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)
121 valsize = sizeof(sbi->ll_dt_desc);
122 memset(&sbi->ll_dt_desc, 0, sizeof(sbi->ll_dt_desc));
123 rc = obd_get_info(sbi->ll_dt_exp, strlen("lovdesc") + 1,
124 "lovdesc", &valsize, &sbi->ll_dt_desc);
128 extern struct dentry_operations ll_d_ops;
130 int lustre_common_fill_super(struct super_block *sb, char *lmv, char *lov,
131 int async, char *security, __u32 *nllu,
134 struct ll_sb_info *sbi = ll_s2sbi(sb);
135 struct ptlrpc_request *request = NULL;
136 struct lustre_handle dt_conn = {0, };
137 struct lustre_handle md_conn = {0, };
138 struct obd_connect_data *data;
139 struct inode *root = NULL;
140 struct obd_device *obd;
141 struct obd_statfs osfs;
148 obd = class_name2obd(lmv);
150 CERROR("MDC %s: not setup or attached\n", lmv);
153 obd_set_info(obd->obd_self_export, strlen("async"), "async",
154 sizeof(async), &async);
156 if ((*remote & (OBD_CONNECT_LOCAL | OBD_CONNECT_REMOTE)) ==
157 (OBD_CONNECT_LOCAL | OBD_CONNECT_REMOTE)) {
158 CERROR("wrong remote flag "LPX64"\n", *remote);
162 OBD_ALLOC(data, sizeof(*data));
166 data->ocd_connect_flags |= *remote & (OBD_CONNECT_LOCAL |
168 memcpy(data->ocd_nllu, nllu, sizeof(data->ocd_nllu));
170 if (security == NULL)
173 err = obd_set_info(obd->obd_self_export, strlen("sec"), "sec",
174 strlen(security), security);
176 CERROR("LMV %s: failed to set security %s, err %d\n",
178 OBD_FREE(data, sizeof(*data));
182 if (proc_lustre_fs_root) {
183 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
186 CERROR("could not register mount in /proc/lustre");
189 err = obd_connect(&md_conn, obd, &sbi->ll_sb_uuid, data,
190 OBD_OPT_REAL_CLIENT);
192 CERROR("An MDS (lmv %s) is performing recovery, of which this"
193 " client is not a part. Please wait for recovery to "
194 "complete, abort, or time out.\n", lmv);
197 CERROR("cannot connect to %s: rc = %d\n", lmv, err);
200 sbi->ll_md_exp = class_conn2export(&md_conn);
201 err = obd_statfs(obd, &osfs, jiffies - HZ);
205 if (!osfs.os_bsize) {
206 CERROR("Invalid block size is detected.");
210 sb->s_magic = LL_SUPER_MAGIC;
211 sb->s_blocksize = osfs.os_bsize;
212 sb->s_blocksize_bits = log2(osfs.os_bsize);
213 sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
215 devno = get_uuid2int(sbi->ll_md_exp->exp_obd->obd_uuid.uuid,
216 strlen(sbi->ll_md_exp->exp_obd->obd_uuid.uuid));
220 /* after statfs, we are supposed to have connected to MDSs,
221 * so it's ok to check remote flag returned.
223 valsize = sizeof(&sbi->ll_remote);
224 err = obd_get_info(sbi->ll_md_exp, strlen("remote_flag"), "remote_flag",
225 &valsize, &sbi->ll_remote);
227 CERROR("fail to obtain remote flag\n");
231 obd = class_name2obd(lov);
233 CERROR("OSC %s: not setup or attached\n", lov);
236 obd_set_info(obd->obd_self_export, strlen("async"), "async",
237 sizeof(async), &async);
239 err = obd_connect(&dt_conn, obd, &sbi->ll_sb_uuid, data,
240 OBD_OPT_REAL_CLIENT);
242 CERROR("An OST (lov %s) is performing recovery, of which this"
243 " client is not a part. Please wait for recovery to "
244 "complete, abort, or time out.\n", lov);
247 CERROR("cannot connect to %s: rc = %d\n", lov, err);
250 sbi->ll_dt_exp = class_conn2export(&dt_conn);
252 err = lustre_init_dt_desc(sbi);
254 int mdsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
255 obd_init_ea_size(sbi->ll_md_exp, mdsize,
256 sbi->ll_dt_desc.ld_tgt_count *
257 sizeof(struct llog_cookie));
260 err = md_getstatus(sbi->ll_md_exp, &sbi->ll_rootid);
262 CERROR("cannot mds_connect: rc = %d\n", err);
265 CDEBUG(D_SUPER, "rootid "DLID4"\n", OLID4(&sbi->ll_rootid));
267 sb->s_op = &lustre_super_operations;
270 * XXX: move this to after cbd setup? */
271 err = md_getattr(sbi->ll_md_exp, &sbi->ll_rootid,
272 (OBD_MD_FLNOTOBD | OBD_MD_FLBLOCKS | OBD_MD_FID), NULL, 0,
275 CERROR("md_getattr failed for root: rc = %d\n", err);
279 err = mdc_req2lustre_md(sbi->ll_md_exp, request, 0,
280 sbi->ll_dt_exp, &md);
282 CERROR("failed to understand root inode md: rc = %d\n", err);
283 ptlrpc_req_finished(request);
287 LASSERT(id_ino(&sbi->ll_rootid) != 0);
288 root = ll_iget(sb, id_ino(&sbi->ll_rootid), &md);
290 ptlrpc_req_finished(request);
292 if (root == NULL || is_bad_inode(root)) {
294 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
296 obd_free_memmd(sbi->ll_md_exp,
297 (struct lov_stripe_md**)&md.mea);
298 CERROR("lustre_lite: bad iget4 for root\n");
299 GOTO(out_root, err = -EBADF);
302 err = ll_close_thread_start(&sbi->ll_lcq);
304 CERROR("cannot start close thread: rc %d\n", err);
308 ll_gns_add_timer(sbi);
310 /* making vm readahead 0 for 2.4.x. In the case of 2.6.x,
311 backing dev info assigned to inode mapping is used for
312 determining maximal readahead. */
313 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0)) && \
314 !defined(KERNEL_HAS_AS_MAX_READAHEAD)
315 /* bug 2805 - set VM readahead to zero */
316 vm_max_readahead = vm_min_readahead = 0;
319 sb->s_root = d_alloc_root(root);
320 sb->s_root->d_op = &ll_d_ops;
322 sb->s_flags |= MS_POSIXACL;
324 CWARN("Enabling PDIROPS\n");
325 sb->s_flags |= S_PDIROPS;
329 OBD_FREE(data, sizeof(*data));
335 obd_disconnect(sbi->ll_dt_exp, 0);
337 obd_disconnect(sbi->ll_md_exp, 0);
340 OBD_FREE(data, sizeof(*data));
341 lprocfs_unregister_mountpoint(sbi);
345 void lustre_common_put_super(struct super_block *sb)
347 struct ll_sb_info *sbi = ll_s2sbi(sb);
348 struct hlist_node *tmp, *next;
351 ll_gns_del_timer(sbi);
352 ll_close_thread_shutdown(sbi->ll_lcq);
354 list_del(&sbi->ll_conn_chain);
355 obd_disconnect(sbi->ll_dt_exp, 0);
357 lprocfs_unregister_mountpoint(sbi);
358 if (sbi->ll_proc_root) {
359 lprocfs_remove(sbi->ll_proc_root);
360 sbi->ll_proc_root = NULL;
363 obd_disconnect(sbi->ll_md_exp, 0);
365 // We do this to get rid of orphaned dentries. That is not really trw.
