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) - 1] = '\0';
79 /* this later may be reset via /proc/fs/... */
80 memset(sbi->ll_gns_upcall, 0, sizeof(sbi->ll_gns_upcall));
82 /* default values, may be changed via /proc/fs/... */
83 sbi->ll_gns_state = LL_GNS_IDLE;
84 sbi->ll_gns_tick = GNS_TICK_TIMEOUT;
85 sbi->ll_gns_timeout = GNS_MOUNT_TIMEOUT;
87 sbi->ll_gns_timer.data = (unsigned long)sbi;
88 sbi->ll_gns_timer.function = ll_gns_timer_callback;
89 init_timer(&sbi->ll_gns_timer);
93 generate_random_uuid(uuid);
94 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
98 void lustre_free_sbi(struct super_block *sb)
100 struct ll_sb_info *sbi = ll_s2sbi(sb);
104 list_del(&sbi->ll_gns_sbi_head);
105 del_timer(&sbi->ll_gns_timer);
106 OBD_FREE(sbi, sizeof(*sbi));
108 ll_set_sbi(sb, NULL);
112 int lustre_init_dt_desc(struct ll_sb_info *sbi)
117 valsize = sizeof(sbi->ll_dt_desc);
118 memset(&sbi->ll_dt_desc, 0, sizeof(sbi->ll_dt_desc));
119 rc = obd_get_info(sbi->ll_dt_exp, strlen("lovdesc") + 1,
120 "lovdesc", &valsize, &sbi->ll_dt_desc);
124 extern struct dentry_operations ll_d_ops;
126 int lustre_common_fill_super(struct super_block *sb, char *lmv, char *lov,
127 char *security, __u32 *nllu)
129 struct ll_sb_info *sbi = ll_s2sbi(sb);
130 struct ptlrpc_request *request = NULL;
131 struct lustre_handle dt_conn = {0, };
132 struct lustre_handle md_conn = {0, };
133 struct inode *root = NULL;
134 struct obd_device *obd;
135 struct obd_statfs osfs;
141 obd = class_name2obd(lmv);
143 CERROR("MDC %s: not setup or attached\n", lmv);
147 if (security == NULL)
150 err = obd_set_info(obd->obd_self_export, strlen("sec"), "sec",
151 strlen(security), security);
153 CERROR("LMV %s: failed to set security %s, err %d\n",
158 err = obd_set_info(obd->obd_self_export, strlen("nllu"), "nllu",
159 sizeof(__u32) * 2, nllu);
161 CERROR("LMV %s: failed to set NLLU, err %d\n",
166 if (proc_lustre_fs_root) {
167 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
170 CERROR("could not register mount in /proc/lustre");
173 err = obd_connect(&md_conn, obd, &sbi->ll_sb_uuid, OBD_OPT_REAL_CLIENT);
175 CERROR("An MDS (mdc %s) is performing recovery, of which this"
176 " client is not a part. Please wait for recovery to "
177 "complete, abort, or time out.\n", lmv);
180 CERROR("cannot connect to %s: rc = %d\n", lmv, err);
183 sbi->ll_md_exp = class_conn2export(&md_conn);
184 err = obd_statfs(obd, &osfs, jiffies - HZ);
188 if (!osfs.os_bsize) {
189 CERROR("Invalid block size is detected.");
193 sb->s_magic = LL_SUPER_MAGIC;
194 sb->s_blocksize = osfs.os_bsize;
195 sb->s_blocksize_bits = log2(osfs.os_bsize);
196 sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
198 devno = get_uuid2int(sbi->ll_md_exp->exp_obd->obd_uuid.uuid,
199 strlen(sbi->ll_md_exp->exp_obd->obd_uuid.uuid));
203 obd = class_name2obd(lov);
205 CERROR("OSC %s: not setup or attached\n", lov);
209 err = obd_connect(&dt_conn, obd, &sbi->ll_sb_uuid, OBD_OPT_REAL_CLIENT);
211 CERROR("An OST (osc %s) is performing recovery, of which this"
212 " client is not a part. Please wait for recovery to "
213 "complete, abort, or time out.\n", lov);
216 CERROR("cannot connect to %s: rc = %d\n", lov, err);
219 sbi->ll_dt_exp = class_conn2export(&dt_conn);
221 err = lustre_init_dt_desc(sbi);
223 int mdsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
224 obd_init_ea_size(sbi->ll_md_exp, mdsize,
225 sbi->ll_dt_desc.ld_tgt_count *
226 sizeof(struct llog_cookie));
229 err = md_getstatus(sbi->ll_md_exp, &sbi->ll_rootid);
231 CERROR("cannot mds_connect: rc = %d\n", err);
234 CDEBUG(D_SUPER, "rootid "DLID4"\n", OLID4(&sbi->ll_rootid));
236 sb->s_op = &lustre_super_operations;
239 * XXX: move this to after cbd setup? */
240 err = md_getattr(sbi->ll_md_exp, &sbi->ll_rootid,
241 (OBD_MD_FLNOTOBD | OBD_MD_FLBLOCKS | OBD_MD_FID), NULL, 0,
244 CERROR("md_getattr failed for root: rc = %d\n", err);
248 err = mdc_req2lustre_md(sbi->ll_md_exp, request, 0,
249 sbi->ll_dt_exp, &md);
251 CERROR("failed to understand root inode md: rc = %d\n", err);
252 ptlrpc_req_finished(request);
256 LASSERT(id_ino(&sbi->ll_rootid) != 0);
257 root = ll_iget(sb, id_ino(&sbi->ll_rootid), &md);
259 ptlrpc_req_finished(request);
261 if (root == NULL || is_bad_inode(root)) {
262 CERROR("lustre_lite: bad iget4 for root\n");
263 GOTO(out_root, err = -EBADF);
266 err = ll_close_thread_start(&sbi->ll_lcq);
268 CERROR("cannot start close thread: rc %d\n", err);
272 ll_gns_add_timer(sbi);
274 /* making vm readahead 0 for 2.4.x. In the case of 2.6.x, backing dev
275 info assigned to inode mapping is used for determining maximal
277 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0))
278 /* bug 2805 - set VM readahead to zero */
279 vm_max_readahead = vm_min_readahead = 0;
282 sb->s_root = d_alloc_root(root);
283 sb->s_root->d_op = &ll_d_ops;
286 CWARN("Enabling PDIROPS\n");
287 sb->s_flags |= S_PDIROPS;
295 obd_disconnect(sbi->ll_dt_exp, 0);
297 obd_disconnect(sbi->ll_md_exp, 0);
299 lprocfs_unregister_mountpoint(sbi);
303 void lustre_common_put_super(struct super_block *sb)
305 struct ll_sb_info *sbi = ll_s2sbi(sb);
306 struct hlist_node *tmp, *next;
309 ll_gns_del_timer(sbi);
310 ll_close_thread_shutdown(sbi->ll_lcq);
312 list_del(&sbi->ll_conn_chain);
313 obd_disconnect(sbi->ll_dt_exp, 0);
315 lprocfs_unregister_mountpoint(sbi);
316 if (sbi->ll_proc_root) {
317 lprocfs_remove(sbi->ll_proc_root);
318 sbi->ll_proc_root = NULL;
321 obd_disconnect(sbi->ll_md_exp, 0);
323 // We do this to get rid of orphaned dentries. That is not really trw.
