1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Lustre Light Super operations
6 * Copyright (c) 2002, 2003 Cluster File Systems, Inc.
8 * This file is part of Lustre, http://www.lustre.org.
10 * Lustre is free software; you can redistribute it and/or
11 * modify it under the terms of version 2 of the GNU General Public
12 * License as published by the Free Software Foundation.
14 * Lustre is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
17 * GNU General Public License for more details.
19 * You should have received a copy of the GNU General Public License
20 * along with Lustre; if not, write to the Free Software
21 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
24 #define DEBUG_SUBSYSTEM S_LLITE
26 #include <linux/module.h>
27 #include <linux/types.h>
28 #include <linux/random.h>
29 #include <linux/version.h>
31 #include <linux/lustre_lite.h>
32 #include <linux/lustre_ha.h>
33 #include <linux/lustre_dlm.h>
34 #include <linux/lprocfs_status.h>
35 #include <linux/lustre_acl.h>
36 #include "llite_internal.h"
38 kmem_cache_t *ll_file_data_slab;
39 kmem_cache_t *ll_intent_slab;
41 extern struct address_space_operations ll_aops;
42 extern struct address_space_operations ll_dir_aops;
45 #define log2(n) ffz(~(n))
48 struct ll_sb_info *lustre_init_sbi(struct super_block *sb)
50 struct ll_sb_info *sbi = NULL;
54 OBD_ALLOC(sbi, sizeof(*sbi));
58 spin_lock_init(&sbi->ll_lock);
59 INIT_LIST_HEAD(&sbi->ll_pglist);
60 sbi->ll_pglist_gen = 0;
61 if (num_physpages < SBI_DEFAULT_RA_MAX / 4)
62 sbi->ll_ra_info.ra_max_pages = num_physpages / 4;
64 sbi->ll_ra_info.ra_max_pages = SBI_DEFAULT_RA_MAX;
65 INIT_LIST_HEAD(&sbi->ll_conn_chain);
66 INIT_HLIST_HEAD(&sbi->ll_orphan_dentry_list);
67 INIT_LIST_HEAD(&sbi->ll_mnt_list);
69 sema_init(&sbi->ll_gns_sem, 1);
70 spin_lock_init(&sbi->ll_gns_lock);
71 INIT_LIST_HEAD(&sbi->ll_gns_sbi_head);
72 init_waitqueue_head(&sbi->ll_gns_waitq);
73 init_completion(&sbi->ll_gns_mount_finished);
75 /* this later may be reset via /proc/fs/... */
76 memcpy(sbi->ll_gns_oname, ".mntinfo", strlen(".mntinfo"));
77 sbi->ll_gns_oname[strlen(sbi->ll_gns_oname)] = '\0';
79 /* this later may be reset via /proc/fs/... */
80 memcpy(sbi->ll_gns_upcall, "/usr/sbin/gns_upcall",
81 strlen("/usr/sbin/gns_upcall"));
82 sbi->ll_gns_upcall[strlen(sbi->ll_gns_upcall)] = '\0';
84 /* default values, may be changed via /proc/fs/... */
85 sbi->ll_gns_state = LL_GNS_IDLE;
86 sbi->ll_gns_pending_dentry = NULL;
87 atomic_set(&sbi->ll_gns_enabled, 1);
88 sbi->ll_gns_tick = GNS_TICK_TIMEOUT;
89 sbi->ll_gns_timeout = GNS_MOUNT_TIMEOUT;
91 sbi->ll_gns_timer.data = (unsigned long)sbi;
92 sbi->ll_gns_timer.function = ll_gns_timer_callback;
93 init_timer(&sbi->ll_gns_timer);
97 generate_random_uuid(uuid);
98 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
102 void lustre_free_sbi(struct super_block *sb)
104 struct ll_sb_info *sbi = ll_s2sbi(sb);
108 list_del(&sbi->ll_gns_sbi_head);
109 del_timer(&sbi->ll_gns_timer);
110 OBD_FREE(sbi, sizeof(*sbi));
112 ll_set_sbi(sb, NULL);
116 int lustre_init_dt_desc(struct ll_sb_info *sbi)
122 valsize = sizeof(sbi->ll_dt_desc);
123 memset(&sbi->ll_dt_desc, 0, sizeof(sbi->ll_dt_desc));
124 rc = obd_get_info(sbi->ll_dt_exp, strlen("lovdesc") + 1,
125 "lovdesc", &valsize, &sbi->ll_dt_desc);
129 extern struct dentry_operations ll_d_ops;
131 int lustre_common_fill_super(struct super_block *sb, char *lmv, char *lov,
132 int async, char *security, __u32 *nllu,
135 struct ll_sb_info *sbi = ll_s2sbi(sb);
136 struct ptlrpc_request *request = NULL;
137 struct lustre_handle dt_conn = {0, };
138 struct lustre_handle md_conn = {0, };
139 struct obd_connect_data *data;
140 struct inode *root = NULL;
141 struct obd_device *obd;
142 struct obd_statfs osfs;
149 obd = class_name2obd(lmv);
151 CERROR("MDC %s: not setup or attached\n", lmv);
154 obd_set_info(obd->obd_self_export, strlen("async"), "async",
155 sizeof(async), &async);
157 if ((*remote & (OBD_CONNECT_LOCAL | OBD_CONNECT_REMOTE)) ==
158 (OBD_CONNECT_LOCAL | OBD_CONNECT_REMOTE)) {
159 CERROR("wrong remote flag "LPX64"\n", *remote);
163 OBD_ALLOC(data, sizeof(*data));
167 data->ocd_connect_flags |= *remote & (OBD_CONNECT_LOCAL |
169 memcpy(data->ocd_nllu, nllu, sizeof(data->ocd_nllu));
171 if (security == NULL)
174 err = obd_set_info(obd->obd_self_export, strlen("sec"), "sec",
175 strlen(security), security);
177 CERROR("LMV %s: failed to set security %s, err %d\n",
179 OBD_FREE(data, sizeof(*data));
183 if (proc_lustre_fs_root) {
184 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
187 CERROR("could not register mount in /proc/lustre");
190 err = obd_connect(&md_conn, obd, &sbi->ll_sb_uuid, data,
191 OBD_OPT_REAL_CLIENT);
193 CERROR("An MDS (lmv %s) is performing recovery, of which this"
194 " client is not a part. Please wait for recovery to "
195 "complete, abort, or time out.\n", lmv);
198 CERROR("cannot connect to %s: rc = %d\n", lmv, err);
201 sbi->ll_md_exp = class_conn2export(&md_conn);
202 err = obd_statfs(obd, &osfs, jiffies - HZ);
206 if (!osfs.os_bsize) {
207 CERROR("Invalid block size is detected.");
211 sb->s_magic = LL_SUPER_MAGIC;
212 sb->s_blocksize = osfs.os_bsize;
213 sb->s_blocksize_bits = log2(osfs.os_bsize);
214 sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
216 devno = get_uuid2int((char *)sbi->ll_md_exp->exp_obd->obd_uuid.uuid,
217 strlen((char *)sbi->ll_md_exp->exp_obd->obd_uuid.uuid));
221 /* after statfs, we are supposed to have connected to MDSs,
222 * so it's ok to check remote flag returned.
