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;
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_lock);
57 INIT_LIST_HEAD(&sbi->ll_pglist);
58 sbi->ll_pglist_gen = 0;
59 sbi->ll_max_read_ahead_pages = SBI_DEFAULT_RA_MAX;
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 /* making vm readahead 0 for 2.4.x. In the case of 2.6.x,
197 backing dev info assigned to inode mapping is used for
198 determining maximal readahead. */
199 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0))
200 /* bug 2805 - set VM readahead to zero */
201 vm_max_readahead = vm_min_readahead = 0;
204 sb->s_root = d_alloc_root(root);
211 obd_disconnect(sbi->ll_osc_exp, 0);
213 obd_disconnect(sbi->ll_mdc_exp, 0);
215 lprocfs_unregister_mountpoint(sbi);
219 void lustre_common_put_super(struct super_block *sb)
221 struct ll_sb_info *sbi = ll_s2sbi(sb);
222 struct hlist_node *tmp, *next;
225 ll_close_thread_shutdown(sbi->ll_lcq);
227 list_del(&sbi->ll_conn_chain);
228 obd_disconnect(sbi->ll_osc_exp, 0);
230 lprocfs_unregister_mountpoint(sbi);
231 if (sbi->ll_proc_root) {
232 lprocfs_remove(sbi->ll_proc_root);
233 sbi->ll_proc_root = NULL;
236 obd_disconnect(sbi->ll_mdc_exp, 0);
238 // We do this to get rid of orphaned dentries. That is not really trw.
239 spin_lock(&dcache_lock);
240 hlist_for_each_safe(tmp, next, &sbi->ll_orphan_dentry_list) {
241 struct dentry *dentry = hlist_entry(tmp, struct dentry, d_hash);
242 CWARN("orphan dentry %*s (%p) at unmount\n",
243 dentry->d_name.len, dentry->d_name.name, dentry);
244 shrink_dcache_parent(dentry);
246 spin_unlock(&dcache_lock);
251 char *ll_read_opt(const char *opt, char *data)
257 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
258 if (strncmp(opt, data, strlen(opt)))
260 if ((value = strchr(data, '=')) == NULL)
264 OBD_ALLOC(retval, strlen(value) + 1);
266 CERROR("out of memory!\n");
270 memcpy(retval, value, strlen(value)+1);
271 CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
275 int ll_set_opt(const char *opt, char *data, int fl)
279 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
280 if (strncmp(opt, data, strlen(opt)))
286 void ll_options(char *options, char **ost, char **mdc, int *flags)
289 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
290 char *opt_ptr = options;
299 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
300 for (this_char = strtok (options, ",");
302 this_char = strtok (NULL, ",")) {
304 while ((this_char = strsep (&opt_ptr, ",")) != NULL) {
306 CDEBUG(D_SUPER, "this_char %s\n", this_char);
307 if (!*ost && (*ost = ll_read_opt("osc", this_char)))
309 if (!*mdc && (*mdc = ll_read_opt("mdc", this_char)))
311 if (!(*flags & LL_SBI_NOLCK) &&
312 ((*flags) = (*flags) |
313 ll_set_opt("nolock", this_char,
320 void ll_lli_init(struct ll_inode_info *lli)
322 sema_init(&lli->lli_open_sem, 1);
324 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
325 spin_lock_init(&lli->lli_lock);
326 INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
329 int ll_fill_super(struct super_block *sb, void *data, int silent)
331 struct ll_sb_info *sbi;
337 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
