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.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2001, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2016, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
33 #define DEBUG_SUBSYSTEM S_MDC
35 #include <linux/init.h>
36 #include <linux/kthread.h>
37 #include <linux/module.h>
38 #include <linux/pagemap.h>
39 #include <linux/user_namespace.h>
40 #include <linux/utsname.h>
41 #ifdef HAVE_UIDGID_HEADER
42 # include <linux/uidgid.h>
45 #include <lustre/lustre_errno.h>
47 #include <cl_object.h>
48 #include <llog_swab.h>
49 #include <lprocfs_status.h>
50 #include <lustre_acl.h>
51 #include <lustre_fid.h>
52 #include <lustre_ioctl.h>
53 #include <lustre_kernelcomm.h>
54 #include <lustre_lmv.h>
55 #include <lustre_log.h>
56 #include <lustre_param.h>
57 #include <lustre_swab.h>
58 #include <obd_class.h>
60 #include "mdc_internal.h"
62 #define REQUEST_MINOR 244
64 static int mdc_cleanup(struct obd_device *obd);
66 static inline int mdc_queue_wait(struct ptlrpc_request *req)
68 struct client_obd *cli = &req->rq_import->imp_obd->u.cli;
71 /* obd_get_request_slot() ensures that this client has no more
72 * than cl_max_rpcs_in_flight RPCs simultaneously inf light
74 rc = obd_get_request_slot(cli);
78 rc = ptlrpc_queue_wait(req);
79 obd_put_request_slot(cli);
85 * Send MDS_GET_ROOT RPC to fetch root FID.
87 * If \a fileset is not NULL it should contain a subdirectory off
88 * the ROOT/ directory to be mounted on the client. Return the FID
89 * of the subdirectory to the client to mount onto its mountpoint.
91 * \param[in] imp MDC import
92 * \param[in] fileset fileset name, which could be NULL
93 * \param[out] rootfid root FID of this mountpoint
94 * \param[out] pc root capa will be unpacked and saved in this pointer
96 * \retval 0 on success, negative errno on failure
98 static int mdc_get_root(struct obd_export *exp, const char *fileset,
99 struct lu_fid *rootfid)
101 struct ptlrpc_request *req;
102 struct mdt_body *body;
107 if (fileset && !(exp_connect_flags(exp) & OBD_CONNECT_SUBTREE))
110 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
116 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
117 strlen(fileset) + 1);
118 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GET_ROOT);
120 ptlrpc_request_free(req);
123 mdc_pack_body(req, NULL, 0, 0, -1, 0);
124 if (fileset != NULL) {
125 char *name = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
127 memcpy(name, fileset, strlen(fileset));
129 lustre_msg_add_flags(req->rq_reqmsg, LUSTRE_IMP_FULL);
130 req->rq_send_state = LUSTRE_IMP_FULL;
132 ptlrpc_request_set_replen(req);
134 rc = ptlrpc_queue_wait(req);
138 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
140 GOTO(out, rc = -EPROTO);
142 *rootfid = body->mbo_fid1;
143 CDEBUG(D_NET, "root fid="DFID", last_committed=%llu\n",
144 PFID(rootfid), lustre_msg_get_last_committed(req->rq_repmsg));
147 ptlrpc_req_finished(req);
153 * This function now is known to always saying that it will receive 4 buffers
154 * from server. Even for cases when acl_size and md_size is zero, RPC header
155 * will contain 4 fields and RPC itself will contain zero size fields. This is
156 * because mdt_getattr*() _always_ returns 4 fields, but if acl is not needed
157 * and thus zero, it shrinks it, making zero size. The same story about
158 * md_size. And this is course of problem when client waits for smaller number
159 * of fields. This issue will be fixed later when client gets aware of RPC
162 static int mdc_getattr_common(struct obd_export *exp,
163 struct ptlrpc_request *req)
165 struct req_capsule *pill = &req->rq_pill;
166 struct mdt_body *body;
171 /* Request message already built. */
172 rc = ptlrpc_queue_wait(req);
176 /* sanity check for the reply */
177 body = req_capsule_server_get(pill, &RMF_MDT_BODY);
181 CDEBUG(D_NET, "mode: %o\n", body->mbo_mode);
183 mdc_update_max_ea_from_body(exp, body);
184 if (body->mbo_eadatasize != 0) {
185 eadata = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
186 body->mbo_eadatasize);
194 static int mdc_getattr(struct obd_export *exp, struct md_op_data *op_data,
195 struct ptlrpc_request **request)
197 struct ptlrpc_request *req;
201 /* Single MDS without an LMV case */
202 if (op_data->op_flags & MF_GET_MDT_IDX) {
207 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
211 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
213 ptlrpc_request_free(req);
217 mdc_pack_body(req, &op_data->op_fid1, op_data->op_valid,
218 op_data->op_mode, -1, 0);
220 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
222 ptlrpc_request_set_replen(req);
224 rc = mdc_getattr_common(exp, req);
226 ptlrpc_req_finished(req);
232 static int mdc_getattr_name(struct obd_export *exp, struct md_op_data *op_data,
233 struct ptlrpc_request **request)
235 struct ptlrpc_request *req;
240 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
241 &RQF_MDS_GETATTR_NAME);
245 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
246 op_data->op_namelen + 1);
248 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR_NAME);
250 ptlrpc_request_free(req);
254 mdc_pack_body(req, &op_data->op_fid1, op_data->op_valid,
255 op_data->op_mode, op_data->op_suppgids[0], 0);
257 if (op_data->op_name) {
258 char *name = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
259 LASSERT(strnlen(op_data->op_name, op_data->op_namelen) ==
260 op_data->op_namelen);
261 memcpy(name, op_data->op_name, op_data->op_namelen);
264 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
266 ptlrpc_request_set_replen(req);
268 rc = mdc_getattr_common(exp, req);
270 ptlrpc_req_finished(req);
276 static int mdc_xattr_common(struct obd_export *exp,const struct req_format *fmt,
277 const struct lu_fid *fid, int opcode, u64 valid,
278 const char *xattr_name, const char *input,
279 int input_size, int output_size, int flags,
280 __u32 suppgid, struct ptlrpc_request **request)
282 struct ptlrpc_request *req;
283 int xattr_namelen = 0;
289 req = ptlrpc_request_alloc(class_exp2cliimp(exp), fmt);
294 xattr_namelen = strlen(xattr_name) + 1;
295 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
300 req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT,
304 /* Flush local XATTR locks to get rid of a possible cancel RPC */
305 if (opcode == MDS_REINT && fid_is_sane(fid) &&
306 exp->exp_connect_data.ocd_ibits_known & MDS_INODELOCK_XATTR) {
307 struct list_head cancels = LIST_HEAD_INIT(cancels);
310 /* Without that packing would fail */
312 req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
315 count = mdc_resource_get_unused(exp, fid,
317 MDS_INODELOCK_XATTR);
319 rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
321 ptlrpc_request_free(req);
325 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, opcode);
327 ptlrpc_request_free(req);
332 if (opcode == MDS_REINT) {
333 struct mdt_rec_setxattr *rec;
335 CLASSERT(sizeof(struct mdt_rec_setxattr) ==
336 sizeof(struct mdt_rec_reint));
337 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
338 rec->sx_opcode = REINT_SETXATTR;
339 rec->sx_fsuid = from_kuid(&init_user_ns, current_fsuid());
340 rec->sx_fsgid = from_kgid(&init_user_ns, current_fsgid());
341 rec->sx_cap = cfs_curproc_cap_pack();
342 rec->sx_suppgid1 = suppgid;
343 rec->sx_suppgid2 = -1;
345 rec->sx_valid = valid | OBD_MD_FLCTIME;
346 rec->sx_time = ktime_get_real_seconds();
347 rec->sx_size = output_size;
348 rec->sx_flags = flags;
350 mdc_pack_body(req, fid, valid, output_size, suppgid, flags);
354 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
355 memcpy(tmp, xattr_name, xattr_namelen);
358 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
359 memcpy(tmp, input, input_size);
362 if (req_capsule_has_field(&req->rq_pill, &RMF_EADATA, RCL_SERVER))
363 req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
364 RCL_SERVER, output_size);
365 ptlrpc_request_set_replen(req);
368 if (opcode == MDS_REINT)
369 mdc_get_mod_rpc_slot(req, NULL);
371 rc = ptlrpc_queue_wait(req);
