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 obd_device *obddev = class_exp2obd(exp);
112 struct mds_body *body;
114 int size[4] = { sizeof(struct ptlrpc_body), sizeof(*body) };
115 int bufcount = 2, rc;
118 /* request message already built */
121 size[bufcount++] = ea_size;
122 CDEBUG(D_INODE, "reserved %u bytes for MD/symlink in packet\n",
126 size[bufcount++] = acl_size;
127 CDEBUG(D_INODE, "reserved %u bytes for ACL\n", acl_size);
130 ptlrpc_req_set_repsize(req, bufcount, size);
132 mdc_enter_request(&obddev->u.cli);
133 rc = ptlrpc_queue_wait(req);
134 mdc_exit_request(&obddev->u.cli);
138 body = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*body),
139 lustre_swab_mds_body);
141 CERROR ("Can't unpack mds_body\n");
145 CDEBUG(D_NET, "mode: %o\n", body->mode);
147 LASSERT_REPSWAB(req, REPLY_REC_OFF + 1);
148 if (body->eadatasize != 0) {
149 /* reply indicates presence of eadata; check it's there... */
150 eadata = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF + 1,
152 if (eadata == NULL) {
153 CERROR ("Missing/short eadata\n");
158 if (body->valid & OBD_MD_FLMODEASIZE) {
159 if (exp->exp_obd->u.cli.cl_max_mds_easize < body->max_mdsize)
160 exp->exp_obd->u.cli.cl_max_mds_easize =
162 if (exp->exp_obd->u.cli.cl_max_mds_cookiesize <
163 body->max_cookiesize)
164 exp->exp_obd->u.cli.cl_max_mds_cookiesize =
165 body->max_cookiesize;
171 int mdc_getattr(struct obd_export *exp, struct ll_fid *fid,
172 obd_valid valid, unsigned int ea_size,
173 struct ptlrpc_request **request)
175 struct ptlrpc_request *req;
176 int size[2] = { sizeof(struct ptlrpc_body), sizeof(struct mds_body) };
177 int acl_size = 0, rc;
180 /* XXX do we need to make another request here? We just did a getattr
181 * to do the lookup in the first place.
183 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
184 MDS_GETATTR, 2, size, NULL);
186 GOTO(out, rc = -ENOMEM);
188 mdc_pack_req_body(req, REQ_REC_OFF, valid, fid, ea_size,
189 MDS_BFLAG_EXT_FLAGS/*request "new" flags(bug 9486)*/);
191 /* currently only root inode will call us with FLACL */
192 if (valid & OBD_MD_FLACL)
193 acl_size = LUSTRE_POSIX_ACL_MAX_SIZE;
195 rc = mdc_getattr_common(exp, ea_size, acl_size, req);
197 ptlrpc_req_finished (req);
205 int mdc_getattr_name(struct obd_export *exp, struct ll_fid *fid,
206 const char *filename, int namelen, unsigned long valid,
207 unsigned int ea_size, struct ptlrpc_request **request)
209 struct ptlrpc_request *req;
210 struct mds_body *body;
211 int rc, size[3] = { sizeof(struct ptlrpc_body), sizeof(*body), namelen};
214 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
215 MDS_GETATTR_NAME, 3, size, NULL);
217 GOTO(out, rc = -ENOMEM);
219 mdc_pack_req_body(req, REQ_REC_OFF, valid, fid, ea_size,
220 MDS_BFLAG_EXT_FLAGS/*request "new" flags(bug 9486)*/);
222 LASSERT(strnlen(filename, namelen) == namelen - 1);
223 memcpy(lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF + 1, namelen),
226 rc = mdc_getattr_common(exp, ea_size, 0, req);
228 ptlrpc_req_finished (req);
237 int mdc_xattr_common(struct obd_export *exp, struct ll_fid *fid,
238 int opcode, obd_valid valid, const char *xattr_name,
239 const char *input, int input_size, int output_size,
240 int flags, struct ptlrpc_request **request)
242 struct obd_device *obddev = class_exp2obd(exp);
243 struct ptlrpc_request *req;
244 int size[4] = { sizeof(struct ptlrpc_body), sizeof(struct mds_body) };
245 // int size[3] = {sizeof(struct mds_body)}, bufcnt = 1;
246 int rc, xattr_namelen = 0, bufcnt = 2, offset;
251 xattr_namelen = strlen(xattr_name) + 1;
252 size[bufcnt++] = xattr_namelen;
256 size[bufcnt++] = input_size;
259 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
260 opcode, bufcnt, size, NULL);
262 GOTO(out, rc = -ENOMEM);
265 mdc_pack_req_body(req, REQ_REC_OFF, valid, fid, output_size, flags);
267 offset = REQ_REC_OFF + 1;
