#include <linux/fs.h>
/* XATTR_{REPLACE,CREATE} */
#include <linux/xattr.h>
-/*
- * XXX temporary stuff: direct access to ldiskfs/jdb. Interface between osd
- * and file system is not yet specified.
- */
-/* handle_t, journal_start(), journal_stop() */
-#include <linux/jbd.h>
-/* LDISKFS_SB() */
-#include <linux/ldiskfs_fs.h>
-#include <linux/ldiskfs_jbd.h>
/* simple_mkdir() */
#include <lvfs.h>
/* fid_is_local() */
#include <lustre_fid.h>
-#include <linux/lustre_iam.h>
#include "osd_internal.h"
#include "osd_igif.h"
struct osd_directory {
struct iam_container od_container;
struct iam_descr od_descr;
- struct semaphore od_sem;
};
struct osd_object {
* creation, or assigned by osd_object_create() under write lock).
*/
struct inode *oo_inode;
+ /**
+ * to protect index ops.
+ */
+ struct rw_semaphore oo_ext_idx_sem;
struct rw_semaphore oo_sem;
struct osd_directory *oo_dir;
/** protects inode attributes. */
#endif
};
-static int osd_root_get (const struct lu_env *env,
- struct dt_device *dev, struct lu_fid *f);
-
-static int lu_device_is_osd (const struct lu_device *d);
-static void osd_mod_exit (void) __exit;
-static int osd_mod_init (void) __init;
-static int osd_type_init (struct lu_device_type *t);
-static void osd_type_fini (struct lu_device_type *t);
-static int osd_object_init (const struct lu_env *env,
- struct lu_object *l,
- const struct lu_object_conf *_);
-static void osd_object_release(const struct lu_env *env,
- struct lu_object *l);
-static int osd_object_print (const struct lu_env *env, void *cookie,
- lu_printer_t p, const struct lu_object *o);
-static struct lu_device *osd_device_free (const struct lu_env *env,
- struct lu_device *m);
-static void *osd_key_init (const struct lu_context *ctx,
- struct lu_context_key *key);
-static void osd_key_fini (const struct lu_context *ctx,
- struct lu_context_key *key, void *data);
-static void osd_key_exit (const struct lu_context *ctx,
- struct lu_context_key *key, void *data);
-static int osd_has_index (const struct osd_object *obj);
-static void osd_object_init0 (struct osd_object *obj);
-static int osd_device_init (const struct lu_env *env,
- struct lu_device *d, const char *,
- struct lu_device *);
-static int osd_fid_lookup (const struct lu_env *env,
- struct osd_object *obj,
- const struct lu_fid *fid);
-static void osd_inode_getattr (const struct lu_env *env,
- struct inode *inode, struct lu_attr *attr);
-static int osd_inode_setattr (const struct lu_env *env,
- struct inode *inode, const struct lu_attr *attr);
-static int osd_param_is_sane (const struct osd_device *dev,
- const struct txn_param *param);
-static int osd_index_iam_lookup(const struct lu_env *env,
- struct dt_object *dt,
- struct dt_rec *rec, const struct dt_key *key,
- struct lustre_capa *capa);
-static int osd_index_ea_lookup(const struct lu_env *env,
- struct dt_object *dt,
- struct dt_rec *rec, const struct dt_key *key,
- struct lustre_capa *capa);
-static int osd_index_iam_insert(const struct lu_env *env,
- struct dt_object *dt,
- const struct dt_rec *rec,
- const struct dt_key *key,
- struct thandle *handle,
- struct lustre_capa *capa,
- int ingore_quota);
-static int osd_index_ea_insert (const struct lu_env *env,
- struct dt_object *dt,
- const struct dt_rec *rec,
- const struct dt_key *key,
- struct thandle *handle,
- struct lustre_capa *capa,
- int ingore_quota);
-static int osd_index_iam_delete(const struct lu_env *env,
- struct dt_object *dt, const struct dt_key *key,
- struct thandle *handle,
- struct lustre_capa *capa);
-static int osd_index_ea_delete (const struct lu_env *env,
- struct dt_object *dt, const struct dt_key *key,
- struct thandle *handle,
- struct lustre_capa *capa);
-
-static int osd_iam_index_probe (const struct lu_env *env,
- struct osd_object *o,
- const struct dt_index_features *feat);
-static int osd_index_try (const struct lu_env *env,
- struct dt_object *dt,
- const struct dt_index_features *feat);
-static void osd_index_fini (struct osd_object *o);
-
-static void osd_it_iam_fini (const struct lu_env *env, struct dt_it *di);
-static int osd_it_iam_get (const struct lu_env *env,
- struct dt_it *di, const struct dt_key *key);
-static void osd_it_iam_put (const struct lu_env *env, struct dt_it *di);
-static int osd_it_iam_next (const struct lu_env *env, struct dt_it *di);
-static int osd_it_iam_key_size (const struct lu_env *env,
- const struct dt_it *di);
-static void osd_it_ea_fini (const struct lu_env *env, struct dt_it *di);
-static int osd_it_ea_get (const struct lu_env *env,
- struct dt_it *di, const struct dt_key *key);
-static void osd_it_ea_put (const struct lu_env *env, struct dt_it *di);
-static int osd_it_ea_next (const struct lu_env *env, struct dt_it *di);
-static int osd_it_ea_key_size(const struct lu_env *env,
- const struct dt_it *di);
-
-static void osd_conf_get (const struct lu_env *env,
- const struct dt_device *dev,
- struct dt_device_param *param);
-static void osd_trans_stop (const struct lu_env *env,
- struct thandle *th);
-static int osd_object_is_root(const struct osd_object *obj);
-
-static struct osd_object *osd_obj (const struct lu_object *o);
-static struct osd_device *osd_dev (const struct lu_device *d);
-static struct osd_device *osd_dt_dev (const struct dt_device *d);
-static struct osd_object *osd_dt_obj (const struct dt_object *d);
-static struct osd_device *osd_obj2dev (const struct osd_object *o);
-static struct lu_device *osd2lu_dev (struct osd_device *osd);
-static struct lu_device *osd_device_fini (const struct lu_env *env,
- struct lu_device *d);
-static struct lu_device *osd_device_alloc (const struct lu_env *env,
- struct lu_device_type *t,
- struct lustre_cfg *cfg);
-static struct lu_object *osd_object_alloc (const struct lu_env *env,
- const struct lu_object_header *hdr,
- struct lu_device *d);
-static struct inode *osd_iget (struct osd_thread_info *info,
- struct osd_device *dev,
- const struct osd_inode_id *id);
-static struct super_block *osd_sb (const struct osd_device *dev);
-static struct dt_it *osd_it_iam_init (const struct lu_env *env,
- struct dt_object *dt,
- struct lustre_capa *capa);
-static struct dt_key *osd_it_iam_key (const struct lu_env *env,
- const struct dt_it *di);
-static struct dt_rec *osd_it_iam_rec (const struct lu_env *env,
- const struct dt_it *di);
-static struct dt_it *osd_it_ea_init (const struct lu_env *env,
- struct dt_object *dt,
- struct lustre_capa *capa);
-static struct dt_key *osd_it_ea_key (const struct lu_env *env,
- const struct dt_it *di);
-static struct dt_rec *osd_it_ea_rec (const struct lu_env *env,
- const struct dt_it *di);
-
-static struct timespec *osd_inode_time (const struct lu_env *env,
- struct inode *inode,
- __u64 seconds);
-static struct thandle *osd_trans_start (const struct lu_env *env,
- struct dt_device *d,
- struct txn_param *p);
-static journal_t *osd_journal (const struct osd_device *dev);
-
-static int __osd_ea_add_rec(struct osd_thread_info *info,
- struct osd_object *pobj,
- struct osd_object *cobj,
- const char *name,
- struct thandle *th);
-
-static const struct lu_device_type_operations osd_device_type_ops;
-static struct lu_device_type osd_device_type;
static const struct lu_object_operations osd_lu_obj_ops;
-static struct obd_ops osd_obd_device_ops;
static const struct lu_device_operations osd_lu_ops;
static struct lu_context_key osd_key;
static const struct dt_object_operations osd_obj_ops;
};
+/*
+ * Helpers.
+ */
+static int lu_device_is_osd(const struct lu_device *d)
+{
+ return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &osd_lu_ops);
+}
+
+static struct osd_device *osd_dt_dev(const struct dt_device *d)
+{
+ LASSERT(lu_device_is_osd(&d->dd_lu_dev));
+ return container_of0(d, struct osd_device, od_dt_dev);
+}
+
+static struct osd_device *osd_dev(const struct lu_device *d)
+{
+ LASSERT(lu_device_is_osd(d));
+ return osd_dt_dev(container_of0(d, struct dt_device, dd_lu_dev));
+}
+
+static struct osd_device *osd_obj2dev(const struct osd_object *o)
+{
+ return osd_dev(o->oo_dt.do_lu.lo_dev);
+}
+
+static struct super_block *osd_sb(const struct osd_device *dev)
+{
+ return dev->od_mount->lmi_mnt->mnt_sb;
+}
+
+static int osd_object_is_root(const struct osd_object *obj)
+{
+ return osd_sb(osd_obj2dev(obj))->s_root->d_inode == obj->oo_inode;
+}
+
+static struct osd_object *osd_obj(const struct lu_object *o)
+{
+ LASSERT(lu_device_is_osd(o->lo_dev));
+ return container_of0(o, struct osd_object, oo_dt.do_lu);
+}
+
+static struct osd_object *osd_dt_obj(const struct dt_object *d)
+{
+ return osd_obj(&d->do_lu);
+}
+
+static struct lu_device *osd2lu_dev(struct osd_device *osd)
+{
+ return &osd->od_dt_dev.dd_lu_dev;
+}
+
+static journal_t *osd_journal(const struct osd_device *dev)
+{
+ return LDISKFS_SB(osd_sb(dev))->s_journal;
+}
+
+static int osd_has_index(const struct osd_object *obj)
+{
+ return obj->oo_dt.do_index_ops != NULL;
+}
+
+static int osd_object_invariant(const struct lu_object *l)
+{
+ return osd_invariant(osd_obj(l));
+}
+
#ifdef HAVE_QUOTA_SUPPORT
static inline void
osd_push_ctxt(const struct lu_env *env, struct osd_ctxt *save)
}
#endif
-/*
- * Invariants, assertions.
