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/obd_lov.h>
20 #include <linux/lustre_dlm.h>
21 #include <linux/init.h>
23 #include <linux/lprocfs_status.h>
25 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
26 kmem_cache_t *ll_file_data_slab;
27 extern struct address_space_operations ll_aops;
28 extern struct address_space_operations ll_dir_aops;
29 struct super_operations ll_super_operations;
31 extern int ll_init_inodecache(void);
32 extern void ll_destroy_inodecache(void);
33 extern int ll_recover(struct recovd_data *, int);
34 extern int ll_commitcbd_setup(struct ll_sb_info *);
35 extern int ll_commitcbd_cleanup(struct ll_sb_info *);
36 int ll_read_inode2(struct inode *inode, void *opaque);
38 extern void ll_proc_namespace(struct super_block* sb, char* osc, char* mdc)
40 static char *ll_read_opt(const char *opt, char *data)
46 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
47 if ( strncmp(opt, data, strlen(opt)) )
49 if ( (value = strchr(data, '=')) == NULL )
53 OBD_ALLOC(retval, strlen(value) + 1);
55 CERROR("out of memory!\n");
59 memcpy(retval, value, strlen(value)+1);
60 CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
64 static int ll_set_opt(const char *opt, char *data, int fl)
68 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
69 if ( strncmp(opt, data, strlen(opt)) )
75 static void ll_options(char *options, char **ost, char **mds, int *flags)
77 char *opt_ptr = options;
86 while ((this_char = strsep (&opt_ptr, ",")) != NULL) {
87 CDEBUG(D_SUPER, "this_char %s\n", this_char);
88 if ( (!*ost && (*ost = ll_read_opt("osc", this_char)))||
89 (!*mds && (*mds = ll_read_opt("mdc", this_char)))||
90 (!(*flags & LL_SBI_NOLCK) && ((*flags) = (*flags) |
91 ll_set_opt("nolock", this_char, LL_SBI_NOLCK))) )
98 #define log2(n) ffz(~(n))
102 static int ll_fill_super(struct super_block *sb, void *data, int silent)
104 struct inode *root = 0;
105 struct obd_device *obd;
106 struct ll_sb_info *sbi;
110 struct ll_fid rootfid;
111 struct obd_statfs osfs;
112 struct ptlrpc_request *request = NULL;
113 struct ptlrpc_connection *mdc_conn;
114 struct ll_read_inode2_cookie lic;
120 OBD_ALLOC(sbi, sizeof(*sbi));
126 INIT_LIST_HEAD(&sbi->ll_conn_chain);
127 generate_random_uuid(uuid);
128 class_uuid_unparse(uuid, sbi->ll_sb_uuid);
132 ll_options(data, &osc, &mdc, &sbi->ll_flags);
136 GOTO(out_free, sb = NULL);
141 GOTO(out_free, sb = NULL);
144 obd = class_uuid2obd(mdc);
146 CERROR("MDC %s: not setup or attached\n", mdc);
147 GOTO(out_free, sb = NULL);
150 err = obd_connect(&sbi->ll_mdc_conn, obd, sbi->ll_sb_uuid,
151 ptlrpc_recovd, ll_recover);
153 CERROR("cannot connect to %s: rc = %d\n", mdc, err);
154 GOTO(out_free, sb = NULL);
157 #warning Peter: is this the right place to raise the connection level?
