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 err = obd_connect(&mdc_conn, obd, &sbi->ll_sb_uuid);
126 CERROR("An MDS (mdc %s) is performing recovery, of which this"
127 " client is not a part. Please wait for recovery to "
128 "complete, abort, or time out.\n", mdc);
131 CERROR("cannot connect to %s: rc = %d\n", mdc, err);
134 sbi->ll_mdc_exp = class_conn2export(&mdc_conn);
136 err = obd_statfs(obd, &osfs, jiffies - HZ);
140 LASSERT(osfs.os_bsize);
141 sb->s_blocksize = osfs.os_bsize;
142 sb->s_blocksize_bits = log2(osfs.os_bsize);
143 sb->s_magic = LL_SUPER_MAGIC;
144 sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
145 sbi->ll_namelen = osfs.os_namelen;
147 devno = get_uuid2int(sbi2mdc(sbi)->cl_import->imp_target_uuid.uuid,
148 strlen(sbi2mdc(sbi)->cl_import->imp_target_uuid.uuid));
149 /* s_dev is also used in lt_compare() to compare two fs */
152 obd = class_name2obd(osc);
154 CERROR("OSC %s: not setup or attached\n", osc);
158 err = obd_connect(&osc_conn, obd, &sbi->ll_sb_uuid);
160 CERROR("An OST (osc %s) is performing recovery, of which this"
161 " client is not a part. Please wait for recovery to "
162 "complete, abort, or time out.\n", osc);
165 CERROR("cannot connect to %s: rc = %d\n", osc, err);
168 sbi->ll_osc_exp = class_conn2export(&osc_conn);
170 mdc_init_ea_size(sbi->ll_mdc_exp, sbi->ll_osc_exp);
172 if (!ll_async_page_slab) {
173 ll_async_page_slab_size =
174 size_round(sizeof(struct ll_async_page)) +
175 obd_prep_async_page(sbi->ll_osc_exp, NULL, NULL, NULL,
176 0, NULL, NULL, NULL);
177 ll_async_page_slab = kmem_cache_create("ll_async_page",
178 ll_async_page_slab_size,
180 if (!ll_async_page_slab)
181 GOTO(out_osc, -ENOMEM);
184 err = mdc_getstatus(sbi->ll_mdc_exp, &rootfid);
186 CERROR("cannot mds_connect: rc = %d\n", err);
189 CDEBUG(D_SUPER, "rootfid "LPU64"\n", rootfid.id);
190 sbi->ll_rootino = rootfid.id;
192 sb->s_op = &lustre_super_operations;
195 * XXX: move this to after cbd setup? */
196 err = mdc_getattr(sbi->ll_mdc_exp, &rootfid,
197 OBD_MD_FLNOTOBD|OBD_MD_FLBLOCKS, 0, &request);
199 CERROR("mdc_getattr failed for root: rc = %d\n", err);
203 err = mdc_req2lustre_md(request, 0, sbi->ll_osc_exp, &md);
205 CERROR("failed to understand root inode md: rc = %d\n",err);
206 ptlrpc_req_finished (request);
210 LASSERT(sbi->ll_rootino != 0);
211 root = ll_iget(sb, sbi->ll_rootino, &md);
213 ptlrpc_req_finished(request);
215 if (root == NULL || is_bad_inode(root)) {
216 /* XXX might need iput() for bad inode */
217 CERROR("lustre_lite: bad iget4 for root\n");
218 GOTO(out_root, err = -EBADF);
221 err = ll_close_thread_start(&sbi->ll_lcq);
223 CERROR("cannot start close thread: rc %d\n", err);
227 /* making vm readahead 0 for 2.4.x. In the case of 2.6.x,
228 backing dev info assigned to inode mapping is used for
229 determining maximal readahead. */
230 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0)) && \
231 !defined(KERNEL_HAS_AS_MAX_READAHEAD)
232 /* bug 2805 - set VM readahead to zero */
233 vm_max_readahead = vm_min_readahead = 0;
236 sb->s_root = d_alloc_root(root);
243 obd_disconnect(sbi->ll_osc_exp, 0);
245 obd_disconnect(sbi->ll_mdc_exp, 0);
247 lprocfs_unregister_mountpoint(sbi);
251 void lustre_dump_inode(struct inode *inode)
253 struct list_head *tmp;
254 int dentry_count = 0;
255 BDEVNAME_DECLARE_STORAGE(buf);
257 LASSERT(inode != NULL);
259 list_for_each(tmp, &inode->i_dentry)
262 CERROR("inode %p dump: dev=%s:%lu, mode=%o, count=%u, %d dentries\n",
263 inode, ll_bdevname(inode->i_sb, buf), inode->i_ino,
264 inode->i_mode, atomic_read(&inode->i_count), dentry_count);
267 void lustre_dump_dentry(struct dentry *dentry, int recur)
269 struct list_head *tmp;
272 LASSERT(dentry != NULL);
274 list_for_each(tmp, &dentry->d_subdirs)
277 CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
278 " flags=0x%x, vfs_flags=0x%lx, fsdata=%p, %d subdirs\n", dentry,
279 dentry->d_name.len, dentry->d_name.name,
280 dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
281 dentry->d_parent, dentry->d_inode, atomic_read(&dentry->d_count),
282 dentry->d_flags, dentry->d_vfs_flags, dentry->d_fsdata, subdirs);
283 if (dentry->d_inode != NULL)
284 lustre_dump_inode(dentry->d_inode);
289 list_for_each(tmp, &dentry->d_subdirs) {
290 struct dentry *d = list_entry(tmp, struct dentry, d_child);
291 lustre_dump_dentry(d, recur - 1);
295 void lustre_common_put_super(struct super_block *sb)
297 struct ll_sb_info *sbi = ll_s2sbi(sb);
298 struct hlist_node *tmp, *next;
301 ll_close_thread_shutdown(sbi->ll_lcq);
303 list_del(&sbi->ll_conn_chain);
304 obd_disconnect(sbi->ll_osc_exp, 0);
306 lprocfs_unregister_mountpoint(sbi);
307 if (sbi->ll_proc_root) {
308 lprocfs_remove(sbi->ll_proc_root);
309 sbi->ll_proc_root = NULL;
312 obd_disconnect(sbi->ll_mdc_exp, 0);
314 // We do this to get rid of orphaned dentries. That is not really trw.
