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 (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
30 * Use is subject to license terms.
33 * Copyright (c) 2011 Whamcloud, Inc.
36 * This file is part of Lustre, http://www.lustre.org/
37 * Lustre is a trademark of Sun Microsystems, Inc.
39 * lustre/mdt/mdt_handler.c
41 * Lustre Metadata Target (mdt) request handler
43 * Author: Peter Braam <braam@clusterfs.com>
44 * Author: Andreas Dilger <adilger@clusterfs.com>
45 * Author: Phil Schwan <phil@clusterfs.com>
46 * Author: Mike Shaver <shaver@clusterfs.com>
47 * Author: Nikita Danilov <nikita@clusterfs.com>
48 * Author: Huang Hua <huanghua@clusterfs.com>
49 * Author: Yury Umanets <umka@clusterfs.com>
53 # define EXPORT_SYMTAB
55 #define DEBUG_SUBSYSTEM S_MDS
57 #include <linux/module.h>
59 * struct OBD_{ALLOC,FREE}*()
61 #include <obd_support.h>
62 /* struct ptlrpc_request */
63 #include <lustre_net.h>
64 /* struct obd_export */
65 #include <lustre_export.h>
66 /* struct obd_device */
69 #include <dt_object.h>
70 #include <lustre_mds.h>
71 #include <lustre_mdt.h>
72 #include "mdt_internal.h"
73 #ifdef HAVE_QUOTA_SUPPORT
74 # include <lustre_quota.h>
76 #include <lustre_acl.h>
77 #include <lustre_param.h>
78 #include <lustre_fsfilt.h>
80 mdl_mode_t mdt_mdl_lock_modes[] = {
81 [LCK_MINMODE] = MDL_MINMODE,
88 [LCK_GROUP] = MDL_GROUP
91 ldlm_mode_t mdt_dlm_lock_modes[] = {
92 [MDL_MINMODE] = LCK_MINMODE,
99 [MDL_GROUP] = LCK_GROUP
103 * Initialized in mdt_mod_init().
105 static unsigned long mdt_num_threads;
106 static unsigned long mdt_min_threads;
107 static unsigned long mdt_max_threads;
109 /* ptlrpc request handler for MDT. All handlers are
110 * grouped into several slices - struct mdt_opc_slice,
111 * and stored in an array - mdt_handlers[].
114 /* The name of this handler. */
116 /* Fail id for this handler, checked at the beginning of this handler*/
118 /* Operation code for this handler */
120 /* flags are listed in enum mdt_handler_flags below. */
122 /* The actual handler function to execute. */
123 int (*mh_act)(struct mdt_thread_info *info);
124 /* Request format for this request. */
125 const struct req_format *mh_fmt;
128 enum mdt_handler_flags {
130 * struct mdt_body is passed in the incoming message, and object
131 * identified by this fid exists on disk.
133 * "habeo corpus" == "I have a body"
135 HABEO_CORPUS = (1 << 0),
137 * struct ldlm_request is passed in the incoming message.
139 * "habeo clavis" == "I have a key"
141 HABEO_CLAVIS = (1 << 1),
143 * this request has fixed reply format, so that reply message can be
144 * packed by generic code.
146 * "habeo refero" == "I have a reply"
148 HABEO_REFERO = (1 << 2),
150 * this request will modify something, so check whether the filesystem
151 * is readonly or not, then return -EROFS to client asap if necessary.
153 * "mutabor" == "I shall modify"
158 struct mdt_opc_slice {
161 struct mdt_handler *mos_hs;
164 static struct mdt_opc_slice mdt_regular_handlers[];
165 static struct mdt_opc_slice mdt_readpage_handlers[];
166 static struct mdt_opc_slice mdt_xmds_handlers[];
167 static struct mdt_opc_slice mdt_seq_handlers[];
168 static struct mdt_opc_slice mdt_fld_handlers[];
170 static struct mdt_device *mdt_dev(struct lu_device *d);
171 static int mdt_regular_handle(struct ptlrpc_request *req);
172 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags);
173 static int mdt_fid2path(const struct lu_env *env, struct mdt_device *mdt,
174 struct getinfo_fid2path *fp);
176 static const struct lu_object_operations mdt_obj_ops;
178 int mdt_get_disposition(struct ldlm_reply *rep, int flag)
182 return (rep->lock_policy_res1 & flag);
185 void mdt_clear_disposition(struct mdt_thread_info *info,
186 struct ldlm_reply *rep, int flag)
189 info->mti_opdata &= ~flag;
191 rep->lock_policy_res1 &= ~flag;
194 void mdt_set_disposition(struct mdt_thread_info *info,
195 struct ldlm_reply *rep, int flag)
198 info->mti_opdata |= flag;
200 rep->lock_policy_res1 |= flag;
203 void mdt_lock_reg_init(struct mdt_lock_handle *lh, ldlm_mode_t lm)
205 lh->mlh_pdo_hash = 0;
206 lh->mlh_reg_mode = lm;
207 lh->mlh_type = MDT_REG_LOCK;
210 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, ldlm_mode_t lm,
211 const char *name, int namelen)
213 lh->mlh_reg_mode = lm;
214 lh->mlh_type = MDT_PDO_LOCK;
217 LASSERT(namelen > 0);
218 lh->mlh_pdo_hash = full_name_hash(name, namelen);
220 LASSERT(namelen == 0);
221 lh->mlh_pdo_hash = 0ull;
225 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
226 struct mdt_lock_handle *lh)
232 * Any dir access needs couple of locks:
234 * 1) on part of dir we gonna take lookup/modify;
236 * 2) on whole dir to protect it from concurrent splitting and/or to
237 * flush client's cache for readdir().
239 * so, for a given mode and object this routine decides what lock mode
240 * to use for lock #2:
242 * 1) if caller's gonna lookup in dir then we need to protect dir from
243 * being splitted only - LCK_CR
245 * 2) if caller's gonna modify dir then we need to protect dir from
246 * being splitted and to flush cache - LCK_CW
248 * 3) if caller's gonna modify dir and that dir seems ready for
249 * splitting then we need to protect it from any type of access
250 * (lookup/modify/split) - LCK_EX --bzzz
253 LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
254 LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
257 * Ask underlaying level its opinion about preferable PDO lock mode
258 * having access type passed as regular lock mode:
260 * - MDL_MINMODE means that lower layer does not want to specify lock
263 * - MDL_NL means that no PDO lock should be taken. This is used in some
264 * cases. Say, for non-splittable directories no need to use PDO locks
267 mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
268 mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
270 if (mode != MDL_MINMODE) {
271 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
274 * Lower layer does not want to specify locking mode. We do it
275 * our selves. No special protection is needed, just flush
276 * client's cache on modification and allow concurrent
279 switch (lh->mlh_reg_mode) {
281 lh->mlh_pdo_mode = LCK_EX;
284 lh->mlh_pdo_mode = LCK_CR;
287 lh->mlh_pdo_mode = LCK_CW;
290 CERROR("Not expected lock type (0x%x)\n",
291 (int)lh->mlh_reg_mode);
296 LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
300 static int mdt_getstatus(struct mdt_thread_info *info)
302 struct mdt_device *mdt = info->mti_mdt;
303 struct md_device *next = mdt->mdt_child;
304 struct mdt_body *repbody;
309 rc = mdt_check_ucred(info);
311 RETURN(err_serious(rc));
313 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETSTATUS_PACK))
314 RETURN(err_serious(-ENOMEM));
316 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
317 rc = next->md_ops->mdo_root_get(info->mti_env, next, &repbody->fid1);
321 repbody->valid |= OBD_MD_FLID;
323 if (mdt->mdt_opts.mo_mds_capa &&
324 info->mti_exp->exp_connect_flags & OBD_CONNECT_MDS_CAPA) {
325 struct mdt_object *root;
326 struct lustre_capa *capa;
328 root = mdt_object_find(info->mti_env, mdt, &repbody->fid1);
330 RETURN(PTR_ERR(root));
332 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA1);
334 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
335 rc = mo_capa_get(info->mti_env, mdt_object_child(root), capa,
337 mdt_object_put(info->mti_env, root);
339 repbody->valid |= OBD_MD_FLMDSCAPA;
345 static int mdt_statfs(struct mdt_thread_info *info)
347 struct ptlrpc_request *req = mdt_info_req(info);
348 struct md_device *next = info->mti_mdt->mdt_child;
349 struct ptlrpc_service *svc;
350 struct obd_statfs *osfs;
355 svc = info->mti_pill->rc_req->rq_rqbd->rqbd_service;
357 /* This will trigger a watchdog timeout */
358 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
359 (MDT_SERVICE_WATCHDOG_FACTOR *
360 at_get(&svc->srv_at_estimate)) + 1);
362 rc = mdt_check_ucred(info);
364 RETURN(err_serious(rc));
366 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK)) {
367 rc = err_serious(-ENOMEM);
369 osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
370 rc = next->md_ops->mdo_statfs(info->mti_env, next,
372 statfs_pack(osfs, &info->mti_u.ksfs);
376 mdt_counter_incr(req->rq_export, LPROC_MDT_STATFS);
382 * Pack SOM attributes into the reply.
383 * Call under a DLM UPDATE lock.
385 static void mdt_pack_size2body(struct mdt_thread_info *info,
386 struct mdt_object *mo)
389 struct md_attr *ma = &info->mti_attr;
391 LASSERT(ma->ma_attr.la_valid & LA_MODE);
392 b = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
394 /* Check if Size-on-MDS is supported, if this is a regular file,
395 * if SOM is enabled on the object and if SOM cache exists and valid.
396 * Otherwise do not pack Size-on-MDS attributes to the reply. */
397 if (!(mdt_conn_flags(info) & OBD_CONNECT_SOM) ||
398 !S_ISREG(ma->ma_attr.la_mode) ||
399 !mdt_object_is_som_enabled(mo) ||
400 !(ma->ma_valid & MA_SOM))
403 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
404 b->size = ma->ma_som->msd_size;
405 b->blocks = ma->ma_som->msd_blocks;
408 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
409 const struct lu_attr *attr, const struct lu_fid *fid)
411 struct md_attr *ma = &info->mti_attr;
413 LASSERT(ma->ma_valid & MA_INODE);
415 b->atime = attr->la_atime;
416 b->mtime = attr->la_mtime;
417 b->ctime = attr->la_ctime;
418 b->mode = attr->la_mode;
419 b->size = attr->la_size;
420 b->blocks = attr->la_blocks;
421 b->uid = attr->la_uid;
422 b->gid = attr->la_gid;
423 b->flags = attr->la_flags;
424 b->nlink = attr->la_nlink;
425 b->rdev = attr->la_rdev;
427 /*XXX should pack the reply body according to lu_valid*/
428 b->valid |= OBD_MD_FLCTIME | OBD_MD_FLUID |
429 OBD_MD_FLGID | OBD_MD_FLTYPE |
430 OBD_MD_FLMODE | OBD_MD_FLNLINK | OBD_MD_FLFLAGS |
431 OBD_MD_FLATIME | OBD_MD_FLMTIME ;
433 if (!S_ISREG(attr->la_mode)) {
434 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
435 } else if (ma->ma_need & MA_LOV && ma->ma_lmm_size == 0) {
436 /* means no objects are allocated on osts. */
437 LASSERT(!(ma->ma_valid & MA_LOV));
438 /* just ignore blocks occupied by extend attributes on MDS */
440 /* if no object is allocated on osts, the size on mds is valid. b=22272 */
441 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
446 b->valid |= OBD_MD_FLID;
448 /* FIXME: these should be fixed when new igif ready.*/
449 b->ino = fid_oid(fid); /* 1.6 compatibility */
450 b->generation = fid_ver(fid); /* 1.6 compatibility */
451 b->valid |= OBD_MD_FLGENER; /* 1.6 compatibility */
453 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, size="LPU64"\n",
454 PFID(fid), b->nlink, b->mode, b->size);
458 mdt_body_reverse_idmap(info, b);
460 if (b->valid & OBD_MD_FLSIZE)
461 CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
462 PFID(fid), (unsigned long long)b->size);
465 static inline int mdt_body_has_lov(const struct lu_attr *la,
466 const struct mdt_body *body)
468 return ((S_ISREG(la->la_mode) && (body->valid & OBD_MD_FLEASIZE)) ||
469 (S_ISDIR(la->la_mode) && (body->valid & OBD_MD_FLDIREA )) );
472 static int mdt_getattr_internal(struct mdt_thread_info *info,
473 struct mdt_object *o, int ma_need)
475 struct md_object *next = mdt_object_child(o);
476 const struct mdt_body *reqbody = info->mti_body;
477 struct ptlrpc_request *req = mdt_info_req(info);
478 struct md_attr *ma = &info->mti_attr;
479 struct lu_attr *la = &ma->ma_attr;
480 struct req_capsule *pill = info->mti_pill;
481 const struct lu_env *env = info->mti_env;
482 struct mdt_body *repbody;
483 struct lu_buf *buffer = &info->mti_buf;
487 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
488 RETURN(err_serious(-ENOMEM));
490 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
494 rc = mdt_object_exists(o);
496 /* This object is located on remote node.*/
497 repbody->fid1 = *mdt_object_fid(o);
498 repbody->valid = OBD_MD_FLID | OBD_MD_MDS;
502 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
503 buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD, RCL_SERVER);
505 /* If it is dir object and client require MEA, then we got MEA */
506 if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
507 reqbody->valid & OBD_MD_MEA) {
508 /* Assumption: MDT_MD size is enough for lmv size. */
509 ma->ma_lmv = buffer->lb_buf;
510 ma->ma_lmv_size = buffer->lb_len;
511 ma->ma_need = MA_LMV | MA_INODE;
513 ma->ma_lmm = buffer->lb_buf;
514 ma->ma_lmm_size = buffer->lb_len;
515 ma->ma_need = MA_LOV | MA_INODE;
518 if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
519 reqbody->valid & OBD_MD_FLDIREA &&
520 lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
521 /* get default stripe info for this dir. */
522 ma->ma_need |= MA_LOV_DEF;
524 ma->ma_need |= ma_need;
525 if (ma->ma_need & MA_SOM)
526 ma->ma_som = &info->mti_u.som.data;
528 rc = mo_attr_get(env, next, ma);
530 CERROR("getattr error for "DFID": %d\n",
531 PFID(mdt_object_fid(o)), rc);
535 if (likely(ma->ma_valid & MA_INODE))
536 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
540 if (mdt_body_has_lov(la, reqbody)) {
541 if (ma->ma_valid & MA_LOV) {
542 LASSERT(ma->ma_lmm_size);
543 mdt_dump_lmm(D_INFO, ma->ma_lmm);
544 repbody->eadatasize = ma->ma_lmm_size;
545 if (S_ISDIR(la->la_mode))
546 repbody->valid |= OBD_MD_FLDIREA;
548 repbody->valid |= OBD_MD_FLEASIZE;
550 if (ma->ma_valid & MA_LMV) {
551 LASSERT(S_ISDIR(la->la_mode));
552 repbody->eadatasize = ma->ma_lmv_size;
553 repbody->valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
555 } else if (S_ISLNK(la->la_mode) &&
556 reqbody->valid & OBD_MD_LINKNAME) {
557 buffer->lb_buf = ma->ma_lmm;
558 buffer->lb_len = reqbody->eadatasize;
559 rc = mo_readlink(env, next, buffer);
560 if (unlikely(rc <= 0)) {
561 CERROR("readlink failed: %d\n", rc);
564 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
566 repbody->valid |= OBD_MD_LINKNAME;
567 repbody->eadatasize = rc;
569 ((char*)ma->ma_lmm)[rc - 1] = 0;
570 CDEBUG(D_INODE, "symlink dest %s, len = %d\n",
571 (char*)ma->ma_lmm, rc);
576 if (reqbody->valid & OBD_MD_FLMODEASIZE) {
577 repbody->max_cookiesize = info->mti_mdt->mdt_max_cookiesize;
578 repbody->max_mdsize = info->mti_mdt->mdt_max_mdsize;
579 repbody->valid |= OBD_MD_FLMODEASIZE;
580 CDEBUG(D_INODE, "I am going to change the MAX_MD_SIZE & "
581 "MAX_COOKIE to : %d:%d\n", repbody->max_mdsize,
582 repbody->max_cookiesize);
585 if (exp_connect_rmtclient(info->mti_exp) &&
586 reqbody->valid & OBD_MD_FLRMTPERM) {
587 void *buf = req_capsule_server_get(pill, &RMF_ACL);
589 /* mdt_getattr_lock only */
590 rc = mdt_pack_remote_perm(info, o, buf);
592 repbody->valid &= ~OBD_MD_FLRMTPERM;
593 repbody->aclsize = 0;
596 repbody->valid |= OBD_MD_FLRMTPERM;
597 repbody->aclsize = sizeof(struct mdt_remote_perm);
600 #ifdef CONFIG_FS_POSIX_ACL
601 else if ((req->rq_export->exp_connect_flags & OBD_CONNECT_ACL) &&
602 (reqbody->valid & OBD_MD_FLACL)) {
603 buffer->lb_buf = req_capsule_server_get(pill, &RMF_ACL);
604 buffer->lb_len = req_capsule_get_size(pill,
605 &RMF_ACL, RCL_SERVER);
606 if (buffer->lb_len > 0) {
607 rc = mo_xattr_get(env, next, buffer,
608 XATTR_NAME_ACL_ACCESS);
610 if (rc == -ENODATA) {
611 repbody->aclsize = 0;
612 repbody->valid |= OBD_MD_FLACL;
614 } else if (rc == -EOPNOTSUPP) {
617 CERROR("got acl size: %d\n", rc);
620 repbody->aclsize = rc;
621 repbody->valid |= OBD_MD_FLACL;
628 if (reqbody->valid & OBD_MD_FLMDSCAPA &&
629 info->mti_mdt->mdt_opts.mo_mds_capa &&
630 info->mti_exp->exp_connect_flags & OBD_CONNECT_MDS_CAPA) {
631 struct lustre_capa *capa;
633 capa = req_capsule_server_get(pill, &RMF_CAPA1);
635 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
636 rc = mo_capa_get(env, next, capa, 0);
639 repbody->valid |= OBD_MD_FLMDSCAPA;
644 static int mdt_renew_capa(struct mdt_thread_info *info)
646 struct mdt_object *obj = info->mti_object;
647 struct mdt_body *body;
648 struct lustre_capa *capa, *c;
652 /* if object doesn't exist, or server has disabled capability,
653 * return directly, client will find body->valid OBD_MD_FLOSSCAPA
656 if (!obj || !info->mti_mdt->mdt_opts.mo_oss_capa ||
657 !(info->mti_exp->exp_connect_flags & OBD_CONNECT_OSS_CAPA))
660 body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
661 LASSERT(body != NULL);
663 c = req_capsule_client_get(info->mti_pill, &RMF_CAPA1);
666 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
670 rc = mo_capa_get(info->mti_env, mdt_object_child(obj), capa, 1);
672 body->valid |= OBD_MD_FLOSSCAPA;
676 static int mdt_getattr(struct mdt_thread_info *info)
678 struct ptlrpc_request *req = mdt_info_req(info);
679 struct mdt_object *obj = info->mti_object;
680 struct req_capsule *pill = info->mti_pill;
681 struct mdt_body *reqbody;
682 struct mdt_body *repbody;
688 reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
691 if (reqbody->valid & OBD_MD_FLOSSCAPA) {
692 rc = req_capsule_server_pack(pill);
694 RETURN(err_serious(rc));
695 rc = mdt_renew_capa(info);
696 GOTO(out_shrink, rc);
699 LASSERT(obj != NULL);
700 LASSERT(lu_object_assert_exists(&obj->mot_obj.mo_lu));
702 mode = lu_object_attr(&obj->mot_obj.mo_lu);
703 if (S_ISLNK(mode) && (reqbody->valid & OBD_MD_LINKNAME) &&
704 (reqbody->eadatasize > info->mti_mdt->mdt_max_mdsize))
705 md_size = reqbody->eadatasize;
707 md_size = info->mti_mdt->mdt_max_mdsize;
709 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, md_size);
711 rc = req_capsule_server_pack(pill);
712 if (unlikely(rc != 0))
713 RETURN(err_serious(rc));
715 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
716 LASSERT(repbody != NULL);
717 repbody->eadatasize = 0;
718 repbody->aclsize = 0;
720 if (reqbody->valid & OBD_MD_FLRMTPERM)
721 rc = mdt_init_ucred(info, reqbody);
723 rc = mdt_check_ucred(info);
725 GOTO(out_shrink, rc);
727 info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
728 info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
731 * Don't check capability at all, because rename might getattr for
732 * remote obj, and at that time no capability is available.
