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 lustre_set_rep_swabbed(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(lustre_rep_swabbed(req, offset));
425 if (md->body->valid & OBD_MD_FLEASIZE) {
427 struct lov_mds_md *lmm;
429 if (!S_ISREG(md->body->mode)) {
430 CERROR("OBD_MD_FLEASIZE set, should be a regular file, "
432 GOTO(err_out, rc = -EPROTO);
435 if (md->body->eadatasize == 0) {
436 CERROR ("OBD_MD_FLEASIZE set, but eadatasize 0\n");
437 GOTO(err_out, rc = -EPROTO);
439 lmmsize = md->body->eadatasize;
440 lmm = lustre_msg_buf(req->rq_repmsg, offset, lmmsize);
442 CERROR ("incorrect message: lmm == 0\n");
443 GOTO(err_out, rc = -EPROTO);
445 LASSERT(lustre_rep_swabbed(req, offset));
447 rc = obd_unpackmd(exp, &md->lsm, lmm, lmmsize);
451 if (rc < sizeof(*md->lsm)) {
452 CERROR ("lsm size too small: rc < sizeof (*md->lsm) "
453 "(%d < "LPSZ")\n", rc, sizeof(*md->lsm));
454 GOTO(err_out, rc = -EPROTO);
461 if (md->body->valid & OBD_MD_FLDIREA) {
462 if(!S_ISDIR(md->body->mode)) {
463 CERROR("OBD_MD_FLDIREA set, should be a directory, but "
465 GOTO(err_out, rc = -EPROTO);
470 /* for ACL, it's possible that FLACL is set but aclsize is zero.
471 * only when aclsize != 0 there's an actual segment for ACL in
474 if ((md->body->valid & OBD_MD_FLACL) && md->body->aclsize) {
475 rc = mdc_unpack_acl(exp, req, md, offset);
485 obd_free_memmd(exp, &md->lsm);
489 void mdc_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
492 obd_free_memmd(exp, &md->lsm);
494 #ifdef CONFIG_FS_POSIX_ACL
496 posix_acl_release(md->posix_acl);
497 md->posix_acl = NULL;
502 static void mdc_commit_open(struct ptlrpc_request *req)
504 struct mdc_open_data *mod = req->rq_cb_data;
508 if (mod->mod_close_req != NULL)
509 mod->mod_close_req->rq_cb_data = NULL;
511 if (mod->mod_och != NULL)
512 mod->mod_och->och_mod = NULL;
514 OBD_FREE(mod, sizeof(*mod));
515 req->rq_cb_data = NULL;
518 static void mdc_replay_open(struct ptlrpc_request *req)
520 struct mdc_open_data *mod = req->rq_cb_data;
521 struct obd_client_handle *och;
522 struct ptlrpc_request *close_req;
523 struct lustre_handle old;
524 struct mds_body *body;
527 body = lustre_swab_repbuf(req, DLM_REPLY_REC_OFF, sizeof(*body),
528 lustre_swab_mds_body);
529 LASSERT (body != NULL);
532 DEBUG_REQ(D_ERROR, req,
533 "can't properly replay without open data");
540 struct lustre_handle *file_fh;
541 LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
542 file_fh = &och->och_fh;
543 CDEBUG(D_RPCTRACE, "updating handle from "LPX64" to "LPX64"\n",
544 file_fh->cookie, body->handle.cookie);
545 memcpy(&old, file_fh, sizeof(old));
546 memcpy(file_fh, &body->handle, sizeof(*file_fh));
549 close_req = mod->mod_close_req;
550 if (close_req != NULL) {
551 struct mds_body *close_body;
552 LASSERT(lustre_msg_get_opc(close_req->rq_reqmsg) == MDS_CLOSE);
553 close_body = lustre_msg_buf(close_req->rq_reqmsg, REQ_REC_OFF,
554 sizeof(*close_body));
556 LASSERT(!memcmp(&old, &close_body->handle, sizeof old));
557 DEBUG_REQ(D_RPCTRACE, close_req, "updating close with new fh");
