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 *mds_security, char *oss_security,
133 __u32 *nllu, __u64 *remote)
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 (mds_security == NULL)
172 mds_security = "null";
174 err = obd_set_info(obd->obd_self_export, strlen("sec"), "sec",
175 strlen(mds_security), mds_security);
177 CERROR("LMV %s: failed to set security %s, err %d\n",
178 lmv, mds_security, err);
179 OBD_FREE(data, sizeof(*data));
183 if (proc_lustre_fs_root) {
184 err = lprocfs_register_mountpoint(proc_lustre_fs_root,
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 if (oss_security == NULL)
241 oss_security = "null";
243 err = obd_set_info(obd->obd_self_export, strlen("sec"), "sec",
244 strlen(oss_security), oss_security);
246 CERROR("LOV %s: failed to set security %s, err %d\n",
247 lov, oss_security, err);
248 OBD_FREE(data, sizeof(*data));
252 err = obd_connect(&dt_conn, obd, &sbi->ll_sb_uuid, data,
253 OBD_OPT_REAL_CLIENT);
255 CERROR("An OST (lov %s) is performing recovery, of which this"
256 " client is not a part. Please wait for recovery to "
257 "complete, abort, or time out.\n", lov);
260 CERROR("cannot connect to %s: rc = %d\n", lov, err);
263 sbi->ll_dt_exp = class_conn2export(&dt_conn);
265 err = lustre_init_dt_desc(sbi);
267 int mdsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
268 obd_init_ea_size(sbi->ll_md_exp, mdsize,
269 sbi->ll_dt_desc.ld_tgt_count *
270 sizeof(struct llog_cookie));
273 err = md_getstatus(sbi->ll_md_exp, &sbi->ll_rootid);
275 CERROR("cannot mds_connect: rc = %d\n", err);
278 CDEBUG(D_SUPER, "rootid "DLID4"\n", OLID4(&sbi->ll_rootid));
280 sb->s_op = &lustre_super_operations;
282 /* make root inode */
283 err = md_getattr(sbi->ll_md_exp, &sbi->ll_rootid,
284 (OBD_MD_FLNOTOBD | OBD_MD_FLBLOCKS | OBD_MD_FID),
285 NULL, 0, 0, &request);
287 CERROR("md_getattr failed for root: rc = %d\n", err);
291 err = mdc_req2lustre_md(sbi->ll_md_exp, request, 0,
292 sbi->ll_dt_exp, &md);
294 CERROR("failed to understand root inode md: rc = %d\n", err);
295 ptlrpc_req_finished(request);
299 LASSERT(id_ino(&sbi->ll_rootid) != 0);
300 root = ll_iget(sb, id_ino(&sbi->ll_rootid), &md);
302 ptlrpc_req_finished(request);
304 if (root == NULL || is_bad_inode(root)) {
306 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
308 obd_free_memmd(sbi->ll_md_exp,
309 (struct lov_stripe_md**)&md.mea);
310 CERROR("lustre_lite: bad iget4 for root\n");
311 GOTO(out_root, err = -EBADF);
314 err = ll_close_thread_start(&sbi->ll_lcq);
316 CERROR("cannot start close thread: rc %d\n", err);
320 ll_gns_add_timer(sbi);
322 /* making vm readahead 0 for 2.4.x. In the case of 2.6.x,
323 backing dev info assigned to inode mapping is used for
324 determining maximal readahead. */
325 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0)) && \
326 !defined(KERNEL_HAS_AS_MAX_READAHEAD)
327 /* bug 2805 - set VM readahead to zero */
328 vm_max_readahead = vm_min_readahead = 0;
331 sb->s_root = d_alloc_root(root);
332 sb->s_root->d_op = &ll_d_ops;
334 sb->s_flags |= MS_POSIXACL;
336 CWARN("Enabling PDIROPS\n");
337 sb->s_flags |= S_PDIROPS;
341 OBD_FREE(data, sizeof(*data));
347 obd_disconnect(sbi->ll_dt_exp, 0);
349 obd_disconnect(sbi->ll_md_exp, 0);
352 OBD_FREE(data, sizeof(*data));
353 lprocfs_unregister_mountpoint(sbi);
357 void lustre_common_put_super(struct super_block *sb)
359 struct ll_sb_info *sbi = ll_s2sbi(sb);
360 struct hlist_node *tmp, *next;
363 ll_gns_del_timer(sbi);
364 ll_close_thread_shutdown(sbi->ll_lcq);
366 list_del(&sbi->ll_conn_chain);
367 obd_disconnect(sbi->ll_dt_exp, 0);
369 lprocfs_unregister_mountpoint(sbi);
370 if (sbi->ll_proc_root) {
371 lprocfs_remove(sbi->ll_proc_root);
372 sbi->ll_proc_root = NULL;
375 obd_disconnect(sbi->ll_md_exp, 0);
377 // We do this to get rid of orphaned dentries. That is not really trw.