158 mdc_conn = sbi2mdc(sbi)->cl_import.imp_connection;
159 mdc_conn->c_level = LUSTRE_CONN_FULL;
160 list_add(&mdc_conn->c_sb_chain, &sbi->ll_conn_chain);
162 obd = class_uuid2obd(osc);
164 CERROR("OSC %s: not setup or attached\n", osc);
165 GOTO(out_mdc, sb = NULL);
168 err = obd_connect(&sbi->ll_osc_conn, obd, sbi->ll_sb_uuid,
169 ptlrpc_recovd, ll_recover);
171 CERROR("cannot connect to %s: rc = %d\n", osc, err);
172 GOTO(out_mdc, sb = NULL);
175 err = mdc_getstatus(&sbi->ll_mdc_conn, &rootfid);
177 CERROR("cannot mds_connect: rc = %d\n", err);
178 GOTO(out_mdc, sb = NULL);
180 CDEBUG(D_SUPER, "rootfid "LPU64"\n", rootfid.id);
181 sbi->ll_rootino = rootfid.id;
183 memset(&osfs, 0, sizeof(osfs));
184 err = mdc_statfs(&sbi->ll_mdc_conn, &osfs);
185 sb->s_blocksize = osfs.os_bsize;
186 sb->s_blocksize_bits = log2(osfs.os_bsize);
187 sb->s_magic = LL_SUPER_MAGIC;
188 sb->s_maxbytes = (1ULL << (32 + 9)) - osfs.os_bsize;
190 sb->s_op = &ll_super_operations;
192 /* make root inode */
193 err = mdc_getattr(&sbi->ll_mdc_conn, sbi->ll_rootino, S_IFDIR,
194 OBD_MD_FLNOTOBD|OBD_MD_FLBLOCKS, 0, &request);
196 CERROR("mdc_getattr failed for root: rc = %d\n", err);
197 GOTO(out_request, sb = NULL);
200 /* initialize committed transaction callback daemon */
201 spin_lock_init(&sbi->ll_commitcbd_lock);
202 init_waitqueue_head(&sbi->ll_commitcbd_waitq);
203 init_waitqueue_head(&sbi->ll_commitcbd_ctl_waitq);
204 sbi->ll_commitcbd_flags = 0;
205 err = ll_commitcbd_setup(sbi);
207 CERROR("failed to start commit callback daemon: rc = %d\n",err);
208 GOTO(out_request, sb = NULL);
211 lic.lic_body = lustre_msg_buf(request->rq_repmsg, 0);
213 root = iget5_locked(sb, sbi->ll_rootino, NULL,
214 ll_read_inode2, &lic);
217 sb->s_root = d_alloc_root(root);
219 CERROR("lustre_lite: bad iget4 for root\n");
220 GOTO(out_cdb, sb = NULL);
223 ptlrpc_req_finished(request);
225 ll_proc_namespace(sb, osc, mdc)
228 OBD_FREE(mdc, strlen(mdc) + 1);
230 OBD_FREE(osc, strlen(osc) + 1);
235 ll_commitcbd_cleanup(sbi);
237 ptlrpc_req_finished(request);
238 obd_disconnect(&sbi->ll_osc_conn);
240 obd_disconnect(&sbi->ll_mdc_conn);
242 OBD_FREE(sbi, sizeof(*sbi));
246 } /* ll_fill_super */
248 struct super_block * ll_get_sb(struct file_system_type *fs_type,
249 int flags, char *devname, void * data)
251 return get_sb_nodev(fs_type, flags, data, ll_fill_super);
254 static void ll_put_super(struct super_block *sb)
256 struct ll_sb_info *sbi = ll_s2sbi(sb);
257 struct ll_fid rootfid;
260 list_del(&sbi->ll_conn_chain);
261 ll_commitcbd_cleanup(sbi);
262 obd_disconnect(&sbi->ll_osc_conn);
264 /* NULL request to force sync on the MDS, and get the last_committed
265 * value to flush remaining RPCs from the pending queue on client.
267 * XXX This should be an mdc_sync() call to sync the whole MDS fs,
268 * which we can call for other reasons as well.
