4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2010, 2013, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/mdt/mdt_handler.c
38 * Lustre Metadata Target (mdt) request handler
40 * Author: Peter Braam <braam@clusterfs.com>
41 * Author: Andreas Dilger <adilger@clusterfs.com>
42 * Author: Phil Schwan <phil@clusterfs.com>
43 * Author: Mike Shaver <shaver@clusterfs.com>
44 * Author: Nikita Danilov <nikita@clusterfs.com>
45 * Author: Huang Hua <huanghua@clusterfs.com>
46 * Author: Yury Umanets <umka@clusterfs.com>
49 #define DEBUG_SUBSYSTEM S_MDS
51 #include <linux/module.h>
53 * struct OBD_{ALLOC,FREE}*()
55 #include <obd_support.h>
56 /* struct ptlrpc_request */
57 #include <lustre_net.h>
58 /* struct obd_export */
59 #include <lustre_export.h>
60 /* struct obd_device */
63 #include <dt_object.h>
64 #include <lustre_mds.h>
65 #include <lustre_mdt.h>
66 #include <lustre_log.h>
67 #include "mdt_internal.h"
68 #include <lustre_acl.h>
69 #include <lustre_param.h>
70 #include <lustre_quota.h>
71 #include <lustre_linkea.h>
72 #include <lustre_lfsck.h>
74 mdl_mode_t mdt_mdl_lock_modes[] = {
75 [LCK_MINMODE] = MDL_MINMODE,
82 [LCK_GROUP] = MDL_GROUP
85 ldlm_mode_t mdt_dlm_lock_modes[] = {
86 [MDL_MINMODE] = LCK_MINMODE,
93 [MDL_GROUP] = LCK_GROUP
96 static struct mdt_device *mdt_dev(struct lu_device *d);
97 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags);
99 static const struct lu_object_operations mdt_obj_ops;
101 /* Slab for MDT object allocation */
102 static cfs_mem_cache_t *mdt_object_kmem;
104 static struct lu_kmem_descr mdt_caches[] = {
106 .ckd_cache = &mdt_object_kmem,
107 .ckd_name = "mdt_obj",
108 .ckd_size = sizeof(struct mdt_object)
115 int mdt_get_disposition(struct ldlm_reply *rep, int flag)
119 return (rep->lock_policy_res1 & flag);
122 void mdt_clear_disposition(struct mdt_thread_info *info,
123 struct ldlm_reply *rep, int flag)
126 info->mti_opdata &= ~flag;
128 rep->lock_policy_res1 &= ~flag;
131 void mdt_set_disposition(struct mdt_thread_info *info,
132 struct ldlm_reply *rep, int flag)
135 info->mti_opdata |= flag;
137 rep->lock_policy_res1 |= flag;
140 void mdt_lock_reg_init(struct mdt_lock_handle *lh, ldlm_mode_t lm)
142 lh->mlh_pdo_hash = 0;
143 lh->mlh_reg_mode = lm;
144 lh->mlh_rreg_mode = lm;
145 lh->mlh_type = MDT_REG_LOCK;
148 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, ldlm_mode_t lm,
149 const char *name, int namelen)
151 lh->mlh_reg_mode = lm;
152 lh->mlh_rreg_mode = lm;
153 lh->mlh_type = MDT_PDO_LOCK;
155 if (name != NULL && (name[0] != '\0')) {
156 LASSERT(namelen > 0);
157 lh->mlh_pdo_hash = full_name_hash(name, namelen);
158 /* XXX Workaround for LU-2856
159 * Zero is a valid return value of full_name_hash, but several
160 * users of mlh_pdo_hash assume a non-zero hash value. We
161 * therefore map zero onto an arbitrary, but consistent
162 * value (1) to avoid problems further down the road. */
163 if (unlikely(!lh->mlh_pdo_hash))
164 lh->mlh_pdo_hash = 1;
166 LASSERT(namelen == 0);
167 lh->mlh_pdo_hash = 0ull;
171 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
172 struct mdt_lock_handle *lh)
178 * Any dir access needs couple of locks:
180 * 1) on part of dir we gonna take lookup/modify;
182 * 2) on whole dir to protect it from concurrent splitting and/or to
183 * flush client's cache for readdir().
185 * so, for a given mode and object this routine decides what lock mode
186 * to use for lock #2:
188 * 1) if caller's gonna lookup in dir then we need to protect dir from
189 * being splitted only - LCK_CR
191 * 2) if caller's gonna modify dir then we need to protect dir from
192 * being splitted and to flush cache - LCK_CW
194 * 3) if caller's gonna modify dir and that dir seems ready for
195 * splitting then we need to protect it from any type of access
196 * (lookup/modify/split) - LCK_EX --bzzz
199 LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
200 LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
203 * Ask underlaying level its opinion about preferable PDO lock mode
204 * having access type passed as regular lock mode:
206 * - MDL_MINMODE means that lower layer does not want to specify lock
209 * - MDL_NL means that no PDO lock should be taken. This is used in some
210 * cases. Say, for non-splittable directories no need to use PDO locks
213 mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
214 mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
216 if (mode != MDL_MINMODE) {
217 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
220 * Lower layer does not want to specify locking mode. We do it
221 * our selves. No special protection is needed, just flush
222 * client's cache on modification and allow concurrent
225 switch (lh->mlh_reg_mode) {
227 lh->mlh_pdo_mode = LCK_EX;
230 lh->mlh_pdo_mode = LCK_CR;
233 lh->mlh_pdo_mode = LCK_CW;
236 CERROR("Not expected lock type (0x%x)\n",
237 (int)lh->mlh_reg_mode);
242 LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
246 int mdt_getstatus(struct mdt_thread_info *info)
248 struct mdt_device *mdt = info->mti_mdt;
249 struct mdt_body *repbody;
253 rc = mdt_check_ucred(info);
255 RETURN(err_serious(rc));
257 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETSTATUS_PACK))
258 RETURN(err_serious(-ENOMEM));
260 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
261 repbody->fid1 = mdt->mdt_md_root_fid;
262 repbody->valid |= OBD_MD_FLID;
264 if (mdt->mdt_opts.mo_mds_capa &&
265 exp_connect_flags(info->mti_exp) & OBD_CONNECT_MDS_CAPA) {
266 struct mdt_object *root;
267 struct lustre_capa *capa;
269 root = mdt_object_find(info->mti_env, mdt, &repbody->fid1);
271 RETURN(PTR_ERR(root));
273 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA1);
275 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
276 rc = mo_capa_get(info->mti_env, mdt_object_child(root), capa,
278 mdt_object_put(info->mti_env, root);
280 repbody->valid |= OBD_MD_FLMDSCAPA;
286 int mdt_statfs(struct mdt_thread_info *info)
288 struct ptlrpc_request *req = mdt_info_req(info);
289 struct md_device *next = info->mti_mdt->mdt_child;
290 struct ptlrpc_service_part *svcpt;
291 struct obd_statfs *osfs;
296 svcpt = info->mti_pill->rc_req->rq_rqbd->rqbd_svcpt;
298 /* This will trigger a watchdog timeout */
299 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
300 (MDT_SERVICE_WATCHDOG_FACTOR *
301 at_get(&svcpt->scp_at_estimate)) + 1);
303 rc = mdt_check_ucred(info);
305 RETURN(err_serious(rc));
307 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK))
308 RETURN(err_serious(-ENOMEM));
310 osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
314 /** statfs information are cached in the mdt_device */
315 if (cfs_time_before_64(info->mti_mdt->mdt_osfs_age,
316 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS))) {
317 /** statfs data is too old, get up-to-date one */
318 rc = next->md_ops->mdo_statfs(info->mti_env, next, osfs);
321 spin_lock(&info->mti_mdt->mdt_osfs_lock);
322 info->mti_mdt->mdt_osfs = *osfs;
323 info->mti_mdt->mdt_osfs_age = cfs_time_current_64();
324 spin_unlock(&info->mti_mdt->mdt_osfs_lock);
326 /** use cached statfs data */
327 spin_lock(&info->mti_mdt->mdt_osfs_lock);
328 *osfs = info->mti_mdt->mdt_osfs;
329 spin_unlock(&info->mti_mdt->mdt_osfs_lock);
333 mdt_counter_incr(req, LPROC_MDT_STATFS);
339 * Pack SOM attributes into the reply.
340 * Call under a DLM UPDATE lock.
342 static void mdt_pack_size2body(struct mdt_thread_info *info,
343 struct mdt_object *mo)
346 struct md_attr *ma = &info->mti_attr;
348 LASSERT(ma->ma_attr.la_valid & LA_MODE);
349 b = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
351 /* Check if Size-on-MDS is supported, if this is a regular file,
352 * if SOM is enabled on the object and if SOM cache exists and valid.
353 * Otherwise do not pack Size-on-MDS attributes to the reply. */
354 if (!(mdt_conn_flags(info) & OBD_CONNECT_SOM) ||
355 !S_ISREG(ma->ma_attr.la_mode) ||
356 !mdt_object_is_som_enabled(mo) ||
357 !(ma->ma_valid & MA_SOM))
360 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
361 b->size = ma->ma_som->msd_size;
362 b->blocks = ma->ma_som->msd_blocks;
365 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
366 const struct lu_attr *attr, const struct lu_fid *fid)
368 struct md_attr *ma = &info->mti_attr;
370 LASSERT(ma->ma_valid & MA_INODE);
372 b->atime = attr->la_atime;
373 b->mtime = attr->la_mtime;
374 b->ctime = attr->la_ctime;
375 b->mode = attr->la_mode;
376 b->size = attr->la_size;
377 b->blocks = attr->la_blocks;
378 b->uid = attr->la_uid;
379 b->gid = attr->la_gid;
380 b->flags = attr->la_flags;
381 b->nlink = attr->la_nlink;
382 b->rdev = attr->la_rdev;
384 /*XXX should pack the reply body according to lu_valid*/
385 b->valid |= OBD_MD_FLCTIME | OBD_MD_FLUID |
386 OBD_MD_FLGID | OBD_MD_FLTYPE |
387 OBD_MD_FLMODE | OBD_MD_FLNLINK | OBD_MD_FLFLAGS |
388 OBD_MD_FLATIME | OBD_MD_FLMTIME ;
390 if (!S_ISREG(attr->la_mode)) {
391 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
392 } else if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV)) {
393 /* means no objects are allocated on osts. */
394 LASSERT(!(ma->ma_valid & MA_LOV));
395 /* just ignore blocks occupied by extend attributes on MDS */
397 /* if no object is allocated on osts, the size on mds is valid. b=22272 */
398 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
403 b->valid |= OBD_MD_FLID;
404 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, size="LPU64"\n",
405 PFID(fid), b->nlink, b->mode, b->size);
409 mdt_body_reverse_idmap(info, b);
411 if (b->valid & OBD_MD_FLSIZE)
412 CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
413 PFID(fid), (unsigned long long)b->size);
416 static inline int mdt_body_has_lov(const struct lu_attr *la,
417 const struct mdt_body *body)
419 return ((S_ISREG(la->la_mode) && (body->valid & OBD_MD_FLEASIZE)) ||
420 (S_ISDIR(la->la_mode) && (body->valid & OBD_MD_FLDIREA )) );
423 void mdt_client_compatibility(struct mdt_thread_info *info)
425 struct mdt_body *body;
426 struct ptlrpc_request *req = mdt_info_req(info);
427 struct obd_export *exp = req->rq_export;
428 struct md_attr *ma = &info->mti_attr;
429 struct lu_attr *la = &ma->ma_attr;
432 if (exp_connect_layout(exp))
433 /* the client can deal with 16-bit lmm_stripe_count */
436 body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
438 if (!mdt_body_has_lov(la, body))
441 /* now we have a reply with a lov for a client not compatible with the
442 * layout lock so we have to clean the layout generation number */
443 if (S_ISREG(la->la_mode))
444 ma->ma_lmm->lmm_layout_gen = 0;
448 static int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
451 const struct lu_env *env = info->mti_env;
455 LASSERT(info->mti_big_lmm_used == 0);
456 rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL, name);
460 /* big_lmm may need to be grown */
461 if (info->mti_big_lmmsize < rc) {
462 int size = size_roundup_power2(rc);
464 if (info->mti_big_lmmsize > 0) {
465 /* free old buffer */
466 LASSERT(info->mti_big_lmm);
467 OBD_FREE_LARGE(info->mti_big_lmm,
468 info->mti_big_lmmsize);
469 info->mti_big_lmm = NULL;
470 info->mti_big_lmmsize = 0;
473 OBD_ALLOC_LARGE(info->mti_big_lmm, size);
474 if (info->mti_big_lmm == NULL)
476 info->mti_big_lmmsize = size;
478 LASSERT(info->mti_big_lmmsize >= rc);
480 info->mti_buf.lb_buf = info->mti_big_lmm;
481 info->mti_buf.lb_len = info->mti_big_lmmsize;
482 rc = mo_xattr_get(env, mdt_object_child(o), &info->mti_buf, name);
487 int mdt_attr_get_lov(struct mdt_thread_info *info,
488 struct mdt_object *o, struct md_attr *ma)
490 struct md_object *next = mdt_object_child(o);
491 struct lu_buf *buf = &info->mti_buf;
494 buf->lb_buf = ma->ma_lmm;
495 buf->lb_len = ma->ma_lmm_size;
496 rc = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_LOV);
498 ma->ma_lmm_size = rc;
499 ma->ma_valid |= MA_LOV;
501 } else if (rc == -ENODATA) {
504 } else if (rc == -ERANGE) {
505 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LOV);
507 info->mti_big_lmm_used = 1;
508 ma->ma_valid |= MA_LOV;
509 ma->ma_lmm = info->mti_big_lmm;
510 ma->ma_lmm_size = rc;
511 /* update mdt_max_mdsize so all clients
512 * will be aware about that */
513 if (info->mti_mdt->mdt_max_mdsize < rc)
514 info->mti_mdt->mdt_max_mdsize = rc;
522 int mdt_attr_get_pfid(struct mdt_thread_info *info,
523 struct mdt_object *o, struct lu_fid *pfid)
525 struct lu_buf *buf = &info->mti_buf;
526 struct link_ea_header *leh;
527 struct link_ea_entry *lee;
531 buf->lb_buf = info->mti_big_lmm;
532 buf->lb_len = info->mti_big_lmmsize;
533 rc = mo_xattr_get(info->mti_env, mdt_object_child(o),
534 buf, XATTR_NAME_LINK);
535 /* ignore errors, MA_PFID won't be set and it is
536 * up to the caller to treat this as an error */
537 if (rc == -ERANGE || buf->lb_len == 0) {
538 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
539 buf->lb_buf = info->mti_big_lmm;
540 buf->lb_len = info->mti_big_lmmsize;
545 if (rc < sizeof(*leh)) {
546 CERROR("short LinkEA on "DFID": rc = %d\n",
547 PFID(mdt_object_fid(o)), rc);
551 leh = (struct link_ea_header *) buf->lb_buf;
552 lee = (struct link_ea_entry *)(leh + 1);
553 if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
554 leh->leh_magic = LINK_EA_MAGIC;
555 leh->leh_reccount = __swab32(leh->leh_reccount);
556 leh->leh_len = __swab64(leh->leh_len);
558 if (leh->leh_magic != LINK_EA_MAGIC)
560 if (leh->leh_reccount == 0)
563 memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
564 fid_be_to_cpu(pfid, pfid);
569 int mdt_attr_get_complex(struct mdt_thread_info *info,
570 struct mdt_object *o, struct md_attr *ma)
572 const struct lu_env *env = info->mti_env;
573 struct md_object *next = mdt_object_child(o);
574 struct lu_buf *buf = &info->mti_buf;
575 u32 mode = lu_object_attr(&next->mo_lu);
576 int need = ma->ma_need;
582 if (need & MA_INODE) {
583 ma->ma_need = MA_INODE;
584 rc = mo_attr_get(env, next, ma);
587 ma->ma_valid |= MA_INODE;
590 if (need & MA_PFID) {
591 rc = mdt_attr_get_pfid(info, o, &ma->ma_pfid);
593 ma->ma_valid |= MA_PFID;
594 /* ignore this error, parent fid is not mandatory */
598 if (need & MA_LOV && (S_ISREG(mode) || S_ISDIR(mode))) {
599 rc = mdt_attr_get_lov(info, o, ma);
604 if (need & MA_LMV && S_ISDIR(mode)) {
605 buf->lb_buf = ma->ma_lmv;
606 buf->lb_len = ma->ma_lmv_size;
607 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_LMV);
609 ma->ma_lmv_size = rc2;
610 ma->ma_valid |= MA_LMV;
611 } else if (rc2 == -ENODATA) {
618 if (need & MA_SOM && S_ISREG(mode)) {
619 buf->lb_buf = info->mti_xattr_buf;
620 buf->lb_len = sizeof(info->mti_xattr_buf);
621 CLASSERT(sizeof(struct som_attrs) <=
622 sizeof(info->mti_xattr_buf));
623 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_SOM);
624 rc2 = lustre_buf2som(info->mti_xattr_buf, rc2, ma->ma_som);
626 ma->ma_valid |= MA_SOM;
627 else if (rc2 < 0 && rc2 != -ENODATA)
631 if (need & MA_HSM && S_ISREG(mode)) {
632 buf->lb_buf = info->mti_xattr_buf;
633 buf->lb_len = sizeof(info->mti_xattr_buf);
634 CLASSERT(sizeof(struct hsm_attrs) <=
635 sizeof(info->mti_xattr_buf));
636 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_HSM);
637 rc2 = lustre_buf2hsm(info->mti_xattr_buf, rc2, &ma->ma_hsm);
639 ma->ma_valid |= MA_HSM;
640 else if (rc2 < 0 && rc2 != -ENODATA)
644 #ifdef CONFIG_FS_POSIX_ACL
645 if (need & MA_ACL_DEF && S_ISDIR(mode)) {
646 buf->lb_buf = ma->ma_acl;
647 buf->lb_len = ma->ma_acl_size;
648 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
650 ma->ma_acl_size = rc2;
651 ma->ma_valid |= MA_ACL_DEF;
652 } else if (rc2 == -ENODATA) {
661 CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = "LPX64" ma_lmm=%p\n",
662 rc, ma->ma_valid, ma->ma_lmm);
666 static int mdt_getattr_internal(struct mdt_thread_info *info,
667 struct mdt_object *o, int ma_need)
669 struct md_object *next = mdt_object_child(o);
670 const struct mdt_body *reqbody = info->mti_body;
671 struct ptlrpc_request *req = mdt_info_req(info);
672 struct md_attr *ma = &info->mti_attr;
673 struct lu_attr *la = &ma->ma_attr;
674 struct req_capsule *pill = info->mti_pill;
675 const struct lu_env *env = info->mti_env;
676 struct mdt_body *repbody;
677 struct lu_buf *buffer = &info->mti_buf;
682 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
683 RETURN(err_serious(-ENOMEM));
685 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
689 if (mdt_object_remote(o)) {
690 /* This object is located on remote node.*/
691 /* Return -EIO for old client */
692 if (!mdt_is_dne_client(req->rq_export))
693 GOTO(out, rc = -EIO);
695 repbody->fid1 = *mdt_object_fid(o);
696 repbody->valid = OBD_MD_FLID | OBD_MD_MDS;
700 buffer->lb_len = reqbody->eadatasize;
701 if (buffer->lb_len > 0)
702 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
704 buffer->lb_buf = NULL;
706 /* If it is dir object and client require MEA, then we got MEA */
707 if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
708 reqbody->valid & OBD_MD_MEA) {
709 /* Assumption: MDT_MD size is enough for lmv size. */
710 ma->ma_lmv = buffer->lb_buf;
711 ma->ma_lmv_size = buffer->lb_len;
712 ma->ma_need = MA_LMV | MA_INODE;
714 ma->ma_lmm = buffer->lb_buf;
715 ma->ma_lmm_size = buffer->lb_len;
716 ma->ma_need = MA_LOV | MA_INODE | MA_HSM;
719 if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
720 reqbody->valid & OBD_MD_FLDIREA &&
721 lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
722 /* get default stripe info for this dir. */
723 ma->ma_need |= MA_LOV_DEF;
725 ma->ma_need |= ma_need;
726 if (ma->ma_need & MA_SOM)
727 ma->ma_som = &info->mti_u.som.data;
729 rc = mdt_attr_get_complex(info, o, ma);
731 CERROR("%s: getattr error for "DFID": rc = %d\n",
732 mdt_obd_name(info->mti_mdt),
733 PFID(mdt_object_fid(o)), rc);
737 is_root = lu_fid_eq(mdt_object_fid(o), &info->mti_mdt->mdt_md_root_fid);
739 /* the Lustre protocol supposes to return default striping
740 * on the user-visible root if explicitly requested */
741 if ((ma->ma_valid & MA_LOV) == 0 && S_ISDIR(la->la_mode) &&
742 (ma->ma_need & MA_LOV_DEF && is_root) && (ma->ma_need & MA_LOV)) {
743 struct lu_fid rootfid;
744 struct mdt_object *root;
745 struct mdt_device *mdt = info->mti_mdt;
747 rc = dt_root_get(env, mdt->mdt_bottom, &rootfid);
750 root = mdt_object_find(env, mdt, &rootfid);
752 RETURN(PTR_ERR(root));
753 rc = mdt_attr_get_lov(info, root, ma);
754 mdt_object_put(info->mti_env, root);
756 CERROR("%s: getattr error for "DFID": rc = %d\n",
757 mdt_obd_name(info->mti_mdt),
758 PFID(mdt_object_fid(o)), rc);
763 if (likely(ma->ma_valid & MA_INODE))
764 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
768 if (mdt_body_has_lov(la, reqbody)) {
769 if (ma->ma_valid & MA_LOV) {
770 LASSERT(ma->ma_lmm_size);
771 mdt_dump_lmm(D_INFO, ma->ma_lmm);
772 repbody->eadatasize = ma->ma_lmm_size;
773 if (S_ISDIR(la->la_mode))
774 repbody->valid |= OBD_MD_FLDIREA;
776 repbody->valid |= OBD_MD_FLEASIZE;
778 if (ma->ma_valid & MA_LMV) {
779 LASSERT(S_ISDIR(la->la_mode));
780 repbody->eadatasize = ma->ma_lmv_size;
781 repbody->valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
783 } else if (S_ISLNK(la->la_mode) &&
784 reqbody->valid & OBD_MD_LINKNAME) {
785 buffer->lb_buf = ma->ma_lmm;
786 /* eadatasize from client includes NULL-terminator, so
787 * there is no need to read it */
788 buffer->lb_len = reqbody->eadatasize - 1;
789 rc = mo_readlink(env, next, buffer);
790 if (unlikely(rc <= 0)) {
791 CERROR("%s: readlink failed for "DFID": rc = %d\n",
792 mdt_obd_name(info->mti_mdt),
793 PFID(mdt_object_fid(o)), rc);
796 int print_limit = min_t(int, CFS_PAGE_SIZE - 128, rc);
798 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
800 repbody->valid |= OBD_MD_LINKNAME;
801 /* we need to report back size with NULL-terminator
802 * because client expects that */
803 repbody->eadatasize = rc + 1;
804 if (repbody->eadatasize != reqbody->eadatasize)
805 CDEBUG(D_INODE, "%s: Read shorter symlink %d "
806 "on "DFID ", expected %d\n",
807 mdt_obd_name(info->mti_mdt),
808 rc, PFID(mdt_object_fid(o)),
809 reqbody->eadatasize - 1);
811 ((char *)ma->ma_lmm)[rc] = 0;
813 /* If the total CDEBUG() size is larger than a page, it
814 * will print a warning to the console, avoid this by
815 * printing just the last part of the symlink. */
816 CDEBUG(D_INODE, "symlink dest %s%.*s, len = %d\n",
817 print_limit < rc ? "..." : "", print_limit,
818 (char *)ma->ma_lmm + rc - print_limit, rc);
823 if (reqbody->valid & OBD_MD_FLMODEASIZE) {
824 repbody->max_cookiesize = 0;
825 repbody->max_mdsize = info->mti_mdt->mdt_max_mdsize;
826 repbody->valid |= OBD_MD_FLMODEASIZE;
827 CDEBUG(D_INODE, "I am going to change the MAX_MD_SIZE & "
828 "MAX_COOKIE to : %d:%d\n", repbody->max_mdsize,
829 repbody->max_cookiesize);
832 if (exp_connect_rmtclient(info->mti_exp) &&
833 reqbody->valid & OBD_MD_FLRMTPERM) {
834 void *buf = req_capsule_server_get(pill, &RMF_ACL);
836 /* mdt_getattr_lock only */
837 rc = mdt_pack_remote_perm(info, o, buf);
839 repbody->valid &= ~OBD_MD_FLRMTPERM;
840 repbody->aclsize = 0;
843 repbody->valid |= OBD_MD_FLRMTPERM;
844 repbody->aclsize = sizeof(struct mdt_remote_perm);
847 #ifdef CONFIG_FS_POSIX_ACL
848 else if ((exp_connect_flags(req->rq_export) & OBD_CONNECT_ACL) &&
849 (reqbody->valid & OBD_MD_FLACL)) {
850 buffer->lb_buf = req_capsule_server_get(pill, &RMF_ACL);
851 buffer->lb_len = req_capsule_get_size(pill,
852 &RMF_ACL, RCL_SERVER);
853 if (buffer->lb_len > 0) {
854 rc = mo_xattr_get(env, next, buffer,
855 XATTR_NAME_ACL_ACCESS);
857 if (rc == -ENODATA) {
858 repbody->aclsize = 0;
859 repbody->valid |= OBD_MD_FLACL;
861 } else if (rc == -EOPNOTSUPP) {
864 CERROR("%s: unable to read "DFID
866 mdt_obd_name(info->mti_mdt),
867 PFID(mdt_object_fid(o)), rc);
870 repbody->aclsize = rc;
871 repbody->valid |= OBD_MD_FLACL;
878 if (reqbody->valid & OBD_MD_FLMDSCAPA &&
879 info->mti_mdt->mdt_opts.mo_mds_capa &&
880 exp_connect_flags(info->mti_exp) & OBD_CONNECT_MDS_CAPA) {
881 struct lustre_capa *capa;
883 capa = req_capsule_server_get(pill, &RMF_CAPA1);
885 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
886 rc = mo_capa_get(env, next, capa, 0);
889 repbody->valid |= OBD_MD_FLMDSCAPA;
894 mdt_counter_incr(req, LPROC_MDT_GETATTR);
899 static int mdt_renew_capa(struct mdt_thread_info *info)
901 struct mdt_object *obj = info->mti_object;
902 struct mdt_body *body;
903 struct lustre_capa *capa, *c;
907 /* if object doesn't exist, or server has disabled capability,
908 * return directly, client will find body->valid OBD_MD_FLOSSCAPA
911 if (!obj || !info->mti_mdt->mdt_opts.mo_oss_capa ||
912 !(exp_connect_flags(info->mti_exp) & OBD_CONNECT_OSS_CAPA))
915 body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
916 LASSERT(body != NULL);
918 c = req_capsule_client_get(info->mti_pill, &RMF_CAPA1);
921 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
925 rc = mo_capa_get(info->mti_env, mdt_object_child(obj), capa, 1);
927 body->valid |= OBD_MD_FLOSSCAPA;
931 int mdt_getattr(struct mdt_thread_info *info)
933 struct mdt_object *obj = info->mti_object;
934 struct req_capsule *pill = info->mti_pill;
935 struct mdt_body *reqbody;
936 struct mdt_body *repbody;
941 reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
944 if (reqbody->valid & OBD_MD_FLOSSCAPA) {
945 rc = req_capsule_server_pack(pill);
947 RETURN(err_serious(rc));
948 rc = mdt_renew_capa(info);
949 GOTO(out_shrink, rc);
952 LASSERT(obj != NULL);
953 LASSERT(lu_object_assert_exists(&obj->mot_obj.mo_lu));
955 mode = lu_object_attr(&obj->mot_obj.mo_lu);
957 /* old clients may not report needed easize, use max value then */
958 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
959 reqbody->eadatasize == 0 ?
