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 = NULL;
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_recover(struct recovd_data *, int);
50 extern int ll_commitcbd_setup(struct ll_sb_info *);
51 extern int ll_commitcbd_cleanup(struct ll_sb_info *);
53 static char *ll_read_opt(const char *opt, char *data)
59 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
60 if ( strncmp(opt, data, strlen(opt)) )
62 if ( (value = strchr(data, '=')) == NULL )
66 OBD_ALLOC(retval, strlen(value) + 1);
68 CERROR("out of memory!\n");
72 memcpy(retval, value, strlen(value)+1);
73 CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
77 static int ll_set_opt(const char *opt, char *data, int fl)
81 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
82 if ( strncmp(opt, data, strlen(opt)) )
88 static void ll_options(char *options, char **ost, char **mds, int *flags)
98 for (this_char = strtok (options, ",");
100 this_char = strtok (NULL, ",")) {
101 CDEBUG(D_SUPER, "this_char %s\n", this_char);
102 if ( (!*ost && (*ost = ll_read_opt("osc", this_char)))||
103 (!*mds && (*mds = ll_read_opt("mdc", this_char)))||
104 (!(*flags & LL_SBI_NOLCK) && ((*flags) = (*flags) |
105 ll_set_opt("nolock", this_char, LL_SBI_NOLCK))) )
112 #define log2(n) ffz(~(n))
115 static struct super_block *ll_read_super(struct super_block *sb,
116 void *data, int silent)
118 struct inode *root = 0;
119 struct obd_device *obd;
120 struct ll_sb_info *sbi;
124 struct ll_fid rootfid;
125 struct obd_statfs osfs;
126 struct ptlrpc_request *request = NULL;
127 struct ptlrpc_connection *mdc_conn;
128 struct ll_read_inode2_cookie lic;
130 struct obd_uuid param_uuid;
134 CDEBUG(D_VFSTRACE, "VFS Op\n");
135 OBD_ALLOC(sbi, sizeof(*sbi));
139 INIT_LIST_HEAD(&sbi->ll_conn_chain);
140 INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
141 generate_random_uuid(uuid);
142 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
144 sb->u.generic_sbp = sbi;
146 ll_options(data, &osc, &mdc, &sbi->ll_flags);
150 GOTO(out_free, sb = NULL);
155 GOTO(out_free, sb = NULL);
158 strncpy(param_uuid.uuid, mdc, sizeof(param_uuid.uuid));
159 obd = class_uuid2obd(¶m_uuid);
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 strncpy(param_uuid.uuid, osc, sizeof(param_uuid.uuid));
176 obd = class_uuid2obd(¶m_uuid);
178 CERROR("OSC %s: not setup or attached\n", osc);
179 GOTO(out_mdc, sb = NULL);
182 err = obd_connect(&sbi->ll_osc_conn, obd, &sbi->ll_sb_uuid,
183 ptlrpc_recovd, ll_recover);
185 CERROR("cannot connect to %s: rc = %d\n", osc, err);
186 GOTO(out_mdc, sb = NULL);
189 err = mdc_getstatus(&sbi->ll_mdc_conn, &rootfid);
191 CERROR("cannot mds_connect: rc = %d\n", err);
192 GOTO(out_mdc, sb = NULL);
194 CDEBUG(D_SUPER, "rootfid "LPU64"\n", rootfid.id);
195 sbi->ll_rootino = rootfid.id;
197 memset(&osfs, 0, sizeof(osfs));
198 err = obd_statfs(&sbi->ll_mdc_conn, &osfs);
199 sb->s_blocksize = osfs.os_bsize;
200 sb->s_blocksize_bits = log2(osfs.os_bsize);
201 sb->s_magic = LL_SUPER_MAGIC;
202 sb->s_maxbytes = (1ULL << (32 + 9)) - osfs.os_bsize;
