4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2001, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2015, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
37 #define DEBUG_SUBSYSTEM S_MDC
39 #include <linux/init.h>
40 #include <linux/kthread.h>
41 #include <linux/miscdevice.h>
42 #include <linux/module.h>
43 #include <linux/pagemap.h>
44 #include <linux/user_namespace.h>
45 #include <linux/utsname.h>
46 #ifdef HAVE_UIDGID_HEADER
47 # include <linux/uidgid.h>
50 #include <cl_object.h>
51 #include <llog_swab.h>
52 #include <lprocfs_status.h>
53 #include <lustre_acl.h>
54 #include <lustre_fid.h>
55 #include <lustre_ioctl.h>
56 #include <lustre_kernelcomm.h>
57 #include <lustre_lmv.h>
58 #include <lustre_log.h>
59 #include <lustre_param.h>
60 #include <lustre_swab.h>
61 #include <obd_class.h>
63 #include "mdc_internal.h"
65 #define REQUEST_MINOR 244
67 static int mdc_cleanup(struct obd_device *obd);
69 static inline int mdc_queue_wait(struct ptlrpc_request *req)
71 struct client_obd *cli = &req->rq_import->imp_obd->u.cli;
74 /* obd_get_request_slot() ensures that this client has no more
75 * than cl_max_rpcs_in_flight RPCs simultaneously inf light
77 rc = obd_get_request_slot(cli);
81 rc = ptlrpc_queue_wait(req);
82 obd_put_request_slot(cli);
88 * Send MDS_GET_ROOT RPC to fetch root FID.
90 * If \a fileset is not NULL it should contain a subdirectory off
91 * the ROOT/ directory to be mounted on the client. Return the FID
92 * of the subdirectory to the client to mount onto its mountpoint.
94 * \param[in] imp MDC import
95 * \param[in] fileset fileset name, which could be NULL
96 * \param[out] rootfid root FID of this mountpoint
97 * \param[out] pc root capa will be unpacked and saved in this pointer
99 * \retval 0 on success, negative errno on failure
101 static int mdc_get_root(struct obd_export *exp, const char *fileset,
102 struct lu_fid *rootfid)
104 struct ptlrpc_request *req;
105 struct mdt_body *body;
110 if (fileset && !(exp_connect_flags(exp) & OBD_CONNECT_SUBTREE))
113 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
119 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
120 strlen(fileset) + 1);
121 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GET_ROOT);
123 ptlrpc_request_free(req);
126 mdc_pack_body(req, NULL, 0, 0, -1, 0);
127 if (fileset != NULL) {
128 char *name = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
130 memcpy(name, fileset, strlen(fileset));
132 lustre_msg_add_flags(req->rq_reqmsg, LUSTRE_IMP_FULL);
133 req->rq_send_state = LUSTRE_IMP_FULL;
135 ptlrpc_request_set_replen(req);
137 rc = ptlrpc_queue_wait(req);
141 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
143 GOTO(out, rc = -EPROTO);
145 *rootfid = body->mbo_fid1;
146 CDEBUG(D_NET, "root fid="DFID", last_committed="LPU64"\n",
147 PFID(rootfid), lustre_msg_get_last_committed(req->rq_repmsg));
150 ptlrpc_req_finished(req);
156 * This function now is known to always saying that it will receive 4 buffers
157 * from server. Even for cases when acl_size and md_size is zero, RPC header
158 * will contain 4 fields and RPC itself will contain zero size fields. This is
159 * because mdt_getattr*() _always_ returns 4 fields, but if acl is not needed
160 * and thus zero, it shrinks it, making zero size. The same story about
161 * md_size. And this is course of problem when client waits for smaller number
162 * of fields. This issue will be fixed later when client gets aware of RPC
165 static int mdc_getattr_common(struct obd_export *exp,
166 struct ptlrpc_request *req)
168 struct req_capsule *pill = &req->rq_pill;
169 struct mdt_body *body;
174 /* Request message already built. */
175 rc = ptlrpc_queue_wait(req);
179 /* sanity check for the reply */
180 body = req_capsule_server_get(pill, &RMF_MDT_BODY);
184 CDEBUG(D_NET, "mode: %o\n", body->mbo_mode);
186 mdc_update_max_ea_from_body(exp, body);
187 if (body->mbo_eadatasize != 0) {
188 eadata = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
189 body->mbo_eadatasize);
194 if (body->mbo_valid & OBD_MD_FLRMTPERM) {
195 struct mdt_remote_perm *perm;
197 LASSERT(client_is_remote(exp));
198 perm = req_capsule_server_swab_get(pill, &RMF_ACL,
199 lustre_swab_mdt_remote_perm);
207 static int mdc_getattr(struct obd_export *exp, struct md_op_data *op_data,
208 struct ptlrpc_request **request)
210 struct ptlrpc_request *req;
214 /* Single MDS without an LMV case */
215 if (op_data->op_flags & MF_GET_MDT_IDX) {
220 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
224 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
226 ptlrpc_request_free(req);
230 mdc_pack_body(req, &op_data->op_fid1, op_data->op_valid,
231 op_data->op_mode, -1, 0);
233 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
235 if (op_data->op_valid & OBD_MD_FLRMTPERM) {
236 LASSERT(client_is_remote(exp));
237 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
238 sizeof(struct mdt_remote_perm));
240 ptlrpc_request_set_replen(req);
242 rc = mdc_getattr_common(exp, req);
244 ptlrpc_req_finished(req);
250 static int mdc_getattr_name(struct obd_export *exp, struct md_op_data *op_data,
251 struct ptlrpc_request **request)
253 struct ptlrpc_request *req;
258 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
259 &RQF_MDS_GETATTR_NAME);
263 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
264 op_data->op_namelen + 1);
266 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR_NAME);
268 ptlrpc_request_free(req);
272 mdc_pack_body(req, &op_data->op_fid1, op_data->op_valid,
273 op_data->op_mode, op_data->op_suppgids[0], 0);
275 if (op_data->op_name) {
276 char *name = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
277 LASSERT(strnlen(op_data->op_name, op_data->op_namelen) ==
278 op_data->op_namelen);
279 memcpy(name, op_data->op_name, op_data->op_namelen);
282 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
284 ptlrpc_request_set_replen(req);
286 rc = mdc_getattr_common(exp, req);
288 ptlrpc_req_finished(req);
294 static int mdc_xattr_common(struct obd_export *exp,const struct req_format *fmt,
295 const struct lu_fid *fid, int opcode, u64 valid,
296 const char *xattr_name, const char *input,
297 int input_size, int output_size, int flags,
298 __u32 suppgid, struct ptlrpc_request **request)
300 struct ptlrpc_request *req;
301 int xattr_namelen = 0;
307 req = ptlrpc_request_alloc(class_exp2cliimp(exp), fmt);
312 xattr_namelen = strlen(xattr_name) + 1;
313 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
318 req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT,
322 /* Flush local XATTR locks to get rid of a possible cancel RPC */
323 if (opcode == MDS_REINT && fid_is_sane(fid) &&
324 exp->exp_connect_data.ocd_ibits_known & MDS_INODELOCK_XATTR) {
325 struct list_head cancels = LIST_HEAD_INIT(cancels);
328 /* Without that packing would fail */
330 req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
333 count = mdc_resource_get_unused(exp, fid,
335 MDS_INODELOCK_XATTR);
337 rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
339 ptlrpc_request_free(req);
343 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, opcode);
345 ptlrpc_request_free(req);
350 if (opcode == MDS_REINT) {
351 struct mdt_rec_setxattr *rec;
353 CLASSERT(sizeof(struct mdt_rec_setxattr) ==
354 sizeof(struct mdt_rec_reint));
355 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
356 rec->sx_opcode = REINT_SETXATTR;
357 rec->sx_fsuid = from_kuid(&init_user_ns, current_fsuid());
358 rec->sx_fsgid = from_kgid(&init_user_ns, current_fsgid());
359 rec->sx_cap = cfs_curproc_cap_pack();
360 rec->sx_suppgid1 = suppgid;
361 rec->sx_suppgid2 = -1;
363 rec->sx_valid = valid | OBD_MD_FLCTIME;
364 rec->sx_time = cfs_time_current_sec();
365 rec->sx_size = output_size;
366 rec->sx_flags = flags;
368 mdc_pack_body(req, fid, valid, output_size, suppgid, flags);
372 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
373 memcpy(tmp, xattr_name, xattr_namelen);
376 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
377 memcpy(tmp, input, input_size);
380 if (req_capsule_has_field(&req->rq_pill, &RMF_EADATA, RCL_SERVER))
381 req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
382 RCL_SERVER, output_size);
383 ptlrpc_request_set_replen(req);
386 if (opcode == MDS_REINT)
387 mdc_get_mod_rpc_slot(req, NULL);
389 rc = ptlrpc_queue_wait(req);
391 if (opcode == MDS_REINT)
392 mdc_put_mod_rpc_slot(req, NULL);
395 ptlrpc_req_finished(req);
401 static int mdc_setxattr(struct obd_export *exp, const struct lu_fid *fid,
402 u64 valid, const char *xattr_name,
403 const char *input, int input_size, int output_size,
404 int flags, __u32 suppgid,
405 struct ptlrpc_request **request)
407 return mdc_xattr_common(exp, &RQF_MDS_REINT_SETXATTR,
408 fid, MDS_REINT, valid, xattr_name,
409 input, input_size, output_size, flags,
413 static int mdc_getxattr(struct obd_export *exp, const struct lu_fid *fid,
414 u64 valid, const char *xattr_name,
415 const char *input, int input_size, int output_size,
416 int flags, struct ptlrpc_request **request)
418 return mdc_xattr_common(exp, &RQF_MDS_GETXATTR,
419 fid, MDS_GETXATTR, valid, xattr_name,
420 input, input_size, output_size, flags,
424 #ifdef CONFIG_FS_POSIX_ACL
425 static int mdc_unpack_acl(struct ptlrpc_request *req, struct lustre_md *md)
427 struct req_capsule *pill = &req->rq_pill;
428 struct mdt_body *body = md->body;
429 struct posix_acl *acl;
434 if (!body->mbo_aclsize)
437 buf = req_capsule_server_sized_get(pill, &RMF_ACL, body->mbo_aclsize);
442 acl = posix_acl_from_xattr(&init_user_ns, buf, body->mbo_aclsize);
447 CERROR("convert xattr to acl: %d\n", rc);
451 rc = posix_acl_valid(acl);
453 CERROR("validate acl: %d\n", rc);
454 posix_acl_release(acl);
462 #define mdc_unpack_acl(req, md) 0
465 int mdc_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
466 struct obd_export *dt_exp, struct obd_export *md_exp,
467 struct lustre_md *md)
469 struct req_capsule *pill = &req->rq_pill;
474 memset(md, 0, sizeof(*md));
476 md->body = req_capsule_server_get(pill, &RMF_MDT_BODY);
477 LASSERT(md->body != NULL);
479 if (md->body->mbo_valid & OBD_MD_FLEASIZE) {
480 if (!S_ISREG(md->body->mbo_mode)) {
481 CDEBUG(D_INFO, "OBD_MD_FLEASIZE set, should be a "
482 "regular file, but is not\n");
483 GOTO(out, rc = -EPROTO);
486 if (md->body->mbo_eadatasize == 0) {
487 CDEBUG(D_INFO, "OBD_MD_FLEASIZE set, "
488 "but eadatasize 0\n");
489 GOTO(out, rc = -EPROTO);
492 md->layout.lb_len = md->body->mbo_eadatasize;
493 md->layout.lb_buf = req_capsule_server_sized_get(pill,
496 if (md->layout.lb_buf == NULL)
497 GOTO(out, rc = -EPROTO);
498 } else if (md->body->mbo_valid & OBD_MD_FLDIREA) {
499 const union lmv_mds_md *lmv;
502 if (!S_ISDIR(md->body->mbo_mode)) {
503 CDEBUG(D_INFO, "OBD_MD_FLDIREA set, should be a "
504 "directory, but is not\n");
505 GOTO(out, rc = -EPROTO);
508 lmv_size = md->body->mbo_eadatasize;
510 CDEBUG(D_INFO, "OBD_MD_FLDIREA is set, "
511 "but eadatasize 0\n");
515 if (md->body->mbo_valid & OBD_MD_MEA) {
516 lmv = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
519 GOTO(out, rc = -EPROTO);
521 rc = md_unpackmd(md_exp, &md->lmv, lmv, lmv_size);
525 if (rc < (typeof(rc))sizeof(*md->lmv)) {
526 CDEBUG(D_INFO, "size too small: "
527 "rc < sizeof(*md->lmv) (%d < %d)\n",
528 rc, (int)sizeof(*md->lmv));
529 GOTO(out, rc = -EPROTO);
535 if (md->body->mbo_valid & OBD_MD_FLRMTPERM) {
536 /* remote permission */
537 LASSERT(client_is_remote(exp));
538 md->remote_perm = req_capsule_server_swab_get(pill, &RMF_ACL,
539 lustre_swab_mdt_remote_perm);
540 if (!md->remote_perm)
541 GOTO(out, rc = -EPROTO);
542 } else if (md->body->mbo_valid & OBD_MD_FLACL) {
543 /* for ACL, it's possible that FLACL is set but aclsize is zero.
