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, int async)
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);
146 obd_set_info(obd->obd_self_export, strlen("async"), "async",
147 sizeof(async), &async);
149 if (security == NULL)
152 err = obd_set_info(obd->obd_self_export, strlen("sec"), "sec",
153 strlen(security), security);
155 CERROR("LMV %s: failed to set security %s, err %d\n",
160 err = obd_set_info(obd->obd_self_export, strlen("nllu"), "nllu",
161 sizeof(__u32) * 2, nllu);
163 CERROR("LMV %s: failed to set NLLU, err %d\n",
168 if (proc_lustre_fs_root) {
169 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
172 CERROR("could not register mount in /proc/lustre");
175 err = obd_connect(&md_conn, obd, &sbi->ll_sb_uuid, OBD_OPT_REAL_CLIENT);
177 CERROR("An MDS (lmv %s) is performing recovery, of which this"
178 " client is not a part. Please wait for recovery to "
179 "complete, abort, or time out.\n", lmv);
182 CERROR("cannot connect to %s: rc = %d\n", lmv, err);
185 sbi->ll_md_exp = class_conn2export(&md_conn);
186 err = obd_statfs(obd, &osfs, jiffies - HZ);
190 if (!osfs.os_bsize) {
191 CERROR("Invalid block size is detected.");
195 sb->s_magic = LL_SUPER_MAGIC;
196 sb->s_blocksize = osfs.os_bsize;
197 sb->s_blocksize_bits = log2(osfs.os_bsize);
198 sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
200 devno = get_uuid2int(sbi->ll_md_exp->exp_obd->obd_uuid.uuid,
201 strlen(sbi->ll_md_exp->exp_obd->obd_uuid.uuid));
205 obd = class_name2obd(lov);
207 CERROR("OSC %s: not setup or attached\n", lov);
210 obd_set_info(obd->obd_self_export, strlen("async"), "async",
211 sizeof(async), &async);
213 err = obd_connect(&dt_conn, obd, &sbi->ll_sb_uuid, OBD_OPT_REAL_CLIENT);
215 CERROR("An OST (lov %s) is performing recovery, of which this"
216 " client is not a part. Please wait for recovery to "
217 "complete, abort, or time out.\n", lov);
220 CERROR("cannot connect to %s: rc = %d\n", lov, err);
223 sbi->ll_dt_exp = class_conn2export(&dt_conn);
225 err = lustre_init_dt_desc(sbi);
227 int mdsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
228 obd_init_ea_size(sbi->ll_md_exp, mdsize,
229 sbi->ll_dt_desc.ld_tgt_count *
230 sizeof(struct llog_cookie));
233 err = md_getstatus(sbi->ll_md_exp, &sbi->ll_rootid);
235 CERROR("cannot mds_connect: rc = %d\n", err);
238 CDEBUG(D_SUPER, "rootid "DLID4"\n", OLID4(&sbi->ll_rootid));
240 sb->s_op = &lustre_super_operations;
243 * XXX: move this to after cbd setup? */
244 err = md_getattr(sbi->ll_md_exp, &sbi->ll_rootid,
245 (OBD_MD_FLNOTOBD | OBD_MD_FLBLOCKS | OBD_MD_FID), NULL, 0,
248 CERROR("md_getattr failed for root: rc = %d\n", err);
252 err = mdc_req2lustre_md(sbi->ll_md_exp, request, 0,
253 sbi->ll_dt_exp, &md);
255 CERROR("failed to understand root inode md: rc = %d\n", err);
256 ptlrpc_req_finished(request);
260 LASSERT(id_ino(&sbi->ll_rootid) != 0);
261 root = ll_iget(sb, id_ino(&sbi->ll_rootid), &md);
263 ptlrpc_req_finished(request);
265 if (root == NULL || is_bad_inode(root)) {
266 CERROR("lustre_lite: bad iget4 for root\n");
267 GOTO(out_root, err = -EBADF);
270 err = ll_close_thread_start(&sbi->ll_lcq);
272 CERROR("cannot start close thread: rc %d\n", err);
276 ll_gns_add_timer(sbi);
278 /* making vm readahead 0 for 2.4.x. In the case of 2.6.x,
279 backing dev info assigned to inode mapping is used for
280 determining maximal readahead. */
281 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0)) && \
282 !defined(KERNEL_HAS_AS_MAX_READAHEAD)
283 /* bug 2805 - set VM readahead to zero */
284 vm_max_readahead = vm_min_readahead = 0;
287 sb->s_root = d_alloc_root(root);
288 sb->s_root->d_op = &ll_d_ops;
291 CWARN("Enabling PDIROPS\n");
292 sb->s_flags |= S_PDIROPS;
300 obd_disconnect(sbi->ll_dt_exp, 0);
302 obd_disconnect(sbi->ll_md_exp, 0);
304 lprocfs_unregister_mountpoint(sbi);
308 void lustre_common_put_super(struct super_block *sb)
310 struct ll_sb_info *sbi = ll_s2sbi(sb);
311 struct hlist_node *tmp, *next;
314 ll_gns_del_timer(sbi);
315 ll_close_thread_shutdown(sbi->ll_lcq);
317 list_del(&sbi->ll_conn_chain);
318 obd_disconnect(sbi->ll_dt_exp, 0);
320 lprocfs_unregister_mountpoint(sbi);
321 if (sbi->ll_proc_root) {
322 lprocfs_remove(sbi->ll_proc_root);
323 sbi->ll_proc_root = NULL;
326 obd_disconnect(sbi->ll_md_exp, 0);
328 // We do this to get rid of orphaned dentries. That is not really trw.
