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 if (mds_security == NULL)
157 mds_security = "null";
159 err = obd_set_info(md_obd->obd_self_export, strlen("sec"), "sec",
160 strlen(mds_security), mds_security);
163 CERROR("LMV %s: failed to set security %s, err %d\n",
164 lmv, mds_security, err);
169 sec_flags = PTLRPC_SEC_FL_PAG;
170 err = obd_set_info(md_obd->obd_self_export,
171 strlen("sec_flags"), "sec_flags",
172 sizeof(sec_flags), &sec_flags);
174 OBD_FREE(data, sizeof(*data));
179 err = obd_connect(&md_conn, md_obd, &sbi->ll_sb_uuid, data,
180 OBD_OPT_REAL_CLIENT);
182 CERROR("An MDS (lmv %s) is performing recovery, of which this"
183 " client is not a part. Please wait for recovery to "
184 "complete, abort, or time out.\n", lmv);
187 CERROR("cannot connect to %s: rc = %d\n", lmv, err);
191 sbi->ll_md_exp = class_conn2export(&md_conn);
193 err = obd_statfs(md_obd, &osfs, jiffies - HZ);
195 GOTO(out_disconnect, err);
197 if (!osfs.os_bsize) {
198 CERROR("Invalid block size is detected.");
199 GOTO(out_disconnect, err);
202 sb->s_magic = LL_SUPER_MAGIC;
203 sb->s_blocksize = osfs.os_bsize;
204 sb->s_blocksize_bits = log2(osfs.os_bsize);
205 sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
207 /* in 2.6.x FS is not allowed to form s_dev */
208 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
212 devno = get_uuid2int((char *)sbi->ll_md_exp->exp_obd->obd_uuid.uuid,
213 strlen((char *)sbi->ll_md_exp->exp_obd->obd_uuid.uuid));
219 /* after statfs, we are supposed to have connected to MDSs,
220 * so it's ok to check remote flag returned.
222 valsize = sizeof(&sbi->ll_remote);
223 err = obd_get_info(sbi->ll_md_exp, strlen("remote_flag"), "remote_flag",
224 &valsize, &sbi->ll_remote);
226 CERROR("fail to obtain remote flag\n");
227 GOTO(out_disconnect, err);
232 obd_disconnect(sbi->ll_md_exp, 0);
237 static int lustre_connect_ost(struct super_block *sb, char *lov,
238 struct obd_connect_data *data,
239 char *oss_security, int async, int pag)
241 struct ll_sb_info *sbi = ll_s2sbi(sb);
242 struct lustre_handle dt_conn = {0, };
243 struct obd_device *obd = NULL;
244 unsigned long sec_flags;
247 obd = class_name2obd(lov);
249 CERROR("OSC %s: not setup or attached\n", lov);
250 GOTO(out, err = -EINVAL);
252 obd_set_info(obd->obd_self_export, strlen("async"), "async",
253 sizeof(async), &async);
255 if (oss_security == NULL)
256 oss_security = "null";
258 err = obd_set_info(obd->obd_self_export, strlen("sec"), "sec",
259 strlen(oss_security), oss_security);
261 CERROR("LOV %s: failed to set security %s, err %d\n",
262 lov, oss_security, err);
267 sec_flags = PTLRPC_SEC_FL_PAG;
268 err = obd_set_info(obd->obd_self_export,
269 strlen("sec_flags"), "sec_flags",
270 sizeof(sec_flags), &sec_flags);
272 OBD_FREE(data, sizeof(*data));
277 err = obd_connect(&dt_conn, obd, &sbi->ll_sb_uuid, data, 0);
279 CERROR("An OST (lov %s) is performing recovery, of which this"
280 " client is not a part. Please wait for recovery to "
281 "complete, abort, or time out.\n", lov);
284 CERROR("cannot connect to %s: rc = %d\n", lov, err);
287 sbi->ll_dt_exp = class_conn2export(&dt_conn);
289 err = lustre_init_dt_desc(sbi);
292 CWARN("init dt_desc error %d \n", err);
295 mdsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
296 obd_init_ea_size(sbi->ll_md_exp, mdsize, sbi->ll_dt_desc.ld_tgt_count *
297 sizeof(struct llog_cookie));
302 extern struct dentry_operations ll_d_ops;
304 static int lustre_init_root_inode(struct super_block *sb)
306 struct ll_sb_info *sbi = ll_s2sbi(sb);
307 struct ptlrpc_request *request = NULL;
308 struct inode *root = NULL;
313 err = md_getstatus(sbi->ll_md_exp, &sbi->ll_rootid);
315 CERROR("cannot mds_connect: rc = %d\n", err);
318 CDEBUG(D_SUPER, "rootid "DLID4"\n", OLID4(&sbi->ll_rootid));
320 sb->s_op = &lustre_super_operations;
322 /* make root inode */
323 err = md_getattr(sbi->ll_md_exp, &sbi->ll_rootid,
324 (OBD_MD_FLNOTOBD | OBD_MD_FLBLOCKS | OBD_MD_FID),
325 NULL, NULL, 0, 0, NULL, &request);
327 CERROR("md_getattr failed for root: rc = %d\n", err);
331 err = mdc_req2lustre_md(sbi->ll_md_exp, request, 0,
332 sbi->ll_dt_exp, &md);
334 CERROR("failed to understand root inode md: rc = %d\n", err);
335 ptlrpc_req_finished(request);
339 LASSERT(id_ino(&sbi->ll_rootid) != 0);
340 root = ll_iget(sb, id_ino(&sbi->ll_rootid), &md);
342 ptlrpc_req_finished(request);
344 if (root == NULL || is_bad_inode(root)) {
346 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
348 obd_free_memmd(sbi->ll_md_exp,
349 (struct lov_stripe_md**)&md.mea);
350 CERROR("lustre_lite: bad iget4 for root\n");
351 GOTO(out_root, err = -EBADF);
353 sb->s_root = d_alloc_root(root);
354 sb->s_root->d_op = &ll_d_ops;
362 int lustre_common_fill_super(struct super_block *sb, char *lmv, char *lov,
363 char *gkc, int async, char *mds_security,
364 char *oss_security, __u32 *nllu, int pag,
367 struct ll_sb_info *sbi = ll_s2sbi(sb);
368 struct obd_connect_data *data;
372 /*process the connect flags*/
373 if ((*remote & (OBD_CONNECT_LOCAL | OBD_CONNECT_REMOTE)) ==
374 (OBD_CONNECT_LOCAL | OBD_CONNECT_REMOTE)) {
375 CERROR("wrong remote flag "LPX64"\n", *remote);
379 OBD_ALLOC(data, sizeof(*data));
383 data->ocd_connect_flags |= *remote & (OBD_CONNECT_LOCAL |
385 memcpy(data->ocd_nllu, nllu, sizeof(data->ocd_nllu));
387 if (proc_lustre_fs_root) {
388 err = lprocfs_register_mountpoint(proc_lustre_fs_root,
391 CERROR("could not register mount in /proc/lustre");
395 err = lustre_connect_mds(sb, lmv, data, mds_security, async, pag);
400 err = lustre_connect_ost(sb, lov, data, oss_security, async, pag);
404 err = lustre_init_crypto(sb, gkc, data, async);
406 CERROR("Could not connect to GSS err %d\n", err);
410 err = lustre_init_root_inode(sb);
414 err = ll_close_thread_start(&sbi->ll_lcq);
416 CERROR("cannot start close thread: rc %d\n", err);
420 ll_gns_add_timer(sbi);
422 /* making vm readahead 0 for 2.4.x. In the case of 2.6.x,
423 backing dev info assigned to inode mapping is used for
424 determining maximal readahead. */
425 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0)) && \
426 !defined(KERNEL_HAS_AS_MAX_READAHEAD)
427 /* bug 2805 - set VM readahead to zero */
428 vm_max_readahead = vm_min_readahead = 0;
430 sb->s_flags |= MS_POSIXACL;
432 CWARN("Enabling PDIROPS\n");
433 sb->s_flags |= S_PDIROPS;
436 OBD_FREE(data, sizeof(*data));
442 lustre_destroy_crypto(sb);
444 obd_disconnect(sbi->ll_md_exp, 0);
447 OBD_FREE(data, sizeof(*data));
448 lprocfs_unregister_mountpoint(sbi);
452 void lustre_common_put_super(struct super_block *sb)
454 struct ll_sb_info *sbi = ll_s2sbi(sb);
455 struct hlist_node *tmp, *next;
458 ll_gns_del_timer(sbi);
459 ll_close_thread_stop(sbi->ll_lcq);
461 lustre_destroy_crypto(sb);
463 list_del(&sbi->ll_conn_chain);
464 obd_disconnect(sbi->ll_dt_exp, 0);
466 lprocfs_unregister_mountpoint(sbi);
467 if (sbi->ll_proc_root) {
468 lprocfs_remove(sbi->ll_proc_root);
469 sbi->ll_proc_root = NULL;
472 obd_disconnect(sbi->ll_md_exp, 0);
474 // We do this to get rid of orphaned dentries. That is not really trw.