544 * only when aclsize != 0 there's an actual segment for ACL
547 if (md->body->mbo_aclsize) {
548 rc = mdc_unpack_acl(req, md);
551 #ifdef CONFIG_FS_POSIX_ACL
553 md->posix_acl = NULL;
561 #ifdef CONFIG_FS_POSIX_ACL
562 posix_acl_release(md->posix_acl);
568 int mdc_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
574 void mdc_replay_open(struct ptlrpc_request *req)
576 struct md_open_data *mod = req->rq_cb_data;
577 struct ptlrpc_request *close_req;
578 struct obd_client_handle *och;
579 struct lustre_handle old;
580 struct mdt_body *body;
584 DEBUG_REQ(D_ERROR, req,
585 "Can't properly replay without open data.");
590 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
591 LASSERT(body != NULL);
595 struct lustre_handle *file_fh;
597 LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
599 file_fh = &och->och_fh;
600 CDEBUG(D_HA, "updating handle from "LPX64" to "LPX64"\n",
601 file_fh->cookie, body->mbo_handle.cookie);
603 *file_fh = body->mbo_handle;
605 close_req = mod->mod_close_req;
606 if (close_req != NULL) {
607 __u32 opc = lustre_msg_get_opc(close_req->rq_reqmsg);
608 struct mdt_ioepoch *epoch;
610 LASSERT(opc == MDS_CLOSE);
611 epoch = req_capsule_client_get(&close_req->rq_pill,
616 LASSERT(!memcmp(&old, &epoch->mio_handle, sizeof(old)));
618 DEBUG_REQ(D_HA, close_req, "updating close body with new fh");
619 epoch->mio_handle = body->mbo_handle;
624 void mdc_commit_open(struct ptlrpc_request *req)
626 struct md_open_data *mod = req->rq_cb_data;
631 * No need to touch md_open_data::mod_och, it holds a reference on
632 * \var mod and will zero references to each other, \var mod will be
633 * freed after that when md_open_data::mod_och will put the reference.
637 * Do not let open request to disappear as it still may be needed
638 * for close rpc to happen (it may happen on evict only, otherwise
639 * ptlrpc_request::rq_replay does not let mdc_commit_open() to be
640 * called), just mark this rpc as committed to distinguish these 2
641 * cases, see mdc_close() for details. The open request reference will
642 * be put along with freeing \var mod.
644 ptlrpc_request_addref(req);
645 spin_lock(&req->rq_lock);
646 req->rq_committed = 1;
647 spin_unlock(&req->rq_lock);
648 req->rq_cb_data = NULL;
652 int mdc_set_open_replay_data(struct obd_export *exp,
653 struct obd_client_handle *och,
654 struct lookup_intent *it)
656 struct md_open_data *mod;
657 struct mdt_rec_create *rec;
658 struct mdt_body *body;
659 struct ptlrpc_request *open_req = it->it_data;
660 struct obd_import *imp = open_req->rq_import;
663 if (!open_req->rq_replay)
666 rec = req_capsule_client_get(&open_req->rq_pill, &RMF_REC_REINT);
667 body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
668 LASSERT(rec != NULL);
669 /* Incoming message in my byte order (it's been swabbed). */
670 /* Outgoing messages always in my byte order. */
671 LASSERT(body != NULL);
673 /* Only if the import is replayable, we set replay_open data */
674 if (och && imp->imp_replayable) {
675 mod = obd_mod_alloc();
677 DEBUG_REQ(D_ERROR, open_req,
678 "Can't allocate md_open_data");
683 * Take a reference on \var mod, to be freed on mdc_close().
684 * It protects \var mod from being freed on eviction (commit
685 * callback is called despite rq_replay flag).
686 * Another reference for \var och.
691 spin_lock(&open_req->rq_lock);
694 mod->mod_is_create = it_disposition(it, DISP_OPEN_CREATE) ||
695 it_disposition(it, DISP_OPEN_STRIPE);
696 mod->mod_open_req = open_req;
697 open_req->rq_cb_data = mod;
698 open_req->rq_commit_cb = mdc_commit_open;
699 spin_unlock(&open_req->rq_lock);
702 rec->cr_fid2 = body->mbo_fid1;
703 rec->cr_ioepoch = body->mbo_ioepoch;
704 rec->cr_old_handle.cookie = body->mbo_handle.cookie;
705 open_req->rq_replay_cb = mdc_replay_open;
706 if (!fid_is_sane(&body->mbo_fid1)) {
707 DEBUG_REQ(D_ERROR, open_req, "Saving replay request with "
712 DEBUG_REQ(D_RPCTRACE, open_req, "Set up open replay data");
716 static void mdc_free_open(struct md_open_data *mod)
720 if (mod->mod_is_create == 0 &&
721 imp_connect_disp_stripe(mod->mod_open_req->rq_import))
724 LASSERT(mod->mod_open_req->rq_replay == 0);
726 DEBUG_REQ(D_RPCTRACE, mod->mod_open_req, "free open request\n");
728 ptlrpc_request_committed(mod->mod_open_req, committed);
729 if (mod->mod_close_req)
730 ptlrpc_request_committed(mod->mod_close_req, committed);
733 int mdc_clear_open_replay_data(struct obd_export *exp,
734 struct obd_client_handle *och)
736 struct md_open_data *mod = och->och_mod;
740 * It is possible to not have \var mod in a case of eviction between
741 * lookup and ll_file_open().
