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.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
33 #define DEBUG_SUBSYSTEM S_MDC
35 #include <linux/module.h>
38 #include <obd_class.h>
39 #include <lustre_dlm.h>
40 #include <lustre_fid.h>
41 #include <lustre_intent.h>
42 #include <lustre_mdc.h>
43 #include <lustre_net.h>
44 #include <lustre_req_layout.h>
45 #include <lustre_swab.h>
46 #include <lustre_acl.h>
48 #include "mdc_internal.h"
50 struct mdc_getattr_args {
51 struct obd_export *ga_exp;
52 struct md_enqueue_info *ga_minfo;
55 int it_open_error(int phase, struct lookup_intent *it)
57 if (it_disposition(it, DISP_OPEN_LEASE)) {
58 if (phase >= DISP_OPEN_LEASE)
63 if (it_disposition(it, DISP_OPEN_OPEN)) {
64 if (phase >= DISP_OPEN_OPEN)
70 if (it_disposition(it, DISP_OPEN_CREATE)) {
71 if (phase >= DISP_OPEN_CREATE)
77 if (it_disposition(it, DISP_LOOKUP_EXECD)) {
78 if (phase >= DISP_LOOKUP_EXECD)
84 if (it_disposition(it, DISP_IT_EXECD)) {
85 if (phase >= DISP_IT_EXECD)
91 CERROR("it disp: %X, status: %d\n", it->it_disposition, it->it_status);
96 EXPORT_SYMBOL(it_open_error);
98 /* this must be called on a lockh that is known to have a referenced lock */
99 int mdc_set_lock_data(struct obd_export *exp, const struct lustre_handle *lockh,
100 void *data, __u64 *bits)
102 struct ldlm_lock *lock;
103 struct inode *new_inode = data;
109 if (!lustre_handle_is_used(lockh))
112 lock = ldlm_handle2lock(lockh);
114 LASSERT(lock != NULL);
115 lock_res_and_lock(lock);
116 if (lock->l_resource->lr_lvb_inode &&
117 lock->l_resource->lr_lvb_inode != data) {
118 struct inode *old_inode = lock->l_resource->lr_lvb_inode;
120 LASSERTF(old_inode->i_state & I_FREEING,
121 "Found existing inode %p/%lu/%u state %lu in lock: setting data to %p/%lu/%u\n",
122 old_inode, old_inode->i_ino, old_inode->i_generation,
124 new_inode, new_inode->i_ino, new_inode->i_generation);
126 lock->l_resource->lr_lvb_inode = new_inode;
128 *bits = lock->l_policy_data.l_inodebits.bits;
130 unlock_res_and_lock(lock);
136 enum ldlm_mode mdc_lock_match(struct obd_export *exp, __u64 flags,
137 const struct lu_fid *fid, enum ldlm_type type,
138 union ldlm_policy_data *policy,
139 enum ldlm_mode mode, struct lustre_handle *lockh)
141 struct ldlm_res_id res_id;
145 fid_build_reg_res_name(fid, &res_id);
146 /* LU-4405: Clear bits not supported by server */
147 policy->l_inodebits.bits &= exp_connect_ibits(exp);
148 rc = ldlm_lock_match(class_exp2obd(exp)->obd_namespace, flags,
149 &res_id, type, policy, mode, lockh, 0);
153 int mdc_cancel_unused(struct obd_export *exp, const struct lu_fid *fid,
154 union ldlm_policy_data *policy, enum ldlm_mode mode,
155 enum ldlm_cancel_flags flags, void *opaque)
157 struct obd_device *obd = class_exp2obd(exp);
158 struct ldlm_res_id res_id;
162 fid_build_reg_res_name(fid, &res_id);
163 rc = ldlm_cli_cancel_unused_resource(obd->obd_namespace, &res_id,
164 policy, mode, flags, opaque);
168 int mdc_null_inode(struct obd_export *exp,
169 const struct lu_fid *fid)
171 struct ldlm_res_id res_id;
172 struct ldlm_resource *res;
173 struct ldlm_namespace *ns = class_exp2obd(exp)->obd_namespace;
176 LASSERTF(ns != NULL, "no namespace passed\n");
178 fid_build_reg_res_name(fid, &res_id);
180 res = ldlm_resource_get(ns, NULL, &res_id, 0, 0);
185 res->lr_lvb_inode = NULL;
188 ldlm_resource_putref(res);
192 static inline void mdc_clear_replay_flag(struct ptlrpc_request *req, int rc)
194 /* Don't hold error requests for replay. */
195 if (req->rq_replay) {
196 spin_lock(&req->rq_lock);
198 spin_unlock(&req->rq_lock);
200 if (rc && req->rq_transno != 0) {
201 DEBUG_REQ(D_ERROR, req, "transno returned on error: rc = %d",
207 /* Save a large LOV EA into the request buffer so that it is available
208 * for replay. We don't do this in the initial request because the
209 * original request doesn't need this buffer (at most it sends just the
210 * lov_mds_md) and it is a waste of RAM/bandwidth to send the empty
211 * buffer and may also be difficult to allocate and save a very large
212 * request buffer for each open. (b=5707)
214 * OOM here may cause recovery failure if lmm is needed (only for the
215 * original open if the MDS crashed just when this client also OOM'd)
216 * but this is incredibly unlikely, and questionable whether the client
217 * could do MDS recovery under OOM anyways...