329 hlist_for_each_safe(tmp, next, &sbi->ll_orphan_dentry_list) {
330 struct dentry *dentry = hlist_entry(tmp, struct dentry, d_hash);
331 CWARN("orphan dentry %.*s (%p->%p) at unmount\n",
332 dentry->d_name.len, dentry->d_name.name, dentry, next);
333 shrink_dcache_parent(dentry);
338 char *ll_read_opt(const char *opt, char *data)
344 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
345 if (strncmp(opt, data, strlen(opt)))
347 if ((value = strchr(data, '=')) == NULL)
351 OBD_ALLOC(retval, strlen(value) + 1);
353 CERROR("out of memory!\n");
357 memcpy(retval, value, strlen(value)+1);
358 CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
362 int ll_set_opt(const char *opt, char *data, int fl)
366 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
367 if (strncmp(opt, data, strlen(opt)))
373 void ll_options(char *options, char **lov, char **lmv, char **sec,
374 int *async, int *flags)
377 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
378 char *opt_ptr = options;
388 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
389 for (this_char = strtok (options, ",");
391 this_char = strtok (NULL, ",")) {
393 while ((this_char = strsep (&opt_ptr, ",")) != NULL) {
395 CDEBUG(D_SUPER, "this_char %s\n", this_char);
396 if (!*lov && (*lov = ll_read_opt("osc", this_char)))
398 if (!*lmv && (*lmv = ll_read_opt("mdc", this_char)))
400 if (!strncmp(this_char, "lasync", strlen("lasync"))) {
404 if (!*sec && (*sec = ll_read_opt("sec", this_char)))
406 if (!(*flags & LL_SBI_NOLCK) &&
407 ((*flags) = (*flags) |
408 ll_set_opt("nolock", this_char,
416 void ll_lli_init(struct ll_inode_info *lli)
418 sema_init(&lli->lli_open_sem, 1);
419 sema_init(&lli->lli_size_sem, 1);
421 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
422 spin_lock_init(&lli->lli_lock);
423 INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
424 lli->lli_inode_magic = LLI_INODE_MAGIC;
425 memset(&lli->lli_id, 0, sizeof(lli->lli_id));
426 sema_init(&lli->lli_och_sem, 1);
427 lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
428 lli->lli_mds_exec_och = NULL;
429 lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
430 lli->lli_open_fd_exec_count = 0;
433 int ll_fill_super(struct super_block *sb, void *data, int silent)
435 struct ll_sb_info *sbi;
440 __u32 nllu[2] = { 99, 99 };
443 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
445 sbi = lustre_init_sbi(sb);
449 sbi->ll_flags |= LL_SBI_READAHEAD;
450 ll_options(data, &lov, &lmv, &sec, &async, &sbi->ll_flags);
454 GOTO(out, err = -EINVAL);
459 GOTO(out, err = -EINVAL);
462 err = lustre_common_fill_super(sb, lmv, lov, sec, nllu, async);
469 OBD_FREE(sec, strlen(sec) + 1);
471 OBD_FREE(lmv, strlen(lmv) + 1);
473 OBD_FREE(lov, strlen(lov) + 1);
475 } /* ll_read_super */
477 static int lustre_process_log(struct lustre_mount_data *lmd, char *profile,
478 struct config_llog_instance *cfg, int allow_recov)
480 struct lustre_cfg lcfg;
481 struct portals_cfg pcfg;
482 char *peer = "MDS_PEER_UUID";
483 struct obd_device *obd;
484 struct lustre_handle md_conn = {0, };
485 struct obd_export *exp;
486 char *name = "mdc_dev";
488 struct obd_uuid lmv_uuid;
489 struct llog_ctxt *ctxt;
493 if (lmd_bad_magic(lmd))
496 generate_random_uuid(uuid);
497 class_uuid_unparse(uuid, &lmv_uuid);
499 if (lmd->lmd_local_nid) {
500 PCFG_INIT(pcfg, NAL_CMD_REGISTER_MYNID);
501 pcfg.pcfg_nal = lmd->lmd_nal;
502 pcfg.pcfg_nid = lmd->lmd_local_nid;
503 rc = libcfs_nal_cmd(&pcfg);
508 if (lmd->lmd_nal == SOCKNAL ||
509 lmd->lmd_nal == OPENIBNAL ||
510 lmd->lmd_nal == IIBNAL ||
511 lmd->lmd_nal == VIBNAL ||
512 lmd->lmd_nal == RANAL) {
513 PCFG_INIT(pcfg, NAL_CMD_ADD_PEER);
514 pcfg.pcfg_nal = lmd->lmd_nal;
515 pcfg.pcfg_nid = lmd->lmd_server_nid;
516 pcfg.pcfg_id = lmd->lmd_server_ipaddr;
517 pcfg.pcfg_misc = lmd->lmd_port;
518 rc = libcfs_nal_cmd(&pcfg);
523 LCFG_INIT(lcfg, LCFG_ADD_UUID, name);
524 lcfg.lcfg_nid = lmd->lmd_server_nid;
525 lcfg.lcfg_inllen1 = strlen(peer) + 1;
526 lcfg.lcfg_inlbuf1 = peer;
527 lcfg.lcfg_nal = lmd->lmd_nal;
528 rc = class_process_config(&lcfg);
530 GOTO(out_del_conn, rc);
532 LCFG_INIT(lcfg, LCFG_ATTACH, name);
533 lcfg.lcfg_inlbuf1 = "mdc";
534 lcfg.lcfg_inllen1 = strlen(lcfg.lcfg_inlbuf1) + 1;
535 lcfg.lcfg_inlbuf2 = lmv_uuid.uuid;
536 lcfg.lcfg_inllen2 = strlen(lcfg.lcfg_inlbuf2) + 1;
537 err = class_process_config(&lcfg);
539 GOTO(out_del_uuid, rc);
541 LCFG_INIT(lcfg, LCFG_SETUP, name);
542 lcfg.lcfg_inlbuf1 = lmd->lmd_mds;
543 lcfg.lcfg_inllen1 = strlen(lcfg.lcfg_inlbuf1) + 1;
544 lcfg.lcfg_inlbuf2 = peer;
545 lcfg.lcfg_inllen2 = strlen(lcfg.lcfg_inlbuf2) + 1;
546 rc = class_process_config(&lcfg);
548 GOTO(out_detach, rc);
550 obd = class_name2obd(name);
552 GOTO(out_cleanup, rc = -EINVAL);
554 rc = obd_set_info(obd->obd_self_export, strlen("sec"), "sec",
555 strlen(lmd->lmd_security), lmd->lmd_security);