224 valsize = sizeof(&sbi->ll_remote);
225 err = obd_get_info(sbi->ll_md_exp, strlen("remote_flag"), "remote_flag",
226 &valsize, &sbi->ll_remote);
228 CERROR("fail to obtain remote flag\n");
232 obd = class_name2obd(lov);
234 CERROR("OSC %s: not setup or attached\n", lov);
237 obd_set_info(obd->obd_self_export, strlen("async"), "async",
238 sizeof(async), &async);
240 err = obd_connect(&dt_conn, obd, &sbi->ll_sb_uuid, data,
241 OBD_OPT_REAL_CLIENT);
243 CERROR("An OST (lov %s) is performing recovery, of which this"
244 " client is not a part. Please wait for recovery to "
245 "complete, abort, or time out.\n", lov);
248 CERROR("cannot connect to %s: rc = %d\n", lov, err);
251 sbi->ll_dt_exp = class_conn2export(&dt_conn);
253 err = lustre_init_dt_desc(sbi);
255 int mdsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
256 obd_init_ea_size(sbi->ll_md_exp, mdsize,
257 sbi->ll_dt_desc.ld_tgt_count *
258 sizeof(struct llog_cookie));
261 err = md_getstatus(sbi->ll_md_exp, &sbi->ll_rootid);
263 CERROR("cannot mds_connect: rc = %d\n", err);
266 CDEBUG(D_SUPER, "rootid "DLID4"\n", OLID4(&sbi->ll_rootid));
268 sb->s_op = &lustre_super_operations;
271 * XXX: move this to after cbd setup? */
272 err = md_getattr(sbi->ll_md_exp, &sbi->ll_rootid,
273 (OBD_MD_FLNOTOBD | OBD_MD_FLBLOCKS | OBD_MD_FID), NULL, 0,
276 CERROR("md_getattr failed for root: rc = %d\n", err);
280 err = mdc_req2lustre_md(sbi->ll_md_exp, request, 0,
281 sbi->ll_dt_exp, &md);
283 CERROR("failed to understand root inode md: rc = %d\n", err);
284 ptlrpc_req_finished(request);
288 LASSERT(id_ino(&sbi->ll_rootid) != 0);
289 root = ll_iget(sb, id_ino(&sbi->ll_rootid), &md);
291 ptlrpc_req_finished(request);
293 if (root == NULL || is_bad_inode(root)) {
295 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
297 obd_free_memmd(sbi->ll_md_exp,
298 (struct lov_stripe_md**)&md.mea);
299 CERROR("lustre_lite: bad iget4 for root\n");
300 GOTO(out_root, err = -EBADF);
303 err = ll_close_thread_start(&sbi->ll_lcq);
305 CERROR("cannot start close thread: rc %d\n", err);
309 ll_gns_add_timer(sbi);
311 /* making vm readahead 0 for 2.4.x. In the case of 2.6.x,
312 backing dev info assigned to inode mapping is used for
313 determining maximal readahead. */
314 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0)) && \
315 !defined(KERNEL_HAS_AS_MAX_READAHEAD)
316 /* bug 2805 - set VM readahead to zero */
317 vm_max_readahead = vm_min_readahead = 0;
320 sb->s_root = d_alloc_root(root);
321 sb->s_root->d_op = &ll_d_ops;
323 sb->s_flags |= MS_POSIXACL;
325 CWARN("Enabling PDIROPS\n");
326 sb->s_flags |= S_PDIROPS;
330 OBD_FREE(data, sizeof(*data));
336 obd_disconnect(sbi->ll_dt_exp, 0);
338 obd_disconnect(sbi->ll_md_exp, 0);
341 OBD_FREE(data, sizeof(*data));
342 lprocfs_unregister_mountpoint(sbi);
346 void lustre_common_put_super(struct super_block *sb)
348 struct ll_sb_info *sbi = ll_s2sbi(sb);
349 struct hlist_node *tmp, *next;
352 ll_gns_del_timer(sbi);
353 ll_close_thread_shutdown(sbi->ll_lcq);
355 list_del(&sbi->ll_conn_chain);
356 obd_disconnect(sbi->ll_dt_exp, 0);
358 lprocfs_unregister_mountpoint(sbi);
359 if (sbi->ll_proc_root) {
360 lprocfs_remove(sbi->ll_proc_root);
361 sbi->ll_proc_root = NULL;
364 obd_disconnect(sbi->ll_md_exp, 0);
366 // We do this to get rid of orphaned dentries. That is not really trw.
