1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Lustre Light Super operations
6 * Copyright (c) 2002, 2003 Cluster File Systems, Inc.
8 * This file is part of Lustre, http://www.lustre.org.
10 * Lustre is free software; you can redistribute it and/or
11 * modify it under the terms of version 2 of the GNU General Public
12 * License as published by the Free Software Foundation.
14 * Lustre is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
17 * GNU General Public License for more details.
19 * You should have received a copy of the GNU General Public License
20 * along with Lustre; if not, write to the Free Software
21 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
24 #define DEBUG_SUBSYSTEM S_LLITE
26 #include <linux/module.h>
27 #include <linux/types.h>
28 #include <linux/random.h>
29 #include <linux/version.h>
30 #include <linux/seq_file.h>
32 #include <linux/lustre_lite.h>
33 #include <linux/lustre_ha.h>
34 #include <linux/lustre_dlm.h>
35 #include <linux/lprocfs_status.h>
36 #include <linux/lustre_acl.h>
37 #include "llite_internal.h"
39 kmem_cache_t *ll_file_data_slab;
40 kmem_cache_t *ll_intent_slab;
42 extern struct address_space_operations ll_aops;
43 extern struct address_space_operations ll_dir_aops;
46 #define log2(n) ffz(~(n))
49 struct ll_sb_info *lustre_init_sbi(struct super_block *sb)
51 struct ll_sb_info *sbi = NULL;
55 OBD_ALLOC(sbi, sizeof(*sbi));
59 spin_lock_init(&sbi->ll_lock);
60 INIT_LIST_HEAD(&sbi->ll_pglist);
61 sbi->ll_pglist_gen = 0;
62 if (num_physpages < SBI_DEFAULT_RA_MAX / 4)
63 sbi->ll_ra_info.ra_max_pages = num_physpages / 4;
65 sbi->ll_ra_info.ra_max_pages = SBI_DEFAULT_RA_MAX;
66 INIT_LIST_HEAD(&sbi->ll_conn_chain);
67 INIT_HLIST_HEAD(&sbi->ll_orphan_dentry_list);
68 INIT_LIST_HEAD(&sbi->ll_mnt_list);
70 sema_init(&sbi->ll_gns_sem, 1);
71 spin_lock_init(&sbi->ll_gns_lock);
72 INIT_LIST_HEAD(&sbi->ll_gns_sbi_head);
73 init_waitqueue_head(&sbi->ll_gns_waitq);
74 init_completion(&sbi->ll_gns_mount_finished);
76 /* this later may be reset via /proc/fs/... */
77 memcpy(sbi->ll_gns_oname, ".mntinfo", strlen(".mntinfo"));
78 sbi->ll_gns_oname[strlen(sbi->ll_gns_oname)] = '\0';
80 /* this later may be reset via /proc/fs/... */
81 memcpy(sbi->ll_gns_upcall, "/usr/sbin/gns_upcall",
82 strlen("/usr/sbin/gns_upcall"));
83 sbi->ll_gns_upcall[strlen(sbi->ll_gns_upcall)] = '\0';
85 /* default values, may be changed via /proc/fs/... */
86 sbi->ll_gns_state = LL_GNS_IDLE;
87 sbi->ll_gns_pending_dentry = NULL;
88 atomic_set(&sbi->ll_gns_enabled, 1);
89 sbi->ll_gns_tick = GNS_TICK_TIMEOUT;
90 sbi->ll_gns_timeout = GNS_MOUNT_TIMEOUT;
92 sbi->ll_gns_timer.data = (unsigned long)sbi;
93 sbi->ll_gns_timer.function = ll_gns_timer_callback;
94 init_timer(&sbi->ll_gns_timer);
98 generate_random_uuid(uuid);
99 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
103 void lustre_free_sbi(struct super_block *sb)
105 struct ll_sb_info *sbi = ll_s2sbi(sb);
109 list_del(&sbi->ll_gns_sbi_head);
110 del_timer(&sbi->ll_gns_timer);
111 OBD_FREE(sbi, sizeof(*sbi));
113 ll_set_sbi(sb, NULL);
117 int lustre_init_dt_desc(struct ll_sb_info *sbi)
123 valsize = sizeof(sbi->ll_dt_desc);
124 memset(&sbi->ll_dt_desc, 0, sizeof(sbi->ll_dt_desc));
125 rc = obd_get_info(sbi->ll_dt_exp, strlen("lovdesc") + 1,
126 "lovdesc", &valsize, &sbi->ll_dt_desc);
130 extern struct dentry_operations ll_d_ops;
132 int lustre_common_fill_super(struct super_block *sb, char *lmv, char *lov,
133 int async, char *mds_security, char *oss_security,
134 __u32 *nllu, __u64 *remote)
136 struct ll_sb_info *sbi = ll_s2sbi(sb);
137 struct ptlrpc_request *request = NULL;
138 struct lustre_handle dt_conn = {0, };
139 struct lustre_handle md_conn = {0, };
140 struct obd_connect_data *data;
141 struct inode *root = NULL;
142 struct obd_device *obd;
143 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 /* in 2.6.x FS is not allowed to form s_dev */
217 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
221 devno = get_uuid2int((char *)sbi->ll_md_exp->exp_obd->obd_uuid.uuid,
222 strlen((char *)sbi->ll_md_exp->exp_obd->obd_uuid.uuid));
228 /* after statfs, we are supposed to have connected to MDSs,
229 * so it's ok to check remote flag returned.
