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");
563 mod->mod_open_req = open_req;
564 open_req->rq_cb_data = mod;
565 open_req->rq_commit_cb = mdc_commit_open;
568 memcpy(&rec->cr_replayfid, &body->fid1, sizeof rec->cr_replayfid);
569 open_req->rq_replay_cb = mdc_replay_open;
570 if (body->fid1.id == 0) {
571 DEBUG_REQ(D_ERROR, open_req, "saving replay request with "
572 "id = 0 gen = %u", body->fid1.generation);
576 DEBUG_REQ(D_HA, open_req, "set up replay data");
579 void mdc_clear_open_replay_data(struct obd_client_handle *och)
581 struct mdc_open_data *mod = och->och_mod;
583 /* Don't free the structure now (it happens in mdc_commit_open, after
584 * we're sure we won't need to fix up the close request in the future),
585 * but make sure that replay doesn't poke at the och, which is about to
587 LASSERT(mod != LP_POISON);
593 static void mdc_commit_close(struct ptlrpc_request *req)
595 struct mdc_open_data *mod = req->rq_cb_data;
596 struct ptlrpc_request *open_req;
597 struct obd_import *imp = req->rq_import;
599 DEBUG_REQ(D_HA, req, "close req committed");
603 mod->mod_close_req = NULL;
604 req->rq_cb_data = NULL;
605 req->rq_commit_cb = NULL;
607 open_req = mod->mod_open_req;
608 LASSERT(open_req != NULL);
609 LASSERT(open_req != LP_POISON);
610 LASSERT(open_req->rq_type != LI_POISON);
612 DEBUG_REQ(D_HA, open_req, "open req balanced");
613 LASSERT(open_req->rq_transno != 0);
614 LASSERT(open_req->rq_import == imp);
616 /* We no longer want to preserve this for transno-unconditional
618 spin_lock(&open_req->rq_lock);
619 open_req->rq_replay = 0;
620 spin_unlock(&open_req->rq_lock);
623 int mdc_close(struct obd_export *exp, struct obdo *oa,
624 struct obd_client_handle *och, struct ptlrpc_request **request)
626 struct obd_device *obd = class_exp2obd(exp);
627 int reqsize[2] = { sizeof(struct ptlrpc_body),
628 sizeof(struct mds_body) };
629 int rc, repsize[4] = { sizeof(struct ptlrpc_body),
630 sizeof(struct mds_body),
631 obd->u.cli.cl_max_mds_easize,
632 obd->u.cli.cl_max_mds_cookiesize };
633 struct ptlrpc_request *req;
634 struct mdc_open_data *mod;
637 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
638 MDS_CLOSE, 2, reqsize, NULL);
640 GOTO(out, rc = -ENOMEM);
642 /* To avoid a livelock (bug 7034), we need to send CLOSE RPCs to a
643 * portal whose threads are not taking any DLM locks and are therefore
644 * always progressing */
645 /* XXX FIXME bug 249 */
646 req->rq_request_portal = MDS_READPAGE_PORTAL;
648 /* Ensure that this close's handle is fixed up during replay. */
649 LASSERT(och != NULL);
650 LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
652 if (likely(mod != NULL)) {
653 if (mod->mod_open_req->rq_type == LI_POISON) {
654 /* FIXME This should be an ASSERT, but until we
655 figure out why it can be poisoned here, give
656 a reasonable return. bug 6155 */
657 CERROR("LBUG POISONED open %p!\n", mod->mod_open_req);
658 ptlrpc_req_finished(req);
660 GOTO(out, rc = -EIO);
662 mod->mod_close_req = req;
663 DEBUG_REQ(D_HA, mod->mod_open_req, "matched open");
665 CDEBUG(D_HA, "couldn't find open req; expecting close error\n");
668 mdc_close_pack(req, REQ_REC_OFF, oa, oa->o_valid, och);
670 ptlrpc_req_set_repsize(req, 4, repsize);