324 spin_lock(&dcache_lock);
325 hlist_for_each_safe(tmp, next, &sbi->ll_orphan_dentry_list) {
326 struct dentry *dentry = hlist_entry(tmp, struct dentry, d_hash);
327 CWARN("orphan dentry %*s (%p) at unmount\n",
328 dentry->d_name.len, dentry->d_name.name, dentry);
329 shrink_dcache_parent(dentry);
331 spin_unlock(&dcache_lock);
335 char *ll_read_opt(const char *opt, char *data)
341 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
342 if (strncmp(opt, data, strlen(opt)))
344 if ((value = strchr(data, '=')) == NULL)
348 OBD_ALLOC(retval, strlen(value) + 1);
350 CERROR("out of memory!\n");
354 memcpy(retval, value, strlen(value)+1);
355 CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
359 int ll_set_opt(const char *opt, char *data, int fl)
363 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
364 if (strncmp(opt, data, strlen(opt)))
370 void ll_options(char *options, char **lov, char **lmv, char **sec, int *flags)
373 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
374 char *opt_ptr = options;
383 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
384 for (this_char = strtok (options, ",");
386 this_char = strtok (NULL, ",")) {
388 while ((this_char = strsep (&opt_ptr, ",")) != NULL) {
390 CDEBUG(D_SUPER, "this_char %s\n", this_char);
391 if (!*lov && (*lov = ll_read_opt("osc", this_char)))
393 if (!*lmv && (*lmv = ll_read_opt("mdc", this_char)))
395 if (!*sec && (*sec = ll_read_opt("sec", this_char)))
397 if (!(*flags & LL_SBI_NOLCK) &&
398 ((*flags) = (*flags) |
399 ll_set_opt("nolock", this_char,
407 void ll_lli_init(struct ll_inode_info *lli)
409 sema_init(&lli->lli_open_sem, 1);
411 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
412 spin_lock_init(&lli->lli_lock);
413 INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
414 lli->lli_inode_magic = LLI_INODE_MAGIC;
415 memset(&lli->lli_id, 0, sizeof(lli->lli_id));
418 int ll_fill_super(struct super_block *sb, void *data, int silent)
420 struct ll_sb_info *sbi;
424 __u32 nllu[2] = { 99, 99 };
428 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
430 sbi = lustre_init_sbi(sb);
434 sbi->ll_flags |= LL_SBI_READAHEAD;
435 ll_options(data, &lov, &lmv, &sec, &sbi->ll_flags);
439 GOTO(out, err = -EINVAL);
444 GOTO(out, err = -EINVAL);
447 err = lustre_common_fill_super(sb, lmv, lov, sec, nllu);
454 OBD_FREE(sec, strlen(sec) + 1);
456 OBD_FREE(lmv, strlen(lmv) + 1);
458 OBD_FREE(lov, strlen(lov) + 1);
460 } /* ll_read_super */
462 static int lustre_process_log(struct lustre_mount_data *lmd, char *profile,
463 struct config_llog_instance *cfg, int allow_recov)
465 struct lustre_cfg lcfg;
466 struct portals_cfg pcfg;
467 char *peer = "MDS_PEER_UUID";
468 struct obd_device *obd;
469 struct lustre_handle md_conn = {0, };
470 struct obd_export *exp;
471 char *name = "mdc_dev";
473 struct obd_uuid lmv_uuid;
474 struct llog_ctxt *ctxt;
478 if (lmd_bad_magic(lmd))
481 generate_random_uuid(uuid);
482 class_uuid_unparse(uuid, &lmv_uuid);
484 if (lmd->lmd_local_nid) {
485 PCFG_INIT(pcfg, NAL_CMD_REGISTER_MYNID);
486 pcfg.pcfg_nal = lmd->lmd_nal;
487 pcfg.pcfg_nid = lmd->lmd_local_nid;
488 rc = libcfs_nal_cmd(&pcfg);
493 if (lmd->lmd_nal == SOCKNAL ||
494 lmd->lmd_nal == OPENIBNAL ||
495 lmd->lmd_nal == IIBNAL ||
496 lmd->lmd_nal == VIBNAL ||
497 lmd->lmd_nal == RANAL) {
498 PCFG_INIT(pcfg, NAL_CMD_ADD_PEER);
499 pcfg.pcfg_nal = lmd->lmd_nal;
500 pcfg.pcfg_nid = lmd->lmd_server_nid;
501 pcfg.pcfg_id = lmd->lmd_server_ipaddr;
502 pcfg.pcfg_misc = lmd->lmd_port;
503 rc = libcfs_nal_cmd(&pcfg);
508 LCFG_INIT(lcfg, LCFG_ADD_UUID, name);
509 lcfg.lcfg_nid = lmd->lmd_server_nid;
510 lcfg.lcfg_inllen1 = strlen(peer) + 1;
511 lcfg.lcfg_inlbuf1 = peer;
512 lcfg.lcfg_nal = lmd->lmd_nal;
513 rc = class_process_config(&lcfg);
515 GOTO(out_del_conn, rc);
517 LCFG_INIT(lcfg, LCFG_ATTACH, name);
518 lcfg.lcfg_inlbuf1 = "mdc";
519 lcfg.lcfg_inllen1 = strlen(lcfg.lcfg_inlbuf1) + 1;
520 lcfg.lcfg_inlbuf2 = lmv_uuid.uuid;
521 lcfg.lcfg_inllen2 = strlen(lcfg.lcfg_inlbuf2) + 1;
522 err = class_process_config(&lcfg);
524 GOTO(out_del_uuid, rc);
526 LCFG_INIT(lcfg, LCFG_SETUP, name);
527 lcfg.lcfg_inlbuf1 = lmd->lmd_mds;
528 lcfg.lcfg_inllen1 = strlen(lcfg.lcfg_inlbuf1) + 1;
529 lcfg.lcfg_inlbuf2 = peer;
530 lcfg.lcfg_inllen2 = strlen(lcfg.lcfg_inlbuf2) + 1;
531 rc = class_process_config(&lcfg);
533 GOTO(out_detach, rc);
535 obd = class_name2obd(name);
537 GOTO(out_cleanup, rc = -EINVAL);
539 rc = obd_set_info(obd->obd_self_export, strlen("sec"), "sec",
540 strlen(lmd->lmd_security), lmd->lmd_security);
