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);
74 /* this later may be reset via /proc/fs/... */
75 memcpy(sbi->ll_gns_oname, ".mntinfo", strlen(".mntinfo"));
76 sbi->ll_gns_oname[strlen(sbi->ll_gns_oname)] = '\0';
78 /* this later may be reset via /proc/fs/... */
79 memcpy(sbi->ll_gns_upcall, "/usr/sbin/gns_upcall",
80 strlen("/usr/sbin/gns_upcall"));
81 sbi->ll_gns_upcall[strlen(sbi->ll_gns_upcall)] = '\0';
83 /* default values, may be changed via /proc/fs/... */
84 sbi->ll_gns_state = LL_GNS_IDLE;
85 sbi->ll_gns_tick = GNS_TICK_TIMEOUT;
86 sbi->ll_gns_timeout = GNS_MOUNT_TIMEOUT;
88 sbi->ll_gns_timer.data = (unsigned long)sbi;
89 sbi->ll_gns_timer.function = ll_gns_timer_callback;
90 init_timer(&sbi->ll_gns_timer);
94 generate_random_uuid(uuid);
95 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
99 void lustre_free_sbi(struct super_block *sb)
101 struct ll_sb_info *sbi = ll_s2sbi(sb);
105 list_del(&sbi->ll_gns_sbi_head);
106 del_timer(&sbi->ll_gns_timer);
107 OBD_FREE(sbi, sizeof(*sbi));
109 ll_set_sbi(sb, NULL);
113 int lustre_init_dt_desc(struct ll_sb_info *sbi)
118 valsize = sizeof(sbi->ll_dt_desc);
119 memset(&sbi->ll_dt_desc, 0, sizeof(sbi->ll_dt_desc));
120 rc = obd_get_info(sbi->ll_dt_exp, strlen("lovdesc") + 1,
121 "lovdesc", &valsize, &sbi->ll_dt_desc);
125 extern struct dentry_operations ll_d_ops;
127 int lustre_common_fill_super(struct super_block *sb, char *lmv, char *lov,
128 int async, char *security, __u32 *nllu,
131 struct ll_sb_info *sbi = ll_s2sbi(sb);
132 struct ptlrpc_request *request = NULL;
133 struct lustre_handle dt_conn = {0, };
134 struct lustre_handle md_conn = {0, };
135 struct obd_connect_data *data;
136 struct inode *root = NULL;
137 struct obd_device *obd;
138 struct obd_statfs osfs;
145 obd = class_name2obd(lmv);
147 CERROR("MDC %s: not setup or attached\n", lmv);
150 obd_set_info(obd->obd_self_export, strlen("async"), "async",
151 sizeof(async), &async);
153 if ((*remote & (OBD_CONNECT_LOCAL | OBD_CONNECT_REMOTE)) ==
154 (OBD_CONNECT_LOCAL | OBD_CONNECT_REMOTE)) {
155 CERROR("wrong remote flag "LPX64"\n", *remote);
159 OBD_ALLOC(data, sizeof(*data));
163 data->ocd_connect_flags |= *remote & (OBD_CONNECT_LOCAL |
165 memcpy(data->ocd_nllu, nllu, sizeof(data->ocd_nllu));
167 if (security == NULL)
170 err = obd_set_info(obd->obd_self_export, strlen("sec"), "sec",
171 strlen(security), security);
173 CERROR("LMV %s: failed to set security %s, err %d\n",
175 OBD_FREE(data, sizeof(*data));
179 if (proc_lustre_fs_root) {
180 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
183 CERROR("could not register mount in /proc/lustre");
186 err = obd_connect(&md_conn, obd, &sbi->ll_sb_uuid, data,
187 OBD_OPT_REAL_CLIENT);
189 CERROR("An MDS (lmv %s) is performing recovery, of which this"
190 " client is not a part. Please wait for recovery to "
191 "complete, abort, or time out.\n", lmv);
194 CERROR("cannot connect to %s: rc = %d\n", lmv, err);
197 sbi->ll_md_exp = class_conn2export(&md_conn);
198 err = obd_statfs(obd, &osfs, jiffies - HZ);
202 if (!osfs.os_bsize) {
203 CERROR("Invalid block size is detected.");
207 sb->s_magic = LL_SUPER_MAGIC;
208 sb->s_blocksize = osfs.os_bsize;
209 sb->s_blocksize_bits = log2(osfs.os_bsize);
210 sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
212 devno = get_uuid2int(sbi->ll_md_exp->exp_obd->obd_uuid.uuid,
213 strlen(sbi->ll_md_exp->exp_obd->obd_uuid.uuid));
217 /* after statfs, we are supposed to have connected to MDSs,
218 * so it's ok to check remote flag returned.
