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 /* bug 2805 - set VM readahead to zero */
196 vm_max_readahead = vm_min_readahead = 0;
197 sb->s_root = d_alloc_root(root);
204 obd_disconnect(sbi->ll_osc_exp, 0);
206 obd_disconnect(sbi->ll_mdc_exp, 0);
208 lprocfs_unregister_mountpoint(sbi);
212 void lustre_common_put_super(struct super_block *sb)
214 struct ll_sb_info *sbi = ll_s2sbi(sb);
215 struct hlist_node *tmp, *next;
218 ll_close_thread_shutdown(sbi->ll_lcq);
220 list_del(&sbi->ll_conn_chain);
221 obd_disconnect(sbi->ll_osc_exp, 0);
223 lprocfs_unregister_mountpoint(sbi);
224 if (sbi->ll_proc_root) {
225 lprocfs_remove(sbi->ll_proc_root);
226 sbi->ll_proc_root = NULL;
229 obd_disconnect(sbi->ll_mdc_exp, 0);
231 // We do this to get rid of orphaned dentries. That is not really trw.
232 spin_lock(&dcache_lock);
233 hlist_for_each_safe(tmp, next, &sbi->ll_orphan_dentry_list) {
234 struct dentry *dentry = hlist_entry(tmp, struct dentry, d_hash);
235 shrink_dcache_parent(dentry);
237 spin_unlock(&dcache_lock);
242 char *ll_read_opt(const char *opt, char *data)
248 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
249 if (strncmp(opt, data, strlen(opt)))
251 if ((value = strchr(data, '=')) == NULL)
255 OBD_ALLOC(retval, strlen(value) + 1);
257 CERROR("out of memory!\n");
261 memcpy(retval, value, strlen(value)+1);
262 CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
266 int ll_set_opt(const char *opt, char *data, int fl)
270 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
271 if (strncmp(opt, data, strlen(opt)))
277 void ll_options(char *options, char **ost, char **mdc, int *flags)
280 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
281 char *opt_ptr = options;
290 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
291 for (this_char = strtok (options, ",");
293 this_char = strtok (NULL, ",")) {
295 while ((this_char = strsep (&opt_ptr, ",")) != NULL) {
297 CDEBUG(D_SUPER, "this_char %s\n", this_char);
298 if (!*ost && (*ost = ll_read_opt("osc", this_char)))
300 if (!*mdc && (*mdc = ll_read_opt("mdc", this_char)))
302 if (!(*flags & LL_SBI_NOLCK) &&
303 ((*flags) = (*flags) |
304 ll_set_opt("nolock", this_char,
311 void ll_lli_init(struct ll_inode_info *lli)
313 sema_init(&lli->lli_open_sem, 1);
315 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
316 spin_lock_init(&lli->lli_lock);
317 INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
320 int ll_fill_super(struct super_block *sb, void *data, int silent)
322 struct ll_sb_info *sbi;
328 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
330 sbi = lustre_init_sbi(sb);
334 sbi->ll_flags |= LL_SBI_READAHEAD;
335 ll_options(data, &osc, &mdc, &sbi->ll_flags);
339 GOTO(out, err = -EINVAL);
344 GOTO(out, err = -EINVAL);
347 err = lustre_common_fill_super(sb, mdc, osc);
353 OBD_FREE(mdc, strlen(mdc) + 1);
355 OBD_FREE(osc, strlen(osc) + 1);
358 } /* ll_read_super */
360 void ll_put_super(struct super_block *sb)