231 valsize = sizeof(&sbi->ll_remote);
232 err = obd_get_info(sbi->ll_md_exp, strlen("remote_flag"), "remote_flag",
233 &valsize, &sbi->ll_remote);
235 CERROR("fail to obtain remote flag\n");
239 obd = class_name2obd(lov);
241 CERROR("OSC %s: not setup or attached\n", lov);
244 obd_set_info(obd->obd_self_export, strlen("async"), "async",
245 sizeof(async), &async);
247 if (oss_security == NULL)
248 oss_security = "null";
250 err = obd_set_info(obd->obd_self_export, strlen("sec"), "sec",
251 strlen(oss_security), oss_security);
253 CERROR("LOV %s: failed to set security %s, err %d\n",
254 lov, oss_security, err);
255 OBD_FREE(data, sizeof(*data));
259 err = obd_connect(&dt_conn, obd, &sbi->ll_sb_uuid, data, 0);
261 CERROR("An OST (lov %s) is performing recovery, of which this"
262 " client is not a part. Please wait for recovery to "
263 "complete, abort, or time out.\n", lov);
266 CERROR("cannot connect to %s: rc = %d\n", lov, err);
269 sbi->ll_dt_exp = class_conn2export(&dt_conn);
271 err = lustre_init_dt_desc(sbi);
273 int mdsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
274 obd_init_ea_size(sbi->ll_md_exp, mdsize,
275 sbi->ll_dt_desc.ld_tgt_count *
276 sizeof(struct llog_cookie));
279 err = md_getstatus(sbi->ll_md_exp, &sbi->ll_rootid);
281 CERROR("cannot mds_connect: rc = %d\n", err);
284 CDEBUG(D_SUPER, "rootid "DLID4"\n", OLID4(&sbi->ll_rootid));
286 sb->s_op = &lustre_super_operations;
288 /* make root inode */
289 err = md_getattr(sbi->ll_md_exp, &sbi->ll_rootid,
290 (OBD_MD_FLNOTOBD | OBD_MD_FLBLOCKS | OBD_MD_FID),
291 NULL, 0, 0, &request);
293 CERROR("md_getattr failed for root: rc = %d\n", err);
297 err = mdc_req2lustre_md(sbi->ll_md_exp, request, 0,
298 sbi->ll_dt_exp, &md);
300 CERROR("failed to understand root inode md: rc = %d\n", err);
301 ptlrpc_req_finished(request);
305 LASSERT(id_ino(&sbi->ll_rootid) != 0);
306 root = ll_iget(sb, id_ino(&sbi->ll_rootid), &md);
308 ptlrpc_req_finished(request);
310 if (root == NULL || is_bad_inode(root)) {
312 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
314 obd_free_memmd(sbi->ll_md_exp,
315 (struct lov_stripe_md**)&md.mea);
316 CERROR("lustre_lite: bad iget4 for root\n");
317 GOTO(out_root, err = -EBADF);
320 err = ll_close_thread_start(&sbi->ll_lcq);
322 CERROR("cannot start close thread: rc %d\n", err);
326 ll_gns_add_timer(sbi);
328 /* making vm readahead 0 for 2.4.x. In the case of 2.6.x,
329 backing dev info assigned to inode mapping is used for
330 determining maximal readahead. */
331 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0)) && \
332 !defined(KERNEL_HAS_AS_MAX_READAHEAD)
333 /* bug 2805 - set VM readahead to zero */
334 vm_max_readahead = vm_min_readahead = 0;
337 sb->s_root = d_alloc_root(root);
338 sb->s_root->d_op = &ll_d_ops;
340 sb->s_flags |= MS_POSIXACL;
342 CWARN("Enabling PDIROPS\n");
343 sb->s_flags |= S_PDIROPS;
347 OBD_FREE(data, sizeof(*data));
353 obd_disconnect(sbi->ll_dt_exp, 0);
355 obd_disconnect(sbi->ll_md_exp, 0);
358 OBD_FREE(data, sizeof(*data));
359 lprocfs_unregister_mountpoint(sbi);
363 void lustre_common_put_super(struct super_block *sb)
365 struct ll_sb_info *sbi = ll_s2sbi(sb);
366 struct hlist_node *tmp, *next;
369 ll_gns_del_timer(sbi);
370 ll_close_thread_shutdown(sbi->ll_lcq);
372 list_del(&sbi->ll_conn_chain);
373 obd_disconnect(sbi->ll_dt_exp, 0);
375 lprocfs_unregister_mountpoint(sbi);
376 if (sbi->ll_proc_root) {
377 lprocfs_remove(sbi->ll_proc_root);
378 sbi->ll_proc_root = NULL;
381 obd_disconnect(sbi->ll_md_exp, 0);
383 // We do this to get rid of orphaned dentries. That is not really trw.
384 hlist_for_each_safe(tmp, next, &sbi->ll_orphan_dentry_list) {
385 struct dentry *dentry = hlist_entry(tmp, struct dentry, d_hash);
386 CWARN("orphan dentry %.*s (%p->%p) at unmount\n",
387 dentry->d_name.len, dentry->d_name.name, dentry, next);
388 shrink_dcache_parent(dentry);
393 char *ll_read_opt(const char *opt, char *data)
399 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
400 if (strncmp(opt, data, strlen(opt)))
402 if ((value = strchr(data, '=')) == NULL)
406 OBD_ALLOC(retval, strlen(value) + 1);
408 CERROR("out of memory!\n");
412 memcpy(retval, value, strlen(value)+1);
413 CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
417 int ll_set_opt(const char *opt, char *data, int fl)
421 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
422 if (strncmp(opt, data, strlen(opt)))
428 void ll_options(char *options, char **lov, char **lmv, char **mds_sec,
429 char **oss_sec, int *async, int *flags)
432 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
433 char *opt_ptr = options;
443 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
444 for (this_char = strtok (options, ",");
446 this_char = strtok (NULL, ",")) {
448 while ((this_char = strsep (&opt_ptr, ",")) != NULL) {
450 CDEBUG(D_SUPER, "this_char %s\n", this_char);
451 if (!*lov && (*lov = ll_read_opt("osc", this_char)))
453 if (!*lmv && (*lmv = ll_read_opt("mdc", this_char)))
455 if (!strncmp(this_char, "lasync", strlen("lasync"))) {
459 if (!*mds_sec && (*mds_sec = ll_read_opt("mds_sec", this_char)))
461 if (!*oss_sec && (*oss_sec = ll_read_opt("oss_sec", this_char)))
463 if (!(*flags & LL_SBI_NOLCK) &&
464 ((*flags) = (*flags) |
465 ll_set_opt("nolock", this_char,
473 void ll_lli_init(struct ll_inode_info *lli)
475 sema_init(&lli->lli_open_sem, 1);
476 sema_init(&lli->lli_size_sem, 1);
478 lli->lli_size_pid = 0;
479 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
480 spin_lock_init(&lli->lli_lock);
481 INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
482 lli->lli_inode_magic = LLI_INODE_MAGIC;
483 memset(&lli->lli_id, 0, sizeof(lli->lli_id));
484 sema_init(&lli->lli_och_sem, 1);