558 memcpy(&close_body->handle, &body->handle,
559 sizeof(close_body->handle));
565 void mdc_set_open_replay_data(struct obd_client_handle *och,
566 struct ptlrpc_request *open_req)
568 struct mdc_open_data *mod;
569 struct mds_rec_create *rec = lustre_msg_buf(open_req->rq_reqmsg,
572 struct mds_body *body = lustre_msg_buf(open_req->rq_repmsg,
576 /* If request is not eligible for replay, just bail out */
577 if (!open_req->rq_replay)
580 /* incoming message in my byte order (it's been swabbed) */
581 LASSERT(rec != NULL);
582 LASSERT(lustre_rep_swabbed(open_req, DLM_REPLY_REC_OFF));
583 /* outgoing messages always in my byte order */
584 LASSERT(body != NULL);
587 OBD_ALLOC(mod, sizeof(*mod));
589 DEBUG_REQ(D_ERROR, open_req, "can't allocate mdc_open_data");
593 spin_lock(&open_req->rq_lock);
596 mod->mod_open_req = open_req;
597 open_req->rq_cb_data = mod;
598 open_req->rq_commit_cb = mdc_commit_open;
599 spin_unlock(&open_req->rq_lock);
602 memcpy(&rec->cr_replayfid, &body->fid1, sizeof rec->cr_replayfid);
603 open_req->rq_replay_cb = mdc_replay_open;
604 if (body->fid1.id == 0) {
605 DEBUG_REQ(D_ERROR, open_req, "saving replay request with "
606 "id = 0 gen = %u", body->fid1.generation);
610 DEBUG_REQ(D_RPCTRACE, open_req, "set up replay data");
613 void mdc_clear_open_replay_data(struct obd_client_handle *och)
615 struct mdc_open_data *mod = och->och_mod;
617 /* Don't free the structure now (it happens in mdc_commit_open, after
618 * we're sure we won't need to fix up the close request in the future),
619 * but make sure that replay doesn't poke at the och, which is about to
621 LASSERT(mod != LP_POISON);
627 static void mdc_commit_close(struct ptlrpc_request *req)
629 struct mdc_open_data *mod = req->rq_cb_data;
630 struct ptlrpc_request *open_req;
631 struct obd_import *imp = req->rq_import;
633 DEBUG_REQ(D_RPCTRACE, req, "close req committed");
637 mod->mod_close_req = NULL;
638 req->rq_cb_data = NULL;
639 req->rq_commit_cb = NULL;
641 open_req = mod->mod_open_req;
642 LASSERT(open_req != NULL);
643 LASSERT(open_req != LP_POISON);
644 LASSERT(open_req->rq_type != LI_POISON);
646 DEBUG_REQ(D_RPCTRACE, open_req, "open req balanced");
647 LASSERT(open_req->rq_transno != 0);
648 LASSERT(open_req->rq_import == imp);
650 /* We no longer want to preserve this for transno-unconditional
652 spin_lock(&open_req->rq_lock);
653 open_req->rq_replay = 0;
654 spin_unlock(&open_req->rq_lock);
657 int mdc_close(struct obd_export *exp, struct obdo *oa,
658 struct obd_client_handle *och, struct ptlrpc_request **request)
660 struct obd_device *obd = class_exp2obd(exp);
661 int reqsize[2] = { sizeof(struct ptlrpc_body),
662 sizeof(struct mds_body) };
663 int rc, repsize[4] = { sizeof(struct ptlrpc_body),
664 sizeof(struct mds_body),
665 obd->u.cli.cl_max_mds_easize,
666 obd->u.cli.cl_max_mds_cookiesize };
667 struct ptlrpc_request *req;
668 struct mdc_open_data *mod;
671 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
672 MDS_CLOSE, 2, reqsize, NULL);
674 GOTO(out, rc = -ENOMEM);
