1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright 2008 Sun Microsystems, Inc. All rights reserved
30 * Use is subject to license terms.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/mdt/mdt_handler.c
38 * Lustre Metadata Target (mdt) request handler
40 * Author: Peter Braam <braam@clusterfs.com>
41 * Author: Andreas Dilger <adilger@clusterfs.com>
42 * Author: Phil Schwan <phil@clusterfs.com>
43 * Author: Mike Shaver <shaver@clusterfs.com>
44 * Author: Nikita Danilov <nikita@clusterfs.com>
45 * Author: Huang Hua <huanghua@clusterfs.com>
46 * Author: Yury Umanets <umka@clusterfs.com>
50 # define EXPORT_SYMTAB
52 #define DEBUG_SUBSYSTEM S_MDS
54 #include <linux/module.h>
56 * struct OBD_{ALLOC,FREE}*()
58 #include <obd_support.h>
59 /* struct ptlrpc_request */
60 #include <lustre_net.h>
61 /* struct obd_export */
62 #include <lustre_export.h>
63 /* struct obd_device */
66 #include <dt_object.h>
67 #include <lustre_mds.h>
68 #include <lustre_mdt.h>
69 #include "mdt_internal.h"
70 #ifdef HAVE_QUOTA_SUPPORT
71 # include <lustre_quota.h>
73 #include <lustre_acl.h>
74 #include <lustre_param.h>
75 #include <lustre_fsfilt.h>
77 mdl_mode_t mdt_mdl_lock_modes[] = {
78 [LCK_MINMODE] = MDL_MINMODE,
85 [LCK_GROUP] = MDL_GROUP
88 ldlm_mode_t mdt_dlm_lock_modes[] = {
89 [MDL_MINMODE] = LCK_MINMODE,
96 [MDL_GROUP] = LCK_GROUP
100 * Initialized in mdt_mod_init().
102 static unsigned long mdt_num_threads;
103 static unsigned long mdt_min_threads;
104 static unsigned long mdt_max_threads;
106 /* ptlrpc request handler for MDT. All handlers are
107 * grouped into several slices - struct mdt_opc_slice,
108 * and stored in an array - mdt_handlers[].
111 /* The name of this handler. */
113 /* Fail id for this handler, checked at the beginning of this handler*/
115 /* Operation code for this handler */
117 /* flags are listed in enum mdt_handler_flags below. */
119 /* The actual handler function to execute. */
120 int (*mh_act)(struct mdt_thread_info *info);
121 /* Request format for this request. */
122 const struct req_format *mh_fmt;
125 enum mdt_handler_flags {
127 * struct mdt_body is passed in the incoming message, and object
128 * identified by this fid exists on disk.
130 * "habeo corpus" == "I have a body"
132 HABEO_CORPUS = (1 << 0),
134 * struct ldlm_request is passed in the incoming message.
136 * "habeo clavis" == "I have a key"
138 HABEO_CLAVIS = (1 << 1),
140 * this request has fixed reply format, so that reply message can be
141 * packed by generic code.
143 * "habeo refero" == "I have a reply"
145 HABEO_REFERO = (1 << 2),
147 * this request will modify something, so check whether the filesystem
148 * is readonly or not, then return -EROFS to client asap if necessary.
150 * "mutabor" == "I shall modify"
155 struct mdt_opc_slice {
158 struct mdt_handler *mos_hs;
161 static struct mdt_opc_slice mdt_regular_handlers[];
162 static struct mdt_opc_slice mdt_readpage_handlers[];
163 static struct mdt_opc_slice mdt_xmds_handlers[];
164 static struct mdt_opc_slice mdt_seq_handlers[];
165 static struct mdt_opc_slice mdt_fld_handlers[];
167 static struct mdt_device *mdt_dev(struct lu_device *d);
168 static int mdt_regular_handle(struct ptlrpc_request *req);
169 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags);
170 static int mdt_fid2path(const struct lu_env *env, struct mdt_device *mdt,
171 struct getinfo_fid2path *fp);
173 static const struct lu_object_operations mdt_obj_ops;
175 int mdt_get_disposition(struct ldlm_reply *rep, int flag)
179 return (rep->lock_policy_res1 & flag);
182 void mdt_clear_disposition(struct mdt_thread_info *info,
183 struct ldlm_reply *rep, int flag)
186 info->mti_opdata &= ~flag;
188 rep->lock_policy_res1 &= ~flag;
191 void mdt_set_disposition(struct mdt_thread_info *info,
192 struct ldlm_reply *rep, int flag)
195 info->mti_opdata |= flag;
197 rep->lock_policy_res1 |= flag;
200 void mdt_lock_reg_init(struct mdt_lock_handle *lh, ldlm_mode_t lm)
202 lh->mlh_pdo_hash = 0;
203 lh->mlh_reg_mode = lm;
204 lh->mlh_type = MDT_REG_LOCK;
207 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, ldlm_mode_t lm,
208 const char *name, int namelen)
210 lh->mlh_reg_mode = lm;
211 lh->mlh_type = MDT_PDO_LOCK;
214 LASSERT(namelen > 0);
215 lh->mlh_pdo_hash = full_name_hash(name, namelen);
217 LASSERT(namelen == 0);
218 lh->mlh_pdo_hash = 0ull;
222 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
223 struct mdt_lock_handle *lh)
229 * Any dir access needs couple of locks:
231 * 1) on part of dir we gonna take lookup/modify;
233 * 2) on whole dir to protect it from concurrent splitting and/or to
234 * flush client's cache for readdir().
236 * so, for a given mode and object this routine decides what lock mode
237 * to use for lock #2:
239 * 1) if caller's gonna lookup in dir then we need to protect dir from
240 * being splitted only - LCK_CR
242 * 2) if caller's gonna modify dir then we need to protect dir from
243 * being splitted and to flush cache - LCK_CW
245 * 3) if caller's gonna modify dir and that dir seems ready for
246 * splitting then we need to protect it from any type of access
247 * (lookup/modify/split) - LCK_EX --bzzz
250 LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
251 LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
254 * Ask underlaying level its opinion about preferable PDO lock mode
255 * having access type passed as regular lock mode:
257 * - MDL_MINMODE means that lower layer does not want to specify lock
260 * - MDL_NL means that no PDO lock should be taken. This is used in some
261 * cases. Say, for non-splittable directories no need to use PDO locks
264 mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
265 mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
267 if (mode != MDL_MINMODE) {
268 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
271 * Lower layer does not want to specify locking mode. We do it
272 * our selves. No special protection is needed, just flush
273 * client's cache on modification and allow concurrent
276 switch (lh->mlh_reg_mode) {
278 lh->mlh_pdo_mode = LCK_EX;
281 lh->mlh_pdo_mode = LCK_CR;
284 lh->mlh_pdo_mode = LCK_CW;
287 CERROR("Not expected lock type (0x%x)\n",
288 (int)lh->mlh_reg_mode);
293 LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
297 static int mdt_getstatus(struct mdt_thread_info *info)
299 struct mdt_device *mdt = info->mti_mdt;
300 struct md_device *next = mdt->mdt_child;
301 struct mdt_body *repbody;
306 rc = mdt_check_ucred(info);
308 RETURN(err_serious(rc));
310 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETSTATUS_PACK))
311 RETURN(err_serious(-ENOMEM));
313 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
314 rc = next->md_ops->mdo_root_get(info->mti_env, next, &repbody->fid1);
318 repbody->valid |= OBD_MD_FLID;
320 if (mdt->mdt_opts.mo_mds_capa &&
321 info->mti_exp->exp_connect_flags & OBD_CONNECT_MDS_CAPA) {
322 struct mdt_object *root;
323 struct lustre_capa *capa;
325 root = mdt_object_find(info->mti_env, mdt, &repbody->fid1);
327 RETURN(PTR_ERR(root));
329 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA1);
331 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
332 rc = mo_capa_get(info->mti_env, mdt_object_child(root), capa,
334 mdt_object_put(info->mti_env, root);
336 repbody->valid |= OBD_MD_FLMDSCAPA;
342 static int mdt_statfs(struct mdt_thread_info *info)
344 struct md_device *next = info->mti_mdt->mdt_child;
345 struct ptlrpc_service *svc;
346 struct obd_statfs *osfs;
351 svc = info->mti_pill->rc_req->rq_rqbd->rqbd_service;
353 /* This will trigger a watchdog timeout */
354 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
355 (MDT_SERVICE_WATCHDOG_FACTOR *
356 at_get(&svc->srv_at_estimate)) + 1);
358 rc = mdt_check_ucred(info);
360 RETURN(err_serious(rc));
362 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK)) {
363 rc = err_serious(-ENOMEM);
365 osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
366 rc = next->md_ops->mdo_statfs(info->mti_env, next,
368 statfs_pack(osfs, &info->mti_u.ksfs);
374 * Pack SOM attributes into the reply.
375 * Call under a DLM UPDATE lock.
377 static void mdt_pack_size2body(struct mdt_thread_info *info,
378 struct mdt_object *mo)
381 struct md_attr *ma = &info->mti_attr;
383 LASSERT(ma->ma_attr.la_valid & LA_MODE);
384 b = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
386 /* Check if Size-on-MDS is supported, if this is a regular file,
387 * if SOM is enabled on the object and if SOM cache exists and valid.
388 * Otherwise do not pack Size-on-MDS attributes to the reply. */
389 if (!(mdt_conn_flags(info) & OBD_CONNECT_SOM) ||
390 !S_ISREG(ma->ma_attr.la_mode) ||
391 !mdt_object_is_som_enabled(mo) ||
392 !(ma->ma_valid & MA_SOM))
395 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
396 b->size = ma->ma_som->msd_size;
397 b->blocks = ma->ma_som->msd_blocks;
400 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
401 const struct lu_attr *attr, const struct lu_fid *fid)
403 struct md_attr *ma = &info->mti_attr;
405 LASSERT(ma->ma_valid & MA_INODE);
407 b->atime = attr->la_atime;
408 b->mtime = attr->la_mtime;
409 b->ctime = attr->la_ctime;
410 b->mode = attr->la_mode;
411 b->size = attr->la_size;
412 b->blocks = attr->la_blocks;
413 b->uid = attr->la_uid;
414 b->gid = attr->la_gid;
415 b->flags = attr->la_flags;
416 b->nlink = attr->la_nlink;
417 b->rdev = attr->la_rdev;
419 /*XXX should pack the reply body according to lu_valid*/
420 b->valid |= OBD_MD_FLCTIME | OBD_MD_FLUID |
421 OBD_MD_FLGID | OBD_MD_FLTYPE |
422 OBD_MD_FLMODE | OBD_MD_FLNLINK | OBD_MD_FLFLAGS |
423 OBD_MD_FLATIME | OBD_MD_FLMTIME ;
425 if (!S_ISREG(attr->la_mode)) {
426 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
427 } else if (ma->ma_need & MA_LOV && ma->ma_lmm_size == 0) {
428 /* means no objects are allocated on osts. */
429 LASSERT(!(ma->ma_valid & MA_LOV));
430 /* just ignore blocks occupied by extend attributes on MDS */
432 /* if no object is allocated on osts, the size on mds is valid. b=22272 */
433 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
438 b->valid |= OBD_MD_FLID;
440 /* FIXME: these should be fixed when new igif ready.*/
441 b->ino = fid_oid(fid); /* 1.6 compatibility */
442 b->generation = fid_ver(fid); /* 1.6 compatibility */
443 b->valid |= OBD_MD_FLGENER; /* 1.6 compatibility */
445 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, size="LPU64"\n",
446 PFID(fid), b->nlink, b->mode, b->size);
450 mdt_body_reverse_idmap(info, b);
453 static inline int mdt_body_has_lov(const struct lu_attr *la,
454 const struct mdt_body *body)
456 return ((S_ISREG(la->la_mode) && (body->valid & OBD_MD_FLEASIZE)) ||
457 (S_ISDIR(la->la_mode) && (body->valid & OBD_MD_FLDIREA )) );
460 static int mdt_getattr_internal(struct mdt_thread_info *info,
461 struct mdt_object *o, int ma_need)
463 struct md_object *next = mdt_object_child(o);
464 const struct mdt_body *reqbody = info->mti_body;
465 struct ptlrpc_request *req = mdt_info_req(info);
466 struct md_attr *ma = &info->mti_attr;
467 struct lu_attr *la = &ma->ma_attr;
468 struct req_capsule *pill = info->mti_pill;
469 const struct lu_env *env = info->mti_env;
470 struct mdt_body *repbody;
471 struct lu_buf *buffer = &info->mti_buf;
475 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
476 RETURN(err_serious(-ENOMEM));
478 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
482 rc = mdt_object_exists(o);
484 /* This object is located on remote node.*/
485 repbody->fid1 = *mdt_object_fid(o);
486 repbody->valid = OBD_MD_FLID | OBD_MD_MDS;
490 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
491 buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD, RCL_SERVER);
493 /* If it is dir object and client require MEA, then we got MEA */
494 if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
495 reqbody->valid & OBD_MD_MEA) {
496 /* Assumption: MDT_MD size is enough for lmv size. */
497 ma->ma_lmv = buffer->lb_buf;
498 ma->ma_lmv_size = buffer->lb_len;
499 ma->ma_need = MA_LMV | MA_INODE;
501 ma->ma_lmm = buffer->lb_buf;
502 ma->ma_lmm_size = buffer->lb_len;
503 ma->ma_need = MA_LOV | MA_INODE;
506 if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
507 reqbody->valid & OBD_MD_FLDIREA &&
508 lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
509 /* get default stripe info for this dir. */
510 ma->ma_need |= MA_LOV_DEF;
512 ma->ma_need |= ma_need;
513 if (ma->ma_need & MA_SOM)
514 ma->ma_som = &info->mti_u.som.data;
516 rc = mo_attr_get(env, next, ma);
518 CERROR("getattr error for "DFID": %d\n",
519 PFID(mdt_object_fid(o)), rc);
523 if (likely(ma->ma_valid & MA_INODE))
524 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
528 if (mdt_body_has_lov(la, reqbody)) {
529 if (ma->ma_valid & MA_LOV) {
530 LASSERT(ma->ma_lmm_size);
531 mdt_dump_lmm(D_INFO, ma->ma_lmm);
532 repbody->eadatasize = ma->ma_lmm_size;
533 if (S_ISDIR(la->la_mode))
534 repbody->valid |= OBD_MD_FLDIREA;
536 repbody->valid |= OBD_MD_FLEASIZE;
538 if (ma->ma_valid & MA_LMV) {
539 LASSERT(S_ISDIR(la->la_mode));
540 repbody->eadatasize = ma->ma_lmv_size;
541 repbody->valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
543 } else if (S_ISLNK(la->la_mode) &&
544 reqbody->valid & OBD_MD_LINKNAME) {
545 buffer->lb_buf = ma->ma_lmm;
546 buffer->lb_len = reqbody->eadatasize;
547 rc = mo_readlink(env, next, buffer);
548 if (unlikely(rc <= 0)) {
549 CERROR("readlink failed: %d\n", rc);
552 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
554 repbody->valid |= OBD_MD_LINKNAME;
555 repbody->eadatasize = rc;
557 ((char*)ma->ma_lmm)[rc - 1] = 0;
558 CDEBUG(D_INODE, "symlink dest %s, len = %d\n",
559 (char*)ma->ma_lmm, rc);
564 if (reqbody->valid & OBD_MD_FLMODEASIZE) {
565 repbody->max_cookiesize = info->mti_mdt->mdt_max_cookiesize;
566 repbody->max_mdsize = info->mti_mdt->mdt_max_mdsize;
567 repbody->valid |= OBD_MD_FLMODEASIZE;
568 CDEBUG(D_INODE, "I am going to change the MAX_MD_SIZE & "
569 "MAX_COOKIE to : %d:%d\n", repbody->max_mdsize,
570 repbody->max_cookiesize);
573 if (exp_connect_rmtclient(info->mti_exp) &&
574 reqbody->valid & OBD_MD_FLRMTPERM) {
575 void *buf = req_capsule_server_get(pill, &RMF_ACL);
577 /* mdt_getattr_lock only */
578 rc = mdt_pack_remote_perm(info, o, buf);
580 repbody->valid &= ~OBD_MD_FLRMTPERM;
581 repbody->aclsize = 0;
584 repbody->valid |= OBD_MD_FLRMTPERM;
585 repbody->aclsize = sizeof(struct mdt_remote_perm);
588 #ifdef CONFIG_FS_POSIX_ACL
589 else if ((req->rq_export->exp_connect_flags & OBD_CONNECT_ACL) &&
590 (reqbody->valid & OBD_MD_FLACL)) {
591 buffer->lb_buf = req_capsule_server_get(pill, &RMF_ACL);
592 buffer->lb_len = req_capsule_get_size(pill,
593 &RMF_ACL, RCL_SERVER);
594 if (buffer->lb_len > 0) {
595 rc = mo_xattr_get(env, next, buffer,
596 XATTR_NAME_ACL_ACCESS);
598 if (rc == -ENODATA) {
599 repbody->aclsize = 0;
600 repbody->valid |= OBD_MD_FLACL;
602 } else if (rc == -EOPNOTSUPP) {
605 CERROR("got acl size: %d\n", rc);
608 repbody->aclsize = rc;
609 repbody->valid |= OBD_MD_FLACL;
616 if (reqbody->valid & OBD_MD_FLMDSCAPA &&
617 info->mti_mdt->mdt_opts.mo_mds_capa &&
618 info->mti_exp->exp_connect_flags & OBD_CONNECT_MDS_CAPA) {
619 struct lustre_capa *capa;
621 capa = req_capsule_server_get(pill, &RMF_CAPA1);
623 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
624 rc = mo_capa_get(env, next, capa, 0);
627 repbody->valid |= OBD_MD_FLMDSCAPA;
632 static int mdt_renew_capa(struct mdt_thread_info *info)
634 struct mdt_object *obj = info->mti_object;
635 struct mdt_body *body;
636 struct lustre_capa *capa, *c;
640 /* if object doesn't exist, or server has disabled capability,
641 * return directly, client will find body->valid OBD_MD_FLOSSCAPA
644 if (!obj || !info->mti_mdt->mdt_opts.mo_oss_capa ||
645 !(info->mti_exp->exp_connect_flags & OBD_CONNECT_OSS_CAPA))
648 body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
649 LASSERT(body != NULL);
651 c = req_capsule_client_get(info->mti_pill, &RMF_CAPA1);
654 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
658 rc = mo_capa_get(info->mti_env, mdt_object_child(obj), capa, 1);
660 body->valid |= OBD_MD_FLOSSCAPA;
664 static int mdt_getattr(struct mdt_thread_info *info)
666 struct mdt_object *obj = info->mti_object;
667 struct req_capsule *pill = info->mti_pill;
668 struct mdt_body *reqbody;
669 struct mdt_body *repbody;
675 reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
678 if (reqbody->valid & OBD_MD_FLOSSCAPA) {
679 rc = req_capsule_server_pack(pill);
681 RETURN(err_serious(rc));
682 rc = mdt_renew_capa(info);
683 GOTO(out_shrink, rc);
686 LASSERT(obj != NULL);
687 LASSERT(lu_object_assert_exists(&obj->mot_obj.mo_lu));
689 mode = lu_object_attr(&obj->mot_obj.mo_lu);
690 if (S_ISLNK(mode) && (reqbody->valid & OBD_MD_LINKNAME) &&
691 (reqbody->eadatasize > info->mti_mdt->mdt_max_mdsize))
692 md_size = reqbody->eadatasize;
694 md_size = info->mti_mdt->mdt_max_mdsize;
696 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, md_size);
698 rc = req_capsule_server_pack(pill);
699 if (unlikely(rc != 0))
700 RETURN(err_serious(rc));
702 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
703 LASSERT(repbody != NULL);
704 repbody->eadatasize = 0;
705 repbody->aclsize = 0;
707 if (reqbody->valid & OBD_MD_FLRMTPERM)
708 rc = mdt_init_ucred(info, reqbody);
710 rc = mdt_check_ucred(info);
712 GOTO(out_shrink, rc);
714 info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
715 info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
718 * Don't check capability at all, because rename might getattr for
719 * remote obj, and at that time no capability is available.
721 mdt_set_capainfo(info, 1, &reqbody->fid1, BYPASS_CAPA);
722 rc = mdt_getattr_internal(info, obj, 0);
723 if (reqbody->valid & OBD_MD_FLRMTPERM)
724 mdt_exit_ucred(info);
727 mdt_shrink_reply(info);
731 static int mdt_is_subdir(struct mdt_thread_info *info)
733 struct mdt_object *o = info->mti_object;
734 struct req_capsule *pill = info->mti_pill;
735 const struct mdt_body *body = info->mti_body;
736 struct mdt_body *repbody;
742 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
745 * We save last checked parent fid to @repbody->fid1 for remote
748 LASSERT(fid_is_sane(&body->fid2));
749 LASSERT(mdt_object_exists(o) > 0);
750 rc = mdo_is_subdir(info->mti_env, mdt_object_child(o),
751 &body->fid2, &repbody->fid1);
752 if (rc == 0 || rc == -EREMOTE)
753 repbody->valid |= OBD_MD_FLID;
758 static int mdt_raw_lookup(struct mdt_thread_info *info,
759 struct mdt_object *parent,
760 const struct lu_name *lname,
761 struct ldlm_reply *ldlm_rep)
763 struct md_object *next = mdt_object_child(info->mti_object);
764 const struct mdt_body *reqbody = info->mti_body;
765 struct lu_fid *child_fid = &info->mti_tmp_fid1;
766 struct mdt_body *repbody;
770 if (reqbody->valid != OBD_MD_FLID)
773 LASSERT(!info->mti_cross_ref);
775 /* Only got the fid of this obj by name */
776 rc = mdo_lookup(info->mti_env, next, lname, child_fid,
779 /* XXX is raw_lookup possible as intent operation? */
782 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
785 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
787 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
790 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
791 repbody->fid1 = *child_fid;
792 repbody->valid = OBD_MD_FLID;
798 * UPDATE lock should be taken against parent, and be release before exit;
799 * child_bits lock should be taken against child, and be returned back:
800 * (1)normal request should release the child lock;
801 * (2)intent request will grant the lock to client.
