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 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0))
196 #warning "Please fix this"
198 /* bug 2805 - set VM readahead to zero */
199 vm_max_readahead = vm_min_readahead = 0;
201 sb->s_root = d_alloc_root(root);
208 obd_disconnect(sbi->ll_osc_exp, 0);
210 obd_disconnect(sbi->ll_mdc_exp, 0);
212 lprocfs_unregister_mountpoint(sbi);
216 void lustre_common_put_super(struct super_block *sb)
218 struct ll_sb_info *sbi = ll_s2sbi(sb);
219 struct hlist_node *tmp, *next;
222 ll_close_thread_shutdown(sbi->ll_lcq);
224 list_del(&sbi->ll_conn_chain);
225 obd_disconnect(sbi->ll_osc_exp, 0);
227 lprocfs_unregister_mountpoint(sbi);
228 if (sbi->ll_proc_root) {
229 lprocfs_remove(sbi->ll_proc_root);
230 sbi->ll_proc_root = NULL;
233 obd_disconnect(sbi->ll_mdc_exp, 0);
235 // We do this to get rid of orphaned dentries. That is not really trw.
236 spin_lock(&dcache_lock);
237 hlist_for_each_safe(tmp, next, &sbi->ll_orphan_dentry_list) {
238 struct dentry *dentry = hlist_entry(tmp, struct dentry, d_hash);
239 shrink_dcache_parent(dentry);
241 spin_unlock(&dcache_lock);
246 char *ll_read_opt(const char *opt, char *data)
252 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
253 if (strncmp(opt, data, strlen(opt)))
255 if ((value = strchr(data, '=')) == NULL)
259 OBD_ALLOC(retval, strlen(value) + 1);
261 CERROR("out of memory!\n");
265 memcpy(retval, value, strlen(value)+1);
266 CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
270 int ll_set_opt(const char *opt, char *data, int fl)
274 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
275 if (strncmp(opt, data, strlen(opt)))
281 void ll_options(char *options, char **ost, char **mdc, int *flags)
284 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
285 char *opt_ptr = options;
294 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
295 for (this_char = strtok (options, ",");
297 this_char = strtok (NULL, ",")) {
299 while ((this_char = strsep (&opt_ptr, ",")) != NULL) {
301 CDEBUG(D_SUPER, "this_char %s\n", this_char);
302 if (!*ost && (*ost = ll_read_opt("osc", this_char)))
304 if (!*mdc && (*mdc = ll_read_opt("mdc", this_char)))
306 if (!(*flags & LL_SBI_NOLCK) &&
307 ((*flags) = (*flags) |
308 ll_set_opt("nolock", this_char,
315 void ll_lli_init(struct ll_inode_info *lli)
317 sema_init(&lli->lli_open_sem, 1);
319 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
320 spin_lock_init(&lli->lli_lock);
321 INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
324 int ll_fill_super(struct super_block *sb, void *data, int silent)
326 struct ll_sb_info *sbi;
332 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
334 sbi = lustre_init_sbi(sb);
338 sbi->ll_flags |= LL_SBI_READAHEAD;
339 ll_options(data, &osc, &mdc, &sbi->ll_flags);
343 GOTO(out, err = -EINVAL);
348 GOTO(out, err = -EINVAL);
351 err = lustre_common_fill_super(sb, mdc, osc);
357 OBD_FREE(mdc, strlen(mdc) + 1);
359 OBD_FREE(osc, strlen(osc) + 1);
362 } /* ll_read_super */
