1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Lustre Light Super operations
6 * Copyright (c) 2002, 2003 Cluster File Systems, Inc.
8 * This file is part of Lustre, http://www.lustre.org.
10 * Lustre is free software; you can redistribute it and/or
11 * modify it under the terms of version 2 of the GNU General Public
12 * License as published by the Free Software Foundation.
14 * Lustre is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
17 * GNU General Public License for more details.
19 * You should have received a copy of the GNU General Public License
20 * along with Lustre; if not, write to the Free Software
21 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
24 #define DEBUG_SUBSYSTEM S_LLITE
26 #include <linux/module.h>
27 #include <linux/types.h>
28 #include <linux/random.h>
29 #include <linux/version.h>
31 #include <linux/lustre_lite.h>
32 #include <linux/lustre_ha.h>
33 #include <linux/lustre_dlm.h>
34 #include <linux/lprocfs_status.h>
35 #include <linux/lustre_acl.h>
36 #include "llite_internal.h"
38 kmem_cache_t *ll_file_data_slab;
39 kmem_cache_t *ll_intent_slab;
41 extern struct address_space_operations ll_aops;
42 extern struct address_space_operations ll_dir_aops;
45 #define log2(n) ffz(~(n))
48 struct ll_sb_info *lustre_init_sbi(struct super_block *sb)
50 struct ll_sb_info *sbi = NULL;
54 OBD_ALLOC(sbi, sizeof(*sbi));
58 spin_lock_init(&sbi->ll_lock);
59 INIT_LIST_HEAD(&sbi->ll_pglist);
60 sbi->ll_pglist_gen = 0;
61 if (num_physpages < SBI_DEFAULT_RA_MAX / 4)
62 sbi->ll_ra_info.ra_max_pages = num_physpages / 4;
64 sbi->ll_ra_info.ra_max_pages = SBI_DEFAULT_RA_MAX;
65 INIT_LIST_HEAD(&sbi->ll_conn_chain);
66 INIT_HLIST_HEAD(&sbi->ll_orphan_dentry_list);
67 INIT_LIST_HEAD(&sbi->ll_mnt_list);
69 sema_init(&sbi->ll_gns_sem, 1);
70 spin_lock_init(&sbi->ll_gns_lock);
71 INIT_LIST_HEAD(&sbi->ll_gns_sbi_head);
72 init_waitqueue_head(&sbi->ll_gns_waitq);
73 init_completion(&sbi->ll_gns_mount_finished);
75 /* this later may be reset via /proc/fs/... */
76 memcpy(sbi->ll_gns_oname, ".mntinfo", strlen(".mntinfo"));
77 sbi->ll_gns_oname[strlen(sbi->ll_gns_oname)] = '\0';
79 /* this later may be reset via /proc/fs/... */
80 memcpy(sbi->ll_gns_upcall, "/usr/sbin/gns_upcall",
81 strlen("/usr/sbin/gns_upcall"));
82 sbi->ll_gns_upcall[strlen(sbi->ll_gns_upcall)] = '\0';
84 /* default values, may be changed via /proc/fs/... */
85 sbi->ll_gns_state = LL_GNS_IDLE;
86 sbi->ll_gns_pending_dentry = NULL;
87 atomic_set(&sbi->ll_gns_enabled, 1);
88 sbi->ll_gns_tick = GNS_TICK_TIMEOUT;
89 sbi->ll_gns_timeout = GNS_MOUNT_TIMEOUT;
91 sbi->ll_gns_timer.data = (unsigned long)sbi;
92 sbi->ll_gns_timer.function = ll_gns_timer_callback;
93 init_timer(&sbi->ll_gns_timer);
97 generate_random_uuid(uuid);
98 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
102 void lustre_free_sbi(struct super_block *sb)
104 struct ll_sb_info *sbi = ll_s2sbi(sb);
108 list_del(&sbi->ll_gns_sbi_head);
109 del_timer(&sbi->ll_gns_timer);
110 OBD_FREE(sbi, sizeof(*sbi));
112 ll_set_sbi(sb, NULL);
116 int lustre_init_dt_desc(struct ll_sb_info *sbi)
122 valsize = sizeof(sbi->ll_dt_desc);
123 memset(&sbi->ll_dt_desc, 0, sizeof(sbi->ll_dt_desc));
124 rc = obd_get_info(sbi->ll_dt_exp, strlen("lovdesc") + 1,
125 "lovdesc", &valsize, &sbi->ll_dt_desc);
129 extern struct dentry_operations ll_d_ops;
131 int lustre_common_fill_super(struct super_block *sb, char *lmv, char *lov,
132 int async, char *mds_security, char *oss_security,
133 __u32 *nllu, __u64 *remote)
135 struct ll_sb_info *sbi = ll_s2sbi(sb);
136 struct ptlrpc_request *request = NULL;
137 struct lustre_handle dt_conn = {0, };
138 struct lustre_handle md_conn = {0, };
139 struct obd_connect_data *data;
140 struct inode *root = NULL;
141 struct obd_device *obd;
142 struct obd_statfs osfs;
148 obd = class_name2obd(lmv);
150 CERROR("MDC %s: not setup or attached\n", lmv);
153 obd_set_info(obd->obd_self_export, strlen("async"), "async",
154 sizeof(async), &async);
156 if ((*remote & (OBD_CONNECT_LOCAL | OBD_CONNECT_REMOTE)) ==
157 (OBD_CONNECT_LOCAL | OBD_CONNECT_REMOTE)) {
158 CERROR("wrong remote flag "LPX64"\n", *remote);
162 OBD_ALLOC(data, sizeof(*data));
166 data->ocd_connect_flags |= *remote & (OBD_CONNECT_LOCAL |
168 memcpy(data->ocd_nllu, nllu, sizeof(data->ocd_nllu));
170 if (mds_security == NULL)
171 mds_security = "null";
173 err = obd_set_info(obd->obd_self_export, strlen("sec"), "sec",
174 strlen(mds_security), mds_security);
176 CERROR("LMV %s: failed to set security %s, err %d\n",
177 lmv, mds_security, err);
178 OBD_FREE(data, sizeof(*data));
182 if (proc_lustre_fs_root) {
183 err = lprocfs_register_mountpoint(proc_lustre_fs_root,
186 CERROR("could not register mount in /proc/lustre");
189 err = obd_connect(&md_conn, obd, &sbi->ll_sb_uuid, data,
190 OBD_OPT_REAL_CLIENT);
192 CERROR("An MDS (lmv %s) is performing recovery, of which this"
193 " client is not a part. Please wait for recovery to "
194 "complete, abort, or time out.\n", lmv);
197 CERROR("cannot connect to %s: rc = %d\n", lmv, err);
200 sbi->ll_md_exp = class_conn2export(&md_conn);
201 err = obd_statfs(obd, &osfs, jiffies - HZ);
205 if (!osfs.os_bsize) {
206 CERROR("Invalid block size is detected.");
210 sb->s_magic = LL_SUPER_MAGIC;
211 sb->s_blocksize = osfs.os_bsize;
212 sb->s_blocksize_bits = log2(osfs.os_bsize);
213 sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
215 /* in 2.6.x FS is not allowed to form s_dev */
216 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
220 devno = get_uuid2int((char *)sbi->ll_md_exp->exp_obd->obd_uuid.uuid,
221 strlen((char *)sbi->ll_md_exp->exp_obd->obd_uuid.uuid));
227 /* after statfs, we are supposed to have connected to MDSs,
228 * so it's ok to check remote flag returned.