367 hlist_for_each_safe(tmp, next, &sbi->ll_orphan_dentry_list) {
368 struct dentry *dentry = hlist_entry(tmp, struct dentry, d_hash);
369 CWARN("orphan dentry %.*s (%p->%p) at unmount\n",
370 dentry->d_name.len, dentry->d_name.name, dentry, next);
371 shrink_dcache_parent(dentry);
376 char *ll_read_opt(const char *opt, char *data)
382 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
383 if (strncmp(opt, data, strlen(opt)))
385 if ((value = strchr(data, '=')) == NULL)
389 OBD_ALLOC(retval, strlen(value) + 1);
391 CERROR("out of memory!\n");
395 memcpy(retval, value, strlen(value)+1);
396 CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
400 int ll_set_opt(const char *opt, char *data, int fl)
404 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
405 if (strncmp(opt, data, strlen(opt)))
411 void ll_options(char *options, char **lov, char **lmv, char **sec,
412 int *async, int *flags)
415 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
416 char *opt_ptr = options;
426 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
427 for (this_char = strtok (options, ",");
429 this_char = strtok (NULL, ",")) {
431 while ((this_char = strsep (&opt_ptr, ",")) != NULL) {
433 CDEBUG(D_SUPER, "this_char %s\n", this_char);
434 if (!*lov && (*lov = ll_read_opt("osc", this_char)))
436 if (!*lmv && (*lmv = ll_read_opt("mdc", this_char)))
438 if (!strncmp(this_char, "lasync", strlen("lasync"))) {
442 if (!*sec && (*sec = ll_read_opt("sec", this_char)))
444 if (!(*flags & LL_SBI_NOLCK) &&
445 ((*flags) = (*flags) |
446 ll_set_opt("nolock", this_char,
454 void ll_lli_init(struct ll_inode_info *lli)
456 sema_init(&lli->lli_open_sem, 1);
457 sema_init(&lli->lli_size_sem, 1);
459 lli->lli_size_pid = 0;
460 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
461 spin_lock_init(&lli->lli_lock);
462 INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
463 lli->lli_inode_magic = LLI_INODE_MAGIC;
464 memset(&lli->lli_id, 0, sizeof(lli->lli_id));
465 sema_init(&lli->lli_och_sem, 1);
466 lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
467 lli->lli_mds_exec_och = NULL;
468 lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
469 lli->lli_open_fd_exec_count = 0;
472 int ll_fill_super(struct super_block *sb, void *data, int silent)
474 struct ll_sb_info *sbi;
479 __u32 nllu[2] = { NOBODY_UID, NOBODY_GID };
480 __u64 remote_flag = 0;
483 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
485 sbi = lustre_init_sbi(sb);
489 sbi->ll_flags |= LL_SBI_READAHEAD;
490 ll_options(data, &lov, &lmv, &sec, &async, &sbi->ll_flags);
494 GOTO(out, err = -EINVAL);
499 GOTO(out, err = -EINVAL);
502 err = lustre_common_fill_super(sb, lmv, lov, async, sec, nllu,
510 OBD_FREE(sec, strlen(sec) + 1);
512 OBD_FREE(lmv, strlen(lmv) + 1);
514 OBD_FREE(lov, strlen(lov) + 1);
516 } /* ll_read_super */
518 static int lustre_process_log(struct lustre_mount_data *lmd, char *profile,
519 struct config_llog_instance *cfg, int allow_recov)
521 struct lustre_cfg *lcfg = NULL;
522 struct lustre_cfg_bufs bufs;
523 struct portals_cfg pcfg;
524 char *peer = "MDS_PEER_UUID";
525 struct obd_device *obd;
526 struct lustre_handle md_conn = {0, };
527 struct obd_export *exp;
528 char *name = "mdc_dev";
530 struct obd_uuid lmv_uuid;
531 struct llog_ctxt *ctxt;
535 if (lmd_bad_magic(lmd))
538 generate_random_uuid(uuid);
539 class_uuid_unparse(uuid, &lmv_uuid);
541 if (lmd->lmd_local_nid) {
542 PCFG_INIT(pcfg, NAL_CMD_REGISTER_MYNID);
543 pcfg.pcfg_nal = lmd->lmd_nal;
544 pcfg.pcfg_nid = lmd->lmd_local_nid;
545 rc = libcfs_nal_cmd(&pcfg);
550 if (lmd->lmd_nal == SOCKNAL ||
551 lmd->lmd_nal == OPENIBNAL ||
552 lmd->lmd_nal == IIBNAL ||
553 lmd->lmd_nal == VIBNAL ||
554 lmd->lmd_nal == RANAL) {
555 PCFG_INIT(pcfg, NAL_CMD_ADD_PEER);
556 pcfg.pcfg_nal = lmd->lmd_nal;
557 pcfg.pcfg_nid = lmd->lmd_server_nid;
558 pcfg.pcfg_id = lmd->lmd_server_ipaddr;
559 pcfg.pcfg_misc = lmd->lmd_port;
560 rc = libcfs_nal_cmd(&pcfg);
564 lustre_cfg_bufs_reset(&bufs, name);
565 lustre_cfg_bufs_set_string(&bufs, 1, peer);
567 lcfg = lustre_cfg_new(LCFG_ADD_UUID, &bufs);
568 lcfg->lcfg_nal = lmd->lmd_nal;
569 lcfg->lcfg_nid = lmd->lmd_server_nid;
570 LASSERT(lcfg->lcfg_nal);
571 LASSERT(lcfg->lcfg_nid);
572 err = class_process_config(lcfg);
573 lustre_cfg_free(lcfg);
575 GOTO(out_del_conn, err);
577 lustre_cfg_bufs_reset(&bufs, name);
578 lustre_cfg_bufs_set_string(&bufs, 1, LUSTRE_MDC_NAME);
579 lustre_cfg_bufs_set_string(&bufs, 2, (char *)lmv_uuid.uuid);
581 lcfg = lustre_cfg_new(LCFG_ATTACH, &bufs);
582 err = class_process_config(lcfg);
583 lustre_cfg_free(lcfg);
585 GOTO(out_del_uuid, err);
587 lustre_cfg_bufs_reset(&bufs, name);
588 lustre_cfg_bufs_set_string(&bufs, 1, lmd->lmd_mds);
589 lustre_cfg_bufs_set_string(&bufs, 2, peer);
591 lcfg = lustre_cfg_new(LCFG_SETUP, &bufs);
592 err = class_process_config(lcfg);
593 lustre_cfg_free(lcfg);
595 GOTO(out_detach, err);
597 obd = class_name2obd(name);
599 GOTO(out_cleanup, rc = -EINVAL);
601 rc = obd_set_info(obd->obd_self_export, strlen("sec"), "sec",
602 strlen(lmd->lmd_security), lmd->lmd_security);
604 GOTO(out_cleanup, rc);
606 /* Disable initial recovery on this import */
607 rc = obd_set_info(obd->obd_self_export,
608 strlen("initial_recov"), "initial_recov",
609 sizeof(allow_recov), &allow_recov);
611 GOTO(out_cleanup, rc);
613 rc = obd_connect(&md_conn, obd, &lmv_uuid, NULL, 0);
615 CERROR("cannot connect to %s: rc = %d\n", lmd->lmd_mds, rc);
616 GOTO(out_cleanup, rc);
619 exp = class_conn2export(&md_conn);
621 ctxt = llog_get_context(&exp->exp_obd->obd_llogs,LLOG_CONFIG_REPL_CTXT);
622 rc = class_config_process_llog(ctxt, profile, cfg);
624 CERROR("class_config_process_llog failed: rc = %d\n", rc);
626 err = obd_disconnect(exp, 0);
630 lustre_cfg_bufs_reset(&bufs, name);
631 lcfg = lustre_cfg_new(LCFG_CLEANUP, &bufs);
632 err = class_process_config(lcfg);
633 lustre_cfg_free(lcfg);
637 lustre_cfg_bufs_reset(&bufs, name);
638 lcfg = lustre_cfg_new(LCFG_DETACH, &bufs);