475 hlist_for_each_safe(tmp, next, &sbi->ll_orphan_dentry_list) {
476 struct dentry *dentry = hlist_entry(tmp, struct dentry, d_hash);
477 CWARN("orphan dentry %.*s (%p->%p) at unmount\n",
478 dentry->d_name.len, dentry->d_name.name, dentry, next);
479 shrink_dcache_parent(dentry);
484 char *ll_read_opt(const char *opt, char *data)
490 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
491 if (strncmp(opt, data, strlen(opt)))
493 if ((value = strchr(data, '=')) == NULL)
497 OBD_ALLOC(retval, strlen(value) + 1);
499 CERROR("out of memory!\n");
503 memcpy(retval, value, strlen(value)+1);
504 CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
508 int ll_set_opt(const char *opt, char *data, int fl)
512 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
513 if (strncmp(opt, data, strlen(opt)))
519 void ll_options(char *options, char **lov, char **lmv, char **gkc,
520 char **mds_sec, char **oss_sec, int *async, int *flags)
523 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
524 char *opt_ptr = options;
534 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
535 for (this_char = strtok (options, ",");
537 this_char = strtok (NULL, ",")) {
539 while ((this_char = strsep (&opt_ptr, ",")) != NULL) {
541 CDEBUG(D_SUPER, "this_char %s\n", this_char);
542 if (!*lov && (*lov = ll_read_opt("osc", this_char)))
544 if (!*lmv && (*lmv = ll_read_opt("mdc", this_char)))
546 if (!*gkc && (*gkc = ll_read_opt("gkc", this_char)))
548 if (!strncmp(this_char, "lasync", strlen("lasync"))) {
552 if (!*mds_sec && (*mds_sec = ll_read_opt("mds_sec", this_char)))
554 if (!*oss_sec && (*oss_sec = ll_read_opt("oss_sec", this_char)))
556 if (!(*flags & LL_SBI_NOLCK) &&
557 ((*flags) = (*flags) |
558 ll_set_opt("nolock", this_char,
566 void ll_lli_init(struct ll_inode_info *lli)
568 sema_init(&lli->lli_open_sem, 1);
569 sema_init(&lli->lli_size_sem, 1);
571 lli->lli_size_pid = 0;
572 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
573 spin_lock_init(&lli->lli_lock);
574 INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
575 INIT_LIST_HEAD(&lli->lli_close_item);
576 lli->lli_inode_magic = LLI_INODE_MAGIC;
577 memset(&lli->lli_id, 0, sizeof(lli->lli_id));
578 sema_init(&lli->lli_och_sem, 1);
579 lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
580 lli->lli_mds_exec_och = NULL;
581 lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
582 lli->lli_open_fd_exec_count = 0;
583 lli->lli_audit_mask = AUDIT_OFF;
584 lli->lli_key_info = NULL;
585 init_waitqueue_head(&lli->lli_dirty_wait);
588 int ll_fill_super(struct super_block *sb, void *data, int silent)
590 struct ll_sb_info *sbi;
591 char *lov = NULL, *lmv = NULL, *gkc = NULL;
592 char *mds_sec = NULL;
593 char *oss_sec = NULL;
595 __u32 nllu[2] = { NOBODY_UID, NOBODY_GID };
596 __u64 remote_flag = 0;
599 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
601 sbi = lustre_init_sbi(sb);
605 sbi->ll_flags |= LL_SBI_READAHEAD;
606 ll_options(data, &lov, &lmv, &gkc, &mds_sec, &oss_sec,
607 &async, &sbi->ll_flags);
610 CERROR("no osc %p or no mdc %p\n", lov, lmv);
611 GOTO(out, err = -EINVAL);
614 err = lustre_common_fill_super(sb, lmv, lov, gkc, async, mds_sec,
615 oss_sec, nllu, 0, &remote_flag);
622 OBD_FREE(lmv, strlen(lmv) + 1);
624 OBD_FREE(lov, strlen(lov) + 1);
626 OBD_FREE(mds_sec, strlen(mds_sec) + 1);
628 OBD_FREE(oss_sec, strlen(oss_sec) + 1);
630 OBD_FREE(gkc, strlen(gkc) + 1);
633 } /* ll_read_super */
635 static int lustre_process_log(struct lustre_mount_data *lmd, char *profile,
636 struct config_llog_instance *cfg, int allow_recov)
638 struct lustre_cfg *lcfg = NULL;
639 struct lustre_cfg_bufs bufs;
640 struct portals_cfg pcfg;
641 char *peer = "MDS_PEER_UUID";
642 struct obd_device *obd;
643 struct lustre_handle md_conn = {0, };
644 struct obd_export *exp;
645 char *name = "mdc_dev";
647 struct obd_uuid lmv_uuid;
648 struct llog_ctxt *ctxt;
652 if (lmd_bad_magic(lmd))
655 generate_random_uuid(uuid);
656 class_uuid_unparse(uuid, &lmv_uuid);
658 if (lmd->lmd_local_nid) {
659 PCFG_INIT(pcfg, NAL_CMD_REGISTER_MYNID);
660 pcfg.pcfg_nal = lmd->lmd_nal;
661 pcfg.pcfg_nid = lmd->lmd_local_nid;
662 rc = libcfs_nal_cmd(&pcfg);
667 if (lmd->lmd_nal == SOCKNAL ||
668 lmd->lmd_nal == OPENIBNAL ||
669 lmd->lmd_nal == IIBNAL ||
670 lmd->lmd_nal == VIBNAL ||
671 lmd->lmd_nal == RANAL) {
672 PCFG_INIT(pcfg, NAL_CMD_ADD_PEER);
673 pcfg.pcfg_nal = lmd->lmd_nal;
674 pcfg.pcfg_nid = lmd->lmd_server_nid;
675 pcfg.pcfg_id = lmd->lmd_server_ipaddr;
676 pcfg.pcfg_misc = lmd->lmd_port;
677 rc = libcfs_nal_cmd(&pcfg);
681 lustre_cfg_bufs_reset(&bufs, name);
682 lustre_cfg_bufs_set_string(&bufs, 1, peer);
684 lcfg = lustre_cfg_new(LCFG_ADD_UUID, &bufs);
685 lcfg->lcfg_nal = lmd->lmd_nal;
686 lcfg->lcfg_nid = lmd->lmd_server_nid;
687 LASSERT(lcfg->lcfg_nal);
688 LASSERT(lcfg->lcfg_nid);
689 err = class_process_config(lcfg);
690 lustre_cfg_free(lcfg);
692 GOTO(out_del_conn, err);
694 lustre_cfg_bufs_reset(&bufs, name);
695 lustre_cfg_bufs_set_string(&bufs, 1, OBD_MDC_DEVICENAME);
696 lustre_cfg_bufs_set_string(&bufs, 2, (char *)lmv_uuid.uuid);
698 lcfg = lustre_cfg_new(LCFG_ATTACH, &bufs);
699 err = class_process_config(lcfg);
700 lustre_cfg_free(lcfg);
702 GOTO(out_del_uuid, err);
704 lustre_cfg_bufs_reset(&bufs, name);
705 lustre_cfg_bufs_set_string(&bufs, 1, lmd->lmd_mds);
706 lustre_cfg_bufs_set_string(&bufs, 2, peer);
708 lcfg = lustre_cfg_new(LCFG_SETUP, &bufs);
709 err = class_process_config(lcfg);
710 lustre_cfg_free(lcfg);
712 GOTO(out_detach, err);
714 obd = class_name2obd(name);
716 GOTO(out_cleanup, rc = -EINVAL);
718 rc = obd_set_info(obd->obd_self_export, strlen("sec"), "sec",
719 strlen(lmd->lmd_mds_security), lmd->lmd_mds_security);
721 GOTO(out_cleanup, rc);
724 unsigned long sec_flags = PTLRPC_SEC_FL_PAG;
725 rc = obd_set_info(obd->obd_self_export,
726 strlen("sec_flags"), "sec_flags",
727 sizeof(sec_flags), &sec_flags);
729 GOTO(out_cleanup, rc);
732 /* Disable initial recovery on this import */
733 rc = obd_set_info(obd->obd_self_export,
734 strlen("initial_recov"), "initial_recov",
735 sizeof(allow_recov), &allow_recov);
737 GOTO(out_cleanup, rc);
739 rc = obd_connect(&md_conn, obd, &lmv_uuid, NULL, OBD_OPT_REAL_CLIENT);
741 CERROR("cannot connect to %s: rc = %d\n", lmd->lmd_mds, rc);
742 GOTO(out_cleanup, rc);
745 exp = class_conn2export(&md_conn);
747 ctxt = llog_get_context(&exp->exp_obd->obd_llogs,LLOG_CONFIG_REPL_CTXT);
748 rc = class_config_process_llog(ctxt, profile, cfg);
750 CERROR("class_config_process_llog failed: rc = %d\n", rc);
752 err = obd_disconnect(exp, 0);
756 lustre_cfg_bufs_reset(&bufs, name);
757 lcfg = lustre_cfg_new(LCFG_CLEANUP, &bufs);
758 err = class_process_config(lcfg);
759 lustre_cfg_free(lcfg);
763 lustre_cfg_bufs_reset(&bufs, name);
764 lcfg = lustre_cfg_new(LCFG_DETACH, &bufs);
765 err = class_process_config(lcfg);
766 lustre_cfg_free(lcfg);
771 lustre_cfg_bufs_reset(&bufs, name);
772 lustre_cfg_bufs_set_string(&bufs, 1, peer);
