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 LIST_HEAD(ll_super_blocks);
40 spinlock_t ll_sb_lock = SPIN_LOCK_UNLOCKED;
42 extern struct address_space_operations ll_aops;
43 extern struct address_space_operations ll_dir_aops;
46 #define log2(n) ffz(~(n))
49 struct ll_sb_info *lustre_init_sbi(struct super_block *sb)
51 struct ll_sb_info *sbi = NULL;
55 OBD_ALLOC(sbi, sizeof(*sbi));
59 spin_lock_init(&sbi->ll_lock);
60 INIT_LIST_HEAD(&sbi->ll_pglist);
61 sbi->ll_pglist_gen = 0;
62 if (num_physpages >> (20 - PAGE_SHIFT) < 512)
63 sbi->ll_async_page_max = num_physpages / 2;
65 sbi->ll_async_page_max = (num_physpages / 4) * 3;
66 sbi->ll_ra_info.ra_max_pages = min(num_physpages / 8,
67 SBI_DEFAULT_READAHEAD_MAX);
69 INIT_LIST_HEAD(&sbi->ll_conn_chain);
70 INIT_HLIST_HEAD(&sbi->ll_orphan_dentry_list);
71 ll_s2sbi_nocast(sb) = sbi;
73 generate_random_uuid(uuid);
74 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
76 spin_lock(&ll_sb_lock);
77 list_add_tail(&sbi->ll_list, &ll_super_blocks);
78 spin_unlock(&ll_sb_lock);
82 void lustre_free_sbi(struct super_block *sb)
84 struct ll_sb_info *sbi = ll_s2sbi(sb);
88 spin_lock(&ll_sb_lock);
89 list_del(&sbi->ll_list);
90 spin_unlock(&ll_sb_lock);
91 OBD_FREE(sbi, sizeof(*sbi));
93 ll_s2sbi_nocast(sb) = NULL;
97 int lustre_common_fill_super(struct super_block *sb, char *mdc, char *osc)
99 struct inode *root = 0;
100 struct ll_sb_info *sbi = ll_s2sbi(sb);
101 struct obd_device *obd;
102 struct ll_fid rootfid;
103 struct obd_statfs osfs;
104 struct ptlrpc_request *request = NULL;
105 struct lustre_handle osc_conn = {0, };
106 struct lustre_handle mdc_conn = {0, };
111 obd = class_name2obd(mdc);
113 CERROR("MDC %s: not setup or attached\n", mdc);
117 if (proc_lustre_fs_root) {
118 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
121 CERROR("could not register mount in /proc/lustre");
124 mdc_init_ea_size(obd, osc);
126 err = obd_connect(&mdc_conn, obd, &sbi->ll_sb_uuid);
128 CERROR("An MDS (mdc %s) is performing recovery, of which this"
129 " client is not a part. Please wait for recovery to "
130 "complete, abort, or time out.\n", mdc);
133 CERROR("cannot connect to %s: rc = %d\n", mdc, err);
136 sbi->ll_mdc_exp = class_conn2export(&mdc_conn);
138 err = obd_statfs(obd, &osfs, jiffies - HZ);
142 LASSERT(osfs.os_bsize);
143 sb->s_blocksize = osfs.os_bsize;
144 sb->s_blocksize_bits = log2(osfs.os_bsize);
145 sb->s_magic = LL_SUPER_MAGIC;
146 sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
147 sbi->ll_namelen = osfs.os_namelen;
149 devno = get_uuid2int(sbi2mdc(sbi)->cl_import->imp_target_uuid.uuid,
150 strlen(sbi2mdc(sbi)->cl_import->imp_target_uuid.uuid));
153 obd = class_name2obd(osc);
155 CERROR("OSC %s: not setup or attached\n", osc);
159 err = obd_connect(&osc_conn, obd, &sbi->ll_sb_uuid);
161 CERROR("An OST (osc %s) is performing recovery, of which this"
162 " client is not a part. Please wait for recovery to "
163 "complete, abort, or time out.\n", osc);
166 CERROR("cannot connect to %s: rc = %d\n", osc, err);
169 sbi->ll_osc_exp = class_conn2export(&osc_conn);
171 if (!ll_async_page_slab) {
172 ll_async_page_slab_size =
173 size_round(sizeof(struct ll_async_page)) +
174 obd_prep_async_page(sbi->ll_osc_exp, NULL, NULL, NULL,
175 0, NULL, NULL, NULL);
176 ll_async_page_slab = kmem_cache_create("ll_async_page",
177 ll_async_page_slab_size,
179 if (!ll_async_page_slab)
180 GOTO(out_osc, -ENOMEM);
183 err = mdc_getstatus(sbi->ll_mdc_exp, &rootfid);
185 CERROR("cannot mds_connect: rc = %d\n", err);
188 CDEBUG(D_SUPER, "rootfid "LPU64"\n", rootfid.id);
189 sbi->ll_rootino = rootfid.id;
191 sb->s_op = &lustre_super_operations;
194 * XXX: move this to after cbd setup? */
195 err = mdc_getattr(sbi->ll_mdc_exp, &rootfid,
196 OBD_MD_FLNOTOBD|OBD_MD_FLBLOCKS, 0, &request);
198 CERROR("mdc_getattr failed for root: rc = %d\n", err);
202 err = mdc_req2lustre_md(request, 0, sbi->ll_osc_exp, &md);
204 CERROR("failed to understand root inode md: rc = %d\n",err);
205 ptlrpc_req_finished (request);
209 LASSERT(sbi->ll_rootino != 0);
210 root = ll_iget(sb, sbi->ll_rootino, &md);
212 ptlrpc_req_finished(request);
214 if (root == NULL || is_bad_inode(root)) {
215 /* XXX might need iput() for bad inode */
216 CERROR("lustre_lite: bad iget4 for root\n");