339 sbi = lustre_init_sbi(sb);
343 sbi->ll_flags |= LL_SBI_READAHEAD;
344 ll_options(data, &osc, &mdc, &sbi->ll_flags);
348 GOTO(out, err = -EINVAL);
353 GOTO(out, err = -EINVAL);
356 err = lustre_common_fill_super(sb, mdc, osc);
362 OBD_FREE(mdc, strlen(mdc) + 1);
364 OBD_FREE(osc, strlen(osc) + 1);
367 } /* ll_read_super */
369 int lustre_process_log(struct lustre_mount_data *lmd, char * profile,
370 struct config_llog_instance *cfg, int allow_recov)
372 struct lustre_cfg lcfg;
373 struct portals_cfg pcfg;
374 char * peer = "MDS_PEER_UUID";
375 struct obd_device *obd;
376 struct lustre_handle mdc_conn = {0, };
377 struct obd_export *exp;
378 char * name = "mdc_dev";
380 struct obd_uuid mdc_uuid;
381 struct llog_ctxt *ctxt;
386 if (lmd_bad_magic(lmd))
389 generate_random_uuid(uuid);
390 class_uuid_unparse(uuid, &mdc_uuid);
392 if (lmd->lmd_local_nid) {
393 PCFG_INIT(pcfg, NAL_CMD_REGISTER_MYNID);
394 pcfg.pcfg_nal = lmd->lmd_nal;
395 pcfg.pcfg_nid = lmd->lmd_local_nid;
396 err = libcfs_nal_cmd(&pcfg);
401 if (lmd->lmd_nal == SOCKNAL) {
402 PCFG_INIT(pcfg, NAL_CMD_ADD_AUTOCONN);
403 pcfg.pcfg_nal = lmd->lmd_nal;
404 pcfg.pcfg_nid = lmd->lmd_server_nid;
405 pcfg.pcfg_id = lmd->lmd_server_ipaddr;
406 pcfg.pcfg_misc = lmd->lmd_port;
407 pcfg.pcfg_size = 8388608;
408 pcfg.pcfg_flags = 0x4; /*share*/
409 err = libcfs_nal_cmd(&pcfg);
414 LCFG_INIT(lcfg, LCFG_ADD_UUID, name);
415 lcfg.lcfg_nid = lmd->lmd_server_nid;
416 lcfg.lcfg_inllen1 = strlen(peer) + 1;
417 lcfg.lcfg_inlbuf1 = peer;
418 lcfg.lcfg_nal = lmd->lmd_nal;
419 err = class_process_config(&lcfg);
421 GOTO(out_del_conn, err);
423 LCFG_INIT(lcfg, LCFG_ATTACH, name);
424 lcfg.lcfg_inlbuf1 = "mdc";
425 lcfg.lcfg_inllen1 = strlen(lcfg.lcfg_inlbuf1) + 1;
426 lcfg.lcfg_inlbuf2 = mdc_uuid.uuid;
427 lcfg.lcfg_inllen2 = strlen(lcfg.lcfg_inlbuf2) + 1;
428 err = class_process_config(&lcfg);
430 GOTO(out_del_uuid, err);
432 LCFG_INIT(lcfg, LCFG_SETUP, name);
433 lcfg.lcfg_inlbuf1 = lmd->lmd_mds;
434 lcfg.lcfg_inllen1 = strlen(lcfg.lcfg_inlbuf1) + 1;
435 lcfg.lcfg_inlbuf2 = peer;
436 lcfg.lcfg_inllen2 = strlen(lcfg.lcfg_inlbuf2) + 1;
437 err = class_process_config(&lcfg);
439 GOTO(out_detach, err);
441 obd = class_name2obd(name);
443 GOTO(out_cleanup, err = -EINVAL);
445 /* Disable initial recovery on this import */
446 err = obd_set_info(obd->obd_self_export,
447 strlen("initial_recov"), "initial_recov",
448 sizeof(allow_recov), &allow_recov);
450 GOTO(out_cleanup, err);
452 err = obd_connect(&mdc_conn, obd, &mdc_uuid);
454 CERROR("cannot connect to %s: rc = %d\n", lmd->lmd_mds, err);
455 GOTO(out_cleanup, err);
458 exp = class_conn2export(&mdc_conn);
460 ctxt = llog_get_context(exp->exp_obd, LLOG_CONFIG_REPL_CTXT);
462 rc = class_config_parse_llog(ctxt, profile, cfg);
465 * For debugging, it's useful to just dump the log
467 rc = class_config_dump_llog(ctxt, profile, cfg);
470 CERROR("class_config_parse_llog failed: rc = %d\n", rc);
473 err = obd_disconnect(exp, 0);
476 LCFG_INIT(lcfg, LCFG_CLEANUP, name);
477 err = class_process_config(&lcfg);
482 LCFG_INIT(lcfg, LCFG_DETACH, name);
