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, 2014, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/mdt/mdt_handler.c
38 * Lustre Metadata Target (mdt) request handler
40 * Author: Peter Braam <braam@clusterfs.com>
41 * Author: Andreas Dilger <adilger@clusterfs.com>
42 * Author: Phil Schwan <phil@clusterfs.com>
43 * Author: Mike Shaver <shaver@clusterfs.com>
44 * Author: Nikita Danilov <nikita@clusterfs.com>
45 * Author: Huang Hua <huanghua@clusterfs.com>
46 * Author: Yury Umanets <umka@clusterfs.com>
49 #define DEBUG_SUBSYSTEM S_MDS
51 #include <linux/module.h>
53 * struct OBD_{ALLOC,FREE}*()
55 #include <obd_support.h>
56 #include <lustre_ioctl.h>
57 /* struct ptlrpc_request */
58 #include <lustre_net.h>
59 /* struct obd_export */
60 #include <lustre_export.h>
61 /* struct obd_device */
64 #include <dt_object.h>
65 #include <lustre_mds.h>
66 #include <lustre_log.h>
67 #include "mdt_internal.h"
68 #include <lustre_acl.h>
69 #include <lustre_param.h>
70 #include <lustre_quota.h>
71 #include <lustre_lfsck.h>
72 #include <lustre_nodemap.h>
74 mdl_mode_t mdt_mdl_lock_modes[] = {
75 [LCK_MINMODE] = MDL_MINMODE,
82 [LCK_GROUP] = MDL_GROUP
85 ldlm_mode_t mdt_dlm_lock_modes[] = {
86 [MDL_MINMODE] = LCK_MINMODE,
93 [MDL_GROUP] = LCK_GROUP
96 static struct mdt_device *mdt_dev(struct lu_device *d);
97 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags);
99 static const struct lu_object_operations mdt_obj_ops;
101 /* Slab for MDT object allocation */
102 static struct kmem_cache *mdt_object_kmem;
104 /* For HSM restore handles */
105 struct kmem_cache *mdt_hsm_cdt_kmem;
107 /* For HSM request handles */
108 struct kmem_cache *mdt_hsm_car_kmem;
110 static struct lu_kmem_descr mdt_caches[] = {
112 .ckd_cache = &mdt_object_kmem,
113 .ckd_name = "mdt_obj",
114 .ckd_size = sizeof(struct mdt_object)
117 .ckd_cache = &mdt_hsm_cdt_kmem,
118 .ckd_name = "mdt_cdt_restore_handle",
119 .ckd_size = sizeof(struct cdt_restore_handle)
122 .ckd_cache = &mdt_hsm_car_kmem,
123 .ckd_name = "mdt_cdt_agent_req",
124 .ckd_size = sizeof(struct cdt_agent_req)
131 int mdt_get_disposition(struct ldlm_reply *rep, int flag)
135 return (rep->lock_policy_res1 & flag);
138 void mdt_clear_disposition(struct mdt_thread_info *info,
139 struct ldlm_reply *rep, int flag)
142 info->mti_opdata &= ~flag;
143 tgt_opdata_clear(info->mti_env, flag);
146 rep->lock_policy_res1 &= ~flag;
149 void mdt_set_disposition(struct mdt_thread_info *info,
150 struct ldlm_reply *rep, int flag)
153 info->mti_opdata |= flag;
154 tgt_opdata_set(info->mti_env, flag);
157 rep->lock_policy_res1 |= flag;
160 void mdt_lock_reg_init(struct mdt_lock_handle *lh, ldlm_mode_t lm)
162 lh->mlh_pdo_hash = 0;
163 lh->mlh_reg_mode = lm;
164 lh->mlh_rreg_mode = lm;
165 lh->mlh_type = MDT_REG_LOCK;
168 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, ldlm_mode_t lock_mode,
169 const struct lu_name *lname)
171 lh->mlh_reg_mode = lock_mode;
172 lh->mlh_rreg_mode = lock_mode;
173 lh->mlh_type = MDT_PDO_LOCK;
175 if (lu_name_is_valid(lname)) {
176 lh->mlh_pdo_hash = full_name_hash(lname->ln_name,
178 /* XXX Workaround for LU-2856
180 * Zero is a valid return value of full_name_hash, but
181 * several users of mlh_pdo_hash assume a non-zero
182 * hash value. We therefore map zero onto an
183 * arbitrary, but consistent value (1) to avoid
184 * problems further down the road. */
185 if (unlikely(lh->mlh_pdo_hash == 0))
186 lh->mlh_pdo_hash = 1;
188 lh->mlh_pdo_hash = 0;
192 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
193 struct mdt_lock_handle *lh)
199 * Any dir access needs couple of locks:
201 * 1) on part of dir we gonna take lookup/modify;
203 * 2) on whole dir to protect it from concurrent splitting and/or to
204 * flush client's cache for readdir().
206 * so, for a given mode and object this routine decides what lock mode
207 * to use for lock #2:
209 * 1) if caller's gonna lookup in dir then we need to protect dir from
210 * being splitted only - LCK_CR
212 * 2) if caller's gonna modify dir then we need to protect dir from
213 * being splitted and to flush cache - LCK_CW
215 * 3) if caller's gonna modify dir and that dir seems ready for
216 * splitting then we need to protect it from any type of access
217 * (lookup/modify/split) - LCK_EX --bzzz
220 LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
221 LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
224 * Ask underlaying level its opinion about preferable PDO lock mode
225 * having access type passed as regular lock mode:
227 * - MDL_MINMODE means that lower layer does not want to specify lock
230 * - MDL_NL means that no PDO lock should be taken. This is used in some
231 * cases. Say, for non-splittable directories no need to use PDO locks
234 mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
235 mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
237 if (mode != MDL_MINMODE) {
238 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
241 * Lower layer does not want to specify locking mode. We do it
242 * our selves. No special protection is needed, just flush
243 * client's cache on modification and allow concurrent
246 switch (lh->mlh_reg_mode) {
248 lh->mlh_pdo_mode = LCK_EX;
251 lh->mlh_pdo_mode = LCK_CR;
254 lh->mlh_pdo_mode = LCK_CW;
257 CERROR("Not expected lock type (0x%x)\n",
258 (int)lh->mlh_reg_mode);
263 LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
267 static int mdt_getstatus(struct tgt_session_info *tsi)
269 struct mdt_thread_info *info = tsi2mdt_info(tsi);
270 struct mdt_device *mdt = info->mti_mdt;
271 struct mdt_body *repbody;
276 rc = mdt_check_ucred(info);
278 GOTO(out, rc = err_serious(rc));
280 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETSTATUS_PACK))
281 GOTO(out, rc = err_serious(-ENOMEM));
283 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
284 repbody->mbo_fid1 = mdt->mdt_md_root_fid;
285 repbody->mbo_valid |= OBD_MD_FLID;
287 if (tsi->tsi_tgt->lut_mds_capa &&
288 exp_connect_flags(info->mti_exp) & OBD_CONNECT_MDS_CAPA) {
289 struct mdt_object *root;
290 struct lustre_capa *capa;
292 root = mdt_object_find(info->mti_env, mdt, &repbody->mbo_fid1);
294 GOTO(out, rc = PTR_ERR(root));
296 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA1);
298 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
299 rc = mo_capa_get(info->mti_env, mdt_object_child(root), capa,
301 mdt_object_put(info->mti_env, root);
303 repbody->mbo_valid |= OBD_MD_FLMDSCAPA;
307 mdt_thread_info_fini(info);
311 static int mdt_statfs(struct tgt_session_info *tsi)
313 struct ptlrpc_request *req = tgt_ses_req(tsi);
314 struct mdt_thread_info *info = tsi2mdt_info(tsi);
315 struct md_device *next = info->mti_mdt->mdt_child;
316 struct ptlrpc_service_part *svcpt;
317 struct obd_statfs *osfs;
322 svcpt = req->rq_rqbd->rqbd_svcpt;
324 /* This will trigger a watchdog timeout */
325 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
326 (MDT_SERVICE_WATCHDOG_FACTOR *
327 at_get(&svcpt->scp_at_estimate)) + 1);
329 rc = mdt_check_ucred(info);
331 GOTO(out, rc = err_serious(rc));
333 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK))
334 GOTO(out, rc = err_serious(-ENOMEM));
336 osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
338 GOTO(out, rc = -EPROTO);
340 /** statfs information are cached in the mdt_device */
341 if (cfs_time_before_64(info->mti_mdt->mdt_osfs_age,
342 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS))) {
343 /** statfs data is too old, get up-to-date one */
344 rc = next->md_ops->mdo_statfs(info->mti_env, next, osfs);
347 spin_lock(&info->mti_mdt->mdt_osfs_lock);
348 info->mti_mdt->mdt_osfs = *osfs;
349 info->mti_mdt->mdt_osfs_age = cfs_time_current_64();
350 spin_unlock(&info->mti_mdt->mdt_osfs_lock);
352 /** use cached statfs data */
353 spin_lock(&info->mti_mdt->mdt_osfs_lock);
354 *osfs = info->mti_mdt->mdt_osfs;
355 spin_unlock(&info->mti_mdt->mdt_osfs_lock);
359 mdt_counter_incr(req, LPROC_MDT_STATFS);
361 mdt_thread_info_fini(info);
366 * Pack SOM attributes into the reply.
367 * Call under a DLM UPDATE lock.
369 static void mdt_pack_size2body(struct mdt_thread_info *info,
370 struct mdt_object *mo)
373 struct md_attr *ma = &info->mti_attr;
375 LASSERT(ma->ma_attr.la_valid & LA_MODE);
376 b = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
378 /* Check if Size-on-MDS is supported, if this is a regular file,
379 * if SOM is enabled on the object and if SOM cache exists and valid.
380 * Otherwise do not pack Size-on-MDS attributes to the reply. */
381 if (!(mdt_conn_flags(info) & OBD_CONNECT_SOM) ||
382 !S_ISREG(ma->ma_attr.la_mode) ||
383 !mdt_object_is_som_enabled(mo) ||
384 !(ma->ma_valid & MA_SOM))
387 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
388 b->mbo_size = ma->ma_som->msd_size;
389 b->mbo_blocks = ma->ma_som->msd_blocks;
392 #ifdef CONFIG_FS_POSIX_ACL
394 * Pack ACL data into the reply. UIDs/GIDs are mapped and filtered by nodemap.
396 * \param info thread info object
397 * \param repbody reply to pack ACLs into
398 * \param o mdt object of file to examine
399 * \param nodemap nodemap of client to reply to
401 * \retval -errno error getting or parsing ACL from disk
403 int mdt_pack_acl2body(struct mdt_thread_info *info, struct mdt_body *repbody,
404 struct mdt_object *o, struct lu_nodemap *nodemap)
406 const struct lu_env *env = info->mti_env;
407 struct md_object *next = mdt_object_child(o);
408 struct lu_buf *buf = &info->mti_buf;
411 buf->lb_buf = req_capsule_server_get(info->mti_pill, &RMF_ACL);
412 buf->lb_len = req_capsule_get_size(info->mti_pill, &RMF_ACL,
414 if (buf->lb_len == 0)
417 rc = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_ACCESS);
419 if (rc == -ENODATA) {
420 repbody->mbo_aclsize = 0;
421 repbody->mbo_valid |= OBD_MD_FLACL;
423 } else if (rc == -EOPNOTSUPP) {
426 CERROR("%s: unable to read "DFID" ACL: rc = %d\n",
427 mdt_obd_name(info->mti_mdt),
428 PFID(mdt_object_fid(o)), rc);
431 rc = nodemap_map_acl(nodemap, buf->lb_buf,
432 rc, NODEMAP_FS_TO_CLIENT);
433 /* if all ACLs mapped out, rc is still >= 0 */
435 CERROR("%s: nodemap_map_acl unable to parse "DFID
436 " ACL: rc = %d\n", mdt_obd_name(info->mti_mdt),
437 PFID(mdt_object_fid(o)), rc);
439 repbody->mbo_aclsize = rc;
440 repbody->mbo_valid |= OBD_MD_FLACL;
448 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
449 const struct lu_attr *attr, const struct lu_fid *fid)
451 struct md_attr *ma = &info->mti_attr;
452 struct obd_export *exp = info->mti_exp;
453 struct lu_nodemap *nodemap = exp->exp_target_data.ted_nodemap;
455 LASSERT(ma->ma_valid & MA_INODE);
457 b->mbo_atime = attr->la_atime;
458 b->mbo_mtime = attr->la_mtime;
459 b->mbo_ctime = attr->la_ctime;
460 b->mbo_mode = attr->la_mode;
461 b->mbo_size = attr->la_size;
462 b->mbo_blocks = attr->la_blocks;
463 b->mbo_uid = nodemap_map_id(nodemap, NODEMAP_UID,
464 NODEMAP_FS_TO_CLIENT,
466 b->mbo_gid = nodemap_map_id(nodemap, NODEMAP_GID,
467 NODEMAP_FS_TO_CLIENT,
469 b->mbo_flags = attr->la_flags;
470 b->mbo_nlink = attr->la_nlink;
471 b->mbo_rdev = attr->la_rdev;
473 /* XXX should pack the reply body according to lu_valid */
474 b->mbo_valid |= OBD_MD_FLCTIME | OBD_MD_FLUID |
475 OBD_MD_FLGID | OBD_MD_FLTYPE |
476 OBD_MD_FLMODE | OBD_MD_FLNLINK | OBD_MD_FLFLAGS |
477 OBD_MD_FLATIME | OBD_MD_FLMTIME ;
479 if (!S_ISREG(attr->la_mode)) {
480 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
481 } else if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV)) {
482 /* means no objects are allocated on osts. */
483 LASSERT(!(ma->ma_valid & MA_LOV));
484 /* just ignore blocks occupied by extend attributes on MDS */
486 /* if no object is allocated on osts, the size on mds is valid.
488 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
489 } else if ((ma->ma_valid & MA_LOV) && ma->ma_lmm != NULL &&
490 ma->ma_lmm->lmm_pattern & LOV_PATTERN_F_RELEASED) {
491 /* A released file stores its size on MDS. */
492 /* But return 1 block for released file, unless tools like tar
493 * will consider it fully sparse. (LU-3864)
495 if (unlikely(b->mbo_size == 0))
499 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
504 b->mbo_valid |= OBD_MD_FLID;
505 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, size="LPU64"\n",
506 PFID(fid), b->mbo_nlink,
507 b->mbo_mode, b->mbo_size);
511 mdt_body_reverse_idmap(info, b);
513 if (fid != NULL && (b->mbo_valid & OBD_MD_FLSIZE))
514 CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
515 PFID(fid), (unsigned long long)b->mbo_size);
518 static inline int mdt_body_has_lov(const struct lu_attr *la,
519 const struct mdt_body *body)
521 return (S_ISREG(la->la_mode) && (body->mbo_valid & OBD_MD_FLEASIZE)) ||
522 (S_ISDIR(la->la_mode) && (body->mbo_valid & OBD_MD_FLDIREA));
525 void mdt_client_compatibility(struct mdt_thread_info *info)
527 struct mdt_body *body;
528 struct ptlrpc_request *req = mdt_info_req(info);
529 struct obd_export *exp = req->rq_export;
530 struct md_attr *ma = &info->mti_attr;
531 struct lu_attr *la = &ma->ma_attr;
534 if (exp_connect_layout(exp))
535 /* the client can deal with 16-bit lmm_stripe_count */
538 body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
540 if (!mdt_body_has_lov(la, body))
543 /* now we have a reply with a lov for a client not compatible with the
544 * layout lock so we have to clean the layout generation number */
545 if (S_ISREG(la->la_mode))
546 ma->ma_lmm->lmm_layout_gen = 0;
550 int mdt_attr_get_eabuf_size(struct mdt_thread_info *info, struct mdt_object *o)
552 const struct lu_env *env = info->mti_env;
555 rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
564 /* Is it a directory? Let's check for the LMV as well */
565 if (S_ISDIR(lu_object_attr(&mdt_object_child(o)->mo_lu))) {
566 rc2 = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
570 rc2 = mo_xattr_get(env, mdt_object_child(o),
572 XATTR_NAME_DEFAULT_LMV);
574 if ((rc2 < 0 && rc2 != -ENODATA) || (rc2 > rc))
582 int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
585 const struct lu_env *env = info->mti_env;
589 LASSERT(info->mti_big_lmm_used == 0);
590 rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL, name);
594 /* big_lmm may need to be grown */
595 if (info->mti_big_lmmsize < rc) {
596 int size = size_roundup_power2(rc);
598 if (info->mti_big_lmmsize > 0) {
599 /* free old buffer */
600 LASSERT(info->mti_big_lmm);
601 OBD_FREE_LARGE(info->mti_big_lmm,
602 info->mti_big_lmmsize);
603 info->mti_big_lmm = NULL;
604 info->mti_big_lmmsize = 0;
607 OBD_ALLOC_LARGE(info->mti_big_lmm, size);
608 if (info->mti_big_lmm == NULL)
610 info->mti_big_lmmsize = size;
612 LASSERT(info->mti_big_lmmsize >= rc);
614 info->mti_buf.lb_buf = info->mti_big_lmm;
615 info->mti_buf.lb_len = info->mti_big_lmmsize;
616 rc = mo_xattr_get(env, mdt_object_child(o), &info->mti_buf, name);
621 int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
622 struct md_attr *ma, const char *name)
624 struct md_object *next = mdt_object_child(o);
625 struct lu_buf *buf = &info->mti_buf;
628 if (strcmp(name, XATTR_NAME_LOV) == 0) {
629 buf->lb_buf = ma->ma_lmm;
630 buf->lb_len = ma->ma_lmm_size;
631 LASSERT(!(ma->ma_valid & MA_LOV));
632 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
633 buf->lb_buf = ma->ma_lmv;
634 buf->lb_len = ma->ma_lmv_size;
635 LASSERT(!(ma->ma_valid & MA_LMV));
636 } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
637 buf->lb_buf = ma->ma_lmv;
638 buf->lb_len = ma->ma_lmv_size;
639 LASSERT(!(ma->ma_valid & MA_LMV_DEF));
644 rc = mo_xattr_get(info->mti_env, next, buf, name);
648 if (strcmp(name, XATTR_NAME_LOV) == 0) {
649 if (info->mti_big_lmm_used)
650 ma->ma_lmm = info->mti_big_lmm;
652 /* NOT return LOV EA with hole to old client. */
653 if (unlikely(le32_to_cpu(ma->ma_lmm->lmm_pattern) &
654 LOV_PATTERN_F_HOLE) &&
655 !(exp_connect_flags(info->mti_exp) &
656 OBD_CONNECT_LFSCK)) {
659 ma->ma_lmm_size = rc;
660 ma->ma_valid |= MA_LOV;
662 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
663 if (info->mti_big_lmm_used)
664 ma->ma_lmv = info->mti_big_lmm;
666 ma->ma_lmv_size = rc;
667 ma->ma_valid |= MA_LMV;
668 } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
669 ma->ma_lmv_size = rc;
670 ma->ma_valid |= MA_LMV_DEF;
673 /* Update mdt_max_mdsize so all clients will be aware that */
674 if (info->mti_mdt->mdt_max_mdsize < rc)
675 info->mti_mdt->mdt_max_mdsize = rc;
678 } else if (rc == -ENODATA) {
681 } else if (rc == -ERANGE) {
682 /* Default LMV has fixed size, so it must be able to fit
683 * in the original buffer */
684 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
686 rc = mdt_big_xattr_get(info, o, name);
688 info->mti_big_lmm_used = 1;
696 static int mdt_attr_get_pfid(struct mdt_thread_info *info,
697 struct mdt_object *o, struct lu_fid *pfid)
699 struct lu_buf *buf = &info->mti_buf;
700 struct link_ea_header *leh;
701 struct link_ea_entry *lee;
705 buf->lb_buf = info->mti_big_lmm;
706 buf->lb_len = info->mti_big_lmmsize;
707 rc = mo_xattr_get(info->mti_env, mdt_object_child(o),
708 buf, XATTR_NAME_LINK);
709 /* ignore errors, MA_PFID won't be set and it is
710 * up to the caller to treat this as an error */
711 if (rc == -ERANGE || buf->lb_len == 0) {
712 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
713 buf->lb_buf = info->mti_big_lmm;
714 buf->lb_len = info->mti_big_lmmsize;
719 if (rc < sizeof(*leh)) {
720 CERROR("short LinkEA on "DFID": rc = %d\n",
721 PFID(mdt_object_fid(o)), rc);
725 leh = (struct link_ea_header *) buf->lb_buf;
726 lee = (struct link_ea_entry *)(leh + 1);
727 if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
728 leh->leh_magic = LINK_EA_MAGIC;
729 leh->leh_reccount = __swab32(leh->leh_reccount);
730 leh->leh_len = __swab64(leh->leh_len);
732 if (leh->leh_magic != LINK_EA_MAGIC)
734 if (leh->leh_reccount == 0)
737 memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
738 fid_be_to_cpu(pfid, pfid);
743 int mdt_attr_get_complex(struct mdt_thread_info *info,
744 struct mdt_object *o, struct md_attr *ma)
746 const struct lu_env *env = info->mti_env;
747 struct md_object *next = mdt_object_child(o);
748 struct lu_buf *buf = &info->mti_buf;
749 int need = ma->ma_need;
756 if (mdt_object_exists(o) == 0)
757 GOTO(out, rc = -ENOENT);
758 mode = lu_object_attr(&next->mo_lu);
760 if (need & MA_INODE) {
761 ma->ma_need = MA_INODE;
762 rc = mo_attr_get(env, next, ma);
765 ma->ma_valid |= MA_INODE;
768 if (need & MA_PFID) {
769 rc = mdt_attr_get_pfid(info, o, &ma->ma_pfid);
771 ma->ma_valid |= MA_PFID;
772 /* ignore this error, parent fid is not mandatory */
776 if (need & MA_LOV && (S_ISREG(mode) || S_ISDIR(mode))) {
777 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LOV);
782 if (need & MA_LMV && S_ISDIR(mode)) {
783 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LMV);
788 if (need & MA_LMV_DEF && S_ISDIR(mode)) {