364 void ll_put_super(struct super_block *sb)
368 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
370 lustre_common_put_super(sb);
377 int lustre_process_log(struct lustre_mount_data *lmd, char * profile,
378 struct config_llog_instance *cfg, int allow_recov)
380 struct lustre_cfg lcfg;
381 struct portals_cfg pcfg;
382 char * peer = "MDS_PEER_UUID";
383 struct obd_device *obd;
384 struct lustre_handle mdc_conn = {0, };
385 struct obd_export *exp;
386 char * name = "mdc_dev";
388 struct obd_uuid mdc_uuid;
389 struct llog_ctxt *ctxt;
394 if (lmd_bad_magic(lmd))
397 generate_random_uuid(uuid);
398 class_uuid_unparse(uuid, &mdc_uuid);
400 if (lmd->lmd_local_nid) {
401 PCFG_INIT(pcfg, NAL_CMD_REGISTER_MYNID);
402 pcfg.pcfg_nal = lmd->lmd_nal;
403 pcfg.pcfg_nid = lmd->lmd_local_nid;
404 err = kportal_nal_cmd(&pcfg);
409 if (lmd->lmd_nal == SOCKNAL) {
410 PCFG_INIT(pcfg, NAL_CMD_ADD_AUTOCONN);
411 pcfg.pcfg_nal = lmd->lmd_nal;
412 pcfg.pcfg_nid = lmd->lmd_server_nid;
413 pcfg.pcfg_id = lmd->lmd_server_ipaddr;
414 pcfg.pcfg_misc = lmd->lmd_port;
415 pcfg.pcfg_size = 8388608;
416 pcfg.pcfg_flags = 0x4; /*share*/
417 err = kportal_nal_cmd(&pcfg);
422 LCFG_INIT(lcfg, LCFG_ADD_UUID, name);
423 lcfg.lcfg_nid = lmd->lmd_server_nid;
424 lcfg.lcfg_inllen1 = strlen(peer) + 1;
425 lcfg.lcfg_inlbuf1 = peer;
426 lcfg.lcfg_nal = lmd->lmd_nal;
427 err = class_process_config(&lcfg);
429 GOTO(out_del_conn, err);
431 LCFG_INIT(lcfg, LCFG_ATTACH, name);
432 lcfg.lcfg_inlbuf1 = "mdc";
433 lcfg.lcfg_inllen1 = strlen(lcfg.lcfg_inlbuf1) + 1;
434 lcfg.lcfg_inlbuf2 = mdc_uuid.uuid;
435 lcfg.lcfg_inllen2 = strlen(lcfg.lcfg_inlbuf2) + 1;
436 err = class_process_config(&lcfg);
438 GOTO(out_del_uuid, err);
440 LCFG_INIT(lcfg, LCFG_SETUP, name);
441 lcfg.lcfg_inlbuf1 = lmd->lmd_mds;
442 lcfg.lcfg_inllen1 = strlen(lcfg.lcfg_inlbuf1) + 1;
443 lcfg.lcfg_inlbuf2 = peer;
444 lcfg.lcfg_inllen2 = strlen(lcfg.lcfg_inlbuf2) + 1;
445 err = class_process_config(&lcfg);
447 GOTO(out_detach, err);
449 obd = class_name2obd(name);
451 GOTO(out_cleanup, err = -EINVAL);
453 /* Disable initial recovery on this import */
454 err = obd_set_info(obd->obd_self_export,
455 strlen("initial_recov"), "initial_recov",
456 sizeof(allow_recov), &allow_recov);
458 GOTO(out_cleanup, err);
460 err = obd_connect(&mdc_conn, obd, &mdc_uuid);
462 CERROR("cannot connect to %s: rc = %d\n", lmd->lmd_mds, err);
463 GOTO(out_cleanup, err);
466 exp = class_conn2export(&mdc_conn);
468 ctxt = llog_get_context(exp->exp_obd, LLOG_CONFIG_REPL_CTXT);
470 rc = class_config_parse_llog(ctxt, profile, cfg);
473 * For debugging, it's useful to just dump the log
475 rc = class_config_dump_llog(ctxt, profile, cfg);
478 CERROR("class_config_parse_llog failed: rc = %d\n", rc);
481 err = obd_disconnect(exp, 0);
484 LCFG_INIT(lcfg, LCFG_CLEANUP, name);
485 err = class_process_config(&lcfg);
490 LCFG_INIT(lcfg, LCFG_DETACH, name);
491 err = class_process_config(&lcfg);
496 LCFG_INIT(lcfg, LCFG_DEL_UUID, name);
497 lcfg.lcfg_inllen1 = strlen(peer) + 1;
498 lcfg.lcfg_inlbuf1 = peer;