485 lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
486 lli->lli_mds_exec_och = NULL;
487 lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
488 lli->lli_open_fd_exec_count = 0;
491 int ll_fill_super(struct super_block *sb, void *data, int silent)
493 struct ll_sb_info *sbi;
496 char *mds_sec = NULL;
497 char *oss_sec = NULL;
499 __u32 nllu[2] = { NOBODY_UID, NOBODY_GID };
500 __u64 remote_flag = 0;
503 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
505 sbi = lustre_init_sbi(sb);
509 sbi->ll_flags |= LL_SBI_READAHEAD;
510 ll_options(data, &lov, &lmv, &mds_sec, &oss_sec,
511 &async, &sbi->ll_flags);
515 GOTO(out, err = -EINVAL);
520 GOTO(out, err = -EINVAL);
523 err = lustre_common_fill_super(sb, lmv, lov, async, mds_sec, oss_sec,
531 OBD_FREE(lmv, strlen(lmv) + 1);
533 OBD_FREE(lov, strlen(lov) + 1);
535 OBD_FREE(mds_sec, strlen(mds_sec) + 1);
537 OBD_FREE(oss_sec, strlen(oss_sec) + 1);
540 } /* ll_read_super */
542 static int lustre_process_log(struct lustre_mount_data *lmd, char *profile,
543 struct config_llog_instance *cfg, int allow_recov)
545 struct lustre_cfg *lcfg = NULL;
546 struct lustre_cfg_bufs bufs;
547 struct portals_cfg pcfg;
548 char *peer = "MDS_PEER_UUID";
549 struct obd_device *obd;
550 struct lustre_handle md_conn = {0, };
551 struct obd_export *exp;
552 char *name = "mdc_dev";
554 struct obd_uuid lmv_uuid;
555 struct llog_ctxt *ctxt;
559 if (lmd_bad_magic(lmd))
562 generate_random_uuid(uuid);
563 class_uuid_unparse(uuid, &lmv_uuid);
565 if (lmd->lmd_local_nid) {
566 PCFG_INIT(pcfg, NAL_CMD_REGISTER_MYNID);
567 pcfg.pcfg_nal = lmd->lmd_nal;
568 pcfg.pcfg_nid = lmd->lmd_local_nid;
569 rc = libcfs_nal_cmd(&pcfg);
574 if (lmd->lmd_nal == SOCKNAL ||
575 lmd->lmd_nal == OPENIBNAL ||
576 lmd->lmd_nal == IIBNAL ||
577 lmd->lmd_nal == VIBNAL ||
578 lmd->lmd_nal == RANAL) {
579 PCFG_INIT(pcfg, NAL_CMD_ADD_PEER);
580 pcfg.pcfg_nal = lmd->lmd_nal;
581 pcfg.pcfg_nid = lmd->lmd_server_nid;
582 pcfg.pcfg_id = lmd->lmd_server_ipaddr;
583 pcfg.pcfg_misc = lmd->lmd_port;
584 rc = libcfs_nal_cmd(&pcfg);
588 lustre_cfg_bufs_reset(&bufs, name);
589 lustre_cfg_bufs_set_string(&bufs, 1, peer);
591 lcfg = lustre_cfg_new(LCFG_ADD_UUID, &bufs);
592 lcfg->lcfg_nal = lmd->lmd_nal;
593 lcfg->lcfg_nid = lmd->lmd_server_nid;
594 LASSERT(lcfg->lcfg_nal);
595 LASSERT(lcfg->lcfg_nid);
596 err = class_process_config(lcfg);
597 lustre_cfg_free(lcfg);
599 GOTO(out_del_conn, err);
601 lustre_cfg_bufs_reset(&bufs, name);
602 lustre_cfg_bufs_set_string(&bufs, 1, LUSTRE_MDC_NAME);
603 lustre_cfg_bufs_set_string(&bufs, 2, (char *)lmv_uuid.uuid);
605 lcfg = lustre_cfg_new(LCFG_ATTACH, &bufs);
606 err = class_process_config(lcfg);
607 lustre_cfg_free(lcfg);
609 GOTO(out_del_uuid, err);
611 lustre_cfg_bufs_reset(&bufs, name);
612 lustre_cfg_bufs_set_string(&bufs, 1, lmd->lmd_mds);
613 lustre_cfg_bufs_set_string(&bufs, 2, peer);
615 lcfg = lustre_cfg_new(LCFG_SETUP, &bufs);
616 err = class_process_config(lcfg);
617 lustre_cfg_free(lcfg);
619 GOTO(out_detach, err);
621 obd = class_name2obd(name);
623 GOTO(out_cleanup, rc = -EINVAL);
625 rc = obd_set_info(obd->obd_self_export, strlen("sec"), "sec",
626 strlen(lmd->lmd_mds_security), lmd->lmd_mds_security);
628 GOTO(out_cleanup, rc);
630 /* Disable initial recovery on this import */
631 rc = obd_set_info(obd->obd_self_export,
632 strlen("initial_recov"), "initial_recov",
633 sizeof(allow_recov), &allow_recov);
635 GOTO(out_cleanup, rc);
637 rc = obd_connect(&md_conn, obd, &lmv_uuid, NULL, 0);
639 CERROR("cannot connect to %s: rc = %d\n", lmd->lmd_mds, rc);
640 GOTO(out_cleanup, rc);
643 exp = class_conn2export(&md_conn);
645 ctxt = llog_get_context(&exp->exp_obd->obd_llogs,LLOG_CONFIG_REPL_CTXT);
646 rc = class_config_process_llog(ctxt, profile, cfg);
648 CERROR("class_config_process_llog failed: rc = %d\n", rc);
650 err = obd_disconnect(exp, 0);
654 lustre_cfg_bufs_reset(&bufs, name);
655 lcfg = lustre_cfg_new(LCFG_CLEANUP, &bufs);
656 err = class_process_config(lcfg);
657 lustre_cfg_free(lcfg);
661 lustre_cfg_bufs_reset(&bufs, name);
662 lcfg = lustre_cfg_new(LCFG_DETACH, &bufs);
663 err = class_process_config(lcfg);
664 lustre_cfg_free(lcfg);
669 lustre_cfg_bufs_reset(&bufs, name);
670 lustre_cfg_bufs_set_string(&bufs, 1, peer);
671 lcfg = lustre_cfg_new(LCFG_DEL_UUID, &bufs);
672 err = class_process_config(lcfg);
673 lustre_cfg_free(lcfg);
676 if (lmd->lmd_nal == SOCKNAL ||
677 lmd->lmd_nal == OPENIBNAL ||
678 lmd->lmd_nal == IIBNAL ||
679 lmd->lmd_nal == VIBNAL ||
680 lmd->lmd_nal == RANAL) {
683 PCFG_INIT(pcfg, NAL_CMD_DEL_PEER);
684 pcfg.pcfg_nal = lmd->lmd_nal;
685 pcfg.pcfg_nid = lmd->lmd_server_nid;
686 pcfg.pcfg_flags = 1; /* single_share */
687 err2 = libcfs_nal_cmd(&pcfg);
700 static void lustre_manual_cleanup(struct ll_sb_info *sbi)
702 struct lustre_cfg *lcfg;
703 struct lustre_cfg_bufs bufs;
704 struct obd_device *obd;
707 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) != NULL)
711 lustre_cfg_bufs_reset(&bufs, obd->obd_name);
712 lcfg = lustre_cfg_new(LCFG_CLEANUP, &bufs);
713 err = class_process_config(lcfg);
715 CERROR("cleanup failed: %s\n", obd->obd_name);
719 lcfg->lcfg_command = LCFG_DETACH;
720 err = class_process_config(lcfg);
721 lustre_cfg_free(lcfg);
723 CERROR("detach failed: %s\n", obd->obd_name);
728 if (sbi->ll_lmd != NULL)
729 class_del_profile(sbi->ll_lmd->lmd_profile);
732 int lustre_fill_super(struct super_block *sb, void *data, int silent)
734 struct lustre_mount_data * lmd = data;
735 char *lov = NULL, *lmv = NULL;
736 struct ll_sb_info *sbi;
740 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
741 if (lmd_bad_magic(lmd))
744 sbi = lustre_init_sbi(sb);
748 sbi->ll_flags |= LL_SBI_READAHEAD;
