1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * lustre/mdt/mdt_handler.c
5 * Lustre Metadata Target (mdt) request handler
7 * Copyright (c) 2006 Cluster File Systems, Inc.
8 * Author: Peter Braam <braam@clusterfs.com>
9 * Author: Andreas Dilger <adilger@clusterfs.com>
10 * Author: Phil Schwan <phil@clusterfs.com>
11 * Author: Mike Shaver <shaver@clusterfs.com>
12 * Author: Nikita Danilov <nikita@clusterfs.com>
13 * Author: Huang Hua <huanghua@clusterfs.com>
14 * Author: Yury Umanets <umka@clusterfs.com>
16 * This file is part of the Lustre file system, http://www.lustre.org
17 * Lustre is a trademark of Cluster File Systems, Inc.
19 * You may have signed or agreed to another license before downloading
20 * this software. If so, you are bound by the terms and conditions
21 * of that agreement, and the following does not apply to you. See the
22 * LICENSE file included with this distribution for more information.
24 * If you did not agree to a different license, then this copy of Lustre
25 * is open source software; you can redistribute it and/or modify it
26 * under the terms of version 2 of the GNU General Public License as
27 * published by the Free Software Foundation.
29 * In either case, Lustre is distributed in the hope that it will be
30 * useful, but WITHOUT ANY WARRANTY; without even the implied warranty
31 * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
32 * license text for more details.
36 # define EXPORT_SYMTAB
38 #define DEBUG_SUBSYSTEM S_MDS
40 #include <linux/module.h>
42 * struct OBD_{ALLOC,FREE}*()
44 #include <obd_support.h>
45 /* struct ptlrpc_request */
46 #include <lustre_net.h>
47 /* struct obd_export */
48 #include <lustre_export.h>
49 /* struct obd_device */
52 #include <dt_object.h>
53 #include <lustre_mds.h>
54 #include <lustre_mdt.h>
55 #include "mdt_internal.h"
56 #include <linux/lustre_acl.h>
57 #include <lustre_param.h>
59 mdl_mode_t mdt_mdl_lock_modes[] = {
60 [LCK_MINMODE] = MDL_MINMODE,
67 [LCK_GROUP] = MDL_GROUP
70 ldlm_mode_t mdt_dlm_lock_modes[] = {
71 [MDL_MINMODE] = LCK_MINMODE,
78 [MDL_GROUP] = LCK_GROUP
82 * Initialized in mdt_mod_init().
84 unsigned long mdt_num_threads;
86 /* ptlrpc request handler for MDT. All handlers are
87 * grouped into several slices - struct mdt_opc_slice,
88 * and stored in an array - mdt_handlers[].
91 /* The name of this handler. */
93 /* Fail id for this handler, checked at the beginning of this handler*/
95 /* Operation code for this handler */
97 /* flags are listed in enum mdt_handler_flags below. */
99 /* The actual handler function to execute. */
100 int (*mh_act)(struct mdt_thread_info *info);
101 /* Request format for this request. */
102 const struct req_format *mh_fmt;
105 enum mdt_handler_flags {
107 * struct mdt_body is passed in the incoming message, and object
108 * identified by this fid exists on disk.
110 * "habeo corpus" == "I have a body"
112 HABEO_CORPUS = (1 << 0),
114 * struct ldlm_request is passed in the incoming message.
116 * "habeo clavis" == "I have a key"
118 HABEO_CLAVIS = (1 << 1),
120 * this request has fixed reply format, so that reply message can be
121 * packed by generic code.
123 * "habeo refero" == "I have a reply"
125 HABEO_REFERO = (1 << 2),
127 * this request will modify something, so check whether the filesystem
128 * is readonly or not, then return -EROFS to client asap if necessary.
130 * "mutabor" == "I shall modify"
135 struct mdt_opc_slice {
138 struct mdt_handler *mos_hs;
141 static struct mdt_opc_slice mdt_regular_handlers[];
142 static struct mdt_opc_slice mdt_readpage_handlers[];
143 static struct mdt_opc_slice mdt_xmds_handlers[];
144 static struct mdt_opc_slice mdt_seq_handlers[];
145 static struct mdt_opc_slice mdt_fld_handlers[];
147 static struct mdt_device *mdt_dev(struct lu_device *d);
148 static int mdt_regular_handle(struct ptlrpc_request *req);
149 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags);
151 static struct lu_object_operations mdt_obj_ops;
153 int mdt_get_disposition(struct ldlm_reply *rep, int flag)
157 return (rep->lock_policy_res1 & flag);
160 void mdt_clear_disposition(struct mdt_thread_info *info,
161 struct ldlm_reply *rep, int flag)
164 info->mti_opdata &= ~flag;
166 rep->lock_policy_res1 &= ~flag;
169 void mdt_set_disposition(struct mdt_thread_info *info,
170 struct ldlm_reply *rep, int flag)
173 info->mti_opdata |= flag;
175 rep->lock_policy_res1 |= flag;
178 void mdt_lock_reg_init(struct mdt_lock_handle *lh, ldlm_mode_t lm)
180 lh->mlh_pdo_hash = 0;
181 lh->mlh_reg_mode = lm;
182 lh->mlh_type = MDT_REG_LOCK;
185 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, ldlm_mode_t lm,
186 const char *name, int namelen)
188 lh->mlh_reg_mode = lm;
189 lh->mlh_type = MDT_PDO_LOCK;
192 LASSERT(namelen > 0);
193 lh->mlh_pdo_hash = full_name_hash(name, namelen);
195 LASSERT(namelen == 0);
196 lh->mlh_pdo_hash = 0ull;
200 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
201 struct mdt_lock_handle *lh)
207 * Any dir access needs couple of locks:
209 * 1) on part of dir we gonna take lookup/modify;
211 * 2) on whole dir to protect it from concurrent splitting and/or to
212 * flush client's cache for readdir().
214 * so, for a given mode and object this routine decides what lock mode
215 * to use for lock #2:
217 * 1) if caller's gonna lookup in dir then we need to protect dir from
218 * being splitted only - LCK_CR
220 * 2) if caller's gonna modify dir then we need to protect dir from
221 * being splitted and to flush cache - LCK_CW
223 * 3) if caller's gonna modify dir and that dir seems ready for
224 * splitting then we need to protect it from any type of access
225 * (lookup/modify/split) - LCK_EX --bzzz
228 LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
229 LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
232 * Ask underlaying level its opinion about preferable PDO lock mode
233 * having access type passed as regular lock mode:
235 * - MDL_MINMODE means that lower layer does not want to specify lock
238 * - MDL_NL means that no PDO lock should be taken. This is used in some
239 * cases. Say, for non-splittable directories no need to use PDO locks
242 mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
243 mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
245 if (mode != MDL_MINMODE) {
246 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
249 * Lower layer does not want to specify locking mode. We do it
250 * our selves. No special protection is needed, just flush
251 * client's cache on modification and allow concurrent
254 switch (lh->mlh_reg_mode) {
256 lh->mlh_pdo_mode = LCK_EX;
259 lh->mlh_pdo_mode = LCK_CR;
262 lh->mlh_pdo_mode = LCK_CW;
265 CERROR("Not expected lock type (0x%x)\n",
266 (int)lh->mlh_reg_mode);
271 LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
275 static int mdt_getstatus(struct mdt_thread_info *info)
277 struct mdt_device *mdt = info->mti_mdt;
278 struct md_device *next = mdt->mdt_child;
279 struct mdt_body *repbody;
284 rc = mdt_check_ucred(info);
286 RETURN(err_serious(rc));
288 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETSTATUS_PACK))
289 RETURN(err_serious(-ENOMEM));
291 repbody = req_capsule_server_get(&info->mti_pill, &RMF_MDT_BODY);
292 rc = next->md_ops->mdo_root_get(info->mti_env, next, &repbody->fid1);
296 repbody->valid |= OBD_MD_FLID;
298 if (mdt->mdt_opts.mo_mds_capa) {
299 struct mdt_object *root;
300 struct lustre_capa *capa;
302 root = mdt_object_find(info->mti_env, mdt, &repbody->fid1);
304 RETURN(PTR_ERR(root));
306 capa = req_capsule_server_get(&info->mti_pill, &RMF_CAPA1);
308 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
310 rc = mo_capa_get(info->mti_env, mdt_object_child(root), capa,
312 mdt_object_put(info->mti_env, root);
314 repbody->valid |= OBD_MD_FLMDSCAPA;
320 static int mdt_statfs(struct mdt_thread_info *info)
322 struct md_device *next = info->mti_mdt->mdt_child;
323 struct obd_statfs *osfs;
328 /* This will trigger a watchdog timeout */
329 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
330 (MDT_SERVICE_WATCHDOG_TIMEOUT / 1000) + 1);
332 rc = mdt_check_ucred(info);
334 RETURN(err_serious(rc));
336 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK)) {
337 rc = err_serious(-ENOMEM);
339 osfs = req_capsule_server_get(&info->mti_pill,&RMF_OBD_STATFS);
340 rc = next->md_ops->mdo_statfs(info->mti_env, next,
342 statfs_pack(osfs, &info->mti_u.ksfs);
347 void mdt_pack_size2body(struct mdt_thread_info *info, struct mdt_object *o)
350 struct lu_attr *attr = &info->mti_attr.ma_attr;
352 b = req_capsule_server_get(&info->mti_pill, &RMF_MDT_BODY);
354 /* Check if Size-on-MDS is enabled. */
355 if ((mdt_conn_flags(info) & OBD_CONNECT_SOM) &&
356 S_ISREG(attr->la_mode) && mdt_sizeonmds_enabled(o)) {
357 b->valid |= (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS);
358 b->size = attr->la_size;
359 b->blocks = attr->la_blocks;
363 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
364 const struct lu_attr *attr, const struct lu_fid *fid)
366 /*XXX should pack the reply body according to lu_valid*/
367 b->valid |= OBD_MD_FLCTIME | OBD_MD_FLUID |
368 OBD_MD_FLGID | OBD_MD_FLTYPE |
369 OBD_MD_FLMODE | OBD_MD_FLNLINK | OBD_MD_FLFLAGS |
370 OBD_MD_FLATIME | OBD_MD_FLMTIME ;
372 if (!S_ISREG(attr->la_mode))
373 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
375 b->atime = attr->la_atime;
376 b->mtime = attr->la_mtime;
377 b->ctime = attr->la_ctime;
378 b->mode = attr->la_mode;
379 b->size = attr->la_size;
380 b->blocks = attr->la_blocks;
381 b->uid = attr->la_uid;
382 b->gid = attr->la_gid;
383 b->flags = attr->la_flags;
384 b->nlink = attr->la_nlink;
385 b->rdev = attr->la_rdev;
389 b->valid |= OBD_MD_FLID;
390 CDEBUG(D_INODE, ""DFID": nlink=%d, mode=%o, size="LPU64"\n",
391 PFID(fid), b->nlink, b->mode, b->size);
395 mdt_body_reverse_idmap(info, b);
398 static inline int mdt_body_has_lov(const struct lu_attr *la,
399 const struct mdt_body *body)
401 return ((S_ISREG(la->la_mode) && (body->valid & OBD_MD_FLEASIZE)) ||
402 (S_ISDIR(la->la_mode) && (body->valid & OBD_MD_FLDIREA )) );
405 static int mdt_getattr_internal(struct mdt_thread_info *info,
406 struct mdt_object *o)
408 struct md_object *next = mdt_object_child(o);
409 const struct mdt_body *reqbody = info->mti_body;
410 struct ptlrpc_request *req = mdt_info_req(info);
411 struct mdt_export_data *med = &req->rq_export->exp_mdt_data;
412 struct md_attr *ma = &info->mti_attr;
413 struct lu_attr *la = &ma->ma_attr;
414 struct req_capsule *pill = &info->mti_pill;
415 const struct lu_env *env = info->mti_env;
416 struct mdt_body *repbody;
417 struct lu_buf *buffer = &info->mti_buf;
421 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
422 RETURN(err_serious(-ENOMEM));
424 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
428 rc = mdt_object_exists(o);
430 /* This object is located on remote node.*/
431 repbody->fid1 = *mdt_object_fid(o);
432 repbody->valid = OBD_MD_FLID | OBD_MD_MDS;
436 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
437 buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD, RCL_SERVER);
439 /* If it is dir object and client require MEA, then we got MEA */
440 if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
441 reqbody->valid & OBD_MD_MEA) {
442 /* Assumption: MDT_MD size is enough for lmv size. */
443 ma->ma_lmv = buffer->lb_buf;
444 ma->ma_lmv_size = buffer->lb_len;
445 ma->ma_need = MA_LMV | MA_INODE;
447 ma->ma_lmm = buffer->lb_buf;
448 ma->ma_lmm_size = buffer->lb_len;
449 ma->ma_need = MA_LOV | MA_INODE;
452 rc = mo_attr_get(env, next, ma);
454 CERROR("getattr error for "DFID": %d\n",
455 PFID(mdt_object_fid(o)), rc);
459 if (likely(ma->ma_valid & MA_INODE))
460 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
464 if (mdt_body_has_lov(la, reqbody)) {
465 if (ma->ma_valid & MA_LOV) {
466 LASSERT(ma->ma_lmm_size);
467 mdt_dump_lmm(D_INFO, ma->ma_lmm);
468 repbody->eadatasize = ma->ma_lmm_size;
469 if (S_ISDIR(la->la_mode))
470 repbody->valid |= OBD_MD_FLDIREA;
472 repbody->valid |= OBD_MD_FLEASIZE;
474 if (ma->ma_valid & MA_LMV) {
475 LASSERT(S_ISDIR(la->la_mode));
476 repbody->eadatasize = ma->ma_lmv_size;
477 repbody->valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
479 if (!(ma->ma_valid & MA_LOV) && !(ma->ma_valid & MA_LMV)) {
480 repbody->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
482 } else if (S_ISLNK(la->la_mode) &&
483 reqbody->valid & OBD_MD_LINKNAME) {
484 buffer->lb_buf = ma->ma_lmm;
485 buffer->lb_len = reqbody->eadatasize;
486 rc = mo_readlink(env, next, buffer);
487 if (unlikely(rc <= 0)) {
488 CERROR("readlink failed: %d\n", rc);
491 repbody->valid |= OBD_MD_LINKNAME;
492 repbody->eadatasize = rc;
494 ((char*)ma->ma_lmm)[rc - 1] = 0;
495 CDEBUG(D_INODE, "symlink dest %s, len = %d\n",
496 (char*)ma->ma_lmm, rc);
501 if (reqbody->valid & OBD_MD_FLMODEASIZE) {
502 repbody->max_cookiesize = info->mti_mdt->mdt_max_cookiesize;
503 repbody->max_mdsize = info->mti_mdt->mdt_max_mdsize;
504 repbody->valid |= OBD_MD_FLMODEASIZE;
505 CDEBUG(D_INODE, "I am going to change the MAX_MD_SIZE & "
506 "MAX_COOKIE to : %d:%d\n", repbody->max_mdsize,
507 repbody->max_cookiesize);
510 if (med->med_rmtclient && (reqbody->valid & OBD_MD_FLRMTPERM)) {
511 void *buf = req_capsule_server_get(pill, &RMF_ACL);
513 /* mdt_getattr_lock only */
514 rc = mdt_pack_remote_perm(info, o, buf);
516 repbody->valid &= ~OBD_MD_FLRMTPERM;
517 repbody->aclsize = 0;
520 repbody->valid |= OBD_MD_FLRMTPERM;
521 repbody->aclsize = sizeof(struct mdt_remote_perm);
524 #ifdef CONFIG_FS_POSIX_ACL
525 else if ((req->rq_export->exp_connect_flags & OBD_CONNECT_ACL) &&
526 (reqbody->valid & OBD_MD_FLACL)) {
527 buffer->lb_buf = req_capsule_server_get(pill, &RMF_ACL);
528 buffer->lb_len = req_capsule_get_size(pill,
529 &RMF_ACL, RCL_SERVER);
530 if (buffer->lb_len > 0) {
531 rc = mo_xattr_get(env, next, buffer,
532 XATTR_NAME_ACL_ACCESS);
534 if (rc == -ENODATA) {
535 repbody->aclsize = 0;
536 repbody->valid |= OBD_MD_FLACL;
538 } else if (rc == -EOPNOTSUPP) {
541 CERROR("got acl size: %d\n", rc);
544 repbody->aclsize = rc;
545 repbody->valid |= OBD_MD_FLACL;
552 if ((reqbody->valid & OBD_MD_FLMDSCAPA) &&
553 info->mti_mdt->mdt_opts.mo_mds_capa) {
554 struct lustre_capa *capa;
556 capa = req_capsule_server_get(&info->mti_pill, &RMF_CAPA1);
558 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
559 rc = mo_capa_get(env, next, capa, 0);
562 repbody->valid |= OBD_MD_FLMDSCAPA;
567 static int mdt_renew_capa(struct mdt_thread_info *info)
569 struct mdt_device *mdt = info->mti_mdt;
570 struct mdt_object *obj = info->mti_object;
571 struct mdt_body *body;
572 struct lustre_capa *capa, *c;
576 /* if object doesn't exist, or server has disabled capability,
577 * return directly, client will find body->valid OBD_MD_FLOSSCAPA
580 if (!obj || !mdt->mdt_opts.mo_mds_capa)
583 body = req_capsule_server_get(&info->mti_pill, &RMF_MDT_BODY);
584 LASSERT(body != NULL);
586 c = req_capsule_client_get(&info->mti_pill, &RMF_CAPA1);
589 capa = req_capsule_server_get(&info->mti_pill, &RMF_CAPA1);
593 rc = mo_capa_get(info->mti_env, mdt_object_child(obj), capa, 1);
595 body->valid |= OBD_MD_FLOSSCAPA;
600 static int mdt_getattr(struct mdt_thread_info *info)
602 struct mdt_object *obj = info->mti_object;
603 struct req_capsule *pill = &info->mti_pill;
604 struct mdt_body *reqbody;
605 struct mdt_body *repbody;
611 reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
614 if (reqbody->valid & OBD_MD_FLOSSCAPA) {
615 rc = req_capsule_pack(pill);
617 rc = err_serious(rc);
619 rc = mdt_renew_capa(info);
620 mdt_shrink_reply(info);
625 LASSERT(obj != NULL);
626 LASSERT(lu_object_assert_exists(&obj->mot_obj.mo_lu));
628 mode = lu_object_attr(&obj->mot_obj.mo_lu);
629 if (S_ISLNK(mode) && (reqbody->valid & OBD_MD_LINKNAME) &&
630 (reqbody->eadatasize > info->mti_mdt->mdt_max_mdsize))
631 md_size = reqbody->eadatasize;
633 md_size = info->mti_mdt->mdt_max_mdsize;
635 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, md_size);
637 rc = req_capsule_pack(pill);
638 if (unlikely(rc != 0))
639 GOTO(out, rc = err_serious(rc));
641 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
642 LASSERT(repbody != NULL);
643 repbody->eadatasize = 0;
644 repbody->aclsize = 0;
646 if (reqbody->valid & OBD_MD_FLRMTPERM)
647 rc = mdt_init_ucred(info, reqbody);
649 rc = mdt_check_ucred(info);
651 GOTO(out_shrink, rc);
653 info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
654 info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
657 * Don't check capability at all, because rename might getattr for
658 * remote obj, and at that time no capability is available.