960 info->mti_mdt->mdt_max_mdsize :
961 reqbody->eadatasize);
963 rc = req_capsule_server_pack(pill);
964 if (unlikely(rc != 0))
965 RETURN(err_serious(rc));
967 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
968 LASSERT(repbody != NULL);
969 repbody->eadatasize = 0;
970 repbody->aclsize = 0;
972 if (reqbody->valid & OBD_MD_FLRMTPERM)
973 rc = mdt_init_ucred(info, reqbody);
975 rc = mdt_check_ucred(info);
977 GOTO(out_shrink, rc);
979 info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
982 * Don't check capability at all, because rename might getattr for
983 * remote obj, and at that time no capability is available.
985 mdt_set_capainfo(info, 1, &reqbody->fid1, BYPASS_CAPA);
986 rc = mdt_getattr_internal(info, obj, 0);
987 if (reqbody->valid & OBD_MD_FLRMTPERM)
988 mdt_exit_ucred(info);
991 mdt_client_compatibility(info);
992 rc2 = mdt_fix_reply(info);
998 int mdt_is_subdir(struct mdt_thread_info *info)
1000 struct mdt_object *o = info->mti_object;
1001 struct req_capsule *pill = info->mti_pill;
1002 const struct mdt_body *body = info->mti_body;
1003 struct mdt_body *repbody;
1009 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1012 * We save last checked parent fid to @repbody->fid1 for remote
1015 LASSERT(fid_is_sane(&body->fid2));
1016 LASSERT(mdt_object_exists(o) && !mdt_object_remote(o));
1017 rc = mdo_is_subdir(info->mti_env, mdt_object_child(o),
1018 &body->fid2, &repbody->fid1);
1019 if (rc == 0 || rc == -EREMOTE)
1020 repbody->valid |= OBD_MD_FLID;
1025 int mdt_swap_layouts(struct mdt_thread_info *info)
1027 struct ptlrpc_request *req = mdt_info_req(info);
1028 struct obd_export *exp = req->rq_export;
1029 struct mdt_object *o1, *o2, *o;
1030 struct mdt_lock_handle *lh1, *lh2;
1031 struct mdc_swap_layouts *msl;
1035 /* client does not support layout lock, so layout swaping
1037 * FIXME: there is a problem for old clients which don't support
1038 * layout lock yet. If those clients have already opened the file
1039 * they won't be notified at all so that old layout may still be
1040 * used to do IO. This can be fixed after file release is landed by
1041 * doing exclusive open and taking full EX ibits lock. - Jinshan */
1042 if (!exp_connect_layout(exp))
1043 RETURN(-EOPNOTSUPP);
1045 if (req_capsule_get_size(info->mti_pill, &RMF_CAPA1, RCL_CLIENT))
1046 mdt_set_capainfo(info, 0, &info->mti_body->fid1,
1047 req_capsule_client_get(info->mti_pill,
1050 if (req_capsule_get_size(info->mti_pill, &RMF_CAPA2, RCL_CLIENT))
1051 mdt_set_capainfo(info, 1, &info->mti_body->fid2,
1052 req_capsule_client_get(info->mti_pill,
1055 o1 = info->mti_object;
1056 o = o2 = mdt_object_find(info->mti_env, info->mti_mdt,
1057 &info->mti_body->fid2);
1059 GOTO(out, rc = PTR_ERR(o));
1061 if (mdt_object_remote(o) || !mdt_object_exists(o)) /* remote object */
1062 GOTO(put, rc = -ENOENT);
1064 rc = lu_fid_cmp(&info->mti_body->fid1, &info->mti_body->fid2);
1065 if (unlikely(rc == 0)) /* same file, you kidding me? no-op. */
1071 /* permission check. Make sure the calling process having permission
1072 * to write both files. */
1073 rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
1078 rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL,
1083 msl = req_capsule_client_get(info->mti_pill, &RMF_SWAP_LAYOUTS);
1085 GOTO(put, rc = -EPROTO);
1087 lh1 = &info->mti_lh[MDT_LH_NEW];
1088 mdt_lock_reg_init(lh1, LCK_EX);
1089 lh2 = &info->mti_lh[MDT_LH_OLD];
1090 mdt_lock_reg_init(lh2, LCK_EX);
1092 rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT,
1097 rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT,
1102 rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1),
1103 mdt_object_child(o2), msl->msl_flags);
1106 mdt_object_unlock(info, o2, lh2, rc);
1108 mdt_object_unlock(info, o1, lh1, rc);
1110 mdt_object_put(info->mti_env, o);
1115 static int mdt_raw_lookup(struct mdt_thread_info *info,
1116 struct mdt_object *parent,
1117 const struct lu_name *lname,
1118 struct ldlm_reply *ldlm_rep)
1120 struct md_object *next = mdt_object_child(info->mti_object);
1121 const struct mdt_body *reqbody = info->mti_body;
1122 struct lu_fid *child_fid = &info->mti_tmp_fid1;
1123 struct mdt_body *repbody;
1127 if (reqbody->valid != OBD_MD_FLID)
1130 LASSERT(!info->mti_cross_ref);
1132 /* Only got the fid of this obj by name */
1133 fid_zero(child_fid);
1134 rc = mdo_lookup(info->mti_env, next, lname, child_fid,
1137 /* XXX is raw_lookup possible as intent operation? */
1140 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1143 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1145 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1148 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1149 repbody->fid1 = *child_fid;
1150 repbody->valid = OBD_MD_FLID;
1156 * UPDATE lock should be taken against parent, and be release before exit;
1157 * child_bits lock should be taken against child, and be returned back:
1158 * (1)normal request should release the child lock;
1159 * (2)intent request will grant the lock to client.
1161 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
1162 struct mdt_lock_handle *lhc,
1164 struct ldlm_reply *ldlm_rep)
1166 struct ptlrpc_request *req = mdt_info_req(info);
1167 struct mdt_body *reqbody = NULL;
1168 struct mdt_object *parent = info->mti_object;
1169 struct mdt_object *child;
1170 struct md_object *next = mdt_object_child(parent);
1171 struct lu_fid *child_fid = &info->mti_tmp_fid1;
1172 struct lu_name *lname = NULL;
1173 const char *name = NULL;
1175 struct mdt_lock_handle *lhp = NULL;
1176 struct ldlm_lock *lock;
1177 struct ldlm_res_id *res_id;
1184 is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
1185 LASSERT(ergo(is_resent,
1186 lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
1188 LASSERT(parent != NULL);
1189 name = req_capsule_client_get(info->mti_pill, &RMF_NAME);
1191 RETURN(err_serious(-EFAULT));
1193 namelen = req_capsule_get_size(info->mti_pill, &RMF_NAME,
1195 if (!info->mti_cross_ref) {
1197 * XXX: Check for "namelen == 0" is for getattr by fid
1198 * (OBD_CONNECT_ATTRFID), otherwise do not allow empty name,
1199 * that is the name must contain at least one character and
1200 * the terminating '\0'
1203 reqbody = req_capsule_client_get(info->mti_pill,
1205 if (unlikely(reqbody == NULL))
1206 RETURN(err_serious(-EFAULT));
1208 if (unlikely(!fid_is_sane(&reqbody->fid2)))
1209 RETURN(err_serious(-EINVAL));
1212 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
1214 PFID(mdt_object_fid(parent)),
1215 PFID(&reqbody->fid2), ldlm_rep);
1217 lname = mdt_name(info->mti_env, (char *)name, namelen);
1218 CDEBUG(D_INODE, "getattr with lock for "DFID"/%s, "
1219 "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
1223 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
1225 if (unlikely(!mdt_object_exists(parent))) {
1226 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1227 &parent->mot_obj.mo_lu,
1228 "Parent doesn't exist!\n");
1230 } else if (!info->mti_cross_ref) {
1231 LASSERTF(!mdt_object_remote(parent),
1232 "Parent "DFID" is on remote server\n",
1233 PFID(mdt_object_fid(parent)));
1236 rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
1244 if (info->mti_cross_ref) {
1245 /* Only getattr on the child. Parent is on another node. */
1246 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1248 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
1249 "ldlm_rep=%p\n", PFID(mdt_object_fid(child)), ldlm_rep);
1252 /* Do not take lock for resent case. */
1253 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1254 LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
1255 lhc->mlh_reg_lh.cookie);
1256 LASSERT(fid_res_name_eq(mdt_object_fid(child),
1257 &lock->l_resource->lr_name));
1258 LDLM_LOCK_PUT(lock);
1261 mdt_lock_handle_init(lhc);
1262 mdt_lock_reg_init(lhc, LCK_PR);
1265 * Object's name is on another MDS, no lookup lock is
1266 * needed here but update is.
1268 child_bits &= ~MDS_INODELOCK_LOOKUP;
1269 child_bits |= MDS_INODELOCK_PERM | MDS_INODELOCK_UPDATE;
1271 rc = mdt_object_lock(info, child, lhc, child_bits,
1275 /* Finally, we can get attr for child. */
1276 mdt_set_capainfo(info, 0, mdt_object_fid(child),
1278 rc = mdt_getattr_internal(info, child, 0);
1279 if (unlikely(rc != 0))
1280 mdt_object_unlock(info, child, lhc, 1);
1286 /* step 1: lock parent only if parent is a directory */
1287 if (S_ISDIR(lu_object_attr(&parent->mot_obj.mo_lu))) {
1288 lhp = &info->mti_lh[MDT_LH_PARENT];
1289 mdt_lock_pdo_init(lhp, LCK_PR, name, namelen);
1290 rc = mdt_object_lock(info, parent, lhp,
1291 MDS_INODELOCK_UPDATE,
1293 if (unlikely(rc != 0))
1297 /* step 2: lookup child's fid by name */
1298 fid_zero(child_fid);
1299 rc = mdo_lookup(info->mti_env, next, lname, child_fid,
1304 mdt_set_disposition(info, ldlm_rep,
1306 GOTO(out_parent, rc);
1308 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1310 *child_fid = reqbody->fid2;
1311 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1315 *step 3: find the child object by fid & lock it.
1316 * regardless if it is local or remote.
1318 child = mdt_object_find(info->mti_env, info->mti_mdt, child_fid);
1320 if (unlikely(IS_ERR(child)))
1321 GOTO(out_parent, rc = PTR_ERR(child));
1323 /* Do not take lock for resent case. */
1324 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1325 LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
1326 lhc->mlh_reg_lh.cookie);
1328 res_id = &lock->l_resource->lr_name;
1329 if (!fid_res_name_eq(mdt_object_fid(child),
1330 &lock->l_resource->lr_name)) {
1331 LASSERTF(fid_res_name_eq(mdt_object_fid(parent),
1332 &lock->l_resource->lr_name),
1333 "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n",
1334 (unsigned long)res_id->name[0],
1335 (unsigned long)res_id->name[1],
1336 (unsigned long)res_id->name[2],
1337 PFID(mdt_object_fid(parent)));
1338 CWARN("Although resent, but still not get child lock"
1339 "parent:"DFID" child:"DFID"\n",
1340 PFID(mdt_object_fid(parent)),
1341 PFID(mdt_object_fid(child)));
1342 lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
1343 LDLM_LOCK_PUT(lock);
1346 LDLM_LOCK_PUT(lock);
1349 bool try_layout = false;
1352 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout*2);
1353 mdt_lock_handle_init(lhc);
1354 mdt_lock_reg_init(lhc, LCK_PR);
1356 if (!mdt_object_exists(child)) {
1357 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1358 &child->mot_obj.mo_lu,
1359 "Object doesn't exist!\n");
1360 GOTO(out_child, rc = -ENOENT);
1363 if (!(child_bits & MDS_INODELOCK_UPDATE) &&
1364 mdt_object_exists(child) && !mdt_object_remote(child)) {
1365 struct md_attr *ma = &info->mti_attr;
1368 ma->ma_need = MA_INODE;
1369 rc = mdt_attr_get_complex(info, child, ma);
1370 if (unlikely(rc != 0))
1371 GOTO(out_child, rc);
1373 /* If the file has not been changed for some time, we
1374 * return not only a LOOKUP lock, but also an UPDATE
1375 * lock and this might save us RPC on later STAT. For
1376 * directories, it also let negative dentry starts
1377 * working for this dir. */
1378 if (ma->ma_valid & MA_INODE &&
1379 ma->ma_attr.la_valid & LA_CTIME &&
1380 info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
1381 ma->ma_attr.la_ctime < cfs_time_current_sec())
1382 child_bits |= MDS_INODELOCK_UPDATE;
1385 /* layout lock must be granted in a best-effort way
1386 * for IT operations */
1387 LASSERT(!(child_bits & MDS_INODELOCK_LAYOUT));
1388 if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) &&
1389 exp_connect_layout(info->mti_exp) &&
1390 S_ISREG(lu_object_attr(&child->mot_obj.mo_lu)) &&
1392 /* try to grant layout lock for regular file. */
1398 child_bits |= MDS_INODELOCK_LAYOUT;
1399 /* try layout lock, it may fail to be granted due to
1400 * contention at LOOKUP or UPDATE */
1401 if (!mdt_object_lock_try(info, child, lhc, child_bits,
1403 child_bits &= ~MDS_INODELOCK_LAYOUT;
1404 LASSERT(child_bits != 0);
1405 rc = mdt_object_lock(info, child, lhc,
1406 child_bits, MDT_CROSS_LOCK);
1411 rc = mdt_object_lock(info, child, lhc, child_bits,
1414 if (unlikely(rc != 0))
1415 GOTO(out_child, rc);
1418 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1419 /* Get MA_SOM attributes if update lock is given. */
1421 lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_UPDATE &&
1422 S_ISREG(lu_object_attr(&mdt_object_child(child)->mo_lu)))
1425 /* finally, we can get attr for child. */
1426 mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
1427 rc = mdt_getattr_internal(info, child, ma_need);
1428 if (unlikely(rc != 0)) {
1429 mdt_object_unlock(info, child, lhc, 1);
1431 /* Debugging code. */
1432 res_id = &lock->l_resource->lr_name;
1433 LDLM_DEBUG(lock, "Returning lock to client");
1434 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1435 &lock->l_resource->lr_name),
1436 "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n",
1437 (unsigned long)res_id->name[0],
1438 (unsigned long)res_id->name[1],
1439 (unsigned long)res_id->name[2],
1440 PFID(mdt_object_fid(child)));
1441 if (mdt_object_exists(child) && !mdt_object_remote(child))
1442 mdt_pack_size2body(info, child);
1445 LDLM_LOCK_PUT(lock);
1449 mdt_object_put(info->mti_env, child);
1452 mdt_object_unlock(info, parent, lhp, 1);
1456 /* normal handler: should release the child lock */
1457 int mdt_getattr_name(struct mdt_thread_info *info)
1459 struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1460 struct mdt_body *reqbody;
1461 struct mdt_body *repbody;
1465 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1466 LASSERT(reqbody != NULL);
1467 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1468 LASSERT(repbody != NULL);
1470 info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
1471 repbody->eadatasize = 0;
1472 repbody->aclsize = 0;
1474 rc = mdt_init_ucred(info, reqbody);
1476 GOTO(out_shrink, rc);
1478 rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1479 if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1480 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1481 lhc->mlh_reg_lh.cookie = 0;
1483 mdt_exit_ucred(info);
1486 mdt_client_compatibility(info);
1487 rc2 = mdt_fix_reply(info);
1493 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1494 void *karg, void *uarg);
1496 int mdt_set_info(struct mdt_thread_info *info)
1498 struct ptlrpc_request *req = mdt_info_req(info);
1501 int keylen, vallen, rc = 0;
1504 rc = req_capsule_server_pack(info->mti_pill);
1508 key = req_capsule_client_get(info->mti_pill, &RMF_SETINFO_KEY);
1510 DEBUG_REQ(D_HA, req, "no set_info key");
1514 keylen = req_capsule_get_size(info->mti_pill, &RMF_SETINFO_KEY,
1517 val = req_capsule_client_get(info->mti_pill, &RMF_SETINFO_VAL);
1519 DEBUG_REQ(D_HA, req, "no set_info val");
1523 vallen = req_capsule_get_size(info->mti_pill, &RMF_SETINFO_VAL,
1526 /* Swab any part of val you need to here */
1527 if (KEY_IS(KEY_READ_ONLY)) {
1529 lustre_msg_set_status(req->rq_repmsg, 0);
1531 spin_lock(&req->rq_export->exp_lock);
1533 *exp_connect_flags_ptr(req->rq_export) |=
1536 *exp_connect_flags_ptr(req->rq_export) &=
1537 ~OBD_CONNECT_RDONLY;
1538 spin_unlock(&req->rq_export->exp_lock);
1540 } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1541 struct changelog_setinfo *cs =
1542 (struct changelog_setinfo *)val;
1543 if (vallen != sizeof(*cs)) {
1544 CERROR("Bad changelog_clear setinfo size %d\n", vallen);
1547 if (ptlrpc_req_need_swab(req)) {
1548 __swab64s(&cs->cs_recno);
1549 __swab32s(&cs->cs_id);
1552 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, info->mti_exp,
1554 lustre_msg_set_status(req->rq_repmsg, rc);
1563 * Top-level handler for MDT connection requests.