219 int mdc_save_lovea(struct ptlrpc_request *req,
220 const struct req_msg_field *field, void *data, u32 size)
222 struct req_capsule *pill = &req->rq_pill;
223 struct lov_user_md *lmm;
226 if (req_capsule_get_size(pill, field, RCL_CLIENT) < size) {
227 rc = sptlrpc_cli_enlarge_reqbuf(req, field, size);
229 CERROR("%s: Can't enlarge ea size to %d: rc = %d\n",
230 req->rq_export->exp_obd->obd_name,
235 req_capsule_shrink(pill, field, size, RCL_CLIENT);
238 req_capsule_set_size(pill, field, RCL_CLIENT, size);
239 lmm = req_capsule_client_get(pill, field);
241 memcpy(lmm, data, size);
242 /* overwrite layout generation returned from the MDS */
243 lmm->lmm_stripe_offset =
244 (typeof(lmm->lmm_stripe_offset))LOV_OFFSET_DEFAULT;
250 static struct ptlrpc_request *
251 mdc_intent_open_pack(struct obd_export *exp, struct lookup_intent *it,
252 struct md_op_data *op_data, __u32 acl_bufsize)
254 struct ptlrpc_request *req;
255 struct obd_device *obd = class_exp2obd(exp);
256 struct ldlm_intent *lit;
257 const void *lmm = op_data->op_data;
258 __u32 lmmsize = op_data->op_data_size;
259 __u32 mdt_md_capsule_size;
263 int repsize, repsize_estimate;
268 mdt_md_capsule_size = obd->u.cli.cl_default_mds_easize;
270 it->it_create_mode = (it->it_create_mode & ~S_IFMT) | S_IFREG;
272 /* XXX: openlock is not cancelled for cross-refs. */
273 /* If inode is known, cancel conflicting OPEN locks. */
274 if (fid_is_sane(&op_data->op_fid2)) {
275 if (it->it_flags & MDS_OPEN_LEASE) { /* try to get lease */
276 if (it->it_flags & MDS_FMODE_WRITE)
281 if (it->it_flags & (MDS_FMODE_WRITE | MDS_OPEN_TRUNC))
284 else if (it->it_flags & FMODE_EXEC)
290 count = mdc_resource_get_unused(exp, &op_data->op_fid2,
295 /* If CREATE, cancel parent's UPDATE lock. */
296 if (it->it_op & IT_CREAT)
300 count += mdc_resource_get_unused(exp, &op_data->op_fid1,
302 MDS_INODELOCK_UPDATE);
304 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
305 &RQF_LDLM_INTENT_OPEN);
307 ldlm_lock_list_put(&cancels, l_bl_ast, count);
308 RETURN(ERR_PTR(-ENOMEM));
311 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
312 op_data->op_namelen + 1);
313 if (cl_is_lov_delay_create(it->it_flags)) {
314 /* open(O_LOV_DELAY_CREATE) won't pack lmm */
315 LASSERT(lmmsize == 0);
316 req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT, 0);
318 req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT,
319 max(lmmsize, obd->u.cli.cl_default_mds_easize));
322 req_capsule_set_size(&req->rq_pill, &RMF_FILE_SECCTX_NAME,
323 RCL_CLIENT, op_data->op_file_secctx_name != NULL ?
324 op_data->op_file_secctx_name_size : 0);
326 req_capsule_set_size(&req->rq_pill, &RMF_FILE_SECCTX, RCL_CLIENT,
327 op_data->op_file_secctx_size);
329 req_capsule_set_size(&req->rq_pill, &RMF_FILE_ENCCTX, RCL_CLIENT,
330 op_data->op_file_encctx_size);
332 /* get SELinux policy info if any */
333 rc = sptlrpc_get_sepol(req);
335 ptlrpc_request_free(req);
338 req_capsule_set_size(&req->rq_pill, &RMF_SELINUX_POL, RCL_CLIENT,
339 strlen(req->rq_sepol) ?
340 strlen(req->rq_sepol) + 1 : 0);
342 rc = ldlm_prep_enqueue_req(exp, req, &cancels, count);
344 ptlrpc_request_free(req);
348 spin_lock(&req->rq_lock);
349 req->rq_replay = req->rq_import->imp_replayable;
350 spin_unlock(&req->rq_lock);
352 /* pack the intent */
353 lit = req_capsule_client_get(&req->rq_pill, &RMF_LDLM_INTENT);
354 lit->opc = (__u64)it->it_op;
356 /* pack the intended request */
357 mdc_open_pack(req, op_data, it->it_create_mode, 0, it->it_flags, lmm,
360 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
361 mdt_md_capsule_size);
362 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER, acl_bufsize);
364 if (!(it->it_op & IT_CREAT) && it->it_op & IT_OPEN &&
365 req_capsule_has_field(&req->rq_pill, &RMF_FILE_SECCTX_NAME,
367 op_data->op_file_secctx_name_size > 0 &&
368 op_data->op_file_secctx_name != NULL) {
371 secctx_name = req_capsule_client_get(&req->rq_pill,
372 &RMF_FILE_SECCTX_NAME);
373 memcpy(secctx_name, op_data->op_file_secctx_name,
374 op_data->op_file_secctx_name_size);
375 req_capsule_set_size(&req->rq_pill, &RMF_FILE_SECCTX,
377 obd->u.cli.cl_max_mds_easize);
379 CDEBUG(D_SEC, "packed '%.*s' as security xattr name\n",
380 op_data->op_file_secctx_name_size,
381 op_data->op_file_secctx_name);
384 req_capsule_set_size(&req->rq_pill, &RMF_FILE_SECCTX,
388 if (exp_connect_encrypt(exp) && !(it->it_op & IT_CREAT) &&
390 req_capsule_set_size(&req->rq_pill, &RMF_FILE_ENCCTX,
392 obd->u.cli.cl_max_mds_easize);
394 req_capsule_set_size(&req->rq_pill, &RMF_FILE_ENCCTX,
398 * Inline buffer for possible data from Data-on-MDT files.
400 req_capsule_set_size(&req->rq_pill, &RMF_NIOBUF_INLINE, RCL_SERVER,
401 sizeof(struct niobuf_remote));
402 ptlrpc_request_set_replen(req);
404 /* Get real repbuf allocated size as rounded up power of 2 */
405 repsize = size_roundup_power2(req->rq_replen +
406 lustre_msg_early_size());
407 /* Estimate free space for DoM files in repbuf */
408 repsize_estimate = repsize - (req->rq_replen -
409 mdt_md_capsule_size +
410 sizeof(struct lov_comp_md_v1) +
411 sizeof(struct lov_comp_md_entry_v1) +
412 lov_mds_md_size(0, LOV_MAGIC_V3));
414 if (repsize_estimate < obd->u.cli.cl_dom_min_inline_repsize) {
415 repsize = obd->u.cli.cl_dom_min_inline_repsize -
416 repsize_estimate + sizeof(struct niobuf_remote);
417 req_capsule_set_size(&req->rq_pill, &RMF_NIOBUF_INLINE,
419 sizeof(struct niobuf_remote) + repsize);
420 ptlrpc_request_set_replen(req);
421 CDEBUG(D_INFO, "Increase repbuf by %d bytes, total: %d\n",
422 repsize, req->rq_replen);
423 repsize = size_roundup_power2(req->rq_replen +
424 lustre_msg_early_size());
426 /* The only way to report real allocated repbuf size to the server
427 * is the lm_repsize but it must be set prior buffer allocation itself
428 * due to security reasons - it is part of buffer used in signature
429 * calculation (see LU-11414). Therefore the saved size is predicted
430 * value as rq_replen rounded to the next higher power of 2.
