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)) {
293 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
295 obd_free_memmd(sbi->ll_md_exp,
296 (struct lov_stripe_md**)&md.mea);
297 CERROR("lustre_lite: bad iget4 for root\n");
298 GOTO(out_root, err = -EBADF);
301 err = ll_close_thread_start(&sbi->ll_lcq);
303 CERROR("cannot start close thread: rc %d\n", err);
307 ll_gns_add_timer(sbi);
309 /* making vm readahead 0 for 2.4.x. In the case of 2.6.x,
310 backing dev info assigned to inode mapping is used for
311 determining maximal readahead. */
312 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0)) && \
313 !defined(KERNEL_HAS_AS_MAX_READAHEAD)
314 /* bug 2805 - set VM readahead to zero */
315 vm_max_readahead = vm_min_readahead = 0;
318 sb->s_root = d_alloc_root(root);
319 sb->s_root->d_op = &ll_d_ops;
321 sb->s_flags |= MS_POSIXACL;
323 CWARN("Enabling PDIROPS\n");
324 sb->s_flags |= S_PDIROPS;
328 OBD_FREE(data, sizeof(*data));
334 obd_disconnect(sbi->ll_dt_exp, 0);
336 obd_disconnect(sbi->ll_md_exp, 0);
339 OBD_FREE(data, sizeof(*data));
340 lprocfs_unregister_mountpoint(sbi);
344 void lustre_common_put_super(struct super_block *sb)
346 struct ll_sb_info *sbi = ll_s2sbi(sb);
347 struct hlist_node *tmp, *next;
350 ll_gns_del_timer(sbi);
351 ll_close_thread_shutdown(sbi->ll_lcq);
353 list_del(&sbi->ll_conn_chain);
354 obd_disconnect(sbi->ll_dt_exp, 0);
356 lprocfs_unregister_mountpoint(sbi);
357 if (sbi->ll_proc_root) {
358 lprocfs_remove(sbi->ll_proc_root);
359 sbi->ll_proc_root = NULL;
362 obd_disconnect(sbi->ll_md_exp, 0);
364 // We do this to get rid of orphaned dentries. That is not really trw.
365 hlist_for_each_safe(tmp, next, &sbi->ll_orphan_dentry_list) {
366 struct dentry *dentry = hlist_entry(tmp, struct dentry, d_hash);
367 CWARN("orphan dentry %.*s (%p->%p) at unmount\n",
368 dentry->d_name.len, dentry->d_name.name, dentry, next);
369 shrink_dcache_parent(dentry);
374 char *ll_read_opt(const char *opt, char *data)
380 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
381 if (strncmp(opt, data, strlen(opt)))
383 if ((value = strchr(data, '=')) == NULL)
387 OBD_ALLOC(retval, strlen(value) + 1);
389 CERROR("out of memory!\n");
393 memcpy(retval, value, strlen(value)+1);
394 CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
398 int ll_set_opt(const char *opt, char *data, int fl)
402 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
403 if (strncmp(opt, data, strlen(opt)))
409 void ll_options(char *options, char **lov, char **lmv, char **sec,
410 int *async, int *flags)
413 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
414 char *opt_ptr = options;
424 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
425 for (this_char = strtok (options, ",");
427 this_char = strtok (NULL, ",")) {
429 while ((this_char = strsep (&opt_ptr, ",")) != NULL) {
431 CDEBUG(D_SUPER, "this_char %s\n", this_char);
432 if (!*lov && (*lov = ll_read_opt("osc", this_char)))
434 if (!*lmv && (*lmv = ll_read_opt("mdc", this_char)))
436 if (!strncmp(this_char, "lasync", strlen("lasync"))) {
440 if (!*sec && (*sec = ll_read_opt("sec", this_char)))
442 if (!(*flags & LL_SBI_NOLCK) &&
443 ((*flags) = (*flags) |
444 ll_set_opt("nolock", this_char,
452 void ll_lli_init(struct ll_inode_info *lli)
454 sema_init(&lli->lli_open_sem, 1);
455 sema_init(&lli->lli_size_sem, 1);
457 lli->lli_size_pid = 0;
458 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
459 spin_lock_init(&lli->lli_lock);
460 INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
461 lli->lli_inode_magic = LLI_INODE_MAGIC;
462 memset(&lli->lli_id, 0, sizeof(lli->lli_id));
463 sema_init(&lli->lli_och_sem, 1);
464 lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
465 lli->lli_mds_exec_och = NULL;
466 lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
467 lli->lli_open_fd_exec_count = 0;
470 int ll_fill_super(struct super_block *sb, void *data, int silent)
472 struct ll_sb_info *sbi;
477 __u32 nllu[2] = { NOBODY_UID, NOBODY_GID };
478 __u64 remote_flag = 0;
481 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
483 sbi = lustre_init_sbi(sb);
