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/random.h>
28 #include <linux/version.h>
30 #include <linux/lustre_lite.h>
31 #include <linux/lustre_ha.h>
32 #include <linux/lustre_dlm.h>
33 #include <linux/lprocfs_status.h>
34 #include "llite_internal.h"
36 kmem_cache_t *ll_file_data_slab;
38 extern struct address_space_operations ll_aops;
39 extern struct address_space_operations ll_dir_aops;
40 extern struct super_operations ll_super_operations;
43 #define log2(n) ffz(~(n))
46 struct ll_sb_info *lustre_init_sbi(struct super_block *sb)
48 struct ll_sb_info *sbi = NULL;
52 OBD_ALLOC(sbi, sizeof(*sbi));
56 spin_lock_init(&sbi->ll_pglist_lock);
57 INIT_LIST_HEAD(&sbi->ll_pglist);
58 sbi->ll_pglist_gen = 0;
59 INIT_LIST_HEAD(&sbi->ll_conn_chain);
60 INIT_HLIST_HEAD(&sbi->ll_orphan_dentry_list);
63 generate_random_uuid(uuid);
64 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
68 void lustre_free_sbi(struct super_block *sb)
70 struct ll_sb_info *sbi = ll_s2sbi(sb);
74 OBD_FREE(sbi, sizeof(*sbi));
79 int lustre_common_fill_super(struct super_block *sb, char *mdc, char *osc)
81 struct inode *root = 0;
82 struct ll_sb_info *sbi = ll_s2sbi(sb);
83 struct obd_device *obd;
84 struct ll_fid rootfid;
85 struct obd_statfs osfs;
86 struct ptlrpc_request *request = NULL;
87 struct lustre_handle osc_conn = {0, };
88 struct lustre_handle mdc_conn = {0, };
93 obd = class_name2obd(mdc);
95 CERROR("MDC %s: not setup or attached\n", mdc);
99 if (proc_lustre_fs_root) {
100 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
103 CERROR("could not register mount in /proc/lustre");
106 mdc_init_ea_size(obd, osc);
108 err = obd_connect(&mdc_conn, obd, &sbi->ll_sb_uuid);
110 CERROR("An MDS (mdc %s) is performing recovery, of which this"
111 " client is not a part. Please wait for recovery to "
112 "complete, abort, or time out.\n", mdc);
115 CERROR("cannot connect to %s: rc = %d\n", mdc, err);
118 sbi->ll_mdc_exp = class_conn2export(&mdc_conn);
120 err = obd_statfs(obd, &osfs, jiffies - HZ);
124 LASSERT(osfs.os_bsize);
125 sb->s_blocksize = osfs.os_bsize;
126 sb->s_blocksize_bits = log2(osfs.os_bsize);
127 sb->s_magic = LL_SUPER_MAGIC;
128 sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
130 devno = get_uuid2int(sbi2mdc(sbi)->cl_import->imp_target_uuid.uuid,
131 strlen(sbi2mdc(sbi)->cl_import->imp_target_uuid.uuid));
134 obd = class_name2obd(osc);
136 CERROR("OSC %s: not setup or attached\n", osc);
140 err = obd_connect(&osc_conn, obd, &sbi->ll_sb_uuid);
142 CERROR("An OST (osc %s) is performing recovery, of which this"
143 " client is not a part. Please wait for recovery to "
144 "complete, abort, or time out.\n", osc);
147 CERROR("cannot connect to %s: rc = %d\n", osc, err);
150 sbi->ll_osc_exp = class_conn2export(&osc_conn);
152 err = mdc_getstatus(sbi->ll_mdc_exp, &rootfid);
154 CERROR("cannot mds_connect: rc = %d\n", err);
157 CDEBUG(D_SUPER, "rootfid "LPU64"\n", rootfid.id);
158 sbi->ll_rootino = rootfid.id;
160 sb->s_op = &lustre_super_operations;
163 * XXX: move this to after cbd setup? */
164 err = mdc_getattr(sbi->ll_mdc_exp, &rootfid,
165 OBD_MD_FLNOTOBD|OBD_MD_FLBLOCKS, 0, &request);
167 CERROR("mdc_getattr failed for root: rc = %d\n", err);
171 err = mdc_req2lustre_md(request, 0, sbi->ll_osc_exp, &md);
173 CERROR("failed to understand root inode md: rc = %d\n",err);
174 ptlrpc_req_finished (request);
178 LASSERT(sbi->ll_rootino != 0);
179 root = ll_iget(sb, sbi->ll_rootino, &md);
181 ptlrpc_req_finished(request);
183 if (root == NULL || is_bad_inode(root)) {
184 /* XXX might need iput() for bad inode */
185 CERROR("lustre_lite: bad iget4 for root\n");
186 GOTO(out_root, err = -EBADF);
189 err = ll_close_thread_start(&sbi->ll_lcq);
191 CERROR("cannot start close thread: rc %d\n", err);
195 sb->s_root = d_alloc_root(root);
202 obd_disconnect(sbi->ll_osc_exp, 0);
204 obd_disconnect(sbi->ll_mdc_exp, 0);
206 lprocfs_unregister_mountpoint(sbi);
210 void lustre_common_put_super(struct super_block *sb)
212 struct ll_sb_info *sbi = ll_s2sbi(sb);
213 struct hlist_node *tmp, *next;
216 ll_close_thread_shutdown(sbi->ll_lcq);
218 list_del(&sbi->ll_conn_chain);
219 obd_disconnect(sbi->ll_osc_exp, 0);
221 lprocfs_unregister_mountpoint(sbi);
222 if (sbi->ll_proc_root) {
223 lprocfs_remove(sbi->ll_proc_root);
224 sbi->ll_proc_root = NULL;
227 obd_disconnect(sbi->ll_mdc_exp, 0);
229 // We do this to get rid of orphaned dentries. That is not really trw.
