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_recover(struct recovd_data *, int);
31 extern int ll_commitcbd_setup(struct ll_sb_info *);
32 extern int ll_commitcbd_cleanup(struct ll_sb_info *);
34 extern void ll_proc_namespace(struct super_block* sb, char* osc, char* mdc);
36 static char *ll_read_opt(const char *opt, char *data)
42 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
43 if ( strncmp(opt, data, strlen(opt)) )
45 if ( (value = strchr(data, '=')) == NULL )
49 OBD_ALLOC(retval, strlen(value) + 1);
51 CERROR("out of memory!\n");
55 memcpy(retval, value, strlen(value)+1);
56 CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
60 static int ll_set_opt(const char *opt, char *data, int fl)
64 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
65 if ( strncmp(opt, data, strlen(opt)) )
71 static void ll_options(char *options, char **ost, char **mds, int *flags)
81 for (this_char = strtok (options, ",");
83 this_char = strtok (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))
98 static struct super_block *ll_read_super(struct super_block *sb,
99 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 struct ptlrpc_request *request = NULL;
110 struct ptlrpc_connection *mdc_conn;
111 struct ll_read_inode2_cookie lic;
116 OBD_ALLOC(sbi, sizeof(*sbi));
120 INIT_LIST_HEAD(&sbi->ll_conn_chain);
121 INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
122 generate_random_uuid(uuid);
123 class_uuid_unparse(uuid, sbi->ll_sb_uuid);
125 sb->u.generic_sbp = sbi;
127 ll_options(data, &osc, &mdc, &sbi->ll_flags);
131 GOTO(out_free, sb = NULL);
136 GOTO(out_free, sb = NULL);
139 obd = class_uuid2obd(mdc);
141 CERROR("MDC %s: not setup or attached\n", mdc);
142 GOTO(out_free, sb = NULL);
145 err = obd_connect(&sbi->ll_mdc_conn, obd, sbi->ll_sb_uuid,
146 ptlrpc_recovd, ll_recover);
148 CERROR("cannot connect to %s: rc = %d\n", mdc, err);
149 GOTO(out_free, sb = NULL);
152 mdc_conn = sbi2mdc(sbi)->cl_import.imp_connection;
153 list_add(&mdc_conn->c_sb_chain, &sbi->ll_conn_chain);
155 obd = class_uuid2obd(osc);
157 CERROR("OSC %s: not setup or attached\n", osc);
158 GOTO(out_mdc, sb = NULL);
161 err = obd_connect(&sbi->ll_osc_conn, obd, sbi->ll_sb_uuid,
162 ptlrpc_recovd, ll_recover);
164 CERROR("cannot connect to %s: rc = %d\n", osc, err);
165 GOTO(out_mdc, sb = NULL);
168 err = mdc_getstatus(&sbi->ll_mdc_conn, &rootfid);
170 CERROR("cannot mds_connect: rc = %d\n", err);
171 GOTO(out_mdc, sb = NULL);
173 CDEBUG(D_SUPER, "rootfid "LPU64"\n", rootfid.id);
174 sbi->ll_rootino = rootfid.id;
176 memset(&osfs, 0, sizeof(osfs));
177 err = obd_statfs(&sbi->ll_mdc_conn, &osfs);
178 sb->s_blocksize = osfs.os_bsize;
179 sb->s_blocksize_bits = log2(osfs.os_bsize);
180 sb->s_magic = LL_SUPER_MAGIC;
181 sb->s_maxbytes = (1ULL << (32 + 9)) - osfs.os_bsize;
183 sb->s_op = &ll_super_operations;
185 /* make root inode */
186 err = mdc_getattr(&sbi->ll_mdc_conn, sbi->ll_rootino, S_IFDIR,
187 OBD_MD_FLNOTOBD|OBD_MD_FLBLOCKS, 0, &request);
189 CERROR("mdc_getattr failed for root: rc = %d\n", err);
190 GOTO(out_request, sb = NULL);
193 /* initialize committed transaction callback daemon */
194 spin_lock_init(&sbi->ll_commitcbd_lock);
195 init_waitqueue_head(&sbi->ll_commitcbd_waitq);
196 init_waitqueue_head(&sbi->ll_commitcbd_ctl_waitq);
197 sbi->ll_commitcbd_flags = 0;