270 tmp = lustre_msg_buf(req->rq_reqmsg, offset++, xattr_namelen);
271 memcpy(tmp, xattr_name, xattr_namelen);
274 tmp = lustre_msg_buf(req->rq_reqmsg, offset++, input_size);
275 memcpy(tmp, input, input_size);
279 if (opcode == MDS_GETXATTR) {
280 size[REPLY_REC_OFF] = sizeof(struct mds_body);
286 /* we do this even output_size is 0, because server is doing that */
287 size[bufcnt++] = output_size;
289 ptlrpc_req_set_repsize(req, bufcnt, size);
292 if (opcode == MDS_SETXATTR)
293 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
295 mdc_enter_request(&obddev->u.cli);
297 rc = ptlrpc_queue_wait(req);
299 if (opcode == MDS_SETXATTR)
300 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
302 mdc_exit_request(&obddev->u.cli);
307 if (opcode == MDS_GETXATTR) {
308 struct mds_body * body = lustre_swab_repbuf(req, REPLY_REC_OFF,
310 lustre_swab_mds_body);
312 CERROR ("Can't unpack mds_body\n");
313 GOTO(err_out, rc = -EPROTO);
320 ptlrpc_req_finished(req);
325 int mdc_setxattr(struct obd_export *exp, struct ll_fid *fid,
326 obd_valid valid, const char *xattr_name,
327 const char *input, int input_size,
328 int output_size, int flags,
329 struct ptlrpc_request **request)
331 return mdc_xattr_common(exp, fid, MDS_SETXATTR, valid, xattr_name,
332 input, input_size, output_size, flags, request);
335 int mdc_getxattr(struct obd_export *exp, struct ll_fid *fid,
336 obd_valid valid, const char *xattr_name,
337 const char *input, int input_size,
338 int output_size, struct ptlrpc_request **request)
340 return mdc_xattr_common(exp, fid, MDS_GETXATTR, valid, xattr_name,
341 input, input_size, output_size, 0, request);
344 /* This should be called with both the request and the reply still packed. */
345 void mdc_store_inode_generation(struct ptlrpc_request *req, int reqoff,
348 struct mds_rec_create *rec = lustre_msg_buf(req->rq_reqmsg, reqoff,
350 struct mds_body *body = lustre_msg_buf(req->rq_repmsg, repoff,
353 LASSERT (rec != NULL);
354 LASSERT (body != NULL);
356 memcpy(&rec->cr_replayfid, &body->fid1, sizeof rec->cr_replayfid);
357 if (body->fid1.id == 0) {
358 DEBUG_REQ(D_ERROR, req, "saving replay request with id = 0 "
359 "gen = %u", body->fid1.generation);
363 DEBUG_REQ(D_INODE, req, "storing generation %u for ino "LPU64,
364 rec->cr_replayfid.generation, rec->cr_replayfid.id);
367 #ifdef CONFIG_FS_POSIX_ACL
369 int mdc_unpack_acl(struct obd_export *exp, struct ptlrpc_request *req,
370 struct lustre_md *md, unsigned int offset)
372 struct mds_body *body = md->body;
373 struct posix_acl *acl;
380 buf = lustre_msg_buf(req->rq_repmsg, offset, body->aclsize);
382 CERROR("aclsize %u, bufcount %u, bufsize %u\n",
383 body->aclsize, lustre_msg_bufcount(req->rq_repmsg),
384 (lustre_msg_bufcount(req->rq_repmsg) <= offset) ?
385 -1 : lustre_msg_buflen(req->rq_repmsg, offset));
389 acl = posix_acl_from_xattr(buf, body->aclsize);
392 CERROR("convert xattr to acl: %d\n", rc);
396 rc = posix_acl_valid(acl);
398 CERROR("validate acl: %d\n", rc);
399 posix_acl_release(acl);
407 #define mdc_unpack_acl(exp, req, md, offset) 0
410 int mdc_req2lustre_md(struct ptlrpc_request *req, int offset,
411 struct obd_export *exp,
412 struct lustre_md *md)
418 memset(md, 0, sizeof(*md));
420 md->body = lustre_msg_buf(req->rq_repmsg, offset, sizeof (*md->body));
421 LASSERT (md->body != NULL);
422 LASSERT_REPSWABBED(req, offset);
425 if (md->body->valid & OBD_MD_FLEASIZE) {
427 struct lov_mds_md *lmm;
429 LASSERT(S_ISREG(md->body->mode));
431 if (md->body->eadatasize == 0) {
432 CERROR ("OBD_MD_FLEASIZE set, but eadatasize 0\n");
435 lmmsize = md->body->eadatasize;
436 lmm = lustre_msg_buf(req->rq_repmsg, offset, lmmsize);
437 LASSERT (lmm != NULL);
438 LASSERT_REPSWABBED(req, offset);
440 rc = obd_unpackmd(exp, &md->lsm, lmm, lmmsize);
444 LASSERT (rc >= sizeof (*md->lsm));
450 /* for ACL, it's possible that FLACL is set but aclsize is zero.