542 GOTO(out_cleanup, rc);
544 /* Disable initial recovery on this import */
545 rc = obd_set_info(obd->obd_self_export,
546 strlen("initial_recov"), "initial_recov",
547 sizeof(allow_recov), &allow_recov);
549 GOTO(out_cleanup, rc);
551 rc = obd_connect(&md_conn, obd, &lmv_uuid, 0);
553 CERROR("cannot connect to %s: rc = %d\n", lmd->lmd_mds, rc);
554 GOTO(out_cleanup, rc);
557 exp = class_conn2export(&md_conn);
559 ctxt = llog_get_context(&exp->exp_obd->obd_llogs,LLOG_CONFIG_REPL_CTXT);
560 rc = class_config_process_llog(ctxt, profile, cfg);
562 CERROR("class_config_process_llog failed: rc = %d\n", rc);
564 rc = obd_disconnect(exp, 0);
568 LCFG_INIT(lcfg, LCFG_CLEANUP, name);
569 err = class_process_config(&lcfg);
574 LCFG_INIT(lcfg, LCFG_DETACH, name);
575 err = class_process_config(&lcfg);
580 LCFG_INIT(lcfg, LCFG_DEL_UUID, name);
581 lcfg.lcfg_inllen1 = strlen(peer) + 1;
582 lcfg.lcfg_inlbuf1 = peer;
583 err = class_process_config(&lcfg);
586 if (lmd->lmd_nal == SOCKNAL ||
587 lmd->lmd_nal == OPENIBNAL ||
588 lmd->lmd_nal == IIBNAL ||
589 lmd->lmd_nal == VIBNAL ||
590 lmd->lmd_nal == RANAL) {
593 PCFG_INIT(pcfg, NAL_CMD_DEL_PEER);
594 pcfg.pcfg_nal = lmd->lmd_nal;
595 pcfg.pcfg_nid = lmd->lmd_server_nid;
596 pcfg.pcfg_flags = 1; /* single_share */
597 err2 = libcfs_nal_cmd(&pcfg);
610 int lustre_fill_super(struct super_block *sb, void *data, int silent)
612 struct lustre_mount_data * lmd = data;
613 char *lov = NULL, *lmv = NULL;
614 struct ll_sb_info *sbi;
618 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
619 if (lmd_bad_magic(lmd))
622 sbi = lustre_init_sbi(sb);
626 sbi->ll_flags |= LL_SBI_READAHEAD;
628 if (lmd->lmd_profile) {
629 struct lustre_profile *lprof;
630 struct config_llog_instance cfg;
633 if (lmd->lmd_mds[0] == '\0') {
634 CERROR("no mds name\n");
635 GOTO(out_free, err = -EINVAL);
637 lmd->lmd_security[sizeof(lmd->lmd_security) - 1] = 0;
639 OBD_ALLOC(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
640 if (sbi->ll_lmd == NULL)
641 GOTO(out_free, err = -ENOMEM);
642 memcpy(sbi->ll_lmd, lmd, sizeof(*lmd));
644 /* generate a string unique to this super, let's try
645 the address of the super itself.*/
646 len = (sizeof(sb) * 2) + 1;
647 OBD_ALLOC(sbi->ll_instance, len);
648 if (sbi->ll_instance == NULL)
649 GOTO(out_free, err = -ENOMEM);
650 sprintf(sbi->ll_instance, "%p", sb);
652 cfg.cfg_instance = sbi->ll_instance;
653 cfg.cfg_uuid = sbi->ll_sb_uuid;
654 cfg.cfg_local_nid = lmd->lmd_local_nid;
655 err = lustre_process_log(lmd, lmd->lmd_profile, &cfg, 0);
657 CERROR("Unable to process log: %s\n", lmd->lmd_profile);
661 lprof = class_get_profile(lmd->lmd_profile);
663 CERROR("No profile found: %s\n", lmd->lmd_profile);
664 GOTO(out_free, err = -EINVAL);
667 OBD_FREE(lov, strlen(lov) + 1);
668 OBD_ALLOC(lov, strlen(lprof->lp_lov) +
669 strlen(sbi->ll_instance) + 2);
670 sprintf(lov, "%s-%s", lprof->lp_lov, sbi->ll_instance);
673 OBD_FREE(lmv, strlen(lmv) + 1);
674 OBD_ALLOC(lmv, strlen(lprof->lp_lmv) +
675 strlen(sbi->ll_instance) + 2);
676 sprintf(lmv, "%s-%s", lprof->lp_lmv, sbi->ll_instance);
681 GOTO(out_free, err = -EINVAL);
686 GOTO(out_free, err = -EINVAL);
689 err = lustre_common_fill_super(sb, lmv, lov, lmd->lmd_security,
697 OBD_FREE(lmv, strlen(lmv) + 1);
699 OBD_FREE(lov, strlen(lov) + 1);
705 int len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean")+1;
708 if (sbi->ll_instance != NULL) {
709 struct lustre_mount_data *lmd = sbi->ll_lmd;
711 struct config_llog_instance cfg;
713 cfg.cfg_instance = sbi->ll_instance;
714 cfg.cfg_uuid = sbi->ll_sb_uuid;
716 OBD_ALLOC(cln_prof, len);
717 sprintf(cln_prof, "%s-clean", lmd->lmd_profile);
719 err = lustre_process_log(lmd, cln_prof, &cfg, 0);
721 CERROR("Unable to process log: %s\n", cln_prof);
722 OBD_FREE(cln_prof, len);
723 OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance)+ 1);
725 OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
730 } /* lustre_fill_super */
732 static void lustre_manual_cleanup(struct ll_sb_info *sbi)
734 struct lustre_cfg lcfg;
735 struct obd_device *obd;
738 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) != NULL)
742 LCFG_INIT(lcfg, LCFG_CLEANUP, obd->obd_name);
743 err = class_process_config(&lcfg);
745 CERROR("cleanup failed: %s\n", obd->obd_name);
749 LCFG_INIT(lcfg, LCFG_DETACH, obd->obd_name);
750 err = class_process_config(&lcfg);
752 CERROR("detach failed: %s\n", obd->obd_name);
757 if (sbi->ll_lmd != NULL)
758 class_del_profile(sbi->ll_lmd->lmd_profile);
