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, 2015, 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/module.h>
36 # include <linux/pagemap.h>
37 # include <linux/miscdevice.h>
38 # include <linux/init.h>
39 # include <linux/utsname.h>
41 #include "../include/cl_object.h"
42 #include "../include/llog_swab.h"
43 #include "../include/lprocfs_status.h"
44 #include "../include/lustre_acl.h"
45 #include "../include/lustre_fid.h"
46 #include "../include/lustre/lustre_ioctl.h"
47 #include "../include/lustre_kernelcomm.h"
48 #include "../include/lustre_lmv.h"
49 #include "../include/lustre_log.h"
50 #include "../include/lustre_param.h"
51 #include "../include/lustre_swab.h"
52 #include "../include/obd_class.h"
54 #include "mdc_internal.h"
56 #define REQUEST_MINOR 244
58 static int mdc_cleanup(struct obd_device *obd);
60 static inline int mdc_queue_wait(struct ptlrpc_request *req)
62 struct client_obd *cli = &req->rq_import->imp_obd->u.cli;
65 /* obd_get_request_slot() ensures that this client has no more
66 * than cl_max_rpcs_in_flight RPCs simultaneously inf light
69 rc = obd_get_request_slot(cli);
73 rc = ptlrpc_queue_wait(req);
74 obd_put_request_slot(cli);
79 static int mdc_getstatus(struct obd_export *exp, struct lu_fid *rootfid)
81 struct ptlrpc_request *req;
82 struct mdt_body *body;
85 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
87 LUSTRE_MDS_VERSION, MDS_GETSTATUS);
91 mdc_pack_body(req, NULL, 0, 0, -1, 0);
92 req->rq_send_state = LUSTRE_IMP_FULL;
94 ptlrpc_request_set_replen(req);
96 rc = ptlrpc_queue_wait(req);
100 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
106 *rootfid = body->mbo_fid1;
108 "root fid="DFID", last_committed=%llu\n",
110 lustre_msg_get_last_committed(req->rq_repmsg));
112 ptlrpc_req_finished(req);
117 * This function now is known to always saying that it will receive 4 buffers
118 * from server. Even for cases when acl_size and md_size is zero, RPC header
119 * will contain 4 fields and RPC itself will contain zero size fields. This is
120 * because mdt_getattr*() _always_ returns 4 fields, but if acl is not needed
121 * and thus zero, it shrinks it, making zero size. The same story about
122 * md_size. And this is course of problem when client waits for smaller number
123 * of fields. This issue will be fixed later when client gets aware of RPC
126 static int mdc_getattr_common(struct obd_export *exp,
127 struct ptlrpc_request *req)
129 struct req_capsule *pill = &req->rq_pill;
130 struct mdt_body *body;
134 /* Request message already built. */
135 rc = ptlrpc_queue_wait(req);
139 /* sanity check for the reply */
140 body = req_capsule_server_get(pill, &RMF_MDT_BODY);
144 CDEBUG(D_NET, "mode: %o\n", body->mbo_mode);
146 mdc_update_max_ea_from_body(exp, body);
147 if (body->mbo_eadatasize != 0) {
148 eadata = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
149 body->mbo_eadatasize);
157 static int mdc_getattr(struct obd_export *exp, struct md_op_data *op_data,
158 struct ptlrpc_request **request)
160 struct ptlrpc_request *req;
163 /* Single MDS without an LMV case */
164 if (op_data->op_flags & MF_GET_MDT_IDX) {
169 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
173 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
175 ptlrpc_request_free(req);
179 mdc_pack_body(req, &op_data->op_fid1, op_data->op_valid,
180 op_data->op_mode, -1, 0);
182 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
184 ptlrpc_request_set_replen(req);
186 rc = mdc_getattr_common(exp, req);
188 ptlrpc_req_finished(req);
194 static int mdc_getattr_name(struct obd_export *exp, struct md_op_data *op_data,
195 struct ptlrpc_request **request)
197 struct ptlrpc_request *req;
201 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
202 &RQF_MDS_GETATTR_NAME);
206 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
207 op_data->op_namelen + 1);
209 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR_NAME);
211 ptlrpc_request_free(req);
215 mdc_pack_body(req, &op_data->op_fid1, op_data->op_valid,
216 op_data->op_mode, op_data->op_suppgids[0], 0);
218 if (op_data->op_name) {
219 char *name = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
221 LASSERT(strnlen(op_data->op_name, op_data->op_namelen) ==
222 op_data->op_namelen);
223 memcpy(name, op_data->op_name, op_data->op_namelen);
226 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
228 ptlrpc_request_set_replen(req);
230 rc = mdc_getattr_common(exp, req);
232 ptlrpc_req_finished(req);
238 static int mdc_xattr_common(struct obd_export *exp,
239 const struct req_format *fmt,
240 const struct lu_fid *fid,
241 int opcode, u64 valid,
242 const char *xattr_name, const char *input,
243 int input_size, int output_size, int flags,
244 __u32 suppgid, struct ptlrpc_request **request)
246 struct ptlrpc_request *req;
247 int xattr_namelen = 0;
252 req = ptlrpc_request_alloc(class_exp2cliimp(exp), fmt);
257 xattr_namelen = strlen(xattr_name) + 1;
258 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
263 req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT,
267 /* Flush local XATTR locks to get rid of a possible cancel RPC */
268 if (opcode == MDS_REINT && fid_is_sane(fid) &&
269 exp->exp_connect_data.ocd_ibits_known & MDS_INODELOCK_XATTR) {
273 /* Without that packing would fail */
275 req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
278 count = mdc_resource_get_unused(exp, fid,
280 MDS_INODELOCK_XATTR);
282 rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
284 ptlrpc_request_free(req);
288 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, opcode);
290 ptlrpc_request_free(req);
295 if (opcode == MDS_REINT) {
296 struct mdt_rec_setxattr *rec;
298 CLASSERT(sizeof(struct mdt_rec_setxattr) ==
299 sizeof(struct mdt_rec_reint));
300 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
301 rec->sx_opcode = REINT_SETXATTR;
302 rec->sx_fsuid = from_kuid(&init_user_ns, current_fsuid());
303 rec->sx_fsgid = from_kgid(&init_user_ns, current_fsgid());
304 rec->sx_cap = cfs_curproc_cap_pack();
305 rec->sx_suppgid1 = suppgid;
306 rec->sx_suppgid2 = -1;
308 rec->sx_valid = valid | OBD_MD_FLCTIME;
309 rec->sx_time = ktime_get_real_seconds();
310 rec->sx_size = output_size;
311 rec->sx_flags = flags;
314 mdc_pack_body(req, fid, valid, output_size, suppgid, flags);
318 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
319 memcpy(tmp, xattr_name, xattr_namelen);
322 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
323 memcpy(tmp, input, input_size);
326 if (req_capsule_has_field(&req->rq_pill, &RMF_EADATA, RCL_SERVER))
327 req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
328 RCL_SERVER, output_size);
329 ptlrpc_request_set_replen(req);
332 if (opcode == MDS_REINT)
333 mdc_get_mod_rpc_slot(req, NULL);
335 rc = ptlrpc_queue_wait(req);
337 if (opcode == MDS_REINT)
338 mdc_put_mod_rpc_slot(req, NULL);
341 ptlrpc_req_finished(req);
347 static int mdc_setxattr(struct obd_export *exp, const struct lu_fid *fid,
348 u64 valid, const char *xattr_name,
349 const char *input, int input_size, int output_size,
350 int flags, __u32 suppgid,
351 struct ptlrpc_request **request)
353 return mdc_xattr_common(exp, &RQF_MDS_REINT_SETXATTR,
354 fid, MDS_REINT, valid, xattr_name,
355 input, input_size, output_size, flags,
359 static int mdc_getxattr(struct obd_export *exp, const struct lu_fid *fid,
360 u64 valid, const char *xattr_name,
361 const char *input, int input_size, int output_size,
362 int flags, struct ptlrpc_request **request)
364 return mdc_xattr_common(exp, &RQF_MDS_GETXATTR,
365 fid, MDS_GETXATTR, valid, xattr_name,
366 input, input_size, output_size, flags,
370 #ifdef CONFIG_FS_POSIX_ACL
371 static int mdc_unpack_acl(struct ptlrpc_request *req, struct lustre_md *md)
373 struct req_capsule *pill = &req->rq_pill;
374 struct mdt_body *body = md->body;
375 struct posix_acl *acl;
379 if (!body->mbo_aclsize)
382 buf = req_capsule_server_sized_get(pill, &RMF_ACL, body->mbo_aclsize);
387 acl = posix_acl_from_xattr(&init_user_ns, buf, body->mbo_aclsize);
393 CERROR("convert xattr to acl: %d\n", rc);
397 rc = posix_acl_valid(&init_user_ns, acl);
399 CERROR("validate acl: %d\n", rc);
400 posix_acl_release(acl);
408 #define mdc_unpack_acl(req, md) 0
411 static int mdc_get_lustre_md(struct obd_export *exp,
412 struct ptlrpc_request *req,
413 struct obd_export *dt_exp,
414 struct obd_export *md_exp,
415 struct lustre_md *md)
417 struct req_capsule *pill = &req->rq_pill;
421 memset(md, 0, sizeof(*md));
423 md->body = req_capsule_server_get(pill, &RMF_MDT_BODY);
425 if (md->body->mbo_valid & OBD_MD_FLEASIZE) {
426 if (!S_ISREG(md->body->mbo_mode)) {
428 "OBD_MD_FLEASIZE set, should be a regular file, but is not\n");
433 if (md->body->mbo_eadatasize == 0) {
435 "OBD_MD_FLEASIZE set, but eadatasize 0\n");
440 md->layout.lb_len = md->body->mbo_eadatasize;
441 md->layout.lb_buf = req_capsule_server_sized_get(pill,
444 if (!md->layout.lb_buf) {
448 } else if (md->body->mbo_valid & OBD_MD_FLDIREA) {
449 const union lmv_mds_md *lmv;
452 if (!S_ISDIR(md->body->mbo_mode)) {
454 "OBD_MD_FLDIREA set, should be a directory, but is not\n");
459 lmv_size = md->body->mbo_eadatasize;
462 "OBD_MD_FLDIREA is set, but eadatasize 0\n");
465 if (md->body->mbo_valid & OBD_MD_MEA) {
466 lmv = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
473 rc = md_unpackmd(md_exp, &md->lmv, lmv, lmv_size);
477 if (rc < (typeof(rc))sizeof(*md->lmv)) {
479 "size too small: rc < sizeof(*md->lmv) (%d < %d)\n",
480 rc, (int)sizeof(*md->lmv));
488 if (md->body->mbo_valid & OBD_MD_FLACL) {
489 /* for ACL, it's possible that FLACL is set but aclsize is zero.
