#include <linux/init.h>
#include <linux/fs.h>
#include <linux/lprocfs_status.h>
+#include "llite_internal.h"
#if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
#include <asm/statfs.h>
struct obd_uuid param_uuid;
ENTRY;
+ CDEBUG(D_VFSTRACE, "VFS Op:\n");
OBD_ALLOC(sbi, sizeof(*sbi));
if (!sbi)
GOTO(out_free, sb = NULL);
}
- err = obd_connect(&sbi->ll_mdc_conn, obd, &sbi->ll_sb_uuid,
- ptlrpc_recovd, ll_recover);
+ err = obd_connect(&sbi->ll_mdc_conn, obd, &sbi->ll_sb_uuid);
if (err) {
CERROR("cannot connect to %s: rc = %d\n", mdc, err);
GOTO(out_free, sb = NULL);
}
- mdc_conn = sbi2mdc(sbi)->cl_import.imp_connection;
- list_add(&mdc_conn->c_sb_chain, &sbi->ll_conn_chain);
+ mdc_conn = sbi2mdc(sbi)->cl_import->imp_connection;
+ strncpy(param_uuid.uuid, osc, sizeof(param_uuid.uuid));
- obd = class_uuid2obd(osc);
+ obd = class_uuid2obd(¶m_uuid);
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,
- ptlrpc_recovd, ll_recover);
+ err = obd_connect(&sbi->ll_osc_conn, obd, &sbi->ll_sb_uuid);
if (err) {
CERROR("cannot connect to %s: rc = %d\n", osc, err);
GOTO(out_mdc, sb = NULL);
err = mdc_getstatus(&sbi->ll_mdc_conn, &rootfid);
if (err) {
CERROR("cannot mds_connect: rc = %d\n", err);
- GOTO(out_mdc, sb = NULL);
+ GOTO(out_osc, sb = NULL);
}
CDEBUG(D_SUPER, "rootfid "LPU64"\n", rootfid.id);
sbi->ll_rootino = rootfid.id;
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)) - osfs.os_bsize;
+ sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
sb->s_op = &ll_super_operations;
- /* make root inode */
- err = mdc_getattr(&sbi->ll_mdc_conn, sbi->ll_rootino, S_IFDIR,
+ /* make root inode
+ * XXX: move this to after cbd setup? */
+ err = mdc_getattr(&sbi->ll_mdc_conn, &rootfid,
OBD_MD_FLNOTOBD|OBD_MD_FLBLOCKS, 0, &request);
if (err) {
CERROR("mdc_getattr failed for root: rc = %d\n", err);
- GOTO(out_request, sb = NULL);
+ GOTO(out_osc, sb = NULL);
}
/* initialize committed transaction callback daemon */
err = ll_commitcbd_setup(sbi);
if (err) {
CERROR("failed to start commit callback daemon: rc = %d\n",err);
- GOTO(out_request, sb = NULL);
+ ptlrpc_req_finished (request);
+ GOTO(out_osc, sb = NULL);
}
- lic.lic_body = lustre_msg_buf(request->rq_repmsg, 0);
- lic.lic_lmm = NULL;
+ lic.lic_body = lustre_msg_buf(request->rq_repmsg, 0, sizeof (*lic.lic_body));
+ LASSERT (lic.lic_body != NULL); /* checked by mdc_getattr() */
+ LASSERT_REPSWABBED (request, 0); /* swabbed by mdc_getattr() */
+
+ lic.lic_lsm = NULL;
+
root = iget5_locked(sb, sbi->ll_rootino, NULL,
ll_read_inode2, &lic);
- if (root) {
- sb->s_root = d_alloc_root(root);
- root->i_state &= ~(I_LOCK | I_NEW);
- } else {
- CERROR("lustre_lite: bad iget4 for root\n");
- GOTO(out_cdb, sb = NULL);
- }
-
ptlrpc_req_finished(request);
- request = NULL;
+ if (root == NULL || is_bad_inode(root)) {
+ /* XXX might need iput() for bad inode */
+ CERROR("lustre_lite: bad iget5 for root\n");
+ GOTO(out_cbd, sb = NULL);
+ }
+
+ sb->s_root = d_alloc_root(root);
+ root->i_state &= ~(I_LOCK | I_NEW);
+ printk("AMRUT 1\n");
if (proc_lustre_fs_root) {
err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
osc, mdc);
OBD_FREE(mdc, strlen(mdc) + 1);
if (osc)
OBD_FREE(osc, strlen(osc) + 1);
+ printk("AMRUT 2\n");
RETURN(0);
-out_cdb:
+out_cbd:
ll_commitcbd_cleanup(sbi);
-out_request:
- ptlrpc_req_finished(request);
- obd_disconnect(&sbi->ll_osc_conn);
+out_osc:
+ obd_disconnect(&sbi->ll_osc_conn, 0);
out_mdc:
- obd_disconnect(&sbi->ll_mdc_conn);
+ obd_disconnect(&sbi->ll_mdc_conn, 0);
out_free:
OBD_FREE(sbi, sizeof(*sbi));
goto out_dev;
} /* ll_fill_super */
+
+int ll_setattr_raw(struct inode *inode, struct iattr *attr)
+{
+ struct ptlrpc_request *request = NULL;
+ struct ll_sb_info *sbi = ll_i2sbi(inode);
+ struct mdc_op_data op_data;
+ int err = 0;
+ ENTRY;
+ CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu\n", inode->i_ino);
+
+ if ((attr->ia_valid & ATTR_SIZE)) {
+ /* writeback uses inode->i_size to determine how far out
+ * its cached pages go. ll_truncate gets a PW lock, canceling
+ * our lock, _after_ it has updated i_size. this can confuse
+ * us into zero extending the file to the newly truncated
+ * size, and this has bad implications for a racing o_append.
