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 "mdc_internal.h"
44 #define REQUEST_MINOR 244
46 static int mdc_cleanup(struct obd_device *obd, int flags);
48 extern int mds_queue_req(struct ptlrpc_request *);
49 /* Helper that implements most of mdc_getstatus and signal_completed_replay. */
50 /* XXX this should become mdc_get_info("key"), sending MDS_GET_INFO RPC */
51 static int send_getstatus(struct obd_import *imp, struct lustre_id *rootid,
52 int level, int msg_flags)
54 struct ptlrpc_request *req;
55 struct mds_body *body;
56 int rc, size[2] = {0, sizeof(*body)};
59 //size[0] = lustre_secdesc_size();
61 req = ptlrpc_prep_req(imp, LUSTRE_MDS_VERSION, MDS_GETSTATUS,
64 GOTO(out, rc = -ENOMEM);
66 //lustre_pack_secdesc(req, size[0]);
68 body = lustre_msg_buf(req->rq_reqmsg, MDS_REQ_REC_OFF, sizeof (*body));
69 req->rq_send_state = level;
70 req->rq_replen = lustre_msg_size(1, &size[1]);
72 req->rq_reqmsg->flags |= msg_flags;
73 rc = ptlrpc_queue_wait(req);
76 body = lustre_swab_repbuf (req, 0, sizeof (*body),
77 lustre_swab_mds_body);
79 CERROR ("Can't extract mds_body\n");
80 GOTO (out, rc = -EPROTO);
83 memcpy(rootid, &body->id1, sizeof(*rootid));
85 CDEBUG(D_NET, "root ino="LPU64", last_committed="LPU64
86 ", last_xid="LPU64"\n", rootid->li_stc.u.e3s.l3s_ino,
87 req->rq_repmsg->last_committed, req->rq_repmsg->last_xid);
92 ptlrpc_req_finished(req);
96 /* This should be mdc_get_info("rootid") */
97 int mdc_getstatus(struct obd_export *exp, struct lustre_id *rootid)
99 return send_getstatus(class_exp2cliimp(exp), rootid,
103 int mdc_getattr_common(struct obd_export *exp, unsigned int ea_size,
104 struct ptlrpc_request *req)
106 struct mds_body *body, *reqbody;
109 int repsize[4] = {sizeof(*body)};
113 /* request message already built */
116 repsize[bufcount++] = ea_size;
117 CDEBUG(D_INODE, "reserved %u bytes for MD/symlink in packet\n",
121 reqbody = lustre_msg_buf(req->rq_reqmsg, 1, sizeof(*reqbody));
123 if (reqbody->valid & OBD_MD_FLACL_ACCESS) {
124 repsize[bufcount++] = 4;
125 repsize[bufcount++] = xattr_acl_size(LL_ACL_MAX_ENTRIES);
128 req->rq_replen = lustre_msg_size(bufcount, repsize);
130 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
131 rc = ptlrpc_queue_wait(req);
132 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
136 body = lustre_swab_repbuf (req, 0, sizeof (*body),
137 lustre_swab_mds_body);
139 CERROR ("Can't unpack mds_body\n");
143 CDEBUG(D_NET, "mode: %o\n", body->mode);
145 LASSERT_REPSWAB (req, 1);
147 /* Skip the check if getxattr/listxattr are called with no buffers */
148 if ((reqbody->valid & (OBD_MD_FLEA | OBD_MD_FLEALIST)) &&
149 (reqbody->eadatasize != 0)){
150 if (body->eadatasize != 0) {
151 /* reply indicates presence of eadata; check it's there... */
152 eadata = lustre_msg_buf (req->rq_repmsg, 1,
154 if (eadata == NULL) {
155 CERROR ("Missing/short eadata\n");
164 static int mdc_cancel_unused(struct obd_export *exp,
165 struct lov_stripe_md *lsm,
166 int flags, void *opaque)
168 struct obd_device *obd = class_exp2obd(exp);
171 RETURN(ldlm_cli_cancel_unused(obd->obd_namespace,
172 NULL, flags, opaque));
175 int mdc_getattr(struct obd_export *exp, struct lustre_id *id,
176 __u64 valid, const char *ea_name, int ea_namelen,
177 unsigned int ea_size, struct ptlrpc_request **request)
179 struct ptlrpc_request *req;
180 struct mds_body *body;
182 int size[3] = {0, sizeof(*body)};
186 /* XXX do we need to make another request here? We just did a getattr
187 * to do the lookup in the first place.