451 * only when aclsize != 0 there's an actual segment for ACL in
454 if ((md->body->valid & OBD_MD_FLACL) && md->body->aclsize) {
455 rc = mdc_unpack_acl(exp, req, md, offset);
465 obd_free_memmd(exp, &md->lsm);
469 void mdc_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
472 obd_free_memmd(exp, &md->lsm);
474 #ifdef CONFIG_FS_POSIX_ACL
476 posix_acl_release(md->posix_acl);
477 md->posix_acl = NULL;
482 static void mdc_commit_open(struct ptlrpc_request *req)
484 struct mdc_open_data *mod = req->rq_cb_data;
488 if (mod->mod_close_req != NULL)
489 mod->mod_close_req->rq_cb_data = NULL;
491 if (mod->mod_och != NULL)
492 mod->mod_och->och_mod = NULL;
494 OBD_FREE(mod, sizeof(*mod));
495 req->rq_cb_data = NULL;
498 static void mdc_replay_open(struct ptlrpc_request *req)
500 struct mdc_open_data *mod = req->rq_cb_data;
501 struct obd_client_handle *och;
502 struct ptlrpc_request *close_req;
503 struct lustre_handle old;
504 struct mds_body *body;
507 body = lustre_swab_repbuf(req, DLM_REPLY_REC_OFF, sizeof(*body),
508 lustre_swab_mds_body);
509 LASSERT (body != NULL);
512 DEBUG_REQ(D_ERROR, req,
513 "can't properly replay without open data");
520 struct lustre_handle *file_fh;
521 LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
522 file_fh = &och->och_fh;
523 CDEBUG(D_RPCTRACE, "updating handle from "LPX64" to "LPX64"\n",
524 file_fh->cookie, body->handle.cookie);
525 memcpy(&old, file_fh, sizeof(old));
526 memcpy(file_fh, &body->handle, sizeof(*file_fh));
529 close_req = mod->mod_close_req;
530 if (close_req != NULL) {
531 struct mds_body *close_body;
532 LASSERT(lustre_msg_get_opc(close_req->rq_reqmsg) == MDS_CLOSE);
533 close_body = lustre_msg_buf(close_req->rq_reqmsg, REQ_REC_OFF,
534 sizeof(*close_body));
536 LASSERT(!memcmp(&old, &close_body->handle, sizeof old));
537 DEBUG_REQ(D_RPCTRACE, close_req, "updating close with new fh");
538 memcpy(&close_body->handle, &body->handle,
539 sizeof(close_body->handle));
545 void mdc_set_open_replay_data(struct obd_client_handle *och,
546 struct ptlrpc_request *open_req)
548 struct mdc_open_data *mod;
549 struct mds_rec_create *rec = lustre_msg_buf(open_req->rq_reqmsg,
552 struct mds_body *body = lustre_msg_buf(open_req->rq_repmsg,
556 /* incoming message in my byte order (it's been swabbed) */
557 LASSERT(rec != NULL);
558 LASSERT_REPSWABBED(open_req, DLM_REPLY_REC_OFF);
559 /* outgoing messages always in my byte order */
560 LASSERT(body != NULL);
563 OBD_ALLOC(mod, sizeof(*mod));
565 DEBUG_REQ(D_ERROR, open_req, "can't allocate mdc_open_data");
569 spin_lock(&open_req->rq_lock);
570 if (!open_req->rq_replay) {
571 OBD_FREE(mod, sizeof(*mod));
572 spin_unlock(&open_req->rq_lock);
578 mod->mod_open_req = open_req;
579 open_req->rq_cb_data = mod;
580 open_req->rq_commit_cb = mdc_commit_open;
581 spin_unlock(&open_req->rq_lock);
584 memcpy(&rec->cr_replayfid, &body->fid1, sizeof rec->cr_replayfid);
585 open_req->rq_replay_cb = mdc_replay_open;
586 if (body->fid1.id == 0) {
587 DEBUG_REQ(D_ERROR, open_req, "saving replay request with "
588 "id = 0 gen = %u", body->fid1.generation);
592 DEBUG_REQ(D_RPCTRACE, open_req, "set up replay data");
595 void mdc_clear_open_replay_data(struct obd_client_handle *och)
597 struct mdc_open_data *mod = och->och_mod;
599 /* Don't free the structure now (it happens in mdc_commit_open, after
600 * we're sure we won't need to fix up the close request in the future),
601 * but make sure that replay doesn't poke at the och, which is about to
603 LASSERT(mod != LP_POISON);
609 static void mdc_commit_close(struct ptlrpc_request *req)
611 struct mdc_open_data *mod = req->rq_cb_data;
612 struct ptlrpc_request *open_req;
613 struct obd_import *imp = req->rq_import;
615 DEBUG_REQ(D_RPCTRACE, req, "close req committed");
619 mod->mod_close_req = NULL;
620 req->rq_cb_data = NULL;
621 req->rq_commit_cb = NULL;
623 open_req = mod->mod_open_req;
624 LASSERT(open_req != NULL);
625 LASSERT(open_req != LP_POISON);
626 LASSERT(open_req->rq_type != LI_POISON);
628 DEBUG_REQ(D_RPCTRACE, open_req, "open req balanced");
629 LASSERT(open_req->rq_transno != 0);