- */
-
-/*
- * XXX: do not enable this, until invariant checking code is made thread safe
- * in the face of pdirops locking.
- */
-#define OSD_INVARIANT_CHECKS (0)
-
-#if OSD_INVARIANT_CHECKS
-static int osd_invariant(const struct osd_object *obj)
-{
- return
- obj != NULL &&
- ergo(obj->oo_inode != NULL,
- obj->oo_inode->i_sb == osd_sb(osd_obj2dev(obj)) &&
- atomic_read(&obj->oo_inode->i_count) > 0) &&
- ergo(obj->oo_dir != NULL &&
- obj->oo_dir->od_conationer.ic_object != NULL,
- obj->oo_dir->od_conationer.ic_object == obj->oo_inode);
-}
-#else
-#define osd_invariant(obj) (1)
-#endif
-
static inline struct osd_thread_info *osd_oti_get(const struct lu_env *env)
{
return lu_context_key_get(&env->le_ctx, &osd_key);
l->lo_ops = &osd_lu_obj_ops;
init_rwsem(&mo->oo_sem);
+ init_rwsem(&mo->oo_ext_idx_sem);
spin_lock_init(&mo->oo_guard);
return l;
} else
}
/*
+ * retrieve object from backend ext fs.
+ **/
+static struct inode *osd_iget(struct osd_thread_info *info,
+ struct osd_device *dev,
+ const struct osd_inode_id *id)
+{
+ struct inode *inode = NULL;
+
+#ifdef HAVE_EXT4_LDISKFS
+ inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
+ if (IS_ERR(inode))
+ /* Newer kernels return an error instead of a NULL pointer */
+ inode = NULL;
+#else
+ inode = iget(osd_sb(dev), id->oii_ino);
+#endif
+ if (inode == NULL) {
+ CERROR("no inode\n");
+ inode = ERR_PTR(-EACCES);
+ } else if (id->oii_gen != OSD_OII_NOGEN &&
+ inode->i_generation != id->oii_gen) {
+ iput(inode);
+ inode = ERR_PTR(-ESTALE);
+ } else if (inode->i_nlink == 0) {
+ /* due to parallel readdir and unlink,
+ * we can have dead inode here. */
+ CWARN("stale inode\n");
+ make_bad_inode(inode);
+ iput(inode);
+ inode = ERR_PTR(-ESTALE);
+ } else if (is_bad_inode(inode)) {
+ CERROR("bad inode %lx\n",inode->i_ino);
+ iput(inode);
+ inode = ERR_PTR(-ENOENT);
+ }
+ return inode;
+}
+
+static int osd_fid_lookup(const struct lu_env *env,
+ struct osd_object *obj, const struct lu_fid *fid)
+{
+ struct osd_thread_info *info;
+ struct lu_device *ldev = obj->oo_dt.do_lu.lo_dev;
+ struct osd_device *dev;
+ struct osd_inode_id *id;
+ struct osd_oi *oi;
+ struct inode *inode;
+ int result;
+
+ LINVRNT(osd_invariant(obj));
+ LASSERT(obj->oo_inode == NULL);
+ LASSERT(fid_is_sane(fid));
+ /*
+ * This assertion checks that osd layer sees only local
+ * fids. Unfortunately it is somewhat expensive (does a
+ * cache-lookup). Disabling it for production/acceptance-testing.
+ */
+ LASSERT(1 || fid_is_local(env, ldev->ld_site, fid));
+
+ ENTRY;
+
+ info = osd_oti_get(env);
+ dev = osd_dev(ldev);
+ id = &info->oti_id;
+ oi = &dev->od_oi;
+
+ if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT))
+ RETURN(-ENOENT);
+
+ result = osd_oi_lookup(info, oi, fid, id);
+ if (result == 0) {
+ inode = osd_iget(info, dev, id);
+ if (!IS_ERR(inode)) {
+ obj->oo_inode = inode;
+ LASSERT(obj->oo_inode->i_sb == osd_sb(dev));
+ if (dev->od_iop_mode) {
+ obj->oo_compat_dot_created = 1;
+ obj->oo_compat_dotdot_created = 1;
+ }
+ result = 0;
+ } else
+ /*
+ * If fid wasn't found in oi, inode-less object is
+ * created, for which lu_object_exists() returns
+ * false. This is used in a (frequent) case when
+ * objects are created as locking anchors or
+ * place holders for objects yet to be created.
+ */
+ result = PTR_ERR(inode);
+ } else if (result == -ENOENT)
+ result = 0;
+ LINVRNT(osd_invariant(obj));
+
+ RETURN(result);
+}
+
+/*
* Concurrency: shouldn't matter.
*/
static void osd_object_init0(struct osd_object *obj)
* life-cycle.
*/
static int osd_object_init(const struct lu_env *env, struct lu_object *l,
- const struct lu_object_conf *_)
+ const struct lu_object_conf *unused)
{
struct osd_object *obj = osd_obj(l);
int result;
OBD_FREE_PTR(obj);
}
+/**
+ * IAM Iterator
+ */
static struct iam_path_descr *osd_it_ipd_get(const struct lu_env *env,
const struct iam_container *bag)
{
};
/*
+ * Journal
+ */
+
+/*
+ * Concurrency: doesn't access mutable data.
+ */
+static int osd_param_is_sane(const struct osd_device *dev,
+ const struct txn_param *param)
+{
+ return param->tp_credits <= osd_journal(dev)->j_max_transaction_buffers;
+}
+
+/*
+ * Concurrency: shouldn't matter.
+ */
+static void osd_trans_commit_cb(struct journal_callback *jcb, int error)
+{
+ struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
+ struct thandle *th = &oh->ot_super;
+ struct dt_device *dev = th->th_dev;
+ struct lu_device *lud = &dev->dd_lu_dev;
+
+ LASSERT(dev != NULL);
+ LASSERT(oh->ot_handle == NULL);
+
+ if (error) {
+ CERROR("transaction @0x%p commit error: %d\n", th, error);
+ } else {
+ struct lu_env *env = &osd_dt_dev(dev)->od_env_for_commit;
+ /*
+ * This od_env_for_commit is only for commit usage. see
+ * "struct dt_device"
+ */
+ lu_context_enter(&env->le_ctx);
+ dt_txn_hook_commit(env, th);
+ lu_context_exit(&env->le_ctx);
+ }
+
+ lu_ref_del_at(&lud->ld_reference, oh->ot_dev_link, "osd-tx", th);
+ lu_device_put(lud);
+ th->th_dev = NULL;
+
+ lu_context_exit(&th->th_ctx);
+ lu_context_fini(&th->th_ctx);
+ OBD_FREE_PTR(oh);
+}
+
+/*
+ * Concurrency: shouldn't matter.
+ */
+static struct thandle *osd_trans_start(const struct lu_env *env,
+ struct dt_device *d,
+ struct txn_param *p)
+{
+ struct osd_device *dev = osd_dt_dev(d);
+ handle_t *jh;
+ struct osd_thandle *oh;
+ struct thandle *th;
+ int hook_res;
+
+ ENTRY;
+
+ hook_res = dt_txn_hook_start(env, d, p);
+ if (hook_res != 0)
+ RETURN(ERR_PTR(hook_res));
+
+ if (osd_param_is_sane(dev, p)) {
+ OBD_ALLOC_GFP(oh, sizeof *oh, CFS_ALLOC_IO);
+ if (oh != NULL) {
+ struct osd_thread_info *oti = osd_oti_get(env);
+
+ /*
+ * XXX temporary stuff. Some abstraction layer should
+ * be used.
+ */
+
+ jh = ldiskfs_journal_start_sb(osd_sb(dev), p->tp_credits);
+ if (!IS_ERR(jh)) {
+ oh->ot_handle = jh;
+ th = &oh->ot_super;
+ th->th_dev = d;
+ th->th_result = 0;
+ jh->h_sync = p->tp_sync;
+ lu_device_get(&d->dd_lu_dev);
+ oh->ot_dev_link = lu_ref_add
+ (&d->dd_lu_dev.ld_reference,
+ "osd-tx", th);
+ /* add commit callback */
+ lu_context_init(&th->th_ctx, LCT_TX_HANDLE);
+ lu_context_enter(&th->th_ctx);
+ osd_journal_callback_set(jh, osd_trans_commit_cb,
+ (struct journal_callback *)&oh->ot_jcb);
+ LASSERT(oti->oti_txns == 0);
+ LASSERT(oti->oti_r_locks == 0);
+ LASSERT(oti->oti_w_locks == 0);
+ oti->oti_txns++;
+ } else {
+ OBD_FREE_PTR(oh);
+ th = (void *)jh;
+ }
+ } else
+ th = ERR_PTR(-ENOMEM);
+ } else {
+ CERROR("Invalid transaction parameters\n");
+ th = ERR_PTR(-EINVAL);
+ }
+
+ RETURN(th);
+}
+
+/*
+ * Concurrency: shouldn't matter.