431 * Such estimation is safe. Though the final allocated buffer might
432 * be even larger, it is not possible to know that at this point.
434 req->rq_reqmsg->lm_repsize = repsize;
438 #define GA_DEFAULT_EA_NAME_LEN 20
439 #define GA_DEFAULT_EA_VAL_LEN 250
440 #define GA_DEFAULT_EA_NUM 10
442 static struct ptlrpc_request *
443 mdc_intent_getxattr_pack(struct obd_export *exp, struct lookup_intent *it,
444 struct md_op_data *op_data)
446 struct ptlrpc_request *req;
447 struct ldlm_intent *lit;
450 u32 ea_vals_buf_size = GA_DEFAULT_EA_VAL_LEN * GA_DEFAULT_EA_NUM;
453 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
454 &RQF_LDLM_INTENT_GETXATTR);
456 RETURN(ERR_PTR(-ENOMEM));
458 /* get SELinux policy info if any */
459 rc = sptlrpc_get_sepol(req);
461 ptlrpc_request_free(req);
464 req_capsule_set_size(&req->rq_pill, &RMF_SELINUX_POL, RCL_CLIENT,
465 strlen(req->rq_sepol) ?
466 strlen(req->rq_sepol) + 1 : 0);
468 rc = ldlm_prep_enqueue_req(exp, req, &cancels, count);
470 ptlrpc_request_free(req);
474 /* pack the intent */
475 lit = req_capsule_client_get(&req->rq_pill, &RMF_LDLM_INTENT);
476 lit->opc = IT_GETXATTR;
477 /* Message below is checked in sanity-selinux test_20d
478 * and sanity-sec test_49
480 CDEBUG(D_INFO, "%s: get xattrs for "DFID"\n",
481 exp->exp_obd->obd_name, PFID(&op_data->op_fid1));
483 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 53, 0)
484 /* If the supplied buffer is too small then the server will return
485 * -ERANGE and llite will fallback to using non cached xattr
486 * operations. On servers before 2.10.1 a (non-cached) listxattr RPC
487 * for an orphan or dead file causes an oops. So let's try to avoid
488 * sending too small a buffer to too old a server. This is effectively
489 * undoing the memory conservation of LU-9417 when it would be *more*
490 * likely to crash the server. See LU-9856.
492 if (exp->exp_connect_data.ocd_version < OBD_OCD_VERSION(2, 10, 1, 0))
493 ea_vals_buf_size = max_t(u32, ea_vals_buf_size,
494 exp->exp_connect_data.ocd_max_easize);
497 /* pack the intended request */
498 mdc_pack_body(req, &op_data->op_fid1, op_data->op_valid,
499 ea_vals_buf_size, -1, 0);
501 /* get SELinux policy info if any */
502 mdc_file_sepol_pack(req);
504 req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_SERVER,
505 GA_DEFAULT_EA_NAME_LEN * GA_DEFAULT_EA_NUM);
507 req_capsule_set_size(&req->rq_pill, &RMF_EAVALS, RCL_SERVER,
510 req_capsule_set_size(&req->rq_pill, &RMF_EAVALS_LENS, RCL_SERVER,
511 sizeof(u32) * GA_DEFAULT_EA_NUM);
513 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER, 0);
515 ptlrpc_request_set_replen(req);
520 static struct ptlrpc_request *
521 mdc_intent_getattr_pack(struct obd_export *exp, struct lookup_intent *it,
522 struct md_op_data *op_data, __u32 acl_bufsize)
524 struct ptlrpc_request *req;
525 struct obd_device *obd = class_exp2obd(exp);
526 u64 valid = OBD_MD_FLGETATTR | OBD_MD_FLEASIZE | OBD_MD_FLMODEASIZE |
527 OBD_MD_FLDIREA | OBD_MD_MEA | OBD_MD_FLACL |
529 struct ldlm_intent *lit;
531 bool have_secctx = false;
535 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
536 &RQF_LDLM_INTENT_GETATTR);
538 RETURN(ERR_PTR(-ENOMEM));
540 /* send name of security xattr to get upon intent */
541 if (it->it_op & (IT_LOOKUP | IT_GETATTR) &&
542 req_capsule_has_field(&req->rq_pill, &RMF_FILE_SECCTX_NAME,
544 op_data->op_file_secctx_name_size > 0 &&
545 op_data->op_file_secctx_name != NULL) {
547 req_capsule_set_size(&req->rq_pill, &RMF_FILE_SECCTX_NAME,
549 op_data->op_file_secctx_name_size);
552 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
553 op_data->op_namelen + 1);
555 rc = ldlm_prep_enqueue_req(exp, req, NULL, 0);
557 ptlrpc_request_free(req);
561 /* pack the intent */
562 lit = req_capsule_client_get(&req->rq_pill, &RMF_LDLM_INTENT);
563 lit->opc = (__u64)it->it_op;
565 easize = obd->u.cli.cl_default_mds_easize;
567 /* pack the intended request */
568 mdc_getattr_pack(req, valid, it->it_flags, op_data, easize);
570 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER, easize);
571 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER, acl_bufsize);
572 req_capsule_set_size(&req->rq_pill, &RMF_DEFAULT_MDT_MD, RCL_SERVER,
573 sizeof(struct lmv_user_md));
578 secctx_name = req_capsule_client_get(&req->rq_pill,
579 &RMF_FILE_SECCTX_NAME);
580 memcpy(secctx_name, op_data->op_file_secctx_name,
581 op_data->op_file_secctx_name_size);
583 req_capsule_set_size(&req->rq_pill, &RMF_FILE_SECCTX,
586 CDEBUG(D_SEC, "packed '%.*s' as security xattr name\n",
587 op_data->op_file_secctx_name_size,
588 op_data->op_file_secctx_name);
590 req_capsule_set_size(&req->rq_pill, &RMF_FILE_SECCTX,
594 if (exp_connect_encrypt(exp) && it->it_op & (IT_LOOKUP | IT_GETATTR))
595 req_capsule_set_size(&req->rq_pill, &RMF_FILE_ENCCTX,
598 req_capsule_set_size(&req->rq_pill, &RMF_FILE_ENCCTX,
601 ptlrpc_request_set_replen(req);
