_id);
163 /* LU-4405: Clear bits not supported by server */
164 policy->l_inodebits.bits &= exp_connect_ibits(exp);
165 rc = ldlm_lock_match(class_exp2obd(exp)->obd_namespace, flags,
166 &res_id, type, policy, mode, lockh, 0);
170 int mdc_cancel_unused(struct obd_export *exp,
171 const struct lu_fid *fid,
172 ldlm_policy_data_t *policy,
174 ldlm_cancel_flags_t flags,
177 struct ldlm_res_id res_id;
178 struct obd_device *obd = class_exp2obd(exp);
181 fid_build_reg_res_name(fid, &res_id);
182 rc = ldlm_cli_cancel_unused_resource(obd->obd_namespace, &res_id,
183 policy, mode, flags, opaque);
187 int mdc_null_inode(struct obd_export *exp,
188 const struct lu_fid *fid)
190 struct ldlm_res_id res_id;
191 struct ldlm_resource *res;
192 struct ldlm_namespace *ns = class_exp2obd(exp)->obd_namespace;
194 LASSERTF(ns != NULL, "no namespace passed\n");
196 fid_build_reg_res_name(fid, &res_id);
198 res = ldlm_resource_get(ns, NULL, &res_id, 0, 0);
203 res->lr_lvb_inode = NULL;
206 ldlm_resource_putref(res);
210 /* find any ldlm lock of the inode in mdc
214 int mdc_find_cbdata(struct obd_export *exp,
215 const struct lu_fid *fid,
216 ldlm_iterator_t it, void *data)
218 struct ldlm_res_id res_id;
221 fid_build_reg_res_name((struct lu_fid *)fid, &res_id);
222 rc = ldlm_resource_iterate(class_exp2obd(exp)->obd_namespace, &res_id,
224 if (rc == LDLM_ITER_STOP)
226 else if (rc == LDLM_ITER_CONTINUE)
231 static inline void mdc_clear_replay_flag(struct ptlrpc_request *req, int rc)
233 /* Don't hold error requests for replay. */
234 if (req->rq_replay) {
235 spin_lock(&req->rq_lock);
237 spin_unlock(&req->rq_lock);
239 if (rc && req->rq_transno != 0) {
240 DEBUG_REQ(D_ERROR, req, "transno returned on error rc %d", rc);
245 /* Save a large LOV EA into the request buffer so that it is available
246 * for replay. We don't do this in the initial request because the
247 * original request doesn't need this buffer (at most it sends just the
248 * lov_mds_md) and it is a waste of RAM/bandwidth to send the empty
249 * buffer and may also be difficult to allocate and save a very large
250 * request buffer for each open. (bug 5707)
252 * OOM here may cause recovery failure if lmm is needed (only for the
253 * original open if the MDS crashed just when this client also OOM'd)
254 * but this is incredibly unlikely, and questionable whether the client
255 * could do MDS recovery under OOM anyways... */
256 static void mdc_realloc_openmsg(struct ptlrpc_request *req,
257 struct mdt_body *body)
261 /* FIXME: remove this explicit offset. */
262 rc = sptlrpc_cli_enlarge_reqbuf(req, DLM_INTENT_REC_OFF + 4,
265 CERROR("Can't enlarge segment %d size to %d\n",
266 DLM_INTENT_REC_OFF + 4, body->eadatasize);
267 body->valid &= ~OBD_MD_FLEASIZE;
268 body->eadatasize = 0;
272 static struct ptlrpc_request *mdc_intent_open_pack(struct obd_export *exp,
273 struct lookup_intent *it,
274 struct md_op_data *op_data,
275 void *lmm, int lmmsize,
278 struct ptlrpc_request *req;
279 struct obd_device *obddev = class_exp2obd(exp);
280 struct ldlm_intent *lit;
286 it->it_create_mode = (it->it_create_mode & ~S_IFMT) | S_IFREG;
288 /* XXX: openlock is not cancelled for cross-refs. */
289 /* If inode is known, cancel conflicting OPEN locks. */
290 if (fid_is_sane(&op_data->op_fid2)) {
291 if (it->it_flags & MDS_OPEN_LEASE) { /* try to get lease */
292 if (it->it_flags & FMODE_WRITE)
297 if (it->it_flags & (FMODE_WRITE|MDS_OPEN_TRUNC))
299 else if (it->it_flags & __FMODE_EXEC)
304 count = mdc_resource_get_unused(exp, &op_data->op_fid2,
309 /* If CREATE, cancel parent's UPDATE lock. */
310 if (it->it_op & IT_CREAT)
314 count += mdc_resource_get_unused(exp, &op_data->op_fid1,
316 MDS_INODELOCK_UPDATE);
318 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