483 err = class_process_config(&lcfg);
488 LCFG_INIT(lcfg, LCFG_DEL_UUID, name);
489 lcfg.lcfg_inllen1 = strlen(peer) + 1;
490 lcfg.lcfg_inlbuf1 = peer;
491 err = class_process_config(&lcfg);
494 if (lmd->lmd_nal == SOCKNAL) {
495 PCFG_INIT(pcfg, NAL_CMD_DEL_AUTOCONN);
496 pcfg.pcfg_nal = lmd->lmd_nal;
497 pcfg.pcfg_nid = lmd->lmd_server_nid;
498 pcfg.pcfg_id = lmd->lmd_server_ipaddr;
499 pcfg.pcfg_flags = 1; /*share*/
500 err = libcfs_nal_cmd(&pcfg);
511 int lustre_fill_super(struct super_block *sb, void *data, int silent)
513 struct lustre_mount_data * lmd = data;
514 struct ll_sb_info *sbi;
520 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
521 if (lmd_bad_magic(lmd))
524 sbi = lustre_init_sbi(sb);
528 sbi->ll_flags |= LL_SBI_READAHEAD;
530 if (lmd->lmd_profile) {
531 struct lustre_profile *lprof;
532 struct config_llog_instance cfg;
535 if (lmd->lmd_mds[0] == '\0') {
536 CERROR("no mds name\n");
537 GOTO(out_free, err = -EINVAL);
540 OBD_ALLOC(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
541 if (sbi->ll_lmd == NULL)
542 GOTO(out_free, err = -ENOMEM);
543 memcpy(sbi->ll_lmd, lmd, sizeof(*lmd));
545 /* generate a string unique to this super, let's try
546 the address of the super itself.*/
547 len = (sizeof(sb) * 2) + 1;
548 OBD_ALLOC(sbi->ll_instance, len);
549 if (sbi->ll_instance == NULL)
550 GOTO(out_free, err = -ENOMEM);
551 sprintf(sbi->ll_instance, "%p", sb);
553 cfg.cfg_instance = sbi->ll_instance;
554 cfg.cfg_uuid = sbi->ll_sb_uuid;
555 cfg.cfg_local_nid = lmd->lmd_local_nid;
556 err = lustre_process_log(lmd, lmd->lmd_profile, &cfg, 0);
558 CERROR("Unable to process log: %s\n", lmd->lmd_profile);
563 lprof = class_get_profile(lmd->lmd_profile);
565 CERROR("No profile found: %s\n", lmd->lmd_profile);
566 GOTO(out_free, err = -EINVAL);
569 OBD_FREE(osc, strlen(osc) + 1);
570 OBD_ALLOC(osc, strlen(lprof->lp_osc) +
571 strlen(sbi->ll_instance) + 2);
572 sprintf(osc, "%s-%s", lprof->lp_osc, sbi->ll_instance);
575 OBD_FREE(mdc, strlen(mdc) + 1);
576 OBD_ALLOC(mdc, strlen(lprof->lp_mdc) +
577 strlen(sbi->ll_instance) + 2);
578 sprintf(mdc, "%s-%s", lprof->lp_mdc, sbi->ll_instance);
583 GOTO(out_free, err = -EINVAL);
588 GOTO(out_free, err = -EINVAL);
591 err = lustre_common_fill_super(sb, mdc, osc);
598 OBD_FREE(mdc, strlen(mdc) + 1);
600 OBD_FREE(osc, strlen(osc) + 1);
606 int len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean")+1;
609 if (sbi->ll_instance != NULL) {
611 struct config_llog_instance cfg;
613 cfg.cfg_instance = sbi->ll_instance;
614 cfg.cfg_uuid = sbi->ll_sb_uuid;
616 OBD_ALLOC(cln_prof, len);
617 sprintf(cln_prof, "%s-clean", sbi->ll_lmd->lmd_profile);
619 err = lustre_process_log(sbi->ll_lmd, cln_prof, &cfg,
622 CERROR("Unable to process log: %s\n", cln_prof);
623 OBD_FREE(cln_prof, len);
624 OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance)+ 1);
626 OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
631 } /* lustre_fill_super */
633 static void lustre_manual_cleanup(struct ll_sb_info *sbi)
635 struct lustre_cfg lcfg;
636 struct obd_device *obd;
639 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) != NULL)