671 req->rq_commit_cb = mdc_commit_close;
672 LASSERT(req->rq_cb_data == NULL);
673 req->rq_cb_data = mod;
675 mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
676 rc = ptlrpc_queue_wait(req);
677 mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
679 if (req->rq_repmsg == NULL) {
680 CDEBUG(D_HA, "request failed to send: %p, %d\n", req,
683 rc = req->rq_status ? req->rq_status : -EIO;
684 } else if (rc == 0) {
685 rc = lustre_msg_get_status(req->rq_repmsg);
686 if (lustre_msg_get_type(req->rq_repmsg) == PTL_RPC_MSG_ERR) {
687 DEBUG_REQ(D_ERROR, req, "type == PTL_RPC_MSG_ERR, err "
691 } else if (mod == NULL) {
692 CERROR("Unexpected: can't find mdc_open_data, but the "
693 "close succeeded. Please tell CFS.\n");
695 if (!lustre_swab_repbuf(req, REPLY_REC_OFF,
696 sizeof(struct mds_body),
697 lustre_swab_mds_body)) {
698 CERROR("Error unpacking mds_body\n");
706 if (rc != 0 && req && req->rq_commit_cb)
707 req->rq_commit_cb(req);
712 int mdc_done_writing(struct obd_export *exp, struct obdo *obdo)
714 struct ptlrpc_request *req;
715 struct mds_body *body;
716 int rc, size[2] = { sizeof(struct ptlrpc_body), sizeof(*body) };
719 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
720 MDS_DONE_WRITING, 2, size, NULL);
724 body = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF, sizeof(*body));
725 mdc_pack_fid(&body->fid1, obdo->o_id, 0, obdo->o_mode);
726 body->size = obdo->o_size;
727 body->blocks = obdo->o_blocks;
728 body->flags = obdo->o_flags;
729 body->valid = obdo->o_valid;
730 // memcpy(&body->handle, &och->och_fh, sizeof(body->handle));
732 ptlrpc_req_set_repsize(req, 2, size);
734 rc = ptlrpc_queue_wait(req);
735 ptlrpc_req_finished(req);
739 int mdc_readpage(struct obd_export *exp, struct ll_fid *fid, __u64 offset,
740 struct page *page, struct ptlrpc_request **request)
742 struct obd_import *imp = class_exp2cliimp(exp);
743 struct ptlrpc_request *req = NULL;
744 struct ptlrpc_bulk_desc *desc = NULL;
745 struct mds_body *body;
746 int rc, size[2] = { sizeof(struct ptlrpc_body), sizeof(*body) };
749 CDEBUG(D_INODE, "inode: "LPU64"\n", fid->id);
751 req = ptlrpc_prep_req(imp, LUSTRE_MDS_VERSION, MDS_READPAGE, 2, size,
754 GOTO(out, rc = -ENOMEM);
756 /* XXX FIXME bug 249 */
757 req->rq_request_portal = MDS_READPAGE_PORTAL;
759 desc = ptlrpc_prep_bulk_imp(req, 1, BULK_PUT_SINK, MDS_BULK_PORTAL);
761 GOTO(out, rc = -ENOMEM);
762 /* NB req now owns desc and will free it when it gets freed */
764 ptlrpc_prep_bulk_page(desc, page, 0, CFS_PAGE_SIZE);
766 mdc_readdir_pack(req, REQ_REC_OFF, offset, CFS_PAGE_SIZE, fid);
768 ptlrpc_req_set_repsize(req, 2, size);
769 rc = ptlrpc_queue_wait(req);
772 body = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*body),
773 lustre_swab_mds_body);
775 CERROR("Can't unpack mds_body\n");
776 GOTO(out, rc = -EPROTO);
779 if (req->rq_bulk->bd_nob_transferred != CFS_PAGE_SIZE) {
780 CERROR ("Unexpected # bytes transferred: %d"
782 req->rq_bulk->bd_nob_transferred,
784 GOTO (out, rc = -EPROTO);
795 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
796 void *karg, void *uarg)
798 struct obd_device *obd = exp->exp_obd;
799 struct obd_ioctl_data *data = karg;
800 struct obd_import *imp = obd->u.cli.cl_import;