319 &RQF_LDLM_INTENT_OPEN);
321 ldlm_lock_list_put(&cancels, l_bl_ast, count);
322 return ERR_PTR(-ENOMEM);
325 /* parent capability */
326 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
327 /* child capability, reserve the size according to parent capa, it will
328 * be filled after we get the reply */
329 mdc_set_capa_size(req, &RMF_CAPA2, op_data->op_capa1);
331 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
332 op_data->op_namelen + 1);
333 req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT,
334 max(lmmsize, obddev->u.cli.cl_default_mds_easize));
336 rc = ldlm_prep_enqueue_req(exp, req, &cancels, count);
338 ptlrpc_request_free(req);
342 spin_lock(&req->rq_lock);
343 req->rq_replay = req->rq_import->imp_replayable;
344 spin_unlock(&req->rq_lock);
346 /* pack the intent */
347 lit = req_capsule_client_get(&req->rq_pill, &RMF_LDLM_INTENT);
348 lit->opc = (__u64)it->it_op;
350 /* pack the intended request */
351 mdc_open_pack(req, op_data, it->it_create_mode, 0, it->it_flags, lmm,
354 /* for remote client, fetch remote perm for current user */
355 if (client_is_remote(exp))
356 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
357 sizeof(struct mdt_remote_perm));
358 ptlrpc_request_set_replen(req);
362 static struct ptlrpc_request *
363 mdc_intent_getxattr_pack(struct obd_export *exp,
364 struct lookup_intent *it,
365 struct md_op_data *op_data)
367 struct ptlrpc_request *req;
368 struct ldlm_intent *lit;
369 int rc, count = 0, maxdata;
374 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
375 &RQF_LDLM_INTENT_GETXATTR);
377 return ERR_PTR(-ENOMEM);
379 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
381 rc = ldlm_prep_enqueue_req(exp, req, &cancels, count);
383 ptlrpc_request_free(req);
387 /* pack the intent */
388 lit = req_capsule_client_get(&req->rq_pill, &RMF_LDLM_INTENT);
389 lit->opc = IT_GETXATTR;
391 maxdata = class_exp2cliimp(exp)->imp_connect_data.ocd_max_easize;
393 /* pack the intended request */
394 mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
395 op_data->op_valid, maxdata, -1, 0);
397 req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
398 RCL_SERVER, maxdata);
400 req_capsule_set_size(&req->rq_pill, &RMF_EAVALS,
401 RCL_SERVER, maxdata);
403 req_capsule_set_size(&req->rq_pill, &RMF_EAVALS_LENS,
404 RCL_SERVER, maxdata);
406 ptlrpc_request_set_replen(req);
411 static struct ptlrpc_request *mdc_intent_unlink_pack(struct obd_export *exp,
412 struct lookup_intent *it,
413 struct md_op_data *op_data)
415 struct ptlrpc_request *req;
416 struct obd_device *obddev = class_exp2obd(exp);
417 struct ldlm_intent *lit;
420 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
421 &RQF_LDLM_INTENT_UNLINK);
423 return ERR_PTR(-ENOMEM);
425 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
426 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
427 op_data->op_namelen + 1);
429 rc = ldlm_prep_enqueue_req(exp, req, NULL, 0);
431 ptlrpc_request_free(req);
435 /* pack the intent */
436 lit = req_capsule_client_get(&req->rq_pill, &RMF_LDLM_INTENT);
437 lit->opc = (__u64)it->it_op;
439 /* pack the intended request */
440 mdc_unlink_pack(req, op_data);
442 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
443 obddev->u.cli.cl_default_mds_easize);
444 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
445 obddev->u.cli.cl_default_mds_cookiesize);
446 ptlrpc_request_set_replen(req);
450 static struct ptlrpc_request *mdc_intent_getattr_pack(struct obd_export *exp,
451 struct lookup_intent *it,
452 struct md_op_data *op_data)
454 struct ptlrpc_request *req;
455 struct obd_device *obddev = class_exp2obd(exp);
456 u64 valid = OBD_MD_FLGETATTR | OBD_MD_FLEASIZE |
457 OBD_MD_FLMODEASIZE | OBD_MD_FLDIREA |
458 OBD_MD_FLMDSCAPA | OBD_MD_MEA |
459 (client_is_remote(exp) ?