366 hlist_for_each_safe(tmp, next, &sbi->ll_orphan_dentry_list) {
367 struct dentry *dentry = hlist_entry(tmp, struct dentry, d_hash);
368 CWARN("orphan dentry %.*s (%p->%p) at unmount\n",
369 dentry->d_name.len, dentry->d_name.name, dentry, next);
370 shrink_dcache_parent(dentry);
375 char *ll_read_opt(const char *opt, char *data)
381 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
382 if (strncmp(opt, data, strlen(opt)))
384 if ((value = strchr(data, '=')) == NULL)
388 OBD_ALLOC(retval, strlen(value) + 1);
390 CERROR("out of memory!\n");
394 memcpy(retval, value, strlen(value)+1);
395 CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
399 int ll_set_opt(const char *opt, char *data, int fl)
403 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
404 if (strncmp(opt, data, strlen(opt)))
410 void ll_options(char *options, char **lov, char **lmv, char **sec,
411 int *async, int *flags)
414 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
415 char *opt_ptr = options;
425 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
426 for (this_char = strtok (options, ",");
428 this_char = strtok (NULL, ",")) {
430 while ((this_char = strsep (&opt_ptr, ",")) != NULL) {
432 CDEBUG(D_SUPER, "this_char %s\n", this_char);
433 if (!*lov && (*lov = ll_read_opt("osc", this_char)))
435 if (!*lmv && (*lmv = ll_read_opt("mdc", this_char)))
437 if (!strncmp(this_char, "lasync", strlen("lasync"))) {
441 if (!*sec && (*sec = ll_read_opt("sec", this_char)))
443 if (!(*flags & LL_SBI_NOLCK) &&
444 ((*flags) = (*flags) |
445 ll_set_opt("nolock", this_char,
453 void ll_lli_init(struct ll_inode_info *lli)
455 sema_init(&lli->lli_open_sem, 1);
456 sema_init(&lli->lli_size_sem, 1);
458 lli->lli_size_pid = 0;
459 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
460 spin_lock_init(&lli->lli_lock);
461 INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
462 lli->lli_inode_magic = LLI_INODE_MAGIC;
463 memset(&lli->lli_id, 0, sizeof(lli->lli_id));
464 sema_init(&lli->lli_och_sem, 1);
465 lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
466 lli->lli_mds_exec_och = NULL;
467 lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
468 lli->lli_open_fd_exec_count = 0;
471 int ll_fill_super(struct super_block *sb, void *data, int silent)
473 struct ll_sb_info *sbi;
478 __u32 nllu[2] = { NOBODY_UID, NOBODY_GID };
479 __u64 remote_flag = 0;
482 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
484 sbi = lustre_init_sbi(sb);
488 sbi->ll_flags |= LL_SBI_READAHEAD;
489 ll_options(data, &lov, &lmv, &sec, &async, &sbi->ll_flags);
493 GOTO(out, err = -EINVAL);
498 GOTO(out, err = -EINVAL);
501 err = lustre_common_fill_super(sb, lmv, lov, async, sec, nllu,
509 OBD_FREE(sec, strlen(sec) + 1);
511 OBD_FREE(lmv, strlen(lmv) + 1);
513 OBD_FREE(lov, strlen(lov) + 1);
515 } /* ll_read_super */
517 static int lustre_process_log(struct lustre_mount_data *lmd, char *profile,
518 struct config_llog_instance *cfg, int allow_recov)
520 struct lustre_cfg *lcfg = NULL;
521 struct lustre_cfg_bufs bufs;
522 struct portals_cfg pcfg;
523 char *peer = "MDS_PEER_UUID";
524 struct obd_device *obd;
525 struct lustre_handle md_conn = {0, };
526 struct obd_export *exp;
527 char *name = "mdc_dev";
529 struct obd_uuid lmv_uuid;
530 struct llog_ctxt *ctxt;
534 if (lmd_bad_magic(lmd))
537 generate_random_uuid(uuid);
538 class_uuid_unparse(uuid, &lmv_uuid);
540 if (lmd->lmd_local_nid) {
541 PCFG_INIT(pcfg, NAL_CMD_REGISTER_MYNID);
542 pcfg.pcfg_nal = lmd->lmd_nal;
543 pcfg.pcfg_nid = lmd->lmd_local_nid;
544 rc = libcfs_nal_cmd(&pcfg);
549 if (lmd->lmd_nal == SOCKNAL ||
550 lmd->lmd_nal == OPENIBNAL ||
551 lmd->lmd_nal == IIBNAL ||
552 lmd->lmd_nal == VIBNAL ||
553 lmd->lmd_nal == RANAL) {
554 PCFG_INIT(pcfg, NAL_CMD_ADD_PEER);
555 pcfg.pcfg_nal = lmd->lmd_nal;
556 pcfg.pcfg_nid = lmd->lmd_server_nid;
557 pcfg.pcfg_id = lmd->lmd_server_ipaddr;
558 pcfg.pcfg_misc = lmd->lmd_port;
559 rc = libcfs_nal_cmd(&pcfg);
563 lustre_cfg_bufs_reset(&bufs, name);
564 lustre_cfg_bufs_set_string(&bufs, 1, peer);
566 lcfg = lustre_cfg_new(LCFG_ADD_UUID, &bufs);
567 lcfg->lcfg_nal = lmd->lmd_nal;
568 lcfg->lcfg_nid = lmd->lmd_server_nid;
569 LASSERT(lcfg->lcfg_nal);
570 LASSERT(lcfg->lcfg_nid);
571 err = class_process_config(lcfg);
572 lustre_cfg_free(lcfg);
574 GOTO(out_del_conn, err);
576 lustre_cfg_bufs_reset(&bufs, name);
577 lustre_cfg_bufs_set_string(&bufs, 1, LUSTRE_MDC_NAME);
578 lustre_cfg_bufs_set_string(&bufs, 2, lmv_uuid.uuid);
580 lcfg = lustre_cfg_new(LCFG_ATTACH, &bufs);
581 err = class_process_config(lcfg);
582 lustre_cfg_free(lcfg);
584 GOTO(out_del_uuid, err);
586 lustre_cfg_bufs_reset(&bufs, name);
587 lustre_cfg_bufs_set_string(&bufs, 1, lmd->lmd_mds);