773 lcfg = lustre_cfg_new(LCFG_DEL_UUID, &bufs);
774 err = class_process_config(lcfg);
775 lustre_cfg_free(lcfg);
778 if (lmd->lmd_nal == SOCKNAL ||
779 lmd->lmd_nal == OPENIBNAL ||
780 lmd->lmd_nal == IIBNAL ||
781 lmd->lmd_nal == VIBNAL ||
782 lmd->lmd_nal == RANAL) {
785 PCFG_INIT(pcfg, NAL_CMD_DEL_PEER);
786 pcfg.pcfg_nal = lmd->lmd_nal;
787 pcfg.pcfg_nid = lmd->lmd_server_nid;
788 pcfg.pcfg_flags = 1; /* single_share */
789 err2 = libcfs_nal_cmd(&pcfg);
802 static void lustre_manual_cleanup(struct ll_sb_info *sbi)
804 struct lustre_cfg *lcfg;
805 struct lustre_cfg_bufs bufs;
806 struct obd_device *obd;
809 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) != NULL) {
812 lustre_cfg_bufs_reset(&bufs, obd->obd_name);
813 lcfg = lustre_cfg_new(LCFG_CLEANUP, &bufs);
814 err = class_process_config(lcfg);
816 CERROR("cleanup failed: %s\n", obd->obd_name);
820 lcfg->lcfg_command = LCFG_DETACH;
821 err = class_process_config(lcfg);
822 lustre_cfg_free(lcfg);
824 CERROR("detach failed: %s\n", obd->obd_name);
829 if (sbi->ll_lmd != NULL)
830 class_del_profile(sbi->ll_lmd->lmd_profile);
833 static int lustre_process_profile(struct super_block *sb,
834 struct lustre_mount_data *lmd,
835 char **lov, char **lmv, char **gkc)
837 struct ll_sb_info *sbi = ll_s2sbi(sb);
838 struct config_llog_instance cfg;
839 struct lustre_profile *lprof;
843 if (!lmd->lmd_profile)
846 if (lmd->lmd_mds[0] == '\0') {
847 CERROR("no mds name\n");
848 GOTO(out, err = -EINVAL);
850 lmd->lmd_mds_security[sizeof(lmd->lmd_mds_security) - 1] = 0;
851 lmd->lmd_oss_security[sizeof(lmd->lmd_oss_security) - 1] = 0;
853 OBD_ALLOC(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
854 if (sbi->ll_lmd == NULL)
855 GOTO(out, err = -ENOMEM);
856 memcpy(sbi->ll_lmd, lmd, sizeof(*lmd));
858 /* generate a string unique to this super, let's try the address of the
860 len = (sizeof(sb) * 2) + 1;
861 OBD_ALLOC(sbi->ll_instance, len);
862 if (sbi->ll_instance == NULL)
863 GOTO(out, err = -ENOMEM);
864 sprintf(sbi->ll_instance, "%p", sb);
866 cfg.cfg_instance = sbi->ll_instance;
867 cfg.cfg_uuid = sbi->ll_sb_uuid;
868 cfg.cfg_local_nid = lmd->lmd_local_nid;
869 err = lustre_process_log(lmd, lmd->lmd_profile, &cfg, 0);
871 CERROR("Unable to process log: %s\n", lmd->lmd_profile);
875 lprof = class_get_profile(lmd->lmd_profile);
877 CERROR("No profile found: %s\n", lmd->lmd_profile);
878 GOTO(out, err = -EINVAL);
881 OBD_ALLOC(*lov, strlen(lprof->lp_lov) +
882 strlen(sbi->ll_instance) + 2);
884 GOTO(out, err = -ENOMEM);
886 sprintf(*lov, "%s-%s", lprof->lp_lov, sbi->ll_instance);
888 OBD_ALLOC(*lmv, strlen(lprof->lp_lmv) +
889 strlen(sbi->ll_instance) + 2);
891 GOTO(out_free_lov, err = -ENOMEM);
893 sprintf(*lmv, "%s-%s", lprof->lp_lmv, sbi->ll_instance);
896 OBD_ALLOC(*gkc, strlen(lprof->lp_gkc) +
897 strlen(sbi->ll_instance) + 2);
899 GOTO(out_free_lmv, err = -ENOMEM);
901 sprintf(*gkc, "%s-%s", lprof->lp_gkc, sbi->ll_instance);
906 OBD_FREE(*lmv, strlen(lprof->lp_lmv) +
907 strlen(sbi->ll_instance) + 2);
909 OBD_FREE(*lov, strlen(lprof->lp_lov) +
910 strlen(sbi->ll_instance) + 2);
915 static int lustre_clean_profile(struct ll_sb_info *sbi, int force_umount)
917 struct lustre_mount_data *lmd = sbi->ll_lmd;
918 struct config_llog_instance cfg;
926 len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean") + 1;
929 CERROR("force umount, doing manual cleanup\n");
930 lustre_manual_cleanup(sbi);
934 if (sbi->ll_instance != NULL) {
935 cfg.cfg_instance = sbi->ll_instance;
936 cfg.cfg_uuid = sbi->ll_sb_uuid;
938 OBD_ALLOC(cl_prof, len);
940 CERROR("can't allocate memory, "
941 "skipping processing cleanup profile.\n");
942 GOTO(free_lmd, err = -ENOMEM);
945 sprintf(cl_prof, "%s-clean", lmd->lmd_profile);
946 err = lustre_process_log(lmd, cl_prof, &cfg, 0);
948 CERROR("Unable to process log: %s\n", cl_prof);
949 lustre_manual_cleanup(sbi);
951 OBD_FREE(cl_prof, len);
955 if (sbi->ll_instance)
956 OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance) + 1);
957 OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
961 int lustre_fill_super(struct super_block *sb, void *data, int silent)
963 struct lustre_mount_data * lmd = data;
964 char *lov = NULL, *lmv = NULL, *gkc = NULL;
965 struct ll_sb_info *sbi;
969 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
970 if (lmd_bad_magic(lmd))
973 sbi = lustre_init_sbi(sb);
977 sbi->ll_flags |= LL_SBI_READAHEAD;
979 err = lustre_process_profile(sb, lmd, &lov, &lmv, &gkc);
981 CERROR("Can not process the profile err %d \n", err);
985 CERROR("no osc %p or no mdc %p \n", lov, lmv);
986 GOTO(out_free, err = -EINVAL);
989 err = lustre_common_fill_super(sb, lmv, lov, gkc, lmd->lmd_async,
990 lmd->lmd_mds_security,
991 lmd->lmd_oss_security,
992 &lmd->lmd_nllu, lmd->lmd_pag,
993 &lmd->lmd_remote_flag);
1001 OBD_FREE(lmv, strlen(lmv) + 1);
1003 OBD_FREE(lov, strlen(lov) + 1);
1005 OBD_FREE(gkc, strlen(gkc) + 1);
1009 lustre_clean_profile(sbi, 0);
1010 lustre_free_sbi(sb);
1013 } /* lustre_fill_super */
1015 void lustre_put_super(struct super_block *sb)
1017 struct obd_device *obd;
1018 struct ll_sb_info *sbi = ll_s2sbi(sb);
1019 int force_umount = 0;
1022 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1023 obd = class_exp2obd(sbi->ll_md_exp);
1025 force_umount = obd->obd_no_recov;
1028 lustre_common_put_super(sb);
1029 lustre_clean_profile(sbi, force_umount);
1030 lustre_free_sbi(sb);
1033 } /* lustre_put_super */
1035 int ll_process_config_update(struct ll_sb_info *sbi, int clean)
1037 struct lustre_mount_data *lmd = sbi->ll_lmd;
1038 char *profile = lmd->lmd_profile, *name = NULL;
1039 struct config_llog_instance cfg;
1040 int rc, namelen = 0, version;
1041 struct llog_ctxt *ctxt;
1044 if (profile == NULL)
1047 CERROR("Client not mounted with zero-conf; cannot "
1048 "process update log.\n");
1052 cfg.cfg_instance = sbi->ll_instance;
1053 cfg.cfg_uuid = sbi->ll_sb_uuid;
1054 cfg.cfg_local_nid = lmd->lmd_local_nid;
1056 namelen = strlen(profile) + 20; /* -clean-######### */
1057 OBD_ALLOC(name, namelen);
1062 version = sbi->ll_config_version - 1;
1063 sprintf(name, "%s-clean-%d", profile, version);
1065 version = sbi->ll_config_version + 1;
1066 sprintf(name, "%s-%d", profile, version);
1069 CWARN("Applying configuration log %s\n", name);
1071 ctxt = llog_get_context(&sbi->ll_md_exp->exp_obd->obd_llogs,
1072 LLOG_CONFIG_REPL_CTXT);
1073 rc = class_config_process_llog(ctxt, name, &cfg);
1075 sbi->ll_config_version = version;
1076 CWARN("Finished applying configuration log %s: %d\n", name, rc);
1078 if (rc == 0 && clean == 0) {
1079 struct lov_desc desc;
1083 valsize = sizeof(desc);
1084 rc = obd_get_info(sbi->ll_dt_exp, strlen("lovdesc") + 1,
1085 "lovdesc", &valsize, &desc);
1087 rc = obd_init_ea_size(sbi->ll_md_exp,
1088 obd_size_diskmd(sbi->ll_dt_exp, NULL),
1089 (desc.ld_tgt_count *
1090 sizeof(struct llog_cookie)));
1092 OBD_FREE(name, namelen);
1096 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1098 struct inode *inode = NULL;