605 static struct ptlrpc_request *mdc_intent_layout_pack(struct obd_export *exp,
606 struct lookup_intent *it,
607 struct md_op_data *op_data)
609 struct obd_device *obd = class_exp2obd(exp);
610 struct ptlrpc_request *req;
611 struct ldlm_intent *lit;
612 struct layout_intent *layout;
617 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
618 &RQF_LDLM_INTENT_LAYOUT);
620 RETURN(ERR_PTR(-ENOMEM));
622 if (fid_is_sane(&op_data->op_fid2) && (it->it_op & IT_LAYOUT) &&
623 (it->it_flags & FMODE_WRITE)) {
624 count = mdc_resource_get_unused(exp, &op_data->op_fid2,
626 MDS_INODELOCK_LAYOUT);
629 req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT, 0);
630 rc = ldlm_prep_enqueue_req(exp, req, &cancels, count);
632 ptlrpc_request_free(req);
636 /* pack the intent */
637 lit = req_capsule_client_get(&req->rq_pill, &RMF_LDLM_INTENT);
638 lit->opc = (__u64)it->it_op;
640 /* pack the layout intent request */
641 layout = req_capsule_client_get(&req->rq_pill, &RMF_LAYOUT_INTENT);
642 LASSERT(op_data->op_data != NULL);
643 LASSERT(op_data->op_data_size == sizeof(*layout));
644 memcpy(layout, op_data->op_data, sizeof(*layout));
646 req_capsule_set_size(&req->rq_pill, &RMF_DLM_LVB, RCL_SERVER,
647 obd->u.cli.cl_default_mds_easize);
648 ptlrpc_request_set_replen(req);
652 static struct ptlrpc_request *mdc_enqueue_pack(struct obd_export *exp,
655 struct ptlrpc_request *req;
659 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LDLM_ENQUEUE);
661 RETURN(ERR_PTR(-ENOMEM));
663 rc = ldlm_prep_enqueue_req(exp, req, NULL, 0);
665 ptlrpc_request_free(req);
669 req_capsule_set_size(&req->rq_pill, &RMF_DLM_LVB, RCL_SERVER, lvb_len);
670 ptlrpc_request_set_replen(req);
674 static int mdc_finish_enqueue(struct obd_export *exp,
675 struct ptlrpc_request *req,
676 struct ldlm_enqueue_info *einfo,
677 struct lookup_intent *it,
678 struct lustre_handle *lockh, int rc)
680 struct req_capsule *pill = &req->rq_pill;
681 struct ldlm_request *lockreq;
682 struct ldlm_reply *lockrep;
683 struct ldlm_lock *lock;
684 struct mdt_body *body = NULL;
685 void *lvb_data = NULL;
690 /* Similarly, if we're going to replay this request, we don't want to
691 * actually get a lock, just perform the intent.
693 if (req->rq_transno || req->rq_replay) {
694 lockreq = req_capsule_client_get(pill, &RMF_DLM_REQ);
695 lockreq->lock_flags |= ldlm_flags_to_wire(LDLM_FL_INTENT_ONLY);
698 if (rc == ELDLM_LOCK_ABORTED) {
700 memset(lockh, 0, sizeof(*lockh));
702 } else { /* rc = 0 */
703 lock = ldlm_handle2lock(lockh);
704 LASSERT(lock != NULL);
706 /* If server returned a different lock mode, fix up variables */
707 if (lock->l_req_mode != einfo->ei_mode) {
708 ldlm_lock_addref(lockh, lock->l_req_mode);
709 ldlm_lock_decref(lockh, einfo->ei_mode);
710 einfo->ei_mode = lock->l_req_mode;
715 lockrep = req_capsule_server_get(pill, &RMF_DLM_REP);
716 LASSERT(lockrep != NULL); /* checked by ldlm_cli_enqueue() */
718 it->it_disposition = (int)lockrep->lock_policy_res1;
719 it->it_status = (int)lockrep->lock_policy_res2;
720 it->it_lock_mode = einfo->ei_mode;
721 it->it_lock_handle = lockh->cookie;
722 it->it_request = req;
724 /* Technically speaking rq_transno must already be zero if
725 * it_status is in error, so the check is a bit redundant.
727 if ((!req->rq_transno || it->it_status < 0) && req->rq_replay)
728 mdc_clear_replay_flag(req, it->it_status);
730 /* If we're doing an IT_OPEN which did not result in an actual
731 * successful open, then we need to remove the bit which saves
732 * this request for unconditional replay.
734 * It's important that we do this first! Otherwise we might exit the
735 * function without doing so, and try to replay a failed create.
738 if (it->it_op & IT_OPEN && req->rq_replay &&
739 (!it_disposition(it, DISP_OPEN_OPEN) || it->it_status != 0))
740 mdc_clear_replay_flag(req, it->it_status);
742 DEBUG_REQ(D_RPCTRACE, req, "op=%x disposition=%x, status=%d",
743 it->it_op, it->it_disposition, it->it_status);
745 /* We know what to expect, so we do any byte flipping required here */
746 if (it_has_reply_body(it)) {
747 body = req_capsule_server_get(pill, &RMF_MDT_BODY);
750 CERROR("%s: cannot swab mdt_body: rc = %d\n",
751 exp->exp_obd->obd_name, rc);
755 if (it_disposition(it, DISP_OPEN_OPEN) &&
756 !it_open_error(DISP_OPEN_OPEN, it)) {
758 * If this is a successful OPEN request, we need to set
759 * replay handler and data early, so that if replay
760 * happens immediately after swabbing below, new reply
761 * is swabbed by that handler correctly.
763 mdc_set_open_replay_data(NULL, NULL, it);
766 if (it_disposition(it, DISP_OPEN_CREATE) &&
767 !it_open_error(DISP_OPEN_CREATE, it)) {
768 lprocfs_counter_incr(exp->exp_obd->obd_md_stats,
772 if (body->mbo_valid & (OBD_MD_FLDIREA | OBD_MD_FLEASIZE)) {
775 mdc_update_max_ea_from_body(exp, body);
778 * The eadata is opaque; just check that it is there.
