1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Lustre Light Super operations
6 * This code is issued under the GNU General Public License.
7 * See the file COPYING in this distribution
9 * Copryright (C) 2002 Cluster File Systems, Inc.
12 #define DEBUG_SUBSYSTEM S_LLITE
14 #include <linux/module.h>
15 #include <linux/random.h>
16 #include <linux/version.h>
17 #include <linux/lustre_lite.h>
18 #include <linux/lustre_ha.h>
19 #include <linux/lustre_dlm.h>
20 #include <linux/init.h>
22 #include <linux/lprocfs_status.h>
24 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
25 kmem_cache_t *ll_file_data_slab;
26 extern struct address_space_operations ll_aops;
27 extern struct address_space_operations ll_dir_aops;
28 struct super_operations ll_super_operations;
30 extern int ll_init_inodecache(void);
31 extern void ll_destroy_inodecache(void);
32 extern int ll_recover(struct recovd_data *, int);
33 extern int ll_commitcbd_setup(struct ll_sb_info *);
34 extern int ll_commitcbd_cleanup(struct ll_sb_info *);
35 int ll_read_inode2(struct inode *inode, void *opaque);
37 extern void ll_proc_namespace(struct super_block* sb, char* osc, char* mdc)
39 static char *ll_read_opt(const char *opt, char *data)
45 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
46 if ( strncmp(opt, data, strlen(opt)) )
48 if ( (value = strchr(data, '=')) == NULL )
52 OBD_ALLOC(retval, strlen(value) + 1);
54 CERROR("out of memory!\n");
58 memcpy(retval, value, strlen(value)+1);
59 CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
63 static int ll_set_opt(const char *opt, char *data, int fl)
67 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
68 if ( strncmp(opt, data, strlen(opt)) )
74 static void ll_options(char *options, char **ost, char **mds, int *flags)
76 char *opt_ptr = options;
85 while ((this_char = strsep (&opt_ptr, ",")) != NULL) {
86 CDEBUG(D_SUPER, "this_char %s\n", this_char);
87 if ( (!*ost && (*ost = ll_read_opt("osc", this_char)))||
88 (!*mds && (*mds = ll_read_opt("mdc", this_char)))||
89 (!(*flags & LL_SBI_NOLCK) && ((*flags) = (*flags) |
90 ll_set_opt("nolock", this_char, LL_SBI_NOLCK))) )
97 #define log2(n) ffz(~(n))
101 static int ll_fill_super(struct super_block *sb, void *data, int silent)
103 struct inode *root = 0;
104 struct obd_device *obd;
105 struct ll_sb_info *sbi;
109 struct ll_fid rootfid;
110 struct obd_statfs osfs;
111 struct ptlrpc_request *request = NULL;
112 struct ptlrpc_connection *mdc_conn;
113 struct ll_read_inode2_cookie lic;
118 OBD_ALLOC(sbi, sizeof(*sbi));
122 INIT_LIST_HEAD(&sbi->ll_conn_chain);
123 generate_random_uuid(uuid);
124 class_uuid_unparse(uuid, sbi->ll_sb_uuid);
128 ll_options(data, &osc, &mdc, &sbi->ll_flags);
132 GOTO(out_free, sb = NULL);
137 GOTO(out_free, sb = NULL);
140 obd = class_uuid2obd(mdc);
142 CERROR("MDC %s: not setup or attached\n", mdc);
143 GOTO(out_free, sb = NULL);
146 err = obd_connect(&sbi->ll_mdc_conn, obd, sbi->ll_sb_uuid,
147 ptlrpc_recovd, ll_recover);
149 CERROR("cannot connect to %s: rc = %d\n", mdc, err);
150 GOTO(out_free, sb = NULL);
153 mdc_conn = sbi2mdc(sbi)->cl_import.imp_connection;
