1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Copyright (C) 2001-2004 Cluster File Systems, Inc.
6 * This file is part of Lustre, http://www.sf.net/projects/lustre/
8 * Lustre is free software; you can redistribute it and/or
9 * modify it under the terms of version 2 of the GNU General Public
10 * License as published by the Free Software Foundation.
12 * Lustre is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
17 * You should have received a copy of the GNU General Public License
18 * along with Lustre; if not, write to the Free Software
19 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
23 # define EXPORT_SYMTAB
25 #define DEBUG_SUBSYSTEM S_MDC
28 # include <linux/module.h>
29 # include <linux/pagemap.h>
30 # include <linux/miscdevice.h>
31 # include <linux/init.h>
33 # include <liblustre.h>
36 #include <linux/obd_class.h>
37 #include <linux/lustre_mds.h>
38 #include <linux/lustre_dlm.h>
39 #include <linux/lustre_sec.h>
40 #include <linux/lprocfs_status.h>
41 #include <linux/lustre_acl.h>
42 #include <linux/lustre_lite.h>
43 #include <linux/lustre_gs.h>
44 #include "mdc_internal.h"
46 #define REQUEST_MINOR 244
48 static int mdc_cleanup(struct obd_device *obd, int flags);
50 extern int mds_queue_req(struct ptlrpc_request *);
51 /* Helper that implements most of mdc_getstatus and signal_completed_replay. */
52 /* XXX this should become mdc_get_info("key"), sending MDS_GET_INFO RPC */
53 static int send_getstatus(struct obd_import *imp, struct lustre_id *rootid,
54 int level, int msg_flags)
56 struct ptlrpc_request *req;
57 struct mds_body *body;
58 int rc, size[2] = {0, sizeof(*body)};
61 //size[0] = lustre_secdesc_size();
63 req = ptlrpc_prep_req(imp, LUSTRE_MDS_VERSION, MDS_GETSTATUS,
66 GOTO(out, rc = -ENOMEM);
68 //lustre_pack_secdesc(req, size[0]);
70 body = lustre_msg_buf(req->rq_reqmsg, MDS_REQ_REC_OFF, sizeof (*body));
71 req->rq_send_state = level;
72 req->rq_replen = lustre_msg_size(1, &size[1]);
74 req->rq_reqmsg->flags |= msg_flags;
75 rc = ptlrpc_queue_wait(req);
78 body = lustre_swab_repbuf (req, 0, sizeof (*body),
79 lustre_swab_mds_body);
81 CERROR ("Can't extract mds_body\n");
82 GOTO (out, rc = -EPROTO);
85 memcpy(rootid, &body->id1, sizeof(*rootid));
87 CDEBUG(D_NET, "root ino="LPU64", last_committed="LPU64
88 ", last_xid="LPU64"\n", rootid->li_stc.u.e3s.l3s_ino,
89 req->rq_repmsg->last_committed, req->rq_repmsg->last_xid);
94 ptlrpc_req_finished(req);
98 /* This should be mdc_get_info("rootid") */
99 int mdc_getstatus(struct obd_export *exp, struct lustre_id *rootid)
101 return send_getstatus(class_exp2cliimp(exp), rootid,
106 mdc_interpret_getattr(struct ptlrpc_request *req, void *unused, int rc)
108 struct mds_body *body = NULL;
109 struct obd_capa *ocapa;
110 struct lustre_capa *capa = NULL;
111 unsigned long expiry;
118 DEBUG_REQ(rc == -ENOENT ? D_INFO : D_ERROR , req,
119 "async getattr failed: rc = %d", rc);
123 body = lustre_swab_repbuf(req, 0, sizeof (*body), lustre_swab_mds_body);
125 CERROR ("Can't unpack mds_body\n");
129 if (!(body->valid & OBD_MD_CAPA)) {
130 CDEBUG(D_INFO, "MDS has disabled capability\n");
134 capa = lustre_swab_repbuf(req, 1, sizeof(*capa),
135 lustre_swab_lustre_capa);
136 if (capa == NULL && rc != 0) {
137 CERROR ("Can't unpack lustre_capa\n");
141 ocapa = capa_renew(capa, CLIENT_CAPA);
145 spin_lock(&capa_lock);
146 expiry = expiry_to_jiffies(capa->lc_expiry - capa_pre_expiry(capa));
147 CDEBUG(D_INFO, "expiry %lu vs timer %lu, base %p\n",
148 expiry, ll_capa_timer.expires, ll_capa_timer.base);
149 if (time_before(expiry, ll_capa_timer.expires) ||
150 !timer_pending(&ll_capa_timer)) {
151 mod_timer(&ll_capa_timer, expiry);
152 CDEBUG(D_INFO, "ll_capa_timer new expiry: %lu\n", expiry);
154 spin_unlock(&capa_lock);
159 int mdc_getattr_async(struct obd_export *exp, struct ptlrpc_request *req)
161 int repsize[2] = {sizeof(struct mds_body), sizeof(struct lustre_capa)};
164 req->rq_replen = lustre_msg_size(2, repsize);
165 req->rq_interpret_reply = mdc_interpret_getattr;
166 ptlrpcd_add_req(req);
171 int mdc_getattr_common(struct obd_export *exp, unsigned int ea_size,
172 struct ptlrpc_request *req)
174 struct mds_body *body, *reqbody;
177 int repsize[2] = {sizeof(*body)};
181 /* request message already built */
184 repsize[bufcount++] = ea_size;
185 CDEBUG(D_INODE, "reserved %u bytes for MD/symlink in packet\n",
189 reqbody = lustre_msg_buf(req->rq_reqmsg, 1, sizeof(*reqbody));
190 LASSERT(!(reqbody->valid & OBD_MD_FLACL));
192 if (reqbody->valid & OBD_MD_FLKEY) {
193 repsize[bufcount++] = 5;
194 repsize[bufcount++] = sizeof(struct lustre_key);
195 } else if (reqbody->valid & OBD_MD_CAPA) {
196 LASSERT(ea_size == 0);
197 repsize[bufcount++] = sizeof(struct lustre_capa);
200 req->rq_replen = lustre_msg_size(bufcount, repsize);
202 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
203 rc = ptlrpc_queue_wait(req);
204 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
208 body = lustre_swab_repbuf (req, 0, sizeof (*body),
209 lustre_swab_mds_body);
211 CERROR ("Can't unpack mds_body\n");
215 CDEBUG(D_NET, "mode: %o\n", body->mode);
217 LASSERT_REPSWAB (req, 1);
219 /* Skip the check if getxattr/listxattr are called with no buffers */
220 if ((reqbody->eadatasize != 0) &&