+ * if we're extending our size we need to flush the pages
+ * with the correct i_size before vmtruncate stomps on
+ * the new i_size. again, this can only find pages to
+ * purge if the PW lock that generated them is still held.
+ */
+ if ( attr->ia_size > inode->i_size ) {
+ filemap_fdatasync(inode->i_mapping);
+ filemap_fdatawait(inode->i_mapping);
+ }
+ err = vmtruncate(inode, attr->ia_size);
+ if (err)
+ RETURN(err);
+ }
+
+ /* Don't send size changes to MDS to avoid "fast EA" problems, and
+ * also avoid a pointless RPC (we get file size from OST anyways).
+ */
+ attr->ia_valid &= ~ATTR_SIZE;
+ if (!attr->ia_valid)
+ RETURN(0);
+
+ ll_prepare_mdc_op_data(&op_data, inode, NULL, NULL, 0, 0);
+
+ err = mdc_setattr(&sbi->ll_mdc_conn, &op_data,
+ attr, NULL, 0, &request);
+ if (err)
+ CERROR("mdc_setattr fails: err = %d\n", err);
+
+ ptlrpc_req_finished(request);
+
+ if (S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_MTIME_SET) {
+ struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
+ struct obdo oa;
+ int err2;
+
+ CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
+ inode->i_ino, attr->ia_mtime);
+ oa.o_id = lsm->lsm_object_id;
+ oa.o_mode = S_IFREG;
+ oa.o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLMTIME;
+ oa.o_mtime = LTIME_S(attr->ia_mtime);
+ err2 = obd_setattr(&sbi->ll_osc_conn, &oa, lsm, NULL);
+ if (err2) {
+ CERROR("obd_setattr fails: rc=%d\n", err);
+ if (!err)
+ err = err2;
+ }
+ }
+ RETURN(err);
+}
struct super_block * ll_get_sb(struct file_system_type *fs_type,
int flags, char *devname, void * data)
{
struct list_head *tmp, *next;
struct ll_fid rootfid;
ENTRY;
+ CDEBUG(D_VFSTRACE, "VFS Op:\n");
list_del(&sbi->ll_conn_chain);
ll_commitcbd_cleanup(sbi);
- obd_disconnect(&sbi->ll_osc_conn);
+ obd_disconnect(&sbi->ll_osc_conn, 0);
/* NULL request to force sync on the MDS, and get the last_committed
* value to flush remaining RPCs from the pending queue on client.
sbi->ll_proc_root = NULL;
}
- obd_disconnect(&sbi->ll_mdc_conn);
+ obd_disconnect(&sbi->ll_mdc_conn, 0);
spin_lock(&dcache_lock);
list_for_each_safe(tmp, next, &sbi->ll_orphan_dentry_list){
struct ll_inode_info *lli = ll_i2info(inode);
int rc;
ENTRY;
+ CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu\n", inode->i_ino);
#warning "Is there a reason we don't do this in 2.5, but we do in 2.4?"