734 mdt_set_capainfo(info, 1, &reqbody->fid1, BYPASS_CAPA);
735 rc = mdt_getattr_internal(info, obj, 0);
736 if (reqbody->valid & OBD_MD_FLRMTPERM)
737 mdt_exit_ucred(info);
741 mdt_counter_incr(req->rq_export, LPROC_MDT_GETATTR);
743 mdt_shrink_reply(info);
747 static int mdt_is_subdir(struct mdt_thread_info *info)
749 struct mdt_object *o = info->mti_object;
750 struct req_capsule *pill = info->mti_pill;
751 const struct mdt_body *body = info->mti_body;
752 struct mdt_body *repbody;
758 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
761 * We save last checked parent fid to @repbody->fid1 for remote
764 LASSERT(fid_is_sane(&body->fid2));
765 LASSERT(mdt_object_exists(o) > 0);
766 rc = mdo_is_subdir(info->mti_env, mdt_object_child(o),
767 &body->fid2, &repbody->fid1);
768 if (rc == 0 || rc == -EREMOTE)
769 repbody->valid |= OBD_MD_FLID;
774 static int mdt_raw_lookup(struct mdt_thread_info *info,
775 struct mdt_object *parent,
776 const struct lu_name *lname,
777 struct ldlm_reply *ldlm_rep)
779 struct md_object *next = mdt_object_child(info->mti_object);
780 const struct mdt_body *reqbody = info->mti_body;
781 struct lu_fid *child_fid = &info->mti_tmp_fid1;
782 struct mdt_body *repbody;
786 if (reqbody->valid != OBD_MD_FLID)
789 LASSERT(!info->mti_cross_ref);
791 /* Only got the fid of this obj by name */
792 rc = mdo_lookup(info->mti_env, next, lname, child_fid,
795 /* XXX is raw_lookup possible as intent operation? */
798 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
801 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
803 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
806 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
807 repbody->fid1 = *child_fid;
808 repbody->valid = OBD_MD_FLID;
814 * UPDATE lock should be taken against parent, and be release before exit;
815 * child_bits lock should be taken against child, and be returned back:
816 * (1)normal request should release the child lock;
817 * (2)intent request will grant the lock to client.
819 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
820 struct mdt_lock_handle *lhc,
822 struct ldlm_reply *ldlm_rep)
824 struct ptlrpc_request *req = mdt_info_req(info);
825 struct mdt_body *reqbody = NULL;
826 struct mdt_object *parent = info->mti_object;
827 struct mdt_object *child;
828 struct md_object *next = mdt_object_child(parent);
829 struct lu_fid *child_fid = &info->mti_tmp_fid1;
830 struct lu_name *lname = NULL;
831 const char *name = NULL;
833 struct mdt_lock_handle *lhp = NULL;
834 struct ldlm_lock *lock;
835 struct ldlm_res_id *res_id;
842 is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
843 LASSERT(ergo(is_resent,
844 lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
846 LASSERT(parent != NULL);
847 name = req_capsule_client_get(info->mti_pill, &RMF_NAME);
849 RETURN(err_serious(-EFAULT));
851 namelen = req_capsule_get_size(info->mti_pill, &RMF_NAME,
853 if (!info->mti_cross_ref) {
855 * XXX: Check for "namelen == 0" is for getattr by fid
856 * (OBD_CONNECT_ATTRFID), otherwise do not allow empty name,
857 * that is the name must contain at least one character and
858 * the terminating '\0'
861 reqbody = req_capsule_client_get(info->mti_pill,
863 if (unlikely(reqbody == NULL))
864 RETURN(err_serious(-EFAULT));
866 if (unlikely(!fid_is_sane(&reqbody->fid2)))
867 RETURN(err_serious(-EINVAL));
870 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
872 PFID(mdt_object_fid(parent)), PFID(&reqbody->fid2),
875 lname = mdt_name(info->mti_env, (char *)name, namelen);
876 CDEBUG(D_INODE, "getattr with lock for "DFID"/%s, "
877 "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
881 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
883 rc = mdt_object_exists(parent);
884 if (unlikely(rc == 0)) {
885 LU_OBJECT_DEBUG(D_WARNING, info->mti_env,
886 &parent->mot_obj.mo_lu,
887 "Parent doesn't exist!\n");
889 } else if (!info->mti_cross_ref) {
890 LASSERTF(rc > 0, "Parent "DFID" is on remote server\n",
891 PFID(mdt_object_fid(parent)));
894 rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
902 if (info->mti_cross_ref) {
903 /* Only getattr on the child. Parent is on another node. */
904 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
906 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
907 "ldlm_rep=%p\n", PFID(mdt_object_fid(child)), ldlm_rep);
910 /* Do not take lock for resent case. */
911 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
912 LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
913 lhc->mlh_reg_lh.cookie);
914 LASSERT(fid_res_name_eq(mdt_object_fid(child),
915 &lock->l_resource->lr_name));
919 mdt_lock_handle_init(lhc);
920 mdt_lock_reg_init(lhc, LCK_PR);
923 * Object's name is on another MDS, no lookup lock is
924 * needed here but update is.
926 child_bits &= ~MDS_INODELOCK_LOOKUP;
927 child_bits |= MDS_INODELOCK_UPDATE;
929 rc = mdt_object_lock(info, child, lhc, child_bits,
933 /* Finally, we can get attr for child. */
934 mdt_set_capainfo(info, 0, mdt_object_fid(child),
936 rc = mdt_getattr_internal(info, child, 0);
937 if (unlikely(rc != 0))
938 mdt_object_unlock(info, child, lhc, 1);
944 /* step 1: lock parent */
945 lhp = &info->mti_lh[MDT_LH_PARENT];
946 mdt_lock_pdo_init(lhp, LCK_PR, name, namelen);
947 rc = mdt_object_lock(info, parent, lhp, MDS_INODELOCK_UPDATE,
949 if (unlikely(rc != 0))
952 /* step 2: lookup child's fid by name */
953 rc = mdo_lookup(info->mti_env, next, lname, child_fid,
958 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
959 GOTO(out_parent, rc);
961 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
963 *child_fid = reqbody->fid2;
964 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
968 *step 3: find the child object by fid & lock it.
969 * regardless if it is local or remote.
971 child = mdt_object_find(info->mti_env, info->mti_mdt, child_fid);
973 if (unlikely(IS_ERR(child)))
974 GOTO(out_parent, rc = PTR_ERR(child));
976 /* Do not take lock for resent case. */
977 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
978 LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
979 lhc->mlh_reg_lh.cookie);
981 res_id = &lock->l_resource->lr_name;
982 if (!fid_res_name_eq(mdt_object_fid(child),
983 &lock->l_resource->lr_name)) {
984 LASSERTF(fid_res_name_eq(mdt_object_fid(parent),
985 &lock->l_resource->lr_name),
986 "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n",
987 (unsigned long)res_id->name[0],
988 (unsigned long)res_id->name[1],
989 (unsigned long)res_id->name[2],
990 PFID(mdt_object_fid(parent)));
991 CWARN("Although resent, but still not get child lock"
992 "parent:"DFID" child:"DFID"\n",
993 PFID(mdt_object_fid(parent)),
994 PFID(mdt_object_fid(child)));
995 lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
1003 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout*2);
1004 mdt_lock_handle_init(lhc);
1005 mdt_lock_reg_init(lhc, LCK_PR);
1007 if (mdt_object_exists(child) == 0) {
1008 LU_OBJECT_DEBUG(D_WARNING, info->mti_env,
1009 &child->mot_obj.mo_lu,
1010 "Object doesn't exist!\n");
1011 GOTO(out_child, rc = -ENOENT);
1014 if (!(child_bits & MDS_INODELOCK_UPDATE)) {
1015 struct md_attr *ma = &info->mti_attr;
1018 ma->ma_need = MA_INODE;
1019 rc = mo_attr_get(info->mti_env,
1020 mdt_object_child(child), ma);
1021 if (unlikely(rc != 0))
1022 GOTO(out_child, rc);
1024 /* If the file has not been changed for some time, we
1025 * return not only a LOOKUP lock, but also an UPDATE
1026 * lock and this might save us RPC on later STAT. For
1027 * directories, it also let negative dentry starts
1028 * working for this dir. */
1029 if (ma->ma_valid & MA_INODE &&
1030 ma->ma_attr.la_valid & LA_CTIME &&
1031 info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
1032 ma->ma_attr.la_ctime < cfs_time_current_sec())
1033 child_bits |= MDS_INODELOCK_UPDATE;
1036 rc = mdt_object_lock(info, child, lhc, child_bits,
1039 if (unlikely(rc != 0))
1040 GOTO(out_child, rc);
1043 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1044 /* Get MA_SOM attributes if update lock is given. */
1046 lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_UPDATE &&
1047 S_ISREG(lu_object_attr(&mdt_object_child(child)->mo_lu)))
1050 /* finally, we can get attr for child. */
1051 mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
1052 rc = mdt_getattr_internal(info, child, ma_need);
1053 if (unlikely(rc != 0)) {
1054 mdt_object_unlock(info, child, lhc, 1);
1056 /* Debugging code. */
1057 res_id = &lock->l_resource->lr_name;
1058 LDLM_DEBUG(lock, "Returning lock to client");
1059 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1060 &lock->l_resource->lr_name),
1061 "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n",
1062 (unsigned long)res_id->name[0],
1063 (unsigned long)res_id->name[1],
1064 (unsigned long)res_id->name[2],
1065 PFID(mdt_object_fid(child)));
1066 mdt_pack_size2body(info, child);
1069 LDLM_LOCK_PUT(lock);
1073 mdt_object_put(info->mti_env, child);
1076 mdt_object_unlock(info, parent, lhp, 1);
1080 /* normal handler: should release the child lock */
1081 static int mdt_getattr_name(struct mdt_thread_info *info)
1083 struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1084 struct mdt_body *reqbody;
1085 struct mdt_body *repbody;
1089 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1090 LASSERT(reqbody != NULL);
1091 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1092 LASSERT(repbody != NULL);
1094 info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
1095 info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
1096 repbody->eadatasize = 0;
1097 repbody->aclsize = 0;
1099 rc = mdt_init_ucred(info, reqbody);
1101 GOTO(out_shrink, rc);
1103 rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1104 if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1105 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1106 lhc->mlh_reg_lh.cookie = 0;
1108 mdt_exit_ucred(info);
1111 mdt_shrink_reply(info);
1115 static const struct lu_device_operations mdt_lu_ops;
1117 static int lu_device_is_mdt(struct lu_device *d)
1119 return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &mdt_lu_ops);
1122 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1123 void *karg, void *uarg);
1125 static int mdt_set_info(struct mdt_thread_info *info)
1127 struct ptlrpc_request *req = mdt_info_req(info);
1130 int keylen, vallen, rc = 0;
1133 rc = req_capsule_server_pack(info->mti_pill);
1137 key = req_capsule_client_get(info->mti_pill, &RMF_SETINFO_KEY);
1139 DEBUG_REQ(D_HA, req, "no set_info key");
1143 keylen = req_capsule_get_size(info->mti_pill, &RMF_SETINFO_KEY,
1146 val = req_capsule_client_get(info->mti_pill, &RMF_SETINFO_VAL);
1148 DEBUG_REQ(D_HA, req, "no set_info val");
1152 vallen = req_capsule_get_size(info->mti_pill, &RMF_SETINFO_VAL,
1155 /* Swab any part of val you need to here */
1156 if (KEY_IS(KEY_READ_ONLY)) {
1158 lustre_msg_set_status(req->rq_repmsg, 0);
1160 cfs_spin_lock(&req->rq_export->exp_lock);
1162 req->rq_export->exp_connect_flags |= OBD_CONNECT_RDONLY;
1164 req->rq_export->exp_connect_flags &=~OBD_CONNECT_RDONLY;
1165 cfs_spin_unlock(&req->rq_export->exp_lock);
1167 } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1168 struct changelog_setinfo *cs =
1169 (struct changelog_setinfo *)val;
1170 if (vallen != sizeof(*cs)) {
1171 CERROR("Bad changelog_clear setinfo size %d\n", vallen);
1174 if (ptlrpc_req_need_swab(req)) {
1175 __swab64s(&cs->cs_recno);
1176 __swab32s(&cs->cs_id);
1179 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, info->mti_exp,
1181 lustre_msg_set_status(req->rq_repmsg, rc);
1189 static int mdt_connect(struct mdt_thread_info *info)
1192 struct ptlrpc_request *req;
1194 req = mdt_info_req(info);
1195 rc = target_handle_connect(req);
1197 LASSERT(req->rq_export != NULL);
1198 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
1199 rc = mdt_init_sec_level(info);
1201 rc = mdt_init_idmap(info);
1203 obd_disconnect(class_export_get(req->rq_export));
1205 rc = err_serious(rc);
1210 static int mdt_disconnect(struct mdt_thread_info *info)
1215 rc = target_handle_disconnect(mdt_info_req(info));
1217 rc = err_serious(rc);
1221 static int mdt_sendpage(struct mdt_thread_info *info,
1222 struct lu_rdpg *rdpg, int nob)
1224 struct ptlrpc_request *req = mdt_info_req(info);
1225 struct obd_export *exp = req->rq_export;
1226 struct ptlrpc_bulk_desc *desc;
1227 struct l_wait_info *lwi = &info->mti_u.rdpg.mti_wait_info;
1234 desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, BULK_PUT_SOURCE,
1239 for (i = 0, tmpcount = nob;
1240 i < rdpg->rp_npages && tmpcount > 0; i++, tmpcount -= tmpsize) {
1241 tmpsize = min_t(int, tmpcount, CFS_PAGE_SIZE);
1242 ptlrpc_prep_bulk_page(desc, rdpg->rp_pages[i], 0, tmpsize);
1245 LASSERT(desc->bd_nob == nob);
1246 rc = target_bulk_io(exp, desc, lwi);
1247 ptlrpc_free_bulk(desc);
1251 #ifdef HAVE_SPLIT_SUPPORT
1253 * Retrieve dir entry from the page and insert it to the slave object, actually,
1254 * this should be in osd layer, but since it will not in the final product, so
1255 * just do it here and do not define more moo api anymore for this.
1257 static int mdt_write_dir_page(struct mdt_thread_info *info, struct page *page,
1260 struct mdt_object *object = info->mti_object;
1261 struct lu_fid *lf = &info->mti_tmp_fid2;
1262 struct md_attr *ma = &info->mti_attr;
1263 struct lu_dirpage *dp;
1264 struct lu_dirent *ent;
1265 int rc = 0, offset = 0;
1268 /* Make sure we have at least one entry. */
1273 * Disable trans for this name insert, since it will include many trans
1276 info->mti_no_need_trans = 1;
1278 * When write_dir_page, no need update parent's ctime,
1279 * and no permission check for name_insert.
1281 ma->ma_attr.la_ctime = 0;
1282 ma->ma_attr.la_valid = LA_MODE;
1283 ma->ma_valid = MA_INODE;
1286 dp = page_address(page);
1287 offset = (int)((__u32)lu_dirent_start(dp) - (__u32)dp);
1289 for (ent = lu_dirent_start(dp); ent != NULL;
1290 ent = lu_dirent_next(ent)) {
1291 struct lu_name *lname;
1294 if (le16_to_cpu(ent->lde_namelen) == 0)
1297 fid_le_to_cpu(lf, &ent->lde_fid);
1298 if (le64_to_cpu(ent->lde_hash) & MAX_HASH_HIGHEST_BIT)
1299 ma->ma_attr.la_mode = S_IFDIR;
1301 ma->ma_attr.la_mode = 0;
1302 OBD_ALLOC(name, le16_to_cpu(ent->lde_namelen) + 1);
1304 GOTO(out, rc = -ENOMEM);
1306 memcpy(name, ent->lde_name, le16_to_cpu(ent->lde_namelen));
1307 lname = mdt_name(info->mti_env, name,
1308 le16_to_cpu(ent->lde_namelen));
1309 ma->ma_attr_flags |= (MDS_PERM_BYPASS | MDS_QUOTA_IGNORE);
1310 rc = mdo_name_insert(info->mti_env,
1311 md_object_next(&object->mot_obj),
1313 OBD_FREE(name, le16_to_cpu(ent->lde_namelen) + 1);
1315 CERROR("Can't insert %*.*s, rc %d\n",
1316 le16_to_cpu(ent->lde_namelen),
1317 le16_to_cpu(ent->lde_namelen),
1322 offset += lu_dirent_size(ent);
1332 static int mdt_bulk_timeout(void *data)
1336 CERROR("mdt bulk transfer timeout \n");
1341 static int mdt_writepage(struct mdt_thread_info *info)
1343 struct ptlrpc_request *req = mdt_info_req(info);
1344 struct mdt_body *reqbody;
1345 struct l_wait_info *lwi;
1346 struct ptlrpc_bulk_desc *desc;
1352 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1353 if (reqbody == NULL)
1354 RETURN(err_serious(-EFAULT));
1356 desc = ptlrpc_prep_bulk_exp(req, 1, BULK_GET_SINK, MDS_BULK_PORTAL);
1358 RETURN(err_serious(-ENOMEM));
1360 /* allocate the page for the desc */
1361 page = cfs_alloc_page(CFS_ALLOC_STD);
1363 GOTO(desc_cleanup, rc = -ENOMEM);
1365 CDEBUG(D_INFO, "Received page offset %d size %d \n",
1366 (int)reqbody->size, (int)reqbody->nlink);
1368 ptlrpc_prep_bulk_page(desc, page, (int)reqbody->size,
1369 (int)reqbody->nlink);
1371 rc = sptlrpc_svc_prep_bulk(req, desc);
1373 GOTO(cleanup_page, rc);
1375 * Check if client was evicted while we were doing i/o before touching
1380 GOTO(cleanup_page, rc = -ENOMEM);
1382 if (desc->bd_export->exp_failed)
1385 rc = ptlrpc_start_bulk_transfer (desc);
1387 *lwi = LWI_TIMEOUT_INTERVAL(obd_timeout * CFS_HZ / 4, CFS_HZ,
1388 mdt_bulk_timeout, desc);
1389 rc = l_wait_event(desc->bd_waitq, !ptlrpc_bulk_active(desc) ||
1390 desc->bd_export->exp_failed, lwi);
1391 LASSERT(rc == 0 || rc == -ETIMEDOUT);
1392 if (rc == -ETIMEDOUT) {
1393 DEBUG_REQ(D_ERROR, req, "timeout on bulk GET");
1394 ptlrpc_abort_bulk(desc);
1395 } else if (desc->bd_export->exp_failed) {
1396 DEBUG_REQ(D_ERROR, req, "Eviction on bulk GET");
1398 ptlrpc_abort_bulk(desc);
1399 } else if (!desc->bd_success ||
1400 desc->bd_nob_transferred != desc->bd_nob) {
1401 DEBUG_REQ(D_ERROR, req, "%s bulk GET %d(%d)",
1403 "truncated" : "network error on",
1404 desc->bd_nob_transferred, desc->bd_nob);
1405 /* XXX should this be a different errno? */
1409 DEBUG_REQ(D_ERROR, req, "ptlrpc_bulk_get failed: rc %d", rc);
1412 GOTO(cleanup_lwi, rc);
1413 rc = mdt_write_dir_page(info, page, reqbody->nlink);
1418 cfs_free_page(page);
1420 ptlrpc_free_bulk(desc);
1425 static int mdt_readpage(struct mdt_thread_info *info)
1427 struct mdt_object *object = info->mti_object;
1428 struct lu_rdpg *rdpg = &info->mti_u.rdpg.mti_rdpg;
1429 struct mdt_body *reqbody;
1430 struct mdt_body *repbody;
1435 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
1436 RETURN(err_serious(-ENOMEM));
1438 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1439 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1440 if (reqbody == NULL || repbody == NULL)
1441 RETURN(err_serious(-EFAULT));
1444 * prepare @rdpg before calling lower layers and transfer itself. Here
1445 * reqbody->size contains offset of where to start to read and
1446 * reqbody->nlink contains number bytes to read.