487 sbi->ll_flags |= LL_SBI_READAHEAD;
488 ll_options(data, &lov, &lmv, &sec, &async, &sbi->ll_flags);
492 GOTO(out, err = -EINVAL);
497 GOTO(out, err = -EINVAL);
500 err = lustre_common_fill_super(sb, lmv, lov, async, sec, nllu,
508 OBD_FREE(sec, strlen(sec) + 1);
510 OBD_FREE(lmv, strlen(lmv) + 1);
512 OBD_FREE(lov, strlen(lov) + 1);
514 } /* ll_read_super */
516 static int lustre_process_log(struct lustre_mount_data *lmd, char *profile,
517 struct config_llog_instance *cfg, int allow_recov)
519 struct lustre_cfg lcfg;
520 struct portals_cfg pcfg;
521 char *peer = "MDS_PEER_UUID";
522 struct obd_device *obd;
523 struct lustre_handle md_conn = {0, };
524 struct obd_export *exp;
525 char *name = "mdc_dev";
527 struct obd_uuid lmv_uuid;
528 struct llog_ctxt *ctxt;
532 if (lmd_bad_magic(lmd))
535 generate_random_uuid(uuid);
536 class_uuid_unparse(uuid, &lmv_uuid);
538 if (lmd->lmd_local_nid) {
539 PCFG_INIT(pcfg, NAL_CMD_REGISTER_MYNID);
540 pcfg.pcfg_nal = lmd->lmd_nal;
541 pcfg.pcfg_nid = lmd->lmd_local_nid;
542 rc = libcfs_nal_cmd(&pcfg);
547 if (lmd->lmd_nal == SOCKNAL ||
548 lmd->lmd_nal == OPENIBNAL ||
549 lmd->lmd_nal == IIBNAL ||
550 lmd->lmd_nal == VIBNAL ||
551 lmd->lmd_nal == RANAL) {
552 PCFG_INIT(pcfg, NAL_CMD_ADD_PEER);
553 pcfg.pcfg_nal = lmd->lmd_nal;
554 pcfg.pcfg_nid = lmd->lmd_server_nid;
555 pcfg.pcfg_id = lmd->lmd_server_ipaddr;
556 pcfg.pcfg_misc = lmd->lmd_port;
557 rc = libcfs_nal_cmd(&pcfg);
562 LCFG_INIT(lcfg, LCFG_ADD_UUID, name);
563 lcfg.lcfg_nid = lmd->lmd_server_nid;
564 lcfg.lcfg_inllen1 = strlen(peer) + 1;
565 lcfg.lcfg_inlbuf1 = peer;
566 lcfg.lcfg_nal = lmd->lmd_nal;
567 rc = class_process_config(&lcfg);
569 GOTO(out_del_conn, rc);
571 LCFG_INIT(lcfg, LCFG_ATTACH, name);
572 lcfg.lcfg_inlbuf1 = "mdc";
573 lcfg.lcfg_inllen1 = strlen(lcfg.lcfg_inlbuf1) + 1;
574 lcfg.lcfg_inlbuf2 = lmv_uuid.uuid;
575 lcfg.lcfg_inllen2 = strlen(lcfg.lcfg_inlbuf2) + 1;
576 err = class_process_config(&lcfg);
578 GOTO(out_del_uuid, rc);
580 LCFG_INIT(lcfg, LCFG_SETUP, name);
581 lcfg.lcfg_inlbuf1 = lmd->lmd_mds;
582 lcfg.lcfg_inllen1 = strlen(lcfg.lcfg_inlbuf1) + 1;
583 lcfg.lcfg_inlbuf2 = peer;
584 lcfg.lcfg_inllen2 = strlen(lcfg.lcfg_inlbuf2) + 1;
585 rc = class_process_config(&lcfg);
587 GOTO(out_detach, rc);
589 obd = class_name2obd(name);
591 GOTO(out_cleanup, rc = -EINVAL);
593 rc = obd_set_info(obd->obd_self_export, strlen("sec"), "sec",
594 strlen(lmd->lmd_security), lmd->lmd_security);
596 GOTO(out_cleanup, rc);
598 /* Disable initial recovery on this import */
599 rc = obd_set_info(obd->obd_self_export,
600 strlen("initial_recov"), "initial_recov",
601 sizeof(allow_recov), &allow_recov);
603 GOTO(out_cleanup, rc);
605 rc = obd_connect(&md_conn, obd, &lmv_uuid, NULL, 0);
607 CERROR("cannot connect to %s: rc = %d\n", lmd->lmd_mds, rc);
608 GOTO(out_cleanup, rc);
611 exp = class_conn2export(&md_conn);
613 ctxt = llog_get_context(&exp->exp_obd->obd_llogs,LLOG_CONFIG_REPL_CTXT);
614 rc = class_config_process_llog(ctxt, profile, cfg);
616 CERROR("class_config_process_llog failed: rc = %d\n", rc);
618 err = obd_disconnect(exp, 0);
622 LCFG_INIT(lcfg, LCFG_CLEANUP, name);
623 err = class_process_config(&lcfg);
628 LCFG_INIT(lcfg, LCFG_DETACH, name);
629 err = class_process_config(&lcfg);
634 LCFG_INIT(lcfg, LCFG_DEL_UUID, name);
635 lcfg.lcfg_inllen1 = strlen(peer) + 1;
636 lcfg.lcfg_inlbuf1 = peer;
637 err = class_process_config(&lcfg);
640 if (lmd->lmd_nal == SOCKNAL ||
641 lmd->lmd_nal == OPENIBNAL ||
642 lmd->lmd_nal == IIBNAL ||
643 lmd->lmd_nal == VIBNAL ||
644 lmd->lmd_nal == RANAL) {
647 PCFG_INIT(pcfg, NAL_CMD_DEL_PEER);
648 pcfg.pcfg_nal = lmd->lmd_nal;
649 pcfg.pcfg_nid = lmd->lmd_server_nid;
650 pcfg.pcfg_flags = 1; /* single_share */
651 err2 = libcfs_nal_cmd(&pcfg);
664 static void lustre_manual_cleanup(struct ll_sb_info *sbi)
666 struct lustre_cfg lcfg;
667 struct obd_device *obd;
670 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) != NULL)
674 LCFG_INIT(lcfg, LCFG_CLEANUP, obd->obd_name);
675 err = class_process_config(&lcfg);
677 CERROR("cleanup failed: %s\n", obd->obd_name);
681 LCFG_INIT(lcfg, LCFG_DETACH, obd->obd_name);