588 lustre_cfg_bufs_set_string(&bufs, 2, peer);
590 lcfg = lustre_cfg_new(LCFG_SETUP, &bufs);
591 err = class_process_config(lcfg);
592 lustre_cfg_free(lcfg);
594 GOTO(out_detach, err);
596 obd = class_name2obd(name);
598 GOTO(out_cleanup, rc = -EINVAL);
600 rc = obd_set_info(obd->obd_self_export, strlen("sec"), "sec",
601 strlen(lmd->lmd_security), lmd->lmd_security);
603 GOTO(out_cleanup, rc);
605 /* Disable initial recovery on this import */
606 rc = obd_set_info(obd->obd_self_export,
607 strlen("initial_recov"), "initial_recov",
608 sizeof(allow_recov), &allow_recov);
610 GOTO(out_cleanup, rc);
612 rc = obd_connect(&md_conn, obd, &lmv_uuid, NULL, 0);
614 CERROR("cannot connect to %s: rc = %d\n", lmd->lmd_mds, rc);
615 GOTO(out_cleanup, rc);
618 exp = class_conn2export(&md_conn);
620 ctxt = llog_get_context(&exp->exp_obd->obd_llogs,LLOG_CONFIG_REPL_CTXT);
621 rc = class_config_process_llog(ctxt, profile, cfg);
623 CERROR("class_config_process_llog failed: rc = %d\n", rc);
625 err = obd_disconnect(exp, 0);
629 lustre_cfg_bufs_reset(&bufs, name);
630 lcfg = lustre_cfg_new(LCFG_CLEANUP, &bufs);
631 err = class_process_config(lcfg);
632 lustre_cfg_free(lcfg);
636 lustre_cfg_bufs_reset(&bufs, name);
637 lcfg = lustre_cfg_new(LCFG_DETACH, &bufs);
638 err = class_process_config(lcfg);
639 lustre_cfg_free(lcfg);
644 lustre_cfg_bufs_reset(&bufs, name);
645 lustre_cfg_bufs_set_string(&bufs, 1, peer);
646 lcfg = lustre_cfg_new(LCFG_DEL_UUID, &bufs);
647 err = class_process_config(lcfg);
648 lustre_cfg_free(lcfg);
651 if (lmd->lmd_nal == SOCKNAL ||
652 lmd->lmd_nal == OPENIBNAL ||
653 lmd->lmd_nal == IIBNAL ||
654 lmd->lmd_nal == VIBNAL ||
655 lmd->lmd_nal == RANAL) {
658 PCFG_INIT(pcfg, NAL_CMD_DEL_PEER);
659 pcfg.pcfg_nal = lmd->lmd_nal;
660 pcfg.pcfg_nid = lmd->lmd_server_nid;
661 pcfg.pcfg_flags = 1; /* single_share */
662 err2 = libcfs_nal_cmd(&pcfg);
675 static void lustre_manual_cleanup(struct ll_sb_info *sbi)
677 struct lustre_cfg *lcfg;
678 struct lustre_cfg_bufs bufs;
679 struct obd_device *obd;
682 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) != NULL)
686 lustre_cfg_bufs_reset(&bufs, obd->obd_name);
687 lcfg = lustre_cfg_new(LCFG_CLEANUP, &bufs);
688 err = class_process_config(lcfg);
690 CERROR("cleanup failed: %s\n", obd->obd_name);
694 lcfg->lcfg_command = LCFG_DETACH;
695 err = class_process_config(lcfg);
696 lustre_cfg_free(lcfg);
698 CERROR("detach failed: %s\n", obd->obd_name);
703 if (sbi->ll_lmd != NULL)
704 class_del_profile(sbi->ll_lmd->lmd_profile);
707 int lustre_fill_super(struct super_block *sb, void *data, int silent)
709 struct lustre_mount_data * lmd = data;
710 char *lov = NULL, *lmv = NULL;
711 struct ll_sb_info *sbi;
715 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
716 if (lmd_bad_magic(lmd))
719 sbi = lustre_init_sbi(sb);
723 sbi->ll_flags |= LL_SBI_READAHEAD;
725 if (lmd->lmd_profile) {
726 struct lustre_profile *lprof;
727 struct config_llog_instance cfg;
730 if (lmd->lmd_mds[0] == '\0') {
731 CERROR("no mds name\n");
732 GOTO(out_free, err = -EINVAL);
734 lmd->lmd_security[sizeof(lmd->lmd_security) - 1] = 0;
736 OBD_ALLOC(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
737 if (sbi->ll_lmd == NULL)
738 GOTO(out_free, err = -ENOMEM);
739 memcpy(sbi->ll_lmd, lmd, sizeof(*lmd));
741 /* generate a string unique to this super, let's try
742 the address of the super itself.*/
743 len = (sizeof(sb) * 2) + 1;
744 OBD_ALLOC(sbi->ll_instance, len);
745 if (sbi->ll_instance == NULL)
746 GOTO(out_free, err = -ENOMEM);
747 sprintf(sbi->ll_instance, "%p", sb);
749 cfg.cfg_instance = sbi->ll_instance;
750 cfg.cfg_uuid = sbi->ll_sb_uuid;
751 cfg.cfg_local_nid = lmd->lmd_local_nid;
752 err = lustre_process_log(lmd, lmd->lmd_profile, &cfg, 0);
754 CERROR("Unable to process log: %s\n", lmd->lmd_profile);
758 lprof = class_get_profile(lmd->lmd_profile);
760 CERROR("No profile found: %s\n", lmd->lmd_profile);
761 GOTO(out_free, err = -EINVAL);
764 OBD_FREE(lov, strlen(lov) + 1);
765 OBD_ALLOC(lov, strlen(lprof->lp_lov) +
766 strlen(sbi->ll_instance) + 2);
767 sprintf(lov, "%s-%s", lprof->lp_lov, sbi->ll_instance);
770 OBD_FREE(lmv, strlen(lmv) + 1);
771 OBD_ALLOC(lmv, strlen(lprof->lp_lmv) +
772 strlen(sbi->ll_instance) + 2);
773 sprintf(lmv, "%s-%s", lprof->lp_lmv, sbi->ll_instance);
778 GOTO(out_free, err = -EINVAL);
783 GOTO(out_free, err = -EINVAL);
786 err = lustre_common_fill_super(sb, lmv, lov, lmd->lmd_async,
787 lmd->lmd_security, &lmd->lmd_nllu,
788 &lmd->lmd_remote_flag);
795 OBD_FREE(lmv, strlen(lmv) + 1);
797 OBD_FREE(lov, strlen(lov) + 1);
803 int len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean")+1;
806 if (sbi->ll_instance != NULL) {