630 LASSERT(open_req->rq_import == imp);
632 /* We no longer want to preserve this for transno-unconditional
634 spin_lock(&open_req->rq_lock);
635 open_req->rq_replay = 0;
636 spin_unlock(&open_req->rq_lock);
639 int mdc_close(struct obd_export *exp, struct obdo *oa,
640 struct obd_client_handle *och, struct ptlrpc_request **request)
642 struct obd_device *obd = class_exp2obd(exp);
643 int reqsize[2] = { sizeof(struct ptlrpc_body),
644 sizeof(struct mds_body) };
645 int rc, repsize[4] = { sizeof(struct ptlrpc_body),
646 sizeof(struct mds_body),
647 obd->u.cli.cl_max_mds_easize,
648 obd->u.cli.cl_max_mds_cookiesize };
649 struct ptlrpc_request *req;
650 struct mdc_open_data *mod;
653 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
654 MDS_CLOSE, 2, reqsize, NULL);
656 GOTO(out, rc = -ENOMEM);
658 /* To avoid a livelock (bug 7034), we need to send CLOSE RPCs to a
659 * portal whose threads are not taking any DLM locks and are therefore
660 * always progressing */
661 /* XXX FIXME bug 249 */
662 req->rq_request_portal = MDS_READPAGE_PORTAL;
664 /* Ensure that this close's handle is fixed up during replay. */
665 LASSERT(och != NULL);
666 LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
668 if (likely(mod != NULL)) {
669 if (mod->mod_open_req->rq_type == LI_POISON) {
670 CERROR("LBUG POISONED open %p!\n", mod->mod_open_req);
672 ptlrpc_req_finished(req);
674 GOTO(out, rc = -EIO);
676 mod->mod_close_req = req;
677 DEBUG_REQ(D_RPCTRACE, mod->mod_open_req, "matched open");
679 CDEBUG(D_RPCTRACE, "couldn't find open req; expecting error\n");
682 mdc_close_pack(req, REQ_REC_OFF, oa, oa->o_valid, och);
684 ptlrpc_req_set_repsize(req, 4, repsize);
685 req->rq_commit_cb = mdc_commit_close;
686 LASSERT(req->rq_cb_data == NULL);
687 req->rq_cb_data = mod;
689 mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
690 rc = ptlrpc_queue_wait(req);
691 mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
693 if (req->rq_repmsg == NULL) {
694 CDEBUG(D_RPCTRACE, "request failed to send: %p, %d\n", req,
697 rc = req->rq_status ? req->rq_status : -EIO;
698 } else if (rc == 0) {
699 rc = lustre_msg_get_status(req->rq_repmsg);
700 if (lustre_msg_get_type(req->rq_repmsg) == PTL_RPC_MSG_ERR) {
701 DEBUG_REQ(D_ERROR, req, "type == PTL_RPC_MSG_ERR, err "
705 } else if (mod == NULL) {
706 CERROR("Unexpected: can't find mdc_open_data, but the "
707 "close succeeded. Please tell CFS.\n");
709 if (!lustre_swab_repbuf(req, REPLY_REC_OFF,
710 sizeof(struct mds_body),
711 lustre_swab_mds_body)) {
712 CERROR("Error unpacking mds_body\n");
720 if (rc != 0 && req && req->rq_commit_cb)
721 req->rq_commit_cb(req);
726 int mdc_done_writing(struct obd_export *exp, struct obdo *obdo)
728 struct ptlrpc_request *req;
729 struct mds_body *body;
730 int rc, size[2] = { sizeof(struct ptlrpc_body), sizeof(*body) };
733 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
734 MDS_DONE_WRITING, 2, size, NULL);
738 body = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF, sizeof(*body));
739 mdc_pack_fid(&body->fid1, obdo->o_id, 0, obdo->o_mode);
740 body->size = obdo->o_size;
741 body->blocks = obdo->o_blocks;
742 body->flags = obdo->o_flags;
743 body->valid = obdo->o_valid;
744 // memcpy(&body->handle, &och->och_fh, sizeof(body->handle));
746 ptlrpc_req_set_repsize(req, 2, size);
748 rc = ptlrpc_queue_wait(req);
749 ptlrpc_req_finished(req);
753 int mdc_readpage(struct obd_export *exp, struct ll_fid *fid, __u64 offset,
754 struct page *page, struct ptlrpc_request **request)
756 struct obd_import *imp = class_exp2cliimp(exp);
757 struct ptlrpc_request *req = NULL;
758 struct ptlrpc_bulk_desc *desc = NULL;
759 struct mds_body *body;
760 int rc, size[2] = { sizeof(struct ptlrpc_body), sizeof(*body) };
763 CDEBUG(D_INODE, "inode: "LPU64"\n", fid->id);
765 req = ptlrpc_prep_req(imp, LUSTRE_MDS_VERSION, MDS_READPAGE, 2, size,
768 GOTO(out, rc = -ENOMEM);
770 /* XXX FIXME bug 249 */
771 req->rq_request_portal = MDS_READPAGE_PORTAL;
773 desc = ptlrpc_prep_bulk_imp(req, 1, BULK_PUT_SINK, MDS_BULK_PORTAL);
775 GOTO(out, rc = -ENOMEM);
776 /* NB req now owns desc and will free it when it gets freed */