1565 int mdt_connect(struct mdt_thread_info *info)
1568 struct obd_connect_data *reply;
1569 struct obd_export *exp;
1570 struct ptlrpc_request *req = mdt_info_req(info);
1572 rc = target_handle_connect(req);
1574 return err_serious(rc);
1576 LASSERT(req->rq_export != NULL);
1577 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
1578 rc = mdt_init_sec_level(info);
1580 obd_disconnect(class_export_get(req->rq_export));
1584 /* To avoid exposing partially initialized connection flags, changes up
1585 * to this point have been staged in reply->ocd_connect_flags. Now that
1586 * connection handling has completed successfully, atomically update
1587 * the connect flags in the shared export data structure. LU-1623 */
1588 reply = req_capsule_server_get(info->mti_pill, &RMF_CONNECT_DATA);
1589 exp = req->rq_export;
1590 spin_lock(&exp->exp_lock);
1591 *exp_connect_flags_ptr(exp) = reply->ocd_connect_flags;
1592 spin_unlock(&exp->exp_lock);
1594 rc = mdt_init_idmap(info);
1596 obd_disconnect(class_export_get(req->rq_export));
1601 int mdt_disconnect(struct mdt_thread_info *info)
1606 rc = target_handle_disconnect(mdt_info_req(info));
1608 rc = err_serious(rc);
1612 static int mdt_sendpage(struct mdt_thread_info *info,
1613 struct lu_rdpg *rdpg, int nob)
1615 struct ptlrpc_request *req = mdt_info_req(info);
1616 struct obd_export *exp = req->rq_export;
1617 struct ptlrpc_bulk_desc *desc;
1618 struct l_wait_info *lwi = &info->mti_u.rdpg.mti_wait_info;
1625 desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, 1, BULK_PUT_SOURCE,
1630 if (!(exp_connect_flags(exp) & OBD_CONNECT_BRW_SIZE))
1631 /* old client requires reply size in it's PAGE_SIZE,
1632 * which is rdpg->rp_count */
1633 nob = rdpg->rp_count;
1635 for (i = 0, tmpcount = nob; i < rdpg->rp_npages && tmpcount > 0;
1636 i++, tmpcount -= tmpsize) {
1637 tmpsize = min_t(int, tmpcount, CFS_PAGE_SIZE);
1638 ptlrpc_prep_bulk_page_pin(desc, rdpg->rp_pages[i], 0, tmpsize);
1641 LASSERT(desc->bd_nob == nob);
1642 rc = target_bulk_io(exp, desc, lwi);
1643 ptlrpc_free_bulk_pin(desc);
1647 int mdt_readpage(struct mdt_thread_info *info)
1649 struct mdt_object *object = info->mti_object;
1650 struct lu_rdpg *rdpg = &info->mti_u.rdpg.mti_rdpg;
1651 struct mdt_body *reqbody;
1652 struct mdt_body *repbody;
1657 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
1658 RETURN(err_serious(-ENOMEM));
1660 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1661 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1662 if (reqbody == NULL || repbody == NULL)
1663 RETURN(err_serious(-EFAULT));
1666 * prepare @rdpg before calling lower layers and transfer itself. Here
1667 * reqbody->size contains offset of where to start to read and
1668 * reqbody->nlink contains number bytes to read.
1670 rdpg->rp_hash = reqbody->size;
1671 if (rdpg->rp_hash != reqbody->size) {
1672 CERROR("Invalid hash: "LPX64" != "LPX64"\n",
1673 rdpg->rp_hash, reqbody->size);
1677 rdpg->rp_attrs = reqbody->mode;
1678 if (exp_connect_flags(info->mti_exp) & OBD_CONNECT_64BITHASH)
1679 rdpg->rp_attrs |= LUDA_64BITHASH;
1680 rdpg->rp_count = min_t(unsigned int, reqbody->nlink,
1681 exp_max_brw_size(info->mti_exp));
1682 rdpg->rp_npages = (rdpg->rp_count + CFS_PAGE_SIZE - 1) >>
1684 OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1685 if (rdpg->rp_pages == NULL)
1688 for (i = 0; i < rdpg->rp_npages; ++i) {
1689 rdpg->rp_pages[i] = cfs_alloc_page(CFS_ALLOC_STD);
1690 if (rdpg->rp_pages[i] == NULL)
1691 GOTO(free_rdpg, rc = -ENOMEM);
1694 /* call lower layers to fill allocated pages with directory data */
1695 rc = mo_readpage(info->mti_env, mdt_object_child(object), rdpg);
1697 GOTO(free_rdpg, rc);
1699 /* send pages to client */
1700 rc = mdt_sendpage(info, rdpg, rc);
1705 for (i = 0; i < rdpg->rp_npages; i++)
1706 if (rdpg->rp_pages[i] != NULL)
1707 cfs_free_page(rdpg->rp_pages[i]);
1708 OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1710 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1716 static int mdt_reint_internal(struct mdt_thread_info *info,
1717 struct mdt_lock_handle *lhc,
1720 struct req_capsule *pill = info->mti_pill;
1721 struct mdt_body *repbody;
1726 rc = mdt_reint_unpack(info, op);
1728 CERROR("Can't unpack reint, rc %d\n", rc);
1729 RETURN(err_serious(rc));
1732 /* for replay (no_create) lmm is not needed, client has it already */
1733 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1734 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1735 info->mti_rr.rr_eadatalen);
1737 /* llog cookies are always 0, the field is kept for compatibility */
1738 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1739 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);
1741 rc = req_capsule_server_pack(pill);
1743 CERROR("Can't pack response, rc %d\n", rc);
1744 RETURN(err_serious(rc));
1747 if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
1748 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1750 repbody->eadatasize = 0;
1751 repbody->aclsize = 0;
1754 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
1756 /* for replay no cookkie / lmm need, because client have this already */
1757 if (info->mti_spec.no_create)
1758 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1759 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
1761 rc = mdt_init_ucred_reint(info);
1763 GOTO(out_shrink, rc);
1765 rc = mdt_fix_attr_ucred(info, op);
1767 GOTO(out_ucred, rc = err_serious(rc));
1769 if (mdt_check_resent(info, mdt_reconstruct, lhc)) {
1770 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
1771 GOTO(out_ucred, rc);
1773 rc = mdt_reint_rec(info, lhc);
1776 mdt_exit_ucred(info);
1778 mdt_client_compatibility(info);
1779 rc2 = mdt_fix_reply(info);
1785 static long mdt_reint_opcode(struct mdt_thread_info *info,
1786 const struct req_format **fmt)
1788 struct mdt_rec_reint *rec;
1791 rec = req_capsule_client_get(info->mti_pill, &RMF_REC_REINT);
1793 opc = rec->rr_opcode;
1794 DEBUG_REQ(D_INODE, mdt_info_req(info), "reint opt = %ld", opc);
1795 if (opc < REINT_MAX && fmt[opc] != NULL)
1796 req_capsule_extend(info->mti_pill, fmt[opc]);
1798 CERROR("%s: Unsupported opcode '%ld' from client '%s': "
1799 "rc = %d\n", mdt_obd_name(info->mti_mdt), opc,
1800 info->mti_mdt->mdt_ldlm_client->cli_name,
1802 opc = err_serious(-EFAULT);
1805 opc = err_serious(-EFAULT);
1810 int mdt_reint(struct mdt_thread_info *info)
1815 static const struct req_format *reint_fmts[REINT_MAX] = {
1816 [REINT_SETATTR] = &RQF_MDS_REINT_SETATTR,
1817 [REINT_CREATE] = &RQF_MDS_REINT_CREATE,
1818 [REINT_LINK] = &RQF_MDS_REINT_LINK,
1819 [REINT_UNLINK] = &RQF_MDS_REINT_UNLINK,
1820 [REINT_RENAME] = &RQF_MDS_REINT_RENAME,
1821 [REINT_OPEN] = &RQF_MDS_REINT_OPEN,
1822 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR,
1823 [REINT_RMENTRY] = &RQF_MDS_REINT_UNLINK
1828 opc = mdt_reint_opcode(info, reint_fmts);
1831 * No lock possible here from client to pass it to reint code
1834 rc = mdt_reint_internal(info, NULL, opc);
1839 info->mti_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
1843 /* this should sync the whole device */
1844 static int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
1846 struct dt_device *dt = mdt->mdt_bottom;
1850 rc = dt->dd_ops->dt_sync(env, dt);
1854 /* this should sync this object */
1855 static int mdt_object_sync(struct mdt_thread_info *info)
1857 struct md_object *next;
1861 if (!mdt_object_exists(info->mti_object)) {
1862 CWARN("Non existing object "DFID"!\n",
1863 PFID(mdt_object_fid(info->mti_object)));
1866 next = mdt_object_child(info->mti_object);
1867 rc = mo_object_sync(info->mti_env, next);
1872 int mdt_sync(struct mdt_thread_info *info)
1874 struct ptlrpc_request *req = mdt_info_req(info);
1875 struct req_capsule *pill = info->mti_pill;
1876 struct mdt_body *body;
1880 /* The fid may be zero, so we req_capsule_set manually */
1881 req_capsule_set(pill, &RQF_MDS_SYNC);
1883 body = req_capsule_client_get(pill, &RMF_MDT_BODY);
1885 RETURN(err_serious(-EINVAL));
1887 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
1888 RETURN(err_serious(-ENOMEM));
1890 if (fid_seq(&body->fid1) == 0) {
1891 /* sync the whole device */
1892 rc = req_capsule_server_pack(pill);
1894 rc = mdt_device_sync(info->mti_env, info->mti_mdt);
1896 rc = err_serious(rc);
1898 /* sync an object */
1899 rc = mdt_unpack_req_pack_rep(info, HABEO_CORPUS|HABEO_REFERO);
1901 rc = mdt_object_sync(info);
1903 const struct lu_fid *fid;
1904 struct lu_attr *la = &info->mti_attr.ma_attr;
1906 info->mti_attr.ma_need = MA_INODE;
1907 info->mti_attr.ma_valid = 0;
1908 rc = mdt_attr_get_complex(info, info->mti_object,
1911 body = req_capsule_server_get(pill,
1913 fid = mdt_object_fid(info->mti_object);
1914 mdt_pack_attr2body(info, body, la, fid);
1918 rc = err_serious(rc);
1921 mdt_counter_incr(req, LPROC_MDT_SYNC);
1927 * Quotacheck handler.
1928 * in-kernel quotacheck isn't supported any more.
1930 int mdt_quotacheck(struct mdt_thread_info *info)
1932 struct obd_quotactl *oqctl;
1936 oqctl = req_capsule_client_get(info->mti_pill, &RMF_OBD_QUOTACTL);
1938 RETURN(err_serious(-EPROTO));
1940 rc = req_capsule_server_pack(info->mti_pill);
1942 RETURN(err_serious(rc));
1944 /* deprecated, not used any more */
1945 RETURN(-EOPNOTSUPP);
1949 * Handle quota control requests to consult current usage/limit, but also
1950 * to configure quota enforcement
1952 int mdt_quotactl(struct mdt_thread_info *info)
1954 struct obd_export *exp = info->mti_exp;
1955 struct req_capsule *pill = info->mti_pill;
1956 struct obd_quotactl *oqctl, *repoqc;
1958 struct lu_device *qmt = info->mti_mdt->mdt_qmt_dev;
1961 oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
1963 RETURN(err_serious(-EPROTO));
1965 rc = req_capsule_server_pack(pill);
1967 RETURN(err_serious(rc));
1969 switch (oqctl->qc_cmd) {
1971 case LUSTRE_Q_INVALIDATE:
1972 case LUSTRE_Q_FINVALIDATE:
1976 /* deprecated, not used any more */
1977 RETURN(-EOPNOTSUPP);
1978 /* master quotactl */
1984 RETURN(-EOPNOTSUPP);
1985 /* slave quotactl */
1990 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
1994 /* map uid/gid for remote client */
1996 if (exp_connect_rmtclient(exp)) {
1997 struct lustre_idmap_table *idmap;
1999 idmap = mdt_req2med(mdt_info_req(info))->med_idmap;
2001 if (unlikely(oqctl->qc_cmd != Q_GETQUOTA &&
2002 oqctl->qc_cmd != Q_GETINFO))
2005 if (oqctl->qc_type == USRQUOTA)
2006 id = lustre_idmap_lookup_uid(NULL, idmap, 0,
2008 else if (oqctl->qc_type == GRPQUOTA)
2009 id = lustre_idmap_lookup_gid(NULL, idmap, 0,
2014 if (id == CFS_IDMAP_NOTFOUND) {
2015 CDEBUG(D_QUOTA, "no mapping for id %u\n", oqctl->qc_id);
2020 repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
2022 RETURN(err_serious(-EFAULT));
2024 if (oqctl->qc_id != id)
2025 swap(oqctl->qc_id, id);
2027 switch (oqctl->qc_cmd) {
2033 /* forward quotactl request to QMT */
2034 rc = qmt_hdls.qmth_quotactl(info->mti_env, qmt, oqctl);
2039 /* slave quotactl */
2040 rc = lquotactl_slv(info->mti_env, info->mti_mdt->mdt_bottom,
2045 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2049 if (oqctl->qc_id != id)
2050 swap(oqctl->qc_id, id);
2057 * OBD PING and other handlers.
2059 int mdt_obd_ping(struct mdt_thread_info *info)
2064 req_capsule_set(info->mti_pill, &RQF_OBD_PING);
2066 rc = target_handle_ping(mdt_info_req(info));
2068 rc = err_serious(rc);
2073 * OBD_IDX_READ handler
2075 int mdt_obd_idx_read(struct mdt_thread_info *info)
2077 struct mdt_device *mdt = info->mti_mdt;
2078 struct lu_rdpg *rdpg = &info->mti_u.rdpg.mti_rdpg;
2079 struct idx_info *req_ii, *rep_ii;
2083 memset(rdpg, 0, sizeof(*rdpg));
2084 req_capsule_set(info->mti_pill, &RQF_OBD_IDX_READ);
2086 /* extract idx_info buffer from request & reply */
2087 req_ii = req_capsule_client_get(info->mti_pill, &RMF_IDX_INFO);
2088 if (req_ii == NULL || req_ii->ii_magic != IDX_INFO_MAGIC)
2089 RETURN(err_serious(-EPROTO));
2091 rc = req_capsule_server_pack(info->mti_pill);
2093 RETURN(err_serious(rc));
2095 rep_ii = req_capsule_server_get(info->mti_pill, &RMF_IDX_INFO);
2097 RETURN(err_serious(-EFAULT));
2098 rep_ii->ii_magic = IDX_INFO_MAGIC;
2100 /* extract hash to start with */
2101 rdpg->rp_hash = req_ii->ii_hash_start;
2103 /* extract requested attributes */
2104 rdpg->rp_attrs = req_ii->ii_attrs;
2106 /* check that fid packed in request is valid and supported */
2107 if (!fid_is_sane(&req_ii->ii_fid))
2109 rep_ii->ii_fid = req_ii->ii_fid;
2112 rep_ii->ii_flags = req_ii->ii_flags;
2114 /* compute number of pages to allocate, ii_count is the number of 4KB
2116 if (req_ii->ii_count <= 0)
2117 GOTO(out, rc = -EFAULT);
2118 rdpg->rp_count = min_t(unsigned int, req_ii->ii_count << LU_PAGE_SHIFT,
2119 exp_max_brw_size(info->mti_exp));
2120 rdpg->rp_npages = (rdpg->rp_count + CFS_PAGE_SIZE -1) >> CFS_PAGE_SHIFT;
2122 /* allocate pages to store the containers */
2123 OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
2124 if (rdpg->rp_pages == NULL)
2125 GOTO(out, rc = -ENOMEM);
2126 for (i = 0; i < rdpg->rp_npages; i++) {
2127 rdpg->rp_pages[i] = cfs_alloc_page(CFS_ALLOC_STD);
2128 if (rdpg->rp_pages[i] == NULL)
2129 GOTO(out, rc = -ENOMEM);
2132 /* populate pages with key/record pairs */
2133 rc = dt_index_read(info->mti_env, mdt->mdt_bottom, rep_ii, rdpg);
2137 LASSERTF(rc <= rdpg->rp_count, "dt_index_read() returned more than "
2138 "asked %d > %d\n", rc, rdpg->rp_count);
2140 /* send pages to client */
2141 rc = mdt_sendpage(info, rdpg, rc);
2145 if (rdpg->rp_pages) {
2146 for (i = 0; i < rdpg->rp_npages; i++)
2147 if (rdpg->rp_pages[i])
2148 cfs_free_page(rdpg->rp_pages[i]);
2149 OBD_FREE(rdpg->rp_pages,
2150 rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
2155 int mdt_obd_log_cancel(struct mdt_thread_info *info)
2157 return err_serious(-EOPNOTSUPP);
2160 int mdt_obd_qc_callback(struct mdt_thread_info *info)
2162 return err_serious(-EOPNOTSUPP);
2169 /** clone llog ctxt from child (mdd)
2170 * This allows remote llog (replicator) access.
2171 * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
2172 * context was originally set up, or we can handle them directly.
2173 * I choose the latter, but that means I need any llog
2174 * contexts set up by child to be accessable by the mdt. So we clone the
2175 * context into our context list here.
2177 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
2180 struct md_device *next = mdt->mdt_child;
2181 struct llog_ctxt *ctxt;
2184 if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
2187 rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
2188 if (rc || ctxt == NULL) {
2192 rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
2194 CERROR("Can't set mdt ctxt %d\n", rc);
2199 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
2200 struct mdt_device *mdt, int idx)
2202 struct llog_ctxt *ctxt;
2204 ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
2207 /* Put once for the get we just did, and once for the clone */
2208 llog_ctxt_put(ctxt);
2209 llog_ctxt_put(ctxt);
2213 int mdt_llog_create(struct mdt_thread_info *info)
2217 req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_CREATE);
2218 rc = llog_origin_handle_open(mdt_info_req(info));
2219 return (rc < 0 ? err_serious(rc) : rc);
2222 int mdt_llog_destroy(struct mdt_thread_info *info)
2226 req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_DESTROY);
2227 rc = llog_origin_handle_destroy(mdt_info_req(info));
2228 return (rc < 0 ? err_serious(rc) : rc);
2231 int mdt_llog_read_header(struct mdt_thread_info *info)
2235 req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_READ_HEADER);
2236 rc = llog_origin_handle_read_header(mdt_info_req(info));
2237 return (rc < 0 ? err_serious(rc) : rc);
2240 int mdt_llog_next_block(struct mdt_thread_info *info)
2244 req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_NEXT_BLOCK);
2245 rc = llog_origin_handle_next_block(mdt_info_req(info));
2246 return (rc < 0 ? err_serious(rc) : rc);
2249 int mdt_llog_prev_block(struct mdt_thread_info *info)
2253 req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_PREV_BLOCK);
2254 rc = llog_origin_handle_prev_block(mdt_info_req(info));
2255 return (rc < 0 ? err_serious(rc) : rc);
2263 static struct ldlm_callback_suite cbs = {
2264 .lcs_completion = ldlm_server_completion_ast,
2265 .lcs_blocking = ldlm_server_blocking_ast,
2266 .lcs_glimpse = ldlm_server_glimpse_ast
2269 int mdt_enqueue(struct mdt_thread_info *info)
2271 struct ptlrpc_request *req;
2275 * info->mti_dlm_req already contains swapped and (if necessary)
2276 * converted dlm request.