378 hlist_for_each_safe(tmp, next, &sbi->ll_orphan_dentry_list) {
379 struct dentry *dentry = hlist_entry(tmp, struct dentry, d_hash);
380 CWARN("orphan dentry %.*s (%p->%p) at unmount\n",
381 dentry->d_name.len, dentry->d_name.name, dentry, next);
382 shrink_dcache_parent(dentry);
387 char *ll_read_opt(const char *opt, char *data)
393 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
394 if (strncmp(opt, data, strlen(opt)))
396 if ((value = strchr(data, '=')) == NULL)
400 OBD_ALLOC(retval, strlen(value) + 1);
402 CERROR("out of memory!\n");
406 memcpy(retval, value, strlen(value)+1);
407 CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
411 int ll_set_opt(const char *opt, char *data, int fl)
415 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
416 if (strncmp(opt, data, strlen(opt)))
422 void ll_options(char *options, char **lov, char **lmv, char **mds_sec,
423 char **oss_sec, int *async, int *flags)
426 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
427 char *opt_ptr = options;
437 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
438 for (this_char = strtok (options, ",");
440 this_char = strtok (NULL, ",")) {
442 while ((this_char = strsep (&opt_ptr, ",")) != NULL) {
444 CDEBUG(D_SUPER, "this_char %s\n", this_char);
445 if (!*lov && (*lov = ll_read_opt("osc", this_char)))
447 if (!*lmv && (*lmv = ll_read_opt("mdc", this_char)))
449 if (!strncmp(this_char, "lasync", strlen("lasync"))) {
453 if (!*mds_sec && (*mds_sec = ll_read_opt("mds_sec", this_char)))
455 if (!*oss_sec && (*oss_sec = ll_read_opt("oss_sec", this_char)))
457 if (!(*flags & LL_SBI_NOLCK) &&
458 ((*flags) = (*flags) |
459 ll_set_opt("nolock", this_char,
467 void ll_lli_init(struct ll_inode_info *lli)
469 sema_init(&lli->lli_open_sem, 1);
470 sema_init(&lli->lli_size_sem, 1);
472 lli->lli_size_pid = 0;
473 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
474 spin_lock_init(&lli->lli_lock);
475 INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
476 lli->lli_inode_magic = LLI_INODE_MAGIC;
477 memset(&lli->lli_id, 0, sizeof(lli->lli_id));
478 sema_init(&lli->lli_och_sem, 1);
479 lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
480 lli->lli_mds_exec_och = NULL;
481 lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
482 lli->lli_open_fd_exec_count = 0;
485 int ll_fill_super(struct super_block *sb, void *data, int silent)
487 struct ll_sb_info *sbi;
490 char *mds_sec = NULL;
491 char *oss_sec = NULL;
493 __u32 nllu[2] = { NOBODY_UID, NOBODY_GID };
494 __u64 remote_flag = 0;
497 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
499 sbi = lustre_init_sbi(sb);
503 sbi->ll_flags |= LL_SBI_READAHEAD;
504 ll_options(data, &lov, &lmv, &mds_sec, &oss_sec,
505 &async, &sbi->ll_flags);
509 GOTO(out, err = -EINVAL);
514 GOTO(out, err = -EINVAL);
517 err = lustre_common_fill_super(sb, lmv, lov, async, mds_sec, oss_sec,
525 OBD_FREE(lmv, strlen(lmv) + 1);
527 OBD_FREE(lov, strlen(lov) + 1);
529 OBD_FREE(mds_sec, strlen(mds_sec) + 1);
531 OBD_FREE(oss_sec, strlen(oss_sec) + 1);
534 } /* ll_read_super */
536 static int lustre_process_log(struct lustre_mount_data *lmd, char *profile,
537 struct config_llog_instance *cfg, int allow_recov)
539 struct lustre_cfg *lcfg = NULL;
540 struct lustre_cfg_bufs bufs;
541 struct portals_cfg pcfg;
542 char *peer = "MDS_PEER_UUID";
543 struct obd_device *obd;
544 struct lustre_handle md_conn = {0, };
545 struct obd_export *exp;
546 char *name = "mdc_dev";
548 struct obd_uuid lmv_uuid;
549 struct llog_ctxt *ctxt;
553 if (lmd_bad_magic(lmd))
556 generate_random_uuid(uuid);
557 class_uuid_unparse(uuid, &lmv_uuid);
559 if (lmd->lmd_local_nid) {
560 PCFG_INIT(pcfg, NAL_CMD_REGISTER_MYNID);
561 pcfg.pcfg_nal = lmd->lmd_nal;
562 pcfg.pcfg_nid = lmd->lmd_local_nid;
563 rc = libcfs_nal_cmd(&pcfg);
568 if (lmd->lmd_nal == SOCKNAL ||
569 lmd->lmd_nal == OPENIBNAL ||
570 lmd->lmd_nal == IIBNAL ||
571 lmd->lmd_nal == VIBNAL ||
572 lmd->lmd_nal == RANAL) {
573 PCFG_INIT(pcfg, NAL_CMD_ADD_PEER);
574 pcfg.pcfg_nal = lmd->lmd_nal;
575 pcfg.pcfg_nid = lmd->lmd_server_nid;
576 pcfg.pcfg_id = lmd->lmd_server_ipaddr;
577 pcfg.pcfg_misc = lmd->lmd_port;
578 rc = libcfs_nal_cmd(&pcfg);
582 lustre_cfg_bufs_reset(&bufs, name);
583 lustre_cfg_bufs_set_string(&bufs, 1, peer);
585 lcfg = lustre_cfg_new(LCFG_ADD_UUID, &bufs);
586 lcfg->lcfg_nal = lmd->lmd_nal;
587 lcfg->lcfg_nid = lmd->lmd_server_nid;
588 LASSERT(lcfg->lcfg_nal);
589 LASSERT(lcfg->lcfg_nid);
590 err = class_process_config(lcfg);
591 lustre_cfg_free(lcfg);
593 GOTO(out_del_conn, err);
595 lustre_cfg_bufs_reset(&bufs, name);
596 lustre_cfg_bufs_set_string(&bufs, 1, LUSTRE_MDC_NAME);
597 lustre_cfg_bufs_set_string(&bufs, 2, (char *)lmv_uuid.uuid);
599 lcfg = lustre_cfg_new(LCFG_ATTACH, &bufs);
600 err = class_process_config(lcfg);
601 lustre_cfg_free(lcfg);
603 GOTO(out_del_uuid, err);
605 lustre_cfg_bufs_reset(&bufs, name);
606 lustre_cfg_bufs_set_string(&bufs, 1, lmd->lmd_mds);
607 lustre_cfg_bufs_set_string(&bufs, 2, peer);
609 lcfg = lustre_cfg_new(LCFG_SETUP, &bufs);
610 err = class_process_config(lcfg);
611 lustre_cfg_free(lcfg);
613 GOTO(out_detach, err);