682 err = class_process_config(&lcfg);
684 CERROR("detach failed: %s\n", obd->obd_name);
689 if (sbi->ll_lmd != NULL)
690 class_del_profile(sbi->ll_lmd->lmd_profile);
693 int lustre_fill_super(struct super_block *sb, void *data, int silent)
695 struct lustre_mount_data * lmd = data;
696 char *lov = NULL, *lmv = NULL;
697 struct ll_sb_info *sbi;
701 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
702 if (lmd_bad_magic(lmd))
705 sbi = lustre_init_sbi(sb);
709 sbi->ll_flags |= LL_SBI_READAHEAD;
711 if (lmd->lmd_profile) {
712 struct lustre_profile *lprof;
713 struct config_llog_instance cfg;
716 if (lmd->lmd_mds[0] == '\0') {
717 CERROR("no mds name\n");
718 GOTO(out_free, err = -EINVAL);
720 lmd->lmd_security[sizeof(lmd->lmd_security) - 1] = 0;
722 OBD_ALLOC(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
723 if (sbi->ll_lmd == NULL)
724 GOTO(out_free, err = -ENOMEM);
725 memcpy(sbi->ll_lmd, lmd, sizeof(*lmd));
727 /* generate a string unique to this super, let's try
728 the address of the super itself.*/
729 len = (sizeof(sb) * 2) + 1;
730 OBD_ALLOC(sbi->ll_instance, len);
731 if (sbi->ll_instance == NULL)
732 GOTO(out_free, err = -ENOMEM);
733 sprintf(sbi->ll_instance, "%p", sb);
735 cfg.cfg_instance = sbi->ll_instance;
736 cfg.cfg_uuid = sbi->ll_sb_uuid;
737 cfg.cfg_local_nid = lmd->lmd_local_nid;
738 err = lustre_process_log(lmd, lmd->lmd_profile, &cfg, 0);
740 CERROR("Unable to process log: %s\n", lmd->lmd_profile);
744 lprof = class_get_profile(lmd->lmd_profile);
746 CERROR("No profile found: %s\n", lmd->lmd_profile);
747 GOTO(out_free, err = -EINVAL);
750 OBD_FREE(lov, strlen(lov) + 1);
751 OBD_ALLOC(lov, strlen(lprof->lp_lov) +
752 strlen(sbi->ll_instance) + 2);
753 sprintf(lov, "%s-%s", lprof->lp_lov, sbi->ll_instance);
756 OBD_FREE(lmv, strlen(lmv) + 1);
757 OBD_ALLOC(lmv, strlen(lprof->lp_lmv) +
758 strlen(sbi->ll_instance) + 2);
759 sprintf(lmv, "%s-%s", lprof->lp_lmv, sbi->ll_instance);
764 GOTO(out_free, err = -EINVAL);
769 GOTO(out_free, err = -EINVAL);
772 err = lustre_common_fill_super(sb, lmv, lov, lmd->lmd_async,
773 lmd->lmd_security, &lmd->lmd_nllu,
774 &lmd->lmd_remote_flag);
781 OBD_FREE(lmv, strlen(lmv) + 1);
783 OBD_FREE(lov, strlen(lov) + 1);
789 int len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean")+1;
792 if (sbi->ll_instance != NULL) {
793 struct lustre_mount_data *lmd = sbi->ll_lmd;
794 struct config_llog_instance cfg;
797 cfg.cfg_instance = sbi->ll_instance;
798 cfg.cfg_uuid = sbi->ll_sb_uuid;
800 OBD_ALLOC(cl_prof, len);
801 sprintf(cl_prof, "%s-clean", lmd->lmd_profile);
802 err = lustre_process_log(lmd, cl_prof, &cfg, 0);
804 CERROR("Unable to process log: %s\n", cl_prof);
805 lustre_manual_cleanup(sbi);
807 OBD_FREE(cl_prof, len);
808 OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance) + 1);
810 OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
814 } /* lustre_fill_super */
816 void lustre_put_super(struct super_block *sb)
818 struct obd_device *obd;
819 struct ll_sb_info *sbi = ll_s2sbi(sb);
820 int force_umount = 0;
823 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
824 obd = class_exp2obd(sbi->ll_md_exp);
826 force_umount = obd->obd_no_recov;
829 lustre_common_put_super(sb);
830 if (sbi->ll_lmd != NULL) {
832 int len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean")+1;
834 struct config_llog_instance cfg;
837 CERROR("force umount, doing manual cleanup\n");
838 lustre_manual_cleanup(sbi);
842 cfg.cfg_instance = sbi->ll_instance;
843 cfg.cfg_uuid = sbi->ll_sb_uuid;
845 OBD_ALLOC(cl_prof, len);
846 sprintf(cl_prof, "%s-clean", sbi->ll_lmd->lmd_profile);
847 err = lustre_process_log(sbi->ll_lmd, cl_prof, &cfg, 0);
849 CERROR("Unable to process log: %s, doing manual cleanup"
851 lustre_manual_cleanup(sbi);
854 OBD_FREE(cl_prof, len);
856 OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
857 OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance) + 1);
863 } /* lustre_put_super */
865 int ll_process_config_update(struct ll_sb_info *sbi, int clean)
867 struct obd_export *md_exp = sbi->ll_md_exp;
868 struct lustre_mount_data *lmd = sbi->ll_lmd;
869 struct llog_ctxt *ctxt;