221 !(reqbody->valid & (OBD_MD_FLXATTR | OBD_MD_FLXATTRLIST))) {
222 /* reply indicates presence of eadata; check it's there... */
223 eadata = lustre_msg_buf (req->rq_repmsg, 1,
225 if (eadata == NULL) {
226 CERROR ("Missing/short eadata\n");
234 static int mdc_cancel_unused(struct obd_export *exp,
235 struct lov_stripe_md *lsm,
236 int flags, void *opaque)
238 struct obd_device *obd = class_exp2obd(exp);
241 RETURN(ldlm_cli_cancel_unused(obd->obd_namespace,
242 NULL, flags, opaque));
245 int mdc_getattr(struct obd_export *exp, struct lustre_id *id,
246 __u64 valid, const char *xattr_name,
247 const void *xattr_data, unsigned int xattr_datalen,
248 unsigned int ea_size, struct obd_capa *ocapa,
249 struct ptlrpc_request **request)
251 struct ptlrpc_request *req;
252 struct mds_body *body;
253 int xattr_namelen = xattr_name ? strlen(xattr_name) + 1 : 0;
254 int size[4] = {0, sizeof(*body)};
259 size[0] = lustre_secdesc_size();
261 if (valid & OBD_MD_FLXATTR) {
262 size[bufcount++] = xattr_namelen;
264 if (xattr_datalen > 0) {
266 size[bufcount++] = xattr_datalen;
268 } else if (valid & OBD_MD_CAPA) {
269 LASSERT(valid == OBD_MD_CAPA);
271 size[bufcount++] = sizeof(*ocapa);
273 LASSERT(!xattr_data && !xattr_datalen);
276 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
277 MDS_GETATTR, bufcount, size, NULL);
279 GOTO(out, rc = -ENOMEM);
281 lustre_pack_secdesc(req, size[0]);
283 body = lustre_msg_buf(req->rq_reqmsg, MDS_REQ_REC_OFF, sizeof (*body));
284 memcpy(&body->id1, id, sizeof(*id));
286 body->eadatasize = ea_size;
288 if (valid & OBD_MD_FLXATTR) {
289 memcpy(lustre_msg_buf(req->rq_reqmsg, 2, xattr_namelen),
290 xattr_name, xattr_namelen);
292 memcpy(lustre_msg_buf(req->rq_reqmsg, 3, xattr_datalen),
293 xattr_data, xattr_datalen);
296 if (valid & OBD_MD_CAPA) {
297 /* renew capability */
298 memcpy(&body->handle, &ocapa->c_handle, sizeof(body->handle));
299 memcpy(lustre_msg_buf(req->rq_reqmsg, 2, sizeof(ocapa->c_capa)),
300 &ocapa->c_capa, sizeof(ocapa->c_capa));
302 rc = mdc_getattr_async(exp, req);
303 req = NULL; /* ptlrpcd will finish request */
305 rc = mdc_getattr_common(exp, ea_size, req);
307 ptlrpc_req_finished (req);
316 int mdc_access_check(struct obd_export *exp, struct lustre_id *id,
317 struct ptlrpc_request **request)
320 struct ptlrpc_request *req;
321 struct mds_body *body;
322 int size[2] = {0, sizeof(*body)};
326 size[0] = lustre_secdesc_size();
327 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
328 MDS_ACCESS_CHECK, 2, size, NULL);
330 GOTO(out, rc = -ENOMEM);
332 lustre_pack_secdesc(req, size[0]);
333 body = lustre_msg_buf(req->rq_reqmsg, MDS_REQ_REC_OFF, sizeof (*body));
334 memcpy(&body->id1, id, sizeof(*id));
336 size[0] = sizeof(*body);
337 size[1] = sizeof(struct mds_remote_perm);
338 req->rq_replen = lustre_msg_size(2, size);
340 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
341 rc = ptlrpc_queue_wait(req);
342 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
344 ptlrpc_req_finished (req);
347 body = lustre_swab_repbuf (req, 0, sizeof (*body),
348 lustre_swab_mds_body);
350 CERROR ("Can't unpack mds_body\n");
360 int mdc_getattr_lock(struct obd_export *exp, struct lustre_id *id,
361 char *filename, int namelen, __u64 valid,
362 unsigned int ea_size, struct ptlrpc_request **request)
364 struct ptlrpc_request *req;
365 struct mds_body *body;
366 int rc, size[3] = {0, sizeof(*body), namelen};
369 size[0] = lustre_secdesc_size();
371 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
372 MDS_GETATTR_LOCK, 3, size, NULL);
374 GOTO(out, rc = -ENOMEM);
376 lustre_pack_secdesc(req, size[0]);
378 body = lustre_msg_buf(req->rq_reqmsg, MDS_REQ_REC_OFF, sizeof (*body));
379 memcpy(&body->id1, id, sizeof(*id));
381 body->eadatasize = ea_size;
383 if (filename != NULL) {
384 LASSERT (strnlen (filename, namelen) == namelen - 1);
385 memcpy(lustre_msg_buf(req->rq_reqmsg, 2, namelen),
388 LASSERT(namelen == 1);
391 rc = mdc_getattr_common(exp, ea_size, req);
393 ptlrpc_req_finished (req);
401 /* This should be called with both the request and the reply still packed. */
402 int mdc_store_inode_generation(struct obd_export *exp,
403 struct ptlrpc_request *req,
404 int reqoff, int repoff)
406 struct mds_rec_create *rec =
407 lustre_msg_buf(req->rq_reqmsg, reqoff, sizeof(*rec));
408 struct mds_body *body =
409 lustre_msg_buf(req->rq_repmsg, repoff, sizeof(*body));
411 LASSERT (rec != NULL);
412 LASSERT (body != NULL);
414 memcpy(&rec->cr_replayid, &body->id1, sizeof(rec->cr_replayid));
415 DEBUG_REQ(D_HA, req, "storing generation for ino "DLID4,
416 OLID4(&rec->cr_replayid));
421 int mdc_unpack_acl(struct obd_export *exp_lmv, struct ptlrpc_request *req,
422 unsigned int offset, struct lustre_md *md)
424 struct posix_acl *acl;
425 struct mds_remote_perm *perm;
430 if (!(md->body->valid & OBD_MD_FLACL))
433 if (md->body->valid & OBD_MD_FLRMTACL) {
434 offset++; /* first 'size' is not used */
436 buf = lustre_swab_repbuf(req, offset, sizeof(*perm),
437 lustre_swab_remote_perm);
439 CERROR("Can't unpack remote perm\n");
443 OBD_ALLOC(perm, sizeof(*perm));
446 memcpy(perm, buf, sizeof(*perm));
447 md->remote_perm = perm;
449 size = le32_to_cpu(*(__u32 *) lustre_msg_buf(
450 req->rq_repmsg, offset, 4));
451 buf = lustre_msg_buf(req->rq_repmsg, offset + 1, size);