803 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
804 struct mdt_lock_handle *lhc,
806 struct ldlm_reply *ldlm_rep)
808 struct ptlrpc_request *req = mdt_info_req(info);
809 struct mdt_body *reqbody = NULL;
810 struct mdt_object *parent = info->mti_object;
811 struct mdt_object *child;
812 struct md_object *next = mdt_object_child(parent);
813 struct lu_fid *child_fid = &info->mti_tmp_fid1;
814 struct lu_name *lname = NULL;
815 const char *name = NULL;
817 struct mdt_lock_handle *lhp;
818 struct ldlm_lock *lock;
819 struct ldlm_res_id *res_id;
826 is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
827 LASSERT(ergo(is_resent,
828 lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
830 LASSERT(parent != NULL);
831 name = req_capsule_client_get(info->mti_pill, &RMF_NAME);
833 RETURN(err_serious(-EFAULT));
835 namelen = req_capsule_get_size(info->mti_pill, &RMF_NAME,
837 if (!info->mti_cross_ref) {
839 * XXX: Check for "namelen == 0" is for getattr by fid
840 * (OBD_CONNECT_ATTRFID), otherwise do not allow empty name,
841 * that is the name must contain at least one character and
842 * the terminating '\0'
845 reqbody = req_capsule_client_get(info->mti_pill,
847 LASSERT(fid_is_sane(&reqbody->fid2));
850 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
852 PFID(mdt_object_fid(parent)), PFID(&reqbody->fid2),
855 lname = mdt_name(info->mti_env, (char *)name, namelen);
856 CDEBUG(D_INODE, "getattr with lock for "DFID"/%s, "
857 "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
861 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
863 rc = mdt_object_exists(parent);
864 if (unlikely(rc == 0)) {
865 LU_OBJECT_DEBUG(D_WARNING, info->mti_env,
866 &parent->mot_obj.mo_lu,
867 "Parent doesn't exist!\n");
869 } else if (!info->mti_cross_ref) {
870 LASSERTF(rc > 0, "Parent "DFID" is on remote server\n",
871 PFID(mdt_object_fid(parent)));
874 rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
882 if (info->mti_cross_ref) {
883 /* Only getattr on the child. Parent is on another node. */
884 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
886 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
887 "ldlm_rep=%p\n", PFID(mdt_object_fid(child)), ldlm_rep);
890 /* Do not take lock for resent case. */
891 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
892 LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
893 lhc->mlh_reg_lh.cookie);
894 LASSERT(fid_res_name_eq(mdt_object_fid(child),
895 &lock->l_resource->lr_name));
899 mdt_lock_handle_init(lhc);
900 mdt_lock_reg_init(lhc, LCK_PR);
903 * Object's name is on another MDS, no lookup lock is
904 * needed here but update is.
906 child_bits &= ~MDS_INODELOCK_LOOKUP;
907 child_bits |= MDS_INODELOCK_UPDATE;
909 rc = mdt_object_lock(info, child, lhc, child_bits,
913 /* Finally, we can get attr for child. */
914 mdt_set_capainfo(info, 0, mdt_object_fid(child),
916 rc = mdt_getattr_internal(info, child, 0);
917 if (unlikely(rc != 0))
918 mdt_object_unlock(info, child, lhc, 1);
923 /* step 1: lock parent */
924 lhp = &info->mti_lh[MDT_LH_PARENT];
925 mdt_lock_pdo_init(lhp, LCK_PR, name, namelen);
926 rc = mdt_object_lock(info, parent, lhp, MDS_INODELOCK_UPDATE,
929 if (unlikely(rc != 0))
933 /* step 2: lookup child's fid by name */
934 rc = mdo_lookup(info->mti_env, next, lname, child_fid,
939 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
940 GOTO(out_parent, rc);
942 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
944 *child_fid = reqbody->fid2;
945 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
949 *step 3: find the child object by fid & lock it.
950 * regardless if it is local or remote.
952 child = mdt_object_find(info->mti_env, info->mti_mdt, child_fid);
954 if (unlikely(IS_ERR(child)))
955 GOTO(out_parent, rc = PTR_ERR(child));
957 /* Do not take lock for resent case. */
958 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
959 LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
960 lhc->mlh_reg_lh.cookie);
962 res_id = &lock->l_resource->lr_name;
963 if (!fid_res_name_eq(mdt_object_fid(child),
964 &lock->l_resource->lr_name)) {
965 LASSERTF(fid_res_name_eq(mdt_object_fid(parent),
966 &lock->l_resource->lr_name),
967 "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n",
968 (unsigned long)res_id->name[0],
969 (unsigned long)res_id->name[1],
970 (unsigned long)res_id->name[2],
971 PFID(mdt_object_fid(parent)));
972 CWARN("Although resent, but still not get child lock"
973 "parent:"DFID" child:"DFID"\n",
974 PFID(mdt_object_fid(parent)),
975 PFID(mdt_object_fid(child)));
976 lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
985 ma = &info->mti_attr;
987 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout*2);
988 mdt_lock_handle_init(lhc);
989 mdt_lock_reg_init(lhc, LCK_PR);
991 if (mdt_object_exists(child) == 0) {
992 LU_OBJECT_DEBUG(D_WARNING, info->mti_env,
993 &child->mot_obj.mo_lu,
994 "Object doesn't exist!\n");
995 GOTO(out_child, rc = -ESTALE);
999 ma->ma_need = MA_INODE;
1000 rc = mo_attr_get(info->mti_env, next, ma);
1001 if (unlikely(rc != 0))
1002 GOTO(out_child, rc);
1004 /* If the file has not been changed for some time, we return
1005 * not only a LOOKUP lock, but also an UPDATE lock and this
1006 * might save us RPC on later STAT. For directories, it also
1007 * let negative dentry starts working for this dir. */
1008 if (ma->ma_valid & MA_INODE &&
1009 ma->ma_attr.la_valid & LA_CTIME &&
1010 info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
1011 ma->ma_attr.la_ctime < cfs_time_current_sec())
1012 child_bits |= MDS_INODELOCK_UPDATE;
1014 rc = mdt_object_lock(info, child, lhc, child_bits,
1017 if (unlikely(rc != 0))
1018 GOTO(out_child, rc);
1021 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1022 /* Get MA_SOM attributes if update lock is given. */
1024 lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_UPDATE &&
1025 S_ISREG(lu_object_attr(&mdt_object_child(child)->mo_lu)))
1028 /* finally, we can get attr for child. */
1029 mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
1030 rc = mdt_getattr_internal(info, child, ma_need);
1031 if (unlikely(rc != 0)) {
1032 mdt_object_unlock(info, child, lhc, 1);
1034 /* Debugging code. */
1035 res_id = &lock->l_resource->lr_name;
1036 LDLM_DEBUG(lock, "Returning lock to client");
1037 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1038 &lock->l_resource->lr_name),
1039 "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n",
1040 (unsigned long)res_id->name[0],
1041 (unsigned long)res_id->name[1],
1042 (unsigned long)res_id->name[2],
1043 PFID(mdt_object_fid(child)));
1044 mdt_pack_size2body(info, child);
1047 LDLM_LOCK_PUT(lock);
1051 mdt_object_put(info->mti_env, child);
1053 mdt_object_unlock(info, parent, lhp, 1);
1057 /* normal handler: should release the child lock */
1058 static int mdt_getattr_name(struct mdt_thread_info *info)
1060 struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1061 struct mdt_body *reqbody;
1062 struct mdt_body *repbody;
1066 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1067 LASSERT(reqbody != NULL);
1068 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1069 LASSERT(repbody != NULL);
1071 info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
1072 info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
1073 repbody->eadatasize = 0;
1074 repbody->aclsize = 0;
1076 rc = mdt_init_ucred(info, reqbody);
1078 GOTO(out_shrink, rc);
1080 rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1081 if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1082 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1083 lhc->mlh_reg_lh.cookie = 0;
1085 mdt_exit_ucred(info);
1088 mdt_shrink_reply(info);
1092 static const struct lu_device_operations mdt_lu_ops;
1094 static int lu_device_is_mdt(struct lu_device *d)
1096 return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &mdt_lu_ops);
1099 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1100 void *karg, void *uarg);
1102 static int mdt_set_info(struct mdt_thread_info *info)
1104 struct ptlrpc_request *req = mdt_info_req(info);
1107 int keylen, vallen, rc = 0;
1110 rc = req_capsule_server_pack(info->mti_pill);
1114 key = req_capsule_client_get(info->mti_pill, &RMF_SETINFO_KEY);
1116 DEBUG_REQ(D_HA, req, "no set_info key");
1120 keylen = req_capsule_get_size(info->mti_pill, &RMF_SETINFO_KEY,
1123 val = req_capsule_client_get(info->mti_pill, &RMF_SETINFO_VAL);
1125 DEBUG_REQ(D_HA, req, "no set_info val");
1129 vallen = req_capsule_get_size(info->mti_pill, &RMF_SETINFO_VAL,
1132 /* Swab any part of val you need to here */
1133 if (KEY_IS(KEY_READ_ONLY)) {
1135 lustre_msg_set_status(req->rq_repmsg, 0);
1137 cfs_spin_lock(&req->rq_export->exp_lock);
1139 req->rq_export->exp_connect_flags |= OBD_CONNECT_RDONLY;
1141 req->rq_export->exp_connect_flags &=~OBD_CONNECT_RDONLY;
1142 cfs_spin_unlock(&req->rq_export->exp_lock);
1144 } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1145 struct changelog_setinfo *cs =
1146 (struct changelog_setinfo *)val;
1147 if (vallen != sizeof(*cs)) {
1148 CERROR("Bad changelog_clear setinfo size %d\n", vallen);
1151 if (ptlrpc_req_need_swab(req)) {
1152 __swab64s(&cs->cs_recno);
1153 __swab32s(&cs->cs_id);
1156 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, info->mti_exp,
1158 lustre_msg_set_status(req->rq_repmsg, rc);
1166 static int mdt_connect(struct mdt_thread_info *info)
1169 struct ptlrpc_request *req;
1171 req = mdt_info_req(info);
1172 rc = target_handle_connect(req);
1174 LASSERT(req->rq_export != NULL);
1175 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
1176 rc = mdt_init_sec_level(info);
1178 rc = mdt_init_idmap(info);
1180 obd_disconnect(class_export_get(req->rq_export));
1182 rc = err_serious(rc);
1187 static int mdt_disconnect(struct mdt_thread_info *info)
1192 rc = target_handle_disconnect(mdt_info_req(info));
1194 rc = err_serious(rc);
1198 static int mdt_sendpage(struct mdt_thread_info *info,
1199 struct lu_rdpg *rdpg)
1201 struct ptlrpc_request *req = mdt_info_req(info);
1202 struct obd_export *exp = req->rq_export;
1203 struct ptlrpc_bulk_desc *desc;
1204 struct l_wait_info *lwi = &info->mti_u.rdpg.mti_wait_info;
1212 desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, BULK_PUT_SOURCE,
1217 for (i = 0, tmpcount = rdpg->rp_count;
1218 i < rdpg->rp_npages; i++, tmpcount -= tmpsize) {
1219 tmpsize = min_t(int, tmpcount, CFS_PAGE_SIZE);
1220 ptlrpc_prep_bulk_page(desc, rdpg->rp_pages[i], 0, tmpsize);
1223 LASSERT(desc->bd_nob == rdpg->rp_count);
1224 rc = sptlrpc_svc_wrap_bulk(req, desc);
1226 GOTO(free_desc, rc);
1228 rc = ptlrpc_start_bulk_transfer(desc);
1230 GOTO(free_desc, rc);
1232 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1233 GOTO(abort_bulk, rc = 0);
1235 timeout = (int) req->rq_deadline - cfs_time_current_sec();
1237 CERROR("Req deadline already passed %lu (now: %lu)\n",
1238 req->rq_deadline, cfs_time_current_sec());
1239 *lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(max(timeout, 1)),
1240 cfs_time_seconds(1), NULL, NULL);
1241 rc = l_wait_event(desc->bd_waitq, !ptlrpc_server_bulk_active(desc) ||
1242 exp->exp_failed || exp->exp_abort_active_req, lwi);
1243 LASSERT (rc == 0 || rc == -ETIMEDOUT);
1246 if (desc->bd_success &&
1247 desc->bd_nob_transferred == rdpg->rp_count)
1248 GOTO(free_desc, rc);
1251 if (exp->exp_abort_active_req || exp->exp_failed)
1252 GOTO(abort_bulk, rc);
1255 DEBUG_REQ(D_ERROR, req, "bulk failed: %s %d(%d), evicting %s@%s",
1256 (rc == -ETIMEDOUT) ? "timeout" : "network error",
1257 desc->bd_nob_transferred, rdpg->rp_count,
1258 exp->exp_client_uuid.uuid,
1259 exp->exp_connection->c_remote_uuid.uuid);
1261 class_fail_export(exp);
1265 ptlrpc_abort_bulk(desc);
1267 ptlrpc_free_bulk(desc);
1271 #ifdef HAVE_SPLIT_SUPPORT
1273 * Retrieve dir entry from the page and insert it to the slave object, actually,
1274 * this should be in osd layer, but since it will not in the final product, so
1275 * just do it here and do not define more moo api anymore for this.
1277 static int mdt_write_dir_page(struct mdt_thread_info *info, struct page *page,
1280 struct mdt_object *object = info->mti_object;
1281 struct lu_fid *lf = &info->mti_tmp_fid2;
1282 struct md_attr *ma = &info->mti_attr;
1283 struct lu_dirpage *dp;
1284 struct lu_dirent *ent;
1285 int rc = 0, offset = 0;
1288 /* Make sure we have at least one entry. */
1293 * Disable trans for this name insert, since it will include many trans
1296 info->mti_no_need_trans = 1;
1298 * When write_dir_page, no need update parent's ctime,
1299 * and no permission check for name_insert.
1301 ma->ma_attr.la_ctime = 0;
1302 ma->ma_attr.la_valid = LA_MODE;
1303 ma->ma_valid = MA_INODE;
1306 dp = page_address(page);
1307 offset = (int)((__u32)lu_dirent_start(dp) - (__u32)dp);
1309 for (ent = lu_dirent_start(dp); ent != NULL;
1310 ent = lu_dirent_next(ent)) {
1311 struct lu_name *lname;
1314 if (le16_to_cpu(ent->lde_namelen) == 0)
1317 fid_le_to_cpu(lf, &ent->lde_fid);
1318 if (le64_to_cpu(ent->lde_hash) & MAX_HASH_HIGHEST_BIT)
1319 ma->ma_attr.la_mode = S_IFDIR;
1321 ma->ma_attr.la_mode = 0;
1322 OBD_ALLOC(name, le16_to_cpu(ent->lde_namelen) + 1);
1324 GOTO(out, rc = -ENOMEM);
1326 memcpy(name, ent->lde_name, le16_to_cpu(ent->lde_namelen));
1327 lname = mdt_name(info->mti_env, name,
1328 le16_to_cpu(ent->lde_namelen));
1329 ma->ma_attr_flags |= (MDS_PERM_BYPASS | MDS_QUOTA_IGNORE);
1330 rc = mdo_name_insert(info->mti_env,
1331 md_object_next(&object->mot_obj),
1333 OBD_FREE(name, le16_to_cpu(ent->lde_namelen) + 1);
1335 CERROR("Can't insert %*.*s, rc %d\n",
1336 le16_to_cpu(ent->lde_namelen),
1337 le16_to_cpu(ent->lde_namelen),
1342 offset += lu_dirent_size(ent);
1352 static int mdt_bulk_timeout(void *data)
1356 CERROR("mdt bulk transfer timeout \n");
1361 static int mdt_writepage(struct mdt_thread_info *info)
1363 struct ptlrpc_request *req = mdt_info_req(info);
1364 struct mdt_body *reqbody;
1365 struct l_wait_info *lwi;
1366 struct ptlrpc_bulk_desc *desc;
1372 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1373 if (reqbody == NULL)
1374 RETURN(err_serious(-EFAULT));
1376 desc = ptlrpc_prep_bulk_exp(req, 1, BULK_GET_SINK, MDS_BULK_PORTAL);
1378 RETURN(err_serious(-ENOMEM));
1380 /* allocate the page for the desc */
1381 page = cfs_alloc_page(CFS_ALLOC_STD);
1383 GOTO(desc_cleanup, rc = -ENOMEM);
1385 CDEBUG(D_INFO, "Received page offset %d size %d \n",
1386 (int)reqbody->size, (int)reqbody->nlink);
1388 ptlrpc_prep_bulk_page(desc, page, (int)reqbody->size,
1389 (int)reqbody->nlink);
1391 rc = sptlrpc_svc_prep_bulk(req, desc);
1393 GOTO(cleanup_page, rc);
1395 * Check if client was evicted while we were doing i/o before touching
1400 GOTO(cleanup_page, rc = -ENOMEM);
1402 if (desc->bd_export->exp_failed)
1405 rc = ptlrpc_start_bulk_transfer (desc);
1407 *lwi = LWI_TIMEOUT_INTERVAL(obd_timeout * CFS_HZ / 4, CFS_HZ,
1408 mdt_bulk_timeout, desc);
1409 rc = l_wait_event(desc->bd_waitq, !ptlrpc_bulk_active(desc) ||
1410 desc->bd_export->exp_failed, lwi);
1411 LASSERT(rc == 0 || rc == -ETIMEDOUT);
1412 if (rc == -ETIMEDOUT) {
1413 DEBUG_REQ(D_ERROR, req, "timeout on bulk GET");
1414 ptlrpc_abort_bulk(desc);
1415 } else if (desc->bd_export->exp_failed) {
1416 DEBUG_REQ(D_ERROR, req, "Eviction on bulk GET");
1418 ptlrpc_abort_bulk(desc);
1419 } else if (!desc->bd_success ||
1420 desc->bd_nob_transferred != desc->bd_nob) {
1421 DEBUG_REQ(D_ERROR, req, "%s bulk GET %d(%d)",
1423 "truncated" : "network error on",
1424 desc->bd_nob_transferred, desc->bd_nob);
1425 /* XXX should this be a different errno? */
1429 DEBUG_REQ(D_ERROR, req, "ptlrpc_bulk_get failed: rc %d", rc);
1432 GOTO(cleanup_lwi, rc);
1433 rc = mdt_write_dir_page(info, page, reqbody->nlink);
1438 cfs_free_page(page);
1440 ptlrpc_free_bulk(desc);
1445 static int mdt_readpage(struct mdt_thread_info *info)
1447 struct mdt_object *object = info->mti_object;
1448 struct lu_rdpg *rdpg = &info->mti_u.rdpg.mti_rdpg;
1449 struct mdt_body *reqbody;
1450 struct mdt_body *repbody;
1455 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
1456 RETURN(err_serious(-ENOMEM));
1458 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1459 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1460 if (reqbody == NULL || repbody == NULL)
1461 RETURN(err_serious(-EFAULT));
1464 * prepare @rdpg before calling lower layers and transfer itself. Here
1465 * reqbody->size contains offset of where to start to read and
1466 * reqbody->nlink contains number bytes to read.
