4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2010, 2013, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/mdt/mdt_handler.c
38 * Lustre Metadata Target (mdt) request handler
40 * Author: Peter Braam <braam@clusterfs.com>
41 * Author: Andreas Dilger <adilger@clusterfs.com>
42 * Author: Phil Schwan <phil@clusterfs.com>
43 * Author: Mike Shaver <shaver@clusterfs.com>
44 * Author: Nikita Danilov <nikita@clusterfs.com>
45 * Author: Huang Hua <huanghua@clusterfs.com>
46 * Author: Yury Umanets <umka@clusterfs.com>
49 #define DEBUG_SUBSYSTEM S_MDS
51 #include <linux/module.h>
53 * struct OBD_{ALLOC,FREE}*()
55 #include <obd_support.h>
56 /* struct ptlrpc_request */
57 #include <lustre_net.h>
58 /* struct obd_export */
59 #include <lustre_export.h>
60 /* struct obd_device */
63 #include <dt_object.h>
64 #include <lustre_mds.h>
65 #include <lustre_log.h>
66 #include "mdt_internal.h"
67 #include <lustre_acl.h>
68 #include <lustre_param.h>
69 #include <lustre_quota.h>
70 #include <lustre_lfsck.h>
72 mdl_mode_t mdt_mdl_lock_modes[] = {
73 [LCK_MINMODE] = MDL_MINMODE,
80 [LCK_GROUP] = MDL_GROUP
83 ldlm_mode_t mdt_dlm_lock_modes[] = {
84 [MDL_MINMODE] = LCK_MINMODE,
91 [MDL_GROUP] = LCK_GROUP
94 static struct mdt_device *mdt_dev(struct lu_device *d);
95 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags);
97 static const struct lu_object_operations mdt_obj_ops;
99 /* Slab for MDT object allocation */
100 static struct kmem_cache *mdt_object_kmem;
102 /* For HSM restore handles */
103 struct kmem_cache *mdt_hsm_cdt_kmem;
105 /* For HSM request handles */
106 struct kmem_cache *mdt_hsm_car_kmem;
108 static struct lu_kmem_descr mdt_caches[] = {
110 .ckd_cache = &mdt_object_kmem,
111 .ckd_name = "mdt_obj",
112 .ckd_size = sizeof(struct mdt_object)
115 .ckd_cache = &mdt_hsm_cdt_kmem,
116 .ckd_name = "mdt_cdt_restore_handle",
117 .ckd_size = sizeof(struct cdt_restore_handle)
120 .ckd_cache = &mdt_hsm_car_kmem,
121 .ckd_name = "mdt_cdt_agent_req",
122 .ckd_size = sizeof(struct cdt_agent_req)
129 int mdt_get_disposition(struct ldlm_reply *rep, int flag)
133 return (rep->lock_policy_res1 & flag);
136 void mdt_clear_disposition(struct mdt_thread_info *info,
137 struct ldlm_reply *rep, int flag)
140 info->mti_opdata &= ~flag;
141 tgt_opdata_clear(info->mti_env, flag);
144 rep->lock_policy_res1 &= ~flag;
147 void mdt_set_disposition(struct mdt_thread_info *info,
148 struct ldlm_reply *rep, int flag)
151 info->mti_opdata |= flag;
152 tgt_opdata_set(info->mti_env, flag);
155 rep->lock_policy_res1 |= flag;
158 void mdt_lock_reg_init(struct mdt_lock_handle *lh, ldlm_mode_t lm)
160 lh->mlh_pdo_hash = 0;
161 lh->mlh_reg_mode = lm;
162 lh->mlh_rreg_mode = lm;
163 lh->mlh_type = MDT_REG_LOCK;
166 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, ldlm_mode_t lock_mode,
167 const struct lu_name *lname)
169 lh->mlh_reg_mode = lock_mode;
170 lh->mlh_rreg_mode = lock_mode;
171 lh->mlh_type = MDT_PDO_LOCK;
173 if (lu_name_is_valid(lname)) {
174 lh->mlh_pdo_hash = full_name_hash(lname->ln_name,
176 /* XXX Workaround for LU-2856
178 * Zero is a valid return value of full_name_hash, but
179 * several users of mlh_pdo_hash assume a non-zero
180 * hash value. We therefore map zero onto an
181 * arbitrary, but consistent value (1) to avoid
182 * problems further down the road. */
183 if (unlikely(lh->mlh_pdo_hash == 0))
184 lh->mlh_pdo_hash = 1;
186 lh->mlh_pdo_hash = 0;
190 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
191 struct mdt_lock_handle *lh)
197 * Any dir access needs couple of locks:
199 * 1) on part of dir we gonna take lookup/modify;
201 * 2) on whole dir to protect it from concurrent splitting and/or to
202 * flush client's cache for readdir().
204 * so, for a given mode and object this routine decides what lock mode
205 * to use for lock #2:
207 * 1) if caller's gonna lookup in dir then we need to protect dir from
208 * being splitted only - LCK_CR
210 * 2) if caller's gonna modify dir then we need to protect dir from
211 * being splitted and to flush cache - LCK_CW
213 * 3) if caller's gonna modify dir and that dir seems ready for
214 * splitting then we need to protect it from any type of access
215 * (lookup/modify/split) - LCK_EX --bzzz
218 LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
219 LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
222 * Ask underlaying level its opinion about preferable PDO lock mode
223 * having access type passed as regular lock mode:
225 * - MDL_MINMODE means that lower layer does not want to specify lock
228 * - MDL_NL means that no PDO lock should be taken. This is used in some
229 * cases. Say, for non-splittable directories no need to use PDO locks
232 mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
233 mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
235 if (mode != MDL_MINMODE) {
236 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
239 * Lower layer does not want to specify locking mode. We do it
240 * our selves. No special protection is needed, just flush
241 * client's cache on modification and allow concurrent
244 switch (lh->mlh_reg_mode) {
246 lh->mlh_pdo_mode = LCK_EX;
249 lh->mlh_pdo_mode = LCK_CR;
252 lh->mlh_pdo_mode = LCK_CW;
255 CERROR("Not expected lock type (0x%x)\n",
256 (int)lh->mlh_reg_mode);
261 LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
265 static int mdt_getstatus(struct tgt_session_info *tsi)
267 struct mdt_thread_info *info = tsi2mdt_info(tsi);
268 struct mdt_device *mdt = info->mti_mdt;
269 struct mdt_body *repbody;
274 rc = mdt_check_ucred(info);
276 GOTO(out, rc = err_serious(rc));
278 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETSTATUS_PACK))
279 GOTO(out, rc = err_serious(-ENOMEM));
281 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
282 repbody->fid1 = mdt->mdt_md_root_fid;
283 repbody->valid |= OBD_MD_FLID;
285 if (tsi->tsi_tgt->lut_mds_capa &&
286 exp_connect_flags(info->mti_exp) & OBD_CONNECT_MDS_CAPA) {
287 struct mdt_object *root;
288 struct lustre_capa *capa;
290 root = mdt_object_find(info->mti_env, mdt, &repbody->fid1);
292 GOTO(out, rc = PTR_ERR(root));
294 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA1);
296 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
297 rc = mo_capa_get(info->mti_env, mdt_object_child(root), capa,
299 mdt_object_put(info->mti_env, root);
301 repbody->valid |= OBD_MD_FLMDSCAPA;
305 mdt_thread_info_fini(info);
309 static int mdt_statfs(struct tgt_session_info *tsi)
311 struct ptlrpc_request *req = tgt_ses_req(tsi);
312 struct mdt_thread_info *info = tsi2mdt_info(tsi);
313 struct md_device *next = info->mti_mdt->mdt_child;
314 struct ptlrpc_service_part *svcpt;
315 struct obd_statfs *osfs;
320 svcpt = req->rq_rqbd->rqbd_svcpt;
322 /* This will trigger a watchdog timeout */
323 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
324 (MDT_SERVICE_WATCHDOG_FACTOR *
325 at_get(&svcpt->scp_at_estimate)) + 1);
327 rc = mdt_check_ucred(info);
329 GOTO(out, rc = err_serious(rc));
331 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK))
332 GOTO(out, rc = err_serious(-ENOMEM));
334 osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
336 GOTO(out, rc = -EPROTO);
338 /** statfs information are cached in the mdt_device */
339 if (cfs_time_before_64(info->mti_mdt->mdt_osfs_age,
340 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS))) {
341 /** statfs data is too old, get up-to-date one */
342 rc = next->md_ops->mdo_statfs(info->mti_env, next, osfs);
345 spin_lock(&info->mti_mdt->mdt_osfs_lock);
346 info->mti_mdt->mdt_osfs = *osfs;
347 info->mti_mdt->mdt_osfs_age = cfs_time_current_64();
348 spin_unlock(&info->mti_mdt->mdt_osfs_lock);
350 /** use cached statfs data */
351 spin_lock(&info->mti_mdt->mdt_osfs_lock);
352 *osfs = info->mti_mdt->mdt_osfs;
353 spin_unlock(&info->mti_mdt->mdt_osfs_lock);
357 mdt_counter_incr(req, LPROC_MDT_STATFS);
359 mdt_thread_info_fini(info);
364 * Pack SOM attributes into the reply.
365 * Call under a DLM UPDATE lock.
367 static void mdt_pack_size2body(struct mdt_thread_info *info,
368 struct mdt_object *mo)
371 struct md_attr *ma = &info->mti_attr;
373 LASSERT(ma->ma_attr.la_valid & LA_MODE);
374 b = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
376 /* Check if Size-on-MDS is supported, if this is a regular file,
377 * if SOM is enabled on the object and if SOM cache exists and valid.
378 * Otherwise do not pack Size-on-MDS attributes to the reply. */
379 if (!(mdt_conn_flags(info) & OBD_CONNECT_SOM) ||
380 !S_ISREG(ma->ma_attr.la_mode) ||
381 !mdt_object_is_som_enabled(mo) ||
382 !(ma->ma_valid & MA_SOM))
385 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
386 b->size = ma->ma_som->msd_size;
387 b->blocks = ma->ma_som->msd_blocks;
390 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
391 const struct lu_attr *attr, const struct lu_fid *fid)
393 struct md_attr *ma = &info->mti_attr;
395 LASSERT(ma->ma_valid & MA_INODE);
397 b->atime = attr->la_atime;
398 b->mtime = attr->la_mtime;
399 b->ctime = attr->la_ctime;
400 b->mode = attr->la_mode;
401 b->size = attr->la_size;
402 b->blocks = attr->la_blocks;
403 b->uid = attr->la_uid;
404 b->gid = attr->la_gid;
405 b->flags = attr->la_flags;
406 b->nlink = attr->la_nlink;
407 b->rdev = attr->la_rdev;
409 /*XXX should pack the reply body according to lu_valid*/
410 b->valid |= OBD_MD_FLCTIME | OBD_MD_FLUID |
411 OBD_MD_FLGID | OBD_MD_FLTYPE |
412 OBD_MD_FLMODE | OBD_MD_FLNLINK | OBD_MD_FLFLAGS |
413 OBD_MD_FLATIME | OBD_MD_FLMTIME ;
415 if (!S_ISREG(attr->la_mode)) {
416 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
417 } else if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV)) {
418 /* means no objects are allocated on osts. */
419 LASSERT(!(ma->ma_valid & MA_LOV));
420 /* just ignore blocks occupied by extend attributes on MDS */
422 /* if no object is allocated on osts, the size on mds is valid. b=22272 */
423 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
424 } else if ((ma->ma_valid & MA_LOV) && ma->ma_lmm != NULL &&
425 ma->ma_lmm->lmm_pattern & LOV_PATTERN_F_RELEASED) {
426 /* A released file stores its size on MDS. */
427 /* But return 1 block for released file, unless tools like tar
428 * will consider it fully sparse. (LU-3864)
430 if (unlikely(b->size == 0))
434 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
439 b->valid |= OBD_MD_FLID;
440 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, size="LPU64"\n",
441 PFID(fid), b->nlink, b->mode, b->size);
445 mdt_body_reverse_idmap(info, b);
447 if (fid != NULL && (b->valid & OBD_MD_FLSIZE))
448 CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
449 PFID(fid), (unsigned long long)b->size);
452 static inline int mdt_body_has_lov(const struct lu_attr *la,
453 const struct mdt_body *body)
455 return ((S_ISREG(la->la_mode) && (body->valid & OBD_MD_FLEASIZE)) ||
456 (S_ISDIR(la->la_mode) && (body->valid & OBD_MD_FLDIREA )) );
459 void mdt_client_compatibility(struct mdt_thread_info *info)
461 struct mdt_body *body;
462 struct ptlrpc_request *req = mdt_info_req(info);
463 struct obd_export *exp = req->rq_export;
464 struct md_attr *ma = &info->mti_attr;
465 struct lu_attr *la = &ma->ma_attr;
468 if (exp_connect_layout(exp))
469 /* the client can deal with 16-bit lmm_stripe_count */
472 body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
474 if (!mdt_body_has_lov(la, body))
477 /* now we have a reply with a lov for a client not compatible with the
478 * layout lock so we have to clean the layout generation number */
479 if (S_ISREG(la->la_mode))
480 ma->ma_lmm->lmm_layout_gen = 0;
484 int mdt_attr_get_eabuf_size(struct mdt_thread_info *info, struct mdt_object *o)
486 const struct lu_env *env = info->mti_env;
489 rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
498 /* Is it a directory? Let's check for the LMV as well */
499 if (S_ISDIR(lu_object_attr(&mdt_object_child(o)->mo_lu))) {
500 rc2 = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
502 if ((rc2 < 0 && rc2 != -ENODATA) || (rc2 > rc))
510 static int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
513 const struct lu_env *env = info->mti_env;
517 LASSERT(info->mti_big_lmm_used == 0);
518 rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL, name);
522 /* big_lmm may need to be grown */
523 if (info->mti_big_lmmsize < rc) {
524 int size = size_roundup_power2(rc);
526 if (info->mti_big_lmmsize > 0) {
527 /* free old buffer */
528 LASSERT(info->mti_big_lmm);
529 OBD_FREE_LARGE(info->mti_big_lmm,
530 info->mti_big_lmmsize);
531 info->mti_big_lmm = NULL;
532 info->mti_big_lmmsize = 0;
535 OBD_ALLOC_LARGE(info->mti_big_lmm, size);
536 if (info->mti_big_lmm == NULL)
538 info->mti_big_lmmsize = size;
540 LASSERT(info->mti_big_lmmsize >= rc);
542 info->mti_buf.lb_buf = info->mti_big_lmm;
543 info->mti_buf.lb_len = info->mti_big_lmmsize;
544 rc = mo_xattr_get(env, mdt_object_child(o), &info->mti_buf, name);
549 int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
550 struct md_attr *ma, const char *name)
552 struct md_object *next = mdt_object_child(o);
553 struct lu_buf *buf = &info->mti_buf;
556 if (strcmp(name, XATTR_NAME_LOV) == 0) {
557 buf->lb_buf = ma->ma_lmm;
558 buf->lb_len = ma->ma_lmm_size;
559 LASSERT(!(ma->ma_valid & MA_LOV));
560 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
561 buf->lb_buf = ma->ma_lmv;
562 buf->lb_len = ma->ma_lmv_size;
563 LASSERT(!(ma->ma_valid & MA_LMV));
564 } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
565 buf->lb_buf = ma->ma_lmv;
566 buf->lb_len = ma->ma_lmv_size;
567 LASSERT(!(ma->ma_valid & MA_LMV_DEF));
572 rc = mo_xattr_get(info->mti_env, next, buf, name);
574 if (strcmp(name, XATTR_NAME_LOV) == 0) {
575 ma->ma_lmm_size = rc;
576 ma->ma_valid |= MA_LOV;
577 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
578 ma->ma_lmv_size = rc;
579 ma->ma_valid |= MA_LMV;
580 } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
581 ma->ma_lmv_size = rc;
582 ma->ma_valid |= MA_LMV_DEF;
586 } else if (rc == -ENODATA) {
589 } else if (rc == -ERANGE) {
590 /* Default LMV has fixed size, so it must be able to fit
591 * in the original buffer */
592 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
594 rc = mdt_big_xattr_get(info, o, name);
596 info->mti_big_lmm_used = 1;
597 if (!strcmp(name, XATTR_NAME_LOV)) {
598 ma->ma_valid |= MA_LOV;
599 ma->ma_lmm = info->mti_big_lmm;
600 ma->ma_lmm_size = rc;
601 } else if (!strcmp(name, XATTR_NAME_LMV)) {
602 ma->ma_valid |= MA_LMV;
603 ma->ma_lmv = info->mti_big_lmm;
604 ma->ma_lmv_size = rc;
609 /* update mdt_max_mdsize so all clients
610 * will be aware about that */
611 if (info->mti_mdt->mdt_max_mdsize < rc)
612 info->mti_mdt->mdt_max_mdsize = rc;
620 static int mdt_attr_get_pfid(struct mdt_thread_info *info,
621 struct mdt_object *o, struct lu_fid *pfid)
623 struct lu_buf *buf = &info->mti_buf;
624 struct link_ea_header *leh;
625 struct link_ea_entry *lee;
629 buf->lb_buf = info->mti_big_lmm;
630 buf->lb_len = info->mti_big_lmmsize;
631 rc = mo_xattr_get(info->mti_env, mdt_object_child(o),
632 buf, XATTR_NAME_LINK);
633 /* ignore errors, MA_PFID won't be set and it is
634 * up to the caller to treat this as an error */
635 if (rc == -ERANGE || buf->lb_len == 0) {
636 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
637 buf->lb_buf = info->mti_big_lmm;
638 buf->lb_len = info->mti_big_lmmsize;
643 if (rc < sizeof(*leh)) {
644 CERROR("short LinkEA on "DFID": rc = %d\n",
645 PFID(mdt_object_fid(o)), rc);
649 leh = (struct link_ea_header *) buf->lb_buf;
650 lee = (struct link_ea_entry *)(leh + 1);
651 if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
652 leh->leh_magic = LINK_EA_MAGIC;
653 leh->leh_reccount = __swab32(leh->leh_reccount);
654 leh->leh_len = __swab64(leh->leh_len);
656 if (leh->leh_magic != LINK_EA_MAGIC)
658 if (leh->leh_reccount == 0)
661 memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
662 fid_be_to_cpu(pfid, pfid);
667 int mdt_attr_get_complex(struct mdt_thread_info *info,
668 struct mdt_object *o, struct md_attr *ma)
670 const struct lu_env *env = info->mti_env;
671 struct md_object *next = mdt_object_child(o);
672 struct lu_buf *buf = &info->mti_buf;
673 u32 mode = lu_object_attr(&next->mo_lu);
674 int need = ma->ma_need;
680 if (need & MA_INODE) {
681 ma->ma_need = MA_INODE;
682 rc = mo_attr_get(env, next, ma);
685 ma->ma_valid |= MA_INODE;
688 if (need & MA_PFID) {
689 rc = mdt_attr_get_pfid(info, o, &ma->ma_pfid);
691 ma->ma_valid |= MA_PFID;
692 /* ignore this error, parent fid is not mandatory */
696 if (need & MA_LOV && (S_ISREG(mode) || S_ISDIR(mode))) {
697 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LOV);
702 if (need & MA_LMV && S_ISDIR(mode)) {
703 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LMV);
708 if (need & MA_LMV_DEF && S_ISDIR(mode)) {
709 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_DEFAULT_LMV);
714 if (need & MA_SOM && S_ISREG(mode)) {
715 buf->lb_buf = info->mti_xattr_buf;
716 buf->lb_len = sizeof(info->mti_xattr_buf);
717 CLASSERT(sizeof(struct som_attrs) <=
718 sizeof(info->mti_xattr_buf));
719 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_SOM);
720 rc2 = lustre_buf2som(info->mti_xattr_buf, rc2, ma->ma_som);
722 ma->ma_valid |= MA_SOM;
723 else if (rc2 < 0 && rc2 != -ENODATA)
727 if (need & MA_HSM && S_ISREG(mode)) {
728 buf->lb_buf = info->mti_xattr_buf;