807 struct lustre_mount_data *lmd = sbi->ll_lmd;
808 struct config_llog_instance cfg;
811 cfg.cfg_instance = sbi->ll_instance;
812 cfg.cfg_uuid = sbi->ll_sb_uuid;
814 OBD_ALLOC(cl_prof, len);
815 sprintf(cl_prof, "%s-clean", lmd->lmd_profile);
816 err = lustre_process_log(lmd, cl_prof, &cfg, 0);
818 CERROR("Unable to process log: %s\n", cl_prof);
819 lustre_manual_cleanup(sbi);
821 OBD_FREE(cl_prof, len);
822 OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance) + 1);
824 OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
828 } /* lustre_fill_super */
830 void lustre_put_super(struct super_block *sb)
832 struct obd_device *obd;
833 struct ll_sb_info *sbi = ll_s2sbi(sb);
834 int force_umount = 0;
837 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
838 obd = class_exp2obd(sbi->ll_md_exp);
840 force_umount = obd->obd_no_recov;
843 lustre_common_put_super(sb);
844 if (sbi->ll_lmd != NULL) {
846 int len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean")+1;
848 struct config_llog_instance cfg;
851 CERROR("force umount, doing manual cleanup\n");
852 lustre_manual_cleanup(sbi);
856 cfg.cfg_instance = sbi->ll_instance;
857 cfg.cfg_uuid = sbi->ll_sb_uuid;
859 OBD_ALLOC(cl_prof, len);
860 sprintf(cl_prof, "%s-clean", sbi->ll_lmd->lmd_profile);
861 err = lustre_process_log(sbi->ll_lmd, cl_prof, &cfg, 0);
863 CERROR("Unable to process log: %s, doing manual cleanup"
865 lustre_manual_cleanup(sbi);
868 OBD_FREE(cl_prof, len);
870 OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
871 OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance) + 1);
877 } /* lustre_put_super */
879 int ll_process_config_update(struct ll_sb_info *sbi, int clean)
881 struct obd_export *md_exp = sbi->ll_md_exp;
882 struct lustre_mount_data *lmd = sbi->ll_lmd;
883 struct llog_ctxt *ctxt;
884 struct config_llog_instance cfg;
885 char *profile = lmd->lmd_profile, *name = NULL;
886 int rc, namelen = 0, version;
892 CERROR("Client not mounted with zero-conf; cannot process "
897 rc = ldlm_cli_cancel_unused(md_exp->exp_obd->obd_namespace, NULL,
898 LDLM_FL_CONFIG_CHANGE, NULL);
900 CWARN("ldlm_cli_cancel_unused(mdc): %d\n", rc);
902 rc = obd_cancel_unused(sbi->ll_dt_exp, NULL, LDLM_FL_CONFIG_CHANGE,
905 CWARN("obd_cancel_unused(lov): %d\n", rc);
907 cfg.cfg_instance = sbi->ll_instance;
908 cfg.cfg_uuid = sbi->ll_sb_uuid;
909 cfg.cfg_local_nid = lmd->lmd_local_nid;
911 namelen = strlen(profile) + 20; /* -clean-######### */
912 OBD_ALLOC(name, namelen);
917 version = sbi->ll_config_version - 1;
918 sprintf(name, "%s-clean-%d", profile, version);
920 version = sbi->ll_config_version + 1;
921 sprintf(name, "%s-%d", profile, version);
924 CWARN("Applying configuration log %s\n", name);
926 ctxt = llog_get_context(&md_exp->exp_obd->obd_llogs,
927 LLOG_CONFIG_REPL_CTXT);
928 rc = class_config_process_llog(ctxt, name, &cfg);
930 sbi->ll_config_version = version;
931 CWARN("Finished applying configuration log %s: %d\n", name, rc);
933 if (rc == 0 && clean == 0) {
934 struct lov_desc desc;
936 valsize = sizeof(desc);
937 rc = obd_get_info(sbi->ll_dt_exp, strlen("lovdesc") + 1,
938 "lovdesc", &valsize, &desc);
940 rc = obd_init_ea_size(md_exp,
941 obd_size_diskmd(sbi->ll_dt_exp, NULL),
943 sizeof(struct llog_cookie)));
945 OBD_FREE(name, namelen);
949 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
951 struct inode *inode = NULL;
952 l_lock(&lock->l_resource->lr_namespace->ns_lock);
953 if (lock->l_ast_data) {
954 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
955 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
956 inode = igrab(lock->l_ast_data);
958 inode = lock->l_ast_data;
959 CDEBUG(inode->i_state & I_FREEING ? D_INFO : D_WARNING,
960 "l_ast_data %p is bogus: magic %0x8\n",
961 lock->l_ast_data, lli->lli_inode_magic);
965 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
969 int null_if_equal(struct ldlm_lock *lock, void *data)
971 if (data == lock->l_ast_data) {
972 lock->l_ast_data = NULL;
974 if (lock->l_req_mode != lock->l_granted_mode)
975 LDLM_ERROR(lock,"clearing inode with ungranted lock\n");
978 return LDLM_ITER_CONTINUE;
981 void ll_clear_inode(struct inode *inode)
984 struct ll_inode_info *lli = ll_i2info(inode);
985 struct ll_sb_info *sbi = ll_i2sbi(inode);
988 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
989 inode->i_generation, inode);
991 lli->lli_inode_magic = LLI_INODE_DEAD;
992 ll_inode2id(&id, inode);
994 clear_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &(ll_i2info(inode)->lli_flags));
995 md_change_cbdata(sbi->ll_md_exp, &id, null_if_equal, inode);
997 LASSERT(!lli->lli_open_fd_write_count);
998 LASSERT(!lli->lli_open_fd_read_count);