1468 rdpg->rp_hash = reqbody->size;
1469 if (rdpg->rp_hash != reqbody->size) {
1470 CERROR("Invalid hash: "LPX64" != "LPX64"\n",
1471 rdpg->rp_hash, reqbody->size);
1475 rdpg->rp_attrs = reqbody->mode;
1476 rdpg->rp_count = reqbody->nlink;
1477 rdpg->rp_npages = (rdpg->rp_count + CFS_PAGE_SIZE - 1)>>CFS_PAGE_SHIFT;
1478 OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1479 if (rdpg->rp_pages == NULL)
1482 for (i = 0; i < rdpg->rp_npages; ++i) {
1483 rdpg->rp_pages[i] = cfs_alloc_page(CFS_ALLOC_STD);
1484 if (rdpg->rp_pages[i] == NULL)
1485 GOTO(free_rdpg, rc = -ENOMEM);
1488 /* call lower layers to fill allocated pages with directory data */
1489 rc = mo_readpage(info->mti_env, mdt_object_child(object), rdpg);
1491 GOTO(free_rdpg, rc);
1493 /* send pages to client */
1494 rc = mdt_sendpage(info, rdpg);
1499 for (i = 0; i < rdpg->rp_npages; i++)
1500 if (rdpg->rp_pages[i] != NULL)
1501 cfs_free_page(rdpg->rp_pages[i]);
1502 OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1504 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1510 static int mdt_reint_internal(struct mdt_thread_info *info,
1511 struct mdt_lock_handle *lhc,
1514 struct req_capsule *pill = info->mti_pill;
1515 struct mdt_device *mdt = info->mti_mdt;
1516 struct md_quota *mq = md_quota(info->mti_env);
1517 struct mdt_body *repbody;
1522 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1523 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1524 mdt->mdt_max_mdsize);
1525 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1526 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER,
1527 mdt->mdt_max_cookiesize);
1529 rc = req_capsule_server_pack(pill);
1531 CERROR("Can't pack response, rc %d\n", rc);
1532 RETURN(err_serious(rc));
1535 if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
1536 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1538 repbody->eadatasize = 0;
1539 repbody->aclsize = 0;
1542 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNPACK))
1543 GOTO(out_shrink, rc = err_serious(-EFAULT));
1545 rc = mdt_reint_unpack(info, op);
1547 CERROR("Can't unpack reint, rc %d\n", rc);
1548 GOTO(out_shrink, rc = err_serious(rc));
1551 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
1553 /* for replay no cookkie / lmm need, because client have this already */
1554 if (info->mti_spec.no_create == 1) {
1555 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1556 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
1558 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1559 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER,
1563 rc = mdt_init_ucred_reint(info);
1565 GOTO(out_shrink, rc);
1567 rc = mdt_fix_attr_ucred(info, op);
1569 GOTO(out_ucred, rc = err_serious(rc));
1571 if (mdt_check_resent(info, mdt_reconstruct, lhc)) {
1572 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
1573 GOTO(out_ucred, rc);
1575 mq->mq_exp = info->mti_exp;
1576 rc = mdt_reint_rec(info, lhc);
1579 mdt_exit_ucred(info);
1581 mdt_shrink_reply(info);
1585 static long mdt_reint_opcode(struct mdt_thread_info *info,
1586 const struct req_format **fmt)
1588 struct mdt_rec_reint *rec;
1591 opc = err_serious(-EFAULT);
1592 rec = req_capsule_client_get(info->mti_pill, &RMF_REC_REINT);
1594 opc = rec->rr_opcode;
1595 DEBUG_REQ(D_INODE, mdt_info_req(info), "reint opt = %ld", opc);
1596 if (opc < REINT_MAX && fmt[opc] != NULL)
1597 req_capsule_extend(info->mti_pill, fmt[opc]);
1599 CERROR("Unsupported opc: %ld\n", opc);
1600 opc = err_serious(opc);
1606 static int mdt_reint(struct mdt_thread_info *info)
1611 static const struct req_format *reint_fmts[REINT_MAX] = {
1612 [REINT_SETATTR] = &RQF_MDS_REINT_SETATTR,
1613 [REINT_CREATE] = &RQF_MDS_REINT_CREATE,
1614 [REINT_LINK] = &RQF_MDS_REINT_LINK,
1615 [REINT_UNLINK] = &RQF_MDS_REINT_UNLINK,
1616 [REINT_RENAME] = &RQF_MDS_REINT_RENAME,
1617 [REINT_OPEN] = &RQF_MDS_REINT_OPEN,
1618 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR
1623 opc = mdt_reint_opcode(info, reint_fmts);
1626 * No lock possible here from client to pass it to reint code
1629 rc = mdt_reint_internal(info, NULL, opc);
1634 info->mti_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
1638 /* this should sync the whole device */
1639 static int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
1641 struct dt_device *dt = mdt->mdt_bottom;
1645 rc = dt->dd_ops->dt_sync(env, dt);
1649 /* this should sync this object */
1650 static int mdt_object_sync(struct mdt_thread_info *info)
1652 struct md_object *next;
1656 if (!mdt_object_exists(info->mti_object)) {
1657 CWARN("Non existing object "DFID"!\n",
1658 PFID(mdt_object_fid(info->mti_object)));
1661 next = mdt_object_child(info->mti_object);
1662 rc = mo_object_sync(info->mti_env, next);
1667 static int mdt_sync(struct mdt_thread_info *info)
1669 struct req_capsule *pill = info->mti_pill;
1670 struct mdt_body *body;
1674 /* The fid may be zero, so we req_capsule_set manually */
1675 req_capsule_set(pill, &RQF_MDS_SYNC);
1677 body = req_capsule_client_get(pill, &RMF_MDT_BODY);
1679 RETURN(err_serious(-EINVAL));
1681 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
1682 RETURN(err_serious(-ENOMEM));
1684 if (fid_seq(&body->fid1) == 0) {
1685 /* sync the whole device */
1686 rc = req_capsule_server_pack(pill);
1688 rc = mdt_device_sync(info->mti_env, info->mti_mdt);
1690 rc = err_serious(rc);
1692 /* sync an object */
1693 rc = mdt_unpack_req_pack_rep(info, HABEO_CORPUS|HABEO_REFERO);
1695 rc = mdt_object_sync(info);
1697 struct md_object *next;
1698 const struct lu_fid *fid;
1699 struct lu_attr *la = &info->mti_attr.ma_attr;
1701 next = mdt_object_child(info->mti_object);
1702 info->mti_attr.ma_need = MA_INODE;
1703 info->mti_attr.ma_valid = 0;
1704 rc = mo_attr_get(info->mti_env, next,
1707 body = req_capsule_server_get(pill,
1709 fid = mdt_object_fid(info->mti_object);
1710 mdt_pack_attr2body(info, body, la, fid);
1714 rc = err_serious(rc);
1719 #ifdef HAVE_QUOTA_SUPPORT
1720 static int mdt_quotacheck_handle(struct mdt_thread_info *info)
1722 struct obd_quotactl *oqctl;
1723 struct req_capsule *pill = info->mti_pill;
1724 struct obd_export *exp = info->mti_exp;
1725 struct md_quota *mq = md_quota(info->mti_env);
1726 struct md_device *next = info->mti_mdt->mdt_child;
1730 oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
1734 /* remote client has no permission for quotacheck */
1735 if (unlikely(exp_connect_rmtclient(exp)))
1738 rc = req_capsule_server_pack(pill);
1743 rc = next->md_ops->mdo_quota.mqo_check(info->mti_env, next,
1748 static int mdt_quotactl_handle(struct mdt_thread_info *info)
1750 struct obd_quotactl *oqctl, *repoqc;
1751 struct req_capsule *pill = info->mti_pill;
1752 struct obd_export *exp = info->mti_exp;
1753 struct md_quota *mq = md_quota(info->mti_env);
1754 struct md_device *next = info->mti_mdt->mdt_child;
1755 const struct md_quota_operations *mqo = &next->md_ops->mdo_quota;
1759 oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
1764 if (exp_connect_rmtclient(exp)) {
1765 struct ptlrpc_request *req = mdt_info_req(info);
1766 struct mdt_export_data *med = mdt_req2med(req);
1767 struct lustre_idmap_table *idmap = med->med_idmap;
1769 if (unlikely(oqctl->qc_cmd != Q_GETQUOTA &&
1770 oqctl->qc_cmd != Q_GETINFO))
1774 if (oqctl->qc_type == USRQUOTA)
1775 id = lustre_idmap_lookup_uid(NULL, idmap, 0,
1777 else if (oqctl->qc_type == GRPQUOTA)
1778 id = lustre_idmap_lookup_gid(NULL, idmap, 0,
1783 if (id == CFS_IDMAP_NOTFOUND) {
1784 CDEBUG(D_QUOTA, "no mapping for id %u\n",
1790 rc = req_capsule_server_pack(pill);
1794 repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
1795 LASSERT(repoqc != NULL);
1798 switch (oqctl->qc_cmd) {
1800 rc = mqo->mqo_on(info->mti_env, next, oqctl->qc_type);
1803 rc = mqo->mqo_off(info->mti_env, next, oqctl->qc_type);
1806 rc = mqo->mqo_setinfo(info->mti_env, next, oqctl->qc_type, id,
1810 rc = mqo->mqo_getinfo(info->mti_env, next, oqctl->qc_type, id,
1814 rc = mqo->mqo_setquota(info->mti_env, next, oqctl->qc_type, id,
1818 rc = mqo->mqo_getquota(info->mti_env, next, oqctl->qc_type, id,
1822 rc = mqo->mqo_getoinfo(info->mti_env, next, oqctl->qc_type, id,
1826 rc = mqo->mqo_getoquota(info->mti_env, next, oqctl->qc_type, id,
1829 case LUSTRE_Q_INVALIDATE:
1830 rc = mqo->mqo_invalidate(info->mti_env, next, oqctl->qc_type);
1832 case LUSTRE_Q_FINVALIDATE:
1833 rc = mqo->mqo_finvalidate(info->mti_env, next, oqctl->qc_type);
1836 CERROR("unsupported mdt_quotactl command: %d\n",
1848 * OBD PING and other handlers.
1850 static int mdt_obd_ping(struct mdt_thread_info *info)
1855 req_capsule_set(info->mti_pill, &RQF_OBD_PING);
1857 rc = target_handle_ping(mdt_info_req(info));
1859 rc = err_serious(rc);
1863 static int mdt_obd_log_cancel(struct mdt_thread_info *info)
1865 return err_serious(-EOPNOTSUPP);
1868 static int mdt_obd_qc_callback(struct mdt_thread_info *info)
1870 return err_serious(-EOPNOTSUPP);
1878 /** clone llog ctxt from child (mdd)
1879 * This allows remote llog (replicator) access.
1880 * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
1881 * context was originally set up, or we can handle them directly.
1882 * I choose the latter, but that means I need any llog
1883 * contexts set up by child to be accessable by the mdt. So we clone the
1884 * context into our context list here.
1886 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
1889 struct md_device *next = mdt->mdt_child;
1890 struct llog_ctxt *ctxt;
1893 if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
1896 rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
1897 if (rc || ctxt == NULL) {
1898 CERROR("Can't get mdd ctxt %d\n", rc);
1902 rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
1904 CERROR("Can't set mdt ctxt %d\n", rc);
1909 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
1910 struct mdt_device *mdt, int idx)
1912 struct llog_ctxt *ctxt;
1914 ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
1917 /* Put once for the get we just did, and once for the clone */
1918 llog_ctxt_put(ctxt);
1919 llog_ctxt_put(ctxt);
1923 static int mdt_llog_create(struct mdt_thread_info *info)
1927 req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_CREATE);
1928 rc = llog_origin_handle_create(mdt_info_req(info));
1929 return (rc < 0 ? err_serious(rc) : rc);
1932 static int mdt_llog_destroy(struct mdt_thread_info *info)
1936 req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_DESTROY);
1937 rc = llog_origin_handle_destroy(mdt_info_req(info));
1938 return (rc < 0 ? err_serious(rc) : rc);
1941 static int mdt_llog_read_header(struct mdt_thread_info *info)
1945 req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_READ_HEADER);
1946 rc = llog_origin_handle_read_header(mdt_info_req(info));
1947 return (rc < 0 ? err_serious(rc) : rc);
1950 static int mdt_llog_next_block(struct mdt_thread_info *info)
1954 req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_NEXT_BLOCK);
1955 rc = llog_origin_handle_next_block(mdt_info_req(info));
1956 return (rc < 0 ? err_serious(rc) : rc);
1959 static int mdt_llog_prev_block(struct mdt_thread_info *info)
1963 req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_PREV_BLOCK);
1964 rc = llog_origin_handle_prev_block(mdt_info_req(info));
1965 return (rc < 0 ? err_serious(rc) : rc);
1972 static struct ldlm_callback_suite cbs = {
1973 .lcs_completion = ldlm_server_completion_ast,
1974 .lcs_blocking = ldlm_server_blocking_ast,
1978 static int mdt_enqueue(struct mdt_thread_info *info)
1980 struct ptlrpc_request *req;
1984 * info->mti_dlm_req already contains swapped and (if necessary)
1985 * converted dlm request.
1987 LASSERT(info->mti_dlm_req != NULL);
1989 req = mdt_info_req(info);
1990 rc = ldlm_handle_enqueue0(info->mti_mdt->mdt_namespace,
1991 req, info->mti_dlm_req, &cbs);
1992 info->mti_fail_id = OBD_FAIL_LDLM_REPLY;
1993 return rc ? err_serious(rc) : req->rq_status;
1996 static int mdt_convert(struct mdt_thread_info *info)
1999 struct ptlrpc_request *req;
2001 LASSERT(info->mti_dlm_req);
2002 req = mdt_info_req(info);
2003 rc = ldlm_handle_convert0(req, info->mti_dlm_req);
2004 return rc ? err_serious(rc) : req->rq_status;
2007 static int mdt_bl_callback(struct mdt_thread_info *info)
2009 CERROR("bl callbacks should not happen on MDS\n");
2011 return err_serious(-EOPNOTSUPP);
2014 static int mdt_cp_callback(struct mdt_thread_info *info)
2016 CERROR("cp callbacks should not happen on MDS\n");
2018 return err_serious(-EOPNOTSUPP);
2022 * sec context handlers
2024 static int mdt_sec_ctx_handle(struct mdt_thread_info *info)
2028 rc = mdt_handle_idmap(info);
2031 struct ptlrpc_request *req = mdt_info_req(info);
2034 opc = lustre_msg_get_opc(req->rq_reqmsg);
2035 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT)
2036 sptlrpc_svc_ctx_invalidate(req);
2039 OBD_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, obd_fail_val);
2044 static struct mdt_object *mdt_obj(struct lu_object *o)
2046 LASSERT(lu_device_is_mdt(o->lo_dev));
2047 return container_of0(o, struct mdt_object, mot_obj.mo_lu);
2050 struct mdt_object *mdt_object_find(const struct lu_env *env,
2051 struct mdt_device *d,
2052 const struct lu_fid *f)
2054 struct lu_object *o;
2055 struct mdt_object *m;
2058 CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2059 o = lu_object_find(env, &d->mdt_md_dev.md_lu_dev, f, NULL);
2060 if (unlikely(IS_ERR(o)))
2061 m = (struct mdt_object *)o;
2068 * Asyncronous commit for mdt device.
2070 * Pass asynchonous commit call down the MDS stack.
2072 * \param env environment
2073 * \param mdt the mdt device
2075 static void mdt_device_commit_async(const struct lu_env *env,
2076 struct mdt_device *mdt)
2078 struct dt_device *dt = mdt->mdt_bottom;
2081 rc = dt->dd_ops->dt_commit_async(env, dt);
2082 if (unlikely(rc != 0))
2083 CWARN("async commit start failed with rc = %d", rc);
2087 * Mark the lock as "synchonous".
2089 * Mark the lock to deffer transaction commit to the unlock time.
2091 * \param lock the lock to mark as "synchonous"
2093 * \see mdt_is_lock_sync
2094 * \see mdt_save_lock
2096 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2098 lock->l_ast_data = (void*)1;
2102 * Check whehter the lock "synchonous" or not.
2104 * \param lock the lock to check
2105 * \retval 1 the lock is "synchonous"
2106 * \retval 0 the lock isn't "synchronous"
2108 * \see mdt_set_lock_sync
2109 * \see mdt_save_lock
2111 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2113 return lock->l_ast_data != NULL;
2117 * Blocking AST for mdt locks.
2119 * Starts transaction commit if in case of COS lock conflict or
2120 * deffers such a commit to the mdt_save_lock.
2122 * \param lock the lock which blocks a request or cancelling lock
2123 * \param desc unused
2124 * \param data unused
2125 * \param flag indicates whether this cancelling or blocking callback
2127 * \see ldlm_blocking_ast_nocheck
2129 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2130 void *data, int flag)
2132 struct obd_device *obd = lock->l_resource->lr_namespace->ns_obd;
2133 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2137 if (flag == LDLM_CB_CANCELING)
2139 lock_res_and_lock(lock);
2140 if (lock->l_blocking_ast != mdt_blocking_ast) {
2141 unlock_res_and_lock(lock);
2144 if (mdt_cos_is_enabled(mdt) &&
2145 lock->l_req_mode & (LCK_PW | LCK_EX) &&
2146 lock->l_blocking_lock != NULL &&
2147 lock->l_client_cookie != lock->l_blocking_lock->l_client_cookie) {
2148 mdt_set_lock_sync(lock);
2150 rc = ldlm_blocking_ast_nocheck(lock);
2152 /* There is no lock conflict if l_blocking_lock == NULL,
2153 * it indicates a blocking ast sent from ldlm_lock_decref_internal
2154 * when the last reference to a local lock was released */
2155 if (lock->l_req_mode == LCK_COS && lock->l_blocking_lock != NULL) {
2158 rc = lu_env_init(&env, LCT_MD_THREAD);
2159 if (unlikely(rc != 0))
2160 CWARN("lu_env initialization failed with rc = %d,"
2161 "cannot start asynchronous commit\n", rc);
2163 mdt_device_commit_async(&env, mdt);
2169 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2170 struct mdt_lock_handle *lh, __u64 ibits, int locality)
2172 struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
2173 ldlm_policy_data_t *policy = &info->mti_policy;
2174 struct ldlm_res_id *res_id = &info->mti_res_id;
2178 LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2179 LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2180 LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
2181 LASSERT(lh->mlh_type != MDT_NUL_LOCK);
2183 if (mdt_object_exists(o) < 0) {
2184 if (locality == MDT_CROSS_LOCK) {
2185 /* cross-ref object fix */
2186 ibits &= ~MDS_INODELOCK_UPDATE;
2187 ibits |= MDS_INODELOCK_LOOKUP;
2189 LASSERT(!(ibits & MDS_INODELOCK_UPDATE));
2190 LASSERT(ibits & MDS_INODELOCK_LOOKUP);
2192 /* No PDO lock on remote object */
2193 LASSERT(lh->mlh_type != MDT_PDO_LOCK);
2196 if (lh->mlh_type == MDT_PDO_LOCK) {
2197 /* check for exists after object is locked */
2198 if (mdt_object_exists(o) == 0) {
2199 /* Non-existent object shouldn't have PDO lock */
2202 /* Non-dir object shouldn't have PDO lock */
2203 LASSERT(S_ISDIR(lu_object_attr(&o->mot_obj.mo_lu)));
2207 memset(policy, 0, sizeof(*policy));
2208 fid_build_reg_res_name(mdt_object_fid(o), res_id);
2211 * Take PDO lock on whole directory and build correct @res_id for lock
2212 * on part of directory.
2214 if (lh->mlh_pdo_hash != 0) {
2215 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
2216 mdt_lock_pdo_mode(info, o, lh);
2217 if (lh->mlh_pdo_mode != LCK_NL) {
2219 * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
2220 * is never going to be sent to client and we do not
2221 * want it slowed down due to possible cancels.
2223 policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
2224 rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
2225 policy, res_id, LDLM_FL_ATOMIC_CB,
2226 &info->mti_exp->exp_handle.h_cookie);
2232 * Finish res_id initializing by name hash marking part of
2233 * directory which is taking modification.
2235 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
2238 policy->l_inodebits.bits = ibits;
2241 * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
2242 * going to be sent to client. If it is - mdt_intent_policy() path will
2243 * fix it up and turn FL_LOCAL flag off.
2245 rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
2246 res_id, LDLM_FL_LOCAL_ONLY | LDLM_FL_ATOMIC_CB,
2247 &info->mti_exp->exp_handle.h_cookie);
2249 mdt_object_unlock(info, o, lh, 1);
2250 else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
2251 lh->mlh_pdo_hash != 0 &&
2252 (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX)) {
2253 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 10);
2260 * Save a lock within request object.
2262 * Keep the lock referenced until whether client ACK or transaction
2263 * commit happens or release the lock immediately depending on input
2264 * parameters. If COS is ON, a write lock is converted to COS lock
2267 * \param info thead info object
2268 * \param h lock handle
2269 * \param mode lock mode
2270 * \param decref force immediate lock releasing
2273 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
2274 ldlm_mode_t mode, int decref)
2278 if (lustre_handle_is_used(h)) {
2279 if (decref || !info->mti_has_trans ||
2280 !(mode & (LCK_PW | LCK_EX))){
2281 mdt_fid_unlock(h, mode);
2283 struct mdt_device *mdt = info->mti_mdt;
2284 struct ldlm_lock *lock = ldlm_handle2lock(h);
2285 struct ptlrpc_request *req = mdt_info_req(info);
2288 LASSERTF(lock != NULL, "no lock for cookie "LPX64"\n",
2290 CDEBUG(D_HA, "request = %p reply state = %p"
2291 " transno = "LPD64"\n",
2292 req, req->rq_reply_state, req->rq_transno);
2293 if (mdt_cos_is_enabled(mdt)) {
2295 ldlm_lock_downgrade(lock, LCK_COS);
2298 ptlrpc_save_lock(req, h, mode, no_ack);
2299 if (mdt_is_lock_sync(lock)) {
2300 CDEBUG(D_HA, "found sync-lock,"
2301 " async commit started\n");
2302 mdt_device_commit_async(info->mti_env,
2305 LDLM_LOCK_PUT(lock);
2314 * Unlock mdt object.
2316 * Immeditely release the regular lock and the PDO lock or save the
2317 * lock in reqeuest and keep them referenced until client ACK or
2318 * transaction commit.
2320 * \param info thread info object
2321 * \param o mdt object
2322 * \param lh mdt lock handle referencing regular and PDO locks
2323 * \param decref force immediate lock releasing
2325 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
2326 struct mdt_lock_handle *lh, int decref)
2330 mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
2331 mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
2336 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
2337 const struct lu_fid *f,
2338 struct mdt_lock_handle *lh,
2341 struct mdt_object *o;
2343 o = mdt_object_find(info->mti_env, info->mti_mdt, f);
2347 rc = mdt_object_lock(info, o, lh, ibits,
2350 mdt_object_put(info->mti_env, o);
2357 void mdt_object_unlock_put(struct mdt_thread_info * info,
2358 struct mdt_object * o,
2359 struct mdt_lock_handle *lh,
2362 mdt_object_unlock(info, o, lh, decref);
2363 mdt_object_put(info->mti_env, o);
2366 static struct mdt_handler *mdt_handler_find(__u32 opc,
2367 struct mdt_opc_slice *supported)
2369 struct mdt_opc_slice *s;
2370 struct mdt_handler *h;
2373 for (s = supported; s->mos_hs != NULL; s++) {
2374 if (s->mos_opc_start <= opc && opc < s->mos_opc_end) {
2375 h = s->mos_hs + (opc - s->mos_opc_start);
2376 if (likely(h->mh_opc != 0))
2377 LASSERTF(h->mh_opc == opc,
2378 "opcode mismatch %d != %d\n",
2381 h = NULL; /* unsupported opc */
2388 static int mdt_lock_resname_compat(struct mdt_device *m,
2389 struct ldlm_request *req)
2391 /* XXX something... later. */
2395 static int mdt_lock_reply_compat(struct mdt_device *m, struct ldlm_reply *rep)
2397 /* XXX something... later. */
2402 * Generic code handling requests that have struct mdt_body passed in:
2404 * - extract mdt_body from request and save it in @info, if present;
2406 * - create lu_object, corresponding to the fid in mdt_body, and save it in
2409 * - if HABEO_CORPUS flag is set for this request type check whether object
2410 * actually exists on storage (lu_object_exists()).