643 LCFG_INIT(lcfg, LCFG_CLEANUP, obd->obd_name);
644 err = class_process_config(&lcfg);
646 CERROR("cleanup failed: %s\n", obd->obd_name);
650 LCFG_INIT(lcfg, LCFG_DETACH, obd->obd_name);
651 err = class_process_config(&lcfg);
653 CERROR("detach failed: %s\n", obd->obd_name);
658 if (sbi->ll_lmd != NULL)
659 class_del_profile(sbi->ll_lmd->lmd_profile);
662 void lustre_put_super(struct super_block *sb)
664 struct obd_device *obd;
665 struct ll_sb_info *sbi = ll_s2sbi(sb);
666 int force_umount = 0;
669 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
670 obd = class_exp2obd(sbi->ll_mdc_exp);
672 force_umount = obd->obd_no_recov;
675 lustre_common_put_super(sb);
677 if (sbi->ll_lmd != NULL) {
679 int len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean")+1;
681 struct config_llog_instance cfg;
684 CERROR("force umount, doing manual cleanup\n");
685 lustre_manual_cleanup(sbi);
689 cfg.cfg_instance = sbi->ll_instance;
690 cfg.cfg_uuid = sbi->ll_sb_uuid;
692 OBD_ALLOC(cln_prof, len);
693 sprintf(cln_prof, "%s-clean", sbi->ll_lmd->lmd_profile);
695 err = lustre_process_log(sbi->ll_lmd, cln_prof, &cfg, 0);
697 CERROR("Unable to process log: %s, doing manual cleanup"
699 lustre_manual_cleanup(sbi);
702 OBD_FREE(cln_prof, len);
704 OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
705 OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance) + 1);
711 } /* lustre_put_super */
714 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
717 l_lock(&lock->l_resource->lr_namespace->ns_lock);
718 if (lock->l_ast_data)
719 inode = igrab(lock->l_ast_data);
722 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
726 static int null_if_equal(struct ldlm_lock *lock, void *data)
728 if (data == lock->l_ast_data)
729 lock->l_ast_data = NULL;
731 if (lock->l_req_mode != lock->l_granted_mode)
732 return LDLM_ITER_STOP;
734 return LDLM_ITER_CONTINUE;
737 void ll_clear_inode(struct inode *inode)
740 struct ll_inode_info *lli = ll_i2info(inode);
741 struct ll_sb_info *sbi = ll_i2sbi(inode);
744 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
745 inode->i_generation, inode);
747 ll_inode2fid(&fid, inode);
748 clear_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &(ll_i2info(inode)->lli_flags));
749 mdc_change_cbdata(sbi->ll_mdc_exp, &fid, null_if_equal, inode);
752 obd_change_cbdata(sbi->ll_osc_exp, lli->lli_smd,
753 null_if_equal, inode);
756 obd_free_memmd(sbi->ll_osc_exp, &lli->lli_smd);
760 if (lli->lli_symlink_name) {
761 OBD_FREE(lli->lli_symlink_name,
762 strlen(lli->lli_symlink_name) + 1);
763 lli->lli_symlink_name = NULL;
769 /* If this inode has objects allocated to it (lsm != NULL), then the OST
770 * object(s) determine the file size and mtime. Otherwise, the MDS will
771 * keep these values until such a time that objects are allocated for it.
772 * We do the MDS operations first, as it is checking permissions for us.
773 * We don't to the MDS RPC if there is nothing that we want to store there,
774 * otherwise there is no harm in updating mtime/atime on the MDS if we are
775 * going to do an RPC anyways.
