-/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
- * vim:expandtab:shiftwidth=8:tabstop=8:
- *
+/*
* GPL HEADER START
*
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
* Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*
- * Copyright (c) 2011, 2012, Whamcloud, Inc.
+ * Copyright (c) 2011, 2013, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
#ifndef _OSD_INTERNAL_H
#define _OSD_INTERNAL_H
-#if defined(__KERNEL__)
+/* struct mutex */
+#include <linux/mutex.h>
/* struct rw_semaphore */
#include <linux/rwsem.h>
/* struct dentry */
#include <linux/dcache.h>
/* struct dirent64 */
#include <linux/dirent.h>
-
+#include <linux/statfs.h>
#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
-
/* LUSTRE_OSD_NAME */
#include <obd.h>
#include <obd_class.h>
#include <lustre_disk.h>
#include <dt_object.h>
+#include <lustre_quota.h>
#include "osd_oi.h"
#include "osd_iam.h"
+#include "osd_scrub.h"
+#include "osd_quota_fmt.h"
struct inode;
-#define OSD_OII_NOGEN (0)
#define OSD_COUNTERS (0)
+/* ldiskfs special inode::i_state_flags need to be accessed with
+ * ldiskfs_{set,clear,test}_inode_state() only */
+
+/* OI scrub should skip this inode. */
+#define LDISKFS_STATE_LUSTRE_NOSCRUB 31
+
/** Enable thandle usage statistics */
#define OSD_THANDLE_STATS (0)
-#ifdef HAVE_QUOTA_SUPPORT
-struct osd_ctxt {
- __u32 oc_uid;
- __u32 oc_gid;
- cfs_kernel_cap_t oc_cap;
-};
-#endif
+#define MAX_OBJID_GROUP (FID_SEQ_ECHO + 1)
+
+#define OBJECTS "OBJECTS"
+#define ADMIN_USR "admin_quotafile_v2.usr"
+#define ADMIN_GRP "admin_quotafile_v2.grp"
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;
/**
* 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;
+ struct rw_semaphore oo_ext_idx_sem;
+ struct rw_semaphore oo_sem;
+ struct osd_directory *oo_dir;
+ /** protects inode attributes. */
+ 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
#endif
};
-#ifdef HAVE_LDISKFS_PDO
-
-#define osd_ldiskfs_find_entry(dir, dentry, de, lock) \
- ll_ldiskfs_find_entry(dir, dentry, de, lock)
-#define osd_ldiskfs_add_entry(handle, child, cinode, hlock) \
- ldiskfs_add_entry(handle, child, cinode, hlock)
+struct osd_obj_seq {
+ /* protects on-fly initialization */
+ int oos_subdir_count; /* subdir count for each seq */
+ struct dentry *oos_root; /* O/<seq> */
+ struct dentry **oos_dirs; /* O/<seq>/d0-dXX */
+ obd_seq oos_seq; /* seq number */
+ struct list_head oos_seq_list; /* list to seq_list */
+};
-#else /* HAVE_LDISKFS_PDO */
+struct osd_obj_map {
+ struct dentry *om_root; /* dentry for /O */
+ rwlock_t om_seq_list_lock; /* lock for seq_list */
+ struct list_head om_seq_list; /* list head for seq */
+ int om_subdir_count;
+ struct mutex om_dir_init_mutex;
+};
-struct htree_lock {
- int dummy;
+struct osd_mdobj {
+ struct dentry *om_root; /* AGENT/<index> */
+ obd_seq om_index; /* mdt index */
+ struct list_head om_list; /* list to omm_list */
};
-struct htree_lock_head {
- int dummy;
+struct osd_mdobj_map {
+ struct dentry *omm_remote_parent;
};
-#define ldiskfs_htree_lock(lock, head, inode, op) do { LBUG(); } while (0)
-#define ldiskfs_htree_unlock(lock) do { LBUG(); } while (0)