+ */
+static void osd_trans_stop(const struct lu_env *env, struct thandle *th)
+{
+ int result;
+ struct osd_thandle *oh;
+ struct osd_thread_info *oti = osd_oti_get(env);
+
+ ENTRY;
+
+ oh = container_of0(th, struct osd_thandle, ot_super);
+ if (oh->ot_handle != NULL) {
+ handle_t *hdl = oh->ot_handle;
+
+ LASSERT(oti->oti_txns == 1);
+ oti->oti_txns--;
+ LASSERT(oti->oti_r_locks == 0);
+ LASSERT(oti->oti_w_locks == 0);
+ result = dt_txn_hook_stop(env, th);
+ if (result != 0)
+ CERROR("Failure in transaction hook: %d\n", result);
+ oh->ot_handle = NULL;
+ result = ldiskfs_journal_stop(hdl);
+ if (result != 0)
+ CERROR("Failure to stop transaction: %d\n", result);
+ }
+ EXIT;
+}
+
+/*
* Concurrency: no concurrent access is possible that late in object
* life-cycle.
*/
struct osd_device *osd = osd_obj2dev(obj);
struct osd_thread_info *oti = osd_oti_get(env);
struct txn_param *prm = &oti->oti_txn;
+ struct lu_env *env_del_obj = &oti->oti_obj_delete_tx_env;
struct thandle *th;
int result;
+ lu_env_init(env_del_obj, LCT_DT_THREAD);
txn_param_init(prm, OSD_TXN_OI_DELETE_CREDITS +
OSD_TXN_INODE_DELETE_CREDITS);
- th = osd_trans_start(env, &osd->od_dt_dev, prm);
+ th = osd_trans_start(env_del_obj, &osd->od_dt_dev, prm);
if (!IS_ERR(th)) {
- result = osd_oi_delete(oti, &osd->od_oi, fid, th);
- osd_trans_stop(env, th);
+ result = osd_oi_delete(osd_oti_get(env_del_obj),
+ &osd->od_oi, fid, th);
+ osd_trans_stop(env_del_obj, th);
} else
result = PTR_ERR(th);
+
+ lu_env_fini(env_del_obj);
return result;
}
}
/*
- * Journal
- */
-
-/*
- * Concurrency: doesn't access mutable data.
- */
-static int osd_param_is_sane(const struct osd_device *dev,
- const struct txn_param *param)
-{
- return param->tp_credits <= osd_journal(dev)->j_max_transaction_buffers;
-}
-
-/*
- * Concurrency: shouldn't matter.
- */
-static void osd_trans_commit_cb(struct journal_callback *jcb, int error)
-{
- struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
- struct thandle *th = &oh->ot_super;
- struct dt_device *dev = th->th_dev;
- struct lu_device *lud = &dev->dd_lu_dev;
-
- LASSERT(dev != NULL);
- LASSERT(oh->ot_handle == NULL);
-
- if (error) {
- CERROR("transaction @0x%p commit error: %d\n", th, error);
- } else {
- struct lu_env *env = &osd_dt_dev(dev)->od_env_for_commit;
- /*
- * This od_env_for_commit is only for commit usage. see
- * "struct dt_device"
- */
- lu_context_enter(&env->le_ctx);
- dt_txn_hook_commit(env, th);
- lu_context_exit(&env->le_ctx);
- }
-
- lu_ref_del_at(&lud->ld_reference, oh->ot_dev_link, "osd-tx", th);
- lu_device_put(lud);
- th->th_dev = NULL;
-
- lu_context_exit(&th->th_ctx);
- lu_context_fini(&th->th_ctx);
- OBD_FREE_PTR(oh);
-}
-
-/*
- * Concurrency: shouldn't matter.
- */
-static struct thandle *osd_trans_start(const struct lu_env *env,
- struct dt_device *d,
- struct txn_param *p)
-{
- struct osd_device *dev = osd_dt_dev(d);
- handle_t *jh;
- struct osd_thandle *oh;
- struct thandle *th;
- int hook_res;
-
- ENTRY;
-
- hook_res = dt_txn_hook_start(env, d, p);
- if (hook_res != 0)
- RETURN(ERR_PTR(hook_res));
-
- if (osd_param_is_sane(dev, p)) {
- OBD_ALLOC_GFP(oh, sizeof *oh, CFS_ALLOC_IO);
- if (oh != NULL) {
- struct osd_thread_info *oti = osd_oti_get(env);
-
- /*
- * XXX temporary stuff. Some abstraction layer should
- * be used.
- */
-
- jh = journal_start(osd_journal(dev), p->tp_credits);
- if (!IS_ERR(jh)) {
- oh->ot_handle = jh;
- th = &oh->ot_super;
- th->th_dev = d;
- th->th_result = 0;
- jh->h_sync = p->tp_sync;
- lu_device_get(&d->dd_lu_dev);
- oh->ot_dev_link = lu_ref_add
- (&d->dd_lu_dev.ld_reference,
- "osd-tx", th);
- /* add commit callback */
- lu_context_init(&th->th_ctx, LCT_TX_HANDLE);
- lu_context_enter(&th->th_ctx);
- journal_callback_set(jh, osd_trans_commit_cb,
- (struct journal_callback *)&oh->ot_jcb);
- LASSERT(oti->oti_txns == 0);
- LASSERT(oti->oti_r_locks == 0);
- LASSERT(oti->oti_w_locks == 0);
- oti->oti_txns++;
- } else {
- OBD_FREE_PTR(oh);
- th = (void *)jh;
- }
- } else
- th = ERR_PTR(-ENOMEM);
- } else {
- CERROR("Invalid transaction parameters\n");
- th = ERR_PTR(-EINVAL);
- }
-
- RETURN(th);
-}
-
-/*
- * Concurrency: shouldn't matter.
- */
-static void osd_trans_stop(const struct lu_env *env, struct thandle *th)
-{
- int result;
- struct osd_thandle *oh;
- struct osd_thread_info *oti = osd_oti_get(env);
-
- ENTRY;
-
- oh = container_of0(th, struct osd_thandle, ot_super);
- if (oh->ot_handle != NULL) {
- handle_t *hdl = oh->ot_handle;
-
- LASSERT(oti->oti_txns == 1);
- oti->oti_txns--;
- LASSERT(oti->oti_r_locks == 0);
- LASSERT(oti->oti_w_locks == 0);
- result = dt_txn_hook_stop(env, th);
- if (result != 0)
- CERROR("Failure in transaction hook: %d\n", result);
- oh->ot_handle = NULL;
- result = journal_stop(hdl);
- if (result != 0)
- CERROR("Failure to stop transaction: %d\n", result);
- }
- EXIT;
-}
-
-/*
* Concurrency: shouldn't matter.
*/
static int osd_sync(const struct lu_env *env, struct dt_device *d)
/**
* Xattr set. The same as xattr of EXT3.
* DATA_TRANS_BLOCKS(14)
- * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS are
- * also counted in. Do not know why?
+ * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
+ * are also counted in. Do not know why?
*/
[DTO_XATTR_SET] = 14,
[DTO_LOG_REC] = 14,
[DTO_WRITE_BLOCK] = 14,
/**
* Attr set credits for chown.
- * 3 (inode bit, group, GDT)
+ * This is extra credits for setattr, and it is null without quota
*/
- [DTO_ATTR_SET_CHOWN]= 3
+ [DTO_ATTR_SET_CHOWN]= 0
};
/**
[DTO_INDEX_DELETE] = 20,
/**
* Unused now.
- */
+ */
[DTO_IDNEX_UPDATE] = 16,
/*
* Create a object. Same as create object in EXT3 filesystem.
* INDEX_EXTRA_BLOCKS(8) +
* 3(inode bits, groups, GDT) +
* QUOTA(?)
- */
+ */
[DTO_OBJECT_DELETE] = 27,
/**
* Attr set credits.
[DTO_WRITE_BLOCK] = 16,
/**
* Attr set credits for chown.
- * 3 (inode bit, group, GDT) +
+ * It is added to already set setattr credits
* 2 * QUOTA_INIT_BLOCKS(25) +
* 2 * QUOTA_DEL_BLOCKS(9)
*/
- [DTO_ATTR_SET_CHOWN]= 71
+ [DTO_ATTR_SET_CHOWN]= 68,
};
static int osd_credit_get(const struct lu_env *env, struct dt_device *d,
up_write(&obj->oo_sem);
}
+static int osd_object_write_locked(const struct lu_env *env,
+ struct dt_object *dt)
+{
+ struct osd_object *obj = osd_dt_obj(dt);
+
+ LINVRNT(osd_invariant(obj));
+
+ return obj->oo_owner == env;
+}
+
static int capa_is_sane(const struct lu_env *env,
struct osd_device *dev,
struct lustre_capa *capa,
return 0;
}
+static struct timespec *osd_inode_time(const struct lu_env *env,
+ struct inode *inode, __u64 seconds)
+{
+ struct osd_thread_info *oti = osd_oti_get(env);
+ struct timespec *t = &oti->oti_time;
+
+ t->tv_sec = seconds;
+ t->tv_nsec = 0;
+ *t = timespec_trunc(*t, get_sb_time_gran(inode->i_sb));
+ return t;
+}
+
+
+static void osd_inode_getattr(const struct lu_env *env,
+ struct inode *inode, struct lu_attr *attr)
+{
+ attr->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
+ LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
+ LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE;
+
+ attr->la_atime = LTIME_S(inode->i_atime);
+ attr->la_mtime = LTIME_S(inode->i_mtime);
+ attr->la_ctime = LTIME_S(inode->i_ctime);
+ attr->la_mode = inode->i_mode;
+ attr->la_size = i_size_read(inode);
+ attr->la_blocks = inode->i_blocks;
+ attr->la_uid = inode->i_uid;
+ attr->la_gid = inode->i_gid;
+ attr->la_flags = LDISKFS_I(inode)->i_flags;
+ attr->la_nlink = inode->i_nlink;
+ attr->la_rdev = inode->i_rdev;
+ attr->la_blksize = ll_inode_blksize(inode);
+ attr->la_blkbits = inode->i_blkbits;
+}
+
static int osd_attr_get(const struct lu_env *env,
struct dt_object *dt,
struct lu_attr *attr,
return 0;
}
-static int osd_attr_set(const struct lu_env *env,
- struct dt_object *dt,
- const struct lu_attr *attr,
- struct thandle *handle,
- struct lustre_capa *capa)
-{
- struct osd_object *obj = osd_dt_obj(dt);
- int rc;
-
- LASSERT(handle != NULL);
- LASSERT(dt_object_exists(dt));
- LASSERT(osd_invariant(obj));
-
- if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
- return -EACCES;
-
- spin_lock(&obj->oo_guard);
- rc = osd_inode_setattr(env, obj->oo_inode, attr);
- spin_unlock(&obj->oo_guard);
-
- if (!rc)
- mark_inode_dirty(obj->oo_inode);
- return rc;
-}
-
-static struct timespec *osd_inode_time(const struct lu_env *env,
- struct inode *inode, __u64 seconds)
-{
- struct osd_thread_info *oti = osd_oti_get(env);
- struct timespec *t = &oti->oti_time;
-
- t->tv_sec = seconds;
- t->tv_nsec = 0;
- *t = timespec_trunc(*t, get_sb_time_gran(inode->i_sb));
- return t;
-}
-
static int osd_inode_setattr(const struct lu_env *env,
struct inode *inode, const struct lu_attr *attr)
{
LDISKFS_I(inode)->i_disksize = attr->la_size;
i_size_write(inode, attr->la_size);
}
-# if 0
- /*
- * OSD should not change "i_blocks" which is used by quota.