189 size[0] = lustre_secdesc_size();
191 LASSERT((ea_name != NULL) == (ea_namelen != 0));
192 if (valid & (OBD_MD_FLEA | OBD_MD_FLEALIST)) {
193 size[bufcount] = ea_namelen;
197 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
198 MDS_GETATTR, bufcount, size, NULL);
200 GOTO(out, rc = -ENOMEM);
202 lustre_pack_secdesc(req, size[0]);
204 body = lustre_msg_buf(req->rq_reqmsg, MDS_REQ_REC_OFF, sizeof (*body));
205 memcpy(&body->id1, id, sizeof(*id));
207 body->eadatasize = ea_size;
210 if (valid & OBD_MD_FLEA) {
211 LASSERT(strnlen(ea_name, ea_namelen) == (ea_namelen - 1));
212 memcpy(lustre_msg_buf(req->rq_reqmsg, 2, ea_namelen),
213 ea_name, ea_namelen);
216 rc = mdc_getattr_common(exp, ea_size, req);
218 ptlrpc_req_finished (req);
226 int mdc_getattr_lock(struct obd_export *exp, struct lustre_id *id,
227 char *filename, int namelen, __u64 valid,
228 unsigned int ea_size, struct ptlrpc_request **request)
230 struct ptlrpc_request *req;
231 struct mds_body *body;
232 int rc, size[3] = {0, sizeof(*body), namelen};
235 size[0] = lustre_secdesc_size();
237 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
238 MDS_GETATTR_LOCK, 3, size, NULL);
240 GOTO(out, rc = -ENOMEM);
242 lustre_pack_secdesc(req, size[0]);
244 body = lustre_msg_buf(req->rq_reqmsg, MDS_REQ_REC_OFF, sizeof (*body));
245 memcpy(&body->id1, id, sizeof(*id));
247 body->eadatasize = ea_size;
249 if (filename != NULL) {
250 LASSERT (strnlen (filename, namelen) == namelen - 1);
251 memcpy(lustre_msg_buf(req->rq_reqmsg, 2, namelen),
254 LASSERT(namelen == 1);
257 rc = mdc_getattr_common(exp, ea_size, req);
259 ptlrpc_req_finished (req);
267 /* This should be called with both the request and the reply still packed. */
268 int mdc_store_inode_generation(struct obd_export *exp,
269 struct ptlrpc_request *req,
270 int reqoff, int repoff)
272 struct mds_rec_create *rec =
273 lustre_msg_buf(req->rq_reqmsg, reqoff, sizeof(*rec));
274 struct mds_body *body =
275 lustre_msg_buf(req->rq_repmsg, repoff, sizeof(*body));
277 LASSERT (rec != NULL);
278 LASSERT (body != NULL);
280 memcpy(&rec->cr_replayid, &body->id1, sizeof(rec->cr_replayid));
281 DEBUG_REQ(D_HA, req, "storing generation for ino "DLID4,
282 OLID4(&rec->cr_replayid));
286 int mdc_req2lustre_md(struct obd_export *exp_lmv, struct ptlrpc_request *req,
287 unsigned int offset, struct obd_export *exp_lov,
288 struct lustre_md *md)
293 struct posix_acl *acl;
294 struct lov_mds_md *lmm;
298 memset(md, 0, sizeof(*md));
300 md->body = lustre_msg_buf(req->rq_repmsg, offset,
305 LASSERT_REPSWABBED(req, offset);
307 if (!(md->body->valid & OBD_MD_FLEASIZE) &&
308 !(md->body->valid & OBD_MD_FLDIREA))
311 if (S_ISREG(md->body->mode)) {
312 if (md->body->eadatasize == 0) {
313 CERROR("invalid EA size (0) is detected\n");
317 lmm = lustre_msg_buf(req->rq_repmsg, offset + 1,
318 md->body->eadatasize);
322 LASSERT(exp_lov != NULL);
324 rc = obd_unpackmd(exp_lov, &md->lsm, lmm,
325 md->body->eadatasize);
327 LASSERT(rc >= sizeof(*md->lsm));
330 } else if (S_ISDIR(md->body->mode)) {
331 /* dir can be non-splitted */
332 if (md->body->eadatasize == 0)
335 lmm = lustre_msg_buf(req->rq_repmsg, offset + 1,
336 md->body->eadatasize);
340 if (md->body->valid & OBD_MD_MEA) {
341 LASSERT(exp_lmv != NULL);
343 rc = obd_unpackmd(exp_lmv, (void *)&md->mea,
344 lmm, md->body->eadatasize);
346 LASSERT(rc >= sizeof(*md->mea));
351 LASSERT(S_ISCHR(md->body->mode) ||
352 S_ISBLK(md->body->mode) ||
353 S_ISFIFO(md->body->mode)||
354 S_ISLNK(md->body->mode) ||
355 S_ISSOCK(md->body->mode));
358 acl_off = (md->body->valid & OBD_MD_FLEASIZE) ? (offset + 2) :
361 if (md->body->valid & OBD_MD_FLACL_ACCESS) {
362 size = le32_to_cpu(*(__u32 *) lustre_msg_buf(req->rq_repmsg,
364 buf = lustre_msg_buf(req->rq_repmsg, acl_off + 1, size);
366 acl = posix_acl_from_xattr(buf, size);
369 CERROR("convert xattr to acl failed: %d\n", rc);
372 rc = posix_acl_valid(acl);
374 CERROR("acl valid error: %d\n", rc);
375 posix_acl_release(acl);
380 md->acl_access = acl;
385 static void mdc_commit_open(struct ptlrpc_request *req)
387 struct mdc_open_data *mod = req->rq_cb_data;
391 if (mod->mod_close_req != NULL)
392 mod->mod_close_req->rq_cb_data = NULL;
394 if (mod->mod_och != NULL)
395 mod->mod_och->och_mod = NULL;
397 OBD_FREE(mod, sizeof(*mod));
398 req->rq_cb_data = NULL;
399 LASSERT(atomic_read(&req->rq_refcount) > 1);
400 ptlrpc_req_finished(req);
403 static void mdc_replay_open(struct ptlrpc_request *req)
405 struct mdc_open_data *mod = req->rq_cb_data;
