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;
391 /* FIXME: these should be fixed when new igif ready.*/
392 b->ino = fid_oid(fid); /* 1.6 compatibility */
393 b->generation = fid_ver(fid); /* 1.6 compatibility */
394 b->valid |= OBD_MD_FLGENER; /* 1.6 compatibility */
396 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, size="LPU64"\n",
397 PFID(fid), b->nlink, b->mode, b->size);
401 mdt_body_reverse_idmap(info, b);
404 static inline int mdt_body_has_lov(const struct lu_attr *la,
405 const struct mdt_body *body)
407 return ((S_ISREG(la->la_mode) && (body->valid & OBD_MD_FLEASIZE)) ||
408 (S_ISDIR(la->la_mode) && (body->valid & OBD_MD_FLDIREA )) );
411 static int mdt_getattr_internal(struct mdt_thread_info *info,
412 struct mdt_object *o)
414 struct md_object *next = mdt_object_child(o);
415 const struct mdt_body *reqbody = info->mti_body;
416 struct ptlrpc_request *req = mdt_info_req(info);
417 struct mdt_export_data *med = &req->rq_export->exp_mdt_data;
418 struct md_attr *ma = &info->mti_attr;
419 struct lu_attr *la = &ma->ma_attr;
420 struct req_capsule *pill = info->mti_pill;
421 const struct lu_env *env = info->mti_env;
422 struct mdt_body *repbody;
423 struct lu_buf *buffer = &info->mti_buf;
427 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
428 RETURN(err_serious(-ENOMEM));
430 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
434 rc = mdt_object_exists(o);
436 /* This object is located on remote node.*/
437 repbody->fid1 = *mdt_object_fid(o);
438 repbody->valid = OBD_MD_FLID | OBD_MD_MDS;
442 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
443 buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD, RCL_SERVER);
445 /* If it is dir object and client require MEA, then we got MEA */
446 if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
447 reqbody->valid & OBD_MD_MEA) {
448 /* Assumption: MDT_MD size is enough for lmv size. */
449 ma->ma_lmv = buffer->lb_buf;
450 ma->ma_lmv_size = buffer->lb_len;
451 ma->ma_need = MA_LMV | MA_INODE;
453 ma->ma_lmm = buffer->lb_buf;
454 ma->ma_lmm_size = buffer->lb_len;
455 ma->ma_need = MA_LOV | MA_INODE;
458 if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
459 reqbody->valid & OBD_MD_FLDIREA &&
460 lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
461 /* get default stripe info for this dir. */
462 ma->ma_need |= MA_LOV_DEF;
464 rc = mo_attr_get(env, next, ma);
466 CERROR("getattr error for "DFID": %d\n",
467 PFID(mdt_object_fid(o)), rc);
471 if (likely(ma->ma_valid & MA_INODE))
472 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
476 if (mdt_body_has_lov(la, reqbody)) {
477 if (ma->ma_valid & MA_LOV) {
478 LASSERT(ma->ma_lmm_size);
479 mdt_dump_lmm(D_INFO, ma->ma_lmm);
480 repbody->eadatasize = ma->ma_lmm_size;
481 if (S_ISDIR(la->la_mode))
482 repbody->valid |= OBD_MD_FLDIREA;
484 repbody->valid |= OBD_MD_FLEASIZE;
486 if (ma->ma_valid & MA_LMV) {
487 LASSERT(S_ISDIR(la->la_mode));
488 repbody->eadatasize = ma->ma_lmv_size;
489 repbody->valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
491 if (!(ma->ma_valid & MA_LOV) && !(ma->ma_valid & MA_LMV)) {
492 repbody->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
494 } else if (S_ISLNK(la->la_mode) &&
495 reqbody->valid & OBD_MD_LINKNAME) {
496 buffer->lb_buf = ma->ma_lmm;
497 buffer->lb_len = reqbody->eadatasize;
498 rc = mo_readlink(env, next, buffer);
499 if (unlikely(rc <= 0)) {
500 CERROR("readlink failed: %d\n", rc);
503 repbody->valid |= OBD_MD_LINKNAME;
504 repbody->eadatasize = rc;
506 ((char*)ma->ma_lmm)[rc - 1] = 0;
507 CDEBUG(D_INODE, "symlink dest %s, len = %d\n",
508 (char*)ma->ma_lmm, rc);
513 if (reqbody->valid & OBD_MD_FLMODEASIZE) {
514 repbody->max_cookiesize = info->mti_mdt->mdt_max_cookiesize;
515 repbody->max_mdsize = info->mti_mdt->mdt_max_mdsize;
516 repbody->valid |= OBD_MD_FLMODEASIZE;
517 CDEBUG(D_INODE, "I am going to change the MAX_MD_SIZE & "
518 "MAX_COOKIE to : %d:%d\n", repbody->max_mdsize,
519 repbody->max_cookiesize);
522 if (med->med_rmtclient && (reqbody->valid & OBD_MD_FLRMTPERM)) {
523 void *buf = req_capsule_server_get(pill, &RMF_ACL);
525 /* mdt_getattr_lock only */
526 rc = mdt_pack_remote_perm(info, o, buf);
528 repbody->valid &= ~OBD_MD_FLRMTPERM;
529 repbody->aclsize = 0;
532 repbody->valid |= OBD_MD_FLRMTPERM;
533 repbody->aclsize = sizeof(struct mdt_remote_perm);
536 #ifdef CONFIG_FS_POSIX_ACL
537 else if ((req->rq_export->exp_connect_flags & OBD_CONNECT_ACL) &&
538 (reqbody->valid & OBD_MD_FLACL)) {
539 buffer->lb_buf = req_capsule_server_get(pill, &RMF_ACL);
540 buffer->lb_len = req_capsule_get_size(pill,
541 &RMF_ACL, RCL_SERVER);
542 if (buffer->lb_len > 0) {
543 rc = mo_xattr_get(env, next, buffer,
544 XATTR_NAME_ACL_ACCESS);
546 if (rc == -ENODATA) {
547 repbody->aclsize = 0;
548 repbody->valid |= OBD_MD_FLACL;
550 } else if (rc == -EOPNOTSUPP) {
553 CERROR("got acl size: %d\n", rc);
556 repbody->aclsize = rc;
557 repbody->valid |= OBD_MD_FLACL;
564 if ((reqbody->valid & OBD_MD_FLMDSCAPA) &&
565 info->mti_mdt->mdt_opts.mo_mds_capa) {
566 struct lustre_capa *capa;
568 capa = req_capsule_server_get(pill, &RMF_CAPA1);
570 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
571 rc = mo_capa_get(env, next, capa, 0);
574 repbody->valid |= OBD_MD_FLMDSCAPA;
579 static int mdt_renew_capa(struct mdt_thread_info *info)
581 struct mdt_device *mdt = info->mti_mdt;
582 struct mdt_object *obj = info->mti_object;
583 struct mdt_body *body;
584 struct lustre_capa *capa, *c;
588 /* if object doesn't exist, or server has disabled capability,
589 * return directly, client will find body->valid OBD_MD_FLOSSCAPA
592 if (!obj || !mdt->mdt_opts.mo_mds_capa)
595 body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
596 LASSERT(body != NULL);
598 c = req_capsule_client_get(info->mti_pill, &RMF_CAPA1);
601 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
605 rc = mo_capa_get(info->mti_env, mdt_object_child(obj), capa, 1);
607 body->valid |= OBD_MD_FLOSSCAPA;
611 static int mdt_getattr(struct mdt_thread_info *info)
613 struct mdt_object *obj = info->mti_object;
614 struct req_capsule *pill = info->mti_pill;
615 struct mdt_body *reqbody;
616 struct mdt_body *repbody;
622 reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
625 if (reqbody->valid & OBD_MD_FLOSSCAPA) {
626 rc = req_capsule_server_pack(pill);
628 RETURN(err_serious(rc));
629 rc = mdt_renew_capa(info);
630 GOTO(out_shrink, rc);
633 LASSERT(obj != NULL);
634 LASSERT(lu_object_assert_exists(&obj->mot_obj.mo_lu));
636 mode = lu_object_attr(&obj->mot_obj.mo_lu);
637 if (S_ISLNK(mode) && (reqbody->valid & OBD_MD_LINKNAME) &&
638 (reqbody->eadatasize > info->mti_mdt->mdt_max_mdsize))
639 md_size = reqbody->eadatasize;
641 md_size = info->mti_mdt->mdt_max_mdsize;
643 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, md_size);
645 rc = req_capsule_server_pack(pill);
646 if (unlikely(rc != 0))
647 RETURN(err_serious(rc));
649 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
650 LASSERT(repbody != NULL);
651 repbody->eadatasize = 0;
652 repbody->aclsize = 0;
654 if (reqbody->valid & OBD_MD_FLRMTPERM)
655 rc = mdt_init_ucred(info, reqbody);
657 rc = mdt_check_ucred(info);
659 GOTO(out_shrink, rc);
661 info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
662 info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
665 * Don't check capability at all, because rename might getattr for
666 * remote obj, and at that time no capability is available.
668 mdt_set_capainfo(info, 1, &reqbody->fid1, BYPASS_CAPA);
669 rc = mdt_getattr_internal(info, obj);
670 if (reqbody->valid & OBD_MD_FLRMTPERM)
671 mdt_exit_ucred(info);
674 mdt_shrink_reply(info);
678 static int mdt_is_subdir(struct mdt_thread_info *info)
680 struct mdt_object *o = info->mti_object;
681 struct req_capsule *pill = info->mti_pill;
682 const struct mdt_body *body = info->mti_body;
683 struct mdt_body *repbody;
689 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
692 * We save last checked parent fid to @repbody->fid1 for remote
695 LASSERT(fid_is_sane(&body->fid2));
696 LASSERT(mdt_object_exists(o) > 0);
697 rc = mdo_is_subdir(info->mti_env, mdt_object_child(o),
698 &body->fid2, &repbody->fid1);
699 if (rc == 0 || rc == -EREMOTE)
700 repbody->valid |= OBD_MD_FLID;
705 static int mdt_raw_lookup(struct mdt_thread_info *info,
706 struct mdt_object *parent,
707 const struct lu_name *lname,
708 struct ldlm_reply *ldlm_rep)
710 struct md_object *next = mdt_object_child(info->mti_object);
711 const struct mdt_body *reqbody = info->mti_body;
712 struct lu_fid *child_fid = &info->mti_tmp_fid1;
713 struct mdt_body *repbody;
717 if (reqbody->valid != OBD_MD_FLID)
720 LASSERT(!info->mti_cross_ref);
722 /* Only got the fid of this obj by name */
723 rc = mdo_lookup(info->mti_env, next, lname, child_fid,
726 /* XXX is raw_lookup possible as intent operation? */
729 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
732 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
734 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
737 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
738 repbody->fid1 = *child_fid;
739 repbody->valid = OBD_MD_FLID;
745 * UPDATE lock should be taken against parent, and be release before exit;
746 * child_bits lock should be taken against child, and be returned back:
747 * (1)normal request should release the child lock;
748 * (2)intent request will grant the lock to client.
750 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
751 struct mdt_lock_handle *lhc,
753 struct ldlm_reply *ldlm_rep)
755 struct ptlrpc_request *req = mdt_info_req(info);
756 struct mdt_body *reqbody = NULL;
757 struct mdt_object *parent = info->mti_object;
758 struct mdt_object *child;
759 struct md_object *next = mdt_object_child(parent);
760 struct lu_fid *child_fid = &info->mti_tmp_fid1;
761 struct lu_name *lname = NULL;
764 struct mdt_lock_handle *lhp;
765 struct ldlm_lock *lock;
766 struct ldlm_res_id *res_id;
772 is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
773 LASSERT(ergo(is_resent,
774 lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
776 LASSERT(parent != NULL);
777 name = req_capsule_client_get(info->mti_pill, &RMF_NAME);
779 RETURN(err_serious(-EFAULT));
781 namelen = req_capsule_get_size(info->mti_pill, &RMF_NAME,
783 LASSERT(namelen >= 0);
785 /* XXX: "namelen == 0" is for getattr by fid (OBD_CONNECT_ATTRFID),
786 * otherwise do not allow empty name, that is the name must contain
787 * at least one character and the terminating '\0'*/
789 reqbody =req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
790 LASSERT(fid_is_sane(&reqbody->fid2));
793 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
795 PFID(mdt_object_fid(parent)), PFID(&reqbody->fid2),
798 lname = mdt_name(info->mti_env, (char *)name, namelen);
799 CDEBUG(D_INODE, "getattr with lock for "DFID"/%s, "
801 PFID(mdt_object_fid(parent)), name, ldlm_rep);
804 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
806 rc = mdt_object_exists(parent);
807 if (unlikely(rc == 0)) {
808 LU_OBJECT_DEBUG(D_WARNING, info->mti_env,
809 &parent->mot_obj.mo_lu,
810 "Parent doesn't exist!\n");
813 LASSERTF(rc > 0, "Parent "DFID" is on remote server\n",
814 PFID(mdt_object_fid(parent)));
817 rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
825 if (info->mti_cross_ref) {
826 /* Only getattr on the child. Parent is on another node. */
827 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
829 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
830 "ldlm_rep=%p\n", PFID(mdt_object_fid(child)), ldlm_rep);
833 /* Do not take lock for resent case. */
834 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
835 LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
836 lhc->mlh_reg_lh.cookie);
837 LASSERT(fid_res_name_eq(mdt_object_fid(child),
838 &lock->l_resource->lr_name));
842 mdt_lock_handle_init(lhc);
843 mdt_lock_reg_init(lhc, LCK_PR);
846 * Object's name is on another MDS, no lookup lock is
847 * needed here but update is.
849 child_bits &= ~MDS_INODELOCK_LOOKUP;
850 child_bits |= MDS_INODELOCK_UPDATE;
852 rc = mdt_object_lock(info, child, lhc, child_bits,
856 /* Finally, we can get attr for child. */
857 mdt_set_capainfo(info, 0, mdt_object_fid(child),
859 rc = mdt_getattr_internal(info, child);
860 if (unlikely(rc != 0))
861 mdt_object_unlock(info, child, lhc, 1);
866 /* step 1: lock parent */
867 lhp = &info->mti_lh[MDT_LH_PARENT];
868 mdt_lock_pdo_init(lhp, LCK_PR, name, namelen);
869 rc = mdt_object_lock(info, parent, lhp, MDS_INODELOCK_UPDATE,
872 if (unlikely(rc != 0))
876 /* step 2: lookup child's fid by name */
877 rc = mdo_lookup(info->mti_env, next, lname, child_fid,
882 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
883 GOTO(out_parent, rc);
885 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
887 *child_fid = reqbody->fid2;
888 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
892 *step 3: find the child object by fid & lock it.
893 * regardless if it is local or remote.
895 child = mdt_object_find(info->mti_env, info->mti_mdt, child_fid);
897 if (unlikely(IS_ERR(child)))
898 GOTO(out_parent, rc = PTR_ERR(child));
900 /* Do not take lock for resent case. */
901 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
902 LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
903 lhc->mlh_reg_lh.cookie);
905 res_id = &lock->l_resource->lr_name;
906 if (!fid_res_name_eq(mdt_object_fid(child),
907 &lock->l_resource->lr_name)) {
908 LASSERTF(fid_res_name_eq(mdt_object_fid(parent),
909 &lock->l_resource->lr_name),
910 "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n",
911 (unsigned long)res_id->name[0],
912 (unsigned long)res_id->name[1],
913 (unsigned long)res_id->name[2],
914 PFID(mdt_object_fid(parent)));
915 CWARN("Although resent, but still not get child lock"
916 "parent:"DFID" child:"DFID"\n",
917 PFID(mdt_object_fid(parent)),
918 PFID(mdt_object_fid(child)));
919 lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
926 struct md_attr *ma = &info->mti_attr;
928 mdt_lock_handle_init(lhc);
929 mdt_lock_reg_init(lhc, LCK_PR);
931 if (mdt_object_exists(child) == 0) {
932 LU_OBJECT_DEBUG(D_WARNING, info->mti_env,
933 &child->mot_obj.mo_lu,
934 "Object doesn't exist!\n");
938 ma->ma_need = MA_INODE;
939 rc = mo_attr_get(info->mti_env, next, ma);
940 if (unlikely(rc != 0))
943 /* If the file has not been changed for some time, we return
944 * not only a LOOKUP lock, but also an UPDATE lock and this
945 * might save us RPC on later STAT. For directories, it also
946 * let negative dentry starts working for this dir. */
947 if (ma->ma_valid & MA_INODE &&
948 ma->ma_attr.la_valid & LA_CTIME &&
949 info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
950 ma->ma_attr.la_ctime < cfs_time_current_sec())
951 child_bits |= MDS_INODELOCK_UPDATE;
953 rc = mdt_object_lock(info, child, lhc, child_bits,
956 if (unlikely(rc != 0))
960 /* finally, we can get attr for child. */
961 mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
962 rc = mdt_getattr_internal(info, child);
963 if (unlikely(rc != 0)) {
964 mdt_object_unlock(info, child, lhc, 1);
966 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
968 struct mdt_body *repbody;
971 /* Debugging code. */
972 res_id = &lock->l_resource->lr_name;
973 LDLM_DEBUG(lock, "Returning lock to client\n");
974 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
975 &lock->l_resource->lr_name),
976 "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n",
977 (unsigned long)res_id->name[0],
978 (unsigned long)res_id->name[1],
979 (unsigned long)res_id->name[2],
980 PFID(mdt_object_fid(child)));
982 * Pack Size-on-MDS inode attributes to the body if
983 * update lock is given.