373 if (opcode == MDS_REINT)
374 mdc_put_mod_rpc_slot(req, NULL);
377 ptlrpc_req_finished(req);
383 static int mdc_setxattr(struct obd_export *exp, const struct lu_fid *fid,
384 u64 valid, const char *xattr_name,
385 const char *input, int input_size, int output_size,
386 int flags, __u32 suppgid,
387 struct ptlrpc_request **request)
389 return mdc_xattr_common(exp, &RQF_MDS_REINT_SETXATTR,
390 fid, MDS_REINT, valid, xattr_name,
391 input, input_size, output_size, flags,
395 static int mdc_getxattr(struct obd_export *exp, const struct lu_fid *fid,
396 u64 valid, const char *xattr_name,
397 const char *input, int input_size, int output_size,
398 int flags, struct ptlrpc_request **request)
400 return mdc_xattr_common(exp, &RQF_MDS_GETXATTR,
401 fid, MDS_GETXATTR, valid, xattr_name,
402 input, input_size, output_size, flags,
406 #ifdef CONFIG_FS_POSIX_ACL
407 static int mdc_unpack_acl(struct ptlrpc_request *req, struct lustre_md *md)
409 struct req_capsule *pill = &req->rq_pill;
410 struct mdt_body *body = md->body;
411 struct posix_acl *acl;
416 if (!body->mbo_aclsize)
419 buf = req_capsule_server_sized_get(pill, &RMF_ACL, body->mbo_aclsize);
424 acl = posix_acl_from_xattr(&init_user_ns, buf, body->mbo_aclsize);
429 CERROR("convert xattr to acl: %d\n", rc);
433 rc = posix_acl_valid(&init_user_ns, acl);
435 CERROR("validate acl: %d\n", rc);
436 posix_acl_release(acl);
444 #define mdc_unpack_acl(req, md) 0
447 int mdc_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
448 struct obd_export *dt_exp, struct obd_export *md_exp,
449 struct lustre_md *md)
451 struct req_capsule *pill = &req->rq_pill;
456 memset(md, 0, sizeof(*md));
458 md->body = req_capsule_server_get(pill, &RMF_MDT_BODY);
459 LASSERT(md->body != NULL);
461 if (md->body->mbo_valid & OBD_MD_FLEASIZE) {
462 if (!S_ISREG(md->body->mbo_mode)) {
463 CDEBUG(D_INFO, "OBD_MD_FLEASIZE set, should be a "
464 "regular file, but is not\n");
465 GOTO(out, rc = -EPROTO);
468 if (md->body->mbo_eadatasize == 0) {
469 CDEBUG(D_INFO, "OBD_MD_FLEASIZE set, "
470 "but eadatasize 0\n");
471 GOTO(out, rc = -EPROTO);
474 md->layout.lb_len = md->body->mbo_eadatasize;
475 md->layout.lb_buf = req_capsule_server_sized_get(pill,
478 if (md->layout.lb_buf == NULL)
479 GOTO(out, rc = -EPROTO);
480 } else if (md->body->mbo_valid & OBD_MD_FLDIREA) {
481 const union lmv_mds_md *lmv;
484 if (!S_ISDIR(md->body->mbo_mode)) {
485 CDEBUG(D_INFO, "OBD_MD_FLDIREA set, should be a "
486 "directory, but is not\n");
487 GOTO(out, rc = -EPROTO);
490 lmv_size = md->body->mbo_eadatasize;
492 CDEBUG(D_INFO, "OBD_MD_FLDIREA is set, "
493 "but eadatasize 0\n");
497 if (md->body->mbo_valid & OBD_MD_MEA) {
498 lmv = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
501 GOTO(out, rc = -EPROTO);
503 rc = md_unpackmd(md_exp, &md->lmv, lmv, lmv_size);
507 if (rc < (typeof(rc))sizeof(*md->lmv)) {
508 CDEBUG(D_INFO, "size too small: "
509 "rc < sizeof(*md->lmv) (%d < %d)\n",
510 rc, (int)sizeof(*md->lmv));
511 GOTO(out, rc = -EPROTO);
517 if (md->body->mbo_valid & OBD_MD_FLACL) {
518 /* for ACL, it's possible that FLACL is set but aclsize is zero.
519 * only when aclsize != 0 there's an actual segment for ACL
522 if (md->body->mbo_aclsize) {
523 rc = mdc_unpack_acl(req, md);
526 #ifdef CONFIG_FS_POSIX_ACL
528 md->posix_acl = NULL;
536 #ifdef CONFIG_FS_POSIX_ACL
537 posix_acl_release(md->posix_acl);
543 int mdc_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
549 void mdc_replay_open(struct ptlrpc_request *req)
551 struct md_open_data *mod = req->rq_cb_data;
552 struct ptlrpc_request *close_req;
553 struct obd_client_handle *och;
554 struct lustre_handle old;
555 struct mdt_body *body;
559 DEBUG_REQ(D_ERROR, req,
560 "Can't properly replay without open data.");
565 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
566 LASSERT(body != NULL);
570 struct lustre_handle *file_fh;
572 LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
574 file_fh = &och->och_fh;
575 CDEBUG(D_HA, "updating handle from %#llx to %#llx\n",
576 file_fh->cookie, body->mbo_handle.cookie);
578 *file_fh = body->mbo_handle;
580 close_req = mod->mod_close_req;
581 if (close_req != NULL) {
582 __u32 opc = lustre_msg_get_opc(close_req->rq_reqmsg);
583 struct mdt_ioepoch *epoch;
585 LASSERT(opc == MDS_CLOSE);
586 epoch = req_capsule_client_get(&close_req->rq_pill,
591 LASSERT(!memcmp(&old, &epoch->mio_handle, sizeof(old)));
593 DEBUG_REQ(D_HA, close_req, "updating close body with new fh");
594 epoch->mio_handle = body->mbo_handle;
599 void mdc_commit_open(struct ptlrpc_request *req)
601 struct md_open_data *mod = req->rq_cb_data;
606 * No need to touch md_open_data::mod_och, it holds a reference on
607 * \var mod and will zero references to each other, \var mod will be
608 * freed after that when md_open_data::mod_och will put the reference.
612 * Do not let open request to disappear as it still may be needed
613 * for close rpc to happen (it may happen on evict only, otherwise
614 * ptlrpc_request::rq_replay does not let mdc_commit_open() to be
615 * called), just mark this rpc as committed to distinguish these 2
616 * cases, see mdc_close() for details. The open request reference will
617 * be put along with freeing \var mod.
619 ptlrpc_request_addref(req);
620 spin_lock(&req->rq_lock);
621 req->rq_committed = 1;
622 spin_unlock(&req->rq_lock);
623 req->rq_cb_data = NULL;
627 int mdc_set_open_replay_data(struct obd_export *exp,
628 struct obd_client_handle *och,
629 struct lookup_intent *it)
631 struct md_open_data *mod;
632 struct mdt_rec_create *rec;
633 struct mdt_body *body;
634 struct ptlrpc_request *open_req = it->it_request;
635 struct obd_import *imp = open_req->rq_import;
638 if (!open_req->rq_replay)
641 rec = req_capsule_client_get(&open_req->rq_pill, &RMF_REC_REINT);
642 body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
643 LASSERT(rec != NULL);
644 /* Incoming message in my byte order (it's been swabbed). */
645 /* Outgoing messages always in my byte order. */
646 LASSERT(body != NULL);
648 /* Only if the import is replayable, we set replay_open data */
649 if (och && imp->imp_replayable) {
650 mod = obd_mod_alloc();
652 DEBUG_REQ(D_ERROR, open_req,
653 "Can't allocate md_open_data");
658 * Take a reference on \var mod, to be freed on mdc_close().
659 * It protects \var mod from being freed on eviction (commit
660 * callback is called despite rq_replay flag).
661 * Another reference for \var och.
666 spin_lock(&open_req->rq_lock);
669 mod->mod_is_create = it_disposition(it, DISP_OPEN_CREATE) ||
670 it_disposition(it, DISP_OPEN_STRIPE);
671 mod->mod_open_req = open_req;
672 open_req->rq_cb_data = mod;
673 open_req->rq_commit_cb = mdc_commit_open;
674 spin_unlock(&open_req->rq_lock);
677 rec->cr_fid2 = body->mbo_fid1;
678 rec->cr_ioepoch = body->mbo_ioepoch;
679 rec->cr_old_handle.cookie = body->mbo_handle.cookie;
680 open_req->rq_replay_cb = mdc_replay_open;
681 if (!fid_is_sane(&body->mbo_fid1)) {
682 DEBUG_REQ(D_ERROR, open_req, "Saving replay request with "
687 DEBUG_REQ(D_RPCTRACE, open_req, "Set up open replay data");
691 static void mdc_free_open(struct md_open_data *mod)
695 if (mod->mod_is_create == 0 &&
696 imp_connect_disp_stripe(mod->mod_open_req->rq_import))
700 * No reason to asssert here if the open request has
701 * rq_replay == 1. It means that mdc_close failed, and
702 * close request wasn`t sent. It is not fatal to client.
703 * The worst thing is eviction if the client gets open lock
706 DEBUG_REQ(D_RPCTRACE, mod->mod_open_req, "free open request rq_replay"
707 "= %d\n", mod->mod_open_req->rq_replay);
709 ptlrpc_request_committed(mod->mod_open_req, committed);
710 if (mod->mod_close_req)
711 ptlrpc_request_committed(mod->mod_close_req, committed);
714 int mdc_clear_open_replay_data(struct obd_export *exp,
715 struct obd_client_handle *och)
717 struct md_open_data *mod = och->och_mod;
721 * It is possible to not have \var mod in a case of eviction between
722 * lookup and ll_file_open().