779 * Eventually, obd_unpackmd() will check the contents.
781 eadata = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
782 body->mbo_eadatasize);
786 /* save LVB data and length if for layout lock */
788 lvb_len = body->mbo_eadatasize;
791 * We save the reply LOV EA in case we have to replay a
792 * create for recovery. If we didn't allocate a large
793 * enough request buffer above we need to reallocate it
794 * here to hold the actual LOV EA.
796 * To not save LOV EA if request is not going to replay
797 * (for example error one).
799 if ((it->it_op & IT_OPEN) && req->rq_replay) {
800 rc = mdc_save_lovea(req, &RMF_EADATA, eadata,
801 body->mbo_eadatasize);
803 body->mbo_valid &= ~OBD_MD_FLEASIZE;
804 body->mbo_eadatasize = 0;
809 } else if (it->it_op & IT_LAYOUT) {
810 /* maybe the lock was granted right away and layout
811 * is packed into RMF_DLM_LVB of req
813 lvb_len = req_capsule_get_size(pill, &RMF_DLM_LVB, RCL_SERVER);
814 CDEBUG(D_INFO, "%s: layout return lvb %d transno %lld\n",
815 class_exp2obd(exp)->obd_name, lvb_len, req->rq_transno);
817 lvb_data = req_capsule_server_sized_get(pill,
818 &RMF_DLM_LVB, lvb_len);
819 if (lvb_data == NULL)
823 * save replied layout data to the request buffer for
824 * recovery consideration (lest MDS reinitialize
825 * another set of OST objects).
828 (void)mdc_save_lovea(req, &RMF_EADATA, lvb_data,
833 /* fill in stripe data for layout lock.
834 * LU-6581: trust layout data only if layout lock is granted. The MDT
835 * has stopped sending layout unless the layout lock is granted. The
836 * client still does this checking in case it's talking with an old
839 lock = ldlm_handle2lock(lockh);
843 if (ldlm_has_layout(lock) && lvb_data != NULL &&
844 !(lockrep->lock_flags & LDLM_FL_BLOCKED_MASK)) {
847 LDLM_DEBUG(lock, "layout lock returned by: %s, lvb_len: %d",
848 ldlm_it2str(it->it_op), lvb_len);
850 OBD_ALLOC_LARGE(lmm, lvb_len);
852 GOTO(out_lock, rc = -ENOMEM);
854 memcpy(lmm, lvb_data, lvb_len);
856 /* install lvb_data */
857 lock_res_and_lock(lock);
858 if (lock->l_lvb_data == NULL) {
859 lock->l_lvb_type = LVB_T_LAYOUT;
860 lock->l_lvb_data = lmm;
861 lock->l_lvb_len = lvb_len;
864 unlock_res_and_lock(lock);
866 OBD_FREE_LARGE(lmm, lvb_len);
869 if (ldlm_has_dom(lock)) {
870 LASSERT(lock->l_glimpse_ast == mdc_ldlm_glimpse_ast);
872 body = req_capsule_server_get(pill, &RMF_MDT_BODY);
873 if (!(body->mbo_valid & OBD_MD_DOM_SIZE)) {
874 LDLM_ERROR(lock, "%s: DoM lock without size.",
875 exp->exp_obd->obd_name);
876 GOTO(out_lock, rc = -EPROTO);
879 LDLM_DEBUG(lock, "DoM lock is returned by: %s, size: %llu",
880 ldlm_it2str(it->it_op), body->mbo_dom_size);
882 rc = mdc_fill_lvb(req, &lock->l_ost_lvb);
890 static inline bool mdc_skip_mod_rpc_slot(const struct lookup_intent *it)
893 (it->it_op == IT_GETATTR || it->it_op == IT_LOOKUP ||
894 it->it_op == IT_READDIR ||
895 (it->it_op == IT_LAYOUT && !(it->it_flags & MDS_FMODE_WRITE))))
900 /* We always reserve enough space in the reply packet for a stripe MD, because
901 * we don't know in advance the file type.
903 static int mdc_enqueue_base(struct obd_export *exp,
904 struct ldlm_enqueue_info *einfo,
905 const union ldlm_policy_data *policy,
906 struct lookup_intent *it,
907 struct md_op_data *op_data,
908 struct lustre_handle *lockh,
909 __u64 extra_lock_flags)
911 struct obd_device *obd = class_exp2obd(exp);
912 struct ptlrpc_request *req;
913 __u64 flags, saved_flags = extra_lock_flags;
914 struct ldlm_res_id res_id;
915 static const union ldlm_policy_data lookup_policy = {
916 .l_inodebits = { MDS_INODELOCK_LOOKUP } };
917 static const union ldlm_policy_data update_policy = {
918 .l_inodebits = { MDS_INODELOCK_UPDATE } };
919 static const union ldlm_policy_data layout_policy = {
920 .l_inodebits = { MDS_INODELOCK_LAYOUT } };
921 static const union ldlm_policy_data getxattr_policy = {
922 .l_inodebits = { MDS_INODELOCK_XATTR } };
923 int generation, resends = 0;
924 struct ldlm_reply *lockrep;
925 struct obd_import *imp = class_exp2cliimp(exp);
927 enum lvb_type lvb_type = 0;
931 LASSERTF(!it || einfo->ei_type == LDLM_IBITS, "lock type %d\n",
933 fid_build_reg_res_name(&op_data->op_fid1, &res_id);
936 LASSERT(policy == NULL);
938 saved_flags |= LDLM_FL_HAS_INTENT;
939 if (it->it_op & (IT_GETATTR | IT_READDIR))
940 policy = &update_policy;
941 else if (it->it_op & IT_LAYOUT)
942 policy = &layout_policy;
943 else if (it->it_op & IT_GETXATTR)
944 policy = &getxattr_policy;
946 policy = &lookup_policy;
949 generation = obd->u.cli.cl_import->imp_generation;
950 if (!it || (it->it_op & (IT_OPEN | IT_CREAT)))
951 acl_bufsize = min_t(__u32, imp->imp_connect_data.ocd_max_easize,
954 acl_bufsize = LUSTRE_POSIX_ACL_MAX_SIZE_OLD;
959 /* The only way right now is FLOCK. */
960 LASSERTF(einfo->ei_type == LDLM_FLOCK, "lock type %d\n",
962 res_id.name[3] = LDLM_FLOCK;
963 req = ldlm_enqueue_pack(exp, 0);
964 } else if (it->it_op & IT_OPEN) {
965 req = mdc_intent_open_pack(exp, it, op_data, acl_bufsize);
966 } else if (it->it_op & (IT_GETATTR | IT_LOOKUP)) {
967 req = mdc_intent_getattr_pack(exp, it, op_data, acl_bufsize);
968 } else if (it->it_op & IT_READDIR) {
969 req = mdc_enqueue_pack(exp, 0);
970 } else if (it->it_op & IT_LAYOUT) {
971 if (!imp_connect_lvb_type(imp))
973 req = mdc_intent_layout_pack(exp, it, op_data);
974 lvb_type = LVB_T_LAYOUT;
975 } else if (it->it_op & IT_GETXATTR) {
976 req = mdc_intent_getxattr_pack(exp, it, op_data);
983 RETURN(PTR_ERR(req));
986 req->rq_generation_set = 1;
987 req->rq_import_generation = generation;
988 req->rq_sent = ktime_get_real_seconds() + resends;
991 einfo->ei_enq_slot = !mdc_skip_mod_rpc_slot(it);
993 /* With Data-on-MDT the glimpse callback is needed too.