557 GOTO(out_cleanup, rc);
559 /* Disable initial recovery on this import */
560 rc = obd_set_info(obd->obd_self_export,
561 strlen("initial_recov"), "initial_recov",
562 sizeof(allow_recov), &allow_recov);
564 GOTO(out_cleanup, rc);
566 rc = obd_connect(&md_conn, obd, &lmv_uuid, 0);
568 CERROR("cannot connect to %s: rc = %d\n", lmd->lmd_mds, rc);
569 GOTO(out_cleanup, rc);
572 exp = class_conn2export(&md_conn);
574 ctxt = llog_get_context(&exp->exp_obd->obd_llogs,LLOG_CONFIG_REPL_CTXT);
575 rc = class_config_process_llog(ctxt, profile, cfg);
577 CERROR("class_config_process_llog failed: rc = %d\n", rc);
579 rc = obd_disconnect(exp, 0);
583 LCFG_INIT(lcfg, LCFG_CLEANUP, name);
584 err = class_process_config(&lcfg);
589 LCFG_INIT(lcfg, LCFG_DETACH, name);
590 err = class_process_config(&lcfg);
595 LCFG_INIT(lcfg, LCFG_DEL_UUID, name);
596 lcfg.lcfg_inllen1 = strlen(peer) + 1;
597 lcfg.lcfg_inlbuf1 = peer;
598 err = class_process_config(&lcfg);
601 if (lmd->lmd_nal == SOCKNAL ||
602 lmd->lmd_nal == OPENIBNAL ||
603 lmd->lmd_nal == IIBNAL ||
604 lmd->lmd_nal == VIBNAL ||
605 lmd->lmd_nal == RANAL) {
608 PCFG_INIT(pcfg, NAL_CMD_DEL_PEER);
609 pcfg.pcfg_nal = lmd->lmd_nal;
610 pcfg.pcfg_nid = lmd->lmd_server_nid;
611 pcfg.pcfg_flags = 1; /* single_share */
612 err2 = libcfs_nal_cmd(&pcfg);
625 static void lustre_manual_cleanup(struct ll_sb_info *sbi)
627 struct lustre_cfg lcfg;
628 struct obd_device *obd;
631 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) != NULL)
635 LCFG_INIT(lcfg, LCFG_CLEANUP, obd->obd_name);
636 err = class_process_config(&lcfg);
638 CERROR("cleanup failed: %s\n", obd->obd_name);
642 LCFG_INIT(lcfg, LCFG_DETACH, obd->obd_name);
643 err = class_process_config(&lcfg);
645 CERROR("detach failed: %s\n", obd->obd_name);
650 if (sbi->ll_lmd != NULL)
651 class_del_profile(sbi->ll_lmd->lmd_profile);
654 int lustre_fill_super(struct super_block *sb, void *data, int silent)
656 struct lustre_mount_data * lmd = data;
657 char *lov = NULL, *lmv = NULL;
658 struct ll_sb_info *sbi;
662 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
663 if (lmd_bad_magic(lmd))
666 sbi = lustre_init_sbi(sb);
670 sbi->ll_flags |= LL_SBI_READAHEAD;
672 if (lmd->lmd_profile) {
673 struct lustre_profile *lprof;
674 struct config_llog_instance cfg;
677 if (lmd->lmd_mds[0] == '\0') {
678 CERROR("no mds name\n");
679 GOTO(out_free, err = -EINVAL);
681 lmd->lmd_security[sizeof(lmd->lmd_security) - 1] = 0;
683 OBD_ALLOC(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
684 if (sbi->ll_lmd == NULL)
685 GOTO(out_free, err = -ENOMEM);
686 memcpy(sbi->ll_lmd, lmd, sizeof(*lmd));
688 /* generate a string unique to this super, let's try
689 the address of the super itself.*/
690 len = (sizeof(sb) * 2) + 1;
691 OBD_ALLOC(sbi->ll_instance, len);
692 if (sbi->ll_instance == NULL)
693 GOTO(out_free, err = -ENOMEM);
694 sprintf(sbi->ll_instance, "%p", sb);
696 cfg.cfg_instance = sbi->ll_instance;
697 cfg.cfg_uuid = sbi->ll_sb_uuid;
698 cfg.cfg_local_nid = lmd->lmd_local_nid;
699 err = lustre_process_log(lmd, lmd->lmd_profile, &cfg, 0);
701 CERROR("Unable to process log: %s\n", lmd->lmd_profile);
705 lprof = class_get_profile(lmd->lmd_profile);
707 CERROR("No profile found: %s\n", lmd->lmd_profile);
708 GOTO(out_free, err = -EINVAL);
711 OBD_FREE(lov, strlen(lov) + 1);
712 OBD_ALLOC(lov, strlen(lprof->lp_lov) +
713 strlen(sbi->ll_instance) + 2);
714 sprintf(lov, "%s-%s", lprof->lp_lov, sbi->ll_instance);
717 OBD_FREE(lmv, strlen(lmv) + 1);
718 OBD_ALLOC(lmv, strlen(lprof->lp_lmv) +
719 strlen(sbi->ll_instance) + 2);
720 sprintf(lmv, "%s-%s", lprof->lp_lmv, sbi->ll_instance);
725 GOTO(out_free, err = -EINVAL);
730 GOTO(out_free, err = -EINVAL);
733 err = lustre_common_fill_super(sb, lmv, lov, lmd->lmd_security,
734 &lmd->lmd_nllu, lmd->lmd_async);
741 OBD_FREE(lmv, strlen(lmv) + 1);
743 OBD_FREE(lov, strlen(lov) + 1);
749 int len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean")+1;
752 if (sbi->ll_instance != NULL) {
753 struct lustre_mount_data *lmd = sbi->ll_lmd;
754 struct config_llog_instance cfg;
757 cfg.cfg_instance = sbi->ll_instance;
758 cfg.cfg_uuid = sbi->ll_sb_uuid;
760 OBD_ALLOC(cl_prof, len);
761 sprintf(cl_prof, "%s-clean", lmd->lmd_profile);
762 err = lustre_process_log(lmd, cl_prof, &cfg, 0);
764 CERROR("Unable to process log: %s\n", cl_prof);
765 lustre_manual_cleanup(sbi);
767 OBD_FREE(cl_prof, len);
768 OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance) + 1);
770 OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
774 } /* lustre_fill_super */
776 void lustre_put_super(struct super_block *sb)
778 struct obd_device *obd;
779 struct ll_sb_info *sbi = ll_s2sbi(sb);
780 int force_umount = 0;
783 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
784 obd = class_exp2obd(sbi->ll_md_exp);