1100 /* NOTE: we depend on atomic igrab() -bzzz */
1101 lock_res_and_lock(lock);
1102 if (lock->l_ast_data) {
1103 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1104 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1105 inode = igrab(lock->l_ast_data);
1108 do_gettimeofday(&now);
1109 inode = lock->l_ast_data;
1110 LDLM_ERROR(lock, "granted at %lu.%lu, now %lu.%lu",
1111 lock->l_enqueued_time.tv_sec,
1112 lock->l_enqueued_time.tv_usec,
1113 now.tv_sec, now.tv_usec);
1114 CDEBUG(inode->i_state & I_FREEING ? D_INFO : D_WARNING,
1115 "l_ast_data %p is bogus: magic %0x8\n",
1116 lock->l_ast_data, lli->lli_inode_magic);
1117 CDEBUG(D_ERROR, "i_state = 0x%lx, l_ast_data %p is bogus: magic %0x8\n",
1118 inode->i_state, lock->l_ast_data, lli->lli_inode_magic);
1120 unlock_res_and_lock(lock);
1124 unlock_res_and_lock(lock);
1128 int null_if_equal(struct ldlm_lock *lock, void *data)
1130 if (data == lock->l_ast_data) {
1131 lock->l_ast_data = NULL;
1133 if (lock->l_req_mode != lock->l_granted_mode)
1134 LDLM_ERROR(lock,"clearing inode with ungranted lock\n");
1137 return LDLM_ITER_CONTINUE;
1140 static void remote_acl_free(struct remote_acl *racl);
1142 void ll_clear_inode(struct inode *inode)
1144 struct lustre_id id;
1145 struct ll_inode_info *lli = ll_i2info(inode);
1146 struct ll_sb_info *sbi = ll_i2sbi(inode);
1149 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1150 inode->i_generation, inode);
1152 LASSERT(ll_is_inode_dirty(inode) == 0);
1153 ll_inode2id(&id, inode);
1155 clear_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &(ll_i2info(inode)->lli_flags));
1156 md_change_cbdata(sbi->ll_md_exp, &id, null_if_equal, inode);
1158 LASSERT(!lli->lli_open_fd_write_count);
1159 LASSERT(!lli->lli_open_fd_read_count);
1160 LASSERT(!lli->lli_open_fd_exec_count);
1161 if (lli->lli_mds_write_och)
1162 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_WRITE);
1163 if (lli->lli_mds_exec_och)
1164 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_EXEC);
1165 if (lli->lli_mds_read_och)
1166 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_READ);
1168 obd_change_cbdata(sbi->ll_dt_exp, lli->lli_smd,
1169 null_if_equal, inode);
1172 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1173 lli->lli_smd = NULL;
1177 obd_free_memmd(sbi->ll_md_exp,
1178 (struct lov_stripe_md **) &lli->lli_mea);
1179 lli->lli_mea = NULL;
1181 ll_crypto_destroy_inode_key(inode);
1182 if (lli->lli_symlink_name) {
1183 OBD_FREE(lli->lli_symlink_name,
1184 strlen(lli->lli_symlink_name) + 1);
1185 lli->lli_symlink_name = NULL;
1188 if (lli->lli_posix_acl) {
1189 LASSERT(lli->lli_remote_acl == NULL);
1190 posix_acl_release(lli->lli_posix_acl);
1191 lli->lli_posix_acl = NULL;
1194 if (lli->lli_remote_acl) {
1195 LASSERT(lli->lli_posix_acl == NULL);
1196 remote_acl_free(lli->lli_remote_acl);
1197 lli->lli_remote_acl = NULL;
1200 if (lli->lli_trunc_capa) {
1201 OBD_FREE(lli->lli_trunc_capa, sizeof(struct lustre_capa));
1202 lli->lli_trunc_capa = NULL;
1205 lli->lli_inode_magic = LLI_INODE_DEAD;
1209 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1210 * object(s) determine the file size and mtime. Otherwise, the MDS will
1211 * keep these values until such a time that objects are allocated for it.
1212 * We do the MDS operations first, as it is checking permissions for us.
1213 * We don't to the MDS RPC if there is nothing that we want to store there,
1214 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1215 * going to do an RPC anyways.
1217 * If we are doing a truncate, we will send the mtime and ctime updates
1218 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1219 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1222 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1224 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1225 struct ll_inode_info *lli = ll_i2info(inode);
1226 struct ll_sb_info *sbi = ll_i2sbi(inode);
1227 struct ptlrpc_request *request = NULL;
1228 struct mdc_op_data *op_data;
1229 struct lustre_capa *trunc_capa = NULL;
1230 int ia_valid = attr->ia_valid;
1234 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu\n", inode->i_ino);
1235 lprocfs_counter_incr(ll_i2sbi(inode)->ll_stats, LPROC_LL_SETATTR);
1237 if (ia_valid & ATTR_SIZE) {
1238 if (attr->ia_size > ll_file_maxbytes(inode)) {
1239 CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1240 attr->ia_size, ll_file_maxbytes(inode));
1244 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1247 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1248 if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1249 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1253 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1254 if (attr->ia_valid & ATTR_CTIME) {
1255 attr->ia_ctime = CURRENT_TIME;
1256 attr->ia_valid |= ATTR_CTIME_SET;
1258 if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1259 attr->ia_atime = CURRENT_TIME;
1260 attr->ia_valid |= ATTR_ATIME_SET;
1262 if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1263 attr->ia_mtime = CURRENT_TIME;
1264 attr->ia_valid |= ATTR_MTIME_SET;
1267 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1268 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1269 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1270 LTIME_S(CURRENT_TIME));
1272 /* If only OST attributes being set on objects, don't do MDS RPC.
1273 * In that case, we need to check permissions and update the local
1274 * inode ourselves so we can call obdo_from_inode() always. */
1275 if (ia_valid & (lsm ? ~(ATTR_SIZE | ATTR_FROM_OPEN /*| ATTR_RAW*/) : ~0)) {
1276 struct lustre_md md;
1280 OBD_ALLOC(op_data, sizeof(*op_data));
1281 if (op_data == NULL)
1283 ll_inode2mdc_data(op_data, inode, (OBD_MD_FLID | OBD_MD_MEA));
1285 if (ia_valid & (ATTR_UID | ATTR_GID | ATTR_MODE)) {
1286 rc = ll_crypto_get_mac(inode, attr, NULL, 0, &key,
1289 CERROR("can not get right mac, rc=%d\n", rc);
1290 if (key && key_size)
1291 OBD_FREE(key, key_size);
1295 rc = md_setattr(sbi->ll_md_exp, op_data,
1296 attr, key, key_size, NULL, 0, NULL,
1298 OBD_FREE(op_data, sizeof(*op_data));
1300 if (key && key_size)
1301 OBD_FREE(key, key_size);
1303 ptlrpc_req_finished(request);
1304 if (rc != -EPERM && rc != -EACCES)
1305 CERROR("md_setattr fails: rc = %d\n", rc);
1308 rc = mdc_req2lustre_md(sbi->ll_md_exp, request, 0,
1309 sbi->ll_dt_exp, &md);
1311 ptlrpc_req_finished(request);
1315 if (attr->ia_valid & ATTR_SIZE) {
1316 /* XXX: hack for truncate capa */
1317 rc = mdc_req2lustre_capa(request, 0, &trunc_capa);
1319 ptlrpc_req_finished(request);
1323 spin_lock(&lli->lli_lock);
1325 if (lli->lli_trunc_capa)
1326 OBD_FREE(lli->lli_trunc_capa,
1327 sizeof(*trunc_capa));
1328 lli->lli_trunc_capa = trunc_capa;
1329 DEBUG_CAPA(D_INFO, trunc_capa, "truncate");
1331 spin_unlock(&lli->lli_lock);
1334 /* We call inode_setattr to adjust timestamps, but we first
1335 * clear ATTR_SIZE to avoid invoking vmtruncate.
