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>
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 static char *ll_read_opt(const char *opt, char *data)
43 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
44 if ( strncmp(opt, data, strlen(opt)) )
46 if ( (value = strchr(data, '=')) == NULL )
50 OBD_ALLOC(retval, strlen(value) + 1);
52 CERROR("out of memory!\n");
56 memcpy(retval, value, strlen(value)+1);
57 CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
61 static int ll_set_opt(const char *opt, char *data, int fl)
65 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
66 if ( strncmp(opt, data, strlen(opt)) )
72 static void ll_options(char *options, char **ost, char **mds, int *flags)
74 char *opt_ptr = options;
83 while ((this_char = strsep (&opt_ptr, ",")) != NULL) {
84 CDEBUG(D_SUPER, "this_char %s\n", this_char);
85 if ( (!*ost && (*ost = ll_read_opt("osc", this_char)))||
86 (!*mds && (*mds = ll_read_opt("mdc", this_char)))||
87 (!(*flags & LL_SBI_NOLCK) && ((*flags) = (*flags) |
88 ll_set_opt("nolock", this_char, LL_SBI_NOLCK))) )
95 #define log2(n) ffz(~(n))
99 static int ll_fill_super(struct super_block *sb, void *data, int silent)
101 struct inode *root = 0;
102 struct obd_device *obd;
103 struct ll_sb_info *sbi;
107 struct ll_fid rootfid;
108 struct obd_statfs osfs;
109 __u64 last_committed;
111 struct ptlrpc_request *request = NULL;
112 struct ptlrpc_connection *mdc_conn;
113 struct ll_read_inode2_cookie lic;
119 OBD_ALLOC(sbi, sizeof(*sbi));
125 INIT_LIST_HEAD(&sbi->ll_conn_chain);
126 sbi->ll_mount_epoch = 0;
127 generate_random_uuid(uuid);
128 class_uuid_unparse(uuid, sbi->ll_sb_uuid);
130 sb->u.generic_sbp = sbi;
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 /* XXX: need to store the last_* values somewhere */
176 err = mdc_getstatus(&sbi->ll_mdc_conn, &rootfid, &last_committed,
177 &last_xid, &request);
178 ptlrpc_req_finished(request);
180 CERROR("cannot mds_connect: rc = %d\n", err);
181 GOTO(out_mdc, sb = NULL);
183 CDEBUG(D_SUPER, "rootfid %Ld\n", (unsigned long long)rootfid.id);
184 sbi->ll_rootino = rootfid.id;
186 memset(&osfs, 0, sizeof(osfs));
188 err = mdc_statfs(&sbi->ll_mdc_conn, &osfs, &request);
189 ptlrpc_req_finished(request);
190 sb->s_blocksize = osfs.os_bsize;
191 sb->s_blocksize_bits = log2(osfs.os_bsize);
192 sb->s_magic = LL_SUPER_MAGIC;
193 sb->s_maxbytes = (1ULL << (32 + 9)) - osfs.os_bsize;
195 sb->s_op = &ll_super_operations;
197 /* make root inode */
199 err = mdc_getattr(&sbi->ll_mdc_conn, sbi->ll_rootino, S_IFDIR,
200 OBD_MD_FLNOTOBD|OBD_MD_FLBLOCKS, 0, &request);
202 CERROR("mdc_getattr failed for root: rc = %d\n", err);
203 GOTO(out_request, sb = NULL);
206 /* initialize committed transaction callback daemon */
207 spin_lock_init(&sbi->ll_commitcbd_lock);
208 init_waitqueue_head(&sbi->ll_commitcbd_waitq);
209 init_waitqueue_head(&sbi->ll_commitcbd_ctl_waitq);
210 sbi->ll_commitcbd_flags = 0;
211 err = ll_commitcbd_setup(sbi);
213 CERROR("failed to start commit callback daemon: rc = %d\n",err);
214 GOTO(out_request, sb = NULL);
217 lic.lic_body = lustre_msg_buf(request->rq_repmsg, 0);
219 root = iget5_locked(sb, sbi->ll_rootino, NULL,
220 ll_read_inode2, &lic);
223 sb->s_root = d_alloc_root(root);
225 CERROR("lustre_lite: bad iget4 for root\n");
226 GOTO(out_cdb, sb = NULL);