2278 LASSERT(info->mti_dlm_req != NULL);
2280 req = mdt_info_req(info);
2281 rc = ldlm_handle_enqueue0(info->mti_mdt->mdt_namespace,
2282 req, info->mti_dlm_req, &cbs);
2283 info->mti_fail_id = OBD_FAIL_LDLM_REPLY;
2284 return rc ? err_serious(rc) : req->rq_status;
2287 int mdt_convert(struct mdt_thread_info *info)
2290 struct ptlrpc_request *req;
2292 LASSERT(info->mti_dlm_req);
2293 req = mdt_info_req(info);
2294 rc = ldlm_handle_convert0(req, info->mti_dlm_req);
2295 return rc ? err_serious(rc) : req->rq_status;
2298 int mdt_bl_callback(struct mdt_thread_info *info)
2300 CERROR("bl callbacks should not happen on MDS\n");
2302 return err_serious(-EOPNOTSUPP);
2305 int mdt_cp_callback(struct mdt_thread_info *info)
2307 CERROR("cp callbacks should not happen on MDS\n");
2309 return err_serious(-EOPNOTSUPP);
2313 * sec context handlers
2315 int mdt_sec_ctx_handle(struct mdt_thread_info *info)
2319 rc = mdt_handle_idmap(info);
2322 struct ptlrpc_request *req = mdt_info_req(info);
2325 opc = lustre_msg_get_opc(req->rq_reqmsg);
2326 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT)
2327 sptlrpc_svc_ctx_invalidate(req);
2330 CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
2336 * quota request handlers
2338 int mdt_quota_dqacq(struct mdt_thread_info *info)
2340 struct lu_device *qmt = info->mti_mdt->mdt_qmt_dev;
2345 RETURN(err_serious(-EOPNOTSUPP));
2347 rc = qmt_hdls.qmth_dqacq(info->mti_env, qmt, mdt_info_req(info));
2351 static struct mdt_object *mdt_obj(struct lu_object *o)
2353 LASSERT(lu_device_is_mdt(o->lo_dev));
2354 return container_of0(o, struct mdt_object, mot_obj.mo_lu);
2357 struct mdt_object *mdt_object_new(const struct lu_env *env,
2358 struct mdt_device *d,
2359 const struct lu_fid *f)
2361 struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2362 struct lu_object *o;
2363 struct mdt_object *m;
2366 CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
2367 o = lu_object_find(env, &d->mdt_md_dev.md_lu_dev, f, &conf);
2368 if (unlikely(IS_ERR(o)))
2369 m = (struct mdt_object *)o;
2375 struct mdt_object *mdt_object_find(const struct lu_env *env,
2376 struct mdt_device *d,
2377 const struct lu_fid *f)
2379 struct lu_object *o;
2380 struct mdt_object *m;
2383 CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2384 o = lu_object_find(env, &d->mdt_md_dev.md_lu_dev, f, NULL);
2385 if (unlikely(IS_ERR(o)))
2386 m = (struct mdt_object *)o;
2393 * Asyncronous commit for mdt device.
2395 * Pass asynchonous commit call down the MDS stack.
2397 * \param env environment
2398 * \param mdt the mdt device
2400 static void mdt_device_commit_async(const struct lu_env *env,
2401 struct mdt_device *mdt)
2403 struct dt_device *dt = mdt->mdt_bottom;
2406 rc = dt->dd_ops->dt_commit_async(env, dt);
2407 if (unlikely(rc != 0))
2408 CWARN("async commit start failed with rc = %d", rc);
2412 * Mark the lock as "synchonous".
2414 * Mark the lock to deffer transaction commit to the unlock time.
2416 * \param lock the lock to mark as "synchonous"
2418 * \see mdt_is_lock_sync
2419 * \see mdt_save_lock
2421 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2423 lock->l_ast_data = (void*)1;
2427 * Check whehter the lock "synchonous" or not.
2429 * \param lock the lock to check
2430 * \retval 1 the lock is "synchonous"
2431 * \retval 0 the lock isn't "synchronous"
2433 * \see mdt_set_lock_sync
2434 * \see mdt_save_lock
2436 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2438 return lock->l_ast_data != NULL;
2442 * Blocking AST for mdt locks.
2444 * Starts transaction commit if in case of COS lock conflict or
2445 * deffers such a commit to the mdt_save_lock.
2447 * \param lock the lock which blocks a request or cancelling lock
2448 * \param desc unused
2449 * \param data unused
2450 * \param flag indicates whether this cancelling or blocking callback
2452 * \see ldlm_blocking_ast_nocheck
2454 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2455 void *data, int flag)
2457 struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2458 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2462 if (flag == LDLM_CB_CANCELING)
2464 lock_res_and_lock(lock);
2465 if (lock->l_blocking_ast != mdt_blocking_ast) {
2466 unlock_res_and_lock(lock);
2469 if (mdt_cos_is_enabled(mdt) &&
2470 lock->l_req_mode & (LCK_PW | LCK_EX) &&
2471 lock->l_blocking_lock != NULL &&
2472 lock->l_client_cookie != lock->l_blocking_lock->l_client_cookie) {
2473 mdt_set_lock_sync(lock);
2475 rc = ldlm_blocking_ast_nocheck(lock);
2477 /* There is no lock conflict if l_blocking_lock == NULL,
2478 * it indicates a blocking ast sent from ldlm_lock_decref_internal
2479 * when the last reference to a local lock was released */
2480 if (lock->l_req_mode == LCK_COS && lock->l_blocking_lock != NULL) {
2483 rc = lu_env_init(&env, LCT_LOCAL);
2484 if (unlikely(rc != 0))
2485 CWARN("lu_env initialization failed with rc = %d,"
2486 "cannot start asynchronous commit\n", rc);
2488 mdt_device_commit_async(&env, mdt);
2494 int mdt_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2495 void *data, int flag)
2497 struct lustre_handle lockh;
2501 case LDLM_CB_BLOCKING:
2502 ldlm_lock2handle(lock, &lockh);
2503 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
2505 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
2509 case LDLM_CB_CANCELING:
2510 LDLM_DEBUG(lock, "Revoke remote lock\n");
2518 int mdt_remote_object_lock(struct mdt_thread_info *mti,
2519 struct mdt_object *o, struct lustre_handle *lh,
2520 ldlm_mode_t mode, __u64 ibits)
2522 struct ldlm_enqueue_info *einfo = &mti->mti_einfo;
2523 ldlm_policy_data_t *policy = &mti->mti_policy;
2527 LASSERT(mdt_object_remote(o));
2529 LASSERT((ibits & MDS_INODELOCK_UPDATE));
2531 memset(einfo, 0, sizeof(*einfo));
2532 einfo->ei_type = LDLM_IBITS;
2533 einfo->ei_mode = mode;
2534 einfo->ei_cb_bl = mdt_md_blocking_ast;
2535 einfo->ei_cb_cp = ldlm_completion_ast;
2537 memset(policy, 0, sizeof(*policy));
2538 policy->l_inodebits.bits = ibits;
2540 rc = mo_object_lock(mti->mti_env, mdt_object_child(o), lh, einfo,
2545 static int mdt_object_lock0(struct mdt_thread_info *info, struct mdt_object *o,
2546 struct mdt_lock_handle *lh, __u64 ibits,
2547 bool nonblock, int locality)
2549 struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
2550 ldlm_policy_data_t *policy = &info->mti_policy;
2551 struct ldlm_res_id *res_id = &info->mti_res_id;
2556 LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2557 LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2558 LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
2559 LASSERT(lh->mlh_type != MDT_NUL_LOCK);
2561 if (mdt_object_remote(o)) {
2562 if (locality == MDT_CROSS_LOCK) {
2563 ibits &= ~(MDS_INODELOCK_UPDATE | MDS_INODELOCK_PERM);
2564 ibits |= MDS_INODELOCK_LOOKUP;
2567 (MDS_INODELOCK_UPDATE | MDS_INODELOCK_PERM)),
2568 "%s: wrong bit "LPX64" for remote obj "DFID"\n",
2569 mdt_obd_name(info->mti_mdt), ibits,
2570 PFID(mdt_object_fid(o)));
2571 LASSERT(ibits & MDS_INODELOCK_LOOKUP);
2573 /* No PDO lock on remote object */
2574 LASSERT(lh->mlh_type != MDT_PDO_LOCK);
2577 if (lh->mlh_type == MDT_PDO_LOCK) {
2578 /* check for exists after object is locked */
2579 if (mdt_object_exists(o) == 0) {
2580 /* Non-existent object shouldn't have PDO lock */
2583 /* Non-dir object shouldn't have PDO lock */
2584 if (!S_ISDIR(lu_object_attr(&o->mot_obj.mo_lu)))
2589 memset(policy, 0, sizeof(*policy));
2590 fid_build_reg_res_name(mdt_object_fid(o), res_id);
2592 dlmflags = LDLM_FL_ATOMIC_CB;
2594 dlmflags |= LDLM_FL_BLOCK_NOWAIT;
2597 * Take PDO lock on whole directory and build correct @res_id for lock
2598 * on part of directory.
2600 if (lh->mlh_pdo_hash != 0) {
2601 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
2602 mdt_lock_pdo_mode(info, o, lh);
2603 if (lh->mlh_pdo_mode != LCK_NL) {
2605 * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
2606 * is never going to be sent to client and we do not
2607 * want it slowed down due to possible cancels.
2609 policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
2610 rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
2611 policy, res_id, dlmflags,
2612 &info->mti_exp->exp_handle.h_cookie);
2618 * Finish res_id initializing by name hash marking part of
2619 * directory which is taking modification.
2621 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
2624 policy->l_inodebits.bits = ibits;
2627 * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
2628 * going to be sent to client. If it is - mdt_intent_policy() path will
2629 * fix it up and turn FL_LOCAL flag off.
2631 rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
2632 res_id, LDLM_FL_LOCAL_ONLY | dlmflags,
2633 &info->mti_exp->exp_handle.h_cookie);
2635 mdt_object_unlock(info, o, lh, 1);
2636 else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
2637 lh->mlh_pdo_hash != 0 &&
2638 (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX)) {
2639 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
2645 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2646 struct mdt_lock_handle *lh, __u64 ibits, int locality)
2648 return mdt_object_lock0(info, o, lh, ibits, false, locality);
2651 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o,
2652 struct mdt_lock_handle *lh, __u64 ibits, int locality)
2654 struct mdt_lock_handle tmp = *lh;
2657 rc = mdt_object_lock0(info, o, &tmp, ibits, true, locality);
2665 * Save a lock within request object.
2667 * Keep the lock referenced until whether client ACK or transaction
2668 * commit happens or release the lock immediately depending on input
2669 * parameters. If COS is ON, a write lock is converted to COS lock
2672 * \param info thead info object
2673 * \param h lock handle
2674 * \param mode lock mode
2675 * \param decref force immediate lock releasing
2678 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
2679 ldlm_mode_t mode, int decref)
2683 if (lustre_handle_is_used(h)) {
2684 if (decref || !info->mti_has_trans ||
2685 !(mode & (LCK_PW | LCK_EX))){
2686 mdt_fid_unlock(h, mode);
2688 struct mdt_device *mdt = info->mti_mdt;
2689 struct ldlm_lock *lock = ldlm_handle2lock(h);
2690 struct ptlrpc_request *req = mdt_info_req(info);
2693 LASSERTF(lock != NULL, "no lock for cookie "LPX64"\n",
2695 CDEBUG(D_HA, "request = %p reply state = %p"
2696 " transno = "LPD64"\n",
2697 req, req->rq_reply_state, req->rq_transno);
2698 if (mdt_cos_is_enabled(mdt)) {
2700 ldlm_lock_downgrade(lock, LCK_COS);
2703 ptlrpc_save_lock(req, h, mode, no_ack);
2704 if (mdt_is_lock_sync(lock)) {
2705 CDEBUG(D_HA, "found sync-lock,"
2706 " async commit started\n");
2707 mdt_device_commit_async(info->mti_env,
2710 LDLM_LOCK_PUT(lock);
2719 * Unlock mdt object.
2721 * Immeditely release the regular lock and the PDO lock or save the
2722 * lock in reqeuest and keep them referenced until client ACK or
2723 * transaction commit.
2725 * \param info thread info object
2726 * \param o mdt object
2727 * \param lh mdt lock handle referencing regular and PDO locks
2728 * \param decref force immediate lock releasing
2730 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
2731 struct mdt_lock_handle *lh, int decref)
2735 mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
2736 mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
2738 if (lustre_handle_is_used(&lh->mlh_rreg_lh))
2739 ldlm_lock_decref(&lh->mlh_rreg_lh, lh->mlh_rreg_mode);
2744 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
2745 const struct lu_fid *f,
2746 struct mdt_lock_handle *lh,
2749 struct mdt_object *o;
2751 o = mdt_object_find(info->mti_env, info->mti_mdt, f);
2755 rc = mdt_object_lock(info, o, lh, ibits,
2758 mdt_object_put(info->mti_env, o);
2765 void mdt_object_unlock_put(struct mdt_thread_info * info,
2766 struct mdt_object * o,
2767 struct mdt_lock_handle *lh,
2770 mdt_object_unlock(info, o, lh, decref);
2771 mdt_object_put(info->mti_env, o);
2774 struct mdt_handler *mdt_handler_find(__u32 opc, struct mdt_opc_slice *supported)
2776 struct mdt_opc_slice *s;
2777 struct mdt_handler *h;
2780 for (s = supported; s->mos_hs != NULL; s++) {
2781 if (s->mos_opc_start <= opc && opc < s->mos_opc_end) {
2782 h = s->mos_hs + (opc - s->mos_opc_start);
2783 if (likely(h->mh_opc != 0))
2784 LASSERTF(h->mh_opc == opc,
2785 "opcode mismatch %d != %d\n",
2788 h = NULL; /* unsupported opc */
2795 static int mdt_lock_resname_compat(struct mdt_device *m,
2796 struct ldlm_request *req)
2798 /* XXX something... later. */
2802 static int mdt_lock_reply_compat(struct mdt_device *m, struct ldlm_reply *rep)
2804 /* XXX something... later. */
2809 * Generic code handling requests that have struct mdt_body passed in:
2811 * - extract mdt_body from request and save it in @info, if present;
2813 * - create lu_object, corresponding to the fid in mdt_body, and save it in
2816 * - if HABEO_CORPUS flag is set for this request type check whether object
2817 * actually exists on storage (lu_object_exists()).
2820 static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags)
2822 const struct mdt_body *body;
2823 struct mdt_object *obj;
2824 const struct lu_env *env;
2825 struct req_capsule *pill;
2829 env = info->mti_env;
2830 pill = info->mti_pill;
2832 body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
2836 if (!(body->valid & OBD_MD_FLID))
2839 if (!fid_is_sane(&body->fid1)) {
2840 CERROR("Invalid fid: "DFID"\n", PFID(&body->fid1));
2845 * Do not get size or any capa fields before we check that request
2846 * contains capa actually. There are some requests which do not, for
2847 * instance MDS_IS_SUBDIR.
2849 if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_CLIENT) &&
2850 req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
2851 mdt_set_capainfo(info, 0, &body->fid1,
2852 req_capsule_client_get(pill, &RMF_CAPA1));
2854 obj = mdt_object_find(env, info->mti_mdt, &body->fid1);
2856 if ((flags & HABEO_CORPUS) &&
2857 !mdt_object_exists(obj)) {
2858 mdt_object_put(env, obj);
2859 /* for capability renew ENOENT will be handled in
2861 if (body->valid & OBD_MD_FLOSSCAPA)
2866 info->mti_object = obj;
2875 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags)
2877 struct req_capsule *pill = info->mti_pill;
2881 if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
2882 rc = mdt_body_unpack(info, flags);
2886 if (rc == 0 && (flags & HABEO_REFERO)) {
2888 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2889 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2890 info->mti_body->eadatasize);
2891 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2892 req_capsule_set_size(pill, &RMF_LOGCOOKIES,
2895 rc = req_capsule_server_pack(pill);
2900 static int mdt_init_capa_ctxt(const struct lu_env *env, struct mdt_device *m)
2902 struct md_device *next = m->mdt_child;
2904 return next->md_ops->mdo_init_capa_ctxt(env, next,
2905 m->mdt_opts.mo_mds_capa,
2906 m->mdt_capa_timeout,
2912 * Invoke handler for this request opc. Also do necessary preprocessing
2913 * (according to handler ->mh_flags), and post-processing (setting of
2914 * ->last_{xid,committed}).
2916 static int mdt_req_handle(struct mdt_thread_info *info,
2917 struct mdt_handler *h, struct ptlrpc_request *req)
2919 int rc, serious = 0;
2924 LASSERT(h->mh_act != NULL);
2925 LASSERT(h->mh_opc == lustre_msg_get_opc(req->rq_reqmsg));
2926 LASSERT(current->journal_info == NULL);
2929 * Checking for various OBD_FAIL_$PREF_$OPC_NET codes. _Do_ not try
2930 * to put same checks into handlers like mdt_close(), mdt_reint(),
2931 * etc., without talking to mdt authors first. Checking same thing
2932 * there again is useless and returning 0 error without packing reply
2933 * is buggy! Handlers either pack reply or return error.
2935 * We return 0 here and do not send any reply in order to emulate
2936 * network failure. Do not send any reply in case any of NET related
2937 * fail_id has occured.
2939 if (OBD_FAIL_CHECK_ORSET(h->mh_fail_id, OBD_FAIL_ONCE))
2943 flags = h->mh_flags;
2944 LASSERT(ergo(flags & (HABEO_CORPUS|HABEO_REFERO), h->mh_fmt != NULL));
2946 if (h->mh_fmt != NULL) {
2947 req_capsule_set(info->mti_pill, h->mh_fmt);
2948 rc = mdt_unpack_req_pack_rep(info, flags);
2951 if (rc == 0 && flags & MUTABOR &&
2952 exp_connect_flags(req->rq_export) & OBD_CONNECT_RDONLY)
2953 /* should it be rq_status? */
2956 if (rc == 0 && flags & HABEO_CLAVIS) {
2957 struct ldlm_request *dlm_req;
2959 LASSERT(h->mh_fmt != NULL);
2961 dlm_req = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
2962 if (dlm_req != NULL) {
2963 if (unlikely(dlm_req->lock_desc.l_resource.lr_type ==
2965 dlm_req->lock_desc.l_policy_data.\
2966 l_inodebits.bits == 0)) {
2968 * Lock without inodebits makes no sense and
2969 * will oops later in ldlm. If client miss to
2970 * set such bits, do not trigger ASSERTION.
2972 * For liblustre flock case, it maybe zero.
2976 if (info->mti_mdt->mdt_opts.mo_compat_resname)
2977 rc = mdt_lock_resname_compat(
2980 info->mti_dlm_req = dlm_req;
2987 /* capability setting changed via /proc, needs reinitialize ctxt */
2988 if (info->mti_mdt && info->mti_mdt->mdt_capa_conf) {
2989 mdt_init_capa_ctxt(info->mti_env, info->mti_mdt);
2990 info->mti_mdt->mdt_capa_conf = 0;
2993 if (likely(rc == 0)) {
2995 * Process request, there can be two types of rc:
2996 * 1) errors with msg unpack/pack, other failures outside the
2997 * operation itself. This is counted as serious errors;
2998 * 2) errors during fs operation, should be placed in rq_status
3001 rc = h->mh_act(info);
3003 !req->rq_no_reply && req->rq_reply_state == NULL) {
3004 DEBUG_REQ(D_ERROR, req, "MDT \"handler\" %s did not "
3005 "pack reply and returned 0 error\n",
3009 serious = is_serious(rc);
3010 rc = clear_serious(rc);
3014 req->rq_status = rc;
3017 * ELDLM_* codes which > 0 should be in rq_status only as well as
3018 * all non-serious errors.
3020 if (rc > 0 || !serious)
3023 LASSERT(current->journal_info == NULL);
3025 if (rc == 0 && (flags & HABEO_CLAVIS) &&
3026 info->mti_mdt->mdt_opts.mo_compat_resname) {
3027 struct ldlm_reply *dlmrep;
3029 dlmrep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3031 rc = mdt_lock_reply_compat(info->mti_mdt, dlmrep);
3034 /* If we're DISCONNECTing, the mdt_export_data is already freed */
3035 if (likely(rc == 0 && req->rq_export && h->mh_opc != MDS_DISCONNECT))
3036 target_committed_to_req(req);
3038 if (unlikely(req_is_replay(req) &&
3039 lustre_msg_get_transno(req->rq_reqmsg) == 0)) {
3040 DEBUG_REQ(D_ERROR, req, "transno is 0 during REPLAY");
3044 target_send_reply(req, rc, info->mti_fail_id);
3048 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
3050 lh->mlh_type = MDT_NUL_LOCK;
3051 lh->mlh_reg_lh.cookie = 0ull;
3052 lh->mlh_reg_mode = LCK_MINMODE;
3053 lh->mlh_pdo_lh.cookie = 0ull;
3054 lh->mlh_pdo_mode = LCK_MINMODE;
3055 lh->mlh_rreg_lh.cookie = 0ull;
3056 lh->mlh_rreg_mode = LCK_MINMODE;
3059 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
3061 LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
3062 LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
3066 * Initialize fields of struct mdt_thread_info. Other fields are left in
3067 * uninitialized state, because it's too expensive to zero out whole
3068 * mdt_thread_info (> 1K) on each request arrival.
3070 static void mdt_thread_info_init(struct ptlrpc_request *req,
3071 struct mdt_thread_info *info)
3075 req_capsule_init(&req->rq_pill, req, RCL_SERVER);
3076 info->mti_pill = &req->rq_pill;
3079 for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
3080 mdt_lock_handle_init(&info->mti_lh[i]);
3082 /* mdt device: it can be NULL while CONNECT */
3083 if (req->rq_export) {
3084 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
3085 info->mti_exp = req->rq_export;
3087 info->mti_mdt = NULL;
3088 info->mti_env = req->rq_svc_thread->t_env;
3089 info->mti_fail_id = OBD_FAIL_MDS_ALL_REPLY_NET;
3090 info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
3091 info->mti_mos = NULL;
3093 memset(&info->mti_attr, 0, sizeof(info->mti_attr));
3094 info->mti_big_buf = LU_BUF_NULL;
3095 info->mti_body = NULL;
3096 info->mti_object = NULL;
3097 info->mti_dlm_req = NULL;
3098 info->mti_has_trans = 0;
3099 info->mti_cross_ref = 0;
3100 info->mti_opdata = 0;
3101 info->mti_big_lmm_used = 0;
3103 /* To not check for split by default. */
3104 info->mti_spec.no_create = 0;
3105 info->mti_spec.sp_rm_entry = 0;
3108 static void mdt_thread_info_fini(struct mdt_thread_info *info)
3112 req_capsule_fini(info->mti_pill);
3113 if (info->mti_object != NULL) {
3114 mdt_object_put(info->mti_env, info->mti_object);
3115 info->mti_object = NULL;
3118 for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
3119 mdt_lock_handle_fini(&info->mti_lh[i]);
3120 info->mti_env = NULL;
3122 if (unlikely(info->mti_big_buf.lb_buf != NULL))
3123 lu_buf_free(&info->mti_big_buf);
3126 static int mdt_filter_recovery_request(struct ptlrpc_request *req,
3127 struct obd_device *obd, int *process)
3129 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
3130 case MDS_CONNECT: /* This will never get here, but for completeness. */
3131 case OST_CONNECT: /* This will never get here, but for completeness. */
3132 case MDS_DISCONNECT:
3133 case OST_DISCONNECT:
3139 case MDS_DONE_WRITING:
3140 case MDS_SYNC: /* used in unmounting */
3147 *process = target_queue_recovery_request(req, obd);
3151 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
3158 * Handle recovery. Return:
3159 * +1: continue request processing;
3160 * -ve: abort immediately with the given error code;
3161 * 0: send reply with error code in req->rq_status;
3163 static int mdt_recovery(struct mdt_thread_info *info)
3165 struct ptlrpc_request *req = mdt_info_req(info);
3166 struct obd_device *obd;
3170 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
3173 case SEC_CTX_INIT_CONT:
3179 rc = mdt_handle_idmap(info);
3188 if (unlikely(!class_connected_export(req->rq_export))) {
3189 CDEBUG(D_HA, "operation %d on unconnected MDS from %s\n",
3190 lustre_msg_get_opc(req->rq_reqmsg),
3191 libcfs_id2str(req->rq_peer));
3192 /* FIXME: For CMD cleanup, when mds_B stop, the req from
3193 * mds_A will get -ENOTCONN(especially for ping req),
3194 * which will cause that mds_A deactive timeout, then when
3195 * mds_A cleanup, the cleanup process will be suspended since
3196 * deactive timeout is not zero.
