X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fosd-ldiskfs%2Fosd_internal.h;h=7e573d1e2c6f5485de026d60442ff7605dcc28d0;hp=30f44f3693c1793966d82c8ec0999cf4075090a9;hb=0081d9de639979c0d0e341ad6ded442b9043a379;hpb=8b3a9994e41b24e20ca839388e0ea47fbab06c53 diff --git a/lustre/osd-ldiskfs/osd_internal.h b/lustre/osd-ldiskfs/osd_internal.h index 30f44f3..7e573d1 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. @@ -28,6 +26,8 @@ /* * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. + * + * Copyright (c) 2011, 2012, Whamcloud, Inc. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -52,33 +52,38 @@ /* struct dirent64 */ #include -#ifdef HAVE_EXT4_LDISKFS #include #include -#define osd_journal_callback_set(handle, func, jcb) jbd2_journal_callback_set(handle, func, jcb) +#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 "osd_oi.h" #include "osd_iam.h" +#include "osd_scrub.h" struct inode; -#define OSD_OII_NOGEN (0) #define OSD_COUNTERS (0) +/* Lustre special inode::i_state to indicate OI scrub skip this inode. */ +#define I_LUSTRE_NOSCRUB (1 << 31) + /** Enable thandle usage statistics */ #define OSD_THANDLE_STATS (0) @@ -86,10 +91,161 @@ struct inode; struct osd_ctxt { __u32 oc_uid; __u32 oc_gid; - __u32 oc_cap; + cfs_kernel_cap_t oc_cap; }; #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) \ + 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; +}; + +struct htree_lock_head { + int dummy; +}; + +#define ldiskfs_htree_lock(lock, head, inode, op) do { LBUG(); } while (0) +#define ldiskfs_htree_unlock(lock) do { LBUG(); } while (0) + +static inline struct htree_lock_head *ldiskfs_htree_lock_head_alloc(int dep) +{ + LBUG(); + return NULL; +} + +#define ldiskfs_htree_lock_head_free(lh) do { LBUG(); } while (0) + +#define LDISKFS_DUMMY_HTREE_LOCK 0xbabecafe + +static inline struct htree_lock *ldiskfs_htree_lock_alloc(void) +{ + return (struct htree_lock *)LDISKFS_DUMMY_HTREE_LOCK; +} + +static inline void ldiskfs_htree_lock_free(struct htree_lock *lk) +{ + LASSERT((unsigned long)lk == LDISKFS_DUMMY_HTREE_LOCK); +} + +#define HTREE_HBITS_DEF 0 + +#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) + +#endif /* HAVE_LDISKFS_PDO */ + +#define OSD_OTABLE_IT_CACHE_SIZE 128 +#define OSD_OTABLE_IT_CACHE_MASK (~(OSD_OTABLE_IT_CACHE_SIZE - 1)) + +struct osd_inconsistent_item { + /* link into osd_scrub::os_inconsistent_items, + * protected by osd_scrub::os_lock. */ + cfs_list_t oii_list; + + /* The right FID <=> ino#/gen mapping. */ + struct osd_idmap_cache oii_cache; + + unsigned int oii_insert:1; /* insert or update mapping. */ +}; + +struct osd_otable_cache { + struct osd_idmap_cache ooc_cache[OSD_OTABLE_IT_CACHE_SIZE]; + + /* Index for next cache slot to be filled. */ + int ooc_producer_idx; + + /* Index for next cache slot to be returned by it::next(). */ + int ooc_consumer_idx; + + /* How many items in ooc_cache. */ + int ooc_cached_items; + + /* 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; + + /* 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. */ +}; + +extern const int osd_dto_credits_noquota[]; + /* * osd device. */ @@ -98,22 +254,11 @@ struct osd_device { struct dt_device od_dt_dev; /* information about underlying file system */ struct lustre_mount_info *od_mount; + struct vfsmount *od_mnt; /* object index */ - struct osd_oi od_oi; - /* - * XXX temporary stuff for object index: directory where every object - * is named by its fid. - */ - struct dt_object *od_obj_area; - - /* Environment for transaction commit callback. - * Currently, OSD is based on ext3/JBD. Transaction commit in ext3/JBD - * is serialized, that is there is no more than one transaction commit - * at a time (JBD journal_commit_transaction() is serialized). - * This means that it's enough to have _one_ lu_context. - */ - struct lu_env od_env_for_commit; - + struct osd_oi **od_oi_table; + /* total number of OI containers */ + int od_oi_count; /* * Fid Capability */ @@ -129,7 +274,7 @@ struct osd_device { * 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; /** @@ -137,6 +282,109 @@ struct osd_device { * 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; + + cfs_mutex_t od_otable_mutex; + struct osd_otable_it *od_otable_it; + struct osd_scrub od_scrub; +}; + +#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 }; /* @@ -145,12 +393,20 @@ 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 @@ -209,6 +465,25 @@ struct osd_it_iam { 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; /** @@ -219,15 +494,18 @@ struct osd_thread_info { /** dentry for Iterator context. */ struct dentry oti_it_dentry; + 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. */ - struct txn_param oti_txn; struct timespec oti_time; - struct timespec oti_time2; /* * XXX temporary: fake struct file for osd_object_sync */ @@ -259,6 +537,8 @@ struct osd_thread_info { /** 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; @@ -273,6 +553,7 @@ struct osd_thread_info { long long oti_alignment_lieutenant_colonel; }; + struct osd_idmap_cache oti_cache; int oti_r_locks; int oti_w_locks; @@ -281,6 +562,10 @@ struct osd_thread_info { 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 @@ -290,6 +575,8 @@ struct osd_thread_info { char oti_ldp2[OSD_FID_REC_SZ]; }; +extern int ldiskfs_pdo; + #ifdef LPROCFS /* osd_lproc.c */ void lprocfs_osd_init_vars(struct lprocfs_static_vars *lvars); @@ -298,9 +585,49 @@ 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); +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, + struct osd_inode_id *id); +struct inode *osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev, + struct osd_inode_id *id, struct lu_fid *fid); + +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); + +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); /* * Invariants, assertions. @@ -328,20 +655,168 @@ 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. +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_igif(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_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 + */ +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, + 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; +} + +/** + * 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; } #endif /* __KERNEL__ */