985 repbody = req_capsule_server_get(info->mti_pill,
987 ma = &info->mti_attr.ma_attr;
988 if (lock->l_policy_data.l_inodebits.bits &
989 MDS_INODELOCK_UPDATE)
990 mdt_pack_size2body(info, child);
996 mdt_object_put(info->mti_env, child);
998 mdt_object_unlock(info, parent, lhp, 1);
1002 /* normal handler: should release the child lock */
1003 static int mdt_getattr_name(struct mdt_thread_info *info)
1005 struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1006 struct mdt_body *reqbody;
1007 struct mdt_body *repbody;
1011 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1012 LASSERT(reqbody != NULL);
1013 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1014 LASSERT(repbody != NULL);
1016 info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
1017 info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
1018 repbody->eadatasize = 0;
1019 repbody->aclsize = 0;
1021 rc = mdt_init_ucred(info, reqbody);
1023 GOTO(out_shrink, rc);
1025 rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1026 if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1027 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1028 lhc->mlh_reg_lh.cookie = 0;
1030 mdt_exit_ucred(info);
1033 mdt_shrink_reply(info);
1037 static struct lu_device_operations mdt_lu_ops;
1039 static int lu_device_is_mdt(struct lu_device *d)
1041 return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &mdt_lu_ops);
1044 static int mdt_set_info(struct mdt_thread_info *info)
1046 struct ptlrpc_request *req = mdt_info_req(info);
1052 rc = req_capsule_server_pack(info->mti_pill);
1056 key = req_capsule_client_get(info->mti_pill, &RMF_SETINFO_KEY);
1058 DEBUG_REQ(D_HA, req, "no set_info key");
1062 keylen = req_capsule_get_size(info->mti_pill, &RMF_SETINFO_KEY,
1065 val = req_capsule_client_get(info->mti_pill, &RMF_SETINFO_VAL);
1067 DEBUG_REQ(D_HA, req, "no set_info val");
1071 if (keylen != (sizeof(KEY_READ_ONLY) - 1) ||
1072 memcmp(key, KEY_READ_ONLY, keylen) != 0)
1076 lustre_msg_set_status(req->rq_repmsg, 0);
1078 spin_lock(&req->rq_export->exp_lock);
1080 req->rq_export->exp_connect_flags |= OBD_CONNECT_RDONLY;
1082 req->rq_export->exp_connect_flags &= ~OBD_CONNECT_RDONLY;
1083 spin_unlock(&req->rq_export->exp_lock);
1088 static int mdt_connect(struct mdt_thread_info *info)
1091 struct ptlrpc_request *req;
1093 req = mdt_info_req(info);
1094 rc = target_handle_connect(req);
1096 LASSERT(req->rq_export != NULL);
1097 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
1098 rc = mdt_init_idmap(info);
1100 /* if mdt_init_idmap failed, revocation for connect */
1101 obd_disconnect(class_export_get(req->rq_export));
1103 rc = err_serious(rc);
1107 static int mdt_disconnect(struct mdt_thread_info *info)
1112 rc = target_handle_disconnect(mdt_info_req(info));
1114 rc = err_serious(rc);
1118 static int mdt_sendpage(struct mdt_thread_info *info,
1119 struct lu_rdpg *rdpg)
1121 struct ptlrpc_request *req = mdt_info_req(info);
1122 struct ptlrpc_bulk_desc *desc;
1123 struct l_wait_info *lwi = &info->mti_u.rdpg.mti_wait_info;
1130 desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, BULK_PUT_SOURCE,
1135 for (i = 0, tmpcount = rdpg->rp_count;
1136 i < rdpg->rp_npages; i++, tmpcount -= tmpsize) {
1137 tmpsize = min_t(int, tmpcount, CFS_PAGE_SIZE);
1138 ptlrpc_prep_bulk_page(desc, rdpg->rp_pages[i], 0, tmpsize);
1141 LASSERT(desc->bd_nob == rdpg->rp_count);
1142 rc = ptlrpc_start_bulk_transfer(desc);
1144 GOTO(free_desc, rc);
1146 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1147 GOTO(abort_bulk, rc = 0);
1149 *lwi = LWI_TIMEOUT(obd_timeout * HZ / 4, NULL, NULL);
1150 rc = l_wait_event(desc->bd_waitq, !ptlrpc_bulk_active(desc), lwi);
1151 LASSERT (rc == 0 || rc == -ETIMEDOUT);
1154 if (desc->bd_success &&
1155 desc->bd_nob_transferred == rdpg->rp_count)
1156 GOTO(free_desc, rc);
1158 rc = -ETIMEDOUT; /* XXX should this be a different errno? */
1161 DEBUG_REQ(D_ERROR, req, "bulk failed: %s %d(%d), evicting %s@%s",
1162 (rc == -ETIMEDOUT) ? "timeout" : "network error",
1163 desc->bd_nob_transferred, rdpg->rp_count,
1164 req->rq_export->exp_client_uuid.uuid,
1165 req->rq_export->exp_connection->c_remote_uuid.uuid);
1167 class_fail_export(req->rq_export);
1171 ptlrpc_abort_bulk(desc);
1173 ptlrpc_free_bulk(desc);
1177 #ifdef HAVE_SPLIT_SUPPORT
1179 * Retrieve dir entry from the page and insert it to the slave object, actually,
1180 * this should be in osd layer, but since it will not in the final product, so
1181 * just do it here and do not define more moo api anymore for this.
1183 static int mdt_write_dir_page(struct mdt_thread_info *info, struct page *page,
1186 struct mdt_object *object = info->mti_object;
1187 struct lu_fid *lf = &info->mti_tmp_fid2;
1188 struct md_attr *ma = &info->mti_attr;
1189 struct lu_dirpage *dp;
1190 struct lu_dirent *ent;
1191 int rc = 0, offset = 0;
1194 /* Make sure we have at least one entry. */
1199 * Disable trans for this name insert, since it will include many trans
1202 info->mti_no_need_trans = 1;
1204 * When write_dir_page, no need update parent's ctime,
1205 * and no permission check for name_insert.
1207 ma->ma_attr.la_ctime = 0;
1208 ma->ma_attr.la_valid = LA_MODE;
1209 ma->ma_valid = MA_INODE;
1212 dp = page_address(page);
1213 offset = (int)((__u32)lu_dirent_start(dp) - (__u32)dp);
1215 for (ent = lu_dirent_start(dp); ent != NULL;
1216 ent = lu_dirent_next(ent)) {
1217 struct lu_name *lname;
1220 if (le16_to_cpu(ent->lde_namelen) == 0)
1223 fid_le_to_cpu(lf, &ent->lde_fid);
1224 if (le32_to_cpu(ent->lde_hash) & MAX_HASH_HIGHEST_BIT)
1225 ma->ma_attr.la_mode = S_IFDIR;
1227 ma->ma_attr.la_mode = 0;
1228 OBD_ALLOC(name, le16_to_cpu(ent->lde_namelen) + 1);
1230 GOTO(out, rc = -ENOMEM);
1232 memcpy(name, ent->lde_name, le16_to_cpu(ent->lde_namelen));
1233 lname = mdt_name(info->mti_env, name,
1234 le16_to_cpu(ent->lde_namelen) + 1);
1235 ma->ma_attr_flags |= MDS_PERM_BYPASS;
1236 rc = mdo_name_insert(info->mti_env,
1237 md_object_next(&object->mot_obj),
1239 OBD_FREE(name, le16_to_cpu(ent->lde_namelen) + 1);
1241 CERROR("Can't insert %*.*s, rc %d\n",
1242 le16_to_cpu(ent->lde_namelen),
1243 le16_to_cpu(ent->lde_namelen),
1248 offset += lu_dirent_size(ent);
1258 static int mdt_bulk_timeout(void *data)
1262 CERROR("mdt bulk transfer timeout \n");
1267 static int mdt_writepage(struct mdt_thread_info *info)
1269 struct ptlrpc_request *req = mdt_info_req(info);
1270 struct mdt_body *reqbody;
1271 struct l_wait_info *lwi;
1272 struct ptlrpc_bulk_desc *desc;
1278 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1279 if (reqbody == NULL)
1280 RETURN(err_serious(-EFAULT));
1282 desc = ptlrpc_prep_bulk_exp(req, 1, BULK_GET_SINK, MDS_BULK_PORTAL);
1284 RETURN(err_serious(-ENOMEM));
1286 /* allocate the page for the desc */
1287 page = cfs_alloc_page(CFS_ALLOC_STD);
1289 GOTO(desc_cleanup, rc = -ENOMEM);
1291 CDEBUG(D_INFO, "Received page offset %d size %d \n",
1292 (int)reqbody->size, (int)reqbody->nlink);
1294 ptlrpc_prep_bulk_page(desc, page, (int)reqbody->size,
1295 (int)reqbody->nlink);
1298 * Check if client was evicted while we were doing i/o before touching
1303 GOTO(cleanup_page, rc = -ENOMEM);
1305 if (desc->bd_export->exp_failed)
1308 rc = ptlrpc_start_bulk_transfer (desc);
1310 *lwi = LWI_TIMEOUT_INTERVAL(obd_timeout * HZ / 4, HZ,
1311 mdt_bulk_timeout, desc);
1312 rc = l_wait_event(desc->bd_waitq, !ptlrpc_bulk_active(desc) ||
1313 desc->bd_export->exp_failed, lwi);
1314 LASSERT(rc == 0 || rc == -ETIMEDOUT);
1315 if (rc == -ETIMEDOUT) {
1316 DEBUG_REQ(D_ERROR, req, "timeout on bulk GET");
1317 ptlrpc_abort_bulk(desc);
1318 } else if (desc->bd_export->exp_failed) {
1319 DEBUG_REQ(D_ERROR, req, "Eviction on bulk GET");
1321 ptlrpc_abort_bulk(desc);
1322 } else if (!desc->bd_success ||
1323 desc->bd_nob_transferred != desc->bd_nob) {
1324 DEBUG_REQ(D_ERROR, req, "%s bulk GET %d(%d)",
1326 "truncated" : "network error on",
1327 desc->bd_nob_transferred, desc->bd_nob);
1328 /* XXX should this be a different errno? */
1332 DEBUG_REQ(D_ERROR, req, "ptlrpc_bulk_get failed: rc %d", rc);
1335 GOTO(cleanup_lwi, rc);
1336 rc = mdt_write_dir_page(info, page, reqbody->nlink);
1341 cfs_free_page(page);
1343 ptlrpc_free_bulk(desc);
1348 static int mdt_readpage(struct mdt_thread_info *info)
1350 struct mdt_object *object = info->mti_object;
1351 struct lu_rdpg *rdpg = &info->mti_u.rdpg.mti_rdpg;
1352 struct mdt_body *reqbody;
1353 struct mdt_body *repbody;
1358 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
1359 RETURN(err_serious(-ENOMEM));
1361 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1362 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1363 if (reqbody == NULL || repbody == NULL)
1364 RETURN(err_serious(-EFAULT));
1367 * prepare @rdpg before calling lower layers and transfer itself. Here
1368 * reqbody->size contains offset of where to start to read and
1369 * reqbody->nlink contains number bytes to read.
1371 rdpg->rp_hash = reqbody->size;
1372 if ((__u64)rdpg->rp_hash != reqbody->size) {
1373 CERROR("Invalid hash: %#llx != %#llx\n",
1374 (__u64)rdpg->rp_hash, reqbody->size);
1377 rdpg->rp_count = reqbody->nlink;
1378 rdpg->rp_npages = (rdpg->rp_count + CFS_PAGE_SIZE - 1)>>CFS_PAGE_SHIFT;
1379 OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1380 if (rdpg->rp_pages == NULL)
1383 for (i = 0; i < rdpg->rp_npages; ++i) {
1384 rdpg->rp_pages[i] = cfs_alloc_page(CFS_ALLOC_STD);
1385 if (rdpg->rp_pages[i] == NULL)
1386 GOTO(free_rdpg, rc = -ENOMEM);
1389 /* call lower layers to fill allocated pages with directory data */
1390 rc = mo_readpage(info->mti_env, mdt_object_child(object), rdpg);
1392 GOTO(free_rdpg, rc);
1394 /* send pages to client */
1395 rc = mdt_sendpage(info, rdpg);
1400 for (i = 0; i < rdpg->rp_npages; i++)
1401 if (rdpg->rp_pages[i] != NULL)
1402 cfs_free_page(rdpg->rp_pages[i]);
1403 OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1405 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1411 static int mdt_reint_internal(struct mdt_thread_info *info,
1412 struct mdt_lock_handle *lhc,
1415 struct req_capsule *pill = info->mti_pill;
1416 struct mdt_device *mdt = info->mti_mdt;
1417 struct mdt_body *repbody;
1422 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1423 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1424 mdt->mdt_max_mdsize);
1425 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1426 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER,
1427 mdt->mdt_max_cookiesize);
1429 rc = req_capsule_server_pack(pill);
1431 CERROR("Can't pack response, rc %d\n", rc);
1432 RETURN(err_serious(rc));
1435 if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
1436 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1438 repbody->eadatasize = 0;
1439 repbody->aclsize = 0;
1442 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNPACK))
1443 GOTO(out_shrink, rc = err_serious(-EFAULT));
1445 rc = mdt_reint_unpack(info, op);
1447 CERROR("Can't unpack reint, rc %d\n", rc);
1448 GOTO(out_shrink, rc = err_serious(rc));
1451 rc = mdt_init_ucred_reint(info);
1453 GOTO(out_shrink, rc);
1455 rc = mdt_fix_attr_ucred(info, op);
1457 GOTO(out_ucred, rc = err_serious(rc));
1459 if (mdt_check_resent(info, mdt_reconstruct, lhc)) {
1460 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
1461 GOTO(out_ucred, rc);
1463 rc = mdt_reint_rec(info, lhc);
1466 mdt_exit_ucred(info);
1468 mdt_shrink_reply(info);
1472 static long mdt_reint_opcode(struct mdt_thread_info *info,
1473 const struct req_format **fmt)
1475 struct mdt_rec_reint *rec;
1478 opc = err_serious(-EFAULT);
1479 rec = req_capsule_client_get(info->mti_pill, &RMF_REC_REINT);
1481 opc = rec->rr_opcode;
1482 DEBUG_REQ(D_INODE, mdt_info_req(info), "reint opt = %ld", opc);
1483 if (opc < REINT_MAX && fmt[opc] != NULL)
1484 req_capsule_extend(info->mti_pill, fmt[opc]);
1486 CERROR("Unsupported opc: %ld\n", opc);
1487 opc = err_serious(opc);
1493 static int mdt_reint(struct mdt_thread_info *info)
1498 static const struct req_format *reint_fmts[REINT_MAX] = {
1499 [REINT_SETATTR] = &RQF_MDS_REINT_SETATTR,
1500 [REINT_CREATE] = &RQF_MDS_REINT_CREATE,
1501 [REINT_LINK] = &RQF_MDS_REINT_LINK,
1502 [REINT_UNLINK] = &RQF_MDS_REINT_UNLINK,
1503 [REINT_RENAME] = &RQF_MDS_REINT_RENAME,
1504 [REINT_OPEN] = &RQF_MDS_REINT_OPEN,
1505 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR
1510 opc = mdt_reint_opcode(info, reint_fmts);
1513 * No lock possible here from client to pass it to reint code
1516 rc = mdt_reint_internal(info, NULL, opc);
1521 info->mti_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
1525 /* TODO these two methods not available now. */
1527 /* this should sync the whole device */
1528 static int mdt_device_sync(struct mdt_thread_info *info)
1533 /* this should sync this object */
1534 static int mdt_object_sync(struct mdt_thread_info *info)
1539 static int mdt_sync(struct mdt_thread_info *info)
1541 struct req_capsule *pill = info->mti_pill;
1542 struct mdt_body *body;
1546 /* The fid may be zero, so we req_capsule_set manually */
1547 req_capsule_set(pill, &RQF_MDS_SYNC);
1549 body = req_capsule_client_get(pill, &RMF_MDT_BODY);
1551 RETURN(err_serious(-EINVAL));
1553 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
1554 RETURN(err_serious(-ENOMEM));
1556 if (fid_seq(&body->fid1) == 0) {
1557 /* sync the whole device */
1558 rc = req_capsule_server_pack(pill);
1560 rc = mdt_device_sync(info);
1562 rc = err_serious(rc);
1564 /* sync an object */
1565 rc = mdt_unpack_req_pack_rep(info, HABEO_CORPUS|HABEO_REFERO);
1567 rc = mdt_object_sync(info);
1569 struct md_object *next;
1570 const struct lu_fid *fid;
1571 struct lu_attr *la = &info->mti_attr.ma_attr;
1573 next = mdt_object_child(info->mti_object);
1574 info->mti_attr.ma_need = MA_INODE;
1575 info->mti_attr.ma_valid = 0;
1576 rc = mo_attr_get(info->mti_env, next,
1579 body = req_capsule_server_get(pill,
1581 fid = mdt_object_fid(info->mti_object);
1582 mdt_pack_attr2body(info, body, la, fid);
1586 rc = err_serious(rc);
1591 static int mdt_quotacheck_handle(struct mdt_thread_info *info)
1593 return err_serious(-EOPNOTSUPP);
1596 static int mdt_quotactl_handle(struct mdt_thread_info *info)
1598 return err_serious(-EOPNOTSUPP);
1602 * OBD PING and other handlers.
