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 #include <lustre_ioctl.h>
57 /* struct ptlrpc_request */
58 #include <lustre_net.h>
59 /* struct obd_export */
60 #include <lustre_export.h>
61 /* struct obd_device */
64 #include <dt_object.h>
65 #include <lustre_mds.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_lfsck.h>
73 mdl_mode_t mdt_mdl_lock_modes[] = {
74 [LCK_MINMODE] = MDL_MINMODE,
81 [LCK_GROUP] = MDL_GROUP
84 ldlm_mode_t mdt_dlm_lock_modes[] = {
85 [MDL_MINMODE] = LCK_MINMODE,
92 [MDL_GROUP] = LCK_GROUP
95 static struct mdt_device *mdt_dev(struct lu_device *d);
96 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags);
98 static const struct lu_object_operations mdt_obj_ops;
100 /* Slab for MDT object allocation */
101 static struct kmem_cache *mdt_object_kmem;
103 /* For HSM restore handles */
104 struct kmem_cache *mdt_hsm_cdt_kmem;
106 /* For HSM request handles */
107 struct kmem_cache *mdt_hsm_car_kmem;
109 static struct lu_kmem_descr mdt_caches[] = {
111 .ckd_cache = &mdt_object_kmem,
112 .ckd_name = "mdt_obj",
113 .ckd_size = sizeof(struct mdt_object)
116 .ckd_cache = &mdt_hsm_cdt_kmem,
117 .ckd_name = "mdt_cdt_restore_handle",
118 .ckd_size = sizeof(struct cdt_restore_handle)
121 .ckd_cache = &mdt_hsm_car_kmem,
122 .ckd_name = "mdt_cdt_agent_req",
123 .ckd_size = sizeof(struct cdt_agent_req)
130 int mdt_get_disposition(struct ldlm_reply *rep, int flag)
134 return (rep->lock_policy_res1 & flag);
137 void mdt_clear_disposition(struct mdt_thread_info *info,
138 struct ldlm_reply *rep, int flag)
141 info->mti_opdata &= ~flag;
142 tgt_opdata_clear(info->mti_env, flag);
145 rep->lock_policy_res1 &= ~flag;
148 void mdt_set_disposition(struct mdt_thread_info *info,
149 struct ldlm_reply *rep, int flag)
152 info->mti_opdata |= flag;
153 tgt_opdata_set(info->mti_env, flag);
156 rep->lock_policy_res1 |= flag;
159 void mdt_lock_reg_init(struct mdt_lock_handle *lh, ldlm_mode_t lm)
161 lh->mlh_pdo_hash = 0;
162 lh->mlh_reg_mode = lm;
163 lh->mlh_rreg_mode = lm;
164 lh->mlh_type = MDT_REG_LOCK;
167 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, ldlm_mode_t lock_mode,
168 const struct lu_name *lname)
170 lh->mlh_reg_mode = lock_mode;
171 lh->mlh_rreg_mode = lock_mode;
172 lh->mlh_type = MDT_PDO_LOCK;
174 if (lu_name_is_valid(lname)) {
175 lh->mlh_pdo_hash = full_name_hash(lname->ln_name,
177 /* XXX Workaround for LU-2856
179 * Zero is a valid return value of full_name_hash, but
180 * several users of mlh_pdo_hash assume a non-zero
181 * hash value. We therefore map zero onto an
182 * arbitrary, but consistent value (1) to avoid
183 * problems further down the road. */
184 if (unlikely(lh->mlh_pdo_hash == 0))
185 lh->mlh_pdo_hash = 1;
187 lh->mlh_pdo_hash = 0;
191 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
192 struct mdt_lock_handle *lh)
198 * Any dir access needs couple of locks:
200 * 1) on part of dir we gonna take lookup/modify;
202 * 2) on whole dir to protect it from concurrent splitting and/or to
203 * flush client's cache for readdir().
205 * so, for a given mode and object this routine decides what lock mode
206 * to use for lock #2:
208 * 1) if caller's gonna lookup in dir then we need to protect dir from
209 * being splitted only - LCK_CR
211 * 2) if caller's gonna modify dir then we need to protect dir from
212 * being splitted and to flush cache - LCK_CW
214 * 3) if caller's gonna modify dir and that dir seems ready for
215 * splitting then we need to protect it from any type of access
216 * (lookup/modify/split) - LCK_EX --bzzz
219 LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
220 LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
223 * Ask underlaying level its opinion about preferable PDO lock mode
224 * having access type passed as regular lock mode:
226 * - MDL_MINMODE means that lower layer does not want to specify lock
229 * - MDL_NL means that no PDO lock should be taken. This is used in some
230 * cases. Say, for non-splittable directories no need to use PDO locks
233 mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
234 mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
236 if (mode != MDL_MINMODE) {
237 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
240 * Lower layer does not want to specify locking mode. We do it
241 * our selves. No special protection is needed, just flush
242 * client's cache on modification and allow concurrent
245 switch (lh->mlh_reg_mode) {
247 lh->mlh_pdo_mode = LCK_EX;
250 lh->mlh_pdo_mode = LCK_CR;
253 lh->mlh_pdo_mode = LCK_CW;
256 CERROR("Not expected lock type (0x%x)\n",
257 (int)lh->mlh_reg_mode);
262 LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
266 static int mdt_getstatus(struct tgt_session_info *tsi)
268 struct mdt_thread_info *info = tsi2mdt_info(tsi);
269 struct mdt_device *mdt = info->mti_mdt;
270 struct mdt_body *repbody;
275 rc = mdt_check_ucred(info);
277 GOTO(out, rc = err_serious(rc));
279 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETSTATUS_PACK))
280 GOTO(out, rc = err_serious(-ENOMEM));
282 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
283 repbody->mbo_fid1 = mdt->mdt_md_root_fid;
284 repbody->mbo_valid |= OBD_MD_FLID;
286 if (tsi->tsi_tgt->lut_mds_capa &&
287 exp_connect_flags(info->mti_exp) & OBD_CONNECT_MDS_CAPA) {
288 struct mdt_object *root;
289 struct lustre_capa *capa;
291 root = mdt_object_find(info->mti_env, mdt, &repbody->mbo_fid1);
293 GOTO(out, rc = PTR_ERR(root));
295 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA1);
297 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
298 rc = mo_capa_get(info->mti_env, mdt_object_child(root), capa,
300 mdt_object_put(info->mti_env, root);
302 repbody->mbo_valid |= OBD_MD_FLMDSCAPA;
306 mdt_thread_info_fini(info);
310 static int mdt_statfs(struct tgt_session_info *tsi)
312 struct ptlrpc_request *req = tgt_ses_req(tsi);
313 struct mdt_thread_info *info = tsi2mdt_info(tsi);
314 struct md_device *next = info->mti_mdt->mdt_child;
315 struct ptlrpc_service_part *svcpt;
316 struct obd_statfs *osfs;
321 svcpt = req->rq_rqbd->rqbd_svcpt;
323 /* This will trigger a watchdog timeout */
324 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
325 (MDT_SERVICE_WATCHDOG_FACTOR *
326 at_get(&svcpt->scp_at_estimate)) + 1);
328 rc = mdt_check_ucred(info);
330 GOTO(out, rc = err_serious(rc));
332 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK))
333 GOTO(out, rc = err_serious(-ENOMEM));
335 osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
337 GOTO(out, rc = -EPROTO);
339 /** statfs information are cached in the mdt_device */
340 if (cfs_time_before_64(info->mti_mdt->mdt_osfs_age,
341 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS))) {
342 /** statfs data is too old, get up-to-date one */
343 rc = next->md_ops->mdo_statfs(info->mti_env, next, osfs);
346 spin_lock(&info->mti_mdt->mdt_osfs_lock);
347 info->mti_mdt->mdt_osfs = *osfs;
348 info->mti_mdt->mdt_osfs_age = cfs_time_current_64();
349 spin_unlock(&info->mti_mdt->mdt_osfs_lock);
351 /** use cached statfs data */
352 spin_lock(&info->mti_mdt->mdt_osfs_lock);
353 *osfs = info->mti_mdt->mdt_osfs;
354 spin_unlock(&info->mti_mdt->mdt_osfs_lock);
358 mdt_counter_incr(req, LPROC_MDT_STATFS);
360 mdt_thread_info_fini(info);
365 * Pack SOM attributes into the reply.
366 * Call under a DLM UPDATE lock.
368 static void mdt_pack_size2body(struct mdt_thread_info *info,
369 struct mdt_object *mo)
372 struct md_attr *ma = &info->mti_attr;
374 LASSERT(ma->ma_attr.la_valid & LA_MODE);
375 b = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
377 /* Check if Size-on-MDS is supported, if this is a regular file,
378 * if SOM is enabled on the object and if SOM cache exists and valid.
379 * Otherwise do not pack Size-on-MDS attributes to the reply. */
380 if (!(mdt_conn_flags(info) & OBD_CONNECT_SOM) ||
381 !S_ISREG(ma->ma_attr.la_mode) ||
382 !mdt_object_is_som_enabled(mo) ||
383 !(ma->ma_valid & MA_SOM))
386 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
387 b->mbo_size = ma->ma_som->msd_size;
388 b->mbo_blocks = ma->ma_som->msd_blocks;
391 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
392 const struct lu_attr *attr, const struct lu_fid *fid)
394 struct md_attr *ma = &info->mti_attr;
396 LASSERT(ma->ma_valid & MA_INODE);
398 b->mbo_atime = attr->la_atime;
399 b->mbo_mtime = attr->la_mtime;
400 b->mbo_ctime = attr->la_ctime;
401 b->mbo_mode = attr->la_mode;
402 b->mbo_size = attr->la_size;
403 b->mbo_blocks = attr->la_blocks;
404 b->mbo_uid = attr->la_uid;
405 b->mbo_gid = attr->la_gid;
406 b->mbo_flags = attr->la_flags;
407 b->mbo_nlink = attr->la_nlink;
408 b->mbo_rdev = attr->la_rdev;
410 /* XXX should pack the reply body according to lu_valid */
411 b->mbo_valid |= OBD_MD_FLCTIME | OBD_MD_FLUID |
412 OBD_MD_FLGID | OBD_MD_FLTYPE |
413 OBD_MD_FLMODE | OBD_MD_FLNLINK | OBD_MD_FLFLAGS |
414 OBD_MD_FLATIME | OBD_MD_FLMTIME ;
416 if (!S_ISREG(attr->la_mode)) {
417 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
418 } else if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV)) {
419 /* means no objects are allocated on osts. */
420 LASSERT(!(ma->ma_valid & MA_LOV));
421 /* just ignore blocks occupied by extend attributes on MDS */
423 /* if no object is allocated on osts, the size on mds is valid.
425 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
426 } else if ((ma->ma_valid & MA_LOV) && ma->ma_lmm != NULL &&
427 ma->ma_lmm->lmm_pattern & LOV_PATTERN_F_RELEASED) {
428 /* A released file stores its size on MDS. */
429 /* But return 1 block for released file, unless tools like tar
430 * will consider it fully sparse. (LU-3864)
432 if (unlikely(b->mbo_size == 0))
436 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
441 b->mbo_valid |= OBD_MD_FLID;
442 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, size="LPU64"\n",
443 PFID(fid), b->mbo_nlink,
444 b->mbo_mode, b->mbo_size);
448 mdt_body_reverse_idmap(info, b);
450 if (fid != NULL && (b->mbo_valid & OBD_MD_FLSIZE))
451 CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
452 PFID(fid), (unsigned long long)b->mbo_size);
455 static inline int mdt_body_has_lov(const struct lu_attr *la,
456 const struct mdt_body *body)
458 return (S_ISREG(la->la_mode) && (body->mbo_valid & OBD_MD_FLEASIZE)) ||
459 (S_ISDIR(la->la_mode) && (body->mbo_valid & OBD_MD_FLDIREA));
462 void mdt_client_compatibility(struct mdt_thread_info *info)
464 struct mdt_body *body;
465 struct ptlrpc_request *req = mdt_info_req(info);
466 struct obd_export *exp = req->rq_export;
467 struct md_attr *ma = &info->mti_attr;
468 struct lu_attr *la = &ma->ma_attr;
471 if (exp_connect_layout(exp))
472 /* the client can deal with 16-bit lmm_stripe_count */
475 body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
477 if (!mdt_body_has_lov(la, body))
480 /* now we have a reply with a lov for a client not compatible with the
481 * layout lock so we have to clean the layout generation number */
482 if (S_ISREG(la->la_mode))
483 ma->ma_lmm->lmm_layout_gen = 0;
487 int mdt_attr_get_eabuf_size(struct mdt_thread_info *info, struct mdt_object *o)
489 const struct lu_env *env = info->mti_env;
492 rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
501 /* Is it a directory? Let's check for the LMV as well */
502 if (S_ISDIR(lu_object_attr(&mdt_object_child(o)->mo_lu))) {
503 rc2 = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
507 rc2 = mo_xattr_get(env, mdt_object_child(o),
509 XATTR_NAME_DEFAULT_LMV);
511 if ((rc2 < 0 && rc2 != -ENODATA) || (rc2 > rc))
519 int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
522 const struct lu_env *env = info->mti_env;
526 LASSERT(info->mti_big_lmm_used == 0);
527 rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL, name);
531 /* big_lmm may need to be grown */
532 if (info->mti_big_lmmsize < rc) {
533 int size = size_roundup_power2(rc);
535 if (info->mti_big_lmmsize > 0) {
536 /* free old buffer */
537 LASSERT(info->mti_big_lmm);
538 OBD_FREE_LARGE(info->mti_big_lmm,
539 info->mti_big_lmmsize);
540 info->mti_big_lmm = NULL;
541 info->mti_big_lmmsize = 0;
544 OBD_ALLOC_LARGE(info->mti_big_lmm, size);
545 if (info->mti_big_lmm == NULL)
547 info->mti_big_lmmsize = size;
549 LASSERT(info->mti_big_lmmsize >= rc);
551 info->mti_buf.lb_buf = info->mti_big_lmm;
552 info->mti_buf.lb_len = info->mti_big_lmmsize;
553 rc = mo_xattr_get(env, mdt_object_child(o), &info->mti_buf, name);
558 int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
559 struct md_attr *ma, const char *name)
561 struct md_object *next = mdt_object_child(o);
562 struct lu_buf *buf = &info->mti_buf;
565 if (strcmp(name, XATTR_NAME_LOV) == 0) {
566 buf->lb_buf = ma->ma_lmm;
567 buf->lb_len = ma->ma_lmm_size;
568 LASSERT(!(ma->ma_valid & MA_LOV));
569 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
570 buf->lb_buf = ma->ma_lmv;
571 buf->lb_len = ma->ma_lmv_size;
572 LASSERT(!(ma->ma_valid & MA_LMV));
573 } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
574 buf->lb_buf = ma->ma_lmv;
575 buf->lb_len = ma->ma_lmv_size;
576 LASSERT(!(ma->ma_valid & MA_LMV_DEF));
581 rc = mo_xattr_get(info->mti_env, next, buf, name);
585 if (strcmp(name, XATTR_NAME_LOV) == 0) {
586 if (info->mti_big_lmm_used)
587 ma->ma_lmm = info->mti_big_lmm;
589 /* NOT return LOV EA with hole to old client. */
590 if (unlikely(le32_to_cpu(ma->ma_lmm->lmm_pattern) &
591 LOV_PATTERN_F_HOLE) &&
592 !(exp_connect_flags(info->mti_exp) &
593 OBD_CONNECT_LFSCK)) {
596 ma->ma_lmm_size = rc;
597 ma->ma_valid |= MA_LOV;
599 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
600 if (info->mti_big_lmm_used)
601 ma->ma_lmv = info->mti_big_lmm;
603 ma->ma_lmv_size = rc;
604 ma->ma_valid |= MA_LMV;
605 } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
606 ma->ma_lmv_size = rc;
607 ma->ma_valid |= MA_LMV_DEF;
610 /* Update mdt_max_mdsize so all clients will be aware that */
611 if (info->mti_mdt->mdt_max_mdsize < rc)
612 info->mti_mdt->mdt_max_mdsize = rc;
615 } else if (rc == -ENODATA) {
618 } else if (rc == -ERANGE) {
619 /* Default LMV has fixed size, so it must be able to fit
620 * in the original buffer */
621 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
623 rc = mdt_big_xattr_get(info, o, name);
625 info->mti_big_lmm_used = 1;
633 static int mdt_attr_get_pfid(struct mdt_thread_info *info,
634 struct mdt_object *o, struct lu_fid *pfid)
636 struct lu_buf *buf = &info->mti_buf;
637 struct link_ea_header *leh;
638 struct link_ea_entry *lee;
642 buf->lb_buf = info->mti_big_lmm;
643 buf->lb_len = info->mti_big_lmmsize;
644 rc = mo_xattr_get(info->mti_env, mdt_object_child(o),
645 buf, XATTR_NAME_LINK);
646 /* ignore errors, MA_PFID won't be set and it is
647 * up to the caller to treat this as an error */
648 if (rc == -ERANGE || buf->lb_len == 0) {
649 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
650 buf->lb_buf = info->mti_big_lmm;
651 buf->lb_len = info->mti_big_lmmsize;
656 if (rc < sizeof(*leh)) {
657 CERROR("short LinkEA on "DFID": rc = %d\n",
658 PFID(mdt_object_fid(o)), rc);
662 leh = (struct link_ea_header *) buf->lb_buf;
663 lee = (struct link_ea_entry *)(leh + 1);
664 if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
665 leh->leh_magic = LINK_EA_MAGIC;
666 leh->leh_reccount = __swab32(leh->leh_reccount);
667 leh->leh_len = __swab64(leh->leh_len);
669 if (leh->leh_magic != LINK_EA_MAGIC)
671 if (leh->leh_reccount == 0)
674 memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
675 fid_be_to_cpu(pfid, pfid);
680 int mdt_attr_get_complex(struct mdt_thread_info *info,
681 struct mdt_object *o, struct md_attr *ma)
683 const struct lu_env *env = info->mti_env;
684 struct md_object *next = mdt_object_child(o);
685 struct lu_buf *buf = &info->mti_buf;
686 u32 mode = lu_object_attr(&next->mo_lu);
687 int need = ma->ma_need;
693 if (need & MA_INODE) {
694 ma->ma_need = MA_INODE;
695 rc = mo_attr_get(env, next, ma);
698 ma->ma_valid |= MA_INODE;
701 if (need & MA_PFID) {
702 rc = mdt_attr_get_pfid(info, o, &ma->ma_pfid);
704 ma->ma_valid |= MA_PFID;
705 /* ignore this error, parent fid is not mandatory */
709 if (need & MA_LOV && (S_ISREG(mode) || S_ISDIR(mode))) {
710 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LOV);
715 if (need & MA_LMV && S_ISDIR(mode)) {
716 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LMV);
721 if (need & MA_LMV_DEF && S_ISDIR(mode)) {
722 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_DEFAULT_LMV);
727 if (need & MA_SOM && S_ISREG(mode)) {
728 buf->lb_buf = info->mti_xattr_buf;
729 buf->lb_len = sizeof(info->mti_xattr_buf);
730 CLASSERT(sizeof(struct som_attrs) <=
731 sizeof(info->mti_xattr_buf));
732 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_SOM);
733 rc2 = lustre_buf2som(info->mti_xattr_buf, rc2, ma->ma_som);
735 ma->ma_valid |= MA_SOM;
736 else if (rc2 < 0 && rc2 != -ENODATA)
740 if (need & MA_HSM && S_ISREG(mode)) {
741 buf->lb_buf = info->mti_xattr_buf;
742 buf->lb_len = sizeof(info->mti_xattr_buf);
743 CLASSERT(sizeof(struct hsm_attrs) <=
744 sizeof(info->mti_xattr_buf));
745 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_HSM);
746 rc2 = lustre_buf2hsm(info->mti_xattr_buf, rc2, &ma->ma_hsm);
748 ma->ma_valid |= MA_HSM;
749 else if (rc2 < 0 && rc2 != -ENODATA)
753 #ifdef CONFIG_FS_POSIX_ACL
754 if (need & MA_ACL_DEF && S_ISDIR(mode)) {
755 buf->lb_buf = ma->ma_acl;
756 buf->lb_len = ma->ma_acl_size;
757 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
759 ma->ma_acl_size = rc2;
760 ma->ma_valid |= MA_ACL_DEF;
761 } else if (rc2 == -ENODATA) {
770 CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = "LPX64" ma_lmm=%p\n",
771 rc, ma->ma_valid, ma->ma_lmm);
775 static int mdt_getattr_internal(struct mdt_thread_info *info,
776 struct mdt_object *o, int ma_need)
778 struct md_object *next = mdt_object_child(o);
779 const struct mdt_body *reqbody = info->mti_body;
780 struct ptlrpc_request *req = mdt_info_req(info);
781 struct md_attr *ma = &info->mti_attr;
782 struct lu_attr *la = &ma->ma_attr;
783 struct req_capsule *pill = info->mti_pill;
784 const struct lu_env *env = info->mti_env;
785 struct mdt_body *repbody;
786 struct lu_buf *buffer = &info->mti_buf;