746 LASSERT(mod != LP_POISON);
747 LASSERT(mod->mod_open_req != NULL);
757 static int mdc_close(struct obd_export *exp, struct md_op_data *op_data,
758 struct md_open_data *mod, struct ptlrpc_request **request)
760 struct obd_device *obd = class_exp2obd(exp);
761 struct ptlrpc_request *req;
762 struct req_format *req_fmt;
767 if (op_data->op_bias & MDS_HSM_RELEASE) {
768 req_fmt = &RQF_MDS_INTENT_CLOSE;
770 /* allocate a FID for volatile file */
771 rc = mdc_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
773 CERROR("%s: "DFID" failed to allocate FID: %d\n",
774 obd->obd_name, PFID(&op_data->op_fid1), rc);
775 /* save the errcode and proceed to close */
778 } else if (op_data->op_bias & MDS_CLOSE_LAYOUT_SWAP) {
779 req_fmt = &RQF_MDS_INTENT_CLOSE;
781 req_fmt = &RQF_MDS_CLOSE;
785 req = ptlrpc_request_alloc(class_exp2cliimp(exp), req_fmt);
789 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_CLOSE);
791 ptlrpc_request_free(req);
795 /* To avoid a livelock (bug 7034), we need to send CLOSE RPCs to a
796 * portal whose threads are not taking any DLM locks and are therefore
797 * always progressing */
798 req->rq_request_portal = MDS_READPAGE_PORTAL;
799 ptlrpc_at_set_req_timeout(req);
801 /* Ensure that this close's handle is fixed up during replay. */
802 if (likely(mod != NULL)) {
803 LASSERTF(mod->mod_open_req != NULL &&
804 mod->mod_open_req->rq_type != LI_POISON,
805 "POISONED open %p!\n", mod->mod_open_req);
807 mod->mod_close_req = req;
809 DEBUG_REQ(D_HA, mod->mod_open_req, "matched open");
810 /* We no longer want to preserve this open for replay even
811 * though the open was committed. b=3632, b=3633 */
812 spin_lock(&mod->mod_open_req->rq_lock);
813 mod->mod_open_req->rq_replay = 0;
814 spin_unlock(&mod->mod_open_req->rq_lock);
816 CDEBUG(D_HA, "couldn't find open req; expecting close error\n");
819 mdc_close_pack(req, op_data);
821 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
822 obd->u.cli.cl_default_mds_easize);
824 ptlrpc_request_set_replen(req);
826 mdc_get_mod_rpc_slot(req, NULL);
827 rc = ptlrpc_queue_wait(req);
828 mdc_put_mod_rpc_slot(req, NULL);
830 if (req->rq_repmsg == NULL) {
831 CDEBUG(D_RPCTRACE, "request failed to send: %p, %d\n", req,
834 rc = req->rq_status ?: -EIO;
835 } else if (rc == 0 || rc == -EAGAIN) {
836 struct mdt_body *body;
838 rc = lustre_msg_get_status(req->rq_repmsg);
839 if (lustre_msg_get_type(req->rq_repmsg) == PTL_RPC_MSG_ERR) {
840 DEBUG_REQ(D_ERROR, req, "type == PTL_RPC_MSG_ERR, err "
845 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
848 } else if (rc == -ESTALE) {
850 * it can be allowed error after 3633 if open was committed and
851 * server failed before close was sent. Let's check if mod
852 * exists and return no error in that case
855 DEBUG_REQ(D_HA, req, "Reset ESTALE = %d", rc);
856 LASSERT(mod->mod_open_req != NULL);
857 if (mod->mod_open_req->rq_committed)
864 mod->mod_close_req = NULL;
865 /* Since now, mod is accessed through open_req only,
866 * thus close req does not keep a reference on mod anymore. */
871 RETURN(rc < 0 ? rc : saved_rc);
874 static int mdc_getpage(struct obd_export *exp, const struct lu_fid *fid,
875 u64 offset, struct page **pages, int npages,
876 struct ptlrpc_request **request)
878 struct ptlrpc_request *req;
879 struct ptlrpc_bulk_desc *desc;
881 wait_queue_head_t waitq;
883 struct l_wait_info lwi;
888 init_waitqueue_head(&waitq);
891 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_READPAGE);
895 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_READPAGE);
897 ptlrpc_request_free(req);
901 req->rq_request_portal = MDS_READPAGE_PORTAL;
902 ptlrpc_at_set_req_timeout(req);
904 desc = ptlrpc_prep_bulk_imp(req, npages, 1,
905 PTLRPC_BULK_PUT_SINK | PTLRPC_BULK_BUF_KIOV,
907 &ptlrpc_bulk_kiov_pin_ops);
909 ptlrpc_request_free(req);
913 /* NB req now owns desc and will free it when it gets freed */
914 for (i = 0; i < npages; i++)
915 desc->bd_frag_ops->add_kiov_frag(desc, pages[i], 0,
918 mdc_readdir_pack(req, offset, PAGE_CACHE_SIZE * npages, fid);
920 ptlrpc_request_set_replen(req);
921 rc = ptlrpc_queue_wait(req);
923 ptlrpc_req_finished(req);
924 if (rc != -ETIMEDOUT)
928 if (!client_should_resend(resends, &exp->exp_obd->u.cli)) {
929 CERROR("%s: too many resend retries: rc = %d\n",
930 exp->exp_obd->obd_name, -EIO);
933 lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(resends), NULL, NULL,
935 l_wait_event(waitq, 0, &lwi);
940 rc = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk,
941 req->rq_bulk->bd_nob_transferred);
943 ptlrpc_req_finished(req);
947 if (req->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK) {
948 CERROR("%s: unexpected bytes transferred: %d (%ld expected)\n",
949 exp->exp_obd->obd_name, req->rq_bulk->bd_nob_transferred,
950 PAGE_CACHE_SIZE * npages);
951 ptlrpc_req_finished(req);
959 static void mdc_release_page(struct page *page, int remove)
963 if (likely(page->mapping != NULL))
964 truncate_complete_page(page->mapping, page);
967 page_cache_release(page);
970 static struct page *mdc_page_locate(struct address_space *mapping, __u64 *hash,
971 __u64 *start, __u64 *end, int hash64)
974 * Complement of hash is used as an index so that
975 * radix_tree_gang_lookup() can be used to find a page with starting
976 * hash _smaller_ than one we are looking for.
978 unsigned long offset = hash_x_index(*hash, hash64);
982 spin_lock_irq(&mapping->tree_lock);
983 found = radix_tree_gang_lookup(&mapping->page_tree,
984 (void **)&page, offset, 1);
985 if (found > 0 && !radix_tree_exceptional_entry(page)) {
986 struct lu_dirpage *dp;
988 page_cache_get(page);
989 spin_unlock_irq(&mapping->tree_lock);
991 * In contrast to find_lock_page() we are sure that directory
992 * page cannot be truncated (while DLM lock is held) and,
993 * hence, can avoid restart.
995 * In fact, page cannot be locked here at all, because
996 * mdc_read_page_remote does synchronous io.
998 wait_on_page_locked(page);
999 if (PageUptodate(page)) {
1001 if (BITS_PER_LONG == 32 && hash64) {
1002 *start = le64_to_cpu(dp->ldp_hash_start) >> 32;
1003 *end = le64_to_cpu(dp->ldp_hash_end) >> 32;
1004 *hash = *hash >> 32;
1006 *start = le64_to_cpu(dp->ldp_hash_start);
1007 *end = le64_to_cpu(dp->ldp_hash_end);
1009 if (unlikely(*start == 1 && *hash == 0))
1012 LASSERTF(*start <= *hash, "start = "LPX64
1013 ",end = "LPX64",hash = "LPX64"\n",
1014 *start, *end, *hash);
1015 CDEBUG(D_VFSTRACE, "offset %lx ["LPX64" "LPX64"],"
1016 " hash "LPX64"\n", offset, *start, *end, *hash);
1019 mdc_release_page(page, 0);
1021 } else if (*end != *start && *hash == *end) {
1023 * upon hash collision, remove this page,
1024 * otherwise put page reference, and
1025 * mdc_read_page_remote() will issue RPC to
1026 * fetch the page we want.
1029 mdc_release_page(page,
1030 le32_to_cpu(dp->ldp_flags) & LDF_COLLIDE);
1034 page_cache_release(page);
1035 page = ERR_PTR(-EIO);
1038 spin_unlock_irq(&mapping->tree_lock);
1045 * Adjust a set of pages, each page containing an array of lu_dirpages,
1046 * so that each page can be used as a single logical lu_dirpage.
1048 * A lu_dirpage is laid out as follows, where s = ldp_hash_start,
1049 * e = ldp_hash_end, f = ldp_flags, p = padding, and each "ent" is a
1050 * struct lu_dirent. It has size up to LU_PAGE_SIZE. The ldp_hash_end
1051 * value is used as a cookie to request the next lu_dirpage in a
1052 * directory listing that spans multiple pages (two in this example):
1055 * .|--------v------- -----.
1056 * |s|e|f|p|ent|ent| ... |ent|
1057 * '--|-------------- -----' Each PAGE contains a single
1058 * '------. lu_dirpage.
1059 * .---------v------- -----.
1060 * |s|e|f|p|ent| 0 | ... | 0 |
1061 * '----------------- -----'
1063 * However, on hosts where the native VM page size (PAGE_CACHE_SIZE) is
1064 * larger than LU_PAGE_SIZE, a single host page may contain multiple
1065 * lu_dirpages. After reading the lu_dirpages from the MDS, the
1066 * ldp_hash_end of the first lu_dirpage refers to the one immediately
1067 * after it in the same PAGE (arrows simplified for brevity, but
1068 * in general e0==s1, e1==s2, etc.):
1070 * .-------------------- -----.
1071 * |s0|e0|f0|p|ent|ent| ... |ent|
1072 * |---v---------------- -----|
1073 * |s1|e1|f1|p|ent|ent| ... |ent|
1074 * |---v---------------- -----| Here, each PAGE contains
1075 * ... multiple lu_dirpages.
1076 * |---v---------------- -----|
1077 * |s'|e'|f'|p|ent|ent| ... |ent|
1078 * '---|---------------- -----'
1080 * .----------------------------.
1083 * This structure is transformed into a single logical lu_dirpage as follows:
1085 * - Replace e0 with e' so the request for the next lu_dirpage gets the page
1086 * labeled 'next PAGE'.
1088 * - Copy the LDF_COLLIDE flag from f' to f0 to correctly reflect whether
1089 * a hash collision with the next page exists.
1091 * - Adjust the lde_reclen of the ending entry of each lu_dirpage to span
1092 * to the first entry of the next lu_dirpage.
1094 #if PAGE_CACHE_SIZE > LU_PAGE_SIZE
1095 static void mdc_adjust_dirpages(struct page **pages, int cfs_pgs, int lu_pgs)
1099 for (i = 0; i < cfs_pgs; i++) {
1100 struct lu_dirpage *dp = kmap(pages[i]);
1101 struct lu_dirpage *first = dp;
1102 struct lu_dirent *end_dirent = NULL;
1103 struct lu_dirent *ent;
1104 __u64 hash_end = le64_to_cpu(dp->ldp_hash_end);
1105 __u32 flags = le32_to_cpu(dp->ldp_flags);
1107 while (--lu_pgs > 0) {
1108 ent = lu_dirent_start(dp);
1109 for (end_dirent = ent; ent != NULL;
1110 end_dirent = ent, ent = lu_dirent_next(ent));
1112 /* Advance dp to next lu_dirpage. */
1113 dp = (struct lu_dirpage *)((char *)dp + LU_PAGE_SIZE);
1115 /* Check if we've reached the end of the PAGE. */
1116 if (!((unsigned long)dp & ~PAGE_MASK))
1119 /* Save the hash and flags of this lu_dirpage. */
1120 hash_end = le64_to_cpu(dp->ldp_hash_end);
1121 flags = le32_to_cpu(dp->ldp_flags);
1123 /* Check if lu_dirpage contains no entries. */
1124 if (end_dirent == NULL)
1127 /* Enlarge the end entry lde_reclen from 0 to
1128 * first entry of next lu_dirpage. */
1129 LASSERT(le16_to_cpu(end_dirent->lde_reclen) == 0);
1130 end_dirent->lde_reclen =
1131 cpu_to_le16((char *)(dp->ldp_entries) -
1132 (char *)end_dirent);
1135 first->ldp_hash_end = hash_end;
1136 first->ldp_flags &= ~cpu_to_le32(LDF_COLLIDE);
1137 first->ldp_flags |= flags & cpu_to_le32(LDF_COLLIDE);
1141 LASSERTF(lu_pgs == 0, "left = %d\n", lu_pgs);
1144 #define mdc_adjust_dirpages(pages, cfs_pgs, lu_pgs) do {} while (0)
1145 #endif /* PAGE_CACHE_SIZE > LU_PAGE_SIZE */
1147 /* parameters for readdir page */
1148 struct readpage_param {
1149 struct md_op_data *rp_mod;
1152 struct obd_export *rp_exp;
1153 struct md_callback *rp_cb;
1156 #ifndef HAVE_DELETE_FROM_PAGE_CACHE
1157 static inline void delete_from_page_cache(struct page *page)
1159 remove_from_page_cache(page);
1160 page_cache_release(page);
1165 * Read pages from server.