994 * It is set here in advance but not in mdc_finish_enqueue()
995 * to avoid possible races. It is safe to have glimpse handler
996 * for non-DOM locks and costs nothing.
998 if (einfo->ei_cb_gl == NULL)
999 einfo->ei_cb_gl = mdc_ldlm_glimpse_ast;
1001 rc = ldlm_cli_enqueue(exp, &req, einfo, &res_id, policy, &flags, NULL,
1002 0, lvb_type, lockh, 0);
1005 /* For flock requests we immediatelly return without further
1006 * delay and let caller deal with the rest, since rest of
1007 * this function metadata processing makes no sense for flock
1008 * requests anyway. But in case of problem during comms with
1009 * server (-ETIMEDOUT) or any signal/kill attempt (-EINTR),
1010 * we cannot rely on caller and this mainly for F_UNLCKs
1011 * (explicits or automatically generated by kernel to clean
1012 * current flocks upon exit) that can't be trashed.
1014 ptlrpc_req_finished(req);
1015 if (((rc == -EINTR) || (rc == -ETIMEDOUT)) &&
1016 (einfo->ei_type == LDLM_FLOCK) &&
1017 (einfo->ei_mode == LCK_NL))
1024 "%s: ldlm_cli_enqueue "DFID":"DFID"=%s failed: rc = %d\n",
1025 obd->obd_name, PFID(&op_data->op_fid1),
1026 PFID(&op_data->op_fid2), op_data->op_name ?: "", rc);
1028 mdc_clear_replay_flag(req, rc);
1029 ptlrpc_req_finished(req);
1033 lockrep = req_capsule_server_get(&req->rq_pill, &RMF_DLM_REP);
1034 LASSERT(lockrep != NULL);
1036 lockrep->lock_policy_res2 =
1037 ptlrpc_status_ntoh(lockrep->lock_policy_res2);
1039 /* Retry infinitely when the server returns -EINPROGRESS for the
1040 * intent operation, when server returns -EINPROGRESS for acquiring
1041 * intent lock, we'll retry in after_reply().
1043 if (it && (int)lockrep->lock_policy_res2 == -EINPROGRESS) {
1044 mdc_clear_replay_flag(req, rc);
1045 ptlrpc_req_finished(req);
1046 if (generation == obd->u.cli.cl_import->imp_generation) {
1047 if (signal_pending(current))
1051 CDEBUG(D_HA, "%s: resend:%d op:%d "DFID"/"DFID"\n",
1052 obd->obd_name, resends, it->it_op,
1053 PFID(&op_data->op_fid1),
1054 PFID(&op_data->op_fid2));
1057 CDEBUG(D_HA, "resend cross eviction\n");
1062 if ((int)lockrep->lock_policy_res2 == -ERANGE &&
1063 it->it_op & (IT_OPEN | IT_GETATTR | IT_LOOKUP) &&
1064 acl_bufsize == LUSTRE_POSIX_ACL_MAX_SIZE_OLD) {
1065 mdc_clear_replay_flag(req, -ERANGE);
1066 ptlrpc_req_finished(req);
1067 acl_bufsize = min_t(__u32, imp->imp_connect_data.ocd_max_easize,
1072 rc = mdc_finish_enqueue(exp, req, einfo, it, lockh, rc);
1074 if (lustre_handle_is_used(lockh)) {
1075 ldlm_lock_decref(lockh, einfo->ei_mode);
1076 memset(lockh, 0, sizeof(*lockh));
1078 ptlrpc_req_finished(req);
1080 it->it_lock_handle = 0;
1081 it->it_lock_mode = 0;
1082 it->it_request = NULL;
1088 int mdc_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1089 const union ldlm_policy_data *policy,
1090 struct md_op_data *op_data,
1091 struct lustre_handle *lockh, __u64 extra_lock_flags)
1093 return mdc_enqueue_base(exp, einfo, policy, NULL,
1094 op_data, lockh, extra_lock_flags);
1097 static int mdc_finish_intent_lock(struct obd_export *exp,
1098 struct ptlrpc_request *request,
1099 struct md_op_data *op_data,
1100 struct lookup_intent *it,
1101 struct lustre_handle *lockh)
1103 struct lustre_handle old_lock;
1104 struct ldlm_lock *lock;
1108 LASSERT(request != NULL);
1109 LASSERT(request != LP_POISON);
1110 LASSERT(request->rq_repmsg != LP_POISON);
1112 if (it->it_op & IT_READDIR)
1115 if (it->it_op & (IT_GETXATTR | IT_LAYOUT)) {
1116 if (it->it_status != 0)
1117 GOTO(out, rc = it->it_status);
1119 if (!it_disposition(it, DISP_IT_EXECD)) {
1120 /* The server failed before it even started executing
1121 * the intent, i.e. because it couldn't unpack the
1124 LASSERT(it->it_status != 0);
1125 GOTO(out, rc = it->it_status);
1127 rc = it_open_error(DISP_IT_EXECD, it);
1131 rc = it_open_error(DISP_LOOKUP_EXECD, it);
1135 /* keep requests around for the multiple phases of the call
1136 * this shows the DISP_XX must guarantee we make it into the
1139 if (!it_disposition(it, DISP_ENQ_CREATE_REF) &&
1140 it_disposition(it, DISP_OPEN_CREATE) &&
1141 !it_open_error(DISP_OPEN_CREATE, it)) {
1142 it_set_disposition(it, DISP_ENQ_CREATE_REF);
1143 /* balanced in ll_create_node */
1144 ptlrpc_request_addref(request);
1146 if (!it_disposition(it, DISP_ENQ_OPEN_REF) &&
1147 it_disposition(it, DISP_OPEN_OPEN) &&