999 LASSERT(!lli->lli_open_fd_exec_count);
1000 if (lli->lli_mds_write_och)
1001 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_WRITE);
1002 if (lli->lli_mds_exec_och)
1003 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_EXEC);
1004 if (lli->lli_mds_read_och)
1005 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_READ);
1007 obd_change_cbdata(sbi->ll_dt_exp, lli->lli_smd,
1008 null_if_equal, inode);
1011 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1012 lli->lli_smd = NULL;
1016 obd_free_memmd(sbi->ll_md_exp,
1017 (struct lov_stripe_md **) &lli->lli_mea);
1018 lli->lli_mea = NULL;
1021 if (lli->lli_symlink_name) {
1022 OBD_FREE(lli->lli_symlink_name,
1023 strlen(lli->lli_symlink_name) + 1);
1024 lli->lli_symlink_name = NULL;
1026 lli->lli_inode_magic = LLI_INODE_DEAD;
1031 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1032 * object(s) determine the file size and mtime. Otherwise, the MDS will
1033 * keep these values until such a time that objects are allocated for it.
1034 * We do the MDS operations first, as it is checking permissions for us.
1035 * We don't to the MDS RPC if there is nothing that we want to store there,
1036 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1037 * going to do an RPC anyways.
1039 * If we are doing a truncate, we will send the mtime and ctime updates
1040 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1041 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1044 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1046 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1047 struct ll_sb_info *sbi = ll_i2sbi(inode);
1048 struct ptlrpc_request *request = NULL;
1049 struct mdc_op_data *op_data;
1050 int ia_valid = attr->ia_valid;
1054 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu\n", inode->i_ino);
1055 lprocfs_counter_incr(ll_i2sbi(inode)->ll_stats, LPROC_LL_SETATTR);
1057 if (ia_valid & ATTR_SIZE) {
1058 if (attr->ia_size > ll_file_maxbytes(inode)) {
1059 CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1060 attr->ia_size, ll_file_maxbytes(inode));
1064 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1067 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1068 if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1069 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1073 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1074 if (attr->ia_valid & ATTR_CTIME) {
1075 attr->ia_ctime = CURRENT_TIME;
1076 attr->ia_valid |= ATTR_CTIME_SET;
1078 if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1079 attr->ia_atime = CURRENT_TIME;
1080 attr->ia_valid |= ATTR_ATIME_SET;
1082 if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1083 attr->ia_mtime = CURRENT_TIME;
1084 attr->ia_valid |= ATTR_MTIME_SET;
1087 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1088 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1089 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1090 LTIME_S(CURRENT_TIME));
1093 attr->ia_valid &= ~ATTR_SIZE;
1095 /* If only OST attributes being set on objects, don't do MDS RPC.
1096 * In that case, we need to check permissions and update the local
1097 * inode ourselves so we can call obdo_from_inode() always. */
1098 if (ia_valid & (lsm ? ~(ATTR_SIZE | ATTR_FROM_OPEN /*| ATTR_RAW*/) : ~0)) {
1099 struct lustre_md md;
1101 OBD_ALLOC(op_data, sizeof(*op_data));
1102 if (op_data == NULL)
1104 ll_prepare_mdc_data(op_data, inode, NULL, NULL, 0, 0);
1106 rc = md_setattr(sbi->ll_md_exp, op_data,
1107 attr, NULL, 0, NULL, 0, &request);
1108 OBD_FREE(op_data, sizeof(*op_data));
1110 ptlrpc_req_finished(request);
1111 if (rc != -EPERM && rc != -EACCES)
1112 CERROR("md_setattr fails: rc = %d\n", rc);
1116 rc = mdc_req2lustre_md(sbi->ll_md_exp, request, 0,
1117 sbi->ll_dt_exp, &md);
1119 ptlrpc_req_finished(request);
1123 /* We call inode_setattr to adjust timestamps, but we first
1124 * clear ATTR_SIZE to avoid invoking vmtruncate.
1126 * NB: ATTR_SIZE will only be set at this point if the size
1127 * resides on the MDS, ie, this file has no objects. */
1128 attr->ia_valid &= ~ATTR_SIZE;
1131 * assigning inode_setattr() to @err to disable warning that
1132 * function's result should be checked by by caller. error is
1133 * impossible here, as vmtruncate() control path is disabled.
1135 err = inode_setattr(inode, attr);
1136 ll_update_inode(inode, &md);
1137 ptlrpc_req_finished(request);
1139 if (!lsm || !S_ISREG(inode->i_mode)) {
1140 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1144 /* The OST doesn't check permissions, but the alternative is
1145 * a gratuitous RPC to the MDS. We already rely on the client
1146 * to do read/write/truncate permission checks, so is mtime OK?