154 list_add(&mdc_conn->c_sb_chain, &sbi->ll_conn_chain);
156 obd = class_uuid2obd(osc);
158 CERROR("OSC %s: not setup or attached\n", osc);
159 GOTO(out_mdc, sb = NULL);
162 err = obd_connect(&sbi->ll_osc_conn, obd, sbi->ll_sb_uuid,
163 ptlrpc_recovd, ll_recover);
165 CERROR("cannot connect to %s: rc = %d\n", osc, err);
166 GOTO(out_mdc, sb = NULL);
169 err = mdc_getstatus(&sbi->ll_mdc_conn, &rootfid);
171 CERROR("cannot mds_connect: rc = %d\n", err);
172 GOTO(out_mdc, sb = NULL);
174 CDEBUG(D_SUPER, "rootfid "LPU64"\n", rootfid.id);
175 sbi->ll_rootino = rootfid.id;
177 memset(&osfs, 0, sizeof(osfs));
178 err = mdc_statfs(&sbi->ll_mdc_conn, &osfs);
179 sb->s_blocksize = osfs.os_bsize;
180 sb->s_blocksize_bits = log2(osfs.os_bsize);
181 sb->s_magic = LL_SUPER_MAGIC;
182 sb->s_maxbytes = (1ULL << (32 + 9)) - osfs.os_bsize;
184 sb->s_op = &ll_super_operations;
186 /* make root inode */
187 err = mdc_getattr(&sbi->ll_mdc_conn, sbi->ll_rootino, S_IFDIR,
188 OBD_MD_FLNOTOBD|OBD_MD_FLBLOCKS, 0, &request);
190 CERROR("mdc_getattr failed for root: rc = %d\n", err);
191 GOTO(out_request, sb = NULL);
194 /* initialize committed transaction callback daemon */
195 spin_lock_init(&sbi->ll_commitcbd_lock);
196 init_waitqueue_head(&sbi->ll_commitcbd_waitq);
197 init_waitqueue_head(&sbi->ll_commitcbd_ctl_waitq);
198 sbi->ll_commitcbd_flags = 0;
199 err = ll_commitcbd_setup(sbi);
201 CERROR("failed to start commit callback daemon: rc = %d\n",err);
202 GOTO(out_request, sb = NULL);
205 lic.lic_body = lustre_msg_buf(request->rq_repmsg, 0);
207 root = iget5_locked(sb, sbi->ll_rootino, NULL,
208 ll_read_inode2, &lic);
211 sb->s_root = d_alloc_root(root);
213 CERROR("lustre_lite: bad iget4 for root\n");
214 GOTO(out_cdb, sb = NULL);
217 ptlrpc_req_finished(request);
219 ll_proc_namespace(sb, osc, mdc)
222 OBD_FREE(mdc, strlen(mdc) + 1);
224 OBD_FREE(osc, strlen(osc) + 1);
229 ll_commitcbd_cleanup(sbi);
231 ptlrpc_req_finished(request);
232 obd_disconnect(&sbi->ll_osc_conn);
234 obd_disconnect(&sbi->ll_mdc_conn);
236 OBD_FREE(sbi, sizeof(*sbi));
239 } /* ll_fill_super */
241 struct super_block * ll_get_sb(struct file_system_type *fs_type,
242 int flags, char *devname, void * data)
244 return get_sb_nodev(fs_type, flags, data, ll_fill_super);
247 static void ll_put_super(struct super_block *sb)
249 struct ll_sb_info *sbi = ll_s2sbi(sb);
250 struct ll_fid rootfid;
253 list_del(&sbi->ll_conn_chain);
254 ll_commitcbd_cleanup(sbi);
255 obd_disconnect(&sbi->ll_osc_conn);
257 /* NULL request to force sync on the MDS, and get the last_committed
258 * value to flush remaining RPCs from the pending queue on client.
260 * XXX This should be an mdc_sync() call to sync the whole MDS fs,
261 * which we can call for other reasons as well.
263 mdc_getstatus(&sbi->ll_mdc_conn, &rootfid);
265 lprocfs_dereg_mnt(sbi->ll_proc_root);
266 sbi->ll_proc_root = NULL;
268 obd_disconnect(&sbi->ll_mdc_conn);
269 OBD_FREE(sbi, sizeof(*sbi));
274 static void ll_clear_inode(struct inode *inode)
276 struct ll_sb_info *sbi = ll_i2sbi(inode);
277 struct ll_inode_info *lli = ll_i2info(inode);
281 #warning "Is there a reason we don't do this in 2.5, but we do in 2.4?"