453 acl = posix_acl_from_xattr(buf, size);
456 CERROR("convert xattr to acl failed: %d\n", rc);
459 rc = posix_acl_valid(acl);
461 CERROR("acl valid error: %d\n", rc);
462 posix_acl_release(acl);
473 static int mdc_unpack_gskey(struct obd_export *exp_lmv, struct ptlrpc_request *req,
474 unsigned int *offset, struct lustre_md *md)
476 int key_off = 0, rc = 0, size = 0;
480 if (md->body->valid & OBD_MD_FLKEY) {
481 size = le32_to_cpu(*(__u32 *) lustre_msg_buf(req->rq_repmsg,
483 buf = lustre_msg_buf(req->rq_repmsg, key_off++, size);
485 CDEBUG(D_INFO, "buf %p key_off %d size %d \n",
487 md->key = (struct lustre_key *)buf;
495 int mdc_req2lustre_md(struct obd_export *exp_lmv, struct ptlrpc_request *req,
496 unsigned int offset, struct obd_export *exp_lov,
497 struct lustre_md *md)
499 struct lov_mds_md *lmm;
500 int rc = 0, reply_off;
504 memset(md, 0, sizeof(*md));
506 md->body = lustre_msg_buf(req->rq_repmsg, offset,
511 LASSERT_REPSWABBED(req, offset);
513 if (!(md->body->valid & OBD_MD_FLEASIZE) &&
514 !(md->body->valid & OBD_MD_FLDIREA))
517 if (S_ISREG(md->body->mode)) {
518 if (md->body->eadatasize == 0) {
519 CERROR("invalid EA size (0) is detected\n");
523 lmm = lustre_msg_buf(req->rq_repmsg, offset + 1,
524 md->body->eadatasize);
528 LASSERT(exp_lov != NULL);
530 rc = obd_unpackmd(exp_lov, &md->lsm, lmm,
531 md->body->eadatasize);
533 LASSERT(rc >= sizeof(*md->lsm));
536 } else if (S_ISDIR(md->body->mode)) {
537 /* dir can be non-splitted */
538 if (md->body->eadatasize == 0)
541 lmm = lustre_msg_buf(req->rq_repmsg, offset + 1,
542 md->body->eadatasize);
546 if (md->body->valid & OBD_MD_MEA) {
547 LASSERT(exp_lmv != NULL);
549 rc = obd_unpackmd(exp_lmv, (void *)&md->mea,
550 lmm, md->body->eadatasize);
552 LASSERT(rc >= sizeof(*md->mea));
557 LASSERT(S_ISCHR(md->body->mode) ||
558 S_ISBLK(md->body->mode) ||
559 S_ISFIFO(md->body->mode)||
560 S_ISLNK(md->body->mode) ||
561 S_ISSOCK(md->body->mode));
564 /* if anything wrong when unpacking md, we don't check acl
565 * stuff, for simplicity
570 reply_off = (md->body->valid & OBD_MD_FLEASIZE) ?
571 (offset + 2) : (offset + 1);
572 rc = mdc_unpack_acl(exp_lmv, req, reply_off, md);
574 CERROR("upack acl error %d \n", rc);
579 rc = mdc_unpack_gskey(exp_lmv, req, &reply_off, md);
586 static void mdc_commit_open(struct ptlrpc_request *req)
588 struct mdc_open_data *mod = req->rq_cb_data;
592 if (mod->mod_close_req != NULL)
593 mod->mod_close_req->rq_cb_data = NULL;
595 if (mod->mod_och != NULL)
596 mod->mod_och->och_mod = NULL;
598 OBD_FREE(mod, sizeof(*mod));
599 req->rq_cb_data = NULL;
600 LASSERT(atomic_read(&req->rq_refcount) > 1);
601 ptlrpc_req_finished(req);
604 static void mdc_replay_open(struct ptlrpc_request *req)
606 struct mdc_open_data *mod = req->rq_cb_data;
607 struct obd_client_handle *och;
608 struct ptlrpc_request *close_req;
609 struct lustre_handle old;
610 struct mds_body *body;
613 body = lustre_swab_repbuf(req, 1, sizeof(*body), lustre_swab_mds_body);
614 LASSERT (body != NULL);
617 if (body->valid & OBD_MD_CAPA) {
618 DEBUG_REQ(D_HA, req, "swap capa data from open");
619 lustre_swab_repbuf(req, 7, sizeof(struct lustre_capa),
620 lustre_swab_lustre_capa);
623 DEBUG_REQ(D_ERROR, req,
624 "can't properly replay without open data");
631 struct lustre_handle *file_fh;
632 LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
633 file_fh = &och->och_fh;
634 CDEBUG(D_HA, "updating handle from "LPX64" to "LPX64"\n",
635 file_fh->cookie, body->handle.cookie);
636 memcpy(&old, file_fh, sizeof(old));
637 memcpy(file_fh, &body->handle, sizeof(*file_fh));
640 close_req = mod->mod_close_req;
641 if (close_req != NULL) {
642 struct mds_body *close_body;
643 LASSERT(close_req->rq_reqmsg->opc == MDS_CLOSE);
644 close_body = lustre_msg_buf(close_req->rq_reqmsg,
646 sizeof(*close_body));
648 LASSERT(!memcmp(&old, &close_body->handle, sizeof old));
649 DEBUG_REQ(D_HA, close_req, "updating close body with new fh");
650 memcpy(&close_body->handle, &body->handle,
651 sizeof(close_body->handle));
657 int mdc_set_open_replay_data(struct obd_export *exp,
658 struct obd_client_handle *och,
659 struct ptlrpc_request *open_req)
661 struct mdc_open_data *mod;
662 struct mds_rec_create *rec;
663 struct mds_body *body;
665 rec = lustre_msg_buf(open_req->rq_reqmsg, MDS_REQ_INTENT_REC_OFF,
667 body = lustre_msg_buf(open_req->rq_repmsg, 1, sizeof(*body));
669 LASSERT(rec != NULL);
670 /* outgoing messages always in my byte order */
671 LASSERT(body != NULL);
672 /* incoming message in my byte order (it's been swabbed) */
673 LASSERT_REPSWABBED(open_req, 1);
675 OBD_ALLOC(mod, sizeof(*mod));
677 DEBUG_REQ(D_ERROR, open_req, "can't allocate mdc_open_data");
683 mod->mod_open_req = ptlrpc_request_addref(open_req);
685 if (memcmp(&rec->cr_replayid, &body->id1, sizeof(rec->cr_replayid)))
686 CDEBUG(D_ERROR, DLID4" != "DLID4"\n", OLID4(&rec->cr_replayid),
688 LASSERT(id_ino(&rec->cr_replayid) != 0);
689 memcpy(&rec->cr_replayid, &body->id1, sizeof rec->cr_replayid);
690 memcpy(&rec->cr_ioepoch, &body->io_epoch, sizeof rec->cr_ioepoch);
691 open_req->rq_replay_cb = mdc_replay_open;
692 open_req->rq_commit_cb = mdc_commit_open;
693 open_req->rq_cb_data = mod;
694 DEBUG_REQ(D_HA, open_req, "set up replay data");
698 int mdc_clear_open_replay_data(struct obd_export *exp,