230 spin_lock(&dcache_lock);
231 hlist_for_each_safe(tmp, next, &sbi->ll_orphan_dentry_list) {
232 struct dentry *dentry = hlist_entry(tmp, struct dentry, d_hash);
233 shrink_dcache_parent(dentry);
235 spin_unlock(&dcache_lock);
240 char *ll_read_opt(const char *opt, char *data)
246 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
247 if (strncmp(opt, data, strlen(opt)))
249 if ((value = strchr(data, '=')) == NULL)
253 OBD_ALLOC(retval, strlen(value) + 1);
255 CERROR("out of memory!\n");
259 memcpy(retval, value, strlen(value)+1);
260 CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
264 int ll_set_opt(const char *opt, char *data, int fl)
268 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
269 if (strncmp(opt, data, strlen(opt)))
275 void ll_options(char *options, char **ost, char **mdc, int *flags)
278 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
279 char *opt_ptr = options;
288 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
289 for (this_char = strtok (options, ",");
291 this_char = strtok (NULL, ",")) {
293 while ((this_char = strsep (&opt_ptr, ",")) != NULL) {
295 CDEBUG(D_SUPER, "this_char %s\n", this_char);
296 if (!*ost && (*ost = ll_read_opt("osc", this_char)))
298 if (!*mdc && (*mdc = ll_read_opt("mdc", this_char)))
300 if (!(*flags & LL_SBI_NOLCK) &&
301 ((*flags) = (*flags) |
302 ll_set_opt("nolock", this_char,
309 void ll_lli_init(struct ll_inode_info *lli)
311 sema_init(&lli->lli_open_sem, 1);
313 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
314 spin_lock_init(&lli->lli_lock);
315 INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
318 int ll_fill_super(struct super_block *sb, void *data, int silent)
320 struct ll_sb_info *sbi;
326 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
328 sbi = lustre_init_sbi(sb);
332 sbi->ll_flags |= LL_SBI_READAHEAD;
333 ll_options(data, &osc, &mdc, &sbi->ll_flags);
337 GOTO(out, err = -EINVAL);
342 GOTO(out, err = -EINVAL);
345 err = lustre_common_fill_super(sb, mdc, osc);
351 OBD_FREE(mdc, strlen(mdc) + 1);
353 OBD_FREE(osc, strlen(osc) + 1);
356 } /* ll_read_super */
358 void ll_put_super(struct super_block *sb)
362 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
364 lustre_common_put_super(sb);
371 int lustre_process_log(struct lustre_mount_data *lmd, char * profile,
372 struct config_llog_instance *cfg, int allow_recov)
374 struct lustre_cfg lcfg;
375 struct portals_cfg pcfg;
376 char * peer = "MDS_PEER_UUID";
377 struct obd_device *obd;
378 struct lustre_handle mdc_conn = {0, };
379 struct obd_export *exp;
380 char * name = "mdc_dev";
382 struct obd_uuid mdc_uuid;
383 struct llog_ctxt *ctxt;
388 generate_random_uuid(uuid);
389 class_uuid_unparse(uuid, &mdc_uuid);
391 if (lmd->lmd_local_nid) {
392 PCFG_INIT(pcfg, NAL_CMD_REGISTER_MYNID);
393 pcfg.pcfg_nal = lmd->lmd_nal;
394 pcfg.pcfg_nid = lmd->lmd_local_nid;
395 err = kportal_nal_cmd(&pcfg);
400 if (lmd->lmd_nal == SOCKNAL) {
401 PCFG_INIT(pcfg, NAL_CMD_ADD_AUTOCONN);
402 pcfg.pcfg_nal = lmd->lmd_nal;
403 pcfg.pcfg_nid = lmd->lmd_server_nid;
404 pcfg.pcfg_id = lmd->lmd_server_ipaddr;
405 pcfg.pcfg_misc = lmd->lmd_port;
406 pcfg.pcfg_size = 8388608;
407 pcfg.pcfg_flags = 0x4; /*share*/
408 err = kportal_nal_cmd(&pcfg);
413 LCFG_INIT(lcfg, LCFG_ADD_UUID, name);
414 lcfg.lcfg_nid = lmd->lmd_server_nid;
415 lcfg.lcfg_inllen1 = strlen(peer) + 1;
416 lcfg.lcfg_inlbuf1 = peer;
417 lcfg.lcfg_nal = lmd->lmd_nal;
418 err = class_process_config(&lcfg);
420 GOTO(out_del_conn, err);