406 struct obd_client_handle *och;
407 struct ptlrpc_request *close_req;
408 struct lustre_handle old;
409 struct mds_body *body;
412 body = lustre_swab_repbuf(req, 1, sizeof(*body), lustre_swab_mds_body);
413 LASSERT (body != NULL);
416 DEBUG_REQ(D_ERROR, req,
417 "can't properly replay without open data");
424 struct lustre_handle *file_fh;
425 LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
426 file_fh = &och->och_fh;
427 CDEBUG(D_HA, "updating handle from "LPX64" to "LPX64"\n",
428 file_fh->cookie, body->handle.cookie);
429 memcpy(&old, file_fh, sizeof(old));
430 memcpy(file_fh, &body->handle, sizeof(*file_fh));
433 close_req = mod->mod_close_req;
434 if (close_req != NULL) {
435 struct mds_body *close_body;
436 LASSERT(close_req->rq_reqmsg->opc == MDS_CLOSE);
437 close_body = lustre_msg_buf(close_req->rq_reqmsg,
439 sizeof(*close_body));
441 LASSERT(!memcmp(&old, &close_body->handle, sizeof old));
442 DEBUG_REQ(D_HA, close_req, "updating close body with new fh");
443 memcpy(&close_body->handle, &body->handle,
444 sizeof(close_body->handle));
450 int mdc_set_open_replay_data(struct obd_export *exp,
451 struct obd_client_handle *och,
452 struct ptlrpc_request *open_req)
454 struct mdc_open_data *mod;
455 struct mds_rec_create *rec;
456 struct mds_body *body;
458 rec = lustre_msg_buf(open_req->rq_reqmsg, MDS_REQ_INTENT_REC_OFF,
460 body = lustre_msg_buf(open_req->rq_repmsg, 1, sizeof(*body));
462 LASSERT(rec != NULL);
463 /* outgoing messages always in my byte order */
464 LASSERT(body != NULL);
465 /* incoming message in my byte order (it's been swabbed) */
466 LASSERT_REPSWABBED(open_req, 1);
468 OBD_ALLOC(mod, sizeof(*mod));
470 DEBUG_REQ(D_ERROR, open_req, "can't allocate mdc_open_data");
476 mod->mod_open_req = ptlrpc_request_addref(open_req);
478 memcpy(&rec->cr_replayid, &body->id1, sizeof rec->cr_replayid);
479 open_req->rq_replay_cb = mdc_replay_open;
480 open_req->rq_commit_cb = mdc_commit_open;
481 open_req->rq_cb_data = mod;
482 DEBUG_REQ(D_HA, open_req, "set up replay data");
486 int mdc_clear_open_replay_data(struct obd_export *exp,
487 struct obd_client_handle *och)
489 struct mdc_open_data *mod = och->och_mod;
491 /* Don't free the structure now (it happens in mdc_commit_open, after
492 * we're sure we won't need to fix up the close request in the future),
493 * but make sure that replay doesn't poke at the och, which is about to
495 LASSERT(mod != LP_POISON);
502 static void mdc_commit_close(struct ptlrpc_request *req)
504 struct mdc_open_data *mod = req->rq_cb_data;
505 struct ptlrpc_request *open_req;
506 struct obd_import *imp = req->rq_import;
508 DEBUG_REQ(D_HA, req, "close req committed");
512 mod->mod_close_req = NULL;
513 req->rq_cb_data = NULL;
514 req->rq_commit_cb = NULL;
516 open_req = mod->mod_open_req;
517 LASSERT(open_req != NULL);
518 LASSERT(open_req != LP_POISON);
519 LASSERT(open_req->rq_type != LI_POISON);
521 DEBUG_REQ(D_HA, open_req, "open req balanced");
522 if (open_req->rq_transno == 0) {
523 DEBUG_REQ(D_ERROR, open_req, "BUG 3892 open");
524 DEBUG_REQ(D_ERROR, req, "BUG 3892 close");
525 LASSERTF(open_req->rq_transno != 0, "BUG 3892");
527 LASSERT(open_req->rq_import == imp);
529 /* We no longer want to preserve this for transno-unconditional
531 spin_lock(&open_req->rq_lock);
532 open_req->rq_replay = 0;
533 spin_unlock(&open_req->rq_lock);
536 int mdc_close(struct obd_export *exp, struct obdo *oa,
537 struct obd_client_handle *och,
538 struct ptlrpc_request **request)
540 struct obd_device *obd = class_exp2obd(exp);
541 struct obd_import *imp = class_exp2cliimp(exp);
542 int reqsize[3] = {0, sizeof(struct mds_body),
543 obd->u.cli.cl_max_mds_cookiesize};
544 int rc, repsize[3] = {sizeof(struct mds_body),
545 obd->u.cli.cl_max_mds_easize,
546 obd->u.cli.cl_max_mds_cookiesize};
547 struct ptlrpc_request *req;
548 struct mdc_open_data *mod;
551 if (imp->imp_connection == NULL) {
552 CERROR("request on not connected import %s\n",
553 imp->imp_obd->obd_name);
557 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
558 MDS_CLOSE, 3, reqsize, NULL);
560 GOTO(out, rc = -ENOMEM);
562 //reqsize[0] = lustre_secdesc_size();
563 //lustre_pack_secdesc(req, reqsize[0]);
565 /* Ensure that this close's handle is fixed up during replay. */
566 LASSERT(och != NULL);
568 if (likely(mod != NULL)) {
569 mod->mod_close_req = req;
570 LASSERT(mod->mod_open_req->rq_type != LI_POISON);
571 DEBUG_REQ(D_HA, mod->mod_open_req, "matched open");