460 OBD_MD_FLRMTPERM : OBD_MD_FLACL);
461 struct ldlm_intent *lit;
465 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
466 &RQF_LDLM_INTENT_GETATTR);
468 return ERR_PTR(-ENOMEM);
470 mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
471 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
472 op_data->op_namelen + 1);
474 rc = ldlm_prep_enqueue_req(exp, req, NULL, 0);
476 ptlrpc_request_free(req);
480 /* pack the intent */
481 lit = req_capsule_client_get(&req->rq_pill, &RMF_LDLM_INTENT);
482 lit->opc = (__u64)it->it_op;
484 if (obddev->u.cli.cl_default_mds_easize > 0)
485 easize = obddev->u.cli.cl_default_mds_easize;
487 easize = obddev->u.cli.cl_max_mds_easize;
489 /* pack the intended request */
490 mdc_getattr_pack(req, valid, it->it_flags, op_data, easize);
492 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER, easize);
493 if (client_is_remote(exp))
494 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
495 sizeof(struct mdt_remote_perm));
496 ptlrpc_request_set_replen(req);
500 static struct ptlrpc_request *mdc_intent_layout_pack(struct obd_export *exp,
501 struct lookup_intent *it,
502 struct md_op_data *unused)
504 struct obd_device *obd = class_exp2obd(exp);
505 struct ptlrpc_request *req;
506 struct ldlm_intent *lit;
507 struct layout_intent *layout;
510 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
511 &RQF_LDLM_INTENT_LAYOUT);
513 return ERR_PTR(-ENOMEM);
515 req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT, 0);
516 rc = ldlm_prep_enqueue_req(exp, req, NULL, 0);
518 ptlrpc_request_free(req);
522 /* pack the intent */
523 lit = req_capsule_client_get(&req->rq_pill, &RMF_LDLM_INTENT);
524 lit->opc = (__u64)it->it_op;
526 /* pack the layout intent request */
527 layout = req_capsule_client_get(&req->rq_pill, &RMF_LAYOUT_INTENT);
528 /* LAYOUT_INTENT_ACCESS is generic, specific operation will be
529 * set for replication */
530 layout->li_opc = LAYOUT_INTENT_ACCESS;
532 req_capsule_set_size(&req->rq_pill, &RMF_DLM_LVB, RCL_SERVER,
533 obd->u.cli.cl_default_mds_easize);
534 ptlrpc_request_set_replen(req);
538 static struct ptlrpc_request *
539 mdc_enqueue_pack(struct obd_export *exp, int lvb_len)
541 struct ptlrpc_request *req;
544 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LDLM_ENQUEUE);
546 return ERR_PTR(-ENOMEM);
548 rc = ldlm_prep_enqueue_req(exp, req, NULL, 0);
550 ptlrpc_request_free(req);
554 req_capsule_set_size(&req->rq_pill, &RMF_DLM_LVB, RCL_SERVER, lvb_len);
555 ptlrpc_request_set_replen(req);
559 static int mdc_finish_enqueue(struct obd_export *exp,
560 struct ptlrpc_request *req,
561 struct ldlm_enqueue_info *einfo,
562 struct lookup_intent *it,
563 struct lustre_handle *lockh,
566 struct req_capsule *pill = &req->rq_pill;
567 struct ldlm_request *lockreq;
568 struct ldlm_reply *lockrep;
569 struct lustre_intent_data *intent = &it->d.lustre;
570 struct ldlm_lock *lock;
571 void *lvb_data = NULL;
575 /* Similarly, if we're going to replay this request, we don't want to
576 * actually get a lock, just perform the intent. */
577 if (req->rq_transno || req->rq_replay) {
578 lockreq = req_capsule_client_get(pill, &RMF_DLM_REQ);
579 lockreq->lock_flags |= ldlm_flags_to_wire(LDLM_FL_INTENT_ONLY);
582 if (rc == ELDLM_LOCK_ABORTED) {
584 memset(lockh, 0, sizeof(*lockh));
586 } else { /* rc = 0 */
587 lock = ldlm_handle2lock(lockh);
588 LASSERT(lock != NULL);
590 /* If the server gave us back a different lock mode, we should
591 * fix up our variables. */
592 if (lock->l_req_mode != einfo->ei_mode) {
593 ldlm_lock_addref(lockh, lock->l_req_mode);
594 ldlm_lock_decref(lockh, einfo->ei_mode);
595 einfo->ei_mode = lock->l_req_mode;
600 lockrep = req_capsule_server_get(pill, &RMF_DLM_REP);
601 LASSERT(lockrep != NULL); /* checked by ldlm_cli_enqueue() */
603 intent->it_disposition = (int)lockrep->lock_policy_res1;
604 intent->it_status = (int)lockrep->lock_policy_res2;
605 intent->it_lock_mode = einfo->ei_mode;
606 intent->it_lock_handle = lockh->cookie;
607 intent->it_data = req;
609 /* Technically speaking rq_transno must already be zero if
610 * it_status is in error, so the check is a bit redundant */
611 if ((!req->rq_transno || intent->it_status < 0) && req->rq_replay)
612 mdc_clear_replay_flag(req, intent->it_status);
614 /* If we're doing an IT_OPEN which did not result in an actual
615 * successful open, then we need to remove the bit which saves
616 * this request for unconditional replay.
618 * It's important that we do this first! Otherwise we might exit the
619 * function without doing so, and try to replay a failed create
621 if (it->it_op & IT_OPEN && req->rq_replay &&
622 (!it_disposition(it, DISP_OPEN_OPEN) || intent->it_status != 0))
623 mdc_clear_replay_flag(req, intent->it_status);
625 DEBUG_REQ(D_RPCTRACE, req, "op: %d disposition: %x, status: %d",
626 it->it_op, intent->it_disposition, intent->it_status);
628 /* We know what to expect, so we do any byte flipping required here */
629 if (it->it_op & (IT_OPEN | IT_UNLINK | IT_LOOKUP | IT_GETATTR)) {
630 struct mdt_body *body;
632 body = req_capsule_server_get(pill, &RMF_MDT_BODY);
634 CERROR("Can't swab mdt_body\n");
638 if (it_disposition(it, DISP_OPEN_OPEN) &&
639 !it_open_error(DISP_OPEN_OPEN, it)) {
641 * If this is a successful OPEN request, we need to set
642 * replay handler and data early, so that if replay
643 * happens immediately after swabbing below, new reply
644 * is swabbed by that handler correctly.