230 valsize = sizeof(&sbi->ll_remote);
231 err = obd_get_info(sbi->ll_md_exp, strlen("remote_flag"), "remote_flag",
232 &valsize, &sbi->ll_remote);
234 CERROR("fail to obtain remote flag\n");
238 obd = class_name2obd(lov);
240 CERROR("OSC %s: not setup or attached\n", lov);
243 obd_set_info(obd->obd_self_export, strlen("async"), "async",
244 sizeof(async), &async);
246 if (oss_security == NULL)
247 oss_security = "null";
249 err = obd_set_info(obd->obd_self_export, strlen("sec"), "sec",
250 strlen(oss_security), oss_security);
252 CERROR("LOV %s: failed to set security %s, err %d\n",
253 lov, oss_security, err);
254 OBD_FREE(data, sizeof(*data));
258 err = obd_connect(&dt_conn, obd, &sbi->ll_sb_uuid, data, 0);
260 CERROR("An OST (lov %s) is performing recovery, of which this"
261 " client is not a part. Please wait for recovery to "
262 "complete, abort, or time out.\n", lov);
265 CERROR("cannot connect to %s: rc = %d\n", lov, err);
268 sbi->ll_dt_exp = class_conn2export(&dt_conn);
270 err = lustre_init_dt_desc(sbi);
272 int mdsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
273 obd_init_ea_size(sbi->ll_md_exp, mdsize,
274 sbi->ll_dt_desc.ld_tgt_count *
275 sizeof(struct llog_cookie));
278 err = md_getstatus(sbi->ll_md_exp, &sbi->ll_rootid);
280 CERROR("cannot mds_connect: rc = %d\n", err);
283 CDEBUG(D_SUPER, "rootid "DLID4"\n", OLID4(&sbi->ll_rootid));
285 sb->s_op = &lustre_super_operations;
287 /* make root inode */
288 err = md_getattr(sbi->ll_md_exp, &sbi->ll_rootid,
289 (OBD_MD_FLNOTOBD | OBD_MD_FLBLOCKS | OBD_MD_FID),
290 NULL, 0, 0, &request);
292 CERROR("md_getattr failed for root: rc = %d\n", err);
296 err = mdc_req2lustre_md(sbi->ll_md_exp, request, 0,
297 sbi->ll_dt_exp, &md);
299 CERROR("failed to understand root inode md: rc = %d\n", err);
300 ptlrpc_req_finished(request);
304 LASSERT(id_ino(&sbi->ll_rootid) != 0);
305 root = ll_iget(sb, id_ino(&sbi->ll_rootid), &md);
307 ptlrpc_req_finished(request);
309 if (root == NULL || is_bad_inode(root)) {
311 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
313 obd_free_memmd(sbi->ll_md_exp,
314 (struct lov_stripe_md**)&md.mea);
315 CERROR("lustre_lite: bad iget4 for root\n");
316 GOTO(out_root, err = -EBADF);
319 err = ll_close_thread_start(&sbi->ll_lcq);
321 CERROR("cannot start close thread: rc %d\n", err);
325 ll_gns_add_timer(sbi);
327 /* making vm readahead 0 for 2.4.x. In the case of 2.6.x,
328 backing dev info assigned to inode mapping is used for
329 determining maximal readahead. */
330 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0)) && \
331 !defined(KERNEL_HAS_AS_MAX_READAHEAD)
332 /* bug 2805 - set VM readahead to zero */
333 vm_max_readahead = vm_min_readahead = 0;
336 sb->s_root = d_alloc_root(root);
337 sb->s_root->d_op = &ll_d_ops;
339 sb->s_flags |= MS_POSIXACL;
341 CWARN("Enabling PDIROPS\n");
342 sb->s_flags |= S_PDIROPS;
346 OBD_FREE(data, sizeof(*data));
352 obd_disconnect(sbi->ll_dt_exp, 0);
354 obd_disconnect(sbi->ll_md_exp, 0);
357 OBD_FREE(data, sizeof(*data));
358 lprocfs_unregister_mountpoint(sbi);
362 void lustre_common_put_super(struct super_block *sb)
364 struct ll_sb_info *sbi = ll_s2sbi(sb);
365 struct hlist_node *tmp, *next;
368 ll_gns_del_timer(sbi);
369 ll_close_thread_shutdown(sbi->ll_lcq);
371 list_del(&sbi->ll_conn_chain);
372 obd_disconnect(sbi->ll_dt_exp, 0);
374 lprocfs_unregister_mountpoint(sbi);
375 if (sbi->ll_proc_root) {
376 lprocfs_remove(sbi->ll_proc_root);
377 sbi->ll_proc_root = NULL;
380 obd_disconnect(sbi->ll_md_exp, 0);
382 // We do this to get rid of orphaned dentries. That is not really trw.