1167 * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
1168 * a header lu_dirpage which describes the start/end hash, and whether this
1169 * page is empty (contains no dir entry) or hash collide with next page.
1170 * After client receives reply, several pages will be integrated into dir page
1171 * in PAGE_SIZE (if PAGE_SIZE greater than LU_PAGE_SIZE), and the
1172 * lu_dirpage for this integrated page will be adjusted.
1174 static int mdc_read_page_remote(void *data, struct page *page0)
1176 struct readpage_param *rp = data;
1177 struct page **page_pool;
1179 struct lu_dirpage *dp;
1180 int rd_pgs = 0; /* number of pages read actually */
1182 struct md_op_data *op_data = rp->rp_mod;
1183 struct ptlrpc_request *req;
1184 int max_pages = op_data->op_max_pages;
1185 struct inode *inode;
1191 LASSERT(max_pages > 0 && max_pages <= PTLRPC_MAX_BRW_PAGES);
1192 inode = op_data->op_data;
1193 fid = &op_data->op_fid1;
1194 LASSERT(inode != NULL);
1196 OBD_ALLOC(page_pool, sizeof(page_pool[0]) * max_pages);
1197 if (page_pool != NULL) {
1198 page_pool[0] = page0;
1204 for (npages = 1; npages < max_pages; npages++) {
1205 page = page_cache_alloc_cold(inode->i_mapping);
1208 page_pool[npages] = page;
1211 rc = mdc_getpage(rp->rp_exp, fid, rp->rp_off, page_pool, npages, &req);
1213 /* page0 is special, which was added into page cache early */
1214 delete_from_page_cache(page0);
1218 rd_pgs = (req->rq_bulk->bd_nob_transferred +
1219 PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1220 lu_pgs = req->rq_bulk->bd_nob_transferred >>
1222 LASSERT(!(req->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK));
1224 CDEBUG(D_INODE, "read %d(%d) pages\n", rd_pgs, lu_pgs);
1226 mdc_adjust_dirpages(page_pool, rd_pgs, lu_pgs);
1228 SetPageUptodate(page0);
1232 ptlrpc_req_finished(req);
1233 CDEBUG(D_CACHE, "read %d/%d pages\n", rd_pgs, npages);
1234 for (i = 1; i < npages; i++) {
1235 unsigned long offset;
1239 page = page_pool[i];
1241 if (rc < 0 || i >= rd_pgs) {
1242 page_cache_release(page);
1246 SetPageUptodate(page);
1249 hash = le64_to_cpu(dp->ldp_hash_start);
1252 offset = hash_x_index(hash, rp->rp_hash64);
1254 prefetchw(&page->flags);
1255 ret = add_to_page_cache_lru(page, inode->i_mapping, offset,
1260 CDEBUG(D_VFSTRACE, "page %lu add to page cache failed:"
1261 " rc = %d\n", offset, ret);
1262 page_cache_release(page);
1265 if (page_pool != &page0)
1266 OBD_FREE(page_pool, sizeof(page_pool[0]) * max_pages);
1272 * Read dir page from cache first, if it can not find it, read it from
1273 * server and add into the cache.
1275 * \param[in] exp MDC export
1276 * \param[in] op_data client MD stack parameters, transfering parameters
1277 * between different layers on client MD stack.
1278 * \param[in] cb_op callback required for ldlm lock enqueue during
1280 * \param[in] hash_offset the hash offset of the page to be read
1281 * \param[in] ppage the page to be read
1283 * retval = 0 get the page successfully
1284 * errno(<0) get the page failed
1286 static int mdc_read_page(struct obd_export *exp, struct md_op_data *op_data,
1287 struct md_callback *cb_op, __u64 hash_offset,
1288 struct page **ppage)
1290 struct lookup_intent it = { .it_op = IT_READDIR };
1292 struct inode *dir = op_data->op_data;
1293 struct address_space *mapping;
1294 struct lu_dirpage *dp;
1297 struct lustre_handle lockh;
1298 struct ptlrpc_request *enq_req = NULL;
1299 struct readpage_param rp_param;
1306 LASSERT(dir != NULL);
1307 mapping = dir->i_mapping;
1309 rc = mdc_intent_lock(exp, op_data, &it, &enq_req,
1310 cb_op->md_blocking_ast, 0);
1311 if (enq_req != NULL)
1312 ptlrpc_req_finished(enq_req);
1315 CERROR("%s: "DFID" lock enqueue fails: rc = %d\n",
1316 exp->exp_obd->obd_name, PFID(&op_data->op_fid1), rc);
1321 mdc_set_lock_data(exp, &it.it_lock_handle, dir, NULL);
1323 rp_param.rp_off = hash_offset;
1324 rp_param.rp_hash64 = op_data->op_cli_flags & CLI_HASH64;
1325 page = mdc_page_locate(mapping, &rp_param.rp_off, &start, &end,
1326 rp_param.rp_hash64);
1328 CERROR("%s: dir page locate: "DFID" at "LPU64": rc %ld\n",
1329 exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1330 rp_param.rp_off, PTR_ERR(page));
1331 GOTO(out_unlock, rc = PTR_ERR(page));
1332 } else if (page != NULL) {
1334 * XXX nikita: not entirely correct handling of a corner case:
1335 * suppose hash chain of entries with hash value HASH crosses
1336 * border between pages P0 and P1. First both P0 and P1 are
1337 * cached, seekdir() is called for some entry from the P0 part
1338 * of the chain. Later P0 goes out of cache. telldir(HASH)
1339 * happens and finds P1, as it starts with matching hash
1340 * value. Remaining entries from P0 part of the chain are
1341 * skipped. (Is that really a bug?)
1343 * Possible solutions: 0. don't cache P1 is such case, handle
1344 * it as an "overflow" page. 1. invalidate all pages at
1345 * once. 2. use HASH|1 as an index for P1.
1347 GOTO(hash_collision, page);
1350 rp_param.rp_exp = exp;
1351 rp_param.rp_mod = op_data;
1352 page = read_cache_page(mapping,
1353 hash_x_index(rp_param.rp_off,
1354 rp_param.rp_hash64),
1355 mdc_read_page_remote, &rp_param);
1357 CDEBUG(D_INFO, "%s: read cache page: "DFID" at "LPU64": %ld\n",
1358 exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1359 rp_param.rp_off, PTR_ERR(page));
1360 GOTO(out_unlock, rc = PTR_ERR(page));
1363 wait_on_page_locked(page);
1365 if (!PageUptodate(page)) {
1366 CERROR("%s: page not updated: "DFID" at "LPU64": rc %d\n",
1367 exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1368 rp_param.rp_off, -5);
1371 if (!PageChecked(page))
1372 SetPageChecked(page);
1373 if (PageError(page)) {
1374 CERROR("%s: page error: "DFID" at "LPU64": rc %d\n",
1375 exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1376 rp_param.rp_off, -5);
1381 dp = page_address(page);
1382 if (BITS_PER_LONG == 32 && rp_param.rp_hash64) {
1383 start = le64_to_cpu(dp->ldp_hash_start) >> 32;
1384 end = le64_to_cpu(dp->ldp_hash_end) >> 32;
1385 rp_param.rp_off = hash_offset >> 32;
1387 start = le64_to_cpu(dp->ldp_hash_start);
1388 end = le64_to_cpu(dp->ldp_hash_end);
1389 rp_param.rp_off = hash_offset;
1392 LASSERT(start == rp_param.rp_off);
1393 CWARN("Page-wide hash collision: %#lx\n", (unsigned long)end);
1394 #if BITS_PER_LONG == 32
1395 CWARN("Real page-wide hash collision at ["LPU64" "LPU64"] with "
1396 "hash "LPU64"\n", le64_to_cpu(dp->ldp_hash_start),
1397 le64_to_cpu(dp->ldp_hash_end), hash_offset);
1401 * Fetch whole overflow chain...