660 mdt_set_capainfo(info, 1, &reqbody->fid1, BYPASS_CAPA);
661 rc = mdt_getattr_internal(info, obj);
662 if (reqbody->valid & OBD_MD_FLRMTPERM)
663 mdt_exit_ucred(info);
666 mdt_shrink_reply(info);
671 static int mdt_is_subdir(struct mdt_thread_info *info)
673 struct mdt_object *o = info->mti_object;
674 struct req_capsule *pill = &info->mti_pill;
675 const struct mdt_body *body = info->mti_body;
676 struct mdt_body *repbody;
682 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
685 * We save last checked parent fid to @repbody->fid1 for remote
688 LASSERT(fid_is_sane(&body->fid2));
689 LASSERT(mdt_object_exists(o) > 0);
690 rc = mdo_is_subdir(info->mti_env, mdt_object_child(o),
691 &body->fid2, &repbody->fid1);
692 if (rc == 0 || rc == -EREMOTE)
693 repbody->valid |= OBD_MD_FLID;
698 static int mdt_raw_lookup(struct mdt_thread_info *info,
699 struct mdt_object *parent,
700 const struct lu_name *lname,
701 struct ldlm_reply *ldlm_rep)
703 struct md_object *next = mdt_object_child(info->mti_object);
704 const struct mdt_body *reqbody = info->mti_body;
705 struct lu_fid *child_fid = &info->mti_tmp_fid1;
706 struct mdt_body *repbody;
710 if (reqbody->valid != OBD_MD_FLID)
713 LASSERT(!info->mti_cross_ref);
715 /* Only got the fid of this obj by name */
716 rc = mdo_lookup(info->mti_env, next, lname, child_fid,
719 /* XXX is raw_lookup possible as intent operation? */
722 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
725 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
727 repbody = req_capsule_server_get(&info->mti_pill, &RMF_MDT_BODY);
730 repbody = req_capsule_server_get(&info->mti_pill,
732 repbody->fid1 = *child_fid;
733 repbody->valid = OBD_MD_FLID;
739 * UPDATE lock should be taken against parent, and be release before exit;
740 * child_bits lock should be taken against child, and be returned back:
741 * (1)normal request should release the child lock;
742 * (2)intent request will grant the lock to client.
744 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
745 struct mdt_lock_handle *lhc,
747 struct ldlm_reply *ldlm_rep)
749 struct ptlrpc_request *req = mdt_info_req(info);
750 struct mdt_body *reqbody = NULL;
751 struct mdt_object *parent = info->mti_object;
752 struct mdt_object *child;
753 struct md_object *next = mdt_object_child(parent);
754 struct lu_fid *child_fid = &info->mti_tmp_fid1;
755 struct lu_name *lname = NULL;
758 struct mdt_lock_handle *lhp;
759 struct ldlm_lock *lock;
760 struct ldlm_res_id *res_id;
766 is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
767 LASSERT(ergo(is_resent,
768 lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
770 LASSERT(parent != NULL);
771 name = req_capsule_client_get(&info->mti_pill, &RMF_NAME);
773 RETURN(err_serious(-EFAULT));
775 namelen = req_capsule_get_size(&info->mti_pill, &RMF_NAME,
777 LASSERT(namelen >= 0);
779 /* XXX: "namelen == 0" is for getattr by fid (OBD_CONNECT_ATTRFID),
780 * otherwise do not allow empty name, that is the name must contain
781 * at least one character and the terminating '\0'*/
783 reqbody =req_capsule_client_get(&info->mti_pill, &RMF_MDT_BODY);
784 LASSERT(fid_is_sane(&reqbody->fid2));
787 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
789 PFID(mdt_object_fid(parent)), PFID(&reqbody->fid2),
792 lname = mdt_name(info->mti_env, (char *)name, namelen);
793 CDEBUG(D_INODE, "getattr with lock for "DFID"/%s, "
795 PFID(mdt_object_fid(parent)), name, ldlm_rep);
798 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
800 rc = mdt_object_exists(parent);
801 if (unlikely(rc == 0)) {
802 LU_OBJECT_DEBUG(D_WARNING, info->mti_env,
803 &parent->mot_obj.mo_lu,
804 "Parent doesn't exist!\n");
807 LASSERTF(rc > 0, "Parent "DFID" is on remote server\n",
808 PFID(mdt_object_fid(parent)));
811 rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
819 if (info->mti_cross_ref) {
820 /* Only getattr on the child. Parent is on another node. */
821 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
823 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
824 "ldlm_rep=%p\n", PFID(mdt_object_fid(child)), ldlm_rep);
827 /* Do not take lock for resent case. */
828 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
829 LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
830 lhc->mlh_reg_lh.cookie);
831 LASSERT(fid_res_name_eq(mdt_object_fid(child),
832 &lock->l_resource->lr_name));
836 mdt_lock_handle_init(lhc);
837 mdt_lock_reg_init(lhc, LCK_PR);
840 * Object's name is on another MDS, no lookup lock is
841 * needed here but update is.
843 child_bits &= ~MDS_INODELOCK_LOOKUP;
844 child_bits |= MDS_INODELOCK_UPDATE;
846 rc = mdt_object_lock(info, child, lhc, child_bits,
850 /* Finally, we can get attr for child. */
851 mdt_set_capainfo(info, 0, mdt_object_fid(child),
853 rc = mdt_getattr_internal(info, child);
854 if (unlikely(rc != 0))
855 mdt_object_unlock(info, child, lhc, 1);
860 /* step 1: lock parent */
861 lhp = &info->mti_lh[MDT_LH_PARENT];
862 mdt_lock_pdo_init(lhp, LCK_PR, name, namelen);
863 rc = mdt_object_lock(info, parent, lhp, MDS_INODELOCK_UPDATE,
866 if (unlikely(rc != 0))
870 /* step 2: lookup child's fid by name */
871 rc = mdo_lookup(info->mti_env, next, lname, child_fid,
876 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
877 GOTO(out_parent, rc);
879 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
881 *child_fid = reqbody->fid2;
882 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
886 *step 3: find the child object by fid & lock it.
887 * regardless if it is local or remote.
889 child = mdt_object_find(info->mti_env, info->mti_mdt, child_fid);
891 if (unlikely(IS_ERR(child)))
892 GOTO(out_parent, rc = PTR_ERR(child));
894 /* Do not take lock for resent case. */
895 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
896 LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
897 lhc->mlh_reg_lh.cookie);
899 res_id = &lock->l_resource->lr_name;
900 if (!fid_res_name_eq(mdt_object_fid(child),
901 &lock->l_resource->lr_name)) {
902 LASSERTF(fid_res_name_eq(mdt_object_fid(parent),
903 &lock->l_resource->lr_name),
904 "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n",
905 (unsigned long)res_id->name[0],
906 (unsigned long)res_id->name[1],
907 (unsigned long)res_id->name[2],
908 PFID(mdt_object_fid(parent)));
909 CWARN("Although resent, but still not get child lock"
910 "parent:"DFID" child:"DFID"\n",
911 PFID(mdt_object_fid(parent)),
912 PFID(mdt_object_fid(child)));
913 lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
921 mdt_lock_handle_init(lhc);
922 mdt_lock_reg_init(lhc, LCK_PR);
924 if (mdt_object_exists(child) == 0) {
925 LU_OBJECT_DEBUG(D_WARNING, info->mti_env,
926 &child->mot_obj.mo_lu,
927 "Object doesn't exist!\n");
929 rc = mdt_object_lock(info, child, lhc, child_bits,
932 if (unlikely(rc != 0))
936 /* finally, we can get attr for child. */
937 mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
938 rc = mdt_getattr_internal(info, child);
939 if (unlikely(rc != 0)) {
940 mdt_object_unlock(info, child, lhc, 1);
942 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
944 struct mdt_body *repbody;
947 /* Debugging code. */
948 res_id = &lock->l_resource->lr_name;
949 LDLM_DEBUG(lock, "Returning lock to client\n");
950 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
951 &lock->l_resource->lr_name),
952 "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n",
953 (unsigned long)res_id->name[0],
954 (unsigned long)res_id->name[1],
955 (unsigned long)res_id->name[2],
956 PFID(mdt_object_fid(child)));
958 * Pack Size-on-MDS inode attributes to the body if
959 * update lock is given.
961 repbody = req_capsule_server_get(&info->mti_pill,
963 ma = &info->mti_attr.ma_attr;
964 if (lock->l_policy_data.l_inodebits.bits &
965 MDS_INODELOCK_UPDATE)
966 mdt_pack_size2body(info, child);
972 mdt_object_put(info->mti_env, child);
974 mdt_object_unlock(info, parent, lhp, 1);
979 /* normal handler: should release the child lock */
980 static int mdt_getattr_name(struct mdt_thread_info *info)
982 struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
983 struct mdt_body *reqbody;
984 struct mdt_body *repbody;
988 reqbody = req_capsule_client_get(&info->mti_pill, &RMF_MDT_BODY);
989 LASSERT(reqbody != NULL);
990 repbody = req_capsule_server_get(&info->mti_pill, &RMF_MDT_BODY);
991 LASSERT(repbody != NULL);
993 info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
994 info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
995 repbody->eadatasize = 0;
996 repbody->aclsize = 0;
998 rc = mdt_init_ucred(info, reqbody);
1002 rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1003 if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1004 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1005 lhc->mlh_reg_lh.cookie = 0;
1007 mdt_exit_ucred(info);
1010 mdt_shrink_reply(info);
1014 static struct lu_device_operations mdt_lu_ops;
1016 static int lu_device_is_mdt(struct lu_device *d)
1018 return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &mdt_lu_ops);
1021 static int mdt_set_info(struct mdt_thread_info *info)
1023 struct ptlrpc_request *req = mdt_info_req(info);
1029 rc = lustre_pack_reply(req, 1, NULL, NULL);
1033 key = req_capsule_client_get(&info->mti_pill, &RMF_SETINFO_KEY);
1035 DEBUG_REQ(D_HA, req, "no set_info key");
1039 keylen = req_capsule_get_size(&info->mti_pill, &RMF_SETINFO_KEY,
1042 val = req_capsule_client_get(&info->mti_pill, &RMF_SETINFO_VAL);
1044 DEBUG_REQ(D_HA, req, "no set_info val");
1048 if (keylen != (sizeof(KEY_READ_ONLY) - 1) ||
1049 memcmp(key, KEY_READ_ONLY, keylen) != 0)
1053 lustre_msg_set_status(req->rq_repmsg, 0);
1055 spin_lock(&req->rq_export->exp_lock);
1057 req->rq_export->exp_connect_flags |= OBD_CONNECT_RDONLY;
1059 req->rq_export->exp_connect_flags &= ~OBD_CONNECT_RDONLY;
1060 spin_unlock(&req->rq_export->exp_lock);
1065 static int mdt_connect(struct mdt_thread_info *info)
1068 struct ptlrpc_request *req;
1070 req = mdt_info_req(info);
1071 rc = target_handle_connect(req);
1073 LASSERT(req->rq_export != NULL);
1074 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
1075 rc = mdt_init_idmap(info);
1077 /* if mdt_init_idmap failed, revocation for connect */
1078 obd_disconnect(class_export_get(req->rq_export));
1080 rc = err_serious(rc);
1084 static int mdt_disconnect(struct mdt_thread_info *info)
1089 rc = target_handle_disconnect(mdt_info_req(info));
1091 rc = err_serious(rc);
1095 static int mdt_sendpage(struct mdt_thread_info *info,
1096 struct lu_rdpg *rdpg)
1098 struct ptlrpc_request *req = mdt_info_req(info);
1099 struct ptlrpc_bulk_desc *desc;
1100 struct l_wait_info *lwi = &info->mti_u.rdpg.mti_wait_info;
1107 desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, BULK_PUT_SOURCE,
1110 GOTO(out, rc = -ENOMEM);
1112 for (i = 0, tmpcount = rdpg->rp_count;
1113 i < rdpg->rp_npages; i++, tmpcount -= tmpsize) {
1114 tmpsize = min_t(int, tmpcount, CFS_PAGE_SIZE);
1115 ptlrpc_prep_bulk_page(desc, rdpg->rp_pages[i], 0, tmpsize);
1118 LASSERT(desc->bd_nob == rdpg->rp_count);
1119 rc = ptlrpc_start_bulk_transfer(desc);
1121 GOTO(free_desc, rc);
1123 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1124 GOTO(abort_bulk, rc);
1126 *lwi = LWI_TIMEOUT(obd_timeout * HZ / 4, NULL, NULL);
1127 rc = l_wait_event(desc->bd_waitq, !ptlrpc_bulk_active(desc), lwi);
1128 LASSERT (rc == 0 || rc == -ETIMEDOUT);
1131 if (desc->bd_success &&
1132 desc->bd_nob_transferred == rdpg->rp_count)
1133 GOTO(free_desc, rc);
1135 rc = -ETIMEDOUT; /* XXX should this be a different errno? */
1138 DEBUG_REQ(D_ERROR, req, "bulk failed: %s %d(%d), evicting %s@%s",
1139 (rc == -ETIMEDOUT) ? "timeout" : "network error",
1140 desc->bd_nob_transferred, rdpg->rp_count,
1141 req->rq_export->exp_client_uuid.uuid,
1142 req->rq_export->exp_connection->c_remote_uuid.uuid);
1144 class_fail_export(req->rq_export);
1148 ptlrpc_abort_bulk(desc);
1150 ptlrpc_free_bulk(desc);
1155 #ifdef HAVE_SPLIT_SUPPORT
1157 * Retrieve dir entry from the page and insert it to the slave object, actually,
1158 * this should be in osd layer, but since it will not in the final product, so
1159 * just do it here and do not define more moo api anymore for this.
1161 static int mdt_write_dir_page(struct mdt_thread_info *info, struct page *page,
1164 struct mdt_object *object = info->mti_object;
1165 struct lu_fid *lf = &info->mti_tmp_fid2;
1166 struct md_attr *ma = &info->mti_attr;
1167 struct lu_dirpage *dp;
1168 struct lu_dirent *ent;
1169 int rc = 0, offset = 0;
1172 /* Make sure we have at least one entry. */
1177 * Disable trans for this name insert, since it will include many trans
1180 info->mti_no_need_trans = 1;
1182 * When write_dir_page, no need update parent's ctime,
1183 * and no permission check for name_insert.