870 struct config_llog_instance cfg;
871 char *profile = lmd->lmd_profile, *name = NULL;
872 int rc, namelen = 0, version;
878 CERROR("Client not mounted with zero-conf; cannot process "
883 rc = ldlm_cli_cancel_unused(md_exp->exp_obd->obd_namespace, NULL,
884 LDLM_FL_CONFIG_CHANGE, NULL);
886 CWARN("ldlm_cli_cancel_unused(mdc): %d\n", rc);
888 rc = obd_cancel_unused(sbi->ll_dt_exp, NULL, LDLM_FL_CONFIG_CHANGE,
891 CWARN("obd_cancel_unused(lov): %d\n", rc);
893 cfg.cfg_instance = sbi->ll_instance;
894 cfg.cfg_uuid = sbi->ll_sb_uuid;
895 cfg.cfg_local_nid = lmd->lmd_local_nid;
897 namelen = strlen(profile) + 20; /* -clean-######### */
898 OBD_ALLOC(name, namelen);
903 version = sbi->ll_config_version - 1;
904 sprintf(name, "%s-clean-%d", profile, version);
906 version = sbi->ll_config_version + 1;
907 sprintf(name, "%s-%d", profile, version);
910 CWARN("Applying configuration log %s\n", name);
912 ctxt = llog_get_context(&md_exp->exp_obd->obd_llogs,
913 LLOG_CONFIG_REPL_CTXT);
914 rc = class_config_process_llog(ctxt, name, &cfg);
916 sbi->ll_config_version = version;
917 CWARN("Finished applying configuration log %s: %d\n", name, rc);
919 if (rc == 0 && clean == 0) {
920 struct lov_desc desc;
922 valsize = sizeof(desc);
923 rc = obd_get_info(sbi->ll_dt_exp, strlen("lovdesc") + 1,
924 "lovdesc", &valsize, &desc);
926 rc = obd_init_ea_size(md_exp,
927 obd_size_diskmd(sbi->ll_dt_exp, NULL),
929 sizeof(struct llog_cookie)));
931 OBD_FREE(name, namelen);
935 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
937 struct inode *inode = NULL;
938 l_lock(&lock->l_resource->lr_namespace->ns_lock);
939 if (lock->l_ast_data) {
940 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
941 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
942 inode = igrab(lock->l_ast_data);
944 inode = lock->l_ast_data;
945 CDEBUG(inode->i_state & I_FREEING ? D_INFO : D_WARNING,
946 "l_ast_data %p is bogus: magic %0x8\n",
947 lock->l_ast_data, lli->lli_inode_magic);
951 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
955 int null_if_equal(struct ldlm_lock *lock, void *data)
957 if (data == lock->l_ast_data) {
958 lock->l_ast_data = NULL;
960 if (lock->l_req_mode != lock->l_granted_mode)
961 LDLM_ERROR(lock,"clearing inode with ungranted lock\n");
964 return LDLM_ITER_CONTINUE;
967 void ll_clear_inode(struct inode *inode)
970 struct ll_inode_info *lli = ll_i2info(inode);
971 struct ll_sb_info *sbi = ll_i2sbi(inode);
974 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
975 inode->i_generation, inode);
977 lli->lli_inode_magic = LLI_INODE_DEAD;
978 ll_inode2id(&id, inode);
980 clear_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &(ll_i2info(inode)->lli_flags));
981 md_change_cbdata(sbi->ll_md_exp, &id, null_if_equal, inode);
983 LASSERT(!lli->lli_open_fd_write_count);
984 LASSERT(!lli->lli_open_fd_read_count);
985 LASSERT(!lli->lli_open_fd_exec_count);
987 if (lli->lli_mds_write_och)
988 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_WRITE);
989 if (lli->lli_mds_exec_och)
990 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_EXEC);
991 if (lli->lli_mds_read_och)
992 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_READ);
995 obd_change_cbdata(sbi->ll_dt_exp, lli->lli_smd,
996 null_if_equal, inode);
999 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1000 lli->lli_smd = NULL;
1004 obd_free_memmd(sbi->ll_md_exp,
1005 (struct lov_stripe_md **) &lli->lli_mea);
1006 lli->lli_mea = NULL;
1009 if (lli->lli_symlink_name) {
1010 OBD_FREE(lli->lli_symlink_name,
1011 strlen(lli->lli_symlink_name) + 1);
1012 lli->lli_symlink_name = NULL;
1014 lli->lli_inode_magic = LLI_INODE_DEAD;
1019 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1020 * object(s) determine the file size and mtime. Otherwise, the MDS will
1021 * keep these values until such a time that objects are allocated for it.
1022 * We do the MDS operations first, as it is checking permissions for us.
1023 * We don't to the MDS RPC if there is nothing that we want to store there,
1024 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1025 * going to do an RPC anyways.
