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_tick = GNS_TICK_TIMEOUT;
87 sbi->ll_gns_timeout = GNS_MOUNT_TIMEOUT;
89 sbi->ll_gns_timer.data = (unsigned long)sbi;
90 sbi->ll_gns_timer.function = ll_gns_timer_callback;
91 init_timer(&sbi->ll_gns_timer);
95 generate_random_uuid(uuid);
96 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
100 void lustre_free_sbi(struct super_block *sb)
102 struct ll_sb_info *sbi = ll_s2sbi(sb);
106 list_del(&sbi->ll_gns_sbi_head);
107 del_timer(&sbi->ll_gns_timer);
108 OBD_FREE(sbi, sizeof(*sbi));
110 ll_set_sbi(sb, NULL);
114 int lustre_init_dt_desc(struct ll_sb_info *sbi)
119 valsize = sizeof(sbi->ll_dt_desc);
120 memset(&sbi->ll_dt_desc, 0, sizeof(sbi->ll_dt_desc));
121 rc = obd_get_info(sbi->ll_dt_exp, strlen("lovdesc") + 1,
122 "lovdesc", &valsize, &sbi->ll_dt_desc);
126 extern struct dentry_operations ll_d_ops;
128 int lustre_common_fill_super(struct super_block *sb, char *lmv, char *lov,
129 int async, char *security, __u32 *nllu,
132 struct ll_sb_info *sbi = ll_s2sbi(sb);
133 struct ptlrpc_request *request = NULL;
134 struct lustre_handle dt_conn = {0, };
135 struct lustre_handle md_conn = {0, };
136 struct obd_connect_data *data;
137 struct inode *root = NULL;
138 struct obd_device *obd;
139 struct obd_statfs osfs;
146 obd = class_name2obd(lmv);
148 CERROR("MDC %s: not setup or attached\n", lmv);
151 obd_set_info(obd->obd_self_export, strlen("async"), "async",
152 sizeof(async), &async);
154 if ((*remote & (OBD_CONNECT_LOCAL | OBD_CONNECT_REMOTE)) ==
155 (OBD_CONNECT_LOCAL | OBD_CONNECT_REMOTE)) {
156 CERROR("wrong remote flag "LPX64"\n", *remote);
160 OBD_ALLOC(data, sizeof(*data));
164 data->ocd_connect_flags |= *remote & (OBD_CONNECT_LOCAL |
166 memcpy(data->ocd_nllu, nllu, sizeof(data->ocd_nllu));
168 if (security == NULL)
171 err = obd_set_info(obd->obd_self_export, strlen("sec"), "sec",
172 strlen(security), security);
174 CERROR("LMV %s: failed to set security %s, err %d\n",
176 OBD_FREE(data, sizeof(*data));
180 if (proc_lustre_fs_root) {
181 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
184 CERROR("could not register mount in /proc/lustre");
187 err = obd_connect(&md_conn, obd, &sbi->ll_sb_uuid, data,
188 OBD_OPT_REAL_CLIENT);
190 CERROR("An MDS (lmv %s) is performing recovery, of which this"
191 " client is not a part. Please wait for recovery to "
192 "complete, abort, or time out.\n", lmv);
195 CERROR("cannot connect to %s: rc = %d\n", lmv, err);
198 sbi->ll_md_exp = class_conn2export(&md_conn);
199 err = obd_statfs(obd, &osfs, jiffies - HZ);
203 if (!osfs.os_bsize) {
204 CERROR("Invalid block size is detected.");
208 sb->s_magic = LL_SUPER_MAGIC;
209 sb->s_blocksize = osfs.os_bsize;
210 sb->s_blocksize_bits = log2(osfs.os_bsize);
211 sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
213 devno = get_uuid2int(sbi->ll_md_exp->exp_obd->obd_uuid.uuid,
214 strlen(sbi->ll_md_exp->exp_obd->obd_uuid.uuid));
218 /* after statfs, we are supposed to have connected to MDSs,
219 * so it's ok to check remote flag returned.
221 valsize = sizeof(&sbi->ll_remote);
222 err = obd_get_info(sbi->ll_md_exp, strlen("remote_flag"), "remote_flag",
223 &valsize, &sbi->ll_remote);
225 CERROR("fail to obtain remote flag\n");
229 obd = class_name2obd(lov);
231 CERROR("OSC %s: not setup or attached\n", lov);
234 obd_set_info(obd->obd_self_export, strlen("async"), "async",
235 sizeof(async), &async);
237 err = obd_connect(&dt_conn, obd, &sbi->ll_sb_uuid, data,
238 OBD_OPT_REAL_CLIENT);
240 CERROR("An OST (lov %s) is performing recovery, of which this"
241 " client is not a part. Please wait for recovery to "
242 "complete, abort, or time out.\n", lov);
245 CERROR("cannot connect to %s: rc = %d\n", lov, err);
248 sbi->ll_dt_exp = class_conn2export(&dt_conn);
250 err = lustre_init_dt_desc(sbi);
252 int mdsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
253 obd_init_ea_size(sbi->ll_md_exp, mdsize,
254 sbi->ll_dt_desc.ld_tgt_count *
255 sizeof(struct llog_cookie));
258 err = md_getstatus(sbi->ll_md_exp, &sbi->ll_rootid);
260 CERROR("cannot mds_connect: rc = %d\n", err);
263 CDEBUG(D_SUPER, "rootid "DLID4"\n", OLID4(&sbi->ll_rootid));
265 sb->s_op = &lustre_super_operations;
268 * XXX: move this to after cbd setup? */
269 err = md_getattr(sbi->ll_md_exp, &sbi->ll_rootid,
270 (OBD_MD_FLNOTOBD | OBD_MD_FLBLOCKS | OBD_MD_FID), NULL, 0,
273 CERROR("md_getattr failed for root: rc = %d\n", err);
277 err = mdc_req2lustre_md(sbi->ll_md_exp, request, 0,
278 sbi->ll_dt_exp, &md);
280 CERROR("failed to understand root inode md: rc = %d\n", err);
281 ptlrpc_req_finished(request);
285 LASSERT(id_ino(&sbi->ll_rootid) != 0);
286 root = ll_iget(sb, id_ino(&sbi->ll_rootid), &md);
288 ptlrpc_req_finished(request);
290 if (root == NULL || is_bad_inode(root)) {
291 CERROR("lustre_lite: bad iget4 for root\n");
292 GOTO(out_root, err = -EBADF);
295 err = ll_close_thread_start(&sbi->ll_lcq);