729 buf->lb_len = sizeof(info->mti_xattr_buf);
730 CLASSERT(sizeof(struct hsm_attrs) <=
731 sizeof(info->mti_xattr_buf));
732 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_HSM);
733 rc2 = lustre_buf2hsm(info->mti_xattr_buf, rc2, &ma->ma_hsm);
735 ma->ma_valid |= MA_HSM;
736 else if (rc2 < 0 && rc2 != -ENODATA)
740 #ifdef CONFIG_FS_POSIX_ACL
741 if (need & MA_ACL_DEF && S_ISDIR(mode)) {
742 buf->lb_buf = ma->ma_acl;
743 buf->lb_len = ma->ma_acl_size;
744 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
746 ma->ma_acl_size = rc2;
747 ma->ma_valid |= MA_ACL_DEF;
748 } else if (rc2 == -ENODATA) {
757 CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = "LPX64" ma_lmm=%p\n",
758 rc, ma->ma_valid, ma->ma_lmm);
762 static int mdt_getattr_internal(struct mdt_thread_info *info,
763 struct mdt_object *o, int ma_need)
765 struct md_object *next = mdt_object_child(o);
766 const struct mdt_body *reqbody = info->mti_body;
767 struct ptlrpc_request *req = mdt_info_req(info);
768 struct md_attr *ma = &info->mti_attr;
769 struct lu_attr *la = &ma->ma_attr;
770 struct req_capsule *pill = info->mti_pill;
771 const struct lu_env *env = info->mti_env;
772 struct mdt_body *repbody;
773 struct lu_buf *buffer = &info->mti_buf;
778 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
779 RETURN(err_serious(-ENOMEM));
781 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
785 if (mdt_object_remote(o)) {
786 /* This object is located on remote node.*/
787 /* Return -EIO for old client */
788 if (!mdt_is_dne_client(req->rq_export))
789 GOTO(out, rc = -EIO);
791 repbody->fid1 = *mdt_object_fid(o);
792 repbody->valid = OBD_MD_FLID | OBD_MD_MDS;
796 if (reqbody->eadatasize > 0) {
797 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
798 if (buffer->lb_buf == NULL)
799 GOTO(out, rc = -EPROTO);
800 buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD,
803 buffer->lb_buf = NULL;
805 ma_need &= ~(MA_LOV | MA_LMV);
806 CDEBUG(D_INFO, "%s: RPC from %s: does not need LOVEA.\n",
807 mdt_obd_name(info->mti_mdt),
808 req->rq_export->exp_client_uuid.uuid);
811 /* If it is dir object and client require MEA, then we got MEA */
812 if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
813 (reqbody->valid & (OBD_MD_MEA | OBD_MD_DEFAULT_MEA))) {
814 /* Assumption: MDT_MD size is enough for lmv size. */
815 ma->ma_lmv = buffer->lb_buf;
816 ma->ma_lmv_size = buffer->lb_len;
817 ma->ma_need = MA_INODE;
818 if (ma->ma_lmv_size > 0) {
819 if (reqbody->valid & OBD_MD_MEA)
820 ma->ma_need |= MA_LMV;
821 else if (reqbody->valid & OBD_MD_DEFAULT_MEA)
822 ma->ma_need |= MA_LMV_DEF;
825 ma->ma_lmm = buffer->lb_buf;
826 ma->ma_lmm_size = buffer->lb_len;
827 ma->ma_need = MA_INODE | MA_HSM;
828 if (ma->ma_lmm_size > 0)
829 ma->ma_need |= MA_LOV;
832 if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
833 reqbody->valid & OBD_MD_FLDIREA &&
834 lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
835 /* get default stripe info for this dir. */
836 ma->ma_need |= MA_LOV_DEF;
838 ma->ma_need |= ma_need;
839 if (ma->ma_need & MA_SOM)
840 ma->ma_som = &info->mti_u.som.data;
842 rc = mdt_attr_get_complex(info, o, ma);
844 CERROR("%s: getattr error for "DFID": rc = %d\n",
845 mdt_obd_name(info->mti_mdt),
846 PFID(mdt_object_fid(o)), rc);
850 /* if file is released, check if a restore is running */
851 if ((ma->ma_valid & MA_HSM) && (ma->ma_hsm.mh_flags & HS_RELEASED) &&
852 mdt_hsm_restore_is_running(info, mdt_object_fid(o))) {
853 repbody->t_state = MS_RESTORE;
854 repbody->valid |= OBD_MD_TSTATE;
857 is_root = lu_fid_eq(mdt_object_fid(o), &info->mti_mdt->mdt_md_root_fid);
859 /* the Lustre protocol supposes to return default striping
860 * on the user-visible root if explicitly requested */
861 if ((ma->ma_valid & MA_LOV) == 0 && S_ISDIR(la->la_mode) &&
862 (ma->ma_need & MA_LOV_DEF && is_root) && ma->ma_need & MA_LOV) {
863 struct lu_fid rootfid;
864 struct mdt_object *root;
865 struct mdt_device *mdt = info->mti_mdt;
867 rc = dt_root_get(env, mdt->mdt_bottom, &rootfid);
870 root = mdt_object_find(env, mdt, &rootfid);
872 RETURN(PTR_ERR(root));
873 rc = mdt_stripe_get(info, root, ma, XATTR_NAME_LOV);
874 mdt_object_put(info->mti_env, root);
876 CERROR("%s: getattr error for "DFID": rc = %d\n",
877 mdt_obd_name(info->mti_mdt),
878 PFID(mdt_object_fid(o)), rc);
883 if (likely(ma->ma_valid & MA_INODE))
884 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
888 if (mdt_body_has_lov(la, reqbody)) {
889 if (ma->ma_valid & MA_LOV) {
890 LASSERT(ma->ma_lmm_size);
891 repbody->eadatasize = ma->ma_lmm_size;
892 if (S_ISDIR(la->la_mode))
893 repbody->valid |= OBD_MD_FLDIREA;
895 repbody->valid |= OBD_MD_FLEASIZE;
896 mdt_dump_lmm(D_INFO, ma->ma_lmm, repbody->valid);
898 if (ma->ma_valid & MA_LMV) {
899 LASSERT(S_ISDIR(la->la_mode));
900 mdt_dump_lmv(D_INFO, ma->ma_lmv);
901 repbody->eadatasize = ma->ma_lmv_size;
902 repbody->valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
904 if (ma->ma_valid & MA_LMV_DEF) {
905 LASSERT(S_ISDIR(la->la_mode));
906 repbody->eadatasize = ma->ma_lmv_size;
907 repbody->valid |= (OBD_MD_FLDIREA|OBD_MD_DEFAULT_MEA);
909 } else if (S_ISLNK(la->la_mode) &&
910 reqbody->valid & OBD_MD_LINKNAME) {
911 buffer->lb_buf = ma->ma_lmm;
912 /* eadatasize from client includes NULL-terminator, so
913 * there is no need to read it */
914 buffer->lb_len = reqbody->eadatasize - 1;
915 rc = mo_readlink(env, next, buffer);
916 if (unlikely(rc <= 0)) {
917 CERROR("%s: readlink failed for "DFID": rc = %d\n",
918 mdt_obd_name(info->mti_mdt),
919 PFID(mdt_object_fid(o)), rc);
922 int print_limit = min_t(int, PAGE_CACHE_SIZE - 128, rc);
924 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
926 repbody->valid |= OBD_MD_LINKNAME;
927 /* we need to report back size with NULL-terminator
928 * because client expects that */
929 repbody->eadatasize = rc + 1;
930 if (repbody->eadatasize != reqbody->eadatasize)
931 CDEBUG(D_INODE, "%s: Read shorter symlink %d "
932 "on "DFID ", expected %d\n",
933 mdt_obd_name(info->mti_mdt),
934 rc, PFID(mdt_object_fid(o)),
935 reqbody->eadatasize - 1);
937 ((char *)ma->ma_lmm)[rc] = 0;
939 /* If the total CDEBUG() size is larger than a page, it
940 * will print a warning to the console, avoid this by
941 * printing just the last part of the symlink. */
942 CDEBUG(D_INODE, "symlink dest %s%.*s, len = %d\n",
943 print_limit < rc ? "..." : "", print_limit,
944 (char *)ma->ma_lmm + rc - print_limit, rc);
949 if (reqbody->valid & OBD_MD_FLMODEASIZE) {
950 repbody->max_cookiesize = 0;
951 repbody->max_mdsize = info->mti_mdt->mdt_max_mdsize;
952 repbody->valid |= OBD_MD_FLMODEASIZE;
953 CDEBUG(D_INODE, "I am going to change the MAX_MD_SIZE & "
954 "MAX_COOKIE to : %d:%d\n", repbody->max_mdsize,
955 repbody->max_cookiesize);
958 if (exp_connect_rmtclient(info->mti_exp) &&
959 reqbody->valid & OBD_MD_FLRMTPERM) {
960 void *buf = req_capsule_server_get(pill, &RMF_ACL);
962 /* mdt_getattr_lock only */
963 rc = mdt_pack_remote_perm(info, o, buf);
965 repbody->valid &= ~OBD_MD_FLRMTPERM;
966 repbody->aclsize = 0;
969 repbody->valid |= OBD_MD_FLRMTPERM;
970 repbody->aclsize = sizeof(struct mdt_remote_perm);
973 #ifdef CONFIG_FS_POSIX_ACL
974 else if ((exp_connect_flags(req->rq_export) & OBD_CONNECT_ACL) &&
975 (reqbody->valid & OBD_MD_FLACL)) {
976 buffer->lb_buf = req_capsule_server_get(pill, &RMF_ACL);
977 buffer->lb_len = req_capsule_get_size(pill,
978 &RMF_ACL, RCL_SERVER);
979 if (buffer->lb_len > 0) {
980 rc = mo_xattr_get(env, next, buffer,
981 XATTR_NAME_ACL_ACCESS);
983 if (rc == -ENODATA) {
984 repbody->aclsize = 0;
985 repbody->valid |= OBD_MD_FLACL;
987 } else if (rc == -EOPNOTSUPP) {
990 CERROR("%s: unable to read "DFID
992 mdt_obd_name(info->mti_mdt),
993 PFID(mdt_object_fid(o)), rc);
996 repbody->aclsize = rc;
997 repbody->valid |= OBD_MD_FLACL;
1004 if (reqbody->valid & OBD_MD_FLMDSCAPA &&
1005 info->mti_mdt->mdt_lut.lut_mds_capa &&
1006 exp_connect_flags(info->mti_exp) & OBD_CONNECT_MDS_CAPA) {
1007 struct lustre_capa *capa;
1009 capa = req_capsule_server_get(pill, &RMF_CAPA1);
1011 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
1012 rc = mo_capa_get(env, next, capa, 0);
1015 repbody->valid |= OBD_MD_FLMDSCAPA;
1020 mdt_counter_incr(req, LPROC_MDT_GETATTR);
1025 static int mdt_renew_capa(struct mdt_thread_info *info)
1027 struct mdt_object *obj = info->mti_object;
1028 struct mdt_body *body;
1029 struct lustre_capa *capa, *c;
1033 /* if object doesn't exist, or server has disabled capability,
1034 * return directly, client will find body->valid OBD_MD_FLOSSCAPA
1037 if (!obj || !info->mti_mdt->mdt_lut.lut_oss_capa ||
1038 !(exp_connect_flags(info->mti_exp) & OBD_CONNECT_OSS_CAPA))
1041 body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1042 LASSERT(body != NULL);
1044 c = req_capsule_client_get(info->mti_pill, &RMF_CAPA1);
1047 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
1051 rc = mo_capa_get(info->mti_env, mdt_object_child(obj), capa, 1);
1053 body->valid |= OBD_MD_FLOSSCAPA;
1057 static int mdt_getattr(struct tgt_session_info *tsi)
1059 struct mdt_thread_info *info = tsi2mdt_info(tsi);
1060 struct mdt_object *obj = info->mti_object;
1061 struct req_capsule *pill = info->mti_pill;
1062 struct mdt_body *reqbody;
1063 struct mdt_body *repbody;
1067 reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
1070 if (reqbody->valid & OBD_MD_FLOSSCAPA) {
1071 rc = req_capsule_server_pack(pill);
1073 RETURN(err_serious(rc));
1074 rc = mdt_renew_capa(info);
1075 GOTO(out_shrink, rc);
1078 LASSERT(obj != NULL);
1079 LASSERT(lu_object_assert_exists(&obj->mot_obj));
1082 if (reqbody->valid & OBD_MD_LINKNAME) {
1083 /* No easy way to know how long is the symlink, but it cannot
1084 * be more than PATH_MAX, so we allocate +1 */
1087 /* A special case for fs ROOT: getattr there might fetch
1088 * default EA for entire fs, not just for this dir!
1090 } else if (lu_fid_eq(mdt_object_fid(obj),
1091 &info->mti_mdt->mdt_md_root_fid) &&
1092 (reqbody->valid & OBD_MD_FLDIREA) &&
1093 (lustre_msg_get_opc(mdt_info_req(info)->rq_reqmsg) ==
1095 /* Should the default strping be bigger, mdt_fix_reply
1096 * will reallocate */
1097 rc = DEF_REP_MD_SIZE;
1099 /* Hopefully no race in EA change for either file or directory?
1101 rc = mdt_attr_get_eabuf_size(info, obj);
1105 GOTO(out_shrink, rc);
1107 /* old clients may not report needed easize, use max value then */
1108 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, rc);
1110 rc = req_capsule_server_pack(pill);
1111 if (unlikely(rc != 0))
1112 GOTO(out, rc = err_serious(rc));
1114 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1115 LASSERT(repbody != NULL);
1116 repbody->eadatasize = 0;
1117 repbody->aclsize = 0;
1119 if (reqbody->valid & OBD_MD_FLRMTPERM)
1120 rc = mdt_init_ucred(info, reqbody);
1122 rc = mdt_check_ucred(info);
1124 GOTO(out_shrink, rc);
1126 info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
1129 * Don't check capability at all, because rename might getattr for
1130 * remote obj, and at that time no capability is available.
1132 mdt_set_capainfo(info, 1, &reqbody->fid1, BYPASS_CAPA);
1133 rc = mdt_getattr_internal(info, obj, 0);
1134 if (reqbody->valid & OBD_MD_FLRMTPERM)
1135 mdt_exit_ucred(info);
1138 mdt_client_compatibility(info);
1139 rc2 = mdt_fix_reply(info);
1143 mdt_thread_info_fini(info);
1147 static int mdt_is_subdir(struct tgt_session_info *tsi)
1149 struct mdt_thread_info *info = tsi2mdt_info(tsi);
1150 struct mdt_object *o = info->mti_object;
1151 struct req_capsule *pill = info->mti_pill;
1152 const struct mdt_body *body = info->mti_body;
1153 struct mdt_body *repbody;
1159 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1162 * We save last checked parent fid to @repbody->fid1 for remote
1165 LASSERT(fid_is_sane(&body->fid2));
1166 LASSERT(mdt_object_exists(o) && !mdt_object_remote(o));
1167 rc = mdo_is_subdir(info->mti_env, mdt_object_child(o),
1168 &body->fid2, &repbody->fid1);
1169 if (rc == 0 || rc == -EREMOTE)
1170 repbody->valid |= OBD_MD_FLID;
1172 mdt_thread_info_fini(info);
1176 static int mdt_swap_layouts(struct tgt_session_info *tsi)
1178 struct mdt_thread_info *info;
1179 struct ptlrpc_request *req = tgt_ses_req(tsi);
1180 struct obd_export *exp = req->rq_export;
1181 struct mdt_object *o1, *o2, *o;
1182 struct mdt_lock_handle *lh1, *lh2;
1183 struct mdc_swap_layouts *msl;
1187 /* client does not support layout lock, so layout swaping
1189 * FIXME: there is a problem for old clients which don't support
1190 * layout lock yet. If those clients have already opened the file
1191 * they won't be notified at all so that old layout may still be
1192 * used to do IO. This can be fixed after file release is landed by
1193 * doing exclusive open and taking full EX ibits lock. - Jinshan */
1194 if (!exp_connect_layout(exp))
1195 RETURN(-EOPNOTSUPP);
1197 info = tsi2mdt_info(tsi);
1199 if (info->mti_dlm_req != NULL)
1200 ldlm_request_cancel(req, info->mti_dlm_req, 0);
1202 if (req_capsule_get_size(info->mti_pill, &RMF_CAPA1, RCL_CLIENT))
1203 mdt_set_capainfo(info, 0, &info->mti_body->fid1,
1204 req_capsule_client_get(info->mti_pill,
1207 if (req_capsule_get_size(info->mti_pill, &RMF_CAPA2, RCL_CLIENT))
1208 mdt_set_capainfo(info, 1, &info->mti_body->fid2,
1209 req_capsule_client_get(info->mti_pill,
1212 o1 = info->mti_object;
1213 o = o2 = mdt_object_find(info->mti_env, info->mti_mdt,
1214 &info->mti_body->fid2);
1216 GOTO(out, rc = PTR_ERR(o));
1218 if (mdt_object_remote(o) || !mdt_object_exists(o)) /* remote object */
1219 GOTO(put, rc = -ENOENT);
1221 rc = lu_fid_cmp(&info->mti_body->fid1, &info->mti_body->fid2);
1222 if (unlikely(rc == 0)) /* same file, you kidding me? no-op. */
1228 /* permission check. Make sure the calling process having permission
1229 * to write both files. */
1230 rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
1235 rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL,
1240 msl = req_capsule_client_get(info->mti_pill, &RMF_SWAP_LAYOUTS);
1242 GOTO(put, rc = -EPROTO);
1244 lh1 = &info->mti_lh[MDT_LH_NEW];
1245 mdt_lock_reg_init(lh1, LCK_EX);
1246 lh2 = &info->mti_lh[MDT_LH_OLD];
1247 mdt_lock_reg_init(lh2, LCK_EX);
1249 rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT |
1250 MDS_INODELOCK_XATTR, MDT_LOCAL_LOCK);
1254 rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT |
1255 MDS_INODELOCK_XATTR, MDT_LOCAL_LOCK);
1259 rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1),
1260 mdt_object_child(o2), msl->msl_flags);
1263 mdt_object_unlock(info, o2, lh2, rc);
1265 mdt_object_unlock(info, o1, lh1, rc);
1267 mdt_object_put(info->mti_env, o);
1269 mdt_thread_info_fini(info);
1273 static int mdt_raw_lookup(struct mdt_thread_info *info,
1274 struct mdt_object *parent,
1275 const struct lu_name *lname,
1276 struct ldlm_reply *ldlm_rep)
1278 struct md_object *next = mdt_object_child(info->mti_object);
1279 const struct mdt_body *reqbody = info->mti_body;
1280 struct lu_fid *child_fid = &info->mti_tmp_fid1;
1281 struct mdt_body *repbody;
1285 if (reqbody->valid != OBD_MD_FLID)
1288 LASSERT(!info->mti_cross_ref);
1290 /* Only got the fid of this obj by name */
1291 fid_zero(child_fid);
1292 rc = mdo_lookup(info->mti_env, next, lname, child_fid,
1295 /* XXX is raw_lookup possible as intent operation? */
1298 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1301 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1303 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1306 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1307 repbody->fid1 = *child_fid;
1308 repbody->valid = OBD_MD_FLID;
1314 * UPDATE lock should be taken against parent, and be release before exit;
1315 * child_bits lock should be taken against child, and be returned back:
1316 * (1)normal request should release the child lock;
1317 * (2)intent request will grant the lock to client.
1319 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
1320 struct mdt_lock_handle *lhc,
1322 struct ldlm_reply *ldlm_rep)
1324 struct ptlrpc_request *req = mdt_info_req(info);
1325 struct mdt_body *reqbody = NULL;
1326 struct mdt_object *parent = info->mti_object;
1327 struct mdt_object *child;
1328 struct md_object *next = mdt_object_child(parent);
1329 struct lu_fid *child_fid = &info->mti_tmp_fid1;
1330 struct lu_name *lname = NULL;
1331 struct mdt_lock_handle *lhp = NULL;
1332 struct ldlm_lock *lock;
1339 is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
1340 LASSERT(ergo(is_resent,
1341 lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
1343 LASSERT(parent != NULL);
1345 if (info->mti_cross_ref) {
1346 /* Only getattr on the child. Parent is on another node. */
1347 mdt_set_disposition(info, ldlm_rep,
1348 DISP_LOOKUP_EXECD | DISP_LOOKUP_POS);
1350 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
1352 PFID(mdt_object_fid(child)), ldlm_rep);
1355 /* Do not take lock for resent case. */
1356 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1357 LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
1358 lhc->mlh_reg_lh.cookie);
1359 LASSERT(fid_res_name_eq(mdt_object_fid(child),
1360 &lock->l_resource->lr_name));
1361 LDLM_LOCK_PUT(lock);
1363 mdt_lock_handle_init(lhc);
1364 mdt_lock_reg_init(lhc, LCK_PR);
1367 * Object's name is on another MDS, no lookup or layout
1368 * lock is needed here but update lock is.
