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, 2014, 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 <obd_class.h>
42 #include <lustre_dlm.h>
43 #include <lustre_fid.h> /* fid_res_name_eq() */
44 #include <lustre_intent.h>
45 #include <lustre_mdc.h>
46 #include <lustre_net.h>
47 #include <lustre_req_layout.h>
48 #include "mdc_internal.h"
50 struct mdc_getattr_args {
51 struct obd_export *ga_exp;
52 struct md_enqueue_info *ga_minfo;
53 struct ldlm_enqueue_info *ga_einfo;
56 int it_open_error(int phase, struct lookup_intent *it)
58 if (it_disposition(it, DISP_OPEN_LEASE)) {
59 if (phase >= DISP_OPEN_LEASE)
60 return it->d.lustre.it_status;
64 if (it_disposition(it, DISP_OPEN_OPEN)) {
65 if (phase >= DISP_OPEN_OPEN)
66 return it->d.lustre.it_status;
71 if (it_disposition(it, DISP_OPEN_CREATE)) {
72 if (phase >= DISP_OPEN_CREATE)
73 return it->d.lustre.it_status;
78 if (it_disposition(it, DISP_LOOKUP_EXECD)) {
79 if (phase >= DISP_LOOKUP_EXECD)
80 return it->d.lustre.it_status;
85 if (it_disposition(it, DISP_IT_EXECD)) {
86 if (phase >= DISP_IT_EXECD)
87 return it->d.lustre.it_status;
91 CERROR("it disp: %X, status: %d\n", it->d.lustre.it_disposition,
92 it->d.lustre.it_status);
96 EXPORT_SYMBOL(it_open_error);
98 /* this must be called on a lockh that is known to have a referenced lock */
99 int mdc_set_lock_data(struct obd_export *exp, __u64 *lockh, void *data,
102 struct ldlm_lock *lock;
103 struct inode *new_inode = data;
112 lock = ldlm_handle2lock((struct lustre_handle *)lockh);
114 LASSERT(lock != NULL);
115 lock_res_and_lock(lock);
116 if (lock->l_resource->lr_lvb_inode &&
117 lock->l_resource->lr_lvb_inode != data) {
118 struct inode *old_inode = lock->l_resource->lr_lvb_inode;
119 LASSERTF(old_inode->i_state & I_FREEING,
120 "Found existing inode %p/%lu/%u state %lu in lock: "
121 "setting data to %p/%lu/%u\n", old_inode,
122 old_inode->i_ino, old_inode->i_generation,
124 new_inode, new_inode->i_ino, new_inode->i_generation);
126 lock->l_resource->lr_lvb_inode = new_inode;
128 *bits = lock->l_policy_data.l_inodebits.bits;
130 unlock_res_and_lock(lock);
136 ldlm_mode_t mdc_lock_match(struct obd_export *exp, __u64 flags,
137 const struct lu_fid *fid, ldlm_type_t type,
138 ldlm_policy_data_t *policy, ldlm_mode_t mode,
139 struct lustre_handle *lockh)
141 struct ldlm_res_id res_id;
145 fid_build_reg_res_name(fid, &res_id);
146 /* LU-4405: Clear bits not supported by server */
147 policy->l_inodebits.bits &= exp_connect_ibits(exp);
148 rc = ldlm_lock_match(class_exp2obd(exp)->obd_namespace, flags,
149 &res_id, type, policy, mode, lockh, 0);
153 int mdc_cancel_unused(struct obd_export *exp,
154 const struct lu_fid *fid,
155 ldlm_policy_data_t *policy,
157 ldlm_cancel_flags_t flags,
160 struct ldlm_res_id res_id;
161 struct obd_device *obd = class_exp2obd(exp);
166 fid_build_reg_res_name(fid, &res_id);
167 rc = ldlm_cli_cancel_unused_resource(obd->obd_namespace, &res_id,
168 policy, mode, flags, opaque);
172 int mdc_null_inode(struct obd_export *exp,
173 const struct lu_fid *fid)
175 struct ldlm_res_id res_id;
176 struct ldlm_resource *res;
177 struct ldlm_namespace *ns = class_exp2obd(exp)->obd_namespace;
180 LASSERTF(ns != NULL, "no namespace passed\n");
182 fid_build_reg_res_name(fid, &res_id);
184 res = ldlm_resource_get(ns, NULL, &res_id, 0, 0);
189 res->lr_lvb_inode = NULL;
192 ldlm_resource_putref(res);
196 /* find any ldlm lock of the inode in mdc
200 int mdc_find_cbdata(struct obd_export *exp,
201 const struct lu_fid *fid,
202 ldlm_iterator_t it, void *data)
204 struct ldlm_res_id res_id;
208 fid_build_reg_res_name((struct lu_fid*)fid, &res_id);
209 rc = ldlm_resource_iterate(class_exp2obd(exp)->obd_namespace, &res_id,
211 if (rc == LDLM_ITER_STOP)
213 else if (rc == LDLM_ITER_CONTINUE)
218 static inline void mdc_clear_replay_flag(struct ptlrpc_request *req, int rc)
220 /* Don't hold error requests for replay. */
221 if (req->rq_replay) {
222 spin_lock(&req->rq_lock);
224 spin_unlock(&req->rq_lock);
226 if (rc && req->rq_transno != 0) {
227 DEBUG_REQ(D_ERROR, req, "transno returned on error rc %d", rc);
232 /* Save a large LOV EA into the request buffer so that it is available
233 * for replay. We don't do this in the initial request because the
234 * original request doesn't need this buffer (at most it sends just the
235 * lov_mds_md) and it is a waste of RAM/bandwidth to send the empty
236 * buffer and may also be difficult to allocate and save a very large
237 * request buffer for each open. (bug 5707)
239 * OOM here may cause recovery failure if lmm is needed (only for the
240 * original open if the MDS crashed just when this client also OOM'd)
241 * but this is incredibly unlikely, and questionable whether the client
242 * could do MDS recovery under OOM anyways... */
243 static void mdc_realloc_openmsg(struct ptlrpc_request *req,
