#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))
kmem_cache_t *ll_file_data_slab;
ENTRY;
- CDEBUG(D_VFSTRACE, "VFS Op\n");
+ CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
OBD_ALLOC(sbi, sizeof(*sbi));
if (!sbi)
RETURN(NULL);
INIT_LIST_HEAD(&sbi->ll_conn_chain);
INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
generate_random_uuid(uuid);
+ spin_lock_init(&sbi->ll_iostats.fis_lock);
class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
sb->u.generic_sbp = 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(¶m_uuid);
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;
+
LASSERT(sbi->ll_rootino != 0);
root = iget4(sb, sbi->ll_rootino, NULL, &lic);
- if (root) {
- sb->s_root = d_alloc_root(root);
- } else {
+ ptlrpc_req_finished(request);
+
+ if (root == NULL || is_bad_inode(root)) {
+ /* XXX might need iput() for bad inode */
CERROR("lustre_lite: bad iget4 for root\n");
- GOTO(out_cdb, sb = NULL);
+ GOTO(out_cbd, sb = NULL);
}
- ptlrpc_req_finished(request);
- request = NULL;
+ sb->s_root = d_alloc_root(root);
if (proc_lustre_fs_root) {
err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
RETURN(sb);
-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));
struct ll_sb_info *sbi = ll_s2sbi(sb);
struct list_head *tmp, *next;
struct ll_fid rootfid;
+ struct obd_device *obd = class_conn2obd(&sbi->ll_mdc_conn);
ENTRY;
- CDEBUG(D_VFSTRACE, "VFS Op\n");
+ CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
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 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);
+ if (!obd->obd_no_recov)
+ mdc_getstatus(&sbi->ll_mdc_conn, &rootfid);
if (sbi->ll_proc_root) {
lprocfs_remove(sbi->ll_proc_root);
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) {
int rc;
ENTRY;
- CDEBUG(D_VFSTRACE, "VFS Op\n");
- rc = mdc_cancel_unused(&sbi->ll_mdc_conn, inode, LDLM_FL_NO_CALLBACK);
+ CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
+ inode->i_generation, inode);
+ rc = ll_mdc_cancel_unused(&sbi->ll_mdc_conn, inode,
+ LDLM_FL_NO_CALLBACK, inode);
if (rc < 0) {
- CERROR("mdc_cancel_unused: %d\n", rc);
+ CERROR("ll_mdc_cancel_unused: %d\n", rc);
/* XXX FIXME do something dramatic */
}
+ if (atomic_read(&inode->i_count) != 0)
+ CERROR("clearing in-use inode %lu: count = %d\n",
+ inode->i_ino, atomic_read(&inode->i_count));
+
if (lli->lli_smd) {
- rc = obd_cancel_unused(&sbi->ll_osc_conn, lli->lli_smd, 0);
+ rc = obd_cancel_unused(&sbi->ll_osc_conn, lli->lli_smd,
+ LDLM_FL_WARN, inode);
if (rc < 0) {
CERROR("obd_cancel_unused: %d\n", rc);
/* XXX FIXME do something dramatic */
}
- }
-
- if (atomic_read(&inode->i_count) != 0)
- CERROR("clearing in-use inode %lu: count = %d\n",
- inode->i_ino, atomic_read(&inode->i_count));
-
- 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,
static void ll_delete_inode(struct inode *inode)
{
ENTRY;
- CDEBUG(D_VFSTRACE, "VFS Op\n");
+ CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
+ inode->i_generation, inode);
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);
int ll_setattr_raw(struct inode *inode, struct iattr *attr)
{
- struct ptlrpc_request *request = NULL;
+ struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
struct ll_sb_info *sbi = ll_i2sbi(inode);
- int err = 0;
+ struct ptlrpc_request *request = NULL;
+ struct mdc_op_data op_data;
+ int rc = 0, err;
ENTRY;
+ CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
+ inode->i_generation, inode);
if ((attr->ia_valid & ATTR_SIZE)) {
+ struct ldlm_extent extent = {attr->ia_size, OBD_OBJECT_EOF};
+ struct lustre_handle lockh = { 0 };
+
+ if (attr->ia_size > ll_file_maxbytes(inode))
+ RETURN(-EFBIG);
+
/* 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);
+ *
+ * If this file doesn't have stripes yet, it is already,
+ * by definition, truncated. */
+ if ((attr->ia_valid & ATTR_FROM_OPEN) && lsm == NULL) {
+ LASSERT(attr->ia_size == 0);
+ GOTO(skip_extent_lock, rc = 0);
+ }
+
+ /* we really need to get our PW lock before we change
+ * inode->i_size. if we don't we can race with other
+ * i_size updaters on our node, like ll_file_read. we
+ * can also race with i_size propogation to other
+ * nodes through dirtying and writeback of final cached
+ * pages. this last one is especially bad for racing
+ * o_append users on other nodes. */
+ rc = ll_extent_lock_no_validate(NULL, inode, lsm, LCK_PW,
+ &extent, &lockh);
+ if (rc != ELDLM_OK) {
+ if (rc > 0)
+ RETURN(-ENOLCK);
+ RETURN(rc);
}
- err = vmtruncate(inode, attr->ia_size);
+
+ rc = vmtruncate(inode, attr->ia_size);
+ if (rc == 0)
+ set_bit(LLI_F_HAVE_SIZE_LOCK,
+ &ll_i2info(inode)->lli_flags);
+
+ /* unlock now as we don't mind others file lockers racing with
+ * the mds updates below? */
+ err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
if (err)
- RETURN(err);
+ CERROR("ll_extent_unlock failed: %d\n", err);
+ if (rc)
+ RETURN(rc);
}
+skip_extent_lock:
/* 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).