676 /* To avoid a livelock (bug 7034), we need to send CLOSE RPCs to a
677 * portal whose threads are not taking any DLM locks and are therefore
678 * always progressing */
679 req->rq_request_portal = MDS_READPAGE_PORTAL;
680 ptlrpc_at_set_req_timeout(req);
682 /* Ensure that this close's handle is fixed up during replay. */
683 LASSERT(och != NULL);
684 LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
686 if (likely(mod != NULL)) {
687 if (mod->mod_open_req->rq_type == LI_POISON) {
688 CERROR("LBUG POISONED open %p!\n", mod->mod_open_req);
690 ptlrpc_req_finished(req);
692 GOTO(out, rc = -EIO);
694 mod->mod_close_req = req;
695 DEBUG_REQ(D_RPCTRACE, mod->mod_open_req, "matched open");
697 CDEBUG(D_RPCTRACE, "couldn't find open req; expecting error\n");
700 mdc_close_pack(req, REQ_REC_OFF, oa, oa->o_valid, och);
702 ptlrpc_req_set_repsize(req, 4, repsize);
703 req->rq_commit_cb = mdc_commit_close;
704 LASSERT(req->rq_cb_data == NULL);
705 req->rq_cb_data = mod;
707 mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
708 rc = ptlrpc_queue_wait(req);
709 mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
711 if (req->rq_repmsg == NULL) {
712 CDEBUG(D_RPCTRACE, "request failed to send: %p, %d\n", req,
715 rc = req->rq_status ? req->rq_status : -EIO;
716 } else if (rc == 0) {
717 rc = lustre_msg_get_status(req->rq_repmsg);
718 if (lustre_msg_get_type(req->rq_repmsg) == PTL_RPC_MSG_ERR) {
719 DEBUG_REQ(D_ERROR, req, "type == PTL_RPC_MSG_ERR, err "
723 } else if (mod == NULL) {
724 CERROR("Unexpected: can't find mdc_open_data, but the "
725 "close succeeded. Please tell CFS.\n");
727 if (!lustre_swab_repbuf(req, REPLY_REC_OFF,
728 sizeof(struct mds_body),
729 lustre_swab_mds_body)) {
730 CERROR("Error unpacking mds_body\n");
738 if (rc != 0 && req && req->rq_commit_cb)
739 req->rq_commit_cb(req);
744 int mdc_done_writing(struct obd_export *exp, struct obdo *obdo)
746 struct ptlrpc_request *req;
747 struct mds_body *body;
748 int rc, size[2] = { sizeof(struct ptlrpc_body), sizeof(*body) };
751 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
752 MDS_DONE_WRITING, 2, size, NULL);
756 body = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF, sizeof(*body));
757 mdc_pack_fid(&body->fid1, obdo->o_id, 0, obdo->o_mode);
758 body->size = obdo->o_size;
759 body->blocks = obdo->o_blocks;
760 body->flags = obdo->o_flags;
761 body->valid = obdo->o_valid;
762 // memcpy(&body->handle, &och->och_fh, sizeof(body->handle));
764 ptlrpc_req_set_repsize(req, 2, size);
766 rc = ptlrpc_queue_wait(req);
767 ptlrpc_req_finished(req);
771 int mdc_readpage(struct obd_export *exp, struct ll_fid *fid, __u64 offset,
772 struct page *page, struct ptlrpc_request **request)
774 struct obd_import *imp = class_exp2cliimp(exp);
775 struct ptlrpc_request *req = NULL;
776 struct ptlrpc_bulk_desc *desc = NULL;
777 struct mds_body *body;
778 int rc, size[2] = { sizeof(struct ptlrpc_body), sizeof(*body) };
781 CDEBUG(D_INODE, "inode: "LPU64"\n", fid->id);
783 req = ptlrpc_prep_req(imp, LUSTRE_MDS_VERSION, MDS_READPAGE, 2, size,
