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 atomic_set(&sbi->ll_gns_enabled, 1);
87 sbi->ll_gns_tick = GNS_TICK_TIMEOUT;
88 sbi->ll_gns_timeout = GNS_MOUNT_TIMEOUT;
90 sbi->ll_gns_timer.data = (unsigned long)sbi;
91 sbi->ll_gns_timer.function = ll_gns_timer_callback;
92 init_timer(&sbi->ll_gns_timer);
96 generate_random_uuid(uuid);
97 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
101 void lustre_free_sbi(struct super_block *sb)
103 struct ll_sb_info *sbi = ll_s2sbi(sb);
107 list_del(&sbi->ll_gns_sbi_head);
108 del_timer(&sbi->ll_gns_timer);
109 OBD_FREE(sbi, sizeof(*sbi));
111 ll_set_sbi(sb, NULL);
115 int lustre_init_dt_desc(struct ll_sb_info *sbi)
120 valsize = sizeof(sbi->ll_dt_desc);
121 memset(&sbi->ll_dt_desc, 0, sizeof(sbi->ll_dt_desc));
122 rc = obd_get_info(sbi->ll_dt_exp, strlen("lovdesc") + 1,
123 "lovdesc", &valsize, &sbi->ll_dt_desc);
127 extern struct dentry_operations ll_d_ops;
129 int lustre_common_fill_super(struct super_block *sb, char *lmv, char *lov,
130 int async, char *security, __u32 *nllu,
133 struct ll_sb_info *sbi = ll_s2sbi(sb);
134 struct ptlrpc_request *request = NULL;
135 struct lustre_handle dt_conn = {0, };
136 struct lustre_handle md_conn = {0, };
137 struct obd_connect_data *data;
138 struct inode *root = NULL;
139 struct obd_device *obd;
140 struct obd_statfs osfs;
147 obd = class_name2obd(lmv);
149 CERROR("MDC %s: not setup or attached\n", lmv);
152 obd_set_info(obd->obd_self_export, strlen("async"), "async",
153 sizeof(async), &async);
155 if ((*remote & (OBD_CONNECT_LOCAL | OBD_CONNECT_REMOTE)) ==
156 (OBD_CONNECT_LOCAL | OBD_CONNECT_REMOTE)) {
157 CERROR("wrong remote flag "LPX64"\n", *remote);
161 OBD_ALLOC(data, sizeof(*data));
165 data->ocd_connect_flags |= *remote & (OBD_CONNECT_LOCAL |
167 memcpy(data->ocd_nllu, nllu, sizeof(data->ocd_nllu));
169 if (security == NULL)
172 err = obd_set_info(obd->obd_self_export, strlen("sec"), "sec",
173 strlen(security), security);
175 CERROR("LMV %s: failed to set security %s, err %d\n",
177 OBD_FREE(data, sizeof(*data));
181 if (proc_lustre_fs_root) {
182 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
185 CERROR("could not register mount in /proc/lustre");
188 err = obd_connect(&md_conn, obd, &sbi->ll_sb_uuid, data,
189 OBD_OPT_REAL_CLIENT);
191 CERROR("An MDS (lmv %s) is performing recovery, of which this"
192 " client is not a part. Please wait for recovery to "
193 "complete, abort, or time out.\n", lmv);
196 CERROR("cannot connect to %s: rc = %d\n", lmv, err);
199 sbi->ll_md_exp = class_conn2export(&md_conn);
200 err = obd_statfs(obd, &osfs, jiffies - HZ);
204 if (!osfs.os_bsize) {
205 CERROR("Invalid block size is detected.");
209 sb->s_magic = LL_SUPER_MAGIC;
210 sb->s_blocksize = osfs.os_bsize;
211 sb->s_blocksize_bits = log2(osfs.os_bsize);
212 sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
214 devno = get_uuid2int(sbi->ll_md_exp->exp_obd->obd_uuid.uuid,
215 strlen(sbi->ll_md_exp->exp_obd->obd_uuid.uuid));
219 /* after statfs, we are supposed to have connected to MDSs,
220 * so it's ok to check remote flag returned.
222 valsize = sizeof(&sbi->ll_remote);
223 err = obd_get_info(sbi->ll_md_exp, strlen("remote_flag"), "remote_flag",
224 &valsize, &sbi->ll_remote);
226 CERROR("fail to obtain remote flag\n");
230 obd = class_name2obd(lov);
232 CERROR("OSC %s: not setup or attached\n", lov);
235 obd_set_info(obd->obd_self_export, strlen("async"), "async",
236 sizeof(async), &async);
238 err = obd_connect(&dt_conn, obd, &sbi->ll_sb_uuid, data,
239 OBD_OPT_REAL_CLIENT);
241 CERROR("An OST (lov %s) is performing recovery, of which this"
242 " client is not a part. Please wait for recovery to "
243 "complete, abort, or time out.\n", lov);
246 CERROR("cannot connect to %s: rc = %d\n", lov, err);
249 sbi->ll_dt_exp = class_conn2export(&dt_conn);
251 err = lustre_init_dt_desc(sbi);
253 int mdsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
254 obd_init_ea_size(sbi->ll_md_exp, mdsize,
255 sbi->ll_dt_desc.ld_tgt_count *
256 sizeof(struct llog_cookie));
259 err = md_getstatus(sbi->ll_md_exp, &sbi->ll_rootid);
261 CERROR("cannot mds_connect: rc = %d\n", err);
264 CDEBUG(D_SUPER, "rootid "DLID4"\n", OLID4(&sbi->ll_rootid));
266 sb->s_op = &lustre_super_operations;
269 * XXX: move this to after cbd setup? */
270 err = md_getattr(sbi->ll_md_exp, &sbi->ll_rootid,
271 (OBD_MD_FLNOTOBD | OBD_MD_FLBLOCKS | OBD_MD_FID), NULL, 0,
274 CERROR("md_getattr failed for root: rc = %d\n", err);
278 err = mdc_req2lustre_md(sbi->ll_md_exp, request, 0,
279 sbi->ll_dt_exp, &md);
281 CERROR("failed to understand root inode md: rc = %d\n", err);
282 ptlrpc_req_finished(request);
286 LASSERT(id_ino(&sbi->ll_rootid) != 0);
287 root = ll_iget(sb, id_ino(&sbi->ll_rootid), &md);
289 ptlrpc_req_finished(request);
291 if (root == NULL || is_bad_inode(root)) {