+#define osd_ldiskfs_add_entry(handle, child, cinode, hlock) \
+ ldiskfs_add_entry(handle, child, cinode, hlock)
-static inline struct htree_lock_head *ldiskfs_htree_lock_head_alloc(int dep)
-{
- LBUG();
- return NULL;
-}
+#define OSD_OTABLE_IT_CACHE_SIZE 64
+#define OSD_OTABLE_IT_CACHE_MASK (~(OSD_OTABLE_IT_CACHE_SIZE - 1))
-#define ldiskfs_htree_lock_head_free(lh) do { LBUG(); } while (0)
+struct osd_inconsistent_item {
+ /* link into osd_scrub::os_inconsistent_items,
+ * protected by osd_scrub::os_lock. */
+ struct list_head oii_list;
-#define LDISKFS_DUMMY_HTREE_LOCK 0xbabecafe
+ /* The right FID <=> ino#/gen mapping. */
+ struct osd_idmap_cache oii_cache;
-static inline struct htree_lock *ldiskfs_htree_lock_alloc(void)
-{
- return (struct htree_lock *)LDISKFS_DUMMY_HTREE_LOCK;
-}
+ unsigned int oii_insert:1; /* insert or update mapping. */
+};
-static inline void ldiskfs_htree_lock_free(struct htree_lock *lk)
-{
- LASSERT((unsigned long)lk == LDISKFS_DUMMY_HTREE_LOCK);
-}
+struct osd_otable_cache {
+ struct osd_idmap_cache ooc_cache[OSD_OTABLE_IT_CACHE_SIZE];
-#define HTREE_HBITS_DEF 0
+ /* Index for next cache slot to be filled. */
+ int ooc_producer_idx;
-#define osd_ldiskfs_find_entry(dir, dentry, de, lock) \
- ll_ldiskfs_find_entry(dir, dentry, de)
-#define osd_ldiskfs_add_entry(handle, child, cinode, lock) \
- ldiskfs_add_entry(handle, child, cinode)
+ /* Index for next cache slot to be returned by it::next(). */
+ int ooc_consumer_idx;
-#endif /* HAVE_LDISKFS_PDO */
+ /* How many items in ooc_cache. */
+ int ooc_cached_items;
-extern const int osd_dto_credits_noquota[];
+ /* Position for up layer LFSCK iteration pre-loading. */
+ __u32 ooc_pos_preload;
+};
+
+struct osd_otable_it {
+ struct osd_device *ooi_dev;
+ struct osd_otable_cache ooi_cache;
+
+ /* The following bits can be updated/checked w/o lock protection.
+ * If more bits will be introduced in the future and need lock to
+ * protect, please add comment. */
+ unsigned long ooi_used_outside:1, /* Some user out of OSD
+ * uses the iteration. */
+ ooi_all_cached:1, /* No more entries can be
+ * filled into cache. */
+ ooi_user_ready:1, /* The user out of OSD is
+ * ready to iterate. */
+ ooi_waiting:1; /* it::next is waiting. */
+};
/*
* osd device.
/* super-class */
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;
/*
* Fid Capability
*/
- unsigned int od_fl_capa:1;
+ unsigned int od_fl_capa:1,
+ od_maybe_new:1,
+ od_noscrub:1,
+ od_dirent_journal:1,
+ od_igif_inoi:1,
+ od_check_ff:1,
+ od_is_ost:1,
+ od_lma_self_repair:1;
+
unsigned long od_capa_timeout;
__u32 od_capa_alg;
struct lustre_capa_key *od_capa_keys;
- cfs_hlist_head_t *od_capa_hash;
+ struct hlist_head *od_capa_hash;
- cfs_proc_dir_entry_t *od_proc_entry;
+ struct proc_dir_entry *od_proc_entry;
struct lprocfs_stats *od_stats;
- /*
- * statfs optimization: we cache a bit.
- */
- cfs_time_t od_osfs_age;
- cfs_kstatfs_t od_kstatfs;
- cfs_spinlock_t od_osfs_lock;
- /**
- * The following flag indicates, if it is interop mode or not.
- * It will be initialized, using mount param.