3198 req->rq_status = -ENOTCONN;
3199 target_send_reply(req, -ENOTCONN, info->mti_fail_id);
3203 /* sanity check: if the xid matches, the request must be marked as a
3204 * resent or replayed */
3205 if (req_xid_is_last(req)) {
3206 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
3207 (MSG_RESENT | MSG_REPLAY))) {
3208 DEBUG_REQ(D_WARNING, req, "rq_xid "LPU64" matches last_xid, "
3209 "expected REPLAY or RESENT flag (%x)", req->rq_xid,
3210 lustre_msg_get_flags(req->rq_reqmsg));
3212 req->rq_status = -ENOTCONN;
3217 /* else: note the opposite is not always true; a RESENT req after a
3218 * failover will usually not match the last_xid, since it was likely
3219 * never committed. A REPLAYed request will almost never match the
3220 * last xid, however it could for a committed, but still retained,
3223 obd = req->rq_export->exp_obd;
3225 /* Check for aborted recovery... */
3226 if (unlikely(obd->obd_recovering)) {
3229 DEBUG_REQ(D_INFO, req, "Got new replay");
3230 rc = mdt_filter_recovery_request(req, obd, &should_process);
3231 if (rc != 0 || !should_process)
3233 else if (should_process < 0) {
3234 req->rq_status = should_process;
3235 rc = ptlrpc_error(req);
3242 static int mdt_msg_check_version(struct lustre_msg *msg)
3246 switch (lustre_msg_get_opc(msg)) {
3248 case MDS_DISCONNECT:
3251 case SEC_CTX_INIT_CONT:
3254 rc = lustre_msg_check_version(msg, LUSTRE_OBD_VERSION);
3256 CERROR("bad opc %u version %08x, expecting %08x\n",
3257 lustre_msg_get_opc(msg),
3258 lustre_msg_get_version(msg),
3259 LUSTRE_OBD_VERSION);
3263 case MDS_GETATTR_NAME:
3270 case MDS_DONE_WRITING:
3277 case MDS_HSM_PROGRESS:
3278 case MDS_HSM_REQUEST:
3279 case MDS_HSM_CT_REGISTER:
3280 case MDS_HSM_CT_UNREGISTER:
3281 case MDS_HSM_STATE_GET:
3282 case MDS_HSM_STATE_SET:
3283 case MDS_HSM_ACTION:
3284 case MDS_QUOTACHECK:
3287 case MDS_SWAP_LAYOUTS:
3292 rc = lustre_msg_check_version(msg, LUSTRE_MDS_VERSION);
3294 CERROR("bad opc %u version %08x, expecting %08x\n",
3295 lustre_msg_get_opc(msg),
3296 lustre_msg_get_version(msg),
3297 LUSTRE_MDS_VERSION);
3301 case LDLM_BL_CALLBACK:
3302 case LDLM_CP_CALLBACK:
3303 rc = lustre_msg_check_version(msg, LUSTRE_DLM_VERSION);
3305 CERROR("bad opc %u version %08x, expecting %08x\n",
3306 lustre_msg_get_opc(msg),
3307 lustre_msg_get_version(msg),
3308 LUSTRE_DLM_VERSION);
3310 case OBD_LOG_CANCEL:
3311 case LLOG_ORIGIN_HANDLE_CREATE:
3312 case LLOG_ORIGIN_HANDLE_NEXT_BLOCK:
3313 case LLOG_ORIGIN_HANDLE_READ_HEADER:
3314 case LLOG_ORIGIN_HANDLE_CLOSE:
3315 case LLOG_ORIGIN_HANDLE_DESTROY:
3316 case LLOG_ORIGIN_HANDLE_PREV_BLOCK:
3318 rc = lustre_msg_check_version(msg, LUSTRE_LOG_VERSION);
3320 CERROR("bad opc %u version %08x, expecting %08x\n",
3321 lustre_msg_get_opc(msg),
3322 lustre_msg_get_version(msg),
3323 LUSTRE_LOG_VERSION);
3326 CERROR("MDS unknown opcode %d\n", lustre_msg_get_opc(msg));
3332 static int mdt_handle0(struct ptlrpc_request *req,
3333 struct mdt_thread_info *info,
3334 struct mdt_opc_slice *supported)
3336 struct mdt_handler *h;
3337 struct lustre_msg *msg;
3342 if (OBD_FAIL_CHECK_ORSET(OBD_FAIL_MDS_ALL_REQUEST_NET, OBD_FAIL_ONCE))
3345 LASSERT(current->journal_info == NULL);
3347 msg = req->rq_reqmsg;
3348 rc = mdt_msg_check_version(msg);
3349 if (likely(rc == 0)) {
3350 rc = mdt_recovery(info);
3351 if (likely(rc == +1)) {
3352 h = mdt_handler_find(lustre_msg_get_opc(msg),
3354 if (likely(h != NULL)) {
3355 rc = mdt_req_handle(info, h, req);
3357 CERROR("The unsupported opc: 0x%x\n",
3358 lustre_msg_get_opc(msg) );
3359 req->rq_status = -ENOTSUPP;
3360 rc = ptlrpc_error(req);
3365 CERROR(LUSTRE_MDT_NAME" drops mal-formed request\n");
3370 * MDT handler function called by ptlrpc service thread when request comes.
3372 * XXX common "target" functionality should be factored into separate module
3373 * shared by mdt, ost and stand-alone services like fld.
3375 int mdt_handle_common(struct ptlrpc_request *req,
3376 struct mdt_opc_slice *supported)
3379 struct mdt_thread_info *info;
3383 env = req->rq_svc_thread->t_env;
3384 /* Refill(initilize) the context(mdt_thread_info), in case it is
3385 * not initialized yet. Usually it happens during start up, after
3386 * MDS(ptlrpc threads) is start up, it gets the first CONNECT request,
3387 * before MDT_thread_info is initialized */
3389 LASSERT(env != NULL);
3390 LASSERT(env->le_ses != NULL);
3391 LASSERT(env->le_ctx.lc_thread == req->rq_svc_thread);
3392 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3393 LASSERT(info != NULL);
3395 mdt_thread_info_init(req, info);
3397 rc = mdt_handle0(req, info, supported);
3399 mdt_thread_info_fini(info);
3404 * This is called from recovery code as handler of _all_ RPC types, FLD and SEQ
3407 int mdt_recovery_handle(struct ptlrpc_request *req)
3412 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
3414 rc = mdt_handle_common(req, mdt_fld_handlers);
3417 rc = mdt_handle_common(req, mdt_seq_handlers);
3420 rc = mdt_handle_common(req, mdt_regular_handlers);
3442 static int mdt_intent_getattr(enum mdt_it_code opcode,
3443 struct mdt_thread_info *info,
3444 struct ldlm_lock **,
3446 static int mdt_intent_layout(enum mdt_it_code opcode,
3447 struct mdt_thread_info *info,
3448 struct ldlm_lock **,
3450 static int mdt_intent_reint(enum mdt_it_code opcode,
3451 struct mdt_thread_info *info,
3452 struct ldlm_lock **,
3455 static struct mdt_it_flavor {
3456 const struct req_format *it_fmt;
3458 int (*it_act)(enum mdt_it_code ,
3459 struct mdt_thread_info *,
3460 struct ldlm_lock **,
3463 } mdt_it_flavor[] = {
3465 .it_fmt = &RQF_LDLM_INTENT,
3466 /*.it_flags = HABEO_REFERO,*/
3468 .it_act = mdt_intent_reint,
3469 .it_reint = REINT_OPEN
3472 .it_fmt = &RQF_LDLM_INTENT,
3473 .it_flags = MUTABOR,
3474 .it_act = mdt_intent_reint,
3475 .it_reint = REINT_OPEN
3478 .it_fmt = &RQF_LDLM_INTENT,
3479 .it_flags = MUTABOR,
3480 .it_act = mdt_intent_reint,
3481 .it_reint = REINT_CREATE
3483 [MDT_IT_GETATTR] = {
3484 .it_fmt = &RQF_LDLM_INTENT_GETATTR,
3485 .it_flags = HABEO_REFERO,
3486 .it_act = mdt_intent_getattr
3488 [MDT_IT_READDIR] = {
3494 .it_fmt = &RQF_LDLM_INTENT_GETATTR,
3495 .it_flags = HABEO_REFERO,
3496 .it_act = mdt_intent_getattr
3499 .it_fmt = &RQF_LDLM_INTENT_UNLINK,
3500 .it_flags = MUTABOR,
3502 .it_reint = REINT_UNLINK
3506 .it_flags = MUTABOR,
3509 [MDT_IT_GETXATTR] = {
3515 .it_fmt = &RQF_LDLM_INTENT_LAYOUT,
3517 .it_act = mdt_intent_layout
3521 int mdt_intent_lock_replace(struct mdt_thread_info *info,
3522 struct ldlm_lock **lockp,
3523 struct ldlm_lock *new_lock,
3524 struct mdt_lock_handle *lh,
3527 struct ptlrpc_request *req = mdt_info_req(info);
3528 struct ldlm_lock *lock = *lockp;
3531 * Get new lock only for cases when possible resent did not find any
3534 if (new_lock == NULL)
3535 new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
3537 if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
3538 lh->mlh_reg_lh.cookie = 0;
3542 LASSERTF(new_lock != NULL,
3543 "lockh "LPX64"\n", lh->mlh_reg_lh.cookie);
3546 * If we've already given this lock to a client once, then we should
3547 * have no readers or writers. Otherwise, we should have one reader
3548 * _or_ writer ref (which will be zeroed below) before returning the
3551 if (new_lock->l_export == req->rq_export) {
3552 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
3554 LASSERT(new_lock->l_export == NULL);
3555 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
3560 if (new_lock->l_export == req->rq_export) {
3562 * Already gave this to the client, which means that we
3563 * reconstructed a reply.
3565 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
3567 lh->mlh_reg_lh.cookie = 0;
3568 RETURN(ELDLM_LOCK_REPLACED);
3572 * Fixup the lock to be given to the client.
3574 lock_res_and_lock(new_lock);
3575 /* Zero new_lock->l_readers and new_lock->l_writers without triggering
3576 * possible blocking AST. */
3577 while (new_lock->l_readers > 0) {
3578 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
3579 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3580 new_lock->l_readers--;
3582 while (new_lock->l_writers > 0) {
3583 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
3584 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3585 new_lock->l_writers--;
3588 new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
3589 new_lock->l_blocking_ast = lock->l_blocking_ast;
3590 new_lock->l_completion_ast = lock->l_completion_ast;
3591 new_lock->l_remote_handle = lock->l_remote_handle;
3592 new_lock->l_flags &= ~LDLM_FL_LOCAL;
3594 unlock_res_and_lock(new_lock);
3596 cfs_hash_add(new_lock->l_export->exp_lock_hash,
3597 &new_lock->l_remote_handle,
3598 &new_lock->l_exp_hash);
3600 LDLM_LOCK_RELEASE(new_lock);
3601 lh->mlh_reg_lh.cookie = 0;
3603 RETURN(ELDLM_LOCK_REPLACED);
3606 static void mdt_intent_fixup_resent(struct mdt_thread_info *info,
3607 struct ldlm_lock *new_lock,
3608 struct ldlm_lock **old_lock,
3609 struct mdt_lock_handle *lh)
3611 struct ptlrpc_request *req = mdt_info_req(info);
3612 struct obd_export *exp = req->rq_export;
3613 struct lustre_handle remote_hdl;
3614 struct ldlm_request *dlmreq;
3615 struct ldlm_lock *lock;
3617 if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
3620 dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
3621 remote_hdl = dlmreq->lock_handle[0];
3623 /* In the function below, .hs_keycmp resolves to
3624 * ldlm_export_lock_keycmp() */
3625 /* coverity[overrun-buffer-val] */
3626 lock = cfs_hash_lookup(exp->exp_lock_hash, &remote_hdl);
3628 if (lock != new_lock) {
3629 lh->mlh_reg_lh.cookie = lock->l_handle.h_cookie;
3630 lh->mlh_reg_mode = lock->l_granted_mode;
3632 LDLM_DEBUG(lock, "Restoring lock cookie");
3633 DEBUG_REQ(D_DLMTRACE, req,
3634 "restoring lock cookie "LPX64,
3635 lh->mlh_reg_lh.cookie);
3637 *old_lock = LDLM_LOCK_GET(lock);
3638 cfs_hash_put(exp->exp_lock_hash, &lock->l_exp_hash);
3642 cfs_hash_put(exp->exp_lock_hash, &lock->l_exp_hash);
3646 * If the xid matches, then we know this is a resent request, and allow
3647 * it. (It's probably an OPEN, for which we don't send a lock.
3649 if (req_xid_is_last(req))
3653 * This remote handle isn't enqueued, so we never received or processed
3654 * this request. Clear MSG_RESENT, because it can be handled like any
3655 * normal request now.
3657 lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
3659 DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle "LPX64,
3663 static int mdt_intent_getattr(enum mdt_it_code opcode,
3664 struct mdt_thread_info *info,
3665 struct ldlm_lock **lockp,
3668 struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3669 struct ldlm_lock *new_lock = NULL;
3671 struct ldlm_reply *ldlm_rep;
3672 struct ptlrpc_request *req;
3673 struct mdt_body *reqbody;
3674 struct mdt_body *repbody;
3678 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
3681 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
3684 info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
3685 repbody->eadatasize = 0;
3686 repbody->aclsize = 0;
3690 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM;
3692 case MDT_IT_GETATTR:
3693 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
3697 CERROR("Unsupported intent (%d)\n", opcode);
3698 GOTO(out_shrink, rc = -EINVAL);
3701 rc = mdt_init_ucred(info, reqbody);
3703 GOTO(out_shrink, rc);
3705 req = info->mti_pill->rc_req;
3706 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3707 mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
3709 /* Get lock from request for possible resent case. */
3710 mdt_intent_fixup_resent(info, *lockp, &new_lock, lhc);
3712 rc = mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
3713 ldlm_rep->lock_policy_res2 = clear_serious(rc);
3715 if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
3716 ldlm_rep->lock_policy_res2 = 0;
3717 if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
3718 ldlm_rep->lock_policy_res2) {
3719 lhc->mlh_reg_lh.cookie = 0ull;
3720 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
3723 rc = mdt_intent_lock_replace(info, lockp, new_lock, lhc, flags);
3726 mdt_exit_ucred(info);
3728 mdt_client_compatibility(info);
3729 rc2 = mdt_fix_reply(info);
3735 static int mdt_intent_layout(enum mdt_it_code opcode,
3736 struct mdt_thread_info *info,
3737 struct ldlm_lock **lockp,
3740 struct layout_intent *layout;
3742 struct mdt_object *obj = NULL;
3743 struct md_object *child = NULL;
3747 if (opcode != MDT_IT_LAYOUT) {
3748 CERROR("%s: Unknown intent (%d)\n", mdt_obd_name(info->mti_mdt),
3753 fid = &info->mti_tmp_fid2;
3754 fid_extract_from_res_name(fid, &(*lockp)->l_resource->lr_name);
3756 obj = mdt_object_find(info->mti_env, info->mti_mdt, fid);
3758 RETURN(PTR_ERR(obj));
3760 if (mdt_object_exists(obj) && !mdt_object_remote(obj)) {
3761 child = mdt_object_child(obj);
3763 /* get the length of lsm */
3764 rc = mo_xattr_get(info->mti_env, child, &LU_BUF_NULL,
3767 if (rc > info->mti_mdt->mdt_max_mdsize)
3768 info->mti_mdt->mdt_max_mdsize = rc;
3771 mdt_object_put(info->mti_env, obj);
3773 (*lockp)->l_lvb_type = LVB_T_LAYOUT;
3774 req_capsule_set_size(info->mti_pill, &RMF_DLM_LVB, RCL_SERVER,
3775 ldlm_lvbo_size(*lockp));
3776 rc = req_capsule_server_pack(info->mti_pill);
3780 layout = req_capsule_client_get(info->mti_pill, &RMF_LAYOUT_INTENT);
3781 LASSERT(layout != NULL);
3782 if (layout->li_opc == LAYOUT_INTENT_ACCESS)
3783 /* return to normal ldlm handling */
3786 CERROR("%s: Unsupported layout intent (%d)\n",
3787 mdt_obd_name(info->mti_mdt), layout->li_opc);
3791 static int mdt_intent_reint(enum mdt_it_code opcode,
3792 struct mdt_thread_info *info,
3793 struct ldlm_lock **lockp,
3796 struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3797 struct ldlm_reply *rep = NULL;
3801 static const struct req_format *intent_fmts[REINT_MAX] = {
3802 [REINT_CREATE] = &RQF_LDLM_INTENT_CREATE,
3803 [REINT_OPEN] = &RQF_LDLM_INTENT_OPEN
3808 opc = mdt_reint_opcode(info, intent_fmts);
3812 if (mdt_it_flavor[opcode].it_reint != opc) {
3813 CERROR("Reint code %ld doesn't match intent: %d\n",
3815 RETURN(err_serious(-EPROTO));
3818 /* Get lock from request for possible resent case. */
3819 mdt_intent_fixup_resent(info, *lockp, NULL, lhc);
3821 rc = mdt_reint_internal(info, lhc, opc);
3823 /* Check whether the reply has been packed successfully. */
3824 if (mdt_info_req(info)->rq_repmsg != NULL)
3825 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3827 RETURN(err_serious(-EFAULT));
3829 /* MDC expects this in any case */
3831 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
3833 /* the open lock or the lock for cross-ref object should be
3834 * returned to the client */
3835 if (rc == -EREMOTE || mdt_get_disposition(rep, DISP_OPEN_LOCK)) {
3836 LASSERT(lustre_handle_is_used(&lhc->mlh_reg_lh));
3837 rep->lock_policy_res2 = 0;
3838 rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
3842 rep->lock_policy_res2 = clear_serious(rc);
3844 if (rep->lock_policy_res2 == -ENOENT &&
3845 mdt_get_disposition(rep, DISP_LOOKUP_NEG))
3846 rep->lock_policy_res2 = 0;
3848 if (rc == -ENOTCONN || rc == -ENODEV ||
3849 rc == -EOVERFLOW) { /**< if VBR failure then return error */
3851 * If it is the disconnect error (ENODEV & ENOCONN), the error
3852 * will be returned by rq_status, and client at ptlrpc layer
3853 * will detect this, then disconnect, reconnect the import
3854 * immediately, instead of impacting the following the rpc.
3856 lhc->mlh_reg_lh.cookie = 0ull;
3860 * For other cases, the error will be returned by intent.
3861 * and client will retrieve the result from intent.
3864 * FIXME: when open lock is finished, that should be
3867 if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3868 LASSERTF(rc == 0, "Error occurred but lock handle "
3869 "is still in use, rc = %d\n", rc);
3870 rep->lock_policy_res2 = 0;
3871 rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
3874 lhc->mlh_reg_lh.cookie = 0ull;
3875 RETURN(ELDLM_LOCK_ABORTED);
3880 static int mdt_intent_code(long itcode)
3888 case IT_OPEN|IT_CREAT:
3895 rc = MDT_IT_READDIR;
3898 rc = MDT_IT_GETATTR;
3910 rc = MDT_IT_GETXATTR;
3915 case IT_QUOTA_DQACQ:
3920 CERROR("Unknown intent opcode: %ld\n", itcode);
3927 static int mdt_intent_opc(long itopc, struct mdt_thread_info *info,
3928 struct ldlm_lock **lockp, __u64 flags)
3930 struct req_capsule *pill;
3931 struct mdt_it_flavor *flv;
3936 opc = mdt_intent_code(itopc);
3940 pill = info->mti_pill;
3942 if (opc == MDT_IT_QUOTA) {
3943 struct lu_device *qmt = info->mti_mdt->mdt_qmt_dev;
3946 RETURN(-EOPNOTSUPP);
3948 (*lockp)->l_lvb_type = LVB_T_LQUOTA;
3949 /* pass the request to quota master */
3950 rc = qmt_hdls.qmth_intent_policy(info->mti_env, qmt,
3951 mdt_info_req(info), lockp,
3956 flv = &mdt_it_flavor[opc];
3957 if (flv->it_fmt != NULL)
3958 req_capsule_extend(pill, flv->it_fmt);
3960 rc = mdt_unpack_req_pack_rep(info, flv->it_flags);
3962 struct ptlrpc_request *req = mdt_info_req(info);
3963 if (flv->it_flags & MUTABOR &&
3964 exp_connect_flags(req->rq_export) & OBD_CONNECT_RDONLY)
3967 if (rc == 0 && flv->it_act != NULL) {
3968 struct ldlm_reply *rep;
3970 /* execute policy */
3971 rc = flv->it_act(opc, info, lockp, flags);
3973 /* Check whether the reply has been packed successfully. */
3974 if (mdt_info_req(info)->rq_repmsg != NULL) {
3975 rep = req_capsule_server_get(info->mti_pill,
3977 rep->lock_policy_res2 =
3978 ptlrpc_status_hton(rep->lock_policy_res2);
3986 static int mdt_intent_policy(struct ldlm_namespace *ns,
3987 struct ldlm_lock **lockp, void *req_cookie,
3988 ldlm_mode_t mode, __u64 flags, void *data)
3990 struct mdt_thread_info *info;
3991 struct ptlrpc_request *req = req_cookie;
3992 struct ldlm_intent *it;
3993 struct req_capsule *pill;
3998 LASSERT(req != NULL);
4000 info = lu_context_key_get(&req->rq_svc_thread->t_env->le_ctx,
4002 LASSERT(info != NULL);
4003 pill = info->mti_pill;
4004 LASSERT(pill->rc_req == req);
4006 if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
4007 req_capsule_extend(pill, &RQF_LDLM_INTENT_BASIC);
4008 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
4010 rc = mdt_intent_opc(it->opc, info, lockp, flags);
4014 /* Lock without inodebits makes no sense and will oops
4015 * later in ldlm. Let's check it now to see if we have
4016 * ibits corrupted somewhere in mdt_intent_opc().