217 GOTO(out_root, err = -EBADF);
220 err = ll_close_thread_start(&sbi->ll_lcq);
222 CERROR("cannot start close thread: rc %d\n", err);
226 /* making vm readahead 0 for 2.4.x. In the case of 2.6.x,
227 backing dev info assigned to inode mapping is used for
228 determining maximal readahead. */
229 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0)) && \
230 !defined(KERNEL_HAS_AS_MAX_READAHEAD)
231 /* bug 2805 - set VM readahead to zero */
232 vm_max_readahead = vm_min_readahead = 0;
235 sb->s_root = d_alloc_root(root);
242 obd_disconnect(sbi->ll_osc_exp, 0);
244 obd_disconnect(sbi->ll_mdc_exp, 0);
246 lprocfs_unregister_mountpoint(sbi);
250 void lustre_dump_inode(struct inode *inode)
252 struct list_head *tmp;
253 int dentry_count = 0;
254 BDEVNAME_DECLARE_STORAGE(buf);
256 LASSERT(inode != NULL);
258 list_for_each(tmp, &inode->i_dentry)
261 CERROR("inode %p dump: dev=%s:%lu, mode=%o, count=%u, %d dentries\n",
262 inode, ll_bdevname(inode->i_sb, buf), inode->i_ino,
263 inode->i_mode, atomic_read(&inode->i_count), dentry_count);
266 void lustre_dump_dentry(struct dentry *dentry, int recur)
268 struct list_head *tmp;
271 LASSERT(dentry != NULL);
273 list_for_each(tmp, &dentry->d_subdirs)
276 CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
277 " flags=0x%x, vfs_flags=0x%lx, fsdata=%p, %d subdirs\n", dentry,
278 dentry->d_name.len, dentry->d_name.name,
279 dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
280 dentry->d_parent, dentry->d_inode, atomic_read(&dentry->d_count),
281 dentry->d_flags, dentry->d_vfs_flags, dentry->d_fsdata, subdirs);
282 if (dentry->d_inode != NULL)
283 lustre_dump_inode(dentry->d_inode);
288 list_for_each(tmp, &dentry->d_subdirs) {
289 struct dentry *d = list_entry(tmp, struct dentry, d_child);
290 lustre_dump_dentry(d, recur - 1);
294 void lustre_common_put_super(struct super_block *sb)
296 struct ll_sb_info *sbi = ll_s2sbi(sb);
297 struct hlist_node *tmp, *next;
300 ll_close_thread_shutdown(sbi->ll_lcq);
302 list_del(&sbi->ll_conn_chain);
303 obd_disconnect(sbi->ll_osc_exp, 0);
305 lprocfs_unregister_mountpoint(sbi);
306 if (sbi->ll_proc_root) {
307 lprocfs_remove(sbi->ll_proc_root);
308 sbi->ll_proc_root = NULL;
311 obd_disconnect(sbi->ll_mdc_exp, 0);
313 // We do this to get rid of orphaned dentries. That is not really trw.
314 hlist_for_each_safe(tmp, next, &sbi->ll_orphan_dentry_list) {
315 struct dentry *dentry = hlist_entry(tmp, struct dentry, d_hash);
316 CWARN("found orphan dentry %.*s (%p->%p) at unmount, dumping "
317 "before and after shrink_dcache_parent\n",
318 dentry->d_name.len, dentry->d_name.name, dentry, next);
319 lustre_dump_dentry(dentry, 1);
320 shrink_dcache_parent(dentry);
321 lustre_dump_dentry(dentry, 1);
327 char *ll_read_opt(const char *opt, char *data)
333 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
334 if (strncmp(opt, data, strlen(opt)))
336 if ((value = strchr(data, '=')) == NULL)
340 OBD_ALLOC(retval, strlen(value) + 1);
342 CERROR("out of memory!\n");
346 memcpy(retval, value, strlen(value)+1);
347 CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
351 int ll_set_opt(const char *opt, char *data, int fl)
355 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
356 if (strncmp(opt, data, strlen(opt)))
362 void ll_options(char *options, char **ost, char **mdc, int *flags)
365 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
366 char *opt_ptr = options;
375 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
376 for (this_char = strtok (options, ",");
378 this_char = strtok (NULL, ",")) {
380 while ((this_char = strsep (&opt_ptr, ",")) != NULL) {
382 CDEBUG(D_SUPER, "this_char %s\n", this_char);
383 if (!*ost && (*ost = ll_read_opt("osc", this_char)))
385 if (!*mdc && (*mdc = ll_read_opt("mdc", this_char)))
387 if (!(*flags & LL_SBI_NOLCK) &&
388 ((*flags) = (*flags) |
389 ll_set_opt("nolock", this_char,
396 void ll_lli_init(struct ll_inode_info *lli)
398 sema_init(&lli->lli_open_sem, 1);
400 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
401 spin_lock_init(&lli->lli_lock);
402 INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
403 lli->lli_inode_magic = LLI_INODE_MAGIC;
406 int ll_fill_super(struct super_block *sb, void *data, int silent)
408 struct ll_sb_info *sbi;
