1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Copyright (C) 2001-2003 Cluster File Systems, Inc.
6 * This file is part of the Lustre file system, http://www.lustre.org
7 * Lustre is a trademark of Cluster File Systems, Inc.
9 * You may have signed or agreed to another license before downloading
10 * this software. If so, you are bound by the terms and conditions
11 * of that agreement, and the following does not apply to you. See the
12 * LICENSE file included with this distribution for more information.
14 * If you did not agree to a different license, then this copy of Lustre
15 * is open source software; you can redistribute it and/or modify it
16 * under the terms of version 2 of the GNU General Public License as
17 * published by the Free Software Foundation.
19 * In either case, Lustre is distributed in the hope that it will be
20 * useful, but WITHOUT ANY WARRANTY; without even the implied warranty
21 * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
22 * license text for more details.
26 # define EXPORT_SYMTAB
28 #define DEBUG_SUBSYSTEM S_MDC
31 # include <linux/module.h>
32 # include <linux/pagemap.h>
33 # include <linux/miscdevice.h>
34 # include <linux/init.h>
36 # include <liblustre.h>
39 #include <obd_class.h>
40 #include <lustre_dlm.h>
41 #include <lprocfs_status.h>
42 #include <lustre_param.h>
43 #include "mdc_internal.h"
45 static quota_interface_t *quota_interface;
47 #define REQUEST_MINOR 244
49 static quota_interface_t *quota_interface;
50 extern quota_interface_t mdc_quota_interface;
52 static int mdc_cleanup(struct obd_device *obd);
54 extern int mds_queue_req(struct ptlrpc_request *);
55 /* Helper that implements most of mdc_getstatus and signal_completed_replay. */
56 /* XXX this should become mdc_get_info("key"), sending MDS_GET_INFO RPC */
57 static int send_getstatus(struct obd_import *imp, struct ll_fid *rootfid,
58 int level, int msg_flags)
60 struct ptlrpc_request *req;
61 struct mds_body *body;
62 int rc, size[2] = { sizeof(struct ptlrpc_body), sizeof(*body) };
65 req = ptlrpc_prep_req(imp, LUSTRE_MDS_VERSION, MDS_GETSTATUS, 2, size,
68 GOTO(out, rc = -ENOMEM);
70 req->rq_send_state = level;
71 ptlrpc_req_set_repsize(req, 2, size);
73 mdc_pack_req_body(req, REQ_REC_OFF, 0, NULL, 0, 0);
74 lustre_msg_add_flags(req->rq_reqmsg, msg_flags);
75 rc = ptlrpc_queue_wait(req);
78 body = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*body),
79 lustre_swab_mds_body);
81 CERROR ("Can't extract mds_body\n");
82 GOTO (out, rc = -EPROTO);
85 memcpy(rootfid, &body->fid1, sizeof(*rootfid));
87 CDEBUG(D_NET, "root ino="LPU64", last_committed="LPU64
88 ", last_xid="LPU64"\n",
90 lustre_msg_get_last_committed(req->rq_repmsg),
91 lustre_msg_get_last_xid(req->rq_repmsg));
96 ptlrpc_req_finished(req);
100 /* This should be mdc_get_info("rootfid") */
101 int mdc_getstatus(struct obd_export *exp, struct ll_fid *rootfid)
103 return send_getstatus(class_exp2cliimp(exp), rootfid, LUSTRE_IMP_FULL,
108 int mdc_getattr_common(struct obd_export *exp, unsigned int ea_size,
109 unsigned int acl_size, struct ptlrpc_request *req)
111 struct mds_body *body;
113 int size[4] = { sizeof(struct ptlrpc_body), sizeof(*body) };
114 int bufcount = 2, rc;
117 /* request message already built */
120 size[bufcount++] = ea_size;
121 CDEBUG(D_INODE, "reserved %u bytes for MD/symlink in packet\n",
125 size[bufcount++] = acl_size;
126 CDEBUG(D_INODE, "reserved %u bytes for ACL\n", acl_size);
129 ptlrpc_req_set_repsize(req, bufcount, size);
131 rc = ptlrpc_queue_wait(req);
135 body = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*body),
136 lustre_swab_mds_body);
138 CERROR ("Can't unpack mds_body\n");
142 CDEBUG(D_NET, "mode: %o\n", body->mode);
144 LASSERT_REPSWAB(req, REPLY_REC_OFF + 1);
145 if (body->eadatasize != 0) {
146 /* reply indicates presence of eadata; check it's there... */
147 eadata = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF + 1,
149 if (eadata == NULL) {
150 CERROR ("Missing/short eadata\n");
155 if (body->valid & OBD_MD_FLMODEASIZE) {
156 if (exp->exp_obd->u.cli.cl_max_mds_easize < body->max_mdsize)
157 exp->exp_obd->u.cli.cl_max_mds_easize =
159 if (exp->exp_obd->u.cli.cl_max_mds_cookiesize <
160 body->max_cookiesize)
161 exp->exp_obd->u.cli.cl_max_mds_cookiesize =
162 body->max_cookiesize;
168 int mdc_getattr(struct obd_export *exp, struct ll_fid *fid,
169 obd_valid valid, unsigned int ea_size,
170 struct ptlrpc_request **request)
172 struct ptlrpc_request *req;
173 int size[2] = { sizeof(struct ptlrpc_body), sizeof(struct mds_body) };
174 int acl_size = 0, rc;
177 /* XXX do we need to make another request here? We just did a getattr
178 * to do the lookup in the first place.