777 * If we are doing a truncate, we will send the mtime and ctime updates
778 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
779 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
782 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
784 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
785 struct ll_sb_info *sbi = ll_i2sbi(inode);
786 struct ptlrpc_request *request = NULL;
787 struct mdc_op_data op_data;
788 int ia_valid = attr->ia_valid;
792 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu\n", inode->i_ino);
793 lprocfs_counter_incr(ll_i2sbi(inode)->ll_stats, LPROC_LL_SETATTR);
795 if (ia_valid & ATTR_SIZE) {
796 if (attr->ia_size > ll_file_maxbytes(inode)) {
797 CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
798 attr->ia_size, ll_file_maxbytes(inode));
802 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
805 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
806 if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
807 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
811 /* We mark all of the fields "set" so MDS/OST does not re-set them */
812 if (attr->ia_valid & ATTR_CTIME) {
813 attr->ia_ctime = CURRENT_TIME;
814 attr->ia_valid |= ATTR_CTIME_SET;
816 if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
817 attr->ia_atime = CURRENT_TIME;
818 attr->ia_valid |= ATTR_ATIME_SET;
820 if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
821 attr->ia_mtime = CURRENT_TIME;
822 attr->ia_valid |= ATTR_MTIME_SET;
825 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
826 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
827 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
828 LTIME_S(CURRENT_TIME));
830 attr->ia_valid &= ~ATTR_SIZE;
832 /* If only OST attributes being set on objects, don't do MDS RPC.
833 * In that case, we need to check permissions and update the local
834 * inode ourselves so we can call obdo_from_inode() always. */
835 if (ia_valid & (lsm ? ~(ATTR_SIZE | ATTR_FROM_OPEN | ATTR_RAW) : ~0)) {
837 ll_prepare_mdc_op_data(&op_data, inode, NULL, NULL, 0, 0);
839 rc = mdc_setattr(sbi->ll_mdc_exp, &op_data,
840 attr, NULL, 0, NULL, 0, &request);
843 ptlrpc_req_finished(request);
844 if (rc != -EPERM && rc != -EACCES)
845 CERROR("mdc_setattr fails: rc = %d\n", rc);
849 rc = mdc_req2lustre_md(request, 0, sbi->ll_osc_exp, &md);
851 ptlrpc_req_finished(request);
855 /* Won't invoke vmtruncate as we already cleared ATTR_SIZE,
856 * but needed to set timestamps backwards on utime. */
857 inode_setattr(inode, attr);
858 ll_update_inode(inode, md.body, md.lsm);
859 ptlrpc_req_finished(request);
861 if (!lsm || !S_ISREG(inode->i_mode)) {
862 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
866 /* The OST doesn't check permissions, but the alternative is
867 * a gratuitous RPC to the MDS. We already rely on the client
868 * to do read/write/truncate permission checks, so is mtime OK?
870 if (ia_valid & (ATTR_MTIME | ATTR_ATIME)) {
871 /* from sys_utime() */
872 if (!(ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET))) {
873 if (current->fsuid != inode->i_uid &&
874 (rc=ll_permission(inode,MAY_WRITE,NULL))!=0)
877 /* from inode_change_ok() */
878 if (current->fsuid != inode->i_uid &&
879 !capable(CAP_FOWNER))
884 /* Won't invoke vmtruncate, as we already cleared ATTR_SIZE */
885 inode_setattr(inode, attr);
888 /* We really need to get our PW lock before we change inode->i_size.