244 struct mdt_body *body)
248 /* FIXME: remove this explicit offset. */
249 rc = sptlrpc_cli_enlarge_reqbuf(req, DLM_INTENT_REC_OFF + 4,
250 body->mbo_eadatasize);
252 CERROR("Can't enlarge segment %d size to %d\n",
253 DLM_INTENT_REC_OFF + 4, body->mbo_eadatasize);
254 body->mbo_valid &= ~OBD_MD_FLEASIZE;
255 body->mbo_eadatasize = 0;
259 static struct ptlrpc_request *
260 mdc_intent_open_pack(struct obd_export *exp, struct lookup_intent *it,
261 struct md_op_data *op_data)
263 struct ptlrpc_request *req;
264 struct obd_device *obddev = class_exp2obd(exp);
265 struct ldlm_intent *lit;
266 const void *lmm = op_data->op_data;
267 __u32 lmmsize = op_data->op_data_size;
268 struct list_head cancels = LIST_HEAD_INIT(cancels);
274 it->it_create_mode = (it->it_create_mode & ~S_IFMT) | S_IFREG;
276 /* XXX: openlock is not cancelled for cross-refs. */
277 /* If inode is known, cancel conflicting OPEN locks. */
278 if (fid_is_sane(&op_data->op_fid2)) {
279 if (it->it_flags & MDS_OPEN_LEASE) { /* try to get lease */
280 if (it->it_flags & FMODE_WRITE)
285 if (it->it_flags & (FMODE_WRITE|MDS_OPEN_TRUNC))
288 else if (it->it_flags & FMODE_EXEC)
294 count = mdc_resource_get_unused(exp, &op_data->op_fid2,
299 /* If CREATE, cancel parent's UPDATE lock. */
300 if (it->it_op & IT_CREAT)
304 count += mdc_resource_get_unused(exp, &op_data->op_fid1,
306 MDS_INODELOCK_UPDATE);
308 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
309 &RQF_LDLM_INTENT_OPEN);
311 ldlm_lock_list_put(&cancels, l_bl_ast, count);
312 RETURN(ERR_PTR(-ENOMEM));
315 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
316 op_data->op_namelen + 1);
317 req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT,
318 max(lmmsize, obddev->u.cli.cl_default_mds_easize));
320 rc = ldlm_prep_enqueue_req(exp, req, &cancels, count);
322 ptlrpc_request_free(req);
326 spin_lock(&req->rq_lock);
327 req->rq_replay = req->rq_import->imp_replayable;
328 spin_unlock(&req->rq_lock);
330 /* pack the intent */
331 lit = req_capsule_client_get(&req->rq_pill, &RMF_LDLM_INTENT);
332 lit->opc = (__u64)it->it_op;
334 /* pack the intended request */
335 mdc_open_pack(req, op_data, it->it_create_mode, 0, it->it_flags, lmm,
338 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
339 obddev->u.cli.cl_max_mds_easize);
341 /* for remote client, fetch remote perm for current user */
342 if (client_is_remote(exp))
343 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
344 sizeof(struct mdt_remote_perm));
345 ptlrpc_request_set_replen(req);
349 static struct ptlrpc_request *
350 mdc_intent_getxattr_pack(struct obd_export *exp,
351 struct lookup_intent *it,
352 struct md_op_data *op_data)
354 struct ptlrpc_request *req;
355 struct ldlm_intent *lit;
358 struct list_head cancels = LIST_HEAD_INIT(cancels);
362 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
363 &RQF_LDLM_INTENT_GETXATTR);
365 RETURN(ERR_PTR(-ENOMEM));
367 rc = ldlm_prep_enqueue_req(exp, req, &cancels, count);
369 ptlrpc_request_free(req);
373 /* pack the intent */
374 lit = req_capsule_client_get(&req->rq_pill, &RMF_LDLM_INTENT);
375 lit->opc = IT_GETXATTR;
377 maxdata = class_exp2cliimp(exp)->imp_connect_data.ocd_max_easize;
379 /* pack the intended request */
380 mdc_pack_body(req, &op_data->op_fid1, op_data->op_valid, maxdata, -1,
383 req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
384 RCL_SERVER, maxdata);
386 req_capsule_set_size(&req->rq_pill, &RMF_EAVALS,
387 RCL_SERVER, maxdata);
389 req_capsule_set_size(&req->rq_pill, &RMF_EAVALS_LENS,
390 RCL_SERVER, maxdata);
392 ptlrpc_request_set_replen(req);
397 static struct ptlrpc_request *mdc_intent_unlink_pack(struct obd_export *exp,
398 struct lookup_intent *it,
399 struct md_op_data *op_data)
401 struct ptlrpc_request *req;
402 struct obd_device *obddev = class_exp2obd(exp);
403 struct ldlm_intent *lit;
407 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
408 &RQF_LDLM_INTENT_UNLINK);
410 RETURN(ERR_PTR(-ENOMEM));
412 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
413 op_data->op_namelen + 1);
415 rc = ldlm_prep_enqueue_req(exp, req, NULL, 0);
417 ptlrpc_request_free(req);
421 /* pack the intent */
422 lit = req_capsule_client_get(&req->rq_pill, &RMF_LDLM_INTENT);
423 lit->opc = (__u64)it->it_op;
425 /* pack the intended request */
426 mdc_unlink_pack(req, op_data);
428 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
429 obddev->u.cli.cl_default_mds_easize);
430 ptlrpc_request_set_replen(req);
434 static struct ptlrpc_request *mdc_intent_getattr_pack(struct obd_export *exp,
435 struct lookup_intent *it,
436 struct md_op_data *op_data)
438 struct ptlrpc_request *req;
439 struct obd_device *obddev = class_exp2obd(exp);
440 u64 valid = OBD_MD_FLGETATTR | OBD_MD_FLEASIZE |
441 OBD_MD_FLMODEASIZE | OBD_MD_FLDIREA |
443 (client_is_remote(exp) ?