180 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
181 MDS_GETATTR, 2, size, NULL);
183 GOTO(out, rc = -ENOMEM);
185 mdc_pack_req_body(req, REQ_REC_OFF, valid, fid, ea_size,
186 MDS_BFLAG_EXT_FLAGS/*request "new" flags(bug 9486)*/);
188 /* currently only root inode will call us with FLACL */
189 if (valid & OBD_MD_FLACL)
190 acl_size = LUSTRE_POSIX_ACL_MAX_SIZE;
192 rc = mdc_getattr_common(exp, ea_size, acl_size, req);
194 ptlrpc_req_finished (req);
202 int mdc_getattr_name(struct obd_export *exp, struct ll_fid *fid,
203 const char *filename, int namelen, unsigned long valid,
204 unsigned int ea_size, struct ptlrpc_request **request)
206 struct ptlrpc_request *req;
207 struct mds_body *body;
208 int rc, size[3] = { sizeof(struct ptlrpc_body), sizeof(*body), namelen};
211 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
212 MDS_GETATTR_NAME, 3, size, NULL);
214 GOTO(out, rc = -ENOMEM);
216 mdc_pack_req_body(req, REQ_REC_OFF, valid, fid, ea_size,
217 MDS_BFLAG_EXT_FLAGS/*request "new" flags(bug 9486)*/);
219 LASSERT(strnlen(filename, namelen) == namelen - 1);
220 memcpy(lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF + 1, namelen),
223 rc = mdc_getattr_common(exp, ea_size, 0, req);
225 ptlrpc_req_finished (req);
234 int mdc_xattr_common(struct obd_export *exp, struct ll_fid *fid,
235 int opcode, obd_valid valid, const char *xattr_name,
236 const char *input, int input_size, int output_size,
237 int flags, struct ptlrpc_request **request)
239 struct ptlrpc_request *req;
240 int size[4] = { sizeof(struct ptlrpc_body), sizeof(struct mds_body) };
241 // int size[3] = {sizeof(struct mds_body)}, bufcnt = 1;
242 int rc, xattr_namelen = 0, bufcnt = 2, offset;
247 xattr_namelen = strlen(xattr_name) + 1;
248 size[bufcnt++] = xattr_namelen;
252 size[bufcnt++] = input_size;
255 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
256 opcode, bufcnt, size, NULL);
258 GOTO(out, rc = -ENOMEM);
261 mdc_pack_req_body(req, REQ_REC_OFF, valid, fid, output_size, flags);
263 offset = REQ_REC_OFF + 1;
266 tmp = lustre_msg_buf(req->rq_reqmsg, offset++, xattr_namelen);
267 memcpy(tmp, xattr_name, xattr_namelen);
270 tmp = lustre_msg_buf(req->rq_reqmsg, offset++, input_size);
271 memcpy(tmp, input, input_size);
275 if (opcode == MDS_GETXATTR) {
276 size[REPLY_REC_OFF] = sizeof(struct mds_body);
282 /* we do this even output_size is 0, because server is doing that */
283 size[bufcnt++] = output_size;
285 ptlrpc_req_set_repsize(req, bufcnt, size);
288 if (opcode == MDS_SETXATTR)
289 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
291 rc = ptlrpc_queue_wait(req);
293 if (opcode == MDS_SETXATTR)
294 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
299 if (opcode == MDS_GETXATTR) {
300 struct mds_body * body = lustre_swab_repbuf(req, REPLY_REC_OFF,
302 lustre_swab_mds_body);
304 CERROR ("Can't unpack mds_body\n");
305 GOTO(err_out, rc = -EPROTO);
312 ptlrpc_req_finished(req);
317 int mdc_setxattr(struct obd_export *exp, struct ll_fid *fid,
318 obd_valid valid, const char *xattr_name,
319 const char *input, int input_size,
320 int output_size, int flags,
321 struct ptlrpc_request **request)
323 return mdc_xattr_common(exp, fid, MDS_SETXATTR, valid, xattr_name,
324 input, input_size, output_size, flags, request);
327 int mdc_getxattr(struct obd_export *exp, struct ll_fid *fid,
328 obd_valid valid, const char *xattr_name,
329 const char *input, int input_size,
330 int output_size, struct ptlrpc_request **request)
332 return mdc_xattr_common(exp, fid, MDS_GETXATTR, valid, xattr_name,
333 input, input_size, output_size, 0, request);
336 /* This should be called with both the request and the reply still packed. */
337 void mdc_store_inode_generation(struct ptlrpc_request *req, int reqoff,
340 struct mds_rec_create *rec = lustre_msg_buf(req->rq_reqmsg, reqoff,
342 struct mds_body *body = lustre_msg_buf(req->rq_repmsg, repoff,
345 LASSERT (rec != NULL);
346 LASSERT (body != NULL);
348 memcpy(&rec->cr_replayfid, &body->fid1, sizeof rec->cr_replayfid);
349 if (body->fid1.id == 0) {
350 DEBUG_REQ(D_ERROR, req, "saving replay request with id = 0 "
351 "gen = %u", body->fid1.generation);
355 DEBUG_REQ(D_HA, req, "storing generation %u for ino "LPU64,
356 rec->cr_replayfid.generation, rec->cr_replayfid.id);
359 #ifdef CONFIG_FS_POSIX_ACL
361 int mdc_unpack_acl(struct obd_export *exp, struct ptlrpc_request *req,
362 struct lustre_md *md, unsigned int offset)
364 struct mds_body *body = md->body;
365 struct posix_acl *acl;
372 buf = lustre_msg_buf(req->rq_repmsg, offset, body->aclsize);
374 CERROR("aclsize %u, bufcount %u, bufsize %u\n",
375 body->aclsize, lustre_msg_bufcount(req->rq_repmsg),
376 (lustre_msg_bufcount(req->rq_repmsg) <= offset) ?