889 * If we don't we can race with other i_size updaters on our node, like
890 * ll_file_read. We can also race with i_size propogation to other
891 * nodes through dirtying and writeback of final cached pages. This
892 * last one is especially bad for racing o_append users on other
894 if (ia_valid & ATTR_SIZE) {
895 ldlm_policy_data_t policy = { .l_extent = {attr->ia_size,
897 struct lustre_handle lockh = { 0 };
898 int err, ast_flags = 0;
899 /* XXX when we fix the AST intents to pass the discard-range
900 * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
902 if (attr->ia_size == 0)
903 ast_flags = LDLM_AST_DISCARD_DATA;
905 /* bug 1639: avoid write/truncate i_sem/DLM deadlock */
906 LASSERT(atomic_read(&inode->i_sem.count) <= 0);
908 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy, &lockh,
914 rc = vmtruncate(inode, attr->ia_size);
916 /* We need to drop the semaphore here, because this unlock may
917 * result in a cancellation, which will need the i_sem */
919 /* unlock now as we don't mind others file lockers racing with
920 * the mds updates below? */
921 err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
924 CERROR("ll_extent_unlock failed: %d\n", err);
928 } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
931 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
932 inode->i_ino, LTIME_S(attr->ia_mtime));
933 oa.o_id = lsm->lsm_object_id;
934 oa.o_valid = OBD_MD_FLID;
935 obdo_from_inode(&oa, inode, OBD_MD_FLTYPE | OBD_MD_FLATIME |
936 OBD_MD_FLMTIME | OBD_MD_FLCTIME);
937 rc = obd_setattr(sbi->ll_osc_exp, &oa, lsm, NULL);
939 CERROR("obd_setattr fails: rc=%d\n", rc);
944 int ll_setattr(struct dentry *de, struct iattr *attr)
946 LBUG(); /* code is unused, but leave this in case of VFS changes */
950 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
951 unsigned long max_age)
953 struct ll_sb_info *sbi = ll_s2sbi(sb);
954 struct obd_statfs obd_osfs;
958 rc = obd_statfs(class_exp2obd(sbi->ll_mdc_exp), osfs, max_age);
960 CERROR("mdc_statfs fails: rc = %d\n", rc);
964 osfs->os_type = sb->s_magic;
966 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
967 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
969 rc = obd_statfs(class_exp2obd(sbi->ll_osc_exp), &obd_osfs, max_age);
971 CERROR("obd_statfs fails: rc = %d\n", rc);
975 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
976 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
979 osfs->os_blocks = obd_osfs.os_blocks;
980 osfs->os_bfree = obd_osfs.os_bfree;
981 osfs->os_bavail = obd_osfs.os_bavail;
983 /* If we don't have as many objects free on the OST as inodes
984 * on the MDS, we reduce the total number of inodes to
985 * compensate, so that the "inodes in use" number is correct.
987 if (obd_osfs.os_ffree < osfs->os_ffree) {
988 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
990 osfs->os_ffree = obd_osfs.os_ffree;
996 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
998 struct obd_statfs osfs;
1001 CDEBUG(D_VFSTRACE, "VFS Op:\n");
1002 lprocfs_counter_incr(ll_s2sbi(sb)->ll_stats, LPROC_LL_STAFS);
1004 /* For now we will always get up-to-date statfs values, but in the
1005 * future we may allow some amount of caching on the client (e.g.
1006 * from QOS or lprocfs updates). */
1007 rc = ll_statfs_internal(sb, &osfs, jiffies - 1);
1011 statfs_unpack(sfs, &osfs);
1013 if (sizeof(sfs->f_blocks) == 4) {
1014 while (osfs.os_blocks > ~0UL) {
1017 osfs.os_blocks >>= 1;
1018 osfs.os_bfree >>= 1;
1019 osfs.os_bavail >>= 1;
1023 sfs->f_blocks = osfs.os_blocks;
1024 sfs->f_bfree = osfs.os_bfree;
1025 sfs->f_bavail = osfs.os_bavail;
1030 void dump_lsm(int level, struct lov_stripe_md *lsm)
1032 CDEBUG(level, "objid "LPX64", maxbytes "LPX64", magic 0x%08X, "
1033 "stripe_size %u, stripe_count %u\n",