1448 rdpg->rp_hash = reqbody->size;
1449 if (rdpg->rp_hash != reqbody->size) {
1450 CERROR("Invalid hash: "LPX64" != "LPX64"\n",
1451 rdpg->rp_hash, reqbody->size);
1455 rdpg->rp_attrs = reqbody->mode;
1456 if (info->mti_exp->exp_connect_flags & OBD_CONNECT_64BITHASH)
1457 rdpg->rp_attrs |= LUDA_64BITHASH;
1458 rdpg->rp_count = min_t(unsigned int, reqbody->nlink,
1459 PTLRPC_MAX_BRW_SIZE);
1460 rdpg->rp_npages = (rdpg->rp_count + CFS_PAGE_SIZE - 1) >>
1462 OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1463 if (rdpg->rp_pages == NULL)
1466 for (i = 0; i < rdpg->rp_npages; ++i) {
1467 rdpg->rp_pages[i] = cfs_alloc_page(CFS_ALLOC_STD);
1468 if (rdpg->rp_pages[i] == NULL)
1469 GOTO(free_rdpg, rc = -ENOMEM);
1472 /* call lower layers to fill allocated pages with directory data */
1473 rc = mo_readpage(info->mti_env, mdt_object_child(object), rdpg);
1475 GOTO(free_rdpg, rc);
1477 /* send pages to client */
1478 rc = mdt_sendpage(info, rdpg, rc);
1483 for (i = 0; i < rdpg->rp_npages; i++)
1484 if (rdpg->rp_pages[i] != NULL)
1485 cfs_free_page(rdpg->rp_pages[i]);
1486 OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1488 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1494 static int mdt_reint_internal(struct mdt_thread_info *info,
1495 struct mdt_lock_handle *lhc,
1498 struct req_capsule *pill = info->mti_pill;
1499 struct mdt_device *mdt = info->mti_mdt;
1500 struct md_quota *mq = md_quota(info->mti_env);
1501 struct mdt_body *repbody;
1506 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1507 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1508 mdt->mdt_max_mdsize);
1509 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1510 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER,
1511 mdt->mdt_max_cookiesize);
1513 rc = req_capsule_server_pack(pill);
1515 CERROR("Can't pack response, rc %d\n", rc);
1516 RETURN(err_serious(rc));
1519 if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
1520 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1522 repbody->eadatasize = 0;
1523 repbody->aclsize = 0;
1526 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNPACK))
1527 GOTO(out_shrink, rc = err_serious(-EFAULT));
1529 rc = mdt_reint_unpack(info, op);
1531 CERROR("Can't unpack reint, rc %d\n", rc);
1532 GOTO(out_shrink, rc = err_serious(rc));
1535 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
1537 /* for replay no cookkie / lmm need, because client have this already */
1538 if (info->mti_spec.no_create == 1) {
1539 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1540 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
1542 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1543 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER,
1547 rc = mdt_init_ucred_reint(info);
1549 GOTO(out_shrink, rc);
1551 rc = mdt_fix_attr_ucred(info, op);
1553 GOTO(out_ucred, rc = err_serious(rc));
1555 if (mdt_check_resent(info, mdt_reconstruct, lhc)) {
1556 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
1557 GOTO(out_ucred, rc);
1559 mq->mq_exp = info->mti_exp;
1560 rc = mdt_reint_rec(info, lhc);
1563 mdt_exit_ucred(info);
1565 mdt_shrink_reply(info);
1569 static long mdt_reint_opcode(struct mdt_thread_info *info,
1570 const struct req_format **fmt)
1572 struct mdt_rec_reint *rec;
1575 opc = err_serious(-EFAULT);
1576 rec = req_capsule_client_get(info->mti_pill, &RMF_REC_REINT);
1578 opc = rec->rr_opcode;
1579 DEBUG_REQ(D_INODE, mdt_info_req(info), "reint opt = %ld", opc);
1580 if (opc < REINT_MAX && fmt[opc] != NULL)
1581 req_capsule_extend(info->mti_pill, fmt[opc]);
1583 CERROR("Unsupported opc: %ld\n", opc);
1584 opc = err_serious(opc);
1590 static int mdt_reint(struct mdt_thread_info *info)
1595 static const struct req_format *reint_fmts[REINT_MAX] = {
1596 [REINT_SETATTR] = &RQF_MDS_REINT_SETATTR,
1597 [REINT_CREATE] = &RQF_MDS_REINT_CREATE,
1598 [REINT_LINK] = &RQF_MDS_REINT_LINK,
1599 [REINT_UNLINK] = &RQF_MDS_REINT_UNLINK,
1600 [REINT_RENAME] = &RQF_MDS_REINT_RENAME,
1601 [REINT_OPEN] = &RQF_MDS_REINT_OPEN,
1602 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR
1607 opc = mdt_reint_opcode(info, reint_fmts);
1610 * No lock possible here from client to pass it to reint code
1613 rc = mdt_reint_internal(info, NULL, opc);
1618 info->mti_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
1622 /* this should sync the whole device */
1623 static int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
1625 struct dt_device *dt = mdt->mdt_bottom;
1629 rc = dt->dd_ops->dt_sync(env, dt);
1633 /* this should sync this object */
1634 static int mdt_object_sync(struct mdt_thread_info *info)
1636 struct md_object *next;
1640 if (!mdt_object_exists(info->mti_object)) {
1641 CWARN("Non existing object "DFID"!\n",
1642 PFID(mdt_object_fid(info->mti_object)));
1645 next = mdt_object_child(info->mti_object);
1646 rc = mo_object_sync(info->mti_env, next);
1651 static int mdt_sync(struct mdt_thread_info *info)
1653 struct ptlrpc_request *req = mdt_info_req(info);
1654 struct req_capsule *pill = info->mti_pill;
1655 struct mdt_body *body;
1659 /* The fid may be zero, so we req_capsule_set manually */
1660 req_capsule_set(pill, &RQF_MDS_SYNC);
1662 body = req_capsule_client_get(pill, &RMF_MDT_BODY);
1664 RETURN(err_serious(-EINVAL));
1666 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
1667 RETURN(err_serious(-ENOMEM));
1669 if (fid_seq(&body->fid1) == 0) {
1670 /* sync the whole device */
1671 rc = req_capsule_server_pack(pill);
1673 rc = mdt_device_sync(info->mti_env, info->mti_mdt);
1675 rc = err_serious(rc);
1677 /* sync an object */
1678 rc = mdt_unpack_req_pack_rep(info, HABEO_CORPUS|HABEO_REFERO);
1680 rc = mdt_object_sync(info);
1682 struct md_object *next;
1683 const struct lu_fid *fid;
1684 struct lu_attr *la = &info->mti_attr.ma_attr;
1686 next = mdt_object_child(info->mti_object);
1687 info->mti_attr.ma_need = MA_INODE;
1688 info->mti_attr.ma_valid = 0;
1689 rc = mo_attr_get(info->mti_env, next,
1692 body = req_capsule_server_get(pill,
1694 fid = mdt_object_fid(info->mti_object);
1695 mdt_pack_attr2body(info, body, la, fid);
1699 rc = err_serious(rc);
1702 mdt_counter_incr(req->rq_export, LPROC_MDT_SYNC);
1707 #ifdef HAVE_QUOTA_SUPPORT
1708 static int mdt_quotacheck_handle(struct mdt_thread_info *info)
1710 struct obd_quotactl *oqctl;
1711 struct req_capsule *pill = info->mti_pill;
1712 struct obd_export *exp = info->mti_exp;
1713 struct md_quota *mq = md_quota(info->mti_env);
1714 struct md_device *next = info->mti_mdt->mdt_child;
1718 oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
1722 /* remote client has no permission for quotacheck */
1723 if (unlikely(exp_connect_rmtclient(exp)))
1726 rc = req_capsule_server_pack(pill);
1731 rc = next->md_ops->mdo_quota.mqo_check(info->mti_env, next,
1736 static int mdt_quotactl_handle(struct mdt_thread_info *info)
1738 struct obd_quotactl *oqctl, *repoqc;
1739 struct req_capsule *pill = info->mti_pill;
1740 struct obd_export *exp = info->mti_exp;
1741 struct md_quota *mq = md_quota(info->mti_env);
1742 struct md_device *next = info->mti_mdt->mdt_child;
1743 const struct md_quota_operations *mqo = &next->md_ops->mdo_quota;
1747 oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
1752 if (exp_connect_rmtclient(exp)) {
1753 struct ptlrpc_request *req = mdt_info_req(info);
1754 struct mdt_export_data *med = mdt_req2med(req);
1755 struct lustre_idmap_table *idmap = med->med_idmap;
1757 if (unlikely(oqctl->qc_cmd != Q_GETQUOTA &&
1758 oqctl->qc_cmd != Q_GETINFO))
1762 if (oqctl->qc_type == USRQUOTA)
1763 id = lustre_idmap_lookup_uid(NULL, idmap, 0,
1765 else if (oqctl->qc_type == GRPQUOTA)
1766 id = lustre_idmap_lookup_gid(NULL, idmap, 0,
1771 if (id == CFS_IDMAP_NOTFOUND) {
1772 CDEBUG(D_QUOTA, "no mapping for id %u\n",
1778 rc = req_capsule_server_pack(pill);
1782 repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
1783 LASSERT(repoqc != NULL);
1786 switch (oqctl->qc_cmd) {
1788 rc = mqo->mqo_on(info->mti_env, next, oqctl->qc_type);
1791 rc = mqo->mqo_off(info->mti_env, next, oqctl->qc_type);
1794 rc = mqo->mqo_setinfo(info->mti_env, next, oqctl->qc_type, id,
1798 rc = mqo->mqo_getinfo(info->mti_env, next, oqctl->qc_type, id,
1802 rc = mqo->mqo_setquota(info->mti_env, next, oqctl->qc_type, id,
1806 rc = mqo->mqo_getquota(info->mti_env, next, oqctl->qc_type, id,
1810 rc = mqo->mqo_getoinfo(info->mti_env, next, oqctl->qc_type, id,
1814 rc = mqo->mqo_getoquota(info->mti_env, next, oqctl->qc_type, id,
1817 case LUSTRE_Q_INVALIDATE:
1818 rc = mqo->mqo_invalidate(info->mti_env, next, oqctl->qc_type);
1820 case LUSTRE_Q_FINVALIDATE:
1821 rc = mqo->mqo_finvalidate(info->mti_env, next, oqctl->qc_type);
1824 CERROR("unsupported mdt_quotactl command: %d\n",
1836 * OBD PING and other handlers.
1838 static int mdt_obd_ping(struct mdt_thread_info *info)
1843 req_capsule_set(info->mti_pill, &RQF_OBD_PING);
1845 rc = target_handle_ping(mdt_info_req(info));
1847 rc = err_serious(rc);
1851 static int mdt_obd_log_cancel(struct mdt_thread_info *info)
1853 return err_serious(-EOPNOTSUPP);
1856 static int mdt_obd_qc_callback(struct mdt_thread_info *info)
1858 return err_serious(-EOPNOTSUPP);
1866 /** clone llog ctxt from child (mdd)
1867 * This allows remote llog (replicator) access.
1868 * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
1869 * context was originally set up, or we can handle them directly.
1870 * I choose the latter, but that means I need any llog
1871 * contexts set up by child to be accessable by the mdt. So we clone the
1872 * context into our context list here.
1874 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
1877 struct md_device *next = mdt->mdt_child;
1878 struct llog_ctxt *ctxt;
1881 if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
1884 rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
1885 if (rc || ctxt == NULL) {
1886 CERROR("Can't get mdd ctxt %d\n", rc);
1890 rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
1892 CERROR("Can't set mdt ctxt %d\n", rc);
1897 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
1898 struct mdt_device *mdt, int idx)
1900 struct llog_ctxt *ctxt;
1902 ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
1905 /* Put once for the get we just did, and once for the clone */
1906 llog_ctxt_put(ctxt);
1907 llog_ctxt_put(ctxt);
1911 static int mdt_llog_create(struct mdt_thread_info *info)
1915 req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_CREATE);
1916 rc = llog_origin_handle_create(mdt_info_req(info));
1917 return (rc < 0 ? err_serious(rc) : rc);
1920 static int mdt_llog_destroy(struct mdt_thread_info *info)
1924 req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_DESTROY);
1925 rc = llog_origin_handle_destroy(mdt_info_req(info));
1926 return (rc < 0 ? err_serious(rc) : rc);
1929 static int mdt_llog_read_header(struct mdt_thread_info *info)
1933 req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_READ_HEADER);
1934 rc = llog_origin_handle_read_header(mdt_info_req(info));
1935 return (rc < 0 ? err_serious(rc) : rc);
1938 static int mdt_llog_next_block(struct mdt_thread_info *info)
1942 req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_NEXT_BLOCK);
1943 rc = llog_origin_handle_next_block(mdt_info_req(info));
1944 return (rc < 0 ? err_serious(rc) : rc);
1947 static int mdt_llog_prev_block(struct mdt_thread_info *info)
1951 req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_PREV_BLOCK);
1952 rc = llog_origin_handle_prev_block(mdt_info_req(info));
1953 return (rc < 0 ? err_serious(rc) : rc);
1960 static struct ldlm_callback_suite cbs = {
1961 .lcs_completion = ldlm_server_completion_ast,
1962 .lcs_blocking = ldlm_server_blocking_ast,
1966 static int mdt_enqueue(struct mdt_thread_info *info)
1968 struct ptlrpc_request *req;
1972 * info->mti_dlm_req already contains swapped and (if necessary)
1973 * converted dlm request.
1975 LASSERT(info->mti_dlm_req != NULL);
1977 req = mdt_info_req(info);
1978 rc = ldlm_handle_enqueue0(info->mti_mdt->mdt_namespace,
1979 req, info->mti_dlm_req, &cbs);
1980 info->mti_fail_id = OBD_FAIL_LDLM_REPLY;
1981 return rc ? err_serious(rc) : req->rq_status;
1984 static int mdt_convert(struct mdt_thread_info *info)
1987 struct ptlrpc_request *req;
1989 LASSERT(info->mti_dlm_req);
1990 req = mdt_info_req(info);
1991 rc = ldlm_handle_convert0(req, info->mti_dlm_req);
1992 return rc ? err_serious(rc) : req->rq_status;
1995 static int mdt_bl_callback(struct mdt_thread_info *info)
1997 CERROR("bl callbacks should not happen on MDS\n");
1999 return err_serious(-EOPNOTSUPP);
2002 static int mdt_cp_callback(struct mdt_thread_info *info)
2004 CERROR("cp callbacks should not happen on MDS\n");
2006 return err_serious(-EOPNOTSUPP);
2010 * sec context handlers
2012 static int mdt_sec_ctx_handle(struct mdt_thread_info *info)
2016 rc = mdt_handle_idmap(info);
2019 struct ptlrpc_request *req = mdt_info_req(info);
2022 opc = lustre_msg_get_opc(req->rq_reqmsg);
2023 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT)
2024 sptlrpc_svc_ctx_invalidate(req);
2027 CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
2032 static struct mdt_object *mdt_obj(struct lu_object *o)
2034 LASSERT(lu_device_is_mdt(o->lo_dev));
2035 return container_of0(o, struct mdt_object, mot_obj.mo_lu);
2038 struct mdt_object *mdt_object_find(const struct lu_env *env,
2039 struct mdt_device *d,
2040 const struct lu_fid *f)
2042 struct lu_object *o;
2043 struct mdt_object *m;
2046 CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2047 o = lu_object_find(env, &d->mdt_md_dev.md_lu_dev, f, NULL);
2048 if (unlikely(IS_ERR(o)))
2049 m = (struct mdt_object *)o;
2056 * Asyncronous commit for mdt device.
2058 * Pass asynchonous commit call down the MDS stack.
2060 * \param env environment
2061 * \param mdt the mdt device
2063 static void mdt_device_commit_async(const struct lu_env *env,
2064 struct mdt_device *mdt)
2066 struct dt_device *dt = mdt->mdt_bottom;
2069 rc = dt->dd_ops->dt_commit_async(env, dt);
2070 if (unlikely(rc != 0))
2071 CWARN("async commit start failed with rc = %d", rc);
2075 * Mark the lock as "synchonous".
2077 * Mark the lock to deffer transaction commit to the unlock time.
2079 * \param lock the lock to mark as "synchonous"
2081 * \see mdt_is_lock_sync
2082 * \see mdt_save_lock
2084 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2086 lock->l_ast_data = (void*)1;
2090 * Check whehter the lock "synchonous" or not.
2092 * \param lock the lock to check
2093 * \retval 1 the lock is "synchonous"
2094 * \retval 0 the lock isn't "synchronous"
2096 * \see mdt_set_lock_sync
2097 * \see mdt_save_lock
2099 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2101 return lock->l_ast_data != NULL;
2105 * Blocking AST for mdt locks.
2107 * Starts transaction commit if in case of COS lock conflict or
2108 * deffers such a commit to the mdt_save_lock.
2110 * \param lock the lock which blocks a request or cancelling lock
2111 * \param desc unused
2112 * \param data unused
2113 * \param flag indicates whether this cancelling or blocking callback
2115 * \see ldlm_blocking_ast_nocheck
2117 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2118 void *data, int flag)
2120 struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2121 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2125 if (flag == LDLM_CB_CANCELING)
2127 lock_res_and_lock(lock);
2128 if (lock->l_blocking_ast != mdt_blocking_ast) {
2129 unlock_res_and_lock(lock);
2132 if (mdt_cos_is_enabled(mdt) &&
2133 lock->l_req_mode & (LCK_PW | LCK_EX) &&
2134 lock->l_blocking_lock != NULL &&
2135 lock->l_client_cookie != lock->l_blocking_lock->l_client_cookie) {
2136 mdt_set_lock_sync(lock);
2138 rc = ldlm_blocking_ast_nocheck(lock);
2140 /* There is no lock conflict if l_blocking_lock == NULL,
2141 * it indicates a blocking ast sent from ldlm_lock_decref_internal
2142 * when the last reference to a local lock was released */
2143 if (lock->l_req_mode == LCK_COS && lock->l_blocking_lock != NULL) {
2146 rc = lu_env_init(&env, LCT_MD_THREAD);
2147 if (unlikely(rc != 0))
2148 CWARN("lu_env initialization failed with rc = %d,"
2149 "cannot start asynchronous commit\n", rc);
2151 mdt_device_commit_async(&env, mdt);
2157 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2158 struct mdt_lock_handle *lh, __u64 ibits, int locality)
2160 struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
2161 ldlm_policy_data_t *policy = &info->mti_policy;
2162 struct ldlm_res_id *res_id = &info->mti_res_id;
2166 LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2167 LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2168 LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
2169 LASSERT(lh->mlh_type != MDT_NUL_LOCK);
2171 if (mdt_object_exists(o) < 0) {
2172 if (locality == MDT_CROSS_LOCK) {
2173 /* cross-ref object fix */
2174 ibits &= ~MDS_INODELOCK_UPDATE;
2175 ibits |= MDS_INODELOCK_LOOKUP;
2177 LASSERT(!(ibits & MDS_INODELOCK_UPDATE));
2178 LASSERT(ibits & MDS_INODELOCK_LOOKUP);
2180 /* No PDO lock on remote object */
2181 LASSERT(lh->mlh_type != MDT_PDO_LOCK);
2184 if (lh->mlh_type == MDT_PDO_LOCK) {
2185 /* check for exists after object is locked */
2186 if (mdt_object_exists(o) == 0) {
2187 /* Non-existent object shouldn't have PDO lock */
2190 /* Non-dir object shouldn't have PDO lock */
2191 if (!S_ISDIR(lu_object_attr(&o->mot_obj.mo_lu)))
2196 memset(policy, 0, sizeof(*policy));
2197 fid_build_reg_res_name(mdt_object_fid(o), res_id);
2200 * Take PDO lock on whole directory and build correct @res_id for lock
2201 * on part of directory.
2203 if (lh->mlh_pdo_hash != 0) {
2204 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
2205 mdt_lock_pdo_mode(info, o, lh);
2206 if (lh->mlh_pdo_mode != LCK_NL) {
2208 * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
2209 * is never going to be sent to client and we do not
2210 * want it slowed down due to possible cancels.
2212 policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
2213 rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
2214 policy, res_id, LDLM_FL_ATOMIC_CB,
2215 &info->mti_exp->exp_handle.h_cookie);
2221 * Finish res_id initializing by name hash marking part of
2222 * directory which is taking modification.
2224 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
2227 policy->l_inodebits.bits = ibits;
2230 * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
2231 * going to be sent to client. If it is - mdt_intent_policy() path will
2232 * fix it up and turn FL_LOCAL flag off.