1185 ma->ma_attr.la_ctime = 0;
1186 ma->ma_attr.la_valid = LA_MODE;
1187 ma->ma_valid = MA_INODE;
1190 dp = page_address(page);
1191 offset = (int)((__u32)lu_dirent_start(dp) - (__u32)dp);
1193 for (ent = lu_dirent_start(dp); ent != NULL;
1194 ent = lu_dirent_next(ent)) {
1195 struct lu_name *lname;
1198 if (le16_to_cpu(ent->lde_namelen) == 0)
1201 fid_le_to_cpu(lf, &ent->lde_fid);
1202 if (le32_to_cpu(ent->lde_hash) & MAX_HASH_HIGHEST_BIT)
1203 ma->ma_attr.la_mode = S_IFDIR;
1205 ma->ma_attr.la_mode = 0;
1206 OBD_ALLOC(name, le16_to_cpu(ent->lde_namelen) + 1);
1208 GOTO(out, rc = -ENOMEM);
1210 memcpy(name, ent->lde_name, le16_to_cpu(ent->lde_namelen));
1211 lname = mdt_name(info->mti_env, name,
1212 le16_to_cpu(ent->lde_namelen) + 1);
1213 ma->ma_attr_flags |= MDS_PERM_BYPASS;
1214 rc = mdo_name_insert(info->mti_env,
1215 md_object_next(&object->mot_obj),
1217 OBD_FREE(name, le16_to_cpu(ent->lde_namelen) + 1);
1219 CERROR("Can't insert %*.*s, rc %d\n",
1220 le16_to_cpu(ent->lde_namelen),
1221 le16_to_cpu(ent->lde_namelen),
1226 offset += lu_dirent_size(ent);
1236 static int mdt_bulk_timeout(void *data)
1240 CERROR("mdt bulk transfer timeout \n");
1245 static int mdt_writepage(struct mdt_thread_info *info)
1247 struct ptlrpc_request *req = mdt_info_req(info);
1248 struct mdt_body *reqbody;
1249 struct l_wait_info *lwi;
1250 struct ptlrpc_bulk_desc *desc;
1256 reqbody = req_capsule_client_get(&info->mti_pill, &RMF_MDT_BODY);
1257 if (reqbody == NULL)
1258 RETURN(err_serious(-EFAULT));
1260 desc = ptlrpc_prep_bulk_exp(req, 1, BULK_GET_SINK, MDS_BULK_PORTAL);
1262 RETURN(err_serious(-ENOMEM));
1264 /* allocate the page for the desc */
1265 page = cfs_alloc_page(CFS_ALLOC_STD);
1267 GOTO(desc_cleanup, rc = -ENOMEM);
1269 CDEBUG(D_INFO, "Received page offset %d size %d \n",
1270 (int)reqbody->size, (int)reqbody->nlink);
1272 ptlrpc_prep_bulk_page(desc, page, (int)reqbody->size,
1273 (int)reqbody->nlink);
1276 * Check if client was evicted while we were doing i/o before touching
1281 GOTO(cleanup_page, rc = -ENOMEM);
1283 if (desc->bd_export->exp_failed)
1286 rc = ptlrpc_start_bulk_transfer (desc);
1288 *lwi = LWI_TIMEOUT_INTERVAL(obd_timeout * HZ / 4, HZ,
1289 mdt_bulk_timeout, desc);
1290 rc = l_wait_event(desc->bd_waitq, !ptlrpc_bulk_active(desc) ||
1291 desc->bd_export->exp_failed, lwi);
1292 LASSERT(rc == 0 || rc == -ETIMEDOUT);
1293 if (rc == -ETIMEDOUT) {
1294 DEBUG_REQ(D_ERROR, req, "timeout on bulk GET");
1295 ptlrpc_abort_bulk(desc);
1296 } else if (desc->bd_export->exp_failed) {
1297 DEBUG_REQ(D_ERROR, req, "Eviction on bulk GET");
1299 ptlrpc_abort_bulk(desc);
1300 } else if (!desc->bd_success ||
1301 desc->bd_nob_transferred != desc->bd_nob) {
1302 DEBUG_REQ(D_ERROR, req, "%s bulk GET %d(%d)",
1304 "truncated" : "network error on",
1305 desc->bd_nob_transferred, desc->bd_nob);
1306 /* XXX should this be a different errno? */
1310 DEBUG_REQ(D_ERROR, req, "ptlrpc_bulk_get failed: rc %d", rc);
1313 GOTO(cleanup_lwi, rc);
1314 rc = mdt_write_dir_page(info, page, reqbody->nlink);
1319 __cfs_free_page(page);
1321 ptlrpc_free_bulk(desc);
1326 static int mdt_readpage(struct mdt_thread_info *info)
1328 struct mdt_object *object = info->mti_object;
1329 struct lu_rdpg *rdpg = &info->mti_u.rdpg.mti_rdpg;
1330 struct mdt_body *reqbody;
1331 struct mdt_body *repbody;
1336 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
1337 RETURN(err_serious(-ENOMEM));
1339 reqbody = req_capsule_client_get(&info->mti_pill, &RMF_MDT_BODY);
1340 repbody = req_capsule_server_get(&info->mti_pill, &RMF_MDT_BODY);
1341 if (reqbody == NULL || repbody == NULL)
1342 RETURN(err_serious(-EFAULT));
1344 rc = mdt_check_ucred(info);
1346 RETURN(err_serious(rc));
1349 * prepare @rdpg before calling lower layers and transfer itself. Here
1350 * reqbody->size contains offset of where to start to read and
1351 * reqbody->nlink contains number bytes to read.
1353 rdpg->rp_hash = reqbody->size;
1354 if ((__u64)rdpg->rp_hash != reqbody->size) {
1355 CERROR("Invalid hash: %#llx != %#llx\n",
1356 (__u64)rdpg->rp_hash, reqbody->size);
1359 rdpg->rp_count = reqbody->nlink;
1360 rdpg->rp_npages = (rdpg->rp_count + CFS_PAGE_SIZE - 1)>>CFS_PAGE_SHIFT;
1361 OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1362 if (rdpg->rp_pages == NULL)
1365 for (i = 0; i < rdpg->rp_npages; ++i) {
1366 rdpg->rp_pages[i] = cfs_alloc_page(CFS_ALLOC_STD);
1367 if (rdpg->rp_pages[i] == NULL)
1368 GOTO(free_rdpg, rc = -ENOMEM);
1371 /* call lower layers to fill allocated pages with directory data */
1372 rc = mo_readpage(info->mti_env, mdt_object_child(object), rdpg);
1374 GOTO(free_rdpg, rc);
1376 /* send pages to client */
1377 rc = mdt_sendpage(info, rdpg);
1382 for (i = 0; i < rdpg->rp_npages; i++)
1383 if (rdpg->rp_pages[i] != NULL)
1384 __cfs_free_page(rdpg->rp_pages[i]);
1385 OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1387 OBD_FAIL_RETURN(OBD_FAIL_MDS_SENDPAGE, 0);
1392 static int mdt_reint_internal(struct mdt_thread_info *info,
1393 struct mdt_lock_handle *lhc,
1396 struct req_capsule *pill = &info->mti_pill;
1397 struct mdt_device *mdt = info->mti_mdt;
1398 struct mdt_body *repbody;
1399 int need_shrink = 0;
1404 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER)) {
1405 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1406 mdt->mdt_max_mdsize);
1409 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER)) {
1410 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER,
1411 mdt->mdt_max_cookiesize);
1414 rc = req_capsule_pack(pill);
1416 CERROR("Can't pack response, rc %d\n", rc);
1417 RETURN(err_serious(rc));
1420 if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
1421 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1423 repbody->eadatasize = 0;
1424 repbody->aclsize = 0;
1427 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNPACK))
1428 GOTO(out_shrink, rc = err_serious(-EFAULT));
1430 rc = mdt_reint_unpack(info, op);
1432 CERROR("Can't unpack reint, rc %d\n", rc);
1433 GOTO(out_shrink, rc = err_serious(rc));
1436 rc = mdt_init_ucred_reint(info);
1438 GOTO(out_shrink, rc);
1440 rc = mdt_fix_attr_ucred(info, op);
1442 GOTO(out_ucred, rc = err_serious(rc));
1445 if (mdt_check_resent(info, mdt_reconstruct, lhc)) {
1446 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
1447 GOTO(out_ucred, rc);
1450 rc = mdt_reint_rec(info, lhc);
1453 mdt_exit_ucred(info);
1456 mdt_shrink_reply(info);
1460 static long mdt_reint_opcode(struct mdt_thread_info *info,
1461 const struct req_format **fmt)
1466 opc = err_serious(-EFAULT);
1467 ptr = req_capsule_client_get(&info->mti_pill, &RMF_REINT_OPC);
1470 DEBUG_REQ(D_INODE, mdt_info_req(info), "reint opt = %ld", opc);
1471 if (opc < REINT_MAX && fmt[opc] != NULL)
1472 req_capsule_extend(&info->mti_pill, fmt[opc]);
1474 CERROR("Unsupported opc: %ld\n", opc);
1475 opc = err_serious(opc);
1481 static int mdt_reint(struct mdt_thread_info *info)
1486 static const struct req_format *reint_fmts[REINT_MAX] = {
1487 [REINT_SETATTR] = &RQF_MDS_REINT_SETATTR,
1488 [REINT_CREATE] = &RQF_MDS_REINT_CREATE,
1489 [REINT_LINK] = &RQF_MDS_REINT_LINK,
1490 [REINT_UNLINK] = &RQF_MDS_REINT_UNLINK,
1491 [REINT_RENAME] = &RQF_MDS_REINT_RENAME,
1492 [REINT_OPEN] = &RQF_MDS_REINT_OPEN
1497 opc = mdt_reint_opcode(info, reint_fmts);
1500 * No lock possible here from client to pass it to reint code
1503 rc = mdt_reint_internal(info, NULL, opc);
1508 info->mti_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
1512 /* TODO these two methods not available now. */
1514 /* this should sync the whole device */
1515 static int mdt_device_sync(struct mdt_thread_info *info)
1520 /* this should sync this object */
1521 static int mdt_object_sync(struct mdt_thread_info *info)
1526 static int mdt_sync(struct mdt_thread_info *info)
1528 struct req_capsule *pill = &info->mti_pill;
1529 struct mdt_body *body;
1533 /* The fid may be zero, so we req_capsule_set manually */
1534 req_capsule_set(pill, &RQF_MDS_SYNC);
1536 body = req_capsule_client_get(pill, &RMF_MDT_BODY);
1538 RETURN(err_serious(-EINVAL));
1540 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
1541 RETURN(err_serious(-ENOMEM));
1543 if (fid_seq(&body->fid1) == 0) {
1544 /* sync the whole device */
1545 rc = req_capsule_pack(pill);
1547 rc = mdt_device_sync(info);
1549 rc = err_serious(rc);
1551 /* sync an object */
1552 rc = mdt_unpack_req_pack_rep(info, HABEO_CORPUS|HABEO_REFERO);
1554 rc = mdt_object_sync(info);
1556 struct md_object *next;
1557 const struct lu_fid *fid;
1558 struct lu_attr *la = &info->mti_attr.ma_attr;
1560 next = mdt_object_child(info->mti_object);
1561 info->mti_attr.ma_need = MA_INODE;
1562 info->mti_attr.ma_valid = 0;
1563 rc = mo_attr_get(info->mti_env, next,
1566 body = req_capsule_server_get(pill,
1568 fid = mdt_object_fid(info->mti_object);
1569 mdt_pack_attr2body(info, body, la, fid);
1573 rc = err_serious(rc);
1578 static int mdt_quotacheck_handle(struct mdt_thread_info *info)
1580 return err_serious(-EOPNOTSUPP);
1583 static int mdt_quotactl_handle(struct mdt_thread_info *info)
1585 return err_serious(-EOPNOTSUPP);
1589 * OBD PING and other handlers.
1591 static int mdt_obd_ping(struct mdt_thread_info *info)
1595 rc = target_handle_ping(mdt_info_req(info));
1597 rc = err_serious(rc);
1601 static int mdt_obd_log_cancel(struct mdt_thread_info *info)
1603 return err_serious(-EOPNOTSUPP);
1606 static int mdt_obd_qc_callback(struct mdt_thread_info *info)
1608 return err_serious(-EOPNOTSUPP);
1615 static struct ldlm_callback_suite cbs = {
1616 .lcs_completion = ldlm_server_completion_ast,
1617 .lcs_blocking = ldlm_server_blocking_ast,
1621 static int mdt_enqueue(struct mdt_thread_info *info)
1623 struct ptlrpc_request *req;
1628 * info->mti_dlm_req already contains swapped and (if necessary)
1629 * converted dlm request.
1631 LASSERT(info->mti_dlm_req != NULL);
1633 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_ENQUEUE)) {
1634 info->mti_fail_id = OBD_FAIL_LDLM_ENQUEUE;
1638 req = mdt_info_req(info);
1641 * Lock without inodebits makes no sense and will oops later in
1642 * ldlm. Let's check it now to see if we have wrong lock from client or
1643 * bits get corrupted somewhere in mdt_intent_policy().
1645 req_bits = info->mti_dlm_req->lock_desc.l_policy_data.l_inodebits.bits;
1646 /* This is disabled because we need to support liblustre flock.
1647 * LASSERT(req_bits != 0);
1650 rc = ldlm_handle_enqueue0(info->mti_mdt->mdt_namespace,
1651 req, info->mti_dlm_req, &cbs);
1652 info->mti_fail_id = OBD_FAIL_LDLM_REPLY;
1653 return rc ? err_serious(rc) : req->rq_status;
1656 static int mdt_convert(struct mdt_thread_info *info)
1659 struct ptlrpc_request *req;
1661 LASSERT(info->mti_dlm_req);
1662 req = mdt_info_req(info);
1663 rc = ldlm_handle_convert0(req, info->mti_dlm_req);
1664 return rc ? err_serious(rc) : req->rq_status;
1667 static int mdt_bl_callback(struct mdt_thread_info *info)
1669 CERROR("bl callbacks should not happen on MDS\n");
1671 return err_serious(-EOPNOTSUPP);
1674 static int mdt_cp_callback(struct mdt_thread_info *info)
1676 CERROR("cp callbacks should not happen on MDS\n");
1678 return err_serious(-EOPNOTSUPP);
1682 * sec context handlers
1684 static int mdt_sec_ctx_handle(struct mdt_thread_info *info)
1688 rc = mdt_handle_idmap(info);
1691 struct ptlrpc_request *req = mdt_info_req(info);
1694 opc = lustre_msg_get_opc(req->rq_reqmsg);
1695 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT)
1696 sptlrpc_svc_ctx_invalidate(req);
1702 static struct mdt_object *mdt_obj(struct lu_object *o)
1704 LASSERT(lu_device_is_mdt(o->lo_dev));
1705 return container_of0(o, struct mdt_object, mot_obj.mo_lu);
1708 struct mdt_object *mdt_object_find(const struct lu_env *env,
1709 struct mdt_device *d,
1710 const struct lu_fid *f)
1712 struct lu_object *o;
1713 struct mdt_object *m;
1716 o = lu_object_find(env, d->mdt_md_dev.md_lu_dev.ld_site, f);
1717 if (unlikely(IS_ERR(o)))
1718 m = (struct mdt_object *)o;
1724 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
1725 struct mdt_lock_handle *lh, __u64 ibits, int locality)
1727 struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
1728 ldlm_policy_data_t *policy = &info->mti_policy;
1729 struct ldlm_res_id *res_id = &info->mti_res_id;
1733 LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
1734 LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
1735 LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
1736 LASSERT(lh->mlh_type != MDT_NUL_LOCK);
1738 if (mdt_object_exists(o) < 0) {
1739 if (locality == MDT_CROSS_LOCK) {
1740 /* cross-ref object fix */
1741 ibits &= ~MDS_INODELOCK_UPDATE;
1742 ibits |= MDS_INODELOCK_LOOKUP;
1744 LASSERT(!(ibits & MDS_INODELOCK_UPDATE));
1745 LASSERT(ibits & MDS_INODELOCK_LOOKUP);
1747 /* No PDO lock on remote object */
1748 LASSERT(lh->mlh_type != MDT_PDO_LOCK);
1751 memset(policy, 0, sizeof(*policy));
1752 fid_build_reg_res_name(mdt_object_fid(o), res_id);
1755 * Take PDO lock on whole directory and build correct @res_id for lock
1756 * on part of directory.
1758 if (lh->mlh_pdo_hash != 0) {
1759 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
1760 mdt_lock_pdo_mode(info, o, lh);
1761 if (lh->mlh_pdo_mode != LCK_NL) {
1763 * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
1764 * is never going to be sent to client and we do not
1765 * want it slowed down due to possible cancels.
1767 policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
1768 rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
1769 policy, res_id, LDLM_FL_ATOMIC_CB);
1775 * Finish res_id initializing by name hash marking patr of
1776 * directory which is taking modification.
1778 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
1781 policy->l_inodebits.bits = ibits;
1784 * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
1785 * going to be sent to client. If it is - mdt_intent_policy() path will
1786 * fix it up and turns FL_LOCAL flag off.
1788 rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
1789 res_id, LDLM_FL_LOCAL_ONLY | LDLM_FL_ATOMIC_CB);
1794 if (lh->mlh_type == MDT_PDO_LOCK) {
1795 /* check for exists after object is locked */
1796 if (mdt_object_exists(o) == 0) {
1797 /* Non-existent object shouldn't have PDO lock */
1800 /* Non-dir object shouldn't have PDO lock */
1801 LASSERT(S_ISDIR(lu_object_attr(&o->mot_obj.mo_lu)));
1806 mdt_object_unlock(info, o, lh, 1);
1813 * Just call ldlm_lock_decref() if decref, else we only call ptlrpc_save_lock()
1814 * to save this lock in req. when transaction committed, req will be released,
1815 * and lock will, too.
1817 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
1818 struct mdt_lock_handle *lh, int decref)
1820 struct ptlrpc_request *req = mdt_info_req(info);
1823 if (lustre_handle_is_used(&lh->mlh_pdo_lh)) {
1824 /* Do not save PDO locks to request, just decref. */
1825 mdt_fid_unlock(&lh->mlh_pdo_lh,
1827 lh->mlh_pdo_lh.cookie = 0ull;
1830 if (lustre_handle_is_used(&lh->mlh_reg_lh)) {
1832 mdt_fid_unlock(&lh->mlh_reg_lh,
1835 ptlrpc_save_lock(req, &lh->mlh_reg_lh,
1838 lh->mlh_reg_lh.cookie = 0ull;
1844 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
1845 const struct lu_fid *f,
1846 struct mdt_lock_handle *lh,
1849 struct mdt_object *o;
1851 o = mdt_object_find(info->mti_env, info->mti_mdt, f);
1855 rc = mdt_object_lock(info, o, lh, ibits,
1858 mdt_object_put(info->mti_env, o);
1865 void mdt_object_unlock_put(struct mdt_thread_info * info,
1866 struct mdt_object * o,
1867 struct mdt_lock_handle *lh,
1870 mdt_object_unlock(info, o, lh, decref);
1871 mdt_object_put(info->mti_env, o);
1874 static struct mdt_handler *mdt_handler_find(__u32 opc,
1875 struct mdt_opc_slice *supported)
1877 struct mdt_opc_slice *s;
1878 struct mdt_handler *h;
1881 for (s = supported; s->mos_hs != NULL; s++) {
1882 if (s->mos_opc_start <= opc && opc < s->mos_opc_end) {
1883 h = s->mos_hs + (opc - s->mos_opc_start);
1884 if (likely(h->mh_opc != 0))
1885 LASSERT(h->mh_opc == opc);
1887 h = NULL; /* unsupported opc */
1894 static int mdt_lock_resname_compat(struct mdt_device *m,
1895 struct ldlm_request *req)
1897 /* XXX something... later. */
1901 static int mdt_lock_reply_compat(struct mdt_device *m, struct ldlm_reply *rep)
1903 /* XXX something... later. */
1908 * Generic code handling requests that have struct mdt_body passed in:
1910 * - extract mdt_body from request and save it in @info, if present;
1912 * - create lu_object, corresponding to the fid in mdt_body, and save it in
1915 * - if HABEO_CORPUS flag is set for this request type check whether object
1916 * actually exists on storage (lu_object_exists()).