+
+ /* OSD should not change "i_blocks" which is used by quota.
* "i_blocks" should be changed by ldiskfs only.
- * Disable this assignment until SOM to fix some EA field. */
+ * Enable this assignment for SOM purpose now, until it is
+ * stored in SOM EA. */
if (bits & LA_BLOCKS)
inode->i_blocks = attr->la_blocks;
-#endif
+
if (bits & LA_MODE)
inode->i_mode = (inode->i_mode & S_IFMT) |
(attr->la_mode & ~S_IFMT);
return 0;
}
+static int osd_attr_set(const struct lu_env *env,
+ struct dt_object *dt,
+ const struct lu_attr *attr,
+ struct thandle *handle,
+ struct lustre_capa *capa)
+{
+ struct osd_object *obj = osd_dt_obj(dt);
+ int rc;
+
+ LASSERT(handle != NULL);
+ LASSERT(dt_object_exists(dt));
+ LASSERT(osd_invariant(obj));
+
+ if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
+ return -EACCES;
+
+ spin_lock(&obj->oo_guard);
+ rc = osd_inode_setattr(env, obj->oo_inode, attr);
+ spin_unlock(&obj->oo_guard);
+
+ if (!rc)
+ mark_inode_dirty(obj->oo_inode);
+ return rc;
+}
+
/*
* Object creation.
*
* XXX temporary solution.
*/
-
static int osd_create_pre(struct osd_thread_info *info, struct osd_object *obj,
struct lu_attr *attr, struct thandle *th)
{
static int osd_create_post(struct osd_thread_info *info, struct osd_object *obj,
struct lu_attr *attr, struct thandle *th)
{
- LASSERT(obj->oo_inode != NULL);
-
osd_object_init0(obj);
return 0;
}
-extern struct inode *ldiskfs_create_inode(handle_t *handle,
- struct inode * dir, int mode);
-extern int ldiskfs_add_entry(handle_t *handle, struct dentry *dentry,
- struct inode *inode);
-extern int ldiskfs_delete_entry(handle_t *handle,
- struct inode * dir,
- struct ldiskfs_dir_entry_2 * de_del,
- struct buffer_head * bh);
-extern struct buffer_head * ldiskfs_find_entry(struct dentry *dentry,
- struct ldiskfs_dir_entry_2
- ** res_dir);
-extern int ldiskfs_add_dot_dotdot(handle_t *handle, struct inode *dir,
- struct inode *inode);
-
-extern int ldiskfs_xattr_set_handle(handle_t *handle, struct inode *inode,
- int name_index, const char *name,
- const void *value, size_t value_len,
- int flags);
-
static struct dentry * osd_child_dentry_get(const struct lu_env *env,
struct osd_object *obj,
const char *name,
return result;
}
-
-extern int iam_lvar_create(struct inode *obj, int keysize, int ptrsize,
- int recsize, handle_t *handle);
-
-extern int iam_lfix_create(struct inode *obj, int keysize, int ptrsize,
- int recsize, handle_t *handle);
-
-
enum {
OSD_NAME_LEN = 255
};
*/
result = iam_lvar_create(obj->oo_inode, OSD_NAME_LEN, 4,
- sizeof (struct lu_fid_pack),
+ sizeof (struct osd_fid_pack),
oth->ot_handle);
}
return result;
*t = inode->i_ctime;
rc = inode->i_op->setxattr(dentry, name, buf->lb_buf,
buf->lb_len, fs_flags);
- if (likely(rc == 0)) {
- spin_lock(&obj->oo_guard);
- inode->i_ctime = *t;
- spin_unlock(&obj->oo_guard);
- mark_inode_dirty(inode);
- }
+ /* ctime should not be updated with server-side time. */
+ spin_lock(&obj->oo_guard);
+ inode->i_ctime = *t;
+ spin_unlock(&obj->oo_guard);
+ mark_inode_dirty(inode);
return rc;
}
/**
* Helper function to form igif
*/
-static inline void osd_igif_get(const struct lu_env *env, struct dentry *dentry,
+static inline void osd_igif_get(const struct lu_env *env, struct inode *inode,
struct lu_fid *fid)
{
- struct inode *inode = dentry->d_inode;
lu_igif_build(fid, inode->i_ino, inode->i_generation);
}
/**
* Helper function to pack the fid
*/
-static inline void osd_fid_pack(const struct lu_env *env, const struct lu_fid *fid,
- struct lu_fid_pack *pack)
+void osd_fid_pack(struct osd_fid_pack *pack, const struct dt_rec *fid,
+ struct lu_fid *befider)
{
- fid_pack(pack, fid, &osd_oti_get(env)->oti_fid);
+ fid_cpu_to_be(befider, (struct lu_fid *)fid);
+ memcpy(pack->fp_area, befider, sizeof(*befider));
+ pack->fp_len = sizeof(*befider) + 1;
+}
+
+int osd_fid_unpack(struct lu_fid *fid, const struct osd_fid_pack *pack)
+{
+ int result;
+
+ result = 0;
+ switch (pack->fp_len) {
+ case sizeof *fid + 1:
+ memcpy(fid, pack->fp_area, sizeof *fid);
+ fid_be_to_cpu(fid, fid);
+ break;
+ default:
+ CERROR("Unexpected packed fid size: %d\n", pack->fp_len);
+ result = -EIO;
+ }
+ return result;
}
/**
* Try to read the fid from inode ea into dt_rec, if return value
* i.e. rc is +ve, then we got fid, otherwise we will have to form igif
*
- * \param rec, the data-structure into which fid/igif is read
+ * \param fid, object fid.
*
* \retval 0, on success
*/
-static int osd_ea_fid_get(const struct lu_env *env, struct dentry *dentry,
- struct dt_rec *rec)
+static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
+ __u32 ino, struct lu_fid *fid)
{
- struct inode *inode = dentry->d_inode;
struct osd_thread_info *info = osd_oti_get(env);
struct lustre_mdt_attrs *mdt_attrs = &info->oti_mdt_attrs;
- struct lu_fid *fid = &info->oti_fid;
- int rc;
+ struct lu_device *ldev = obj->oo_dt.do_lu.lo_dev;
+ struct dentry *dentry = &info->oti_child_dentry;
+ struct osd_inode_id *id = &info->oti_id;
+ struct osd_device *dev;
+ struct inode *inode;
+ int rc;
- LASSERT(inode->i_op != NULL && inode->i_op->getxattr != NULL);
+ ENTRY;
+ dev = osd_dev(ldev);
+
+ id->oii_ino = ino;
+ id->oii_gen = OSD_OII_NOGEN;
+
+ inode = osd_iget(info, dev, id);
+ if (IS_ERR(inode)) {
+ rc = PTR_ERR(inode);
+ GOTO(out,rc);
+ }
+ dentry->d_inode = inode;
+ LASSERT(inode->i_op != NULL && inode->i_op->getxattr != NULL);
rc = inode->i_op->getxattr(dentry, XATTR_NAME_LMA, (void *)mdt_attrs,
sizeof *mdt_attrs);
+ /* Check LMA compatibility */
+ if (rc > 0 &&
+ (mdt_attrs->lma_incompat & ~cpu_to_be32(LMA_INCOMPAT_SUPP))) {
+ CWARN("Inode %lx: Unsupported incompat LMA feature(s) %#x\n",
+ inode->i_ino, be32_to_cpu(mdt_attrs->lma_incompat) &
+ ~LMA_INCOMPAT_SUPP);
+ return -ENOSYS;
+ }
+
if (rc > 0) {
fid_be_to_cpu(fid, &mdt_attrs->lma_self_fid);
rc = 0;
} else if (rc == -ENODATA) {
- osd_igif_get(env, dentry, fid);
+ osd_igif_get(env, inode, fid);
rc = 0;
}
+ iput(inode);
- if (rc == 0)
- osd_fid_pack(env, fid, (struct lu_fid_pack*)rec);
-
- return rc;
+out:
+ RETURN(rc);
}
/**
return inode->i_op->getxattr(dentry, name, buf->lb_buf, buf->lb_len);
}
-
/*
* Concurrency: @dt is write locked.
*/
dentry->d_inode = inode;
*t = inode->i_ctime;
rc = inode->i_op->removexattr(dentry, name);
- if (likely(rc == 0)) {
- /* ctime should not be updated with server-side time. */
- spin_lock(&obj->oo_guard);
- inode->i_ctime = *t;
- spin_unlock(&obj->oo_guard);
- mark_inode_dirty(inode);
- }
+ /* ctime should not be updated with server-side time. */
+ spin_lock(&obj->oo_guard);
+ inode->i_ctime = *t;
+ spin_unlock(&obj->oo_guard);
+ mark_inode_dirty(inode);
return rc;
}
RETURN(rc);
}
+/*
+ * Get the 64-bit version for an inode.