2234 rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
2235 res_id, LDLM_FL_LOCAL_ONLY | LDLM_FL_ATOMIC_CB,
2236 &info->mti_exp->exp_handle.h_cookie);
2238 mdt_object_unlock(info, o, lh, 1);
2239 else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
2240 lh->mlh_pdo_hash != 0 &&
2241 (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX)) {
2242 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
2249 * Save a lock within request object.
2251 * Keep the lock referenced until whether client ACK or transaction
2252 * commit happens or release the lock immediately depending on input
2253 * parameters. If COS is ON, a write lock is converted to COS lock
2256 * \param info thead info object
2257 * \param h lock handle
2258 * \param mode lock mode
2259 * \param decref force immediate lock releasing
2262 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
2263 ldlm_mode_t mode, int decref)
2267 if (lustre_handle_is_used(h)) {
2268 if (decref || !info->mti_has_trans ||
2269 !(mode & (LCK_PW | LCK_EX))){
2270 mdt_fid_unlock(h, mode);
2272 struct mdt_device *mdt = info->mti_mdt;
2273 struct ldlm_lock *lock = ldlm_handle2lock(h);
2274 struct ptlrpc_request *req = mdt_info_req(info);
2277 LASSERTF(lock != NULL, "no lock for cookie "LPX64"\n",
2279 CDEBUG(D_HA, "request = %p reply state = %p"
2280 " transno = "LPD64"\n",
2281 req, req->rq_reply_state, req->rq_transno);
2282 if (mdt_cos_is_enabled(mdt)) {
2284 ldlm_lock_downgrade(lock, LCK_COS);
2287 ptlrpc_save_lock(req, h, mode, no_ack);
2288 if (mdt_is_lock_sync(lock)) {
2289 CDEBUG(D_HA, "found sync-lock,"
2290 " async commit started\n");
2291 mdt_device_commit_async(info->mti_env,
2294 LDLM_LOCK_PUT(lock);
2303 * Unlock mdt object.
2305 * Immeditely release the regular lock and the PDO lock or save the
2306 * lock in reqeuest and keep them referenced until client ACK or
2307 * transaction commit.
2309 * \param info thread info object
2310 * \param o mdt object
2311 * \param lh mdt lock handle referencing regular and PDO locks
2312 * \param decref force immediate lock releasing
2314 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
2315 struct mdt_lock_handle *lh, int decref)
2319 mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
2320 mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
2325 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
2326 const struct lu_fid *f,
2327 struct mdt_lock_handle *lh,
2330 struct mdt_object *o;
2332 o = mdt_object_find(info->mti_env, info->mti_mdt, f);
2336 rc = mdt_object_lock(info, o, lh, ibits,
2339 mdt_object_put(info->mti_env, o);
2346 void mdt_object_unlock_put(struct mdt_thread_info * info,
2347 struct mdt_object * o,
2348 struct mdt_lock_handle *lh,
2351 mdt_object_unlock(info, o, lh, decref);
2352 mdt_object_put(info->mti_env, o);
2355 static struct mdt_handler *mdt_handler_find(__u32 opc,
2356 struct mdt_opc_slice *supported)
2358 struct mdt_opc_slice *s;
2359 struct mdt_handler *h;
2362 for (s = supported; s->mos_hs != NULL; s++) {
2363 if (s->mos_opc_start <= opc && opc < s->mos_opc_end) {
2364 h = s->mos_hs + (opc - s->mos_opc_start);
2365 if (likely(h->mh_opc != 0))
2366 LASSERTF(h->mh_opc == opc,
2367 "opcode mismatch %d != %d\n",
2370 h = NULL; /* unsupported opc */
2377 static int mdt_lock_resname_compat(struct mdt_device *m,
2378 struct ldlm_request *req)
2380 /* XXX something... later. */
2384 static int mdt_lock_reply_compat(struct mdt_device *m, struct ldlm_reply *rep)
2386 /* XXX something... later. */
2391 * Generic code handling requests that have struct mdt_body passed in:
2393 * - extract mdt_body from request and save it in @info, if present;
2395 * - create lu_object, corresponding to the fid in mdt_body, and save it in
2398 * - if HABEO_CORPUS flag is set for this request type check whether object
2399 * actually exists on storage (lu_object_exists()).
2402 static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags)
2404 const struct mdt_body *body;
2405 struct mdt_object *obj;
2406 const struct lu_env *env;
2407 struct req_capsule *pill;
2411 env = info->mti_env;
2412 pill = info->mti_pill;
2414 body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
2418 if (!(body->valid & OBD_MD_FLID))
2421 if (!fid_is_sane(&body->fid1)) {
2422 CERROR("Invalid fid: "DFID"\n", PFID(&body->fid1));
2427 * Do not get size or any capa fields before we check that request
2428 * contains capa actually. There are some requests which do not, for
2429 * instance MDS_IS_SUBDIR.
2431 if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_CLIENT) &&
2432 req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
2433 mdt_set_capainfo(info, 0, &body->fid1,
2434 req_capsule_client_get(pill, &RMF_CAPA1));
2436 obj = mdt_object_find(env, info->mti_mdt, &body->fid1);
2438 if ((flags & HABEO_CORPUS) &&
2439 !mdt_object_exists(obj)) {
2440 mdt_object_put(env, obj);
2441 /* for capability renew ENOENT will be handled in
2443 if (body->valid & OBD_MD_FLOSSCAPA)
2448 info->mti_object = obj;
2457 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags)
2459 struct req_capsule *pill = info->mti_pill;
2463 if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
2464 rc = mdt_body_unpack(info, flags);
2468 if (rc == 0 && (flags & HABEO_REFERO)) {
2469 struct mdt_device *mdt = info->mti_mdt;
2473 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2474 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2475 mdt->mdt_max_mdsize);
2476 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2477 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER,
2478 mdt->mdt_max_cookiesize);
2480 rc = req_capsule_server_pack(pill);
2485 static int mdt_init_capa_ctxt(const struct lu_env *env, struct mdt_device *m)
2487 struct md_device *next = m->mdt_child;
2489 return next->md_ops->mdo_init_capa_ctxt(env, next,
2490 m->mdt_opts.mo_mds_capa,
2491 m->mdt_capa_timeout,
2497 * Invoke handler for this request opc. Also do necessary preprocessing
2498 * (according to handler ->mh_flags), and post-processing (setting of
2499 * ->last_{xid,committed}).
2501 static int mdt_req_handle(struct mdt_thread_info *info,
2502 struct mdt_handler *h, struct ptlrpc_request *req)
2504 int rc, serious = 0;
2509 LASSERT(h->mh_act != NULL);
2510 LASSERT(h->mh_opc == lustre_msg_get_opc(req->rq_reqmsg));
2511 LASSERT(current->journal_info == NULL);
2514 * Checking for various OBD_FAIL_$PREF_$OPC_NET codes. _Do_ not try
2515 * to put same checks into handlers like mdt_close(), mdt_reint(),
2516 * etc., without talking to mdt authors first. Checking same thing
2517 * there again is useless and returning 0 error without packing reply
2518 * is buggy! Handlers either pack reply or return error.
2520 * We return 0 here and do not send any reply in order to emulate
2521 * network failure. Do not send any reply in case any of NET related
2522 * fail_id has occured.
2524 if (OBD_FAIL_CHECK_ORSET(h->mh_fail_id, OBD_FAIL_ONCE))
2528 flags = h->mh_flags;
2529 LASSERT(ergo(flags & (HABEO_CORPUS|HABEO_REFERO), h->mh_fmt != NULL));
2531 if (h->mh_fmt != NULL) {
2532 req_capsule_set(info->mti_pill, h->mh_fmt);
2533 rc = mdt_unpack_req_pack_rep(info, flags);
2536 if (rc == 0 && flags & MUTABOR &&
2537 req->rq_export->exp_connect_flags & OBD_CONNECT_RDONLY)
2538 /* should it be rq_status? */
2541 if (rc == 0 && flags & HABEO_CLAVIS) {
2542 struct ldlm_request *dlm_req;
2544 LASSERT(h->mh_fmt != NULL);
2546 dlm_req = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
2547 if (dlm_req != NULL) {
2548 if (unlikely(dlm_req->lock_desc.l_resource.lr_type ==
2550 dlm_req->lock_desc.l_policy_data.\
2551 l_inodebits.bits == 0)) {
2553 * Lock without inodebits makes no sense and
2554 * will oops later in ldlm. If client miss to
2555 * set such bits, do not trigger ASSERTION.
2557 * For liblustre flock case, it maybe zero.
2561 if (info->mti_mdt->mdt_opts.mo_compat_resname)
2562 rc = mdt_lock_resname_compat(
2565 info->mti_dlm_req = dlm_req;
2572 /* capability setting changed via /proc, needs reinitialize ctxt */
2573 if (info->mti_mdt && info->mti_mdt->mdt_capa_conf) {
2574 mdt_init_capa_ctxt(info->mti_env, info->mti_mdt);
2575 info->mti_mdt->mdt_capa_conf = 0;
2578 if (likely(rc == 0)) {
2580 * Process request, there can be two types of rc:
2581 * 1) errors with msg unpack/pack, other failures outside the
2582 * operation itself. This is counted as serious errors;
2583 * 2) errors during fs operation, should be placed in rq_status
2586 rc = h->mh_act(info);
2588 !req->rq_no_reply && req->rq_reply_state == NULL) {
2589 DEBUG_REQ(D_ERROR, req, "MDT \"handler\" %s did not "
2590 "pack reply and returned 0 error\n",
2594 serious = is_serious(rc);
2595 rc = clear_serious(rc);
2599 req->rq_status = rc;
2602 * ELDLM_* codes which > 0 should be in rq_status only as well as
2603 * all non-serious errors.
2605 if (rc > 0 || !serious)
2608 LASSERT(current->journal_info == NULL);
2610 if (rc == 0 && (flags & HABEO_CLAVIS) &&
2611 info->mti_mdt->mdt_opts.mo_compat_resname) {
2612 struct ldlm_reply *dlmrep;
2614 dlmrep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
2616 rc = mdt_lock_reply_compat(info->mti_mdt, dlmrep);
2619 /* If we're DISCONNECTing, the mdt_export_data is already freed */
2620 if (likely(rc == 0 && req->rq_export && h->mh_opc != MDS_DISCONNECT))
2621 target_committed_to_req(req);
2623 if (unlikely(req_is_replay(req) &&
2624 lustre_msg_get_transno(req->rq_reqmsg) == 0)) {
2625 DEBUG_REQ(D_ERROR, req, "transno is 0 during REPLAY");
2629 target_send_reply(req, rc, info->mti_fail_id);
2633 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
2635 lh->mlh_type = MDT_NUL_LOCK;
2636 lh->mlh_reg_lh.cookie = 0ull;
2637 lh->mlh_reg_mode = LCK_MINMODE;
2638 lh->mlh_pdo_lh.cookie = 0ull;
2639 lh->mlh_pdo_mode = LCK_MINMODE;
2642 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
2644 LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2645 LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2649 * Initialize fields of struct mdt_thread_info. Other fields are left in
2650 * uninitialized state, because it's too expensive to zero out whole
2651 * mdt_thread_info (> 1K) on each request arrival.
2653 static void mdt_thread_info_init(struct ptlrpc_request *req,
2654 struct mdt_thread_info *info)
2657 struct md_capainfo *ci;
2659 req_capsule_init(&req->rq_pill, req, RCL_SERVER);
2660 info->mti_pill = &req->rq_pill;
2663 for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2664 mdt_lock_handle_init(&info->mti_lh[i]);
2666 /* mdt device: it can be NULL while CONNECT */
2667 if (req->rq_export) {
2668 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
2669 info->mti_exp = req->rq_export;
2671 info->mti_mdt = NULL;
2672 info->mti_env = req->rq_svc_thread->t_env;
2673 ci = md_capainfo(info->mti_env);
2674 memset(ci, 0, sizeof *ci);
2675 if (req->rq_export) {
2676 if (exp_connect_rmtclient(req->rq_export))
2677 ci->mc_auth = LC_ID_CONVERT;
2678 else if (req->rq_export->exp_connect_flags &
2679 OBD_CONNECT_MDS_CAPA)
2680 ci->mc_auth = LC_ID_PLAIN;
2682 ci->mc_auth = LC_ID_NONE;
2685 info->mti_fail_id = OBD_FAIL_MDS_ALL_REPLY_NET;
2686 info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
2687 info->mti_mos = NULL;
2689 memset(&info->mti_attr, 0, sizeof(info->mti_attr));
2690 info->mti_body = NULL;
2691 info->mti_object = NULL;
2692 info->mti_dlm_req = NULL;
2693 info->mti_has_trans = 0;
2694 info->mti_no_need_trans = 0;
2695 info->mti_cross_ref = 0;
2696 info->mti_opdata = 0;
2698 /* To not check for split by default. */
2699 info->mti_spec.sp_ck_split = 0;
2700 info->mti_spec.no_create = 0;
2703 static void mdt_thread_info_fini(struct mdt_thread_info *info)
2707 req_capsule_fini(info->mti_pill);
2708 if (info->mti_object != NULL) {
2710 * freeing an object may lead to OSD level transaction, do not
2711 * let it mess with MDT. bz19385.
2713 info->mti_no_need_trans = 1;
2714 mdt_object_put(info->mti_env, info->mti_object);
2715 info->mti_object = NULL;
2717 for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2718 mdt_lock_handle_fini(&info->mti_lh[i]);
2719 info->mti_env = NULL;
2722 static int mdt_filter_recovery_request(struct ptlrpc_request *req,
2723 struct obd_device *obd, int *process)
2725 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2726 case MDS_CONNECT: /* This will never get here, but for completeness. */
2727 case OST_CONNECT: /* This will never get here, but for completeness. */
2728 case MDS_DISCONNECT:
2729 case OST_DISCONNECT:
2734 case MDS_DONE_WRITING:
2735 case MDS_SYNC: /* used in unmounting */
2741 *process = target_queue_recovery_request(req, obd);
2745 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
2752 * Handle recovery. Return:
2753 * +1: continue request processing;
2754 * -ve: abort immediately with the given error code;
2755 * 0: send reply with error code in req->rq_status;
2757 static int mdt_recovery(struct mdt_thread_info *info)
2759 struct ptlrpc_request *req = mdt_info_req(info);
2760 struct obd_device *obd;
2764 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2767 case SEC_CTX_INIT_CONT:
2773 rc = mdt_handle_idmap(info);
2782 if (unlikely(!class_connected_export(req->rq_export))) {
2783 CERROR("operation %d on unconnected MDS from %s\n",
2784 lustre_msg_get_opc(req->rq_reqmsg),
2785 libcfs_id2str(req->rq_peer));
2786 /* FIXME: For CMD cleanup, when mds_B stop, the req from
2787 * mds_A will get -ENOTCONN(especially for ping req),
2788 * which will cause that mds_A deactive timeout, then when
2789 * mds_A cleanup, the cleanup process will be suspended since
2790 * deactive timeout is not zero.
2792 req->rq_status = -ENOTCONN;
2793 target_send_reply(req, -ENOTCONN, info->mti_fail_id);
2797 /* sanity check: if the xid matches, the request must be marked as a
2798 * resent or replayed */
2799 if (req_xid_is_last(req)) {
2800 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
2801 (MSG_RESENT | MSG_REPLAY))) {
2802 DEBUG_REQ(D_WARNING, req, "rq_xid "LPU64" matches last_xid, "
2803 "expected REPLAY or RESENT flag (%x)", req->rq_xid,
2804 lustre_msg_get_flags(req->rq_reqmsg));
2806 req->rq_status = -ENOTCONN;
2811 /* else: note the opposite is not always true; a RESENT req after a
2812 * failover will usually not match the last_xid, since it was likely
2813 * never committed. A REPLAYed request will almost never match the
2814 * last xid, however it could for a committed, but still retained,
2817 obd = req->rq_export->exp_obd;
2819 /* Check for aborted recovery... */
2820 if (unlikely(obd->obd_recovering)) {
2823 DEBUG_REQ(D_INFO, req, "Got new replay");
2824 rc = mdt_filter_recovery_request(req, obd, &should_process);
2825 if (rc != 0 || !should_process)
2827 else if (should_process < 0) {
2828 req->rq_status = should_process;
2829 rc = ptlrpc_error(req);
2836 static int mdt_msg_check_version(struct lustre_msg *msg)
2840 switch (lustre_msg_get_opc(msg)) {
2842 case MDS_DISCONNECT:
2845 case SEC_CTX_INIT_CONT:
2847 rc = lustre_msg_check_version(msg, LUSTRE_OBD_VERSION);
2849 CERROR("bad opc %u version %08x, expecting %08x\n",
2850 lustre_msg_get_opc(msg),
2851 lustre_msg_get_version(msg),
2852 LUSTRE_OBD_VERSION);
2856 case MDS_GETATTR_NAME:
2863 case MDS_DONE_WRITING:
2870 case MDS_QUOTACHECK:
2876 rc = lustre_msg_check_version(msg, LUSTRE_MDS_VERSION);
2878 CERROR("bad opc %u version %08x, expecting %08x\n",
2879 lustre_msg_get_opc(msg),
2880 lustre_msg_get_version(msg),
2881 LUSTRE_MDS_VERSION);
2885 case LDLM_BL_CALLBACK:
2886 case LDLM_CP_CALLBACK:
2887 rc = lustre_msg_check_version(msg, LUSTRE_DLM_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_DLM_VERSION);
2894 case OBD_LOG_CANCEL:
2895 case LLOG_ORIGIN_HANDLE_CREATE:
2896 case LLOG_ORIGIN_HANDLE_NEXT_BLOCK:
2897 case LLOG_ORIGIN_HANDLE_READ_HEADER:
2898 case LLOG_ORIGIN_HANDLE_CLOSE:
2899 case LLOG_ORIGIN_HANDLE_DESTROY:
2900 case LLOG_ORIGIN_HANDLE_PREV_BLOCK:
2902 rc = lustre_msg_check_version(msg, LUSTRE_LOG_VERSION);
2904 CERROR("bad opc %u version %08x, expecting %08x\n",
2905 lustre_msg_get_opc(msg),
2906 lustre_msg_get_version(msg),
2907 LUSTRE_LOG_VERSION);
2910 CERROR("MDS unknown opcode %d\n", lustre_msg_get_opc(msg));
2916 static int mdt_handle0(struct ptlrpc_request *req,
2917 struct mdt_thread_info *info,
2918 struct mdt_opc_slice *supported)
2920 struct mdt_handler *h;
2921 struct lustre_msg *msg;
2926 if (OBD_FAIL_CHECK_ORSET(OBD_FAIL_MDS_ALL_REQUEST_NET, OBD_FAIL_ONCE))
2929 LASSERT(current->journal_info == NULL);
2931 msg = req->rq_reqmsg;
2932 rc = mdt_msg_check_version(msg);
2933 if (likely(rc == 0)) {
2934 rc = mdt_recovery(info);
2935 if (likely(rc == +1)) {
2936 h = mdt_handler_find(lustre_msg_get_opc(msg),
2938 if (likely(h != NULL)) {
2939 rc = mdt_req_handle(info, h, req);
2941 CERROR("The unsupported opc: 0x%x\n",
2942 lustre_msg_get_opc(msg) );
2943 req->rq_status = -ENOTSUPP;
2944 rc = ptlrpc_error(req);
2949 CERROR(LUSTRE_MDT_NAME" drops mal-formed request\n");
2954 * MDT handler function called by ptlrpc service thread when request comes.
2956 * XXX common "target" functionality should be factored into separate module
2957 * shared by mdt, ost and stand-alone services like fld.