220 valsize = sizeof(&sbi->ll_remote);
221 err = obd_get_info(sbi->ll_md_exp, strlen("remote_flag"), "remote_flag",
222 &valsize, &sbi->ll_remote);
224 CERROR("fail to obtain remote flag\n");
228 obd = class_name2obd(lov);
230 CERROR("OSC %s: not setup or attached\n", lov);
233 obd_set_info(obd->obd_self_export, strlen("async"), "async",
234 sizeof(async), &async);
236 err = obd_connect(&dt_conn, obd, &sbi->ll_sb_uuid, data,
237 OBD_OPT_REAL_CLIENT);
239 CERROR("An OST (lov %s) is performing recovery, of which this"
240 " client is not a part. Please wait for recovery to "
241 "complete, abort, or time out.\n", lov);
244 CERROR("cannot connect to %s: rc = %d\n", lov, err);
247 sbi->ll_dt_exp = class_conn2export(&dt_conn);
249 err = lustre_init_dt_desc(sbi);
251 int mdsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
252 obd_init_ea_size(sbi->ll_md_exp, mdsize,
253 sbi->ll_dt_desc.ld_tgt_count *
254 sizeof(struct llog_cookie));
257 err = md_getstatus(sbi->ll_md_exp, &sbi->ll_rootid);
259 CERROR("cannot mds_connect: rc = %d\n", err);
262 CDEBUG(D_SUPER, "rootid "DLID4"\n", OLID4(&sbi->ll_rootid));
264 sb->s_op = &lustre_super_operations;
267 * XXX: move this to after cbd setup? */
268 err = md_getattr(sbi->ll_md_exp, &sbi->ll_rootid,
269 (OBD_MD_FLNOTOBD | OBD_MD_FLBLOCKS | OBD_MD_FID), NULL, 0,
272 CERROR("md_getattr failed for root: rc = %d\n", err);
276 err = mdc_req2lustre_md(sbi->ll_md_exp, request, 0,
277 sbi->ll_dt_exp, &md);
279 CERROR("failed to understand root inode md: rc = %d\n", err);
280 ptlrpc_req_finished(request);
284 LASSERT(id_ino(&sbi->ll_rootid) != 0);
285 root = ll_iget(sb, id_ino(&sbi->ll_rootid), &md);
287 ptlrpc_req_finished(request);
289 if (root == NULL || is_bad_inode(root)) {
290 CERROR("lustre_lite: bad iget4 for root\n");
291 GOTO(out_root, err = -EBADF);
294 err = ll_close_thread_start(&sbi->ll_lcq);
296 CERROR("cannot start close thread: rc %d\n", err);
300 ll_gns_add_timer(sbi);
302 /* making vm readahead 0 for 2.4.x. In the case of 2.6.x,
303 backing dev info assigned to inode mapping is used for
304 determining maximal readahead. */
305 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0)) && \
306 !defined(KERNEL_HAS_AS_MAX_READAHEAD)
307 /* bug 2805 - set VM readahead to zero */
308 vm_max_readahead = vm_min_readahead = 0;
311 sb->s_root = d_alloc_root(root);
312 sb->s_root->d_op = &ll_d_ops;
314 sb->s_flags |= MS_POSIXACL;
316 CWARN("Enabling PDIROPS\n");
317 sb->s_flags |= S_PDIROPS;
321 OBD_FREE(data, sizeof(*data));
327 obd_disconnect(sbi->ll_dt_exp, 0);
329 obd_disconnect(sbi->ll_md_exp, 0);
332 OBD_FREE(data, sizeof(*data));
333 lprocfs_unregister_mountpoint(sbi);
337 void lustre_common_put_super(struct super_block *sb)
339 struct ll_sb_info *sbi = ll_s2sbi(sb);
340 struct hlist_node *tmp, *next;
343 ll_gns_del_timer(sbi);
344 ll_close_thread_shutdown(sbi->ll_lcq);
346 list_del(&sbi->ll_conn_chain);
347 obd_disconnect(sbi->ll_dt_exp, 0);
349 lprocfs_unregister_mountpoint(sbi);
350 if (sbi->ll_proc_root) {
351 lprocfs_remove(sbi->ll_proc_root);
352 sbi->ll_proc_root = NULL;
355 obd_disconnect(sbi->ll_md_exp, 0);
357 // We do this to get rid of orphaned dentries. That is not really trw.
358 hlist_for_each_safe(tmp, next, &sbi->ll_orphan_dentry_list) {
359 struct dentry *dentry = hlist_entry(tmp, struct dentry, d_hash);
360 CWARN("orphan dentry %.*s (%p->%p) at unmount\n",
361 dentry->d_name.len, dentry->d_name.name, dentry, next);
362 shrink_dcache_parent(dentry);
367 char *ll_read_opt(const char *opt, char *data)
373 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
374 if (strncmp(opt, data, strlen(opt)))
376 if ((value = strchr(data, '=')) == NULL)
380 OBD_ALLOC(retval, strlen(value) + 1);
382 CERROR("out of memory!\n");
386 memcpy(retval, value, strlen(value)+1);
387 CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
391 int ll_set_opt(const char *opt, char *data, int fl)
395 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
396 if (strncmp(opt, data, strlen(opt)))
402 void ll_options(char *options, char **lov, char **lmv, char **sec,
403 int *async, int *flags)
406 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
407 char *opt_ptr = options;
417 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
418 for (this_char = strtok (options, ",");
420 this_char = strtok (NULL, ",")) {
422 while ((this_char = strsep (&opt_ptr, ",")) != NULL) {
424 CDEBUG(D_SUPER, "this_char %s\n", this_char);
425 if (!*lov && (*lov = ll_read_opt("osc", this_char)))
427 if (!*lmv && (*lmv = ll_read_opt("mdc", this_char)))
429 if (!strncmp(this_char, "lasync", strlen("lasync"))) {
433 if (!*sec && (*sec = ll_read_opt("sec", this_char)))
435 if (!(*flags & LL_SBI_NOLCK) &&
436 ((*flags) = (*flags) |
437 ll_set_opt("nolock", this_char,
445 void ll_lli_init(struct ll_inode_info *lli)
447 sema_init(&lli->lli_open_sem, 1);
448 sema_init(&lli->lli_size_sem, 1);
450 lli->lli_size_pid = 0;
451 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
452 spin_lock_init(&lli->lli_lock);
453 INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
454 lli->lli_inode_magic = LLI_INODE_MAGIC;
455 memset(&lli->lli_id, 0, sizeof(lli->lli_id));
456 sema_init(&lli->lli_och_sem, 1);
457 lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
458 lli->lli_mds_exec_och = NULL;