414 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
416 sbi = lustre_init_sbi(sb);
420 ll_options(data, &osc, &mdc, &sbi->ll_flags);
424 GOTO(out, err = -EINVAL);
429 GOTO(out, err = -EINVAL);
432 err = lustre_common_fill_super(sb, mdc, osc);
438 OBD_FREE(mdc, strlen(mdc) + 1);
440 OBD_FREE(osc, strlen(osc) + 1);
443 } /* ll_read_super */
445 int lustre_process_log(struct lustre_mount_data *lmd, char * profile,
446 struct config_llog_instance *cfg, int allow_recov)
448 struct lustre_cfg lcfg;
449 struct portals_cfg pcfg;
450 char * peer = "MDS_PEER_UUID";
451 struct obd_device *obd;
452 struct lustre_handle mdc_conn = {0, };
453 struct obd_export *exp;
454 char * name = "mdc_dev";
456 struct obd_uuid mdc_uuid;
457 struct llog_ctxt *ctxt;
462 if (lmd_bad_magic(lmd))
465 generate_random_uuid(uuid);
466 class_uuid_unparse(uuid, &mdc_uuid);
468 if (lmd->lmd_local_nid) {
469 PCFG_INIT(pcfg, NAL_CMD_REGISTER_MYNID);
470 pcfg.pcfg_nal = lmd->lmd_nal;
471 pcfg.pcfg_nid = lmd->lmd_local_nid;
472 err = libcfs_nal_cmd(&pcfg);
477 if (lmd->lmd_nal == SOCKNAL ||
478 lmd->lmd_nal == OPENIBNAL ||
479 lmd->lmd_nal == IIBNAL) {
480 PCFG_INIT(pcfg, NAL_CMD_ADD_PEER);
481 pcfg.pcfg_nal = lmd->lmd_nal;
482 pcfg.pcfg_nid = lmd->lmd_server_nid;
483 pcfg.pcfg_id = lmd->lmd_server_ipaddr;
484 pcfg.pcfg_misc = lmd->lmd_port;
485 err = libcfs_nal_cmd(&pcfg);
490 LCFG_INIT(lcfg, LCFG_ADD_UUID, name);
491 lcfg.lcfg_nid = lmd->lmd_server_nid;
492 lcfg.lcfg_inllen1 = strlen(peer) + 1;
493 lcfg.lcfg_inlbuf1 = peer;
494 lcfg.lcfg_nal = lmd->lmd_nal;
495 err = class_process_config(&lcfg);
497 GOTO(out_del_conn, err);
499 LCFG_INIT(lcfg, LCFG_ATTACH, name);
500 lcfg.lcfg_inlbuf1 = "mdc";
501 lcfg.lcfg_inllen1 = strlen(lcfg.lcfg_inlbuf1) + 1;
502 lcfg.lcfg_inlbuf2 = mdc_uuid.uuid;
503 lcfg.lcfg_inllen2 = strlen(lcfg.lcfg_inlbuf2) + 1;
504 err = class_process_config(&lcfg);
506 GOTO(out_del_uuid, err);
508 LCFG_INIT(lcfg, LCFG_SETUP, name);
509 lcfg.lcfg_inlbuf1 = lmd->lmd_mds;
510 lcfg.lcfg_inllen1 = strlen(lcfg.lcfg_inlbuf1) + 1;
511 lcfg.lcfg_inlbuf2 = peer;
512 lcfg.lcfg_inllen2 = strlen(lcfg.lcfg_inlbuf2) + 1;
513 err = class_process_config(&lcfg);
515 GOTO(out_detach, err);
517 obd = class_name2obd(name);
519 GOTO(out_cleanup, err = -EINVAL);
521 /* Disable initial recovery on this import */
522 err = obd_set_info(obd->obd_self_export,
523 strlen("initial_recov"), "initial_recov",
524 sizeof(allow_recov), &allow_recov);
526 GOTO(out_cleanup, err);
528 err = obd_connect(&mdc_conn, obd, &mdc_uuid);
530 CERROR("cannot connect to %s: rc = %d\n", lmd->lmd_mds, err);
531 GOTO(out_cleanup, err);
534 exp = class_conn2export(&mdc_conn);
536 ctxt = llog_get_context(exp->exp_obd, LLOG_CONFIG_REPL_CTXT);
538 rc = class_config_parse_llog(ctxt, profile, cfg);
541 * For debugging, it's useful to just dump the log
543 rc = class_config_dump_llog(ctxt, profile, cfg);
546 CERROR("class_config_parse_llog failed: rc = %d\n", rc);
549 err = obd_disconnect(exp, 0);
552 LCFG_INIT(lcfg, LCFG_CLEANUP, name);
553 err = class_process_config(&lcfg);
558 LCFG_INIT(lcfg, LCFG_DETACH, name);
559 err = class_process_config(&lcfg);
564 LCFG_INIT(lcfg, LCFG_DEL_UUID, name);
565 lcfg.lcfg_inllen1 = strlen(peer) + 1;
566 lcfg.lcfg_inlbuf1 = peer;
567 err = class_process_config(&lcfg);
570 if (lmd->lmd_nal == SOCKNAL ||
571 lmd->lmd_nal == OPENIBNAL ||
572 lmd->lmd_nal == IIBNAL) {
573 PCFG_INIT(pcfg, NAL_CMD_DEL_PEER);
574 pcfg.pcfg_nal = lmd->lmd_nal;
575 pcfg.pcfg_nid = lmd->lmd_server_nid;
576 pcfg.pcfg_flags = 1; /* single_share */
577 err = libcfs_nal_cmd(&pcfg);
588 static void lustre_manual_cleanup(struct ll_sb_info *sbi)
590 struct lustre_cfg lcfg;
591 struct obd_device *obd;
594 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) != NULL)
598 LCFG_INIT(lcfg, LCFG_CLEANUP, obd->obd_name);
599 err = class_process_config(&lcfg);
601 CERROR("cleanup failed: %s\n", obd->obd_name);
605 LCFG_INIT(lcfg, LCFG_DETACH, obd->obd_name);
606 err = class_process_config(&lcfg);
608 CERROR("detach failed: %s\n", obd->obd_name);
613 if (sbi->ll_lmd != NULL)
614 class_del_profile(sbi->ll_lmd->lmd_profile);
617 int lustre_fill_super(struct super_block *sb, void *data, int silent)
619 struct lustre_mount_data * lmd = data;
620 struct ll_sb_info *sbi;