801 struct llog_ctxt *ctxt;
805 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
808 if (!try_module_get(THIS_MODULE)) {
809 CERROR("Can't get module. Is it alive?");
814 case OBD_IOC_CLIENT_RECOVER:
815 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1);
819 case IOC_OSC_SET_ACTIVE:
820 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
822 case OBD_IOC_PARSE: {
823 ctxt = llog_get_context(exp->exp_obd, LLOG_CONFIG_REPL_CTXT);
824 rc = class_config_parse_llog(ctxt, data->ioc_inlbuf1, NULL);
828 case OBD_IOC_LLOG_INFO:
829 case OBD_IOC_LLOG_PRINT: {
830 ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT);
831 rc = llog_ioctl(ctxt, cmd, data);
836 case OBD_IOC_POLL_QUOTACHECK:
837 rc = lquota_poll_check(quota_interface, exp,
838 (struct if_quotacheck *)karg);
841 CERROR("mdc_ioctl(): unrecognised ioctl %#x\n", cmd);
842 GOTO(out, rc = -ENOTTY);
845 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
848 module_put(THIS_MODULE);
854 int mdc_set_info_async(struct obd_export *exp, obd_count keylen,
855 void *key, obd_count vallen, void *val,
856 struct ptlrpc_request_set *set)
858 struct obd_import *imp = class_exp2cliimp(exp);
861 if (KEY_IS(KEY_INIT_RECOV)) {
862 if (vallen != sizeof(int))
864 imp->imp_initial_recov = *(int *)val;
865 CDEBUG(D_HA, "%s: set imp_initial_recov = %d\n",
866 exp->exp_obd->obd_name, imp->imp_initial_recov);
869 /* Turn off initial_recov after we try all backup servers once */
870 if (KEY_IS(KEY_INIT_RECOV_BACKUP)) {
871 if (vallen != sizeof(int))
873 imp->imp_initial_recov_bk = *(int *)val;
874 if (imp->imp_initial_recov_bk)
875 imp->imp_initial_recov = 1;
876 CDEBUG(D_HA, "%s: set imp_initial_recov_bk = %d\n",
877 exp->exp_obd->obd_name, imp->imp_initial_recov_bk);
880 if (KEY_IS("read-only")) {
881 struct ptlrpc_request *req;
882 int size[3] = { sizeof(struct ptlrpc_body), keylen, vallen };
883 char *bufs[3] = { NULL, key, val };
885 if (vallen != sizeof(int))
889 imp->imp_connect_flags_orig |= OBD_CONNECT_RDONLY;
890 imp->imp_connect_data.ocd_connect_flags |=
893 imp->imp_connect_flags_orig &= ~OBD_CONNECT_RDONLY;
894 imp->imp_connect_data.ocd_connect_flags &=
898 req = ptlrpc_prep_req(imp, LUSTRE_MDS_VERSION, MDS_SET_INFO,
903 ptlrpc_req_set_repsize(req, 1, NULL);
906 ptlrpc_set_add_req(set, req);
907 ptlrpc_check_set(set);
909 rc = ptlrpc_queue_wait(req);
910 ptlrpc_req_finished(req);
919 int mdc_get_info(struct obd_export *exp, __u32 keylen, void *key,
920 __u32 *vallen, void *val)
924 if (keylen == strlen("max_easize") &&
925 memcmp(key, "max_easize", strlen("max_easize")) == 0) {
926 int mdsize, *max_easize;
928 if (*vallen != sizeof(int))
931 if (mdsize > exp->exp_obd->u.cli.cl_max_mds_easize)
932 exp->exp_obd->u.cli.cl_max_mds_easize = mdsize;
934 *max_easize = exp->exp_obd->u.cli.cl_max_mds_easize;
940 static int mdc_statfs(struct obd_device *obd, struct obd_statfs *osfs,
943 struct ptlrpc_request *req;
944 struct obd_statfs *msfs;
945 int rc, size[2] = { sizeof(struct ptlrpc_body), sizeof(*msfs) };
948 /* We could possibly pass max_age in the request (as an absolute
949 * timestamp or a "seconds.usec ago") so the target can avoid doing
950 * extra calls into the filesystem if that isn't necessary (e.g.