499 err = class_process_config(&lcfg);
502 if (lmd->lmd_nal == SOCKNAL) {
503 PCFG_INIT(pcfg, NAL_CMD_DEL_AUTOCONN);
504 pcfg.pcfg_nal = lmd->lmd_nal;
505 pcfg.pcfg_nid = lmd->lmd_server_nid;
506 pcfg.pcfg_id = lmd->lmd_server_ipaddr;
507 pcfg.pcfg_flags = 1; /*share*/
508 err = kportal_nal_cmd(&pcfg);
519 int lustre_fill_super(struct super_block *sb, void *data, int silent)
521 struct lustre_mount_data * lmd = data;
522 struct ll_sb_info *sbi;
528 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
529 if (lmd_bad_magic(lmd))
532 sbi = lustre_init_sbi(sb);
536 sbi->ll_flags |= LL_SBI_READAHEAD;
538 if (lmd->lmd_profile) {
539 struct lustre_profile *lprof;
540 struct config_llog_instance cfg;
544 CERROR("no mds name\n");
545 GOTO(out_free, err = -EINVAL);
548 OBD_ALLOC(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
549 if (sbi->ll_lmd == NULL)
550 GOTO(out_free, err = -ENOMEM);
551 memcpy(sbi->ll_lmd, lmd, sizeof(*lmd));
553 /* generate a string unique to this super, let's try
554 the address of the super itself.*/
555 len = (sizeof(sb) * 2) + 1;
556 OBD_ALLOC(sbi->ll_instance, len);
557 if (sbi->ll_instance == NULL)
558 GOTO(out_free, err = -ENOMEM);
559 sprintf(sbi->ll_instance, "%p", sb);
561 cfg.cfg_instance = sbi->ll_instance;
562 cfg.cfg_uuid = sbi->ll_sb_uuid;
563 cfg.cfg_local_nid = lmd->lmd_local_nid;
564 err = lustre_process_log(lmd, lmd->lmd_profile, &cfg, 0);
566 CERROR("Unable to process log: %s\n", lmd->lmd_profile);
571 lprof = class_get_profile(lmd->lmd_profile);
573 CERROR("No profile found: %s\n", lmd->lmd_profile);
574 GOTO(out_free, err = -EINVAL);
577 OBD_FREE(osc, strlen(osc) + 1);
578 OBD_ALLOC(osc, strlen(lprof->lp_osc) +
579 strlen(sbi->ll_instance) + 2);
580 sprintf(osc, "%s-%s", lprof->lp_osc, sbi->ll_instance);
583 OBD_FREE(mdc, strlen(mdc) + 1);
584 OBD_ALLOC(mdc, strlen(lprof->lp_mdc) +
585 strlen(sbi->ll_instance) + 2);
586 sprintf(mdc, "%s-%s", lprof->lp_mdc, sbi->ll_instance);
591 GOTO(out_free, err = -EINVAL);
596 GOTO(out_free, err = -EINVAL);
599 err = lustre_common_fill_super(sb, mdc, osc);
606 OBD_FREE(mdc, strlen(mdc) + 1);
608 OBD_FREE(osc, strlen(osc) + 1);
614 int len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean")+1;
617 if (sbi->ll_instance != NULL) {
619 struct config_llog_instance cfg;
621 cfg.cfg_instance = sbi->ll_instance;
622 cfg.cfg_uuid = sbi->ll_sb_uuid;
624 OBD_ALLOC(cln_prof, len);
625 sprintf(cln_prof, "%s-clean", sbi->ll_lmd->lmd_profile);
627 err = lustre_process_log(sbi->ll_lmd, cln_prof, &cfg,
630 CERROR("Unable to process log: %s\n", cln_prof);
631 OBD_FREE(cln_prof, len);
632 OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance)+ 1);
634 OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
639 } /* lustre_fill_super */
641 static void lustre_manual_cleanup(struct ll_sb_info *sbi)
643 struct lustre_cfg lcfg;
644 struct obd_device *obd;
647 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) != NULL)
651 LCFG_INIT(lcfg, LCFG_CLEANUP, obd->obd_name);