459 lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
460 lli->lli_open_fd_exec_count = 0;
463 int ll_fill_super(struct super_block *sb, void *data, int silent)
465 struct ll_sb_info *sbi;
470 __u32 nllu[2] = { NOBODY_UID, NOBODY_GID };
471 __u64 remote_flag = 0;
474 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
476 sbi = lustre_init_sbi(sb);
480 sbi->ll_flags |= LL_SBI_READAHEAD;
481 ll_options(data, &lov, &lmv, &sec, &async, &sbi->ll_flags);
485 GOTO(out, err = -EINVAL);
490 GOTO(out, err = -EINVAL);
493 err = lustre_common_fill_super(sb, lmv, lov, async, sec, nllu,
501 OBD_FREE(sec, strlen(sec) + 1);
503 OBD_FREE(lmv, strlen(lmv) + 1);
505 OBD_FREE(lov, strlen(lov) + 1);
507 } /* ll_read_super */
509 static int lustre_process_log(struct lustre_mount_data *lmd, char *profile,
510 struct config_llog_instance *cfg, int allow_recov)
512 struct lustre_cfg lcfg;
513 struct portals_cfg pcfg;
514 char *peer = "MDS_PEER_UUID";
515 struct obd_device *obd;
516 struct lustre_handle md_conn = {0, };
517 struct obd_export *exp;
518 char *name = "mdc_dev";
520 struct obd_uuid lmv_uuid;
521 struct llog_ctxt *ctxt;
525 if (lmd_bad_magic(lmd))
528 generate_random_uuid(uuid);
529 class_uuid_unparse(uuid, &lmv_uuid);
531 if (lmd->lmd_local_nid) {
532 PCFG_INIT(pcfg, NAL_CMD_REGISTER_MYNID);
533 pcfg.pcfg_nal = lmd->lmd_nal;
534 pcfg.pcfg_nid = lmd->lmd_local_nid;
535 rc = libcfs_nal_cmd(&pcfg);
540 if (lmd->lmd_nal == SOCKNAL ||
541 lmd->lmd_nal == OPENIBNAL ||
542 lmd->lmd_nal == IIBNAL ||
543 lmd->lmd_nal == VIBNAL ||
544 lmd->lmd_nal == RANAL) {
545 PCFG_INIT(pcfg, NAL_CMD_ADD_PEER);
546 pcfg.pcfg_nal = lmd->lmd_nal;
547 pcfg.pcfg_nid = lmd->lmd_server_nid;
548 pcfg.pcfg_id = lmd->lmd_server_ipaddr;
549 pcfg.pcfg_misc = lmd->lmd_port;
550 rc = libcfs_nal_cmd(&pcfg);
555 LCFG_INIT(lcfg, LCFG_ADD_UUID, name);
556 lcfg.lcfg_nid = lmd->lmd_server_nid;
557 lcfg.lcfg_inllen1 = strlen(peer) + 1;
558 lcfg.lcfg_inlbuf1 = peer;
559 lcfg.lcfg_nal = lmd->lmd_nal;
560 rc = class_process_config(&lcfg);
562 GOTO(out_del_conn, rc);
564 LCFG_INIT(lcfg, LCFG_ATTACH, name);
565 lcfg.lcfg_inlbuf1 = "mdc";
566 lcfg.lcfg_inllen1 = strlen(lcfg.lcfg_inlbuf1) + 1;
567 lcfg.lcfg_inlbuf2 = lmv_uuid.uuid;
568 lcfg.lcfg_inllen2 = strlen(lcfg.lcfg_inlbuf2) + 1;
569 err = class_process_config(&lcfg);
571 GOTO(out_del_uuid, rc);
573 LCFG_INIT(lcfg, LCFG_SETUP, name);
574 lcfg.lcfg_inlbuf1 = lmd->lmd_mds;
575 lcfg.lcfg_inllen1 = strlen(lcfg.lcfg_inlbuf1) + 1;
576 lcfg.lcfg_inlbuf2 = peer;
577 lcfg.lcfg_inllen2 = strlen(lcfg.lcfg_inlbuf2) + 1;
578 rc = class_process_config(&lcfg);
580 GOTO(out_detach, rc);
582 obd = class_name2obd(name);
584 GOTO(out_cleanup, rc = -EINVAL);
586 rc = obd_set_info(obd->obd_self_export, strlen("sec"), "sec",
587 strlen(lmd->lmd_security), lmd->lmd_security);
589 GOTO(out_cleanup, rc);
591 /* Disable initial recovery on this import */
592 rc = obd_set_info(obd->obd_self_export,
593 strlen("initial_recov"), "initial_recov",
594 sizeof(allow_recov), &allow_recov);
596 GOTO(out_cleanup, rc);
598 rc = obd_connect(&md_conn, obd, &lmv_uuid, NULL, 0);
600 CERROR("cannot connect to %s: rc = %d\n", lmd->lmd_mds, rc);
601 GOTO(out_cleanup, rc);
604 exp = class_conn2export(&md_conn);
606 ctxt = llog_get_context(&exp->exp_obd->obd_llogs,LLOG_CONFIG_REPL_CTXT);
607 rc = class_config_process_llog(ctxt, profile, cfg);
609 CERROR("class_config_process_llog failed: rc = %d\n", rc);
611 err = obd_disconnect(exp, 0);
615 LCFG_INIT(lcfg, LCFG_CLEANUP, name);
616 err = class_process_config(&lcfg);
621 LCFG_INIT(lcfg, LCFG_DETACH, name);
622 err = class_process_config(&lcfg);
627 LCFG_INIT(lcfg, LCFG_DEL_UUID, name);
628 lcfg.lcfg_inllen1 = strlen(peer) + 1;
629 lcfg.lcfg_inlbuf1 = peer;
630 err = class_process_config(&lcfg);
633 if (lmd->lmd_nal == SOCKNAL ||
634 lmd->lmd_nal == OPENIBNAL ||
635 lmd->lmd_nal == IIBNAL ||
636 lmd->lmd_nal == VIBNAL ||
637 lmd->lmd_nal == RANAL) {
640 PCFG_INIT(pcfg, NAL_CMD_DEL_PEER);
641 pcfg.pcfg_nal = lmd->lmd_nal;
642 pcfg.pcfg_nid = lmd->lmd_server_nid;
643 pcfg.pcfg_flags = 1; /* single_share */
644 err2 = libcfs_nal_cmd(&pcfg);
657 static void lustre_manual_cleanup(struct ll_sb_info *sbi)
659 struct lustre_cfg lcfg;
660 struct obd_device *obd;
663 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) != NULL)
667 LCFG_INIT(lcfg, LCFG_CLEANUP, obd->obd_name);
668 err = class_process_config(&lcfg);
670 CERROR("cleanup failed: %s\n", obd->obd_name);
674 LCFG_INIT(lcfg, LCFG_DETACH, obd->obd_name);
675 err = class_process_config(&lcfg);
677 CERROR("detach failed: %s\n", obd->obd_name);
682 if (sbi->ll_lmd != NULL)
683 class_del_profile(sbi->ll_lmd->lmd_profile);
686 int lustre_fill_super(struct super_block *sb, void *data, int silent)
688 struct lustre_mount_data * lmd = data;
689 char *lov = NULL, *lmv = NULL;
690 struct ll_sb_info *sbi;
694 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
695 if (lmd_bad_magic(lmd))
698 sbi = lustre_init_sbi(sb);
702 sbi->ll_flags |= LL_SBI_READAHEAD;
704 if (lmd->lmd_profile) {
705 struct lustre_profile *lprof;
706 struct config_llog_instance cfg;
709 if (lmd->lmd_mds[0] == '\0') {
710 CERROR("no mds name\n");
711 GOTO(out_free, err = -EINVAL);
713 lmd->lmd_security[sizeof(lmd->lmd_security) - 1] = 0;
715 OBD_ALLOC(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