315 hlist_for_each_safe(tmp, next, &sbi->ll_orphan_dentry_list) {
316 struct dentry *dentry = hlist_entry(tmp, struct dentry, d_hash);
317 CWARN("found orphan dentry %.*s (%p->%p) at unmount, dumping "
318 "before and after shrink_dcache_parent\n",
319 dentry->d_name.len, dentry->d_name.name, dentry, next);
320 lustre_dump_dentry(dentry, 1);
321 shrink_dcache_parent(dentry);
322 lustre_dump_dentry(dentry, 1);
328 char *ll_read_opt(const char *opt, char *data)
334 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
335 if (strncmp(opt, data, strlen(opt)))
337 if ((value = strchr(data, '=')) == NULL)
341 OBD_ALLOC(retval, strlen(value) + 1);
343 CERROR("out of memory!\n");
347 memcpy(retval, value, strlen(value)+1);
348 CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
352 int ll_set_opt(const char *opt, char *data, int fl)
356 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
357 if (strncmp(opt, data, strlen(opt)))
363 void ll_options(char *options, char **ost, char **mdc, int *flags)
366 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
367 char *opt_ptr = options;
376 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
377 for (this_char = strtok (options, ",");
379 this_char = strtok (NULL, ",")) {
381 while ((this_char = strsep (&opt_ptr, ",")) != NULL) {
383 CDEBUG(D_SUPER, "this_char %s\n", this_char);
384 if (!*ost && (*ost = ll_read_opt("osc", this_char)))
386 if (!*mdc && (*mdc = ll_read_opt("mdc", this_char)))
388 if (!(*flags & LL_SBI_NOLCK) &&
389 ((*flags) = (*flags) |
390 ll_set_opt("nolock", this_char,
397 void ll_lli_init(struct ll_inode_info *lli)
399 sema_init(&lli->lli_open_sem, 1);
400 sema_init(&lli->lli_size_sem, 1);
402 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
403 spin_lock_init(&lli->lli_lock);
404 INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
405 lli->lli_inode_magic = LLI_INODE_MAGIC;
408 int ll_fill_super(struct super_block *sb, void *data, int silent)
410 struct ll_sb_info *sbi;
416 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
418 sbi = lustre_init_sbi(sb);
422 ll_options(data, &osc, &mdc, &sbi->ll_flags);
426 GOTO(out, err = -EINVAL);
431 GOTO(out, err = -EINVAL);
434 err = lustre_common_fill_super(sb, mdc, osc);
440 OBD_FREE(mdc, strlen(mdc) + 1);
442 OBD_FREE(osc, strlen(osc) + 1);
445 } /* ll_read_super */
447 int lustre_process_log(struct lustre_mount_data *lmd, char * profile,
448 struct config_llog_instance *cfg, int allow_recov)
450 struct lustre_cfg lcfg;
451 struct portals_cfg pcfg;
452 char * peer = "MDS_PEER_UUID";
453 struct obd_device *obd;
454 struct lustre_handle mdc_conn = {0, };
455 struct obd_export *exp;
456 char * name = "mdc_dev";
458 struct obd_uuid mdc_uuid;
459 struct llog_ctxt *ctxt;
464 if (lmd_bad_magic(lmd))
467 generate_random_uuid(uuid);
468 class_uuid_unparse(uuid, &mdc_uuid);
470 if (lmd->lmd_local_nid) {
471 PCFG_INIT(pcfg, NAL_CMD_REGISTER_MYNID);
472 pcfg.pcfg_nal = lmd->lmd_nal;
473 pcfg.pcfg_nid = lmd->lmd_local_nid;
474 err = libcfs_nal_cmd(&pcfg);
479 if (lmd->lmd_nal == SOCKNAL ||
480 lmd->lmd_nal == OPENIBNAL ||
481 lmd->lmd_nal == IIBNAL ||
482 lmd->lmd_nal == VIBNAL ||
483 lmd->lmd_nal == RANAL) {
484 PCFG_INIT(pcfg, NAL_CMD_ADD_PEER);
485 pcfg.pcfg_nal = lmd->lmd_nal;
486 pcfg.pcfg_nid = lmd->lmd_server_nid;
487 pcfg.pcfg_id = lmd->lmd_server_ipaddr;
488 pcfg.pcfg_misc = lmd->lmd_port;
489 err = libcfs_nal_cmd(&pcfg);
494 LCFG_INIT(lcfg, LCFG_ADD_UUID, name);
495 lcfg.lcfg_nid = lmd->lmd_server_nid;
496 lcfg.lcfg_inllen1 = strlen(peer) + 1;