652 err = class_process_config(&lcfg);
654 CERROR("cleanup failed: %s\n", obd->obd_name);
658 LCFG_INIT(lcfg, LCFG_DETACH, obd->obd_name);
659 err = class_process_config(&lcfg);
661 CERROR("detach failed: %s\n", obd->obd_name);
666 if (sbi->ll_lmd != NULL)
667 class_del_profile(sbi->ll_lmd->lmd_profile);
670 void lustre_put_super(struct super_block *sb)
672 struct obd_device *obd;
673 struct ll_sb_info *sbi = ll_s2sbi(sb);
674 int force_umount = 0;
677 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
678 obd = class_exp2obd(sbi->ll_mdc_exp);
680 force_umount = obd->obd_no_recov;
683 lustre_common_put_super(sb);
685 if (sbi->ll_lmd != NULL) {
687 int len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean")+1;
689 struct config_llog_instance cfg;
692 CERROR("force umount, doing manual cleanup\n");
693 lustre_manual_cleanup(sbi);
697 cfg.cfg_instance = sbi->ll_instance;
698 cfg.cfg_uuid = sbi->ll_sb_uuid;
700 OBD_ALLOC(cln_prof, len);
701 sprintf(cln_prof, "%s-clean", sbi->ll_lmd->lmd_profile);
703 err = lustre_process_log(sbi->ll_lmd, cln_prof, &cfg, 0);
705 CERROR("Unable to process log: %s, doing manual cleanup"
707 lustre_manual_cleanup(sbi);
710 OBD_FREE(cln_prof, len);
712 OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
713 OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance) + 1);
719 } /* lustre_put_super */
722 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
725 l_lock(&lock->l_resource->lr_namespace->ns_lock);
726 if (lock->l_ast_data)
727 inode = igrab(lock->l_ast_data);
730 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
734 static int null_if_equal(struct ldlm_lock *lock, void *data)
736 if (data == lock->l_ast_data)
737 lock->l_ast_data = NULL;
739 if (lock->l_req_mode != lock->l_granted_mode)
740 return LDLM_ITER_STOP;
742 return LDLM_ITER_CONTINUE;
745 void ll_clear_inode(struct inode *inode)
748 struct ll_inode_info *lli = ll_i2info(inode);
749 struct ll_sb_info *sbi = ll_i2sbi(inode);
752 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
753 inode->i_generation, inode);
755 ll_inode2fid(&fid, inode);
756 clear_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &(ll_i2info(inode)->lli_flags));
757 mdc_change_cbdata(sbi->ll_mdc_exp, &fid, null_if_equal, inode);
760 obd_change_cbdata(sbi->ll_osc_exp, lli->lli_smd,
761 null_if_equal, inode);
764 obd_free_memmd(sbi->ll_osc_exp, &lli->lli_smd);
768 if (lli->lli_symlink_name) {
769 OBD_FREE(lli->lli_symlink_name,
770 strlen(lli->lli_symlink_name) + 1);
771 lli->lli_symlink_name = NULL;
777 /* If this inode has objects allocated to it (lsm != NULL), then the OST
778 * object(s) determine the file size and mtime. Otherwise, the MDS will
779 * keep these values until such a time that objects are allocated for it.
780 * We do the MDS operations first, as it is checking permissions for us.
781 * We don't to the MDS RPC if there is nothing that we want to store there,
782 * otherwise there is no harm in updating mtime/atime on the MDS if we are
783 * going to do an RPC anyways.