*/
if (!attr->ia_valid)
RETURN(0);
- err = mdc_setattr(&sbi->ll_mdc_conn, inode, attr, NULL, 0,
- &request);
+ 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) {
+ 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;
+ if (lsm == NULL) {
+ CDEBUG(D_INODE, "no lsm: not setting mtime on OSTs\n");
+ RETURN(err);
+ }
+
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;
int ll_setattr(struct dentry *de, struct iattr *attr)
{
int rc = inode_change_ok(de->d_inode, attr);
-
- CDEBUG(D_VFSTRACE, "VFS Op\n");
+ CDEBUG(D_VFSTRACE, "VFS Op:name=%s\n", de->d_name.name);
if (rc)
return rc;
int rc;
ENTRY;
- CDEBUG(D_VFSTRACE, "VFS Op\n");
+ CDEBUG(D_VFSTRACE, "VFS Op:\n");
memset(sfs, 0, sizeof(*sfs));
rc = obd_statfs(&sbi->ll_mdc_conn, &osfs);
statfs_unpack(sfs, &osfs);
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)
+
+ /* If we don't have as many objects free on the OST as inodes
+ * on the MDS, we reduce the total number of inodes to
+ * compensate, so that the "inodes in use" number is correct.
+ */
+ 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:
RETURN(rc);
}
+void dump_lsm(int level, struct lov_stripe_md *lsm)
+{
+ CDEBUG(level, "objid "LPX64", maxbytes "LPX64", magic %#08x, "
+ "stripe_size %#08x, offset %u, stripe_count %u\n",
+ lsm->lsm_object_id, lsm->lsm_maxbytes, lsm->lsm_magic,
+ lsm->lsm_stripe_size, lsm->lsm_stripe_offset,
+ lsm->lsm_stripe_count);
+}
+
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_maxbytes = lsm->lsm_maxbytes;
+ if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
+ lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
+ lli->lli_smd = lsm;
+ } else {
+ if (memcmp(lli->lli_smd, lsm, sizeof(*lsm))) {
+ CERROR("lsm mismatch for inode %ld\n",
+ inode->i_ino);
+ CERROR("lli_smd:\n");
+ dump_lsm(D_ERROR, lli->lli_smd);
+ CERROR("lsm:\n");
+ dump_lsm(D_ERROR, lsm);
+ LBUG();
+ }
+ }
+ }
if (body->valid & OBD_MD_FLID)
inode->i_ino = body->ino;
if (body->valid & OBD_MD_FLATIME)
- inode->i_atime = body->atime;
+ LTIME_S(inode->i_atime) = body->atime;
if (body->valid & OBD_MD_FLMTIME)
- inode->i_mtime = body->mtime;
+ LTIME_S(inode->i_mtime) = body->mtime;
if (body->valid & OBD_MD_FLCTIME)
- inode->i_ctime = 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 mds_body *body = lic->lic_body;
struct ll_inode_info *lli = ll_i2info(inode);
ENTRY;
+ CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
+ inode->i_generation, inode);
- CDEBUG(D_VFSTRACE, "VFS Op\n");
sema_init(&lli->lli_open_sem, 1);
- atomic_set(&lli->lli_open_count, 0);
- lli->lli_flags = 0;
- init_MUTEX(&lli->lli_getattr_sem);
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;
+ /* We default to 2T-4k until the LSM is created/read, at which point
+ * it'll be updated. */
+ lli->lli_maxbytes = LUSTRE_STRIPE_MAXBYTES;
LASSERT(!lli->lli_smd);
/* core attributes from the MDS first */
- ll_update_inode(inode, body, lic ? lic->lic_lmm : NULL);
-
- /* Get the authoritative file size */
- if (lli->lli_smd && (inode->i_mode & S_IFREG)) {
- struct ldlm_extent extent = {0, OBD_OBJECT_EOF};
- struct lustre_handle lockh = {0, 0};
- struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
- ldlm_error_t rc;
-
- LASSERT(lli->lli_smd->lsm_object_id != 0);
-
- rc = ll_extent_lock(NULL, inode, lsm, LCK_PR, &extent, &lockh);
- if (rc != ELDLM_OK && rc != ELDLM_LOCK_MATCHED) {
- ll_clear_inode(inode);
- make_bad_inode(inode);
- } else {
- ll_extent_unlock(NULL, inode, lsm, LCK_PR, &lockh);
- }
- }
+ ll_update_inode(inode, body, lic->lic_lsm);
/* OIDEBUG(inode); */
}
}
-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 "LPU64" op %d to %s:%d\n",
- req->rq_xid, req->rq_reqmsg->opc,
- req->rq_connection->c_remote_uuid.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;
- CDEBUG(D_VFSTRACE, "VFS Op\n");
-
- 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);
- }
+ 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;
}
git://git.whamcloud.com / fs / lustre-release.git / blobdiff