490 * only when aclsize != 0 there's an actual segment for ACL
493 if (md->body->mbo_aclsize) {
494 rc = mdc_unpack_acl(req, md);
497 #ifdef CONFIG_FS_POSIX_ACL
499 md->posix_acl = NULL;
506 #ifdef CONFIG_FS_POSIX_ACL
507 posix_acl_release(md->posix_acl);
513 static int mdc_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
518 void mdc_replay_open(struct ptlrpc_request *req)
520 struct md_open_data *mod = req->rq_cb_data;
521 struct ptlrpc_request *close_req;
522 struct obd_client_handle *och;
523 struct lustre_handle old;
524 struct mdt_body *body;
527 DEBUG_REQ(D_ERROR, req,
528 "Can't properly replay without open data.");
532 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
536 struct lustre_handle *file_fh;
538 LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
540 file_fh = &och->och_fh;
541 CDEBUG(D_HA, "updating handle from %#llx to %#llx\n",
542 file_fh->cookie, body->mbo_handle.cookie);
544 *file_fh = body->mbo_handle;
546 close_req = mod->mod_close_req;
548 __u32 opc = lustre_msg_get_opc(close_req->rq_reqmsg);
549 struct mdt_ioepoch *epoch;
551 LASSERT(opc == MDS_CLOSE);
552 epoch = req_capsule_client_get(&close_req->rq_pill,
557 LASSERT(!memcmp(&old, &epoch->mio_handle, sizeof(old)));
558 DEBUG_REQ(D_HA, close_req, "updating close body with new fh");
559 epoch->mio_handle = body->mbo_handle;
563 void mdc_commit_open(struct ptlrpc_request *req)
565 struct md_open_data *mod = req->rq_cb_data;
571 * No need to touch md_open_data::mod_och, it holds a reference on
572 * \var mod and will zero references to each other, \var mod will be
573 * freed after that when md_open_data::mod_och will put the reference.
577 * Do not let open request to disappear as it still may be needed
578 * for close rpc to happen (it may happen on evict only, otherwise
579 * ptlrpc_request::rq_replay does not let mdc_commit_open() to be
580 * called), just mark this rpc as committed to distinguish these 2
581 * cases, see mdc_close() for details. The open request reference will
582 * be put along with freeing \var mod.
584 ptlrpc_request_addref(req);
585 spin_lock(&req->rq_lock);
586 req->rq_committed = 1;
587 spin_unlock(&req->rq_lock);
588 req->rq_cb_data = NULL;
592 int mdc_set_open_replay_data(struct obd_export *exp,
593 struct obd_client_handle *och,
594 struct lookup_intent *it)
596 struct md_open_data *mod;
597 struct mdt_rec_create *rec;
598 struct mdt_body *body;
599 struct ptlrpc_request *open_req = it->it_request;
600 struct obd_import *imp = open_req->rq_import;
602 if (!open_req->rq_replay)
605 rec = req_capsule_client_get(&open_req->rq_pill, &RMF_REC_REINT);
606 body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
608 /* Incoming message in my byte order (it's been swabbed). */
609 /* Outgoing messages always in my byte order. */
612 /* Only if the import is replayable, we set replay_open data */
613 if (och && imp->imp_replayable) {
614 mod = obd_mod_alloc();
616 DEBUG_REQ(D_ERROR, open_req,
617 "Can't allocate md_open_data");
622 * Take a reference on \var mod, to be freed on mdc_close().
623 * It protects \var mod from being freed on eviction (commit
624 * callback is called despite rq_replay flag).
625 * Another reference for \var och.
630 spin_lock(&open_req->rq_lock);
633 mod->mod_is_create = it_disposition(it, DISP_OPEN_CREATE) ||
634 it_disposition(it, DISP_OPEN_STRIPE);
635 mod->mod_open_req = open_req;
636 open_req->rq_cb_data = mod;
637 open_req->rq_commit_cb = mdc_commit_open;
638 spin_unlock(&open_req->rq_lock);
641 rec->cr_fid2 = body->mbo_fid1;
642 rec->cr_ioepoch = body->mbo_ioepoch;
643 rec->cr_old_handle.cookie = body->mbo_handle.cookie;
644 open_req->rq_replay_cb = mdc_replay_open;
645 if (!fid_is_sane(&body->mbo_fid1)) {
646 DEBUG_REQ(D_ERROR, open_req,
647 "Saving replay request with insane fid");
651 DEBUG_REQ(D_RPCTRACE, open_req, "Set up open replay data");
655 static void mdc_free_open(struct md_open_data *mod)
659 if (mod->mod_is_create == 0 &&
660 imp_connect_disp_stripe(mod->mod_open_req->rq_import))
664 * No reason to asssert here if the open request has
665 * rq_replay == 1. It means that mdc_close failed, and
666 * close request wasn`t sent. It is not fatal to client.
667 * The worst thing is eviction if the client gets open lock
669 DEBUG_REQ(D_RPCTRACE, mod->mod_open_req,
670 "free open request rq_replay = %d\n",
671 mod->mod_open_req->rq_replay);
673 ptlrpc_request_committed(mod->mod_open_req, committed);
674 if (mod->mod_close_req)
675 ptlrpc_request_committed(mod->mod_close_req, committed);
678 static int mdc_clear_open_replay_data(struct obd_export *exp,
679 struct obd_client_handle *och)
681 struct md_open_data *mod = och->och_mod;
684 * It is possible to not have \var mod in a case of eviction between
685 * lookup and ll_file_open().
690 LASSERT(mod != LP_POISON);
691 LASSERT(mod->mod_open_req);
701 static int mdc_close(struct obd_export *exp, struct md_op_data *op_data,
702 struct md_open_data *mod, struct ptlrpc_request **request)
704 struct obd_device *obd = class_exp2obd(exp);
705 struct ptlrpc_request *req;
706 struct req_format *req_fmt;
710 if (op_data->op_bias & MDS_HSM_RELEASE) {
711 req_fmt = &RQF_MDS_INTENT_CLOSE;
713 /* allocate a FID for volatile file */
714 rc = mdc_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
716 CERROR("%s: "DFID" failed to allocate FID: %d\n",
717 obd->obd_name, PFID(&op_data->op_fid1), rc);
718 /* save the errcode and proceed to close */
721 } else if (op_data->op_bias & MDS_CLOSE_LAYOUT_SWAP) {
722 req_fmt = &RQF_MDS_INTENT_CLOSE;
724 req_fmt = &RQF_MDS_CLOSE;
728 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_CLOSE))
731 req = ptlrpc_request_alloc(class_exp2cliimp(exp), req_fmt);
733 /* Ensure that this close's handle is fixed up during replay. */
735 LASSERTF(mod->mod_open_req &&
736 mod->mod_open_req->rq_type != LI_POISON,
737 "POISONED open %p!\n", mod->mod_open_req);
739 mod->mod_close_req = req;
741 DEBUG_REQ(D_HA, mod->mod_open_req, "matched open");
742 /* We no longer want to preserve this open for replay even
743 * though the open was committed. b=3632, b=3633
745 spin_lock(&mod->mod_open_req->rq_lock);
746 mod->mod_open_req->rq_replay = 0;
747 spin_unlock(&mod->mod_open_req->rq_lock);
750 "couldn't find open req; expecting close error\n");
754 * TODO: repeat close after errors
756 CWARN("%s: close of FID "DFID" failed, file reference will be dropped when this client unmounts or is evicted\n",
757 obd->obd_name, PFID(&op_data->op_fid1));
762 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_CLOSE);
764 ptlrpc_request_free(req);
769 * To avoid a livelock (bug 7034), we need to send CLOSE RPCs to a
770 * portal whose threads are not taking any DLM locks and are therefore
773 req->rq_request_portal = MDS_READPAGE_PORTAL;
774 ptlrpc_at_set_req_timeout(req);
776 mdc_close_pack(req, op_data);
778 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
779 obd->u.cli.cl_default_mds_easize);
781 ptlrpc_request_set_replen(req);
783 mdc_get_mod_rpc_slot(req, NULL);
784 rc = ptlrpc_queue_wait(req);
785 mdc_put_mod_rpc_slot(req, NULL);
787 if (!req->rq_repmsg) {
788 CDEBUG(D_RPCTRACE, "request failed to send: %p, %d\n", req,
791 rc = req->rq_status ?: -EIO;
792 } else if (rc == 0 || rc == -EAGAIN) {
793 struct mdt_body *body;
795 rc = lustre_msg_get_status(req->rq_repmsg);
796 if (lustre_msg_get_type(req->rq_repmsg) == PTL_RPC_MSG_ERR) {
797 DEBUG_REQ(D_ERROR, req,
798 "type == PTL_RPC_MSG_ERR, err = %d", rc);
802 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
805 } else if (rc == -ESTALE) {
807 * it can be allowed error after 3633 if open was committed and
808 * server failed before close was sent. Let's check if mod
809 * exists and return no error in that case
812 DEBUG_REQ(D_HA, req, "Reset ESTALE = %d", rc);
813 if (mod->mod_open_req->rq_committed)
821 mod->mod_close_req = NULL;
822 /* Since now, mod is accessed through open_req only,
823 * thus close req does not keep a reference on mod anymore.