444 OBD_MD_FLRMTPERM : OBD_MD_FLACL);
445 struct ldlm_intent *lit;
450 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
451 &RQF_LDLM_INTENT_GETATTR);
453 RETURN(ERR_PTR(-ENOMEM));
455 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
456 op_data->op_namelen + 1);
458 rc = ldlm_prep_enqueue_req(exp, req, NULL, 0);
460 ptlrpc_request_free(req);
464 /* pack the intent */
465 lit = req_capsule_client_get(&req->rq_pill, &RMF_LDLM_INTENT);
466 lit->opc = (__u64)it->it_op;
468 if (obddev->u.cli.cl_default_mds_easize > 0)
469 easize = obddev->u.cli.cl_default_mds_easize;
471 easize = obddev->u.cli.cl_max_mds_easize;
473 /* pack the intended request */
474 mdc_getattr_pack(req, valid, it->it_flags, op_data, easize);
476 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER, easize);
477 if (client_is_remote(exp))
478 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
479 sizeof(struct mdt_remote_perm));
480 ptlrpc_request_set_replen(req);
484 static struct ptlrpc_request *mdc_intent_layout_pack(struct obd_export *exp,
485 struct lookup_intent *it,
486 struct md_op_data *unused)
488 struct obd_device *obd = class_exp2obd(exp);
489 struct ptlrpc_request *req;
490 struct ldlm_intent *lit;
491 struct layout_intent *layout;
495 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
496 &RQF_LDLM_INTENT_LAYOUT);
498 RETURN(ERR_PTR(-ENOMEM));
500 req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT, 0);
501 rc = ldlm_prep_enqueue_req(exp, req, NULL, 0);
503 ptlrpc_request_free(req);
507 /* pack the intent */
508 lit = req_capsule_client_get(&req->rq_pill, &RMF_LDLM_INTENT);
509 lit->opc = (__u64)it->it_op;
511 /* pack the layout intent request */
512 layout = req_capsule_client_get(&req->rq_pill, &RMF_LAYOUT_INTENT);
513 /* LAYOUT_INTENT_ACCESS is generic, specific operation will be
514 * set for replication */
515 layout->li_opc = LAYOUT_INTENT_ACCESS;
517 req_capsule_set_size(&req->rq_pill, &RMF_DLM_LVB, RCL_SERVER,
518 obd->u.cli.cl_default_mds_easize);
519 ptlrpc_request_set_replen(req);
523 static struct ptlrpc_request *
524 mdc_enqueue_pack(struct obd_export *exp, int lvb_len)
526 struct ptlrpc_request *req;
530 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LDLM_ENQUEUE);
532 RETURN(ERR_PTR(-ENOMEM));
534 rc = ldlm_prep_enqueue_req(exp, req, NULL, 0);
536 ptlrpc_request_free(req);
540 req_capsule_set_size(&req->rq_pill, &RMF_DLM_LVB, RCL_SERVER, lvb_len);
541 ptlrpc_request_set_replen(req);
545 static int mdc_finish_enqueue(struct obd_export *exp,
546 struct ptlrpc_request *req,
547 struct ldlm_enqueue_info *einfo,
548 struct lookup_intent *it,
549 struct lustre_handle *lockh,
552 struct req_capsule *pill = &req->rq_pill;
553 struct ldlm_request *lockreq;
554 struct ldlm_reply *lockrep;
555 struct lustre_intent_data *intent = &it->d.lustre;
556 struct ldlm_lock *lock;
557 void *lvb_data = NULL;
562 /* Similarly, if we're going to replay this request, we don't want to
563 * actually get a lock, just perform the intent. */
564 if (req->rq_transno || req->rq_replay) {
565 lockreq = req_capsule_client_get(pill, &RMF_DLM_REQ);
566 lockreq->lock_flags |= ldlm_flags_to_wire(LDLM_FL_INTENT_ONLY);
569 if (rc == ELDLM_LOCK_ABORTED) {
571 memset(lockh, 0, sizeof(*lockh));
573 } else { /* rc = 0 */
574 lock = ldlm_handle2lock(lockh);
575 LASSERT(lock != NULL);
577 /* If the server gave us back a different lock mode, we should
578 * fix up our variables. */
579 if (lock->l_req_mode != einfo->ei_mode) {
580 ldlm_lock_addref(lockh, lock->l_req_mode);
581 ldlm_lock_decref(lockh, einfo->ei_mode);