198 err = ll_commitcbd_setup(sbi);
200 CERROR("failed to start commit callback daemon: rc = %d\n",err);
201 GOTO(out_request, sb = NULL);
204 lic.lic_body = lustre_msg_buf(request->rq_repmsg, 0);
206 LASSERT(sbi->ll_rootino != 0);
207 root = iget4(sb, sbi->ll_rootino, NULL, &lic);
210 sb->s_root = d_alloc_root(root);
212 CERROR("lustre_lite: bad iget4 for root\n");
213 GOTO(out_cdb, sb = NULL);
216 ptlrpc_req_finished(request);
218 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_read_super */
241 static void ll_put_super(struct super_block *sb)
243 struct ll_sb_info *sbi = ll_s2sbi(sb);
244 struct list_head *tmp, *next;
245 struct ll_fid rootfid;
248 list_del(&sbi->ll_conn_chain);
249 ll_commitcbd_cleanup(sbi);
250 obd_disconnect(&sbi->ll_osc_conn);
252 /* NULL request to force sync on the MDS, and get the last_committed
253 * value to flush remaining RPCs from the sending queue on client.
255 * XXX This should be an mdc_sync() call to sync the whole MDS fs,
256 * which we can call for other reasons as well.
258 mdc_getstatus(&sbi->ll_mdc_conn, &rootfid);
260 lprocfs_dereg_mnt(sbi->ll_proc_root);
261 sbi->ll_proc_root = NULL;
263 obd_disconnect(&sbi->ll_mdc_conn);
265 spin_lock(&dcache_lock);
266 list_for_each_safe(tmp, next, &sbi->ll_orphan_dentry_list) {
267 struct dentry *dentry = list_entry(tmp, struct dentry, d_hash);
268 shrink_dcache_parent(dentry);
270 spin_unlock(&dcache_lock);
272 OBD_FREE(sbi, sizeof(*sbi));
277 static void ll_clear_inode(struct inode *inode)
279 struct ll_sb_info *sbi = ll_i2sbi(inode);
280 struct ll_inode_info *lli = ll_i2info(inode);
284 rc = mdc_cancel_unused(&sbi->ll_mdc_conn, inode, LDLM_FL_NO_CALLBACK);
286 CERROR("mdc_cancel_unused: %d\n", rc);
287 /* XXX FIXME do something dramatic */
291 rc = obd_cancel_unused(&sbi->ll_osc_conn, lli->lli_smd, 0);
293 CERROR("obd_cancel_unused: %d\n", rc);
294 /* 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 obdo_from_inode(oa, inode, OBD_MD_FLID | OBD_MD_FLTYPE);
338 err = obd_destroy(ll_i2obdconn(inode), oa, lsm);
341 CDEBUG(D_SUPER, "obd destroy objid "LPX64" error %d\n",
342 lsm->lsm_object_id, err);
349 /* like inode_setattr, but doesn't mark the inode dirty */
350 static int ll_attr2inode(struct inode * inode, struct iattr * attr, int trunc)
352 unsigned int ia_valid = attr->ia_valid;
355 if ((ia_valid & ATTR_SIZE) && trunc) {
356 error = vmtruncate(inode, attr->ia_size);
359 } else if (ia_valid & ATTR_SIZE)
360 inode->i_size = attr->ia_size;
362 if (ia_valid & ATTR_UID)
363 inode->i_uid = attr->ia_uid;
364 if (ia_valid & ATTR_GID)
365 inode->i_gid = attr->ia_gid;
366 if (ia_valid & ATTR_ATIME)
367 inode->i_atime = attr->ia_atime;
368 if (ia_valid & ATTR_MTIME)
369 inode->i_mtime = attr->ia_mtime;
370 if (ia_valid & ATTR_CTIME)
371 inode->i_ctime = attr->ia_ctime;
372 if (ia_valid & ATTR_MODE) {
373 inode->i_mode = attr->ia_mode;
374 if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID))
375 inode->i_mode &= ~S_ISGID;
381 int ll_inode_setattr(struct inode *inode, struct iattr *attr, int do_trunc)
383 struct ptlrpc_request *request = NULL;
384 struct ll_sb_info *sbi = ll_i2sbi(inode);
388 /* change incore inode */
389 ll_attr2inode(inode, attr, do_trunc);
391 /* Don't send size changes to MDS to avoid "fast EA" problems, and
392 * also avoid a pointless RPC (we get file size from OST anyways).
