1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Lustre Light Super operations
6 * Copyright (c) 2002, 2003 Cluster File Systems, Inc.
8 * This file is part of Lustre, http://www.lustre.org.
10 * Lustre is free software; you can redistribute it and/or
11 * modify it under the terms of version 2 of the GNU General Public
12 * License as published by the Free Software Foundation.
14 * Lustre is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
17 * GNU General Public License for more details.
19 * You should have received a copy of the GNU General Public License
20 * along with Lustre; if not, write to the Free Software
21 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
24 #define DEBUG_SUBSYSTEM S_LLITE
26 #include <linux/module.h>
27 #include <linux/random.h>
28 #include <linux/version.h>
29 #include <linux/lustre_lite.h>
30 #include <linux/lustre_ha.h>
31 #include <linux/lustre_dlm.h>
32 #include <linux/init.h>
34 #include <linux/lprocfs_status.h>
36 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
37 kmem_cache_t *ll_file_data_slab;
38 extern struct address_space_operations ll_aops;
39 extern struct address_space_operations ll_dir_aops;
40 struct super_operations ll_super_operations;
42 /* /proc/lustre/llite root that tracks llite mount points */
43 struct proc_dir_entry *proc_lustre_fs_root;
45 extern int lprocfs_register_mountpoint(struct proc_dir_entry *parent,
46 struct super_block *sb,
47 char *osc, char *mdc);
49 extern int ll_init_inodecache(void);
50 extern void ll_destroy_inodecache(void);
51 extern int ll_recover(struct recovd_data *, int);
52 extern int ll_commitcbd_setup(struct ll_sb_info *);
53 extern int ll_commitcbd_cleanup(struct ll_sb_info *);
54 int ll_read_inode2(struct inode *inode, void *opaque);
56 extern int ll_proc_namespace(struct super_block* sb, char* osc, char* mdc)
58 static char *ll_read_opt(const char *opt, char *data)
64 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
65 if ( strncmp(opt, data, strlen(opt)) )
67 if ( (value = strchr(data, '=')) == NULL )
71 OBD_ALLOC(retval, strlen(value) + 1);
73 CERROR("out of memory!\n");
77 memcpy(retval, value, strlen(value)+1);
78 CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
82 static int ll_set_opt(const char *opt, char *data, int fl)
86 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
87 if ( strncmp(opt, data, strlen(opt)) )
93 static void ll_options(char *options, char **ost, char **mds, int *flags)
95 char *opt_ptr = options;
104 while ((this_char = strsep (&opt_ptr, ",")) != NULL) {
105 CDEBUG(D_SUPER, "this_char %s\n", this_char);
106 if ( (!*ost && (*ost = ll_read_opt("osc", this_char)))||
107 (!*mds && (*mds = ll_read_opt("mdc", this_char)))||
108 (!(*flags & LL_SBI_NOLCK) && ((*flags) = (*flags) |
109 ll_set_opt("nolock", this_char, LL_SBI_NOLCK))) )
116 #define log2(n) ffz(~(n))
120 static int ll_fill_super(struct super_block *sb, void *data, int silent)
122 struct inode *root = 0;
123 struct obd_device *obd;
124 struct ll_sb_info *sbi;
128 struct ll_fid rootfid;
129 struct obd_statfs osfs;
130 struct ptlrpc_request *request = NULL;
131 struct ptlrpc_connection *mdc_conn;
132 struct ll_read_inode2_cookie lic;
137 OBD_ALLOC(sbi, sizeof(*sbi));
141 INIT_LIST_HEAD(&sbi->ll_conn_chain);
142 generate_random_uuid(uuid);
143 class_uuid_unparse(uuid, sbi->ll_sb_uuid);
147 ll_options(data, &osc, &mdc, &sbi->ll_flags);
151 GOTO(out_free, sb = NULL);
156 GOTO(out_free, sb = NULL);
159 obd = class_uuid2obd(mdc);
161 CERROR("MDC %s: not setup or attached\n", mdc);
162 GOTO(out_free, sb = NULL);