292 CERROR("lustre_lite: bad iget4 for root\n");
293 GOTO(out_root, err = -EBADF);
296 err = ll_close_thread_start(&sbi->ll_lcq);
298 CERROR("cannot start close thread: rc %d\n", err);
302 ll_gns_add_timer(sbi);
304 /* making vm readahead 0 for 2.4.x. In the case of 2.6.x,
305 backing dev info assigned to inode mapping is used for
306 determining maximal readahead. */
307 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0)) && \
308 !defined(KERNEL_HAS_AS_MAX_READAHEAD)
309 /* bug 2805 - set VM readahead to zero */
310 vm_max_readahead = vm_min_readahead = 0;
313 sb->s_root = d_alloc_root(root);
314 sb->s_root->d_op = &ll_d_ops;
316 sb->s_flags |= MS_POSIXACL;
318 CWARN("Enabling PDIROPS\n");
319 sb->s_flags |= S_PDIROPS;
323 OBD_FREE(data, sizeof(*data));
329 obd_disconnect(sbi->ll_dt_exp, 0);
331 obd_disconnect(sbi->ll_md_exp, 0);
334 OBD_FREE(data, sizeof(*data));
335 lprocfs_unregister_mountpoint(sbi);
339 void lustre_common_put_super(struct super_block *sb)
341 struct ll_sb_info *sbi = ll_s2sbi(sb);
342 struct hlist_node *tmp, *next;
345 ll_gns_del_timer(sbi);
346 ll_close_thread_shutdown(sbi->ll_lcq);
348 list_del(&sbi->ll_conn_chain);
349 obd_disconnect(sbi->ll_dt_exp, 0);
351 lprocfs_unregister_mountpoint(sbi);
352 if (sbi->ll_proc_root) {
353 lprocfs_remove(sbi->ll_proc_root);
354 sbi->ll_proc_root = NULL;
357 obd_disconnect(sbi->ll_md_exp, 0);
359 // We do this to get rid of orphaned dentries. That is not really trw.
360 hlist_for_each_safe(tmp, next, &sbi->ll_orphan_dentry_list) {
361 struct dentry *dentry = hlist_entry(tmp, struct dentry, d_hash);
362 CWARN("orphan dentry %.*s (%p->%p) at unmount\n",
363 dentry->d_name.len, dentry->d_name.name, dentry, next);
364 shrink_dcache_parent(dentry);
369 char *ll_read_opt(const char *opt, char *data)
375 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
376 if (strncmp(opt, data, strlen(opt)))
378 if ((value = strchr(data, '=')) == NULL)
382 OBD_ALLOC(retval, strlen(value) + 1);
384 CERROR("out of memory!\n");
388 memcpy(retval, value, strlen(value)+1);
389 CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
393 int ll_set_opt(const char *opt, char *data, int fl)
397 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
398 if (strncmp(opt, data, strlen(opt)))
404 void ll_options(char *options, char **lov, char **lmv, char **sec,
405 int *async, int *flags)
408 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
409 char *opt_ptr = options;
419 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
420 for (this_char = strtok (options, ",");
422 this_char = strtok (NULL, ",")) {
424 while ((this_char = strsep (&opt_ptr, ",")) != NULL) {
426 CDEBUG(D_SUPER, "this_char %s\n", this_char);
427 if (!*lov && (*lov = ll_read_opt("osc", this_char)))
429 if (!*lmv && (*lmv = ll_read_opt("mdc", this_char)))
431 if (!strncmp(this_char, "lasync", strlen("lasync"))) {
435 if (!*sec && (*sec = ll_read_opt("sec", this_char)))
437 if (!(*flags & LL_SBI_NOLCK) &&
438 ((*flags) = (*flags) |
439 ll_set_opt("nolock", this_char,
447 void ll_lli_init(struct ll_inode_info *lli)
449 sema_init(&lli->lli_open_sem, 1);
450 sema_init(&lli->lli_size_sem, 1);
452 lli->lli_size_pid = 0;
453 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
454 spin_lock_init(&lli->lli_lock);
455 INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
456 lli->lli_inode_magic = LLI_INODE_MAGIC;
457 memset(&lli->lli_id, 0, sizeof(lli->lli_id));
458 sema_init(&lli->lli_och_sem, 1);
459 lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
460 lli->lli_mds_exec_och = NULL;
461 lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
462 lli->lli_open_fd_exec_count = 0;
465 int ll_fill_super(struct super_block *sb, void *data, int silent)
467 struct ll_sb_info *sbi;
472 __u32 nllu[2] = { NOBODY_UID, NOBODY_GID };
473 __u64 remote_flag = 0;
476 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
478 sbi = lustre_init_sbi(sb);
482 sbi->ll_flags |= LL_SBI_READAHEAD;
483 ll_options(data, &lov, &lmv, &sec, &async, &sbi->ll_flags);
487 GOTO(out, err = -EINVAL);
492 GOTO(out, err = -EINVAL);
495 err = lustre_common_fill_super(sb, lmv, lov, async, sec, nllu,
503 OBD_FREE(sec, strlen(sec) + 1);
505 OBD_FREE(lmv, strlen(lmv) + 1);
507 OBD_FREE(lov, strlen(lov) + 1);
509 } /* ll_read_super */
511 static int lustre_process_log(struct lustre_mount_data *lmd, char *profile,
512 struct config_llog_instance *cfg, int allow_recov)
514 struct lustre_cfg lcfg;
515 struct portals_cfg pcfg;
516 char *peer = "MDS_PEER_UUID";
517 struct obd_device *obd;
518 struct lustre_handle md_conn = {0, };
519 struct obd_export *exp;
520 char *name = "mdc_dev";
522 struct obd_uuid lmv_uuid;
523 struct llog_ctxt *ctxt;
527 if (lmd_bad_magic(lmd))
530 generate_random_uuid(uuid);
531 class_uuid_unparse(uuid, &lmv_uuid);
533 if (lmd->lmd_local_nid) {