789 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_DEFAULT_LMV);
794 if (need & MA_SOM && S_ISREG(mode)) {
795 buf->lb_buf = info->mti_xattr_buf;
796 buf->lb_len = sizeof(info->mti_xattr_buf);
797 CLASSERT(sizeof(struct som_attrs) <=
798 sizeof(info->mti_xattr_buf));
799 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_SOM);
800 rc2 = lustre_buf2som(info->mti_xattr_buf, rc2, ma->ma_som);
802 ma->ma_valid |= MA_SOM;
803 else if (rc2 < 0 && rc2 != -ENODATA)
807 if (need & MA_HSM && S_ISREG(mode)) {
808 buf->lb_buf = info->mti_xattr_buf;
809 buf->lb_len = sizeof(info->mti_xattr_buf);
810 CLASSERT(sizeof(struct hsm_attrs) <=
811 sizeof(info->mti_xattr_buf));
812 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_HSM);
813 rc2 = lustre_buf2hsm(info->mti_xattr_buf, rc2, &ma->ma_hsm);
815 ma->ma_valid |= MA_HSM;
816 else if (rc2 < 0 && rc2 != -ENODATA)
820 #ifdef CONFIG_FS_POSIX_ACL
821 if (need & MA_ACL_DEF && S_ISDIR(mode)) {
822 buf->lb_buf = ma->ma_acl;
823 buf->lb_len = ma->ma_acl_size;
824 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
826 ma->ma_acl_size = rc2;
827 ma->ma_valid |= MA_ACL_DEF;
828 } else if (rc2 == -ENODATA) {
837 CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = "LPX64" ma_lmm=%p\n",
838 rc, ma->ma_valid, ma->ma_lmm);
842 static int mdt_getattr_internal(struct mdt_thread_info *info,
843 struct mdt_object *o, int ma_need)
845 struct md_object *next = mdt_object_child(o);
846 const struct mdt_body *reqbody = info->mti_body;
847 struct ptlrpc_request *req = mdt_info_req(info);
848 struct md_attr *ma = &info->mti_attr;
849 struct lu_attr *la = &ma->ma_attr;
850 struct req_capsule *pill = info->mti_pill;
851 const struct lu_env *env = info->mti_env;
852 struct mdt_body *repbody;
853 struct lu_buf *buffer = &info->mti_buf;
854 struct obd_export *exp = info->mti_exp;
855 struct lu_nodemap *nodemap = exp->exp_target_data.ted_nodemap;
860 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
861 RETURN(err_serious(-ENOMEM));
863 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
867 if (mdt_object_remote(o)) {
868 /* This object is located on remote node.*/
869 /* Return -ENOTSUPP for old client */
870 if (!mdt_is_dne_client(req->rq_export))
871 GOTO(out, rc = -ENOTSUPP);
873 repbody->mbo_fid1 = *mdt_object_fid(o);
874 repbody->mbo_valid = OBD_MD_FLID | OBD_MD_MDS;
878 if (reqbody->mbo_eadatasize > 0) {
879 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
880 if (buffer->lb_buf == NULL)
881 GOTO(out, rc = -EPROTO);
882 buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD,
885 buffer->lb_buf = NULL;
887 ma_need &= ~(MA_LOV | MA_LMV);
888 CDEBUG(D_INFO, "%s: RPC from %s: does not need LOVEA.\n",
889 mdt_obd_name(info->mti_mdt),
890 req->rq_export->exp_client_uuid.uuid);
893 /* If it is dir object and client require MEA, then we got MEA */
894 if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
895 (reqbody->mbo_valid & (OBD_MD_MEA | OBD_MD_DEFAULT_MEA))) {
896 /* Assumption: MDT_MD size is enough for lmv size. */
897 ma->ma_lmv = buffer->lb_buf;
898 ma->ma_lmv_size = buffer->lb_len;
899 ma->ma_need = MA_INODE;
900 if (ma->ma_lmv_size > 0) {
901 if (reqbody->mbo_valid & OBD_MD_MEA)
902 ma->ma_need |= MA_LMV;
903 else if (reqbody->mbo_valid & OBD_MD_DEFAULT_MEA)
904 ma->ma_need |= MA_LMV_DEF;
907 ma->ma_lmm = buffer->lb_buf;
908 ma->ma_lmm_size = buffer->lb_len;
909 ma->ma_need = MA_INODE | MA_HSM;
910 if (ma->ma_lmm_size > 0)
911 ma->ma_need |= MA_LOV;
914 if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
915 reqbody->mbo_valid & OBD_MD_FLDIREA &&
916 lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
917 /* get default stripe info for this dir. */
918 ma->ma_need |= MA_LOV_DEF;
920 ma->ma_need |= ma_need;
921 if (ma->ma_need & MA_SOM)
922 ma->ma_som = &info->mti_u.som.data;
924 rc = mdt_attr_get_complex(info, o, ma);
926 CERROR("%s: getattr error for "DFID": rc = %d\n",
927 mdt_obd_name(info->mti_mdt),
928 PFID(mdt_object_fid(o)), rc);
932 /* if file is released, check if a restore is running */
933 if ((ma->ma_valid & MA_HSM) && (ma->ma_hsm.mh_flags & HS_RELEASED) &&
934 mdt_hsm_restore_is_running(info, mdt_object_fid(o))) {
935 repbody->mbo_t_state = MS_RESTORE;
936 repbody->mbo_valid |= OBD_MD_TSTATE;
939 is_root = lu_fid_eq(mdt_object_fid(o), &info->mti_mdt->mdt_md_root_fid);
941 /* the Lustre protocol supposes to return default striping
942 * on the user-visible root if explicitly requested */
943 if ((ma->ma_valid & MA_LOV) == 0 && S_ISDIR(la->la_mode) &&
944 (ma->ma_need & MA_LOV_DEF && is_root) && ma->ma_need & MA_LOV) {
945 struct lu_fid rootfid;
946 struct mdt_object *root;
947 struct mdt_device *mdt = info->mti_mdt;
949 rc = dt_root_get(env, mdt->mdt_bottom, &rootfid);
952 root = mdt_object_find(env, mdt, &rootfid);
954 RETURN(PTR_ERR(root));
955 rc = mdt_stripe_get(info, root, ma, XATTR_NAME_LOV);
956 mdt_object_put(info->mti_env, root);
958 CERROR("%s: getattr error for "DFID": rc = %d\n",
959 mdt_obd_name(info->mti_mdt),
960 PFID(mdt_object_fid(o)), rc);
965 if (likely(ma->ma_valid & MA_INODE))
966 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
970 if (mdt_body_has_lov(la, reqbody)) {
971 if (ma->ma_valid & MA_LOV) {
972 LASSERT(ma->ma_lmm_size);
973 repbody->mbo_eadatasize = ma->ma_lmm_size;
974 if (S_ISDIR(la->la_mode))
975 repbody->mbo_valid |= OBD_MD_FLDIREA;
977 repbody->mbo_valid |= OBD_MD_FLEASIZE;
978 mdt_dump_lmm(D_INFO, ma->ma_lmm, repbody->mbo_valid);
980 if (ma->ma_valid & MA_LMV) {
981 /* Return -ENOTSUPP for old client */
982 if (!mdt_is_striped_client(req->rq_export))
985 LASSERT(S_ISDIR(la->la_mode));
986 mdt_dump_lmv(D_INFO, ma->ma_lmv);
987 repbody->mbo_eadatasize = ma->ma_lmv_size;
988 repbody->mbo_valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
990 if (ma->ma_valid & MA_LMV_DEF) {
991 /* Return -ENOTSUPP for old client */
992 if (!mdt_is_striped_client(req->rq_export))
994 LASSERT(S_ISDIR(la->la_mode));
995 repbody->mbo_eadatasize = ma->ma_lmv_size;
996 repbody->mbo_valid |= (OBD_MD_FLDIREA |
999 } else if (S_ISLNK(la->la_mode) &&
1000 reqbody->mbo_valid & OBD_MD_LINKNAME) {
1001 buffer->lb_buf = ma->ma_lmm;
1002 /* eadatasize from client includes NULL-terminator, so
1003 * there is no need to read it */
1004 buffer->lb_len = reqbody->mbo_eadatasize - 1;
1005 rc = mo_readlink(env, next, buffer);
1006 if (unlikely(rc <= 0)) {
1007 CERROR("%s: readlink failed for "DFID": rc = %d\n",
1008 mdt_obd_name(info->mti_mdt),
1009 PFID(mdt_object_fid(o)), rc);
1012 int print_limit = min_t(int, PAGE_CACHE_SIZE - 128, rc);
1014 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
1016 repbody->mbo_valid |= OBD_MD_LINKNAME;
1017 /* we need to report back size with NULL-terminator
1018 * because client expects that */
1019 repbody->mbo_eadatasize = rc + 1;
1020 if (repbody->mbo_eadatasize != reqbody->mbo_eadatasize)
1021 CDEBUG(D_INODE, "%s: Read shorter symlink %d "
1022 "on "DFID ", expected %d\n",
1023 mdt_obd_name(info->mti_mdt),
1024 rc, PFID(mdt_object_fid(o)),
1025 reqbody->mbo_eadatasize - 1);
1026 /* NULL terminate */
1027 ((char *)ma->ma_lmm)[rc] = 0;
1029 /* If the total CDEBUG() size is larger than a page, it
1030 * will print a warning to the console, avoid this by
1031 * printing just the last part of the symlink. */
1032 CDEBUG(D_INODE, "symlink dest %s%.*s, len = %d\n",
1033 print_limit < rc ? "..." : "", print_limit,
1034 (char *)ma->ma_lmm + rc - print_limit, rc);
1039 if (reqbody->mbo_valid & OBD_MD_FLMODEASIZE) {
1040 repbody->mbo_max_cookiesize = 0;
1041 repbody->mbo_max_mdsize = info->mti_mdt->mdt_max_mdsize;
1042 repbody->mbo_valid |= OBD_MD_FLMODEASIZE;
1043 CDEBUG(D_INODE, "I am going to change the MAX_MD_SIZE & "
1044 "MAX_COOKIE to : %d:%d\n", repbody->mbo_max_mdsize,
1045 repbody->mbo_max_cookiesize);
1048 if (exp_connect_rmtclient(info->mti_exp) &&
1049 reqbody->mbo_valid & OBD_MD_FLRMTPERM) {
1050 void *buf = req_capsule_server_get(pill, &RMF_ACL);
1052 /* mdt_getattr_lock only */
1053 rc = mdt_pack_remote_perm(info, o, buf);
1055 repbody->mbo_valid &= ~OBD_MD_FLRMTPERM;
1056 repbody->mbo_aclsize = 0;
1059 repbody->mbo_valid |= OBD_MD_FLRMTPERM;
1060 repbody->mbo_aclsize = sizeof(struct mdt_remote_perm);
1063 #ifdef CONFIG_FS_POSIX_ACL
1064 else if ((exp_connect_flags(req->rq_export) & OBD_CONNECT_ACL) &&
1065 (reqbody->mbo_valid & OBD_MD_FLACL))
1066 rc = mdt_pack_acl2body(info, repbody, o, nodemap);
1069 if (reqbody->mbo_valid & OBD_MD_FLMDSCAPA &&
1070 info->mti_mdt->mdt_lut.lut_mds_capa &&
1071 exp_connect_flags(info->mti_exp) & OBD_CONNECT_MDS_CAPA) {
1072 struct lustre_capa *capa;
1074 capa = req_capsule_server_get(pill, &RMF_CAPA1);
1076 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
1077 rc = mo_capa_get(env, next, capa, 0);
1080 repbody->mbo_valid |= OBD_MD_FLMDSCAPA;
1085 mdt_counter_incr(req, LPROC_MDT_GETATTR);
1090 static int mdt_renew_capa(struct mdt_thread_info *info)
1092 struct mdt_object *obj = info->mti_object;
1093 struct mdt_body *body;
1094 struct lustre_capa *capa, *c;
1098 /* if object doesn't exist, or server has disabled capability,
1099 * return directly, client will find body->valid OBD_MD_FLOSSCAPA
1102 if (!obj || !info->mti_mdt->mdt_lut.lut_oss_capa ||
1103 !(exp_connect_flags(info->mti_exp) & OBD_CONNECT_OSS_CAPA))
1106 body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1107 LASSERT(body != NULL);
1109 c = req_capsule_client_get(info->mti_pill, &RMF_CAPA1);
1112 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
1116 rc = mo_capa_get(info->mti_env, mdt_object_child(obj), capa, 1);
1118 body->mbo_valid |= OBD_MD_FLOSSCAPA;
1122 static int mdt_getattr(struct tgt_session_info *tsi)
1124 struct mdt_thread_info *info = tsi2mdt_info(tsi);
1125 struct mdt_object *obj = info->mti_object;
1126 struct req_capsule *pill = info->mti_pill;
1127 struct mdt_body *reqbody;
1128 struct mdt_body *repbody;
1132 reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
1135 if (reqbody->mbo_valid & OBD_MD_FLOSSCAPA) {
1136 rc = req_capsule_server_pack(pill);
1138 RETURN(err_serious(rc));
1139 rc = mdt_renew_capa(info);
1140 GOTO(out_shrink, rc);
1143 LASSERT(obj != NULL);
1144 LASSERT(lu_object_assert_exists(&obj->mot_obj));
1146 /* Unlike intent case where we need to pre-fill out buffers early on
1147 * in intent policy for ldlm reasons, here we can have a much better
1148 * guess at EA size by just reading it from disk.
1149 * Exceptions are readdir and (missing) directory striping */
1151 if (reqbody->mbo_valid & OBD_MD_LINKNAME) {
1152 /* No easy way to know how long is the symlink, but it cannot
1153 * be more than PATH_MAX, so we allocate +1 */
1156 /* A special case for fs ROOT: getattr there might fetch
1157 * default EA for entire fs, not just for this dir!
1159 } else if (lu_fid_eq(mdt_object_fid(obj),
1160 &info->mti_mdt->mdt_md_root_fid) &&
1161 (reqbody->mbo_valid & OBD_MD_FLDIREA) &&
1162 (lustre_msg_get_opc(mdt_info_req(info)->rq_reqmsg) ==
1164 /* Should the default strping be bigger, mdt_fix_reply
1165 * will reallocate */
1166 rc = DEF_REP_MD_SIZE;
1168 /* Read the actual EA size from disk */
1169 rc = mdt_attr_get_eabuf_size(info, obj);
1173 GOTO(out, rc = err_serious(rc));
1175 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, rc);
1177 rc = req_capsule_server_pack(pill);
1178 if (unlikely(rc != 0))
1179 GOTO(out, rc = err_serious(rc));
1181 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1182 LASSERT(repbody != NULL);
1183 repbody->mbo_eadatasize = 0;
1184 repbody->mbo_aclsize = 0;
1186 if (reqbody->mbo_valid & OBD_MD_FLRMTPERM)
1187 rc = mdt_init_ucred(info, reqbody);
1189 rc = mdt_check_ucred(info);
1191 GOTO(out_shrink, rc);
1193 info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1196 * Don't check capability at all, because rename might getattr for
1197 * remote obj, and at that time no capability is available.
1199 mdt_set_capainfo(info, 1, &reqbody->mbo_fid1, BYPASS_CAPA);
1200 rc = mdt_getattr_internal(info, obj, 0);
1201 if (reqbody->mbo_valid & OBD_MD_FLRMTPERM)
1202 mdt_exit_ucred(info);
1205 mdt_client_compatibility(info);
1206 rc2 = mdt_fix_reply(info);
1210 mdt_thread_info_fini(info);
1214 static int mdt_swap_layouts(struct tgt_session_info *tsi)
1216 struct mdt_thread_info *info;
1217 struct ptlrpc_request *req = tgt_ses_req(tsi);
1218 struct obd_export *exp = req->rq_export;
1219 struct mdt_object *o1, *o2, *o;
1220 struct mdt_lock_handle *lh1, *lh2;
1221 struct mdc_swap_layouts *msl;
1225 /* client does not support layout lock, so layout swaping
1227 * FIXME: there is a problem for old clients which don't support
1228 * layout lock yet. If those clients have already opened the file
1229 * they won't be notified at all so that old layout may still be
1230 * used to do IO. This can be fixed after file release is landed by
1231 * doing exclusive open and taking full EX ibits lock. - Jinshan */
1232 if (!exp_connect_layout(exp))
1233 RETURN(-EOPNOTSUPP);
1235 info = tsi2mdt_info(tsi);
1237 if (info->mti_dlm_req != NULL)
1238 ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
1240 if (req_capsule_get_size(info->mti_pill, &RMF_CAPA1, RCL_CLIENT))
1241 mdt_set_capainfo(info, 0, &info->mti_body->mbo_fid1,
1242 req_capsule_client_get(info->mti_pill,
1245 if (req_capsule_get_size(info->mti_pill, &RMF_CAPA2, RCL_CLIENT))
1246 mdt_set_capainfo(info, 1, &info->mti_body->mbo_fid2,
1247 req_capsule_client_get(info->mti_pill,
1250 o1 = info->mti_object;
1251 o = o2 = mdt_object_find(info->mti_env, info->mti_mdt,
1252 &info->mti_body->mbo_fid2);
1254 GOTO(out, rc = PTR_ERR(o));
1256 if (mdt_object_remote(o) || !mdt_object_exists(o)) /* remote object */
1257 GOTO(put, rc = -ENOENT);
1259 rc = lu_fid_cmp(&info->mti_body->mbo_fid1, &info->mti_body->mbo_fid2);
1260 if (unlikely(rc == 0)) /* same file, you kidding me? no-op. */
1266 /* permission check. Make sure the calling process having permission
1267 * to write both files. */
1268 rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
1273 rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL,
1278 msl = req_capsule_client_get(info->mti_pill, &RMF_SWAP_LAYOUTS);
1280 GOTO(put, rc = -EPROTO);
1282 lh1 = &info->mti_lh[MDT_LH_NEW];
1283 mdt_lock_reg_init(lh1, LCK_EX);
1284 lh2 = &info->mti_lh[MDT_LH_OLD];
1285 mdt_lock_reg_init(lh2, LCK_EX);
1287 rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT |
1288 MDS_INODELOCK_XATTR, MDT_LOCAL_LOCK);
1292 rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT |
1293 MDS_INODELOCK_XATTR, MDT_LOCAL_LOCK);
1297 rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1),
1298 mdt_object_child(o2), msl->msl_flags);
1301 mdt_object_unlock(info, o2, lh2, rc);
1303 mdt_object_unlock(info, o1, lh1, rc);
1305 mdt_object_put(info->mti_env, o);
1307 mdt_thread_info_fini(info);
1311 static int mdt_raw_lookup(struct mdt_thread_info *info,
1312 struct mdt_object *parent,
1313 const struct lu_name *lname,
1314 struct ldlm_reply *ldlm_rep)
1316 struct lu_fid *child_fid = &info->mti_tmp_fid1;
1320 LASSERT(!info->mti_cross_ref);
1322 /* Only got the fid of this obj by name */
1323 fid_zero(child_fid);
1324 rc = mdo_lookup(info->mti_env, mdt_object_child(info->mti_object),
1325 lname, child_fid, &info->mti_spec);
1327 struct mdt_body *repbody;
1329 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1330 repbody->mbo_fid1 = *child_fid;
1331 repbody->mbo_valid = OBD_MD_FLID;
1332 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1333 } else if (rc == -ENOENT) {
1334 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1341 * UPDATE lock should be taken against parent, and be release before exit;
1342 * child_bits lock should be taken against child, and be returned back:
1343 * (1)normal request should release the child lock;
1344 * (2)intent request will grant the lock to client.
1346 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
1347 struct mdt_lock_handle *lhc,
1349 struct ldlm_reply *ldlm_rep)
1351 struct ptlrpc_request *req = mdt_info_req(info);
1352 struct mdt_body *reqbody = NULL;
1353 struct mdt_object *parent = info->mti_object;
1354 struct mdt_object *child;
1355 struct lu_fid *child_fid = &info->mti_tmp_fid1;
1356 struct lu_name *lname = NULL;
1357 struct mdt_lock_handle *lhp = NULL;
1358 struct ldlm_lock *lock;
1365 is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
1366 LASSERT(ergo(is_resent,
1367 lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
1372 if (info->mti_cross_ref) {
1373 /* Only getattr on the child. Parent is on another node. */
1374 mdt_set_disposition(info, ldlm_rep,
1375 DISP_LOOKUP_EXECD | DISP_LOOKUP_POS);
1377 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
1379 PFID(mdt_object_fid(child)), ldlm_rep);
1381 rc = mdt_check_resent_lock(info, child, lhc);
1384 } else if (rc > 0) {
1385 mdt_lock_handle_init(lhc);
1386 mdt_lock_reg_init(lhc, LCK_PR);
1389 * Object's name is on another MDS, no lookup or layout
1390 * lock is needed here but update lock is.
1392 child_bits &= ~(MDS_INODELOCK_LOOKUP |
1393 MDS_INODELOCK_LAYOUT);
1394 child_bits |= MDS_INODELOCK_PERM | MDS_INODELOCK_UPDATE;
1396 rc = mdt_object_lock(info, child, lhc, child_bits,
1402 /* Finally, we can get attr for child. */
1403 if (!mdt_object_exists(child)) {
1404 LU_OBJECT_DEBUG(D_INFO, info->mti_env,
1406 "remote object doesn't exist.\n");
1407 mdt_object_unlock(info, child, lhc, 1);
1411 mdt_set_capainfo(info, 0, mdt_object_fid(child), BYPASS_CAPA);
1412 rc = mdt_getattr_internal(info, child, 0);
1413 if (unlikely(rc != 0))
1414 mdt_object_unlock(info, child, lhc, 1);
1419 lname = &info->mti_name;
1420 mdt_name_unpack(info->mti_pill, &RMF_NAME, lname, MNF_FIX_ANON);
1422 if (lu_name_is_valid(lname)) {
1423 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DNAME", "
1424 "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
1425 PNAME(lname), ldlm_rep);
1427 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1428 if (unlikely(reqbody == NULL))
1429 RETURN(err_serious(-EPROTO));
1431 *child_fid = reqbody->mbo_fid2;
1433 if (unlikely(!fid_is_sane(child_fid)))
1434 RETURN(err_serious(-EINVAL));
1436 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
1438 PFID(mdt_object_fid(parent)),
1439 PFID(&reqbody->mbo_fid2), ldlm_rep);
1442 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
1444 if (unlikely(!mdt_object_exists(parent)) && lu_name_is_valid(lname)) {
1445 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1447 "Parent doesn't exist!\n");
1451 if (mdt_object_remote(parent)) {
1452 CERROR("%s: parent "DFID" is on remote target\n",
1453 mdt_obd_name(info->mti_mdt),
1454 PFID(mdt_object_fid(parent)));
1458 if (lu_name_is_valid(lname)) {
1459 /* Always allow to lookup ".." */
1460 if (unlikely(lname->ln_namelen == 2 &&
1461 lname->ln_name[0] == '.' &&
1462 lname->ln_name[1] == '.'))
