1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Lustre Light Super operations
6 * Copyright (c) 2002, 2003 Cluster File Systems, Inc.
8 * This file is part of Lustre, http://www.lustre.org.
10 * Lustre is free software; you can redistribute it and/or
11 * modify it under the terms of version 2 of the GNU General Public
12 * License as published by the Free Software Foundation.
14 * Lustre is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
17 * GNU General Public License for more details.
19 * You should have received a copy of the GNU General Public License
20 * along with Lustre; if not, write to the Free Software
21 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
24 #define DEBUG_SUBSYSTEM S_LLITE
26 #include <linux/module.h>
27 #include <linux/types.h>
28 #include <linux/random.h>
29 #include <linux/version.h>
31 #include <linux/lustre_lite.h>
32 #include <linux/lustre_ha.h>
33 #include <linux/lustre_dlm.h>
34 #include <linux/lprocfs_status.h>
35 #include "llite_internal.h"
37 kmem_cache_t *ll_file_data_slab;
39 extern struct address_space_operations ll_aops;
40 extern struct address_space_operations ll_dir_aops;
43 #define log2(n) ffz(~(n))
46 struct ll_sb_info *lustre_init_sbi(struct super_block *sb)
48 struct ll_sb_info *sbi = NULL;
52 OBD_ALLOC(sbi, sizeof(*sbi));
56 spin_lock_init(&sbi->ll_lock);
57 INIT_LIST_HEAD(&sbi->ll_pglist);
58 sbi->ll_pglist_gen = 0;
59 if (num_physpages < SBI_DEFAULT_RA_MAX / 4)
60 sbi->ll_ra_info.ra_max_pages = num_physpages / 4;
62 sbi->ll_ra_info.ra_max_pages = SBI_DEFAULT_RA_MAX;
63 INIT_LIST_HEAD(&sbi->ll_conn_chain);
64 INIT_HLIST_HEAD(&sbi->ll_orphan_dentry_list);
65 INIT_LIST_HEAD(&sbi->ll_mnt_list);
66 sema_init(&sbi->ll_gns_sem, 1);
67 init_completion(&sbi->ll_gns_completion);
68 sbi->ll_gns_state = LL_GNS_STATE_IDLE;
69 sbi->ll_gns_timer.data = (unsigned long)sbi;
70 sbi->ll_gns_timer.function = ll_gns_timer_callback;
71 init_timer(&sbi->ll_gns_timer);
72 INIT_LIST_HEAD(&sbi->ll_gns_sbi_head);
76 generate_random_uuid(uuid);
77 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
81 void lustre_free_sbi(struct super_block *sb)
83 struct ll_sb_info *sbi = ll_s2sbi(sb);
87 list_del(&sbi->ll_gns_sbi_head);
88 del_timer(&sbi->ll_gns_timer);
89 OBD_FREE(sbi, sizeof(*sbi));
95 int lustre_init_ea_size(struct ll_sb_info *sbi)
100 /* OSC may have larger ea size */
101 valsize = sizeof(desc);
102 rc = obd_get_info(sbi->ll_osc_exp, strlen("lovdesc") + 1, "lovdesc",
106 obd_init_ea_size(sbi->ll_mdc_exp, obd_size_diskmd(sbi->ll_osc_exp, NULL),
107 desc.ld_tgt_count*sizeof(struct llog_cookie));
112 int lustre_common_fill_super(struct super_block *sb, char *mdc, char *osc)
114 struct inode *root = 0;
115 struct ll_sb_info *sbi = ll_s2sbi(sb);
116 struct obd_device *obd;
117 struct ll_fid rootfid;
118 struct obd_statfs osfs;
119 struct ptlrpc_request *request = NULL;
120 struct lustre_handle osc_conn = {0, };
121 struct lustre_handle mdc_conn = {0, };
126 obd = class_name2obd(mdc);
128 CERROR("MDC %s: not setup or attached\n", mdc);
132 if (proc_lustre_fs_root) {
133 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
136 CERROR("could not register mount in /proc/lustre");
139 err = obd_connect(&mdc_conn, obd, &sbi->ll_sb_uuid, 0);
141 CERROR("An MDS (mdc %s) is performing recovery, of which this"
142 " client is not a part. Please wait for recovery to "
143 "complete, abort, or time out.\n", mdc);
146 CERROR("cannot connect to %s: rc = %d\n", mdc, err);
149 sbi->ll_mdc_exp = class_conn2export(&mdc_conn);
151 err = obd_statfs(obd, &osfs, jiffies - HZ);
155 LASSERT(osfs.os_bsize);
156 sb->s_blocksize = osfs.os_bsize;
157 sb->s_blocksize_bits = log2(osfs.os_bsize);
158 sb->s_magic = LL_SUPER_MAGIC;
159 sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
161 devno = get_uuid2int(sbi->ll_mdc_exp->exp_obd->obd_uuid.uuid,
162 strlen(sbi->ll_mdc_exp->exp_obd->obd_uuid.uuid));
166 obd = class_name2obd(osc);
168 CERROR("OSC %s: not setup or attached\n", osc);
172 err = obd_connect(&osc_conn, obd, &sbi->ll_sb_uuid, 0);
174 CERROR("An OST (osc %s) is performing recovery, of which this"
175 " client is not a part. Please wait for recovery to "
176 "complete, abort, or time out.\n", osc);
179 CERROR("cannot connect to %s: rc = %d\n", osc, err);
182 sbi->ll_osc_exp = class_conn2export(&osc_conn);
184 lustre_init_ea_size(sbi);
185 err = md_getstatus(sbi->ll_mdc_exp, &rootfid);
187 CERROR("cannot mds_connect: rc = %d\n", err);
190 CDEBUG(D_SUPER, "rootfid "LPU64"\n", rootfid.id);
191 sbi->ll_rootino = rootfid.id;
