*
* You should have received a copy of the GNU General Public License
* version 2 along with this program; If not, see
- * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
- *
- * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
- * CA 95054 USA or visit www.sun.com if you need additional information or
- * have any questions.
+ * http://www.gnu.org/licenses/gpl-2.0.html
*
* GPL HEADER END
*/
/*
* Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
- */
-/*
- * Copyright (c) 2012, 2013, Intel Corporation.
- * Use is subject to license terms.
+ *
+ * Copyright (c) 2012, 2016, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
#include <dt_object.h>
#include <md_object.h>
#include <lustre_quota.h>
-
-#define _SPL_KMEM_H
-#include <sys/kstat.h>
-#define kmem_zalloc(a, b) kzalloc(a, b)
-#define kmem_free(ptr, sz) ((void)(sz), kfree(ptr))
-#ifndef KM_SLEEP
-#define KM_SLEEP GFP_KERNEL
+#ifdef SHRINK_STOP
+#undef SHRINK_STOP
#endif
-
#include <sys/arc.h>
#include <sys/nvpair.h>
#include <sys/zfs_znode.h>
#include <sys/zap.h>
+#include <sys/dbuf.h>
+#include <sys/dmu_objset.h>
+
+/**
+ * By design including kmem.h overrides the Linux slab interfaces to provide
+ * the Illumos kmem cache interfaces. To override this and gain access to
+ * the Linux interfaces these preprocessor macros must be undefined.
+ */
+#ifdef kmem_cache_destroy
+#undef kmem_cache_destroy
+#endif
+
+#ifdef kmem_cache_create
+#undef kmem_cache_create
+#endif
+
+#ifdef kmem_cache_alloc
+#undef kmem_cache_alloc
+#endif
+
+#ifdef kmem_cache_free
+#undef kmem_cache_free
+#endif
#define LUSTRE_ROOT_FID_SEQ 0
#define DMU_OSD_SVNAME "svname"
#define OSD_GFP_IO (GFP_NOFS | __GFP_HIGHMEM)
/* Statfs space reservation for grant, fragmentation, and unlink space. */
-#define OSD_STATFS_RESERVED_BLKS (1ULL << (22 - SPA_MAXBLOCKSHIFT)) /* 4MB */
-#define OSD_STATFS_RESERVED_SHIFT (7) /* reserve 0.78% of all space */
+#define OSD_STATFS_RESERVED_SIZE (16ULL << 20) /* reserve 16MB minimum */
+#define OSD_STATFS_RESERVED_SHIFT (7) /* reserve 0.78% of all space */
/* Statfs {minimum, safe estimate, and maximum} dnodes per block */
-#define OSD_DNODE_MIN_BLKSHIFT (SPA_MAXBLOCKSHIFT - DNODE_SHIFT) /* 17-9 =8 */
-#define OSD_DNODE_EST_BLKSHIFT (SPA_MAXBLOCKSHIFT - 12) /* 17-12=5 */
-#define OSD_DNODE_EST_COUNT 1024
+#define OSD_DNODE_MIN_BLKSHIFT (DNODES_PER_BLOCK_SHIFT)
+#define OSD_DNODE_EST_BLKSHIFT (12) /* est 4KB/dnode */
+#define OSD_DNODE_EST_COUNT 4096
#define OSD_GRANT_FOR_LOCAL_OIDS (2ULL << 20) /* 2MB for last_rcvd, ... */
struct osd_zap_it {
zap_cursor_t *ozi_zc;
struct osd_object *ozi_obj;
- struct lustre_capa *ozi_capa;
unsigned ozi_reset:1; /* 1 -- no need to advance */
/* ozi_pos - position of the cursor:
* 0 - before any record
uint64_t ctime[2];
};
+
+#define OSD_INS_CACHE_SIZE 8
+
+/* OI cache entry */
+struct osd_idmap_cache {
+ struct osd_device *oic_dev;
+ struct lu_fid oic_fid;
+ /** max 2^48 dnodes per dataset, avoid spilling into another word */
+ uint64_t oic_dnode:DN_MAX_OBJECT_SHIFT,
+ oic_remote:1; /* FID isn't local */
+};
+
+/* max.number of regular attrubites the callers may ask for */
+#define OSD_MAX_IN_BULK 13
+
struct osd_thread_info {
const struct lu_env *oti_env;
* XXX temporary: for ->i_op calls.