422 LCFG_INIT(lcfg, LCFG_ATTACH, name);
423 lcfg.lcfg_inlbuf1 = "mdc";
424 lcfg.lcfg_inllen1 = strlen(lcfg.lcfg_inlbuf1) + 1;
425 lcfg.lcfg_inlbuf2 = mdc_uuid.uuid;
426 lcfg.lcfg_inllen2 = strlen(lcfg.lcfg_inlbuf2) + 1;
427 err = class_process_config(&lcfg);
429 GOTO(out_del_uuid, err);
431 LCFG_INIT(lcfg, LCFG_SETUP, name);
432 lcfg.lcfg_inlbuf1 = lmd->lmd_mds;
433 lcfg.lcfg_inllen1 = strlen(lcfg.lcfg_inlbuf1) + 1;
434 lcfg.lcfg_inlbuf2 = peer;
435 lcfg.lcfg_inllen2 = strlen(lcfg.lcfg_inlbuf2) + 1;
436 err = class_process_config(&lcfg);
438 GOTO(out_detach, err);
440 obd = class_name2obd(name);
442 GOTO(out_cleanup, err = -EINVAL);
444 /* Disable initial recovery on this import */
445 err = obd_set_info(obd->obd_self_export,
446 strlen("initial_recov"), "initial_recov",
447 sizeof(allow_recov), &allow_recov);
449 GOTO(out_cleanup, err);
451 err = obd_connect(&mdc_conn, obd, &mdc_uuid);
453 CERROR("cannot connect to %s: rc = %d\n", lmd->lmd_mds, err);
454 GOTO(out_cleanup, err);
457 exp = class_conn2export(&mdc_conn);
459 ctxt = llog_get_context(exp->exp_obd, LLOG_CONFIG_REPL_CTXT);
460 rc = class_config_parse_llog(ctxt, profile, cfg);
462 CERROR("class_config_parse_llog failed: rc = %d\n", rc);
465 err = obd_disconnect(exp, 0);
468 LCFG_INIT(lcfg, LCFG_CLEANUP, name);
469 err = class_process_config(&lcfg);
474 LCFG_INIT(lcfg, LCFG_DETACH, name);
475 err = class_process_config(&lcfg);
480 LCFG_INIT(lcfg, LCFG_DEL_UUID, name);
481 lcfg.lcfg_inllen1 = strlen(peer) + 1;
482 lcfg.lcfg_inlbuf1 = peer;
483 err = class_process_config(&lcfg);
486 if (lmd->lmd_nal == SOCKNAL) {
487 PCFG_INIT(pcfg, NAL_CMD_DEL_AUTOCONN);
488 pcfg.pcfg_nal = lmd->lmd_nal;
489 pcfg.pcfg_nid = lmd->lmd_server_nid;
490 pcfg.pcfg_id = lmd->lmd_server_ipaddr;
491 pcfg.pcfg_flags = 1; /*share*/
492 err = kportal_nal_cmd(&pcfg);
503 int lustre_fill_super(struct super_block *sb, void *data, int silent)
505 struct lustre_mount_data * lmd = data;
506 struct ll_sb_info *sbi;
512 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
513 sbi = lustre_init_sbi(sb);
517 sbi->ll_flags |= LL_SBI_READAHEAD;
519 if (lmd->lmd_profile) {
520 struct lustre_profile *lprof;
521 struct config_llog_instance cfg;
525 CERROR("no mds name\n");
526 GOTO(out_free, err = -EINVAL);
529 OBD_ALLOC(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
530 if (sbi->ll_lmd == NULL)
531 GOTO(out_free, err = -ENOMEM);
532 memcpy(sbi->ll_lmd, lmd, sizeof(*lmd));
534 /* generate a string unique to this super, let's try
535 the address of the super itself.*/
536 len = (sizeof(sb) * 2) + 1;
537 OBD_ALLOC(sbi->ll_instance, len);
538 if (sbi->ll_instance == NULL)
539 GOTO(out_free, err = -ENOMEM);
540 sprintf(sbi->ll_instance, "%p", sb);
542 cfg.cfg_instance = sbi->ll_instance;
543 cfg.cfg_uuid = sbi->ll_sb_uuid;
544 cfg.cfg_local_nid = lmd->lmd_local_nid;
545 err = lustre_process_log(lmd, lmd->lmd_profile, &cfg, 1);
547 CERROR("Unable to process log: %s\n", lmd->lmd_profile);
552 lprof = class_get_profile(lmd->lmd_profile);
554 CERROR("No profile found: %s\n", lmd->lmd_profile);
555 GOTO(out_free, err = -EINVAL);
558 OBD_FREE(osc, strlen(osc) + 1);
559 OBD_ALLOC(osc, strlen(lprof->lp_osc) +
560 strlen(sbi->ll_instance) + 2);
561 sprintf(osc, "%s-%s", lprof->lp_osc, sbi->ll_instance);
564 OBD_FREE(mdc, strlen(mdc) + 1);
565 OBD_ALLOC(mdc, strlen(lprof->lp_mdc) +
566 strlen(sbi->ll_instance) + 2);
567 sprintf(mdc, "%s-%s", lprof->lp_mdc, sbi->ll_instance);