1463 info->mti_spec.sp_permitted = 1;
1465 if (info->mti_body->mbo_valid == OBD_MD_FLID) {
1466 rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
1471 /* step 1: lock parent only if parent is a directory */
1472 if (S_ISDIR(lu_object_attr(&parent->mot_obj))) {
1473 lhp = &info->mti_lh[MDT_LH_PARENT];
1474 mdt_lock_pdo_init(lhp, LCK_PR, lname);
1475 rc = mdt_object_lock(info, parent, lhp,
1476 MDS_INODELOCK_UPDATE,
1478 if (unlikely(rc != 0))
1482 /* step 2: lookup child's fid by name */
1483 fid_zero(child_fid);
1484 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
1485 child_fid, &info->mti_spec);
1487 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1490 GOTO(out_parent, rc);
1493 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1496 *step 3: find the child object by fid & lock it.
1497 * regardless if it is local or remote.
1499 *Note: LU-3240 (commit 762f2114d282a98ebfa4dbbeea9298a8088ad24e)
1500 * set parent dir fid the same as child fid in getattr by fid case
1501 * we should not lu_object_find() the object again, could lead
1502 * to hung if there is a concurrent unlink destroyed the object.
1504 if (lu_fid_eq(mdt_object_fid(parent), child_fid)) {
1505 mdt_object_get(info->mti_env, parent);
1508 child = mdt_object_find(info->mti_env, info->mti_mdt,
1512 if (unlikely(IS_ERR(child)))
1513 GOTO(out_parent, rc = PTR_ERR(child));
1515 rc = mdt_check_resent_lock(info, child, lhc);
1517 GOTO(out_child, rc);
1518 } else if (rc > 0) {
1519 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout*2);
1520 mdt_lock_handle_init(lhc);
1521 mdt_lock_reg_init(lhc, LCK_PR);
1524 if (!mdt_object_exists(child)) {
1525 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1527 "Object doesn't exist!\n");
1528 GOTO(out_child, rc = -ENOENT);
1531 if (!(child_bits & MDS_INODELOCK_UPDATE) &&
1532 mdt_object_exists(child) && !mdt_object_remote(child)) {
1533 struct md_attr *ma = &info->mti_attr;
1536 ma->ma_need = MA_INODE;
1537 rc = mdt_attr_get_complex(info, child, ma);
1538 if (unlikely(rc != 0))
1539 GOTO(out_child, rc);
1541 /* If the file has not been changed for some time, we
1542 * return not only a LOOKUP lock, but also an UPDATE
1543 * lock and this might save us RPC on later STAT. For
1544 * directories, it also let negative dentry cache start
1545 * working for this dir. */
1546 if (ma->ma_valid & MA_INODE &&
1547 ma->ma_attr.la_valid & LA_CTIME &&
1548 info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
1549 ma->ma_attr.la_ctime < cfs_time_current_sec())
1550 child_bits |= MDS_INODELOCK_UPDATE;
1553 /* layout lock must be granted in a best-effort way
1554 * for IT operations */
1555 LASSERT(!(child_bits & MDS_INODELOCK_LAYOUT));
1556 if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) &&
1557 exp_connect_layout(info->mti_exp) &&
1558 S_ISREG(lu_object_attr(&child->mot_obj)) &&
1559 !mdt_object_remote(child) && ldlm_rep != NULL) {
1560 /* try to grant layout lock for regular file. */
1566 child_bits |= MDS_INODELOCK_LAYOUT;
1567 /* try layout lock, it may fail to be granted due to
1568 * contention at LOOKUP or UPDATE */
1569 if (!mdt_object_lock_try(info, child, lhc, child_bits,
1571 child_bits &= ~MDS_INODELOCK_LAYOUT;
1572 LASSERT(child_bits != 0);
1573 rc = mdt_object_lock(info, child, lhc,
1574 child_bits, MDT_CROSS_LOCK);
1579 /* Do not enqueue the UPDATE lock from MDT(cross-MDT),
1580 * client will enqueue the lock to the remote MDT */
1581 if (mdt_object_remote(child))
1582 child_bits &= ~MDS_INODELOCK_UPDATE;
1583 rc = mdt_object_lock(info, child, lhc, child_bits,
1586 if (unlikely(rc != 0))
1587 GOTO(out_child, rc);
1590 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1591 /* Get MA_SOM attributes if update lock is given. */
1593 lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_UPDATE &&
1594 S_ISREG(lu_object_attr(&mdt_object_child(child)->mo_lu)))
1597 /* finally, we can get attr for child. */
1598 mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
1599 rc = mdt_getattr_internal(info, child, ma_need);
1600 if (unlikely(rc != 0)) {
1601 mdt_object_unlock(info, child, lhc, 1);
1603 /* Debugging code. */
1604 LDLM_DEBUG(lock, "Returning lock to client");
1605 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1606 &lock->l_resource->lr_name),
1607 "Lock res_id: "DLDLMRES", fid: "DFID"\n",
1608 PLDLMRES(lock->l_resource),
1609 PFID(mdt_object_fid(child)));
1610 if (mdt_object_exists(child) && !mdt_object_remote(child))
1611 mdt_pack_size2body(info, child);
1614 LDLM_LOCK_PUT(lock);
1618 mdt_object_put(info->mti_env, child);
1621 mdt_object_unlock(info, parent, lhp, 1);
1625 /* normal handler: should release the child lock */
1626 static int mdt_getattr_name(struct tgt_session_info *tsi)
1628 struct mdt_thread_info *info = tsi2mdt_info(tsi);
1629 struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1630 struct mdt_body *reqbody;
1631 struct mdt_body *repbody;
1635 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1636 LASSERT(reqbody != NULL);
1637 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1638 LASSERT(repbody != NULL);
1640 info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1641 repbody->mbo_eadatasize = 0;
1642 repbody->mbo_aclsize = 0;
1644 rc = mdt_init_ucred(info, reqbody);
1646 GOTO(out_shrink, rc);
1648 rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1649 if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1650 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1651 lhc->mlh_reg_lh.cookie = 0;
1653 mdt_exit_ucred(info);
1656 mdt_client_compatibility(info);
1657 rc2 = mdt_fix_reply(info);
1660 mdt_thread_info_fini(info);
1664 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1665 void *karg, void *uarg);
1667 static int mdt_set_info(struct tgt_session_info *tsi)
1669 struct ptlrpc_request *req = tgt_ses_req(tsi);
1672 int keylen, vallen, rc = 0;
1676 key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
1678 DEBUG_REQ(D_HA, req, "no set_info key");
1679 RETURN(err_serious(-EFAULT));
1682 keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
1685 val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
1687 DEBUG_REQ(D_HA, req, "no set_info val");
1688 RETURN(err_serious(-EFAULT));
1691 vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
1694 /* Swab any part of val you need to here */
1695 if (KEY_IS(KEY_READ_ONLY)) {
1696 spin_lock(&req->rq_export->exp_lock);
1698 *exp_connect_flags_ptr(req->rq_export) |=
1701 *exp_connect_flags_ptr(req->rq_export) &=
1702 ~OBD_CONNECT_RDONLY;
1703 spin_unlock(&req->rq_export->exp_lock);
1704 } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1705 struct changelog_setinfo *cs = val;
1707 if (vallen != sizeof(*cs)) {
1708 CERROR("%s: bad changelog_clear setinfo size %d\n",
1709 tgt_name(tsi->tsi_tgt), vallen);
1712 if (ptlrpc_req_need_swab(req)) {
1713 __swab64s(&cs->cs_recno);
1714 __swab32s(&cs->cs_id);
1717 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, req->rq_export,
1725 static int mdt_readpage(struct tgt_session_info *tsi)
1727 struct mdt_thread_info *info = mdt_th_info(tsi->tsi_env);
1728 struct mdt_object *object = mdt_obj(tsi->tsi_corpus);
1729 struct lu_rdpg *rdpg = &info->mti_u.rdpg.mti_rdpg;
1730 const struct mdt_body *reqbody = tsi->tsi_mdt_body;
1731 struct mdt_body *repbody;
1737 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
1738 RETURN(err_serious(-ENOMEM));
1740 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_MDT_BODY);
1741 if (repbody == NULL || reqbody == NULL)
1742 RETURN(err_serious(-EFAULT));
1745 * prepare @rdpg before calling lower layers and transfer itself. Here
1746 * reqbody->size contains offset of where to start to read and
1747 * reqbody->nlink contains number bytes to read.
1749 rdpg->rp_hash = reqbody->mbo_size;
1750 if (rdpg->rp_hash != reqbody->mbo_size) {
1751 CERROR("Invalid hash: "LPX64" != "LPX64"\n",
1752 rdpg->rp_hash, reqbody->mbo_size);
1756 rdpg->rp_attrs = reqbody->mbo_mode;
1757 if (exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_64BITHASH)
1758 rdpg->rp_attrs |= LUDA_64BITHASH;
1759 rdpg->rp_count = min_t(unsigned int, reqbody->mbo_nlink,
1760 exp_max_brw_size(tsi->tsi_exp));
1761 rdpg->rp_npages = (rdpg->rp_count + PAGE_CACHE_SIZE - 1) >>
1763 OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1764 if (rdpg->rp_pages == NULL)
1767 for (i = 0; i < rdpg->rp_npages; ++i) {
1768 rdpg->rp_pages[i] = alloc_page(GFP_IOFS);
1769 if (rdpg->rp_pages[i] == NULL)
1770 GOTO(free_rdpg, rc = -ENOMEM);
1773 /* call lower layers to fill allocated pages with directory data */
1774 rc = mo_readpage(tsi->tsi_env, mdt_object_child(object), rdpg);
1776 GOTO(free_rdpg, rc);
1778 /* send pages to client */
1779 rc = tgt_sendpage(tsi, rdpg, rc);
1784 for (i = 0; i < rdpg->rp_npages; i++)
1785 if (rdpg->rp_pages[i] != NULL)
1786 __free_page(rdpg->rp_pages[i]);
1787 OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1789 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1795 static int mdt_reint_internal(struct mdt_thread_info *info,
1796 struct mdt_lock_handle *lhc,
1799 struct req_capsule *pill = info->mti_pill;
1800 struct mdt_body *repbody;
1805 rc = mdt_reint_unpack(info, op);
1807 CERROR("Can't unpack reint, rc %d\n", rc);
1808 RETURN(err_serious(rc));
1811 /* for replay (no_create) lmm is not needed, client has it already */
1812 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1813 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1816 /* llog cookies are always 0, the field is kept for compatibility */
1817 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1818 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);
1820 rc = req_capsule_server_pack(pill);
1822 CERROR("Can't pack response, rc %d\n", rc);
1823 RETURN(err_serious(rc));
1826 if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
1827 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1829 repbody->mbo_eadatasize = 0;
1830 repbody->mbo_aclsize = 0;
1833 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
1835 /* for replay no cookkie / lmm need, because client have this already */
1836 if (info->mti_spec.no_create)
1837 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1838 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
1840 rc = mdt_init_ucred_reint(info);
1842 GOTO(out_shrink, rc);
1844 rc = mdt_fix_attr_ucred(info, op);
1846 GOTO(out_ucred, rc = err_serious(rc));
1848 if (mdt_check_resent(info, mdt_reconstruct, lhc)) {
1849 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt.");
1850 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
1851 GOTO(out_ucred, rc);
1853 rc = mdt_reint_rec(info, lhc);
1856 mdt_exit_ucred(info);
1858 mdt_client_compatibility(info);
1859 rc2 = mdt_fix_reply(info);
1865 static long mdt_reint_opcode(struct ptlrpc_request *req,
1866 const struct req_format **fmt)
1868 struct mdt_device *mdt;
1869 struct mdt_rec_reint *rec;
1872 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
1874 opc = rec->rr_opcode;
1875 DEBUG_REQ(D_INODE, req, "reint opt = %ld", opc);
1876 if (opc < REINT_MAX && fmt[opc] != NULL)
1877 req_capsule_extend(&req->rq_pill, fmt[opc]);
1879 mdt = mdt_exp2dev(req->rq_export);
1880 CERROR("%s: Unsupported opcode '%ld' from client '%s':"
1881 " rc = %d\n", req->rq_export->exp_obd->obd_name,
1882 opc, mdt->mdt_ldlm_client->cli_name, -EFAULT);
1883 opc = err_serious(-EFAULT);
1886 opc = err_serious(-EFAULT);
1891 static int mdt_reint(struct tgt_session_info *tsi)
1895 static const struct req_format *reint_fmts[REINT_MAX] = {
1896 [REINT_SETATTR] = &RQF_MDS_REINT_SETATTR,
1897 [REINT_CREATE] = &RQF_MDS_REINT_CREATE,
1898 [REINT_LINK] = &RQF_MDS_REINT_LINK,
1899 [REINT_UNLINK] = &RQF_MDS_REINT_UNLINK,
1900 [REINT_RENAME] = &RQF_MDS_REINT_RENAME,
1901 [REINT_OPEN] = &RQF_MDS_REINT_OPEN,
1902 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR,
1903 [REINT_RMENTRY] = &RQF_MDS_REINT_UNLINK,
1904 [REINT_MIGRATE] = &RQF_MDS_REINT_RENAME
1909 opc = mdt_reint_opcode(tgt_ses_req(tsi), reint_fmts);
1911 struct mdt_thread_info *info = tsi2mdt_info(tsi);
1913 * No lock possible here from client to pass it to reint code
1916 rc = mdt_reint_internal(info, NULL, opc);
1917 mdt_thread_info_fini(info);
1922 tsi->tsi_reply_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
1926 /* this should sync the whole device */
1927 static int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
1929 struct dt_device *dt = mdt->mdt_bottom;
1933 rc = dt->dd_ops->dt_sync(env, dt);
1937 /* this should sync this object */
1938 static int mdt_object_sync(struct mdt_thread_info *info)
1940 struct md_object *next;
1944 if (!mdt_object_exists(info->mti_object)) {
1945 CWARN("%s: non existing object "DFID": rc = %d\n",
1946 mdt_obd_name(info->mti_mdt),
1947 PFID(mdt_object_fid(info->mti_object)), -ESTALE);
1950 next = mdt_object_child(info->mti_object);
1951 rc = mo_object_sync(info->mti_env, next);
1956 static int mdt_sync(struct tgt_session_info *tsi)
1958 struct ptlrpc_request *req = tgt_ses_req(tsi);
1959 struct req_capsule *pill = tsi->tsi_pill;
1960 struct mdt_body *body;
1965 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
1966 RETURN(err_serious(-ENOMEM));
1968 if (fid_seq(&tsi->tsi_mdt_body->mbo_fid1) == 0) {
1969 rc = mdt_device_sync(tsi->tsi_env, mdt_exp2dev(tsi->tsi_exp));
1971 struct mdt_thread_info *info = tsi2mdt_info(tsi);
1973 /* sync an object */
1974 rc = mdt_object_sync(info);
1976 const struct lu_fid *fid;
1977 struct lu_attr *la = &info->mti_attr.ma_attr;
1979 info->mti_attr.ma_need = MA_INODE;
1980 info->mti_attr.ma_valid = 0;
1981 rc = mdt_attr_get_complex(info, info->mti_object,
1984 body = req_capsule_server_get(pill,
1986 fid = mdt_object_fid(info->mti_object);
1987 mdt_pack_attr2body(info, body, la, fid);
1990 mdt_thread_info_fini(info);
1993 mdt_counter_incr(req, LPROC_MDT_SYNC);
1999 * Handle quota control requests to consult current usage/limit, but also
2000 * to configure quota enforcement
2002 static int mdt_quotactl(struct tgt_session_info *tsi)
2004 struct obd_export *exp = tsi->tsi_exp;
2005 struct req_capsule *pill = tsi->tsi_pill;
2006 struct obd_quotactl *oqctl, *repoqc;
2008 struct mdt_device *mdt = mdt_exp2dev(exp);
2009 struct lu_device *qmt = mdt->mdt_qmt_dev;
2010 struct lu_nodemap *nodemap = exp->exp_target_data.ted_nodemap;
2013 oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
2015 RETURN(err_serious(-EPROTO));
2017 rc = req_capsule_server_pack(pill);
2019 RETURN(err_serious(rc));
2021 switch (oqctl->qc_cmd) {
2023 case LUSTRE_Q_INVALIDATE:
2024 case LUSTRE_Q_FINVALIDATE:
2028 /* deprecated, not used any more */
2029 RETURN(-EOPNOTSUPP);
2030 /* master quotactl */
2033 if (!nodemap_can_setquota(nodemap))
2038 RETURN(-EOPNOTSUPP);
2039 /* slave quotactl */
2044 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2048 /* map uid/gid for remote client */
2050 if (exp_connect_rmtclient(exp)) {
2051 struct lustre_idmap_table *idmap;
2053 idmap = exp->exp_mdt_data.med_idmap;
2055 if (unlikely(oqctl->qc_cmd != Q_GETQUOTA &&
2056 oqctl->qc_cmd != Q_GETINFO))
2059 if (oqctl->qc_type == USRQUOTA)
2060 id = lustre_idmap_lookup_uid(NULL, idmap, 0,
2062 else if (oqctl->qc_type == GRPQUOTA)
2063 id = lustre_idmap_lookup_gid(NULL, idmap, 0,
2068 if (id == CFS_IDMAP_NOTFOUND) {
2069 CDEBUG(D_QUOTA, "no mapping for id %u\n", oqctl->qc_id);
2074 if (oqctl->qc_type == USRQUOTA)
2075 id = nodemap_map_id(nodemap, NODEMAP_UID,
2076 NODEMAP_CLIENT_TO_FS, id);
2077 else if (oqctl->qc_type == GRPQUOTA)
2078 id = nodemap_map_id(nodemap, NODEMAP_UID,
2079 NODEMAP_CLIENT_TO_FS, id);
2081 repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
2083 RETURN(err_serious(-EFAULT));
2085 if (oqctl->qc_id != id)
2086 swap(oqctl->qc_id, id);
2088 switch (oqctl->qc_cmd) {
2094 /* forward quotactl request to QMT */
2095 rc = qmt_hdls.qmth_quotactl(tsi->tsi_env, qmt, oqctl);
2100 /* slave quotactl */
2101 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom,
2106 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2110 if (oqctl->qc_id != id)
2111 swap(oqctl->qc_id, id);
2117 /** clone llog ctxt from child (mdd)
2118 * This allows remote llog (replicator) access.
2119 * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
2120 * context was originally set up, or we can handle them directly.
2121 * I choose the latter, but that means I need any llog
2122 * contexts set up by child to be accessable by the mdt. So we clone the
2123 * context into our context list here.
2125 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
2128 struct md_device *next = mdt->mdt_child;
2129 struct llog_ctxt *ctxt;
2132 if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
2135 rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
2136 if (rc || ctxt == NULL) {
2140 rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
2142 CERROR("Can't set mdt ctxt %d\n", rc);
2147 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
2148 struct mdt_device *mdt, int idx)
2150 struct llog_ctxt *ctxt;
2152 ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
2155 /* Put once for the get we just did, and once for the clone */
2156 llog_ctxt_put(ctxt);
2157 llog_ctxt_put(ctxt);
2162 * sec context handlers
2164 static int mdt_sec_ctx_handle(struct tgt_session_info *tsi)
2168 rc = mdt_handle_idmap(tsi);
2170 struct ptlrpc_request *req = tgt_ses_req(tsi);
2173 opc = lustre_msg_get_opc(req->rq_reqmsg);
2174 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT)
2175 sptlrpc_svc_ctx_invalidate(req);
2178 CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
2184 * quota request handlers
2186 static int mdt_quota_dqacq(struct tgt_session_info *tsi)
2188 struct mdt_device *mdt = mdt_exp2dev(tsi->tsi_exp);
2189 struct lu_device *qmt = mdt->mdt_qmt_dev;
2194 RETURN(err_serious(-EOPNOTSUPP));
2196 rc = qmt_hdls.qmth_dqacq(tsi->tsi_env, qmt, tgt_ses_req(tsi));
2200 struct mdt_object *mdt_object_new(const struct lu_env *env,
2201 struct mdt_device *d,
2202 const struct lu_fid *f)
2204 struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2205 struct lu_object *o;
2206 struct mdt_object *m;
2209 CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
2210 o = lu_object_find(env, &d->mdt_lu_dev, f, &conf);
2211 if (unlikely(IS_ERR(o)))
2212 m = (struct mdt_object *)o;
2218 struct mdt_object *mdt_object_find(const struct lu_env *env,
2219 struct mdt_device *d,
2220 const struct lu_fid *f)
2222 struct lu_object *o;
2223 struct mdt_object *m;
2226 CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2227 o = lu_object_find(env, &d->mdt_lu_dev, f, NULL);
2228 if (unlikely(IS_ERR(o)))
2229 m = (struct mdt_object *)o;
2237 * Asyncronous commit for mdt device.
2239 * Pass asynchonous commit call down the MDS stack.