1027 * If we are doing a truncate, we will send the mtime and ctime updates
1028 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1029 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1032 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1034 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1035 struct ll_sb_info *sbi = ll_i2sbi(inode);
1036 struct ptlrpc_request *request = NULL;
1037 struct mdc_op_data *op_data;
1038 int ia_valid = attr->ia_valid;
1042 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu\n", inode->i_ino);
1043 lprocfs_counter_incr(ll_i2sbi(inode)->ll_stats, LPROC_LL_SETATTR);
1045 if (ia_valid & ATTR_SIZE) {
1046 if (attr->ia_size > ll_file_maxbytes(inode)) {
1047 CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1048 attr->ia_size, ll_file_maxbytes(inode));
1052 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1055 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1056 if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1057 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1061 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1062 if (attr->ia_valid & ATTR_CTIME) {
1063 attr->ia_ctime = CURRENT_TIME;
1064 attr->ia_valid |= ATTR_CTIME_SET;
1066 if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1067 attr->ia_atime = CURRENT_TIME;
1068 attr->ia_valid |= ATTR_ATIME_SET;
1070 if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1071 attr->ia_mtime = CURRENT_TIME;
1072 attr->ia_valid |= ATTR_MTIME_SET;
1075 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1076 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1077 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1078 LTIME_S(CURRENT_TIME));
1081 attr->ia_valid &= ~ATTR_SIZE;
1083 /* If only OST attributes being set on objects, don't do MDS RPC.
1084 * In that case, we need to check permissions and update the local
1085 * inode ourselves so we can call obdo_from_inode() always. */
1086 if (ia_valid & (lsm ? ~(ATTR_SIZE | ATTR_FROM_OPEN /*| ATTR_RAW*/) : ~0)) {
1087 struct lustre_md md;
1089 OBD_ALLOC(op_data, sizeof(*op_data));
1090 if (op_data == NULL)
1092 ll_prepare_mdc_data(op_data, inode, NULL, NULL, 0, 0);
1094 rc = md_setattr(sbi->ll_md_exp, op_data,
1095 attr, NULL, 0, NULL, 0, &request);
1096 OBD_FREE(op_data, sizeof(*op_data));
1098 ptlrpc_req_finished(request);
1099 if (rc != -EPERM && rc != -EACCES)
1100 CERROR("md_setattr fails: rc = %d\n", rc);
1104 rc = mdc_req2lustre_md(sbi->ll_md_exp, request, 0,
1105 sbi->ll_dt_exp, &md);
1107 ptlrpc_req_finished(request);
1111 /* We call inode_setattr to adjust timestamps, but we first
1112 * clear ATTR_SIZE to avoid invoking vmtruncate.
1114 * NB: ATTR_SIZE will only be set at this point if the size
1115 * resides on the MDS, ie, this file has no objects. */
1116 attr->ia_valid &= ~ATTR_SIZE;
1119 * assigning inode_setattr() to @err to disable warning that
1120 * function's result should be checked by by caller. error is
1121 * impossible here, as vmtruncate() control path is disabled.
1123 err = inode_setattr(inode, attr);
1124 ll_update_inode(inode, &md);
1125 ptlrpc_req_finished(request);
1127 if (!lsm || !S_ISREG(inode->i_mode)) {
1128 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1132 /* The OST doesn't check permissions, but the alternative is
1133 * a gratuitous RPC to the MDS. We already rely on the client
1134 * to do read/write/truncate permission checks, so is mtime OK?
1136 if (ia_valid & (ATTR_MTIME | ATTR_ATIME)) {
1137 /* from sys_utime() */
1138 if (!(ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET))) {
1139 if (current->fsuid != inode->i_uid &&
1140 (rc=ll_permission(inode,MAY_WRITE,NULL))!=0)
1143 /* from inode_change_ok() */
1144 if (current->fsuid != inode->i_uid &&
1145 !capable(CAP_FOWNER))
1150 /* won't invoke vmtruncate, as we already cleared ATTR_SIZE */
1151 err = inode_setattr(inode, attr);
1153 * assigning inode_setattr() to @err to disable warning that
1154 * function's result should be checked by by caller. error is
1155 * impossible here, as vmtruncate() control path is disabled.
1159 /* We really need to get our PW lock before we change inode->i_size.
1160 * If we don't we can race with other i_size updaters on our node, like
1161 * ll_file_read. We can also race with i_size propogation to other
1162 * nodes through dirtying and writeback of final cached pages. This
1163 * last one is especially bad for racing o_append users on other
1165 if (ia_valid & ATTR_SIZE) {
1166 ldlm_policy_data_t policy = { .l_extent = {attr->ia_size,
1168 struct lustre_handle lockh = { 0 };
1169 struct ll_inode_info *lli = ll_i2info(inode);
1170 int err, ast_flags = 0;
1171 /* XXX when we fix the AST intents to pass the discard-range
1172 * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
1174 if (attr->ia_size == 0)
1175 ast_flags = LDLM_AST_DISCARD_DATA;
1177 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy, &lockh,
1178 ast_flags, &ll_i2sbi(inode)->ll_seek_stime);
1183 down(&lli->lli_size_sem);
1184 lli->lli_size_pid = current->pid;
1185 rc = vmtruncate(inode, attr->ia_size);
1187 LASSERT(atomic_read(&lli->lli_size_sem.count) <= 0);
1188 lli->lli_size_pid = 0;
1189 up(&lli->lli_size_sem);
1192 err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
1194 CERROR("ll_extent_unlock failed: %d\n", err);
1198 } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
1199 struct obdo *oa = NULL;
1201 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1202 inode->i_ino, LTIME_S(attr->ia_mtime));
1208 oa->o_id = lsm->lsm_object_id;
1209 oa->o_gr = lsm->lsm_object_gr;
1210 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1211 obdo_from_inode(oa, inode, OBD_MD_FLTYPE | OBD_MD_FLATIME |
1212 OBD_MD_FLMTIME | OBD_MD_FLCTIME);
1213 rc = obd_setattr(sbi->ll_dt_exp, oa, lsm, NULL);
1216 CERROR("obd_setattr fails: rc = %d\n", rc);
1221 int ll_setattr(struct dentry *de, struct iattr *attr)
1223 LASSERT(de->d_inode);
1224 return ll_setattr_raw(de->d_inode, attr);
1227 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1228 unsigned long max_age)
1230 struct ll_sb_info *sbi = ll_s2sbi(sb);
1231 struct obd_statfs obd_osfs;
1235 rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age);
1237 CERROR("obd_statfs fails: rc = %d\n", rc);
1241 osfs->os_type = sb->s_magic;
1243 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1244 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1246 rc = obd_statfs(class_exp2obd(sbi->ll_dt_exp), &obd_osfs, max_age);
1248 CERROR("obd_statfs fails: rc = %d\n", rc);
1252 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1253 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1256 osfs->os_blocks = obd_osfs.os_blocks;
1257 osfs->os_bfree = obd_osfs.os_bfree;
1258 osfs->os_bavail = obd_osfs.os_bavail;
1260 /* If we don't have as many objects free on the OST as inodes
1261 * on the MDS, we reduce the total number of inodes to
1262 * compensate, so that the "inodes in use" number is correct.