626 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
627 if (lmd_bad_magic(lmd))
630 sbi = lustre_init_sbi(sb);
634 if (lmd->lmd_profile) {
635 struct lustre_profile *lprof;
636 struct config_llog_instance cfg;
639 if (lmd->lmd_mds[0] == '\0') {
640 CERROR("no mds name\n");
641 GOTO(out_free, err = -EINVAL);
644 OBD_ALLOC(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
645 if (sbi->ll_lmd == NULL)
646 GOTO(out_free, err = -ENOMEM);
647 memcpy(sbi->ll_lmd, lmd, sizeof(*lmd));
649 /* generate a string unique to this super, let's try
650 the address of the super itself.*/
651 len = (sizeof(sb) * 2) + 1;
652 OBD_ALLOC(sbi->ll_instance, len);
653 if (sbi->ll_instance == NULL)
654 GOTO(out_free, err = -ENOMEM);
655 sprintf(sbi->ll_instance, "%p", sb);
657 cfg.cfg_instance = sbi->ll_instance;
658 cfg.cfg_uuid = sbi->ll_sb_uuid;
659 cfg.cfg_local_nid = lmd->lmd_local_nid;
660 err = lustre_process_log(lmd, lmd->lmd_profile, &cfg, 0);
662 CERROR("Unable to process log: %s\n", lmd->lmd_profile);
667 lprof = class_get_profile(lmd->lmd_profile);
669 CERROR("No profile found: %s\n", lmd->lmd_profile);
670 GOTO(out_free, err = -EINVAL);
673 OBD_FREE(osc, strlen(osc) + 1);
674 OBD_ALLOC(osc, strlen(lprof->lp_osc) +
675 strlen(sbi->ll_instance) + 2);
676 sprintf(osc, "%s-%s", lprof->lp_osc, sbi->ll_instance);
679 OBD_FREE(mdc, strlen(mdc) + 1);
680 OBD_ALLOC(mdc, strlen(lprof->lp_mdc) +
681 strlen(sbi->ll_instance) + 2);
682 sprintf(mdc, "%s-%s", lprof->lp_mdc, sbi->ll_instance);
687 GOTO(out_free, err = -EINVAL);
692 GOTO(out_free, err = -EINVAL);
695 err = lustre_common_fill_super(sb, mdc, osc);
702 OBD_FREE(mdc, strlen(mdc) + 1);
704 OBD_FREE(osc, strlen(osc) + 1);
710 int len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean")+1;
713 if (sbi->ll_instance != NULL) {
715 struct config_llog_instance cfg;
717 cfg.cfg_instance = sbi->ll_instance;
718 cfg.cfg_uuid = sbi->ll_sb_uuid;
720 OBD_ALLOC(cln_prof, len);
721 sprintf(cln_prof, "%s-clean", sbi->ll_lmd->lmd_profile);
723 err = lustre_process_log(sbi->ll_lmd, cln_prof, &cfg,
726 CERROR("Unable to process log: %s\n", cln_prof);
727 lustre_manual_cleanup(sbi);
729 OBD_FREE(cln_prof, len);
730 OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance)+ 1);
732 OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
737 } /* lustre_fill_super */
739 void lustre_put_super(struct super_block *sb)
741 struct obd_device *obd;
742 struct ll_sb_info *sbi = ll_s2sbi(sb);
743 int force_umount = 0;
746 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
747 obd = class_exp2obd(sbi->ll_mdc_exp);
749 force_umount = obd->obd_no_recov;
752 lustre_common_put_super(sb);
754 if (sbi->ll_lmd != NULL) {
756 int len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean")+1;
758 struct config_llog_instance cfg;
761 CERROR("force umount, doing manual cleanup\n");
762 lustre_manual_cleanup(sbi);
766 cfg.cfg_instance = sbi->ll_instance;
767 cfg.cfg_uuid = sbi->ll_sb_uuid;
769 OBD_ALLOC(cln_prof, len);
770 sprintf(cln_prof, "%s-clean", sbi->ll_lmd->lmd_profile);
772 err = lustre_process_log(sbi->ll_lmd, cln_prof, &cfg, 0);
774 CERROR("Unable to process log: %s, doing manual cleanup"
776 lustre_manual_cleanup(sbi);
779 OBD_FREE(cln_prof, len);
781 OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
782 OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance) + 1);
788 } /* lustre_put_super */
790 #ifdef HAVE_REGISTER_CACHE
791 #include <linux/cache_def.h>
792 #ifdef HAVE_CACHE_RETURN_INT
797 ll_shrink_cache(int priority, unsigned int gfp_mask)
799 struct ll_sb_info *sbi;
802 list_for_each_entry(sbi, &ll_super_blocks, ll_list)
803 count += llap_shrink_cache(sbi, priority);
805 #ifdef HAVE_CACHE_RETURN_INT
810 struct cache_definition ll_cache_definition = {
811 .name = "llap_cache",
812 .shrink = ll_shrink_cache
814 #endif /* HAVE_REGISTER_CACHE */
816 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
818 struct inode *inode = NULL;
819 l_lock(&lock->l_resource->lr_namespace->ns_lock);
820 if (lock->l_ast_data) {
821 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
822 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
823 inode = igrab(lock->l_ast_data);
825 inode = lock->l_ast_data;
826 __LDLM_DEBUG(inode->i_state & I_FREEING ?