1337 * NB: ATTR_SIZE will only be set at this point if the size
1338 * resides on the MDS, ie, this file has no objects. */
1339 attr->ia_valid &= ~ATTR_SIZE;
1342 * assigning inode_setattr() to @err to disable warning that
1343 * function's result should be checked by by caller. error is
1344 * impossible here, as vmtruncate() control path is disabled.
1346 err = inode_setattr(inode, attr);
1347 ll_update_inode(inode, &md);
1348 ptlrpc_req_finished(request);
1350 if (!lsm || !S_ISREG(inode->i_mode)) {
1351 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1355 /* The OST doesn't check permissions, but the alternative is
1356 * a gratuitous RPC to the MDS. We already rely on the client
1357 * to do read/write/truncate permission checks, so is mtime OK?
1359 if (ia_valid & (ATTR_MTIME | ATTR_ATIME)) {
1360 /* from sys_utime() */
1361 if (!(ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET))) {
1362 if (current->fsuid != inode->i_uid &&
1363 (rc = ll_permission(inode, MAY_WRITE, NULL)) != 0)
1366 /* from inode_change_ok() */
1367 if (current->fsuid != inode->i_uid &&
1368 !capable(CAP_FOWNER))
1374 attr->ia_valid &= ~ATTR_SIZE;
1376 /* won't invoke vmtruncate, as we already cleared ATTR_SIZE */
1377 err = inode_setattr(inode, attr);
1379 * assigning inode_setattr() to @err to disable warning that
1380 * function's result should be checked by by caller. error is
1381 * impossible here, as vmtruncate() control path is disabled.
1385 /* We really need to get our PW lock before we change inode->i_size.
1386 * If we don't we can race with other i_size updaters on our node, like
1387 * ll_file_read. We can also race with i_size propogation to other
1388 * nodes through dirtying and writeback of final cached pages. This
1389 * last one is especially bad for racing o_append users on other
1391 if (ia_valid & ATTR_SIZE) {
1392 ldlm_policy_data_t policy = { .l_extent = {attr->ia_size,
1394 struct lustre_handle lockh = { 0 };
1395 int err, ast_flags = 0;
1396 /* XXX when we fix the AST intents to pass the discard-range
1397 * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
1399 if (attr->ia_size == 0)
1400 ast_flags = LDLM_AST_DISCARD_DATA;
1402 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy, &lockh,
1403 ast_flags, &ll_i2sbi(inode)->ll_seek_stime);
1408 down(&lli->lli_size_sem);
1409 lli->lli_size_pid = current->pid;
1410 rc = vmtruncate(inode, attr->ia_size);
1412 LASSERT(atomic_read(&lli->lli_size_sem.count) <= 0);
1413 lli->lli_size_pid = 0;
1414 up(&lli->lli_size_sem);
1417 err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
1419 CERROR("ll_extent_unlock failed: %d\n", err);
1423 } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET | ATTR_UID | ATTR_GID)) {
1424 struct obdo *oa = NULL;
1426 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1427 inode->i_ino, LTIME_S(attr->ia_mtime));
1433 oa->o_id = lsm->lsm_object_id;
1434 oa->o_gr = lsm->lsm_object_gr;
1435 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1437 /* adding uid and gid, needed for quota */
1438 if (ia_valid & ATTR_UID) {
1439 oa->o_uid = inode->i_uid;
1440 oa->o_valid |= OBD_MD_FLUID;
1443 if (ia_valid & ATTR_GID) {
1444 oa->o_gid = inode->i_gid;
1445 oa->o_valid |= OBD_MD_FLGID;
1448 *(obdo_id(oa)) = lli->lli_id;
1449 oa->o_valid |= OBD_MD_FLIFID;
1451 obdo_from_inode(oa, inode, OBD_MD_FLTYPE | OBD_MD_FLATIME |
1452 OBD_MD_FLMTIME | OBD_MD_FLCTIME);
1453 rc = obd_setattr(sbi->ll_dt_exp, oa, lsm, NULL, NULL);
1456 CERROR("obd_setattr fails: rc = %d\n", rc);
1462 int ll_setattr(struct dentry *de, struct iattr *attr)
1464 LASSERT(de->d_inode);
1465 return ll_setattr_raw(de->d_inode, attr);
1468 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1469 unsigned long max_age)
1471 struct ll_sb_info *sbi = ll_s2sbi(sb);
1472 struct obd_statfs obd_osfs;
1476 rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age);
1478 CERROR("obd_statfs fails: rc = %d\n", rc);
1482 osfs->os_type = sb->s_magic;
1484 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1485 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1487 rc = obd_statfs(class_exp2obd(sbi->ll_dt_exp), &obd_osfs, max_age);
1489 CERROR("obd_statfs fails: rc = %d\n", rc);
1493 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1494 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1497 osfs->os_blocks = obd_osfs.os_blocks;
1498 osfs->os_bfree = obd_osfs.os_bfree;
1499 osfs->os_bavail = obd_osfs.os_bavail;
1501 /* If we don't have as many objects free on the OST as inodes
1502 * on the MDS, we reduce the total number of inodes to
1503 * compensate, so that the "inodes in use" number is correct.
1505 if (obd_osfs.os_ffree < osfs->os_ffree) {
1506 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1508 osfs->os_ffree = obd_osfs.os_ffree;
1514 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1516 struct obd_statfs osfs;
1519 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p\n", sb);
1520 lprocfs_counter_incr(ll_s2sbi(sb)->ll_stats, LPROC_LL_STAFS);
1522 /* For now we will always get up-to-date statfs values, but in the
1523 * future we may allow some amount of caching on the client (e.g.
1524 * from QOS or lprocfs updates). */
1525 rc = ll_statfs_internal(sb, &osfs, jiffies - 1);
1529 statfs_unpack(sfs, &osfs);
1531 if (sizeof(sfs->f_blocks) == 4) {
1532 while (osfs.os_blocks > ~0UL) {
1535 osfs.os_blocks >>= 1;
1536 osfs.os_bfree >>= 1;
1537 osfs.os_bavail >>= 1;
1541 sfs->f_blocks = osfs.os_blocks;
1542 sfs->f_bfree = osfs.os_bfree;
1543 sfs->f_bavail = osfs.os_bavail;
1549 /********************************
1551 ********************************/
1553 static struct remote_acl *remote_acl_alloc(void)
1555 struct remote_acl *racl;
1558 OBD_ALLOC(racl, sizeof(*racl));
1562 spin_lock_init(&racl->ra_lock);
1563 init_MUTEX(&racl->ra_update_sem);
1565 for (i = 0; i < REMOTE_ACL_HASHSIZE; i++)
1566 INIT_LIST_HEAD(&racl->ra_perm_cache[i]);
1572 * caller should guarantee no race here.
1574 static void remote_perm_flush_xperms(struct lustre_remote_perm *perm)
1576 struct remote_perm_setxid *xperm;
1578 while (!list_empty(&perm->lrp_setxid_perms)) {
1579 xperm = list_entry(perm->lrp_setxid_perms.next,
1580 struct remote_perm_setxid,
1582 list_del(&xperm->list);
1583 OBD_FREE(xperm, sizeof(*xperm));
1588 * caller should guarantee no race here.