646 mdc_set_open_replay_data(NULL, NULL, it);
649 if ((body->valid & (OBD_MD_FLDIREA | OBD_MD_FLEASIZE)) != 0) {
652 mdc_update_max_ea_from_body(exp, body);
655 * The eadata is opaque; just check that it is there.
656 * Eventually, obd_unpackmd() will check the contents.
658 eadata = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
663 /* save lvb data and length in case this is for layout
666 lvb_len = body->eadatasize;
669 * We save the reply LOV EA in case we have to replay a
670 * create for recovery. If we didn't allocate a large
671 * enough request buffer above we need to reallocate it
672 * here to hold the actual LOV EA.
674 * To not save LOV EA if request is not going to replay
675 * (for example error one).
677 if ((it->it_op & IT_OPEN) && req->rq_replay) {
680 if (req_capsule_get_size(pill, &RMF_EADATA,
683 mdc_realloc_openmsg(req, body);
685 req_capsule_shrink(pill, &RMF_EADATA,
689 req_capsule_set_size(pill, &RMF_EADATA,
693 lmm = req_capsule_client_get(pill, &RMF_EADATA);
695 memcpy(lmm, eadata, body->eadatasize);
699 if (body->valid & OBD_MD_FLRMTPERM) {
700 struct mdt_remote_perm *perm;
702 LASSERT(client_is_remote(exp));
703 perm = req_capsule_server_swab_get(pill, &RMF_ACL,
704 lustre_swab_mdt_remote_perm);
708 if (body->valid & OBD_MD_FLMDSCAPA) {
709 struct lustre_capa *capa, *p;
711 capa = req_capsule_server_get(pill, &RMF_CAPA1);
715 if (it->it_op & IT_OPEN) {
716 /* client fid capa will be checked in replay */
717 p = req_capsule_client_get(pill, &RMF_CAPA2);
722 if (body->valid & OBD_MD_FLOSSCAPA) {
723 struct lustre_capa *capa;
725 capa = req_capsule_server_get(pill, &RMF_CAPA2);
729 } else if (it->it_op & IT_LAYOUT) {
730 /* maybe the lock was granted right away and layout
731 * is packed into RMF_DLM_LVB of req */
732 lvb_len = req_capsule_get_size(pill, &RMF_DLM_LVB, RCL_SERVER);
734 lvb_data = req_capsule_server_sized_get(pill,
735 &RMF_DLM_LVB, lvb_len);
736 if (lvb_data == NULL)
741 /* fill in stripe data for layout lock */
742 lock = ldlm_handle2lock(lockh);
743 if (lock != NULL && ldlm_has_layout(lock) && lvb_data != NULL) {
746 LDLM_DEBUG(lock, "layout lock returned by: %s, lvb_len: %d\n",
747 ldlm_it2str(it->it_op), lvb_len);
749 lmm = libcfs_kvzalloc(lvb_len, GFP_NOFS);
754 memcpy(lmm, lvb_data, lvb_len);
756 /* install lvb_data */
757 lock_res_and_lock(lock);
758 if (lock->l_lvb_data == NULL) {
759 lock->l_lvb_type = LVB_T_LAYOUT;
760 lock->l_lvb_data = lmm;
761 lock->l_lvb_len = lvb_len;
764 unlock_res_and_lock(lock);
774 /* We always reserve enough space in the reply packet for a stripe MD, because
775 * we don't know in advance the file type. */
776 int mdc_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
777 struct lookup_intent *it, struct md_op_data *op_data,
778 struct lustre_handle *lockh, void *lmm, int lmmsize,
779 struct ptlrpc_request **reqp, u64 extra_lock_flags)
781 static const ldlm_policy_data_t lookup_policy = {
782 .l_inodebits = { MDS_INODELOCK_LOOKUP }
784 static const ldlm_policy_data_t update_policy = {
785 .l_inodebits = { MDS_INODELOCK_UPDATE }
787 static const ldlm_policy_data_t layout_policy = {
788 .l_inodebits = { MDS_INODELOCK_LAYOUT }
790 static const ldlm_policy_data_t getxattr_policy = {
791 .l_inodebits = { MDS_INODELOCK_XATTR }
793 ldlm_policy_data_t const *policy = &lookup_policy;
794 struct obd_device *obddev = class_exp2obd(exp);
795 struct ptlrpc_request *req;
796 u64 flags, saved_flags = extra_lock_flags;
797 struct ldlm_res_id res_id;
798 int generation, resends = 0;
799 struct ldlm_reply *lockrep;
800 enum lvb_type lvb_type = LVB_T_NONE;
803 LASSERTF(!it || einfo->ei_type == LDLM_IBITS, "lock type %d\n",
806 fid_build_reg_res_name(&op_data->op_fid1, &res_id);
809 saved_flags |= LDLM_FL_HAS_INTENT;