786 force_umount = obd->obd_no_recov;
789 lustre_common_put_super(sb);
790 if (sbi->ll_lmd != NULL) {
792 int len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean")+1;
794 struct config_llog_instance cfg;
797 CERROR("force umount, doing manual cleanup\n");
798 lustre_manual_cleanup(sbi);
802 cfg.cfg_instance = sbi->ll_instance;
803 cfg.cfg_uuid = sbi->ll_sb_uuid;
805 OBD_ALLOC(cl_prof, len);
806 sprintf(cl_prof, "%s-clean", sbi->ll_lmd->lmd_profile);
807 err = lustre_process_log(sbi->ll_lmd, cl_prof, &cfg, 0);
809 CERROR("Unable to process log: %s, doing manual cleanup"
811 lustre_manual_cleanup(sbi);
814 OBD_FREE(cl_prof, len);
816 OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
817 OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance) + 1);
823 } /* lustre_put_super */
825 int ll_process_config_update(struct ll_sb_info *sbi, int clean)
827 struct obd_export *md_exp = sbi->ll_md_exp;
828 struct lustre_mount_data *lmd = sbi->ll_lmd;
829 struct llog_ctxt *ctxt;
830 struct config_llog_instance cfg;
831 char *profile = lmd->lmd_profile, *name = NULL;
832 int rc, namelen = 0, version;
838 CERROR("Client not mounted with zero-conf; cannot process "
843 rc = ldlm_cli_cancel_unused(md_exp->exp_obd->obd_namespace, NULL,
844 LDLM_FL_CONFIG_CHANGE, NULL);
846 CWARN("ldlm_cli_cancel_unused(mdc): %d\n", rc);
848 rc = obd_cancel_unused(sbi->ll_dt_exp, NULL, LDLM_FL_CONFIG_CHANGE,
851 CWARN("obd_cancel_unused(lov): %d\n", rc);
853 cfg.cfg_instance = sbi->ll_instance;
854 cfg.cfg_uuid = sbi->ll_sb_uuid;
855 cfg.cfg_local_nid = lmd->lmd_local_nid;
857 namelen = strlen(profile) + 20; /* -clean-######### */
858 OBD_ALLOC(name, namelen);
863 version = sbi->ll_config_version - 1;
864 sprintf(name, "%s-clean-%d", profile, version);
866 version = sbi->ll_config_version + 1;
867 sprintf(name, "%s-%d", profile, version);
870 CWARN("Applying configuration log %s\n", name);
872 ctxt = llog_get_context(&md_exp->exp_obd->obd_llogs,
873 LLOG_CONFIG_REPL_CTXT);
874 rc = class_config_process_llog(ctxt, name, &cfg);
876 sbi->ll_config_version = version;
877 CWARN("Finished applying configuration log %s: %d\n", name, rc);
879 if (rc == 0 && clean == 0) {
880 struct lov_desc desc;
882 valsize = sizeof(desc);
883 rc = obd_get_info(sbi->ll_dt_exp, strlen("lovdesc") + 1,
884 "lovdesc", &valsize, &desc);
886 rc = obd_init_ea_size(md_exp,
887 obd_size_diskmd(sbi->ll_dt_exp, NULL),
889 sizeof(struct llog_cookie)));
891 OBD_FREE(name, namelen);
895 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
897 struct inode *inode = NULL;
898 l_lock(&lock->l_resource->lr_namespace->ns_lock);
899 if (lock->l_ast_data) {
900 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
901 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
902 inode = igrab(lock->l_ast_data);
904 inode = lock->l_ast_data;
905 CDEBUG(inode->i_state & I_FREEING ? D_INFO : D_WARNING,
906 "l_ast_data %p is bogus: magic %0x8\n",
907 lock->l_ast_data, lli->lli_inode_magic);
911 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
915 int null_if_equal(struct ldlm_lock *lock, void *data)
917 if (data == lock->l_ast_data) {
918 lock->l_ast_data = NULL;
920 if (lock->l_req_mode != lock->l_granted_mode)
921 LDLM_ERROR(lock,"clearing inode with ungranted lock\n");
924 return LDLM_ITER_CONTINUE;
927 void ll_clear_inode(struct inode *inode)
930 struct ll_inode_info *lli = ll_i2info(inode);
931 struct ll_sb_info *sbi = ll_i2sbi(inode);
934 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
935 inode->i_generation, inode);
937 lli->lli_inode_magic = LLI_INODE_DEAD;
938 ll_inode2id(&id, inode);
940 clear_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &(ll_i2info(inode)->lli_flags));
941 md_change_cbdata(sbi->ll_md_exp, &id, null_if_equal, inode);
943 LASSERT(!lli->lli_open_fd_write_count);
944 LASSERT(!lli->lli_open_fd_read_count);
945 LASSERT(!lli->lli_open_fd_exec_count);
947 if (lli->lli_mds_write_och)
948 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_WRITE);
949 if (lli->lli_mds_exec_och)
950 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_EXEC);
951 if (lli->lli_mds_read_och)
952 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_READ);
955 obd_change_cbdata(sbi->ll_dt_exp, lli->lli_smd,
956 null_if_equal, inode);
959 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
964 obd_free_memmd(sbi->ll_md_exp,
965 (struct lov_stripe_md **) &lli->lli_mea);
969 if (lli->lli_symlink_name) {
970 OBD_FREE(lli->lli_symlink_name,
971 strlen(lli->lli_symlink_name) + 1);
972 lli->lli_symlink_name = NULL;
974 lli->lli_inode_magic = LLI_INODE_DEAD;
979 /* If this inode has objects allocated to it (lsm != NULL), then the OST
980 * object(s) determine the file size and mtime. Otherwise, the MDS will
981 * keep these values until such a time that objects are allocated for it.