727 LASSERT(mod != LP_POISON);
728 LASSERT(mod->mod_open_req != NULL);
738 static int mdc_close(struct obd_export *exp, struct md_op_data *op_data,
739 struct md_open_data *mod, struct ptlrpc_request **request)
741 struct obd_device *obd = class_exp2obd(exp);
742 struct ptlrpc_request *req;
743 struct req_format *req_fmt;
748 if (op_data->op_bias & MDS_HSM_RELEASE) {
749 req_fmt = &RQF_MDS_INTENT_CLOSE;
751 /* allocate a FID for volatile file */
752 rc = mdc_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
754 CERROR("%s: "DFID" failed to allocate FID: %d\n",
755 obd->obd_name, PFID(&op_data->op_fid1), rc);
756 /* save the errcode and proceed to close */
759 } else if (op_data->op_bias & MDS_CLOSE_LAYOUT_SWAP) {
760 req_fmt = &RQF_MDS_INTENT_CLOSE;
762 req_fmt = &RQF_MDS_CLOSE;
766 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_CLOSE))
769 req = ptlrpc_request_alloc(class_exp2cliimp(exp), req_fmt);
771 /* Ensure that this close's handle is fixed up during replay. */
772 if (likely(mod != NULL)) {
773 LASSERTF(mod->mod_open_req != NULL &&
774 mod->mod_open_req->rq_type != LI_POISON,
775 "POISONED open %p!\n", mod->mod_open_req);
777 mod->mod_close_req = req;
779 DEBUG_REQ(D_HA, mod->mod_open_req, "matched open");
780 /* We no longer want to preserve this open for replay even
781 * though the open was committed. b=3632, b=3633 */
782 spin_lock(&mod->mod_open_req->rq_lock);
783 mod->mod_open_req->rq_replay = 0;
784 spin_unlock(&mod->mod_open_req->rq_lock);
786 CDEBUG(D_HA, "couldn't find open req; expecting close error\n");
790 * TODO: repeat close after errors
792 CWARN("%s: close of FID "DFID" failed, file reference will be "
793 "dropped when this client unmounts or is evicted\n",
794 obd->obd_name, PFID(&op_data->op_fid1));
795 GOTO(out, rc = -ENOMEM);
798 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_CLOSE);
800 ptlrpc_request_free(req);
805 /* To avoid a livelock (bug 7034), we need to send CLOSE RPCs to a
806 * portal whose threads are not taking any DLM locks and are therefore
807 * always progressing */
808 req->rq_request_portal = MDS_READPAGE_PORTAL;
809 ptlrpc_at_set_req_timeout(req);
812 mdc_close_pack(req, op_data);
814 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
815 obd->u.cli.cl_default_mds_easize);
817 ptlrpc_request_set_replen(req);
819 mdc_get_mod_rpc_slot(req, NULL);
820 rc = ptlrpc_queue_wait(req);
821 mdc_put_mod_rpc_slot(req, NULL);
823 if (req->rq_repmsg == NULL) {
824 CDEBUG(D_RPCTRACE, "request failed to send: %p, %d\n", req,
827 rc = req->rq_status ?: -EIO;
828 } else if (rc == 0 || rc == -EAGAIN) {
829 struct mdt_body *body;
831 rc = lustre_msg_get_status(req->rq_repmsg);
832 if (lustre_msg_get_type(req->rq_repmsg) == PTL_RPC_MSG_ERR) {
833 DEBUG_REQ(D_ERROR, req, "type == PTL_RPC_MSG_ERR, err "
838 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
841 } else if (rc == -ESTALE) {
843 * it can be allowed error after 3633 if open was committed and
844 * server failed before close was sent. Let's check if mod
845 * exists and return no error in that case
848 DEBUG_REQ(D_HA, req, "Reset ESTALE = %d", rc);
849 LASSERT(mod->mod_open_req != NULL);
850 if (mod->mod_open_req->rq_committed)
858 mod->mod_close_req = NULL;
859 /* Since now, mod is accessed through open_req only,
860 * thus close req does not keep a reference on mod anymore. */
865 RETURN(rc < 0 ? rc : saved_rc);
868 static int mdc_getpage(struct obd_export *exp, const struct lu_fid *fid,
869 u64 offset, struct page **pages, int npages,
870 struct ptlrpc_request **request)
872 struct ptlrpc_request *req;
873 struct ptlrpc_bulk_desc *desc;
875 wait_queue_head_t waitq;
877 struct l_wait_info lwi;
882 init_waitqueue_head(&waitq);
885 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_READPAGE);
889 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_READPAGE);
891 ptlrpc_request_free(req);
895 req->rq_request_portal = MDS_READPAGE_PORTAL;
896 ptlrpc_at_set_req_timeout(req);
898 desc = ptlrpc_prep_bulk_imp(req, npages, 1,
899 PTLRPC_BULK_PUT_SINK | PTLRPC_BULK_BUF_KIOV,
901 &ptlrpc_bulk_kiov_pin_ops);
903 ptlrpc_req_finished(req);
907 /* NB req now owns desc and will free it when it gets freed */
908 for (i = 0; i < npages; i++)
909 desc->bd_frag_ops->add_kiov_frag(desc, pages[i], 0,
912 mdc_readdir_pack(req, offset, PAGE_SIZE * npages, fid);
914 ptlrpc_request_set_replen(req);
915 rc = ptlrpc_queue_wait(req);
917 ptlrpc_req_finished(req);
918 if (rc != -ETIMEDOUT)
922 if (!client_should_resend(resends, &exp->exp_obd->u.cli)) {
923 CERROR("%s: too many resend retries: rc = %d\n",
924 exp->exp_obd->obd_name, -EIO);
927 lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(resends), NULL, NULL,
929 l_wait_event(waitq, 0, &lwi);
934 rc = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk,
935 req->rq_bulk->bd_nob_transferred);
937 ptlrpc_req_finished(req);
941 if (req->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK) {
942 CERROR("%s: unexpected bytes transferred: %d (%ld expected)\n",
943 exp->exp_obd->obd_name, req->rq_bulk->bd_nob_transferred,
945 ptlrpc_req_finished(req);
953 static void mdc_release_page(struct page *page, int remove)
957 if (likely(page->mapping != NULL))
958 truncate_complete_page(page->mapping, page);
964 static struct page *mdc_page_locate(struct address_space *mapping, __u64 *hash,
965 __u64 *start, __u64 *end, int hash64)
968 * Complement of hash is used as an index so that
969 * radix_tree_gang_lookup() can be used to find a page with starting
970 * hash _smaller_ than one we are looking for.
972 unsigned long offset = hash_x_index(*hash, hash64);
976 spin_lock_irq(&mapping->tree_lock);
977 found = radix_tree_gang_lookup(&mapping->page_tree,
978 (void **)&page, offset, 1);
979 if (found > 0 && !radix_tree_exceptional_entry(page)) {
980 struct lu_dirpage *dp;
983 spin_unlock_irq(&mapping->tree_lock);
985 * In contrast to find_lock_page() we are sure that directory
986 * page cannot be truncated (while DLM lock is held) and,
987 * hence, can avoid restart.
989 * In fact, page cannot be locked here at all, because
990 * mdc_read_page_remote does synchronous io.
992 wait_on_page_locked(page);
993 if (PageUptodate(page)) {
995 if (BITS_PER_LONG == 32 && hash64) {
996 *start = le64_to_cpu(dp->ldp_hash_start) >> 32;
997 *end = le64_to_cpu(dp->ldp_hash_end) >> 32;
1000 *start = le64_to_cpu(dp->ldp_hash_start);
1001 *end = le64_to_cpu(dp->ldp_hash_end);
1003 if (unlikely(*start == 1 && *hash == 0))
1006 LASSERTF(*start <= *hash, "start = %#llx"
1007 ",end = %#llx,hash = %#llx\n",
1008 *start, *end, *hash);
1009 CDEBUG(D_VFSTRACE, "offset %lx [%#llx %#llx],"
1010 " hash %#llx\n", offset, *start, *end, *hash);
1013 mdc_release_page(page, 0);
1015 } else if (*end != *start && *hash == *end) {
1017 * upon hash collision, remove this page,
1018 * otherwise put page reference, and
1019 * mdc_read_page_remote() will issue RPC to
1020 * fetch the page we want.
1023 mdc_release_page(page,
1024 le32_to_cpu(dp->ldp_flags) & LDF_COLLIDE);
1029 page = ERR_PTR(-EIO);
1032 spin_unlock_irq(&mapping->tree_lock);
1039 * Adjust a set of pages, each page containing an array of lu_dirpages,
1040 * so that each page can be used as a single logical lu_dirpage.
1042 * A lu_dirpage is laid out as follows, where s = ldp_hash_start,
1043 * e = ldp_hash_end, f = ldp_flags, p = padding, and each "ent" is a
1044 * struct lu_dirent. It has size up to LU_PAGE_SIZE. The ldp_hash_end
1045 * value is used as a cookie to request the next lu_dirpage in a
1046 * directory listing that spans multiple pages (two in this example):
1049 * .|--------v------- -----.
1050 * |s|e|f|p|ent|ent| ... |ent|
1051 * '--|-------------- -----' Each PAGE contains a single
1052 * '------. lu_dirpage.
1053 * .---------v------- -----.
1054 * |s|e|f|p|ent| 0 | ... | 0 |
1055 * '----------------- -----'
1057 * However, on hosts where the native VM page size (PAGE_SIZE) is
1058 * larger than LU_PAGE_SIZE, a single host page may contain multiple
1059 * lu_dirpages. After reading the lu_dirpages from the MDS, the
1060 * ldp_hash_end of the first lu_dirpage refers to the one immediately
1061 * after it in the same PAGE (arrows simplified for brevity, but
1062 * in general e0==s1, e1==s2, etc.):
1064 * .-------------------- -----.
1065 * |s0|e0|f0|p|ent|ent| ... |ent|
1066 * |---v---------------- -----|
1067 * |s1|e1|f1|p|ent|ent| ... |ent|
1068 * |---v---------------- -----| Here, each PAGE contains
1069 * ... multiple lu_dirpages.
1070 * |---v---------------- -----|
1071 * |s'|e'|f'|p|ent|ent| ... |ent|
1072 * '---|---------------- -----'
1074 * .----------------------------.