699 struct obd_client_handle *och)
701 struct mdc_open_data *mod = och->och_mod;
703 /* Don't free the structure now (it happens in mdc_commit_open, after
704 * we're sure we won't need to fix up the close request in the future),
705 * but make sure that replay doesn't poke at the och, which is about to
707 LASSERT(mod != LP_POISON);
714 static void mdc_commit_close(struct ptlrpc_request *req)
716 struct mdc_open_data *mod = req->rq_cb_data;
717 struct obd_import *imp = req->rq_import;
718 struct ptlrpc_request *open_req;
720 DEBUG_REQ(D_HA, req, "close req committed");
724 mod->mod_close_req = NULL;
725 req->rq_cb_data = NULL;
726 req->rq_commit_cb = NULL;
728 open_req = mod->mod_open_req;
729 LASSERT(open_req != NULL);
730 LASSERT(open_req != LP_POISON);
731 LASSERT(open_req->rq_type != LI_POISON);
733 DEBUG_REQ(D_HA, open_req, "open req balanced");
734 if (open_req->rq_transno == 0) {
735 DEBUG_REQ(D_ERROR, open_req, "BUG 3892 open");
736 DEBUG_REQ(D_ERROR, req, "BUG 3892 close");
737 LASSERTF(open_req->rq_transno != 0, "BUG 3892\n");
739 LASSERT(open_req->rq_import == imp);
741 /* We no longer want to preserve this for transno-unconditional
743 spin_lock(&open_req->rq_lock);
744 open_req->rq_replay = 0;
745 spin_unlock(&open_req->rq_lock);
748 int mdc_close(struct obd_export *exp, struct mdc_op_data *op_data,
749 struct obd_client_handle *och, struct ptlrpc_request **request)
751 struct obd_device *obd = class_exp2obd(exp);
752 struct obd_import *imp = class_exp2cliimp(exp);
753 int reqsize[3] = {0, sizeof(struct mds_body),
754 obd->u.cli.cl_max_mds_cookiesize};
755 int rc, repsize[3] = {sizeof(struct mds_body),
756 obd->u.cli.cl_max_mds_easize,
757 obd->u.cli.cl_max_mds_cookiesize};
758 struct ptlrpc_request *req;
759 struct mdc_open_data *mod;
762 if (imp->imp_connection == NULL) {
763 CERROR("request on not connected import %s\n",
764 imp->imp_obd->obd_name);
768 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
769 MDS_CLOSE, 3, reqsize, NULL);
771 GOTO(out, rc = -ENOMEM);
772 //req->rq_request_portal = MDS_CLOSE_PORTAL;
774 /* ensure that this close's handle is fixed up during replay. */
775 LASSERT(och != NULL);
777 if (likely(mod != NULL)) {
778 mod->mod_close_req = req;
779 LASSERT(mod->mod_open_req->rq_type != LI_POISON);
780 DEBUG_REQ(D_HA, mod->mod_open_req, "matched open");
782 CDEBUG(D_HA, "couldn't find open req; "
783 "expecting close error\n");
786 mdc_close_pack(req, 1, op_data, och);
788 req->rq_replen = lustre_msg_size(3, repsize);
789 req->rq_commit_cb = mdc_commit_close;
790 LASSERT(req->rq_cb_data == NULL);
791 req->rq_cb_data = mod;
793 mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
794 rc = ptlrpc_queue_wait(req);
795 mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
797 if (req->rq_repmsg == NULL) {
798 CDEBUG(D_HA, "request failed to send: %p, %d\n", req,
801 rc = req->rq_status ? req->rq_status : -EIO;
802 } else if (rc == 0) {
803 rc = req->rq_repmsg->status;
804 if (req->rq_repmsg->type == PTL_RPC_MSG_ERR) {
805 DEBUG_REQ(D_ERROR, req, "type == PTL_RPC_MSG_ERR, "
811 CERROR("Unexpected: can't find mdc_open_data, but "
812 "close succeeded. Please tell CFS.\n");
813 if (!lustre_swab_repbuf(req, 0, sizeof(struct mds_body),
814 lustre_swab_mds_body))
816 CERROR("Error unpacking mds_body\n");
828 int mdc_done_writing(struct obd_export *exp, struct obdo *obdo)
830 struct ptlrpc_request *req;
831 struct mds_body *body;
832 int rc, size[2] = {0, sizeof(*body)};
835 size[0] = lustre_secdesc_size();
837 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
838 MDS_DONE_WRITING, 2, size, NULL);
842 lustre_pack_secdesc(req, size[0]);
844 body = lustre_msg_buf(req->rq_reqmsg, MDS_REQ_REC_OFF,
847 mdc_pack_id(&body->id1, obdo->o_id, 0, obdo->o_mode,
848 obdo->o_mds, obdo->o_fid);
850 body->size = obdo->o_size;
851 body->blocks = obdo->o_blocks;
852 body->flags = obdo->o_flags;
853 body->valid = obdo->o_valid;
855 req->rq_replen = lustre_msg_size(1, &size[1]);
857 rc = ptlrpc_queue_wait(req);
858 ptlrpc_req_finished(req);
862 int mdc_readpage(struct obd_export *exp,
863 struct lustre_id *id,
864 __u64 offset, struct page *page,
865 struct ptlrpc_request **request)
867 struct obd_import *imp = class_exp2cliimp(exp);
868 struct ptlrpc_request *req = NULL;
869 struct ptlrpc_bulk_desc *desc = NULL;
870 struct mds_body *body;
871 int rc, size[2] = {0, sizeof(*body)};
874 CDEBUG(D_INODE, "inode: %ld\n", (long)id->li_stc.u.e3s.l3s_ino);
876 size[0] = lustre_secdesc_size();
878 req = ptlrpc_prep_req(imp, LUSTRE_MDS_VERSION, MDS_READPAGE,
881 GOTO(out, rc = -ENOMEM);
882 /* XXX FIXME bug 249 */
883 req->rq_request_portal = MDS_READPAGE_PORTAL;
885 lustre_pack_secdesc(req, size[0]);
887 desc = ptlrpc_prep_bulk_imp(req, 1, BULK_PUT_SINK, MDS_BULK_PORTAL);
889 GOTO(out, rc = -ENOMEM);
890 /* NB req now owns desc and will free it when it gets freed */
892 ptlrpc_prep_bulk_page(desc, page, 0, PAGE_CACHE_SIZE);
893 mdc_readdir_pack(req, 1, offset, PAGE_CACHE_SIZE, id);
895 req->rq_replen = lustre_msg_size(1, &size[1]);
896 rc = ptlrpc_queue_wait(req);
899 body = lustre_swab_repbuf(req, 0, sizeof (*body),
900 lustre_swab_mds_body);
902 CERROR("Can't unpack mds_body\n");
903 GOTO(out, rc = -EPROTO);
906 if (req->rq_bulk->bd_nob_transferred != PAGE_CACHE_SIZE) {