270 mdc_getstatus(&sbi->ll_mdc_conn, &rootfid);
272 lprocfs_dereg_mnt(sbi->ll_proc_root);
273 sbi->ll_proc_root = NULL;
275 obd_disconnect(&sbi->ll_mdc_conn);
276 OBD_FREE(sbi, sizeof(*sbi));
282 static void ll_clear_inode(struct inode *inode)
286 if (atomic_read(&inode->i_count) == 0) {
287 struct ll_inode_info *lli = ll_i2info(inode);
288 struct lov_stripe_md *lsm = lli->lli_smd;
289 char *symlink_name = lli->lli_symlink_name;
292 OBD_FREE(lsm, ll_ost_easize(inode->i_sb));
296 OBD_FREE(symlink_name, strlen(symlink_name) + 1);
297 lli->lli_symlink_name = NULL;
303 static void ll_delete_inode(struct inode *inode)
306 if (S_ISREG(inode->i_mode)) {
309 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
314 if (lsm->lsm_object_id == 0) {
315 CERROR("This really happens\n");
316 /* No obdo was ever created */
324 oa->o_id = lsm->lsm_object_id;
325 oa->o_easize = ll_mds_easize(inode->i_sb);
326 oa->o_mode = inode->i_mode;
327 oa->o_valid = OBD_MD_FLID | OBD_MD_FLEASIZE | OBD_MD_FLTYPE;
329 err = obd_destroy(ll_i2obdconn(inode), oa, lsm);
331 CDEBUG(D_SUPER, "obd destroy of objid "LPX64" error %d\n",
332 lsm->lsm_object_id, err);
339 /* like inode_setattr, but doesn't mark the inode dirty */
340 static int ll_attr2inode(struct inode * inode, struct iattr * attr, int trunc)
342 unsigned int ia_valid = attr->ia_valid;
345 if ((ia_valid & ATTR_SIZE) && trunc) {
346 error = vmtruncate(inode, attr->ia_size);
349 } else if (ia_valid & ATTR_SIZE)
350 inode->i_size = attr->ia_size;
352 if (ia_valid & ATTR_UID)
353 inode->i_uid = attr->ia_uid;
354 if (ia_valid & ATTR_GID)
355 inode->i_gid = attr->ia_gid;
356 if (ia_valid & ATTR_ATIME)
357 inode->i_atime = attr->ia_atime;
358 if (ia_valid & ATTR_MTIME)
359 inode->i_mtime = attr->ia_mtime;
360 if (ia_valid & ATTR_CTIME)
361 inode->i_ctime = attr->ia_ctime;
362 if (ia_valid & ATTR_MODE) {
363 inode->i_mode = attr->ia_mode;
364 if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID))
365 inode->i_mode &= ~S_ISGID;
371 int ll_inode_setattr(struct inode *inode, struct iattr *attr, int do_trunc)
373 struct ptlrpc_request *request = NULL;
374 struct ll_sb_info *sbi = ll_i2sbi(inode);
379 /* change incore inode */
380 ll_attr2inode(inode, attr, do_trunc);
382 err = mdc_setattr(&sbi->ll_mdc_conn, inode, attr, &request);
384 CERROR("mdc_setattr fails (%d)\n", err);
386 ptlrpc_req_finished(request);
391 int ll_setattr(struct dentry *de, struct iattr *attr)
393 int rc = inode_change_ok(de->d_inode, attr);
398 return ll_inode_setattr(de->d_inode, attr, 1);
401 static int ll_statfs(struct super_block *sb, struct statfs *sfs)
403 struct ll_sb_info *sbi = ll_s2sbi(sb);
404 struct obd_statfs osfs;
408 memset(sfs, 0, sizeof(*sfs));
409 rc = obd_statfs(&sbi->ll_mdc_conn, &osfs);
410 statfs_unpack(sfs, &osfs);
412 CERROR("mdc_statfs fails: rc = %d\n", rc);
414 CDEBUG(D_SUPER, "mdc_statfs shows blocks "LPU64"/"LPU64
415 " objects "LPU64"/"LPU64"\n",
416 osfs.os_bavail, osfs.os_blocks,
417 osfs.os_ffree, osfs.os_files);
419 /* temporary until mds_statfs returns statfs info for all OSTs */
421 rc = obd_statfs(&sbi->ll_osc_conn, &osfs);