828 return rc < 0 ? rc : saved_rc;
831 static int mdc_getpage(struct obd_export *exp, const struct lu_fid *fid,
832 u64 offset, struct page **pages, int npages,
833 struct ptlrpc_request **request)
835 struct ptlrpc_bulk_desc *desc;
836 struct ptlrpc_request *req;
837 wait_queue_head_t waitq;
838 struct l_wait_info lwi;
844 init_waitqueue_head(&waitq);
847 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_READPAGE);
851 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_READPAGE);
853 ptlrpc_request_free(req);
857 req->rq_request_portal = MDS_READPAGE_PORTAL;
858 ptlrpc_at_set_req_timeout(req);
860 desc = ptlrpc_prep_bulk_imp(req, npages, 1,
861 PTLRPC_BULK_PUT_SINK | PTLRPC_BULK_BUF_KIOV,
863 &ptlrpc_bulk_kiov_pin_ops);
865 ptlrpc_request_free(req);
869 /* NB req now owns desc and will free it when it gets freed */
870 for (i = 0; i < npages; i++)
871 desc->bd_frag_ops->add_kiov_frag(desc, pages[i], 0, PAGE_SIZE);
873 mdc_readdir_pack(req, offset, PAGE_SIZE * npages, fid);
875 ptlrpc_request_set_replen(req);
876 rc = ptlrpc_queue_wait(req);
878 ptlrpc_req_finished(req);
879 if (rc != -ETIMEDOUT)
883 if (!client_should_resend(resends, &exp->exp_obd->u.cli)) {
884 CERROR("%s: too many resend retries: rc = %d\n",
885 exp->exp_obd->obd_name, -EIO);
888 lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(resends), NULL, NULL,
890 l_wait_event(waitq, 0, &lwi);
895 rc = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk,
896 req->rq_bulk->bd_nob_transferred);
898 ptlrpc_req_finished(req);
902 if (req->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK) {
903 CERROR("%s: unexpected bytes transferred: %d (%ld expected)\n",
904 exp->exp_obd->obd_name, req->rq_bulk->bd_nob_transferred,
906 ptlrpc_req_finished(req);
914 static void mdc_release_page(struct page *page, int remove)
918 if (likely(page->mapping))
919 truncate_complete_page(page->mapping, page);
925 static struct page *mdc_page_locate(struct address_space *mapping, __u64 *hash,
926 __u64 *start, __u64 *end, int hash64)
929 * Complement of hash is used as an index so that
930 * radix_tree_gang_lookup() can be used to find a page with starting
931 * hash _smaller_ than one we are looking for.
933 unsigned long offset = hash_x_index(*hash, hash64);
937 spin_lock_irq(&mapping->tree_lock);
938 found = radix_tree_gang_lookup(&mapping->page_tree,
939 (void **)&page, offset, 1);
940 if (found > 0 && !radix_tree_exceptional_entry(page)) {
941 struct lu_dirpage *dp;
944 spin_unlock_irq(&mapping->tree_lock);
946 * In contrast to find_lock_page() we are sure that directory
947 * page cannot be truncated (while DLM lock is held) and,
948 * hence, can avoid restart.
950 * In fact, page cannot be locked here at all, because
951 * mdc_read_page_remote does synchronous io.
953 wait_on_page_locked(page);
954 if (PageUptodate(page)) {
956 if (BITS_PER_LONG == 32 && hash64) {
957 *start = le64_to_cpu(dp->ldp_hash_start) >> 32;
958 *end = le64_to_cpu(dp->ldp_hash_end) >> 32;
961 *start = le64_to_cpu(dp->ldp_hash_start);
962 *end = le64_to_cpu(dp->ldp_hash_end);
964 if (unlikely(*start == 1 && *hash == 0))
967 LASSERTF(*start <= *hash, "start = %#llx,end = %#llx,hash = %#llx\n",
968 *start, *end, *hash);
969 CDEBUG(D_VFSTRACE, "offset %lx [%#llx %#llx], hash %#llx\n",
970 offset, *start, *end, *hash);
973 mdc_release_page(page, 0);
975 } else if (*end != *start && *hash == *end) {
977 * upon hash collision, remove this page,
978 * otherwise put page reference, and
979 * mdc_read_page_remote() will issue RPC to
980 * fetch the page we want.
983 mdc_release_page(page,
984 le32_to_cpu(dp->ldp_flags) & LDF_COLLIDE);
989 page = ERR_PTR(-EIO);
992 spin_unlock_irq(&mapping->tree_lock);
999 * Adjust a set of pages, each page containing an array of lu_dirpages,
1000 * so that each page can be used as a single logical lu_dirpage.
1002 * A lu_dirpage is laid out as follows, where s = ldp_hash_start,
1003 * e = ldp_hash_end, f = ldp_flags, p = padding, and each "ent" is a
1004 * struct lu_dirent. It has size up to LU_PAGE_SIZE. The ldp_hash_end
1005 * value is used as a cookie to request the next lu_dirpage in a
1006 * directory listing that spans multiple pages (two in this example):
1009 * .|--------v------- -----.
1010 * |s|e|f|p|ent|ent| ... |ent|
1011 * '--|-------------- -----' Each PAGE contains a single
1012 * '------. lu_dirpage.
1013 * .---------v------- -----.
1014 * |s|e|f|p|ent| 0 | ... | 0 |
1015 * '----------------- -----'
1017 * However, on hosts where the native VM page size (PAGE_SIZE) is
1018 * larger than LU_PAGE_SIZE, a single host page may contain multiple
1019 * lu_dirpages. After reading the lu_dirpages from the MDS, the
1020 * ldp_hash_end of the first lu_dirpage refers to the one immediately
1021 * after it in the same PAGE (arrows simplified for brevity, but
1022 * in general e0==s1, e1==s2, etc.):
1024 * .-------------------- -----.
1025 * |s0|e0|f0|p|ent|ent| ... |ent|
1026 * |---v---------------- -----|
1027 * |s1|e1|f1|p|ent|ent| ... |ent|
1028 * |---v---------------- -----| Here, each PAGE contains
1029 * ... multiple lu_dirpages.
1030 * |---v---------------- -----|
1031 * |s'|e'|f'|p|ent|ent| ... |ent|
1032 * '---|---------------- -----'
1034 * .----------------------------.
1037 * This structure is transformed into a single logical lu_dirpage as follows:
1039 * - Replace e0 with e' so the request for the next lu_dirpage gets the page
1040 * labeled 'next PAGE'.
1042 * - Copy the LDF_COLLIDE flag from f' to f0 to correctly reflect whether
1043 * a hash collision with the next page exists.
1045 * - Adjust the lde_reclen of the ending entry of each lu_dirpage to span
1046 * to the first entry of the next lu_dirpage.
1048 #if PAGE_SIZE > LU_PAGE_SIZE
1049 static void mdc_adjust_dirpages(struct page **pages, int cfs_pgs, int lu_pgs)
1053 for (i = 0; i < cfs_pgs; i++) {
1054 struct lu_dirpage *dp = kmap(pages[i]);
1055 __u64 hash_end = le64_to_cpu(dp->ldp_hash_end);
1056 __u32 flags = le32_to_cpu(dp->ldp_flags);
1057 struct lu_dirpage *first = dp;
1058 struct lu_dirent *end_dirent = NULL;
1059 struct lu_dirent *ent;
1061 while (--lu_pgs > 0) {
1062 ent = lu_dirent_start(dp);
1063 for (end_dirent = ent; ent;
1064 end_dirent = ent, ent = lu_dirent_next(ent));
1066 /* Advance dp to next lu_dirpage. */
1067 dp = (struct lu_dirpage *)((char *)dp + LU_PAGE_SIZE);
1069 /* Check if we've reached the end of the CFS_PAGE. */
1070 if (!((unsigned long)dp & ~PAGE_MASK))
1073 /* Save the hash and flags of this lu_dirpage. */
1074 hash_end = le64_to_cpu(dp->ldp_hash_end);
1075 flags = le32_to_cpu(dp->ldp_flags);
1077 /* Check if lu_dirpage contains no entries. */
1082 * Enlarge the end entry lde_reclen from 0 to
1083 * first entry of next lu_dirpage.