810 if (it->it_op & (IT_UNLINK | IT_GETATTR | IT_READDIR))
811 policy = &update_policy;
812 else if (it->it_op & IT_LAYOUT)
813 policy = &layout_policy;
814 else if (it->it_op & (IT_GETXATTR | IT_SETXATTR))
815 policy = &getxattr_policy;
818 LASSERT(reqp == NULL);
820 generation = obddev->u.cli.cl_import->imp_generation;
824 /* The only way right now is FLOCK, in this case we hide flock
825 policy as lmm, but lmmsize is 0 */
826 LASSERT(lmm && lmmsize == 0);
827 LASSERTF(einfo->ei_type == LDLM_FLOCK, "lock type %d\n",
829 policy = (ldlm_policy_data_t *)lmm;
830 res_id.name[3] = LDLM_FLOCK;
832 } else if (it->it_op & IT_OPEN) {
833 req = mdc_intent_open_pack(exp, it, op_data, lmm, lmmsize,
835 policy = &update_policy;
836 einfo->ei_cbdata = NULL;
838 } else if (it->it_op & IT_UNLINK) {
839 req = mdc_intent_unlink_pack(exp, it, op_data);
840 } else if (it->it_op & (IT_GETATTR | IT_LOOKUP)) {
841 req = mdc_intent_getattr_pack(exp, it, op_data);
842 } else if (it->it_op & IT_READDIR) {
843 req = mdc_enqueue_pack(exp, 0);
844 } else if (it->it_op & IT_LAYOUT) {
845 if (!imp_connect_lvb_type(class_exp2cliimp(exp)))
847 req = mdc_intent_layout_pack(exp, it, op_data);
848 lvb_type = LVB_T_LAYOUT;
849 } else if (it->it_op & IT_GETXATTR) {
850 req = mdc_intent_getxattr_pack(exp, it, op_data);
859 if (req != NULL && it && it->it_op & IT_CREAT)
860 /* ask ptlrpc not to resend on EINPROGRESS since we have our own
862 req->rq_no_retry_einprogress = 1;
865 req->rq_generation_set = 1;
866 req->rq_import_generation = generation;
867 req->rq_sent = get_seconds() + resends;
870 /* It is important to obtain rpc_lock first (if applicable), so that
871 * threads that are serialised with rpc_lock are not polluting our
872 * rpcs in flight counter. We do not do flock request limiting, though*/
874 mdc_get_rpc_lock(obddev->u.cli.cl_rpc_lock, it);
875 rc = mdc_enter_request(&obddev->u.cli);
877 mdc_put_rpc_lock(obddev->u.cli.cl_rpc_lock, it);
878 mdc_clear_replay_flag(req, 0);
879 ptlrpc_req_finished(req);
884 rc = ldlm_cli_enqueue(exp, &req, einfo, &res_id, policy, &flags, NULL,
885 0, lvb_type, lockh, 0);
887 /* For flock requests we immediately return without further
888 delay and let caller deal with the rest, since rest of
889 this function metadata processing makes no sense for flock
890 requests anyway. But in case of problem during comms with
891 Server (ETIMEDOUT) or any signal/kill attempt (EINTR), we
892 can not rely on caller and this mainly for F_UNLCKs
893 (explicits or automatically generated by Kernel to clean
894 current FLocks upon exit) that can't be trashed */
895 if ((rc == -EINTR) || (rc == -ETIMEDOUT))
900 mdc_exit_request(&obddev->u.cli);
901 mdc_put_rpc_lock(obddev->u.cli.cl_rpc_lock, it);
904 CDEBUG_LIMIT((rc == -EACCES || rc == -EIDRM) ? D_INFO : D_ERROR,
905 "%s: ldlm_cli_enqueue failed: rc = %d\n",
906 obddev->obd_name, rc);
908 mdc_clear_replay_flag(req, rc);
909 ptlrpc_req_finished(req);
913 lockrep = req_capsule_server_get(&req->rq_pill, &RMF_DLM_REP);
914 LASSERT(lockrep != NULL);
916 lockrep->lock_policy_res2 =
917 ptlrpc_status_ntoh(lockrep->lock_policy_res2);
919 /* Retry the create infinitely when we get -EINPROGRESS from
920 * server. This is required by the new quota design. */
921 if (it && it->it_op & IT_CREAT &&
922 (int)lockrep->lock_policy_res2 == -EINPROGRESS) {
923 mdc_clear_replay_flag(req, rc);
924 ptlrpc_req_finished(req);
927 CDEBUG(D_HA, "%s: resend:%d op:%d "DFID"/"DFID"\n",
928 obddev->obd_name, resends, it->it_op,
929 PFID(&op_data->op_fid1), PFID(&op_data->op_fid2));
931 if (generation == obddev->u.cli.cl_import->imp_generation) {
934 CDEBUG(D_HA, "resend cross eviction\n");
939 rc = mdc_finish_enqueue(exp, req, einfo, it, lockh, rc);
941 if (lustre_handle_is_used(lockh)) {