615 obd = class_name2obd(name);
617 GOTO(out_cleanup, rc = -EINVAL);
619 rc = obd_set_info(obd->obd_self_export, strlen("sec"), "sec",
620 strlen(lmd->lmd_mds_security), lmd->lmd_mds_security);
622 GOTO(out_cleanup, rc);
624 /* Disable initial recovery on this import */
625 rc = obd_set_info(obd->obd_self_export,
626 strlen("initial_recov"), "initial_recov",
627 sizeof(allow_recov), &allow_recov);
629 GOTO(out_cleanup, rc);
631 rc = obd_connect(&md_conn, obd, &lmv_uuid, NULL, 0);
633 CERROR("cannot connect to %s: rc = %d\n", lmd->lmd_mds, rc);
634 GOTO(out_cleanup, rc);
637 exp = class_conn2export(&md_conn);
639 ctxt = llog_get_context(&exp->exp_obd->obd_llogs,LLOG_CONFIG_REPL_CTXT);
640 rc = class_config_process_llog(ctxt, profile, cfg);
642 CERROR("class_config_process_llog failed: rc = %d\n", rc);
644 err = obd_disconnect(exp, 0);
648 lustre_cfg_bufs_reset(&bufs, name);
649 lcfg = lustre_cfg_new(LCFG_CLEANUP, &bufs);
650 err = class_process_config(lcfg);
651 lustre_cfg_free(lcfg);
655 lustre_cfg_bufs_reset(&bufs, name);
656 lcfg = lustre_cfg_new(LCFG_DETACH, &bufs);
657 err = class_process_config(lcfg);
658 lustre_cfg_free(lcfg);
663 lustre_cfg_bufs_reset(&bufs, name);
664 lustre_cfg_bufs_set_string(&bufs, 1, peer);
665 lcfg = lustre_cfg_new(LCFG_DEL_UUID, &bufs);
666 err = class_process_config(lcfg);
667 lustre_cfg_free(lcfg);
670 if (lmd->lmd_nal == SOCKNAL ||
671 lmd->lmd_nal == OPENIBNAL ||
672 lmd->lmd_nal == IIBNAL ||
673 lmd->lmd_nal == VIBNAL ||
674 lmd->lmd_nal == RANAL) {
677 PCFG_INIT(pcfg, NAL_CMD_DEL_PEER);
678 pcfg.pcfg_nal = lmd->lmd_nal;
679 pcfg.pcfg_nid = lmd->lmd_server_nid;
680 pcfg.pcfg_flags = 1; /* single_share */
681 err2 = libcfs_nal_cmd(&pcfg);
694 static void lustre_manual_cleanup(struct ll_sb_info *sbi)
696 struct lustre_cfg *lcfg;
697 struct lustre_cfg_bufs bufs;
698 struct obd_device *obd;
701 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) != NULL)
705 lustre_cfg_bufs_reset(&bufs, obd->obd_name);
706 lcfg = lustre_cfg_new(LCFG_CLEANUP, &bufs);
707 err = class_process_config(lcfg);
709 CERROR("cleanup failed: %s\n", obd->obd_name);
713 lcfg->lcfg_command = LCFG_DETACH;
714 err = class_process_config(lcfg);
715 lustre_cfg_free(lcfg);
717 CERROR("detach failed: %s\n", obd->obd_name);
722 if (sbi->ll_lmd != NULL)
723 class_del_profile(sbi->ll_lmd->lmd_profile);
726 int lustre_fill_super(struct super_block *sb, void *data, int silent)
728 struct lustre_mount_data * lmd = data;
729 char *lov = NULL, *lmv = NULL;
730 struct ll_sb_info *sbi;
734 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
735 if (lmd_bad_magic(lmd))
738 sbi = lustre_init_sbi(sb);
742 sbi->ll_flags |= LL_SBI_READAHEAD;
744 if (lmd->lmd_profile) {
745 struct lustre_profile *lprof;
746 struct config_llog_instance cfg;
749 if (lmd->lmd_mds[0] == '\0') {
750 CERROR("no mds name\n");
751 GOTO(out_free, err = -EINVAL);
753 lmd->lmd_mds_security[sizeof(lmd->lmd_mds_security) - 1] = 0;
754 lmd->lmd_oss_security[sizeof(lmd->lmd_oss_security) - 1] = 0;
756 OBD_ALLOC(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
757 if (sbi->ll_lmd == NULL)
758 GOTO(out_free, err = -ENOMEM);
759 memcpy(sbi->ll_lmd, lmd, sizeof(*lmd));
761 /* generate a string unique to this super, let's try
762 the address of the super itself.*/
763 len = (sizeof(sb) * 2) + 1;
764 OBD_ALLOC(sbi->ll_instance, len);
765 if (sbi->ll_instance == NULL)
766 GOTO(out_free, err = -ENOMEM);
767 sprintf(sbi->ll_instance, "%p", sb);
769 cfg.cfg_instance = sbi->ll_instance;
770 cfg.cfg_uuid = sbi->ll_sb_uuid;
771 cfg.cfg_local_nid = lmd->lmd_local_nid;
772 err = lustre_process_log(lmd, lmd->lmd_profile, &cfg, 0);
774 CERROR("Unable to process log: %s\n", lmd->lmd_profile);
778 lprof = class_get_profile(lmd->lmd_profile);
780 CERROR("No profile found: %s\n", lmd->lmd_profile);
781 GOTO(out_free, err = -EINVAL);
784 OBD_FREE(lov, strlen(lov) + 1);
785 OBD_ALLOC(lov, strlen(lprof->lp_lov) +
786 strlen(sbi->ll_instance) + 2);
787 sprintf(lov, "%s-%s", lprof->lp_lov, sbi->ll_instance);
790 OBD_FREE(lmv, strlen(lmv) + 1);
791 OBD_ALLOC(lmv, strlen(lprof->lp_lmv) +
792 strlen(sbi->ll_instance) + 2);
793 sprintf(lmv, "%s-%s", lprof->lp_lmv, sbi->ll_instance);
798 GOTO(out_free, err = -EINVAL);
803 GOTO(out_free, err = -EINVAL);
806 err = lustre_common_fill_super(sb, lmv, lov, lmd->lmd_async,
807 lmd->lmd_mds_security,
808 lmd->lmd_oss_security,
809 &lmd->lmd_nllu, &lmd->lmd_remote_flag);
816 OBD_FREE(lmv, strlen(lmv) + 1);
818 OBD_FREE(lov, strlen(lov) + 1);
824 int len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean")+1;
827 if (sbi->ll_instance != NULL) {
828 struct lustre_mount_data *lmd = sbi->ll_lmd;
829 struct config_llog_instance cfg;
832 cfg.cfg_instance = sbi->ll_instance;
833 cfg.cfg_uuid = sbi->ll_sb_uuid;
835 OBD_ALLOC(cl_prof, len);
836 sprintf(cl_prof, "%s-clean", lmd->lmd_profile);
837 err = lustre_process_log(lmd, cl_prof, &cfg, 0);
839 CERROR("Unable to process log: %s\n", cl_prof);
840 lustre_manual_cleanup(sbi);
842 OBD_FREE(cl_prof, len);
843 OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance) + 1);
845 OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