4017 * The case for client miss to set ibits has been
4018 * processed by others. */
4019 LASSERT(ergo(info->mti_dlm_req->lock_desc.l_resource.\
4020 lr_type == LDLM_IBITS,
4021 info->mti_dlm_req->lock_desc.\
4022 l_policy_data.l_inodebits.bits != 0));
4024 rc = err_serious(-EFAULT);
4026 /* No intent was provided */
4027 LASSERT(pill->rc_fmt == &RQF_LDLM_ENQUEUE);
4028 req_capsule_set_size(pill, &RMF_DLM_LVB, RCL_SERVER, 0);
4029 rc = req_capsule_server_pack(pill);
4031 rc = err_serious(rc);
4036 static int mdt_seq_fini(const struct lu_env *env,
4037 struct mdt_device *m)
4039 return seq_site_fini(env, mdt_seq_site(m));
4042 static int mdt_seq_init(const struct lu_env *env,
4044 struct mdt_device *m)
4046 struct seq_server_site *ss;
4051 ss = mdt_seq_site(m);
4054 * This is sequence-controller node. Init seq-controller server on local
4057 if (ss->ss_node_id == 0) {
4058 LASSERT(ss->ss_control_seq == NULL);
4060 OBD_ALLOC_PTR(ss->ss_control_seq);
4061 if (ss->ss_control_seq == NULL)
4064 rc = seq_server_init(ss->ss_control_seq,
4065 m->mdt_bottom, uuid,
4066 LUSTRE_SEQ_CONTROLLER,
4071 GOTO(out_seq_fini, rc);
4073 OBD_ALLOC_PTR(ss->ss_client_seq);
4074 if (ss->ss_client_seq == NULL)
4075 GOTO(out_seq_fini, rc = -ENOMEM);
4077 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
4078 if (prefix == NULL) {
4079 OBD_FREE_PTR(ss->ss_client_seq);
4080 GOTO(out_seq_fini, rc = -ENOMEM);
4083 snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s",
4087 * Init seq-controller client after seq-controller server is
4088 * ready. Pass ss->ss_control_seq to it for direct talking.
4090 rc = seq_client_init(ss->ss_client_seq, NULL,
4091 LUSTRE_SEQ_METADATA, prefix,
4092 ss->ss_control_seq);
4093 OBD_FREE(prefix, MAX_OBD_NAME + 5);
4096 GOTO(out_seq_fini, rc);
4099 /* Init seq-server on local MDT */
4100 LASSERT(ss->ss_server_seq == NULL);
4102 OBD_ALLOC_PTR(ss->ss_server_seq);
4103 if (ss->ss_server_seq == NULL)
4104 GOTO(out_seq_fini, rc = -ENOMEM);
4106 rc = seq_server_init(ss->ss_server_seq,
4107 m->mdt_bottom, uuid,
4112 GOTO(out_seq_fini, rc = -ENOMEM);
4114 /* Assign seq-controller client to local seq-server. */
4115 if (ss->ss_node_id == 0) {
4116 LASSERT(ss->ss_client_seq != NULL);
4118 rc = seq_server_set_cli(ss->ss_server_seq,
4126 mdt_seq_fini(env, m);
4134 static int mdt_fld_fini(const struct lu_env *env,
4135 struct mdt_device *m)
4137 struct seq_server_site *ss = mdt_seq_site(m);
4140 if (ss && ss->ss_server_fld) {
4141 fld_server_fini(env, ss->ss_server_fld);
4142 OBD_FREE_PTR(ss->ss_server_fld);
4143 ss->ss_server_fld = NULL;
4149 static int mdt_fld_init(const struct lu_env *env,
4151 struct mdt_device *m)
4153 struct seq_server_site *ss;
4157 ss = mdt_seq_site(m);
4159 OBD_ALLOC_PTR(ss->ss_server_fld);
4160 if (ss->ss_server_fld == NULL)
4161 RETURN(rc = -ENOMEM);
4163 rc = fld_server_init(env, ss->ss_server_fld, m->mdt_bottom, uuid,
4164 ss->ss_node_id, LU_SEQ_RANGE_MDT);
4166 OBD_FREE_PTR(ss->ss_server_fld);
4167 ss->ss_server_fld = NULL;
4174 static void mdt_stack_pre_fini(const struct lu_env *env,
4175 struct mdt_device *m, struct lu_device *top)
4177 struct obd_device *obd;
4178 struct lustre_cfg_bufs *bufs;
4179 struct lustre_cfg *lcfg;
4180 struct mdt_thread_info *info;
4185 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4186 LASSERT(info != NULL);
4188 bufs = &info->mti_u.bufs;
4190 LASSERT(m->mdt_child_exp);
4191 LASSERT(m->mdt_child_exp->exp_obd);
4192 obd = m->mdt_child_exp->exp_obd;
4194 /* process cleanup, pass mdt obd name to get obd umount flags */
4195 /* XXX: this is needed because all layers are referenced by
4196 * objects (some of them are pinned by osd, for example *
4197 * the proper solution should be a model where object used
4198 * by osd only doesn't have mdt/mdd slices -bzzz */
4199 lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
4200 lustre_cfg_bufs_set_string(bufs, 1, NULL);
4201 lcfg = lustre_cfg_new(LCFG_PRE_CLEANUP, bufs);
4203 CERROR("%s:Cannot alloc lcfg!\n", mdt_obd_name(m));
4206 top->ld_ops->ldo_process_config(env, top, lcfg);
4207 lustre_cfg_free(lcfg);
4211 static void mdt_stack_fini(const struct lu_env *env,
4212 struct mdt_device *m, struct lu_device *top)
4214 struct obd_device *obd = mdt2obd_dev(m);
4215 struct lustre_cfg_bufs *bufs;
4216 struct lustre_cfg *lcfg;
4217 struct mdt_thread_info *info;
4221 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4222 LASSERT(info != NULL);
4224 lu_dev_del_linkage(top->ld_site, top);
4226 lu_site_purge(env, top->ld_site, -1);
4228 bufs = &info->mti_u.bufs;
4229 /* process cleanup, pass mdt obd name to get obd umount flags */
4230 /* another purpose is to let all layers to release their objects */
4231 lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
4236 lustre_cfg_bufs_set_string(bufs, 1, flags);
4237 lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
4239 CERROR("Cannot alloc lcfg!\n");
4243 top->ld_ops->ldo_process_config(env, top, lcfg);
4244 lustre_cfg_free(lcfg);
4246 lu_site_purge(env, top->ld_site, -1);
4248 m->mdt_child = NULL;
4249 m->mdt_bottom = NULL;
4251 obd_disconnect(m->mdt_child_exp);
4252 m->mdt_child_exp = NULL;
4254 obd_disconnect(m->mdt_bottom_exp);
4255 m->mdt_child_exp = NULL;
4258 static int mdt_connect_to_next(const struct lu_env *env, struct mdt_device *m,
4259 const char *next, struct obd_export **exp)
4261 struct obd_connect_data *data = NULL;
4262 struct obd_device *obd;
4266 OBD_ALLOC_PTR(data);
4268 GOTO(out, rc = -ENOMEM);
4270 obd = class_name2obd(next);
4272 CERROR("%s: can't locate next device: %s\n",
4273 mdt_obd_name(m), next);
4274 GOTO(out, rc = -ENOTCONN);
4277 data->ocd_connect_flags = OBD_CONNECT_VERSION;
4278 data->ocd_version = LUSTRE_VERSION_CODE;
4280 rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
4282 CERROR("%s: cannot connect to next dev %s (%d)\n",
4283 mdt_obd_name(m), next, rc);
4293 static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt,
4294 struct lustre_cfg *cfg)
4296 char *dev = lustre_cfg_string(cfg, 0);
4297 int rc, name_size, uuid_size;
4298 char *name, *uuid, *p;
4299 struct lustre_cfg_bufs *bufs;
4300 struct lustre_cfg *lcfg;
4301 struct obd_device *obd;
4302 struct lustre_profile *lprof;
4303 struct lu_site *site;
4306 /* in 1.8 we had the only device in the stack - MDS.
4307 * 2.0 introduces MDT, MDD, OSD; MDT starts others internally.
4308 * in 2.3 OSD is instantiated by obd_mount.c, so we need
4309 * to generate names and setup MDT, MDD. MDT will be using
4310 * generated name to connect to MDD. for MDD the next device
4311 * will be LOD with name taken from so called "profile" which
4312 * is generated by mount_option line
4314 * 1.8 MGS generates config. commands like this:
4315 * #06 (104)mount_option 0: 1:lustre-MDT0000 2:lustre-mdtlov
4316 * #08 (120)setup 0:lustre-MDT0000 1:dev 2:type 3:lustre-MDT0000
4317 * 2.0 MGS generates config. commands like this:
4318 * #07 (112)mount_option 0: 1:lustre-MDT0000 2:lustre-MDT0000-mdtlov
4319 * #08 (160)setup 0:lustre-MDT0000 1:lustre-MDT0000_UUID 2:0
4320 * 3:lustre-MDT0000-mdtlov 4:f
4322 * we generate MDD name from MDT one, just replacing T with D
4324 * after all the preparations, the logical equivalent will be
4325 * #01 (160)setup 0:lustre-MDD0000 1:lustre-MDD0000_UUID 2:0
4326 * 3:lustre-MDT0000-mdtlov 4:f
4327 * #02 (160)setup 0:lustre-MDT0000 1:lustre-MDT0000_UUID 2:0
4328 * 3:lustre-MDD0000 4:f
4330 * notice we build the stack from down to top: MDD first, then MDT */
4332 name_size = MAX_OBD_NAME;
4333 uuid_size = MAX_OBD_NAME;
4335 OBD_ALLOC(name, name_size);
4336 OBD_ALLOC(uuid, uuid_size);
4337 if (name == NULL || uuid == NULL)
4338 GOTO(cleanup_mem, rc = -ENOMEM);
4340 OBD_ALLOC_PTR(bufs);
4342 GOTO(cleanup_mem, rc = -ENOMEM);
4345 p = strstr(name, "-MDT");
4347 GOTO(cleanup_mem, rc = -ENOMEM);
4350 snprintf(uuid, MAX_OBD_NAME, "%s_UUID", name);
4352 lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4353 if (lprof == NULL || lprof->lp_dt == NULL) {
4354 CERROR("can't find the profile: %s\n",
4355 lustre_cfg_string(cfg, 0));
4356 GOTO(cleanup_mem, rc = -EINVAL);
4359 lustre_cfg_bufs_reset(bufs, name);
4360 lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_MDD_NAME);
4361 lustre_cfg_bufs_set_string(bufs, 2, uuid);
4362 lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4364 lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
4366 GOTO(free_bufs, rc = -ENOMEM);
4368 rc = class_attach(lcfg);
4370 GOTO(lcfg_cleanup, rc);
4372 obd = class_name2obd(name);
4374 CERROR("Can not find obd %s (%s in config)\n",
4375 MDD_OBD_NAME, lustre_cfg_string(cfg, 0));
4376 GOTO(class_detach, rc = -EINVAL);
4379 lustre_cfg_free(lcfg);
4381 lustre_cfg_bufs_reset(bufs, name);
4382 lustre_cfg_bufs_set_string(bufs, 1, uuid);
4383 lustre_cfg_bufs_set_string(bufs, 2, dev);
4384 lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4386 lcfg = lustre_cfg_new(LCFG_SETUP, bufs);
4388 rc = class_setup(obd, lcfg);
4390 GOTO(class_detach, rc);
4392 /* connect to MDD we just setup */
4393 rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_child_exp);
4397 site = mdt->mdt_child_exp->exp_obd->obd_lu_dev->ld_site;
4399 LASSERT(mdt_lu_site(mdt) == NULL);
4400 mdt->mdt_md_dev.md_lu_dev.ld_site = site;
4401 site->ls_top_dev = &mdt->mdt_md_dev.md_lu_dev;
4402 mdt->mdt_child = lu2md_dev(mdt->mdt_child_exp->exp_obd->obd_lu_dev);
4405 /* now connect to bottom OSD */
4406 snprintf(name, MAX_OBD_NAME, "%s-osd", dev);
4407 rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_bottom_exp);
4411 lu2dt_dev(mdt->mdt_bottom_exp->exp_obd->obd_lu_dev);
4414 rc = lu_env_refill((struct lu_env *)env);
4416 CERROR("Failure to refill session: '%d'\n", rc);
4418 lu_dev_add_linkage(site, &mdt->mdt_md_dev.md_lu_dev);
4423 class_detach(obd, lcfg);
4425 lustre_cfg_free(lcfg);
4430 OBD_FREE(name, name_size);
4432 OBD_FREE(uuid, uuid_size);
4436 /* setup quota master target on MDT0 */
4437 static int mdt_quota_init(const struct lu_env *env, struct mdt_device *mdt,
4438 struct lustre_cfg *cfg)
4440 struct obd_device *obd;
4441 char *dev = lustre_cfg_string(cfg, 0);
4442 char *qmtname, *uuid, *p;
4443 struct lustre_cfg_bufs *bufs;
4444 struct lustre_cfg *lcfg;
4445 struct lustre_profile *lprof;
4446 struct obd_connect_data *data;
4450 LASSERT(mdt->mdt_qmt_exp == NULL);
4451 LASSERT(mdt->mdt_qmt_dev == NULL);
4453 /* quota master is on MDT0 only for now */
4454 if (mdt->mdt_seq_site.ss_node_id != 0)
4457 /* MGS generates config commands which look as follows:
4458 * #01 (160)setup 0:lustre-MDT0000 1:lustre-MDT0000_UUID 2:0
4459 * 3:lustre-MDT0000-mdtlov 4:f
4461 * We generate the QMT name from the MDT one, just replacing MD with QM
4462 * after all the preparations, the logical equivalent will be:
4463 * #01 (160)setup 0:lustre-QMT0000 1:lustre-QMT0000_UUID 2:0
4464 * 3:lustre-MDT0000-osd 4:f */
4465 OBD_ALLOC(qmtname, MAX_OBD_NAME);
4466 OBD_ALLOC(uuid, UUID_MAX);
4467 OBD_ALLOC_PTR(bufs);
4468 OBD_ALLOC_PTR(data);
4469 if (qmtname == NULL || uuid == NULL || bufs == NULL || data == NULL)
4470 GOTO(cleanup_mem, rc = -ENOMEM);
4472 strcpy(qmtname, dev);
4473 p = strstr(qmtname, "-MDT");
4475 GOTO(cleanup_mem, rc = -ENOMEM);
4476 /* replace MD with QM */
4480 snprintf(uuid, UUID_MAX, "%s_UUID", qmtname);
4482 lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4483 if (lprof == NULL || lprof->lp_dt == NULL) {
4484 CERROR("can't find profile for %s\n",
4485 lustre_cfg_string(cfg, 0));
4486 GOTO(cleanup_mem, rc = -EINVAL);
4489 lustre_cfg_bufs_reset(bufs, qmtname);
4490 lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_QMT_NAME);
4491 lustre_cfg_bufs_set_string(bufs, 2, uuid);
4492 lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4494 lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
4496 GOTO(cleanup_mem, rc = -ENOMEM);
4498 rc = class_attach(lcfg);
4500 GOTO(lcfg_cleanup, rc);
4502 obd = class_name2obd(qmtname);
4504 CERROR("Can not find obd %s (%s in config)\n", qmtname,
4505 lustre_cfg_string(cfg, 0));
4506 GOTO(class_detach, rc = -EINVAL);
4509 lustre_cfg_free(lcfg);
4511 lustre_cfg_bufs_reset(bufs, qmtname);
4512 lustre_cfg_bufs_set_string(bufs, 1, uuid);
4513 lustre_cfg_bufs_set_string(bufs, 2, dev);
4515 /* for quota, the next device should be the OSD device */
4516 lustre_cfg_bufs_set_string(bufs, 3,
4517 mdt->mdt_bottom->dd_lu_dev.ld_obd->obd_name);
4519 lcfg = lustre_cfg_new(LCFG_SETUP, bufs);
4521 rc = class_setup(obd, lcfg);
4523 GOTO(class_detach, rc);
4525 mdt->mdt_qmt_dev = obd->obd_lu_dev;
4527 /* configure local quota objects */
4528 rc = mdt->mdt_qmt_dev->ld_ops->ldo_prepare(env,
4529 &mdt->mdt_md_dev.md_lu_dev,
4532 GOTO(class_cleanup, rc);
4534 /* connect to quota master target */
4535 data->ocd_connect_flags = OBD_CONNECT_VERSION;
4536 data->ocd_version = LUSTRE_VERSION_CODE;
4537 rc = obd_connect(NULL, &mdt->mdt_qmt_exp, obd, &obd->obd_uuid,
4540 CERROR("cannot connect to quota master device %s (%d)\n",
4542 GOTO(class_cleanup, rc);
4548 class_manual_cleanup(obd);
4549 mdt->mdt_qmt_dev = NULL;
4553 class_detach(obd, lcfg);
4555 lustre_cfg_free(lcfg);
4560 OBD_FREE(qmtname, MAX_OBD_NAME);
4562 OBD_FREE(uuid, UUID_MAX);
4568 /* Shutdown quota master target associated with mdt */
4569 static void mdt_quota_fini(const struct lu_env *env, struct mdt_device *mdt)
4573 if (mdt->mdt_qmt_exp == NULL)
4575 LASSERT(mdt->mdt_qmt_dev != NULL);
4577 /* the qmt automatically shuts down when the mdt disconnects */
4578 obd_disconnect(mdt->mdt_qmt_exp);
4579 mdt->mdt_qmt_exp = NULL;
4580 mdt->mdt_qmt_dev = NULL;
4584 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
4586 struct md_device *next = m->mdt_child;
4587 struct lu_device *d = &m->mdt_md_dev.md_lu_dev;
4588 struct obd_device *obd = mdt2obd_dev(m);
4591 target_recovery_fini(obd);
4593 ping_evictor_stop();
4596 mdt_stack_pre_fini(env, m, md2lu_dev(m->mdt_child));
4597 mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4598 obd_exports_barrier(obd);
4599 obd_zombie_barrier();
4603 tgt_fini(env, &m->mdt_lut);
4604 mdt_fs_cleanup(env, m);
4605 upcall_cache_cleanup(m->mdt_identity_cache);
4606 m->mdt_identity_cache = NULL;
4608 if (m->mdt_namespace != NULL) {
4609 ldlm_namespace_free(m->mdt_namespace, NULL,
4610 d->ld_obd->obd_force);
4611 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
4614 mdt_quota_fini(env, m);
4616 cfs_free_nidlist(&m->mdt_nosquash_nids);
4617 if (m->mdt_nosquash_str) {
4618 OBD_FREE(m->mdt_nosquash_str, m->mdt_nosquash_strlen);
4619 m->mdt_nosquash_str = NULL;
4620 m->mdt_nosquash_strlen = 0;
4623 next->md_ops->mdo_iocontrol(env, next, OBD_IOC_PAUSE_LFSCK, 0, NULL);
4625 mdt_seq_fini(env, m);
4626 mdt_fld_fini(env, m);
4627 sptlrpc_rule_set_free(&m->mdt_sptlrpc_rset);
4629 next->md_ops->mdo_init_capa_ctxt(env, next, 0, 0, 0, NULL);
4630 cfs_timer_disarm(&m->mdt_ck_timer);
4631 mdt_ck_thread_stop(m);
4636 mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4638 LASSERT(cfs_atomic_read(&d->ld_ref) == 0);
4640 server_put_mount(mdt_obd_name(m), NULL);
4645 static int mdt_adapt_sptlrpc_conf(struct obd_device *obd, int initial)
4647 struct mdt_device *m = mdt_dev(obd->obd_lu_dev);
4648 struct sptlrpc_rule_set tmp_rset;
4651 sptlrpc_rule_set_init(&tmp_rset);
4652 rc = sptlrpc_conf_target_get_rules(obd, &tmp_rset, initial);
4654 CERROR("mdt %s: failed get sptlrpc rules: %d\n",
4655 mdt_obd_name(m), rc);
4659 sptlrpc_target_update_exp_flavor(obd, &tmp_rset);
4661 write_lock(&m->mdt_sptlrpc_lock);
4662 sptlrpc_rule_set_free(&m->mdt_sptlrpc_rset);
4663 m->mdt_sptlrpc_rset = tmp_rset;
4664 write_unlock(&m->mdt_sptlrpc_lock);
4669 int mdt_postrecov(const struct lu_env *, struct mdt_device *);
4671 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
4672 struct lu_device_type *ldt, struct lustre_cfg *cfg)
4674 struct mdt_thread_info *info;
4675 struct obd_device *obd;
4676 const char *dev = lustre_cfg_string(cfg, 0);
4677 const char *num = lustre_cfg_string(cfg, 2);
4678 struct lustre_mount_info *lmi = NULL;
4679 struct lustre_sb_info *lsi;
4681 struct seq_server_site *ss_site;
4682 const char *identity_upcall = "NONE";
4683 struct md_device *next;
4689 md_device_init(&m->mdt_md_dev, ldt);
4691 * Environment (env) might be missing mdt_thread_key values at that
4692 * point, if device is allocated when mdt_thread_key is in QUIESCENT
4695 * Usually device allocation path doesn't use module key values, but
4696 * mdt has to do a lot of work here, so allocate key value.