1370 child_bits &= ~(MDS_INODELOCK_LOOKUP |
1371 MDS_INODELOCK_LAYOUT);
1372 child_bits |= MDS_INODELOCK_PERM | MDS_INODELOCK_UPDATE;
1374 rc = mdt_object_lock(info, child, lhc, child_bits,
1380 /* Finally, we can get attr for child. */
1381 if (!mdt_object_exists(child)) {
1382 LU_OBJECT_DEBUG(D_INFO, info->mti_env,
1384 "remote object doesn't exist.\n");
1385 mdt_object_unlock(info, child, lhc, 1);
1389 mdt_set_capainfo(info, 0, mdt_object_fid(child), BYPASS_CAPA);
1390 rc = mdt_getattr_internal(info, child, 0);
1391 if (unlikely(rc != 0))
1392 mdt_object_unlock(info, child, lhc, 1);
1397 lname = &info->mti_name;
1398 mdt_name_unpack(info->mti_pill, &RMF_NAME, lname, MNF_FIX_ANON);
1400 if (lu_name_is_valid(lname)) {
1401 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DNAME", "
1402 "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
1403 PNAME(lname), ldlm_rep);
1405 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1406 if (unlikely(reqbody == NULL))
1407 RETURN(err_serious(-EPROTO));
1409 *child_fid = reqbody->fid2;
1411 if (unlikely(!fid_is_sane(child_fid)))
1412 RETURN(err_serious(-EINVAL));
1414 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
1416 PFID(mdt_object_fid(parent)),
1417 PFID(&reqbody->fid2), ldlm_rep);
1420 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
1422 if (unlikely(!mdt_object_exists(parent)) && lu_name_is_valid(lname)) {
1423 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1425 "Parent doesn't exist!\n");
1429 if (mdt_object_remote(parent)) {
1430 CERROR("%s: parent "DFID" is on remote target\n",
1431 mdt_obd_name(info->mti_mdt),
1432 PFID(mdt_object_fid(parent)));
1436 if (lu_name_is_valid(lname)) {
1437 rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
1444 /* step 1: lock parent only if parent is a directory */
1445 if (S_ISDIR(lu_object_attr(&parent->mot_obj))) {
1446 lhp = &info->mti_lh[MDT_LH_PARENT];
1447 mdt_lock_pdo_init(lhp, LCK_PR, lname);
1448 rc = mdt_object_lock(info, parent, lhp,
1449 MDS_INODELOCK_UPDATE,
1451 if (unlikely(rc != 0))
1455 /* step 2: lookup child's fid by name */
1456 fid_zero(child_fid);
1457 rc = mdo_lookup(info->mti_env, next, lname, child_fid,
1460 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1463 GOTO(out_parent, rc);
1466 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1469 *step 3: find the child object by fid & lock it.
1470 * regardless if it is local or remote.
1472 *Note: LU-3240 (commit 762f2114d282a98ebfa4dbbeea9298a8088ad24e)
1473 * set parent dir fid the same as child fid in getattr by fid case
1474 * we should not lu_object_find() the object again, could lead
1475 * to hung if there is a concurrent unlink destroyed the object.
1477 if (lu_fid_eq(mdt_object_fid(parent), child_fid)) {
1478 mdt_object_get(info->mti_env, parent);
1481 child = mdt_object_find(info->mti_env, info->mti_mdt,
1485 if (unlikely(IS_ERR(child)))
1486 GOTO(out_parent, rc = PTR_ERR(child));
1488 /* Do not take lock for resent case. */
1489 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1490 LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
1491 lhc->mlh_reg_lh.cookie);
1493 if (!fid_res_name_eq(mdt_object_fid(child),
1494 &lock->l_resource->lr_name)) {
1495 LASSERTF(fid_res_name_eq(mdt_object_fid(parent),
1496 &lock->l_resource->lr_name),
1497 "Lock res_id: "DLDLMRES", fid: "DFID"\n",
1498 PLDLMRES(lock->l_resource),
1499 PFID(mdt_object_fid(parent)));
1500 CWARN("Although resent, but still not get child lock"
1501 "parent:"DFID" child:"DFID"\n",
1502 PFID(mdt_object_fid(parent)),
1503 PFID(mdt_object_fid(child)));
1504 lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
1505 LDLM_LOCK_PUT(lock);
1508 LDLM_LOCK_PUT(lock);
1511 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout*2);
1512 mdt_lock_handle_init(lhc);
1513 mdt_lock_reg_init(lhc, LCK_PR);
1516 if (!mdt_object_exists(child)) {
1517 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1519 "Object doesn't exist!\n");
1520 GOTO(out_child, rc = -ENOENT);
1523 if (!(child_bits & MDS_INODELOCK_UPDATE) &&
1524 mdt_object_exists(child) && !mdt_object_remote(child)) {
1525 struct md_attr *ma = &info->mti_attr;
1528 ma->ma_need = MA_INODE;
1529 rc = mdt_attr_get_complex(info, child, ma);
1530 if (unlikely(rc != 0))
1531 GOTO(out_child, rc);
1533 /* If the file has not been changed for some time, we
1534 * return not only a LOOKUP lock, but also an UPDATE
1535 * lock and this might save us RPC on later STAT. For
1536 * directories, it also let negative dentry cache start
1537 * working for this dir. */
1538 if (ma->ma_valid & MA_INODE &&
1539 ma->ma_attr.la_valid & LA_CTIME &&
1540 info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
1541 ma->ma_attr.la_ctime < cfs_time_current_sec())
1542 child_bits |= MDS_INODELOCK_UPDATE;
1545 /* layout lock must be granted in a best-effort way
1546 * for IT operations */
1547 LASSERT(!(child_bits & MDS_INODELOCK_LAYOUT));
1548 if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) &&
1549 exp_connect_layout(info->mti_exp) &&
1550 S_ISREG(lu_object_attr(&child->mot_obj)) &&
1551 !mdt_object_remote(child) && ldlm_rep != NULL) {
1552 /* try to grant layout lock for regular file. */
1558 child_bits |= MDS_INODELOCK_LAYOUT;
1559 /* try layout lock, it may fail to be granted due to
1560 * contention at LOOKUP or UPDATE */
1561 if (!mdt_object_lock_try(info, child, lhc, child_bits,
1563 child_bits &= ~MDS_INODELOCK_LAYOUT;
1564 LASSERT(child_bits != 0);
1565 rc = mdt_object_lock(info, child, lhc,
1566 child_bits, MDT_CROSS_LOCK);
1571 /* Do not enqueue the UPDATE lock from MDT(cross-MDT),
1572 * client will enqueue the lock to the remote MDT */
1573 if (mdt_object_remote(child))
1574 child_bits &= ~MDS_INODELOCK_UPDATE;
1575 rc = mdt_object_lock(info, child, lhc, child_bits,
1578 if (unlikely(rc != 0))
1579 GOTO(out_child, rc);
1582 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1583 /* Get MA_SOM attributes if update lock is given. */
1585 lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_UPDATE &&
1586 S_ISREG(lu_object_attr(&mdt_object_child(child)->mo_lu)))
1589 /* finally, we can get attr for child. */
1590 mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
1591 rc = mdt_getattr_internal(info, child, ma_need);
1592 if (unlikely(rc != 0)) {
1593 mdt_object_unlock(info, child, lhc, 1);
1595 /* Debugging code. */
1596 LDLM_DEBUG(lock, "Returning lock to client");
1597 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1598 &lock->l_resource->lr_name),
1599 "Lock res_id: "DLDLMRES", fid: "DFID"\n",
1600 PLDLMRES(lock->l_resource),
1601 PFID(mdt_object_fid(child)));
1602 if (mdt_object_exists(child) && !mdt_object_remote(child))
1603 mdt_pack_size2body(info, child);
1606 LDLM_LOCK_PUT(lock);
1610 mdt_object_put(info->mti_env, child);
1613 mdt_object_unlock(info, parent, lhp, 1);
1617 /* normal handler: should release the child lock */
1618 static int mdt_getattr_name(struct tgt_session_info *tsi)
1620 struct mdt_thread_info *info = tsi2mdt_info(tsi);
1621 struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1622 struct mdt_body *reqbody;
1623 struct mdt_body *repbody;
1627 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1628 LASSERT(reqbody != NULL);
1629 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1630 LASSERT(repbody != NULL);
1632 info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
1633 repbody->eadatasize = 0;
1634 repbody->aclsize = 0;
1636 rc = mdt_init_ucred(info, reqbody);
1638 GOTO(out_shrink, rc);
1640 rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1641 if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1642 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1643 lhc->mlh_reg_lh.cookie = 0;
1645 mdt_exit_ucred(info);
1648 mdt_client_compatibility(info);
1649 rc2 = mdt_fix_reply(info);
1652 mdt_thread_info_fini(info);
1656 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1657 void *karg, void *uarg);
1659 static int mdt_set_info(struct tgt_session_info *tsi)
1661 struct ptlrpc_request *req = tgt_ses_req(tsi);
1664 int keylen, vallen, rc = 0;
1668 key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
1670 DEBUG_REQ(D_HA, req, "no set_info key");
1671 RETURN(err_serious(-EFAULT));
1674 keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
1677 val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
1679 DEBUG_REQ(D_HA, req, "no set_info val");
1680 RETURN(err_serious(-EFAULT));
1683 vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
1686 /* Swab any part of val you need to here */
1687 if (KEY_IS(KEY_READ_ONLY)) {
1688 spin_lock(&req->rq_export->exp_lock);
1690 *exp_connect_flags_ptr(req->rq_export) |=
1693 *exp_connect_flags_ptr(req->rq_export) &=
1694 ~OBD_CONNECT_RDONLY;
1695 spin_unlock(&req->rq_export->exp_lock);
1696 } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1697 struct changelog_setinfo *cs = val;
1699 if (vallen != sizeof(*cs)) {
1700 CERROR("%s: bad changelog_clear setinfo size %d\n",
1701 tgt_name(tsi->tsi_tgt), vallen);
1704 if (ptlrpc_req_need_swab(req)) {
1705 __swab64s(&cs->cs_recno);
1706 __swab32s(&cs->cs_id);
1709 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, req->rq_export,
1717 static int mdt_readpage(struct tgt_session_info *tsi)
1719 struct mdt_thread_info *info = mdt_th_info(tsi->tsi_env);
1720 struct mdt_object *object = mdt_obj(tsi->tsi_corpus);
1721 struct lu_rdpg *rdpg = &info->mti_u.rdpg.mti_rdpg;
1722 const struct mdt_body *reqbody = tsi->tsi_mdt_body;
1723 struct mdt_body *repbody;
1729 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
1730 RETURN(err_serious(-ENOMEM));
1732 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_MDT_BODY);
1733 if (repbody == NULL || reqbody == NULL)
1734 RETURN(err_serious(-EFAULT));
1737 * prepare @rdpg before calling lower layers and transfer itself. Here
1738 * reqbody->size contains offset of where to start to read and
1739 * reqbody->nlink contains number bytes to read.
1741 rdpg->rp_hash = reqbody->size;
1742 if (rdpg->rp_hash != reqbody->size) {
1743 CERROR("Invalid hash: "LPX64" != "LPX64"\n",
1744 rdpg->rp_hash, reqbody->size);
1748 rdpg->rp_attrs = reqbody->mode;
1749 if (exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_64BITHASH)
1750 rdpg->rp_attrs |= LUDA_64BITHASH;
1751 rdpg->rp_count = min_t(unsigned int, reqbody->nlink,
1752 exp_max_brw_size(tsi->tsi_exp));
1753 rdpg->rp_npages = (rdpg->rp_count + PAGE_CACHE_SIZE - 1) >>
1755 OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1756 if (rdpg->rp_pages == NULL)
1759 for (i = 0; i < rdpg->rp_npages; ++i) {
1760 rdpg->rp_pages[i] = alloc_page(GFP_IOFS);
1761 if (rdpg->rp_pages[i] == NULL)
1762 GOTO(free_rdpg, rc = -ENOMEM);
1765 /* call lower layers to fill allocated pages with directory data */
1766 rc = mo_readpage(tsi->tsi_env, mdt_object_child(object), rdpg);
1768 GOTO(free_rdpg, rc);
1770 /* send pages to client */
1771 rc = tgt_sendpage(tsi, rdpg, rc);
1776 for (i = 0; i < rdpg->rp_npages; i++)
1777 if (rdpg->rp_pages[i] != NULL)
1778 __free_page(rdpg->rp_pages[i]);
1779 OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1781 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1787 static int mdt_reint_internal(struct mdt_thread_info *info,
1788 struct mdt_lock_handle *lhc,
1791 struct req_capsule *pill = info->mti_pill;
1792 struct mdt_body *repbody;
1797 rc = mdt_reint_unpack(info, op);
1799 CERROR("Can't unpack reint, rc %d\n", rc);
1800 RETURN(err_serious(rc));
1803 /* for replay (no_create) lmm is not needed, client has it already */
1804 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1805 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1808 /* llog cookies are always 0, the field is kept for compatibility */
1809 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1810 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);
1812 rc = req_capsule_server_pack(pill);
1814 CERROR("Can't pack response, rc %d\n", rc);
1815 RETURN(err_serious(rc));
1818 if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
1819 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1821 repbody->eadatasize = 0;
1822 repbody->aclsize = 0;
1825 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
1827 /* for replay no cookkie / lmm need, because client have this already */
1828 if (info->mti_spec.no_create)
1829 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1830 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
1832 rc = mdt_init_ucred_reint(info);
1834 GOTO(out_shrink, rc);
1836 rc = mdt_fix_attr_ucred(info, op);
1838 GOTO(out_ucred, rc = err_serious(rc));
1840 if (mdt_check_resent(info, mdt_reconstruct, lhc)) {
1841 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt.");
1842 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
1843 GOTO(out_ucred, rc);
1845 rc = mdt_reint_rec(info, lhc);
1848 mdt_exit_ucred(info);
1850 mdt_client_compatibility(info);
1851 rc2 = mdt_fix_reply(info);
1857 static long mdt_reint_opcode(struct ptlrpc_request *req,
1858 const struct req_format **fmt)
1860 struct mdt_device *mdt;
1861 struct mdt_rec_reint *rec;
1864 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
1866 opc = rec->rr_opcode;
1867 DEBUG_REQ(D_INODE, req, "reint opt = %ld", opc);
1868 if (opc < REINT_MAX && fmt[opc] != NULL)
1869 req_capsule_extend(&req->rq_pill, fmt[opc]);
1871 mdt = mdt_exp2dev(req->rq_export);
1872 CERROR("%s: Unsupported opcode '%ld' from client '%s':"
1873 " rc = %d\n", req->rq_export->exp_obd->obd_name,
1874 opc, mdt->mdt_ldlm_client->cli_name, -EFAULT);
1875 opc = err_serious(-EFAULT);
1878 opc = err_serious(-EFAULT);
1883 static int mdt_reint(struct tgt_session_info *tsi)
1887 static const struct req_format *reint_fmts[REINT_MAX] = {
1888 [REINT_SETATTR] = &RQF_MDS_REINT_SETATTR,
1889 [REINT_CREATE] = &RQF_MDS_REINT_CREATE,
1890 [REINT_LINK] = &RQF_MDS_REINT_LINK,
1891 [REINT_UNLINK] = &RQF_MDS_REINT_UNLINK,
1892 [REINT_RENAME] = &RQF_MDS_REINT_RENAME,
1893 [REINT_OPEN] = &RQF_MDS_REINT_OPEN,
1894 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR,
1895 [REINT_RMENTRY] = &RQF_MDS_REINT_UNLINK,
1896 [REINT_MIGRATE] = &RQF_MDS_REINT_RENAME
1901 opc = mdt_reint_opcode(tgt_ses_req(tsi), reint_fmts);
1903 struct mdt_thread_info *info = tsi2mdt_info(tsi);
1905 * No lock possible here from client to pass it to reint code
1908 rc = mdt_reint_internal(info, NULL, opc);
1909 mdt_thread_info_fini(info);
1914 tsi->tsi_reply_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
1918 /* this should sync the whole device */
1919 static int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
1921 struct dt_device *dt = mdt->mdt_bottom;
1925 rc = dt->dd_ops->dt_sync(env, dt);
1929 /* this should sync this object */
1930 static int mdt_object_sync(struct mdt_thread_info *info)
1932 struct md_object *next;
1936 if (!mdt_object_exists(info->mti_object)) {
1937 CWARN("Non existing object "DFID"!\n",
1938 PFID(mdt_object_fid(info->mti_object)));
1941 next = mdt_object_child(info->mti_object);
1942 rc = mo_object_sync(info->mti_env, next);
1947 static int mdt_sync(struct tgt_session_info *tsi)
1949 struct ptlrpc_request *req = tgt_ses_req(tsi);
1950 struct req_capsule *pill = tsi->tsi_pill;
1951 struct mdt_body *body;
1956 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
1957 RETURN(err_serious(-ENOMEM));
1959 if (fid_seq(&tsi->tsi_mdt_body->fid1) == 0) {
1960 rc = mdt_device_sync(tsi->tsi_env, mdt_exp2dev(tsi->tsi_exp));
1962 struct mdt_thread_info *info = tsi2mdt_info(tsi);
1964 /* sync an object */
1965 rc = mdt_object_sync(info);
1967 const struct lu_fid *fid;
1968 struct lu_attr *la = &info->mti_attr.ma_attr;
1970 info->mti_attr.ma_need = MA_INODE;
1971 info->mti_attr.ma_valid = 0;
1972 rc = mdt_attr_get_complex(info, info->mti_object,
1975 body = req_capsule_server_get(pill,
1977 fid = mdt_object_fid(info->mti_object);
1978 mdt_pack_attr2body(info, body, la, fid);
1981 mdt_thread_info_fini(info);
1984 mdt_counter_incr(req, LPROC_MDT_SYNC);
1990 * Handle quota control requests to consult current usage/limit, but also
1991 * to configure quota enforcement
1993 static int mdt_quotactl(struct tgt_session_info *tsi)
1995 struct obd_export *exp = tsi->tsi_exp;
1996 struct req_capsule *pill = tsi->tsi_pill;
1997 struct obd_quotactl *oqctl, *repoqc;
1999 struct mdt_device *mdt = mdt_exp2dev(exp);
2000 struct lu_device *qmt = mdt->mdt_qmt_dev;
2003 oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
2005 RETURN(err_serious(-EPROTO));
2007 rc = req_capsule_server_pack(pill);
2009 RETURN(err_serious(rc));
2011 switch (oqctl->qc_cmd) {
2013 case LUSTRE_Q_INVALIDATE:
2014 case LUSTRE_Q_FINVALIDATE:
2018 /* deprecated, not used any more */
2019 RETURN(-EOPNOTSUPP);
2020 /* master quotactl */
2026 RETURN(-EOPNOTSUPP);
2027 /* slave quotactl */
2032 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2036 /* map uid/gid for remote client */
2038 if (exp_connect_rmtclient(exp)) {
2039 struct lustre_idmap_table *idmap;
2041 idmap = exp->exp_mdt_data.med_idmap;
2043 if (unlikely(oqctl->qc_cmd != Q_GETQUOTA &&
2044 oqctl->qc_cmd != Q_GETINFO))
2047 if (oqctl->qc_type == USRQUOTA)
2048 id = lustre_idmap_lookup_uid(NULL, idmap, 0,
2050 else if (oqctl->qc_type == GRPQUOTA)
2051 id = lustre_idmap_lookup_gid(NULL, idmap, 0,
2056 if (id == CFS_IDMAP_NOTFOUND) {
2057 CDEBUG(D_QUOTA, "no mapping for id %u\n", oqctl->qc_id);
2062 repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
2064 RETURN(err_serious(-EFAULT));
2066 if (oqctl->qc_id != id)
2067 swap(oqctl->qc_id, id);
2069 switch (oqctl->qc_cmd) {
2075 /* forward quotactl request to QMT */
2076 rc = qmt_hdls.qmth_quotactl(tsi->tsi_env, qmt, oqctl);
2081 /* slave quotactl */
2082 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom,
2087 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2091 if (oqctl->qc_id != id)
2092 swap(oqctl->qc_id, id);
2098 /** clone llog ctxt from child (mdd)
2099 * This allows remote llog (replicator) access.
2100 * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
2101 * context was originally set up, or we can handle them directly.
2102 * I choose the latter, but that means I need any llog
2103 * contexts set up by child to be accessable by the mdt. So we clone the
2104 * context into our context list here.
2106 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
2109 struct md_device *next = mdt->mdt_child;
2110 struct llog_ctxt *ctxt;
2113 if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
2116 rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
2117 if (rc || ctxt == NULL) {
2121 rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
2123 CERROR("Can't set mdt ctxt %d\n", rc);
2128 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
2129 struct mdt_device *mdt, int idx)
2131 struct llog_ctxt *ctxt;
2133 ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
2136 /* Put once for the get we just did, and once for the clone */
2137 llog_ctxt_put(ctxt);
2138 llog_ctxt_put(ctxt);
2143 * sec context handlers
2145 static int mdt_sec_ctx_handle(struct tgt_session_info *tsi)
2149 rc = mdt_handle_idmap(tsi);
2151 struct ptlrpc_request *req = tgt_ses_req(tsi);
2154 opc = lustre_msg_get_opc(req->rq_reqmsg);
2155 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT)
2156 sptlrpc_svc_ctx_invalidate(req);
2159 CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
2165 * quota request handlers
2167 static int mdt_quota_dqacq(struct tgt_session_info *tsi)
2169 struct mdt_device *mdt = mdt_exp2dev(tsi->tsi_exp);
2170 struct lu_device *qmt = mdt->mdt_qmt_dev;
2175 RETURN(err_serious(-EOPNOTSUPP));
2177 rc = qmt_hdls.qmth_dqacq(tsi->tsi_env, qmt, tgt_ses_req(tsi));
2181 struct mdt_object *mdt_object_new(const struct lu_env *env,
2182 struct mdt_device *d,
2183 const struct lu_fid *f)
2185 struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2186 struct lu_object *o;
2187 struct mdt_object *m;
2190 CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
2191 o = lu_object_find(env, &d->mdt_lu_dev, f, &conf);
2192 if (unlikely(IS_ERR(o)))
2193 m = (struct mdt_object *)o;
2199 struct mdt_object *mdt_object_find(const struct lu_env *env,
2200 struct mdt_device *d,
2201 const struct lu_fid *f)
2203 struct lu_object *o;
2204 struct mdt_object *m;
2207 CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2208 o = lu_object_find(env, &d->mdt_lu_dev, f, NULL);
2209 if (unlikely(IS_ERR(o)))
2210 m = (struct mdt_object *)o;
2218 * Asyncronous commit for mdt device.