383 hlist_for_each_safe(tmp, next, &sbi->ll_orphan_dentry_list) {
384 struct dentry *dentry = hlist_entry(tmp, struct dentry, d_hash);
385 CWARN("orphan dentry %.*s (%p->%p) at unmount\n",
386 dentry->d_name.len, dentry->d_name.name, dentry, next);
387 shrink_dcache_parent(dentry);
392 char *ll_read_opt(const char *opt, char *data)
398 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
399 if (strncmp(opt, data, strlen(opt)))
401 if ((value = strchr(data, '=')) == NULL)
405 OBD_ALLOC(retval, strlen(value) + 1);
407 CERROR("out of memory!\n");
411 memcpy(retval, value, strlen(value)+1);
412 CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
416 int ll_set_opt(const char *opt, char *data, int fl)
420 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
421 if (strncmp(opt, data, strlen(opt)))
427 void ll_options(char *options, char **lov, char **lmv, char **mds_sec,
428 char **oss_sec, int *async, int *flags)
431 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
432 char *opt_ptr = options;
442 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
443 for (this_char = strtok (options, ",");
445 this_char = strtok (NULL, ",")) {
447 while ((this_char = strsep (&opt_ptr, ",")) != NULL) {
449 CDEBUG(D_SUPER, "this_char %s\n", this_char);
450 if (!*lov && (*lov = ll_read_opt("osc", this_char)))
452 if (!*lmv && (*lmv = ll_read_opt("mdc", this_char)))
454 if (!strncmp(this_char, "lasync", strlen("lasync"))) {
458 if (!*mds_sec && (*mds_sec = ll_read_opt("mds_sec", this_char)))
460 if (!*oss_sec && (*oss_sec = ll_read_opt("oss_sec", this_char)))
462 if (!(*flags & LL_SBI_NOLCK) &&
463 ((*flags) = (*flags) |
464 ll_set_opt("nolock", this_char,
472 void ll_lli_init(struct ll_inode_info *lli)
474 sema_init(&lli->lli_open_sem, 1);
475 sema_init(&lli->lli_size_sem, 1);
477 lli->lli_size_pid = 0;
478 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
479 spin_lock_init(&lli->lli_lock);
480 INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
481 lli->lli_inode_magic = LLI_INODE_MAGIC;
482 memset(&lli->lli_id, 0, sizeof(lli->lli_id));
483 sema_init(&lli->lli_och_sem, 1);
484 lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
485 lli->lli_mds_exec_och = NULL;
486 lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
487 lli->lli_open_fd_exec_count = 0;
490 int ll_fill_super(struct super_block *sb, void *data, int silent)
492 struct ll_sb_info *sbi;
495 char *mds_sec = NULL;
496 char *oss_sec = NULL;
498 __u32 nllu[2] = { NOBODY_UID, NOBODY_GID };
499 __u64 remote_flag = 0;
502 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
504 sbi = lustre_init_sbi(sb);
508 sbi->ll_flags |= LL_SBI_READAHEAD;
509 ll_options(data, &lov, &lmv, &mds_sec, &oss_sec,
510 &async, &sbi->ll_flags);
514 GOTO(out, err = -EINVAL);
519 GOTO(out, err = -EINVAL);
522 err = lustre_common_fill_super(sb, lmv, lov, async, mds_sec, oss_sec,
530 OBD_FREE(lmv, strlen(lmv) + 1);
532 OBD_FREE(lov, strlen(lov) + 1);
534 OBD_FREE(mds_sec, strlen(mds_sec) + 1);
536 OBD_FREE(oss_sec, strlen(oss_sec) + 1);
539 } /* ll_read_super */
541 static int lustre_process_log(struct lustre_mount_data *lmd, char *profile,
542 struct config_llog_instance *cfg, int allow_recov)
544 struct lustre_cfg *lcfg = NULL;
545 struct lustre_cfg_bufs bufs;
546 struct portals_cfg pcfg;
547 char *peer = "MDS_PEER_UUID";
548 struct obd_device *obd;
549 struct lustre_handle md_conn = {0, };
550 struct obd_export *exp;
551 char *name = "mdc_dev";
553 struct obd_uuid lmv_uuid;
554 struct llog_ctxt *ctxt;
558 if (lmd_bad_magic(lmd))
561 generate_random_uuid(uuid);
562 class_uuid_unparse(uuid, &lmv_uuid);
564 if (lmd->lmd_local_nid) {
565 PCFG_INIT(pcfg, NAL_CMD_REGISTER_MYNID);
566 pcfg.pcfg_nal = lmd->lmd_nal;
567 pcfg.pcfg_nid = lmd->lmd_local_nid;
568 rc = libcfs_nal_cmd(&pcfg);
573 if (lmd->lmd_nal == SOCKNAL ||
574 lmd->lmd_nal == OPENIBNAL ||
575 lmd->lmd_nal == IIBNAL ||
576 lmd->lmd_nal == VIBNAL ||
577 lmd->lmd_nal == RANAL) {
578 PCFG_INIT(pcfg, NAL_CMD_ADD_PEER);
579 pcfg.pcfg_nal = lmd->lmd_nal;
580 pcfg.pcfg_nid = lmd->lmd_server_nid;
581 pcfg.pcfg_id = lmd->lmd_server_ipaddr;
582 pcfg.pcfg_misc = lmd->lmd_port;
583 rc = libcfs_nal_cmd(&pcfg);
587 lustre_cfg_bufs_reset(&bufs, name);
588 lustre_cfg_bufs_set_string(&bufs, 1, peer);
590 lcfg = lustre_cfg_new(LCFG_ADD_UUID, &bufs);
591 lcfg->lcfg_nal = lmd->lmd_nal;
592 lcfg->lcfg_nid = lmd->lmd_server_nid;
593 LASSERT(lcfg->lcfg_nal);
594 LASSERT(lcfg->lcfg_nid);
595 err = class_process_config(lcfg);
596 lustre_cfg_free(lcfg);
598 GOTO(out_del_conn, err);
600 lustre_cfg_bufs_reset(&bufs, name);
601 lustre_cfg_bufs_set_string(&bufs, 1, LUSTRE_MDC_NAME);
602 lustre_cfg_bufs_set_string(&bufs, 2, (char *)lmv_uuid.uuid);
604 lcfg = lustre_cfg_new(LCFG_ATTACH, &bufs);
605 err = class_process_config(lcfg);
606 lustre_cfg_free(lcfg);
608 GOTO(out_del_uuid, err);
610 lustre_cfg_bufs_reset(&bufs, name);
611 lustre_cfg_bufs_set_string(&bufs, 1, lmd->lmd_mds);
612 lustre_cfg_bufs_set_string(&bufs, 2, peer);
614 lcfg = lustre_cfg_new(LCFG_SETUP, &bufs);
615 err = class_process_config(lcfg);
616 lustre_cfg_free(lcfg);
618 GOTO(out_detach, err);
620 obd = class_name2obd(name);
622 GOTO(out_cleanup, rc = -EINVAL);
624 rc = obd_set_info(obd->obd_self_export, strlen("sec"), "sec",
625 strlen(lmd->lmd_mds_security), lmd->lmd_mds_security);
627 GOTO(out_cleanup, rc);
629 /* Disable initial recovery on this import */
630 rc = obd_set_info(obd->obd_self_export,
631 strlen("initial_recov"), "initial_recov",
632 sizeof(allow_recov), &allow_recov);
634 GOTO(out_cleanup, rc);
636 rc = obd_connect(&md_conn, obd, &lmv_uuid, NULL, 0);
638 CERROR("cannot connect to %s: rc = %d\n", lmd->lmd_mds, rc);
639 GOTO(out_cleanup, rc);
642 exp = class_conn2export(&md_conn);
644 ctxt = llog_get_context(&exp->exp_obd->obd_llogs,LLOG_CONFIG_REPL_CTXT);
645 rc = class_config_process_llog(ctxt, profile, cfg);
647 CERROR("class_config_process_llog failed: rc = %d\n", rc);
649 err = obd_disconnect(exp, 0);
653 lustre_cfg_bufs_reset(&bufs, name);
654 lcfg = lustre_cfg_new(LCFG_CLEANUP, &bufs);
655 err = class_process_config(lcfg);