1085 LASSERT(!le16_to_cpu(end_dirent->lde_reclen));
1086 end_dirent->lde_reclen =
1087 cpu_to_le16((char *)(dp->ldp_entries) -
1088 (char *)end_dirent);
1091 first->ldp_hash_end = hash_end;
1092 first->ldp_flags &= ~cpu_to_le32(LDF_COLLIDE);
1093 first->ldp_flags |= flags & cpu_to_le32(LDF_COLLIDE);
1097 LASSERTF(lu_pgs == 0, "left = %d", lu_pgs);
1100 #define mdc_adjust_dirpages(pages, cfs_pgs, lu_pgs) do {} while (0)
1101 #endif /* PAGE_SIZE > LU_PAGE_SIZE */
1103 /* parameters for readdir page */
1104 struct readpage_param {
1105 struct md_op_data *rp_mod;
1108 struct obd_export *rp_exp;
1109 struct md_callback *rp_cb;
1113 * Read pages from server.
1115 * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
1116 * a header lu_dirpage which describes the start/end hash, and whether this
1117 * page is empty (contains no dir entry) or hash collide with next page.
1118 * After client receives reply, several pages will be integrated into dir page
1119 * in PAGE_SIZE (if PAGE_SIZE greater than LU_PAGE_SIZE), and the
1120 * lu_dirpage for this integrated page will be adjusted.
1122 static int mdc_read_page_remote(void *data, struct page *page0)
1124 struct readpage_param *rp = data;
1125 struct page **page_pool;
1127 struct lu_dirpage *dp;
1128 int rd_pgs = 0; /* number of pages read actually */
1130 struct md_op_data *op_data = rp->rp_mod;
1131 struct ptlrpc_request *req;
1132 int max_pages = op_data->op_max_pages;
1133 struct inode *inode;
1138 LASSERT(max_pages > 0 && max_pages <= PTLRPC_MAX_BRW_PAGES);
1139 inode = op_data->op_data;
1140 fid = &op_data->op_fid1;
1143 page_pool = kcalloc(max_pages, sizeof(page), GFP_NOFS);
1145 page_pool[0] = page0;
1151 for (npages = 1; npages < max_pages; npages++) {
1152 page = page_cache_alloc_cold(inode->i_mapping);
1155 page_pool[npages] = page;
1158 rc = mdc_getpage(rp->rp_exp, fid, rp->rp_off, page_pool, npages, &req);
1160 int lu_pgs = req->rq_bulk->bd_nob_transferred;
1162 rd_pgs = (req->rq_bulk->bd_nob_transferred +
1163 PAGE_SIZE - 1) >> PAGE_SHIFT;
1164 lu_pgs >>= LU_PAGE_SHIFT;
1165 LASSERT(!(req->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK));
1167 CDEBUG(D_INODE, "read %d(%d) pages\n", rd_pgs, lu_pgs);
1169 mdc_adjust_dirpages(page_pool, rd_pgs, lu_pgs);
1171 SetPageUptodate(page0);
1175 ptlrpc_req_finished(req);
1176 CDEBUG(D_CACHE, "read %d/%d pages\n", rd_pgs, npages);
1177 for (i = 1; i < npages; i++) {
1178 unsigned long offset;
1182 page = page_pool[i];
1184 if (rc < 0 || i >= rd_pgs) {
1189 SetPageUptodate(page);
1192 hash = le64_to_cpu(dp->ldp_hash_start);
1195 offset = hash_x_index(hash, rp->rp_hash64);
1197 prefetchw(&page->flags);
1198 ret = add_to_page_cache_lru(page, inode->i_mapping, offset,
1203 CDEBUG(D_VFSTRACE, "page %lu add to page cache failed: rc = %d\n",
1208 if (page_pool != &page0)
1215 * Read dir page from cache first, if it can not find it, read it from
1216 * server and add into the cache.
1218 * \param[in] exp MDC export
1219 * \param[in] op_data client MD stack parameters, transferring parameters
1220 * between different layers on client MD stack.
1221 * \param[in] cb_op callback required for ldlm lock enqueue during
1223 * \param[in] hash_offset the hash offset of the page to be read
1224 * \param[in] ppage the page to be read
1226 * retval = 0 get the page successfully
1227 * errno(<0) get the page failed
1229 static int mdc_read_page(struct obd_export *exp, struct md_op_data *op_data,
1230 struct md_callback *cb_op, __u64 hash_offset,
1231 struct page **ppage)
1233 struct lookup_intent it = { .it_op = IT_READDIR };
1235 struct inode *dir = op_data->op_data;
1236 struct address_space *mapping;
1237 struct lu_dirpage *dp;
1240 struct lustre_handle lockh;
1241 struct ptlrpc_request *enq_req = NULL;
1242 struct readpage_param rp_param;
1248 mapping = dir->i_mapping;
1250 rc = mdc_intent_lock(exp, op_data, &it, &enq_req,
1251 cb_op->md_blocking_ast, 0);
1253 ptlrpc_req_finished(enq_req);
1256 CERROR("%s: "DFID" lock enqueue fails: rc = %d\n",
1257 exp->exp_obd->obd_name, PFID(&op_data->op_fid1), rc);
1262 lockh.cookie = it.it_lock_handle;
1263 mdc_set_lock_data(exp, &lockh, dir, NULL);
1265 rp_param.rp_off = hash_offset;
1266 rp_param.rp_hash64 = op_data->op_cli_flags & CLI_HASH64;
1267 page = mdc_page_locate(mapping, &rp_param.rp_off, &start, &end,
1268 rp_param.rp_hash64);
1270 CDEBUG(D_INFO, "%s: dir page locate: " DFID " at %llu: rc %ld\n",
1271 exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1272 rp_param.rp_off, PTR_ERR(page));
1277 * XXX nikita: not entirely correct handling of a corner case:
1278 * suppose hash chain of entries with hash value HASH crosses
1279 * border between pages P0 and P1. First both P0 and P1 are
1280 * cached, seekdir() is called for some entry from the P0 part
1281 * of the chain. Later P0 goes out of cache. telldir(HASH)
1282 * happens and finds P1, as it starts with matching hash
1283 * value. Remaining entries from P0 part of the chain are
1284 * skipped. (Is that really a bug?)
1286 * Possible solutions: 0. don't cache P1 is such case, handle
1287 * it as an "overflow" page. 1. invalidate all pages at
1288 * once. 2. use HASH|1 as an index for P1.
1290 goto hash_collision;
1293 rp_param.rp_exp = exp;
1294 rp_param.rp_mod = op_data;
1295 page = read_cache_page(mapping,
1296 hash_x_index(rp_param.rp_off,
1297 rp_param.rp_hash64),
1298 mdc_read_page_remote, &rp_param);
1300 CERROR("%s: read cache page: "DFID" at %llu: rc %ld\n",
1301 exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1302 rp_param.rp_off, PTR_ERR(page));
1307 wait_on_page_locked(page);
1309 if (!PageUptodate(page)) {
1310 CERROR("%s: page not updated: "DFID" at %llu: rc %d\n",
1311 exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1312 rp_param.rp_off, -5);
1315 if (!PageChecked(page))
1316 SetPageChecked(page);
1317 if (PageError(page)) {
1318 CERROR("%s: page error: "DFID" at %llu: rc %d\n",
1319 exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1320 rp_param.rp_off, -5);
1325 dp = page_address(page);
1326 if (BITS_PER_LONG == 32 && rp_param.rp_hash64) {
1327 start = le64_to_cpu(dp->ldp_hash_start) >> 32;
1328 end = le64_to_cpu(dp->ldp_hash_end) >> 32;
1329 rp_param.rp_off = hash_offset >> 32;
1331 start = le64_to_cpu(dp->ldp_hash_start);
1332 end = le64_to_cpu(dp->ldp_hash_end);
1333 rp_param.rp_off = hash_offset;
1336 LASSERT(start == rp_param.rp_off);
1337 CWARN("Page-wide hash collision: %#lx\n", (unsigned long)end);
1338 #if BITS_PER_LONG == 32
1339 CWARN("Real page-wide hash collision at [%llu %llu] with hash %llu\n",
1340 le64_to_cpu(dp->ldp_hash_start),
1341 le64_to_cpu(dp->ldp_hash_end), hash_offset);
1344 * Fetch whole overflow chain...
