/*
* Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
- */
-/*
- * Copyright (c) 2011 Whamcloud, Inc.
+ *
+ * Copyright (c) 2011, 2012, Whamcloud, Inc.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
/* struct dirent64 */
#include <linux/dirent.h>
-#ifdef HAVE_EXT4_LDISKFS
#include <ldiskfs/ldiskfs.h>
#include <ldiskfs/ldiskfs_jbd2.h>
-# ifdef HAVE_LDISKFS_JOURNAL_CALLBACK_ADD
-# define journal_callback ldiskfs_journal_cb_entry
-# define osd_journal_callback_set(handle, func, jcb) ldiskfs_journal_callback_add(handle, func, jcb)
-# else
-# define osd_journal_callback_set(handle, func, jcb) jbd2_journal_callback_set(handle, func, jcb)
-# endif
+#ifdef HAVE_LDISKFS_JOURNAL_CALLBACK_ADD
+# define journal_callback ldiskfs_journal_cb_entry
+# define osd_journal_callback_set(handle, func, jcb) \
+ ldiskfs_journal_callback_add(handle, func, jcb)
#else
-#include <linux/jbd.h>
-#include <linux/ldiskfs_fs.h>
-#include <linux/ldiskfs_jbd.h>
-#define osd_journal_callback_set(handle, func, jcb) journal_callback_set(handle, func, jcb)
+# define osd_journal_callback_set(handle, func, jcb) \
+ jbd2_journal_callback_set(handle, func, jcb)
#endif
+/* fsfilt_{get|put}_ops */
+#include <lustre_fsfilt.h>
/* LUSTRE_OSD_NAME */
#include <obd.h>
/* class_register_type(), class_unregister_type(), class_get_type() */
#include <obd_class.h>
#include <lustre_disk.h>
-
#include <dt_object.h>
+
#include "osd_oi.h"
#include "osd_iam.h"
};
#endif
+struct osd_directory {
+ struct iam_container od_container;
+ struct iam_descr od_descr;
+};
+
+/*
+ * Object Index (oi) instance.
+ */
+struct osd_oi {
+ /*
+ * underlying index object, where fid->id mapping in stored.
+ */
+ struct inode *oi_inode;
+ struct osd_directory oi_dir;
+};
+
+extern const int osd_dto_credits_noquota[];
+
+struct osd_object {
+ struct dt_object oo_dt;
+ /**
+ * Inode for file system object represented by this osd_object. This
+ * inode is pinned for the whole duration of lu_object life.
+ *
+ * Not modified concurrently (either setup early during object
+ * creation, or assigned by osd_object_create() under write lock).
+ */
+ struct inode *oo_inode;
+ /**
+ * to protect index ops.
+ */
+ struct htree_lock_head *oo_hl_head;
+ cfs_rw_semaphore_t oo_ext_idx_sem;
+ cfs_rw_semaphore_t oo_sem;
+ struct osd_directory *oo_dir;
+ /** protects inode attributes. */
+ cfs_spinlock_t oo_guard;
+ /**
+ * Following two members are used to indicate the presence of dot and
+ * dotdot in the given directory. This is required for interop mode
+ * (b11826).
+ */
+ int oo_compat_dot_created;
+ int oo_compat_dotdot_created;
+
+ const struct lu_env *oo_owner;
+#ifdef CONFIG_LOCKDEP
+ struct lockdep_map oo_dep_map;
+#endif
+};
+
#ifdef HAVE_LDISKFS_PDO
#define osd_ldiskfs_find_entry(dir, dentry, de, lock) \
#endif /* HAVE_LDISKFS_PDO */
+extern const int osd_dto_credits_noquota[];
+
/*
* osd device.
*/
struct dt_device od_dt_dev;
/* information about underlying file system */
struct lustre_mount_info *od_mount;
- /*
- * XXX temporary stuff for object index: directory where every object
- * is named by its fid.
- */
- struct dt_object *od_obj_area;
+ struct vfsmount *od_mnt;
/* object index */
- struct osd_oi *od_oi_table;
+ struct osd_oi **od_oi_table;
/* total number of OI containers */
int od_oi_count;
/*
* statfs optimization: we cache a bit.
*/
cfs_time_t od_osfs_age;
- cfs_kstatfs_t od_kstatfs;
+ struct obd_statfs od_statfs;
cfs_spinlock_t od_osfs_lock;
/**
* It will be initialized, using mount param.