827 D_INFO : D_WARNING, lock,
828 "l_ast_data %p is bogus: magic %08x",
829 lock->l_ast_data, lli->lli_inode_magic);
833 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
837 static int null_if_equal(struct ldlm_lock *lock, void *data)
839 if (data == lock->l_ast_data) {
840 lock->l_ast_data = NULL;
842 if (lock->l_req_mode != lock->l_granted_mode)
843 LDLM_ERROR(lock,"clearing inode with ungranted lock");
846 return LDLM_ITER_CONTINUE;
849 void ll_clear_inode(struct inode *inode)
852 struct ll_inode_info *lli = ll_i2info(inode);
853 struct ll_sb_info *sbi = ll_i2sbi(inode);
856 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
857 inode->i_generation, inode);
859 ll_inode2fid(&fid, inode);
860 clear_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &(ll_i2info(inode)->lli_flags));
861 mdc_change_cbdata(sbi->ll_mdc_exp, &fid, null_if_equal, inode);
864 obd_change_cbdata(sbi->ll_osc_exp, lli->lli_smd,
865 null_if_equal, inode);
868 obd_free_memmd(sbi->ll_osc_exp, &lli->lli_smd);
872 if (lli->lli_symlink_name) {
873 OBD_FREE(lli->lli_symlink_name,
874 strlen(lli->lli_symlink_name) + 1);
875 lli->lli_symlink_name = NULL;
877 lli->lli_inode_magic = LLI_INODE_DEAD;
882 /* If this inode has objects allocated to it (lsm != NULL), then the OST
883 * object(s) determine the file size and mtime. Otherwise, the MDS will
884 * keep these values until such a time that objects are allocated for it.
885 * We do the MDS operations first, as it is checking permissions for us.
886 * We don't to the MDS RPC if there is nothing that we want to store there,
887 * otherwise there is no harm in updating mtime/atime on the MDS if we are
888 * going to do an RPC anyways.
890 * If we are doing a truncate, we will send the mtime and ctime updates
891 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
892 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
895 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
897 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
898 struct ll_sb_info *sbi = ll_i2sbi(inode);
899 struct ptlrpc_request *request = NULL;
900 struct mdc_op_data op_data;
901 int ia_valid = attr->ia_valid;
905 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu\n", inode->i_ino);
906 lprocfs_counter_incr(ll_i2sbi(inode)->ll_stats, LPROC_LL_SETATTR);
908 if (ia_valid & ATTR_SIZE) {
909 if (attr->ia_size > ll_file_maxbytes(inode)) {
910 CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
911 attr->ia_size, ll_file_maxbytes(inode));
915 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
918 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
919 if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
920 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
924 /* We mark all of the fields "set" so MDS/OST does not re-set them */
925 if (attr->ia_valid & ATTR_CTIME) {
926 attr->ia_ctime = CURRENT_TIME;
927 attr->ia_valid |= ATTR_CTIME_SET;
929 if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
930 attr->ia_atime = CURRENT_TIME;
931 attr->ia_valid |= ATTR_ATIME_SET;
933 if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
934 attr->ia_mtime = CURRENT_TIME;
935 attr->ia_valid |= ATTR_MTIME_SET;
938 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
939 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
940 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
943 attr->ia_valid &= ~ATTR_SIZE;
945 /* If only OST attributes being set on objects, don't do MDS RPC.
946 * In that case, we need to check permissions and update the local
947 * inode ourselves so we can call obdo_from_inode() always. */
948 if (ia_valid & (lsm ? ~(ATTR_SIZE | ATTR_FROM_OPEN | ATTR_RAW) : ~0)) {
950 ll_prepare_mdc_op_data(&op_data, inode, NULL, NULL, 0, 0);
952 rc = mdc_setattr(sbi->ll_mdc_exp, &op_data,
953 attr, NULL, 0, NULL, 0, &request);
956 ptlrpc_req_finished(request);
957 if (rc != -EPERM && rc != -EACCES)
958 CERROR("mdc_setattr fails: rc = %d\n", rc);
962 rc = mdc_req2lustre_md(request, 0, sbi->ll_osc_exp, &md);
964 ptlrpc_req_finished(request);
968 /* Won't invoke vmtruncate as we already cleared ATTR_SIZE,
969 * but needed to set timestamps backwards on utime. */
970 inode_setattr(inode, attr);
971 ll_update_inode(inode, md.body, md.lsm);
972 ptlrpc_req_finished(request);
974 if (!lsm || !S_ISREG(inode->i_mode)) {
975 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
979 /* The OST doesn't check permissions, but the alternative is
980 * a gratuitous RPC to the MDS. We already rely on the client
981 * to do read/write/truncate permission checks, so is mtime OK?