*/
struct timespec oti_time;
- /*
- * XXX temporary: for capa operations.
- */
- struct lustre_capa_key oti_capa_key;
- struct lustre_capa oti_capa;
struct ost_id oti_ostid;
char oti_buf[64];
- /** osd iterator context used for iterator session */
- union {
- struct osd_zap_it oti_it_zap;
- struct osd_it_quota oti_it_quota;
- };
-
char oti_str[64];
union {
char oti_key[MAXNAMELEN + 1];
__u64 oti_key64[(MAXNAMELEN + 1)/sizeof(__u64)];
+ sa_bulk_attr_t oti_attr_bulk[OSD_MAX_IN_BULK];
};
struct lustre_mdt_attrs oti_mdt_attrs;
struct lquota_id_info oti_qi;
struct lu_seq_range oti_seq_range;
- unsigned int oti_it_inline:1;
+
+ /* dedicated OI cache for insert (which needs inum) */
+ struct osd_idmap_cache *oti_ins_cache;
+ int oti_ins_cache_size;
+ int oti_ins_cache_used;
};
extern struct lu_context_key osd_key;
struct osd_thandle {
struct thandle ot_super;
struct list_head ot_dcb_list;
+ struct list_head ot_stop_dcb_list;
+ struct list_head ot_unlinked_list;
struct list_head ot_sa_list;
struct semaphore ot_sa_lock;
dmu_tx_t *ot_tx;
struct osd_oi {
char oi_name[OSD_OI_NAME_SIZE]; /* unused */
uint64_t oi_zapid;
+ dnode_t *oi_dn;
};
struct osd_seq {
/* information about underlying file system */
struct objset *od_os;
uint64_t od_rootid; /* id of root znode */
+ uint64_t od_unlinkedid; /* id of unlinked zapobj */
/* SA attr mapping->id,
* name is the same as in ZFS to use defines SA_ZPL_...*/
sa_attr_type_t *z_attr_table;
- /*
- * Fid Capability
- */
- unsigned int od_fl_capa:1;
- unsigned long od_capa_timeout;
- __u32 od_capa_alg;
- struct lustre_capa_key *od_capa_keys;
- struct hlist_head *od_capa_hash;
-
struct proc_dir_entry *od_proc_entry;
struct lprocfs_stats *od_stats;
+ uint64_t od_max_blksz;
uint64_t od_root;
uint64_t od_O_id;
struct osd_oi **od_oi_table;
unsigned int od_oi_count;
struct osd_seq_list od_seq_list;
- unsigned int od_rdonly:1,
+ unsigned int od_dev_set_rdonly:1, /**< osd_ro() called */
+ od_prop_rdonly:1, /**< ZFS property readonly */
od_xattr_in_sa:1,
- od_quota_iused_est:1,
od_is_ost:1,
od_posix_acl:1;
/* quota slave instance */
struct qsd_instance *od_quota_slave;
+ struct brw_stats od_brw_stats;
+ atomic_t od_r_in_flight;
+ atomic_t od_w_in_flight;
+
/* used to debug zerocopy logic: the fields track all
* allocated, loaned and referenced buffers in use.
* to be removed once the change is tested well. */
struct lu_client_seq *od_cl_seq;
};
+enum osd_destroy_type {
+ OSD_DESTROY_NONE = 0,
+ OSD_DESTROY_SYNC = 1,
+ OSD_DESTROY_ASYNC = 2,
+};
+
struct osd_object {
struct dt_object oo_dt;
/*
* Not modified concurrently (either setup early during object
* creation, or assigned by osd_object_create() under write lock).
