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 "llite_internal.h"
37 kmem_cache_t *ll_file_data_slab;
39 extern struct address_space_operations ll_aops;
40 extern struct address_space_operations ll_dir_aops;
41 extern struct super_operations ll_super_operations;
44 #define log2(n) ffz(~(n))
47 struct ll_sb_info *lustre_init_sbi(struct super_block *sb)
49 struct ll_sb_info *sbi = NULL;
53 OBD_ALLOC(sbi, sizeof(*sbi));
57 spin_lock_init(&sbi->ll_pglist_lock);
58 INIT_LIST_HEAD(&sbi->ll_pglist);
59 sbi->ll_pglist_gen = 0;
60 INIT_LIST_HEAD(&sbi->ll_conn_chain);
61 INIT_HLIST_HEAD(&sbi->ll_orphan_dentry_list);
64 generate_random_uuid(uuid);
65 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
69 void lustre_free_sbi(struct super_block *sb)
71 struct ll_sb_info *sbi = ll_s2sbi(sb);
75 OBD_FREE(sbi, sizeof(*sbi));
80 int lustre_common_fill_super(struct super_block *sb, char *mdc, char *osc)
82 struct inode *root = 0;
83 struct ll_sb_info *sbi = ll_s2sbi(sb);
84 struct obd_device *obd;
85 struct ll_fid rootfid;
86 struct obd_statfs osfs;
87 struct ptlrpc_request *request = NULL;
88 struct lustre_handle osc_conn = {0, };
89 struct lustre_handle mdc_conn = {0, };
94 obd = class_name2obd(mdc);
96 CERROR("MDC %s: not setup or attached\n", mdc);
100 if (proc_lustre_fs_root) {
101 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
104 CERROR("could not register mount in /proc/lustre");
107 mdc_init_ea_size(obd, osc);
109 err = obd_connect(&mdc_conn, obd, &sbi->ll_sb_uuid);
111 CERROR("An MDS (mdc %s) is performing recovery, of which this"
112 " client is not a part. Please wait for recovery to "
113 "complete, abort, or time out.\n", mdc);
116 CERROR("cannot connect to %s: rc = %d\n", mdc, err);
119 sbi->ll_mdc_exp = class_conn2export(&mdc_conn);
121 err = obd_statfs(obd, &osfs, jiffies - HZ);
125 LASSERT(osfs.os_bsize);
126 sb->s_blocksize = osfs.os_bsize;
127 sb->s_blocksize_bits = log2(osfs.os_bsize);
128 sb->s_magic = LL_SUPER_MAGIC;
129 sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
131 devno = get_uuid2int(sbi2mdc(sbi)->cl_import->imp_target_uuid.uuid,
132 strlen(sbi2mdc(sbi)->cl_import->imp_target_uuid.uuid));
135 obd = class_name2obd(osc);
137 CERROR("OSC %s: not setup or attached\n", osc);
141 err = obd_connect(&osc_conn, obd, &sbi->ll_sb_uuid);
143 CERROR("An OST (osc %s) is performing recovery, of which this"
144 " client is not a part. Please wait for recovery to "
145 "complete, abort, or time out.\n", osc);
148 CERROR("cannot connect to %s: rc = %d\n", osc, err);
151 sbi->ll_osc_exp = class_conn2export(&osc_conn);
153 err = mdc_getstatus(sbi->ll_mdc_exp, &rootfid);
155 CERROR("cannot mds_connect: rc = %d\n", err);
158 CDEBUG(D_SUPER, "rootfid "LPU64"\n", rootfid.id);
159 sbi->ll_rootino = rootfid.id;
161 sb->s_op = &lustre_super_operations;
164 * XXX: move this to after cbd setup? */
165 err = mdc_getattr(sbi->ll_mdc_exp, &rootfid,
166 OBD_MD_FLNOTOBD|OBD_MD_FLBLOCKS, 0, &request);
168 CERROR("mdc_getattr failed for root: rc = %d\n", err);
172 err = mdc_req2lustre_md(request, 0, sbi->ll_osc_exp, &md);
174 CERROR("failed to understand root inode md: rc = %d\n",err);
175 ptlrpc_req_finished (request);
179 LASSERT(sbi->ll_rootino != 0);
180 root = ll_iget(sb, sbi->ll_rootino, &md);
182 ptlrpc_req_finished(request);
184 if (root == NULL || is_bad_inode(root)) {
185 /* XXX might need iput() for bad inode */
186 CERROR("lustre_lite: bad iget4 for root\n");
187 GOTO(out_root, err = -EBADF);
190 err = ll_close_thread_start(&sbi->ll_lcq);
192 CERROR("cannot start close thread: rc %d\n", err);
196 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0))
197 #warning "Please fix this"
199 /* bug 2805 - set VM readahead to zero */
200 vm_max_readahead = vm_min_readahead = 0;
202 sb->s_root = d_alloc_root(root);
209 obd_disconnect(sbi->ll_osc_exp, 0);
211 obd_disconnect(sbi->ll_mdc_exp, 0);
213 lprocfs_unregister_mountpoint(sbi);
217 void lustre_common_put_super(struct super_block *sb)
219 struct ll_sb_info *sbi = ll_s2sbi(sb);
220 struct hlist_node *tmp, *next;
223 ll_close_thread_shutdown(sbi->ll_lcq);
225 list_del(&sbi->ll_conn_chain);
226 obd_disconnect(sbi->ll_osc_exp, 0);
228 lprocfs_unregister_mountpoint(sbi);
229 if (sbi->ll_proc_root) {
230 lprocfs_remove(sbi->ll_proc_root);
231 sbi->ll_proc_root = NULL;
234 obd_disconnect(sbi->ll_mdc_exp, 0);
236 // We do this to get rid of orphaned dentries. That is not really trw.