572 GOTO(out_free, err = -EINVAL);
577 GOTO(out_free, err = -EINVAL);
580 err = lustre_common_fill_super(sb, mdc, osc);
587 OBD_FREE(mdc, strlen(mdc) + 1);
589 OBD_FREE(osc, strlen(osc) + 1);
595 int len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean")+1;
598 if (sbi->ll_instance != NULL) {
600 struct config_llog_instance cfg;
602 cfg.cfg_instance = sbi->ll_instance;
603 cfg.cfg_uuid = sbi->ll_sb_uuid;
605 OBD_ALLOC(cln_prof, len);
606 sprintf(cln_prof, "%s-clean", sbi->ll_lmd->lmd_profile);
608 err = lustre_process_log(sbi->ll_lmd, cln_prof, &cfg,
611 CERROR("Unable to process log: %s\n", cln_prof);
612 OBD_FREE(cln_prof, len);
613 OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance)+ 1);
615 OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
620 } /* lustre_fill_super */
622 static void lustre_manual_cleanup(struct ll_sb_info *sbi)
624 struct lustre_cfg lcfg;
625 struct obd_device *obd;
628 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) != NULL)
632 LCFG_INIT(lcfg, LCFG_CLEANUP, obd->obd_name);
633 err = class_process_config(&lcfg);
635 CERROR("cleanup failed: %s\n", obd->obd_name);
639 LCFG_INIT(lcfg, LCFG_DETACH, obd->obd_name);
640 err = class_process_config(&lcfg);
642 CERROR("detach failed: %s\n", obd->obd_name);
647 if (sbi->ll_lmd != NULL)
648 class_del_profile(sbi->ll_lmd->lmd_profile);
651 void lustre_put_super(struct super_block *sb)
653 struct obd_device *obd;
654 struct ll_sb_info *sbi = ll_s2sbi(sb);
655 int force_umount = 0;
658 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
659 obd = class_exp2obd(sbi->ll_mdc_exp);
661 force_umount = obd->obd_no_recov;
664 lustre_common_put_super(sb);
666 if (sbi->ll_lmd != NULL) {
668 int len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean")+1;
670 struct config_llog_instance cfg;
673 CERROR("force umount, doing manual cleanup\n");
674 lustre_manual_cleanup(sbi);
678 cfg.cfg_instance = sbi->ll_instance;
679 cfg.cfg_uuid = sbi->ll_sb_uuid;
681 OBD_ALLOC(cln_prof, len);
682 sprintf(cln_prof, "%s-clean", sbi->ll_lmd->lmd_profile);
684 err = lustre_process_log(sbi->ll_lmd, cln_prof, &cfg, 0);
686 CERROR("Unable to process log: %s, doing manual cleanup"
688 lustre_manual_cleanup(sbi);
691 OBD_FREE(cln_prof, len);
693 OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
694 OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance) + 1);
700 } /* lustre_put_super */
703 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
706 l_lock(&lock->l_resource->lr_namespace->ns_lock);
707 if (lock->l_ast_data)
708 inode = igrab(lock->l_ast_data);
711 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
715 static int null_if_equal(struct ldlm_lock *lock, void *data)
717 if (data == lock->l_ast_data)
718 lock->l_ast_data = NULL;
720 if (lock->l_req_mode != lock->l_granted_mode)
721 return LDLM_ITER_STOP;
723 return LDLM_ITER_CONTINUE;
726 void ll_clear_inode(struct inode *inode)
729 struct ll_inode_info *lli = ll_i2info(inode);
730 struct ll_sb_info *sbi = ll_i2sbi(inode);
733 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
734 inode->i_generation, inode);
736 ll_inode2fid(&fid, inode);
737 clear_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &(ll_i2info(inode)->lli_flags));
738 mdc_change_cbdata(sbi->ll_mdc_exp, &fid, null_if_equal, inode);
741 obd_change_cbdata(sbi->ll_osc_exp, lli->lli_smd,
742 null_if_equal, inode);
745 obd_free_memmd(sbi->ll_osc_exp, &lli->lli_smd);
749 if (lli->lli_symlink_name) {
750 OBD_FREE(lli->lli_symlink_name,