297 CERROR("cannot start close thread: rc %d\n", err);
301 ll_gns_add_timer(sbi);
303 /* making vm readahead 0 for 2.4.x. In the case of 2.6.x,
304 backing dev info assigned to inode mapping is used for
305 determining maximal readahead. */
306 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0)) && \
307 !defined(KERNEL_HAS_AS_MAX_READAHEAD)
308 /* bug 2805 - set VM readahead to zero */
309 vm_max_readahead = vm_min_readahead = 0;
312 sb->s_root = d_alloc_root(root);
313 sb->s_root->d_op = &ll_d_ops;
315 sb->s_flags |= MS_POSIXACL;
317 CWARN("Enabling PDIROPS\n");
318 sb->s_flags |= S_PDIROPS;
322 OBD_FREE(data, sizeof(*data));
328 obd_disconnect(sbi->ll_dt_exp, 0);
330 obd_disconnect(sbi->ll_md_exp, 0);
333 OBD_FREE(data, sizeof(*data));
334 lprocfs_unregister_mountpoint(sbi);
338 void lustre_common_put_super(struct super_block *sb)
340 struct ll_sb_info *sbi = ll_s2sbi(sb);
341 struct hlist_node *tmp, *next;
344 ll_gns_del_timer(sbi);
345 ll_close_thread_shutdown(sbi->ll_lcq);
347 list_del(&sbi->ll_conn_chain);
348 obd_disconnect(sbi->ll_dt_exp, 0);
350 lprocfs_unregister_mountpoint(sbi);
351 if (sbi->ll_proc_root) {
352 lprocfs_remove(sbi->ll_proc_root);
353 sbi->ll_proc_root = NULL;
356 obd_disconnect(sbi->ll_md_exp, 0);
358 // We do this to get rid of orphaned dentries. That is not really trw.
359 hlist_for_each_safe(tmp, next, &sbi->ll_orphan_dentry_list) {
360 struct dentry *dentry = hlist_entry(tmp, struct dentry, d_hash);
361 CWARN("orphan dentry %.*s (%p->%p) at unmount\n",
362 dentry->d_name.len, dentry->d_name.name, dentry, next);
363 shrink_dcache_parent(dentry);
368 char *ll_read_opt(const char *opt, char *data)
374 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
375 if (strncmp(opt, data, strlen(opt)))
377 if ((value = strchr(data, '=')) == NULL)
381 OBD_ALLOC(retval, strlen(value) + 1);
383 CERROR("out of memory!\n");
387 memcpy(retval, value, strlen(value)+1);
388 CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
392 int ll_set_opt(const char *opt, char *data, int fl)
396 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
397 if (strncmp(opt, data, strlen(opt)))
403 void ll_options(char *options, char **lov, char **lmv, char **sec,
404 int *async, int *flags)
407 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
408 char *opt_ptr = options;
418 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
419 for (this_char = strtok (options, ",");
421 this_char = strtok (NULL, ",")) {
423 while ((this_char = strsep (&opt_ptr, ",")) != NULL) {
425 CDEBUG(D_SUPER, "this_char %s\n", this_char);
426 if (!*lov && (*lov = ll_read_opt("osc", this_char)))
428 if (!*lmv && (*lmv = ll_read_opt("mdc", this_char)))
430 if (!strncmp(this_char, "lasync", strlen("lasync"))) {
434 if (!*sec && (*sec = ll_read_opt("sec", this_char)))
436 if (!(*flags & LL_SBI_NOLCK) &&
437 ((*flags) = (*flags) |
438 ll_set_opt("nolock", this_char,
446 void ll_lli_init(struct ll_inode_info *lli)
448 sema_init(&lli->lli_open_sem, 1);
449 sema_init(&lli->lli_size_sem, 1);
451 lli->lli_size_pid = 0;
452 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
453 spin_lock_init(&lli->lli_lock);
454 INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
455 lli->lli_inode_magic = LLI_INODE_MAGIC;
456 memset(&lli->lli_id, 0, sizeof(lli->lli_id));
457 sema_init(&lli->lli_och_sem, 1);
458 lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
459 lli->lli_mds_exec_och = NULL;
460 lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
461 lli->lli_open_fd_exec_count = 0;
464 int ll_fill_super(struct super_block *sb, void *data, int silent)
466 struct ll_sb_info *sbi;
471 __u32 nllu[2] = { NOBODY_UID, NOBODY_GID };
472 __u64 remote_flag = 0;
475 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
477 sbi = lustre_init_sbi(sb);
481 sbi->ll_flags |= LL_SBI_READAHEAD;
482 ll_options(data, &lov, &lmv, &sec, &async, &sbi->ll_flags);
486 GOTO(out, err = -EINVAL);
491 GOTO(out, err = -EINVAL);
494 err = lustre_common_fill_super(sb, lmv, lov, async, sec, nllu,
502 OBD_FREE(sec, strlen(sec) + 1);
504 OBD_FREE(lmv, strlen(lmv) + 1);
506 OBD_FREE(lov, strlen(lov) + 1);
508 } /* ll_read_super */
510 static int lustre_process_log(struct lustre_mount_data *lmd, char *profile,
511 struct config_llog_instance *cfg, int allow_recov)
513 struct lustre_cfg lcfg;
514 struct portals_cfg pcfg;
515 char *peer = "MDS_PEER_UUID";
516 struct obd_device *obd;
517 struct lustre_handle md_conn = {0, };
518 struct obd_export *exp;
519 char *name = "mdc_dev";
521 struct obd_uuid lmv_uuid;
522 struct llog_ctxt *ctxt;
526 if (lmd_bad_magic(lmd))
529 generate_random_uuid(uuid);
530 class_uuid_unparse(uuid, &lmv_uuid);
532 if (lmd->lmd_local_nid) {
533 PCFG_INIT(pcfg, NAL_CMD_REGISTER_MYNID);
534 pcfg.pcfg_nal = lmd->lmd_nal;
535 pcfg.pcfg_nid = lmd->lmd_local_nid;
536 rc = libcfs_nal_cmd(&pcfg);
541 if (lmd->lmd_nal == SOCKNAL ||