237 spin_lock(&dcache_lock);
238 hlist_for_each_safe(tmp, next, &sbi->ll_orphan_dentry_list) {
239 struct dentry *dentry = hlist_entry(tmp, struct dentry, d_hash);
240 shrink_dcache_parent(dentry);
242 spin_unlock(&dcache_lock);
247 char *ll_read_opt(const char *opt, char *data)
253 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
254 if (strncmp(opt, data, strlen(opt)))
256 if ((value = strchr(data, '=')) == NULL)
260 OBD_ALLOC(retval, strlen(value) + 1);
262 CERROR("out of memory!\n");
266 memcpy(retval, value, strlen(value)+1);
267 CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
271 int ll_set_opt(const char *opt, char *data, int fl)
275 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
276 if (strncmp(opt, data, strlen(opt)))
282 void ll_options(char *options, char **ost, char **mdc, int *flags)
285 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
286 char *opt_ptr = options;
295 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
296 for (this_char = strtok (options, ",");
298 this_char = strtok (NULL, ",")) {
300 while ((this_char = strsep (&opt_ptr, ",")) != NULL) {
302 CDEBUG(D_SUPER, "this_char %s\n", this_char);
303 if (!*ost && (*ost = ll_read_opt("osc", this_char)))
305 if (!*mdc && (*mdc = ll_read_opt("mdc", this_char)))
307 if (!(*flags & LL_SBI_NOLCK) &&
308 ((*flags) = (*flags) |
309 ll_set_opt("nolock", this_char,
316 void ll_lli_init(struct ll_inode_info *lli)
318 sema_init(&lli->lli_open_sem, 1);
320 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
321 spin_lock_init(&lli->lli_lock);
322 INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
325 int ll_fill_super(struct super_block *sb, void *data, int silent)
327 struct ll_sb_info *sbi;
333 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
335 sbi = lustre_init_sbi(sb);
339 sbi->ll_flags |= LL_SBI_READAHEAD;
340 ll_options(data, &osc, &mdc, &sbi->ll_flags);
344 GOTO(out, err = -EINVAL);
349 GOTO(out, err = -EINVAL);
352 err = lustre_common_fill_super(sb, mdc, osc);
358 OBD_FREE(mdc, strlen(mdc) + 1);
360 OBD_FREE(osc, strlen(osc) + 1);
363 } /* ll_read_super */
365 void ll_put_super(struct super_block *sb)
369 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
371 lustre_common_put_super(sb);
378 int lustre_process_log(struct lustre_mount_data *lmd, char * profile,
379 struct config_llog_instance *cfg, int allow_recov)
381 struct lustre_cfg lcfg;
382 struct portals_cfg pcfg;
383 char * peer = "MDS_PEER_UUID";
384 struct obd_device *obd;
385 struct lustre_handle mdc_conn = {0, };
386 struct obd_export *exp;
387 char * name = "mdc_dev";
389 struct obd_uuid mdc_uuid;
390 struct llog_ctxt *ctxt;
395 if (lmd_bad_magic(lmd))
398 generate_random_uuid(uuid);
399 class_uuid_unparse(uuid, &mdc_uuid);
401 if (lmd->lmd_local_nid) {
402 PCFG_INIT(pcfg, NAL_CMD_REGISTER_MYNID);
403 pcfg.pcfg_nal = lmd->lmd_nal;
404 pcfg.pcfg_nid = lmd->lmd_local_nid;
405 err = kportal_nal_cmd(&pcfg);
410 if (lmd->lmd_nal == SOCKNAL) {
411 PCFG_INIT(pcfg, NAL_CMD_ADD_AUTOCONN);
412 pcfg.pcfg_nal = lmd->lmd_nal;
413 pcfg.pcfg_nid = lmd->lmd_server_nid;
414 pcfg.pcfg_id = lmd->lmd_server_ipaddr;
415 pcfg.pcfg_misc = lmd->lmd_port;
416 pcfg.pcfg_size = 8388608;
417 pcfg.pcfg_flags = 0x4; /*share*/
418 err = kportal_nal_cmd(&pcfg);
423 LCFG_INIT(lcfg, LCFG_ADD_UUID, name);
424 lcfg.lcfg_nid = lmd->lmd_server_nid;
425 lcfg.lcfg_inllen1 = strlen(peer) + 1;
426 lcfg.lcfg_inlbuf1 = peer;
427 lcfg.lcfg_nal = lmd->lmd_nal;
428 err = class_process_config(&lcfg);
430 GOTO(out_del_conn, err);
432 LCFG_INIT(lcfg, LCFG_ATTACH, name);
433 lcfg.lcfg_inlbuf1 = "mdc";
434 lcfg.lcfg_inllen1 = strlen(lcfg.lcfg_inlbuf1) + 1;