- */
- __u32 od_iop_mode;
-};
+ spinlock_t od_osfs_lock;
-#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);\
- LASSERT((oh)->ot_declare_ ##op > 0); \
- ((oh)->ot_declare_ ##op)--; }
-#else
-#define OSD_DECLARE_OP(oh, op)
-#define OSD_EXEC_OP(oh, op)
-#endif
+ int od_connects;
+ struct lu_site od_site;
+
+ struct osd_obj_map *od_ost_map;
+ struct osd_mdobj_map *od_mdt_map;
+
+ unsigned long long od_readcache_max_filesize;
+ int od_read_cache;
+ int od_writethrough_cache;
+
+ struct brw_stats od_brw_stats;
+ atomic_t od_r_in_flight;
+ atomic_t od_w_in_flight;
+
+ struct mutex od_otable_mutex;
+ struct osd_otable_it *od_otable_it;
+ struct osd_scrub od_scrub;
+ struct list_head od_ios_list;
+
+ /* service name associated with the osd device */
+ char od_svname[MAX_OBD_NAME];
+ char od_mntdev[MAX_OBD_NAME];
+
+ /* quota slave instance */
+ struct qsd_instance *od_quota_slave;
+
+ /* osd seq instance */
+ struct lu_client_seq *od_cl_seq;
+};
/* 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 source child uid & gid ('..' entry update when 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);
*/
#define OSD_MAX_UGID_CNT 10
+enum {
+ OSD_OT_ATTR_SET = 0,
+ OSD_OT_PUNCH = 1,
+ OSD_OT_XATTR_SET = 2,
+ OSD_OT_CREATE = 3,
+ OSD_OT_DESTROY = 4,
+ OSD_OT_REF_ADD = 5,
+ OSD_OT_REF_DEL = 6,
+ OSD_OT_WRITE = 7,
+ OSD_OT_INSERT = 8,
+ OSD_OT_DELETE = 9,
+ OSD_OT_UPDATE = 10,
+ OSD_OT_QUOTA = 11,
+ OSD_OT_MAX = 12
+};
+
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;
+ struct ldiskfs_journal_cb_entry ot_jcb;
+ struct list_head 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
-
+ struct lquota_trans *ot_quota_trans;
#if OSD_THANDLE_STATS
/** time when this handle was allocated */
cfs_time_t oth_alloced;
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,
#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
* there would be one ext3 readdir for every mdd readdir page.
*/
-#define OSD_IT_EA_BUFSIZE (CFS_PAGE_SIZE + CFS_PAGE_SIZE/4)
+#define OSD_IT_EA_BUFSIZE (PAGE_CACHE_SIZE + PAGE_CACHE_SIZE/4)
/**
* This is iterator's in-memory data structure in interoperability
struct iam_iterator oi_it;
};
+struct osd_quota_leaf {
+ struct list_head oql_link;
+ uint oql_blk;
+};
+
+/**
+ * Iterator's in-memory data structure for quota file.
+ */
+struct osd_it_quota {
+ struct osd_object *oiq_obj;
+ /** tree blocks path to where the entry is stored */
+ uint oiq_blk[LUSTRE_DQTREEDEPTH + 1];
+ /** on-disk offset for current key where quota record can be found */
+ loff_t oiq_offset;
+ /** identifier for current quota record */
+ __u64 oiq_id;
+ /** the record index in the leaf/index block */
+ uint oiq_index[LUSTRE_DQTREEDEPTH + 1];
+ /** list of already processed leaf blocks */
+ struct list_head oiq_list;
+};
+
+#define MAX_BLOCKS_PER_PAGE (PAGE_CACHE_SIZE / 512)
+
+struct osd_iobuf {
+ wait_queue_head_t dr_wait;
+ 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 lu_buf dr_pg_buf;
+ struct page **dr_pages;
+ struct lu_buf dr_bl_buf;
+ unsigned long *dr_blocks;
+ unsigned long dr_start_time;
+ unsigned long dr_elapsed; /* how long io took */
+ struct osd_device *dr_dev;
+ unsigned int dr_init_at; /* the line iobuf was initialized */
+};
+
struct osd_thread_info {
const struct lu_env *oti_env;
/**
struct htree_lock *oti_hlock;
struct lu_fid oti_fid;
- struct osd_inode_id oti_id;
+ struct lu_fid oti_fid2;
+ struct lu_fid oti_fid3;
+ struct osd_inode_id oti_id;
+ struct osd_inode_id oti_id2;
+ struct osd_inode_id oti_id3;
+ struct ost_id oti_ostid;
+
/*
* XXX temporary: for ->i_op calls.