942 ldlm_lock_decref(lockh, einfo->ei_mode);
943 memset(lockh, 0, sizeof(*lockh));
945 ptlrpc_req_finished(req);
947 it->d.lustre.it_lock_handle = 0;
948 it->d.lustre.it_lock_mode = 0;
949 it->d.lustre.it_data = NULL;
955 static int mdc_finish_intent_lock(struct obd_export *exp,
956 struct ptlrpc_request *request,
957 struct md_op_data *op_data,
958 struct lookup_intent *it,
959 struct lustre_handle *lockh)
961 struct lustre_handle old_lock;
962 struct mdt_body *mdt_body;
963 struct ldlm_lock *lock;
966 LASSERT(request != NULL);
967 LASSERT(request != LP_POISON);
968 LASSERT(request->rq_repmsg != LP_POISON);
970 if (!it_disposition(it, DISP_IT_EXECD)) {
971 /* The server failed before it even started executing the
972 * intent, i.e. because it couldn't unpack the request. */
973 LASSERT(it->d.lustre.it_status != 0);
974 return it->d.lustre.it_status;
976 rc = it_open_error(DISP_IT_EXECD, it);
980 mdt_body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
981 LASSERT(mdt_body != NULL); /* mdc_enqueue checked */
983 /* If we were revalidating a fid/name pair, mark the intent in
984 * case we fail and get called again from lookup */
985 if (fid_is_sane(&op_data->op_fid2) &&
986 it->it_create_mode & M_CHECK_STALE &&
987 it->it_op != IT_GETATTR) {
989 /* Also: did we find the same inode? */
990 /* sever can return one of two fids:
991 * op_fid2 - new allocated fid - if file is created.
992 * op_fid3 - existent fid - if file only open.
993 * op_fid3 is saved in lmv_intent_open */
994 if ((!lu_fid_eq(&op_data->op_fid2, &mdt_body->fid1)) &&
995 (!lu_fid_eq(&op_data->op_fid3, &mdt_body->fid1))) {
996 CDEBUG(D_DENTRY, "Found stale data "DFID"("DFID")/"DFID
997 "\n", PFID(&op_data->op_fid2),
998 PFID(&op_data->op_fid2), PFID(&mdt_body->fid1));
1003 rc = it_open_error(DISP_LOOKUP_EXECD, it);
1007 /* keep requests around for the multiple phases of the call
1008 * this shows the DISP_XX must guarantee we make it into the call
1010 if (!it_disposition(it, DISP_ENQ_CREATE_REF) &&
1011 it_disposition(it, DISP_OPEN_CREATE) &&
1012 !it_open_error(DISP_OPEN_CREATE, it)) {
1013 it_set_disposition(it, DISP_ENQ_CREATE_REF);
1014 ptlrpc_request_addref(request); /* balanced in ll_create_node */
1016 if (!it_disposition(it, DISP_ENQ_OPEN_REF) &&
1017 it_disposition(it, DISP_OPEN_OPEN) &&
1018 !it_open_error(DISP_OPEN_OPEN, it)) {
1019 it_set_disposition(it, DISP_ENQ_OPEN_REF);
1020 ptlrpc_request_addref(request); /* balanced in ll_file_open */
1021 /* BUG 11546 - eviction in the middle of open rpc processing */
1022 OBD_FAIL_TIMEOUT(OBD_FAIL_MDC_ENQUEUE_PAUSE, obd_timeout);
1025 if (it->it_op & IT_CREAT) {
1026 /* XXX this belongs in ll_create_it */
1027 } else if (it->it_op == IT_OPEN) {
1028 LASSERT(!it_disposition(it, DISP_OPEN_CREATE));
1030 LASSERT(it->it_op & (IT_GETATTR | IT_LOOKUP | IT_LAYOUT));
1033 /* If we already have a matching lock, then cancel the new
1034 * one. We have to set the data here instead of in
1035 * mdc_enqueue, because we need to use the child's inode as
1036 * the l_ast_data to match, and that's not available until
1037 * intent_finish has performed the iget().) */
1038 lock = ldlm_handle2lock(lockh);
1040 ldlm_policy_data_t policy = lock->l_policy_data;
1042 LDLM_DEBUG(lock, "matching against this");
1044 LASSERTF(fid_res_name_eq(&mdt_body->fid1,
1045 &lock->l_resource->lr_name),
1046 "Lock res_id: "DLDLMRES", fid: "DFID"\n",
1047 PLDLMRES(lock->l_resource), PFID(&mdt_body->fid1));
1048 LDLM_LOCK_PUT(lock);
1050 memcpy(&old_lock, lockh, sizeof(*lockh));
1051 if (ldlm_lock_match(NULL, LDLM_FL_BLOCK_GRANTED, NULL,
1052 LDLM_IBITS, &policy, LCK_NL,
1054 ldlm_lock_decref_and_cancel(lockh,
1055 it->d.lustre.it_lock_mode);
1056 memcpy(lockh, &old_lock, sizeof(old_lock));
