#include <linux/module.h>
#include <linux/random.h>
+#include <linux/version.h>
#include <linux/lustre_lite.h>
#include <linux/lustre_ha.h>
#include <linux/obd_lov.h>
#include <linux/lustre_dlm.h>
#include <linux/init.h>
+#include <linux/fs.h>
+#include <linux/lprocfs_status.h>
+
+#if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
kmem_cache_t *ll_file_data_slab;
extern struct address_space_operations ll_aops;
extern struct address_space_operations ll_dir_aops;
struct super_operations ll_super_operations;
-extern int ll_recover(struct ptlrpc_client *);
+extern int ll_recover(struct recovd_data *, int);
extern int ll_commitcbd_setup(struct ll_sb_info *);
extern int ll_commitcbd_cleanup(struct ll_sb_info *);
+extern int rd_dev_uuid(char* page, char **start, off_t off,
+ int count, int *eof, void *data);
+
+extern int rd_dev_name(char* page, char **start, off_t off,
+ int count, int *eof, void *data);
+
+
+extern lprocfs_vars_t status_var_nm_1[];
+
static char *ll_read_opt(const char *opt, char *data)
{
char *value;
CERROR("out of memory!\n");
RETURN(NULL);
}
-
+
memcpy(retval, value, strlen(value)+1);
CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
RETURN(retval);
}
-static void ll_options(char *options, char **ost, char **mds)
+static int ll_set_opt(const char *opt, char *data, int fl)
+{
+ ENTRY;
+
+ CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
+ if ( strncmp(opt, data, strlen(opt)) )
+ RETURN(0);
+ else
+ RETURN(fl);
+}
+
+static void ll_options(char *options, char **ost, char **mds, int *flags)
{
char *this_char;
ENTRY;
this_char = strtok (NULL, ",")) {
CDEBUG(D_SUPER, "this_char %s\n", this_char);
if ( (!*ost && (*ost = ll_read_opt("osc", this_char)))||
- (!*mds && (*mds = ll_read_opt("mdc", this_char))) )
+ (!*mds && (*mds = ll_read_opt("mdc", this_char)))||
+ (!(*flags & LL_SBI_NOLCK) && ((*flags) = (*flags) |
+ ll_set_opt("nolock", this_char, LL_SBI_NOLCK))) )
continue;
}
EXIT;
char *mdc = NULL;
int err;
struct ll_fid rootfid;
- struct statfs sfs;
- __u64 last_committed;
- __u32 last_xid;
+ struct obd_statfs osfs;
struct ptlrpc_request *request = NULL;
- struct ll_inode_md md;
+ struct ptlrpc_connection *mdc_conn;
+ struct ll_read_inode2_cookie lic;
class_uuid_t uuid;
+
+ /* Lprocfs variables */
+ char mnt_name[100];
+ char uuid_name[100];
+ lprocfs_vars_t d_vars[3];
+
+
+
ENTRY;
MOD_INC_USE_COUNT;
RETURN(NULL);
}
+ INIT_LIST_HEAD(&sbi->ll_conn_chain);
generate_random_uuid(uuid);
class_uuid_unparse(uuid, sbi->ll_sb_uuid);
sb->u.generic_sbp = sbi;
- ll_options(data, &osc, &mdc);
+ ll_options(data, &osc, &mdc, &sbi->ll_flags);
if (!osc) {
CERROR("no osc\n");
GOTO(out_free, sb = NULL);
}
- obd = class_uuid2obd(mdc);
+ obd = class_uuid2obd(mdc);
if (!obd) {
CERROR("MDC %s: not setup or attached\n", mdc);
GOTO(out_free, sb = NULL);
}
- err = obd_connect(&sbi->ll_mdc_conn, obd, sbi->ll_sb_uuid);
+ err = obd_connect(&sbi->ll_mdc_conn, obd, sbi->ll_sb_uuid,
+ ptlrpc_recovd, ll_recover);
if (err) {
CERROR("cannot connect to %s: rc = %d\n", mdc, err);
GOTO(out_free, sb = NULL);
}
- sbi2mdc(sbi)->cl_conn->c_level = LUSTRE_CONN_FULL;
+
+#warning Peter: is this the right place to raise the connection level?