*/
__u32 od_iop_mode;
+
+ struct fsfilt_operations *od_fsops;
+
+ /*
+ * mapping for legacy OST objids
+ */
+ struct osd_compat_objid *od_ost_map;
+
+ unsigned long long od_readcache_max_filesize;
+ int od_read_cache;
+ int od_writethrough_cache;
+
+ struct brw_stats od_brw_stats;
+ cfs_atomic_t od_r_in_flight;
+ cfs_atomic_t od_w_in_flight;
+};
+
+#define OSD_TRACK_DECLARES
+#ifdef OSD_TRACK_DECLARES
+#define OSD_DECLARE_OP(oh, op) { \
+ LASSERT(oh->ot_handle == NULL); \
+ ((oh)->ot_declare_ ##op)++; }
+#define OSD_EXEC_OP(handle,op) { \
+ struct osd_thandle *oh; \
+ oh = container_of0(handle, struct osd_thandle, ot_super);\
+ if (((oh)->ot_declare_ ##op) > 0) { \
+ ((oh)->ot_declare_ ##op)--; \
+ } \
+ }
+#else
+#define OSD_DECLARE_OP(oh, op)
+#define OSD_EXEC_OP(oh, op)
+#endif
+
+/* There are at most 10 uid/gids are affected in a transaction, and
+ * that's rename case:
+ * - 2 for source parent uid & gid;
+ * - 2 for source child uid & gid ('..' entry update when the child
+ * is directory);
+ * - 2 for target parent uid & gid;
+ * - 2 for target child uid & gid (if the target child exists);
+ * - 2 for root uid & gid (last_rcvd, llog, etc);
+ *
+ * The 0 to (OSD_MAX_UGID_CNT - 1) bits of ot_id_type is for indicating
+ * the id type of each id in the ot_id_array.
+ */
+#define OSD_MAX_UGID_CNT 10
+
+struct osd_thandle {
+ struct thandle ot_super;
+ handle_t *ot_handle;
+ struct journal_callback ot_jcb;
+ cfs_list_t ot_dcb_list;
+ /* Link to the device, for debugging. */
+ struct lu_ref_link *ot_dev_link;
+ unsigned short ot_credits;
+ unsigned short ot_id_cnt;
+ unsigned short ot_id_type;
+ uid_t ot_id_array[OSD_MAX_UGID_CNT];
+
+#ifdef OSD_TRACK_DECLARES
+ unsigned char ot_declare_attr_set;
+ unsigned char ot_declare_punch;
+ unsigned char ot_declare_xattr_set;
+ unsigned char ot_declare_create;
+ unsigned char ot_declare_destroy;
+ unsigned char ot_declare_ref_add;
+ unsigned char ot_declare_ref_del;
+ unsigned char ot_declare_write;
+ unsigned char ot_declare_insert;
+ unsigned char ot_declare_delete;
+#endif
+
+#if OSD_THANDLE_STATS
+ /** time when this handle was allocated */
+ cfs_time_t oth_alloced;
+
+ /** time when this thanle was started */
+ cfs_time_t oth_started;
+#endif
+};
+
+/**
+ * Basic transaction credit op
+ */
+enum dt_txn_op {
+ DTO_INDEX_INSERT,
+ DTO_INDEX_DELETE,
+ DTO_INDEX_UPDATE,
+ DTO_OBJECT_CREATE,
+ DTO_OBJECT_DELETE,
+ DTO_ATTR_SET_BASE,
+ DTO_XATTR_SET,
+ DTO_LOG_REC, /**< XXX temporary: dt layer knows nothing about llog. */
+ DTO_WRITE_BASE,
+ DTO_WRITE_BLOCK,
+ DTO_ATTR_SET_CHOWN,
+
+ DTO_NR
};
/*
#ifdef LPROCFS
enum {
+ LPROC_OSD_READ_BYTES = 0,
+ LPROC_OSD_WRITE_BYTES = 1,
+ LPROC_OSD_GET_PAGE = 2,
+ LPROC_OSD_NO_PAGE = 3,
+ LPROC_OSD_CACHE_ACCESS = 4,
+ LPROC_OSD_CACHE_HIT = 5,
+ LPROC_OSD_CACHE_MISS = 6,
+
#if OSD_THANDLE_STATS
LPROC_OSD_THANDLE_STARTING,
LPROC_OSD_THANDLE_OPEN,
LPROC_OSD_THANDLE_CLOSING,
#endif
- LPROC_OSD_NR
+ LPROC_OSD_LAST,
};
#endif
struct iam_iterator oi_it;
};
+#define MAX_BLOCKS_PER_PAGE (CFS_PAGE_SIZE / 512)
+
+struct osd_iobuf {
+ cfs_waitq_t dr_wait;
+ cfs_atomic_t dr_numreqs; /* number of reqs being processed */
+ int dr_max_pages;
+ int dr_npages;
+ int dr_error;
+ int dr_frags;
+ unsigned int dr_ignore_quota:1;
+ unsigned int dr_elapsed_valid:1; /* we really did count time */
+ unsigned int dr_rw:1;
+ struct page *dr_pages[PTLRPC_MAX_BRW_PAGES];
+ unsigned long dr_blocks[PTLRPC_MAX_BRW_PAGES*MAX_BLOCKS_PER_PAGE];
+ unsigned long dr_start_time;
+ unsigned long dr_elapsed; /* how long io took */
+ struct osd_device *dr_dev;
+};
+
struct osd_thread_info {
const struct lu_env *oti_env;
/**
struct lu_fid oti_fid;
struct osd_inode_id oti_id;
+ struct ost_id oti_ostid;
+
/*
* XXX temporary: for ->i_op calls.