*/
- dmu_buf_t *oo_db;
+ dnode_t *oo_dn;
sa_handle_t *oo_sa_hdl;
nvlist_t *oo_sa_xattr;
struct list_head oo_sa_linkage;
+ /* used to implement osd_object_*_{lock|unlock} */
struct rw_semaphore oo_sem;
+ /* to serialize some updates: destroy vs. others,
+ * xattr_set, object block size change etc */
+ struct rw_semaphore oo_guard;
+
+ /* protected by oo_guard */
+ struct list_head oo_unlinked_linkage;
+
/* cached attributes */
rwlock_t oo_attr_lock;
struct lu_attr oo_attr;
- /* protects extended attributes */
- struct semaphore oo_guard;
+ /* external dnode holding large EAs, protected by oo_guard */
uint64_t oo_xattr;
+ enum osd_destroy_type oo_destroy;
- /* record size for index file */
- unsigned char oo_keysize;
- unsigned char oo_recsize;
- unsigned char oo_recusize; /* unit size */
+ __u32 oo_destroyed:1;
+
+ /* the i_flags in LMA */
+ __u32 oo_lma_flags;
+ union {
+ int oo_ea_in_bonus; /* EA bytes we expect */
+ struct {
+ /* record size for index file */
+ unsigned char oo_keysize;
+ unsigned char oo_recsize;
+ unsigned char oo_recusize; /* unit size */
+ };
+ };
};
int osd_statfs(const struct lu_env *, struct dt_device *, struct obd_statfs *);
extern const struct dt_index_operations osd_acct_index_ops;
-uint64_t osd_quota_fid2dmu(const struct lu_fid *fid);
+int osd_quota_fid2dmu(const struct lu_fid *fid, uint64_t *oid);
extern struct lu_device_operations osd_lu_ops;
extern struct dt_index_operations osd_dir_ops;
int osd_declare_quota(const struct lu_env *env, struct osd_device *osd,
struct osd_thandle *oh, bool is_blk, int *flags,
bool force);
uint64_t osd_objs_count_estimate(uint64_t refdbytes, uint64_t usedobjs,
- uint64_t nrblocks);
+ uint64_t nrblocks, uint64_t est_maxblockshift);
/*
* Helpers.
return 1;
}
+/**
+ * Put the osd object once done with it.
+ *
+ * \param obj osd object that needs to be put
+ */
+static inline void osd_object_put(const struct lu_env *env,
+ struct osd_object *obj)
+{
+ dt_object_put(env, &obj->oo_dt);
+}
+
static inline int osd_object_invariant(const struct lu_object *l)
{
return osd_invariant(osd_obj(l));
return osd->od_dt_dev.dd_lu_dev.ld_obd->obd_name;
}
-#ifdef LPROCFS
+#ifdef CONFIG_PROC_FS
enum {
LPROC_OSD_READ_BYTES = 0,
LPROC_OSD_WRITE_BYTES = 1,
LPROC_OSD_LAST,
};
+extern struct kmem_cache *osd_zapit_cachep;
/* osd_lproc.c */
-extern struct lprocfs_seq_vars lprocfs_osd_obd_vars[];
+extern struct lprocfs_vars lprocfs_osd_obd_vars[];
int osd_procfs_init(struct osd_device *osd, const char *name);
int osd_procfs_fini(struct osd_device *osd);
/* osd_object.c */
extern char *osd_obj_tag;
void osd_object_sa_dirty_rele(struct osd_thandle *oh);
-int __osd_obj2dbuf(const struct lu_env *env, objset_t *os,
- uint64_t oid, dmu_buf_t **dbp);
+int __osd_obj2dnode(const struct lu_env *env, objset_t *os,
+ uint64_t oid, dnode_t **dnp);
struct lu_object *osd_object_alloc(const struct lu_env *env,
const struct lu_object_header *hdr,
struct lu_device *d);