229 ptlrpc_req_finished(request);
233 OBD_FREE(mdc, strlen(mdc) + 1);
235 OBD_FREE(osc, strlen(osc) + 1);
240 ll_commitcbd_cleanup(sbi);
242 ptlrpc_req_finished(request);
243 obd_disconnect(&sbi->ll_osc_conn);
245 obd_disconnect(&sbi->ll_mdc_conn);
247 OBD_FREE(sbi, sizeof(*sbi));
251 } /* ll_fill_super */
253 struct super_block * ll_get_sb(struct file_system_type *fs_type,
254 int flags, char *devname, void * data)
256 return get_sb_nodev(fs_type, flags, data, ll_fill_super);
259 static void ll_put_super(struct super_block *sb)
261 struct ll_sb_info *sbi = ll_s2sbi(sb);
264 list_del(&sbi->ll_conn_chain);
265 ll_commitcbd_cleanup(sbi);
266 obd_disconnect(&sbi->ll_osc_conn);
267 obd_disconnect(&sbi->ll_mdc_conn);
268 OBD_FREE(sbi, sizeof(*sbi));
274 static void ll_clear_inode(struct inode *inode)
278 if (atomic_read(&inode->i_count) == 0) {
279 struct ll_inode_info *lli = ll_i2info(inode);
280 struct lov_stripe_md *lsm = lli->lli_smd;
281 char *symlink_name = lli->lli_symlink_name;
284 OBD_FREE(lsm, ll_ost_easize(inode->i_sb));
288 OBD_FREE(symlink_name, strlen(symlink_name) + 1);
289 lli->lli_symlink_name = NULL;
295 static void ll_delete_inode(struct inode *inode)
298 if (S_ISREG(inode->i_mode)) {
301 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
306 if (lsm->lsm_object_id == 0) {
307 CERROR("This really happens\n");
308 /* No obdo was ever created */
316 oa->o_id = lsm->lsm_object_id;
317 oa->o_easize = ll_mds_easize(inode->i_sb);
318 oa->o_mode = inode->i_mode;
319 oa->o_valid = OBD_MD_FLID | OBD_MD_FLEASIZE | OBD_MD_FLTYPE;
321 err = obd_destroy(ll_i2obdconn(inode), oa, lsm);
323 CDEBUG(D_SUPER, "obd destroy of objid "LPX64" error %d\n",
324 lsm->lsm_object_id, err);
331 /* like inode_setattr, but doesn't mark the inode dirty */
332 static int ll_attr2inode(struct inode * inode, struct iattr * attr, int trunc)
334 unsigned int ia_valid = attr->ia_valid;
337 if ((ia_valid & ATTR_SIZE) && trunc) {
338 error = vmtruncate(inode, attr->ia_size);
341 } else if (ia_valid & ATTR_SIZE)
342 inode->i_size = attr->ia_size;
344 if (ia_valid & ATTR_UID)
345 inode->i_uid = attr->ia_uid;
346 if (ia_valid & ATTR_GID)
347 inode->i_gid = attr->ia_gid;
348 if (ia_valid & ATTR_ATIME)
349 inode->i_atime = attr->ia_atime;
350 if (ia_valid & ATTR_MTIME)
351 inode->i_mtime = attr->ia_mtime;
352 if (ia_valid & ATTR_CTIME)
353 inode->i_ctime = attr->ia_ctime;
354 if (ia_valid & ATTR_MODE) {
355 inode->i_mode = attr->ia_mode;
356 if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID))
357 inode->i_mode &= ~S_ISGID;
363 int ll_inode_setattr(struct inode *inode, struct iattr *attr, int do_trunc)
365 struct ptlrpc_request *request = NULL;
366 struct ll_sb_info *sbi = ll_i2sbi(inode);
371 /* change incore inode */
372 ll_attr2inode(inode, attr, do_trunc);
374 err = mdc_setattr(&sbi->ll_mdc_conn, inode, attr, &request);
376 CERROR("mdc_setattr fails (%d)\n", err);
378 ptlrpc_req_finished(request);
383 int ll_setattr(struct dentry *de, struct iattr *attr)
385 int rc = inode_change_ok(de->d_inode, attr);
390 return ll_inode_setattr(de->d_inode, attr, 1);
393 static int ll_statfs(struct super_block *sb, struct statfs *sfs)
395 struct ptlrpc_request *request = NULL;
396 struct ll_sb_info *sbi = ll_s2sbi(sb);
397 struct obd_statfs osfs;
401 memset(sfs, 0, sizeof(*sfs));
402 rc = mdc_statfs(&sbi->ll_mdc_conn, &osfs, &request);