435 lcfg.lcfg_inlbuf2 = mdc_uuid.uuid;
436 lcfg.lcfg_inllen2 = strlen(lcfg.lcfg_inlbuf2) + 1;
437 err = class_process_config(&lcfg);
439 GOTO(out_del_uuid, err);
441 LCFG_INIT(lcfg, LCFG_SETUP, name);
442 lcfg.lcfg_inlbuf1 = lmd->lmd_mds;
443 lcfg.lcfg_inllen1 = strlen(lcfg.lcfg_inlbuf1) + 1;
444 lcfg.lcfg_inlbuf2 = peer;
445 lcfg.lcfg_inllen2 = strlen(lcfg.lcfg_inlbuf2) + 1;
446 err = class_process_config(&lcfg);
448 GOTO(out_detach, err);
450 obd = class_name2obd(name);
452 GOTO(out_cleanup, err = -EINVAL);
454 /* Disable initial recovery on this import */
455 err = obd_set_info(obd->obd_self_export,
456 strlen("initial_recov"), "initial_recov",
457 sizeof(allow_recov), &allow_recov);
459 GOTO(out_cleanup, err);
461 err = obd_connect(&mdc_conn, obd, &mdc_uuid);
463 CERROR("cannot connect to %s: rc = %d\n", lmd->lmd_mds, err);
464 GOTO(out_cleanup, err);
467 exp = class_conn2export(&mdc_conn);
469 ctxt = llog_get_context(exp->exp_obd, LLOG_CONFIG_REPL_CTXT);
471 rc = class_config_parse_llog(ctxt, profile, cfg);
474 * For debugging, it's useful to just dump the log
476 rc = class_config_dump_llog(ctxt, profile, cfg);
479 CERROR("class_config_parse_llog failed: rc = %d\n", rc);
482 err = obd_disconnect(exp, 0);
485 LCFG_INIT(lcfg, LCFG_CLEANUP, name);
486 err = class_process_config(&lcfg);
491 LCFG_INIT(lcfg, LCFG_DETACH, name);
492 err = class_process_config(&lcfg);
497 LCFG_INIT(lcfg, LCFG_DEL_UUID, name);
498 lcfg.lcfg_inllen1 = strlen(peer) + 1;
499 lcfg.lcfg_inlbuf1 = peer;
500 err = class_process_config(&lcfg);
503 if (lmd->lmd_nal == SOCKNAL) {
504 PCFG_INIT(pcfg, NAL_CMD_DEL_AUTOCONN);
505 pcfg.pcfg_nal = lmd->lmd_nal;
506 pcfg.pcfg_nid = lmd->lmd_server_nid;
507 pcfg.pcfg_id = lmd->lmd_server_ipaddr;
508 pcfg.pcfg_flags = 1; /*share*/
509 err = kportal_nal_cmd(&pcfg);
520 int lustre_fill_super(struct super_block *sb, void *data, int silent)
522 struct lustre_mount_data * lmd = data;
523 struct ll_sb_info *sbi;
529 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
530 if (lmd_bad_magic(lmd))
533 sbi = lustre_init_sbi(sb);
537 sbi->ll_flags |= LL_SBI_READAHEAD;
539 if (lmd->lmd_profile) {
540 struct lustre_profile *lprof;
541 struct config_llog_instance cfg;
545 CERROR("no mds name\n");
546 GOTO(out_free, err = -EINVAL);
549 OBD_ALLOC(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
550 if (sbi->ll_lmd == NULL)
551 GOTO(out_free, err = -ENOMEM);
552 memcpy(sbi->ll_lmd, lmd, sizeof(*lmd));
554 /* generate a string unique to this super, let's try
555 the address of the super itself.*/
556 len = (sizeof(sb) * 2) + 1;
557 OBD_ALLOC(sbi->ll_instance, len);
558 if (sbi->ll_instance == NULL)
559 GOTO(out_free, err = -ENOMEM);
560 sprintf(sbi->ll_instance, "%p", sb);
562 cfg.cfg_instance = sbi->ll_instance;
563 cfg.cfg_uuid = sbi->ll_sb_uuid;
564 cfg.cfg_local_nid = lmd->lmd_local_nid;
565 err = lustre_process_log(lmd, lmd->lmd_profile, &cfg, 0);
567 CERROR("Unable to process log: %s\n", lmd->lmd_profile);
572 lprof = class_get_profile(lmd->lmd_profile);
574 CERROR("No profile found: %s\n", lmd->lmd_profile);
575 GOTO(out_free, err = -EINVAL);
578 OBD_FREE(osc, strlen(osc) + 1);
579 OBD_ALLOC(osc, strlen(lprof->lp_osc) +
580 strlen(sbi->ll_instance) + 2);
581 sprintf(osc, "%s-%s", lprof->lp_osc, sbi->ll_instance);
584 OBD_FREE(mdc, strlen(mdc) + 1);
585 OBD_ALLOC(mdc, strlen(lprof->lp_mdc) +
586 strlen(sbi->ll_instance) + 2);
587 sprintf(mdc, "%s-%s", lprof->lp_mdc, sbi->ll_instance);
592 GOTO(out_free, err = -EINVAL);
597 GOTO(out_free, err = -EINVAL);
600 err = lustre_common_fill_super(sb, mdc, osc);
607 OBD_FREE(mdc, strlen(mdc) + 1);