982 * We do the MDS operations first, as it is checking permissions for us.
983 * We don't to the MDS RPC if there is nothing that we want to store there,
984 * otherwise there is no harm in updating mtime/atime on the MDS if we are
985 * going to do an RPC anyways.
987 * If we are doing a truncate, we will send the mtime and ctime updates
988 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
989 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
992 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
994 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
995 struct ll_sb_info *sbi = ll_i2sbi(inode);
996 struct ptlrpc_request *request = NULL;
997 struct mdc_op_data *op_data;
998 int ia_valid = attr->ia_valid;
1002 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu\n", inode->i_ino);
1003 lprocfs_counter_incr(ll_i2sbi(inode)->ll_stats, LPROC_LL_SETATTR);
1005 if (ia_valid & ATTR_SIZE) {
1006 if (attr->ia_size > ll_file_maxbytes(inode)) {
1007 CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1008 attr->ia_size, ll_file_maxbytes(inode));
1012 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1015 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1016 if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1017 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1021 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1022 if (attr->ia_valid & ATTR_CTIME) {
1023 attr->ia_ctime = CURRENT_TIME;
1024 attr->ia_valid |= ATTR_CTIME_SET;
1026 if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1027 attr->ia_atime = CURRENT_TIME;
1028 attr->ia_valid |= ATTR_ATIME_SET;
1030 if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1031 attr->ia_mtime = CURRENT_TIME;
1032 attr->ia_valid |= ATTR_MTIME_SET;
1035 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1036 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1037 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1038 LTIME_S(CURRENT_TIME));
1040 attr->ia_valid &= ~ATTR_SIZE;
1042 /* If only OST attributes being set on objects, don't do MDS RPC.
1043 * In that case, we need to check permissions and update the local
1044 * inode ourselves so we can call obdo_from_inode() always. */
1045 if (ia_valid & (lsm ? ~(ATTR_SIZE | ATTR_FROM_OPEN /*| ATTR_RAW*/) : ~0)) {
1046 struct lustre_md md;
1049 OBD_ALLOC(op_data, sizeof(*op_data));
1050 if (op_data == NULL)
1052 ll_prepare_mdc_data(op_data, inode, NULL, NULL, 0, 0);
1054 rc = md_setattr(sbi->ll_md_exp, op_data,
1055 attr, NULL, 0, NULL, 0, &request);
1056 OBD_FREE(op_data, sizeof(*op_data));
1058 ptlrpc_req_finished(request);
1059 if (rc != -EPERM && rc != -EACCES)
1060 CERROR("md_setattr fails: rc = %d\n", rc);
1064 rc = mdc_req2lustre_md(sbi->ll_md_exp, request, 0,
1065 sbi->ll_dt_exp, &md);
1067 ptlrpc_req_finished(request);
1071 /* We call inode_setattr to adjust timestamps, but we first
1072 * clear ATTR_SIZE to avoid invoking vmtruncate.
1074 * NB: ATTR_SIZE will only be set at this point if the size
1075 * resides on the MDS, ie, this file has no objects. */
1076 save_valid = attr->ia_valid;
1077 attr->ia_valid &= ~ATTR_SIZE;
1078 inode_setattr(inode, attr);
1079 attr->ia_valid = save_valid;
1081 ll_update_inode(inode, &md);
1082 ptlrpc_req_finished(request);
1084 if (!lsm || !S_ISREG(inode->i_mode)) {
1085 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1089 /* The OST doesn't check permissions, but the alternative is
1090 * a gratuitous RPC to the MDS. We already rely on the client
1091 * to do read/write/truncate permission checks, so is mtime OK?
1093 if (ia_valid & (ATTR_MTIME | ATTR_ATIME)) {
1094 /* from sys_utime() */
1095 if (!(ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET))) {
1096 if (current->fsuid != inode->i_uid &&
1097 (rc=ll_permission(inode,MAY_WRITE,NULL))!=0)
1100 /* from inode_change_ok() */
1101 if (current->fsuid != inode->i_uid &&
1102 !capable(CAP_FOWNER))
1107 /* Won't invoke vmtruncate, as we already cleared ATTR_SIZE */
1108 inode_setattr(inode, attr);
1111 /* We really need to get our PW lock before we change inode->i_size.