+ */
+static dt_obj_version_t osd_object_version_get(const struct lu_env *env,
+ struct dt_object *dt)
+{
+ struct inode *inode = osd_dt_obj(dt)->oo_inode;
+
+ CDEBUG(D_INFO, "Get version "LPX64" for inode %lu\n",
+ LDISKFS_I(inode)->i_fs_version, inode->i_ino);
+ return LDISKFS_I(inode)->i_fs_version;
+}
+
+/*
+ * Set the 64-bit version and return the old version.
+ */
+static void osd_object_version_set(const struct lu_env *env, struct dt_object *dt,
+ dt_obj_version_t new_version)
+{
+ struct inode *inode = osd_dt_obj(dt)->oo_inode;
+
+ CDEBUG(D_INFO, "Set version "LPX64" (old "LPX64") for inode %lu\n",
+ new_version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
+ LDISKFS_I(inode)->i_fs_version = new_version;
+ /** Version is set after all inode operations are finished,
+ * so we should mark it dirty here */
+ inode->i_sb->s_op->dirty_inode(inode);
+}
+
static int osd_data_get(const struct lu_env *env, struct dt_object *dt,
void **data)
{
struct osd_object *obj = osd_dt_obj(dt);
ENTRY;
- *data = (void *)obj->oo_inode;
- RETURN(0);
+ *data = (void *)obj->oo_inode;
+ RETURN(0);
+}
+
+/*
+ * Index operations.
+ */
+
+static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
+ const struct dt_index_features *feat)
+{
+ struct iam_descr *descr;
+
+ if (osd_object_is_root(o))
+ return feat == &dt_directory_features;
+
+ LASSERT(o->oo_dir != NULL);
+
+ descr = o->oo_dir->od_container.ic_descr;
+ if (feat == &dt_directory_features) {
+ if (descr->id_rec_size == sizeof(struct osd_fid_pack))
+ return 1;
+ else
+ return 0;
+ } else {
+ return
+ feat->dif_keysize_min <= descr->id_key_size &&
+ descr->id_key_size <= feat->dif_keysize_max &&
+ feat->dif_recsize_min <= descr->id_rec_size &&
+ descr->id_rec_size <= feat->dif_recsize_max &&
+ !(feat->dif_flags & (DT_IND_VARKEY |
+ DT_IND_VARREC | DT_IND_NONUNQ)) &&
+ ergo(feat->dif_flags & DT_IND_UPDATE,
+ 1 /* XXX check that object (and file system) is
+ * writable */);
+ }
+}
+
+static int osd_iam_container_init(const struct lu_env *env,
+ struct osd_object *obj,
+ struct osd_directory *dir)
+{
+ int result;
+ struct iam_container *bag;
+
+ bag = &dir->od_container;
+ result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
+ if (result == 0) {
+ result = iam_container_setup(bag);
+ if (result == 0)
+ obj->oo_dt.do_index_ops = &osd_index_iam_ops;
+ else
+ iam_container_fini(bag);
+ }
+ return result;
+}
+
+
+/*
+ * Concurrency: no external locking is necessary.
+ */
+static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
+ const struct dt_index_features *feat)
+{
+ int result;
+ int ea_dir = 0;
+ struct osd_object *obj = osd_dt_obj(dt);
+ struct osd_device *osd = osd_obj2dev(obj);
+
+ LINVRNT(osd_invariant(obj));
+ LASSERT(dt_object_exists(dt));
+
+ if (osd_object_is_root(obj)) {
+ dt->do_index_ops = &osd_index_ea_ops;
+ result = 0;
+ } else if (feat == &dt_directory_features && osd->od_iop_mode) {
+ dt->do_index_ops = &osd_index_ea_ops;
+ if (S_ISDIR(obj->oo_inode->i_mode))
+ result = 0;
+ else
+ result = -ENOTDIR;
+ ea_dir = 1;
+ } else if (!osd_has_index(obj)) {
+ struct osd_directory *dir;
+
+ OBD_ALLOC_PTR(dir);
+ if (dir != NULL) {
+
+ spin_lock(&obj->oo_guard);
+ if (obj->oo_dir == NULL)
+ obj->oo_dir = dir;
+ else
+ /*
+ * Concurrent thread allocated container data.
+ */
+ OBD_FREE_PTR(dir);
+ spin_unlock(&obj->oo_guard);
+ /*
+ * Now, that we have container data, serialize its
+ * initialization.
+ */
+ down_write(&obj->oo_ext_idx_sem);
+ /*
+ * recheck under lock.
+ */
+ if (!osd_has_index(obj))
+ result = osd_iam_container_init(env, obj, dir);
+ else
+ result = 0;
+ up_write(&obj->oo_ext_idx_sem);
+ } else
+ result = -ENOMEM;
+ } else
+ result = 0;
+
+ if (result == 0 && ea_dir == 0) {
+ if (!osd_iam_index_probe(env, obj, feat))
+ result = -ENOTDIR;
+ }
+ LINVRNT(osd_invariant(obj));
+
+ return result;
}
static const struct dt_object_operations osd_obj_ops = {
.do_write_lock = osd_object_write_lock,
.do_read_unlock = osd_object_read_unlock,
.do_write_unlock = osd_object_write_unlock,
+ .do_write_locked = osd_object_write_locked,
.do_attr_get = osd_attr_get,
.do_attr_set = osd_attr_set,
.do_ah_init = osd_ah_init,
.do_xattr_list = osd_xattr_list,
.do_capa_get = osd_capa_get,
.do_object_sync = osd_object_sync,
+ .do_version_get = osd_object_version_get,
+ .do_version_set = osd_object_version_set,
.do_data_get = osd_data_get,
};
.do_write_lock = osd_object_write_lock,
.do_read_unlock = osd_object_read_unlock,
.do_write_unlock = osd_object_write_unlock,
+ .do_write_locked = osd_object_write_locked,
.do_attr_get = osd_attr_get,
.do_attr_set = osd_attr_set,
.do_ah_init = osd_ah_init,
.do_xattr_list = osd_xattr_list,
.do_capa_get = osd_capa_get,
.do_object_sync = osd_object_sync,
+ .do_version_get = osd_object_version_get,
+ .do_version_set = osd_object_version_set,
.do_data_get = osd_data_get,
};
.dbo_write = osd_write
};
-/*
- * Index operations.
- */
-
-static int osd_object_is_root(const struct osd_object *obj)
-{
- return osd_sb(osd_obj2dev(obj))->s_root->d_inode == obj->oo_inode;
-}
-
-static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
- const struct dt_index_features *feat)
-{
- struct iam_descr *descr;
- struct dt_object *dt = &o->oo_dt;
-
- if (osd_object_is_root(o))
- return feat == &dt_directory_features;
-
- LASSERT(o->oo_dir != NULL);
-
- descr = o->oo_dir->od_container.ic_descr;
- if (feat == &dt_directory_features) {
- if (descr->id_rec_size == sizeof(struct lu_fid_pack))
- return 1;
-
- if (descr == &iam_htree_compat_param) {
- /* if it is a HTREE dir then there is good chance that,
- * we dealing with ext3 directory here with no FIDs. */
-
- if (descr->id_rec_size ==
- sizeof ((struct ldiskfs_dir_entry_2 *)NULL)->inode) {
-
- dt->do_index_ops = &osd_index_ea_ops;
- return 1;
- }
- }
- return 0;
- } else {
- return
- feat->dif_keysize_min <= descr->id_key_size &&
- descr->id_key_size <= feat->dif_keysize_max &&
- feat->dif_recsize_min <= descr->id_rec_size &&
- descr->id_rec_size <= feat->dif_recsize_max &&
- !(feat->dif_flags & (DT_IND_VARKEY |
- DT_IND_VARREC | DT_IND_NONUNQ)) &&
- ergo(feat->dif_flags & DT_IND_UPDATE,
- 1 /* XXX check that object (and file system) is
- * writable */);
- }
-}
-
-static int osd_iam_container_init(const struct lu_env *env,
- struct osd_object *obj,
- struct osd_directory *dir)
-{
- int result;
- struct iam_container *bag;
-
- bag = &dir->od_container;
- result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
- if (result == 0) {
- result = iam_container_setup(bag);
- if (result == 0)
- obj->oo_dt.do_index_ops = &osd_index_iam_ops;
- else
- iam_container_fini(bag);
- }
- return result;
-}
-
-/*
- * Concurrency: no external locking is necessary.
- */
-static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
- const struct dt_index_features *feat)
-{
- int result;
- int ea_dir = 0;
- struct osd_object *obj = osd_dt_obj(dt);
- struct osd_device *osd = osd_obj2dev(obj);
-
- LINVRNT(osd_invariant(obj));
- LASSERT(dt_object_exists(dt));
-
- if (osd_object_is_root(obj)) {
- dt->do_index_ops = &osd_index_ea_ops;
- result = 0;
- } else if (feat == &dt_directory_features && osd->od_iop_mode) {
- dt->do_index_ops = &osd_index_ea_ops;
- if (S_ISDIR(obj->oo_inode->i_mode))
- result = 0;
- else
- result = -ENOTDIR;
- ea_dir = 1;
- } else if (!osd_has_index(obj)) {
- struct osd_directory *dir;
-
- OBD_ALLOC_PTR(dir);
- if (dir != NULL) {
- sema_init(&dir->od_sem, 1);
-
- spin_lock(&obj->oo_guard);
- if (obj->oo_dir == NULL)
- obj->oo_dir = dir;
- else
- /*
- * Concurrent thread allocated container data.
- */
- OBD_FREE_PTR(dir);
- spin_unlock(&obj->oo_guard);
- /*
- * Now, that we have container data, serialize its
- * initialization.
- */
- down(&obj->oo_dir->od_sem);
- /*
- * recheck under lock.