+ mdc_conn = sbi2mdc(sbi)->cl_import.imp_connection;
+ mdc_conn->c_level = LUSTRE_CONN_FULL;
+ list_add(&mdc_conn->c_sb_chain, &sbi->ll_conn_chain);
obd = class_uuid2obd(osc);
if (!obd) {
CERROR("OSC %s: not setup or attached\n", osc);
GOTO(out_mdc, sb = NULL);
}
- err = obd_connect(&sbi->ll_osc_conn, obd, sbi->ll_sb_uuid);
+
+ err = obd_connect(&sbi->ll_osc_conn, obd, sbi->ll_sb_uuid,
+ ptlrpc_recovd, ll_recover);
if (err) {
CERROR("cannot connect to %s: rc = %d\n", osc, err);
GOTO(out_mdc, sb = NULL);
}
- /* XXX: need to store the last_* values somewhere */
- err = mdc_getstatus(&sbi->ll_mdc_conn, &rootfid, &last_committed,
- &last_xid, &request);
- ptlrpc_req_finished(request);
+ err = mdc_getstatus(&sbi->ll_mdc_conn, &rootfid);
if (err) {
CERROR("cannot mds_connect: rc = %d\n", err);
GOTO(out_mdc, sb = NULL);
}
- CDEBUG(D_SUPER, "rootfid %Ld\n", (unsigned long long)rootfid.id);
+ CDEBUG(D_SUPER, "rootfid "LPU64"\n", rootfid.id);
sbi->ll_rootino = rootfid.id;
- memset(&sfs, 0, sizeof(sfs));
- request = NULL;
- err = mdc_statfs(&sbi->ll_mdc_conn, &sfs, &request);
- ptlrpc_req_finished(request);
- sb->s_blocksize = sfs.f_bsize;
- sb->s_blocksize_bits = log2(sfs.f_bsize);
+ memset(&osfs, 0, sizeof(osfs));
+ err = obd_statfs(&sbi->ll_mdc_conn, &osfs);
+ sb->s_blocksize = osfs.os_bsize;
+ sb->s_blocksize_bits = log2(osfs.os_bsize);
sb->s_magic = LL_SUPER_MAGIC;
- sb->s_maxbytes = (1ULL << (32 + 9)) - sfs.f_bsize;
+ sb->s_maxbytes = (1ULL << (32 + 9)) - osfs.os_bsize;
sb->s_op = &ll_super_operations;
/* make root inode */
- request = NULL;
err = mdc_getattr(&sbi->ll_mdc_conn, sbi->ll_rootino, S_IFDIR,
OBD_MD_FLNOTOBD|OBD_MD_FLBLOCKS, 0, &request);
if (err) {
GOTO(out_request, sb = NULL);
}
- md.body = lustre_msg_buf(request->rq_repmsg, 0);
- md.md = NULL;
- root = iget4(sb, sbi->ll_rootino, NULL, &md);
+ lic.lic_body = lustre_msg_buf(request->rq_repmsg, 0);
+ lic.lic_lmm = NULL;
+ LASSERT(sbi->ll_rootino != 0);
+ root = iget4(sb, sbi->ll_rootino, NULL, &lic);
if (root) {
sb->s_root = d_alloc_root(root);
}
ptlrpc_req_finished(request);
+ request = NULL;
+
+
+ /* Register this mount instance with LProcFS */
+
+ snprintf(mnt_name, 100, "mount_%s", sbi->ll_sb_uuid);
+ sbi->ll_mnt_root=lprocfs_reg_mnt(mnt_name);
+ if(!sbi->ll_mnt_root)
+ goto out_dev;
+
+ /* Add the static configuration info */
+ lprocfs_add_vars(sbi->ll_mnt_root, (lprocfs_vars_t*)status_var_nm_1,
+ (void*)sb);
+
+ /* Add the dynamic configuration stuff */
+ /* MDC */
+ obd = class_uuid2obd(mdc);
+
+ /* Reuse mnt_name */
+
+ sprintf(mnt_name, "status/%s/common_name", obd->obd_type->typ_name);
+
+ memset(d_vars, 0, sizeof(d_vars));
+ d_vars[0].read_fptr=rd_dev_name;
+ d_vars[0].write_fptr=0;
+ d_vars[0].name=(char*)mnt_name;
+
+ memset(uuid_name, '\0', strlen(uuid_name));
+ sprintf(uuid_name, "status/%s/uuid", obd->obd_type->typ_name);
+ d_vars[1].read_fptr=rd_dev_uuid;
+ d_vars[1].write_fptr=0;
+ d_vars[1].name=(char*)uuid_name;
+
+ err=lprocfs_add_vars(sbi->ll_mnt_root, (lprocfs_vars_t*)d_vars,
+ (void*)obd);
+ if (err) {
+ CDEBUG(D_OTHER, "Unable to add fs proc dynamic variables");
+ }
+ /* OSC or LOV*/
+ obd = class_uuid2obd(osc);
+ /* Reuse mnt_name */
+ memset(mnt_name, '\0', strlen(mnt_name));
+ sprintf(mnt_name, "status/%s/common_name", obd->obd_type->typ_name);
+
+ memset(d_vars, 0, sizeof(d_vars));
+ d_vars[0].read_fptr=rd_dev_name;