403 statfs_unpack(sfs, &osfs);
404 ptlrpc_req_finished(request);
406 CERROR("mdc_statfs fails: rc = %d\n", rc);
408 CDEBUG(D_SUPER, "mdc_statfs shows blocks "LPU64"/"LPU64
409 " objects "LPU64"/"LPU64"\n",
410 osfs.os_bavail, osfs.os_blocks,
411 osfs.os_ffree, osfs.os_files);
413 /* temporary until mds_statfs returns statfs info for all OSTs */
415 rc = obd_statfs(&sbi->ll_osc_conn, &osfs);
417 CERROR("obd_statfs fails: rc = %d\n", rc);
420 CDEBUG(D_SUPER, "obd_statfs shows blocks "LPU64"/"LPU64
421 " objects "LPU64"/"LPU64"\n",
422 osfs.os_bavail, osfs.os_blocks,
423 osfs.os_ffree, osfs.os_files);
425 while (osfs.os_blocks > ~0UL) {
428 osfs.os_blocks >>= 1;
430 osfs.os_bavail >>= 1;
432 sfs->f_blocks = osfs.os_blocks;
433 sfs->f_bfree = osfs.os_bfree;
434 sfs->f_bavail = osfs.os_bavail;
435 if (osfs.os_ffree < (__u64)sfs->f_ffree)
436 sfs->f_ffree = osfs.os_ffree;
443 void ll_update_inode(struct inode *inode, struct mds_body *body)
445 if (body->valid & OBD_MD_FLID)
446 inode->i_ino = body->ino;
447 if (body->valid & OBD_MD_FLATIME)
448 inode->i_atime = body->atime;
449 if (body->valid & OBD_MD_FLMTIME)
450 inode->i_mtime = body->mtime;
451 if (body->valid & OBD_MD_FLCTIME)
452 inode->i_ctime = body->ctime;
453 if (body->valid & OBD_MD_FLMODE)
454 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
455 if (body->valid & OBD_MD_FLTYPE)
456 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
457 if (body->valid & OBD_MD_FLUID)
458 inode->i_uid = body->uid;
459 if (body->valid & OBD_MD_FLGID)
460 inode->i_gid = body->gid;
461 if (body->valid & OBD_MD_FLFLAGS)
462 inode->i_flags = body->flags;
463 if (body->valid & OBD_MD_FLNLINK)
464 inode->i_nlink = body->nlink;
465 if (body->valid & OBD_MD_FLGENER)
466 inode->i_generation = body->generation;
467 if (body->valid & OBD_MD_FLRDEV)
468 inode->i_rdev = to_kdev_t(body->rdev);
469 if (body->valid & OBD_MD_FLSIZE)
470 inode->i_size = body->size;
473 int ll_read_inode2(struct inode *inode, void *opaque)
475 struct ll_read_inode2_cookie *lic = opaque;
476 struct mds_body *body = lic->lic_body;
477 struct ll_inode_info *lli = ll_i2info(inode);
480 sema_init(&lli->lli_open_sem, 1);
481 lli->lli_mount_epoch = ll_i2sbi(inode)->ll_mount_epoch;
483 /* core attributes first */
484 ll_update_inode(inode, body);
486 //if (body->valid & OBD_MD_FLEASIZE)
487 if (lic && lic->lic_lmm) {
488 struct lov_mds_md *lmm = lic->lic_lmm;
491 if (lmm->lmm_easize != ll_mds_easize(inode->i_sb)) {
492 CERROR("Striping metadata size error %ld\n",
496 size = ll_ost_easize(inode->i_sb);
497 OBD_ALLOC(lli->lli_smd, size);
499 CERROR("No memory for %d\n", size);
502 lov_unpackmd(lli->lli_smd, lmm);
507 /* Get the authoritative file size */
508 if (lli->lli_smd && (inode->i_mode & S_IFREG)) {
511 rc = ll_file_size(inode, lli->lli_smd);
513 CERROR("ll_file_size: %d\n", rc);
514 /* FIXME: need to somehow prevent inode creation */
519 /* OIDEBUG(inode); */
521 if (S_ISREG(inode->i_mode)) {
522 inode->i_op = &ll_file_inode_operations;
523 inode->i_fop = &ll_file_operations;
524 inode->i_mapping->a_ops = &ll_aops;
526 } else if (S_ISDIR(inode->i_mode)) {
527 inode->i_op = &ll_dir_inode_operations;
528 inode->i_fop = &ll_dir_operations;
529 inode->i_mapping->a_ops = &ll_dir_aops;
531 } else if (S_ISLNK(inode->i_mode)) {