165 err = obd_connect(&sbi->ll_mdc_conn, obd, sbi->ll_sb_uuid,
166 ptlrpc_recovd, ll_recover);
168 CERROR("cannot connect to %s: rc = %d\n", mdc, err);
169 GOTO(out_free, sb = NULL);
172 mdc_conn = sbi2mdc(sbi)->cl_import.imp_connection;
173 list_add(&mdc_conn->c_sb_chain, &sbi->ll_conn_chain);
175 obd = class_uuid2obd(osc);
177 CERROR("OSC %s: not setup or attached\n", osc);
178 GOTO(out_mdc, sb = NULL);
181 err = obd_connect(&sbi->ll_osc_conn, obd, sbi->ll_sb_uuid,
182 ptlrpc_recovd, ll_recover);
184 CERROR("cannot connect to %s: rc = %d\n", osc, err);
185 GOTO(out_mdc, sb = NULL);
188 err = mdc_getstatus(&sbi->ll_mdc_conn, &rootfid);
190 CERROR("cannot mds_connect: rc = %d\n", err);
191 GOTO(out_mdc, sb = NULL);
193 CDEBUG(D_SUPER, "rootfid "LPU64"\n", rootfid.id);
194 sbi->ll_rootino = rootfid.id;
196 memset(&osfs, 0, sizeof(osfs));
197 err = mdc_statfs(&sbi->ll_mdc_conn, &osfs);
198 sb->s_blocksize = osfs.os_bsize;
199 sb->s_blocksize_bits = log2(osfs.os_bsize);
200 sb->s_magic = LL_SUPER_MAGIC;
201 sb->s_maxbytes = (1ULL << (32 + 9)) - osfs.os_bsize;
203 sb->s_op = &ll_super_operations;
205 /* make root inode */
206 err = mdc_getattr(&sbi->ll_mdc_conn, sbi->ll_rootino, S_IFDIR,
207 OBD_MD_FLNOTOBD|OBD_MD_FLBLOCKS, 0, &request);
209 CERROR("mdc_getattr failed for root: rc = %d\n", err);
210 GOTO(out_request, sb = NULL);
213 /* initialize committed transaction callback daemon */
214 spin_lock_init(&sbi->ll_commitcbd_lock);
215 init_waitqueue_head(&sbi->ll_commitcbd_waitq);
216 init_waitqueue_head(&sbi->ll_commitcbd_ctl_waitq);
217 sbi->ll_commitcbd_flags = 0;
218 err = ll_commitcbd_setup(sbi);
220 CERROR("failed to start commit callback daemon: rc = %d\n",err);
221 GOTO(out_request, sb = NULL);
224 lic.lic_body = lustre_msg_buf(request->rq_repmsg, 0);
226 root = iget5_locked(sb, sbi->ll_rootino, NULL,
227 ll_read_inode2, &lic);
230 sb->s_root = d_alloc_root(root);
232 CERROR("lustre_lite: bad iget4 for root\n");
233 GOTO(out_cdb, sb = NULL);
236 ptlrpc_req_finished(request);
239 if (proc_lustre_fs_root) {
240 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
243 CERROR("could not register mount in /proc/lustre");
248 OBD_FREE(mdc, strlen(mdc) + 1);
250 OBD_FREE(osc, strlen(osc) + 1);
255 ll_commitcbd_cleanup(sbi);
257 ptlrpc_req_finished(request);
258 obd_disconnect(&sbi->ll_osc_conn);
260 obd_disconnect(&sbi->ll_mdc_conn);
262 OBD_FREE(sbi, sizeof(*sbi));
265 } /* ll_fill_super */
267 struct super_block * ll_get_sb(struct file_system_type *fs_type,
268 int flags, char *devname, void * data)
270 return get_sb_nodev(fs_type, flags, data, ll_fill_super);
273 static void ll_put_super(struct super_block *sb)
275 struct ll_sb_info *sbi = ll_s2sbi(sb);
276 struct ll_fid rootfid;
279 list_del(&sbi->ll_conn_chain);
280 ll_commitcbd_cleanup(sbi);
281 obd_disconnect(&sbi->ll_osc_conn);
283 /* NULL request to force sync on the MDS, and get the last_committed
284 * value to flush remaining RPCs from the pending queue on client.
286 * XXX This should be an mdc_sync() call to sync the whole MDS fs,
287 * which we can call for other reasons as well.
289 mdc_getstatus(&sbi->ll_mdc_conn, &rootfid);
291 if (sbi->ll_proc_root) {
292 lprocfs_remove(sbi->ll_proc_root);
293 sbi->ll_proc_root = NULL;
296 obd_disconnect(&sbi->ll_mdc_conn);
297 OBD_FREE(sbi, sizeof(*sbi));
302 static void ll_clear_inode(struct inode *inode)
304 struct ll_sb_info *sbi = ll_i2sbi(inode);
305 struct ll_inode_info *lli = ll_i2info(inode);
309 #warning "Is there a reason we don't do this in 2.5, but we do in 2.4?"