786 GOTO(out, rc = -ENOMEM);
788 req->rq_request_portal = MDS_READPAGE_PORTAL;
789 ptlrpc_at_set_req_timeout(req);
791 desc = ptlrpc_prep_bulk_imp(req, 1, BULK_PUT_SINK, MDS_BULK_PORTAL);
793 GOTO(out, rc = -ENOMEM);
794 /* NB req now owns desc and will free it when it gets freed */
796 ptlrpc_prep_bulk_page(desc, page, 0, CFS_PAGE_SIZE);
798 mdc_readdir_pack(req, REQ_REC_OFF, offset, CFS_PAGE_SIZE, fid);
800 ptlrpc_req_set_repsize(req, 2, size);
801 rc = ptlrpc_queue_wait(req);
804 body = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*body),
805 lustre_swab_mds_body);
807 CERROR("Can't unpack mds_body\n");
808 GOTO(out, rc = -EPROTO);
811 if (req->rq_bulk->bd_nob_transferred != CFS_PAGE_SIZE) {
812 CERROR ("Unexpected # bytes transferred: %d"
814 req->rq_bulk->bd_nob_transferred,
816 GOTO (out, rc = -EPROTO);
827 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
828 void *karg, void *uarg)
830 struct obd_device *obd = exp->exp_obd;
831 struct obd_ioctl_data *data = karg;
832 struct obd_import *imp = obd->u.cli.cl_import;
833 struct llog_ctxt *ctxt;
837 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
840 if (!try_module_get(THIS_MODULE)) {
841 CERROR("Can't get module. Is it alive?");
846 case OBD_IOC_CLIENT_RECOVER:
847 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1);
851 case IOC_OSC_SET_ACTIVE:
852 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
854 case OBD_IOC_PARSE: {
855 ctxt = llog_get_context(exp->exp_obd, LLOG_CONFIG_REPL_CTXT);
856 rc = class_config_parse_llog(ctxt, data->ioc_inlbuf1, NULL);
861 case OBD_IOC_LLOG_INFO:
862 case OBD_IOC_LLOG_PRINT: {
863 ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT);
864 rc = llog_ioctl(ctxt, cmd, data);
869 case OBD_IOC_POLL_QUOTACHECK:
870 rc = lquota_poll_check(quota_interface, exp,
871 (struct if_quotacheck *)karg);
874 CERROR("mdc_ioctl(): unrecognised ioctl %#x\n", cmd);
875 GOTO(out, rc = -ENOTTY);
878 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
881 module_put(THIS_MODULE);
887 int mdc_set_info_async(struct obd_export *exp, obd_count keylen,
888 void *key, obd_count vallen, void *val,
889 struct ptlrpc_request_set *set)
891 struct obd_import *imp = class_exp2cliimp(exp);
894 if (KEY_IS(KEY_INIT_RECOV)) {
895 if (vallen != sizeof(int))
897 spin_lock(&imp->imp_lock);
898 imp->imp_initial_recov = *(int *)val;
899 spin_unlock(&imp->imp_lock);
901 CDEBUG(D_HA, "%s: set imp_initial_recov = %d\n",
902 exp->exp_obd->obd_name, imp->imp_initial_recov);
905 /* Turn off initial_recov after we try all backup servers once */
906 if (KEY_IS(KEY_INIT_RECOV_BACKUP)) {
907 if (vallen != sizeof(int))
910 spin_lock(&imp->imp_lock);
911 imp->imp_initial_recov_bk = *(int *)val;
912 if (imp->imp_initial_recov_bk)
913 imp->imp_initial_recov = 1;
914 spin_unlock(&imp->imp_lock);
916 CDEBUG(D_HA, "%s: set imp_initial_recov_bk = %d\n",
917 exp->exp_obd->obd_name, imp->imp_initial_recov_bk);
920 if (KEY_IS("read-only")) {
921 struct ptlrpc_request *req;
922 int size[3] = { sizeof(struct ptlrpc_body), keylen, vallen };
