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>
30 #include <linux/seq_file.h>
32 #include <linux/lustre_lite.h>
33 #include <linux/lustre_ha.h>
34 #include <linux/lustre_dlm.h>
35 #include <linux/lprocfs_status.h>
36 #include <linux/lustre_acl.h>
37 #include <linux/lustre_audit.h>
38 #include <linux/lustre_gs.h>
39 #include <linux/lustre_sec.h>
40 #include "llite_internal.h"
42 kmem_cache_t *ll_file_data_slab;
43 kmem_cache_t *ll_intent_slab;
45 extern struct address_space_operations ll_aops;
46 extern struct address_space_operations ll_dir_aops;
49 #define log2(n) ffz(~(n))
52 struct ll_sb_info *lustre_init_sbi(struct super_block *sb)
54 struct ll_sb_info *sbi = NULL;
58 OBD_ALLOC(sbi, sizeof(*sbi));
62 spin_lock_init(&sbi->ll_lock);
63 INIT_LIST_HEAD(&sbi->ll_pglist);
64 sbi->ll_pglist_gen = 0;
65 if (num_physpages < SBI_DEFAULT_RA_MAX / 4)
66 sbi->ll_ra_info.ra_max_pages = num_physpages / 4;
68 sbi->ll_ra_info.ra_max_pages = SBI_DEFAULT_RA_MAX;
69 INIT_LIST_HEAD(&sbi->ll_conn_chain);
70 INIT_HLIST_HEAD(&sbi->ll_orphan_dentry_list);
71 INIT_LIST_HEAD(&sbi->ll_mnt_list);
73 sema_init(&sbi->ll_gns_sem, 1);
74 spin_lock_init(&sbi->ll_gns_lock);
75 INIT_LIST_HEAD(&sbi->ll_gns_sbi_head);
76 init_waitqueue_head(&sbi->ll_gns_waitq);
77 init_completion(&sbi->ll_gns_mount_finished);
79 /* this later may be reset via /proc/fs/... */
80 memcpy(sbi->ll_gns_oname, ".mntinfo", strlen(".mntinfo"));
81 sbi->ll_gns_oname[strlen(sbi->ll_gns_oname)] = '\0';
83 /* this later may be reset via /proc/fs/... */
84 memcpy(sbi->ll_gns_upcall, "/usr/sbin/gns_upcall",
85 strlen("/usr/sbin/gns_upcall"));
86 sbi->ll_gns_upcall[strlen(sbi->ll_gns_upcall)] = '\0';
88 /* default values, may be changed via /proc/fs/... */
89 sbi->ll_gns_state = LL_GNS_IDLE;
90 sbi->ll_gns_pending_dentry = NULL;
91 atomic_set(&sbi->ll_gns_enabled, 1);
92 sbi->ll_gns_tick = GNS_TICK_TIMEOUT;
93 sbi->ll_gns_timeout = GNS_MOUNT_TIMEOUT;
95 sbi->ll_gns_timer.data = (unsigned long)sbi;
96 sbi->ll_gns_timer.function = ll_gns_timer_callback;
97 init_timer(&sbi->ll_gns_timer);
99 sbi->ll_audit_mask = AUDIT_OFF;
102 generate_random_uuid(uuid);
103 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
107 void lustre_free_sbi(struct super_block *sb)
109 struct ll_sb_info *sbi = ll_s2sbi(sb);
113 list_del(&sbi->ll_gns_sbi_head);
114 del_timer(&sbi->ll_gns_timer);
115 OBD_FREE(sbi, sizeof(*sbi));
117 ll_set_sbi(sb, NULL);
121 int lustre_init_dt_desc(struct ll_sb_info *sbi)
127 valsize = sizeof(sbi->ll_dt_desc);
128 memset(&sbi->ll_dt_desc, 0, sizeof(sbi->ll_dt_desc));
129 rc = obd_get_info(sbi->ll_dt_exp, strlen("lovdesc") + 1,
130 "lovdesc", &valsize, &sbi->ll_dt_desc);
134 static int lustre_connect_mds(struct super_block *sb, char *lmv,
135 struct obd_connect_data *data,
136 char *mds_security, int async, int pag)
138 struct ll_sb_info *sbi = ll_s2sbi(sb);
139 struct lustre_handle md_conn = {0, };
140 struct obd_device *md_obd;
141 struct obd_statfs osfs;
142 unsigned long sec_flags;
147 md_obd = class_name2obd(lmv);
149 CERROR("MDC %s: not setup or attached\n", lmv);
153 obd_set_info(md_obd->obd_self_export, strlen("async"), "async",
154 sizeof(async), &async);
156 err = obd_set_info(md_obd->obd_self_export, strlen("sec"), "sec",
157 strlen(mds_security), mds_security);
160 CERROR("LMV %s: failed to set security %s, err %d\n",
161 lmv, mds_security, err);
166 sec_flags = PTLRPC_SEC_FL_PAG;
167 err = obd_set_info(md_obd->obd_self_export,
168 strlen("sec_flags"), "sec_flags",
169 sizeof(sec_flags), &sec_flags);
171 OBD_FREE(data, sizeof(*data));
176 err = obd_connect(&md_conn, md_obd, &sbi->ll_sb_uuid, data,
177 OBD_OPT_REAL_CLIENT);
179 CERROR("An MDS (lmv %s) is performing recovery, of which this"
180 " client is not a part. Please wait for recovery to "
181 "complete, abort, or time out.\n", lmv);
184 CERROR("cannot connect to %s: rc = %d\n", lmv, err);
188 sbi->ll_md_exp = class_conn2export(&md_conn);
190 err = obd_statfs(md_obd, &osfs, jiffies - HZ);
192 GOTO(out_disconnect, err);
194 if (!osfs.os_bsize) {
195 CERROR("Invalid block size is detected.");
196 GOTO(out_disconnect, err);
199 sb->s_magic = LL_SUPER_MAGIC;
200 sb->s_blocksize = osfs.os_bsize;
201 sb->s_blocksize_bits = log2(osfs.os_bsize);
202 sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
204 /* in 2.6.x FS is not allowed to form s_dev */
205 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
209 devno = get_uuid2int((char *)sbi->ll_md_exp->exp_obd->obd_uuid.uuid,
210 strlen((char *)sbi->ll_md_exp->exp_obd->obd_uuid.uuid));
216 /* after statfs, we are supposed to have connected to MDSs,
217 * so it's ok to check remote flag returned.
219 valsize = sizeof(&sbi->ll_remote);
220 err = obd_get_info(sbi->ll_md_exp, strlen("remote_flag"), "remote_flag",
221 &valsize, &sbi->ll_remote);
223 CERROR("fail to obtain remote flag\n");
224 GOTO(out_disconnect, err);
229 obd_disconnect(sbi->ll_md_exp, 0);
233 static int lustre_connect_ost(struct super_block *sb, char *lov,
234 struct obd_connect_data *data,
235 char *oss_security, int async, int pag)
237 struct ll_sb_info *sbi = ll_s2sbi(sb);
238 struct lustre_handle dt_conn = {0, };
239 struct obd_device *obd = NULL;
240 unsigned long sec_flags;
243 obd = class_name2obd(lov);
245 CERROR("OSC %s: not setup or attached\n", lov);
246 GOTO(out, err = -EINVAL);
248 obd_set_info(obd->obd_self_export, strlen("async"), "async",
249 sizeof(async), &async);
251 if (oss_security == NULL)
252 oss_security = "null";
254 err = obd_set_info(obd->obd_self_export, strlen("sec"), "sec",
255 strlen(oss_security), oss_security);
257 CERROR("LOV %s: failed to set security %s, err %d\n",
258 lov, oss_security, err);
263 sec_flags = PTLRPC_SEC_FL_PAG;
264 err = obd_set_info(obd->obd_self_export,
265 strlen("sec_flags"), "sec_flags",
266 sizeof(sec_flags), &sec_flags);
268 OBD_FREE(data, sizeof(*data));
273 err = obd_connect(&dt_conn, obd, &sbi->ll_sb_uuid, data, 0);
275 CERROR("An OST (lov %s) is performing recovery, of which this"
276 " client is not a part. Please wait for recovery to "
277 "complete, abort, or time out.\n", lov);
280 CERROR("cannot connect to %s: rc = %d\n", lov, err);
283 sbi->ll_dt_exp = class_conn2export(&dt_conn);
285 err = lustre_init_dt_desc(sbi);
288 CWARN("init dt_desc error %d \n", err);
291 mdsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
292 obd_init_ea_size(sbi->ll_md_exp, mdsize, sbi->ll_dt_desc.ld_tgt_count *
293 sizeof(struct llog_cookie));
298 extern struct dentry_operations ll_d_ops;
300 static int lustre_init_root_inode(struct super_block *sb)
302 struct ll_sb_info *sbi = ll_s2sbi(sb);
303 struct ptlrpc_request *request = NULL;
304 struct inode *root = NULL;
309 err = md_getstatus(sbi->ll_md_exp, &sbi->ll_rootid);
311 CERROR("cannot mds_connect: rc = %d\n", err);
