X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fosd-ldiskfs%2Fosd_internal.h;h=f28af744486f3a83d2b2b11bf70601176cbd9d2b;hp=a9450a41262df6f94a0a463ca52ae0fbee4350f7;hb=f4ea7b630b8adc9856ee67c6d16549f36e14efd1;hpb=4db03b974810789133e354b08029f0eb261fad5e diff --git a/lustre/osd-ldiskfs/osd_internal.h b/lustre/osd-ldiskfs/osd_internal.h index a9450a4..f28af74 100644 --- a/lustre/osd-ldiskfs/osd_internal.h +++ b/lustre/osd-ldiskfs/osd_internal.h @@ -1,6 +1,4 @@ -/* -*- 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. @@ -28,9 +26,8 @@ /* * 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, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -55,97 +52,178 @@ /* struct dirent64 */ #include -#ifdef HAVE_EXT4_LDISKFS #include #include -# 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 -#include -#include -#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_OSD_NAME */ #include /* class_register_type(), class_unregister_type(), class_get_type() */ #include #include - #include +#include + #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 + +/* Do not add OI mapping for this inode. */ +#define LDISKFS_STATE_LUSTRE_NO_OI 30 + /** 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; +#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; + /** + * 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; + 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 + * (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 +}; + +struct osd_obj_seq { + /* protects on-fly initialization */ + int oos_subdir_count; /* subdir count for each seq */ + struct dentry *oos_root; /* O/ */ + struct dentry **oos_dirs; /* O//d0-dXX */ + obd_seq oos_seq; /* seq number */ + cfs_list_t oos_seq_list; /* list to seq_list */ +}; -#ifdef HAVE_LDISKFS_PDO +struct osd_obj_map { + struct dentry *om_root; /* dentry for /O */ + rwlock_t om_seq_list_lock; /* lock for seq_list */ + cfs_list_t om_seq_list; /* list head for seq */ + int om_subdir_count; + struct semaphore om_dir_init_sem; +}; #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) -#else /* HAVE_LDISKFS_PDO */ - -struct htree_lock { - int dummy; -}; +#define OSD_OTABLE_IT_CACHE_SIZE 128 +#define OSD_OTABLE_IT_CACHE_MASK (~(OSD_OTABLE_IT_CACHE_SIZE - 1)) -struct htree_lock_head { - int dummy; -}; +struct osd_inconsistent_item { + /* link into osd_scrub::os_inconsistent_items, + * protected by osd_scrub::os_lock. */ + cfs_list_t oii_list; -#define ldiskfs_htree_lock(lock, head, inode, op) do { LBUG(); } while (0) -#define ldiskfs_htree_unlock(lock) do { LBUG(); } while (0) + /* The right FID <=> ino#/gen mapping. */ + struct osd_idmap_cache oii_cache; -static inline struct htree_lock_head *ldiskfs_htree_lock_head_alloc(int dep) -{ - LBUG(); - return NULL; -} + unsigned int oii_insert:1; /* insert or update mapping. */ +}; -#define ldiskfs_htree_lock_head_free(lh) do { LBUG(); } while (0) +struct osd_otable_cache { + struct osd_idmap_cache ooc_cache[OSD_OTABLE_IT_CACHE_SIZE]; -#define LDISKFS_DUMMY_HTREE_LOCK 0xbabecafe + /* Index for next cache slot to be filled. */ + int ooc_producer_idx; -static inline struct htree_lock *ldiskfs_htree_lock_alloc(void) -{ - return (struct htree_lock *)LDISKFS_DUMMY_HTREE_LOCK; -} + /* Index for next cache slot to be returned by it::next(). */ + int ooc_consumer_idx; -static inline void ldiskfs_htree_lock_free(struct htree_lock *lk) -{ - LASSERT((unsigned long)lk == LDISKFS_DUMMY_HTREE_LOCK); -} + /* How many items in ooc_cache. */ + int ooc_cached_items; -#define HTREE_HBITS_DEF 0 + /* Position for up layer LFSCK iteration pre-loading. */ + __u32 ooc_pos_preload; +}; -#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) +struct osd_otable_it { + struct osd_device *ooi_dev; + struct osd_otable_cache ooi_cache; + + /* For osd_otable_it_key. */ + __u8 ooi_key[16]; + + /* 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. */ +}; -#endif /* HAVE_LDISKFS_PDO */ +extern const int osd_dto_credits_noquota[]; /* * osd device. @@ -154,20 +232,16 @@ struct 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_is_md:1; /* set in ->ldo_prepare */ unsigned long od_capa_timeout; __u32 od_capa_alg; struct lustre_capa_key *od_capa_keys; @@ -179,14 +253,106 @@ struct osd_device { * statfs optimization: we cache a bit. */ cfs_time_t od_osfs_age; - cfs_kstatfs_t od_kstatfs; - cfs_spinlock_t od_osfs_lock; + struct obd_statfs od_statfs; + 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; + unsigned int od_noscrub:1, + od_handle_nolma:1; + + struct fsfilt_operations *od_fsops; + int od_connects; + struct lu_site od_site; + + struct osd_obj_map *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; + + struct mutex od_otable_mutex; + struct osd_otable_it *od_otable_it; + struct osd_scrub od_scrub; + + /* 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; +}; + +/* 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 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 + +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_QUOTA = 10, + OSD_OT_MAX = 11 +}; + +#define OSD_TRACK_DECLARES + +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]; + struct lquota_trans *ot_quota_trans; +#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 }; /* @@ -195,12 +361,20 @@ struct osd_device { #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 @@ -259,6 +433,48 @@ struct osd_it_iam { struct iam_iterator oi_it; }; +struct osd_quota_leaf { + cfs_list_t 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 */ + cfs_list_t oiq_list; +}; + +#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; + unsigned int dr_init_at; /* the line iobuf was initialized */ +}; + struct osd_thread_info { const