907 CERROR ("Unexpected # bytes transferred: %d"
909 req->rq_bulk->bd_nob_transferred,
911 GOTO (out, rc = -EPROTO);
921 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
922 void *karg, void *uarg)
924 struct obd_device *obd = exp->exp_obd;
925 struct obd_ioctl_data *data = karg;
926 struct obd_import *imp = obd->u.cli.cl_import;
927 struct llog_ctxt *ctxt;
931 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
934 if (!try_module_get(THIS_MODULE)) {
935 CERROR("Can't get module. Is it alive?");
940 case OBD_IOC_CLIENT_RECOVER:
941 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1);
945 case IOC_OSC_SET_ACTIVE:
946 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
948 case IOC_OSC_CTL_RECOVERY:
949 rc = ptlrpc_import_control_recovery(imp, data->ioc_offset);
951 case OBD_IOC_PARSE: {
952 ctxt = llog_get_context(&exp->exp_obd->obd_llogs,
953 LLOG_CONFIG_REPL_CTXT);
954 rc = class_config_process_llog(ctxt, data->ioc_inlbuf1, NULL);
958 case OBD_IOC_LLOG_INFO:
959 case OBD_IOC_LLOG_PRINT: {
960 ctxt = llog_get_context(&obd->obd_llogs, LLOG_CONFIG_REPL_CTXT);
961 rc = llog_ioctl(ctxt, cmd, data);
967 CERROR("mdc_ioctl(): unrecognised ioctl %#x\n", cmd);
968 GOTO(out, rc = -ENOTTY);
971 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
974 module_put(THIS_MODULE);
980 int mdc_set_info(struct obd_export *exp, obd_count keylen,
981 void *key, obd_count vallen, void *val)
985 if (keylen == strlen("initial_recov") &&
986 memcmp(key, "initial_recov", strlen("initial_recov")) == 0) {
987 struct obd_import *imp = exp->exp_obd->u.cli.cl_import;
988 if (vallen != sizeof(int))
990 imp->imp_initial_recov = *(int *)val;
991 CDEBUG(D_HA, "%s: set imp_no_init_recov = %d\n",
992 exp->exp_obd->obd_name,
993 imp->imp_initial_recov);
995 } else if ((keylen >= strlen("crypto_type")) &&
996 strcmp(key, "crypto_type") == 0) {
997 struct ptlrpc_request *req;
998 char *bufs[2] = {key, val};
999 int rc, size[2] = {keylen, vallen};
1001 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_OBD_VERSION,
1002 OST_SET_INFO, 2, size, bufs);
1006 req->rq_replen = lustre_msg_size(0, NULL);
1007 rc = ptlrpc_queue_wait(req);
1008 ptlrpc_req_finished(req);
1010 } else if (keylen >= strlen("inter_mds") && strcmp(key, "inter_mds") == 0) {
1011 struct obd_import *imp = class_exp2cliimp(exp);
1012 imp->imp_server_timeout = 1;
1013 CDEBUG(D_OTHER, "%s: timeout / 2\n", exp->exp_obd->obd_name);
1015 } else if (keylen == strlen("sec") &&
1016 memcmp(key, "sec", keylen) == 0) {
1017 struct client_obd *cli = &exp->exp_obd->u.cli;
1019 cli->cl_sec_flavor = ptlrpcs_name2flavor(val);
1020 if (cli->cl_sec_flavor == PTLRPCS_FLVR_INVALID) {
1021 CERROR("unrecognized security type %s\n", (char*) val);
1026 } else if (keylen == strlen("sec_flags") &&
1027 memcmp(key, "sec_flags", keylen) == 0) {
1028 struct client_obd *cli = &exp->exp_obd->u.cli;
1030 cli->cl_sec_flags = *((unsigned long *) val);
1032 } else if (keylen == strlen("flush_cred") &&
1033 memcmp(key, "flush_cred", keylen) == 0) {
1034 struct client_obd *cli = &exp->exp_obd->u.cli;
1037 ptlrpcs_import_flush_current_creds(cli->cl_import);
1039 } else if (keylen == strlen("async") && memcmp(key, "async", keylen) == 0) {
1040 struct client_obd *cl = &exp->exp_obd->u.cli;
1041 if (vallen != sizeof(int))
1043 cl->cl_async = *(int *)val;
1044 CDEBUG(D_HA, "%s: set async = %d\n",
1045 exp->exp_obd->obd_name, cl->cl_async);
1047 } else if (keylen == strlen("setext") && memcmp(key, "setext", keylen) == 0) {
1048 struct ptlrpc_request *req;
1049 char *bufs[2] = {key, val};
1050 int rc, size[2] = {keylen, vallen};
1052 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_OBD_VERSION,
1053 OST_SET_INFO, 2, size, bufs);
1057 req->rq_replen = lustre_msg_size(0, NULL);
1058 rc = ptlrpc_queue_wait(req);
1059 ptlrpc_req_finished(req);
1061 } else if (keylen == 5 && strcmp(key, "audit") == 0) {
1062 struct ptlrpc_request *req;
1063 char *bufs[2] = {key, val};
1064 int rc, size[2] = {keylen, vallen};
1066 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_OBD_VERSION,
1067 OST_SET_INFO, 2, size, bufs);
1071 req->rq_replen = lustre_msg_size(0, NULL);
1072 lustre_swab_reqbuf(req, 1, sizeof(struct audit_attr_msg),
1073 lustre_swab_audit_attr);
1074 rc = ptlrpc_queue_wait(req);
1075 ptlrpc_req_finished(req);
1078 } else if (keylen == strlen("ids") && memcmp(key, "ids", keylen) == 0) {
1079 struct ptlrpc_request *req;
1080 struct lustre_id *ids = (struct lustre_id *)val;
1081 char *bufs[3] = {key, (char *)ids, (char *)(ids + 1)};
1082 int rc, size[3] = {keylen, sizeof(struct lustre_id),
1083 sizeof(struct lustre_id)};
1085 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_OBD_VERSION,
1086 OST_SET_INFO, 3, size, bufs);
1090 req->rq_replen = lustre_msg_size(0, NULL);
1091 rc = ptlrpc_queue_wait(req);
1092 ptlrpc_req_finished(req);
1098 static int mdc_statfs(struct obd_device *obd, struct obd_statfs *osfs,
1099 unsigned long max_age)
1101 struct obd_statfs *msfs;
1102 struct ptlrpc_request *req;
1103 int rc, size = sizeof(*msfs);
1106 /* We could possibly pass max_age in the request (as an absolute
1107 * timestamp or a "seconds.usec ago") so the target can avoid doing
1108 * extra calls into the filesystem if that isn't necessary (e.g.