849 } /* lustre_fill_super */
851 void lustre_put_super(struct super_block *sb)
853 struct obd_device *obd;
854 struct ll_sb_info *sbi = ll_s2sbi(sb);
855 int force_umount = 0;
858 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
859 obd = class_exp2obd(sbi->ll_md_exp);
861 force_umount = obd->obd_no_recov;
864 lustre_common_put_super(sb);
865 if (sbi->ll_lmd != NULL) {
867 int len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean")+1;
869 struct config_llog_instance cfg;
872 CERROR("force umount, doing manual cleanup\n");
873 lustre_manual_cleanup(sbi);
877 cfg.cfg_instance = sbi->ll_instance;
878 cfg.cfg_uuid = sbi->ll_sb_uuid;
880 OBD_ALLOC(cl_prof, len);
881 sprintf(cl_prof, "%s-clean", sbi->ll_lmd->lmd_profile);
882 err = lustre_process_log(sbi->ll_lmd, cl_prof, &cfg, 0);
884 CERROR("Unable to process log: %s, doing manual cleanup"
886 lustre_manual_cleanup(sbi);
889 OBD_FREE(cl_prof, len);
891 OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
892 OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance) + 1);
898 } /* lustre_put_super */
900 int ll_process_config_update(struct ll_sb_info *sbi, int clean)
902 struct lustre_mount_data *lmd = sbi->ll_lmd;
903 char *profile = lmd->lmd_profile, *name = NULL;
904 struct config_llog_instance cfg;
905 int rc, namelen = 0, version;
906 struct llog_ctxt *ctxt;
912 CERROR("Client not mounted with zero-conf; cannot "
913 "process update log.\n");
917 rc = obd_cancel_unused(sbi->ll_md_exp, NULL,
918 LDLM_FL_CONFIG_CHANGE, NULL);
920 CWARN("obd_cancel_unused(mdc): %d\n", rc);
922 rc = obd_cancel_unused(sbi->ll_dt_exp, NULL,
923 LDLM_FL_CONFIG_CHANGE, NULL);
925 CWARN("obd_cancel_unused(lov): %d\n", rc);
927 cfg.cfg_instance = sbi->ll_instance;
928 cfg.cfg_uuid = sbi->ll_sb_uuid;
929 cfg.cfg_local_nid = lmd->lmd_local_nid;
931 namelen = strlen(profile) + 20; /* -clean-######### */
932 OBD_ALLOC(name, namelen);
937 version = sbi->ll_config_version - 1;
938 sprintf(name, "%s-clean-%d", profile, version);
940 version = sbi->ll_config_version + 1;
941 sprintf(name, "%s-%d", profile, version);
944 CWARN("Applying configuration log %s\n", name);
946 ctxt = llog_get_context(&sbi->ll_md_exp->exp_obd->obd_llogs,
947 LLOG_CONFIG_REPL_CTXT);
948 rc = class_config_process_llog(ctxt, name, &cfg);
950 sbi->ll_config_version = version;
951 CWARN("Finished applying configuration log %s: %d\n", name, rc);
953 if (rc == 0 && clean == 0) {
954 struct lov_desc desc;
958 valsize = sizeof(desc);
959 rc = obd_get_info(sbi->ll_dt_exp, strlen("lovdesc") + 1,
960 "lovdesc", &valsize, &desc);
962 rc = obd_init_ea_size(sbi->ll_md_exp,
963 obd_size_diskmd(sbi->ll_dt_exp, NULL),
965 sizeof(struct llog_cookie)));
967 OBD_FREE(name, namelen);
971 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
973 struct inode *inode = NULL;
974 l_lock(&lock->l_resource->lr_namespace->ns_lock);
975 if (lock->l_ast_data) {
976 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
977 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
978 inode = igrab(lock->l_ast_data);
980 inode = lock->l_ast_data;
981 CDEBUG(inode->i_state & I_FREEING ? D_INFO : D_WARNING,
982 "l_ast_data %p is bogus: magic %0x8\n",
983 lock->l_ast_data, lli->lli_inode_magic);
987 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
991 int null_if_equal(struct ldlm_lock *lock, void *data)
993 if (data == lock->l_ast_data) {
994 lock->l_ast_data = NULL;
996 if (lock->l_req_mode != lock->l_granted_mode)
997 LDLM_ERROR(lock,"clearing inode with ungranted lock\n");
1000 return LDLM_ITER_CONTINUE;
1003 void ll_clear_inode(struct inode *inode)
1005 struct lustre_id id;
1006 struct ll_inode_info *lli = ll_i2info(inode);
1007 struct ll_sb_info *sbi = ll_i2sbi(inode);
1010 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1011 inode->i_generation, inode);
1013 lli->lli_inode_magic = LLI_INODE_DEAD;
1014 ll_inode2id(&id, inode);
1016 clear_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &(ll_i2info(inode)->lli_flags));
1017 md_change_cbdata(sbi->ll_md_exp, &id, null_if_equal, inode);
1019 LASSERT(!lli->lli_open_fd_write_count);
1020 LASSERT(!lli->lli_open_fd_read_count);
1021 LASSERT(!lli->lli_open_fd_exec_count);
1022 if (lli->lli_mds_write_och)
1023 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_WRITE);
1024 if (lli->lli_mds_exec_och)
1025 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_EXEC);
1026 if (lli->lli_mds_read_och)
1027 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_READ);
1029 obd_change_cbdata(sbi->ll_dt_exp, lli->lli_smd,
1030 null_if_equal, inode);
1033 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1034 lli->lli_smd = NULL;
1038 obd_free_memmd(sbi->ll_md_exp,
1039 (struct lov_stripe_md **) &lli->lli_mea);
1040 lli->lli_mea = NULL;
1043 if (lli->lli_symlink_name) {
1044 OBD_FREE(lli->lli_symlink_name,
1045 strlen(lli->lli_symlink_name) + 1);
1046 lli->lli_symlink_name = NULL;
1048 lli->lli_inode_magic = LLI_INODE_DEAD;
1053 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1054 * object(s) determine the file size and mtime. Otherwise, the MDS will
1055 * keep these values until such a time that objects are allocated for it.