1590 static void remote_acl_flush(struct remote_acl *racl)
1592 struct list_head *head;
1593 struct lustre_remote_perm *perm, *tmp;
1596 for (i = 0; i < REMOTE_ACL_HASHSIZE; i++) {
1597 head = &racl->ra_perm_cache[i];
1599 list_for_each_entry_safe(perm, tmp, head, lrp_list) {
1600 remote_perm_flush_xperms(perm);
1601 list_del(&perm->lrp_list);
1602 OBD_FREE(perm, sizeof(*perm));
1607 static void remote_acl_free(struct remote_acl *racl)
1612 down(&racl->ra_update_sem);
1613 spin_lock(&racl->ra_lock);
1614 remote_acl_flush(racl);
1615 spin_unlock(&racl->ra_lock);
1616 up(&racl->ra_update_sem);
1618 OBD_FREE(racl, sizeof(*racl));
1621 static inline int remote_acl_hashfunc(__u32 id)
1623 return (id & (REMOTE_ACL_HASHSIZE - 1));
1627 int __remote_acl_check(struct remote_acl *racl, unsigned int *perm)
1629 struct list_head *head;
1630 struct lustre_remote_perm *lperm;
1631 struct remote_perm_setxid *xperm;
1632 int found = 0, rc = -ENOENT;
1635 head = &racl->ra_perm_cache[remote_acl_hashfunc(current->uid)];
1636 spin_lock(&racl->ra_lock);
1638 list_for_each_entry(lperm, head, lrp_list) {
1639 if (lperm->lrp_auth_uid == current->uid) {
1648 if (lperm->lrp_auth_uid == current->fsuid &&
1649 lperm->lrp_auth_gid == current->fsgid) {
1650 if (lperm->lrp_valid) {
1651 *perm = lperm->lrp_perm;
1655 } else if ((!lperm->lrp_setuid &&
1656 lperm->lrp_auth_uid != current->fsuid) ||
1657 (!lperm->lrp_setgid &&
1658 lperm->lrp_auth_gid != current->fsgid)) {
1664 list_for_each_entry(xperm, &lperm->lrp_setxid_perms, list) {
1665 if (xperm->uid == current->fsuid &&
1666 xperm->gid == current->fsgid) {
1667 *perm = xperm->perm;
1674 spin_unlock(&racl->ra_lock);
1679 int __remote_acl_update(struct remote_acl *racl,
1680 struct mds_remote_perm *mperm,
1681 struct lustre_remote_perm *lperm,
1682 struct remote_perm_setxid *xperm)
1684 struct list_head *head;
1685 struct lustre_remote_perm *lp;
1686 struct remote_perm_setxid *xp;
1687 int found = 0, setuid = 0, setgid = 0;
1692 LASSERT(current->uid == mperm->mrp_auth_uid);
1694 if (current->fsuid != mperm->mrp_auth_uid)
1696 if (current->fsgid != mperm->mrp_auth_gid)
1699 head = &racl->ra_perm_cache[remote_acl_hashfunc(current->uid)];
1700 spin_lock(&racl->ra_lock);
1702 list_for_each_entry(lp, head, lrp_list) {
1703 if (lp->lrp_auth_uid == current->uid) {
1710 OBD_FREE(lperm, sizeof(*lperm));
1712 if (!lp->lrp_valid && !setuid && !setgid) {
1713 lp->lrp_perm = mperm->mrp_perm;
1717 /* sanity check for changes of setxid rules */
1718 if ((lp->lrp_setuid != 0) != (mperm->mrp_allow_setuid != 0)) {
1719 CWARN("setuid changes: %d => %d\n",
1720 (lp->lrp_setuid != 0),
1721 (mperm->mrp_allow_setuid != 0));
1722 lp->lrp_setuid = (mperm->mrp_allow_setuid != 0);
1725 if ((lp->lrp_setgid != 0) != (mperm->mrp_allow_setgid != 0)) {
1726 CWARN("setgid changes: %d => %d\n",
1727 (lp->lrp_setgid != 0),
1728 (mperm->mrp_allow_setgid != 0));
1729 lp->lrp_setgid = (mperm->mrp_allow_setgid != 0);
1732 if (!lp->lrp_setuid && !lp->lrp_setgid &&
1733 !list_empty(&lp->lrp_setxid_perms)) {
1734 remote_perm_flush_xperms(lp);
1737 /* initialize lperm and linked into hashtable
1739 INIT_LIST_HEAD(&lperm->lrp_setxid_perms);
1740 lperm->lrp_auth_uid = mperm->mrp_auth_uid;
1741 lperm->lrp_auth_gid = mperm->mrp_auth_gid;
1742 lperm->lrp_setuid = (mperm->mrp_allow_setuid != 0);
1743 lperm->lrp_setgid = (mperm->mrp_allow_setgid != 0);
1744 list_add(&lperm->lrp_list, head);
1746 if (!setuid && !setgid) {
1747 /* in this case, i'm the authenticated user,
1748 * and mrp_perm is for me.
1750 lperm->lrp_perm = mperm->mrp_perm;
1751 lperm->lrp_valid = 1;
1752 spin_unlock(&racl->ra_lock);
1755 OBD_FREE(xperm, sizeof(*xperm));
1763 LASSERT(lp->lrp_setuid || lp->lrp_setgid ||
1764 list_empty(&lp->lrp_setxid_perms));
1766 /* if no xperm supplied, we are all done here */
1768 spin_unlock(&racl->ra_lock);
1772 /* whether we allow setuid/setgid */
1773 if ((!lp->lrp_setuid && setuid) || (!lp->lrp_setgid && setgid)) {
1774 OBD_FREE(xperm, sizeof(*xperm));
1775 spin_unlock(&racl->ra_lock);
1779 /* traverse xperm list */
1780 list_for_each_entry(xp, &lp->lrp_setxid_perms, list) {
1781 if (xp->uid == current->fsuid &&
1782 xp->gid == current->fsgid) {
1783 if (xp->perm != mperm->mrp_perm) {
1784 /* actually this should not happen */
1785 CWARN("perm changed: %o => %o\n",
1786 xp->perm, mperm->mrp_perm);
1787 xp->perm = mperm->mrp_perm;
1789 OBD_FREE(xperm, sizeof(*xperm));
1790 spin_unlock(&racl->ra_lock);
1795 /* finally insert this xperm */
1796 xperm->uid = current->fsuid;
1797 xperm->gid = current->fsgid;
1798 xperm->perm = mperm->mrp_perm;
1799 list_add(&xperm->list, &lp->lrp_setxid_perms);
1801 spin_unlock(&racl->ra_lock);
1806 * remote_acl semaphore must be held by caller
1809 int remote_acl_update_locked(struct remote_acl *racl,
1810 struct mds_remote_perm *mperm)
1812 struct lustre_remote_perm *lperm;
1813 struct remote_perm_setxid *xperm;
1814 int setuid = 0, setgid = 0;
1818 if (current->uid != mperm->mrp_auth_uid) {
1819 CERROR("current uid %u while authenticated as %u\n",
1820 current->uid, mperm->mrp_auth_uid);
1824 if (current->fsuid != mperm->mrp_auth_uid)
1826 if (current->fsgid == mperm->mrp_auth_gid)
1829 OBD_ALLOC(lperm, sizeof(*lperm));
1833 if ((setuid || setgid) &&
1834 !(setuid && !mperm->mrp_allow_setuid) &&
1835 !(setgid && !mperm->mrp_allow_setgid)) {
1836 OBD_ALLOC(xperm, sizeof(*xperm));
1838 OBD_FREE(lperm, sizeof(*lperm));
1844 return __remote_acl_update(racl, mperm, lperm, xperm);
1848 * return -EACCES at any error cases
1850 int ll_remote_acl_permission(struct inode *inode, int mode)
1852 struct ll_sb_info *sbi = ll_i2sbi(inode);
1853 struct remote_acl *racl = ll_i2info(inode)->lli_remote_acl;
1854 struct ptlrpc_request *req = NULL;
1855 struct lustre_id id;
1856 struct mds_remote_perm *mperm;
1857 int rc = -EACCES, perm;
1862 if (__remote_acl_check(racl, &perm) == 0) {
1863 return ((perm & mode) == mode ? 0 : -EACCES);
1870 down(&racl->ra_update_sem);
1872 /* we might lose the race when obtain semaphore,
1875 if (__remote_acl_check(racl, &perm) == 0) {
1876 if ((perm & mode) == mode)
1881 /* really fetch from mds
1883 ll_inode2id(&id, inode);
1884 if (md_access_check(sbi->ll_md_exp, &id, &req))
1887 /* status non-zero indicate there's more apparent error
1888 * detected by mds, e.g. didn't allow this user at all.
1889 * we simply ignore and didn't cache it.
1891 if (req->rq_repmsg->status)
1894 mperm = lustre_swab_repbuf(req, 1, sizeof(*mperm),
1895 lustre_swab_remote_perm);
1897 LASSERT_REPSWABBED(req, 1);
1899 if ((mperm->mrp_perm & mode) == mode)
1902 remote_acl_update_locked(racl, mperm);
1905 ptlrpc_req_finished(req);
1907 up(&racl->ra_update_sem);
1911 int ll_remote_acl_update(struct inode *inode, struct mds_remote_perm *perm)
1913 struct remote_acl *racl = ll_i2info(inode)->lli_remote_acl;
1921 down(&racl->ra_update_sem);
1922 rc = remote_acl_update_locked(racl, perm);
1923 up(&racl->ra_update_sem);
1928 void ll_inode_invalidate_acl(struct inode *inode)
1930 struct ll_sb_info *sbi = ll_i2sbi(inode);
1931 struct ll_inode_info *lli = ll_i2info(inode);
1933 if (sbi->ll_remote) {
1934 struct remote_acl *racl = lli->lli_remote_acl;
1936 LASSERT(!lli->lli_posix_acl);
1938 down(&racl->ra_update_sem);
1939 spin_lock(&racl->ra_lock);
1940 remote_acl_flush(lli->lli_remote_acl);
1941 spin_unlock(&racl->ra_lock);
1942 up(&racl->ra_update_sem);
1945 /* we can't invalide acl here: suppose we touch a new file
1946 * under a dir, blocking ast on dir will lead to open failure
1947 * on client, although succeed on mds. it's kind of weird,
1948 * the real fix i think is improve client-vfs interaction.