377 -1 : lustre_msg_buflen(req->rq_repmsg, offset));
381 acl = posix_acl_from_xattr(buf, body->aclsize);
384 CERROR("convert xattr to acl: %d\n", rc);
388 rc = posix_acl_valid(acl);
390 CERROR("validate acl: %d\n", rc);
391 posix_acl_release(acl);
399 #define mdc_unpack_acl(exp, req, md, offset) 0
402 int mdc_req2lustre_md(struct ptlrpc_request *req, int offset,
403 struct obd_export *exp,
404 struct lustre_md *md)
410 memset(md, 0, sizeof(*md));
412 md->body = lustre_msg_buf(req->rq_repmsg, offset, sizeof (*md->body));
413 LASSERT (md->body != NULL);
414 LASSERT_REPSWABBED(req, offset);
417 if (md->body->valid & OBD_MD_FLEASIZE) {
419 struct lov_mds_md *lmm;
421 LASSERT(S_ISREG(md->body->mode));
423 if (md->body->eadatasize == 0) {
424 CERROR ("OBD_MD_FLEASIZE set, but eadatasize 0\n");
427 lmmsize = md->body->eadatasize;
428 lmm = lustre_msg_buf(req->rq_repmsg, offset, lmmsize);
429 LASSERT (lmm != NULL);
430 LASSERT_REPSWABBED(req, offset);
432 rc = obd_unpackmd(exp, &md->lsm, lmm, lmmsize);
436 LASSERT (rc >= sizeof (*md->lsm));
442 /* for ACL, it's possible that FLACL is set but aclsize is zero.
443 * only when aclsize != 0 there's an actual segment for ACL in
446 if ((md->body->valid & OBD_MD_FLACL) && md->body->aclsize) {
447 rc = mdc_unpack_acl(exp, req, md, offset);
457 obd_free_memmd(exp, &md->lsm);
461 void mdc_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
464 obd_free_memmd(exp, &md->lsm);
466 #ifdef CONFIG_FS_POSIX_ACL
468 posix_acl_release(md->posix_acl);
469 md->posix_acl = NULL;
474 static void mdc_commit_open(struct ptlrpc_request *req)
476 struct mdc_open_data *mod = req->rq_cb_data;
480 if (mod->mod_close_req != NULL)
481 mod->mod_close_req->rq_cb_data = NULL;
483 if (mod->mod_och != NULL)
484 mod->mod_och->och_mod = NULL;
486 OBD_FREE(mod, sizeof(*mod));
487 req->rq_cb_data = NULL;
490 static void mdc_replay_open(struct ptlrpc_request *req)
492 struct mdc_open_data *mod = req->rq_cb_data;
493 struct obd_client_handle *och;
494 struct ptlrpc_request *close_req;
495 struct lustre_handle old;
496 struct mds_body *body;
499 body = lustre_swab_repbuf(req, DLM_REPLY_REC_OFF, sizeof(*body),
500 lustre_swab_mds_body);
501 LASSERT (body != NULL);
504 DEBUG_REQ(D_ERROR, req,
505 "can't properly replay without open data");
512 struct lustre_handle *file_fh;
513 LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
514 file_fh = &och->och_fh;
515 CDEBUG(D_HA, "updating handle from "LPX64" to "LPX64"\n",
516 file_fh->cookie, body->handle.cookie);
517 memcpy(&old, file_fh, sizeof(old));
518 memcpy(file_fh, &body->handle, sizeof(*file_fh));
521 close_req = mod->mod_close_req;
522 if (close_req != NULL) {
523 struct mds_body *close_body;
524 LASSERT(lustre_msg_get_opc(close_req->rq_reqmsg) == MDS_CLOSE);
525 close_body = lustre_msg_buf(close_req->rq_reqmsg, REQ_REC_OFF,
526 sizeof(*close_body));
528 LASSERT(!memcmp(&old, &close_body->handle, sizeof old));
529 DEBUG_REQ(D_HA, close_req, "updating close body with new fh");
530 memcpy(&close_body->handle, &body->handle,
531 sizeof(close_body->handle));
537 void mdc_set_open_replay_data(struct obd_client_handle *och,
538 struct ptlrpc_request *open_req)
540 struct mdc_open_data *mod;
541 struct mds_rec_create *rec = lustre_msg_buf(open_req->rq_reqmsg,
544 struct mds_body *body = lustre_msg_buf(open_req->rq_repmsg,
548 /* incoming message in my byte order (it's been swabbed) */
549 LASSERT(rec != NULL);
550 LASSERT_REPSWABBED(open_req, DLM_REPLY_REC_OFF);
551 /* outgoing messages always in my byte order */
552 LASSERT(body != NULL);
555 OBD_ALLOC(mod, sizeof(*mod));
557 DEBUG_REQ(D_ERROR, open_req, "can't allocate mdc_open_data");
561 spin_lock(&open_req->rq_lock);
562 if (!open_req->rq_replay) {
563 OBD_FREE(mod, sizeof(*mod));
564 spin_unlock(&open_req->rq_lock);
570 mod->mod_open_req = open_req;
571 open_req->rq_cb_data = mod;
572 open_req->rq_commit_cb = mdc_commit_open;
573 spin_unlock(&open_req->rq_lock);
576 memcpy(&rec->cr_replayfid, &body->fid1, sizeof rec->cr_replayfid);