751 strlen(lli->lli_symlink_name) + 1);
752 lli->lli_symlink_name = NULL;
758 /* If this inode has objects allocated to it (lsm != NULL), then the OST
759 * object(s) determine the file size and mtime. Otherwise, the MDS will
760 * keep these values until such a time that objects are allocated for it.
761 * We do the MDS operations first, as it is checking permissions for us.
762 * We don't to the MDS RPC if there is nothing that we want to store there,
763 * otherwise there is no harm in updating mtime/atime on the MDS if we are
764 * going to do an RPC anyways.
766 * If we are doing a truncate, we will send the mtime and ctime updates
767 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
768 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
771 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
773 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
774 struct ll_sb_info *sbi = ll_i2sbi(inode);
775 struct ptlrpc_request *request = NULL;
776 struct mdc_op_data op_data;
777 int ia_valid = attr->ia_valid;
781 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu\n", inode->i_ino);
782 lprocfs_counter_incr(ll_i2sbi(inode)->ll_stats, LPROC_LL_SETATTR);
784 if (ia_valid & ATTR_SIZE) {
785 if (attr->ia_size > ll_file_maxbytes(inode)) {
786 CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
787 attr->ia_size, ll_file_maxbytes(inode));
791 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
794 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
795 if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
796 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
800 /* We mark all of the fields "set" so MDS/OST does not re-set them */
801 if (attr->ia_valid & ATTR_CTIME) {
802 attr->ia_ctime = CURRENT_TIME;
803 attr->ia_valid |= ATTR_CTIME_SET;
805 if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
806 attr->ia_atime = CURRENT_TIME;
807 attr->ia_valid |= ATTR_ATIME_SET;
809 if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
810 attr->ia_mtime = CURRENT_TIME;
811 attr->ia_valid |= ATTR_MTIME_SET;
814 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
815 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
816 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
817 LTIME_S(CURRENT_TIME));
819 attr->ia_valid &= ~ATTR_SIZE;
821 /* If only OST attributes being set on objects, don't do MDS RPC.
822 * In that case, we need to check permissions and update the local
823 * inode ourselves so we can call obdo_from_inode() always. */
824 if (ia_valid & (lsm ? ~(ATTR_SIZE | ATTR_FROM_OPEN | ATTR_RAW) : ~0)) {
826 ll_prepare_mdc_op_data(&op_data, inode, NULL, NULL, 0, 0);
828 rc = mdc_setattr(sbi->ll_mdc_exp, &op_data,
829 attr, NULL, 0, NULL, 0, &request);
832 ptlrpc_req_finished(request);
833 if (rc != -EPERM && rc != -EACCES)
834 CERROR("mdc_setattr fails: rc = %d\n", rc);
838 rc = mdc_req2lustre_md(request, 0, sbi->ll_osc_exp, &md);
840 ptlrpc_req_finished(request);
844 /* Won't invoke vmtruncate as we already cleared ATTR_SIZE,
845 * but needed to set timestamps backwards on utime. */
846 inode_setattr(inode, attr);
847 ll_update_inode(inode, md.body, md.lsm);
848 ptlrpc_req_finished(request);
850 if (!lsm || !S_ISREG(inode->i_mode)) {
851 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
855 /* The OST doesn't check permissions, but the alternative is
856 * a gratuitous RPC to the MDS. We already rely on the client
857 * to do read/write/truncate permission checks, so is mtime OK?