2959 static int mdt_handle_common(struct ptlrpc_request *req,
2960 struct mdt_opc_slice *supported)
2963 struct mdt_thread_info *info;
2967 env = req->rq_svc_thread->t_env;
2968 LASSERT(env != NULL);
2969 LASSERT(env->le_ses != NULL);
2970 LASSERT(env->le_ctx.lc_thread == req->rq_svc_thread);
2971 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
2972 LASSERT(info != NULL);
2974 mdt_thread_info_init(req, info);
2976 rc = mdt_handle0(req, info, supported);
2978 mdt_thread_info_fini(info);
2983 * This is called from recovery code as handler of _all_ RPC types, FLD and SEQ
2986 int mdt_recovery_handle(struct ptlrpc_request *req)
2991 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2993 rc = mdt_handle_common(req, mdt_fld_handlers);
2996 rc = mdt_handle_common(req, mdt_seq_handlers);
2999 rc = mdt_handle_common(req, mdt_regular_handlers);
3006 static int mdt_regular_handle(struct ptlrpc_request *req)
3008 return mdt_handle_common(req, mdt_regular_handlers);
3011 static int mdt_readpage_handle(struct ptlrpc_request *req)
3013 return mdt_handle_common(req, mdt_readpage_handlers);
3016 static int mdt_xmds_handle(struct ptlrpc_request *req)
3018 return mdt_handle_common(req, mdt_xmds_handlers);
3021 static int mdt_mdsc_handle(struct ptlrpc_request *req)
3023 return mdt_handle_common(req, mdt_seq_handlers);
3026 static int mdt_mdss_handle(struct ptlrpc_request *req)
3028 return mdt_handle_common(req, mdt_seq_handlers);
3031 static int mdt_dtss_handle(struct ptlrpc_request *req)
3033 return mdt_handle_common(req, mdt_seq_handlers);
3036 static int mdt_fld_handle(struct ptlrpc_request *req)
3038 return mdt_handle_common(req, mdt_fld_handlers);
3054 static int mdt_intent_getattr(enum mdt_it_code opcode,
3055 struct mdt_thread_info *info,
3056 struct ldlm_lock **,
3058 static int mdt_intent_reint(enum mdt_it_code opcode,
3059 struct mdt_thread_info *info,
3060 struct ldlm_lock **,
3063 static struct mdt_it_flavor {
3064 const struct req_format *it_fmt;
3066 int (*it_act)(enum mdt_it_code ,
3067 struct mdt_thread_info *,
3068 struct ldlm_lock **,
3071 } mdt_it_flavor[] = {
3073 .it_fmt = &RQF_LDLM_INTENT,
3074 /*.it_flags = HABEO_REFERO,*/
3076 .it_act = mdt_intent_reint,
3077 .it_reint = REINT_OPEN
3080 .it_fmt = &RQF_LDLM_INTENT,
3081 .it_flags = MUTABOR,
3082 .it_act = mdt_intent_reint,
3083 .it_reint = REINT_OPEN
3086 .it_fmt = &RQF_LDLM_INTENT,
3087 .it_flags = MUTABOR,
3088 .it_act = mdt_intent_reint,
3089 .it_reint = REINT_CREATE
3091 [MDT_IT_GETATTR] = {
3092 .it_fmt = &RQF_LDLM_INTENT_GETATTR,
3093 .it_flags = HABEO_REFERO,
3094 .it_act = mdt_intent_getattr
3096 [MDT_IT_READDIR] = {
3102 .it_fmt = &RQF_LDLM_INTENT_GETATTR,
3103 .it_flags = HABEO_REFERO,
3104 .it_act = mdt_intent_getattr
3107 .it_fmt = &RQF_LDLM_INTENT_UNLINK,
3108 .it_flags = MUTABOR,
3110 .it_reint = REINT_UNLINK
3114 .it_flags = MUTABOR,
3117 [MDT_IT_GETXATTR] = {
3124 int mdt_intent_lock_replace(struct mdt_thread_info *info,
3125 struct ldlm_lock **lockp,
3126 struct ldlm_lock *new_lock,
3127 struct mdt_lock_handle *lh,
3130 struct ptlrpc_request *req = mdt_info_req(info);
3131 struct ldlm_lock *lock = *lockp;
3134 * Get new lock only for cases when possible resent did not find any
3137 if (new_lock == NULL)
3138 new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
3140 if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
3141 lh->mlh_reg_lh.cookie = 0;
3145 LASSERTF(new_lock != NULL,
3146 "lockh "LPX64"\n", lh->mlh_reg_lh.cookie);
3149 * If we've already given this lock to a client once, then we should
3150 * have no readers or writers. Otherwise, we should have one reader
3151 * _or_ writer ref (which will be zeroed below) before returning the
3154 if (new_lock->l_export == req->rq_export) {
3155 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
3157 LASSERT(new_lock->l_export == NULL);
3158 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
3163 if (new_lock->l_export == req->rq_export) {
3165 * Already gave this to the client, which means that we
3166 * reconstructed a reply.
3168 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
3170 lh->mlh_reg_lh.cookie = 0;
3171 RETURN(ELDLM_LOCK_REPLACED);
3175 * Fixup the lock to be given to the client.
3177 lock_res_and_lock(new_lock);
3178 /* Zero new_lock->l_readers and new_lock->l_writers without triggering
3179 * possible blocking AST. */
3180 while (new_lock->l_readers > 0) {
3181 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
3182 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3183 new_lock->l_readers--;
3185 while (new_lock->l_writers > 0) {
3186 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
3187 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3188 new_lock->l_writers--;
3191 new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
3192 new_lock->l_blocking_ast = lock->l_blocking_ast;
3193 new_lock->l_completion_ast = lock->l_completion_ast;
3194 new_lock->l_remote_handle = lock->l_remote_handle;
3195 new_lock->l_flags &= ~LDLM_FL_LOCAL;
3197 unlock_res_and_lock(new_lock);
3199 cfs_hash_add(new_lock->l_export->exp_lock_hash,
3200 &new_lock->l_remote_handle,
3201 &new_lock->l_exp_hash);
3203 LDLM_LOCK_RELEASE(new_lock);
3204 lh->mlh_reg_lh.cookie = 0;
3206 RETURN(ELDLM_LOCK_REPLACED);
3209 static void mdt_intent_fixup_resent(struct mdt_thread_info *info,
3210 struct ldlm_lock *new_lock,
3211 struct ldlm_lock **old_lock,
3212 struct mdt_lock_handle *lh)
3214 struct ptlrpc_request *req = mdt_info_req(info);
3215 struct obd_export *exp = req->rq_export;
3216 struct lustre_handle remote_hdl;
3217 struct ldlm_request *dlmreq;
3218 struct ldlm_lock *lock;
3220 if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
3223 dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
3224 remote_hdl = dlmreq->lock_handle[0];
3226 lock = cfs_hash_lookup(exp->exp_lock_hash, &remote_hdl);
3228 if (lock != new_lock) {
3229 lh->mlh_reg_lh.cookie = lock->l_handle.h_cookie;
3230 lh->mlh_reg_mode = lock->l_granted_mode;
3232 LDLM_DEBUG(lock, "Restoring lock cookie");
3233 DEBUG_REQ(D_DLMTRACE, req,
3234 "restoring lock cookie "LPX64,
3235 lh->mlh_reg_lh.cookie);
3237 *old_lock = LDLM_LOCK_GET(lock);
3238 cfs_hash_put(exp->exp_lock_hash, &lock->l_exp_hash);
3242 cfs_hash_put(exp->exp_lock_hash, &lock->l_exp_hash);
3246 * If the xid matches, then we know this is a resent request, and allow
3247 * it. (It's probably an OPEN, for which we don't send a lock.
3249 if (req_xid_is_last(req))
3253 * This remote handle isn't enqueued, so we never received or processed
3254 * this request. Clear MSG_RESENT, because it can be handled like any
3255 * normal request now.
3257 lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
3259 DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle "LPX64,
3263 static int mdt_intent_getattr(enum mdt_it_code opcode,
3264 struct mdt_thread_info *info,
3265 struct ldlm_lock **lockp,
3268 struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3269 struct ldlm_lock *new_lock = NULL;
3271 struct ldlm_reply *ldlm_rep;
3272 struct ptlrpc_request *req;
3273 struct mdt_body *reqbody;
3274 struct mdt_body *repbody;
3278 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
3281 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
3284 info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT);
3285 info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
3286 repbody->eadatasize = 0;
3287 repbody->aclsize = 0;
3291 child_bits = MDS_INODELOCK_LOOKUP;
3293 case MDT_IT_GETATTR:
3294 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE;
3297 CERROR("Unhandled till now");
3298 GOTO(out_shrink, rc = -EINVAL);
3301 rc = mdt_init_ucred(info, reqbody);
3303 GOTO(out_shrink, rc);
3305 req = info->mti_pill->rc_req;
3306 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3307 mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
3309 /* Get lock from request for possible resent case. */
3310 mdt_intent_fixup_resent(info, *lockp, &new_lock, lhc);
3312 ldlm_rep->lock_policy_res2 =
3313 mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
3315 if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
3316 ldlm_rep->lock_policy_res2 = 0;
3317 if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
3318 ldlm_rep->lock_policy_res2) {
3319 lhc->mlh_reg_lh.cookie = 0ull;
3320 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
3323 rc = mdt_intent_lock_replace(info, lockp, new_lock, lhc, flags);
3326 mdt_exit_ucred(info);
3328 mdt_shrink_reply(info);
3332 static int mdt_intent_reint(enum mdt_it_code opcode,
3333 struct mdt_thread_info *info,
3334 struct ldlm_lock **lockp,
3337 struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3338 struct ldlm_reply *rep = NULL;
3342 static const struct req_format *intent_fmts[REINT_MAX] = {
3343 [REINT_CREATE] = &RQF_LDLM_INTENT_CREATE,
3344 [REINT_OPEN] = &RQF_LDLM_INTENT_OPEN
3349 opc = mdt_reint_opcode(info, intent_fmts);
3353 if (mdt_it_flavor[opcode].it_reint != opc) {
3354 CERROR("Reint code %ld doesn't match intent: %d\n",
3356 RETURN(err_serious(-EPROTO));
3359 /* Get lock from request for possible resent case. */
3360 mdt_intent_fixup_resent(info, *lockp, NULL, lhc);
3362 rc = mdt_reint_internal(info, lhc, opc);
3364 /* Check whether the reply has been packed successfully. */
3365 if (mdt_info_req(info)->rq_repmsg != NULL)
3366 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3368 RETURN(err_serious(-EFAULT));
3370 /* MDC expects this in any case */
3372 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
3374 /* Cross-ref case, the lock should be returned to the client */
3375 if (rc == -EREMOTE) {
3376 LASSERT(lustre_handle_is_used(&lhc->mlh_reg_lh));
3377 rep->lock_policy_res2 = 0;
3378 rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
3381 rep->lock_policy_res2 = clear_serious(rc);
3383 if (rep->lock_policy_res2 == -ENOENT &&
3384 mdt_get_disposition(rep, DISP_LOOKUP_NEG))
3385 rep->lock_policy_res2 = 0;
3387 if (rc == -ENOTCONN || rc == -ENODEV ||
3388 rc == -EOVERFLOW) { /**< if VBR failure then return error */
3390 * If it is the disconnect error (ENODEV & ENOCONN), the error
3391 * will be returned by rq_status, and client at ptlrpc layer
3392 * will detect this, then disconnect, reconnect the import
3393 * immediately, instead of impacting the following the rpc.
3395 lhc->mlh_reg_lh.cookie = 0ull;
3399 * For other cases, the error will be returned by intent.
3400 * and client will retrieve the result from intent.
3403 * FIXME: when open lock is finished, that should be
3406 if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3407 LASSERTF(rc == 0, "Error occurred but lock handle "
3408 "is still in use\n");
3409 rep->lock_policy_res2 = 0;
3410 rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
3413 lhc->mlh_reg_lh.cookie = 0ull;
3414 RETURN(ELDLM_LOCK_ABORTED);
3419 static int mdt_intent_code(long itcode)
3427 case IT_OPEN|IT_CREAT:
3434 rc = MDT_IT_READDIR;
3437 rc = MDT_IT_GETATTR;
3449 rc = MDT_IT_GETXATTR;
3452 CERROR("Unknown intent opcode: %ld\n", itcode);
3459 static int mdt_intent_opc(long itopc, struct mdt_thread_info *info,
3460 struct ldlm_lock **lockp, int flags)
3462 struct req_capsule *pill;
3463 struct mdt_it_flavor *flv;
3468 opc = mdt_intent_code(itopc);
3472 pill = info->mti_pill;
3473 flv = &mdt_it_flavor[opc];
3475 if (flv->it_fmt != NULL)
3476 req_capsule_extend(pill, flv->it_fmt);
3478 rc = mdt_unpack_req_pack_rep(info, flv->it_flags);
3480 struct ptlrpc_request *req = mdt_info_req(info);
3481 if (flv->it_flags & MUTABOR &&
3482 req->rq_export->exp_connect_flags & OBD_CONNECT_RDONLY)
3485 if (rc == 0 && flv->it_act != NULL) {
3486 /* execute policy */
3487 rc = flv->it_act(opc, info, lockp, flags);
3494 static int mdt_intent_policy(struct ldlm_namespace *ns,
3495 struct ldlm_lock **lockp, void *req_cookie,
3496 ldlm_mode_t mode, int flags, void *data)
3498 struct mdt_thread_info *info;
3499 struct ptlrpc_request *req = req_cookie;
3500 struct ldlm_intent *it;
3501 struct req_capsule *pill;
3506 LASSERT(req != NULL);
3508 info = lu_context_key_get(&req->rq_svc_thread->t_env->le_ctx,
3510 LASSERT(info != NULL);
3511 pill = info->mti_pill;
3512 LASSERT(pill->rc_req == req);
3514 if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
3515 req_capsule_extend(pill, &RQF_LDLM_INTENT);
3516 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
3518 rc = mdt_intent_opc(it->opc, info, lockp, flags);
3522 /* Lock without inodebits makes no sense and will oops
3523 * later in ldlm. Let's check it now to see if we have
3524 * ibits corrupted somewhere in mdt_intent_opc().
3525 * The case for client miss to set ibits has been
3526 * processed by others. */
3527 LASSERT(ergo(info->mti_dlm_req->lock_desc.l_resource.\
3528 lr_type == LDLM_IBITS,
3529 info->mti_dlm_req->lock_desc.\
3530 l_policy_data.l_inodebits.bits != 0));
3532 rc = err_serious(-EFAULT);
3534 /* No intent was provided */
3535 LASSERT(pill->rc_fmt == &RQF_LDLM_ENQUEUE);
3536 rc = req_capsule_server_pack(pill);
3538 rc = err_serious(rc);
3543 static int mdt_seq_fini(const struct lu_env *env,
3544 struct mdt_device *m)
3546 struct md_site *ms = mdt_md_site(m);
3550 if (ms->ms_server_seq) {
3551 seq_server_fini(ms->ms_server_seq, env);
3552 OBD_FREE_PTR(ms->ms_server_seq);
3553 ms->ms_server_seq = NULL;
3556 if (ms->ms_control_seq) {
3557 seq_server_fini(ms->ms_control_seq, env);
3558 OBD_FREE_PTR(ms->ms_control_seq);
3559 ms->ms_control_seq = NULL;
3562 if (ms->ms_client_seq) {
3563 seq_client_fini(ms->ms_client_seq);
3564 OBD_FREE_PTR(ms->ms_client_seq);
3565 ms->ms_client_seq = NULL;
3572 static int mdt_seq_init(const struct lu_env *env,
3574 struct mdt_device *m)
3581 ms = mdt_md_site(m);
3584 * This is sequence-controller node. Init seq-controller server on local
3587 if (ms->ms_node_id == 0) {
3588 LASSERT(ms->ms_control_seq == NULL);
3590 OBD_ALLOC_PTR(ms->ms_control_seq);
3591 if (ms->ms_control_seq == NULL)
3594 rc = seq_server_init(ms->ms_control_seq,
3595 m->mdt_bottom, uuid,
3596 LUSTRE_SEQ_CONTROLLER,
3601 GOTO(out_seq_fini, rc);
3603 OBD_ALLOC_PTR(ms->ms_client_seq);
3604 if (ms->ms_client_seq == NULL)
3605 GOTO(out_seq_fini, rc = -ENOMEM);
3607 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3608 if (prefix == NULL) {
3609 OBD_FREE_PTR(ms->ms_client_seq);
3610 GOTO(out_seq_fini, rc = -ENOMEM);
3613 snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s",
3617 * Init seq-controller client after seq-controller server is
3618 * ready. Pass ms->ms_control_seq to it for direct talking.
3620 rc = seq_client_init(ms->ms_client_seq, NULL,
3621 LUSTRE_SEQ_METADATA, prefix,
3622 ms->ms_control_seq);
3623 OBD_FREE(prefix, MAX_OBD_NAME + 5);
3626 GOTO(out_seq_fini, rc);
3629 /* Init seq-server on local MDT */
3630 LASSERT(ms->ms_server_seq == NULL);
3632 OBD_ALLOC_PTR(ms->ms_server_seq);
3633 if (ms->ms_server_seq == NULL)
3634 GOTO(out_seq_fini, rc = -ENOMEM);
3636 rc = seq_server_init(ms->ms_server_seq,
3637 m->mdt_bottom, uuid,
3642 GOTO(out_seq_fini, rc = -ENOMEM);
3644 /* Assign seq-controller client to local seq-server. */
3645 if (ms->ms_node_id == 0) {
3646 LASSERT(ms->ms_client_seq != NULL);
3648 rc = seq_server_set_cli(ms->ms_server_seq,
3656 mdt_seq_fini(env, m);
3661 * Init client sequence manager which is used by local MDS to talk to sequence
3662 * controller on remote node.
3664 static int mdt_seq_init_cli(const struct lu_env *env,
3665 struct mdt_device *m,
3666 struct lustre_cfg *cfg)
3668 struct md_site *ms = mdt_md_site(m);
3669 struct obd_device *mdc;
3670 struct obd_uuid *uuidp, *mdcuuidp;
3671 char *uuid_str, *mdc_uuid_str;
3674 struct mdt_thread_info *info;
3675 char *p, *index_string = lustre_cfg_string(cfg, 2);
3678 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3679 uuidp = &info->mti_u.uuid[0];
3680 mdcuuidp = &info->mti_u.uuid[1];
3682 LASSERT(index_string);
3684 index = simple_strtol(index_string, &p, 10);
3686 CERROR("Invalid index in lustre_cgf, offset 2\n");
3690 /* check if this is adding the first MDC and controller is not yet
3692 if (index != 0 || ms->ms_client_seq)
3695 uuid_str = lustre_cfg_string(cfg, 1);
3696 mdc_uuid_str = lustre_cfg_string(cfg, 4);
3697 obd_str2uuid(uuidp, uuid_str);
3698 obd_str2uuid(mdcuuidp, mdc_uuid_str);
3700 mdc = class_find_client_obd(uuidp, LUSTRE_MDC_NAME, mdcuuidp);
3702 CERROR("can't find controller MDC by uuid %s\n",
3705 } else if (!mdc->obd_set_up) {
3706 CERROR("target %s not set up\n", mdc->obd_name);
3709 LASSERT(ms->ms_control_exp);
3710 OBD_ALLOC_PTR(ms->ms_client_seq);
3711 if (ms->ms_client_seq != NULL) {
3714 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3718 snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s",
3721 rc = seq_client_init(ms->ms_client_seq,
3723 LUSTRE_SEQ_METADATA,
3725 OBD_FREE(prefix, MAX_OBD_NAME + 5);
3732 LASSERT(ms->ms_server_seq != NULL);
3733 rc = seq_server_set_cli(ms->ms_server_seq, ms->ms_client_seq,
3740 static void mdt_seq_fini_cli(struct mdt_device *m)
3746 ms = mdt_md_site(m);
3749 if (ms->ms_server_seq)
3750 seq_server_set_cli(ms->ms_server_seq,
3753 if (ms->ms_control_exp) {
3754 class_export_put(ms->ms_control_exp);
3755 ms->ms_control_exp = NULL;
3764 static int mdt_fld_fini(const struct lu_env *env,
3765 struct mdt_device *m)
3767 struct md_site *ms = mdt_md_site(m);
3770 if (ms && ms->ms_server_fld) {
3771 fld_server_fini(ms->ms_server_fld, env);
3772 OBD_FREE_PTR(ms->ms_server_fld);
3773 ms->ms_server_fld = NULL;
3779 static int mdt_fld_init(const struct lu_env *env,
3781 struct mdt_device *m)
3787 ms = mdt_md_site(m);
3789 OBD_ALLOC_PTR(ms->ms_server_fld);
3790 if (ms->ms_server_fld == NULL)
3791 RETURN(rc = -ENOMEM);
3793 rc = fld_server_init(ms->ms_server_fld,
3794 m->mdt_bottom, uuid,
3795 env, ms->ms_node_id);
3797 OBD_FREE_PTR(ms->ms_server_fld);
3798 ms->ms_server_fld = NULL;
3805 /* device init/fini methods */
3806 static void mdt_stop_ptlrpc_service(struct mdt_device *m)
3809 if (m->mdt_regular_service != NULL) {
3810 ptlrpc_unregister_service(m->mdt_regular_service);
3811 m->mdt_regular_service = NULL;
3813 if (m->mdt_readpage_service != NULL) {
3814 ptlrpc_unregister_service(m->mdt_readpage_service);
3815 m->mdt_readpage_service = NULL;
3817 if (m->mdt_xmds_service != NULL) {
3818 ptlrpc_unregister_service(m->mdt_xmds_service);
3819 m->mdt_xmds_service = NULL;
3821 if (m->mdt_setattr_service != NULL) {
3822 ptlrpc_unregister_service(m->mdt_setattr_service);
3823 m->mdt_setattr_service = NULL;
3825 if (m->mdt_mdsc_service != NULL) {
3826 ptlrpc_unregister_service(m->mdt_mdsc_service);
3827 m->mdt_mdsc_service = NULL;
3829 if (m->mdt_mdss_service != NULL) {
3830 ptlrpc_unregister_service(m->mdt_mdss_service);
3831 m->mdt_mdss_service = NULL;
3833 if (m->mdt_dtss_service != NULL) {
3834 ptlrpc_unregister_service(m->mdt_dtss_service);
3835 m->mdt_dtss_service = NULL;
3837 if (m->mdt_fld_service != NULL) {
3838 ptlrpc_unregister_service(m->mdt_fld_service);
3839 m->mdt_fld_service = NULL;
3844 static int mdt_start_ptlrpc_service(struct mdt_device *m)
3847 static struct ptlrpc_service_conf conf;
3848 cfs_proc_dir_entry_t *procfs_entry;
3851 procfs_entry = m->mdt_md_dev.md_lu_dev.ld_obd->obd_proc_entry;
3853 conf = (typeof(conf)) {
3854 .psc_nbufs = MDS_NBUFS,
3855 .psc_bufsize = MDS_BUFSIZE,
3856 .psc_max_req_size = MDS_MAXREQSIZE,
3857 .psc_max_reply_size = MDS_MAXREPSIZE,
3858 .psc_req_portal = MDS_REQUEST_PORTAL,
3859 .psc_rep_portal = MDC_REPLY_PORTAL,
3860 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3862 * We'd like to have a mechanism to set this on a per-device
3863 * basis, but alas...