struct lu_env *oti_env; /** @@ -272,7 +488,11 @@ struct osd_thread_info { struct htree_lock *oti_hlock; struct lu_fid oti_fid; - struct osd_inode_id oti_id; + struct lu_fid oti_fid2; + struct osd_inode_id oti_id; + struct osd_inode_id oti_id2; + struct ost_id oti_ostid; + /* * XXX temporary: for ->i_op calls. */ @@ -299,15 +519,19 @@ struct osd_thread_info { /** 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; + cfs_kstatfs_t oti_ksfs; + /** IAM iterator for index operation. */ struct iam_iterator oti_idx_it; @@ -322,6 +546,7 @@ struct osd_thread_info { long long oti_alignment_lieutenant_colonel; }; + struct osd_idmap_cache oti_cache; int oti_r_locks; int oti_w_locks; @@ -329,14 +554,41 @@ struct osd_thread_info { /** used in osd_fid_set() to put xattr */ struct lu_buf oti_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 + 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; + int oti_created[PTLRPC_MAX_BRW_PAGES]; struct lu_env oti_obj_delete_tx_env; #define OSD_FID_REC_SZ 32 char oti_ldp[OSD_FID_REC_SZ]; char oti_ldp2[OSD_FID_REC_SZ]; + + /* used by quota code */ + union { + struct if_dqblk oti_dqblk; + 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; + +#ifdef OSD_TRACK_DECLARES + /* Tracking for transaction credits, to allow debugging and optimizing + * cases where a large number of credits are being allocated for + * single transaction. */ + unsigned char oti_declare_ops[OSD_OT_MAX]; + unsigned char oti_declare_ops_rb[OSD_OT_MAX]; + unsigned short oti_declare_ops_cred[OSD_OT_MAX]; + bool oti_rollback; +#endif + }; extern int ldiskfs_pdo; @@ -346,12 +598,77 @@ extern int ldiskfs_pdo; void lprocfs_osd_init_vars(struct lprocfs_static_vars *lvars); 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_get_lma(struct osd_thread_info *info, struct inode *inode, + struct dentry *dentry, struct lustre_mdt_attrs *lma); + +int osd_obj_map_init(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, + struct thandle *th); +int osd_obj_map_delete(struct osd_thread_info *info, struct osd_device *osd, + const struct lu_fid *fid, struct thandle *th); +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, + struct thandle *th); + +void osd_scrub_file_reset(struct osd_scrub *scrub, __u8 *uuid, __u64 flags); +int osd_scrub_file_store(struct osd_scrub *scrub); +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 osd_device *dev, char *buf, int len); + +int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd, + const struct lu_fid *fid, struct lu_seq_range *range); +/* 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, bool allocated, 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, + bool is_blk, bool allocated, 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; +} /* * Invariants, assertions. @@ -379,30 +696,279 @@ static inline int osd_invariant(const struct osd_object *obj) #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_MAX_CACHE_SIZE OBD_OBJECT_EOF + +extern const struct dt_index_operations osd_otable_ops; + +static inline int osd_oi_fid2idx(struct osd_device *dev, + const struct lu_fid *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) +{ + LASSERTF(!fid_is_idif(fid), DFID"\n", PFID(fid)); + LASSERTF(!fid_is_last_id(fid), DFID"\n", PFID(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[osd_oi_fid2idx(osd, fid)]; +} + +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_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; +} + +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; + +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 + */ +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; +} + +#ifdef OSD_TRACK_DECLARES +extern int osd_trans_declare_op2rb[]; + +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); + LASSERT(op < OSD_OT_MAX); + + 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); + LASSERT(op < OSD_OT_MAX); + + 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]; + LASSERTF(rb < OSD_OT_MAX, "op = %u\n", op); + LASSERTF(oti->oti_declare_ops_rb[rb] > 0, "rb = %u\n", rb); + 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); + LASSERT(op < OSD_OT_MAX); + + oti->oti_declare_ops_rb[op]++; +} +#else +static inline void osd_trans_declare_op(const struct lu_env *env, + struct osd_thandle *oh, + unsigned int op, int credits) +{ + oh->ot_credits += credits; +} + +static inline void osd_trans_exec_op(const struct lu_env *env, + struct thandle *th, unsigned int op) +{ +} + +static inline void osd_trans_declare_rb(const struct lu_env *env, + struct thandle *th, unsigned int op) +{ +} +#endif + +/** + * Helper function to pack the fid, ldiskfs stores fid in packed format. */ -#define OSD_ROOT_SEQ 2 -static inline int osd_fid_is_root(const struct lu_fid *fid) +static inline +void osd_fid_pack(struct osd_fid_pack *pack, const struct dt_rec *fid, + struct lu_fid *befider) { - return fid_seq(fid) == OSD_ROOT_SEQ; + 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_is_igif(const struct lu_fid *fid) +static inline +int osd_fid_unpack(struct lu_fid *fid, const struct osd_fid_pack *pack) { - return fid_is_igif(fid) || osd_fid_is_root(fid); + 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; } -static inline struct osd_oi * -osd_fid2oi(struct osd_device *osd, const struct lu_fid *fid) +/** + * 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) { - if (!fid_is_norm(fid)) - return NULL; +#ifdef CONFIG_COMPAT + return is_compat_task(); +#else + return (BITS_PER_LONG == 32); +#endif +} - LASSERT(osd->od_oi_table != NULL && osd->od_oi_count >= 1); - return &osd->od_oi_table[fid->f_seq % osd->od_oi_count]; +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; } #endif /* __KERNEL__ */