716 if (sbi->ll_lmd == NULL)
717 GOTO(out_free, err = -ENOMEM);
718 memcpy(sbi->ll_lmd, lmd, sizeof(*lmd));
720 /* generate a string unique to this super, let's try
721 the address of the super itself.*/
722 len = (sizeof(sb) * 2) + 1;
723 OBD_ALLOC(sbi->ll_instance, len);
724 if (sbi->ll_instance == NULL)
725 GOTO(out_free, err = -ENOMEM);
726 sprintf(sbi->ll_instance, "%p", sb);
728 cfg.cfg_instance = sbi->ll_instance;
729 cfg.cfg_uuid = sbi->ll_sb_uuid;
730 cfg.cfg_local_nid = lmd->lmd_local_nid;
731 err = lustre_process_log(lmd, lmd->lmd_profile, &cfg, 0);
733 CERROR("Unable to process log: %s\n", lmd->lmd_profile);
737 lprof = class_get_profile(lmd->lmd_profile);
739 CERROR("No profile found: %s\n", lmd->lmd_profile);
740 GOTO(out_free, err = -EINVAL);
743 OBD_FREE(lov, strlen(lov) + 1);
744 OBD_ALLOC(lov, strlen(lprof->lp_lov) +
745 strlen(sbi->ll_instance) + 2);
746 sprintf(lov, "%s-%s", lprof->lp_lov, sbi->ll_instance);
749 OBD_FREE(lmv, strlen(lmv) + 1);
750 OBD_ALLOC(lmv, strlen(lprof->lp_lmv) +
751 strlen(sbi->ll_instance) + 2);
752 sprintf(lmv, "%s-%s", lprof->lp_lmv, sbi->ll_instance);
757 GOTO(out_free, err = -EINVAL);
762 GOTO(out_free, err = -EINVAL);
765 err = lustre_common_fill_super(sb, lmv, lov, lmd->lmd_async,
766 lmd->lmd_security, &lmd->lmd_nllu,
767 &lmd->lmd_remote_flag);
774 OBD_FREE(lmv, strlen(lmv) + 1);
776 OBD_FREE(lov, strlen(lov) + 1);
782 int len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean")+1;
785 if (sbi->ll_instance != NULL) {
786 struct lustre_mount_data *lmd = sbi->ll_lmd;
787 struct config_llog_instance cfg;
790 cfg.cfg_instance = sbi->ll_instance;
791 cfg.cfg_uuid = sbi->ll_sb_uuid;
793 OBD_ALLOC(cl_prof, len);
794 sprintf(cl_prof, "%s-clean", lmd->lmd_profile);
795 err = lustre_process_log(lmd, cl_prof, &cfg, 0);
797 CERROR("Unable to process log: %s\n", cl_prof);
798 lustre_manual_cleanup(sbi);
800 OBD_FREE(cl_prof, len);
801 OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance) + 1);
803 OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
807 } /* lustre_fill_super */
809 void lustre_put_super(struct super_block *sb)
811 struct obd_device *obd;
812 struct ll_sb_info *sbi = ll_s2sbi(sb);
813 int force_umount = 0;
816 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
817 obd = class_exp2obd(sbi->ll_md_exp);
819 force_umount = obd->obd_no_recov;
822 lustre_common_put_super(sb);
823 if (sbi->ll_lmd != NULL) {
825 int len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean")+1;
827 struct config_llog_instance cfg;
830 CERROR("force umount, doing manual cleanup\n");
831 lustre_manual_cleanup(sbi);
835 cfg.cfg_instance = sbi->ll_instance;
836 cfg.cfg_uuid = sbi->ll_sb_uuid;
838 OBD_ALLOC(cl_prof, len);
839 sprintf(cl_prof, "%s-clean", sbi->ll_lmd->lmd_profile);
840 err = lustre_process_log(sbi->ll_lmd, cl_prof, &cfg, 0);
842 CERROR("Unable to process log: %s, doing manual cleanup"
844 lustre_manual_cleanup(sbi);
847 OBD_FREE(cl_prof, len);
849 OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
850 OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance) + 1);
856 } /* lustre_put_super */
858 int ll_process_config_update(struct ll_sb_info *sbi, int clean)
860 struct obd_export *md_exp = sbi->ll_md_exp;
861 struct lustre_mount_data *lmd = sbi->ll_lmd;
862 struct llog_ctxt *ctxt;
863 struct config_llog_instance cfg;
864 char *profile = lmd->lmd_profile, *name = NULL;
865 int rc, namelen = 0, version;
871 CERROR("Client not mounted with zero-conf; cannot process "
876 rc = ldlm_cli_cancel_unused(md_exp->exp_obd->obd_namespace, NULL,
877 LDLM_FL_CONFIG_CHANGE, NULL);
879 CWARN("ldlm_cli_cancel_unused(mdc): %d\n", rc);
881 rc = obd_cancel_unused(sbi->ll_dt_exp, NULL, LDLM_FL_CONFIG_CHANGE,
884 CWARN("obd_cancel_unused(lov): %d\n", rc);
886 cfg.cfg_instance = sbi->ll_instance;
887 cfg.cfg_uuid = sbi->ll_sb_uuid;
888 cfg.cfg_local_nid = lmd->lmd_local_nid;
890 namelen = strlen(profile) + 20; /* -clean-######### */
891 OBD_ALLOC(name, namelen);
896 version = sbi->ll_config_version - 1;
897 sprintf(name, "%s-clean-%d", profile, version);
899 version = sbi->ll_config_version + 1;
900 sprintf(name, "%s-%d", profile, version);
903 CWARN("Applying configuration log %s\n", name);
905 ctxt = llog_get_context(&md_exp->exp_obd->obd_llogs,
906 LLOG_CONFIG_REPL_CTXT);
907 rc = class_config_process_llog(ctxt, name, &cfg);
909 sbi->ll_config_version = version;
910 CWARN("Finished applying configuration log %s: %d\n", name, rc);
912 if (rc == 0 && clean == 0) {
913 struct lov_desc desc;
915 valsize = sizeof(desc);
916 rc = obd_get_info(sbi->ll_dt_exp, strlen("lovdesc") + 1,
917 "lovdesc", &valsize, &desc);
919 rc = obd_init_ea_size(md_exp,
920 obd_size_diskmd(sbi->ll_dt_exp, NULL),
922 sizeof(struct llog_cookie)));
924 OBD_FREE(name, namelen);
928 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
930 struct inode *inode = NULL;
931 l_lock(&lock->l_resource->lr_namespace->ns_lock);
932 if (lock->l_ast_data) {
933 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
934 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
935 inode = igrab(lock->l_ast_data);
937 inode = lock->l_ast_data;
938 CDEBUG(inode->i_state & I_FREEING ? D_INFO : D_WARNING,
939 "l_ast_data %p is bogus: magic %0x8\n",
940 lock->l_ast_data, lli->lli_inode_magic);