791 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
792 RETURN(err_serious(-ENOMEM));
794 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
798 if (mdt_object_remote(o)) {
799 /* This object is located on remote node.*/
800 /* Return -ENOTSUPP for old client */
801 if (!mdt_is_dne_client(req->rq_export))
802 GOTO(out, rc = -ENOTSUPP);
804 repbody->mbo_fid1 = *mdt_object_fid(o);
805 repbody->mbo_valid = OBD_MD_FLID | OBD_MD_MDS;
809 if (reqbody->mbo_eadatasize > 0) {
810 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
811 if (buffer->lb_buf == NULL)
812 GOTO(out, rc = -EPROTO);
813 buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD,
816 buffer->lb_buf = NULL;
818 ma_need &= ~(MA_LOV | MA_LMV);
819 CDEBUG(D_INFO, "%s: RPC from %s: does not need LOVEA.\n",
820 mdt_obd_name(info->mti_mdt),
821 req->rq_export->exp_client_uuid.uuid);
824 /* If it is dir object and client require MEA, then we got MEA */
825 if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
826 (reqbody->mbo_valid & (OBD_MD_MEA | OBD_MD_DEFAULT_MEA))) {
827 /* Assumption: MDT_MD size is enough for lmv size. */
828 ma->ma_lmv = buffer->lb_buf;
829 ma->ma_lmv_size = buffer->lb_len;
830 ma->ma_need = MA_INODE;
831 if (ma->ma_lmv_size > 0) {
832 if (reqbody->mbo_valid & OBD_MD_MEA)
833 ma->ma_need |= MA_LMV;
834 else if (reqbody->mbo_valid & OBD_MD_DEFAULT_MEA)
835 ma->ma_need |= MA_LMV_DEF;
838 ma->ma_lmm = buffer->lb_buf;
839 ma->ma_lmm_size = buffer->lb_len;
840 ma->ma_need = MA_INODE | MA_HSM;
841 if (ma->ma_lmm_size > 0)
842 ma->ma_need |= MA_LOV;
845 if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
846 reqbody->mbo_valid & OBD_MD_FLDIREA &&
847 lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
848 /* get default stripe info for this dir. */
849 ma->ma_need |= MA_LOV_DEF;
851 ma->ma_need |= ma_need;
852 if (ma->ma_need & MA_SOM)
853 ma->ma_som = &info->mti_u.som.data;
855 rc = mdt_attr_get_complex(info, o, ma);
857 CERROR("%s: getattr error for "DFID": rc = %d\n",
858 mdt_obd_name(info->mti_mdt),
859 PFID(mdt_object_fid(o)), rc);
863 /* if file is released, check if a restore is running */
864 if ((ma->ma_valid & MA_HSM) && (ma->ma_hsm.mh_flags & HS_RELEASED) &&
865 mdt_hsm_restore_is_running(info, mdt_object_fid(o))) {
866 repbody->mbo_t_state = MS_RESTORE;
867 repbody->mbo_valid |= OBD_MD_TSTATE;
870 is_root = lu_fid_eq(mdt_object_fid(o), &info->mti_mdt->mdt_md_root_fid);
872 /* the Lustre protocol supposes to return default striping
873 * on the user-visible root if explicitly requested */
874 if ((ma->ma_valid & MA_LOV) == 0 && S_ISDIR(la->la_mode) &&
875 (ma->ma_need & MA_LOV_DEF && is_root) && ma->ma_need & MA_LOV) {
876 struct lu_fid rootfid;
877 struct mdt_object *root;
878 struct mdt_device *mdt = info->mti_mdt;
880 rc = dt_root_get(env, mdt->mdt_bottom, &rootfid);
883 root = mdt_object_find(env, mdt, &rootfid);
885 RETURN(PTR_ERR(root));
886 rc = mdt_stripe_get(info, root, ma, XATTR_NAME_LOV);
887 mdt_object_put(info->mti_env, root);
889 CERROR("%s: getattr error for "DFID": rc = %d\n",
890 mdt_obd_name(info->mti_mdt),
891 PFID(mdt_object_fid(o)), rc);
896 if (likely(ma->ma_valid & MA_INODE))
897 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
901 if (mdt_body_has_lov(la, reqbody)) {
902 if (ma->ma_valid & MA_LOV) {
903 LASSERT(ma->ma_lmm_size);
904 repbody->mbo_eadatasize = ma->ma_lmm_size;
905 if (S_ISDIR(la->la_mode))
906 repbody->mbo_valid |= OBD_MD_FLDIREA;
908 repbody->mbo_valid |= OBD_MD_FLEASIZE;
909 mdt_dump_lmm(D_INFO, ma->ma_lmm, repbody->mbo_valid);
911 if (ma->ma_valid & MA_LMV) {
912 /* Return -ENOTSUPP for old client */
913 if (!mdt_is_striped_client(req->rq_export))
916 LASSERT(S_ISDIR(la->la_mode));
917 mdt_dump_lmv(D_INFO, ma->ma_lmv);
918 repbody->mbo_eadatasize = ma->ma_lmv_size;
919 repbody->mbo_valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
921 if (ma->ma_valid & MA_LMV_DEF) {
922 /* Return -ENOTSUPP for old client */
923 if (!mdt_is_striped_client(req->rq_export))
925 LASSERT(S_ISDIR(la->la_mode));
926 repbody->mbo_eadatasize = ma->ma_lmv_size;
927 repbody->mbo_valid |= (OBD_MD_FLDIREA |
930 } else if (S_ISLNK(la->la_mode) &&
931 reqbody->mbo_valid & OBD_MD_LINKNAME) {
932 buffer->lb_buf = ma->ma_lmm;
933 /* eadatasize from client includes NULL-terminator, so
934 * there is no need to read it */
935 buffer->lb_len = reqbody->mbo_eadatasize - 1;
936 rc = mo_readlink(env, next, buffer);
937 if (unlikely(rc <= 0)) {
938 CERROR("%s: readlink failed for "DFID": rc = %d\n",
939 mdt_obd_name(info->mti_mdt),
940 PFID(mdt_object_fid(o)), rc);
943 int print_limit = min_t(int, PAGE_CACHE_SIZE - 128, rc);
945 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
947 repbody->mbo_valid |= OBD_MD_LINKNAME;
948 /* we need to report back size with NULL-terminator
949 * because client expects that */
950 repbody->mbo_eadatasize = rc + 1;
951 if (repbody->mbo_eadatasize != reqbody->mbo_eadatasize)
952 CDEBUG(D_INODE, "%s: Read shorter symlink %d "
953 "on "DFID ", expected %d\n",
954 mdt_obd_name(info->mti_mdt),
955 rc, PFID(mdt_object_fid(o)),
956 reqbody->mbo_eadatasize - 1);
958 ((char *)ma->ma_lmm)[rc] = 0;
960 /* If the total CDEBUG() size is larger than a page, it
961 * will print a warning to the console, avoid this by
962 * printing just the last part of the symlink. */
963 CDEBUG(D_INODE, "symlink dest %s%.*s, len = %d\n",
964 print_limit < rc ? "..." : "", print_limit,
965 (char *)ma->ma_lmm + rc - print_limit, rc);
970 if (reqbody->mbo_valid & OBD_MD_FLMODEASIZE) {
971 repbody->mbo_max_cookiesize = 0;
972 repbody->mbo_max_mdsize = info->mti_mdt->mdt_max_mdsize;
973 repbody->mbo_valid |= OBD_MD_FLMODEASIZE;
974 CDEBUG(D_INODE, "I am going to change the MAX_MD_SIZE & "
975 "MAX_COOKIE to : %d:%d\n", repbody->mbo_max_mdsize,
976 repbody->mbo_max_cookiesize);
979 if (exp_connect_rmtclient(info->mti_exp) &&
980 reqbody->mbo_valid & OBD_MD_FLRMTPERM) {
981 void *buf = req_capsule_server_get(pill, &RMF_ACL);
983 /* mdt_getattr_lock only */
984 rc = mdt_pack_remote_perm(info, o, buf);
986 repbody->mbo_valid &= ~OBD_MD_FLRMTPERM;
987 repbody->mbo_aclsize = 0;
990 repbody->mbo_valid |= OBD_MD_FLRMTPERM;
991 repbody->mbo_aclsize = sizeof(struct mdt_remote_perm);
994 #ifdef CONFIG_FS_POSIX_ACL
995 else if ((exp_connect_flags(req->rq_export) & OBD_CONNECT_ACL) &&
996 (reqbody->mbo_valid & OBD_MD_FLACL)) {
997 buffer->lb_buf = req_capsule_server_get(pill, &RMF_ACL);
998 buffer->lb_len = req_capsule_get_size(pill,
999 &RMF_ACL, RCL_SERVER);
1000 if (buffer->lb_len > 0) {
1001 rc = mo_xattr_get(env, next, buffer,
1002 XATTR_NAME_ACL_ACCESS);
1004 if (rc == -ENODATA) {
1005 repbody->mbo_aclsize = 0;
1006 repbody->mbo_valid |= OBD_MD_FLACL;
1008 } else if (rc == -EOPNOTSUPP) {
1011 CERROR("%s: unable to read "DFID
1013 mdt_obd_name(info->mti_mdt),
1014 PFID(mdt_object_fid(o)), rc);
1017 repbody->mbo_aclsize = rc;
1018 repbody->mbo_valid |= OBD_MD_FLACL;
1025 if (reqbody->mbo_valid & OBD_MD_FLMDSCAPA &&
1026 info->mti_mdt->mdt_lut.lut_mds_capa &&
1027 exp_connect_flags(info->mti_exp) & OBD_CONNECT_MDS_CAPA) {
1028 struct lustre_capa *capa;
1030 capa = req_capsule_server_get(pill, &RMF_CAPA1);
1032 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
1033 rc = mo_capa_get(env, next, capa, 0);
1036 repbody->mbo_valid |= OBD_MD_FLMDSCAPA;
1041 mdt_counter_incr(req, LPROC_MDT_GETATTR);
1046 static int mdt_renew_capa(struct mdt_thread_info *info)
1048 struct mdt_object *obj = info->mti_object;
1049 struct mdt_body *body;
1050 struct lustre_capa *capa, *c;
1054 /* if object doesn't exist, or server has disabled capability,
1055 * return directly, client will find body->valid OBD_MD_FLOSSCAPA
1058 if (!obj || !info->mti_mdt->mdt_lut.lut_oss_capa ||
1059 !(exp_connect_flags(info->mti_exp) & OBD_CONNECT_OSS_CAPA))
1062 body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1063 LASSERT(body != NULL);
1065 c = req_capsule_client_get(info->mti_pill, &RMF_CAPA1);
1068 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
1072 rc = mo_capa_get(info->mti_env, mdt_object_child(obj), capa, 1);
1074 body->mbo_valid |= OBD_MD_FLOSSCAPA;
1078 static int mdt_getattr(struct tgt_session_info *tsi)
1080 struct mdt_thread_info *info = tsi2mdt_info(tsi);
1081 struct mdt_object *obj = info->mti_object;
1082 struct req_capsule *pill = info->mti_pill;
1083 struct mdt_body *reqbody;
1084 struct mdt_body *repbody;
1088 reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
1091 if (reqbody->mbo_valid & OBD_MD_FLOSSCAPA) {
1092 rc = req_capsule_server_pack(pill);
1094 RETURN(err_serious(rc));
1095 rc = mdt_renew_capa(info);
1096 GOTO(out_shrink, rc);
1099 LASSERT(obj != NULL);
1100 LASSERT(lu_object_assert_exists(&obj->mot_obj));
1102 /* Unlike intent case where we need to pre-fill out buffers early on
1103 * in intent policy for ldlm reasons, here we can have a much better
1104 * guess at EA size by just reading it from disk.
1105 * Exceptions are readdir and (missing) directory striping */
1107 if (reqbody->mbo_valid & OBD_MD_LINKNAME) {
1108 /* No easy way to know how long is the symlink, but it cannot
1109 * be more than PATH_MAX, so we allocate +1 */
1112 /* A special case for fs ROOT: getattr there might fetch
1113 * default EA for entire fs, not just for this dir!
1115 } else if (lu_fid_eq(mdt_object_fid(obj),
1116 &info->mti_mdt->mdt_md_root_fid) &&
1117 (reqbody->mbo_valid & OBD_MD_FLDIREA) &&
1118 (lustre_msg_get_opc(mdt_info_req(info)->rq_reqmsg) ==
1120 /* Should the default strping be bigger, mdt_fix_reply
1121 * will reallocate */
1122 rc = DEF_REP_MD_SIZE;
1124 /* Read the actual EA size from disk */
1125 rc = mdt_attr_get_eabuf_size(info, obj);
1129 GOTO(out_shrink, rc);
1131 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, rc);
1133 rc = req_capsule_server_pack(pill);
1134 if (unlikely(rc != 0))
1135 GOTO(out, rc = err_serious(rc));
1137 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1138 LASSERT(repbody != NULL);
1139 repbody->mbo_eadatasize = 0;
1140 repbody->mbo_aclsize = 0;
1142 if (reqbody->mbo_valid & OBD_MD_FLRMTPERM)
1143 rc = mdt_init_ucred(info, reqbody);
1145 rc = mdt_check_ucred(info);
1147 GOTO(out_shrink, rc);
1149 info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1152 * Don't check capability at all, because rename might getattr for
1153 * remote obj, and at that time no capability is available.
1155 mdt_set_capainfo(info, 1, &reqbody->mbo_fid1, BYPASS_CAPA);
1156 rc = mdt_getattr_internal(info, obj, 0);
1157 if (reqbody->mbo_valid & OBD_MD_FLRMTPERM)
1158 mdt_exit_ucred(info);
1161 mdt_client_compatibility(info);
1162 rc2 = mdt_fix_reply(info);
1166 mdt_thread_info_fini(info);
1170 static int mdt_swap_layouts(struct tgt_session_info *tsi)
1172 struct mdt_thread_info *info;
1173 struct ptlrpc_request *req = tgt_ses_req(tsi);
1174 struct obd_export *exp = req->rq_export;
1175 struct mdt_object *o1, *o2, *o;
1176 struct mdt_lock_handle *lh1, *lh2;
1177 struct mdc_swap_layouts *msl;
1181 /* client does not support layout lock, so layout swaping
1183 * FIXME: there is a problem for old clients which don't support
1184 * layout lock yet. If those clients have already opened the file
1185 * they won't be notified at all so that old layout may still be
1186 * used to do IO. This can be fixed after file release is landed by
1187 * doing exclusive open and taking full EX ibits lock. - Jinshan */
1188 if (!exp_connect_layout(exp))
1189 RETURN(-EOPNOTSUPP);
1191 info = tsi2mdt_info(tsi);
1193 if (info->mti_dlm_req != NULL)
1194 ldlm_request_cancel(req, info->mti_dlm_req, 0);
1196 if (req_capsule_get_size(info->mti_pill, &RMF_CAPA1, RCL_CLIENT))
1197 mdt_set_capainfo(info, 0, &info->mti_body->mbo_fid1,
1198 req_capsule_client_get(info->mti_pill,
1201 if (req_capsule_get_size(info->mti_pill, &RMF_CAPA2, RCL_CLIENT))
1202 mdt_set_capainfo(info, 1, &info->mti_body->mbo_fid2,
1203 req_capsule_client_get(info->mti_pill,
1206 o1 = info->mti_object;
1207 o = o2 = mdt_object_find(info->mti_env, info->mti_mdt,
1208 &info->mti_body->mbo_fid2);
1210 GOTO(out, rc = PTR_ERR(o));
1212 if (mdt_object_remote(o) || !mdt_object_exists(o)) /* remote object */
1213 GOTO(put, rc = -ENOENT);
1215 rc = lu_fid_cmp(&info->mti_body->mbo_fid1, &info->mti_body->mbo_fid2);
1216 if (unlikely(rc == 0)) /* same file, you kidding me? no-op. */
1222 /* permission check. Make sure the calling process having permission
1223 * to write both files. */
1224 rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
1229 rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL,
1234 msl = req_capsule_client_get(info->mti_pill, &RMF_SWAP_LAYOUTS);
1236 GOTO(put, rc = -EPROTO);
1238 lh1 = &info->mti_lh[MDT_LH_NEW];
1239 mdt_lock_reg_init(lh1, LCK_EX);
1240 lh2 = &info->mti_lh[MDT_LH_OLD];
1241 mdt_lock_reg_init(lh2, LCK_EX);
1243 rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT |
1244 MDS_INODELOCK_XATTR, MDT_LOCAL_LOCK);
1248 rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT |
1249 MDS_INODELOCK_XATTR, MDT_LOCAL_LOCK);
1253 rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1),
1254 mdt_object_child(o2), msl->msl_flags);
1257 mdt_object_unlock(info, o2, lh2, rc);
1259 mdt_object_unlock(info, o1, lh1, rc);
1261 mdt_object_put(info->mti_env, o);
1263 mdt_thread_info_fini(info);
1267 static int mdt_raw_lookup(struct mdt_thread_info *info,
1268 struct mdt_object *parent,
1269 const struct lu_name *lname,
1270 struct ldlm_reply *ldlm_rep)
1272 struct md_object *next = mdt_object_child(info->mti_object);
1273 const struct mdt_body *reqbody = info->mti_body;
1274 struct lu_fid *child_fid = &info->mti_tmp_fid1;
1275 struct mdt_body *repbody;
1279 if (reqbody->mbo_valid != OBD_MD_FLID)
1282 LASSERT(!info->mti_cross_ref);
1284 /* Only got the fid of this obj by name */
1285 fid_zero(child_fid);
1286 rc = mdo_lookup(info->mti_env, next, lname, child_fid,
1289 /* XXX is raw_lookup possible as intent operation? */
1292 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1295 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1297 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1300 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1301 repbody->mbo_fid1 = *child_fid;
1302 repbody->mbo_valid = OBD_MD_FLID;
1308 * UPDATE lock should be taken against parent, and be release before exit;
1309 * child_bits lock should be taken against child, and be returned back:
1310 * (1)normal request should release the child lock;
1311 * (2)intent request will grant the lock to client.
1313 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
1314 struct mdt_lock_handle *lhc,
1316 struct ldlm_reply *ldlm_rep)
1318 struct ptlrpc_request *req = mdt_info_req(info);
1319 struct mdt_body *reqbody = NULL;
1320 struct mdt_object *parent = info->mti_object;
1321 struct mdt_object *child;
1322 struct md_object *next = mdt_object_child(parent);
1323 struct lu_fid *child_fid = &info->mti_tmp_fid1;
1324 struct lu_name *lname = NULL;
1325 struct mdt_lock_handle *lhp = NULL;
1326 struct ldlm_lock *lock;
1333 is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
1334 LASSERT(ergo(is_resent,
1335 lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
1337 LASSERT(parent != NULL);
1339 if (info->mti_cross_ref) {
1340 /* Only getattr on the child. Parent is on another node. */
1341 mdt_set_disposition(info, ldlm_rep,
1342 DISP_LOOKUP_EXECD | DISP_LOOKUP_POS);
1344 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
1346 PFID(mdt_object_fid(child)), ldlm_rep);
1348 rc = mdt_check_resent_lock(info, child, lhc);
1351 } else if (rc > 0) {
1352 mdt_lock_handle_init(lhc);
1353 mdt_lock_reg_init(lhc, LCK_PR);
1356 * Object's name is on another MDS, no lookup or layout
1357 * lock is needed here but update lock is.
1359 child_bits &= ~(MDS_INODELOCK_LOOKUP |
1360 MDS_INODELOCK_LAYOUT);
1361 child_bits |= MDS_INODELOCK_PERM | MDS_INODELOCK_UPDATE;
1363 rc = mdt_object_lock(info, child, lhc, child_bits,
1369 /* Finally, we can get attr for child. */
1370 if (!mdt_object_exists(child)) {
1371 LU_OBJECT_DEBUG(D_INFO, info->mti_env,
1373 "remote object doesn't exist.\n");
1374 mdt_object_unlock(info, child, lhc, 1);
1378 mdt_set_capainfo(info, 0, mdt_object_fid(child), BYPASS_CAPA);
1379 rc = mdt_getattr_internal(info, child, 0);
1380 if (unlikely(rc != 0))
1381 mdt_object_unlock(info, child, lhc, 1);
1386 lname = &info->mti_name;
1387 mdt_name_unpack(info->mti_pill, &RMF_NAME, lname, MNF_FIX_ANON);
1389 if (lu_name_is_valid(lname)) {
1390 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DNAME", "
1391 "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
1392 PNAME(lname), ldlm_rep);
1394 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1395 if (unlikely(reqbody == NULL))
1396 RETURN(err_serious(-EPROTO));
1398 *child_fid = reqbody->mbo_fid2;
1400 if (unlikely(!fid_is_sane(child_fid)))
1401 RETURN(err_serious(-EINVAL));
1403 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
1405 PFID(mdt_object_fid(parent)),
1406 PFID(&reqbody->mbo_fid2), ldlm_rep);
1409 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
1411 if (unlikely(!mdt_object_exists(parent)) && lu_name_is_valid(lname)) {
1412 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1414 "Parent doesn't exist!\n");
1418 if (mdt_object_remote(parent)) {
1419 CERROR("%s: parent "DFID" is on remote target\n",
1420 mdt_obd_name(info->mti_mdt),
1421 PFID(mdt_object_fid(parent)));
1425 if (lu_name_is_valid(lname)) {
1426 rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
1433 /* step 1: lock parent only if parent is a directory */
1434 if (S_ISDIR(lu_object_attr(&parent->mot_obj))) {
1435 lhp = &info->mti_lh[MDT_LH_PARENT];
1436 mdt_lock_pdo_init(lhp, LCK_PR, lname);
1437 rc = mdt_object_lock(info, parent, lhp,
1438 MDS_INODELOCK_UPDATE,
1440 if (unlikely(rc != 0))
1444 /* step 2: lookup child's fid by name */
1445 fid_zero(child_fid);
1446 rc = mdo_lookup(info->mti_env, next, lname, child_fid,
1449 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1452 GOTO(out_parent, rc);
1455 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1458 *step 3: find the child object by fid & lock it.