*/
/** osd iterator context used for iterator session */
- union {
- struct osd_it_iam oti_it;
- /** ldiskfs iterator data structure, see osd_it_ea_{init, fini} */
- struct osd_it_ea oti_it_ea;
- };
+ union {
+ struct osd_it_iam oti_it;
+ /* ldiskfs iterator data structure,
+ * see osd_it_ea_{init, fini} */
+ struct osd_it_ea oti_it_ea;
+ struct osd_it_quota oti_it_quota;
+ };
+
+ /** pre-allocated buffer used by oti_it_ea, size OSD_IT_EA_BUFSIZE */
+ void *oti_it_ea_buf;
- /** pre-allocated buffer used by oti_it_ea, size OSD_IT_EA_BUFSIZE */
- void *oti_it_ea_buf;
+ struct kstatfs oti_ksfs;
/** IAM iterator for index operation. */
struct iam_iterator oti_idx_it;
long long oti_alignment_lieutenant_colonel;
};
+ struct osd_idmap_cache oti_cache;
int oti_r_locks;
int oti_w_locks;
int oti_txns;
/** used in osd_fid_set() to put xattr */
struct lu_buf oti_buf;
+ struct lu_buf oti_big_buf;
/** used in osd_ea_fid_set() to set fid into common ea */
- struct lustre_mdt_attrs oti_mdt_attrs;
-#ifdef HAVE_QUOTA_SUPPORT
- struct osd_ctxt oti_ctxt;
-#endif
- struct lu_env oti_obj_delete_tx_env;
+ union {
+ struct lustre_mdt_attrs oti_mdt_attrs;
+ /* old LMA for compatibility */
+ char oti_mdt_attrs_old[LMA_OLD_SIZE];
+ };
+ /** 0-copy IO */
+ struct osd_iobuf oti_iobuf;
+ struct inode oti_inode;
#define OSD_FID_REC_SZ 32
- char oti_ldp[OSD_FID_REC_SZ];
- char oti_ldp2[OSD_FID_REC_SZ];
+ char oti_ldp[OSD_FID_REC_SZ];
+ char oti_ldp2[OSD_FID_REC_SZ];
+
+ /* used by quota code */
+ union {
+#ifdef HAVE_DQUOT_FS_DISK_QUOTA
+ struct fs_disk_quota oti_fdq;
+#else
+ struct if_dqblk oti_dqblk;
+#endif
+ struct if_dqinfo oti_dqinfo;
+ };
+ struct lquota_id_info oti_qi;
+ struct lquota_trans oti_quota_trans;
+ union lquota_rec oti_quota_rec;
+ __u64 oti_quota_id;
+ struct lu_seq_range oti_seq_range;
+
+ /* Tracking for transaction credits, to allow debugging and optimizing
+ * cases where a large number of credits are being allocated for
+ * single transaction. */
+ unsigned short oti_declare_ops[OSD_OT_MAX];
+ unsigned short oti_declare_ops_rb[OSD_OT_MAX];
+ unsigned short oti_declare_ops_cred[OSD_OT_MAX];
+ bool oti_rollback;
+
+ char oti_name[48];
+ union {
+ struct filter_fid_old oti_ff;
+ struct filter_fid oti_ff_new;
+ };
};
extern int ldiskfs_pdo;
+static inline int __osd_xattr_get(struct inode *inode, struct dentry *dentry,
+ const char *name, void *buf, int len)
+{
+ if (inode == NULL)
+ return -EINVAL;
+
+ dentry->d_inode = inode;
+ dentry->d_sb = inode->i_sb;
+ return inode->i_op->getxattr(dentry, name, buf, len);
+}
+
+static inline int __osd_xattr_set(struct osd_thread_info *info,
+ struct inode *inode, const char *name,
+ const void *buf, int buflen, int fl)
+{
+ struct dentry *dentry = &info->oti_child_dentry;
+
+ ll_vfs_dq_init(inode);
+ dentry->d_inode = inode;
+ dentry->d_sb = inode->i_sb;
+ return inode->i_op->setxattr(dentry, name, buf, buflen, fl);
+}
+
#ifdef LPROCFS
/* osd_lproc.c */
-void lprocfs_osd_init_vars(struct lprocfs_static_vars *lvars);
+extern struct lprocfs_seq_vars lprocfs_osd_obd_vars[];