3865 .psc_min_threads = mdt_min_threads,
3866 .psc_max_threads = mdt_max_threads,
3867 .psc_ctx_tags = LCT_MD_THREAD
3870 m->mdt_ldlm_client = &m->mdt_md_dev.md_lu_dev.ld_obd->obd_ldlm_client;
3871 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
3872 "mdt_ldlm_client", m->mdt_ldlm_client);
3874 m->mdt_regular_service =
3875 ptlrpc_init_svc_conf(&conf, mdt_regular_handle, LUSTRE_MDT_NAME,
3876 procfs_entry, target_print_req,
3878 if (m->mdt_regular_service == NULL)
3881 rc = ptlrpc_start_threads(m->mdt_regular_service);
3883 GOTO(err_mdt_svc, rc);
3886 * readpage service configuration. Parameters have to be adjusted,
3889 conf = (typeof(conf)) {
3890 .psc_nbufs = MDS_NBUFS,
3891 .psc_bufsize = MDS_BUFSIZE,
3892 .psc_max_req_size = MDS_MAXREQSIZE,
3893 .psc_max_reply_size = MDS_MAXREPSIZE,
3894 .psc_req_portal = MDS_READPAGE_PORTAL,
3895 .psc_rep_portal = MDC_REPLY_PORTAL,
3896 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3897 .psc_min_threads = mdt_min_threads,
3898 .psc_max_threads = mdt_max_threads,
3899 .psc_ctx_tags = LCT_MD_THREAD
3901 m->mdt_readpage_service =
3902 ptlrpc_init_svc_conf(&conf, mdt_readpage_handle,
3903 LUSTRE_MDT_NAME "_readpage",
3904 procfs_entry, target_print_req,"mdt_rdpg");
3906 if (m->mdt_readpage_service == NULL) {
3907 CERROR("failed to start readpage service\n");
3908 GOTO(err_mdt_svc, rc = -ENOMEM);
3911 rc = ptlrpc_start_threads(m->mdt_readpage_service);
3914 * setattr service configuration.
3916 * XXX To keep the compatibility with old client(< 2.2), we need to
3917 * preserve this portal for a certain time, it should be removed
3918 * eventually. LU-617.
3920 conf = (typeof(conf)) {
3921 .psc_nbufs = MDS_NBUFS,
3922 .psc_bufsize = MDS_BUFSIZE,
3923 .psc_max_req_size = MDS_MAXREQSIZE,
3924 .psc_max_reply_size = MDS_MAXREPSIZE,
3925 .psc_req_portal = MDS_SETATTR_PORTAL,
3926 .psc_rep_portal = MDC_REPLY_PORTAL,
3927 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3928 .psc_min_threads = mdt_min_threads,
3929 .psc_max_threads = mdt_max_threads,
3930 .psc_ctx_tags = LCT_MD_THREAD
3933 m->mdt_setattr_service =
3934 ptlrpc_init_svc_conf(&conf, mdt_regular_handle,
3935 LUSTRE_MDT_NAME "_setattr",
3936 procfs_entry, target_print_req,"mdt_attr");
3938 if (!m->mdt_setattr_service) {
3939 CERROR("failed to start setattr service\n");
3940 GOTO(err_mdt_svc, rc = -ENOMEM);
3943 rc = ptlrpc_start_threads(m->mdt_setattr_service);
3945 GOTO(err_mdt_svc, rc);
3948 * sequence controller service configuration
3950 conf = (typeof(conf)) {
3951 .psc_nbufs = MDS_NBUFS,
3952 .psc_bufsize = MDS_BUFSIZE,
3953 .psc_max_req_size = SEQ_MAXREQSIZE,
3954 .psc_max_reply_size = SEQ_MAXREPSIZE,
3955 .psc_req_portal = SEQ_CONTROLLER_PORTAL,
3956 .psc_rep_portal = MDC_REPLY_PORTAL,
3957 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3958 .psc_min_threads = mdt_min_threads,
3959 .psc_max_threads = mdt_max_threads,
3960 .psc_ctx_tags = LCT_MD_THREAD|LCT_DT_THREAD
3963 m->mdt_mdsc_service =
3964 ptlrpc_init_svc_conf(&conf, mdt_mdsc_handle,
3965 LUSTRE_MDT_NAME"_mdsc",
3966 procfs_entry, target_print_req,"mdt_mdsc");
3967 if (!m->mdt_mdsc_service) {
3968 CERROR("failed to start seq controller service\n");
3969 GOTO(err_mdt_svc, rc = -ENOMEM);
3972 rc = ptlrpc_start_threads(m->mdt_mdsc_service);
3974 GOTO(err_mdt_svc, rc);
3977 * metadata sequence server service configuration
3979 conf = (typeof(conf)) {
3980 .psc_nbufs = MDS_NBUFS,
3981 .psc_bufsize = MDS_BUFSIZE,
3982 .psc_max_req_size = SEQ_MAXREQSIZE,
3983 .psc_max_reply_size = SEQ_MAXREPSIZE,
3984 .psc_req_portal = SEQ_METADATA_PORTAL,
3985 .psc_rep_portal = MDC_REPLY_PORTAL,
3986 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
3987 .psc_min_threads = mdt_min_threads,
3988 .psc_max_threads = mdt_max_threads,
3989 .psc_ctx_tags = LCT_MD_THREAD|LCT_DT_THREAD
3992 m->mdt_mdss_service =
3993 ptlrpc_init_svc_conf(&conf, mdt_mdss_handle,
3994 LUSTRE_MDT_NAME"_mdss",
3995 procfs_entry, target_print_req,"mdt_mdss");
3996 if (!m->mdt_mdss_service) {
3997 CERROR("failed to start metadata seq server service\n");
3998 GOTO(err_mdt_svc, rc = -ENOMEM);
4001 rc = ptlrpc_start_threads(m->mdt_mdss_service);
4003 GOTO(err_mdt_svc, rc);
4007 * Data sequence server service configuration. We want to have really
4008 * cluster-wide sequences space. This is why we start only one sequence
4009 * controller which manages space.
4011 conf = (typeof(conf)) {
4012 .psc_nbufs = MDS_NBUFS,
4013 .psc_bufsize = MDS_BUFSIZE,
4014 .psc_max_req_size = SEQ_MAXREQSIZE,
4015 .psc_max_reply_size = SEQ_MAXREPSIZE,
4016 .psc_req_portal = SEQ_DATA_PORTAL,
4017 .psc_rep_portal = OSC_REPLY_PORTAL,
4018 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
4019 .psc_min_threads = mdt_min_threads,
4020 .psc_max_threads = mdt_max_threads,
4021 .psc_ctx_tags = LCT_MD_THREAD|LCT_DT_THREAD
4024 m->mdt_dtss_service =
4025 ptlrpc_init_svc_conf(&conf, mdt_dtss_handle,
4026 LUSTRE_MDT_NAME"_dtss",
4027 procfs_entry, target_print_req,"mdt_dtss");
4028 if (!m->mdt_dtss_service) {
4029 CERROR("failed to start data seq server service\n");
4030 GOTO(err_mdt_svc, rc = -ENOMEM);
4033 rc = ptlrpc_start_threads(m->mdt_dtss_service);
4035 GOTO(err_mdt_svc, rc);
4037 /* FLD service start */
4038 conf = (typeof(conf)) {
4039 .psc_nbufs = MDS_NBUFS,
4040 .psc_bufsize = MDS_BUFSIZE,
4041 .psc_max_req_size = FLD_MAXREQSIZE,
4042 .psc_max_reply_size = FLD_MAXREPSIZE,
4043 .psc_req_portal = FLD_REQUEST_PORTAL,
4044 .psc_rep_portal = MDC_REPLY_PORTAL,
4045 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
4046 .psc_min_threads = mdt_min_threads,
4047 .psc_max_threads = mdt_max_threads,
4048 .psc_ctx_tags = LCT_DT_THREAD|LCT_MD_THREAD
4051 m->mdt_fld_service =
4052 ptlrpc_init_svc_conf(&conf, mdt_fld_handle,
4053 LUSTRE_MDT_NAME"_fld",
4054 procfs_entry, target_print_req, "mdt_fld");
4055 if (!m->mdt_fld_service) {
4056 CERROR("failed to start fld service\n");
4057 GOTO(err_mdt_svc, rc = -ENOMEM);
4060 rc = ptlrpc_start_threads(m->mdt_fld_service);
4062 GOTO(err_mdt_svc, rc);
4065 * mds-mds service configuration. Separate portal is used to allow
4066 * mds-mds requests be not blocked during recovery.
4068 conf = (typeof(conf)) {
4069 .psc_nbufs = MDS_NBUFS,
4070 .psc_bufsize = MDS_BUFSIZE,
4071 .psc_max_req_size = MDS_MAXREQSIZE,
4072 .psc_max_reply_size = MDS_MAXREPSIZE,
4073 .psc_req_portal = MDS_MDS_PORTAL,
4074 .psc_rep_portal = MDC_REPLY_PORTAL,
4075 .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR,
4076 .psc_min_threads = mdt_min_threads,
4077 .psc_max_threads = mdt_max_threads,
4078 .psc_ctx_tags = LCT_MD_THREAD
4080 m->mdt_xmds_service =
4081 ptlrpc_init_svc_conf(&conf, mdt_xmds_handle,
4082 LUSTRE_MDT_NAME "_mds",
4083 procfs_entry, target_print_req,"mdt_xmds");
4085 if (m->mdt_xmds_service == NULL) {
4086 CERROR("failed to start xmds service\n");
4087 GOTO(err_mdt_svc, rc = -ENOMEM);
4090 rc = ptlrpc_start_threads(m->mdt_xmds_service);
4092 GOTO(err_mdt_svc, rc);
4097 mdt_stop_ptlrpc_service(m);
4102 static void mdt_stack_fini(const struct lu_env *env,
4103 struct mdt_device *m, struct lu_device *top)
4105 struct obd_device *obd = mdt2obd_dev(m);
4106 struct lustre_cfg_bufs *bufs;
4107 struct lustre_cfg *lcfg;
4108 struct mdt_thread_info *info;
4112 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4113 LASSERT(info != NULL);
4115 bufs = &info->mti_u.bufs;
4116 /* process cleanup, pass mdt obd name to get obd umount flags */
4117 lustre_cfg_bufs_reset(bufs, obd->obd_name);
4122 lustre_cfg_bufs_set_string(bufs, 1, flags);
4123 lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
4125 CERROR("Cannot alloc lcfg!\n");
4130 top->ld_ops->ldo_process_config(env, top, lcfg);
4131 lustre_cfg_free(lcfg);
4133 lu_stack_fini(env, top);
4134 m->mdt_child = NULL;
4135 m->mdt_bottom = NULL;
4138 static struct lu_device *mdt_layer_setup(struct lu_env *env,
4139 const char *typename,
4140 struct lu_device *child,
4141 struct lustre_cfg *cfg)
4143 const char *dev = lustre_cfg_string(cfg, 0);
4144 struct obd_type *type;
4145 struct lu_device_type *ldt;
4146 struct lu_device *d;
4151 type = class_get_type(typename);
4153 CERROR("Unknown type: '%s'\n", typename);
4154 GOTO(out, rc = -ENODEV);
4157 rc = lu_env_refill((struct lu_env *)env);
4159 CERROR("Failure to refill session: '%d'\n", rc);
4165 CERROR("type: '%s'\n", typename);
4166 GOTO(out_type, rc = -EINVAL);
4169 ldt->ldt_obd_type = type;
4170 d = ldt->ldt_ops->ldto_device_alloc(env, ldt, cfg);
4172 CERROR("Cannot allocate device: '%s'\n", typename);
4173 GOTO(out_type, rc = -ENODEV);
4176 LASSERT(child->ld_site);
4177 d->ld_site = child->ld_site;
4180 rc = ldt->ldt_ops->ldto_device_init(env, d, dev, child);
4182 CERROR("can't init device '%s', rc %d\n", typename, rc);
4183 GOTO(out_alloc, rc);
4186 lu_ref_add(&d->ld_reference, "lu-stack", &lu_site_init);
4188 cfs_spin_lock(&d->ld_site->ls_ld_lock);
4189 cfs_list_add_tail(&d->ld_linkage, &d->ld_site->ls_ld_linkage);
4190 cfs_spin_unlock(&d->ld_site->ls_ld_lock);
4194 ldt->ldt_ops->ldto_device_free(env, d);
4197 class_put_type(type);
4202 static int mdt_stack_init(struct lu_env *env,
4203 struct mdt_device *m,
4204 struct lustre_cfg *cfg,
4205 struct lustre_mount_info *lmi)
4207 struct lu_device *d = &m->mdt_md_dev.md_lu_dev;
4208 struct lu_device *tmp;
4209 struct md_device *md;
4210 struct lu_device *child_lu_dev;
4214 /* init the stack */
4215 tmp = mdt_layer_setup(env, LUSTRE_OSD_NAME, d, cfg);
4217 RETURN(PTR_ERR(tmp));
4219 m->mdt_bottom = lu2dt_dev(tmp);
4221 tmp = mdt_layer_setup(env, LUSTRE_MDD_NAME, d, cfg);
4223 GOTO(out, rc = PTR_ERR(tmp));
4228 tmp = mdt_layer_setup(env, LUSTRE_CMM_NAME, d, cfg);
4230 GOTO(out, rc = PTR_ERR(tmp));
4233 /*set mdd upcall device*/
4234 md_upcall_dev_set(md, lu2md_dev(d));
4237 /*set cmm upcall device*/
4238 md_upcall_dev_set(md, &m->mdt_md_dev);
4240 m->mdt_child = lu2md_dev(d);
4242 /* process setup config */
4243 tmp = &m->mdt_md_dev.md_lu_dev;
4244 rc = tmp->ld_ops->ldo_process_config(env, tmp, cfg);
4248 /* initialize local objects */
4249 child_lu_dev = &m->mdt_child->md_lu_dev;
4251 rc = child_lu_dev->ld_ops->ldo_prepare(env,
4252 &m->mdt_md_dev.md_lu_dev,
4255 /* fini from last known good lu_device */
4257 mdt_stack_fini(env, m, d);
4263 * setup CONFIG_ORIG context, used to access local config log.
4264 * this may need to be rewrite as part of llog rewrite for lu-api.
4266 static int mdt_obd_llog_setup(struct obd_device *obd,
4267 struct lustre_sb_info *lsi)
4271 LASSERT(obd->obd_fsops == NULL);
4273 obd->obd_fsops = fsfilt_get_ops(MT_STR(lsi->lsi_ldd));
4274 if (IS_ERR(obd->obd_fsops))
4275 return PTR_ERR(obd->obd_fsops);
4277 rc = fsfilt_setup(obd, lsi->lsi_srv_mnt->mnt_sb);
4279 fsfilt_put_ops(obd->obd_fsops);
4283 OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
4284 obd->obd_lvfs_ctxt.pwdmnt = lsi->lsi_srv_mnt;
4285 obd->obd_lvfs_ctxt.pwd = lsi->lsi_srv_mnt->mnt_root;
4286 obd->obd_lvfs_ctxt.fs = get_ds();
4288 rc = llog_setup(obd, &obd->obd_olg, LLOG_CONFIG_ORIG_CTXT, obd,
4289 0, NULL, &llog_lvfs_ops);
4291 CERROR("llog_setup() failed: %d\n", rc);
4292 fsfilt_put_ops(obd->obd_fsops);
4298 static void mdt_obd_llog_cleanup(struct obd_device *obd)
4300 struct llog_ctxt *ctxt;
4302 ctxt = llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT);
4306 if (obd->obd_fsops) {
4307 fsfilt_put_ops(obd->obd_fsops);
4308 obd->obd_fsops = NULL;
4312 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
4314 struct md_device *next = m->mdt_child;
4315 struct lu_device *d = &m->mdt_md_dev.md_lu_dev;
4316 struct lu_site *ls = d->ld_site;
4317 struct obd_device *obd = mdt2obd_dev(m);
4320 target_recovery_fini(obd);
4322 ping_evictor_stop();
4324 mdt_stop_ptlrpc_service(m);
4325 mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4326 mdt_obd_llog_cleanup(obd);
4327 obd_exports_barrier(obd);
4328 obd_zombie_barrier();
4332 #ifdef HAVE_QUOTA_SUPPORT
4333 next->md_ops->mdo_quota.mqo_cleanup(env, next);
4335 lut_fini(env, &m->mdt_lut);
4336 mdt_fs_cleanup(env, m);
4337 upcall_cache_cleanup(m->mdt_identity_cache);
4338 m->mdt_identity_cache = NULL;
4340 if (m->mdt_namespace != NULL) {
4341 ldlm_namespace_free(m->mdt_namespace, NULL,
4342 d->ld_obd->obd_force);
4343 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
4346 cfs_free_nidlist(&m->mdt_nosquash_nids);
4347 if (m->mdt_nosquash_str) {
4348 OBD_FREE(m->mdt_nosquash_str, m->mdt_nosquash_strlen);
4349 m->mdt_nosquash_str = NULL;
4350 m->mdt_nosquash_strlen = 0;
4353 mdt_seq_fini(env, m);
4354 mdt_seq_fini_cli(m);
4355 mdt_fld_fini(env, m);
4356 sptlrpc_rule_set_free(&m->mdt_sptlrpc_rset);
4358 next->md_ops->mdo_init_capa_ctxt(env, next, 0, 0, 0, NULL);
4359 cfs_timer_disarm(&m->mdt_ck_timer);
4360 mdt_ck_thread_stop(m);
4365 mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4368 struct md_site *mite;
4371 mite = lu_site2md(ls);
4375 LASSERT(cfs_atomic_read(&d->ld_ref) == 0);
4380 static int mdt_adapt_sptlrpc_conf(struct obd_device *obd, int initial)
4382 struct mdt_device *m = mdt_dev(obd->obd_lu_dev);
4383 struct sptlrpc_rule_set tmp_rset;
4386 sptlrpc_rule_set_init(&tmp_rset);
4387 rc = sptlrpc_conf_target_get_rules(obd, &tmp_rset, initial);
4389 CERROR("mdt %s: failed get sptlrpc rules: %d\n",
4394 sptlrpc_target_update_exp_flavor(obd, &tmp_rset);
4396 cfs_write_lock(&m->mdt_sptlrpc_lock);
4397 sptlrpc_rule_set_free(&m->mdt_sptlrpc_rset);
4398 m->mdt_sptlrpc_rset = tmp_rset;
4399 cfs_write_unlock(&m->mdt_sptlrpc_lock);
4404 int mdt_postrecov(const struct lu_env *, struct mdt_device *);
4406 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
4407 struct lu_device_type *ldt, struct lustre_cfg *cfg)
4409 struct mdt_thread_info *info;
4410 struct obd_device *obd;
4411 const char *dev = lustre_cfg_string(cfg, 0);
4412 const char *num = lustre_cfg_string(cfg, 2);
4413 struct lustre_mount_info *lmi = NULL;
4414 struct lustre_sb_info *lsi;
4415 struct lustre_disk_data *ldd;
4417 struct md_site *mite;
4418 const char *identity_upcall = "NONE";
4419 struct md_device *next;
4425 md_device_init(&m->mdt_md_dev, ldt);
4427 * Environment (env) might be missing mdt_thread_key values at that
4428 * point, if device is allocated when mdt_thread_key is in QUIESCENT
4431 * Usually device allocation path doesn't use module key values, but
4432 * mdt has to do a lot of work here, so allocate key value.