639 err = class_process_config(lcfg);
640 lustre_cfg_free(lcfg);
645 lustre_cfg_bufs_reset(&bufs, name);
646 lustre_cfg_bufs_set_string(&bufs, 1, peer);
647 lcfg = lustre_cfg_new(LCFG_DEL_UUID, &bufs);
648 err = class_process_config(lcfg);
649 lustre_cfg_free(lcfg);
652 if (lmd->lmd_nal == SOCKNAL ||
653 lmd->lmd_nal == OPENIBNAL ||
654 lmd->lmd_nal == IIBNAL ||
655 lmd->lmd_nal == VIBNAL ||
656 lmd->lmd_nal == RANAL) {
659 PCFG_INIT(pcfg, NAL_CMD_DEL_PEER);
660 pcfg.pcfg_nal = lmd->lmd_nal;
661 pcfg.pcfg_nid = lmd->lmd_server_nid;
662 pcfg.pcfg_flags = 1; /* single_share */
663 err2 = libcfs_nal_cmd(&pcfg);
676 static void lustre_manual_cleanup(struct ll_sb_info *sbi)
678 struct lustre_cfg *lcfg;
679 struct lustre_cfg_bufs bufs;
680 struct obd_device *obd;
683 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) != NULL)
687 lustre_cfg_bufs_reset(&bufs, obd->obd_name);
688 lcfg = lustre_cfg_new(LCFG_CLEANUP, &bufs);
689 err = class_process_config(lcfg);
691 CERROR("cleanup failed: %s\n", obd->obd_name);
695 lcfg->lcfg_command = LCFG_DETACH;
696 err = class_process_config(lcfg);
697 lustre_cfg_free(lcfg);
699 CERROR("detach failed: %s\n", obd->obd_name);
704 if (sbi->ll_lmd != NULL)
705 class_del_profile(sbi->ll_lmd->lmd_profile);
708 int lustre_fill_super(struct super_block *sb, void *data, int silent)
710 struct lustre_mount_data * lmd = data;
711 char *lov = NULL, *lmv = NULL;
712 struct ll_sb_info *sbi;
716 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
717 if (lmd_bad_magic(lmd))
720 sbi = lustre_init_sbi(sb);
724 sbi->ll_flags |= LL_SBI_READAHEAD;
726 if (lmd->lmd_profile) {
727 struct lustre_profile *lprof;
728 struct config_llog_instance cfg;
731 if (lmd->lmd_mds[0] == '\0') {
732 CERROR("no mds name\n");
733 GOTO(out_free, err = -EINVAL);
735 lmd->lmd_security[sizeof(lmd->lmd_security) - 1] = 0;
737 OBD_ALLOC(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
738 if (sbi->ll_lmd == NULL)
739 GOTO(out_free, err = -ENOMEM);
740 memcpy(sbi->ll_lmd, lmd, sizeof(*lmd));
742 /* generate a string unique to this super, let's try
743 the address of the super itself.*/
744 len = (sizeof(sb) * 2) + 1;
745 OBD_ALLOC(sbi->ll_instance, len);
746 if (sbi->ll_instance == NULL)
747 GOTO(out_free, err = -ENOMEM);
748 sprintf(sbi->ll_instance, "%p", sb);
750 cfg.cfg_instance = sbi->ll_instance;
751 cfg.cfg_uuid = sbi->ll_sb_uuid;
752 cfg.cfg_local_nid = lmd->lmd_local_nid;
753 err = lustre_process_log(lmd, lmd->lmd_profile, &cfg, 0);
755 CERROR("Unable to process log: %s\n", lmd->lmd_profile);
759 lprof = class_get_profile(lmd->lmd_profile);
761 CERROR("No profile found: %s\n", lmd->lmd_profile);
762 GOTO(out_free, err = -EINVAL);
765 OBD_FREE(lov, strlen(lov) + 1);
766 OBD_ALLOC(lov, strlen(lprof->lp_lov) +
767 strlen(sbi->ll_instance) + 2);
768 sprintf(lov, "%s-%s", lprof->lp_lov, sbi->ll_instance);
771 OBD_FREE(lmv, strlen(lmv) + 1);
772 OBD_ALLOC(lmv, strlen(lprof->lp_lmv) +
773 strlen(sbi->ll_instance) + 2);
774 sprintf(lmv, "%s-%s", lprof->lp_lmv, sbi->ll_instance);
779 GOTO(out_free, err = -EINVAL);
784 GOTO(out_free, err = -EINVAL);
787 err = lustre_common_fill_super(sb, lmv, lov, lmd->lmd_async,
788 lmd->lmd_security, &lmd->lmd_nllu,
789 &lmd->lmd_remote_flag);
796 OBD_FREE(lmv, strlen(lmv) + 1);
798 OBD_FREE(lov, strlen(lov) + 1);
804 int len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean")+1;
807 if (sbi->ll_instance != NULL) {
808 struct lustre_mount_data *lmd = sbi->ll_lmd;
809 struct config_llog_instance cfg;
812 cfg.cfg_instance = sbi->ll_instance;
813 cfg.cfg_uuid = sbi->ll_sb_uuid;
815 OBD_ALLOC(cl_prof, len);
816 sprintf(cl_prof, "%s-clean", lmd->lmd_profile);
817 err = lustre_process_log(lmd, cl_prof, &cfg, 0);
819 CERROR("Unable to process log: %s\n", cl_prof);
820 lustre_manual_cleanup(sbi);
822 OBD_FREE(cl_prof, len);
823 OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance) + 1);
825 OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
829 } /* lustre_fill_super */
831 void lustre_put_super(struct super_block *sb)
833 struct obd_device *obd;
834 struct ll_sb_info *sbi = ll_s2sbi(sb);
835 int force_umount = 0;
838 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
839 obd = class_exp2obd(sbi->ll_md_exp);
841 force_umount = obd->obd_no_recov;
844 lustre_common_put_super(sb);
845 if (sbi->ll_lmd != NULL) {
847 int len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean")+1;
849 struct config_llog_instance cfg;
852 CERROR("force umount, doing manual cleanup\n");
853 lustre_manual_cleanup(sbi);
857 cfg.cfg_instance = sbi->ll_instance;
858 cfg.cfg_uuid = sbi->ll_sb_uuid;
860 OBD_ALLOC(cl_prof, len);
861 sprintf(cl_prof, "%s-clean", sbi->ll_lmd->lmd_profile);
862 err = lustre_process_log(sbi->ll_lmd, cl_prof, &cfg, 0);
864 CERROR("Unable to process log: %s, doing manual cleanup"
866 lustre_manual_cleanup(sbi);
869 OBD_FREE(cl_prof, len);
871 OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
872 OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance) + 1);
878 } /* lustre_put_super */
880 int ll_process_config_update(struct ll_sb_info *sbi, int clean)
882 struct obd_export *md_exp = sbi->ll_md_exp;
883 struct lustre_mount_data *lmd = sbi->ll_lmd;
884 char *profile = lmd->lmd_profile, *name = NULL;
885 struct config_llog_instance cfg;
886 int rc, namelen = 0, version;
887 struct llog_ctxt *ctxt;
893 CERROR("Client not mounted with zero-conf; cannot process "
898 rc = ldlm_cli_cancel_unused(md_exp->exp_obd->obd_namespace, NULL,
899 LDLM_FL_CONFIG_CHANGE, NULL);
901 CWARN("ldlm_cli_cancel_unused(mdc): %d\n", rc);
903 rc = obd_cancel_unused(sbi->ll_dt_exp, NULL, LDLM_FL_CONFIG_CHANGE,
906 CWARN("obd_cancel_unused(lov): %d\n", rc);
908 cfg.cfg_instance = sbi->ll_instance;
909 cfg.cfg_uuid = sbi->ll_sb_uuid;
910 cfg.cfg_local_nid = lmd->lmd_local_nid;
912 namelen = strlen(profile) + 20; /* -clean-######### */
913 OBD_ALLOC(name, namelen);
918 version = sbi->ll_config_version - 1;
919 sprintf(name, "%s-clean-%d", profile, version);
921 version = sbi->ll_config_version + 1;