534 PCFG_INIT(pcfg, NAL_CMD_REGISTER_MYNID);
535 pcfg.pcfg_nal = lmd->lmd_nal;
536 pcfg.pcfg_nid = lmd->lmd_local_nid;
537 rc = libcfs_nal_cmd(&pcfg);
542 if (lmd->lmd_nal == SOCKNAL ||
543 lmd->lmd_nal == OPENIBNAL ||
544 lmd->lmd_nal == IIBNAL ||
545 lmd->lmd_nal == VIBNAL ||
546 lmd->lmd_nal == RANAL) {
547 PCFG_INIT(pcfg, NAL_CMD_ADD_PEER);
548 pcfg.pcfg_nal = lmd->lmd_nal;
549 pcfg.pcfg_nid = lmd->lmd_server_nid;
550 pcfg.pcfg_id = lmd->lmd_server_ipaddr;
551 pcfg.pcfg_misc = lmd->lmd_port;
552 rc = libcfs_nal_cmd(&pcfg);
557 LCFG_INIT(lcfg, LCFG_ADD_UUID, name);
558 lcfg.lcfg_nid = lmd->lmd_server_nid;
559 lcfg.lcfg_inllen1 = strlen(peer) + 1;
560 lcfg.lcfg_inlbuf1 = peer;
561 lcfg.lcfg_nal = lmd->lmd_nal;
562 rc = class_process_config(&lcfg);
564 GOTO(out_del_conn, rc);
566 LCFG_INIT(lcfg, LCFG_ATTACH, name);
567 lcfg.lcfg_inlbuf1 = "mdc";
568 lcfg.lcfg_inllen1 = strlen(lcfg.lcfg_inlbuf1) + 1;
569 lcfg.lcfg_inlbuf2 = lmv_uuid.uuid;
570 lcfg.lcfg_inllen2 = strlen(lcfg.lcfg_inlbuf2) + 1;
571 err = class_process_config(&lcfg);
573 GOTO(out_del_uuid, rc);
575 LCFG_INIT(lcfg, LCFG_SETUP, name);
576 lcfg.lcfg_inlbuf1 = lmd->lmd_mds;
577 lcfg.lcfg_inllen1 = strlen(lcfg.lcfg_inlbuf1) + 1;
578 lcfg.lcfg_inlbuf2 = peer;
579 lcfg.lcfg_inllen2 = strlen(lcfg.lcfg_inlbuf2) + 1;
580 rc = class_process_config(&lcfg);
582 GOTO(out_detach, rc);
584 obd = class_name2obd(name);
586 GOTO(out_cleanup, rc = -EINVAL);
588 rc = obd_set_info(obd->obd_self_export, strlen("sec"), "sec",
589 strlen(lmd->lmd_security), lmd->lmd_security);
591 GOTO(out_cleanup, rc);
593 /* Disable initial recovery on this import */
594 rc = obd_set_info(obd->obd_self_export,
595 strlen("initial_recov"), "initial_recov",
596 sizeof(allow_recov), &allow_recov);
598 GOTO(out_cleanup, rc);
600 rc = obd_connect(&md_conn, obd, &lmv_uuid, NULL, 0);
602 CERROR("cannot connect to %s: rc = %d\n", lmd->lmd_mds, rc);
603 GOTO(out_cleanup, rc);
606 exp = class_conn2export(&md_conn);
608 ctxt = llog_get_context(&exp->exp_obd->obd_llogs,LLOG_CONFIG_REPL_CTXT);
609 rc = class_config_process_llog(ctxt, profile, cfg);
611 CERROR("class_config_process_llog failed: rc = %d\n", rc);
613 err = obd_disconnect(exp, 0);
617 LCFG_INIT(lcfg, LCFG_CLEANUP, name);
618 err = class_process_config(&lcfg);
623 LCFG_INIT(lcfg, LCFG_DETACH, name);
624 err = class_process_config(&lcfg);
629 LCFG_INIT(lcfg, LCFG_DEL_UUID, name);
630 lcfg.lcfg_inllen1 = strlen(peer) + 1;
631 lcfg.lcfg_inlbuf1 = peer;
632 err = class_process_config(&lcfg);
635 if (lmd->lmd_nal == SOCKNAL ||
636 lmd->lmd_nal == OPENIBNAL ||
637 lmd->lmd_nal == IIBNAL ||
638 lmd->lmd_nal == VIBNAL ||
639 lmd->lmd_nal == RANAL) {
642 PCFG_INIT(pcfg, NAL_CMD_DEL_PEER);
643 pcfg.pcfg_nal = lmd->lmd_nal;
644 pcfg.pcfg_nid = lmd->lmd_server_nid;
645 pcfg.pcfg_flags = 1; /* single_share */
646 err2 = libcfs_nal_cmd(&pcfg);
659 static void lustre_manual_cleanup(struct ll_sb_info *sbi)
661 struct lustre_cfg lcfg;
662 struct obd_device *obd;
665 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) != NULL)
669 LCFG_INIT(lcfg, LCFG_CLEANUP, obd->obd_name);
670 err = class_process_config(&lcfg);
672 CERROR("cleanup failed: %s\n", obd->obd_name);
676 LCFG_INIT(lcfg, LCFG_DETACH, obd->obd_name);
677 err = class_process_config(&lcfg);
679 CERROR("detach failed: %s\n", obd->obd_name);
684 if (sbi->ll_lmd != NULL)
685 class_del_profile(sbi->ll_lmd->lmd_profile);
688 int lustre_fill_super(struct super_block *sb, void *data, int silent)
690 struct lustre_mount_data * lmd = data;
691 char *lov = NULL, *lmv = NULL;
692 struct ll_sb_info *sbi;
696 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
697 if (lmd_bad_magic(lmd))
700 sbi = lustre_init_sbi(sb);
704 sbi->ll_flags |= LL_SBI_READAHEAD;
706 if (lmd->lmd_profile) {
707 struct lustre_profile *lprof;
708 struct config_llog_instance cfg;
711 if (lmd->lmd_mds[0] == '\0') {
712 CERROR("no mds name\n");
713 GOTO(out_free, err = -EINVAL);
715 lmd->lmd_security[sizeof(lmd->lmd_security) - 1] = 0;
717 OBD_ALLOC(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
718 if (sbi->ll_lmd == NULL)
719 GOTO(out_free, err = -ENOMEM);
720 memcpy(sbi->ll_lmd, lmd, sizeof(*lmd));
722 /* generate a string unique to this super, let's try
723 the address of the super itself.*/
724 len = (sizeof(sb) * 2) + 1;
725 OBD_ALLOC(sbi->ll_instance, len);
726 if (sbi->ll_instance == NULL)
727 GOTO(out_free, err = -ENOMEM);
728 sprintf(sbi->ll_instance, "%p", sb);
730 cfg.cfg_instance = sbi->ll_instance;
731 cfg.cfg_uuid = sbi->ll_sb_uuid;
732 cfg.cfg_local_nid = lmd->lmd_local_nid;
733 err = lustre_process_log(lmd, lmd->lmd_profile, &cfg, 0);
735 CERROR("Unable to process log: %s\n", lmd->lmd_profile);
739 lprof = class_get_profile(lmd->lmd_profile);
741 CERROR("No profile found: %s\n", lmd->lmd_profile);
742 GOTO(out_free, err = -EINVAL);