204 sb->s_op = &ll_super_operations;
206 /* make root inode */
207 err = mdc_getattr(&sbi->ll_mdc_conn, sbi->ll_rootino, S_IFDIR,
208 OBD_MD_FLNOTOBD|OBD_MD_FLBLOCKS, 0, &request);
210 CERROR("mdc_getattr failed for root: rc = %d\n", err);
211 GOTO(out_request, sb = NULL);
214 /* initialize committed transaction callback daemon */
215 spin_lock_init(&sbi->ll_commitcbd_lock);
216 init_waitqueue_head(&sbi->ll_commitcbd_waitq);
217 init_waitqueue_head(&sbi->ll_commitcbd_ctl_waitq);
218 sbi->ll_commitcbd_flags = 0;
219 err = ll_commitcbd_setup(sbi);
221 CERROR("failed to start commit callback daemon: rc = %d\n",err);
222 GOTO(out_request, sb = NULL);
225 lic.lic_body = lustre_msg_buf(request->rq_repmsg, 0);
227 LASSERT(sbi->ll_rootino != 0);
228 root = iget4(sb, sbi->ll_rootino, NULL, &lic);
231 sb->s_root = d_alloc_root(root);
233 CERROR("lustre_lite: bad iget4 for root\n");
234 GOTO(out_cdb, sb = NULL);
237 ptlrpc_req_finished(request);
240 if (proc_lustre_fs_root) {
241 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
244 CERROR("could not register mount in /proc/lustre");
249 OBD_FREE(mdc, strlen(mdc) + 1);
251 OBD_FREE(osc, strlen(osc) + 1);
256 ll_commitcbd_cleanup(sbi);
258 ptlrpc_req_finished(request);
259 obd_disconnect(&sbi->ll_osc_conn);
261 obd_disconnect(&sbi->ll_mdc_conn);
263 OBD_FREE(sbi, sizeof(*sbi));
266 } /* ll_read_super */
268 static void ll_put_super(struct super_block *sb)
270 struct ll_sb_info *sbi = ll_s2sbi(sb);
271 struct list_head *tmp, *next;
272 struct ll_fid rootfid;
275 CDEBUG(D_VFSTRACE, "VFS Op\n");
276 list_del(&sbi->ll_conn_chain);
277 ll_commitcbd_cleanup(sbi);
278 obd_disconnect(&sbi->ll_osc_conn);
280 /* NULL request to force sync on the MDS, and get the last_committed
281 * value to flush remaining RPCs from the sending queue on client.
283 * XXX This should be an mdc_sync() call to sync the whole MDS fs,
284 * which we can call for other reasons as well.
286 mdc_getstatus(&sbi->ll_mdc_conn, &rootfid);
288 if (sbi->ll_proc_root) {
289 lprocfs_remove(sbi->ll_proc_root);
290 sbi->ll_proc_root = NULL;
293 obd_disconnect(&sbi->ll_mdc_conn);
295 spin_lock(&dcache_lock);
296 list_for_each_safe(tmp, next, &sbi->ll_orphan_dentry_list) {
297 struct dentry *dentry = list_entry(tmp, struct dentry, d_hash);
298 shrink_dcache_parent(dentry);
300 spin_unlock(&dcache_lock);
302 OBD_FREE(sbi, sizeof(*sbi));
307 static void ll_clear_inode(struct inode *inode)
309 struct ll_sb_info *sbi = ll_i2sbi(inode);
310 struct ll_inode_info *lli = ll_i2info(inode);
314 CDEBUG(D_VFSTRACE, "VFS Op\n");
315 rc = mdc_cancel_unused(&sbi->ll_mdc_conn, inode, LDLM_FL_NO_CALLBACK);
317 CERROR("mdc_cancel_unused: %d\n", rc);
318 /* XXX FIXME do something dramatic */
322 rc = obd_cancel_unused(&sbi->ll_osc_conn, lli->lli_smd, 0);
324 CERROR("obd_cancel_unused: %d\n", rc);
325 /* XXX FIXME do something dramatic */
329 if (atomic_read(&inode->i_count) != 0)
330 CERROR("clearing in-use inode %lu: count = %d\n",
331 inode->i_ino, atomic_read(&inode->i_count));