577 open_req->rq_replay_cb = mdc_replay_open;
578 if (body->fid1.id == 0) {
579 DEBUG_REQ(D_ERROR, open_req, "saving replay request with "
580 "id = 0 gen = %u", body->fid1.generation);
584 DEBUG_REQ(D_HA, open_req, "set up replay data");
587 void mdc_clear_open_replay_data(struct obd_client_handle *och)
589 struct mdc_open_data *mod = och->och_mod;
591 /* Don't free the structure now (it happens in mdc_commit_open, after
592 * we're sure we won't need to fix up the close request in the future),
593 * but make sure that replay doesn't poke at the och, which is about to
595 LASSERT(mod != LP_POISON);
601 static void mdc_commit_close(struct ptlrpc_request *req)
603 struct mdc_open_data *mod = req->rq_cb_data;
604 struct ptlrpc_request *open_req;
605 struct obd_import *imp = req->rq_import;
607 DEBUG_REQ(D_HA, req, "close req committed");
611 mod->mod_close_req = NULL;
612 req->rq_cb_data = NULL;
613 req->rq_commit_cb = NULL;
615 open_req = mod->mod_open_req;
616 LASSERT(open_req != NULL);
617 LASSERT(open_req != LP_POISON);
618 LASSERT(open_req->rq_type != LI_POISON);
620 DEBUG_REQ(D_HA, open_req, "open req balanced");
621 LASSERT(open_req->rq_transno != 0);
622 LASSERT(open_req->rq_import == imp);
624 /* We no longer want to preserve this for transno-unconditional
626 spin_lock(&open_req->rq_lock);
627 open_req->rq_replay = 0;
628 spin_unlock(&open_req->rq_lock);
631 int mdc_close(struct obd_export *exp, struct obdo *oa,
632 struct obd_client_handle *och, struct ptlrpc_request **request)
634 struct obd_device *obd = class_exp2obd(exp);
635 int reqsize[2] = { sizeof(struct ptlrpc_body),
636 sizeof(struct mds_body) };
637 int rc, repsize[4] = { sizeof(struct ptlrpc_body),
638 sizeof(struct mds_body),
639 obd->u.cli.cl_max_mds_easize,
640 obd->u.cli.cl_max_mds_cookiesize };
641 struct ptlrpc_request *req;
642 struct mdc_open_data *mod;
645 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
646 MDS_CLOSE, 2, reqsize, NULL);
648 GOTO(out, rc = -ENOMEM);
650 /* To avoid a livelock (bug 7034), we need to send CLOSE RPCs to a
651 * portal whose threads are not taking any DLM locks and are therefore
652 * always progressing */
653 /* XXX FIXME bug 249 */
654 req->rq_request_portal = MDS_READPAGE_PORTAL;
656 /* Ensure that this close's handle is fixed up during replay. */
657 LASSERT(och != NULL);
658 LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
660 if (likely(mod != NULL)) {
661 if (mod->mod_open_req->rq_type == LI_POISON) {
662 CERROR("LBUG POISONED open %p!\n", mod->mod_open_req);
664 ptlrpc_req_finished(req);
666 GOTO(out, rc = -EIO);
668 mod->mod_close_req = req;
669 DEBUG_REQ(D_HA, mod->mod_open_req, "matched open");
671 CDEBUG(D_HA, "couldn't find open req; expecting close error\n");
674 mdc_close_pack(req, REQ_REC_OFF, oa, oa->o_valid, och);
676 ptlrpc_req_set_repsize(req, 4, repsize);
677 req->rq_commit_cb = mdc_commit_close;
678 LASSERT(req->rq_cb_data == NULL);
679 req->rq_cb_data = mod;
681 mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
682 rc = ptlrpc_queue_wait(req);
683 mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
685 if (req->rq_repmsg == NULL) {
686 CDEBUG(D_HA, "request failed to send: %p, %d\n", req,
689 rc = req->rq_status ? req->rq_status : -EIO;
690 } else if (rc == 0) {
691 rc = lustre_msg_get_status(req->rq_repmsg);
692 if (lustre_msg_get_type(req->rq_repmsg) == PTL_RPC_MSG_ERR) {
693 DEBUG_REQ(D_ERROR, req, "type == PTL_RPC_MSG_ERR, err "
697 } else if (mod == NULL) {
698 CERROR("Unexpected: can't find mdc_open_data, but the "
699 "close succeeded. Please tell CFS.\n");
701 if (!lustre_swab_repbuf(req, REPLY_REC_OFF,
702 sizeof(struct mds_body),
703 lustre_swab_mds_body)) {
704 CERROR("Error unpacking mds_body\n");
712 if (rc != 0 && req && req->rq_commit_cb)
713 req->rq_commit_cb(req);
718 int mdc_done_writing(struct obd_export *exp, struct obdo *obdo)
720 struct ptlrpc_request *req;
721 struct mds_body *body;
722 int rc, size[2] = { sizeof(struct ptlrpc_body), sizeof(*body) };