193 sb->s_op = &lustre_super_operations;
196 * XXX: move this to after cbd setup? */
197 err = md_getattr(sbi->ll_mdc_exp, &rootfid,
198 OBD_MD_FLNOTOBD | OBD_MD_FLBLOCKS, 0, &request);
200 CERROR("md_getattr failed for root: rc = %d\n", err);
204 err = mdc_req2lustre_md(sbi->ll_mdc_exp, request, 0,
205 sbi->ll_osc_exp, &md);
207 CERROR("failed to understand root inode md: rc = %d\n",err);
208 ptlrpc_req_finished (request);
212 LASSERT(sbi->ll_rootino != 0);
213 root = ll_iget(sb, sbi->ll_rootino, &md);
215 ptlrpc_req_finished(request);
217 if (root == NULL || is_bad_inode(root)) {
218 /* XXX might need iput() for bad inode */
219 CERROR("lustre_lite: bad iget4 for root\n");
220 GOTO(out_root, err = -EBADF);
223 err = ll_close_thread_start(&sbi->ll_lcq);
225 CERROR("cannot start close thread: rc %d\n", err);
229 ll_gns_add_timer(sbi);
231 /* making vm readahead 0 for 2.4.x. In the case of 2.6.x,
232 backing dev info assigned to inode mapping is used for
233 determining maximal readahead. */
234 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0))
235 /* bug 2805 - set VM readahead to zero */
236 vm_max_readahead = vm_min_readahead = 0;
239 sb->s_root = d_alloc_root(root);
242 CWARN("Enabling PDIROPS\n");
243 sb->s_flags |= S_PDIROPS;
252 obd_disconnect(sbi->ll_osc_exp, 0);
254 obd_disconnect(sbi->ll_mdc_exp, 0);
256 lprocfs_unregister_mountpoint(sbi);
260 void lustre_common_put_super(struct super_block *sb)
262 struct ll_sb_info *sbi = ll_s2sbi(sb);
263 struct hlist_node *tmp, *next;
266 ll_gns_del_timer(sbi);
268 ll_close_thread_shutdown(sbi->ll_lcq);
270 list_del(&sbi->ll_conn_chain);
271 obd_disconnect(sbi->ll_osc_exp, 0);
273 lprocfs_unregister_mountpoint(sbi);
274 if (sbi->ll_proc_root) {
275 lprocfs_remove(sbi->ll_proc_root);
276 sbi->ll_proc_root = NULL;
279 obd_disconnect(sbi->ll_mdc_exp, 0);
281 // We do this to get rid of orphaned dentries. That is not really trw.
282 spin_lock(&dcache_lock);
283 hlist_for_each_safe(tmp, next, &sbi->ll_orphan_dentry_list) {
284 struct dentry *dentry = hlist_entry(tmp, struct dentry, d_hash);
285 CWARN("orphan dentry %*s (%p) at unmount\n",
286 dentry->d_name.len, dentry->d_name.name, dentry);
287 shrink_dcache_parent(dentry);
289 spin_unlock(&dcache_lock);
294 char *ll_read_opt(const char *opt, char *data)
300 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
301 if (strncmp(opt, data, strlen(opt)))
303 if ((value = strchr(data, '=')) == NULL)
307 OBD_ALLOC(retval, strlen(value) + 1);
309 CERROR("out of memory!\n");
313 memcpy(retval, value, strlen(value)+1);
314 CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
318 int ll_set_opt(const char *opt, char *data, int fl)
322 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
323 if (strncmp(opt, data, strlen(opt)))
329 void ll_options(char *options, char **ost, char **mdc, int *flags)
332 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
333 char *opt_ptr = options;
342 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
343 for (this_char = strtok (options, ",");
345 this_char = strtok (NULL, ",")) {
347 while ((this_char = strsep (&opt_ptr, ",")) != NULL) {
349 CDEBUG(D_SUPER, "this_char %s\n", this_char);
350 if (!*ost && (*ost = ll_read_opt("osc", this_char)))
352 if (!*mdc && (*mdc = ll_read_opt("mdc", this_char)))
354 if (!(*flags & LL_SBI_NOLCK) &&
355 ((*flags) = (*flags) |
356 ll_set_opt("nolock", this_char,
364 void ll_lli_init(struct ll_inode_info *lli)
366 sema_init(&lli->lli_open_sem, 1);
368 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
369 spin_lock_init(&lli->lli_lock);
370 INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
371 lli->lli_inode_magic = LLI_INODE_MAGIC;
374 int ll_fill_super(struct super_block *sb, void *data, int silent)
376 struct ll_sb_info *sbi;
382 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
384 sbi = lustre_init_sbi(sb);
388 sbi->ll_flags |= LL_SBI_READAHEAD;
389 ll_options(data, &osc, &mdc, &sbi->ll_flags);
393 GOTO(out, err = -EINVAL);
398 GOTO(out, err = -EINVAL);
401 err = lustre_common_fill_super(sb, mdc, osc);
407 OBD_FREE(mdc, strlen(mdc) + 1);
409 OBD_FREE(osc, strlen(osc) + 1);
412 } /* ll_read_super */
414 static int lustre_process_log(struct lustre_mount_data *lmd, char *profile,
415 struct config_llog_instance *cfg, int allow_recov)
417 struct lustre_cfg lcfg;
418 struct portals_cfg pcfg;
419 char * peer = "MDS_PEER_UUID";
420 struct obd_device *obd;
421 struct lustre_handle mdc_conn = {0, };
422 struct obd_export *exp;
423 char * name = "mdc_dev";
425 struct obd_uuid mdc_uuid;