364 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
366 lustre_common_put_super(sb);
373 int lustre_process_log(struct lustre_mount_data *lmd, char * profile,
374 struct config_llog_instance *cfg, int allow_recov)
376 struct lustre_cfg lcfg;
377 struct portals_cfg pcfg;
378 char * peer = "MDS_PEER_UUID";
379 struct obd_device *obd;
380 struct lustre_handle mdc_conn = {0, };
381 struct obd_export *exp;
382 char * name = "mdc_dev";
384 struct obd_uuid mdc_uuid;
385 struct llog_ctxt *ctxt;
390 if (lmd_bad_magic(lmd))
393 generate_random_uuid(uuid);
394 class_uuid_unparse(uuid, &mdc_uuid);
396 if (lmd->lmd_local_nid) {
397 PCFG_INIT(pcfg, NAL_CMD_REGISTER_MYNID);
398 pcfg.pcfg_nal = lmd->lmd_nal;
399 pcfg.pcfg_nid = lmd->lmd_local_nid;
400 err = kportal_nal_cmd(&pcfg);
405 if (lmd->lmd_nal == SOCKNAL) {
406 PCFG_INIT(pcfg, NAL_CMD_ADD_AUTOCONN);
407 pcfg.pcfg_nal = lmd->lmd_nal;
408 pcfg.pcfg_nid = lmd->lmd_server_nid;
409 pcfg.pcfg_id = lmd->lmd_server_ipaddr;
410 pcfg.pcfg_misc = lmd->lmd_port;
411 pcfg.pcfg_size = 8388608;
412 pcfg.pcfg_flags = 0x4; /*share*/
413 err = kportal_nal_cmd(&pcfg);
418 LCFG_INIT(lcfg, LCFG_ADD_UUID, name);
419 lcfg.lcfg_nid = lmd->lmd_server_nid;
420 lcfg.lcfg_inllen1 = strlen(peer) + 1;
421 lcfg.lcfg_inlbuf1 = peer;
422 lcfg.lcfg_nal = lmd->lmd_nal;
423 err = class_process_config(&lcfg);
425 GOTO(out_del_conn, err);
427 LCFG_INIT(lcfg, LCFG_ATTACH, name);
428 lcfg.lcfg_inlbuf1 = "mdc";
429 lcfg.lcfg_inllen1 = strlen(lcfg.lcfg_inlbuf1) + 1;
430 lcfg.lcfg_inlbuf2 = mdc_uuid.uuid;
431 lcfg.lcfg_inllen2 = strlen(lcfg.lcfg_inlbuf2) + 1;
432 err = class_process_config(&lcfg);
434 GOTO(out_del_uuid, err);
436 LCFG_INIT(lcfg, LCFG_SETUP, name);
437 lcfg.lcfg_inlbuf1 = lmd->lmd_mds;
438 lcfg.lcfg_inllen1 = strlen(lcfg.lcfg_inlbuf1) + 1;
439 lcfg.lcfg_inlbuf2 = peer;
440 lcfg.lcfg_inllen2 = strlen(lcfg.lcfg_inlbuf2) + 1;
441 err = class_process_config(&lcfg);
443 GOTO(out_detach, err);
445 obd = class_name2obd(name);
447 GOTO(out_cleanup, err = -EINVAL);
449 /* Disable initial recovery on this import */
450 err = obd_set_info(obd->obd_self_export,
451 strlen("initial_recov"), "initial_recov",
452 sizeof(allow_recov), &allow_recov);
454 GOTO(out_cleanup, err);
456 err = obd_connect(&mdc_conn, obd, &mdc_uuid);
458 CERROR("cannot connect to %s: rc = %d\n", lmd->lmd_mds, err);
459 GOTO(out_cleanup, err);
462 exp = class_conn2export(&mdc_conn);
464 ctxt = llog_get_context(exp->exp_obd, LLOG_CONFIG_REPL_CTXT);
465 rc = class_config_parse_llog(ctxt, profile, cfg);
467 CERROR("class_config_parse_llog failed: rc = %d\n", rc);
470 err = obd_disconnect(exp, 0);
473 LCFG_INIT(lcfg, LCFG_CLEANUP, name);
474 err = class_process_config(&lcfg);
479 LCFG_INIT(lcfg, LCFG_DETACH, name);
480 err = class_process_config(&lcfg);
485 LCFG_INIT(lcfg, LCFG_DEL_UUID, name);
486 lcfg.lcfg_inllen1 = strlen(peer) + 1;
487 lcfg.lcfg_inlbuf1 = peer;
488 err = class_process_config(&lcfg);
491 if (lmd->lmd_nal == SOCKNAL) {