1604 static int mdt_obd_ping(struct mdt_thread_info *info)
1609 req_capsule_set(info->mti_pill, &RQF_OBD_PING);
1611 rc = target_handle_ping(mdt_info_req(info));
1613 rc = err_serious(rc);
1617 static int mdt_obd_log_cancel(struct mdt_thread_info *info)
1619 return err_serious(-EOPNOTSUPP);
1622 static int mdt_obd_qc_callback(struct mdt_thread_info *info)
1624 return err_serious(-EOPNOTSUPP);
1631 static struct ldlm_callback_suite cbs = {
1632 .lcs_completion = ldlm_server_completion_ast,
1633 .lcs_blocking = ldlm_server_blocking_ast,
1637 static int mdt_enqueue(struct mdt_thread_info *info)
1639 struct ptlrpc_request *req;
1644 * info->mti_dlm_req already contains swapped and (if necessary)
1645 * converted dlm request.
1647 LASSERT(info->mti_dlm_req != NULL);
1649 req = mdt_info_req(info);
1652 * Lock without inodebits makes no sense and will oops later in
1653 * ldlm. Let's check it now to see if we have wrong lock from client or
1654 * bits get corrupted somewhere in mdt_intent_policy().
1656 req_bits = info->mti_dlm_req->lock_desc.l_policy_data.l_inodebits.bits;
1657 /* This is disabled because we need to support liblustre flock.
1658 * LASSERT(req_bits != 0);
1661 rc = ldlm_handle_enqueue0(info->mti_mdt->mdt_namespace,
1662 req, info->mti_dlm_req, &cbs);
1663 info->mti_fail_id = OBD_FAIL_LDLM_REPLY;
1664 return rc ? err_serious(rc) : req->rq_status;
1667 static int mdt_convert(struct mdt_thread_info *info)
1670 struct ptlrpc_request *req;
1672 LASSERT(info->mti_dlm_req);
1673 req = mdt_info_req(info);
1674 rc = ldlm_handle_convert0(req, info->mti_dlm_req);
1675 return rc ? err_serious(rc) : req->rq_status;
1678 static int mdt_bl_callback(struct mdt_thread_info *info)
1680 CERROR("bl callbacks should not happen on MDS\n");
1682 return err_serious(-EOPNOTSUPP);
1685 static int mdt_cp_callback(struct mdt_thread_info *info)
1687 CERROR("cp callbacks should not happen on MDS\n");
1689 return err_serious(-EOPNOTSUPP);
1693 * sec context handlers
1695 static int mdt_sec_ctx_handle(struct mdt_thread_info *info)
1699 rc = mdt_handle_idmap(info);
1702 struct ptlrpc_request *req = mdt_info_req(info);
1705 opc = lustre_msg_get_opc(req->rq_reqmsg);
1706 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT)
1707 sptlrpc_svc_ctx_invalidate(req);
1713 static struct mdt_object *mdt_obj(struct lu_object *o)
1715 LASSERT(lu_device_is_mdt(o->lo_dev));
1716 return container_of0(o, struct mdt_object, mot_obj.mo_lu);
1719 struct mdt_object *mdt_object_find(const struct lu_env *env,
1720 struct mdt_device *d,
1721 const struct lu_fid *f)
1723 struct lu_object *o;
1724 struct mdt_object *m;
1727 CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
1728 o = lu_object_find(env, d->mdt_md_dev.md_lu_dev.ld_site, f);
1729 if (unlikely(IS_ERR(o)))
1730 m = (struct mdt_object *)o;
1736 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
1737 struct mdt_lock_handle *lh, __u64 ibits, int locality)
1739 struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
1740 ldlm_policy_data_t *policy = &info->mti_policy;
1741 struct ldlm_res_id *res_id = &info->mti_res_id;
1745 LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
1746 LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
1747 LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
1748 LASSERT(lh->mlh_type != MDT_NUL_LOCK);
1750 if (mdt_object_exists(o) < 0) {
1751 if (locality == MDT_CROSS_LOCK) {
1752 /* cross-ref object fix */
1753 ibits &= ~MDS_INODELOCK_UPDATE;
1754 ibits |= MDS_INODELOCK_LOOKUP;
1756 LASSERT(!(ibits & MDS_INODELOCK_UPDATE));
1757 LASSERT(ibits & MDS_INODELOCK_LOOKUP);
1759 /* No PDO lock on remote object */
1760 LASSERT(lh->mlh_type != MDT_PDO_LOCK);
1763 memset(policy, 0, sizeof(*policy));
1764 fid_build_reg_res_name(mdt_object_fid(o), res_id);
1767 * Take PDO lock on whole directory and build correct @res_id for lock
1768 * on part of directory.
1770 if (lh->mlh_pdo_hash != 0) {
1771 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
1772 mdt_lock_pdo_mode(info, o, lh);
1773 if (lh->mlh_pdo_mode != LCK_NL) {
1775 * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
1776 * is never going to be sent to client and we do not
1777 * want it slowed down due to possible cancels.
1779 policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
1780 rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
1781 policy, res_id, LDLM_FL_ATOMIC_CB);
1787 * Finish res_id initializing by name hash marking patr of
1788 * directory which is taking modification.
1790 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
1793 policy->l_inodebits.bits = ibits;
1796 * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
1797 * going to be sent to client. If it is - mdt_intent_policy() path will
1798 * fix it up and turns FL_LOCAL flag off.
1800 rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
1801 res_id, LDLM_FL_LOCAL_ONLY | LDLM_FL_ATOMIC_CB);
1806 if (lh->mlh_type == MDT_PDO_LOCK) {
1807 /* check for exists after object is locked */
1808 if (mdt_object_exists(o) == 0) {
1809 /* Non-existent object shouldn't have PDO lock */
1812 /* Non-dir object shouldn't have PDO lock */
1813 LASSERT(S_ISDIR(lu_object_attr(&o->mot_obj.mo_lu)));
1818 mdt_object_unlock(info, o, lh, 1);
1825 * Just call ldlm_lock_decref() if decref, else we only call ptlrpc_save_lock()
1826 * to save this lock in req. when transaction committed, req will be released,
1827 * and lock will, too.
1829 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
1830 struct mdt_lock_handle *lh, int decref)
1832 struct ptlrpc_request *req = mdt_info_req(info);
1835 if (lustre_handle_is_used(&lh->mlh_pdo_lh)) {
1836 /* Do not save PDO locks to request, just decref. */
1837 mdt_fid_unlock(&lh->mlh_pdo_lh,
1839 lh->mlh_pdo_lh.cookie = 0ull;
1842 if (lustre_handle_is_used(&lh->mlh_reg_lh)) {
1844 mdt_fid_unlock(&lh->mlh_reg_lh,
1847 ptlrpc_save_lock(req, &lh->mlh_reg_lh,
1850 lh->mlh_reg_lh.cookie = 0ull;
1856 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
1857 const struct lu_fid *f,
1858 struct mdt_lock_handle *lh,
1861 struct mdt_object *o;
1863 o = mdt_object_find(info->mti_env, info->mti_mdt, f);
1867 rc = mdt_object_lock(info, o, lh, ibits,
1870 mdt_object_put(info->mti_env, o);
1877 void mdt_object_unlock_put(struct mdt_thread_info * info,
1878 struct mdt_object * o,
1879 struct mdt_lock_handle *lh,
1882 mdt_object_unlock(info, o, lh, decref);
1883 mdt_object_put(info->mti_env, o);
1886 static struct mdt_handler *mdt_handler_find(__u32 opc,
1887 struct mdt_opc_slice *supported)
1889 struct mdt_opc_slice *s;
1890 struct mdt_handler *h;
1893 for (s = supported; s->mos_hs != NULL; s++) {
1894 if (s->mos_opc_start <= opc && opc < s->mos_opc_end) {
1895 h = s->mos_hs + (opc - s->mos_opc_start);
1896 if (likely(h->mh_opc != 0))
1897 LASSERT(h->mh_opc == opc);
1899 h = NULL; /* unsupported opc */
1906 static int mdt_lock_resname_compat(struct mdt_device *m,
1907 struct ldlm_request *req)
1909 /* XXX something... later. */
1913 static int mdt_lock_reply_compat(struct mdt_device *m, struct ldlm_reply *rep)
1915 /* XXX something... later. */
1920 * Generic code handling requests that have struct mdt_body passed in:
1922 * - extract mdt_body from request and save it in @info, if present;
1924 * - create lu_object, corresponding to the fid in mdt_body, and save it in
1927 * - if HABEO_CORPUS flag is set for this request type check whether object
1928 * actually exists on storage (lu_object_exists()).
1931 static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags)
1933 const struct mdt_body *body;
1934 struct mdt_object *obj;
1935 const struct lu_env *env;
1936 struct req_capsule *pill;
1940 env = info->mti_env;
1941 pill = info->mti_pill;
1943 body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
1947 if (!(body->valid & OBD_MD_FLID))
1950 if (!fid_is_sane(&body->fid1)) {
1951 CERROR("Invalid fid: "DFID"\n", PFID(&body->fid1));
1956 * Do not get size or any capa fields before we check that request
1957 * contains capa actually. There are some requests which do not, for
1958 * instance MDS_IS_SUBDIR.
1960 if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_CLIENT) &&
1961 req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
1962 mdt_set_capainfo(info, 0, &body->fid1,
1963 req_capsule_client_get(pill, &RMF_CAPA1));
1965 obj = mdt_object_find(env, info->mti_mdt, &body->fid1);
1967 if ((flags & HABEO_CORPUS) &&
1968 !mdt_object_exists(obj)) {
1969 mdt_object_put(env, obj);
1970 /* for capability renew ENOENT will be handled in
1972 if (body->valid & OBD_MD_FLOSSCAPA)
1977 info->mti_object = obj;
1986 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags)
1988 struct req_capsule *pill = info->mti_pill;
1992 if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
1993 rc = mdt_body_unpack(info, flags);
1997 if (rc == 0 && (flags & HABEO_REFERO)) {
1998 struct mdt_device *mdt = info->mti_mdt;
2001 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2002 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2003 mdt->mdt_max_mdsize);
2004 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2005 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER,
2006 mdt->mdt_max_cookiesize);
2008 rc = req_capsule_server_pack(pill);
2013 static int mdt_init_capa_ctxt(const struct lu_env *env, struct mdt_device *m)
2015 struct md_device *next = m->mdt_child;
2017 return next->md_ops->mdo_init_capa_ctxt(env, next,
2018 m->mdt_opts.mo_mds_capa,
2019 m->mdt_capa_timeout,
2025 * Invoke handler for this request opc. Also do necessary preprocessing
2026 * (according to handler ->mh_flags), and post-processing (setting of
2027 * ->last_{xid,committed}).
2029 static int mdt_req_handle(struct mdt_thread_info *info,
2030 struct mdt_handler *h, struct ptlrpc_request *req)
2032 int rc, serious = 0;
2037 LASSERT(h->mh_act != NULL);
2038 LASSERT(h->mh_opc == lustre_msg_get_opc(req->rq_reqmsg));
2039 LASSERT(current->journal_info == NULL);
2042 * Checking for various OBD_FAIL_$PREF_$OPC_NET codes. _Do_ not try
2043 * to put same checks into handlers like mdt_close(), mdt_reint(),
2044 * etc., without talking to mdt authors first. Checking same thing
2045 * there again is useless and returning 0 error wihtout packing reply
2046 * is buggy! Handlers either pack reply or return error.
2048 * We return 0 here and do not send any reply in order to emulate
2049 * network failure. Do not send any reply in case any of NET related
2050 * fail_id has occured.
2052 if (OBD_FAIL_CHECK_ORSET(h->mh_fail_id, OBD_FAIL_ONCE))
2056 flags = h->mh_flags;
2057 LASSERT(ergo(flags & (HABEO_CORPUS|HABEO_REFERO), h->mh_fmt != NULL));
2059 if (h->mh_fmt != NULL) {
2060 req_capsule_set(info->mti_pill, h->mh_fmt);
2061 rc = mdt_unpack_req_pack_rep(info, flags);
2064 if (rc == 0 && flags & MUTABOR &&
2065 req->rq_export->exp_connect_flags & OBD_CONNECT_RDONLY)
2066 /* should it be rq_status? */
2069 if (rc == 0 && flags & HABEO_CLAVIS) {
2070 struct ldlm_request *dlm_req;
2072 LASSERT(h->mh_fmt != NULL);
2074 dlm_req = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
2075 if (dlm_req != NULL) {
2076 if (info->mti_mdt->mdt_opts.mo_compat_resname)
2077 rc = mdt_lock_resname_compat(info->mti_mdt,
2079 info->mti_dlm_req = dlm_req;
2085 /* capability setting changed via /proc, needs reinitialize ctxt */
2086 if (info->mti_mdt && info->mti_mdt->mdt_capa_conf) {
2087 mdt_init_capa_ctxt(info->mti_env, info->mti_mdt);
2088 info->mti_mdt->mdt_capa_conf = 0;
2091 if (likely(rc == 0)) {
2093 * Process request, there can be two types of rc:
2094 * 1) errors with msg unpack/pack, other failures outside the
2095 * operation itself. This is counted as serious errors;
2096 * 2) errors during fs operation, should be placed in rq_status
2099 rc = h->mh_act(info);
2100 if (rc == 0 && req->rq_reply_state == NULL) {
2101 DEBUG_REQ(D_ERROR, req, "MDT \"handler\" %s did not pack "
2102 "reply and returned 0 error\n", h->mh_name);
2105 serious = is_serious(rc);
2106 rc = clear_serious(rc);
2110 req->rq_status = rc;
2113 * ELDLM_* codes which > 0 should be in rq_status only as well as
2114 * all non-serious errors.
2116 if (rc > 0 || !serious)
2119 LASSERT(current->journal_info == NULL);
2121 if (rc == 0 && (flags & HABEO_CLAVIS) &&
2122 info->mti_mdt->mdt_opts.mo_compat_resname) {
2123 struct ldlm_reply *dlmrep;
2125 dlmrep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
2127 rc = mdt_lock_reply_compat(info->mti_mdt, dlmrep);
2130 /* If we're DISCONNECTing, the mdt_export_data is already freed */
2131 if (likely(rc == 0 && h->mh_opc != MDS_DISCONNECT))
2132 target_committed_to_req(req);
2134 if (unlikely((lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) &&
2135 lustre_msg_get_transno(req->rq_reqmsg) == 0)) {
2136 DEBUG_REQ(D_ERROR, req, "transno is 0 during REPLAY");
2140 target_send_reply(req, rc, info->mti_fail_id);
2144 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
2146 lh->mlh_type = MDT_NUL_LOCK;
2147 lh->mlh_reg_lh.cookie = 0ull;
2148 lh->mlh_reg_mode = LCK_MINMODE;
2149 lh->mlh_pdo_lh.cookie = 0ull;
2150 lh->mlh_pdo_mode = LCK_MINMODE;
2153 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
2155 LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2156 LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2160 * Initialize fields of struct mdt_thread_info. Other fields are left in
2161 * uninitialized state, because it's too expensive to zero out whole
2162 * mdt_thread_info (> 1K) on each request arrival.