1109 * during mount that would help a bit). Having relative timestamps
1110 * is not so great if request processing is slow, while absolute
1111 * timestamps are not ideal because they need time synchronization. */
1112 req = ptlrpc_prep_req(obd->u.cli.cl_import, LUSTRE_MDS_VERSION,
1113 MDS_STATFS, 0, NULL, NULL);
1117 req->rq_replen = lustre_msg_size(1, &size);
1119 mdc_get_rpc_lock(obd->u.cli.cl_rpc_lock, NULL);
1120 rc = ptlrpc_queue_wait(req);
1121 mdc_put_rpc_lock(obd->u.cli.cl_rpc_lock, NULL);
1124 /* this can be LMV fake import, whcih is not connected. */
1125 if (!req->rq_import->imp_connection)
1126 memset(osfs, 0, sizeof(*osfs));
1130 msfs = lustre_swab_repbuf(req, 0, sizeof(*msfs),
1131 lustre_swab_obd_statfs);
1133 CERROR("Can't unpack obd_statfs\n");
1134 GOTO(out, rc = -EPROTO);
1137 memcpy(osfs, msfs, sizeof (*msfs));
1140 ptlrpc_req_finished(req);
1144 static int mdc_pin(struct obd_export *exp, obd_id ino, __u32 gen, int type,
1145 struct obd_client_handle *handle, int flag)
1147 struct ptlrpc_request *req;
1148 struct mds_body *body;
1149 int rc, size[2] = {0, sizeof(*body)};
1152 //size[0] = lustre_secdesc_size();
1154 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
1155 MDS_PIN, 2, size, NULL);
1159 //lustre_pack_secdesc(req, size[0]);
1161 body = lustre_msg_buf(req->rq_reqmsg,
1162 MDS_REQ_REC_OFF, sizeof(*body));
1164 /* FIXME-UMKA: here should be also mdsnum and fid. */
1165 mdc_pack_id(&body->id1, ino, gen, type, 0, 0);
1168 req->rq_replen = lustre_msg_size(1, &size[1]);
1170 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1171 rc = ptlrpc_queue_wait(req);
1172 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1174 CERROR("pin failed: %d\n", rc);
1175 ptlrpc_req_finished(req);
1179 body = lustre_swab_repbuf(req, 0, sizeof(*body), lustre_swab_mds_body);
1181 ptlrpc_req_finished(req);
1185 memcpy(&handle->och_fh, &body->handle, sizeof(body->handle));
1186 handle->och_magic = OBD_CLIENT_HANDLE_MAGIC;
1188 OBD_ALLOC(handle->och_mod, sizeof(*handle->och_mod));
1189 if (handle->och_mod == NULL) {
1190 DEBUG_REQ(D_ERROR, req, "can't allocate mdc_open_data");
1193 handle->och_mod->mod_open_req = req; /* will be dropped by unpin */
1198 static int mdc_unpin(struct obd_export *exp,
1199 struct obd_client_handle *handle, int flag)
1201 struct ptlrpc_request *req;
1202 struct mds_body *body;
1203 int rc, size[2] = {0, sizeof(*body)};
1206 if (handle->och_magic != OBD_CLIENT_HANDLE_MAGIC)
1209 //size[0] = lustre_secdesc_size();
1211 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
1212 MDS_CLOSE, 2, size, NULL);
1216 //lustre_pack_secdesc(req, size[0]);
1218 body = lustre_msg_buf(req->rq_reqmsg, MDS_REQ_REC_OFF, sizeof(*body));
1219 memcpy(&body->handle, &handle->och_fh, sizeof(body->handle));
1222 req->rq_replen = lustre_msg_size(0, NULL);
1223 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1224 rc = ptlrpc_queue_wait(req);
1225 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1228 CERROR("unpin failed: %d\n", rc);
1230 ptlrpc_req_finished(req);
1231 ptlrpc_req_finished(handle->och_mod->mod_open_req);
1232 OBD_FREE(handle->och_mod, sizeof(*handle->och_mod));
1236 int mdc_sync(struct obd_export *exp, struct lustre_id *id,
1237 struct ptlrpc_request **request)
1239 struct ptlrpc_request *req;
1240 struct mds_body *body;
1241 int size[2] = {0, sizeof(*body)};
1245 //size[0] = lustre_secdesc_size();
1247 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
1248 MDS_SYNC, 2, size, NULL);
1250 RETURN(rc = -ENOMEM);
1252 //lustre_pack_secdesc(req, size[0]);
1255 body = lustre_msg_buf(req->rq_reqmsg, MDS_REQ_REC_OFF,
1257 memcpy(&body->id1, id, sizeof(*id));
1260 req->rq_replen = lustre_msg_size(1, &size[1]);
1262 rc = ptlrpc_queue_wait(req);
1263 if (rc || request == NULL)
1264 ptlrpc_req_finished(req);
1271 static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
1272 enum obd_import_event event)
1276 LASSERT(imp->imp_obd == obd);
1279 case IMP_EVENT_DISCON: {
1282 case IMP_EVENT_INACTIVE: {
1283 if (obd->obd_observer)
1284 rc = obd_notify(obd->obd_observer, obd, 0, 0);
1287 case IMP_EVENT_INVALIDATE: {
1288 struct ldlm_namespace *ns = obd->obd_namespace;
1290 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
1294 case IMP_EVENT_ACTIVE: {
1295 if (obd->obd_observer)
1296 rc = obd_notify(obd->obd_observer, obd, 1, 0);
1300 CERROR("Unknown import event %d\n", event);
1306 static int mdc_attach(struct obd_device *dev, obd_count len, void *data)
1308 struct lprocfs_static_vars lvars;
1310 lprocfs_init_vars(mdc, &lvars);
1311 return lprocfs_obd_attach(dev, lvars.obd_vars);
1314 static int mdc_detach(struct obd_device *dev)
1316 return lprocfs_obd_detach(dev);
1319 static int mdc_setup(struct obd_device *obd, obd_count len, void *buf)
1321 struct client_obd *cli = &obd->u.cli;
1325 OBD_ALLOC(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1326 if (!cli->cl_rpc_lock)
1328 mdc_init_rpc_lock(cli->cl_rpc_lock);
1332 OBD_ALLOC(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1333 if (!cli->cl_setattr_lock)
1334 GOTO(err_rpc_lock, rc = -ENOMEM);
1335 mdc_init_rpc_lock(cli->cl_setattr_lock);
1337 OBD_ALLOC(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1338 if (!cli->cl_close_lock)
1339 GOTO(err_setattr_lock, rc = -ENOMEM);
1340 mdc_init_rpc_lock(cli->cl_close_lock);
1342 rc = client_obd_setup(obd, len, buf);
1344 GOTO(err_close_lock, rc);
1346 rc = obd_llog_init(obd, &obd->obd_llogs, obd, 0, NULL);
1348 mdc_cleanup(obd, 0);
1349 CERROR("failed to setup llogging subsystems\n");