497 lcfg.lcfg_inlbuf1 = peer;
498 lcfg.lcfg_nal = lmd->lmd_nal;
499 err = class_process_config(&lcfg);
501 GOTO(out_del_conn, err);
503 LCFG_INIT(lcfg, LCFG_ATTACH, name);
504 lcfg.lcfg_inlbuf1 = "mdc";
505 lcfg.lcfg_inllen1 = strlen(lcfg.lcfg_inlbuf1) + 1;
506 lcfg.lcfg_inlbuf2 = mdc_uuid.uuid;
507 lcfg.lcfg_inllen2 = strlen(lcfg.lcfg_inlbuf2) + 1;
508 err = class_process_config(&lcfg);
510 GOTO(out_del_uuid, err);
512 LCFG_INIT(lcfg, LCFG_SETUP, name);
513 lcfg.lcfg_inlbuf1 = lmd->lmd_mds;
514 lcfg.lcfg_inllen1 = strlen(lcfg.lcfg_inlbuf1) + 1;
515 lcfg.lcfg_inlbuf2 = peer;
516 lcfg.lcfg_inllen2 = strlen(lcfg.lcfg_inlbuf2) + 1;
517 err = class_process_config(&lcfg);
519 GOTO(out_detach, err);
521 obd = class_name2obd(name);
523 GOTO(out_cleanup, err = -EINVAL);
525 /* Disable initial recovery on this import */
526 err = obd_set_info(obd->obd_self_export,
527 strlen("initial_recov"), "initial_recov",
528 sizeof(allow_recov), &allow_recov);
530 GOTO(out_cleanup, err);
532 err = obd_connect(&mdc_conn, obd, &mdc_uuid);
534 CERROR("cannot connect to %s: rc = %d\n", lmd->lmd_mds, err);
535 GOTO(out_cleanup, err);
538 exp = class_conn2export(&mdc_conn);
540 ctxt = llog_get_context(exp->exp_obd, LLOG_CONFIG_REPL_CTXT);
542 rc = class_config_parse_llog(ctxt, profile, cfg);
545 * For debugging, it's useful to just dump the log
547 rc = class_config_dump_llog(ctxt, profile, cfg);
550 CERROR("class_config_parse_llog failed: rc = %d\n", rc);
553 err = obd_disconnect(exp, 0);
556 LCFG_INIT(lcfg, LCFG_CLEANUP, name);
557 err = class_process_config(&lcfg);
562 LCFG_INIT(lcfg, LCFG_DETACH, name);
563 err = class_process_config(&lcfg);
568 LCFG_INIT(lcfg, LCFG_DEL_UUID, name);
569 lcfg.lcfg_inllen1 = strlen(peer) + 1;
570 lcfg.lcfg_inlbuf1 = peer;
571 err = class_process_config(&lcfg);
574 if (lmd->lmd_nal == SOCKNAL ||
575 lmd->lmd_nal == OPENIBNAL ||
576 lmd->lmd_nal == IIBNAL ||
577 lmd->lmd_nal == VIBNAL ||
578 lmd->lmd_nal == RANAL) {
579 PCFG_INIT(pcfg, NAL_CMD_DEL_PEER);
580 pcfg.pcfg_nal = lmd->lmd_nal;
581 pcfg.pcfg_nid = lmd->lmd_server_nid;
582 pcfg.pcfg_flags = 1; /* single_share */
583 err = libcfs_nal_cmd(&pcfg);
594 static void lustre_manual_cleanup(struct ll_sb_info *sbi)
596 struct lustre_cfg lcfg;
597 struct obd_device *obd;
600 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) != NULL)
604 LCFG_INIT(lcfg, LCFG_CLEANUP, obd->obd_name);
605 err = class_process_config(&lcfg);
607 CERROR("cleanup failed: %s\n", obd->obd_name);
611 LCFG_INIT(lcfg, LCFG_DETACH, obd->obd_name);
612 err = class_process_config(&lcfg);
614 CERROR("detach failed: %s\n", obd->obd_name);
619 if (sbi->ll_lmd != NULL)
620 class_del_profile(sbi->ll_lmd->lmd_profile);
623 int lustre_fill_super(struct super_block *sb, void *data, int silent)
625 struct lustre_mount_data * lmd = data;
626 struct ll_sb_info *sbi;
632 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
633 if (lmd_bad_magic(lmd))
636 sbi = lustre_init_sbi(sb);
640 if (lmd->lmd_profile) {
641 struct lustre_profile *lprof;
642 struct config_llog_instance cfg;
645 if (lmd->lmd_mds[0] == '\0') {
646 CERROR("no mds name\n");
647 GOTO(out_free, err = -EINVAL);
650 OBD_ALLOC(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
651 if (sbi->ll_lmd == NULL)
652 GOTO(out_free, err = -ENOMEM);
653 memcpy(sbi->ll_lmd, lmd, sizeof(*lmd));
655 /* generate a string unique to this super, let's try
656 the address of the super itself.*/
657 len = (sizeof(sb) * 2) + 1;
658 OBD_ALLOC(sbi->ll_instance, len);
659 if (sbi->ll_instance == NULL)
660 GOTO(out_free, err = -ENOMEM);