2164 static void mdt_thread_info_init(struct ptlrpc_request *req,
2165 struct mdt_thread_info *info)
2168 struct md_capainfo *ci;
2170 req_capsule_init(&req->rq_pill, req, RCL_SERVER);
2171 info->mti_pill = &req->rq_pill;
2174 for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2175 mdt_lock_handle_init(&info->mti_lh[i]);
2177 /* mdt device: it can be NULL while CONNECT */
2178 if (req->rq_export) {
2179 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
2180 info->mti_exp = req->rq_export;
2182 info->mti_mdt = NULL;
2183 info->mti_env = req->rq_svc_thread->t_env;
2184 ci = md_capainfo(info->mti_env);
2185 memset(ci, 0, sizeof *ci);
2187 info->mti_fail_id = OBD_FAIL_MDS_ALL_REPLY_NET;
2188 info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
2190 memset(&info->mti_attr, 0, sizeof(info->mti_attr));
2191 info->mti_body = NULL;
2192 info->mti_object = NULL;
2193 info->mti_dlm_req = NULL;
2194 info->mti_has_trans = 0;
2195 info->mti_no_need_trans = 0;
2196 info->mti_cross_ref = 0;
2197 info->mti_opdata = 0;
2199 /* To not check for split by default. */
2200 info->mti_spec.sp_ck_split = 0;
2203 static void mdt_thread_info_fini(struct mdt_thread_info *info)
2207 req_capsule_fini(info->mti_pill);
2208 if (info->mti_object != NULL) {
2209 mdt_object_put(info->mti_env, info->mti_object);
2210 info->mti_object = NULL;
2212 for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2213 mdt_lock_handle_fini(&info->mti_lh[i]);
2214 info->mti_env = NULL;
2217 static int mdt_filter_recovery_request(struct ptlrpc_request *req,
2218 struct obd_device *obd, int *process)
2220 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2221 case MDS_CONNECT: /* This will never get here, but for completeness. */
2222 case OST_CONNECT: /* This will never get here, but for completeness. */
2223 case MDS_DISCONNECT:
2224 case OST_DISCONNECT:
2229 case MDS_DONE_WRITING:
2230 case MDS_SYNC: /* used in unmounting */
2236 *process = target_queue_recovery_request(req, obd);
2240 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
2247 * Handle recovery. Return:
2248 * +1: continue request processing;
2249 * -ve: abort immediately with the given error code;
2250 * 0: send reply with error code in req->rq_status;
2252 static int mdt_recovery(struct mdt_thread_info *info)
2254 struct ptlrpc_request *req = mdt_info_req(info);
2256 struct obd_device *obd;
2260 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2263 case SEC_CTX_INIT_CONT:
2269 rc = mdt_handle_idmap(info);
2278 if (unlikely(req->rq_export == NULL)) {
2279 CERROR("operation %d on unconnected MDS from %s\n",
2280 lustre_msg_get_opc(req->rq_reqmsg),
2281 libcfs_id2str(req->rq_peer));
2282 /* FIXME: For CMD cleanup, when mds_B stop, the req from
2283 * mds_A will get -ENOTCONN(especially for ping req),
2284 * which will cause that mds_A deactive timeout, then when
2285 * mds_A cleanup, the cleanup process will be suspended since
2286 * deactive timeout is not zero.
2288 req->rq_status = -ENOTCONN;
2289 target_send_reply(req, -ENOTCONN, info->mti_fail_id);
2293 /* sanity check: if the xid matches, the request must be marked as a
2294 * resent or replayed */
2295 if (req_xid_is_last(req)) {
2296 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
2297 (MSG_RESENT | MSG_REPLAY))) {
2298 DEBUG_REQ(D_WARNING, req, "rq_xid "LPU64" matches last_xid, "
2299 "expected REPLAY or RESENT flag (%x)", req->rq_xid,
2300 lustre_msg_get_flags(req->rq_reqmsg));
2302 req->rq_status = -ENOTCONN;
2307 /* else: note the opposite is not always true; a RESENT req after a
2308 * failover will usually not match the last_xid, since it was likely
2309 * never committed. A REPLAYed request will almost never match the
2310 * last xid, however it could for a committed, but still retained,
2313 obd = req->rq_export->exp_obd;
2315 /* Check for aborted recovery... */
2316 spin_lock_bh(&obd->obd_processing_task_lock);
2317 recovering = obd->obd_recovering;
2318 spin_unlock_bh(&obd->obd_processing_task_lock);
2319 if (unlikely(recovering)) {
2322 DEBUG_REQ(D_INFO, req, "Got new replay");
2323 rc = mdt_filter_recovery_request(req, obd, &should_process);
2324 if (rc != 0 || !should_process)
2326 else if (should_process < 0) {
2327 req->rq_status = should_process;
2328 rc = ptlrpc_error(req);
2335 static int mdt_msg_check_version(struct lustre_msg *msg)
2339 switch (lustre_msg_get_opc(msg)) {
2341 case MDS_DISCONNECT:
2344 case SEC_CTX_INIT_CONT:
2346 rc = lustre_msg_check_version(msg, LUSTRE_OBD_VERSION);
2348 CERROR("bad opc %u version %08x, expecting %08x\n",
2349 lustre_msg_get_opc(msg),
2350 lustre_msg_get_version(msg),
2351 LUSTRE_OBD_VERSION);
2355 case MDS_GETATTR_NAME:
2362 case MDS_DONE_WRITING:
2368 case MDS_QUOTACHECK:
2374 rc = lustre_msg_check_version(msg, LUSTRE_MDS_VERSION);
2376 CERROR("bad opc %u version %08x, expecting %08x\n",
2377 lustre_msg_get_opc(msg),
2378 lustre_msg_get_version(msg),
2379 LUSTRE_MDS_VERSION);
2383 case LDLM_BL_CALLBACK:
2384 case LDLM_CP_CALLBACK:
2385 rc = lustre_msg_check_version(msg, LUSTRE_DLM_VERSION);
2387 CERROR("bad opc %u version %08x, expecting %08x\n",
2388 lustre_msg_get_opc(msg),
2389 lustre_msg_get_version(msg),
2390 LUSTRE_DLM_VERSION);
2392 case OBD_LOG_CANCEL:
2393 case LLOG_ORIGIN_HANDLE_CREATE:
2394 case LLOG_ORIGIN_HANDLE_NEXT_BLOCK:
2395 case LLOG_ORIGIN_HANDLE_READ_HEADER:
2396 case LLOG_ORIGIN_HANDLE_CLOSE:
2397 case LLOG_ORIGIN_HANDLE_DESTROY:
2398 case LLOG_ORIGIN_HANDLE_PREV_BLOCK:
2400 rc = lustre_msg_check_version(msg, LUSTRE_LOG_VERSION);
2402 CERROR("bad opc %u version %08x, expecting %08x\n",
2403 lustre_msg_get_opc(msg),
2404 lustre_msg_get_version(msg),
2405 LUSTRE_LOG_VERSION);
2408 CERROR("MDS unknown opcode %d\n", lustre_msg_get_opc(msg));
2414 static int mdt_handle0(struct ptlrpc_request *req,
2415 struct mdt_thread_info *info,
2416 struct mdt_opc_slice *supported)
2418 struct mdt_handler *h;
2419 struct lustre_msg *msg;
2424 if (OBD_FAIL_CHECK_ORSET(OBD_FAIL_MDS_ALL_REQUEST_NET, OBD_FAIL_ONCE))
2427 LASSERT(current->journal_info == NULL);
2429 msg = req->rq_reqmsg;
2430 rc = mdt_msg_check_version(msg);
2431 if (likely(rc == 0)) {
2432 rc = mdt_recovery(info);
2433 if (likely(rc == +1)) {
2434 h = mdt_handler_find(lustre_msg_get_opc(msg),
2436 if (likely(h != NULL)) {
2437 rc = mdt_req_handle(info, h, req);
2439 CERROR("The unsupported opc: 0x%x\n", lustre_msg_get_opc(msg) );
2440 req->rq_status = -ENOTSUPP;
2441 rc = ptlrpc_error(req);
2446 CERROR(LUSTRE_MDT_NAME" drops mal-formed request\n");
2451 * MDT handler function called by ptlrpc service thread when request comes.
2453 * XXX common "target" functionality should be factored into separate module
2454 * shared by mdt, ost and stand-alone services like fld.
2456 static int mdt_handle_common(struct ptlrpc_request *req,
2457 struct mdt_opc_slice *supported)
2460 struct mdt_thread_info *info;
2464 env = req->rq_svc_thread->t_env;
2465 LASSERT(env != NULL);
2466 LASSERT(env->le_ses != NULL);
2467 LASSERT(env->le_ctx.lc_thread == req->rq_svc_thread);
2468 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
2469 LASSERT(info != NULL);
2471 mdt_thread_info_init(req, info);
2473 rc = mdt_handle0(req, info, supported);
2475 mdt_thread_info_fini(info);
2480 * This is called from recovery code as handler of _all_ RPC types, FLD and SEQ
2483 int mdt_recovery_handle(struct ptlrpc_request *req)
2488 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2490 rc = mdt_handle_common(req, mdt_fld_handlers);
2493 rc = mdt_handle_common(req, mdt_seq_handlers);
2496 rc = mdt_handle_common(req, mdt_regular_handlers);
2503 static int mdt_regular_handle(struct ptlrpc_request *req)
2505 return mdt_handle_common(req, mdt_regular_handlers);
2508 static int mdt_readpage_handle(struct ptlrpc_request *req)
2510 return mdt_handle_common(req, mdt_readpage_handlers);
2513 static int mdt_xmds_handle(struct ptlrpc_request *req)
2515 return mdt_handle_common(req, mdt_xmds_handlers);
2518 static int mdt_mdsc_handle(struct ptlrpc_request *req)
2520 return mdt_handle_common(req, mdt_seq_handlers);
2523 static int mdt_mdss_handle(struct ptlrpc_request *req)
2525 return mdt_handle_common(req, mdt_seq_handlers);
2528 static int mdt_dtss_handle(struct ptlrpc_request *req)
2530 return mdt_handle_common(req, mdt_seq_handlers);
2533 static int mdt_fld_handle(struct ptlrpc_request *req)
2535 return mdt_handle_common(req, mdt_fld_handlers);
2551 static int mdt_intent_getattr(enum mdt_it_code opcode,
2552 struct mdt_thread_info *info,
2553 struct ldlm_lock **,
2555 static int mdt_intent_reint(enum mdt_it_code opcode,
2556 struct mdt_thread_info *info,
2557 struct ldlm_lock **,
2560 static struct mdt_it_flavor {
2561 const struct req_format *it_fmt;
2563 int (*it_act)(enum mdt_it_code ,
2564 struct mdt_thread_info *,
2565 struct ldlm_lock **,
2568 } mdt_it_flavor[] = {
2570 .it_fmt = &RQF_LDLM_INTENT,
2571 /*.it_flags = HABEO_REFERO,*/
2573 .it_act = mdt_intent_reint,
2574 .it_reint = REINT_OPEN
2577 .it_fmt = &RQF_LDLM_INTENT,
2578 .it_flags = MUTABOR,
2579 .it_act = mdt_intent_reint,
2580 .it_reint = REINT_OPEN
2583 .it_fmt = &RQF_LDLM_INTENT,
2584 .it_flags = MUTABOR,
2585 .it_act = mdt_intent_reint,
2586 .it_reint = REINT_CREATE
2588 [MDT_IT_GETATTR] = {
2589 .it_fmt = &RQF_LDLM_INTENT_GETATTR,
2590 .it_flags = HABEO_REFERO,
2591 .it_act = mdt_intent_getattr
2593 [MDT_IT_READDIR] = {
2599 .it_fmt = &RQF_LDLM_INTENT_GETATTR,
2600 .it_flags = HABEO_REFERO,
2601 .it_act = mdt_intent_getattr
2604 .it_fmt = &RQF_LDLM_INTENT_UNLINK,
2605 .it_flags = MUTABOR,
2607 .it_reint = REINT_UNLINK
2611 .it_flags = MUTABOR,
2614 [MDT_IT_GETXATTR] = {
2621 int mdt_intent_lock_replace(struct mdt_thread_info *info,
2622 struct ldlm_lock **lockp,
2623 struct ldlm_lock *new_lock,
2624 struct mdt_lock_handle *lh,
2627 struct ptlrpc_request *req = mdt_info_req(info);
2628 struct ldlm_lock *lock = *lockp;
2631 * Get new lock only for cases when possible resent did not find any
2634 if (new_lock == NULL)
2635 new_lock = ldlm_handle2lock(&lh->mlh_reg_lh);
2637 if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
2638 lh->mlh_reg_lh.cookie = 0;
2642 LASSERTF(new_lock != NULL,
2643 "lockh "LPX64"\n", lh->mlh_reg_lh.cookie);
2646 * If we've already given this lock to a client once, then we should
2647 * have no readers or writers. Otherwise, we should have one reader
2648 * _or_ writer ref (which will be zeroed below) before returning the
2651 if (new_lock->l_export == req->rq_export) {
2652 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
2654 LASSERT(new_lock->l_export == NULL);
2655 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
2660 if (new_lock->l_export == req->rq_export) {
2662 * Already gave this to the client, which means that we
2663 * reconstructed a reply.
2665 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
2667 lh->mlh_reg_lh.cookie = 0;
2668 RETURN(ELDLM_LOCK_REPLACED);
2671 /* This lock might already be given to the client by an resent req,
2672 * in this case we should return ELDLM_LOCK_ABORTED,
2673 * so we should check led_held_locks here, but it will affect
2674 * performance, FIXME
2676 /* Fixup the lock to be given to the client */
2677 lock_res_and_lock(new_lock);
2678 new_lock->l_readers = 0;
2679 new_lock->l_writers = 0;
2681 new_lock->l_export = class_export_get(req->rq_export);
2682 spin_lock(&req->rq_export->exp_ldlm_data.led_lock);
2683 list_add(&new_lock->l_export_chain,
2684 &new_lock->l_export->exp_ldlm_data.led_held_locks);
2685 spin_unlock(&req->rq_export->exp_ldlm_data.led_lock);
2687 new_lock->l_blocking_ast = lock->l_blocking_ast;
2688 new_lock->l_completion_ast = lock->l_completion_ast;
2689 new_lock->l_remote_handle = lock->l_remote_handle;
2690 new_lock->l_flags &= ~LDLM_FL_LOCAL;
2692 unlock_res_and_lock(new_lock);
2693 LDLM_LOCK_PUT(new_lock);
2694 lh->mlh_reg_lh.cookie = 0;
2696 RETURN(ELDLM_LOCK_REPLACED);
2699 static void mdt_intent_fixup_resent(struct mdt_thread_info *info,
2700 struct ldlm_lock *new_lock,
2701 struct ldlm_lock **old_lock,
2702 struct mdt_lock_handle *lh)
2704 struct ptlrpc_request *req = mdt_info_req(info);
2705 struct obd_export *exp = req->rq_export;
2706 struct lustre_handle remote_hdl;
2707 struct ldlm_request *dlmreq;
2708 struct list_head *iter;
2710 if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
2713 dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
2714 remote_hdl = dlmreq->lock_handle[0];
2716 spin_lock(&exp->exp_ldlm_data.led_lock);
2717 list_for_each(iter, &exp->exp_ldlm_data.led_held_locks) {
2718 struct ldlm_lock *lock;
2719 lock = list_entry(iter, struct ldlm_lock, l_export_chain);
2720 if (lock == new_lock)
2722 if (lock->l_remote_handle.cookie == remote_hdl.cookie) {
2723 lh->mlh_reg_lh.cookie = lock->l_handle.h_cookie;
2724 lh->mlh_reg_mode = lock->l_granted_mode;
2726 LDLM_DEBUG(lock, "restoring lock cookie");
2727 DEBUG_REQ(D_DLMTRACE, req,
2728 "restoring lock cookie "LPX64,
2729 lh->mlh_reg_lh.cookie);
2731 *old_lock = LDLM_LOCK_GET(lock);
2732 spin_unlock(&exp->exp_ldlm_data.led_lock);
2736 spin_unlock(&exp->exp_ldlm_data.led_lock);
2739 * If the xid matches, then we know this is a resent request, and allow
2740 * it. (It's probably an OPEN, for which we don't send a lock.
2742 if (req_xid_is_last(req))
2746 * This remote handle isn't enqueued, so we never received or processed
2747 * this request. Clear MSG_RESENT, because it can be handled like any
2748 * normal request now.