492 PCFG_INIT(pcfg, NAL_CMD_DEL_AUTOCONN);
493 pcfg.pcfg_nal = lmd->lmd_nal;
494 pcfg.pcfg_nid = lmd->lmd_server_nid;
495 pcfg.pcfg_id = lmd->lmd_server_ipaddr;
496 pcfg.pcfg_flags = 1; /*share*/
497 err = kportal_nal_cmd(&pcfg);
508 int lustre_fill_super(struct super_block *sb, void *data, int silent)
510 struct lustre_mount_data * lmd = data;
511 struct ll_sb_info *sbi;
517 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
518 if (lmd_bad_magic(lmd))
521 sbi = lustre_init_sbi(sb);
525 sbi->ll_flags |= LL_SBI_READAHEAD;
527 if (lmd->lmd_profile) {
528 struct lustre_profile *lprof;
529 struct config_llog_instance cfg;
533 CERROR("no mds name\n");
534 GOTO(out_free, err = -EINVAL);
537 OBD_ALLOC(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
538 if (sbi->ll_lmd == NULL)
539 GOTO(out_free, err = -ENOMEM);
540 memcpy(sbi->ll_lmd, lmd, sizeof(*lmd));
542 /* generate a string unique to this super, let's try
543 the address of the super itself.*/
544 len = (sizeof(sb) * 2) + 1;
545 OBD_ALLOC(sbi->ll_instance, len);
546 if (sbi->ll_instance == NULL)
547 GOTO(out_free, err = -ENOMEM);
548 sprintf(sbi->ll_instance, "%p", sb);
550 cfg.cfg_instance = sbi->ll_instance;
551 cfg.cfg_uuid = sbi->ll_sb_uuid;
552 cfg.cfg_local_nid = lmd->lmd_local_nid;
553 err = lustre_process_log(lmd, lmd->lmd_profile, &cfg, 0);
555 CERROR("Unable to process log: %s\n", lmd->lmd_profile);
560 lprof = class_get_profile(lmd->lmd_profile);
562 CERROR("No profile found: %s\n", lmd->lmd_profile);
563 GOTO(out_free, err = -EINVAL);
566 OBD_FREE(osc, strlen(osc) + 1);
567 OBD_ALLOC(osc, strlen(lprof->lp_osc) +
568 strlen(sbi->ll_instance) + 2);
569 sprintf(osc, "%s-%s", lprof->lp_osc, sbi->ll_instance);
572 OBD_FREE(mdc, strlen(mdc) + 1);
573 OBD_ALLOC(mdc, strlen(lprof->lp_mdc) +
574 strlen(sbi->ll_instance) + 2);
575 sprintf(mdc, "%s-%s", lprof->lp_mdc, sbi->ll_instance);
580 GOTO(out_free, err = -EINVAL);
585 GOTO(out_free, err = -EINVAL);
588 err = lustre_common_fill_super(sb, mdc, osc);
595 OBD_FREE(mdc, strlen(mdc) + 1);
597 OBD_FREE(osc, strlen(osc) + 1);
603 int len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean")+1;
606 if (sbi->ll_instance != NULL) {
608 struct config_llog_instance cfg;
610 cfg.cfg_instance = sbi->ll_instance;
611 cfg.cfg_uuid = sbi->ll_sb_uuid;
613 OBD_ALLOC(cln_prof, len);
614 sprintf(cln_prof, "%s-clean", sbi->ll_lmd->lmd_profile);
616 err = lustre_process_log(sbi->ll_lmd, cln_prof, &cfg,
619 CERROR("Unable to process log: %s\n", cln_prof);
620 OBD_FREE(cln_prof, len);
621 OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance)+ 1);
623 OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
628 } /* lustre_fill_super */
630 static void lustre_manual_cleanup(struct ll_sb_info *sbi)
632 struct lustre_cfg lcfg;
633 struct obd_device *obd;
636 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) != NULL)
640 LCFG_INIT(lcfg, LCFG_CLEANUP, obd->obd_name);
641 err = class_process_config(&lcfg);