1148 if (ia_valid & (ATTR_MTIME | ATTR_ATIME)) {
1149 /* from sys_utime() */
1150 if (!(ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET))) {
1151 if (current->fsuid != inode->i_uid &&
1152 (rc=ll_permission(inode,MAY_WRITE,NULL))!=0)
1155 /* from inode_change_ok() */
1156 if (current->fsuid != inode->i_uid &&
1157 !capable(CAP_FOWNER))
1162 /* won't invoke vmtruncate, as we already cleared ATTR_SIZE */
1163 err = inode_setattr(inode, attr);
1165 * assigning inode_setattr() to @err to disable warning that
1166 * function's result should be checked by by caller. error is
1167 * impossible here, as vmtruncate() control path is disabled.
1171 /* We really need to get our PW lock before we change inode->i_size.
1172 * If we don't we can race with other i_size updaters on our node, like
1173 * ll_file_read. We can also race with i_size propogation to other
1174 * nodes through dirtying and writeback of final cached pages. This
1175 * last one is especially bad for racing o_append users on other
1177 if (ia_valid & ATTR_SIZE) {
1178 ldlm_policy_data_t policy = { .l_extent = {attr->ia_size,
1180 struct lustre_handle lockh = { 0 };
1181 struct ll_inode_info *lli = ll_i2info(inode);
1182 int err, ast_flags = 0;
1183 /* XXX when we fix the AST intents to pass the discard-range
1184 * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
1186 if (attr->ia_size == 0)
1187 ast_flags = LDLM_AST_DISCARD_DATA;
1189 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy, &lockh,
1190 ast_flags, &ll_i2sbi(inode)->ll_seek_stime);
1195 down(&lli->lli_size_sem);
1196 lli->lli_size_pid = current->pid;
1197 rc = vmtruncate(inode, attr->ia_size);
1199 LASSERT(atomic_read(&lli->lli_size_sem.count) <= 0);
1200 lli->lli_size_pid = 0;
1201 up(&lli->lli_size_sem);
1204 err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
1206 CERROR("ll_extent_unlock failed: %d\n", err);
1210 } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
1211 struct obdo *oa = NULL;
1213 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1214 inode->i_ino, LTIME_S(attr->ia_mtime));
1220 oa->o_id = lsm->lsm_object_id;
1221 oa->o_gr = lsm->lsm_object_gr;
1222 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1223 obdo_from_inode(oa, inode, OBD_MD_FLTYPE | OBD_MD_FLATIME |
1224 OBD_MD_FLMTIME | OBD_MD_FLCTIME);
1225 rc = obd_setattr(sbi->ll_dt_exp, oa, lsm, NULL);
1228 CERROR("obd_setattr fails: rc = %d\n", rc);
1233 int ll_setattr(struct dentry *de, struct iattr *attr)
1235 LASSERT(de->d_inode);
1236 return ll_setattr_raw(de->d_inode, attr);
1239 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1240 unsigned long max_age)
1242 struct ll_sb_info *sbi = ll_s2sbi(sb);
1243 struct obd_statfs obd_osfs;
1247 rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age);
1249 CERROR("obd_statfs fails: rc = %d\n", rc);
1253 osfs->os_type = sb->s_magic;
1255 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1256 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1258 rc = obd_statfs(class_exp2obd(sbi->ll_dt_exp), &obd_osfs, max_age);
1260 CERROR("obd_statfs fails: rc = %d\n", rc);
1264 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1265 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1268 osfs->os_blocks = obd_osfs.os_blocks;
1269 osfs->os_bfree = obd_osfs.os_bfree;
1270 osfs->os_bavail = obd_osfs.os_bavail;
1272 /* If we don't have as many objects free on the OST as inodes
1273 * on the MDS, we reduce the total number of inodes to
1274 * compensate, so that the "inodes in use" number is correct.
1276 if (obd_osfs.os_ffree < osfs->os_ffree) {
1277 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1279 osfs->os_ffree = obd_osfs.os_ffree;
1285 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1287 struct obd_statfs osfs;
1290 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p\n", sb);
1291 lprocfs_counter_incr(ll_s2sbi(sb)->ll_stats, LPROC_LL_STAFS);
1293 /* For now we will always get up-to-date statfs values, but in the
1294 * future we may allow some amount of caching on the client (e.g.
1295 * from QOS or lprocfs updates). */
1296 rc = ll_statfs_internal(sb, &osfs, jiffies - 1);
1300 statfs_unpack(sfs, &osfs);
1302 if (sizeof(sfs->f_blocks) == 4) {
1303 while (osfs.os_blocks > ~0UL) {
1306 osfs.os_blocks >>= 1;
1307 osfs.os_bfree >>= 1;
1308 osfs.os_bavail >>= 1;
1312 sfs->f_blocks = osfs.os_blocks;
1313 sfs->f_bfree = osfs.os_bfree;
1314 sfs->f_bavail = osfs.os_bavail;
1319 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1321 struct ll_inode_info *lli = ll_i2info(inode);
1322 struct lov_stripe_md *lsm = md->lsm;
1323 struct mds_body *body = md->body;
1324 struct mea *mea = md->mea;
1325 struct posix_acl *ll_acl_access = md->acl_access;
1328 LASSERT((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1330 if (md->lsm && md->lsm->lsm_magic != LOV_MAGIC) {
1331 /* check for default striping info for dir. */
1332 LASSERT((mea != NULL) == ((body->valid & OBD_MD_FLDIREA) != 0));
1336 LASSERT(lsm->lsm_object_gr > 0);
1337 if (lli->lli_smd == NULL) {
1339 lli->lli_maxbytes = lsm->lsm_maxbytes;