2241 * \param env environment
2242 * \param mdt the mdt device
2244 static void mdt_device_commit_async(const struct lu_env *env,
2245 struct mdt_device *mdt)
2247 struct dt_device *dt = mdt->mdt_bottom;
2250 rc = dt->dd_ops->dt_commit_async(env, dt);
2251 if (unlikely(rc != 0))
2252 CWARN("%s: async commit start failed: rc = %d\n",
2253 mdt_obd_name(mdt), rc);
2257 * Mark the lock as "synchonous".
2259 * Mark the lock to deffer transaction commit to the unlock time.
2261 * \param lock the lock to mark as "synchonous"
2263 * \see mdt_is_lock_sync
2264 * \see mdt_save_lock
2266 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2268 lock->l_ast_data = (void*)1;
2272 * Check whehter the lock "synchonous" or not.
2274 * \param lock the lock to check
2275 * \retval 1 the lock is "synchonous"
2276 * \retval 0 the lock isn't "synchronous"
2278 * \see mdt_set_lock_sync
2279 * \see mdt_save_lock
2281 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2283 return lock->l_ast_data != NULL;
2287 * Blocking AST for mdt locks.
2289 * Starts transaction commit if in case of COS lock conflict or
2290 * deffers such a commit to the mdt_save_lock.
2292 * \param lock the lock which blocks a request or cancelling lock
2293 * \param desc unused
2294 * \param data unused
2295 * \param flag indicates whether this cancelling or blocking callback
2297 * \see ldlm_blocking_ast_nocheck
2299 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2300 void *data, int flag)
2302 struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2303 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2307 if (flag == LDLM_CB_CANCELING)
2309 lock_res_and_lock(lock);
2310 if (lock->l_blocking_ast != mdt_blocking_ast) {
2311 unlock_res_and_lock(lock);
2314 if (mdt_cos_is_enabled(mdt) &&
2315 lock->l_req_mode & (LCK_PW | LCK_EX) &&
2316 lock->l_blocking_lock != NULL &&
2317 lock->l_client_cookie != lock->l_blocking_lock->l_client_cookie) {
2318 mdt_set_lock_sync(lock);
2320 rc = ldlm_blocking_ast_nocheck(lock);
2322 /* There is no lock conflict if l_blocking_lock == NULL,
2323 * it indicates a blocking ast sent from ldlm_lock_decref_internal
2324 * when the last reference to a local lock was released */
2325 if (lock->l_req_mode == LCK_COS && lock->l_blocking_lock != NULL) {
2328 rc = lu_env_init(&env, LCT_LOCAL);
2329 if (unlikely(rc != 0))
2330 CWARN("%s: lu_env initialization failed, cannot "
2331 "start asynchronous commit: rc = %d\n",
2334 mdt_device_commit_async(&env, mdt);
2340 /* Used for cross-MDT lock */
2341 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2342 void *data, int flag)
2344 struct lustre_handle lockh;
2348 case LDLM_CB_BLOCKING:
2349 ldlm_lock2handle(lock, &lockh);
2350 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
2352 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
2356 case LDLM_CB_CANCELING:
2357 LDLM_DEBUG(lock, "Revoke remote lock\n");
2365 int mdt_check_resent_lock(struct mdt_thread_info *info,
2366 struct mdt_object *mo,
2367 struct mdt_lock_handle *lhc)
2369 /* the lock might already be gotten in ldlm_handle_enqueue() */
2370 if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
2371 struct ptlrpc_request *req = mdt_info_req(info);
2372 struct ldlm_lock *lock;
2374 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
2375 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT);
2377 /* Lock is pinned by ldlm_handle_enqueue0() as it is
2378 * a resend case, however, it could be already destroyed
2379 * due to client eviction or a raced cancel RPC. */
2380 LDLM_DEBUG_NOLOCK("Invalid lock handle "LPX64"\n",
2381 lhc->mlh_reg_lh.cookie);
2385 if (!fid_res_name_eq(mdt_object_fid(mo),
2386 &lock->l_resource->lr_name)) {
2387 CWARN("%s: Although resent, but still not "
2388 "get child lock:"DFID"\n",
2389 info->mti_exp->exp_obd->obd_name,
2390 PFID(mdt_object_fid(mo)));
2391 LDLM_LOCK_PUT(lock);
2394 LDLM_LOCK_PUT(lock);
2400 int mdt_remote_object_lock(struct mdt_thread_info *mti,
2401 struct mdt_object *o, const struct lu_fid *fid,
2402 struct lustre_handle *lh, ldlm_mode_t mode,
2405 struct ldlm_enqueue_info *einfo = &mti->mti_einfo;
2406 ldlm_policy_data_t *policy = &mti->mti_policy;
2407 struct ldlm_res_id *res_id = &mti->mti_res_id;
2411 LASSERT(mdt_object_remote(o));
2413 LASSERT(ibits == MDS_INODELOCK_UPDATE);
2415 fid_build_reg_res_name(fid, res_id);
2417 memset(einfo, 0, sizeof(*einfo));
2418 einfo->ei_type = LDLM_IBITS;
2419 einfo->ei_mode = mode;
2420 einfo->ei_cb_bl = mdt_remote_blocking_ast;
2421 einfo->ei_cb_cp = ldlm_completion_ast;
2422 einfo->ei_enq_slave = 0;
2423 einfo->ei_res_id = res_id;
2425 memset(policy, 0, sizeof(*policy));
2426 policy->l_inodebits.bits = ibits;
2428 rc = mo_object_lock(mti->mti_env, mdt_object_child(o), lh, einfo,
2433 static int mdt_object_local_lock(struct mdt_thread_info *info,
2434 struct mdt_object *o,
2435 struct mdt_lock_handle *lh, __u64 ibits,
2436 bool nonblock, int locality)
2438 struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
2439 ldlm_policy_data_t *policy = &info->mti_policy;
2440 struct ldlm_res_id *res_id = &info->mti_res_id;
2445 LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2446 LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2447 LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
2448 LASSERT(lh->mlh_type != MDT_NUL_LOCK);
2450 /* Only enqueue LOOKUP lock for remote object */
2451 if (mdt_object_remote(o))
2452 LASSERT(ibits == MDS_INODELOCK_LOOKUP);
2454 if (lh->mlh_type == MDT_PDO_LOCK) {
2455 /* check for exists after object is locked */
2456 if (mdt_object_exists(o) == 0) {
2457 /* Non-existent object shouldn't have PDO lock */
2460 /* Non-dir object shouldn't have PDO lock */
2461 if (!S_ISDIR(lu_object_attr(&o->mot_obj)))
2466 memset(policy, 0, sizeof(*policy));
2467 fid_build_reg_res_name(mdt_object_fid(o), res_id);
2469 dlmflags = LDLM_FL_ATOMIC_CB;
2471 dlmflags |= LDLM_FL_BLOCK_NOWAIT;
2474 * Take PDO lock on whole directory and build correct @res_id for lock
2475 * on part of directory.
2477 if (lh->mlh_pdo_hash != 0) {
2478 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
2479 mdt_lock_pdo_mode(info, o, lh);
2480 if (lh->mlh_pdo_mode != LCK_NL) {
2482 * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
2483 * is never going to be sent to client and we do not
2484 * want it slowed down due to possible cancels.
2486 policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
2487 rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
2488 policy, res_id, dlmflags,
2489 info->mti_exp == NULL ? NULL :
2490 &info->mti_exp->exp_handle.h_cookie);
2491 if (unlikely(rc != 0))
2492 GOTO(out_unlock, rc);
2496 * Finish res_id initializing by name hash marking part of
2497 * directory which is taking modification.
2499 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
2502 policy->l_inodebits.bits = ibits;
2505 * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
2506 * going to be sent to client. If it is - mdt_intent_policy() path will
2507 * fix it up and turn FL_LOCAL flag off.
2509 rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
2510 res_id, LDLM_FL_LOCAL_ONLY | dlmflags,
2511 info->mti_exp == NULL ? NULL :
2512 &info->mti_exp->exp_handle.h_cookie);
2515 mdt_object_unlock(info, o, lh, 1);
2516 else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
2517 lh->mlh_pdo_hash != 0 &&
2518 (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX))
2519 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
2525 mdt_object_lock_internal(struct mdt_thread_info *info, struct mdt_object *o,
2526 struct mdt_lock_handle *lh, __u64 ibits,
2527 bool nonblock, int locality)
2532 if (!mdt_object_remote(o))
2533 return mdt_object_local_lock(info, o, lh, ibits, nonblock,
2536 if (locality == MDT_LOCAL_LOCK) {
2537 CERROR("%s: try to get local lock for remote object"
2538 DFID".\n", mdt_obd_name(info->mti_mdt),
2539 PFID(mdt_object_fid(o)));
2543 /* XXX do not support PERM/LAYOUT/XATTR lock for remote object yet */
2544 ibits &= ~(MDS_INODELOCK_PERM | MDS_INODELOCK_LAYOUT |
2545 MDS_INODELOCK_XATTR);
2546 if (ibits & MDS_INODELOCK_UPDATE) {
2547 /* Sigh, PDO needs to enqueue 2 locks right now, but
2548 * enqueue RPC can only request 1 lock, to avoid extra
2549 * RPC, so it will instead enqueue EX lock for remote
2550 * object anyway XXX*/
2551 if (lh->mlh_type == MDT_PDO_LOCK &&
2552 lh->mlh_pdo_hash != 0) {
2553 CDEBUG(D_INFO, "%s: "DFID" convert PDO lock to"
2554 "EX lock.\n", mdt_obd_name(info->mti_mdt),
2555 PFID(mdt_object_fid(o)));
2556 lh->mlh_pdo_hash = 0;
2557 lh->mlh_rreg_mode = LCK_EX;
2558 lh->mlh_type = MDT_REG_LOCK;
2560 rc = mdt_remote_object_lock(info, o, mdt_object_fid(o),
2563 MDS_INODELOCK_UPDATE);
2568 /* Only enqueue LOOKUP lock for remote object */
2569 if (ibits & MDS_INODELOCK_LOOKUP) {
2570 rc = mdt_object_local_lock(info, o, lh,
2571 MDS_INODELOCK_LOOKUP,
2572 nonblock, locality);
2580 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2581 struct mdt_lock_handle *lh, __u64 ibits, int locality)
2583 return mdt_object_lock_internal(info, o, lh, ibits, false, locality);
2586 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o,
2587 struct mdt_lock_handle *lh, __u64 ibits, int locality)
2589 struct mdt_lock_handle tmp = *lh;
2592 rc = mdt_object_lock_internal(info, o, &tmp, ibits, true, locality);
2600 * Save a lock within request object.
2602 * Keep the lock referenced until whether client ACK or transaction
2603 * commit happens or release the lock immediately depending on input
2604 * parameters. If COS is ON, a write lock is converted to COS lock
2607 * \param info thead info object
2608 * \param h lock handle
2609 * \param mode lock mode
2610 * \param decref force immediate lock releasing
2613 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
2614 ldlm_mode_t mode, int decref)
2618 if (lustre_handle_is_used(h)) {
2619 if (decref || !info->mti_has_trans ||
2620 !(mode & (LCK_PW | LCK_EX))){
2621 mdt_fid_unlock(h, mode);
2623 struct mdt_device *mdt = info->mti_mdt;
2624 struct ldlm_lock *lock = ldlm_handle2lock(h);
2625 struct ptlrpc_request *req = mdt_info_req(info);
2628 LASSERTF(lock != NULL, "no lock for cookie "LPX64"\n",
2630 /* there is no request if mdt_object_unlock() is called
2631 * from mdt_export_cleanup()->mdt_add_dirty_flag() */
2632 if (likely(req != NULL)) {
2633 CDEBUG(D_HA, "request = %p reply state = %p"
2634 " transno = "LPD64"\n", req,
2635 req->rq_reply_state, req->rq_transno);
2636 if (mdt_cos_is_enabled(mdt)) {
2638 ldlm_lock_downgrade(lock, LCK_COS);
2641 ptlrpc_save_lock(req, h, mode, no_ack);
2643 ldlm_lock_decref(h, mode);
2645 if (mdt_is_lock_sync(lock)) {
2646 CDEBUG(D_HA, "found sync-lock,"
2647 " async commit started\n");
2648 mdt_device_commit_async(info->mti_env,
2651 LDLM_LOCK_PUT(lock);
2660 * Unlock mdt object.
2662 * Immeditely release the regular lock and the PDO lock or save the
2663 * lock in reqeuest and keep them referenced until client ACK or
2664 * transaction commit.
2666 * \param info thread info object
2667 * \param o mdt object
2668 * \param lh mdt lock handle referencing regular and PDO locks
2669 * \param decref force immediate lock releasing
2671 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
2672 struct mdt_lock_handle *lh, int decref)
2676 mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
2677 mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
2679 if (lustre_handle_is_used(&lh->mlh_rreg_lh))
2680 ldlm_lock_decref(&lh->mlh_rreg_lh, lh->mlh_rreg_mode);
2685 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
2686 const struct lu_fid *f,
2687 struct mdt_lock_handle *lh,
2690 struct mdt_object *o;
2692 o = mdt_object_find(info->mti_env, info->mti_mdt, f);
2696 rc = mdt_object_lock(info, o, lh, ibits,
2699 mdt_object_put(info->mti_env, o);
2706 void mdt_object_unlock_put(struct mdt_thread_info * info,
2707 struct mdt_object * o,
2708 struct mdt_lock_handle *lh,
2711 mdt_object_unlock(info, o, lh, decref);
2712 mdt_object_put(info->mti_env, o);
2716 * Generic code handling requests that have struct mdt_body passed in:
2718 * - extract mdt_body from request and save it in @info, if present;
2720 * - create lu_object, corresponding to the fid in mdt_body, and save it in
2723 * - if HABEO_CORPUS flag is set for this request type check whether object
2724 * actually exists on storage (lu_object_exists()).
2727 static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags)
2729 const struct mdt_body *body;
2730 struct mdt_object *obj;
2731 const struct lu_env *env;
2732 struct req_capsule *pill;
2736 env = info->mti_env;
2737 pill = info->mti_pill;
2739 body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
2743 if (!(body->mbo_valid & OBD_MD_FLID))
2746 if (!fid_is_sane(&body->mbo_fid1)) {
2747 CERROR("Invalid fid: "DFID"\n", PFID(&body->mbo_fid1));
2752 * Do not get size or any capa fields before we check that request
2753 * contains capa actually. There are some requests which do not, for
2754 * instance MDS_IS_SUBDIR.
2756 if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_CLIENT) &&
2757 req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
2758 mdt_set_capainfo(info, 0, &body->mbo_fid1,
2759 req_capsule_client_get(pill, &RMF_CAPA1));
2761 obj = mdt_object_find(env, info->mti_mdt, &body->mbo_fid1);
2763 if ((flags & HABEO_CORPUS) &&
2764 !mdt_object_exists(obj)) {
2765 mdt_object_put(env, obj);
2766 /* for capability renew ENOENT will be handled in
2768 if (body->mbo_valid & OBD_MD_FLOSSCAPA)
2773 info->mti_object = obj;
2782 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags)
2784 struct req_capsule *pill = info->mti_pill;
2788 if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
2789 rc = mdt_body_unpack(info, flags);
2793 if (rc == 0 && (flags & HABEO_REFERO)) {
2795 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2796 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2798 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2799 req_capsule_set_size(pill, &RMF_LOGCOOKIES,
2802 rc = req_capsule_server_pack(pill);
2807 static int mdt_init_capa_ctxt(const struct lu_env *env, struct mdt_device *m)
2809 struct md_device *next = m->mdt_child;
2811 return next->md_ops->mdo_init_capa_ctxt(env, next,
2812 m->mdt_lut.lut_mds_capa,
2813 m->mdt_capa_timeout,
2818 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
2820 lh->mlh_type = MDT_NUL_LOCK;
2821 lh->mlh_reg_lh.cookie = 0ull;
2822 lh->mlh_reg_mode = LCK_MINMODE;
2823 lh->mlh_pdo_lh.cookie = 0ull;
2824 lh->mlh_pdo_mode = LCK_MINMODE;
2825 lh->mlh_rreg_lh.cookie = 0ull;
2826 lh->mlh_rreg_mode = LCK_MINMODE;
2829 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
2831 LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2832 LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2836 * Initialize fields of struct mdt_thread_info. Other fields are left in
2837 * uninitialized state, because it's too expensive to zero out whole
2838 * mdt_thread_info (> 1K) on each request arrival.
2840 void mdt_thread_info_init(struct ptlrpc_request *req,
2841 struct mdt_thread_info *info)
2845 info->mti_pill = &req->rq_pill;
2848 for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2849 mdt_lock_handle_init(&info->mti_lh[i]);
2851 /* mdt device: it can be NULL while CONNECT */
2852 if (req->rq_export) {
2853 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
2854 info->mti_exp = req->rq_export;
2856 info->mti_mdt = NULL;
2857 info->mti_env = req->rq_svc_thread->t_env;
2858 info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
2860 memset(&info->mti_attr, 0, sizeof(info->mti_attr));
2861 info->mti_big_buf = LU_BUF_NULL;
2862 info->mti_body = NULL;
2863 info->mti_object = NULL;
2864 info->mti_dlm_req = NULL;
2865 info->mti_has_trans = 0;
2866 info->mti_cross_ref = 0;
2867 info->mti_opdata = 0;
2868 info->mti_big_lmm_used = 0;
2870 info->mti_spec.no_create = 0;
2871 info->mti_spec.sp_rm_entry = 0;
2872 info->mti_spec.sp_permitted = 0;
2874 info->mti_spec.u.sp_ea.eadata = NULL;
2875 info->mti_spec.u.sp_ea.eadatalen = 0;
2878 void mdt_thread_info_fini(struct mdt_thread_info *info)
2882 if (info->mti_object != NULL) {
2883 mdt_object_put(info->mti_env, info->mti_object);
2884 info->mti_object = NULL;
2887 for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2888 mdt_lock_handle_fini(&info->mti_lh[i]);
2889 info->mti_env = NULL;
2890 info->mti_pill = NULL;
2891 info->mti_exp = NULL;
2893 if (unlikely(info->mti_big_buf.lb_buf != NULL))
2894 lu_buf_free(&info->mti_big_buf);
2897 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi)
2899 struct mdt_thread_info *mti;
2900 struct lustre_capa *lc;
2902 mti = mdt_th_info(tsi->tsi_env);
2903 LASSERT(mti != NULL);
2905 mdt_thread_info_init(tgt_ses_req(tsi), mti);
2906 if (tsi->tsi_corpus != NULL) {
2907 struct req_capsule *pill = tsi->tsi_pill;
2909 mti->mti_object = mdt_obj(tsi->tsi_corpus);
2910 lu_object_get(tsi->tsi_corpus);
2913 * XXX: must be part of tgt_mdt_body_unpack but moved here
2914 * due to mdt_set_capainfo().
2916 if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_CLIENT) &&
2917 req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT) > 0) {
2918 lc = req_capsule_client_get(pill, &RMF_CAPA1);
2919 mdt_set_capainfo(mti, 0, &tsi->tsi_mdt_body->mbo_fid1,
2923 mti->mti_body = tsi->tsi_mdt_body;
2924 mti->mti_dlm_req = tsi->tsi_dlm_req;
2929 static int mdt_tgt_connect(struct tgt_session_info *tsi)
2931 struct ptlrpc_request *req = tgt_ses_req(tsi);
2936 if (OBD_FAIL_CHECK(OBD_FAIL_TGT_DELAY_CONDITIONAL) &&
2938 tsi2mdt_info(tsi)->mti_mdt->mdt_seq_site.ss_node_id)
2939 schedule_timeout_and_set_state(TASK_UNINTERRUPTIBLE,
2940 msecs_to_jiffies(3 * MSEC_PER_SEC));
2942 rc = tgt_connect(tsi);
2946 rc = mdt_init_idmap(tsi);
2951 obd_disconnect(class_export_get(req->rq_export));
2970 static int mdt_intent_getattr(enum mdt_it_code opcode,
2971 struct mdt_thread_info *info,
2972 struct ldlm_lock **,
2975 static int mdt_intent_getxattr(enum mdt_it_code opcode,
2976 struct mdt_thread_info *info,
2977 struct ldlm_lock **lockp,
2980 static int mdt_intent_layout(enum mdt_it_code opcode,
2981 struct mdt_thread_info *info,
2982 struct ldlm_lock **,
2984 static int mdt_intent_reint(enum mdt_it_code opcode,
2985 struct mdt_thread_info *info,
2986 struct ldlm_lock **,
2989 static struct mdt_it_flavor {
2990 const struct req_format *it_fmt;
2992 int (*it_act)(enum mdt_it_code ,
2993 struct mdt_thread_info *,
2994 struct ldlm_lock **,
2997 } mdt_it_flavor[] = {
2999 .it_fmt = &RQF_LDLM_INTENT,
3000 /*.it_flags = HABEO_REFERO,*/
3002 .it_act = mdt_intent_reint,
3003 .it_reint = REINT_OPEN
3006 .it_fmt = &RQF_LDLM_INTENT,
3008 * OCREAT is not a MUTABOR request as if the file
3010 * We do the extra check of OBD_CONNECT_RDONLY in
3011 * mdt_reint_open() when we really need to create
3015 .it_act = mdt_intent_reint,
3016 .it_reint = REINT_OPEN
3019 .it_fmt = &RQF_LDLM_INTENT,
3020 .it_flags = MUTABOR,
3021 .it_act = mdt_intent_reint,
3022 .it_reint = REINT_CREATE
3024 [MDT_IT_GETATTR] = {
3025 .it_fmt = &RQF_LDLM_INTENT_GETATTR,
3026 .it_flags = HABEO_REFERO,
3027 .it_act = mdt_intent_getattr
3029 [MDT_IT_READDIR] = {
3035 .it_fmt = &RQF_LDLM_INTENT_GETATTR,
3036 .it_flags = HABEO_REFERO,
3037 .it_act = mdt_intent_getattr
3040 .it_fmt = &RQF_LDLM_INTENT_UNLINK,
3041 .it_flags = MUTABOR,
3043 .it_reint = REINT_UNLINK
3047 .it_flags = MUTABOR,
3050 [MDT_IT_GETXATTR] = {
3051 .it_fmt = &RQF_LDLM_INTENT_GETXATTR,
3052 .it_flags = HABEO_CORPUS,
3053 .it_act = mdt_intent_getxattr
3056 .it_fmt = &RQF_LDLM_INTENT_LAYOUT,
3058 .it_act = mdt_intent_layout
3063 mdt_intent_lock_replace(struct mdt_thread_info *info,
3064 struct ldlm_lock **lockp,
3065 struct mdt_lock_handle *lh,
3068 struct ptlrpc_request *req = mdt_info_req(info);
3069 struct ldlm_lock *lock = *lockp;
3070 struct ldlm_lock *new_lock;
3072 /* If possible resent found a lock, @lh is set to its handle */
3073 new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
3075 if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
3076 lh->mlh_reg_lh.cookie = 0;
3080 LASSERTF(new_lock != NULL,
3081 "lockh "LPX64"\n", lh->mlh_reg_lh.cookie);
3084 * If we've already given this lock to a client once, then we should
3085 * have no readers or writers. Otherwise, we should have one reader
3086 * _or_ writer ref (which will be zeroed below) before returning the
3089 if (new_lock->l_export == req->rq_export) {
3090 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
3092 LASSERT(new_lock->l_export == NULL);
3093 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
3098 if (new_lock->l_export == req->rq_export) {
3100 * Already gave this to the client, which means that we
3101 * reconstructed a reply.