426 struct llog_ctxt *ctxt;
431 if (lmd_bad_magic(lmd))
434 generate_random_uuid(uuid);
435 class_uuid_unparse(uuid, &mdc_uuid);
437 if (lmd->lmd_local_nid) {
438 PCFG_INIT(pcfg, NAL_CMD_REGISTER_MYNID);
439 pcfg.pcfg_nal = lmd->lmd_nal;
440 pcfg.pcfg_nid = lmd->lmd_local_nid;
441 err = libcfs_nal_cmd(&pcfg);
446 if (lmd->lmd_nal == SOCKNAL ||
447 lmd->lmd_nal == OPENIBNAL) {
448 PCFG_INIT(pcfg, NAL_CMD_ADD_PEER);
449 pcfg.pcfg_nal = lmd->lmd_nal;
450 pcfg.pcfg_nid = lmd->lmd_server_nid;
451 pcfg.pcfg_id = lmd->lmd_server_ipaddr;
452 pcfg.pcfg_misc = lmd->lmd_port;
453 pcfg.pcfg_size = 8388608;
454 pcfg.pcfg_flags = 0x4; /*share*/
455 err = libcfs_nal_cmd(&pcfg);
460 LCFG_INIT(lcfg, LCFG_ADD_UUID, name);
461 lcfg.lcfg_nid = lmd->lmd_server_nid;
462 lcfg.lcfg_inllen1 = strlen(peer) + 1;
463 lcfg.lcfg_inlbuf1 = peer;
464 lcfg.lcfg_nal = lmd->lmd_nal;
465 err = class_process_config(&lcfg);
467 GOTO(out_del_conn, err);
469 LCFG_INIT(lcfg, LCFG_ATTACH, name);
470 lcfg.lcfg_inlbuf1 = "mdc";
471 lcfg.lcfg_inllen1 = strlen(lcfg.lcfg_inlbuf1) + 1;
472 lcfg.lcfg_inlbuf2 = mdc_uuid.uuid;
473 lcfg.lcfg_inllen2 = strlen(lcfg.lcfg_inlbuf2) + 1;
474 err = class_process_config(&lcfg);
476 GOTO(out_del_uuid, err);
478 LCFG_INIT(lcfg, LCFG_SETUP, name);
479 lcfg.lcfg_inlbuf1 = lmd->lmd_mds;
480 lcfg.lcfg_inllen1 = strlen(lcfg.lcfg_inlbuf1) + 1;
481 lcfg.lcfg_inlbuf2 = peer;
482 lcfg.lcfg_inllen2 = strlen(lcfg.lcfg_inlbuf2) + 1;
483 err = class_process_config(&lcfg);
485 GOTO(out_detach, err);
487 obd = class_name2obd(name);
489 GOTO(out_cleanup, err = -EINVAL);
491 /* Disable initial recovery on this import */
492 err = obd_set_info(obd->obd_self_export,
493 strlen("initial_recov"), "initial_recov",
494 sizeof(allow_recov), &allow_recov);
496 GOTO(out_cleanup, err);
498 err = obd_connect(&mdc_conn, obd, &mdc_uuid, 0);
500 CERROR("cannot connect to %s: rc = %d\n", lmd->lmd_mds, err);
501 GOTO(out_cleanup, err);
504 exp = class_conn2export(&mdc_conn);
506 ctxt = llog_get_context(&exp->exp_obd->obd_llogs,LLOG_CONFIG_REPL_CTXT);
507 rc = class_config_process_llog(ctxt, profile, cfg);
509 CERROR("class_config_process_llog failed: rc = %d\n", rc);
511 err = obd_disconnect(exp, 0);
514 LCFG_INIT(lcfg, LCFG_CLEANUP, name);
515 err = class_process_config(&lcfg);
520 LCFG_INIT(lcfg, LCFG_DETACH, name);
521 err = class_process_config(&lcfg);
526 LCFG_INIT(lcfg, LCFG_DEL_UUID, name);
527 lcfg.lcfg_inllen1 = strlen(peer) + 1;
528 lcfg.lcfg_inlbuf1 = peer;
529 err = class_process_config(&lcfg);
532 if (lmd->lmd_nal == SOCKNAL ||
533 lmd->lmd_nal == OPENIBNAL) {
534 PCFG_INIT(pcfg, NAL_CMD_DEL_PEER);
535 pcfg.pcfg_nal = lmd->lmd_nal;
536 pcfg.pcfg_nid = lmd->lmd_server_nid;
537 pcfg.pcfg_flags = 1; /* single_share */
538 err = libcfs_nal_cmd(&pcfg);
549 int lustre_fill_super(struct super_block *sb, void *data, int silent)
551 struct lustre_mount_data * lmd = data;
552 struct ll_sb_info *sbi;
558 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
559 if (lmd_bad_magic(lmd))
562 sbi = lustre_init_sbi(sb);
566 sbi->ll_flags |= LL_SBI_READAHEAD;
568 if (lmd->lmd_profile) {
569 struct lustre_profile *lprof;
570 struct config_llog_instance cfg;
573 if (lmd->lmd_mds[0] == '\0') {
574 CERROR("no mds name\n");
575 GOTO(out_free, err = -EINVAL);
578 OBD_ALLOC(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
579 if (sbi->ll_lmd == NULL)
580 GOTO(out_free, err = -ENOMEM);
581 memcpy(sbi->ll_lmd, lmd, sizeof(*lmd));
583 /* generate a string unique to this super, let's try
584 the address of the super itself.*/
585 len = (sizeof(sb) * 2) + 1;
586 OBD_ALLOC(sbi->ll_instance, len);
587 if (sbi->ll_instance == NULL)
588 GOTO(out_free, err = -ENOMEM);
589 sprintf(sbi->ll_instance, "%p", sb);
591 cfg.cfg_instance = sbi->ll_instance;
592 cfg.cfg_uuid = sbi->ll_sb_uuid;
593 cfg.cfg_local_nid = lmd->lmd_local_nid;
594 err = lustre_process_log(lmd, lmd->lmd_profile, &cfg, 0);
596 CERROR("Unable to process log: %s\n", lmd->lmd_profile);
600 lprof = class_get_profile(lmd->lmd_profile);
602 CERROR("No profile found: %s\n", lmd->lmd_profile);
603 GOTO(out_free, err = -EINVAL);
606 OBD_FREE(osc, strlen(osc) + 1);
607 OBD_ALLOC(osc, strlen(lprof->lp_osc) +
608 strlen(sbi->ll_instance) + 2);
609 sprintf(osc, "%s-%s", lprof->lp_osc, sbi->ll_instance);
612 OBD_FREE(mdc, strlen(mdc) + 1);
613 OBD_ALLOC(mdc, strlen(lprof->lp_mdc) +
614 strlen(sbi->ll_instance) + 2);
615 sprintf(mdc, "%s-%s", lprof->lp_mdc, sbi->ll_instance);
620 GOTO(out_free, err = -EINVAL);
625 GOTO(out_free, err = -EINVAL);