922 sprintf(name, "%s-%d", profile, version);
925 CWARN("Applying configuration log %s\n", name);
927 ctxt = llog_get_context(&md_exp->exp_obd->obd_llogs,
928 LLOG_CONFIG_REPL_CTXT);
929 rc = class_config_process_llog(ctxt, name, &cfg);
931 sbi->ll_config_version = version;
932 CWARN("Finished applying configuration log %s: %d\n", name, rc);
934 if (rc == 0 && clean == 0) {
935 struct lov_desc desc;
939 valsize = sizeof(desc);
940 rc = obd_get_info(sbi->ll_dt_exp, strlen("lovdesc") + 1,
941 "lovdesc", &valsize, &desc);
943 rc = obd_init_ea_size(md_exp,
944 obd_size_diskmd(sbi->ll_dt_exp, NULL),
946 sizeof(struct llog_cookie)));
948 OBD_FREE(name, namelen);
952 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
954 struct inode *inode = NULL;
955 l_lock(&lock->l_resource->lr_namespace->ns_lock);
956 if (lock->l_ast_data) {
957 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
958 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
959 inode = igrab(lock->l_ast_data);
961 inode = lock->l_ast_data;
962 CDEBUG(inode->i_state & I_FREEING ? D_INFO : D_WARNING,
963 "l_ast_data %p is bogus: magic %0x8\n",
964 lock->l_ast_data, lli->lli_inode_magic);
968 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
972 int null_if_equal(struct ldlm_lock *lock, void *data)
974 if (data == lock->l_ast_data) {
975 lock->l_ast_data = NULL;
977 if (lock->l_req_mode != lock->l_granted_mode)
978 LDLM_ERROR(lock,"clearing inode with ungranted lock\n");
981 return LDLM_ITER_CONTINUE;
984 void ll_clear_inode(struct inode *inode)
987 struct ll_inode_info *lli = ll_i2info(inode);
988 struct ll_sb_info *sbi = ll_i2sbi(inode);
991 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
992 inode->i_generation, inode);
994 lli->lli_inode_magic = LLI_INODE_DEAD;
995 ll_inode2id(&id, inode);
997 clear_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &(ll_i2info(inode)->lli_flags));
998 md_change_cbdata(sbi->ll_md_exp, &id, null_if_equal, inode);
1000 LASSERT(!lli->lli_open_fd_write_count);
1001 LASSERT(!lli->lli_open_fd_read_count);
1002 LASSERT(!lli->lli_open_fd_exec_count);
1003 if (lli->lli_mds_write_och)
1004 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_WRITE);
1005 if (lli->lli_mds_exec_och)
1006 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_EXEC);
1007 if (lli->lli_mds_read_och)
1008 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_READ);
1010 obd_change_cbdata(sbi->ll_dt_exp, lli->lli_smd,
1011 null_if_equal, inode);
1014 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1015 lli->lli_smd = NULL;
1019 obd_free_memmd(sbi->ll_md_exp,
1020 (struct lov_stripe_md **) &lli->lli_mea);
1021 lli->lli_mea = NULL;
1024 if (lli->lli_symlink_name) {
1025 OBD_FREE(lli->lli_symlink_name,
1026 strlen(lli->lli_symlink_name) + 1);
1027 lli->lli_symlink_name = NULL;
1029 lli->lli_inode_magic = LLI_INODE_DEAD;
1034 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1035 * object(s) determine the file size and mtime. Otherwise, the MDS will
1036 * keep these values until such a time that objects are allocated for it.
1037 * We do the MDS operations first, as it is checking permissions for us.
1038 * We don't to the MDS RPC if there is nothing that we want to store there,
1039 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1040 * going to do an RPC anyways.
1042 * If we are doing a truncate, we will send the mtime and ctime updates
1043 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1044 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1047 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1049 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1050 struct ll_sb_info *sbi = ll_i2sbi(inode);
1051 struct ptlrpc_request *request = NULL;
1052 struct mdc_op_data *op_data;
1053 int ia_valid = attr->ia_valid;
1057 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu\n", inode->i_ino);
1058 lprocfs_counter_incr(ll_i2sbi(inode)->ll_stats, LPROC_LL_SETATTR);
1060 if (ia_valid & ATTR_SIZE) {
1061 if (attr->ia_size > ll_file_maxbytes(inode)) {
1062 CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1063 attr->ia_size, ll_file_maxbytes(inode));
1067 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1070 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1071 if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1072 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1076 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1077 if (attr->ia_valid & ATTR_CTIME) {
1078 attr->ia_ctime = CURRENT_TIME;
1079 attr->ia_valid |= ATTR_CTIME_SET;
1081 if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1082 attr->ia_atime = CURRENT_TIME;
1083 attr->ia_valid |= ATTR_ATIME_SET;
1085 if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1086 attr->ia_mtime = CURRENT_TIME;
1087 attr->ia_valid |= ATTR_MTIME_SET;
1090 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1091 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1092 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1093 LTIME_S(CURRENT_TIME));
1096 attr->ia_valid &= ~ATTR_SIZE;
1098 /* If only OST attributes being set on objects, don't do MDS RPC.
1099 * In that case, we need to check permissions and update the local
1100 * inode ourselves so we can call obdo_from_inode() always. */
1101 if (ia_valid & (lsm ? ~(ATTR_SIZE | ATTR_FROM_OPEN /*| ATTR_RAW*/) : ~0)) {
1102 struct lustre_md md;
1104 OBD_ALLOC(op_data, sizeof(*op_data));
1105 if (op_data == NULL)
1107 ll_prepare_mdc_data(op_data, inode, NULL, NULL, 0, 0);
1109 rc = md_setattr(sbi->ll_md_exp, op_data,
1110 attr, NULL, 0, NULL, 0, &request);
1111 OBD_FREE(op_data, sizeof(*op_data));
1113 ptlrpc_req_finished(request);
1114 if (rc != -EPERM && rc != -EACCES)
1115 CERROR("md_setattr fails: rc = %d\n", rc);
1119 rc = mdc_req2lustre_md(sbi->ll_md_exp, request, 0,
1120 sbi->ll_dt_exp, &md);
1122 ptlrpc_req_finished(request);
1126 /* We call inode_setattr to adjust timestamps, but we first
1127 * clear ATTR_SIZE to avoid invoking vmtruncate.