983 if (ia_valid & (ATTR_MTIME | ATTR_ATIME)) {
984 /* from sys_utime() */
985 if (!(ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET))) {
986 if (current->fsuid != inode->i_uid &&
987 (rc=ll_permission(inode,MAY_WRITE,NULL))!=0)
990 /* from inode_change_ok() */
991 if (current->fsuid != inode->i_uid &&
992 !capable(CAP_FOWNER))
997 /* Won't invoke vmtruncate, as we already cleared ATTR_SIZE */
998 inode_setattr(inode, attr);
1001 /* We really need to get our PW lock before we change inode->i_size.
1002 * If we don't we can race with other i_size updaters on our node, like
1003 * ll_file_read. We can also race with i_size propogation to other
1004 * nodes through dirtying and writeback of final cached pages. This
1005 * last one is especially bad for racing o_append users on other
1007 if (ia_valid & ATTR_SIZE) {
1008 ldlm_policy_data_t policy = { .l_extent = {attr->ia_size,
1010 struct lustre_handle lockh = { 0 };
1011 int err, ast_flags = 0;
1012 /* XXX when we fix the AST intents to pass the discard-range
1013 * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
1015 if (attr->ia_size == 0)
1016 ast_flags = LDLM_AST_DISCARD_DATA;
1018 /* bug 1639: avoid write/truncate i_sem/DLM deadlock */
1019 LASSERT(atomic_read(&inode->i_sem.count) <= 0);
1021 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy, &lockh,
1023 down(&inode->i_sem);
1027 rc = vmtruncate(inode, attr->ia_size);
1029 /* We need to drop the semaphore here, because this unlock may
1030 * result in a cancellation, which will need the i_sem */
1032 /* unlock now as we don't mind others file lockers racing with
1033 * the mds updates below? */
1034 err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
1035 down(&inode->i_sem);
1037 CERROR("ll_extent_unlock failed: %d\n", err);
1041 } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
1044 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1045 inode->i_ino, LTIME_S(attr->ia_mtime));
1046 oa.o_id = lsm->lsm_object_id;
1047 oa.o_valid = OBD_MD_FLID;
1048 obdo_from_inode(&oa, inode, OBD_MD_FLTYPE | OBD_MD_FLATIME |
1049 OBD_MD_FLMTIME | OBD_MD_FLCTIME);
1050 rc = obd_setattr(sbi->ll_osc_exp, &oa, lsm, NULL);
1052 CERROR("obd_setattr fails: rc=%d\n", rc);
1057 int ll_setattr(struct dentry *de, struct iattr *attr)
1059 LBUG(); /* code is unused, but leave this in case of VFS changes */
1063 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1064 unsigned long max_age)
1066 struct ll_sb_info *sbi = ll_s2sbi(sb);
1067 struct obd_statfs obd_osfs;
1071 rc = obd_statfs(class_exp2obd(sbi->ll_mdc_exp), osfs, max_age);
1073 CERROR("mdc_statfs fails: rc = %d\n", rc);
1077 osfs->os_type = sb->s_magic;
1079 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1080 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1082 rc = obd_statfs(class_exp2obd(sbi->ll_osc_exp), &obd_osfs, max_age);
1084 CERROR("obd_statfs fails: rc = %d\n", rc);
1088 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1089 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1092 osfs->os_blocks = obd_osfs.os_blocks;
1093 osfs->os_bfree = obd_osfs.os_bfree;
1094 osfs->os_bavail = obd_osfs.os_bavail;
1096 /* If we don't have as many objects free on the OST as inodes
1097 * on the MDS, we reduce the total number of inodes to
1098 * compensate, so that the "inodes in use" number is correct.
1100 if (obd_osfs.os_ffree < osfs->os_ffree) {
1101 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1103 osfs->os_ffree = obd_osfs.os_ffree;
1109 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1111 struct obd_statfs osfs;
1114 CDEBUG(D_VFSTRACE, "VFS Op:\n");
1115 lprocfs_counter_incr(ll_s2sbi(sb)->ll_stats, LPROC_LL_STAFS);
1117 /* For now we will always get up-to-date statfs values, but in the
1118 * future we may allow some amount of caching on the client (e.g.