1409 lockh.cookie = it.it_lock_handle;
1410 ldlm_lock_decref(&lockh, it.it_lock_mode);
1411 it.it_lock_handle = 0;
1415 mdc_release_page(page, 1);
1421 static int mdc_statfs(const struct lu_env *env,
1422 struct obd_export *exp, struct obd_statfs *osfs,
1423 __u64 max_age, __u32 flags)
1425 struct obd_device *obd = class_exp2obd(exp);
1426 struct ptlrpc_request *req;
1427 struct obd_statfs *msfs;
1428 struct obd_import *imp = NULL;
1433 * Since the request might also come from lprocfs, so we need
1434 * sync this with client_disconnect_export Bug15684
1436 down_read(&obd->u.cli.cl_sem);
1437 if (obd->u.cli.cl_import)
1438 imp = class_import_get(obd->u.cli.cl_import);
1439 up_read(&obd->u.cli.cl_sem);
1443 req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_STATFS,
1444 LUSTRE_MDS_VERSION, MDS_STATFS);
1446 GOTO(output, rc = -ENOMEM);
1448 ptlrpc_request_set_replen(req);
1450 if (flags & OBD_STATFS_NODELAY) {
1451 /* procfs requests not want stay in wait for avoid deadlock */
1452 req->rq_no_resend = 1;
1453 req->rq_no_delay = 1;
1456 rc = ptlrpc_queue_wait(req);
1458 /* check connection error first */
1459 if (imp->imp_connect_error)
1460 rc = imp->imp_connect_error;
1464 msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
1466 GOTO(out, rc = -EPROTO);
1471 ptlrpc_req_finished(req);
1473 class_import_put(imp);
1477 static int mdc_ioc_fid2path(struct obd_export *exp, struct getinfo_fid2path *gf)
1479 __u32 keylen, vallen;
1483 if (gf->gf_pathlen > PATH_MAX)
1484 RETURN(-ENAMETOOLONG);
1485 if (gf->gf_pathlen < 2)
1488 /* Key is KEY_FID2PATH + getinfo_fid2path description */
1489 keylen = cfs_size_round(sizeof(KEY_FID2PATH) + sizeof(*gf) +
1490 sizeof(struct lu_fid));
1491 OBD_ALLOC(key, keylen);
1494 memcpy(key, KEY_FID2PATH, sizeof(KEY_FID2PATH));
1495 memcpy(key + cfs_size_round(sizeof(KEY_FID2PATH)), gf, sizeof(*gf));
1496 memcpy(key + cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*gf),
1497 gf->gf_u.gf_root_fid, sizeof(struct lu_fid));
1498 CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n",
1499 PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno);
1501 if (!fid_is_sane(&gf->gf_fid))
1502 GOTO(out, rc = -EINVAL);
1504 /* Val is struct getinfo_fid2path result plus path */
1505 vallen = sizeof(*gf) + gf->gf_pathlen;
1507 rc = obd_get_info(NULL, exp, keylen, key, &vallen, gf);
1508 if (rc != 0 && rc != -EREMOTE)
1511 if (vallen <= sizeof(*gf))
1512 GOTO(out, rc = -EPROTO);
1513 if (vallen > sizeof(*gf) + gf->gf_pathlen)
1514 GOTO(out, rc = -EOVERFLOW);
1516 CDEBUG(D_IOCTL, "path got "DFID" from "LPU64" #%d: %s\n",
1517 PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno,
1518 gf->gf_pathlen < 512 ? gf->gf_u.gf_path :
1519 /* only log the last 512 characters of the path */
1520 gf->gf_u.gf_path + gf->gf_pathlen - 512);
1523 OBD_FREE(key, keylen);
1527 static int mdc_ioc_hsm_progress(struct obd_export *exp,
1528 struct hsm_progress_kernel *hpk)
1530 struct obd_import *imp = class_exp2cliimp(exp);
1531 struct hsm_progress_kernel *req_hpk;
1532 struct ptlrpc_request *req;
1536 req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_PROGRESS,
1537 LUSTRE_MDS_VERSION, MDS_HSM_PROGRESS);
1539 GOTO(out, rc = -ENOMEM);
1541 mdc_pack_body(req, NULL, OBD_MD_FLRMTPERM, 0, -1, 0);
1543 /* Copy hsm_progress struct */
1544 req_hpk = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_PROGRESS);
1545 if (req_hpk == NULL)
1546 GOTO(out, rc = -EPROTO);
1549 req_hpk->hpk_errval = lustre_errno_hton(hpk->hpk_errval);
1551 ptlrpc_request_set_replen(req);
1553 mdc_get_mod_rpc_slot(req, NULL);
1554 rc = ptlrpc_queue_wait(req);
1555 mdc_put_mod_rpc_slot(req, NULL);
1559 ptlrpc_req_finished(req);
1563 static int mdc_ioc_hsm_ct_register(struct obd_import *imp, __u32 archives)
1565 __u32 *archive_mask;
1566 struct ptlrpc_request *req;
1570 req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_REGISTER,
1572 MDS_HSM_CT_REGISTER);
1574 GOTO(out, rc = -ENOMEM);
1576 mdc_pack_body(req, NULL, OBD_MD_FLRMTPERM, 0, -1, 0);
1578 /* Copy hsm_progress struct */
1579 archive_mask = req_capsule_client_get(&req->rq_pill,
1580 &RMF_MDS_HSM_ARCHIVE);
1581 if (archive_mask == NULL)
1582 GOTO(out, rc = -EPROTO);
1584 *archive_mask = archives;
1586 ptlrpc_request_set_replen(req);
1588 rc = mdc_queue_wait(req);
1591 ptlrpc_req_finished(req);
1595 static int mdc_ioc_hsm_current_action(struct obd_export *exp,
1596 struct md_op_data *op_data)
1598 struct hsm_current_action *hca = op_data->op_data;
1599 struct hsm_current_action *req_hca;
1600 struct ptlrpc_request *req;
1604 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1605 &RQF_MDS_HSM_ACTION);
1609 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_ACTION);
1611 ptlrpc_request_free(req);
1615 mdc_pack_body(req, &op_data->op_fid1, OBD_MD_FLRMTPERM, 0,
1616 op_data->op_suppgids[0], 0);
1618 ptlrpc_request_set_replen(req);
1620 rc = mdc_queue_wait(req);
1624 req_hca = req_capsule_server_get(&req->rq_pill,
1625 &RMF_MDS_HSM_CURRENT_ACTION);
1626 if (req_hca == NULL)
1627 GOTO(out, rc = -EPROTO);
1633 ptlrpc_req_finished(req);
1637 static int mdc_ioc_hsm_ct_unregister(struct obd_import *imp)
1639 struct ptlrpc_request *req;
1643 req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_UNREGISTER,
1645 MDS_HSM_CT_UNREGISTER);
1647 GOTO(out, rc = -ENOMEM);
1649 mdc_pack_body(req, NULL, OBD_MD_FLRMTPERM, 0, -1, 0);
1651 ptlrpc_request_set_replen(req);
1653 rc = mdc_queue_wait(req);
1656 ptlrpc_req_finished(req);
1660 static int mdc_ioc_hsm_state_get(struct obd_export *exp,
1661 struct md_op_data *op_data)
1663 struct hsm_user_state *hus = op_data->op_data;
1664 struct hsm_user_state *req_hus;
1665 struct ptlrpc_request *req;
1669 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1670 &RQF_MDS_HSM_STATE_GET);
1674 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_GET);
1676 ptlrpc_request_free(req);
1680 mdc_pack_body(req, &op_data->op_fid1, OBD_MD_FLRMTPERM, 0,
1681 op_data->op_suppgids[0], 0);
1683 ptlrpc_request_set_replen(req);
1685 rc = mdc_queue_wait(req);
1689 req_hus = req_capsule_server_get(&req->rq_pill, &RMF_HSM_USER_STATE);
1690 if (req_hus == NULL)
1691 GOTO(out, rc = -EPROTO);
1697 ptlrpc_req_finished(req);
1701 static int mdc_ioc_hsm_state_set(struct obd_export *exp,
1702 struct md_op_data *op_data)
1704 struct hsm_state_set *hss = op_data->op_data;
1705 struct hsm_state_set *req_hss;
1706 struct ptlrpc_request *req;
1710 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1711 &RQF_MDS_HSM_STATE_SET);
1715 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_SET);
1717 ptlrpc_request_free(req);
1721 mdc_pack_body(req, &op_data->op_fid1, OBD_MD_FLRMTPERM, 0,
1722 op_data->op_suppgids[0], 0);
1725 req_hss = req_capsule_client_get(&req->rq_pill, &RMF_HSM_STATE_SET);
1726 if (req_hss == NULL)
1727 GOTO(out, rc = -EPROTO);
1730 ptlrpc_request_set_replen(req);
1732 mdc_get_mod_rpc_slot(req, NULL);
1733 rc = ptlrpc_queue_wait(req);
1734 mdc_put_mod_rpc_slot(req, NULL);
1738 ptlrpc_req_finished(req);
1742 static int mdc_ioc_hsm_request(struct obd_export *exp,
1743 struct hsm_user_request *hur)
1745 struct obd_import *imp = class_exp2cliimp(exp);
1746 struct ptlrpc_request *req;
1747 struct hsm_request *req_hr;
1748 struct hsm_user_item *req_hui;
1753 req = ptlrpc_request_alloc(imp, &RQF_MDS_HSM_REQUEST);
1755 GOTO(out, rc = -ENOMEM);
1757 req_capsule_set_size(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM, RCL_CLIENT,
1758 hur->hur_request.hr_itemcount
1759 * sizeof(struct hsm_user_item));
1760 req_capsule_set_size(&req->rq_pill, &RMF_GENERIC_DATA, RCL_CLIENT,
1761 hur->hur_request.hr_data_len);
1763 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_REQUEST);
1765 ptlrpc_request_free(req);
1769 mdc_pack_body(req, NULL, OBD_MD_FLRMTPERM, 0, -1, 0);
1771 /* Copy hsm_request struct */
1772 req_hr = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_REQUEST);
1774 GOTO(out, rc = -EPROTO);
1775 *req_hr = hur->hur_request;
1777 /* Copy hsm_user_item structs */
1778 req_hui = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM);
1779 if (req_hui == NULL)
1780 GOTO(out, rc = -EPROTO);
1781 memcpy(req_hui, hur->hur_user_item,