725 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
726 MDS_DONE_WRITING, 2, size, NULL);
730 body = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF, sizeof(*body));
731 mdc_pack_fid(&body->fid1, obdo->o_id, 0, obdo->o_mode);
732 body->size = obdo->o_size;
733 body->blocks = obdo->o_blocks;
734 body->flags = obdo->o_flags;
735 body->valid = obdo->o_valid;
736 // memcpy(&body->handle, &och->och_fh, sizeof(body->handle));
738 ptlrpc_req_set_repsize(req, 2, size);
740 rc = ptlrpc_queue_wait(req);
741 ptlrpc_req_finished(req);
745 int mdc_readpage(struct obd_export *exp, struct ll_fid *fid, __u64 offset,
746 struct page *page, struct ptlrpc_request **request)
748 struct obd_import *imp = class_exp2cliimp(exp);
749 struct ptlrpc_request *req = NULL;
750 struct ptlrpc_bulk_desc *desc = NULL;
751 struct mds_body *body;
752 int rc, size[2] = { sizeof(struct ptlrpc_body), sizeof(*body) };
755 CDEBUG(D_INODE, "inode: "LPU64"\n", fid->id);
757 req = ptlrpc_prep_req(imp, LUSTRE_MDS_VERSION, MDS_READPAGE, 2, size,
760 GOTO(out, rc = -ENOMEM);
762 /* XXX FIXME bug 249 */
763 req->rq_request_portal = MDS_READPAGE_PORTAL;
765 desc = ptlrpc_prep_bulk_imp(req, 1, BULK_PUT_SINK, MDS_BULK_PORTAL);
767 GOTO(out, rc = -ENOMEM);
768 /* NB req now owns desc and will free it when it gets freed */
770 ptlrpc_prep_bulk_page(desc, page, 0, CFS_PAGE_SIZE);
772 mdc_readdir_pack(req, REQ_REC_OFF, offset, CFS_PAGE_SIZE, fid);
774 ptlrpc_req_set_repsize(req, 2, size);
775 rc = ptlrpc_queue_wait(req);
778 body = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*body),
779 lustre_swab_mds_body);
781 CERROR("Can't unpack mds_body\n");
782 GOTO(out, rc = -EPROTO);
785 if (req->rq_bulk->bd_nob_transferred != CFS_PAGE_SIZE) {
786 CERROR ("Unexpected # bytes transferred: %d"
788 req->rq_bulk->bd_nob_transferred,
790 GOTO (out, rc = -EPROTO);
801 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
802 void *karg, void *uarg)
804 struct obd_device *obd = exp->exp_obd;
805 struct obd_ioctl_data *data = karg;
806 struct obd_import *imp = obd->u.cli.cl_import;
807 struct llog_ctxt *ctxt;
811 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
814 if (!try_module_get(THIS_MODULE)) {
815 CERROR("Can't get module. Is it alive?");
820 case OBD_IOC_CLIENT_RECOVER:
821 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1);
825 case IOC_OSC_SET_ACTIVE:
826 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
828 case OBD_IOC_PARSE: {
829 ctxt = llog_get_context(exp->exp_obd, LLOG_CONFIG_REPL_CTXT);
830 rc = class_config_parse_llog(ctxt, data->ioc_inlbuf1, NULL);
834 case OBD_IOC_LLOG_INFO:
835 case OBD_IOC_LLOG_PRINT: {
836 ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT);
837 rc = llog_ioctl(ctxt, cmd, data);
842 case OBD_IOC_POLL_QUOTACHECK:
843 rc = lquota_poll_check(quota_interface, exp,
844 (struct if_quotacheck *)karg);
847 CERROR("mdc_ioctl(): unrecognised ioctl %#x\n", cmd);
848 GOTO(out, rc = -ENOTTY);
851 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
854 module_put(THIS_MODULE);
860 int mdc_set_info_async(struct obd_export *exp, obd_count keylen,
861 void *key, obd_count vallen, void *val,
862 struct ptlrpc_request_set *set)
864 struct obd_import *imp = class_exp2cliimp(exp);
867 if (KEY_IS(KEY_INIT_RECOV)) {
868 if (vallen != sizeof(int))
870 spin_lock(&imp->imp_lock);
871 imp->imp_initial_recov = *(int *)val;
872 spin_unlock(&imp->imp_lock);
874 CDEBUG(D_HA, "%s: set imp_initial_recov = %d\n",
875 exp->exp_obd->obd_name, imp->imp_initial_recov);
878 /* Turn off initial_recov after we try all backup servers once */
879 if (KEY_IS(KEY_INIT_RECOV_BACKUP)) {
880 if (vallen != sizeof(int))
883 spin_lock(&imp->imp_lock);
884 imp->imp_initial_recov_bk = *(int *)val;
885 if (imp->imp_initial_recov_bk)
886 imp->imp_initial_recov = 1;
887 spin_unlock(&imp->imp_lock);
889 CDEBUG(D_HA, "%s: set imp_initial_recov_bk = %d\n",
890 exp->exp_obd->obd_name, imp->imp_initial_recov_bk);
893 if (KEY_IS("read-only")) {