#if 0
- rc = mdc_cancel_unused(&sbi->ll_mdc_conn, inode, LDLM_FL_NO_CALLBACK);
+ rc = ll_mdc_cancel_unused(&sbi->ll_mdc_conn, inode, LDLM_FL_NO_CALLBACK);
if (rc < 0) {
- CERROR("mdc_cancel_unused: %d\n", rc);
+ CERROR("ll_mdc_cancel_unused: %d\n", rc);
/* XXX FIXME do something dramatic */
}
CERROR("clearing in-use inode %lu: count = %d\n",
inode->i_ino, atomic_read(&inode->i_count));
- if (lli->lli_smd)
+ if (lli->lli_smd) {
obd_free_memmd(&sbi->ll_osc_conn, &lli->lli_smd);
+ lli->lli_smd = NULL;
+ }
if (lli->lli_symlink_name) {
OBD_FREE(lli->lli_symlink_name,strlen(lli->lli_symlink_name)+1);
static void ll_delete_inode(struct inode *inode)
{
ENTRY;
+ CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu\n", inode->i_ino);
if (S_ISREG(inode->i_mode)) {
int err;
struct obdo *oa;
int error = 0;
if ((ia_valid & ATTR_SIZE) && trunc) {
+ if (attr->ia_size > ll_file_maxbytes(inode)) {
+ error = -EFBIG;
+ goto out;
+ }
error = vmtruncate(inode, attr->ia_size);
if (error)
goto out;
ENTRY;
/* change incore inode */
- ll_attr2inode(inode, attr, do_trunc);
+ err = ll_attr2inode(inode, attr, do_trunc);
+ if (err)
+ RETURN(err);
/* Don't send size changes to MDS to avoid "fast EA" problems, and
* also avoid a pointless RPC (we get file size from OST anyways).
*/
attr->ia_valid &= ~ATTR_SIZE;
if (attr->ia_valid) {
- err = mdc_setattr(&sbi->ll_mdc_conn, inode, attr, NULL, 0,
- &request);
+ struct mdc_op_data op_data;
+
+ ll_prepare_mdc_op_data(&op_data, inode, NULL, NULL, 0, 0);
+
+ err = mdc_setattr(&sbi->ll_mdc_conn, &op_data,
+ attr, NULL, 0, &request);
if (err)
CERROR("mdc_setattr fails: err = %d\n", err);
oa.o_id = lsm->lsm_object_id;
oa.o_mode = S_IFREG;
oa.o_valid = OBD_MD_FLID |OBD_MD_FLTYPE |OBD_MD_FLMTIME;
- oa.o_mtime = attr->ia_mtime.tv_sec;
+ oa.o_mtime = LTIME_S(attr->ia_mtime);
err2 = obd_setattr(&sbi->ll_osc_conn, &oa, lsm, NULL);
if (err2) {
CERROR("obd_setattr fails: rc=%d\n", err);
int ll_setattr(struct dentry *de, struct iattr *attr)
{
int rc = inode_change_ok(de->d_inode, attr);
-
+ CDEBUG(D_VFSTRACE, "VFS Op:name=%s\n", de->d_name.name);
if (rc)
return rc;
struct obd_statfs osfs;
int rc;
ENTRY;
+ CDEBUG(D_VFSTRACE, "VFS Op:\n");
memset(sfs, 0, sizeof(*sfs));
rc = obd_statfs(&sbi->ll_mdc_conn, &osfs);
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)
+ if (osfs.os_ffree < (__u64)sfs->f_ffree) {
+ sfs->f_files = (sfs->f_files - sfs->f_ffree) +
+ osfs.os_ffree;
sfs->f_ffree = osfs.os_ffree;
+ }
}
out:
}
void ll_update_inode(struct inode *inode, struct mds_body *body,
- struct lov_mds_md *lmm)
+ struct lov_stripe_md *lsm)
{
struct ll_inode_info *lli = ll_i2info(inode);
- if (lmm != NULL)
- obd_unpackmd(ll_i2obdconn(inode), &lli->lli_smd, lmm);
+ LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
+ if (lsm != NULL) {
+ if (lli->lli_smd == NULL) {
+ lli->lli_smd = lsm;
+ lli->lli_maxbytes = lsm->lsm_maxbytes;
+ if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
+ lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
+ } else {
+ LASSERT (!memcmp (lli->lli_smd, lsm, sizeof (*lsm)));
+ }
+ }
if (body->valid & OBD_MD_FLID)
inode->i_ino = body->ino;
if (body->valid & OBD_MD_FLATIME)
- inode->i_atime.tv_sec = body->atime;
+ LTIME_S(inode->i_atime) = body->atime;
if (body->valid & OBD_MD_FLMTIME)
- inode->i_mtime.tv_sec = body->mtime;
+ LTIME_S(inode->i_mtime) = body->mtime;
if (body->valid & OBD_MD_FLCTIME)