1077 * This structure is transformed into a single logical lu_dirpage as follows:
1079 * - Replace e0 with e' so the request for the next lu_dirpage gets the page
1080 * labeled 'next PAGE'.
1082 * - Copy the LDF_COLLIDE flag from f' to f0 to correctly reflect whether
1083 * a hash collision with the next page exists.
1085 * - Adjust the lde_reclen of the ending entry of each lu_dirpage to span
1086 * to the first entry of the next lu_dirpage.
1088 #if PAGE_SIZE > LU_PAGE_SIZE
1089 static void mdc_adjust_dirpages(struct page **pages, int cfs_pgs, int lu_pgs)
1093 for (i = 0; i < cfs_pgs; i++) {
1094 struct lu_dirpage *dp = kmap(pages[i]);
1095 struct lu_dirpage *first = dp;
1096 struct lu_dirent *end_dirent = NULL;
1097 struct lu_dirent *ent;
1098 __u64 hash_end = le64_to_cpu(dp->ldp_hash_end);
1099 __u32 flags = le32_to_cpu(dp->ldp_flags);
1101 while (--lu_pgs > 0) {
1102 ent = lu_dirent_start(dp);
1103 for (end_dirent = ent; ent != NULL;
1104 end_dirent = ent, ent = lu_dirent_next(ent));
1106 /* Advance dp to next lu_dirpage. */
1107 dp = (struct lu_dirpage *)((char *)dp + LU_PAGE_SIZE);
1109 /* Check if we've reached the end of the PAGE. */
1110 if (!((unsigned long)dp & ~PAGE_MASK))
1113 /* Save the hash and flags of this lu_dirpage. */
1114 hash_end = le64_to_cpu(dp->ldp_hash_end);
1115 flags = le32_to_cpu(dp->ldp_flags);
1117 /* Check if lu_dirpage contains no entries. */
1118 if (end_dirent == NULL)
1121 /* Enlarge the end entry lde_reclen from 0 to
1122 * first entry of next lu_dirpage. */
1123 LASSERT(le16_to_cpu(end_dirent->lde_reclen) == 0);
1124 end_dirent->lde_reclen =
1125 cpu_to_le16((char *)(dp->ldp_entries) -
1126 (char *)end_dirent);
1129 first->ldp_hash_end = hash_end;
1130 first->ldp_flags &= ~cpu_to_le32(LDF_COLLIDE);
1131 first->ldp_flags |= flags & cpu_to_le32(LDF_COLLIDE);
1135 LASSERTF(lu_pgs == 0, "left = %d\n", lu_pgs);
1138 #define mdc_adjust_dirpages(pages, cfs_pgs, lu_pgs) do {} while (0)
1139 #endif /* PAGE_SIZE > LU_PAGE_SIZE */
1141 /* parameters for readdir page */
1142 struct readpage_param {
1143 struct md_op_data *rp_mod;
1146 struct obd_export *rp_exp;
1147 struct md_callback *rp_cb;
1150 #ifndef HAVE_DELETE_FROM_PAGE_CACHE
1151 static inline void delete_from_page_cache(struct page *page)
1153 remove_from_page_cache(page);
1159 * Read pages from server.
1161 * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
1162 * a header lu_dirpage which describes the start/end hash, and whether this
1163 * page is empty (contains no dir entry) or hash collide with next page.
1164 * After client receives reply, several pages will be integrated into dir page
1165 * in PAGE_SIZE (if PAGE_SIZE greater than LU_PAGE_SIZE), and the
1166 * lu_dirpage for this integrated page will be adjusted.
1168 static int mdc_read_page_remote(void *data, struct page *page0)
1170 struct readpage_param *rp = data;
1171 struct page **page_pool;
1173 struct lu_dirpage *dp;
1174 int rd_pgs = 0; /* number of pages read actually */
1176 struct md_op_data *op_data = rp->rp_mod;
1177 struct ptlrpc_request *req;
1178 int max_pages = op_data->op_max_pages;
1179 struct inode *inode;
1185 LASSERT(max_pages > 0 && max_pages <= PTLRPC_MAX_BRW_PAGES);
1186 inode = op_data->op_data;
1187 fid = &op_data->op_fid1;
1188 LASSERT(inode != NULL);
1190 OBD_ALLOC(page_pool, sizeof(page_pool[0]) * max_pages);
1191 if (page_pool != NULL) {
1192 page_pool[0] = page0;
1198 for (npages = 1; npages < max_pages; npages++) {
1199 page = page_cache_alloc_cold(inode->i_mapping);
1202 page_pool[npages] = page;
1205 rc = mdc_getpage(rp->rp_exp, fid, rp->rp_off, page_pool, npages, &req);
1207 /* page0 is special, which was added into page cache early */
1208 delete_from_page_cache(page0);
1212 rd_pgs = (req->rq_bulk->bd_nob_transferred +
1213 PAGE_SIZE - 1) >> PAGE_SHIFT;
1214 lu_pgs = req->rq_bulk->bd_nob_transferred >>
1216 LASSERT(!(req->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK));
1218 CDEBUG(D_INODE, "read %d(%d) pages\n", rd_pgs, lu_pgs);
1220 mdc_adjust_dirpages(page_pool, rd_pgs, lu_pgs);
1222 SetPageUptodate(page0);
1226 ptlrpc_req_finished(req);
1227 CDEBUG(D_CACHE, "read %d/%d pages\n", rd_pgs, npages);
1228 for (i = 1; i < npages; i++) {
1229 unsigned long offset;
1233 page = page_pool[i];
1235 if (rc < 0 || i >= rd_pgs) {
1240 SetPageUptodate(page);
1243 hash = le64_to_cpu(dp->ldp_hash_start);
1246 offset = hash_x_index(hash, rp->rp_hash64);
1248 prefetchw(&page->flags);
1249 ret = add_to_page_cache_lru(page, inode->i_mapping, offset,
1254 CDEBUG(D_VFSTRACE, "page %lu add to page cache failed:"
1255 " rc = %d\n", offset, ret);
1259 if (page_pool != &page0)
1260 OBD_FREE(page_pool, sizeof(page_pool[0]) * max_pages);
1266 * Read dir page from cache first, if it can not find it, read it from
1267 * server and add into the cache.
1269 * \param[in] exp MDC export
1270 * \param[in] op_data client MD stack parameters, transfering parameters
1271 * between different layers on client MD stack.
1272 * \param[in] cb_op callback required for ldlm lock enqueue during
1274 * \param[in] hash_offset the hash offset of the page to be read
1275 * \param[in] ppage the page to be read
1277 * retval = 0 get the page successfully
1278 * errno(<0) get the page failed
1280 static int mdc_read_page(struct obd_export *exp, struct md_op_data *op_data,
1281 struct md_callback *cb_op, __u64 hash_offset,
1282 struct page **ppage)
1284 struct lookup_intent it = { .it_op = IT_READDIR };
1286 struct inode *dir = op_data->op_data;
1287 struct address_space *mapping;
1288 struct lu_dirpage *dp;
1291 struct lustre_handle lockh;
1292 struct ptlrpc_request *enq_req = NULL;
1293 struct readpage_param rp_param;
1300 LASSERT(dir != NULL);
1301 mapping = dir->i_mapping;
1303 rc = mdc_intent_lock(exp, op_data, &it, &enq_req,
1304 cb_op->md_blocking_ast, 0);
1305 if (enq_req != NULL)
1306 ptlrpc_req_finished(enq_req);
1309 CERROR("%s: "DFID" lock enqueue fails: rc = %d\n",
1310 exp->exp_obd->obd_name, PFID(&op_data->op_fid1), rc);
1315 lockh.cookie = it.it_lock_handle;
1316 mdc_set_lock_data(exp, &lockh, dir, NULL);
1318 rp_param.rp_off = hash_offset;
1319 rp_param.rp_hash64 = op_data->op_cli_flags & CLI_HASH64;
1320 page = mdc_page_locate(mapping, &rp_param.rp_off, &start, &end,
1321 rp_param.rp_hash64);
1323 CERROR("%s: dir page locate: "DFID" at %llu: rc %ld\n",
1324 exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1325 rp_param.rp_off, PTR_ERR(page));
1326 GOTO(out_unlock, rc = PTR_ERR(page));
1327 } else if (page != NULL) {
1329 * XXX nikita: not entirely correct handling of a corner case:
1330 * suppose hash chain of entries with hash value HASH crosses
1331 * border between pages P0 and P1. First both P0 and P1 are
1332 * cached, seekdir() is called for some entry from the P0 part
1333 * of the chain. Later P0 goes out of cache. telldir(HASH)
1334 * happens and finds P1, as it starts with matching hash
1335 * value. Remaining entries from P0 part of the chain are
1336 * skipped. (Is that really a bug?)
1338 * Possible solutions: 0. don't cache P1 is such case, handle
1339 * it as an "overflow" page. 1. invalidate all pages at
1340 * once. 2. use HASH|1 as an index for P1.