923 char *bufs[3] = { NULL, key, val };
925 if (vallen != sizeof(int))
929 imp->imp_connect_flags_orig |= OBD_CONNECT_RDONLY;
930 imp->imp_connect_data.ocd_connect_flags |=
933 imp->imp_connect_flags_orig &= ~OBD_CONNECT_RDONLY;
934 imp->imp_connect_data.ocd_connect_flags &=
938 req = ptlrpc_prep_req(imp, LUSTRE_MDS_VERSION, MDS_SET_INFO,
943 ptlrpc_req_set_repsize(req, 1, NULL);
946 ptlrpc_set_add_req(set, req);
947 ptlrpc_check_set(set);
949 rc = ptlrpc_queue_wait(req);
950 ptlrpc_req_finished(req);
959 int mdc_get_info(struct obd_export *exp, __u32 keylen, void *key,
960 __u32 *vallen, void *val)
964 if (keylen == strlen("max_easize") &&
965 memcmp(key, "max_easize", strlen("max_easize")) == 0) {
966 int mdsize, *max_easize;
968 if (*vallen != sizeof(int))
971 if (mdsize > exp->exp_obd->u.cli.cl_max_mds_easize)
972 exp->exp_obd->u.cli.cl_max_mds_easize = mdsize;
974 *max_easize = exp->exp_obd->u.cli.cl_max_mds_easize;
980 static int mdc_statfs(struct obd_device *obd, struct obd_statfs *osfs,
983 struct ptlrpc_request *req;
984 struct obd_statfs *msfs;
985 int rc, size[2] = { sizeof(struct ptlrpc_body), sizeof(*msfs) };
988 /* We could possibly pass max_age in the request (as an absolute
989 * timestamp or a "seconds.usec ago") so the target can avoid doing
990 * extra calls into the filesystem if that isn't necessary (e.g.
991 * during mount that would help a bit). Having relative timestamps
992 * is not so great if request processing is slow, while absolute
993 * timestamps are not ideal because they need time synchronization. */
994 req = ptlrpc_prep_req(obd->u.cli.cl_import, LUSTRE_MDS_VERSION,
995 MDS_STATFS, 1, NULL, NULL);
999 ptlrpc_req_set_repsize(req, 2, size);
1001 rc = ptlrpc_queue_wait(req);
1006 msfs = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*msfs),
1007 lustre_swab_obd_statfs);
1009 CERROR("Can't unpack obd_statfs\n");
1010 GOTO(out, rc = -EPROTO);
1013 memcpy(osfs, msfs, sizeof(*msfs));
1016 ptlrpc_req_finished(req);
1021 static int mdc_pin(struct obd_export *exp, obd_id ino, __u32 gen, int type,
1022 struct obd_client_handle *handle, int flag)
1024 struct ptlrpc_request *req;
1025 struct mds_body *body;
1026 int rc, size[2] = { sizeof(struct ptlrpc_body), sizeof(*body) };
1029 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
1030 MDS_PIN, 2, size, NULL);
1034 body = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF, sizeof(*body));
1035 mdc_pack_fid(&body->fid1, ino, gen, type);
1038 ptlrpc_req_set_repsize(req, 2, size);
1040 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1041 rc = ptlrpc_queue_wait(req);
1042 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1044 CERROR("pin failed: %d\n", rc);
1045 ptlrpc_req_finished(req);
1049 body = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*body),
1050 lustre_swab_mds_body);
1052 ptlrpc_req_finished(req);
1056 memcpy(&handle->och_fh, &body->handle, sizeof(body->handle));
1057 handle->och_magic = OBD_CLIENT_HANDLE_MAGIC;
1059 OBD_ALLOC(handle->och_mod, sizeof(*handle->och_mod));