314 CDEBUG(D_SUPER, "rootid "DLID4"\n", OLID4(&sbi->ll_rootid));
316 sb->s_op = &lustre_super_operations;
318 /* make root inode */
319 err = md_getattr(sbi->ll_md_exp, &sbi->ll_rootid,
320 (OBD_MD_FLNOTOBD | OBD_MD_FLBLOCKS | OBD_MD_FID),
321 NULL, NULL, 0, 0, &request);
323 CERROR("md_getattr failed for root: rc = %d\n", err);
327 err = mdc_req2lustre_md(sbi->ll_md_exp, request, 0,
328 sbi->ll_dt_exp, &md);
330 CERROR("failed to understand root inode md: rc = %d\n", err);
331 ptlrpc_req_finished(request);
335 LASSERT(id_ino(&sbi->ll_rootid) != 0);
336 root = ll_iget(sb, id_ino(&sbi->ll_rootid), &md);
338 ptlrpc_req_finished(request);
340 if (root == NULL || is_bad_inode(root)) {
342 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
344 obd_free_memmd(sbi->ll_md_exp,
345 (struct lov_stripe_md**)&md.mea);
346 CERROR("lustre_lite: bad iget4 for root\n");
347 GOTO(out_root, err = -EBADF);
349 sb->s_root = d_alloc_root(root);
350 sb->s_root->d_op = &ll_d_ops;
358 int lustre_common_fill_super(struct super_block *sb, char *lmv, char *lov,
359 char *gkc, int async, char *mds_security,
360 char *oss_security, __u32 *nllu, int pag,
363 struct ll_sb_info *sbi = ll_s2sbi(sb);
364 struct obd_connect_data *data;
368 /*process the connect flags*/
369 if ((*remote & (OBD_CONNECT_LOCAL | OBD_CONNECT_REMOTE)) ==
370 (OBD_CONNECT_LOCAL | OBD_CONNECT_REMOTE)) {
371 CERROR("wrong remote flag "LPX64"\n", *remote);
375 OBD_ALLOC(data, sizeof(*data));
379 data->ocd_connect_flags |= *remote & (OBD_CONNECT_LOCAL |
381 memcpy(data->ocd_nllu, nllu, sizeof(data->ocd_nllu));
383 if (proc_lustre_fs_root) {
384 err = lprocfs_register_mountpoint(proc_lustre_fs_root,
387 CERROR("could not register mount in /proc/lustre");
391 err = lustre_connect_mds(sb, lmv, data, mds_security, async, pag);
396 err = lustre_connect_ost(sb, lov, data, oss_security, async, pag);
400 err = lustre_init_crypto(sb, gkc, data, async);
402 CERROR("Could not connect to GSS err %d\n", err);
406 err = lustre_init_root_inode(sb);
410 err = ll_close_thread_start(&sbi->ll_lcq);
412 CERROR("cannot start close thread: rc %d\n", err);
416 ll_gns_add_timer(sbi);
418 /* making vm readahead 0 for 2.4.x. In the case of 2.6.x,
419 backing dev info assigned to inode mapping is used for
420 determining maximal readahead. */
421 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0)) && \
422 !defined(KERNEL_HAS_AS_MAX_READAHEAD)
423 /* bug 2805 - set VM readahead to zero */
424 vm_max_readahead = vm_min_readahead = 0;
426 sb->s_flags |= MS_POSIXACL;
428 CWARN("Enabling PDIROPS\n");
429 sb->s_flags |= S_PDIROPS;
432 OBD_FREE(data, sizeof(*data));
438 lustre_destroy_crypto(sb);
440 obd_disconnect(sbi->ll_md_exp, 0);
443 OBD_FREE(data, sizeof(*data));
444 lprocfs_unregister_mountpoint(sbi);
448 void lustre_common_put_super(struct super_block *sb)
450 struct ll_sb_info *sbi = ll_s2sbi(sb);
451 struct hlist_node *tmp, *next;
454 ll_gns_del_timer(sbi);
455 ll_close_thread_stop(sbi->ll_lcq);
457 lustre_destroy_crypto(sb);
459 list_del(&sbi->ll_conn_chain);
460 obd_disconnect(sbi->ll_dt_exp, 0);
462 lprocfs_unregister_mountpoint(sbi);
463 if (sbi->ll_proc_root) {
464 lprocfs_remove(sbi->ll_proc_root);
465 sbi->ll_proc_root = NULL;
468 obd_disconnect(sbi->ll_md_exp, 0);
470 // We do this to get rid of orphaned dentries. That is not really trw.
471 hlist_for_each_safe(tmp, next, &sbi->ll_orphan_dentry_list) {
472 struct dentry *dentry = hlist_entry(tmp, struct dentry, d_hash);
473 CWARN("orphan dentry %.*s (%p->%p) at unmount\n",
474 dentry->d_name.len, dentry->d_name.name, dentry, next);
475 shrink_dcache_parent(dentry);
480 char *ll_read_opt(const char *opt, char *data)
486 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
487 if (strncmp(opt, data, strlen(opt)))
489 if ((value = strchr(data, '=')) == NULL)
493 OBD_ALLOC(retval, strlen(value) + 1);
495 CERROR("out of memory!\n");
499 memcpy(retval, value, strlen(value)+1);
500 CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
504 int ll_set_opt(const char *opt, char *data, int fl)
508 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
509 if (strncmp(opt, data, strlen(opt)))
515 void ll_options(char *options, char **lov, char **lmv, char **gkc,
516 char **mds_sec, char **oss_sec, int *async, int *flags)
519 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
520 char *opt_ptr = options;
530 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
531 for (this_char = strtok (options, ",");
533 this_char = strtok (NULL, ",")) {
535 while ((this_char = strsep (&opt_ptr, ",")) != NULL) {
537 CDEBUG(D_SUPER, "this_char %s\n", this_char);
538 if (!*lov && (*lov = ll_read_opt("osc", this_char)))
540 if (!*lmv && (*lmv = ll_read_opt("mdc", this_char)))
542 if (!*gkc && (*gkc = ll_read_opt("gkc", this_char)))
544 if (!strncmp(this_char, "lasync", strlen("lasync"))) {
548 if (!*mds_sec && (*mds_sec = ll_read_opt("mds_sec", this_char)))
550 if (!*oss_sec && (*oss_sec = ll_read_opt("oss_sec", this_char)))
552 if (!(*flags & LL_SBI_NOLCK) &&
553 ((*flags) = (*flags) |
554 ll_set_opt("nolock", this_char,
562 void ll_lli_init(struct ll_inode_info *lli)
564 sema_init(&lli->lli_open_sem, 1);
565 sema_init(&lli->lli_size_sem, 1);
567 lli->lli_size_pid = 0;
568 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
569 spin_lock_init(&lli->lli_lock);
570 INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
571 INIT_LIST_HEAD(&lli->lli_close_item);
572 lli->lli_inode_magic = LLI_INODE_MAGIC;
573 memset(&lli->lli_id, 0, sizeof(lli->lli_id));
574 sema_init(&lli->lli_och_sem, 1);
575 lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
576 lli->lli_mds_exec_och = NULL;
577 lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
578 lli->lli_open_fd_exec_count = 0;
579 lli->lli_audit_mask = AUDIT_OFF;
580 lli->lli_key_info = NULL;
583 int ll_fill_super(struct super_block *sb, void *data, int silent)
585 struct ll_sb_info *sbi;
586 char *lov = NULL, *lmv = NULL, *gkc = NULL;
587 char *mds_sec = NULL;
588 char *oss_sec = NULL;
590 __u32 nllu[2] = { NOBODY_UID, NOBODY_GID };
591 __u64 remote_flag = 0;
594 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
596 sbi = lustre_init_sbi(sb);
600 sbi->ll_flags |= LL_SBI_READAHEAD;
601 ll_options(data, &lov, &lmv, &gkc, &mds_sec, &oss_sec,
602 &async, &sbi->ll_flags);
605 CERROR("no osc %p or no mdc %p\n", lov, lmv);
606 GOTO(out, err = -EINVAL);
609 err = lustre_common_fill_super(sb, lmv, lov, gkc, async, mds_sec,
610 oss_sec, nllu, 0, &remote_flag);
617 OBD_FREE(lmv, strlen(lmv) + 1);
619 OBD_FREE(lov, strlen(lov) + 1);
621 OBD_FREE(mds_sec, strlen(mds_sec) + 1);
623 OBD_FREE(oss_sec, strlen(oss_sec) + 1);
625 OBD_FREE(gkc, strlen(gkc) + 1);
628 } /* ll_read_super */
630 static int lustre_process_log(struct lustre_mount_data *lmd, char *profile,