542 lmd->lmd_nal == OPENIBNAL ||
543 lmd->lmd_nal == IIBNAL ||
544 lmd->lmd_nal == VIBNAL ||
545 lmd->lmd_nal == RANAL) {
546 PCFG_INIT(pcfg, NAL_CMD_ADD_PEER);
547 pcfg.pcfg_nal = lmd->lmd_nal;
548 pcfg.pcfg_nid = lmd->lmd_server_nid;
549 pcfg.pcfg_id = lmd->lmd_server_ipaddr;
550 pcfg.pcfg_misc = lmd->lmd_port;
551 rc = libcfs_nal_cmd(&pcfg);
556 LCFG_INIT(lcfg, LCFG_ADD_UUID, name);
557 lcfg.lcfg_nid = lmd->lmd_server_nid;
558 lcfg.lcfg_inllen1 = strlen(peer) + 1;
559 lcfg.lcfg_inlbuf1 = peer;
560 lcfg.lcfg_nal = lmd->lmd_nal;
561 rc = class_process_config(&lcfg);
563 GOTO(out_del_conn, rc);
565 LCFG_INIT(lcfg, LCFG_ATTACH, name);
566 lcfg.lcfg_inlbuf1 = "mdc";
567 lcfg.lcfg_inllen1 = strlen(lcfg.lcfg_inlbuf1) + 1;
568 lcfg.lcfg_inlbuf2 = lmv_uuid.uuid;
569 lcfg.lcfg_inllen2 = strlen(lcfg.lcfg_inlbuf2) + 1;
570 err = class_process_config(&lcfg);
572 GOTO(out_del_uuid, rc);
574 LCFG_INIT(lcfg, LCFG_SETUP, name);
575 lcfg.lcfg_inlbuf1 = lmd->lmd_mds;
576 lcfg.lcfg_inllen1 = strlen(lcfg.lcfg_inlbuf1) + 1;
577 lcfg.lcfg_inlbuf2 = peer;
578 lcfg.lcfg_inllen2 = strlen(lcfg.lcfg_inlbuf2) + 1;
579 rc = class_process_config(&lcfg);
581 GOTO(out_detach, rc);
583 obd = class_name2obd(name);
585 GOTO(out_cleanup, rc = -EINVAL);
587 rc = obd_set_info(obd->obd_self_export, strlen("sec"), "sec",
588 strlen(lmd->lmd_security), lmd->lmd_security);
590 GOTO(out_cleanup, rc);
592 /* Disable initial recovery on this import */
593 rc = obd_set_info(obd->obd_self_export,
594 strlen("initial_recov"), "initial_recov",
595 sizeof(allow_recov), &allow_recov);
597 GOTO(out_cleanup, rc);
599 rc = obd_connect(&md_conn, obd, &lmv_uuid, NULL, 0);
601 CERROR("cannot connect to %s: rc = %d\n", lmd->lmd_mds, rc);
602 GOTO(out_cleanup, rc);
605 exp = class_conn2export(&md_conn);
607 ctxt = llog_get_context(&exp->exp_obd->obd_llogs,LLOG_CONFIG_REPL_CTXT);
608 rc = class_config_process_llog(ctxt, profile, cfg);
610 CERROR("class_config_process_llog failed: rc = %d\n", rc);
612 err = obd_disconnect(exp, 0);
616 LCFG_INIT(lcfg, LCFG_CLEANUP, name);
617 err = class_process_config(&lcfg);
622 LCFG_INIT(lcfg, LCFG_DETACH, name);
623 err = class_process_config(&lcfg);
628 LCFG_INIT(lcfg, LCFG_DEL_UUID, name);
629 lcfg.lcfg_inllen1 = strlen(peer) + 1;
630 lcfg.lcfg_inlbuf1 = peer;
631 err = class_process_config(&lcfg);
634 if (lmd->lmd_nal == SOCKNAL ||
635 lmd->lmd_nal == OPENIBNAL ||
636 lmd->lmd_nal == IIBNAL ||
637 lmd->lmd_nal == VIBNAL ||
638 lmd->lmd_nal == RANAL) {
641 PCFG_INIT(pcfg, NAL_CMD_DEL_PEER);
642 pcfg.pcfg_nal = lmd->lmd_nal;
643 pcfg.pcfg_nid = lmd->lmd_server_nid;
644 pcfg.pcfg_flags = 1; /* single_share */
645 err2 = libcfs_nal_cmd(&pcfg);
658 static void lustre_manual_cleanup(struct ll_sb_info *sbi)
660 struct lustre_cfg lcfg;
661 struct obd_device *obd;
664 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) != NULL)
668 LCFG_INIT(lcfg, LCFG_CLEANUP, obd->obd_name);
669 err = class_process_config(&lcfg);
671 CERROR("cleanup failed: %s\n", obd->obd_name);
675 LCFG_INIT(lcfg, LCFG_DETACH, obd->obd_name);
676 err = class_process_config(&lcfg);
678 CERROR("detach failed: %s\n", obd->obd_name);
683 if (sbi->ll_lmd != NULL)
684 class_del_profile(sbi->ll_lmd->lmd_profile);
687 int lustre_fill_super(struct super_block *sb, void *data, int silent)
689 struct lustre_mount_data * lmd = data;
690 char *lov = NULL, *lmv = NULL;
691 struct ll_sb_info *sbi;
695 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
696 if (lmd_bad_magic(lmd))
699 sbi = lustre_init_sbi(sb);
703 sbi->ll_flags |= LL_SBI_READAHEAD;
705 if (lmd->lmd_profile) {
706 struct lustre_profile *lprof;
707 struct config_llog_instance cfg;
710 if (lmd->lmd_mds[0] == '\0') {
711 CERROR("no mds name\n");
712 GOTO(out_free, err = -EINVAL);
714 lmd->lmd_security[sizeof(lmd->lmd_security) - 1] = 0;
716 OBD_ALLOC(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
717 if (sbi->ll_lmd == NULL)
718 GOTO(out_free, err = -ENOMEM);
719 memcpy(sbi->ll_lmd, lmd, sizeof(*lmd));
721 /* generate a string unique to this super, let's try
722 the address of the super itself.*/
723 len = (sizeof(sb) * 2) + 1;
724 OBD_ALLOC(sbi->ll_instance, len);
725 if (sbi->ll_instance == NULL)
726 GOTO(out_free, err = -ENOMEM);
727 sprintf(sbi->ll_instance, "%p", sb);
729 cfg.cfg_instance = sbi->ll_instance;
730 cfg.cfg_uuid = sbi->ll_sb_uuid;
731 cfg.cfg_local_nid = lmd->lmd_local_nid;
732 err = lustre_process_log(lmd, lmd->lmd_profile, &cfg, 0);
734 CERROR("Unable to process log: %s\n", lmd->lmd_profile);
738 lprof = class_get_profile(lmd->lmd_profile);
740 CERROR("No profile found: %s\n", lmd->lmd_profile);
741 GOTO(out_free, err = -EINVAL);
744 OBD_FREE(lov, strlen(lov) + 1);
745 OBD_ALLOC(lov, strlen(lprof->lp_lov) +
746 strlen(sbi->ll_instance) + 2);
747 sprintf(lov, "%s-%s", lprof->lp_lov, sbi->ll_instance);
750 OBD_FREE(lmv, strlen(lmv) + 1);
751 OBD_ALLOC(lmv, strlen(lprof->lp_lmv) +