1352 ldlm_lock_decref(&lockh, it.it_lock_mode);
1356 mdc_release_page(page, 1);
1361 static int mdc_statfs(const struct lu_env *env,
1362 struct obd_export *exp, struct obd_statfs *osfs,
1363 __u64 max_age, __u32 flags)
1365 struct obd_device *obd = class_exp2obd(exp);
1366 struct ptlrpc_request *req;
1367 struct obd_statfs *msfs;
1368 struct obd_import *imp = NULL;
1372 * Since the request might also come from lprocfs, so we need
1373 * sync this with client_disconnect_export Bug15684
1375 down_read(&obd->u.cli.cl_sem);
1376 if (obd->u.cli.cl_import)
1377 imp = class_import_get(obd->u.cli.cl_import);
1378 up_read(&obd->u.cli.cl_sem);
1382 req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_STATFS,
1383 LUSTRE_MDS_VERSION, MDS_STATFS);
1389 ptlrpc_request_set_replen(req);
1391 if (flags & OBD_STATFS_NODELAY) {
1392 /* procfs requests not want stay in wait for avoid deadlock */
1393 req->rq_no_resend = 1;
1394 req->rq_no_delay = 1;
1397 rc = ptlrpc_queue_wait(req);
1399 /* check connection error first */
1400 if (imp->imp_connect_error)
1401 rc = imp->imp_connect_error;
1405 msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
1413 ptlrpc_req_finished(req);
1415 class_import_put(imp);
1419 static int mdc_ioc_fid2path(struct obd_export *exp, struct getinfo_fid2path *gf)
1421 __u32 keylen, vallen;
1425 if (gf->gf_pathlen > PATH_MAX)
1426 return -ENAMETOOLONG;
1427 if (gf->gf_pathlen < 2)
1430 /* Key is KEY_FID2PATH + getinfo_fid2path description */
1431 keylen = cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*gf);
1432 key = kzalloc(keylen, GFP_NOFS);
1435 memcpy(key, KEY_FID2PATH, sizeof(KEY_FID2PATH));
1436 memcpy(key + cfs_size_round(sizeof(KEY_FID2PATH)), gf, sizeof(*gf));
1438 CDEBUG(D_IOCTL, "path get "DFID" from %llu #%d\n",
1439 PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno);
1441 if (!fid_is_sane(&gf->gf_fid)) {
1446 /* Val is struct getinfo_fid2path result plus path */
1447 vallen = sizeof(*gf) + gf->gf_pathlen;
1449 rc = obd_get_info(NULL, exp, keylen, key, &vallen, gf);
1450 if (rc != 0 && rc != -EREMOTE)
1453 if (vallen <= sizeof(*gf)) {
1456 } else if (vallen > sizeof(*gf) + gf->gf_pathlen) {
1461 CDEBUG(D_IOCTL, "path got " DFID " from %llu #%d: %s\n",
1462 PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno,
1463 gf->gf_pathlen < 512 ? gf->gf_path :
1464 /* only log the last 512 characters of the path */
1465 gf->gf_path + gf->gf_pathlen - 512);
1472 static int mdc_ioc_hsm_progress(struct obd_export *exp,
1473 struct hsm_progress_kernel *hpk)
1475 struct obd_import *imp = class_exp2cliimp(exp);
1476 struct hsm_progress_kernel *req_hpk;
1477 struct ptlrpc_request *req;
1480 req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_PROGRESS,
1481 LUSTRE_MDS_VERSION, MDS_HSM_PROGRESS);
1487 mdc_pack_body(req, NULL, 0, 0, -1, 0);
1489 /* Copy hsm_progress struct */
1490 req_hpk = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_PROGRESS);
1497 req_hpk->hpk_errval = lustre_errno_hton(hpk->hpk_errval);
1499 ptlrpc_request_set_replen(req);
1501 mdc_get_mod_rpc_slot(req, NULL);
1502 rc = ptlrpc_queue_wait(req);
1503 mdc_put_mod_rpc_slot(req, NULL);
1505 ptlrpc_req_finished(req);
1509 static int mdc_ioc_hsm_ct_register(struct obd_import *imp, __u32 archives)
1511 __u32 *archive_mask;
1512 struct ptlrpc_request *req;
1515 req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_REGISTER,
1517 MDS_HSM_CT_REGISTER);
1523 mdc_pack_body(req, NULL, 0, 0, -1, 0);
1525 /* Copy hsm_progress struct */
1526 archive_mask = req_capsule_client_get(&req->rq_pill,
1527 &RMF_MDS_HSM_ARCHIVE);
1528 if (!archive_mask) {
1533 *archive_mask = archives;
1535 ptlrpc_request_set_replen(req);
1537 rc = mdc_queue_wait(req);
1539 ptlrpc_req_finished(req);
1543 static int mdc_ioc_hsm_current_action(struct obd_export *exp,
1544 struct md_op_data *op_data)
1546 struct hsm_current_action *hca = op_data->op_data;
1547 struct hsm_current_action *req_hca;
1548 struct ptlrpc_request *req;
1551 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1552 &RQF_MDS_HSM_ACTION);
1556 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_ACTION);
1558 ptlrpc_request_free(req);
1562 mdc_pack_body(req, &op_data->op_fid1, 0, 0,
1563 op_data->op_suppgids[0], 0);
1565 ptlrpc_request_set_replen(req);
1567 rc = mdc_queue_wait(req);
1571 req_hca = req_capsule_server_get(&req->rq_pill,
1572 &RMF_MDS_HSM_CURRENT_ACTION);
1581 ptlrpc_req_finished(req);
1585 static int mdc_ioc_hsm_ct_unregister(struct obd_import *imp)
1587 struct ptlrpc_request *req;
1590 req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_UNREGISTER,
1592 MDS_HSM_CT_UNREGISTER);
1598 mdc_pack_body(req, NULL, 0, 0, -1, 0);
1600 ptlrpc_request_set_replen(req);
1602 rc = mdc_queue_wait(req);
1604 ptlrpc_req_finished(req);
1608 static int mdc_ioc_hsm_state_get(struct obd_export *exp,
1609 struct md_op_data *op_data)
1611 struct hsm_user_state *hus = op_data->op_data;
1612 struct hsm_user_state *req_hus;
1613 struct ptlrpc_request *req;
1616 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1617 &RQF_MDS_HSM_STATE_GET);
1621 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_GET);
1623 ptlrpc_request_free(req);
1627 mdc_pack_body(req, &op_data->op_fid1, 0, 0,
1628 op_data->op_suppgids[0], 0);
1630 ptlrpc_request_set_replen(req);
1632 rc = mdc_queue_wait(req);
1636 req_hus = req_capsule_server_get(&req->rq_pill, &RMF_HSM_USER_STATE);
1645 ptlrpc_req_finished(req);
1649 static int mdc_ioc_hsm_state_set(struct obd_export *exp,
1650 struct md_op_data *op_data)
1652 struct hsm_state_set *hss = op_data->op_data;
1653 struct hsm_state_set *req_hss;
1654 struct ptlrpc_request *req;
1657 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1658 &RQF_MDS_HSM_STATE_SET);
1662 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_SET);
1664 ptlrpc_request_free(req);
1668 mdc_pack_body(req, &op_data->op_fid1, 0, 0,
1669 op_data->op_suppgids[0], 0);
1672 req_hss = req_capsule_client_get(&req->rq_pill, &RMF_HSM_STATE_SET);
1679 ptlrpc_request_set_replen(req);
1681 mdc_get_mod_rpc_slot(req, NULL);
1682 rc = ptlrpc_queue_wait(req);
1683 mdc_put_mod_rpc_slot(req, NULL);
1685 ptlrpc_req_finished(req);
1689 static int mdc_ioc_hsm_request(struct obd_export *exp,
1690 struct hsm_user_request *hur)
1692 struct obd_import *imp = class_exp2cliimp(exp);
1693 struct ptlrpc_request *req;
1694 struct hsm_request *req_hr;
1695 struct hsm_user_item *req_hui;
1699 req = ptlrpc_request_alloc(imp, &RQF_MDS_HSM_REQUEST);
1705 req_capsule_set_size(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM, RCL_CLIENT,
1706 hur->hur_request.hr_itemcount
1707 * sizeof(struct hsm_user_item));
1708 req_capsule_set_size(&req->rq_pill, &RMF_GENERIC_DATA, RCL_CLIENT,
1709 hur->hur_request.hr_data_len);
1711 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_REQUEST);
1713 ptlrpc_request_free(req);
1717 mdc_pack_body(req, NULL, 0, 0, -1, 0);
1719 /* Copy hsm_request struct */
1720 req_hr = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_REQUEST);
1725 *req_hr = hur->hur_request;
1727 /* Copy hsm_user_item structs */
1728 req_hui = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM);
1733 memcpy(req_hui, hur->hur_user_item,
1734 hur->hur_request.hr_itemcount * sizeof(struct hsm_user_item));
1736 /* Copy opaque field */
1737 req_opaque = req_capsule_client_get(&req->rq_pill, &RMF_GENERIC_DATA);
1742 memcpy(req_opaque, hur_data(hur), hur->hur_request.hr_data_len);
1744 ptlrpc_request_set_replen(req);
1746 mdc_get_mod_rpc_slot(req, NULL);
1747 rc = ptlrpc_queue_wait(req);
1748 mdc_put_mod_rpc_slot(req, NULL);
1750 ptlrpc_req_finished(req);