573 CDEBUG(D_HA, "couldn't find open req; "
574 "expecting close error\n");
577 mdc_close_pack(req, 1, oa, oa->o_valid, och);
579 req->rq_replen = lustre_msg_size(3, repsize);
580 req->rq_commit_cb = mdc_commit_close;
581 LASSERT(req->rq_cb_data == NULL);
582 req->rq_cb_data = mod;
584 mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
585 rc = ptlrpc_queue_wait(req);
586 mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
588 if (req->rq_repmsg == NULL) {
589 CDEBUG(D_HA, "request failed to send: %p, %d\n", req,
592 rc = req->rq_status ? req->rq_status : -EIO;
593 } else if (rc == 0) {
594 rc = req->rq_repmsg->status;
595 if (req->rq_repmsg->type == PTL_RPC_MSG_ERR) {
596 DEBUG_REQ(D_ERROR, req, "type == PTL_RPC_MSG_ERR, "
602 CERROR("Unexpected: can't find mdc_open_data, but "
603 "close succeeded. Please tell CFS.\n");
604 if (!lustre_swab_repbuf(req, 0, sizeof(struct mds_body),
605 lustre_swab_mds_body))
607 CERROR("Error unpacking mds_body\n");
619 int mdc_done_writing(struct obd_export *exp, struct obdo *obdo)
621 struct ptlrpc_request *req;
622 struct mds_body *body;
623 int rc, size[2] = {0, sizeof(*body)};
626 size[0] = lustre_secdesc_size();
628 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
629 MDS_DONE_WRITING, 2, size, NULL);
633 lustre_pack_secdesc(req, size[0]);
635 body = lustre_msg_buf(req->rq_reqmsg, MDS_REQ_REC_OFF,
638 mdc_pack_id(&body->id1, obdo->o_id, 0, obdo->o_mode,
639 obdo->o_mds, obdo->o_fid);
641 body->size = obdo->o_size;
642 body->blocks = obdo->o_blocks;
643 body->flags = obdo->o_flags;
644 body->valid = obdo->o_valid;
646 req->rq_replen = lustre_msg_size(1, &size[1]);
648 rc = ptlrpc_queue_wait(req);
649 ptlrpc_req_finished(req);
653 int mdc_readpage(struct obd_export *exp,
654 struct lustre_id *id,
655 __u64 offset, struct page *page,
656 struct ptlrpc_request **request)
658 struct obd_import *imp = class_exp2cliimp(exp);
659 struct ptlrpc_request *req = NULL;
660 struct ptlrpc_bulk_desc *desc = NULL;
661 struct mds_body *body;
662 int rc, size[2] = {0, sizeof(*body)};
665 CDEBUG(D_INODE, "inode: %ld\n", (long)id->li_stc.u.e3s.l3s_ino);
667 size[0] = lustre_secdesc_size();
669 req = ptlrpc_prep_req(imp, LUSTRE_MDS_VERSION, MDS_READPAGE,
672 GOTO(out, rc = -ENOMEM);
673 /* XXX FIXME bug 249 */
674 req->rq_request_portal = MDS_READPAGE_PORTAL;
676 lustre_pack_secdesc(req, size[0]);
678 desc = ptlrpc_prep_bulk_imp(req, 1, BULK_PUT_SINK, MDS_BULK_PORTAL);
680 GOTO(out, rc = -ENOMEM);
681 /* NB req now owns desc and will free it when it gets freed */
683 ptlrpc_prep_bulk_page(desc, page, 0, PAGE_CACHE_SIZE);
684 mdc_readdir_pack(req, 1, offset, PAGE_CACHE_SIZE, id);
686 req->rq_replen = lustre_msg_size(1, &size[1]);
687 rc = ptlrpc_queue_wait(req);
690 body = lustre_swab_repbuf(req, 0, sizeof (*body),
691 lustre_swab_mds_body);
693 CERROR("Can't unpack mds_body\n");
694 GOTO(out, rc = -EPROTO);
697 if (req->rq_bulk->bd_nob_transferred != PAGE_CACHE_SIZE) {
698 CERROR ("Unexpected # bytes transferred: %d"
700 req->rq_bulk->bd_nob_transferred,
702 GOTO (out, rc = -EPROTO);
712 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
713 void *karg, void *uarg)
715 struct obd_device *obd = exp->exp_obd;
716 struct obd_ioctl_data *data = karg;
717 struct obd_import *imp = obd->u.cli.cl_import;
718 struct llog_ctxt *ctxt;
722 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
725 if (!try_module_get(THIS_MODULE)) {
726 CERROR("Can't get module. Is it alive?");
731 case OBD_IOC_CLIENT_RECOVER:
732 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1);
736 case IOC_OSC_SET_ACTIVE:
737 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
739 case IOC_OSC_CTL_RECOVERY:
740 rc = ptlrpc_import_control_recovery(imp, data->ioc_offset);
742 case OBD_IOC_PARSE: {
743 ctxt = llog_get_context(&exp->exp_obd->obd_llogs,
744 LLOG_CONFIG_REPL_CTXT);
745 rc = class_config_process_llog(ctxt, data->ioc_inlbuf1, NULL);
749 case OBD_IOC_LLOG_INFO:
750 case OBD_IOC_LLOG_PRINT: {
751 ctxt = llog_get_context(&obd->obd_llogs, LLOG_CONFIG_REPL_CTXT);
752 rc = llog_ioctl(ctxt, cmd, data);
758 CERROR("mdc_ioctl(): unrecognised ioctl %#x\n", cmd);
759 GOTO(out, rc = -ENOTTY);
762 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
765 module_put(THIS_MODULE);
771 int mdc_set_info(struct obd_export *exp, obd_count keylen,
772 void *key, obd_count vallen, void *val)
776 if (keylen == strlen("initial_recov") &&
777 memcmp(key, "initial_recov", strlen("initial_recov")) == 0) {