- */
- if (!osd_has_index(obj))
- result = osd_iam_container_init(env, obj, dir);
- else
- result = 0;
- up(&obj->oo_dir->od_sem);
- } else
- result = -ENOMEM;
- } else
- result = 0;
-
- if (result == 0 && ea_dir == 0) {
- if (!osd_iam_index_probe(env, obj, feat))
- result = -ENOTDIR;
- }
- LINVRNT(osd_invariant(obj));
-
- return result;
-}
/**
* delete a (key, value) pair from index \a dt specified by \a key
dentry = osd_child_dentry_get(env, obj,
(char *)key, strlen((char *)key));
+
+ down_write(&obj->oo_ext_idx_sem);
bh = ldiskfs_find_entry(dentry, &de);
if (bh) {
+ struct osd_thread_info *oti = osd_oti_get(env);
+ struct timespec *ctime = &oti->oti_time;
+ struct timespec *mtime = &oti->oti_time2;
+
+ *ctime = dir->i_ctime;
+ *mtime = dir->i_mtime;
rc = ldiskfs_delete_entry(oh->ot_handle,
dir, de, bh);
- if (!rc)
- mark_inode_dirty(dir);
+ /* xtime should not be updated with server-side time. */
+ spin_lock(&obj->oo_guard);
+ dir->i_ctime = *ctime;
+ dir->i_mtime = *mtime;
+ spin_unlock(&obj->oo_guard);
+ mark_inode_dirty(dir);
brelse(bh);
} else
rc = -ENOENT;
+ up_write(&obj->oo_ext_idx_sem);
LASSERT(osd_invariant(obj));
RETURN(rc);
}
struct iam_container *bag = &obj->oo_dir->od_container;
struct osd_thread_info *oti = osd_oti_get(env);
struct iam_iterator *it = &oti->oti_idx_it;
+ struct iam_rec *iam_rec;
int rc;
ENTRY;
iam_it_init(it, bag, 0, ipd);
rc = iam_it_get(it, (struct iam_key *)key);
- if (rc >= 0)
- iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)rec);
+ if (rc >= 0) {
+ if (S_ISDIR(obj->oo_inode->i_mode))
+ iam_rec = (struct iam_rec *)oti->oti_fid_packed;
+ else
+ iam_rec = (struct iam_rec *) rec;
+ iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
+ if (S_ISDIR(obj->oo_inode->i_mode))
+ osd_fid_unpack((struct lu_fid *) rec,
+ (struct osd_fid_pack *)iam_rec);
+ }
iam_it_put(it);
iam_it_fini(it);
osd_ipd_put(env, bag, ipd);
#ifdef HAVE_QUOTA_SUPPORT
cfs_cap_t save = current->cap_effective;
#endif
+ struct osd_thread_info *oti = osd_oti_get(env);
+ struct iam_rec *iam_rec = (struct iam_rec *)oti->oti_fid_packed;
int rc;
ENTRY;
else
current->cap_effective &= ~CFS_CAP_SYS_RESOURCE_MASK;
#endif
+ if (S_ISDIR(obj->oo_inode->i_mode))
+ osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid);
+ else
+ iam_rec = (struct iam_rec *) rec;
rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
- (struct iam_rec *)rec, ipd);
+ iam_rec, ipd);
#ifdef HAVE_QUOTA_SUPPORT
current->cap_effective = save;
#endif
}
/**
+ * Calls ldiskfs_add_entry() to add directory entry
+ * into the directory. This is required for
+ * interoperability mode (b11826)
+ *
+ * \retval 0, on success
+ * \retval -ve, on error
+ */
+static int __osd_ea_add_rec(struct osd_thread_info *info,
+ struct osd_object *pobj,
+ struct osd_object *cobj,
+ const char *name,
+ struct thandle *th)
+{
+ struct dentry *child;
+ struct osd_thandle *oth;
+ struct inode *cinode = cobj->oo_inode;
+ int rc;
+
+ oth = container_of(th, struct osd_thandle, ot_super);
+ LASSERT(oth->ot_handle != NULL);
+ LASSERT(oth->ot_handle->h_transaction != NULL);
+
+ child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
+ rc = ldiskfs_add_entry(oth->ot_handle, child, cinode);
+
+ RETURN(rc);
+}
+
+/**
* Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
* into the directory.Also sets flags into osd object to
* indicate dot and dotdot are created. This is required for
return result;
}
-/**
- * Calls ldiskfs_add_entry() to add directory entry
- * into the directory. This is required for
- * interoperability mode (b11826)
- *
- * \retval 0, on success
- * \retval -ve, on error
- */
-static int __osd_ea_add_rec(struct osd_thread_info *info,
- struct osd_object *pobj,
- struct osd_object *cobj,
- const char *name,
- struct thandle *th)
-{
- struct dentry *child;
- struct osd_thandle *oth;
- struct inode *cinode = cobj->oo_inode;
- int rc;
-
- oth = container_of(th, struct osd_thandle, ot_super);
- LASSERT(oth->ot_handle != NULL);
- LASSERT(oth->ot_handle->h_transaction != NULL);
-
- child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
- rc = ldiskfs_add_entry(oth->ot_handle, child, cinode);
-
- RETURN(rc);
-}
/**
* It will call the appropriate osd_add* function and return the
static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
struct dt_rec *rec, const struct dt_key *key)
{
- struct inode *dir = obj->oo_inode;
- struct osd_thread_info *info = osd_oti_get(env);
- struct dentry *dentry;
- struct osd_device *dev = osd_dev(obj->oo_dt.do_lu.lo_dev);
- struct osd_inode_id *id = &info->oti_id;
+ struct inode *dir = obj->oo_inode;
+ struct dentry *dentry;
struct ldiskfs_dir_entry_2 *de;
struct buffer_head *bh;
- struct inode *inode;
+ struct lu_fid *fid = (struct lu_fid *) rec;
int ino;
int rc;
dentry = osd_child_dentry_get(env, obj,
(char *)key, strlen((char *)key));
+
+ down_read(&obj->oo_ext_idx_sem);
bh = ldiskfs_find_entry(dentry, &de);
if (bh) {
ino = le32_to_cpu(de->inode);
brelse(bh);
- id->oii_ino = ino;
- id->oii_gen = OSD_OII_NOGEN;
-
- inode = osd_iget(info, dev, id);
- if (!IS_ERR(inode)) {
- dentry->d_inode = inode;
-
- rc = osd_ea_fid_get(env, dentry, rec);
- iput(inode);
- } else
- rc = -ENOENT;
+ rc = osd_ea_fid_get(env, obj, ino, fid);
} else
rc = -ENOENT;
+ up_read(&obj->oo_ext_idx_sem);
RETURN (rc);
}
struct lustre_capa *capa, int ignore_quota)
{
struct osd_object *obj = osd_dt_obj(dt);
- struct lu_fid *fid = &osd_oti_get(env)->oti_fid;
- const struct lu_fid_pack *pack = (const struct lu_fid_pack *)rec;
+ struct lu_fid *fid = (struct lu_fid *) rec;
const char *name = (const char *)key;
struct osd_object *child;
#ifdef HAVE_QUOTA_SUPPORT
- cfs_cap_t save = current->cap_effective;
+ cfs_cap_t save = current->cap_effective;
#endif
int rc;
if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
RETURN(-EACCES);
- rc = fid_unpack(pack, fid);
- if (rc != 0)
- RETURN(rc);
child = osd_object_find(env, dt, fid);
if (!IS_ERR(child)) {
+ struct inode *inode = obj->oo_inode;
+ struct osd_thread_info *oti = osd_oti_get(env);
+ struct timespec *ctime = &oti->oti_time;
+ struct timespec *mtime = &oti->oti_time2;
+
+ *ctime = inode->i_ctime;
+ *mtime = inode->i_mtime;
#ifdef HAVE_QUOTA_SUPPORT
if (ignore_quota)
current->cap_effective |= CFS_CAP_SYS_RESOURCE_MASK;
else
current->cap_effective &= ~CFS_CAP_SYS_RESOURCE_MASK;
#endif
+ down_write(&obj->oo_ext_idx_sem);
rc = osd_ea_add_rec(env, obj, child, name, th);
-
+ up_write(&obj->oo_ext_idx_sem);
#ifdef HAVE_QUOTA_SUPPORT
current->cap_effective = save;
#endif
osd_object_put(env, child);
+ /* xtime should not be updated with server-side time. */
+ spin_lock(&obj->oo_guard);
+ inode->i_ctime = *ctime;
+ inode->i_mtime = *mtime;
+ spin_unlock(&obj->oo_guard);
+ mark_inode_dirty(inode);
} else {
rc = PTR_ERR(child);
}
return iam_it_key_size(&it->oi_it);
}
+static inline void osd_it_append_attrs(struct lu_dirent*ent,
+ __u32 attr,
+ int len,
+ __u16 type)
+{
+ struct luda_type *lt;
+ const unsigned align = sizeof(struct luda_type) - 1;
+
+ /* check if file type is required */
+ if (attr & LUDA_TYPE) {
+ len = (len + align) & ~align;
+
+ lt = (void *) ent->lde_name + len;
+ lt->lt_type = cpu_to_le16(CFS_DTTOIF(type));
+ ent->lde_attrs |= LUDA_TYPE;
+ }
+
+ ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
+}
+
+/**
+ * build lu direct from backend fs dirent.
+ */
+
+static inline void osd_it_pack_dirent(struct lu_dirent *ent,
+ struct lu_fid *fid,
+ __u64 offset,
+ char *name,
+ __u16 namelen,
+ __u16 type,
+ __u32 attr)
+{
+ fid_cpu_to_le(&ent->lde_fid, fid);
+ ent->lde_attrs = LUDA_FID;
+
+ ent->lde_hash = cpu_to_le64(offset);
+ ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
+
+ strncpy(ent->lde_name, name, namelen);
+ ent->lde_namelen = cpu_to_le16(namelen);
+
+ /* append lustre attributes */
+ osd_it_append_attrs(ent, attr, namelen, type);
+}
+
/**
* Return pointer to the record under iterator.