643 CERROR("cleanup failed: %s\n", obd->obd_name);
647 LCFG_INIT(lcfg, LCFG_DETACH, obd->obd_name);
648 err = class_process_config(&lcfg);
650 CERROR("detach failed: %s\n", obd->obd_name);
655 if (sbi->ll_lmd != NULL)
656 class_del_profile(sbi->ll_lmd->lmd_profile);
659 void lustre_put_super(struct super_block *sb)
661 struct obd_device *obd;
662 struct ll_sb_info *sbi = ll_s2sbi(sb);
663 int force_umount = 0;
666 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
667 obd = class_exp2obd(sbi->ll_mdc_exp);
669 force_umount = obd->obd_no_recov;
672 lustre_common_put_super(sb);
674 if (sbi->ll_lmd != NULL) {
676 int len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean")+1;
678 struct config_llog_instance cfg;
681 CERROR("force umount, doing manual cleanup\n");
682 lustre_manual_cleanup(sbi);
686 cfg.cfg_instance = sbi->ll_instance;
687 cfg.cfg_uuid = sbi->ll_sb_uuid;
689 OBD_ALLOC(cln_prof, len);
690 sprintf(cln_prof, "%s-clean", sbi->ll_lmd->lmd_profile);
692 err = lustre_process_log(sbi->ll_lmd, cln_prof, &cfg, 0);
694 CERROR("Unable to process log: %s, doing manual cleanup"
696 lustre_manual_cleanup(sbi);
699 OBD_FREE(cln_prof, len);
701 OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
702 OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance) + 1);
708 } /* lustre_put_super */
711 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
714 l_lock(&lock->l_resource->lr_namespace->ns_lock);
715 if (lock->l_ast_data)
716 inode = igrab(lock->l_ast_data);
719 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
723 static int null_if_equal(struct ldlm_lock *lock, void *data)
725 if (data == lock->l_ast_data)
726 lock->l_ast_data = NULL;
728 if (lock->l_req_mode != lock->l_granted_mode)
729 return LDLM_ITER_STOP;
731 return LDLM_ITER_CONTINUE;
734 void ll_clear_inode(struct inode *inode)
737 struct ll_inode_info *lli = ll_i2info(inode);
738 struct ll_sb_info *sbi = ll_i2sbi(inode);
741 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
742 inode->i_generation, inode);
744 ll_inode2fid(&fid, inode);
745 clear_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &(ll_i2info(inode)->lli_flags));
746 mdc_change_cbdata(sbi->ll_mdc_exp, &fid, null_if_equal, inode);
749 obd_change_cbdata(sbi->ll_osc_exp, lli->lli_smd,
750 null_if_equal, inode);
753 obd_free_memmd(sbi->ll_osc_exp, &lli->lli_smd);
757 if (lli->lli_symlink_name) {
758 OBD_FREE(lli->lli_symlink_name,
759 strlen(lli->lli_symlink_name) + 1);
760 lli->lli_symlink_name = NULL;
766 /* If this inode has objects allocated to it (lsm != NULL), then the OST
767 * object(s) determine the file size and mtime. Otherwise, the MDS will
768 * keep these values until such a time that objects are allocated for it.
769 * We do the MDS operations first, as it is checking permissions for us.
770 * We don't to the MDS RPC if there is nothing that we want to store there,
771 * otherwise there is no harm in updating mtime/atime on the MDS if we are
772 * going to do an RPC anyways.