778 struct obd_import *imp = exp->exp_obd->u.cli.cl_import;
779 if (vallen != sizeof(int))
781 imp->imp_initial_recov = *(int *)val;
782 CDEBUG(D_HA, "%s: set imp_no_init_recov = %d\n",
783 exp->exp_obd->obd_name,
784 imp->imp_initial_recov);
786 } else if (keylen >= strlen("mds_type") &&
787 strcmp(key, "mds_type") == 0) {
788 struct ptlrpc_request *req;
789 char *bufs[2] = {key, val};
790 int rc, size[2] = {keylen, vallen};
792 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_OBD_VERSION,
793 OST_SET_INFO, 2, size, bufs);
797 req->rq_replen = lustre_msg_size(0, NULL);
798 rc = ptlrpc_queue_wait(req);
799 ptlrpc_req_finished(req);
801 } else if (keylen >= strlen("inter_mds") && strcmp(key, "inter_mds") == 0) {
802 struct obd_import *imp = class_exp2cliimp(exp);
803 imp->imp_server_timeout = 1;
804 CDEBUG(D_OTHER, "%s: timeout / 2\n", exp->exp_obd->obd_name);
806 } else if (keylen == strlen("sec") &&
807 memcmp(key, "sec", keylen) == 0) {
808 struct client_obd *cli = &exp->exp_obd->u.cli;
810 cli->cl_sec_flavor = ptlrpcs_name2flavor(val);
811 if (cli->cl_sec_flavor == PTLRPCS_FLVR_INVALID) {
812 CERROR("unrecognized security type %s\n", (char*) val);
817 } else if (keylen == strlen("sec_flags") &&
818 memcmp(key, "sec_flags", keylen) == 0) {
819 struct client_obd *cli = &exp->exp_obd->u.cli;
821 cli->cl_sec_flags = *((unsigned long *) val);
823 } else if (keylen == strlen("flush_cred") &&
824 memcmp(key, "flush_cred", keylen) == 0) {
825 struct client_obd *cli = &exp->exp_obd->u.cli;
828 ptlrpcs_import_flush_creds(cli->cl_import,
831 } else if (keylen == strlen("async") && memcmp(key, "async", keylen) == 0) {
832 struct client_obd *cl = &exp->exp_obd->u.cli;
833 if (vallen != sizeof(int))
835 cl->cl_async = *(int *)val;
836 CDEBUG(D_HA, "%s: set async = %d\n",
837 exp->exp_obd->obd_name, cl->cl_async);
843 static int mdc_statfs(struct obd_device *obd, struct obd_statfs *osfs,
844 unsigned long max_age)
846 struct obd_statfs *msfs;
847 struct ptlrpc_request *req;
848 int rc, size = sizeof(*msfs);
851 /* We could possibly pass max_age in the request (as an absolute
852 * timestamp or a "seconds.usec ago") so the target can avoid doing
853 * extra calls into the filesystem if that isn't necessary (e.g.
854 * during mount that would help a bit). Having relative timestamps
855 * is not so great if request processing is slow, while absolute
856 * timestamps are not ideal because they need time synchronization. */
857 req = ptlrpc_prep_req(obd->u.cli.cl_import, LUSTRE_MDS_VERSION,
858 MDS_STATFS, 0, NULL, NULL);
862 req->rq_replen = lustre_msg_size(1, &size);
864 mdc_get_rpc_lock(obd->u.cli.cl_rpc_lock, NULL);
865 rc = ptlrpc_queue_wait(req);
866 mdc_put_rpc_lock(obd->u.cli.cl_rpc_lock, NULL);
869 /* this can be LMV fake import, whcih is not connected. */
870 if (!req->rq_import->imp_connection)
871 memset(osfs, 0, sizeof(*osfs));
875 msfs = lustre_swab_repbuf(req, 0, sizeof(*msfs),
876 lustre_swab_obd_statfs);
878 CERROR("Can't unpack obd_statfs\n");
879 GOTO(out, rc = -EPROTO);
882 memcpy(osfs, msfs, sizeof (*msfs));
885 ptlrpc_req_finished(req);
889 static int mdc_pin(struct obd_export *exp, obd_id ino, __u32 gen, int type,
890 struct obd_client_handle *handle, int flag)
892 struct ptlrpc_request *req;
893 struct mds_body *body;
894 int rc, size[2] = {0, sizeof(*body)};
897 //size[0] = lustre_secdesc_size();
899 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
900 MDS_PIN, 2, size, NULL);
904 //lustre_pack_secdesc(req, size[0]);
906 body = lustre_msg_buf(req->rq_reqmsg,
907 MDS_REQ_REC_OFF, sizeof(*body));
909 /* FIXME-UMKA: here should be also mdsnum and fid. */
910 mdc_pack_id(&body->id1, ino, gen, type, 0, 0);
913 req->rq_replen = lustre_msg_size(1, &size[1]);
915 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
916 rc = ptlrpc_queue_wait(req);
917 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
919 CERROR("pin failed: %d\n", rc);
920 ptlrpc_req_finished(req);
924 body = lustre_swab_repbuf(req, 0, sizeof(*body), lustre_swab_mds_body);
926 ptlrpc_req_finished(req);
930 memcpy(&handle->och_fh, &body->handle, sizeof(body->handle));
931 handle->och_magic = OBD_CLIENT_HANDLE_MAGIC;
933 OBD_ALLOC(handle->och_mod, sizeof(*handle->och_mod));
934 if (handle->och_mod == NULL) {
935 DEBUG_REQ(D_ERROR, req, "can't allocate mdc_open_data");
938 handle->och_mod->mod_open_req = req; /* will be dropped by unpin */
943 static int mdc_unpin(struct obd_export *exp,
944 struct obd_client_handle *handle, int flag)