656 lustre_cfg_free(lcfg);
660 lustre_cfg_bufs_reset(&bufs, name);
661 lcfg = lustre_cfg_new(LCFG_DETACH, &bufs);
662 err = class_process_config(lcfg);
663 lustre_cfg_free(lcfg);
668 lustre_cfg_bufs_reset(&bufs, name);
669 lustre_cfg_bufs_set_string(&bufs, 1, peer);
670 lcfg = lustre_cfg_new(LCFG_DEL_UUID, &bufs);
671 err = class_process_config(lcfg);
672 lustre_cfg_free(lcfg);
675 if (lmd->lmd_nal == SOCKNAL ||
676 lmd->lmd_nal == OPENIBNAL ||
677 lmd->lmd_nal == IIBNAL ||
678 lmd->lmd_nal == VIBNAL ||
679 lmd->lmd_nal == RANAL) {
682 PCFG_INIT(pcfg, NAL_CMD_DEL_PEER);
683 pcfg.pcfg_nal = lmd->lmd_nal;
684 pcfg.pcfg_nid = lmd->lmd_server_nid;
685 pcfg.pcfg_flags = 1; /* single_share */
686 err2 = libcfs_nal_cmd(&pcfg);
699 static void lustre_manual_cleanup(struct ll_sb_info *sbi)
701 struct lustre_cfg *lcfg;
702 struct lustre_cfg_bufs bufs;
703 struct obd_device *obd;
706 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) != NULL)
710 lustre_cfg_bufs_reset(&bufs, obd->obd_name);
711 lcfg = lustre_cfg_new(LCFG_CLEANUP, &bufs);
712 err = class_process_config(lcfg);
714 CERROR("cleanup failed: %s\n", obd->obd_name);
718 lcfg->lcfg_command = LCFG_DETACH;
719 err = class_process_config(lcfg);
720 lustre_cfg_free(lcfg);
722 CERROR("detach failed: %s\n", obd->obd_name);
727 if (sbi->ll_lmd != NULL)
728 class_del_profile(sbi->ll_lmd->lmd_profile);
731 int lustre_fill_super(struct super_block *sb, void *data, int silent)
733 struct lustre_mount_data * lmd = data;
734 char *lov = NULL, *lmv = NULL;
735 struct ll_sb_info *sbi;
739 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
740 if (lmd_bad_magic(lmd))
743 sbi = lustre_init_sbi(sb);
747 sbi->ll_flags |= LL_SBI_READAHEAD;
749 if (lmd->lmd_profile) {
750 struct lustre_profile *lprof;
751 struct config_llog_instance cfg;
754 if (lmd->lmd_mds[0] == '\0') {
755 CERROR("no mds name\n");
756 GOTO(out_free, err = -EINVAL);
758 lmd->lmd_mds_security[sizeof(lmd->lmd_mds_security) - 1] = 0;
759 lmd->lmd_oss_security[sizeof(lmd->lmd_oss_security) - 1] = 0;
761 OBD_ALLOC(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
762 if (sbi->ll_lmd == NULL)
763 GOTO(out_free, err = -ENOMEM);
764 memcpy(sbi->ll_lmd, lmd, sizeof(*lmd));
766 /* generate a string unique to this super, let's try
767 the address of the super itself.*/
768 len = (sizeof(sb) * 2) + 1;
769 OBD_ALLOC(sbi->ll_instance, len);
770 if (sbi->ll_instance == NULL)
771 GOTO(out_free, err = -ENOMEM);
772 sprintf(sbi->ll_instance, "%p", sb);
774 cfg.cfg_instance = sbi->ll_instance;
775 cfg.cfg_uuid = sbi->ll_sb_uuid;
776 cfg.cfg_local_nid = lmd->lmd_local_nid;
777 err = lustre_process_log(lmd, lmd->lmd_profile, &cfg, 0);
779 CERROR("Unable to process log: %s\n", lmd->lmd_profile);
783 lprof = class_get_profile(lmd->lmd_profile);
785 CERROR("No profile found: %s\n", lmd->lmd_profile);
786 GOTO(out_free, err = -EINVAL);
789 OBD_FREE(lov, strlen(lov) + 1);
790 OBD_ALLOC(lov, strlen(lprof->lp_lov) +
791 strlen(sbi->ll_instance) + 2);
792 sprintf(lov, "%s-%s", lprof->lp_lov, sbi->ll_instance);
795 OBD_FREE(lmv, strlen(lmv) + 1);
796 OBD_ALLOC(lmv, strlen(lprof->lp_lmv) +
797 strlen(sbi->ll_instance) + 2);
798 sprintf(lmv, "%s-%s", lprof->lp_lmv, sbi->ll_instance);
803 GOTO(out_free, err = -EINVAL);
808 GOTO(out_free, err = -EINVAL);
811 err = lustre_common_fill_super(sb, lmv, lov, lmd->lmd_async,
812 lmd->lmd_mds_security,
813 lmd->lmd_oss_security,
814 &lmd->lmd_nllu, &lmd->lmd_remote_flag);
821 OBD_FREE(lmv, strlen(lmv) + 1);
823 OBD_FREE(lov, strlen(lov) + 1);
829 int len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean")+1;
832 if (sbi->ll_instance != NULL) {
833 struct lustre_mount_data *lmd = sbi->ll_lmd;
834 struct config_llog_instance cfg;
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", lmd->lmd_profile);
842 err = lustre_process_log(lmd, cl_prof, &cfg, 0);
844 CERROR("Unable to process log: %s\n", cl_prof);
845 lustre_manual_cleanup(sbi);
847 OBD_FREE(cl_prof, len);
848 OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance) + 1);
850 OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
854 } /* lustre_fill_super */
856 void lustre_put_super(struct super_block *sb)
858 struct obd_device *obd;
859 struct ll_sb_info *sbi = ll_s2sbi(sb);
860 int force_umount = 0;
863 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
864 obd = class_exp2obd(sbi->ll_md_exp);
866 force_umount = obd->obd_no_recov;
869 lustre_common_put_super(sb);
870 if (sbi->ll_lmd != NULL) {
872 int len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean")+1;
874 struct config_llog_instance cfg;
877 CERROR("force umount, doing manual cleanup\n");
878 lustre_manual_cleanup(sbi);
882 cfg.cfg_instance = sbi->ll_instance;
883 cfg.cfg_uuid = sbi->ll_sb_uuid;
885 OBD_ALLOC(cl_prof, len);
886 sprintf(cl_prof, "%s-clean", sbi->ll_lmd->lmd_profile);
887 err = lustre_process_log(sbi->ll_lmd, cl_prof, &cfg, 0);
889 CERROR("Unable to process log: %s, doing manual cleanup"
891 lustre_manual_cleanup(sbi);
894 OBD_FREE(cl_prof, len);
896 OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
897 OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance) + 1);
903 } /* lustre_put_super */
905 int ll_process_config_update(struct ll_sb_info *sbi, int clean)
907 struct lustre_mount_data *lmd = sbi->ll_lmd;
908 char *profile = lmd->lmd_profile, *name = NULL;
909 struct config_llog_instance cfg;
910 int rc, namelen = 0, version;
911 struct llog_ctxt *ctxt;
917 CERROR("Client not mounted with zero-conf; cannot "
918 "process update log.\n");
922 rc = obd_cancel_unused(sbi->ll_md_exp, NULL,
923 LDLM_FL_CONFIG_CHANGE, NULL);
925 CWARN("obd_cancel_unused(mdc): %d\n", rc);
927 rc = obd_cancel_unused(sbi->ll_dt_exp, NULL,
928 LDLM_FL_CONFIG_CHANGE, NULL);
930 CWARN("obd_cancel_unused(lov): %d\n", rc);
932 cfg.cfg_instance = sbi->ll_instance;
933 cfg.cfg_uuid = sbi->ll_sb_uuid;
934 cfg.cfg_local_nid = lmd->lmd_local_nid;
936 namelen = strlen(profile) + 20; /* -clean-######### */
937 OBD_ALLOC(name, namelen);
942 version = sbi->ll_config_version - 1;
943 sprintf(name, "%s-clean-%d", profile, version);