1754 static struct kuc_hdr *changelog_kuc_hdr(char *buf, size_t len, u32 flags)
1756 struct kuc_hdr *lh = (struct kuc_hdr *)buf;
1758 LASSERT(len <= KUC_CHANGELOG_MSG_MAXSIZE);
1760 lh->kuc_magic = KUC_MAGIC;
1761 lh->kuc_transport = KUC_TRANSPORT_CHANGELOG;
1762 lh->kuc_flags = flags;
1763 lh->kuc_msgtype = CL_RECORD;
1764 lh->kuc_msglen = len;
1768 struct changelog_show {
1770 enum changelog_send_flag cs_flags;
1773 struct obd_device *cs_obd;
1776 static inline char *cs_obd_name(struct changelog_show *cs)
1778 return cs->cs_obd->obd_name;
1781 static int changelog_kkuc_cb(const struct lu_env *env, struct llog_handle *llh,
1782 struct llog_rec_hdr *hdr, void *data)
1784 struct changelog_show *cs = data;
1785 struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
1790 if (rec->cr_hdr.lrh_type != CHANGELOG_REC) {
1792 CERROR("%s: not a changelog rec %x/%d: rc = %d\n",
1793 cs_obd_name(cs), rec->cr_hdr.lrh_type,
1794 rec->cr.cr_type, rc);
1798 if (rec->cr.cr_index < cs->cs_startrec) {
1799 /* Skip entries earlier than what we are interested in */
1800 CDEBUG(D_HSM, "rec=%llu start=%llu\n",
1801 rec->cr.cr_index, cs->cs_startrec);
1805 CDEBUG(D_HSM, "%llu %02d%-5s %llu 0x%x t=" DFID " p=" DFID
1806 " %.*s\n", rec->cr.cr_index, rec->cr.cr_type,
1807 changelog_type2str(rec->cr.cr_type), rec->cr.cr_time,
1808 rec->cr.cr_flags & CLF_FLAGMASK,
1809 PFID(&rec->cr.cr_tfid), PFID(&rec->cr.cr_pfid),
1810 rec->cr.cr_namelen, changelog_rec_name(&rec->cr));
1812 len = sizeof(*lh) + changelog_rec_size(&rec->cr) + rec->cr.cr_namelen;
1814 /* Set up the message */
1815 lh = changelog_kuc_hdr(cs->cs_buf, len, cs->cs_flags);
1816 memcpy(lh + 1, &rec->cr, len - sizeof(*lh));
1818 rc = libcfs_kkuc_msg_put(cs->cs_fp, lh);
1819 CDEBUG(D_HSM, "kucmsg fp %p len %zu rc %d\n", cs->cs_fp, len, rc);
1824 static int mdc_changelog_send_thread(void *csdata)
1826 enum llog_flag flags = LLOG_F_IS_CAT;
1827 struct changelog_show *cs = csdata;
1828 struct llog_ctxt *ctxt = NULL;
1829 struct llog_handle *llh = NULL;
1830 struct kuc_hdr *kuch;
1833 CDEBUG(D_HSM, "changelog to fp=%p start %llu\n",
1834 cs->cs_fp, cs->cs_startrec);
1836 cs->cs_buf = kzalloc(KUC_CHANGELOG_MSG_MAXSIZE, GFP_NOFS);
1842 /* Set up the remote catalog handle */
1843 ctxt = llog_get_context(cs->cs_obd, LLOG_CHANGELOG_REPL_CTXT);
1848 rc = llog_open(NULL, ctxt, &llh, NULL, CHANGELOG_CATALOG,
1851 CERROR("%s: fail to open changelog catalog: rc = %d\n",
1852 cs_obd_name(cs), rc);
1856 if (cs->cs_flags & CHANGELOG_FLAG_JOBID)
1857 flags |= LLOG_F_EXT_JOBID;
1859 rc = llog_init_handle(NULL, llh, flags, NULL);
1861 CERROR("llog_init_handle failed %d\n", rc);
1865 rc = llog_cat_process(NULL, llh, changelog_kkuc_cb, cs, 0, 0);
1867 /* Send EOF no matter what our result */
1868 kuch = changelog_kuc_hdr(cs->cs_buf, sizeof(*kuch), cs->cs_flags);
1869 kuch->kuc_msgtype = CL_EOF;
1870 libcfs_kkuc_msg_put(cs->cs_fp, kuch);
1875 llog_cat_close(NULL, llh);
1877 llog_ctxt_put(ctxt);
1883 static int mdc_ioc_changelog_send(struct obd_device *obd,
1884 struct ioc_changelog *icc)
1886 struct changelog_show *cs;
1887 struct task_struct *task;
1890 /* Freed in mdc_changelog_send_thread */
1891 cs = kzalloc(sizeof(*cs), GFP_NOFS);
1896 cs->cs_startrec = icc->icc_recno;
1897 /* matching fput in mdc_changelog_send_thread */
1898 cs->cs_fp = fget(icc->icc_id);
1899 cs->cs_flags = icc->icc_flags;
1902 * New thread because we should return to user app before
1903 * writing into our pipe
1905 task = kthread_run(mdc_changelog_send_thread, cs,
1906 "mdc_clg_send_thread");
1909 CERROR("%s: can't start changelog thread: rc = %d\n",
1910 cs_obd_name(cs), rc);
1914 CDEBUG(D_HSM, "%s: started changelog thread\n",
1918 CERROR("Failed to start changelog thread: %d\n", rc);
1922 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
1923 struct lustre_kernelcomm *lk);
1925 static int mdc_quotactl(struct obd_device *unused, struct obd_export *exp,
1926 struct obd_quotactl *oqctl)
1928 struct ptlrpc_request *req;
1929 struct obd_quotactl *oqc;
1932 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
1933 &RQF_MDS_QUOTACTL, LUSTRE_MDS_VERSION,
1938 oqc = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1941 ptlrpc_request_set_replen(req);
1942 ptlrpc_at_set_req_timeout(req);
1943 req->rq_no_resend = 1;
1945 rc = ptlrpc_queue_wait(req);
1947 CERROR("ptlrpc_queue_wait failed, rc: %d\n", rc);
1949 if (req->rq_repmsg) {
1950 oqc = req_capsule_server_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1954 CERROR("Can't unpack obd_quotactl\n");
1958 CERROR("Can't unpack obd_quotactl\n");
1961 ptlrpc_req_finished(req);
1966 static int mdc_ioc_swap_layouts(struct obd_export *exp,
1967 struct md_op_data *op_data)
1970 struct ptlrpc_request *req;
1972 struct mdc_swap_layouts *msl, *payload;
1974 msl = op_data->op_data;
1976 /* When the MDT will get the MDS_SWAP_LAYOUTS RPC the
1977 * first thing it will do is to cancel the 2 layout
1978 * locks hold by this client.
1979 * So the client must cancel its layout locks on the 2 fids
1980 * with the request RPC to avoid extra RPC round trips
1982 count = mdc_resource_get_unused(exp, &op_data->op_fid1, &cancels,
1983 LCK_CR, MDS_INODELOCK_LAYOUT |
1984 MDS_INODELOCK_XATTR);
1985 count += mdc_resource_get_unused(exp, &op_data->op_fid2, &cancels,
1986 LCK_CR, MDS_INODELOCK_LAYOUT |
1987 MDS_INODELOCK_XATTR);
1989 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1990 &RQF_MDS_SWAP_LAYOUTS);
1992 ldlm_lock_list_put(&cancels, l_bl_ast, count);
1996 rc = mdc_prep_elc_req(exp, req, MDS_SWAP_LAYOUTS, &cancels, count);
1998 ptlrpc_request_free(req);
2002 mdc_swap_layouts_pack(req, op_data);
2004 payload = req_capsule_client_get(&req->rq_pill, &RMF_SWAP_LAYOUTS);
2009 ptlrpc_request_set_replen(req);
2011 rc = ptlrpc_queue_wait(req);
2013 ptlrpc_req_finished(req);
2017 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2018 void *karg, void __user *uarg)
2020 struct obd_device *obd = exp->exp_obd;
2021 struct obd_ioctl_data *data = karg;
2022 struct obd_import *imp = obd->u.cli.cl_import;
2025 if (!try_module_get(THIS_MODULE)) {
2026 CERROR("%s: cannot get module '%s'\n", obd->obd_name,
2027 module_name(THIS_MODULE));
2031 case OBD_IOC_CHANGELOG_SEND:
2032 rc = mdc_ioc_changelog_send(obd, karg);
2034 case OBD_IOC_CHANGELOG_CLEAR: {
2035 struct ioc_changelog *icc = karg;
2036 struct changelog_setinfo cs = {
2037 .cs_recno = icc->icc_recno,
2038 .cs_id = icc->icc_id
2041 rc = obd_set_info_async(NULL, exp, strlen(KEY_CHANGELOG_CLEAR),
2042 KEY_CHANGELOG_CLEAR, sizeof(cs), &cs,
2046 case OBD_IOC_FID2PATH:
2047 rc = mdc_ioc_fid2path(exp, karg);
2049 case LL_IOC_HSM_CT_START:
2050 rc = mdc_ioc_hsm_ct_start(exp, karg);
2051 /* ignore if it was already registered on this MDS. */
2055 case LL_IOC_HSM_PROGRESS:
2056 rc = mdc_ioc_hsm_progress(exp, karg);
2058 case LL_IOC_HSM_STATE_GET:
2059 rc = mdc_ioc_hsm_state_get(exp, karg);
2061 case LL_IOC_HSM_STATE_SET:
2062 rc = mdc_ioc_hsm_state_set(exp, karg);
2064 case LL_IOC_HSM_ACTION:
2065 rc = mdc_ioc_hsm_current_action(exp, karg);
2067 case LL_IOC_HSM_REQUEST:
2068 rc = mdc_ioc_hsm_request(exp, karg);
2070 case OBD_IOC_CLIENT_RECOVER:
2071 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1, 0);
2076 case IOC_OSC_SET_ACTIVE:
2077 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
2079 case OBD_IOC_PING_TARGET:
2080 rc = ptlrpc_obd_ping(obd);
2083 * Normally IOC_OBD_STATFS, OBD_IOC_QUOTACTL iocontrol are handled by
2084 * LMV instead of MDC. But when the cluster is upgraded from 1.8,
2085 * there'd be no LMV layer thus we might be called here. Eventually
2086 * this code should be removed.