951 * during mount that would help a bit). Having relative timestamps
952 * is not so great if request processing is slow, while absolute
953 * timestamps are not ideal because they need time synchronization. */
954 req = ptlrpc_prep_req(obd->u.cli.cl_import, LUSTRE_MDS_VERSION,
955 MDS_STATFS, 1, NULL, NULL);
959 ptlrpc_req_set_repsize(req, 2, size);
961 rc = ptlrpc_queue_wait(req);
966 msfs = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*msfs),
967 lustre_swab_obd_statfs);
969 CERROR("Can't unpack obd_statfs\n");
970 GOTO(out, rc = -EPROTO);
973 memcpy(osfs, msfs, sizeof(*msfs));
976 ptlrpc_req_finished(req);
981 static int mdc_pin(struct obd_export *exp, obd_id ino, __u32 gen, int type,
982 struct obd_client_handle *handle, int flag)
984 struct ptlrpc_request *req;
985 struct mds_body *body;
986 int rc, size[2] = { sizeof(struct ptlrpc_body), sizeof(*body) };
989 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
990 MDS_PIN, 2, size, NULL);
994 body = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF, sizeof(*body));
995 mdc_pack_fid(&body->fid1, ino, gen, type);
998 ptlrpc_req_set_repsize(req, 2, size);
1000 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1001 rc = ptlrpc_queue_wait(req);
1002 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1004 CERROR("pin failed: %d\n", rc);
1005 ptlrpc_req_finished(req);
1009 body = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*body),
1010 lustre_swab_mds_body);
1012 ptlrpc_req_finished(req);
1016 memcpy(&handle->och_fh, &body->handle, sizeof(body->handle));
1017 handle->och_magic = OBD_CLIENT_HANDLE_MAGIC;
1019 OBD_ALLOC(handle->och_mod, sizeof(*handle->och_mod));
1020 if (handle->och_mod == NULL) {
1021 DEBUG_REQ(D_ERROR, req, "can't allocate mdc_open_data");
1024 handle->och_mod->mod_open_req = req; /* will be dropped by unpin */
1029 static int mdc_unpin(struct obd_export *exp,
1030 struct obd_client_handle *handle, int flag)
1032 struct ptlrpc_request *req;
1033 struct mds_body *body;
1034 int rc, size[2] = { sizeof(struct ptlrpc_body), sizeof(*body) };
1037 if (handle->och_magic != OBD_CLIENT_HANDLE_MAGIC)
1040 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
1041 MDS_CLOSE, 2, size, NULL);
1045 body = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF, sizeof(*body));
1046 memcpy(&body->handle, &handle->och_fh, sizeof(body->handle));
1049 ptlrpc_req_set_repsize(req, 1, NULL);
1050 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1051 rc = ptlrpc_queue_wait(req);
1052 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1055 CERROR("unpin failed: %d\n", rc);
1057 ptlrpc_req_finished(req);
1058 ptlrpc_req_finished(handle->och_mod->mod_open_req);
1059 OBD_FREE(handle->och_mod, sizeof(*handle->och_mod));
1063 int mdc_sync(struct obd_export *exp, struct ll_fid *fid,
1064 struct ptlrpc_request **request)
1066 struct ptlrpc_request *req;
1067 int size[2] = { sizeof(struct ptlrpc_body), sizeof(struct mds_body) };
1071 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
1072 MDS_SYNC, 2, size, NULL);
1074 RETURN(rc = -ENOMEM);
1076 mdc_pack_req_body(req, REQ_REC_OFF, 0, fid, 0, 0);
1078 ptlrpc_req_set_repsize(req, 2, size);
1080 rc = ptlrpc_queue_wait(req);
1081 if (rc || request == NULL)
1082 ptlrpc_req_finished(req);
1089 static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
1090 enum obd_import_event event)
1094 LASSERT(imp->imp_obd == obd);
1097 case IMP_EVENT_DISCON: {
1100 case IMP_EVENT_INACTIVE: {
1101 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE, NULL);