945 version = sbi->ll_config_version + 1;
946 sprintf(name, "%s-%d", profile, version);
949 CWARN("Applying configuration log %s\n", name);
951 ctxt = llog_get_context(&sbi->ll_md_exp->exp_obd->obd_llogs,
952 LLOG_CONFIG_REPL_CTXT);
953 rc = class_config_process_llog(ctxt, name, &cfg);
955 sbi->ll_config_version = version;
956 CWARN("Finished applying configuration log %s: %d\n", name, rc);
958 if (rc == 0 && clean == 0) {
959 struct lov_desc desc;
963 valsize = sizeof(desc);
964 rc = obd_get_info(sbi->ll_dt_exp, strlen("lovdesc") + 1,
965 "lovdesc", &valsize, &desc);
967 rc = obd_init_ea_size(sbi->ll_md_exp,
968 obd_size_diskmd(sbi->ll_dt_exp, NULL),
970 sizeof(struct llog_cookie)));
972 OBD_FREE(name, namelen);
976 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
978 struct inode *inode = NULL;
979 l_lock(&lock->l_resource->lr_namespace->ns_lock);
980 if (lock->l_ast_data) {
981 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
982 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
983 inode = igrab(lock->l_ast_data);
985 inode = lock->l_ast_data;
986 CDEBUG(inode->i_state & I_FREEING ? D_INFO : D_WARNING,
987 "l_ast_data %p is bogus: magic %0x8\n",
988 lock->l_ast_data, lli->lli_inode_magic);
992 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
996 int null_if_equal(struct ldlm_lock *lock, void *data)
998 if (data == lock->l_ast_data) {
999 lock->l_ast_data = NULL;
1001 if (lock->l_req_mode != lock->l_granted_mode)
1002 LDLM_ERROR(lock,"clearing inode with ungranted lock\n");
1005 return LDLM_ITER_CONTINUE;
1008 void ll_clear_inode(struct inode *inode)
1010 struct lustre_id id;
1011 struct ll_inode_info *lli = ll_i2info(inode);
1012 struct ll_sb_info *sbi = ll_i2sbi(inode);
1015 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1016 inode->i_generation, inode);
1018 lli->lli_inode_magic = LLI_INODE_DEAD;
1019 ll_inode2id(&id, inode);
1021 clear_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &(ll_i2info(inode)->lli_flags));
1022 md_change_cbdata(sbi->ll_md_exp, &id, null_if_equal, inode);
1024 LASSERT(!lli->lli_open_fd_write_count);
1025 LASSERT(!lli->lli_open_fd_read_count);
1026 LASSERT(!lli->lli_open_fd_exec_count);
1027 if (lli->lli_mds_write_och)
1028 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_WRITE);
1029 if (lli->lli_mds_exec_och)
1030 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_EXEC);
1031 if (lli->lli_mds_read_och)
1032 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_READ);
1034 obd_change_cbdata(sbi->ll_dt_exp, lli->lli_smd,
1035 null_if_equal, inode);
1038 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1039 lli->lli_smd = NULL;
1043 obd_free_memmd(sbi->ll_md_exp,
1044 (struct lov_stripe_md **) &lli->lli_mea);
1045 lli->lli_mea = NULL;
1048 if (lli->lli_symlink_name) {
1049 OBD_FREE(lli->lli_symlink_name,
1050 strlen(lli->lli_symlink_name) + 1);
1051 lli->lli_symlink_name = NULL;
1053 lli->lli_inode_magic = LLI_INODE_DEAD;
1058 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1059 * object(s) determine the file size and mtime. Otherwise, the MDS will
1060 * keep these values until such a time that objects are allocated for it.
1061 * We do the MDS operations first, as it is checking permissions for us.
1062 * We don't to the MDS RPC if there is nothing that we want to store there,
1063 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1064 * going to do an RPC anyways.
1066 * If we are doing a truncate, we will send the mtime and ctime updates
1067 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1068 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1071 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1073 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1074 struct ll_sb_info *sbi = ll_i2sbi(inode);
1075 struct ptlrpc_request *request = NULL;
1076 struct mdc_op_data *op_data;
1077 int ia_valid = attr->ia_valid;
1081 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu\n", inode->i_ino);
1082 lprocfs_counter_incr(ll_i2sbi(inode)->ll_stats, LPROC_LL_SETATTR);
1084 if (ia_valid & ATTR_SIZE) {
1085 if (attr->ia_size > ll_file_maxbytes(inode)) {
1086 CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1087 attr->ia_size, ll_file_maxbytes(inode));
1091 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1094 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1095 if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1096 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1100 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1101 if (attr->ia_valid & ATTR_CTIME) {
1102 attr->ia_ctime = CURRENT_TIME;
1103 attr->ia_valid |= ATTR_CTIME_SET;
1105 if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1106 attr->ia_atime = CURRENT_TIME;
1107 attr->ia_valid |= ATTR_ATIME_SET;
1109 if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1110 attr->ia_mtime = CURRENT_TIME;
1111 attr->ia_valid |= ATTR_MTIME_SET;
1114 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1115 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1116 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1117 LTIME_S(CURRENT_TIME));
1120 attr->ia_valid &= ~ATTR_SIZE;
1122 /* If only OST attributes being set on objects, don't do MDS RPC.
1123 * In that case, we need to check permissions and update the local
1124 * inode ourselves so we can call obdo_from_inode() always. */
1125 if (ia_valid & (lsm ? ~(ATTR_SIZE | ATTR_FROM_OPEN /*| ATTR_RAW*/) : ~0)) {
1126 struct lustre_md md;
1128 OBD_ALLOC(op_data, sizeof(*op_data));
1129 if (op_data == NULL)
1131 ll_prepare_mdc_data(op_data, inode, NULL, NULL, 0, 0);
1133 rc = md_setattr(sbi->ll_md_exp, op_data,
1134 attr, NULL, 0, NULL, 0, &request);
1135 OBD_FREE(op_data, sizeof(*op_data));
1137 ptlrpc_req_finished(request);
1138 if (rc != -EPERM && rc != -EACCES)
1139 CERROR("md_setattr fails: rc = %d\n", rc);
1143 rc = mdc_req2lustre_md(sbi->ll_md_exp, request, 0,
1144 sbi->ll_dt_exp, &md);
1146 ptlrpc_req_finished(request);
1150 /* We call inode_setattr to adjust timestamps, but we first
1151 * clear ATTR_SIZE to avoid invoking vmtruncate.