609 OBD_FREE(osc, strlen(osc) + 1);
615 int len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean")+1;
618 if (sbi->ll_instance != NULL) {
620 struct config_llog_instance cfg;
622 cfg.cfg_instance = sbi->ll_instance;
623 cfg.cfg_uuid = sbi->ll_sb_uuid;
625 OBD_ALLOC(cln_prof, len);
626 sprintf(cln_prof, "%s-clean", sbi->ll_lmd->lmd_profile);
628 err = lustre_process_log(sbi->ll_lmd, cln_prof, &cfg,
631 CERROR("Unable to process log: %s\n", cln_prof);
632 OBD_FREE(cln_prof, len);
633 OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance)+ 1);
635 OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
640 } /* lustre_fill_super */
642 static void lustre_manual_cleanup(struct ll_sb_info *sbi)
644 struct lustre_cfg lcfg;
645 struct obd_device *obd;
648 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) != NULL)
652 LCFG_INIT(lcfg, LCFG_CLEANUP, obd->obd_name);
653 err = class_process_config(&lcfg);
655 CERROR("cleanup failed: %s\n", obd->obd_name);
659 LCFG_INIT(lcfg, LCFG_DETACH, obd->obd_name);
660 err = class_process_config(&lcfg);
662 CERROR("detach failed: %s\n", obd->obd_name);
667 if (sbi->ll_lmd != NULL)
668 class_del_profile(sbi->ll_lmd->lmd_profile);
671 void lustre_put_super(struct super_block *sb)
673 struct obd_device *obd;
674 struct ll_sb_info *sbi = ll_s2sbi(sb);
675 int force_umount = 0;
678 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
679 obd = class_exp2obd(sbi->ll_mdc_exp);
681 force_umount = obd->obd_no_recov;
684 lustre_common_put_super(sb);
686 if (sbi->ll_lmd != NULL) {
688 int len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean")+1;
690 struct config_llog_instance cfg;
693 CERROR("force umount, doing manual cleanup\n");
694 lustre_manual_cleanup(sbi);
698 cfg.cfg_instance = sbi->ll_instance;
699 cfg.cfg_uuid = sbi->ll_sb_uuid;
701 OBD_ALLOC(cln_prof, len);
702 sprintf(cln_prof, "%s-clean", sbi->ll_lmd->lmd_profile);
704 err = lustre_process_log(sbi->ll_lmd, cln_prof, &cfg, 0);
706 CERROR("Unable to process log: %s, doing manual cleanup"
708 lustre_manual_cleanup(sbi);
711 OBD_FREE(cln_prof, len);
713 OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
714 OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance) + 1);
720 } /* lustre_put_super */
723 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
726 l_lock(&lock->l_resource->lr_namespace->ns_lock);
727 if (lock->l_ast_data)
728 inode = igrab(lock->l_ast_data);
731 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
735 static int null_if_equal(struct ldlm_lock *lock, void *data)
737 if (data == lock->l_ast_data)
738 lock->l_ast_data = NULL;
740 if (lock->l_req_mode != lock->l_granted_mode)
741 return LDLM_ITER_STOP;
743 return LDLM_ITER_CONTINUE;
746 void ll_clear_inode(struct inode *inode)
749 struct ll_inode_info *lli = ll_i2info(inode);
750 struct ll_sb_info *sbi = ll_i2sbi(inode);
753 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
754 inode->i_generation, inode);
756 ll_inode2fid(&fid, inode);
757 clear_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &(ll_i2info(inode)->lli_flags));
758 mdc_change_cbdata(sbi->ll_mdc_exp, &fid, null_if_equal, inode);
761 obd_change_cbdata(sbi->ll_osc_exp, lli->lli_smd,
762 null_if_equal, inode);
765 obd_free_memmd(sbi->ll_osc_exp, &lli->lli_smd);
769 if (lli->lli_symlink_name) {
770 OBD_FREE(lli->lli_symlink_name,
771 strlen(lli->lli_symlink_name) + 1);
772 lli->lli_symlink_name = NULL;
778 /* If this inode has objects allocated to it (lsm != NULL), then the OST
779 * object(s) determine the file size and mtime. Otherwise, the MDS will
780 * keep these values until such a time that objects are allocated for it.