1459 * regardless if it is local or remote.
1461 *Note: LU-3240 (commit 762f2114d282a98ebfa4dbbeea9298a8088ad24e)
1462 * set parent dir fid the same as child fid in getattr by fid case
1463 * we should not lu_object_find() the object again, could lead
1464 * to hung if there is a concurrent unlink destroyed the object.
1466 if (lu_fid_eq(mdt_object_fid(parent), child_fid)) {
1467 mdt_object_get(info->mti_env, parent);
1470 child = mdt_object_find(info->mti_env, info->mti_mdt,
1474 if (unlikely(IS_ERR(child)))
1475 GOTO(out_parent, rc = PTR_ERR(child));
1477 rc = mdt_check_resent_lock(info, child, lhc);
1479 GOTO(out_child, rc);
1480 } else if (rc > 0) {
1481 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout*2);
1482 mdt_lock_handle_init(lhc);
1483 mdt_lock_reg_init(lhc, LCK_PR);
1486 if (!mdt_object_exists(child)) {
1487 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1489 "Object doesn't exist!\n");
1490 GOTO(out_child, rc = -ENOENT);
1493 if (!(child_bits & MDS_INODELOCK_UPDATE) &&
1494 mdt_object_exists(child) && !mdt_object_remote(child)) {
1495 struct md_attr *ma = &info->mti_attr;
1498 ma->ma_need = MA_INODE;
1499 rc = mdt_attr_get_complex(info, child, ma);
1500 if (unlikely(rc != 0))
1501 GOTO(out_child, rc);
1503 /* If the file has not been changed for some time, we
1504 * return not only a LOOKUP lock, but also an UPDATE
1505 * lock and this might save us RPC on later STAT. For
1506 * directories, it also let negative dentry cache start
1507 * working for this dir. */
1508 if (ma->ma_valid & MA_INODE &&
1509 ma->ma_attr.la_valid & LA_CTIME &&
1510 info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
1511 ma->ma_attr.la_ctime < cfs_time_current_sec())
1512 child_bits |= MDS_INODELOCK_UPDATE;
1515 /* layout lock must be granted in a best-effort way
1516 * for IT operations */
1517 LASSERT(!(child_bits & MDS_INODELOCK_LAYOUT));
1518 if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) &&
1519 exp_connect_layout(info->mti_exp) &&
1520 S_ISREG(lu_object_attr(&child->mot_obj)) &&
1521 !mdt_object_remote(child) && ldlm_rep != NULL) {
1522 /* try to grant layout lock for regular file. */
1528 child_bits |= MDS_INODELOCK_LAYOUT;
1529 /* try layout lock, it may fail to be granted due to
1530 * contention at LOOKUP or UPDATE */
1531 if (!mdt_object_lock_try(info, child, lhc, child_bits,
1533 child_bits &= ~MDS_INODELOCK_LAYOUT;
1534 LASSERT(child_bits != 0);
1535 rc = mdt_object_lock(info, child, lhc,
1536 child_bits, MDT_CROSS_LOCK);
1541 /* Do not enqueue the UPDATE lock from MDT(cross-MDT),
1542 * client will enqueue the lock to the remote MDT */
1543 if (mdt_object_remote(child))
1544 child_bits &= ~MDS_INODELOCK_UPDATE;
1545 rc = mdt_object_lock(info, child, lhc, child_bits,
1548 if (unlikely(rc != 0))
1549 GOTO(out_child, rc);
1552 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1553 /* Get MA_SOM attributes if update lock is given. */
1555 lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_UPDATE &&
1556 S_ISREG(lu_object_attr(&mdt_object_child(child)->mo_lu)))
1559 /* finally, we can get attr for child. */
1560 mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
1561 rc = mdt_getattr_internal(info, child, ma_need);
1562 if (unlikely(rc != 0)) {
1563 mdt_object_unlock(info, child, lhc, 1);
1565 /* Debugging code. */
1566 LDLM_DEBUG(lock, "Returning lock to client");
1567 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1568 &lock->l_resource->lr_name),
1569 "Lock res_id: "DLDLMRES", fid: "DFID"\n",
1570 PLDLMRES(lock->l_resource),
1571 PFID(mdt_object_fid(child)));
1572 if (mdt_object_exists(child) && !mdt_object_remote(child))
1573 mdt_pack_size2body(info, child);
1576 LDLM_LOCK_PUT(lock);
1580 mdt_object_put(info->mti_env, child);
1583 mdt_object_unlock(info, parent, lhp, 1);
1587 /* normal handler: should release the child lock */
1588 static int mdt_getattr_name(struct tgt_session_info *tsi)
1590 struct mdt_thread_info *info = tsi2mdt_info(tsi);
1591 struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1592 struct mdt_body *reqbody;
1593 struct mdt_body *repbody;
1597 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1598 LASSERT(reqbody != NULL);
1599 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1600 LASSERT(repbody != NULL);
1602 info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1603 repbody->mbo_eadatasize = 0;
1604 repbody->mbo_aclsize = 0;
1606 rc = mdt_init_ucred(info, reqbody);
1608 GOTO(out_shrink, rc);
1610 rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1611 if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1612 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1613 lhc->mlh_reg_lh.cookie = 0;
1615 mdt_exit_ucred(info);
1618 mdt_client_compatibility(info);
1619 rc2 = mdt_fix_reply(info);
1622 mdt_thread_info_fini(info);
1626 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1627 void *karg, void *uarg);
1629 static int mdt_set_info(struct tgt_session_info *tsi)
1631 struct ptlrpc_request *req = tgt_ses_req(tsi);
1634 int keylen, vallen, rc = 0;
1638 key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
1640 DEBUG_REQ(D_HA, req, "no set_info key");
1641 RETURN(err_serious(-EFAULT));
1644 keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
1647 val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
1649 DEBUG_REQ(D_HA, req, "no set_info val");
1650 RETURN(err_serious(-EFAULT));
1653 vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
1656 /* Swab any part of val you need to here */
1657 if (KEY_IS(KEY_READ_ONLY)) {
1658 spin_lock(&req->rq_export->exp_lock);
1660 *exp_connect_flags_ptr(req->rq_export) |=
1663 *exp_connect_flags_ptr(req->rq_export) &=
1664 ~OBD_CONNECT_RDONLY;
1665 spin_unlock(&req->rq_export->exp_lock);
1666 } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1667 struct changelog_setinfo *cs = val;
1669 if (vallen != sizeof(*cs)) {
1670 CERROR("%s: bad changelog_clear setinfo size %d\n",
1671 tgt_name(tsi->tsi_tgt), vallen);
1674 if (ptlrpc_req_need_swab(req)) {
1675 __swab64s(&cs->cs_recno);
1676 __swab32s(&cs->cs_id);
1679 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, req->rq_export,
1687 static int mdt_readpage(struct tgt_session_info *tsi)
1689 struct mdt_thread_info *info = mdt_th_info(tsi->tsi_env);
1690 struct mdt_object *object = mdt_obj(tsi->tsi_corpus);
1691 struct lu_rdpg *rdpg = &info->mti_u.rdpg.mti_rdpg;
1692 const struct mdt_body *reqbody = tsi->tsi_mdt_body;
1693 struct mdt_body *repbody;
1699 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
1700 RETURN(err_serious(-ENOMEM));
1702 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_MDT_BODY);
1703 if (repbody == NULL || reqbody == NULL)
1704 RETURN(err_serious(-EFAULT));
1707 * prepare @rdpg before calling lower layers and transfer itself. Here
1708 * reqbody->size contains offset of where to start to read and
1709 * reqbody->nlink contains number bytes to read.
1711 rdpg->rp_hash = reqbody->mbo_size;
1712 if (rdpg->rp_hash != reqbody->mbo_size) {
1713 CERROR("Invalid hash: "LPX64" != "LPX64"\n",
1714 rdpg->rp_hash, reqbody->mbo_size);
1718 rdpg->rp_attrs = reqbody->mbo_mode;
1719 if (exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_64BITHASH)
1720 rdpg->rp_attrs |= LUDA_64BITHASH;
1721 rdpg->rp_count = min_t(unsigned int, reqbody->mbo_nlink,
1722 exp_max_brw_size(tsi->tsi_exp));
1723 rdpg->rp_npages = (rdpg->rp_count + PAGE_CACHE_SIZE - 1) >>
1725 OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1726 if (rdpg->rp_pages == NULL)
1729 for (i = 0; i < rdpg->rp_npages; ++i) {
1730 rdpg->rp_pages[i] = alloc_page(GFP_IOFS);
1731 if (rdpg->rp_pages[i] == NULL)
1732 GOTO(free_rdpg, rc = -ENOMEM);
1735 /* call lower layers to fill allocated pages with directory data */
1736 rc = mo_readpage(tsi->tsi_env, mdt_object_child(object), rdpg);
1738 GOTO(free_rdpg, rc);
1740 /* send pages to client */
1741 rc = tgt_sendpage(tsi, rdpg, rc);
1746 for (i = 0; i < rdpg->rp_npages; i++)
1747 if (rdpg->rp_pages[i] != NULL)
1748 __free_page(rdpg->rp_pages[i]);
1749 OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1751 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1757 static int mdt_reint_internal(struct mdt_thread_info *info,
1758 struct mdt_lock_handle *lhc,
1761 struct req_capsule *pill = info->mti_pill;
1762 struct mdt_body *repbody;
1767 rc = mdt_reint_unpack(info, op);
1769 CERROR("Can't unpack reint, rc %d\n", rc);
1770 RETURN(err_serious(rc));
1773 /* for replay (no_create) lmm is not needed, client has it already */
1774 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1775 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1778 /* llog cookies are always 0, the field is kept for compatibility */
1779 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1780 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);
1782 rc = req_capsule_server_pack(pill);
1784 CERROR("Can't pack response, rc %d\n", rc);
1785 RETURN(err_serious(rc));
1788 if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
1789 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1791 repbody->mbo_eadatasize = 0;
1792 repbody->mbo_aclsize = 0;
1795 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
1797 /* for replay no cookkie / lmm need, because client have this already */
1798 if (info->mti_spec.no_create)
1799 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1800 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
1802 rc = mdt_init_ucred_reint(info);
1804 GOTO(out_shrink, rc);
1806 rc = mdt_fix_attr_ucred(info, op);
1808 GOTO(out_ucred, rc = err_serious(rc));
1810 if (mdt_check_resent(info, mdt_reconstruct, lhc)) {
1811 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt.");
1812 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
1813 GOTO(out_ucred, rc);
1815 rc = mdt_reint_rec(info, lhc);
1818 mdt_exit_ucred(info);
1820 mdt_client_compatibility(info);
1821 rc2 = mdt_fix_reply(info);
1827 static long mdt_reint_opcode(struct ptlrpc_request *req,
1828 const struct req_format **fmt)
1830 struct mdt_device *mdt;
1831 struct mdt_rec_reint *rec;
1834 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
1836 opc = rec->rr_opcode;
1837 DEBUG_REQ(D_INODE, req, "reint opt = %ld", opc);
1838 if (opc < REINT_MAX && fmt[opc] != NULL)
1839 req_capsule_extend(&req->rq_pill, fmt[opc]);
1841 mdt = mdt_exp2dev(req->rq_export);
1842 CERROR("%s: Unsupported opcode '%ld' from client '%s':"
1843 " rc = %d\n", req->rq_export->exp_obd->obd_name,
1844 opc, mdt->mdt_ldlm_client->cli_name, -EFAULT);
1845 opc = err_serious(-EFAULT);
1848 opc = err_serious(-EFAULT);
1853 static int mdt_reint(struct tgt_session_info *tsi)
1857 static const struct req_format *reint_fmts[REINT_MAX] = {
1858 [REINT_SETATTR] = &RQF_MDS_REINT_SETATTR,
1859 [REINT_CREATE] = &RQF_MDS_REINT_CREATE,
1860 [REINT_LINK] = &RQF_MDS_REINT_LINK,
1861 [REINT_UNLINK] = &RQF_MDS_REINT_UNLINK,
1862 [REINT_RENAME] = &RQF_MDS_REINT_RENAME,
1863 [REINT_OPEN] = &RQF_MDS_REINT_OPEN,
1864 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR,
1865 [REINT_RMENTRY] = &RQF_MDS_REINT_UNLINK,
1866 [REINT_MIGRATE] = &RQF_MDS_REINT_RENAME
1871 opc = mdt_reint_opcode(tgt_ses_req(tsi), reint_fmts);
1873 struct mdt_thread_info *info = tsi2mdt_info(tsi);
1875 * No lock possible here from client to pass it to reint code
1878 rc = mdt_reint_internal(info, NULL, opc);
1879 mdt_thread_info_fini(info);
1884 tsi->tsi_reply_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
1888 /* this should sync the whole device */
1889 static int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
1891 struct dt_device *dt = mdt->mdt_bottom;
1895 rc = dt->dd_ops->dt_sync(env, dt);
1899 /* this should sync this object */
1900 static int mdt_object_sync(struct mdt_thread_info *info)
1902 struct md_object *next;
1906 if (!mdt_object_exists(info->mti_object)) {
1907 CWARN("Non existing object "DFID"!\n",
1908 PFID(mdt_object_fid(info->mti_object)));
1911 next = mdt_object_child(info->mti_object);
1912 rc = mo_object_sync(info->mti_env, next);
1917 static int mdt_sync(struct tgt_session_info *tsi)
1919 struct ptlrpc_request *req = tgt_ses_req(tsi);
1920 struct req_capsule *pill = tsi->tsi_pill;
1921 struct mdt_body *body;
1926 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
1927 RETURN(err_serious(-ENOMEM));
1929 if (fid_seq(&tsi->tsi_mdt_body->mbo_fid1) == 0) {
1930 rc = mdt_device_sync(tsi->tsi_env, mdt_exp2dev(tsi->tsi_exp));
1932 struct mdt_thread_info *info = tsi2mdt_info(tsi);
1934 /* sync an object */
1935 rc = mdt_object_sync(info);
1937 const struct lu_fid *fid;
1938 struct lu_attr *la = &info->mti_attr.ma_attr;
1940 info->mti_attr.ma_need = MA_INODE;
1941 info->mti_attr.ma_valid = 0;
1942 rc = mdt_attr_get_complex(info, info->mti_object,
1945 body = req_capsule_server_get(pill,
1947 fid = mdt_object_fid(info->mti_object);
1948 mdt_pack_attr2body(info, body, la, fid);
1951 mdt_thread_info_fini(info);
1954 mdt_counter_incr(req, LPROC_MDT_SYNC);
1960 * Handle quota control requests to consult current usage/limit, but also
1961 * to configure quota enforcement
1963 static int mdt_quotactl(struct tgt_session_info *tsi)
1965 struct obd_export *exp = tsi->tsi_exp;
1966 struct req_capsule *pill = tsi->tsi_pill;
1967 struct obd_quotactl *oqctl, *repoqc;
1969 struct mdt_device *mdt = mdt_exp2dev(exp);
1970 struct lu_device *qmt = mdt->mdt_qmt_dev;
1973 oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
1975 RETURN(err_serious(-EPROTO));
1977 rc = req_capsule_server_pack(pill);
1979 RETURN(err_serious(rc));
1981 switch (oqctl->qc_cmd) {
1983 case LUSTRE_Q_INVALIDATE:
1984 case LUSTRE_Q_FINVALIDATE:
1988 /* deprecated, not used any more */
1989 RETURN(-EOPNOTSUPP);
1990 /* master quotactl */
1996 RETURN(-EOPNOTSUPP);
1997 /* slave quotactl */
2002 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2006 /* map uid/gid for remote client */
2008 if (exp_connect_rmtclient(exp)) {
2009 struct lustre_idmap_table *idmap;
2011 idmap = exp->exp_mdt_data.med_idmap;
2013 if (unlikely(oqctl->qc_cmd != Q_GETQUOTA &&
2014 oqctl->qc_cmd != Q_GETINFO))
2017 if (oqctl->qc_type == USRQUOTA)
2018 id = lustre_idmap_lookup_uid(NULL, idmap, 0,
2020 else if (oqctl->qc_type == GRPQUOTA)
2021 id = lustre_idmap_lookup_gid(NULL, idmap, 0,
2026 if (id == CFS_IDMAP_NOTFOUND) {
2027 CDEBUG(D_QUOTA, "no mapping for id %u\n", oqctl->qc_id);
2032 repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
2034 RETURN(err_serious(-EFAULT));
2036 if (oqctl->qc_id != id)
2037 swap(oqctl->qc_id, id);
2039 switch (oqctl->qc_cmd) {
2045 /* forward quotactl request to QMT */
2046 rc = qmt_hdls.qmth_quotactl(tsi->tsi_env, qmt, oqctl);
2051 /* slave quotactl */
2052 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom,
2057 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2061 if (oqctl->qc_id != id)
2062 swap(oqctl->qc_id, id);
2068 /** clone llog ctxt from child (mdd)
2069 * This allows remote llog (replicator) access.
2070 * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
2071 * context was originally set up, or we can handle them directly.
2072 * I choose the latter, but that means I need any llog
2073 * contexts set up by child to be accessable by the mdt. So we clone the
2074 * context into our context list here.
2076 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
2079 struct md_device *next = mdt->mdt_child;
2080 struct llog_ctxt *ctxt;
2083 if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
2086 rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
2087 if (rc || ctxt == NULL) {
2091 rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
2093 CERROR("Can't set mdt ctxt %d\n", rc);
2098 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
2099 struct mdt_device *mdt, int idx)
2101 struct llog_ctxt *ctxt;
2103 ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
2106 /* Put once for the get we just did, and once for the clone */
2107 llog_ctxt_put(ctxt);
2108 llog_ctxt_put(ctxt);
2113 * sec context handlers
2115 static int mdt_sec_ctx_handle(struct tgt_session_info *tsi)
2119 rc = mdt_handle_idmap(tsi);
2121 struct ptlrpc_request *req = tgt_ses_req(tsi);
2124 opc = lustre_msg_get_opc(req->rq_reqmsg);
2125 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT)
2126 sptlrpc_svc_ctx_invalidate(req);
2129 CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
2135 * quota request handlers
2137 static int mdt_quota_dqacq(struct tgt_session_info *tsi)
2139 struct mdt_device *mdt = mdt_exp2dev(tsi->tsi_exp);
2140 struct lu_device *qmt = mdt->mdt_qmt_dev;
2145 RETURN(err_serious(-EOPNOTSUPP));
2147 rc = qmt_hdls.qmth_dqacq(tsi->tsi_env, qmt, tgt_ses_req(tsi));
2151 struct mdt_object *mdt_object_new(const struct lu_env *env,
2152 struct mdt_device *d,
2153 const struct lu_fid *f)
2155 struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2156 struct lu_object *o;
2157 struct mdt_object *m;
2160 CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
2161 o = lu_object_find(env, &d->mdt_lu_dev, f, &conf);
2162 if (unlikely(IS_ERR(o)))
2163 m = (struct mdt_object *)o;
2169 struct mdt_object *mdt_object_find(const struct lu_env *env,
2170 struct mdt_device *d,
2171 const struct lu_fid *f)
2173 struct lu_object *o;
2174 struct mdt_object *m;
2177 CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2178 o = lu_object_find(env, &d->mdt_lu_dev, f, NULL);
2179 if (unlikely(IS_ERR(o)))
2180 m = (struct mdt_object *)o;
2188 * Asyncronous commit for mdt device.
2190 * Pass asynchonous commit call down the MDS stack.
2192 * \param env environment
2193 * \param mdt the mdt device
2195 static void mdt_device_commit_async(const struct lu_env *env,
2196 struct mdt_device *mdt)
2198 struct dt_device *dt = mdt->mdt_bottom;
2201 rc = dt->dd_ops->dt_commit_async(env, dt);
2202 if (unlikely(rc != 0))
2203 CWARN("async commit start failed with rc = %d", rc);
2207 * Mark the lock as "synchonous".
2209 * Mark the lock to deffer transaction commit to the unlock time.
2211 * \param lock the lock to mark as "synchonous"
2213 * \see mdt_is_lock_sync
2214 * \see mdt_save_lock
2216 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2218 lock->l_ast_data = (void*)1;
2222 * Check whehter the lock "synchonous" or not.
2224 * \param lock the lock to check
2225 * \retval 1 the lock is "synchonous"
2226 * \retval 0 the lock isn't "synchronous"
2228 * \see mdt_set_lock_sync
2229 * \see mdt_save_lock
2231 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2233 return lock->l_ast_data != NULL;
2237 * Blocking AST for mdt locks.
2239 * Starts transaction commit if in case of COS lock conflict or
2240 * deffers such a commit to the mdt_save_lock.