*/
/** pre-allocated buffer used by oti_it_ea, size OSD_IT_EA_BUFSIZE */
void *oti_it_ea_buf;
+ cfs_kstatfs_t oti_ksfs;
+
/** IAM iterator for index operation. */
struct iam_iterator oti_idx_it;
struct lu_buf oti_buf;
/** used in osd_ea_fid_set() to set fid into common ea */
struct lustre_mdt_attrs oti_mdt_attrs;
+ /** 0-copy IO */
+ struct osd_iobuf oti_iobuf;
+ struct inode oti_inode;
+ int oti_created[PTLRPC_MAX_BRW_PAGES];
#ifdef HAVE_QUOTA_SUPPORT
struct osd_ctxt oti_ctxt;
#endif
void osd_lprocfs_time_start(const struct lu_env *env);
void osd_lprocfs_time_end(const struct lu_env *env,
struct osd_device *osd, int op);
+void osd_brw_stats_update(struct osd_device *osd, struct osd_iobuf *iobuf);
+
#endif
int osd_statfs(const struct lu_env *env, struct dt_device *dev,
- cfs_kstatfs_t *sfs);
+ struct obd_statfs *sfs);
+int osd_object_auth(const struct lu_env *env, struct dt_object *dt,
+ struct lustre_capa *capa, __u64 opc);
+void osd_declare_qid(struct dt_object *dt, struct osd_thandle *oh,
+ int type, uid_t id, struct inode *inode);
+struct inode *osd_iget(struct osd_thread_info *info,
+ struct osd_device *dev,
+ const struct osd_inode_id *id);
+
+int osd_compat_init(struct osd_device *dev);
+void osd_compat_fini(struct osd_device *dev);
+int osd_compat_objid_lookup(struct osd_thread_info *info,
+ struct osd_device *osd,
+ const struct lu_fid *fid, struct osd_inode_id *id);
+int osd_compat_objid_insert(struct osd_thread_info *info,
+ struct osd_device *osd,
+ const struct lu_fid *fid,
+ const struct osd_inode_id *id, struct thandle *th);
+int osd_compat_objid_delete(struct osd_thread_info *info,
+ struct osd_device *osd,
+ const struct lu_fid *fid, struct thandle *th);
+int osd_compat_spec_lookup(struct osd_thread_info *info,
+ struct osd_device *osd,
+ const struct lu_fid *fid, struct osd_inode_id *id);
+int osd_compat_spec_insert(struct osd_thread_info *info,
+ struct osd_device *osd,
+ const struct lu_fid *fid,
+ const struct osd_inode_id *id, struct thandle *th);
/*
* Invariants, assertions.
#define osd_invariant(obj) (1)
#endif
-/* The on-disk extN format reserves inodes 0-11 for internal filesystem
- * use, and these inodes will be invisible on client side, so the valid
- * sequence for IGIF fid is 12-0xffffffff. But root inode (2#) will be seen
- * on server side (osd), and it should be valid too here.