2413 static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags)
2415 const struct mdt_body *body;
2416 struct mdt_object *obj;
2417 const struct lu_env *env;
2418 struct req_capsule *pill;
2422 env = info->mti_env;
2423 pill = info->mti_pill;
2425 body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
2429 if (!(body->valid & OBD_MD_FLID))
2432 if (!fid_is_sane(&body->fid1)) {
2433 CERROR("Invalid fid: "DFID"\n", PFID(&body->fid1));
2438 * Do not get size or any capa fields before we check that request
2439 * contains capa actually. There are some requests which do not, for
2440 * instance MDS_IS_SUBDIR.
2442 if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_CLIENT) &&
2443 req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
2444 mdt_set_capainfo(info, 0, &body->fid1,
2445 req_capsule_client_get(pill, &RMF_CAPA1));
2447 obj = mdt_object_find(env, info->mti_mdt, &body->fid1);
2449 if ((flags & HABEO_CORPUS) &&
2450 !mdt_object_exists(obj)) {
2451 mdt_object_put(env, obj);
2452 /* for capability renew ENOENT will be handled in
2454 if (body->valid & OBD_MD_FLOSSCAPA)
2459 info->mti_object = obj;
2468 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags)
2470 struct req_capsule *pill = info->mti_pill;
2474 if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
2475 rc = mdt_body_unpack(info, flags);
2479 if (rc == 0 && (flags & HABEO_REFERO)) {
2480 struct mdt_device *mdt = info->mti_mdt;
2484 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2485 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2486 mdt->mdt_max_mdsize);
2487 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2488 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER,
2489 mdt->mdt_max_cookiesize);
2491 rc = req_capsule_server_pack(pill);
2496 static int mdt_init_capa_ctxt(const struct lu_env *env, struct mdt_device *m)
2498 struct md_device *next = m->mdt_child;
2500 return next->md_ops->mdo_init_capa_ctxt(env, next,
2501 m->mdt_opts.mo_mds_capa,
2502 m->mdt_capa_timeout,
2508 * Invoke handler for this request opc. Also do necessary preprocessing
2509 * (according to handler ->mh_flags), and post-processing (setting of
2510 * ->last_{xid,committed}).
2512 static int mdt_req_handle(struct mdt_thread_info *info,
2513 struct mdt_handler *h, struct ptlrpc_request *req)
2515 int rc, serious = 0;
2520 LASSERT(h->mh_act != NULL);
2521 LASSERT(h->mh_opc == lustre_msg_get_opc(req->rq_reqmsg));
2522 LASSERT(current->journal_info == NULL);
2525 * Checking for various OBD_FAIL_$PREF_$OPC_NET codes. _Do_ not try
2526 * to put same checks into handlers like mdt_close(), mdt_reint(),
2527 * etc., without talking to mdt authors first. Checking same thing
2528 * there again is useless and returning 0 error without packing reply
2529 * is buggy! Handlers either pack reply or return error.
2531 * We return 0 here and do not send any reply in order to emulate
2532 * network failure. Do not send any reply in case any of NET related
2533 * fail_id has occured.
2535 if (OBD_FAIL_CHECK_ORSET(h->mh_fail_id, OBD_FAIL_ONCE))
2539 flags = h->mh_flags;
2540 LASSERT(ergo(flags & (HABEO_CORPUS|HABEO_REFERO), h->mh_fmt != NULL));
2542 if (h->mh_fmt != NULL) {
2543 req_capsule_set(info->mti_pill, h->mh_fmt);
2544 rc = mdt_unpack_req_pack_rep(info, flags);
2547 if (rc == 0 && flags & MUTABOR &&
2548 req->rq_export->exp_connect_flags & OBD_CONNECT_RDONLY)
2549 /* should it be rq_status? */
2552 if (rc == 0 && flags & HABEO_CLAVIS) {
2553 struct ldlm_request *dlm_req;
2555 LASSERT(h->mh_fmt != NULL);
2557 dlm_req = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
2558 if (dlm_req != NULL) {
2559 if (unlikely(dlm_req->lock_desc.l_resource.lr_type ==
2561 dlm_req->lock_desc.l_policy_data.\
2562 l_inodebits.bits == 0)) {
2564 * Lock without inodebits makes no sense and
2565 * will oops later in ldlm. If client miss to
2566 * set such bits, do not trigger ASSERTION.
2568 * For liblustre flock case, it maybe zero.
2572 if (info->mti_mdt->mdt_opts.mo_compat_resname)
2573 rc = mdt_lock_resname_compat(
2576 info->mti_dlm_req = dlm_req;
2583 /* capability setting changed via /proc, needs reinitialize ctxt */
2584 if (info->mti_mdt && info->mti_mdt->mdt_capa_conf) {
2585 mdt_init_capa_ctxt(info->mti_env, info->mti_mdt);
2586 info->mti_mdt->mdt_capa_conf = 0;
2589 if (likely(rc == 0)) {
2591 * Process request, there can be two types of rc:
2592 * 1) errors with msg unpack/pack, other failures outside the
2593 * operation itself. This is counted as serious errors;
2594 * 2) errors during fs operation, should be placed in rq_status
2597 rc = h->mh_act(info);
2599 !req->rq_no_reply && req->rq_reply_state == NULL) {
2600 DEBUG_REQ(D_ERROR, req, "MDT \"handler\" %s did not "
2601 "pack reply and returned 0 error\n",
2605 serious = is_serious(rc);
2606 rc = clear_serious(rc);
2610 req->rq_status = rc;
2613 * ELDLM_* codes which > 0 should be in rq_status only as well as
2614 * all non-serious errors.
2616 if (rc > 0 || !serious)
2619 LASSERT(current->journal_info == NULL);
2621 if (rc == 0 && (flags & HABEO_CLAVIS) &&
2622 info->mti_mdt->mdt_opts.mo_compat_resname) {
2623 struct ldlm_reply *dlmrep;
2625 dlmrep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
2627 rc = mdt_lock_reply_compat(info->mti_mdt, dlmrep);
2630 /* If we're DISCONNECTing, the mdt_export_data is already freed */
2631 if (likely(rc == 0 && req->rq_export && h->mh_opc != MDS_DISCONNECT))
2632 target_committed_to_req(req);
2634 if (unlikely(req_is_replay(req) &&
2635 lustre_msg_get_transno(req->rq_reqmsg) == 0)) {
2636 DEBUG_REQ(D_ERROR, req, "transno is 0 during REPLAY");
2640 target_send_reply(req, rc, info->mti_fail_id);
2644 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
2646 lh->mlh_type = MDT_NUL_LOCK;
2647 lh->mlh_reg_lh.cookie = 0ull;
2648 lh->mlh_reg_mode = LCK_MINMODE;
2649 lh->mlh_pdo_lh.cookie = 0ull;
2650 lh->mlh_pdo_mode = LCK_MINMODE;
2653 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
2655 LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2656 LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2660 * Initialize fields of struct mdt_thread_info. Other fields are left in
2661 * uninitialized state, because it's too expensive to zero out whole
2662 * mdt_thread_info (> 1K) on each request arrival.
2664 static void mdt_thread_info_init(struct ptlrpc_request *req,
2665 struct mdt_thread_info *info)
2668 struct md_capainfo *ci;
2670 req_capsule_init(&req->rq_pill, req, RCL_SERVER);
2671 info->mti_pill = &req->rq_pill;
2674 for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2675 mdt_lock_handle_init(&info->mti_lh[i]);
2677 /* mdt device: it can be NULL while CONNECT */
2678 if (req->rq_export) {
2679 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
2680 info->mti_exp = req->rq_export;
2682 info->mti_mdt = NULL;
2683 info->mti_env = req->rq_svc_thread->t_env;
2684 ci = md_capainfo(info->mti_env);
2685 memset(ci, 0, sizeof *ci);
2686 if (req->rq_export) {
2687 if (exp_connect_rmtclient(req->rq_export))
2688 ci->mc_auth = LC_ID_CONVERT;
2689 else if (req->rq_export->exp_connect_flags &
2690 OBD_CONNECT_MDS_CAPA)
2691 ci->mc_auth = LC_ID_PLAIN;
2693 ci->mc_auth = LC_ID_NONE;
2696 info->mti_fail_id = OBD_FAIL_MDS_ALL_REPLY_NET;
2697 info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
2698 info->mti_mos = NULL;
2700 memset(&info->mti_attr, 0, sizeof(info->mti_attr));
2701 info->mti_body = NULL;
2702 info->mti_object = NULL;
2703 info->mti_dlm_req = NULL;
2704 info->mti_has_trans = 0;
2705 info->mti_no_need_trans = 0;
2706 info->mti_cross_ref = 0;
2707 info->mti_opdata = 0;
2709 /* To not check for split by default. */
2710 info->mti_spec.sp_ck_split = 0;
2711 info->mti_spec.no_create = 0;
2714 static void mdt_thread_info_fini(struct mdt_thread_info *info)
2718 req_capsule_fini(info->mti_pill);
2719 if (info->mti_object != NULL) {
2721 * freeing an object may lead to OSD level transaction, do not
2722 * let it mess with MDT. bz19385.
2724 info->mti_no_need_trans = 1;
2725 mdt_object_put(info->mti_env, info->mti_object);
2726 info->mti_object = NULL;
2728 for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2729 mdt_lock_handle_fini(&info->mti_lh[i]);
2730 info->mti_env = NULL;
2733 static int mdt_filter_recovery_request(struct ptlrpc_request *req,
2734 struct obd_device *obd, int *process)
2736 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2737 case MDS_CONNECT: /* This will never get here, but for completeness. */
2738 case OST_CONNECT: /* This will never get here, but for completeness. */
2739 case MDS_DISCONNECT:
2740 case OST_DISCONNECT:
2745 case MDS_DONE_WRITING:
2746 case MDS_SYNC: /* used in unmounting */
2752 *process = target_queue_recovery_request(req, obd);
2756 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
2763 * Handle recovery. Return:
2764 * +1: continue request processing;
2765 * -ve: abort immediately with the given error code;
2766 * 0: send reply with error code in req->rq_status;
2768 static int mdt_recovery(struct mdt_thread_info *info)
2770 struct ptlrpc_request *req = mdt_info_req(info);
2771 struct obd_device *obd;
2775 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2778 case SEC_CTX_INIT_CONT:
2784 rc = mdt_handle_idmap(info);
2793 if (unlikely(!class_connected_export(req->rq_export))) {
2794 CERROR("operation %d on unconnected MDS from %s\n",
2795 lustre_msg_get_opc(req->rq_reqmsg),
2796 libcfs_id2str(req->rq_peer));
2797 /* FIXME: For CMD cleanup, when mds_B stop, the req from
2798 * mds_A will get -ENOTCONN(especially for ping req),
2799 * which will cause that mds_A deactive timeout, then when
2800 * mds_A cleanup, the cleanup process will be suspended since
2801 * deactive timeout is not zero.
2803 req->rq_status = -ENOTCONN;
2804 target_send_reply(req, -ENOTCONN, info->mti_fail_id);
2808 /* sanity check: if the xid matches, the request must be marked as a
2809 * resent or replayed */
2810 if (req_xid_is_last(req)) {
2811 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
2812 (MSG_RESENT | MSG_REPLAY))) {
2813 DEBUG_REQ(D_WARNING, req, "rq_xid "LPU64" matches last_xid, "
2814 "expected REPLAY or RESENT flag (%x)", req->rq_xid,
2815 lustre_msg_get_flags(req->rq_reqmsg));
2817 req->rq_status = -ENOTCONN;
2822 /* else: note the opposite is not always true; a RESENT req after a
2823 * failover will usually not match the last_xid, since it was likely
2824 * never committed. A REPLAYed request will almost never match the
2825 * last xid, however it could for a committed, but still retained,
2828 obd = req->rq_export->exp_obd;
2830 /* Check for aborted recovery... */
2831 if (unlikely(obd->obd_recovering)) {
2834 DEBUG_REQ(D_INFO, req, "Got new replay");
2835 rc = mdt_filter_recovery_request(req, obd, &should_process);
2836 if (rc != 0 || !should_process)
2838 else if (should_process < 0) {
2839 req->rq_status = should_process;
2840 rc = ptlrpc_error(req);
2847 static int mdt_msg_check_version(struct lustre_msg *msg)
2851 switch (lustre_msg_get_opc(msg)) {
2853 case MDS_DISCONNECT:
2856 case SEC_CTX_INIT_CONT:
2858 rc = lustre_msg_check_version(msg, LUSTRE_OBD_VERSION);
2860 CERROR("bad opc %u version %08x, expecting %08x\n",
2861 lustre_msg_get_opc(msg),
2862 lustre_msg_get_version(msg),
2863 LUSTRE_OBD_VERSION);
2867 case MDS_GETATTR_NAME:
2874 case MDS_DONE_WRITING:
2881 case MDS_QUOTACHECK:
2887 rc = lustre_msg_check_version(msg, LUSTRE_MDS_VERSION);
2889 CERROR("bad opc %u version %08x, expecting %08x\n",
2890 lustre_msg_get_opc(msg),
2891 lustre_msg_get_version(msg),
2892 LUSTRE_MDS_VERSION);
2896 case LDLM_BL_CALLBACK:
2897 case LDLM_CP_CALLBACK:
2898 rc = lustre_msg_check_version(msg, LUSTRE_DLM_VERSION);
2900 CERROR("bad opc %u version %08x, expecting %08x\n",
2901 lustre_msg_get_opc(msg),
2902 lustre_msg_get_version(msg),
2903 LUSTRE_DLM_VERSION);
2905 case OBD_LOG_CANCEL:
2906 case LLOG_ORIGIN_HANDLE_CREATE:
2907 case LLOG_ORIGIN_HANDLE_NEXT_BLOCK:
2908 case LLOG_ORIGIN_HANDLE_READ_HEADER:
2909 case LLOG_ORIGIN_HANDLE_CLOSE:
2910 case LLOG_ORIGIN_HANDLE_DESTROY:
2911 case LLOG_ORIGIN_HANDLE_PREV_BLOCK:
2913 rc = lustre_msg_check_version(msg, LUSTRE_LOG_VERSION);
2915 CERROR("bad opc %u version %08x, expecting %08x\n",
2916 lustre_msg_get_opc(msg),
2917 lustre_msg_get_version(msg),
2918 LUSTRE_LOG_VERSION);
2921 CERROR("MDS unknown opcode %d\n", lustre_msg_get_opc(msg));
2927 static int mdt_handle0(struct ptlrpc_request *req,
2928 struct mdt_thread_info *info,
2929 struct mdt_opc_slice *supported)
2931 struct mdt_handler *h;
2932 struct lustre_msg *msg;
2937 if (OBD_FAIL_CHECK_ORSET(OBD_FAIL_MDS_ALL_REQUEST_NET, OBD_FAIL_ONCE))
2940 LASSERT(current->journal_info == NULL);
2942 msg = req->rq_reqmsg;
2943 rc = mdt_msg_check_version(msg);
2944 if (likely(rc == 0)) {
2945 rc = mdt_recovery(info);
2946 if (likely(rc == +1)) {
2947 h = mdt_handler_find(lustre_msg_get_opc(msg),
2949 if (likely(h != NULL)) {
2950 rc = mdt_req_handle(info, h, req);
2952 CERROR("The unsupported opc: 0x%x\n",
2953 lustre_msg_get_opc(msg) );
2954 req->rq_status = -ENOTSUPP;
2955 rc = ptlrpc_error(req);
2960 CERROR(LUSTRE_MDT_NAME" drops mal-formed request\n");
2965 * MDT handler function called by ptlrpc service thread when request comes.
2967 * XXX common "target" functionality should be factored into separate module
2968 * shared by mdt, ost and stand-alone services like fld.
2970 static int mdt_handle_common(struct ptlrpc_request *req,
2971 struct mdt_opc_slice *supported)
2974 struct mdt_thread_info *info;
2978 env = req->rq_svc_thread->t_env;
2979 LASSERT(env != NULL);
2980 LASSERT(env->le_ses != NULL);
2981 LASSERT(env->le_ctx.lc_thread == req->rq_svc_thread);
2982 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
2983 LASSERT(info != NULL);
2985 mdt_thread_info_init(req, info);
2987 rc = mdt_handle0(req, info, supported);
2989 mdt_thread_info_fini(info);
2994 * This is called from recovery code as handler of _all_ RPC types, FLD and SEQ
2997 int mdt_recovery_handle(struct ptlrpc_request *req)
3002 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
3004 rc = mdt_handle_common(req, mdt_fld_handlers);
3007 rc = mdt_handle_common(req, mdt_seq_handlers);
3010 rc = mdt_handle_common(req, mdt_regular_handlers);
3017 static int mdt_regular_handle(struct ptlrpc_request *req)
3019 return mdt_handle_common(req, mdt_regular_handlers);
3022 static int mdt_readpage_handle(struct ptlrpc_request *req)
3024 return mdt_handle_common(req, mdt_readpage_handlers);
3027 static int mdt_xmds_handle(struct ptlrpc_request *req)
3029 return mdt_handle_common(req, mdt_xmds_handlers);
3032 static int mdt_mdsc_handle(struct ptlrpc_request *req)
3034 return mdt_handle_common(req, mdt_seq_handlers);
3037 static int mdt_mdss_handle(struct ptlrpc_request *req)
3039 return mdt_handle_common(req, mdt_seq_handlers);
3042 static int mdt_dtss_handle(struct ptlrpc_request *req)
3044 return mdt_handle_common(req, mdt_seq_handlers);
3047 static int mdt_fld_handle(struct ptlrpc_request *req)
3049 return mdt_handle_common(req, mdt_fld_handlers);
3065 static int mdt_intent_getattr(enum mdt_it_code opcode,
3066 struct mdt_thread_info *info,
3067 struct ldlm_lock **,
3069 static int mdt_intent_reint(enum mdt_it_code opcode,
3070 struct mdt_thread_info *info,
3071 struct ldlm_lock **,
3074 static struct mdt_it_flavor {
3075 const struct req_format *it_fmt;
3077 int (*it_act)(enum mdt_it_code ,
3078 struct mdt_thread_info *,
3079 struct ldlm_lock **,
3082 } mdt_it_flavor[] = {
3084 .it_fmt = &RQF_LDLM_INTENT,
3085 /*.it_flags = HABEO_REFERO,*/
3087 .it_act = mdt_intent_reint,
3088 .it_reint = REINT_OPEN
3091 .it_fmt = &RQF_LDLM_INTENT,
3092 .it_flags = MUTABOR,
3093 .it_act = mdt_intent_reint,
3094 .it_reint = REINT_OPEN
3097 .it_fmt = &RQF_LDLM_INTENT,
3098 .it_flags = MUTABOR,
3099 .it_act = mdt_intent_reint,
3100 .it_reint = REINT_CREATE
3102 [MDT_IT_GETATTR] = {
3103 .it_fmt = &RQF_LDLM_INTENT_GETATTR,
3104 .it_flags = HABEO_REFERO,
3105 .it_act = mdt_intent_getattr
3107 [MDT_IT_READDIR] = {
3113 .it_fmt = &RQF_LDLM_INTENT_GETATTR,
3114 .it_flags = HABEO_REFERO,
3115 .it_act = mdt_intent_getattr
3118 .it_fmt = &RQF_LDLM_INTENT_UNLINK,
3119 .it_flags = MUTABOR,
3121 .it_reint = REINT_UNLINK
3125 .it_flags = MUTABOR,
3128 [MDT_IT_GETXATTR] = {
3135 int mdt_intent_lock_replace(struct mdt_thread_info *info,
3136 struct ldlm_lock **lockp,
3137 struct ldlm_lock *new_lock,
3138 struct mdt_lock_handle *lh,
3141 struct ptlrpc_request *req = mdt_info_req(info);
3142 struct ldlm_lock *lock = *lockp;
3145 * Get new lock only for cases when possible resent did not find any
3148 if (new_lock == NULL)
3149 new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
3151 if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
3152 lh->mlh_reg_lh.cookie = 0;
3156 LASSERTF(new_lock != NULL,
3157 "lockh "LPX64"\n", lh->mlh_reg_lh.cookie);
3160 * If we've already given this lock to a client once, then we should
3161 * have no readers or writers. Otherwise, we should have one reader
3162 * _or_ writer ref (which will be zeroed below) before returning the
3165 if (new_lock->l_export == req->rq_export) {
3166 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
3168 LASSERT(new_lock->l_export == NULL);
3169 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
3174 if (new_lock->l_export == req->rq_export) {
3176 * Already gave this to the client, which means that we
3177 * reconstructed a reply.
3179 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
3181 lh->mlh_reg_lh.cookie = 0;
3182 RETURN(ELDLM_LOCK_REPLACED);
3186 * Fixup the lock to be given to the client.