631 struct config_llog_instance *cfg, int allow_recov)
633 struct lustre_cfg *lcfg = NULL;
634 struct lustre_cfg_bufs bufs;
635 struct portals_cfg pcfg;
636 char *peer = "MDS_PEER_UUID";
637 struct obd_device *obd;
638 struct lustre_handle md_conn = {0, };
639 struct obd_export *exp;
640 char *name = "mdc_dev";
642 struct obd_uuid lmv_uuid;
643 struct llog_ctxt *ctxt;
647 if (lmd_bad_magic(lmd))
650 generate_random_uuid(uuid);
651 class_uuid_unparse(uuid, &lmv_uuid);
653 if (lmd->lmd_local_nid) {
654 PCFG_INIT(pcfg, NAL_CMD_REGISTER_MYNID);
655 pcfg.pcfg_nal = lmd->lmd_nal;
656 pcfg.pcfg_nid = lmd->lmd_local_nid;
657 rc = libcfs_nal_cmd(&pcfg);
662 if (lmd->lmd_nal == SOCKNAL ||
663 lmd->lmd_nal == OPENIBNAL ||
664 lmd->lmd_nal == IIBNAL ||
665 lmd->lmd_nal == VIBNAL ||
666 lmd->lmd_nal == RANAL) {
667 PCFG_INIT(pcfg, NAL_CMD_ADD_PEER);
668 pcfg.pcfg_nal = lmd->lmd_nal;
669 pcfg.pcfg_nid = lmd->lmd_server_nid;
670 pcfg.pcfg_id = lmd->lmd_server_ipaddr;
671 pcfg.pcfg_misc = lmd->lmd_port;
672 rc = libcfs_nal_cmd(&pcfg);
676 lustre_cfg_bufs_reset(&bufs, name);
677 lustre_cfg_bufs_set_string(&bufs, 1, peer);
679 lcfg = lustre_cfg_new(LCFG_ADD_UUID, &bufs);
680 lcfg->lcfg_nal = lmd->lmd_nal;
681 lcfg->lcfg_nid = lmd->lmd_server_nid;
682 LASSERT(lcfg->lcfg_nal);
683 LASSERT(lcfg->lcfg_nid);
684 err = class_process_config(lcfg);
685 lustre_cfg_free(lcfg);
687 GOTO(out_del_conn, err);
689 lustre_cfg_bufs_reset(&bufs, name);
690 lustre_cfg_bufs_set_string(&bufs, 1, OBD_MDC_DEVICENAME);
691 lustre_cfg_bufs_set_string(&bufs, 2, (char *)lmv_uuid.uuid);
693 lcfg = lustre_cfg_new(LCFG_ATTACH, &bufs);
694 err = class_process_config(lcfg);
695 lustre_cfg_free(lcfg);
697 GOTO(out_del_uuid, err);
699 lustre_cfg_bufs_reset(&bufs, name);
700 lustre_cfg_bufs_set_string(&bufs, 1, lmd->lmd_mds);
701 lustre_cfg_bufs_set_string(&bufs, 2, peer);
703 lcfg = lustre_cfg_new(LCFG_SETUP, &bufs);
704 err = class_process_config(lcfg);
705 lustre_cfg_free(lcfg);
707 GOTO(out_detach, err);
709 obd = class_name2obd(name);
711 GOTO(out_cleanup, rc = -EINVAL);
713 rc = obd_set_info(obd->obd_self_export, strlen("sec"), "sec",
714 strlen(lmd->lmd_mds_security), lmd->lmd_mds_security);
716 GOTO(out_cleanup, rc);
719 unsigned long sec_flags = PTLRPC_SEC_FL_PAG;
720 rc = obd_set_info(obd->obd_self_export,
721 strlen("sec_flags"), "sec_flags",
722 sizeof(sec_flags), &sec_flags);
724 GOTO(out_cleanup, rc);
727 /* Disable initial recovery on this import */
728 rc = obd_set_info(obd->obd_self_export,
729 strlen("initial_recov"), "initial_recov",
730 sizeof(allow_recov), &allow_recov);
732 GOTO(out_cleanup, rc);
734 rc = obd_connect(&md_conn, obd, &lmv_uuid, NULL, 0);
736 CERROR("cannot connect to %s: rc = %d\n", lmd->lmd_mds, rc);
737 GOTO(out_cleanup, rc);
740 exp = class_conn2export(&md_conn);
742 ctxt = llog_get_context(&exp->exp_obd->obd_llogs,LLOG_CONFIG_REPL_CTXT);
743 rc = class_config_process_llog(ctxt, profile, cfg);
745 CERROR("class_config_process_llog failed: rc = %d\n", rc);
747 err = obd_disconnect(exp, 0);
751 lustre_cfg_bufs_reset(&bufs, name);
752 lcfg = lustre_cfg_new(LCFG_CLEANUP, &bufs);
753 err = class_process_config(lcfg);
754 lustre_cfg_free(lcfg);
758 lustre_cfg_bufs_reset(&bufs, name);
759 lcfg = lustre_cfg_new(LCFG_DETACH, &bufs);
760 err = class_process_config(lcfg);
761 lustre_cfg_free(lcfg);
766 lustre_cfg_bufs_reset(&bufs, name);
767 lustre_cfg_bufs_set_string(&bufs, 1, peer);
768 lcfg = lustre_cfg_new(LCFG_DEL_UUID, &bufs);
769 err = class_process_config(lcfg);
770 lustre_cfg_free(lcfg);
773 if (lmd->lmd_nal == SOCKNAL ||
774 lmd->lmd_nal == OPENIBNAL ||
775 lmd->lmd_nal == IIBNAL ||
776 lmd->lmd_nal == VIBNAL ||
777 lmd->lmd_nal == RANAL) {
780 PCFG_INIT(pcfg, NAL_CMD_DEL_PEER);
781 pcfg.pcfg_nal = lmd->lmd_nal;
782 pcfg.pcfg_nid = lmd->lmd_server_nid;
783 pcfg.pcfg_flags = 1; /* single_share */
784 err2 = libcfs_nal_cmd(&pcfg);
797 static void lustre_manual_cleanup(struct ll_sb_info *sbi)
799 struct lustre_cfg *lcfg;
800 struct lustre_cfg_bufs bufs;
801 struct obd_device *obd;
804 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) != NULL) {
807 lustre_cfg_bufs_reset(&bufs, obd->obd_name);
808 lcfg = lustre_cfg_new(LCFG_CLEANUP, &bufs);
809 err = class_process_config(lcfg);
811 CERROR("cleanup failed: %s\n", obd->obd_name);
815 lcfg->lcfg_command = LCFG_DETACH;
816 err = class_process_config(lcfg);
817 lustre_cfg_free(lcfg);
819 CERROR("detach failed: %s\n", obd->obd_name);
824 if (sbi->ll_lmd != NULL)
825 class_del_profile(sbi->ll_lmd->lmd_profile);
828 static int lustre_process_profile(struct super_block *sb,
829 struct lustre_mount_data *lmd,
830 char **lov, char **lmv, char **gkc)
832 struct ll_sb_info *sbi = ll_s2sbi(sb);
833 struct lustre_profile *lprof;
834 struct config_llog_instance cfg;
839 if (!lmd->lmd_profile)
842 if (lmd->lmd_mds[0] == '\0') {
843 CERROR("no mds name\n");
844 GOTO(out, err = -EINVAL);
846 lmd->lmd_mds_security[sizeof(lmd->lmd_mds_security) - 1] = 0;
847 lmd->lmd_oss_security[sizeof(lmd->lmd_oss_security) - 1] = 0;
849 OBD_ALLOC(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
850 if (sbi->ll_lmd == NULL)
851 GOTO(out, err = -ENOMEM);
852 memcpy(sbi->ll_lmd, lmd, sizeof(*lmd));
854 /* generate a string unique to this super, let's try
855 the address of the super itself.*/
856 len = (sizeof(sb) * 2) + 1;
857 OBD_ALLOC(sbi->ll_instance, len);
858 if (sbi->ll_instance == NULL)
859 GOTO(out, err = -ENOMEM);
860 sprintf(sbi->ll_instance, "%p", sb);
862 cfg.cfg_instance = sbi->ll_instance;
863 cfg.cfg_uuid = sbi->ll_sb_uuid;
864 cfg.cfg_local_nid = lmd->lmd_local_nid;
865 err = lustre_process_log(lmd, lmd->lmd_profile, &cfg, 0);
867 CERROR("Unable to process log: %s\n", lmd->lmd_profile);
871 lprof = class_get_profile(lmd->lmd_profile);
873 CERROR("No profile found: %s\n", lmd->lmd_profile);
874 GOTO(out, err = -EINVAL);
877 OBD_ALLOC(*lov, strlen(lprof->lp_lov) +
878 strlen(sbi->ll_instance) + 2);
879 sprintf(*lov, "%s-%s", lprof->lp_lov, sbi->ll_instance);
881 OBD_ALLOC(*lmv, strlen(lprof->lp_lmv) +
882 strlen(sbi->ll_instance) + 2);
883 sprintf(*lmv, "%s-%s", lprof->lp_lmv, sbi->ll_instance);
886 OBD_ALLOC(*gkc, strlen(lprof->lp_gkc) +
887 strlen(sbi->ll_instance) + 2);
888 sprintf(*gkc, "%s-%s", lprof->lp_gkc, sbi->ll_instance);
894 static int lustre_clean_profile(struct ll_sb_info *sbi, int force_umount)
896 struct lustre_mount_data *lmd = sbi->ll_lmd;
897 struct config_llog_instance cfg;
905 len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean") + 1;
908 CERROR("force umount, doing manual cleanup\n");
909 lustre_manual_cleanup(sbi);
913 if (sbi->ll_instance != NULL) {
914 cfg.cfg_instance = sbi->ll_instance;
915 cfg.cfg_uuid = sbi->ll_sb_uuid;
917 OBD_ALLOC(cl_prof, len);
918 sprintf(cl_prof, "%s-clean", lmd->lmd_profile);