1782 hur->hur_request.hr_itemcount * sizeof(struct hsm_user_item));
1784 /* Copy opaque field */
1785 req_opaque = req_capsule_client_get(&req->rq_pill, &RMF_GENERIC_DATA);
1786 if (req_opaque == NULL)
1787 GOTO(out, rc = -EPROTO);
1788 memcpy(req_opaque, hur_data(hur), hur->hur_request.hr_data_len);
1790 ptlrpc_request_set_replen(req);
1792 mdc_get_mod_rpc_slot(req, NULL);
1793 rc = ptlrpc_queue_wait(req);
1794 mdc_put_mod_rpc_slot(req, NULL);
1799 ptlrpc_req_finished(req);
1803 static struct kuc_hdr *changelog_kuc_hdr(char *buf, size_t len, __u32 flags)
1805 struct kuc_hdr *lh = (struct kuc_hdr *)buf;
1807 LASSERT(len <= KUC_CHANGELOG_MSG_MAXSIZE);
1809 lh->kuc_magic = KUC_MAGIC;
1810 lh->kuc_transport = KUC_TRANSPORT_CHANGELOG;
1811 lh->kuc_flags = flags;
1812 lh->kuc_msgtype = CL_RECORD;
1813 lh->kuc_msglen = len;
1817 struct changelog_show {
1819 enum changelog_send_flag cs_flags;
1822 struct obd_device *cs_obd;
1825 static inline char *cs_obd_name(struct changelog_show *cs)
1827 return cs->cs_obd->obd_name;
1830 static int changelog_kkuc_cb(const struct lu_env *env, struct llog_handle *llh,
1831 struct llog_rec_hdr *hdr, void *data)
1833 struct changelog_show *cs = data;
1834 struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
1840 if (rec->cr_hdr.lrh_type != CHANGELOG_REC) {
1842 CERROR("%s: not a changelog rec %x/%d: rc = %d\n",
1843 cs_obd_name(cs), rec->cr_hdr.lrh_type,
1844 rec->cr.cr_type, rc);
1848 if (rec->cr.cr_index < cs->cs_startrec) {
1849 /* Skip entries earlier than what we are interested in */
1850 CDEBUG(D_HSM, "rec="LPU64" start="LPU64"\n",
1851 rec->cr.cr_index, cs->cs_startrec);
1855 CDEBUG(D_HSM, LPU64" %02d%-5s "LPU64" 0x%x t="DFID" p="DFID" %.*s\n",
1856 rec->cr.cr_index, rec->cr.cr_type,
1857 changelog_type2str(rec->cr.cr_type), rec->cr.cr_time,
1858 rec->cr.cr_flags & CLF_FLAGMASK,
1859 PFID(&rec->cr.cr_tfid), PFID(&rec->cr.cr_pfid),
1860 rec->cr.cr_namelen, changelog_rec_name(&rec->cr));
1862 len = sizeof(*lh) + changelog_rec_size(&rec->cr) + rec->cr.cr_namelen;
1864 /* Set up the message */
1865 lh = changelog_kuc_hdr(cs->cs_buf, len, cs->cs_flags);
1866 memcpy(lh + 1, &rec->cr, len - sizeof(*lh));
1868 rc = libcfs_kkuc_msg_put(cs->cs_fp, lh);
1869 CDEBUG(D_HSM, "kucmsg fp %p len %zu rc %d\n", cs->cs_fp, len, rc);
1874 static int mdc_changelog_send_thread(void *csdata)
1876 struct changelog_show *cs = csdata;
1877 struct llog_ctxt *ctxt = NULL;
1878 struct llog_handle *llh = NULL;
1879 struct kuc_hdr *kuch;
1880 enum llog_flag flags = LLOG_F_IS_CAT;
1883 CDEBUG(D_HSM, "changelog to fp=%p start "LPU64"\n",
1884 cs->cs_fp, cs->cs_startrec);
1886 OBD_ALLOC(cs->cs_buf, KUC_CHANGELOG_MSG_MAXSIZE);
1887 if (cs->cs_buf == NULL)
1888 GOTO(out, rc = -ENOMEM);
1890 /* Set up the remote catalog handle */
1891 ctxt = llog_get_context(cs->cs_obd, LLOG_CHANGELOG_REPL_CTXT);
1893 GOTO(out, rc = -ENOENT);
1894 rc = llog_open(NULL, ctxt, &llh, NULL, CHANGELOG_CATALOG,
1897 CERROR("%s: fail to open changelog catalog: rc = %d\n",
1898 cs_obd_name(cs), rc);
1902 if (cs->cs_flags & CHANGELOG_FLAG_JOBID)
1903 flags |= LLOG_F_EXT_JOBID;
1905 rc = llog_init_handle(NULL, llh, flags, NULL);
1907 CERROR("llog_init_handle failed %d\n", rc);
1911 rc = llog_cat_process(NULL, llh, changelog_kkuc_cb, cs, 0, 0);
1913 /* Send EOF no matter what our result */
1914 kuch = changelog_kuc_hdr(cs->cs_buf, sizeof(*kuch), cs->cs_flags);
1915 kuch->kuc_msgtype = CL_EOF;
1916 libcfs_kkuc_msg_put(cs->cs_fp, kuch);
1921 llog_cat_close(NULL, llh);
1923 llog_ctxt_put(ctxt);
1925 OBD_FREE(cs->cs_buf, KUC_CHANGELOG_MSG_MAXSIZE);
1930 static int mdc_ioc_changelog_send(struct obd_device *obd,
1931 struct ioc_changelog *icc)
1933 struct changelog_show *cs;
1934 struct task_struct *task;
1937 /* Freed in mdc_changelog_send_thread */
1943 cs->cs_startrec = icc->icc_recno;
1944 /* matching fput in mdc_changelog_send_thread */
1945 cs->cs_fp = fget(icc->icc_id);
1946 cs->cs_flags = icc->icc_flags;
1949 * New thread because we should return to user app before
1950 * writing into our pipe
1952 task = kthread_run(mdc_changelog_send_thread, cs,
1953 "mdc_clg_send_thread");
1956 CERROR("%s: cannot start changelog thread: rc = %d\n",
1957 cs_obd_name(cs), rc);
1961 CDEBUG(D_HSM, "%s: started changelog thread\n",
1968 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
1969 struct lustre_kernelcomm *lk);
1971 static int mdc_quotactl(struct obd_device *unused, struct obd_export *exp,
1972 struct obd_quotactl *oqctl)
1974 struct ptlrpc_request *req;
1975 struct obd_quotactl *oqc;
1979 req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
1980 &RQF_MDS_QUOTACTL, LUSTRE_MDS_VERSION,
1985 oqc = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1988 ptlrpc_request_set_replen(req);
1989 ptlrpc_at_set_req_timeout(req);
1990 req->rq_no_resend = 1;
1992 rc = ptlrpc_queue_wait(req);
1994 CERROR("ptlrpc_queue_wait failed, rc: %d\n", rc);
1996 if (req->rq_repmsg &&
1997 (oqc = req_capsule_server_get(&req->rq_pill, &RMF_OBD_QUOTACTL))) {
2000 CERROR ("Can't unpack obd_quotactl\n");
2003 ptlrpc_req_finished(req);
2008 static int mdc_ioc_swap_layouts(struct obd_export *exp,
2009 struct md_op_data *op_data)
2011 struct list_head cancels = LIST_HEAD_INIT(cancels);
2012 struct ptlrpc_request *req;
2014 struct mdc_swap_layouts *msl, *payload;
2017 msl = op_data->op_data;
2019 /* When the MDT will get the MDS_SWAP_LAYOUTS RPC the
2020 * first thing it will do is to cancel the 2 layout
2021 * locks held by this client.
2022 * So the client must cancel its layout locks on the 2 fids
2023 * with the request RPC to avoid extra RPC round trips.
2025 count = mdc_resource_get_unused(exp, &op_data->op_fid1, &cancels,
2026 LCK_EX, MDS_INODELOCK_LAYOUT |
2027 MDS_INODELOCK_XATTR);
2028 count += mdc_resource_get_unused(exp, &op_data->op_fid2, &cancels,
2029 LCK_EX, MDS_INODELOCK_LAYOUT |
2030 MDS_INODELOCK_XATTR);
2032 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
2033 &RQF_MDS_SWAP_LAYOUTS);
2035 ldlm_lock_list_put(&cancels, l_bl_ast, count);
2039 rc = mdc_prep_elc_req(exp, req, MDS_SWAP_LAYOUTS, &cancels, count);
2041 ptlrpc_request_free(req);
2045 mdc_swap_layouts_pack(req, op_data);
2047 payload = req_capsule_client_get(&req->rq_pill, &RMF_SWAP_LAYOUTS);
2052 ptlrpc_request_set_replen(req);
2054 rc = ptlrpc_queue_wait(req);
2060 ptlrpc_req_finished(req);
2064 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2065 void *karg, void __user *uarg)
2067 struct obd_device *obd = exp->exp_obd;
2068 struct obd_ioctl_data *data = karg;
2069 struct obd_import *imp = obd->u.cli.cl_import;
2073 if (!try_module_get(THIS_MODULE)) {
2074 CERROR("%s: cannot get module '%s'\n", obd->obd_name,
2075 module_name(THIS_MODULE));
2079 case OBD_IOC_CHANGELOG_SEND:
2080 rc = mdc_ioc_changelog_send(obd, karg);
2082 case OBD_IOC_CHANGELOG_CLEAR: {
2083 struct ioc_changelog *icc = karg;
2084 struct changelog_setinfo cs =
2085 {.cs_recno = icc->icc_recno, .cs_id = icc->icc_id};
2086 rc = obd_set_info_async(NULL, exp, strlen(KEY_CHANGELOG_CLEAR),
2087 KEY_CHANGELOG_CLEAR, sizeof(cs), &cs,
2091 case OBD_IOC_FID2PATH:
2092 rc = mdc_ioc_fid2path(exp, karg);
2094 case LL_IOC_HSM_CT_START:
2095 rc = mdc_ioc_hsm_ct_start(exp, karg);
2096 /* ignore if it was already registered on this MDS. */
2100 case LL_IOC_HSM_PROGRESS:
2101 rc = mdc_ioc_hsm_progress(exp, karg);
2103 case LL_IOC_HSM_STATE_GET:
2104 rc = mdc_ioc_hsm_state_get(exp, karg);
2106 case LL_IOC_HSM_STATE_SET:
2107 rc = mdc_ioc_hsm_state_set(exp, karg);
2109 case LL_IOC_HSM_ACTION:
2110 rc = mdc_ioc_hsm_current_action(exp, karg);
2112 case LL_IOC_HSM_REQUEST:
2113 rc = mdc_ioc_hsm_request(exp, karg);
2115 case OBD_IOC_CLIENT_RECOVER:
2116 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1, 0);
2120 case IOC_OSC_SET_ACTIVE:
2121 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
2123 case OBD_IOC_PING_TARGET:
2124 rc = ptlrpc_obd_ping(obd);
2127 * Normally IOC_OBD_STATFS, OBD_IOC_QUOTACTL iocontrol are handled by
2128 * LMV instead of MDC. But when the cluster is upgraded from 1.8,
2129 * there'd be no LMV layer thus we might be called here. Eventually
2130 * this code should be removed.