745 OBD_FREE(lov, strlen(lov) + 1);
746 OBD_ALLOC(lov, strlen(lprof->lp_lov) +
747 strlen(sbi->ll_instance) + 2);
748 sprintf(lov, "%s-%s", lprof->lp_lov, sbi->ll_instance);
751 OBD_FREE(lmv, strlen(lmv) + 1);
752 OBD_ALLOC(lmv, strlen(lprof->lp_lmv) +
753 strlen(sbi->ll_instance) + 2);
754 sprintf(lmv, "%s-%s", lprof->lp_lmv, sbi->ll_instance);
759 GOTO(out_free, err = -EINVAL);
764 GOTO(out_free, err = -EINVAL);
767 err = lustre_common_fill_super(sb, lmv, lov, lmd->lmd_async,
768 lmd->lmd_security, &lmd->lmd_nllu,
769 &lmd->lmd_remote_flag);
776 OBD_FREE(lmv, strlen(lmv) + 1);
778 OBD_FREE(lov, strlen(lov) + 1);
784 int len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean")+1;
787 if (sbi->ll_instance != NULL) {
788 struct lustre_mount_data *lmd = sbi->ll_lmd;
789 struct config_llog_instance cfg;
792 cfg.cfg_instance = sbi->ll_instance;
793 cfg.cfg_uuid = sbi->ll_sb_uuid;
795 OBD_ALLOC(cl_prof, len);
796 sprintf(cl_prof, "%s-clean", lmd->lmd_profile);
797 err = lustre_process_log(lmd, cl_prof, &cfg, 0);
799 CERROR("Unable to process log: %s\n", cl_prof);
800 lustre_manual_cleanup(sbi);
802 OBD_FREE(cl_prof, len);
803 OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance) + 1);
805 OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
809 } /* lustre_fill_super */
811 void lustre_put_super(struct super_block *sb)
813 struct obd_device *obd;
814 struct ll_sb_info *sbi = ll_s2sbi(sb);
815 int force_umount = 0;
818 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
819 obd = class_exp2obd(sbi->ll_md_exp);
821 force_umount = obd->obd_no_recov;
824 lustre_common_put_super(sb);
825 if (sbi->ll_lmd != NULL) {
827 int len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean")+1;
829 struct config_llog_instance cfg;
832 CERROR("force umount, doing manual cleanup\n");
833 lustre_manual_cleanup(sbi);
837 cfg.cfg_instance = sbi->ll_instance;
838 cfg.cfg_uuid = sbi->ll_sb_uuid;
840 OBD_ALLOC(cl_prof, len);
841 sprintf(cl_prof, "%s-clean", sbi->ll_lmd->lmd_profile);
842 err = lustre_process_log(sbi->ll_lmd, cl_prof, &cfg, 0);
844 CERROR("Unable to process log: %s, doing manual cleanup"
846 lustre_manual_cleanup(sbi);
849 OBD_FREE(cl_prof, len);
851 OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
852 OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance) + 1);
858 } /* lustre_put_super */
860 int ll_process_config_update(struct ll_sb_info *sbi, int clean)
862 struct obd_export *md_exp = sbi->ll_md_exp;
863 struct lustre_mount_data *lmd = sbi->ll_lmd;
864 struct llog_ctxt *ctxt;
865 struct config_llog_instance cfg;
866 char *profile = lmd->lmd_profile, *name = NULL;
867 int rc, namelen = 0, version;
873 CERROR("Client not mounted with zero-conf; cannot process "
878 rc = ldlm_cli_cancel_unused(md_exp->exp_obd->obd_namespace, NULL,
879 LDLM_FL_CONFIG_CHANGE, NULL);
881 CWARN("ldlm_cli_cancel_unused(mdc): %d\n", rc);
883 rc = obd_cancel_unused(sbi->ll_dt_exp, NULL, LDLM_FL_CONFIG_CHANGE,
886 CWARN("obd_cancel_unused(lov): %d\n", rc);
888 cfg.cfg_instance = sbi->ll_instance;
889 cfg.cfg_uuid = sbi->ll_sb_uuid;
890 cfg.cfg_local_nid = lmd->lmd_local_nid;
892 namelen = strlen(profile) + 20; /* -clean-######### */
893 OBD_ALLOC(name, namelen);
898 version = sbi->ll_config_version - 1;
899 sprintf(name, "%s-clean-%d", profile, version);
901 version = sbi->ll_config_version + 1;
902 sprintf(name, "%s-%d", profile, version);
905 CWARN("Applying configuration log %s\n", name);
907 ctxt = llog_get_context(&md_exp->exp_obd->obd_llogs,
908 LLOG_CONFIG_REPL_CTXT);
909 rc = class_config_process_llog(ctxt, name, &cfg);
911 sbi->ll_config_version = version;
912 CWARN("Finished applying configuration log %s: %d\n", name, rc);
914 if (rc == 0 && clean == 0) {
915 struct lov_desc desc;
917 valsize = sizeof(desc);
918 rc = obd_get_info(sbi->ll_dt_exp, strlen("lovdesc") + 1,
919 "lovdesc", &valsize, &desc);
921 rc = obd_init_ea_size(md_exp,
922 obd_size_diskmd(sbi->ll_dt_exp, NULL),
924 sizeof(struct llog_cookie)));
926 OBD_FREE(name, namelen);
930 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
932 struct inode *inode = NULL;
933 l_lock(&lock->l_resource->lr_namespace->ns_lock);
934 if (lock->l_ast_data) {
935 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
936 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
937 inode = igrab(lock->l_ast_data);
939 inode = lock->l_ast_data;
940 CDEBUG(inode->i_state & I_FREEING ? D_INFO : D_WARNING,
941 "l_ast_data %p is bogus: magic %0x8\n",
942 lock->l_ast_data, lli->lli_inode_magic);
946 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
950 int null_if_equal(struct ldlm_lock *lock, void *data)
952 if (data == lock->l_ast_data) {
953 lock->l_ast_data = NULL;
955 if (lock->l_req_mode != lock->l_granted_mode)
956 LDLM_ERROR(lock,"clearing inode with ungranted lock\n");