785 * If we are doing a truncate, we will send the mtime and ctime updates
786 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
787 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
790 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
792 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
793 struct ll_sb_info *sbi = ll_i2sbi(inode);
794 struct ptlrpc_request *request = NULL;
795 struct mdc_op_data op_data;
796 int ia_valid = attr->ia_valid;
800 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu\n", inode->i_ino);
801 lprocfs_counter_incr(ll_i2sbi(inode)->ll_stats, LPROC_LL_SETATTR);
803 if (ia_valid & ATTR_SIZE) {
804 if (attr->ia_size > ll_file_maxbytes(inode)) {
805 CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
806 attr->ia_size, ll_file_maxbytes(inode));
810 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
813 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
814 if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
815 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
819 /* We mark all of the fields "set" so MDS/OST does not re-set them */
820 if (attr->ia_valid & ATTR_CTIME) {
821 attr->ia_ctime = CURRENT_TIME;
822 attr->ia_valid |= ATTR_CTIME_SET;
824 if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
825 attr->ia_atime = CURRENT_TIME;
826 attr->ia_valid |= ATTR_ATIME_SET;
828 if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
829 attr->ia_mtime = CURRENT_TIME;
830 attr->ia_valid |= ATTR_MTIME_SET;
833 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
834 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
835 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
836 LTIME_S(CURRENT_TIME));
838 attr->ia_valid &= ~ATTR_SIZE;
840 /* If only OST attributes being set on objects, don't do MDS RPC.
841 * In that case, we need to check permissions and update the local
842 * inode ourselves so we can call obdo_from_inode() always. */
843 if (ia_valid & (lsm ? ~(ATTR_SIZE | ATTR_FROM_OPEN | ATTR_RAW) : ~0)) {
845 ll_prepare_mdc_op_data(&op_data, inode, NULL, NULL, 0, 0);
847 rc = mdc_setattr(sbi->ll_mdc_exp, &op_data,
848 attr, NULL, 0, NULL, 0, &request);
851 ptlrpc_req_finished(request);
852 if (rc != -EPERM && rc != -EACCES)
853 CERROR("mdc_setattr fails: rc = %d\n", rc);
857 rc = mdc_req2lustre_md(request, 0, sbi->ll_osc_exp, &md);
859 ptlrpc_req_finished(request);
863 /* Won't invoke vmtruncate as we already cleared ATTR_SIZE,
864 * but needed to set timestamps backwards on utime. */
865 inode_setattr(inode, attr);
866 ll_update_inode(inode, md.body, md.lsm);
867 ptlrpc_req_finished(request);
869 if (!lsm || !S_ISREG(inode->i_mode)) {
870 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
874 /* The OST doesn't check permissions, but the alternative is
875 * a gratuitous RPC to the MDS. We already rely on the client
876 * to do read/write/truncate permission checks, so is mtime OK?
878 if (ia_valid & (ATTR_MTIME | ATTR_ATIME)) {
879 /* from sys_utime() */
880 if (!(ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET))) {
881 if (current->fsuid != inode->i_uid &&
882 (rc=ll_permission(inode,MAY_WRITE,NULL))!=0)
885 /* from inode_change_ok() */
886 if (current->fsuid != inode->i_uid &&
887 !capable(CAP_FOWNER))
892 /* Won't invoke vmtruncate, as we already cleared ATTR_SIZE */
893 inode_setattr(inode, attr);
896 /* We really need to get our PW lock before we change inode->i_size.
897 * If we don't we can race with other i_size updaters on our node, like
898 * ll_file_read. We can also race with i_size propogation to other
899 * nodes through dirtying and writeback of final cached pages. This