334 obd_free_memmd(&sbi->ll_osc_conn, &lli->lli_smd);
336 if (lli->lli_symlink_name) {
337 OBD_FREE(lli->lli_symlink_name,
338 strlen(lli->lli_symlink_name) + 1);
339 lli->lli_symlink_name = NULL;
346 static void ll_delete_inode(struct inode *inode)
349 CDEBUG(D_VFSTRACE, "VFS Op\n");
350 if (S_ISREG(inode->i_mode)) {
353 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
355 /* mcreate with no open */
359 if (lsm->lsm_object_id == 0) {
360 CERROR("This really happens\n");
361 /* No obdo was ever created */
369 oa->o_id = lsm->lsm_object_id;
370 obdo_from_inode(oa, inode, OBD_MD_FLID | OBD_MD_FLTYPE);
372 err = obd_destroy(ll_i2obdconn(inode), oa, lsm, NULL);
376 "inode %lu obd_destroy objid "LPX64" error %d\n",
377 inode->i_ino, lsm->lsm_object_id, err);
385 /* like inode_setattr, but doesn't mark the inode dirty */
386 static int ll_attr2inode(struct inode *inode, struct iattr *attr, int trunc)
388 unsigned int ia_valid = attr->ia_valid;
391 if ((ia_valid & ATTR_SIZE) && trunc) {
392 error = vmtruncate(inode, attr->ia_size);
395 } else if (ia_valid & ATTR_SIZE)
396 inode->i_size = attr->ia_size;
398 if (ia_valid & ATTR_UID)
399 inode->i_uid = attr->ia_uid;
400 if (ia_valid & ATTR_GID)
401 inode->i_gid = attr->ia_gid;
402 if (ia_valid & ATTR_ATIME)
403 inode->i_atime = attr->ia_atime;
404 if (ia_valid & ATTR_MTIME)
405 inode->i_mtime = attr->ia_mtime;
406 if (ia_valid & ATTR_CTIME)
407 inode->i_ctime = attr->ia_ctime;
408 if (ia_valid & ATTR_MODE) {
409 inode->i_mode = attr->ia_mode;
410 if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID))
411 inode->i_mode &= ~S_ISGID;
417 int ll_inode_setattr(struct inode *inode, struct iattr *attr, int do_trunc)
419 struct ptlrpc_request *request = NULL;
420 struct ll_sb_info *sbi = ll_i2sbi(inode);
424 /* change incore inode */
425 ll_attr2inode(inode, attr, do_trunc);
427 /* Don't send size changes to MDS to avoid "fast EA" problems, and
428 * also avoid a pointless RPC (we get file size from OST anyways).
430 attr->ia_valid &= ~ATTR_SIZE;
431 if (attr->ia_valid) {
432 err = mdc_setattr(&sbi->ll_mdc_conn, inode, attr, NULL, 0,
435 CERROR("mdc_setattr fails: err = %d\n", err);
437 ptlrpc_req_finished(request);
438 if (S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_MTIME_SET) {
439 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
443 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
444 inode->i_ino, attr->ia_mtime);
445 oa.o_id = lsm->lsm_object_id;
447 oa.o_valid = OBD_MD_FLID |OBD_MD_FLTYPE |OBD_MD_FLMTIME;
448 oa.o_mtime = attr->ia_mtime;
449 err2 = obd_setattr(&sbi->ll_osc_conn, &oa, lsm, NULL);
451 CERROR("obd_setattr fails: rc=%d\n", err);
461 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
463 struct ptlrpc_request *request = NULL;
464 struct ll_sb_info *sbi = ll_i2sbi(inode);
468 if ((attr->ia_valid & ATTR_SIZE)) {
469 err = vmtruncate(inode, attr->ia_size);
474 /* Don't send size changes to MDS to avoid "fast EA" problems, and
475 * also avoid a pointless RPC (we get file size from OST anyways).