283 rc = mdc_cancel_unused(&sbi->ll_mdc_conn, inode, LDLM_FL_NO_CALLBACK);
285 CERROR("mdc_cancel_unused: %d\n", rc);
286 /* XXX FIXME do something dramatic */
290 rc = obd_cancel_unused(&sbi->ll_osc_conn, lli->lli_smd, 0);
292 CERROR("obd_cancel_unused: %d\n", rc);
293 /* XXX FIXME do something dramatic */
298 if (atomic_read(&inode->i_count) != 0)
299 CERROR("clearing in-use inode %lu: count = %d\n",
300 inode->i_ino, atomic_read(&inode->i_count));
303 obd_free_memmd(&sbi->ll_osc_conn, &lli->lli_smd);
305 if (lli->lli_symlink_name) {
306 OBD_FREE(lli->lli_symlink_name,strlen(lli->lli_symlink_name)+1);
307 lli->lli_symlink_name = NULL;
313 static void ll_delete_inode(struct inode *inode)
316 if (S_ISREG(inode->i_mode)) {
319 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
321 /* mcreate with no open */
325 if (lsm->lsm_object_id == 0) {
326 CERROR("This really happens\n");
327 /* No obdo was ever created */
335 oa->o_id = lsm->lsm_object_id;
336 oa->o_mode = inode->i_mode;
337 oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE;
339 err = obd_destroy(ll_i2obdconn(inode), oa, lsm);
342 CDEBUG(D_SUPER, "obd destroy objid "LPX64" error %d\n",
343 lsm->lsm_object_id, err);
350 /* like inode_setattr, but doesn't mark the inode dirty */
351 static int ll_attr2inode(struct inode * inode, struct iattr * attr, int trunc)
353 unsigned int ia_valid = attr->ia_valid;
356 if ((ia_valid & ATTR_SIZE) && trunc) {
357 error = vmtruncate(inode, attr->ia_size);
360 } else if (ia_valid & ATTR_SIZE)
361 inode->i_size = attr->ia_size;
363 if (ia_valid & ATTR_UID)
364 inode->i_uid = attr->ia_uid;
365 if (ia_valid & ATTR_GID)
366 inode->i_gid = attr->ia_gid;
367 if (ia_valid & ATTR_ATIME)
368 inode->i_atime = attr->ia_atime;
369 if (ia_valid & ATTR_MTIME)
370 inode->i_mtime = attr->ia_mtime;
371 if (ia_valid & ATTR_CTIME)
372 inode->i_ctime = attr->ia_ctime;
373 if (ia_valid & ATTR_MODE) {
374 inode->i_mode = attr->ia_mode;
375 if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID))
376 inode->i_mode &= ~S_ISGID;
382 int ll_inode_setattr(struct inode *inode, struct iattr *attr, int do_trunc)
384 struct ptlrpc_request *request = NULL;
385 struct ll_sb_info *sbi = ll_i2sbi(inode);
390 /* change incore inode */
391 ll_attr2inode(inode, attr, do_trunc);
393 /* Don't send size changes to MDS to avoid "fast EA" problems, and
394 * also avoid a pointless RPC (we get file size from OST anyways).