+extern struct lprocfs_seq_vars lprocfs_osd_module_vars[];
int osd_procfs_init(struct osd_device *osd, const char *name);
int osd_procfs_fini(struct osd_device *osd);
-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);
+struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
+ struct osd_inode_id *id);
+int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
+ const struct lu_fid *fid, __u32 compat, __u32 incompat);
+int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
+ struct dentry *dentry, struct lustre_mdt_attrs *lma);
+int osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
+ struct osd_inode_id *id, const struct lu_fid *fid);
+int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
+ struct dentry *dentry, struct lu_fid *fid);
+
+int osd_obj_map_init(const struct lu_env *env, struct osd_device *osd);
+void osd_obj_map_fini(struct osd_device *dev);
+int osd_obj_map_lookup(struct osd_thread_info *info, struct osd_device *osd,
+ const struct lu_fid *fid, struct osd_inode_id *id);
+int osd_obj_map_insert(struct osd_thread_info *info, struct osd_device *osd,
+ const struct lu_fid *fid, const struct osd_inode_id *id,
+ handle_t *th);
+int osd_obj_map_delete(struct osd_thread_info *info, struct osd_device *osd,
+ const struct lu_fid *fid, handle_t *th);
+int osd_obj_map_update(struct osd_thread_info *info, struct osd_device *osd,
+ const struct lu_fid *fid, const struct osd_inode_id *id,
+ handle_t *th);
+int osd_obj_map_recover(struct osd_thread_info *info, struct osd_device *osd,
+ struct inode *src_parent, struct dentry *src_child,
+ const struct lu_fid *fid);
+int osd_obj_spec_lookup(struct osd_thread_info *info, struct osd_device *osd,
+ const struct lu_fid *fid, struct osd_inode_id *id);
+int osd_obj_spec_insert(struct osd_thread_info *info, struct osd_device *osd,
+ const struct lu_fid *fid, const struct osd_inode_id *id,
+ handle_t *th);
+int osd_obj_spec_update(struct osd_thread_info *info, struct osd_device *osd,
+ const struct lu_fid *fid, const struct osd_inode_id *id,
+ handle_t *th);
+
+void osd_scrub_file_reset(struct osd_scrub *scrub, __u8 *uuid, __u64 flags);
+int osd_scrub_file_store(struct osd_scrub *scrub);
+char *osd_lf_fid2name(const struct lu_fid *fid);
+int osd_scrub_start(struct osd_device *dev);
+int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev);
+void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev);
+int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic,
+ int insert);
+int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
+ struct osd_inode_id *id);
+int osd_scrub_dump(struct seq_file *m, struct osd_device *dev);
+
+int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
+ obd_seq seq, struct lu_seq_range *range);
+
+int osd_delete_from_remote_parent(const struct lu_env *env,
+ struct osd_device *osd,
+ struct osd_object *obj,
+ struct osd_thandle *oh);
+int osd_add_to_remote_parent(const struct lu_env *env, struct osd_device *osd,
+ struct osd_object *obj, struct osd_thandle *oh);
+int osd_lookup_in_remote_parent(struct osd_thread_info *oti,
+ struct osd_device *osd,