2750 lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
2752 DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle "LPX64,
2756 static int mdt_intent_getattr(enum mdt_it_code opcode,
2757 struct mdt_thread_info *info,
2758 struct ldlm_lock **lockp,
2761 struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
2762 struct ldlm_lock *new_lock = NULL;
2764 struct ldlm_reply *ldlm_rep;
2765 struct ptlrpc_request *req;
2766 struct mdt_body *reqbody;
2767 struct mdt_body *repbody;
2771 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
2774 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
2777 info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
2778 info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
2779 repbody->eadatasize = 0;
2780 repbody->aclsize = 0;
2784 child_bits = MDS_INODELOCK_LOOKUP;
2786 case MDT_IT_GETATTR:
2787 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE;
2790 CERROR("Unhandled till now");
2791 GOTO(out_shrink, rc = -EINVAL);
2794 rc = mdt_init_ucred(info, reqbody);
2796 GOTO(out_shrink, rc);
2798 req = info->mti_pill->rc_req;
2799 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
2800 mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
2802 /* Get lock from request for possible resent case. */
2803 mdt_intent_fixup_resent(info, *lockp, &new_lock, lhc);
2805 ldlm_rep->lock_policy_res2 =
2806 mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
2808 if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
2809 ldlm_rep->lock_policy_res2 = 0;
2810 if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
2811 ldlm_rep->lock_policy_res2) {
2812 lhc->mlh_reg_lh.cookie = 0ull;
2813 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
2816 rc = mdt_intent_lock_replace(info, lockp, new_lock, lhc, flags);
2819 mdt_exit_ucred(info);
2821 mdt_shrink_reply(info);
2825 static int mdt_intent_reint(enum mdt_it_code opcode,
2826 struct mdt_thread_info *info,
2827 struct ldlm_lock **lockp,
2830 struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
2831 struct ldlm_reply *rep = NULL;
2835 static const struct req_format *intent_fmts[REINT_MAX] = {
2836 [REINT_CREATE] = &RQF_LDLM_INTENT_CREATE,
2837 [REINT_OPEN] = &RQF_LDLM_INTENT_OPEN
2842 opc = mdt_reint_opcode(info, intent_fmts);
2846 if (mdt_it_flavor[opcode].it_reint != opc) {
2847 CERROR("Reint code %ld doesn't match intent: %d\n",
2849 RETURN(err_serious(-EPROTO));
2852 /* Get lock from request for possible resent case. */
2853 mdt_intent_fixup_resent(info, *lockp, NULL, lhc);
2855 rc = mdt_reint_internal(info, lhc, opc);
2857 /* Check whether the reply has been packed successfully. */
2858 if (mdt_info_req(info)->rq_repmsg != NULL)
2859 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
2861 RETURN(err_serious(-EFAULT));
2863 /* MDC expects this in any case */
2865 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
2867 /* Cross-ref case, the lock should be returned to the client */
2868 if (rc == -EREMOTE) {
2869 LASSERT(lustre_handle_is_used(&lhc->mlh_reg_lh));
2870 rep->lock_policy_res2 = 0;
2871 rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
2874 rep->lock_policy_res2 = clear_serious(rc);
2876 lhc->mlh_reg_lh.cookie = 0ull;
2877 rc = ELDLM_LOCK_ABORTED;
2881 static int mdt_intent_code(long itcode)
2889 case IT_OPEN|IT_CREAT:
2896 rc = MDT_IT_READDIR;
2899 rc = MDT_IT_GETATTR;
2911 rc = MDT_IT_GETXATTR;
2914 CERROR("Unknown intent opcode: %ld\n", itcode);
2921 static int mdt_intent_opc(long itopc, struct mdt_thread_info *info,
2922 struct ldlm_lock **lockp, int flags)
2924 struct req_capsule *pill;
2925 struct mdt_it_flavor *flv;
2930 opc = mdt_intent_code(itopc);
2934 pill = info->mti_pill;
2935 flv = &mdt_it_flavor[opc];
2937 if (flv->it_fmt != NULL)
2938 req_capsule_extend(pill, flv->it_fmt);
2940 rc = mdt_unpack_req_pack_rep(info, flv->it_flags);
2942 struct ptlrpc_request *req = mdt_info_req(info);
2943 if (flv->it_flags & MUTABOR &&
2944 req->rq_export->exp_connect_flags & OBD_CONNECT_RDONLY)
2947 if (rc == 0 && flv->it_act != NULL) {
2948 /* execute policy */
2949 rc = flv->it_act(opc, info, lockp, flags);
2956 static int mdt_intent_policy(struct ldlm_namespace *ns,
2957 struct ldlm_lock **lockp, void *req_cookie,
2958 ldlm_mode_t mode, int flags, void *data)
2960 struct mdt_thread_info *info;
2961 struct ptlrpc_request *req = req_cookie;
2962 struct ldlm_intent *it;
2963 struct req_capsule *pill;
2968 LASSERT(req != NULL);
2970 info = lu_context_key_get(&req->rq_svc_thread->t_env->le_ctx,
2972 LASSERT(info != NULL);
2973 pill = info->mti_pill;
2974 LASSERT(pill->rc_req == req);
2976 if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
2977 req_capsule_extend(pill, &RQF_LDLM_INTENT);
2978 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
2980 const struct ldlm_request *dlmreq;
2983 struct ldlm_lock *lock = *lockp;
2985 LDLM_DEBUG(lock, "intent policy opc: %s\n",
2986 ldlm_it2str(it->opc));
2989 rc = mdt_intent_opc(it->opc, info, lockp, flags);
2994 * Lock without inodebits makes no sense and will oops
2995 * later in ldlm. Let's check it now to see if we have
2996 * wrong lock from client or bits get corrupted
2997 * somewhere in mdt_intent_opc().
2999 dlmreq = info->mti_dlm_req;
3000 req_bits = dlmreq->lock_desc.l_policy_data.l_inodebits.bits;
3001 LASSERT(req_bits != 0);
3004 rc = err_serious(-EFAULT);
3006 /* No intent was provided */
3007 LASSERT(pill->rc_fmt == &RQF_LDLM_ENQUEUE);
3008 rc = req_capsule_server_pack(pill);
3010 rc = err_serious(rc);
3018 static void mdt_seq_adjust(const struct lu_env *env,
3019 struct mdt_device *m, int lost)
3021 struct lu_site *ls = m->mdt_md_dev.md_lu_dev.ld_site;
3022 struct lu_range out;
3025 LASSERT(ls && ls->ls_server_seq);
3027 /* get extra seq from seq_server, moving it's range up */
3028 while (lost-- > 0) {
3029 seq_server_alloc_meta(ls->ls_server_seq, NULL, &out, env);
3034 static int mdt_seq_fini(const struct lu_env *env,
3035 struct mdt_device *m)
3037 struct lu_site *ls = m->mdt_md_dev.md_lu_dev.ld_site;
3040 if (ls && ls->ls_server_seq) {
3041 seq_server_fini(ls->ls_server_seq, env);
3042 OBD_FREE_PTR(ls->ls_server_seq);
3043 ls->ls_server_seq = NULL;
3046 if (ls && ls->ls_control_seq) {
3047 seq_server_fini(ls->ls_control_seq, env);
3048 OBD_FREE_PTR(ls->ls_control_seq);
3049 ls->ls_control_seq = NULL;
3052 if (ls && ls->ls_client_seq) {
3053 seq_client_fini(ls->ls_client_seq);
3054 OBD_FREE_PTR(ls->ls_client_seq);
3055 ls->ls_client_seq = NULL;
3061 static int mdt_seq_init(const struct lu_env *env,
3063 struct mdt_device *m)
3070 ls = m->mdt_md_dev.md_lu_dev.ld_site;
3073 * This is sequence-controller node. Init seq-controller server on local
3076 if (ls->ls_node_id == 0) {
3077 LASSERT(ls->ls_control_seq == NULL);
3079 OBD_ALLOC_PTR(ls->ls_control_seq);
3080 if (ls->ls_control_seq == NULL)
3083 rc = seq_server_init(ls->ls_control_seq,
3084 m->mdt_bottom, uuid,
3085 LUSTRE_SEQ_CONTROLLER,
3089 GOTO(out_seq_fini, rc);
3091 OBD_ALLOC_PTR(ls->ls_client_seq);
3092 if (ls->ls_client_seq == NULL)
3093 GOTO(out_seq_fini, rc = -ENOMEM);
3095 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3096 if (prefix == NULL) {
3097 OBD_FREE_PTR(ls->ls_client_seq);
3098 GOTO(out_seq_fini, rc = -ENOMEM);
3101 snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s",
3105 * Init seq-controller client after seq-controller server is
3106 * ready. Pass ls->ls_control_seq to it for direct talking.
3108 rc = seq_client_init(ls->ls_client_seq, NULL,
3109 LUSTRE_SEQ_METADATA, prefix,
3110 ls->ls_control_seq);
3111 OBD_FREE(prefix, MAX_OBD_NAME + 5);
3114 GOTO(out_seq_fini, rc);
3117 /* Init seq-server on local MDT */
3118 LASSERT(ls->ls_server_seq == NULL);
3120 OBD_ALLOC_PTR(ls->ls_server_seq);
3121 if (ls->ls_server_seq == NULL)
3122 GOTO(out_seq_fini, rc = -ENOMEM);
3124 rc = seq_server_init(ls->ls_server_seq,
3125 m->mdt_bottom, uuid,
3129 GOTO(out_seq_fini, rc = -ENOMEM);
3131 /* Assign seq-controller client to local seq-server. */
3132 if (ls->ls_node_id == 0) {
3133 LASSERT(ls->ls_client_seq != NULL);
3135 rc = seq_server_set_cli(ls->ls_server_seq,
3143 mdt_seq_fini(env, m);
3148 * Init client sequence manager which is used by local MDS to talk to sequence
3149 * controller on remote node.
3151 static int mdt_seq_init_cli(const struct lu_env *env,
3152 struct mdt_device *m,
3153 struct lustre_cfg *cfg)
3155 struct lu_site *ls = m->mdt_md_dev.md_lu_dev.ld_site;
3156 struct obd_device *mdc;
3157 struct obd_uuid *uuidp, *mdcuuidp;
3158 char *uuid_str, *mdc_uuid_str;
3161 struct mdt_thread_info *info;
3162 char *p, *index_string = lustre_cfg_string(cfg, 2);
3165 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3166 uuidp = &info->mti_u.uuid[0];
3167 mdcuuidp = &info->mti_u.uuid[1];
3169 LASSERT(index_string);
3171 index = simple_strtol(index_string, &p, 10);
3173 CERROR("Invalid index in lustre_cgf, offset 2\n");
3177 /* check if this is adding the first MDC and controller is not yet
3179 if (index != 0 || ls->ls_client_seq)
3182 uuid_str = lustre_cfg_string(cfg, 1);
3183 mdc_uuid_str = lustre_cfg_string(cfg, 4);
3184 obd_str2uuid(uuidp, uuid_str);
3185 obd_str2uuid(mdcuuidp, mdc_uuid_str);
3187 mdc = class_find_client_obd(uuidp, LUSTRE_MDC_NAME, mdcuuidp);
3189 CERROR("can't find controller MDC by uuid %s\n",
3192 } else if (!mdc->obd_set_up) {
3193 CERROR("target %s not set up\n", mdc->obd_name);
3196 LASSERT(ls->ls_control_exp);
3197 OBD_ALLOC_PTR(ls->ls_client_seq);
3198 if (ls->ls_client_seq != NULL) {
3201 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3205 snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s",
3208 rc = seq_client_init(ls->ls_client_seq,
3210 LUSTRE_SEQ_METADATA,
3212 OBD_FREE(prefix, MAX_OBD_NAME + 5);
3219 LASSERT(ls->ls_server_seq != NULL);
3220 rc = seq_server_set_cli(ls->ls_server_seq, ls->ls_client_seq,
3227 static void mdt_seq_fini_cli(struct mdt_device *m)
3233 ls = m->mdt_md_dev.md_lu_dev.ld_site;
3235 if (ls && ls->ls_server_seq)
3236 seq_server_set_cli(ls->ls_server_seq,
3239 if (ls && ls->ls_control_exp) {
3240 class_export_put(ls->ls_control_exp);
3241 ls->ls_control_exp = NULL;
3249 static int mdt_fld_fini(const struct lu_env *env,
3250 struct mdt_device *m)
3252 struct lu_site *ls = m->mdt_md_dev.md_lu_dev.ld_site;
3255 if (ls && ls->ls_server_fld) {
3256 fld_server_fini(ls->ls_server_fld, env);
3257 OBD_FREE_PTR(ls->ls_server_fld);
3258 ls->ls_server_fld = NULL;
3264 static int mdt_fld_init(const struct lu_env *env,
3266 struct mdt_device *m)
3272 ls = m->mdt_md_dev.md_lu_dev.ld_site;
3274 OBD_ALLOC_PTR(ls->ls_server_fld);
3275 if (ls->ls_server_fld == NULL)
3276 RETURN(rc = -ENOMEM);
3278 rc = fld_server_init(ls->ls_server_fld,
3279 m->mdt_bottom, uuid, env);
3281 OBD_FREE_PTR(ls->ls_server_fld);
3282 ls->ls_server_fld = NULL;
3289 /* device init/fini methods */
3290 static void mdt_stop_ptlrpc_service(struct mdt_device *m)
3293 if (m->mdt_regular_service != NULL) {
3294 ptlrpc_unregister_service(m->mdt_regular_service);
3295 m->mdt_regular_service = NULL;
3297 if (m->mdt_readpage_service != NULL) {
3298 ptlrpc_unregister_service(m->mdt_readpage_service);
3299 m->mdt_readpage_service = NULL;
3301 if (m->mdt_xmds_service != NULL) {
3302 ptlrpc_unregister_service(m->mdt_xmds_service);
3303 m->mdt_xmds_service = NULL;
3305 if (m->mdt_setattr_service != NULL) {
3306 ptlrpc_unregister_service(m->mdt_setattr_service);
3307 m->mdt_setattr_service = NULL;
3309 if (m->mdt_mdsc_service != NULL) {
3310 ptlrpc_unregister_service(m->mdt_mdsc_service);
3311 m->mdt_mdsc_service = NULL;
3313 if (m->mdt_mdss_service != NULL) {
3314 ptlrpc_unregister_service(m->mdt_mdss_service);
3315 m->mdt_mdss_service = NULL;
3317 if (m->mdt_dtss_service != NULL) {
3318 ptlrpc_unregister_service(m->mdt_dtss_service);
3319 m->mdt_dtss_service = NULL;
3321 if (m->mdt_fld_service != NULL) {
3322 ptlrpc_unregister_service(m->mdt_fld_service);
3323 m->mdt_fld_service = NULL;
3328 static int mdt_start_ptlrpc_service(struct mdt_device *m)
3331 static struct ptlrpc_service_conf conf;
3332 cfs_proc_dir_entry_t *procfs_entry;
3335 procfs_entry = m->mdt_md_dev.md_lu_dev.ld_obd->obd_proc_entry;
3337 conf = (typeof(conf)) {
3338 .psc_nbufs = MDS_NBUFS,
3339 .psc_bufsize = MDS_BUFSIZE,
3340 .psc_max_req_size = MDS_MAXREQSIZE,
3341 .psc_max_reply_size = MDS_MAXREPSIZE,
3342 .psc_req_portal = MDS_REQUEST_PORTAL,
3343 .psc_rep_portal = MDC_REPLY_PORTAL,
3344 .psc_watchdog_timeout = MDT_SERVICE_WATCHDOG_TIMEOUT,
3346 * We'd like to have a mechanism to set this on a per-device
3347 * basis, but alas...
3349 .psc_min_threads = min(max(mdt_num_threads, MDT_MIN_THREADS),
3351 .psc_max_threads = MDT_MAX_THREADS,
3352 .psc_ctx_tags = LCT_MD_THREAD
3355 m->mdt_ldlm_client = &m->mdt_md_dev.md_lu_dev.ld_obd->obd_ldlm_client;
3356 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
3357 "mdt_ldlm_client", m->mdt_ldlm_client);
3359 m->mdt_regular_service =
3360 ptlrpc_init_svc_conf(&conf, mdt_regular_handle, LUSTRE_MDT_NAME,
3361 procfs_entry, NULL, LUSTRE_MDT_NAME);
3362 if (m->mdt_regular_service == NULL)
3365 rc = ptlrpc_start_threads(NULL, m->mdt_regular_service);
3367 GOTO(err_mdt_svc, rc);
3370 * readpage service configuration. Parameters have to be adjusted,
3373 conf = (typeof(conf)) {
3374 .psc_nbufs = MDS_NBUFS,
3375 .psc_bufsize = MDS_BUFSIZE,
3376 .psc_max_req_size = MDS_MAXREQSIZE,
3377 .psc_max_reply_size = MDS_MAXREPSIZE,
3378 .psc_req_portal = MDS_READPAGE_PORTAL,
3379 .psc_rep_portal = MDC_REPLY_PORTAL,
3380 .psc_watchdog_timeout = MDT_SERVICE_WATCHDOG_TIMEOUT,
3381 .psc_min_threads = min(max(mdt_num_threads, MDT_MIN_THREADS),
3383 .psc_max_threads = MDT_MAX_THREADS,
3384 .psc_ctx_tags = LCT_MD_THREAD
3386 m->mdt_readpage_service =
3387 ptlrpc_init_svc_conf(&conf, mdt_readpage_handle,
3388 LUSTRE_MDT_NAME "_readpage",
3389 procfs_entry, NULL, "mdt_rdpg");
3391 if (m->mdt_readpage_service == NULL) {
3392 CERROR("failed to start readpage service\n");
3393 GOTO(err_mdt_svc, rc = -ENOMEM);
3396 rc = ptlrpc_start_threads(NULL, m->mdt_readpage_service);
3399 * setattr service configuration.