1119 * from QOS or lprocfs updates). */
1120 rc = ll_statfs_internal(sb, &osfs, jiffies - 1);
1124 statfs_unpack(sfs, &osfs);
1126 if (sizeof(sfs->f_blocks) == 4) {
1127 while (osfs.os_blocks > ~0UL) {
1130 osfs.os_blocks >>= 1;
1131 osfs.os_bfree >>= 1;
1132 osfs.os_bavail >>= 1;
1136 sfs->f_blocks = osfs.os_blocks;
1137 sfs->f_bfree = osfs.os_bfree;
1138 sfs->f_bavail = osfs.os_bavail;
1143 void dump_lsm(int level, struct lov_stripe_md *lsm)
1145 CDEBUG(level, "objid "LPX64", maxbytes "LPX64", magic 0x%08X, "
1146 "stripe_size %u, stripe_count %u\n",
1147 lsm->lsm_object_id, lsm->lsm_maxbytes, lsm->lsm_magic,
1148 lsm->lsm_stripe_size, lsm->lsm_stripe_count);
1151 void ll_update_inode(struct inode *inode, struct mds_body *body,
1152 struct lov_stripe_md *lsm)
1154 struct ll_inode_info *lli = ll_i2info(inode);
1156 LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1158 if (lli->lli_smd == NULL) {
1160 lli->lli_maxbytes = lsm->lsm_maxbytes;
1161 if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1162 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1164 if (memcmp(lli->lli_smd, lsm, sizeof(*lsm))) {
1165 CERROR("lsm mismatch for inode %ld\n",
1167 CERROR("lli_smd:\n");
1168 dump_lsm(D_ERROR, lli->lli_smd);
1170 dump_lsm(D_ERROR, lsm);
1174 /* bug 2844 - limit i_blksize for broken user-space apps */
1175 LASSERTF(lsm->lsm_xfersize != 0, "%lu\n", lsm->lsm_xfersize);
1176 inode->i_blksize = min(lsm->lsm_xfersize, LL_MAX_BLKSIZE);
1177 if (lli->lli_smd != lsm)
1178 obd_free_memmd(ll_i2obdexp(inode), &lsm);
1181 if (body->valid & OBD_MD_FLID)
1182 inode->i_ino = body->ino;
1183 if (body->valid & OBD_MD_FLATIME)
1184 LTIME_S(inode->i_atime) = body->atime;
1185 if (body->valid & OBD_MD_FLMTIME &&
1186 body->mtime > LTIME_S(inode->i_mtime)) {
1187 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to %u\n",
1188 inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
1189 LTIME_S(inode->i_mtime) = body->mtime;
1191 if (body->valid & OBD_MD_FLCTIME &&
1192 body->ctime > LTIME_S(inode->i_ctime))
1193 LTIME_S(inode->i_ctime) = body->ctime;
1194 if (body->valid & OBD_MD_FLMODE)
1195 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1196 if (body->valid & OBD_MD_FLTYPE)
1197 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1198 if (body->valid & OBD_MD_FLUID)
1199 inode->i_uid = body->uid;
1200 if (body->valid & OBD_MD_FLGID)
1201 inode->i_gid = body->gid;
1202 if (body->valid & OBD_MD_FLFLAGS)
1203 inode->i_flags = body->flags;
1204 if (body->valid & OBD_MD_FLNLINK)
1205 inode->i_nlink = body->nlink;
1206 if (body->valid & OBD_MD_FLGENER)
1207 inode->i_generation = body->generation;
1208 if (body->valid & OBD_MD_FLRDEV)
1209 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1210 inode->i_rdev = body->rdev;
1212 inode->i_rdev = old_decode_dev(body->rdev);
1214 if (body->valid & OBD_MD_FLSIZE)
1215 inode->i_size = body->size;
1216 if (body->valid & OBD_MD_FLBLOCKS)
1217 inode->i_blocks = body->blocks;
1219 if (body->valid & OBD_MD_FLSIZE)
1220 set_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &lli->lli_flags);
1223 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
1224 static struct backing_dev_info ll_backing_dev_info = {
1225 .ra_pages = 0, /* No readahead */
1226 .memory_backed = 0, /* Does contribute to dirty memory */
1230 void ll_read_inode2(struct inode *inode, void *opaque)
1232 struct lustre_md *md = opaque;
1233 struct ll_inode_info *lli = ll_i2info(inode);
1236 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1237 inode->i_generation, inode);
1241 LASSERT(!lli->lli_smd);
1243 /* Core attributes from the MDS first. This is a new inode, and
1244 * the VFS doesn't zero times in the core inode so we have to do
1245 * it ourselves. They will be overwritten by either MDS or OST
1246 * attributes - we just need to make sure they aren't newer. */
1247 LTIME_S(inode->i_mtime) = 0;
1248 LTIME_S(inode->i_atime) = 0;
1249 LTIME_S(inode->i_ctime) = 0;
1251 ll_update_inode(inode, md->body, md->lsm);
1253 /* OIDEBUG(inode); */
1255 if (S_ISREG(inode->i_mode)) {
1256 inode->i_op = &ll_file_inode_operations;
1257 inode->i_fop = &ll_file_operations;
1258 inode->i_mapping->a_ops = &ll_aops;
1260 } else if (S_ISDIR(inode->i_mode)) {
1261 inode->i_op = &ll_dir_inode_operations;
1262 inode->i_fop = &ll_dir_operations;
1263 inode->i_mapping->a_ops = &ll_dir_aops;
1265 } else if (S_ISLNK(inode->i_mode)) {
1266 inode->i_op = &ll_fast_symlink_inode_operations;
1269 inode->i_op = &ll_special_inode_operations;