761 void lustre_put_super(struct super_block *sb)
763 struct obd_device *obd;
764 struct ll_sb_info *sbi = ll_s2sbi(sb);
765 int force_umount = 0;
768 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
769 obd = class_exp2obd(sbi->ll_md_exp);
771 force_umount = obd->obd_no_recov;
774 lustre_common_put_super(sb);
775 if (sbi->ll_lmd != NULL) {
777 int len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean")+1;
779 struct config_llog_instance cfg;
782 CERROR("force umount, doing manual cleanup\n");
783 lustre_manual_cleanup(sbi);
787 cfg.cfg_instance = sbi->ll_instance;
788 cfg.cfg_uuid = sbi->ll_sb_uuid;
790 OBD_ALLOC(cln_prof, len);
791 sprintf(cln_prof, "%s-clean", sbi->ll_lmd->lmd_profile);
793 err = lustre_process_log(sbi->ll_lmd, cln_prof, &cfg, 0);
795 CERROR("Unable to process log: %s, doing manual cleanup"
797 lustre_manual_cleanup(sbi);
800 OBD_FREE(cln_prof, len);
802 OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
803 OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance) + 1);
809 } /* lustre_put_super */
811 int ll_process_config_update(struct ll_sb_info *sbi, int clean)
813 struct obd_export *md_exp = sbi->ll_md_exp;
814 struct lustre_mount_data *lmd = sbi->ll_lmd;
815 struct llog_ctxt *ctxt;
816 struct config_llog_instance cfg;
817 char *profile = lmd->lmd_profile, *name = NULL;
818 int rc, namelen = 0, version;
824 CERROR("Client not mounted with zero-conf; cannot process "
829 rc = ldlm_cli_cancel_unused(md_exp->exp_obd->obd_namespace, NULL,
830 LDLM_FL_CONFIG_CHANGE, NULL);
832 CWARN("ldlm_cli_cancel_unused(mdc): %d\n", rc);
834 rc = obd_cancel_unused(sbi->ll_dt_exp, NULL, LDLM_FL_CONFIG_CHANGE,
837 CWARN("obd_cancel_unused(lov): %d\n", rc);
839 cfg.cfg_instance = sbi->ll_instance;
840 cfg.cfg_uuid = sbi->ll_sb_uuid;
841 cfg.cfg_local_nid = lmd->lmd_local_nid;
843 namelen = strlen(profile) + 20; /* -clean-######### */
844 OBD_ALLOC(name, namelen);
849 version = sbi->ll_config_version - 1;
850 sprintf(name, "%s-clean-%d", profile, version);
852 version = sbi->ll_config_version + 1;
853 sprintf(name, "%s-%d", profile, version);
856 CWARN("Applying configuration log %s\n", name);
858 ctxt = llog_get_context(&md_exp->exp_obd->obd_llogs,
859 LLOG_CONFIG_REPL_CTXT);
860 rc = class_config_process_llog(ctxt, name, &cfg);
862 sbi->ll_config_version = version;
863 CWARN("Finished applying configuration log %s: %d\n", name, rc);
865 if (rc == 0 && clean == 0) {
866 struct lov_desc desc;
868 valsize = sizeof(desc);
869 rc = obd_get_info(sbi->ll_dt_exp, strlen("lovdesc") + 1,
870 "lovdesc", &valsize, &desc);
872 rc = obd_init_ea_size(md_exp,
873 obd_size_diskmd(sbi->ll_dt_exp, NULL),
875 sizeof(struct llog_cookie)));
877 OBD_FREE(name, namelen);
881 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
883 struct inode *inode = NULL;
884 l_lock(&lock->l_resource->lr_namespace->ns_lock);
885 if (lock->l_ast_data) {
886 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
887 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
888 inode = igrab(lock->l_ast_data);
890 CERROR("DEBUG: l_ast_data %p is bogus: magic %x\n",
891 lock->l_ast_data, lli->lli_inode_magic);
894 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
898 int null_if_equal(struct ldlm_lock *lock, void *data)
900 if (data == lock->l_ast_data) {
901 lock->l_ast_data = NULL;
903 if (lock->l_req_mode != lock->l_granted_mode)
904 LDLM_ERROR(lock,"clearing inode with ungranted lock\n");
907 return LDLM_ITER_CONTINUE;
910 void ll_clear_inode(struct inode *inode)
913 struct ll_inode_info *lli = ll_i2info(inode);
914 struct ll_sb_info *sbi = ll_i2sbi(inode);
917 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
918 inode->i_generation, inode);
920 lli->lli_inode_magic = LLI_INODE_DEAD;
921 ll_inode2id(&id, inode);
923 clear_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &(ll_i2info(inode)->lli_flags));
924 md_change_cbdata(sbi->ll_md_exp, &id, null_if_equal, inode);
927 obd_change_cbdata(sbi->ll_dt_exp, lli->lli_smd,
928 null_if_equal, inode);
931 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
936 obd_free_memmd(sbi->ll_md_exp,
937 (struct lov_stripe_md **) &lli->lli_mea);
941 if (lli->lli_symlink_name) {
942 OBD_FREE(lli->lli_symlink_name,
943 strlen(lli->lli_symlink_name) + 1);
944 lli->lli_symlink_name = NULL;
950 /* If this inode has objects allocated to it (lsm != NULL), then the OST
951 * object(s) determine the file size and mtime. Otherwise, the MDS will
952 * keep these values until such a time that objects are allocated for it.