959 return LDLM_ITER_CONTINUE;
962 void ll_clear_inode(struct inode *inode)
965 struct ll_inode_info *lli = ll_i2info(inode);
966 struct ll_sb_info *sbi = ll_i2sbi(inode);
969 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
970 inode->i_generation, inode);
972 lli->lli_inode_magic = LLI_INODE_DEAD;
973 ll_inode2id(&id, inode);
975 clear_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &(ll_i2info(inode)->lli_flags));
976 md_change_cbdata(sbi->ll_md_exp, &id, null_if_equal, inode);
978 LASSERT(!lli->lli_open_fd_write_count);
979 LASSERT(!lli->lli_open_fd_read_count);
980 LASSERT(!lli->lli_open_fd_exec_count);
982 if (lli->lli_mds_write_och)
983 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_WRITE);
984 if (lli->lli_mds_exec_och)
985 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_EXEC);
986 if (lli->lli_mds_read_och)
987 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_READ);
990 obd_change_cbdata(sbi->ll_dt_exp, lli->lli_smd,
991 null_if_equal, inode);
994 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
999 obd_free_memmd(sbi->ll_md_exp,
1000 (struct lov_stripe_md **) &lli->lli_mea);
1001 lli->lli_mea = NULL;
1004 if (lli->lli_symlink_name) {
1005 OBD_FREE(lli->lli_symlink_name,
1006 strlen(lli->lli_symlink_name) + 1);
1007 lli->lli_symlink_name = NULL;
1009 lli->lli_inode_magic = LLI_INODE_DEAD;
1014 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1015 * object(s) determine the file size and mtime. Otherwise, the MDS will
1016 * keep these values until such a time that objects are allocated for it.
1017 * We do the MDS operations first, as it is checking permissions for us.
1018 * We don't to the MDS RPC if there is nothing that we want to store there,
1019 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1020 * going to do an RPC anyways.
1022 * If we are doing a truncate, we will send the mtime and ctime updates
1023 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1024 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1027 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1029 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1030 struct ll_sb_info *sbi = ll_i2sbi(inode);
1031 struct ptlrpc_request *request = NULL;
1032 struct mdc_op_data *op_data;
1033 int ia_valid = attr->ia_valid;
1037 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu\n", inode->i_ino);
1038 lprocfs_counter_incr(ll_i2sbi(inode)->ll_stats, LPROC_LL_SETATTR);
1040 if (ia_valid & ATTR_SIZE) {
1041 if (attr->ia_size > ll_file_maxbytes(inode)) {
1042 CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1043 attr->ia_size, ll_file_maxbytes(inode));
1047 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1050 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1051 if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1052 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1056 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1057 if (attr->ia_valid & ATTR_CTIME) {
1058 attr->ia_ctime = CURRENT_TIME;
1059 attr->ia_valid |= ATTR_CTIME_SET;
1061 if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1062 attr->ia_atime = CURRENT_TIME;
1063 attr->ia_valid |= ATTR_ATIME_SET;
1065 if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1066 attr->ia_mtime = CURRENT_TIME;
1067 attr->ia_valid |= ATTR_MTIME_SET;
1070 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1071 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1072 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1073 LTIME_S(CURRENT_TIME));
1076 attr->ia_valid &= ~ATTR_SIZE;
1078 /* If only OST attributes being set on objects, don't do MDS RPC.
1079 * In that case, we need to check permissions and update the local
1080 * inode ourselves so we can call obdo_from_inode() always. */
1081 if (ia_valid & (lsm ? ~(ATTR_SIZE | ATTR_FROM_OPEN /*| ATTR_RAW*/) : ~0)) {
1082 struct lustre_md md;
1084 OBD_ALLOC(op_data, sizeof(*op_data));
1085 if (op_data == NULL)
1087 ll_prepare_mdc_data(op_data, inode, NULL, NULL, 0, 0);
1089 rc = md_setattr(sbi->ll_md_exp, op_data,
1090 attr, NULL, 0, NULL, 0, &request);
1091 OBD_FREE(op_data, sizeof(*op_data));
1093 ptlrpc_req_finished(request);
1094 if (rc != -EPERM && rc != -EACCES)
1095 CERROR("md_setattr fails: rc = %d\n", rc);
1099 rc = mdc_req2lustre_md(sbi->ll_md_exp, request, 0,
1100 sbi->ll_dt_exp, &md);
1102 ptlrpc_req_finished(request);
1106 /* We call inode_setattr to adjust timestamps, but we first
1107 * clear ATTR_SIZE to avoid invoking vmtruncate.
1109 * NB: ATTR_SIZE will only be set at this point if the size
1110 * resides on the MDS, ie, this file has no objects. */
1111 attr->ia_valid &= ~ATTR_SIZE;
1114 * assigning inode_setattr() to @err to disable warning that
1115 * function's result should be checked by by caller. error is
1116 * impossible here, as vmtruncate() control path is disabled.