3103 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
3106 LDLM_LOCK_RELEASE(new_lock);
3107 lh->mlh_reg_lh.cookie = 0;
3108 RETURN(ELDLM_LOCK_REPLACED);
3112 * Fixup the lock to be given to the client.
3114 lock_res_and_lock(new_lock);
3115 /* Zero new_lock->l_readers and new_lock->l_writers without triggering
3116 * possible blocking AST. */
3117 while (new_lock->l_readers > 0) {
3118 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
3119 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3120 new_lock->l_readers--;
3122 while (new_lock->l_writers > 0) {
3123 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
3124 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3125 new_lock->l_writers--;
3128 new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
3129 new_lock->l_blocking_ast = lock->l_blocking_ast;
3130 new_lock->l_completion_ast = lock->l_completion_ast;
3131 new_lock->l_remote_handle = lock->l_remote_handle;
3132 new_lock->l_flags &= ~LDLM_FL_LOCAL;
3134 unlock_res_and_lock(new_lock);
3136 cfs_hash_add(new_lock->l_export->exp_lock_hash,
3137 &new_lock->l_remote_handle,
3138 &new_lock->l_exp_hash);
3140 LDLM_LOCK_RELEASE(new_lock);
3141 lh->mlh_reg_lh.cookie = 0;
3143 RETURN(ELDLM_LOCK_REPLACED);
3146 static void mdt_intent_fixup_resent(struct mdt_thread_info *info,
3147 struct ldlm_lock *new_lock,
3148 struct mdt_lock_handle *lh,
3151 struct ptlrpc_request *req = mdt_info_req(info);
3152 struct ldlm_request *dlmreq;
3154 if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
3157 dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
3159 /* Check if this is a resend case (MSG_RESENT is set on RPC) and a
3160 * lock was found by ldlm_handle_enqueue(); if so @lh must be
3162 if (flags & LDLM_FL_RESENT) {
3163 lh->mlh_reg_lh.cookie = new_lock->l_handle.h_cookie;
3164 lh->mlh_reg_mode = new_lock->l_granted_mode;
3166 LDLM_DEBUG(new_lock, "Restoring lock cookie");
3167 DEBUG_REQ(D_DLMTRACE, req, "restoring lock cookie "LPX64,
3168 lh->mlh_reg_lh.cookie);
3173 * If the xid matches, then we know this is a resent request, and allow
3174 * it. (It's probably an OPEN, for which we don't send a lock.
3176 if (req_xid_is_last(req))
3180 * This remote handle isn't enqueued, so we never received or processed
3181 * this request. Clear MSG_RESENT, because it can be handled like any
3182 * normal request now.
3184 lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
3186 DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle "LPX64,
3187 dlmreq->lock_handle[0].cookie);
3190 static int mdt_intent_getxattr(enum mdt_it_code opcode,
3191 struct mdt_thread_info *info,
3192 struct ldlm_lock **lockp,
3195 struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3196 struct ldlm_reply *ldlm_rep = NULL;
3200 * Initialize lhc->mlh_reg_lh either from a previously granted lock
3201 * (for the resend case) or a new lock. Below we will use it to
3202 * replace the original lock.
3204 mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3205 if (!lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3206 mdt_lock_reg_init(lhc, (*lockp)->l_req_mode);
3207 rc = mdt_object_lock(info, info->mti_object, lhc,
3208 MDS_INODELOCK_XATTR,
3214 grc = mdt_getxattr(info);
3216 rc = mdt_intent_lock_replace(info, lockp, lhc, flags);
3218 if (mdt_info_req(info)->rq_repmsg != NULL)
3219 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3220 if (ldlm_rep == NULL)
3221 RETURN(err_serious(-EFAULT));
3223 ldlm_rep->lock_policy_res2 = grc;
3228 static int mdt_intent_getattr(enum mdt_it_code opcode,
3229 struct mdt_thread_info *info,
3230 struct ldlm_lock **lockp,
3233 struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3235 struct ldlm_reply *ldlm_rep;
3236 struct mdt_body *reqbody;
3237 struct mdt_body *repbody;
3241 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
3244 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
3247 info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
3248 repbody->mbo_eadatasize = 0;
3249 repbody->mbo_aclsize = 0;
3253 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM;
3255 case MDT_IT_GETATTR:
3256 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
3260 CERROR("Unsupported intent (%d)\n", opcode);
3261 GOTO(out_shrink, rc = -EINVAL);
3264 rc = mdt_init_ucred(info, reqbody);
3266 GOTO(out_shrink, rc);
3268 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3269 mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
3271 /* Get lock from request for possible resent case. */
3272 mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3274 rc = mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
3275 ldlm_rep->lock_policy_res2 = clear_serious(rc);
3277 if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
3278 ldlm_rep->lock_policy_res2 = 0;
3279 if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
3280 ldlm_rep->lock_policy_res2) {
3281 lhc->mlh_reg_lh.cookie = 0ull;
3282 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
3285 rc = mdt_intent_lock_replace(info, lockp, lhc, flags);
3288 mdt_exit_ucred(info);
3290 mdt_client_compatibility(info);
3291 rc2 = mdt_fix_reply(info);
3297 static int mdt_intent_layout(enum mdt_it_code opcode,
3298 struct mdt_thread_info *info,
3299 struct ldlm_lock **lockp,
3302 struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_LAYOUT];
3303 struct layout_intent *layout;
3305 struct mdt_object *obj = NULL;
3309 if (opcode != MDT_IT_LAYOUT) {
3310 CERROR("%s: Unknown intent (%d)\n", mdt_obd_name(info->mti_mdt),
3315 fid = &info->mti_tmp_fid2;
3316 fid_extract_from_res_name(fid, &(*lockp)->l_resource->lr_name);
3318 /* Get lock from request for possible resent case. */
3319 mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3321 obj = mdt_object_find(info->mti_env, info->mti_mdt, fid);
3323 RETURN(PTR_ERR(obj));
3325 if (mdt_object_exists(obj) && !mdt_object_remote(obj)) {
3326 /* get the length of lsm */
3327 rc = mdt_attr_get_eabuf_size(info, obj);
3329 mdt_object_put(info->mti_env, obj);
3333 if (rc > info->mti_mdt->mdt_max_mdsize)
3334 info->mti_mdt->mdt_max_mdsize = rc;
3337 mdt_object_put(info->mti_env, obj);
3339 (*lockp)->l_lvb_type = LVB_T_LAYOUT;
3340 req_capsule_set_size(info->mti_pill, &RMF_DLM_LVB, RCL_SERVER, rc);
3341 rc = req_capsule_server_pack(info->mti_pill);
3345 if (lustre_handle_is_used(&lhc->mlh_reg_lh))
3346 rc = mdt_intent_lock_replace(info, lockp, lhc, flags);
3348 layout = req_capsule_client_get(info->mti_pill, &RMF_LAYOUT_INTENT);
3349 LASSERT(layout != NULL);
3350 if (layout->li_opc == LAYOUT_INTENT_ACCESS)
3351 /* return to normal/resent ldlm handling */
3354 CERROR("%s: Unsupported layout intent (%d)\n",
3355 mdt_obd_name(info->mti_mdt), layout->li_opc);
3359 static int mdt_intent_reint(enum mdt_it_code opcode,
3360 struct mdt_thread_info *info,
3361 struct ldlm_lock **lockp,
3364 struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3365 struct ldlm_reply *rep = NULL;
3369 static const struct req_format *intent_fmts[REINT_MAX] = {
3370 [REINT_CREATE] = &RQF_LDLM_INTENT_CREATE,
3371 [REINT_OPEN] = &RQF_LDLM_INTENT_OPEN
3376 opc = mdt_reint_opcode(mdt_info_req(info), intent_fmts);
3380 if (mdt_it_flavor[opcode].it_reint != opc) {
3381 CERROR("Reint code %ld doesn't match intent: %d\n",
3383 RETURN(err_serious(-EPROTO));
3386 /* Get lock from request for possible resent case. */
3387 mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3389 rc = mdt_reint_internal(info, lhc, opc);
3391 /* Check whether the reply has been packed successfully. */
3392 if (mdt_info_req(info)->rq_repmsg != NULL)
3393 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3395 RETURN(err_serious(-EFAULT));
3397 /* MDC expects this in any case */
3399 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
3401 /* the open lock or the lock for cross-ref object should be
3402 * returned to the client */
3403 if (rc == -MDT_EREMOTE_OPEN || mdt_get_disposition(rep, DISP_OPEN_LOCK)) {
3404 LASSERT(lustre_handle_is_used(&lhc->mlh_reg_lh));
3405 rep->lock_policy_res2 = 0;
3406 rc = mdt_intent_lock_replace(info, lockp, lhc, flags);
3410 rep->lock_policy_res2 = clear_serious(rc);
3412 if (rep->lock_policy_res2 == -ENOENT &&
3413 mdt_get_disposition(rep, DISP_LOOKUP_NEG))
3414 rep->lock_policy_res2 = 0;
3416 if (rc == -ENOTCONN || rc == -ENODEV ||
3417 rc == -EOVERFLOW) { /**< if VBR failure then return error */
3419 * If it is the disconnect error (ENODEV & ENOCONN), the error
3420 * will be returned by rq_status, and client at ptlrpc layer
3421 * will detect this, then disconnect, reconnect the import
3422 * immediately, instead of impacting the following the rpc.
3424 lhc->mlh_reg_lh.cookie = 0ull;
3428 * For other cases, the error will be returned by intent.
3429 * and client will retrieve the result from intent.
3432 * FIXME: when open lock is finished, that should be
3435 if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3436 LASSERTF(rc == 0, "Error occurred but lock handle "
3437 "is still in use, rc = %d\n", rc);
3438 rep->lock_policy_res2 = 0;
3439 rc = mdt_intent_lock_replace(info, lockp, lhc, flags);
3442 lhc->mlh_reg_lh.cookie = 0ull;
3443 RETURN(ELDLM_LOCK_ABORTED);
3448 static int mdt_intent_code(long itcode)
3456 case IT_OPEN|IT_CREAT:
3463 rc = MDT_IT_READDIR;
3466 rc = MDT_IT_GETATTR;
3478 rc = MDT_IT_GETXATTR;
3483 case IT_QUOTA_DQACQ:
3488 CERROR("Unknown intent opcode: %ld\n", itcode);
3495 static int mdt_intent_opc(long itopc, struct mdt_thread_info *info,
3496 struct ldlm_lock **lockp, __u64 flags)
3498 struct req_capsule *pill = info->mti_pill;
3499 struct ptlrpc_request *req = mdt_info_req(info);
3500 struct mdt_it_flavor *flv;
3505 opc = mdt_intent_code(itopc);
3509 if (opc == MDT_IT_QUOTA) {
3510 struct lu_device *qmt = info->mti_mdt->mdt_qmt_dev;
3513 RETURN(-EOPNOTSUPP);
3515 (*lockp)->l_lvb_type = LVB_T_LQUOTA;
3516 /* pass the request to quota master */
3517 rc = qmt_hdls.qmth_intent_policy(info->mti_env, qmt,
3518 mdt_info_req(info), lockp,
3523 flv = &mdt_it_flavor[opc];
3524 if (flv->it_fmt != NULL)
3525 req_capsule_extend(pill, flv->it_fmt);
3527 rc = mdt_unpack_req_pack_rep(info, flv->it_flags);
3531 if (flv->it_flags & MUTABOR &&
3532 exp_connect_flags(req->rq_export) & OBD_CONNECT_RDONLY)
3535 if (flv->it_act != NULL) {
3536 struct ldlm_reply *rep;
3538 /* execute policy */
3539 rc = flv->it_act(opc, info, lockp, flags);
3541 /* Check whether the reply has been packed successfully. */
3542 if (req->rq_repmsg != NULL) {
3543 rep = req_capsule_server_get(info->mti_pill,
3545 rep->lock_policy_res2 =
3546 ptlrpc_status_hton(rep->lock_policy_res2);
3553 static int mdt_intent_policy(struct ldlm_namespace *ns,
3554 struct ldlm_lock **lockp, void *req_cookie,
3555 ldlm_mode_t mode, __u64 flags, void *data)
3557 struct tgt_session_info *tsi;
3558 struct mdt_thread_info *info;
3559 struct ptlrpc_request *req = req_cookie;
3560 struct ldlm_intent *it;
3561 struct req_capsule *pill;
3566 LASSERT(req != NULL);
3568 tsi = tgt_ses_info(req->rq_svc_thread->t_env);
3570 info = tsi2mdt_info(tsi);
3571 LASSERT(info != NULL);
3572 pill = info->mti_pill;
3573 LASSERT(pill->rc_req == req);
3575 if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
3576 req_capsule_extend(pill, &RQF_LDLM_INTENT_BASIC);
3577 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
3579 rc = mdt_intent_opc(it->opc, info, lockp, flags);
3583 /* Lock without inodebits makes no sense and will oops
3584 * later in ldlm. Let's check it now to see if we have
3585 * ibits corrupted somewhere in mdt_intent_opc().
3586 * The case for client miss to set ibits has been
3587 * processed by others. */
3588 LASSERT(ergo(info->mti_dlm_req->lock_desc.l_resource.\
3589 lr_type == LDLM_IBITS,
3590 info->mti_dlm_req->lock_desc.\
3591 l_policy_data.l_inodebits.bits != 0));
3593 rc = err_serious(-EFAULT);
3595 /* No intent was provided */
3596 LASSERT(pill->rc_fmt == &RQF_LDLM_ENQUEUE);
3597 req_capsule_set_size(pill, &RMF_DLM_LVB, RCL_SERVER, 0);
3598 rc = req_capsule_server_pack(pill);
3600 rc = err_serious(rc);
3602 mdt_thread_info_fini(info);
3606 static void mdt_deregister_seq_exp(struct mdt_device *mdt)
3608 struct seq_server_site *ss = mdt_seq_site(mdt);
3610 if (ss->ss_node_id == 0)
3613 if (ss->ss_client_seq != NULL) {
3614 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
3615 ss->ss_client_seq->lcs_exp = NULL;
3618 if (ss->ss_server_fld != NULL) {
3619 lustre_deregister_lwp_item(&ss->ss_server_fld->lsf_control_exp);
3620 ss->ss_server_fld->lsf_control_exp = NULL;
3624 static void mdt_seq_fini_cli(struct mdt_device *mdt)
3626 struct seq_server_site *ss = mdt_seq_site(mdt);
3631 if (ss->ss_server_seq != NULL)
3632 seq_server_set_cli(NULL, ss->ss_server_seq, NULL);
3635 static int mdt_seq_fini(const struct lu_env *env, struct mdt_device *mdt)
3637 mdt_seq_fini_cli(mdt);
3638 mdt_deregister_seq_exp(mdt);
3640 return seq_site_fini(env, mdt_seq_site(mdt));
3644 * It will retrieve its FLDB entries from MDT0, and it only happens
3645 * when upgrading existent FS to 2.6 or when local FLDB is corrupted,
3646 * and it needs to refresh FLDB from the MDT0.
3648 static int mdt_register_lwp_callback(void *data)
3651 struct mdt_device *mdt = data;
3652 struct lu_server_fld *fld = mdt_seq_site(mdt)->ss_server_fld;
3656 LASSERT(mdt_seq_site(mdt)->ss_node_id != 0);
3658 if (!likely(fld->lsf_new))
3661 rc = lu_env_init(&env, LCT_MD_THREAD);
3663 CERROR("%s: cannot init env: rc = %d\n", mdt_obd_name(mdt), rc);
3667 rc = fld_update_from_controller(&env, fld);
3669 CERROR("%s: cannot update controller: rc = %d\n",
3670 mdt_obd_name(mdt), rc);
3678 static int mdt_register_seq_exp(struct mdt_device *mdt)
3680 struct seq_server_site *ss = mdt_seq_site(mdt);
3681 char *lwp_name = NULL;
3684 if (ss->ss_node_id == 0)
3687 OBD_ALLOC(lwp_name, MAX_OBD_NAME);
3688 if (lwp_name == NULL)
3689 GOTO(out_free, rc = -ENOMEM);
3691 rc = tgt_name2lwp_name(mdt_obd_name(mdt), lwp_name, MAX_OBD_NAME, 0);
3695 rc = lustre_register_lwp_item(lwp_name, &ss->ss_client_seq->lcs_exp,
3700 rc = lustre_register_lwp_item(lwp_name,
3701 &ss->ss_server_fld->lsf_control_exp,
3702 mdt_register_lwp_callback, mdt);
3704 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
3705 ss->ss_client_seq->lcs_exp = NULL;
3709 if (lwp_name != NULL)
3710 OBD_FREE(lwp_name, MAX_OBD_NAME);
3716 * Init client sequence manager which is used by local MDS to talk to sequence
3717 * controller on remote node.
3719 static int mdt_seq_init_cli(const struct lu_env *env, struct mdt_device *mdt)
3721 struct seq_server_site *ss = mdt_seq_site(mdt);
3726 /* check if this is adding the first MDC and controller is not yet
3728 OBD_ALLOC_PTR(ss->ss_client_seq);
3729 if (ss->ss_client_seq == NULL)
3732 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3733 if (prefix == NULL) {
3734 OBD_FREE_PTR(ss->ss_client_seq);
3735 ss->ss_client_seq = NULL;
3739 /* Note: seq_client_fini will be called in seq_site_fini */
3740 snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s", mdt_obd_name(mdt));
3741 rc = seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_METADATA,
3742 prefix, ss->ss_node_id == 0 ? ss->ss_control_seq :
3744 OBD_FREE(prefix, MAX_OBD_NAME + 5);
3746 OBD_FREE_PTR(ss->ss_client_seq);
3747 ss->ss_client_seq = NULL;
3751 rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq);
3756 static int mdt_seq_init(const struct lu_env *env, struct mdt_device *mdt)
3758 struct seq_server_site *ss;
3762 ss = mdt_seq_site(mdt);
3763 /* init sequence controller server(MDT0) */
3764 if (ss->ss_node_id == 0) {
3765 OBD_ALLOC_PTR(ss->ss_control_seq);
3766 if (ss->ss_control_seq == NULL)
3769 rc = seq_server_init(env, ss->ss_control_seq, mdt->mdt_bottom,
3770 mdt_obd_name(mdt), LUSTRE_SEQ_CONTROLLER,
3773 GOTO(out_seq_fini, rc);
3776 /* Init normal sequence server */
3777 OBD_ALLOC_PTR(ss->ss_server_seq);
3778 if (ss->ss_server_seq == NULL)
3779 GOTO(out_seq_fini, rc = -ENOMEM);
3781 rc = seq_server_init(env, ss->ss_server_seq, mdt->mdt_bottom,
3782 mdt_obd_name(mdt), LUSTRE_SEQ_SERVER, ss);
3784 GOTO(out_seq_fini, rc);
3786 /* init seq client for seq server to talk to seq controller(MDT0) */
3787 rc = mdt_seq_init_cli(env, mdt);
3789 GOTO(out_seq_fini, rc);
3791 if (ss->ss_node_id != 0)
3792 /* register controler export through lwp */
3793 rc = mdt_register_seq_exp(mdt);
3798 mdt_seq_fini(env, mdt);
3806 static int mdt_fld_fini(const struct lu_env *env,
3807 struct mdt_device *m)
3809 struct seq_server_site *ss = mdt_seq_site(m);
3812 if (ss && ss->ss_server_fld) {
3813 fld_server_fini(env, ss->ss_server_fld);
3814 OBD_FREE_PTR(ss->ss_server_fld);
3815 ss->ss_server_fld = NULL;
3821 static int mdt_fld_init(const struct lu_env *env,
3823 struct mdt_device *m)
3825 struct seq_server_site *ss;
3829 ss = mdt_seq_site(m);
3831 OBD_ALLOC_PTR(ss->ss_server_fld);
3832 if (ss->ss_server_fld == NULL)
3833 RETURN(rc = -ENOMEM);
3835 rc = fld_server_init(env, ss->ss_server_fld, m->mdt_bottom, uuid,
3838 OBD_FREE_PTR(ss->ss_server_fld);
3839 ss->ss_server_fld = NULL;
3846 static void mdt_stack_pre_fini(const struct lu_env *env,
3847 struct mdt_device *m, struct lu_device *top)
3849 struct lustre_cfg_bufs *bufs;
3850 struct lustre_cfg *lcfg;
3851 struct mdt_thread_info *info;
3856 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3857 LASSERT(info != NULL);
3859 bufs = &info->mti_u.bufs;
3861 LASSERT(m->mdt_child_exp);
3862 LASSERT(m->mdt_child_exp->exp_obd);
3864 /* process cleanup, pass mdt obd name to get obd umount flags */
3865 /* XXX: this is needed because all layers are referenced by
3866 * objects (some of them are pinned by osd, for example *
3867 * the proper solution should be a model where object used
3868 * by osd only doesn't have mdt/mdd slices -bzzz */
3869 lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
3870 lustre_cfg_bufs_set_string(bufs, 1, NULL);
3871 lcfg = lustre_cfg_new(LCFG_PRE_CLEANUP, bufs);
3875 top->ld_ops->ldo_process_config(env, top, lcfg);
3876 lustre_cfg_free(lcfg);
3880 static void mdt_stack_fini(const struct lu_env *env,
3881 struct mdt_device *m, struct lu_device *top)
3883 struct obd_device *obd = mdt2obd_dev(m);
3884 struct lustre_cfg_bufs *bufs;
3885 struct lustre_cfg *lcfg;
3886 struct mdt_thread_info *info;
3890 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3891 LASSERT(info != NULL);
3893 lu_dev_del_linkage(top->ld_site, top);
3895 lu_site_purge(env, top->ld_site, -1);
3897 bufs = &info->mti_u.bufs;
3898 /* process cleanup, pass mdt obd name to get obd umount flags */
3899 /* another purpose is to let all layers to release their objects */
3900 lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
3905 lustre_cfg_bufs_set_string(bufs, 1, flags);
3906 lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
3911 top->ld_ops->ldo_process_config(env, top, lcfg);
3912 lustre_cfg_free(lcfg);
3914 lu_site_purge(env, top->ld_site, -1);
3916 m->mdt_child = NULL;
3917 m->mdt_bottom = NULL;
3919 obd_disconnect(m->mdt_child_exp);
3920 m->mdt_child_exp = NULL;
3922 obd_disconnect(m->mdt_bottom_exp);
3923 m->mdt_child_exp = NULL;
3926 static int mdt_connect_to_next(const struct lu_env *env, struct mdt_device *m,
3927 const char *next, struct obd_export **exp)
3929 struct obd_connect_data *data = NULL;
3930 struct obd_device *obd;
3934 OBD_ALLOC_PTR(data);
3936 GOTO(out, rc = -ENOMEM);
3938 obd = class_name2obd(next);
3940 CERROR("%s: can't locate next device: %s\n",
3941 mdt_obd_name(m), next);
3942 GOTO(out, rc = -ENOTCONN);
3945 data->ocd_connect_flags = OBD_CONNECT_VERSION;
3946 data->ocd_version = LUSTRE_VERSION_CODE;
3948 rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
3950 CERROR("%s: cannot connect to next dev %s (%d)\n",
3951 mdt_obd_name(m), next, rc);
3961 static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt,
3962 struct lustre_cfg *cfg)
3964 char *dev = lustre_cfg_string(cfg, 0);
3965 int rc, name_size, uuid_size;
3966 char *name, *uuid, *p;
3967 struct lustre_cfg_bufs *bufs;
3968 struct lustre_cfg *lcfg;
3969 struct obd_device *obd;
3970 struct lustre_profile *lprof;
3971 struct lu_site *site;
3974 /* in 1.8 we had the only device in the stack - MDS.