1919 static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags)
1921 const struct mdt_body *body;
1922 struct mdt_object *obj;
1923 const struct lu_env *env;
1924 struct req_capsule *pill;
1927 env = info->mti_env;
1928 pill = &info->mti_pill;
1930 body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
1934 if (!fid_is_sane(&body->fid1)) {
1935 CERROR("Invalid fid: "DFID"\n", PFID(&body->fid1));
1940 * Do not get size or any capa fields before we check that request
1941 * contains capa actually. There are some requests which do not, for
1942 * instance MDS_IS_SUBDIR.
1944 if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_CLIENT) &&
1945 req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1946 mdt_set_capainfo(info, 0, &body->fid1,
1947 req_capsule_client_get(pill, &RMF_CAPA1));
1949 obj = mdt_object_find(env, info->mti_mdt, &body->fid1);
1951 if ((flags & HABEO_CORPUS) &&
1952 !mdt_object_exists(obj)) {
1953 mdt_object_put(env, obj);
1954 /* for capability renew ENOENT will be handled in
1956 if (body->valid & OBD_MD_FLOSSCAPA)
1961 info->mti_object = obj;
1970 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags)
1972 struct req_capsule *pill;
1976 pill = &info->mti_pill;
1978 if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
1979 rc = mdt_body_unpack(info, flags);
1983 if (rc == 0 && (flags & HABEO_REFERO)) {
1984 struct mdt_device *mdt = info->mti_mdt;
1987 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1988 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1989 mdt->mdt_max_mdsize);
1990 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1991 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER,
1992 mdt->mdt_max_cookiesize);
1994 rc = req_capsule_pack(pill);
1999 static int mdt_init_capa_ctxt(const struct lu_env *env, struct mdt_device *m)
2001 struct md_device *next = m->mdt_child;
2003 return next->md_ops->mdo_init_capa_ctxt(env, next,
2004 m->mdt_opts.mo_mds_capa,
2005 m->mdt_capa_timeout,
2011 * Invoke handler for this request opc. Also do necessary preprocessing
2012 * (according to handler ->mh_flags), and post-processing (setting of
2013 * ->last_{xid,committed}).
2015 static int mdt_req_handle(struct mdt_thread_info *info,
2016 struct mdt_handler *h, struct ptlrpc_request *req)
2018 int rc, serious = 0;
2023 LASSERT(h->mh_act != NULL);
2024 LASSERT(h->mh_opc == lustre_msg_get_opc(req->rq_reqmsg));
2025 LASSERT(current->journal_info == NULL);
2028 * Mask out OBD_FAIL_ONCE, because that will stop
2029 * correct handling of failed req later in ldlm due to doing
2030 * obd_fail_loc |= OBD_FAIL_ONCE | OBD_FAILED without actually
2031 * correct actions like it is done in target_send_reply_msg().
2033 if (h->mh_fail_id != 0) {
2035 * Set to info->mti_fail_id to handler fail_id, it will be used
2036 * later, and better than use default fail_id.
2038 if (OBD_FAIL_CHECK(h->mh_fail_id && OBD_FAIL_MASK_LOC)) {
2039 info->mti_fail_id = h->mh_fail_id;
2045 flags = h->mh_flags;
2046 LASSERT(ergo(flags & (HABEO_CORPUS|HABEO_REFERO), h->mh_fmt != NULL));
2048 if (h->mh_fmt != NULL) {
2049 req_capsule_set(&info->mti_pill, h->mh_fmt);
2050 rc = mdt_unpack_req_pack_rep(info, flags);
2053 if (rc == 0 && flags & MUTABOR &&
2054 req->rq_export->exp_connect_flags & OBD_CONNECT_RDONLY)
2055 /* should it be rq_status? */
2058 if (rc == 0 && flags & HABEO_CLAVIS) {
2059 struct ldlm_request *dlm_req;
2061 LASSERT(h->mh_fmt != NULL);
2063 dlm_req = req_capsule_client_get(&info->mti_pill, &RMF_DLM_REQ);
2064 if (dlm_req != NULL) {
2065 if (info->mti_mdt->mdt_opts.mo_compat_resname)
2066 rc = mdt_lock_resname_compat(info->mti_mdt,
2068 info->mti_dlm_req = dlm_req;
2070 CERROR("Can't unpack dlm request\n");
2075 /* capability setting changed via /proc, needs reinitialize ctxt */
2076 if (info->mti_mdt && info->mti_mdt->mdt_capa_conf) {
2077 mdt_init_capa_ctxt(info->mti_env, info->mti_mdt);
2078 info->mti_mdt->mdt_capa_conf = 0;
2081 if (likely(rc == 0)) {
2083 * Process request, there can be two types of rc:
2084 * 1) errors with msg unpack/pack, other failures outside the
2085 * operation itself. This is counted as serious errors;
2086 * 2) errors during fs operation, should be placed in rq_status
2089 rc = h->mh_act(info);
2090 serious = is_serious(rc);
2091 rc = clear_serious(rc);
2095 req->rq_status = rc;
2098 * ELDLM_* codes which > 0 should be in rq_status only as well as
2099 * all non-serious errors.
2101 if (rc > 0 || !serious)
2104 LASSERT(current->journal_info == NULL);
2106 if (rc == 0 && (flags & HABEO_CLAVIS) &&
2107 info->mti_mdt->mdt_opts.mo_compat_resname) {
2108 struct ldlm_reply *dlmrep;
2110 dlmrep = req_capsule_server_get(&info->mti_pill, &RMF_DLM_REP);
2112 rc = mdt_lock_reply_compat(info->mti_mdt, dlmrep);
2115 /* If we're DISCONNECTing, the mdt_export_data is already freed */
2116 if (likely(rc == 0 && h->mh_opc != MDS_DISCONNECT))
2117 target_committed_to_req(req);
2119 if (unlikely((lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) &&
2120 lustre_msg_get_transno(req->rq_reqmsg) == 0)) {
2121 DEBUG_REQ(D_ERROR, req, "transno is 0 during REPLAY");
2128 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
2130 lh->mlh_type = MDT_NUL_LOCK;
2131 lh->mlh_reg_lh.cookie = 0ull;
2132 lh->mlh_reg_mode = LCK_MINMODE;
2133 lh->mlh_pdo_lh.cookie = 0ull;
2134 lh->mlh_pdo_mode = LCK_MINMODE;
2137 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
2139 LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2140 LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2144 * Initialize fields of struct mdt_thread_info. Other fields are left in
2145 * uninitialized state, because it's too expensive to zero out whole
2146 * mdt_thread_info (> 1K) on each request arrival.
2148 static void mdt_thread_info_init(struct ptlrpc_request *req,
2149 struct mdt_thread_info *info)
2152 struct md_capainfo *ci;
2154 info->mti_rep_buf_nr = ARRAY_SIZE(info->mti_rep_buf_size);
2155 for (i = 0; i < ARRAY_SIZE(info->mti_rep_buf_size); i++)
2156 info->mti_rep_buf_size[i] = -1;
2157 req_capsule_init(&info->mti_pill, req, RCL_SERVER,
2158 info->mti_rep_buf_size);
2161 for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2162 mdt_lock_handle_init(&info->mti_lh[i]);
2164 /* mdt device: it can be NULL while CONNECT */
2165 if (req->rq_export) {
2166 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
2167 info->mti_exp = req->rq_export;
2169 info->mti_mdt = NULL;
2170 info->mti_env = req->rq_svc_thread->t_env;
2171 ci = md_capainfo(info->mti_env);
2172 memset(ci, 0, sizeof *ci);
2174 info->mti_fail_id = OBD_FAIL_MDS_ALL_REPLY_NET;
2175 info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
2177 memset(&info->mti_attr, 0, sizeof(info->mti_attr));
2178 info->mti_body = NULL;
2179 info->mti_object = NULL;
2180 info->mti_dlm_req = NULL;
2181 info->mti_has_trans = 0;
2182 info->mti_no_need_trans = 0;
2183 info->mti_cross_ref = 0;
2184 info->mti_opdata = 0;
2186 /* To not check for split by default. */
2187 info->mti_spec.sp_ck_split = 0;
2190 static void mdt_thread_info_fini(struct mdt_thread_info *info)
2194 req_capsule_fini(&info->mti_pill);
2195 if (info->mti_object != NULL) {
2196 mdt_object_put(info->mti_env, info->mti_object);
2197 info->mti_object = NULL;
2199 for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2200 mdt_lock_handle_fini(&info->mti_lh[i]);
2201 info->mti_env = NULL;
2205 extern int mds_filter_recovery_request(struct ptlrpc_request *req,
2206 struct obd_device *obd, int *process);
2208 * Handle recovery. Return:
2209 * +1: continue request processing;
2210 * -ve: abort immediately with the given error code;
2211 * 0: send reply with error code in req->rq_status;
2213 static int mdt_recovery(struct mdt_thread_info *info)
2215 struct ptlrpc_request *req = mdt_info_req(info);
2217 struct obd_device *obd;
2221 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2224 case SEC_CTX_INIT_CONT:
2230 rc = mdt_handle_idmap(info);
2239 if (unlikely(req->rq_export == NULL)) {
2240 CERROR("operation %d on unconnected MDS from %s\n",
2241 lustre_msg_get_opc(req->rq_reqmsg),
2242 libcfs_id2str(req->rq_peer));
2243 /* FIXME: For CMD cleanup, when mds_B stop, the req from
2244 * mds_A will get -ENOTCONN(especially for ping req),
2245 * which will cause that mds_A deactive timeout, then when
2246 * mds_A cleanup, the cleanup process will be suspended since
2247 * deactive timeout is not zero.
2249 req->rq_status = -ENOTCONN;
2250 target_send_reply(req, -ENOTCONN, info->mti_fail_id);
2254 /* sanity check: if the xid matches, the request must be marked as a
2255 * resent or replayed */
2256 if (req_xid_is_last(req)) {
2257 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
2258 (MSG_RESENT | MSG_REPLAY))) {
2259 DEBUG_REQ(D_WARNING, req, "rq_xid "LPU64" matches last_xid, "
2260 "expected REPLAY or RESENT flag (%x)", req->rq_xid,
2261 lustre_msg_get_flags(req->rq_reqmsg));
2263 req->rq_status = -ENOTCONN;
2268 /* else: note the opposite is not always true; a RESENT req after a
2269 * failover will usually not match the last_xid, since it was likely
2270 * never committed. A REPLAYed request will almost never match the
2271 * last xid, however it could for a committed, but still retained,
2274 obd = req->rq_export->exp_obd;
2276 /* Check for aborted recovery... */
2277 spin_lock_bh(&obd->obd_processing_task_lock);
2278 recovering = obd->obd_recovering;
2279 spin_unlock_bh(&obd->obd_processing_task_lock);
2280 if (unlikely(recovering)) {
2283 DEBUG_REQ(D_INFO, req, "Got new replay");
2284 rc = mds_filter_recovery_request(req, obd, &should_process);
2285 if (rc != 0 || !should_process)
2287 else if (should_process < 0) {
2288 req->rq_status = should_process;
2289 rc = ptlrpc_error(req);
2296 static int mdt_reply(struct ptlrpc_request *req, int rc,
2297 struct mdt_thread_info *info)
2302 if (req->rq_reply_state == NULL && rc == 0) {
2303 req->rq_status = rc;
2304 lustre_pack_reply(req, 1, NULL, NULL);
2307 target_send_reply(req, rc, info->mti_fail_id);
2312 extern int mds_msg_check_version(struct lustre_msg *msg);
2314 static int mdt_handle0(struct ptlrpc_request *req,
2315 struct mdt_thread_info *info,
2316 struct mdt_opc_slice *supported)
2318 struct mdt_handler *h;
2319 struct lustre_msg *msg;
2324 OBD_FAIL_RETURN(OBD_FAIL_MDS_ALL_REQUEST_NET | OBD_FAIL_ONCE, 0);
2326 LASSERT(current->journal_info == NULL);
2328 msg = req->rq_reqmsg;
2329 rc = mds_msg_check_version(msg);
2330 if (likely(rc == 0)) {
2331 rc = mdt_recovery(info);
2332 if (likely(rc == +1)) {
2333 h = mdt_handler_find(lustre_msg_get_opc(msg),
2335 if (likely(h != NULL)) {
2336 rc = mdt_req_handle(info, h, req);
2337 rc = mdt_reply(req, rc, info);
2339 CERROR("The unsupported opc: 0x%x\n", lustre_msg_get_opc(msg) );
2340 req->rq_status = -ENOTSUPP;
2341 rc = ptlrpc_error(req);
2346 CERROR(LUSTRE_MDT_NAME" drops mal-formed request\n");
2351 * MDT handler function called by ptlrpc service thread when request comes.
2353 * XXX common "target" functionality should be factored into separate module
2354 * shared by mdt, ost and stand-alone services like fld.
2356 static int mdt_handle_common(struct ptlrpc_request *req,
2357 struct mdt_opc_slice *supported)
2360 struct mdt_thread_info *info;
2364 env = req->rq_svc_thread->t_env;
2365 LASSERT(env != NULL);
2366 LASSERT(env->le_ses != NULL);
2367 LASSERT(env->le_ctx.lc_thread == req->rq_svc_thread);
2368 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
2369 LASSERT(info != NULL);
2371 mdt_thread_info_init(req, info);
2373 rc = mdt_handle0(req, info, supported);
2375 mdt_thread_info_fini(info);
2380 * This is called from recovery code as handler of _all_ RPC types, FLD and SEQ
2383 int mdt_recovery_handle(struct ptlrpc_request *req)
2388 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2390 rc = mdt_handle_common(req, mdt_fld_handlers);
2393 rc = mdt_handle_common(req, mdt_seq_handlers);
2396 rc = mdt_handle_common(req, mdt_regular_handlers);
2403 static int mdt_regular_handle(struct ptlrpc_request *req)
2405 return mdt_handle_common(req, mdt_regular_handlers);
2408 static int mdt_readpage_handle(struct ptlrpc_request *req)
2410 return mdt_handle_common(req, mdt_readpage_handlers);
2413 static int mdt_xmds_handle(struct ptlrpc_request *req)
2415 return mdt_handle_common(req, mdt_xmds_handlers);
2418 static int mdt_mdsc_handle(struct ptlrpc_request *req)
2420 return mdt_handle_common(req, mdt_seq_handlers);
2423 static int mdt_mdss_handle(struct ptlrpc_request *req)
2425 return mdt_handle_common(req, mdt_seq_handlers);
2428 static int mdt_dtss_handle(struct ptlrpc_request *req)
2430 return mdt_handle_common(req, mdt_seq_handlers);
2433 static int mdt_fld_handle(struct ptlrpc_request *req)
2435 return mdt_handle_common(req, mdt_fld_handlers);
2451 static int mdt_intent_getattr(enum mdt_it_code opcode,
2452 struct mdt_thread_info *info,
2453 struct ldlm_lock **,
2455 static int mdt_intent_reint(enum mdt_it_code opcode,
2456 struct mdt_thread_info *info,
2457 struct ldlm_lock **,
2460 static struct mdt_it_flavor {
2461 const struct req_format *it_fmt;
2463 int (*it_act)(enum mdt_it_code ,
2464 struct mdt_thread_info *,
2465 struct ldlm_lock **,
2468 } mdt_it_flavor[] = {
2470 .it_fmt = &RQF_LDLM_INTENT,
2471 /*.it_flags = HABEO_REFERO,*/
2473 .it_act = mdt_intent_reint,
2474 .it_reint = REINT_OPEN
2477 .it_fmt = &RQF_LDLM_INTENT,
2478 .it_flags = MUTABOR,
2479 .it_act = mdt_intent_reint,
2480 .it_reint = REINT_OPEN
2483 .it_fmt = &RQF_LDLM_INTENT,
2484 .it_flags = MUTABOR,
2485 .it_act = mdt_intent_reint,
2486 .it_reint = REINT_CREATE
2488 [MDT_IT_GETATTR] = {
2489 .it_fmt = &RQF_LDLM_INTENT_GETATTR,
2490 .it_flags = HABEO_REFERO,
2491 .it_act = mdt_intent_getattr
2493 [MDT_IT_READDIR] = {
2499 .it_fmt = &RQF_LDLM_INTENT_GETATTR,
2500 .it_flags = HABEO_REFERO,
2501 .it_act = mdt_intent_getattr
2504 .it_fmt = &RQF_LDLM_INTENT_UNLINK,
2505 .it_flags = MUTABOR,
2507 .it_reint = REINT_UNLINK
2511 .it_flags = MUTABOR,
2514 [MDT_IT_GETXATTR] = {
2521 int mdt_intent_lock_replace(struct mdt_thread_info *info,
2522 struct ldlm_lock **lockp,
2523 struct ldlm_lock *new_lock,
2524 struct mdt_lock_handle *lh,
2527 struct ptlrpc_request *req = mdt_info_req(info);
2528 struct ldlm_lock *lock = *lockp;
2531 * Get new lock only for cases when possible resent did not find any
2534 if (new_lock == NULL)
2535 new_lock = ldlm_handle2lock(&lh->mlh_reg_lh);
2537 if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
2538 lh->mlh_reg_lh.cookie = 0;
2542 LASSERTF(new_lock != NULL,
2543 "lockh "LPX64"\n", lh->mlh_reg_lh.cookie);
2546 * If we've already given this lock to a client once, then we should
2547 * have no readers or writers. Otherwise, we should have one reader
2548 * _or_ writer ref (which will be zeroed below) before returning the
2551 if (new_lock->l_export == req->rq_export) {
2552 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
2554 LASSERT(new_lock->l_export == NULL);
2555 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
2560 if (new_lock->l_export == req->rq_export) {
2562 * Already gave this to the client, which means that we
2563 * reconstructed a reply.