752 strlen(sbi->ll_instance) + 2);
753 sprintf(lmv, "%s-%s", lprof->lp_lmv, sbi->ll_instance);
758 GOTO(out_free, err = -EINVAL);
763 GOTO(out_free, err = -EINVAL);
766 err = lustre_common_fill_super(sb, lmv, lov, lmd->lmd_async,
767 lmd->lmd_security, &lmd->lmd_nllu,
768 &lmd->lmd_remote_flag);
775 OBD_FREE(lmv, strlen(lmv) + 1);
777 OBD_FREE(lov, strlen(lov) + 1);
783 int len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean")+1;
786 if (sbi->ll_instance != NULL) {
787 struct lustre_mount_data *lmd = sbi->ll_lmd;
788 struct config_llog_instance cfg;
791 cfg.cfg_instance = sbi->ll_instance;
792 cfg.cfg_uuid = sbi->ll_sb_uuid;
794 OBD_ALLOC(cl_prof, len);
795 sprintf(cl_prof, "%s-clean", lmd->lmd_profile);
796 err = lustre_process_log(lmd, cl_prof, &cfg, 0);
798 CERROR("Unable to process log: %s\n", cl_prof);
799 lustre_manual_cleanup(sbi);
801 OBD_FREE(cl_prof, len);
802 OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance) + 1);
804 OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
808 } /* lustre_fill_super */
810 void lustre_put_super(struct super_block *sb)
812 struct obd_device *obd;
813 struct ll_sb_info *sbi = ll_s2sbi(sb);
814 int force_umount = 0;
817 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
818 obd = class_exp2obd(sbi->ll_md_exp);
820 force_umount = obd->obd_no_recov;
823 lustre_common_put_super(sb);
824 if (sbi->ll_lmd != NULL) {
826 int len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean")+1;
828 struct config_llog_instance cfg;
831 CERROR("force umount, doing manual cleanup\n");
832 lustre_manual_cleanup(sbi);
836 cfg.cfg_instance = sbi->ll_instance;
837 cfg.cfg_uuid = sbi->ll_sb_uuid;
839 OBD_ALLOC(cl_prof, len);
840 sprintf(cl_prof, "%s-clean", sbi->ll_lmd->lmd_profile);
841 err = lustre_process_log(sbi->ll_lmd, cl_prof, &cfg, 0);
843 CERROR("Unable to process log: %s, doing manual cleanup"
845 lustre_manual_cleanup(sbi);
848 OBD_FREE(cl_prof, len);
850 OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
851 OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance) + 1);
857 } /* lustre_put_super */
859 int ll_process_config_update(struct ll_sb_info *sbi, int clean)
861 struct obd_export *md_exp = sbi->ll_md_exp;
862 struct lustre_mount_data *lmd = sbi->ll_lmd;
863 struct llog_ctxt *ctxt;
864 struct config_llog_instance cfg;
865 char *profile = lmd->lmd_profile, *name = NULL;
866 int rc, namelen = 0, version;
872 CERROR("Client not mounted with zero-conf; cannot process "
877 rc = ldlm_cli_cancel_unused(md_exp->exp_obd->obd_namespace, NULL,
878 LDLM_FL_CONFIG_CHANGE, NULL);
880 CWARN("ldlm_cli_cancel_unused(mdc): %d\n", rc);
882 rc = obd_cancel_unused(sbi->ll_dt_exp, NULL, LDLM_FL_CONFIG_CHANGE,
885 CWARN("obd_cancel_unused(lov): %d\n", rc);
887 cfg.cfg_instance = sbi->ll_instance;
888 cfg.cfg_uuid = sbi->ll_sb_uuid;
889 cfg.cfg_local_nid = lmd->lmd_local_nid;
891 namelen = strlen(profile) + 20; /* -clean-######### */
892 OBD_ALLOC(name, namelen);
897 version = sbi->ll_config_version - 1;
898 sprintf(name, "%s-clean-%d", profile, version);
900 version = sbi->ll_config_version + 1;
901 sprintf(name, "%s-%d", profile, version);
904 CWARN("Applying configuration log %s\n", name);
906 ctxt = llog_get_context(&md_exp->exp_obd->obd_llogs,
907 LLOG_CONFIG_REPL_CTXT);
908 rc = class_config_process_llog(ctxt, name, &cfg);
910 sbi->ll_config_version = version;
911 CWARN("Finished applying configuration log %s: %d\n", name, rc);
913 if (rc == 0 && clean == 0) {
914 struct lov_desc desc;
916 valsize = sizeof(desc);
917 rc = obd_get_info(sbi->ll_dt_exp, strlen("lovdesc") + 1,
918 "lovdesc", &valsize, &desc);
920 rc = obd_init_ea_size(md_exp,
921 obd_size_diskmd(sbi->ll_dt_exp, NULL),
923 sizeof(struct llog_cookie)));
925 OBD_FREE(name, namelen);
929 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
931 struct inode *inode = NULL;
932 l_lock(&lock->l_resource->lr_namespace->ns_lock);
933 if (lock->l_ast_data) {
934 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
935 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
936 inode = igrab(lock->l_ast_data);
938 inode = lock->l_ast_data;
939 CDEBUG(inode->i_state & I_FREEING ? D_INFO : D_WARNING,
940 "l_ast_data %p is bogus: magic %0x8\n",
941 lock->l_ast_data, lli->lli_inode_magic);
945 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
949 int null_if_equal(struct ldlm_lock *lock, void *data)
951 if (data == lock->l_ast_data) {
952 lock->l_ast_data = NULL;
954 if (lock->l_req_mode != lock->l_granted_mode)
955 LDLM_ERROR(lock,"clearing inode with ungranted lock\n");
958 return LDLM_ITER_CONTINUE;
961 void ll_clear_inode(struct inode *inode)
964 struct ll_inode_info *lli = ll_i2info(inode);
965 struct ll_sb_info *sbi = ll_i2sbi(inode);
968 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
969 inode->i_generation, inode);
971 lli->lli_inode_magic = LLI_INODE_DEAD;
972 ll_inode2id(&id, inode);