953 * We do the MDS operations first, as it is checking permissions for us.
954 * We don't to the MDS RPC if there is nothing that we want to store there,
955 * otherwise there is no harm in updating mtime/atime on the MDS if we are
956 * going to do an RPC anyways.
958 * If we are doing a truncate, we will send the mtime and ctime updates
959 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
960 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
963 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
965 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
966 struct ll_sb_info *sbi = ll_i2sbi(inode);
967 struct ptlrpc_request *request = NULL;
968 struct mdc_op_data *op_data;
969 int ia_valid = attr->ia_valid;
973 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu\n", inode->i_ino);
974 lprocfs_counter_incr(ll_i2sbi(inode)->ll_stats, LPROC_LL_SETATTR);
976 if (ia_valid & ATTR_SIZE) {
977 if (attr->ia_size > ll_file_maxbytes(inode)) {
978 CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
979 attr->ia_size, ll_file_maxbytes(inode));
983 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
986 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
987 if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
988 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
992 /* We mark all of the fields "set" so MDS/OST does not re-set them */
993 if (attr->ia_valid & ATTR_CTIME) {
994 attr->ia_ctime = CURRENT_TIME;
995 attr->ia_valid |= ATTR_CTIME_SET;
997 if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
998 attr->ia_atime = CURRENT_TIME;
999 attr->ia_valid |= ATTR_ATIME_SET;
1001 if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1002 attr->ia_mtime = CURRENT_TIME;
1003 attr->ia_valid |= ATTR_MTIME_SET;
1006 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1007 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1008 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1009 LTIME_S(CURRENT_TIME));
1011 attr->ia_valid &= ~ATTR_SIZE;
1013 /* If only OST attributes being set on objects, don't do MDS RPC.
1014 * In that case, we need to check permissions and update the local
1015 * inode ourselves so we can call obdo_from_inode() always. */
1016 if (ia_valid & (lsm ? ~(ATTR_SIZE | ATTR_FROM_OPEN /*| ATTR_RAW*/) : ~0)) {
1017 struct lustre_md md;
1019 OBD_ALLOC(op_data, sizeof(*op_data));
1020 if (op_data == NULL)
1022 ll_prepare_mdc_data(op_data, inode, NULL, NULL, 0, 0);
1024 rc = md_setattr(sbi->ll_md_exp, op_data,
1025 attr, NULL, 0, NULL, 0, &request);
1026 OBD_FREE(op_data, sizeof(*op_data));
1028 ptlrpc_req_finished(request);
1029 if (rc != -EPERM && rc != -EACCES)
1030 CERROR("md_setattr fails: rc = %d\n", rc);
1034 rc = mdc_req2lustre_md(sbi->ll_md_exp, request, 0,
1035 sbi->ll_dt_exp, &md);
1037 ptlrpc_req_finished(request);
1041 /* Won't invoke vmtruncate as we already cleared ATTR_SIZE,
1042 * but needed to set timestamps backwards on utime. */
1043 inode_setattr(inode, attr);
1044 ll_update_inode(inode, &md);
1045 ptlrpc_req_finished(request);
1047 if (!lsm || !S_ISREG(inode->i_mode)) {
1048 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1052 /* The OST doesn't check permissions, but the alternative is
1053 * a gratuitous RPC to the MDS. We already rely on the client
1054 * to do read/write/truncate permission checks, so is mtime OK?
1056 if (ia_valid & (ATTR_MTIME | ATTR_ATIME)) {
1057 /* from sys_utime() */
1058 if (!(ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET))) {
1059 if (current->fsuid != inode->i_uid &&
1060 (rc=ll_permission(inode,MAY_WRITE,NULL))!=0)
1063 /* from inode_change_ok() */
1064 if (current->fsuid != inode->i_uid &&
1065 !capable(CAP_FOWNER))
1070 /* Won't invoke vmtruncate, as we already cleared ATTR_SIZE */
1071 inode_setattr(inode, attr);
1074 /* We really need to get our PW lock before we change inode->i_size.