1060 if (handle->och_mod == NULL) {
1061 DEBUG_REQ(D_ERROR, req, "can't allocate mdc_open_data");
1064 handle->och_mod->mod_open_req = req; /* will be dropped by unpin */
1069 static int mdc_unpin(struct obd_export *exp,
1070 struct obd_client_handle *handle, int flag)
1072 struct ptlrpc_request *req;
1073 struct mds_body *body;
1074 int rc, size[2] = { sizeof(struct ptlrpc_body), sizeof(*body) };
1077 if (handle->och_magic != OBD_CLIENT_HANDLE_MAGIC)
1080 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
1081 MDS_CLOSE, 2, size, NULL);
1085 body = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF, sizeof(*body));
1086 memcpy(&body->handle, &handle->och_fh, sizeof(body->handle));
1089 ptlrpc_req_set_repsize(req, 1, NULL);
1090 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1091 rc = ptlrpc_queue_wait(req);
1092 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1095 CERROR("unpin failed: %d\n", rc);
1097 ptlrpc_req_finished(req);
1098 ptlrpc_req_finished(handle->och_mod->mod_open_req);
1099 OBD_FREE(handle->och_mod, sizeof(*handle->och_mod));
1103 int mdc_sync(struct obd_export *exp, struct ll_fid *fid,
1104 struct ptlrpc_request **request)
1106 struct ptlrpc_request *req;
1107 int size[2] = { sizeof(struct ptlrpc_body), sizeof(struct mds_body) };
1111 req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
1112 MDS_SYNC, 2, size, NULL);
1114 RETURN(rc = -ENOMEM);
1116 mdc_pack_req_body(req, REQ_REC_OFF, 0, fid, 0, 0);
1118 ptlrpc_req_set_repsize(req, 2, size);
1120 rc = ptlrpc_queue_wait(req);
1121 if (rc || request == NULL)
1122 ptlrpc_req_finished(req);
1129 static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
1130 enum obd_import_event event)
1134 LASSERT(imp->imp_obd == obd);
1137 case IMP_EVENT_DISCON: {
1138 ptlrpc_import_setasync(imp, -obd->obd_namespace->ns_max_unused);
1141 case IMP_EVENT_INACTIVE: {
1142 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE, NULL);
1145 case IMP_EVENT_INVALIDATE: {
1146 struct ldlm_namespace *ns = obd->obd_namespace;
1148 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
1152 case IMP_EVENT_ACTIVE: {
1153 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE, NULL);
1157 ptlrpc_import_setasync(imp, obd->obd_namespace->ns_max_unused);
1161 CERROR("Unknown import event %x\n", event);
1167 static int mdc_setup(struct obd_device *obd, obd_count len, void *buf)
1169 struct client_obd *cli = &obd->u.cli;
1170 struct lprocfs_static_vars lvars = { 0 };
1174 OBD_ALLOC(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1175 if (!cli->cl_rpc_lock)
1177 mdc_init_rpc_lock(cli->cl_rpc_lock);
1181 OBD_ALLOC(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1182 if (!cli->cl_setattr_lock)
1183 GOTO(err_rpc_lock, rc = -ENOMEM);
1184 mdc_init_rpc_lock(cli->cl_setattr_lock);
1186 OBD_ALLOC(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1187 if (!cli->cl_close_lock)
1188 GOTO(err_setattr_lock, rc = -ENOMEM);
1189 mdc_init_rpc_lock(cli->cl_close_lock);
1191 rc = client_obd_setup(obd, len, buf);
1193 GOTO(err_close_lock, rc);