1056 * We do the MDS operations first, as it is checking permissions for us.
1057 * We don't to the MDS RPC if there is nothing that we want to store there,
1058 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1059 * going to do an RPC anyways.
1061 * If we are doing a truncate, we will send the mtime and ctime updates
1062 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1063 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1066 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1068 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1069 struct ll_sb_info *sbi = ll_i2sbi(inode);
1070 struct ptlrpc_request *request = NULL;
1071 struct mdc_op_data *op_data;
1072 int ia_valid = attr->ia_valid;
1076 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu\n", inode->i_ino);
1077 lprocfs_counter_incr(ll_i2sbi(inode)->ll_stats, LPROC_LL_SETATTR);
1079 if (ia_valid & ATTR_SIZE) {
1080 if (attr->ia_size > ll_file_maxbytes(inode)) {
1081 CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1082 attr->ia_size, ll_file_maxbytes(inode));
1086 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1089 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1090 if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1091 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1095 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1096 if (attr->ia_valid & ATTR_CTIME) {
1097 attr->ia_ctime = CURRENT_TIME;
1098 attr->ia_valid |= ATTR_CTIME_SET;
1100 if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1101 attr->ia_atime = CURRENT_TIME;
1102 attr->ia_valid |= ATTR_ATIME_SET;
1104 if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1105 attr->ia_mtime = CURRENT_TIME;
1106 attr->ia_valid |= ATTR_MTIME_SET;
1109 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1110 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1111 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1112 LTIME_S(CURRENT_TIME));
1115 attr->ia_valid &= ~ATTR_SIZE;
1117 /* If only OST attributes being set on objects, don't do MDS RPC.
1118 * In that case, we need to check permissions and update the local
1119 * inode ourselves so we can call obdo_from_inode() always. */
1120 if (ia_valid & (lsm ? ~(ATTR_SIZE | ATTR_FROM_OPEN /*| ATTR_RAW*/) : ~0)) {
1121 struct lustre_md md;
1123 OBD_ALLOC(op_data, sizeof(*op_data));
1124 if (op_data == NULL)
1126 ll_prepare_mdc_data(op_data, inode, NULL, NULL, 0, 0);
1128 rc = md_setattr(sbi->ll_md_exp, op_data,
1129 attr, NULL, 0, NULL, 0, &request);
1130 OBD_FREE(op_data, sizeof(*op_data));
1132 ptlrpc_req_finished(request);
1133 if (rc != -EPERM && rc != -EACCES)
1134 CERROR("md_setattr fails: rc = %d\n", rc);
1138 rc = mdc_req2lustre_md(sbi->ll_md_exp, request, 0,
1139 sbi->ll_dt_exp, &md);
1141 ptlrpc_req_finished(request);
1145 /* We call inode_setattr to adjust timestamps, but we first
1146 * clear ATTR_SIZE to avoid invoking vmtruncate.
1148 * NB: ATTR_SIZE will only be set at this point if the size
1149 * resides on the MDS, ie, this file has no objects. */
1150 attr->ia_valid &= ~ATTR_SIZE;
1153 * assigning inode_setattr() to @err to disable warning that
1154 * function's result should be checked by by caller. error is
1155 * impossible here, as vmtruncate() control path is disabled.
1157 err = inode_setattr(inode, attr);
1158 ll_update_inode(inode, &md);
1159 ptlrpc_req_finished(request);
1161 if (!lsm || !S_ISREG(inode->i_mode)) {
1162 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1166 /* The OST doesn't check permissions, but the alternative is
1167 * a gratuitous RPC to the MDS. We already rely on the client
1168 * to do read/write/truncate permission checks, so is mtime OK?
1170 if (ia_valid & (ATTR_MTIME | ATTR_ATIME)) {
1171 /* from sys_utime() */
1172 if (!(ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET))) {
1173 if (current->fsuid != inode->i_uid &&
1174 (rc=ll_permission(inode,MAY_WRITE,NULL))!=0)
1177 /* from inode_change_ok() */
1178 if (current->fsuid != inode->i_uid &&
1179 !capable(CAP_FOWNER))
1184 /* won't invoke vmtruncate, as we already cleared ATTR_SIZE */
1185 err = inode_setattr(inode, attr);
1187 * assigning inode_setattr() to @err to disable warning that
1188 * function's result should be checked by by caller. error is
1189 * impossible here, as vmtruncate() control path is disabled.
1193 /* We really need to get our PW lock before we change inode->i_size.