1118 err = inode_setattr(inode, attr);
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 * assigning inode_setattr() to @err to disable warning that
1149 * function's result should be checked by by caller. error is
1150 * impossible here, as vmtruncate() control path is disabled.
1154 /* We really need to get our PW lock before we change inode->i_size.
1155 * If we don't we can race with other i_size updaters on our node, like
1156 * ll_file_read. We can also race with i_size propogation to other
1157 * nodes through dirtying and writeback of final cached pages. This
1158 * last one is especially bad for racing o_append users on other
1160 if (ia_valid & ATTR_SIZE) {
1161 ldlm_policy_data_t policy = { .l_extent = {attr->ia_size,
1163 struct lustre_handle lockh = { 0 };
1164 struct ll_inode_info *lli = ll_i2info(inode);
1165 int err, ast_flags = 0;
1166 /* XXX when we fix the AST intents to pass the discard-range
1167 * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
1169 if (attr->ia_size == 0)
1170 ast_flags = LDLM_AST_DISCARD_DATA;
1172 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy, &lockh,
1173 ast_flags, &ll_i2sbi(inode)->ll_seek_stime);
1178 down(&lli->lli_size_sem);
1179 lli->lli_size_pid = current->pid;
1180 rc = vmtruncate(inode, attr->ia_size);
1182 LASSERT(atomic_read(&lli->lli_size_sem.count) <= 0);
1183 lli->lli_size_pid = 0;
1184 up(&lli->lli_size_sem);
1187 err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
1189 CERROR("ll_extent_unlock failed: %d\n", err);
1193 } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
1194 struct obdo *oa = NULL;
1196 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1197 inode->i_ino, LTIME_S(attr->ia_mtime));
1203 oa->o_id = lsm->lsm_object_id;
1204 oa->o_gr = lsm->lsm_object_gr;
1205 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1206 obdo_from_inode(oa, inode, OBD_MD_FLTYPE | OBD_MD_FLATIME |
1207 OBD_MD_FLMTIME | OBD_MD_FLCTIME);
1208 rc = obd_setattr(sbi->ll_dt_exp, oa, lsm, NULL);
1211 CERROR("obd_setattr fails: rc = %d\n", rc);
1216 int ll_setattr(struct dentry *de, struct iattr *attr)
1218 LASSERT(de->d_inode);
1219 return ll_setattr_raw(de->d_inode, attr);
1222 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1223 unsigned long max_age)
1225 struct ll_sb_info *sbi = ll_s2sbi(sb);
1226 struct obd_statfs obd_osfs;
1230 rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age);
1232 CERROR("obd_statfs fails: rc = %d\n", rc);
1236 osfs->os_type = sb->s_magic;
1238 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1239 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1241 rc = obd_statfs(class_exp2obd(sbi->ll_dt_exp), &obd_osfs, max_age);
1243 CERROR("obd_statfs fails: rc = %d\n", rc);
1247 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1248 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1251 osfs->os_blocks = obd_osfs.os_blocks;
1252 osfs->os_bfree = obd_osfs.os_bfree;
1253 osfs->os_bavail = obd_osfs.os_bavail;
1255 /* If we don't have as many objects free on the OST as inodes
1256 * on the MDS, we reduce the total number of inodes to
1257 * compensate, so that the "inodes in use" number is correct.
1259 if (obd_osfs.os_ffree < osfs->os_ffree) {
1260 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1262 osfs->os_ffree = obd_osfs.os_ffree;
1268 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1270 struct obd_statfs osfs;
1273 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p\n", sb);
1274 lprocfs_counter_incr(ll_s2sbi(sb)->ll_stats, LPROC_LL_STAFS);
1276 /* For now we will always get up-to-date statfs values, but in the
1277 * future we may allow some amount of caching on the client (e.g.
1278 * from QOS or lprocfs updates). */
1279 rc = ll_statfs_internal(sb, &osfs, jiffies - 1);
1283 statfs_unpack(sfs, &osfs);
1285 if (sizeof(sfs->f_blocks) == 4) {
1286 while (osfs.os_blocks > ~0UL) {
1289 osfs.os_blocks >>= 1;
1290 osfs.os_bfree >>= 1;
1291 osfs.os_bavail >>= 1;
1295 sfs->f_blocks = osfs.os_blocks;
1296 sfs->f_bfree = osfs.os_bfree;
1297 sfs->f_bavail = osfs.os_bavail;
1302 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1304 struct ll_inode_info *lli = ll_i2info(inode);
1305 struct lov_stripe_md *lsm = md->lsm;
1306 struct mds_body *body = md->body;
1307 struct mea *mea = md->mea;
1308 struct posix_acl *ll_acl_access = md->acl_access;
1311 LASSERT((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1313 if (md->lsm && md->lsm->lsm_magic != LOV_MAGIC) {
1314 /* check for default striping info for dir. */
1315 LASSERT((mea != NULL) == ((body->valid & OBD_MD_FLDIREA) != 0));
1319 LASSERT(lsm->lsm_object_gr > 0);
1320 if (lli->lli_smd == NULL) {
1322 lli->lli_maxbytes = lsm->lsm_maxbytes;
1323 if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1324 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1327 if (memcmp(lli->lli_smd, lsm, sizeof(*lsm))) {
1328 CERROR("lsm mismatch for inode %ld\n",
1330 CERROR("lli_smd:\n");
1331 dump_lsm(D_ERROR, lli->lli_smd);
1333 dump_lsm(D_ERROR, lsm);