1104 case IMP_EVENT_INVALIDATE: {
1105 struct ldlm_namespace *ns = obd->obd_namespace;
1107 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
1111 case IMP_EVENT_ACTIVE: {
1112 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE, NULL);
1119 CERROR("Unknown import event %x\n", event);
1125 static int mdc_setup(struct obd_device *obd, obd_count len, void *buf)
1127 struct client_obd *cli = &obd->u.cli;
1128 struct lprocfs_static_vars lvars;
1132 OBD_ALLOC(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1133 if (!cli->cl_rpc_lock)
1135 mdc_init_rpc_lock(cli->cl_rpc_lock);
1139 OBD_ALLOC(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1140 if (!cli->cl_setattr_lock)
1141 GOTO(err_rpc_lock, rc = -ENOMEM);
1142 mdc_init_rpc_lock(cli->cl_setattr_lock);
1144 OBD_ALLOC(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1145 if (!cli->cl_close_lock)
1146 GOTO(err_setattr_lock, rc = -ENOMEM);
1147 mdc_init_rpc_lock(cli->cl_close_lock);
1149 rc = client_obd_setup(obd, len, buf);
1151 GOTO(err_close_lock, rc);
1152 lprocfs_init_vars(mdc, &lvars);
1153 lprocfs_obd_setup(obd, lvars.obd_vars);
1155 rc = obd_llog_init(obd, obd, 0, NULL, NULL);
1158 CERROR("failed to setup llogging subsystems\n");
1164 OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1166 OBD_FREE(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1168 OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1173 /* Initialize the default and maximum LOV EA and cookie sizes. This allows
1174 * us to make MDS RPCs with large enough reply buffers to hold the
1175 * maximum-sized (= maximum striped) EA and cookie without having to
1176 * calculate this (via a call into the LOV + OSCs) each time we make an RPC. */
1177 int mdc_init_ea_size(struct obd_export *mdc_exp, struct obd_export *lov_exp)
1179 struct obd_device *obd = mdc_exp->exp_obd;
1180 struct client_obd *cli = &obd->u.cli;
1181 struct lov_stripe_md lsm = { .lsm_magic = LOV_MAGIC };
1182 struct lov_desc desc;
1183 __u32 valsize = sizeof(desc);
1188 rc = obd_get_info(lov_exp, strlen(KEY_LOVDESC) + 1, KEY_LOVDESC,
1193 stripes = min(desc.ld_tgt_count, (__u32)LOV_MAX_STRIPE_COUNT);
1194 lsm.lsm_stripe_count = stripes;
1195 size = obd_size_diskmd(lov_exp, &lsm);
1197 if (cli->cl_max_mds_easize < size)
1198 cli->cl_max_mds_easize = size;
1200 lsm.lsm_stripe_count = desc.ld_default_stripe_count;
1201 size = obd_size_diskmd(lov_exp, &lsm);
1203 if (cli->cl_default_mds_easize < size)
1204 cli->cl_default_mds_easize = size;
1206 size = stripes * sizeof(struct llog_cookie);
1207 if (cli->cl_max_mds_cookiesize < size)
1208 cli->cl_max_mds_cookiesize = size;
1210 CDEBUG(D_HA, "updating max_mdsize/max_cookiesize: %d/%d\n",
1211 cli->cl_max_mds_easize, cli->cl_max_mds_cookiesize);
1216 static int mdc_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
1222 case OBD_CLEANUP_EARLY:
1223 case OBD_CLEANUP_EXPORTS:
1224 /* If we set up but never connected, the
1225 client import will not have been cleaned. */
1226 if (obd->u.cli.cl_import) {
1227 struct obd_import *imp;
1228 imp = obd->u.cli.cl_import;
1229 CERROR("client import never connected\n");
1230 ptlrpc_invalidate_import(imp);
1231 ptlrpc_free_rq_pool(imp->imp_rq_pool);
1232 class_destroy_import(imp);
1233 obd->u.cli.cl_import = NULL;
1236 case OBD_CLEANUP_SELF_EXP:
1237 rc = obd_llog_finish(obd, 0);
1239 CERROR("failed to cleanup llogging subsystems\n");
1240 case OBD_CLEANUP_OBD:
1246 static int mdc_cleanup(struct obd_device *obd)