919 err = lustre_process_log(lmd, cl_prof, &cfg, 0);
921 CERROR("Unable to process log: %s\n", cl_prof);
922 lustre_manual_cleanup(sbi);
924 OBD_FREE(cl_prof, len);
927 if (sbi->ll_instance)
928 OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance) + 1);
929 OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
933 int lustre_fill_super(struct super_block *sb, void *data, int silent)
935 struct lustre_mount_data * lmd = data;
936 char *lov = NULL, *lmv = NULL, *gkc = NULL;
937 struct ll_sb_info *sbi;
941 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
942 if (lmd_bad_magic(lmd))
945 sbi = lustre_init_sbi(sb);
949 sbi->ll_flags |= LL_SBI_READAHEAD;
951 err = lustre_process_profile(sb, lmd, &lov, &lmv, &gkc);
953 CERROR("Can not process the profile err %d \n", err);
957 CERROR("no osc %p or no mdc %p \n", lov, lmv);
958 GOTO(out_free, err = -EINVAL);
961 err = lustre_common_fill_super(sb, lmv, lov, gkc, lmd->lmd_async,
962 lmd->lmd_mds_security,
963 lmd->lmd_oss_security,
964 &lmd->lmd_nllu, lmd->lmd_pag,
965 &lmd->lmd_remote_flag);
972 OBD_FREE(lmv, strlen(lmv) + 1);
974 OBD_FREE(lov, strlen(lov) + 1);
976 OBD_FREE(gkc, strlen(gkc) + 1);
980 lustre_clean_profile(sbi, 0);
984 } /* lustre_fill_super */
986 void lustre_put_super(struct super_block *sb)
988 struct obd_device *obd;
989 struct ll_sb_info *sbi = ll_s2sbi(sb);
990 int force_umount = 0;
993 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
994 obd = class_exp2obd(sbi->ll_md_exp);
996 force_umount = obd->obd_no_recov;
999 lustre_common_put_super(sb);
1000 lustre_clean_profile(sbi, force_umount);
1001 lustre_free_sbi(sb);
1004 } /* lustre_put_super */
1006 int ll_process_config_update(struct ll_sb_info *sbi, int clean)
1008 struct lustre_mount_data *lmd = sbi->ll_lmd;
1009 char *profile = lmd->lmd_profile, *name = NULL;
1010 struct config_llog_instance cfg;
1011 int rc, namelen = 0, version;
1012 struct llog_ctxt *ctxt;
1015 if (profile == NULL)
1018 CERROR("Client not mounted with zero-conf; cannot "
1019 "process update log.\n");
1023 rc = obd_cancel_unused(sbi->ll_md_exp, NULL,
1024 LDLM_FL_CONFIG_CHANGE, NULL);
1026 CWARN("obd_cancel_unused(mdc): %d\n", rc);
1028 rc = obd_cancel_unused(sbi->ll_dt_exp, NULL,
1029 LDLM_FL_CONFIG_CHANGE, NULL);
1031 CWARN("obd_cancel_unused(lov): %d\n", rc);
1033 cfg.cfg_instance = sbi->ll_instance;
1034 cfg.cfg_uuid = sbi->ll_sb_uuid;
1035 cfg.cfg_local_nid = lmd->lmd_local_nid;
1037 namelen = strlen(profile) + 20; /* -clean-######### */
1038 OBD_ALLOC(name, namelen);
1043 version = sbi->ll_config_version - 1;
1044 sprintf(name, "%s-clean-%d", profile, version);
1046 version = sbi->ll_config_version + 1;
1047 sprintf(name, "%s-%d", profile, version);
1050 CWARN("Applying configuration log %s\n", name);
1052 ctxt = llog_get_context(&sbi->ll_md_exp->exp_obd->obd_llogs,
1053 LLOG_CONFIG_REPL_CTXT);
1054 rc = class_config_process_llog(ctxt, name, &cfg);
1056 sbi->ll_config_version = version;
1057 CWARN("Finished applying configuration log %s: %d\n", name, rc);
1059 if (rc == 0 && clean == 0) {
1060 struct lov_desc desc;
1064 valsize = sizeof(desc);
1065 rc = obd_get_info(sbi->ll_dt_exp, strlen("lovdesc") + 1,
1066 "lovdesc", &valsize, &desc);
1068 rc = obd_init_ea_size(sbi->ll_md_exp,
1069 obd_size_diskmd(sbi->ll_dt_exp, NULL),
1070 (desc.ld_tgt_count *
1071 sizeof(struct llog_cookie)));
1073 OBD_FREE(name, namelen);
1077 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1079 struct inode *inode = NULL;
1081 /* NOTE: we depend on atomic igrab() -bzzz */
1082 lock_res_and_lock(lock);
1083 if (lock->l_ast_data) {
1084 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1085 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1086 inode = igrab(lock->l_ast_data);
1089 do_gettimeofday(&now);
1090 inode = lock->l_ast_data;
1091 LDLM_ERROR(lock, "granted at %lu.%lu, now %lu.%lu",
1092 lock->l_enqueued_time.tv_sec,
1093 lock->l_enqueued_time.tv_usec,
1094 now.tv_sec, now.tv_usec);
1095 CDEBUG(inode->i_state & I_FREEING ? D_INFO : D_WARNING,
1096 "l_ast_data %p is bogus: magic %0x8\n",
1097 lock->l_ast_data, lli->lli_inode_magic);
1098 CDEBUG(D_ERROR, "i_state = 0x%lx, l_ast_data %p is bogus: magic %0x8\n",
1099 inode->i_state, lock->l_ast_data, lli->lli_inode_magic);
1101 unlock_res_and_lock(lock);
1105 unlock_res_and_lock(lock);
1109 int null_if_equal(struct ldlm_lock *lock, void *data)
1111 if (data == lock->l_ast_data) {
1112 lock->l_ast_data = NULL;
1114 if (lock->l_req_mode != lock->l_granted_mode)
1115 LDLM_ERROR(lock,"clearing inode with ungranted lock\n");
1118 return LDLM_ITER_CONTINUE;
1121 static void remote_acl_free(struct remote_acl *racl);
1123 void ll_clear_inode(struct inode *inode)
1125 struct lustre_id id;
1126 struct ll_inode_info *lli = ll_i2info(inode);
1127 struct ll_sb_info *sbi = ll_i2sbi(inode);
1130 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1131 inode->i_generation, inode);
1133 ll_inode2id(&id, inode);
1135 clear_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &(ll_i2info(inode)->lli_flags));
1136 md_change_cbdata(sbi->ll_md_exp, &id, null_if_equal, inode);
1138 LASSERT(!lli->lli_open_fd_write_count);
1139 LASSERT(!lli->lli_open_fd_read_count);
1140 LASSERT(!lli->lli_open_fd_exec_count);
1141 if (lli->lli_mds_write_och)
1142 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_WRITE);
1143 if (lli->lli_mds_exec_och)
1144 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_EXEC);
1145 if (lli->lli_mds_read_och)
1146 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_READ);
1148 obd_change_cbdata(sbi->ll_dt_exp, lli->lli_smd,
1149 null_if_equal, inode);
1152 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1153 lli->lli_smd = NULL;
1157 obd_free_memmd(sbi->ll_md_exp,
1158 (struct lov_stripe_md **) &lli->lli_mea);
1159 lli->lli_mea = NULL;
1161 ll_crypto_destroy_inode_key(inode);
1162 if (lli->lli_symlink_name) {
1163 OBD_FREE(lli->lli_symlink_name,
1164 strlen(lli->lli_symlink_name) + 1);
1165 lli->lli_symlink_name = NULL;
1168 if (lli->lli_posix_acl) {
1169 LASSERT(lli->lli_remote_acl == NULL);
1170 posix_acl_release(lli->lli_posix_acl);
1171 lli->lli_posix_acl = NULL;
1174 if (lli->lli_remote_acl) {
1175 LASSERT(lli->lli_posix_acl == NULL);
1176 remote_acl_free(lli->lli_remote_acl);
1177 lli->lli_remote_acl = NULL;
1180 lli->lli_inode_magic = LLI_INODE_DEAD;
1185 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1186 * object(s) determine the file size and mtime. Otherwise, the MDS will
1187 * keep these values until such a time that objects are allocated for it.
1188 * We do the MDS operations first, as it is checking permissions for us.
1189 * We don't to the MDS RPC if there is nothing that we want to store there,
1190 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1191 * going to do an RPC anyways.