+ const struct lu_fid *fid,
+ struct osd_inode_id *id);
+
+int osd_ost_seq_exists(struct osd_thread_info *info, struct osd_device *osd,
+ __u64 seq);
+/* osd_quota_fmt.c */
+int walk_tree_dqentry(const struct lu_env *env, struct osd_object *obj,
+ int type, uint blk, int depth, uint index,
+ struct osd_it_quota *it);
+int walk_block_dqentry(const struct lu_env *env, struct osd_object *obj,
+ int type, uint blk, uint index,
+ struct osd_it_quota *it);
+loff_t find_tree_dqentry(const struct lu_env *env,
+ struct osd_object *obj, int type,
+ qid_t dqid, uint blk, int depth,
+ struct osd_it_quota *it);
+/* osd_quota.c */
+int osd_declare_qid(const struct lu_env *env, struct osd_thandle *oh,
+ struct lquota_id_info *qi, struct osd_object *obj,
+ bool enforce, int *flags);
+int osd_declare_inode_qid(const struct lu_env *env, qid_t uid, qid_t gid,
+ long long space, struct osd_thandle *oh,
+ struct osd_object *obj, bool is_blk, int *flags,
+ bool force);
+const struct dt_rec *osd_quota_pack(struct osd_object *obj,
+ const struct dt_rec *rec,
+ union lquota_rec *quota_rec);
+void osd_quota_unpack(struct osd_object *obj, const struct dt_rec *rec);
+int osd_quota_migration(const struct lu_env *env, struct dt_object *dt,
+ const struct dt_index_features *feat);
+
+static inline bool is_quota_glb_feat(const struct dt_index_features *feat)
+{
+ return (feat == &dt_quota_iusr_features ||
+ feat == &dt_quota_busr_features ||
+ feat == &dt_quota_igrp_features ||
+ feat == &dt_quota_bgrp_features) ? true : false;
+}
+
+#ifndef HAVE_I_UID_READ
+static inline uid_t i_uid_read(const struct inode *inode)
+{
+ return inode->i_uid;
+}
+
+static inline gid_t i_gid_read(const struct inode *inode)
+{
+ return inode->i_gid;
+}
+
+static inline void i_uid_write(struct inode *inode, uid_t uid)
+{
+ inode->i_uid = uid;
+}
+
+static inline void i_gid_write(struct inode *inode, gid_t gid)
+{
+ inode->i_gid = gid;
+}
+#endif
/*
* 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.
- */
-#define OSD_ROOT_SEQ 2
-static inline int osd_fid_is_root(const struct lu_fid *fid)
-{
- return fid_seq(fid) == OSD_ROOT_SEQ;
-}
+#define OSD_MAX_CACHE_SIZE OBD_OBJECT_EOF
+
+extern const struct dt_index_operations osd_otable_ops;
-static inline int osd_fid_is_igif(const struct lu_fid *fid)
+static inline int osd_oi_fid2idx(struct osd_device *dev,
+ const struct lu_fid *fid)
{
- return fid_is_igif(fid) || osd_fid_is_root(fid);
+ return fid->f_seq & (dev->od_oi_count - 1);
}
-static inline struct osd_oi *
-osd_fid2oi(struct osd_device *osd, const struct lu_fid *fid)
+static inline struct osd_oi *osd_fid2oi(struct osd_device *osd,
+ const struct lu_fid *fid)
{
- if (!fid_is_norm(fid))
- return NULL;
-
- 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)];
+ LASSERTF(!fid_is_idif(fid), DFID"\n", PFID(fid));
+ LASSERTF(!fid_is_last_id(fid), DFID"\n", PFID(fid));
+ LASSERTF(osd->od_oi_table != NULL && osd->od_oi_count >= 1,
+ DFID"\n", PFID(fid));
+ /* 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[osd_oi_fid2idx(osd, fid)];
}
-/*
- * Helpers.