750 if (lmd->lmd_profile) {
751 struct lustre_profile *lprof;
752 struct config_llog_instance cfg;
755 if (lmd->lmd_mds[0] == '\0') {
756 CERROR("no mds name\n");
757 GOTO(out_free, err = -EINVAL);
759 lmd->lmd_mds_security[sizeof(lmd->lmd_mds_security) - 1] = 0;
760 lmd->lmd_oss_security[sizeof(lmd->lmd_oss_security) - 1] = 0;
762 OBD_ALLOC(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
763 if (sbi->ll_lmd == NULL)
764 GOTO(out_free, err = -ENOMEM);
765 memcpy(sbi->ll_lmd, lmd, sizeof(*lmd));
767 /* generate a string unique to this super, let's try
768 the address of the super itself.*/
769 len = (sizeof(sb) * 2) + 1;
770 OBD_ALLOC(sbi->ll_instance, len);
771 if (sbi->ll_instance == NULL)
772 GOTO(out_free, err = -ENOMEM);
773 sprintf(sbi->ll_instance, "%p", sb);
775 cfg.cfg_instance = sbi->ll_instance;
776 cfg.cfg_uuid = sbi->ll_sb_uuid;
777 cfg.cfg_local_nid = lmd->lmd_local_nid;
778 err = lustre_process_log(lmd, lmd->lmd_profile, &cfg, 0);
780 CERROR("Unable to process log: %s\n", lmd->lmd_profile);
784 lprof = class_get_profile(lmd->lmd_profile);
786 CERROR("No profile found: %s\n", lmd->lmd_profile);
787 GOTO(out_free, err = -EINVAL);
790 OBD_FREE(lov, strlen(lov) + 1);
791 OBD_ALLOC(lov, strlen(lprof->lp_lov) +
792 strlen(sbi->ll_instance) + 2);
793 sprintf(lov, "%s-%s", lprof->lp_lov, sbi->ll_instance);
796 OBD_FREE(lmv, strlen(lmv) + 1);
797 OBD_ALLOC(lmv, strlen(lprof->lp_lmv) +
798 strlen(sbi->ll_instance) + 2);
799 sprintf(lmv, "%s-%s", lprof->lp_lmv, sbi->ll_instance);
804 GOTO(out_free, err = -EINVAL);
809 GOTO(out_free, err = -EINVAL);
812 err = lustre_common_fill_super(sb, lmv, lov, lmd->lmd_async,
813 lmd->lmd_mds_security,
814 lmd->lmd_oss_security,
815 &lmd->lmd_nllu, &lmd->lmd_remote_flag);
822 OBD_FREE(lmv, strlen(lmv) + 1);
824 OBD_FREE(lov, strlen(lov) + 1);
830 int len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean")+1;
833 if (sbi->ll_instance != NULL) {
834 struct lustre_mount_data *lmd = sbi->ll_lmd;
835 struct config_llog_instance cfg;
838 cfg.cfg_instance = sbi->ll_instance;
839 cfg.cfg_uuid = sbi->ll_sb_uuid;
841 OBD_ALLOC(cl_prof, len);
842 sprintf(cl_prof, "%s-clean", lmd->lmd_profile);
843 err = lustre_process_log(lmd, cl_prof, &cfg, 0);
845 CERROR("Unable to process log: %s\n", cl_prof);
846 lustre_manual_cleanup(sbi);
848 OBD_FREE(cl_prof, len);
849 OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance) + 1);
851 OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
855 } /* lustre_fill_super */
857 void lustre_put_super(struct super_block *sb)
859 struct obd_device *obd;
860 struct ll_sb_info *sbi = ll_s2sbi(sb);
861 int force_umount = 0;
864 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
865 obd = class_exp2obd(sbi->ll_md_exp);
867 force_umount = obd->obd_no_recov;
870 lustre_common_put_super(sb);
871 if (sbi->ll_lmd != NULL) {
873 int len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean")+1;
875 struct config_llog_instance cfg;
878 CERROR("force umount, doing manual cleanup\n");
879 lustre_manual_cleanup(sbi);
883 cfg.cfg_instance = sbi->ll_instance;
884 cfg.cfg_uuid = sbi->ll_sb_uuid;
886 OBD_ALLOC(cl_prof, len);
887 sprintf(cl_prof, "%s-clean", sbi->ll_lmd->lmd_profile);
888 err = lustre_process_log(sbi->ll_lmd, cl_prof, &cfg, 0);
890 CERROR("Unable to process log: %s, doing manual cleanup"
892 lustre_manual_cleanup(sbi);
895 OBD_FREE(cl_prof, len);
897 OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
898 OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance) + 1);
904 } /* lustre_put_super */
906 int ll_process_config_update(struct ll_sb_info *sbi, int clean)
908 struct lustre_mount_data *lmd = sbi->ll_lmd;
909 char *profile = lmd->lmd_profile, *name = NULL;
910 struct config_llog_instance cfg;
911 int rc, namelen = 0, version;
912 struct llog_ctxt *ctxt;
918 CERROR("Client not mounted with zero-conf; cannot "
919 "process update log.\n");
923 rc = obd_cancel_unused(sbi->ll_md_exp, NULL,
924 LDLM_FL_CONFIG_CHANGE, NULL);
926 CWARN("obd_cancel_unused(mdc): %d\n", rc);
928 rc = obd_cancel_unused(sbi->ll_dt_exp, NULL,
929 LDLM_FL_CONFIG_CHANGE, NULL);
931 CWARN("obd_cancel_unused(lov): %d\n", rc);
933 cfg.cfg_instance = sbi->ll_instance;
934 cfg.cfg_uuid = sbi->ll_sb_uuid;
935 cfg.cfg_local_nid = lmd->lmd_local_nid;
937 namelen = strlen(profile) + 20; /* -clean-######### */
938 OBD_ALLOC(name, namelen);
943 version = sbi->ll_config_version - 1;
944 sprintf(name, "%s-clean-%d", profile, version);
946 version = sbi->ll_config_version + 1;
947 sprintf(name, "%s-%d", profile, version);
950 CWARN("Applying configuration log %s\n", name);
952 ctxt = llog_get_context(&sbi->ll_md_exp->exp_obd->obd_llogs,
953 LLOG_CONFIG_REPL_CTXT);
954 rc = class_config_process_llog(ctxt, name, &cfg);
956 sbi->ll_config_version = version;
957 CWARN("Finished applying configuration log %s: %d\n", name, rc);
959 if (rc == 0 && clean == 0) {
960 struct lov_desc desc;
964 valsize = sizeof(desc);
965 rc = obd_get_info(sbi->ll_dt_exp, strlen("lovdesc") + 1,
966 "lovdesc", &valsize, &desc);
968 rc = obd_init_ea_size(sbi->ll_md_exp,
969 obd_size_diskmd(sbi->ll_dt_exp, NULL),
971 sizeof(struct llog_cookie)));
973 OBD_FREE(name, namelen);
977 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
979 struct inode *inode = NULL;
981 /* NOTE: we depend on atomic igrab() -bzzz */
982 lock_res(lock->l_resource);
983 if (lock->l_ast_data) {
984 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
985 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
986 inode = igrab(lock->l_ast_data);
988 inode = lock->l_ast_data;
989 CDEBUG(inode->i_state & I_FREEING ? D_INFO : D_WARNING,
990 "l_ast_data %p is bogus: magic %0x8\n",