4698 rc = lu_env_refill((struct lu_env *)env);
4702 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4703 LASSERT(info != NULL);
4705 obd = class_name2obd(dev);
4706 LASSERT(obd != NULL);
4708 m->mdt_max_mdsize = MAX_MD_SIZE; /* 4 stripes */
4710 m->mdt_som_conf = 0;
4712 m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
4713 lmi = server_get_mount(dev);
4715 CERROR("Cannot get mount info for %s!\n", dev);
4718 lsi = s2lsi(lmi->lmi_sb);
4719 /* CMD is supported only in IAM mode */
4721 node_id = simple_strtol(num, NULL, 10);
4722 obd->u.obt.obt_magic = OBT_MAGIC;
4725 rwlock_init(&m->mdt_sptlrpc_lock);
4726 sptlrpc_rule_set_init(&m->mdt_sptlrpc_rset);
4728 spin_lock_init(&m->mdt_ioepoch_lock);
4729 m->mdt_opts.mo_compat_resname = 0;
4730 m->mdt_opts.mo_mds_capa = 1;
4731 m->mdt_opts.mo_oss_capa = 1;
4732 m->mdt_capa_timeout = CAPA_TIMEOUT;
4733 m->mdt_capa_alg = CAPA_HMAC_ALG_SHA1;
4734 m->mdt_ck_timeout = CAPA_KEY_TIMEOUT;
4735 m->mdt_squash_uid = 0;
4736 m->mdt_squash_gid = 0;
4737 CFS_INIT_LIST_HEAD(&m->mdt_nosquash_nids);
4738 m->mdt_nosquash_str = NULL;
4739 m->mdt_nosquash_strlen = 0;
4740 init_rwsem(&m->mdt_squash_sem);
4741 spin_lock_init(&m->mdt_osfs_lock);
4742 m->mdt_osfs_age = cfs_time_shift_64(-1000);
4743 m->mdt_enable_remote_dir = 0;
4744 m->mdt_enable_remote_dir_gid = 0;
4746 m->mdt_md_dev.md_lu_dev.ld_ops = &mdt_lu_ops;
4747 m->mdt_md_dev.md_lu_dev.ld_obd = obd;
4748 /* set this lu_device to obd, because error handling need it */
4749 obd->obd_lu_dev = &m->mdt_md_dev.md_lu_dev;
4751 /* init the stack */
4752 rc = mdt_stack_init((struct lu_env *)env, m, cfg);
4754 CERROR("%s: Can't init device stack, rc %d\n",
4755 mdt_obd_name(m), rc);
4760 ss_site = mdt_seq_site(m);
4761 s->ld_seq_site = ss_site;
4764 /* set server index */
4765 ss_site->ss_node_id = node_id;
4767 /* failover is the default
4768 * FIXME: we do not failout mds0/mgs, which may cause some problems.
4769 * assumed whose ss_node_id == 0 XXX
4771 obd->obd_replayable = 1;
4772 /* No connection accepted until configurations will finish */
4773 obd->obd_no_conn = 1;
4775 if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
4776 char *str = lustre_cfg_string(cfg, 4);
4777 if (strchr(str, 'n')) {
4778 CWARN("%s: recovery disabled\n", mdt_obd_name(m));
4779 obd->obd_replayable = 0;
4783 rc = tgt_init(env, &m->mdt_lut, obd, m->mdt_bottom);
4785 GOTO(err_fini_stack, rc);
4787 rc = mdt_fld_init(env, mdt_obd_name(m), m);
4791 rc = mdt_seq_init(env, mdt_obd_name(m), m);
4793 GOTO(err_fini_fld, rc);
4795 snprintf(info->mti_u.ns_name, sizeof(info->mti_u.ns_name), "%s-%s",
4796 LUSTRE_MDT_NAME, obd->obd_uuid.uuid);
4797 m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
4798 LDLM_NAMESPACE_SERVER,
4799 LDLM_NAMESPACE_GREEDY,
4801 if (m->mdt_namespace == NULL)
4802 GOTO(err_fini_seq, rc = -ENOMEM);
4804 m->mdt_namespace->ns_lvbp = m;
4805 m->mdt_namespace->ns_lvbo = &mdt_lvbo;
4807 ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
4808 /* set obd_namespace for compatibility with old code */
4809 obd->obd_namespace = m->mdt_namespace;
4811 cfs_timer_init(&m->mdt_ck_timer, mdt_ck_timer_callback, m);
4813 rc = mdt_ck_thread_start(m);
4815 GOTO(err_free_ns, rc);
4817 rc = mdt_fs_setup(env, m, obd, lsi);
4821 mdt_adapt_sptlrpc_conf(obd, 1);
4823 next = m->mdt_child;
4824 rc = next->md_ops->mdo_iocontrol(env, next, OBD_IOC_GET_MNTOPT, 0,
4827 GOTO(err_llog_cleanup, rc);
4829 if (mntopts & MNTOPT_USERXATTR)
4830 m->mdt_opts.mo_user_xattr = 1;
4832 m->mdt_opts.mo_user_xattr = 0;
4834 if (mntopts & MNTOPT_ACL)
4835 m->mdt_opts.mo_acl = 1;
4837 m->mdt_opts.mo_acl = 0;
4839 /* XXX: to support suppgid for ACL, we enable identity_upcall
4840 * by default, otherwise, maybe got unexpected -EACCESS. */
4841 if (m->mdt_opts.mo_acl)
4842 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
4844 m->mdt_identity_cache = upcall_cache_init(mdt_obd_name(m),
4846 &mdt_identity_upcall_cache_ops);
4847 if (IS_ERR(m->mdt_identity_cache)) {
4848 rc = PTR_ERR(m->mdt_identity_cache);
4849 m->mdt_identity_cache = NULL;
4850 GOTO(err_llog_cleanup, rc);
4853 rc = mdt_procfs_init(m, dev);
4855 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
4856 GOTO(err_recovery, rc);
4859 rc = mdt_quota_init(env, m, cfg);
4861 GOTO(err_procfs, rc);
4863 m->mdt_ldlm_client = &mdt2obd_dev(m)->obd_ldlm_client;
4864 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
4865 "mdt_ldlm_client", m->mdt_ldlm_client);
4867 ping_evictor_start();
4869 /* recovery will be started upon mdt_prepare()
4870 * when the whole stack is complete and ready
4871 * to serve the requests */
4873 mdt_init_capa_ctxt(env, m);
4875 /* Reduce the initial timeout on an MDS because it doesn't need such
4876 * a long timeout as an OST does. Adaptive timeouts will adjust this
4877 * value appropriately. */
4878 if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
4879 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
4886 target_recovery_fini(obd);
4887 upcall_cache_cleanup(m->mdt_identity_cache);
4888 m->mdt_identity_cache = NULL;
4890 mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4891 mdt_fs_cleanup(env, m);
4893 cfs_timer_disarm(&m->mdt_ck_timer);
4894 mdt_ck_thread_stop(m);
4896 ldlm_namespace_free(m->mdt_namespace, NULL, 0);
4897 obd->obd_namespace = m->mdt_namespace = NULL;
4899 mdt_seq_fini(env, m);
4901 mdt_fld_fini(env, m);
4903 tgt_fini(env, &m->mdt_lut);
4905 mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4908 server_put_mount(dev, lmi->lmi_mnt);
4912 /* For interoperability, the left element is old parameter, the right one
4913 * is the new version of the parameter, if some parameter is deprecated,
4914 * the new version should be set as NULL. */
4915 static struct cfg_interop_param mdt_interop_param[] = {
4916 { "mdt.group_upcall", NULL },
4917 { "mdt.quota_type", NULL },
4918 { "mdd.quota_type", NULL },
4919 { "mdt.rootsquash", "mdt.root_squash" },
4920 { "mdt.nosquash_nid", "mdt.nosquash_nids" },
4924 /* used by MGS to process specific configurations */
4925 static int mdt_process_config(const struct lu_env *env,
4926 struct lu_device *d, struct lustre_cfg *cfg)
4928 struct mdt_device *m = mdt_dev(d);
4929 struct md_device *md_next = m->mdt_child;
4930 struct lu_device *next = md2lu_dev(md_next);
4934 switch (cfg->lcfg_command) {
4936 struct lprocfs_static_vars lvars;
4937 struct obd_device *obd = d->ld_obd;
4939 /* For interoperability */
4940 struct cfg_interop_param *ptr = NULL;
4941 struct lustre_cfg *old_cfg = NULL;
4944 param = lustre_cfg_string(cfg, 1);
4945 if (param == NULL) {
4946 CERROR("param is empty\n");
4951 ptr = class_find_old_param(param, mdt_interop_param);
4953 if (ptr->new_param == NULL) {
4955 CWARN("For interoperability, skip this %s."
4956 " It is obsolete.\n", ptr->old_param);
4960 CWARN("Found old param %s, changed it to %s.\n",
4961 ptr->old_param, ptr->new_param);
4964 cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
4971 lprocfs_mdt_init_vars(&lvars);
4972 rc = class_process_proc_param(PARAM_MDT, lvars.obd_vars,
4974 if (rc > 0 || rc == -ENOSYS)
4975 /* we don't understand; pass it on */
4976 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4978 if (old_cfg != NULL)
4979 lustre_cfg_free(cfg);
4984 /* others are passed further */
4985 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4991 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
4992 const struct lu_object_header *hdr,
4993 struct lu_device *d)
4995 struct mdt_object *mo;
4999 OBD_SLAB_ALLOC_PTR_GFP(mo, mdt_object_kmem, CFS_ALLOC_IO);
5001 struct lu_object *o;
5002 struct lu_object_header *h;
5004 o = &mo->mot_obj.mo_lu;
5005 h = &mo->mot_header;
5006 lu_object_header_init(h);
5007 lu_object_init(o, h, d);
5008 lu_object_add_top(h, o);
5009 o->lo_ops = &mdt_obj_ops;
5010 mutex_init(&mo->mot_ioepoch_mutex);
5011 mutex_init(&mo->mot_lov_mutex);
5017 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
5018 const struct lu_object_conf *unused)
5020 struct mdt_device *d = mdt_dev(o->lo_dev);
5021 struct lu_device *under;
5022 struct lu_object *below;
5026 CDEBUG(D_INFO, "object init, fid = "DFID"\n",
5027 PFID(lu_object_fid(o)));
5029 under = &d->mdt_child->md_lu_dev;
5030 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
5031 if (below != NULL) {
5032 lu_object_add(o, below);
5039 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
5041 struct mdt_object *mo = mdt_obj(o);
5042 struct lu_object_header *h;
5046 CDEBUG(D_INFO, "object free, fid = "DFID"\n",
5047 PFID(lu_object_fid(o)));
5050 lu_object_header_fini(h);
5051 OBD_SLAB_FREE_PTR(mo, mdt_object_kmem);
5056 static int mdt_object_print(const struct lu_env *env, void *cookie,
5057 lu_printer_t p, const struct lu_object *o)
5059 struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
5060 return (*p)(env, cookie, LUSTRE_MDT_NAME"-object@%p(ioepoch="LPU64" "
5061 "flags="LPX64", epochcount=%d, writecount=%d)",
5062 mdto, mdto->mot_ioepoch, mdto->mot_flags,
5063 mdto->mot_ioepoch_count, mdto->mot_writecount);
5066 static int mdt_prepare(const struct lu_env *env,
5067 struct lu_device *pdev,
5068 struct lu_device *cdev)
5070 struct mdt_device *mdt = mdt_dev(cdev);
5071 struct lu_device *next = &mdt->mdt_child->md_lu_dev;
5072 struct obd_device *obd = cdev->ld_obd;
5073 struct lfsck_start_param lsp;
5080 rc = next->ld_ops->ldo_prepare(env, cdev, next);
5084 rc = mdt_llog_ctxt_clone(env, mdt, LLOG_CHANGELOG_ORIG_CTXT);
5088 lsp.lsp_start = NULL;
5089 lsp.lsp_namespace = mdt->mdt_namespace;
5090 rc = mdt->mdt_child->md_ops->mdo_iocontrol(env, mdt->mdt_child,
5091 OBD_IOC_START_LFSCK,
5094 CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
5095 mdt_obd_name(mdt), rc);
5099 if (mdt->mdt_seq_site.ss_node_id == 0) {
5100 rc = mdt->mdt_child->md_ops->mdo_root_get(env, mdt->mdt_child,
5101 &mdt->mdt_md_root_fid);
5105 LASSERT(!test_bit(MDT_FL_CFGLOG, &mdt->mdt_state));
5106 target_recovery_init(&mdt->mdt_lut, mdt_recovery_handle);
5107 set_bit(MDT_FL_CFGLOG, &mdt->mdt_state);
5108 LASSERT(obd->obd_no_conn);
5109 spin_lock(&obd->obd_dev_lock);
5110 obd->obd_no_conn = 0;
5111 spin_unlock(&obd->obd_dev_lock);
5113 if (obd->obd_recovering == 0)
5114 mdt_postrecov(env, mdt);
5119 const struct lu_device_operations mdt_lu_ops = {
5120 .ldo_object_alloc = mdt_object_alloc,
5121 .ldo_process_config = mdt_process_config,
5122 .ldo_prepare = mdt_prepare,
5125 static const struct lu_object_operations mdt_obj_ops = {
5126 .loo_object_init = mdt_object_init,
5127 .loo_object_free = mdt_object_free,
5128 .loo_object_print = mdt_object_print
5131 static int mdt_obd_set_info_async(const struct lu_env *env,
5132 struct obd_export *exp,
5133 __u32 keylen, void *key,
5134 __u32 vallen, void *val,
5135 struct ptlrpc_request_set *set)
5137 struct obd_device *obd = exp->exp_obd;
5143 if (KEY_IS(KEY_SPTLRPC_CONF)) {
5144 rc = mdt_adapt_sptlrpc_conf(obd, 0);
5152 * Match client and server connection feature flags.
5154 * Compute the compatibility flags for a connection request based on
5155 * features mutually supported by client and server.
5157 * The obd_export::exp_connect_data.ocd_connect_flags field in \a exp
5158 * must not be updated here, otherwise a partially initialized value may
5159 * be exposed. After the connection request is successfully processed,
5160 * the top-level MDT connect request handler atomically updates the export
5161 * connect flags from the obd_connect_data::ocd_connect_flags field of the
5162 * reply. \see mdt_connect().
5164 * \param exp the obd_export associated with this client/target pair
5165 * \param mdt the target device for the connection
5166 * \param data stores data for this connect request
5169 * \retval -EPROTO \a data unexpectedly has zero obd_connect_data::ocd_brw_size
5170 * \retval -EBADE client and server feature requirements are incompatible
5172 static int mdt_connect_internal(struct obd_export *exp,
5173 struct mdt_device *mdt,
5174 struct obd_connect_data *data)
5176 LASSERT(data != NULL);
5178 data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
5179 data->ocd_ibits_known &= MDS_INODELOCK_FULL;
5181 /* If no known bits (which should not happen, probably,
5182 as everybody should support LOOKUP and UPDATE bits at least)
5183 revert to compat mode with plain locks. */
5184 if (!data->ocd_ibits_known &&
5185 data->ocd_connect_flags & OBD_CONNECT_IBITS)
5186 data->ocd_connect_flags &= ~OBD_CONNECT_IBITS;
5188 if (!mdt->mdt_opts.mo_acl)
5189 data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
5191 if (!mdt->mdt_opts.mo_user_xattr)
5192 data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
5194 if (!mdt->mdt_som_conf)
5195 data->ocd_connect_flags &= ~OBD_CONNECT_SOM;
5197 if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE) {
5198 data->ocd_brw_size = min(data->ocd_brw_size,
5199 (__u32)MD_MAX_BRW_SIZE);
5200 if (data->ocd_brw_size == 0) {
5201 CERROR("%s: cli %s/%p ocd_connect_flags: "LPX64
5202 " ocd_version: %x ocd_grant: %d "
5203 "ocd_index: %u ocd_brw_size is "
5204 "unexpectedly zero, network data "
5205 "corruption? Refusing connection of this"
5208 exp->exp_client_uuid.uuid,
5209 exp, data->ocd_connect_flags, data->ocd_version,
5210 data->ocd_grant, data->ocd_index);
5215 /* NB: Disregard the rule against updating
5216 * exp_connect_data.ocd_connect_flags in this case, since
5217 * tgt_client_new() needs to know if this is a lightweight
5218 * connection, and it is safe to expose this flag before
5219 * connection processing completes. */
5220 if (data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) {
5221 spin_lock(&exp->exp_lock);
5222 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_LIGHTWEIGHT;
5223 spin_unlock(&exp->exp_lock);
5226 data->ocd_version = LUSTRE_VERSION_CODE;
5227 exp->exp_connect_data = *data;
5228 exp->exp_mdt_data.med_ibits_known = data->ocd_ibits_known;
5230 if ((data->ocd_connect_flags & OBD_CONNECT_FID) == 0) {
5231 CWARN("%s: MDS requires FID support, but client not\n",
5236 if (mdt->mdt_som_conf &&
5237 !(data->ocd_connect_flags & (OBD_CONNECT_LIGHTWEIGHT |
5238 OBD_CONNECT_MDS_MDS |
5239 OBD_CONNECT_SOM))) {
5240 CWARN("%s: MDS has SOM enabled, but client does not support "
5241 "it\n", mdt_obd_name(mdt));
5245 if (OCD_HAS_FLAG(data, PINGLESS)) {
5246 if (ptlrpc_pinger_suppress_pings()) {
5247 spin_lock(&exp->exp_obd->obd_dev_lock);
5248 list_del_init(&exp->exp_obd_chain_timed);
5249 spin_unlock(&exp->exp_obd->obd_dev_lock);
5251 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
5258 static int mdt_connect_check_sptlrpc(struct mdt_device *mdt,
5259 struct obd_export *exp,
5260 struct ptlrpc_request *req)
5262 struct sptlrpc_flavor flvr;
5265 if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
5266 read_lock(&mdt->mdt_sptlrpc_lock);
5267 sptlrpc_target_choose_flavor(&mdt->mdt_sptlrpc_rset,
5271 read_unlock(&mdt->mdt_sptlrpc_lock);
5273 spin_lock(&exp->exp_lock);
5275 exp->exp_sp_peer = req->rq_sp_from;
5276 exp->exp_flvr = flvr;
5278 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
5279 exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
5280 CERROR("unauthorized rpc flavor %x from %s, "
5281 "expect %x\n", req->rq_flvr.sf_rpc,
5282 libcfs_nid2str(req->rq_peer.nid),
5283 exp->exp_flvr.sf_rpc);
5287 spin_unlock(&exp->exp_lock);
5289 if (exp->exp_sp_peer != req->rq_sp_from) {
5290 CERROR("RPC source %s doesn't match %s\n",
5291 sptlrpc_part2name(req->rq_sp_from),
5292 sptlrpc_part2name(exp->exp_sp_peer));
5295 rc = sptlrpc_target_export_check(exp, req);
5302 /* mds_connect copy */
5303 static int mdt_obd_connect(const struct lu_env *env,
5304 struct obd_export **exp, struct obd_device *obd,
5305 struct obd_uuid *cluuid,
5306 struct obd_connect_data *data,
5309 struct mdt_thread_info *info;
5310 struct obd_export *lexp;
5311 struct lustre_handle conn = { 0 };
5312 struct mdt_device *mdt;
5313 struct ptlrpc_request *req;
5317 LASSERT(env != NULL);
5318 if (!exp || !obd || !cluuid)
5321 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5322 req = info->mti_pill->rc_req;
5323 mdt = mdt_dev(obd->obd_lu_dev);
5326 * first, check whether the stack is ready to handle requests
5327 * XXX: probably not very appropriate method is used now
5328 * at some point we should find a better one
5330 if (!test_bit(MDT_FL_SYNCED, &mdt->mdt_state) &&
5331 !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT)) {
5332 rc = obd_health_check(env, mdt->mdt_child_exp->exp_obd);
5335 set_bit(MDT_FL_SYNCED, &mdt->mdt_state);
5338 rc = class_connect(&conn, obd, cluuid);
5342 lexp = class_conn2export(&conn);
5343 LASSERT(lexp != NULL);
5345 rc = mdt_connect_check_sptlrpc(mdt, lexp, req);
5349 if (OBD_FAIL_CHECK(OBD_FAIL_TGT_RCVG_FLAG))
5350 lustre_msg_add_op_flags(req->rq_repmsg, MSG_CONNECT_RECOVERING);
5352 rc = mdt_connect_internal(lexp, mdt, data);
5354 struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
5357 info->mti_exp = lexp;
5358 memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
5359 rc = tgt_client_new(env, lexp);
5361 mdt_export_stats_init(obd, lexp, localdata);
5366 class_disconnect(lexp);
5375 static int mdt_obd_reconnect(const struct lu_env *env,
5376 struct obd_export *exp, struct obd_device *obd,
5377 struct obd_uuid *cluuid,
5378 struct obd_connect_data *data,
5381 struct mdt_thread_info *info;
5382 struct mdt_device *mdt;
5383 struct ptlrpc_request *req;
5387 if (exp == NULL || obd == NULL || cluuid == NULL)
5390 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5391 req = info->mti_pill->rc_req;
5392 mdt = mdt_dev(obd->obd_lu_dev);
5394 rc = mdt_connect_check_sptlrpc(mdt, exp, req);
5398 rc = mdt_connect_internal(exp, mdt_dev(obd->obd_lu_dev), data);
5400 mdt_export_stats_init(obd, exp, localdata);
5405 static int mdt_export_cleanup(struct obd_export *exp)
5407 struct mdt_export_data *med = &exp->exp_mdt_data;
5408 struct obd_device *obd = exp->exp_obd;
5409 struct mdt_device *mdt;
5410 struct mdt_thread_info *info;
5412 CFS_LIST_HEAD(closing_list);
5413 struct mdt_file_data *mfd, *n;
5417 spin_lock(&med->med_open_lock);
5418 while (!cfs_list_empty(&med->med_open_head)) {
5419 cfs_list_t *tmp = med->med_open_head.next;
5420 mfd = cfs_list_entry(tmp, struct mdt_file_data, mfd_list);
5422 /* Remove mfd handle so it can't be found again.