3188 lock_res_and_lock(new_lock);
3189 /* Zero new_lock->l_readers and new_lock->l_writers without triggering
3190 * possible blocking AST. */
3191 while (new_lock->l_readers > 0) {
3192 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
3193 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3194 new_lock->l_readers--;
3196 while (new_lock->l_writers > 0) {
3197 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
3198 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3199 new_lock->l_writers--;
3202 new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
3203 new_lock->l_blocking_ast = lock->l_blocking_ast;
3204 new_lock->l_completion_ast = lock->l_completion_ast;
3205 new_lock->l_remote_handle = lock->l_remote_handle;
3206 new_lock->l_flags &= ~LDLM_FL_LOCAL;
3208 unlock_res_and_lock(new_lock);
3210 cfs_hash_add(new_lock->l_export->exp_lock_hash,
3211 &new_lock->l_remote_handle,
3212 &new_lock->l_exp_hash);
3214 LDLM_LOCK_RELEASE(new_lock);
3215 lh->mlh_reg_lh.cookie = 0;
3217 RETURN(ELDLM_LOCK_REPLACED);
3220 static void mdt_intent_fixup_resent(struct mdt_thread_info *info,
3221 struct ldlm_lock *new_lock,
3222 struct ldlm_lock **old_lock,
3223 struct mdt_lock_handle *lh)
3225 struct ptlrpc_request *req = mdt_info_req(info);
3226 struct obd_export *exp = req->rq_export;
3227 struct lustre_handle remote_hdl;
3228 struct ldlm_request *dlmreq;
3229 struct ldlm_lock *lock;
3231 if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
3234 dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
3235 remote_hdl = dlmreq->lock_handle[0];
3237 lock = cfs_hash_lookup(exp->exp_lock_hash, &remote_hdl);
3239 if (lock != new_lock) {
3240 lh->mlh_reg_lh.cookie = lock->l_handle.h_cookie;
3241 lh->mlh_reg_mode = lock->l_granted_mode;
3243 LDLM_DEBUG(lock, "Restoring lock cookie");
3244 DEBUG_REQ(D_DLMTRACE, req,
3245 "restoring lock cookie "LPX64,
3246 lh->mlh_reg_lh.cookie);
3248 *old_lock = LDLM_LOCK_GET(lock);
3249 cfs_hash_put(exp->exp_lock_hash, &lock->l_exp_hash);
3253 cfs_hash_put(exp->exp_lock_hash, &lock->l_exp_hash);
3257 * If the xid matches, then we know this is a resent request, and allow
3258 * it. (It's probably an OPEN, for which we don't send a lock.
3260 if (req_xid_is_last(req))
3264 * This remote handle isn't enqueued, so we never received or processed
3265 * this request. Clear MSG_RESENT, because it can be handled like any
3266 * normal request now.
3268 lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
3270 DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle "LPX64,
3274 static int mdt_intent_getattr(enum mdt_it_code opcode,
3275 struct mdt_thread_info *info,
3276 struct ldlm_lock **lockp,
3279 struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3280 struct ldlm_lock *new_lock = NULL;
3282 struct ldlm_reply *ldlm_rep;
3283 struct ptlrpc_request *req;
3284 struct mdt_body *reqbody;
3285 struct mdt_body *repbody;
3289 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
3292 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
3295 info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
3296 info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
3297 repbody->eadatasize = 0;
3298 repbody->aclsize = 0;
3302 child_bits = MDS_INODELOCK_LOOKUP;
3304 case MDT_IT_GETATTR:
3305 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE;
3308 CERROR("Unhandled till now");
3309 GOTO(out_shrink, rc = -EINVAL);
3312 rc = mdt_init_ucred(info, reqbody);
3314 GOTO(out_shrink, rc);
3316 req = info->mti_pill->rc_req;
3317 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3318 mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
3320 /* Get lock from request for possible resent case. */
3321 mdt_intent_fixup_resent(info, *lockp, &new_lock, lhc);
3323 ldlm_rep->lock_policy_res2 =
3324 mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
3326 if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
3327 ldlm_rep->lock_policy_res2 = 0;
3328 if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
3329 ldlm_rep->lock_policy_res2) {
3330 lhc->mlh_reg_lh.cookie = 0ull;
3331 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
3334 rc = mdt_intent_lock_replace(info, lockp, new_lock, lhc, flags);
3337 mdt_exit_ucred(info);
3339 mdt_shrink_reply(info);
3343 static int mdt_intent_reint(enum mdt_it_code opcode,
3344 struct mdt_thread_info *info,
3345 struct ldlm_lock **lockp,
3348 struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3349 struct ldlm_reply *rep = NULL;
3353 static const struct req_format *intent_fmts[REINT_MAX] = {
3354 [REINT_CREATE] = &RQF_LDLM_INTENT_CREATE,
3355 [REINT_OPEN] = &RQF_LDLM_INTENT_OPEN
3360 opc = mdt_reint_opcode(info, intent_fmts);
3364 if (mdt_it_flavor[opcode].it_reint != opc) {
3365 CERROR("Reint code %ld doesn't match intent: %d\n",
3367 RETURN(err_serious(-EPROTO));
3370 /* Get lock from request for possible resent case. */
3371 mdt_intent_fixup_resent(info, *lockp, NULL, lhc);
3373 rc = mdt_reint_internal(info, lhc, opc);
3375 /* Check whether the reply has been packed successfully. */
3376 if (mdt_info_req(info)->rq_repmsg != NULL)
3377 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3379 RETURN(err_serious(-EFAULT));
3381 /* MDC expects this in any case */
3383 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
3385 /* Cross-ref case, the lock should be returned to the client */
3386 if (rc == -EREMOTE) {
3387 LASSERT(lustre_handle_is_used(&lhc->mlh_reg_lh));
3388 rep->lock_policy_res2 = 0;
3389 rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
3392 rep->lock_policy_res2 = clear_serious(rc);
3394 if (rc == -ENOTCONN || rc == -ENODEV ||
3395 rc == -EOVERFLOW) { /**< if VBR failure then return error */
3397 * If it is the disconnect error (ENODEV & ENOCONN), the error
3398 * will be returned by rq_status, and client at ptlrpc layer
3399 * will detect this, then disconnect, reconnect the import
3400 * immediately, instead of impacting the following the rpc.
3402 lhc->mlh_reg_lh.cookie = 0ull;
3406 * For other cases, the error will be returned by intent.
3407 * and client will retrieve the result from intent.
3410 * FIXME: when open lock is finished, that should be
3413 if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3414 rep->lock_policy_res2 = 0;
3415 rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
3418 lhc->mlh_reg_lh.cookie = 0ull;
3419 RETURN(ELDLM_LOCK_ABORTED);
3424 static int mdt_intent_code(long itcode)
3432 case IT_OPEN|IT_CREAT:
3439 rc = MDT_IT_READDIR;
3442 rc = MDT_IT_GETATTR;
3454 rc = MDT_IT_GETXATTR;
3457 CERROR("Unknown intent opcode: %ld\n", itcode);
3464 static int mdt_intent_opc(long itopc, struct mdt_thread_info *info,
3465 struct ldlm_lock **lockp, int flags)
3467 struct req_capsule *pill;
3468 struct mdt_it_flavor *flv;
3473 opc = mdt_intent_code(itopc);
3477 pill = info->mti_pill;
3478 flv = &mdt_it_flavor[opc];
3480 if (flv->it_fmt != NULL)
3481 req_capsule_extend(pill, flv->it_fmt);
3483 rc = mdt_unpack_req_pack_rep(info, flv->it_flags);
3485 struct ptlrpc_request *req = mdt_info_req(info);
3486 if (flv->it_flags & MUTABOR &&
3487 req->rq_export->exp_connect_flags & OBD_CONNECT_RDONLY)
3490 if (rc == 0 && flv->it_act != NULL) {
3491 /* execute policy */
3492 rc = flv->it_act(opc, info, lockp, flags);
3499 static int mdt_intent_policy(struct ldlm_namespace *ns,
3500 struct ldlm_lock **lockp, void *req_cookie,
3501 ldlm_mode_t mode, int flags, void *data)
3503 struct mdt_thread_info *info;
3504 struct ptlrpc_request *req = req_cookie;
3505 struct ldlm_intent *it;
3506 struct req_capsule *pill;
3511 LASSERT(req != NULL);
3513 info = lu_context_key_get(&req->rq_svc_thread->t_env->le_ctx,
3515 LASSERT(info != NULL);
3516 pill = info->mti_pill;
3517 LASSERT(pill->rc_req == req);
3519 if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
3520 req_capsule_extend(pill, &RQF_LDLM_INTENT);
3521 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
3523 rc = mdt_intent_opc(it->opc, info, lockp, flags);
3527 /* Lock without inodebits makes no sense and will oops
3528 * later in ldlm. Let's check it now to see if we have
3529 * ibits corrupted somewhere in mdt_intent_opc().
3530 * The case for client miss to set ibits has been
3531 * processed by others. */
3532 LASSERT(ergo(info->mti_dlm_req->lock_desc.l_resource.\
3533 lr_type == LDLM_IBITS,
3534 info->mti_dlm_req->lock_desc.\
3535 l_policy_data.l_inodebits.bits != 0));
3537 rc = err_serious(-EFAULT);
3539 /* No intent was provided */
3540 LASSERT(pill->rc_fmt == &RQF_LDLM_ENQUEUE);
3541 rc = req_capsule_server_pack(pill);
3543 rc = err_serious(rc);
3548 static int mdt_seq_fini(const struct lu_env *env,
3549 struct mdt_device *m)
3551 struct md_site *ms = mdt_md_site(m);
3555 if (ms->ms_server_seq) {
3556 seq_server_fini(ms->ms_server_seq, env);
3557 OBD_FREE_PTR(ms->ms_server_seq);
3558 ms->ms_server_seq = NULL;
3561 if (ms->ms_control_seq) {
3562 seq_server_fini(ms->ms_control_seq, env);
3563 OBD_FREE_PTR(ms->ms_control_seq);
3564 ms->ms_control_seq = NULL;
3567 if (ms->ms_client_seq) {
3568 seq_client_fini(ms->ms_client_seq);
3569 OBD_FREE_PTR(ms->ms_client_seq);
3570 ms->ms_client_seq = NULL;
3577 static int mdt_seq_init(const struct lu_env *env,
3579 struct mdt_device *m)
3586 ms = mdt_md_site(m);
3589 * This is sequence-controller node. Init seq-controller server on local
3592 if (ms->ms_node_id == 0) {
3593 LASSERT(ms->ms_control_seq == NULL);
3595 OBD_ALLOC_PTR(ms->ms_control_seq);
3596 if (ms->ms_control_seq == NULL)
3599 rc = seq_server_init(ms->ms_control_seq,
3600 m->mdt_bottom, uuid,
3601 LUSTRE_SEQ_CONTROLLER,
3606 GOTO(out_seq_fini, rc);
3608 OBD_ALLOC_PTR(ms->ms_client_seq);
3609 if (ms->ms_client_seq == NULL)
3610 GOTO(out_seq_fini, rc = -ENOMEM);
3612 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3613 if (prefix == NULL) {
3614 OBD_FREE_PTR(ms->ms_client_seq);
3615 GOTO(out_seq_fini, rc = -ENOMEM);
3618 snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s",
3622 * Init seq-controller client after seq-controller server is
3623 * ready. Pass ms->ms_control_seq to it for direct talking.
3625 rc = seq_client_init(ms->ms_client_seq, NULL,
3626 LUSTRE_SEQ_METADATA, prefix,
3627 ms->ms_control_seq);
3628 OBD_FREE(prefix, MAX_OBD_NAME + 5);
3631 GOTO(out_seq_fini, rc);
3634 /* Init seq-server on local MDT */
3635 LASSERT(ms->ms_server_seq == NULL);
3637 OBD_ALLOC_PTR(ms->ms_server_seq);
3638 if (ms->ms_server_seq == NULL)
3639 GOTO(out_seq_fini, rc = -ENOMEM);
3641 rc = seq_server_init(ms->ms_server_seq,
3642 m->mdt_bottom, uuid,
3647 GOTO(out_seq_fini, rc = -ENOMEM);
3649 /* Assign seq-controller client to local seq-server. */
3650 if (ms->ms_node_id == 0) {
3651 LASSERT(ms->ms_client_seq != NULL);
3653 rc = seq_server_set_cli(ms->ms_server_seq,
3661 mdt_seq_fini(env, m);
3666 * Init client sequence manager which is used by local MDS to talk to sequence
3667 * controller on remote node.
3669 static int mdt_seq_init_cli(const struct lu_env *env,
3670 struct mdt_device *m,
3671 struct lustre_cfg *cfg)
3673 struct md_site *ms = mdt_md_site(m);
3674 struct obd_device *mdc;
3675 struct obd_uuid *uuidp, *mdcuuidp;
3676 char *uuid_str, *mdc_uuid_str;
3679 struct mdt_thread_info *info;
3680 char *p, *index_string = lustre_cfg_string(cfg, 2);
3683 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3684 uuidp = &info->mti_u.uuid[0];
3685 mdcuuidp = &info->mti_u.uuid[1];
3687 LASSERT(index_string);
3689 index = simple_strtol(index_string, &p, 10);
3691 CERROR("Invalid index in lustre_cgf, offset 2\n");
3695 /* check if this is adding the first MDC and controller is not yet
3697 if (index != 0 || ms->ms_client_seq)
3700 uuid_str = lustre_cfg_string(cfg, 1);
3701 mdc_uuid_str = lustre_cfg_string(cfg, 4);
3702 obd_str2uuid(uuidp, uuid_str);
3703 obd_str2uuid(mdcuuidp, mdc_uuid_str);
3705 mdc = class_find_client_obd(uuidp, LUSTRE_MDC_NAME, mdcuuidp);
3707 CERROR("can't find controller MDC by uuid %s\n",
3710 } else if (!mdc->obd_set_up) {
3711 CERROR("target %s not set up\n", mdc->obd_name);
3714 LASSERT(ms->ms_control_exp);
3715 OBD_ALLOC_PTR(ms->ms_client_seq);
3716 if (ms->ms_client_seq != NULL) {
3719 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3723 snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s",
3726 rc = seq_client_init(ms->ms_client_seq,
3728 LUSTRE_SEQ_METADATA,
3730 OBD_FREE(prefix, MAX_OBD_NAME + 5);
3737 LASSERT(ms->ms_server_seq != NULL);
3738 rc = seq_server_set_cli(ms->ms_server_seq, ms->ms_client_seq,
3745 static void mdt_seq_fini_cli(struct mdt_device *m)
3751 ms = mdt_md_site(m);
3754 if (ms->ms_server_seq)
3755 seq_server_set_cli(ms->ms_server_seq,
3758 if (ms->ms_control_exp) {
3759 class_export_put(ms->ms_control_exp);
3760 ms->ms_control_exp = NULL;
3769 static int mdt_fld_fini(const struct lu_env *env,
3770 struct mdt_device *m)
3772 struct md_site *ms = mdt_md_site(m);
3775 if (ms && ms->ms_server_fld) {
3776 fld_server_fini(ms->ms_server_fld, env);
3777 OBD_FREE_PTR(ms->ms_server_fld);
3778 ms->ms_server_fld = NULL;
3784 static int mdt_fld_init(const struct lu_env *env,
3786 struct mdt_device *m)
3792 ms = mdt_md_site(m);
3794 OBD_ALLOC_PTR(ms->ms_server_fld);
3795 if (ms->ms_server_fld == NULL)
3796 RETURN(rc = -ENOMEM);
3798 rc = fld_server_init(ms->ms_server_fld,
3799 m->mdt_bottom, uuid,
3800 env, ms->ms_node_id);
3802 OBD_FREE_PTR(ms->ms_server_fld);
3803 ms->ms_server_fld = NULL;
3810 /* device init/fini methods */
3811 static void mdt_stop_ptlrpc_service(struct mdt_device *m)
3814 if (m->mdt_regular_service != NULL) {
3815 ptlrpc_unregister_service(m->mdt_regular_service);
3816 m->mdt_regular_service = NULL;
3818 if (m->mdt_readpage_service != NULL) {
3819 ptlrpc_unregister_service(m->mdt_readpage_service);
3820 m->mdt_readpage_service = NULL;
3822 if (m->mdt_xmds_service != NULL) {
3823 ptlrpc_unregister_service(m->mdt_xmds_service);
3824 m->mdt_xmds_service = NULL;
3826 if (m->mdt_setattr_service != NULL) {
3827 ptlrpc_unregister_service(m->mdt_setattr_service);
3828 m->mdt_setattr_service = NULL;
3830 if (m->mdt_mdsc_service != NULL) {
3831 ptlrpc_unregister_service(m->mdt_mdsc_service);
3832 m->mdt_mdsc_service = NULL;
3834 if (m->mdt_mdss_service != NULL) {
3835 ptlrpc_unregister_service(m->mdt_mdss_service);
3836 m->mdt_mdss_service = NULL;
3838 if (m->mdt_dtss_service != NULL) {
3839 ptlrpc_unregister_service(m->mdt_dtss_service);
3840 m->mdt_dtss_service = NULL;
3842 if (m->mdt_fld_service != NULL) {
3843 ptlrpc_unregister_service(m->mdt_fld_service);
3844 m->mdt_fld_service = NULL;
3849 static int mdt_start_ptlrpc_service(struct mdt_device *m)
3852 static struct ptlrpc_service_conf conf;
3853 cfs_proc_dir_entry_t *procfs_entry;
3856 procfs_entry = m->mdt_md_dev.md_lu_dev.ld_obd->obd_proc_entry;
3858 conf = (typeof(conf)) {
3859 .psc_nbufs = MDS_NBUFS,
3860 .psc_bufsize = MDS_BUFSIZE,
3861 .psc_max_req_size = MDS_MAXREQSIZE,
3862 .psc_max_reply_size = MDS_MAXREPSIZE,
3863 .psc_req_portal = MDS_REQUEST_PORTAL,
3864 .psc_rep_portal = MDC_REPLY_PORTAL,
3865 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3867 * We'd like to have a mechanism to set this on a per-device
3868 * basis, but alas...
3870 .psc_min_threads = mdt_min_threads,
3871 .psc_max_threads = mdt_max_threads,
3872 .psc_ctx_tags = LCT_MD_THREAD
3875 m->mdt_ldlm_client = &m->mdt_md_dev.md_lu_dev.ld_obd->obd_ldlm_client;
3876 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
3877 "mdt_ldlm_client", m->mdt_ldlm_client);
3879 m->mdt_regular_service =
3880 ptlrpc_init_svc_conf(&conf, mdt_regular_handle, LUSTRE_MDT_NAME,
3881 procfs_entry, target_print_req,
3883 if (m->mdt_regular_service == NULL)
3886 rc = ptlrpc_start_threads(NULL, m->mdt_regular_service);
3888 GOTO(err_mdt_svc, rc);
3891 * readpage service configuration. Parameters have to be adjusted,
3894 conf = (typeof(conf)) {
3895 .psc_nbufs = MDS_NBUFS,
3896 .psc_bufsize = MDS_BUFSIZE,
3897 .psc_max_req_size = MDS_MAXREQSIZE,
3898 .psc_max_reply_size = MDS_MAXREPSIZE,
3899 .psc_req_portal = MDS_READPAGE_PORTAL,
3900 .psc_rep_portal = MDC_REPLY_PORTAL,
3901 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3902 .psc_min_threads = mdt_min_threads,
3903 .psc_max_threads = mdt_max_threads,
3904 .psc_ctx_tags = LCT_MD_THREAD
3906 m->mdt_readpage_service =
3907 ptlrpc_init_svc_conf(&conf, mdt_readpage_handle,
3908 LUSTRE_MDT_NAME "_readpage",
3909 procfs_entry, target_print_req,"mdt_rdpg");
3911 if (m->mdt_readpage_service == NULL) {
3912 CERROR("failed to start readpage service\n");
3913 GOTO(err_mdt_svc, rc = -ENOMEM);
3916 rc = ptlrpc_start_threads(NULL, m->mdt_readpage_service);
3919 * setattr service configuration.
3921 conf = (typeof(conf)) {
3922 .psc_nbufs = MDS_NBUFS,
3923 .psc_bufsize = MDS_BUFSIZE,
3924 .psc_max_req_size = MDS_MAXREQSIZE,
3925 .psc_max_reply_size = MDS_MAXREPSIZE,
3926 .psc_req_portal = MDS_SETATTR_PORTAL,
3927 .psc_rep_portal = MDC_REPLY_PORTAL,
3928 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3929 .psc_min_threads = mdt_min_threads,
3930 .psc_max_threads = mdt_max_threads,
3931 .psc_ctx_tags = LCT_MD_THREAD
3934 m->mdt_setattr_service =
3935 ptlrpc_init_svc_conf(&conf, mdt_regular_handle,
3936 LUSTRE_MDT_NAME "_setattr",
3937 procfs_entry, target_print_req,"mdt_attr");
3939 if (!m->mdt_setattr_service) {
3940 CERROR("failed to start setattr service\n");
3941 GOTO(err_mdt_svc, rc = -ENOMEM);
3944 rc = ptlrpc_start_threads(NULL, m->mdt_setattr_service);
3946 GOTO(err_mdt_svc, rc);
3949 * sequence controller service configuration
3951 conf = (typeof(conf)) {
3952 .psc_nbufs = MDS_NBUFS,
3953 .psc_bufsize = MDS_BUFSIZE,
3954 .psc_max_req_size = SEQ_MAXREQSIZE,
3955 .psc_max_reply_size = SEQ_MAXREPSIZE,
3956 .psc_req_portal = SEQ_CONTROLLER_PORTAL,
3957 .psc_rep_portal = MDC_REPLY_PORTAL,
3958 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3959 .psc_min_threads = mdt_min_threads,
3960 .psc_max_threads = mdt_max_threads,
3961 .psc_ctx_tags = LCT_MD_THREAD|LCT_DT_THREAD
3964 m->mdt_mdsc_service =
3965 ptlrpc_init_svc_conf(&conf, mdt_mdsc_handle,
3966 LUSTRE_MDT_NAME"_mdsc",
3967 procfs_entry, target_print_req,"mdt_mdsc");
3968 if (!m->mdt_mdsc_service) {
3969 CERROR("failed to start seq controller service\n");
3970 GOTO(err_mdt_svc, rc = -ENOMEM);
3973 rc = ptlrpc_start_threads(NULL, m->mdt_mdsc_service);
3975 GOTO(err_mdt_svc, rc);
3978 * metadata sequence server service configuration
3980 conf = (typeof(conf)) {
3981 .psc_nbufs = MDS_NBUFS,
3982 .psc_bufsize = MDS_BUFSIZE,
3983 .psc_max_req_size = SEQ_MAXREQSIZE,
3984 .psc_max_reply_size = SEQ_MAXREPSIZE,
3985 .psc_req_portal = SEQ_METADATA_PORTAL,
3986 .psc_rep_portal = MDC_REPLY_PORTAL,
3987 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3988 .psc_min_threads = mdt_min_threads,
3989 .psc_max_threads = mdt_max_threads,
3990 .psc_ctx_tags = LCT_MD_THREAD|LCT_DT_THREAD
3993 m->mdt_mdss_service =
3994 ptlrpc_init_svc_conf(&conf, mdt_mdss_handle,
3995 LUSTRE_MDT_NAME"_mdss",
3996 procfs_entry, target_print_req,"mdt_mdss");
3997 if (!m->mdt_mdss_service) {
3998 CERROR("failed to start metadata seq server service\n");
3999 GOTO(err_mdt_svc, rc = -ENOMEM);
4002 rc = ptlrpc_start_threads(NULL, m->mdt_mdss_service);
4004 GOTO(err_mdt_svc, rc);
4008 * Data sequence server service configuration. We want to have really
4009 * cluster-wide sequences space. This is why we start only one sequence
4010 * controller which manages space.