2242 * \param lock the lock which blocks a request or cancelling lock
2243 * \param desc unused
2244 * \param data unused
2245 * \param flag indicates whether this cancelling or blocking callback
2247 * \see ldlm_blocking_ast_nocheck
2249 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2250 void *data, int flag)
2252 struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2253 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2257 if (flag == LDLM_CB_CANCELING)
2259 lock_res_and_lock(lock);
2260 if (lock->l_blocking_ast != mdt_blocking_ast) {
2261 unlock_res_and_lock(lock);
2264 if (mdt_cos_is_enabled(mdt) &&
2265 lock->l_req_mode & (LCK_PW | LCK_EX) &&
2266 lock->l_blocking_lock != NULL &&
2267 lock->l_client_cookie != lock->l_blocking_lock->l_client_cookie) {
2268 mdt_set_lock_sync(lock);
2270 rc = ldlm_blocking_ast_nocheck(lock);
2272 /* There is no lock conflict if l_blocking_lock == NULL,
2273 * it indicates a blocking ast sent from ldlm_lock_decref_internal
2274 * when the last reference to a local lock was released */
2275 if (lock->l_req_mode == LCK_COS && lock->l_blocking_lock != NULL) {
2278 rc = lu_env_init(&env, LCT_LOCAL);
2279 if (unlikely(rc != 0))
2280 CWARN("lu_env initialization failed with rc = %d,"
2281 "cannot start asynchronous commit\n", rc);
2283 mdt_device_commit_async(&env, mdt);
2289 /* Used for cross-MDT lock */
2290 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2291 void *data, int flag)
2293 struct lustre_handle lockh;
2297 case LDLM_CB_BLOCKING:
2298 ldlm_lock2handle(lock, &lockh);
2299 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
2301 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
2305 case LDLM_CB_CANCELING:
2306 LDLM_DEBUG(lock, "Revoke remote lock\n");
2314 int mdt_check_resent_lock(struct mdt_thread_info *info,
2315 struct mdt_object *mo,
2316 struct mdt_lock_handle *lhc)
2318 /* the lock might already be gotten in ldlm_handle_enqueue() */
2319 if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
2320 struct ptlrpc_request *req = mdt_info_req(info);
2321 struct ldlm_lock *lock;
2323 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
2324 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT);
2325 LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
2326 lhc->mlh_reg_lh.cookie);
2328 if (!fid_res_name_eq(mdt_object_fid(mo),
2329 &lock->l_resource->lr_name)) {
2330 CWARN("%s: Although resent, but still not "
2331 "get child lock:"DFID"\n",
2332 info->mti_exp->exp_obd->obd_name,
2333 PFID(mdt_object_fid(mo)));
2334 LDLM_LOCK_PUT(lock);
2337 LDLM_LOCK_PUT(lock);
2343 int mdt_remote_object_lock(struct mdt_thread_info *mti,
2344 struct mdt_object *o, const struct lu_fid *fid,
2345 struct lustre_handle *lh, ldlm_mode_t mode,
2348 struct ldlm_enqueue_info *einfo = &mti->mti_einfo;
2349 ldlm_policy_data_t *policy = &mti->mti_policy;
2350 struct ldlm_res_id *res_id = &mti->mti_res_id;
2354 LASSERT(mdt_object_remote(o));
2356 LASSERT(ibits == MDS_INODELOCK_UPDATE);
2358 fid_build_reg_res_name(fid, res_id);
2360 memset(einfo, 0, sizeof(*einfo));
2361 einfo->ei_type = LDLM_IBITS;
2362 einfo->ei_mode = mode;
2363 einfo->ei_cb_bl = mdt_remote_blocking_ast;
2364 einfo->ei_cb_cp = ldlm_completion_ast;
2365 einfo->ei_enq_slave = 0;
2366 einfo->ei_res_id = res_id;
2368 memset(policy, 0, sizeof(*policy));
2369 policy->l_inodebits.bits = ibits;
2371 rc = mo_object_lock(mti->mti_env, mdt_object_child(o), lh, einfo,
2376 static int mdt_object_local_lock(struct mdt_thread_info *info,
2377 struct mdt_object *o,
2378 struct mdt_lock_handle *lh, __u64 ibits,
2379 bool nonblock, int locality)
2381 struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
2382 ldlm_policy_data_t *policy = &info->mti_policy;
2383 struct ldlm_res_id *res_id = &info->mti_res_id;
2388 LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2389 LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2390 LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
2391 LASSERT(lh->mlh_type != MDT_NUL_LOCK);
2393 /* Only enqueue LOOKUP lock for remote object */
2394 if (mdt_object_remote(o))
2395 LASSERT(ibits == MDS_INODELOCK_LOOKUP);
2397 if (lh->mlh_type == MDT_PDO_LOCK) {
2398 /* check for exists after object is locked */
2399 if (mdt_object_exists(o) == 0) {
2400 /* Non-existent object shouldn't have PDO lock */
2403 /* Non-dir object shouldn't have PDO lock */
2404 if (!S_ISDIR(lu_object_attr(&o->mot_obj)))
2409 memset(policy, 0, sizeof(*policy));
2410 fid_build_reg_res_name(mdt_object_fid(o), res_id);
2412 dlmflags = LDLM_FL_ATOMIC_CB;
2414 dlmflags |= LDLM_FL_BLOCK_NOWAIT;
2417 * Take PDO lock on whole directory and build correct @res_id for lock
2418 * on part of directory.
2420 if (lh->mlh_pdo_hash != 0) {
2421 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
2422 mdt_lock_pdo_mode(info, o, lh);
2423 if (lh->mlh_pdo_mode != LCK_NL) {
2425 * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
2426 * is never going to be sent to client and we do not
2427 * want it slowed down due to possible cancels.
2429 policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
2430 rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
2431 policy, res_id, dlmflags,
2432 info->mti_exp == NULL ? NULL :
2433 &info->mti_exp->exp_handle.h_cookie);
2439 * Finish res_id initializing by name hash marking part of
2440 * directory which is taking modification.
2442 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
2445 policy->l_inodebits.bits = ibits;
2448 * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
2449 * going to be sent to client. If it is - mdt_intent_policy() path will
2450 * fix it up and turn FL_LOCAL flag off.
2452 rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
2453 res_id, LDLM_FL_LOCAL_ONLY | dlmflags,
2454 info->mti_exp == NULL ? NULL :
2455 &info->mti_exp->exp_handle.h_cookie);
2457 mdt_object_unlock(info, o, lh, 1);
2458 else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
2459 lh->mlh_pdo_hash != 0 &&
2460 (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX)) {
2461 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
2468 mdt_object_lock_internal(struct mdt_thread_info *info, struct mdt_object *o,
2469 struct mdt_lock_handle *lh, __u64 ibits,
2470 bool nonblock, int locality)
2475 if (!mdt_object_remote(o))
2476 return mdt_object_local_lock(info, o, lh, ibits, nonblock,
2479 if (locality == MDT_LOCAL_LOCK) {
2480 CERROR("%s: try to get local lock for remote object"
2481 DFID".\n", mdt_obd_name(info->mti_mdt),
2482 PFID(mdt_object_fid(o)));
2486 /* XXX do not support PERM/LAYOUT/XATTR lock for remote object yet */
2487 ibits &= ~(MDS_INODELOCK_PERM | MDS_INODELOCK_LAYOUT |
2488 MDS_INODELOCK_XATTR);
2489 if (ibits & MDS_INODELOCK_UPDATE) {
2490 /* Sigh, PDO needs to enqueue 2 locks right now, but
2491 * enqueue RPC can only request 1 lock, to avoid extra
2492 * RPC, so it will instead enqueue EX lock for remote
2493 * object anyway XXX*/
2494 if (lh->mlh_type == MDT_PDO_LOCK &&
2495 lh->mlh_pdo_hash != 0) {
2496 CDEBUG(D_INFO, "%s: "DFID" convert PDO lock to"
2497 "EX lock.\n", mdt_obd_name(info->mti_mdt),
2498 PFID(mdt_object_fid(o)));
2499 lh->mlh_pdo_hash = 0;
2500 lh->mlh_rreg_mode = LCK_EX;
2501 lh->mlh_type = MDT_REG_LOCK;
2503 rc = mdt_remote_object_lock(info, o, mdt_object_fid(o),
2506 MDS_INODELOCK_UPDATE);
2511 /* Only enqueue LOOKUP lock for remote object */
2512 if (ibits & MDS_INODELOCK_LOOKUP) {
2513 rc = mdt_object_local_lock(info, o, lh,
2514 MDS_INODELOCK_LOOKUP,
2515 nonblock, locality);
2523 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2524 struct mdt_lock_handle *lh, __u64 ibits, int locality)
2526 return mdt_object_lock_internal(info, o, lh, ibits, false, locality);
2529 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o,
2530 struct mdt_lock_handle *lh, __u64 ibits, int locality)
2532 struct mdt_lock_handle tmp = *lh;
2535 rc = mdt_object_lock_internal(info, o, &tmp, ibits, true, locality);
2543 * Save a lock within request object.
2545 * Keep the lock referenced until whether client ACK or transaction
2546 * commit happens or release the lock immediately depending on input
2547 * parameters. If COS is ON, a write lock is converted to COS lock
2550 * \param info thead info object
2551 * \param h lock handle
2552 * \param mode lock mode
2553 * \param decref force immediate lock releasing
2556 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
2557 ldlm_mode_t mode, int decref)
2561 if (lustre_handle_is_used(h)) {
2562 if (decref || !info->mti_has_trans ||
2563 !(mode & (LCK_PW | LCK_EX))){
2564 mdt_fid_unlock(h, mode);
2566 struct mdt_device *mdt = info->mti_mdt;
2567 struct ldlm_lock *lock = ldlm_handle2lock(h);
2568 struct ptlrpc_request *req = mdt_info_req(info);
2571 LASSERTF(lock != NULL, "no lock for cookie "LPX64"\n",
2573 /* there is no request if mdt_object_unlock() is called
2574 * from mdt_export_cleanup()->mdt_add_dirty_flag() */
2575 if (likely(req != NULL)) {
2576 CDEBUG(D_HA, "request = %p reply state = %p"
2577 " transno = "LPD64"\n", req,
2578 req->rq_reply_state, req->rq_transno);
2579 if (mdt_cos_is_enabled(mdt)) {
2581 ldlm_lock_downgrade(lock, LCK_COS);
2584 ptlrpc_save_lock(req, h, mode, no_ack);
2586 ldlm_lock_decref(h, mode);
2588 if (mdt_is_lock_sync(lock)) {
2589 CDEBUG(D_HA, "found sync-lock,"
2590 " async commit started\n");
2591 mdt_device_commit_async(info->mti_env,
2594 LDLM_LOCK_PUT(lock);
2603 * Unlock mdt object.
2605 * Immeditely release the regular lock and the PDO lock or save the
2606 * lock in reqeuest and keep them referenced until client ACK or
2607 * transaction commit.
2609 * \param info thread info object
2610 * \param o mdt object
2611 * \param lh mdt lock handle referencing regular and PDO locks
2612 * \param decref force immediate lock releasing
2614 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
2615 struct mdt_lock_handle *lh, int decref)
2619 mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
2620 mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
2622 if (lustre_handle_is_used(&lh->mlh_rreg_lh))
2623 ldlm_lock_decref(&lh->mlh_rreg_lh, lh->mlh_rreg_mode);
2628 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
2629 const struct lu_fid *f,
2630 struct mdt_lock_handle *lh,
2633 struct mdt_object *o;
2635 o = mdt_object_find(info->mti_env, info->mti_mdt, f);
2639 rc = mdt_object_lock(info, o, lh, ibits,
2642 mdt_object_put(info->mti_env, o);
2649 void mdt_object_unlock_put(struct mdt_thread_info * info,
2650 struct mdt_object * o,
2651 struct mdt_lock_handle *lh,
2654 mdt_object_unlock(info, o, lh, decref);
2655 mdt_object_put(info->mti_env, o);
2659 * Generic code handling requests that have struct mdt_body passed in:
2661 * - extract mdt_body from request and save it in @info, if present;
2663 * - create lu_object, corresponding to the fid in mdt_body, and save it in
2666 * - if HABEO_CORPUS flag is set for this request type check whether object
2667 * actually exists on storage (lu_object_exists()).
2670 static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags)
2672 const struct mdt_body *body;
2673 struct mdt_object *obj;
2674 const struct lu_env *env;
2675 struct req_capsule *pill;
2679 env = info->mti_env;
2680 pill = info->mti_pill;
2682 body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
2686 if (!(body->mbo_valid & OBD_MD_FLID))
2689 if (!fid_is_sane(&body->mbo_fid1)) {
2690 CERROR("Invalid fid: "DFID"\n", PFID(&body->mbo_fid1));
2695 * Do not get size or any capa fields before we check that request
2696 * contains capa actually. There are some requests which do not, for
2697 * instance MDS_IS_SUBDIR.
2699 if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_CLIENT) &&
2700 req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
2701 mdt_set_capainfo(info, 0, &body->mbo_fid1,
2702 req_capsule_client_get(pill, &RMF_CAPA1));
2704 obj = mdt_object_find(env, info->mti_mdt, &body->mbo_fid1);
2706 if ((flags & HABEO_CORPUS) &&
2707 !mdt_object_exists(obj)) {
2708 mdt_object_put(env, obj);
2709 /* for capability renew ENOENT will be handled in
2711 if (body->mbo_valid & OBD_MD_FLOSSCAPA)
2716 info->mti_object = obj;
2725 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags)
2727 struct req_capsule *pill = info->mti_pill;
2731 if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
2732 rc = mdt_body_unpack(info, flags);
2736 if (rc == 0 && (flags & HABEO_REFERO)) {
2738 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2739 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2741 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2742 req_capsule_set_size(pill, &RMF_LOGCOOKIES,
2745 rc = req_capsule_server_pack(pill);
2750 static int mdt_init_capa_ctxt(const struct lu_env *env, struct mdt_device *m)
2752 struct md_device *next = m->mdt_child;
2754 return next->md_ops->mdo_init_capa_ctxt(env, next,
2755 m->mdt_lut.lut_mds_capa,
2756 m->mdt_capa_timeout,
2761 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
2763 lh->mlh_type = MDT_NUL_LOCK;
2764 lh->mlh_reg_lh.cookie = 0ull;
2765 lh->mlh_reg_mode = LCK_MINMODE;
2766 lh->mlh_pdo_lh.cookie = 0ull;
2767 lh->mlh_pdo_mode = LCK_MINMODE;
2768 lh->mlh_rreg_lh.cookie = 0ull;
2769 lh->mlh_rreg_mode = LCK_MINMODE;
2772 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
2774 LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2775 LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2779 * Initialize fields of struct mdt_thread_info. Other fields are left in
2780 * uninitialized state, because it's too expensive to zero out whole
2781 * mdt_thread_info (> 1K) on each request arrival.
2783 void mdt_thread_info_init(struct ptlrpc_request *req,
2784 struct mdt_thread_info *info)
2788 info->mti_pill = &req->rq_pill;
2791 for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2792 mdt_lock_handle_init(&info->mti_lh[i]);
2794 /* mdt device: it can be NULL while CONNECT */
2795 if (req->rq_export) {
2796 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
2797 info->mti_exp = req->rq_export;
2799 info->mti_mdt = NULL;
2800 info->mti_env = req->rq_svc_thread->t_env;
2801 info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
2803 memset(&info->mti_attr, 0, sizeof(info->mti_attr));
2804 info->mti_big_buf = LU_BUF_NULL;
2805 info->mti_body = NULL;
2806 info->mti_object = NULL;
2807 info->mti_dlm_req = NULL;
2808 info->mti_has_trans = 0;
2809 info->mti_cross_ref = 0;
2810 info->mti_opdata = 0;
2811 info->mti_big_lmm_used = 0;
2813 info->mti_spec.no_create = 0;
2814 info->mti_spec.sp_rm_entry = 0;
2816 info->mti_spec.u.sp_ea.eadata = NULL;
2817 info->mti_spec.u.sp_ea.eadatalen = 0;
2820 void mdt_thread_info_fini(struct mdt_thread_info *info)
2824 if (info->mti_object != NULL) {
2825 mdt_object_put(info->mti_env, info->mti_object);
2826 info->mti_object = NULL;
2829 for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2830 mdt_lock_handle_fini(&info->mti_lh[i]);
2831 info->mti_env = NULL;
2832 info->mti_pill = NULL;
2833 info->mti_exp = NULL;
2835 if (unlikely(info->mti_big_buf.lb_buf != NULL))
2836 lu_buf_free(&info->mti_big_buf);
2839 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi)
2841 struct mdt_thread_info *mti;
2842 struct lustre_capa *lc;
2844 mti = mdt_th_info(tsi->tsi_env);
2845 LASSERT(mti != NULL);
2847 mdt_thread_info_init(tgt_ses_req(tsi), mti);
2848 if (tsi->tsi_corpus != NULL) {
2849 struct req_capsule *pill = tsi->tsi_pill;
2851 mti->mti_object = mdt_obj(tsi->tsi_corpus);
2852 lu_object_get(tsi->tsi_corpus);
2855 * XXX: must be part of tgt_mdt_body_unpack but moved here
2856 * due to mdt_set_capainfo().
2858 if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_CLIENT) &&
2859 req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT) > 0) {
2860 lc = req_capsule_client_get(pill, &RMF_CAPA1);
2861 mdt_set_capainfo(mti, 0, &tsi->tsi_mdt_body->mbo_fid1,
2865 mti->mti_body = tsi->tsi_mdt_body;
2866 mti->mti_dlm_req = tsi->tsi_dlm_req;
2871 static int mdt_tgt_connect(struct tgt_session_info *tsi)
2873 struct ptlrpc_request *req = tgt_ses_req(tsi);
2878 rc = tgt_connect(tsi);
2882 rc = mdt_init_idmap(tsi);
2887 obd_disconnect(class_export_get(req->rq_export));
2906 static int mdt_intent_getattr(enum mdt_it_code opcode,
2907 struct mdt_thread_info *info,
2908 struct ldlm_lock **,
2911 static int mdt_intent_getxattr(enum mdt_it_code opcode,
2912 struct mdt_thread_info *info,
2913 struct ldlm_lock **lockp,
2916 static int mdt_intent_layout(enum mdt_it_code opcode,
2917 struct mdt_thread_info *info,
2918 struct ldlm_lock **,
2920 static int mdt_intent_reint(enum mdt_it_code opcode,
2921 struct mdt_thread_info *info,
2922 struct ldlm_lock **,
2925 static struct mdt_it_flavor {
2926 const struct req_format *it_fmt;
2928 int (*it_act)(enum mdt_it_code ,
2929 struct mdt_thread_info *,
2930 struct ldlm_lock **,
2933 } mdt_it_flavor[] = {
2935 .it_fmt = &RQF_LDLM_INTENT,
2936 /*.it_flags = HABEO_REFERO,*/
2938 .it_act = mdt_intent_reint,
2939 .it_reint = REINT_OPEN
2942 .it_fmt = &RQF_LDLM_INTENT,
2944 * OCREAT is not a MUTABOR request as if the file
2946 * We do the extra check of OBD_CONNECT_RDONLY in
2947 * mdt_reint_open() when we really need to create
2951 .it_act = mdt_intent_reint,
2952 .it_reint = REINT_OPEN
2955 .it_fmt = &RQF_LDLM_INTENT,
2956 .it_flags = MUTABOR,
2957 .it_act = mdt_intent_reint,
2958 .it_reint = REINT_CREATE
2960 [MDT_IT_GETATTR] = {
2961 .it_fmt = &RQF_LDLM_INTENT_GETATTR,
2962 .it_flags = HABEO_REFERO,
2963 .it_act = mdt_intent_getattr
2965 [MDT_IT_READDIR] = {
2971 .it_fmt = &RQF_LDLM_INTENT_GETATTR,
2972 .it_flags = HABEO_REFERO,
2973 .it_act = mdt_intent_getattr
2976 .it_fmt = &RQF_LDLM_INTENT_UNLINK,
2977 .it_flags = MUTABOR,
2979 .it_reint = REINT_UNLINK
2983 .it_flags = MUTABOR,
2986 [MDT_IT_GETXATTR] = {
2987 .it_fmt = &RQF_LDLM_INTENT_GETXATTR,
2988 .it_flags = HABEO_CORPUS,
2989 .it_act = mdt_intent_getxattr
2992 .it_fmt = &RQF_LDLM_INTENT_LAYOUT,
2994 .it_act = mdt_intent_layout
2999 mdt_intent_lock_replace(struct mdt_thread_info *info,
3000 struct ldlm_lock **lockp,
3001 struct mdt_lock_handle *lh,
3004 struct ptlrpc_request *req = mdt_info_req(info);
3005 struct ldlm_lock *lock = *lockp;
3006 struct ldlm_lock *new_lock;
3008 /* If possible resent found a lock, @lh is set to its handle */
3009 new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
3011 if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
3012 lh->mlh_reg_lh.cookie = 0;
3016 LASSERTF(new_lock != NULL,
3017 "lockh "LPX64"\n", lh->mlh_reg_lh.cookie);
3020 * If we've already given this lock to a client once, then we should
3021 * have no readers or writers. Otherwise, we should have one reader
3022 * _or_ writer ref (which will be zeroed below) before returning the
3025 if (new_lock->l_export == req->rq_export) {
3026 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
3028 LASSERT(new_lock->l_export == NULL);
3029 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
3034 if (new_lock->l_export == req->rq_export) {
3036 * Already gave this to the client, which means that we
3037 * reconstructed a reply.
3039 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
3042 LDLM_LOCK_RELEASE(new_lock);
3043 lh->mlh_reg_lh.cookie = 0;
3044 RETURN(ELDLM_LOCK_REPLACED);
3048 * Fixup the lock to be given to the client.