661 sprintf(sbi->ll_instance, "%p", sb);
663 cfg.cfg_instance = sbi->ll_instance;
664 cfg.cfg_uuid = sbi->ll_sb_uuid;
665 cfg.cfg_local_nid = lmd->lmd_local_nid;
666 err = lustre_process_log(lmd, lmd->lmd_profile, &cfg, 0);
668 CERROR("Unable to process log: %s\n", lmd->lmd_profile);
673 lprof = class_get_profile(lmd->lmd_profile);
675 CERROR("No profile found: %s\n", lmd->lmd_profile);
676 GOTO(out_free, err = -EINVAL);
679 OBD_FREE(osc, strlen(osc) + 1);
680 OBD_ALLOC(osc, strlen(lprof->lp_osc) +
681 strlen(sbi->ll_instance) + 2);
682 sprintf(osc, "%s-%s", lprof->lp_osc, sbi->ll_instance);
685 OBD_FREE(mdc, strlen(mdc) + 1);
686 OBD_ALLOC(mdc, strlen(lprof->lp_mdc) +
687 strlen(sbi->ll_instance) + 2);
688 sprintf(mdc, "%s-%s", lprof->lp_mdc, sbi->ll_instance);
693 GOTO(out_free, err = -EINVAL);
698 GOTO(out_free, err = -EINVAL);
701 err = lustre_common_fill_super(sb, mdc, osc);
708 OBD_FREE(mdc, strlen(mdc) + 1);
710 OBD_FREE(osc, strlen(osc) + 1);
716 int len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean")+1;
719 if (sbi->ll_instance != NULL) {
721 struct config_llog_instance cfg;
723 cfg.cfg_instance = sbi->ll_instance;
724 cfg.cfg_uuid = sbi->ll_sb_uuid;
726 OBD_ALLOC(cln_prof, len);
727 sprintf(cln_prof, "%s-clean", sbi->ll_lmd->lmd_profile);
729 err = lustre_process_log(sbi->ll_lmd, cln_prof, &cfg,
732 CERROR("Unable to process log: %s\n", cln_prof);
733 lustre_manual_cleanup(sbi);
735 OBD_FREE(cln_prof, len);
736 OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance)+ 1);
738 OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
743 } /* lustre_fill_super */
745 void lustre_put_super(struct super_block *sb)
747 struct obd_device *obd;
748 struct ll_sb_info *sbi = ll_s2sbi(sb);
749 int force_umount = 0;
752 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
753 obd = class_exp2obd(sbi->ll_mdc_exp);
755 force_umount = obd->obd_no_recov;
758 lustre_common_put_super(sb);
760 if (sbi->ll_lmd != NULL) {
762 int len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean")+1;
764 struct config_llog_instance cfg;
767 CERROR("force umount, doing manual cleanup\n");
768 lustre_manual_cleanup(sbi);
772 cfg.cfg_instance = sbi->ll_instance;
773 cfg.cfg_uuid = sbi->ll_sb_uuid;
775 OBD_ALLOC(cln_prof, len);
776 sprintf(cln_prof, "%s-clean", sbi->ll_lmd->lmd_profile);
778 err = lustre_process_log(sbi->ll_lmd, cln_prof, &cfg, 0);
780 CERROR("Unable to process log: %s, doing manual cleanup"
782 lustre_manual_cleanup(sbi);
785 OBD_FREE(cln_prof, len);
787 OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
788 OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance) + 1);
794 } /* lustre_put_super */
796 #ifdef HAVE_REGISTER_CACHE
797 #include <linux/cache_def.h>
798 #ifdef HAVE_CACHE_RETURN_INT
803 ll_shrink_cache(int priority, unsigned int gfp_mask)
805 struct ll_sb_info *sbi;
808 list_for_each_entry(sbi, &ll_super_blocks, ll_list)
809 count += llap_shrink_cache(sbi, priority);
811 #ifdef HAVE_CACHE_RETURN_INT
816 struct cache_definition ll_cache_definition = {
817 .name = "llap_cache",
818 .shrink = ll_shrink_cache
820 #endif /* HAVE_REGISTER_CACHE */
822 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
824 struct inode *inode = NULL;
825 l_lock(&lock->l_resource->lr_namespace->ns_lock);
826 if (lock->l_ast_data) {
827 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
828 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
829 inode = igrab(lock->l_ast_data);
831 inode = lock->l_ast_data;
832 __LDLM_DEBUG(inode->i_state & I_FREEING ?