1057 it->d.lustre.it_lock_handle = lockh->cookie;
1061 "D_IT dentry %.*s intent: %s status %d disp %x rc %d\n",
1062 op_data->op_namelen, op_data->op_name, ldlm_it2str(it->it_op),
1063 it->d.lustre.it_status, it->d.lustre.it_disposition, rc);
1067 int mdc_revalidate_lock(struct obd_export *exp, struct lookup_intent *it,
1068 struct lu_fid *fid, __u64 *bits)
1070 /* We could just return 1 immediately, but since we should only
1071 * be called in revalidate_it if we already have a lock, let's
1073 struct ldlm_res_id res_id;
1074 struct lustre_handle lockh;
1075 ldlm_policy_data_t policy;
1078 if (it->d.lustre.it_lock_handle) {
1079 lockh.cookie = it->d.lustre.it_lock_handle;
1080 mode = ldlm_revalidate_lock_handle(&lockh, bits);
1082 fid_build_reg_res_name(fid, &res_id);
1083 switch (it->it_op) {
1085 /* File attributes are held under multiple bits:
1086 * nlink is under lookup lock, size and times are
1087 * under UPDATE lock and recently we've also got
1088 * a separate permissions lock for owner/group/acl that
1089 * were protected by lookup lock before.
1090 * Getattr must provide all of that information,
1091 * so we need to ensure we have all of those locks.
1092 * Unfortunately, if the bits are split across multiple
1093 * locks, there's no easy way to match all of them here,
1094 * so an extra RPC would be performed to fetch all
1095 * of those bits at once for now. */
1096 /* For new MDTs(> 2.4), UPDATE|PERM should be enough,
1097 * but for old MDTs (< 2.4), permission is covered
1098 * by LOOKUP lock, so it needs to match all bits here.*/
1099 policy.l_inodebits.bits = MDS_INODELOCK_UPDATE |
1100 MDS_INODELOCK_LOOKUP |
1104 policy.l_inodebits.bits = MDS_INODELOCK_LAYOUT;
1107 policy.l_inodebits.bits = MDS_INODELOCK_LOOKUP;
1111 mode = mdc_lock_match(exp, LDLM_FL_BLOCK_GRANTED, fid,
1112 LDLM_IBITS, &policy,
1113 LCK_CR | LCK_CW | LCK_PR | LCK_PW,
1118 it->d.lustre.it_lock_handle = lockh.cookie;
1119 it->d.lustre.it_lock_mode = mode;
1121 it->d.lustre.it_lock_handle = 0;
1122 it->d.lustre.it_lock_mode = 0;
1129 * This long block is all about fixing up the lock and request state
1130 * so that it is correct as of the moment _before_ the operation was
1131 * applied; that way, the VFS will think that everything is normal and
1132 * call Lustre's regular VFS methods.
1134 * If we're performing a creation, that means that unless the creation
1135 * failed with EEXIST, we should fake up a negative dentry.
1137 * For everything else, we want to lookup to succeed.
1139 * One additional note: if CREATE or OPEN succeeded, we add an extra
1140 * reference to the request because we need to keep it around until
1141 * ll_create/ll_open gets called.
1143 * The server will return to us, in it_disposition, an indication of
1144 * exactly what d.lustre.it_status refers to.
1146 * If DISP_OPEN_OPEN is set, then d.lustre.it_status refers to the open() call,
1147 * otherwise if DISP_OPEN_CREATE is set, then it status is the
1148 * creation failure mode. In either case, one of DISP_LOOKUP_NEG or
1149 * DISP_LOOKUP_POS will be set, indicating whether the child lookup
1152 * Else, if DISP_LOOKUP_EXECD then d.lustre.it_status is the rc of the
1155 int mdc_intent_lock(struct obd_export *exp, struct md_op_data *op_data,
1156 void *lmm, int lmmsize, struct lookup_intent *it,
1157 int lookup_flags, struct ptlrpc_request **reqp,
1158 ldlm_blocking_callback cb_blocking,
1159 __u64 extra_lock_flags)
1161 struct ldlm_enqueue_info einfo = {
1162 .ei_type = LDLM_IBITS,
1163 .ei_mode = it_to_lock_mode(it),
1164 .ei_cb_bl = cb_blocking,
1165 .ei_cb_cp = ldlm_completion_ast,
1167 struct lustre_handle lockh;
1172 CDEBUG(D_DLMTRACE, "(name: %.*s,"DFID") in obj "DFID
1173 ", intent: %s flags %#Lo\n", op_data->op_namelen,