1153 * NB: ATTR_SIZE will only be set at this point if the size
1154 * resides on the MDS, ie, this file has no objects. */
1155 attr->ia_valid &= ~ATTR_SIZE;
1158 * assigning inode_setattr() to @err to disable warning that
1159 * function's result should be checked by by caller. error is
1160 * impossible here, as vmtruncate() control path is disabled.
1162 err = inode_setattr(inode, attr);
1163 ll_update_inode(inode, &md);
1164 ptlrpc_req_finished(request);
1166 if (!lsm || !S_ISREG(inode->i_mode)) {
1167 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1171 /* The OST doesn't check permissions, but the alternative is
1172 * a gratuitous RPC to the MDS. We already rely on the client
1173 * to do read/write/truncate permission checks, so is mtime OK?
1175 if (ia_valid & (ATTR_MTIME | ATTR_ATIME)) {
1176 /* from sys_utime() */
1177 if (!(ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET))) {
1178 if (current->fsuid != inode->i_uid &&
1179 (rc = ll_permission(inode, MAY_WRITE, NULL)) != 0)
1182 /* from inode_change_ok() */
1183 if (current->fsuid != inode->i_uid &&
1184 !capable(CAP_FOWNER))
1189 /* won't invoke vmtruncate, as we already cleared ATTR_SIZE */
1190 err = inode_setattr(inode, attr);
1192 * assigning inode_setattr() to @err to disable warning that
1193 * function's result should be checked by by caller. error is
1194 * impossible here, as vmtruncate() control path is disabled.
1198 /* We really need to get our PW lock before we change inode->i_size.
1199 * If we don't we can race with other i_size updaters on our node, like
1200 * ll_file_read. We can also race with i_size propogation to other
1201 * nodes through dirtying and writeback of final cached pages. This
1202 * last one is especially bad for racing o_append users on other
1204 if (ia_valid & ATTR_SIZE) {
1205 ldlm_policy_data_t policy = { .l_extent = {attr->ia_size,
1207 struct lustre_handle lockh = { 0 };
1208 struct ll_inode_info *lli = ll_i2info(inode);
1209 int err, ast_flags = 0;
1210 /* XXX when we fix the AST intents to pass the discard-range
1211 * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
1213 if (attr->ia_size == 0)
1214 ast_flags = LDLM_AST_DISCARD_DATA;
1216 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy, &lockh,
1217 ast_flags, &ll_i2sbi(inode)->ll_seek_stime);
1222 down(&lli->lli_size_sem);
1223 lli->lli_size_pid = current->pid;
1224 rc = vmtruncate(inode, attr->ia_size);
1226 LASSERT(atomic_read(&lli->lli_size_sem.count) <= 0);
1227 lli->lli_size_pid = 0;
1228 up(&lli->lli_size_sem);
1231 err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
1233 CERROR("ll_extent_unlock failed: %d\n", err);
1237 } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET | ATTR_UID | ATTR_GID)) {
1238 struct obdo *oa = NULL;
1240 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1241 inode->i_ino, LTIME_S(attr->ia_mtime));
1247 oa->o_id = lsm->lsm_object_id;
1248 oa->o_gr = lsm->lsm_object_gr;
1249 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1251 if (ia_valid & ATTR_UID) {
1252 oa->o_uid = inode->i_uid;
1253 oa->o_valid |= OBD_MD_FLUID;
1256 if (ia_valid & ATTR_GID) {
1257 oa->o_gid = inode->i_gid;
1258 oa->o_valid |= OBD_MD_FLGID;
1261 obdo_from_inode(oa, inode, OBD_MD_FLTYPE | OBD_MD_FLATIME |
1262 OBD_MD_FLMTIME | OBD_MD_FLCTIME);
1263 rc = obd_setattr(sbi->ll_dt_exp, oa, lsm, NULL);
1266 CERROR("obd_setattr fails: rc = %d\n", rc);
1272 int ll_setattr(struct dentry *de, struct iattr *attr)
1274 LASSERT(de->d_inode);
1275 return ll_setattr_raw(de->d_inode, attr);
1278 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1279 unsigned long max_age)
1281 struct ll_sb_info *sbi = ll_s2sbi(sb);
1282 struct obd_statfs obd_osfs;
1286 rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age);
1288 CERROR("obd_statfs fails: rc = %d\n", rc);
1292 osfs->os_type = sb->s_magic;
1294 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1295 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1297 rc = obd_statfs(class_exp2obd(sbi->ll_dt_exp), &obd_osfs, max_age);
1299 CERROR("obd_statfs fails: rc = %d\n", rc);
1303 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1304 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1307 osfs->os_blocks = obd_osfs.os_blocks;
1308 osfs->os_bfree = obd_osfs.os_bfree;
1309 osfs->os_bavail = obd_osfs.os_bavail;
1311 /* If we don't have as many objects free on the OST as inodes
1312 * on the MDS, we reduce the total number of inodes to
1313 * compensate, so that the "inodes in use" number is correct.
1315 if (obd_osfs.os_ffree < osfs->os_ffree) {
1316 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1318 osfs->os_ffree = obd_osfs.os_ffree;
1324 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1326 struct obd_statfs osfs;
1329 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p\n", sb);
1330 lprocfs_counter_incr(ll_s2sbi(sb)->ll_stats, LPROC_LL_STAFS);
1332 /* For now we will always get up-to-date statfs values, but in the
1333 * future we may allow some amount of caching on the client (e.g.