859 if (ia_valid & (ATTR_MTIME | ATTR_ATIME)) {
860 /* from sys_utime() */
861 if (!(ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET))) {
862 if (current->fsuid != inode->i_uid &&
863 (rc = ll_permission(inode, MAY_WRITE, NULL)) != 0)
866 /* from inode_change_ok() */
867 if (current->fsuid != inode->i_uid &&
868 !capable(CAP_FOWNER))
873 /* Won't invoke vmtruncate, as we already cleared ATTR_SIZE */
874 inode_setattr(inode, attr);
877 /* We really need to get our PW lock before we change inode->i_size.
878 * If we don't we can race with other i_size updaters on our node, like
879 * ll_file_read. We can also race with i_size propogation to other
880 * nodes through dirtying and writeback of final cached pages. This
881 * last one is especially bad for racing o_append users on other
883 if (ia_valid & ATTR_SIZE) {
884 struct ldlm_extent extent = { .start = attr->ia_size,
885 .end = OBD_OBJECT_EOF };
886 struct lustre_handle lockh = { 0 };
887 int err, ast_flags = 0;
888 /* XXX when we fix the AST intents to pass the discard-range
889 * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
891 if (attr->ia_size == 0)
892 ast_flags = LDLM_AST_DISCARD_DATA;
894 /* bug 1639: avoid write/truncate i_sem/DLM deadlock */
895 LASSERT(atomic_read(&inode->i_sem.count) <= 0);
897 rc = ll_extent_lock_no_validate(NULL, inode, lsm, LCK_PW,
898 &extent, &lockh, ast_flags);
903 rc = vmtruncate(inode, attr->ia_size);
905 set_bit(LLI_F_HAVE_OST_SIZE_LOCK,
906 &ll_i2info(inode)->lli_flags);
908 //ll_try_done_writing(inode);
910 /* unlock now as we don't mind others file lockers racing with
911 * the mds updates below? */
912 err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
914 CERROR("ll_extent_unlock failed: %d\n", err);
918 } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
921 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
922 inode->i_ino, LTIME_S(attr->ia_mtime));
923 oa.o_id = lsm->lsm_object_id;
924 oa.o_valid = OBD_MD_FLID;
925 obdo_from_inode(&oa, inode, OBD_MD_FLTYPE | OBD_MD_FLATIME |
926 OBD_MD_FLMTIME | OBD_MD_FLCTIME);
927 rc = obd_setattr(sbi->ll_osc_exp, &oa, lsm, NULL);
929 CERROR("obd_setattr fails: rc=%d\n", rc);
934 int ll_setattr(struct dentry *de, struct iattr *attr)
936 LBUG(); /* code is unused, but leave this in case of VFS changes */
940 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
941 unsigned long max_age)
943 struct ll_sb_info *sbi = ll_s2sbi(sb);
944 struct obd_statfs obd_osfs;
948 rc = obd_statfs(class_exp2obd(sbi->ll_mdc_exp), osfs, max_age);
950 CERROR("mdc_statfs fails: rc = %d\n", rc);
954 osfs->os_type = sb->s_magic;
956 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
957 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
959 rc = obd_statfs(class_exp2obd(sbi->ll_osc_exp), &obd_osfs, max_age);
961 CERROR("obd_statfs fails: rc = %d\n", rc);
965 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
966 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
969 osfs->os_blocks = obd_osfs.os_blocks;
970 osfs->os_bfree = obd_osfs.os_bfree;
971 osfs->os_bavail = obd_osfs.os_bavail;
973 /* If we don't have as many objects free on the OST as inodes
974 * on the MDS, we reduce the total number of inodes to
975 * compensate, so that the "inodes in use" number is correct.
977 if (obd_osfs.os_ffree < osfs->os_ffree) {
978 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
980 osfs->os_ffree = obd_osfs.os_ffree;
986 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
988 struct obd_statfs osfs;
991 CDEBUG(D_VFSTRACE, "VFS Op:\n");
992 lprocfs_counter_incr(ll_s2sbi(sb)->ll_stats, LPROC_LL_STAFS);
994 /* For now we will always get up-to-date statfs values, but in the
995 * future we may allow some amount of caching on the client (e.g.