2565 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
2567 lh->mlh_reg_lh.cookie = 0;
2568 RETURN(ELDLM_LOCK_REPLACED);
2571 /* This lock might already be given to the client by an resent req,
2572 * in this case we should return ELDLM_LOCK_ABORTED,
2573 * so we should check led_held_locks here, but it will affect
2574 * performance, FIXME
2576 /* Fixup the lock to be given to the client */
2577 lock_res_and_lock(new_lock);
2578 new_lock->l_readers = 0;
2579 new_lock->l_writers = 0;
2581 new_lock->l_export = class_export_get(req->rq_export);
2582 spin_lock(&req->rq_export->exp_ldlm_data.led_lock);
2583 list_add(&new_lock->l_export_chain,
2584 &new_lock->l_export->exp_ldlm_data.led_held_locks);
2585 spin_unlock(&req->rq_export->exp_ldlm_data.led_lock);
2587 new_lock->l_blocking_ast = lock->l_blocking_ast;
2588 new_lock->l_completion_ast = lock->l_completion_ast;
2589 new_lock->l_remote_handle = lock->l_remote_handle;
2590 new_lock->l_flags &= ~LDLM_FL_LOCAL;
2592 unlock_res_and_lock(new_lock);
2593 LDLM_LOCK_PUT(new_lock);
2594 lh->mlh_reg_lh.cookie = 0;
2596 RETURN(ELDLM_LOCK_REPLACED);
2599 static void mdt_intent_fixup_resent(struct mdt_thread_info *info,
2600 struct ldlm_lock *new_lock,
2601 struct ldlm_lock **old_lock,
2602 struct mdt_lock_handle *lh)
2604 struct ptlrpc_request *req = mdt_info_req(info);
2605 struct obd_export *exp = req->rq_export;
2606 struct lustre_handle remote_hdl;
2607 struct ldlm_request *dlmreq;
2608 struct list_head *iter;
2610 if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
2613 dlmreq = req_capsule_client_get(&info->mti_pill, &RMF_DLM_REQ);
2614 remote_hdl = dlmreq->lock_handle[0];
2616 spin_lock(&exp->exp_ldlm_data.led_lock);
2617 list_for_each(iter, &exp->exp_ldlm_data.led_held_locks) {
2618 struct ldlm_lock *lock;
2619 lock = list_entry(iter, struct ldlm_lock, l_export_chain);
2620 if (lock == new_lock)
2622 if (lock->l_remote_handle.cookie == remote_hdl.cookie) {
2623 lh->mlh_reg_lh.cookie = lock->l_handle.h_cookie;
2624 lh->mlh_reg_mode = lock->l_granted_mode;
2626 LDLM_DEBUG(lock, "restoring lock cookie");
2627 DEBUG_REQ(D_DLMTRACE, req,
2628 "restoring lock cookie "LPX64,
2629 lh->mlh_reg_lh.cookie);
2631 *old_lock = LDLM_LOCK_GET(lock);
2632 spin_unlock(&exp->exp_ldlm_data.led_lock);
2636 spin_unlock(&exp->exp_ldlm_data.led_lock);
2639 * If the xid matches, then we know this is a resent request, and allow
2640 * it. (It's probably an OPEN, for which we don't send a lock.
2642 if (req_xid_is_last(req))
2646 * This remote handle isn't enqueued, so we never received or processed
2647 * this request. Clear MSG_RESENT, because it can be handled like any
2648 * normal request now.
2650 lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
2652 DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle "LPX64,
2656 static int mdt_intent_getattr(enum mdt_it_code opcode,
2657 struct mdt_thread_info *info,
2658 struct ldlm_lock **lockp,
2661 struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
2662 struct ldlm_lock *new_lock = NULL;
2664 struct ldlm_reply *ldlm_rep;
2665 struct ptlrpc_request *req;
2666 struct mdt_body *reqbody;
2667 struct mdt_body *repbody;
2671 reqbody = req_capsule_client_get(&info->mti_pill, &RMF_MDT_BODY);
2674 repbody = req_capsule_server_get(&info->mti_pill, &RMF_MDT_BODY);
2677 info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
2678 info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
2679 repbody->eadatasize = 0;
2680 repbody->aclsize = 0;
2684 child_bits = MDS_INODELOCK_LOOKUP;
2686 case MDT_IT_GETATTR:
2687 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE;
2690 CERROR("Unhandled till now");
2691 GOTO(out, rc = -EINVAL);
2694 rc = mdt_init_ucred(info, reqbody);
2698 req = info->mti_pill.rc_req;
2699 ldlm_rep = req_capsule_server_get(&info->mti_pill, &RMF_DLM_REP);
2700 mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
2702 /* Get lock from request for possible resent case. */
2703 mdt_intent_fixup_resent(info, *lockp, &new_lock, lhc);
2705 ldlm_rep->lock_policy_res2 =
2706 mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
2708 if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
2709 ldlm_rep->lock_policy_res2 = 0;
2710 if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
2711 ldlm_rep->lock_policy_res2) {
2712 lhc->mlh_reg_lh.cookie = 0ull;
2713 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
2716 rc = mdt_intent_lock_replace(info, lockp, new_lock, lhc, flags);
2719 mdt_exit_ucred(info);
2721 mdt_shrink_reply(info);
2725 static int mdt_intent_reint(enum mdt_it_code opcode,
2726 struct mdt_thread_info *info,
2727 struct ldlm_lock **lockp,
2730 struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
2731 struct ldlm_reply *rep = NULL;
2735 static const struct req_format *intent_fmts[REINT_MAX] = {
2736 [REINT_CREATE] = &RQF_LDLM_INTENT_CREATE,
2737 [REINT_OPEN] = &RQF_LDLM_INTENT_OPEN
2742 opc = mdt_reint_opcode(info, intent_fmts);
2744 GOTO(out, rc = opc);
2746 if (mdt_it_flavor[opcode].it_reint != opc) {
2747 CERROR("Reint code %ld doesn't match intent: %d\n",
2749 GOTO(out, rc = err_serious(-EPROTO));
2752 /* Get lock from request for possible resent case. */
2753 mdt_intent_fixup_resent(info, *lockp, NULL, lhc);
2755 rc = mdt_reint_internal(info, lhc, opc);
2757 /* Check whether the reply has been packed successfully. */
2758 if (mdt_info_req(info)->rq_repmsg != NULL)
2759 rep = req_capsule_server_get(&info->mti_pill, &RMF_DLM_REP);
2761 GOTO(out, rc = err_serious(-EFAULT));
2763 /* MDC expects this in any case */
2765 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
2767 /* Cross-ref case, the lock should be returned to the client */
2768 if (rc == -EREMOTE) {
2769 LASSERT(lustre_handle_is_used(&lhc->mlh_reg_lh));
2770 rep->lock_policy_res2 = 0;
2771 rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
2774 rep->lock_policy_res2 = clear_serious(rc);
2776 lhc->mlh_reg_lh.cookie = 0ull;
2777 rc = ELDLM_LOCK_ABORTED;
2783 static int mdt_intent_code(long itcode)
2791 case IT_OPEN|IT_CREAT:
2798 rc = MDT_IT_READDIR;
2801 rc = MDT_IT_GETATTR;
2813 rc = MDT_IT_GETXATTR;
2816 CERROR("Unknown intent opcode: %ld\n", itcode);
2823 static int mdt_intent_opc(long itopc, struct mdt_thread_info *info,
2824 struct ldlm_lock **lockp, int flags)
2826 struct req_capsule *pill;
2827 struct mdt_it_flavor *flv;
2832 opc = mdt_intent_code(itopc);
2836 pill = &info->mti_pill;
2837 flv = &mdt_it_flavor[opc];
2839 if (flv->it_fmt != NULL)
2840 req_capsule_extend(pill, flv->it_fmt);
2842 rc = mdt_unpack_req_pack_rep(info, flv->it_flags);
2844 struct ptlrpc_request *req = mdt_info_req(info);
2845 if (flv->it_flags & MUTABOR &&
2846 req->rq_export->exp_connect_flags & OBD_CONNECT_RDONLY)
2849 if (rc == 0 && flv->it_act != NULL) {
2850 /* execute policy */
2851 rc = flv->it_act(opc, info, lockp, flags);
2857 static int mdt_intent_policy(struct ldlm_namespace *ns,
2858 struct ldlm_lock **lockp, void *req_cookie,
2859 ldlm_mode_t mode, int flags, void *data)
2861 struct mdt_thread_info *info;
2862 struct ptlrpc_request *req = req_cookie;
2863 struct ldlm_intent *it;
2864 struct req_capsule *pill;
2869 LASSERT(req != NULL);
2871 info = lu_context_key_get(&req->rq_svc_thread->t_env->le_ctx,
2873 LASSERT(info != NULL);
2874 pill = &info->mti_pill;
2875 LASSERT(pill->rc_req == req);
2877 if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
2878 req_capsule_extend(pill, &RQF_LDLM_INTENT);
2879 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
2881 const struct ldlm_request *dlmreq;
2884 struct ldlm_lock *lock = *lockp;
2886 LDLM_DEBUG(lock, "intent policy opc: %s\n",
2887 ldlm_it2str(it->opc));
2890 rc = mdt_intent_opc(it->opc, info, lockp, flags);
2895 * Lock without inodebits makes no sense and will oops
2896 * later in ldlm. Let's check it now to see if we have
2897 * wrong lock from client or bits get corrupted
2898 * somewhere in mdt_intent_opc().
2900 dlmreq = info->mti_dlm_req;
2901 req_bits = dlmreq->lock_desc.l_policy_data.l_inodebits.bits;
2902 LASSERT(req_bits != 0);
2905 rc = err_serious(-EFAULT);
2907 /* No intent was provided */
2908 LASSERT(pill->rc_fmt == &RQF_LDLM_ENQUEUE);
2909 rc = req_capsule_pack(pill);
2911 rc = err_serious(rc);
2919 static void mdt_seq_adjust(const struct lu_env *env,
2920 struct mdt_device *m, int lost)
2922 struct lu_site *ls = m->mdt_md_dev.md_lu_dev.ld_site;
2923 struct lu_range out;
2926 LASSERT(ls && ls->ls_server_seq);
2928 /* get extra seq from seq_server, moving it's range up */
2929 while (lost-- > 0) {
2930 seq_server_alloc_meta(ls->ls_server_seq, NULL, &out, env);
2935 static int mdt_seq_fini(const struct lu_env *env,
2936 struct mdt_device *m)
2938 struct lu_site *ls = m->mdt_md_dev.md_lu_dev.ld_site;
2941 if (ls && ls->ls_server_seq) {
2942 seq_server_fini(ls->ls_server_seq, env);
2943 OBD_FREE_PTR(ls->ls_server_seq);
2944 ls->ls_server_seq = NULL;
2947 if (ls && ls->ls_control_seq) {
2948 seq_server_fini(ls->ls_control_seq, env);
2949 OBD_FREE_PTR(ls->ls_control_seq);
2950 ls->ls_control_seq = NULL;
2953 if (ls && ls->ls_client_seq) {
2954 seq_client_fini(ls->ls_client_seq);
2955 OBD_FREE_PTR(ls->ls_client_seq);
2956 ls->ls_client_seq = NULL;
2962 static int mdt_seq_init(const struct lu_env *env,
2964 struct mdt_device *m)
2971 ls = m->mdt_md_dev.md_lu_dev.ld_site;
2974 * This is sequence-controller node. Init seq-controller server on local
2977 if (ls->ls_node_id == 0) {
2978 LASSERT(ls->ls_control_seq == NULL);
2980 OBD_ALLOC_PTR(ls->ls_control_seq);
2981 if (ls->ls_control_seq == NULL)
2984 rc = seq_server_init(ls->ls_control_seq,
2985 m->mdt_bottom, uuid,
2986 LUSTRE_SEQ_CONTROLLER,
2990 GOTO(out_seq_fini, rc);
2992 OBD_ALLOC_PTR(ls->ls_client_seq);
2993 if (ls->ls_client_seq == NULL)
2994 GOTO(out_seq_fini, rc = -ENOMEM);
2996 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
2997 if (prefix == NULL) {
2998 OBD_FREE_PTR(ls->ls_client_seq);
2999 GOTO(out_seq_fini, rc = -ENOMEM);
3002 snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s",
3006 * Init seq-controller client after seq-controller server is
3007 * ready. Pass ls->ls_control_seq to it for direct talking.
3009 rc = seq_client_init(ls->ls_client_seq, NULL,
3010 LUSTRE_SEQ_METADATA, prefix,
3011 ls->ls_control_seq);
3012 OBD_FREE(prefix, MAX_OBD_NAME + 5);
3015 GOTO(out_seq_fini, rc);
3018 /* Init seq-server on local MDT */
3019 LASSERT(ls->ls_server_seq == NULL);
3021 OBD_ALLOC_PTR(ls->ls_server_seq);
3022 if (ls->ls_server_seq == NULL)
3023 GOTO(out_seq_fini, rc = -ENOMEM);
3025 rc = seq_server_init(ls->ls_server_seq,
3026 m->mdt_bottom, uuid,
3030 GOTO(out_seq_fini, rc = -ENOMEM);
3032 /* Assign seq-controller client to local seq-server. */
3033 if (ls->ls_node_id == 0) {
3034 LASSERT(ls->ls_client_seq != NULL);
3036 rc = seq_server_set_cli(ls->ls_server_seq,
3044 mdt_seq_fini(env, m);
3049 * Init client sequence manager which is used by local MDS to talk to sequence
3050 * controller on remote node.
3052 static int mdt_seq_init_cli(const struct lu_env *env,
3053 struct mdt_device *m,
3054 struct lustre_cfg *cfg)
3056 struct lu_site *ls = m->mdt_md_dev.md_lu_dev.ld_site;
3057 struct obd_device *mdc;
3058 struct obd_uuid *uuidp, *mdcuuidp;
3059 char *uuid_str, *mdc_uuid_str;
3062 struct mdt_thread_info *info;
3063 char *p, *index_string = lustre_cfg_string(cfg, 2);
3066 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3067 uuidp = &info->mti_u.uuid[0];
3068 mdcuuidp = &info->mti_u.uuid[1];
3070 LASSERT(index_string);
3072 index = simple_strtol(index_string, &p, 10);
3074 CERROR("Invalid index in lustre_cgf, offset 2\n");
3078 /* check if this is adding the first MDC and controller is not yet
3080 if (index != 0 || ls->ls_client_seq)
3083 uuid_str = lustre_cfg_string(cfg, 1);
3084 mdc_uuid_str = lustre_cfg_string(cfg, 4);
3085 obd_str2uuid(uuidp, uuid_str);
3086 obd_str2uuid(mdcuuidp, mdc_uuid_str);
3088 mdc = class_find_client_obd(uuidp, LUSTRE_MDC_NAME, mdcuuidp);
3090 CERROR("can't find controller MDC by uuid %s\n",
3093 } else if (!mdc->obd_set_up) {
3094 CERROR("target %s not set up\n", mdc->obd_name);
3097 LASSERT(ls->ls_control_exp);
3098 OBD_ALLOC_PTR(ls->ls_client_seq);
3099 if (ls->ls_client_seq != NULL) {
3102 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3106 snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s",
3109 rc = seq_client_init(ls->ls_client_seq,
3111 LUSTRE_SEQ_METADATA,
3113 OBD_FREE(prefix, MAX_OBD_NAME + 5);
3120 LASSERT(ls->ls_server_seq != NULL);
3121 rc = seq_server_set_cli(ls->ls_server_seq, ls->ls_client_seq,
3128 static void mdt_seq_fini_cli(struct mdt_device *m)
3134 ls = m->mdt_md_dev.md_lu_dev.ld_site;
3136 if (ls && ls->ls_server_seq)
3137 seq_server_set_cli(ls->ls_server_seq,
3140 if (ls && ls->ls_control_exp) {
3141 class_export_put(ls->ls_control_exp);
3142 ls->ls_control_exp = NULL;
3150 static int mdt_fld_fini(const struct lu_env *env,
3151 struct mdt_device *m)
3153 struct lu_site *ls = m->mdt_md_dev.md_lu_dev.ld_site;
3156 if (ls && ls->ls_server_fld) {
3157 fld_server_fini(ls->ls_server_fld, env);
3158 OBD_FREE_PTR(ls->ls_server_fld);
3159 ls->ls_server_fld = NULL;
3165 static int mdt_fld_init(const struct lu_env *env,
3167 struct mdt_device *m)
3173 ls = m->mdt_md_dev.md_lu_dev.ld_site;
3175 OBD_ALLOC_PTR(ls->ls_server_fld);
3176 if (ls->ls_server_fld == NULL)
3177 RETURN(rc = -ENOMEM);
3179 rc = fld_server_init(ls->ls_server_fld,
3180 m->mdt_bottom, uuid, env);
3182 OBD_FREE_PTR(ls->ls_server_fld);
3183 ls->ls_server_fld = NULL;
3190 /* device init/fini methods */
3191 static void mdt_stop_ptlrpc_service(struct mdt_device *m)
3194 if (m->mdt_regular_service != NULL) {
3195 ptlrpc_unregister_service(m->mdt_regular_service);
3196 m->mdt_regular_service = NULL;
3198 if (m->mdt_readpage_service != NULL) {
3199 ptlrpc_unregister_service(m->mdt_readpage_service);
3200 m->mdt_readpage_service = NULL;
3202 if (m->mdt_xmds_service != NULL) {
3203 ptlrpc_unregister_service(m->mdt_xmds_service);
3204 m->mdt_xmds_service = NULL;
3206 if (m->mdt_setattr_service != NULL) {
3207 ptlrpc_unregister_service(m->mdt_setattr_service);
3208 m->mdt_setattr_service = NULL;
3210 if (m->mdt_mdsc_service != NULL) {
3211 ptlrpc_unregister_service(m->mdt_mdsc_service);
3212 m->mdt_mdsc_service = NULL;
3214 if (m->mdt_mdss_service != NULL) {
3215 ptlrpc_unregister_service(m->mdt_mdss_service);
3216 m->mdt_mdss_service = NULL;
3218 if (m->mdt_dtss_service != NULL) {
3219 ptlrpc_unregister_service(m->mdt_dtss_service);
3220 m->mdt_dtss_service = NULL;
3222 if (m->mdt_fld_service != NULL) {
3223 ptlrpc_unregister_service(m->mdt_fld_service);
3224 m->mdt_fld_service = NULL;
3229 static int mdt_start_ptlrpc_service(struct mdt_device *m)
3232 static struct ptlrpc_service_conf conf;
3233 cfs_proc_dir_entry_t *procfs_entry;
3236 procfs_entry = m->mdt_md_dev.md_lu_dev.ld_obd->obd_proc_entry;
3238 conf = (typeof(conf)) {
3239 .psc_nbufs = MDS_NBUFS,
3240 .psc_bufsize = MDS_BUFSIZE,
3241 .psc_max_req_size = MDS_MAXREQSIZE,
3242 .psc_max_reply_size = MDS_MAXREPSIZE,
3243 .psc_req_portal = MDS_REQUEST_PORTAL,
3244 .psc_rep_portal = MDC_REPLY_PORTAL,
3245 .psc_watchdog_timeout = MDT_SERVICE_WATCHDOG_TIMEOUT,
3247 * We'd like to have a mechanism to set this on a per-device
3248 * basis, but alas...