974 clear_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &(ll_i2info(inode)->lli_flags));
975 md_change_cbdata(sbi->ll_md_exp, &id, null_if_equal, inode);
977 LASSERT(!lli->lli_open_fd_write_count);
978 LASSERT(!lli->lli_open_fd_read_count);
979 LASSERT(!lli->lli_open_fd_exec_count);
981 if (lli->lli_mds_write_och)
982 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_WRITE);
983 if (lli->lli_mds_exec_och)
984 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_EXEC);
985 if (lli->lli_mds_read_och)
986 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_READ);
989 obd_change_cbdata(sbi->ll_dt_exp, lli->lli_smd,
990 null_if_equal, inode);
993 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
998 obd_free_memmd(sbi->ll_md_exp,
999 (struct lov_stripe_md **) &lli->lli_mea);
1000 lli->lli_mea = NULL;
1003 if (lli->lli_symlink_name) {
1004 OBD_FREE(lli->lli_symlink_name,
1005 strlen(lli->lli_symlink_name) + 1);
1006 lli->lli_symlink_name = NULL;
1008 lli->lli_inode_magic = LLI_INODE_DEAD;
1013 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1014 * object(s) determine the file size and mtime. Otherwise, the MDS will
1015 * keep these values until such a time that objects are allocated for it.
1016 * We do the MDS operations first, as it is checking permissions for us.
1017 * We don't to the MDS RPC if there is nothing that we want to store there,
1018 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1019 * going to do an RPC anyways.
1021 * If we are doing a truncate, we will send the mtime and ctime updates
1022 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1023 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1026 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1028 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1029 struct ll_sb_info *sbi = ll_i2sbi(inode);
1030 struct ptlrpc_request *request = NULL;
1031 struct mdc_op_data *op_data;
1032 int ia_valid = attr->ia_valid;
1036 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu\n", inode->i_ino);
1037 lprocfs_counter_incr(ll_i2sbi(inode)->ll_stats, LPROC_LL_SETATTR);
1039 if (ia_valid & ATTR_SIZE) {
1040 if (attr->ia_size > ll_file_maxbytes(inode)) {
1041 CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1042 attr->ia_size, ll_file_maxbytes(inode));
1046 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1049 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1050 if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1051 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1055 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1056 if (attr->ia_valid & ATTR_CTIME) {
1057 attr->ia_ctime = CURRENT_TIME;
1058 attr->ia_valid |= ATTR_CTIME_SET;
1060 if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1061 attr->ia_atime = CURRENT_TIME;
1062 attr->ia_valid |= ATTR_ATIME_SET;
1064 if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1065 attr->ia_mtime = CURRENT_TIME;
1066 attr->ia_valid |= ATTR_MTIME_SET;
1069 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1070 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1071 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1072 LTIME_S(CURRENT_TIME));
1075 attr->ia_valid &= ~ATTR_SIZE;
1077 /* If only OST attributes being set on objects, don't do MDS RPC.
1078 * In that case, we need to check permissions and update the local
1079 * inode ourselves so we can call obdo_from_inode() always. */
1080 if (ia_valid & (lsm ? ~(ATTR_SIZE | ATTR_FROM_OPEN /*| ATTR_RAW*/) : ~0)) {
1081 struct lustre_md md;
1083 OBD_ALLOC(op_data, sizeof(*op_data));
1084 if (op_data == NULL)
1086 ll_prepare_mdc_data(op_data, inode, NULL, NULL, 0, 0);
1088 rc = md_setattr(sbi->ll_md_exp, op_data,
1089 attr, NULL, 0, NULL, 0, &request);
1090 OBD_FREE(op_data, sizeof(*op_data));
1092 ptlrpc_req_finished(request);
1093 if (rc != -EPERM && rc != -EACCES)
1094 CERROR("md_setattr fails: rc = %d\n", rc);
1098 rc = mdc_req2lustre_md(sbi->ll_md_exp, request, 0,
1099 sbi->ll_dt_exp, &md);
1101 ptlrpc_req_finished(request);
1105 /* We call inode_setattr to adjust timestamps, but we first
1106 * clear ATTR_SIZE to avoid invoking vmtruncate.
1108 * NB: ATTR_SIZE will only be set at this point if the size
1109 * resides on the MDS, ie, this file has no objects. */
1110 attr->ia_valid &= ~ATTR_SIZE;
1111 err = inode_setattr(inode, attr);
1113 CERROR("inode_setattr() failed, inode=%lu/%u(%p), "
1114 "err = %d\n", (unsigned long)inode->i_ino,
1115 inode->i_generation, inode, err);
1116 /* should we go to error path here? --umka */
1119 ll_update_inode(inode, &md);
1120 ptlrpc_req_finished(request);
1122 if (!lsm || !S_ISREG(inode->i_mode)) {
1123 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1127 /* The OST doesn't check permissions, but the alternative is
1128 * a gratuitous RPC to the MDS. We already rely on the client
1129 * to do read/write/truncate permission checks, so is mtime OK?