1342 GOTO(hash_collision, page);
1345 rp_param.rp_exp = exp;
1346 rp_param.rp_mod = op_data;
1347 page = read_cache_page(mapping,
1348 hash_x_index(rp_param.rp_off,
1349 rp_param.rp_hash64),
1350 mdc_read_page_remote, &rp_param);
1352 CDEBUG(D_INFO, "%s: read cache page: "DFID" at %llu: %ld\n",
1353 exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1354 rp_param.rp_off, PTR_ERR(page));
1355 GOTO(out_unlock, rc = PTR_ERR(page));
1358 wait_on_page_locked(page);
1360 if (!PageUptodate(page)) {
1361 CERROR("%s: page not updated: "DFID" at %llu: rc %d\n",
1362 exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1363 rp_param.rp_off, -5);
1366 if (!PageChecked(page))
1367 SetPageChecked(page);
1368 if (PageError(page)) {
1369 CERROR("%s: page error: "DFID" at %llu: rc %d\n",
1370 exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1371 rp_param.rp_off, -5);
1376 dp = page_address(page);
1377 if (BITS_PER_LONG == 32 && rp_param.rp_hash64) {
1378 start = le64_to_cpu(dp->ldp_hash_start) >> 32;
1379 end = le64_to_cpu(dp->ldp_hash_end) >> 32;
1380 rp_param.rp_off = hash_offset >> 32;
1382 start = le64_to_cpu(dp->ldp_hash_start);
1383 end = le64_to_cpu(dp->ldp_hash_end);
1384 rp_param.rp_off = hash_offset;
1387 LASSERT(start == rp_param.rp_off);
1388 CWARN("Page-wide hash collision: %#lx\n", (unsigned long)end);
1389 #if BITS_PER_LONG == 32
1390 CWARN("Real page-wide hash collision at [%llu %llu] with "
1391 "hash %llu\n", le64_to_cpu(dp->ldp_hash_start),
1392 le64_to_cpu(dp->ldp_hash_end), hash_offset);
1396 * Fetch whole overflow chain...
1404 ldlm_lock_decref(&lockh, it.it_lock_mode);
1408 mdc_release_page(page, 1);
1414 static int mdc_statfs(const struct lu_env *env,
1415 struct obd_export *exp, struct obd_statfs *osfs,
1416 __u64 max_age, __u32 flags)
1418 struct obd_device *obd = class_exp2obd(exp);
1419 struct ptlrpc_request *req;
1420 struct obd_statfs *msfs;
1421 struct obd_import *imp = NULL;
1426 * Since the request might also come from lprocfs, so we need
1427 * sync this with client_disconnect_export Bug15684
1429 down_read(&obd->u.cli.cl_sem);
1430 if (obd->u.cli.cl_import)
1431 imp = class_import_get(obd->u.cli.cl_import);
1432 up_read(&obd->u.cli.cl_sem);
1436 req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_STATFS,
1437 LUSTRE_MDS_VERSION, MDS_STATFS);
1439 GOTO(output, rc = -ENOMEM);
1441 ptlrpc_request_set_replen(req);
1443 if (flags & OBD_STATFS_NODELAY) {
1444 /* procfs requests not want stay in wait for avoid deadlock */
1445 req->rq_no_resend = 1;
1446 req->rq_no_delay = 1;
1449 rc = ptlrpc_queue_wait(req);
1451 /* check connection error first */
1452 if (imp->imp_connect_error)
1453 rc = imp->imp_connect_error;
1457 msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
1459 GOTO(out, rc = -EPROTO);
1464 ptlrpc_req_finished(req);
1466 class_import_put(imp);
1470 static int mdc_ioc_fid2path(struct obd_export *exp, struct getinfo_fid2path *gf)
1472 __u32 keylen, vallen;
1476 if (gf->gf_pathlen > PATH_MAX)
1477 RETURN(-ENAMETOOLONG);
1478 if (gf->gf_pathlen < 2)
1481 /* Key is KEY_FID2PATH + getinfo_fid2path description */
1482 keylen = cfs_size_round(sizeof(KEY_FID2PATH) + sizeof(*gf) +
1483 sizeof(struct lu_fid));
1484 OBD_ALLOC(key, keylen);
1487 memcpy(key, KEY_FID2PATH, sizeof(KEY_FID2PATH));
1488 memcpy(key + cfs_size_round(sizeof(KEY_FID2PATH)), gf, sizeof(*gf));
1489 memcpy(key + cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*gf),
1490 gf->gf_u.gf_root_fid, sizeof(struct lu_fid));
1491 CDEBUG(D_IOCTL, "path get "DFID" from %llu #%d\n",
1492 PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno);
1494 if (!fid_is_sane(&gf->gf_fid))
1495 GOTO(out, rc = -EINVAL);
1497 /* Val is struct getinfo_fid2path result plus path */
1498 vallen = sizeof(*gf) + gf->gf_pathlen;
1500 rc = obd_get_info(NULL, exp, keylen, key, &vallen, gf);
1501 if (rc != 0 && rc != -EREMOTE)
1504 if (vallen <= sizeof(*gf))
1505 GOTO(out, rc = -EPROTO);
1506 if (vallen > sizeof(*gf) + gf->gf_pathlen)
1507 GOTO(out, rc = -EOVERFLOW);
1509 CDEBUG(D_IOCTL, "path got "DFID" from %llu #%d: %s\n",
1510 PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno,
1511 gf->gf_pathlen < 512 ? gf->gf_u.gf_path :
1512 /* only log the last 512 characters of the path */
1513 gf->gf_u.gf_path + gf->gf_pathlen - 512);
1516 OBD_FREE(key, keylen);
1520 static int mdc_ioc_hsm_progress(struct obd_export *exp,
1521 struct hsm_progress_kernel *hpk)
1523 struct obd_import *imp = class_exp2cliimp(exp);
1524 struct hsm_progress_kernel *req_hpk;
1525 struct ptlrpc_request *req;
1529 req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_PROGRESS,
1530 LUSTRE_MDS_VERSION, MDS_HSM_PROGRESS);
1532 GOTO(out, rc = -ENOMEM);
1534 mdc_pack_body(req, NULL, 0, 0, -1, 0);
1536 /* Copy hsm_progress struct */
1537 req_hpk = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_PROGRESS);
1538 if (req_hpk == NULL)
1539 GOTO(out, rc = -EPROTO);
1542 req_hpk->hpk_errval = lustre_errno_hton(hpk->hpk_errval);
1544 ptlrpc_request_set_replen(req);
1546 mdc_get_mod_rpc_slot(req, NULL);
1547 rc = ptlrpc_queue_wait(req);
1548 mdc_put_mod_rpc_slot(req, NULL);
1552 ptlrpc_req_finished(req);
1556 static int mdc_ioc_hsm_ct_register(struct obd_import *imp, __u32 archives)
1558 __u32 *archive_mask;
1559 struct ptlrpc_request *req;
1563 req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_REGISTER,
1565 MDS_HSM_CT_REGISTER);
1567 GOTO(out, rc = -ENOMEM);
1569 mdc_pack_body(req, NULL, 0, 0, -1, 0);
1571 /* Copy hsm_progress struct */
1572 archive_mask = req_capsule_client_get(&req->rq_pill,
1573 &RMF_MDS_HSM_ARCHIVE);
1574 if (archive_mask == NULL)
1575 GOTO(out, rc = -EPROTO);
1577 *archive_mask = archives;
1579 ptlrpc_request_set_replen(req);
1581 rc = mdc_queue_wait(req);
1584 ptlrpc_req_finished(req);
1588 static int mdc_ioc_hsm_current_action(struct obd_export *exp,
1589 struct md_op_data *op_data)
1591 struct hsm_current_action *hca = op_data->op_data;
1592 struct hsm_current_action *req_hca;
1593 struct ptlrpc_request *req;
1597 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1598 &RQF_MDS_HSM_ACTION);
1602 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_ACTION);
1604 ptlrpc_request_free(req);
1608 mdc_pack_body(req, &op_data->op_fid1, 0, 0,
1609 op_data->op_suppgids[0], 0);
1611 ptlrpc_request_set_replen(req);
1613 rc = mdc_queue_wait(req);
1617 req_hca = req_capsule_server_get(&req->rq_pill,
1618 &RMF_MDS_HSM_CURRENT_ACTION);
1619 if (req_hca == NULL)
1620 GOTO(out, rc = -EPROTO);
1626 ptlrpc_req_finished(req);
1630 static int mdc_ioc_hsm_ct_unregister(struct obd_import *imp)
1632 struct ptlrpc_request *req;
1636 req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_UNREGISTER,
1638 MDS_HSM_CT_UNREGISTER);
1640 GOTO(out, rc = -ENOMEM);