3250 .psc_min_threads = min(max(mdt_num_threads, MDT_MIN_THREADS),
3252 .psc_max_threads = MDT_MAX_THREADS,
3253 .psc_ctx_tags = LCT_MD_THREAD
3256 m->mdt_ldlm_client = &m->mdt_md_dev.md_lu_dev.ld_obd->obd_ldlm_client;
3257 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
3258 "mdt_ldlm_client", m->mdt_ldlm_client);
3260 m->mdt_regular_service =
3261 ptlrpc_init_svc_conf(&conf, mdt_regular_handle, LUSTRE_MDT_NAME,
3262 procfs_entry, NULL, LUSTRE_MDT_NAME);
3263 if (m->mdt_regular_service == NULL)
3266 rc = ptlrpc_start_threads(NULL, m->mdt_regular_service);
3268 GOTO(err_mdt_svc, rc);
3271 * readpage service configuration. Parameters have to be adjusted,
3274 conf = (typeof(conf)) {
3275 .psc_nbufs = MDS_NBUFS,
3276 .psc_bufsize = MDS_BUFSIZE,
3277 .psc_max_req_size = MDS_MAXREQSIZE,
3278 .psc_max_reply_size = MDS_MAXREPSIZE,
3279 .psc_req_portal = MDS_READPAGE_PORTAL,
3280 .psc_rep_portal = MDC_REPLY_PORTAL,
3281 .psc_watchdog_timeout = MDT_SERVICE_WATCHDOG_TIMEOUT,
3282 .psc_min_threads = min(max(mdt_num_threads, MDT_MIN_THREADS),
3284 .psc_max_threads = MDT_MAX_THREADS,
3285 .psc_ctx_tags = LCT_MD_THREAD
3287 m->mdt_readpage_service =
3288 ptlrpc_init_svc_conf(&conf, mdt_readpage_handle,
3289 LUSTRE_MDT_NAME "_readpage",
3290 procfs_entry, NULL, "mdt_rdpg");
3292 if (m->mdt_readpage_service == NULL) {
3293 CERROR("failed to start readpage service\n");
3294 GOTO(err_mdt_svc, rc = -ENOMEM);
3297 rc = ptlrpc_start_threads(NULL, m->mdt_readpage_service);
3300 * setattr service configuration.
3302 conf = (typeof(conf)) {
3303 .psc_nbufs = MDS_NBUFS,
3304 .psc_bufsize = MDS_BUFSIZE,
3305 .psc_max_req_size = MDS_MAXREQSIZE,
3306 .psc_max_reply_size = MDS_MAXREPSIZE,
3307 .psc_req_portal = MDS_SETATTR_PORTAL,
3308 .psc_rep_portal = MDC_REPLY_PORTAL,
3309 .psc_watchdog_timeout = MDT_SERVICE_WATCHDOG_TIMEOUT,
3310 .psc_min_threads = min(max(mdt_num_threads, MDT_MIN_THREADS),
3312 .psc_max_threads = MDT_MAX_THREADS,
3313 .psc_ctx_tags = LCT_MD_THREAD
3316 m->mdt_setattr_service =
3317 ptlrpc_init_svc_conf(&conf, mdt_regular_handle,
3318 LUSTRE_MDT_NAME "_setattr",
3319 procfs_entry, NULL, "mdt_attr");
3321 if (!m->mdt_setattr_service) {
3322 CERROR("failed to start setattr service\n");
3323 GOTO(err_mdt_svc, rc = -ENOMEM);
3326 rc = ptlrpc_start_threads(NULL, m->mdt_setattr_service);
3328 GOTO(err_mdt_svc, rc);
3331 * sequence controller service configuration
3333 conf = (typeof(conf)) {
3334 .psc_nbufs = MDS_NBUFS,
3335 .psc_bufsize = MDS_BUFSIZE,
3336 .psc_max_req_size = SEQ_MAXREQSIZE,
3337 .psc_max_reply_size = SEQ_MAXREPSIZE,
3338 .psc_req_portal = SEQ_CONTROLLER_PORTAL,
3339 .psc_rep_portal = MDC_REPLY_PORTAL,
3340 .psc_watchdog_timeout = MDT_SERVICE_WATCHDOG_TIMEOUT,
3341 .psc_min_threads = SEQ_NUM_THREADS,
3342 .psc_max_threads = SEQ_NUM_THREADS,
3343 .psc_ctx_tags = LCT_MD_THREAD|LCT_DT_THREAD
3346 m->mdt_mdsc_service =
3347 ptlrpc_init_svc_conf(&conf, mdt_mdsc_handle,
3348 LUSTRE_MDT_NAME"_mdsc",
3349 procfs_entry, NULL, "mdt_mdsc");
3350 if (!m->mdt_mdsc_service) {
3351 CERROR("failed to start seq controller service\n");
3352 GOTO(err_mdt_svc, rc = -ENOMEM);
3355 rc = ptlrpc_start_threads(NULL, m->mdt_mdsc_service);
3357 GOTO(err_mdt_svc, rc);
3360 * metadata sequence server service configuration
3362 conf = (typeof(conf)) {
3363 .psc_nbufs = MDS_NBUFS,
3364 .psc_bufsize = MDS_BUFSIZE,
3365 .psc_max_req_size = SEQ_MAXREQSIZE,
3366 .psc_max_reply_size = SEQ_MAXREPSIZE,
3367 .psc_req_portal = SEQ_METADATA_PORTAL,
3368 .psc_rep_portal = MDC_REPLY_PORTAL,
3369 .psc_watchdog_timeout = MDT_SERVICE_WATCHDOG_TIMEOUT,
3370 .psc_min_threads = SEQ_NUM_THREADS,
3371 .psc_max_threads = SEQ_NUM_THREADS,
3372 .psc_ctx_tags = LCT_MD_THREAD|LCT_DT_THREAD
3375 m->mdt_mdss_service =
3376 ptlrpc_init_svc_conf(&conf, mdt_mdss_handle,
3377 LUSTRE_MDT_NAME"_mdss",
3378 procfs_entry, NULL, "mdt_mdss");
3379 if (!m->mdt_mdss_service) {
3380 CERROR("failed to start metadata seq server service\n");
3381 GOTO(err_mdt_svc, rc = -ENOMEM);
3384 rc = ptlrpc_start_threads(NULL, m->mdt_mdss_service);
3386 GOTO(err_mdt_svc, rc);
3390 * Data sequence server service configuration. We want to have really
3391 * cluster-wide sequences space. This is why we start only one sequence
3392 * controller which manages space.
3394 conf = (typeof(conf)) {
3395 .psc_nbufs = MDS_NBUFS,
3396 .psc_bufsize = MDS_BUFSIZE,
3397 .psc_max_req_size = SEQ_MAXREQSIZE,
3398 .psc_max_reply_size = SEQ_MAXREPSIZE,
3399 .psc_req_portal = SEQ_DATA_PORTAL,
3400 .psc_rep_portal = OSC_REPLY_PORTAL,
3401 .psc_watchdog_timeout = MDT_SERVICE_WATCHDOG_TIMEOUT,
3402 .psc_min_threads = SEQ_NUM_THREADS,
3403 .psc_max_threads = SEQ_NUM_THREADS,
3404 .psc_ctx_tags = LCT_MD_THREAD|LCT_DT_THREAD
3407 m->mdt_dtss_service =
3408 ptlrpc_init_svc_conf(&conf, mdt_dtss_handle,
3409 LUSTRE_MDT_NAME"_dtss",
3410 procfs_entry, NULL, "mdt_dtss");
3411 if (!m->mdt_dtss_service) {
3412 CERROR("failed to start data seq server service\n");
3413 GOTO(err_mdt_svc, rc = -ENOMEM);
3416 rc = ptlrpc_start_threads(NULL, m->mdt_dtss_service);
3418 GOTO(err_mdt_svc, rc);
3420 /* FLD service start */
3421 conf = (typeof(conf)) {
3422 .psc_nbufs = MDS_NBUFS,
3423 .psc_bufsize = MDS_BUFSIZE,
3424 .psc_max_req_size = FLD_MAXREQSIZE,
3425 .psc_max_reply_size = FLD_MAXREPSIZE,
3426 .psc_req_portal = FLD_REQUEST_PORTAL,
3427 .psc_rep_portal = MDC_REPLY_PORTAL,
3428 .psc_watchdog_timeout = MDT_SERVICE_WATCHDOG_TIMEOUT,
3429 .psc_min_threads = FLD_NUM_THREADS,
3430 .psc_max_threads = FLD_NUM_THREADS,
3431 .psc_ctx_tags = LCT_DT_THREAD|LCT_MD_THREAD
3434 m->mdt_fld_service =
3435 ptlrpc_init_svc_conf(&conf, mdt_fld_handle,
3436 LUSTRE_MDT_NAME"_fld",
3437 procfs_entry, NULL, "mdt_fld");
3438 if (!m->mdt_fld_service) {
3439 CERROR("failed to start fld service\n");
3440 GOTO(err_mdt_svc, rc = -ENOMEM);
3443 rc = ptlrpc_start_threads(NULL, m->mdt_fld_service);
3445 GOTO(err_mdt_svc, rc);
3448 * mds-mds service configuration. Separate portal is used to allow
3449 * mds-mds requests be not blocked during recovery.
3451 conf = (typeof(conf)) {
3452 .psc_nbufs = MDS_NBUFS,
3453 .psc_bufsize = MDS_BUFSIZE,
3454 .psc_max_req_size = MDS_MAXREQSIZE,
3455 .psc_max_reply_size = MDS_MAXREPSIZE,
3456 .psc_req_portal = MDS_MDS_PORTAL,
3457 .psc_rep_portal = MDC_REPLY_PORTAL,
3458 .psc_watchdog_timeout = MDT_SERVICE_WATCHDOG_TIMEOUT,
3459 .psc_min_threads = min(max(mdt_num_threads, MDT_MIN_THREADS),
3461 .psc_max_threads = MDT_MAX_THREADS,
3462 .psc_ctx_tags = LCT_MD_THREAD
3464 m->mdt_xmds_service = ptlrpc_init_svc_conf(&conf, mdt_xmds_handle,
3465 LUSTRE_MDT_NAME "_mds",
3466 procfs_entry, NULL, "mdt_xmds");
3468 if (m->mdt_xmds_service == NULL) {
3469 CERROR("failed to start readpage service\n");
3470 GOTO(err_mdt_svc, rc = -ENOMEM);
3473 rc = ptlrpc_start_threads(NULL, m->mdt_xmds_service);
3475 GOTO(err_mdt_svc, rc);
3480 mdt_stop_ptlrpc_service(m);
3485 static void mdt_stack_fini(const struct lu_env *env,
3486 struct mdt_device *m, struct lu_device *top)
3488 struct lu_device *d = top, *n;
3489 struct obd_device *obd = m->mdt_md_dev.md_lu_dev.ld_obd;
3490 struct lustre_cfg_bufs *bufs;
3491 struct lustre_cfg *lcfg;
3492 struct mdt_thread_info *info;
3496 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3497 LASSERT(info != NULL);
3499 bufs = &info->mti_u.bufs;
3500 /* process cleanup, pass mdt obd name to get obd umount flags */
3501 lustre_cfg_bufs_reset(bufs, obd->obd_name);
3506 lustre_cfg_bufs_set_string(bufs, 1, flags);
3507 lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
3509 CERROR("Cannot alloc lcfg!\n");
3514 top->ld_ops->ldo_process_config(env, top, lcfg);
3515 lustre_cfg_free(lcfg);
3517 lu_site_purge(env, top->ld_site, ~0);
3519 struct obd_type *type;
3520 struct lu_device_type *ldt = d->ld_type;
3522 /* each fini() returns next device in stack of layers
3523 * * so we can avoid the recursion */
3524 n = ldt->ldt_ops->ldto_device_fini(env, d);
3526 ldt->ldt_ops->ldto_device_free(env, d);
3527 type = ldt->ldt_obd_type;
3529 class_put_type(type);
3531 /* switch to the next device in the layer */
3534 m->mdt_child = NULL;
3535 m->mdt_bottom = NULL;
3538 static struct lu_device *mdt_layer_setup(const struct lu_env *env,
3539 const char *typename,
3540 struct lu_device *child,
3541 struct lustre_cfg *cfg)
3543 const char *dev = lustre_cfg_string(cfg, 0);
3544 struct obd_type *type;
3545 struct lu_device_type *ldt;
3546 struct lu_device *d;
3551 type = class_get_type(typename);
3553 CERROR("Unknown type: '%s'\n", typename);
3554 GOTO(out, rc = -ENODEV);
3557 rc = lu_context_refill(&env->le_ctx);
3559 CERROR("Failure to refill context: '%d'\n", rc);
3563 if (env->le_ses != NULL) {
3564 rc = lu_context_refill(env->le_ses);
3566 CERROR("Failure to refill session: '%d'\n", rc);
3573 CERROR("type: '%s'\n", typename);
3574 GOTO(out_type, rc = -EINVAL);
3577 ldt->ldt_obd_type = type;
3578 d = ldt->ldt_ops->ldto_device_alloc(env, ldt, cfg);
3580 CERROR("Cannot allocate device: '%s'\n", typename);
3581 GOTO(out_type, rc = -ENODEV);
3584 LASSERT(child->ld_site);
3585 d->ld_site = child->ld_site;
3588 rc = ldt->ldt_ops->ldto_device_init(env, d, dev, child);
3590 CERROR("can't init device '%s', rc %d\n", typename, rc);
3591 GOTO(out_alloc, rc);
3598 ldt->ldt_ops->ldto_device_free(env, d);
3601 class_put_type(type);
3606 static int mdt_stack_init(const struct lu_env *env,
3607 struct mdt_device *m, struct lustre_cfg *cfg)
3609 struct lu_device *d = &m->mdt_md_dev.md_lu_dev;
3610 struct lu_device *tmp;
3611 struct md_device *md;
3615 /* init the stack */
3616 tmp = mdt_layer_setup(env, LUSTRE_OSD_NAME, d, cfg);
3618 RETURN(PTR_ERR(tmp));
3620 m->mdt_bottom = lu2dt_dev(tmp);
3622 tmp = mdt_layer_setup(env, LUSTRE_MDD_NAME, d, cfg);
3624 GOTO(out, rc = PTR_ERR(tmp));
3629 tmp = mdt_layer_setup(env, LUSTRE_CMM_NAME, d, cfg);
3631 GOTO(out, rc = PTR_ERR(tmp));
3634 /*set mdd upcall device*/
3635 md_upcall_dev_set(md, lu2md_dev(d));
3638 /*set cmm upcall device*/
3639 md_upcall_dev_set(md, &m->mdt_md_dev);
3641 m->mdt_child = lu2md_dev(d);
3643 /* process setup config */
3644 tmp = &m->mdt_md_dev.md_lu_dev;
3645 rc = tmp->ld_ops->ldo_process_config(env, tmp, cfg);
3648 /* fini from last known good lu_device */
3650 mdt_stack_fini(env, m, d);
3655 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
3657 struct md_device *next = m->mdt_child;
3658 struct lu_device *d = &m->mdt_md_dev.md_lu_dev;
3659 struct lu_site *ls = d->ld_site;
3660 struct obd_device *obd = m->mdt_md_dev.md_lu_dev.ld_obd;
3663 ping_evictor_stop();
3665 target_recovery_fini(obd);
3666 mdt_stop_ptlrpc_service(m);
3668 mdt_fs_cleanup(env, m);
3670 upcall_cache_cleanup(m->mdt_rmtacl_cache);
3671 m->mdt_rmtacl_cache = NULL;
3673 upcall_cache_cleanup(m->mdt_identity_cache);
3674 m->mdt_identity_cache = NULL;
3676 if (m->mdt_namespace != NULL) {
3677 ldlm_namespace_free(m->mdt_namespace, d->ld_obd->obd_force);
3678 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
3681 mdt_seq_fini(env, m);
3682 mdt_seq_fini_cli(m);
3683 mdt_fld_fini(env, m);
3685 ptlrpc_lprocfs_unregister_obd(d->ld_obd);
3686 lprocfs_obd_cleanup(d->ld_obd);
3688 if (m->mdt_rootsquash_info) {
3689 OBD_FREE_PTR(m->mdt_rootsquash_info);
3690 m->mdt_rootsquash_info = NULL;
3693 next->md_ops->mdo_init_capa_ctxt(env, next, 0, 0, 0, NULL);
3694 del_timer(&m->mdt_ck_timer);
3695 mdt_ck_thread_stop(m);
3697 /* finish the stack */
3698 mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
3701 if (!list_empty(&ls->ls_lru) || ls->ls_total != 0) {
3703 * Uh-oh, objects still exist.