1336 /* XXX FIXME -- We should decide on a safer (atomic) and
1337 * more elegant way to update the lsm */
1338 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1339 lli->lli_smd->lsm_oinfo[i].loi_id =
1340 lsm->lsm_oinfo[i].loi_id;
1341 lli->lli_smd->lsm_oinfo[i].loi_gr =
1342 lsm->lsm_oinfo[i].loi_gr;
1343 lli->lli_smd->lsm_oinfo[i].loi_ost_idx =
1344 lsm->lsm_oinfo[i].loi_ost_idx;
1345 lli->lli_smd->lsm_oinfo[i].loi_ost_gen =
1346 lsm->lsm_oinfo[i].loi_ost_gen;
1349 /* bug 2844 - limit i_blksize for broken user-space apps */
1350 LASSERTF(lsm->lsm_xfersize != 0, "%lu\n", lsm->lsm_xfersize);
1351 inode->i_blksize = min(lsm->lsm_xfersize, LL_MAX_BLKSIZE);
1352 if (lli->lli_smd != lsm)
1353 obd_free_memmd(ll_i2dtexp(inode), &lsm);
1357 if (lli->lli_mea == NULL) {
1360 if (memcmp(lli->lli_mea, mea, body->eadatasize)) {
1361 CERROR("mea mismatch for inode %lu\n",
1366 if (lli->lli_mea != mea)
1367 obd_free_memmd(ll_i2mdexp(inode),
1368 (struct lov_stripe_md **) &mea);
1371 if (body->valid & OBD_MD_FID)
1372 id_assign_fid(&lli->lli_id, &body->id1);
1374 if (body->valid & OBD_MD_FLID)
1375 id_ino(&lli->lli_id) = id_ino(&body->id1);
1377 if (body->valid & OBD_MD_FLGENER)
1378 id_gen(&lli->lli_id) = id_gen(&body->id1);
1380 spin_lock(&lli->lli_lock);
1381 if (ll_acl_access != NULL) {
1382 if (lli->lli_acl_access != NULL)
1383 posix_acl_release(lli->lli_acl_access);
1384 lli->lli_acl_access = ll_acl_access;
1386 spin_unlock(&lli->lli_lock);
1388 if (body->valid & OBD_MD_FLID)
1389 inode->i_ino = id_ino(&body->id1);
1390 if (body->valid & OBD_MD_FLGENER)
1391 inode->i_generation = id_gen(&body->id1);
1392 if (body->valid & OBD_MD_FLATIME)
1393 LTIME_S(inode->i_atime) = body->atime;
1394 if (body->valid & OBD_MD_FLMTIME &&
1395 body->mtime > LTIME_S(inode->i_mtime)) {
1396 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to %u\n",
1397 inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
1398 LTIME_S(inode->i_mtime) = body->mtime;
1400 if (body->valid & OBD_MD_FLCTIME &&
1401 body->ctime > LTIME_S(inode->i_ctime))
1402 LTIME_S(inode->i_ctime) = body->ctime;
1403 if (body->valid & OBD_MD_FLMODE) {
1404 inode->i_mode = (inode->i_mode & S_IFMT) |
1405 (body->mode & ~S_IFMT);
1407 if (body->valid & OBD_MD_FLTYPE) {
1408 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1409 (body->mode & S_IFMT);
1411 if (body->valid & OBD_MD_FLUID)
1412 inode->i_uid = body->uid;
1413 if (body->valid & OBD_MD_FLGID)
1414 inode->i_gid = body->gid;
1415 if (body->valid & OBD_MD_FLFLAGS)
1416 inode->i_flags = body->flags;
1417 if (body->valid & OBD_MD_FLNLINK)
1418 inode->i_nlink = body->nlink;
1419 if (body->valid & OBD_MD_FLRDEV)
1420 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1421 inode->i_rdev = body->rdev;
1423 inode->i_rdev = old_decode_dev(body->rdev);
1425 if (body->valid & OBD_MD_FLSIZE)
1426 inode->i_size = body->size;
1427 if (body->valid & OBD_MD_FLBLOCKS)
1428 inode->i_blocks = body->blocks;
1430 if (body->valid & OBD_MD_FLSIZE)
1431 set_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &lli->lli_flags);
1433 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1434 inode->i_dev = (kdev_t)id_group(&lli->lli_id);
1436 LASSERT(id_fid(&lli->lli_id) != 0);
1439 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
1440 static struct backing_dev_info ll_backing_dev_info = {
1441 .ra_pages = 0, /* No readahead */
1442 .memory_backed = 0, /* Does contribute to dirty memory */
1446 void ll_read_inode2(struct inode *inode, void *opaque)
1448 struct lustre_md *md = opaque;
1449 struct ll_inode_info *lli = ll_i2info(inode);
1452 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1453 inode->i_generation, inode);
1457 LASSERT(!lli->lli_smd);
1459 /* Core attributes from the MDS first. This is a new inode, and
1460 * the VFS doesn't zero times in the core inode so we have to do
1461 * it ourselves. They will be overwritten by either MDS or OST
1462 * attributes - we just need to make sure they aren't newer. */
1463 LTIME_S(inode->i_mtime) = 0;
1464 LTIME_S(inode->i_atime) = 0;
1465 LTIME_S(inode->i_ctime) = 0;
1468 ll_update_inode(inode, md);
1470 /* OIDEBUG(inode); */
1472 if (S_ISREG(inode->i_mode)) {
1473 inode->i_op = &ll_file_inode_operations;
1474 inode->i_fop = &ll_file_operations;
1475 inode->i_mapping->a_ops = &ll_aops;
1477 } else if (S_ISDIR(inode->i_mode)) {
1478 inode->i_op = &ll_dir_inode_operations;
1479 inode->i_fop = &ll_dir_operations;
1480 inode->i_mapping->a_ops = &ll_dir_aops;
1482 } else if (S_ISLNK(inode->i_mode)) {
1483 inode->i_op = &ll_fast_symlink_inode_operations;
1486 inode->i_op = &ll_special_inode_operations;
1488 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
1489 init_special_inode(inode, inode->i_mode,
1490 kdev_t_to_nr(inode->i_rdev));
1492 /* initializing backing dev info. */
1493 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
1495 init_special_inode(inode, inode->i_mode, inode->i_rdev);
1497 lli->ll_save_ifop = inode->i_fop;
1499 if (S_ISCHR(inode->i_mode))