- */
extern const struct lu_device_operations osd_lu_ops;
static inline int lu_device_is_osd(const struct lu_device *d)
static inline struct super_block *osd_sb(const struct osd_device *dev)
{
- return dev->od_mount->lmi_mnt->mnt_sb;
+ return dev->od_mnt->mnt_sb;
}
static inline int osd_object_is_root(const struct osd_object *obj)
return LDISKFS_SB(osd_sb(dev))->s_journal;
}
+static inline struct seq_server_site *osd_seq_site(struct osd_device *osd)
+{
+ return osd->od_dt_dev.dd_lu_dev.ld_site->ld_seq_site;
+}
+
+static inline char *osd_name(struct osd_device *osd)
+{
+ return osd->od_dt_dev.dd_lu_dev.ld_obd->obd_name;
+}
+
extern const struct dt_body_operations osd_body_ops;
extern struct lu_context_key osd_key;
return lu_context_key_get(&env->le_ctx, &osd_key);
}
-#endif /* __KERNEL__ */
+extern const struct dt_body_operations osd_body_ops_new;
+
+/**
+ * IAM Iterator
+ */
+static inline
+struct iam_path_descr *osd_it_ipd_get(const struct lu_env *env,
+ const struct iam_container *bag)
+{
+ return bag->ic_descr->id_ops->id_ipd_alloc(bag,
+ osd_oti_get(env)->oti_it_ipd);
+}
+
+static inline
+struct iam_path_descr *osd_idx_ipd_get(const struct lu_env *env,
+ const struct iam_container *bag)
+{
+ return bag->ic_descr->id_ops->id_ipd_alloc(bag,
+ osd_oti_get(env)->oti_idx_ipd);
+}
+
+static inline void osd_ipd_put(const struct lu_env *env,
+ const struct iam_container *bag,
+ struct iam_path_descr *ipd)
+{
+ bag->ic_descr->id_ops->id_ipd_free(ipd);
+}
+
+int osd_ldiskfs_read(struct inode *inode, void *buf, int size, loff_t *offs);
+int osd_ldiskfs_write_record(struct inode *inode, void *buf, int bufsize,
+ int write_NUL, loff_t *offs, handle_t *handle);
+
+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;
+}
+
+extern int osd_trans_declare_op2rb[];
+extern int ldiskfs_track_declares_assert;
+
+static inline void osd_trans_declare_op(const struct lu_env *env,
+ struct osd_thandle *oh,
+ unsigned int op, int credits)
+{
+ struct osd_thread_info *oti = osd_oti_get(env);
+
+ LASSERT(oh->ot_handle == NULL);
+ if (unlikely(op >= OSD_OT_MAX)) {
+ if (unlikely(ldiskfs_track_declares_assert)) {
+ LASSERT(op < OSD_OT_MAX);
+ } else {
+ CWARN("%s: Invalid operation index %d\n",
+ osd_name(oti->oti_dev), op);
+ libcfs_debug_dumpstack(NULL);
+ }
+ } else {
+ oti->oti_declare_ops[op]++;
+ oti->oti_declare_ops_cred[op] += credits;
+ }
+ oh->ot_credits += credits;
+}
+
+static inline void osd_trans_exec_op(const struct lu_env *env,
+ struct thandle *th, unsigned int op)
+{
+ struct osd_thread_info *oti = osd_oti_get(env);
+ struct osd_thandle *oh = container_of(th, struct osd_thandle,
+ ot_super);
+ unsigned int rb;
+
+ LASSERT(oh->ot_handle != NULL);
+ if (unlikely(op >= OSD_OT_MAX)) {
+ if (unlikely(ldiskfs_track_declares_assert))
+ LASSERT(op < OSD_OT_MAX);
+ else {
+ CWARN("%s: Invalid operation index %d\n",
+ osd_name(oti->oti_dev), op);
+ libcfs_debug_dumpstack(NULL);
+ return;
+ }
+ }
+
+ if (likely(!oti->oti_rollback && oti->oti_declare_ops[op] > 0)) {
+ oti->oti_declare_ops[op]--;
+ oti->oti_declare_ops_rb[op]++;
+ } else {
+ /* all future updates are considered rollback */
+ oti->oti_rollback = true;
+ rb = osd_trans_declare_op2rb[op];
+ if (unlikely(rb >= OSD_OT_MAX)) {
+ if (unlikely(ldiskfs_track_declares_assert))
+ LASSERTF(rb < OSD_OT_MAX, "rb = %u\n", rb);
+ else {
+ CWARN("%s: Invalid rollback index %d\n",