2220 * Pass asynchonous commit call down the MDS stack.
2222 * \param env environment
2223 * \param mdt the mdt device
2225 static void mdt_device_commit_async(const struct lu_env *env,
2226 struct mdt_device *mdt)
2228 struct dt_device *dt = mdt->mdt_bottom;
2231 rc = dt->dd_ops->dt_commit_async(env, dt);
2232 if (unlikely(rc != 0))
2233 CWARN("async commit start failed with rc = %d", rc);
2237 * Mark the lock as "synchonous".
2239 * Mark the lock to deffer transaction commit to the unlock time.
2241 * \param lock the lock to mark as "synchonous"
2243 * \see mdt_is_lock_sync
2244 * \see mdt_save_lock
2246 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2248 lock->l_ast_data = (void*)1;
2252 * Check whehter the lock "synchonous" or not.
2254 * \param lock the lock to check
2255 * \retval 1 the lock is "synchonous"
2256 * \retval 0 the lock isn't "synchronous"
2258 * \see mdt_set_lock_sync
2259 * \see mdt_save_lock
2261 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2263 return lock->l_ast_data != NULL;
2267 * Blocking AST for mdt locks.
2269 * Starts transaction commit if in case of COS lock conflict or
2270 * deffers such a commit to the mdt_save_lock.
2272 * \param lock the lock which blocks a request or cancelling lock
2273 * \param desc unused
2274 * \param data unused
2275 * \param flag indicates whether this cancelling or blocking callback
2277 * \see ldlm_blocking_ast_nocheck
2279 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2280 void *data, int flag)
2282 struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2283 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2287 if (flag == LDLM_CB_CANCELING)
2289 lock_res_and_lock(lock);
2290 if (lock->l_blocking_ast != mdt_blocking_ast) {
2291 unlock_res_and_lock(lock);
2294 if (mdt_cos_is_enabled(mdt) &&
2295 lock->l_req_mode & (LCK_PW | LCK_EX) &&
2296 lock->l_blocking_lock != NULL &&
2297 lock->l_client_cookie != lock->l_blocking_lock->l_client_cookie) {
2298 mdt_set_lock_sync(lock);
2300 rc = ldlm_blocking_ast_nocheck(lock);
2302 /* There is no lock conflict if l_blocking_lock == NULL,
2303 * it indicates a blocking ast sent from ldlm_lock_decref_internal
2304 * when the last reference to a local lock was released */
2305 if (lock->l_req_mode == LCK_COS && lock->l_blocking_lock != NULL) {
2308 rc = lu_env_init(&env, LCT_LOCAL);
2309 if (unlikely(rc != 0))
2310 CWARN("lu_env initialization failed with rc = %d,"
2311 "cannot start asynchronous commit\n", rc);
2313 mdt_device_commit_async(&env, mdt);
2319 /* Used for cross-MDT lock */
2320 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2321 void *data, int flag)
2323 struct lustre_handle lockh;
2327 case LDLM_CB_BLOCKING:
2328 ldlm_lock2handle(lock, &lockh);
2329 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
2331 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
2335 case LDLM_CB_CANCELING:
2336 LDLM_DEBUG(lock, "Revoke remote lock\n");
2344 int mdt_remote_object_lock(struct mdt_thread_info *mti,
2345 struct mdt_object *o, const struct lu_fid *fid,
2346 struct lustre_handle *lh, ldlm_mode_t mode,
2349 struct ldlm_enqueue_info *einfo = &mti->mti_einfo;
2350 ldlm_policy_data_t *policy = &mti->mti_policy;
2351 struct ldlm_res_id *res_id = &mti->mti_res_id;
2355 LASSERT(mdt_object_remote(o));
2357 LASSERT(ibits == MDS_INODELOCK_UPDATE);
2359 fid_build_reg_res_name(fid, res_id);
2361 memset(einfo, 0, sizeof(*einfo));
2362 einfo->ei_type = LDLM_IBITS;
2363 einfo->ei_mode = mode;
2364 einfo->ei_cb_bl = mdt_remote_blocking_ast;
2365 einfo->ei_cb_cp = ldlm_completion_ast;
2366 einfo->ei_enq_slave = 0;
2367 einfo->ei_res_id = res_id;
2369 memset(policy, 0, sizeof(*policy));
2370 policy->l_inodebits.bits = ibits;
2372 rc = mo_object_lock(mti->mti_env, mdt_object_child(o), lh, einfo,
2377 static int mdt_object_local_lock(struct mdt_thread_info *info,
2378 struct mdt_object *o,
2379 struct mdt_lock_handle *lh, __u64 ibits,
2380 bool nonblock, int locality)
2382 struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
2383 ldlm_policy_data_t *policy = &info->mti_policy;
2384 struct ldlm_res_id *res_id = &info->mti_res_id;
2389 LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2390 LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2391 LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
2392 LASSERT(lh->mlh_type != MDT_NUL_LOCK);
2394 /* Only enqueue LOOKUP lock for remote object */
2395 if (mdt_object_remote(o))
2396 LASSERT(ibits == MDS_INODELOCK_LOOKUP);
2398 if (lh->mlh_type == MDT_PDO_LOCK) {
2399 /* check for exists after object is locked */
2400 if (mdt_object_exists(o) == 0) {
2401 /* Non-existent object shouldn't have PDO lock */
2404 /* Non-dir object shouldn't have PDO lock */
2405 if (!S_ISDIR(lu_object_attr(&o->mot_obj)))
2410 memset(policy, 0, sizeof(*policy));
2411 fid_build_reg_res_name(mdt_object_fid(o), res_id);
2413 dlmflags = LDLM_FL_ATOMIC_CB;
2415 dlmflags |= LDLM_FL_BLOCK_NOWAIT;
2418 * Take PDO lock on whole directory and build correct @res_id for lock
2419 * on part of directory.
2421 if (lh->mlh_pdo_hash != 0) {
2422 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
2423 mdt_lock_pdo_mode(info, o, lh);
2424 if (lh->mlh_pdo_mode != LCK_NL) {
2426 * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
2427 * is never going to be sent to client and we do not
2428 * want it slowed down due to possible cancels.
2430 policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
2431 rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
2432 policy, res_id, dlmflags,
2433 info->mti_exp == NULL ? NULL :
2434 &info->mti_exp->exp_handle.h_cookie);
2440 * Finish res_id initializing by name hash marking part of
2441 * directory which is taking modification.
2443 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
2446 policy->l_inodebits.bits = ibits;
2449 * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
2450 * going to be sent to client. If it is - mdt_intent_policy() path will
2451 * fix it up and turn FL_LOCAL flag off.
2453 rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
2454 res_id, LDLM_FL_LOCAL_ONLY | dlmflags,
2455 info->mti_exp == NULL ? NULL :
2456 &info->mti_exp->exp_handle.h_cookie);
2458 mdt_object_unlock(info, o, lh, 1);
2459 else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
2460 lh->mlh_pdo_hash != 0 &&
2461 (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX)) {
2462 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
2469 mdt_object_lock_internal(struct mdt_thread_info *info, struct mdt_object *o,
2470 struct mdt_lock_handle *lh, __u64 ibits,
2471 bool nonblock, int locality)
2476 if (!mdt_object_remote(o))
2477 return mdt_object_local_lock(info, o, lh, ibits, nonblock,
2480 if (locality == MDT_LOCAL_LOCK) {
2481 CERROR("%s: try to get local lock for remote object"
2482 DFID".\n", mdt_obd_name(info->mti_mdt),
2483 PFID(mdt_object_fid(o)));
2487 /* XXX do not support PERM/LAYOUT/XATTR lock for remote object yet */
2488 ibits &= ~(MDS_INODELOCK_PERM | MDS_INODELOCK_LAYOUT |
2489 MDS_INODELOCK_XATTR);
2490 if (ibits & MDS_INODELOCK_UPDATE) {
2491 /* Sigh, PDO needs to enqueue 2 locks right now, but
2492 * enqueue RPC can only request 1 lock, to avoid extra
2493 * RPC, so it will instead enqueue EX lock for remote
2494 * object anyway XXX*/
2495 if (lh->mlh_type == MDT_PDO_LOCK &&
2496 lh->mlh_pdo_hash != 0) {
2497 CDEBUG(D_INFO, "%s: "DFID" convert PDO lock to"
2498 "EX lock.\n", mdt_obd_name(info->mti_mdt),
2499 PFID(mdt_object_fid(o)));
2500 lh->mlh_pdo_hash = 0;
2501 lh->mlh_rreg_mode = LCK_EX;
2502 lh->mlh_type = MDT_REG_LOCK;
2504 rc = mdt_remote_object_lock(info, o, mdt_object_fid(o),
2507 MDS_INODELOCK_UPDATE);
2512 /* Only enqueue LOOKUP lock for remote object */
2513 if (ibits & MDS_INODELOCK_LOOKUP) {
2514 rc = mdt_object_local_lock(info, o, lh,
2515 MDS_INODELOCK_LOOKUP,
2516 nonblock, locality);
2524 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2525 struct mdt_lock_handle *lh, __u64 ibits, int locality)
2527 return mdt_object_lock_internal(info, o, lh, ibits, false, locality);
2530 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o,
2531 struct mdt_lock_handle *lh, __u64 ibits, int locality)
2533 struct mdt_lock_handle tmp = *lh;
2536 rc = mdt_object_lock_internal(info, o, &tmp, ibits, true, locality);
2544 * Save a lock within request object.
2546 * Keep the lock referenced until whether client ACK or transaction
2547 * commit happens or release the lock immediately depending on input
2548 * parameters. If COS is ON, a write lock is converted to COS lock
2551 * \param info thead info object
2552 * \param h lock handle
2553 * \param mode lock mode
2554 * \param decref force immediate lock releasing
2557 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
2558 ldlm_mode_t mode, int decref)
2562 if (lustre_handle_is_used(h)) {
2563 if (decref || !info->mti_has_trans ||
2564 !(mode & (LCK_PW | LCK_EX))){
2565 mdt_fid_unlock(h, mode);
2567 struct mdt_device *mdt = info->mti_mdt;
2568 struct ldlm_lock *lock = ldlm_handle2lock(h);
2569 struct ptlrpc_request *req = mdt_info_req(info);
2572 LASSERTF(lock != NULL, "no lock for cookie "LPX64"\n",
2574 /* there is no request if mdt_object_unlock() is called
2575 * from mdt_export_cleanup()->mdt_add_dirty_flag() */
2576 if (likely(req != NULL)) {
2577 CDEBUG(D_HA, "request = %p reply state = %p"
2578 " transno = "LPD64"\n", req,
2579 req->rq_reply_state, req->rq_transno);
2580 if (mdt_cos_is_enabled(mdt)) {
2582 ldlm_lock_downgrade(lock, LCK_COS);
2585 ptlrpc_save_lock(req, h, mode, no_ack);
2587 ldlm_lock_decref(h, mode);
2589 if (mdt_is_lock_sync(lock)) {
2590 CDEBUG(D_HA, "found sync-lock,"
2591 " async commit started\n");
2592 mdt_device_commit_async(info->mti_env,
2595 LDLM_LOCK_PUT(lock);
2604 * Unlock mdt object.
2606 * Immeditely release the regular lock and the PDO lock or save the
2607 * lock in reqeuest and keep them referenced until client ACK or
2608 * transaction commit.
2610 * \param info thread info object
2611 * \param o mdt object
2612 * \param lh mdt lock handle referencing regular and PDO locks
2613 * \param decref force immediate lock releasing
2615 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
2616 struct mdt_lock_handle *lh, int decref)
2620 mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
2621 mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
2623 if (lustre_handle_is_used(&lh->mlh_rreg_lh))
2624 ldlm_lock_decref(&lh->mlh_rreg_lh, lh->mlh_rreg_mode);
2629 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
2630 const struct lu_fid *f,
2631 struct mdt_lock_handle *lh,
2634 struct mdt_object *o;
2636 o = mdt_object_find(info->mti_env, info->mti_mdt, f);
2640 rc = mdt_object_lock(info, o, lh, ibits,
2643 mdt_object_put(info->mti_env, o);
2650 void mdt_object_unlock_put(struct mdt_thread_info * info,
2651 struct mdt_object * o,
2652 struct mdt_lock_handle *lh,
2655 mdt_object_unlock(info, o, lh, decref);
2656 mdt_object_put(info->mti_env, o);
2660 * Generic code handling requests that have struct mdt_body passed in:
2662 * - extract mdt_body from request and save it in @info, if present;
2664 * - create lu_object, corresponding to the fid in mdt_body, and save it in
2667 * - if HABEO_CORPUS flag is set for this request type check whether object
2668 * actually exists on storage (lu_object_exists()).
2671 static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags)
2673 const struct mdt_body *body;
2674 struct mdt_object *obj;
2675 const struct lu_env *env;
2676 struct req_capsule *pill;
2680 env = info->mti_env;
2681 pill = info->mti_pill;
2683 body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
2687 if (!(body->valid & OBD_MD_FLID))
2690 if (!fid_is_sane(&body->fid1)) {
2691 CERROR("Invalid fid: "DFID"\n", PFID(&body->fid1));
2696 * Do not get size or any capa fields before we check that request
2697 * contains capa actually. There are some requests which do not, for
2698 * instance MDS_IS_SUBDIR.
2700 if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_CLIENT) &&
2701 req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
2702 mdt_set_capainfo(info, 0, &body->fid1,
2703 req_capsule_client_get(pill, &RMF_CAPA1));
2705 obj = mdt_object_find(env, info->mti_mdt, &body->fid1);
2707 if ((flags & HABEO_CORPUS) &&
2708 !mdt_object_exists(obj)) {
2709 mdt_object_put(env, obj);
2710 /* for capability renew ENOENT will be handled in
2712 if (body->valid & OBD_MD_FLOSSCAPA)
2717 info->mti_object = obj;
2726 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags)
2728 struct req_capsule *pill = info->mti_pill;
2732 if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
2733 rc = mdt_body_unpack(info, flags);
2737 if (rc == 0 && (flags & HABEO_REFERO)) {
2739 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2740 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2742 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2743 req_capsule_set_size(pill, &RMF_LOGCOOKIES,
2746 rc = req_capsule_server_pack(pill);
2751 static int mdt_init_capa_ctxt(const struct lu_env *env, struct mdt_device *m)
2753 struct md_device *next = m->mdt_child;
2755 return next->md_ops->mdo_init_capa_ctxt(env, next,
2756 m->mdt_lut.lut_mds_capa,
2757 m->mdt_capa_timeout,
2762 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
2764 lh->mlh_type = MDT_NUL_LOCK;
2765 lh->mlh_reg_lh.cookie = 0ull;
2766 lh->mlh_reg_mode = LCK_MINMODE;
2767 lh->mlh_pdo_lh.cookie = 0ull;
2768 lh->mlh_pdo_mode = LCK_MINMODE;
2769 lh->mlh_rreg_lh.cookie = 0ull;
2770 lh->mlh_rreg_mode = LCK_MINMODE;
2773 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
2775 LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2776 LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2780 * Initialize fields of struct mdt_thread_info. Other fields are left in
2781 * uninitialized state, because it's too expensive to zero out whole
2782 * mdt_thread_info (> 1K) on each request arrival.
2784 void mdt_thread_info_init(struct ptlrpc_request *req,
2785 struct mdt_thread_info *info)
2789 info->mti_pill = &req->rq_pill;
2792 for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2793 mdt_lock_handle_init(&info->mti_lh[i]);
2795 /* mdt device: it can be NULL while CONNECT */
2796 if (req->rq_export) {
2797 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
2798 info->mti_exp = req->rq_export;
2800 info->mti_mdt = NULL;
2801 info->mti_env = req->rq_svc_thread->t_env;
2802 info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
2804 memset(&info->mti_attr, 0, sizeof(info->mti_attr));
2805 info->mti_big_buf = LU_BUF_NULL;
2806 info->mti_body = NULL;
2807 info->mti_object = NULL;
2808 info->mti_dlm_req = NULL;
2809 info->mti_has_trans = 0;
2810 info->mti_cross_ref = 0;
2811 info->mti_opdata = 0;
2812 info->mti_big_lmm_used = 0;
2814 info->mti_spec.no_create = 0;
2815 info->mti_spec.sp_rm_entry = 0;
2817 info->mti_spec.u.sp_ea.eadata = NULL;
2818 info->mti_spec.u.sp_ea.eadatalen = 0;
2821 void mdt_thread_info_fini(struct mdt_thread_info *info)
2825 if (info->mti_object != NULL) {
2826 mdt_object_put(info->mti_env, info->mti_object);
2827 info->mti_object = NULL;
2830 for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2831 mdt_lock_handle_fini(&info->mti_lh[i]);
2832 info->mti_env = NULL;
2833 info->mti_pill = NULL;
2834 info->mti_exp = NULL;
2836 if (unlikely(info->mti_big_buf.lb_buf != NULL))
2837 lu_buf_free(&info->mti_big_buf);
2840 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi)
2842 struct mdt_thread_info *mti;
2843 struct lustre_capa *lc;
2845 mti = mdt_th_info(tsi->tsi_env);
2846 LASSERT(mti != NULL);
2848 mdt_thread_info_init(tgt_ses_req(tsi), mti);
2849 if (tsi->tsi_corpus != NULL) {
2850 struct req_capsule *pill = tsi->tsi_pill;
2852 mti->mti_object = mdt_obj(tsi->tsi_corpus);
2853 lu_object_get(tsi->tsi_corpus);
2856 * XXX: must be part of tgt_mdt_body_unpack but moved here
2857 * due to mdt_set_capainfo().
2859 if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_CLIENT) &&
2860 req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT) > 0) {
2861 lc = req_capsule_client_get(pill, &RMF_CAPA1);
2862 mdt_set_capainfo(mti, 0, &tsi->tsi_mdt_body->fid1, lc);
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, struct ldlm_lock **lockp,
3000 struct ldlm_lock *new_lock, struct mdt_lock_handle *lh,
3003 struct ptlrpc_request *req = mdt_info_req(info);
3004 struct ldlm_lock *lock = *lockp;
3007 * Get new lock only for cases when possible resent did not find any
3010 if (new_lock == NULL)
3011 new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
3013 if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
3014 lh->mlh_reg_lh.cookie = 0;
3018 LASSERTF(new_lock != NULL,
3019 "lockh "LPX64"\n", lh->mlh_reg_lh.cookie);
3022 * If we've already given this lock to a client once, then we should
3023 * have no readers or writers. Otherwise, we should have one reader
3024 * _or_ writer ref (which will be zeroed below) before returning the
3027 if (new_lock->l_export == req->rq_export) {
3028 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
3030 LASSERT(new_lock->l_export == NULL);
3031 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
3036 if (new_lock->l_export == req->rq_export) {
3038 * Already gave this to the client, which means that we
3039 * reconstructed a reply.
3041 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
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 ldlm_lock **old_lock,
3085 struct mdt_lock_handle *lh,
3086 enum mdt_it_code opcode)
3088 struct ptlrpc_request *req = mdt_info_req(info);
3089 struct obd_export *exp = req->rq_export;
3090 struct lustre_handle remote_hdl;
3091 struct ldlm_request *dlmreq;
3092 struct ldlm_lock *lock;
3094 if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
3097 dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
3098 remote_hdl = dlmreq->lock_handle[0];
3099 /* If the client does not require open lock, it does not need to
3100 * search lock in exp_lock_hash, since the server thread will
3101 * make sure the lock will be released, and the resend request
3102 * can always re-enqueue the lock */
3103 if ((opcode != MDT_IT_OPEN) || (opcode == MDT_IT_OPEN &&
3104 info->mti_spec.sp_cr_flags & MDS_OPEN_LOCK)) {
3105 /* In the function below, .hs_keycmp resolves to
3106 * ldlm_export_lock_keycmp() */
3107 /* coverity[overrun-buffer-val] */
3108 lock = cfs_hash_lookup(exp->exp_lock_hash, &remote_hdl);
3110 lock_res_and_lock(lock);
3111 if (lock != new_lock) {
3112 lh->mlh_reg_lh.cookie = lock->l_handle.h_cookie;
3113 lh->mlh_reg_mode = lock->l_granted_mode;
3115 LDLM_DEBUG(lock, "Restoring lock cookie");
3116 DEBUG_REQ(D_DLMTRACE, req,
3117 "restoring lock cookie "LPX64,
3118 lh->mlh_reg_lh.cookie);
3120 *old_lock = LDLM_LOCK_GET(lock);
3121 cfs_hash_put(exp->exp_lock_hash,
3123 unlock_res_and_lock(lock);
3126 cfs_hash_put(exp->exp_lock_hash, &lock->l_exp_hash);
3127 unlock_res_and_lock(lock);
3131 * If the xid matches, then we know this is a resent request, and allow
3132 * it. (It's probably an OPEN, for which we don't send a lock.