394 attr->ia_valid &= ~ATTR_SIZE;
395 if (attr->ia_valid) {
396 err = mdc_setattr(&sbi->ll_mdc_conn, inode, attr, &request);
398 CERROR("mdc_setattr fails (%d)\n", err);
400 ptlrpc_req_finished(request);
406 int ll_setattr(struct dentry *de, struct iattr *attr)
408 int rc = inode_change_ok(de->d_inode, attr);
413 return ll_inode_setattr(de->d_inode, attr, 1);
416 static int ll_statfs(struct super_block *sb, struct statfs *sfs)
418 struct ll_sb_info *sbi = ll_s2sbi(sb);
419 struct obd_statfs osfs;
423 memset(sfs, 0, sizeof(*sfs));
424 rc = obd_statfs(&sbi->ll_mdc_conn, &osfs);
425 statfs_unpack(sfs, &osfs);
427 CERROR("mdc_statfs fails: rc = %d\n", rc);
429 CDEBUG(D_SUPER, "mdc_statfs shows blocks "LPU64"/"LPU64
430 " objects "LPU64"/"LPU64"\n",
431 osfs.os_bavail, osfs.os_blocks,
432 osfs.os_ffree, osfs.os_files);
434 /* temporary until mds_statfs returns statfs info for all OSTs */
436 rc = obd_statfs(&sbi->ll_osc_conn, &osfs);
438 CERROR("obd_statfs fails: rc = %d\n", rc);
441 CDEBUG(D_SUPER, "obd_statfs shows blocks "LPU64"/"LPU64
442 " objects "LPU64"/"LPU64"\n",
443 osfs.os_bavail, osfs.os_blocks,
444 osfs.os_ffree, osfs.os_files);
446 while (osfs.os_blocks > ~0UL) {
449 osfs.os_blocks >>= 1;
451 osfs.os_bavail >>= 1;
453 sfs->f_blocks = osfs.os_blocks;
454 sfs->f_bfree = osfs.os_bfree;
455 sfs->f_bavail = osfs.os_bavail;
456 if (osfs.os_ffree < (__u64)sfs->f_ffree)
457 sfs->f_ffree = osfs.os_ffree;
464 void ll_update_inode(struct inode *inode, struct mds_body *body)
466 if (body->valid & OBD_MD_FLID)
467 inode->i_ino = body->ino;
468 if (body->valid & OBD_MD_FLATIME)
469 inode->i_atime = body->atime;
470 if (body->valid & OBD_MD_FLMTIME)
471 inode->i_mtime = body->mtime;
472 if (body->valid & OBD_MD_FLCTIME)
473 inode->i_ctime = body->ctime;
474 if (body->valid & OBD_MD_FLMODE)
475 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
476 if (body->valid & OBD_MD_FLTYPE)
477 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
478 if (body->valid & OBD_MD_FLUID)
479 inode->i_uid = body->uid;
480 if (body->valid & OBD_MD_FLGID)
481 inode->i_gid = body->gid;
482 if (body->valid & OBD_MD_FLFLAGS)
483 inode->i_flags = body->flags;
484 if (body->valid & OBD_MD_FLNLINK)
485 inode->i_nlink = body->nlink;
486 if (body->valid & OBD_MD_FLGENER)
487 inode->i_generation = body->generation;
488 if (body->valid & OBD_MD_FLRDEV)
489 inode->i_rdev = body->rdev;
490 if (body->valid & OBD_MD_FLSIZE)
491 inode->i_size = body->size;
494 static void ll_read_inode2(struct inode *inode, void *opaque)
496 struct ll_read_inode2_cookie *lic = opaque;
497 struct mds_body *body = lic->lic_body;
498 struct ll_inode_info *lli = ll_i2info(inode);
501 sema_init(&lli->lli_open_sem, 1);
502 atomic_set(&lli->lli_open_count, 0);
504 /* core attributes first */
505 ll_update_inode(inode, body);
507 LASSERT(!lli->lli_smd);
508 if (lic && lic->lic_lmm)