944 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
948 int null_if_equal(struct ldlm_lock *lock, void *data)
950 if (data == lock->l_ast_data) {
951 lock->l_ast_data = NULL;
953 if (lock->l_req_mode != lock->l_granted_mode)
954 LDLM_ERROR(lock,"clearing inode with ungranted lock\n");
957 return LDLM_ITER_CONTINUE;
960 void ll_clear_inode(struct inode *inode)
963 struct ll_inode_info *lli = ll_i2info(inode);
964 struct ll_sb_info *sbi = ll_i2sbi(inode);
967 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
968 inode->i_generation, inode);
970 lli->lli_inode_magic = LLI_INODE_DEAD;
971 ll_inode2id(&id, inode);
973 clear_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &(ll_i2info(inode)->lli_flags));
974 md_change_cbdata(sbi->ll_md_exp, &id, null_if_equal, inode);
976 LASSERT(!lli->lli_open_fd_write_count);
977 LASSERT(!lli->lli_open_fd_read_count);
978 LASSERT(!lli->lli_open_fd_exec_count);
980 if (lli->lli_mds_write_och)
981 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_WRITE);
982 if (lli->lli_mds_exec_och)
983 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_EXEC);
984 if (lli->lli_mds_read_och)
985 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_READ);
988 obd_change_cbdata(sbi->ll_dt_exp, lli->lli_smd,
989 null_if_equal, inode);
992 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
997 obd_free_memmd(sbi->ll_md_exp,
998 (struct lov_stripe_md **) &lli->lli_mea);
1002 if (lli->lli_symlink_name) {
1003 OBD_FREE(lli->lli_symlink_name,
1004 strlen(lli->lli_symlink_name) + 1);
1005 lli->lli_symlink_name = NULL;
1007 lli->lli_inode_magic = LLI_INODE_DEAD;
1012 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1013 * object(s) determine the file size and mtime. Otherwise, the MDS will
1014 * keep these values until such a time that objects are allocated for it.
1015 * We do the MDS operations first, as it is checking permissions for us.
1016 * We don't to the MDS RPC if there is nothing that we want to store there,
1017 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1018 * going to do an RPC anyways.
1020 * If we are doing a truncate, we will send the mtime and ctime updates
1021 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1022 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1025 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1027 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1028 struct ll_sb_info *sbi = ll_i2sbi(inode);
1029 struct ptlrpc_request *request = NULL;
1030 struct mdc_op_data *op_data;
1031 int ia_valid = attr->ia_valid;
1035 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu\n", inode->i_ino);
1036 lprocfs_counter_incr(ll_i2sbi(inode)->ll_stats, LPROC_LL_SETATTR);
1038 if (ia_valid & ATTR_SIZE) {
1039 if (attr->ia_size > ll_file_maxbytes(inode)) {
1040 CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1041 attr->ia_size, ll_file_maxbytes(inode));
1045 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1048 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1049 if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1050 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1054 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1055 if (attr->ia_valid & ATTR_CTIME) {
1056 attr->ia_ctime = CURRENT_TIME;
1057 attr->ia_valid |= ATTR_CTIME_SET;
1059 if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1060 attr->ia_atime = CURRENT_TIME;
1061 attr->ia_valid |= ATTR_ATIME_SET;
1063 if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1064 attr->ia_mtime = CURRENT_TIME;
1065 attr->ia_valid |= ATTR_MTIME_SET;
1068 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1069 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1070 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1071 LTIME_S(CURRENT_TIME));
1074 attr->ia_valid &= ~ATTR_SIZE;
1076 /* If only OST attributes being set on objects, don't do MDS RPC.
1077 * In that case, we need to check permissions and update the local
1078 * inode ourselves so we can call obdo_from_inode() always. */
1079 if (ia_valid & (lsm ? ~(ATTR_SIZE | ATTR_FROM_OPEN /*| ATTR_RAW*/) : ~0)) {
1080 struct lustre_md md;
1082 OBD_ALLOC(op_data, sizeof(*op_data));
1083 if (op_data == NULL)
1085 ll_prepare_mdc_data(op_data, inode, NULL, NULL, 0, 0);
1087 rc = md_setattr(sbi->ll_md_exp, op_data,
1088 attr, NULL, 0, NULL, 0, &request);
1089 OBD_FREE(op_data, sizeof(*op_data));
1091 ptlrpc_req_finished(request);
1092 if (rc != -EPERM && rc != -EACCES)
1093 CERROR("md_setattr fails: rc = %d\n", rc);
1097 rc = mdc_req2lustre_md(sbi->ll_md_exp, request, 0,
1098 sbi->ll_dt_exp, &md);
1100 ptlrpc_req_finished(request);
1104 /* We call inode_setattr to adjust timestamps, but we first
1105 * clear ATTR_SIZE to avoid invoking vmtruncate.
1107 * NB: ATTR_SIZE will only be set at this point if the size
1108 * resides on the MDS, ie, this file has no objects. */
1109 attr->ia_valid &= ~ATTR_SIZE;
1110 inode_setattr(inode, attr);
1112 ll_update_inode(inode, &md);
1113 ptlrpc_req_finished(request);
1115 if (!lsm || !S_ISREG(inode->i_mode)) {
1116 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1120 /* The OST doesn't check permissions, but the alternative is
1121 * a gratuitous RPC to the MDS. We already rely on the client
1122 * to do read/write/truncate permission checks, so is mtime OK?