1264 if (obd_osfs.os_ffree < osfs->os_ffree) {
1265 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1267 osfs->os_ffree = obd_osfs.os_ffree;
1273 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1275 struct obd_statfs osfs;
1278 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p\n", sb);
1279 lprocfs_counter_incr(ll_s2sbi(sb)->ll_stats, LPROC_LL_STAFS);
1281 /* For now we will always get up-to-date statfs values, but in the
1282 * future we may allow some amount of caching on the client (e.g.
1283 * from QOS or lprocfs updates). */
1284 rc = ll_statfs_internal(sb, &osfs, jiffies - 1);
1288 statfs_unpack(sfs, &osfs);
1290 if (sizeof(sfs->f_blocks) == 4) {
1291 while (osfs.os_blocks > ~0UL) {
1294 osfs.os_blocks >>= 1;
1295 osfs.os_bfree >>= 1;
1296 osfs.os_bavail >>= 1;
1300 sfs->f_blocks = osfs.os_blocks;
1301 sfs->f_bfree = osfs.os_bfree;
1302 sfs->f_bavail = osfs.os_bavail;
1307 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1309 struct ll_inode_info *lli = ll_i2info(inode);
1310 struct lov_stripe_md *lsm = md->lsm;
1311 struct mds_body *body = md->body;
1312 struct mea *mea = md->mea;
1313 struct posix_acl *ll_acl_access = md->acl_access;
1316 LASSERT((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1318 if (md->lsm && md->lsm->lsm_magic != LOV_MAGIC) {
1319 /* check for default striping info for dir. */
1320 LASSERT((mea != NULL) == ((body->valid & OBD_MD_FLDIREA) != 0));
1324 LASSERT(lsm->lsm_object_gr > 0);
1325 if (lli->lli_smd == NULL) {
1327 lli->lli_maxbytes = lsm->lsm_maxbytes;
1328 if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1329 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1332 if (memcmp(lli->lli_smd, lsm, sizeof(*lsm))) {
1333 CERROR("lsm mismatch for inode %ld\n",
1335 CERROR("lli_smd:\n");
1336 dump_lsm(D_ERROR, lli->lli_smd);
1338 dump_lsm(D_ERROR, lsm);
1341 /* XXX FIXME -- We should decide on a safer (atomic) and
1342 * more elegant way to update the lsm */
1343 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1344 lli->lli_smd->lsm_oinfo[i].loi_id =
1345 lsm->lsm_oinfo[i].loi_id;
1346 lli->lli_smd->lsm_oinfo[i].loi_gr =
1347 lsm->lsm_oinfo[i].loi_gr;
1348 lli->lli_smd->lsm_oinfo[i].loi_ost_idx =
1349 lsm->lsm_oinfo[i].loi_ost_idx;
1350 lli->lli_smd->lsm_oinfo[i].loi_ost_gen =
1351 lsm->lsm_oinfo[i].loi_ost_gen;
1354 /* bug 2844 - limit i_blksize for broken user-space apps */
1355 LASSERTF(lsm->lsm_xfersize != 0, "%lu\n", lsm->lsm_xfersize);
1356 inode->i_blksize = min(lsm->lsm_xfersize, LL_MAX_BLKSIZE);
1357 if (lli->lli_smd != lsm)
1358 obd_free_memmd(ll_i2dtexp(inode), &lsm);
1362 if (lli->lli_mea == NULL) {
1365 if (memcmp(lli->lli_mea, mea, body->eadatasize)) {
1366 CERROR("mea mismatch for inode %lu\n",
1371 if (lli->lli_mea != mea)
1372 obd_free_memmd(ll_i2mdexp(inode),
1373 (struct lov_stripe_md **) &mea);
1376 if (body->valid & OBD_MD_FID)
1377 id_assign_fid(&lli->lli_id, &body->id1);
1379 if (body->valid & OBD_MD_FLID)
1380 id_ino(&lli->lli_id) = id_ino(&body->id1);
1382 if (body->valid & OBD_MD_FLGENER)
1383 id_gen(&lli->lli_id) = id_gen(&body->id1);
1385 spin_lock(&lli->lli_lock);
1386 if (ll_acl_access != NULL) {
1387 if (lli->lli_acl_access != NULL)
1388 posix_acl_release(lli->lli_acl_access);
1389 lli->lli_acl_access = ll_acl_access;
1391 spin_unlock(&lli->lli_lock);
1393 if (body->valid & OBD_MD_FLID)
1394 inode->i_ino = id_ino(&body->id1);
1395 if (body->valid & OBD_MD_FLGENER)
1396 inode->i_generation = id_gen(&body->id1);
1397 if (body->valid & OBD_MD_FLATIME)
1398 LTIME_S(inode->i_atime) = body->atime;
1399 if (body->valid & OBD_MD_FLMTIME &&
1400 body->mtime > LTIME_S(inode->i_mtime)) {
1401 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to %u\n",
1402 inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
1403 LTIME_S(inode->i_mtime) = body->mtime;
1405 if (body->valid & OBD_MD_FLCTIME &&
1406 body->ctime > LTIME_S(inode->i_ctime))
1407 LTIME_S(inode->i_ctime) = body->ctime;