894 struct ptlrpc_request *req;
895 int size[3] = { sizeof(struct ptlrpc_body), keylen, vallen };
896 char *bufs[3] = { NULL, key, val };
898 if (vallen != sizeof(int))
902 imp->imp_connect_flags_orig |= OBD_CONNECT_RDONLY;
903 imp->imp_connect_data.ocd_connect_flags |=
906 imp->imp_connect_flags_orig &= ~OBD_CONNECT_RDONLY;
907 imp->imp_connect_data.ocd_connect_flags &=
911 req = ptlrpc_prep_req(imp, LUSTRE_MDS_VERSION, MDS_SET_INFO,
916 ptlrpc_req_set_repsize(req, 1, NULL);
919 ptlrpc_set_add_req(set, req);
920 ptlrpc_check_set(set);
922 rc = ptlrpc_queue_wait(req);
923 ptlrpc_req_finished(req);
932 int mdc_get_info(struct obd_export *exp, __u32 keylen, void *key,
933 __u32 *vallen, void *val)
937 if (keylen == strlen("max_easize") &&
938 memcmp(key, "max_easize", strlen("max_easize")) == 0) {
939 int mdsize, *max_easize;
941 if (*vallen != sizeof(int))
944 if (mdsize > exp->exp_obd->u.cli.cl_max_mds_easize)
945 exp->exp_obd->u.cli.cl_max_mds_easize = mdsize;
947 *max_easize = exp->exp_obd->u.cli.cl_max_mds_easize;
953 static int mdc_statfs(struct obd_device *obd, struct obd_statfs *osfs,
956 struct ptlrpc_request *req;
957 struct obd_statfs *msfs;
958 int rc, size[2] = { sizeof(struct ptlrpc_body), sizeof(*msfs) };
961 /* We could possibly pass max_age in the request (as an absolute
962 * timestamp or a "seconds.usec ago") so the target can avoid doing
963 * extra calls into the filesystem if that isn't necessary (e.g.
964 * during mount that would help a bit). Having relative timestamps
965 * is not so great if request processing is slow, while absolute
966 * timestamps are not ideal because they need time synchronization. */
967 req = ptlrpc_prep_req(obd->u.cli.cl_import, LUSTRE_MDS_VERSION,
968 MDS_STATFS, 1, NULL, NULL);
972 ptlrpc_req_set_repsize(req, 2, size);
974 rc = ptlrpc_queue_wait(req);
979 msfs = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*msfs),
980 lustre_swab_obd_statfs);
982 CERROR("Can't unpack obd_statfs\n");
983 GOTO(out, rc = -EPROTO);
986 memcpy(osfs, msfs, sizeof(*msfs));
989 ptlrpc_req_finished(req);
994 static int mdc_pin(struct obd_export *exp, obd_id ino, __u32 gen, int type,
995 struct obd_client_handle *handle, int flag)
997 struct ptlrpc_request *req;
998 struct mds_body *body;
999 int rc, size[2] = { sizeof(struct ptlrpc_body), sizeof(*body) };
1002 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
1003 MDS_PIN, 2, size, NULL);
1007 body = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF, sizeof(*body));
1008 mdc_pack_fid(&body->fid1, ino, gen, type);
1011 ptlrpc_req_set_repsize(req, 2, size);
1013 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1014 rc = ptlrpc_queue_wait(req);
1015 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1017 CERROR("pin failed: %d\n", rc);
1018 ptlrpc_req_finished(req);
1022 body = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*body),
1023 lustre_swab_mds_body);
1025 ptlrpc_req_finished(req);
1029 memcpy(&handle->och_fh, &body->handle, sizeof(body->handle));
1030 handle->och_magic = OBD_CLIENT_HANDLE_MAGIC;
1032 OBD_ALLOC(handle->och_mod, sizeof(*handle->och_mod));
1033 if (handle->och_mod == NULL) {
1034 DEBUG_REQ(D_ERROR, req, "can't allocate mdc_open_data");
1037 handle->och_mod->mod_open_req = req; /* will be dropped by unpin */
1042 static int mdc_unpin(struct obd_export *exp,
1043 struct obd_client_handle *handle, int flag)
1045 struct ptlrpc_request *req;
1046 struct mds_body *body;
1047 int rc, size[2] = { sizeof(struct ptlrpc_body), sizeof(*body) };
1050 if (handle->och_magic != OBD_CLIENT_HANDLE_MAGIC)
1053 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
1054 MDS_CLOSE, 2, size, NULL);
1058 body = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF, sizeof(*body));
1059 memcpy(&body->handle, &handle->och_fh, sizeof(body->handle));
1062 ptlrpc_req_set_repsize(req, 1, NULL);
1063 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1064 rc = ptlrpc_queue_wait(req);