996 * from QOS or lprocfs updates). */
997 rc = ll_statfs_internal(sb, &osfs, jiffies - 1);
1001 statfs_unpack(sfs, &osfs);
1003 if (sizeof(sfs->f_blocks) == 4) {
1004 while (osfs.os_blocks > ~0UL) {
1007 osfs.os_blocks >>= 1;
1008 osfs.os_bfree >>= 1;
1009 osfs.os_bavail >>= 1;
1013 sfs->f_blocks = osfs.os_blocks;
1014 sfs->f_bfree = osfs.os_bfree;
1015 sfs->f_bavail = osfs.os_bavail;
1020 void dump_lsm(int level, struct lov_stripe_md *lsm)
1022 CDEBUG(level, "objid "LPX64", maxbytes "LPX64", magic 0x%08X, "
1023 "stripe_size %u, stripe_count %u\n",
1024 lsm->lsm_object_id, lsm->lsm_maxbytes, lsm->lsm_magic,
1025 lsm->lsm_stripe_size, lsm->lsm_stripe_count);
1028 void ll_update_inode(struct inode *inode, struct mds_body *body,
1029 struct lov_stripe_md *lsm)
1031 struct ll_inode_info *lli = ll_i2info(inode);
1033 LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1035 if (lli->lli_smd == NULL) {
1037 lli->lli_maxbytes = lsm->lsm_maxbytes;
1038 if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1039 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1041 if (memcmp(lli->lli_smd, lsm, sizeof(*lsm))) {
1042 CERROR("lsm mismatch for inode %ld\n",
1044 CERROR("lli_smd:\n");
1045 dump_lsm(D_ERROR, lli->lli_smd);
1047 dump_lsm(D_ERROR, lsm);
1051 if (lli->lli_smd != lsm)
1052 obd_free_memmd(ll_i2obdexp(inode), &lsm);
1055 if (body->valid & OBD_MD_FLID)
1056 inode->i_ino = body->ino;
1057 if (body->valid & OBD_MD_FLATIME)
1058 LTIME_S(inode->i_atime) = body->atime;
1059 if (body->valid & OBD_MD_FLMTIME &&
1060 body->mtime > LTIME_S(inode->i_mtime)) {
1061 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to %u\n",
1062 inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
1063 LTIME_S(inode->i_mtime) = body->mtime;
1065 if (body->valid & OBD_MD_FLCTIME &&
1066 body->ctime > LTIME_S(inode->i_ctime))
1067 LTIME_S(inode->i_ctime) = body->ctime;
1068 if (body->valid & OBD_MD_FLMODE)
1069 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1070 if (body->valid & OBD_MD_FLTYPE)
1071 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1072 if (body->valid & OBD_MD_FLUID)
1073 inode->i_uid = body->uid;
1074 if (body->valid & OBD_MD_FLGID)
1075 inode->i_gid = body->gid;
1076 if (body->valid & OBD_MD_FLFLAGS)
1077 inode->i_flags = body->flags;
1078 if (body->valid & OBD_MD_FLNLINK)
1079 inode->i_nlink = body->nlink;
1080 if (body->valid & OBD_MD_FLGENER)
1081 inode->i_generation = body->generation;
1082 if (body->valid & OBD_MD_FLRDEV)
1083 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1084 inode->i_rdev = body->rdev;
1086 inode->i_rdev = old_decode_dev(body->rdev);
1088 if (body->valid & OBD_MD_FLSIZE)
1089 inode->i_size = body->size;
1090 if (body->valid & OBD_MD_FLBLOCKS)
1091 inode->i_blocks = body->blocks;
1093 if (body->valid & OBD_MD_FLSIZE)
1094 set_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &lli->lli_flags);
1097 void ll_read_inode2(struct inode *inode, void *opaque)
1099 struct lustre_md *md = opaque;
1100 struct ll_inode_info *lli = ll_i2info(inode);
1103 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1104 inode->i_generation, inode);
1108 LASSERT(!lli->lli_smd);
1110 /* Core attributes from the MDS first. This is a new inode, and
1111 * the VFS doesn't zero times in the core inode so we have to do
1112 * it ourselves. They will be overwritten by either MDS or OST
1113 * attributes - we just need to make sure they aren't newer. */
1114 LTIME_S(inode->i_mtime) = 0;
1115 LTIME_S(inode->i_atime) = 0;
1116 LTIME_S(inode->i_ctime) = 0;
1117 ll_update_inode(inode, md->body, md->lsm);
1119 /* OIDEBUG(inode); */
1121 if (S_ISREG(inode->i_mode)) {
1122 inode->i_op = &ll_file_inode_operations;
1123 inode->i_fop = &ll_file_operations;
1124 inode->i_mapping->a_ops = &ll_aops;
1126 } else if (S_ISDIR(inode->i_mode)) {