3975 * 2.0 introduces MDT, MDD, OSD; MDT starts others internally.
3976 * in 2.3 OSD is instantiated by obd_mount.c, so we need
3977 * to generate names and setup MDT, MDD. MDT will be using
3978 * generated name to connect to MDD. for MDD the next device
3979 * will be LOD with name taken from so called "profile" which
3980 * is generated by mount_option line
3982 * 1.8 MGS generates config. commands like this:
3983 * #06 (104)mount_option 0: 1:lustre-MDT0000 2:lustre-mdtlov
3984 * #08 (120)setup 0:lustre-MDT0000 1:dev 2:type 3:lustre-MDT0000
3985 * 2.0 MGS generates config. commands like this:
3986 * #07 (112)mount_option 0: 1:lustre-MDT0000 2:lustre-MDT0000-mdtlov
3987 * #08 (160)setup 0:lustre-MDT0000 1:lustre-MDT0000_UUID 2:0
3988 * 3:lustre-MDT0000-mdtlov 4:f
3990 * we generate MDD name from MDT one, just replacing T with D
3992 * after all the preparations, the logical equivalent will be
3993 * #01 (160)setup 0:lustre-MDD0000 1:lustre-MDD0000_UUID 2:0
3994 * 3:lustre-MDT0000-mdtlov 4:f
3995 * #02 (160)setup 0:lustre-MDT0000 1:lustre-MDT0000_UUID 2:0
3996 * 3:lustre-MDD0000 4:f
3998 * notice we build the stack from down to top: MDD first, then MDT */
4000 name_size = MAX_OBD_NAME;
4001 uuid_size = MAX_OBD_NAME;
4003 OBD_ALLOC(name, name_size);
4004 OBD_ALLOC(uuid, uuid_size);
4005 if (name == NULL || uuid == NULL)
4006 GOTO(cleanup_mem, rc = -ENOMEM);
4008 OBD_ALLOC_PTR(bufs);
4010 GOTO(cleanup_mem, rc = -ENOMEM);
4013 p = strstr(name, "-MDT");
4015 GOTO(free_bufs, rc = -ENOMEM);
4018 snprintf(uuid, MAX_OBD_NAME, "%s_UUID", name);
4020 lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4021 if (lprof == NULL || lprof->lp_dt == NULL) {
4022 CERROR("can't find the profile: %s\n",
4023 lustre_cfg_string(cfg, 0));
4024 GOTO(free_bufs, rc = -EINVAL);
4027 lustre_cfg_bufs_reset(bufs, name);
4028 lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_MDD_NAME);
4029 lustre_cfg_bufs_set_string(bufs, 2, uuid);
4030 lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4032 lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
4034 GOTO(free_bufs, rc = -ENOMEM);
4036 rc = class_attach(lcfg);
4038 GOTO(lcfg_cleanup, rc);
4040 obd = class_name2obd(name);
4042 CERROR("Can not find obd %s (%s in config)\n",
4043 MDD_OBD_NAME, lustre_cfg_string(cfg, 0));
4044 GOTO(lcfg_cleanup, rc = -EINVAL);
4047 lustre_cfg_free(lcfg);
4049 lustre_cfg_bufs_reset(bufs, name);
4050 lustre_cfg_bufs_set_string(bufs, 1, uuid);
4051 lustre_cfg_bufs_set_string(bufs, 2, dev);
4052 lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4054 lcfg = lustre_cfg_new(LCFG_SETUP, bufs);
4056 GOTO(class_detach, rc = -ENOMEM);
4058 rc = class_setup(obd, lcfg);
4060 GOTO(class_detach, rc);
4062 /* connect to MDD we just setup */
4063 rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_child_exp);
4065 GOTO(class_detach, rc);
4067 site = mdt->mdt_child_exp->exp_obd->obd_lu_dev->ld_site;
4069 LASSERT(mdt_lu_site(mdt) == NULL);
4070 mdt->mdt_lu_dev.ld_site = site;
4071 site->ls_top_dev = &mdt->mdt_lu_dev;
4072 mdt->mdt_child = lu2md_dev(mdt->mdt_child_exp->exp_obd->obd_lu_dev);
4074 /* now connect to bottom OSD */
4075 snprintf(name, MAX_OBD_NAME, "%s-osd", dev);
4076 rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_bottom_exp);
4078 GOTO(class_detach, rc);
4080 lu2dt_dev(mdt->mdt_bottom_exp->exp_obd->obd_lu_dev);
4082 rc = lu_env_refill((struct lu_env *)env);
4084 CERROR("Failure to refill session: '%d'\n", rc);
4086 lu_dev_add_linkage(site, &mdt->mdt_lu_dev);
4091 class_detach(obd, lcfg);
4093 lustre_cfg_free(lcfg);
4098 OBD_FREE(name, name_size);
4100 OBD_FREE(uuid, uuid_size);
4104 /* setup quota master target on MDT0 */
4105 static int mdt_quota_init(const struct lu_env *env, struct mdt_device *mdt,
4106 struct lustre_cfg *cfg)
4108 struct obd_device *obd;
4109 char *dev = lustre_cfg_string(cfg, 0);
4110 char *qmtname, *uuid, *p;
4111 struct lustre_cfg_bufs *bufs;
4112 struct lustre_cfg *lcfg;
4113 struct lustre_profile *lprof;
4114 struct obd_connect_data *data;
4118 LASSERT(mdt->mdt_qmt_exp == NULL);
4119 LASSERT(mdt->mdt_qmt_dev == NULL);
4121 /* quota master is on MDT0 only for now */
4122 if (mdt->mdt_seq_site.ss_node_id != 0)
4125 /* MGS generates config commands which look as follows:
4126 * #01 (160)setup 0:lustre-MDT0000 1:lustre-MDT0000_UUID 2:0
4127 * 3:lustre-MDT0000-mdtlov 4:f
4129 * We generate the QMT name from the MDT one, just replacing MD with QM
4130 * after all the preparations, the logical equivalent will be:
4131 * #01 (160)setup 0:lustre-QMT0000 1:lustre-QMT0000_UUID 2:0
4132 * 3:lustre-MDT0000-osd 4:f */
4133 OBD_ALLOC(qmtname, MAX_OBD_NAME);
4134 OBD_ALLOC(uuid, UUID_MAX);
4135 OBD_ALLOC_PTR(bufs);
4136 OBD_ALLOC_PTR(data);
4137 if (qmtname == NULL || uuid == NULL || bufs == NULL || data == NULL)
4138 GOTO(cleanup_mem, rc = -ENOMEM);
4140 strcpy(qmtname, dev);
4141 p = strstr(qmtname, "-MDT");
4143 GOTO(cleanup_mem, rc = -ENOMEM);
4144 /* replace MD with QM */
4148 snprintf(uuid, UUID_MAX, "%s_UUID", qmtname);
4150 lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4151 if (lprof == NULL || lprof->lp_dt == NULL) {
4152 CERROR("can't find profile for %s\n",
4153 lustre_cfg_string(cfg, 0));
4154 GOTO(cleanup_mem, rc = -EINVAL);
4157 lustre_cfg_bufs_reset(bufs, qmtname);
4158 lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_QMT_NAME);
4159 lustre_cfg_bufs_set_string(bufs, 2, uuid);
4160 lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4162 lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
4164 GOTO(cleanup_mem, rc = -ENOMEM);
4166 rc = class_attach(lcfg);
4168 GOTO(lcfg_cleanup, rc);
4170 obd = class_name2obd(qmtname);
4172 CERROR("Can not find obd %s (%s in config)\n", qmtname,
4173 lustre_cfg_string(cfg, 0));
4174 GOTO(lcfg_cleanup, rc = -EINVAL);
4177 lustre_cfg_free(lcfg);
4179 lustre_cfg_bufs_reset(bufs, qmtname);
4180 lustre_cfg_bufs_set_string(bufs, 1, uuid);
4181 lustre_cfg_bufs_set_string(bufs, 2, dev);
4183 /* for quota, the next device should be the OSD device */
4184 lustre_cfg_bufs_set_string(bufs, 3,
4185 mdt->mdt_bottom->dd_lu_dev.ld_obd->obd_name);
4187 lcfg = lustre_cfg_new(LCFG_SETUP, bufs);
4189 GOTO(class_detach, rc = -ENOMEM);
4191 rc = class_setup(obd, lcfg);
4193 GOTO(class_detach, rc);
4195 mdt->mdt_qmt_dev = obd->obd_lu_dev;
4197 /* configure local quota objects */
4198 rc = mdt->mdt_qmt_dev->ld_ops->ldo_prepare(env,
4202 GOTO(class_cleanup, rc);
4204 /* connect to quota master target */
4205 data->ocd_connect_flags = OBD_CONNECT_VERSION;
4206 data->ocd_version = LUSTRE_VERSION_CODE;
4207 rc = obd_connect(NULL, &mdt->mdt_qmt_exp, obd, &obd->obd_uuid,
4210 CERROR("cannot connect to quota master device %s (%d)\n",
4212 GOTO(class_cleanup, rc);
4218 class_manual_cleanup(obd);
4219 mdt->mdt_qmt_dev = NULL;
4223 class_detach(obd, lcfg);
4225 lustre_cfg_free(lcfg);
4230 OBD_FREE(qmtname, MAX_OBD_NAME);
4232 OBD_FREE(uuid, UUID_MAX);
4238 /* Shutdown quota master target associated with mdt */
4239 static void mdt_quota_fini(const struct lu_env *env, struct mdt_device *mdt)
4243 if (mdt->mdt_qmt_exp == NULL)
4245 LASSERT(mdt->mdt_qmt_dev != NULL);
4247 /* the qmt automatically shuts down when the mdt disconnects */
4248 obd_disconnect(mdt->mdt_qmt_exp);
4249 mdt->mdt_qmt_exp = NULL;
4250 mdt->mdt_qmt_dev = NULL;
4254 /* mdt_getxattr() is used from mdt_intent_getxattr(), use this wrapper
4255 * for now. This will be removed along with converting rest of MDT code
4256 * to use tgt_session_info */
4257 int mdt_tgt_getxattr(struct tgt_session_info *tsi)
4259 struct mdt_thread_info *info = tsi2mdt_info(tsi);
4262 rc = mdt_getxattr(info);
4264 mdt_thread_info_fini(info);
4268 static struct tgt_handler mdt_tgt_handlers[] = {
4269 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4270 0, MDS_CONNECT, mdt_tgt_connect,
4271 &RQF_CONNECT, LUSTRE_OBD_VERSION),
4272 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4273 0, MDS_DISCONNECT, tgt_disconnect,
4274 &RQF_MDS_DISCONNECT, LUSTRE_OBD_VERSION),
4275 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4276 HABEO_REFERO, MDS_SET_INFO, mdt_set_info,
4277 &RQF_OBD_SET_INFO, LUSTRE_MDS_VERSION),
4278 TGT_MDT_HDL(0, MDS_GET_INFO, mdt_get_info),
4279 TGT_MDT_HDL(0 | HABEO_REFERO, MDS_GETSTATUS, mdt_getstatus),
4280 TGT_MDT_HDL(HABEO_CORPUS, MDS_GETATTR, mdt_getattr),
4281 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_GETATTR_NAME,
4283 TGT_MDT_HDL(HABEO_CORPUS, MDS_GETXATTR, mdt_tgt_getxattr),
4284 TGT_MDT_HDL(0 | HABEO_REFERO, MDS_STATFS, mdt_statfs),
4285 TGT_MDT_HDL(0 | MUTABOR, MDS_REINT, mdt_reint),
4286 TGT_MDT_HDL(HABEO_CORPUS, MDS_CLOSE, mdt_close),
4287 TGT_MDT_HDL(HABEO_CORPUS, MDS_DONE_WRITING,
4289 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_READPAGE, mdt_readpage),
4290 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_SYNC, mdt_sync),
4291 TGT_MDT_HDL(0, MDS_QUOTACTL, mdt_quotactl),
4292 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_PROGRESS,
4294 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_REGISTER,
4295 mdt_hsm_ct_register),
4296 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_UNREGISTER,
4297 mdt_hsm_ct_unregister),
4298 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_STATE_GET,
4300 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_STATE_SET,
4302 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_ACTION, mdt_hsm_action),
4303 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_REQUEST,
4305 TGT_MDT_HDL(HABEO_CLAVIS | HABEO_CORPUS | HABEO_REFERO | MUTABOR,
4310 static struct tgt_handler mdt_sec_ctx_ops[] = {
4311 TGT_SEC_HDL_VAR(0, SEC_CTX_INIT, mdt_sec_ctx_handle),
4312 TGT_SEC_HDL_VAR(0, SEC_CTX_INIT_CONT,mdt_sec_ctx_handle),
4313 TGT_SEC_HDL_VAR(0, SEC_CTX_FINI, mdt_sec_ctx_handle)
4316 static struct tgt_handler mdt_quota_ops[] = {
4317 TGT_QUOTA_HDL(HABEO_REFERO, QUOTA_DQACQ, mdt_quota_dqacq),
4320 static struct tgt_opc_slice mdt_common_slice[] = {
4322 .tos_opc_start = MDS_FIRST_OPC,
4323 .tos_opc_end = MDS_LAST_OPC,
4324 .tos_hs = mdt_tgt_handlers
4327 .tos_opc_start = OBD_FIRST_OPC,
4328 .tos_opc_end = OBD_LAST_OPC,
4329 .tos_hs = tgt_obd_handlers
4332 .tos_opc_start = LDLM_FIRST_OPC,
4333 .tos_opc_end = LDLM_LAST_OPC,
4334 .tos_hs = tgt_dlm_handlers
4337 .tos_opc_start = SEC_FIRST_OPC,
4338 .tos_opc_end = SEC_LAST_OPC,
4339 .tos_hs = mdt_sec_ctx_ops
4342 .tos_opc_start = OUT_UPDATE_FIRST_OPC,
4343 .tos_opc_end = OUT_UPDATE_LAST_OPC,
4344 .tos_hs = tgt_out_handlers
4347 .tos_opc_start = FLD_FIRST_OPC,
4348 .tos_opc_end = FLD_LAST_OPC,
4349 .tos_hs = fld_handlers
4352 .tos_opc_start = SEQ_FIRST_OPC,
4353 .tos_opc_end = SEQ_LAST_OPC,
4354 .tos_hs = seq_handlers
4357 .tos_opc_start = QUOTA_DQACQ,
4358 .tos_opc_end = QUOTA_LAST_OPC,
4359 .tos_hs = mdt_quota_ops
4362 .tos_opc_start = LLOG_FIRST_OPC,
4363 .tos_opc_end = LLOG_LAST_OPC,
4364 .tos_hs = tgt_llog_handlers
4367 .tos_opc_start = LFSCK_FIRST_OPC,
4368 .tos_opc_end = LFSCK_LAST_OPC,
4369 .tos_hs = tgt_lfsck_handlers
4377 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
4379 struct md_device *next = m->mdt_child;
4380 struct lu_device *d = &m->mdt_lu_dev;
4381 struct obd_device *obd = mdt2obd_dev(m);
4382 struct lfsck_stop stop;
4385 stop.ls_status = LS_PAUSED;
4387 next->md_ops->mdo_iocontrol(env, next, OBD_IOC_STOP_LFSCK, 0, &stop);
4389 target_recovery_fini(obd);
4390 ping_evictor_stop();
4391 mdt_stack_pre_fini(env, m, md2lu_dev(m->mdt_child));
4393 if (m->mdt_opts.mo_coordinator)
4394 mdt_hsm_cdt_stop(m);
4396 mdt_hsm_cdt_fini(m);
4398 mdt_llog_ctxt_unclone(env, m, LLOG_AGENT_ORIG_CTXT);
4399 mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4400 obd_exports_barrier(obd);
4401 obd_zombie_barrier();
4405 tgt_fini(env, &m->mdt_lut);
4406 mdt_fs_cleanup(env, m);
4407 upcall_cache_cleanup(m->mdt_identity_cache);
4408 m->mdt_identity_cache = NULL;
4410 if (m->mdt_namespace != NULL) {
4411 ldlm_namespace_free(m->mdt_namespace, NULL,
4412 d->ld_obd->obd_force);
4413 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
4416 mdt_quota_fini(env, m);
4418 cfs_free_nidlist(&m->mdt_squash.rsi_nosquash_nids);
4420 mdt_seq_fini(env, m);
4421 mdt_fld_fini(env, m);
4423 next->md_ops->mdo_init_capa_ctxt(env, next, 0, 0, 0, NULL);
4424 cfs_timer_disarm(&m->mdt_ck_timer);
4425 mdt_ck_thread_stop(m);
4430 mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4432 LASSERT(atomic_read(&d->ld_ref) == 0);
4434 server_put_mount(mdt_obd_name(m), true);
4439 static int mdt_postrecov(const struct lu_env *, struct mdt_device *);
4441 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
4442 struct lu_device_type *ldt, struct lustre_cfg *cfg)
4444 struct mdt_thread_info *info;
4445 struct obd_device *obd;
4446 const char *dev = lustre_cfg_string(cfg, 0);
4447 const char *num = lustre_cfg_string(cfg, 2);
4448 struct lustre_mount_info *lmi = NULL;
4449 struct lustre_sb_info *lsi;
4451 struct seq_server_site *ss_site;
4452 const char *identity_upcall = "NONE";
4453 struct md_device *next;
4459 lu_device_init(&m->mdt_lu_dev, ldt);
4461 * Environment (env) might be missing mdt_thread_key values at that
4462 * point, if device is allocated when mdt_thread_key is in QUIESCENT
4465 * Usually device allocation path doesn't use module key values, but
4466 * mdt has to do a lot of work here, so allocate key value.
4468 rc = lu_env_refill((struct lu_env *)env);
4472 info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4473 LASSERT(info != NULL);
4475 obd = class_name2obd(dev);
4476 LASSERT(obd != NULL);
4478 m->mdt_max_mdsize = MAX_MD_SIZE; /* 4 stripes */
4480 m->mdt_som_conf = 0;
4482 m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
4484 /* default is coordinator off, it is started through conf_param
4486 m->mdt_opts.mo_coordinator = 0;
4488 lmi = server_get_mount(dev);
4490 CERROR("Cannot get mount info for %s!\n", dev);
4493 lsi = s2lsi(lmi->lmi_sb);
4494 /* CMD is supported only in IAM mode */
4496 node_id = simple_strtol(num, NULL, 10);
4497 obd->u.obt.obt_magic = OBT_MAGIC;
4500 spin_lock_init(&m->mdt_ioepoch_lock);
4501 m->mdt_capa_timeout = CAPA_TIMEOUT;
4502 m->mdt_capa_alg = CAPA_HMAC_ALG_SHA1;
4503 m->mdt_ck_timeout = CAPA_KEY_TIMEOUT;
4504 m->mdt_squash.rsi_uid = 0;
4505 m->mdt_squash.rsi_gid = 0;
4506 INIT_LIST_HEAD(&m->mdt_squash.rsi_nosquash_nids);
4507 init_rwsem(&m->mdt_squash.rsi_sem);
4508 spin_lock_init(&m->mdt_osfs_lock);
4509 m->mdt_osfs_age = cfs_time_shift_64(-1000);
4510 m->mdt_enable_remote_dir = 0;
4511 m->mdt_enable_remote_dir_gid = 0;
4513 m->mdt_lu_dev.ld_ops = &mdt_lu_ops;
4514 m->mdt_lu_dev.ld_obd = obd;
4515 /* Set this lu_device to obd for error handling purposes. */
4516 obd->obd_lu_dev = &m->mdt_lu_dev;
4518 /* init the stack */
4519 rc = mdt_stack_init((struct lu_env *)env, m, cfg);
4521 CERROR("%s: Can't init device stack, rc %d\n",
4522 mdt_obd_name(m), rc);
4527 ss_site = mdt_seq_site(m);
4528 s->ld_seq_site = ss_site;
4531 /* set server index */
4532 ss_site->ss_node_id = node_id;
4534 /* failover is the default
4535 * FIXME: we do not failout mds0/mgs, which may cause some problems.