946 struct ptlrpc_request *req;
947 struct mds_body *body;
948 int rc, size[2] = {0, sizeof(*body)};
951 if (handle->och_magic != OBD_CLIENT_HANDLE_MAGIC)
954 //size[0] = lustre_secdesc_size();
956 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
957 MDS_CLOSE, 2, size, NULL);
961 //lustre_pack_secdesc(req, size[0]);
963 body = lustre_msg_buf(req->rq_reqmsg, MDS_REQ_REC_OFF, sizeof(*body));
964 memcpy(&body->handle, &handle->och_fh, sizeof(body->handle));
967 req->rq_replen = lustre_msg_size(0, NULL);
968 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
969 rc = ptlrpc_queue_wait(req);
970 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
973 CERROR("unpin failed: %d\n", rc);
975 ptlrpc_req_finished(req);
976 ptlrpc_req_finished(handle->och_mod->mod_open_req);
977 OBD_FREE(handle->och_mod, sizeof(*handle->och_mod));
981 int mdc_sync(struct obd_export *exp, struct lustre_id *id,
982 struct ptlrpc_request **request)
984 struct ptlrpc_request *req;
985 struct mds_body *body;
986 int size[2] = {0, sizeof(*body)};
990 //size[0] = lustre_secdesc_size();
992 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
993 MDS_SYNC, 2, size, NULL);
995 RETURN(rc = -ENOMEM);
997 //lustre_pack_secdesc(req, size[0]);
1000 body = lustre_msg_buf(req->rq_reqmsg, MDS_REQ_REC_OFF,
1002 memcpy(&body->id1, id, sizeof(*id));
1005 req->rq_replen = lustre_msg_size(1, &size[1]);
1007 rc = ptlrpc_queue_wait(req);
1008 if (rc || request == NULL)
1009 ptlrpc_req_finished(req);
1016 static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
1017 enum obd_import_event event)
1021 LASSERT(imp->imp_obd == obd);
1024 case IMP_EVENT_DISCON: {
1027 case IMP_EVENT_INACTIVE: {
1028 if (obd->obd_observer)
1029 rc = obd_notify(obd->obd_observer, obd, 0, 0);
1032 case IMP_EVENT_INVALIDATE: {
1033 struct ldlm_namespace *ns = obd->obd_namespace;
1035 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
1039 case IMP_EVENT_ACTIVE: {
1040 if (obd->obd_observer)
1041 rc = obd_notify(obd->obd_observer, obd, 1, 0);
1045 CERROR("Unknown import event %d\n", event);
1051 static int mdc_attach(struct obd_device *dev, obd_count len, void *data)
1053 struct lprocfs_static_vars lvars;
1055 lprocfs_init_vars(mdc, &lvars);
1056 return lprocfs_obd_attach(dev, lvars.obd_vars);
1059 static int mdc_detach(struct obd_device *dev)
1061 return lprocfs_obd_detach(dev);
1064 static int mdc_setup(struct obd_device *obd, obd_count len, void *buf)
1066 struct client_obd *cli = &obd->u.cli;
1070 OBD_ALLOC(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1071 if (!cli->cl_rpc_lock)
1073 mdc_init_rpc_lock(cli->cl_rpc_lock);
1077 OBD_ALLOC(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1078 if (!cli->cl_setattr_lock)
1079 GOTO(err_rpc_lock, rc = -ENOMEM);
1080 mdc_init_rpc_lock(cli->cl_setattr_lock);
1082 OBD_ALLOC(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1083 if (!cli->cl_close_lock)
1084 GOTO(err_setattr_lock, rc = -ENOMEM);
1085 mdc_init_rpc_lock(cli->cl_close_lock);
1087 rc = client_obd_setup(obd, len, buf);
1089 GOTO(err_close_lock, rc);
1091 rc = obd_llog_init(obd, &obd->obd_llogs, obd, 0, NULL);
1093 mdc_cleanup(obd, 0);
1094 CERROR("failed to setup llogging subsystems\n");
1100 OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1102 OBD_FREE(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1104 OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1109 static int mdc_init_ea_size(struct obd_export *exp, int easize, int cookiesize)
1111 struct obd_device *obd = exp->exp_obd;
1112 struct client_obd *cli = &obd->u.cli;
1115 if (cli->cl_max_mds_easize < easize)
1116 cli->cl_max_mds_easize = easize;
1117 if (cli->cl_max_mds_cookiesize < cookiesize)
1118 cli->cl_max_mds_cookiesize = cookiesize;
1122 static int mdc_precleanup(struct obd_device *obd, int flags)
1126 rc = obd_llog_finish(obd, &obd->obd_llogs, 0);
1128 CERROR("failed to cleanup llogging subsystems\n");
1133 static int mdc_cleanup(struct obd_device *obd, int flags)
1135 struct client_obd *cli = &obd->u.cli;
1137 OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1138 OBD_FREE(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1139 OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1143 return client_obd_cleanup(obd, flags);
1147 static int mdc_llog_init(struct obd_device *obd, struct obd_llogs *llogs,
1148 struct obd_device *tgt, int count,
1149 struct llog_catid *logid)
1151 struct llog_ctxt *ctxt;
1155 rc = obd_llog_setup(obd, llogs, LLOG_CONFIG_REPL_CTXT, tgt, 0, NULL,