1642 mdc_pack_body(req, NULL, 0, 0, -1, 0);
1644 ptlrpc_request_set_replen(req);
1646 rc = mdc_queue_wait(req);
1649 ptlrpc_req_finished(req);
1653 static int mdc_ioc_hsm_state_get(struct obd_export *exp,
1654 struct md_op_data *op_data)
1656 struct hsm_user_state *hus = op_data->op_data;
1657 struct hsm_user_state *req_hus;
1658 struct ptlrpc_request *req;
1662 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1663 &RQF_MDS_HSM_STATE_GET);
1667 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_GET);
1669 ptlrpc_request_free(req);
1673 mdc_pack_body(req, &op_data->op_fid1, 0, 0,
1674 op_data->op_suppgids[0], 0);
1676 ptlrpc_request_set_replen(req);
1678 rc = mdc_queue_wait(req);
1682 req_hus = req_capsule_server_get(&req->rq_pill, &RMF_HSM_USER_STATE);
1683 if (req_hus == NULL)
1684 GOTO(out, rc = -EPROTO);
1690 ptlrpc_req_finished(req);
1694 static int mdc_ioc_hsm_state_set(struct obd_export *exp,
1695 struct md_op_data *op_data)
1697 struct hsm_state_set *hss = op_data->op_data;
1698 struct hsm_state_set *req_hss;
1699 struct ptlrpc_request *req;
1703 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1704 &RQF_MDS_HSM_STATE_SET);
1708 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_SET);
1710 ptlrpc_request_free(req);
1714 mdc_pack_body(req, &op_data->op_fid1, 0, 0,
1715 op_data->op_suppgids[0], 0);
1718 req_hss = req_capsule_client_get(&req->rq_pill, &RMF_HSM_STATE_SET);
1719 if (req_hss == NULL)
1720 GOTO(out, rc = -EPROTO);
1723 ptlrpc_request_set_replen(req);
1725 mdc_get_mod_rpc_slot(req, NULL);
1726 rc = ptlrpc_queue_wait(req);
1727 mdc_put_mod_rpc_slot(req, NULL);
1731 ptlrpc_req_finished(req);
1735 static int mdc_ioc_hsm_request(struct obd_export *exp,
1736 struct hsm_user_request *hur)
1738 struct obd_import *imp = class_exp2cliimp(exp);
1739 struct ptlrpc_request *req;
1740 struct hsm_request *req_hr;
1741 struct hsm_user_item *req_hui;
1746 req = ptlrpc_request_alloc(imp, &RQF_MDS_HSM_REQUEST);
1748 GOTO(out, rc = -ENOMEM);
1750 req_capsule_set_size(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM, RCL_CLIENT,
1751 hur->hur_request.hr_itemcount
1752 * sizeof(struct hsm_user_item));
1753 req_capsule_set_size(&req->rq_pill, &RMF_GENERIC_DATA, RCL_CLIENT,
1754 hur->hur_request.hr_data_len);
1756 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_REQUEST);
1758 ptlrpc_request_free(req);
1762 mdc_pack_body(req, NULL, 0, 0, -1, 0);
1764 /* Copy hsm_request struct */
1765 req_hr = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_REQUEST);
1767 GOTO(out, rc = -EPROTO);
1768 *req_hr = hur->hur_request;
1770 /* Copy hsm_user_item structs */
1771 req_hui = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM);
1772 if (req_hui == NULL)
1773 GOTO(out, rc = -EPROTO);
1774 memcpy(req_hui, hur->hur_user_item,
1775 hur->hur_request.hr_itemcount * sizeof(struct hsm_user_item));
1777 /* Copy opaque field */
1778 req_opaque = req_capsule_client_get(&req->rq_pill, &RMF_GENERIC_DATA);
1779 if (req_opaque == NULL)
1780 GOTO(out, rc = -EPROTO);
1781 memcpy(req_opaque, hur_data(hur), hur->hur_request.hr_data_len);
1783 ptlrpc_request_set_replen(req);
1785 mdc_get_mod_rpc_slot(req, NULL);
1786 rc = ptlrpc_queue_wait(req);
1787 mdc_put_mod_rpc_slot(req, NULL);
1792 ptlrpc_req_finished(req);
1796 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
1797 struct lustre_kernelcomm *lk);
1799 static int mdc_quotactl(struct obd_device *unused, struct obd_export *exp,
1800 struct obd_quotactl *oqctl)
1802 struct ptlrpc_request *req;
1803 struct obd_quotactl *oqc;
1807 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
1808 &RQF_MDS_QUOTACTL, LUSTRE_MDS_VERSION,
1813 oqc = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1816 ptlrpc_request_set_replen(req);
1817 ptlrpc_at_set_req_timeout(req);
1818 req->rq_no_resend = 1;
1820 rc = ptlrpc_queue_wait(req);
1822 CERROR("ptlrpc_queue_wait failed, rc: %d\n", rc);
1824 if (req->rq_repmsg &&
1825 (oqc = req_capsule_server_get(&req->rq_pill, &RMF_OBD_QUOTACTL))) {
1828 CERROR ("Can't unpack obd_quotactl\n");
1831 ptlrpc_req_finished(req);
1836 static int mdc_ioc_swap_layouts(struct obd_export *exp,
1837 struct md_op_data *op_data)
1839 struct list_head cancels = LIST_HEAD_INIT(cancels);
1840 struct ptlrpc_request *req;
1842 struct mdc_swap_layouts *msl, *payload;
1845 msl = op_data->op_data;
1847 /* When the MDT will get the MDS_SWAP_LAYOUTS RPC the
1848 * first thing it will do is to cancel the 2 layout
1849 * locks held by this client.
1850 * So the client must cancel its layout locks on the 2 fids
1851 * with the request RPC to avoid extra RPC round trips.
1853 count = mdc_resource_get_unused(exp, &op_data->op_fid1, &cancels,
1854 LCK_EX, MDS_INODELOCK_LAYOUT |
1855 MDS_INODELOCK_XATTR);
1856 count += mdc_resource_get_unused(exp, &op_data->op_fid2, &cancels,
1857 LCK_EX, MDS_INODELOCK_LAYOUT |
1858 MDS_INODELOCK_XATTR);
1860 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1861 &RQF_MDS_SWAP_LAYOUTS);
1863 ldlm_lock_list_put(&cancels, l_bl_ast, count);
1867 rc = mdc_prep_elc_req(exp, req, MDS_SWAP_LAYOUTS, &cancels, count);
1869 ptlrpc_request_free(req);
1873 mdc_swap_layouts_pack(req, op_data);
1875 payload = req_capsule_client_get(&req->rq_pill, &RMF_SWAP_LAYOUTS);
1880 ptlrpc_request_set_replen(req);
1882 rc = ptlrpc_queue_wait(req);
1888 ptlrpc_req_finished(req);
1892 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1893 void *karg, void __user *uarg)
1895 struct obd_device *obd = exp->exp_obd;
1896 struct obd_ioctl_data *data = karg;
1897 struct obd_import *imp = obd->u.cli.cl_import;
1901 if (!try_module_get(THIS_MODULE)) {
1902 CERROR("%s: cannot get module '%s'\n", obd->obd_name,
1903 module_name(THIS_MODULE));
1907 case OBD_IOC_FID2PATH:
1908 rc = mdc_ioc_fid2path(exp, karg);
1910 case LL_IOC_HSM_CT_START:
1911 rc = mdc_ioc_hsm_ct_start(exp, karg);
1912 /* ignore if it was already registered on this MDS. */
1916 case LL_IOC_HSM_PROGRESS:
1917 rc = mdc_ioc_hsm_progress(exp, karg);
1919 case LL_IOC_HSM_STATE_GET:
1920 rc = mdc_ioc_hsm_state_get(exp, karg);
1922 case LL_IOC_HSM_STATE_SET:
1923 rc = mdc_ioc_hsm_state_set(exp, karg);
1925 case LL_IOC_HSM_ACTION:
1926 rc = mdc_ioc_hsm_current_action(exp, karg);
1928 case LL_IOC_HSM_REQUEST:
1929 rc = mdc_ioc_hsm_request(exp, karg);
1931 case OBD_IOC_CLIENT_RECOVER:
1932 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1, 0);
1936 case IOC_OSC_SET_ACTIVE:
1937 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
1939 case OBD_IOC_PING_TARGET:
1940 rc = ptlrpc_obd_ping(obd);
1943 * Normally IOC_OBD_STATFS, OBD_IOC_QUOTACTL iocontrol are handled by
1944 * LMV instead of MDC. But when the cluster is upgraded from 1.8,
1945 * there'd be no LMV layer thus we might be called here. Eventually
1946 * this code should be removed.