+static inline struct osd_oi *osd_fid2oi(struct osd_device *osd,
+ const struct lu_fid *fid)
+{
+ LASSERT(!fid_is_idif(fid));
+ LASSERT(!fid_is_igif(fid));
+ LASSERT(osd->od_oi_table != NULL && osd->od_oi_count >= 1);
+ /* It can work even od_oi_count equals to 1 although it's unexpected,
+ * the only reason we set it to 1 is for performance measurement */
+ return osd->od_oi_table[fid->f_seq & (osd->od_oi_count - 1)];
+}
+
+extern const struct lu_device_operations osd_lu_ops;
+
+static inline 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 inline 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 inline 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 inline struct osd_device *osd_obj2dev(const struct osd_object *o)
+{
+ return osd_dev(o->oo_dt.do_lu.lo_dev);
+}
+
+static inline struct super_block *osd_sb(const struct osd_device *dev)
+{
+ return dev->od_mount->lmi_mnt->mnt_sb;
+}
+
+static inline int osd_object_is_root(const struct osd_object *obj)
+{
+ return osd_sb(osd_obj2dev(obj))->s_root->d_inode == obj->oo_inode;
+}
+
+static inline 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 inline struct osd_object *osd_dt_obj(const struct dt_object *d)
+{
+ return osd_obj(&d->do_lu);
+}
+
+static inline struct lu_device *osd2lu_dev(struct osd_device *osd)
+{
+ return &osd->od_dt_dev.dd_lu_dev;
+}
+
+static inline journal_t *osd_journal(const struct osd_device *dev)
+{
+ return LDISKFS_SB(osd_sb(dev))->s_journal;
+}
+
+extern const struct dt_body_operations osd_body_ops;
+extern struct lu_context_key osd_key;
+
+static inline struct osd_thread_info *osd_oti_get(const struct lu_env *env)
+{
+ return lu_context_key_get(&env->le_ctx, &osd_key);
+}
+
+extern const struct dt_body_operations osd_body_ops_new;
+
+/**
+ * IAM Iterator
*/
-#define OSD_ROOT_SEQ 2
-static inline int osd_fid_is_root(const struct lu_fid *fid)
+static inline
+struct iam_path_descr *osd_it_ipd_get(const struct lu_env *env,
+ const struct iam_container *bag)
{
- return fid_seq(fid) == OSD_ROOT_SEQ;
+ return bag->ic_descr->id_ops->id_ipd_alloc(bag,
+ osd_oti_get(env)->oti_it_ipd);
}
-static inline int osd_fid_is_igif(const struct lu_fid *fid)
+static inline
+struct iam_path_descr *osd_idx_ipd_get(const struct lu_env *env,
+ const struct iam_container *bag)
{
- return fid_is_igif(fid) || osd_fid_is_root(fid);
+ return bag->ic_descr->id_ops->id_ipd_alloc(bag,
+ osd_oti_get(env)->oti_idx_ipd);
}
-static inline struct osd_oi *
-osd_fid2oi(struct osd_device *osd, const struct lu_fid *fid)
+static inline void osd_ipd_put(const struct lu_env *env,
+ const struct iam_container *bag,
+ struct iam_path_descr *ipd)
{
- if (!fid_is_norm(fid))
- return NULL;
+ bag->ic_descr->id_ops->id_ipd_free(ipd);
+}
- LASSERT(osd->od_oi_table != NULL && osd->od_oi_count >= 1);
- /* It can work even od_oi_count equals to 1 although it's unexpected,
- * the only reason we set it to 1 is for performance measurement */
- return &osd->od_oi_table[fid->f_seq & (osd->od_oi_count - 1)];
+int osd_ldiskfs_read(struct inode *inode, void *buf, int size, loff_t *offs);
+
+static inline
+struct dentry *osd_child_dentry_by_inode(const struct lu_env *env,
+ struct inode *inode,
+ const char *name, const int namelen)
+{
+ struct osd_thread_info *info = osd_oti_get(env);
+ struct dentry *child_dentry = &info->oti_child_dentry;
+ struct dentry *obj_dentry = &info->oti_obj_dentry;
+
+ obj_dentry->d_inode = inode;
+ obj_dentry->d_sb = inode->i_sb;
+ obj_dentry->d_name.hash = 0;
+
+ child_dentry->d_name.hash = 0;
+ child_dentry->d_parent = obj_dentry;
+ child_dentry->d_name.name = name;
+ child_dentry->d_name.len = namelen;
+ return child_dentry;
+}
+
+/**
+ * Helper function to pack the fid, ldiskfs stores fid in packed format.
+ */
+static inline
+void osd_fid_pack(struct osd_fid_pack *pack, const struct dt_rec *fid,
+ struct lu_fid *befider)
+{
+ fid_cpu_to_be(befider, (struct lu_fid *)fid);
+ memcpy(pack->fp_area, befider, sizeof(*befider));
+ pack->fp_len = sizeof(*befider) + 1;
+}
+
+static inline
+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;
}
#endif /* __KERNEL__ */