1355 OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1357 OBD_FREE(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1359 OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1364 static int mdc_init_ea_size(struct obd_export *exp, int easize, int cookiesize)
1366 struct obd_device *obd = exp->exp_obd;
1367 struct client_obd *cli = &obd->u.cli;
1370 if (cli->cl_max_mds_easize < easize)
1371 cli->cl_max_mds_easize = easize;
1372 if (cli->cl_max_mds_cookiesize < cookiesize)
1373 cli->cl_max_mds_cookiesize = cookiesize;
1377 static int mdc_precleanup(struct obd_device *obd, int flags)
1381 rc = obd_llog_finish(obd, &obd->obd_llogs, 0);
1383 CERROR("failed to cleanup llogging subsystems\n");
1388 static int mdc_cleanup(struct obd_device *obd, int flags)
1390 struct client_obd *cli = &obd->u.cli;
1392 OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1393 OBD_FREE(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1394 OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1398 return client_obd_cleanup(obd, flags);
1402 static int mdc_llog_init(struct obd_device *obd, struct obd_llogs *llogs,
1403 struct obd_device *tgt, int count,
1404 struct llog_catid *logid)
1406 struct llog_ctxt *ctxt;
1410 rc = obd_llog_setup(obd, llogs, LLOG_CONFIG_REPL_CTXT, tgt, 0, NULL,
1413 ctxt = llog_get_context(llogs, LLOG_CONFIG_REPL_CTXT);
1414 ctxt->loc_imp = obd->u.cli.cl_import;
1420 static int mdc_llog_finish(struct obd_device *obd,
1421 struct obd_llogs *llogs, int count)
1426 rc = obd_llog_cleanup(llog_get_context(llogs, LLOG_CONFIG_REPL_CTXT));
1430 static struct obd_device *mdc_get_real_obd(struct obd_export *exp,
1431 struct lustre_id *id)
1434 RETURN(exp->exp_obd);
1437 static int mdc_get_info(struct obd_export *exp, __u32 keylen,
1438 void *key, __u32 *valsize, void *val)
1440 struct ptlrpc_request *req;
1441 char *bufs[1] = {key};
1445 if (!valsize || !val)
1448 if (keylen >= strlen("remote_flag") && !strcmp(key, "remote_flag")) {
1449 struct obd_import *imp;
1450 struct obd_connect_data *data;
1452 imp = class_exp2cliimp(exp);
1458 if (imp->imp_state != LUSTRE_IMP_FULL) {
1459 CERROR("import state not full\n");
1463 data = &imp->imp_connect_data;
1464 if (data->ocd_connect_flags & OBD_CONNECT_REMOTE) {
1467 } else if (data->ocd_connect_flags & OBD_CONNECT_LOCAL) {
1471 CERROR("no remote flag set?\n");
1475 if ((keylen < strlen("mdsize") || strcmp(key, "mdsize") != 0) &&
1476 (keylen < strlen("mdsnum") || strcmp(key, "mdsnum") != 0) &&
1477 (keylen < strlen("lovdesc") || strcmp(key, "lovdesc") != 0) &&
1478 (keylen < strlen("getext") || strcmp(key, "getext") != 0) &&
1479 (keylen < strlen("rootid") || strcmp(key, "rootid") != 0) &&
1480 (keylen < strlen("auditid") || strcmp(key, "auditid") != 0))
1483 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_OBD_VERSION,
1484 OST_GET_INFO, 1, (int *)&keylen, bufs);
1488 req->rq_replen = lustre_msg_size(1, (int *)valsize);
1489 rc = ptlrpc_queue_wait(req);
1493 if ((keylen >= strlen("rootid") && !strcmp(key, "rootid")) ||
1494 (keylen >= strlen("auditid") && !strcmp(key, "auditid"))) {
1495 struct lustre_id *reply;
1497 reply = lustre_swab_repbuf(req, 0, sizeof(*reply),
1498 lustre_swab_lustre_id);
1499 if (reply == NULL) {
1500 CERROR("Can't unpack %s\n", (char *)key);
1501 GOTO(out_req, rc = -EPROTO);
1504 *(struct lustre_id *)val = *reply;
1505 } else if (keylen >= strlen("lovdesc") && !strcmp(key, "lovdesc")) {
1506 struct lov_desc *reply;
1508 reply = lustre_swab_repbuf(req, 0, sizeof(*reply),
1509 lustre_swab_lov_desc);
1510 if (reply == NULL) {
1511 CERROR("Can't unpack %s\n", (char *)key);
1512 GOTO(out_req, rc = -EPROTO);
1515 *(struct lov_desc *)val = *reply;
1517 } else if (keylen >= strlen("getext") && !strcmp(key, "getext")) {
1518 struct fid_extent *reply;
1520 reply = lustre_swab_repbuf(req, 0, sizeof(*reply),
1521 lustre_swab_fid_extent);
1522 if (reply == NULL) {
1523 CERROR("Can't unpack %s\n", (char *)key);
1524 GOTO(out_req, rc = -EPROTO);
1527 *(struct fid_extent *)val = *reply;
1532 reply = lustre_swab_repbuf(req, 0, sizeof(*reply),
1533 lustre_swab_generic_32s);
1534 if (reply == NULL) {
1535 CERROR("Can't unpack %s\n", (char *)key);
1536 GOTO(out_req, rc = -EPROTO);
1538 *((__u32 *)val) = *reply;
1541 ptlrpc_req_finished(req);
1545 int mdc_obj_create(struct obd_export *exp, struct obdo *oa,
1546 void *acl, int acl_size, struct lov_stripe_md **ea,
1547 struct obd_trans_info *oti)
1549 struct ptlrpc_request *request;
1550 struct ost_body *body;
1552 int rc, size[2] = { sizeof(*body), acl_size };
1557 request = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_OBD_VERSION,
1558 OST_CREATE, 2, size, NULL);
1560 GOTO(out_req, rc = -ENOMEM);
1562 body = lustre_msg_buf(request->rq_reqmsg, 0, sizeof (*body));
1563 memcpy(&body->oa, oa, sizeof(body->oa));
1566 acl_buf = lustre_msg_buf(request->rq_reqmsg, 1, acl_size);
1567 memcpy(acl_buf, acl, acl_size);
1570 request->rq_replen = lustre_msg_size(1, size);
1571 rc = ptlrpc_queue_wait(request);
1575 body = lustre_swab_repbuf(request, 0, sizeof(*body),
1576 lustre_swab_ost_body);
1578 CERROR ("can't unpack ost_body\n");
1579 GOTO (out_req, rc = -EPROTO);
1582 memcpy(oa, &body->oa, sizeof(*oa));
1584 /* store ino/generation for recovery */
1585 body = lustre_msg_buf(request->rq_reqmsg, 0, sizeof (*body));
1586 body->oa.o_id = oa->o_id;
1587 body->oa.o_generation = oa->o_generation;
1588 body->oa.o_fid = oa->o_fid;
1589 body->oa.o_mds = oa->o_mds;
1591 CDEBUG(D_HA, "transno: "LPD64"\n", request->rq_repmsg->transno);
1594 ptlrpc_req_finished(request);
1598 int mdc_brw(int rw, struct obd_export *exp, struct obdo *oa,
1599 struct lov_stripe_md *ea, obd_count oa_bufs,