833 D_INFO : D_WARNING, lock,
834 "l_ast_data %p is bogus: magic %08x",
835 lock->l_ast_data, lli->lli_inode_magic);
839 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
843 static int null_if_equal(struct ldlm_lock *lock, void *data)
845 if (data == lock->l_ast_data) {
846 lock->l_ast_data = NULL;
848 if (lock->l_req_mode != lock->l_granted_mode)
849 LDLM_ERROR(lock,"clearing inode with ungranted lock");
852 return LDLM_ITER_CONTINUE;
855 void ll_clear_inode(struct inode *inode)
858 struct ll_inode_info *lli = ll_i2info(inode);
859 struct ll_sb_info *sbi = ll_i2sbi(inode);
862 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
863 inode->i_generation, inode);
865 ll_inode2fid(&fid, inode);
866 clear_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &(ll_i2info(inode)->lli_flags));
867 mdc_change_cbdata(sbi->ll_mdc_exp, &fid, null_if_equal, inode);
870 obd_change_cbdata(sbi->ll_osc_exp, lli->lli_smd,
871 null_if_equal, inode);
874 obd_free_memmd(sbi->ll_osc_exp, &lli->lli_smd);
878 if (lli->lli_symlink_name) {
879 OBD_FREE(lli->lli_symlink_name,
880 strlen(lli->lli_symlink_name) + 1);
881 lli->lli_symlink_name = NULL;
883 lli->lli_inode_magic = LLI_INODE_DEAD;
888 /* If this inode has objects allocated to it (lsm != NULL), then the OST
889 * object(s) determine the file size and mtime. Otherwise, the MDS will
890 * keep these values until such a time that objects are allocated for it.
891 * We do the MDS operations first, as it is checking permissions for us.
892 * We don't to the MDS RPC if there is nothing that we want to store there,
893 * otherwise there is no harm in updating mtime/atime on the MDS if we are
894 * going to do an RPC anyways.
896 * If we are doing a truncate, we will send the mtime and ctime updates
897 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
898 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
901 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
903 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
904 struct ll_sb_info *sbi = ll_i2sbi(inode);
905 struct ptlrpc_request *request = NULL;
906 struct mdc_op_data op_data;
907 int ia_valid = attr->ia_valid;
911 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu\n", inode->i_ino);
912 lprocfs_counter_incr(ll_i2sbi(inode)->ll_stats, LPROC_LL_SETATTR);
914 if (ia_valid & ATTR_SIZE) {
915 if (attr->ia_size > ll_file_maxbytes(inode)) {
916 CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
917 attr->ia_size, ll_file_maxbytes(inode));
921 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
924 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
925 if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
926 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
930 /* We mark all of the fields "set" so MDS/OST does not re-set them */
931 if (attr->ia_valid & ATTR_CTIME) {
932 attr->ia_ctime = CURRENT_TIME;
933 attr->ia_valid |= ATTR_CTIME_SET;
935 if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
936 attr->ia_atime = CURRENT_TIME;
937 attr->ia_valid |= ATTR_ATIME_SET;
939 if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
940 attr->ia_mtime = CURRENT_TIME;
941 attr->ia_valid |= ATTR_MTIME_SET;
944 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
945 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
946 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
949 attr->ia_valid &= ~ATTR_SIZE;
951 /* If only OST attributes being set on objects, don't do MDS RPC.
952 * In that case, we need to check permissions and update the local
953 * inode ourselves so we can call obdo_from_inode() always. */
954 if (ia_valid & (lsm ? ~(ATTR_SIZE | ATTR_FROM_OPEN | ATTR_RAW) : ~0)) {
957 ll_prepare_mdc_op_data(&op_data, inode, NULL, NULL, 0, 0);
959 rc = mdc_setattr(sbi->ll_mdc_exp, &op_data,
960 attr, NULL, 0, NULL, 0, &request);
963 ptlrpc_req_finished(request);
964 if (rc != -EPERM && rc != -EACCES)
965 CERROR("mdc_setattr fails: rc = %d\n", rc);
969 rc = mdc_req2lustre_md(request, 0, sbi->ll_osc_exp, &md);
971 ptlrpc_req_finished(request);
975 /* We call inode_setattr to adjust timestamps, but we first
976 * clear ATTR_SIZE to avoid invoking vmtruncate.
978 * NB: ATTR_SIZE will only be set at this point if the size
979 * resides on the MDS, ie, this file has no objects. */
980 save_valid = attr->ia_valid;
981 attr->ia_valid &= ~ATTR_SIZE;
982 inode_setattr(inode, attr);
983 attr->ia_valid = save_valid;
985 ll_update_inode(inode, md.body, md.lsm);
986 ptlrpc_req_finished(request);
988 if (!lsm || !S_ISREG(inode->i_mode)) {
989 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
993 /* The OST doesn't check permissions, but the alternative is
994 * a gratuitous RPC to the MDS. We already rely on the client
995 * to do read/write/truncate permission checks, so is mtime OK?