3401 conf = (typeof(conf)) {
3402 .psc_nbufs = MDS_NBUFS,
3403 .psc_bufsize = MDS_BUFSIZE,
3404 .psc_max_req_size = MDS_MAXREQSIZE,
3405 .psc_max_reply_size = MDS_MAXREPSIZE,
3406 .psc_req_portal = MDS_SETATTR_PORTAL,
3407 .psc_rep_portal = MDC_REPLY_PORTAL,
3408 .psc_watchdog_timeout = MDT_SERVICE_WATCHDOG_TIMEOUT,
3409 .psc_min_threads = min(max(mdt_num_threads, MDT_MIN_THREADS),
3411 .psc_max_threads = MDT_MAX_THREADS,
3412 .psc_ctx_tags = LCT_MD_THREAD
3415 m->mdt_setattr_service =
3416 ptlrpc_init_svc_conf(&conf, mdt_regular_handle,
3417 LUSTRE_MDT_NAME "_setattr",
3418 procfs_entry, NULL, "mdt_attr");
3420 if (!m->mdt_setattr_service) {
3421 CERROR("failed to start setattr service\n");
3422 GOTO(err_mdt_svc, rc = -ENOMEM);
3425 rc = ptlrpc_start_threads(NULL, m->mdt_setattr_service);
3427 GOTO(err_mdt_svc, rc);
3430 * sequence controller service configuration
3432 conf = (typeof(conf)) {
3433 .psc_nbufs = MDS_NBUFS,
3434 .psc_bufsize = MDS_BUFSIZE,
3435 .psc_max_req_size = SEQ_MAXREQSIZE,
3436 .psc_max_reply_size = SEQ_MAXREPSIZE,
3437 .psc_req_portal = SEQ_CONTROLLER_PORTAL,
3438 .psc_rep_portal = MDC_REPLY_PORTAL,
3439 .psc_watchdog_timeout = MDT_SERVICE_WATCHDOG_TIMEOUT,
3440 .psc_min_threads = SEQ_NUM_THREADS,
3441 .psc_max_threads = SEQ_NUM_THREADS,
3442 .psc_ctx_tags = LCT_MD_THREAD|LCT_DT_THREAD
3445 m->mdt_mdsc_service =
3446 ptlrpc_init_svc_conf(&conf, mdt_mdsc_handle,
3447 LUSTRE_MDT_NAME"_mdsc",
3448 procfs_entry, NULL, "mdt_mdsc");
3449 if (!m->mdt_mdsc_service) {
3450 CERROR("failed to start seq controller service\n");
3451 GOTO(err_mdt_svc, rc = -ENOMEM);
3454 rc = ptlrpc_start_threads(NULL, m->mdt_mdsc_service);
3456 GOTO(err_mdt_svc, rc);
3459 * metadata sequence server service configuration
3461 conf = (typeof(conf)) {
3462 .psc_nbufs = MDS_NBUFS,
3463 .psc_bufsize = MDS_BUFSIZE,
3464 .psc_max_req_size = SEQ_MAXREQSIZE,
3465 .psc_max_reply_size = SEQ_MAXREPSIZE,
3466 .psc_req_portal = SEQ_METADATA_PORTAL,
3467 .psc_rep_portal = MDC_REPLY_PORTAL,
3468 .psc_watchdog_timeout = MDT_SERVICE_WATCHDOG_TIMEOUT,
3469 .psc_min_threads = SEQ_NUM_THREADS,
3470 .psc_max_threads = SEQ_NUM_THREADS,
3471 .psc_ctx_tags = LCT_MD_THREAD|LCT_DT_THREAD
3474 m->mdt_mdss_service =
3475 ptlrpc_init_svc_conf(&conf, mdt_mdss_handle,
3476 LUSTRE_MDT_NAME"_mdss",
3477 procfs_entry, NULL, "mdt_mdss");
3478 if (!m->mdt_mdss_service) {
3479 CERROR("failed to start metadata seq server service\n");
3480 GOTO(err_mdt_svc, rc = -ENOMEM);
3483 rc = ptlrpc_start_threads(NULL, m->mdt_mdss_service);
3485 GOTO(err_mdt_svc, rc);
3489 * Data sequence server service configuration. We want to have really
3490 * cluster-wide sequences space. This is why we start only one sequence
3491 * controller which manages space.
3493 conf = (typeof(conf)) {
3494 .psc_nbufs = MDS_NBUFS,
3495 .psc_bufsize = MDS_BUFSIZE,
3496 .psc_max_req_size = SEQ_MAXREQSIZE,
3497 .psc_max_reply_size = SEQ_MAXREPSIZE,
3498 .psc_req_portal = SEQ_DATA_PORTAL,
3499 .psc_rep_portal = OSC_REPLY_PORTAL,
3500 .psc_watchdog_timeout = MDT_SERVICE_WATCHDOG_TIMEOUT,
3501 .psc_min_threads = SEQ_NUM_THREADS,
3502 .psc_max_threads = SEQ_NUM_THREADS,
3503 .psc_ctx_tags = LCT_MD_THREAD|LCT_DT_THREAD
3506 m->mdt_dtss_service =
3507 ptlrpc_init_svc_conf(&conf, mdt_dtss_handle,
3508 LUSTRE_MDT_NAME"_dtss",
3509 procfs_entry, NULL, "mdt_dtss");
3510 if (!m->mdt_dtss_service) {
3511 CERROR("failed to start data seq server service\n");
3512 GOTO(err_mdt_svc, rc = -ENOMEM);
3515 rc = ptlrpc_start_threads(NULL, m->mdt_dtss_service);
3517 GOTO(err_mdt_svc, rc);
3519 /* FLD service start */
3520 conf = (typeof(conf)) {
3521 .psc_nbufs = MDS_NBUFS,
3522 .psc_bufsize = MDS_BUFSIZE,
3523 .psc_max_req_size = FLD_MAXREQSIZE,
3524 .psc_max_reply_size = FLD_MAXREPSIZE,
3525 .psc_req_portal = FLD_REQUEST_PORTAL,
3526 .psc_rep_portal = MDC_REPLY_PORTAL,
3527 .psc_watchdog_timeout = MDT_SERVICE_WATCHDOG_TIMEOUT,
3528 .psc_min_threads = FLD_NUM_THREADS,
3529 .psc_max_threads = FLD_NUM_THREADS,
3530 .psc_ctx_tags = LCT_DT_THREAD|LCT_MD_THREAD
3533 m->mdt_fld_service =
3534 ptlrpc_init_svc_conf(&conf, mdt_fld_handle,
3535 LUSTRE_MDT_NAME"_fld",
3536 procfs_entry, NULL, "mdt_fld");
3537 if (!m->mdt_fld_service) {
3538 CERROR("failed to start fld service\n");
3539 GOTO(err_mdt_svc, rc = -ENOMEM);
3542 rc = ptlrpc_start_threads(NULL, m->mdt_fld_service);
3544 GOTO(err_mdt_svc, rc);
3547 * mds-mds service configuration. Separate portal is used to allow
3548 * mds-mds requests be not blocked during recovery.
3550 conf = (typeof(conf)) {
3551 .psc_nbufs = MDS_NBUFS,
3552 .psc_bufsize = MDS_BUFSIZE,
3553 .psc_max_req_size = MDS_MAXREQSIZE,
3554 .psc_max_reply_size = MDS_MAXREPSIZE,
3555 .psc_req_portal = MDS_MDS_PORTAL,
3556 .psc_rep_portal = MDC_REPLY_PORTAL,
3557 .psc_watchdog_timeout = MDT_SERVICE_WATCHDOG_TIMEOUT,
3558 .psc_min_threads = min(max(mdt_num_threads, MDT_MIN_THREADS),
3560 .psc_max_threads = MDT_MAX_THREADS,
3561 .psc_ctx_tags = LCT_MD_THREAD
3563 m->mdt_xmds_service = ptlrpc_init_svc_conf(&conf, mdt_xmds_handle,
3564 LUSTRE_MDT_NAME "_mds",
3565 procfs_entry, NULL, "mdt_xmds");
3567 if (m->mdt_xmds_service == NULL) {
3568 CERROR("failed to start readpage service\n");
3569 GOTO(err_mdt_svc, rc = -ENOMEM);
3572 rc = ptlrpc_start_threads(NULL, m->mdt_xmds_service);
3574 GOTO(err_mdt_svc, rc);
3579 mdt_stop_ptlrpc_service(m);
3584 static void mdt_stack_fini(const struct lu_env *env,
3585 struct mdt_device *m, struct lu_device *top)
3587 struct obd_device *obd = m->mdt_md_dev.md_lu_dev.ld_obd;
3588 struct lustre_cfg_bufs *bufs;
3589 struct lustre_cfg *lcfg;
3590 struct mdt_thread_info *info;
3594 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3595 LASSERT(info != NULL);
3597 bufs = &info->mti_u.bufs;
3598 /* process cleanup, pass mdt obd name to get obd umount flags */
3599 lustre_cfg_bufs_reset(bufs, obd->obd_name);
3604 lustre_cfg_bufs_set_string(bufs, 1, flags);
3605 lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
3607 CERROR("Cannot alloc lcfg!\n");
3612 top->ld_ops->ldo_process_config(env, top, lcfg);
3613 lustre_cfg_free(lcfg);
3615 lu_stack_fini(env, top);
3616 m->mdt_child = NULL;
3617 m->mdt_bottom = NULL;
3620 static struct lu_device *mdt_layer_setup(const struct lu_env *env,
3621 const char *typename,
3622 struct lu_device *child,
3623 struct lustre_cfg *cfg)
3625 const char *dev = lustre_cfg_string(cfg, 0);
3626 struct obd_type *type;
3627 struct lu_device_type *ldt;
3628 struct lu_device *d;
3633 type = class_get_type(typename);
3635 CERROR("Unknown type: '%s'\n", typename);
3636 GOTO(out, rc = -ENODEV);
3639 rc = lu_context_refill(&env->le_ctx);
3641 CERROR("Failure to refill context: '%d'\n", rc);
3645 if (env->le_ses != NULL) {
3646 rc = lu_context_refill(env->le_ses);
3648 CERROR("Failure to refill session: '%d'\n", rc);
3655 CERROR("type: '%s'\n", typename);
3656 GOTO(out_type, rc = -EINVAL);
3659 ldt->ldt_obd_type = type;
3660 d = ldt->ldt_ops->ldto_device_alloc(env, ldt, cfg);
3662 CERROR("Cannot allocate device: '%s'\n", typename);
3663 GOTO(out_type, rc = -ENODEV);
3666 LASSERT(child->ld_site);
3667 d->ld_site = child->ld_site;
3670 rc = ldt->ldt_ops->ldto_device_init(env, d, dev, child);
3672 CERROR("can't init device '%s', rc %d\n", typename, rc);
3673 GOTO(out_alloc, rc);
3680 ldt->ldt_ops->ldto_device_free(env, d);
3683 class_put_type(type);
3688 static int mdt_stack_init(const struct lu_env *env,
3689 struct mdt_device *m, struct lustre_cfg *cfg)
3691 struct lu_device *d = &m->mdt_md_dev.md_lu_dev;
3692 struct lu_device *tmp;
3693 struct md_device *md;
3697 /* init the stack */
3698 tmp = mdt_layer_setup(env, LUSTRE_OSD_NAME, d, cfg);
3700 RETURN(PTR_ERR(tmp));
3702 m->mdt_bottom = lu2dt_dev(tmp);
3704 tmp = mdt_layer_setup(env, LUSTRE_MDD_NAME, d, cfg);
3706 GOTO(out, rc = PTR_ERR(tmp));
3711 tmp = mdt_layer_setup(env, LUSTRE_CMM_NAME, d, cfg);
3713 GOTO(out, rc = PTR_ERR(tmp));
3716 /*set mdd upcall device*/
3717 md_upcall_dev_set(md, lu2md_dev(d));
3720 /*set cmm upcall device*/
3721 md_upcall_dev_set(md, &m->mdt_md_dev);
3723 m->mdt_child = lu2md_dev(d);
3725 /* process setup config */
3726 tmp = &m->mdt_md_dev.md_lu_dev;
3727 rc = tmp->ld_ops->ldo_process_config(env, tmp, cfg);
3730 /* fini from last known good lu_device */
3732 mdt_stack_fini(env, m, d);
3737 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
3739 struct md_device *next = m->mdt_child;
3740 struct lu_device *d = &m->mdt_md_dev.md_lu_dev;
3741 struct lu_site *ls = d->ld_site;
3742 struct obd_device *obd = m->mdt_md_dev.md_lu_dev.ld_obd;
3745 ping_evictor_stop();
3747 target_recovery_fini(obd);
3748 mdt_stop_ptlrpc_service(m);
3750 mdt_fs_cleanup(env, m);
3752 upcall_cache_cleanup(m->mdt_identity_cache);
3753 m->mdt_identity_cache = NULL;
3755 if (m->mdt_namespace != NULL) {
3756 ldlm_namespace_free(m->mdt_namespace, NULL, d->ld_obd->obd_force);
3757 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
3760 mdt_seq_fini(env, m);
3761 mdt_seq_fini_cli(m);
3762 mdt_fld_fini(env, m);
3764 ptlrpc_lprocfs_unregister_obd(d->ld_obd);
3765 lprocfs_obd_cleanup(d->ld_obd);
3767 sptlrpc_rule_set_free(&m->mdt_sptlrpc_rset);
3769 next->md_ops->mdo_init_capa_ctxt(env, next, 0, 0, 0, NULL);
3770 cfs_timer_disarm(&m->mdt_ck_timer);
3771 mdt_ck_thread_stop(m);
3773 /* finish the stack */
3774 mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
3781 LASSERT(atomic_read(&d->ld_ref) == 0);
3782 md_device_fini(&m->mdt_md_dev);
3787 static void fsoptions_to_mdt_flags(struct mdt_device *m, char *options)
3791 #ifdef CONFIG_FS_POSIX_ACL
3792 /* ACLs should be enabled by default (b=13829) */
3793 m->mdt_opts.mo_acl = 1;
3794 LCONSOLE_INFO("Enabling ACL\n");
3796 m->mdt_opts.mo_acl = 0;
3797 LCONSOLE_INFO("Disabling ACL\n");
3806 while (*p && *p != ',')
3810 if ((len == sizeof("user_xattr") - 1) &&
3811 (memcmp(options, "user_xattr", len) == 0)) {
3812 m->mdt_opts.mo_user_xattr = 1;
3813 LCONSOLE_INFO("Enabling user_xattr\n");
3814 } else if ((len == sizeof("nouser_xattr") - 1) &&
3815 (memcmp(options, "nouser_xattr", len) == 0)) {
3816 m->mdt_opts.mo_user_xattr = 0;
3817 LCONSOLE_INFO("Disabling user_xattr\n");
3818 } else if ((len == sizeof("noacl") - 1) &&
3819 (memcmp(options, "noacl", len) == 0)) {
3820 m->mdt_opts.mo_acl = 0;
3821 LCONSOLE_INFO("Disabling ACL\n");
3828 int mdt_postrecov(const struct lu_env *, struct mdt_device *);
3830 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
3831 struct lu_device_type *ldt, struct lustre_cfg *cfg)
3833 struct lprocfs_static_vars lvars;
3834 struct mdt_thread_info *info;
3835 struct obd_device *obd;
3836 const char *dev = lustre_cfg_string(cfg, 0);
3837 const char *num = lustre_cfg_string(cfg, 2);
3838 struct lustre_mount_info *lmi;
3839 struct lustre_sb_info *lsi;
3841 const char *identity_upcall = "NONE";
3845 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3846 LASSERT(info != NULL);
3848 obd = class_name2obd(dev);
3849 LASSERT(obd != NULL);
3851 spin_lock_init(&m->mdt_transno_lock);
3853 m->mdt_max_mdsize = MAX_MD_SIZE;
3854 m->mdt_max_cookiesize = sizeof(struct llog_cookie);
3856 m->mdt_opts.mo_user_xattr = 0;
3857 m->mdt_opts.mo_acl = 0;
3858 lmi = server_get_mount_2(dev);
3860 CERROR("Cannot get mount info for %s!\n", dev);
3863 lsi = s2lsi(lmi->lmi_sb);
3864 fsoptions_to_mdt_flags(m, lsi->lsi_lmd->lmd_opts);
3865 server_put_mount_2(dev, lmi->lmi_mnt);
3868 m->mdt_sptlrpc_lock = RW_LOCK_UNLOCKED;
3869 sptlrpc_rule_set_init(&m->mdt_sptlrpc_rset);
3871 spin_lock_init(&m->mdt_ioepoch_lock);
3872 m->mdt_opts.mo_compat_resname = 0;
3873 m->mdt_capa_timeout = CAPA_TIMEOUT;
3874 m->mdt_capa_alg = CAPA_HMAC_ALG_SHA1;
3875 m->mdt_ck_timeout = CAPA_KEY_TIMEOUT;
3877 spin_lock_init(&m->mdt_client_bitmap_lock);
3883 md_device_init(&m->mdt_md_dev, ldt);
3884 m->mdt_md_dev.md_lu_dev.ld_ops = &mdt_lu_ops;
3885 m->mdt_md_dev.md_lu_dev.ld_obd = obd;
3886 /* set this lu_device to obd, because error handling need it */
3887 obd->obd_lu_dev = &m->mdt_md_dev.md_lu_dev;
3889 rc = lu_site_init(s, &m->mdt_md_dev.md_lu_dev);
3891 CERROR("Can't init lu_site, rc %d\n", rc);
3892 GOTO(err_free_site, rc);
3895 lprocfs_mdt_init_vars(&lvars);
3896 rc = lprocfs_obd_setup(obd, lvars.obd_vars);
3898 CERROR("Can't init lprocfs, rc %d\n", rc);
3899 GOTO(err_fini_site, rc);
3901 ptlrpc_lprocfs_register_obd(obd);
3903 rc = mdt_procfs_init(m, dev);
3905 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
3906 GOTO(err_fini_proc, rc);
3909 /* set server index */
3911 s->ls_node_id = simple_strtol(num, NULL, 10);
3913 /* failover is the default
3914 * FIXME: we do not failout mds0/mgs, which may cause some problems.