1124 if (ia_valid & (ATTR_MTIME | ATTR_ATIME)) {
1125 /* from sys_utime() */
1126 if (!(ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET))) {
1127 if (current->fsuid != inode->i_uid &&
1128 (rc=ll_permission(inode,MAY_WRITE,NULL))!=0)
1131 /* from inode_change_ok() */
1132 if (current->fsuid != inode->i_uid &&
1133 !capable(CAP_FOWNER))
1138 /* Won't invoke vmtruncate, as we already cleared ATTR_SIZE */
1139 inode_setattr(inode, attr);
1142 /* We really need to get our PW lock before we change inode->i_size.
1143 * If we don't we can race with other i_size updaters on our node, like
1144 * ll_file_read. We can also race with i_size propogation to other
1145 * nodes through dirtying and writeback of final cached pages. This
1146 * last one is especially bad for racing o_append users on other
1148 if (ia_valid & ATTR_SIZE) {
1149 ldlm_policy_data_t policy = { .l_extent = {attr->ia_size,
1151 struct lustre_handle lockh = { 0 };
1152 struct ll_inode_info *lli = ll_i2info(inode);
1153 int err, ast_flags = 0;
1154 /* XXX when we fix the AST intents to pass the discard-range
1155 * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
1157 if (attr->ia_size == 0)
1158 ast_flags = LDLM_AST_DISCARD_DATA;
1160 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy, &lockh,
1161 ast_flags, &ll_i2sbi(inode)->ll_seek_stime);
1166 down(&lli->lli_size_sem);
1167 lli->lli_size_pid = current->pid;
1168 rc = vmtruncate(inode, attr->ia_size);
1170 LASSERT(atomic_read(&lli->lli_size_sem.count) <= 0);
1171 lli->lli_size_pid = 0;
1172 up(&lli->lli_size_sem);
1175 err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
1177 CERROR("ll_extent_unlock failed: %d\n", err);
1181 } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
1182 struct obdo *oa = NULL;
1184 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1185 inode->i_ino, LTIME_S(attr->ia_mtime));
1191 oa->o_id = lsm->lsm_object_id;
1192 oa->o_gr = lsm->lsm_object_gr;
1193 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1194 obdo_from_inode(oa, inode, OBD_MD_FLTYPE | OBD_MD_FLATIME |
1195 OBD_MD_FLMTIME | OBD_MD_FLCTIME);
1196 rc = obd_setattr(sbi->ll_dt_exp, oa, lsm, NULL);
1199 CERROR("obd_setattr fails: rc = %d\n", rc);
1204 int ll_setattr(struct dentry *de, struct iattr *attr)
1206 LASSERT(de->d_inode);
1207 return ll_setattr_raw(de->d_inode, attr);
1210 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1211 unsigned long max_age)
1213 struct ll_sb_info *sbi = ll_s2sbi(sb);
1214 struct obd_statfs obd_osfs;
1218 rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age);
1220 CERROR("obd_statfs fails: rc = %d\n", rc);
1224 osfs->os_type = sb->s_magic;
1226 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1227 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1229 rc = obd_statfs(class_exp2obd(sbi->ll_dt_exp), &obd_osfs, max_age);
1231 CERROR("obd_statfs fails: rc = %d\n", rc);
1235 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1236 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1239 osfs->os_blocks = obd_osfs.os_blocks;
1240 osfs->os_bfree = obd_osfs.os_bfree;
1241 osfs->os_bavail = obd_osfs.os_bavail;
1243 /* If we don't have as many objects free on the OST as inodes
1244 * on the MDS, we reduce the total number of inodes to
1245 * compensate, so that the "inodes in use" number is correct.
1247 if (obd_osfs.os_ffree < osfs->os_ffree) {
1248 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1250 osfs->os_ffree = obd_osfs.os_ffree;
1256 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1258 struct obd_statfs osfs;
1261 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p\n", sb);
1262 lprocfs_counter_incr(ll_s2sbi(sb)->ll_stats, LPROC_LL_STAFS);
1264 /* For now we will always get up-to-date statfs values, but in the
1265 * future we may allow some amount of caching on the client (e.g.
1266 * from QOS or lprocfs updates). */
1267 rc = ll_statfs_internal(sb, &osfs, jiffies - 1);
1271 statfs_unpack(sfs, &osfs);
1273 if (sizeof(sfs->f_blocks) == 4) {
1274 while (osfs.os_blocks > ~0UL) {
1277 osfs.os_blocks >>= 1;
1278 osfs.os_bfree >>= 1;
1279 osfs.os_bavail >>= 1;
1283 sfs->f_blocks = osfs.os_blocks;
1284 sfs->f_bfree = osfs.os_bfree;
1285 sfs->f_bavail = osfs.os_bavail;
1290 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1292 struct ll_inode_info *lli = ll_i2info(inode);
1293 struct lov_stripe_md *lsm = md->lsm;
1294 struct mds_body *body = md->body;
1295 struct mea *mea = md->mea;
1296 struct posix_acl *ll_acl_access = md->acl_access;
1299 LASSERT((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1301 if (md->lsm && md->lsm->lsm_magic != LOV_MAGIC) {
1302 /* check for default striping info for dir. */
1303 LASSERT((mea != NULL) == ((body->valid & OBD_MD_FLDIREA) != 0));
1307 LASSERT(lsm->lsm_object_gr > 0);
1308 if (lli->lli_smd == NULL) {
1310 lli->lli_maxbytes = lsm->lsm_maxbytes;
1311 if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1312 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1315 if (memcmp(lli->lli_smd, lsm, sizeof(*lsm))) {
1316 CERROR("lsm mismatch for inode %ld\n",