4434 rc = lu_env_refill((struct lu_env *)env);
4438 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4439 LASSERT(info != NULL);
4441 obd = class_name2obd(dev);
4442 LASSERT(obd != NULL);
4444 m->mdt_max_mdsize = MAX_MD_SIZE;
4445 m->mdt_max_cookiesize = sizeof(struct llog_cookie);
4446 m->mdt_som_conf = 0;
4448 m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
4449 lmi = server_get_mount_2(dev);
4451 CERROR("Cannot get mount info for %s!\n", dev);
4454 lsi = s2lsi(lmi->lmi_sb);
4455 /* CMD is supported only in IAM mode */
4458 node_id = simple_strtol(num, NULL, 10);
4459 if (!(ldd->ldd_flags & LDD_F_IAM_DIR) && node_id) {
4460 CERROR("CMD Operation not allowed in IOP mode\n");
4461 GOTO(err_lmi, rc = -EINVAL);
4464 obd->u.obt.obt_magic = OBT_MAGIC;
4467 cfs_rwlock_init(&m->mdt_sptlrpc_lock);
4468 sptlrpc_rule_set_init(&m->mdt_sptlrpc_rset);
4470 cfs_spin_lock_init(&m->mdt_ioepoch_lock);
4471 m->mdt_opts.mo_compat_resname = 0;
4472 m->mdt_opts.mo_mds_capa = 1;
4473 m->mdt_opts.mo_oss_capa = 1;
4474 m->mdt_capa_timeout = CAPA_TIMEOUT;
4475 m->mdt_capa_alg = CAPA_HMAC_ALG_SHA1;
4476 m->mdt_ck_timeout = CAPA_KEY_TIMEOUT;
4477 m->mdt_squash_uid = 0;
4478 m->mdt_squash_gid = 0;
4479 CFS_INIT_LIST_HEAD(&m->mdt_nosquash_nids);
4480 m->mdt_nosquash_str = NULL;
4481 m->mdt_nosquash_strlen = 0;
4482 cfs_init_rwsem(&m->mdt_squash_sem);
4484 OBD_ALLOC_PTR(mite);
4486 GOTO(err_lmi, rc = -ENOMEM);
4490 m->mdt_md_dev.md_lu_dev.ld_ops = &mdt_lu_ops;
4491 m->mdt_md_dev.md_lu_dev.ld_obd = obd;
4492 /* set this lu_device to obd, because error handling need it */
4493 obd->obd_lu_dev = &m->mdt_md_dev.md_lu_dev;
4495 rc = lu_site_init(s, &m->mdt_md_dev.md_lu_dev);
4497 CERROR("Can't init lu_site, rc %d\n", rc);
4498 GOTO(err_free_site, rc);
4501 /* set server index */
4502 lu_site2md(s)->ms_node_id = node_id;
4504 /* failover is the default
4505 * FIXME: we do not failout mds0/mgs, which may cause some problems.
4506 * assumed whose ms_node_id == 0 XXX
4508 obd->obd_replayable = 1;
4509 /* No connection accepted until configurations will finish */
4510 obd->obd_no_conn = 1;
4512 if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
4513 char *str = lustre_cfg_string(cfg, 4);
4514 if (strchr(str, 'n')) {
4515 CWARN("%s: recovery disabled\n", obd->obd_name);
4516 obd->obd_replayable = 0;
4520 /* init the stack */
4521 rc = mdt_stack_init((struct lu_env *)env, m, cfg, lmi);
4523 CERROR("Can't init device stack, rc %d\n", rc);
4524 GOTO(err_lu_site, rc);
4527 rc = lut_init(env, &m->mdt_lut, obd, m->mdt_bottom);
4529 GOTO(err_fini_stack, rc);
4531 rc = mdt_fld_init(env, obd->obd_name, m);
4535 rc = mdt_seq_init(env, obd->obd_name, m);
4537 GOTO(err_fini_fld, rc);
4539 snprintf(info->mti_u.ns_name, sizeof info->mti_u.ns_name,
4540 LUSTRE_MDT_NAME"-%p", m);
4541 m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
4542 LDLM_NAMESPACE_SERVER,
4543 LDLM_NAMESPACE_GREEDY,
4545 if (m->mdt_namespace == NULL)
4546 GOTO(err_fini_seq, rc = -ENOMEM);
4548 ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
4549 /* set obd_namespace for compatibility with old code */
4550 obd->obd_namespace = m->mdt_namespace;
4552 cfs_timer_init(&m->mdt_ck_timer, mdt_ck_timer_callback, m);
4554 rc = mdt_ck_thread_start(m);
4556 GOTO(err_free_ns, rc);
4558 rc = mdt_fs_setup(env, m, obd, lsi);
4562 rc = mdt_obd_llog_setup(obd, lsi);
4564 GOTO(err_fs_cleanup, rc);
4566 rc = mdt_llog_ctxt_clone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4568 GOTO(err_llog_cleanup, rc);
4570 mdt_adapt_sptlrpc_conf(obd, 1);
4572 next = m->mdt_child;
4573 #ifdef HAVE_QUOTA_SUPPORT
4574 rc = next->md_ops->mdo_quota.mqo_setup(env, next, lmi->lmi_mnt);
4576 GOTO(err_llog_cleanup, rc);
4579 server_put_mount_2(dev, lmi->lmi_mnt);
4582 rc = next->md_ops->mdo_iocontrol(env, next, OBD_IOC_GET_MNTOPT, 0,
4585 GOTO(err_quota, rc);
4587 if (mntopts & MNTOPT_USERXATTR)
4588 m->mdt_opts.mo_user_xattr = 1;
4590 m->mdt_opts.mo_user_xattr = 0;
4592 if (mntopts & MNTOPT_ACL)
4593 m->mdt_opts.mo_acl = 1;
4595 m->mdt_opts.mo_acl = 0;
4597 /* XXX: to support suppgid for ACL, we enable identity_upcall
4598 * by default, otherwise, maybe got unexpected -EACCESS. */
4599 if (m->mdt_opts.mo_acl)
4600 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
4602 m->mdt_identity_cache = upcall_cache_init(obd->obd_name,identity_upcall,
4603 &mdt_identity_upcall_cache_ops);
4604 if (IS_ERR(m->mdt_identity_cache)) {
4605 rc = PTR_ERR(m->mdt_identity_cache);
4606 m->mdt_identity_cache = NULL;
4607 GOTO(err_quota, rc);
4610 target_recovery_init(&m->mdt_lut, mdt_recovery_handle);
4612 rc = mdt_procfs_init(m, dev);
4614 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
4615 GOTO(err_recovery, rc);
4618 rc = mdt_start_ptlrpc_service(m);
4620 GOTO(err_procfs, rc);
4622 ping_evictor_start();
4624 rc = lu_site_init_finish(s);
4626 GOTO(err_stop_service, rc);
4628 if (obd->obd_recovering == 0)
4629 mdt_postrecov(env, m);
4631 mdt_init_capa_ctxt(env, m);
4633 /* Reduce the initial timeout on an MDS because it doesn't need such
4634 * a long timeout as an OST does. Adaptive timeouts will adjust this
4635 * value appropriately. */
4636 if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
4637 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
4642 ping_evictor_stop();
4643 mdt_stop_ptlrpc_service(m);
4647 target_recovery_fini(obd);
4648 upcall_cache_cleanup(m->mdt_identity_cache);
4649 m->mdt_identity_cache = NULL;
4651 #ifdef HAVE_QUOTA_SUPPORT
4652 next->md_ops->mdo_quota.mqo_cleanup(env, next);
4655 mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4656 mdt_obd_llog_cleanup(obd);
4658 mdt_fs_cleanup(env, m);
4660 cfs_timer_disarm(&m->mdt_ck_timer);
4661 mdt_ck_thread_stop(m);
4663 ldlm_namespace_free(m->mdt_namespace, NULL, 0);
4664 obd->obd_namespace = m->mdt_namespace = NULL;
4666 mdt_seq_fini(env, m);
4668 mdt_fld_fini(env, m);
4670 lut_fini(env, &m->mdt_lut);
4672 mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4679 server_put_mount_2(dev, lmi->lmi_mnt);
4683 /* used by MGS to process specific configurations */
4684 static int mdt_process_config(const struct lu_env *env,
4685 struct lu_device *d, struct lustre_cfg *cfg)
4687 struct mdt_device *m = mdt_dev(d);
4688 struct md_device *md_next = m->mdt_child;
4689 struct lu_device *next = md2lu_dev(md_next);
4693 switch (cfg->lcfg_command) {
4695 struct lprocfs_static_vars lvars;
4696 struct obd_device *obd = d->ld_obd;
4699 * For interoperability between 1.8 and 2.0,
4702 /* Skip old "mdt.group_upcall" param. */
4703 char *param = lustre_cfg_string(cfg, 1);
4704 if (param && !strncmp("mdt.group_upcall", param, 16)) {
4705 CWARN("For 1.8 interoperability, skip this"
4706 " mdt.group_upcall. It is obsolete\n");
4709 /* Rename old "mdt.quota_type" to "mdd.quota_type. */
4710 if (param && !strncmp("mdt.quota_type", param, 14)) {
4711 CWARN("Found old param mdt.quota_type, changed"
4712 " it to mdd.quota_type.\n");
4717 lprocfs_mdt_init_vars(&lvars);
4718 rc = class_process_proc_param(PARAM_MDT, lvars.obd_vars,
4720 if (rc > 0 || rc == -ENOSYS)
4721 /* we don't understand; pass it on */
4722 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4727 * Add mdc hook to get first MDT uuid and connect it to
4728 * ls->controller to use for seq manager.
4730 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4732 CERROR("Can't add mdc, rc %d\n", rc);
4734 rc = mdt_seq_init_cli(env, mdt_dev(d), cfg);
4737 /* others are passed further */
4738 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4744 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
4745 const struct lu_object_header *hdr,
4746 struct lu_device *d)
4748 struct mdt_object *mo;
4754 struct lu_object *o;
4755 struct lu_object_header *h;
4757 o = &mo->mot_obj.mo_lu;
4758 h = &mo->mot_header;
4759 lu_object_header_init(h);
4760 lu_object_init(o, h, d);
4761 lu_object_add_top(h, o);
4762 o->lo_ops = &mdt_obj_ops;
4763 cfs_sema_init(&mo->mot_ioepoch_sem, 1);
4764 cfs_sema_init(&mo->mot_lov_sem, 1);
4770 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
4771 const struct lu_object_conf *unused)
4773 struct mdt_device *d = mdt_dev(o->lo_dev);
4774 struct lu_device *under;
4775 struct lu_object *below;
4779 CDEBUG(D_INFO, "object init, fid = "DFID"\n",
4780 PFID(lu_object_fid(o)));
4782 under = &d->mdt_child->md_lu_dev;
4783 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
4784 if (below != NULL) {
4785 lu_object_add(o, below);
4792 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
4794 struct mdt_object *mo = mdt_obj(o);
4795 struct lu_object_header *h;
4799 CDEBUG(D_INFO, "object free, fid = "DFID"\n",
4800 PFID(lu_object_fid(o)));
4803 lu_object_header_fini(h);
4808 static int mdt_object_print(const struct lu_env *env, void *cookie,
4809 lu_printer_t p, const struct lu_object *o)
4811 struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
4812 return (*p)(env, cookie, LUSTRE_MDT_NAME"-object@%p(ioepoch="LPU64" "
4813 "flags="LPX64", epochcount=%d, writecount=%d)",
4814 mdto, mdto->mot_ioepoch, mdto->mot_flags,
4815 mdto->mot_ioepoch_count, mdto->mot_writecount);
4818 static const struct lu_device_operations mdt_lu_ops = {
4819 .ldo_object_alloc = mdt_object_alloc,
4820 .ldo_process_config = mdt_process_config,
4823 static const struct lu_object_operations mdt_obj_ops = {
4824 .loo_object_init = mdt_object_init,
4825 .loo_object_free = mdt_object_free,
4826 .loo_object_print = mdt_object_print
4829 static int mdt_obd_set_info_async(struct obd_export *exp,
4830 __u32 keylen, void *key,
4831 __u32 vallen, void *val,
4832 struct ptlrpc_request_set *set)
4834 struct obd_device *obd = exp->exp_obd;
4840 if (KEY_IS(KEY_SPTLRPC_CONF)) {
4841 rc = mdt_adapt_sptlrpc_conf(obd, 0);
4848 /* mds_connect_internal */
4849 static int mdt_connect_internal(struct obd_export *exp,
4850 struct mdt_device *mdt,
4851 struct obd_connect_data *data)
4854 data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
4855 data->ocd_ibits_known &= MDS_INODELOCK_FULL;
4857 /* If no known bits (which should not happen, probably,
4858 as everybody should support LOOKUP and UPDATE bits at least)
4859 revert to compat mode with plain locks. */
4860 if (!data->ocd_ibits_known &&
4861 data->ocd_connect_flags & OBD_CONNECT_IBITS)
4862 data->ocd_connect_flags &= ~OBD_CONNECT_IBITS;
4864 if (!mdt->mdt_opts.mo_acl)
4865 data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
4867 if (!mdt->mdt_opts.mo_user_xattr)
4868 data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
4870 if (!mdt->mdt_som_conf)
4871 data->ocd_connect_flags &= ~OBD_CONNECT_SOM;
4873 if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE) {
4874 data->ocd_brw_size = min(data->ocd_brw_size,
4875 (__u32)(PTLRPC_MAX_BRW_PAGES << CFS_PAGE_SHIFT));
4876 if (data->ocd_brw_size == 0) {
4877 CERROR("%s: cli %s/%p ocd_connect_flags: "LPX64
4878 " ocd_version: %x ocd_grant: %d "
4879 "ocd_index: %u ocd_brw_size is "
4880 "unexpectedly zero, network data "
4881 "corruption? Refusing connection of this"
4883 exp->exp_obd->obd_name,
4884 exp->exp_client_uuid.uuid,
4885 exp, data->ocd_connect_flags, data->ocd_version,
4886 data->ocd_grant, data->ocd_index);
4891 cfs_spin_lock(&exp->exp_lock);
4892 exp->exp_connect_flags = data->ocd_connect_flags;
4893 cfs_spin_unlock(&exp->exp_lock);
4894 data->ocd_version = LUSTRE_VERSION_CODE;
4895 exp->exp_mdt_data.med_ibits_known = data->ocd_ibits_known;
4899 if (mdt->mdt_opts.mo_acl &&
4900 ((exp->exp_connect_flags & OBD_CONNECT_ACL) == 0)) {
4901 CWARN("%s: MDS requires ACL support but client does not\n",
4902 mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4907 if ((exp->exp_connect_flags & OBD_CONNECT_FID) == 0) {
4908 CWARN("%s: MDS requires FID support, but client not\n",
4909 mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4913 if (mdt->mdt_som_conf && !exp_connect_som(exp) &&
4914 !(exp->exp_connect_flags & OBD_CONNECT_MDS_MDS)) {
4915 CWARN("%s: MDS has SOM enabled, but client does not support "
4916 "it\n", mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
4923 static int mdt_connect_check_sptlrpc(struct mdt_device *mdt,
4924 struct obd_export *exp,
4925 struct ptlrpc_request *req)
4927 struct sptlrpc_flavor flvr;
4930 if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
4931 cfs_read_lock(&mdt->mdt_sptlrpc_lock);
4932 sptlrpc_target_choose_flavor(&mdt->mdt_sptlrpc_rset,
4936 cfs_read_unlock(&mdt->mdt_sptlrpc_lock);
4938 cfs_spin_lock(&exp->exp_lock);
4940 exp->exp_sp_peer = req->rq_sp_from;
4941 exp->exp_flvr = flvr;
4943 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
4944 exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
4945 CERROR("unauthorized rpc flavor %x from %s, "
4946 "expect %x\n", req->rq_flvr.sf_rpc,
4947 libcfs_nid2str(req->rq_peer.nid),
4948 exp->exp_flvr.sf_rpc);
4952 cfs_spin_unlock(&exp->exp_lock);
4954 if (exp->exp_sp_peer != req->rq_sp_from) {
4955 CERROR("RPC source %s doesn't match %s\n",
4956 sptlrpc_part2name(req->rq_sp_from),
4957 sptlrpc_part2name(exp->exp_sp_peer));
4960 rc = sptlrpc_target_export_check(exp, req);
4967 /* mds_connect copy */
4968 static int mdt_obd_connect(const struct lu_env *env,
4969 struct obd_export **exp, struct obd_device *obd,
4970 struct obd_uuid *cluuid,
4971 struct obd_connect_data *data,
4974 struct mdt_thread_info *info;
4975 struct obd_export *lexp;
4976 struct lustre_handle conn = { 0 };
4977 struct mdt_device *mdt;
4978 struct ptlrpc_request *req;
4982 LASSERT(env != NULL);
4983 if (!exp || !obd || !cluuid)
4986 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4987 req = info->mti_pill->rc_req;
4988 mdt = mdt_dev(obd->obd_lu_dev);
4990 rc = class_connect(&conn, obd, cluuid);
4994 lexp = class_conn2export(&conn);
4995 LASSERT(lexp != NULL);
4997 rc = mdt_connect_check_sptlrpc(mdt, lexp, req);
5001 if (OBD_FAIL_CHECK(OBD_FAIL_TGT_RCVG_FLAG))
5002 lustre_msg_add_op_flags(req->rq_repmsg, MSG_CONNECT_RECOVERING);
5004 rc = mdt_connect_internal(lexp, mdt, data);
5006 struct mdt_thread_info *mti;
5007 struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
5009 mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5010 LASSERT(mti != NULL);
5011 mti->mti_exp = lexp;
5012 memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
5013 rc = mdt_client_new(env, mdt);
5015 mdt_export_stats_init(obd, lexp, localdata);
5020 class_disconnect(lexp);
5029 static int mdt_obd_reconnect(const struct lu_env *env,
5030 struct obd_export *exp, struct obd_device *obd,
5031 struct obd_uuid *cluuid,
5032 struct obd_connect_data *data,
5035 struct mdt_thread_info *info;
5036 struct mdt_device *mdt;
5037 struct ptlrpc_request *req;
5041 if (exp == NULL || obd == NULL || cluuid == NULL)
5044 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5045 req = info->mti_pill->rc_req;
5046 mdt = mdt_dev(obd->obd_lu_dev);
5048 rc = mdt_connect_check_sptlrpc(mdt, exp, req);
5052 rc = mdt_connect_internal(exp, mdt_dev(obd->obd_lu_dev), data);
5054 mdt_export_stats_init(obd, exp, localdata);
5058 static int mdt_export_cleanup(struct obd_export *exp)
5060 struct mdt_export_data *med = &exp->exp_mdt_data;
5061 struct obd_device *obd = exp->exp_obd;
5062 struct mdt_device *mdt;
5063 struct mdt_thread_info *info;
5065 CFS_LIST_HEAD(closing_list);
5066 struct mdt_file_data *mfd, *n;
5070 cfs_spin_lock(&med->med_open_lock);
5071 while (!cfs_list_empty(&med->med_open_head)) {
5072 cfs_list_t *tmp = med->med_open_head.next;
5073 mfd = cfs_list_entry(tmp, struct mdt_file_data, mfd_list);
5075 /* Remove mfd handle so it can't be found again.