423 CERROR("obd_statfs fails: rc = %d\n", rc);
426 CDEBUG(D_SUPER, "obd_statfs shows blocks "LPU64"/"LPU64
427 " objects "LPU64"/"LPU64"\n",
428 osfs.os_bavail, osfs.os_blocks,
429 osfs.os_ffree, osfs.os_files);
431 while (osfs.os_blocks > ~0UL) {
434 osfs.os_blocks >>= 1;
436 osfs.os_bavail >>= 1;
438 sfs->f_blocks = osfs.os_blocks;
439 sfs->f_bfree = osfs.os_bfree;
440 sfs->f_bavail = osfs.os_bavail;
441 if (osfs.os_ffree < (__u64)sfs->f_ffree)
442 sfs->f_ffree = osfs.os_ffree;
449 void ll_update_inode(struct inode *inode, struct mds_body *body)
451 if (body->valid & OBD_MD_FLID)
452 inode->i_ino = body->ino;
453 if (body->valid & OBD_MD_FLATIME)
454 inode->i_atime = body->atime;
455 if (body->valid & OBD_MD_FLMTIME)
456 inode->i_mtime = body->mtime;
457 if (body->valid & OBD_MD_FLCTIME)
458 inode->i_ctime = body->ctime;
459 if (body->valid & OBD_MD_FLMODE)
460 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
461 if (body->valid & OBD_MD_FLTYPE)
462 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
463 if (body->valid & OBD_MD_FLUID)
464 inode->i_uid = body->uid;
465 if (body->valid & OBD_MD_FLGID)
466 inode->i_gid = body->gid;
467 if (body->valid & OBD_MD_FLFLAGS)
468 inode->i_flags = body->flags;
469 if (body->valid & OBD_MD_FLNLINK)
470 inode->i_nlink = body->nlink;
471 if (body->valid & OBD_MD_FLGENER)
472 inode->i_generation = body->generation;
473 if (body->valid & OBD_MD_FLRDEV)
474 inode->i_rdev = to_kdev_t(body->rdev);
475 if (body->valid & OBD_MD_FLSIZE)
476 inode->i_size = body->size;
479 int ll_read_inode2(struct inode *inode, void *opaque)
481 struct ll_read_inode2_cookie *lic = opaque;
482 struct mds_body *body = lic->lic_body;
483 struct ll_inode_info *lli = ll_i2info(inode);
486 sema_init(&lli->lli_open_sem, 1);
488 /* core attributes first */
489 ll_update_inode(inode, body);
491 //if (body->valid & OBD_MD_FLEASIZE)
492 if (lic && lic->lic_lmm) {
493 struct lov_mds_md *lmm = lic->lic_lmm;
496 /* XXX This should probably not be an error in the future,
497 * when we allow LOV OSTs to be added.
499 if (lmm->lmm_easize != ll_mds_easize(inode->i_sb)) {
500 CERROR("Striping metadata size error %ld\n",
504 size = ll_ost_easize(inode->i_sb);
505 OBD_ALLOC(lli->lli_smd, size);
507 CERROR("No memory for %d\n", size);
510 lov_unpackmd(lli->lli_smd, lmm);
515 /* Get the authoritative file size */
516 if (lli->lli_smd && (inode->i_mode & S_IFREG)) {
519 rc = ll_file_size(inode, lli->lli_smd);
521 CERROR("ll_file_size: %d\n", rc);
522 /* FIXME: need to somehow prevent inode creation */
527 /* OIDEBUG(inode); */
529 if (S_ISREG(inode->i_mode)) {
530 inode->i_op = &ll_file_inode_operations;
531 inode->i_fop = &ll_file_operations;
532 inode->i_mapping->a_ops = &ll_aops;
534 } else if (S_ISDIR(inode->i_mode)) {
535 inode->i_op = &ll_dir_inode_operations;
536 inode->i_fop = &ll_dir_operations;
537 inode->i_mapping->a_ops = &ll_dir_aops;
539 } else if (S_ISLNK(inode->i_mode)) {
540 inode->i_op = &ll_fast_symlink_inode_operations;
543 init_special_inode(inode, inode->i_mode,
544 kdev_t_to_nr(inode->i_rdev));
550 static inline void invalidate_request_list(struct list_head *req_list)