991 lock->l_ast_data, lli->lli_inode_magic);
995 unlock_res(lock->l_resource);
999 int null_if_equal(struct ldlm_lock *lock, void *data)
1001 if (data == lock->l_ast_data) {
1002 lock->l_ast_data = NULL;
1004 if (lock->l_req_mode != lock->l_granted_mode)
1005 LDLM_ERROR(lock,"clearing inode with ungranted lock\n");
1008 return LDLM_ITER_CONTINUE;
1011 void ll_clear_inode(struct inode *inode)
1013 struct lustre_id id;
1014 struct ll_inode_info *lli = ll_i2info(inode);
1015 struct ll_sb_info *sbi = ll_i2sbi(inode);
1018 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1019 inode->i_generation, inode);
1021 lli->lli_inode_magic = LLI_INODE_DEAD;
1022 ll_inode2id(&id, inode);
1024 clear_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &(ll_i2info(inode)->lli_flags));
1025 md_change_cbdata(sbi->ll_md_exp, &id, null_if_equal, inode);
1027 LASSERT(!lli->lli_open_fd_write_count);
1028 LASSERT(!lli->lli_open_fd_read_count);
1029 LASSERT(!lli->lli_open_fd_exec_count);
1030 if (lli->lli_mds_write_och)
1031 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_WRITE);
1032 if (lli->lli_mds_exec_och)
1033 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_EXEC);
1034 if (lli->lli_mds_read_och)
1035 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_READ);
1037 obd_change_cbdata(sbi->ll_dt_exp, lli->lli_smd,
1038 null_if_equal, inode);
1041 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1042 lli->lli_smd = NULL;
1046 obd_free_memmd(sbi->ll_md_exp,
1047 (struct lov_stripe_md **) &lli->lli_mea);
1048 lli->lli_mea = NULL;
1051 if (lli->lli_symlink_name) {
1052 OBD_FREE(lli->lli_symlink_name,
1053 strlen(lli->lli_symlink_name) + 1);
1054 lli->lli_symlink_name = NULL;
1056 lli->lli_inode_magic = LLI_INODE_DEAD;
1061 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1062 * object(s) determine the file size and mtime. Otherwise, the MDS will
1063 * keep these values until such a time that objects are allocated for it.
1064 * We do the MDS operations first, as it is checking permissions for us.
1065 * We don't to the MDS RPC if there is nothing that we want to store there,
1066 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1067 * going to do an RPC anyways.
1069 * If we are doing a truncate, we will send the mtime and ctime updates
1070 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1071 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1074 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1076 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1077 struct ll_inode_info *lli = ll_i2info(inode);
1078 struct ll_sb_info *sbi = ll_i2sbi(inode);
1079 struct ptlrpc_request *request = NULL;
1080 struct mdc_op_data *op_data;
1081 int ia_valid = attr->ia_valid;
1085 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu\n", inode->i_ino);
1086 lprocfs_counter_incr(ll_i2sbi(inode)->ll_stats, LPROC_LL_SETATTR);
1088 if (ia_valid & ATTR_SIZE) {
1089 if (attr->ia_size > ll_file_maxbytes(inode)) {
1090 CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1091 attr->ia_size, ll_file_maxbytes(inode));
1095 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1098 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1099 if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1100 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1104 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1105 if (attr->ia_valid & ATTR_CTIME) {
1106 attr->ia_ctime = CURRENT_TIME;
1107 attr->ia_valid |= ATTR_CTIME_SET;
1109 if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1110 attr->ia_atime = CURRENT_TIME;
1111 attr->ia_valid |= ATTR_ATIME_SET;
1113 if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1114 attr->ia_mtime = CURRENT_TIME;
1115 attr->ia_valid |= ATTR_MTIME_SET;
1118 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1119 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1120 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1121 LTIME_S(CURRENT_TIME));
1124 attr->ia_valid &= ~ATTR_SIZE;
1126 /* If only OST attributes being set on objects, don't do MDS RPC.
1127 * In that case, we need to check permissions and update the local
1128 * inode ourselves so we can call obdo_from_inode() always. */
1129 if (ia_valid & (lsm ? ~(ATTR_SIZE | ATTR_FROM_OPEN /*| ATTR_RAW*/) : ~0)) {
1130 struct lustre_md md;
1132 OBD_ALLOC(op_data, sizeof(*op_data));
1133 if (op_data == NULL)
1135 ll_prepare_mdc_data(op_data, inode, NULL, NULL, 0, 0);
1137 rc = md_setattr(sbi->ll_md_exp, op_data,
1138 attr, NULL, 0, NULL, 0, &request);
1139 OBD_FREE(op_data, sizeof(*op_data));
1141 ptlrpc_req_finished(request);
1142 if (rc != -EPERM && rc != -EACCES)
1143 CERROR("md_setattr fails: rc = %d\n", rc);
1147 rc = mdc_req2lustre_md(sbi->ll_md_exp, request, 0,
1148 sbi->ll_dt_exp, &md);
1150 ptlrpc_req_finished(request);
1154 /* We call inode_setattr to adjust timestamps, but we first
1155 * clear ATTR_SIZE to avoid invoking vmtruncate.
1157 * NB: ATTR_SIZE will only be set at this point if the size
1158 * resides on the MDS, ie, this file has no objects. */
1159 attr->ia_valid &= ~ATTR_SIZE;
1162 * assigning inode_setattr() to @err to disable warning that
1163 * function's result should be checked by by caller. error is
1164 * impossible here, as vmtruncate() control path is disabled.
1166 err = inode_setattr(inode, attr);
1167 ll_update_inode(inode, &md);
1168 ptlrpc_req_finished(request);
1170 if (!lsm || !S_ISREG(inode->i_mode)) {
1171 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1175 /* The OST doesn't check permissions, but the alternative is
1176 * a gratuitous RPC to the MDS. We already rely on the client
1177 * to do read/write/truncate permission checks, so is mtime OK?