3134 if (req_xid_is_last(req))
3138 * This remote handle isn't enqueued, so we never received or processed
3139 * this request. Clear MSG_RESENT, because it can be handled like any
3140 * normal request now.
3142 lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
3144 DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle "LPX64,
3148 static int mdt_intent_getxattr(enum mdt_it_code opcode,
3149 struct mdt_thread_info *info,
3150 struct ldlm_lock **lockp,
3153 struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3154 struct ldlm_reply *ldlm_rep = NULL;
3158 * Initialize lhc->mlh_reg_lh either from a previously granted lock
3159 * (for the resend case) or a new lock. Below we will use it to
3160 * replace the original lock.
3162 mdt_intent_fixup_resent(info, *lockp, NULL, lhc, opcode);
3163 if (!lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3164 mdt_lock_reg_init(lhc, (*lockp)->l_req_mode);
3165 rc = mdt_object_lock(info, info->mti_object, lhc,
3166 MDS_INODELOCK_XATTR,
3172 grc = mdt_getxattr(info);
3174 rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
3176 if (mdt_info_req(info)->rq_repmsg != NULL)
3177 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3178 if (ldlm_rep == NULL)
3179 RETURN(err_serious(-EFAULT));
3181 ldlm_rep->lock_policy_res2 = grc;
3186 static int mdt_intent_getattr(enum mdt_it_code opcode,
3187 struct mdt_thread_info *info,
3188 struct ldlm_lock **lockp,
3191 struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3192 struct ldlm_lock *new_lock = NULL;
3194 struct ldlm_reply *ldlm_rep;
3195 struct mdt_body *reqbody;
3196 struct mdt_body *repbody;
3200 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
3203 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
3206 info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
3207 repbody->eadatasize = 0;
3208 repbody->aclsize = 0;
3212 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM;
3214 case MDT_IT_GETATTR:
3215 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
3219 CERROR("Unsupported intent (%d)\n", opcode);
3220 GOTO(out_shrink, rc = -EINVAL);
3223 rc = mdt_init_ucred(info, reqbody);
3225 GOTO(out_shrink, rc);
3227 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3228 mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
3230 /* Get lock from request for possible resent case. */
3231 mdt_intent_fixup_resent(info, *lockp, &new_lock, lhc, opcode);
3233 rc = mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
3234 ldlm_rep->lock_policy_res2 = clear_serious(rc);
3236 if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
3237 ldlm_rep->lock_policy_res2 = 0;
3238 if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
3239 ldlm_rep->lock_policy_res2) {
3240 lhc->mlh_reg_lh.cookie = 0ull;
3241 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
3244 rc = mdt_intent_lock_replace(info, lockp, new_lock, lhc, flags);
3247 mdt_exit_ucred(info);
3249 mdt_client_compatibility(info);
3250 rc2 = mdt_fix_reply(info);
3256 static int mdt_intent_layout(enum mdt_it_code opcode,
3257 struct mdt_thread_info *info,
3258 struct ldlm_lock **lockp,
3261 struct layout_intent *layout;
3263 struct mdt_object *obj = NULL;
3267 if (opcode != MDT_IT_LAYOUT) {
3268 CERROR("%s: Unknown intent (%d)\n", mdt_obd_name(info->mti_mdt),
3273 fid = &info->mti_tmp_fid2;
3274 fid_extract_from_res_name(fid, &(*lockp)->l_resource->lr_name);
3276 obj = mdt_object_find(info->mti_env, info->mti_mdt, fid);
3278 RETURN(PTR_ERR(obj));
3280 if (mdt_object_exists(obj) && !mdt_object_remote(obj)) {
3281 /* get the length of lsm */
3282 rc = mdt_attr_get_eabuf_size(info, obj);
3286 if (rc > info->mti_mdt->mdt_max_mdsize)
3287 info->mti_mdt->mdt_max_mdsize = rc;
3290 mdt_object_put(info->mti_env, obj);
3292 (*lockp)->l_lvb_type = LVB_T_LAYOUT;
3293 req_capsule_set_size(info->mti_pill, &RMF_DLM_LVB, RCL_SERVER, rc);
3294 rc = req_capsule_server_pack(info->mti_pill);
3298 layout = req_capsule_client_get(info->mti_pill, &RMF_LAYOUT_INTENT);
3299 LASSERT(layout != NULL);
3300 if (layout->li_opc == LAYOUT_INTENT_ACCESS)
3301 /* return to normal ldlm handling */
3304 CERROR("%s: Unsupported layout intent (%d)\n",
3305 mdt_obd_name(info->mti_mdt), layout->li_opc);
3309 static int mdt_intent_reint(enum mdt_it_code opcode,
3310 struct mdt_thread_info *info,
3311 struct ldlm_lock **lockp,
3314 struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3315 struct ldlm_reply *rep = NULL;
3319 static const struct req_format *intent_fmts[REINT_MAX] = {
3320 [REINT_CREATE] = &RQF_LDLM_INTENT_CREATE,
3321 [REINT_OPEN] = &RQF_LDLM_INTENT_OPEN
3326 opc = mdt_reint_opcode(mdt_info_req(info), intent_fmts);
3330 if (mdt_it_flavor[opcode].it_reint != opc) {
3331 CERROR("Reint code %ld doesn't match intent: %d\n",
3333 RETURN(err_serious(-EPROTO));
3336 /* Get lock from request for possible resent case. */
3337 mdt_intent_fixup_resent(info, *lockp, NULL, lhc, opcode);
3339 rc = mdt_reint_internal(info, lhc, opc);
3341 /* Check whether the reply has been packed successfully. */
3342 if (mdt_info_req(info)->rq_repmsg != NULL)
3343 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3345 RETURN(err_serious(-EFAULT));
3347 /* MDC expects this in any case */
3349 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
3351 /* the open lock or the lock for cross-ref object should be
3352 * returned to the client */
3353 if (rc == -EREMOTE || mdt_get_disposition(rep, DISP_OPEN_LOCK)) {
3354 LASSERT(lustre_handle_is_used(&lhc->mlh_reg_lh));
3355 rep->lock_policy_res2 = 0;
3356 rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
3360 rep->lock_policy_res2 = clear_serious(rc);
3362 if (rep->lock_policy_res2 == -ENOENT &&
3363 mdt_get_disposition(rep, DISP_LOOKUP_NEG))
3364 rep->lock_policy_res2 = 0;
3366 if (rc == -ENOTCONN || rc == -ENODEV ||
3367 rc == -EOVERFLOW) { /**< if VBR failure then return error */
3369 * If it is the disconnect error (ENODEV & ENOCONN), the error
3370 * will be returned by rq_status, and client at ptlrpc layer
3371 * will detect this, then disconnect, reconnect the import
3372 * immediately, instead of impacting the following the rpc.
3374 lhc->mlh_reg_lh.cookie = 0ull;
3378 * For other cases, the error will be returned by intent.
3379 * and client will retrieve the result from intent.
3382 * FIXME: when open lock is finished, that should be
3385 if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3386 LASSERTF(rc == 0, "Error occurred but lock handle "
3387 "is still in use, rc = %d\n", rc);
3388 rep->lock_policy_res2 = 0;
3389 rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
3392 lhc->mlh_reg_lh.cookie = 0ull;
3393 RETURN(ELDLM_LOCK_ABORTED);
3398 static int mdt_intent_code(long itcode)
3406 case IT_OPEN|IT_CREAT:
3413 rc = MDT_IT_READDIR;
3416 rc = MDT_IT_GETATTR;
3428 rc = MDT_IT_GETXATTR;
3433 case IT_QUOTA_DQACQ:
3438 CERROR("Unknown intent opcode: %ld\n", itcode);
3445 static int mdt_intent_opc(long itopc, struct mdt_thread_info *info,
3446 struct ldlm_lock **lockp, __u64 flags)
3448 struct req_capsule *pill;
3449 struct mdt_it_flavor *flv;
3454 opc = mdt_intent_code(itopc);
3458 pill = info->mti_pill;
3460 if (opc == MDT_IT_QUOTA) {
3461 struct lu_device *qmt = info->mti_mdt->mdt_qmt_dev;
3464 RETURN(-EOPNOTSUPP);
3466 (*lockp)->l_lvb_type = LVB_T_LQUOTA;
3467 /* pass the request to quota master */
3468 rc = qmt_hdls.qmth_intent_policy(info->mti_env, qmt,
3469 mdt_info_req(info), lockp,
3474 flv = &mdt_it_flavor[opc];
3475 if (flv->it_fmt != NULL)
3476 req_capsule_extend(pill, flv->it_fmt);
3478 rc = mdt_unpack_req_pack_rep(info, flv->it_flags);
3480 struct ptlrpc_request *req = mdt_info_req(info);
3481 if (flv->it_flags & MUTABOR &&
3482 exp_connect_flags(req->rq_export) & OBD_CONNECT_RDONLY)
3485 if (rc == 0 && flv->it_act != NULL) {
3486 struct ldlm_reply *rep;
3488 /* execute policy */
3489 rc = flv->it_act(opc, info, lockp, flags);
3491 /* Check whether the reply has been packed successfully. */
3492 if (mdt_info_req(info)->rq_repmsg != NULL) {
3493 rep = req_capsule_server_get(info->mti_pill,
3495 rep->lock_policy_res2 =
3496 ptlrpc_status_hton(rep->lock_policy_res2);
3504 static int mdt_intent_policy(struct ldlm_namespace *ns,
3505 struct ldlm_lock **lockp, void *req_cookie,
3506 ldlm_mode_t mode, __u64 flags, void *data)
3508 struct tgt_session_info *tsi;
3509 struct mdt_thread_info *info;
3510 struct ptlrpc_request *req = req_cookie;
3511 struct ldlm_intent *it;
3512 struct req_capsule *pill;
3517 LASSERT(req != NULL);
3519 tsi = tgt_ses_info(req->rq_svc_thread->t_env);
3521 info = tsi2mdt_info(tsi);
3522 LASSERT(info != NULL);
3523 pill = info->mti_pill;
3524 LASSERT(pill->rc_req == req);
3526 if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
3527 req_capsule_extend(pill, &RQF_LDLM_INTENT_BASIC);
3528 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
3530 rc = mdt_intent_opc(it->opc, info, lockp, flags);
3534 /* Lock without inodebits makes no sense and will oops
3535 * later in ldlm. Let's check it now to see if we have
3536 * ibits corrupted somewhere in mdt_intent_opc().
3537 * The case for client miss to set ibits has been
3538 * processed by others. */
3539 LASSERT(ergo(info->mti_dlm_req->lock_desc.l_resource.\
3540 lr_type == LDLM_IBITS,
3541 info->mti_dlm_req->lock_desc.\
3542 l_policy_data.l_inodebits.bits != 0));
3544 rc = err_serious(-EFAULT);
3546 /* No intent was provided */
3547 LASSERT(pill->rc_fmt == &RQF_LDLM_ENQUEUE);
3548 req_capsule_set_size(pill, &RMF_DLM_LVB, RCL_SERVER, 0);
3549 rc = req_capsule_server_pack(pill);
3551 rc = err_serious(rc);
3553 mdt_thread_info_fini(info);
3557 static void mdt_deregister_seq_exp(struct mdt_device *mdt)
3559 struct seq_server_site *ss = mdt_seq_site(mdt);
3561 if (ss->ss_node_id == 0)
3564 if (ss->ss_client_seq != NULL) {
3565 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
3566 ss->ss_client_seq->lcs_exp = NULL;
3569 if (ss->ss_server_fld != NULL) {
3570 lustre_deregister_lwp_item(&ss->ss_server_fld->lsf_control_exp);
3571 ss->ss_server_fld->lsf_control_exp = NULL;
3575 static void mdt_seq_fini_cli(struct mdt_device *mdt)
3577 struct seq_server_site *ss = mdt_seq_site(mdt);
3582 if (ss->ss_server_seq != NULL)
3583 seq_server_set_cli(NULL, ss->ss_server_seq, NULL);
3586 static int mdt_seq_fini(const struct lu_env *env, struct mdt_device *mdt)
3588 mdt_seq_fini_cli(mdt);
3589 mdt_deregister_seq_exp(mdt);
3591 return seq_site_fini(env, mdt_seq_site(mdt));
3595 * It will retrieve its FLDB entries from MDT0, and it only happens
3596 * when upgrading existent FS to 2.6 or when local FLDB is corrupted,
3597 * and it needs to refresh FLDB from the MDT0.
3599 static int mdt_register_lwp_callback(void *data)
3602 struct mdt_device *mdt = data;
3603 struct lu_server_fld *fld = mdt_seq_site(mdt)->ss_server_fld;
3607 LASSERT(mdt_seq_site(mdt)->ss_node_id != 0);
3609 if (!likely(fld->lsf_new))
3612 rc = lu_env_init(&env, LCT_MD_THREAD);
3614 CERROR("%s: cannot init env: rc = %d\n", mdt_obd_name(mdt), rc);
3618 rc = fld_update_from_controller(&env, fld);
3620 CERROR("%s: cannot update controller: rc = %d\n",
3621 mdt_obd_name(mdt), rc);
3629 static int mdt_register_seq_exp(struct mdt_device *mdt)
3631 struct seq_server_site *ss = mdt_seq_site(mdt);
3632 char *lwp_name = NULL;
3635 if (ss->ss_node_id == 0)
3638 OBD_ALLOC(lwp_name, MAX_OBD_NAME);
3639 if (lwp_name == NULL)
3640 GOTO(out_free, rc = -ENOMEM);
3642 rc = tgt_name2lwp_name(mdt_obd_name(mdt), lwp_name, MAX_OBD_NAME, 0);
3646 rc = lustre_register_lwp_item(lwp_name, &ss->ss_client_seq->lcs_exp,
3651 rc = lustre_register_lwp_item(lwp_name,
3652 &ss->ss_server_fld->lsf_control_exp,
3653 mdt_register_lwp_callback, mdt);
3655 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
3656 ss->ss_client_seq->lcs_exp = NULL;
3660 if (lwp_name != NULL)
3661 OBD_FREE(lwp_name, MAX_OBD_NAME);
3667 * Init client sequence manager which is used by local MDS to talk to sequence
3668 * controller on remote node.
3670 static int mdt_seq_init_cli(const struct lu_env *env, struct mdt_device *mdt)
3672 struct seq_server_site *ss = mdt_seq_site(mdt);
3677 /* check if this is adding the first MDC and controller is not yet
3679 OBD_ALLOC_PTR(ss->ss_client_seq);
3680 if (ss->ss_client_seq == NULL)
3683 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3684 if (prefix == NULL) {
3685 OBD_FREE_PTR(ss->ss_client_seq);
3686 ss->ss_client_seq = NULL;
3690 /* Note: seq_client_fini will be called in seq_site_fini */
3691 snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s", mdt_obd_name(mdt));
3692 rc = seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_METADATA,
3693 prefix, ss->ss_node_id == 0 ? ss->ss_control_seq :
3695 OBD_FREE(prefix, MAX_OBD_NAME + 5);
3697 OBD_FREE_PTR(ss->ss_client_seq);
3698 ss->ss_client_seq = NULL;
3702 rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq);
3707 static int mdt_seq_init(const struct lu_env *env, struct mdt_device *mdt)
3709 struct seq_server_site *ss;
3713 ss = mdt_seq_site(mdt);
3714 /* init sequence controller server(MDT0) */
3715 if (ss->ss_node_id == 0) {
3716 OBD_ALLOC_PTR(ss->ss_control_seq);
3717 if (ss->ss_control_seq == NULL)
3720 rc = seq_server_init(env, ss->ss_control_seq, mdt->mdt_bottom,
3721 mdt_obd_name(mdt), LUSTRE_SEQ_CONTROLLER,
3724 GOTO(out_seq_fini, rc);
3727 /* Init normal sequence server */
3728 OBD_ALLOC_PTR(ss->ss_server_seq);
3729 if (ss->ss_server_seq == NULL)
3730 GOTO(out_seq_fini, rc = -ENOMEM);
3732 rc = seq_server_init(env, ss->ss_server_seq, mdt->mdt_bottom,
3733 mdt_obd_name(mdt), LUSTRE_SEQ_SERVER, ss);
3735 GOTO(out_seq_fini, rc);
3737 /* init seq client for seq server to talk to seq controller(MDT0) */
3738 rc = mdt_seq_init_cli(env, mdt);
3740 GOTO(out_seq_fini, rc);
3742 if (ss->ss_node_id != 0)
3743 /* register controler export through lwp */
3744 rc = mdt_register_seq_exp(mdt);
3749 mdt_seq_fini(env, mdt);
3757 static int mdt_fld_fini(const struct lu_env *env,
3758 struct mdt_device *m)
3760 struct seq_server_site *ss = mdt_seq_site(m);
3763 if (ss && ss->ss_server_fld) {
3764 fld_server_fini(env, ss->ss_server_fld);
3765 OBD_FREE_PTR(ss->ss_server_fld);
3766 ss->ss_server_fld = NULL;
3772 static int mdt_fld_init(const struct lu_env *env,
3774 struct mdt_device *m)
3776 struct seq_server_site *ss;
3780 ss = mdt_seq_site(m);
3782 OBD_ALLOC_PTR(ss->ss_server_fld);
3783 if (ss->ss_server_fld == NULL)
3784 RETURN(rc = -ENOMEM);
3786 rc = fld_server_init(env, ss->ss_server_fld, m->mdt_bottom, uuid,
3789 OBD_FREE_PTR(ss->ss_server_fld);
3790 ss->ss_server_fld = NULL;
3797 static void mdt_stack_pre_fini(const struct lu_env *env,
3798 struct mdt_device *m, struct lu_device *top)
3800 struct lustre_cfg_bufs *bufs;
3801 struct lustre_cfg *lcfg;
3802 struct mdt_thread_info *info;
3807 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3808 LASSERT(info != NULL);
3810 bufs = &info->mti_u.bufs;
3812 LASSERT(m->mdt_child_exp);
3813 LASSERT(m->mdt_child_exp->exp_obd);
3815 /* process cleanup, pass mdt obd name to get obd umount flags */
3816 /* XXX: this is needed because all layers are referenced by
3817 * objects (some of them are pinned by osd, for example *
3818 * the proper solution should be a model where object used
3819 * by osd only doesn't have mdt/mdd slices -bzzz */
3820 lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
3821 lustre_cfg_bufs_set_string(bufs, 1, NULL);
3822 lcfg = lustre_cfg_new(LCFG_PRE_CLEANUP, bufs);
3824 CERROR("%s:Cannot alloc lcfg!\n", mdt_obd_name(m));
3827 top->ld_ops->ldo_process_config(env, top, lcfg);
3828 lustre_cfg_free(lcfg);
3832 static void mdt_stack_fini(const struct lu_env *env,
3833 struct mdt_device *m, struct lu_device *top)
3835 struct obd_device *obd = mdt2obd_dev(m);
3836 struct lustre_cfg_bufs *bufs;
3837 struct lustre_cfg *lcfg;
3838 struct mdt_thread_info *info;
3842 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3843 LASSERT(info != NULL);
3845 lu_dev_del_linkage(top->ld_site, top);
3847 lu_site_purge(env, top->ld_site, -1);
3849 bufs = &info->mti_u.bufs;
3850 /* process cleanup, pass mdt obd name to get obd umount flags */
3851 /* another purpose is to let all layers to release their objects */
3852 lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
3857 lustre_cfg_bufs_set_string(bufs, 1, flags);
3858 lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
3860 CERROR("Cannot alloc lcfg!\n");
3864 top->ld_ops->ldo_process_config(env, top, lcfg);
3865 lustre_cfg_free(lcfg);
3867 lu_site_purge(env, top->ld_site, -1);
3869 m->mdt_child = NULL;
3870 m->mdt_bottom = NULL;
3872 obd_disconnect(m->mdt_child_exp);
3873 m->mdt_child_exp = NULL;
3875 obd_disconnect(m->mdt_bottom_exp);
3876 m->mdt_child_exp = NULL;
3879 static int mdt_connect_to_next(const struct lu_env *env, struct mdt_device *m,
3880 const char *next, struct obd_export **exp)
3882 struct obd_connect_data *data = NULL;
3883 struct obd_device *obd;
3887 OBD_ALLOC_PTR(data);
3889 GOTO(out, rc = -ENOMEM);
3891 obd = class_name2obd(next);
3893 CERROR("%s: can't locate next device: %s\n",
3894 mdt_obd_name(m), next);
3895 GOTO(out, rc = -ENOTCONN);
3898 data->ocd_connect_flags = OBD_CONNECT_VERSION;
3899 data->ocd_version = LUSTRE_VERSION_CODE;
3901 rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
3903 CERROR("%s: cannot connect to next dev %s (%d)\n",
3904 mdt_obd_name(m), next, rc);
3914 static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt,
3915 struct lustre_cfg *cfg)
3917 char *dev = lustre_cfg_string(cfg, 0);
3918 int rc, name_size, uuid_size;
3919 char *name, *uuid, *p;
3920 struct lustre_cfg_bufs *bufs;
3921 struct lustre_cfg *lcfg;
3922 struct obd_device *obd;
3923 struct lustre_profile *lprof;
3924 struct lu_site *site;
3927 /* in 1.8 we had the only device in the stack - MDS.