396 attr->ia_valid &= ~ATTR_SIZE;
397 if (attr->ia_valid) {
398 err = mdc_setattr(&sbi->ll_mdc_conn, inode, attr, &request);
400 CERROR("mdc_setattr fails (%d)\n", err);
402 ptlrpc_req_finished(request);
408 int ll_setattr(struct dentry *de, struct iattr *attr)
410 int rc = inode_change_ok(de->d_inode, attr);
415 return ll_inode_setattr(de->d_inode, attr, 1);
418 static int ll_statfs(struct super_block *sb, struct statfs *sfs)
420 struct ll_sb_info *sbi = ll_s2sbi(sb);
421 struct obd_statfs osfs;
425 memset(sfs, 0, sizeof(*sfs));
426 rc = obd_statfs(&sbi->ll_mdc_conn, &osfs);
427 statfs_unpack(sfs, &osfs);
429 CERROR("mdc_statfs fails: rc = %d\n", rc);
431 CDEBUG(D_SUPER, "mdc_statfs shows blocks "LPU64"/"LPU64
432 " objects "LPU64"/"LPU64"\n",
433 osfs.os_bavail, osfs.os_blocks,
434 osfs.os_ffree, osfs.os_files);
436 /* temporary until mds_statfs returns statfs info for all OSTs */
438 rc = obd_statfs(&sbi->ll_osc_conn, &osfs);
440 CERROR("obd_statfs fails: rc = %d\n", rc);
443 CDEBUG(D_SUPER, "obd_statfs shows blocks "LPU64"/"LPU64
444 " objects "LPU64"/"LPU64"\n",
445 osfs.os_bavail, osfs.os_blocks,
446 osfs.os_ffree, osfs.os_files);
448 while (osfs.os_blocks > ~0UL) {
451 osfs.os_blocks >>= 1;
453 osfs.os_bavail >>= 1;
455 sfs->f_blocks = osfs.os_blocks;
456 sfs->f_bfree = osfs.os_bfree;
457 sfs->f_bavail = osfs.os_bavail;
458 if (osfs.os_ffree < (__u64)sfs->f_ffree)
459 sfs->f_ffree = osfs.os_ffree;
466 void ll_update_inode(struct inode *inode, struct mds_body *body)
468 if (body->valid & OBD_MD_FLID)
469 inode->i_ino = body->ino;
470 if (body->valid & OBD_MD_FLATIME)
471 inode->i_atime = body->atime;
472 if (body->valid & OBD_MD_FLMTIME)
473 inode->i_mtime = body->mtime;
474 if (body->valid & OBD_MD_FLCTIME)
475 inode->i_ctime = body->ctime;
476 if (body->valid & OBD_MD_FLMODE)
477 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
478 if (body->valid & OBD_MD_FLTYPE)
479 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
480 if (body->valid & OBD_MD_FLUID)
481 inode->i_uid = body->uid;
482 if (body->valid & OBD_MD_FLGID)
483 inode->i_gid = body->gid;
484 if (body->valid & OBD_MD_FLFLAGS)
485 inode->i_flags = body->flags;
486 if (body->valid & OBD_MD_FLNLINK)
487 inode->i_nlink = body->nlink;
488 if (body->valid & OBD_MD_FLGENER)
489 inode->i_generation = body->generation;
490 if (body->valid & OBD_MD_FLRDEV)
491 inode->i_rdev = to_kdev_t(body->rdev);
492 if (body->valid & OBD_MD_FLSIZE)
493 inode->i_size = body->size;
496 int ll_read_inode2(struct inode *inode, void *opaque)
498 struct ll_read_inode2_cookie *lic = opaque;
499 struct mds_body *body = lic->lic_body;
500 struct ll_inode_info *lli = ll_i2info(inode);
504 sema_init(&lli->lli_open_sem, 1);
506 /* core attributes first */
507 ll_update_inode(inode, body);
509 LASSERT(!lli->lli_smd);
510 if (lic && lic->lic_lmm)
511 obd_unpackmd(ll_i2obdconn(inode), &lli->lli_smd, lic->lic_lmm);
513 /* Get the authoritative file size */
514 if (lli->lli_smd && S_ISREG(inode->i_mode)) {
515 rc = ll_file_size(inode, lli->lli_smd);
517 CERROR("ll_file_size: %d\n", rc);
518 ll_clear_inode(inode);
519 make_bad_inode(inode);
524 /* OIDEBUG(inode); */
526 if (S_ISREG(inode->i_mode)) {
527 inode->i_op = &ll_file_inode_operations;
528 inode->i_fop = &ll_file_operations;
529 inode->i_mapping->a_ops = &ll_aops;
531 } else if (S_ISDIR(inode->i_mode)) {
532 inode->i_op = &ll_dir_inode_operations;