5076 * We are consuming the mfd_list reference here. */
5077 class_handle_unhash(&mfd->mfd_handle);
5078 cfs_list_move_tail(&mfd->mfd_list, &closing_list);
5080 cfs_spin_unlock(&med->med_open_lock);
5081 mdt = mdt_dev(obd->obd_lu_dev);
5082 LASSERT(mdt != NULL);
5084 rc = lu_env_init(&env, LCT_MD_THREAD);
5088 info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5089 LASSERT(info != NULL);
5090 memset(info, 0, sizeof *info);
5091 info->mti_env = &env;
5092 info->mti_mdt = mdt;
5093 info->mti_exp = exp;
5095 if (!cfs_list_empty(&closing_list)) {
5096 struct md_attr *ma = &info->mti_attr;
5100 lmm_size = mdt->mdt_max_mdsize;
5101 OBD_ALLOC_LARGE(ma->ma_lmm, lmm_size);
5102 if (ma->ma_lmm == NULL)
5103 GOTO(out_lmm, rc = -ENOMEM);
5105 cookie_size = mdt->mdt_max_cookiesize;
5106 OBD_ALLOC_LARGE(ma->ma_cookie, cookie_size);
5107 if (ma->ma_cookie == NULL)
5108 GOTO(out_cookie, rc = -ENOMEM);
5110 /* Close any open files (which may also cause orphan unlinking). */
5111 cfs_list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
5112 cfs_list_del_init(&mfd->mfd_list);
5113 memset(&ma->ma_attr, 0, sizeof(ma->ma_attr));
5114 ma->ma_lmm_size = lmm_size;
5115 ma->ma_cookie_size = cookie_size;
5117 /* It is not for setattr, just tell MDD to send
5118 * DESTROY RPC to OSS if needed */
5119 ma->ma_valid = MA_FLAGS;
5120 ma->ma_attr_flags = MDS_CLOSE_CLEANUP;
5121 /* Don't unlink orphan on failover umount, LU-184 */
5122 if (exp->exp_flags & OBD_OPT_FAILOVER)
5123 ma->ma_attr_flags |= MDS_KEEP_ORPHAN;
5124 mdt_mfd_close(info, mfd);
5126 OBD_FREE_LARGE(ma->ma_cookie, cookie_size);
5127 ma->ma_cookie = NULL;
5129 OBD_FREE_LARGE(ma->ma_lmm, lmm_size);
5133 info->mti_mdt = NULL;
5134 /* cleanup client slot early */
5135 /* Do not erase record for recoverable client. */
5136 if (!(exp->exp_flags & OBD_OPT_FAILOVER) || exp->exp_failed)
5137 mdt_client_del(&env, mdt);
5143 static int mdt_obd_disconnect(struct obd_export *exp)
5149 class_export_get(exp);
5151 rc = server_disconnect_export(exp);
5153 CDEBUG(D_IOCTL, "server disconnect error: %d\n", rc);
5155 rc = mdt_export_cleanup(exp);
5156 class_export_put(exp);
5160 /* FIXME: Can we avoid using these two interfaces? */
5161 static int mdt_init_export(struct obd_export *exp)
5163 struct mdt_export_data *med = &exp->exp_mdt_data;
5167 CFS_INIT_LIST_HEAD(&med->med_open_head);
5168 cfs_spin_lock_init(&med->med_open_lock);
5169 cfs_sema_init(&med->med_idmap_sem, 1);
5170 med->med_idmap = NULL;
5171 cfs_spin_lock(&exp->exp_lock);
5172 exp->exp_connecting = 1;
5173 cfs_spin_unlock(&exp->exp_lock);
5175 /* self-export doesn't need client data and ldlm initialization */
5176 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5177 &exp->exp_client_uuid)))
5180 rc = lut_client_alloc(exp);
5182 rc = ldlm_init_export(exp);
5184 CERROR("%s: Error %d while initializing export\n",
5185 exp->exp_obd->obd_name, rc);
5189 static int mdt_destroy_export(struct obd_export *exp)
5193 if (exp_connect_rmtclient(exp))
5194 mdt_cleanup_idmap(&exp->exp_mdt_data);
5196 target_destroy_export(exp);
5197 /* destroy can be called from failed obd_setup, so
5198 * checking uuid is safer than obd_self_export */
5199 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5200 &exp->exp_client_uuid)))
5203 ldlm_destroy_export(exp);
5204 lut_client_free(exp);
5206 LASSERT(cfs_list_empty(&exp->exp_outstanding_replies));
5207 LASSERT(cfs_list_empty(&exp->exp_mdt_data.med_open_head));
5212 static void mdt_allow_cli(struct mdt_device *m, unsigned int flag)
5214 if (flag & CONFIG_LOG)
5215 cfs_set_bit(MDT_FL_CFGLOG, &m->mdt_state);
5217 /* also notify active event */
5218 if (flag & CONFIG_SYNC)
5219 cfs_set_bit(MDT_FL_SYNCED, &m->mdt_state);
5221 if (cfs_test_bit(MDT_FL_CFGLOG, &m->mdt_state) &&
5222 cfs_test_bit(MDT_FL_SYNCED, &m->mdt_state)) {
5223 struct obd_device *obd = m->mdt_md_dev.md_lu_dev.ld_obd;
5225 /* Open for clients */
5226 if (obd->obd_no_conn) {
5227 cfs_spin_lock(&obd->obd_dev_lock);
5228 obd->obd_no_conn = 0;
5229 cfs_spin_unlock(&obd->obd_dev_lock);
5234 static int mdt_upcall(const struct lu_env *env, struct md_device *md,
5235 enum md_upcall_event ev, void *data)
5237 struct mdt_device *m = mdt_dev(&md->md_lu_dev);
5238 struct md_device *next = m->mdt_child;
5239 struct mdt_thread_info *mti;
5245 rc = next->md_ops->mdo_maxsize_get(env, next,
5247 &m->mdt_max_cookiesize);
5248 CDEBUG(D_INFO, "get max mdsize %d max cookiesize %d\n",
5249 m->mdt_max_mdsize, m->mdt_max_cookiesize);
5250 mdt_allow_cli(m, CONFIG_SYNC);
5253 m->mdt_lut.lut_obd->u.obt.obt_mount_count;
5256 mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5257 mti->mti_no_need_trans = 1;
5258 CDEBUG(D_INFO, "disable mdt trans for this thread\n");
5261 /* Check that MDT is not yet configured */
5262 LASSERT(!cfs_test_bit(MDT_FL_CFGLOG, &m->mdt_state));
5264 #ifdef HAVE_QUOTA_SUPPORT
5266 if (md->md_lu_dev.ld_obd->obd_recovering == 0 &&
5267 likely(md->md_lu_dev.ld_obd->obd_stopping == 0))
5268 next->md_ops->mdo_quota.mqo_recovery(env, next);
5272 CERROR("invalid event\n");
5279 static int mdt_obd_notify(struct obd_device *obd,
5280 struct obd_device *watched,
5281 enum obd_notify_event ev, void *data)
5283 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
5284 #ifdef HAVE_QUOTA_SUPPORT
5285 struct md_device *next = mdt->mdt_child;
5290 case OBD_NOTIFY_CONFIG:
5291 mdt_allow_cli(mdt, (unsigned long)data);
5293 #ifdef HAVE_QUOTA_SUPPORT
5294 /* quota_type has been processed, we can now handle
5295 * incoming quota requests */
5296 next->md_ops->mdo_quota.mqo_notify(NULL, next);
5300 CDEBUG(D_INFO, "Unhandled notification %#x\n", ev);
5305 static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key,
5306 void *val, int vallen)
5308 struct mdt_device *mdt = mdt_dev(info->mti_exp->exp_obd->obd_lu_dev);
5309 struct getinfo_fid2path *fpout, *fpin;
5312 fpin = key + cfs_size_round(sizeof(KEY_FID2PATH));
5315 if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
5316 lustre_swab_fid2path(fpin);
5318 memcpy(fpout, fpin, sizeof(*fpin));
5319 if (fpout->gf_pathlen != vallen - sizeof(*fpin))
5322 rc = mdt_fid2path(info->mti_env, mdt, fpout);
5326 static int mdt_fid2path(const struct lu_env *env, struct mdt_device *mdt,
5327 struct getinfo_fid2path *fp)
5329 struct mdt_object *obj;
5333 CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n",
5334 PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
5336 if (!fid_is_sane(&fp->gf_fid))
5339 obj = mdt_object_find(env, mdt, &fp->gf_fid);
5340 if (obj == NULL || IS_ERR(obj)) {
5341 CDEBUG(D_IOCTL, "no object "DFID": %ld\n",PFID(&fp->gf_fid),
5346 rc = lu_object_exists(&obj->mot_obj.mo_lu);
5352 mdt_object_put(env, obj);
5353 CDEBUG(D_IOCTL, "nonlocal object "DFID": %d\n",
5354 PFID(&fp->gf_fid), rc);
5358 rc = mo_path(env, md_object_next(&obj->mot_obj), fp->gf_path,
5359 fp->gf_pathlen, &fp->gf_recno, &fp->gf_linkno);
5360 mdt_object_put(env, obj);
5365 static int mdt_get_info(struct mdt_thread_info *info)
5367 struct ptlrpc_request *req = mdt_info_req(info);
5375 key = req_capsule_client_get(info->mti_pill, &RMF_GETINFO_KEY);
5377 CDEBUG(D_IOCTL, "No GETINFO key");
5380 keylen = req_capsule_get_size(info->mti_pill, &RMF_GETINFO_KEY,
5383 vallen = req_capsule_client_get(info->mti_pill, &RMF_GETINFO_VALLEN);
5384 if (vallen == NULL) {
5385 CDEBUG(D_IOCTL, "Unable to get RMF_GETINFO_VALLEN buffer");
5389 req_capsule_set_size(info->mti_pill, &RMF_GETINFO_VAL, RCL_SERVER,
5391 rc = req_capsule_server_pack(info->mti_pill);
5392 valout = req_capsule_server_get(info->mti_pill, &RMF_GETINFO_VAL);
5393 if (valout == NULL) {
5394 CDEBUG(D_IOCTL, "Unable to get get-info RPC out buffer");
5398 if (KEY_IS(KEY_FID2PATH))
5399 rc = mdt_rpc_fid2path(info, key, valout, *vallen);
5403 lustre_msg_set_status(req->rq_repmsg, rc);
5408 /* Pass the ioc down */
5409 static int mdt_ioc_child(struct lu_env *env, struct mdt_device *mdt,
5410 unsigned int cmd, int len, void *data)
5412 struct lu_context ioctl_session;
5413 struct md_device *next = mdt->mdt_child;
5417 rc = lu_context_init(&ioctl_session, LCT_SESSION);
5420 ioctl_session.lc_thread = (struct ptlrpc_thread *)cfs_current();
5421 lu_context_enter(&ioctl_session);
5422 env->le_ses = &ioctl_session;
5424 LASSERT(next->md_ops->mdo_iocontrol);
5425 rc = next->md_ops->mdo_iocontrol(env, next, cmd, len, data);
5427 lu_context_exit(&ioctl_session);
5428 lu_context_fini(&ioctl_session);
5432 static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
5434 struct obd_ioctl_data *data = karg;
5435 struct lu_fid *fid = (struct lu_fid *)data->ioc_inlbuf1;
5437 struct mdt_object *obj;
5438 struct mdt_lock_handle *lh;
5442 CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
5443 if (!fid_is_sane(fid))
5446 lh = &mti->mti_lh[MDT_LH_PARENT];
5447 mdt_lock_reg_init(lh, LCK_CR);
5449 obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE);
5451 RETURN(PTR_ERR(obj));
5453 rc = mdt_object_exists(obj);
5457 * before calling version get the correct MDS should be
5458 * fid, this is error to find remote object here
5460 CERROR("nonlocal object "DFID"\n", PFID(fid));
5461 } else if (rc == 0) {
5462 *(__u64 *)data->ioc_inlbuf2 = ENOENT_VERSION;
5465 version = dt_version_get(mti->mti_env, mdt_obj2dt(obj));
5466 *(__u64 *)data->ioc_inlbuf2 = version;
5469 mdt_object_unlock_put(mti, obj, lh, 1);
5473 /* ioctls on obd dev */
5474 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
5475 void *karg, void *uarg)
5478 struct obd_device *obd = exp->exp_obd;
5479 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
5480 struct dt_device *dt = mdt->mdt_bottom;
5484 CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
5485 rc = lu_env_init(&env, LCT_MD_THREAD);
5491 rc = mdt_device_sync(&env, mdt);
5493 case OBD_IOC_SET_READONLY:
5494 dt->dd_ops->dt_ro(&env, dt);
5496 case OBD_IOC_ABORT_RECOVERY:
5497 CERROR("Aborting recovery for device %s\n", obd->obd_name);
5498 target_stop_recovery_thread(obd);
5501 case OBD_IOC_CHANGELOG_REG:
5502 case OBD_IOC_CHANGELOG_DEREG:
5503 case OBD_IOC_CHANGELOG_CLEAR:
5504 rc = mdt_ioc_child(&env, mdt, cmd, len, karg);
5506 case OBD_IOC_GET_OBJ_VERSION: {
5507 struct mdt_thread_info *mti;
5508 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5509 memset(mti, 0, sizeof *mti);
5510 mti->mti_env = &env;
5514 rc = mdt_ioc_version_get(mti, karg);
5518 CERROR("Not supported cmd = %d for device %s\n",
5519 cmd, obd->obd_name);
5527 int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
5529 struct lu_device *ld = md2lu_dev(mdt->mdt_child);
5530 #ifdef HAVE_QUOTA_SUPPORT
5531 struct obd_device *obd = mdt2obd_dev(mdt);
5532 struct md_device *next = mdt->mdt_child;
5537 rc = ld->ld_ops->ldo_recovery_complete(env, ld);
5538 #ifdef HAVE_QUOTA_SUPPORT
5539 if (likely(obd->obd_stopping == 0))
5540 next->md_ops->mdo_quota.mqo_recovery(env, next);
5545 int mdt_obd_postrecov(struct obd_device *obd)
5550 rc = lu_env_init(&env, LCT_MD_THREAD);
5553 rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
5559 * Send a copytool req to a client
5560 * Note this sends a request RPC from a server (MDT) to a client (MDC),
5561 * backwards of normal comms.
5563 int mdt_hsm_copytool_send(struct obd_export *exp)
5566 struct hsm_action_list *hal;
5567 struct hsm_action_item *hai;
5571 CWARN("%s: writing to mdc at %s\n", exp->exp_obd->obd_name,
5572 libcfs_nid2str(exp->exp_connection->c_peer.nid));
5574 len = sizeof(*lh) + sizeof(*hal) + MTI_NAME_MAXLEN +
5575 /* for mockup below */ 2 * cfs_size_round(sizeof(*hai));
5580 lh->kuc_magic = KUC_MAGIC;
5581 lh->kuc_transport = KUC_TRANSPORT_HSM;
5582 lh->kuc_msgtype = HMT_ACTION_LIST;
5583 lh->kuc_msglen = len;
5585 hal = (struct hsm_action_list *)(lh + 1);
5586 hal->hal_version = HAL_VERSION;
5587 hal->hal_archive_num = 1;
5588 obd_uuid2fsname(hal->hal_fsname, exp->exp_obd->obd_name,
5591 /* mock up an action list */
5593 hai = hai_zero(hal);
5594 hai->hai_action = HSMA_ARCHIVE;
5595 hai->hai_fid.f_oid = 0xA00A;
5596 hai->hai_len = sizeof(*hai);
5597 hai = hai_next(hai);
5598 hai->hai_action = HSMA_RESTORE;
5599 hai->hai_fid.f_oid = 0xB00B;
5600 hai->hai_len = sizeof(*hai);
5602 /* Uses the ldlm reverse import; this rpc will be seen by
5603 the ldlm_callback_handler */
5604 rc = do_set_info_async(exp->exp_imp_reverse,
5605 LDLM_SET_INFO, LUSTRE_OBD_VERSION,
5606 sizeof(KEY_HSM_COPYTOOL_SEND),
5607 KEY_HSM_COPYTOOL_SEND,
5615 static struct obd_ops mdt_obd_device_ops = {
5616 .o_owner = THIS_MODULE,
5617 .o_set_info_async = mdt_obd_set_info_async,
5618 .o_connect = mdt_obd_connect,
5619 .o_reconnect = mdt_obd_reconnect,
5620 .o_disconnect = mdt_obd_disconnect,
5621 .o_init_export = mdt_init_export,
5622 .o_destroy_export = mdt_destroy_export,
5623 .o_iocontrol = mdt_iocontrol,
5624 .o_postrecov = mdt_obd_postrecov,
5625 .o_notify = mdt_obd_notify
5628 static struct lu_device* mdt_device_fini(const struct lu_env *env,
5629 struct lu_device *d)
5631 struct mdt_device *m = mdt_dev(d);
5638 static struct lu_device *mdt_device_free(const struct lu_env *env,
5639 struct lu_device *d)
5641 struct mdt_device *m = mdt_dev(d);
5644 md_device_fini(&m->mdt_md_dev);
5649 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
5650 struct lu_device_type *t,
5651 struct lustre_cfg *cfg)
5653 struct lu_device *l;
5654 struct mdt_device *m;
5660 l = &m->mdt_md_dev.md_lu_dev;
5661 rc = mdt_init0(env, m, t, cfg);
5663 mdt_device_free(env, l);
5667 md_upcall_init(&m->mdt_md_dev, mdt_upcall);
5669 l = ERR_PTR(-ENOMEM);
5673 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
5674 LU_KEY_INIT_FINI(mdt, struct mdt_thread_info);
5676 /* context key: mdt_thread_key */
5677 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
5679 struct md_ucred *mdt_ucred(const struct mdt_thread_info *info)
5681 return md_ucred(info->mti_env);
5685 * Enable/disable COS (Commit On Sharing).
5687 * Set/Clear the COS flag in mdt options.
5689 * \param mdt mdt device
5690 * \param val 0 disables COS, other values enable COS
5692 void mdt_enable_cos(struct mdt_device *mdt, int val)
5697 mdt->mdt_opts.mo_cos = !!val;
5698 rc = lu_env_init(&env, LCT_MD_THREAD);
5699 if (unlikely(rc != 0)) {
5700 CWARN("lu_env initialization failed with rc = %d,"
5701 "cannot sync\n", rc);
5704 mdt_device_sync(&env, mdt);
5709 * Check COS (Commit On Sharing) status.
5711 * Return COS flag status.
5713 * \param mdt mdt device
5715 int mdt_cos_is_enabled(struct mdt_device *mdt)
5717 return mdt->mdt_opts.mo_cos != 0;
5720 /* type constructor/destructor: mdt_type_init, mdt_type_fini */
5721 LU_TYPE_INIT_FINI(mdt, &mdt_thread_key);
5723 static struct lu_device_type_operations mdt_device_type_ops = {
5724 .ldto_init = mdt_type_init,
5725 .ldto_fini = mdt_type_fini,
5727 .ldto_start = mdt_type_start,
5728 .ldto_stop = mdt_type_stop,
5730 .ldto_device_alloc = mdt_device_alloc,
5731 .ldto_device_free = mdt_device_free,
5732 .ldto_device_fini = mdt_device_fini
5735 static struct lu_device_type mdt_device_type = {
5736 .ldt_tags = LU_DEVICE_MD,
5737 .ldt_name = LUSTRE_MDT_NAME,
5738 .ldt_ops = &mdt_device_type_ops,
5739 .ldt_ctx_tags = LCT_MD_THREAD
5742 static struct lu_local_obj_desc mdt_last_recv = {
5743 .llod_name = LAST_RCVD,
5744 .llod_oid = MDT_LAST_RECV_OID,
5748 static int __init mdt_mod_init(void)
5750 struct lprocfs_static_vars lvars;
5753 llo_local_obj_register(&mdt_last_recv);
5755 if (mdt_num_threads > 0) {
5756 if (mdt_num_threads > MDT_MAX_THREADS)
5757 mdt_num_threads = MDT_MAX_THREADS;
5758 if (mdt_num_threads < MDT_MIN_THREADS)
5759 mdt_num_threads = MDT_MIN_THREADS;
5760 mdt_max_threads = mdt_min_threads = mdt_num_threads;
5762 mdt_max_threads = MDT_MAX_THREADS;
5763 mdt_min_threads = MDT_MIN_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);