774 * If we are doing a truncate, we will send the mtime and ctime updates
775 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
776 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
779 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
781 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
782 struct ll_sb_info *sbi = ll_i2sbi(inode);
783 struct ptlrpc_request *request = NULL;
784 struct mdc_op_data op_data;
785 int ia_valid = attr->ia_valid;
789 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu\n", inode->i_ino);
790 lprocfs_counter_incr(ll_i2sbi(inode)->ll_stats, LPROC_LL_SETATTR);
792 if (ia_valid & ATTR_SIZE) {
793 if (attr->ia_size > ll_file_maxbytes(inode)) {
794 CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
795 attr->ia_size, ll_file_maxbytes(inode));
799 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
802 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
803 if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
804 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
808 /* We mark all of the fields "set" so MDS/OST does not re-set them */
809 if (attr->ia_valid & ATTR_CTIME) {
810 attr->ia_ctime = CURRENT_TIME;
811 attr->ia_valid |= ATTR_CTIME_SET;
813 if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
814 attr->ia_atime = CURRENT_TIME;
815 attr->ia_valid |= ATTR_ATIME_SET;
817 if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
818 attr->ia_mtime = CURRENT_TIME;
819 attr->ia_valid |= ATTR_MTIME_SET;
822 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
823 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
824 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
825 LTIME_S(CURRENT_TIME));
827 attr->ia_valid &= ~ATTR_SIZE;
829 /* If only OST attributes being set on objects, don't do MDS RPC.
830 * In that case, we need to check permissions and update the local
831 * inode ourselves so we can call obdo_from_inode() always. */
832 if (ia_valid & (lsm ? ~(ATTR_SIZE | ATTR_FROM_OPEN | ATTR_RAW) : ~0)) {
834 ll_prepare_mdc_op_data(&op_data, inode, NULL, NULL, 0, 0);
836 rc = mdc_setattr(sbi->ll_mdc_exp, &op_data,
837 attr, NULL, 0, NULL, 0, &request);
840 ptlrpc_req_finished(request);
841 if (rc != -EPERM && rc != -EACCES)
842 CERROR("mdc_setattr fails: rc = %d\n", rc);
846 rc = mdc_req2lustre_md(request, 0, sbi->ll_osc_exp, &md);
848 ptlrpc_req_finished(request);
852 /* Won't invoke vmtruncate as we already cleared ATTR_SIZE,
853 * but needed to set timestamps backwards on utime. */
854 inode_setattr(inode, attr);
855 ll_update_inode(inode, md.body, md.lsm);
856 ptlrpc_req_finished(request);
858 if (!lsm || !S_ISREG(inode->i_mode)) {
859 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
863 /* The OST doesn't check permissions, but the alternative is
864 * a gratuitous RPC to the MDS. We already rely on the client
865 * to do read/write/truncate permission checks, so is mtime OK?
867 if (ia_valid & (ATTR_MTIME | ATTR_ATIME)) {
868 /* from sys_utime() */
869 if (!(ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET))) {
870 if (current->fsuid != inode->i_uid &&
871 (rc=ll_permission(inode,MAY_WRITE,NULL))!=0)
874 /* from inode_change_ok() */
875 if (current->fsuid != inode->i_uid &&
876 !capable(CAP_FOWNER))
881 /* Won't invoke vmtruncate, as we already cleared ATTR_SIZE */
882 inode_setattr(inode, attr);
885 /* We really need to get our PW lock before we change inode->i_size.
886 * If we don't we can race with other i_size updaters on our node, like
887 * ll_file_read. We can also race with i_size propogation to other
888 * nodes through dirtying and writeback of final cached pages. This