311 rc = mdc_cancel_unused(&sbi->ll_mdc_conn, inode, LDLM_FL_NO_CALLBACK);
313 CERROR("mdc_cancel_unused: %d\n", rc);
314 /* XXX FIXME do something dramatic */
318 rc = obd_cancel_unused(&sbi->ll_osc_conn, lli->lli_smd, 0);
320 CERROR("obd_cancel_unused: %d\n", rc);
321 /* XXX FIXME do something dramatic */
326 if (atomic_read(&inode->i_count) != 0)
327 CERROR("clearing in-use inode %lu: count = %d\n",
328 inode->i_ino, atomic_read(&inode->i_count));
331 obd_free_memmd(&sbi->ll_osc_conn, &lli->lli_smd);
333 if (lli->lli_symlink_name) {
334 OBD_FREE(lli->lli_symlink_name,strlen(lli->lli_symlink_name)+1);
335 lli->lli_symlink_name = NULL;
341 static void ll_delete_inode(struct inode *inode)
344 if (S_ISREG(inode->i_mode)) {
347 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
349 /* mcreate with no open */
353 if (lsm->lsm_object_id == 0) {
354 CERROR("This really happens\n");
355 /* No obdo was ever created */
363 oa->o_id = lsm->lsm_object_id;
364 oa->o_mode = inode->i_mode;
365 oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE;
367 err = obd_destroy(ll_i2obdconn(inode), oa, lsm);
370 CDEBUG(D_SUPER, "obd destroy objid "LPX64" error %d\n",
371 lsm->lsm_object_id, err);
378 /* like inode_setattr, but doesn't mark the inode dirty */
379 static int ll_attr2inode(struct inode * inode, struct iattr * attr, int trunc)
381 unsigned int ia_valid = attr->ia_valid;
384 if ((ia_valid & ATTR_SIZE) && trunc) {
385 error = vmtruncate(inode, attr->ia_size);
388 } else if (ia_valid & ATTR_SIZE)
389 inode->i_size = attr->ia_size;
391 if (ia_valid & ATTR_UID)
392 inode->i_uid = attr->ia_uid;
393 if (ia_valid & ATTR_GID)
394 inode->i_gid = attr->ia_gid;
395 if (ia_valid & ATTR_ATIME)
396 inode->i_atime = attr->ia_atime;
397 if (ia_valid & ATTR_MTIME)
398 inode->i_mtime = attr->ia_mtime;
399 if (ia_valid & ATTR_CTIME)
400 inode->i_ctime = attr->ia_ctime;
401 if (ia_valid & ATTR_MODE) {
402 inode->i_mode = attr->ia_mode;
403 if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID))
404 inode->i_mode &= ~S_ISGID;
410 int ll_inode_setattr(struct inode *inode, struct iattr *attr, int do_trunc)
412 struct ptlrpc_request *request = NULL;
413 struct ll_sb_info *sbi = ll_i2sbi(inode);
418 /* change incore inode */
419 ll_attr2inode(inode, attr, do_trunc);
421 /* Don't send size changes to MDS to avoid "fast EA" problems, and
422 * also avoid a pointless RPC (we get file size from OST anyways).