1034 lsm->lsm_object_id, lsm->lsm_maxbytes, lsm->lsm_magic,
1035 lsm->lsm_stripe_size, lsm->lsm_stripe_count);
1038 void ll_update_inode(struct inode *inode, struct mds_body *body,
1039 struct lov_stripe_md *lsm)
1041 struct ll_inode_info *lli = ll_i2info(inode);
1043 LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1045 if (lli->lli_smd == NULL) {
1047 lli->lli_maxbytes = lsm->lsm_maxbytes;
1048 if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1049 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1051 if (memcmp(lli->lli_smd, lsm, sizeof(*lsm))) {
1052 CERROR("lsm mismatch for inode %ld\n",
1054 CERROR("lli_smd:\n");
1055 dump_lsm(D_ERROR, lli->lli_smd);
1057 dump_lsm(D_ERROR, lsm);
1061 /* bug 2844 - limit i_blksize for broken user-space apps */
1062 LASSERTF(lsm->lsm_xfersize != 0, "%lu\n", lsm->lsm_xfersize);
1063 inode->i_blksize = min(lsm->lsm_xfersize, LL_MAX_BLKSIZE);
1064 if (lli->lli_smd != lsm)
1065 obd_free_memmd(ll_i2obdexp(inode), &lsm);
1068 if (body->valid & OBD_MD_FLID)
1069 inode->i_ino = body->ino;
1070 if (body->valid & OBD_MD_FLATIME)
1071 LTIME_S(inode->i_atime) = body->atime;
1072 if (body->valid & OBD_MD_FLMTIME &&
1073 body->mtime > LTIME_S(inode->i_mtime)) {
1074 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to %u\n",
1075 inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
1076 LTIME_S(inode->i_mtime) = body->mtime;
1078 if (body->valid & OBD_MD_FLCTIME &&
1079 body->ctime > LTIME_S(inode->i_ctime))
1080 LTIME_S(inode->i_ctime) = body->ctime;
1081 if (body->valid & OBD_MD_FLMODE)
1082 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1083 if (body->valid & OBD_MD_FLTYPE)
1084 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1085 if (body->valid & OBD_MD_FLUID)
1086 inode->i_uid = body->uid;
1087 if (body->valid & OBD_MD_FLGID)
1088 inode->i_gid = body->gid;
1089 if (body->valid & OBD_MD_FLFLAGS)
1090 inode->i_flags = body->flags;
1091 if (body->valid & OBD_MD_FLNLINK)
1092 inode->i_nlink = body->nlink;
1093 if (body->valid & OBD_MD_FLGENER)
1094 inode->i_generation = body->generation;
1095 if (body->valid & OBD_MD_FLRDEV)
1096 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1097 inode->i_rdev = body->rdev;
1099 inode->i_rdev = old_decode_dev(body->rdev);
1101 if (body->valid & OBD_MD_FLSIZE)
1102 inode->i_size = body->size;
1103 if (body->valid & OBD_MD_FLBLOCKS)
1104 inode->i_blocks = body->blocks;
1106 if (body->valid & OBD_MD_FLSIZE)
1107 set_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &lli->lli_flags);
1110 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
1111 static struct backing_dev_info ll_backing_dev_info = {
1112 .ra_pages = 0, /* No readahead */
1113 .memory_backed = 0, /* Does contribute to dirty memory */
1117 void ll_read_inode2(struct inode *inode, void *opaque)
1119 struct lustre_md *md = opaque;
1120 struct ll_inode_info *lli = ll_i2info(inode);
1123 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1124 inode->i_generation, inode);
1128 LASSERT(!lli->lli_smd);
1130 /* Core attributes from the MDS first. This is a new inode, and
1131 * the VFS doesn't zero times in the core inode so we have to do
1132 * it ourselves. They will be overwritten by either MDS or OST
1133 * attributes - we just need to make sure they aren't newer. */
1134 LTIME_S(inode->i_mtime) = 0;
1135 LTIME_S(inode->i_atime) = 0;
1136 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;
1158 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
1159 init_special_inode(inode, inode->i_mode,
1160 kdev_t_to_nr(inode->i_rdev));
1162 /* initializing backing dev info. */
1163 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
1165 init_special_inode(inode, inode->i_mode, inode->i_rdev);
1167 lli->ll_save_ifop = inode->i_fop;
1169 if (S_ISCHR(inode->i_mode))