1340 if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1341 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1344 if (memcmp(lli->lli_smd, lsm, sizeof(*lsm))) {
1345 CERROR("lsm mismatch for inode %ld\n",
1347 CERROR("lli_smd:\n");
1348 dump_lsm(D_ERROR, lli->lli_smd);
1350 dump_lsm(D_ERROR, lsm);
1353 /* XXX FIXME -- We should decide on a safer (atomic) and
1354 * more elegant way to update the lsm */
1355 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1356 lli->lli_smd->lsm_oinfo[i].loi_id =
1357 lsm->lsm_oinfo[i].loi_id;
1358 lli->lli_smd->lsm_oinfo[i].loi_gr =
1359 lsm->lsm_oinfo[i].loi_gr;
1360 lli->lli_smd->lsm_oinfo[i].loi_ost_idx =
1361 lsm->lsm_oinfo[i].loi_ost_idx;
1362 lli->lli_smd->lsm_oinfo[i].loi_ost_gen =
1363 lsm->lsm_oinfo[i].loi_ost_gen;
1366 /* bug 2844 - limit i_blksize for broken user-space apps */
1367 LASSERTF(lsm->lsm_xfersize != 0, "%lu\n", lsm->lsm_xfersize);
1368 inode->i_blksize = min(lsm->lsm_xfersize, LL_MAX_BLKSIZE);
1369 if (lli->lli_smd != lsm)
1370 obd_free_memmd(ll_i2dtexp(inode), &lsm);
1374 if (lli->lli_mea == NULL) {
1377 if (memcmp(lli->lli_mea, mea, body->eadatasize)) {
1378 CERROR("mea mismatch for inode %lu\n",
1383 if (lli->lli_mea != mea)
1384 obd_free_memmd(ll_i2mdexp(inode),
1385 (struct lov_stripe_md **) &mea);
1388 if (body->valid & OBD_MD_FID)
1389 id_assign_fid(&lli->lli_id, &body->id1);
1391 if (body->valid & OBD_MD_FLID)
1392 id_ino(&lli->lli_id) = id_ino(&body->id1);
1394 if (body->valid & OBD_MD_FLGENER)
1395 id_gen(&lli->lli_id) = id_gen(&body->id1);
1397 spin_lock(&lli->lli_lock);
1398 if (ll_acl_access != NULL) {
1399 if (lli->lli_acl_access != NULL)
1400 posix_acl_release(lli->lli_acl_access);
1401 lli->lli_acl_access = ll_acl_access;
1403 spin_unlock(&lli->lli_lock);
1405 if (body->valid & OBD_MD_FLID)
1406 inode->i_ino = id_ino(&body->id1);
1407 if (body->valid & OBD_MD_FLGENER)
1408 inode->i_generation = id_gen(&body->id1);
1409 if (body->valid & OBD_MD_FLATIME)
1410 LTIME_S(inode->i_atime) = body->atime;
1411 if (body->valid & OBD_MD_FLMTIME &&
1412 body->mtime > LTIME_S(inode->i_mtime)) {
1413 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to %u\n",
1414 inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
1415 LTIME_S(inode->i_mtime) = body->mtime;
1417 if (body->valid & OBD_MD_FLCTIME &&
1418 body->ctime > LTIME_S(inode->i_ctime))
1419 LTIME_S(inode->i_ctime) = body->ctime;
1420 if (body->valid & OBD_MD_FLMODE) {
1421 inode->i_mode = (inode->i_mode & S_IFMT) |
1422 (body->mode & ~S_IFMT);
1424 if (body->valid & OBD_MD_FLTYPE) {
1425 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1426 (body->mode & S_IFMT);
1428 if (body->valid & OBD_MD_FLUID)
1429 inode->i_uid = body->uid;
1430 if (body->valid & OBD_MD_FLGID)
1431 inode->i_gid = body->gid;
1432 if (body->valid & OBD_MD_FLFLAGS)
1433 inode->i_flags = body->flags;
1434 if (body->valid & OBD_MD_FLNLINK)
1435 inode->i_nlink = body->nlink;
1436 if (body->valid & OBD_MD_FLRDEV)
1437 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1438 inode->i_rdev = body->rdev;
1440 inode->i_rdev = old_decode_dev(body->rdev);
1442 if (body->valid & OBD_MD_FLSIZE)
1443 inode->i_size = body->size;
1444 if (body->valid & OBD_MD_FLBLOCKS)
1445 inode->i_blocks = body->blocks;
1447 if (body->valid & OBD_MD_FLSIZE)
1448 set_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &lli->lli_flags);
1450 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1451 inode->i_dev = (kdev_t)id_group(&lli->lli_id);
1453 LASSERT(id_fid(&lli->lli_id) != 0);
1456 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
1457 static struct backing_dev_info ll_backing_dev_info = {
1458 .ra_pages = 0, /* No readahead */
1459 .memory_backed = 0, /* Does contribute to dirty memory */
1463 void ll_read_inode2(struct inode *inode, void *opaque)
1465 struct lustre_md *md = opaque;
1466 struct ll_inode_info *lli = ll_i2info(inode);
1469 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1470 inode->i_generation, inode);
1474 LASSERT(!lli->lli_smd);
1476 /* Core attributes from the MDS first. This is a new inode, and
1477 * the VFS doesn't zero times in the core inode so we have to do
1478 * it ourselves. They will be overwritten by either MDS or OST
1479 * attributes - we just need to make sure they aren't newer. */
1480 LTIME_S(inode->i_mtime) = 0;
1481 LTIME_S(inode->i_atime) = 0;
1482 LTIME_S(inode->i_ctime) = 0;
1485 ll_update_inode(inode, md);
1487 /* OIDEBUG(inode); */
1489 if (S_ISREG(inode->i_mode)) {
1490 inode->i_op = &ll_file_inode_operations;
1491 inode->i_fop = &ll_file_operations;
1492 inode->i_mapping->a_ops = &ll_aops;
1494 } else if (S_ISDIR(inode->i_mode)) {
1495 inode->i_op = &ll_dir_inode_operations;
1496 inode->i_fop = &ll_dir_operations;
1497 inode->i_mapping->a_ops = &ll_dir_aops;
1499 } else if (S_ISLNK(inode->i_mode)) {
1500 inode->i_op = &ll_fast_symlink_inode_operations;
1503 inode->i_op = &ll_special_inode_operations;
1505 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
1506 init_special_inode(inode, inode->i_mode,
1507 kdev_t_to_nr(inode->i_rdev));
1509 /* initializing backing dev info. */
1510 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