900 * last one is especially bad for racing o_append users on other
902 if (ia_valid & ATTR_SIZE) {
903 ldlm_policy_data_t policy = { .l_extent = {attr->ia_size,
905 struct lustre_handle lockh = { 0 };
906 int err, ast_flags = 0;
907 /* XXX when we fix the AST intents to pass the discard-range
908 * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
910 if (attr->ia_size == 0)
911 ast_flags = LDLM_AST_DISCARD_DATA;
913 /* bug 1639: avoid write/truncate i_sem/DLM deadlock */
914 LASSERT(atomic_read(&inode->i_sem.count) <= 0);
916 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy, &lockh,
922 rc = vmtruncate(inode, attr->ia_size);
924 /* We need to drop the semaphore here, because this unlock may
925 * result in a cancellation, which will need the i_sem */
927 /* unlock now as we don't mind others file lockers racing with
928 * the mds updates below? */
929 err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
932 CERROR("ll_extent_unlock failed: %d\n", err);
936 } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
939 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
940 inode->i_ino, LTIME_S(attr->ia_mtime));
941 oa.o_id = lsm->lsm_object_id;
942 oa.o_valid = OBD_MD_FLID;
943 obdo_from_inode(&oa, inode, OBD_MD_FLTYPE | OBD_MD_FLATIME |
944 OBD_MD_FLMTIME | OBD_MD_FLCTIME);
945 rc = obd_setattr(sbi->ll_osc_exp, &oa, lsm, NULL);
947 CERROR("obd_setattr fails: rc=%d\n", rc);
952 int ll_setattr(struct dentry *de, struct iattr *attr)
954 LBUG(); /* code is unused, but leave this in case of VFS changes */
958 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
959 unsigned long max_age)
961 struct ll_sb_info *sbi = ll_s2sbi(sb);
962 struct obd_statfs obd_osfs;
966 rc = obd_statfs(class_exp2obd(sbi->ll_mdc_exp), osfs, max_age);
968 CERROR("mdc_statfs fails: rc = %d\n", rc);
972 osfs->os_type = sb->s_magic;
974 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
975 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
977 rc = obd_statfs(class_exp2obd(sbi->ll_osc_exp), &obd_osfs, max_age);
979 CERROR("obd_statfs fails: rc = %d\n", rc);
983 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
984 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
987 osfs->os_blocks = obd_osfs.os_blocks;
988 osfs->os_bfree = obd_osfs.os_bfree;
989 osfs->os_bavail = obd_osfs.os_bavail;
991 /* If we don't have as many objects free on the OST as inodes
992 * on the MDS, we reduce the total number of inodes to
993 * compensate, so that the "inodes in use" number is correct.
995 if (obd_osfs.os_ffree < osfs->os_ffree) {
996 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
998 osfs->os_ffree = obd_osfs.os_ffree;
1004 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1006 struct obd_statfs osfs;
1009 CDEBUG(D_VFSTRACE, "VFS Op:\n");
1010 lprocfs_counter_incr(ll_s2sbi(sb)->ll_stats, LPROC_LL_STAFS);
1012 /* For now we will always get up-to-date statfs values, but in the
1013 * future we may allow some amount of caching on the client (e.g.
1014 * from QOS or lprocfs updates). */
1015 rc = ll_statfs_internal(sb, &osfs, jiffies - 1);
1019 statfs_unpack(sfs, &osfs);
1021 if (sizeof(sfs->f_blocks) == 4) {
1022 while (osfs.os_blocks > ~0UL) {
1025 osfs.os_blocks >>= 1;
1026 osfs.os_bfree >>= 1;
1027 osfs.os_bavail >>= 1;
1031 sfs->f_blocks = osfs.os_blocks;
1032 sfs->f_bfree = osfs.os_bfree;
1033 sfs->f_bavail = osfs.os_bavail;
1038 void dump_lsm(int level, struct lov_stripe_md *lsm)
1040 CDEBUG(level, "objid "LPX64", maxbytes "LPX64", magic 0x%08X, "
1041 "stripe_size %u, stripe_count %u\n",
1042 lsm->lsm_object_id, lsm->lsm_maxbytes, lsm->lsm_magic,