1170 inode->i_fop = &ll_special_chr_inode_fops;
1171 else if (S_ISBLK(inode->i_mode))
1172 inode->i_fop = &ll_special_blk_inode_fops;
1173 else if (S_ISFIFO(inode->i_mode))
1174 inode->i_fop = &ll_special_fifo_inode_fops;
1175 else if (S_ISSOCK(inode->i_mode))
1176 inode->i_fop = &ll_special_sock_inode_fops;
1178 CWARN("saved %p, replaced with %p\n", lli->ll_save_ifop,
1181 if (lli->ll_save_ifop->owner) {
1182 CWARN("%p has owner %p\n", lli->ll_save_ifop,
1183 lli->ll_save_ifop->owner);
1189 int ll_iocontrol(struct inode *inode, struct file *file,
1190 unsigned int cmd, unsigned long arg)
1192 struct ll_sb_info *sbi = ll_i2sbi(inode);
1193 struct ptlrpc_request *req = NULL;
1198 case EXT3_IOC_GETFLAGS: {
1200 unsigned long valid = OBD_MD_FLFLAGS;
1201 struct mds_body *body;
1203 ll_inode2fid(&fid, inode);
1204 rc = mdc_getattr(sbi->ll_mdc_exp, &fid, valid, 0, &req);
1206 CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1210 body = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*body));
1212 if (body->flags & S_APPEND)
1213 flags |= EXT3_APPEND_FL;
1214 if (body->flags & S_IMMUTABLE)
1215 flags |= EXT3_IMMUTABLE_FL;
1216 if (body->flags & S_NOATIME)
1217 flags |= EXT3_NOATIME_FL;
1219 ptlrpc_req_finished (req);
1221 RETURN(put_user(flags, (int *)arg));
1223 case EXT3_IOC_SETFLAGS: {
1224 struct mdc_op_data op_data;
1227 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1229 if (get_user(flags, (int *)arg))
1236 ll_prepare_mdc_op_data(&op_data, inode, NULL, NULL, 0, 0);
1238 memset(&attr, 0x0, sizeof(attr));
1239 attr.ia_attr_flags = flags;
1240 attr.ia_valid |= ATTR_ATTR_FLAG;
1242 rc = mdc_setattr(sbi->ll_mdc_exp, &op_data,
1243 &attr, NULL, 0, NULL, 0, &req);
1245 ptlrpc_req_finished(req);
1246 if (rc != -EPERM && rc != -EACCES)
1247 CERROR("mdc_setattr fails: rc = %d\n", rc);
1251 ptlrpc_req_finished(req);
1253 oa->o_id = lsm->lsm_object_id;
1254 oa->o_flags = flags;
1255 oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS;
1257 rc = obd_setattr(sbi->ll_osc_exp, oa, lsm, NULL);
1260 if (rc != -EPERM && rc != -EACCES)
1261 CERROR("mdc_setattr fails: rc = %d\n", rc);
1265 if (flags & EXT3_APPEND_FL)
1266 inode->i_flags |= S_APPEND;
1268 inode->i_flags &= ~S_APPEND;
1269 if (flags & EXT3_IMMUTABLE_FL)
1270 inode->i_flags |= S_IMMUTABLE;
1272 inode->i_flags &= ~S_IMMUTABLE;
1273 if (flags & EXT3_NOATIME_FL)
1274 inode->i_flags |= S_NOATIME;
1276 inode->i_flags &= ~S_NOATIME;
1287 void ll_umount_begin(struct super_block *sb)
1289 struct ll_sb_info *sbi = ll_s2sbi(sb);
1290 struct obd_device *obd;
1291 struct obd_ioctl_data ioc_data = { 0 };
1293 CDEBUG(D_VFSTRACE, "VFS Op:\n");
1295 obd = class_exp2obd(sbi->ll_mdc_exp);
1297 CERROR("Invalid MDC connection handle "LPX64"\n",
1298 sbi->ll_mdc_exp->exp_handle.h_cookie);
1302 obd->obd_no_recov = 1;
1303 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_mdc_exp, sizeof ioc_data,
1306 obd = class_exp2obd(sbi->ll_osc_exp);
1308 CERROR("Invalid LOV connection handle "LPX64"\n",
1309 sbi->ll_osc_exp->exp_handle.h_cookie);
1314 obd->obd_no_recov = 1;
1315 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_osc_exp, sizeof ioc_data,
1318 /* Really, we'd like to wait until there are no requests outstanding,
1319 * and then continue. For now, we just invalidate the requests,
1320 * schedule, and hope.
1327 int ll_prep_inode(struct obd_export *exp, struct inode **inode,
1328 struct ptlrpc_request *req, int offset,struct super_block *sb)
1330 struct lustre_md md;
1333 rc = mdc_req2lustre_md(req, offset, exp, &md);
1338 ll_update_inode(*inode, md.body, md.lsm);
1341 *inode = ll_iget(sb, md.body->ino, &md);
1342 if (*inode == NULL || is_bad_inode(*inode)) {
1343 /* free the lsm if we allocated one above */
1345 obd_free_memmd(exp, &md.lsm);
1347 CERROR("new_inode -fatal: rc %d\n", rc);