582 einfo->ei_mode = lock->l_req_mode;
587 lockrep = req_capsule_server_get(pill, &RMF_DLM_REP);
588 LASSERT(lockrep != NULL); /* checked by ldlm_cli_enqueue() */
590 intent->it_disposition = (int)lockrep->lock_policy_res1;
591 intent->it_status = (int)lockrep->lock_policy_res2;
592 intent->it_lock_mode = einfo->ei_mode;
593 intent->it_lock_handle = lockh->cookie;
594 intent->it_data = req;
596 /* Technically speaking rq_transno must already be zero if
597 * it_status is in error, so the check is a bit redundant */
598 if ((!req->rq_transno || intent->it_status < 0) && req->rq_replay)
599 mdc_clear_replay_flag(req, intent->it_status);
601 /* If we're doing an IT_OPEN which did not result in an actual
602 * successful open, then we need to remove the bit which saves
603 * this request for unconditional replay.
605 * It's important that we do this first! Otherwise we might exit the
606 * function without doing so, and try to replay a failed create
608 if (it->it_op & IT_OPEN && req->rq_replay &&
609 (!it_disposition(it, DISP_OPEN_OPEN) ||intent->it_status != 0))
610 mdc_clear_replay_flag(req, intent->it_status);
612 DEBUG_REQ(D_RPCTRACE, req, "op: %d disposition: %x, status: %d",
613 it->it_op, intent->it_disposition, intent->it_status);
615 /* We know what to expect, so we do any byte flipping required here */
616 if (it->it_op & (IT_OPEN | IT_UNLINK | IT_LOOKUP | IT_GETATTR)) {
617 struct mdt_body *body;
619 body = req_capsule_server_get(pill, &RMF_MDT_BODY);
621 CERROR ("Can't swab mdt_body\n");
625 if (it_disposition(it, DISP_OPEN_OPEN) &&
626 !it_open_error(DISP_OPEN_OPEN, it)) {
628 * If this is a successful OPEN request, we need to set
629 * replay handler and data early, so that if replay
630 * happens immediately after swabbing below, new reply
631 * is swabbed by that handler correctly.
633 mdc_set_open_replay_data(NULL, NULL, it);
636 if (body->mbo_valid & (OBD_MD_FLDIREA | OBD_MD_FLEASIZE)) {
639 mdc_update_max_ea_from_body(exp, body);
642 * The eadata is opaque; just check that it is there.
643 * Eventually, obd_unpackmd() will check the contents.
645 eadata = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
646 body->mbo_eadatasize);
650 /* save lvb data and length in case this is for layout
653 lvb_len = body->mbo_eadatasize;
656 * We save the reply LOV EA in case we have to replay a
657 * create for recovery. If we didn't allocate a large
658 * enough request buffer above we need to reallocate it
659 * here to hold the actual LOV EA.
661 * To not save LOV EA if request is not going to replay
662 * (for example error one).
664 if ((it->it_op & IT_OPEN) && req->rq_replay) {
666 if (req_capsule_get_size(pill, &RMF_EADATA,
668 body->mbo_eadatasize)
669 mdc_realloc_openmsg(req, body);
671 req_capsule_shrink(pill, &RMF_EADATA,
672 body->mbo_eadatasize,
675 req_capsule_set_size(pill, &RMF_EADATA,
677 body->mbo_eadatasize);
679 lmm = req_capsule_client_get(pill, &RMF_EADATA);
682 body->mbo_eadatasize);
686 if (body->mbo_valid & OBD_MD_FLRMTPERM) {
687 struct mdt_remote_perm *perm;
689 LASSERT(client_is_remote(exp));
690 perm = req_capsule_server_swab_get(pill, &RMF_ACL,
691 lustre_swab_mdt_remote_perm);
695 } else if (it->it_op & IT_LAYOUT) {
696 /* maybe the lock was granted right away and layout
697 * is packed into RMF_DLM_LVB of req */
698 lvb_len = req_capsule_get_size(pill, &RMF_DLM_LVB, RCL_SERVER);
700 lvb_data = req_capsule_server_sized_get(pill,
701 &RMF_DLM_LVB, lvb_len);
702 if (lvb_data == NULL)
707 /* fill in stripe data for layout lock.