1112 * If we don't we can race with other i_size updaters on our node, like
1113 * ll_file_read. We can also race with i_size propogation to other
1114 * nodes through dirtying and writeback of final cached pages. This
1115 * last one is especially bad for racing o_append users on other
1117 if (ia_valid & ATTR_SIZE) {
1118 ldlm_policy_data_t policy = { .l_extent = {attr->ia_size,
1120 struct lustre_handle lockh = { 0 };
1121 struct ll_inode_info *lli = ll_i2info(inode);
1122 int err, ast_flags = 0;
1123 /* XXX when we fix the AST intents to pass the discard-range
1124 * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
1126 if (attr->ia_size == 0)
1127 ast_flags = LDLM_AST_DISCARD_DATA;
1129 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy, &lockh,
1130 ast_flags, &ll_i2sbi(inode)->ll_seek_stime);
1135 down(&lli->lli_size_sem);
1136 rc = vmtruncate(inode, attr->ia_size);
1138 LASSERT(atomic_read(&lli->lli_size_sem.count) <= 0);
1139 up(&lli->lli_size_sem);
1142 err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
1144 CERROR("ll_extent_unlock failed: %d\n", err);
1148 } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
1149 struct obdo *oa = NULL;
1151 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1152 inode->i_ino, LTIME_S(attr->ia_mtime));
1158 oa->o_id = lsm->lsm_object_id;
1159 oa->o_gr = lsm->lsm_object_gr;
1160 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1161 obdo_from_inode(oa, inode, OBD_MD_FLTYPE | OBD_MD_FLATIME |
1162 OBD_MD_FLMTIME | OBD_MD_FLCTIME);
1163 rc = obd_setattr(sbi->ll_dt_exp, oa, lsm, NULL);
1166 CERROR("obd_setattr fails: rc = %d\n", rc);
1171 int ll_setattr(struct dentry *de, struct iattr *attr)
1173 LASSERT(de->d_inode);
1174 return ll_setattr_raw(de->d_inode, attr);
1177 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1178 unsigned long max_age)
1180 struct ll_sb_info *sbi = ll_s2sbi(sb);
1181 struct obd_statfs obd_osfs;
1185 rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age);
1187 CERROR("obd_statfs fails: rc = %d\n", rc);
1191 osfs->os_type = sb->s_magic;
1193 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1194 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1196 rc = obd_statfs(class_exp2obd(sbi->ll_dt_exp), &obd_osfs, max_age);
1198 CERROR("obd_statfs fails: rc = %d\n", rc);
1202 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1203 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1206 osfs->os_blocks = obd_osfs.os_blocks;
1207 osfs->os_bfree = obd_osfs.os_bfree;
1208 osfs->os_bavail = obd_osfs.os_bavail;
1210 /* If we don't have as many objects free on the OST as inodes
1211 * on the MDS, we reduce the total number of inodes to
1212 * compensate, so that the "inodes in use" number is correct.
1214 if (obd_osfs.os_ffree < osfs->os_ffree) {
1215 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1217 osfs->os_ffree = obd_osfs.os_ffree;
1223 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1225 struct obd_statfs osfs;
1228 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p\n", sb);
1229 lprocfs_counter_incr(ll_s2sbi(sb)->ll_stats, LPROC_LL_STAFS);
1231 /* For now we will always get up-to-date statfs values, but in the
1232 * future we may allow some amount of caching on the client (e.g.
1233 * from QOS or lprocfs updates). */
1234 rc = ll_statfs_internal(sb, &osfs, jiffies - 1);
1238 statfs_unpack(sfs, &osfs);
1240 if (sizeof(sfs->f_blocks) == 4) {
1241 while (osfs.os_blocks > ~0UL) {
1244 osfs.os_blocks >>= 1;
1245 osfs.os_bfree >>= 1;
1246 osfs.os_bavail >>= 1;
1250 sfs->f_blocks = osfs.os_blocks;
1251 sfs->f_bfree = osfs.os_bfree;
1252 sfs->f_bavail = osfs.os_bavail;
1257 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1259 struct ll_inode_info *lli = ll_i2info(inode);
1260 struct lov_stripe_md *lsm = md->lsm;
1261 struct mds_body *body = md->body;
1262 struct mea *mea = md->mea;
1263 struct posix_acl *ll_acl_access = md->acl_access;
1266 LASSERT((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1268 if (md->lsm && md->lsm->lsm_magic != LOV_MAGIC) {
1269 /* check for default striping info for dir. */
1270 LASSERT((mea != NULL) == ((body->valid & OBD_MD_FLDIREA) != 0));
1274 LASSERT(lsm->lsm_object_gr > 0);
1275 if (lli->lli_smd == NULL) {
1277 lli->lli_maxbytes = lsm->lsm_maxbytes;
1278 if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1279 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1282 if (memcmp(lli->lli_smd, lsm, sizeof(*lsm))) {
1283 CERROR("lsm mismatch for inode %ld\n",
1285 CERROR("lli_smd:\n");
1286 dump_lsm(D_ERROR, lli->lli_smd);
1288 dump_lsm(D_ERROR, lsm);
1291 /* XXX FIXME -- We should decide on a safer (atomic) and
1292 * more elegant way to update the lsm */
1293 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1294 lli->lli_smd->lsm_oinfo[i].loi_id =
1295 lsm->lsm_oinfo[i].loi_id;
1296 lli->lli_smd->lsm_oinfo[i].loi_gr =