1248 struct client_obd *cli = &obd->u.cli;
1250 OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1251 OBD_FREE(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1252 OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1254 lprocfs_obd_cleanup(obd);
1257 return client_obd_cleanup(obd);
1261 static int mdc_llog_init(struct obd_device *obd, struct obd_device *tgt,
1262 int count, struct llog_catid *logid,
1263 struct obd_uuid *uuid)
1265 struct llog_ctxt *ctxt;
1269 rc = llog_setup(obd, LLOG_CONFIG_REPL_CTXT, tgt, 0, NULL,
1272 ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT);
1273 ctxt->loc_imp = obd->u.cli.cl_import;
1276 rc = llog_setup(obd, LLOG_LOVEA_REPL_CTXT, tgt, 0, NULL,
1279 ctxt = llog_get_context(obd, LLOG_LOVEA_REPL_CTXT);
1280 ctxt->loc_imp = obd->u.cli.cl_import;
1286 static int mdc_llog_finish(struct obd_device *obd, int count)
1291 rc = llog_cleanup(llog_get_context(obd, LLOG_LOVEA_REPL_CTXT));
1293 CERROR("can not cleanup LLOG_CONFIG_REPL_CTXT rc %d\n", rc);
1295 rc = llog_cleanup(llog_get_context(obd, LLOG_CONFIG_REPL_CTXT));
1299 static int mdc_process_config(struct obd_device *obd, obd_count len, void *buf)
1301 struct lustre_cfg *lcfg = buf;
1302 struct lprocfs_static_vars lvars;
1305 lprocfs_init_vars(mdc, &lvars);
1307 rc = class_process_proc_param(PARAM_MDC, lvars.obd_vars, lcfg, obd);
1311 struct obd_ops mdc_obd_ops = {
1312 .o_owner = THIS_MODULE,
1313 .o_setup = mdc_setup,
1314 .o_precleanup = mdc_precleanup,
1315 .o_cleanup = mdc_cleanup,
1316 .o_add_conn = client_import_add_conn,
1317 .o_del_conn = client_import_del_conn,
1318 .o_connect = client_connect_import,
1319 .o_disconnect = client_disconnect_export,
1320 .o_iocontrol = mdc_iocontrol,
1321 .o_set_info_async = mdc_set_info_async,
1322 .o_get_info = mdc_get_info,
1323 .o_statfs = mdc_statfs,
1325 .o_unpin = mdc_unpin,
1326 .o_import_event = mdc_import_event,
1327 .o_llog_init = mdc_llog_init,
1328 .o_llog_finish = mdc_llog_finish,
1329 .o_process_config = mdc_process_config,
1332 int __init mdc_init(void)
1335 struct lprocfs_static_vars lvars;
1336 lprocfs_init_vars(mdc, &lvars);
1337 request_module("lquota");
1338 quota_interface = PORTAL_SYMBOL_GET(mdc_quota_interface);
1339 init_obd_quota_ops(quota_interface, &mdc_obd_ops);
1341 rc = class_register_type(&mdc_obd_ops, lvars.module_vars,
1343 if (rc && quota_interface)
1344 PORTAL_SYMBOL_PUT(mdc_quota_interface);
1350 static void /*__exit*/ mdc_exit(void)
1352 if (quota_interface)
1353 PORTAL_SYMBOL_PUT(mdc_quota_interface);
1355 class_unregister_type(LUSTRE_MDC_NAME);
1358 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
1359 MODULE_DESCRIPTION("Lustre Metadata Client");
1360 MODULE_LICENSE("GPL");
1362 EXPORT_SYMBOL(mdc_req2lustre_md);
1363 EXPORT_SYMBOL(mdc_free_lustre_md);
1364 EXPORT_SYMBOL(mdc_change_cbdata);
1365 EXPORT_SYMBOL(mdc_getstatus);
1366 EXPORT_SYMBOL(mdc_getattr);
1367 EXPORT_SYMBOL(mdc_getattr_name);
1368 EXPORT_SYMBOL(mdc_create);
1369 EXPORT_SYMBOL(mdc_unlink);
1370 EXPORT_SYMBOL(mdc_rename);
1371 EXPORT_SYMBOL(mdc_link);
1372 EXPORT_SYMBOL(mdc_readpage);
1373 EXPORT_SYMBOL(mdc_setattr);
1374 EXPORT_SYMBOL(mdc_close);
1375 EXPORT_SYMBOL(mdc_done_writing);
1376 EXPORT_SYMBOL(mdc_sync);
1377 EXPORT_SYMBOL(mdc_set_open_replay_data);
1378 EXPORT_SYMBOL(mdc_clear_open_replay_data);
1379 EXPORT_SYMBOL(mdc_store_inode_generation);
1380 EXPORT_SYMBOL(mdc_init_ea_size);
1381 EXPORT_SYMBOL(mdc_getxattr);
1382 EXPORT_SYMBOL(mdc_setxattr);
1384 module_init(mdc_init);
1385 module_exit(mdc_exit);