997 if (ia_valid & (ATTR_MTIME | ATTR_ATIME)) {
998 /* from sys_utime() */
999 if (!(ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET))) {
1000 if (current->fsuid != inode->i_uid &&
1001 (rc=ll_permission(inode,MAY_WRITE,NULL))!=0)
1004 /* from inode_change_ok() */
1005 if (current->fsuid != inode->i_uid &&
1006 !capable(CAP_FOWNER))
1011 /* Won't invoke vmtruncate, as we already cleared ATTR_SIZE */
1012 inode_setattr(inode, attr);
1015 /* We really need to get our PW lock before we change inode->i_size.
1016 * If we don't we can race with other i_size updaters on our node, like
1017 * ll_file_read. We can also race with i_size propogation to other
1018 * nodes through dirtying and writeback of final cached pages. This
1019 * last one is especially bad for racing o_append users on other
1021 if (ia_valid & ATTR_SIZE) {
1022 ldlm_policy_data_t policy = { .l_extent = {attr->ia_size,
1024 struct lustre_handle lockh = { 0 };
1025 struct ll_inode_info *lli = ll_i2info(inode);
1026 int err, ast_flags = 0;
1027 /* XXX when we fix the AST intents to pass the discard-range
1028 * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
1030 if (attr->ia_size == 0)
1031 ast_flags = LDLM_AST_DISCARD_DATA;
1033 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy, &lockh,
1038 down(&lli->lli_size_sem);
1039 rc = vmtruncate(inode, attr->ia_size);
1040 // if vmtruncate returned 0, then ll_truncate dropped _size_sem
1042 LASSERT(atomic_read(&lli->lli_size_sem.count) <= 0);
1043 up(&lli->lli_size_sem);
1046 err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
1048 CERROR("ll_extent_unlock failed: %d\n", err);
1052 } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
1055 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1056 inode->i_ino, LTIME_S(attr->ia_mtime));
1057 oa.o_id = lsm->lsm_object_id;
1058 oa.o_valid = OBD_MD_FLID;
1059 obdo_from_inode(&oa, inode, OBD_MD_FLTYPE | OBD_MD_FLATIME |
1060 OBD_MD_FLMTIME | OBD_MD_FLCTIME);
1061 rc = obd_setattr(sbi->ll_osc_exp, &oa, lsm, NULL);
1063 CERROR("obd_setattr fails: rc=%d\n", rc);
1068 int ll_setattr(struct dentry *de, struct iattr *attr)
1070 LBUG(); /* code is unused, but leave this in case of VFS changes */
1074 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1075 unsigned long max_age)
1077 struct ll_sb_info *sbi = ll_s2sbi(sb);
1078 struct obd_statfs obd_osfs;
1082 rc = obd_statfs(class_exp2obd(sbi->ll_mdc_exp), osfs, max_age);
1084 CERROR("mdc_statfs fails: rc = %d\n", rc);
1088 osfs->os_type = sb->s_magic;
1090 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1091 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1093 rc = obd_statfs(class_exp2obd(sbi->ll_osc_exp), &obd_osfs, max_age);
1095 CERROR("obd_statfs fails: rc = %d\n", rc);
1099 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1100 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1103 osfs->os_blocks = obd_osfs.os_blocks;
1104 osfs->os_bfree = obd_osfs.os_bfree;
1105 osfs->os_bavail = obd_osfs.os_bavail;
1107 /* If we don't have as many objects free on the OST as inodes
1108 * on the MDS, we reduce the total number of inodes to
1109 * compensate, so that the "inodes in use" number is correct.
1111 if (obd_osfs.os_ffree < osfs->os_ffree) {
1112 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1114 osfs->os_ffree = obd_osfs.os_ffree;
1120 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1122 struct obd_statfs osfs;
1125 CDEBUG(D_VFSTRACE, "VFS Op:\n");
1126 lprocfs_counter_incr(ll_s2sbi(sb)->ll_stats, LPROC_LL_STAFS);
1128 /* For now we will always get up-to-date statfs values, but in the
1129 * future we may allow some amount of caching on the client (e.g.