4012 conf = (typeof(conf)) {
4013 .psc_nbufs = MDS_NBUFS,
4014 .psc_bufsize = MDS_BUFSIZE,
4015 .psc_max_req_size = SEQ_MAXREQSIZE,
4016 .psc_max_reply_size = SEQ_MAXREPSIZE,
4017 .psc_req_portal = SEQ_DATA_PORTAL,
4018 .psc_rep_portal = OSC_REPLY_PORTAL,
4019 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
4020 .psc_min_threads = mdt_min_threads,
4021 .psc_max_threads = mdt_max_threads,
4022 .psc_ctx_tags = LCT_MD_THREAD|LCT_DT_THREAD
4025 m->mdt_dtss_service =
4026 ptlrpc_init_svc_conf(&conf, mdt_dtss_handle,
4027 LUSTRE_MDT_NAME"_dtss",
4028 procfs_entry, target_print_req,"mdt_dtss");
4029 if (!m->mdt_dtss_service) {
4030 CERROR("failed to start data seq server service\n");
4031 GOTO(err_mdt_svc, rc = -ENOMEM);
4034 rc = ptlrpc_start_threads(NULL, m->mdt_dtss_service);
4036 GOTO(err_mdt_svc, rc);
4038 /* FLD service start */
4039 conf = (typeof(conf)) {
4040 .psc_nbufs = MDS_NBUFS,
4041 .psc_bufsize = MDS_BUFSIZE,
4042 .psc_max_req_size = FLD_MAXREQSIZE,
4043 .psc_max_reply_size = FLD_MAXREPSIZE,
4044 .psc_req_portal = FLD_REQUEST_PORTAL,
4045 .psc_rep_portal = MDC_REPLY_PORTAL,
4046 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
4047 .psc_min_threads = mdt_min_threads,
4048 .psc_max_threads = mdt_max_threads,
4049 .psc_ctx_tags = LCT_DT_THREAD|LCT_MD_THREAD
4052 m->mdt_fld_service =
4053 ptlrpc_init_svc_conf(&conf, mdt_fld_handle,
4054 LUSTRE_MDT_NAME"_fld",
4055 procfs_entry, target_print_req, "mdt_fld");
4056 if (!m->mdt_fld_service) {
4057 CERROR("failed to start fld service\n");
4058 GOTO(err_mdt_svc, rc = -ENOMEM);
4061 rc = ptlrpc_start_threads(NULL, m->mdt_fld_service);
4063 GOTO(err_mdt_svc, rc);
4066 * mds-mds service configuration. Separate portal is used to allow
4067 * mds-mds requests be not blocked during recovery.
4069 conf = (typeof(conf)) {
4070 .psc_nbufs = MDS_NBUFS,
4071 .psc_bufsize = MDS_BUFSIZE,
4072 .psc_max_req_size = MDS_MAXREQSIZE,
4073 .psc_max_reply_size = MDS_MAXREPSIZE,
4074 .psc_req_portal = MDS_MDS_PORTAL,
4075 .psc_rep_portal = MDC_REPLY_PORTAL,
4076 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
4077 .psc_min_threads = mdt_min_threads,
4078 .psc_max_threads = mdt_max_threads,
4079 .psc_ctx_tags = LCT_MD_THREAD
4081 m->mdt_xmds_service =
4082 ptlrpc_init_svc_conf(&conf, mdt_xmds_handle,
4083 LUSTRE_MDT_NAME "_mds",
4084 procfs_entry, target_print_req,"mdt_xmds");
4086 if (m->mdt_xmds_service == NULL) {
4087 CERROR("failed to start xmds service\n");
4088 GOTO(err_mdt_svc, rc = -ENOMEM);
4091 rc = ptlrpc_start_threads(NULL, m->mdt_xmds_service);
4093 GOTO(err_mdt_svc, rc);
4098 mdt_stop_ptlrpc_service(m);
4103 static void mdt_stack_fini(const struct lu_env *env,
4104 struct mdt_device *m, struct lu_device *top)
4106 struct obd_device *obd = mdt2obd_dev(m);
4107 struct lustre_cfg_bufs *bufs;
4108 struct lustre_cfg *lcfg;
4109 struct mdt_thread_info *info;
4113 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4114 LASSERT(info != NULL);
4116 bufs = &info->mti_u.bufs;
4117 /* process cleanup, pass mdt obd name to get obd umount flags */
4118 lustre_cfg_bufs_reset(bufs, obd->obd_name);
4123 lustre_cfg_bufs_set_string(bufs, 1, flags);
4124 lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
4126 CERROR("Cannot alloc lcfg!\n");
4131 top->ld_ops->ldo_process_config(env, top, lcfg);
4132 lustre_cfg_free(lcfg);
4134 lu_stack_fini(env, top);
4135 m->mdt_child = NULL;
4136 m->mdt_bottom = NULL;
4139 static struct lu_device *mdt_layer_setup(struct lu_env *env,
4140 const char *typename,
4141 struct lu_device *child,
4142 struct lustre_cfg *cfg)
4144 const char *dev = lustre_cfg_string(cfg, 0);
4145 struct obd_type *type;
4146 struct lu_device_type *ldt;
4147 struct lu_device *d;
4152 type = class_get_type(typename);
4154 CERROR("Unknown type: '%s'\n", typename);
4155 GOTO(out, rc = -ENODEV);
4158 rc = lu_env_refill((struct lu_env *)env);
4160 CERROR("Failure to refill session: '%d'\n", rc);
4166 CERROR("type: '%s'\n", typename);
4167 GOTO(out_type, rc = -EINVAL);
4170 ldt->ldt_obd_type = type;
4171 d = ldt->ldt_ops->ldto_device_alloc(env, ldt, cfg);
4173 CERROR("Cannot allocate device: '%s'\n", typename);
4174 GOTO(out_type, rc = -ENODEV);
4177 LASSERT(child->ld_site);
4178 d->ld_site = child->ld_site;
4181 rc = ldt->ldt_ops->ldto_device_init(env, d, dev, child);
4183 CERROR("can't init device '%s', rc %d\n", typename, rc);
4184 GOTO(out_alloc, rc);
4187 lu_ref_add(&d->ld_reference, "lu-stack", &lu_site_init);
4192 ldt->ldt_ops->ldto_device_free(env, d);
4195 class_put_type(type);
4200 static int mdt_stack_init(struct lu_env *env,
4201 struct mdt_device *m,
4202 struct lustre_cfg *cfg,
4203 struct lustre_mount_info *lmi)
4205 struct lu_device *d = &m->mdt_md_dev.md_lu_dev;
4206 struct lu_device *tmp;
4207 struct md_device *md;
4208 struct lu_device *child_lu_dev;
4212 /* init the stack */
4213 tmp = mdt_layer_setup(env, LUSTRE_OSD_NAME, d, cfg);
4215 RETURN(PTR_ERR(tmp));
4217 m->mdt_bottom = lu2dt_dev(tmp);
4219 tmp = mdt_layer_setup(env, LUSTRE_MDD_NAME, d, cfg);
4221 GOTO(out, rc = PTR_ERR(tmp));
4226 tmp = mdt_layer_setup(env, LUSTRE_CMM_NAME, d, cfg);
4228 GOTO(out, rc = PTR_ERR(tmp));
4231 /*set mdd upcall device*/
4232 md_upcall_dev_set(md, lu2md_dev(d));
4235 /*set cmm upcall device*/
4236 md_upcall_dev_set(md, &m->mdt_md_dev);
4238 m->mdt_child = lu2md_dev(d);
4240 /* process setup config */
4241 tmp = &m->mdt_md_dev.md_lu_dev;
4242 rc = tmp->ld_ops->ldo_process_config(env, tmp, cfg);
4246 /* initialize local objects */
4247 child_lu_dev = &m->mdt_child->md_lu_dev;
4249 rc = child_lu_dev->ld_ops->ldo_prepare(env,
4250 &m->mdt_md_dev.md_lu_dev,
4253 /* fini from last known good lu_device */
4255 mdt_stack_fini(env, m, d);
4261 * setup CONFIG_ORIG context, used to access local config log.
4262 * this may need to be rewrite as part of llog rewrite for lu-api.
4264 static int mdt_obd_llog_setup(struct obd_device *obd,
4265 struct lustre_sb_info *lsi)
4269 LASSERT(obd->obd_fsops == NULL);
4271 obd->obd_fsops = fsfilt_get_ops(MT_STR(lsi->lsi_ldd));
4272 if (IS_ERR(obd->obd_fsops))
4273 return PTR_ERR(obd->obd_fsops);
4275 rc = fsfilt_setup(obd, lsi->lsi_srv_mnt->mnt_sb);
4277 fsfilt_put_ops(obd->obd_fsops);
4281 OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
4282 obd->obd_lvfs_ctxt.pwdmnt = lsi->lsi_srv_mnt;
4283 obd->obd_lvfs_ctxt.pwd = lsi->lsi_srv_mnt->mnt_root;
4284 obd->obd_lvfs_ctxt.fs = get_ds();
4286 rc = llog_setup(obd, &obd->obd_olg, LLOG_CONFIG_ORIG_CTXT, obd,
4287 0, NULL, &llog_lvfs_ops);
4289 CERROR("llog_setup() failed: %d\n", rc);
4290 fsfilt_put_ops(obd->obd_fsops);
4296 static void mdt_obd_llog_cleanup(struct obd_device *obd)
4298 struct llog_ctxt *ctxt;
4300 ctxt = llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT);
4304 if (obd->obd_fsops) {
4305 fsfilt_put_ops(obd->obd_fsops);
4306 obd->obd_fsops = NULL;
4310 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
4312 struct md_device *next = m->mdt_child;
4313 struct lu_device *d = &m->mdt_md_dev.md_lu_dev;
4314 struct lu_site *ls = d->ld_site;
4315 struct obd_device *obd = mdt2obd_dev(m);
4318 target_recovery_fini(obd);
4320 ping_evictor_stop();
4322 mdt_stop_ptlrpc_service(m);
4323 mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4324 mdt_obd_llog_cleanup(obd);
4325 obd_exports_barrier(obd);
4326 obd_zombie_barrier();
4327 #ifdef HAVE_QUOTA_SUPPORT
4328 next->md_ops->mdo_quota.mqo_cleanup(env, next);
4330 lut_fini(env, &m->mdt_lut);
4331 mdt_fs_cleanup(env, m);
4332 upcall_cache_cleanup(m->mdt_identity_cache);
4333 m->mdt_identity_cache = NULL;
4335 if (m->mdt_namespace != NULL) {
4336 ldlm_namespace_free(m->mdt_namespace, NULL,
4337 d->ld_obd->obd_force);
4338 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
4341 cfs_free_nidlist(&m->mdt_nosquash_nids);
4342 if (m->mdt_nosquash_str) {
4343 OBD_FREE(m->mdt_nosquash_str, m->mdt_nosquash_strlen);
4344 m->mdt_nosquash_str = NULL;
4345 m->mdt_nosquash_strlen = 0;
4348 mdt_seq_fini(env, m);
4349 mdt_seq_fini_cli(m);
4350 mdt_fld_fini(env, m);
4351 sptlrpc_rule_set_free(&m->mdt_sptlrpc_rset);
4353 next->md_ops->mdo_init_capa_ctxt(env, next, 0, 0, 0, NULL);
4354 cfs_timer_disarm(&m->mdt_ck_timer);
4355 mdt_ck_thread_stop(m);
4360 mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4362 lprocfs_free_per_client_stats(obd);
4363 lprocfs_free_obd_stats(obd);
4367 struct md_site *mite;
4370 mite = lu_site2md(ls);
4374 LASSERT(cfs_atomic_read(&d->ld_ref) == 0);
4379 static int mdt_adapt_sptlrpc_conf(struct obd_device *obd, int initial)
4381 struct mdt_device *m = mdt_dev(obd->obd_lu_dev);
4382 struct sptlrpc_rule_set tmp_rset;
4385 sptlrpc_rule_set_init(&tmp_rset);
4386 rc = sptlrpc_conf_target_get_rules(obd, &tmp_rset, initial);
4388 CERROR("mdt %s: failed get sptlrpc rules: %d\n",
4393 sptlrpc_target_update_exp_flavor(obd, &tmp_rset);
4395 cfs_write_lock(&m->mdt_sptlrpc_lock);
4396 sptlrpc_rule_set_free(&m->mdt_sptlrpc_rset);
4397 m->mdt_sptlrpc_rset = tmp_rset;
4398 cfs_write_unlock(&m->mdt_sptlrpc_lock);
4403 static void fsoptions_to_mdt_flags(struct mdt_device *m, char *options)
4407 m->mdt_opts.mo_mds_capa = 1;
4408 m->mdt_opts.mo_oss_capa = 1;
4409 #ifdef CONFIG_FS_POSIX_ACL
4410 /* ACLs should be enabled by default (b=13829) */
4411 m->mdt_opts.mo_acl = 1;
4412 LCONSOLE_INFO("Enabling ACL\n");
4414 m->mdt_opts.mo_acl = 0;
4415 LCONSOLE_INFO("Disabling ACL\n");
4424 while (*p && *p != ',')
4428 if ((len == sizeof("user_xattr") - 1) &&
4429 (memcmp(options, "user_xattr", len) == 0)) {
4430 m->mdt_opts.mo_user_xattr = 1;
4431 LCONSOLE_INFO("Enabling user_xattr\n");
4432 } else if ((len == sizeof("nouser_xattr") - 1) &&
4433 (memcmp(options, "nouser_xattr", len) == 0)) {
4434 m->mdt_opts.mo_user_xattr = 0;
4435 LCONSOLE_INFO("Disabling user_xattr\n");
4436 } else if ((len == sizeof("noacl") - 1) &&
4437 (memcmp(options, "noacl", len) == 0)) {
4438 m->mdt_opts.mo_acl = 0;
4439 LCONSOLE_INFO("Disabling ACL\n");
4446 int mdt_postrecov(const struct lu_env *, struct mdt_device *);
4448 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
4449 struct lu_device_type *ldt, struct lustre_cfg *cfg)
4451 struct mdt_thread_info *info;
4452 struct obd_device *obd;
4453 const char *dev = lustre_cfg_string(cfg, 0);
4454 const char *num = lustre_cfg_string(cfg, 2);
4455 struct lustre_mount_info *lmi = NULL;
4456 struct lustre_sb_info *lsi;
4457 struct lustre_disk_data *ldd;
4459 struct md_site *mite;
4460 const char *identity_upcall = "NONE";
4461 #ifdef HAVE_QUOTA_SUPPORT
4462 struct md_device *next;
4468 md_device_init(&m->mdt_md_dev, ldt);
4470 * Environment (env) might be missing mdt_thread_key values at that
4471 * point, if device is allocated when mdt_thread_key is in QUIESCENT
4474 * Usually device allocation path doesn't use module key values, but
4475 * mdt has to do a lot of work here, so allocate key value.
4477 rc = lu_env_refill((struct lu_env *)env);
4481 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4482 LASSERT(info != NULL);
4484 obd = class_name2obd(dev);
4485 LASSERT(obd != NULL);
4487 m->mdt_max_mdsize = MAX_MD_SIZE;
4488 m->mdt_max_cookiesize = sizeof(struct llog_cookie);
4489 m->mdt_som_conf = 0;
4491 m->mdt_opts.mo_user_xattr = 0;
4492 m->mdt_opts.mo_acl = 0;
4493 m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
4494 lmi = server_get_mount_2(dev);
4496 CERROR("Cannot get mount info for %s!\n", dev);
4499 lsi = s2lsi(lmi->lmi_sb);
4500 fsoptions_to_mdt_flags(m, lsi->lsi_lmd->lmd_opts);
4501 /* CMD is supported only in IAM mode */
4504 node_id = simple_strtol(num, NULL, 10);
4505 if (!(ldd->ldd_flags & LDD_F_IAM_DIR) && node_id) {
4506 CERROR("CMD Operation not allowed in IOP mode\n");
4507 GOTO(err_lmi, rc = -EINVAL);
4509 /* Read recovery timeouts */
4510 if (lsi->lsi_lmd && lsi->lsi_lmd->lmd_recovery_time_soft)
4511 obd->obd_recovery_timeout =
4512 lsi->lsi_lmd->lmd_recovery_time_soft;
4514 if (lsi->lsi_lmd && lsi->lsi_lmd->lmd_recovery_time_hard)
4515 obd->obd_recovery_time_hard =
4516 lsi->lsi_lmd->lmd_recovery_time_hard;
4519 cfs_rwlock_init(&m->mdt_sptlrpc_lock);
4520 sptlrpc_rule_set_init(&m->mdt_sptlrpc_rset);
4522 cfs_spin_lock_init(&m->mdt_ioepoch_lock);
4523 m->mdt_opts.mo_compat_resname = 0;
4524 m->mdt_capa_timeout = CAPA_TIMEOUT;
4525 m->mdt_capa_alg = CAPA_HMAC_ALG_SHA1;
4526 m->mdt_ck_timeout = CAPA_KEY_TIMEOUT;
4527 m->mdt_squash_uid = 0;
4528 m->mdt_squash_gid = 0;
4529 CFS_INIT_LIST_HEAD(&m->mdt_nosquash_nids);
4530 m->mdt_nosquash_str = NULL;
4531 m->mdt_nosquash_strlen = 0;
4532 cfs_init_rwsem(&m->mdt_squash_sem);
4534 OBD_ALLOC_PTR(mite);
4536 GOTO(err_lmi, rc = -ENOMEM);
4540 m->mdt_md_dev.md_lu_dev.ld_ops = &mdt_lu_ops;
4541 m->mdt_md_dev.md_lu_dev.ld_obd = obd;
4542 /* set this lu_device to obd, because error handling need it */
4543 obd->obd_lu_dev = &m->mdt_md_dev.md_lu_dev;
4545 rc = lu_site_init(s, &m->mdt_md_dev.md_lu_dev);
4547 CERROR("Can't init lu_site, rc %d\n", rc);
4548 GOTO(err_free_site, rc);
4551 rc = mdt_procfs_init(m, dev);
4553 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
4554 GOTO(err_fini_proc, rc);
4557 /* set server index */
4558 lu_site2md(s)->ms_node_id = node_id;
4560 /* failover is the default
4561 * FIXME: we do not failout mds0/mgs, which may cause some problems.