509 obd_unpackmd(ll_i2obdconn(inode), &lli->lli_smd, lic->lic_lmm);
511 /* Get the authoritative file size */
512 if (lli->lli_smd && (inode->i_mode & S_IFREG)) {
514 LASSERT(lli->lli_smd->lsm_object_id != 0);
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);
523 /* OIDEBUG(inode); */
525 if (S_ISREG(inode->i_mode)) {
526 inode->i_op = &ll_file_inode_operations;
527 inode->i_fop = &ll_file_operations;
528 inode->i_mapping->a_ops = &ll_aops;
530 } else if (S_ISDIR(inode->i_mode)) {
531 inode->i_op = &ll_dir_inode_operations;
532 inode->i_fop = &ll_dir_operations;
533 inode->i_mapping->a_ops = &ll_dir_aops;
535 } else if (S_ISLNK(inode->i_mode)) {
536 inode->i_op = &ll_fast_symlink_inode_operations;
539 init_special_inode(inode, inode->i_mode, inode->i_rdev);
544 static inline void invalidate_request_list(struct list_head *req_list)
546 struct list_head *tmp, *n;
547 list_for_each_safe(tmp, n, req_list) {
548 struct ptlrpc_request *req =
549 list_entry(tmp, struct ptlrpc_request, rq_list);
550 CERROR("invalidating req xid "LPU64" op %d to %s:%d\n",
551 req->rq_xid, req->rq_reqmsg->opc,
552 req->rq_connection->c_remote_uuid,
553 req->rq_import->imp_client->cli_request_portal);
554 req->rq_flags |= PTL_RPC_FL_ERR;
555 wake_up(&req->rq_wait_for_rep);
559 void ll_umount_begin(struct super_block *sb)
561 struct ll_sb_info *sbi = ll_s2sbi(sb);
562 struct list_head *ctmp;
566 list_for_each(ctmp, &sbi->ll_conn_chain) {
567 struct ptlrpc_connection *conn;
568 conn = list_entry(ctmp, struct ptlrpc_connection, c_sb_chain);
570 spin_lock(&conn->c_lock);
571 /* XXX should just be dealing with imports, probably through
572 * XXX iocontrol, need next-gen recovery! */
573 conn->c_flags |= CONN_INVALID;
574 /* invalidate_request_list(&conn->c_sending_head); */
575 invalidate_request_list(&conn->c_delayed_head);
576 spin_unlock(&conn->c_lock);
582 /* exported operations */
583 struct super_operations ll_super_operations =
585 read_inode2: ll_read_inode2,
586 clear_inode: ll_clear_inode,
587 delete_inode: ll_delete_inode,
588 put_super: ll_put_super,
590 umount_begin: ll_umount_begin
593 static struct file_system_type lustre_lite_fs_type = {
596 read_super: ll_read_super,
600 static int __init init_lustre_lite(void)
602 printk(KERN_INFO "Lustre Lite 0.5.14, info@clusterfs.com\n");
603 ll_file_data_slab = kmem_cache_create("ll_file_data",
604 sizeof(struct ll_file_data), 0,
605 SLAB_HWCACHE_ALIGN, NULL, NULL);
606 if (ll_file_data_slab == NULL)
608 return register_filesystem(&lustre_lite_fs_type);
611 static void __exit exit_lustre_lite(void)
613 unregister_filesystem(&lustre_lite_fs_type);
614 kmem_cache_destroy(ll_file_data_slab);
617 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
618 MODULE_DESCRIPTION("Lustre Lite Client File System v1.0");
619 MODULE_LICENSE("GPL");
621 module_init(init_lustre_lite);
622 module_exit(exit_lustre_lite);