2133 case IOC_OBD_STATFS: {
2134 struct obd_statfs stat_buf = {0};
2136 if (*((__u32 *) data->ioc_inlbuf2) != 0)
2137 GOTO(out, rc = -ENODEV);
2140 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(obd),
2141 min((int)data->ioc_plen2,
2142 (int)sizeof(struct obd_uuid))))
2143 GOTO(out, rc = -EFAULT);
2145 rc = mdc_statfs(NULL, obd->obd_self_export, &stat_buf,
2146 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
2151 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
2152 min((int) data->ioc_plen1,
2153 (int) sizeof(stat_buf))))
2154 GOTO(out, rc = -EFAULT);
2158 case OBD_IOC_QUOTACTL: {
2159 struct if_quotactl *qctl = karg;
2160 struct obd_quotactl *oqctl;
2162 OBD_ALLOC_PTR(oqctl);
2164 GOTO(out, rc = -ENOMEM);
2166 QCTL_COPY(oqctl, qctl);
2167 rc = obd_quotactl(exp, oqctl);
2169 QCTL_COPY(qctl, oqctl);
2170 qctl->qc_valid = QC_MDTIDX;
2171 qctl->obd_uuid = obd->u.cli.cl_target_uuid;
2174 OBD_FREE_PTR(oqctl);
2177 case LL_IOC_GET_CONNECT_FLAGS:
2178 if (copy_to_user(uarg, exp_connect_flags_ptr(exp),
2179 sizeof(*exp_connect_flags_ptr(exp))))
2180 GOTO(out, rc = -EFAULT);
2183 case LL_IOC_LOV_SWAP_LAYOUTS:
2184 rc = mdc_ioc_swap_layouts(exp, karg);
2187 CERROR("unrecognised ioctl: cmd = %#x\n", cmd);
2188 GOTO(out, rc = -ENOTTY);
2191 module_put(THIS_MODULE);
2196 static int mdc_get_info_rpc(struct obd_export *exp,
2197 u32 keylen, void *key,
2198 u32 vallen, void *val)
2200 struct obd_import *imp = class_exp2cliimp(exp);
2201 struct ptlrpc_request *req;
2206 req = ptlrpc_request_alloc(imp, &RQF_MDS_GET_INFO);
2210 req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_KEY,
2211 RCL_CLIENT, keylen);
2212 req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VALLEN,
2213 RCL_CLIENT, sizeof(vallen));
2215 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GET_INFO);
2217 ptlrpc_request_free(req);
2221 tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_KEY);
2222 memcpy(tmp, key, keylen);
2223 tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_VALLEN);
2224 memcpy(tmp, &vallen, sizeof(vallen));
2226 req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VAL,
2227 RCL_SERVER, vallen);
2228 ptlrpc_request_set_replen(req);
2230 rc = ptlrpc_queue_wait(req);
2231 /* -EREMOTE means the get_info result is partial, and it needs to
2232 * continue on another MDT, see fid2path part in lmv_iocontrol */
2233 if (rc == 0 || rc == -EREMOTE) {
2234 tmp = req_capsule_server_get(&req->rq_pill, &RMF_GETINFO_VAL);
2235 memcpy(val, tmp, vallen);
2236 if (ptlrpc_rep_need_swab(req)) {
2237 if (KEY_IS(KEY_FID2PATH))
2238 lustre_swab_fid2path(val);
2241 ptlrpc_req_finished(req);
2246 static void lustre_swab_hai(struct hsm_action_item *h)
2248 __swab32s(&h->hai_len);
2249 __swab32s(&h->hai_action);
2250 lustre_swab_lu_fid(&h->hai_fid);
2251 lustre_swab_lu_fid(&h->hai_dfid);
2252 __swab64s(&h->hai_cookie);
2253 __swab64s(&h->hai_extent.offset);
2254 __swab64s(&h->hai_extent.length);
2255 __swab64s(&h->hai_gid);
2258 static void lustre_swab_hal(struct hsm_action_list *h)
2260 struct hsm_action_item *hai;
2263 __swab32s(&h->hal_version);
2264 __swab32s(&h->hal_count);
2265 __swab32s(&h->hal_archive_id);
2266 __swab64s(&h->hal_flags);
2268 for (i = 0; i < h->hal_count; i++, hai = hai_next(hai))
2269 lustre_swab_hai(hai);
2272 static void lustre_swab_kuch(struct kuc_hdr *l)
2274 __swab16s(&l->kuc_magic);
2275 /* __u8 l->kuc_transport */
2276 __swab16s(&l->kuc_msgtype);
2277 __swab16s(&l->kuc_msglen);
2280 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
2281 struct lustre_kernelcomm *lk)
2283 struct obd_import *imp = class_exp2cliimp(exp);
2284 __u32 archive = lk->lk_data;
2287 if (lk->lk_group != KUC_GRP_HSM) {
2288 CERROR("Bad copytool group %d\n", lk->lk_group);
2292 CDEBUG(D_HSM, "CT start r%d w%d u%d g%d f%#x\n", lk->lk_rfd, lk->lk_wfd,
2293 lk->lk_uid, lk->lk_group, lk->lk_flags);
2295 if (lk->lk_flags & LK_FLG_STOP) {
2296 /* Unregister with the coordinator */
2297 rc = mdc_ioc_hsm_ct_unregister(imp);
2299 rc = mdc_ioc_hsm_ct_register(imp, archive);
2306 * Send a message to any listening copytools
2307 * @param val KUC message (kuc_hdr + hsm_action_list)
2308 * @param len total length of message
2310 static int mdc_hsm_copytool_send(size_t len, void *val)
2312 struct kuc_hdr *lh = (struct kuc_hdr *)val;
2313 struct hsm_action_list *hal = (struct hsm_action_list *)(lh + 1);
2317 if (len < sizeof(*lh) + sizeof(*hal)) {
2318 CERROR("Short HSM message %zu < %zu\n", len,
2319 sizeof(*lh) + sizeof(*hal));
2322 if (lh->kuc_magic == __swab16(KUC_MAGIC)) {
2323 lustre_swab_kuch(lh);
2324 lustre_swab_hal(hal);
2325 } else if (lh->kuc_magic != KUC_MAGIC) {
2326 CERROR("Bad magic %x!=%x\n", lh->kuc_magic, KUC_MAGIC);
2330 CDEBUG(D_HSM, " Received message mg=%x t=%d m=%d l=%d actions=%d "
2332 lh->kuc_magic, lh->kuc_transport, lh->kuc_msgtype,
2333 lh->kuc_msglen, hal->hal_count, hal->hal_fsname);
2335 /* Broadcast to HSM listeners */
2336 rc = libcfs_kkuc_group_put(KUC_GRP_HSM, lh);
2342 * callback function passed to kuc for re-registering each HSM copytool
2343 * running on MDC, after MDT shutdown/recovery.
2344 * @param data copytool registration data
2345 * @param cb_arg callback argument (obd_import)
2347 static int mdc_hsm_ct_reregister(void *data, void *cb_arg)
2349 struct kkuc_ct_data *kcd = data;
2350 struct obd_import *imp = (struct obd_import *)cb_arg;
2353 if (kcd == NULL || kcd->kcd_magic != KKUC_CT_DATA_MAGIC)
2356 if (!obd_uuid_equals(&kcd->kcd_uuid, &imp->imp_obd->obd_uuid))
2359 CDEBUG(D_HA, "%s: recover copytool registration to MDT (archive=%#x)\n",
2360 imp->imp_obd->obd_name, kcd->kcd_archive);
2361 rc = mdc_ioc_hsm_ct_register(imp, kcd->kcd_archive);
2363 /* ignore error if the copytool is already registered */
2364 return (rc == -EEXIST) ? 0 : rc;
2368 * Re-establish all kuc contexts with MDT
2369 * after MDT shutdown/recovery.
2371 static int mdc_kuc_reregister(struct obd_import *imp)
2373 /* re-register HSM agents */
2374 return libcfs_kkuc_group_foreach(KUC_GRP_HSM, mdc_hsm_ct_reregister,
2378 static int mdc_set_info_async(const struct lu_env *env,
2379 struct obd_export *exp,
2380 u32 keylen, void *key,
2381 u32 vallen, void *val,
2382 struct ptlrpc_request_set *set)
2384 struct obd_import *imp = class_exp2cliimp(exp);
2388 if (KEY_IS(KEY_READ_ONLY)) {
2389 if (vallen != sizeof(int))
2392 spin_lock(&imp->imp_lock);
2393 if (*((int *)val)) {
2394 imp->imp_connect_flags_orig |= OBD_CONNECT_RDONLY;
2395 imp->imp_connect_data.ocd_connect_flags |=
2398 imp->imp_connect_flags_orig &= ~OBD_CONNECT_RDONLY;
2399 imp->imp_connect_data.ocd_connect_flags &=
2400 ~OBD_CONNECT_RDONLY;
2402 spin_unlock(&imp->imp_lock);
2404 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2405 keylen, key, vallen, val, set);
2408 if (KEY_IS(KEY_SPTLRPC_CONF)) {
2409 sptlrpc_conf_client_adapt(exp->exp_obd);
2412 if (KEY_IS(KEY_FLUSH_CTX)) {
2413 sptlrpc_import_flush_my_ctx(imp);
2416 if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
2417 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2418 keylen, key, vallen, val, set);
2421 if (KEY_IS(KEY_HSM_COPYTOOL_SEND)) {
2422 rc = mdc_hsm_copytool_send(vallen, val);
2426 if (KEY_IS(KEY_DEFAULT_EASIZE)) {
2427 __u32 *default_easize = val;
2429 exp->exp_obd->u.cli.cl_default_mds_easize = *default_easize;
2433 CERROR("Unknown key %s\n", (char *)key);
2437 static int mdc_get_info(const struct lu_env *env, struct obd_export *exp,
2438 __u32 keylen, void *key, __u32 *vallen, void *val)
2442 if (KEY_IS(KEY_MAX_EASIZE)) {
2443 __u32 mdsize, *max_easize;
2445 if (*vallen != sizeof(int))
2447 mdsize = *(__u32 *)val;
2448 if (mdsize > exp->exp_obd->u.cli.cl_max_mds_easize)
2449 exp->exp_obd->u.cli.cl_max_mds_easize = mdsize;
2451 *max_easize = exp->exp_obd->u.cli.cl_max_mds_easize;
2453 } else if (KEY_IS(KEY_DEFAULT_EASIZE)) {
2454 __u32 *default_easize;
2456 if (*vallen != sizeof(int))
2458 default_easize = val;
2459 *default_easize = exp->exp_obd->u.cli.cl_default_mds_easize;
2461 } else if (KEY_IS(KEY_CONN_DATA)) {
2462 struct obd_import *imp = class_exp2cliimp(exp);
2463 struct obd_connect_data *data = val;
2465 if (*vallen != sizeof(*data))
2468 *data = imp->imp_connect_data;
2470 } else if (KEY_IS(KEY_TGT_COUNT)) {
2471 *((__u32 *)val) = 1;
2475 rc = mdc_get_info_rpc(exp, keylen, key, *vallen, val);
2480 static int mdc_fsync(struct obd_export *exp, const struct lu_fid *fid,
2481 struct ptlrpc_request **request)
2483 struct ptlrpc_request *req;
2488 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_SYNC);
2492 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_SYNC);
2494 ptlrpc_request_free(req);
2498 mdc_pack_body(req, fid, 0, 0, -1, 0);
2500 ptlrpc_request_set_replen(req);
2502 rc = ptlrpc_queue_wait(req);
2504 ptlrpc_req_finished(req);
2510 static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
2511 enum obd_import_event event)
2515 LASSERT(imp->imp_obd == obd);
2518 case IMP_EVENT_DISCON: {
2520 /* XXX Pass event up to OBDs stack. used only for FLD now */
2521 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_DISCON, NULL);
2525 case IMP_EVENT_INACTIVE: {
2526 struct client_obd *cli = &obd->u.cli;
2528 * Flush current sequence to make client obtain new one
2529 * from server in case of disconnect/reconnect.