1065 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1068 CERROR("unpin failed: %d\n", rc);
1070 ptlrpc_req_finished(req);
1071 ptlrpc_req_finished(handle->och_mod->mod_open_req);
1072 OBD_FREE(handle->och_mod, sizeof(*handle->och_mod));
1076 int mdc_sync(struct obd_export *exp, struct ll_fid *fid,
1077 struct ptlrpc_request **request)
1079 struct ptlrpc_request *req;
1080 int size[2] = { sizeof(struct ptlrpc_body), sizeof(struct mds_body) };
1084 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
1085 MDS_SYNC, 2, size, NULL);
1087 RETURN(rc = -ENOMEM);
1089 mdc_pack_req_body(req, REQ_REC_OFF, 0, fid, 0, 0);
1091 ptlrpc_req_set_repsize(req, 2, size);
1093 rc = ptlrpc_queue_wait(req);
1094 if (rc || request == NULL)
1095 ptlrpc_req_finished(req);
1102 static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
1103 enum obd_import_event event)
1107 LASSERT(imp->imp_obd == obd);
1110 case IMP_EVENT_DISCON: {
1113 case IMP_EVENT_INACTIVE: {
1114 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE, NULL);
1117 case IMP_EVENT_INVALIDATE: {
1118 struct ldlm_namespace *ns = obd->obd_namespace;
1120 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
1124 case IMP_EVENT_ACTIVE: {
1125 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE, NULL);
1132 CERROR("Unknown import event %x\n", event);
1138 static int mdc_setup(struct obd_device *obd, obd_count len, void *buf)
1140 struct client_obd *cli = &obd->u.cli;
1141 struct lprocfs_static_vars lvars;
1145 OBD_ALLOC(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1146 if (!cli->cl_rpc_lock)
1148 mdc_init_rpc_lock(cli->cl_rpc_lock);
1152 OBD_ALLOC(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1153 if (!cli->cl_setattr_lock)
1154 GOTO(err_rpc_lock, rc = -ENOMEM);
1155 mdc_init_rpc_lock(cli->cl_setattr_lock);
1157 OBD_ALLOC(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1158 if (!cli->cl_close_lock)
1159 GOTO(err_setattr_lock, rc = -ENOMEM);
1160 mdc_init_rpc_lock(cli->cl_close_lock);
1162 rc = client_obd_setup(obd, len, buf);
1164 GOTO(err_close_lock, rc);
1165 lprocfs_init_vars(mdc, &lvars);
1166 lprocfs_obd_setup(obd, lvars.obd_vars);
1168 rc = obd_llog_init(obd, obd, 0, NULL, NULL);
1171 CERROR("failed to setup llogging subsystems\n");
1177 OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1179 OBD_FREE(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1181 OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1186 /* Initialize the default and maximum LOV EA and cookie sizes. This allows
1187 * us to make MDS RPCs with large enough reply buffers to hold the
1188 * maximum-sized (= maximum striped) EA and cookie without having to
1189 * calculate this (via a call into the LOV + OSCs) each time we make an RPC. */
1190 int mdc_init_ea_size(struct obd_export *mdc_exp, struct obd_export *lov_exp)
1192 struct obd_device *obd = mdc_exp->exp_obd;
1193 struct client_obd *cli = &obd->u.cli;
1194 struct lov_stripe_md lsm = { .lsm_magic = LOV_MAGIC };
1195 struct lov_desc desc;
1196 __u32 valsize = sizeof(desc);
1201 rc = obd_get_info(lov_exp, strlen(KEY_LOVDESC) + 1, KEY_LOVDESC,
1206 stripes = min(desc.ld_tgt_count, (__u32)LOV_MAX_STRIPE_COUNT);
1207 lsm.lsm_stripe_count = stripes;
1208 size = obd_size_diskmd(lov_exp, &lsm);
1210 if (cli->cl_max_mds_easize < size)
1211 cli->cl_max_mds_easize = size;
1213 lsm.lsm_stripe_count = desc.ld_default_stripe_count;
1214 size = obd_size_diskmd(lov_exp, &lsm);
1216 if (cli->cl_default_mds_easize < size)
1217 cli->cl_default_mds_easize = size;
1219 size = stripes * sizeof(struct llog_cookie);
1220 if (cli->cl_max_mds_cookiesize < size)
1221 cli->cl_max_mds_cookiesize = size;
1223 CDEBUG(D_HA, "updating max_mdsize/max_cookiesize: %d/%d\n",
1224 cli->cl_max_mds_easize, cli->cl_max_mds_cookiesize);
1229 static int mdc_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
1235 case OBD_CLEANUP_EARLY:
1236 case OBD_CLEANUP_EXPORTS:
1237 /* If we set up but never connected, the
1238 client import will not have been cleaned. */
1239 if (obd->u.cli.cl_import) {