1194 lprocfs_mdc_init_vars(&lvars);
1195 if (lprocfs_obd_setup(obd, lvars.obd_vars) == 0)
1196 ptlrpc_lprocfs_register_obd(obd);
1198 rc = obd_llog_init(obd, obd, 0, NULL, NULL);
1201 CERROR("failed to setup llogging subsystems\n");
1207 OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1209 OBD_FREE(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1211 OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1216 /* Initialize the default and maximum LOV EA and cookie sizes. This allows
1217 * us to make MDS RPCs with large enough reply buffers to hold the
1218 * maximum-sized (= maximum striped) EA and cookie without having to
1219 * calculate this (via a call into the LOV + OSCs) each time we make an RPC. */
1220 int mdc_init_ea_size(struct obd_export *mdc_exp, struct obd_export *lov_exp)
1222 struct obd_device *obd = mdc_exp->exp_obd;
1223 struct client_obd *cli = &obd->u.cli;
1224 struct lov_stripe_md lsm = { .lsm_magic = LOV_MAGIC };
1225 struct lov_desc desc;
1226 __u32 valsize = sizeof(desc);
1231 rc = obd_get_info(lov_exp, strlen(KEY_LOVDESC) + 1, KEY_LOVDESC,
1236 stripes = min(desc.ld_tgt_count, (__u32)LOV_MAX_STRIPE_COUNT);
1237 lsm.lsm_stripe_count = stripes;
1238 size = obd_size_diskmd(lov_exp, &lsm);
1240 if (cli->cl_max_mds_easize < size)
1241 cli->cl_max_mds_easize = size;
1243 lsm.lsm_stripe_count = desc.ld_default_stripe_count;
1244 size = obd_size_diskmd(lov_exp, &lsm);
1246 if (cli->cl_default_mds_easize < size)
1247 cli->cl_default_mds_easize = size;
1249 size = stripes * sizeof(struct llog_cookie);
1250 if (cli->cl_max_mds_cookiesize < size)
1251 cli->cl_max_mds_cookiesize = size;
1253 CDEBUG(D_HA, "updating max_mdsize/max_cookiesize: %d/%d\n",
1254 cli->cl_max_mds_easize, cli->cl_max_mds_cookiesize);
1259 static int mdc_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
1265 case OBD_CLEANUP_EARLY:
1266 case OBD_CLEANUP_EXPORTS:
1267 /* If we set up but never connected, the
1268 client import will not have been cleaned. */
1269 if (obd->u.cli.cl_import) {
1270 struct obd_import *imp;
1271 imp = obd->u.cli.cl_import;
1272 CERROR("client import never connected\n");
1273 ptlrpc_invalidate_import(imp);
1274 ptlrpc_free_rq_pool(imp->imp_rq_pool);
1275 class_destroy_import(imp);
1276 obd->u.cli.cl_import = NULL;
1279 case OBD_CLEANUP_SELF_EXP:
1280 rc = obd_llog_finish(obd, 0);
1282 CERROR("failed to cleanup llogging subsystems\n");
1283 case OBD_CLEANUP_OBD:
1289 static int mdc_cleanup(struct obd_device *obd)
1291 struct client_obd *cli = &obd->u.cli;
1293 OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1294 OBD_FREE(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1295 OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1297 ptlrpc_lprocfs_unregister_obd(obd);
1298 lprocfs_obd_cleanup(obd);
1301 return client_obd_cleanup(obd);
1305 static int mdc_llog_init(struct obd_device *obd, struct obd_device *tgt,
1306 int count, struct llog_catid *logid,
1307 struct obd_uuid *uuid)
1309 struct llog_ctxt *ctxt;
1313 rc = llog_setup(obd, LLOG_CONFIG_REPL_CTXT, tgt, 0, NULL,