3928 * 2.0 introduces MDT, MDD, OSD; MDT starts others internally.
3929 * in 2.3 OSD is instantiated by obd_mount.c, so we need
3930 * to generate names and setup MDT, MDD. MDT will be using
3931 * generated name to connect to MDD. for MDD the next device
3932 * will be LOD with name taken from so called "profile" which
3933 * is generated by mount_option line
3935 * 1.8 MGS generates config. commands like this:
3936 * #06 (104)mount_option 0: 1:lustre-MDT0000 2:lustre-mdtlov
3937 * #08 (120)setup 0:lustre-MDT0000 1:dev 2:type 3:lustre-MDT0000
3938 * 2.0 MGS generates config. commands like this:
3939 * #07 (112)mount_option 0: 1:lustre-MDT0000 2:lustre-MDT0000-mdtlov
3940 * #08 (160)setup 0:lustre-MDT0000 1:lustre-MDT0000_UUID 2:0
3941 * 3:lustre-MDT0000-mdtlov 4:f
3943 * we generate MDD name from MDT one, just replacing T with D
3945 * after all the preparations, the logical equivalent will be
3946 * #01 (160)setup 0:lustre-MDD0000 1:lustre-MDD0000_UUID 2:0
3947 * 3:lustre-MDT0000-mdtlov 4:f
3948 * #02 (160)setup 0:lustre-MDT0000 1:lustre-MDT0000_UUID 2:0
3949 * 3:lustre-MDD0000 4:f
3951 * notice we build the stack from down to top: MDD first, then MDT */
3953 name_size = MAX_OBD_NAME;
3954 uuid_size = MAX_OBD_NAME;
3956 OBD_ALLOC(name, name_size);
3957 OBD_ALLOC(uuid, uuid_size);
3958 if (name == NULL || uuid == NULL)
3959 GOTO(cleanup_mem, rc = -ENOMEM);
3961 OBD_ALLOC_PTR(bufs);
3963 GOTO(cleanup_mem, rc = -ENOMEM);
3966 p = strstr(name, "-MDT");
3968 GOTO(free_bufs, rc = -ENOMEM);
3971 snprintf(uuid, MAX_OBD_NAME, "%s_UUID", name);
3973 lprof = class_get_profile(lustre_cfg_string(cfg, 0));
3974 if (lprof == NULL || lprof->lp_dt == NULL) {
3975 CERROR("can't find the profile: %s\n",
3976 lustre_cfg_string(cfg, 0));
3977 GOTO(free_bufs, rc = -EINVAL);
3980 lustre_cfg_bufs_reset(bufs, name);
3981 lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_MDD_NAME);
3982 lustre_cfg_bufs_set_string(bufs, 2, uuid);
3983 lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
3985 lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
3987 GOTO(free_bufs, rc = -ENOMEM);
3989 rc = class_attach(lcfg);
3991 GOTO(lcfg_cleanup, rc);
3993 obd = class_name2obd(name);
3995 CERROR("Can not find obd %s (%s in config)\n",
3996 MDD_OBD_NAME, lustre_cfg_string(cfg, 0));
3997 GOTO(class_detach, rc = -EINVAL);
4000 lustre_cfg_free(lcfg);
4002 lustre_cfg_bufs_reset(bufs, name);
4003 lustre_cfg_bufs_set_string(bufs, 1, uuid);
4004 lustre_cfg_bufs_set_string(bufs, 2, dev);
4005 lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4007 lcfg = lustre_cfg_new(LCFG_SETUP, bufs);
4009 rc = class_setup(obd, lcfg);
4011 GOTO(class_detach, rc);
4013 /* connect to MDD we just setup */
4014 rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_child_exp);
4016 GOTO(class_detach, rc);
4018 site = mdt->mdt_child_exp->exp_obd->obd_lu_dev->ld_site;
4020 LASSERT(mdt_lu_site(mdt) == NULL);
4021 mdt->mdt_lu_dev.ld_site = site;
4022 site->ls_top_dev = &mdt->mdt_lu_dev;
4023 mdt->mdt_child = lu2md_dev(mdt->mdt_child_exp->exp_obd->obd_lu_dev);
4025 /* now connect to bottom OSD */
4026 snprintf(name, MAX_OBD_NAME, "%s-osd", dev);
4027 rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_bottom_exp);
4029 GOTO(class_detach, rc);
4031 lu2dt_dev(mdt->mdt_bottom_exp->exp_obd->obd_lu_dev);
4033 rc = lu_env_refill((struct lu_env *)env);
4035 CERROR("Failure to refill session: '%d'\n", rc);
4037 lu_dev_add_linkage(site, &mdt->mdt_lu_dev);
4042 class_detach(obd, lcfg);
4044 lustre_cfg_free(lcfg);
4049 OBD_FREE(name, name_size);
4051 OBD_FREE(uuid, uuid_size);
4055 /* setup quota master target on MDT0 */
4056 static int mdt_quota_init(const struct lu_env *env, struct mdt_device *mdt,
4057 struct lustre_cfg *cfg)
4059 struct obd_device *obd;
4060 char *dev = lustre_cfg_string(cfg, 0);
4061 char *qmtname, *uuid, *p;
4062 struct lustre_cfg_bufs *bufs;
4063 struct lustre_cfg *lcfg;
4064 struct lustre_profile *lprof;
4065 struct obd_connect_data *data;
4069 LASSERT(mdt->mdt_qmt_exp == NULL);
4070 LASSERT(mdt->mdt_qmt_dev == NULL);
4072 /* quota master is on MDT0 only for now */
4073 if (mdt->mdt_seq_site.ss_node_id != 0)
4076 /* MGS generates config commands which look as follows:
4077 * #01 (160)setup 0:lustre-MDT0000 1:lustre-MDT0000_UUID 2:0
4078 * 3:lustre-MDT0000-mdtlov 4:f
4080 * We generate the QMT name from the MDT one, just replacing MD with QM
4081 * after all the preparations, the logical equivalent will be:
4082 * #01 (160)setup 0:lustre-QMT0000 1:lustre-QMT0000_UUID 2:0
4083 * 3:lustre-MDT0000-osd 4:f */
4084 OBD_ALLOC(qmtname, MAX_OBD_NAME);
4085 OBD_ALLOC(uuid, UUID_MAX);
4086 OBD_ALLOC_PTR(bufs);
4087 OBD_ALLOC_PTR(data);
4088 if (qmtname == NULL || uuid == NULL || bufs == NULL || data == NULL)
4089 GOTO(cleanup_mem, rc = -ENOMEM);
4091 strcpy(qmtname, dev);
4092 p = strstr(qmtname, "-MDT");
4094 GOTO(cleanup_mem, rc = -ENOMEM);
4095 /* replace MD with QM */
4099 snprintf(uuid, UUID_MAX, "%s_UUID", qmtname);
4101 lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4102 if (lprof == NULL || lprof->lp_dt == NULL) {
4103 CERROR("can't find profile for %s\n",
4104 lustre_cfg_string(cfg, 0));
4105 GOTO(cleanup_mem, rc = -EINVAL);
4108 lustre_cfg_bufs_reset(bufs, qmtname);
4109 lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_QMT_NAME);
4110 lustre_cfg_bufs_set_string(bufs, 2, uuid);
4111 lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4113 lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
4115 GOTO(cleanup_mem, rc = -ENOMEM);
4117 rc = class_attach(lcfg);
4119 GOTO(lcfg_cleanup, rc);
4121 obd = class_name2obd(qmtname);
4123 CERROR("Can not find obd %s (%s in config)\n", qmtname,
4124 lustre_cfg_string(cfg, 0));
4125 GOTO(class_detach, rc = -EINVAL);
4128 lustre_cfg_free(lcfg);
4130 lustre_cfg_bufs_reset(bufs, qmtname);
4131 lustre_cfg_bufs_set_string(bufs, 1, uuid);
4132 lustre_cfg_bufs_set_string(bufs, 2, dev);
4134 /* for quota, the next device should be the OSD device */
4135 lustre_cfg_bufs_set_string(bufs, 3,
4136 mdt->mdt_bottom->dd_lu_dev.ld_obd->obd_name);
4138 lcfg = lustre_cfg_new(LCFG_SETUP, bufs);
4140 rc = class_setup(obd, lcfg);
4142 GOTO(class_detach, rc);
4144 mdt->mdt_qmt_dev = obd->obd_lu_dev;
4146 /* configure local quota objects */
4147 rc = mdt->mdt_qmt_dev->ld_ops->ldo_prepare(env,
4151 GOTO(class_cleanup, rc);
4153 /* connect to quota master target */
4154 data->ocd_connect_flags = OBD_CONNECT_VERSION;
4155 data->ocd_version = LUSTRE_VERSION_CODE;
4156 rc = obd_connect(NULL, &mdt->mdt_qmt_exp, obd, &obd->obd_uuid,
4159 CERROR("cannot connect to quota master device %s (%d)\n",
4161 GOTO(class_cleanup, rc);
4167 class_manual_cleanup(obd);
4168 mdt->mdt_qmt_dev = NULL;
4172 class_detach(obd, lcfg);
4174 lustre_cfg_free(lcfg);
4179 OBD_FREE(qmtname, MAX_OBD_NAME);
4181 OBD_FREE(uuid, UUID_MAX);
4187 /* Shutdown quota master target associated with mdt */
4188 static void mdt_quota_fini(const struct lu_env *env, struct mdt_device *mdt)
4192 if (mdt->mdt_qmt_exp == NULL)
4194 LASSERT(mdt->mdt_qmt_dev != NULL);
4196 /* the qmt automatically shuts down when the mdt disconnects */
4197 obd_disconnect(mdt->mdt_qmt_exp);
4198 mdt->mdt_qmt_exp = NULL;
4199 mdt->mdt_qmt_dev = NULL;
4203 /* mdt_getxattr() is used from mdt_intent_getxattr(), use this wrapper
4204 * for now. This will be removed along with converting rest of MDT code
4205 * to use tgt_session_info */
4206 int mdt_tgt_getxattr(struct tgt_session_info *tsi)
4208 struct mdt_thread_info *info = tsi2mdt_info(tsi);
4211 rc = mdt_getxattr(info);
4213 mdt_thread_info_fini(info);
4217 static struct tgt_handler mdt_tgt_handlers[] = {
4218 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4219 0, MDS_CONNECT, mdt_tgt_connect,
4220 &RQF_CONNECT, LUSTRE_OBD_VERSION),
4221 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4222 0, MDS_DISCONNECT, tgt_disconnect,
4223 &RQF_MDS_DISCONNECT, LUSTRE_OBD_VERSION),
4224 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4225 HABEO_REFERO, MDS_SET_INFO, mdt_set_info,
4226 &RQF_OBD_SET_INFO, LUSTRE_MDS_VERSION),
4227 TGT_MDT_HDL(0, MDS_GET_INFO, mdt_get_info),
4228 TGT_MDT_HDL(0 | HABEO_REFERO, MDS_GETSTATUS, mdt_getstatus),
4229 TGT_MDT_HDL(HABEO_CORPUS, MDS_GETATTR, mdt_getattr),
4230 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_GETATTR_NAME,
4232 TGT_MDT_HDL(HABEO_CORPUS, MDS_GETXATTR, mdt_tgt_getxattr),
4233 TGT_MDT_HDL(0 | HABEO_REFERO, MDS_STATFS, mdt_statfs),
4234 TGT_MDT_HDL(0 | MUTABOR, MDS_REINT, mdt_reint),
4235 TGT_MDT_HDL(HABEO_CORPUS, MDS_CLOSE, mdt_close),
4236 TGT_MDT_HDL(HABEO_CORPUS, MDS_DONE_WRITING,
4238 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_READPAGE, mdt_readpage),
4239 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_SYNC, mdt_sync),
4240 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_IS_SUBDIR, mdt_is_subdir),
4241 TGT_MDT_HDL(0, MDS_QUOTACTL, mdt_quotactl),
4242 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_PROGRESS,
4244 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_REGISTER,
4245 mdt_hsm_ct_register),
4246 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_UNREGISTER,
4247 mdt_hsm_ct_unregister),
4248 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_STATE_GET,
4250 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_STATE_SET,
4252 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_ACTION, mdt_hsm_action),
4253 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_REQUEST,
4255 TGT_MDT_HDL(HABEO_CLAVIS | HABEO_CORPUS | HABEO_REFERO | MUTABOR,
4260 static struct tgt_handler mdt_sec_ctx_ops[] = {
4261 TGT_SEC_HDL_VAR(0, SEC_CTX_INIT, mdt_sec_ctx_handle),
4262 TGT_SEC_HDL_VAR(0, SEC_CTX_INIT_CONT,mdt_sec_ctx_handle),
4263 TGT_SEC_HDL_VAR(0, SEC_CTX_FINI, mdt_sec_ctx_handle)
4266 static struct tgt_handler mdt_quota_ops[] = {
4267 TGT_QUOTA_HDL(HABEO_REFERO, QUOTA_DQACQ, mdt_quota_dqacq),
4270 static struct tgt_opc_slice mdt_common_slice[] = {
4272 .tos_opc_start = MDS_FIRST_OPC,
4273 .tos_opc_end = MDS_LAST_OPC,
4274 .tos_hs = mdt_tgt_handlers
4277 .tos_opc_start = OBD_FIRST_OPC,
4278 .tos_opc_end = OBD_LAST_OPC,
4279 .tos_hs = tgt_obd_handlers
4282 .tos_opc_start = LDLM_FIRST_OPC,
4283 .tos_opc_end = LDLM_LAST_OPC,
4284 .tos_hs = tgt_dlm_handlers
4287 .tos_opc_start = SEC_FIRST_OPC,
4288 .tos_opc_end = SEC_LAST_OPC,
4289 .tos_hs = mdt_sec_ctx_ops
4292 .tos_opc_start = OUT_UPDATE_FIRST_OPC,
4293 .tos_opc_end = OUT_UPDATE_LAST_OPC,
4294 .tos_hs = tgt_out_handlers
4297 .tos_opc_start = FLD_FIRST_OPC,
4298 .tos_opc_end = FLD_LAST_OPC,
4299 .tos_hs = fld_handlers
4302 .tos_opc_start = SEQ_FIRST_OPC,
4303 .tos_opc_end = SEQ_LAST_OPC,
4304 .tos_hs = seq_handlers
4307 .tos_opc_start = QUOTA_DQACQ,
4308 .tos_opc_end = QUOTA_LAST_OPC,
4309 .tos_hs = mdt_quota_ops
4312 .tos_opc_start = LLOG_FIRST_OPC,
4313 .tos_opc_end = LLOG_LAST_OPC,
4314 .tos_hs = tgt_llog_handlers
4317 .tos_opc_start = LFSCK_FIRST_OPC,
4318 .tos_opc_end = LFSCK_LAST_OPC,
4319 .tos_hs = tgt_lfsck_handlers
4327 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
4329 struct md_device *next = m->mdt_child;
4330 struct lu_device *d = &m->mdt_lu_dev;
4331 struct obd_device *obd = mdt2obd_dev(m);
4332 struct lfsck_stop stop;
4335 stop.ls_status = LS_PAUSED;
4337 next->md_ops->mdo_iocontrol(env, next, OBD_IOC_STOP_LFSCK, 0, &stop);
4339 target_recovery_fini(obd);
4340 ping_evictor_stop();
4341 mdt_stack_pre_fini(env, m, md2lu_dev(m->mdt_child));
4343 if (m->mdt_opts.mo_coordinator)
4344 mdt_hsm_cdt_stop(m);
4346 mdt_hsm_cdt_fini(m);
4348 mdt_llog_ctxt_unclone(env, m, LLOG_AGENT_ORIG_CTXT);
4349 mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4350 obd_exports_barrier(obd);
4351 obd_zombie_barrier();
4355 tgt_fini(env, &m->mdt_lut);
4356 mdt_fs_cleanup(env, m);
4357 upcall_cache_cleanup(m->mdt_identity_cache);
4358 m->mdt_identity_cache = NULL;
4360 if (m->mdt_namespace != NULL) {
4361 ldlm_namespace_free(m->mdt_namespace, NULL,
4362 d->ld_obd->obd_force);
4363 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
4366 mdt_quota_fini(env, m);
4368 cfs_free_nidlist(&m->mdt_nosquash_nids);
4369 if (m->mdt_nosquash_str) {
4370 OBD_FREE(m->mdt_nosquash_str, m->mdt_nosquash_strlen);
4371 m->mdt_nosquash_str = NULL;
4372 m->mdt_nosquash_strlen = 0;
4375 mdt_seq_fini(env, m);
4376 mdt_fld_fini(env, m);
4378 next->md_ops->mdo_init_capa_ctxt(env, next, 0, 0, 0, NULL);
4379 cfs_timer_disarm(&m->mdt_ck_timer);
4380 mdt_ck_thread_stop(m);
4385 mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4387 LASSERT(atomic_read(&d->ld_ref) == 0);
4389 server_put_mount(mdt_obd_name(m));
4394 static int mdt_postrecov(const struct lu_env *, struct mdt_device *);
4396 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
4397 struct lu_device_type *ldt, struct lustre_cfg *cfg)
4399 struct mdt_thread_info *info;
4400 struct obd_device *obd;
4401 const char *dev = lustre_cfg_string(cfg, 0);
4402 const char *num = lustre_cfg_string(cfg, 2);
4403 struct lustre_mount_info *lmi = NULL;
4404 struct lustre_sb_info *lsi;
4406 struct seq_server_site *ss_site;
4407 const char *identity_upcall = "NONE";
4408 struct md_device *next;
4414 lu_device_init(&m->mdt_lu_dev, ldt);
4416 * Environment (env) might be missing mdt_thread_key values at that
4417 * point, if device is allocated when mdt_thread_key is in QUIESCENT
4420 * Usually device allocation path doesn't use module key values, but
4421 * mdt has to do a lot of work here, so allocate key value.
4423 rc = lu_env_refill((struct lu_env *)env);
4427 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4428 LASSERT(info != NULL);
4430 obd = class_name2obd(dev);
4431 LASSERT(obd != NULL);
4433 m->mdt_max_mdsize = MAX_MD_SIZE; /* 4 stripes */
4435 m->mdt_som_conf = 0;
4437 m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
4439 /* default is coordinator off, it is started through conf_param
4441 m->mdt_opts.mo_coordinator = 0;
4443 lmi = server_get_mount(dev);
4445 CERROR("Cannot get mount info for %s!\n", dev);
4448 lsi = s2lsi(lmi->lmi_sb);
4449 /* CMD is supported only in IAM mode */
4451 node_id = simple_strtol(num, NULL, 10);
4452 obd->u.obt.obt_magic = OBT_MAGIC;
4455 spin_lock_init(&m->mdt_ioepoch_lock);
4456 m->mdt_capa_timeout = CAPA_TIMEOUT;
4457 m->mdt_capa_alg = CAPA_HMAC_ALG_SHA1;
4458 m->mdt_ck_timeout = CAPA_KEY_TIMEOUT;
4459 m->mdt_squash_uid = 0;
4460 m->mdt_squash_gid = 0;
4461 CFS_INIT_LIST_HEAD(&m->mdt_nosquash_nids);
4462 m->mdt_nosquash_str = NULL;
4463 m->mdt_nosquash_strlen = 0;
4464 init_rwsem(&m->mdt_squash_sem);
4465 spin_lock_init(&m->mdt_osfs_lock);
4466 m->mdt_osfs_age = cfs_time_shift_64(-1000);
4467 m->mdt_enable_remote_dir = 0;
4468 m->mdt_enable_remote_dir_gid = 0;
4470 m->mdt_lu_dev.ld_ops = &mdt_lu_ops;
4471 m->mdt_lu_dev.ld_obd = obd;
4472 /* Set this lu_device to obd for error handling purposes. */
4473 obd->obd_lu_dev = &m->mdt_lu_dev;
4475 /* init the stack */
4476 rc = mdt_stack_init((struct lu_env *)env, m, cfg);
4478 CERROR("%s: Can't init device stack, rc %d\n",
4479 mdt_obd_name(m), rc);
4484 ss_site = mdt_seq_site(m);
4485 s->ld_seq_site = ss_site;
4488 /* set server index */
4489 ss_site->ss_node_id = node_id;
4491 /* failover is the default
4492 * FIXME: we do not failout mds0/mgs, which may cause some problems.