889 * last one is especially bad for racing o_append users on other
891 if (ia_valid & ATTR_SIZE) {
892 ldlm_policy_data_t policy = { .l_extent = {attr->ia_size,
894 struct lustre_handle lockh = { 0 };
895 int err, ast_flags = 0;
896 /* XXX when we fix the AST intents to pass the discard-range
897 * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
899 if (attr->ia_size == 0)
900 ast_flags = LDLM_AST_DISCARD_DATA;
902 /* bug 1639: avoid write/truncate i_sem/DLM deadlock */
903 LASSERT(atomic_read(&inode->i_sem.count) <= 0);
905 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy, &lockh,
911 rc = vmtruncate(inode, attr->ia_size);
913 /* We need to drop the semaphore here, because this unlock may
914 * result in a cancellation, which will need the i_sem */
916 /* unlock now as we don't mind others file lockers racing with
917 * the mds updates below? */
918 err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
921 CERROR("ll_extent_unlock failed: %d\n", err);
925 } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
928 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
929 inode->i_ino, LTIME_S(attr->ia_mtime));
930 oa.o_id = lsm->lsm_object_id;
931 oa.o_valid = OBD_MD_FLID;
932 obdo_from_inode(&oa, inode, OBD_MD_FLTYPE | OBD_MD_FLATIME |
933 OBD_MD_FLMTIME | OBD_MD_FLCTIME);
934 rc = obd_setattr(sbi->ll_osc_exp, &oa, lsm, NULL);
936 CERROR("obd_setattr fails: rc=%d\n", rc);
941 int ll_setattr(struct dentry *de, struct iattr *attr)
943 LBUG(); /* code is unused, but leave this in case of VFS changes */
947 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
948 unsigned long max_age)
950 struct ll_sb_info *sbi = ll_s2sbi(sb);
951 struct obd_statfs obd_osfs;
955 rc = obd_statfs(class_exp2obd(sbi->ll_mdc_exp), osfs, max_age);
957 CERROR("mdc_statfs fails: rc = %d\n", rc);
961 osfs->os_type = sb->s_magic;
963 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
964 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
966 rc = obd_statfs(class_exp2obd(sbi->ll_osc_exp), &obd_osfs, max_age);
968 CERROR("obd_statfs fails: rc = %d\n", rc);
972 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
973 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
976 osfs->os_blocks = obd_osfs.os_blocks;
977 osfs->os_bfree = obd_osfs.os_bfree;
978 osfs->os_bavail = obd_osfs.os_bavail;
980 /* If we don't have as many objects free on the OST as inodes
981 * on the MDS, we reduce the total number of inodes to
982 * compensate, so that the "inodes in use" number is correct.
984 if (obd_osfs.os_ffree < osfs->os_ffree) {
985 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
987 osfs->os_ffree = obd_osfs.os_ffree;
993 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
995 struct obd_statfs osfs;
998 CDEBUG(D_VFSTRACE, "VFS Op:\n");
999 lprocfs_counter_incr(ll_s2sbi(sb)->ll_stats, LPROC_LL_STAFS);
1001 /* For now we will always get up-to-date statfs values, but in the
1002 * future we may allow some amount of caching on the client (e.g.
1003 * from QOS or lprocfs updates). */
1004 rc = ll_statfs_internal(sb, &osfs, jiffies - 1);
1008 statfs_unpack(sfs, &osfs);
1010 if (sizeof(sfs->f_blocks) == 4) {
1011 while (osfs.os_blocks > ~0UL) {
1014 osfs.os_blocks >>= 1;
1015 osfs.os_bfree >>= 1;
1016 osfs.os_bavail >>= 1;
1020 sfs->f_blocks = osfs.os_blocks;
1021 sfs->f_bfree = osfs.os_bfree;
1022 sfs->f_bavail = osfs.os_bavail;
1027 void dump_lsm(int level, struct lov_stripe_md *lsm)
1029 CDEBUG(level, "objid "LPX64", maxbytes "LPX64", magic 0x%08X, "
1030 "stripe_size %u, stripe_count %u\n",
1031 lsm->lsm_object_id, lsm->lsm_maxbytes, lsm->lsm_magic,