1148 !it_open_error(DISP_OPEN_OPEN, it)) {
1149 it_set_disposition(it, DISP_ENQ_OPEN_REF);
1150 /* balanced in ll_file_open */
1151 ptlrpc_request_addref(request);
1152 /* eviction in middle of open RPC processing b=11546 */
1153 OBD_FAIL_TIMEOUT(OBD_FAIL_MDC_ENQUEUE_PAUSE,
1157 if (it->it_op & IT_CREAT) {
1158 /* XXX this belongs in ll_create_it */
1159 } else if (it->it_op == IT_OPEN) {
1160 LASSERT(!it_disposition(it, DISP_OPEN_CREATE));
1162 LASSERT(it->it_op & (IT_GETATTR | IT_LOOKUP));
1166 /* If we already have a matching lock, then cancel the new
1167 * one. We have to set the data here instead of in
1168 * mdc_enqueue, because we need to use the child's inode as
1169 * the l_ast_data to match, and that's not available until
1170 * intent_finish has performed the iget().
1172 lock = ldlm_handle2lock(lockh);
1174 union ldlm_policy_data policy = lock->l_policy_data;
1176 LDLM_DEBUG(lock, "matching against this");
1178 if (it_has_reply_body(it)) {
1179 struct mdt_body *body;
1181 body = req_capsule_server_get(&request->rq_pill,
1183 /* mdc_enqueue checked */
1184 LASSERT(body != NULL);
1185 LASSERTF(fid_res_name_eq(&body->mbo_fid1,
1186 &lock->l_resource->lr_name),
1187 "Lock res_id: "DLDLMRES", fid: "DFID"\n",
1188 PLDLMRES(lock->l_resource),
1189 PFID(&body->mbo_fid1));
1191 LDLM_LOCK_PUT(lock);
1193 memcpy(&old_lock, lockh, sizeof(*lockh));
1194 if (ldlm_lock_match(NULL, LDLM_FL_BLOCK_GRANTED, NULL,
1195 LDLM_IBITS, &policy, LCK_NL, &old_lock, 0)) {
1196 ldlm_lock_decref_and_cancel(lockh, it->it_lock_mode);
1197 memcpy(lockh, &old_lock, sizeof(old_lock));
1198 it->it_lock_handle = lockh->cookie;
1205 "D_IT dentry=%.*s intent=%s status=%d disp=%x: rc = %d\n",
1206 (int)op_data->op_namelen, op_data->op_name,
1207 ldlm_it2str(it->it_op), it->it_status, it->it_disposition, rc);
1212 int mdc_revalidate_lock(struct obd_export *exp, struct lookup_intent *it,
1213 struct lu_fid *fid, __u64 *bits)
1215 /* We could just return 1 immediately, but as we should only be called
1216 * in revalidate_it if we already have a lock, let's verify that.
1218 struct ldlm_res_id res_id;
1219 struct lustre_handle lockh;
1220 union ldlm_policy_data policy;
1221 enum ldlm_mode mode;
1224 if (it->it_lock_handle) {
1225 lockh.cookie = it->it_lock_handle;
1226 mode = ldlm_revalidate_lock_handle(&lockh, bits);
1228 fid_build_reg_res_name(fid, &res_id);
1229 switch (it->it_op) {
1231 /* File attributes are held under multiple bits:
1232 * nlink is under lookup lock, size and times are
1233 * under UPDATE lock and recently we've also got
1234 * a separate permissions lock for owner/group/acl that
1235 * were protected by lookup lock before.
1236 * Getattr must provide all of that information,
1237 * so we need to ensure we have all of those locks.
1238 * Unfortunately, if the bits are split across multiple
1239 * locks, there's no easy way to match all of them here,
1240 * so an extra RPC would be performed to fetch all
1241 * of those bits at once for now.
1243 /* For new MDTs(> 2.4), UPDATE|PERM should be enough,
1244 * but for old MDTs (< 2.4), permission is covered
1245 * by LOOKUP lock, so it needs to match all bits here.
1247 policy.l_inodebits.bits = MDS_INODELOCK_UPDATE |
1248 MDS_INODELOCK_LOOKUP |
1252 policy.l_inodebits.bits = MDS_INODELOCK_UPDATE;
1255 policy.l_inodebits.bits = MDS_INODELOCK_LAYOUT;
1258 policy.l_inodebits.bits = MDS_INODELOCK_LOOKUP;
1262 mode = mdc_lock_match(exp, LDLM_FL_BLOCK_GRANTED, fid,
1263 LDLM_IBITS, &policy,
1264 LCK_CR | LCK_CW | LCK_PR | LCK_PW,
1269 it->it_lock_handle = lockh.cookie;
1270 it->it_lock_mode = mode;
1272 it->it_lock_handle = 0;
1273 it->it_lock_mode = 0;
1280 * This long block is all about fixing up the lock and request state
1281 * so that it is correct as of the moment _before_ the operation was
1282 * applied; that way, the VFS will think that everything is normal and
1283 * call Lustre's regular VFS methods.
1285 * If we're performing a creation, that means that unless the creation
1286 * failed with EEXIST, we should fake up a negative dentry.
1288 * For everything else, we want to lookup to succeed.
1290 * One additional note: if CREATE or OPEN succeeded, we add an extra
1291 * reference to the request because we need to keep it around until
1292 * ll_create/ll_open gets called.