1179 if (ia_valid & (ATTR_MTIME | ATTR_ATIME)) {
1180 /* from sys_utime() */
1181 if (!(ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET))) {
1182 if (current->fsuid != inode->i_uid &&
1183 (rc = ll_permission(inode, MAY_WRITE, NULL)) != 0)
1186 /* from inode_change_ok() */
1187 if (current->fsuid != inode->i_uid &&
1188 !capable(CAP_FOWNER))
1193 /* won't invoke vmtruncate, as we already cleared ATTR_SIZE */
1194 err = inode_setattr(inode, attr);
1196 * assigning inode_setattr() to @err to disable warning that
1197 * function's result should be checked by by caller. error is
1198 * impossible here, as vmtruncate() control path is disabled.
1202 /* We really need to get our PW lock before we change inode->i_size.
1203 * If we don't we can race with other i_size updaters on our node, like
1204 * ll_file_read. We can also race with i_size propogation to other
1205 * nodes through dirtying and writeback of final cached pages. This
1206 * last one is especially bad for racing o_append users on other
1208 if (ia_valid & ATTR_SIZE) {
1209 ldlm_policy_data_t policy = { .l_extent = {attr->ia_size,
1211 struct lustre_handle lockh = { 0 };
1212 int err, ast_flags = 0;
1213 /* XXX when we fix the AST intents to pass the discard-range
1214 * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
1216 if (attr->ia_size == 0)
1217 ast_flags = LDLM_AST_DISCARD_DATA;
1219 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy, &lockh,
1220 ast_flags, &ll_i2sbi(inode)->ll_seek_stime);
1225 down(&lli->lli_size_sem);
1226 lli->lli_size_pid = current->pid;
1227 rc = vmtruncate(inode, attr->ia_size);
1229 LASSERT(atomic_read(&lli->lli_size_sem.count) <= 0);
1230 lli->lli_size_pid = 0;
1231 up(&lli->lli_size_sem);
1234 err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
1236 CERROR("ll_extent_unlock failed: %d\n", err);
1240 } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET | ATTR_UID | ATTR_GID)) {
1241 struct obdo *oa = NULL;
1243 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1244 inode->i_ino, LTIME_S(attr->ia_mtime));
1250 oa->o_id = lsm->lsm_object_id;
1251 oa->o_gr = lsm->lsm_object_gr;
1252 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1254 /* adding uid and gid, needed for quota */
1255 if (ia_valid & ATTR_UID) {
1256 oa->o_uid = inode->i_uid;
1257 oa->o_valid |= OBD_MD_FLUID;
1260 if (ia_valid & ATTR_GID) {
1261 oa->o_gid = inode->i_gid;
1262 oa->o_valid |= OBD_MD_FLGID;
1265 /* putting there also fid, needed for quota too. */
1266 memcpy(obdo_id(oa), &lli->lli_id, sizeof(lli->lli_id));
1267 oa->o_valid |= OBD_MD_FLINLINE;
1269 obdo_from_inode(oa, inode, OBD_MD_FLTYPE | OBD_MD_FLATIME |
1270 OBD_MD_FLMTIME | OBD_MD_FLCTIME);
1271 rc = obd_setattr(sbi->ll_dt_exp, oa, lsm, NULL);
1274 CERROR("obd_setattr fails: rc = %d\n", rc);
1280 int ll_setattr(struct dentry *de, struct iattr *attr)
1282 LASSERT(de->d_inode);
1283 return ll_setattr_raw(de->d_inode, attr);
1286 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1287 unsigned long max_age)
1289 struct ll_sb_info *sbi = ll_s2sbi(sb);
1290 struct obd_statfs obd_osfs;
1294 rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age);
1296 CERROR("obd_statfs fails: rc = %d\n", rc);
1300 osfs->os_type = sb->s_magic;
1302 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1303 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1305 rc = obd_statfs(class_exp2obd(sbi->ll_dt_exp), &obd_osfs, max_age);
1307 CERROR("obd_statfs fails: rc = %d\n", rc);
1311 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1312 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1315 osfs->os_blocks = obd_osfs.os_blocks;
1316 osfs->os_bfree = obd_osfs.os_bfree;
1317 osfs->os_bavail = obd_osfs.os_bavail;
1319 /* If we don't have as many objects free on the OST as inodes
1320 * on the MDS, we reduce the total number of inodes to
1321 * compensate, so that the "inodes in use" number is correct.
1323 if (obd_osfs.os_ffree < osfs->os_ffree) {
1324 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1326 osfs->os_ffree = obd_osfs.os_ffree;
1332 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1334 struct obd_statfs osfs;
1337 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p\n", sb);
1338 lprocfs_counter_incr(ll_s2sbi(sb)->ll_stats, LPROC_LL_STAFS);
1340 /* For now we will always get up-to-date statfs values, but in the
1341 * future we may allow some amount of caching on the client (e.g.
1342 * from QOS or lprocfs updates). */
1343 rc = ll_statfs_internal(sb, &osfs, jiffies - 1);
1347 statfs_unpack(sfs, &osfs);
1349 if (sizeof(sfs->f_blocks) == 4) {
1350 while (osfs.os_blocks > ~0UL) {
1353 osfs.os_blocks >>= 1;
1354 osfs.os_bfree >>= 1;
1355 osfs.os_bavail >>= 1;
1359 sfs->f_blocks = osfs.os_blocks;
1360 sfs->f_bfree = osfs.os_bfree;
1361 sfs->f_bavail = osfs.os_bavail;
1366 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1368 struct ll_inode_info *lli = ll_i2info(inode);
1369 struct lov_stripe_md *lsm = md->lsm;
1370 struct mds_body *body = md->body;
1371 struct mea *mea = md->mea;
1372 struct posix_acl *ll_acl_access = md->acl_access;
1375 LASSERT((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1377 if (md->lsm && md->lsm->lsm_magic != LOV_MAGIC) {
1378 /* check for default striping info for dir. */
1379 LASSERT((mea != NULL) == ((body->valid & OBD_MD_FLDIREA) != 0));
1383 LASSERT(lsm->lsm_object_gr > 0);
1384 if (lli->lli_smd == NULL) {
1386 lli->lli_maxbytes = lsm->lsm_maxbytes;
1387 if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1388 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1391 if (memcmp(lli->lli_smd, lsm, sizeof(*lsm))) {