4562 * assumed whose ms_node_id == 0 XXX
4564 obd->obd_replayable = 1;
4565 /* No connection accepted until configurations will finish */
4566 obd->obd_no_conn = 1;
4568 if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
4569 char *str = lustre_cfg_string(cfg, 4);
4570 if (strchr(str, 'n')) {
4571 CWARN("%s: recovery disabled\n", obd->obd_name);
4572 obd->obd_replayable = 0;
4576 /* init the stack */
4577 rc = mdt_stack_init((struct lu_env *)env, m, cfg, lmi);
4579 CERROR("Can't init device stack, rc %d\n", rc);
4580 GOTO(err_fini_proc, rc);
4583 rc = lut_init(env, &m->mdt_lut, obd, m->mdt_bottom);
4585 GOTO(err_fini_stack, rc);
4587 rc = mdt_fld_init(env, obd->obd_name, m);
4591 rc = mdt_seq_init(env, obd->obd_name, m);
4593 GOTO(err_fini_fld, rc);
4595 snprintf(info->mti_u.ns_name, sizeof info->mti_u.ns_name,
4596 LUSTRE_MDT_NAME"-%p", m);
4597 m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
4598 LDLM_NAMESPACE_SERVER,
4599 LDLM_NAMESPACE_GREEDY);
4600 if (m->mdt_namespace == NULL)
4601 GOTO(err_fini_seq, rc = -ENOMEM);
4603 ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
4604 /* set obd_namespace for compatibility with old code */
4605 obd->obd_namespace = m->mdt_namespace;
4607 /* XXX: to support suppgid for ACL, we enable identity_upcall
4608 * by default, otherwise, maybe got unexpected -EACCESS. */
4609 if (m->mdt_opts.mo_acl)
4610 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
4612 m->mdt_identity_cache = upcall_cache_init(obd->obd_name, identity_upcall,
4613 &mdt_identity_upcall_cache_ops);
4614 if (IS_ERR(m->mdt_identity_cache)) {
4615 rc = PTR_ERR(m->mdt_identity_cache);
4616 m->mdt_identity_cache = NULL;
4617 GOTO(err_free_ns, rc);
4620 cfs_timer_init(&m->mdt_ck_timer, mdt_ck_timer_callback, m);
4622 rc = mdt_ck_thread_start(m);
4624 GOTO(err_free_ns, rc);
4626 rc = mdt_fs_setup(env, m, obd, lsi);
4630 rc = mdt_obd_llog_setup(obd, lsi);
4632 GOTO(err_fs_cleanup, rc);
4634 rc = mdt_llog_ctxt_clone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4636 GOTO(err_llog_cleanup, rc);
4638 mdt_adapt_sptlrpc_conf(obd, 1);
4640 #ifdef HAVE_QUOTA_SUPPORT
4641 next = m->mdt_child;
4642 rc = next->md_ops->mdo_quota.mqo_setup(env, next, lmi->lmi_mnt);
4644 GOTO(err_llog_cleanup, rc);
4647 server_put_mount_2(dev, lmi->lmi_mnt);
4650 target_recovery_init(&m->mdt_lut, mdt_recovery_handle);
4652 rc = mdt_start_ptlrpc_service(m);
4654 GOTO(err_recovery, rc);
4656 ping_evictor_start();
4658 rc = lu_site_init_finish(s);
4660 GOTO(err_stop_service, rc);
4662 if (obd->obd_recovering == 0)
4663 mdt_postrecov(env, m);
4665 mdt_init_capa_ctxt(env, m);
4667 /* Reduce the initial timeout on an MDS because it doesn't need such
4668 * a long timeout as an OST does. Adaptive timeouts will adjust this
4669 * value appropriately. */
4670 if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
4671 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
4676 ping_evictor_stop();
4677 mdt_stop_ptlrpc_service(m);
4679 target_recovery_fini(obd);
4680 #ifdef HAVE_QUOTA_SUPPORT
4681 next->md_ops->mdo_quota.mqo_cleanup(env, next);
4684 mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4685 mdt_obd_llog_cleanup(obd);
4687 mdt_fs_cleanup(env, m);
4689 cfs_timer_disarm(&m->mdt_ck_timer);
4690 mdt_ck_thread_stop(m);
4692 upcall_cache_cleanup(m->mdt_identity_cache);
4693 m->mdt_identity_cache = NULL;
4694 ldlm_namespace_free(m->mdt_namespace, NULL, 0);
4695 obd->obd_namespace = m->mdt_namespace = NULL;
4697 mdt_seq_fini(env, m);
4699 mdt_fld_fini(env, m);
4701 lut_fini(env, &m->mdt_lut);
4703 mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4711 server_put_mount_2(dev, lmi->lmi_mnt);
4715 /* used by MGS to process specific configurations */
4716 static int mdt_process_config(const struct lu_env *env,
4717 struct lu_device *d, struct lustre_cfg *cfg)
4719 struct mdt_device *m = mdt_dev(d);
4720 struct md_device *md_next = m->mdt_child;
4721 struct lu_device *next = md2lu_dev(md_next);
4725 switch (cfg->lcfg_command) {
4727 struct lprocfs_static_vars lvars;
4728 struct obd_device *obd = d->ld_obd;
4731 * For interoperability between 1.8 and 2.0,
4732 * skip old "mdt.group_upcall" param.
4735 char *param = lustre_cfg_string(cfg, 1);
4736 if (param && !strncmp("mdt.group_upcall", param, 16)) {
4737 CWARN("For 1.8 interoperability, skip this"
4738 " mdt.group_upcall. It is obsolete\n");
4743 lprocfs_mdt_init_vars(&lvars);
4744 rc = class_process_proc_param(PARAM_MDT, lvars.obd_vars,
4746 if (rc > 0 || rc == -ENOSYS)
4747 /* we don't understand; pass it on */
4748 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4753 * Add mdc hook to get first MDT uuid and connect it to
4754 * ls->controller to use for seq manager.
4756 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4758 CERROR("Can't add mdc, rc %d\n", rc);
4760 rc = mdt_seq_init_cli(env, mdt_dev(d), cfg);
4763 /* others are passed further */
4764 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4770 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
4771 const struct lu_object_header *hdr,
4772 struct lu_device *d)
4774 struct mdt_object *mo;
4780 struct lu_object *o;
4781 struct lu_object_header *h;
4783 o = &mo->mot_obj.mo_lu;
4784 h = &mo->mot_header;
4785 lu_object_header_init(h);
4786 lu_object_init(o, h, d);
4787 lu_object_add_top(h, o);
4788 o->lo_ops = &mdt_obj_ops;
4789 cfs_sema_init(&mo->mot_ioepoch_sem, 1);
4795 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
4796 const struct lu_object_conf *unused)
4798 struct mdt_device *d = mdt_dev(o->lo_dev);
4799 struct lu_device *under;
4800 struct lu_object *below;
4804 CDEBUG(D_INFO, "object init, fid = "DFID"\n",
4805 PFID(lu_object_fid(o)));
4807 under = &d->mdt_child->md_lu_dev;
4808 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
4809 if (below != NULL) {
4810 lu_object_add(o, below);
4817 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
4819 struct mdt_object *mo = mdt_obj(o);
4820 struct lu_object_header *h;
4824 CDEBUG(D_INFO, "object free, fid = "DFID"\n",
4825 PFID(lu_object_fid(o)));
4828 lu_object_header_fini(h);
4833 static int mdt_object_print(const struct lu_env *env, void *cookie,
4834 lu_printer_t p, const struct lu_object *o)
4836 struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
4837 return (*p)(env, cookie, LUSTRE_MDT_NAME"-object@%p(ioepoch="LPU64" "
4838 "flags="LPX64", epochcount=%d, writecount=%d)",
4839 mdto, mdto->mot_ioepoch, mdto->mot_flags,
4840 mdto->mot_ioepoch_count, mdto->mot_writecount);
4843 static const struct lu_device_operations mdt_lu_ops = {
4844 .ldo_object_alloc = mdt_object_alloc,
4845 .ldo_process_config = mdt_process_config,
4848 static const struct lu_object_operations mdt_obj_ops = {
4849 .loo_object_init = mdt_object_init,
4850 .loo_object_free = mdt_object_free,
4851 .loo_object_print = mdt_object_print
4854 static int mdt_obd_set_info_async(struct obd_export *exp,
4855 __u32 keylen, void *key,
4856 __u32 vallen, void *val,
4857 struct ptlrpc_request_set *set)
4859 struct obd_device *obd = exp->exp_obd;
4865 if (KEY_IS(KEY_SPTLRPC_CONF)) {
4866 rc = mdt_adapt_sptlrpc_conf(obd, 0);
4873 /* mds_connect_internal */
4874 static int mdt_connect_internal(struct obd_export *exp,
4875 struct mdt_device *mdt,
4876 struct obd_connect_data *data)
4879 data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
4880 data->ocd_ibits_known &= MDS_INODELOCK_FULL;
4882 /* If no known bits (which should not happen, probably,
4883 as everybody should support LOOKUP and UPDATE bits at least)
4884 revert to compat mode with plain locks. */
4885 if (!data->ocd_ibits_known &&
4886 data->ocd_connect_flags & OBD_CONNECT_IBITS)
4887 data->ocd_connect_flags &= ~OBD_CONNECT_IBITS;
4889 if (!mdt->mdt_opts.mo_acl)
4890 data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
4892 if (!mdt->mdt_opts.mo_user_xattr)
4893 data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
4895 if (!mdt->mdt_som_conf)
4896 data->ocd_connect_flags &= ~OBD_CONNECT_SOM;
4898 cfs_spin_lock(&exp->exp_lock);
4899 exp->exp_connect_flags = data->ocd_connect_flags;
4900 cfs_spin_unlock(&exp->exp_lock);
4901 data->ocd_version = LUSTRE_VERSION_CODE;
4902 exp->exp_mdt_data.med_ibits_known = data->ocd_ibits_known;
4906 if (mdt->mdt_opts.mo_acl &&
4907 ((exp->exp_connect_flags & OBD_CONNECT_ACL) == 0)) {
4908 CWARN("%s: MDS requires ACL support but client does not\n",
4909 mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4914 if ((exp->exp_connect_flags & OBD_CONNECT_FID) == 0) {
4915 CWARN("%s: MDS requires FID support, but client not\n",
4916 mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4920 if (mdt->mdt_som_conf && !exp_connect_som(exp) &&
4921 !(exp->exp_connect_flags & OBD_CONNECT_MDS_MDS)) {
4922 CWARN("%s: MDS has SOM enabled, but client does not support "
4923 "it\n", mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4930 static int mdt_connect_check_sptlrpc(struct mdt_device *mdt,
4931 struct obd_export *exp,
4932 struct ptlrpc_request *req)
4934 struct sptlrpc_flavor flvr;
4937 if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
4938 cfs_read_lock(&mdt->mdt_sptlrpc_lock);
4939 sptlrpc_target_choose_flavor(&mdt->mdt_sptlrpc_rset,
4943 cfs_read_unlock(&mdt->mdt_sptlrpc_lock);
4945 cfs_spin_lock(&exp->exp_lock);
4947 exp->exp_sp_peer = req->rq_sp_from;
4948 exp->exp_flvr = flvr;
4950 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
4951 exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
4952 CERROR("unauthorized rpc flavor %x from %s, "
4953 "expect %x\n", req->rq_flvr.sf_rpc,
4954 libcfs_nid2str(req->rq_peer.nid),
4955 exp->exp_flvr.sf_rpc);
4959 cfs_spin_unlock(&exp->exp_lock);
4961 if (exp->exp_sp_peer != req->rq_sp_from) {
4962 CERROR("RPC source %s doesn't match %s\n",
4963 sptlrpc_part2name(req->rq_sp_from),
4964 sptlrpc_part2name(exp->exp_sp_peer));
4967 rc = sptlrpc_target_export_check(exp, req);
4974 /* mds_connect copy */
4975 static int mdt_obd_connect(const struct lu_env *env,
4976 struct obd_export **exp, struct obd_device *obd,
4977 struct obd_uuid *cluuid,
4978 struct obd_connect_data *data,
4981 struct mdt_thread_info *info;
4982 struct obd_export *lexp;
4983 struct lustre_handle conn = { 0 };
4984 struct mdt_device *mdt;
4985 struct ptlrpc_request *req;
4989 LASSERT(env != NULL);
4990 if (!exp || !obd || !cluuid)
4993 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4994 req = info->mti_pill->rc_req;
4995 mdt = mdt_dev(obd->obd_lu_dev);
4997 rc = class_connect(&conn, obd, cluuid);
5001 lexp = class_conn2export(&conn);
5002 LASSERT(lexp != NULL);
5004 rc = mdt_connect_check_sptlrpc(mdt, lexp, req);
5008 rc = mdt_connect_internal(lexp, mdt, data);
5010 struct mdt_thread_info *mti;
5011 struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
5013 mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5014 LASSERT(mti != NULL);
5015 mti->mti_exp = lexp;
5016 memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
5017 rc = mdt_client_new(env, mdt);
5019 mdt_export_stats_init(obd, lexp, localdata);
5024 class_disconnect(lexp);
5033 static int mdt_obd_reconnect(const struct lu_env *env,
5034 struct obd_export *exp, struct obd_device *obd,
5035 struct obd_uuid *cluuid,
5036 struct obd_connect_data *data,
5039 struct mdt_thread_info *info;
5040 struct mdt_device *mdt;
5041 struct ptlrpc_request *req;
5045 if (exp == NULL || obd == NULL || cluuid == NULL)
5048 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5049 req = info->mti_pill->rc_req;
5050 mdt = mdt_dev(obd->obd_lu_dev);
5052 rc = mdt_connect_check_sptlrpc(mdt, exp, req);
5056 rc = mdt_connect_internal(exp, mdt_dev(obd->obd_lu_dev), data);
5058 mdt_export_stats_init(obd, exp, localdata);
5062 static int mdt_export_cleanup(struct obd_export *exp)
5064 struct mdt_export_data *med = &exp->exp_mdt_data;
5065 struct obd_device *obd = exp->exp_obd;
5066 struct mdt_device *mdt;
5067 struct mdt_thread_info *info;
5069 CFS_LIST_HEAD(closing_list);
5070 struct mdt_file_data *mfd, *n;
5074 cfs_spin_lock(&med->med_open_lock);
5075 while (!cfs_list_empty(&med->med_open_head)) {
5076 cfs_list_t *tmp = med->med_open_head.next;
5077 mfd = cfs_list_entry(tmp, struct mdt_file_data, mfd_list);
5079 /* Remove mfd handle so it can't be found again.
5080 * We are consuming the mfd_list reference here. */
5081 class_handle_unhash(&mfd->mfd_handle);
5082 cfs_list_move_tail(&mfd->mfd_list, &closing_list);
5084 cfs_spin_unlock(&med->med_open_lock);
5085 mdt = mdt_dev(obd->obd_lu_dev);
5086 LASSERT(mdt != NULL);
5088 rc = lu_env_init(&env, LCT_MD_THREAD);
5092 info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5093 LASSERT(info != NULL);
5094 memset(info, 0, sizeof *info);
5095 info->mti_env = &env;
5096 info->mti_mdt = mdt;
5097 info->mti_exp = exp;
5099 if (!cfs_list_empty(&closing_list)) {
5100 struct md_attr *ma = &info->mti_attr;
5104 lmm_size = mdt->mdt_max_mdsize;
5105 OBD_ALLOC(ma->ma_lmm, lmm_size);
5106 if (ma->ma_lmm == NULL)
5107 GOTO(out_lmm, rc = -ENOMEM);
5109 cookie_size = mdt->mdt_max_cookiesize;
5110 OBD_ALLOC(ma->ma_cookie, cookie_size);
5111 if (ma->ma_cookie == NULL)
5112 GOTO(out_cookie, rc = -ENOMEM);
5114 /* Close any open files (which may also cause orphan unlinking). */
5115 cfs_list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
5116 cfs_list_del_init(&mfd->mfd_list);
5117 memset(&ma->ma_attr, 0, sizeof(ma->ma_attr));
5118 ma->ma_lmm_size = lmm_size;
5119 ma->ma_cookie_size = cookie_size;
5121 /* It is not for setattr, just tell MDD to send
5122 * DESTROY RPC to OSS if needed */
5123 ma->ma_attr_flags = MDS_CLOSE_CLEANUP;
5124 ma->ma_valid = MA_FLAGS;
5125 mdt_mfd_close(info, mfd);
5127 OBD_FREE(ma->ma_cookie, cookie_size);
5128 ma->ma_cookie = NULL;
5130 OBD_FREE(ma->ma_lmm, lmm_size);
5134 info->mti_mdt = NULL;
5135 /* cleanup client slot early */
5136 /* Do not erase record for recoverable client. */
5137 if (!obd->obd_fail || exp->exp_failed)
5138 mdt_client_del(&env, mdt);
5144 static int mdt_obd_disconnect(struct obd_export *exp)
5150 class_export_get(exp);
5152 rc = server_disconnect_export(exp);
5154 CDEBUG(D_IOCTL, "server disconnect error: %d\n", rc);
5156 rc = mdt_export_cleanup(exp);
5157 class_export_put(exp);
5161 /* FIXME: Can we avoid using these two interfaces? */
5162 static int mdt_init_export(struct obd_export *exp)
5164 struct mdt_export_data *med = &exp->exp_mdt_data;
5168 CFS_INIT_LIST_HEAD(&med->med_open_head);
5169 cfs_spin_lock_init(&med->med_open_lock);
5170 cfs_sema_init(&med->med_idmap_sem, 1);
5171 med->med_idmap = NULL;
5172 cfs_spin_lock(&exp->exp_lock);
5173 exp->exp_connecting = 1;
5174 cfs_spin_unlock(&exp->exp_lock);
5175 rc = lut_client_alloc(exp);
5177 rc = ldlm_init_export(exp);
5180 CERROR("Error %d while initializing export\n", rc);
5184 static int mdt_destroy_export(struct obd_export *exp)
5186 struct mdt_export_data *med;
5190 med = &exp->exp_mdt_data;
5191 if (exp_connect_rmtclient(exp))
5192 mdt_cleanup_idmap(&exp->exp_mdt_data);
5194 target_destroy_export(exp);
5195 ldlm_destroy_export(exp);
5196 lut_client_free(exp);
5198 LASSERT(cfs_list_empty(&exp->exp_outstanding_replies));
5199 LASSERT(cfs_list_empty(&exp->exp_mdt_data.med_open_head));
5200 if (obd_uuid_equals(&exp->exp_client_uuid, &exp->exp_obd->obd_uuid))
5206 static void mdt_allow_cli(struct mdt_device *m, unsigned int flag)
5208 if (flag & CONFIG_LOG)
5209 cfs_set_bit(MDT_FL_CFGLOG, &m->mdt_state);
5211 /* also notify active event */
5212 if (flag & CONFIG_SYNC)
5213 cfs_set_bit(MDT_FL_SYNCED, &m->mdt_state);
5215 if (cfs_test_bit(MDT_FL_CFGLOG, &m->mdt_state) &&
5216 cfs_test_bit(MDT_FL_SYNCED, &m->mdt_state)) {
5217 struct obd_device *obd = m->mdt_md_dev.md_lu_dev.ld_obd;
5219 /* Open for clients */
5220 if (obd->obd_no_conn) {
5221 cfs_spin_lock_bh(&obd->obd_processing_task_lock);
5222 obd->obd_no_conn = 0;
5223 cfs_spin_unlock_bh(&obd->obd_processing_task_lock);
5228 static int mdt_upcall(const struct lu_env *env, struct md_device *md,
5229 enum md_upcall_event ev, void *data)
5231 struct mdt_device *m = mdt_dev(&md->md_lu_dev);
5232 struct md_device *next = m->mdt_child;
5233 struct mdt_thread_info *mti;
5239 rc = next->md_ops->mdo_maxsize_get(env, next,
5241 &m->mdt_max_cookiesize);
5242 CDEBUG(D_INFO, "get max mdsize %d max cookiesize %d\n",
5243 m->mdt_max_mdsize, m->mdt_max_cookiesize);
5244 mdt_allow_cli(m, CONFIG_SYNC);
5246 (*(__u64 *)data) = m->mdt_lut.lut_mount_count;
5249 mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5250 mti->mti_no_need_trans = 1;
5251 CDEBUG(D_INFO, "disable mdt trans for this thread\n");
5254 /* Check that MDT is not yet configured */
5255 LASSERT(!cfs_test_bit(MDT_FL_CFGLOG, &m->mdt_state));
5257 #ifdef HAVE_QUOTA_SUPPORT
5259 if (md->md_lu_dev.ld_obd->obd_recovering == 0 &&
5260 likely(md->md_lu_dev.ld_obd->obd_stopping == 0))
5261 next->md_ops->mdo_quota.mqo_recovery(env, next);
5265 CERROR("invalid event\n");
5272 static int mdt_obd_notify(struct obd_device *host,
5273 struct obd_device *watched,
5274 enum obd_notify_event ev, void *data)
5276 struct mdt_device *mdt = mdt_dev(host->obd_lu_dev);
5277 #ifdef HAVE_QUOTA_SUPPORT
5278 struct md_device *next = mdt->mdt_child;
5283 case OBD_NOTIFY_CONFIG:
5284 mdt_allow_cli(mdt, (unsigned long)data);
5286 #ifdef HAVE_QUOTA_SUPPORT
5287 /* quota_type has been processed, we can now handle
5288 * incoming quota requests */
5289 next->md_ops->mdo_quota.mqo_notify(NULL, next);
5293 CDEBUG(D_INFO, "Unhandled notification %#x\n", ev);
5298 static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key,
5299 void *val, int vallen)
5301 struct mdt_device *mdt = mdt_dev(info->mti_exp->exp_obd->obd_lu_dev);
5302 struct getinfo_fid2path *fpout, *fpin;
5305 fpin = key + cfs_size_round(sizeof(KEY_FID2PATH));
5308 if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
5309 lustre_swab_fid2path(fpin);
5311 memcpy(fpout, fpin, sizeof(*fpin));
5312 if (fpout->gf_pathlen != vallen - sizeof(*fpin))
5315 rc = mdt_fid2path(info->mti_env, mdt, fpout);
5319 static int mdt_fid2path(const struct lu_env *env, struct mdt_device *mdt,
5320 struct getinfo_fid2path *fp)
5322 struct mdt_object *obj;
5326 CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n",
5327 PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
5329 if (!fid_is_sane(&fp->gf_fid))
5332 obj = mdt_object_find(env, mdt, &fp->gf_fid);
5333 if (obj == NULL || IS_ERR(obj)) {
5334 CDEBUG(D_IOCTL, "no object "DFID": %ld\n",PFID(&fp->gf_fid),
5339 rc = lu_object_exists(&obj->mot_obj.mo_lu);
5345 mdt_object_put(env, obj);
5346 CDEBUG(D_IOCTL, "nonlocal object "DFID": %d\n",
5347 PFID(&fp->gf_fid), rc);
5351 rc = mo_path(env, md_object_next(&obj->mot_obj), fp->gf_path,
5352 fp->gf_pathlen, &fp->gf_recno, &fp->gf_linkno);
5353 mdt_object_put(env, obj);
5358 static int mdt_get_info(struct mdt_thread_info *info)
5360 struct ptlrpc_request *req = mdt_info_req(info);
5368 key = req_capsule_client_get(info->mti_pill, &RMF_GETINFO_KEY);
5370 CDEBUG(D_IOCTL, "No GETINFO key");
5373 keylen = req_capsule_get_size(info->mti_pill, &RMF_GETINFO_KEY,
5376 vallen = req_capsule_client_get(info->mti_pill, &RMF_GETINFO_VALLEN);
5377 if (vallen == NULL) {
5378 CDEBUG(D_IOCTL, "Unable to get RMF_GETINFO_VALLEN buffer");
5382 req_capsule_set_size(info->mti_pill, &RMF_GETINFO_VAL, RCL_SERVER,
5384 rc = req_capsule_server_pack(info->mti_pill);
5385 valout = req_capsule_server_get(info->mti_pill, &RMF_GETINFO_VAL);
5386 if (valout == NULL) {
5387 CDEBUG(D_IOCTL, "Unable to get get-info RPC out buffer");
5391 if (KEY_IS(KEY_FID2PATH))
5392 rc = mdt_rpc_fid2path(info, key, valout, *vallen);
5396 lustre_msg_set_status(req->rq_repmsg, rc);
5401 /* Pass the ioc down */
5402 static int mdt_ioc_child(struct lu_env *env, struct mdt_device *mdt,
5403 unsigned int cmd, int len, void *data)
5405 struct lu_context ioctl_session;
5406 struct md_device *next = mdt->mdt_child;
5410 rc = lu_context_init(&ioctl_session, LCT_SESSION);
5413 ioctl_session.lc_thread = (struct ptlrpc_thread *)cfs_current();
5414 lu_context_enter(&ioctl_session);
5415 env->le_ses = &ioctl_session;
5417 LASSERT(next->md_ops->mdo_iocontrol);
5418 rc = next->md_ops->mdo_iocontrol(env, next, cmd, len, data);
5420 lu_context_exit(&ioctl_session);
5421 lu_context_fini(&ioctl_session);
5425 static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
5427 struct obd_ioctl_data *data = karg;
5428 struct lu_fid *fid = (struct lu_fid *)data->ioc_inlbuf1;
5430 struct mdt_object *obj;
5431 struct mdt_lock_handle *lh;
5434 CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
5435 if (!fid_is_sane(fid))
5438 lh = &mti->mti_lh[MDT_LH_PARENT];
5439 mdt_lock_reg_init(lh, LCK_CR);
5441 obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE);
5443 RETURN(PTR_ERR(obj));
5445 rc = mdt_object_exists(obj);
5449 * before calling version get the correct MDS should be
5450 * fid, this is error to find remote object here
5452 CERROR("nonlocal object "DFID"\n", PFID(fid));
5454 version = mo_version_get(mti->mti_env, mdt_object_child(obj));
5455 *(__u64 *)data->ioc_inlbuf2 = version;
5458 mdt_object_unlock_put(mti, obj, lh, 1);
5462 /* ioctls on obd dev */
5463 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
5464 void *karg, void *uarg)
5467 struct obd_device *obd = exp->exp_obd;
5468 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
5469 struct dt_device *dt = mdt->mdt_bottom;
5473 CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
5474 rc = lu_env_init(&env, LCT_MD_THREAD);
5480 rc = mdt_device_sync(&env, mdt);
5482 case OBD_IOC_SET_READONLY:
5483 dt->dd_ops->dt_ro(&env, dt);
5485 case OBD_IOC_ABORT_RECOVERY:
5486 CERROR("Aborting recovery for device %s\n", obd->obd_name);
5487 target_stop_recovery_thread(obd);
5490 case OBD_IOC_CHANGELOG_REG:
5491 case OBD_IOC_CHANGELOG_DEREG:
5492 case OBD_IOC_CHANGELOG_CLEAR:
5493 rc = mdt_ioc_child(&env, mdt, cmd, len, karg);
5495 case OBD_IOC_GET_OBJ_VERSION: {
5496 struct mdt_thread_info *mti;
5497 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5498 memset(mti, 0, sizeof *mti);
5499 mti->mti_env = &env;
5503 rc = mdt_ioc_version_get(mti, karg);
5507 CERROR("Not supported cmd = %d for device %s\n",
5508 cmd, obd->obd_name);
5516 int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
5518 struct lu_device *ld = md2lu_dev(mdt->mdt_child);
5519 struct obd_device *obd = mdt2obd_dev(mdt);
5520 #ifdef HAVE_QUOTA_SUPPORT
5521 struct md_device *next = mdt->mdt_child;
5526 rc = ld->ld_ops->ldo_recovery_complete(env, ld);
5527 #ifdef HAVE_QUOTA_SUPPORT
5528 if (likely(obd->obd_stopping == 0))
5529 next->md_ops->mdo_quota.mqo_recovery(env, next);
5534 int mdt_obd_postrecov(struct obd_device *obd)
5539 rc = lu_env_init(&env, LCT_MD_THREAD);
5542 rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
5548 * Send a copytool req to a client
5549 * Note this sends a request RPC from a server (MDT) to a client (MDC),
5550 * backwards of normal comms.