708 * LU-6581: trust layout data only if layout lock is granted. The MDT
709 * has stopped sending layout unless the layout lock is granted. The
710 * client still does this checking in case it's talking with an old
711 * server. - Jinshan */
712 lock = ldlm_handle2lock(lockh);
713 if (lock != NULL && ldlm_has_layout(lock) && lvb_data != NULL &&
714 !(lockrep->lock_flags & LDLM_FL_BLOCKED_MASK)) {
717 LDLM_DEBUG(lock, "layout lock returned by: %s, lvb_len: %d\n",
718 ldlm_it2str(it->it_op), lvb_len);
720 OBD_ALLOC_LARGE(lmm, lvb_len);
725 memcpy(lmm, lvb_data, lvb_len);
727 /* install lvb_data */
728 lock_res_and_lock(lock);
729 if (lock->l_lvb_data == NULL) {
730 lock->l_lvb_type = LVB_T_LAYOUT;
731 lock->l_lvb_data = lmm;
732 lock->l_lvb_len = lvb_len;
735 unlock_res_and_lock(lock);
737 OBD_FREE_LARGE(lmm, lvb_len);
745 /* We always reserve enough space in the reply packet for a stripe MD, because
746 * we don't know in advance the file type. */
747 int mdc_enqueue(struct obd_export *exp,
748 struct ldlm_enqueue_info *einfo,
749 const union ldlm_policy_data *policy,
750 struct lookup_intent *it, struct md_op_data *op_data,
751 struct lustre_handle *lockh, __u64 extra_lock_flags)
753 struct obd_device *obddev = class_exp2obd(exp);
754 struct ptlrpc_request *req = NULL;
755 __u64 flags, saved_flags = extra_lock_flags;
757 struct ldlm_res_id res_id;
758 static const ldlm_policy_data_t lookup_policy =
759 { .l_inodebits = { MDS_INODELOCK_LOOKUP } };
760 static const ldlm_policy_data_t update_policy =
761 { .l_inodebits = { MDS_INODELOCK_UPDATE } };
762 static const ldlm_policy_data_t layout_policy =
763 { .l_inodebits = { MDS_INODELOCK_LAYOUT } };
764 static const ldlm_policy_data_t getxattr_policy = {
765 .l_inodebits = { MDS_INODELOCK_XATTR } };
766 int generation, resends = 0;
767 struct ldlm_reply *lockrep;
768 enum lvb_type lvb_type = 0;
771 LASSERTF(!it || einfo->ei_type == LDLM_IBITS, "lock type %d\n",
773 fid_build_reg_res_name(&op_data->op_fid1, &res_id);
776 LASSERT(policy == NULL);
778 saved_flags |= LDLM_FL_HAS_INTENT;
779 if (it->it_op & (IT_OPEN | IT_UNLINK | IT_GETATTR | IT_READDIR))
780 policy = &update_policy;
781 else if (it->it_op & IT_LAYOUT)
782 policy = &layout_policy;
783 else if (it->it_op & (IT_GETXATTR | IT_SETXATTR))
784 policy = &getxattr_policy;
786 policy = &lookup_policy;
789 generation = obddev->u.cli.cl_import->imp_generation;
793 /* The only way right now is FLOCK. */
794 LASSERTF(einfo->ei_type == LDLM_FLOCK, "lock type %d\n",
796 res_id.name[3] = LDLM_FLOCK;
797 } else if (it->it_op & IT_OPEN) {
798 req = mdc_intent_open_pack(exp, it, op_data);
799 } else if (it->it_op & IT_UNLINK) {
800 req = mdc_intent_unlink_pack(exp, it, op_data);
801 } else if (it->it_op & (IT_GETATTR | IT_LOOKUP)) {
802 req = mdc_intent_getattr_pack(exp, it, op_data);
803 } else if (it->it_op & IT_READDIR) {
804 req = mdc_enqueue_pack(exp, 0);
805 } else if (it->it_op & IT_LAYOUT) {
806 if (!imp_connect_lvb_type(class_exp2cliimp(exp)))
808 req = mdc_intent_layout_pack(exp, it, op_data);
809 lvb_type = LVB_T_LAYOUT;
810 } else if (it->it_op & IT_GETXATTR) {
811 req = mdc_intent_getxattr_pack(exp, it, op_data);
818 RETURN(PTR_ERR(req));
820 if (req != NULL && it && it->it_op & IT_CREAT)
821 /* ask ptlrpc not to resend on EINPROGRESS since we have our own
823 req->rq_no_retry_einprogress = 1;
826 req->rq_generation_set = 1;
827 req->rq_import_generation = generation;
828 req->rq_sent = cfs_time_current_sec() + resends;
831 /* It is important to obtain modify RPC slot first (if applicable), so
832 * that threads that are waiting for a modify RPC slot are not polluting
833 * our rpcs in flight counter.