1129 * NB: ATTR_SIZE will only be set at this point if the size
1130 * resides on the MDS, ie, this file has no objects. */
1131 attr->ia_valid &= ~ATTR_SIZE;
1134 * assigning inode_setattr() to @err to disable warning that
1135 * function's result should be checked by by caller. error is
1136 * impossible here, as vmtruncate() control path is disabled.
1138 err = inode_setattr(inode, attr);
1139 ll_update_inode(inode, &md);
1140 ptlrpc_req_finished(request);
1142 if (!lsm || !S_ISREG(inode->i_mode)) {
1143 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1147 /* The OST doesn't check permissions, but the alternative is
1148 * a gratuitous RPC to the MDS. We already rely on the client
1149 * to do read/write/truncate permission checks, so is mtime OK?
1151 if (ia_valid & (ATTR_MTIME | ATTR_ATIME)) {
1152 /* from sys_utime() */
1153 if (!(ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET))) {
1154 if (current->fsuid != inode->i_uid &&
1155 (rc=ll_permission(inode,MAY_WRITE,NULL))!=0)
1158 /* from inode_change_ok() */
1159 if (current->fsuid != inode->i_uid &&
1160 !capable(CAP_FOWNER))
1165 /* won't invoke vmtruncate, as we already cleared ATTR_SIZE */
1166 err = inode_setattr(inode, attr);
1168 * assigning inode_setattr() to @err to disable warning that
1169 * function's result should be checked by by caller. error is
1170 * impossible here, as vmtruncate() control path is disabled.
1174 /* We really need to get our PW lock before we change inode->i_size.
1175 * If we don't we can race with other i_size updaters on our node, like
1176 * ll_file_read. We can also race with i_size propogation to other
1177 * nodes through dirtying and writeback of final cached pages. This
1178 * last one is especially bad for racing o_append users on other
1180 if (ia_valid & ATTR_SIZE) {
1181 ldlm_policy_data_t policy = { .l_extent = {attr->ia_size,
1183 struct lustre_handle lockh = { 0 };
1184 struct ll_inode_info *lli = ll_i2info(inode);
1185 int err, ast_flags = 0;
1186 /* XXX when we fix the AST intents to pass the discard-range
1187 * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
1189 if (attr->ia_size == 0)
1190 ast_flags = LDLM_AST_DISCARD_DATA;
1192 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy, &lockh,
1193 ast_flags, &ll_i2sbi(inode)->ll_seek_stime);
1198 down(&lli->lli_size_sem);
1199 lli->lli_size_pid = current->pid;
1200 rc = vmtruncate(inode, attr->ia_size);
1202 LASSERT(atomic_read(&lli->lli_size_sem.count) <= 0);
1203 lli->lli_size_pid = 0;
1204 up(&lli->lli_size_sem);
1207 err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
1209 CERROR("ll_extent_unlock failed: %d\n", err);
1213 } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
1214 struct obdo *oa = NULL;
1216 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1217 inode->i_ino, LTIME_S(attr->ia_mtime));
1223 oa->o_id = lsm->lsm_object_id;
1224 oa->o_gr = lsm->lsm_object_gr;
1225 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1226 obdo_from_inode(oa, inode, OBD_MD_FLTYPE | OBD_MD_FLATIME |
1227 OBD_MD_FLMTIME | OBD_MD_FLCTIME);
1228 rc = obd_setattr(sbi->ll_dt_exp, oa, lsm, NULL);
1231 CERROR("obd_setattr fails: rc = %d\n", rc);
1236 int ll_setattr(struct dentry *de, struct iattr *attr)
1238 LASSERT(de->d_inode);
1239 return ll_setattr_raw(de->d_inode, attr);
1242 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1243 unsigned long max_age)
1245 struct ll_sb_info *sbi = ll_s2sbi(sb);
1246 struct obd_statfs obd_osfs;
1250 rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age);
1252 CERROR("obd_statfs fails: rc = %d\n", rc);
1256 osfs->os_type = sb->s_magic;
1258 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1259 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1261 rc = obd_statfs(class_exp2obd(sbi->ll_dt_exp), &obd_osfs, max_age);
1263 CERROR("obd_statfs fails: rc = %d\n", rc);
1267 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1268 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1271 osfs->os_blocks = obd_osfs.os_blocks;
1272 osfs->os_bfree = obd_osfs.os_bfree;
1273 osfs->os_bavail = obd_osfs.os_bavail;
1275 /* If we don't have as many objects free on the OST as inodes
1276 * on the MDS, we reduce the total number of inodes to
1277 * compensate, so that the "inodes in use" number is correct.
1279 if (obd_osfs.os_ffree < osfs->os_ffree) {
1280 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1282 osfs->os_ffree = obd_osfs.os_ffree;
1288 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1290 struct obd_statfs osfs;
1293 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p\n", sb);
1294 lprocfs_counter_incr(ll_s2sbi(sb)->ll_stats, LPROC_LL_STAFS);
1296 /* For now we will always get up-to-date statfs values, but in the
1297 * future we may allow some amount of caching on the client (e.g.