1316 ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT);
1317 ctxt->loc_imp = obd->u.cli.cl_import;
1318 llog_ctxt_put(ctxt);
1321 rc = llog_setup(obd, LLOG_LOVEA_REPL_CTXT, tgt, 0, NULL,
1324 ctxt = llog_get_context(obd, LLOG_LOVEA_REPL_CTXT);
1325 ctxt->loc_imp = obd->u.cli.cl_import;
1326 llog_ctxt_put(ctxt);
1332 static int mdc_llog_finish(struct obd_device *obd, int count)
1337 rc = llog_cleanup(llog_get_context(obd, LLOG_LOVEA_REPL_CTXT));
1339 CERROR("can not cleanup LLOG_CONFIG_REPL_CTXT rc %d\n", rc);
1341 rc = llog_cleanup(llog_get_context(obd, LLOG_CONFIG_REPL_CTXT));
1345 static int mdc_process_config(struct obd_device *obd, obd_count len, void *buf)
1347 struct lustre_cfg *lcfg = buf;
1348 struct lprocfs_static_vars lvars = { 0 };
1351 lprocfs_mdc_init_vars(&lvars);
1353 rc = class_process_proc_param(PARAM_MDC, lvars.obd_vars, lcfg, obd);
1357 struct obd_ops mdc_obd_ops = {
1358 .o_owner = THIS_MODULE,
1359 .o_setup = mdc_setup,
1360 .o_precleanup = mdc_precleanup,
1361 .o_cleanup = mdc_cleanup,
1362 .o_add_conn = client_import_add_conn,
1363 .o_del_conn = client_import_del_conn,
1364 .o_connect = client_connect_import,
1365 .o_disconnect = client_disconnect_export,
1366 .o_iocontrol = mdc_iocontrol,
1367 .o_set_info_async = mdc_set_info_async,
1368 .o_get_info = mdc_get_info,
1369 .o_statfs = mdc_statfs,
1371 .o_unpin = mdc_unpin,
1372 .o_import_event = mdc_import_event,
1373 .o_llog_init = mdc_llog_init,
1374 .o_llog_finish = mdc_llog_finish,
1375 .o_process_config = mdc_process_config,
1378 int __init mdc_init(void)
1381 struct lprocfs_static_vars lvars = { 0 };
1382 lprocfs_mdc_init_vars(&lvars);
1383 request_module("lquota");
1384 quota_interface = PORTAL_SYMBOL_GET(mdc_quota_interface);
1385 init_obd_quota_ops(quota_interface, &mdc_obd_ops);
1387 rc = class_register_type(&mdc_obd_ops, lvars.module_vars,
1389 if (rc && quota_interface)
1390 PORTAL_SYMBOL_PUT(mdc_quota_interface);
1396 static void /*__exit*/ mdc_exit(void)
1398 if (quota_interface)
1399 PORTAL_SYMBOL_PUT(mdc_quota_interface);
1401 class_unregister_type(LUSTRE_MDC_NAME);
1404 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
1405 MODULE_DESCRIPTION("Lustre Metadata Client");
1406 MODULE_LICENSE("GPL");
1408 EXPORT_SYMBOL(mdc_req2lustre_md);
1409 EXPORT_SYMBOL(mdc_free_lustre_md);
1410 EXPORT_SYMBOL(mdc_change_cbdata);
1411 EXPORT_SYMBOL(mdc_getstatus);
1412 EXPORT_SYMBOL(mdc_getattr);
1413 EXPORT_SYMBOL(mdc_getattr_name);
1414 EXPORT_SYMBOL(mdc_create);
1415 EXPORT_SYMBOL(mdc_unlink);
1416 EXPORT_SYMBOL(mdc_rename);
1417 EXPORT_SYMBOL(mdc_link);
1418 EXPORT_SYMBOL(mdc_readpage);
1419 EXPORT_SYMBOL(mdc_setattr);
1420 EXPORT_SYMBOL(mdc_close);
1421 EXPORT_SYMBOL(mdc_done_writing);
1422 EXPORT_SYMBOL(mdc_sync);
1423 EXPORT_SYMBOL(mdc_set_open_replay_data);
1424 EXPORT_SYMBOL(mdc_clear_open_replay_data);
1425 EXPORT_SYMBOL(mdc_store_inode_generation);
1426 EXPORT_SYMBOL(mdc_init_ea_size);
1427 EXPORT_SYMBOL(mdc_getxattr);
1428 EXPORT_SYMBOL(mdc_setxattr);
1430 module_init(mdc_init);
1431 module_exit(mdc_exit);