834 * We do not do flock request limiting, though */
836 mdc_get_mod_rpc_slot(req, it);
837 rc = obd_get_request_slot(&obddev->u.cli);
839 mdc_put_mod_rpc_slot(req, it);
840 mdc_clear_replay_flag(req, 0);
841 ptlrpc_req_finished(req);
846 rc = ldlm_cli_enqueue(exp, &req, einfo, &res_id, policy, &flags, NULL,
847 0, lvb_type, lockh, 0);
849 /* For flock requests we immediatelly return without further
850 delay and let caller deal with the rest, since rest of
851 this function metadata processing makes no sense for flock
852 requests anyway. But in case of problem during comms with
853 Server (ETIMEDOUT) or any signal/kill attempt (EINTR), we
854 can not rely on caller and this mainly for F_UNLCKs
855 (explicits or automatically generated by Kernel to clean
856 current FLocks upon exit) that can't be trashed */
857 if (((rc == -EINTR) || (rc == -ETIMEDOUT)) &&
858 (einfo->ei_type == LDLM_FLOCK) &&
859 (einfo->ei_mode == LCK_NL))
864 obd_put_request_slot(&obddev->u.cli);
865 mdc_put_mod_rpc_slot(req, it);
868 CDEBUG(D_INFO, "%s: ldlm_cli_enqueue failed: rc = %d\n",
869 obddev->obd_name, rc);
871 mdc_clear_replay_flag(req, rc);
872 ptlrpc_req_finished(req);
876 lockrep = req_capsule_server_get(&req->rq_pill, &RMF_DLM_REP);
877 LASSERT(lockrep != NULL);
879 lockrep->lock_policy_res2 =
880 ptlrpc_status_ntoh(lockrep->lock_policy_res2);
882 /* Retry the create infinitely when we get -EINPROGRESS from
883 * server. This is required by the new quota design. */
884 if (it && it->it_op & IT_CREAT &&
885 (int)lockrep->lock_policy_res2 == -EINPROGRESS) {
886 mdc_clear_replay_flag(req, rc);
887 ptlrpc_req_finished(req);
890 CDEBUG(D_HA, "%s: resend:%d op:%d "DFID"/"DFID"\n",
891 obddev->obd_name, resends, it->it_op,
892 PFID(&op_data->op_fid1), PFID(&op_data->op_fid2));
894 if (generation == obddev->u.cli.cl_import->imp_generation) {
897 CDEBUG(D_HA, "resend cross eviction\n");
902 rc = mdc_finish_enqueue(exp, req, einfo, it, lockh, rc);
904 if (lustre_handle_is_used(lockh)) {
905 ldlm_lock_decref(lockh, einfo->ei_mode);
906 memset(lockh, 0, sizeof(*lockh));
908 ptlrpc_req_finished(req);
910 it->d.lustre.it_lock_handle = 0;
911 it->d.lustre.it_lock_mode = 0;
912 it->d.lustre.it_data = NULL;
918 static int mdc_finish_intent_lock(struct obd_export *exp,
919 struct ptlrpc_request *request,
920 struct md_op_data *op_data,
921 struct lookup_intent *it,
922 struct lustre_handle *lockh)
924 struct lustre_handle old_lock;
925 struct mdt_body *mdt_body;
926 struct ldlm_lock *lock;
930 LASSERT(request != NULL);
931 LASSERT(request != LP_POISON);
932 LASSERT(request->rq_repmsg != LP_POISON);
934 if (it->it_op & IT_READDIR)
937 if (!it_disposition(it, DISP_IT_EXECD)) {
938 /* The server failed before it even started executing the
939 * intent, i.e. because it couldn't unpack the request. */
940 LASSERT(it->d.lustre.it_status != 0);
941 RETURN(it->d.lustre.it_status);
943 rc = it_open_error(DISP_IT_EXECD, it);
947 mdt_body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
948 LASSERT(mdt_body != NULL); /* mdc_enqueue checked */
950 rc = it_open_error(DISP_LOOKUP_EXECD, it);
954 /* keep requests around for the multiple phases of the call
955 * this shows the DISP_XX must guarantee we make it into the call
957 if (!it_disposition(it, DISP_ENQ_CREATE_REF) &&
958 it_disposition(it, DISP_OPEN_CREATE) &&
959 !it_open_error(DISP_OPEN_CREATE, it)) {
960 it_set_disposition(it, DISP_ENQ_CREATE_REF);
961 ptlrpc_request_addref(request); /* balanced in ll_create_node */
963 if (!it_disposition(it, DISP_ENQ_OPEN_REF) &&
964 it_disposition(it, DISP_OPEN_OPEN) &&
965 !it_open_error(DISP_OPEN_OPEN, it)) {
966 it_set_disposition(it, DISP_ENQ_OPEN_REF);
967 ptlrpc_request_addref(request); /* balanced in ll_file_open */
968 /* BUG 11546 - eviction in the middle of open rpc processing */
969 OBD_FAIL_TIMEOUT(OBD_FAIL_MDC_ENQUEUE_PAUSE, obd_timeout);
972 if (it->it_op & IT_CREAT) {
973 /* XXX this belongs in ll_create_it */