1408 if (body->valid & OBD_MD_FLMODE) {
1409 inode->i_mode = (inode->i_mode & S_IFMT) |
1410 (body->mode & ~S_IFMT);
1412 if (body->valid & OBD_MD_FLTYPE) {
1413 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1414 (body->mode & S_IFMT);
1416 if (body->valid & OBD_MD_FLUID)
1417 inode->i_uid = body->uid;
1418 if (body->valid & OBD_MD_FLGID)
1419 inode->i_gid = body->gid;
1420 if (body->valid & OBD_MD_FLFLAGS)
1421 inode->i_flags = body->flags;
1422 if (body->valid & OBD_MD_FLNLINK)
1423 inode->i_nlink = body->nlink;
1424 if (body->valid & OBD_MD_FLRDEV)
1425 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1426 inode->i_rdev = body->rdev;
1428 inode->i_rdev = old_decode_dev(body->rdev);
1430 if (body->valid & OBD_MD_FLSIZE)
1431 inode->i_size = body->size;
1432 if (body->valid & OBD_MD_FLBLOCKS)
1433 inode->i_blocks = body->blocks;
1435 if (body->valid & OBD_MD_FLSIZE)
1436 set_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &lli->lli_flags);
1438 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1439 inode->i_dev = (kdev_t)id_group(&lli->lli_id);
1441 LASSERT(id_fid(&lli->lli_id) != 0);
1444 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
1445 static struct backing_dev_info ll_backing_dev_info = {
1446 .ra_pages = 0, /* No readahead */
1447 .memory_backed = 0, /* Does contribute to dirty memory */
1451 void ll_read_inode2(struct inode *inode, void *opaque)
1453 struct lustre_md *md = opaque;
1454 struct ll_inode_info *lli = ll_i2info(inode);
1457 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1458 inode->i_generation, inode);
1462 LASSERT(!lli->lli_smd);
1464 /* Core attributes from the MDS first. This is a new inode, and
1465 * the VFS doesn't zero times in the core inode so we have to do
1466 * it ourselves. They will be overwritten by either MDS or OST
1467 * attributes - we just need to make sure they aren't newer. */
1468 LTIME_S(inode->i_mtime) = 0;
1469 LTIME_S(inode->i_atime) = 0;
1470 LTIME_S(inode->i_ctime) = 0;
1473 ll_update_inode(inode, md);
1475 /* OIDEBUG(inode); */
1477 if (S_ISREG(inode->i_mode)) {
1478 inode->i_op = &ll_file_inode_operations;
1479 inode->i_fop = &ll_file_operations;
1480 inode->i_mapping->a_ops = &ll_aops;
1482 } else if (S_ISDIR(inode->i_mode)) {
1483 inode->i_op = &ll_dir_inode_operations;
1484 inode->i_fop = &ll_dir_operations;
1485 inode->i_mapping->a_ops = &ll_dir_aops;
1487 } else if (S_ISLNK(inode->i_mode)) {
1488 inode->i_op = &ll_fast_symlink_inode_operations;
1491 inode->i_op = &ll_special_inode_operations;
1493 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
1494 init_special_inode(inode, inode->i_mode,
1495 kdev_t_to_nr(inode->i_rdev));
1497 /* initializing backing dev info. */
1498 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
1500 init_special_inode(inode, inode->i_mode, inode->i_rdev);
1502 lli->ll_save_ifop = inode->i_fop;
1504 if (S_ISCHR(inode->i_mode))
1505 inode->i_fop = &ll_special_chr_inode_fops;
1506 else if (S_ISBLK(inode->i_mode))
1507 inode->i_fop = &ll_special_blk_inode_fops;
1508 else if (S_ISFIFO(inode->i_mode))
1509 inode->i_fop = &ll_special_fifo_inode_fops;
1510 else if (S_ISSOCK(inode->i_mode))
1511 inode->i_fop = &ll_special_sock_inode_fops;
1513 CWARN("saved %p, replaced with %p\n", lli->ll_save_ifop,
1516 if (lli->ll_save_ifop->owner) {
1517 CWARN("%p has owner %p\n", lli->ll_save_ifop,
1518 lli->ll_save_ifop->owner);
1524 void ll_delete_inode(struct inode *inode)
1526 struct ll_sb_info *sbi = ll_i2sbi(inode);
1527 struct lustre_id id;
1531 ll_inode2id(&id, inode);
1533 rc = md_delete_inode(sbi->ll_md_exp, &id);
1535 CERROR("md_delete_inode() failed, error %d\n",
1543 int ll_iocontrol(struct inode *inode, struct file *file,
1544 unsigned int cmd, unsigned long arg)
1546 struct ll_sb_info *sbi = ll_i2sbi(inode);
1547 struct ptlrpc_request *req = NULL;
1552 case EXT3_IOC_GETFLAGS: {
1553 struct lustre_id id;
1554 __u64 valid = OBD_MD_FLFLAGS;
1555 struct mds_body *body;