4536 * assumed whose ss_node_id == 0 XXX
4538 obd->obd_replayable = 1;
4539 /* No connection accepted until configurations will finish */
4540 obd->obd_no_conn = 1;
4542 if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
4543 char *str = lustre_cfg_string(cfg, 4);
4544 if (strchr(str, 'n')) {
4545 CWARN("%s: recovery disabled\n", mdt_obd_name(m));
4546 obd->obd_replayable = 0;
4550 rc = mdt_fld_init(env, mdt_obd_name(m), m);
4552 GOTO(err_fini_stack, rc);
4554 rc = mdt_seq_init(env, m);
4556 GOTO(err_fini_fld, rc);
4558 snprintf(info->mti_u.ns_name, sizeof(info->mti_u.ns_name), "%s-%s",
4559 LUSTRE_MDT_NAME, obd->obd_uuid.uuid);
4560 m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
4561 LDLM_NAMESPACE_SERVER,
4562 LDLM_NAMESPACE_GREEDY,
4564 if (m->mdt_namespace == NULL)
4565 GOTO(err_fini_seq, rc = -ENOMEM);
4567 m->mdt_namespace->ns_lvbp = m;
4568 m->mdt_namespace->ns_lvbo = &mdt_lvbo;
4570 ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
4571 /* set obd_namespace for compatibility with old code */
4572 obd->obd_namespace = m->mdt_namespace;
4574 cfs_timer_init(&m->mdt_ck_timer, mdt_ck_timer_callback, m);
4576 rc = mdt_hsm_cdt_init(m);
4578 CERROR("%s: error initializing coordinator, rc %d\n",
4579 mdt_obd_name(m), rc);
4580 GOTO(err_free_ns, rc);
4583 rc = mdt_ck_thread_start(m);
4585 GOTO(err_free_hsm, rc);
4587 rc = tgt_init(env, &m->mdt_lut, obd, m->mdt_bottom, mdt_common_slice,
4588 OBD_FAIL_MDS_ALL_REQUEST_NET,
4589 OBD_FAIL_MDS_ALL_REPLY_NET);
4593 rc = mdt_fs_setup(env, m, obd, lsi);
4597 tgt_adapt_sptlrpc_conf(&m->mdt_lut, 1);
4599 next = m->mdt_child;
4600 rc = next->md_ops->mdo_iocontrol(env, next, OBD_IOC_GET_MNTOPT, 0,
4603 GOTO(err_fs_cleanup, rc);
4605 if (mntopts & MNTOPT_USERXATTR)
4606 m->mdt_opts.mo_user_xattr = 1;
4608 m->mdt_opts.mo_user_xattr = 0;
4610 rc = next->md_ops->mdo_maxeasize_get(env, next, &m->mdt_max_ea_size);
4612 GOTO(err_fs_cleanup, rc);
4614 if (mntopts & MNTOPT_ACL)
4615 m->mdt_opts.mo_acl = 1;
4617 m->mdt_opts.mo_acl = 0;
4619 /* XXX: to support suppgid for ACL, we enable identity_upcall
4620 * by default, otherwise, maybe got unexpected -EACCESS. */
4621 if (m->mdt_opts.mo_acl)
4622 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
4624 m->mdt_identity_cache = upcall_cache_init(mdt_obd_name(m),
4626 &mdt_identity_upcall_cache_ops);
4627 if (IS_ERR(m->mdt_identity_cache)) {
4628 rc = PTR_ERR(m->mdt_identity_cache);
4629 m->mdt_identity_cache = NULL;
4630 GOTO(err_fs_cleanup, rc);
4633 rc = mdt_procfs_init(m, dev);
4635 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
4636 GOTO(err_recovery, rc);
4639 rc = mdt_quota_init(env, m, cfg);
4641 GOTO(err_procfs, rc);
4643 m->mdt_ldlm_client = &mdt2obd_dev(m)->obd_ldlm_client;
4644 ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
4645 "mdt_ldlm_client", m->mdt_ldlm_client);
4647 ping_evictor_start();
4649 /* recovery will be started upon mdt_prepare()
4650 * when the whole stack is complete and ready
4651 * to serve the requests */
4653 mdt_init_capa_ctxt(env, m);
4655 /* Reduce the initial timeout on an MDS because it doesn't need such
4656 * a long timeout as an OST does. Adaptive timeouts will adjust this
4657 * value appropriately. */
4658 if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
4659 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
4665 target_recovery_fini(obd);
4666 upcall_cache_cleanup(m->mdt_identity_cache);
4667 m->mdt_identity_cache = NULL;
4669 mdt_fs_cleanup(env, m);
4671 tgt_fini(env, &m->mdt_lut);
4673 cfs_timer_disarm(&m->mdt_ck_timer);
4674 mdt_ck_thread_stop(m);
4676 mdt_hsm_cdt_fini(m);
4678 ldlm_namespace_free(m->mdt_namespace, NULL, 0);
4679 obd->obd_namespace = m->mdt_namespace = NULL;
4681 mdt_seq_fini(env, m);
4683 mdt_fld_fini(env, m);
4685 mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4688 server_put_mount(dev, true);
4692 /* For interoperability, the left element is old parameter, the right one
4693 * is the new version of the parameter, if some parameter is deprecated,
4694 * the new version should be set as NULL. */
4695 static struct cfg_interop_param mdt_interop_param[] = {
4696 { "mdt.group_upcall", NULL },
4697 { "mdt.quota_type", NULL },
4698 { "mdd.quota_type", NULL },
4699 { "mdt.rootsquash", "mdt.root_squash" },
4700 { "mdt.nosquash_nid", "mdt.nosquash_nids" },
4704 /* used by MGS to process specific configurations */
4705 static int mdt_process_config(const struct lu_env *env,
4706 struct lu_device *d, struct lustre_cfg *cfg)
4708 struct mdt_device *m = mdt_dev(d);
4709 struct md_device *md_next = m->mdt_child;
4710 struct lu_device *next = md2lu_dev(md_next);
4714 switch (cfg->lcfg_command) {
4716 struct obd_device *obd = d->ld_obd;
4718 /* For interoperability */
4719 struct cfg_interop_param *ptr = NULL;
4720 struct lustre_cfg *old_cfg = NULL;
4723 param = lustre_cfg_string(cfg, 1);
4724 if (param == NULL) {
4725 CERROR("param is empty\n");
4730 ptr = class_find_old_param(param, mdt_interop_param);
4732 if (ptr->new_param == NULL) {
4734 CWARN("For interoperability, skip this %s."
4735 " It is obsolete.\n", ptr->old_param);
4739 CWARN("Found old param %s, changed it to %s.\n",
4740 ptr->old_param, ptr->new_param);
4743 cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
4750 rc = class_process_proc_param(PARAM_MDT, obd->obd_vars,
4752 if (rc > 0 || rc == -ENOSYS) {
4753 /* is it an HSM var ? */
4754 rc = class_process_proc_param(PARAM_HSM,
4755 hsm_cdt_get_proc_vars(),
4757 if (rc > 0 || rc == -ENOSYS)
4758 /* we don't understand; pass it on */
4759 rc = next->ld_ops->ldo_process_config(env, next,
4763 if (old_cfg != NULL)
4764 lustre_cfg_free(cfg);
4769 /* others are passed further */
4770 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4776 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
4777 const struct lu_object_header *hdr,
4778 struct lu_device *d)
4780 struct mdt_object *mo;
4784 OBD_SLAB_ALLOC_PTR_GFP(mo, mdt_object_kmem, GFP_NOFS);
4786 struct lu_object *o;
4787 struct lu_object_header *h;
4790 h = &mo->mot_header;
4791 lu_object_header_init(h);
4792 lu_object_init(o, h, d);
4793 lu_object_add_top(h, o);
4794 o->lo_ops = &mdt_obj_ops;
4795 mutex_init(&mo->mot_ioepoch_mutex);
4796 mutex_init(&mo->mot_lov_mutex);
4797 init_rwsem(&mo->mot_open_sem);
4803 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
4804 const struct lu_object_conf *unused)
4806 struct mdt_device *d = mdt_dev(o->lo_dev);
4807 struct lu_device *under;
4808 struct lu_object *below;
4812 CDEBUG(D_INFO, "object init, fid = "DFID"\n",
4813 PFID(lu_object_fid(o)));
4815 under = &d->mdt_child->md_lu_dev;
4816 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
4817 if (below != NULL) {
4818 lu_object_add(o, below);
4825 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
4827 struct mdt_object *mo = mdt_obj(o);
4828 struct lu_object_header *h;
4832 CDEBUG(D_INFO, "object free, fid = "DFID"\n",
4833 PFID(lu_object_fid(o)));
4835 LASSERT(atomic_read(&mo->mot_open_count) == 0);
4836 LASSERT(atomic_read(&mo->mot_lease_count) == 0);
4839 lu_object_header_fini(h);
4840 OBD_SLAB_FREE_PTR(mo, mdt_object_kmem);
4845 static int mdt_object_print(const struct lu_env *env, void *cookie,
4846 lu_printer_t p, const struct lu_object *o)
4848 struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
4849 return (*p)(env, cookie, LUSTRE_MDT_NAME"-object@%p(ioepoch="LPU64" "
4850 "flags="LPX64", epochcount=%d, writecount=%d)",
4851 mdto, mdto->mot_ioepoch, mdto->mot_flags,
4852 mdto->mot_ioepoch_count, mdto->mot_writecount);
4855 static int mdt_prepare(const struct lu_env *env,
4856 struct lu_device *pdev,
4857 struct lu_device *cdev)
4859 struct mdt_device *mdt = mdt_dev(cdev);
4860 struct lu_device *next = &mdt->mdt_child->md_lu_dev;
4861 struct obd_device *obd = cdev->ld_obd;
4868 rc = next->ld_ops->ldo_prepare(env, cdev, next);
4872 rc = mdt_llog_ctxt_clone(env, mdt, LLOG_CHANGELOG_ORIG_CTXT);
4876 rc = mdt_llog_ctxt_clone(env, mdt, LLOG_AGENT_ORIG_CTXT);
4880 rc = lfsck_register_namespace(env, mdt->mdt_bottom, mdt->mdt_namespace);
4881 /* The LFSCK instance is registered just now, so it must be there when
4882 * register the namespace to such instance. */
4883 LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
4885 if (mdt->mdt_seq_site.ss_node_id == 0) {
4886 rc = mdt->mdt_child->md_ops->mdo_root_get(env, mdt->mdt_child,
4887 &mdt->mdt_md_root_fid);
4892 LASSERT(!test_bit(MDT_FL_CFGLOG, &mdt->mdt_state));
4893 target_recovery_init(&mdt->mdt_lut, tgt_request_handle);
4894 set_bit(MDT_FL_CFGLOG, &mdt->mdt_state);
4895 LASSERT(obd->obd_no_conn);
4896 spin_lock(&obd->obd_dev_lock);
4897 obd->obd_no_conn = 0;
4898 spin_unlock(&obd->obd_dev_lock);
4900 if (obd->obd_recovering == 0)
4901 mdt_postrecov(env, mdt);
4906 const struct lu_device_operations mdt_lu_ops = {
4907 .ldo_object_alloc = mdt_object_alloc,
4908 .ldo_process_config = mdt_process_config,
4909 .ldo_prepare = mdt_prepare,
4912 static const struct lu_object_operations mdt_obj_ops = {
4913 .loo_object_init = mdt_object_init,
4914 .loo_object_free = mdt_object_free,
4915 .loo_object_print = mdt_object_print
4918 static int mdt_obd_set_info_async(const struct lu_env *env,
4919 struct obd_export *exp,
4920 __u32 keylen, void *key,
4921 __u32 vallen, void *val,
4922 struct ptlrpc_request_set *set)
4928 if (KEY_IS(KEY_SPTLRPC_CONF)) {
4929 rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp), 0);
4937 * Match client and server connection feature flags.
4939 * Compute the compatibility flags for a connection request based on
4940 * features mutually supported by client and server.
4942 * The obd_export::exp_connect_data.ocd_connect_flags field in \a exp
4943 * must not be updated here, otherwise a partially initialized value may
4944 * be exposed. After the connection request is successfully processed,
4945 * the top-level MDT connect request handler atomically updates the export
4946 * connect flags from the obd_connect_data::ocd_connect_flags field of the
4947 * reply. \see mdt_connect().
4949 * \param exp the obd_export associated with this client/target pair
4950 * \param mdt the target device for the connection
4951 * \param data stores data for this connect request
4954 * \retval -EPROTO \a data unexpectedly has zero obd_connect_data::ocd_brw_size
4955 * \retval -EBADE client and server feature requirements are incompatible
4957 static int mdt_connect_internal(struct obd_export *exp,
4958 struct mdt_device *mdt,
4959 struct obd_connect_data *data)
4961 LASSERT(data != NULL);
4963 data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
4964 data->ocd_ibits_known &= MDS_INODELOCK_FULL;
4966 if (!(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
4967 !(data->ocd_connect_flags & OBD_CONNECT_IBITS)) {
4968 CWARN("%s: client %s does not support ibits lock, either "
4969 "very old or an invalid client: flags "LPX64"\n",
4970 mdt_obd_name(mdt), exp->exp_client_uuid.uuid,
4971 data->ocd_connect_flags);
4975 if (!mdt->mdt_opts.mo_acl)
4976 data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
4978 if (!mdt->mdt_opts.mo_user_xattr)
4979 data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
4981 if (!mdt->mdt_som_conf)
4982 data->ocd_connect_flags &= ~OBD_CONNECT_SOM;
4984 if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE) {
4985 data->ocd_brw_size = min(data->ocd_brw_size,
4986 (__u32)MD_MAX_BRW_SIZE);
4987 if (data->ocd_brw_size == 0) {
4988 CERROR("%s: cli %s/%p ocd_connect_flags: "LPX64
4989 " ocd_version: %x ocd_grant: %d "
4990 "ocd_index: %u ocd_brw_size is "
4991 "unexpectedly zero, network data "
4992 "corruption? Refusing connection of this"
4995 exp->exp_client_uuid.uuid,
4996 exp, data->ocd_connect_flags, data->ocd_version,
4997 data->ocd_grant, data->ocd_index);
5002 /* NB: Disregard the rule against updating
5003 * exp_connect_data.ocd_connect_flags in this case, since
5004 * tgt_client_new() needs to know if this is a lightweight
5005 * connection, and it is safe to expose this flag before
5006 * connection processing completes. */
5007 if (data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) {
5008 spin_lock(&exp->exp_lock);
5009 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_LIGHTWEIGHT;
5010 spin_unlock(&exp->exp_lock);
5013 data->ocd_version = LUSTRE_VERSION_CODE;
5015 if ((data->ocd_connect_flags & OBD_CONNECT_FID) == 0) {
5016 CWARN("%s: MDS requires FID support, but client not\n",
5021 if (mdt->mdt_som_conf &&
5022 !(data->ocd_connect_flags & (OBD_CONNECT_LIGHTWEIGHT |
5023 OBD_CONNECT_MDS_MDS |
5024 OBD_CONNECT_SOM))) {
5025 CWARN("%s: MDS has SOM enabled, but client does not support "
5026 "it\n", mdt_obd_name(mdt));
5030 if (OCD_HAS_FLAG(data, PINGLESS)) {
5031 if (ptlrpc_pinger_suppress_pings()) {
5032 spin_lock(&exp->exp_obd->obd_dev_lock);
5033 list_del_init(&exp->exp_obd_chain_timed);
5034 spin_unlock(&exp->exp_obd->obd_dev_lock);
5036 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
5040 data->ocd_max_easize = mdt->mdt_max_ea_size;
5045 static int mdt_ctxt_add_dirty_flag(struct lu_env *env,
5046 struct mdt_thread_info *info,
5047 struct mdt_file_data *mfd)
5049 struct lu_context ses;
5053 rc = lu_context_init(&ses, LCT_SERVER_SESSION);
5058 lu_context_enter(&ses);
5060 mdt_ucred(info)->uc_valid = UCRED_OLD;
5061 rc = mdt_add_dirty_flag(info, mfd->mfd_object, &info->mti_attr);
5063 lu_context_exit(&ses);
5064 lu_context_fini(&ses);
5070 static int mdt_export_cleanup(struct obd_export *exp)
5072 struct list_head closing_list;
5073 struct mdt_export_data *med = &exp->exp_mdt_data;
5074 struct obd_device *obd = exp->exp_obd;
5075 struct mdt_device *mdt;
5076 struct mdt_thread_info *info;
5078 struct mdt_file_data *mfd, *n;
5082 INIT_LIST_HEAD(&closing_list);
5083 spin_lock(&med->med_open_lock);
5084 while (!list_empty(&med->med_open_head)) {
5085 struct list_head *tmp = med->med_open_head.next;
5086 mfd = list_entry(tmp, struct mdt_file_data, mfd_list);
5088 /* Remove mfd handle so it can't be found again.
5089 * We are consuming the mfd_list reference here. */
5090 class_handle_unhash(&mfd->mfd_handle);
5091 list_move_tail(&mfd->mfd_list, &closing_list);
5093 spin_unlock(&med->med_open_lock);
5094 mdt = mdt_dev(obd->obd_lu_dev);
5095 LASSERT(mdt != NULL);
5097 rc = lu_env_init(&env, LCT_MD_THREAD);
5101 info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5102 LASSERT(info != NULL);
5103 memset(info, 0, sizeof *info);
5104 info->mti_env = &env;
5105 info->mti_mdt = mdt;
5106 info->mti_exp = exp;
5108 if (!list_empty(&closing_list)) {
5109 struct md_attr *ma = &info->mti_attr;
5111 /* Close any open files (which may also cause orphan
5113 list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
5114 list_del_init(&mfd->mfd_list);
5115 ma->ma_need = ma->ma_valid = 0;
5117 /* This file is being closed due to an eviction, it
5118 * could have been modified and now dirty regarding to
5119 * HSM archive, check this!