3050 lock_res_and_lock(new_lock);
3051 /* Zero new_lock->l_readers and new_lock->l_writers without triggering
3052 * possible blocking AST. */
3053 while (new_lock->l_readers > 0) {
3054 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
3055 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3056 new_lock->l_readers--;
3058 while (new_lock->l_writers > 0) {
3059 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
3060 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3061 new_lock->l_writers--;
3064 new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
3065 new_lock->l_blocking_ast = lock->l_blocking_ast;
3066 new_lock->l_completion_ast = lock->l_completion_ast;
3067 new_lock->l_remote_handle = lock->l_remote_handle;
3068 new_lock->l_flags &= ~LDLM_FL_LOCAL;
3070 unlock_res_and_lock(new_lock);
3072 cfs_hash_add(new_lock->l_export->exp_lock_hash,
3073 &new_lock->l_remote_handle,
3074 &new_lock->l_exp_hash);
3076 LDLM_LOCK_RELEASE(new_lock);
3077 lh->mlh_reg_lh.cookie = 0;
3079 RETURN(ELDLM_LOCK_REPLACED);
3082 static void mdt_intent_fixup_resent(struct mdt_thread_info *info,
3083 struct ldlm_lock *new_lock,
3084 struct mdt_lock_handle *lh,
3087 struct ptlrpc_request *req = mdt_info_req(info);
3088 struct ldlm_request *dlmreq;
3090 if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
3093 dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
3095 /* Check if this is a resend case (MSG_RESENT is set on RPC) and a
3096 * lock was found by ldlm_handle_enqueue(); if so @lh must be
3098 if (flags & LDLM_FL_RESENT) {
3099 lh->mlh_reg_lh.cookie = new_lock->l_handle.h_cookie;
3100 lh->mlh_reg_mode = new_lock->l_granted_mode;
3102 LDLM_DEBUG(new_lock, "Restoring lock cookie");
3103 DEBUG_REQ(D_DLMTRACE, req, "restoring lock cookie "LPX64,
3104 lh->mlh_reg_lh.cookie);
3109 * If the xid matches, then we know this is a resent request, and allow
3110 * it. (It's probably an OPEN, for which we don't send a lock.
3112 if (req_xid_is_last(req))
3116 * This remote handle isn't enqueued, so we never received or processed
3117 * this request. Clear MSG_RESENT, because it can be handled like any
3118 * normal request now.
3120 lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
3122 DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle "LPX64,
3123 dlmreq->lock_handle[0].cookie);
3126 static int mdt_intent_getxattr(enum mdt_it_code opcode,
3127 struct mdt_thread_info *info,
3128 struct ldlm_lock **lockp,
3131 struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3132 struct ldlm_reply *ldlm_rep = NULL;
3136 * Initialize lhc->mlh_reg_lh either from a previously granted lock
3137 * (for the resend case) or a new lock. Below we will use it to
3138 * replace the original lock.
3140 mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3141 if (!lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3142 mdt_lock_reg_init(lhc, (*lockp)->l_req_mode);
3143 rc = mdt_object_lock(info, info->mti_object, lhc,
3144 MDS_INODELOCK_XATTR,
3150 grc = mdt_getxattr(info);
3152 rc = mdt_intent_lock_replace(info, lockp, lhc, flags);
3154 if (mdt_info_req(info)->rq_repmsg != NULL)
3155 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3156 if (ldlm_rep == NULL)
3157 RETURN(err_serious(-EFAULT));
3159 ldlm_rep->lock_policy_res2 = grc;
3164 static int mdt_intent_getattr(enum mdt_it_code opcode,
3165 struct mdt_thread_info *info,
3166 struct ldlm_lock **lockp,
3169 struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3171 struct ldlm_reply *ldlm_rep;
3172 struct mdt_body *reqbody;
3173 struct mdt_body *repbody;
3177 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
3180 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
3183 info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
3184 repbody->mbo_eadatasize = 0;
3185 repbody->mbo_aclsize = 0;
3189 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM;
3191 case MDT_IT_GETATTR:
3192 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
3196 CERROR("Unsupported intent (%d)\n", opcode);
3197 GOTO(out_shrink, rc = -EINVAL);
3200 rc = mdt_init_ucred(info, reqbody);
3202 GOTO(out_shrink, rc);
3204 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3205 mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
3207 /* Get lock from request for possible resent case. */
3208 mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3210 rc = mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
3211 ldlm_rep->lock_policy_res2 = clear_serious(rc);
3213 if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
3214 ldlm_rep->lock_policy_res2 = 0;
3215 if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
3216 ldlm_rep->lock_policy_res2) {
3217 lhc->mlh_reg_lh.cookie = 0ull;
3218 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
3221 rc = mdt_intent_lock_replace(info, lockp, lhc, flags);
3224 mdt_exit_ucred(info);
3226 mdt_client_compatibility(info);
3227 rc2 = mdt_fix_reply(info);
3233 static int mdt_intent_layout(enum mdt_it_code opcode,
3234 struct mdt_thread_info *info,
3235 struct ldlm_lock **lockp,
3238 struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_LAYOUT];
3239 struct layout_intent *layout;
3241 struct mdt_object *obj = NULL;
3245 if (opcode != MDT_IT_LAYOUT) {
3246 CERROR("%s: Unknown intent (%d)\n", mdt_obd_name(info->mti_mdt),
3251 fid = &info->mti_tmp_fid2;
3252 fid_extract_from_res_name(fid, &(*lockp)->l_resource->lr_name);
3254 /* Get lock from request for possible resent case. */
3255 mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3257 obj = mdt_object_find(info->mti_env, info->mti_mdt, fid);
3259 RETURN(PTR_ERR(obj));
3261 if (mdt_object_exists(obj) && !mdt_object_remote(obj)) {
3262 /* get the length of lsm */
3263 rc = mdt_attr_get_eabuf_size(info, obj);
3267 if (rc > info->mti_mdt->mdt_max_mdsize)
3268 info->mti_mdt->mdt_max_mdsize = rc;
3271 mdt_object_put(info->mti_env, obj);
3273 (*lockp)->l_lvb_type = LVB_T_LAYOUT;
3274 req_capsule_set_size(info->mti_pill, &RMF_DLM_LVB, RCL_SERVER, rc);
3275 rc = req_capsule_server_pack(info->mti_pill);
3279 if (lustre_handle_is_used(&lhc->mlh_reg_lh))
3280 rc = mdt_intent_lock_replace(info, lockp, lhc, flags);
3282 layout = req_capsule_client_get(info->mti_pill, &RMF_LAYOUT_INTENT);
3283 LASSERT(layout != NULL);
3284 if (layout->li_opc == LAYOUT_INTENT_ACCESS)
3285 /* return to normal/resent ldlm handling */
3288 CERROR("%s: Unsupported layout intent (%d)\n",
3289 mdt_obd_name(info->mti_mdt), layout->li_opc);
3293 static int mdt_intent_reint(enum mdt_it_code opcode,
3294 struct mdt_thread_info *info,
3295 struct ldlm_lock **lockp,
3298 struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3299 struct ldlm_reply *rep = NULL;
3303 static const struct req_format *intent_fmts[REINT_MAX] = {
3304 [REINT_CREATE] = &RQF_LDLM_INTENT_CREATE,
3305 [REINT_OPEN] = &RQF_LDLM_INTENT_OPEN
3310 opc = mdt_reint_opcode(mdt_info_req(info), intent_fmts);
3314 if (mdt_it_flavor[opcode].it_reint != opc) {
3315 CERROR("Reint code %ld doesn't match intent: %d\n",
3317 RETURN(err_serious(-EPROTO));
3320 /* Get lock from request for possible resent case. */
3321 mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3323 rc = mdt_reint_internal(info, lhc, opc);
3325 /* Check whether the reply has been packed successfully. */
3326 if (mdt_info_req(info)->rq_repmsg != NULL)
3327 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3329 RETURN(err_serious(-EFAULT));
3331 /* MDC expects this in any case */
3333 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
3335 /* the open lock or the lock for cross-ref object should be
3336 * returned to the client */
3337 if (rc == -EREMOTE || mdt_get_disposition(rep, DISP_OPEN_LOCK)) {
3338 LASSERT(lustre_handle_is_used(&lhc->mlh_reg_lh));
3339 rep->lock_policy_res2 = 0;
3340 rc = mdt_intent_lock_replace(info, lockp, lhc, flags);
3344 rep->lock_policy_res2 = clear_serious(rc);
3346 if (rep->lock_policy_res2 == -ENOENT &&
3347 mdt_get_disposition(rep, DISP_LOOKUP_NEG))
3348 rep->lock_policy_res2 = 0;
3350 if (rc == -ENOTCONN || rc == -ENODEV ||
3351 rc == -EOVERFLOW) { /**< if VBR failure then return error */
3353 * If it is the disconnect error (ENODEV & ENOCONN), the error
3354 * will be returned by rq_status, and client at ptlrpc layer
3355 * will detect this, then disconnect, reconnect the import
3356 * immediately, instead of impacting the following the rpc.
3358 lhc->mlh_reg_lh.cookie = 0ull;
3362 * For other cases, the error will be returned by intent.
3363 * and client will retrieve the result from intent.
3366 * FIXME: when open lock is finished, that should be
3369 if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3370 LASSERTF(rc == 0, "Error occurred but lock handle "
3371 "is still in use, rc = %d\n", rc);
3372 rep->lock_policy_res2 = 0;
3373 rc = mdt_intent_lock_replace(info, lockp, lhc, flags);
3376 lhc->mlh_reg_lh.cookie = 0ull;
3377 RETURN(ELDLM_LOCK_ABORTED);
3382 static int mdt_intent_code(long itcode)
3390 case IT_OPEN|IT_CREAT:
3397 rc = MDT_IT_READDIR;
3400 rc = MDT_IT_GETATTR;
3412 rc = MDT_IT_GETXATTR;
3417 case IT_QUOTA_DQACQ:
3422 CERROR("Unknown intent opcode: %ld\n", itcode);
3429 static int mdt_intent_opc(long itopc, struct mdt_thread_info *info,
3430 struct ldlm_lock **lockp, __u64 flags)
3432 struct req_capsule *pill;
3433 struct mdt_it_flavor *flv;
3438 opc = mdt_intent_code(itopc);
3442 pill = info->mti_pill;
3444 if (opc == MDT_IT_QUOTA) {
3445 struct lu_device *qmt = info->mti_mdt->mdt_qmt_dev;
3448 RETURN(-EOPNOTSUPP);
3450 (*lockp)->l_lvb_type = LVB_T_LQUOTA;
3451 /* pass the request to quota master */
3452 rc = qmt_hdls.qmth_intent_policy(info->mti_env, qmt,
3453 mdt_info_req(info), lockp,
3458 flv = &mdt_it_flavor[opc];
3459 if (flv->it_fmt != NULL)
3460 req_capsule_extend(pill, flv->it_fmt);
3462 rc = mdt_unpack_req_pack_rep(info, flv->it_flags);
3464 struct ptlrpc_request *req = mdt_info_req(info);
3465 if (flv->it_flags & MUTABOR &&
3466 exp_connect_flags(req->rq_export) & OBD_CONNECT_RDONLY)
3469 if (rc == 0 && flv->it_act != NULL) {
3470 struct ldlm_reply *rep;
3472 /* execute policy */
3473 rc = flv->it_act(opc, info, lockp, flags);
3475 /* Check whether the reply has been packed successfully. */
3476 if (mdt_info_req(info)->rq_repmsg != NULL) {
3477 rep = req_capsule_server_get(info->mti_pill,
3479 rep->lock_policy_res2 =
3480 ptlrpc_status_hton(rep->lock_policy_res2);
3488 static int mdt_intent_policy(struct ldlm_namespace *ns,
3489 struct ldlm_lock **lockp, void *req_cookie,
3490 ldlm_mode_t mode, __u64 flags, void *data)
3492 struct tgt_session_info *tsi;
3493 struct mdt_thread_info *info;
3494 struct ptlrpc_request *req = req_cookie;
3495 struct ldlm_intent *it;
3496 struct req_capsule *pill;
3501 LASSERT(req != NULL);
3503 tsi = tgt_ses_info(req->rq_svc_thread->t_env);
3505 info = tsi2mdt_info(tsi);
3506 LASSERT(info != NULL);
3507 pill = info->mti_pill;
3508 LASSERT(pill->rc_req == req);
3510 if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
3511 req_capsule_extend(pill, &RQF_LDLM_INTENT_BASIC);
3512 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
3514 rc = mdt_intent_opc(it->opc, info, lockp, flags);
3518 /* Lock without inodebits makes no sense and will oops
3519 * later in ldlm. Let's check it now to see if we have
3520 * ibits corrupted somewhere in mdt_intent_opc().
3521 * The case for client miss to set ibits has been
3522 * processed by others. */
3523 LASSERT(ergo(info->mti_dlm_req->lock_desc.l_resource.\
3524 lr_type == LDLM_IBITS,
3525 info->mti_dlm_req->lock_desc.\
3526 l_policy_data.l_inodebits.bits != 0));
3528 rc = err_serious(-EFAULT);
3530 /* No intent was provided */
3531 LASSERT(pill->rc_fmt == &RQF_LDLM_ENQUEUE);
3532 req_capsule_set_size(pill, &RMF_DLM_LVB, RCL_SERVER, 0);
3533 rc = req_capsule_server_pack(pill);
3535 rc = err_serious(rc);
3537 mdt_thread_info_fini(info);
3541 static void mdt_deregister_seq_exp(struct mdt_device *mdt)
3543 struct seq_server_site *ss = mdt_seq_site(mdt);
3545 if (ss->ss_node_id == 0)
3548 if (ss->ss_client_seq != NULL) {
3549 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
3550 ss->ss_client_seq->lcs_exp = NULL;
3553 if (ss->ss_server_fld != NULL) {
3554 lustre_deregister_lwp_item(&ss->ss_server_fld->lsf_control_exp);
3555 ss->ss_server_fld->lsf_control_exp = NULL;
3559 static void mdt_seq_fini_cli(struct mdt_device *mdt)
3561 struct seq_server_site *ss = mdt_seq_site(mdt);
3566 if (ss->ss_server_seq != NULL)
3567 seq_server_set_cli(NULL, ss->ss_server_seq, NULL);
3570 static int mdt_seq_fini(const struct lu_env *env, struct mdt_device *mdt)
3572 mdt_seq_fini_cli(mdt);
3573 mdt_deregister_seq_exp(mdt);
3575 return seq_site_fini(env, mdt_seq_site(mdt));
3579 * It will retrieve its FLDB entries from MDT0, and it only happens
3580 * when upgrading existent FS to 2.6 or when local FLDB is corrupted,
3581 * and it needs to refresh FLDB from the MDT0.
3583 static int mdt_register_lwp_callback(void *data)
3586 struct mdt_device *mdt = data;
3587 struct lu_server_fld *fld = mdt_seq_site(mdt)->ss_server_fld;
3591 LASSERT(mdt_seq_site(mdt)->ss_node_id != 0);
3593 if (!likely(fld->lsf_new))
3596 rc = lu_env_init(&env, LCT_MD_THREAD);
3598 CERROR("%s: cannot init env: rc = %d\n", mdt_obd_name(mdt), rc);
3602 rc = fld_update_from_controller(&env, fld);
3604 CERROR("%s: cannot update controller: rc = %d\n",
3605 mdt_obd_name(mdt), rc);
3613 static int mdt_register_seq_exp(struct mdt_device *mdt)
3615 struct seq_server_site *ss = mdt_seq_site(mdt);
3616 char *lwp_name = NULL;
3619 if (ss->ss_node_id == 0)
3622 OBD_ALLOC(lwp_name, MAX_OBD_NAME);
3623 if (lwp_name == NULL)
3624 GOTO(out_free, rc = -ENOMEM);
3626 rc = tgt_name2lwp_name(mdt_obd_name(mdt), lwp_name, MAX_OBD_NAME, 0);
3630 rc = lustre_register_lwp_item(lwp_name, &ss->ss_client_seq->lcs_exp,
3635 rc = lustre_register_lwp_item(lwp_name,
3636 &ss->ss_server_fld->lsf_control_exp,
3637 mdt_register_lwp_callback, mdt);
3639 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
3640 ss->ss_client_seq->lcs_exp = NULL;
3644 if (lwp_name != NULL)
3645 OBD_FREE(lwp_name, MAX_OBD_NAME);
3651 * Init client sequence manager which is used by local MDS to talk to sequence
3652 * controller on remote node.
3654 static int mdt_seq_init_cli(const struct lu_env *env, struct mdt_device *mdt)
3656 struct seq_server_site *ss = mdt_seq_site(mdt);
3661 /* check if this is adding the first MDC and controller is not yet
3663 OBD_ALLOC_PTR(ss->ss_client_seq);
3664 if (ss->ss_client_seq == NULL)
3667 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3668 if (prefix == NULL) {
3669 OBD_FREE_PTR(ss->ss_client_seq);
3670 ss->ss_client_seq = NULL;
3674 /* Note: seq_client_fini will be called in seq_site_fini */
3675 snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s", mdt_obd_name(mdt));
3676 rc = seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_METADATA,
3677 prefix, ss->ss_node_id == 0 ? ss->ss_control_seq :
3679 OBD_FREE(prefix, MAX_OBD_NAME + 5);
3681 OBD_FREE_PTR(ss->ss_client_seq);
3682 ss->ss_client_seq = NULL;
3686 rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq);
3691 static int mdt_seq_init(const struct lu_env *env, struct mdt_device *mdt)
3693 struct seq_server_site *ss;
3697 ss = mdt_seq_site(mdt);
3698 /* init sequence controller server(MDT0) */
3699 if (ss->ss_node_id == 0) {
3700 OBD_ALLOC_PTR(ss->ss_control_seq);
3701 if (ss->ss_control_seq == NULL)
3704 rc = seq_server_init(env, ss->ss_control_seq, mdt->mdt_bottom,
3705 mdt_obd_name(mdt), LUSTRE_SEQ_CONTROLLER,
3708 GOTO(out_seq_fini, rc);
3711 /* Init normal sequence server */
3712 OBD_ALLOC_PTR(ss->ss_server_seq);
3713 if (ss->ss_server_seq == NULL)
3714 GOTO(out_seq_fini, rc = -ENOMEM);
3716 rc = seq_server_init(env, ss->ss_server_seq, mdt->mdt_bottom,
3717 mdt_obd_name(mdt), LUSTRE_SEQ_SERVER, ss);
3719 GOTO(out_seq_fini, rc);
3721 /* init seq client for seq server to talk to seq controller(MDT0) */
3722 rc = mdt_seq_init_cli(env, mdt);
3724 GOTO(out_seq_fini, rc);
3726 if (ss->ss_node_id != 0)
3727 /* register controler export through lwp */
3728 rc = mdt_register_seq_exp(mdt);
3733 mdt_seq_fini(env, mdt);
3741 static int mdt_fld_fini(const struct lu_env *env,
3742 struct mdt_device *m)
3744 struct seq_server_site *ss = mdt_seq_site(m);
3747 if (ss && ss->ss_server_fld) {
3748 fld_server_fini(env, ss->ss_server_fld);
3749 OBD_FREE_PTR(ss->ss_server_fld);
3750 ss->ss_server_fld = NULL;
3756 static int mdt_fld_init(const struct lu_env *env,
3758 struct mdt_device *m)
3760 struct seq_server_site *ss;
3764 ss = mdt_seq_site(m);
3766 OBD_ALLOC_PTR(ss->ss_server_fld);
3767 if (ss->ss_server_fld == NULL)
3768 RETURN(rc = -ENOMEM);
3770 rc = fld_server_init(env, ss->ss_server_fld, m->mdt_bottom, uuid,
3773 OBD_FREE_PTR(ss->ss_server_fld);
3774 ss->ss_server_fld = NULL;
3781 static void mdt_stack_pre_fini(const struct lu_env *env,
3782 struct mdt_device *m, struct lu_device *top)
3784 struct lustre_cfg_bufs *bufs;
3785 struct lustre_cfg *lcfg;
3786 struct mdt_thread_info *info;
3791 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3792 LASSERT(info != NULL);
3794 bufs = &info->mti_u.bufs;
3796 LASSERT(m->mdt_child_exp);
3797 LASSERT(m->mdt_child_exp->exp_obd);
3799 /* process cleanup, pass mdt obd name to get obd umount flags */
3800 /* XXX: this is needed because all layers are referenced by
3801 * objects (some of them are pinned by osd, for example *
3802 * the proper solution should be a model where object used
3803 * by osd only doesn't have mdt/mdd slices -bzzz */
3804 lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
3805 lustre_cfg_bufs_set_string(bufs, 1, NULL);
3806 lcfg = lustre_cfg_new(LCFG_PRE_CLEANUP, bufs);
3808 CERROR("%s:Cannot alloc lcfg!\n", mdt_obd_name(m));
3811 top->ld_ops->ldo_process_config(env, top, lcfg);
3812 lustre_cfg_free(lcfg);
3816 static void mdt_stack_fini(const struct lu_env *env,
3817 struct mdt_device *m, struct lu_device *top)
3819 struct obd_device *obd = mdt2obd_dev(m);
3820 struct lustre_cfg_bufs *bufs;
3821 struct lustre_cfg *lcfg;
3822 struct mdt_thread_info *info;
3826 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3827 LASSERT(info != NULL);
3829 lu_dev_del_linkage(top->ld_site, top);
3831 lu_site_purge(env, top->ld_site, -1);
3833 bufs = &info->mti_u.bufs;
3834 /* process cleanup, pass mdt obd name to get obd umount flags */
3835 /* another purpose is to let all layers to release their objects */
3836 lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
3841 lustre_cfg_bufs_set_string(bufs, 1, flags);
3842 lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
3844 CERROR("Cannot alloc lcfg!\n");
3848 top->ld_ops->ldo_process_config(env, top, lcfg);
3849 lustre_cfg_free(lcfg);
3851 lu_site_purge(env, top->ld_site, -1);
3853 m->mdt_child = NULL;
3854 m->mdt_bottom = NULL;
3856 obd_disconnect(m->mdt_child_exp);
3857 m->mdt_child_exp = NULL;
3859 obd_disconnect(m->mdt_bottom_exp);
3860 m->mdt_child_exp = NULL;
3863 static int mdt_connect_to_next(const struct lu_env *env, struct mdt_device *m,
3864 const char *next, struct obd_export **exp)
3866 struct obd_connect_data *data = NULL;
3867 struct obd_device *obd;
3871 OBD_ALLOC_PTR(data);
3873 GOTO(out, rc = -ENOMEM);
3875 obd = class_name2obd(next);
3877 CERROR("%s: can't locate next device: %s\n",
3878 mdt_obd_name(m), next);
3879 GOTO(out, rc = -ENOTCONN);
3882 data->ocd_connect_flags = OBD_CONNECT_VERSION;
3883 data->ocd_version = LUSTRE_VERSION_CODE;
3885 rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
3887 CERROR("%s: cannot connect to next dev %s (%d)\n",
3888 mdt_obd_name(m), next, rc);
3898 static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt,
3899 struct lustre_cfg *cfg)
3901 char *dev = lustre_cfg_string(cfg, 0);
3902 int rc, name_size, uuid_size;
3903 char *name, *uuid, *p;
3904 struct lustre_cfg_bufs *bufs;
3905 struct lustre_cfg *lcfg;
3906 struct obd_device *obd;
3907 struct lustre_profile *lprof;
3908 struct lu_site *site;
3911 /* in 1.8 we had the only device in the stack - MDS.