1127 inode->i_op = &ll_dir_inode_operations;
1128 inode->i_fop = &ll_dir_operations;
1129 inode->i_mapping->a_ops = &ll_dir_aops;
1131 } else if (S_ISLNK(inode->i_mode)) {
1132 inode->i_op = &ll_fast_symlink_inode_operations;
1135 inode->i_op = &ll_special_inode_operations;
1136 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
1137 init_special_inode(inode, inode->i_mode,
1138 kdev_t_to_nr(inode->i_rdev));
1140 init_special_inode(inode, inode->i_mode, inode->i_rdev);
1146 int ll_iocontrol(struct inode *inode, struct file *file,
1147 unsigned int cmd, unsigned long arg)
1149 struct ll_sb_info *sbi = ll_i2sbi(inode);
1150 struct ptlrpc_request *req = NULL;
1155 case EXT3_IOC_GETFLAGS: {
1157 unsigned long valid = OBD_MD_FLFLAGS;
1158 struct mds_body *body;
1160 ll_inode2fid(&fid, inode);
1161 rc = mdc_getattr(sbi->ll_mdc_exp, &fid, valid, 0, &req);
1163 CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1167 body = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*body));
1169 if (body->flags & S_APPEND)
1170 flags |= EXT3_APPEND_FL;
1171 if (body->flags & S_IMMUTABLE)
1172 flags |= EXT3_IMMUTABLE_FL;
1173 if (body->flags & S_NOATIME)
1174 flags |= EXT3_NOATIME_FL;
1176 ptlrpc_req_finished (req);
1178 RETURN(put_user(flags, (int *)arg));
1180 case EXT3_IOC_SETFLAGS: {
1181 struct mdc_op_data op_data;
1184 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1186 if (get_user(flags, (int *)arg))
1193 ll_prepare_mdc_op_data(&op_data, inode, NULL, NULL, 0, 0);
1195 memset(&attr, 0x0, sizeof(attr));
1196 attr.ia_attr_flags = flags;
1197 attr.ia_valid |= ATTR_ATTR_FLAG;
1199 rc = mdc_setattr(sbi->ll_mdc_exp, &op_data,
1200 &attr, NULL, 0, NULL, 0, &req);
1202 ptlrpc_req_finished(req);
1203 if (rc != -EPERM && rc != -EACCES)
1204 CERROR("mdc_setattr fails: rc = %d\n", rc);
1208 ptlrpc_req_finished(req);
1210 oa->o_id = lsm->lsm_object_id;
1211 oa->o_flags = flags;
1212 oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS;
1214 rc = obd_setattr(sbi->ll_osc_exp, oa, lsm, NULL);
1217 if (rc != -EPERM && rc != -EACCES)
1218 CERROR("mdc_setattr fails: rc = %d\n", rc);
1222 if (flags & EXT3_APPEND_FL)
1223 inode->i_flags |= S_APPEND;
1225 inode->i_flags &= ~S_APPEND;
1226 if (flags & EXT3_IMMUTABLE_FL)
1227 inode->i_flags |= S_IMMUTABLE;
1229 inode->i_flags &= ~S_IMMUTABLE;
1230 if (flags & EXT3_NOATIME_FL)
1231 inode->i_flags |= S_NOATIME;
1233 inode->i_flags &= ~S_NOATIME;
1244 void ll_umount_begin(struct super_block *sb)
1246 struct ll_sb_info *sbi = ll_s2sbi(sb);
1247 struct obd_device *obd;
1248 struct obd_ioctl_data ioc_data = { 0 };
1250 CDEBUG(D_VFSTRACE, "VFS Op:\n");
1252 obd = class_exp2obd(sbi->ll_mdc_exp);
1254 CERROR("Invalid MDC connection handle "LPX64"\n",
1255 sbi->ll_mdc_exp->exp_handle.h_cookie);
1259 obd->obd_no_recov = 1;
1260 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_mdc_exp, sizeof ioc_data,
1263 obd = class_exp2obd(sbi->ll_osc_exp);
1265 CERROR("Invalid LOV connection handle "LPX64"\n",
1266 sbi->ll_osc_exp->exp_handle.h_cookie);
1271 obd->obd_no_recov = 1;
1272 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_osc_exp, sizeof ioc_data,
1275 /* Really, we'd like to wait until there are no requests outstanding,
1276 * and then continue. For now, we just invalidate the requests,
1277 * schedule, and hope.
1284 int ll_prep_inode(struct obd_export *exp, struct inode **inode,
1285 struct ptlrpc_request *req, int offset,struct super_block *sb)
1287 struct lustre_md md;
1290 rc = mdc_req2lustre_md(req, offset, exp, &md);
1295 ll_update_inode(*inode, md.body, md.lsm);
1298 *inode = ll_iget(sb, md.body->ino, &md);
1299 if (*inode == NULL || is_bad_inode(*inode)) {
1300 /* free the lsm if we allocated one above */
1302 obd_free_memmd(exp, &md.lsm);
1304 CERROR("new_inode -fatal: rc %d\n", rc);