781 * We do the MDS operations first, as it is checking permissions for us.
782 * We don't to the MDS RPC if there is nothing that we want to store there,
783 * otherwise there is no harm in updating mtime/atime on the MDS if we are
784 * going to do an RPC anyways.
786 * If we are doing a truncate, we will send the mtime and ctime updates
787 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
788 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
791 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
793 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
794 struct ll_sb_info *sbi = ll_i2sbi(inode);
795 struct ptlrpc_request *request = NULL;
796 struct mdc_op_data op_data;
797 int ia_valid = attr->ia_valid;
801 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu\n", inode->i_ino);
802 lprocfs_counter_incr(ll_i2sbi(inode)->ll_stats, LPROC_LL_SETATTR);
804 if (ia_valid & ATTR_SIZE) {
805 if (attr->ia_size > ll_file_maxbytes(inode)) {
806 CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
807 attr->ia_size, ll_file_maxbytes(inode));
811 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
814 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
815 if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
816 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
820 /* We mark all of the fields "set" so MDS/OST does not re-set them */
821 if (attr->ia_valid & ATTR_CTIME) {
822 attr->ia_ctime = CURRENT_TIME;
823 attr->ia_valid |= ATTR_CTIME_SET;
825 if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
826 attr->ia_atime = CURRENT_TIME;
827 attr->ia_valid |= ATTR_ATIME_SET;
829 if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
830 attr->ia_mtime = CURRENT_TIME;
831 attr->ia_valid |= ATTR_MTIME_SET;
834 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
835 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
836 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
837 LTIME_S(CURRENT_TIME));
839 attr->ia_valid &= ~ATTR_SIZE;
841 /* If only OST attributes being set on objects, don't do MDS RPC.
842 * In that case, we need to check permissions and update the local
843 * inode ourselves so we can call obdo_from_inode() always. */
844 if (ia_valid & (lsm ? ~(ATTR_SIZE | ATTR_FROM_OPEN | ATTR_RAW) : ~0)) {
846 ll_prepare_mdc_op_data(&op_data, inode, NULL, NULL, 0, 0);
848 rc = mdc_setattr(sbi->ll_mdc_exp, &op_data,
849 attr, NULL, 0, NULL, 0, &request);
852 ptlrpc_req_finished(request);
853 if (rc != -EPERM && rc != -EACCES)
854 CERROR("mdc_setattr fails: rc = %d\n", rc);
858 rc = mdc_req2lustre_md(request, 0, sbi->ll_osc_exp, &md);
860 ptlrpc_req_finished(request);
864 /* Won't invoke vmtruncate as we already cleared ATTR_SIZE,
865 * but needed to set timestamps backwards on utime. */
866 inode_setattr(inode, attr);
867 ll_update_inode(inode, md.body, md.lsm);
868 ptlrpc_req_finished(request);
870 if (!lsm || !S_ISREG(inode->i_mode)) {
871 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
875 /* The OST doesn't check permissions, but the alternative is
876 * a gratuitous RPC to the MDS. We already rely on the client
877 * to do read/write/truncate permission checks, so is mtime OK?
879 if (ia_valid & (ATTR_MTIME | ATTR_ATIME)) {
880 /* from sys_utime() */
881 if (!(ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET))) {
882 if (current->fsuid != inode->i_uid &&
883 (rc=ll_permission(inode,MAY_WRITE,NULL))!=0)
886 /* from inode_change_ok() */
887 if (current->fsuid != inode->i_uid &&
888 !capable(CAP_FOWNER))
893 /* Won't invoke vmtruncate, as we already cleared ATTR_SIZE */
894 inode_setattr(inode, attr);
897 /* We really need to get our PW lock before we change inode->i_size.