1130 * from QOS or lprocfs updates). */
1131 rc = ll_statfs_internal(sb, &osfs, jiffies - 1);
1135 statfs_unpack(sfs, &osfs);
1137 if (sizeof(sfs->f_blocks) == 4) {
1138 while (osfs.os_blocks > ~0UL) {
1141 osfs.os_blocks >>= 1;
1142 osfs.os_bfree >>= 1;
1143 osfs.os_bavail >>= 1;
1147 sfs->f_blocks = osfs.os_blocks;
1148 sfs->f_bfree = osfs.os_bfree;
1149 sfs->f_bavail = osfs.os_bavail;
1154 void ll_update_inode(struct inode *inode, struct mds_body *body,
1155 struct lov_stripe_md *lsm)
1157 struct ll_inode_info *lli = ll_i2info(inode);
1159 LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1161 if (lli->lli_smd == NULL) {
1163 lli->lli_maxbytes = lsm->lsm_maxbytes;
1164 if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1165 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1167 if (memcmp(lli->lli_smd, lsm, sizeof(*lsm))) {
1168 CERROR("lsm mismatch for inode %ld\n",
1170 CERROR("lli_smd:\n");
1171 dump_lsm(D_ERROR, lli->lli_smd);
1173 dump_lsm(D_ERROR, lsm);
1177 /* bug 2844 - limit i_blksize for broken user-space apps */
1178 LASSERTF(lsm->lsm_xfersize != 0, "%lu\n", lsm->lsm_xfersize);
1179 inode->i_blksize = min(lsm->lsm_xfersize, LL_MAX_BLKSIZE);
1180 if (lli->lli_smd != lsm)
1181 obd_free_memmd(ll_i2obdexp(inode), &lsm);
1184 if (body->valid & OBD_MD_FLID)
1185 inode->i_ino = body->ino;
1186 if (body->valid & OBD_MD_FLATIME)
1187 LTIME_S(inode->i_atime) = body->atime;
1188 if (body->valid & OBD_MD_FLMTIME &&
1189 body->mtime > LTIME_S(inode->i_mtime)) {
1190 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to %u\n",
1191 inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
1192 LTIME_S(inode->i_mtime) = body->mtime;
1194 if (body->valid & OBD_MD_FLCTIME &&
1195 body->ctime > LTIME_S(inode->i_ctime))
1196 LTIME_S(inode->i_ctime) = body->ctime;
1197 if (body->valid & OBD_MD_FLMODE)
1198 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1199 if (body->valid & OBD_MD_FLTYPE)
1200 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1201 if (body->valid & OBD_MD_FLUID)
1202 inode->i_uid = body->uid;
1203 if (body->valid & OBD_MD_FLGID)
1204 inode->i_gid = body->gid;
1205 if (body->valid & OBD_MD_FLFLAGS)
1206 inode->i_flags = body->flags;
1207 if (body->valid & OBD_MD_FLNLINK)
1208 inode->i_nlink = body->nlink;
1209 if (body->valid & OBD_MD_FLGENER)
1210 inode->i_generation = body->generation;
1211 if (body->valid & OBD_MD_FLRDEV)
1212 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1213 inode->i_rdev = body->rdev;
1215 inode->i_rdev = old_decode_dev(body->rdev);
1217 if (body->valid & OBD_MD_FLSIZE)
1218 inode->i_size = body->size;
1219 if (body->valid & OBD_MD_FLBLOCKS)
1220 inode->i_blocks = body->blocks;
1222 if (body->valid & OBD_MD_FLSIZE)
1223 set_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &lli->lli_flags);
1226 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
1227 static struct backing_dev_info ll_backing_dev_info = {
1228 .ra_pages = 0, /* No readahead */
1229 .memory_backed = 0, /* Does contribute to dirty memory */
1233 void ll_read_inode2(struct inode *inode, void *opaque)
1235 struct lustre_md *md = opaque;
1236 struct ll_inode_info *lli = ll_i2info(inode);
1239 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1240 inode->i_generation, inode);
1244 LASSERT(!lli->lli_smd);
1246 /* Core attributes from the MDS first. This is a new inode, and
1247 * the VFS doesn't zero times in the core inode so we have to do
1248 * it ourselves. They will be overwritten by either MDS or OST
1249 * attributes - we just need to make sure they aren't newer. */
1250 LTIME_S(inode->i_mtime) = 0;
1251 LTIME_S(inode->i_atime) = 0;
1252 LTIME_S(inode->i_ctime) = 0;
1254 ll_update_inode(inode, md->body, md->lsm);
1256 /* OIDEBUG(inode); */
1258 if (S_ISREG(inode->i_mode)) {
1259 inode->i_op = &ll_file_inode_operations;
1260 inode->i_fop = &ll_file_operations;
1261 inode->i_mapping->a_ops = &ll_aops;
1263 } else if (S_ISDIR(inode->i_mode)) {
1264 inode->i_op = &ll_dir_inode_operations;
1265 inode->i_fop = &ll_dir_operations;
1266 inode->i_mapping->a_ops = &ll_dir_aops;
1268 } else if (S_ISLNK(inode->i_mode)) {
1269 inode->i_op = &ll_fast_symlink_inode_operations;
1272 inode->i_op = &ll_special_inode_operations;
1274 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
1275 init_special_inode(inode, inode->i_mode,
1276 kdev_t_to_nr(inode->i_rdev));
1278 /* initializing backing dev info. */