1193 * If we are doing a truncate, we will send the mtime and ctime updates
1194 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1195 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1198 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1200 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1201 struct ll_inode_info *lli = ll_i2info(inode);
1202 struct ll_sb_info *sbi = ll_i2sbi(inode);
1203 struct ptlrpc_request *request = NULL;
1204 struct mdc_op_data *op_data;
1205 int ia_valid = attr->ia_valid;
1209 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu\n", inode->i_ino);
1210 lprocfs_counter_incr(ll_i2sbi(inode)->ll_stats, LPROC_LL_SETATTR);
1212 if (ia_valid & ATTR_SIZE) {
1213 if (attr->ia_size > ll_file_maxbytes(inode)) {
1214 CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1215 attr->ia_size, ll_file_maxbytes(inode));
1219 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1222 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1223 if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1224 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1228 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1229 if (attr->ia_valid & ATTR_CTIME) {
1230 attr->ia_ctime = CURRENT_TIME;
1231 attr->ia_valid |= ATTR_CTIME_SET;
1233 if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1234 attr->ia_atime = CURRENT_TIME;
1235 attr->ia_valid |= ATTR_ATIME_SET;
1237 if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1238 attr->ia_mtime = CURRENT_TIME;
1239 attr->ia_valid |= ATTR_MTIME_SET;
1242 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1243 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1244 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1245 LTIME_S(CURRENT_TIME));
1248 attr->ia_valid &= ~ATTR_SIZE;
1250 /* If only OST attributes being set on objects, don't do MDS RPC.
1251 * In that case, we need to check permissions and update the local
1252 * inode ourselves so we can call obdo_from_inode() always. */
1253 if (ia_valid & (lsm ? ~(ATTR_SIZE | ATTR_FROM_OPEN /*| ATTR_RAW*/) : ~0)) {
1254 struct lustre_md md;
1258 OBD_ALLOC(op_data, sizeof(*op_data));
1259 if (op_data == NULL)
1261 ll_prepare_mdc_data(op_data, inode, NULL, NULL, 0, 0);
1263 if (ia_valid & (ATTR_UID | ATTR_GID)) {
1264 rc = ll_crypto_get_mac(inode, attr, NULL, 0, &key,
1267 rc = md_setattr(sbi->ll_md_exp, op_data,
1268 attr, key, key_size, NULL, 0, NULL,
1270 OBD_FREE(op_data, sizeof(*op_data));
1272 if (key && key_size)
1273 OBD_FREE(key, key_size);
1275 ptlrpc_req_finished(request);
1276 if (rc != -EPERM && rc != -EACCES)
1277 CERROR("md_setattr fails: rc = %d\n", rc);
1280 rc = mdc_req2lustre_md(sbi->ll_md_exp, request, 0,
1281 sbi->ll_dt_exp, &md);
1283 ptlrpc_req_finished(request);
1287 /* We call inode_setattr to adjust timestamps, but we first
1288 * clear ATTR_SIZE to avoid invoking vmtruncate.
1290 * NB: ATTR_SIZE will only be set at this point if the size
1291 * resides on the MDS, ie, this file has no objects. */
1292 attr->ia_valid &= ~ATTR_SIZE;
1295 * assigning inode_setattr() to @err to disable warning that
1296 * function's result should be checked by by caller. error is
1297 * impossible here, as vmtruncate() control path is disabled.
1299 err = inode_setattr(inode, attr);
1300 ll_update_inode(inode, &md);
1301 ptlrpc_req_finished(request);
1303 if (!lsm || !S_ISREG(inode->i_mode)) {
1304 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1308 /* The OST doesn't check permissions, but the alternative is
1309 * a gratuitous RPC to the MDS. We already rely on the client
1310 * to do read/write/truncate permission checks, so is mtime OK?
1312 if (ia_valid & (ATTR_MTIME | ATTR_ATIME)) {
1313 /* from sys_utime() */
1314 if (!(ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET))) {
1315 if (current->fsuid != inode->i_uid &&
1316 (rc = ll_permission(inode, MAY_WRITE, NULL)) != 0)
1319 /* from inode_change_ok() */
1320 if (current->fsuid != inode->i_uid &&
1321 !capable(CAP_FOWNER))
1326 /* won't invoke vmtruncate, as we already cleared ATTR_SIZE */
1327 err = inode_setattr(inode, attr);
1329 * assigning inode_setattr() to @err to disable warning that
1330 * function's result should be checked by by caller. error is
1331 * impossible here, as vmtruncate() control path is disabled.
1335 /* We really need to get our PW lock before we change inode->i_size.
1336 * If we don't we can race with other i_size updaters on our node, like
1337 * ll_file_read. We can also race with i_size propogation to other
1338 * nodes through dirtying and writeback of final cached pages. This
1339 * last one is especially bad for racing o_append users on other
1341 if (ia_valid & ATTR_SIZE) {
1342 ldlm_policy_data_t policy = { .l_extent = {attr->ia_size,
1344 struct lustre_handle lockh = { 0 };
1345 int err, ast_flags = 0;
1346 /* XXX when we fix the AST intents to pass the discard-range
1347 * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
1349 if (attr->ia_size == 0)
1350 ast_flags = LDLM_AST_DISCARD_DATA;
1352 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy, &lockh,
1353 ast_flags, &ll_i2sbi(inode)->ll_seek_stime);
1358 down(&lli->lli_size_sem);
1359 lli->lli_size_pid = current->pid;
1360 rc = vmtruncate(inode, attr->ia_size);
1362 LASSERT(atomic_read(&lli->lli_size_sem.count) <= 0);
1363 lli->lli_size_pid = 0;
1364 up(&lli->lli_size_sem);
1367 err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
1369 CERROR("ll_extent_unlock failed: %d\n", err);
1373 } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET | ATTR_UID | ATTR_GID)) {
1374 struct obdo *oa = NULL;
1376 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1377 inode->i_ino, LTIME_S(attr->ia_mtime));
1383 oa->o_id = lsm->lsm_object_id;
1384 oa->o_gr = lsm->lsm_object_gr;
1385 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1387 /* adding uid and gid, needed for quota */
1388 if (ia_valid & ATTR_UID) {
1389 oa->o_uid = inode->i_uid;
1390 oa->o_valid |= OBD_MD_FLUID;
1393 if (ia_valid & ATTR_GID) {
1394 oa->o_gid = inode->i_gid;
1395 oa->o_valid |= OBD_MD_FLGID;
1398 *(obdo_id(oa)) = lli->lli_id;
1399 oa->o_valid |= OBD_MD_FLIFID;
1401 obdo_from_inode(oa, inode, OBD_MD_FLTYPE | OBD_MD_FLATIME |
1402 OBD_MD_FLMTIME | OBD_MD_FLCTIME);
1403 rc = obd_setattr(sbi->ll_dt_exp, oa, lsm, NULL);
1406 CERROR("obd_setattr fails: rc = %d\n", rc);
1412 int ll_setattr(struct dentry *de, struct iattr *attr)
1414 LASSERT(de->d_inode);
1415 return ll_setattr_raw(de->d_inode, attr);
1418 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1419 unsigned long max_age)
1421 struct ll_sb_info *sbi = ll_s2sbi(sb);
1422 struct obd_statfs obd_osfs;
1426 rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age);
1428 CERROR("obd_statfs fails: rc = %d\n", rc);
1432 osfs->os_type = sb->s_magic;
1434 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1435 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1437 rc = obd_statfs(class_exp2obd(sbi->ll_dt_exp), &obd_osfs, max_age);
1439 CERROR("obd_statfs fails: rc = %d\n", rc);
1443 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1444 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1447 osfs->os_blocks = obd_osfs.os_blocks;
1448 osfs->os_bfree = obd_osfs.os_bfree;
1449 osfs->os_bavail = obd_osfs.os_bavail;
1451 /* If we don't have as many objects free on the OST as inodes
1452 * on the MDS, we reduce the total number of inodes to
1453 * compensate, so that the "inodes in use" number is correct.
1455 if (obd_osfs.os_ffree < osfs->os_ffree) {
1456 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1458 osfs->os_ffree = obd_osfs.os_ffree;
1464 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1466 struct obd_statfs osfs;
1469 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p\n", sb);
1470 lprocfs_counter_incr(ll_s2sbi(sb)->ll_stats, LPROC_LL_STAFS);
1472 /* For now we will always get up-to-date statfs values, but in the
1473 * future we may allow some amount of caching on the client (e.g.
1474 * from QOS or lprocfs updates). */
1475 rc = ll_statfs_internal(sb, &osfs, jiffies - 1);
1479 statfs_unpack(sfs, &osfs);
1481 if (sizeof(sfs->f_blocks) == 4) {
1482 while (osfs.os_blocks > ~0UL) {
1485 osfs.os_blocks >>= 1;
1486 osfs.os_bfree >>= 1;
1487 osfs.os_bavail >>= 1;
1491 sfs->f_blocks = osfs.os_blocks;
1492 sfs->f_bfree = osfs.os_bfree;
1493 sfs->f_bavail = osfs.os_bavail;
1499 /********************************
1501 ********************************/
1503 static struct remote_acl *remote_acl_alloc(void)
1505 struct remote_acl *racl;
1508 OBD_ALLOC(racl, sizeof(*racl));
1512 spin_lock_init(&racl->ra_lock);
1513 init_MUTEX(&racl->ra_update_sem);
1515 for (i = 0; i < REMOTE_ACL_HASHSIZE; i++)
1516 INIT_LIST_HEAD(&racl->ra_perm_cache[i]);
1522 * caller should guarantee no race here.
1524 static void remote_perm_flush_xperms(struct lustre_remote_perm *perm)
1526 struct remote_perm_setxid *xperm;
1528 while (!list_empty(&perm->lrp_setxid_perms)) {
1529 xperm = list_entry(perm->lrp_setxid_perms.next,
1530 struct remote_perm_setxid,
1532 list_del(&xperm->list);
1533 OBD_FREE(xperm, sizeof(*xperm));
1538 * caller should guarantee no race here.