*/
-static struct dt_rec *osd_it_iam_rec(const struct lu_env *env,
- const struct dt_it *di)
+static int osd_it_iam_rec(const struct lu_env *env,
+ const struct dt_it *di,
+ struct lu_dirent *lde,
+ __u32 attr)
{
- struct osd_it_iam *it = (struct osd_it_iam *)di;
+ struct osd_it_iam *it = (struct osd_it_iam *)di;
+ struct osd_thread_info *info = osd_oti_get(env);
+ struct lu_fid *fid = &info->oti_fid;
+ const struct osd_fid_pack *rec;
+ char *name;
+ int namelen;
+ __u64 hash;
+ int rc;
+
+ name = (char *)iam_it_key_get(&it->oi_it);
+ if (IS_ERR(name))
+ RETURN(PTR_ERR(name));
+
+ namelen = iam_it_key_size(&it->oi_it);
+
+ rec = (const struct osd_fid_pack *) iam_it_rec_get(&it->oi_it);
+ if (IS_ERR(rec))
+ RETURN(PTR_ERR(rec));
+
+ rc = osd_fid_unpack(fid, rec);
+ if (rc)
+ RETURN(rc);
+
+ hash = iam_it_store(&it->oi_it);
+
+ /* IAM does not store object type in IAM index (dir) */
+ osd_it_pack_dirent(lde, fid, hash, name, namelen,
+ 0, LUDA_FID);
- return (struct dt_rec *)iam_it_rec_get(&it->oi_it);
+ return 0;
}
/**
obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
obj_dentry->d_name.hash = 0;
- it->oie_namelen = 0;
- it->oie_curr_pos = 0;
- it->oie_next_pos = 0;
+ it->oie_rd_dirent = 0;
+ it->oie_it_dirent = 0;
+ it->oie_dirent = NULL;
+ it->oie_buf = info->oti_it_ea_buf;
it->oie_obj = obj;
+ it->oie_file.f_pos = 0;
it->oie_file.f_dentry = obj_dentry;
it->oie_file.f_mapping = obj->oo_inode->i_mapping;
it->oie_file.f_op = obj->oo_inode->i_fop;
it->oie_file.private_data = NULL;
lu_object_get(lo);
-
- RETURN((struct dt_it*) it);
+ RETURN((struct dt_it *) it);
}
/**
{
struct osd_it_ea *it = (struct osd_it_ea *)di;
struct osd_object *obj = it->oie_obj;
-
+ struct inode *inode = obj->oo_inode;
ENTRY;
+ it->oie_file.f_op->release(inode, &it->oie_file);
lu_object_put(env, &obj->oo_dt.do_lu);
EXIT;
}
ENTRY;
LASSERT(((const char *)key)[0] == '\0');
- it->oie_namelen = 0;
- it->oie_curr_pos = 0;
- it->oie_next_pos = 0;
+ it->oie_file.f_pos = 0;
+ it->oie_rd_dirent = 0;
+ it->oie_it_dirent = 0;
+ it->oie_dirent = NULL;
RETURN(+1);
}
* \retval 1, on buffer full
*/
static int osd_ldiskfs_filldir(char *buf, const char *name, int namelen,
- loff_t offset, ino_t ino,
- unsigned int d_type)
+ loff_t offset, __u64 ino,
+ unsigned d_type)
{
- struct osd_it_ea *it = (struct osd_it_ea *)buf;
- struct dirent64 *dirent = &it->oie_dirent64;
- int reclen = LDISKFS_DIR_REC_LEN(namelen);
-
-
+ struct osd_it_ea *it = (struct osd_it_ea *)buf;
+ struct osd_it_ea_dirent *ent = it->oie_dirent;
ENTRY;
- if (it->oie_namelen)
- RETURN(-ENOENT);
- if (namelen == 0 || namelen > LDISKFS_NAME_LEN)
+ /* this should never happen */
+ if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
+ CERROR("ldiskfs return invalid namelen %d\n", namelen);
RETURN(-EIO);
+ }
+
+ if ((void *) ent - it->oie_buf + sizeof(*ent) + namelen >
+ OSD_IT_EA_BUFSIZE)
+ RETURN(1);
- strncpy(dirent->d_name, name, LDISKFS_NAME_LEN);
- dirent->d_name[namelen] = 0;
- dirent->d_ino = ino;
- dirent->d_off = offset;
- dirent->d_reclen = reclen;
- it->oie_namelen = namelen;
- it->oie_curr_pos = offset;
+ ent->oied_ino = ino;
+ ent->oied_off = offset;
+ ent->oied_namelen = namelen;
+ ent->oied_type = d_type;
+ memcpy(ent->oied_name, name, namelen);
+
+ it->oie_rd_dirent++;
+ it->oie_dirent = (void *) ent + size_round(sizeof(*ent) + namelen);
RETURN(0);
}
* \retval 0, on success
* \retval -ve, on error
*/
-int osd_ldiskfs_it_fill(const struct dt_it *di)
+static int osd_ldiskfs_it_fill(const struct dt_it *di)
{
struct osd_it_ea *it = (struct osd_it_ea *)di;
struct osd_object *obj = it->oie_obj;
int result = 0;
ENTRY;
- it->oie_namelen = 0;
- it->oie_file.f_pos = it->oie_curr_pos;
+ it->oie_dirent = it->oie_buf;
+ it->oie_rd_dirent = 0;
+ down_read(&obj->oo_ext_idx_sem);
result = inode->i_fop->readdir(&it->oie_file, it,
(filldir_t) osd_ldiskfs_filldir);
- it->oie_next_pos = it->oie_file.f_pos;
+ up_read(&obj->oo_ext_idx_sem);
- if(!result && it->oie_namelen == 0)
+ if (it->oie_rd_dirent == 0) {
result = -EIO;
+ } else {
+ it->oie_dirent = it->oie_buf;
+ it->oie_it_dirent = 1;
+ }
RETURN(result);
}
int rc;
ENTRY;
- it->oie_curr_pos = it->oie_next_pos;
- if (it->oie_curr_pos == LDISKFS_HTREE_EOF)
- rc = +1;
- else
- rc = osd_ldiskfs_it_fill(di);
+ if (it->oie_it_dirent < it->oie_rd_dirent) {
+ it->oie_dirent = (void *) it->oie_dirent +
+ size_round(sizeof(struct osd_it_ea_dirent) +
+ it->oie_dirent->oied_namelen);
+ it->oie_it_dirent++;
+ RETURN(0);
+ } else {
+ if (it->oie_file.f_pos == LDISKFS_HTREE_EOF)
+ rc = +1;
+ else
+ rc = osd_ldiskfs_it_fill(di);
+ }
RETURN(rc);
}
{
struct osd_it_ea *it = (struct osd_it_ea *)di;
ENTRY;
- RETURN((struct dt_key *)it->oie_dirent64.d_name);
+ RETURN((struct dt_key *)it->oie_dirent->oied_name);
}
/**
{
struct osd_it_ea *it = (struct osd_it_ea *)di;
ENTRY;
- RETURN(it->oie_namelen);
+ RETURN(it->oie_dirent->oied_namelen);
}
+
/**
* Returns the value (i.e. fid/igif) at current position from iterator's
* in memory structure.
*
* \param di, struct osd_it_ea, iterator's in memory structure
+ * \param attr, attr requested for dirent.
+ * \param lde, lustre dirent
*
- * \retval value i.e. struct dt_rec on success
+ * \retval 0, no error and \param lde has correct lustre dirent.
+ * \retval -ve, on error
*/
-static struct dt_rec *osd_it_ea_rec(const struct lu_env *env,
- const struct dt_it *di)
+static inline int osd_it_ea_rec(const struct lu_env *env,
+ const struct dt_it *di,
+ struct lu_dirent *lde,
+ __u32 attr)
{
- struct osd_it_ea *it = (struct osd_it_ea *)di;
- struct osd_object *obj = it->oie_obj;
- struct osd_thread_info *info = osd_oti_get(env);
- struct osd_inode_id *id = &info->oti_id;
- struct lu_fid_pack *rec = &info->oti_pack;
- struct lu_device *ldev = obj->oo_dt.do_lu.lo_dev;
- struct dentry *dentry = &info->oti_child_dentry;
- struct osd_device *dev;
- struct inode *inode;
- int rc;
+ struct osd_it_ea *it = (struct osd_it_ea *)di;
+ struct osd_object *obj = it->oie_obj;
+ struct osd_thread_info *info = osd_oti_get(env);
+ struct lu_fid *fid = &info->oti_fid;
+ int rc;
ENTRY;
- dev = osd_dev(ldev);
- id->oii_ino = it->oie_dirent64.d_ino;
- id->oii_gen = OSD_OII_NOGEN;
- inode = osd_iget(info, dev, id);
- if (!IS_ERR(inode)) {
- dentry->d_inode = inode;
- LASSERT(dentry->d_inode->i_sb == osd_sb(dev));
- } else {
- CERROR("Error getting inode for ino =%d", id->oii_ino);
- RETURN((struct dt_rec *) PTR_ERR(inode));
- }
- rc = osd_ea_fid_get(env, dentry, (struct dt_rec*) rec);
-
- iput(inode);
- RETURN((struct dt_rec *)rec);
+ rc = osd_ea_fid_get(env, obj, it->oie_dirent->oied_ino, fid);
+ if (rc == 0)
+ osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
+ it->oie_dirent->oied_name,
+ it->oie_dirent->oied_namelen,
+ it->oie_dirent->oied_type,
+ attr);
+ RETURN(rc);
}
/**
{
struct osd_it_ea *it = (struct osd_it_ea *)di;
ENTRY;
- RETURN(it->oie_curr_pos);
+ RETURN(it->oie_dirent->oied_off);
}
/**
int rc;
ENTRY;
- it->oie_curr_pos = it->oie_next_pos = hash;
+ it->oie_file.f_pos = hash;
rc = osd_ldiskfs_it_fill(di);
if (rc == 0)
RETURN(rc);
}
-/**
- * Index and Iterator operations for interoperability
- * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
- */
-static const struct dt_index_operations osd_index_ea_ops = {
- .dio_lookup = osd_index_ea_lookup,
- .dio_insert = osd_index_ea_insert,
- .dio_delete = osd_index_ea_delete,
- .dio_it = {
- .init = osd_it_ea_init,
- .fini = osd_it_ea_fini,
- .get = osd_it_ea_get,
- .put = osd_it_ea_put,
- .next = osd_it_ea_next,
- .key = osd_it_ea_key,
- .key_size = osd_it_ea_key_size,
- .rec = osd_it_ea_rec,
- .store = osd_it_ea_store,
- .load = osd_it_ea_load
- }
-};
/**
* Index lookup function for interoperability mode (b11826).