1950 * currently we just do nothing here.
1954 LASSERT(!lli->lli_remote_acl);
1955 spin_lock(&lli->lli_lock);
1956 posix_acl_release(lli->lli_posix_acl);
1957 lli->lli_posix_acl = NULL;
1958 spin_unlock(&lli->lli_lock);
1962 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1964 struct ll_inode_info *lli = ll_i2info(inode);
1965 struct lov_stripe_md *lsm = md->lsm;
1966 struct mds_body *body = md->body;
1967 struct mea *mea = md->mea;
1968 struct posix_acl *posix_acl = md->posix_acl;
1969 struct ll_sb_info *sbi = ll_i2sbi(inode);
1970 struct lustre_key *mkey = md->key;
1973 LASSERT((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1975 if (md->lsm && md->lsm->lsm_magic != LOV_MAGIC) {
1976 /* check for default striping info for dir. */
1977 LASSERT((mea != NULL) == ((body->valid & OBD_MD_FLDIREA) != 0));
1981 LASSERT(lsm->lsm_object_gr > 0);
1982 if (lli->lli_smd == NULL) {
1984 lli->lli_maxbytes = lsm->lsm_maxbytes;
1985 if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1986 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1989 if (memcmp(lli->lli_smd, lsm, sizeof(*lsm))) {
1990 CERROR("lsm mismatch for inode %ld\n",
1992 CERROR("lli_smd:\n");
1993 dump_lsm(D_ERROR, lli->lli_smd);
1995 dump_lsm(D_ERROR, lsm);
1998 /* XXX FIXME -- We should decide on a safer (atomic) and
1999 * more elegant way to update the lsm */
2000 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2001 lli->lli_smd->lsm_oinfo[i].loi_id =
2002 lsm->lsm_oinfo[i].loi_id;
2003 lli->lli_smd->lsm_oinfo[i].loi_gr =
2004 lsm->lsm_oinfo[i].loi_gr;
2005 lli->lli_smd->lsm_oinfo[i].loi_ost_idx =
2006 lsm->lsm_oinfo[i].loi_ost_idx;
2007 lli->lli_smd->lsm_oinfo[i].loi_ost_gen =
2008 lsm->lsm_oinfo[i].loi_ost_gen;
2011 /* bug 2844 - limit i_blksize for broken user-space apps */
2012 LASSERTF(lsm->lsm_xfersize != 0, "%lu\n", lsm->lsm_xfersize);
2013 inode->i_blksize = min(lsm->lsm_xfersize, LL_MAX_BLKSIZE);
2014 if (lli->lli_smd != lsm)
2015 obd_free_memmd(ll_i2dtexp(inode), &lsm);
2019 if (lli->lli_mea == NULL) {
2022 if (memcmp(lli->lli_mea, mea, body->eadatasize)) {
2023 CERROR("mea mismatch for inode %lu\n",
2028 if (lli->lli_mea != mea)
2029 obd_free_memmd(ll_i2mdexp(inode),
2030 (struct lov_stripe_md **) &mea);
2033 if (body->valid & OBD_MD_FID)
2034 id_assign_fid(&lli->lli_id, &body->id1);
2036 if (body->valid & OBD_MD_FLID)
2037 id_ino(&lli->lli_id) = id_ino(&body->id1);
2039 if (body->valid & OBD_MD_FLGENER)
2040 id_gen(&lli->lli_id) = id_gen(&body->id1);
2042 /* local/remote ACL */
2043 if (sbi->ll_remote) {
2044 LASSERT(md->posix_acl == NULL);
2045 if (md->remote_perm) {
2046 ll_remote_acl_update(inode, md->remote_perm);
2047 OBD_FREE(md->remote_perm, sizeof(*md->remote_perm));
2048 md->remote_perm = NULL;
2051 LASSERT(md->remote_perm == NULL);
2052 spin_lock(&lli->lli_lock);
2053 if (posix_acl != NULL) {
2054 if (lli->lli_posix_acl != NULL)
2055 posix_acl_release(lli->lli_posix_acl);
2056 lli->lli_posix_acl = posix_acl;
2058 spin_unlock(&lli->lli_lock);
2061 if (body->valid & OBD_MD_FLID)
2062 inode->i_ino = id_ino(&body->id1);
2063 if (body->valid & OBD_MD_FLGENER)
2064 inode->i_generation = id_gen(&body->id1);
2065 if (body->valid & OBD_MD_FLATIME)
2066 LTIME_S(inode->i_atime) = body->atime;
2067 if (body->valid & OBD_MD_FLMTIME &&
2068 body->mtime > LTIME_S(inode->i_mtime)) {
2069 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to %u\n",
2070 inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
2071 LTIME_S(inode->i_mtime) = body->mtime;
2073 if (body->valid & OBD_MD_FLCTIME &&
2074 body->ctime > LTIME_S(inode->i_ctime))
2075 LTIME_S(inode->i_ctime) = body->ctime;
2076 if (body->valid & OBD_MD_FLMODE) {
2077 inode->i_mode = (inode->i_mode & S_IFMT) |
2078 (body->mode & ~S_IFMT);
2080 if (body->valid & OBD_MD_FLTYPE) {
2081 inode->i_mode = (inode->i_mode & ~S_IFMT) |
2082 (body->mode & S_IFMT);
2084 if (body->valid & OBD_MD_FLUID)
2085 inode->i_uid = body->uid;
2086 if (body->valid & OBD_MD_FLGID)
2087 inode->i_gid = body->gid;
2088 if (body->valid & OBD_MD_FLFLAGS)
2089 inode->i_flags = body->flags;
2090 if (body->valid & OBD_MD_FLNLINK)
2091 inode->i_nlink = body->nlink;
2092 if (body->valid & OBD_MD_FLRDEV)
2093 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
2094 inode->i_rdev = body->rdev;
2096 inode->i_rdev = old_decode_dev(body->rdev);
2098 if (body->valid & OBD_MD_FLSIZE)
2099 inode->i_size = body->size;
2100 if (body->valid & OBD_MD_FLBLOCKS)
2101 inode->i_blocks = body->blocks;
2103 if (body->valid & OBD_MD_FLSIZE)
2104 set_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &lli->lli_flags);
2106 if (body->valid & OBD_MD_FLAUDIT) {
2107 struct ll_sb_info * sbi = ll_s2sbi(inode->i_sb);
2108 if (IS_AUDIT_OP(body->audit, AUDIT_FS))
2109 sbi->ll_audit_mask = body->audit;
2111 lli->lli_audit_mask = body->audit;
2115 LASSERT(body->valid & OBD_MD_FLKEY);
2116 ll_crypto_init_inode_key(inode, mkey);
2119 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
2120 inode->i_dev = (kdev_t)id_group(&lli->lli_id);
2122 LASSERT(id_fid(&lli->lli_id) != 0);
2125 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
2126 static struct backing_dev_info ll_backing_dev_info = {
2127 .ra_pages = 0, /* No readahead */
2128 .memory_backed = 0, /* Does contribute to dirty memory */
2132 void ll_read_inode2(struct inode *inode, void *opaque)
2134 struct lustre_md *md = opaque;
2135 struct ll_inode_info *lli = ll_i2info(inode);
2138 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
2139 inode->i_generation, inode);
2143 LASSERT(!lli->lli_smd);
2145 if (ll_i2sbi(inode)->ll_remote) {
2146 lli->lli_remote_acl = remote_acl_alloc();
2147 /* if failed alloc, nobody will be able to access this inode */
2150 /* Core attributes from the MDS first. This is a new inode, and
2151 * the VFS doesn't zero times in the core inode so we have to do
2152 * it ourselves. They will be overwritten by either MDS or OST
2153 * attributes - we just need to make sure they aren't newer. */
2154 LTIME_S(inode->i_mtime) = 0;
2155 LTIME_S(inode->i_atime) = 0;
2156 LTIME_S(inode->i_ctime) = 0;
2159 ll_update_inode(inode, md);
2161 /* OIDEBUG(inode); */
2163 if (S_ISREG(inode->i_mode)) {
2164 inode->i_op = &ll_file_inode_operations;
2165 inode->i_fop = &ll_file_operations;
2166 inode->i_mapping->a_ops = &ll_aops;
2168 } else if (S_ISDIR(inode->i_mode)) {
2169 inode->i_op = &ll_dir_inode_operations;
2170 inode->i_fop = &ll_dir_operations;
2171 inode->i_mapping->a_ops = &ll_dir_aops;
2173 } else if (S_ISLNK(inode->i_mode)) {