628 err = lustre_common_fill_super(sb, mdc, osc);
635 OBD_FREE(mdc, strlen(mdc) + 1);
637 OBD_FREE(osc, strlen(osc) + 1);
643 int len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean")+1;
646 if (sbi->ll_instance != NULL) {
647 struct lustre_mount_data *lmd = sbi->ll_lmd;
649 struct config_llog_instance cfg;
651 cfg.cfg_instance = sbi->ll_instance;
652 cfg.cfg_uuid = sbi->ll_sb_uuid;
654 OBD_ALLOC(cln_prof, len);
655 sprintf(cln_prof, "%s-clean", lmd->lmd_profile);
657 err = lustre_process_log(lmd, cln_prof, &cfg, 0);
659 CERROR("Unable to process log: %s\n", cln_prof);
660 OBD_FREE(cln_prof, len);
661 OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance)+ 1);
663 OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
668 } /* lustre_fill_super */
670 static void lustre_manual_cleanup(struct ll_sb_info *sbi)
672 struct lustre_cfg lcfg;
673 struct obd_device *obd;
676 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) != NULL)
680 LCFG_INIT(lcfg, LCFG_CLEANUP, obd->obd_name);
681 err = class_process_config(&lcfg);
683 CERROR("cleanup failed: %s\n", obd->obd_name);
687 LCFG_INIT(lcfg, LCFG_DETACH, obd->obd_name);
688 err = class_process_config(&lcfg);
690 CERROR("detach failed: %s\n", obd->obd_name);
695 if (sbi->ll_lmd != NULL)
696 class_del_profile(sbi->ll_lmd->lmd_profile);
699 void lustre_put_super(struct super_block *sb)
701 struct obd_device *obd;
702 struct ll_sb_info *sbi = ll_s2sbi(sb);
703 int force_umount = 0;
706 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
707 obd = class_exp2obd(sbi->ll_mdc_exp);
709 force_umount = obd->obd_no_recov;
712 lustre_common_put_super(sb);
713 if (sbi->ll_lmd != NULL) {
715 int len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean")+1;
717 struct config_llog_instance cfg;
720 CERROR("force umount, doing manual cleanup\n");
721 lustre_manual_cleanup(sbi);
725 cfg.cfg_instance = sbi->ll_instance;
726 cfg.cfg_uuid = sbi->ll_sb_uuid;
728 OBD_ALLOC(cln_prof, len);
729 sprintf(cln_prof, "%s-clean", sbi->ll_lmd->lmd_profile);
731 err = lustre_process_log(sbi->ll_lmd, cln_prof, &cfg, 0);
733 CERROR("Unable to process log: %s, doing manual cleanup"
735 lustre_manual_cleanup(sbi);
738 OBD_FREE(cln_prof, len);
740 OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
741 OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance) + 1);
747 } /* lustre_put_super */
749 int ll_process_config_update(struct ll_sb_info *sbi, int clean)
751 struct obd_export *mdc_exp = sbi->ll_mdc_exp;
752 struct lustre_mount_data *lmd = sbi->ll_lmd;
753 struct llog_ctxt *ctxt;
754 struct config_llog_instance cfg;
755 char *profile = lmd->lmd_profile, *name = NULL;
756 int rc, namelen = 0, version;
762 CERROR("Client not mounted with zero-conf; cannot process "
767 rc = ldlm_cli_cancel_unused(mdc_exp->exp_obd->obd_namespace, NULL,
768 LDLM_FL_CONFIG_CHANGE, NULL);
770 CWARN("ldlm_cli_cancel_unused(mdc): %d\n", rc);
772 rc = obd_cancel_unused(sbi->ll_osc_exp, NULL, LDLM_FL_CONFIG_CHANGE,
775 CWARN("obd_cancel_unused(lov): %d\n", rc);
777 cfg.cfg_instance = sbi->ll_instance;
778 cfg.cfg_uuid = sbi->ll_sb_uuid;
779 cfg.cfg_local_nid = lmd->lmd_local_nid;
781 namelen = strlen(profile) + 20; /* -clean-######### */
782 OBD_ALLOC(name, namelen);
787 version = sbi->ll_config_version - 1;
788 sprintf(name, "%s-clean-%d", profile, version);
790 version = sbi->ll_config_version + 1;
791 sprintf(name, "%s-%d", profile, version);
794 CWARN("Applying configuration log %s\n", name);
796 ctxt = llog_get_context(&mdc_exp->exp_obd->obd_llogs,
797 LLOG_CONFIG_REPL_CTXT);
798 rc = class_config_process_llog(ctxt, name, &cfg);
800 sbi->ll_config_version = version;
801 CWARN("Finished applying configuration log %s: %d\n", name, rc);
803 if (rc == 0 && clean == 0) {
804 struct lov_desc desc;
806 valsize = sizeof(desc);
807 rc = obd_get_info(sbi->ll_osc_exp, strlen("lovdesc") + 1,
808 "lovdesc", &valsize, &desc);
810 rc = obd_init_ea_size(mdc_exp,
811 obd_size_diskmd(sbi->ll_osc_exp, NULL),
813 sizeof(struct llog_cookie)));
815 OBD_FREE(name, namelen);
819 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
821 struct inode *inode = NULL;
822 l_lock(&lock->l_resource->lr_namespace->ns_lock);
823 if (lock->l_ast_data) {
824 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
825 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
826 inode = igrab(lock->l_ast_data);
828 CERROR("DEBUG: l_ast_data %p is bogus: magic %x\n",
829 lock->l_ast_data, lli->lli_inode_magic);
832 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