1131 if (ia_valid & (ATTR_MTIME | ATTR_ATIME)) {
1132 /* from sys_utime() */
1133 if (!(ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET))) {
1134 if (current->fsuid != inode->i_uid &&
1135 (rc=ll_permission(inode,MAY_WRITE,NULL))!=0)
1138 /* from inode_change_ok() */
1139 if (current->fsuid != inode->i_uid &&
1140 !capable(CAP_FOWNER))
1145 /* Won't invoke vmtruncate, as we already cleared ATTR_SIZE */
1146 err = inode_setattr(inode, attr);
1148 CERROR("inode_setattr() failed, inode=%lu/%u(%p), "
1149 "err = %d\n", (unsigned long)inode->i_ino,
1150 inode->i_generation, inode, err);
1151 /* should we go to error path here? --umka */
1155 /* We really need to get our PW lock before we change inode->i_size.
1156 * If we don't we can race with other i_size updaters on our node, like
1157 * ll_file_read. We can also race with i_size propogation to other
1158 * nodes through dirtying and writeback of final cached pages. This
1159 * last one is especially bad for racing o_append users on other
1161 if (ia_valid & ATTR_SIZE) {
1162 ldlm_policy_data_t policy = { .l_extent = {attr->ia_size,
1164 struct lustre_handle lockh = { 0 };
1165 struct ll_inode_info *lli = ll_i2info(inode);
1166 int err, ast_flags = 0;
1167 /* XXX when we fix the AST intents to pass the discard-range
1168 * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
1170 if (attr->ia_size == 0)
1171 ast_flags = LDLM_AST_DISCARD_DATA;
1173 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy, &lockh,
1174 ast_flags, &ll_i2sbi(inode)->ll_seek_stime);
1179 down(&lli->lli_size_sem);
1180 lli->lli_size_pid = current->pid;
1181 rc = vmtruncate(inode, attr->ia_size);
1183 LASSERT(atomic_read(&lli->lli_size_sem.count) <= 0);
1184 lli->lli_size_pid = 0;
1185 up(&lli->lli_size_sem);
1188 err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
1190 CERROR("ll_extent_unlock failed: %d\n", err);
1194 } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
1195 struct obdo *oa = NULL;
1197 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1198 inode->i_ino, LTIME_S(attr->ia_mtime));
1204 oa->o_id = lsm->lsm_object_id;
1205 oa->o_gr = lsm->lsm_object_gr;
1206 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1207 obdo_from_inode(oa, inode, OBD_MD_FLTYPE | OBD_MD_FLATIME |
1208 OBD_MD_FLMTIME | OBD_MD_FLCTIME);
1209 rc = obd_setattr(sbi->ll_dt_exp, oa, lsm, NULL);
1212 CERROR("obd_setattr fails: rc = %d\n", rc);
1217 int ll_setattr(struct dentry *de, struct iattr *attr)
1219 LASSERT(de->d_inode);
1220 return ll_setattr_raw(de->d_inode, attr);
1223 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1224 unsigned long max_age)
1226 struct ll_sb_info *sbi = ll_s2sbi(sb);
1227 struct obd_statfs obd_osfs;
1231 rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age);
1233 CERROR("obd_statfs fails: rc = %d\n", rc);
1237 osfs->os_type = sb->s_magic;
1239 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1240 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1242 rc = obd_statfs(class_exp2obd(sbi->ll_dt_exp), &obd_osfs, max_age);
1244 CERROR("obd_statfs fails: rc = %d\n", rc);
1248 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1249 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1252 osfs->os_blocks = obd_osfs.os_blocks;
1253 osfs->os_bfree = obd_osfs.os_bfree;
1254 osfs->os_bavail = obd_osfs.os_bavail;
1256 /* If we don't have as many objects free on the OST as inodes
1257 * on the MDS, we reduce the total number of inodes to
1258 * compensate, so that the "inodes in use" number is correct.
1260 if (obd_osfs.os_ffree < osfs->os_ffree) {
1261 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1263 osfs->os_ffree = obd_osfs.os_ffree;
1269 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1271 struct obd_statfs osfs;
1274 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p\n", sb);
1275 lprocfs_counter_incr(ll_s2sbi(sb)->ll_stats, LPROC_LL_STAFS);
1277 /* For now we will always get up-to-date statfs values, but in the
1278 * future we may allow some amount of caching on the client (e.g.
1279 * from QOS or lprocfs updates). */
1280 rc = ll_statfs_internal(sb, &osfs, jiffies - 1);
1284 statfs_unpack(sfs, &osfs);
1286 if (sizeof(sfs->f_blocks) == 4) {
1287 while (osfs.os_blocks > ~0UL) {
1290 osfs.os_blocks >>= 1;
1291 osfs.os_bfree >>= 1;
1292 osfs.os_bavail >>= 1;
1296 sfs->f_blocks = osfs.os_blocks;
1297 sfs->f_bfree = osfs.os_bfree;
1298 sfs->f_bavail = osfs.os_bavail;
1303 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1305 struct ll_inode_info *lli = ll_i2info(inode);
1306 struct lov_stripe_md *lsm = md->lsm;
1307 struct mds_body *body = md->body;
1308 struct mea *mea = md->mea;
1309 struct posix_acl *ll_acl_access = md->acl_access;
1312 LASSERT((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1314 if (md->lsm && md->lsm->lsm_magic != LOV_MAGIC) {
1315 /* check for default striping info for dir. */
1316 LASSERT((mea != NULL) == ((body->valid & OBD_MD_FLDIREA) != 0));
1320 LASSERT(lsm->lsm_object_gr > 0);
1321 if (lli->lli_smd == NULL) {
1323 lli->lli_maxbytes = lsm->lsm_maxbytes;