2531 if (cli->cl_seq != NULL)
2532 seq_client_flush(cli->cl_seq);
2534 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE, NULL);
2537 case IMP_EVENT_INVALIDATE: {
2538 struct ldlm_namespace *ns = obd->obd_namespace;
2540 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
2544 case IMP_EVENT_ACTIVE:
2545 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE, NULL);
2546 /* redo the kuc registration after reconnecting */
2548 rc = mdc_kuc_reregister(imp);
2551 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_OCD, NULL);
2553 case IMP_EVENT_DEACTIVATE:
2554 case IMP_EVENT_ACTIVATE:
2557 CERROR("Unknown import event %x\n", event);
2563 int mdc_fid_alloc(const struct lu_env *env, struct obd_export *exp,
2564 struct lu_fid *fid, struct md_op_data *op_data)
2566 struct client_obd *cli = &exp->exp_obd->u.cli;
2567 struct lu_client_seq *seq = cli->cl_seq;
2569 RETURN(seq_client_alloc_fid(env, seq, fid));
2572 static struct obd_uuid *mdc_get_uuid(struct obd_export *exp)
2574 struct client_obd *cli = &exp->exp_obd->u.cli;
2575 return &cli->cl_target_uuid;
2579 * Determine whether the lock can be canceled before replaying it during
2580 * recovery, non zero value will be return if the lock can be canceled,
2581 * or zero returned for not
2583 static int mdc_cancel_weight(struct ldlm_lock *lock)
2585 if (lock->l_resource->lr_type != LDLM_IBITS)
2588 /* FIXME: if we ever get into a situation where there are too many
2589 * opened files with open locks on a single node, then we really
2590 * should replay these open locks to reget it */
2591 if (lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_OPEN)
2597 static int mdc_resource_inode_free(struct ldlm_resource *res)
2599 if (res->lr_lvb_inode)
2600 res->lr_lvb_inode = NULL;
2605 static struct ldlm_valblock_ops inode_lvbo = {
2606 .lvbo_free = mdc_resource_inode_free
2609 static int mdc_llog_init(struct obd_device *obd)
2611 struct obd_llog_group *olg = &obd->obd_olg;
2612 struct llog_ctxt *ctxt;
2617 rc = llog_setup(NULL, obd, olg, LLOG_CHANGELOG_REPL_CTXT, obd,
2622 ctxt = llog_group_get_ctxt(olg, LLOG_CHANGELOG_REPL_CTXT);
2623 llog_initiator_connect(ctxt);
2624 llog_ctxt_put(ctxt);
2629 static void mdc_llog_finish(struct obd_device *obd)
2631 struct llog_ctxt *ctxt;
2635 ctxt = llog_get_context(obd, LLOG_CHANGELOG_REPL_CTXT);
2637 llog_cleanup(NULL, ctxt);
2642 static int mdc_setup(struct obd_device *obd, struct lustre_cfg *cfg)
2647 rc = ptlrpcd_addref();
2651 rc = client_obd_setup(obd, cfg);
2653 GOTO(err_ptlrpcd_decref, rc);
2654 #ifdef CONFIG_PROC_FS
2655 obd->obd_vars = lprocfs_mdc_obd_vars;
2656 lprocfs_obd_setup(obd);
2657 lprocfs_alloc_md_stats(obd, 0);
2659 sptlrpc_lprocfs_cliobd_attach(obd);
2660 ptlrpc_lprocfs_register_obd(obd);
2662 ns_register_cancel(obd->obd_namespace, mdc_cancel_weight);
2664 obd->obd_namespace->ns_lvbo = &inode_lvbo;
2666 rc = mdc_llog_init(obd);
2669 CERROR("failed to setup llogging subsystems\n");
2680 /* Initialize the default and maximum LOV EA sizes. This allows
2681 * us to make MDS RPCs with large enough reply buffers to hold a default
2682 * sized EA without having to calculate this (via a call into the
2683 * LOV + OSCs) each time we make an RPC. The maximum size is also tracked
2684 * but not used to avoid wastefully vmalloc()'ing large reply buffers when
2685 * a large number of stripes is possible. If a larger reply buffer is
2686 * required it will be reallocated in the ptlrpc layer due to overflow.
2688 static int mdc_init_ea_size(struct obd_export *exp, __u32 easize,
2691 struct obd_device *obd = exp->exp_obd;
2692 struct client_obd *cli = &obd->u.cli;
2695 if (cli->cl_max_mds_easize < easize)
2696 cli->cl_max_mds_easize = easize;
2698 if (cli->cl_default_mds_easize < def_easize)
2699 cli->cl_default_mds_easize = def_easize;
2704 static int mdc_precleanup(struct obd_device *obd)
2708 /* Failsafe, ok if racy */
2709 if (obd->obd_type->typ_refcnt <= 1)
2710 libcfs_kkuc_group_rem(0, KUC_GRP_HSM);
2712 obd_cleanup_client_import(obd);
2713 ptlrpc_lprocfs_unregister_obd(obd);
2714 lprocfs_obd_cleanup(obd);
2715 lprocfs_free_md_stats(obd);
2716 mdc_llog_finish(obd);
2720 static int mdc_cleanup(struct obd_device *obd)
2724 return client_obd_cleanup(obd);
2727 static int mdc_process_config(struct obd_device *obd, size_t len, void *buf)
2729 struct lustre_cfg *lcfg = buf;
2730 int rc = class_process_proc_param(PARAM_MDC, obd->obd_vars, lcfg, obd);
2731 return (rc > 0 ? 0: rc);
2735 /* get remote permission for current user on fid */
2736 static int mdc_get_remote_perm(struct obd_export *exp, const struct lu_fid *fid,
2737 u32 suppgid, struct ptlrpc_request **request)
2739 struct ptlrpc_request *req;
2743 LASSERT(client_is_remote(exp));
2746 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
2750 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
2752 ptlrpc_request_free(req);
2756 mdc_pack_body(req, fid, OBD_MD_FLRMTPERM, 0, suppgid, 0);
2758 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
2759 sizeof(struct mdt_remote_perm));
2761 ptlrpc_request_set_replen(req);
2763 rc = ptlrpc_queue_wait(req);
2765 ptlrpc_req_finished(req);
2771 static struct obd_ops mdc_obd_ops = {
2772 .o_owner = THIS_MODULE,
2773 .o_setup = mdc_setup,
2774 .o_precleanup = mdc_precleanup,
2775 .o_cleanup = mdc_cleanup,
2776 .o_add_conn = client_import_add_conn,
2777 .o_del_conn = client_import_del_conn,
2778 .o_connect = client_connect_import,
2779 .o_disconnect = client_disconnect_export,
2780 .o_iocontrol = mdc_iocontrol,
2781 .o_set_info_async = mdc_set_info_async,
2782 .o_statfs = mdc_statfs,
2783 .o_fid_init = client_fid_init,
2784 .o_fid_fini = client_fid_fini,
2785 .o_fid_alloc = mdc_fid_alloc,
2786 .o_import_event = mdc_import_event,
2787 .o_get_info = mdc_get_info,
2788 .o_process_config = mdc_process_config,
2789 .o_get_uuid = mdc_get_uuid,
2790 .o_quotactl = mdc_quotactl,
2793 static struct md_ops mdc_md_ops = {
2794 .m_get_root = mdc_get_root,
2795 .m_null_inode = mdc_null_inode,
2796 .m_close = mdc_close,
2797 .m_create = mdc_create,
2798 .m_enqueue = mdc_enqueue,
2799 .m_getattr = mdc_getattr,
2800 .m_getattr_name = mdc_getattr_name,
2801 .m_intent_lock = mdc_intent_lock,
2803 .m_rename = mdc_rename,
2804 .m_setattr = mdc_setattr,
2805 .m_setxattr = mdc_setxattr,
2806 .m_getxattr = mdc_getxattr,
2807 .m_fsync = mdc_fsync,
2808 .m_read_page = mdc_read_page,
2809 .m_unlink = mdc_unlink,
2810 .m_cancel_unused = mdc_cancel_unused,
2811 .m_init_ea_size = mdc_init_ea_size,
2812 .m_set_lock_data = mdc_set_lock_data,
2813 .m_lock_match = mdc_lock_match,
2814 .m_get_lustre_md = mdc_get_lustre_md,
2815 .m_free_lustre_md = mdc_free_lustre_md,
2816 .m_set_open_replay_data = mdc_set_open_replay_data,
2817 .m_clear_open_replay_data = mdc_clear_open_replay_data,
2818 .m_get_remote_perm = mdc_get_remote_perm,
2819 .m_intent_getattr_async = mdc_intent_getattr_async,
2820 .m_revalidate_lock = mdc_revalidate_lock
2823 static int __init mdc_init(void)
2825 return class_register_type(&mdc_obd_ops, &mdc_md_ops, true, NULL,
2826 LUSTRE_MDC_NAME, NULL);
2829 static void __exit mdc_exit(void)
2831 class_unregister_type(LUSTRE_MDC_NAME);
2834 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
2835 MODULE_DESCRIPTION("Lustre Metadata Client");
2836 MODULE_VERSION(LUSTRE_VERSION_STRING);
2837 MODULE_LICENSE("GPL");
2839 module_init(mdc_init);
2840 module_exit(mdc_exit);