1392 CERROR("lsm mismatch for inode %ld\n",
1394 CERROR("lli_smd:\n");
1395 dump_lsm(D_ERROR, lli->lli_smd);
1397 dump_lsm(D_ERROR, lsm);
1400 /* XXX FIXME -- We should decide on a safer (atomic) and
1401 * more elegant way to update the lsm */
1402 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1403 lli->lli_smd->lsm_oinfo[i].loi_id =
1404 lsm->lsm_oinfo[i].loi_id;
1405 lli->lli_smd->lsm_oinfo[i].loi_gr =
1406 lsm->lsm_oinfo[i].loi_gr;
1407 lli->lli_smd->lsm_oinfo[i].loi_ost_idx =
1408 lsm->lsm_oinfo[i].loi_ost_idx;
1409 lli->lli_smd->lsm_oinfo[i].loi_ost_gen =
1410 lsm->lsm_oinfo[i].loi_ost_gen;
1413 /* bug 2844 - limit i_blksize for broken user-space apps */
1414 LASSERTF(lsm->lsm_xfersize != 0, "%lu\n", lsm->lsm_xfersize);
1415 inode->i_blksize = min(lsm->lsm_xfersize, LL_MAX_BLKSIZE);
1416 if (lli->lli_smd != lsm)
1417 obd_free_memmd(ll_i2dtexp(inode), &lsm);
1421 if (lli->lli_mea == NULL) {
1424 if (memcmp(lli->lli_mea, mea, body->eadatasize)) {
1425 CERROR("mea mismatch for inode %lu\n",
1430 if (lli->lli_mea != mea)
1431 obd_free_memmd(ll_i2mdexp(inode),
1432 (struct lov_stripe_md **) &mea);
1435 if (body->valid & OBD_MD_FID)
1436 id_assign_fid(&lli->lli_id, &body->id1);
1438 if (body->valid & OBD_MD_FLID)
1439 id_ino(&lli->lli_id) = id_ino(&body->id1);
1441 if (body->valid & OBD_MD_FLGENER)
1442 id_gen(&lli->lli_id) = id_gen(&body->id1);
1444 spin_lock(&lli->lli_lock);
1445 if (ll_acl_access != NULL) {
1446 if (lli->lli_acl_access != NULL)
1447 posix_acl_release(lli->lli_acl_access);
1448 lli->lli_acl_access = ll_acl_access;
1450 spin_unlock(&lli->lli_lock);
1452 if (body->valid & OBD_MD_FLID)
1453 inode->i_ino = id_ino(&body->id1);
1454 if (body->valid & OBD_MD_FLGENER)
1455 inode->i_generation = id_gen(&body->id1);
1456 if (body->valid & OBD_MD_FLATIME)
1457 LTIME_S(inode->i_atime) = body->atime;
1458 if (body->valid & OBD_MD_FLMTIME &&
1459 body->mtime > LTIME_S(inode->i_mtime)) {
1460 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to %u\n",
1461 inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
1462 LTIME_S(inode->i_mtime) = body->mtime;
1464 if (body->valid & OBD_MD_FLCTIME &&
1465 body->ctime > LTIME_S(inode->i_ctime))
1466 LTIME_S(inode->i_ctime) = body->ctime;
1467 if (body->valid & OBD_MD_FLMODE) {
1468 inode->i_mode = (inode->i_mode & S_IFMT) |
1469 (body->mode & ~S_IFMT);
1471 if (body->valid & OBD_MD_FLTYPE) {
1472 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1473 (body->mode & S_IFMT);
1475 if (body->valid & OBD_MD_FLUID)
1476 inode->i_uid = body->uid;
1477 if (body->valid & OBD_MD_FLGID)
1478 inode->i_gid = body->gid;
1479 if (body->valid & OBD_MD_FLFLAGS)
1480 inode->i_flags = body->flags;
1481 if (body->valid & OBD_MD_FLNLINK)
1482 inode->i_nlink = body->nlink;
1483 if (body->valid & OBD_MD_FLRDEV)
1484 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1485 inode->i_rdev = body->rdev;
1487 inode->i_rdev = old_decode_dev(body->rdev);
1489 if (body->valid & OBD_MD_FLSIZE)
1490 inode->i_size = body->size;
1491 if (body->valid & OBD_MD_FLBLOCKS)
1492 inode->i_blocks = body->blocks;
1494 if (body->valid & OBD_MD_FLSIZE)
1495 set_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &lli->lli_flags);
1497 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1498 inode->i_dev = (kdev_t)id_group(&lli->lli_id);
1500 LASSERT(id_fid(&lli->lli_id) != 0);
1503 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
1504 static struct backing_dev_info ll_backing_dev_info = {
1505 .ra_pages = 0, /* No readahead */
1506 .memory_backed = 0, /* Does contribute to dirty memory */
1510 void ll_read_inode2(struct inode *inode, void *opaque)
1512 struct lustre_md *md = opaque;
1513 struct ll_inode_info *lli = ll_i2info(inode);
1516 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1517 inode->i_generation, inode);
1521 LASSERT(!lli->lli_smd);
1523 /* Core attributes from the MDS first. This is a new inode, and
1524 * the VFS doesn't zero times in the core inode so we have to do
1525 * it ourselves. They will be overwritten by either MDS or OST
1526 * attributes - we just need to make sure they aren't newer. */
1527 LTIME_S(inode->i_mtime) = 0;
1528 LTIME_S(inode->i_atime) = 0;
1529 LTIME_S(inode->i_ctime) = 0;
1532 ll_update_inode(inode, md);
1534 /* OIDEBUG(inode); */
1536 if (S_ISREG(inode->i_mode)) {
1537 inode->i_op = &ll_file_inode_operations;
1538 inode->i_fop = &ll_file_operations;
1539 inode->i_mapping->a_ops = &ll_aops;
1541 } else if (S_ISDIR(inode->i_mode)) {
1542 inode->i_op = &ll_dir_inode_operations;
1543 inode->i_fop = &ll_dir_operations;
1544 inode->i_mapping->a_ops = &ll_dir_aops;
1546 } else if (S_ISLNK(inode->i_mode)) {
1547 inode->i_op = &ll_fast_symlink_inode_operations;
1550 inode->i_op = &ll_special_inode_operations;
1552 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
1553 init_special_inode(inode, inode->i_mode,
1554 kdev_t_to_nr(inode->i_rdev));
1556 /* initializing backing dev info. */
1557 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
1559 init_special_inode(inode, inode->i_mode, inode->i_rdev);
1561 lli->ll_save_ifop = inode->i_fop;
1563 if (S_ISCHR(inode->i_mode))
1564 inode->i_fop = &ll_special_chr_inode_fops;
1565 else if (S_ISBLK(inode->i_mode))
1566 inode->i_fop = &ll_special_blk_inode_fops;
1567 else if (S_ISFIFO(inode->i_mode))
1568 inode->i_fop = &ll_special_fifo_inode_fops;
1569 else if (S_ISSOCK(inode->i_mode))
1570 inode->i_fop = &ll_special_sock_inode_fops;
1572 CWARN("saved %p, replaced with %p\n", lli->ll_save_ifop,