1297 lsm->lsm_oinfo[i].loi_gr;
1298 lli->lli_smd->lsm_oinfo[i].loi_ost_idx =
1299 lsm->lsm_oinfo[i].loi_ost_idx;
1300 lli->lli_smd->lsm_oinfo[i].loi_ost_gen =
1301 lsm->lsm_oinfo[i].loi_ost_gen;
1304 /* bug 2844 - limit i_blksize for broken user-space apps */
1305 LASSERTF(lsm->lsm_xfersize != 0, "%lu\n", lsm->lsm_xfersize);
1306 inode->i_blksize = min(lsm->lsm_xfersize, LL_MAX_BLKSIZE);
1307 if (lli->lli_smd != lsm)
1308 obd_free_memmd(ll_i2dtexp(inode), &lsm);
1312 if (lli->lli_mea == NULL) {
1315 if (memcmp(lli->lli_mea, mea, body->eadatasize)) {
1316 CERROR("mea mismatch for inode %lu\n",
1321 if (lli->lli_mea != mea)
1322 obd_free_memmd(ll_i2mdexp(inode),
1323 (struct lov_stripe_md **) &mea);
1326 if (body->valid & OBD_MD_FID)
1327 id_assign_fid(&lli->lli_id, &body->id1);
1329 if (body->valid & OBD_MD_FLID)
1330 id_ino(&lli->lli_id) = id_ino(&body->id1);
1332 if (body->valid & OBD_MD_FLGENER)
1333 id_gen(&lli->lli_id) = id_gen(&body->id1);
1335 spin_lock(&lli->lli_lock);
1336 if (ll_acl_access != NULL) {
1337 if (lli->lli_acl_access != NULL)
1338 posix_acl_release(lli->lli_acl_access);
1339 lli->lli_acl_access = ll_acl_access;
1341 spin_unlock(&lli->lli_lock);
1343 if (body->valid & OBD_MD_FLID)
1344 inode->i_ino = id_ino(&body->id1);
1345 if (body->valid & OBD_MD_FLGENER)
1346 inode->i_generation = id_gen(&body->id1);
1347 if (body->valid & OBD_MD_FLATIME)
1348 LTIME_S(inode->i_atime) = body->atime;
1349 if (body->valid & OBD_MD_FLMTIME &&
1350 body->mtime > LTIME_S(inode->i_mtime)) {
1351 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to %u\n",
1352 inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
1353 LTIME_S(inode->i_mtime) = body->mtime;
1355 if (body->valid & OBD_MD_FLCTIME &&
1356 body->ctime > LTIME_S(inode->i_ctime))
1357 LTIME_S(inode->i_ctime) = body->ctime;
1358 if (body->valid & OBD_MD_FLMODE) {
1359 inode->i_mode = (inode->i_mode & S_IFMT) |
1360 (body->mode & ~S_IFMT);
1362 if (body->valid & OBD_MD_FLTYPE) {
1363 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1364 (body->mode & S_IFMT);
1366 if (body->valid & OBD_MD_FLUID)
1367 inode->i_uid = body->uid;
1368 if (body->valid & OBD_MD_FLGID)
1369 inode->i_gid = body->gid;
1370 if (body->valid & OBD_MD_FLFLAGS)
1371 inode->i_flags = body->flags;
1372 if (body->valid & OBD_MD_FLNLINK)
1373 inode->i_nlink = body->nlink;
1374 if (body->valid & OBD_MD_FLRDEV)
1375 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1376 inode->i_rdev = body->rdev;
1378 inode->i_rdev = old_decode_dev(body->rdev);
1380 if (body->valid & OBD_MD_FLSIZE)
1381 inode->i_size = body->size;
1382 if (body->valid & OBD_MD_FLBLOCKS)
1383 inode->i_blocks = body->blocks;
1385 if (body->valid & OBD_MD_FLSIZE)
1386 set_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &lli->lli_flags);
1388 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1389 inode->i_dev = (kdev_t)id_group(&lli->lli_id);
1391 LASSERT(id_fid(&lli->lli_id) != 0);
1394 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
1395 static struct backing_dev_info ll_backing_dev_info = {
1396 .ra_pages = 0, /* No readahead */
1397 .memory_backed = 0, /* Does contribute to dirty memory */
1401 void ll_read_inode2(struct inode *inode, void *opaque)
1403 struct lustre_md *md = opaque;
1404 struct ll_inode_info *lli = ll_i2info(inode);
1407 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1408 inode->i_generation, inode);
1412 LASSERT(!lli->lli_smd);
1414 /* Core attributes from the MDS first. This is a new inode, and
1415 * the VFS doesn't zero times in the core inode so we have to do
1416 * it ourselves. They will be overwritten by either MDS or OST
1417 * attributes - we just need to make sure they aren't newer. */
1418 LTIME_S(inode->i_mtime) = 0;
1419 LTIME_S(inode->i_atime) = 0;
1420 LTIME_S(inode->i_ctime) = 0;
1423 ll_update_inode(inode, md);
1425 /* OIDEBUG(inode); */
1427 if (S_ISREG(inode->i_mode)) {
1428 inode->i_op = &ll_file_inode_operations;
1429 inode->i_fop = &ll_file_operations;
1430 inode->i_mapping->a_ops = &ll_aops;
1432 } else if (S_ISDIR(inode->i_mode)) {
1433 inode->i_op = &ll_dir_inode_operations;
1434 inode->i_fop = &ll_dir_operations;
1435 inode->i_mapping->a_ops = &ll_dir_aops;
1437 } else if (S_ISLNK(inode->i_mode)) {
1438 inode->i_op = &ll_fast_symlink_inode_operations;
1441 inode->i_op = &ll_special_inode_operations;
1443 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
1444 init_special_inode(inode, inode->i_mode,
1445 kdev_t_to_nr(inode->i_rdev));
1447 /* initializing backing dev info. */
1448 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
1450 init_special_inode(inode, inode->i_mode, inode->i_rdev);
1452 lli->ll_save_ifop = inode->i_fop;
1454 if (S_ISCHR(inode->i_mode))
1455 inode->i_fop = &ll_special_chr_inode_fops;
1456 else if (S_ISBLK(inode->i_mode))
1457 inode->i_fop = &ll_special_blk_inode_fops;