836 int null_if_equal(struct ldlm_lock *lock, void *data)
838 if (data == lock->l_ast_data) {
839 lock->l_ast_data = NULL;
841 if (lock->l_req_mode != lock->l_granted_mode)
842 LDLM_ERROR(lock,"clearing inode with ungranted lock\n");
845 return LDLM_ITER_CONTINUE;
848 void ll_clear_inode(struct inode *inode)
851 struct ll_inode_info *lli = ll_i2info(inode);
852 struct ll_sb_info *sbi = ll_i2sbi(inode);
855 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
856 inode->i_generation, inode);
858 lli->lli_inode_magic = LLI_INODE_DEAD;
859 ll_inode2fid(&fid, inode);
860 clear_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &(ll_i2info(inode)->lli_flags));
861 md_change_cbdata(sbi->ll_mdc_exp, &fid, null_if_equal, inode);
864 obd_change_cbdata(sbi->ll_osc_exp, lli->lli_smd,
865 null_if_equal, inode);
868 obd_free_memmd(sbi->ll_osc_exp, &lli->lli_smd);
873 /* FIXME: change cbdata for mea here */
874 obd_free_memmd(sbi->ll_mdc_exp,
875 (struct lov_stripe_md **) &lli->lli_mea);
879 if (lli->lli_symlink_name) {
880 OBD_FREE(lli->lli_symlink_name,
881 strlen(lli->lli_symlink_name) + 1);
882 lli->lli_symlink_name = NULL;
888 /* If this inode has objects allocated to it (lsm != NULL), then the OST
889 * object(s) determine the file size and mtime. Otherwise, the MDS will
890 * keep these values until such a time that objects are allocated for it.
891 * We do the MDS operations first, as it is checking permissions for us.
892 * We don't to the MDS RPC if there is nothing that we want to store there,
893 * otherwise there is no harm in updating mtime/atime on the MDS if we are
894 * going to do an RPC anyways.
896 * If we are doing a truncate, we will send the mtime and ctime updates
897 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
898 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
901 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
903 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
904 struct ll_sb_info *sbi = ll_i2sbi(inode);
905 struct ptlrpc_request *request = NULL;
906 struct mdc_op_data op_data;
907 int ia_valid = attr->ia_valid;
911 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu\n", inode->i_ino);
912 lprocfs_counter_incr(ll_i2sbi(inode)->ll_stats, LPROC_LL_SETATTR);
914 if (ia_valid & ATTR_SIZE) {
915 if (attr->ia_size > ll_file_maxbytes(inode)) {
916 CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
917 attr->ia_size, ll_file_maxbytes(inode));
921 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
924 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
925 if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
926 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
930 /* We mark all of the fields "set" so MDS/OST does not re-set them */
931 if (attr->ia_valid & ATTR_CTIME) {
932 attr->ia_ctime = CURRENT_TIME;
933 attr->ia_valid |= ATTR_CTIME_SET;
935 if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
936 attr->ia_atime = CURRENT_TIME;
937 attr->ia_valid |= ATTR_ATIME_SET;
939 if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
940 attr->ia_mtime = CURRENT_TIME;
941 attr->ia_valid |= ATTR_MTIME_SET;
944 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
945 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
946 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
947 LTIME_S(CURRENT_TIME));
949 attr->ia_valid &= ~ATTR_SIZE;
951 /* If only OST attributes being set on objects, don't do MDS RPC.
952 * In that case, we need to check permissions and update the local
953 * inode ourselves so we can call obdo_from_inode() always. */
954 if (ia_valid & (lsm ? ~(ATTR_SIZE | ATTR_FROM_OPEN | ATTR_RAW) : ~0)) {
956 ll_prepare_mdc_op_data(&op_data, inode, NULL, NULL, 0, 0);
958 rc = md_setattr(sbi->ll_mdc_exp, &op_data,
959 attr, NULL, 0, NULL, 0, &request);
961 ptlrpc_req_finished(request);
962 if (rc != -EPERM && rc != -EACCES)
963 CERROR("md_setattr fails: rc = %d\n", rc);
967 rc = mdc_req2lustre_md(sbi->ll_mdc_exp, request, 0,
968 sbi->ll_osc_exp, &md);
970 ptlrpc_req_finished(request);
974 /* Won't invoke vmtruncate as we already cleared ATTR_SIZE,
975 * but needed to set timestamps backwards on utime. */
976 inode_setattr(inode, attr);
977 ll_update_inode(inode, &md);
978 ptlrpc_req_finished(request);
980 if (!lsm || !S_ISREG(inode->i_mode)) {
981 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
985 /* The OST doesn't check permissions, but the alternative is
986 * a gratuitous RPC to the MDS. We already rely on the client
987 * to do read/write/truncate permission checks, so is mtime OK?