424 attr->ia_valid &= ~ATTR_SIZE;
425 if (attr->ia_valid) {
426 err = mdc_setattr(&sbi->ll_mdc_conn, inode, attr, &request);
428 CERROR("mdc_setattr fails: err = %d\n", err);
430 ptlrpc_req_finished(request);
431 if (S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_MTIME_SET) {
432 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
436 CDEBUG(D_ERROR, "setting mtime on OST\n");
437 oa.o_id = lsm->lsm_object_id;
439 oa.o_valid = OBD_MD_FLID |OBD_MD_FLTYPE |OBD_MD_FLMTIME;
440 oa.o_mtime = attr->ia_mtime;
441 err = obd_setattr(&sbi->ll_osc_conn, &oa, lsm);
443 CERROR("obd_setattr fails: rc=%d\n", err);
453 int ll_setattr(struct dentry *de, struct iattr *attr)
455 int rc = inode_change_ok(de->d_inode, attr);
460 return ll_inode_setattr(de->d_inode, attr, 1);
463 static int ll_statfs(struct super_block *sb, struct statfs *sfs)
465 struct ll_sb_info *sbi = ll_s2sbi(sb);
466 struct obd_statfs osfs;
470 memset(sfs, 0, sizeof(*sfs));
471 rc = obd_statfs(&sbi->ll_mdc_conn, &osfs);
472 statfs_unpack(sfs, &osfs);
474 CERROR("mdc_statfs fails: rc = %d\n", rc);
476 CDEBUG(D_SUPER, "mdc_statfs shows blocks "LPU64"/"LPU64
477 " objects "LPU64"/"LPU64"\n",
478 osfs.os_bavail, osfs.os_blocks,
479 osfs.os_ffree, osfs.os_files);
481 /* temporary until mds_statfs returns statfs info for all OSTs */
483 rc = obd_statfs(&sbi->ll_osc_conn, &osfs);
485 CERROR("obd_statfs fails: rc = %d\n", rc);
488 CDEBUG(D_SUPER, "obd_statfs shows blocks "LPU64"/"LPU64
489 " objects "LPU64"/"LPU64"\n",
490 osfs.os_bavail, osfs.os_blocks,
491 osfs.os_ffree, osfs.os_files);
493 while (osfs.os_blocks > ~0UL) {
496 osfs.os_blocks >>= 1;
498 osfs.os_bavail >>= 1;
500 sfs->f_blocks = osfs.os_blocks;
501 sfs->f_bfree = osfs.os_bfree;
502 sfs->f_bavail = osfs.os_bavail;
503 if (osfs.os_ffree < (__u64)sfs->f_ffree)
504 sfs->f_ffree = osfs.os_ffree;
511 void ll_update_inode(struct inode *inode, struct mds_body *body,
512 struct lov_mds_md *lmm)
514 struct ll_inode_info *lli = ll_i2info(inode);
517 obd_unpackmd(ll_i2obdconn(inode), &lli->lli_smd, lmm);
519 if (body->valid & OBD_MD_FLID)
520 inode->i_ino = body->ino;
521 if (body->valid & OBD_MD_FLATIME)
522 inode->i_atime = body->atime;
523 if (body->valid & OBD_MD_FLMTIME)
524 inode->i_mtime = body->mtime;
525 if (body->valid & OBD_MD_FLCTIME)
526 inode->i_ctime = body->ctime;
527 if (body->valid & OBD_MD_FLMODE)
528 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
529 if (body->valid & OBD_MD_FLTYPE)
530 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
531 if (body->valid & OBD_MD_FLUID)
532 inode->i_uid = body->uid;
533 if (body->valid & OBD_MD_FLGID)
534 inode->i_gid = body->gid;
535 if (body->valid & OBD_MD_FLFLAGS)
536 inode->i_flags = body->flags;
537 if (body->valid & OBD_MD_FLNLINK)
538 inode->i_nlink = body->nlink;
539 if (body->valid & OBD_MD_FLGENER)
540 inode->i_generation = body->generation;
541 if (body->valid & OBD_MD_FLRDEV)
542 inode->i_rdev = to_kdev_t(body->rdev);
543 if (body->valid & OBD_MD_FLSIZE)
544 inode->i_size = body->size;
545 if (body->valid & OBD_MD_FLBLOCKS)
546 inode->i_blocks = body->blocks;
549 int ll_read_inode2(struct inode *inode, void *opaque)
551 struct ll_read_inode2_cookie *lic = opaque;
552 struct mds_body *body = lic->lic_body;
553 struct ll_inode_info *lli = ll_i2info(inode);
557 sema_init(&lli->lli_open_sem, 1);
559 LASSERT(!lli->lli_smd);
561 /* core attributes first */
562 ll_update_inode(inode, body, lic ? lic->lic_lmm : NULL);
564 /* Get the authoritative file size */
565 if (lli->lli_smd && S_ISREG(inode->i_mode)) {
566 rc = ll_file_size(inode, lli->lli_smd, NULL);
568 CERROR("ll_file_size: %d\n", rc);
569 ll_clear_inode(inode);
570 make_bad_inode(inode);
575 /* OIDEBUG(inode); */
577 if (S_ISREG(inode->i_mode)) {
578 inode->i_op = &ll_file_inode_operations;