1334 * from QOS or lprocfs updates). */
1335 rc = ll_statfs_internal(sb, &osfs, jiffies - 1);
1339 statfs_unpack(sfs, &osfs);
1341 if (sizeof(sfs->f_blocks) == 4) {
1342 while (osfs.os_blocks > ~0UL) {
1345 osfs.os_blocks >>= 1;
1346 osfs.os_bfree >>= 1;
1347 osfs.os_bavail >>= 1;
1351 sfs->f_blocks = osfs.os_blocks;
1352 sfs->f_bfree = osfs.os_bfree;
1353 sfs->f_bavail = osfs.os_bavail;
1358 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1360 struct ll_inode_info *lli = ll_i2info(inode);
1361 struct lov_stripe_md *lsm = md->lsm;
1362 struct mds_body *body = md->body;
1363 struct mea *mea = md->mea;
1364 struct posix_acl *ll_acl_access = md->acl_access;
1367 LASSERT((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1369 if (md->lsm && md->lsm->lsm_magic != LOV_MAGIC) {
1370 /* check for default striping info for dir. */
1371 LASSERT((mea != NULL) == ((body->valid & OBD_MD_FLDIREA) != 0));
1375 LASSERT(lsm->lsm_object_gr > 0);
1376 if (lli->lli_smd == NULL) {
1378 lli->lli_maxbytes = lsm->lsm_maxbytes;
1379 if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1380 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1383 if (memcmp(lli->lli_smd, lsm, sizeof(*lsm))) {
1384 CERROR("lsm mismatch for inode %ld\n",
1386 CERROR("lli_smd:\n");
1387 dump_lsm(D_ERROR, lli->lli_smd);
1389 dump_lsm(D_ERROR, lsm);
1392 /* XXX FIXME -- We should decide on a safer (atomic) and
1393 * more elegant way to update the lsm */
1394 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1395 lli->lli_smd->lsm_oinfo[i].loi_id =
1396 lsm->lsm_oinfo[i].loi_id;
1397 lli->lli_smd->lsm_oinfo[i].loi_gr =
1398 lsm->lsm_oinfo[i].loi_gr;
1399 lli->lli_smd->lsm_oinfo[i].loi_ost_idx =
1400 lsm->lsm_oinfo[i].loi_ost_idx;
1401 lli->lli_smd->lsm_oinfo[i].loi_ost_gen =
1402 lsm->lsm_oinfo[i].loi_ost_gen;
1405 /* bug 2844 - limit i_blksize for broken user-space apps */
1406 LASSERTF(lsm->lsm_xfersize != 0, "%lu\n", lsm->lsm_xfersize);
1407 inode->i_blksize = min(lsm->lsm_xfersize, LL_MAX_BLKSIZE);
1408 if (lli->lli_smd != lsm)
1409 obd_free_memmd(ll_i2dtexp(inode), &lsm);
1413 if (lli->lli_mea == NULL) {
1416 if (memcmp(lli->lli_mea, mea, body->eadatasize)) {
1417 CERROR("mea mismatch for inode %lu\n",
1422 if (lli->lli_mea != mea)
1423 obd_free_memmd(ll_i2mdexp(inode),
1424 (struct lov_stripe_md **) &mea);
1427 if (body->valid & OBD_MD_FID)
1428 id_assign_fid(&lli->lli_id, &body->id1);
1430 if (body->valid & OBD_MD_FLID)
1431 id_ino(&lli->lli_id) = id_ino(&body->id1);
1433 if (body->valid & OBD_MD_FLGENER)
1434 id_gen(&lli->lli_id) = id_gen(&body->id1);
1436 spin_lock(&lli->lli_lock);
1437 if (ll_acl_access != NULL) {
1438 if (lli->lli_acl_access != NULL)
1439 posix_acl_release(lli->lli_acl_access);
1440 lli->lli_acl_access = ll_acl_access;
1442 spin_unlock(&lli->lli_lock);
1444 if (body->valid & OBD_MD_FLID)
1445 inode->i_ino = id_ino(&body->id1);
1446 if (body->valid & OBD_MD_FLGENER)
1447 inode->i_generation = id_gen(&body->id1);
1448 if (body->valid & OBD_MD_FLATIME)
1449 LTIME_S(inode->i_atime) = body->atime;
1450 if (body->valid & OBD_MD_FLMTIME &&
1451 body->mtime > LTIME_S(inode->i_mtime)) {
1452 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to %u\n",
1453 inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
1454 LTIME_S(inode->i_mtime) = body->mtime;
1456 if (body->valid & OBD_MD_FLCTIME &&
1457 body->ctime > LTIME_S(inode->i_ctime))
1458 LTIME_S(inode->i_ctime) = body->ctime;
1459 if (body->valid & OBD_MD_FLMODE) {
1460 inode->i_mode = (inode->i_mode & S_IFMT) |
1461 (body->mode & ~S_IFMT);
1463 if (body->valid & OBD_MD_FLTYPE) {
1464 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1465 (body->mode & S_IFMT);
1467 if (body->valid & OBD_MD_FLUID)
1468 inode->i_uid = body->uid;
1469 if (body->valid & OBD_MD_FLGID)
1470 inode->i_gid = body->gid;
1471 if (body->valid & OBD_MD_FLFLAGS)
1472 inode->i_flags = body->flags;
1473 if (body->valid & OBD_MD_FLNLINK)
1474 inode->i_nlink = body->nlink;
1475 if (body->valid & OBD_MD_FLRDEV)
1476 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1477 inode->i_rdev = body->rdev;
1479 inode->i_rdev = old_decode_dev(body->rdev);
1481 if (body->valid & OBD_MD_FLSIZE)
1482 inode->i_size = body->size;
1483 if (body->valid & OBD_MD_FLBLOCKS)
1484 inode->i_blocks = body->blocks;
1486 if (body->valid & OBD_MD_FLSIZE)
1487 set_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &lli->lli_flags);
1489 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1490 inode->i_dev = (kdev_t)id_group(&lli->lli_id);
1492 LASSERT(id_fid(&lli->lli_id) != 0);
1495 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
1496 static struct backing_dev_info ll_backing_dev_info = {
1497 .ra_pages = 0, /* No readahead */
1498 .memory_backed = 0, /* Does contribute to dirty memory */
1502 void ll_read_inode2(struct inode *inode, void *opaque)
1504 struct lustre_md *md = opaque;
1505 struct ll_inode_info *lli = ll_i2info(inode);
1508 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1509 inode->i_generation, inode);
1513 LASSERT(!lli->lli_smd);
1515 /* Core attributes from the MDS first. This is a new inode, and
1516 * the VFS doesn't zero times in the core inode so we have to do
1517 * it ourselves. They will be overwritten by either MDS or OST
1518 * attributes - we just need to make sure they aren't newer. */
1519 LTIME_S(inode->i_mtime) = 0;
1520 LTIME_S(inode->i_atime) = 0;
1521 LTIME_S(inode->i_ctime) = 0;
1524 ll_update_inode(inode, md);
1526 /* OIDEBUG(inode); */
1528 if (S_ISREG(inode->i_mode)) {
1529 inode->i_op = &ll_file_inode_operations;
1530 inode->i_fop = &ll_file_operations;
1531 inode->i_mapping->a_ops = &ll_aops;
1533 } else if (S_ISDIR(inode->i_mode)) {