3705 static DECLARE_LU_CDEBUG_PRINT_INFO(cookie, D_ERROR);
3707 lu_site_print(env, ls, &cookie, lu_cdebug_printer);
3714 LASSERT(atomic_read(&d->ld_ref) == 0);
3715 md_device_fini(&m->mdt_md_dev);
3720 static void fsoptions_to_mdt_flags(struct mdt_device *m, char *options)
3730 while (*p && *p != ',')
3734 if ((len == sizeof("user_xattr") - 1) &&
3735 (memcmp(options, "user_xattr", len) == 0)) {
3736 m->mdt_opts.mo_user_xattr = 1;
3737 LCONSOLE_INFO("Enabling user_xattr\n");
3738 } else if ((len == sizeof("nouser_xattr") - 1) &&
3739 (memcmp(options, "nouser_xattr", len) == 0)) {
3740 m->mdt_opts.mo_user_xattr = 0;
3741 LCONSOLE_INFO("Disabling user_xattr\n");
3742 } else if ((len == sizeof("acl") - 1) &&
3743 (memcmp(options, "acl", len) == 0)) {
3744 #ifdef CONFIG_FS_POSIX_ACL
3745 m->mdt_opts.mo_acl = 1;
3746 LCONSOLE_INFO("Enabling ACL\n");
3748 m->mdt_opts.mo_acl = 0;
3749 CWARN("ignoring unsupported acl mount option\n");
3750 LCONSOLE_INFO("Disabling ACL\n");
3752 } else if ((len == sizeof("noacl") - 1) &&
3753 (memcmp(options, "noacl", len) == 0)) {
3754 m->mdt_opts.mo_acl = 0;
3755 LCONSOLE_INFO("Disabling ACL\n");
3762 int mdt_postrecov(const struct lu_env *, struct mdt_device *);
3764 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
3765 struct lu_device_type *ldt, struct lustre_cfg *cfg)
3767 struct lprocfs_static_vars lvars;
3768 struct mdt_thread_info *info;
3769 struct obd_device *obd;
3770 const char *dev = lustre_cfg_string(cfg, 0);
3771 const char *num = lustre_cfg_string(cfg, 2);
3772 struct lustre_mount_info *lmi;
3773 struct lustre_sb_info *lsi;
3778 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3779 LASSERT(info != NULL);
3781 obd = class_name2obd(dev);
3782 LASSERT(obd != NULL);
3784 spin_lock_init(&m->mdt_transno_lock);
3786 m->mdt_max_mdsize = MAX_MD_SIZE;
3787 m->mdt_max_cookiesize = sizeof(struct llog_cookie);
3789 m->mdt_opts.mo_user_xattr = 0;
3790 m->mdt_opts.mo_acl = 0;
3791 lmi = server_get_mount_2(dev);
3793 CERROR("Cannot get mount info for %s!\n", dev);
3796 lsi = s2lsi(lmi->lmi_sb);
3797 fsoptions_to_mdt_flags(m, lsi->lsi_lmd->lmd_opts);
3798 server_put_mount_2(dev, lmi->lmi_mnt);
3801 spin_lock_init(&m->mdt_ioepoch_lock);
3802 m->mdt_opts.mo_compat_resname = 0;
3803 m->mdt_capa_timeout = CAPA_TIMEOUT;
3804 m->mdt_capa_alg = CAPA_HMAC_ALG_SHA1;
3805 m->mdt_ck_timeout = CAPA_KEY_TIMEOUT;
3807 spin_lock_init(&m->mdt_client_bitmap_lock);
3813 md_device_init(&m->mdt_md_dev, ldt);
3814 m->mdt_md_dev.md_lu_dev.ld_ops = &mdt_lu_ops;
3815 m->mdt_md_dev.md_lu_dev.ld_obd = obd;
3816 /* set this lu_device to obd, because error handling need it */
3817 obd->obd_lu_dev = &m->mdt_md_dev.md_lu_dev;
3819 rc = lu_site_init(s, &m->mdt_md_dev.md_lu_dev);
3821 CERROR("Can't init lu_site, rc %d\n", rc);
3822 GOTO(err_free_site, rc);
3825 lprocfs_init_vars(mdt, &lvars);
3826 rc = lprocfs_obd_setup(obd, lvars.obd_vars);
3828 CERROR("Can't init lprocfs, rc %d\n", rc);
3829 GOTO(err_fini_site, rc);
3831 ptlrpc_lprocfs_register_obd(obd);
3833 rc = mdt_procfs_init(m, dev);
3835 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
3836 GOTO(err_fini_proc, rc);
3839 /* set server index */
3841 s->ls_node_id = simple_strtol(num, NULL, 10);
3843 /* failover is the default
3844 * FIXME: we do not failout mds0/mgs, which may cause some problems.
3845 * assumed whose ls_node_id == 0 XXX
3847 obd->obd_replayable = 1;
3848 /* No connection accepted until configurations will finish */
3849 obd->obd_no_conn = 1;
3851 if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
3852 char *str = lustre_cfg_string(cfg, 4);
3853 if (strchr(str, 'n')) {
3854 CWARN("%s: recovery disabled\n", obd->obd_name);
3855 obd->obd_replayable = 0;
3859 /* init the stack */
3860 rc = mdt_stack_init(env, m, cfg);
3862 CERROR("Can't init device stack, rc %d\n", rc);
3863 GOTO(err_fini_proc, rc);
3866 rc = mdt_fld_init(env, obd->obd_name, m);
3868 GOTO(err_fini_stack, rc);
3870 rc = mdt_seq_init(env, obd->obd_name, m);
3872 GOTO(err_fini_fld, rc);
3874 snprintf(info->mti_u.ns_name, sizeof info->mti_u.ns_name,
3875 LUSTRE_MDT_NAME"-%p", m);
3876 m->mdt_namespace = ldlm_namespace_new(info->mti_u.ns_name,
3877 LDLM_NAMESPACE_SERVER,
3878 LDLM_NAMESPACE_GREEDY);
3879 if (m->mdt_namespace == NULL)
3880 GOTO(err_fini_seq, rc = -ENOMEM);
3882 ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
3883 /* set obd_namespace for compatibility with old code */
3884 obd->obd_namespace = m->mdt_namespace;
3886 m->mdt_identity_cache = upcall_cache_init(obd->obd_name,
3888 &mdt_identity_upcall_cache_ops);
3889 if (IS_ERR(m->mdt_identity_cache)) {
3890 rc = PTR_ERR(m->mdt_identity_cache);
3891 m->mdt_identity_cache = NULL;
3892 GOTO(err_free_ns, rc);
3895 m->mdt_rmtacl_cache = upcall_cache_init(obd->obd_name,
3896 MDT_RMTACL_UPCALL_PATH,
3897 &mdt_rmtacl_upcall_cache_ops);
3898 if (IS_ERR(m->mdt_rmtacl_cache)) {
3899 rc = PTR_ERR(m->mdt_rmtacl_cache);
3900 m->mdt_rmtacl_cache = NULL;
3901 GOTO(err_free_ns, rc);
3904 m->mdt_ck_timer.function = mdt_ck_timer_callback;
3905 m->mdt_ck_timer.data = (unsigned long)m;
3906 init_timer(&m->mdt_ck_timer);
3907 rc = mdt_ck_thread_start(m);
3909 GOTO(err_free_ns, rc);
3911 rc = mdt_fs_setup(env, m, obd);
3915 target_recovery_init(obd, mdt_recovery_handle);
3917 rc = mdt_start_ptlrpc_service(m);
3919 GOTO(err_fs_cleanup, rc);
3921 ping_evictor_start();
3923 rc = lu_site_init_finish(s);
3925 GOTO(err_stop_service, rc);
3927 if (obd->obd_recovering == 0)
3928 mdt_postrecov(env, m);
3930 mdt_init_capa_ctxt(env, m);
3932 if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
3938 ping_evictor_stop();
3939 mdt_stop_ptlrpc_service(m);
3941 target_recovery_fini(obd);
3942 mdt_fs_cleanup(env, m);
3944 del_timer(&m->mdt_ck_timer);
3945 mdt_ck_thread_stop(m);
3947 upcall_cache_cleanup(m->mdt_rmtacl_cache);
3948 m->mdt_rmtacl_cache = NULL;
3949 upcall_cache_cleanup(m->mdt_identity_cache);
3950 m->mdt_identity_cache = NULL;
3951 ldlm_namespace_free(m->mdt_namespace, 0);
3952 obd->obd_namespace = m->mdt_namespace = NULL;
3954 mdt_seq_fini(env, m);
3956 mdt_fld_fini(env, m);
3958 mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
3961 lprocfs_obd_cleanup(obd);
3967 md_device_fini(&m->mdt_md_dev);
3971 /* used by MGS to process specific configurations */
3972 static int mdt_process_config(const struct lu_env *env,
3973 struct lu_device *d, struct lustre_cfg *cfg)
3975 struct mdt_device *m = mdt_dev(d);
3976 struct md_device *md_next = m->mdt_child;
3977 struct lu_device *next = md2lu_dev(md_next);
3981 switch (cfg->lcfg_command) {
3983 struct lprocfs_static_vars lvars;
3984 struct obd_device *obd = d->ld_obd;
3986 lprocfs_init_vars(mdt, &lvars);
3987 rc = class_process_proc_param(PARAM_MDT, lvars.obd_vars, cfg, obd);
3989 /* others are passed further */
3990 rc = next->ld_ops->ldo_process_config(env, next, cfg);
3995 * Add mdc hook to get first MDT uuid and connect it to
3996 * ls->controller to use for seq manager.
3998 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4000 CERROR("Can't add mdc, rc %d\n", rc);
4002 rc = mdt_seq_init_cli(env, mdt_dev(d), cfg);
4005 /* others are passed further */
4006 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4012 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
4013 const struct lu_object_header *hdr,
4014 struct lu_device *d)
4016 struct mdt_object *mo;
4022 struct lu_object *o;
4023 struct lu_object_header *h;
4025 o = &mo->mot_obj.mo_lu;
4026 h = &mo->mot_header;
4027 lu_object_header_init(h);
4028 lu_object_init(o, h, d);
4029 lu_object_add_top(h, o);
4030 o->lo_ops = &mdt_obj_ops;
4036 static int mdt_object_init(const struct lu_env *env, struct lu_object *o)
4038 struct mdt_device *d = mdt_dev(o->lo_dev);
4039 struct lu_device *under;
4040 struct lu_object *below;
4044 CDEBUG(D_INFO, "object init, fid = "DFID"\n",
4045 PFID(lu_object_fid(o)));
4047 under = &d->mdt_child->md_lu_dev;
4048 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
4049 if (below != NULL) {
4050 lu_object_add(o, below);
4057 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
4059 struct mdt_object *mo = mdt_obj(o);
4060 struct lu_object_header *h;
4064 CDEBUG(D_INFO, "object free, fid = "DFID"\n",
4065 PFID(lu_object_fid(o)));
4068 lu_object_header_fini(h);
4073 static int mdt_object_print(const struct lu_env *env, void *cookie,
4074 lu_printer_t p, const struct lu_object *o)
4076 return (*p)(env, cookie, LUSTRE_MDT_NAME"-object@%p", o);
4079 static struct lu_device_operations mdt_lu_ops = {
4080 .ldo_object_alloc = mdt_object_alloc,
4081 .ldo_process_config = mdt_process_config
4084 static struct lu_object_operations mdt_obj_ops = {
4085 .loo_object_init = mdt_object_init,
4086 .loo_object_free = mdt_object_free,
4087 .loo_object_print = mdt_object_print
4090 /* mds_connect_internal */
4091 static int mdt_connect_internal(struct obd_export *exp,
4092 struct mdt_device *mdt,
4093 struct obd_connect_data *data)
4098 data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
4099 data->ocd_ibits_known &= MDS_INODELOCK_FULL;
4101 /* If no known bits (which should not happen, probably,
4102 as everybody should support LOOKUP and UPDATE bits at least)
4103 revert to compat mode with plain locks. */
4104 if (!data->ocd_ibits_known &&
4105 data->ocd_connect_flags & OBD_CONNECT_IBITS)
4106 data->ocd_connect_flags &= ~OBD_CONNECT_IBITS;
4108 if (!mdt->mdt_opts.mo_acl)
4109 data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
4111 if (!mdt->mdt_opts.mo_user_xattr)
4112 data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
4114 if (!mdt->mdt_opts.mo_mds_capa)
4115 data->ocd_connect_flags &= ~OBD_CONNECT_MDS_CAPA;
4117 if (!mdt->mdt_opts.mo_oss_capa)
4118 data->ocd_connect_flags &= ~OBD_CONNECT_OSS_CAPA;
4120 spin_lock(&exp->exp_lock);
4121 exp->exp_connect_flags = data->ocd_connect_flags;
4122 spin_unlock(&exp->exp_lock);
4123 data->ocd_version = LUSTRE_VERSION_CODE;
4124 exp->exp_mdt_data.med_ibits_known = data->ocd_ibits_known;
4128 if (mdt->mdt_opts.mo_acl &&
4129 ((exp->exp_connect_flags & OBD_CONNECT_ACL) == 0)) {
4130 CWARN("%s: MDS requires ACL support but client does not\n",
4131 mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4136 flags = OBD_CONNECT_LCL_CLIENT | OBD_CONNECT_RMT_CLIENT;
4137 if ((exp->exp_connect_flags & flags) == flags) {
4138 CWARN("%s: both local and remote client flags are set\n",
4139 mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4143 if (mdt->mdt_opts.mo_mds_capa &&
4144 ((exp->exp_connect_flags & OBD_CONNECT_MDS_CAPA) == 0)) {
4145 CWARN("%s: MDS requires capability support, but client not\n",
4146 mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4150 if (mdt->mdt_opts.mo_oss_capa &&
4151 ((exp->exp_connect_flags & OBD_CONNECT_OSS_CAPA) == 0)) {
4152 CWARN("%s: MDS requires OSS capability support, "
4154 mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4161 /* mds_connect copy */
4162 static int mdt_obd_connect(const struct lu_env *env,
4163 struct lustre_handle *conn, struct obd_device *obd,
4164 struct obd_uuid *cluuid,
4165 struct obd_connect_data *data)
4167 struct mdt_client_data *mcd;
4168 struct obd_export *exp;
4169 struct mdt_device *mdt;
4173 LASSERT(env != NULL);
4174 if (!conn || !obd || !cluuid)
4177 mdt = mdt_dev(obd->obd_lu_dev);
4179 rc = class_connect(conn, obd, cluuid);
4183 exp = class_conn2export(conn);
4184 LASSERT(exp != NULL);
4186 rc = mdt_connect_internal(exp, mdt, data);
4190 struct mdt_thread_info *mti;
4191 mti = lu_context_key_get(&env->le_ctx,
4193 LASSERT(mti != NULL);
4195 memcpy(mcd->mcd_uuid, cluuid, sizeof mcd->mcd_uuid);
4196 exp->exp_mdt_data.med_mcd = mcd;
4197 rc = mdt_client_new(env, mdt);
4200 exp->exp_mdt_data.med_mcd = NULL;
4207 class_disconnect(exp);
4209 class_export_put(exp);
4214 static int mdt_obd_reconnect(struct obd_export *exp, struct obd_device *obd,
4215 struct obd_uuid *cluuid,
4216 struct obd_connect_data *data)
4221 if (exp == NULL || obd == NULL || cluuid == NULL)
4224 rc = mdt_connect_internal(exp, mdt_dev(obd->obd_lu_dev), data);
4229 static int mdt_obd_disconnect(struct obd_export *exp)
4231 struct mdt_device *mdt = mdt_dev(exp->exp_obd->obd_lu_dev);
4236 class_export_get(exp);
4238 /* Disconnect early so that clients can't keep using export */
4239 rc = class_disconnect(exp);
4240 if (mdt->mdt_namespace != NULL || exp->exp_obd->obd_namespace != NULL)
4241 ldlm_cancel_locks_for_export(exp);
4243 /* complete all outstanding replies */
4244 spin_lock(&exp->exp_lock);
4245 while (!list_empty(&exp->exp_outstanding_replies)) {
4246 struct ptlrpc_reply_state *rs =
4247 list_entry(exp->exp_outstanding_replies.next,
4248 struct ptlrpc_reply_state, rs_exp_list);
4249 struct ptlrpc_service *svc = rs->rs_service;
4251 spin_lock(&svc->srv_lock);
4252 list_del_init(&rs->rs_exp_list);
4253 ptlrpc_schedule_difficult_reply(rs);
4254 spin_unlock(&svc->srv_lock);
4256 spin_unlock(&exp->exp_lock);
4258 class_export_put(exp);
4262 /* FIXME: Can we avoid using these two interfaces? */
4263 static int mdt_init_export(struct obd_export *exp)
4265 struct mdt_export_data *med = &exp->exp_mdt_data;
4268 INIT_LIST_HEAD(&med->med_open_head);
4269 spin_lock_init(&med->med_open_lock);
4270 spin_lock(&exp->exp_lock);
4271 exp->exp_connecting = 1;
4272 spin_unlock(&exp->exp_lock);
4276 static int mdt_destroy_export(struct obd_export *export)
4278 struct mdt_export_data *med;
4279 struct obd_device *obd = export->exp_obd;
4280 struct mdt_device *mdt;
4281 struct mdt_thread_info *info;
4289 med = &export->exp_mdt_data;
4290 if (med->med_rmtclient)
4291 mdt_cleanup_idmap(med);
4293 target_destroy_export(export);
4295 if (obd_uuid_equals(&export->exp_client_uuid, &obd->obd_uuid))
4298 mdt = mdt_dev(obd->obd_lu_dev);
4299 LASSERT(mdt != NULL);
4301 rc = lu_env_init(&env, NULL, LCT_MD_THREAD);
4305 info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
4306 LASSERT(info != NULL);
4307 memset(info, 0, sizeof *info);
4308 info->mti_env = &env;
4309 info->mti_mdt = mdt;
4310 info->mti_exp = export;
4312 ma = &info->mti_attr;
4313 lmm_size = ma->ma_lmm_size = mdt->mdt_max_mdsize;
4314 cookie_size = ma->ma_cookie_size = mdt->mdt_max_cookiesize;
4315 OBD_ALLOC(ma->ma_lmm, lmm_size);
4316 OBD_ALLOC(ma->ma_cookie, cookie_size);
4318 if (ma->ma_lmm == NULL || ma->ma_cookie == NULL)
4319 GOTO(out, rc = -ENOMEM);
4320 ma->ma_need = MA_LOV | MA_COOKIE;
4322 /* Close any open files (which may also cause orphan unlinking). */
4323 spin_lock(&med->med_open_lock);
4324 while (!list_empty(&med->med_open_head)) {
4325 struct list_head *tmp = med->med_open_head.next;
4326 struct mdt_file_data *mfd =
4327 list_entry(tmp, struct mdt_file_data, mfd_list);
4329 /* Remove mfd handle so it can't be found again.