5552 int mdt_hsm_copytool_send(struct obd_export *exp)
5555 struct hsm_action_list *hal;
5556 struct hsm_action_item *hai;
5560 CWARN("%s: writing to mdc at %s\n", exp->exp_obd->obd_name,
5561 libcfs_nid2str(exp->exp_connection->c_peer.nid));
5563 len = sizeof(*lh) + sizeof(*hal) + MTI_NAME_MAXLEN +
5564 /* for mockup below */ 2 * cfs_size_round(sizeof(*hai));
5569 lh->kuc_magic = KUC_MAGIC;
5570 lh->kuc_transport = KUC_TRANSPORT_HSM;
5571 lh->kuc_msgtype = HMT_ACTION_LIST;
5572 lh->kuc_msglen = len;
5574 hal = (struct hsm_action_list *)(lh + 1);
5575 hal->hal_version = HAL_VERSION;
5576 hal->hal_archive_num = 1;
5577 obd_uuid2fsname(hal->hal_fsname, exp->exp_obd->obd_name,
5580 /* mock up an action list */
5582 hai = hai_zero(hal);
5583 hai->hai_action = HSMA_ARCHIVE;
5584 hai->hai_fid.f_oid = 0xA00A;
5585 hai->hai_len = sizeof(*hai);
5586 hai = hai_next(hai);
5587 hai->hai_action = HSMA_RESTORE;
5588 hai->hai_fid.f_oid = 0xB00B;
5589 hai->hai_len = sizeof(*hai);
5591 /* Uses the ldlm reverse import; this rpc will be seen by
5592 the ldlm_callback_handler */
5593 rc = do_set_info_async(exp->exp_imp_reverse,
5594 LDLM_SET_INFO, LUSTRE_OBD_VERSION,
5595 sizeof(KEY_HSM_COPYTOOL_SEND),
5596 KEY_HSM_COPYTOOL_SEND,
5604 static struct obd_ops mdt_obd_device_ops = {
5605 .o_owner = THIS_MODULE,
5606 .o_set_info_async = mdt_obd_set_info_async,
5607 .o_connect = mdt_obd_connect,
5608 .o_reconnect = mdt_obd_reconnect,
5609 .o_disconnect = mdt_obd_disconnect,
5610 .o_init_export = mdt_init_export,
5611 .o_destroy_export = mdt_destroy_export,
5612 .o_iocontrol = mdt_iocontrol,
5613 .o_postrecov = mdt_obd_postrecov,
5614 .o_notify = mdt_obd_notify
5617 static struct lu_device* mdt_device_fini(const struct lu_env *env,
5618 struct lu_device *d)
5620 struct mdt_device *m = mdt_dev(d);
5627 static struct lu_device *mdt_device_free(const struct lu_env *env,
5628 struct lu_device *d)
5630 struct mdt_device *m = mdt_dev(d);
5633 md_device_fini(&m->mdt_md_dev);
5638 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
5639 struct lu_device_type *t,
5640 struct lustre_cfg *cfg)
5642 struct lu_device *l;
5643 struct mdt_device *m;
5649 l = &m->mdt_md_dev.md_lu_dev;
5650 rc = mdt_init0(env, m, t, cfg);
5652 mdt_device_free(env, l);
5656 md_upcall_init(&m->mdt_md_dev, mdt_upcall);
5658 l = ERR_PTR(-ENOMEM);
5662 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
5663 LU_KEY_INIT_FINI(mdt, struct mdt_thread_info);
5665 /* context key: mdt_thread_key */
5666 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
5668 /* context key constructor/destructor: mdt_txn_key_init, mdt_txn_key_fini */
5669 LU_KEY_INIT_FINI(mdt_txn, struct mdt_txn_info);
5671 struct lu_context_key mdt_txn_key = {
5672 .lct_tags = LCT_TX_HANDLE,
5673 .lct_init = mdt_txn_key_init,
5674 .lct_fini = mdt_txn_key_fini
5677 struct md_ucred *mdt_ucred(const struct mdt_thread_info *info)
5679 return md_ucred(info->mti_env);
5683 * Enable/disable COS (Commit On Sharing).
5685 * Set/Clear the COS flag in mdt options.
5687 * \param mdt mdt device
5688 * \param val 0 disables COS, other values enable COS
5690 void mdt_enable_cos(struct mdt_device *mdt, int val)
5695 mdt->mdt_opts.mo_cos = !!val;
5696 rc = lu_env_init(&env, LCT_MD_THREAD);
5697 if (unlikely(rc != 0)) {
5698 CWARN("lu_env initialization failed with rc = %d,"
5699 "cannot sync\n", rc);
5702 mdt_device_sync(&env, mdt);
5707 * Check COS (Commit On Sharing) status.
5709 * Return COS flag status.
5711 * \param mdt mdt device
5713 int mdt_cos_is_enabled(struct mdt_device *mdt)
5715 return mdt->mdt_opts.mo_cos != 0;
5718 /* type constructor/destructor: mdt_type_init, mdt_type_fini */
5719 LU_TYPE_INIT_FINI(mdt, &mdt_thread_key, &mdt_txn_key);
5721 static struct lu_device_type_operations mdt_device_type_ops = {
5722 .ldto_init = mdt_type_init,
5723 .ldto_fini = mdt_type_fini,
5725 .ldto_start = mdt_type_start,
5726 .ldto_stop = mdt_type_stop,
5728 .ldto_device_alloc = mdt_device_alloc,
5729 .ldto_device_free = mdt_device_free,
5730 .ldto_device_fini = mdt_device_fini
5733 static struct lu_device_type mdt_device_type = {
5734 .ldt_tags = LU_DEVICE_MD,
5735 .ldt_name = LUSTRE_MDT_NAME,
5736 .ldt_ops = &mdt_device_type_ops,
5737 .ldt_ctx_tags = LCT_MD_THREAD
5740 static struct lu_local_obj_desc mdt_last_recv = {
5741 .llod_name = LAST_RCVD,
5742 .llod_oid = MDT_LAST_RECV_OID,
5746 static int __init mdt_mod_init(void)
5748 struct lprocfs_static_vars lvars;
5751 llo_local_obj_register(&mdt_last_recv);
5753 if (mdt_num_threads > 0) {
5754 if (mdt_num_threads > MDT_MAX_THREADS)
5755 mdt_num_threads = MDT_MAX_THREADS;
5756 if (mdt_num_threads < MDT_MIN_THREADS)
5757 mdt_num_threads = MDT_MIN_THREADS;
5758 mdt_max_threads = mdt_min_threads = mdt_num_threads;
5760 mdt_max_threads = MDT_MAX_THREADS;
5761 mdt_min_threads = MDT_MIN_THREADS;
5762 if (mdt_min_threads < MDT_NUM_THREADS)
5763 mdt_min_threads = MDT_NUM_THREADS;
5766 lprocfs_mdt_init_vars(&lvars);
5767 rc = class_register_type(&mdt_obd_device_ops, NULL,
5768 lvars.module_vars, LUSTRE_MDT_NAME,
5774 static void __exit mdt_mod_exit(void)
5776 llo_local_obj_unregister(&mdt_last_recv);
5777 class_unregister_type(LUSTRE_MDT_NAME);
5781 #define DEF_HNDL(prefix, base, suffix, flags, opc, fn, fmt) \
5782 [prefix ## _ ## opc - prefix ## _ ## base] = { \
5784 .mh_fail_id = OBD_FAIL_ ## prefix ## _ ## opc ## suffix, \
5785 .mh_opc = prefix ## _ ## opc, \
5786 .mh_flags = flags, \
5791 #define DEF_MDT_HNDL(flags, name, fn, fmt) \
5792 DEF_HNDL(MDS, GETATTR, _NET, flags, name, fn, fmt)
5794 #define DEF_SEQ_HNDL(flags, name, fn, fmt) \
5795 DEF_HNDL(SEQ, QUERY, _NET, flags, name, fn, fmt)
5797 #define DEF_FLD_HNDL(flags, name, fn, fmt) \
5798 DEF_HNDL(FLD, QUERY, _NET, flags, name, fn, fmt)
5800 * Request with a format known in advance
5802 #define DEF_MDT_HNDL_F(flags, name, fn) \
5803 DEF_HNDL(MDS, GETATTR, _NET, flags, name, fn, &RQF_MDS_ ## name)
5805 #define DEF_SEQ_HNDL_F(flags, name, fn) \
5806 DEF_HNDL(SEQ, QUERY, _NET, flags, name, fn, &RQF_SEQ_ ## name)
5808 #define DEF_FLD_HNDL_F(flags, name, fn) \
5809 DEF_HNDL(FLD, QUERY, _NET, flags, name, fn, &RQF_FLD_ ## name)
5811 * Request with a format we do not yet know
5813 #define DEF_MDT_HNDL_0(flags, name, fn) \
5814 DEF_HNDL(MDS, GETATTR, _NET, flags, name, fn, NULL)
5816 static struct mdt_handler mdt_mds_ops[] = {
5817 DEF_MDT_HNDL_F(0, CONNECT, mdt_connect),
5818 DEF_MDT_HNDL_F(0, DISCONNECT, mdt_disconnect),
5819 DEF_MDT_HNDL (0, SET_INFO, mdt_set_info,
5821 DEF_MDT_HNDL_F(0, GET_INFO, mdt_get_info),
5822 DEF_MDT_HNDL_F(0 |HABEO_REFERO, GETSTATUS, mdt_getstatus),
5823 DEF_MDT_HNDL_F(HABEO_CORPUS, GETATTR, mdt_getattr),
5824 DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, GETATTR_NAME, mdt_getattr_name),
5825 DEF_MDT_HNDL_F(HABEO_CORPUS, GETXATTR, mdt_getxattr),
5826 DEF_MDT_HNDL_F(0 |HABEO_REFERO, STATFS, mdt_statfs),
5827 DEF_MDT_HNDL_F(0 |MUTABOR, REINT, mdt_reint),
5828 DEF_MDT_HNDL_F(HABEO_CORPUS, CLOSE, mdt_close),
5829 DEF_MDT_HNDL_F(HABEO_CORPUS, DONE_WRITING, mdt_done_writing),
5830 DEF_MDT_HNDL_F(0 |HABEO_REFERO, PIN, mdt_pin),
5831 DEF_MDT_HNDL_0(0, SYNC, mdt_sync),
5832 DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, IS_SUBDIR, mdt_is_subdir),
5833 #ifdef HAVE_QUOTA_SUPPORT
5834 DEF_MDT_HNDL_F(0, QUOTACHECK, mdt_quotacheck_handle),
5835 DEF_MDT_HNDL_F(0, QUOTACTL, mdt_quotactl_handle)
5839 #define DEF_OBD_HNDL(flags, name, fn) \
5840 DEF_HNDL(OBD, PING, _NET, flags, name, fn, NULL)
5843 static struct mdt_handler mdt_obd_ops[] = {
5844 DEF_OBD_HNDL(0, PING, mdt_obd_ping),
5845 DEF_OBD_HNDL(0, LOG_CANCEL, mdt_obd_log_cancel),
5846 DEF_OBD_HNDL(0, QC_CALLBACK, mdt_obd_qc_callback)
5849 #define DEF_DLM_HNDL_0(flags, name, fn) \
5850 DEF_HNDL(LDLM, ENQUEUE, , flags, name, fn, NULL)
5851 #define DEF_DLM_HNDL_F(flags, name, fn) \
5852 DEF_HNDL(LDLM, ENQUEUE, , flags, name, fn, &RQF_LDLM_ ## name)
5854 static struct mdt_handler mdt_dlm_ops[] = {
5855 DEF_DLM_HNDL_F(HABEO_CLAVIS, ENQUEUE, mdt_enqueue),
5856 DEF_DLM_HNDL_0(HABEO_CLAVIS, CONVERT, mdt_convert),
5857 DEF_DLM_HNDL_0(0, BL_CALLBACK, mdt_bl_callback),
5858 DEF_DLM_HNDL_0(0, CP_CALLBACK, mdt_cp_callback)
5861 #define DEF_LLOG_HNDL(flags, name, fn) \
5862 DEF_HNDL(LLOG, ORIGIN_HANDLE_CREATE, _NET, flags, name, fn, NULL)
5864 static struct mdt_handler mdt_llog_ops[] = {
5865 DEF_LLOG_HNDL(0, ORIGIN_HANDLE_CREATE, mdt_llog_create),
5866 DEF_LLOG_HNDL(0, ORIGIN_HANDLE_NEXT_BLOCK, mdt_llog_next_block),
5867 DEF_LLOG_HNDL(0, ORIGIN_HANDLE_READ_HEADER, mdt_llog_read_header),
5868 DEF_LLOG_HNDL(0, ORIGIN_HANDLE_WRITE_REC, NULL),
5869 DEF_LLOG_HNDL(0, ORIGIN_HANDLE_CLOSE, NULL),
5870 DEF_LLOG_HNDL(0, ORIGIN_CONNECT, NULL),
5871 DEF_LLOG_HNDL(0, CATINFO, NULL),
5872 DEF_LLOG_HNDL(0, ORIGIN_HANDLE_PREV_BLOCK, mdt_llog_prev_block),
5873 DEF_LLOG_HNDL(0, ORIGIN_HANDLE_DESTROY, mdt_llog_destroy),
5876 #define DEF_SEC_CTX_HNDL(name, fn) \
5877 DEF_HNDL(SEC_CTX, INIT, _NET, 0, name, fn, NULL)
5879 static struct mdt_handler mdt_sec_ctx_ops[] = {
5880 DEF_SEC_CTX_HNDL(INIT, mdt_sec_ctx_handle),
5881 DEF_SEC_CTX_HNDL(INIT_CONT, mdt_sec_ctx_handle),
5882 DEF_SEC_CTX_HNDL(FINI, mdt_sec_ctx_handle)
5885 static struct mdt_opc_slice mdt_regular_handlers[] = {
5887 .mos_opc_start = MDS_GETATTR,
5888 .mos_opc_end = MDS_LAST_OPC,
5889 .mos_hs = mdt_mds_ops
5892 .mos_opc_start = OBD_PING,
5893 .mos_opc_end = OBD_LAST_OPC,
5894 .mos_hs = mdt_obd_ops
5897 .mos_opc_start = LDLM_ENQUEUE,
5898 .mos_opc_end = LDLM_LAST_OPC,
5899 .mos_hs = mdt_dlm_ops
5902 .mos_opc_start = LLOG_ORIGIN_HANDLE_CREATE,
5903 .mos_opc_end = LLOG_LAST_OPC,
5904 .mos_hs = mdt_llog_ops
5907 .mos_opc_start = SEC_CTX_INIT,
5908 .mos_opc_end = SEC_LAST_OPC,
5909 .mos_hs = mdt_sec_ctx_ops
5916 static struct mdt_handler mdt_readpage_ops[] = {
5917 DEF_MDT_HNDL_F(0, CONNECT, mdt_connect),
5918 DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, READPAGE, mdt_readpage),
5919 #ifdef HAVE_SPLIT_SUPPORT
5920 DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, WRITEPAGE, mdt_writepage),
5924 * XXX: this is ugly and should be fixed one day, see mdc_close() for
5925 * detailed comments. --umka
5927 DEF_MDT_HNDL_F(HABEO_CORPUS, CLOSE, mdt_close),
5928 DEF_MDT_HNDL_F(HABEO_CORPUS, DONE_WRITING, mdt_done_writing),
5931 static struct mdt_opc_slice mdt_readpage_handlers[] = {
5933 .mos_opc_start = MDS_GETATTR,
5934 .mos_opc_end = MDS_LAST_OPC,
5935 .mos_hs = mdt_readpage_ops
5942 static struct mdt_handler mdt_xmds_ops[] = {
5943 DEF_MDT_HNDL_F(0, CONNECT, mdt_connect),
5944 DEF_MDT_HNDL_F(HABEO_CORPUS , GETATTR, mdt_getattr),
5945 DEF_MDT_HNDL_F(0 | MUTABOR , REINT, mdt_reint),
5946 DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, IS_SUBDIR, mdt_is_subdir),
5949 static struct mdt_opc_slice mdt_xmds_handlers[] = {
5951 .mos_opc_start = MDS_GETATTR,
5952 .mos_opc_end = MDS_LAST_OPC,
5953 .mos_hs = mdt_xmds_ops
5956 .mos_opc_start = OBD_PING,
5957 .mos_opc_end = OBD_LAST_OPC,
5958 .mos_hs = mdt_obd_ops
5961 .mos_opc_start = SEC_CTX_INIT,
5962 .mos_opc_end = SEC_LAST_OPC,
5963 .mos_hs = mdt_sec_ctx_ops
5970 static struct mdt_handler mdt_seq_ops[] = {
5971 DEF_SEQ_HNDL_F(0, QUERY, (int (*)(struct mdt_thread_info *))seq_query)
5974 static struct mdt_opc_slice mdt_seq_handlers[] = {
5976 .mos_opc_start = SEQ_QUERY,
5977 .mos_opc_end = SEQ_LAST_OPC,
5978 .mos_hs = mdt_seq_ops
5985 static struct mdt_handler mdt_fld_ops[] = {
5986 DEF_FLD_HNDL_F(0, QUERY, (int (*)(struct mdt_thread_info *))fld_query)
5989 static struct mdt_opc_slice mdt_fld_handlers[] = {
5991 .mos_opc_start = FLD_QUERY,
5992 .mos_opc_end = FLD_LAST_OPC,
5993 .mos_hs = mdt_fld_ops
6000 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
6001 MODULE_DESCRIPTION("Lustre Meta-data Target ("LUSTRE_MDT_NAME")");
6002 MODULE_LICENSE("GPL");
6004 CFS_MODULE_PARM(mdt_num_threads, "ul", ulong, 0444,
6005 "number of mdt service threads to start");
6007 cfs_module(mdt, "0.2.0", mdt_mod_init, mdt_mod_exit);