1032 lsm->lsm_stripe_size, lsm->lsm_stripe_count);
1035 void ll_update_inode(struct inode *inode, struct mds_body *body,
1036 struct lov_stripe_md *lsm)
1038 struct ll_inode_info *lli = ll_i2info(inode);
1040 LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1042 if (lli->lli_smd == NULL) {
1044 lli->lli_maxbytes = lsm->lsm_maxbytes;
1045 if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1046 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1048 if (memcmp(lli->lli_smd, lsm, sizeof(*lsm))) {
1049 CERROR("lsm mismatch for inode %ld\n",
1051 CERROR("lli_smd:\n");
1052 dump_lsm(D_ERROR, lli->lli_smd);
1054 dump_lsm(D_ERROR, lsm);
1058 /* bug 2844 - limit i_blksize for broken user-space apps */
1059 LASSERTF(lsm->lsm_xfersize != 0, "%lu\n", lsm->lsm_xfersize);
1060 inode->i_blksize = min(lsm->lsm_xfersize, LL_MAX_BLKSIZE);
1061 if (lli->lli_smd != lsm)
1062 obd_free_memmd(ll_i2obdexp(inode), &lsm);
1065 if (body->valid & OBD_MD_FLID)
1066 inode->i_ino = body->ino;
1067 if (body->valid & OBD_MD_FLATIME)
1068 LTIME_S(inode->i_atime) = body->atime;
1069 if (body->valid & OBD_MD_FLMTIME &&
1070 body->mtime > LTIME_S(inode->i_mtime)) {
1071 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to %u\n",
1072 inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
1073 LTIME_S(inode->i_mtime) = body->mtime;
1075 if (body->valid & OBD_MD_FLCTIME &&
1076 body->ctime > LTIME_S(inode->i_ctime))
1077 LTIME_S(inode->i_ctime) = body->ctime;
1078 if (body->valid & OBD_MD_FLMODE)
1079 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1080 if (body->valid & OBD_MD_FLTYPE)
1081 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1082 if (body->valid & OBD_MD_FLUID)
1083 inode->i_uid = body->uid;
1084 if (body->valid & OBD_MD_FLGID)
1085 inode->i_gid = body->gid;
1086 if (body->valid & OBD_MD_FLFLAGS)
1087 inode->i_flags = body->flags;
1088 if (body->valid & OBD_MD_FLNLINK)
1089 inode->i_nlink = body->nlink;
1090 if (body->valid & OBD_MD_FLGENER)
1091 inode->i_generation = body->generation;
1092 if (body->valid & OBD_MD_FLRDEV)
1093 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1094 inode->i_rdev = body->rdev;
1096 inode->i_rdev = old_decode_dev(body->rdev);
1098 if (body->valid & OBD_MD_FLSIZE)
1099 inode->i_size = body->size;
1100 if (body->valid & OBD_MD_FLBLOCKS)
1101 inode->i_blocks = body->blocks;
1103 if (body->valid & OBD_MD_FLSIZE)
1104 set_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &lli->lli_flags);
1107 void ll_read_inode2(struct inode *inode, void *opaque)
1109 struct lustre_md *md = opaque;
1110 struct ll_inode_info *lli = ll_i2info(inode);
1113 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1114 inode->i_generation, inode);
1118 LASSERT(!lli->lli_smd);
1120 /* Core attributes from the MDS first. This is a new inode, and
1121 * the VFS doesn't zero times in the core inode so we have to do
1122 * it ourselves. They will be overwritten by either MDS or OST
1123 * attributes - we just need to make sure they aren't newer. */
1124 LTIME_S(inode->i_mtime) = 0;
1125 LTIME_S(inode->i_atime) = 0;
1126 LTIME_S(inode->i_ctime) = 0;
1127 ll_update_inode(inode, md->body, md->lsm);
1129 /* OIDEBUG(inode); */
1131 if (S_ISREG(inode->i_mode)) {
1132 inode->i_op = &ll_file_inode_operations;
1133 inode->i_fop = &ll_file_operations;
1134 inode->i_mapping->a_ops = &ll_aops;
1136 } else if (S_ISDIR(inode->i_mode)) {
1137 inode->i_op = &ll_dir_inode_operations;
1138 inode->i_fop = &ll_dir_operations;
1139 inode->i_mapping->a_ops = &ll_dir_aops;
1141 } else if (S_ISLNK(inode->i_mode)) {