778 ptlrpc_prep_bulk_page(desc, page, 0, CFS_PAGE_SIZE);
780 mdc_readdir_pack(req, REQ_REC_OFF, offset, CFS_PAGE_SIZE, fid);
782 ptlrpc_req_set_repsize(req, 2, size);
783 rc = ptlrpc_queue_wait(req);
786 body = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*body),
787 lustre_swab_mds_body);
789 CERROR("Can't unpack mds_body\n");
790 GOTO(out, rc = -EPROTO);
793 if (req->rq_bulk->bd_nob_transferred != CFS_PAGE_SIZE) {
794 CERROR ("Unexpected # bytes transferred: %d"
796 req->rq_bulk->bd_nob_transferred,
798 GOTO (out, rc = -EPROTO);
809 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
810 void *karg, void *uarg)
812 struct obd_device *obd = exp->exp_obd;
813 struct obd_ioctl_data *data = karg;
814 struct obd_import *imp = obd->u.cli.cl_import;
815 struct llog_ctxt *ctxt;
819 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
822 if (!try_module_get(THIS_MODULE)) {
823 CERROR("Can't get module. Is it alive?");
828 case OBD_IOC_CLIENT_RECOVER:
829 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1);
833 case IOC_OSC_SET_ACTIVE:
834 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
836 case OBD_IOC_PARSE: {
837 ctxt = llog_get_context(exp->exp_obd, LLOG_CONFIG_REPL_CTXT);
838 rc = class_config_parse_llog(ctxt, data->ioc_inlbuf1, NULL);
843 case OBD_IOC_LLOG_INFO:
844 case OBD_IOC_LLOG_PRINT: {
845 ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT);
846 rc = llog_ioctl(ctxt, cmd, data);
851 case OBD_IOC_POLL_QUOTACHECK:
852 rc = lquota_poll_check(quota_interface, exp,
853 (struct if_quotacheck *)karg);
856 CERROR("mdc_ioctl(): unrecognised ioctl %#x\n", cmd);
857 GOTO(out, rc = -ENOTTY);
860 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
863 module_put(THIS_MODULE);
869 int mdc_set_info_async(struct obd_export *exp, obd_count keylen,
870 void *key, obd_count vallen, void *val,
871 struct ptlrpc_request_set *set)
873 struct obd_import *imp = class_exp2cliimp(exp);
876 if (KEY_IS(KEY_INIT_RECOV)) {
877 if (vallen != sizeof(int))
879 spin_lock(&imp->imp_lock);
880 imp->imp_initial_recov = *(int *)val;
881 spin_unlock(&imp->imp_lock);
883 CDEBUG(D_HA, "%s: set imp_initial_recov = %d\n",
884 exp->exp_obd->obd_name, imp->imp_initial_recov);
887 /* Turn off initial_recov after we try all backup servers once */
888 if (KEY_IS(KEY_INIT_RECOV_BACKUP)) {
889 if (vallen != sizeof(int))
892 spin_lock(&imp->imp_lock);
893 imp->imp_initial_recov_bk = *(int *)val;
894 if (imp->imp_initial_recov_bk)
895 imp->imp_initial_recov = 1;
896 spin_unlock(&imp->imp_lock);
898 CDEBUG(D_HA, "%s: set imp_initial_recov_bk = %d\n",
899 exp->exp_obd->obd_name, imp->imp_initial_recov_bk);
902 if (KEY_IS("read-only")) {
903 struct ptlrpc_request *req;
904 int size[3] = { sizeof(struct ptlrpc_body), keylen, vallen };
905 char *bufs[3] = { NULL, key, val };
907 if (vallen != sizeof(int))
911 imp->imp_connect_flags_orig |= OBD_CONNECT_RDONLY;
912 imp->imp_connect_data.ocd_connect_flags |=
915 imp->imp_connect_flags_orig &= ~OBD_CONNECT_RDONLY;
916 imp->imp_connect_data.ocd_connect_flags &=
920 req = ptlrpc_prep_req(imp, LUSTRE_MDS_VERSION, MDS_SET_INFO,
925 ptlrpc_req_set_repsize(req, 1, NULL);
928 ptlrpc_set_add_req(set, req);
929 ptlrpc_check_set(set);
931 rc = ptlrpc_queue_wait(req);
932 ptlrpc_req_finished(req);
941 int mdc_get_info(struct obd_export *exp, __u32 keylen, void *key,
942 __u32 *vallen, void *val)
946 if (keylen == strlen("max_easize") &&
947 memcmp(key, "max_easize", strlen("max_easize")) == 0) {
948 int mdsize, *max_easize;
950 if (*vallen != sizeof(int))
953 if (mdsize > exp->exp_obd->u.cli.cl_max_mds_easize)
954 exp->exp_obd->u.cli.cl_max_mds_easize = mdsize;
956 *max_easize = exp->exp_obd->u.cli.cl_max_mds_easize;
962 static int mdc_statfs(struct obd_device *obd, struct obd_statfs *osfs,
965 struct ptlrpc_request *req;
966 struct obd_statfs *msfs;
967 int rc, size[2] = { sizeof(struct ptlrpc_body), sizeof(*msfs) };
970 /* We could possibly pass max_age in the request (as an absolute
971 * timestamp or a "seconds.usec ago") so the target can avoid doing
972 * extra calls into the filesystem if that isn't necessary (e.g.