4493 * assumed whose ss_node_id == 0 XXX
4495 obd->obd_replayable = 1;
4496 /* No connection accepted until configurations will finish */
4497 obd->obd_no_conn = 1;
4499 if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
4500 char *str = lustre_cfg_string(cfg, 4);
4501 if (strchr(str, 'n')) {
4502 CWARN("%s: recovery disabled\n", mdt_obd_name(m));
4503 obd->obd_replayable = 0;
4507 rc = mdt_fld_init(env, mdt_obd_name(m), m);
4509 GOTO(err_fini_stack, rc);
4511 rc = mdt_seq_init(env, m);
4513 GOTO(err_fini_fld, rc);
4515 snprintf(info->mti_u.ns_name, sizeof(info->mti_u.ns_name), "%s-%s",
4516 LUSTRE_MDT_NAME, obd->obd_uuid.uuid);
4517 m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
4518 LDLM_NAMESPACE_SERVER,
4519 LDLM_NAMESPACE_GREEDY,
4521 if (m->mdt_namespace == NULL)
4522 GOTO(err_fini_seq, rc = -ENOMEM);
4524 m->mdt_namespace->ns_lvbp = m;
4525 m->mdt_namespace->ns_lvbo = &mdt_lvbo;
4527 ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
4528 /* set obd_namespace for compatibility with old code */
4529 obd->obd_namespace = m->mdt_namespace;
4531 cfs_timer_init(&m->mdt_ck_timer, mdt_ck_timer_callback, m);
4533 rc = mdt_hsm_cdt_init(m);
4535 CERROR("%s: error initializing coordinator, rc %d\n",
4536 mdt_obd_name(m), rc);
4537 GOTO(err_free_ns, rc);
4540 rc = mdt_ck_thread_start(m);
4542 GOTO(err_free_hsm, rc);
4544 rc = tgt_init(env, &m->mdt_lut, obd, m->mdt_bottom, mdt_common_slice,
4545 OBD_FAIL_MDS_ALL_REQUEST_NET,
4546 OBD_FAIL_MDS_ALL_REPLY_NET);
4550 rc = mdt_fs_setup(env, m, obd, lsi);
4554 tgt_adapt_sptlrpc_conf(&m->mdt_lut, 1);
4556 next = m->mdt_child;
4557 rc = next->md_ops->mdo_iocontrol(env, next, OBD_IOC_GET_MNTOPT, 0,
4560 GOTO(err_fs_cleanup, rc);
4562 if (mntopts & MNTOPT_USERXATTR)
4563 m->mdt_opts.mo_user_xattr = 1;
4565 m->mdt_opts.mo_user_xattr = 0;
4567 rc = next->md_ops->mdo_maxeasize_get(env, next, &m->mdt_max_ea_size);
4569 GOTO(err_fs_cleanup, rc);
4571 if (mntopts & MNTOPT_ACL)
4572 m->mdt_opts.mo_acl = 1;
4574 m->mdt_opts.mo_acl = 0;
4576 /* XXX: to support suppgid for ACL, we enable identity_upcall
4577 * by default, otherwise, maybe got unexpected -EACCESS. */
4578 if (m->mdt_opts.mo_acl)
4579 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
4581 m->mdt_identity_cache = upcall_cache_init(mdt_obd_name(m),
4583 &mdt_identity_upcall_cache_ops);
4584 if (IS_ERR(m->mdt_identity_cache)) {
4585 rc = PTR_ERR(m->mdt_identity_cache);
4586 m->mdt_identity_cache = NULL;
4587 GOTO(err_fs_cleanup, rc);
4590 rc = mdt_procfs_init(m, dev);
4592 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
4593 GOTO(err_recovery, rc);
4596 rc = mdt_quota_init(env, m, cfg);
4598 GOTO(err_procfs, rc);
4600 m->mdt_ldlm_client = &mdt2obd_dev(m)->obd_ldlm_client;
4601 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
4602 "mdt_ldlm_client", m->mdt_ldlm_client);
4604 ping_evictor_start();
4606 /* recovery will be started upon mdt_prepare()
4607 * when the whole stack is complete and ready
4608 * to serve the requests */
4610 mdt_init_capa_ctxt(env, m);
4612 /* Reduce the initial timeout on an MDS because it doesn't need such
4613 * a long timeout as an OST does. Adaptive timeouts will adjust this
4614 * value appropriately. */
4615 if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
4616 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
4622 target_recovery_fini(obd);
4623 upcall_cache_cleanup(m->mdt_identity_cache);
4624 m->mdt_identity_cache = NULL;
4626 mdt_fs_cleanup(env, m);
4628 tgt_fini(env, &m->mdt_lut);
4630 cfs_timer_disarm(&m->mdt_ck_timer);
4631 mdt_ck_thread_stop(m);
4633 mdt_hsm_cdt_fini(m);
4635 ldlm_namespace_free(m->mdt_namespace, NULL, 0);
4636 obd->obd_namespace = m->mdt_namespace = NULL;
4638 mdt_seq_fini(env, m);
4640 mdt_fld_fini(env, m);
4642 mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4645 server_put_mount(dev);
4649 /* For interoperability, the left element is old parameter, the right one
4650 * is the new version of the parameter, if some parameter is deprecated,
4651 * the new version should be set as NULL. */
4652 static struct cfg_interop_param mdt_interop_param[] = {
4653 { "mdt.group_upcall", NULL },
4654 { "mdt.quota_type", NULL },
4655 { "mdd.quota_type", NULL },
4656 { "mdt.rootsquash", "mdt.root_squash" },
4657 { "mdt.nosquash_nid", "mdt.nosquash_nids" },
4661 /* used by MGS to process specific configurations */
4662 static int mdt_process_config(const struct lu_env *env,
4663 struct lu_device *d, struct lustre_cfg *cfg)
4665 struct mdt_device *m = mdt_dev(d);
4666 struct md_device *md_next = m->mdt_child;
4667 struct lu_device *next = md2lu_dev(md_next);
4671 switch (cfg->lcfg_command) {
4673 struct lprocfs_static_vars lvars;
4674 struct obd_device *obd = d->ld_obd;
4676 /* For interoperability */
4677 struct cfg_interop_param *ptr = NULL;
4678 struct lustre_cfg *old_cfg = NULL;
4681 param = lustre_cfg_string(cfg, 1);
4682 if (param == NULL) {
4683 CERROR("param is empty\n");
4688 ptr = class_find_old_param(param, mdt_interop_param);
4690 if (ptr->new_param == NULL) {
4692 CWARN("For interoperability, skip this %s."
4693 " It is obsolete.\n", ptr->old_param);
4697 CWARN("Found old param %s, changed it to %s.\n",
4698 ptr->old_param, ptr->new_param);
4701 cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
4708 lprocfs_mdt_init_vars(&lvars);
4709 rc = class_process_proc_param(PARAM_MDT, lvars.obd_vars,
4711 if (rc > 0 || rc == -ENOSYS) {
4712 /* is it an HSM var ? */
4713 rc = class_process_proc_param(PARAM_HSM,
4714 hsm_cdt_get_proc_vars(),
4716 if (rc > 0 || rc == -ENOSYS)
4717 /* we don't understand; pass it on */
4718 rc = next->ld_ops->ldo_process_config(env, next,
4722 if (old_cfg != NULL)
4723 lustre_cfg_free(cfg);
4728 /* others are passed further */
4729 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4735 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
4736 const struct lu_object_header *hdr,
4737 struct lu_device *d)
4739 struct mdt_object *mo;
4743 OBD_SLAB_ALLOC_PTR_GFP(mo, mdt_object_kmem, GFP_NOFS);
4745 struct lu_object *o;
4746 struct lu_object_header *h;
4749 h = &mo->mot_header;
4750 lu_object_header_init(h);
4751 lu_object_init(o, h, d);
4752 lu_object_add_top(h, o);
4753 o->lo_ops = &mdt_obj_ops;
4754 mutex_init(&mo->mot_ioepoch_mutex);
4755 mutex_init(&mo->mot_lov_mutex);
4756 init_rwsem(&mo->mot_open_sem);
4762 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
4763 const struct lu_object_conf *unused)
4765 struct mdt_device *d = mdt_dev(o->lo_dev);
4766 struct lu_device *under;
4767 struct lu_object *below;
4771 CDEBUG(D_INFO, "object init, fid = "DFID"\n",
4772 PFID(lu_object_fid(o)));
4774 under = &d->mdt_child->md_lu_dev;
4775 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
4776 if (below != NULL) {
4777 lu_object_add(o, below);
4784 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
4786 struct mdt_object *mo = mdt_obj(o);
4787 struct lu_object_header *h;
4791 CDEBUG(D_INFO, "object free, fid = "DFID"\n",
4792 PFID(lu_object_fid(o)));
4794 LASSERT(atomic_read(&mo->mot_open_count) == 0);
4795 LASSERT(atomic_read(&mo->mot_lease_count) == 0);
4798 lu_object_header_fini(h);
4799 OBD_SLAB_FREE_PTR(mo, mdt_object_kmem);
4804 static int mdt_object_print(const struct lu_env *env, void *cookie,
4805 lu_printer_t p, const struct lu_object *o)
4807 struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
4808 return (*p)(env, cookie, LUSTRE_MDT_NAME"-object@%p(ioepoch="LPU64" "
4809 "flags="LPX64", epochcount=%d, writecount=%d)",
4810 mdto, mdto->mot_ioepoch, mdto->mot_flags,
4811 mdto->mot_ioepoch_count, mdto->mot_writecount);
4814 static int mdt_prepare(const struct lu_env *env,
4815 struct lu_device *pdev,
4816 struct lu_device *cdev)
4818 struct mdt_device *mdt = mdt_dev(cdev);
4819 struct lu_device *next = &mdt->mdt_child->md_lu_dev;
4820 struct obd_device *obd = cdev->ld_obd;
4821 struct lfsck_start_param lsp;
4828 rc = next->ld_ops->ldo_prepare(env, cdev, next);
4832 rc = mdt_llog_ctxt_clone(env, mdt, LLOG_CHANGELOG_ORIG_CTXT);
4836 rc = mdt_llog_ctxt_clone(env, mdt, LLOG_AGENT_ORIG_CTXT);
4840 rc = lfsck_register_namespace(env, mdt->mdt_bottom, mdt->mdt_namespace);
4841 /* The LFSCK instance is registered just now, so it must be there when
4842 * register the namespace to such instance. */
4843 LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
4845 lsp.lsp_start = NULL;
4846 lsp.lsp_index_valid = 0;
4847 rc = mdt->mdt_child->md_ops->mdo_iocontrol(env, mdt->mdt_child,
4848 OBD_IOC_START_LFSCK,
4851 CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
4852 mdt_obd_name(mdt), rc);
4856 if (mdt->mdt_seq_site.ss_node_id == 0) {
4857 rc = mdt->mdt_child->md_ops->mdo_root_get(env, mdt->mdt_child,
4858 &mdt->mdt_md_root_fid);
4863 LASSERT(!test_bit(MDT_FL_CFGLOG, &mdt->mdt_state));
4864 target_recovery_init(&mdt->mdt_lut, tgt_request_handle);
4865 set_bit(MDT_FL_CFGLOG, &mdt->mdt_state);
4866 LASSERT(obd->obd_no_conn);
4867 spin_lock(&obd->obd_dev_lock);
4868 obd->obd_no_conn = 0;
4869 spin_unlock(&obd->obd_dev_lock);
4871 if (obd->obd_recovering == 0)
4872 mdt_postrecov(env, mdt);
4877 const struct lu_device_operations mdt_lu_ops = {
4878 .ldo_object_alloc = mdt_object_alloc,
4879 .ldo_process_config = mdt_process_config,
4880 .ldo_prepare = mdt_prepare,
4883 static const struct lu_object_operations mdt_obj_ops = {
4884 .loo_object_init = mdt_object_init,
4885 .loo_object_free = mdt_object_free,
4886 .loo_object_print = mdt_object_print
4889 static int mdt_obd_set_info_async(const struct lu_env *env,
4890 struct obd_export *exp,
4891 __u32 keylen, void *key,
4892 __u32 vallen, void *val,
4893 struct ptlrpc_request_set *set)
4899 if (KEY_IS(KEY_SPTLRPC_CONF)) {
4900 rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp), 0);
4908 * Match client and server connection feature flags.
4910 * Compute the compatibility flags for a connection request based on
4911 * features mutually supported by client and server.
4913 * The obd_export::exp_connect_data.ocd_connect_flags field in \a exp
4914 * must not be updated here, otherwise a partially initialized value may
4915 * be exposed. After the connection request is successfully processed,
4916 * the top-level MDT connect request handler atomically updates the export
4917 * connect flags from the obd_connect_data::ocd_connect_flags field of the
4918 * reply. \see mdt_connect().
4920 * \param exp the obd_export associated with this client/target pair
4921 * \param mdt the target device for the connection
4922 * \param data stores data for this connect request
4925 * \retval -EPROTO \a data unexpectedly has zero obd_connect_data::ocd_brw_size
4926 * \retval -EBADE client and server feature requirements are incompatible
4928 static int mdt_connect_internal(struct obd_export *exp,
4929 struct mdt_device *mdt,
4930 struct obd_connect_data *data)
4932 LASSERT(data != NULL);
4934 data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
4935 data->ocd_ibits_known &= MDS_INODELOCK_FULL;
4937 /* If no known bits (which should not happen, probably,
4938 as everybody should support LOOKUP and UPDATE bits at least)
4939 revert to compat mode with plain locks. */
4940 if (!data->ocd_ibits_known &&
4941 data->ocd_connect_flags & OBD_CONNECT_IBITS)
4942 data->ocd_connect_flags &= ~OBD_CONNECT_IBITS;
4944 if (!mdt->mdt_opts.mo_acl)
4945 data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
4947 if (!mdt->mdt_opts.mo_user_xattr)
4948 data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
4950 if (!mdt->mdt_som_conf)
4951 data->ocd_connect_flags &= ~OBD_CONNECT_SOM;
4953 if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE) {
4954 data->ocd_brw_size = min(data->ocd_brw_size,
4955 (__u32)MD_MAX_BRW_SIZE);
4956 if (data->ocd_brw_size == 0) {
4957 CERROR("%s: cli %s/%p ocd_connect_flags: "LPX64
4958 " ocd_version: %x ocd_grant: %d "
4959 "ocd_index: %u ocd_brw_size is "
4960 "unexpectedly zero, network data "
4961 "corruption? Refusing connection of this"
4964 exp->exp_client_uuid.uuid,
4965 exp, data->ocd_connect_flags, data->ocd_version,
4966 data->ocd_grant, data->ocd_index);
4971 /* NB: Disregard the rule against updating
4972 * exp_connect_data.ocd_connect_flags in this case, since
4973 * tgt_client_new() needs to know if this is a lightweight
4974 * connection, and it is safe to expose this flag before
4975 * connection processing completes. */
4976 if (data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) {
4977 spin_lock(&exp->exp_lock);
4978 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_LIGHTWEIGHT;
4979 spin_unlock(&exp->exp_lock);
4982 data->ocd_version = LUSTRE_VERSION_CODE;
4984 if ((data->ocd_connect_flags & OBD_CONNECT_FID) == 0) {
4985 CWARN("%s: MDS requires FID support, but client not\n",
4990 if (mdt->mdt_som_conf &&
4991 !(data->ocd_connect_flags & (OBD_CONNECT_LIGHTWEIGHT |
4992 OBD_CONNECT_MDS_MDS |
4993 OBD_CONNECT_SOM))) {
4994 CWARN("%s: MDS has SOM enabled, but client does not support "
4995 "it\n", mdt_obd_name(mdt));
4999 if (OCD_HAS_FLAG(data, PINGLESS)) {
5000 if (ptlrpc_pinger_suppress_pings()) {
5001 spin_lock(&exp->exp_obd->obd_dev_lock);
5002 list_del_init(&exp->exp_obd_chain_timed);
5003 spin_unlock(&exp->exp_obd->obd_dev_lock);
5005 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
5009 data->ocd_max_easize = mdt->mdt_max_ea_size;
5014 /* mds_connect copy */
5015 static int mdt_obd_connect(const struct lu_env *env,
5016 struct obd_export **exp, struct obd_device *obd,
5017 struct obd_uuid *cluuid,
5018 struct obd_connect_data *data,
5021 struct obd_export *lexp;
5022 struct lustre_handle conn = { 0 };
5023 struct mdt_device *mdt;
5027 LASSERT(env != NULL);
5028 if (!exp || !obd || !cluuid)
5031 mdt = mdt_dev(obd->obd_lu_dev);
5034 * first, check whether the stack is ready to handle requests
5035 * XXX: probably not very appropriate method is used now
5036 * at some point we should find a better one
5038 if (!test_bit(MDT_FL_SYNCED, &mdt->mdt_state) && data != NULL &&
5039 !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT)) {
5040 rc = obd_get_info(env, mdt->mdt_child_exp,
5041 sizeof(KEY_OSP_CONNECTED),
5042 KEY_OSP_CONNECTED, NULL, NULL, NULL);
5045 set_bit(MDT_FL_SYNCED, &mdt->mdt_state);
5048 rc = class_connect(&conn, obd, cluuid);
5052 lexp = class_conn2export(&conn);
5053 LASSERT(lexp != NULL);
5055 rc = mdt_connect_internal(lexp, mdt, data);
5057 struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
5060 memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
5061 rc = tgt_client_new(env, lexp);
5063 mdt_export_stats_init(obd, lexp, localdata);
5065 /* For phase I, sync for cross-ref operation. */
5066 lexp->exp_keep_sync = 1;
5070 class_disconnect(lexp);
5079 static int mdt_obd_reconnect(const struct lu_env *env,
5080 struct obd_export *exp, struct obd_device *obd,
5081 struct obd_uuid *cluuid,
5082 struct obd_connect_data *data,
5088 if (exp == NULL || obd == NULL || cluuid == NULL)
5091 rc = mdt_connect_internal(exp, mdt_dev(obd->obd_lu_dev), data);
5093 mdt_export_stats_init(obd, exp, localdata);
5098 static int mdt_ctxt_add_dirty_flag(struct lu_env *env,
5099 struct mdt_thread_info *info,
5100 struct mdt_file_data *mfd)
5102 struct lu_context ses;
5106 rc = lu_context_init(&ses, LCT_SERVER_SESSION);
5111 lu_context_enter(&ses);
5113 mdt_ucred(info)->uc_valid = UCRED_OLD;
5114 rc = mdt_add_dirty_flag(info, mfd->mfd_object, &info->mti_attr);
5116 lu_context_exit(&ses);
5117 lu_context_fini(&ses);
5123 static int mdt_export_cleanup(struct obd_export *exp)
5125 struct mdt_export_data *med = &exp->exp_mdt_data;
5126 struct obd_device *obd = exp->exp_obd;
5127 struct mdt_device *mdt;
5128 struct mdt_thread_info *info;
5130 CFS_LIST_HEAD(closing_list);
5131 struct mdt_file_data *mfd, *n;
5135 spin_lock(&med->med_open_lock);
5136 while (!cfs_list_empty(&med->med_open_head)) {
5137 cfs_list_t *tmp = med->med_open_head.next;
5138 mfd = cfs_list_entry(tmp, struct mdt_file_data, mfd_list);
5140 /* Remove mfd handle so it can't be found again.
5141 * We are consuming the mfd_list reference here. */
5142 class_handle_unhash(&mfd->mfd_handle);
5143 cfs_list_move_tail(&mfd->mfd_list, &closing_list);
5145 spin_unlock(&med->med_open_lock);
5146 mdt = mdt_dev(obd->obd_lu_dev);
5147 LASSERT(mdt != NULL);
5149 rc = lu_env_init(&env, LCT_MD_THREAD);
5153 info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5154 LASSERT(info != NULL);
5155 memset(info, 0, sizeof *info);
5156 info->mti_env = &env;
5157 info->mti_mdt = mdt;
5158 info->mti_exp = exp;
5160 if (!cfs_list_empty(&closing_list)) {
5161 struct md_attr *ma = &info->mti_attr;
5163 /* Close any open files (which may also cause orphan unlinking). */
5164 cfs_list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
5165 cfs_list_del_init(&mfd->mfd_list);
5166 ma->ma_need = ma->ma_valid = 0;
5168 /* This file is being closed due to an eviction, it
5169 * could have been modified and now dirty regarding to
5170 * HSM archive, check this!