898 * If we don't we can race with other i_size updaters on our node, like
899 * ll_file_read. We can also race with i_size propogation to other
900 * nodes through dirtying and writeback of final cached pages. This
901 * last one is especially bad for racing o_append users on other
903 if (ia_valid & ATTR_SIZE) {
904 ldlm_policy_data_t policy = { .l_extent = {attr->ia_size,
906 struct lustre_handle lockh = { 0 };
907 int err, ast_flags = 0;
908 /* XXX when we fix the AST intents to pass the discard-range
909 * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
911 if (attr->ia_size == 0)
912 ast_flags = LDLM_AST_DISCARD_DATA;
914 /* bug 1639: avoid write/truncate i_sem/DLM deadlock */
915 LASSERT(atomic_read(&inode->i_sem.count) <= 0);
917 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy, &lockh,
923 rc = vmtruncate(inode, attr->ia_size);
925 /* We need to drop the semaphore here, because this unlock may
926 * result in a cancellation, which will need the i_sem */
928 /* unlock now as we don't mind others file lockers racing with
929 * the mds updates below? */
930 err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
933 CERROR("ll_extent_unlock failed: %d\n", err);
937 } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
940 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
941 inode->i_ino, LTIME_S(attr->ia_mtime));
942 oa.o_id = lsm->lsm_object_id;
943 oa.o_valid = OBD_MD_FLID;
944 obdo_from_inode(&oa, inode, OBD_MD_FLTYPE | OBD_MD_FLATIME |
945 OBD_MD_FLMTIME | OBD_MD_FLCTIME);
946 rc = obd_setattr(sbi->ll_osc_exp, &oa, lsm, NULL);
948 CERROR("obd_setattr fails: rc=%d\n", rc);
953 int ll_setattr(struct dentry *de, struct iattr *attr)
955 LBUG(); /* code is unused, but leave this in case of VFS changes */
959 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
960 unsigned long max_age)
962 struct ll_sb_info *sbi = ll_s2sbi(sb);
963 struct obd_statfs obd_osfs;
967 rc = obd_statfs(class_exp2obd(sbi->ll_mdc_exp), osfs, max_age);
969 CERROR("mdc_statfs fails: rc = %d\n", rc);
973 osfs->os_type = sb->s_magic;
975 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
976 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
978 rc = obd_statfs(class_exp2obd(sbi->ll_osc_exp), &obd_osfs, max_age);
980 CERROR("obd_statfs fails: rc = %d\n", rc);
984 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
985 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
988 osfs->os_blocks = obd_osfs.os_blocks;
989 osfs->os_bfree = obd_osfs.os_bfree;
990 osfs->os_bavail = obd_osfs.os_bavail;
992 /* If we don't have as many objects free on the OST as inodes
993 * on the MDS, we reduce the total number of inodes to
994 * compensate, so that the "inodes in use" number is correct.
996 if (obd_osfs.os_ffree < osfs->os_ffree) {
997 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
999 osfs->os_ffree = obd_osfs.os_ffree;
1005 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1007 struct obd_statfs osfs;
1010 CDEBUG(D_VFSTRACE, "VFS Op:\n");
1011 lprocfs_counter_incr(ll_s2sbi(sb)->ll_stats, LPROC_LL_STAFS);
1013 /* For now we will always get up-to-date statfs values, but in the
1014 * future we may allow some amount of caching on the client (e.g.
1015 * from QOS or lprocfs updates). */
1016 rc = ll_statfs_internal(sb, &osfs, jiffies - 1);
1020 statfs_unpack(sfs, &osfs);
1022 if (sizeof(sfs->f_blocks) == 4) {
1023 while (osfs.os_blocks > ~0UL) {
1026 osfs.os_blocks >>= 1;
1027 osfs.os_bfree >>= 1;
1028 osfs.os_bavail >>= 1;
1032 sfs->f_blocks = osfs.os_blocks;
1033 sfs->f_bfree = osfs.os_bfree;
1034 sfs->f_bavail = osfs.os_bavail;
1039 void dump_lsm(int level, struct lov_stripe_md *lsm)
1041 CDEBUG(level, "objid "LPX64", maxbytes "LPX64", magic 0x%08X, "
1042 "stripe_size %u, stripe_count %u\n",
1043 lsm->lsm_object_id, lsm->lsm_maxbytes, lsm->lsm_magic,
1044 lsm->lsm_stripe_size, lsm->lsm_stripe_count);