+ d_vars[0].write_fptr=0;
+ d_vars[0].name=(char*)mnt_name;
+
+ memset(uuid_name, '\0', strlen(uuid_name));
+ sprintf(uuid_name, "status/%s/uuid", obd->obd_type->typ_name);
+ d_vars[1].read_fptr=rd_dev_uuid;
+ d_vars[1].write_fptr=0;
+ d_vars[1].name=(char*)uuid_name;
+
+ err=lprocfs_add_vars(sbi->ll_mnt_root, (lprocfs_vars_t*)d_vars,
+ (void*)obd);
+ if (err) {
+ CDEBUG(D_OTHER, "Unable to add fs proc dynamic variables");
+ }
out_dev:
if (mdc)
OBD_FREE(mdc, strlen(mdc) + 1);
static void ll_put_super(struct super_block *sb)
{
struct ll_sb_info *sbi = ll_s2sbi(sb);
+ struct ll_fid rootfid;
ENTRY;
+
+ list_del(&sbi->ll_conn_chain);
ll_commitcbd_cleanup(sbi);
obd_disconnect(&sbi->ll_osc_conn);
+
+ /* NULL request to force sync on the MDS, and get the last_committed
+ * value to flush remaining RPCs from the sending queue on client.
+ *
+ * XXX This should be an mdc_sync() call to sync the whole MDS fs,
+ * which we can call for other reasons as well.
+ */
+ mdc_getstatus(&sbi->ll_mdc_conn, &rootfid);
+
+ lprocfs_dereg_mnt(sbi->ll_mnt_root);
obd_disconnect(&sbi->ll_mdc_conn);
OBD_FREE(sbi, sizeof(*sbi));
static void ll_clear_inode(struct inode *inode)
{
+ struct ll_sb_info *sbi = ll_i2sbi(inode);
+ int rc;
+ ENTRY;
+
+ rc = mdc_cancel_unused(&sbi->ll_mdc_conn, inode, LDLM_FL_NO_CALLBACK);
+ if (rc < 0) {
+ CERROR("obd_cancel_unused: %d\n", rc);
+ /* XXX FIXME do something dramatic */
+ }
+
if (atomic_read(&inode->i_count) == 0) {
struct ll_inode_info *lli = ll_i2info(inode);
- struct lov_stripe_md *md = lli->lli_smd;
+ struct lov_stripe_md *lsm = lli->lli_smd;
char *symlink_name = lli->lli_symlink_name;
- if (md) {
- int size = sizeof(*md) +
- md->lmd_stripe_count * sizeof(struct lov_oinfo);
- OBD_FREE(md, size);
+ if (lsm) {
+ OBD_FREE(lsm, ll_ost_easize(inode->i_sb));
lli->lli_smd = NULL;
}
if (symlink_name) {
lli->lli_symlink_name = NULL;
}
}
+
+ EXIT;
}
static void ll_delete_inode(struct inode *inode)
{
+ ENTRY;
if (S_ISREG(inode->i_mode)) {
int err;
struct obdo *oa;
- struct lov_stripe_md *md = ll_i2info(inode)->lli_smd;
+ struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
- if (!md)
+ if (!lsm)
GOTO(out, -EINVAL);
- if (md->lmd_object_id == 0) {
+ if (lsm->lsm_object_id == 0) {
CERROR("This really happens\n");
/* No obdo was ever created */
GOTO(out, 0);
if (oa == NULL)
GOTO(out, -ENOMEM);
- oa->o_id = md->lmd_object_id;
- oa->o_easize = md->lmd_easize;
+ oa->o_id = lsm->lsm_object_id;
+ oa->o_easize = ll_mds_easize(inode->i_sb);
oa->o_mode = inode->i_mode;
- oa->o_valid = OBD_MD_FLID | OBD_MD_FLEASIZE | OBD_MD_FLMODE;
+ oa->o_valid = OBD_MD_FLID | OBD_MD_FLEASIZE | OBD_MD_FLTYPE;
- err = obd_destroy(ll_i2obdconn(inode), oa, md);
+ err = obd_destroy(ll_i2obdconn(inode), oa, lsm);
obdo_free(oa);
- CDEBUG(D_SUPER, "obd destroy of %Ld error %d\n",
- md->lmd_object_id, err);
+ CDEBUG(D_SUPER, "obd destroy of objid "LPX64" error %d\n",
+ lsm->lsm_object_id, err);
}
out:
clear_inode(inode);
+ EXIT;
}
/* like inode_setattr, but doesn't mark the inode dirty */
static int ll_statfs(struct super_block *sb, struct statfs *sfs)
{
- struct ptlrpc_request *request = NULL;
struct ll_sb_info *sbi = ll_s2sbi(sb);
+ struct obd_statfs osfs;
int rc;
ENTRY;
memset(sfs, 0, sizeof(*sfs));
- rc = mdc_statfs(&sbi->ll_mdc_conn, sfs, &request);
- ptlrpc_req_finished(request);
+ rc = obd_statfs(&sbi->ll_mdc_conn, &osfs);