1318 CERROR("lli_smd:\n");
1319 dump_lsm(D_ERROR, lli->lli_smd);
1321 dump_lsm(D_ERROR, lsm);
1324 /* XXX FIXME -- We should decide on a safer (atomic) and
1325 * more elegant way to update the lsm */
1326 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1327 lli->lli_smd->lsm_oinfo[i].loi_id =
1328 lsm->lsm_oinfo[i].loi_id;
1329 lli->lli_smd->lsm_oinfo[i].loi_gr =
1330 lsm->lsm_oinfo[i].loi_gr;
1331 lli->lli_smd->lsm_oinfo[i].loi_ost_idx =
1332 lsm->lsm_oinfo[i].loi_ost_idx;
1333 lli->lli_smd->lsm_oinfo[i].loi_ost_gen =
1334 lsm->lsm_oinfo[i].loi_ost_gen;
1337 /* bug 2844 - limit i_blksize for broken user-space apps */
1338 LASSERTF(lsm->lsm_xfersize != 0, "%lu\n", lsm->lsm_xfersize);
1339 inode->i_blksize = min(lsm->lsm_xfersize, LL_MAX_BLKSIZE);
1340 if (lli->lli_smd != lsm)
1341 obd_free_memmd(ll_i2dtexp(inode), &lsm);
1345 if (lli->lli_mea == NULL) {
1348 if (memcmp(lli->lli_mea, mea, body->eadatasize)) {
1349 CERROR("mea mismatch for inode %lu\n",
1354 if (lli->lli_mea != mea)
1355 obd_free_memmd(ll_i2mdexp(inode),
1356 (struct lov_stripe_md **) &mea);
1359 if (body->valid & OBD_MD_FID)
1360 id_assign_fid(&lli->lli_id, &body->id1);
1362 if (body->valid & OBD_MD_FLID)
1363 id_ino(&lli->lli_id) = id_ino(&body->id1);
1365 if (body->valid & OBD_MD_FLGENER)
1366 id_gen(&lli->lli_id) = id_gen(&body->id1);
1368 spin_lock(&lli->lli_lock);
1369 if (ll_acl_access != NULL) {
1370 if (lli->lli_acl_access != NULL)
1371 posix_acl_release(lli->lli_acl_access);
1372 lli->lli_acl_access = ll_acl_access;
1374 spin_unlock(&lli->lli_lock);
1376 if (body->valid & OBD_MD_FLID)
1377 inode->i_ino = id_ino(&body->id1);
1378 if (body->valid & OBD_MD_FLGENER)
1379 inode->i_generation = id_gen(&body->id1);
1380 if (body->valid & OBD_MD_FLATIME)
1381 LTIME_S(inode->i_atime) = body->atime;
1382 if (body->valid & OBD_MD_FLMTIME &&
1383 body->mtime > LTIME_S(inode->i_mtime)) {
1384 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to %u\n",
1385 inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
1386 LTIME_S(inode->i_mtime) = body->mtime;
1388 if (body->valid & OBD_MD_FLCTIME &&
1389 body->ctime > LTIME_S(inode->i_ctime))
1390 LTIME_S(inode->i_ctime) = body->ctime;
1391 if (body->valid & OBD_MD_FLMODE) {
1392 inode->i_mode = (inode->i_mode & S_IFMT) |
1393 (body->mode & ~S_IFMT);
1395 if (body->valid & OBD_MD_FLTYPE) {
1396 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1397 (body->mode & S_IFMT);
1399 if (body->valid & OBD_MD_FLUID)
1400 inode->i_uid = body->uid;
1401 if (body->valid & OBD_MD_FLGID)
1402 inode->i_gid = body->gid;
1403 if (body->valid & OBD_MD_FLFLAGS)
1404 inode->i_flags = body->flags;
1405 if (body->valid & OBD_MD_FLNLINK)
1406 inode->i_nlink = body->nlink;
1407 if (body->valid & OBD_MD_FLRDEV)
1408 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1409 inode->i_rdev = body->rdev;
1411 inode->i_rdev = old_decode_dev(body->rdev);
1413 if (body->valid & OBD_MD_FLSIZE)
1414 inode->i_size = body->size;
1415 if (body->valid & OBD_MD_FLBLOCKS)
1416 inode->i_blocks = body->blocks;
1418 if (body->valid & OBD_MD_FLSIZE)
1419 set_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &lli->lli_flags);
1421 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1422 inode->i_dev = (kdev_t)id_group(&lli->lli_id);
1424 LASSERT(id_fid(&lli->lli_id) != 0);
1427 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
1428 static struct backing_dev_info ll_backing_dev_info = {
1429 .ra_pages = 0, /* No readahead */
1430 .memory_backed = 0, /* Does contribute to dirty memory */
1434 void ll_read_inode2(struct inode *inode, void *opaque)
1436 struct lustre_md *md = opaque;
1437 struct ll_inode_info *lli = ll_i2info(inode);
1440 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1441 inode->i_generation, inode);
1445 LASSERT(!lli->lli_smd);
1447 /* Core attributes from the MDS first. This is a new inode, and
1448 * the VFS doesn't zero times in the core inode so we have to do
1449 * it ourselves. They will be overwritten by either MDS or OST
1450 * attributes - we just need to make sure they aren't newer. */
1451 LTIME_S(inode->i_mtime) = 0;
1452 LTIME_S(inode->i_atime) = 0;
1453 LTIME_S(inode->i_ctime) = 0;
1456 ll_update_inode(inode, md);
1458 /* OIDEBUG(inode); */
1460 if (S_ISREG(inode->i_mode)) {
1461 inode->i_op = &ll_file_inode_operations;
1462 inode->i_fop = &ll_file_operations;
1463 inode->i_mapping->a_ops = &ll_aops;
1465 } else if (S_ISDIR(inode->i_mode)) {
1466 inode->i_op = &ll_dir_inode_operations;
1467 inode->i_fop = &ll_dir_operations;
1468 inode->i_mapping->a_ops = &ll_dir_aops;
1470 } else if (S_ISLNK(inode->i_mode)) {
1471 inode->i_op = &ll_fast_symlink_inode_operations;
1474 inode->i_op = &ll_special_inode_operations;
1476 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
1477 init_special_inode(inode, inode->i_mode,
1478 kdev_t_to_nr(inode->i_rdev));
1480 /* initializing backing dev info. */
1481 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
1483 init_special_inode(inode, inode->i_mode, inode->i_rdev);
1485 lli->ll_save_ifop = inode->i_fop;