477 attr->ia_valid &= ~ATTR_SIZE;
481 err = mdc_setattr(&sbi->ll_mdc_conn, inode, attr, NULL, 0,
484 CERROR("mdc_setattr fails: err = %d\n", err);
486 ptlrpc_req_finished(request);
488 if (S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_MTIME_SET) {
489 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
493 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
494 inode->i_ino, attr->ia_mtime);
495 oa.o_id = lsm->lsm_object_id;
497 oa.o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLMTIME;
498 oa.o_mtime = attr->ia_mtime;
499 err2 = obd_setattr(&sbi->ll_osc_conn, &oa, lsm, NULL);
501 CERROR("obd_setattr fails: rc=%d\n", err);
509 int ll_setattr(struct dentry *de, struct iattr *attr)
511 int rc = inode_change_ok(de->d_inode, attr);
513 CDEBUG(D_VFSTRACE, "VFS Op\n");
517 return ll_inode_setattr(de->d_inode, attr, 1);
520 static int ll_statfs(struct super_block *sb, struct statfs *sfs)
522 struct ll_sb_info *sbi = ll_s2sbi(sb);
523 struct obd_statfs osfs;
527 CDEBUG(D_VFSTRACE, "VFS Op\n");
528 memset(sfs, 0, sizeof(*sfs));
529 rc = obd_statfs(&sbi->ll_mdc_conn, &osfs);
530 statfs_unpack(sfs, &osfs);
532 CERROR("mdc_statfs fails: rc = %d\n", rc);
534 CDEBUG(D_SUPER, "mdc_statfs shows blocks "LPU64"/"LPU64
535 " objects "LPU64"/"LPU64"\n",
536 osfs.os_bavail, osfs.os_blocks,
537 osfs.os_ffree, osfs.os_files);
539 /* temporary until mds_statfs returns statfs info for all OSTs */
541 rc = obd_statfs(&sbi->ll_osc_conn, &osfs);
543 CERROR("obd_statfs fails: rc = %d\n", rc);
546 CDEBUG(D_SUPER, "obd_statfs shows blocks "LPU64"/"LPU64
547 " objects "LPU64"/"LPU64"\n",
548 osfs.os_bavail, osfs.os_blocks,
549 osfs.os_ffree, osfs.os_files);
551 while (osfs.os_blocks > ~0UL) {
554 osfs.os_blocks >>= 1;
556 osfs.os_bavail >>= 1;
558 sfs->f_blocks = osfs.os_blocks;
559 sfs->f_bfree = osfs.os_bfree;
560 sfs->f_bavail = osfs.os_bavail;
561 if (osfs.os_ffree < (__u64)sfs->f_ffree)
562 sfs->f_ffree = osfs.os_ffree;
569 void ll_update_inode(struct inode *inode, struct mds_body *body,
570 struct lov_mds_md *lmm)
572 struct ll_inode_info *lli = ll_i2info(inode);
575 obd_unpackmd(ll_i2obdconn(inode), &lli->lli_smd, lmm);
577 if (body->valid & OBD_MD_FLID)
578 inode->i_ino = body->ino;
579 if (body->valid & OBD_MD_FLATIME)
580 inode->i_atime = body->atime;
581 if (body->valid & OBD_MD_FLMTIME)
582 inode->i_mtime = body->mtime;
583 if (body->valid & OBD_MD_FLCTIME)
584 inode->i_ctime = body->ctime;
585 if (body->valid & OBD_MD_FLMODE)
586 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
587 if (body->valid & OBD_MD_FLTYPE)
588 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
589 if (body->valid & OBD_MD_FLUID)
590 inode->i_uid = body->uid;
591 if (body->valid & OBD_MD_FLGID)
592 inode->i_gid = body->gid;
593 if (body->valid & OBD_MD_FLFLAGS)
594 inode->i_flags = body->flags;
595 if (body->valid & OBD_MD_FLNLINK)
596 inode->i_nlink = body->nlink;
597 if (body->valid & OBD_MD_FLGENER)
598 inode->i_generation = body->generation;
599 if (body->valid & OBD_MD_FLRDEV)
600 inode->i_rdev = body->rdev;
601 if (body->valid & OBD_MD_FLSIZE)
602 inode->i_size = body->size;
603 if (body->valid & OBD_MD_FLBLOCKS)
604 inode->i_blocks = body->blocks;
607 static void ll_read_inode2(struct inode *inode, void *opaque)
609 struct ll_read_inode2_cookie *lic = opaque;
610 struct mds_body *body = lic->lic_body;
611 struct ll_inode_info *lli = ll_i2info(inode);