1075 * If we don't we can race with other i_size updaters on our node, like
1076 * ll_file_read. We can also race with i_size propogation to other
1077 * nodes through dirtying and writeback of final cached pages. This
1078 * last one is especially bad for racing o_append users on other
1080 if (ia_valid & ATTR_SIZE) {
1081 ldlm_policy_data_t policy = { .l_extent = {attr->ia_size,
1083 struct lustre_handle lockh = { 0 };
1084 int err, ast_flags = 0;
1085 /* XXX when we fix the AST intents to pass the discard-range
1086 * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
1088 if (attr->ia_size == 0)
1089 ast_flags = LDLM_AST_DISCARD_DATA;
1091 /* bug 1639: avoid write/truncate i_sem/DLM deadlock */
1092 LASSERT(atomic_read(&inode->i_sem.count) <= 0);
1094 UP_WRITE_I_ALLOC_SEM(inode);
1095 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy, &lockh,
1096 ast_flags, &ll_i2sbi(inode)->ll_seek_stime);
1097 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1098 DOWN_WRITE_I_ALLOC_SEM(inode);
1099 down(&inode->i_sem);
1101 down(&inode->i_sem);
1102 DOWN_WRITE_I_ALLOC_SEM(inode);
1107 rc = vmtruncate(inode, attr->ia_size);
1109 /* We need to drop the semaphore here, because this unlock may
1110 * result in a cancellation, which will need the i_sem */
1112 UP_WRITE_I_ALLOC_SEM(inode);
1113 /* unlock now as we don't mind others file lockers racing with
1114 * the mds updates below? */
1115 err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
1116 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1117 DOWN_WRITE_I_ALLOC_SEM(inode);
1118 down(&inode->i_sem);
1120 down(&inode->i_sem);
1121 DOWN_WRITE_I_ALLOC_SEM(inode);
1124 CERROR("ll_extent_unlock failed: %d\n", err);
1128 } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
1129 struct obdo *oa = NULL;
1131 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1132 inode->i_ino, LTIME_S(attr->ia_mtime));
1138 oa->o_id = lsm->lsm_object_id;
1139 oa->o_gr = lsm->lsm_object_gr;
1140 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1141 obdo_from_inode(oa, inode, OBD_MD_FLTYPE | OBD_MD_FLATIME |
1142 OBD_MD_FLMTIME | OBD_MD_FLCTIME);
1143 rc = obd_setattr(sbi->ll_dt_exp, oa, lsm, NULL);
1146 CERROR("obd_setattr fails: rc = %d\n", rc);
1151 int ll_setattr(struct dentry *de, struct iattr *attr)
1153 LASSERT(de->d_inode);
1154 return ll_setattr_raw(de->d_inode, attr);
1157 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1158 unsigned long max_age)
1160 struct ll_sb_info *sbi = ll_s2sbi(sb);
1161 struct obd_statfs obd_osfs;
1165 rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age);
1167 CERROR("mdc_statfs fails: rc = %d\n", rc);
1171 osfs->os_type = sb->s_magic;
1173 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1174 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1176 rc = obd_statfs(class_exp2obd(sbi->ll_dt_exp), &obd_osfs, max_age);
1178 CERROR("obd_statfs fails: rc = %d\n", rc);
1182 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1183 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1186 osfs->os_blocks = obd_osfs.os_blocks;
1187 osfs->os_bfree = obd_osfs.os_bfree;
1188 osfs->os_bavail = obd_osfs.os_bavail;
1190 /* If we don't have as many objects free on the OST as inodes
1191 * on the MDS, we reduce the total number of inodes to
1192 * compensate, so that the "inodes in use" number is correct.
1194 if (obd_osfs.os_ffree < osfs->os_ffree) {
1195 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1197 osfs->os_ffree = obd_osfs.os_ffree;
1203 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1205 struct obd_statfs osfs;
1208 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p\n", sb);
1209 lprocfs_counter_incr(ll_s2sbi(sb)->ll_stats, LPROC_LL_STAFS);
1211 /* For now we will always get up-to-date statfs values, but in the
1212 * future we may allow some amount of caching on the client (e.g.
1213 * from QOS or lprocfs updates). */
1214 rc = ll_statfs_internal(sb, &osfs, jiffies - 1);
1218 statfs_unpack(sfs, &osfs);
1220 if (sizeof(sfs->f_blocks) == 4) {
1221 while (osfs.os_blocks > ~0UL) {
1224 osfs.os_blocks >>= 1;
1225 osfs.os_bfree >>= 1;
1226 osfs.os_bavail >>= 1;
1230 sfs->f_blocks = osfs.os_blocks;
1231 sfs->f_bfree = osfs.os_bfree;
1232 sfs->f_bavail = osfs.os_bavail;
1237 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1239 struct ll_inode_info *lli = ll_i2info(inode);
1240 struct lov_stripe_md *lsm = md->lsm;
1241 struct mds_body *body = md->body;
1242 struct mea *mea = md->mea;
1243 struct posix_acl *ll_acl_access = md->acl_access;
1246 LASSERT((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1247 LASSERT((mea != NULL) == ((body->valid & OBD_MD_FLDIREA) != 0));
1250 LASSERT(lsm->lsm_object_gr > 0);
1251 if (lli->lli_smd == NULL) {
1253 lli->lli_maxbytes = lsm->lsm_maxbytes;
1254 if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1255 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1258 if (memcmp(lli->lli_smd, lsm, sizeof(*lsm))) {
1259 CERROR("lsm mismatch for inode %ld\n",
1261 CERROR("lli_smd:\n");