579 inode->i_fop = &ll_file_operations;
580 inode->i_mapping->a_ops = &ll_aops;
582 } else if (S_ISDIR(inode->i_mode)) {
583 inode->i_op = &ll_dir_inode_operations;
584 inode->i_fop = &ll_dir_operations;
585 inode->i_mapping->a_ops = &ll_dir_aops;
587 } else if (S_ISLNK(inode->i_mode)) {
588 inode->i_op = &ll_fast_symlink_inode_operations;
591 init_special_inode(inode, inode->i_mode,
592 kdev_t_to_nr(inode->i_rdev));
599 static inline void invalidate_request_list(struct list_head *req_list)
601 struct list_head *tmp, *n;
602 list_for_each_safe(tmp, n, req_list) {
603 struct ptlrpc_request *req =
604 list_entry(tmp, struct ptlrpc_request, rq_list);
605 CERROR("invalidating req xid %d op %d to %s:%d\n",
606 (unsigned long long)req->rq_xid, req->rq_reqmsg->opc,
607 req->rq_connection->c_remote_uuid,
608 req->rq_import->imp_client->cli_request_portal);
609 req->rq_flags |= PTL_RPC_FL_ERR;
610 wake_up(&req->rq_wait_for_rep);
614 void ll_umount_begin(struct super_block *sb)
616 struct ll_sb_info *sbi = ll_s2sbi(sb);
617 struct list_head *ctmp;
621 list_for_each(ctmp, &sbi->ll_conn_chain) {
622 struct ptlrpc_connection *conn;
623 conn = list_entry(ctmp, struct ptlrpc_connection, c_sb_chain);
625 spin_lock(&conn->c_lock);
626 conn->c_flags |= CONN_INVALID;
627 invalidate_request_list(&conn->c_sending_head);
628 invalidate_request_list(&conn->c_delayed_head);
629 spin_unlock(&conn->c_lock);
636 static kmem_cache_t *ll_inode_cachep;
638 static struct inode *ll_alloc_inode(struct super_block *sb)
640 struct ll_inode_info *lli;
641 lli = kmem_cache_alloc(ll_inode_cachep, SLAB_KERNEL);
645 memset(lli, 0, (char *)&lli->lli_vfs_inode - (char *)lli);
646 sema_init(&lli->lli_open_sem, 1);
648 return &lli->lli_vfs_inode;
651 static void ll_destroy_inode(struct inode *inode)
653 kmem_cache_free(ll_inode_cachep, ll_i2info(inode));
656 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
658 struct ll_inode_info *lli = foo;
660 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
661 SLAB_CTOR_CONSTRUCTOR)
662 inode_init_once(&lli->lli_vfs_inode);
665 int ll_init_inodecache(void)
667 ll_inode_cachep = kmem_cache_create("lustre_inode_cache",
668 sizeof(struct ll_inode_info),
669 0, SLAB_HWCACHE_ALIGN,
671 if (ll_inode_cachep == NULL)
676 void ll_destroy_inodecache(void)
678 if (kmem_cache_destroy(ll_inode_cachep))
679 CERROR("ll_inode_cache: not all structures were freed\n");
684 /* exported operations */
685 struct super_operations ll_super_operations =
687 alloc_inode: ll_alloc_inode,
688 destroy_inode: ll_destroy_inode,
689 clear_inode: ll_clear_inode,
690 delete_inode: ll_delete_inode,
691 put_super: ll_put_super,
693 umount_begin: ll_umount_begin
696 struct file_system_type lustre_lite_fs_type = {
697 .owner = THIS_MODULE,
698 .name = "lustre_lite",
700 .kill_sb = kill_litter_super,
703 static int __init init_lustre_lite(void)
706 printk(KERN_INFO "Lustre Lite Client File System; "
707 "info@clusterfs.com\n");
708 rc = ll_init_inodecache();
711 ll_file_data_slab = kmem_cache_create("ll_file_data",
712 sizeof(struct ll_file_data), 0,
713 SLAB_HWCACHE_ALIGN, NULL, NULL);
714 if (ll_file_data_slab == NULL) {
715 ll_destroy_inodecache();
719 proc_lustre_fs_root = proc_lustre_root ?
720 proc_mkdir("llite", proc_lustre_root) : NULL;
722 return register_filesystem(&lustre_lite_fs_type);
725 static void __exit exit_lustre_lite(void)
727 unregister_filesystem(&lustre_lite_fs_type);
728 ll_destroy_inodecache();
729 kmem_cache_destroy(ll_file_data_slab);
730 if (proc_lustre_fs_root) {
731 lprocfs_remove(proc_lustre_fs_root);
732 proc_lustre_fs_root = NULL;
736 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
737 MODULE_DESCRIPTION("Lustre Lite Client File System");
738 MODULE_LICENSE("GPL");
740 module_init(init_lustre_lite);
741 module_exit(exit_lustre_lite);