5171 * The logic here is to mark a file dirty if there's a
5172 * chance it was dirtied before the client was evicted,
5173 * so that we don't have to wait for a release attempt
5174 * before finding out the file was actually dirty and
5175 * fail the release. Aggressively marking it dirty here
5176 * will cause the policy engine to attempt to
5177 * re-archive it; when rearchiving, we can compare the
5178 * current version to the HSM data_version and make the
5179 * archive request into a noop if it's not actually
5182 if (mfd->mfd_mode & (FMODE_WRITE|MDS_FMODE_TRUNC))
5183 rc = mdt_ctxt_add_dirty_flag(&env, info, mfd);
5185 /* Don't unlink orphan on failover umount, LU-184 */
5186 if (exp->exp_flags & OBD_OPT_FAILOVER) {
5187 ma->ma_valid = MA_FLAGS;
5188 ma->ma_attr_flags |= MDS_KEEP_ORPHAN;
5190 mdt_mfd_close(info, mfd);
5193 info->mti_mdt = NULL;
5194 /* cleanup client slot early */
5195 /* Do not erase record for recoverable client. */
5196 if (!(exp->exp_flags & OBD_OPT_FAILOVER) || exp->exp_failed)
5197 tgt_client_del(&env, exp);
5203 static int mdt_obd_disconnect(struct obd_export *exp)
5209 class_export_get(exp);
5211 rc = server_disconnect_export(exp);
5213 CDEBUG(D_IOCTL, "server disconnect error: %d\n", rc);
5215 rc = mdt_export_cleanup(exp);
5216 class_export_put(exp);
5220 /* FIXME: Can we avoid using these two interfaces? */
5221 static int mdt_init_export(struct obd_export *exp)
5223 struct mdt_export_data *med = &exp->exp_mdt_data;
5227 CFS_INIT_LIST_HEAD(&med->med_open_head);
5228 spin_lock_init(&med->med_open_lock);
5229 mutex_init(&med->med_idmap_mutex);
5230 med->med_idmap = NULL;
5231 spin_lock(&exp->exp_lock);
5232 exp->exp_connecting = 1;
5233 spin_unlock(&exp->exp_lock);
5235 /* self-export doesn't need client data and ldlm initialization */
5236 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5237 &exp->exp_client_uuid)))
5240 rc = tgt_client_alloc(exp);
5244 rc = ldlm_init_export(exp);
5251 tgt_client_free(exp);
5253 CERROR("%s: Failed to initialize export: rc = %d\n",
5254 exp->exp_obd->obd_name, rc);
5258 static int mdt_destroy_export(struct obd_export *exp)
5262 if (exp_connect_rmtclient(exp))
5263 mdt_cleanup_idmap(&exp->exp_mdt_data);
5265 target_destroy_export(exp);
5266 /* destroy can be called from failed obd_setup, so
5267 * checking uuid is safer than obd_self_export */
5268 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5269 &exp->exp_client_uuid)))
5272 ldlm_destroy_export(exp);
5273 tgt_client_free(exp);
5275 LASSERT(cfs_list_empty(&exp->exp_outstanding_replies));
5276 LASSERT(cfs_list_empty(&exp->exp_mdt_data.med_open_head));
5281 /** The maximum depth that fid2path() will search.
5282 * This is limited only because we want to store the fids for
5283 * historical path lookup purposes.
5285 #define MAX_PATH_DEPTH 100
5287 /** mdt_path() lookup structure. */
5288 struct path_lookup_info {
5289 __u64 pli_recno; /**< history point */
5290 __u64 pli_currec; /**< current record */
5291 struct lu_fid pli_fid;
5292 struct lu_fid pli_fids[MAX_PATH_DEPTH]; /**< path, in fids */
5293 struct mdt_object *pli_mdt_obj;
5294 char *pli_path; /**< full path */
5296 int pli_linkno; /**< which hardlink to follow */
5297 int pli_fidcount; /**< number of \a pli_fids */
5300 int mdt_links_read(struct mdt_thread_info *info, struct mdt_object *mdt_obj,
5301 struct linkea_data *ldata)
5305 LASSERT(ldata->ld_buf->lb_buf != NULL);
5307 if (!mdt_object_exists(mdt_obj))
5310 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5311 ldata->ld_buf, XATTR_NAME_LINK);
5312 if (rc == -ERANGE) {
5313 /* Buf was too small, figure out what we need. */
5314 lu_buf_free(ldata->ld_buf);
5315 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5316 ldata->ld_buf, XATTR_NAME_LINK);
5319 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
5320 if (ldata->ld_buf->lb_buf == NULL)
5322 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5323 ldata->ld_buf, XATTR_NAME_LINK);
5328 return linkea_init(ldata);
5331 static int mdt_path_current(struct mdt_thread_info *info,
5332 struct path_lookup_info *pli)
5334 struct mdt_device *mdt = info->mti_mdt;
5335 struct mdt_object *mdt_obj;
5336 struct link_ea_header *leh;
5337 struct link_ea_entry *lee;
5338 struct lu_name *tmpname = &info->mti_name;
5339 struct lu_fid *tmpfid = &info->mti_tmp_fid1;
5340 struct lu_buf *buf = &info->mti_big_buf;
5343 struct linkea_data ldata = { 0 };
5347 /* temp buffer for path element, the buffer will be finally freed
5348 * in mdt_thread_info_fini */
5349 buf = lu_buf_check_and_alloc(buf, PATH_MAX);
5350 if (buf->lb_buf == NULL)
5354 ptr = pli->pli_path + pli->pli_pathlen - 1;
5357 pli->pli_fidcount = 0;
5358 pli->pli_fids[0] = *(struct lu_fid *)mdt_object_fid(pli->pli_mdt_obj);
5360 /* root FID only exists on MDT0, and fid2path should also ends at MDT0,
5361 * so checking root_fid can only happen on MDT0. */
5362 while (!lu_fid_eq(&mdt->mdt_md_root_fid,
5363 &pli->pli_fids[pli->pli_fidcount])) {
5364 mdt_obj = mdt_object_find(info->mti_env, mdt,
5365 &pli->pli_fids[pli->pli_fidcount]);
5366 if (IS_ERR(mdt_obj))
5367 GOTO(out, rc = PTR_ERR(mdt_obj));
5368 if (mdt_object_remote(mdt_obj)) {
5369 mdt_object_put(info->mti_env, mdt_obj);
5370 GOTO(remote_out, rc = -EREMOTE);
5372 if (!mdt_object_exists(mdt_obj)) {
5373 mdt_object_put(info->mti_env, mdt_obj);
5374 GOTO(out, rc = -ENOENT);
5377 rc = mdt_links_read(info, mdt_obj, &ldata);
5378 mdt_object_put(info->mti_env, mdt_obj);
5383 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
5384 linkea_entry_unpack(lee, &reclen, tmpname, tmpfid);
5385 /* If set, use link #linkno for path lookup, otherwise use
5386 link #0. Only do this for the final path element. */
5387 if (pli->pli_fidcount == 0 &&
5388 pli->pli_linkno < leh->leh_reccount) {
5390 for (count = 0; count < pli->pli_linkno; count++) {
5391 lee = (struct link_ea_entry *)
5392 ((char *)lee + reclen);
5393 linkea_entry_unpack(lee, &reclen, tmpname,
5396 if (pli->pli_linkno < leh->leh_reccount - 1)
5397 /* indicate to user there are more links */
5401 /* Pack the name in the end of the buffer */
5402 ptr -= tmpname->ln_namelen;
5403 if (ptr - 1 <= pli->pli_path)
5404 GOTO(out, rc = -EOVERFLOW);
5405 strncpy(ptr, tmpname->ln_name, tmpname->ln_namelen);
5408 /* Store the parent fid for historic lookup */
5409 if (++pli->pli_fidcount >= MAX_PATH_DEPTH)
5410 GOTO(out, rc = -EOVERFLOW);
5411 pli->pli_fids[pli->pli_fidcount] = *tmpfid;
5415 ptr++; /* skip leading / */
5416 memmove(pli->pli_path, ptr, pli->pli_path + pli->pli_pathlen - ptr);
5423 /* Returns the full path to this fid, as of changelog record recno. */
5424 static int mdt_path(struct mdt_thread_info *info, struct mdt_object *obj,
5425 char *path, int pathlen, __u64 *recno, int *linkno,
5428 struct mdt_device *mdt = info->mti_mdt;
5429 struct path_lookup_info *pli;
5437 if (lu_fid_eq(&mdt->mdt_md_root_fid, mdt_object_fid(obj))) {
5446 pli->pli_mdt_obj = obj;
5447 pli->pli_recno = *recno;
5448 pli->pli_path = path;
5449 pli->pli_pathlen = pathlen;
5450 pli->pli_linkno = *linkno;
5452 /* Retry multiple times in case file is being moved */
5453 while (tries-- && rc == -EAGAIN)
5454 rc = mdt_path_current(info, pli);
5456 /* return the last resolved fids to the client, so the client will
5457 * build the left path on another MDT for remote object */
5458 *fid = pli->pli_fids[pli->pli_fidcount];
5460 *recno = pli->pli_currec;
5461 /* Return next link index to caller */
5462 *linkno = pli->pli_linkno;
5469 static int mdt_fid2path(struct mdt_thread_info *info,
5470 struct getinfo_fid2path *fp)
5472 struct mdt_device *mdt = info->mti_mdt;
5473 struct mdt_object *obj;
5477 CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n",
5478 PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
5480 if (!fid_is_sane(&fp->gf_fid))
5483 if (!fid_is_namespace_visible(&fp->gf_fid)) {
5484 CWARN("%s: "DFID" is invalid, sequence should be "
5485 ">= "LPX64"\n", mdt_obd_name(mdt),
5486 PFID(&fp->gf_fid), (__u64)FID_SEQ_NORMAL);
5490 obj = mdt_object_find(info->mti_env, mdt, &fp->gf_fid);
5491 if (obj == NULL || IS_ERR(obj)) {
5492 CDEBUG(D_IOCTL, "no object "DFID": %ld\n", PFID(&fp->gf_fid),
5497 if (mdt_object_remote(obj))
5499 else if (!mdt_object_exists(obj))
5505 mdt_object_put(info->mti_env, obj);
5506 CDEBUG(D_IOCTL, "nonlocal object "DFID": %d\n",
5507 PFID(&fp->gf_fid), rc);
5511 rc = mdt_path(info, obj, fp->gf_path, fp->gf_pathlen, &fp->gf_recno,
5512 &fp->gf_linkno, &fp->gf_fid);
5514 CDEBUG(D_INFO, "fid "DFID", path %s recno "LPX64" linkno %u\n",
5515 PFID(&fp->gf_fid), fp->gf_path, fp->gf_recno, fp->gf_linkno);
5517 mdt_object_put(info->mti_env, obj);
5522 static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key,
5523 void *val, int vallen)
5525 struct getinfo_fid2path *fpout, *fpin;
5528 fpin = key + cfs_size_round(sizeof(KEY_FID2PATH));
5531 if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
5532 lustre_swab_fid2path(fpin);
5534 memcpy(fpout, fpin, sizeof(*fpin));
5535 if (fpout->gf_pathlen != vallen - sizeof(*fpin))
5538 rc = mdt_fid2path(info, fpout);
5542 int mdt_get_info(struct tgt_session_info *tsi)
5552 key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
5554 CDEBUG(D_IOCTL, "No GETINFO key");
5555 RETURN(err_serious(-EFAULT));
5557 keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
5560 vallen = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_VALLEN);
5561 if (vallen == NULL) {
5562 CDEBUG(D_IOCTL, "Unable to get RMF_GETINFO_VALLEN buffer");
5563 RETURN(err_serious(-EFAULT));
5566 req_capsule_set_size(tsi->tsi_pill, &RMF_GETINFO_VAL, RCL_SERVER,
5568 rc = req_capsule_server_pack(tsi->tsi_pill);
5570 RETURN(err_serious(rc));
5572 valout = req_capsule_server_get(tsi->tsi_pill, &RMF_GETINFO_VAL);
5573 if (valout == NULL) {
5574 CDEBUG(D_IOCTL, "Unable to get get-info RPC out buffer");
5575 RETURN(err_serious(-EFAULT));
5578 if (KEY_IS(KEY_FID2PATH)) {
5579 struct mdt_thread_info *info = tsi2mdt_info(tsi);
5581 rc = mdt_rpc_fid2path(info, key, valout, *vallen);
5582 mdt_thread_info_fini(info);
5589 /* Pass the ioc down */
5590 static int mdt_ioc_child(struct lu_env *env, struct mdt_device *mdt,
5591 unsigned int cmd, int len, void *data)
5593 struct lu_context ioctl_session;
5594 struct md_device *next = mdt->mdt_child;
5598 rc = lu_context_init(&ioctl_session, LCT_SERVER_SESSION);
5601 ioctl_session.lc_thread = (struct ptlrpc_thread *)current;
5602 lu_context_enter(&ioctl_session);
5603 env->le_ses = &ioctl_session;
5605 LASSERT(next->md_ops->mdo_iocontrol);
5606 rc = next->md_ops->mdo_iocontrol(env, next, cmd, len, data);
5608 lu_context_exit(&ioctl_session);
5609 lu_context_fini(&ioctl_session);
5613 static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
5615 struct obd_ioctl_data *data = karg;
5618 struct mdt_object *obj;
5619 struct mdt_lock_handle *lh;
5623 if (data->ioc_inlbuf1 == NULL || data->ioc_inllen1 != sizeof(*fid) ||
5624 data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
5627 fid = (struct lu_fid *)data->ioc_inlbuf1;
5629 if (!fid_is_sane(fid))
5632 CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
5634 lh = &mti->mti_lh[MDT_LH_PARENT];
5635 mdt_lock_reg_init(lh, LCK_CR);
5637 obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE);
5639 RETURN(PTR_ERR(obj));
5641 if (mdt_object_remote(obj)) {
5644 * before calling version get the correct MDS should be
5645 * fid, this is error to find remote object here
5647 CERROR("nonlocal object "DFID"\n", PFID(fid));
5648 } else if (!mdt_object_exists(obj)) {
5649 *(__u64 *)data->ioc_inlbuf2 = ENOENT_VERSION;
5652 version = dt_version_get(mti->mti_env, mdt_obj2dt(obj));
5653 *(__u64 *)data->ioc_inlbuf2 = version;
5656 mdt_object_unlock_put(mti, obj, lh, 1);
5660 /* ioctls on obd dev */
5661 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
5662 void *karg, void *uarg)
5665 struct obd_device *obd = exp->exp_obd;
5666 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
5667 struct dt_device *dt = mdt->mdt_bottom;
5671 CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
5672 rc = lu_env_init(&env, LCT_MD_THREAD);
5678 rc = mdt_device_sync(&env, mdt);
5680 case OBD_IOC_SET_READONLY:
5681 rc = dt->dd_ops->dt_ro(&env, dt);
5683 case OBD_IOC_ABORT_RECOVERY:
5684 CERROR("%s: Aborting recovery for device\n", mdt_obd_name(mdt));
5685 target_stop_recovery_thread(obd);
5688 case OBD_IOC_CHANGELOG_REG:
5689 case OBD_IOC_CHANGELOG_DEREG:
5690 case OBD_IOC_CHANGELOG_CLEAR:
5691 rc = mdt_ioc_child(&env, mdt, cmd, len, karg);
5693 case OBD_IOC_START_LFSCK: {
5694 struct md_device *next = mdt->mdt_child;
5695 struct obd_ioctl_data *data = karg;
5696 struct lfsck_start_param lsp;
5698 if (unlikely(data == NULL)) {
5703 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
5704 lsp.lsp_index_valid = 0;
5705 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &lsp);
5708 case OBD_IOC_STOP_LFSCK: {
5709 struct md_device *next = mdt->mdt_child;
5710 struct obd_ioctl_data *data = karg;
5711 struct lfsck_stop stop;
5713 stop.ls_status = LS_STOPPED;
5714 /* Old lfsck utils may pass NULL @stop. */
5715 if (data->ioc_inlbuf1 == NULL)
5719 ((struct lfsck_stop *)(data->ioc_inlbuf1))->ls_flags;
5721 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &stop);
5724 case OBD_IOC_GET_OBJ_VERSION: {
5725 struct mdt_thread_info *mti;
5726 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5727 memset(mti, 0, sizeof *mti);
5728 mti->mti_env = &env;
5732 rc = mdt_ioc_version_get(mti, karg);
5735 case OBD_IOC_CATLOGLIST: {
5736 struct mdt_thread_info *mti;
5738 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5739 lu_local_obj_fid(&mti->mti_tmp_fid1, LLOG_CATALOGS_OID);
5740 rc = llog_catalog_list(&env, mdt->mdt_bottom, 0, karg,
5741 &mti->mti_tmp_fid1);
5746 CERROR("%s: Not supported cmd = %d, rc = %d\n",
5747 mdt_obd_name(mdt), cmd, rc);
5754 static int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
5756 struct lu_device *ld = md2lu_dev(mdt->mdt_child);
5760 rc = ld->ld_ops->ldo_recovery_complete(env, ld);
5764 static int mdt_obd_postrecov(struct obd_device *obd)
5769 rc = lu_env_init(&env, LCT_MD_THREAD);
5772 rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
5777 static struct obd_ops mdt_obd_device_ops = {
5778 .o_owner = THIS_MODULE,
5779 .o_set_info_async = mdt_obd_set_info_async,
5780 .o_connect = mdt_obd_connect,
5781 .o_reconnect = mdt_obd_reconnect,
5782 .o_disconnect = mdt_obd_disconnect,
5783 .o_init_export = mdt_init_export,
5784 .o_destroy_export = mdt_destroy_export,
5785 .o_iocontrol = mdt_iocontrol,
5786 .o_postrecov = mdt_obd_postrecov,
5789 static struct lu_device* mdt_device_fini(const struct lu_env *env,
5790 struct lu_device *d)
5792 struct mdt_device *m = mdt_dev(d);
5799 static struct lu_device *mdt_device_free(const struct lu_env *env,
5800 struct lu_device *d)
5802 struct mdt_device *m = mdt_dev(d);
5805 lu_device_fini(&m->mdt_lu_dev);
5811 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
5812 struct lu_device_type *t,
5813 struct lustre_cfg *cfg)
5815 struct lu_device *l;
5816 struct mdt_device *m;
5823 rc = mdt_init0(env, m, t, cfg);
5825 mdt_device_free(env, l);
5830 l = ERR_PTR(-ENOMEM);
5834 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
5835 LU_KEY_INIT(mdt, struct mdt_thread_info);
5837 static void mdt_key_fini(const struct lu_context *ctx,
5838 struct lu_context_key *key, void* data)
5840 struct mdt_thread_info *info = data;
5842 if (info->mti_big_lmm) {
5843 OBD_FREE_LARGE(info->mti_big_lmm, info->mti_big_lmmsize);
5844 info->mti_big_lmm = NULL;
5845 info->mti_big_lmmsize = 0;
5850 /* context key: mdt_thread_key */
5851 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
5853 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info)
5855 return lu_ucred(info->mti_env);
5858 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info)
5860 return lu_ucred_check(info->mti_env);
5864 * Enable/disable COS (Commit On Sharing).
5866 * Set/Clear the COS flag in mdt options.
5868 * \param mdt mdt device
5869 * \param val 0 disables COS, other values enable COS
5871 void mdt_enable_cos(struct mdt_device *mdt, int val)
5876 mdt->mdt_opts.mo_cos = !!val;
5877 rc = lu_env_init(&env, LCT_LOCAL);
5878 if (unlikely(rc != 0)) {
5879 CWARN("lu_env initialization failed with rc = %d,"
5880 "cannot sync\n", rc);
5883 mdt_device_sync(&env, mdt);
5888 * Check COS (Commit On Sharing) status.
5890 * Return COS flag status.
5892 * \param mdt mdt device
5894 int mdt_cos_is_enabled(struct mdt_device *mdt)
5896 return mdt->mdt_opts.mo_cos != 0;
5899 static struct lu_device_type_operations mdt_device_type_ops = {
5900 .ldto_device_alloc = mdt_device_alloc,
5901 .ldto_device_free = mdt_device_free,
5902 .ldto_device_fini = mdt_device_fini
5905 static struct lu_device_type mdt_device_type = {
5906 .ldt_tags = LU_DEVICE_MD,
5907 .ldt_name = LUSTRE_MDT_NAME,
5908 .ldt_ops = &mdt_device_type_ops,
5909 .ldt_ctx_tags = LCT_MD_THREAD
5912 static int __init mdt_mod_init(void)
5914 struct lprocfs_static_vars lvars;
5917 CLASSERT(sizeof("0x0123456789ABCDEF:0x01234567:0x01234567") ==
5918 FID_NOBRACE_LEN + 1);
5919 CLASSERT(sizeof("[0x0123456789ABCDEF:0x01234567:0x01234567]") ==
5921 rc = lu_kmem_init(mdt_caches);
5925 rc = mds_mod_init();
5929 lprocfs_mdt_init_vars(&lvars);
5930 rc = class_register_type(&mdt_obd_device_ops, NULL, true, NULL,
5931 #ifndef HAVE_ONLY_PROCFS_SEQ
5934 LUSTRE_MDT_NAME, &mdt_device_type);
5939 lu_kmem_fini(mdt_caches);
5946 static void __exit mdt_mod_exit(void)
5948 class_unregister_type(LUSTRE_MDT_NAME);
5950 lu_kmem_fini(mdt_caches);
5953 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
5954 MODULE_DESCRIPTION("Lustre Metadata Target ("LUSTRE_MDT_NAME")");
5955 MODULE_LICENSE("GPL");
5957 cfs_module(mdt, LUSTRE_VERSION_STRING, mdt_mod_init, mdt_mod_exit);