1194 * If we don't we can race with other i_size updaters on our node, like
1195 * ll_file_read. We can also race with i_size propogation to other
1196 * nodes through dirtying and writeback of final cached pages. This
1197 * last one is especially bad for racing o_append users on other
1199 if (ia_valid & ATTR_SIZE) {
1200 ldlm_policy_data_t policy = { .l_extent = {attr->ia_size,
1202 struct lustre_handle lockh = { 0 };
1203 struct ll_inode_info *lli = ll_i2info(inode);
1204 int err, ast_flags = 0;
1205 /* XXX when we fix the AST intents to pass the discard-range
1206 * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
1208 if (attr->ia_size == 0)
1209 ast_flags = LDLM_AST_DISCARD_DATA;
1211 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy, &lockh,
1212 ast_flags, &ll_i2sbi(inode)->ll_seek_stime);
1217 down(&lli->lli_size_sem);
1218 lli->lli_size_pid = current->pid;
1219 rc = vmtruncate(inode, attr->ia_size);
1221 LASSERT(atomic_read(&lli->lli_size_sem.count) <= 0);
1222 lli->lli_size_pid = 0;
1223 up(&lli->lli_size_sem);
1226 err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
1228 CERROR("ll_extent_unlock failed: %d\n", err);
1232 } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
1233 struct obdo *oa = NULL;
1235 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1236 inode->i_ino, LTIME_S(attr->ia_mtime));
1242 oa->o_id = lsm->lsm_object_id;
1243 oa->o_gr = lsm->lsm_object_gr;
1244 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1245 obdo_from_inode(oa, inode, OBD_MD_FLTYPE | OBD_MD_FLATIME |
1246 OBD_MD_FLMTIME | OBD_MD_FLCTIME);
1247 rc = obd_setattr(sbi->ll_dt_exp, oa, lsm, NULL);
1250 CERROR("obd_setattr fails: rc = %d\n", rc);
1255 int ll_setattr(struct dentry *de, struct iattr *attr)
1257 LASSERT(de->d_inode);
1258 return ll_setattr_raw(de->d_inode, attr);
1261 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1262 unsigned long max_age)
1264 struct ll_sb_info *sbi = ll_s2sbi(sb);
1265 struct obd_statfs obd_osfs;
1269 rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age);
1271 CERROR("obd_statfs fails: rc = %d\n", rc);
1275 osfs->os_type = sb->s_magic;
1277 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1278 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1280 rc = obd_statfs(class_exp2obd(sbi->ll_dt_exp), &obd_osfs, max_age);
1282 CERROR("obd_statfs fails: rc = %d\n", rc);
1286 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1287 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1290 osfs->os_blocks = obd_osfs.os_blocks;
1291 osfs->os_bfree = obd_osfs.os_bfree;
1292 osfs->os_bavail = obd_osfs.os_bavail;
1294 /* If we don't have as many objects free on the OST as inodes
1295 * on the MDS, we reduce the total number of inodes to
1296 * compensate, so that the "inodes in use" number is correct.
1298 if (obd_osfs.os_ffree < osfs->os_ffree) {
1299 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1301 osfs->os_ffree = obd_osfs.os_ffree;
1307 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1309 struct obd_statfs osfs;
1312 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p\n", sb);
1313 lprocfs_counter_incr(ll_s2sbi(sb)->ll_stats, LPROC_LL_STAFS);
1315 /* For now we will always get up-to-date statfs values, but in the
1316 * future we may allow some amount of caching on the client (e.g.
1317 * from QOS or lprocfs updates). */
1318 rc = ll_statfs_internal(sb, &osfs, jiffies - 1);
1322 statfs_unpack(sfs, &osfs);
1324 if (sizeof(sfs->f_blocks) == 4) {
1325 while (osfs.os_blocks > ~0UL) {
1328 osfs.os_blocks >>= 1;
1329 osfs.os_bfree >>= 1;
1330 osfs.os_bavail >>= 1;
1334 sfs->f_blocks = osfs.os_blocks;
1335 sfs->f_bfree = osfs.os_bfree;
1336 sfs->f_bavail = osfs.os_bavail;
1341 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1343 struct ll_inode_info *lli = ll_i2info(inode);
1344 struct lov_stripe_md *lsm = md->lsm;
1345 struct mds_body *body = md->body;
1346 struct mea *mea = md->mea;
1347 struct posix_acl *ll_acl_access = md->acl_access;
1350 LASSERT((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1352 if (md->lsm && md->lsm->lsm_magic != LOV_MAGIC) {
1353 /* check for default striping info for dir. */
1354 LASSERT((mea != NULL) == ((body->valid & OBD_MD_FLDIREA) != 0));
1358 LASSERT(lsm->lsm_object_gr > 0);
1359 if (lli->lli_smd == NULL) {
1361 lli->lli_maxbytes = lsm->lsm_maxbytes;
1362 if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1363 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1366 if (memcmp(lli->lli_smd, lsm, sizeof(*lsm))) {
1367 CERROR("lsm mismatch for inode %ld\n",
1369 CERROR("lli_smd:\n");
1370 dump_lsm(D_ERROR, lli->lli_smd);
1372 dump_lsm(D_ERROR, lsm);
1375 /* XXX FIXME -- We should decide on a safer (atomic) and
1376 * more elegant way to update the lsm */
1377 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1378 lli->lli_smd->lsm_oinfo[i].loi_id =
1379 lsm->lsm_oinfo[i].loi_id;