+ statfs_unpack(sfs, &osfs);
if (rc)
- CERROR("obd_statfs fails: rc = %d\n", rc);
+ CERROR("mdc_statfs fails: rc = %d\n", rc);
else
- CDEBUG(D_SUPER, "statfs shows blocks %ld/%ld objects %ld/%ld\n",
- sfs->f_bavail, sfs->f_blocks, sfs->f_files,sfs->f_ffree);
+ CDEBUG(D_SUPER, "mdc_statfs shows blocks "LPU64"/"LPU64
+ " objects "LPU64"/"LPU64"\n",
+ osfs.os_bavail, osfs.os_blocks,
+ osfs.os_ffree, osfs.os_files);
/* temporary until mds_statfs returns statfs info for all OSTs */
if (!rc) {
- struct statfs obd_sfs;
-
- rc = obd_statfs(&sbi->ll_osc_conn, &obd_sfs);
+ rc = obd_statfs(&sbi->ll_osc_conn, &osfs);
if (rc) {
CERROR("obd_statfs fails: rc = %d\n", rc);
GOTO(out, rc);
}
- CDEBUG(D_SUPER, "obd_statfs returns blocks %ld/%ld, "
- "objects %ld/%ld\n",
- obd_sfs.f_bavail, obd_sfs.f_blocks,
- obd_sfs.f_ffree, obd_sfs.f_files);
-
- sfs->f_bfree = obd_sfs.f_bfree;
- sfs->f_bavail = obd_sfs.f_bavail;
- sfs->f_blocks = obd_sfs.f_blocks;
- if (obd_sfs.f_ffree < sfs->f_ffree)
- sfs->f_ffree = obd_sfs.f_ffree;
+ CDEBUG(D_SUPER, "obd_statfs shows blocks "LPU64"/"LPU64
+ " objects "LPU64"/"LPU64"\n",
+ osfs.os_bavail, osfs.os_blocks,
+ osfs.os_ffree, osfs.os_files);
+
+ while (osfs.os_blocks > ~0UL) {
+ sfs->f_bsize <<= 1;
+
+ osfs.os_blocks >>= 1;
+ osfs.os_bfree >>= 1;
+ osfs.os_bavail >>= 1;
+ }
+ sfs->f_blocks = osfs.os_blocks;
+ sfs->f_bfree = osfs.os_bfree;
+ sfs->f_bavail = osfs.os_bavail;
+ if (osfs.os_ffree < (__u64)sfs->f_ffree)
+ sfs->f_ffree = osfs.os_ffree;
}
out:
RETURN(rc);
}
-inline int ll_stripe_mds_md_size(struct super_block *sb)
+void ll_update_inode(struct inode *inode, struct mds_body *body)
{
- struct client_obd *mdc = sbi2mdc(ll_s2sbi(sb));
- return mdc->cl_max_mdsize;
-}
-
-static void ll_read_inode2(struct inode *inode, void *opaque)
-{
- struct ll_inode_md *md = opaque;
- struct mds_body *body = md->body;
- struct ll_inode_info *lli = ll_i2info(inode);
- ENTRY;
-
- sema_init(&lli->lli_open_sem, 1);
-
- /* core attributes first */
if (body->valid & OBD_MD_FLID)
inode->i_ino = body->ino;
if (body->valid & OBD_MD_FLATIME)
if (body->valid & OBD_MD_FLCTIME)
inode->i_ctime = body->ctime;
if (body->valid & OBD_MD_FLMODE)
- inode->i_mode = body->mode;
+ inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
+ if (body->valid & OBD_MD_FLTYPE)
+ inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
if (body->valid & OBD_MD_FLUID)
inode->i_uid = body->uid;
if (body->valid & OBD_MD_FLGID)
if (body->valid & OBD_MD_FLGENER)
inode->i_generation = body->generation;
if (body->valid & OBD_MD_FLRDEV)
- inode->i_rdev = body->extra;
+ inode->i_rdev = body->rdev;
if (body->valid & OBD_MD_FLSIZE)
inode->i_size = body->size;
+}
+static void ll_read_inode2(struct inode *inode, void *opaque)
+{
+ struct ll_read_inode2_cookie *lic = opaque;
+ struct mds_body *body = lic->lic_body;
+ struct ll_inode_info *lli = ll_i2info(inode);
+ ENTRY;
+
+ sema_init(&lli->lli_open_sem, 1);
+
+ /* core attributes first */
+ ll_update_inode(inode, body);
//if (body->valid & OBD_MD_FLEASIZE)
- if (md && md->md && md->md->lmd_stripe_count) {
- struct lov_mds_md *smd = md->md;
+ if (lic && lic->lic_lmm) {
+ struct lov_mds_md *lmm = lic->lic_lmm;
int size;
- if (md->md->lmd_easize != ll_stripe_mds_md_size(inode->i_sb)) {
+
+ /* XXX This should probably not be an error in the future,
+ * when we allow LOV OSTs to be added.