1557 ll_inode2id(&id, inode);
1558 rc = md_getattr(sbi->ll_md_exp, &id, valid, NULL, 0, 0, &req);
1560 CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1564 body = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*body));
1566 if (body->flags & S_APPEND)
1567 flags |= EXT3_APPEND_FL;
1568 if (body->flags & S_IMMUTABLE)
1569 flags |= EXT3_IMMUTABLE_FL;
1570 if (body->flags & S_NOATIME)
1571 flags |= EXT3_NOATIME_FL;
1573 ptlrpc_req_finished (req);
1575 RETURN(put_user(flags, (int *)arg));
1577 case EXT3_IOC_SETFLAGS: {
1578 struct mdc_op_data *op_data;
1581 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1583 if (get_user(flags, (int *)arg))
1590 OBD_ALLOC(op_data, sizeof(*op_data));
1591 if (op_data == NULL) {
1595 ll_prepare_mdc_data(op_data, inode, NULL, NULL, 0, 0);
1597 memset(&attr, 0x0, sizeof(attr));
1598 attr.ia_attr_flags = flags;
1599 attr.ia_valid |= ATTR_ATTR_FLAG;
1601 rc = md_setattr(sbi->ll_md_exp, op_data,
1602 &attr, NULL, 0, NULL, 0, &req);
1603 OBD_FREE(op_data, sizeof(*op_data));
1605 ptlrpc_req_finished(req);
1606 if (rc != -EPERM && rc != -EACCES)
1607 CERROR("md_setattr fails: rc = %d\n", rc);
1611 ptlrpc_req_finished(req);
1613 oa->o_id = lsm->lsm_object_id;
1614 oa->o_gr = lsm->lsm_object_gr;
1615 oa->o_flags = flags;
1616 oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS | OBD_MD_FLGROUP;
1618 rc = obd_setattr(sbi->ll_dt_exp, oa, lsm, NULL);
1621 if (rc != -EPERM && rc != -EACCES)
1622 CERROR("md_setattr fails: rc = %d\n", rc);
1626 if (flags & EXT3_APPEND_FL)
1627 inode->i_flags |= S_APPEND;
1629 inode->i_flags &= ~S_APPEND;
1630 if (flags & EXT3_IMMUTABLE_FL)
1631 inode->i_flags |= S_IMMUTABLE;
1633 inode->i_flags &= ~S_IMMUTABLE;
1634 if (flags & EXT3_NOATIME_FL)
1635 inode->i_flags |= S_NOATIME;
1637 inode->i_flags &= ~S_NOATIME;
1648 /* this is only called in the case of forced umount. */
1649 void ll_umount_begin(struct super_block *sb)
1651 struct ll_sb_info *sbi = ll_s2sbi(sb);
1652 struct obd_ioctl_data ioc_data = { 0 };
1653 struct obd_device *obd;
1656 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1657 sb->s_count, atomic_read(&sb->s_active));
1659 obd = class_exp2obd(sbi->ll_md_exp);
1661 CERROR("Invalid MDC connection handle "LPX64"\n",
1662 sbi->ll_md_exp->exp_handle.h_cookie);
1666 obd->obd_no_recov = 1;
1667 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
1668 sizeof(ioc_data), &ioc_data, NULL);
1670 obd = class_exp2obd(sbi->ll_dt_exp);
1672 CERROR("Invalid LOV connection handle "LPX64"\n",
1673 sbi->ll_dt_exp->exp_handle.h_cookie);
1678 obd->obd_no_recov = 1;
1679 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
1680 sizeof(ioc_data), &ioc_data, NULL);
1683 * really, we'd like to wait until there are no requests outstanding,
1684 * and then continue. For now, we just invalidate the requests,
1685 * schedule, and hope.
1692 int ll_prep_inode(struct obd_export *dt_exp, struct obd_export *md_exp,
1693 struct inode **inode, struct ptlrpc_request *req,
1694 int offset, struct super_block *sb)
1696 struct lustre_md md;
1699 rc = mdc_req2lustre_md(md_exp, req, offset, dt_exp, &md);
1704 ll_update_inode(*inode, &md);
1707 *inode = ll_iget(sb, id_ino(&md.body->id1), &md);
1708 if (*inode == NULL || is_bad_inode(*inode)) {
1709 /* free the lsm if we allocated one above */
1711 obd_free_memmd(dt_exp, &md.lsm);
1713 obd_free_memmd(md_exp,
1714 (struct lov_stripe_md**)&md.mea);
1716 CERROR("new_inode -fatal: rc %d\n", rc);
1723 int ll_get_fid(struct obd_export *exp, struct lustre_id *idp,
1724 char *filename, struct lustre_id *ret)
1726 struct ptlrpc_request *request = NULL;
1727 struct mds_body *body;
1730 rc = md_getattr_lock(exp, idp, filename, strlen(filename) + 1,
1731 OBD_MD_FID, 0, &request);
1733 CDEBUG(D_INFO, "md_getattr_lock failed on %s: rc %d\n",
1738 body = lustre_msg_buf(request->rq_repmsg, 0, sizeof(*body));
1739 LASSERT(body != NULL);
1740 LASSERT_REPSWABBED(request, 0);
1743 ptlrpc_req_finished(request);