973 * during mount that would help a bit). Having relative timestamps
974 * is not so great if request processing is slow, while absolute
975 * timestamps are not ideal because they need time synchronization. */
976 req = ptlrpc_prep_req(obd->u.cli.cl_import, LUSTRE_MDS_VERSION,
977 MDS_STATFS, 1, NULL, NULL);
981 ptlrpc_req_set_repsize(req, 2, size);
983 rc = ptlrpc_queue_wait(req);
988 msfs = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*msfs),
989 lustre_swab_obd_statfs);
991 CERROR("Can't unpack obd_statfs\n");
992 GOTO(out, rc = -EPROTO);
995 memcpy(osfs, msfs, sizeof(*msfs));
998 ptlrpc_req_finished(req);
1003 static int mdc_pin(struct obd_export *exp, obd_id ino, __u32 gen, int type,
1004 struct obd_client_handle *handle, int flag)
1006 struct ptlrpc_request *req;
1007 struct mds_body *body;
1008 int rc, size[2] = { sizeof(struct ptlrpc_body), sizeof(*body) };
1011 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
1012 MDS_PIN, 2, size, NULL);
1016 body = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF, sizeof(*body));
1017 mdc_pack_fid(&body->fid1, ino, gen, type);
1020 ptlrpc_req_set_repsize(req, 2, size);
1022 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1023 rc = ptlrpc_queue_wait(req);
1024 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1026 CERROR("pin failed: %d\n", rc);
1027 ptlrpc_req_finished(req);
1031 body = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*body),
1032 lustre_swab_mds_body);
1034 ptlrpc_req_finished(req);
1038 memcpy(&handle->och_fh, &body->handle, sizeof(body->handle));
1039 handle->och_magic = OBD_CLIENT_HANDLE_MAGIC;
1041 OBD_ALLOC(handle->och_mod, sizeof(*handle->och_mod));
1042 if (handle->och_mod == NULL) {
1043 DEBUG_REQ(D_ERROR, req, "can't allocate mdc_open_data");
1046 handle->och_mod->mod_open_req = req; /* will be dropped by unpin */
1051 static int mdc_unpin(struct obd_export *exp,
1052 struct obd_client_handle *handle, int flag)
1054 struct ptlrpc_request *req;
1055 struct mds_body *body;
1056 int rc, size[2] = { sizeof(struct ptlrpc_body), sizeof(*body) };
1059 if (handle->och_magic != OBD_CLIENT_HANDLE_MAGIC)
1062 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
1063 MDS_CLOSE, 2, size, NULL);
1067 body = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF, sizeof(*body));
1068 memcpy(&body->handle, &handle->och_fh, sizeof(body->handle));
1071 ptlrpc_req_set_repsize(req, 1, NULL);
1072 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1073 rc = ptlrpc_queue_wait(req);
1074 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1077 CERROR("unpin failed: %d\n", rc);
1079 ptlrpc_req_finished(req);
1080 ptlrpc_req_finished(handle->och_mod->mod_open_req);
1081 OBD_FREE(handle->och_mod, sizeof(*handle->och_mod));
1085 int mdc_sync(struct obd_export *exp, struct ll_fid *fid,
1086 struct ptlrpc_request **request)
1088 struct ptlrpc_request *req;
1089 int size[2] = { sizeof(struct ptlrpc_body), sizeof(struct mds_body) };
1093 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
1094 MDS_SYNC, 2, size, NULL);
1096 RETURN(rc = -ENOMEM);
1098 mdc_pack_req_body(req, REQ_REC_OFF, 0, fid, 0, 0);
1100 ptlrpc_req_set_repsize(req, 2, size);
1102 rc = ptlrpc_queue_wait(req);
1103 if (rc || request == NULL)
1104 ptlrpc_req_finished(req);
1111 static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
1112 enum obd_import_event event)
1116 LASSERT(imp->imp_obd == obd);
1119 case IMP_EVENT_DISCON: {
1120 ptlrpc_import_setasync(imp, -obd->obd_namespace->ns_max_unused);
1123 case IMP_EVENT_INACTIVE: {