1279 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
1281 init_special_inode(inode, inode->i_mode, inode->i_rdev);
1283 lli->ll_save_ifop = inode->i_fop;
1285 if (S_ISCHR(inode->i_mode))
1286 inode->i_fop = &ll_special_chr_inode_fops;
1287 else if (S_ISBLK(inode->i_mode))
1288 inode->i_fop = &ll_special_blk_inode_fops;
1289 else if (S_ISFIFO(inode->i_mode))
1290 inode->i_fop = &ll_special_fifo_inode_fops;
1291 else if (S_ISSOCK(inode->i_mode))
1292 inode->i_fop = &ll_special_sock_inode_fops;
1297 int ll_iocontrol(struct inode *inode, struct file *file,
1298 unsigned int cmd, unsigned long arg)
1300 struct ll_sb_info *sbi = ll_i2sbi(inode);
1301 struct ptlrpc_request *req = NULL;
1306 case EXT3_IOC_GETFLAGS: {
1308 unsigned long valid = OBD_MD_FLFLAGS;
1309 struct mds_body *body;
1311 ll_inode2fid(&fid, inode);
1312 rc = mdc_getattr(sbi->ll_mdc_exp, &fid, valid, 0, &req);
1314 CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1318 body = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*body));
1320 if (body->flags & S_APPEND)
1321 flags |= EXT3_APPEND_FL;
1322 if (body->flags & S_IMMUTABLE)
1323 flags |= EXT3_IMMUTABLE_FL;
1324 if (body->flags & S_NOATIME)
1325 flags |= EXT3_NOATIME_FL;
1327 ptlrpc_req_finished (req);
1329 RETURN(put_user(flags, (int *)arg));
1331 case EXT3_IOC_SETFLAGS: {
1332 struct mdc_op_data op_data;
1335 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1337 if (get_user(flags, (int *)arg))
1344 ll_prepare_mdc_op_data(&op_data, inode, NULL, NULL, 0, 0);
1346 memset(&attr, 0x0, sizeof(attr));
1347 attr.ia_attr_flags = flags;
1348 attr.ia_valid |= ATTR_ATTR_FLAG;
1350 rc = mdc_setattr(sbi->ll_mdc_exp, &op_data,
1351 &attr, NULL, 0, NULL, 0, &req);
1353 ptlrpc_req_finished(req);
1354 if (rc != -EPERM && rc != -EACCES)
1355 CERROR("mdc_setattr fails: rc = %d\n", rc);
1359 ptlrpc_req_finished(req);
1361 oa->o_id = lsm->lsm_object_id;
1362 oa->o_flags = flags;
1363 oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS;
1365 rc = obd_setattr(sbi->ll_osc_exp, oa, lsm, NULL);
1368 if (rc != -EPERM && rc != -EACCES)
1369 CERROR("mdc_setattr fails: rc = %d\n", rc);
1373 if (flags & EXT3_APPEND_FL)
1374 inode->i_flags |= S_APPEND;
1376 inode->i_flags &= ~S_APPEND;
1377 if (flags & EXT3_IMMUTABLE_FL)
1378 inode->i_flags |= S_IMMUTABLE;
1380 inode->i_flags &= ~S_IMMUTABLE;
1381 if (flags & EXT3_NOATIME_FL)
1382 inode->i_flags |= S_NOATIME;
1384 inode->i_flags &= ~S_NOATIME;
1395 void ll_umount_begin(struct super_block *sb)
1397 struct ll_sb_info *sbi = ll_s2sbi(sb);
1398 struct obd_device *obd;
1399 struct obd_ioctl_data ioc_data = { 0 };
1401 CDEBUG(D_VFSTRACE, "VFS Op:\n");
1403 obd = class_exp2obd(sbi->ll_mdc_exp);
1405 CERROR("Invalid MDC connection handle "LPX64"\n",
1406 sbi->ll_mdc_exp->exp_handle.h_cookie);
1410 obd->obd_no_recov = 1;
1411 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_mdc_exp, sizeof ioc_data,
1414 obd = class_exp2obd(sbi->ll_osc_exp);
1416 CERROR("Invalid LOV connection handle "LPX64"\n",
1417 sbi->ll_osc_exp->exp_handle.h_cookie);
1422 obd->obd_no_recov = 1;
1423 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_osc_exp, sizeof ioc_data,
1426 /* Really, we'd like to wait until there are no requests outstanding,
1427 * and then continue. For now, we just invalidate the requests,
1428 * schedule, and hope.
1435 int ll_prep_inode(struct obd_export *exp, struct inode **inode,
1436 struct ptlrpc_request *req, int offset,struct super_block *sb)
1438 struct lustre_md md;
1441 rc = mdc_req2lustre_md(req, offset, exp, &md);
1446 ll_update_inode(*inode, md.body, md.lsm);
1449 *inode = ll_iget(sb, md.body->ino, &md);
1450 if (*inode == NULL || is_bad_inode(*inode)) {
1451 /* free the lsm if we allocated one above */
1453 obd_free_memmd(exp, &md.lsm);
1455 CERROR("new_inode -fatal: rc %d\n", rc);
1462 char *llap_origins[] = {
1463 [LLAP_ORIGIN_UNKNOWN] = "--",
1464 [LLAP_ORIGIN_READPAGE] = "rp",
1465 [LLAP_ORIGIN_READAHEAD] = "ra",
1466 [LLAP_ORIGIN_COMMIT_WRITE] = "cw",
1467 [LLAP_ORIGIN_WRITEPAGE] = "wp",
1470 struct ll_async_page *llite_pglist_next_llap(struct ll_sb_info *sbi,
1471 struct list_head *list)
1473 struct ll_async_page *llap;
1474 struct list_head *pos;
1476 list_for_each(pos, list) {
1477 if (pos == &sbi->ll_pglist)
1479 llap = list_entry(pos, struct ll_async_page, llap_pglist_item);
1480 if (llap->llap_page == NULL)