989 if (ia_valid & (ATTR_MTIME | ATTR_ATIME)) {
990 /* from sys_utime() */
991 if (!(ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET))) {
992 if (current->fsuid != inode->i_uid &&
993 (rc=ll_permission(inode,MAY_WRITE,NULL))!=0)
996 /* from inode_change_ok() */
997 if (current->fsuid != inode->i_uid &&
998 !capable(CAP_FOWNER))
1003 /* Won't invoke vmtruncate, as we already cleared ATTR_SIZE */
1004 inode_setattr(inode, attr);
1007 /* We really need to get our PW lock before we change inode->i_size.
1008 * If we don't we can race with other i_size updaters on our node, like
1009 * ll_file_read. We can also race with i_size propogation to other
1010 * nodes through dirtying and writeback of final cached pages. This
1011 * last one is especially bad for racing o_append users on other
1013 if (ia_valid & ATTR_SIZE) {
1014 ldlm_policy_data_t policy = { .l_extent = {attr->ia_size,
1016 struct lustre_handle lockh = { 0 };
1017 int err, ast_flags = 0;
1018 /* XXX when we fix the AST intents to pass the discard-range
1019 * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
1021 if (attr->ia_size == 0)
1022 ast_flags = LDLM_AST_DISCARD_DATA;
1024 /* bug 1639: avoid write/truncate i_sem/DLM deadlock */
1025 LASSERT(atomic_read(&inode->i_sem.count) <= 0);
1027 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy, &lockh,
1028 ast_flags, &ll_i2sbi(inode)->ll_seek_stime);
1029 down(&inode->i_sem);
1033 rc = vmtruncate(inode, attr->ia_size);
1035 /* We need to drop the semaphore here, because this unlock may
1036 * result in a cancellation, which will need the i_sem */
1038 /* unlock now as we don't mind others file lockers racing with
1039 * the mds updates below? */
1040 err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
1041 down(&inode->i_sem);
1043 CERROR("ll_extent_unlock failed: %d\n", err);
1047 } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
1050 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1051 inode->i_ino, LTIME_S(attr->ia_mtime));
1052 oa.o_id = lsm->lsm_object_id;
1053 oa.o_gr = lsm->lsm_object_gr;
1054 oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1055 obdo_from_inode(&oa, inode, OBD_MD_FLTYPE | OBD_MD_FLATIME |
1056 OBD_MD_FLMTIME | OBD_MD_FLCTIME);
1057 rc = obd_setattr(sbi->ll_osc_exp, &oa, lsm, NULL);
1059 CERROR("obd_setattr fails: rc=%d\n", rc);
1064 int ll_setattr(struct dentry *de, struct iattr *attr)
1066 LBUG(); /* code is unused, but leave this in case of VFS changes */
1070 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1071 unsigned long max_age)
1073 struct ll_sb_info *sbi = ll_s2sbi(sb);
1074 struct obd_statfs obd_osfs;
1078 rc = obd_statfs(class_exp2obd(sbi->ll_mdc_exp), osfs, max_age);
1080 CERROR("mdc_statfs fails: rc = %d\n", rc);
1084 osfs->os_type = sb->s_magic;
1086 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1087 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1089 rc = obd_statfs(class_exp2obd(sbi->ll_osc_exp), &obd_osfs, max_age);
1091 CERROR("obd_statfs fails: rc = %d\n", rc);
1095 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1096 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1099 osfs->os_blocks = obd_osfs.os_blocks;
1100 osfs->os_bfree = obd_osfs.os_bfree;
1101 osfs->os_bavail = obd_osfs.os_bavail;
1103 /* If we don't have as many objects free on the OST as inodes
1104 * on the MDS, we reduce the total number of inodes to
1105 * compensate, so that the "inodes in use" number is correct.
1107 if (obd_osfs.os_ffree < osfs->os_ffree) {
1108 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1110 osfs->os_ffree = obd_osfs.os_ffree;
1116 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1118 struct obd_statfs osfs;
1121 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p\n", sb);
1122 lprocfs_counter_incr(ll_s2sbi(sb)->ll_stats, LPROC_LL_STAFS);
1124 /* For now we will always get up-to-date statfs values, but in the
1125 * future we may allow some amount of caching on the client (e.g.