1949 case IOC_OBD_STATFS: {
1950 struct obd_statfs stat_buf = {0};
1952 if (*((__u32 *) data->ioc_inlbuf2) != 0)
1953 GOTO(out, rc = -ENODEV);
1956 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(obd),
1957 min((int)data->ioc_plen2,
1958 (int)sizeof(struct obd_uuid))))
1959 GOTO(out, rc = -EFAULT);
1961 rc = mdc_statfs(NULL, obd->obd_self_export, &stat_buf,
1962 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1967 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
1968 min((int) data->ioc_plen1,
1969 (int) sizeof(stat_buf))))
1970 GOTO(out, rc = -EFAULT);
1974 case OBD_IOC_QUOTACTL: {
1975 struct if_quotactl *qctl = karg;
1976 struct obd_quotactl *oqctl;
1978 OBD_ALLOC_PTR(oqctl);
1980 GOTO(out, rc = -ENOMEM);
1982 QCTL_COPY(oqctl, qctl);
1983 rc = obd_quotactl(exp, oqctl);
1985 QCTL_COPY(qctl, oqctl);
1986 qctl->qc_valid = QC_MDTIDX;
1987 qctl->obd_uuid = obd->u.cli.cl_target_uuid;
1990 OBD_FREE_PTR(oqctl);
1993 case LL_IOC_GET_CONNECT_FLAGS:
1994 if (copy_to_user(uarg, exp_connect_flags_ptr(exp),
1995 sizeof(*exp_connect_flags_ptr(exp))))
1996 GOTO(out, rc = -EFAULT);
1999 case LL_IOC_LOV_SWAP_LAYOUTS:
2000 rc = mdc_ioc_swap_layouts(exp, karg);
2003 CERROR("unrecognised ioctl: cmd = %#x\n", cmd);
2004 GOTO(out, rc = -ENOTTY);
2007 module_put(THIS_MODULE);
2012 static int mdc_get_info_rpc(struct obd_export *exp,
2013 u32 keylen, void *key,
2014 u32 vallen, void *val)
2016 struct obd_import *imp = class_exp2cliimp(exp);
2017 struct ptlrpc_request *req;
2022 req = ptlrpc_request_alloc(imp, &RQF_MDS_GET_INFO);
2026 req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_KEY,
2027 RCL_CLIENT, keylen);
2028 req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VALLEN,
2029 RCL_CLIENT, sizeof(vallen));
2031 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GET_INFO);
2033 ptlrpc_request_free(req);
2037 tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_KEY);
2038 memcpy(tmp, key, keylen);
2039 tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_VALLEN);
2040 memcpy(tmp, &vallen, sizeof(vallen));
2042 req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VAL,
2043 RCL_SERVER, vallen);
2044 ptlrpc_request_set_replen(req);
2046 rc = ptlrpc_queue_wait(req);
2047 /* -EREMOTE means the get_info result is partial, and it needs to
2048 * continue on another MDT, see fid2path part in lmv_iocontrol */
2049 if (rc == 0 || rc == -EREMOTE) {
2050 tmp = req_capsule_server_get(&req->rq_pill, &RMF_GETINFO_VAL);
2051 memcpy(val, tmp, vallen);
2052 if (ptlrpc_rep_need_swab(req)) {
2053 if (KEY_IS(KEY_FID2PATH))
2054 lustre_swab_fid2path(val);
2057 ptlrpc_req_finished(req);
2062 static void lustre_swab_hai(struct hsm_action_item *h)
2064 __swab32s(&h->hai_len);
2065 __swab32s(&h->hai_action);
2066 lustre_swab_lu_fid(&h->hai_fid);
2067 lustre_swab_lu_fid(&h->hai_dfid);
2068 __swab64s(&h->hai_cookie);
2069 __swab64s(&h->hai_extent.offset);
2070 __swab64s(&h->hai_extent.length);
2071 __swab64s(&h->hai_gid);
2074 static void lustre_swab_hal(struct hsm_action_list *h)
2076 struct hsm_action_item *hai;
2079 __swab32s(&h->hal_version);
2080 __swab32s(&h->hal_count);
2081 __swab32s(&h->hal_archive_id);
2082 __swab64s(&h->hal_flags);
2084 for (i = 0; i < h->hal_count; i++, hai = hai_next(hai))
2085 lustre_swab_hai(hai);
2088 static void lustre_swab_kuch(struct kuc_hdr *l)
2090 __swab16s(&l->kuc_magic);
2091 /* __u8 l->kuc_transport */
2092 __swab16s(&l->kuc_msgtype);
2093 __swab16s(&l->kuc_msglen);
2096 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
2097 struct lustre_kernelcomm *lk)
2099 struct obd_import *imp = class_exp2cliimp(exp);
2100 __u32 archive = lk->lk_data;
2103 if (lk->lk_group != KUC_GRP_HSM) {
2104 CERROR("Bad copytool group %d\n", lk->lk_group);
2108 CDEBUG(D_HSM, "CT start r%d w%d u%d g%d f%#x\n", lk->lk_rfd, lk->lk_wfd,
2109 lk->lk_uid, lk->lk_group, lk->lk_flags);
2111 if (lk->lk_flags & LK_FLG_STOP) {
2112 /* Unregister with the coordinator */
2113 rc = mdc_ioc_hsm_ct_unregister(imp);
2115 rc = mdc_ioc_hsm_ct_register(imp, archive);
2122 * Send a message to any listening copytools
2123 * @param val KUC message (kuc_hdr + hsm_action_list)
2124 * @param len total length of message
2126 static int mdc_hsm_copytool_send(size_t len, void *val)
2128 struct kuc_hdr *lh = (struct kuc_hdr *)val;
2129 struct hsm_action_list *hal = (struct hsm_action_list *)(lh + 1);
2133 if (len < sizeof(*lh) + sizeof(*hal)) {
2134 CERROR("Short HSM message %zu < %zu\n", len,
2135 sizeof(*lh) + sizeof(*hal));
2138 if (lh->kuc_magic == __swab16(KUC_MAGIC)) {
2139 lustre_swab_kuch(lh);
2140 lustre_swab_hal(hal);
2141 } else if (lh->kuc_magic != KUC_MAGIC) {
2142 CERROR("Bad magic %x!=%x\n", lh->kuc_magic, KUC_MAGIC);
2146 CDEBUG(D_HSM, " Received message mg=%x t=%d m=%d l=%d actions=%d "
2148 lh->kuc_magic, lh->kuc_transport, lh->kuc_msgtype,
2149 lh->kuc_msglen, hal->hal_count, hal->hal_fsname);
2151 /* Broadcast to HSM listeners */
2152 rc = libcfs_kkuc_group_put(KUC_GRP_HSM, lh);
2158 * callback function passed to kuc for re-registering each HSM copytool
2159 * running on MDC, after MDT shutdown/recovery.
2160 * @param data copytool registration data
2161 * @param cb_arg callback argument (obd_import)
2163 static int mdc_hsm_ct_reregister(void *data, void *cb_arg)
2165 struct kkuc_ct_data *kcd = data;
2166 struct obd_import *imp = (struct obd_import *)cb_arg;
2169 if (kcd == NULL || kcd->kcd_magic != KKUC_CT_DATA_MAGIC)
2172 if (!obd_uuid_equals(&kcd->kcd_uuid, &imp->imp_obd->obd_uuid))
2175 CDEBUG(D_HA, "%s: recover copytool registration to MDT (archive=%#x)\n",
2176 imp->imp_obd->obd_name, kcd->kcd_archive);
2177 rc = mdc_ioc_hsm_ct_register(imp, kcd->kcd_archive);
2179 /* ignore error if the copytool is already registered */
2180 return (rc == -EEXIST) ? 0 : rc;
2184 * Re-establish all kuc contexts with MDT
2185 * after MDT shutdown/recovery.
2187 static int mdc_kuc_reregister(struct obd_import *imp)
2189 /* re-register HSM agents */
2190 return libcfs_kkuc_group_foreach(KUC_GRP_HSM, mdc_hsm_ct_reregister,
2194 static int mdc_set_info_async(const struct lu_env *env,
2195 struct obd_export *exp,
2196 u32 keylen, void *key,
2197 u32 vallen, void *val,
2198 struct ptlrpc_request_set *set)
2200 struct obd_import *imp = class_exp2cliimp(exp);
2204 if (KEY_IS(KEY_READ_ONLY)) {
2205 if (vallen != sizeof(int))
2208 spin_lock(&imp->imp_lock);
2209 if (*((int *)val)) {
2210 imp->imp_connect_flags_orig |= OBD_CONNECT_RDONLY;
2211 imp->imp_connect_data.ocd_connect_flags |=
2214 imp->imp_connect_flags_orig &= ~OBD_CONNECT_RDONLY;
2215 imp->imp_connect_data.ocd_connect_flags &=
2216 ~OBD_CONNECT_RDONLY;
2218 spin_unlock(&imp->imp_lock);
2220 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2221 keylen, key, vallen, val, set);
2224 if (KEY_IS(KEY_SPTLRPC_CONF)) {
2225 sptlrpc_conf_client_adapt(exp->exp_obd);
2228 if (KEY_IS(KEY_FLUSH_CTX)) {
2229 sptlrpc_import_flush_my_ctx(imp);
2232 if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
2233 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2234 keylen, key, vallen, val, set);
2237 if (KEY_IS(KEY_HSM_COPYTOOL_SEND)) {
2238 rc = mdc_hsm_copytool_send(vallen, val);
2242 if (KEY_IS(KEY_DEFAULT_EASIZE)) {
2243 __u32 *default_easize = val;
2245 exp->exp_obd->u.cli.cl_default_mds_easize = *default_easize;
2249 CERROR("Unknown key %s\n", (char *)key);
2253 static int mdc_get_info(const struct lu_env *env, struct obd_export *exp,
2254 __u32 keylen, void *key, __u32 *vallen, void *val)
2258 if (KEY_IS(KEY_MAX_EASIZE)) {
2259 __u32 mdsize, *max_easize;
2261 if (*vallen != sizeof(int))
2263 mdsize = *(__u32 *)val;
2264 if (mdsize > exp->exp_obd->u.cli.cl_max_mds_easize)
2265 exp->exp_obd->u.cli.cl_max_mds_easize = mdsize;
2267 *max_easize = exp->exp_obd->u.cli.cl_max_mds_easize;
2269 } else if (KEY_IS(KEY_DEFAULT_EASIZE)) {
2270 __u32 *default_easize;
2272 if (*vallen != sizeof(int))
2274 default_easize = val;
2275 *default_easize = exp->exp_obd->u.cli.cl_default_mds_easize;
2277 } else if (KEY_IS(KEY_CONN_DATA)) {
2278 struct obd_import *imp = class_exp2cliimp(exp);
2279 struct obd_connect_data *data = val;
2281 if (*vallen != sizeof(*data))
2284 *data = imp->imp_connect_data;
2286 } else if (KEY_IS(KEY_TGT_COUNT)) {
2287 *((__u32 *)val) = 1;
2291 rc = mdc_get_info_rpc(exp, keylen, key, *vallen, val);
2296 static int mdc_fsync(struct obd_export *exp, const struct lu_fid *fid,
2297 struct ptlrpc_request **request)
2299 struct ptlrpc_request *req;
2304 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_SYNC);
2308 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_SYNC);
2310 ptlrpc_request_free(req);
2314 mdc_pack_body(req, fid, 0, 0, -1, 0);
2316 ptlrpc_request_set_replen(req);
2318 rc = ptlrpc_queue_wait(req);
2320 ptlrpc_req_finished(req);
2326 static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
2327 enum obd_import_event event)
2331 LASSERT(imp->imp_obd == obd);
2335 case IMP_EVENT_INACTIVE: {
2336 struct client_obd *cli = &obd->u.cli;
2338 * Flush current sequence to make client obtain new one
2339 * from server in case of disconnect/reconnect.