3915 * assumed whose ls_node_id == 0 XXX
3917 obd->obd_replayable = 1;
3918 /* No connection accepted until configurations will finish */
3919 obd->obd_no_conn = 1;
3921 if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
3922 char *str = lustre_cfg_string(cfg, 4);
3923 if (strchr(str, 'n')) {
3924 CWARN("%s: recovery disabled\n", obd->obd_name);
3925 obd->obd_replayable = 0;
3929 /* init the stack */
3930 rc = mdt_stack_init(env, m, cfg);
3932 CERROR("Can't init device stack, rc %d\n", rc);
3933 GOTO(err_fini_proc, rc);
3936 rc = mdt_fld_init(env, obd->obd_name, m);
3938 GOTO(err_fini_stack, rc);
3940 rc = mdt_seq_init(env, obd->obd_name, m);
3942 GOTO(err_fini_fld, rc);
3944 snprintf(info->mti_u.ns_name, sizeof info->mti_u.ns_name,
3945 LUSTRE_MDT_NAME"-%p", m);
3946 m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
3947 LDLM_NAMESPACE_SERVER,
3948 LDLM_NAMESPACE_GREEDY);
3949 if (m->mdt_namespace == NULL)
3950 GOTO(err_fini_seq, rc = -ENOMEM);
3952 ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
3953 /* set obd_namespace for compatibility with old code */
3954 obd->obd_namespace = m->mdt_namespace;
3956 /* XXX: to support suppgid for ACL, we enable identity_upcall
3957 * by default, otherwise, maybe got unexpected -EACCESS. */
3958 if (m->mdt_opts.mo_acl)
3959 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
3961 m->mdt_identity_cache = upcall_cache_init(obd->obd_name, identity_upcall,
3962 &mdt_identity_upcall_cache_ops);
3963 if (IS_ERR(m->mdt_identity_cache)) {
3964 rc = PTR_ERR(m->mdt_identity_cache);
3965 m->mdt_identity_cache = NULL;
3966 GOTO(err_free_ns, rc);
3969 cfs_timer_init(&m->mdt_ck_timer, mdt_ck_timer_callback, m);
3971 rc = mdt_ck_thread_start(m);
3973 GOTO(err_free_ns, rc);
3975 rc = mdt_fs_setup(env, m, obd);
3979 target_recovery_init(obd, mdt_recovery_handle);
3981 rc = mdt_start_ptlrpc_service(m);
3983 GOTO(err_fs_cleanup, rc);
3985 ping_evictor_start();
3987 rc = lu_site_init_finish(s);
3989 GOTO(err_stop_service, rc);
3991 if (obd->obd_recovering == 0)
3992 mdt_postrecov(env, m);
3994 mdt_init_capa_ctxt(env, m);
3996 if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
4002 ping_evictor_stop();
4003 mdt_stop_ptlrpc_service(m);
4005 target_recovery_fini(obd);
4006 mdt_fs_cleanup(env, m);
4008 cfs_timer_disarm(&m->mdt_ck_timer);
4009 mdt_ck_thread_stop(m);
4011 upcall_cache_cleanup(m->mdt_identity_cache);
4012 m->mdt_identity_cache = NULL;
4013 ldlm_namespace_free(m->mdt_namespace, NULL, 0);
4014 obd->obd_namespace = m->mdt_namespace = NULL;
4016 mdt_seq_fini(env, m);
4018 mdt_fld_fini(env, m);
4020 mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4023 lprocfs_obd_cleanup(obd);
4029 md_device_fini(&m->mdt_md_dev);
4033 /* used by MGS to process specific configurations */
4034 static int mdt_process_config(const struct lu_env *env,
4035 struct lu_device *d, struct lustre_cfg *cfg)
4037 struct mdt_device *m = mdt_dev(d);
4038 struct md_device *md_next = m->mdt_child;
4039 struct lu_device *next = md2lu_dev(md_next);
4043 switch (cfg->lcfg_command) {
4044 case LCFG_SPTLRPC_CONF: {
4045 struct sptlrpc_conf_log *log;
4046 struct sptlrpc_rule_set tmp_rset;
4048 log = sptlrpc_conf_log_extract(cfg);
4054 sptlrpc_rule_set_init(&tmp_rset);
4056 rc = sptlrpc_rule_set_from_log(&tmp_rset, log);
4058 CERROR("mdt %p: failed get sptlrpc rules: %d\n", m, rc);
4062 write_lock(&m->mdt_sptlrpc_lock);
4063 sptlrpc_rule_set_free(&m->mdt_sptlrpc_rset);
4064 m->mdt_sptlrpc_rset = tmp_rset;
4065 write_unlock(&m->mdt_sptlrpc_lock);
4067 sptlrpc_target_update_exp_flavor(
4068 md2lu_dev(&m->mdt_md_dev)->ld_obd, &tmp_rset);
4073 struct lprocfs_static_vars lvars;
4074 struct obd_device *obd = d->ld_obd;
4076 lprocfs_mdt_init_vars(&lvars);
4077 rc = class_process_proc_param(PARAM_MDT, lvars.obd_vars, cfg, obd);
4079 /* others are passed further */
4080 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4085 * Add mdc hook to get first MDT uuid and connect it to
4086 * ls->controller to use for seq manager.
4088 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4090 CERROR("Can't add mdc, rc %d\n", rc);
4092 rc = mdt_seq_init_cli(env, mdt_dev(d), cfg);
4095 /* others are passed further */
4096 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4102 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
4103 const struct lu_object_header *hdr,
4104 struct lu_device *d)
4106 struct mdt_object *mo;
4112 struct lu_object *o;
4113 struct lu_object_header *h;
4115 o = &mo->mot_obj.mo_lu;
4116 h = &mo->mot_header;
4117 lu_object_header_init(h);
4118 lu_object_init(o, h, d);
4119 lu_object_add_top(h, o);
4120 o->lo_ops = &mdt_obj_ops;
4126 static int mdt_object_init(const struct lu_env *env, struct lu_object *o)
4128 struct mdt_device *d = mdt_dev(o->lo_dev);
4129 struct lu_device *under;
4130 struct lu_object *below;
4134 CDEBUG(D_INFO, "object init, fid = "DFID"\n",
4135 PFID(lu_object_fid(o)));
4137 under = &d->mdt_child->md_lu_dev;
4138 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
4139 if (below != NULL) {
4140 lu_object_add(o, below);
4147 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
4149 struct mdt_object *mo = mdt_obj(o);
4150 struct lu_object_header *h;
4154 CDEBUG(D_INFO, "object free, fid = "DFID"\n",
4155 PFID(lu_object_fid(o)));
4158 lu_object_header_fini(h);
4163 static int mdt_object_print(const struct lu_env *env, void *cookie,
4164 lu_printer_t p, const struct lu_object *o)
4166 return (*p)(env, cookie, LUSTRE_MDT_NAME"-object@%p", o);
4169 static struct lu_device_operations mdt_lu_ops = {
4170 .ldo_object_alloc = mdt_object_alloc,
4171 .ldo_process_config = mdt_process_config
4174 static struct lu_object_operations mdt_obj_ops = {
4175 .loo_object_init = mdt_object_init,
4176 .loo_object_free = mdt_object_free,
4177 .loo_object_print = mdt_object_print
4180 /* mds_connect_internal */
4181 static int mdt_connect_internal(struct obd_export *exp,
4182 struct mdt_device *mdt,
4183 struct obd_connect_data *data)
4188 data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
4189 data->ocd_ibits_known &= MDS_INODELOCK_FULL;
4191 /* If no known bits (which should not happen, probably,
4192 as everybody should support LOOKUP and UPDATE bits at least)
4193 revert to compat mode with plain locks. */
4194 if (!data->ocd_ibits_known &&
4195 data->ocd_connect_flags & OBD_CONNECT_IBITS)
4196 data->ocd_connect_flags &= ~OBD_CONNECT_IBITS;
4198 if (!mdt->mdt_opts.mo_acl)
4199 data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
4201 if (!mdt->mdt_opts.mo_user_xattr)
4202 data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
4204 if (!mdt->mdt_opts.mo_mds_capa)
4205 data->ocd_connect_flags &= ~OBD_CONNECT_MDS_CAPA;
4207 if (!mdt->mdt_opts.mo_oss_capa)
4208 data->ocd_connect_flags &= ~OBD_CONNECT_OSS_CAPA;
4210 spin_lock(&exp->exp_lock);
4211 exp->exp_connect_flags = data->ocd_connect_flags;
4212 spin_unlock(&exp->exp_lock);
4213 data->ocd_version = LUSTRE_VERSION_CODE;
4214 exp->exp_mdt_data.med_ibits_known = data->ocd_ibits_known;
4218 if (mdt->mdt_opts.mo_acl &&
4219 ((exp->exp_connect_flags & OBD_CONNECT_ACL) == 0)) {
4220 CWARN("%s: MDS requires ACL support but client does not\n",
4221 mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4226 flags = OBD_CONNECT_LCL_CLIENT | OBD_CONNECT_RMT_CLIENT;
4227 if ((exp->exp_connect_flags & flags) == flags) {
4228 CWARN("%s: both local and remote client flags are set\n",
4229 mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4233 if (mdt->mdt_opts.mo_mds_capa &&
4234 ((exp->exp_connect_flags & OBD_CONNECT_MDS_CAPA) == 0)) {
4235 CWARN("%s: MDS requires capability support, but client not\n",
4236 mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4240 if (mdt->mdt_opts.mo_oss_capa &&
4241 ((exp->exp_connect_flags & OBD_CONNECT_OSS_CAPA) == 0)) {
4242 CWARN("%s: MDS requires OSS capability support, "
4244 mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4248 if ((exp->exp_connect_flags & OBD_CONNECT_FID) == 0) {
4249 CWARN("%s: MDS requires FID support, but client not\n",
4250 mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4257 /* mds_connect copy */
4258 static int mdt_obd_connect(const struct lu_env *env,
4259 struct lustre_handle *conn, struct obd_device *obd,
4260 struct obd_uuid *cluuid,
4261 struct obd_connect_data *data,
4264 struct mdt_thread_info *info;
4265 struct mdt_client_data *mcd;
4266 struct obd_export *exp;
4267 struct mdt_device *mdt;
4268 struct ptlrpc_request *req;
4272 LASSERT(env != NULL);
4273 if (!conn || !obd || !cluuid)
4276 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4277 req = info->mti_pill->rc_req;
4278 mdt = mdt_dev(obd->obd_lu_dev);
4280 rc = class_connect(conn, obd, cluuid);
4284 exp = class_conn2export(conn);
4285 LASSERT(exp != NULL);
4287 CDEBUG(D_SEC, "from %s\n", sptlrpc_part2name(req->rq_sp_from));
4289 spin_lock(&exp->exp_lock);
4290 exp->exp_sp_peer = req->rq_sp_from;
4292 read_lock(&mdt->mdt_sptlrpc_lock);
4293 sptlrpc_rule_set_choose(&mdt->mdt_sptlrpc_rset, exp->exp_sp_peer,
4294 req->rq_peer.nid, &exp->exp_flvr);
4295 read_unlock(&mdt->mdt_sptlrpc_lock);
4297 if (exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
4298 CERROR("invalid rpc flavor %x, expect %x, from %s\n",
4299 req->rq_flvr.sf_rpc, exp->exp_flvr.sf_rpc,
4300 libcfs_nid2str(req->rq_peer.nid));
4301 exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_INVALID;
4302 spin_unlock(&exp->exp_lock);
4305 spin_unlock(&exp->exp_lock);
4307 rc = mdt_connect_internal(exp, mdt, data);
4311 struct mdt_thread_info *mti;
4312 mti = lu_context_key_get(&env->le_ctx,
4314 LASSERT(mti != NULL);
4316 memcpy(mcd->mcd_uuid, cluuid, sizeof mcd->mcd_uuid);
4317 exp->exp_mdt_data.med_mcd = mcd;
4318 rc = mdt_client_new(env, mdt);
4321 exp->exp_mdt_data.med_mcd = NULL;
4328 class_disconnect(exp);
4330 class_export_put(exp);
4335 static int mdt_obd_reconnect(const struct lu_env *env,
4336 struct obd_export *exp, struct obd_device *obd,
4337 struct obd_uuid *cluuid,
4338 struct obd_connect_data *data)
4340 struct mdt_thread_info *info;
4341 struct mdt_device *mdt;
4342 struct ptlrpc_request *req;
4346 if (exp == NULL || obd == NULL || cluuid == NULL)
4349 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4350 req = info->mti_pill->rc_req;
4351 mdt = mdt_dev(obd->obd_lu_dev);
4353 CDEBUG(D_SEC, "from %s\n", sptlrpc_part2name(req->rq_sp_from));
4355 spin_lock(&exp->exp_lock);
4356 if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
4357 exp->exp_sp_peer = req->rq_sp_from;
4359 read_lock(&mdt->mdt_sptlrpc_lock);
4360 sptlrpc_rule_set_choose(&mdt->mdt_sptlrpc_rset,
4362 req->rq_peer.nid, &exp->exp_flvr);
4363 read_unlock(&mdt->mdt_sptlrpc_lock);
4365 if (exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
4366 CERROR("invalid rpc flavor %x, expect %x, from %s\n",
4367 req->rq_flvr.sf_rpc, exp->exp_flvr.sf_rpc,
4368 libcfs_nid2str(req->rq_peer.nid));
4369 exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_INVALID;
4370 spin_unlock(&exp->exp_lock);
4374 spin_unlock(&exp->exp_lock);
4376 rc = mdt_connect_internal(exp, mdt_dev(obd->obd_lu_dev), data);
4381 static int mdt_obd_disconnect(struct obd_export *exp)
4383 struct mdt_device *mdt = mdt_dev(exp->exp_obd->obd_lu_dev);
4388 class_export_get(exp);
4390 /* Disconnect early so that clients can't keep using export */
4391 rc = class_disconnect(exp);
4392 if (mdt->mdt_namespace != NULL || exp->exp_obd->obd_namespace != NULL)
4393 ldlm_cancel_locks_for_export(exp);
4395 /* complete all outstanding replies */
4396 spin_lock(&exp->exp_lock);
4397 while (!list_empty(&exp->exp_outstanding_replies)) {
4398 struct ptlrpc_reply_state *rs =
4399 list_entry(exp->exp_outstanding_replies.next,
4400 struct ptlrpc_reply_state, rs_exp_list);
4401 struct ptlrpc_service *svc = rs->rs_service;
4403 spin_lock(&svc->srv_lock);
4404 list_del_init(&rs->rs_exp_list);
4405 ptlrpc_schedule_difficult_reply(rs);
4406 spin_unlock(&svc->srv_lock);
4408 spin_unlock(&exp->exp_lock);
4410 class_export_put(exp);
4414 /* FIXME: Can we avoid using these two interfaces? */
4415 static int mdt_init_export(struct obd_export *exp)
4417 struct mdt_export_data *med = &exp->exp_mdt_data;
4420 CFS_INIT_LIST_HEAD(&med->med_open_head);
4421 spin_lock_init(&med->med_open_lock);
4422 sema_init(&med->med_idmap_sem, 1);
4423 med->med_idmap = NULL;
4424 spin_lock(&exp->exp_lock);
4425 exp->exp_connecting = 1;
4426 spin_unlock(&exp->exp_lock);
4430 static int mdt_destroy_export(struct obd_export *export)
4432 struct mdt_export_data *med;
4433 struct obd_device *obd = export->exp_obd;
4434 struct mdt_device *mdt;
4435 struct mdt_thread_info *info;
4443 med = &export->exp_mdt_data;
4444 if (med->med_rmtclient)
4445 mdt_cleanup_idmap(med);
4447 target_destroy_export(export);
4449 if (obd_uuid_equals(&export->exp_client_uuid, &obd->obd_uuid))
4452 mdt = mdt_dev(obd->obd_lu_dev);
4453 LASSERT(mdt != NULL);
4455 rc = lu_env_init(&env, NULL, LCT_MD_THREAD);
4459 info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
4460 LASSERT(info != NULL);
4461 memset(info, 0, sizeof *info);
4462 info->mti_env = &env;
4463 info->mti_mdt = mdt;
4464 info->mti_exp = export;
4466 ma = &info->mti_attr;
4467 lmm_size = ma->ma_lmm_size = mdt->mdt_max_mdsize;
4468 cookie_size = ma->ma_cookie_size = mdt->mdt_max_cookiesize;
4469 OBD_ALLOC(ma->ma_lmm, lmm_size);
4470 OBD_ALLOC(ma->ma_cookie, cookie_size);
4472 if (ma->ma_lmm == NULL || ma->ma_cookie == NULL)
4473 GOTO(out, rc = -ENOMEM);
4474 ma->ma_need = MA_LOV | MA_COOKIE;
4476 /* Close any open files (which may also cause orphan unlinking). */
4477 spin_lock(&med->med_open_lock);
4478 while (!list_empty(&med->med_open_head)) {
4479 struct list_head *tmp = med->med_open_head.next;
4480 struct mdt_file_data *mfd =
4481 list_entry(tmp, struct mdt_file_data, mfd_list);
4483 /* Remove mfd handle so it can't be found again.