552 struct list_head *tmp, *n;
553 list_for_each_safe(tmp, n, req_list) {
554 struct ptlrpc_request *req =
555 list_entry(tmp, struct ptlrpc_request, rq_list);
556 CERROR("invalidating req xid %d op %d to %s:%d\n",
557 (unsigned long long)req->rq_xid, req->rq_reqmsg->opc,
558 req->rq_connection->c_remote_uuid,
559 req->rq_import->imp_client->cli_request_portal);
560 req->rq_flags |= PTL_RPC_FL_ERR;
561 wake_up(&req->rq_wait_for_rep);
565 void ll_umount_begin(struct super_block *sb)
567 struct ll_sb_info *sbi = ll_s2sbi(sb);
568 struct list_head *ctmp;
572 list_for_each(ctmp, &sbi->ll_conn_chain) {
573 struct ptlrpc_connection *conn;
574 conn = list_entry(ctmp, struct ptlrpc_connection, c_sb_chain);
576 spin_lock(&conn->c_lock);
577 conn->c_flags |= CONN_INVALID;
578 invalidate_request_list(&conn->c_sending_head);
579 invalidate_request_list(&conn->c_delayed_head);
580 spin_unlock(&conn->c_lock);
587 static kmem_cache_t *ll_inode_cachep;
589 static struct inode *ll_alloc_inode(struct super_block *sb)
591 struct ll_inode_info *lli;
592 lli = kmem_cache_alloc(ll_inode_cachep, SLAB_KERNEL);
596 memset(lli, 0, (char *)&lli->lli_vfs_inode - (char *)lli);
597 sema_init(&lli->lli_open_sem, 1);
599 return &lli->lli_vfs_inode;
602 static void ll_destroy_inode(struct inode *inode)
604 kmem_cache_free(ll_inode_cachep, ll_i2info(inode));
607 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
609 struct ll_inode_info *lli = foo;
611 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
612 SLAB_CTOR_CONSTRUCTOR)
613 inode_init_once(&lli->lli_vfs_inode);
616 int ll_init_inodecache(void)
618 ll_inode_cachep = kmem_cache_create("lustre_inode_cache",
619 sizeof(struct ll_inode_info),
620 0, SLAB_HWCACHE_ALIGN,
622 if (ll_inode_cachep == NULL)
627 void ll_destroy_inodecache(void)
629 if (kmem_cache_destroy(ll_inode_cachep))
630 CERROR("ll_inode_cache: not all structures were freed\n");
635 /* exported operations */
636 struct super_operations ll_super_operations =
638 alloc_inode: ll_alloc_inode,
639 destroy_inode: ll_destroy_inode,
640 clear_inode: ll_clear_inode,
641 delete_inode: ll_delete_inode,
642 put_super: ll_put_super,
644 umount_begin: ll_umount_begin
647 struct file_system_type lustre_lite_fs_type = {
648 .owner = THIS_MODULE,
649 .name = "lustre_lite",
651 .kill_sb = kill_litter_super,
654 static int __init init_lustre_lite(void)
657 printk(KERN_INFO "Lustre Lite 0.5.14, info@clusterfs.com\n");
658 rc = ll_init_inodecache();
661 ll_file_data_slab = kmem_cache_create("ll_file_data",
662 sizeof(struct ll_file_data), 0,
663 SLAB_HWCACHE_ALIGN, NULL, NULL);
664 if (ll_file_data_slab == NULL) {
665 ll_destroy_inodecache();
668 return register_filesystem(&lustre_lite_fs_type);
671 static void __exit exit_lustre_lite(void)
673 unregister_filesystem(&lustre_lite_fs_type);
674 ll_destroy_inodecache();
675 kmem_cache_destroy(ll_file_data_slab);
678 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
679 MODULE_DESCRIPTION("Lustre Lite Client File System v1.0");
680 MODULE_LICENSE("GPL");
682 module_init(init_lustre_lite);
683 module_exit(exit_lustre_lite);