1158 ctxt = llog_get_context(llogs, LLOG_CONFIG_REPL_CTXT);
1159 ctxt->loc_imp = obd->u.cli.cl_import;
1165 static int mdc_llog_finish(struct obd_device *obd,
1166 struct obd_llogs *llogs, int count)
1171 rc = obd_llog_cleanup(llog_get_context(llogs, LLOG_CONFIG_REPL_CTXT));
1174 static struct obd_device *mdc_get_real_obd(struct obd_export *exp,
1175 struct lustre_id *id)
1178 RETURN(exp->exp_obd);
1181 static int mdc_get_info(struct obd_export *exp, __u32 keylen,
1182 void *key, __u32 *valsize, void *val)
1184 struct ptlrpc_request *req;
1185 char *bufs[1] = {key};
1189 if (!valsize || !val)
1192 if (keylen == strlen("remote_flag") && !strcmp(key, "remote_flag")) {
1193 struct obd_import *imp;
1194 struct obd_connect_data *data;
1196 imp = class_exp2cliimp(exp);
1202 if (imp->imp_state != LUSTRE_IMP_FULL) {
1203 CERROR("import state not full\n");
1207 data = &imp->imp_connect_data;
1208 if (data->ocd_connect_flags & OBD_CONNECT_REMOTE) {
1211 } else if (data->ocd_connect_flags & OBD_CONNECT_LOCAL) {
1215 CERROR("no remote flag set?\n");
1219 if ((keylen < strlen("mdsize") || strcmp(key, "mdsize") != 0) &&
1220 (keylen < strlen("mdsnum") || strcmp(key, "mdsnum") != 0) &&
1221 (keylen < strlen("rootid") || strcmp(key, "rootid") != 0))
1224 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_OBD_VERSION,
1225 OST_GET_INFO, 1, (int *)&keylen, bufs);
1229 req->rq_replen = lustre_msg_size(1, (int *)valsize);
1230 rc = ptlrpc_queue_wait(req);
1234 if (keylen >= strlen("rootid") && !strcmp(key, "rootid")) {
1235 struct lustre_id *reply;
1237 reply = lustre_swab_repbuf(req, 0, sizeof(*reply),
1238 lustre_swab_lustre_id);
1239 if (reply == NULL) {
1240 CERROR("Can't unpack %s\n", (char *)key);
1241 GOTO(out_req, rc = -EPROTO);
1244 *(struct lustre_id *)val = *reply;
1248 reply = lustre_swab_repbuf(req, 0, sizeof(*reply),
1249 lustre_swab_generic_32s);
1250 if (reply == NULL) {
1251 CERROR("Can't unpack %s\n", (char *)key);
1252 GOTO(out_req, rc = -EPROTO);
1254 *((__u32 *)val) = *reply;
1257 ptlrpc_req_finished(req);
1261 int mdc_obj_create(struct obd_export *exp, struct obdo *oa,
1262 void *acl, int acl_size,
1263 struct lov_stripe_md **ea, struct obd_trans_info *oti)
1265 struct ptlrpc_request *request;
1266 struct ost_body *body;
1268 int rc, size[2] = { sizeof(*body), acl_size };
1273 request = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_OBD_VERSION,
1274 OST_CREATE, 2, size, NULL);
1276 GOTO(out_req, rc = -ENOMEM);
1278 body = lustre_msg_buf(request->rq_reqmsg, 0, sizeof (*body));
1279 memcpy(&body->oa, oa, sizeof(body->oa));
1282 acl_buf = lustre_msg_buf(request->rq_reqmsg, 1, acl_size);
1283 memcpy(acl_buf, acl, acl_size);
1286 request->rq_replen = lustre_msg_size(1, size);
1287 rc = ptlrpc_queue_wait(request);
1291 body = lustre_swab_repbuf(request, 0, sizeof(*body),
1292 lustre_swab_ost_body);
1294 CERROR ("can't unpack ost_body\n");
1295 GOTO (out_req, rc = -EPROTO);
1298 memcpy(oa, &body->oa, sizeof(*oa));
1300 /* store ino/generation for recovery */
1301 body = lustre_msg_buf(request->rq_reqmsg, 0, sizeof (*body));
1302 body->oa.o_id = oa->o_id;
1303 body->oa.o_generation = oa->o_generation;
1304 body->oa.o_fid = oa->o_fid;
1305 body->oa.o_mds = oa->o_mds;
1307 CDEBUG(D_HA, "transno: "LPD64"\n", request->rq_repmsg->transno);
1310 ptlrpc_req_finished(request);
1314 int mdc_brw(int rw, struct obd_export *exp, struct obdo *oa,
1315 struct lov_stripe_md *ea, obd_count oa_bufs,
1316 struct brw_page *pgarr, struct obd_trans_info *oti)
1318 struct ptlrpc_bulk_desc *desc;
1319 struct niobuf_remote *niobuf;
1320 struct ptlrpc_request *req;
1321 struct obd_ioobj *ioobj;
1322 struct ost_body *body;
1326 opc = ((rw & OBD_BRW_WRITE) != 0) ? OST_WRITE : OST_READ;
1328 size[0] = sizeof(*body);
1329 size[1] = sizeof(*ioobj);
1330 size[2] = oa_bufs * sizeof(*niobuf);
1332 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_OBD_VERSION, opc,
1334 LASSERT(req != NULL);
1336 if (opc == OST_WRITE)
1337 desc = ptlrpc_prep_bulk_imp(req, oa_bufs, BULK_GET_SOURCE,
1340 desc = ptlrpc_prep_bulk_imp(req, oa_bufs, BULK_PUT_SINK,
1342 LASSERT(desc != NULL);
1344 body = lustre_msg_buf(req->rq_reqmsg, 0, sizeof(*body));
1345 ioobj = lustre_msg_buf(req->rq_reqmsg, 1, sizeof(*ioobj));
1346 niobuf = lustre_msg_buf(req->rq_reqmsg, 2, oa_bufs * sizeof(*niobuf));
1348 memcpy(&body->oa, oa, sizeof(*oa));
1349 obdo_to_ioobj(oa, ioobj);
1350 ioobj->ioo_bufcnt = oa_bufs;
1352 for (i = 0; i < oa_bufs; i++, niobuf++) {