2089 case IOC_OBD_STATFS: {
2090 struct obd_statfs stat_buf = {0};
2092 if (*((__u32 *)data->ioc_inlbuf2) != 0) {
2098 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(obd),
2099 min_t(size_t, data->ioc_plen2,
2100 sizeof(struct obd_uuid)))) {
2105 rc = mdc_statfs(NULL, obd->obd_self_export, &stat_buf,
2106 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
2111 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
2112 min_t(size_t, data->ioc_plen1,
2113 sizeof(stat_buf)))) {
2121 case OBD_IOC_QUOTACTL: {
2122 struct if_quotactl *qctl = karg;
2123 struct obd_quotactl *oqctl;
2125 oqctl = kzalloc(sizeof(*oqctl), GFP_NOFS);
2131 QCTL_COPY(oqctl, qctl);
2132 rc = obd_quotactl(exp, oqctl);
2134 QCTL_COPY(qctl, oqctl);
2135 qctl->qc_valid = QC_MDTIDX;
2136 qctl->obd_uuid = obd->u.cli.cl_target_uuid;
2142 case LL_IOC_GET_CONNECT_FLAGS:
2143 if (copy_to_user(uarg, exp_connect_flags_ptr(exp),
2144 sizeof(*exp_connect_flags_ptr(exp)))) {
2151 case LL_IOC_LOV_SWAP_LAYOUTS:
2152 rc = mdc_ioc_swap_layouts(exp, karg);
2155 CERROR("unrecognised ioctl: cmd = %#x\n", cmd);
2160 module_put(THIS_MODULE);
2165 static int mdc_get_info_rpc(struct obd_export *exp,
2166 u32 keylen, void *key,
2167 int vallen, void *val)
2169 struct obd_import *imp = class_exp2cliimp(exp);
2170 struct ptlrpc_request *req;
2174 req = ptlrpc_request_alloc(imp, &RQF_MDS_GET_INFO);
2178 req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_KEY,
2179 RCL_CLIENT, keylen);
2180 req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VALLEN,
2181 RCL_CLIENT, sizeof(__u32));
2183 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GET_INFO);
2185 ptlrpc_request_free(req);
2189 tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_KEY);
2190 memcpy(tmp, key, keylen);
2191 tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_VALLEN);
2192 memcpy(tmp, &vallen, sizeof(__u32));
2194 req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VAL,
2195 RCL_SERVER, vallen);
2196 ptlrpc_request_set_replen(req);
2198 rc = ptlrpc_queue_wait(req);
2199 /* -EREMOTE means the get_info result is partial, and it needs to
2200 * continue on another MDT, see fid2path part in lmv_iocontrol
2202 if (rc == 0 || rc == -EREMOTE) {
2203 tmp = req_capsule_server_get(&req->rq_pill, &RMF_GETINFO_VAL);
2204 memcpy(val, tmp, vallen);
2205 if (ptlrpc_rep_need_swab(req)) {
2206 if (KEY_IS(KEY_FID2PATH))
2207 lustre_swab_fid2path(val);
2210 ptlrpc_req_finished(req);
2215 static void lustre_swab_hai(struct hsm_action_item *h)
2217 __swab32s(&h->hai_len);
2218 __swab32s(&h->hai_action);
2219 lustre_swab_lu_fid(&h->hai_fid);
2220 lustre_swab_lu_fid(&h->hai_dfid);
2221 __swab64s(&h->hai_cookie);
2222 __swab64s(&h->hai_extent.offset);
2223 __swab64s(&h->hai_extent.length);
2224 __swab64s(&h->hai_gid);
2227 static void lustre_swab_hal(struct hsm_action_list *h)
2229 struct hsm_action_item *hai;
2232 __swab32s(&h->hal_version);
2233 __swab32s(&h->hal_count);
2234 __swab32s(&h->hal_archive_id);
2235 __swab64s(&h->hal_flags);
2237 for (i = 0; i < h->hal_count; i++, hai = hai_next(hai))
2238 lustre_swab_hai(hai);
2241 static void lustre_swab_kuch(struct kuc_hdr *l)
2243 __swab16s(&l->kuc_magic);
2244 /* __u8 l->kuc_transport */
2245 __swab16s(&l->kuc_msgtype);
2246 __swab16s(&l->kuc_msglen);
2249 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
2250 struct lustre_kernelcomm *lk)
2252 struct obd_import *imp = class_exp2cliimp(exp);
2253 __u32 archive = lk->lk_data;
2256 if (lk->lk_group != KUC_GRP_HSM) {
2257 CERROR("Bad copytool group %d\n", lk->lk_group);
2261 CDEBUG(D_HSM, "CT start r%d w%d u%d g%d f%#x\n", lk->lk_rfd, lk->lk_wfd,
2262 lk->lk_uid, lk->lk_group, lk->lk_flags);
2264 if (lk->lk_flags & LK_FLG_STOP) {
2265 /* Unregister with the coordinator */
2266 rc = mdc_ioc_hsm_ct_unregister(imp);
2268 rc = mdc_ioc_hsm_ct_register(imp, archive);
2275 * Send a message to any listening copytools
2276 * @param val KUC message (kuc_hdr + hsm_action_list)
2277 * @param len total length of message
2279 static int mdc_hsm_copytool_send(size_t len, void *val)
2281 struct kuc_hdr *lh = (struct kuc_hdr *)val;
2282 struct hsm_action_list *hal = (struct hsm_action_list *)(lh + 1);
2284 if (len < sizeof(*lh) + sizeof(*hal)) {
2285 CERROR("Short HSM message %zu < %zu\n", len,
2286 sizeof(*lh) + sizeof(*hal));
2289 if (lh->kuc_magic == __swab16(KUC_MAGIC)) {
2290 lustre_swab_kuch(lh);
2291 lustre_swab_hal(hal);
2292 } else if (lh->kuc_magic != KUC_MAGIC) {
2293 CERROR("Bad magic %x!=%x\n", lh->kuc_magic, KUC_MAGIC);
2298 "Received message mg=%x t=%d m=%d l=%d actions=%d on %s\n",
2299 lh->kuc_magic, lh->kuc_transport, lh->kuc_msgtype,
2300 lh->kuc_msglen, hal->hal_count, hal->hal_fsname);
2302 /* Broadcast to HSM listeners */
2303 return libcfs_kkuc_group_put(KUC_GRP_HSM, lh);
2307 * callback function passed to kuc for re-registering each HSM copytool
2308 * running on MDC, after MDT shutdown/recovery.
2309 * @param data copytool registration data
2310 * @param cb_arg callback argument (obd_import)
2312 static int mdc_hsm_ct_reregister(void *data, void *cb_arg)
2314 struct kkuc_ct_data *kcd = data;
2315 struct obd_import *imp = (struct obd_import *)cb_arg;
2318 if (!kcd || kcd->kcd_magic != KKUC_CT_DATA_MAGIC)
2321 if (!obd_uuid_equals(&kcd->kcd_uuid, &imp->imp_obd->obd_uuid))
2324 CDEBUG(D_HA, "%s: recover copytool registration to MDT (archive=%#x)\n",
2325 imp->imp_obd->obd_name, kcd->kcd_archive);
2326 rc = mdc_ioc_hsm_ct_register(imp, kcd->kcd_archive);
2328 /* ignore error if the copytool is already registered */
2329 return (rc == -EEXIST) ? 0 : rc;
2332 static int mdc_set_info_async(const struct lu_env *env,
2333 struct obd_export *exp,
2334 u32 keylen, void *key,
2335 u32 vallen, void *val,
2336 struct ptlrpc_request_set *set)
2338 struct obd_import *imp = class_exp2cliimp(exp);
2341 if (KEY_IS(KEY_READ_ONLY)) {
2342 if (vallen != sizeof(int))
2345 spin_lock(&imp->imp_lock);
2346 if (*((int *)val)) {
2347 imp->imp_connect_flags_orig |= OBD_CONNECT_RDONLY;
2348 imp->imp_connect_data.ocd_connect_flags |=
2351 imp->imp_connect_flags_orig &= ~OBD_CONNECT_RDONLY;
2352 imp->imp_connect_data.ocd_connect_flags &=
2353 ~OBD_CONNECT_RDONLY;
2355 spin_unlock(&imp->imp_lock);
2357 return do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2358 keylen, key, vallen, val, set);
2360 if (KEY_IS(KEY_SPTLRPC_CONF)) {
2361 sptlrpc_conf_client_adapt(exp->exp_obd);
2364 if (KEY_IS(KEY_FLUSH_CTX)) {
2365 sptlrpc_import_flush_my_ctx(imp);
2368 if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
2369 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2370 keylen, key, vallen, val, set);
2373 if (KEY_IS(KEY_HSM_COPYTOOL_SEND)) {
2374 rc = mdc_hsm_copytool_send(vallen, val);
2377 if (KEY_IS(KEY_DEFAULT_EASIZE)) {
2378 u32 *default_easize = val;
2380 exp->exp_obd->u.cli.cl_default_mds_easize = *default_easize;
2384 CERROR("Unknown key %s\n", (char *)key);
2388 static int mdc_get_info(const struct lu_env *env, struct obd_export *exp,
2389 __u32 keylen, void *key, __u32 *vallen, void *val)
2393 if (KEY_IS(KEY_MAX_EASIZE)) {
2394 u32 mdsize, *max_easize;
2396 if (*vallen != sizeof(int))
2398 mdsize = *(u32 *)val;
2399 if (mdsize > exp->exp_obd->u.cli.cl_max_mds_easize)
2400 exp->exp_obd->u.cli.cl_max_mds_easize = mdsize;
2402 *max_easize = exp->exp_obd->u.cli.cl_max_mds_easize;
2404 } else if (KEY_IS(KEY_DEFAULT_EASIZE)) {
2405 u32 *default_easize;
2407 if (*vallen != sizeof(int))
2409 default_easize = val;
2410 *default_easize = exp->exp_obd->u.cli.cl_default_mds_easize;
2412 } else if (KEY_IS(KEY_CONN_DATA)) {
2413 struct obd_import *imp = class_exp2cliimp(exp);
2414 struct obd_connect_data *data = val;
2416 if (*vallen != sizeof(*data))
2419 *data = imp->imp_connect_data;
2421 } else if (KEY_IS(KEY_TGT_COUNT)) {
2426 rc = mdc_get_info_rpc(exp, keylen, key, *vallen, val);
2431 static int mdc_sync(struct obd_export *exp, const struct lu_fid *fid,
2432 struct ptlrpc_request **request)
2434 struct ptlrpc_request *req;
2438 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_SYNC);
2442 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_SYNC);
2444 ptlrpc_request_free(req);
2448 mdc_pack_body(req, fid, 0, 0, -1, 0);
2450 ptlrpc_request_set_replen(req);
2452 rc = ptlrpc_queue_wait(req);
2454 ptlrpc_req_finished(req);
2460 static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
2461 enum obd_import_event event)
2465 LASSERT(imp->imp_obd == obd);
2468 case IMP_EVENT_DISCON: {
2470 /* XXX Pass event up to OBDs stack. used only for FLD now */
2471 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_DISCON, NULL);
2475 case IMP_EVENT_INACTIVE: {
2476 struct client_obd *cli = &obd->u.cli;
2478 * Flush current sequence to make client obtain new one
2479 * from server in case of disconnect/reconnect.