532 inode->i_op = &ll_fast_symlink_inode_operations;
535 init_special_inode(inode, inode->i_mode,
536 kdev_t_to_nr(inode->i_rdev));
542 static inline void invalidate_request_list(struct list_head *req_list)
544 struct list_head *tmp, *n;
545 list_for_each_safe(tmp, n, req_list) {
546 struct ptlrpc_request *req =
547 list_entry(tmp, struct ptlrpc_request, rq_list);
548 CERROR("invalidating req xid %d op %d to %s:%d\n",
549 (unsigned long long)req->rq_xid, req->rq_reqmsg->opc,
550 req->rq_connection->c_remote_uuid,
551 req->rq_import->imp_client->cli_request_portal);
552 req->rq_flags |= PTL_RPC_FL_ERR;
553 wake_up(&req->rq_wait_for_rep);
557 void ll_umount_begin(struct super_block *sb)
559 struct ll_sb_info *sbi = ll_s2sbi(sb);
560 struct list_head *ctmp;
564 list_for_each(ctmp, &sbi->ll_conn_chain) {
565 struct ptlrpc_connection *conn;
566 conn = list_entry(ctmp, struct ptlrpc_connection, c_sb_chain);
568 spin_lock(&conn->c_lock);
569 conn->c_flags |= CONN_INVALID;
570 invalidate_request_list(&conn->c_sending_head);
571 invalidate_request_list(&conn->c_delayed_head);
572 spin_unlock(&conn->c_lock);
579 static kmem_cache_t *ll_inode_cachep;
581 static struct inode *ll_alloc_inode(struct super_block *sb)
583 struct ll_inode_info *lli;
584 lli = kmem_cache_alloc(ll_inode_cachep, SLAB_KERNEL);
588 memset(lli, 0, sizeof(*lli));
589 sema_init(&lli->lli_open_sem, 1);
590 lli->lli_mount_epoch = ll_i2sbi(&lli->lli_vfs_inode)->ll_mount_epoch;
592 return &lli->lli_vfs_inode;
595 static void ll_destroy_inode(struct inode *inode)
597 kmem_cache_free(ll_inode_cachep, LL_I(inode));
600 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
602 struct ll_inode_info *lli = foo;
604 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
605 SLAB_CTOR_CONSTRUCTOR)
606 inode_init_once(&lli->lli_vfs_inode);
609 int ll_init_inodecache(void)
611 ll_inode_cachep = kmem_cache_create("lustre_inode_cache",
612 sizeof(struct ll_inode_info),
613 0, SLAB_HWCACHE_ALIGN,
615 if (ll_inode_cachep == NULL)
620 void ll_destroy_inodecache(void)
622 if (kmem_cache_destroy(ll_inode_cachep))
623 CERROR("ll_inode_cache: not all structures were freed\n");
628 /* exported operations */
629 struct super_operations ll_super_operations =
631 alloc_inode: ll_alloc_inode,
632 destroy_inode: ll_destroy_inode,
633 clear_inode: ll_clear_inode,
634 delete_inode: ll_delete_inode,
635 put_super: ll_put_super,
637 umount_begin: ll_umount_begin
640 struct file_system_type lustre_lite_fs_type = {
641 .owner = THIS_MODULE,
642 .name = "lustre_lite",
644 .kill_sb = kill_litter_super,
647 static int __init init_lustre_lite(void)
650 printk(KERN_INFO "Lustre Lite 0.5.14, info@clusterfs.com\n");
651 rc = ll_init_inodecache();
654 ll_file_data_slab = kmem_cache_create("ll_file_data",
655 sizeof(struct ll_file_data), 0,
656 SLAB_HWCACHE_ALIGN, NULL, NULL);
657 if (ll_file_data_slab == NULL) {
658 ll_destroy_inodecache();
661 return register_filesystem(&lustre_lite_fs_type);
664 static void __exit exit_lustre_lite(void)
666 unregister_filesystem(&lustre_lite_fs_type);
667 ll_destroy_inodecache();
668 kmem_cache_destroy(ll_file_data_slab);
671 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
672 MODULE_DESCRIPTION("Lustre Lite Client File System v1.0");
673 MODULE_LICENSE("GPL");
675 module_init(init_lustre_lite);
676 module_exit(exit_lustre_lite);