1512 init_special_inode(inode, inode->i_mode, inode->i_rdev);
1514 lli->ll_save_ifop = inode->i_fop;
1516 if (S_ISCHR(inode->i_mode))
1517 inode->i_fop = &ll_special_chr_inode_fops;
1518 else if (S_ISBLK(inode->i_mode))
1519 inode->i_fop = &ll_special_blk_inode_fops;
1520 else if (S_ISFIFO(inode->i_mode))
1521 inode->i_fop = &ll_special_fifo_inode_fops;
1522 else if (S_ISSOCK(inode->i_mode))
1523 inode->i_fop = &ll_special_sock_inode_fops;
1525 CWARN("saved %p, replaced with %p\n", lli->ll_save_ifop,
1528 if (lli->ll_save_ifop->owner) {
1529 CWARN("%p has owner %p\n", lli->ll_save_ifop,
1530 lli->ll_save_ifop->owner);
1536 void ll_delete_inode(struct inode *inode)
1538 struct ll_sb_info *sbi = ll_i2sbi(inode);
1539 struct lustre_id id;
1543 ll_inode2id(&id, inode);
1545 rc = md_delete_inode(sbi->ll_md_exp, &id);
1547 CERROR("md_delete_inode() failed, error %d\n",
1555 int ll_iocontrol(struct inode *inode, struct file *file,
1556 unsigned int cmd, unsigned long arg)
1558 struct ll_sb_info *sbi = ll_i2sbi(inode);
1559 struct ptlrpc_request *req = NULL;
1564 case EXT3_IOC_GETFLAGS: {
1565 struct lustre_id id;
1566 __u64 valid = OBD_MD_FLFLAGS;
1567 struct mds_body *body;
1569 ll_inode2id(&id, inode);
1570 rc = md_getattr(sbi->ll_md_exp, &id, valid, NULL, 0, 0, &req);
1572 CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1576 body = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*body));
1578 if (body->flags & S_APPEND)
1579 flags |= EXT3_APPEND_FL;
1580 if (body->flags & S_IMMUTABLE)
1581 flags |= EXT3_IMMUTABLE_FL;
1582 if (body->flags & S_NOATIME)
1583 flags |= EXT3_NOATIME_FL;
1585 ptlrpc_req_finished (req);
1587 RETURN(put_user(flags, (int *)arg));
1589 case EXT3_IOC_SETFLAGS: {
1590 struct mdc_op_data *op_data;
1593 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1595 if (get_user(flags, (int *)arg))
1602 OBD_ALLOC(op_data, sizeof(*op_data));
1603 if (op_data == NULL) {
1607 ll_prepare_mdc_data(op_data, inode, NULL, NULL, 0, 0);
1609 memset(&attr, 0x0, sizeof(attr));
1610 attr.ia_attr_flags = flags;
1611 attr.ia_valid |= ATTR_ATTR_FLAG;
1613 rc = md_setattr(sbi->ll_md_exp, op_data,
1614 &attr, NULL, 0, NULL, 0, &req);
1615 OBD_FREE(op_data, sizeof(*op_data));
1617 ptlrpc_req_finished(req);
1618 if (rc != -EPERM && rc != -EACCES)
1619 CERROR("md_setattr fails: rc = %d\n", rc);
1623 ptlrpc_req_finished(req);
1625 oa->o_id = lsm->lsm_object_id;
1626 oa->o_gr = lsm->lsm_object_gr;
1627 oa->o_flags = flags;
1628 oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS | OBD_MD_FLGROUP;
1630 rc = obd_setattr(sbi->ll_dt_exp, oa, lsm, NULL);
1633 if (rc != -EPERM && rc != -EACCES)
1634 CERROR("md_setattr fails: rc = %d\n", rc);
1638 if (flags & EXT3_APPEND_FL)
1639 inode->i_flags |= S_APPEND;
1641 inode->i_flags &= ~S_APPEND;
1642 if (flags & EXT3_IMMUTABLE_FL)
1643 inode->i_flags |= S_IMMUTABLE;
1645 inode->i_flags &= ~S_IMMUTABLE;
1646 if (flags & EXT3_NOATIME_FL)
1647 inode->i_flags |= S_NOATIME;
1649 inode->i_flags &= ~S_NOATIME;
1660 /* this is only called in the case of forced umount. */
1661 void ll_umount_begin(struct super_block *sb)
1663 struct ll_sb_info *sbi = ll_s2sbi(sb);
1664 struct obd_ioctl_data ioc_data = { 0 };
1665 struct obd_device *obd;
1668 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1669 sb->s_count, atomic_read(&sb->s_active));
1671 obd = class_exp2obd(sbi->ll_md_exp);
1673 CERROR("Invalid MDC connection handle "LPX64"\n",
1674 sbi->ll_md_exp->exp_handle.h_cookie);
1678 obd->obd_no_recov = 1;
1679 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
1680 sizeof(ioc_data), &ioc_data, NULL);
1682 obd = class_exp2obd(sbi->ll_dt_exp);
1684 CERROR("Invalid LOV connection handle "LPX64"\n",
1685 sbi->ll_dt_exp->exp_handle.h_cookie);
1690 obd->obd_no_recov = 1;
1691 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
1692 sizeof(ioc_data), &ioc_data, NULL);
1695 * really, we'd like to wait until there are no requests outstanding,
1696 * and then continue. For now, we just invalidate the requests,
1697 * schedule, and hope.
1704 int ll_prep_inode(struct obd_export *dt_exp, struct obd_export *md_exp,
1705 struct inode **inode, struct ptlrpc_request *req,
1706 int offset, struct super_block *sb)
1708 struct lustre_md md;
1711 rc = mdc_req2lustre_md(md_exp, req, offset, dt_exp, &md);
1716 ll_update_inode(*inode, &md);
1719 *inode = ll_iget(sb, id_ino(&md.body->id1), &md);
1720 if (*inode == NULL || is_bad_inode(*inode)) {
1721 /* free the lsm if we allocated one above */
1723 obd_free_memmd(dt_exp, &md.lsm);
1725 obd_free_memmd(md_exp,
1726 (struct lov_stripe_md**)&md.mea);
1728 CERROR("new_inode -fatal: rc %d\n", rc);
1735 int ll_get_fid(struct obd_export *exp, struct lustre_id *idp,
1736 char *filename, struct lustre_id *ret)
1738 struct ptlrpc_request *request = NULL;
1739 struct mds_body *body;
1742 rc = md_getattr_lock(exp, idp, filename, strlen(filename) + 1,
1743 OBD_MD_FID, 0, &request);
1745 CDEBUG(D_INFO, "md_getattr_lock failed on %s: rc %d\n",
1750 body = lustre_msg_buf(request->rq_repmsg, 0, sizeof(*body));
1751 LASSERT(body != NULL);
1752 LASSERT_REPSWABBED(request, 0);
1755 ptlrpc_req_finished(request);