1540 static void remote_acl_flush(struct remote_acl *racl)
1542 struct list_head *head;
1543 struct lustre_remote_perm *perm, *tmp;
1546 for (i = 0; i < REMOTE_ACL_HASHSIZE; i++) {
1547 head = &racl->ra_perm_cache[i];
1549 list_for_each_entry_safe(perm, tmp, head, lrp_list) {
1550 remote_perm_flush_xperms(perm);
1551 list_del(&perm->lrp_list);
1552 OBD_FREE(perm, sizeof(*perm));
1557 static void remote_acl_free(struct remote_acl *racl)
1562 down(&racl->ra_update_sem);
1563 spin_lock(&racl->ra_lock);
1564 remote_acl_flush(racl);
1565 spin_unlock(&racl->ra_lock);
1566 up(&racl->ra_update_sem);
1568 OBD_FREE(racl, sizeof(*racl));
1571 static inline int remote_acl_hashfunc(__u32 id)
1573 return (id & (REMOTE_ACL_HASHSIZE - 1));
1577 int __remote_acl_check(struct remote_acl *racl, unsigned int *perm)
1579 struct list_head *head;
1580 struct lustre_remote_perm *lperm;
1581 struct remote_perm_setxid *xperm;
1582 int found = 0, rc = -ENOENT;
1585 head = &racl->ra_perm_cache[remote_acl_hashfunc(current->uid)];
1586 spin_lock(&racl->ra_lock);
1588 list_for_each_entry(lperm, head, lrp_list) {
1589 if (lperm->lrp_auth_uid == current->uid) {
1598 if (lperm->lrp_auth_uid == current->fsuid &&
1599 lperm->lrp_auth_gid == current->fsgid) {
1600 if (lperm->lrp_valid) {
1601 *perm = lperm->lrp_perm;
1605 } else if ((!lperm->lrp_setuid &&
1606 lperm->lrp_auth_uid != current->fsuid) ||
1607 (!lperm->lrp_setgid &&
1608 lperm->lrp_auth_gid != current->fsgid)) {
1614 list_for_each_entry(xperm, &lperm->lrp_setxid_perms, list) {
1615 if (xperm->uid == current->fsuid &&
1616 xperm->gid == current->fsgid) {
1617 *perm = xperm->perm;
1624 spin_unlock(&racl->ra_lock);
1629 int __remote_acl_update(struct remote_acl *racl,
1630 struct mds_remote_perm *mperm,
1631 struct lustre_remote_perm *lperm,
1632 struct remote_perm_setxid *xperm)
1634 struct list_head *head;
1635 struct lustre_remote_perm *lp;
1636 struct remote_perm_setxid *xp;
1637 int found = 0, setuid = 0, setgid = 0;
1642 LASSERT(current->uid == mperm->mrp_auth_uid);
1644 if (current->fsuid != mperm->mrp_auth_uid)
1646 if (current->fsgid != mperm->mrp_auth_gid)
1649 head = &racl->ra_perm_cache[remote_acl_hashfunc(current->uid)];
1650 spin_lock(&racl->ra_lock);
1652 list_for_each_entry(lp, head, lrp_list) {
1653 if (lp->lrp_auth_uid == current->uid) {
1660 OBD_FREE(lperm, sizeof(*lperm));
1662 if (!lp->lrp_valid && !setuid && !setgid) {
1663 lp->lrp_perm = mperm->mrp_perm;
1667 /* sanity check for changes of setxid rules */
1668 if ((lp->lrp_setuid != 0) != (mperm->mrp_allow_setuid != 0)) {
1669 CWARN("setuid changes: %d => %d\n",
1670 (lp->lrp_setuid != 0),
1671 (mperm->mrp_allow_setuid != 0));
1672 lp->lrp_setuid = (mperm->mrp_allow_setuid != 0);
1675 if ((lp->lrp_setgid != 0) != (mperm->mrp_allow_setgid != 0)) {
1676 CWARN("setgid changes: %d => %d\n",
1677 (lp->lrp_setgid != 0),
1678 (mperm->mrp_allow_setgid != 0));
1679 lp->lrp_setgid = (mperm->mrp_allow_setgid != 0);
1682 if (!lp->lrp_setuid && !lp->lrp_setgid &&
1683 !list_empty(&lp->lrp_setxid_perms)) {
1684 remote_perm_flush_xperms(lp);
1687 /* initialize lperm and linked into hashtable
1689 INIT_LIST_HEAD(&lperm->lrp_setxid_perms);
1690 lperm->lrp_auth_uid = mperm->mrp_auth_uid;
1691 lperm->lrp_auth_gid = mperm->mrp_auth_gid;
1692 lperm->lrp_setuid = (mperm->mrp_allow_setuid != 0);
1693 lperm->lrp_setgid = (mperm->mrp_allow_setgid != 0);
1694 list_add(&lperm->lrp_list, head);
1696 if (!setuid && !setgid) {
1697 /* in this case, i'm the authenticated user,
1698 * and mrp_perm is for me.
1700 lperm->lrp_perm = mperm->mrp_perm;
1701 lperm->lrp_valid = 1;
1702 spin_unlock(&racl->ra_lock);
1705 OBD_FREE(xperm, sizeof(*xperm));
1713 LASSERT(lp->lrp_setuid || lp->lrp_setgid ||
1714 list_empty(&lp->lrp_setxid_perms));
1716 /* if no xperm supplied, we are all done here */
1718 spin_unlock(&racl->ra_lock);
1722 /* whether we allow setuid/setgid */
1723 if ((!lp->lrp_setuid && setuid) || (!lp->lrp_setgid && setgid)) {
1724 OBD_FREE(xperm, sizeof(*xperm));
1725 spin_unlock(&racl->ra_lock);
1729 /* traverse xperm list */
1730 list_for_each_entry(xp, &lp->lrp_setxid_perms, list) {
1731 if (xp->uid == current->fsuid &&
1732 xp->gid == current->fsgid) {
1733 if (xp->perm != mperm->mrp_perm) {
1734 /* actually this should not happen */
1735 CWARN("perm changed: %o => %o\n",
1736 xp->perm, mperm->mrp_perm);
1737 xp->perm = mperm->mrp_perm;
1739 OBD_FREE(xperm, sizeof(*xperm));
1740 spin_unlock(&racl->ra_lock);
1745 /* finally insert this xperm */
1746 xperm->uid = current->fsuid;
1747 xperm->gid = current->fsgid;
1748 xperm->perm = mperm->mrp_perm;
1749 list_add(&xperm->list, &lp->lrp_setxid_perms);
1751 spin_unlock(&racl->ra_lock);
1756 * remote_acl semaphore must be held by caller
1759 int remote_acl_update_locked(struct remote_acl *racl,
1760 struct mds_remote_perm *mperm)
1762 struct lustre_remote_perm *lperm;
1763 struct remote_perm_setxid *xperm;
1764 int setuid = 0, setgid = 0;
1768 if (current->uid != mperm->mrp_auth_uid) {
1769 CERROR("current uid %u while authenticated as %u\n",
1770 current->uid, mperm->mrp_auth_uid);
1774 if (current->fsuid != mperm->mrp_auth_uid)
1776 if (current->fsgid == mperm->mrp_auth_gid)
1779 OBD_ALLOC(lperm, sizeof(*lperm));
1783 if ((setuid || setgid) &&
1784 !(setuid && !mperm->mrp_allow_setuid) &&
1785 !(setgid && !mperm->mrp_allow_setgid)) {
1786 OBD_ALLOC(xperm, sizeof(*xperm));
1788 OBD_FREE(lperm, sizeof(*lperm));
1794 return __remote_acl_update(racl, mperm, lperm, xperm);
1798 * return -EACCES at any error cases
1800 int ll_remote_acl_permission(struct inode *inode, int mode)
1802 struct ll_sb_info *sbi = ll_i2sbi(inode);
1803 struct remote_acl *racl = ll_i2info(inode)->lli_remote_acl;
1804 struct ptlrpc_request *req = NULL;
1805 struct lustre_id id;
1806 struct mds_remote_perm *mperm;
1807 int rc = -EACCES, perm;
1812 if (__remote_acl_check(racl, &perm) == 0) {
1813 return ((perm & mode) == mode ? 0 : -EACCES);
1820 down(&racl->ra_update_sem);
1822 /* we might lose the race when obtain semaphore,
1825 if (__remote_acl_check(racl, &perm) == 0) {
1826 if ((perm & mode) == mode)
1831 /* really fetch from mds
1833 ll_inode2id(&id, inode);
1834 if (md_access_check(sbi->ll_md_exp, &id, &req))
1837 /* status non-zero indicate there's more apparent error
1838 * detected by mds, e.g. didn't allow this user at all.
1839 * we simply ignore and didn't cache it.