1575 if (lli->ll_save_ifop->owner) {
1576 CWARN("%p has owner %p\n", lli->ll_save_ifop,
1577 lli->ll_save_ifop->owner);
1583 void ll_delete_inode(struct inode *inode)
1585 struct ll_sb_info *sbi = ll_i2sbi(inode);
1586 struct lustre_id id;
1590 ll_inode2id(&id, inode);
1592 rc = md_delete_inode(sbi->ll_md_exp, &id);
1594 CERROR("md_delete_inode() failed, error %d\n",
1602 int ll_iocontrol(struct inode *inode, struct file *file,
1603 unsigned int cmd, unsigned long arg)
1605 struct ll_sb_info *sbi = ll_i2sbi(inode);
1606 struct ptlrpc_request *req = NULL;
1611 case EXT3_IOC_GETFLAGS: {
1612 struct lustre_id id;
1613 __u64 valid = OBD_MD_FLFLAGS;
1614 struct mds_body *body;
1616 ll_inode2id(&id, inode);
1617 rc = md_getattr(sbi->ll_md_exp, &id, valid, NULL, 0, 0, &req);
1619 CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1623 body = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*body));
1625 if (body->flags & S_APPEND)
1626 flags |= EXT3_APPEND_FL;
1627 if (body->flags & S_IMMUTABLE)
1628 flags |= EXT3_IMMUTABLE_FL;
1629 if (body->flags & S_NOATIME)
1630 flags |= EXT3_NOATIME_FL;
1632 ptlrpc_req_finished (req);
1634 RETURN(put_user(flags, (int *)arg));
1636 case EXT3_IOC_SETFLAGS: {
1637 struct mdc_op_data *op_data;
1640 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1642 if (get_user(flags, (int *)arg))
1649 OBD_ALLOC(op_data, sizeof(*op_data));
1650 if (op_data == NULL) {
1654 ll_prepare_mdc_data(op_data, inode, NULL, NULL, 0, 0);
1656 memset(&attr, 0x0, sizeof(attr));
1657 attr.ia_attr_flags = flags;
1658 attr.ia_valid |= ATTR_ATTR_FLAG;
1660 rc = md_setattr(sbi->ll_md_exp, op_data,
1661 &attr, NULL, 0, NULL, 0, &req);
1662 OBD_FREE(op_data, sizeof(*op_data));
1664 ptlrpc_req_finished(req);
1665 if (rc != -EPERM && rc != -EACCES)
1666 CERROR("md_setattr fails: rc = %d\n", rc);
1670 ptlrpc_req_finished(req);
1672 oa->o_id = lsm->lsm_object_id;
1673 oa->o_gr = lsm->lsm_object_gr;
1674 oa->o_flags = flags;
1675 oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS | OBD_MD_FLGROUP;
1677 rc = obd_setattr(sbi->ll_dt_exp, oa, lsm, NULL);
1680 if (rc != -EPERM && rc != -EACCES)
1681 CERROR("md_setattr fails: rc = %d\n", rc);
1685 if (flags & EXT3_APPEND_FL)
1686 inode->i_flags |= S_APPEND;
1688 inode->i_flags &= ~S_APPEND;
1689 if (flags & EXT3_IMMUTABLE_FL)
1690 inode->i_flags |= S_IMMUTABLE;
1692 inode->i_flags &= ~S_IMMUTABLE;
1693 if (flags & EXT3_NOATIME_FL)
1694 inode->i_flags |= S_NOATIME;
1696 inode->i_flags &= ~S_NOATIME;
1707 /* this is only called in the case of forced umount. */
1708 void ll_umount_begin(struct super_block *sb)
1710 struct ll_sb_info *sbi = ll_s2sbi(sb);
1711 struct obd_ioctl_data ioc_data = { 0 };
1712 struct obd_device *obd;
1715 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1716 sb->s_count, atomic_read(&sb->s_active));
1718 obd = class_exp2obd(sbi->ll_md_exp);
1720 CERROR("Invalid MDC connection handle "LPX64"\n",
1721 sbi->ll_md_exp->exp_handle.h_cookie);
1725 obd->obd_no_recov = 1;
1726 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
1727 sizeof(ioc_data), &ioc_data, NULL);
1729 obd = class_exp2obd(sbi->ll_dt_exp);
1731 CERROR("Invalid LOV connection handle "LPX64"\n",
1732 sbi->ll_dt_exp->exp_handle.h_cookie);
1737 obd->obd_no_recov = 1;
1738 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
1739 sizeof(ioc_data), &ioc_data, NULL);
1742 * really, we'd like to wait until there are no requests outstanding,
1743 * and then continue. For now, we just invalidate the requests,
1744 * schedule, and hope.
1751 int ll_prep_inode(struct obd_export *dt_exp, struct obd_export *md_exp,
1752 struct inode **inode, struct ptlrpc_request *req,
1753 int offset, struct super_block *sb)
1755 struct lustre_md md;
1758 rc = mdc_req2lustre_md(md_exp, req, offset, dt_exp, &md);
1763 ll_update_inode(*inode, &md);
1766 *inode = ll_iget(sb, id_ino(&md.body->id1), &md);
1767 if (*inode == NULL || is_bad_inode(*inode)) {
1768 /* free the lsm if we allocated one above */
1770 obd_free_memmd(dt_exp, &md.lsm);
1772 obd_free_memmd(md_exp,
1773 (struct lov_stripe_md**)&md.mea);
1775 CERROR("new_inode -fatal: rc %d\n", rc);
1782 int ll_show_options(struct seq_file *m, struct vfsmount *mnt)
1784 struct ll_sb_info *sbi = ll_s2sbi(mnt->mnt_sb);
1785 struct lustre_mount_data *lmd = sbi->ll_lmd;
1788 seq_printf(m, ",mds_sec=%s,oss_sec=%s",
1789 lmd->lmd_mds_security, lmd->lmd_oss_security);
1791 seq_printf(m, ",%s", sbi->ll_remote ? "remote" : "local");
1792 if (sbi->ll_remote && lmd) {
1793 seq_printf(m, ",nllu=%u:%u", lmd->lmd_nllu, lmd->lmd_nllg);
1799 int ll_get_fid(struct obd_export *exp, struct lustre_id *idp,
1800 char *filename, struct lustre_id *ret)
1802 struct ptlrpc_request *request = NULL;
1803 struct mds_body *body;
1806 rc = md_getattr_lock(exp, idp, filename, strlen(filename) + 1,
1807 OBD_MD_FID, 0, &request);
1809 CDEBUG(D_INFO, "md_getattr_lock failed on %s: rc %d\n",
1814 body = lustre_msg_buf(request->rq_repmsg, 0, sizeof(*body));
1815 LASSERT(body != NULL);
1816 LASSERT_REPSWABBED(request, 0);
1819 ptlrpc_req_finished(request);
1824 int ll_flush_cred(struct inode *inode)
1826 struct ll_sb_info *sbi = ll_i2sbi(inode);
1827 uid_t uid = current->fsuid;
1830 /* XXX to avoid adding api, we simply use set_info() interface
1831 * to notify underlying obds. set_info() is more like a ioctl() now...
1833 if (sbi->ll_md_exp) {
1834 rc = obd_set_info(sbi->ll_md_exp,
1835 strlen("flush_cred"), "flush_cred",
1841 if (sbi->ll_dt_exp) {
1842 rc = obd_set_info(sbi->ll_dt_exp,
1843 strlen("flush_cred"), "flush_cred",