614 CDEBUG(D_VFSTRACE, "VFS Op\n");
615 sema_init(&lli->lli_open_sem, 1);
617 LASSERT(!lli->lli_smd);
619 /* core attributes first */
620 ll_update_inode(inode, body, lic ? lic->lic_lmm : NULL);
622 /* Get the authoritative file size */
623 if (lli->lli_smd && (inode->i_mode & S_IFREG)) {
625 LASSERT(lli->lli_smd->lsm_object_id != 0);
626 rc = ll_file_size(inode, lli->lli_smd, NULL);
628 CERROR("ll_file_size: %d\n", rc);
629 ll_clear_inode(inode);
630 make_bad_inode(inode);
634 /* OIDEBUG(inode); */
636 if (S_ISREG(inode->i_mode)) {
637 inode->i_op = &ll_file_inode_operations;
638 inode->i_fop = &ll_file_operations;
639 inode->i_mapping->a_ops = &ll_aops;
641 } else if (S_ISDIR(inode->i_mode)) {
642 inode->i_op = &ll_dir_inode_operations;
643 inode->i_fop = &ll_dir_operations;
644 inode->i_mapping->a_ops = &ll_dir_aops;
646 } else if (S_ISLNK(inode->i_mode)) {
647 inode->i_op = &ll_fast_symlink_inode_operations;
650 inode->i_op = &ll_special_inode_operations;
651 init_special_inode(inode, inode->i_mode, inode->i_rdev);
656 static inline void invalidate_request_list(struct list_head *req_list)
658 struct list_head *tmp, *n;
659 list_for_each_safe(tmp, n, req_list) {
660 struct ptlrpc_request *req =
661 list_entry(tmp, struct ptlrpc_request, rq_list);
662 CERROR("invalidating req xid "LPU64" op %d to %s:%d\n",
663 req->rq_xid, req->rq_reqmsg->opc,
664 req->rq_connection->c_remote_uuid.uuid,
665 req->rq_import->imp_client->cli_request_portal);
666 req->rq_flags |= PTL_RPC_FL_ERR;
667 wake_up(&req->rq_wait_for_rep);
671 void ll_umount_begin(struct super_block *sb)
673 struct ll_sb_info *sbi = ll_s2sbi(sb);
674 struct list_head *ctmp;
677 CDEBUG(D_VFSTRACE, "VFS Op\n");
679 list_for_each(ctmp, &sbi->ll_conn_chain) {
680 struct ptlrpc_connection *conn;
681 conn = list_entry(ctmp, struct ptlrpc_connection, c_sb_chain);
683 spin_lock(&conn->c_lock);
684 /* XXX should just be dealing with imports, probably through
685 * XXX iocontrol, need next-gen recovery! */
686 conn->c_flags |= CONN_INVALID;
687 /* invalidate_request_list(&conn->c_sending_head); */
688 invalidate_request_list(&conn->c_delayed_head);
689 spin_unlock(&conn->c_lock);
695 /* exported operations */
696 struct super_operations ll_super_operations =
698 read_inode2: ll_read_inode2,
699 clear_inode: ll_clear_inode,
700 // delete_inode: ll_delete_inode,
701 put_super: ll_put_super,
703 umount_begin: ll_umount_begin
706 static struct file_system_type lustre_lite_fs_type = {
709 read_super: ll_read_super,
713 static int __init init_lustre_lite(void)
715 printk(KERN_INFO "Lustre Lite Client File System; "
716 "info@clusterfs.com\n");
717 ll_file_data_slab = kmem_cache_create("ll_file_data",
718 sizeof(struct ll_file_data), 0,
719 SLAB_HWCACHE_ALIGN, NULL, NULL);
720 if (ll_file_data_slab == NULL)
723 proc_lustre_fs_root = proc_lustre_root ? proc_mkdir("llite", proc_lustre_root) : NULL;
725 return register_filesystem(&lustre_lite_fs_type);
728 static void __exit exit_lustre_lite(void)
730 unregister_filesystem(&lustre_lite_fs_type);
731 kmem_cache_destroy(ll_file_data_slab);
733 if (proc_lustre_fs_root) {
734 lprocfs_remove(proc_lustre_fs_root);
735 proc_lustre_fs_root = NULL;
739 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
740 MODULE_DESCRIPTION("Lustre Lite Client File System");
741 MODULE_LICENSE("GPL");
743 module_init(init_lustre_lite);
744 module_exit(exit_lustre_lite);