1294 * The server will return to us, in it_disposition, an indication of
1295 * exactly what it_status refers to.
1297 * If DISP_OPEN_OPEN is set, then it_status refers to the open() call,
1298 * otherwise if DISP_OPEN_CREATE is set, then it status is the
1299 * creation failure mode. In either case, one of DISP_LOOKUP_NEG or
1300 * DISP_LOOKUP_POS will be set, indicating whether the child lookup
1303 * Else, if DISP_LOOKUP_EXECD then it_status is the rc of the
1306 int mdc_intent_lock(struct obd_export *exp, struct md_op_data *op_data,
1307 struct lookup_intent *it, struct ptlrpc_request **reqp,
1308 ldlm_blocking_callback cb_blocking, __u64 extra_lock_flags)
1310 struct ldlm_enqueue_info einfo = {
1311 .ei_type = LDLM_IBITS,
1312 .ei_mode = it_to_lock_mode(it),
1313 .ei_cb_bl = cb_blocking,
1314 .ei_cb_cp = ldlm_completion_ast,
1315 .ei_cb_gl = mdc_ldlm_glimpse_ast,
1317 struct lustre_handle lockh;
1322 CDEBUG(D_DLMTRACE, "(name: %.*s,"DFID") in obj "DFID
1323 ", intent: %s flags %#llo\n", (int)op_data->op_namelen,
1324 op_data->op_name, PFID(&op_data->op_fid2),
1325 PFID(&op_data->op_fid1), ldlm_it2str(it->it_op),
1329 if (fid_is_sane(&op_data->op_fid2) &&
1330 (it->it_op & (IT_LOOKUP | IT_GETATTR | IT_READDIR))) {
1331 /* We could just return 1 immediately, but since we should only
1332 * be called in revalidate_it if we already have a lock, let's
1335 it->it_lock_handle = 0;
1336 rc = mdc_revalidate_lock(exp, it, &op_data->op_fid2, NULL);
1337 /* Only return failure if it was not GETATTR by cfid
1338 * (from inode_revalidate()).
1340 if (rc || op_data->op_namelen != 0)
1344 /* For case if upper layer did not alloc fid, do it now. */
1345 if (!fid_is_sane(&op_data->op_fid2) && it->it_op & IT_CREAT) {
1346 rc = mdc_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
1348 CERROR("%s: cannot allocate new FID: rc=%d\n",
1349 exp->exp_obd->obd_name, rc);
1354 rc = mdc_enqueue_base(exp, &einfo, NULL, it, op_data, &lockh,
1359 *reqp = it->it_request;
1360 rc = mdc_finish_intent_lock(exp, *reqp, op_data, it, &lockh);
1364 static int mdc_intent_getattr_async_interpret(const struct lu_env *env,
1365 struct ptlrpc_request *req,
1368 struct mdc_getattr_args *ga = args;
1369 struct obd_export *exp = ga->ga_exp;
1370 struct md_enqueue_info *minfo = ga->ga_minfo;
1371 struct ldlm_enqueue_info *einfo = &minfo->mi_einfo;
1372 struct lookup_intent *it = &minfo->mi_it;
1373 struct lustre_handle *lockh = &minfo->mi_lockh;
1374 struct ldlm_reply *lockrep;
1375 __u64 flags = LDLM_FL_HAS_INTENT;
1378 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_GETATTR_ENQUEUE))
1381 rc = ldlm_cli_enqueue_fini(exp, req, einfo->ei_type, 1, einfo->ei_mode,
1382 &flags, NULL, 0, lockh, rc);
1384 CERROR("%s: ldlm_cli_enqueue_fini() failed: rc = %d\n",
1385 exp->exp_obd->obd_name, rc);
1386 mdc_clear_replay_flag(req, rc);
1390 lockrep = req_capsule_server_get(&req->rq_pill, &RMF_DLM_REP);
1391 LASSERT(lockrep != NULL);
1393 lockrep->lock_policy_res2 =
1394 ptlrpc_status_ntoh(lockrep->lock_policy_res2);
1396 rc = mdc_finish_enqueue(exp, req, einfo, it, lockh, rc);
1400 rc = mdc_finish_intent_lock(exp, req, &minfo->mi_data, it, lockh);
1404 minfo->mi_cb(req, minfo, rc);
1408 int mdc_intent_getattr_async(struct obd_export *exp,
1409 struct md_enqueue_info *minfo)
1411 struct md_op_data *op_data = &minfo->mi_data;
1412 struct lookup_intent *it = &minfo->mi_it;
1413 struct ptlrpc_request *req;
1414 struct mdc_getattr_args *ga;
1415 struct ldlm_res_id res_id;
1416 union ldlm_policy_data policy = {
1417 .l_inodebits = { MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE }
1419 __u64 flags = LDLM_FL_HAS_INTENT;
1424 "name: %.*s in inode "DFID", intent: %s flags %#llo\n",
1425 (int)op_data->op_namelen, op_data->op_name,
1426 PFID(&op_data->op_fid1), ldlm_it2str(it->it_op), it->it_flags);
1428 fid_build_reg_res_name(&op_data->op_fid1, &res_id);
1429 /* If the MDT return -ERANGE because of large ACL, then the sponsor
1430 * of the async getattr RPC will handle that by itself.
1432 req = mdc_intent_getattr_pack(exp, it, op_data,
1433 LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
1435 RETURN(PTR_ERR(req));
1437 /* With Data-on-MDT the glimpse callback is needed too.
1438 * It is set here in advance but not in mdc_finish_enqueue()
1439 * to avoid possible races. It is safe to have glimpse handler
1440 * for non-DOM locks and costs nothing.
1442 if (minfo->mi_einfo.ei_cb_gl == NULL)
1443 minfo->mi_einfo.ei_cb_gl = mdc_ldlm_glimpse_ast;
1445 rc = ldlm_cli_enqueue(exp, &req, &minfo->mi_einfo, &res_id, &policy,
1446 &flags, NULL, 0, LVB_T_NONE, &minfo->mi_lockh, 1);
1448 ptlrpc_req_finished(req);
1452 ga = ptlrpc_req_async_args(ga, req);
1454 ga->ga_minfo = minfo;
1456 req->rq_interpret_reply = mdc_intent_getattr_async_interpret;
1457 ptlrpcd_add_req(req);