1298 * from QOS or lprocfs updates). */
1299 rc = ll_statfs_internal(sb, &osfs, jiffies - 1);
1303 statfs_unpack(sfs, &osfs);
1305 if (sizeof(sfs->f_blocks) == 4) {
1306 while (osfs.os_blocks > ~0UL) {
1309 osfs.os_blocks >>= 1;
1310 osfs.os_bfree >>= 1;
1311 osfs.os_bavail >>= 1;
1315 sfs->f_blocks = osfs.os_blocks;
1316 sfs->f_bfree = osfs.os_bfree;
1317 sfs->f_bavail = osfs.os_bavail;
1322 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1324 struct ll_inode_info *lli = ll_i2info(inode);
1325 struct lov_stripe_md *lsm = md->lsm;
1326 struct mds_body *body = md->body;
1327 struct mea *mea = md->mea;
1328 struct posix_acl *ll_acl_access = md->acl_access;
1331 LASSERT((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1333 if (md->lsm && md->lsm->lsm_magic != LOV_MAGIC) {
1334 /* check for default striping info for dir. */
1335 LASSERT((mea != NULL) == ((body->valid & OBD_MD_FLDIREA) != 0));
1339 LASSERT(lsm->lsm_object_gr > 0);
1340 if (lli->lli_smd == NULL) {
1342 lli->lli_maxbytes = lsm->lsm_maxbytes;
1343 if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1344 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1347 if (memcmp(lli->lli_smd, lsm, sizeof(*lsm))) {
1348 CERROR("lsm mismatch for inode %ld\n",
1350 CERROR("lli_smd:\n");
1351 dump_lsm(D_ERROR, lli->lli_smd);
1353 dump_lsm(D_ERROR, lsm);
1356 /* XXX FIXME -- We should decide on a safer (atomic) and
1357 * more elegant way to update the lsm */
1358 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1359 lli->lli_smd->lsm_oinfo[i].loi_id =
1360 lsm->lsm_oinfo[i].loi_id;
1361 lli->lli_smd->lsm_oinfo[i].loi_gr =
1362 lsm->lsm_oinfo[i].loi_gr;
1363 lli->lli_smd->lsm_oinfo[i].loi_ost_idx =
1364 lsm->lsm_oinfo[i].loi_ost_idx;
1365 lli->lli_smd->lsm_oinfo[i].loi_ost_gen =
1366 lsm->lsm_oinfo[i].loi_ost_gen;
1369 /* bug 2844 - limit i_blksize for broken user-space apps */
1370 LASSERTF(lsm->lsm_xfersize != 0, "%lu\n", lsm->lsm_xfersize);
1371 inode->i_blksize = min(lsm->lsm_xfersize, LL_MAX_BLKSIZE);
1372 if (lli->lli_smd != lsm)
1373 obd_free_memmd(ll_i2dtexp(inode), &lsm);
1377 if (lli->lli_mea == NULL) {
1380 if (memcmp(lli->lli_mea, mea, body->eadatasize)) {
1381 CERROR("mea mismatch for inode %lu\n",
1386 if (lli->lli_mea != mea)
1387 obd_free_memmd(ll_i2mdexp(inode),
1388 (struct lov_stripe_md **) &mea);
1391 if (body->valid & OBD_MD_FID)
1392 id_assign_fid(&lli->lli_id, &body->id1);
1394 if (body->valid & OBD_MD_FLID)
1395 id_ino(&lli->lli_id) = id_ino(&body->id1);
1397 if (body->valid & OBD_MD_FLGENER)
1398 id_gen(&lli->lli_id) = id_gen(&body->id1);
1400 spin_lock(&lli->lli_lock);
1401 if (ll_acl_access != NULL) {
1402 if (lli->lli_acl_access != NULL)
1403 posix_acl_release(lli->lli_acl_access);
1404 lli->lli_acl_access = ll_acl_access;
1406 spin_unlock(&lli->lli_lock);
1408 if (body->valid & OBD_MD_FLID)
1409 inode->i_ino = id_ino(&body->id1);
1410 if (body->valid & OBD_MD_FLGENER)
1411 inode->i_generation = id_gen(&body->id1);
1412 if (body->valid & OBD_MD_FLATIME)
1413 LTIME_S(inode->i_atime) = body->atime;
1414 if (body->valid & OBD_MD_FLMTIME &&
1415 body->mtime > LTIME_S(inode->i_mtime)) {
1416 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to %u\n",
1417 inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
1418 LTIME_S(inode->i_mtime) = body->mtime;
1420 if (body->valid & OBD_MD_FLCTIME &&
1421 body->ctime > LTIME_S(inode->i_ctime))
1422 LTIME_S(inode->i_ctime) = body->ctime;
1423 if (body->valid & OBD_MD_FLMODE) {
1424 inode->i_mode = (inode->i_mode & S_IFMT) |
1425 (body->mode & ~S_IFMT);
1427 if (body->valid & OBD_MD_FLTYPE) {
1428 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1429 (body->mode & S_IFMT);
1431 if (body->valid & OBD_MD_FLUID)
1432 inode->i_uid = body->uid;
1433 if (body->valid & OBD_MD_FLGID)
1434 inode->i_gid = body->gid;
1435 if (body->valid & OBD_MD_FLFLAGS)
1436 inode->i_flags = body->flags;
1437 if (body->valid & OBD_MD_FLNLINK)
1438 inode->i_nlink = body->nlink;
1439 if (body->valid & OBD_MD_FLRDEV)
1440 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1441 inode->i_rdev = body->rdev;
1443 inode->i_rdev = old_decode_dev(body->rdev);
1445 if (body->valid & OBD_MD_FLSIZE)
1446 inode->i_size = body->size;
1447 if (body->valid & OBD_MD_FLBLOCKS)
1448 inode->i_blocks = body->blocks;
1450 if (body->valid & OBD_MD_FLSIZE)
1451 set_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &lli->lli_flags);
1453 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1454 inode->i_dev = (kdev_t)id_group(&lli->lli_id);
1456 LASSERT(id_fid(&lli->lli_id) != 0);
1459 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
1460 static struct backing_dev_info ll_backing_dev_info = {
1461 .ra_pages = 0, /* No readahead */
1462 .memory_backed = 0, /* Does contribute to dirty memory */
1466 void ll_read_inode2(struct inode *inode, void *opaque)
1468 struct lustre_md *md = opaque;
1469 struct ll_inode_info *lli = ll_i2info(inode);
1472 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1473 inode->i_generation, inode);
1477 LASSERT(!lli->lli_smd);
1479 /* Core attributes from the MDS first. This is a new inode, and
1480 * the VFS doesn't zero times in the core inode so we have to do
1481 * it ourselves. They will be overwritten by either MDS or OST
1482 * attributes - we just need to make sure they aren't newer. */
1483 LTIME_S(inode->i_mtime) = 0;
1484 LTIME_S(inode->i_atime) = 0;
1485 LTIME_S(inode->i_ctime) = 0;
1488 ll_update_inode(inode, md);
1490 /* OIDEBUG(inode); */
1492 if (S_ISREG(inode->i_mode)) {
1493 inode->i_op = &ll_file_inode_operations;
1494 inode->i_fop = &ll_file_operations;
1495 inode->i_mapping->a_ops = &ll_aops;
1497 } else if (S_ISDIR(inode->i_mode)) {