1271 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
1272 init_special_inode(inode, inode->i_mode,
1273 kdev_t_to_nr(inode->i_rdev));
1275 /* initializing backing dev info. */
1276 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
1278 init_special_inode(inode, inode->i_mode, inode->i_rdev);
1280 lli->ll_save_ifop = inode->i_fop;
1282 if (S_ISCHR(inode->i_mode))
1283 inode->i_fop = &ll_special_chr_inode_fops;
1284 else if (S_ISBLK(inode->i_mode))
1285 inode->i_fop = &ll_special_blk_inode_fops;
1286 else if (S_ISFIFO(inode->i_mode))
1287 inode->i_fop = &ll_special_fifo_inode_fops;
1288 else if (S_ISSOCK(inode->i_mode))
1289 inode->i_fop = &ll_special_sock_inode_fops;
1294 int ll_iocontrol(struct inode *inode, struct file *file,
1295 unsigned int cmd, unsigned long arg)
1297 struct ll_sb_info *sbi = ll_i2sbi(inode);
1298 struct ptlrpc_request *req = NULL;
1303 case EXT3_IOC_GETFLAGS: {
1305 unsigned long valid = OBD_MD_FLFLAGS;
1306 struct mds_body *body;
1308 ll_inode2fid(&fid, inode);
1309 rc = mdc_getattr(sbi->ll_mdc_exp, &fid, valid, 0, &req);
1311 CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1315 body = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*body));
1317 if (body->flags & S_APPEND)
1318 flags |= EXT3_APPEND_FL;
1319 if (body->flags & S_IMMUTABLE)
1320 flags |= EXT3_IMMUTABLE_FL;
1321 if (body->flags & S_NOATIME)
1322 flags |= EXT3_NOATIME_FL;
1324 ptlrpc_req_finished (req);
1326 RETURN(put_user(flags, (int *)arg));
1328 case EXT3_IOC_SETFLAGS: {
1329 struct mdc_op_data op_data;
1332 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1334 if (get_user(flags, (int *)arg))
1341 ll_prepare_mdc_op_data(&op_data, inode, NULL, NULL, 0, 0);
1343 memset(&attr, 0x0, sizeof(attr));
1344 attr.ia_attr_flags = flags;
1345 attr.ia_valid |= ATTR_ATTR_FLAG;
1347 rc = mdc_setattr(sbi->ll_mdc_exp, &op_data,
1348 &attr, NULL, 0, NULL, 0, &req);
1350 ptlrpc_req_finished(req);
1351 if (rc != -EPERM && rc != -EACCES)
1352 CERROR("mdc_setattr fails: rc = %d\n", rc);
1356 ptlrpc_req_finished(req);
1358 oa->o_id = lsm->lsm_object_id;
1359 oa->o_flags = flags;
1360 oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS;
1362 rc = obd_setattr(sbi->ll_osc_exp, oa, lsm, NULL);
1365 if (rc != -EPERM && rc != -EACCES)
1366 CERROR("mdc_setattr fails: rc = %d\n", rc);
1370 if (flags & EXT3_APPEND_FL)
1371 inode->i_flags |= S_APPEND;
1373 inode->i_flags &= ~S_APPEND;
1374 if (flags & EXT3_IMMUTABLE_FL)
1375 inode->i_flags |= S_IMMUTABLE;
1377 inode->i_flags &= ~S_IMMUTABLE;
1378 if (flags & EXT3_NOATIME_FL)
1379 inode->i_flags |= S_NOATIME;
1381 inode->i_flags &= ~S_NOATIME;
1392 void ll_umount_begin(struct super_block *sb)
1394 struct ll_sb_info *sbi = ll_s2sbi(sb);
1395 struct obd_device *obd;
1396 struct obd_ioctl_data ioc_data = { 0 };
1398 CDEBUG(D_VFSTRACE, "VFS Op:\n");
1400 obd = class_exp2obd(sbi->ll_mdc_exp);
1402 CERROR("Invalid MDC connection handle "LPX64"\n",
1403 sbi->ll_mdc_exp->exp_handle.h_cookie);
1407 obd->obd_no_recov = 1;
1408 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_mdc_exp, sizeof ioc_data,
1411 obd = class_exp2obd(sbi->ll_osc_exp);
1413 CERROR("Invalid LOV connection handle "LPX64"\n",
1414 sbi->ll_osc_exp->exp_handle.h_cookie);
1419 obd->obd_no_recov = 1;
1420 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_osc_exp, sizeof ioc_data,
1423 /* Really, we'd like to wait until there are no requests outstanding,
1424 * and then continue. For now, we just invalidate the requests,
1425 * schedule, and hope.
1432 int ll_prep_inode(struct obd_export *exp, struct inode **inode,
1433 struct ptlrpc_request *req, int offset,struct super_block *sb)
1435 struct lustre_md md;
1438 rc = mdc_req2lustre_md(req, offset, exp, &md);
1443 ll_update_inode(*inode, md.body, md.lsm);
1446 *inode = ll_iget(sb, md.body->ino, &md);
1447 if (*inode == NULL || is_bad_inode(*inode)) {
1448 /* free the lsm if we allocated one above */
1450 obd_free_memmd(exp, &md.lsm);
1452 CERROR("new_inode -fatal: rc %d\n", rc);
1459 char *llap_origins[] = {
1460 [LLAP_ORIGIN_UNKNOWN] = "--",
1461 [LLAP_ORIGIN_READPAGE] = "rp",
1462 [LLAP_ORIGIN_READAHEAD] = "ra",
1463 [LLAP_ORIGIN_COMMIT_WRITE] = "cw",
1464 [LLAP_ORIGIN_WRITEPAGE] = "wp",
1467 struct ll_async_page *llite_pglist_next_llap(struct ll_sb_info *sbi,
1468 struct list_head *list)
1470 struct ll_async_page *llap;
1471 struct list_head *pos;
1473 list_for_each(pos, list) {
1474 if (pos == &sbi->ll_pglist)
1476 llap = list_entry(pos, struct ll_async_page, llap_pglist_item);
1477 if (llap->llap_page == NULL)