1380 lli->lli_smd->lsm_oinfo[i].loi_gr =
1381 lsm->lsm_oinfo[i].loi_gr;
1382 lli->lli_smd->lsm_oinfo[i].loi_ost_idx =
1383 lsm->lsm_oinfo[i].loi_ost_idx;
1384 lli->lli_smd->lsm_oinfo[i].loi_ost_gen =
1385 lsm->lsm_oinfo[i].loi_ost_gen;
1388 /* bug 2844 - limit i_blksize for broken user-space apps */
1389 LASSERTF(lsm->lsm_xfersize != 0, "%lu\n", lsm->lsm_xfersize);
1390 inode->i_blksize = min(lsm->lsm_xfersize, LL_MAX_BLKSIZE);
1391 if (lli->lli_smd != lsm)
1392 obd_free_memmd(ll_i2dtexp(inode), &lsm);
1396 if (lli->lli_mea == NULL) {
1399 if (memcmp(lli->lli_mea, mea, body->eadatasize)) {
1400 CERROR("mea mismatch for inode %lu\n",
1405 if (lli->lli_mea != mea)
1406 obd_free_memmd(ll_i2mdexp(inode),
1407 (struct lov_stripe_md **) &mea);
1410 if (body->valid & OBD_MD_FID)
1411 id_assign_fid(&lli->lli_id, &body->id1);
1413 if (body->valid & OBD_MD_FLID)
1414 id_ino(&lli->lli_id) = id_ino(&body->id1);
1416 if (body->valid & OBD_MD_FLGENER)
1417 id_gen(&lli->lli_id) = id_gen(&body->id1);
1419 spin_lock(&lli->lli_lock);
1420 if (ll_acl_access != NULL) {
1421 if (lli->lli_acl_access != NULL)
1422 posix_acl_release(lli->lli_acl_access);
1423 lli->lli_acl_access = ll_acl_access;
1425 spin_unlock(&lli->lli_lock);
1427 if (body->valid & OBD_MD_FLID)
1428 inode->i_ino = id_ino(&body->id1);
1429 if (body->valid & OBD_MD_FLGENER)
1430 inode->i_generation = id_gen(&body->id1);
1431 if (body->valid & OBD_MD_FLATIME)
1432 LTIME_S(inode->i_atime) = body->atime;
1433 if (body->valid & OBD_MD_FLMTIME &&
1434 body->mtime > LTIME_S(inode->i_mtime)) {
1435 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to %u\n",
1436 inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
1437 LTIME_S(inode->i_mtime) = body->mtime;
1439 if (body->valid & OBD_MD_FLCTIME &&
1440 body->ctime > LTIME_S(inode->i_ctime))
1441 LTIME_S(inode->i_ctime) = body->ctime;
1442 if (body->valid & OBD_MD_FLMODE) {
1443 inode->i_mode = (inode->i_mode & S_IFMT) |
1444 (body->mode & ~S_IFMT);
1446 if (body->valid & OBD_MD_FLTYPE) {
1447 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1448 (body->mode & S_IFMT);
1450 if (body->valid & OBD_MD_FLUID)
1451 inode->i_uid = body->uid;
1452 if (body->valid & OBD_MD_FLGID)
1453 inode->i_gid = body->gid;
1454 if (body->valid & OBD_MD_FLFLAGS)
1455 inode->i_flags = body->flags;
1456 if (body->valid & OBD_MD_FLNLINK)
1457 inode->i_nlink = body->nlink;
1458 if (body->valid & OBD_MD_FLRDEV)
1459 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1460 inode->i_rdev = body->rdev;
1462 inode->i_rdev = old_decode_dev(body->rdev);
1464 if (body->valid & OBD_MD_FLSIZE)
1465 inode->i_size = body->size;
1466 if (body->valid & OBD_MD_FLBLOCKS)
1467 inode->i_blocks = body->blocks;
1469 if (body->valid & OBD_MD_FLSIZE)
1470 set_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &lli->lli_flags);
1472 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1473 inode->i_dev = (kdev_t)id_group(&lli->lli_id);
1475 LASSERT(id_fid(&lli->lli_id) != 0);
1478 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
1479 static struct backing_dev_info ll_backing_dev_info = {
1480 .ra_pages = 0, /* No readahead */
1481 .memory_backed = 0, /* Does contribute to dirty memory */
1485 void ll_read_inode2(struct inode *inode, void *opaque)
1487 struct lustre_md *md = opaque;
1488 struct ll_inode_info *lli = ll_i2info(inode);
1491 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1492 inode->i_generation, inode);
1496 LASSERT(!lli->lli_smd);
1498 /* Core attributes from the MDS first. This is a new inode, and
1499 * the VFS doesn't zero times in the core inode so we have to do
1500 * it ourselves. They will be overwritten by either MDS or OST
1501 * attributes - we just need to make sure they aren't newer. */
1502 LTIME_S(inode->i_mtime) = 0;
1503 LTIME_S(inode->i_atime) = 0;
1504 LTIME_S(inode->i_ctime) = 0;
1507 ll_update_inode(inode, md);
1509 /* OIDEBUG(inode); */
1511 if (S_ISREG(inode->i_mode)) {
1512 inode->i_op = &ll_file_inode_operations;
1513 inode->i_fop = &ll_file_operations;
1514 inode->i_mapping->a_ops = &ll_aops;
1516 } else if (S_ISDIR(inode->i_mode)) {
1517 inode->i_op = &ll_dir_inode_operations;
1518 inode->i_fop = &ll_dir_operations;
1519 inode->i_mapping->a_ops = &ll_dir_aops;
1521 } else if (S_ISLNK(inode->i_mode)) {
1522 inode->i_op = &ll_fast_symlink_inode_operations;
1525 inode->i_op = &ll_special_inode_operations;
1527 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
1528 init_special_inode(inode, inode->i_mode,
1529 kdev_t_to_nr(inode->i_rdev));
1531 /* initializing backing dev info. */
1532 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
1534 init_special_inode(inode, inode->i_mode, inode->i_rdev);
1536 lli->ll_save_ifop = inode->i_fop;
1538 if (S_ISCHR(inode->i_mode))
1539 inode->i_fop = &ll_special_chr_inode_fops;
1540 else if (S_ISBLK(inode->i_mode))
1541 inode->i_fop = &ll_special_blk_inode_fops;