1126 * from QOS or lprocfs updates). */
1127 rc = ll_statfs_internal(sb, &osfs, jiffies - 1);
1131 statfs_unpack(sfs, &osfs);
1133 if (sizeof(sfs->f_blocks) == 4) {
1134 while (osfs.os_blocks > ~0UL) {
1137 osfs.os_blocks >>= 1;
1138 osfs.os_bfree >>= 1;
1139 osfs.os_bavail >>= 1;
1143 sfs->f_blocks = osfs.os_blocks;
1144 sfs->f_bfree = osfs.os_bfree;
1145 sfs->f_bavail = osfs.os_bavail;
1150 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1152 struct ll_inode_info *lli = ll_i2info(inode);
1153 struct mds_body *body = md->body;
1154 struct lov_stripe_md *lsm = md->lsm;
1155 struct mea *mea = md->mea;
1158 LASSERT((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1159 LASSERT((mea != NULL) == ((body->valid & OBD_MD_FLDIREA) != 0));
1161 LASSERT(lsm->lsm_object_gr > 0);
1162 if (lli->lli_smd == NULL) {
1164 lli->lli_maxbytes = lsm->lsm_maxbytes;
1165 if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1166 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1169 if (memcmp(lli->lli_smd, lsm, sizeof(*lsm))) {
1170 CERROR("lsm mismatch for inode %ld\n",
1172 CERROR("lli_smd:\n");
1173 dump_lsm(D_ERROR, lli->lli_smd);
1175 dump_lsm(D_ERROR, lsm);
1178 /* XXX FIXME -- We should decide on a safer (atomic) and
1179 * more elegant way to update the lsm */
1180 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1181 lli->lli_smd->lsm_oinfo[i].loi_id =
1182 lsm->lsm_oinfo[i].loi_id;
1183 lli->lli_smd->lsm_oinfo[i].loi_gr =
1184 lsm->lsm_oinfo[i].loi_gr;
1185 lli->lli_smd->lsm_oinfo[i].loi_ost_idx =
1186 lsm->lsm_oinfo[i].loi_ost_idx;
1187 lli->lli_smd->lsm_oinfo[i].loi_ost_gen =
1188 lsm->lsm_oinfo[i].loi_ost_gen;
1191 /* bug 2844 - limit i_blksize for broken user-space apps */
1192 LASSERTF(lsm->lsm_xfersize != 0, "%lu\n", lsm->lsm_xfersize);
1193 inode->i_blksize = min(lsm->lsm_xfersize, LL_MAX_BLKSIZE);
1194 if (lli->lli_smd != lsm)
1195 obd_free_memmd(ll_i2obdexp(inode), &lsm);
1199 if (lli->lli_mea == NULL) {
1202 if (memcmp(lli->lli_mea, mea, body->eadatasize)) {
1203 CERROR("mea mismatch for inode %lu\n",
1208 if (lli->lli_mea != mea)
1209 obd_free_memmd(ll_i2mdcexp(inode),
1210 (struct lov_stripe_md **) &mea);
1214 if (body->valid & OBD_MD_FLID)
1215 inode->i_ino = body->ino;
1216 if (body->valid & OBD_MD_FLATIME)
1217 LTIME_S(inode->i_atime) = body->atime;
1218 if (body->valid & OBD_MD_FLMTIME &&
1219 body->mtime > LTIME_S(inode->i_mtime)) {
1220 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to %u\n",
1221 inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
1222 LTIME_S(inode->i_mtime) = body->mtime;
1224 if (body->valid & OBD_MD_FLCTIME &&
1225 body->ctime > LTIME_S(inode->i_ctime))
1226 LTIME_S(inode->i_ctime) = body->ctime;
1227 if (body->valid & OBD_MD_FLMODE)
1228 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1229 if (body->valid & OBD_MD_FLTYPE)
1230 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1231 if (body->valid & OBD_MD_FLUID)
1232 inode->i_uid = body->uid;
1233 if (body->valid & OBD_MD_FLGID)
1234 inode->i_gid = body->gid;
1235 if (body->valid & OBD_MD_FLFLAGS)
1236 inode->i_flags = body->flags;
1237 if (body->valid & OBD_MD_FLNLINK)
1238 inode->i_nlink = body->nlink;
1239 if (body->valid & OBD_MD_FLGENER)
1240 inode->i_generation = body->generation;
1241 if (body->valid & OBD_MD_FLRDEV)
1242 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1243 inode->i_rdev = body->rdev;
1245 inode->i_rdev = old_decode_dev(body->rdev);
1247 if (body->valid & OBD_MD_FLSIZE)
1248 inode->i_size = body->size;
1249 if (body->valid & OBD_MD_FLBLOCKS)
1250 inode->i_blocks = body->blocks;
1252 if (body->valid & OBD_MD_FLSIZE)
1253 set_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &lli->lli_flags);
1255 lli->lli_mds = body->mds;
1256 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1257 inode->i_dev = (kdev_t) body->mds;
1259 LASSERT(body->mds < 1000);
1262 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
1263 static struct backing_dev_info ll_backing_dev_info = {
1264 .ra_pages = 0, /* No readahead */
1265 .memory_backed = 0, /* Does contribute to dirty memory */
1269 void ll_read_inode2(struct inode *inode, void *opaque)
1271 struct lustre_md *md = opaque;
1272 struct ll_inode_info *lli = ll_i2info(inode);
1275 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1276 inode->i_generation, inode);
1280 LASSERT(!lli->lli_smd);
1282 /* Core attributes from the MDS first. This is a new inode, and
1283 * the VFS doesn't zero times in the core inode so we have to do
1284 * it ourselves. They will be overwritten by either MDS or OST
1285 * attributes - we just need to make sure they aren't newer. */
1286 LTIME_S(inode->i_mtime) = 0;
1287 LTIME_S(inode->i_atime) = 0;
1288 LTIME_S(inode->i_ctime) = 0;
1291 ll_update_inode(inode, md);
1293 /* OIDEBUG(inode); */
1295 if (S_ISREG(inode->i_mode)) {
1296 inode->i_op = &ll_file_inode_operations;
1297 inode->i_fop = &ll_file_operations;
1298 inode->i_mapping->a_ops = &ll_aops;