1047 void ll_update_inode(struct inode *inode, struct mds_body *body,
1048 struct lov_stripe_md *lsm)
1050 struct ll_inode_info *lli = ll_i2info(inode);
1052 LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1054 if (lli->lli_smd == NULL) {
1056 lli->lli_maxbytes = lsm->lsm_maxbytes;
1057 if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1058 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1060 if (memcmp(lli->lli_smd, lsm, sizeof(*lsm))) {
1061 CERROR("lsm mismatch for inode %ld\n",
1063 CERROR("lli_smd:\n");
1064 dump_lsm(D_ERROR, lli->lli_smd);
1066 dump_lsm(D_ERROR, lsm);
1070 /* bug 2844 - limit i_blksize for broken user-space apps */
1071 LASSERTF(lsm->lsm_xfersize != 0, "%lu\n", lsm->lsm_xfersize);
1072 inode->i_blksize = min(lsm->lsm_xfersize, LL_MAX_BLKSIZE);
1073 if (lli->lli_smd != lsm)
1074 obd_free_memmd(ll_i2obdexp(inode), &lsm);
1077 if (body->valid & OBD_MD_FLID)
1078 inode->i_ino = body->ino;
1079 if (body->valid & OBD_MD_FLATIME)
1080 LTIME_S(inode->i_atime) = body->atime;
1081 if (body->valid & OBD_MD_FLMTIME &&
1082 body->mtime > LTIME_S(inode->i_mtime)) {
1083 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to %u\n",
1084 inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
1085 LTIME_S(inode->i_mtime) = body->mtime;
1087 if (body->valid & OBD_MD_FLCTIME &&
1088 body->ctime > LTIME_S(inode->i_ctime))
1089 LTIME_S(inode->i_ctime) = body->ctime;
1090 if (body->valid & OBD_MD_FLMODE)
1091 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1092 if (body->valid & OBD_MD_FLTYPE)
1093 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1094 if (body->valid & OBD_MD_FLUID)
1095 inode->i_uid = body->uid;
1096 if (body->valid & OBD_MD_FLGID)
1097 inode->i_gid = body->gid;
1098 if (body->valid & OBD_MD_FLFLAGS)
1099 inode->i_flags = body->flags;
1100 if (body->valid & OBD_MD_FLNLINK)
1101 inode->i_nlink = body->nlink;
1102 if (body->valid & OBD_MD_FLGENER)
1103 inode->i_generation = body->generation;
1104 if (body->valid & OBD_MD_FLRDEV)
1105 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1106 inode->i_rdev = body->rdev;
1108 inode->i_rdev = old_decode_dev(body->rdev);
1110 if (body->valid & OBD_MD_FLSIZE)
1111 inode->i_size = body->size;
1112 if (body->valid & OBD_MD_FLBLOCKS)
1113 inode->i_blocks = body->blocks;
1115 if (body->valid & OBD_MD_FLSIZE)
1116 set_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &lli->lli_flags);
1119 void ll_read_inode2(struct inode *inode, void *opaque)
1121 struct lustre_md *md = opaque;
1122 struct ll_inode_info *lli = ll_i2info(inode);
1125 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1126 inode->i_generation, inode);
1130 LASSERT(!lli->lli_smd);
1132 /* Core attributes from the MDS first. This is a new inode, and
1133 * the VFS doesn't zero times in the core inode so we have to do
1134 * it ourselves. They will be overwritten by either MDS or OST
1135 * attributes - we just need to make sure they aren't newer. */
1136 LTIME_S(inode->i_mtime) = 0;
1137 LTIME_S(inode->i_atime) = 0;
1138 LTIME_S(inode->i_ctime) = 0;
1139 ll_update_inode(inode, md->body, md->lsm);
1141 /* OIDEBUG(inode); */
1143 if (S_ISREG(inode->i_mode)) {
1144 inode->i_op = &ll_file_inode_operations;
1145 inode->i_fop = &ll_file_operations;
1146 inode->i_mapping->a_ops = &ll_aops;
1148 } else if (S_ISDIR(inode->i_mode)) {
1149 inode->i_op = &ll_dir_inode_operations;
1150 inode->i_fop = &ll_dir_operations;
1151 inode->i_mapping->a_ops = &ll_dir_aops;
1153 } else if (S_ISLNK(inode->i_mode)) {
1154 inode->i_op = &ll_fast_symlink_inode_operations;
1157 inode->i_op = &ll_special_inode_operations;
1158 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
1159 #warning "need to fix this for 2.6 also"
1160 init_special_inode(inode, inode->i_mode,
1161 kdev_t_to_nr(inode->i_rdev));
1163 init_special_inode(inode, inode->i_mode, inode->i_rdev);