1043 lsm->lsm_stripe_size, lsm->lsm_stripe_count);
1046 void ll_update_inode(struct inode *inode, struct mds_body *body,
1047 struct lov_stripe_md *lsm)
1049 struct ll_inode_info *lli = ll_i2info(inode);
1051 LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1053 if (lli->lli_smd == NULL) {
1055 lli->lli_maxbytes = lsm->lsm_maxbytes;
1056 if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1057 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1059 if (memcmp(lli->lli_smd, lsm, sizeof(*lsm))) {
1060 CERROR("lsm mismatch for inode %ld\n",
1062 CERROR("lli_smd:\n");
1063 dump_lsm(D_ERROR, lli->lli_smd);
1065 dump_lsm(D_ERROR, lsm);
1069 /* bug 2844 - limit i_blksize for broken user-space apps */
1070 LASSERTF(lsm->lsm_xfersize != 0, "%lu\n", lsm->lsm_xfersize);
1071 inode->i_blksize = min(lsm->lsm_xfersize, LL_MAX_BLKSIZE);
1072 if (lli->lli_smd != lsm)
1073 obd_free_memmd(ll_i2obdexp(inode), &lsm);
1076 if (body->valid & OBD_MD_FLID)
1077 inode->i_ino = body->ino;
1078 if (body->valid & OBD_MD_FLATIME)
1079 LTIME_S(inode->i_atime) = body->atime;
1080 if (body->valid & OBD_MD_FLMTIME &&
1081 body->mtime > LTIME_S(inode->i_mtime)) {
1082 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to %u\n",
1083 inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
1084 LTIME_S(inode->i_mtime) = body->mtime;
1086 if (body->valid & OBD_MD_FLCTIME &&
1087 body->ctime > LTIME_S(inode->i_ctime))
1088 LTIME_S(inode->i_ctime) = body->ctime;
1089 if (body->valid & OBD_MD_FLMODE)
1090 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1091 if (body->valid & OBD_MD_FLTYPE)
1092 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1093 if (body->valid & OBD_MD_FLUID)
1094 inode->i_uid = body->uid;
1095 if (body->valid & OBD_MD_FLGID)
1096 inode->i_gid = body->gid;
1097 if (body->valid & OBD_MD_FLFLAGS)
1098 inode->i_flags = body->flags;
1099 if (body->valid & OBD_MD_FLNLINK)
1100 inode->i_nlink = body->nlink;
1101 if (body->valid & OBD_MD_FLGENER)
1102 inode->i_generation = body->generation;
1103 if (body->valid & OBD_MD_FLRDEV)
1104 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1105 inode->i_rdev = body->rdev;
1107 inode->i_rdev = old_decode_dev(body->rdev);
1109 if (body->valid & OBD_MD_FLSIZE)
1110 inode->i_size = body->size;
1111 if (body->valid & OBD_MD_FLBLOCKS)
1112 inode->i_blocks = body->blocks;
1114 if (body->valid & OBD_MD_FLSIZE)
1115 set_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &lli->lli_flags);
1118 void ll_read_inode2(struct inode *inode, void *opaque)
1120 struct lustre_md *md = opaque;
1121 struct ll_inode_info *lli = ll_i2info(inode);
1124 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1125 inode->i_generation, inode);
1129 LASSERT(!lli->lli_smd);
1131 /* Core attributes from the MDS first. This is a new inode, and
1132 * the VFS doesn't zero times in the core inode so we have to do
1133 * it ourselves. They will be overwritten by either MDS or OST
1134 * attributes - we just need to make sure they aren't newer. */
1135 LTIME_S(inode->i_mtime) = 0;
1136 LTIME_S(inode->i_atime) = 0;
1137 LTIME_S(inode->i_ctime) = 0;
1138 ll_update_inode(inode, md->body, md->lsm);
1140 /* OIDEBUG(inode); */
1142 if (S_ISREG(inode->i_mode)) {
1143 inode->i_op = &ll_file_inode_operations;
1144 inode->i_fop = &ll_file_operations;
1145 inode->i_mapping->a_ops = &ll_aops;
1147 } else if (S_ISDIR(inode->i_mode)) {
1148 inode->i_op = &ll_dir_inode_operations;
1149 inode->i_fop = &ll_dir_operations;
1150 inode->i_mapping->a_ops = &ll_dir_aops;
1152 } else if (S_ISLNK(inode->i_mode)) {