1300 } else if (S_ISDIR(inode->i_mode)) {
1301 inode->i_op = &ll_dir_inode_operations;
1302 inode->i_fop = &ll_dir_operations;
1303 inode->i_mapping->a_ops = &ll_dir_aops;
1305 } else if (S_ISLNK(inode->i_mode)) {
1306 inode->i_op = &ll_fast_symlink_inode_operations;
1309 inode->i_op = &ll_special_inode_operations;
1311 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
1312 init_special_inode(inode, inode->i_mode,
1313 kdev_t_to_nr(inode->i_rdev));
1315 /* initializing backing dev info. */
1316 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
1318 init_special_inode(inode, inode->i_mode, inode->i_rdev);
1320 lli->ll_save_ifop = inode->i_fop;
1322 if (S_ISCHR(inode->i_mode))
1323 inode->i_fop = &ll_special_chr_inode_fops;
1324 else if (S_ISBLK(inode->i_mode))
1325 inode->i_fop = &ll_special_blk_inode_fops;
1326 else if (S_ISFIFO(inode->i_mode))
1327 inode->i_fop = &ll_special_fifo_inode_fops;
1328 else if (S_ISSOCK(inode->i_mode))
1329 inode->i_fop = &ll_special_sock_inode_fops;
1331 CWARN("saved %p, replaced with %p\n", lli->ll_save_ifop,
1334 if (lli->ll_save_ifop->owner) {
1335 CWARN("%p has owner %p\n", lli->ll_save_ifop,
1336 lli->ll_save_ifop->owner);
1342 void ll_delete_inode(struct inode *inode)
1346 struct ll_sb_info *sbi = ll_i2sbi(inode);
1349 ll_inode2fid(&fid, inode);
1351 rc = md_delete_object(sbi->ll_mdc_exp, &fid);
1353 CERROR("md_delete_object() failed, error %d.\n",
1361 int ll_iocontrol(struct inode *inode, struct file *file,
1362 unsigned int cmd, unsigned long arg)
1364 struct ll_sb_info *sbi = ll_i2sbi(inode);
1365 struct ptlrpc_request *req = NULL;
1370 case EXT3_IOC_GETFLAGS: {
1372 unsigned long valid = OBD_MD_FLFLAGS;
1373 struct mds_body *body;
1375 ll_inode2fid(&fid, inode);
1376 rc = md_getattr(sbi->ll_mdc_exp, &fid, valid, 0, &req);
1378 CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1382 body = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*body));
1384 if (body->flags & S_APPEND)
1385 flags |= EXT3_APPEND_FL;
1386 if (body->flags & S_IMMUTABLE)
1387 flags |= EXT3_IMMUTABLE_FL;
1388 if (body->flags & S_NOATIME)
1389 flags |= EXT3_NOATIME_FL;
1391 ptlrpc_req_finished (req);
1393 RETURN(put_user(flags, (int *)arg));
1395 case EXT3_IOC_SETFLAGS: {
1396 struct mdc_op_data op_data;
1399 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1401 if (get_user(flags, (int *)arg))
1408 ll_prepare_mdc_op_data(&op_data, inode, NULL, NULL, 0, 0);
1410 memset(&attr, 0x0, sizeof(attr));
1411 attr.ia_attr_flags = flags;
1412 attr.ia_valid |= ATTR_ATTR_FLAG;
1414 rc = md_setattr(sbi->ll_mdc_exp, &op_data,
1415 &attr, NULL, 0, NULL, 0, &req);
1417 ptlrpc_req_finished(req);
1418 if (rc != -EPERM && rc != -EACCES)
1419 CERROR("md_setattr fails: rc = %d\n", rc);
1423 ptlrpc_req_finished(req);
1425 oa->o_id = lsm->lsm_object_id;
1426 oa->o_gr = lsm->lsm_object_gr;
1427 oa->o_flags = flags;
1428 oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS | OBD_MD_FLGROUP;
1430 rc = obd_setattr(sbi->ll_osc_exp, oa, lsm, NULL);
1433 if (rc != -EPERM && rc != -EACCES)
1434 CERROR("md_setattr fails: rc = %d\n", rc);
1438 if (flags & EXT3_APPEND_FL)
1439 inode->i_flags |= S_APPEND;
1441 inode->i_flags &= ~S_APPEND;
1442 if (flags & EXT3_IMMUTABLE_FL)
1443 inode->i_flags |= S_IMMUTABLE;
1445 inode->i_flags &= ~S_IMMUTABLE;
1446 if (flags & EXT3_NOATIME_FL)
1447 inode->i_flags |= S_NOATIME;
1449 inode->i_flags &= ~S_NOATIME;
1460 void ll_umount_begin(struct super_block *sb)
1462 struct ll_sb_info *sbi = ll_s2sbi(sb);
1463 struct obd_device *obd;
1464 struct obd_ioctl_data ioc_data = { 0 };
1467 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1468 sb->s_count, atomic_read(&sb->s_active));
1470 obd = class_exp2obd(sbi->ll_mdc_exp);
1472 CERROR("Invalid MDC connection handle "LPX64"\n",
1473 sbi->ll_mdc_exp->exp_handle.h_cookie);
1477 obd->obd_no_recov = 1;
1478 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_mdc_exp, sizeof ioc_data,
1481 obd = class_exp2obd(sbi->ll_osc_exp);
1483 CERROR("Invalid LOV connection handle "LPX64"\n",
1484 sbi->ll_osc_exp->exp_handle.h_cookie);
1489 obd->obd_no_recov = 1;
1490 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_osc_exp, sizeof ioc_data,
1493 /* Really, we'd like to wait until there are no requests outstanding,
1494 * and then continue. For now, we just invalidate the requests,
1495 * schedule, and hope.
1502 int ll_prep_inode(struct obd_export *osc_exp, struct obd_export *mdc_exp,
1503 struct inode **inode, struct ptlrpc_request *req,
1504 int offset, struct super_block *sb)
1506 struct lustre_md md;
1509 rc = mdc_req2lustre_md(mdc_exp, req, offset, osc_exp, &md);
1514 ll_update_inode(*inode, &md);
1517 *inode = ll_iget(sb, md.body->ino, &md);
1518 if (*inode == NULL || is_bad_inode(*inode)) {
1519 /* free the lsm if we allocated one above */
1521 obd_free_memmd(osc_exp, &md.lsm);
1523 obd_free_memmd(mdc_exp,
1524 (struct lov_stripe_md**)&md.mea);
1526 CERROR("new_inode -fatal: rc %d\n", rc);