3912 * 2.0 introduces MDT, MDD, OSD; MDT starts others internally.
3913 * in 2.3 OSD is instantiated by obd_mount.c, so we need
3914 * to generate names and setup MDT, MDD. MDT will be using
3915 * generated name to connect to MDD. for MDD the next device
3916 * will be LOD with name taken from so called "profile" which
3917 * is generated by mount_option line
3919 * 1.8 MGS generates config. commands like this:
3920 * #06 (104)mount_option 0: 1:lustre-MDT0000 2:lustre-mdtlov
3921 * #08 (120)setup 0:lustre-MDT0000 1:dev 2:type 3:lustre-MDT0000
3922 * 2.0 MGS generates config. commands like this:
3923 * #07 (112)mount_option 0: 1:lustre-MDT0000 2:lustre-MDT0000-mdtlov
3924 * #08 (160)setup 0:lustre-MDT0000 1:lustre-MDT0000_UUID 2:0
3925 * 3:lustre-MDT0000-mdtlov 4:f
3927 * we generate MDD name from MDT one, just replacing T with D
3929 * after all the preparations, the logical equivalent will be
3930 * #01 (160)setup 0:lustre-MDD0000 1:lustre-MDD0000_UUID 2:0
3931 * 3:lustre-MDT0000-mdtlov 4:f
3932 * #02 (160)setup 0:lustre-MDT0000 1:lustre-MDT0000_UUID 2:0
3933 * 3:lustre-MDD0000 4:f
3935 * notice we build the stack from down to top: MDD first, then MDT */
3937 name_size = MAX_OBD_NAME;
3938 uuid_size = MAX_OBD_NAME;
3940 OBD_ALLOC(name, name_size);
3941 OBD_ALLOC(uuid, uuid_size);
3942 if (name == NULL || uuid == NULL)
3943 GOTO(cleanup_mem, rc = -ENOMEM);
3945 OBD_ALLOC_PTR(bufs);
3947 GOTO(cleanup_mem, rc = -ENOMEM);
3950 p = strstr(name, "-MDT");
3952 GOTO(free_bufs, rc = -ENOMEM);
3955 snprintf(uuid, MAX_OBD_NAME, "%s_UUID", name);
3957 lprof = class_get_profile(lustre_cfg_string(cfg, 0));
3958 if (lprof == NULL || lprof->lp_dt == NULL) {
3959 CERROR("can't find the profile: %s\n",
3960 lustre_cfg_string(cfg, 0));
3961 GOTO(free_bufs, rc = -EINVAL);
3964 lustre_cfg_bufs_reset(bufs, name);
3965 lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_MDD_NAME);
3966 lustre_cfg_bufs_set_string(bufs, 2, uuid);
3967 lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
3969 lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
3971 GOTO(free_bufs, rc = -ENOMEM);
3973 rc = class_attach(lcfg);
3975 GOTO(lcfg_cleanup, rc);
3977 obd = class_name2obd(name);
3979 CERROR("Can not find obd %s (%s in config)\n",
3980 MDD_OBD_NAME, lustre_cfg_string(cfg, 0));
3981 GOTO(class_detach, rc = -EINVAL);
3984 lustre_cfg_free(lcfg);
3986 lustre_cfg_bufs_reset(bufs, name);
3987 lustre_cfg_bufs_set_string(bufs, 1, uuid);
3988 lustre_cfg_bufs_set_string(bufs, 2, dev);
3989 lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
3991 lcfg = lustre_cfg_new(LCFG_SETUP, bufs);
3993 rc = class_setup(obd, lcfg);
3995 GOTO(class_detach, rc);
3997 /* connect to MDD we just setup */
3998 rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_child_exp);
4000 GOTO(class_detach, rc);
4002 site = mdt->mdt_child_exp->exp_obd->obd_lu_dev->ld_site;
4004 LASSERT(mdt_lu_site(mdt) == NULL);
4005 mdt->mdt_lu_dev.ld_site = site;
4006 site->ls_top_dev = &mdt->mdt_lu_dev;
4007 mdt->mdt_child = lu2md_dev(mdt->mdt_child_exp->exp_obd->obd_lu_dev);
4009 /* now connect to bottom OSD */
4010 snprintf(name, MAX_OBD_NAME, "%s-osd", dev);
4011 rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_bottom_exp);
4013 GOTO(class_detach, rc);
4015 lu2dt_dev(mdt->mdt_bottom_exp->exp_obd->obd_lu_dev);
4017 rc = lu_env_refill((struct lu_env *)env);
4019 CERROR("Failure to refill session: '%d'\n", rc);
4021 lu_dev_add_linkage(site, &mdt->mdt_lu_dev);
4026 class_detach(obd, lcfg);
4028 lustre_cfg_free(lcfg);
4033 OBD_FREE(name, name_size);
4035 OBD_FREE(uuid, uuid_size);
4039 /* setup quota master target on MDT0 */
4040 static int mdt_quota_init(const struct lu_env *env, struct mdt_device *mdt,
4041 struct lustre_cfg *cfg)
4043 struct obd_device *obd;
4044 char *dev = lustre_cfg_string(cfg, 0);
4045 char *qmtname, *uuid, *p;
4046 struct lustre_cfg_bufs *bufs;
4047 struct lustre_cfg *lcfg;
4048 struct lustre_profile *lprof;
4049 struct obd_connect_data *data;
4053 LASSERT(mdt->mdt_qmt_exp == NULL);
4054 LASSERT(mdt->mdt_qmt_dev == NULL);
4056 /* quota master is on MDT0 only for now */
4057 if (mdt->mdt_seq_site.ss_node_id != 0)
4060 /* MGS generates config commands which look as follows:
4061 * #01 (160)setup 0:lustre-MDT0000 1:lustre-MDT0000_UUID 2:0
4062 * 3:lustre-MDT0000-mdtlov 4:f
4064 * We generate the QMT name from the MDT one, just replacing MD with QM
4065 * after all the preparations, the logical equivalent will be:
4066 * #01 (160)setup 0:lustre-QMT0000 1:lustre-QMT0000_UUID 2:0
4067 * 3:lustre-MDT0000-osd 4:f */
4068 OBD_ALLOC(qmtname, MAX_OBD_NAME);
4069 OBD_ALLOC(uuid, UUID_MAX);
4070 OBD_ALLOC_PTR(bufs);
4071 OBD_ALLOC_PTR(data);
4072 if (qmtname == NULL || uuid == NULL || bufs == NULL || data == NULL)
4073 GOTO(cleanup_mem, rc = -ENOMEM);
4075 strcpy(qmtname, dev);
4076 p = strstr(qmtname, "-MDT");
4078 GOTO(cleanup_mem, rc = -ENOMEM);
4079 /* replace MD with QM */
4083 snprintf(uuid, UUID_MAX, "%s_UUID", qmtname);
4085 lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4086 if (lprof == NULL || lprof->lp_dt == NULL) {
4087 CERROR("can't find profile for %s\n",
4088 lustre_cfg_string(cfg, 0));
4089 GOTO(cleanup_mem, rc = -EINVAL);
4092 lustre_cfg_bufs_reset(bufs, qmtname);
4093 lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_QMT_NAME);
4094 lustre_cfg_bufs_set_string(bufs, 2, uuid);
4095 lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4097 lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
4099 GOTO(cleanup_mem, rc = -ENOMEM);
4101 rc = class_attach(lcfg);
4103 GOTO(lcfg_cleanup, rc);
4105 obd = class_name2obd(qmtname);
4107 CERROR("Can not find obd %s (%s in config)\n", qmtname,
4108 lustre_cfg_string(cfg, 0));
4109 GOTO(class_detach, rc = -EINVAL);
4112 lustre_cfg_free(lcfg);
4114 lustre_cfg_bufs_reset(bufs, qmtname);
4115 lustre_cfg_bufs_set_string(bufs, 1, uuid);
4116 lustre_cfg_bufs_set_string(bufs, 2, dev);
4118 /* for quota, the next device should be the OSD device */
4119 lustre_cfg_bufs_set_string(bufs, 3,
4120 mdt->mdt_bottom->dd_lu_dev.ld_obd->obd_name);
4122 lcfg = lustre_cfg_new(LCFG_SETUP, bufs);
4124 rc = class_setup(obd, lcfg);
4126 GOTO(class_detach, rc);
4128 mdt->mdt_qmt_dev = obd->obd_lu_dev;
4130 /* configure local quota objects */
4131 rc = mdt->mdt_qmt_dev->ld_ops->ldo_prepare(env,
4135 GOTO(class_cleanup, rc);
4137 /* connect to quota master target */
4138 data->ocd_connect_flags = OBD_CONNECT_VERSION;
4139 data->ocd_version = LUSTRE_VERSION_CODE;
4140 rc = obd_connect(NULL, &mdt->mdt_qmt_exp, obd, &obd->obd_uuid,
4143 CERROR("cannot connect to quota master device %s (%d)\n",
4145 GOTO(class_cleanup, rc);
4151 class_manual_cleanup(obd);
4152 mdt->mdt_qmt_dev = NULL;
4156 class_detach(obd, lcfg);
4158 lustre_cfg_free(lcfg);
4163 OBD_FREE(qmtname, MAX_OBD_NAME);
4165 OBD_FREE(uuid, UUID_MAX);
4171 /* Shutdown quota master target associated with mdt */
4172 static void mdt_quota_fini(const struct lu_env *env, struct mdt_device *mdt)
4176 if (mdt->mdt_qmt_exp == NULL)
4178 LASSERT(mdt->mdt_qmt_dev != NULL);
4180 /* the qmt automatically shuts down when the mdt disconnects */
4181 obd_disconnect(mdt->mdt_qmt_exp);
4182 mdt->mdt_qmt_exp = NULL;
4183 mdt->mdt_qmt_dev = NULL;
4187 /* mdt_getxattr() is used from mdt_intent_getxattr(), use this wrapper
4188 * for now. This will be removed along with converting rest of MDT code
4189 * to use tgt_session_info */
4190 int mdt_tgt_getxattr(struct tgt_session_info *tsi)
4192 struct mdt_thread_info *info = tsi2mdt_info(tsi);
4195 rc = mdt_getxattr(info);
4197 mdt_thread_info_fini(info);
4201 static struct tgt_handler mdt_tgt_handlers[] = {
4202 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4203 0, MDS_CONNECT, mdt_tgt_connect,
4204 &RQF_CONNECT, LUSTRE_OBD_VERSION),
4205 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4206 0, MDS_DISCONNECT, tgt_disconnect,
4207 &RQF_MDS_DISCONNECT, LUSTRE_OBD_VERSION),
4208 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4209 HABEO_REFERO, MDS_SET_INFO, mdt_set_info,
4210 &RQF_OBD_SET_INFO, LUSTRE_MDS_VERSION),
4211 TGT_MDT_HDL(0, MDS_GET_INFO, mdt_get_info),
4212 TGT_MDT_HDL(0 | HABEO_REFERO, MDS_GETSTATUS, mdt_getstatus),
4213 TGT_MDT_HDL(HABEO_CORPUS, MDS_GETATTR, mdt_getattr),
4214 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_GETATTR_NAME,
4216 TGT_MDT_HDL(HABEO_CORPUS, MDS_GETXATTR, mdt_tgt_getxattr),
4217 TGT_MDT_HDL(0 | HABEO_REFERO, MDS_STATFS, mdt_statfs),
4218 TGT_MDT_HDL(0 | MUTABOR, MDS_REINT, mdt_reint),
4219 TGT_MDT_HDL(HABEO_CORPUS, MDS_CLOSE, mdt_close),
4220 TGT_MDT_HDL(HABEO_CORPUS, MDS_DONE_WRITING,
4222 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_READPAGE, mdt_readpage),
4223 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_SYNC, mdt_sync),
4224 TGT_MDT_HDL(0, MDS_QUOTACTL, mdt_quotactl),
4225 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_PROGRESS,
4227 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_REGISTER,
4228 mdt_hsm_ct_register),
4229 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_UNREGISTER,
4230 mdt_hsm_ct_unregister),
4231 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_STATE_GET,
4233 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_STATE_SET,
4235 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_ACTION, mdt_hsm_action),
4236 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_REQUEST,
4238 TGT_MDT_HDL(HABEO_CLAVIS | HABEO_CORPUS | HABEO_REFERO | MUTABOR,
4243 static struct tgt_handler mdt_sec_ctx_ops[] = {
4244 TGT_SEC_HDL_VAR(0, SEC_CTX_INIT, mdt_sec_ctx_handle),
4245 TGT_SEC_HDL_VAR(0, SEC_CTX_INIT_CONT,mdt_sec_ctx_handle),
4246 TGT_SEC_HDL_VAR(0, SEC_CTX_FINI, mdt_sec_ctx_handle)
4249 static struct tgt_handler mdt_quota_ops[] = {
4250 TGT_QUOTA_HDL(HABEO_REFERO, QUOTA_DQACQ, mdt_quota_dqacq),
4253 static struct tgt_opc_slice mdt_common_slice[] = {
4255 .tos_opc_start = MDS_FIRST_OPC,
4256 .tos_opc_end = MDS_LAST_OPC,
4257 .tos_hs = mdt_tgt_handlers
4260 .tos_opc_start = OBD_FIRST_OPC,
4261 .tos_opc_end = OBD_LAST_OPC,
4262 .tos_hs = tgt_obd_handlers
4265 .tos_opc_start = LDLM_FIRST_OPC,
4266 .tos_opc_end = LDLM_LAST_OPC,
4267 .tos_hs = tgt_dlm_handlers
4270 .tos_opc_start = SEC_FIRST_OPC,
4271 .tos_opc_end = SEC_LAST_OPC,
4272 .tos_hs = mdt_sec_ctx_ops
4275 .tos_opc_start = OUT_UPDATE_FIRST_OPC,
4276 .tos_opc_end = OUT_UPDATE_LAST_OPC,
4277 .tos_hs = tgt_out_handlers
4280 .tos_opc_start = FLD_FIRST_OPC,
4281 .tos_opc_end = FLD_LAST_OPC,
4282 .tos_hs = fld_handlers
4285 .tos_opc_start = SEQ_FIRST_OPC,
4286 .tos_opc_end = SEQ_LAST_OPC,
4287 .tos_hs = seq_handlers
4290 .tos_opc_start = QUOTA_DQACQ,
4291 .tos_opc_end = QUOTA_LAST_OPC,
4292 .tos_hs = mdt_quota_ops
4295 .tos_opc_start = LLOG_FIRST_OPC,
4296 .tos_opc_end = LLOG_LAST_OPC,
4297 .tos_hs = tgt_llog_handlers
4300 .tos_opc_start = LFSCK_FIRST_OPC,
4301 .tos_opc_end = LFSCK_LAST_OPC,
4302 .tos_hs = tgt_lfsck_handlers
4310 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
4312 struct md_device *next = m->mdt_child;
4313 struct lu_device *d = &m->mdt_lu_dev;
4314 struct obd_device *obd = mdt2obd_dev(m);
4315 struct lfsck_stop stop;
4318 stop.ls_status = LS_PAUSED;
4320 next->md_ops->mdo_iocontrol(env, next, OBD_IOC_STOP_LFSCK, 0, &stop);
4322 target_recovery_fini(obd);
4323 ping_evictor_stop();
4324 mdt_stack_pre_fini(env, m, md2lu_dev(m->mdt_child));
4326 if (m->mdt_opts.mo_coordinator)
4327 mdt_hsm_cdt_stop(m);
4329 mdt_hsm_cdt_fini(m);
4331 mdt_llog_ctxt_unclone(env, m, LLOG_AGENT_ORIG_CTXT);
4332 mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4333 obd_exports_barrier(obd);
4334 obd_zombie_barrier();
4338 tgt_fini(env, &m->mdt_lut);
4339 mdt_fs_cleanup(env, m);
4340 upcall_cache_cleanup(m->mdt_identity_cache);
4341 m->mdt_identity_cache = NULL;
4343 if (m->mdt_namespace != NULL) {
4344 ldlm_namespace_free(m->mdt_namespace, NULL,
4345 d->ld_obd->obd_force);
4346 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
4349 mdt_quota_fini(env, m);
4351 cfs_free_nidlist(&m->mdt_squash.rsi_nosquash_nids);
4353 mdt_seq_fini(env, m);
4354 mdt_fld_fini(env, m);
4356 next->md_ops->mdo_init_capa_ctxt(env, next, 0, 0, 0, NULL);
4357 cfs_timer_disarm(&m->mdt_ck_timer);
4358 mdt_ck_thread_stop(m);
4363 mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4365 LASSERT(atomic_read(&d->ld_ref) == 0);
4367 server_put_mount(mdt_obd_name(m), true);
4372 static int mdt_postrecov(const struct lu_env *, struct mdt_device *);
4374 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
4375 struct lu_device_type *ldt, struct lustre_cfg *cfg)
4377 struct mdt_thread_info *info;
4378 struct obd_device *obd;
4379 const char *dev = lustre_cfg_string(cfg, 0);
4380 const char *num = lustre_cfg_string(cfg, 2);
4381 struct lustre_mount_info *lmi = NULL;
4382 struct lustre_sb_info *lsi;
4384 struct seq_server_site *ss_site;
4385 const char *identity_upcall = "NONE";
4386 struct md_device *next;
4392 lu_device_init(&m->mdt_lu_dev, ldt);
4394 * Environment (env) might be missing mdt_thread_key values at that
4395 * point, if device is allocated when mdt_thread_key is in QUIESCENT
4398 * Usually device allocation path doesn't use module key values, but
4399 * mdt has to do a lot of work here, so allocate key value.
4401 rc = lu_env_refill((struct lu_env *)env);
4405 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4406 LASSERT(info != NULL);
4408 obd = class_name2obd(dev);
4409 LASSERT(obd != NULL);
4411 m->mdt_max_mdsize = MAX_MD_SIZE; /* 4 stripes */
4413 m->mdt_som_conf = 0;
4415 m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
4417 /* default is coordinator off, it is started through conf_param
4419 m->mdt_opts.mo_coordinator = 0;
4421 lmi = server_get_mount(dev);
4423 CERROR("Cannot get mount info for %s!\n", dev);
4426 lsi = s2lsi(lmi->lmi_sb);
4427 /* CMD is supported only in IAM mode */
4429 node_id = simple_strtol(num, NULL, 10);
4430 obd->u.obt.obt_magic = OBT_MAGIC;
4433 spin_lock_init(&m->mdt_ioepoch_lock);
4434 m->mdt_capa_timeout = CAPA_TIMEOUT;
4435 m->mdt_capa_alg = CAPA_HMAC_ALG_SHA1;
4436 m->mdt_ck_timeout = CAPA_KEY_TIMEOUT;
4437 m->mdt_squash.rsi_uid = 0;
4438 m->mdt_squash.rsi_gid = 0;
4439 INIT_LIST_HEAD(&m->mdt_squash.rsi_nosquash_nids);
4440 init_rwsem(&m->mdt_squash.rsi_sem);
4441 spin_lock_init(&m->mdt_osfs_lock);
4442 m->mdt_osfs_age = cfs_time_shift_64(-1000);
4443 m->mdt_enable_remote_dir = 0;
4444 m->mdt_enable_remote_dir_gid = 0;
4446 m->mdt_lu_dev.ld_ops = &mdt_lu_ops;
4447 m->mdt_lu_dev.ld_obd = obd;
4448 /* Set this lu_device to obd for error handling purposes. */
4449 obd->obd_lu_dev = &m->mdt_lu_dev;
4451 /* init the stack */
4452 rc = mdt_stack_init((struct lu_env *)env, m, cfg);
4454 CERROR("%s: Can't init device stack, rc %d\n",
4455 mdt_obd_name(m), rc);
4460 ss_site = mdt_seq_site(m);
4461 s->ld_seq_site = ss_site;
4464 /* set server index */
4465 ss_site->ss_node_id = node_id;
4467 /* failover is the default
4468 * FIXME: we do not failout mds0/mgs, which may cause some problems.