1324 if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1325 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1328 if (memcmp(lli->lli_smd, lsm, sizeof(*lsm))) {
1329 CERROR("lsm mismatch for inode %ld\n",
1331 CERROR("lli_smd:\n");
1332 dump_lsm(D_ERROR, lli->lli_smd);
1334 dump_lsm(D_ERROR, lsm);
1337 /* XXX FIXME -- We should decide on a safer (atomic) and
1338 * more elegant way to update the lsm */
1339 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1340 lli->lli_smd->lsm_oinfo[i].loi_id =
1341 lsm->lsm_oinfo[i].loi_id;
1342 lli->lli_smd->lsm_oinfo[i].loi_gr =
1343 lsm->lsm_oinfo[i].loi_gr;
1344 lli->lli_smd->lsm_oinfo[i].loi_ost_idx =
1345 lsm->lsm_oinfo[i].loi_ost_idx;
1346 lli->lli_smd->lsm_oinfo[i].loi_ost_gen =
1347 lsm->lsm_oinfo[i].loi_ost_gen;
1350 /* bug 2844 - limit i_blksize for broken user-space apps */
1351 LASSERTF(lsm->lsm_xfersize != 0, "%lu\n", lsm->lsm_xfersize);
1352 inode->i_blksize = min(lsm->lsm_xfersize, LL_MAX_BLKSIZE);
1353 if (lli->lli_smd != lsm)
1354 obd_free_memmd(ll_i2dtexp(inode), &lsm);
1358 if (lli->lli_mea == NULL) {
1361 if (memcmp(lli->lli_mea, mea, body->eadatasize)) {
1362 CERROR("mea mismatch for inode %lu\n",
1367 if (lli->lli_mea != mea)
1368 obd_free_memmd(ll_i2mdexp(inode),
1369 (struct lov_stripe_md **) &mea);
1372 if (body->valid & OBD_MD_FID)
1373 id_assign_fid(&lli->lli_id, &body->id1);
1375 if (body->valid & OBD_MD_FLID)
1376 id_ino(&lli->lli_id) = id_ino(&body->id1);
1378 if (body->valid & OBD_MD_FLGENER)
1379 id_gen(&lli->lli_id) = id_gen(&body->id1);
1381 spin_lock(&lli->lli_lock);
1382 if (ll_acl_access != NULL) {
1383 if (lli->lli_acl_access != NULL)
1384 posix_acl_release(lli->lli_acl_access);
1385 lli->lli_acl_access = ll_acl_access;
1387 spin_unlock(&lli->lli_lock);
1389 if (body->valid & OBD_MD_FLID)
1390 inode->i_ino = id_ino(&body->id1);
1391 if (body->valid & OBD_MD_FLGENER)
1392 inode->i_generation = id_gen(&body->id1);
1393 if (body->valid & OBD_MD_FLATIME)
1394 LTIME_S(inode->i_atime) = body->atime;
1395 if (body->valid & OBD_MD_FLMTIME &&
1396 body->mtime > LTIME_S(inode->i_mtime)) {
1397 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to %u\n",
1398 inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
1399 LTIME_S(inode->i_mtime) = body->mtime;
1401 if (body->valid & OBD_MD_FLCTIME &&
1402 body->ctime > LTIME_S(inode->i_ctime))
1403 LTIME_S(inode->i_ctime) = body->ctime;
1404 if (body->valid & OBD_MD_FLMODE) {
1405 inode->i_mode = (inode->i_mode & S_IFMT) |
1406 (body->mode & ~S_IFMT);
1408 if (body->valid & OBD_MD_FLTYPE) {
1409 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1410 (body->mode & S_IFMT);
1412 if (body->valid & OBD_MD_FLUID)
1413 inode->i_uid = body->uid;
1414 if (body->valid & OBD_MD_FLGID)
1415 inode->i_gid = body->gid;
1416 if (body->valid & OBD_MD_FLFLAGS)
1417 inode->i_flags = body->flags;
1418 if (body->valid & OBD_MD_FLNLINK)
1419 inode->i_nlink = body->nlink;
1420 if (body->valid & OBD_MD_FLRDEV)
1421 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1422 inode->i_rdev = body->rdev;
1424 inode->i_rdev = old_decode_dev(body->rdev);
1426 if (body->valid & OBD_MD_FLSIZE)
1427 inode->i_size = body->size;
1428 if (body->valid & OBD_MD_FLBLOCKS)
1429 inode->i_blocks = body->blocks;
1431 if (body->valid & OBD_MD_FLSIZE)
1432 set_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &lli->lli_flags);
1434 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1435 inode->i_dev = (kdev_t)id_group(&lli->lli_id);
1437 LASSERT(id_fid(&lli->lli_id) != 0);
1440 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
1441 static struct backing_dev_info ll_backing_dev_info = {
1442 .ra_pages = 0, /* No readahead */
1443 .memory_backed = 0, /* Does contribute to dirty memory */
1447 void ll_read_inode2(struct inode *inode, void *opaque)
1449 struct lustre_md *md = opaque;
1450 struct ll_inode_info *lli = ll_i2info(inode);
1453 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1454 inode->i_generation, inode);
1458 LASSERT(!lli->lli_smd);
1460 /* Core attributes from the MDS first. This is a new inode, and
1461 * the VFS doesn't zero times in the core inode so we have to do
1462 * it ourselves. They will be overwritten by either MDS or OST
1463 * attributes - we just need to make sure they aren't newer. */
1464 LTIME_S(inode->i_mtime) = 0;
1465 LTIME_S(inode->i_atime) = 0;
1466 LTIME_S(inode->i_ctime) = 0;
1469 ll_update_inode(inode, md);
1471 /* OIDEBUG(inode); */
1473 if (S_ISREG(inode->i_mode)) {
1474 inode->i_op = &ll_file_inode_operations;
1475 inode->i_fop = &ll_file_operations;
1476 inode->i_mapping->a_ops = &ll_aops;
1478 } else if (S_ISDIR(inode->i_mode)) {
1479 inode->i_op = &ll_dir_inode_operations;
1480 inode->i_fop = &ll_dir_operations;
1481 inode->i_mapping->a_ops = &ll_dir_aops;
1483 } else if (S_ISLNK(inode->i_mode)) {
1484 inode->i_op = &ll_fast_symlink_inode_operations;
1487 inode->i_op = &ll_special_inode_operations;
1489 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
1490 init_special_inode(inode, inode->i_mode,
1491 kdev_t_to_nr(inode->i_rdev));
1493 /* initializing backing dev info. */
1494 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