+ */
+ if (lmm->lmm_easize != ll_mds_easize(inode->i_sb)) {
CERROR("Striping metadata size error %ld\n",
inode->i_ino);
LBUG();
}
- size = sizeof(*lli->lli_smd) +
- md->md->lmd_stripe_count * sizeof(struct lov_oinfo);
+ size = ll_ost_easize(inode->i_sb);
OBD_ALLOC(lli->lli_smd, size);
if (!lli->lli_smd) {
CERROR("No memory for %d\n", size);
LBUG();
}
- lov_unpackmd(lli->lli_smd, smd);
+ lov_unpackmd(lli->lli_smd, lmm);
} else {
lli->lli_smd = NULL;
}
init_special_inode(inode, inode->i_mode, inode->i_rdev);
EXIT;
}
+}
- return;
+static inline void invalidate_request_list(struct list_head *req_list)
+{
+ struct list_head *tmp, *n;
+ list_for_each_safe(tmp, n, req_list) {
+ struct ptlrpc_request *req =
+ list_entry(tmp, struct ptlrpc_request, rq_list);
+ CERROR("invalidating req xid "LPD64" op %d to %s:%d\n",
+ (unsigned long long)req->rq_xid, req->rq_reqmsg->opc,
+ req->rq_connection->c_remote_uuid,
+ req->rq_import->imp_client->cli_request_portal);
+ req->rq_flags |= PTL_RPC_FL_ERR;
+ wake_up(&req->rq_wait_for_rep);
+ }
+}
+
+void ll_umount_begin(struct super_block *sb)
+{
+ struct ll_sb_info *sbi = ll_s2sbi(sb);
+ struct list_head *ctmp;
+
+ ENTRY;
+
+ list_for_each(ctmp, &sbi->ll_conn_chain) {
+ struct ptlrpc_connection *conn;
+ conn = list_entry(ctmp, struct ptlrpc_connection, c_sb_chain);
+
+ spin_lock(&conn->c_lock);
+ /* XXX should just be dealing with imports, probably through
+ * XXX iocontrol, need next-gen recovery! */
+ conn->c_flags |= CONN_INVALID;
+ invalidate_request_list(&conn->c_sending_head);
+ invalidate_request_list(&conn->c_delayed_head);
+ spin_unlock(&conn->c_lock);
+ }
+
+ EXIT;
}
/* exported operations */
clear_inode: ll_clear_inode,
delete_inode: ll_delete_inode,
put_super: ll_put_super,
- statfs: ll_statfs
+ statfs: ll_statfs,
+ umount_begin: ll_umount_begin
};
struct file_system_type lustre_lite_fs_type = {
static int __init init_lustre_lite(void)
{
- printk(KERN_INFO "Lustre Lite 0.0.1, info@clusterfs.com\n");
+ printk(KERN_INFO "Lustre Lite 0.5.14, info@clusterfs.com\n");
ll_file_data_slab = kmem_cache_create("ll_file_data",
sizeof(struct ll_file_data), 0,
- SLAB_HWCACHE_ALIGN, NULL, NULL);
+ SLAB_HWCACHE_ALIGN, NULL, NULL);
if (ll_file_data_slab == NULL)
return -ENOMEM;
return register_filesystem(&lustre_lite_fs_type);
module_init(init_lustre_lite);
module_exit(exit_lustre_lite);
+#endif