974 } else if (it->it_op == IT_OPEN) {
975 LASSERT(!it_disposition(it, DISP_OPEN_CREATE));
977 LASSERT(it->it_op & (IT_GETATTR | IT_LOOKUP | IT_LAYOUT));
980 /* If we already have a matching lock, then cancel the new
981 * one. We have to set the data here instead of in
982 * mdc_enqueue, because we need to use the child's inode as
983 * the l_ast_data to match, and that's not available until
984 * intent_finish has performed the iget().) */
985 lock = ldlm_handle2lock(lockh);
987 ldlm_policy_data_t policy = lock->l_policy_data;
988 LDLM_DEBUG(lock, "matching against this");
990 LASSERTF(fid_res_name_eq(&mdt_body->mbo_fid1,
991 &lock->l_resource->lr_name),
992 "Lock res_id: "DLDLMRES", fid: "DFID"\n",
993 PLDLMRES(lock->l_resource), PFID(&mdt_body->mbo_fid1));
996 memcpy(&old_lock, lockh, sizeof(*lockh));
997 if (ldlm_lock_match(NULL, LDLM_FL_BLOCK_GRANTED, NULL,
998 LDLM_IBITS, &policy, LCK_NL, &old_lock, 0)) {
999 ldlm_lock_decref_and_cancel(lockh,
1000 it->d.lustre.it_lock_mode);
1001 memcpy(lockh, &old_lock, sizeof(old_lock));
1002 it->d.lustre.it_lock_handle = lockh->cookie;
1005 CDEBUG(D_DENTRY,"D_IT dentry %.*s intent: %s status %d disp %x rc %d\n",
1006 (int)op_data->op_namelen, op_data->op_name,
1007 ldlm_it2str(it->it_op), it->d.lustre.it_status,
1008 it->d.lustre.it_disposition, rc);
1012 int mdc_revalidate_lock(struct obd_export *exp, struct lookup_intent *it,
1013 struct lu_fid *fid, __u64 *bits)
1015 /* We could just return 1 immediately, but since we should only
1016 * be called in revalidate_it if we already have a lock, let's
1018 struct ldlm_res_id res_id;
1019 struct lustre_handle lockh;
1020 ldlm_policy_data_t policy;
1024 if (it->d.lustre.it_lock_handle) {
1025 lockh.cookie = it->d.lustre.it_lock_handle;
1026 mode = ldlm_revalidate_lock_handle(&lockh, bits);
1028 fid_build_reg_res_name(fid, &res_id);
1029 switch (it->it_op) {
1031 /* File attributes are held under multiple bits:
1032 * nlink is under lookup lock, size and times are
1033 * under UPDATE lock and recently we've also got
1034 * a separate permissions lock for owner/group/acl that
1035 * were protected by lookup lock before.
1036 * Getattr must provide all of that information,
1037 * so we need to ensure we have all of those locks.
1038 * Unfortunately, if the bits are split across multiple
1039 * locks, there's no easy way to match all of them here,
1040 * so an extra RPC would be performed to fetch all
1041 * of those bits at once for now. */
1042 /* For new MDTs(> 2.4), UPDATE|PERM should be enough,
1043 * but for old MDTs (< 2.4), permission is covered
1044 * by LOOKUP lock, so it needs to match all bits here.*/
1045 policy.l_inodebits.bits = MDS_INODELOCK_UPDATE |
1046 MDS_INODELOCK_LOOKUP |
1050 policy.l_inodebits.bits = MDS_INODELOCK_UPDATE;
1053 policy.l_inodebits.bits = MDS_INODELOCK_LAYOUT;
1056 policy.l_inodebits.bits = MDS_INODELOCK_LOOKUP;
1060 mode = mdc_lock_match(exp, LDLM_FL_BLOCK_GRANTED, fid,
1061 LDLM_IBITS, &policy,
1062 LCK_CR | LCK_CW | LCK_PR | LCK_PW,
1067 it->d.lustre.it_lock_handle = lockh.cookie;
1068 it->d.lustre.it_lock_mode = mode;
1070 it->d.lustre.it_lock_handle = 0;
1071 it->d.lustre.it_lock_mode = 0;
1078 * This long block is all about fixing up the lock and request state
1079 * so that it is correct as of the moment _before_ the operation was
1080 * applied; that way, the VFS will think that everything is normal and
1081 * call Lustre's regular VFS methods.
1083 * If we're performing a creation, that means that unless the creation
1084 * failed with EEXIST, we should fake up a negative dentry.
1086 * For everything else, we want to lookup to succeed.
1088 * One additional note: if CREATE or OPEN succeeded, we add an extra
1089 * reference to the request because we need to keep it around until
1090 * ll_create/ll_open gets called.
1092 * The server will return to us, in it_disposition, an indication of
1093 * exactly what d.lustre.it_status refers to.