1153 inode->i_op = &ll_fast_symlink_inode_operations;
1156 inode->i_op = &ll_special_inode_operations;
1157 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
1158 init_special_inode(inode, inode->i_mode,
1159 kdev_t_to_nr(inode->i_rdev));
1161 init_special_inode(inode, inode->i_mode, inode->i_rdev);
1167 int ll_iocontrol(struct inode *inode, struct file *file,
1168 unsigned int cmd, unsigned long arg)
1170 struct ll_sb_info *sbi = ll_i2sbi(inode);
1171 struct ptlrpc_request *req = NULL;
1176 case EXT3_IOC_GETFLAGS: {
1178 unsigned long valid = OBD_MD_FLFLAGS;
1179 struct mds_body *body;
1181 ll_inode2fid(&fid, inode);
1182 rc = mdc_getattr(sbi->ll_mdc_exp, &fid, valid, 0, &req);
1184 CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1188 body = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*body));
1190 if (body->flags & S_APPEND)
1191 flags |= EXT3_APPEND_FL;
1192 if (body->flags & S_IMMUTABLE)
1193 flags |= EXT3_IMMUTABLE_FL;
1194 if (body->flags & S_NOATIME)
1195 flags |= EXT3_NOATIME_FL;
1197 ptlrpc_req_finished (req);
1199 RETURN(put_user(flags, (int *)arg));
1201 case EXT3_IOC_SETFLAGS: {
1202 struct mdc_op_data op_data;
1205 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1207 if (get_user(flags, (int *)arg))
1214 ll_prepare_mdc_op_data(&op_data, inode, NULL, NULL, 0, 0);
1216 memset(&attr, 0x0, sizeof(attr));
1217 attr.ia_attr_flags = flags;
1218 attr.ia_valid |= ATTR_ATTR_FLAG;
1220 rc = mdc_setattr(sbi->ll_mdc_exp, &op_data,
1221 &attr, NULL, 0, NULL, 0, &req);
1223 ptlrpc_req_finished(req);
1224 if (rc != -EPERM && rc != -EACCES)
1225 CERROR("mdc_setattr fails: rc = %d\n", rc);
1229 ptlrpc_req_finished(req);
1231 oa->o_id = lsm->lsm_object_id;
1232 oa->o_flags = flags;
1233 oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS;
1235 rc = obd_setattr(sbi->ll_osc_exp, oa, lsm, NULL);
1238 if (rc != -EPERM && rc != -EACCES)
1239 CERROR("mdc_setattr fails: rc = %d\n", rc);
1243 if (flags & EXT3_APPEND_FL)
1244 inode->i_flags |= S_APPEND;
1246 inode->i_flags &= ~S_APPEND;
1247 if (flags & EXT3_IMMUTABLE_FL)
1248 inode->i_flags |= S_IMMUTABLE;
1250 inode->i_flags &= ~S_IMMUTABLE;
1251 if (flags & EXT3_NOATIME_FL)
1252 inode->i_flags |= S_NOATIME;
1254 inode->i_flags &= ~S_NOATIME;
1265 void ll_umount_begin(struct super_block *sb)
1267 struct ll_sb_info *sbi = ll_s2sbi(sb);
1268 struct obd_device *obd;
1269 struct obd_ioctl_data ioc_data = { 0 };
1271 CDEBUG(D_VFSTRACE, "VFS Op:\n");
1273 obd = class_exp2obd(sbi->ll_mdc_exp);
1275 CERROR("Invalid MDC connection handle "LPX64"\n",
1276 sbi->ll_mdc_exp->exp_handle.h_cookie);
1280 obd->obd_no_recov = 1;
1281 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_mdc_exp, sizeof ioc_data,
1284 obd = class_exp2obd(sbi->ll_osc_exp);
1286 CERROR("Invalid LOV connection handle "LPX64"\n",
1287 sbi->ll_osc_exp->exp_handle.h_cookie);
1292 obd->obd_no_recov = 1;
1293 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_osc_exp, sizeof ioc_data,
1296 /* Really, we'd like to wait until there are no requests outstanding,
1297 * and then continue. For now, we just invalidate the requests,
1298 * schedule, and hope.
1305 int ll_prep_inode(struct obd_export *exp, struct inode **inode,
1306 struct ptlrpc_request *req, int offset,struct super_block *sb)
1308 struct lustre_md md;
1311 rc = mdc_req2lustre_md(req, offset, exp, &md);
1316 ll_update_inode(*inode, md.body, md.lsm);
1319 *inode = ll_iget(sb, md.body->ino, &md);
1320 if (*inode == NULL || is_bad_inode(*inode)) {
1321 /* free the lsm if we allocated one above */
1323 obd_free_memmd(exp, &md.lsm);
1325 CERROR("new_inode -fatal: rc %d\n", rc);