1174 op_data->op_name, PFID(&op_data->op_fid2),
1175 PFID(&op_data->op_fid1), ldlm_it2str(it->it_op),
1179 if (fid_is_sane(&op_data->op_fid2) &&
1180 (it->it_op & (IT_LOOKUP | IT_GETATTR))) {
1181 /* We could just return 1 immediately, but since we should only
1182 * be called in revalidate_it if we already have a lock, let's
1184 it->d.lustre.it_lock_handle = 0;
1185 rc = mdc_revalidate_lock(exp, it, &op_data->op_fid2, NULL);
1186 /* Only return failure if it was not GETATTR by cfid
1187 (from inode_revalidate) */
1188 if (rc || op_data->op_namelen != 0)
1192 /* For case if upper layer did not alloc fid, do it now. */
1193 if (!fid_is_sane(&op_data->op_fid2) && it->it_op & IT_CREAT) {
1194 rc = mdc_fid_alloc(exp, &op_data->op_fid2, op_data);
1196 CERROR("Can't alloc new fid, rc %d\n", rc);
1200 rc = mdc_enqueue(exp, &einfo, it, op_data, &lockh, lmm, lmmsize, NULL,
1205 *reqp = it->d.lustre.it_data;
1206 rc = mdc_finish_intent_lock(exp, *reqp, op_data, it, &lockh);
1210 static int mdc_intent_getattr_async_interpret(const struct lu_env *env,
1211 struct ptlrpc_request *req,
1214 struct mdc_getattr_args *ga = args;
1215 struct obd_export *exp = ga->ga_exp;
1216 struct md_enqueue_info *minfo = ga->ga_minfo;
1217 struct ldlm_enqueue_info *einfo = ga->ga_einfo;
1218 struct lookup_intent *it;
1219 struct lustre_handle *lockh;
1220 struct obd_device *obddev;
1221 struct ldlm_reply *lockrep;
1222 __u64 flags = LDLM_FL_HAS_INTENT;
1225 lockh = &minfo->mi_lockh;
1227 obddev = class_exp2obd(exp);
1229 mdc_exit_request(&obddev->u.cli);
1230 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_GETATTR_ENQUEUE))
1233 rc = ldlm_cli_enqueue_fini(exp, req, einfo->ei_type, 1, einfo->ei_mode,
1234 &flags, NULL, 0, lockh, rc);
1236 CERROR("ldlm_cli_enqueue_fini: %d\n", rc);
1237 mdc_clear_replay_flag(req, rc);
1241 lockrep = req_capsule_server_get(&req->rq_pill, &RMF_DLM_REP);
1242 LASSERT(lockrep != NULL);
1244 lockrep->lock_policy_res2 =
1245 ptlrpc_status_ntoh(lockrep->lock_policy_res2);
1247 rc = mdc_finish_enqueue(exp, req, einfo, it, lockh, rc);
1251 rc = mdc_finish_intent_lock(exp, req, &minfo->mi_data, it, lockh);
1255 minfo->mi_cb(req, minfo, rc);
1259 int mdc_intent_getattr_async(struct obd_export *exp,
1260 struct md_enqueue_info *minfo,
1261 struct ldlm_enqueue_info *einfo)
1263 struct md_op_data *op_data = &minfo->mi_data;
1264 struct lookup_intent *it = &minfo->mi_it;
1265 struct ptlrpc_request *req;
1266 struct mdc_getattr_args *ga;
1267 struct obd_device *obddev = class_exp2obd(exp);
1268 struct ldlm_res_id res_id;
1269 /*XXX: Both MDS_INODELOCK_LOOKUP and MDS_INODELOCK_UPDATE are needed
1270 * for statahead currently. Consider CMD in future, such two bits
1271 * maybe managed by different MDS, should be adjusted then. */
1272 ldlm_policy_data_t policy = {
1273 .l_inodebits = { MDS_INODELOCK_LOOKUP |
1274 MDS_INODELOCK_UPDATE }
1277 __u64 flags = LDLM_FL_HAS_INTENT;
1280 "name: %.*s in inode "DFID", intent: %s flags %#Lo\n",
1281 op_data->op_namelen, op_data->op_name, PFID(&op_data->op_fid1),
1282 ldlm_it2str(it->it_op), it->it_flags);
1284 fid_build_reg_res_name(&op_data->op_fid1, &res_id);
1285 req = mdc_intent_getattr_pack(exp, it, op_data);
1287 return PTR_ERR(req);
1289 rc = mdc_enter_request(&obddev->u.cli);
1291 ptlrpc_req_finished(req);
1295 rc = ldlm_cli_enqueue(exp, &req, einfo, &res_id, &policy, &flags, NULL,
1296 0, LVB_T_NONE, &minfo->mi_lockh, 1);
1298 mdc_exit_request(&obddev->u.cli);
1299 ptlrpc_req_finished(req);
1303 CLASSERT(sizeof(*ga) <= sizeof(req->rq_async_args));
1304 ga = ptlrpc_req_async_args(req);
1306 ga->ga_minfo = minfo;
1307 ga->ga_einfo = einfo;
1309 req->rq_interpret_reply = mdc_intent_getattr_async_interpret;
1310 ptlrpcd_add_req(req, PDL_POLICY_LOCAL, -1);