1841 if (req->rq_repmsg->status)
1844 mperm = lustre_swab_repbuf(req, 1, sizeof(*mperm),
1845 lustre_swab_remote_perm);
1847 LASSERT_REPSWABBED(req, 1);
1849 if ((mperm->mrp_perm & mode) == mode)
1852 remote_acl_update_locked(racl, mperm);
1855 ptlrpc_req_finished(req);
1857 up(&racl->ra_update_sem);
1861 int ll_remote_acl_update(struct inode *inode, struct mds_remote_perm *perm)
1863 struct remote_acl *racl = ll_i2info(inode)->lli_remote_acl;
1871 down(&racl->ra_update_sem);
1872 rc = remote_acl_update_locked(racl, perm);
1873 up(&racl->ra_update_sem);
1878 void ll_inode_invalidate_acl(struct inode *inode)
1880 struct ll_sb_info *sbi = ll_i2sbi(inode);
1881 struct ll_inode_info *lli = ll_i2info(inode);
1883 if (sbi->ll_remote) {
1884 struct remote_acl *racl = lli->lli_remote_acl;
1886 LASSERT(!lli->lli_posix_acl);
1888 down(&racl->ra_update_sem);
1889 spin_lock(&racl->ra_lock);
1890 remote_acl_flush(lli->lli_remote_acl);
1891 spin_unlock(&racl->ra_lock);
1892 up(&racl->ra_update_sem);
1895 /* we can't invalide acl here: suppose we touch a new file
1896 * under a dir, blocking ast on dir will lead to open failure
1897 * on client, although succeed on mds. it's kind of weird,
1898 * the real fix i think is improve client-vfs interaction.
1900 * currently we just do nothing here.
1904 LASSERT(!lli->lli_remote_acl);
1905 spin_lock(&lli->lli_lock);
1906 posix_acl_release(lli->lli_posix_acl);
1907 lli->lli_posix_acl = NULL;
1908 spin_unlock(&lli->lli_lock);
1912 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1914 struct ll_inode_info *lli = ll_i2info(inode);
1915 struct lov_stripe_md *lsm = md->lsm;
1916 struct mds_body *body = md->body;
1917 struct mea *mea = md->mea;
1918 struct posix_acl *posix_acl = md->posix_acl;
1919 struct ll_sb_info *sbi = ll_i2sbi(inode);
1920 struct lustre_key *mkey = md->key;
1923 LASSERT((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1925 if (md->lsm && md->lsm->lsm_magic != LOV_MAGIC) {
1926 /* check for default striping info for dir. */
1927 LASSERT((mea != NULL) == ((body->valid & OBD_MD_FLDIREA) != 0));
1931 LASSERT(lsm->lsm_object_gr > 0);
1932 if (lli->lli_smd == NULL) {
1934 lli->lli_maxbytes = lsm->lsm_maxbytes;
1935 if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1936 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1939 if (memcmp(lli->lli_smd, lsm, sizeof(*lsm))) {
1940 CERROR("lsm mismatch for inode %ld\n",
1942 CERROR("lli_smd:\n");
1943 dump_lsm(D_ERROR, lli->lli_smd);
1945 dump_lsm(D_ERROR, lsm);
1948 /* XXX FIXME -- We should decide on a safer (atomic) and
1949 * more elegant way to update the lsm */
1950 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1951 lli->lli_smd->lsm_oinfo[i].loi_id =
1952 lsm->lsm_oinfo[i].loi_id;
1953 lli->lli_smd->lsm_oinfo[i].loi_gr =
1954 lsm->lsm_oinfo[i].loi_gr;
1955 lli->lli_smd->lsm_oinfo[i].loi_ost_idx =
1956 lsm->lsm_oinfo[i].loi_ost_idx;
1957 lli->lli_smd->lsm_oinfo[i].loi_ost_gen =
1958 lsm->lsm_oinfo[i].loi_ost_gen;
1961 /* bug 2844 - limit i_blksize for broken user-space apps */
1962 LASSERTF(lsm->lsm_xfersize != 0, "%lu\n", lsm->lsm_xfersize);
1963 inode->i_blksize = min(lsm->lsm_xfersize, LL_MAX_BLKSIZE);
1964 if (lli->lli_smd != lsm)
1965 obd_free_memmd(ll_i2dtexp(inode), &lsm);
1969 if (lli->lli_mea == NULL) {
1972 if (memcmp(lli->lli_mea, mea, body->eadatasize)) {
1973 CERROR("mea mismatch for inode %lu\n",
1978 if (lli->lli_mea != mea)
1979 obd_free_memmd(ll_i2mdexp(inode),
1980 (struct lov_stripe_md **) &mea);
1983 if (body->valid & OBD_MD_FID)
1984 id_assign_fid(&lli->lli_id, &body->id1);
1986 if (body->valid & OBD_MD_FLID)
1987 id_ino(&lli->lli_id) = id_ino(&body->id1);
1989 if (body->valid & OBD_MD_FLGENER)
1990 id_gen(&lli->lli_id) = id_gen(&body->id1);
1992 /* local/remote ACL */
1993 if (sbi->ll_remote) {
1994 LASSERT(md->posix_acl == NULL);
1995 if (md->remote_perm) {
1996 ll_remote_acl_update(inode, md->remote_perm);
1997 OBD_FREE(md->remote_perm, sizeof(*md->remote_perm));
1998 md->remote_perm = NULL;
2001 LASSERT(md->remote_perm == NULL);
2002 spin_lock(&lli->lli_lock);
2003 if (posix_acl != NULL) {
2004 if (lli->lli_posix_acl != NULL)
2005 posix_acl_release(lli->lli_posix_acl);
2006 lli->lli_posix_acl = posix_acl;
2008 spin_unlock(&lli->lli_lock);
2011 if (body->valid & OBD_MD_FLID)
2012 inode->i_ino = id_ino(&body->id1);
2013 if (body->valid & OBD_MD_FLGENER)
2014 inode->i_generation = id_gen(&body->id1);
2015 if (body->valid & OBD_MD_FLATIME)
2016 LTIME_S(inode->i_atime) = body->atime;
2017 if (body->valid & OBD_MD_FLMTIME &&
2018 body->mtime > LTIME_S(inode->i_mtime)) {
2019 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to %u\n",
2020 inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
2021 LTIME_S(inode->i_mtime) = body->mtime;
2023 if (body->valid & OBD_MD_FLCTIME &&
2024 body->ctime > LTIME_S(inode->i_ctime))
2025 LTIME_S(inode->i_ctime) = body->ctime;
2026 if (body->valid & OBD_MD_FLMODE) {
2027 inode->i_mode = (inode->i_mode & S_IFMT) |
2028 (body->mode & ~S_IFMT);
2030 if (body->valid & OBD_MD_FLTYPE) {
2031 inode->i_mode = (inode->i_mode & ~S_IFMT) |
2032 (body->mode & S_IFMT);
2034 if (body->valid & OBD_MD_FLUID)
2035 inode->i_uid = body->uid;
2036 if (body->valid & OBD_MD_FLGID)
2037 inode->i_gid = body->gid;
2038 if (body->valid & OBD_MD_FLFLAGS)
2039 inode->i_flags = body->flags;
2040 if (body->valid & OBD_MD_FLNLINK)
2041 inode->i_nlink = body->nlink;
2042 if (body->valid & OBD_MD_FLRDEV)
2043 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
2044 inode->i_rdev = body->rdev;
2046 inode->i_rdev = old_decode_dev(body->rdev);
2048 if (body->valid & OBD_MD_FLSIZE)
2049 inode->i_size = body->size;
2050 if (body->valid & OBD_MD_FLBLOCKS)
2051 inode->i_blocks = body->blocks;
2053 if (body->valid & OBD_MD_FLSIZE)
2054 set_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &lli->lli_flags);
2056 if (body->valid & OBD_MD_FLAUDIT) {
2057 struct ll_sb_info * sbi = ll_s2sbi(inode->i_sb);
2058 if (IS_AUDIT_OP(body->audit, AUDIT_FS))
2059 sbi->ll_audit_mask = body->audit;
2061 lli->lli_audit_mask = body->audit;
2065 LASSERT(body->valid & OBD_MD_FLKEY);
2066 ll_crypto_init_inode_key(inode, mkey);
2069 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
2070 inode->i_dev = (kdev_t)id_group(&lli->lli_id);
2072 LASSERT(id_fid(&lli->lli_id) != 0);
2075 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
2076 static struct backing_dev_info ll_backing_dev_info = {
2077 .ra_pages = 0, /* No readahead */
2078 .memory_backed = 0, /* Does contribute to dirty memory */
2082 void ll_read_inode2(struct inode *inode, void *opaque)
2084 struct lustre_md *md = opaque;
2085 struct ll_inode_info *lli = ll_i2info(inode);
2088 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
2089 inode->i_generation, inode);
2093 LASSERT(!lli->lli_smd);
2095 if (ll_i2sbi(inode)->ll_remote) {
2096 lli->lli_remote_acl = remote_acl_alloc();
2097 /* if failed alloc, nobody will be able to access this inode */
2100 /* Core attributes from the MDS first. This is a new inode, and
2101 * the VFS doesn't zero times in the core inode so we have to do