1142 inode->i_op = &ll_fast_symlink_inode_operations;
1145 inode->i_op = &ll_special_inode_operations;
1146 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
1147 init_special_inode(inode, inode->i_mode,
1148 kdev_t_to_nr(inode->i_rdev));
1150 init_special_inode(inode, inode->i_mode, inode->i_rdev);
1156 int ll_iocontrol(struct inode *inode, struct file *file,
1157 unsigned int cmd, unsigned long arg)
1159 struct ll_sb_info *sbi = ll_i2sbi(inode);
1160 struct ptlrpc_request *req = NULL;
1165 case EXT3_IOC_GETFLAGS: {
1167 unsigned long valid = OBD_MD_FLFLAGS;
1168 struct mds_body *body;
1170 ll_inode2fid(&fid, inode);
1171 rc = mdc_getattr(sbi->ll_mdc_exp, &fid, valid, 0, &req);
1173 CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1177 body = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*body));
1179 if (body->flags & S_APPEND)
1180 flags |= EXT3_APPEND_FL;
1181 if (body->flags & S_IMMUTABLE)
1182 flags |= EXT3_IMMUTABLE_FL;
1183 if (body->flags & S_NOATIME)
1184 flags |= EXT3_NOATIME_FL;
1186 ptlrpc_req_finished (req);
1188 RETURN(put_user(flags, (int *)arg));
1190 case EXT3_IOC_SETFLAGS: {
1191 struct mdc_op_data op_data;
1194 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1196 if (get_user(flags, (int *)arg))
1203 ll_prepare_mdc_op_data(&op_data, inode, NULL, NULL, 0, 0);
1205 memset(&attr, 0x0, sizeof(attr));
1206 attr.ia_attr_flags = flags;
1207 attr.ia_valid |= ATTR_ATTR_FLAG;
1209 rc = mdc_setattr(sbi->ll_mdc_exp, &op_data,
1210 &attr, NULL, 0, NULL, 0, &req);
1212 ptlrpc_req_finished(req);
1213 if (rc != -EPERM && rc != -EACCES)
1214 CERROR("mdc_setattr fails: rc = %d\n", rc);
1218 ptlrpc_req_finished(req);
1220 oa->o_id = lsm->lsm_object_id;
1221 oa->o_flags = flags;
1222 oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS;
1224 rc = obd_setattr(sbi->ll_osc_exp, oa, lsm, NULL);
1227 if (rc != -EPERM && rc != -EACCES)
1228 CERROR("mdc_setattr fails: rc = %d\n", rc);
1232 if (flags & EXT3_APPEND_FL)
1233 inode->i_flags |= S_APPEND;
1235 inode->i_flags &= ~S_APPEND;
1236 if (flags & EXT3_IMMUTABLE_FL)
1237 inode->i_flags |= S_IMMUTABLE;
1239 inode->i_flags &= ~S_IMMUTABLE;
1240 if (flags & EXT3_NOATIME_FL)
1241 inode->i_flags |= S_NOATIME;
1243 inode->i_flags &= ~S_NOATIME;
1254 void ll_umount_begin(struct super_block *sb)
1256 struct ll_sb_info *sbi = ll_s2sbi(sb);
1257 struct obd_device *obd;
1258 struct obd_ioctl_data ioc_data = { 0 };
1260 CDEBUG(D_VFSTRACE, "VFS Op:\n");
1262 obd = class_exp2obd(sbi->ll_mdc_exp);
1264 CERROR("Invalid MDC connection handle "LPX64"\n",
1265 sbi->ll_mdc_exp->exp_handle.h_cookie);
1269 obd->obd_no_recov = 1;
1270 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_mdc_exp, sizeof ioc_data,
1273 obd = class_exp2obd(sbi->ll_osc_exp);
1275 CERROR("Invalid LOV connection handle "LPX64"\n",
1276 sbi->ll_osc_exp->exp_handle.h_cookie);
1281 obd->obd_no_recov = 1;
1282 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_osc_exp, sizeof ioc_data,
1285 /* Really, we'd like to wait until there are no requests outstanding,
1286 * and then continue. For now, we just invalidate the requests,
1287 * schedule, and hope.
1294 int ll_prep_inode(struct obd_export *exp, struct inode **inode,
1295 struct ptlrpc_request *req, int offset,struct super_block *sb)
1297 struct lustre_md md;
1300 rc = mdc_req2lustre_md(req, offset, exp, &md);
1305 ll_update_inode(*inode, md.body, md.lsm);
1308 *inode = ll_iget(sb, md.body->ino, &md);
1309 if (*inode == NULL || is_bad_inode(*inode)) {
1310 /* free the lsm if we allocated one above */
1312 obd_free_memmd(exp, &md.lsm);
1314 CERROR("new_inode -fatal: rc %d\n", rc);