- inode->i_ctime.tv_sec = body->ctime;
+ LTIME_S(inode->i_ctime) = body->ctime;
if (body->valid & OBD_MD_FLMODE)
inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
if (body->valid & OBD_MD_FLTYPE)
struct ll_inode_info *lli = ll_i2info(inode);
int rc = 0;
ENTRY;
+ CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu\n", inode->i_ino);
sema_init(&lli->lli_open_sem, 1);
- lli->flags = 0;
- init_MUTEX(&lli->lli_getattr_sem);
/* these are 2.4 only, but putting them here for consistency.. */
spin_lock_init(&lli->lli_read_extent_lock);
INIT_LIST_HEAD(&lli->lli_read_extents);
+ ll_lldo_init(&lli->lli_dirty);
+ lli->lli_flags = 0;
+ lli->lli_maxbytes = LUSTRE_STRIPE_MAXBYTES;
LASSERT(!lli->lli_smd);
/* core attributes first */
- ll_update_inode(inode, body, lic ? lic->lic_lmm : NULL);
-
- /* Get the authoritative file size */
- if (lli->lli_smd && S_ISREG(inode->i_mode)) {
- struct ll_file_data *fd = file->private_data;
- struct ldlm_extent extent = {0, OBD_OBJECT_EOF};
- struct lustre_handle lockh = {0, 0};
-
- LASSERT(lli->lli_smd->lsm_object_id != 0);
-
- rc = ll_extent_lock(fd, inode, lsm, LCK_PR, &extent, &lockh);
- if (err != ELDLM_OK && err != ELDLM_MATCHED) {
- ll_clear_inode(inode);
- make_bad_inode(inode);
- } else {
- l_extent_unlock(fd, inode, lsm, LCK_PR, &extent,
- &lockh);
- }
- }
+ ll_update_inode(inode, body, lic ? lic->lic_lsm : NULL);
/* OIDEBUG(inode); */
inode->i_op = &ll_fast_symlink_inode_operations;
EXIT;
} else {
+ inode->i_op = &ll_special_inode_operations;
init_special_inode(inode, inode->i_mode,
kdev_t_to_nr(inode->i_rdev));
EXIT;
return rc;
}
-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 %d 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;
+ struct obd_device *obd;
+ struct obd_ioctl_data ioc_data = { 0 };
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);
- conn->c_flags |= CONN_INVALID;
- /*invalidate_request_list(&conn->c_sending_head);*/
- invalidate_request_list(&conn->c_delayed_head);
- spin_unlock(&conn->c_lock);
- }
+ CDEBUG(D_VFSTRACE, "VFS Op:\n");
+
+ obd = class_conn2obd(&sbi->ll_mdc_conn);
+ obd->obd_no_recov = 1;
+ obd_iocontrol(IOC_OSC_SET_ACTIVE, &sbi->ll_mdc_conn, sizeof ioc_data,
+ &ioc_data, NULL);
+
+ obd = class_conn2obd(&sbi->ll_osc_conn);
+ obd->obd_no_recov = 1;
+ obd_iocontrol(IOC_OSC_SET_ACTIVE, &sbi->ll_osc_conn, sizeof ioc_data,
+ &ioc_data, NULL);
+
+ /* Really, we'd like to wait until there are no requests outstanding,
+ * and then continue. For now, we just invalidate the requests,
+ * schedule, and hope.
+ */
+ schedule();
EXIT;
}
-
static kmem_cache_t *ll_inode_cachep;
static struct inode *ll_alloc_inode(struct super_block *sb)
{
struct ll_inode_info *lli;
- lli = kmem_cache_alloc(ll_inode_cachep, SLAB_KERNEL);
- if (!lli)
+ OBD_SLAB_ALLOC(lli, ll_inode_cachep, SLAB_KERNEL, sizeof *lli);
+ if (lli == NULL)
return NULL;
memset(lli, 0, (char *)&lli->lli_vfs_inode - (char *)lli);
sema_init(&lli->lli_open_sem, 1);
init_MUTEX(&lli->lli_size_valid_sem);
+ lli->lli_maxbytes = LUSTRE_STRIPE_MAXBYTES;
return &lli->lli_vfs_inode;
}
static void ll_destroy_inode(struct inode *inode)
{
- kmem_cache_free(ll_inode_cachep, ll_i2info(inode));
+ OBD_SLAB_FREE(ll_inode_cachep, ll_i2info(inode),
+ sizeof(struct ll_inode_info));
}
static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
git://git.whamcloud.com / fs / lustre-release.git / blobdiff