1165 lli->ll_save_ifop = inode->i_fop;
1166 if (S_ISCHR(inode->i_mode))
1167 inode->i_fop = &ll_special_chr_inode_fops;
1168 else if (S_ISBLK(inode->i_mode))
1169 inode->i_fop = &ll_special_blk_inode_fops;
1170 else if (S_ISFIFO(inode->i_mode))
1171 inode->i_fop = &ll_special_fifo_inode_fops;
1172 else if (S_ISSOCK(inode->i_mode))
1173 inode->i_fop = &ll_special_sock_inode_fops;
1174 CWARN("saved %p, replaced with %p\n", lli->ll_save_ifop,
1176 if (lli->ll_save_ifop->owner)
1177 CWARN("%p has owner %p\n", lli->ll_save_ifop,
1178 lli->ll_save_ifop->owner);
1184 int ll_iocontrol(struct inode *inode, struct file *file,
1185 unsigned int cmd, unsigned long arg)
1187 struct ll_sb_info *sbi = ll_i2sbi(inode);
1188 struct ptlrpc_request *req = NULL;
1193 case EXT3_IOC_GETFLAGS: {
1195 unsigned long valid = OBD_MD_FLFLAGS;
1196 struct mds_body *body;
1198 ll_inode2fid(&fid, inode);
1199 rc = mdc_getattr(sbi->ll_mdc_exp, &fid, valid, 0, &req);
1201 CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1205 body = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*body));
1207 if (body->flags & S_APPEND)
1208 flags |= EXT3_APPEND_FL;
1209 if (body->flags & S_IMMUTABLE)
1210 flags |= EXT3_IMMUTABLE_FL;
1211 if (body->flags & S_NOATIME)
1212 flags |= EXT3_NOATIME_FL;
1214 ptlrpc_req_finished (req);
1216 RETURN(put_user(flags, (int *)arg));
1218 case EXT3_IOC_SETFLAGS: {
1219 struct mdc_op_data op_data;
1222 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1224 if (get_user(flags, (int *)arg))
1231 ll_prepare_mdc_op_data(&op_data, inode, NULL, NULL, 0, 0);
1233 memset(&attr, 0x0, sizeof(attr));
1234 attr.ia_attr_flags = flags;
1235 attr.ia_valid |= ATTR_ATTR_FLAG;
1237 rc = mdc_setattr(sbi->ll_mdc_exp, &op_data,
1238 &attr, NULL, 0, NULL, 0, &req);
1240 ptlrpc_req_finished(req);
1241 if (rc != -EPERM && rc != -EACCES)
1242 CERROR("mdc_setattr fails: rc = %d\n", rc);
1246 ptlrpc_req_finished(req);
1248 oa->o_id = lsm->lsm_object_id;
1249 oa->o_flags = flags;
1250 oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS;
1252 rc = obd_setattr(sbi->ll_osc_exp, oa, lsm, NULL);
1255 if (rc != -EPERM && rc != -EACCES)
1256 CERROR("mdc_setattr fails: rc = %d\n", rc);
1260 if (flags & EXT3_APPEND_FL)
1261 inode->i_flags |= S_APPEND;
1263 inode->i_flags &= ~S_APPEND;
1264 if (flags & EXT3_IMMUTABLE_FL)
1265 inode->i_flags |= S_IMMUTABLE;
1267 inode->i_flags &= ~S_IMMUTABLE;
1268 if (flags & EXT3_NOATIME_FL)
1269 inode->i_flags |= S_NOATIME;
1271 inode->i_flags &= ~S_NOATIME;
1282 void ll_umount_begin(struct super_block *sb)
1284 struct ll_sb_info *sbi = ll_s2sbi(sb);
1285 struct obd_device *obd;
1286 struct obd_ioctl_data ioc_data = { 0 };
1288 CDEBUG(D_VFSTRACE, "VFS Op:\n");
1290 obd = class_exp2obd(sbi->ll_mdc_exp);
1292 CERROR("Invalid MDC connection handle "LPX64"\n",
1293 sbi->ll_mdc_exp->exp_handle.h_cookie);
1297 obd->obd_no_recov = 1;
1298 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_mdc_exp, sizeof ioc_data,
1301 obd = class_exp2obd(sbi->ll_osc_exp);
1303 CERROR("Invalid LOV connection handle "LPX64"\n",
1304 sbi->ll_osc_exp->exp_handle.h_cookie);
1309 obd->obd_no_recov = 1;
1310 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_osc_exp, sizeof ioc_data,
1313 /* Really, we'd like to wait until there are no requests outstanding,
1314 * and then continue. For now, we just invalidate the requests,
1315 * schedule, and hope.
1322 int ll_prep_inode(struct obd_export *exp, struct inode **inode,
1323 struct ptlrpc_request *req, int offset,struct super_block *sb)
1325 struct lustre_md md;
1328 rc = mdc_req2lustre_md(req, offset, exp, &md);
1333 ll_update_inode(*inode, md.body, md.lsm);
1336 *inode = ll_iget(sb, md.body->ino, &md);
1337 if (*inode == NULL || is_bad_inode(*inode)) {
1338 /* free the lsm if we allocated one above */
1340 obd_free_memmd(exp, &md.lsm);
1342 CERROR("new_inode -fatal: rc %d\n", rc);