1124 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE, NULL);
1127 case IMP_EVENT_INVALIDATE: {
1128 struct ldlm_namespace *ns = obd->obd_namespace;
1130 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
1134 case IMP_EVENT_ACTIVE: {
1135 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE, NULL);
1139 ptlrpc_import_setasync(imp, obd->obd_namespace->ns_max_unused);
1143 CERROR("Unknown import event %x\n", event);
1149 static int mdc_setup(struct obd_device *obd, obd_count len, void *buf)
1151 struct client_obd *cli = &obd->u.cli;
1152 struct lprocfs_static_vars lvars;
1156 OBD_ALLOC(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1157 if (!cli->cl_rpc_lock)
1159 mdc_init_rpc_lock(cli->cl_rpc_lock);
1163 OBD_ALLOC(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1164 if (!cli->cl_setattr_lock)
1165 GOTO(err_rpc_lock, rc = -ENOMEM);
1166 mdc_init_rpc_lock(cli->cl_setattr_lock);
1168 OBD_ALLOC(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1169 if (!cli->cl_close_lock)
1170 GOTO(err_setattr_lock, rc = -ENOMEM);
1171 mdc_init_rpc_lock(cli->cl_close_lock);
1173 rc = client_obd_setup(obd, len, buf);
1175 GOTO(err_close_lock, rc);
1176 lprocfs_init_vars(mdc, &lvars);
1177 lprocfs_obd_setup(obd, lvars.obd_vars);
1179 rc = obd_llog_init(obd, obd, 0, NULL, NULL);
1182 CERROR("failed to setup llogging subsystems\n");
1188 OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1190 OBD_FREE(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1192 OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1197 /* Initialize the default and maximum LOV EA and cookie sizes. This allows
1198 * us to make MDS RPCs with large enough reply buffers to hold the
1199 * maximum-sized (= maximum striped) EA and cookie without having to
1200 * calculate this (via a call into the LOV + OSCs) each time we make an RPC. */
1201 int mdc_init_ea_size(struct obd_export *mdc_exp, struct obd_export *lov_exp)
1203 struct obd_device *obd = mdc_exp->exp_obd;
1204 struct client_obd *cli = &obd->u.cli;
1205 struct lov_stripe_md lsm = { .lsm_magic = LOV_MAGIC };
1206 struct lov_desc desc;
1207 __u32 valsize = sizeof(desc);
1212 rc = obd_get_info(lov_exp, strlen(KEY_LOVDESC) + 1, KEY_LOVDESC,
1217 stripes = min(desc.ld_tgt_count, (__u32)LOV_MAX_STRIPE_COUNT);
1218 lsm.lsm_stripe_count = stripes;
1219 size = obd_size_diskmd(lov_exp, &lsm);
1221 if (cli->cl_max_mds_easize < size)
1222 cli->cl_max_mds_easize = size;
1224 lsm.lsm_stripe_count = desc.ld_default_stripe_count;
1225 size = obd_size_diskmd(lov_exp, &lsm);
1227 if (cli->cl_default_mds_easize < size)
1228 cli->cl_default_mds_easize = size;
1230 size = stripes * sizeof(struct llog_cookie);
1231 if (cli->cl_max_mds_cookiesize < size)
1232 cli->cl_max_mds_cookiesize = size;
1234 CDEBUG(D_HA, "updating max_mdsize/max_cookiesize: %d/%d\n",
1235 cli->cl_max_mds_easize, cli->cl_max_mds_cookiesize);
1240 static int mdc_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
1246 case OBD_CLEANUP_EARLY:
1247 case OBD_CLEANUP_EXPORTS:
1248 /* If we set up but never connected, the
1249 client import will not have been cleaned. */
1250 if (obd->u.cli.cl_import) {
1251 struct obd_import *imp;
1252 imp = obd->u.cli.cl_import;
1253 CERROR("client import never connected\n");
1254 ptlrpc_invalidate_import(imp);
1255 ptlrpc_free_rq_pool(imp->imp_rq_pool);
1256 class_destroy_import(imp);
1257 obd->u.cli.cl_import = NULL;
1260 case OBD_CLEANUP_SELF_EXP:
1261 rc = obd_llog_finish(obd, 0);
1263 CERROR("failed to cleanup llogging subsystems\n");
1264 case OBD_CLEANUP_OBD:
1270 static int mdc_cleanup(struct obd_device *obd)