2482 seq_client_flush(cli->cl_seq);
2484 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE, NULL);
2487 case IMP_EVENT_INVALIDATE: {
2488 struct ldlm_namespace *ns = obd->obd_namespace;
2490 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
2494 case IMP_EVENT_ACTIVE:
2495 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE, NULL);
2496 /* redo the kuc registration after reconnecting */
2498 /* re-register HSM agents */
2499 rc = libcfs_kkuc_group_foreach(KUC_GRP_HSM,
2500 mdc_hsm_ct_reregister,
2504 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_OCD, NULL);
2506 case IMP_EVENT_DEACTIVATE:
2507 case IMP_EVENT_ACTIVATE:
2510 CERROR("Unknown import event %x\n", event);
2516 int mdc_fid_alloc(const struct lu_env *env, struct obd_export *exp,
2517 struct lu_fid *fid, struct md_op_data *op_data)
2519 struct client_obd *cli = &exp->exp_obd->u.cli;
2520 struct lu_client_seq *seq = cli->cl_seq;
2522 return seq_client_alloc_fid(env, seq, fid);
2525 static struct obd_uuid *mdc_get_uuid(struct obd_export *exp)
2527 struct client_obd *cli = &exp->exp_obd->u.cli;
2529 return &cli->cl_target_uuid;
2533 * Determine whether the lock can be canceled before replaying it during
2534 * recovery, non zero value will be return if the lock can be canceled,
2535 * or zero returned for not
2537 static int mdc_cancel_weight(struct ldlm_lock *lock)
2539 if (lock->l_resource->lr_type != LDLM_IBITS)
2542 /* FIXME: if we ever get into a situation where there are too many
2543 * opened files with open locks on a single node, then we really
2544 * should replay these open locks to reget it
2546 if (lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_OPEN)
2552 static int mdc_resource_inode_free(struct ldlm_resource *res)
2554 if (res->lr_lvb_inode)
2555 res->lr_lvb_inode = NULL;
2560 static struct ldlm_valblock_ops inode_lvbo = {
2561 .lvbo_free = mdc_resource_inode_free,
2564 static int mdc_llog_init(struct obd_device *obd)
2566 struct obd_llog_group *olg = &obd->obd_olg;
2567 struct llog_ctxt *ctxt;
2570 rc = llog_setup(NULL, obd, olg, LLOG_CHANGELOG_REPL_CTXT, obd,
2575 ctxt = llog_group_get_ctxt(olg, LLOG_CHANGELOG_REPL_CTXT);
2576 llog_initiator_connect(ctxt);
2577 llog_ctxt_put(ctxt);
2582 static void mdc_llog_finish(struct obd_device *obd)
2584 struct llog_ctxt *ctxt;
2586 ctxt = llog_get_context(obd, LLOG_CHANGELOG_REPL_CTXT);
2588 llog_cleanup(NULL, ctxt);
2591 static int mdc_setup(struct obd_device *obd, struct lustre_cfg *cfg)
2593 struct lprocfs_static_vars lvars = { NULL };
2596 rc = ptlrpcd_addref();
2600 rc = client_obd_setup(obd, cfg);
2602 goto err_ptlrpcd_decref;
2604 lprocfs_mdc_init_vars(&lvars);
2605 lprocfs_obd_setup(obd, lvars.obd_vars, lvars.sysfs_vars);
2606 sptlrpc_lprocfs_cliobd_attach(obd);
2607 ptlrpc_lprocfs_register_obd(obd);
2609 ns_register_cancel(obd->obd_namespace, mdc_cancel_weight);
2611 obd->obd_namespace->ns_lvbo = &inode_lvbo;
2613 rc = mdc_llog_init(obd);
2616 CERROR("failed to setup llogging subsystems\n");
2627 /* Initialize the default and maximum LOV EA sizes. This allows
2628 * us to make MDS RPCs with large enough reply buffers to hold a default
2629 * sized EA without having to calculate this (via a call into the
2630 * LOV + OSCs) each time we make an RPC. The maximum size is also tracked
2631 * but not used to avoid wastefully vmalloc()'ing large reply buffers when
2632 * a large number of stripes is possible. If a larger reply buffer is
2633 * required it will be reallocated in the ptlrpc layer due to overflow.
2635 static int mdc_init_ea_size(struct obd_export *exp, u32 easize, u32 def_easize)
2637 struct obd_device *obd = exp->exp_obd;
2638 struct client_obd *cli = &obd->u.cli;
2640 if (cli->cl_max_mds_easize < easize)
2641 cli->cl_max_mds_easize = easize;
2643 if (cli->cl_default_mds_easize < def_easize)
2644 cli->cl_default_mds_easize = def_easize;
2649 static int mdc_precleanup(struct obd_device *obd)
2651 /* Failsafe, ok if racy */
2652 if (obd->obd_type->typ_refcnt <= 1)
2653 libcfs_kkuc_group_rem(0, KUC_GRP_HSM);
2655 obd_cleanup_client_import(obd);
2656 ptlrpc_lprocfs_unregister_obd(obd);
2657 lprocfs_obd_cleanup(obd);
2658 mdc_llog_finish(obd);
2662 static int mdc_cleanup(struct obd_device *obd)
2666 return client_obd_cleanup(obd);
2669 static int mdc_process_config(struct obd_device *obd, u32 len, void *buf)
2671 struct lustre_cfg *lcfg = buf;
2672 struct lprocfs_static_vars lvars = { NULL };
2675 lprocfs_mdc_init_vars(&lvars);
2676 switch (lcfg->lcfg_command) {
2678 rc = class_process_proc_param(PARAM_MDC, lvars.obd_vars,
2687 static struct obd_ops mdc_obd_ops = {
2688 .owner = THIS_MODULE,
2690 .precleanup = mdc_precleanup,
2691 .cleanup = mdc_cleanup,
2692 .add_conn = client_import_add_conn,
2693 .del_conn = client_import_del_conn,
2694 .connect = client_connect_import,
2695 .disconnect = client_disconnect_export,
2696 .iocontrol = mdc_iocontrol,
2697 .set_info_async = mdc_set_info_async,
2698 .statfs = mdc_statfs,
2699 .fid_init = client_fid_init,
2700 .fid_fini = client_fid_fini,
2701 .fid_alloc = mdc_fid_alloc,
2702 .import_event = mdc_import_event,
2703 .get_info = mdc_get_info,
2704 .process_config = mdc_process_config,
2705 .get_uuid = mdc_get_uuid,
2706 .quotactl = mdc_quotactl,
2709 static struct md_ops mdc_md_ops = {
2710 .getstatus = mdc_getstatus,
2711 .null_inode = mdc_null_inode,
2713 .create = mdc_create,
2714 .enqueue = mdc_enqueue,
2715 .getattr = mdc_getattr,
2716 .getattr_name = mdc_getattr_name,
2717 .intent_lock = mdc_intent_lock,
2719 .rename = mdc_rename,
2720 .setattr = mdc_setattr,
2721 .setxattr = mdc_setxattr,
2722 .getxattr = mdc_getxattr,
2724 .read_page = mdc_read_page,
2725 .unlink = mdc_unlink,
2726 .cancel_unused = mdc_cancel_unused,
2727 .init_ea_size = mdc_init_ea_size,
2728 .set_lock_data = mdc_set_lock_data,
2729 .lock_match = mdc_lock_match,
2730 .get_lustre_md = mdc_get_lustre_md,
2731 .free_lustre_md = mdc_free_lustre_md,
2732 .set_open_replay_data = mdc_set_open_replay_data,
2733 .clear_open_replay_data = mdc_clear_open_replay_data,
2734 .intent_getattr_async = mdc_intent_getattr_async,
2735 .revalidate_lock = mdc_revalidate_lock
2738 static int __init mdc_init(void)
2740 struct lprocfs_static_vars lvars = { NULL };
2742 lprocfs_mdc_init_vars(&lvars);
2744 return class_register_type(&mdc_obd_ops, &mdc_md_ops,
2745 LUSTRE_MDC_NAME, NULL);
2748 static void /*__exit*/ mdc_exit(void)
2750 class_unregister_type(LUSTRE_MDC_NAME);
2753 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
2754 MODULE_DESCRIPTION("Lustre Metadata Client");
2755 MODULE_VERSION(LUSTRE_VERSION_STRING);
2756 MODULE_LICENSE("GPL");
2758 module_init(mdc_init);
2759 module_exit(mdc_exit);