1353 struct brw_page *pg = &pgarr[i];
1355 LASSERT(pg->count > 0);
1356 LASSERT((pg->disk_offset & ~PAGE_MASK) + pg->count <= PAGE_SIZE);
1358 ptlrpc_prep_bulk_page(desc, pg->pg, pg->disk_offset & ~PAGE_MASK,
1361 niobuf->offset = pg->disk_offset;
1362 niobuf->len = pg->count;
1363 niobuf->flags = pg->flag;
1366 /* size[0] still sizeof (*body) */
1367 if (opc == OST_WRITE) {
1368 /* 1 RC per niobuf */
1369 size[1] = sizeof(__u32) * oa_bufs;
1370 req->rq_replen = lustre_msg_size(2, size);
1372 /* 1 RC for the whole I/O */
1373 req->rq_replen = lustre_msg_size(1, size);
1375 err = ptlrpc_queue_wait(req);
1378 ptlrpc_req_finished(req);
1382 static int mdc_valid_attrs(struct obd_export *exp,
1383 struct lustre_id *id)
1385 struct ldlm_res_id res_id = { .name = {0} };
1386 struct obd_device *obd = exp->exp_obd;
1387 struct lustre_handle lockh;
1388 ldlm_policy_data_t policy;
1392 res_id.name[0] = id_fid(id);
1393 res_id.name[1] = id_group(id);
1394 policy.l_inodebits.bits = MDS_INODELOCK_UPDATE;
1396 CDEBUG(D_INFO, "trying to match res "LPU64"\n",
1399 /* FIXME use LDLM_FL_TEST_LOCK instead */
1400 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING;
1401 if (ldlm_lock_match(obd->obd_namespace, flags, &res_id,
1402 LDLM_IBITS, &policy, LCK_PR, &lockh)) {
1403 ldlm_lock_decref(&lockh, LCK_PR);
1407 if (ldlm_lock_match(obd->obd_namespace, flags, &res_id,
1408 LDLM_IBITS, &policy, LCK_PW, &lockh)) {
1409 ldlm_lock_decref(&lockh, LCK_PW);
1415 static int mdc_change_cbdata_name(struct obd_export *exp,
1416 struct lustre_id *pid,
1417 char *name, int len,
1418 struct lustre_id *cid,
1419 ldlm_iterator_t it, void *data)
1422 rc = mdc_change_cbdata(exp, cid, it, data);
1426 struct obd_ops mdc_obd_ops = {
1427 .o_owner = THIS_MODULE,
1428 .o_attach = mdc_attach,
1429 .o_detach = mdc_detach,
1430 .o_setup = mdc_setup,
1431 .o_precleanup = mdc_precleanup,
1432 .o_cleanup = mdc_cleanup,
1433 .o_add_conn = client_import_add_conn,
1434 .o_del_conn = client_import_del_conn,
1435 .o_connect = client_connect_import,
1436 .o_disconnect = client_disconnect_export,
1437 .o_iocontrol = mdc_iocontrol,
1438 .o_packmd = mdc_packmd,
1439 .o_unpackmd = mdc_unpackmd,
1440 .o_statfs = mdc_statfs,
1442 .o_unpin = mdc_unpin,
1443 .o_import_event = mdc_import_event,
1444 .o_llog_init = mdc_llog_init,
1445 .o_llog_finish = mdc_llog_finish,
1446 .o_create = mdc_obj_create,
1447 .o_set_info = mdc_set_info,
1448 .o_get_info = mdc_get_info,
1450 .o_cancel_unused = mdc_cancel_unused,
1451 .o_init_ea_size = mdc_init_ea_size,
1454 struct md_ops mdc_md_ops = {
1455 .m_getstatus = mdc_getstatus,
1456 .m_getattr = mdc_getattr,
1457 .m_close = mdc_close,
1458 .m_create = mdc_create,
1459 .m_done_writing = mdc_done_writing,
1460 .m_enqueue = mdc_enqueue,
1461 .m_getattr_lock = mdc_getattr_lock,
1462 .m_intent_lock = mdc_intent_lock,
1464 .m_rename = mdc_rename,
1465 .m_setattr = mdc_setattr,
1467 .m_readpage = mdc_readpage,
1468 .m_unlink = mdc_unlink,
1469 .m_valid_attrs = mdc_valid_attrs,
1470 .m_req2lustre_md = mdc_req2lustre_md,
1471 .m_set_open_replay_data = mdc_set_open_replay_data,
1472 .m_clear_open_replay_data = mdc_clear_open_replay_data,
1473 .m_store_inode_generation = mdc_store_inode_generation,
1474 .m_set_lock_data = mdc_set_lock_data,
1475 .m_get_real_obd = mdc_get_real_obd,
1476 .m_change_cbdata_name = mdc_change_cbdata_name,
1477 .m_change_cbdata = mdc_change_cbdata,
1480 int __init mdc_init(void)
1482 struct lprocfs_static_vars lvars;
1483 lprocfs_init_vars(mdc, &lvars);
1484 return class_register_type(&mdc_obd_ops, &mdc_md_ops, lvars.module_vars,
1489 static void /*__exit*/ mdc_exit(void)
1491 class_unregister_type(LUSTRE_MDC_NAME);
1494 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
1495 MODULE_DESCRIPTION("Lustre Metadata Client");
1496 MODULE_LICENSE("GPL");
1498 EXPORT_SYMBOL(mdc_req2lustre_md);
1499 EXPORT_SYMBOL(mdc_change_cbdata);
1500 EXPORT_SYMBOL(mdc_getstatus);
1501 EXPORT_SYMBOL(mdc_getattr);
1502 EXPORT_SYMBOL(mdc_getattr_lock);
1503 EXPORT_SYMBOL(mdc_create);
1504 EXPORT_SYMBOL(mdc_unlink);
1505 EXPORT_SYMBOL(mdc_rename);
1506 EXPORT_SYMBOL(mdc_link);
1507 EXPORT_SYMBOL(mdc_readpage);
1508 EXPORT_SYMBOL(mdc_setattr);
1509 EXPORT_SYMBOL(mdc_close);
1510 EXPORT_SYMBOL(mdc_done_writing);
1511 EXPORT_SYMBOL(mdc_sync);
1512 EXPORT_SYMBOL(mdc_set_open_replay_data);
1513 EXPORT_SYMBOL(mdc_clear_open_replay_data);
1514 EXPORT_SYMBOL(mdc_store_inode_generation);
1516 module_init(mdc_init);
1517 module_exit(mdc_exit);