2174 inode->i_op = &ll_fast_symlink_inode_operations;
2177 inode->i_op = &ll_special_inode_operations;
2179 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
2180 init_special_inode(inode, inode->i_mode,
2181 kdev_t_to_nr(inode->i_rdev));
2183 /* initializing backing dev info. */
2184 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
2186 init_special_inode(inode, inode->i_mode, inode->i_rdev);
2188 lli->ll_save_ifop = inode->i_fop;
2190 if (S_ISCHR(inode->i_mode))
2191 inode->i_fop = &ll_special_chr_inode_fops;
2192 else if (S_ISBLK(inode->i_mode))
2193 inode->i_fop = &ll_special_blk_inode_fops;
2194 else if (S_ISFIFO(inode->i_mode))
2195 inode->i_fop = &ll_special_fifo_inode_fops;
2196 else if (S_ISSOCK(inode->i_mode))
2197 inode->i_fop = &ll_special_sock_inode_fops;
2199 CWARN("saved %p, replaced with %p\n", lli->ll_save_ifop,
2202 if (lli->ll_save_ifop->owner) {
2203 CWARN("%p has owner %p\n", lli->ll_save_ifop,
2204 lli->ll_save_ifop->owner);
2210 void ll_delete_inode(struct inode *inode)
2212 struct ll_sb_info *sbi = ll_i2sbi(inode);
2213 struct lustre_id id;
2217 ll_inode2id(&id, inode);
2219 rc = md_delete_inode(sbi->ll_md_exp, &id);
2221 CERROR("md_delete_inode() failed, error %d\n",
2229 int ll_iocontrol(struct inode *inode, struct file *file,
2230 unsigned int cmd, unsigned long arg)
2232 struct ll_sb_info *sbi = ll_i2sbi(inode);
2233 struct ptlrpc_request *req = NULL;
2238 case EXT3_IOC_GETFLAGS: {
2239 struct lustre_id id;
2240 __u64 valid = OBD_MD_FLFLAGS;
2241 struct mds_body *body;
2243 ll_inode2id(&id, inode);
2244 rc = md_getattr(sbi->ll_md_exp, &id, valid, NULL, NULL,
2247 CERROR("failure %d inode %lu\n", rc, inode->i_ino);
2251 body = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*body));
2253 if (body->flags & S_APPEND)
2254 flags |= EXT3_APPEND_FL;
2255 if (body->flags & S_IMMUTABLE)
2256 flags |= EXT3_IMMUTABLE_FL;
2257 if (body->flags & S_NOATIME)
2258 flags |= EXT3_NOATIME_FL;
2260 ptlrpc_req_finished (req);
2262 RETURN(put_user(flags, (int *)arg));
2264 case EXT3_IOC_SETFLAGS: {
2265 struct mdc_op_data *op_data;
2268 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
2270 if (get_user(flags, (int *)arg))
2277 OBD_ALLOC(op_data, sizeof(*op_data));
2278 if (op_data == NULL) {
2282 ll_inode2mdc_data(op_data, inode, (OBD_MD_FLID | OBD_MD_MEA));
2284 memset(&attr, 0x0, sizeof(attr));
2285 attr.ia_attr_flags = flags;
2286 attr.ia_valid |= ATTR_ATTR_FLAG;
2288 rc = md_setattr(sbi->ll_md_exp, op_data,
2289 &attr, NULL, 0, NULL, 0, NULL, 0, &req);
2290 OBD_FREE(op_data, sizeof(*op_data));
2292 ptlrpc_req_finished(req);
2293 if (rc != -EPERM && rc != -EACCES)
2294 CERROR("md_setattr fails: rc = %d\n", rc);
2298 ptlrpc_req_finished(req);
2300 oa->o_id = lsm->lsm_object_id;
2301 oa->o_gr = lsm->lsm_object_gr;
2302 oa->o_flags = flags;
2303 *(obdo_id(oa)) = ll_i2info(inode)->lli_id;
2304 oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS | OBD_MD_FLGROUP
2307 rc = obd_setattr(sbi->ll_dt_exp, oa, lsm, NULL, NULL);
2310 if (rc != -EPERM && rc != -EACCES)
2311 CERROR("md_setattr fails: rc = %d\n", rc);
2315 if (flags & EXT3_APPEND_FL)
2316 inode->i_flags |= S_APPEND;
2318 inode->i_flags &= ~S_APPEND;
2319 if (flags & EXT3_IMMUTABLE_FL)
2320 inode->i_flags |= S_IMMUTABLE;
2322 inode->i_flags &= ~S_IMMUTABLE;
2323 if (flags & EXT3_NOATIME_FL)
2324 inode->i_flags |= S_NOATIME;
2326 inode->i_flags &= ~S_NOATIME;
2337 /* this is only called in the case of forced umount. */
2338 void ll_umount_begin(struct super_block *sb)
2340 struct ll_sb_info *sbi = ll_s2sbi(sb);
2341 struct obd_ioctl_data ioc_data = { 0 };
2342 struct obd_device *obd;
2345 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2346 sb->s_count, atomic_read(&sb->s_active));
2348 obd = class_exp2obd(sbi->ll_md_exp);
2350 CERROR("Invalid MDC connection handle "LPX64"\n",
2351 sbi->ll_md_exp->exp_handle.h_cookie);
2355 obd->obd_no_recov = 1;
2356 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2357 sizeof(ioc_data), &ioc_data, NULL);
2359 obd = class_exp2obd(sbi->ll_dt_exp);
2361 CERROR("Invalid LOV connection handle "LPX64"\n",
2362 sbi->ll_dt_exp->exp_handle.h_cookie);
2367 obd->obd_no_recov = 1;
2368 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2369 sizeof(ioc_data), &ioc_data, NULL);
2372 * really, we'd like to wait until there are no requests outstanding,
2373 * and then continue. For now, we just invalidate the requests,
2374 * schedule, and hope.
2381 int ll_prep_inode(struct obd_export *dt_exp, struct obd_export *md_exp,
2382 struct inode **inode, struct ptlrpc_request *req,
2383 int offset, struct super_block *sb)
2385 struct lustre_md md;
2388 rc = mdc_req2lustre_md(md_exp, req, offset, dt_exp, &md);
2393 ll_update_inode(*inode, &md);
2396 *inode = ll_iget(sb, id_ino(&md.body->id1), &md);
2397 if (*inode == NULL || is_bad_inode(*inode)) {
2398 /* free the lsm if we allocated one above */
2400 obd_free_memmd(dt_exp, &md.lsm);
2402 obd_free_memmd(md_exp,
2403 (struct lov_stripe_md**)&md.mea);
2405 CERROR("new_inode -fatal: rc %d\n", rc);
2412 int ll_show_options(struct seq_file *m, struct vfsmount *mnt)
2414 struct ll_sb_info *sbi = ll_s2sbi(mnt->mnt_sb);
2415 struct lustre_mount_data *lmd = sbi->ll_lmd;
2418 seq_printf(m, ",mds_sec=%s,oss_sec=%s",
2419 lmd->lmd_mds_security, lmd->lmd_oss_security);
2421 seq_printf(m, ",%s", sbi->ll_remote ? "remote" : "local");
2422 if (sbi->ll_remote && lmd)
2423 seq_printf(m, ",nllu=%u:%u", lmd->lmd_nllu, lmd->lmd_nllg);
2425 if (lmd && lmd->lmd_pag)
2426 seq_printf(m, ",pag");
2431 int ll_get_fid(struct obd_export *exp, struct lustre_id *idp,
2432 char *filename, struct lustre_id *ret)
2434 struct ptlrpc_request *request = NULL;
2435 struct mds_body *body;
2438 rc = md_getattr_lock(exp, idp, filename, strlen(filename) + 1,
2439 OBD_MD_FID, 0, &request);
2441 CDEBUG(D_INFO, "md_getattr_lock failed on %s: rc %d\n",
2446 body = lustre_msg_buf(request->rq_repmsg, 0, sizeof(*body));
2447 LASSERT(body != NULL);
2448 LASSERT_REPSWABBED(request, 0);
2451 ptlrpc_req_finished(request);
2455 int ll_flush_cred(struct inode *inode)
2457 struct ll_sb_info *sbi = ll_i2sbi(inode);
2460 /* XXX to avoid adding api, we simply use set_info() interface
2461 * to notify underlying obds. set_info() is more like a ioctl() now...
2463 if (sbi->ll_md_exp) {
2464 rc = obd_set_info(sbi->ll_md_exp,
2465 strlen("flush_cred"), "flush_cred",
2471 if (sbi->ll_dt_exp) {
2472 rc = obd_set_info(sbi->ll_dt_exp,
2473 strlen("flush_cred"), "flush_cred",