1487 if (S_ISCHR(inode->i_mode))
1488 inode->i_fop = &ll_special_chr_inode_fops;
1489 else if (S_ISBLK(inode->i_mode))
1490 inode->i_fop = &ll_special_blk_inode_fops;
1491 else if (S_ISFIFO(inode->i_mode))
1492 inode->i_fop = &ll_special_fifo_inode_fops;
1493 else if (S_ISSOCK(inode->i_mode))
1494 inode->i_fop = &ll_special_sock_inode_fops;
1496 CWARN("saved %p, replaced with %p\n", lli->ll_save_ifop,
1499 if (lli->ll_save_ifop->owner) {
1500 CWARN("%p has owner %p\n", lli->ll_save_ifop,
1501 lli->ll_save_ifop->owner);
1507 void ll_delete_inode(struct inode *inode)
1509 struct ll_sb_info *sbi = ll_i2sbi(inode);
1510 struct lustre_id id;
1514 ll_inode2id(&id, inode);
1516 rc = md_delete_inode(sbi->ll_md_exp, &id);
1518 CERROR("md_delete_inode() failed, error %d\n",
1526 int ll_iocontrol(struct inode *inode, struct file *file,
1527 unsigned int cmd, unsigned long arg)
1529 struct ll_sb_info *sbi = ll_i2sbi(inode);
1530 struct ptlrpc_request *req = NULL;
1535 case EXT3_IOC_GETFLAGS: {
1536 struct lustre_id id;
1537 __u64 valid = OBD_MD_FLFLAGS;
1538 struct mds_body *body;
1540 ll_inode2id(&id, inode);
1541 rc = md_getattr(sbi->ll_md_exp, &id, valid, NULL, 0, 0, &req);
1543 CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1547 body = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*body));
1549 if (body->flags & S_APPEND)
1550 flags |= EXT3_APPEND_FL;
1551 if (body->flags & S_IMMUTABLE)
1552 flags |= EXT3_IMMUTABLE_FL;
1553 if (body->flags & S_NOATIME)
1554 flags |= EXT3_NOATIME_FL;
1556 ptlrpc_req_finished (req);
1558 RETURN(put_user(flags, (int *)arg));
1560 case EXT3_IOC_SETFLAGS: {
1561 struct mdc_op_data *op_data;
1564 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1566 if (get_user(flags, (int *)arg))
1573 OBD_ALLOC(op_data, sizeof(*op_data));
1574 if (op_data == NULL) {
1578 ll_prepare_mdc_data(op_data, inode, NULL, NULL, 0, 0);
1580 memset(&attr, 0x0, sizeof(attr));
1581 attr.ia_attr_flags = flags;
1582 attr.ia_valid |= ATTR_ATTR_FLAG;
1584 rc = md_setattr(sbi->ll_md_exp, op_data,
1585 &attr, NULL, 0, NULL, 0, &req);
1586 OBD_FREE(op_data, sizeof(*op_data));
1588 ptlrpc_req_finished(req);
1589 if (rc != -EPERM && rc != -EACCES)
1590 CERROR("md_setattr fails: rc = %d\n", rc);
1594 ptlrpc_req_finished(req);
1596 oa->o_id = lsm->lsm_object_id;
1597 oa->o_gr = lsm->lsm_object_gr;
1598 oa->o_flags = flags;
1599 oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS | OBD_MD_FLGROUP;
1601 rc = obd_setattr(sbi->ll_dt_exp, oa, lsm, NULL);
1604 if (rc != -EPERM && rc != -EACCES)
1605 CERROR("md_setattr fails: rc = %d\n", rc);
1609 if (flags & EXT3_APPEND_FL)
1610 inode->i_flags |= S_APPEND;
1612 inode->i_flags &= ~S_APPEND;
1613 if (flags & EXT3_IMMUTABLE_FL)
1614 inode->i_flags |= S_IMMUTABLE;
1616 inode->i_flags &= ~S_IMMUTABLE;
1617 if (flags & EXT3_NOATIME_FL)
1618 inode->i_flags |= S_NOATIME;
1620 inode->i_flags &= ~S_NOATIME;
1631 /* this is only called in the case of forced umount. */
1632 void ll_umount_begin(struct super_block *sb)
1634 struct ll_sb_info *sbi = ll_s2sbi(sb);
1635 struct obd_ioctl_data ioc_data = { 0 };
1636 struct obd_device *obd;
1639 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1640 sb->s_count, atomic_read(&sb->s_active));
1642 obd = class_exp2obd(sbi->ll_md_exp);
1644 CERROR("Invalid MDC connection handle "LPX64"\n",
1645 sbi->ll_md_exp->exp_handle.h_cookie);
1649 obd->obd_no_recov = 1;
1650 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
1651 sizeof(ioc_data), &ioc_data, NULL);
1653 obd = class_exp2obd(sbi->ll_dt_exp);
1655 CERROR("Invalid LOV connection handle "LPX64"\n",
1656 sbi->ll_dt_exp->exp_handle.h_cookie);
1661 obd->obd_no_recov = 1;
1662 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
1663 sizeof(ioc_data), &ioc_data, NULL);
1666 * really, we'd like to wait until there are no requests outstanding,
1667 * and then continue. For now, we just invalidate the requests,
1668 * schedule, and hope.
1675 int ll_prep_inode(struct obd_export *dt_exp, struct obd_export *md_exp,
1676 struct inode **inode, struct ptlrpc_request *req,
1677 int offset, struct super_block *sb)
1679 struct lustre_md md;
1682 rc = mdc_req2lustre_md(md_exp, req, offset, dt_exp, &md);
1687 ll_update_inode(*inode, &md);
1690 *inode = ll_iget(sb, id_ino(&md.body->id1), &md);
1691 if (*inode == NULL || is_bad_inode(*inode)) {
1692 /* free the lsm if we allocated one above */
1694 obd_free_memmd(dt_exp, &md.lsm);
1696 obd_free_memmd(md_exp,
1697 (struct lov_stripe_md**)&md.mea);
1699 CERROR("new_inode -fatal: rc %d\n", rc);
1706 int ll_get_fid(struct obd_export *exp, struct lustre_id *idp,
1707 char *filename, struct lustre_id *ret)
1709 struct ptlrpc_request *request = NULL;
1710 struct mds_body *body;
1713 rc = md_getattr_lock(exp, idp, filename, strlen(filename) + 1,
1714 OBD_MD_FID, 0, &request);
1716 CDEBUG(D_INFO, "md_getattr_lock failed on %s: rc %d\n",
1721 body = lustre_msg_buf(request->rq_repmsg, 0, sizeof(*body));
1722 LASSERT(body != NULL);
1723 LASSERT_REPSWABBED(request, 0);
1726 ptlrpc_req_finished(request);