1272 struct client_obd *cli = &obd->u.cli;
1274 OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1275 OBD_FREE(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1276 OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1278 lprocfs_obd_cleanup(obd);
1281 return client_obd_cleanup(obd);
1285 static int mdc_llog_init(struct obd_device *obd, struct obd_device *tgt,
1286 int count, struct llog_catid *logid,
1287 struct obd_uuid *uuid)
1289 struct llog_ctxt *ctxt;
1293 rc = llog_setup(obd, LLOG_CONFIG_REPL_CTXT, tgt, 0, NULL,
1296 ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT);
1297 ctxt->loc_imp = obd->u.cli.cl_import;
1298 llog_ctxt_put(ctxt);
1301 rc = llog_setup(obd, LLOG_LOVEA_REPL_CTXT, tgt, 0, NULL,
1304 ctxt = llog_get_context(obd, LLOG_LOVEA_REPL_CTXT);
1305 ctxt->loc_imp = obd->u.cli.cl_import;
1306 llog_ctxt_put(ctxt);
1312 static int mdc_llog_finish(struct obd_device *obd, int count)
1317 rc = llog_cleanup(llog_get_context(obd, LLOG_LOVEA_REPL_CTXT));
1319 CERROR("can not cleanup LLOG_CONFIG_REPL_CTXT rc %d\n", rc);
1321 rc = llog_cleanup(llog_get_context(obd, LLOG_CONFIG_REPL_CTXT));
1325 static int mdc_process_config(struct obd_device *obd, obd_count len, void *buf)
1327 struct lustre_cfg *lcfg = buf;
1328 struct lprocfs_static_vars lvars;
1331 lprocfs_init_vars(mdc, &lvars);
1333 rc = class_process_proc_param(PARAM_MDC, lvars.obd_vars, lcfg, obd);
1337 struct obd_ops mdc_obd_ops = {
1338 .o_owner = THIS_MODULE,
1339 .o_setup = mdc_setup,
1340 .o_precleanup = mdc_precleanup,
1341 .o_cleanup = mdc_cleanup,
1342 .o_add_conn = client_import_add_conn,
1343 .o_del_conn = client_import_del_conn,
1344 .o_connect = client_connect_import,
1345 .o_disconnect = client_disconnect_export,
1346 .o_iocontrol = mdc_iocontrol,
1347 .o_set_info_async = mdc_set_info_async,
1348 .o_get_info = mdc_get_info,
1349 .o_statfs = mdc_statfs,
1351 .o_unpin = mdc_unpin,
1352 .o_import_event = mdc_import_event,
1353 .o_llog_init = mdc_llog_init,
1354 .o_llog_finish = mdc_llog_finish,
1355 .o_process_config = mdc_process_config,
1358 int __init mdc_init(void)
1361 struct lprocfs_static_vars lvars;
1362 lprocfs_init_vars(mdc, &lvars);
1363 request_module("lquota");
1364 quota_interface = PORTAL_SYMBOL_GET(mdc_quota_interface);
1365 init_obd_quota_ops(quota_interface, &mdc_obd_ops);
1367 rc = class_register_type(&mdc_obd_ops, lvars.module_vars,
1369 if (rc && quota_interface)
1370 PORTAL_SYMBOL_PUT(mdc_quota_interface);
1376 static void /*__exit*/ mdc_exit(void)
1378 if (quota_interface)
1379 PORTAL_SYMBOL_PUT(mdc_quota_interface);
1381 class_unregister_type(LUSTRE_MDC_NAME);
1384 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
1385 MODULE_DESCRIPTION("Lustre Metadata Client");
1386 MODULE_LICENSE("GPL");
1388 EXPORT_SYMBOL(mdc_req2lustre_md);
1389 EXPORT_SYMBOL(mdc_free_lustre_md);
1390 EXPORT_SYMBOL(mdc_change_cbdata);
1391 EXPORT_SYMBOL(mdc_getstatus);
1392 EXPORT_SYMBOL(mdc_getattr);
1393 EXPORT_SYMBOL(mdc_getattr_name);
1394 EXPORT_SYMBOL(mdc_create);
1395 EXPORT_SYMBOL(mdc_unlink);
1396 EXPORT_SYMBOL(mdc_rename);
1397 EXPORT_SYMBOL(mdc_link);
1398 EXPORT_SYMBOL(mdc_readpage);
1399 EXPORT_SYMBOL(mdc_setattr);
1400 EXPORT_SYMBOL(mdc_close);
1401 EXPORT_SYMBOL(mdc_done_writing);
1402 EXPORT_SYMBOL(mdc_sync);
1403 EXPORT_SYMBOL(mdc_set_open_replay_data);
1404 EXPORT_SYMBOL(mdc_clear_open_replay_data);
1405 EXPORT_SYMBOL(mdc_store_inode_generation);
1406 EXPORT_SYMBOL(mdc_init_ea_size);
1407 EXPORT_SYMBOL(mdc_getxattr);
1408 EXPORT_SYMBOL(mdc_setxattr);
1410 module_init(mdc_init);
1411 module_exit(mdc_exit);