1240 struct obd_import *imp;
1241 imp = obd->u.cli.cl_import;
1242 CERROR("client import never connected\n");
1243 ptlrpc_invalidate_import(imp);
1244 ptlrpc_free_rq_pool(imp->imp_rq_pool);
1245 class_destroy_import(imp);
1246 obd->u.cli.cl_import = NULL;
1249 case OBD_CLEANUP_SELF_EXP:
1250 rc = obd_llog_finish(obd, 0);
1252 CERROR("failed to cleanup llogging subsystems\n");
1253 case OBD_CLEANUP_OBD:
1259 static int mdc_cleanup(struct obd_device *obd)
1261 struct client_obd *cli = &obd->u.cli;
1263 OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1264 OBD_FREE(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1265 OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1267 lprocfs_obd_cleanup(obd);
1270 return client_obd_cleanup(obd);
1274 static int mdc_llog_init(struct obd_device *obd, struct obd_device *tgt,
1275 int count, struct llog_catid *logid,
1276 struct obd_uuid *uuid)
1278 struct llog_ctxt *ctxt;
1282 rc = llog_setup(obd, LLOG_CONFIG_REPL_CTXT, tgt, 0, NULL,
1285 ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT);
1286 ctxt->loc_imp = obd->u.cli.cl_import;
1289 rc = llog_setup(obd, LLOG_LOVEA_REPL_CTXT, tgt, 0, NULL,
1292 ctxt = llog_get_context(obd, LLOG_LOVEA_REPL_CTXT);
1293 ctxt->loc_imp = obd->u.cli.cl_import;
1299 static int mdc_llog_finish(struct obd_device *obd, int count)
1304 rc = llog_cleanup(llog_get_context(obd, LLOG_LOVEA_REPL_CTXT));
1306 CERROR("can not cleanup LLOG_CONFIG_REPL_CTXT rc %d\n", rc);
1308 rc = llog_cleanup(llog_get_context(obd, LLOG_CONFIG_REPL_CTXT));
1312 static int mdc_process_config(struct obd_device *obd, obd_count len, void *buf)
1314 struct lustre_cfg *lcfg = buf;
1315 struct lprocfs_static_vars lvars;
1318 lprocfs_init_vars(mdc, &lvars);
1320 rc = class_process_proc_param(PARAM_MDC, lvars.obd_vars, lcfg, obd);
1324 struct obd_ops mdc_obd_ops = {
1325 .o_owner = THIS_MODULE,
1326 .o_setup = mdc_setup,
1327 .o_precleanup = mdc_precleanup,
1328 .o_cleanup = mdc_cleanup,
1329 .o_add_conn = client_import_add_conn,
1330 .o_del_conn = client_import_del_conn,
1331 .o_connect = client_connect_import,
1332 .o_disconnect = client_disconnect_export,
1333 .o_iocontrol = mdc_iocontrol,
1334 .o_set_info_async = mdc_set_info_async,
1335 .o_get_info = mdc_get_info,
1336 .o_statfs = mdc_statfs,
1338 .o_unpin = mdc_unpin,
1339 .o_import_event = mdc_import_event,
1340 .o_llog_init = mdc_llog_init,
1341 .o_llog_finish = mdc_llog_finish,
1342 .o_process_config = mdc_process_config,
1345 int __init mdc_init(void)
1348 struct lprocfs_static_vars lvars;
1349 lprocfs_init_vars(mdc, &lvars);
1350 request_module("lquota");
1351 quota_interface = PORTAL_SYMBOL_GET(mdc_quota_interface);
1352 init_obd_quota_ops(quota_interface, &mdc_obd_ops);
1354 rc = class_register_type(&mdc_obd_ops, lvars.module_vars,
1356 if (rc && quota_interface)
1357 PORTAL_SYMBOL_PUT(mdc_quota_interface);
1363 static void /*__exit*/ mdc_exit(void)
1365 if (quota_interface)
1366 PORTAL_SYMBOL_PUT(mdc_quota_interface);
1368 class_unregister_type(LUSTRE_MDC_NAME);
1371 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
1372 MODULE_DESCRIPTION("Lustre Metadata Client");
1373 MODULE_LICENSE("GPL");
1375 EXPORT_SYMBOL(mdc_req2lustre_md);
1376 EXPORT_SYMBOL(mdc_free_lustre_md);
1377 EXPORT_SYMBOL(mdc_change_cbdata);
1378 EXPORT_SYMBOL(mdc_getstatus);
1379 EXPORT_SYMBOL(mdc_getattr);
1380 EXPORT_SYMBOL(mdc_getattr_name);
1381 EXPORT_SYMBOL(mdc_create);
1382 EXPORT_SYMBOL(mdc_unlink);
1383 EXPORT_SYMBOL(mdc_rename);
1384 EXPORT_SYMBOL(mdc_link);
1385 EXPORT_SYMBOL(mdc_readpage);
1386 EXPORT_SYMBOL(mdc_setattr);
1387 EXPORT_SYMBOL(mdc_close);
1388 EXPORT_SYMBOL(mdc_done_writing);
1389 EXPORT_SYMBOL(mdc_sync);
1390 EXPORT_SYMBOL(mdc_set_open_replay_data);
1391 EXPORT_SYMBOL(mdc_clear_open_replay_data);
1392 EXPORT_SYMBOL(mdc_store_inode_generation);
1393 EXPORT_SYMBOL(mdc_init_ea_size);
1394 EXPORT_SYMBOL(mdc_getxattr);
1395 EXPORT_SYMBOL(mdc_setxattr);
1397 module_init(mdc_init);
1398 module_exit(mdc_exit);