1542 else if (S_ISFIFO(inode->i_mode))
1543 inode->i_fop = &ll_special_fifo_inode_fops;
1544 else if (S_ISSOCK(inode->i_mode))
1545 inode->i_fop = &ll_special_sock_inode_fops;
1547 CWARN("saved %p, replaced with %p\n", lli->ll_save_ifop,
1550 if (lli->ll_save_ifop->owner) {
1551 CWARN("%p has owner %p\n", lli->ll_save_ifop,
1552 lli->ll_save_ifop->owner);
1558 void ll_delete_inode(struct inode *inode)
1560 struct ll_sb_info *sbi = ll_i2sbi(inode);
1561 struct lustre_id id;
1565 ll_inode2id(&id, inode);
1567 rc = md_delete_inode(sbi->ll_md_exp, &id);
1569 CERROR("md_delete_inode() failed, error %d\n",
1577 int ll_iocontrol(struct inode *inode, struct file *file,
1578 unsigned int cmd, unsigned long arg)
1580 struct ll_sb_info *sbi = ll_i2sbi(inode);
1581 struct ptlrpc_request *req = NULL;
1586 case EXT3_IOC_GETFLAGS: {
1587 struct lustre_id id;
1588 __u64 valid = OBD_MD_FLFLAGS;
1589 struct mds_body *body;
1591 ll_inode2id(&id, inode);
1592 rc = md_getattr(sbi->ll_md_exp, &id, valid, NULL, 0, 0, &req);
1594 CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1598 body = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*body));
1600 if (body->flags & S_APPEND)
1601 flags |= EXT3_APPEND_FL;
1602 if (body->flags & S_IMMUTABLE)
1603 flags |= EXT3_IMMUTABLE_FL;
1604 if (body->flags & S_NOATIME)
1605 flags |= EXT3_NOATIME_FL;
1607 ptlrpc_req_finished (req);
1609 RETURN(put_user(flags, (int *)arg));
1611 case EXT3_IOC_SETFLAGS: {
1612 struct mdc_op_data *op_data;
1615 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1617 if (get_user(flags, (int *)arg))
1624 OBD_ALLOC(op_data, sizeof(*op_data));
1625 if (op_data == NULL) {
1629 ll_prepare_mdc_data(op_data, inode, NULL, NULL, 0, 0);
1631 memset(&attr, 0x0, sizeof(attr));
1632 attr.ia_attr_flags = flags;
1633 attr.ia_valid |= ATTR_ATTR_FLAG;
1635 rc = md_setattr(sbi->ll_md_exp, op_data,
1636 &attr, NULL, 0, NULL, 0, &req);
1637 OBD_FREE(op_data, sizeof(*op_data));
1639 ptlrpc_req_finished(req);
1640 if (rc != -EPERM && rc != -EACCES)
1641 CERROR("md_setattr fails: rc = %d\n", rc);
1645 ptlrpc_req_finished(req);
1647 oa->o_id = lsm->lsm_object_id;
1648 oa->o_gr = lsm->lsm_object_gr;
1649 oa->o_flags = flags;
1650 oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS | OBD_MD_FLGROUP;
1652 rc = obd_setattr(sbi->ll_dt_exp, oa, lsm, NULL);
1655 if (rc != -EPERM && rc != -EACCES)
1656 CERROR("md_setattr fails: rc = %d\n", rc);
1660 if (flags & EXT3_APPEND_FL)
1661 inode->i_flags |= S_APPEND;
1663 inode->i_flags &= ~S_APPEND;
1664 if (flags & EXT3_IMMUTABLE_FL)
1665 inode->i_flags |= S_IMMUTABLE;
1667 inode->i_flags &= ~S_IMMUTABLE;
1668 if (flags & EXT3_NOATIME_FL)
1669 inode->i_flags |= S_NOATIME;
1671 inode->i_flags &= ~S_NOATIME;
1682 /* this is only called in the case of forced umount. */
1683 void ll_umount_begin(struct super_block *sb)
1685 struct ll_sb_info *sbi = ll_s2sbi(sb);
1686 struct obd_ioctl_data ioc_data = { 0 };
1687 struct obd_device *obd;
1690 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1691 sb->s_count, atomic_read(&sb->s_active));
1693 obd = class_exp2obd(sbi->ll_md_exp);
1695 CERROR("Invalid MDC connection handle "LPX64"\n",
1696 sbi->ll_md_exp->exp_handle.h_cookie);
1700 obd->obd_no_recov = 1;
1701 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
1702 sizeof(ioc_data), &ioc_data, NULL);
1704 obd = class_exp2obd(sbi->ll_dt_exp);
1706 CERROR("Invalid LOV connection handle "LPX64"\n",
1707 sbi->ll_dt_exp->exp_handle.h_cookie);
1712 obd->obd_no_recov = 1;
1713 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
1714 sizeof(ioc_data), &ioc_data, NULL);
1717 * really, we'd like to wait until there are no requests outstanding,
1718 * and then continue. For now, we just invalidate the requests,
1719 * schedule, and hope.
1726 int ll_prep_inode(struct obd_export *dt_exp, struct obd_export *md_exp,
1727 struct inode **inode, struct ptlrpc_request *req,
1728 int offset, struct super_block *sb)
1730 struct lustre_md md;
1733 rc = mdc_req2lustre_md(md_exp, req, offset, dt_exp, &md);
1738 ll_update_inode(*inode, &md);
1741 *inode = ll_iget(sb, id_ino(&md.body->id1), &md);
1742 if (*inode == NULL || is_bad_inode(*inode)) {
1743 /* free the lsm if we allocated one above */
1745 obd_free_memmd(dt_exp, &md.lsm);
1747 obd_free_memmd(md_exp,
1748 (struct lov_stripe_md**)&md.mea);
1750 CERROR("new_inode -fatal: rc %d\n", rc);
1757 int ll_get_fid(struct obd_export *exp, struct lustre_id *idp,
1758 char *filename, struct lustre_id *ret)
1760 struct ptlrpc_request *request = NULL;
1761 struct mds_body *body;
1764 rc = md_getattr_lock(exp, idp, filename, strlen(filename) + 1,
1765 OBD_MD_FID, 0, &request);
1767 CDEBUG(D_INFO, "md_getattr_lock failed on %s: rc %d\n",
1772 body = lustre_msg_buf(request->rq_repmsg, 0, sizeof(*body));
1773 LASSERT(body != NULL);
1774 LASSERT_REPSWABBED(request, 0);
1777 ptlrpc_req_finished(request);