int osd_object_sa_update(struct osd_object *obj, sa_attr_type_t type,
void *buf, uint32_t buflen, struct osd_thandle *oh);
int __osd_zap_create(const struct lu_env *env, struct osd_device *osd,
- dmu_buf_t **zap_dbp, dmu_tx_t *tx, struct lu_attr *la,
- uint64_t parent, zap_flags_t flags);
-int __osd_object_create(const struct lu_env *env, struct osd_device *osd,
- dmu_buf_t **dbp, dmu_tx_t *tx, struct lu_attr *la,
- uint64_t parent);
+ dnode_t **zap_dnp, dmu_tx_t *tx, struct lu_attr *la,
+ zap_flags_t flags);
+int __osd_object_create(const struct lu_env *env, struct osd_object *obj,
+ dnode_t **dnp, dmu_tx_t *tx, struct lu_attr *la);
+int __osd_attr_init(const struct lu_env *env, struct osd_device *osd,
+ sa_handle_t *sa_hdl, dmu_tx_t *tx,
+ struct lu_attr *la, uint64_t parent);
/* osd_oi.c */
int osd_oi_init(const struct lu_env *env, struct osd_device *o);
int osd_fid_lookup(const struct lu_env *env,
struct osd_device *, const struct lu_fid *, uint64_t *);
uint64_t osd_get_name_n_idx(const struct lu_env *env, struct osd_device *osd,
- const struct lu_fid *fid, char *buf);
+ const struct lu_fid *fid, char *buf, int bufsize);
int osd_options_init(void);
-int osd_convert_root_to_new_seq(const struct lu_env *env,
- struct osd_device *o);
int osd_ost_seq_exists(const struct lu_env *env, struct osd_device *osd,
__u64 seq);
+int osd_idc_find_and_init(const struct lu_env *env, struct osd_device *osd,
+ struct osd_object *obj);
+struct osd_idmap_cache *osd_idc_find_or_init(const struct lu_env *env,
+ struct osd_device *osd,
+ const struct lu_fid *fid);
+struct osd_idmap_cache *osd_idc_find(const struct lu_env *env,
+ struct osd_device *osd,
+ const struct lu_fid *fid);
+
/* osd_index.c */
int osd_index_try(const struct lu_env *env, struct dt_object *dt,
const struct dt_index_features *feat);
uint64_t id, uint64_t dirhash);
void osd_zap_cursor_fini(zap_cursor_t *zc);
uint64_t osd_zap_cursor_serialize(zap_cursor_t *zc);
+int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
+ const struct lu_fid *fid);
/* osd_xattr.c */
int __osd_xattr_load(struct osd_device *osd, uint64_t dnode,
uint64_t xattr, struct lu_buf *buf,
const char *name, int *sizep);
int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
- struct lu_buf *buf, const char *name,
- struct lustre_capa *capa);
+ struct lu_buf *buf, const char *name);
int osd_declare_xattr_set(const struct lu_env *env, struct dt_object *dt,
const struct lu_buf *buf, const char *name,
int fl, struct thandle *handle);
int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
const struct lu_buf *buf, const char *name, int fl,
- struct thandle *handle, struct lustre_capa *capa);
+ struct thandle *handle);
int osd_declare_xattr_del(const struct lu_env *env, struct dt_object *dt,
const char *name, struct thandle *handle);
int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
- const char *name, struct thandle *handle,
- struct lustre_capa *capa);