1262 dump_lsm(D_ERROR, lli->lli_smd);
1264 dump_lsm(D_ERROR, lsm);
1267 /* XXX FIXME -- We should decide on a safer (atomic) and
1268 * more elegant way to update the lsm */
1269 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1270 lli->lli_smd->lsm_oinfo[i].loi_id =
1271 lsm->lsm_oinfo[i].loi_id;
1272 lli->lli_smd->lsm_oinfo[i].loi_gr =
1273 lsm->lsm_oinfo[i].loi_gr;
1274 lli->lli_smd->lsm_oinfo[i].loi_ost_idx =
1275 lsm->lsm_oinfo[i].loi_ost_idx;
1276 lli->lli_smd->lsm_oinfo[i].loi_ost_gen =
1277 lsm->lsm_oinfo[i].loi_ost_gen;
1280 /* bug 2844 - limit i_blksize for broken user-space apps */
1281 LASSERTF(lsm->lsm_xfersize != 0, "%lu\n", lsm->lsm_xfersize);
1282 inode->i_blksize = min(lsm->lsm_xfersize, LL_MAX_BLKSIZE);
1283 if (lli->lli_smd != lsm)
1284 obd_free_memmd(ll_i2dtexp(inode), &lsm);
1288 if (lli->lli_mea == NULL) {
1291 if (memcmp(lli->lli_mea, mea, body->eadatasize)) {
1292 CERROR("mea mismatch for inode %lu\n",
1297 if (lli->lli_mea != mea)
1298 obd_free_memmd(ll_i2mdexp(inode),
1299 (struct lov_stripe_md **) &mea);
1302 if (body->valid & OBD_MD_FID)
1303 id_assign_fid(&lli->lli_id, &body->id1);
1305 if (body->valid & OBD_MD_FLID)
1306 id_ino(&lli->lli_id) = id_ino(&body->id1);
1308 if (body->valid & OBD_MD_FLGENER)
1309 id_gen(&lli->lli_id) = id_gen(&body->id1);
1311 spin_lock(&lli->lli_lock);
1312 if (ll_acl_access != NULL) {
1313 if (lli->lli_acl_access != NULL)
1314 posix_acl_release(lli->lli_acl_access);
1315 lli->lli_acl_access = ll_acl_access;
1317 spin_unlock(&lli->lli_lock);
1319 if (body->valid & OBD_MD_FLID)
1320 inode->i_ino = id_ino(&body->id1);
1321 if (body->valid & OBD_MD_FLGENER)
1322 inode->i_generation = id_gen(&body->id1);
1323 if (body->valid & OBD_MD_FLATIME)
1324 LTIME_S(inode->i_atime) = body->atime;
1325 if (body->valid & OBD_MD_FLMTIME &&
1326 body->mtime > LTIME_S(inode->i_mtime)) {
1327 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to %u\n",
1328 inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
1329 LTIME_S(inode->i_mtime) = body->mtime;
1331 if (body->valid & OBD_MD_FLCTIME &&
1332 body->ctime > LTIME_S(inode->i_ctime))
1333 LTIME_S(inode->i_ctime) = body->ctime;
1334 if (body->valid & OBD_MD_FLMODE) {
1335 inode->i_mode = (inode->i_mode & S_IFMT) |
1336 (body->mode & ~S_IFMT);
1338 if (body->valid & OBD_MD_FLTYPE) {
1339 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1340 (body->mode & S_IFMT);
1342 if (body->valid & OBD_MD_FLUID)
1343 inode->i_uid = body->uid;
1344 if (body->valid & OBD_MD_FLGID)
1345 inode->i_gid = body->gid;
1346 if (body->valid & OBD_MD_FLFLAGS)
1347 inode->i_flags = body->flags;
1348 if (body->valid & OBD_MD_FLNLINK)
1349 inode->i_nlink = body->nlink;
1350 if (body->valid & OBD_MD_FLRDEV)
1351 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1352 inode->i_rdev = body->rdev;
1354 inode->i_rdev = old_decode_dev(body->rdev);
1356 if (body->valid & OBD_MD_FLSIZE)
1357 inode->i_size = body->size;
1358 if (body->valid & OBD_MD_FLBLOCKS)
1359 inode->i_blocks = body->blocks;
1361 if (body->valid & OBD_MD_FLSIZE)
1362 set_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &lli->lli_flags);
1364 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1365 inode->i_dev = (kdev_t)id_group(&lli->lli_id);
1367 LASSERT(id_fid(&lli->lli_id) != 0);
1370 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
1371 static struct backing_dev_info ll_backing_dev_info = {
1372 .ra_pages = 0, /* No readahead */
1373 .memory_backed = 0, /* Does contribute to dirty memory */
1377 void ll_read_inode2(struct inode *inode, void *opaque)
1379 struct lustre_md *md = opaque;
1380 struct ll_inode_info *lli = ll_i2info(inode);
1383 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1384 inode->i_generation, inode);
1388 LASSERT(!lli->lli_smd);
1390 /* Core attributes from the MDS first. This is a new inode, and
1391 * the VFS doesn't zero times in the core inode so we have to do
1392 * it ourselves. They will be overwritten by either MDS or OST
1393 * attributes - we just need to make sure they aren't newer. */
1394 LTIME_S(inode->i_mtime) = 0;
1395 LTIME_S(inode->i_atime) = 0;
1396 LTIME_S(inode->i_ctime) = 0;
1399 ll_update_inode(inode, md);
1401 /* OIDEBUG(inode); */
1403 if (S_ISREG(inode->i_mode)) {
1404 inode->i_op = &ll_file_inode_operations;
1405 inode->i_fop = &ll_file_operations;
1406 inode->i_mapping->a_ops = &ll_aops;
1408 } else if (S_ISDIR(inode->i_mode)) {
1409 inode->i_op = &ll_dir_inode_operations;
1410 inode->i_fop = &ll_dir_operations;
1411 inode->i_mapping->a_ops = &ll_dir_aops;
1413 } else if (S_ISLNK(inode->i_mode)) {
1414 inode->i_op = &ll_fast_symlink_inode_operations;
1417 inode->i_op = &ll_special_inode_operations;
1419 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
1420 init_special_inode(inode, inode->i_mode,
1421 kdev_t_to_nr(inode->i_rdev));
1423 /* initializing backing dev info. */
1424 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
1426 init_special_inode(inode, inode->i_mode, inode->i_rdev);
1428 lli->ll_save_ifop = inode->i_fop;