4469 * assumed whose ss_node_id == 0 XXX
4471 obd->obd_replayable = 1;
4472 /* No connection accepted until configurations will finish */
4473 obd->obd_no_conn = 1;
4475 if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
4476 char *str = lustre_cfg_string(cfg, 4);
4477 if (strchr(str, 'n')) {
4478 CWARN("%s: recovery disabled\n", mdt_obd_name(m));
4479 obd->obd_replayable = 0;
4483 rc = mdt_fld_init(env, mdt_obd_name(m), m);
4485 GOTO(err_fini_stack, rc);
4487 rc = mdt_seq_init(env, m);
4489 GOTO(err_fini_fld, rc);
4491 snprintf(info->mti_u.ns_name, sizeof(info->mti_u.ns_name), "%s-%s",
4492 LUSTRE_MDT_NAME, obd->obd_uuid.uuid);
4493 m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
4494 LDLM_NAMESPACE_SERVER,
4495 LDLM_NAMESPACE_GREEDY,
4497 if (m->mdt_namespace == NULL)
4498 GOTO(err_fini_seq, rc = -ENOMEM);
4500 m->mdt_namespace->ns_lvbp = m;
4501 m->mdt_namespace->ns_lvbo = &mdt_lvbo;
4503 ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
4504 /* set obd_namespace for compatibility with old code */
4505 obd->obd_namespace = m->mdt_namespace;
4507 cfs_timer_init(&m->mdt_ck_timer, mdt_ck_timer_callback, m);
4509 rc = mdt_hsm_cdt_init(m);
4511 CERROR("%s: error initializing coordinator, rc %d\n",
4512 mdt_obd_name(m), rc);
4513 GOTO(err_free_ns, rc);
4516 rc = mdt_ck_thread_start(m);
4518 GOTO(err_free_hsm, rc);
4520 rc = tgt_init(env, &m->mdt_lut, obd, m->mdt_bottom, mdt_common_slice,
4521 OBD_FAIL_MDS_ALL_REQUEST_NET,
4522 OBD_FAIL_MDS_ALL_REPLY_NET);
4526 rc = mdt_fs_setup(env, m, obd, lsi);
4530 tgt_adapt_sptlrpc_conf(&m->mdt_lut, 1);
4532 next = m->mdt_child;
4533 rc = next->md_ops->mdo_iocontrol(env, next, OBD_IOC_GET_MNTOPT, 0,
4536 GOTO(err_fs_cleanup, rc);
4538 if (mntopts & MNTOPT_USERXATTR)
4539 m->mdt_opts.mo_user_xattr = 1;
4541 m->mdt_opts.mo_user_xattr = 0;
4543 rc = next->md_ops->mdo_maxeasize_get(env, next, &m->mdt_max_ea_size);
4545 GOTO(err_fs_cleanup, rc);
4547 if (mntopts & MNTOPT_ACL)
4548 m->mdt_opts.mo_acl = 1;
4550 m->mdt_opts.mo_acl = 0;
4552 /* XXX: to support suppgid for ACL, we enable identity_upcall
4553 * by default, otherwise, maybe got unexpected -EACCESS. */
4554 if (m->mdt_opts.mo_acl)
4555 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
4557 m->mdt_identity_cache = upcall_cache_init(mdt_obd_name(m),
4559 &mdt_identity_upcall_cache_ops);
4560 if (IS_ERR(m->mdt_identity_cache)) {
4561 rc = PTR_ERR(m->mdt_identity_cache);
4562 m->mdt_identity_cache = NULL;
4563 GOTO(err_fs_cleanup, rc);
4566 rc = mdt_procfs_init(m, dev);
4568 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
4569 GOTO(err_recovery, rc);
4572 rc = mdt_quota_init(env, m, cfg);
4574 GOTO(err_procfs, rc);
4576 m->mdt_ldlm_client = &mdt2obd_dev(m)->obd_ldlm_client;
4577 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
4578 "mdt_ldlm_client", m->mdt_ldlm_client);
4580 ping_evictor_start();
4582 /* recovery will be started upon mdt_prepare()
4583 * when the whole stack is complete and ready
4584 * to serve the requests */
4586 mdt_init_capa_ctxt(env, m);
4588 /* Reduce the initial timeout on an MDS because it doesn't need such
4589 * a long timeout as an OST does. Adaptive timeouts will adjust this
4590 * value appropriately. */
4591 if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
4592 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
4598 target_recovery_fini(obd);
4599 upcall_cache_cleanup(m->mdt_identity_cache);
4600 m->mdt_identity_cache = NULL;
4602 mdt_fs_cleanup(env, m);
4604 tgt_fini(env, &m->mdt_lut);
4606 cfs_timer_disarm(&m->mdt_ck_timer);
4607 mdt_ck_thread_stop(m);
4609 mdt_hsm_cdt_fini(m);
4611 ldlm_namespace_free(m->mdt_namespace, NULL, 0);
4612 obd->obd_namespace = m->mdt_namespace = NULL;
4614 mdt_seq_fini(env, m);
4616 mdt_fld_fini(env, m);
4618 mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4621 server_put_mount(dev, true);
4625 /* For interoperability, the left element is old parameter, the right one
4626 * is the new version of the parameter, if some parameter is deprecated,
4627 * the new version should be set as NULL. */
4628 static struct cfg_interop_param mdt_interop_param[] = {
4629 { "mdt.group_upcall", NULL },
4630 { "mdt.quota_type", NULL },
4631 { "mdd.quota_type", NULL },
4632 { "mdt.rootsquash", "mdt.root_squash" },
4633 { "mdt.nosquash_nid", "mdt.nosquash_nids" },
4637 /* used by MGS to process specific configurations */
4638 static int mdt_process_config(const struct lu_env *env,
4639 struct lu_device *d, struct lustre_cfg *cfg)
4641 struct mdt_device *m = mdt_dev(d);
4642 struct md_device *md_next = m->mdt_child;
4643 struct lu_device *next = md2lu_dev(md_next);
4647 switch (cfg->lcfg_command) {
4649 struct obd_device *obd = d->ld_obd;
4651 /* For interoperability */
4652 struct cfg_interop_param *ptr = NULL;
4653 struct lustre_cfg *old_cfg = NULL;
4656 param = lustre_cfg_string(cfg, 1);
4657 if (param == NULL) {
4658 CERROR("param is empty\n");
4663 ptr = class_find_old_param(param, mdt_interop_param);
4665 if (ptr->new_param == NULL) {
4667 CWARN("For interoperability, skip this %s."
4668 " It is obsolete.\n", ptr->old_param);
4672 CWARN("Found old param %s, changed it to %s.\n",
4673 ptr->old_param, ptr->new_param);
4676 cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
4683 rc = class_process_proc_seq_param(PARAM_MDT, obd->obd_vars,
4685 if (rc > 0 || rc == -ENOSYS) {
4686 /* is it an HSM var ? */
4687 rc = class_process_proc_seq_param(PARAM_HSM,
4688 hsm_cdt_get_proc_vars(),
4690 if (rc > 0 || rc == -ENOSYS)
4691 /* we don't understand; pass it on */
4692 rc = next->ld_ops->ldo_process_config(env, next,
4696 if (old_cfg != NULL)
4697 lustre_cfg_free(cfg);
4702 /* others are passed further */
4703 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4709 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
4710 const struct lu_object_header *hdr,
4711 struct lu_device *d)
4713 struct mdt_object *mo;
4717 OBD_SLAB_ALLOC_PTR_GFP(mo, mdt_object_kmem, GFP_NOFS);
4719 struct lu_object *o;
4720 struct lu_object_header *h;
4723 h = &mo->mot_header;
4724 lu_object_header_init(h);
4725 lu_object_init(o, h, d);
4726 lu_object_add_top(h, o);
4727 o->lo_ops = &mdt_obj_ops;
4728 mutex_init(&mo->mot_ioepoch_mutex);
4729 mutex_init(&mo->mot_lov_mutex);
4730 init_rwsem(&mo->mot_open_sem);
4736 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
4737 const struct lu_object_conf *unused)
4739 struct mdt_device *d = mdt_dev(o->lo_dev);
4740 struct lu_device *under;
4741 struct lu_object *below;
4745 CDEBUG(D_INFO, "object init, fid = "DFID"\n",
4746 PFID(lu_object_fid(o)));
4748 under = &d->mdt_child->md_lu_dev;
4749 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
4750 if (below != NULL) {
4751 lu_object_add(o, below);
4758 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
4760 struct mdt_object *mo = mdt_obj(o);
4761 struct lu_object_header *h;
4765 CDEBUG(D_INFO, "object free, fid = "DFID"\n",
4766 PFID(lu_object_fid(o)));
4768 LASSERT(atomic_read(&mo->mot_open_count) == 0);
4769 LASSERT(atomic_read(&mo->mot_lease_count) == 0);
4772 lu_object_header_fini(h);
4773 OBD_SLAB_FREE_PTR(mo, mdt_object_kmem);
4778 static int mdt_object_print(const struct lu_env *env, void *cookie,
4779 lu_printer_t p, const struct lu_object *o)
4781 struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
4782 return (*p)(env, cookie, LUSTRE_MDT_NAME"-object@%p(ioepoch="LPU64" "
4783 "flags="LPX64", epochcount=%d, writecount=%d)",
4784 mdto, mdto->mot_ioepoch, mdto->mot_flags,
4785 mdto->mot_ioepoch_count, mdto->mot_writecount);
4788 static int mdt_prepare(const struct lu_env *env,
4789 struct lu_device *pdev,
4790 struct lu_device *cdev)
4792 struct mdt_device *mdt = mdt_dev(cdev);
4793 struct lu_device *next = &mdt->mdt_child->md_lu_dev;
4794 struct obd_device *obd = cdev->ld_obd;
4801 rc = next->ld_ops->ldo_prepare(env, cdev, next);
4805 rc = mdt_llog_ctxt_clone(env, mdt, LLOG_CHANGELOG_ORIG_CTXT);
4809 rc = mdt_llog_ctxt_clone(env, mdt, LLOG_AGENT_ORIG_CTXT);
4813 rc = lfsck_register_namespace(env, mdt->mdt_bottom, mdt->mdt_namespace);
4814 /* The LFSCK instance is registered just now, so it must be there when
4815 * register the namespace to such instance. */
4816 LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
4818 if (mdt->mdt_seq_site.ss_node_id == 0) {
4819 rc = mdt->mdt_child->md_ops->mdo_root_get(env, mdt->mdt_child,
4820 &mdt->mdt_md_root_fid);
4825 LASSERT(!test_bit(MDT_FL_CFGLOG, &mdt->mdt_state));
4826 target_recovery_init(&mdt->mdt_lut, tgt_request_handle);
4827 set_bit(MDT_FL_CFGLOG, &mdt->mdt_state);
4828 LASSERT(obd->obd_no_conn);
4829 spin_lock(&obd->obd_dev_lock);
4830 obd->obd_no_conn = 0;
4831 spin_unlock(&obd->obd_dev_lock);
4833 if (obd->obd_recovering == 0)
4834 mdt_postrecov(env, mdt);
4839 const struct lu_device_operations mdt_lu_ops = {
4840 .ldo_object_alloc = mdt_object_alloc,
4841 .ldo_process_config = mdt_process_config,
4842 .ldo_prepare = mdt_prepare,
4845 static const struct lu_object_operations mdt_obj_ops = {
4846 .loo_object_init = mdt_object_init,
4847 .loo_object_free = mdt_object_free,
4848 .loo_object_print = mdt_object_print
4851 static int mdt_obd_set_info_async(const struct lu_env *env,
4852 struct obd_export *exp,
4853 __u32 keylen, void *key,
4854 __u32 vallen, void *val,
4855 struct ptlrpc_request_set *set)
4861 if (KEY_IS(KEY_SPTLRPC_CONF)) {
4862 rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp), 0);
4870 * Match client and server connection feature flags.
4872 * Compute the compatibility flags for a connection request based on
4873 * features mutually supported by client and server.
4875 * The obd_export::exp_connect_data.ocd_connect_flags field in \a exp
4876 * must not be updated here, otherwise a partially initialized value may
4877 * be exposed. After the connection request is successfully processed,
4878 * the top-level MDT connect request handler atomically updates the export
4879 * connect flags from the obd_connect_data::ocd_connect_flags field of the
4880 * reply. \see mdt_connect().
4882 * \param exp the obd_export associated with this client/target pair
4883 * \param mdt the target device for the connection
4884 * \param data stores data for this connect request
4887 * \retval -EPROTO \a data unexpectedly has zero obd_connect_data::ocd_brw_size
4888 * \retval -EBADE client and server feature requirements are incompatible
4890 static int mdt_connect_internal(struct obd_export *exp,
4891 struct mdt_device *mdt,
4892 struct obd_connect_data *data)
4894 LASSERT(data != NULL);
4896 data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
4897 data->ocd_ibits_known &= MDS_INODELOCK_FULL;
4899 /* If no known bits (which should not happen, probably,
4900 as everybody should support LOOKUP and UPDATE bits at least)
4901 revert to compat mode with plain locks. */
4902 if (!data->ocd_ibits_known &&
4903 data->ocd_connect_flags & OBD_CONNECT_IBITS)
4904 data->ocd_connect_flags &= ~OBD_CONNECT_IBITS;
4906 if (!mdt->mdt_opts.mo_acl)
4907 data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
4909 if (!mdt->mdt_opts.mo_user_xattr)
4910 data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
4912 if (!mdt->mdt_som_conf)
4913 data->ocd_connect_flags &= ~OBD_CONNECT_SOM;
4915 if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE) {
4916 data->ocd_brw_size = min(data->ocd_brw_size,
4917 (__u32)MD_MAX_BRW_SIZE);
4918 if (data->ocd_brw_size == 0) {
4919 CERROR("%s: cli %s/%p ocd_connect_flags: "LPX64
4920 " ocd_version: %x ocd_grant: %d "
4921 "ocd_index: %u ocd_brw_size is "
4922 "unexpectedly zero, network data "
4923 "corruption? Refusing connection of this"
4926 exp->exp_client_uuid.uuid,
4927 exp, data->ocd_connect_flags, data->ocd_version,
4928 data->ocd_grant, data->ocd_index);
4933 /* NB: Disregard the rule against updating
4934 * exp_connect_data.ocd_connect_flags in this case, since
4935 * tgt_client_new() needs to know if this is a lightweight
4936 * connection, and it is safe to expose this flag before
4937 * connection processing completes. */
4938 if (data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) {
4939 spin_lock(&exp->exp_lock);
4940 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_LIGHTWEIGHT;
4941 spin_unlock(&exp->exp_lock);
4944 data->ocd_version = LUSTRE_VERSION_CODE;
4946 if ((data->ocd_connect_flags & OBD_CONNECT_FID) == 0) {
4947 CWARN("%s: MDS requires FID support, but client not\n",
4952 if (mdt->mdt_som_conf &&
4953 !(data->ocd_connect_flags & (OBD_CONNECT_LIGHTWEIGHT |
4954 OBD_CONNECT_MDS_MDS |
4955 OBD_CONNECT_SOM))) {
4956 CWARN("%s: MDS has SOM enabled, but client does not support "
4957 "it\n", mdt_obd_name(mdt));
4961 if (OCD_HAS_FLAG(data, PINGLESS)) {
4962 if (ptlrpc_pinger_suppress_pings()) {
4963 spin_lock(&exp->exp_obd->obd_dev_lock);
4964 list_del_init(&exp->exp_obd_chain_timed);
4965 spin_unlock(&exp->exp_obd->obd_dev_lock);
4967 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
4971 data->ocd_max_easize = mdt->mdt_max_ea_size;
4976 /* mds_connect copy */
4977 static int mdt_obd_connect(const struct lu_env *env,
4978 struct obd_export **exp, struct obd_device *obd,
4979 struct obd_uuid *cluuid,
4980 struct obd_connect_data *data,
4983 struct obd_export *lexp;
4984 struct lustre_handle conn = { 0 };
4985 struct mdt_device *mdt;
4989 LASSERT(env != NULL);
4990 if (!exp || !obd || !cluuid)
4993 mdt = mdt_dev(obd->obd_lu_dev);
4996 * first, check whether the stack is ready to handle requests
4997 * XXX: probably not very appropriate method is used now
4998 * at some point we should find a better one
5000 if (!test_bit(MDT_FL_SYNCED, &mdt->mdt_state) && data != NULL &&
5001 !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT)) {
5002 rc = obd_get_info(env, mdt->mdt_child_exp,
5003 sizeof(KEY_OSP_CONNECTED),
5004 KEY_OSP_CONNECTED, NULL, NULL, NULL);
5007 set_bit(MDT_FL_SYNCED, &mdt->mdt_state);
5010 rc = class_connect(&conn, obd, cluuid);
5014 lexp = class_conn2export(&conn);
5015 LASSERT(lexp != NULL);
5017 rc = mdt_connect_internal(lexp, mdt, data);
5019 struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
5022 memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
5023 rc = tgt_client_new(env, lexp);
5025 mdt_export_stats_init(obd, lexp, localdata);
5027 /* For phase I, sync for cross-ref operation. */
5028 lexp->exp_keep_sync = 1;
5032 class_disconnect(lexp);
5041 static int mdt_obd_reconnect(const struct lu_env *env,
5042 struct obd_export *exp, struct obd_device *obd,
5043 struct obd_uuid *cluuid,
5044 struct obd_connect_data *data,
5050 if (exp == NULL || obd == NULL || cluuid == NULL)
5053 rc = mdt_connect_internal(exp, mdt_dev(obd->obd_lu_dev), data);
5055 mdt_export_stats_init(obd, exp, localdata);
5060 static int mdt_ctxt_add_dirty_flag(struct lu_env *env,
5061 struct mdt_thread_info *info,
5062 struct mdt_file_data *mfd)
5064 struct lu_context ses;
5068 rc = lu_context_init(&ses, LCT_SERVER_SESSION);
5073 lu_context_enter(&ses);
5075 mdt_ucred(info)->uc_valid = UCRED_OLD;
5076 rc = mdt_add_dirty_flag(info, mfd->mfd_object, &info->mti_attr);
5078 lu_context_exit(&ses);
5079 lu_context_fini(&ses);
5085 static int mdt_export_cleanup(struct obd_export *exp)
5087 struct mdt_export_data *med = &exp->exp_mdt_data;
5088 struct obd_device *obd = exp->exp_obd;
5089 struct mdt_device *mdt;
5090 struct mdt_thread_info *info;
5092 CFS_LIST_HEAD(closing_list);
5093 struct mdt_file_data *mfd, *n;
5097 spin_lock(&med->med_open_lock);
5098 while (!cfs_list_empty(&med->med_open_head)) {
5099 cfs_list_t *tmp = med->med_open_head.next;
5100 mfd = cfs_list_entry(tmp, struct mdt_file_data, mfd_list);
5102 /* Remove mfd handle so it can't be found again.
5103 * We are consuming the mfd_list reference here. */
5104 class_handle_unhash(&mfd->mfd_handle);
5105 cfs_list_move_tail(&mfd->mfd_list, &closing_list);
5107 spin_unlock(&med->med_open_lock);
5108 mdt = mdt_dev(obd->obd_lu_dev);
5109 LASSERT(mdt != NULL);
5111 rc = lu_env_init(&env, LCT_MD_THREAD);
5115 info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5116 LASSERT(info != NULL);
5117 memset(info, 0, sizeof *info);
5118 info->mti_env = &env;
5119 info->mti_mdt = mdt;
5120 info->mti_exp = exp;
5122 if (!cfs_list_empty(&closing_list)) {
5123 struct md_attr *ma = &info->mti_attr;
5125 /* Close any open files (which may also cause orphan unlinking). */
5126 cfs_list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
5127 cfs_list_del_init(&mfd->mfd_list);
5128 ma->ma_need = ma->ma_valid = 0;
5130 /* This file is being closed due to an eviction, it
5131 * could have been modified and now dirty regarding to
5132 * HSM archive, check this!