5423 * We are consuming the mfd_list reference here. */
5424 class_handle_unhash(&mfd->mfd_handle);
5425 cfs_list_move_tail(&mfd->mfd_list, &closing_list);
5427 spin_unlock(&med->med_open_lock);
5428 mdt = mdt_dev(obd->obd_lu_dev);
5429 LASSERT(mdt != NULL);
5431 rc = lu_env_init(&env, LCT_MD_THREAD);
5435 info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5436 LASSERT(info != NULL);
5437 memset(info, 0, sizeof *info);
5438 info->mti_env = &env;
5439 info->mti_mdt = mdt;
5440 info->mti_exp = exp;
5442 if (!cfs_list_empty(&closing_list)) {
5443 struct md_attr *ma = &info->mti_attr;
5445 /* Close any open files (which may also cause orphan unlinking). */
5446 cfs_list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
5447 cfs_list_del_init(&mfd->mfd_list);
5448 ma->ma_need = ma->ma_valid = 0;
5449 /* Don't unlink orphan on failover umount, LU-184 */
5450 if (exp->exp_flags & OBD_OPT_FAILOVER) {
5451 ma->ma_valid = MA_FLAGS;
5452 ma->ma_attr_flags |= MDS_KEEP_ORPHAN;
5454 mdt_mfd_close(info, mfd);
5457 info->mti_mdt = NULL;
5458 /* cleanup client slot early */
5459 /* Do not erase record for recoverable client. */
5460 if (!(exp->exp_flags & OBD_OPT_FAILOVER) || exp->exp_failed)
5461 tgt_client_del(&env, exp);
5467 static int mdt_obd_disconnect(struct obd_export *exp)
5473 class_export_get(exp);
5475 rc = server_disconnect_export(exp);
5477 CDEBUG(D_IOCTL, "server disconnect error: %d\n", rc);
5479 rc = mdt_export_cleanup(exp);
5480 class_export_put(exp);
5484 /* FIXME: Can we avoid using these two interfaces? */
5485 static int mdt_init_export(struct obd_export *exp)
5487 struct mdt_export_data *med = &exp->exp_mdt_data;
5491 CFS_INIT_LIST_HEAD(&med->med_open_head);
5492 spin_lock_init(&med->med_open_lock);
5493 mutex_init(&med->med_idmap_mutex);
5494 med->med_idmap = NULL;
5495 spin_lock(&exp->exp_lock);
5496 exp->exp_connecting = 1;
5497 spin_unlock(&exp->exp_lock);
5499 /* self-export doesn't need client data and ldlm initialization */
5500 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5501 &exp->exp_client_uuid)))
5504 rc = tgt_client_alloc(exp);
5508 rc = ldlm_init_export(exp);
5515 tgt_client_free(exp);
5517 CERROR("%s: Failed to initialize export: rc = %d\n",
5518 exp->exp_obd->obd_name, rc);
5522 static int mdt_destroy_export(struct obd_export *exp)
5526 if (exp_connect_rmtclient(exp))
5527 mdt_cleanup_idmap(&exp->exp_mdt_data);
5529 target_destroy_export(exp);
5530 /* destroy can be called from failed obd_setup, so
5531 * checking uuid is safer than obd_self_export */
5532 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5533 &exp->exp_client_uuid)))
5536 ldlm_destroy_export(exp);
5537 tgt_client_free(exp);
5539 LASSERT(cfs_list_empty(&exp->exp_outstanding_replies));
5540 LASSERT(cfs_list_empty(&exp->exp_mdt_data.med_open_head));
5545 /** The maximum depth that fid2path() will search.
5546 * This is limited only because we want to store the fids for
5547 * historical path lookup purposes.
5549 #define MAX_PATH_DEPTH 100
5551 /** mdt_path() lookup structure. */
5552 struct path_lookup_info {
5553 __u64 pli_recno; /**< history point */
5554 __u64 pli_currec; /**< current record */
5555 struct lu_fid pli_fid;
5556 struct lu_fid pli_fids[MAX_PATH_DEPTH]; /**< path, in fids */
5557 struct mdt_object *pli_mdt_obj;
5558 char *pli_path; /**< full path */
5560 int pli_linkno; /**< which hardlink to follow */
5561 int pli_fidcount; /**< number of \a pli_fids */
5564 static int mdt_links_read(struct mdt_thread_info *info,
5565 struct mdt_object *mdt_obj, struct linkea_data *ldata)
5569 LASSERT(ldata->ld_buf->lb_buf != NULL);
5571 if (!mdt_object_exists(mdt_obj))
5574 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5575 ldata->ld_buf, XATTR_NAME_LINK);
5576 if (rc == -ERANGE) {
5577 /* Buf was too small, figure out what we need. */
5578 lu_buf_free(ldata->ld_buf);
5579 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5580 ldata->ld_buf, XATTR_NAME_LINK);
5583 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
5584 if (ldata->ld_buf->lb_buf == NULL)
5586 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5587 ldata->ld_buf, XATTR_NAME_LINK);
5597 static int mdt_path_current(struct mdt_thread_info *info,
5598 struct path_lookup_info *pli)
5600 struct mdt_device *mdt = info->mti_mdt;
5601 struct mdt_object *mdt_obj;
5602 struct link_ea_header *leh;
5603 struct link_ea_entry *lee;
5604 struct lu_name *tmpname = &info->mti_name;
5605 struct lu_fid *tmpfid = &info->mti_tmp_fid1;
5606 struct lu_buf *buf = &info->mti_big_buf;
5609 struct linkea_data ldata = { 0 };
5613 /* temp buffer for path element, the buffer will be finally freed
5614 * in mdt_thread_info_fini */
5615 buf = lu_buf_check_and_alloc(buf, PATH_MAX);
5616 if (buf->lb_buf == NULL)
5620 ptr = pli->pli_path + pli->pli_pathlen - 1;
5623 pli->pli_fidcount = 0;
5624 pli->pli_fids[0] = *(struct lu_fid *)mdt_object_fid(pli->pli_mdt_obj);
5626 /* root FID only exists on MDT0, and fid2path should also ends at MDT0,
5627 * so checking root_fid can only happen on MDT0. */
5628 while (!lu_fid_eq(&mdt->mdt_md_root_fid,
5629 &pli->pli_fids[pli->pli_fidcount])) {
5630 mdt_obj = mdt_object_find(info->mti_env, mdt,
5631 &pli->pli_fids[pli->pli_fidcount]);
5632 if (IS_ERR(mdt_obj))
5633 GOTO(out, rc = PTR_ERR(mdt_obj));
5634 if (mdt_object_remote(mdt_obj)) {
5635 mdt_object_put(info->mti_env, mdt_obj);
5636 GOTO(remote_out, rc = -EREMOTE);
5638 if (!mdt_object_exists(mdt_obj)) {
5639 mdt_object_put(info->mti_env, mdt_obj);
5640 GOTO(out, rc = -ENOENT);
5643 rc = mdt_links_read(info, mdt_obj, &ldata);
5644 mdt_object_put(info->mti_env, mdt_obj);
5646 GOTO(out, rc = PTR_ERR(buf));
5649 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
5650 linkea_entry_unpack(lee, &reclen, tmpname, tmpfid);
5651 /* If set, use link #linkno for path lookup, otherwise use
5652 link #0. Only do this for the final path element. */
5653 if ((pli->pli_fidcount == 0) &&
5654 (pli->pli_linkno < leh->leh_reccount)) {
5656 for (count = 0; count < pli->pli_linkno; count++) {
5657 lee = (struct link_ea_entry *)
5658 ((char *)lee + reclen);
5659 linkea_entry_unpack(lee, &reclen, tmpname,
5662 if (pli->pli_linkno < leh->leh_reccount - 1)
5663 /* indicate to user there are more links */
5667 /* Pack the name in the end of the buffer */
5668 ptr -= tmpname->ln_namelen;
5669 if (ptr - 1 <= pli->pli_path)
5670 GOTO(out, rc = -EOVERFLOW);
5671 strncpy(ptr, tmpname->ln_name, tmpname->ln_namelen);
5674 /* Store the parent fid for historic lookup */
5675 if (++pli->pli_fidcount >= MAX_PATH_DEPTH)
5676 GOTO(out, rc = -EOVERFLOW);
5677 pli->pli_fids[pli->pli_fidcount] = *tmpfid;
5681 ptr++; /* skip leading / */
5682 memmove(pli->pli_path, ptr, pli->pli_path + pli->pli_pathlen - ptr);
5689 /* Returns the full path to this fid, as of changelog record recno. */
5690 static int mdt_path(struct mdt_thread_info *info, struct mdt_object *obj,
5691 char *path, int pathlen, __u64 *recno, int *linkno,
5694 struct mdt_device *mdt = info->mti_mdt;
5695 struct path_lookup_info *pli;
5703 if (lu_fid_eq(&mdt->mdt_md_root_fid, mdt_object_fid(obj))) {
5712 pli->pli_mdt_obj = obj;
5713 pli->pli_recno = *recno;
5714 pli->pli_path = path;
5715 pli->pli_pathlen = pathlen;
5716 pli->pli_linkno = *linkno;
5718 /* Retry multiple times in case file is being moved */
5719 while (tries-- && rc == -EAGAIN)
5720 rc = mdt_path_current(info, pli);
5722 /* return the last resolved fids to the client, so the client will
5723 * build the left path on another MDT for remote object */
5724 *fid = pli->pli_fids[pli->pli_fidcount];
5726 *recno = pli->pli_currec;
5727 /* Return next link index to caller */
5728 *linkno = pli->pli_linkno;
5735 static int mdt_fid2path(struct mdt_thread_info *info,
5736 struct getinfo_fid2path *fp)
5738 struct mdt_device *mdt = info->mti_mdt;
5739 struct mdt_object *obj;
5743 CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n",
5744 PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
5746 if (!fid_is_sane(&fp->gf_fid))
5749 if (!fid_is_namespace_visible(&fp->gf_fid)) {
5750 CWARN("%s: "DFID" is invalid, sequence should be "
5751 ">= "LPX64"\n", mdt_obd_name(mdt),
5752 PFID(&fp->gf_fid), (__u64)FID_SEQ_NORMAL);
5756 obj = mdt_object_find(info->mti_env, mdt, &fp->gf_fid);
5757 if (obj == NULL || IS_ERR(obj)) {
5758 CDEBUG(D_IOCTL, "no object "DFID": %ld\n", PFID(&fp->gf_fid),
5763 if (mdt_object_remote(obj))
5765 else if (!mdt_object_exists(obj))
5771 mdt_object_put(info->mti_env, obj);
5772 CDEBUG(D_IOCTL, "nonlocal object "DFID": %d\n",
5773 PFID(&fp->gf_fid), rc);
5777 rc = mdt_path(info, obj, fp->gf_path, fp->gf_pathlen, &fp->gf_recno,
5778 &fp->gf_linkno, &fp->gf_fid);
5780 CDEBUG(D_INFO, "fid "DFID", path %s recno "LPX64" linkno %u\n",
5781 PFID(&fp->gf_fid), fp->gf_path, fp->gf_recno, fp->gf_linkno);
5783 mdt_object_put(info->mti_env, obj);
5788 static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key,
5789 void *val, int vallen)
5791 struct getinfo_fid2path *fpout, *fpin;
5794 fpin = key + cfs_size_round(sizeof(KEY_FID2PATH));
5797 if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
5798 lustre_swab_fid2path(fpin);
5800 memcpy(fpout, fpin, sizeof(*fpin));
5801 if (fpout->gf_pathlen != vallen - sizeof(*fpin))
5804 rc = mdt_fid2path(info, fpout);
5808 int mdt_get_info(struct mdt_thread_info *info)
5810 struct ptlrpc_request *req = mdt_info_req(info);
5818 key = req_capsule_client_get(info->mti_pill, &RMF_GETINFO_KEY);
5820 CDEBUG(D_IOCTL, "No GETINFO key");
5823 keylen = req_capsule_get_size(info->mti_pill, &RMF_GETINFO_KEY,
5826 vallen = req_capsule_client_get(info->mti_pill, &RMF_GETINFO_VALLEN);
5827 if (vallen == NULL) {
5828 CDEBUG(D_IOCTL, "Unable to get RMF_GETINFO_VALLEN buffer");
5832 req_capsule_set_size(info->mti_pill, &RMF_GETINFO_VAL, RCL_SERVER,
5834 rc = req_capsule_server_pack(info->mti_pill);
5835 valout = req_capsule_server_get(info->mti_pill, &RMF_GETINFO_VAL);
5836 if (valout == NULL) {
5837 CDEBUG(D_IOCTL, "Unable to get get-info RPC out buffer");
5841 if (KEY_IS(KEY_FID2PATH))
5842 rc = mdt_rpc_fid2path(info, key, valout, *vallen);
5846 lustre_msg_set_status(req->rq_repmsg, rc);
5851 /* Pass the ioc down */
5852 static int mdt_ioc_child(struct lu_env *env, struct mdt_device *mdt,
5853 unsigned int cmd, int len, void *data)
5855 struct lu_context ioctl_session;
5856 struct md_device *next = mdt->mdt_child;
5860 rc = lu_context_init(&ioctl_session, LCT_SESSION);
5863 ioctl_session.lc_thread = (struct ptlrpc_thread *)cfs_current();
5864 lu_context_enter(&ioctl_session);
5865 env->le_ses = &ioctl_session;
5867 LASSERT(next->md_ops->mdo_iocontrol);
5868 rc = next->md_ops->mdo_iocontrol(env, next, cmd, len, data);
5870 lu_context_exit(&ioctl_session);
5871 lu_context_fini(&ioctl_session);
5875 static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
5877 struct obd_ioctl_data *data = karg;
5880 struct mdt_object *obj;
5881 struct mdt_lock_handle *lh;
5885 if (data->ioc_inlbuf1 == NULL || data->ioc_inllen1 != sizeof(*fid) ||
5886 data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
5889 fid = (struct lu_fid *)data->ioc_inlbuf1;
5891 if (!fid_is_sane(fid))
5894 CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
5896 lh = &mti->mti_lh[MDT_LH_PARENT];
5897 mdt_lock_reg_init(lh, LCK_CR);
5899 obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE);
5901 RETURN(PTR_ERR(obj));
5903 if (mdt_object_remote(obj)) {
5906 * before calling version get the correct MDS should be
5907 * fid, this is error to find remote object here
5909 CERROR("nonlocal object "DFID"\n", PFID(fid));
5910 } else if (!mdt_object_exists(obj)) {
5911 *(__u64 *)data->ioc_inlbuf2 = ENOENT_VERSION;
5914 version = dt_version_get(mti->mti_env, mdt_obj2dt(obj));
5915 *(__u64 *)data->ioc_inlbuf2 = version;
5918 mdt_object_unlock_put(mti, obj, lh, 1);
5922 /* ioctls on obd dev */
5923 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
5924 void *karg, void *uarg)
5927 struct obd_device *obd = exp->exp_obd;
5928 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
5929 struct dt_device *dt = mdt->mdt_bottom;
5933 CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
5934 rc = lu_env_init(&env, LCT_MD_THREAD);
5940 rc = mdt_device_sync(&env, mdt);
5942 case OBD_IOC_SET_READONLY:
5943 rc = dt->dd_ops->dt_ro(&env, dt);
5945 case OBD_IOC_ABORT_RECOVERY:
5946 CERROR("%s: Aborting recovery for device\n", mdt_obd_name(mdt));
5947 target_stop_recovery_thread(obd);
5950 case OBD_IOC_CHANGELOG_REG:
5951 case OBD_IOC_CHANGELOG_DEREG:
5952 case OBD_IOC_CHANGELOG_CLEAR:
5953 rc = mdt_ioc_child(&env, mdt, cmd, len, karg);
5955 case OBD_IOC_START_LFSCK: {
5956 struct md_device *next = mdt->mdt_child;
5957 struct obd_ioctl_data *data = karg;
5958 struct lfsck_start_param lsp;
5960 if (unlikely(data == NULL)) {
5965 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
5966 lsp.lsp_namespace = mdt->mdt_namespace;
5967 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &lsp);
5970 case OBD_IOC_STOP_LFSCK: {
5971 struct md_device *next = mdt->mdt_child;
5973 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, NULL);
5976 case OBD_IOC_GET_OBJ_VERSION: {
5977 struct mdt_thread_info *mti;
5978 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5979 memset(mti, 0, sizeof *mti);
5980 mti->mti_env = &env;
5984 rc = mdt_ioc_version_get(mti, karg);
5989 CERROR("%s: Not supported cmd = %d, rc = %d\n",
5990 mdt_obd_name(mdt), cmd, rc);
5997 int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
5999 struct lu_device *ld = md2lu_dev(mdt->mdt_child);
6003 rc = ld->ld_ops->ldo_recovery_complete(env, ld);
6007 int mdt_obd_postrecov(struct obd_device *obd)
6012 rc = lu_env_init(&env, LCT_MD_THREAD);
6015 rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
6020 static struct obd_ops mdt_obd_device_ops = {
6021 .o_owner = THIS_MODULE,
6022 .o_set_info_async = mdt_obd_set_info_async,
6023 .o_connect = mdt_obd_connect,
6024 .o_reconnect = mdt_obd_reconnect,
6025 .o_disconnect = mdt_obd_disconnect,
6026 .o_init_export = mdt_init_export,
6027 .o_destroy_export = mdt_destroy_export,
6028 .o_iocontrol = mdt_iocontrol,
6029 .o_postrecov = mdt_obd_postrecov,
6032 static struct lu_device* mdt_device_fini(const struct lu_env *env,
6033 struct lu_device *d)
6035 struct mdt_device *m = mdt_dev(d);
6042 static struct lu_device *mdt_device_free(const struct lu_env *env,
6043 struct lu_device *d)
6045 struct mdt_device *m = mdt_dev(d);
6048 md_device_fini(&m->mdt_md_dev);
6053 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
6054 struct lu_device_type *t,
6055 struct lustre_cfg *cfg)
6057 struct lu_device *l;
6058 struct mdt_device *m;
6064 l = &m->mdt_md_dev.md_lu_dev;
6065 rc = mdt_init0(env, m, t, cfg);
6067 mdt_device_free(env, l);
6072 l = ERR_PTR(-ENOMEM);
6076 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
6077 LU_KEY_INIT(mdt, struct mdt_thread_info);
6079 static void mdt_key_fini(const struct lu_context *ctx,
6080 struct lu_context_key *key, void* data)
6082 struct mdt_thread_info *info = data;
6084 if (info->mti_big_lmm) {
6085 OBD_FREE_LARGE(info->mti_big_lmm, info->mti_big_lmmsize);
6086 info->mti_big_lmm = NULL;
6087 info->mti_big_lmmsize = 0;
6092 /* context key: mdt_thread_key */
6093 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
6095 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info)
6097 return lu_ucred(info->mti_env);
6100 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info)
6102 return lu_ucred_check(info->mti_env);
6106 * Enable/disable COS (Commit On Sharing).
6108 * Set/Clear the COS flag in mdt options.
6110 * \param mdt mdt device
6111 * \param val 0 disables COS, other values enable COS
6113 void mdt_enable_cos(struct mdt_device *mdt, int val)
6118 mdt->mdt_opts.mo_cos = !!val;
6119 rc = lu_env_init(&env, LCT_LOCAL);
6120 if (unlikely(rc != 0)) {
6121 CWARN("lu_env initialization failed with rc = %d,"
6122 "cannot sync\n", rc);
6125 mdt_device_sync(&env, mdt);
6130 * Check COS (Commit On Sharing) status.
6132 * Return COS flag status.
6134 * \param mdt mdt device
6136 int mdt_cos_is_enabled(struct mdt_device *mdt)
6138 return mdt->mdt_opts.mo_cos != 0;
6141 static struct lu_device_type_operations mdt_device_type_ops = {
6142 .ldto_device_alloc = mdt_device_alloc,
6143 .ldto_device_free = mdt_device_free,
6144 .ldto_device_fini = mdt_device_fini
6147 static struct lu_device_type mdt_device_type = {
6148 .ldt_tags = LU_DEVICE_MD,
6149 .ldt_name = LUSTRE_MDT_NAME,
6150 .ldt_ops = &mdt_device_type_ops,
6151 .ldt_ctx_tags = LCT_MD_THREAD
6154 static int __init mdt_mod_init(void)
6156 struct lprocfs_static_vars lvars;
6159 CLASSERT(sizeof("0x0123456789ABCDEF:0x01234567:0x01234567") ==
6160 FID_NOBRACE_LEN + 1);
6161 CLASSERT(sizeof("[0x0123456789ABCDEF:0x01234567:0x01234567]") ==
6163 rc = lu_kmem_init(mdt_caches);
6167 rc = mds_mod_init();
6171 lprocfs_mdt_init_vars(&lvars);
6172 rc = class_register_type(&mdt_obd_device_ops, NULL,
6173 lvars.module_vars, LUSTRE_MDT_NAME,
6179 lu_kmem_fini(mdt_caches);
6186 static void __exit mdt_mod_exit(void)
6188 class_unregister_type(LUSTRE_MDT_NAME);
6190 lu_kmem_fini(mdt_caches);
6193 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
6194 MODULE_DESCRIPTION("Lustre Metadata Target ("LUSTRE_MDT_NAME")");
6195 MODULE_LICENSE("GPL");
6197 cfs_module(mdt, LUSTRE_VERSION_STRING, mdt_mod_init, mdt_mod_exit);