4484 * We are consuming the mfd_list reference here. */
4485 class_handle_unhash(&mfd->mfd_handle);
4486 list_del_init(&mfd->mfd_list);
4487 spin_unlock(&med->med_open_lock);
4488 mdt_mfd_close(info, mfd);
4489 /* TODO: if we close the unlinked file,
4490 * we need to remove it's objects from OST */
4491 memset(&ma->ma_attr, 0, sizeof(ma->ma_attr));
4492 spin_lock(&med->med_open_lock);
4493 ma->ma_lmm_size = lmm_size;
4494 ma->ma_cookie_size = cookie_size;
4495 ma->ma_need = MA_LOV | MA_COOKIE;
4498 spin_unlock(&med->med_open_lock);
4499 info->mti_mdt = NULL;
4500 mdt_client_del(&env, mdt);
4505 OBD_FREE(ma->ma_lmm, lmm_size);
4509 OBD_FREE(ma->ma_cookie, cookie_size);
4510 ma->ma_cookie = NULL;
4517 static void mdt_allow_cli(struct mdt_device *m, unsigned int flag)
4519 if (flag & CONFIG_LOG)
4520 m->mdt_fl_cfglog = 1;
4521 if (flag & CONFIG_SYNC)
4522 m->mdt_fl_synced = 1;
4524 if (m->mdt_fl_cfglog /* bz11778: && m->mdt_fl_synced */)
4525 /* Open for clients */
4526 m->mdt_md_dev.md_lu_dev.ld_obd->obd_no_conn = 0;
4529 static int mdt_upcall(const struct lu_env *env, struct md_device *md,
4530 enum md_upcall_event ev)
4532 struct mdt_device *m = mdt_dev(&md->md_lu_dev);
4533 struct md_device *next = m->mdt_child;
4534 struct mdt_thread_info *mti;
4540 rc = next->md_ops->mdo_maxsize_get(env, next,
4542 &m->mdt_max_cookiesize);
4543 CDEBUG(D_INFO, "get max mdsize %d max cookiesize %d\n",
4544 m->mdt_max_mdsize, m->mdt_max_cookiesize);
4545 mdt_allow_cli(m, CONFIG_SYNC);
4548 mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4549 mti->mti_no_need_trans = 1;
4550 CDEBUG(D_INFO, "disable mdt trans for this thread\n");
4553 /* Check that MDT is not yet configured */
4554 LASSERT(!m->mdt_fl_cfglog);
4557 CERROR("invalid event\n");
4564 static int mdt_obd_notify(struct obd_device *host,
4565 struct obd_device *watched,
4566 enum obd_notify_event ev, void *data)
4571 case OBD_NOTIFY_CONFIG:
4572 mdt_allow_cli(mdt_dev(host->obd_lu_dev), (unsigned int)data);
4575 CDEBUG(D_INFO, "Unhandled notification %#x\n", ev);
4580 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
4581 void *karg, void *uarg)
4584 struct obd_device *obd= exp->exp_obd;
4585 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
4586 struct dt_device *dt = mdt->mdt_bottom;
4590 CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
4591 rc = lu_env_init(&env, NULL, LCT_MD_THREAD);
4597 rc = dt->dd_ops->dt_sync(&env, dt);
4599 case OBD_IOC_SET_READONLY:
4600 rc = dt->dd_ops->dt_sync(&env, dt);
4601 dt->dd_ops->dt_ro(&env, dt);
4603 case OBD_IOC_ABORT_RECOVERY:
4604 CERROR("Aborting recovery for device %s\n", obd->obd_name);
4605 target_stop_recovery_thread(obd);
4609 CERROR("Not supported cmd = %d for device %s\n",
4610 cmd, obd->obd_name);
4618 int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
4620 struct lu_device *ld = md2lu_dev(mdt->mdt_child);
4621 struct obd_device *obd = mdt->mdt_md_dev.md_lu_dev.ld_obd;
4624 /* if some clients didn't participate in recovery then we can possibly
4625 * lost sequence. Now we should increase sequence for safe value */
4626 lost = obd->obd_max_recoverable_clients - obd->obd_connected_clients;
4627 mdt_seq_adjust(env, mdt, lost);
4629 rc = ld->ld_ops->ldo_recovery_complete(env, ld);
4633 int mdt_obd_postrecov(struct obd_device *obd)
4638 rc = lu_env_init(&env, NULL, LCT_MD_THREAD);
4641 rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
4646 static struct obd_ops mdt_obd_device_ops = {
4647 .o_owner = THIS_MODULE,
4648 .o_connect = mdt_obd_connect,
4649 .o_reconnect = mdt_obd_reconnect,
4650 .o_disconnect = mdt_obd_disconnect,
4651 .o_init_export = mdt_init_export,
4652 .o_destroy_export = mdt_destroy_export,
4653 .o_iocontrol = mdt_iocontrol,
4654 .o_postrecov = mdt_obd_postrecov,
4655 .o_notify = mdt_obd_notify
4658 static struct lu_device* mdt_device_fini(const struct lu_env *env,
4659 struct lu_device *d)
4661 struct mdt_device *m = mdt_dev(d);
4668 static struct lu_device *mdt_device_free(const struct lu_env *env,
4669 struct lu_device *d)
4671 struct mdt_device *m = mdt_dev(d);
4678 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
4679 struct lu_device_type *t,
4680 struct lustre_cfg *cfg)
4682 struct lu_device *l;
4683 struct mdt_device *m;
4689 l = &m->mdt_md_dev.md_lu_dev;
4690 rc = mdt_init0(env, m, t, cfg);
4696 md_upcall_init(&m->mdt_md_dev, mdt_upcall);
4698 l = ERR_PTR(-ENOMEM);
4702 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
4703 LU_KEY_INIT_FINI(mdt, struct mdt_thread_info);
4705 /* context key: mdt_thread_key */
4706 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
4708 /* context key constructor/destructor: mdt_txn_key_init, mdt_txn_key_fini */
4709 LU_KEY_INIT_FINI(mdt_txn, struct mdt_txn_info);
4711 struct lu_context_key mdt_txn_key = {
4712 .lct_tags = LCT_TX_HANDLE,
4713 .lct_init = mdt_txn_key_init,
4714 .lct_fini = mdt_txn_key_fini
4717 struct md_ucred *mdt_ucred(const struct mdt_thread_info *info)
4719 return md_ucred(info->mti_env);
4722 /* type constructor/destructor: mdt_type_init, mdt_type_fini */
4723 LU_TYPE_INIT_FINI(mdt, &mdt_thread_key, &mdt_txn_key);
4725 static struct lu_device_type_operations mdt_device_type_ops = {
4726 .ldto_init = mdt_type_init,
4727 .ldto_fini = mdt_type_fini,
4729 .ldto_device_alloc = mdt_device_alloc,
4730 .ldto_device_free = mdt_device_free,
4731 .ldto_device_fini = mdt_device_fini
4734 static struct lu_device_type mdt_device_type = {
4735 .ldt_tags = LU_DEVICE_MD,
4736 .ldt_name = LUSTRE_MDT_NAME,
4737 .ldt_ops = &mdt_device_type_ops,
4738 .ldt_ctx_tags = LCT_MD_THREAD
4741 static int __init mdt_mod_init(void)
4743 struct lprocfs_static_vars lvars;
4746 mdt_num_threads = MDT_NUM_THREADS;
4747 lprocfs_mdt_init_vars(&lvars);
4748 rc = class_register_type(&mdt_obd_device_ops, NULL,
4749 lvars.module_vars, LUSTRE_MDT_NAME,
4755 static void __exit mdt_mod_exit(void)
4757 class_unregister_type(LUSTRE_MDT_NAME);
4761 #define DEF_HNDL(prefix, base, suffix, flags, opc, fn, fmt) \
4762 [prefix ## _ ## opc - prefix ## _ ## base] = { \
4764 .mh_fail_id = OBD_FAIL_ ## prefix ## _ ## opc ## suffix, \
4765 .mh_opc = prefix ## _ ## opc, \
4766 .mh_flags = flags, \
4771 #define DEF_MDT_HNDL(flags, name, fn, fmt) \
4772 DEF_HNDL(MDS, GETATTR, _NET, flags, name, fn, fmt)
4774 #define DEF_SEQ_HNDL(flags, name, fn, fmt) \
4775 DEF_HNDL(SEQ, QUERY, _NET, flags, name, fn, fmt)
4777 #define DEF_FLD_HNDL(flags, name, fn, fmt) \
4778 DEF_HNDL(FLD, QUERY, _NET, flags, name, fn, fmt)
4780 * Request with a format known in advance
4782 #define DEF_MDT_HNDL_F(flags, name, fn) \
4783 DEF_HNDL(MDS, GETATTR, _NET, flags, name, fn, &RQF_MDS_ ## name)
4785 #define DEF_SEQ_HNDL_F(flags, name, fn) \
4786 DEF_HNDL(SEQ, QUERY, _NET, flags, name, fn, &RQF_SEQ_ ## name)
4788 #define DEF_FLD_HNDL_F(flags, name, fn) \
4789 DEF_HNDL(FLD, QUERY, _NET, flags, name, fn, &RQF_FLD_ ## name)
4791 * Request with a format we do not yet know
4793 #define DEF_MDT_HNDL_0(flags, name, fn) \
4794 DEF_HNDL(MDS, GETATTR, _NET, flags, name, fn, NULL)
4796 static struct mdt_handler mdt_mds_ops[] = {
4797 DEF_MDT_HNDL_F(0, CONNECT, mdt_connect),
4798 DEF_MDT_HNDL_F(0, DISCONNECT, mdt_disconnect),
4799 DEF_MDT_HNDL_F(0, SET_INFO, mdt_set_info),
4800 DEF_MDT_HNDL_F(0 |HABEO_REFERO, GETSTATUS, mdt_getstatus),
4801 DEF_MDT_HNDL_F(HABEO_CORPUS, GETATTR, mdt_getattr),
4802 DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, GETATTR_NAME, mdt_getattr_name),
4803 DEF_MDT_HNDL_F(HABEO_CORPUS, GETXATTR, mdt_getxattr),
4804 DEF_MDT_HNDL_F(0 |HABEO_REFERO, STATFS, mdt_statfs),
4805 DEF_MDT_HNDL_F(0 |MUTABOR, REINT, mdt_reint),
4806 DEF_MDT_HNDL_F(HABEO_CORPUS, CLOSE, mdt_close),
4807 DEF_MDT_HNDL_F(HABEO_CORPUS, DONE_WRITING, mdt_done_writing),
4808 DEF_MDT_HNDL_F(0 |HABEO_REFERO, PIN, mdt_pin),
4809 DEF_MDT_HNDL_0(0, SYNC, mdt_sync),
4810 DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, IS_SUBDIR, mdt_is_subdir),
4811 DEF_MDT_HNDL_F(0, QUOTACHECK, mdt_quotacheck_handle),
4812 DEF_MDT_HNDL_F(0, QUOTACTL, mdt_quotactl_handle)
4815 #define DEF_OBD_HNDL(flags, name, fn) \
4816 DEF_HNDL(OBD, PING, _NET, flags, name, fn, NULL)
4819 static struct mdt_handler mdt_obd_ops[] = {
4820 DEF_OBD_HNDL(0, PING, mdt_obd_ping),
4821 DEF_OBD_HNDL(0, LOG_CANCEL, mdt_obd_log_cancel),
4822 DEF_OBD_HNDL(0, QC_CALLBACK, mdt_obd_qc_callback)
4825 #define DEF_DLM_HNDL_0(flags, name, fn) \
4826 DEF_HNDL(LDLM, ENQUEUE, , flags, name, fn, NULL)
4827 #define DEF_DLM_HNDL_F(flags, name, fn) \
4828 DEF_HNDL(LDLM, ENQUEUE, , flags, name, fn, &RQF_LDLM_ ## name)
4830 static struct mdt_handler mdt_dlm_ops[] = {
4831 DEF_DLM_HNDL_F(HABEO_CLAVIS, ENQUEUE, mdt_enqueue),
4832 DEF_DLM_HNDL_0(HABEO_CLAVIS, CONVERT, mdt_convert),
4833 DEF_DLM_HNDL_0(0, BL_CALLBACK, mdt_bl_callback),
4834 DEF_DLM_HNDL_0(0, CP_CALLBACK, mdt_cp_callback)
4837 static struct mdt_handler mdt_llog_ops[] = {
4840 #define DEF_SEC_CTX_HNDL(name, fn) \
4841 DEF_HNDL(SEC_CTX, INIT, _NET, 0, name, fn, NULL)
4843 static struct mdt_handler mdt_sec_ctx_ops[] = {
4844 DEF_SEC_CTX_HNDL(INIT, mdt_sec_ctx_handle),
4845 DEF_SEC_CTX_HNDL(INIT_CONT, mdt_sec_ctx_handle),
4846 DEF_SEC_CTX_HNDL(FINI, mdt_sec_ctx_handle)
4849 static struct mdt_opc_slice mdt_regular_handlers[] = {
4851 .mos_opc_start = MDS_GETATTR,
4852 .mos_opc_end = MDS_LAST_OPC,
4853 .mos_hs = mdt_mds_ops
4856 .mos_opc_start = OBD_PING,
4857 .mos_opc_end = OBD_LAST_OPC,
4858 .mos_hs = mdt_obd_ops
4861 .mos_opc_start = LDLM_ENQUEUE,
4862 .mos_opc_end = LDLM_LAST_OPC,
4863 .mos_hs = mdt_dlm_ops
4866 .mos_opc_start = LLOG_ORIGIN_HANDLE_CREATE,
4867 .mos_opc_end = LLOG_LAST_OPC,
4868 .mos_hs = mdt_llog_ops
4871 .mos_opc_start = SEC_CTX_INIT,
4872 .mos_opc_end = SEC_LAST_OPC,
4873 .mos_hs = mdt_sec_ctx_ops
4880 static struct mdt_handler mdt_readpage_ops[] = {
4881 DEF_MDT_HNDL_F(0, CONNECT, mdt_connect),
4882 DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, READPAGE, mdt_readpage),
4883 #ifdef HAVE_SPLIT_SUPPORT
4884 DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, WRITEPAGE, mdt_writepage),
4888 * XXX: this is ugly and should be fixed one day, see mdc_close() for
4889 * detailed comments. --umka
4891 DEF_MDT_HNDL_F(HABEO_CORPUS, CLOSE, mdt_close),
4892 DEF_MDT_HNDL_F(HABEO_CORPUS, DONE_WRITING, mdt_done_writing),
4895 static struct mdt_opc_slice mdt_readpage_handlers[] = {
4897 .mos_opc_start = MDS_GETATTR,
4898 .mos_opc_end = MDS_LAST_OPC,
4899 .mos_hs = mdt_readpage_ops
4906 static struct mdt_handler mdt_xmds_ops[] = {
4907 DEF_MDT_HNDL_F(0, CONNECT, mdt_connect),
4908 DEF_MDT_HNDL_F(HABEO_CORPUS , GETATTR, mdt_getattr),
4909 DEF_MDT_HNDL_F(0 | MUTABOR , REINT, mdt_reint),
4910 DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, IS_SUBDIR, mdt_is_subdir),
4913 static struct mdt_opc_slice mdt_xmds_handlers[] = {
4915 .mos_opc_start = MDS_GETATTR,
4916 .mos_opc_end = MDS_LAST_OPC,
4917 .mos_hs = mdt_xmds_ops
4920 .mos_opc_start = OBD_PING,
4921 .mos_opc_end = OBD_LAST_OPC,
4922 .mos_hs = mdt_obd_ops
4925 .mos_opc_start = SEC_CTX_INIT,
4926 .mos_opc_end = SEC_LAST_OPC,
4927 .mos_hs = mdt_sec_ctx_ops
4934 static struct mdt_handler mdt_seq_ops[] = {
4935 DEF_SEQ_HNDL_F(0, QUERY, (int (*)(struct mdt_thread_info *))seq_query)
4938 static struct mdt_opc_slice mdt_seq_handlers[] = {
4940 .mos_opc_start = SEQ_QUERY,
4941 .mos_opc_end = SEQ_LAST_OPC,
4942 .mos_hs = mdt_seq_ops
4949 static struct mdt_handler mdt_fld_ops[] = {
4950 DEF_FLD_HNDL_F(0, QUERY, (int (*)(struct mdt_thread_info *))fld_query)
4953 static struct mdt_opc_slice mdt_fld_handlers[] = {
4955 .mos_opc_start = FLD_QUERY,
4956 .mos_opc_end = FLD_LAST_OPC,
4957 .mos_hs = mdt_fld_ops
4964 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
4965 MODULE_DESCRIPTION("Lustre Meta-data Target ("LUSTRE_MDT_NAME")");
4966 MODULE_LICENSE("GPL");
4968 CFS_MODULE_PARM(mdt_num_threads, "ul", ulong, 0444,
4969 "number of mdt service threads to start");
4971 cfs_module(mdt, "0.2.0", mdt_mod_init, mdt_mod_exit);