1458 else if (S_ISFIFO(inode->i_mode))
1459 inode->i_fop = &ll_special_fifo_inode_fops;
1460 else if (S_ISSOCK(inode->i_mode))
1461 inode->i_fop = &ll_special_sock_inode_fops;
1463 CWARN("saved %p, replaced with %p\n", lli->ll_save_ifop,
1466 if (lli->ll_save_ifop->owner) {
1467 CWARN("%p has owner %p\n", lli->ll_save_ifop,
1468 lli->ll_save_ifop->owner);
1474 void ll_delete_inode(struct inode *inode)
1476 struct ll_sb_info *sbi = ll_i2sbi(inode);
1477 struct lustre_id id;
1481 ll_inode2id(&id, inode);
1483 rc = md_delete_inode(sbi->ll_md_exp, &id);
1485 CERROR("md_delete_inode() failed, error %d\n",
1493 int ll_iocontrol(struct inode *inode, struct file *file,
1494 unsigned int cmd, unsigned long arg)
1496 struct ll_sb_info *sbi = ll_i2sbi(inode);
1497 struct ptlrpc_request *req = NULL;
1502 case EXT3_IOC_GETFLAGS: {
1503 struct lustre_id id;
1504 __u64 valid = OBD_MD_FLFLAGS;
1505 struct mds_body *body;
1507 ll_inode2id(&id, inode);
1508 rc = md_getattr(sbi->ll_md_exp, &id, valid, NULL, 0, 0, &req);
1510 CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1514 body = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*body));
1516 if (body->flags & S_APPEND)
1517 flags |= EXT3_APPEND_FL;
1518 if (body->flags & S_IMMUTABLE)
1519 flags |= EXT3_IMMUTABLE_FL;
1520 if (body->flags & S_NOATIME)
1521 flags |= EXT3_NOATIME_FL;
1523 ptlrpc_req_finished (req);
1525 RETURN(put_user(flags, (int *)arg));
1527 case EXT3_IOC_SETFLAGS: {
1528 struct mdc_op_data *op_data;
1531 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1533 if (get_user(flags, (int *)arg))
1540 OBD_ALLOC(op_data, sizeof(*op_data));
1541 if (op_data == NULL) {
1545 ll_prepare_mdc_data(op_data, inode, NULL, NULL, 0, 0);
1547 memset(&attr, 0x0, sizeof(attr));
1548 attr.ia_attr_flags = flags;
1549 attr.ia_valid |= ATTR_ATTR_FLAG;
1551 rc = md_setattr(sbi->ll_md_exp, op_data,
1552 &attr, NULL, 0, NULL, 0, &req);
1553 OBD_FREE(op_data, sizeof(*op_data));
1555 ptlrpc_req_finished(req);
1556 if (rc != -EPERM && rc != -EACCES)
1557 CERROR("md_setattr fails: rc = %d\n", rc);
1561 ptlrpc_req_finished(req);
1563 oa->o_id = lsm->lsm_object_id;
1564 oa->o_gr = lsm->lsm_object_gr;
1565 oa->o_flags = flags;
1566 oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS | OBD_MD_FLGROUP;
1568 rc = obd_setattr(sbi->ll_dt_exp, oa, lsm, NULL);
1571 if (rc != -EPERM && rc != -EACCES)
1572 CERROR("md_setattr fails: rc = %d\n", rc);
1576 if (flags & EXT3_APPEND_FL)
1577 inode->i_flags |= S_APPEND;
1579 inode->i_flags &= ~S_APPEND;
1580 if (flags & EXT3_IMMUTABLE_FL)
1581 inode->i_flags |= S_IMMUTABLE;
1583 inode->i_flags &= ~S_IMMUTABLE;
1584 if (flags & EXT3_NOATIME_FL)
1585 inode->i_flags |= S_NOATIME;
1587 inode->i_flags &= ~S_NOATIME;
1598 /* this is only called in the case of forced umount. */
1599 void ll_umount_begin(struct super_block *sb)
1601 struct ll_sb_info *sbi = ll_s2sbi(sb);
1602 struct obd_ioctl_data ioc_data = { 0 };
1603 struct obd_device *obd;
1606 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1607 sb->s_count, atomic_read(&sb->s_active));
1609 obd = class_exp2obd(sbi->ll_md_exp);
1611 CERROR("Invalid MDC connection handle "LPX64"\n",
1612 sbi->ll_md_exp->exp_handle.h_cookie);
1616 obd->obd_no_recov = 1;
1617 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
1618 sizeof(ioc_data), &ioc_data, NULL);
1620 obd = class_exp2obd(sbi->ll_dt_exp);
1622 CERROR("Invalid LOV connection handle "LPX64"\n",
1623 sbi->ll_dt_exp->exp_handle.h_cookie);
1628 obd->obd_no_recov = 1;
1629 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
1630 sizeof(ioc_data), &ioc_data, NULL);
1633 * really, we'd like to wait until there are no requests outstanding,
1634 * and then continue. For now, we just invalidate the requests,
1635 * schedule, and hope.
1642 int ll_prep_inode(struct obd_export *dt_exp, struct obd_export *md_exp,
1643 struct inode **inode, struct ptlrpc_request *req,
1644 int offset, struct super_block *sb)
1646 struct lustre_md md;
1649 rc = mdc_req2lustre_md(md_exp, req, offset, dt_exp, &md);
1654 ll_update_inode(*inode, &md);
1657 *inode = ll_iget(sb, id_ino(&md.body->id1), &md);
1658 if (*inode == NULL || is_bad_inode(*inode)) {
1659 /* free the lsm if we allocated one above */
1661 obd_free_memmd(dt_exp, &md.lsm);
1663 obd_free_memmd(md_exp,
1664 (struct lov_stripe_md**)&md.mea);
1666 CERROR("new_inode -fatal: rc %d\n", rc);
1673 int ll_get_fid(struct obd_export *exp, struct lustre_id *idp,
1674 char *filename, struct lustre_id *ret)
1676 struct ptlrpc_request *request = NULL;
1677 struct mds_body *body;
1680 rc = md_getattr_lock(exp, idp, filename, strlen(filename) + 1,
1681 OBD_MD_FID, 0, &request);
1683 CDEBUG(D_INFO, "md_getattr_lock failed on %s: rc %d\n",
1688 body = lustre_msg_buf(request->rq_repmsg, 0, sizeof(*body));
1689 LASSERT(body != NULL);
1690 LASSERT_REPSWABBED(request, 0);
1693 ptlrpc_req_finished(request);