1498 inode->i_op = &ll_dir_inode_operations;
1499 inode->i_fop = &ll_dir_operations;
1500 inode->i_mapping->a_ops = &ll_dir_aops;
1502 } else if (S_ISLNK(inode->i_mode)) {
1503 inode->i_op = &ll_fast_symlink_inode_operations;
1506 inode->i_op = &ll_special_inode_operations;
1508 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
1509 init_special_inode(inode, inode->i_mode,
1510 kdev_t_to_nr(inode->i_rdev));
1512 /* initializing backing dev info. */
1513 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
1515 init_special_inode(inode, inode->i_mode, inode->i_rdev);
1517 lli->ll_save_ifop = inode->i_fop;
1519 if (S_ISCHR(inode->i_mode))
1520 inode->i_fop = &ll_special_chr_inode_fops;
1521 else if (S_ISBLK(inode->i_mode))
1522 inode->i_fop = &ll_special_blk_inode_fops;
1523 else if (S_ISFIFO(inode->i_mode))
1524 inode->i_fop = &ll_special_fifo_inode_fops;
1525 else if (S_ISSOCK(inode->i_mode))
1526 inode->i_fop = &ll_special_sock_inode_fops;
1528 CWARN("saved %p, replaced with %p\n", lli->ll_save_ifop,
1531 if (lli->ll_save_ifop->owner) {
1532 CWARN("%p has owner %p\n", lli->ll_save_ifop,
1533 lli->ll_save_ifop->owner);
1539 void ll_delete_inode(struct inode *inode)
1541 struct ll_sb_info *sbi = ll_i2sbi(inode);
1542 struct lustre_id id;
1546 ll_inode2id(&id, inode);
1548 rc = md_delete_inode(sbi->ll_md_exp, &id);
1550 CERROR("md_delete_inode() failed, error %d\n",
1558 int ll_iocontrol(struct inode *inode, struct file *file,
1559 unsigned int cmd, unsigned long arg)
1561 struct ll_sb_info *sbi = ll_i2sbi(inode);
1562 struct ptlrpc_request *req = NULL;
1567 case EXT3_IOC_GETFLAGS: {
1568 struct lustre_id id;
1569 __u64 valid = OBD_MD_FLFLAGS;
1570 struct mds_body *body;
1572 ll_inode2id(&id, inode);
1573 rc = md_getattr(sbi->ll_md_exp, &id, valid, NULL, 0, 0, &req);
1575 CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1579 body = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*body));
1581 if (body->flags & S_APPEND)
1582 flags |= EXT3_APPEND_FL;
1583 if (body->flags & S_IMMUTABLE)
1584 flags |= EXT3_IMMUTABLE_FL;
1585 if (body->flags & S_NOATIME)
1586 flags |= EXT3_NOATIME_FL;
1588 ptlrpc_req_finished (req);
1590 RETURN(put_user(flags, (int *)arg));
1592 case EXT3_IOC_SETFLAGS: {
1593 struct mdc_op_data *op_data;
1596 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1598 if (get_user(flags, (int *)arg))
1605 OBD_ALLOC(op_data, sizeof(*op_data));
1606 if (op_data == NULL) {
1610 ll_prepare_mdc_data(op_data, inode, NULL, NULL, 0, 0);
1612 memset(&attr, 0x0, sizeof(attr));
1613 attr.ia_attr_flags = flags;
1614 attr.ia_valid |= ATTR_ATTR_FLAG;
1616 rc = md_setattr(sbi->ll_md_exp, op_data,
1617 &attr, NULL, 0, NULL, 0, &req);
1618 OBD_FREE(op_data, sizeof(*op_data));
1620 ptlrpc_req_finished(req);
1621 if (rc != -EPERM && rc != -EACCES)
1622 CERROR("md_setattr fails: rc = %d\n", rc);
1626 ptlrpc_req_finished(req);
1628 oa->o_id = lsm->lsm_object_id;
1629 oa->o_gr = lsm->lsm_object_gr;
1630 oa->o_flags = flags;
1631 oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS | OBD_MD_FLGROUP;
1633 rc = obd_setattr(sbi->ll_dt_exp, oa, lsm, NULL);
1636 if (rc != -EPERM && rc != -EACCES)
1637 CERROR("md_setattr fails: rc = %d\n", rc);
1641 if (flags & EXT3_APPEND_FL)
1642 inode->i_flags |= S_APPEND;
1644 inode->i_flags &= ~S_APPEND;
1645 if (flags & EXT3_IMMUTABLE_FL)
1646 inode->i_flags |= S_IMMUTABLE;
1648 inode->i_flags &= ~S_IMMUTABLE;
1649 if (flags & EXT3_NOATIME_FL)
1650 inode->i_flags |= S_NOATIME;
1652 inode->i_flags &= ~S_NOATIME;
1663 /* this is only called in the case of forced umount. */
1664 void ll_umount_begin(struct super_block *sb)
1666 struct ll_sb_info *sbi = ll_s2sbi(sb);
1667 struct obd_ioctl_data ioc_data = { 0 };
1668 struct obd_device *obd;
1671 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1672 sb->s_count, atomic_read(&sb->s_active));
1674 obd = class_exp2obd(sbi->ll_md_exp);
1676 CERROR("Invalid MDC connection handle "LPX64"\n",
1677 sbi->ll_md_exp->exp_handle.h_cookie);
1681 obd->obd_no_recov = 1;
1682 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
1683 sizeof(ioc_data), &ioc_data, NULL);
1685 obd = class_exp2obd(sbi->ll_dt_exp);
1687 CERROR("Invalid LOV connection handle "LPX64"\n",
1688 sbi->ll_dt_exp->exp_handle.h_cookie);
1693 obd->obd_no_recov = 1;
1694 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
1695 sizeof(ioc_data), &ioc_data, NULL);
1698 * really, we'd like to wait until there are no requests outstanding,
1699 * and then continue. For now, we just invalidate the requests,
1700 * schedule, and hope.
1707 int ll_prep_inode(struct obd_export *dt_exp, struct obd_export *md_exp,
1708 struct inode **inode, struct ptlrpc_request *req,
1709 int offset, struct super_block *sb)
1711 struct lustre_md md;
1714 rc = mdc_req2lustre_md(md_exp, req, offset, dt_exp, &md);
1719 ll_update_inode(*inode, &md);
1722 *inode = ll_iget(sb, id_ino(&md.body->id1), &md);
1723 if (*inode == NULL || is_bad_inode(*inode)) {
1724 /* free the lsm if we allocated one above */
1726 obd_free_memmd(dt_exp, &md.lsm);
1728 obd_free_memmd(md_exp,
1729 (struct lov_stripe_md**)&md.mea);
1731 CERROR("new_inode -fatal: rc %d\n", rc);
1738 int ll_get_fid(struct obd_export *exp, struct lustre_id *idp,
1739 char *filename, struct lustre_id *ret)
1741 struct ptlrpc_request *request = NULL;
1742 struct mds_body *body;
1745 rc = md_getattr_lock(exp, idp, filename, strlen(filename) + 1,
1746 OBD_MD_FID, 0, &request);
1748 CDEBUG(D_INFO, "md_getattr_lock failed on %s: rc %d\n",
1753 body = lustre_msg_buf(request->rq_repmsg, 0, sizeof(*body));
1754 LASSERT(body != NULL);
1755 LASSERT_REPSWABBED(request, 0);
1758 ptlrpc_req_finished(request);