4330 * We are consuming the mfd_list reference here. */
4331 class_handle_unhash(&mfd->mfd_handle);
4332 list_del_init(&mfd->mfd_list);
4333 spin_unlock(&med->med_open_lock);
4334 mdt_mfd_close(info, mfd);
4335 /* TODO: if we close the unlinked file,
4336 * we need to remove it's objects from OST */
4337 memset(&ma->ma_attr, 0, sizeof(ma->ma_attr));
4338 spin_lock(&med->med_open_lock);
4339 ma->ma_lmm_size = lmm_size;
4340 ma->ma_cookie_size = cookie_size;
4341 ma->ma_need = MA_LOV | MA_COOKIE;
4344 spin_unlock(&med->med_open_lock);
4345 info->mti_mdt = NULL;
4346 mdt_client_del(&env, mdt);
4351 OBD_FREE(ma->ma_lmm, lmm_size);
4355 OBD_FREE(ma->ma_cookie, cookie_size);
4356 ma->ma_cookie = NULL;
4363 static void mdt_allow_cli(struct mdt_device *m, unsigned int flag)
4365 if (flag & CONFIG_LOG)
4366 m->mdt_fl_cfglog = 1;
4367 if (flag & CONFIG_SYNC)
4368 m->mdt_fl_synced = 1;
4370 if (m->mdt_fl_cfglog /* bz11778: && m->mdt_fl_synced */)
4371 /* Open for clients */
4372 m->mdt_md_dev.md_lu_dev.ld_obd->obd_no_conn = 0;
4375 static int mdt_upcall(const struct lu_env *env, struct md_device *md,
4376 enum md_upcall_event ev)
4378 struct mdt_device *m = mdt_dev(&md->md_lu_dev);
4379 struct md_device *next = m->mdt_child;
4380 struct mdt_thread_info *mti;
4386 rc = next->md_ops->mdo_maxsize_get(env, next,
4388 &m->mdt_max_cookiesize);
4389 CDEBUG(D_INFO, "get max mdsize %d max cookiesize %d\n",
4390 m->mdt_max_mdsize, m->mdt_max_cookiesize);
4391 mdt_allow_cli(m, CONFIG_SYNC);
4394 mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4395 mti->mti_no_need_trans = 1;
4396 CDEBUG(D_INFO, "disable mdt trans for this thread\n");
4399 /* Check that MDT is not yet configured */
4400 LASSERT(!m->mdt_fl_cfglog);
4403 CERROR("invalid event\n");
4410 static int mdt_obd_notify(struct obd_device *host,
4411 struct obd_device *watched,
4412 enum obd_notify_event ev, void *data)
4417 case OBD_NOTIFY_CONFIG:
4418 mdt_allow_cli(mdt_dev(host->obd_lu_dev), (unsigned int)data);
4421 CDEBUG(D_INFO, "Unhandled notification %#x\n", ev);
4426 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
4427 void *karg, void *uarg)
4430 struct obd_device *obd= exp->exp_obd;
4431 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
4432 struct dt_device *dt = mdt->mdt_bottom;
4436 CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
4437 rc = lu_env_init(&env, NULL, LCT_MD_THREAD);
4443 rc = dt->dd_ops->dt_sync(&env, dt);
4445 case OBD_IOC_SET_READONLY:
4446 rc = dt->dd_ops->dt_sync(&env, dt);
4447 dt->dd_ops->dt_ro(&env, dt);
4449 case OBD_IOC_ABORT_RECOVERY:
4450 CERROR("Aborting recovery for device %s\n", obd->obd_name);
4451 target_stop_recovery_thread(obd);
4455 CERROR("Not supported cmd = %d for device %s\n",
4456 cmd, obd->obd_name);
4464 int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
4466 struct lu_device *ld = md2lu_dev(mdt->mdt_child);
4467 struct obd_device *obd = mdt->mdt_md_dev.md_lu_dev.ld_obd;
4470 /* if some clients didn't participate in recovery then we can possibly
4471 * lost sequence. Now we should increase sequence for safe value */
4472 lost = obd->obd_max_recoverable_clients - obd->obd_connected_clients;
4473 mdt_seq_adjust(env, mdt, lost);
4475 rc = ld->ld_ops->ldo_recovery_complete(env, ld);
4479 int mdt_obd_postrecov(struct obd_device *obd)
4484 rc = lu_env_init(&env, NULL, LCT_MD_THREAD);
4487 rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
4492 static struct obd_ops mdt_obd_device_ops = {
4493 .o_owner = THIS_MODULE,
4494 .o_connect = mdt_obd_connect,
4495 .o_reconnect = mdt_obd_reconnect,
4496 .o_disconnect = mdt_obd_disconnect,
4497 .o_init_export = mdt_init_export,
4498 .o_destroy_export = mdt_destroy_export,
4499 .o_iocontrol = mdt_iocontrol,
4500 .o_postrecov = mdt_obd_postrecov,
4501 .o_notify = mdt_obd_notify
4504 static struct lu_device* mdt_device_fini(const struct lu_env *env,
4505 struct lu_device *d)
4507 struct mdt_device *m = mdt_dev(d);
4514 static void mdt_device_free(const struct lu_env *env, struct lu_device *d)
4516 struct mdt_device *m = mdt_dev(d);
4521 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
4522 struct lu_device_type *t,
4523 struct lustre_cfg *cfg)
4525 struct lu_device *l;
4526 struct mdt_device *m;
4532 l = &m->mdt_md_dev.md_lu_dev;
4533 rc = mdt_init0(env, m, t, cfg);
4539 md_upcall_init(&m->mdt_md_dev, mdt_upcall);
4541 l = ERR_PTR(-ENOMEM);
4545 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
4546 LU_KEY_INIT_FINI(mdt, struct mdt_thread_info);
4548 /* context key: mdt_thread_key */
4549 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
4551 /* context key constructor/destructor: mdt_txn_key_init, mdt_txn_key_fini */
4552 LU_KEY_INIT_FINI(mdt_txn, struct mdt_txn_info);
4554 struct lu_context_key mdt_txn_key = {
4555 .lct_tags = LCT_TX_HANDLE,
4556 .lct_init = mdt_txn_key_init,
4557 .lct_fini = mdt_txn_key_fini
4560 struct md_ucred *mdt_ucred(const struct mdt_thread_info *info)
4562 return md_ucred(info->mti_env);
4565 /* type constructor/destructor: mdt_type_init, mdt_type_fini */
4566 LU_TYPE_INIT_FINI(mdt, &mdt_thread_key, &mdt_txn_key);
4568 static struct lu_device_type_operations mdt_device_type_ops = {
4569 .ldto_init = mdt_type_init,
4570 .ldto_fini = mdt_type_fini,
4572 .ldto_device_alloc = mdt_device_alloc,
4573 .ldto_device_free = mdt_device_free,
4574 .ldto_device_fini = mdt_device_fini
4577 static struct lu_device_type mdt_device_type = {
4578 .ldt_tags = LU_DEVICE_MD,
4579 .ldt_name = LUSTRE_MDT_NAME,
4580 .ldt_ops = &mdt_device_type_ops,
4581 .ldt_ctx_tags = LCT_MD_THREAD
4584 static int __init mdt_mod_init(void)
4586 struct lprocfs_static_vars lvars;
4589 mdt_num_threads = MDT_NUM_THREADS;
4590 lprocfs_init_vars(mdt, &lvars);
4591 rc = class_register_type(&mdt_obd_device_ops, NULL,
4592 lvars.module_vars, LUSTRE_MDT_NAME,
4598 static void __exit mdt_mod_exit(void)
4600 class_unregister_type(LUSTRE_MDT_NAME);
4604 #define DEF_HNDL(prefix, base, suffix, flags, opc, fn, fmt) \
4605 [prefix ## _ ## opc - prefix ## _ ## base] = { \
4607 .mh_fail_id = OBD_FAIL_ ## prefix ## _ ## opc ## suffix, \
4608 .mh_opc = prefix ## _ ## opc, \
4609 .mh_flags = flags, \
4614 #define DEF_MDT_HNDL(flags, name, fn, fmt) \
4615 DEF_HNDL(MDS, GETATTR, _NET, flags, name, fn, fmt)
4617 #define DEF_SEQ_HNDL(flags, name, fn, fmt) \
4618 DEF_HNDL(SEQ, QUERY, _NET, flags, name, fn, fmt)
4620 #define DEF_FLD_HNDL(flags, name, fn, fmt) \
4621 DEF_HNDL(FLD, QUERY, _NET, flags, name, fn, fmt)
4623 * Request with a format known in advance
4625 #define DEF_MDT_HNDL_F(flags, name, fn) \
4626 DEF_HNDL(MDS, GETATTR, _NET, flags, name, fn, &RQF_MDS_ ## name)
4628 #define DEF_SEQ_HNDL_F(flags, name, fn) \
4629 DEF_HNDL(SEQ, QUERY, _NET, flags, name, fn, &RQF_SEQ_ ## name)
4631 #define DEF_FLD_HNDL_F(flags, name, fn) \
4632 DEF_HNDL(FLD, QUERY, _NET, flags, name, fn, &RQF_FLD_ ## name)
4634 * Request with a format we do not yet know
4636 #define DEF_MDT_HNDL_0(flags, name, fn) \
4637 DEF_HNDL(MDS, GETATTR, _NET, flags, name, fn, NULL)
4639 static struct mdt_handler mdt_mds_ops[] = {
4640 DEF_MDT_HNDL_F(0, CONNECT, mdt_connect),
4641 DEF_MDT_HNDL_F(0, DISCONNECT, mdt_disconnect),
4642 DEF_MDT_HNDL_F(0, SET_INFO, mdt_set_info),
4643 DEF_MDT_HNDL_F(0 |HABEO_REFERO, GETSTATUS, mdt_getstatus),
4644 DEF_MDT_HNDL_F(HABEO_CORPUS, GETATTR, mdt_getattr),
4645 DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, GETATTR_NAME, mdt_getattr_name),
4646 DEF_MDT_HNDL_F(HABEO_CORPUS|MUTABOR, SETXATTR, mdt_setxattr),
4647 DEF_MDT_HNDL_F(HABEO_CORPUS, GETXATTR, mdt_getxattr),
4648 DEF_MDT_HNDL_F(0 |HABEO_REFERO, STATFS, mdt_statfs),
4649 DEF_MDT_HNDL_F(0 |MUTABOR, REINT, mdt_reint),
4650 DEF_MDT_HNDL_F(HABEO_CORPUS, CLOSE, mdt_close),
4651 DEF_MDT_HNDL_F(HABEO_CORPUS, DONE_WRITING, mdt_done_writing),
4652 DEF_MDT_HNDL_F(0 |HABEO_REFERO, PIN, mdt_pin),
4653 DEF_MDT_HNDL_0(0, SYNC, mdt_sync),
4654 DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, IS_SUBDIR, mdt_is_subdir),
4655 DEF_MDT_HNDL_0(0, QUOTACHECK, mdt_quotacheck_handle),
4656 DEF_MDT_HNDL_0(0, QUOTACTL, mdt_quotactl_handle)
4659 #define DEF_OBD_HNDL(flags, name, fn) \
4660 DEF_HNDL(OBD, PING, _NET, flags, name, fn, NULL)
4663 static struct mdt_handler mdt_obd_ops[] = {
4664 DEF_OBD_HNDL(0, PING, mdt_obd_ping),
4665 DEF_OBD_HNDL(0, LOG_CANCEL, mdt_obd_log_cancel),
4666 DEF_OBD_HNDL(0, QC_CALLBACK, mdt_obd_qc_callback)
4669 #define DEF_DLM_HNDL_0(flags, name, fn) \
4670 DEF_HNDL(LDLM, ENQUEUE, , flags, name, fn, NULL)
4671 #define DEF_DLM_HNDL_F(flags, name, fn) \
4672 DEF_HNDL(LDLM, ENQUEUE, , flags, name, fn, &RQF_LDLM_ ## name)
4674 static struct mdt_handler mdt_dlm_ops[] = {
4675 DEF_DLM_HNDL_F(HABEO_CLAVIS, ENQUEUE, mdt_enqueue),
4676 DEF_DLM_HNDL_0(HABEO_CLAVIS, CONVERT, mdt_convert),
4677 DEF_DLM_HNDL_0(0, BL_CALLBACK, mdt_bl_callback),
4678 DEF_DLM_HNDL_0(0, CP_CALLBACK, mdt_cp_callback)
4681 static struct mdt_handler mdt_llog_ops[] = {
4684 #define DEF_SEC_CTX_HNDL(name, fn) \
4685 DEF_HNDL(SEC_CTX, INIT, _NET, 0, name, fn, NULL)
4687 static struct mdt_handler mdt_sec_ctx_ops[] = {
4688 DEF_SEC_CTX_HNDL(INIT, mdt_sec_ctx_handle),
4689 DEF_SEC_CTX_HNDL(INIT_CONT, mdt_sec_ctx_handle),
4690 DEF_SEC_CTX_HNDL(FINI, mdt_sec_ctx_handle)
4693 static struct mdt_opc_slice mdt_regular_handlers[] = {
4695 .mos_opc_start = MDS_GETATTR,
4696 .mos_opc_end = MDS_LAST_OPC,
4697 .mos_hs = mdt_mds_ops
4700 .mos_opc_start = OBD_PING,
4701 .mos_opc_end = OBD_LAST_OPC,
4702 .mos_hs = mdt_obd_ops
4705 .mos_opc_start = LDLM_ENQUEUE,
4706 .mos_opc_end = LDLM_LAST_OPC,
4707 .mos_hs = mdt_dlm_ops
4710 .mos_opc_start = LLOG_ORIGIN_HANDLE_CREATE,
4711 .mos_opc_end = LLOG_LAST_OPC,
4712 .mos_hs = mdt_llog_ops
4715 .mos_opc_start = SEC_CTX_INIT,
4716 .mos_opc_end = SEC_LAST_OPC,
4717 .mos_hs = mdt_sec_ctx_ops
4724 static struct mdt_handler mdt_readpage_ops[] = {
4725 DEF_MDT_HNDL_F(0, CONNECT, mdt_connect),
4726 DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, READPAGE, mdt_readpage),
4727 #ifdef HAVE_SPLIT_SUPPORT
4728 DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, WRITEPAGE, mdt_writepage),
4732 * XXX: this is ugly and should be fixed one day, see mdc_close() for
4733 * detailed comments. --umka
4735 DEF_MDT_HNDL_F(HABEO_CORPUS, CLOSE, mdt_close),
4736 DEF_MDT_HNDL_F(HABEO_CORPUS, DONE_WRITING, mdt_done_writing),
4739 static struct mdt_opc_slice mdt_readpage_handlers[] = {
4741 .mos_opc_start = MDS_GETATTR,
4742 .mos_opc_end = MDS_LAST_OPC,
4743 .mos_hs = mdt_readpage_ops
4750 static struct mdt_handler mdt_xmds_ops[] = {
4751 DEF_MDT_HNDL_F(0, CONNECT, mdt_connect),
4752 DEF_MDT_HNDL_F(HABEO_CORPUS , GETATTR, mdt_getattr),
4753 DEF_MDT_HNDL_F(0 | MUTABOR , REINT, mdt_reint),
4754 DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, IS_SUBDIR, mdt_is_subdir),
4757 static struct mdt_opc_slice mdt_xmds_handlers[] = {
4759 .mos_opc_start = MDS_GETATTR,
4760 .mos_opc_end = MDS_LAST_OPC,
4761 .mos_hs = mdt_xmds_ops
4764 .mos_opc_start = OBD_PING,
4765 .mos_opc_end = OBD_LAST_OPC,
4766 .mos_hs = mdt_obd_ops
4773 static struct mdt_handler mdt_seq_ops[] = {
4774 DEF_SEQ_HNDL_F(0, QUERY, (int (*)(struct mdt_thread_info *))seq_query)
4777 static struct mdt_opc_slice mdt_seq_handlers[] = {
4779 .mos_opc_start = SEQ_QUERY,
4780 .mos_opc_end = SEQ_LAST_OPC,
4781 .mos_hs = mdt_seq_ops
4788 static struct mdt_handler mdt_fld_ops[] = {
4789 DEF_FLD_HNDL_F(0, QUERY, (int (*)(struct mdt_thread_info *))fld_query)
4792 static struct mdt_opc_slice mdt_fld_handlers[] = {
4794 .mos_opc_start = FLD_QUERY,
4795 .mos_opc_end = FLD_LAST_OPC,
4796 .mos_hs = mdt_fld_ops
4803 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
4804 MODULE_DESCRIPTION("Lustre Meta-data Target ("LUSTRE_MDT_NAME")");
4805 MODULE_LICENSE("GPL");
4807 CFS_MODULE_PARM(mdt_num_threads, "ul", ulong, 0444,
4808 "number of mdt service threads to start");
4810 cfs_module(mdt, "0.2.0", mdt_mod_init, mdt_mod_exit);