1534 inode->i_op = &ll_dir_inode_operations;
1535 inode->i_fop = &ll_dir_operations;
1536 inode->i_mapping->a_ops = &ll_dir_aops;
1538 } else if (S_ISLNK(inode->i_mode)) {
1539 inode->i_op = &ll_fast_symlink_inode_operations;
1542 inode->i_op = &ll_special_inode_operations;
1544 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
1545 init_special_inode(inode, inode->i_mode,
1546 kdev_t_to_nr(inode->i_rdev));
1548 /* initializing backing dev info. */
1549 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
1551 init_special_inode(inode, inode->i_mode, inode->i_rdev);
1553 lli->ll_save_ifop = inode->i_fop;
1555 if (S_ISCHR(inode->i_mode))
1556 inode->i_fop = &ll_special_chr_inode_fops;
1557 else if (S_ISBLK(inode->i_mode))
1558 inode->i_fop = &ll_special_blk_inode_fops;
1559 else if (S_ISFIFO(inode->i_mode))
1560 inode->i_fop = &ll_special_fifo_inode_fops;
1561 else if (S_ISSOCK(inode->i_mode))
1562 inode->i_fop = &ll_special_sock_inode_fops;
1564 CWARN("saved %p, replaced with %p\n", lli->ll_save_ifop,
1567 if (lli->ll_save_ifop->owner) {
1568 CWARN("%p has owner %p\n", lli->ll_save_ifop,
1569 lli->ll_save_ifop->owner);
1575 void ll_delete_inode(struct inode *inode)
1577 struct ll_sb_info *sbi = ll_i2sbi(inode);
1578 struct lustre_id id;
1582 ll_inode2id(&id, inode);
1584 rc = md_delete_inode(sbi->ll_md_exp, &id);
1586 CERROR("md_delete_inode() failed, error %d\n",
1594 int ll_iocontrol(struct inode *inode, struct file *file,
1595 unsigned int cmd, unsigned long arg)
1597 struct ll_sb_info *sbi = ll_i2sbi(inode);
1598 struct ptlrpc_request *req = NULL;
1603 case EXT3_IOC_GETFLAGS: {
1604 struct lustre_id id;
1605 __u64 valid = OBD_MD_FLFLAGS;
1606 struct mds_body *body;
1608 ll_inode2id(&id, inode);
1609 rc = md_getattr(sbi->ll_md_exp, &id, valid, NULL, 0, 0, &req);
1611 CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1615 body = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*body));
1617 if (body->flags & S_APPEND)
1618 flags |= EXT3_APPEND_FL;
1619 if (body->flags & S_IMMUTABLE)
1620 flags |= EXT3_IMMUTABLE_FL;
1621 if (body->flags & S_NOATIME)
1622 flags |= EXT3_NOATIME_FL;
1624 ptlrpc_req_finished (req);
1626 RETURN(put_user(flags, (int *)arg));
1628 case EXT3_IOC_SETFLAGS: {
1629 struct mdc_op_data *op_data;
1632 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1634 if (get_user(flags, (int *)arg))
1641 OBD_ALLOC(op_data, sizeof(*op_data));
1642 if (op_data == NULL) {
1646 ll_prepare_mdc_data(op_data, inode, NULL, NULL, 0, 0);
1648 memset(&attr, 0x0, sizeof(attr));
1649 attr.ia_attr_flags = flags;
1650 attr.ia_valid |= ATTR_ATTR_FLAG;
1652 rc = md_setattr(sbi->ll_md_exp, op_data,
1653 &attr, NULL, 0, NULL, 0, &req);
1654 OBD_FREE(op_data, sizeof(*op_data));
1656 ptlrpc_req_finished(req);
1657 if (rc != -EPERM && rc != -EACCES)
1658 CERROR("md_setattr fails: rc = %d\n", rc);
1662 ptlrpc_req_finished(req);
1664 oa->o_id = lsm->lsm_object_id;
1665 oa->o_gr = lsm->lsm_object_gr;
1666 oa->o_flags = flags;
1667 oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS | OBD_MD_FLGROUP;
1669 rc = obd_setattr(sbi->ll_dt_exp, oa, lsm, NULL);
1672 if (rc != -EPERM && rc != -EACCES)
1673 CERROR("md_setattr fails: rc = %d\n", rc);
1677 if (flags & EXT3_APPEND_FL)
1678 inode->i_flags |= S_APPEND;
1680 inode->i_flags &= ~S_APPEND;
1681 if (flags & EXT3_IMMUTABLE_FL)
1682 inode->i_flags |= S_IMMUTABLE;
1684 inode->i_flags &= ~S_IMMUTABLE;
1685 if (flags & EXT3_NOATIME_FL)
1686 inode->i_flags |= S_NOATIME;
1688 inode->i_flags &= ~S_NOATIME;
1699 /* this is only called in the case of forced umount. */
1700 void ll_umount_begin(struct super_block *sb)
1702 struct ll_sb_info *sbi = ll_s2sbi(sb);
1703 struct obd_ioctl_data ioc_data = { 0 };
1704 struct obd_device *obd;
1707 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1708 sb->s_count, atomic_read(&sb->s_active));
1710 obd = class_exp2obd(sbi->ll_md_exp);
1712 CERROR("Invalid MDC connection handle "LPX64"\n",
1713 sbi->ll_md_exp->exp_handle.h_cookie);
1717 obd->obd_no_recov = 1;
1718 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
1719 sizeof(ioc_data), &ioc_data, NULL);
1721 obd = class_exp2obd(sbi->ll_dt_exp);
1723 CERROR("Invalid LOV connection handle "LPX64"\n",
1724 sbi->ll_dt_exp->exp_handle.h_cookie);
1729 obd->obd_no_recov = 1;
1730 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
1731 sizeof(ioc_data), &ioc_data, NULL);
1734 * really, we'd like to wait until there are no requests outstanding,
1735 * and then continue. For now, we just invalidate the requests,
1736 * schedule, and hope.
1743 int ll_prep_inode(struct obd_export *dt_exp, struct obd_export *md_exp,
1744 struct inode **inode, struct ptlrpc_request *req,
1745 int offset, struct super_block *sb)
1747 struct lustre_md md;
1750 rc = mdc_req2lustre_md(md_exp, req, offset, dt_exp, &md);
1755 ll_update_inode(*inode, &md);
1758 *inode = ll_iget(sb, id_ino(&md.body->id1), &md);
1759 if (*inode == NULL || is_bad_inode(*inode)) {
1760 /* free the lsm if we allocated one above */
1762 obd_free_memmd(dt_exp, &md.lsm);
1764 obd_free_memmd(md_exp,
1765 (struct lov_stripe_md**)&md.mea);
1767 CERROR("new_inode -fatal: rc %d\n", rc);
1774 int ll_get_fid(struct obd_export *exp, struct lustre_id *idp,
1775 char *filename, struct lustre_id *ret)
1777 struct ptlrpc_request *request = NULL;
1778 struct mds_body *body;
1781 rc = md_getattr_lock(exp, idp, filename, strlen(filename) + 1,
1782 OBD_MD_FID, 0, &request);
1784 CDEBUG(D_INFO, "md_getattr_lock failed on %s: rc %d\n",
1789 body = lustre_msg_buf(request->rq_repmsg, 0, sizeof(*body));
1790 LASSERT(body != NULL);
1791 LASSERT_REPSWABBED(request, 0);
1794 ptlrpc_req_finished(request);