+ const char *name, struct thandle *handle);
+void osd_declare_xattrs_destroy(const struct lu_env *env,
+ struct osd_object *obj,
+ struct osd_thandle *oh);
+int osd_xattrs_destroy(const struct lu_env *env,
+ struct osd_object *obj, struct osd_thandle *oh);
int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
- const struct lu_buf *lb, struct lustre_capa *capa);
+ const struct lu_buf *lb);
void __osd_xattr_declare_set(const struct lu_env *env, struct osd_object *obj,
int vallen, const char *name, struct osd_thandle *oh);
int __osd_sa_xattr_set(const struct lu_env *env, struct osd_object *obj,
int __osd_xattr_set(const struct lu_env *env, struct osd_object *obj,
const struct lu_buf *buf, const char *name, int fl,
struct osd_thandle *oh);
+int __osd_sa_xattr_update(const struct lu_env *env, struct osd_object *obj,
+ struct osd_thandle *oh);
static inline int
osd_xattr_set_internal(const struct lu_env *env, struct osd_object *obj,
const struct lu_buf *buf, const char *name, int fl,
- struct osd_thandle *oh, struct lustre_capa *capa)
+ struct osd_thandle *oh)
{
int rc;
+ if (unlikely(!dt_object_exists(&obj->oo_dt) || obj->oo_destroyed))
+ return -ENOENT;
+
+ LASSERT(obj->oo_dn);
if (osd_obj2dev(obj)->od_xattr_in_sa) {
rc = __osd_sa_xattr_set(env, obj, buf, name, fl, oh);
if (rc == -EFBIG)
static inline uint64_t attrs_fs2zfs(const uint32_t flags)
{
- return (((flags & FS_APPEND_FL) ? ZFS_APPENDONLY : 0) |
- ((flags & FS_NODUMP_FL) ? ZFS_NODUMP : 0) |
- ((flags & FS_IMMUTABLE_FL) ? ZFS_IMMUTABLE : 0));
+ return (flags & LUSTRE_APPEND_FL ? ZFS_APPENDONLY : 0) |
+ (flags & LUSTRE_NODUMP_FL ? ZFS_NODUMP : 0) |
+ (flags & LUSTRE_IMMUTABLE_FL ? ZFS_IMMUTABLE : 0);
}
static inline uint32_t attrs_zfs2fs(const uint64_t flags)
{
- return (((flags & ZFS_APPENDONLY) ? FS_APPEND_FL : 0) |
- ((flags & ZFS_NODUMP) ? FS_NODUMP_FL : 0) |
- ((flags & ZFS_IMMUTABLE) ? FS_IMMUTABLE_FL : 0));
+ return (flags & ZFS_APPENDONLY ? LUSTRE_APPEND_FL : 0) |
+ (flags & ZFS_NODUMP ? LUSTRE_NODUMP_FL : 0) |
+ (flags & ZFS_IMMUTABLE ? LUSTRE_IMMUTABLE_FL : 0);
}
#endif
static inline void dsl_pool_config_exit(dsl_pool_t *dp, char *name)
{
}
+#endif
+#ifdef HAVE_SPA_MAXBLOCKSIZE
+#define osd_spa_maxblocksize(spa) spa_maxblocksize(spa)
+#define osd_spa_maxblockshift(spa) fls64(spa_maxblocksize(spa) - 1)
+#else
+#define osd_spa_maxblocksize(spa) SPA_MAXBLOCKSIZE
+#define osd_spa_maxblockshift(spa) SPA_MAXBLOCKSHIFT
+#define SPA_OLD_MAXBLOCKSIZE SPA_MAXBLOCKSIZE
#endif
+#ifdef HAVE_SA_SPILL_ALLOC
+static inline void *
+osd_zio_buf_alloc(size_t size)
+{
+ return sa_spill_alloc(KM_SLEEP);
+}
+
+static inline void
+osd_zio_buf_free(void *buf, size_t size)
+{
+ sa_spill_free(buf);
+}
+#else
+#define osd_zio_buf_alloc(size) zio_buf_alloc(size)
+#define osd_zio_buf_free(buf, size) zio_buf_free(buf, size)
+#endif
+
+#ifdef HAVE_DMU_OBJECT_ALLOC_DNSIZE
+static inline uint64_t
+osd_dmu_object_alloc(objset_t *os, dmu_object_type_t objtype, int blocksize,