1600 struct brw_page *pgarr, struct obd_trans_info *oti)
1602 struct ptlrpc_bulk_desc *desc;
1603 struct niobuf_remote *niobuf;
1604 struct ptlrpc_request *req;
1605 struct obd_ioobj *ioobj;
1606 struct ost_body *body;
1610 opc = ((rw & OBD_BRW_WRITE) != 0) ? OST_WRITE : OST_READ;
1612 size[0] = sizeof(*body);
1613 size[1] = sizeof(*ioobj);
1614 size[2] = oa_bufs * sizeof(*niobuf);
1616 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_OBD_VERSION, opc,
1618 LASSERT(req != NULL);
1620 if (opc == OST_WRITE)
1621 desc = ptlrpc_prep_bulk_imp(req, oa_bufs, BULK_GET_SOURCE,
1624 desc = ptlrpc_prep_bulk_imp(req, oa_bufs, BULK_PUT_SINK,
1626 LASSERT(desc != NULL);
1628 body = lustre_msg_buf(req->rq_reqmsg, 0, sizeof(*body));
1629 ioobj = lustre_msg_buf(req->rq_reqmsg, 1, sizeof(*ioobj));
1630 niobuf = lustre_msg_buf(req->rq_reqmsg, 2, oa_bufs * sizeof(*niobuf));
1632 memcpy(&body->oa, oa, sizeof(*oa));
1633 obdo_to_ioobj(oa, ioobj);
1634 ioobj->ioo_bufcnt = oa_bufs;
1636 for (i = 0; i < oa_bufs; i++, niobuf++) {
1637 struct brw_page *pg = &pgarr[i];
1639 LASSERT(pg->count > 0);
1640 LASSERT((pg->disk_offset & ~PAGE_MASK) + pg->count <= PAGE_SIZE);
1642 ptlrpc_prep_bulk_page(desc, pg->pg, pg->disk_offset & ~PAGE_MASK,
1645 niobuf->offset = pg->disk_offset;
1646 niobuf->len = pg->count;
1647 niobuf->flags = pg->flag;
1650 /* size[0] still sizeof (*body) */
1651 if (opc == OST_WRITE) {
1652 /* 1 RC per niobuf */
1653 size[1] = sizeof(__u32) * oa_bufs;
1654 req->rq_replen = lustre_msg_size(2, size);
1656 /* 1 RC for the whole I/O */
1657 req->rq_replen = lustre_msg_size(1, size);
1659 err = ptlrpc_queue_wait(req);
1662 ptlrpc_req_finished(req);
1666 static int mdc_valid_attrs(struct obd_export *exp,
1667 struct lustre_id *id)
1669 struct ldlm_res_id res_id = { .name = {0} };
1670 struct obd_device *obd = exp->exp_obd;
1671 struct lustre_handle lockh;
1672 ldlm_policy_data_t policy;
1676 res_id.name[0] = id_fid(id);
1677 res_id.name[1] = id_group(id);
1678 policy.l_inodebits.bits = MDS_INODELOCK_UPDATE;
1680 CDEBUG(D_INFO, "trying to match res "LPU64"\n",
1683 /* FIXME use LDLM_FL_TEST_LOCK instead */
1684 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING;
1685 if (ldlm_lock_match(obd->obd_namespace, flags, &res_id,
1686 LDLM_IBITS, &policy, LCK_PR, &lockh)) {
1687 ldlm_lock_decref(&lockh, LCK_PR);
1691 if (ldlm_lock_match(obd->obd_namespace, flags, &res_id,
1692 LDLM_IBITS, &policy, LCK_PW, &lockh)) {
1693 ldlm_lock_decref(&lockh, LCK_PW);
1699 static int mdc_change_cbdata_name(struct obd_export *exp,
1700 struct lustre_id *pid,
1701 char *name, int len,
1702 struct lustre_id *cid,
1703 ldlm_iterator_t it, void *data)
1706 rc = mdc_change_cbdata(exp, cid, it, data);
1710 struct obd_ops mdc_obd_ops = {
1711 .o_owner = THIS_MODULE,
1712 .o_attach = mdc_attach,
1713 .o_detach = mdc_detach,
1714 .o_setup = mdc_setup,
1715 .o_precleanup = mdc_precleanup,
1716 .o_cleanup = mdc_cleanup,
1717 .o_add_conn = client_import_add_conn,
1718 .o_del_conn = client_import_del_conn,
1719 .o_connect = client_connect_import,
1720 .o_disconnect = client_disconnect_export,
1721 .o_iocontrol = mdc_iocontrol,
1722 .o_packmd = mdc_packmd,
1723 .o_unpackmd = mdc_unpackmd,
1724 .o_statfs = mdc_statfs,
1726 .o_unpin = mdc_unpin,
1727 .o_import_event = mdc_import_event,
1728 .o_llog_init = mdc_llog_init,
1729 .o_llog_finish = mdc_llog_finish,
1730 .o_create = mdc_obj_create,
1731 .o_set_info = mdc_set_info,
1732 .o_get_info = mdc_get_info,
1734 .o_cancel_unused = mdc_cancel_unused,
1735 .o_init_ea_size = mdc_init_ea_size,
1738 struct md_ops mdc_md_ops = {
1739 .m_getstatus = mdc_getstatus,
1740 .m_getattr = mdc_getattr,
1741 .m_close = mdc_close,
1742 .m_create = mdc_create,
1743 .m_done_writing = mdc_done_writing,
1744 .m_enqueue = mdc_enqueue,
1745 .m_getattr_lock = mdc_getattr_lock,
1746 .m_intent_lock = mdc_intent_lock,
1748 .m_rename = mdc_rename,
1749 .m_setattr = mdc_setattr,
1751 .m_readpage = mdc_readpage,
1752 .m_unlink = mdc_unlink,
1753 .m_valid_attrs = mdc_valid_attrs,
1754 .m_req2lustre_md = mdc_req2lustre_md,
1755 .m_set_open_replay_data = mdc_set_open_replay_data,
1756 .m_clear_open_replay_data = mdc_clear_open_replay_data,
1757 .m_store_inode_generation = mdc_store_inode_generation,
1758 .m_set_lock_data = mdc_set_lock_data,
1759 .m_get_real_obd = mdc_get_real_obd,
1760 .m_change_cbdata_name = mdc_change_cbdata_name,
1761 .m_change_cbdata = mdc_change_cbdata,
1762 .m_access_check = mdc_access_check,
1765 int __init mdc_init(void)
1767 struct lprocfs_static_vars lvars;
1769 lprocfs_init_vars(mdc, &lvars);
1770 return class_register_type(&mdc_obd_ops, &mdc_md_ops,
1771 lvars.module_vars, OBD_MDC_DEVICENAME);
1775 static void /*__exit*/ mdc_exit(void)
1777 class_unregister_type(OBD_MDC_DEVICENAME);
1780 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
1781 MODULE_DESCRIPTION("Lustre Metadata Client");
1782 MODULE_LICENSE("GPL");
1784 EXPORT_SYMBOL(mdc_req2lustre_md);
1785 EXPORT_SYMBOL(mdc_change_cbdata);
1786 EXPORT_SYMBOL(mdc_getstatus);
1787 EXPORT_SYMBOL(mdc_getattr);
1788 EXPORT_SYMBOL(mdc_getattr_lock);
1789 EXPORT_SYMBOL(mdc_create);
1790 EXPORT_SYMBOL(mdc_unlink);
1791 EXPORT_SYMBOL(mdc_rename);
1792 EXPORT_SYMBOL(mdc_link);
1793 EXPORT_SYMBOL(mdc_readpage);
1794 EXPORT_SYMBOL(mdc_setattr);
1795 EXPORT_SYMBOL(mdc_close);
1796 EXPORT_SYMBOL(mdc_done_writing);
1797 EXPORT_SYMBOL(mdc_sync);
1798 EXPORT_SYMBOL(mdc_set_open_replay_data);
1799 EXPORT_SYMBOL(mdc_clear_open_replay_data);
1800 EXPORT_SYMBOL(mdc_store_inode_generation);
1802 module_init(mdc_init);
1803 module_exit(mdc_exit);