RETURN(rc);
}
-/* type constructor/destructor: osd_type_init, osd_type_fini */
-LU_TYPE_INIT_FINI(osd, &osd_key);
-
-static struct lu_context_key osd_key = {
- .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD,
- .lct_init = osd_key_init,
- .lct_fini = osd_key_fini,
- .lct_exit = osd_key_exit
+/**
+ * Index and Iterator operations for interoperability
+ * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
+ */
+static const struct dt_index_operations osd_index_ea_ops = {
+ .dio_lookup = osd_index_ea_lookup,
+ .dio_insert = osd_index_ea_insert,
+ .dio_delete = osd_index_ea_delete,
+ .dio_it = {
+ .init = osd_it_ea_init,
+ .fini = osd_it_ea_fini,
+ .get = osd_it_ea_get,
+ .put = osd_it_ea_put,
+ .next = osd_it_ea_next,
+ .key = osd_it_ea_key,
+ .key_size = osd_it_ea_key_size,
+ .rec = osd_it_ea_rec,
+ .store = osd_it_ea_store,
+ .load = osd_it_ea_load
+ }
};
static void *osd_key_init(const struct lu_context *ctx,
struct osd_thread_info *info;
OBD_ALLOC_PTR(info);
- if (info != NULL)
- info->oti_env = container_of(ctx, struct lu_env, le_ctx);
- else
+ if (info != NULL) {
+ OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
+ if (info->oti_it_ea_buf != NULL) {
+ info->oti_env = container_of(ctx, struct lu_env,
+ le_ctx);
+ } else {
+ OBD_FREE_PTR(info);
+ info = ERR_PTR(-ENOMEM);
+ }
+ } else {
info = ERR_PTR(-ENOMEM);
+ }
return info;
}
-/* context key destructor: osd_key_fini */
-LU_KEY_FINI(osd, struct osd_thread_info);
+static void osd_key_fini(const struct lu_context *ctx,
+ struct lu_context_key *key, void* data)
+{
+ struct osd_thread_info *info = data;
+
+ OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
+ OBD_FREE_PTR(info);
+}
static void osd_key_exit(const struct lu_context *ctx,
struct lu_context_key *key, void *data)
LASSERT(info->oti_txns == 0);
}
+/* type constructor/destructor: osd_type_init, osd_type_fini */
+LU_TYPE_INIT_FINI(osd, &osd_key);
+
+static struct lu_context_key osd_key = {
+ .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD,
+ .lct_init = osd_key_init,
+ .lct_fini = osd_key_fini,
+ .lct_exit = osd_key_exit
+};
+
+
static int osd_device_init(const struct lu_env *env, struct lu_device *d,
const char *name, struct lu_device *next)
{
struct lustre_sb_info *lsi;
ENTRY;
-
if (o->od_mount != NULL) {
CERROR("Already mounted (%s)\n", dev);
RETURN(-EEXIST);
RETURN(err);
}
-extern void ldiskfs_orphan_cleanup (struct super_block * sb,
- struct ldiskfs_super_block * es);
-
static int osd_recovery_complete(const struct lu_env *env,
struct lu_device *d)
{
- struct osd_device *o = osd_dev(d);
- ENTRY;
- /* TODO: orphans handling */
- ldiskfs_orphan_cleanup(osd_sb(o), LDISKFS_SB(osd_sb(o))->s_es);
RETURN(0);
}
RETURN(result);
}
-static struct inode *osd_iget(struct osd_thread_info *info,
- struct osd_device *dev,
- const struct osd_inode_id *id)
-{
- struct inode *inode;
-
- inode = iget(osd_sb(dev), id->oii_ino);
- if (inode == NULL) {
- CERROR("no inode\n");
- inode = ERR_PTR(-EACCES);
- } else if (is_bad_inode(inode)) {
- CERROR("bad inode\n");
- iput(inode);
- inode = ERR_PTR(-ENOENT);
- } else if (id->oii_gen != OSD_OII_NOGEN &&
- inode->i_generation != id->oii_gen) {
- CERROR("stale inode\n");
- iput(inode);
- inode = ERR_PTR(-ESTALE);
- }
-
- return inode;
-
-}
-
-static int osd_fid_lookup(const struct lu_env *env,
- struct osd_object *obj, const struct lu_fid *fid)
-{
- struct osd_thread_info *info;
- struct lu_device *ldev = obj->oo_dt.do_lu.lo_dev;
- struct osd_device *dev;
- struct osd_inode_id *id;
- struct osd_oi *oi;
- struct inode *inode;
- int result;
-
- LINVRNT(osd_invariant(obj));
- LASSERT(obj->oo_inode == NULL);
- LASSERT(fid_is_sane(fid));
- /*
- * This assertion checks that osd layer sees only local
- * fids. Unfortunately it is somewhat expensive (does a
- * cache-lookup). Disabling it for production/acceptance-testing.
- */
- LASSERT(1 || fid_is_local(env, ldev->ld_site, fid));
-
- ENTRY;
-
- info = osd_oti_get(env);
- dev = osd_dev(ldev);
- id = &info->oti_id;
- oi = &dev->od_oi;
-
- if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT))
- RETURN(-ENOENT);
-
- result = osd_oi_lookup(info, oi, fid, id);
- if (result == 0) {
- inode = osd_iget(info, dev, id);
- if (!IS_ERR(inode)) {
- obj->oo_inode = inode;
- LASSERT(obj->oo_inode->i_sb == osd_sb(dev));
- if (dev->od_iop_mode) {
- obj->oo_compat_dot_created = 1;
- obj->oo_compat_dotdot_created = 1;
- }
- result = 0;
- } else
- /*
- * If fid wasn't found in oi, inode-less object is
- * created, for which lu_object_exists() returns
- * false. This is used in a (frequent) case when
- * objects are created as locking anchors or
- * place holders for objects yet to be created.
- */
- result = PTR_ERR(inode);
- } else if (result == -ENOENT)
- result = 0;
- LINVRNT(osd_invariant(obj));
-
- RETURN(result);
-}
-
-static void osd_inode_getattr(const struct lu_env *env,
- struct inode *inode, struct lu_attr *attr)
-{
- attr->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
- LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
- LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE;
-
- attr->la_atime = LTIME_S(inode->i_atime);
- attr->la_mtime = LTIME_S(inode->i_mtime);
- attr->la_ctime = LTIME_S(inode->i_ctime);
- attr->la_mode = inode->i_mode;
- attr->la_size = i_size_read(inode);
- attr->la_blocks = inode->i_blocks;
- attr->la_uid = inode->i_uid;
- attr->la_gid = inode->i_gid;
- attr->la_flags = LDISKFS_I(inode)->i_flags;
- attr->la_nlink = inode->i_nlink;
- attr->la_rdev = inode->i_rdev;
- attr->la_blksize = ll_inode_blksize(inode);
- attr->la_blkbits = inode->i_blkbits;
-}
-
-/*
- * Helpers.
- */
-
-static int lu_device_is_osd(const struct lu_device *d)
-{
- return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &osd_lu_ops);
-}
-
-static struct osd_object *osd_obj(const struct lu_object *o)
-{
- LASSERT(lu_device_is_osd(o->lo_dev));
- return container_of0(o, struct osd_object, oo_dt.do_lu);
-}
-
-static struct osd_device *osd_dt_dev(const struct dt_device *d)
-{
- LASSERT(lu_device_is_osd(&d->dd_lu_dev));
- return container_of0(d, struct osd_device, od_dt_dev);
-}
-
-static struct osd_device *osd_dev(const struct lu_device *d)
-{
- LASSERT(lu_device_is_osd(d));
- return osd_dt_dev(container_of0(d, struct dt_device, dd_lu_dev));
-}
-
-static struct osd_object *osd_dt_obj(const struct dt_object *d)
-{
- return osd_obj(&d->do_lu);
-}
-
-static struct osd_device *osd_obj2dev(const struct osd_object *o)
-{
- return osd_dev(o->oo_dt.do_lu.lo_dev);
-}
-
-static struct lu_device *osd2lu_dev(struct osd_device *osd)
-{
- return &osd->od_dt_dev.dd_lu_dev;
-}
-
-static struct super_block *osd_sb(const struct osd_device *dev)
-{
- return dev->od_mount->lmi_mnt->mnt_sb;
-}
-
-static journal_t *osd_journal(const struct osd_device *dev)
-{
- return LDISKFS_SB(osd_sb(dev))->s_journal;
-}
-
-static int osd_has_index(const struct osd_object *obj)
-{
- return obj->oo_dt.do_index_ops != NULL;
-}
-
-static int osd_object_invariant(const struct lu_object *l)
-{
- return osd_invariant(osd_obj(l));
-}
-
static const struct lu_object_operations osd_lu_obj_ops = {
.loo_object_init = osd_object_init,
.loo_object_delete = osd_object_delete,
git://git.whamcloud.com / fs / lustre-release.git / blobdiff