+ int dnodesize, dmu_tx_t *tx)
+{
+ if (dnodesize == 0)
+ dnodesize = MAX(dmu_objset_dnodesize(os), DNODE_MIN_SIZE);
+
+ return dmu_object_alloc_dnsize(os, objtype, blocksize, DMU_OT_SA,
+ DN_BONUS_SIZE(dnodesize), dnodesize, tx);
+}
+
+static inline uint64_t
+osd_zap_create_flags(objset_t *os, int normflags, zap_flags_t flags,
+ dmu_object_type_t ot, int leaf_blockshift,
+ int indirect_blockshift, int dnodesize, dmu_tx_t *tx)
+{
+ if (dnodesize == 0)
+ dnodesize = MAX(dmu_objset_dnodesize(os), DNODE_MIN_SIZE);
+
+ return zap_create_flags_dnsize(os, normflags, flags, ot,
+ leaf_blockshift, indirect_blockshift,
+ DMU_OT_SA, DN_BONUS_SIZE(dnodesize),
+ dnodesize, tx);
+}
+
+static inline int
+osd_obj_bonuslen(struct osd_object *obj)
+{
+ int bonuslen = DN_BONUS_SIZE(DNODE_MIN_SIZE);
+
+ if (obj->oo_dn != NULL && obj->oo_dn->dn_num_slots != 0) {
+ bonuslen = DN_SLOTS_TO_BONUSLEN(obj->oo_dn->dn_num_slots);
+ } else {
+ objset_t *os = osd_dtobj2objset(&obj->oo_dt);
+ int dnodesize;
+
+ if (os != NULL) {
+ dnodesize = dmu_objset_dnodesize(os);
+ if (dnodesize != 0)
+ bonuslen = DN_BONUS_SIZE(dnodesize);
+ }
+ }
+
+ return bonuslen;
+}
+#else
+static inline uint64_t
+osd_dmu_object_alloc(objset_t *os, dmu_object_type_t objtype, int blocksize,
+ int dnodesize, dmu_tx_t *tx)
+{
+ return dmu_object_alloc(os, objtype, blocksize, DMU_OT_SA,
+ DN_MAX_BONUSLEN, tx);
+}
+
+static inline uint64_t
+osd_zap_create_flags(objset_t *os, int normflags, zap_flags_t flags,
+ dmu_object_type_t ot, int leaf_blockshift,
+ int indirect_blockshift, int dnodesize, dmu_tx_t *tx)
+{
+ return zap_create_flags(os, normflags, flags, ot, leaf_blockshift,
+ indirect_blockshift, DMU_OT_SA,
+ DN_MAX_BONUSLEN, tx);
+}
+
+static inline int
+osd_obj_bonuslen(struct osd_object *obj)
+{
+ return DN_MAX_BONUSLEN;
+}
+#endif /* HAVE_DMU_OBJECT_ALLOC_DNSIZE */
+
+#ifdef HAVE_DMU_PREFETCH_6ARG
+#define osd_dmu_prefetch(os, obj, lvl, off, len, pri) \
+ dmu_prefetch((os), (obj), (lvl), (off), (len), (pri))
+#else
+#define osd_dmu_prefetch(os, obj, lvl, off, len, pri) \
+ dmu_prefetch((os), (obj), (lvl), (off))
+#endif
+
+static inline void osd_dnode_rele(dnode_t *dn)
+{
+ dmu_buf_impl_t *db;
+ LASSERT(dn);
+ LASSERT(dn->dn_bonus);
+ db = dn->dn_bonus;
+
+ DB_DNODE_EXIT(db);
+ dmu_buf_rele(&db->db, osd_obj_tag);
+}
+
+#ifdef HAVE_DMU_USEROBJ_ACCOUNTING
+
+#define OSD_DMU_USEROBJ_PREFIX DMU_OBJACCT_PREFIX
+
+static inline bool osd_dmu_userobj_accounting_available(struct osd_device *osd)
+{
+ if (unlikely(dmu_objset_userobjspace_upgradable(osd->od_os)))
+ dmu_objset_userobjspace_upgrade(osd->od_os);
+
+ return dmu_objset_userobjspace_present(osd->od_os);
+}
+#else
+
+#define OSD_DMU_USEROBJ_PREFIX "obj-"
+
+static inline bool osd_dmu_userobj_accounting_available(struct osd_device *osd)
+{
+ return false;
+}
+#endif /* #ifdef HAVE_DMU_USEROBJ_ACCOUNTING */
+
#endif /* _OSD_INTERNAL_H */
git://git.whamcloud.com / fs / lustre-release.git / blobdiff