533 inode->i_fop = &ll_dir_operations;
534 inode->i_mapping->a_ops = &ll_dir_aops;
536 } else if (S_ISLNK(inode->i_mode)) {
537 inode->i_op = &ll_fast_symlink_inode_operations;
540 init_special_inode(inode, inode->i_mode,
541 kdev_t_to_nr(inode->i_rdev));
548 static inline void invalidate_request_list(struct list_head *req_list)
550 struct list_head *tmp, *n;
551 list_for_each_safe(tmp, n, req_list) {
552 struct ptlrpc_request *req =
553 list_entry(tmp, struct ptlrpc_request, rq_list);
554 CERROR("invalidating req xid %d op %d to %s:%d\n",
555 (unsigned long long)req->rq_xid, req->rq_reqmsg->opc,
556 req->rq_connection->c_remote_uuid,
557 req->rq_import->imp_client->cli_request_portal);
558 req->rq_flags |= PTL_RPC_FL_ERR;
559 wake_up(&req->rq_wait_for_rep);
563 void ll_umount_begin(struct super_block *sb)
565 struct ll_sb_info *sbi = ll_s2sbi(sb);
566 struct list_head *ctmp;
570 list_for_each(ctmp, &sbi->ll_conn_chain) {
571 struct ptlrpc_connection *conn;
572 conn = list_entry(ctmp, struct ptlrpc_connection, c_sb_chain);
574 spin_lock(&conn->c_lock);
575 conn->c_flags |= CONN_INVALID;
576 invalidate_request_list(&conn->c_sending_head);
577 invalidate_request_list(&conn->c_delayed_head);
578 spin_unlock(&conn->c_lock);
585 static kmem_cache_t *ll_inode_cachep;
587 static struct inode *ll_alloc_inode(struct super_block *sb)
589 struct ll_inode_info *lli;
590 lli = kmem_cache_alloc(ll_inode_cachep, SLAB_KERNEL);
594 memset(lli, 0, (char *)&lli->lli_vfs_inode - (char *)lli);
595 sema_init(&lli->lli_open_sem, 1);
597 return &lli->lli_vfs_inode;
600 static void ll_destroy_inode(struct inode *inode)
602 kmem_cache_free(ll_inode_cachep, ll_i2info(inode));
605 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
607 struct ll_inode_info *lli = foo;
609 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
610 SLAB_CTOR_CONSTRUCTOR)
611 inode_init_once(&lli->lli_vfs_inode);
614 int ll_init_inodecache(void)
616 ll_inode_cachep = kmem_cache_create("lustre_inode_cache",
617 sizeof(struct ll_inode_info),
618 0, SLAB_HWCACHE_ALIGN,
620 if (ll_inode_cachep == NULL)
625 void ll_destroy_inodecache(void)
627 if (kmem_cache_destroy(ll_inode_cachep))
628 CERROR("ll_inode_cache: not all structures were freed\n");
633 /* exported operations */
634 struct super_operations ll_super_operations =
636 alloc_inode: ll_alloc_inode,
637 destroy_inode: ll_destroy_inode,
638 clear_inode: ll_clear_inode,
639 delete_inode: ll_delete_inode,
640 put_super: ll_put_super,
642 umount_begin: ll_umount_begin
645 struct file_system_type lustre_lite_fs_type = {
646 .owner = THIS_MODULE,
647 .name = "lustre_lite",
649 .kill_sb = kill_litter_super,
652 static int __init init_lustre_lite(void)
655 printk(KERN_INFO "Lustre Lite 0.5.14, info@clusterfs.com\n");
656 rc = ll_init_inodecache();
659 ll_file_data_slab = kmem_cache_create("ll_file_data",
660 sizeof(struct ll_file_data), 0,
661 SLAB_HWCACHE_ALIGN, NULL, NULL);
662 if (ll_file_data_slab == NULL) {
663 ll_destroy_inodecache();
666 return register_filesystem(&lustre_lite_fs_type);
669 static void __exit exit_lustre_lite(void)
671 unregister_filesystem(&lustre_lite_fs_type);
672 ll_destroy_inodecache();
673 kmem_cache_destroy(ll_file_data_slab);
676 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
677 MODULE_DESCRIPTION("Lustre Lite Client File System v1.0");
678 MODULE_LICENSE("GPL");
680 module_init(init_lustre_lite);
681 module_exit(exit_lustre_lite);