5133 * The logic here is to mark a file dirty if there's a
5134 * chance it was dirtied before the client was evicted,
5135 * so that we don't have to wait for a release attempt
5136 * before finding out the file was actually dirty and
5137 * fail the release. Aggressively marking it dirty here
5138 * will cause the policy engine to attempt to
5139 * re-archive it; when rearchiving, we can compare the
5140 * current version to the HSM data_version and make the
5141 * archive request into a noop if it's not actually
5144 if (mfd->mfd_mode & (FMODE_WRITE|MDS_FMODE_TRUNC))
5145 rc = mdt_ctxt_add_dirty_flag(&env, info, mfd);
5147 /* Don't unlink orphan on failover umount, LU-184 */
5148 if (exp->exp_flags & OBD_OPT_FAILOVER) {
5149 ma->ma_valid = MA_FLAGS;
5150 ma->ma_attr_flags |= MDS_KEEP_ORPHAN;
5152 mdt_mfd_close(info, mfd);
5155 info->mti_mdt = NULL;
5156 /* cleanup client slot early */
5157 /* Do not erase record for recoverable client. */
5158 if (!(exp->exp_flags & OBD_OPT_FAILOVER) || exp->exp_failed)
5159 tgt_client_del(&env, exp);
5165 static int mdt_obd_disconnect(struct obd_export *exp)
5171 class_export_get(exp);
5173 rc = server_disconnect_export(exp);
5175 CDEBUG(D_IOCTL, "server disconnect error: %d\n", rc);
5177 rc = mdt_export_cleanup(exp);
5178 class_export_put(exp);
5182 /* FIXME: Can we avoid using these two interfaces? */
5183 static int mdt_init_export(struct obd_export *exp)
5185 struct mdt_export_data *med = &exp->exp_mdt_data;
5189 CFS_INIT_LIST_HEAD(&med->med_open_head);
5190 spin_lock_init(&med->med_open_lock);
5191 mutex_init(&med->med_idmap_mutex);
5192 med->med_idmap = NULL;
5193 spin_lock(&exp->exp_lock);
5194 exp->exp_connecting = 1;
5195 spin_unlock(&exp->exp_lock);
5197 /* self-export doesn't need client data and ldlm initialization */
5198 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5199 &exp->exp_client_uuid)))
5202 rc = tgt_client_alloc(exp);
5206 rc = ldlm_init_export(exp);
5213 tgt_client_free(exp);
5215 CERROR("%s: Failed to initialize export: rc = %d\n",
5216 exp->exp_obd->obd_name, rc);
5220 static int mdt_destroy_export(struct obd_export *exp)
5224 if (exp_connect_rmtclient(exp))
5225 mdt_cleanup_idmap(&exp->exp_mdt_data);
5227 target_destroy_export(exp);
5228 /* destroy can be called from failed obd_setup, so
5229 * checking uuid is safer than obd_self_export */
5230 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5231 &exp->exp_client_uuid)))
5234 ldlm_destroy_export(exp);
5235 tgt_client_free(exp);
5237 LASSERT(cfs_list_empty(&exp->exp_outstanding_replies));
5238 LASSERT(cfs_list_empty(&exp->exp_mdt_data.med_open_head));
5243 /** The maximum depth that fid2path() will search.
5244 * This is limited only because we want to store the fids for
5245 * historical path lookup purposes.
5247 #define MAX_PATH_DEPTH 100
5249 /** mdt_path() lookup structure. */
5250 struct path_lookup_info {
5251 __u64 pli_recno; /**< history point */
5252 __u64 pli_currec; /**< current record */
5253 struct lu_fid pli_fid;
5254 struct lu_fid pli_fids[MAX_PATH_DEPTH]; /**< path, in fids */
5255 struct mdt_object *pli_mdt_obj;
5256 char *pli_path; /**< full path */
5258 int pli_linkno; /**< which hardlink to follow */
5259 int pli_fidcount; /**< number of \a pli_fids */
5262 int mdt_links_read(struct mdt_thread_info *info, struct mdt_object *mdt_obj,
5263 struct linkea_data *ldata)
5267 LASSERT(ldata->ld_buf->lb_buf != NULL);
5269 if (!mdt_object_exists(mdt_obj))
5272 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5273 ldata->ld_buf, XATTR_NAME_LINK);
5274 if (rc == -ERANGE) {
5275 /* Buf was too small, figure out what we need. */
5276 lu_buf_free(ldata->ld_buf);
5277 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5278 ldata->ld_buf, XATTR_NAME_LINK);
5281 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
5282 if (ldata->ld_buf->lb_buf == NULL)
5284 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5285 ldata->ld_buf, XATTR_NAME_LINK);
5290 return linkea_init(ldata);
5293 static int mdt_path_current(struct mdt_thread_info *info,
5294 struct path_lookup_info *pli)
5296 struct mdt_device *mdt = info->mti_mdt;
5297 struct mdt_object *mdt_obj;
5298 struct link_ea_header *leh;
5299 struct link_ea_entry *lee;
5300 struct lu_name *tmpname = &info->mti_name;
5301 struct lu_fid *tmpfid = &info->mti_tmp_fid1;
5302 struct lu_buf *buf = &info->mti_big_buf;
5305 struct linkea_data ldata = { 0 };
5309 /* temp buffer for path element, the buffer will be finally freed
5310 * in mdt_thread_info_fini */
5311 buf = lu_buf_check_and_alloc(buf, PATH_MAX);
5312 if (buf->lb_buf == NULL)
5316 ptr = pli->pli_path + pli->pli_pathlen - 1;
5319 pli->pli_fidcount = 0;
5320 pli->pli_fids[0] = *(struct lu_fid *)mdt_object_fid(pli->pli_mdt_obj);
5321 *tmpfid = pli->pli_fids[0];
5322 /* root FID only exists on MDT0, and fid2path should also ends at MDT0,
5323 * so checking root_fid can only happen on MDT0. */
5324 while (!lu_fid_eq(&mdt->mdt_md_root_fid,
5325 &pli->pli_fids[pli->pli_fidcount])) {
5326 struct lu_buf lmv_buf;
5328 mdt_obj = mdt_object_find(info->mti_env, mdt, tmpfid);
5329 if (IS_ERR(mdt_obj))
5330 GOTO(out, rc = PTR_ERR(mdt_obj));
5332 if (mdt_object_remote(mdt_obj)) {
5333 mdt_object_put(info->mti_env, mdt_obj);
5334 GOTO(remote_out, rc = -EREMOTE);
5337 lmv_buf.lb_buf = info->mti_xattr_buf;
5338 lmv_buf.lb_len = sizeof(info->mti_xattr_buf);
5340 /* Check if it is slave stripes */
5341 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5342 &lmv_buf, XATTR_NAME_LMV);
5344 union lmv_mds_md *lmm = lmv_buf.lb_buf;
5346 /* For slave stripes, get its master */
5347 if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE) {
5348 struct lmv_mds_md_v1 *lmm1 = &lmm->lmv_md_v1;
5350 fid_le_to_cpu(tmpfid, &lmm1->lmv_master_fid);
5351 if (!fid_is_sane(tmpfid)) {
5352 mdt_object_put(info->mti_env, mdt_obj);
5353 GOTO(out, rc = -EINVAL);
5355 mdt_object_put(info->mti_env, mdt_obj);
5356 pli->pli_fids[pli->pli_fidcount] = *tmpfid;
5361 if (!mdt_object_exists(mdt_obj)) {
5362 mdt_object_put(info->mti_env, mdt_obj);
5363 GOTO(out, rc = -ENOENT);
5366 rc = mdt_links_read(info, mdt_obj, &ldata);
5367 mdt_object_put(info->mti_env, mdt_obj);
5372 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
5373 linkea_entry_unpack(lee, &reclen, tmpname, tmpfid);
5374 /* If set, use link #linkno for path lookup, otherwise use
5375 link #0. Only do this for the final path element. */
5376 if (pli->pli_fidcount == 0 &&
5377 pli->pli_linkno < leh->leh_reccount) {
5379 for (count = 0; count < pli->pli_linkno; count++) {
5380 lee = (struct link_ea_entry *)
5381 ((char *)lee + reclen);
5382 linkea_entry_unpack(lee, &reclen, tmpname,
5385 if (pli->pli_linkno < leh->leh_reccount - 1)
5386 /* indicate to user there are more links */
5390 /* Pack the name in the end of the buffer */
5391 ptr -= tmpname->ln_namelen;
5392 if (ptr - 1 <= pli->pli_path)
5393 GOTO(out, rc = -EOVERFLOW);
5394 strncpy(ptr, tmpname->ln_name, tmpname->ln_namelen);
5397 /* Store the parent fid for historic lookup */
5398 if (++pli->pli_fidcount >= MAX_PATH_DEPTH)
5399 GOTO(out, rc = -EOVERFLOW);
5400 pli->pli_fids[pli->pli_fidcount] = *tmpfid;
5404 ptr++; /* skip leading / */
5405 memmove(pli->pli_path, ptr, pli->pli_path + pli->pli_pathlen - ptr);
5412 /* Returns the full path to this fid, as of changelog record recno. */
5413 static int mdt_path(struct mdt_thread_info *info, struct mdt_object *obj,
5414 char *path, int pathlen, __u64 *recno, int *linkno,
5417 struct mdt_device *mdt = info->mti_mdt;
5418 struct path_lookup_info *pli;
5426 if (lu_fid_eq(&mdt->mdt_md_root_fid, mdt_object_fid(obj))) {
5435 pli->pli_mdt_obj = obj;
5436 pli->pli_recno = *recno;
5437 pli->pli_path = path;
5438 pli->pli_pathlen = pathlen;
5439 pli->pli_linkno = *linkno;
5441 /* Retry multiple times in case file is being moved */
5442 while (tries-- && rc == -EAGAIN)
5443 rc = mdt_path_current(info, pli);
5445 /* return the last resolved fids to the client, so the client will
5446 * build the left path on another MDT for remote object */
5447 *fid = pli->pli_fids[pli->pli_fidcount];
5449 *recno = pli->pli_currec;
5450 /* Return next link index to caller */
5451 *linkno = pli->pli_linkno;
5458 static int mdt_fid2path(struct mdt_thread_info *info,
5459 struct getinfo_fid2path *fp)
5461 struct mdt_device *mdt = info->mti_mdt;
5462 struct mdt_object *obj;
5466 CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n",
5467 PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
5469 if (!fid_is_sane(&fp->gf_fid))
5472 if (!fid_is_namespace_visible(&fp->gf_fid)) {
5473 CWARN("%s: "DFID" is invalid, sequence should be "
5474 ">= "LPX64"\n", mdt_obd_name(mdt),
5475 PFID(&fp->gf_fid), (__u64)FID_SEQ_NORMAL);
5479 obj = mdt_object_find(info->mti_env, mdt, &fp->gf_fid);
5480 if (obj == NULL || IS_ERR(obj)) {
5481 CDEBUG(D_IOCTL, "no object "DFID": %ld\n", PFID(&fp->gf_fid),
5486 if (mdt_object_remote(obj))
5488 else if (!mdt_object_exists(obj))
5494 mdt_object_put(info->mti_env, obj);
5495 CDEBUG(D_IOCTL, "nonlocal object "DFID": %d\n",
5496 PFID(&fp->gf_fid), rc);
5500 rc = mdt_path(info, obj, fp->gf_path, fp->gf_pathlen, &fp->gf_recno,
5501 &fp->gf_linkno, &fp->gf_fid);
5503 CDEBUG(D_INFO, "fid "DFID", path %s recno "LPX64" linkno %u\n",
5504 PFID(&fp->gf_fid), fp->gf_path, fp->gf_recno, fp->gf_linkno);
5506 mdt_object_put(info->mti_env, obj);
5511 static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key,
5512 void *val, int vallen)
5514 struct getinfo_fid2path *fpout, *fpin;
5517 fpin = key + cfs_size_round(sizeof(KEY_FID2PATH));
5520 if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
5521 lustre_swab_fid2path(fpin);
5523 memcpy(fpout, fpin, sizeof(*fpin));
5524 if (fpout->gf_pathlen != vallen - sizeof(*fpin))
5527 rc = mdt_fid2path(info, fpout);
5531 int mdt_get_info(struct tgt_session_info *tsi)
5541 key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
5543 CDEBUG(D_IOCTL, "No GETINFO key");
5544 RETURN(err_serious(-EFAULT));
5546 keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
5549 vallen = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_VALLEN);
5550 if (vallen == NULL) {
5551 CDEBUG(D_IOCTL, "Unable to get RMF_GETINFO_VALLEN buffer");
5552 RETURN(err_serious(-EFAULT));
5555 req_capsule_set_size(tsi->tsi_pill, &RMF_GETINFO_VAL, RCL_SERVER,
5557 rc = req_capsule_server_pack(tsi->tsi_pill);
5559 RETURN(err_serious(rc));
5561 valout = req_capsule_server_get(tsi->tsi_pill, &RMF_GETINFO_VAL);
5562 if (valout == NULL) {
5563 CDEBUG(D_IOCTL, "Unable to get get-info RPC out buffer");
5564 RETURN(err_serious(-EFAULT));
5567 if (KEY_IS(KEY_FID2PATH)) {
5568 struct mdt_thread_info *info = tsi2mdt_info(tsi);
5570 rc = mdt_rpc_fid2path(info, key, valout, *vallen);
5571 mdt_thread_info_fini(info);
5578 /* Pass the ioc down */
5579 static int mdt_ioc_child(struct lu_env *env, struct mdt_device *mdt,
5580 unsigned int cmd, int len, void *data)
5582 struct lu_context ioctl_session;
5583 struct md_device *next = mdt->mdt_child;
5587 rc = lu_context_init(&ioctl_session, LCT_SERVER_SESSION);
5590 ioctl_session.lc_thread = (struct ptlrpc_thread *)current;
5591 lu_context_enter(&ioctl_session);
5592 env->le_ses = &ioctl_session;
5594 LASSERT(next->md_ops->mdo_iocontrol);
5595 rc = next->md_ops->mdo_iocontrol(env, next, cmd, len, data);
5597 lu_context_exit(&ioctl_session);
5598 lu_context_fini(&ioctl_session);
5602 static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
5604 struct obd_ioctl_data *data = karg;
5607 struct mdt_object *obj;
5608 struct mdt_lock_handle *lh;
5612 if (data->ioc_inlbuf1 == NULL || data->ioc_inllen1 != sizeof(*fid) ||
5613 data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
5616 fid = (struct lu_fid *)data->ioc_inlbuf1;
5618 if (!fid_is_sane(fid))
5621 CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
5623 lh = &mti->mti_lh[MDT_LH_PARENT];
5624 mdt_lock_reg_init(lh, LCK_CR);
5626 obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE);
5628 RETURN(PTR_ERR(obj));
5630 if (mdt_object_remote(obj)) {
5633 * before calling version get the correct MDS should be
5634 * fid, this is error to find remote object here
5636 CERROR("nonlocal object "DFID"\n", PFID(fid));
5637 } else if (!mdt_object_exists(obj)) {
5638 *(__u64 *)data->ioc_inlbuf2 = ENOENT_VERSION;
5641 version = dt_version_get(mti->mti_env, mdt_obj2dt(obj));
5642 *(__u64 *)data->ioc_inlbuf2 = version;
5645 mdt_object_unlock_put(mti, obj, lh, 1);
5649 /* ioctls on obd dev */
5650 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
5651 void *karg, void *uarg)
5654 struct obd_device *obd = exp->exp_obd;
5655 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
5656 struct dt_device *dt = mdt->mdt_bottom;
5660 CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
5661 rc = lu_env_init(&env, LCT_MD_THREAD);
5667 rc = mdt_device_sync(&env, mdt);
5669 case OBD_IOC_SET_READONLY:
5670 rc = dt->dd_ops->dt_ro(&env, dt);
5672 case OBD_IOC_ABORT_RECOVERY:
5673 CERROR("%s: Aborting recovery for device\n", mdt_obd_name(mdt));
5674 target_stop_recovery_thread(obd);
5677 case OBD_IOC_CHANGELOG_REG:
5678 case OBD_IOC_CHANGELOG_DEREG:
5679 case OBD_IOC_CHANGELOG_CLEAR:
5680 rc = mdt_ioc_child(&env, mdt, cmd, len, karg);
5682 case OBD_IOC_START_LFSCK: {
5683 struct md_device *next = mdt->mdt_child;
5684 struct obd_ioctl_data *data = karg;
5685 struct lfsck_start_param lsp;
5687 if (unlikely(data == NULL)) {
5692 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
5693 lsp.lsp_index_valid = 0;
5694 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &lsp);
5697 case OBD_IOC_STOP_LFSCK: {
5698 struct md_device *next = mdt->mdt_child;
5699 struct obd_ioctl_data *data = karg;
5700 struct lfsck_stop stop;
5702 stop.ls_status = LS_STOPPED;
5703 /* Old lfsck utils may pass NULL @stop. */
5704 if (data->ioc_inlbuf1 == NULL)
5708 ((struct lfsck_stop *)(data->ioc_inlbuf1))->ls_flags;
5710 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &stop);
5713 case OBD_IOC_GET_OBJ_VERSION: {
5714 struct mdt_thread_info *mti;
5715 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5716 memset(mti, 0, sizeof *mti);
5717 mti->mti_env = &env;
5721 rc = mdt_ioc_version_get(mti, karg);
5724 case OBD_IOC_CATLOGLIST: {
5725 struct mdt_thread_info *mti;
5727 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5728 lu_local_obj_fid(&mti->mti_tmp_fid1, LLOG_CATALOGS_OID);
5729 rc = llog_catalog_list(&env, mdt->mdt_bottom, 0, karg,
5730 &mti->mti_tmp_fid1);
5735 CERROR("%s: Not supported cmd = %d, rc = %d\n",
5736 mdt_obd_name(mdt), cmd, rc);
5743 static int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
5745 struct lu_device *ld = md2lu_dev(mdt->mdt_child);
5746 struct lfsck_start_param lsp;
5750 lsp.lsp_start = NULL;
5751 lsp.lsp_index_valid = 0;
5752 rc = mdt->mdt_child->md_ops->mdo_iocontrol(env, mdt->mdt_child,
5753 OBD_IOC_START_LFSCK,
5755 if (rc != 0 && rc != -EALREADY)
5756 CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
5757 mdt_obd_name(mdt), rc);
5759 rc = ld->ld_ops->ldo_recovery_complete(env, ld);
5763 static int mdt_obd_postrecov(struct obd_device *obd)
5768 rc = lu_env_init(&env, LCT_MD_THREAD);
5771 rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
5776 static struct obd_ops mdt_obd_device_ops = {
5777 .o_owner = THIS_MODULE,
5778 .o_set_info_async = mdt_obd_set_info_async,
5779 .o_connect = mdt_obd_connect,
5780 .o_reconnect = mdt_obd_reconnect,
5781 .o_disconnect = mdt_obd_disconnect,
5782 .o_init_export = mdt_init_export,
5783 .o_destroy_export = mdt_destroy_export,
5784 .o_iocontrol = mdt_iocontrol,
5785 .o_postrecov = mdt_obd_postrecov,
5788 static struct lu_device* mdt_device_fini(const struct lu_env *env,
5789 struct lu_device *d)
5791 struct mdt_device *m = mdt_dev(d);
5798 static struct lu_device *mdt_device_free(const struct lu_env *env,
5799 struct lu_device *d)
5801 struct mdt_device *m = mdt_dev(d);
5804 lu_device_fini(&m->mdt_lu_dev);
5810 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
5811 struct lu_device_type *t,
5812 struct lustre_cfg *cfg)
5814 struct lu_device *l;
5815 struct mdt_device *m;
5822 rc = mdt_init0(env, m, t, cfg);
5824 mdt_device_free(env, l);
5829 l = ERR_PTR(-ENOMEM);
5833 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
5834 LU_KEY_INIT(mdt, struct mdt_thread_info);
5836 static void mdt_key_fini(const struct lu_context *ctx,
5837 struct lu_context_key *key, void* data)
5839 struct mdt_thread_info *info = data;
5841 if (info->mti_big_lmm) {
5842 OBD_FREE_LARGE(info->mti_big_lmm, info->mti_big_lmmsize);
5843 info->mti_big_lmm = NULL;
5844 info->mti_big_lmmsize = 0;
5849 /* context key: mdt_thread_key */
5850 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
5852 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info)
5854 return lu_ucred(info->mti_env);
5857 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info)
5859 return lu_ucred_check(info->mti_env);
5863 * Enable/disable COS (Commit On Sharing).
5865 * Set/Clear the COS flag in mdt options.
5867 * \param mdt mdt device
5868 * \param val 0 disables COS, other values enable COS
5870 void mdt_enable_cos(struct mdt_device *mdt, int val)
5875 mdt->mdt_opts.mo_cos = !!val;
5876 rc = lu_env_init(&env, LCT_LOCAL);
5877 if (unlikely(rc != 0)) {
5878 CWARN("lu_env initialization failed with rc = %d,"
5879 "cannot sync\n", rc);
5882 mdt_device_sync(&env, mdt);
5887 * Check COS (Commit On Sharing) status.
5889 * Return COS flag status.
5891 * \param mdt mdt device
5893 int mdt_cos_is_enabled(struct mdt_device *mdt)
5895 return mdt->mdt_opts.mo_cos != 0;
5898 static struct lu_device_type_operations mdt_device_type_ops = {
5899 .ldto_device_alloc = mdt_device_alloc,
5900 .ldto_device_free = mdt_device_free,
5901 .ldto_device_fini = mdt_device_fini
5904 static struct lu_device_type mdt_device_type = {
5905 .ldt_tags = LU_DEVICE_MD,
5906 .ldt_name = LUSTRE_MDT_NAME,
5907 .ldt_ops = &mdt_device_type_ops,
5908 .ldt_ctx_tags = LCT_MD_THREAD
5911 static int __init mdt_mod_init(void)
5915 CLASSERT(sizeof("0x0123456789ABCDEF:0x01234567:0x01234567") ==
5916 FID_NOBRACE_LEN + 1);
5917 CLASSERT(sizeof("[0x0123456789ABCDEF:0x01234567:0x01234567]") ==
5919 rc = lu_kmem_init(mdt_caches);
5923 rc = mds_mod_init();
5927 rc = class_register_type(&mdt_obd_device_ops, NULL, true, NULL,
5928 #ifndef HAVE_ONLY_PROCFS_SEQ
5931 LUSTRE_MDT_NAME, &mdt_device_type);
5936 lu_kmem_fini(mdt_caches);
5943 static void __exit mdt_mod_exit(void)
5945 class_unregister_type(LUSTRE_MDT_NAME);
5947 lu_kmem_fini(mdt_caches);
5950 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
5951 MODULE_DESCRIPTION("Lustre Metadata Target ("LUSTRE_MDT_NAME")");
5952 MODULE_LICENSE("GPL");
5954 cfs_module(mdt, LUSTRE_VERSION_STRING, mdt_mod_init, mdt_mod_exit);