1095 * If DISP_OPEN_OPEN is set, then d.lustre.it_status refers to the open() call,
1096 * otherwise if DISP_OPEN_CREATE is set, then it status is the
1097 * creation failure mode. In either case, one of DISP_LOOKUP_NEG or
1098 * DISP_LOOKUP_POS will be set, indicating whether the child lookup
1101 * Else, if DISP_LOOKUP_EXECD then d.lustre.it_status is the rc of the
1104 int mdc_intent_lock(struct obd_export *exp, struct md_op_data *op_data,
1105 struct lookup_intent *it, struct ptlrpc_request **reqp,
1106 ldlm_blocking_callback cb_blocking, __u64 extra_lock_flags)
1108 struct ldlm_enqueue_info einfo = {
1109 .ei_type = LDLM_IBITS,
1110 .ei_mode = it_to_lock_mode(it),
1111 .ei_cb_bl = cb_blocking,
1112 .ei_cb_cp = ldlm_completion_ast,
1114 struct lustre_handle lockh;
1119 CDEBUG(D_DLMTRACE, "(name: %.*s,"DFID") in obj "DFID
1120 ", intent: %s flags %#"LPF64"o\n", (int)op_data->op_namelen,
1121 op_data->op_name, PFID(&op_data->op_fid2),
1122 PFID(&op_data->op_fid1), ldlm_it2str(it->it_op),
1126 if (fid_is_sane(&op_data->op_fid2) &&
1127 (it->it_op & (IT_LOOKUP | IT_GETATTR | IT_READDIR))) {
1128 /* We could just return 1 immediately, but since we should only
1129 * be called in revalidate_it if we already have a lock, let's
1131 it->d.lustre.it_lock_handle = 0;
1132 rc = mdc_revalidate_lock(exp, it, &op_data->op_fid2, NULL);
1133 /* Only return failure if it was not GETATTR by cfid
1134 (from inode_revalidate) */
1135 if (rc || op_data->op_namelen != 0)
1139 /* For case if upper layer did not alloc fid, do it now. */
1140 if (!fid_is_sane(&op_data->op_fid2) && it->it_op & IT_CREAT) {
1141 rc = mdc_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
1143 CERROR("Can't alloc new fid, rc %d\n", rc);
1148 rc = mdc_enqueue(exp, &einfo, NULL, it, op_data, &lockh,
1153 *reqp = it->d.lustre.it_data;
1154 rc = mdc_finish_intent_lock(exp, *reqp, op_data, it, &lockh);
1158 static int mdc_intent_getattr_async_interpret(const struct lu_env *env,
1159 struct ptlrpc_request *req,
1162 struct mdc_getattr_args *ga = args;
1163 struct obd_export *exp = ga->ga_exp;
1164 struct md_enqueue_info *minfo = ga->ga_minfo;
1165 struct ldlm_enqueue_info *einfo = ga->ga_einfo;
1166 struct lookup_intent *it;
1167 struct lustre_handle *lockh;
1168 struct obd_device *obddev;
1169 struct ldlm_reply *lockrep;
1170 __u64 flags = LDLM_FL_HAS_INTENT;
1174 lockh = &minfo->mi_lockh;
1176 obddev = class_exp2obd(exp);
1178 obd_put_request_slot(&obddev->u.cli);
1179 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_GETATTR_ENQUEUE))
1182 rc = ldlm_cli_enqueue_fini(exp, req, einfo->ei_type, 1, einfo->ei_mode,
1183 &flags, NULL, 0, lockh, rc);
1185 CERROR("ldlm_cli_enqueue_fini: %d\n", rc);
1186 mdc_clear_replay_flag(req, rc);
1190 lockrep = req_capsule_server_get(&req->rq_pill, &RMF_DLM_REP);
1191 LASSERT(lockrep != NULL);
1193 lockrep->lock_policy_res2 =
1194 ptlrpc_status_ntoh(lockrep->lock_policy_res2);
1196 rc = mdc_finish_enqueue(exp, req, einfo, it, lockh, rc);
1200 rc = mdc_finish_intent_lock(exp, req, &minfo->mi_data, it, lockh);
1204 OBD_FREE_PTR(einfo);
1205 minfo->mi_cb(req, minfo, rc);
1209 int mdc_intent_getattr_async(struct obd_export *exp,
1210 struct md_enqueue_info *minfo,
1211 struct ldlm_enqueue_info *einfo)
1213 struct md_op_data *op_data = &minfo->mi_data;
1214 struct lookup_intent *it = &minfo->mi_it;
1215 struct ptlrpc_request *req;
1216 struct mdc_getattr_args *ga;
1217 struct obd_device *obddev = class_exp2obd(exp);
1218 struct ldlm_res_id res_id;
1219 /*XXX: Both MDS_INODELOCK_LOOKUP and MDS_INODELOCK_UPDATE are needed
1220 * for statahead currently. Consider CMD in future, such two bits
1221 * maybe managed by different MDS, should be adjusted then. */
1222 ldlm_policy_data_t policy = {
1223 .l_inodebits = { MDS_INODELOCK_LOOKUP |
1224 MDS_INODELOCK_UPDATE }
1227 __u64 flags = LDLM_FL_HAS_INTENT;
1230 CDEBUG(D_DLMTRACE, "name: %.*s in inode "DFID", intent: %s flags %#"
1232 (int)op_data->op_namelen, op_data->op_name,
1233 PFID(&op_data->op_fid1), ldlm_it2str(it->it_op), it->it_flags);
1235 fid_build_reg_res_name(&op_data->op_fid1, &res_id);
1236 req = mdc_intent_getattr_pack(exp, it, op_data);
1238 RETURN(PTR_ERR(req));
1240 rc = obd_get_request_slot(&obddev->u.cli);
1242 ptlrpc_req_finished(req);
1246 rc = ldlm_cli_enqueue(exp, &req, einfo, &res_id, &policy, &flags, NULL,
1247 0, LVB_T_NONE, &minfo->mi_lockh, 1);
1249 obd_put_request_slot(&obddev->u.cli);
1250 ptlrpc_req_finished(req);
1254 CLASSERT(sizeof(*ga) <= sizeof(req->rq_async_args));
1255 ga = ptlrpc_req_async_args(req);
1257 ga->ga_minfo = minfo;
1258 ga->ga_einfo = einfo;
1260 req->rq_interpret_reply = mdc_intent_getattr_async_interpret;
1261 ptlrpcd_add_req(req);