2341 down_read(&cli->cl_seq_rwsem);
2343 seq_client_flush(cli->cl_seq);
2344 up_read(&cli->cl_seq_rwsem);
2346 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE);
2349 case IMP_EVENT_INVALIDATE: {
2350 struct ldlm_namespace *ns = obd->obd_namespace;
2352 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
2356 case IMP_EVENT_ACTIVE:
2357 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE);
2358 /* redo the kuc registration after reconnecting */
2360 rc = mdc_kuc_reregister(imp);
2363 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_OCD);
2365 case IMP_EVENT_DISCON:
2366 case IMP_EVENT_DEACTIVATE:
2367 case IMP_EVENT_ACTIVATE:
2370 CERROR("Unknown import event %x\n", event);
2376 int mdc_fid_alloc(const struct lu_env *env, struct obd_export *exp,
2377 struct lu_fid *fid, struct md_op_data *op_data)
2379 struct client_obd *cli = &exp->exp_obd->u.cli;
2384 down_read(&cli->cl_seq_rwsem);
2386 rc = seq_client_alloc_fid(env, cli->cl_seq, fid);
2387 up_read(&cli->cl_seq_rwsem);
2392 static struct obd_uuid *mdc_get_uuid(struct obd_export *exp)
2394 struct client_obd *cli = &exp->exp_obd->u.cli;
2395 return &cli->cl_target_uuid;
2399 * Determine whether the lock can be canceled before replaying it during
2400 * recovery, non zero value will be return if the lock can be canceled,
2401 * or zero returned for not
2403 static int mdc_cancel_weight(struct ldlm_lock *lock)
2405 if (lock->l_resource->lr_type != LDLM_IBITS)
2408 /* FIXME: if we ever get into a situation where there are too many
2409 * opened files with open locks on a single node, then we really
2410 * should replay these open locks to reget it */
2411 if (lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_OPEN)
2417 static int mdc_resource_inode_free(struct ldlm_resource *res)
2419 if (res->lr_lvb_inode)
2420 res->lr_lvb_inode = NULL;
2425 static struct ldlm_valblock_ops inode_lvbo = {
2426 .lvbo_free = mdc_resource_inode_free
2429 static int mdc_llog_init(struct obd_device *obd)
2431 struct obd_llog_group *olg = &obd->obd_olg;
2432 struct llog_ctxt *ctxt;
2437 rc = llog_setup(NULL, obd, olg, LLOG_CHANGELOG_REPL_CTXT, obd,
2442 ctxt = llog_group_get_ctxt(olg, LLOG_CHANGELOG_REPL_CTXT);
2443 llog_initiator_connect(ctxt);
2444 llog_ctxt_put(ctxt);
2449 static void mdc_llog_finish(struct obd_device *obd)
2451 struct llog_ctxt *ctxt;
2455 ctxt = llog_get_context(obd, LLOG_CHANGELOG_REPL_CTXT);
2457 llog_cleanup(NULL, ctxt);
2462 static int mdc_setup(struct obd_device *obd, struct lustre_cfg *cfg)
2467 rc = ptlrpcd_addref();
2471 rc = client_obd_setup(obd, cfg);
2473 GOTO(err_ptlrpcd_decref, rc);
2474 #ifdef CONFIG_PROC_FS
2475 obd->obd_vars = lprocfs_mdc_obd_vars;
2476 lprocfs_obd_setup(obd);
2477 lprocfs_alloc_md_stats(obd, 0);
2479 sptlrpc_lprocfs_cliobd_attach(obd);
2480 ptlrpc_lprocfs_register_obd(obd);
2482 ns_register_cancel(obd->obd_namespace, mdc_cancel_weight);
2484 obd->obd_namespace->ns_lvbo = &inode_lvbo;
2486 rc = mdc_llog_init(obd);
2488 CERROR("%s: failed to setup llogging subsystems: rc = %d\n",
2490 GOTO(err_mdc_cleanup, rc);
2493 rc = mdc_changelog_cdev_init(obd);
2495 CERROR("%s: failed to setup changelog char device: rc = %d\n",
2497 GOTO(err_mdc_cleanup, rc);
2503 client_obd_cleanup(obd);
2512 /* Initialize the default and maximum LOV EA sizes. This allows
2513 * us to make MDS RPCs with large enough reply buffers to hold a default
2514 * sized EA without having to calculate this (via a call into the
2515 * LOV + OSCs) each time we make an RPC. The maximum size is also tracked
2516 * but not used to avoid wastefully vmalloc()'ing large reply buffers when
2517 * a large number of stripes is possible. If a larger reply buffer is
2518 * required it will be reallocated in the ptlrpc layer due to overflow.
2520 static int mdc_init_ea_size(struct obd_export *exp, __u32 easize,
2523 struct obd_device *obd = exp->exp_obd;
2524 struct client_obd *cli = &obd->u.cli;
2527 if (cli->cl_max_mds_easize < easize)
2528 cli->cl_max_mds_easize = easize;
2530 if (cli->cl_default_mds_easize < def_easize)
2531 cli->cl_default_mds_easize = def_easize;
2536 static int mdc_precleanup(struct obd_device *obd)
2540 /* Failsafe, ok if racy */
2541 if (obd->obd_type->typ_refcnt <= 1)
2542 libcfs_kkuc_group_rem(0, KUC_GRP_HSM);
2544 mdc_changelog_cdev_finish(obd);
2546 obd_cleanup_client_import(obd);
2547 ptlrpc_lprocfs_unregister_obd(obd);
2548 lprocfs_obd_cleanup(obd);
2549 lprocfs_free_md_stats(obd);
2550 mdc_llog_finish(obd);
2554 static int mdc_cleanup(struct obd_device *obd)
2558 return client_obd_cleanup(obd);
2561 static int mdc_process_config(struct obd_device *obd, size_t len, void *buf)
2563 struct lustre_cfg *lcfg = buf;
2564 int rc = class_process_proc_param(PARAM_MDC, obd->obd_vars, lcfg, obd);
2565 return (rc > 0 ? 0: rc);
2568 static struct obd_ops mdc_obd_ops = {
2569 .o_owner = THIS_MODULE,
2570 .o_setup = mdc_setup,
2571 .o_precleanup = mdc_precleanup,
2572 .o_cleanup = mdc_cleanup,
2573 .o_add_conn = client_import_add_conn,
2574 .o_del_conn = client_import_del_conn,
2575 .o_connect = client_connect_import,
2576 .o_disconnect = client_disconnect_export,
2577 .o_iocontrol = mdc_iocontrol,
2578 .o_set_info_async = mdc_set_info_async,
2579 .o_statfs = mdc_statfs,
2580 .o_fid_init = client_fid_init,
2581 .o_fid_fini = client_fid_fini,
2582 .o_fid_alloc = mdc_fid_alloc,
2583 .o_import_event = mdc_import_event,
2584 .o_get_info = mdc_get_info,
2585 .o_process_config = mdc_process_config,
2586 .o_get_uuid = mdc_get_uuid,
2587 .o_quotactl = mdc_quotactl,
2590 static struct md_ops mdc_md_ops = {
2591 .m_get_root = mdc_get_root,
2592 .m_null_inode = mdc_null_inode,
2593 .m_close = mdc_close,
2594 .m_create = mdc_create,
2595 .m_enqueue = mdc_enqueue,
2596 .m_getattr = mdc_getattr,
2597 .m_getattr_name = mdc_getattr_name,
2598 .m_intent_lock = mdc_intent_lock,
2600 .m_rename = mdc_rename,
2601 .m_setattr = mdc_setattr,
2602 .m_setxattr = mdc_setxattr,
2603 .m_getxattr = mdc_getxattr,
2604 .m_fsync = mdc_fsync,
2605 .m_read_page = mdc_read_page,
2606 .m_unlink = mdc_unlink,
2607 .m_cancel_unused = mdc_cancel_unused,
2608 .m_init_ea_size = mdc_init_ea_size,
2609 .m_set_lock_data = mdc_set_lock_data,
2610 .m_lock_match = mdc_lock_match,
2611 .m_get_lustre_md = mdc_get_lustre_md,
2612 .m_free_lustre_md = mdc_free_lustre_md,
2613 .m_set_open_replay_data = mdc_set_open_replay_data,
2614 .m_clear_open_replay_data = mdc_clear_open_replay_data,
2615 .m_intent_getattr_async = mdc_intent_getattr_async,
2616 .m_revalidate_lock = mdc_revalidate_lock
2619 static int __init mdc_init(void)
2621 return class_register_type(&mdc_obd_ops, &mdc_md_ops, true, NULL,
2622 LUSTRE_MDC_NAME, NULL);
2625 static void __exit mdc_exit(void)
2627 class_unregister_type(LUSTRE_MDC_NAME);
2630 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
2631 MODULE_DESCRIPTION("Lustre Metadata Client");
2632 MODULE_VERSION(LUSTRE_VERSION_STRING);
2633 MODULE_LICENSE("GPL");
2635 module_init(mdc_init);
2636 module_exit(mdc_exit);