1430 if (S_ISCHR(inode->i_mode))
1431 inode->i_fop = &ll_special_chr_inode_fops;
1432 else if (S_ISBLK(inode->i_mode))
1433 inode->i_fop = &ll_special_blk_inode_fops;
1434 else if (S_ISFIFO(inode->i_mode))
1435 inode->i_fop = &ll_special_fifo_inode_fops;
1436 else if (S_ISSOCK(inode->i_mode))
1437 inode->i_fop = &ll_special_sock_inode_fops;
1439 CWARN("saved %p, replaced with %p\n", lli->ll_save_ifop,
1442 if (lli->ll_save_ifop->owner) {
1443 CWARN("%p has owner %p\n", lli->ll_save_ifop,
1444 lli->ll_save_ifop->owner);
1450 void ll_delete_inode(struct inode *inode)
1452 struct ll_sb_info *sbi = ll_i2sbi(inode);
1453 struct lustre_id id;
1457 ll_inode2id(&id, inode);
1459 rc = md_delete_inode(sbi->ll_md_exp, &id);
1461 CERROR("md_delete_inode() failed, error %d\n",
1469 int ll_iocontrol(struct inode *inode, struct file *file,
1470 unsigned int cmd, unsigned long arg)
1472 struct ll_sb_info *sbi = ll_i2sbi(inode);
1473 struct ptlrpc_request *req = NULL;
1478 case EXT3_IOC_GETFLAGS: {
1479 struct lustre_id id;
1480 __u64 valid = OBD_MD_FLFLAGS;
1481 struct mds_body *body;
1483 ll_inode2id(&id, inode);
1484 rc = md_getattr(sbi->ll_md_exp, &id, valid, NULL, 0, 0, &req);
1486 CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1490 body = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*body));
1492 if (body->flags & S_APPEND)
1493 flags |= EXT3_APPEND_FL;
1494 if (body->flags & S_IMMUTABLE)
1495 flags |= EXT3_IMMUTABLE_FL;
1496 if (body->flags & S_NOATIME)
1497 flags |= EXT3_NOATIME_FL;
1499 ptlrpc_req_finished (req);
1501 RETURN(put_user(flags, (int *)arg));
1503 case EXT3_IOC_SETFLAGS: {
1504 struct mdc_op_data *op_data;
1507 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1509 if (get_user(flags, (int *)arg))
1516 OBD_ALLOC(op_data, sizeof(*op_data));
1517 if (op_data == NULL) {
1521 ll_prepare_mdc_data(op_data, inode, NULL, NULL, 0, 0);
1523 memset(&attr, 0x0, sizeof(attr));
1524 attr.ia_attr_flags = flags;
1525 attr.ia_valid |= ATTR_ATTR_FLAG;
1527 rc = md_setattr(sbi->ll_md_exp, op_data,
1528 &attr, NULL, 0, NULL, 0, &req);
1529 OBD_FREE(op_data, sizeof(*op_data));
1531 ptlrpc_req_finished(req);
1532 if (rc != -EPERM && rc != -EACCES)
1533 CERROR("md_setattr fails: rc = %d\n", rc);
1537 ptlrpc_req_finished(req);
1539 oa->o_id = lsm->lsm_object_id;
1540 oa->o_gr = lsm->lsm_object_gr;
1541 oa->o_flags = flags;
1542 oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS | OBD_MD_FLGROUP;
1544 rc = obd_setattr(sbi->ll_dt_exp, oa, lsm, NULL);
1547 if (rc != -EPERM && rc != -EACCES)
1548 CERROR("md_setattr fails: rc = %d\n", rc);
1552 if (flags & EXT3_APPEND_FL)
1553 inode->i_flags |= S_APPEND;
1555 inode->i_flags &= ~S_APPEND;
1556 if (flags & EXT3_IMMUTABLE_FL)
1557 inode->i_flags |= S_IMMUTABLE;
1559 inode->i_flags &= ~S_IMMUTABLE;
1560 if (flags & EXT3_NOATIME_FL)
1561 inode->i_flags |= S_NOATIME;
1563 inode->i_flags &= ~S_NOATIME;
1574 /* this is only called in the case of forced umount. */
1575 void ll_umount_begin(struct super_block *sb)
1577 struct ll_sb_info *sbi = ll_s2sbi(sb);
1578 struct obd_device *obd;
1579 struct obd_ioctl_data ioc_data = { 0 };
1582 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1583 sb->s_count, atomic_read(&sb->s_active));
1585 obd = class_exp2obd(sbi->ll_md_exp);
1587 CERROR("Invalid MDC connection handle "LPX64"\n",
1588 sbi->ll_md_exp->exp_handle.h_cookie);
1592 obd->obd_no_recov = 1;
1593 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
1594 sizeof(ioc_data), &ioc_data, NULL);
1596 obd = class_exp2obd(sbi->ll_dt_exp);
1598 CERROR("Invalid LOV connection handle "LPX64"\n",
1599 sbi->ll_dt_exp->exp_handle.h_cookie);
1604 obd->obd_no_recov = 1;
1605 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
1606 sizeof(ioc_data), &ioc_data, NULL);
1609 * really, we'd like to wait until there are no requests outstanding,
1610 * and then continue. For now, we just invalidate the requests,
1611 * schedule, and hope.
1618 int ll_prep_inode(struct obd_export *dt_exp, struct obd_export *md_exp,
1619 struct inode **inode, struct ptlrpc_request *req,
1620 int offset, struct super_block *sb)
1622 struct lustre_md md;
1625 rc = mdc_req2lustre_md(md_exp, req, offset, dt_exp, &md);
1630 ll_update_inode(*inode, &md);
1633 *inode = ll_iget(sb, id_ino(&md.body->id1), &md);
1634 if (*inode == NULL || is_bad_inode(*inode)) {
1635 /* free the lsm if we allocated one above */
1637 obd_free_memmd(dt_exp, &md.lsm);
1639 obd_free_memmd(md_exp,
1640 (struct lov_stripe_md**)&md.mea);
1642 CERROR("new_inode -fatal: rc %d\n", rc);
1649 int ll_get_fid(struct obd_export *exp, struct lustre_id *idp,
1650 char *filename, struct lustre_id *ret)
1652 struct ptlrpc_request *request = NULL;
1653 struct mds_body *body;
1656 rc = md_getattr_lock(exp, idp, filename, strlen(filename) + 1,
1657 OBD_MD_FID, 0, &request);
1659 CDEBUG(D_INFO, "md_getattr_lock failed on %s: rc %d\n",
1664 body = lustre_msg_buf(request->rq_repmsg, 0, sizeof(*body));
1665 LASSERT(body != NULL);
1666 LASSERT_REPSWABBED(request, 0);
1669 ptlrpc_req_finished(request);