2102 * it ourselves. They will be overwritten by either MDS or OST
2103 * attributes - we just need to make sure they aren't newer. */
2104 LTIME_S(inode->i_mtime) = 0;
2105 LTIME_S(inode->i_atime) = 0;
2106 LTIME_S(inode->i_ctime) = 0;
2109 ll_update_inode(inode, md);
2111 /* OIDEBUG(inode); */
2113 if (S_ISREG(inode->i_mode)) {
2114 inode->i_op = &ll_file_inode_operations;
2115 inode->i_fop = &ll_file_operations;
2116 inode->i_mapping->a_ops = &ll_aops;
2118 } else if (S_ISDIR(inode->i_mode)) {
2119 inode->i_op = &ll_dir_inode_operations;
2120 inode->i_fop = &ll_dir_operations;
2121 inode->i_mapping->a_ops = &ll_dir_aops;
2123 } else if (S_ISLNK(inode->i_mode)) {
2124 inode->i_op = &ll_fast_symlink_inode_operations;
2127 inode->i_op = &ll_special_inode_operations;
2129 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
2130 init_special_inode(inode, inode->i_mode,
2131 kdev_t_to_nr(inode->i_rdev));
2133 /* initializing backing dev info. */
2134 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
2136 init_special_inode(inode, inode->i_mode, inode->i_rdev);
2138 lli->ll_save_ifop = inode->i_fop;
2140 if (S_ISCHR(inode->i_mode))
2141 inode->i_fop = &ll_special_chr_inode_fops;
2142 else if (S_ISBLK(inode->i_mode))
2143 inode->i_fop = &ll_special_blk_inode_fops;
2144 else if (S_ISFIFO(inode->i_mode))
2145 inode->i_fop = &ll_special_fifo_inode_fops;
2146 else if (S_ISSOCK(inode->i_mode))
2147 inode->i_fop = &ll_special_sock_inode_fops;
2149 CWARN("saved %p, replaced with %p\n", lli->ll_save_ifop,
2152 if (lli->ll_save_ifop->owner) {
2153 CWARN("%p has owner %p\n", lli->ll_save_ifop,
2154 lli->ll_save_ifop->owner);
2160 void ll_delete_inode(struct inode *inode)
2162 struct ll_sb_info *sbi = ll_i2sbi(inode);
2163 struct lustre_id id;
2167 ll_inode2id(&id, inode);
2169 rc = md_delete_inode(sbi->ll_md_exp, &id);
2171 CERROR("md_delete_inode() failed, error %d\n",
2179 int ll_iocontrol(struct inode *inode, struct file *file,
2180 unsigned int cmd, unsigned long arg)
2182 struct ll_sb_info *sbi = ll_i2sbi(inode);
2183 struct ptlrpc_request *req = NULL;
2188 case EXT3_IOC_GETFLAGS: {
2189 struct lustre_id id;
2190 __u64 valid = OBD_MD_FLFLAGS;
2191 struct mds_body *body;
2193 ll_inode2id(&id, inode);
2194 rc = md_getattr(sbi->ll_md_exp, &id, valid, NULL, NULL, 0, 0,
2197 CERROR("failure %d inode %lu\n", rc, inode->i_ino);
2201 body = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*body));
2203 if (body->flags & S_APPEND)
2204 flags |= EXT3_APPEND_FL;
2205 if (body->flags & S_IMMUTABLE)
2206 flags |= EXT3_IMMUTABLE_FL;
2207 if (body->flags & S_NOATIME)
2208 flags |= EXT3_NOATIME_FL;
2210 ptlrpc_req_finished (req);
2212 RETURN(put_user(flags, (int *)arg));
2214 case EXT3_IOC_SETFLAGS: {
2215 struct mdc_op_data *op_data;
2218 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
2220 if (get_user(flags, (int *)arg))
2227 OBD_ALLOC(op_data, sizeof(*op_data));
2228 if (op_data == NULL) {
2232 ll_prepare_mdc_data(op_data, inode, NULL, NULL, 0, 0);
2234 memset(&attr, 0x0, sizeof(attr));
2235 attr.ia_attr_flags = flags;
2236 attr.ia_valid |= ATTR_ATTR_FLAG;
2238 rc = md_setattr(sbi->ll_md_exp, op_data,
2239 &attr, NULL, 0, NULL, 0, NULL, 0, &req);
2240 OBD_FREE(op_data, sizeof(*op_data));
2242 ptlrpc_req_finished(req);
2243 if (rc != -EPERM && rc != -EACCES)
2244 CERROR("md_setattr fails: rc = %d\n", rc);
2248 ptlrpc_req_finished(req);
2250 oa->o_id = lsm->lsm_object_id;
2251 oa->o_gr = lsm->lsm_object_gr;
2252 oa->o_flags = flags;
2253 *(obdo_id(oa)) = ll_i2info(inode)->lli_id;
2254 oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS | OBD_MD_FLGROUP
2257 rc = obd_setattr(sbi->ll_dt_exp, oa, lsm, NULL);
2260 if (rc != -EPERM && rc != -EACCES)
2261 CERROR("md_setattr fails: rc = %d\n", rc);
2265 if (flags & EXT3_APPEND_FL)
2266 inode->i_flags |= S_APPEND;
2268 inode->i_flags &= ~S_APPEND;
2269 if (flags & EXT3_IMMUTABLE_FL)
2270 inode->i_flags |= S_IMMUTABLE;
2272 inode->i_flags &= ~S_IMMUTABLE;
2273 if (flags & EXT3_NOATIME_FL)
2274 inode->i_flags |= S_NOATIME;
2276 inode->i_flags &= ~S_NOATIME;
2287 /* this is only called in the case of forced umount. */
2288 void ll_umount_begin(struct super_block *sb)
2290 struct ll_sb_info *sbi = ll_s2sbi(sb);
2291 struct obd_ioctl_data ioc_data = { 0 };
2292 struct obd_device *obd;
2295 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2296 sb->s_count, atomic_read(&sb->s_active));
2298 obd = class_exp2obd(sbi->ll_md_exp);
2300 CERROR("Invalid MDC connection handle "LPX64"\n",
2301 sbi->ll_md_exp->exp_handle.h_cookie);
2305 obd->obd_no_recov = 1;
2306 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2307 sizeof(ioc_data), &ioc_data, NULL);
2309 obd = class_exp2obd(sbi->ll_dt_exp);
2311 CERROR("Invalid LOV connection handle "LPX64"\n",
2312 sbi->ll_dt_exp->exp_handle.h_cookie);
2317 obd->obd_no_recov = 1;
2318 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2319 sizeof(ioc_data), &ioc_data, NULL);
2322 * really, we'd like to wait until there are no requests outstanding,
2323 * and then continue. For now, we just invalidate the requests,
2324 * schedule, and hope.
2331 int ll_prep_inode(struct obd_export *dt_exp, struct obd_export *md_exp,
2332 struct inode **inode, struct ptlrpc_request *req,
2333 int offset, struct super_block *sb)
2335 struct lustre_md md;
2338 rc = mdc_req2lustre_md(md_exp, req, offset, dt_exp, &md);
2343 ll_update_inode(*inode, &md);
2346 *inode = ll_iget(sb, id_ino(&md.body->id1), &md);
2347 if (*inode == NULL || is_bad_inode(*inode)) {
2348 /* free the lsm if we allocated one above */
2350 obd_free_memmd(dt_exp, &md.lsm);
2352 obd_free_memmd(md_exp,
2353 (struct lov_stripe_md**)&md.mea);
2355 CERROR("new_inode -fatal: rc %d\n", rc);
2362 int ll_show_options(struct seq_file *m, struct vfsmount *mnt)
2364 struct ll_sb_info *sbi = ll_s2sbi(mnt->mnt_sb);
2365 struct lustre_mount_data *lmd = sbi->ll_lmd;
2368 seq_printf(m, ",mds_sec=%s,oss_sec=%s",
2369 lmd->lmd_mds_security, lmd->lmd_oss_security);
2371 seq_printf(m, ",%s", sbi->ll_remote ? "remote" : "local");
2372 if (sbi->ll_remote && lmd)
2373 seq_printf(m, ",nllu=%u:%u", lmd->lmd_nllu, lmd->lmd_nllg);
2375 if (lmd && lmd->lmd_pag)
2376 seq_printf(m, ",pag");
2381 int ll_get_fid(struct obd_export *exp, struct lustre_id *idp,
2382 char *filename, struct lustre_id *ret)
2384 struct ptlrpc_request *request = NULL;
2385 struct mds_body *body;
2388 rc = md_getattr_lock(exp, idp, filename, strlen(filename) + 1,
2389 OBD_MD_FID, 0, &request);
2391 CDEBUG(D_INFO, "md_getattr_lock failed on %s: rc %d\n",
2396 body = lustre_msg_buf(request->rq_repmsg, 0, sizeof(*body));
2397 LASSERT(body != NULL);
2398 LASSERT_REPSWABBED(request, 0);
2401 ptlrpc_req_finished(request);
2405 int ll_flush_cred(struct inode *inode)
2407 struct ll_sb_info *sbi = ll_i2sbi(inode);
2410 /* XXX to avoid adding api, we simply use set_info() interface
2411 * to notify underlying obds. set_info() is more like a ioctl() now...
2413 if (sbi->ll_md_exp) {
2414 rc = obd_set_info(sbi->ll_md_exp,
2415 strlen("flush_cred"), "flush_cred",
2421 if (sbi->ll_dt_exp) {
2422 rc = obd_set_info(sbi->ll_dt_exp,
2423 strlen("flush_cred"), "flush_cred",