5120 * The logic here is to mark a file dirty if there's a
5121 * chance it was dirtied before the client was evicted,
5122 * so that we don't have to wait for a release attempt
5123 * before finding out the file was actually dirty and
5124 * fail the release. Aggressively marking it dirty here
5125 * will cause the policy engine to attempt to
5126 * re-archive it; when rearchiving, we can compare the
5127 * current version to the HSM data_version and make the
5128 * archive request into a noop if it's not actually
5131 if (mfd->mfd_mode & (FMODE_WRITE|MDS_FMODE_TRUNC))
5132 rc = mdt_ctxt_add_dirty_flag(&env, info, mfd);
5134 /* Don't unlink orphan on failover umount, LU-184 */
5135 if (exp->exp_flags & OBD_OPT_FAILOVER) {
5136 ma->ma_valid = MA_FLAGS;
5137 ma->ma_attr_flags |= MDS_KEEP_ORPHAN;
5139 mdt_mfd_close(info, mfd);
5142 info->mti_mdt = NULL;
5143 /* cleanup client slot early */
5144 /* Do not erase record for recoverable client. */
5145 if (!(exp->exp_flags & OBD_OPT_FAILOVER) || exp->exp_failed)
5146 tgt_client_del(&env, exp);
5152 static int mdt_obd_disconnect(struct obd_export *exp)
5158 class_export_get(exp);
5160 nodemap_del_member(exp);
5161 rc = server_disconnect_export(exp);
5163 CDEBUG(D_IOCTL, "server disconnect error: rc = %d\n", rc);
5165 rc = mdt_export_cleanup(exp);
5166 class_export_put(exp);
5170 /* mds_connect copy */
5171 static int mdt_obd_connect(const struct lu_env *env,
5172 struct obd_export **exp, struct obd_device *obd,
5173 struct obd_uuid *cluuid,
5174 struct obd_connect_data *data,
5177 struct obd_export *lexp;
5178 struct lustre_handle conn = { 0 };
5179 struct mdt_device *mdt;
5181 lnet_nid_t *client_nid = localdata;
5184 LASSERT(env != NULL);
5185 if (!exp || !obd || !cluuid)
5188 mdt = mdt_dev(obd->obd_lu_dev);
5191 * first, check whether the stack is ready to handle requests
5192 * XXX: probably not very appropriate method is used now
5193 * at some point we should find a better one
5195 if (!test_bit(MDT_FL_SYNCED, &mdt->mdt_state) && data != NULL &&
5196 !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT)) {
5197 rc = obd_get_info(env, mdt->mdt_child_exp,
5198 sizeof(KEY_OSP_CONNECTED),
5199 KEY_OSP_CONNECTED, NULL, NULL, NULL);
5202 set_bit(MDT_FL_SYNCED, &mdt->mdt_state);
5205 rc = class_connect(&conn, obd, cluuid);
5209 lexp = class_conn2export(&conn);
5210 LASSERT(lexp != NULL);
5212 rc = mdt_connect_internal(lexp, mdt, data);
5214 struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
5217 memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
5218 rc = tgt_client_new(env, lexp);
5220 rc = nodemap_add_member(*client_nid, lexp);
5221 if (rc != 0 && rc != -EEXIST)
5224 mdt_export_stats_init(obd, lexp, localdata);
5227 /* For phase I, sync for cross-ref operation. */
5228 spin_lock(&lexp->exp_lock);
5229 lexp->exp_keep_sync = 1;
5230 spin_unlock(&lexp->exp_lock);
5234 class_disconnect(lexp);
5243 static int mdt_obd_reconnect(const struct lu_env *env,
5244 struct obd_export *exp, struct obd_device *obd,
5245 struct obd_uuid *cluuid,
5246 struct obd_connect_data *data,
5249 lnet_nid_t *client_nid = localdata;
5253 if (exp == NULL || obd == NULL || cluuid == NULL)
5256 rc = mdt_connect_internal(exp, mdt_dev(obd->obd_lu_dev), data);
5258 rc = nodemap_add_member(*client_nid, exp);
5259 if (rc == 0 || rc == -EEXIST)
5260 mdt_export_stats_init(obd, exp, localdata);
5266 /* FIXME: Can we avoid using these two interfaces? */
5267 static int mdt_init_export(struct obd_export *exp)
5269 struct mdt_export_data *med = &exp->exp_mdt_data;
5273 INIT_LIST_HEAD(&med->med_open_head);
5274 spin_lock_init(&med->med_open_lock);
5275 mutex_init(&med->med_idmap_mutex);
5276 med->med_idmap = NULL;
5277 spin_lock(&exp->exp_lock);
5278 exp->exp_connecting = 1;
5279 spin_unlock(&exp->exp_lock);
5281 /* self-export doesn't need client data and ldlm initialization */
5282 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5283 &exp->exp_client_uuid)))
5286 rc = tgt_client_alloc(exp);
5290 rc = ldlm_init_export(exp);
5297 tgt_client_free(exp);
5299 CERROR("%s: Failed to initialize export: rc = %d\n",
5300 exp->exp_obd->obd_name, rc);
5304 static int mdt_destroy_export(struct obd_export *exp)
5308 if (exp_connect_rmtclient(exp))
5309 mdt_cleanup_idmap(&exp->exp_mdt_data);
5311 target_destroy_export(exp);
5312 /* destroy can be called from failed obd_setup, so
5313 * checking uuid is safer than obd_self_export */
5314 if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5315 &exp->exp_client_uuid)))
5318 ldlm_destroy_export(exp);
5319 tgt_client_free(exp);
5321 LASSERT(list_empty(&exp->exp_outstanding_replies));
5322 LASSERT(list_empty(&exp->exp_mdt_data.med_open_head));
5327 int mdt_links_read(struct mdt_thread_info *info, struct mdt_object *mdt_obj,
5328 struct linkea_data *ldata)
5332 LASSERT(ldata->ld_buf->lb_buf != NULL);
5334 if (!mdt_object_exists(mdt_obj))
5337 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5338 ldata->ld_buf, XATTR_NAME_LINK);
5339 if (rc == -ERANGE) {
5340 /* Buf was too small, figure out what we need. */
5341 lu_buf_free(ldata->ld_buf);
5342 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5343 ldata->ld_buf, XATTR_NAME_LINK);
5346 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
5347 if (ldata->ld_buf->lb_buf == NULL)
5349 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5350 ldata->ld_buf, XATTR_NAME_LINK);
5355 return linkea_init(ldata);
5359 * Given an MDT object, try to look up the full path to the object.
5360 * Part of the MDT layer implementation of lfs fid2path.
5362 * \param[in] info Per-thread common data shared by MDT level handlers.
5363 * \param[in] obj Object to do path lookup of
5364 * \param[in,out] fp User-provided struct to store path information
5366 * \retval 0 Lookup successful, path information stored in fp
5367 * \retval -EAGAIN Lookup failed, usually because object is being moved
5368 * \retval negative errno if there was a problem
5370 static int mdt_path_current(struct mdt_thread_info *info,
5371 struct mdt_object *obj,
5372 struct getinfo_fid2path *fp)
5374 struct mdt_device *mdt = info->mti_mdt;
5375 struct mdt_object *mdt_obj;
5376 struct link_ea_header *leh;
5377 struct link_ea_entry *lee;
5378 struct lu_name *tmpname = &info->mti_name;
5379 struct lu_fid *tmpfid = &info->mti_tmp_fid1;
5380 struct lu_buf *buf = &info->mti_big_buf;
5383 struct linkea_data ldata = { 0 };
5388 /* temp buffer for path element, the buffer will be finally freed
5389 * in mdt_thread_info_fini */
5390 buf = lu_buf_check_and_alloc(buf, PATH_MAX);
5391 if (buf->lb_buf == NULL)
5395 ptr = fp->gf_path + fp->gf_pathlen - 1;
5398 *tmpfid = fp->gf_fid = *mdt_object_fid(obj);
5400 /* root FID only exists on MDT0, and fid2path should also ends at MDT0,
5401 * so checking root_fid can only happen on MDT0. */
5402 while (!lu_fid_eq(&mdt->mdt_md_root_fid, &fp->gf_fid)) {
5403 struct lu_buf lmv_buf;
5405 mdt_obj = mdt_object_find(info->mti_env, mdt, tmpfid);
5406 if (IS_ERR(mdt_obj))
5407 GOTO(out, rc = PTR_ERR(mdt_obj));
5409 if (!mdt_object_exists(mdt_obj)) {
5410 mdt_object_put(info->mti_env, mdt_obj);
5411 GOTO(out, rc = -ENOENT);
5414 if (mdt_object_remote(mdt_obj)) {
5415 mdt_object_put(info->mti_env, mdt_obj);
5416 GOTO(remote_out, rc = -EREMOTE);
5419 rc = mdt_links_read(info, mdt_obj, &ldata);
5421 mdt_object_put(info->mti_env, mdt_obj);
5426 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
5427 linkea_entry_unpack(lee, &reclen, tmpname, tmpfid);
5428 /* If set, use link #linkno for path lookup, otherwise use
5429 link #0. Only do this for the final path element. */
5430 if (first && fp->gf_linkno < leh->leh_reccount) {
5432 for (count = 0; count < fp->gf_linkno; count++) {
5433 lee = (struct link_ea_entry *)
5434 ((char *)lee + reclen);
5435 linkea_entry_unpack(lee, &reclen, tmpname,
5438 if (fp->gf_linkno < leh->leh_reccount - 1)
5439 /* indicate to user there are more links */
5443 lmv_buf.lb_buf = info->mti_xattr_buf;
5444 lmv_buf.lb_len = sizeof(info->mti_xattr_buf);
5445 /* Check if it is slave stripes */
5446 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5447 &lmv_buf, XATTR_NAME_LMV);
5448 mdt_object_put(info->mti_env, mdt_obj);
5450 union lmv_mds_md *lmm = lmv_buf.lb_buf;
5452 /* For slave stripes, get its master */
5453 if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE) {
5454 fp->gf_fid = *tmpfid;
5457 } else if (rc < 0 && rc != -ENODATA) {
5463 /* Pack the name in the end of the buffer */
5464 ptr -= tmpname->ln_namelen;
5465 if (ptr - 1 <= fp->gf_path)
5466 GOTO(out, rc = -EOVERFLOW);
5467 strncpy(ptr, tmpname->ln_name, tmpname->ln_namelen);
5470 /* keep the last resolved fid to the client, so the
5471 * client will build the left path on another MDT for
5473 fp->gf_fid = *tmpfid;
5479 ptr++; /* skip leading / */
5480 memmove(fp->gf_path, ptr, fp->gf_path + fp->gf_pathlen - ptr);
5487 * Given an MDT object, use mdt_path_current to get the path.
5488 * Essentially a wrapper to retry mdt_path_current a set number of times
5489 * if -EAGAIN is returned (usually because an object is being moved).
5491 * Part of the MDT layer implementation of lfs fid2path.
5493 * \param[in] info Per-thread common data shared by mdt level handlers.
5494 * \param[in] obj Object to do path lookup of
5495 * \param[in,out] fp User-provided struct for arguments and to store path
5498 * \retval 0 Lookup successful, path information stored in fp
5499 * \retval negative errno if there was a problem
5501 static int mdt_path(struct mdt_thread_info *info, struct mdt_object *obj,
5502 struct getinfo_fid2path *fp)
5504 struct mdt_device *mdt = info->mti_mdt;
5509 if (fp->gf_pathlen < 3)
5512 if (lu_fid_eq(&mdt->mdt_md_root_fid, mdt_object_fid(obj))) {
5513 fp->gf_path[0] = '\0';
5517 /* Retry multiple times in case file is being moved */
5518 while (tries-- && rc == -EAGAIN)
5519 rc = mdt_path_current(info, obj, fp);
5525 * Get the full path of the provided FID, as of changelog record recno.
5527 * This checks sanity and looks up object for user provided FID
5528 * before calling the actual path lookup code.
5530 * Part of the MDT layer implementation of lfs fid2path.
5532 * \param[in] info Per-thread common data shared by mdt level handlers.
5533 * \param[in,out] fp User-provided struct for arguments and to store path
5536 * \retval 0 Lookup successful, path information and recno stored in fp
5537 * \retval -ENOENT, object does not exist
5538 * \retval negative errno if there was a problem
5540 static int mdt_fid2path(struct mdt_thread_info *info,
5541 struct getinfo_fid2path *fp)
5543 struct mdt_device *mdt = info->mti_mdt;
5544 struct mdt_object *obj;
5548 CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n",
5549 PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
5551 if (!fid_is_sane(&fp->gf_fid))
5554 if (!fid_is_namespace_visible(&fp->gf_fid)) {
5555 CWARN("%s: "DFID" is invalid, sequence should be "
5556 ">= "LPX64"\n", mdt_obd_name(mdt),
5557 PFID(&fp->gf_fid), (__u64)FID_SEQ_NORMAL);
5561 obj = mdt_object_find(info->mti_env, mdt, &fp->gf_fid);
5562 if (obj == NULL || IS_ERR(obj)) {
5563 CDEBUG(D_IOCTL, "no object "DFID": %ld\n", PFID(&fp->gf_fid),
5568 if (mdt_object_remote(obj))
5570 else if (!mdt_object_exists(obj))
5576 mdt_object_put(info->mti_env, obj);
5577 CDEBUG(D_IOCTL, "nonlocal object "DFID": rc = %d\n",
5578 PFID(&fp->gf_fid), rc);
5582 rc = mdt_path(info, obj, fp);
5584 CDEBUG(D_INFO, "fid "DFID", path %s recno "LPX64" linkno %u\n",
5585 PFID(&fp->gf_fid), fp->gf_path, fp->gf_recno, fp->gf_linkno);
5587 mdt_object_put(info->mti_env, obj);
5592 static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key,
5593 void *val, int vallen)
5595 struct getinfo_fid2path *fpout, *fpin;
5598 fpin = key + cfs_size_round(sizeof(KEY_FID2PATH));
5601 if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
5602 lustre_swab_fid2path(fpin);
5604 memcpy(fpout, fpin, sizeof(*fpin));
5605 if (fpout->gf_pathlen != vallen - sizeof(*fpin))
5608 rc = mdt_fid2path(info, fpout);
5612 int mdt_get_info(struct tgt_session_info *tsi)
5622 key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
5624 CDEBUG(D_IOCTL, "No GETINFO key\n");
5625 RETURN(err_serious(-EFAULT));
5627 keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
5630 vallen = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_VALLEN);
5631 if (vallen == NULL) {
5632 CDEBUG(D_IOCTL, "%s: cannot get RMF_GETINFO_VALLEN buffer\n",
5633 tgt_name(tsi->tsi_tgt));
5634 RETURN(err_serious(-EFAULT));
5637 req_capsule_set_size(tsi->tsi_pill, &RMF_GETINFO_VAL, RCL_SERVER,
5639 rc = req_capsule_server_pack(tsi->tsi_pill);
5641 RETURN(err_serious(rc));
5643 valout = req_capsule_server_get(tsi->tsi_pill, &RMF_GETINFO_VAL);
5644 if (valout == NULL) {
5645 CDEBUG(D_IOCTL, "%s: cannot get get-info RPC out buffer\n",
5646 tgt_name(tsi->tsi_tgt));
5647 RETURN(err_serious(-EFAULT));
5650 if (KEY_IS(KEY_FID2PATH)) {
5651 struct mdt_thread_info *info = tsi2mdt_info(tsi);
5653 rc = mdt_rpc_fid2path(info, key, valout, *vallen);
5654 mdt_thread_info_fini(info);
5661 /* Pass the ioc down */
5662 static int mdt_ioc_child(struct lu_env *env, struct mdt_device *mdt,
5663 unsigned int cmd, int len, void *data)
5665 struct lu_context ioctl_session;
5666 struct md_device *next = mdt->mdt_child;
5670 rc = lu_context_init(&ioctl_session, LCT_SERVER_SESSION);
5673 ioctl_session.lc_thread = (struct ptlrpc_thread *)current;
5674 lu_context_enter(&ioctl_session);
5675 env->le_ses = &ioctl_session;
5677 LASSERT(next->md_ops->mdo_iocontrol);
5678 rc = next->md_ops->mdo_iocontrol(env, next, cmd, len, data);
5680 lu_context_exit(&ioctl_session);
5681 lu_context_fini(&ioctl_session);
5685 static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
5687 struct obd_ioctl_data *data = karg;
5690 struct mdt_object *obj;
5691 struct mdt_lock_handle *lh;
5695 if (data->ioc_inlbuf1 == NULL || data->ioc_inllen1 != sizeof(*fid) ||
5696 data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
5699 fid = (struct lu_fid *)data->ioc_inlbuf1;
5701 if (!fid_is_sane(fid))
5704 CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
5706 lh = &mti->mti_lh[MDT_LH_PARENT];
5707 mdt_lock_reg_init(lh, LCK_CR);
5709 obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE);
5711 RETURN(PTR_ERR(obj));
5713 if (mdt_object_remote(obj)) {
5716 * before calling version get the correct MDS should be
5717 * fid, this is error to find remote object here
5719 CERROR("nonlocal object "DFID"\n", PFID(fid));
5720 } else if (!mdt_object_exists(obj)) {
5721 *(__u64 *)data->ioc_inlbuf2 = ENOENT_VERSION;
5724 version = dt_version_get(mti->mti_env, mdt_obj2dt(obj));
5725 *(__u64 *)data->ioc_inlbuf2 = version;
5728 mdt_object_unlock_put(mti, obj, lh, 1);
5732 /* ioctls on obd dev */
5733 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
5734 void *karg, void *uarg)
5737 struct obd_device *obd = exp->exp_obd;
5738 struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
5739 struct dt_device *dt = mdt->mdt_bottom;
5743 CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
5744 rc = lu_env_init(&env, LCT_MD_THREAD);
5750 rc = mdt_device_sync(&env, mdt);
5752 case OBD_IOC_SET_READONLY:
5753 rc = dt->dd_ops->dt_ro(&env, dt);
5755 case OBD_IOC_ABORT_RECOVERY:
5756 CERROR("%s: Aborting recovery for device\n", mdt_obd_name(mdt));
5757 target_stop_recovery_thread(obd);
5760 case OBD_IOC_CHANGELOG_REG:
5761 case OBD_IOC_CHANGELOG_DEREG:
5762 case OBD_IOC_CHANGELOG_CLEAR:
5763 rc = mdt_ioc_child(&env, mdt, cmd, len, karg);
5765 case OBD_IOC_START_LFSCK: {
5766 struct md_device *next = mdt->mdt_child;
5767 struct obd_ioctl_data *data = karg;
5768 struct lfsck_start_param lsp;
5770 if (unlikely(data == NULL)) {
5775 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
5776 lsp.lsp_index_valid = 0;
5777 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &lsp);
5780 case OBD_IOC_STOP_LFSCK: {
5781 struct md_device *next = mdt->mdt_child;
5782 struct obd_ioctl_data *data = karg;
5783 struct lfsck_stop stop;
5785 stop.ls_status = LS_STOPPED;
5786 /* Old lfsck utils may pass NULL @stop. */
5787 if (data->ioc_inlbuf1 == NULL)
5791 ((struct lfsck_stop *)(data->ioc_inlbuf1))->ls_flags;
5793 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &stop);
5796 case OBD_IOC_GET_OBJ_VERSION: {
5797 struct mdt_thread_info *mti;
5798 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5799 memset(mti, 0, sizeof *mti);
5800 mti->mti_env = &env;
5804 rc = mdt_ioc_version_get(mti, karg);
5807 case OBD_IOC_CATLOGLIST: {
5808 struct mdt_thread_info *mti;
5810 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5811 lu_local_obj_fid(&mti->mti_tmp_fid1, LLOG_CATALOGS_OID);
5812 rc = llog_catalog_list(&env, mdt->mdt_bottom, 0, karg,
5813 &mti->mti_tmp_fid1);
5818 CERROR("%s: Not supported cmd = %d, rc = %d\n",
5819 mdt_obd_name(mdt), cmd, rc);
5826 static int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
5828 struct lu_device *ld = md2lu_dev(mdt->mdt_child);
5829 struct lfsck_start_param lsp;
5833 lsp.lsp_start = NULL;
5834 lsp.lsp_index_valid = 0;
5835 rc = mdt->mdt_child->md_ops->mdo_iocontrol(env, mdt->mdt_child,
5836 OBD_IOC_START_LFSCK,
5838 if (rc != 0 && rc != -EALREADY)
5839 CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
5840 mdt_obd_name(mdt), rc);
5842 rc = ld->ld_ops->ldo_recovery_complete(env, ld);
5846 static int mdt_obd_postrecov(struct obd_device *obd)
5851 rc = lu_env_init(&env, LCT_MD_THREAD);
5854 rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
5859 static struct obd_ops mdt_obd_device_ops = {
5860 .o_owner = THIS_MODULE,
5861 .o_set_info_async = mdt_obd_set_info_async,
5862 .o_connect = mdt_obd_connect,
5863 .o_reconnect = mdt_obd_reconnect,
5864 .o_disconnect = mdt_obd_disconnect,
5865 .o_init_export = mdt_init_export,
5866 .o_destroy_export = mdt_destroy_export,
5867 .o_iocontrol = mdt_iocontrol,
5868 .o_postrecov = mdt_obd_postrecov,
5871 static struct lu_device* mdt_device_fini(const struct lu_env *env,
5872 struct lu_device *d)
5874 struct mdt_device *m = mdt_dev(d);
5881 static struct lu_device *mdt_device_free(const struct lu_env *env,
5882 struct lu_device *d)
5884 struct mdt_device *m = mdt_dev(d);
5887 lu_device_fini(&m->mdt_lu_dev);
5893 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
5894 struct lu_device_type *t,
5895 struct lustre_cfg *cfg)
5897 struct lu_device *l;
5898 struct mdt_device *m;
5905 rc = mdt_init0(env, m, t, cfg);
5907 mdt_device_free(env, l);
5912 l = ERR_PTR(-ENOMEM);
5916 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
5917 LU_KEY_INIT(mdt, struct mdt_thread_info);
5919 static void mdt_key_fini(const struct lu_context *ctx,
5920 struct lu_context_key *key, void* data)
5922 struct mdt_thread_info *info = data;
5924 if (info->mti_big_lmm) {
5925 OBD_FREE_LARGE(info->mti_big_lmm, info->mti_big_lmmsize);
5926 info->mti_big_lmm = NULL;
5927 info->mti_big_lmmsize = 0;
5932 /* context key: mdt_thread_key */
5933 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
5935 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info)
5937 return lu_ucred(info->mti_env);
5940 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info)
5942 return lu_ucred_check(info->mti_env);
5946 * Enable/disable COS (Commit On Sharing).
5948 * Set/Clear the COS flag in mdt options.
5950 * \param mdt mdt device
5951 * \param val 0 disables COS, other values enable COS
5953 void mdt_enable_cos(struct mdt_device *mdt, int val)
5958 mdt->mdt_opts.mo_cos = !!val;
5959 rc = lu_env_init(&env, LCT_LOCAL);
5960 if (unlikely(rc != 0)) {
5961 CWARN("%s: lu_env initialization failed, cannot "
5962 "sync: rc = %d\n", mdt_obd_name(mdt), rc);
5965 mdt_device_sync(&env, mdt);
5970 * Check COS (Commit On Sharing) status.
5972 * Return COS flag status.
5974 * \param mdt mdt device
5976 int mdt_cos_is_enabled(struct mdt_device *mdt)
5978 return mdt->mdt_opts.mo_cos != 0;
5981 static struct lu_device_type_operations mdt_device_type_ops = {
5982 .ldto_device_alloc = mdt_device_alloc,
5983 .ldto_device_free = mdt_device_free,
5984 .ldto_device_fini = mdt_device_fini
5987 static struct lu_device_type mdt_device_type = {
5988 .ldt_tags = LU_DEVICE_MD,
5989 .ldt_name = LUSTRE_MDT_NAME,
5990 .ldt_ops = &mdt_device_type_ops,
5991 .ldt_ctx_tags = LCT_MD_THREAD
5994 static int __init mdt_mod_init(void)
5998 CLASSERT(sizeof("0x0123456789ABCDEF:0x01234567:0x01234567") ==
5999 FID_NOBRACE_LEN + 1);
6000 CLASSERT(sizeof("[0x0123456789ABCDEF:0x01234567:0x01234567]") ==
6002 rc = lu_kmem_init(mdt_caches);
6006 rc = mds_mod_init();
6010 rc = class_register_type(&mdt_obd_device_ops, NULL, true, NULL,
6011 LUSTRE_MDT_NAME, &mdt_device_type);
6016 lu_kmem_fini(mdt_caches);
6023 static void __exit mdt_mod_exit(void)
6025 class_unregister_type(LUSTRE_MDT_NAME);
6027 lu_kmem_fini(mdt_caches);
6030 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
6031 MODULE_DESCRIPTION("Lustre Metadata Target ("LUSTRE_MDT_NAME")");
6032 MODULE_LICENSE("GPL");
6034 cfs_module(mdt, LUSTRE_VERSION_STRING, mdt_mod_init, mdt_mod_exit);