1500 inode->i_fop = &ll_special_chr_inode_fops;
1501 else if (S_ISBLK(inode->i_mode))
1502 inode->i_fop = &ll_special_blk_inode_fops;
1503 else if (S_ISFIFO(inode->i_mode))
1504 inode->i_fop = &ll_special_fifo_inode_fops;
1505 else if (S_ISSOCK(inode->i_mode))
1506 inode->i_fop = &ll_special_sock_inode_fops;
1508 CWARN("saved %p, replaced with %p\n", lli->ll_save_ifop,
1511 if (lli->ll_save_ifop->owner) {
1512 CWARN("%p has owner %p\n", lli->ll_save_ifop,
1513 lli->ll_save_ifop->owner);
1519 void ll_delete_inode(struct inode *inode)
1521 struct ll_sb_info *sbi = ll_i2sbi(inode);
1522 struct lustre_id id;
1526 ll_inode2id(&id, inode);
1528 rc = md_delete_inode(sbi->ll_md_exp, &id);
1530 CERROR("md_delete_inode() failed, error %d\n",
1538 int ll_iocontrol(struct inode *inode, struct file *file,
1539 unsigned int cmd, unsigned long arg)
1541 struct ll_sb_info *sbi = ll_i2sbi(inode);
1542 struct ptlrpc_request *req = NULL;
1547 case EXT3_IOC_GETFLAGS: {
1548 struct lustre_id id;
1549 __u64 valid = OBD_MD_FLFLAGS;
1550 struct mds_body *body;
1552 ll_inode2id(&id, inode);
1553 rc = md_getattr(sbi->ll_md_exp, &id, valid, NULL, 0, 0, &req);
1555 CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1559 body = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*body));
1561 if (body->flags & S_APPEND)
1562 flags |= EXT3_APPEND_FL;
1563 if (body->flags & S_IMMUTABLE)
1564 flags |= EXT3_IMMUTABLE_FL;
1565 if (body->flags & S_NOATIME)
1566 flags |= EXT3_NOATIME_FL;
1568 ptlrpc_req_finished (req);
1570 RETURN(put_user(flags, (int *)arg));
1572 case EXT3_IOC_SETFLAGS: {
1573 struct mdc_op_data *op_data;
1576 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1578 if (get_user(flags, (int *)arg))
1585 OBD_ALLOC(op_data, sizeof(*op_data));
1586 if (op_data == NULL) {
1590 ll_prepare_mdc_data(op_data, inode, NULL, NULL, 0, 0);
1592 memset(&attr, 0x0, sizeof(attr));
1593 attr.ia_attr_flags = flags;
1594 attr.ia_valid |= ATTR_ATTR_FLAG;
1596 rc = md_setattr(sbi->ll_md_exp, op_data,
1597 &attr, NULL, 0, NULL, 0, &req);
1598 OBD_FREE(op_data, sizeof(*op_data));
1600 ptlrpc_req_finished(req);
1601 if (rc != -EPERM && rc != -EACCES)
1602 CERROR("md_setattr fails: rc = %d\n", rc);
1606 ptlrpc_req_finished(req);
1608 oa->o_id = lsm->lsm_object_id;
1609 oa->o_gr = lsm->lsm_object_gr;
1610 oa->o_flags = flags;
1611 oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS | OBD_MD_FLGROUP;
1613 rc = obd_setattr(sbi->ll_dt_exp, oa, lsm, NULL);
1616 if (rc != -EPERM && rc != -EACCES)
1617 CERROR("md_setattr fails: rc = %d\n", rc);
1621 if (flags & EXT3_APPEND_FL)
1622 inode->i_flags |= S_APPEND;
1624 inode->i_flags &= ~S_APPEND;
1625 if (flags & EXT3_IMMUTABLE_FL)
1626 inode->i_flags |= S_IMMUTABLE;
1628 inode->i_flags &= ~S_IMMUTABLE;
1629 if (flags & EXT3_NOATIME_FL)
1630 inode->i_flags |= S_NOATIME;
1632 inode->i_flags &= ~S_NOATIME;
1643 /* this is only called in the case of forced umount. */
1644 void ll_umount_begin(struct super_block *sb)
1646 struct ll_sb_info *sbi = ll_s2sbi(sb);
1647 struct obd_ioctl_data ioc_data = { 0 };
1648 struct obd_device *obd;
1651 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1652 sb->s_count, atomic_read(&sb->s_active));
1654 obd = class_exp2obd(sbi->ll_md_exp);
1656 CERROR("Invalid MDC connection handle "LPX64"\n",
1657 sbi->ll_md_exp->exp_handle.h_cookie);
1661 obd->obd_no_recov = 1;
1662 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
1663 sizeof(ioc_data), &ioc_data, NULL);
1665 obd = class_exp2obd(sbi->ll_dt_exp);
1667 CERROR("Invalid LOV connection handle "LPX64"\n",
1668 sbi->ll_dt_exp->exp_handle.h_cookie);
1673 obd->obd_no_recov = 1;
1674 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
1675 sizeof(ioc_data), &ioc_data, NULL);
1678 * really, we'd like to wait until there are no requests outstanding,
1679 * and then continue. For now, we just invalidate the requests,
1680 * schedule, and hope.
1687 int ll_prep_inode(struct obd_export *dt_exp, struct obd_export *md_exp,
1688 struct inode **inode, struct ptlrpc_request *req,
1689 int offset, struct super_block *sb)
1691 struct lustre_md md;
1694 rc = mdc_req2lustre_md(md_exp, req, offset, dt_exp, &md);
1699 ll_update_inode(*inode, &md);
1702 *inode = ll_iget(sb, id_ino(&md.body->id1), &md);
1703 if (*inode == NULL || is_bad_inode(*inode)) {
1704 /* free the lsm if we allocated one above */
1706 obd_free_memmd(dt_exp, &md.lsm);
1708 obd_free_memmd(md_exp,
1709 (struct lov_stripe_md**)&md.mea);
1711 CERROR("new_inode -fatal: rc %d\n", rc);
1718 int ll_get_fid(struct obd_export *exp, struct lustre_id *idp,
1719 char *filename, struct lustre_id *ret)
1721 struct ptlrpc_request *request = NULL;
1722 struct mds_body *body;
1725 rc = md_getattr_lock(exp, idp, filename, strlen(filename) + 1,
1726 OBD_MD_FID, 0, &request);
1728 CDEBUG(D_INFO, "md_getattr_lock failed on %s: rc %d\n",
1733 body = lustre_msg_buf(request->rq_repmsg, 0, sizeof(*body));
1734 LASSERT(body != NULL);
1735 LASSERT_REPSWABBED(request, 0);
1738 ptlrpc_req_finished(request);