1496 init_special_inode(inode, inode->i_mode, inode->i_rdev);
1498 lli->ll_save_ifop = inode->i_fop;
1500 if (S_ISCHR(inode->i_mode))
1501 inode->i_fop = &ll_special_chr_inode_fops;
1502 else if (S_ISBLK(inode->i_mode))
1503 inode->i_fop = &ll_special_blk_inode_fops;
1504 else if (S_ISFIFO(inode->i_mode))
1505 inode->i_fop = &ll_special_fifo_inode_fops;
1506 else if (S_ISSOCK(inode->i_mode))
1507 inode->i_fop = &ll_special_sock_inode_fops;
1509 CWARN("saved %p, replaced with %p\n", lli->ll_save_ifop,
1512 if (lli->ll_save_ifop->owner) {
1513 CWARN("%p has owner %p\n", lli->ll_save_ifop,
1514 lli->ll_save_ifop->owner);
1520 void ll_delete_inode(struct inode *inode)
1522 struct ll_sb_info *sbi = ll_i2sbi(inode);
1523 struct lustre_id id;
1527 ll_inode2id(&id, inode);
1529 rc = md_delete_inode(sbi->ll_md_exp, &id);
1531 CERROR("md_delete_inode() failed, error %d\n",
1539 int ll_iocontrol(struct inode *inode, struct file *file,
1540 unsigned int cmd, unsigned long arg)
1542 struct ll_sb_info *sbi = ll_i2sbi(inode);
1543 struct ptlrpc_request *req = NULL;
1548 case EXT3_IOC_GETFLAGS: {
1549 struct lustre_id id;
1550 __u64 valid = OBD_MD_FLFLAGS;
1551 struct mds_body *body;
1553 ll_inode2id(&id, inode);
1554 rc = md_getattr(sbi->ll_md_exp, &id, valid, NULL, 0, 0, &req);
1556 CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1560 body = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*body));
1562 if (body->flags & S_APPEND)
1563 flags |= EXT3_APPEND_FL;
1564 if (body->flags & S_IMMUTABLE)
1565 flags |= EXT3_IMMUTABLE_FL;
1566 if (body->flags & S_NOATIME)
1567 flags |= EXT3_NOATIME_FL;
1569 ptlrpc_req_finished (req);
1571 RETURN(put_user(flags, (int *)arg));
1573 case EXT3_IOC_SETFLAGS: {
1574 struct mdc_op_data *op_data;
1577 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1579 if (get_user(flags, (int *)arg))
1586 OBD_ALLOC(op_data, sizeof(*op_data));
1587 if (op_data == NULL) {
1591 ll_prepare_mdc_data(op_data, inode, NULL, NULL, 0, 0);
1593 memset(&attr, 0x0, sizeof(attr));
1594 attr.ia_attr_flags = flags;
1595 attr.ia_valid |= ATTR_ATTR_FLAG;
1597 rc = md_setattr(sbi->ll_md_exp, op_data,
1598 &attr, NULL, 0, NULL, 0, &req);
1599 OBD_FREE(op_data, sizeof(*op_data));
1601 ptlrpc_req_finished(req);
1602 if (rc != -EPERM && rc != -EACCES)
1603 CERROR("md_setattr fails: rc = %d\n", rc);
1607 ptlrpc_req_finished(req);
1609 oa->o_id = lsm->lsm_object_id;
1610 oa->o_gr = lsm->lsm_object_gr;
1611 oa->o_flags = flags;
1612 oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS | OBD_MD_FLGROUP;
1614 rc = obd_setattr(sbi->ll_dt_exp, oa, lsm, NULL);
1617 if (rc != -EPERM && rc != -EACCES)
1618 CERROR("md_setattr fails: rc = %d\n", rc);
1622 if (flags & EXT3_APPEND_FL)
1623 inode->i_flags |= S_APPEND;
1625 inode->i_flags &= ~S_APPEND;
1626 if (flags & EXT3_IMMUTABLE_FL)
1627 inode->i_flags |= S_IMMUTABLE;
1629 inode->i_flags &= ~S_IMMUTABLE;
1630 if (flags & EXT3_NOATIME_FL)
1631 inode->i_flags |= S_NOATIME;
1633 inode->i_flags &= ~S_NOATIME;
1644 /* this is only called in the case of forced umount. */
1645 void ll_umount_begin(struct super_block *sb)
1647 struct ll_sb_info *sbi = ll_s2sbi(sb);
1648 struct obd_ioctl_data ioc_data = { 0 };
1649 struct obd_device *obd;
1652 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1653 sb->s_count, atomic_read(&sb->s_active));
1655 obd = class_exp2obd(sbi->ll_md_exp);
1657 CERROR("Invalid MDC connection handle "LPX64"\n",
1658 sbi->ll_md_exp->exp_handle.h_cookie);
1662 obd->obd_no_recov = 1;
1663 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
1664 sizeof(ioc_data), &ioc_data, NULL);
1666 obd = class_exp2obd(sbi->ll_dt_exp);
1668 CERROR("Invalid LOV connection handle "LPX64"\n",
1669 sbi->ll_dt_exp->exp_handle.h_cookie);
1674 obd->obd_no_recov = 1;
1675 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
1676 sizeof(ioc_data), &ioc_data, NULL);
1679 * really, we'd like to wait until there are no requests outstanding,
1680 * and then continue. For now, we just invalidate the requests,
1681 * schedule, and hope.
1688 int ll_prep_inode(struct obd_export *dt_exp, struct obd_export *md_exp,
1689 struct inode **inode, struct ptlrpc_request *req,
1690 int offset, struct super_block *sb)
1692 struct lustre_md md;
1695 rc = mdc_req2lustre_md(md_exp, req, offset, dt_exp, &md);
1700 ll_update_inode(*inode, &md);
1703 *inode = ll_iget(sb, id_ino(&md.body->id1), &md);
1704 if (*inode == NULL || is_bad_inode(*inode)) {
1705 /* free the lsm if we allocated one above */
1707 obd_free_memmd(dt_exp, &md.lsm);
1709 obd_free_memmd(md_exp,
1710 (struct lov_stripe_md**)&md.mea);
1712 CERROR("new_inode -fatal: rc %d\n", rc);
1719 int ll_get_fid(struct obd_export *exp, struct lustre_id *idp,
1720 char *filename, struct lustre_id *ret)
1722 struct ptlrpc_request *request = NULL;
1723 struct mds_body *body;
1726 rc = md_getattr_lock(exp, idp, filename, strlen(filename) + 1,
1727 OBD_MD_FID, 0, &request);
1729 CDEBUG(D_INFO, "md_getattr_lock failed on %s: rc %d\n",
1734 body = lustre_msg_buf(request->rq_repmsg, 0, sizeof(*body));
1735 LASSERT(body != NULL);
1736 LASSERT_REPSWABBED(request, 0);
1739 ptlrpc_req_finished(request);