+ osd_name(oti->oti_dev), rb);
+ libcfs_debug_dumpstack(NULL);
+ return;
+ }
+ }
+ if (unlikely(oti->oti_declare_ops_rb[rb] == 0)) {
+ if (unlikely(ldiskfs_track_declares_assert))
+ LASSERTF(oti->oti_declare_ops_rb[rb] > 0,
+ "rb = %u\n", rb);
+ else {
+ CWARN("%s: Overflow in tracking declares for "
+ "index, rb = %d\n",
+ osd_name(oti->oti_dev), rb);
+ libcfs_debug_dumpstack(NULL);
+ return;
+ }
+ }
+ oti->oti_declare_ops_rb[rb]--;
+ }
+}
+
+static inline void osd_trans_declare_rb(const struct lu_env *env,
+ struct thandle *th, unsigned int op)
+{
+ struct osd_thread_info *oti = osd_oti_get(env);
+ struct osd_thandle *oh = container_of(th, struct osd_thandle,
+ ot_super);
+
+ LASSERT(oh->ot_handle != NULL);
+ if (unlikely(op >= OSD_OT_MAX)) {
+ if (unlikely(ldiskfs_track_declares_assert))
+ LASSERT(op < OSD_OT_MAX);
+ else {
+ CWARN("%s: Invalid operation index %d\n",
+ osd_name(oti->oti_dev), op);
+ libcfs_debug_dumpstack(NULL);
+ }
+
+ } else {
+ oti->oti_declare_ops_rb[op]++;
+ }
+}
+
+/**
+ * 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;
+}
+
+/**
+ * Quota/Accounting handling
+ */
+extern const struct dt_index_operations osd_acct_index_ops;
+int osd_acct_obj_lookup(struct osd_thread_info *info, struct osd_device *osd,
+ const struct lu_fid *fid, struct osd_inode_id *id);
+
+/* copy from fs/ext4/dir.c */
+static inline int is_32bit_api(void)
+{
+#ifdef CONFIG_COMPAT
+ return is_compat_task();
+#else
+ return (BITS_PER_LONG == 32);
+#endif
+}
+
+static inline loff_t ldiskfs_get_htree_eof(struct file *filp)
+{
+ if ((filp->f_mode & FMODE_32BITHASH) ||
+ (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
+ return LDISKFS_HTREE_EOF_32BIT;
+ else
+ return LDISKFS_HTREE_EOF_64BIT;
+}
+
+static inline int fid_is_internal(const struct lu_fid *fid)
+{
+ return (!fid_is_namespace_visible(fid) && !fid_is_idif(fid));
+}
+
+static inline unsigned long osd_remote_parent_ino(struct osd_device *dev)
+{
+ return dev->od_mdt_map->omm_remote_parent->d_inode->i_ino;
+}
+
+#ifdef JOURNAL_START_HAS_3ARGS
+# define osd_journal_start_sb(sb, type, nblock) \
+ ldiskfs_journal_start_sb(sb, type, nblock)
+# define osd_ldiskfs_append(handle, inode, nblock, err) \
+ ldiskfs_append(handle, inode, nblock)
+# define osd_ldiskfs_find_entry(dir, name, de, inlined, lock) \
+ ldiskfs_find_entry(dir, name, de, inlined, lock)
+# define osd_journal_start(inode, type, nblocks) \
+ ldiskfs_journal_start(inode, type, nblocks)
+# define osd_transaction_size(dev) \
+ (osd_journal(dev)->j_max_transaction_buffers / 2)
+#else
+# define LDISKFS_HT_MISC 0
+# define osd_journal_start_sb(sb, type, nblock) \
+ ldiskfs_journal_start_sb(sb, nblock)
+# define osd_ldiskfs_append(handle, inode, nblock, err) \
+ ldiskfs_append(handle, inode, nblock, err)
+# define osd_ldiskfs_find_entry(dir, name, de, inlined, lock) \
+ ldiskfs_find_entry(dir, name, de, lock)
+# define osd_journal_start(inode, type, nblocks) \
+ ldiskfs_journal_start(inode, nblocks)
+# define osd_transaction_size(dev) \
+ (osd_journal(dev)->j_max_transaction_buffers)
+#endif
+
+void ldiskfs_inc_count(handle_t *handle, struct inode *inode);
+void ldiskfs_dec_count(handle_t *handle, struct inode *inode);
+
+void osd_fini_iobuf(struct osd_device *d, struct osd_iobuf *iobuf);
+
#endif /* _OSD_INTERNAL_H */
git://git.whamcloud.com / fs / lustre-release.git / blobdiff