X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fosd-ldiskfs%2Fosd_internal.h;h=90fa9a26b41b224ade59d1012b2708b1ff3db3bf;hp=e9dccbcd24ba364e5f86df2b8846f54f77a194d8;hb=74d0cf95c147ac8dc5a1ace9387b92711e5a6a68;hpb=b1491d26271f074dc6f99cde037403337c0b2151 diff --git a/lustre/osd-ldiskfs/osd_internal.h b/lustre/osd-ldiskfs/osd_internal.h index e9dccbc..90fa9a2 100644 --- a/lustre/osd-ldiskfs/osd_internal.h +++ b/lustre/osd-ldiskfs/osd_internal.h @@ -15,11 +15,7 @@ * * You should have received a copy of the GNU General Public License * version 2 along with this program; If not, see - * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf - * - * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, - * CA 95054 USA or visit www.sun.com if you need additional information or - * have any questions. + * http://www.gnu.org/licenses/gpl-2.0.html * * GPL HEADER END */ @@ -27,7 +23,7 @@ * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. * - * Copyright (c) 2011, 2013, Intel Corporation. + * Copyright (c) 2011, 2016, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -43,15 +39,16 @@ #ifndef _OSD_INTERNAL_H #define _OSD_INTERNAL_H -#if defined(__KERNEL__) +/* struct mutex */ +#include /* struct rw_semaphore */ #include /* struct dentry */ #include /* struct dirent64 */ #include - +#include #include #include @@ -69,6 +66,7 @@ #include "osd_quota_fmt.h" struct inode; +extern struct kmem_cache *dynlock_cachep; #define OSD_COUNTERS (0) @@ -77,9 +75,7 @@ struct inode; /* 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 +#define LDISKFS_STATE_LUSTRE_DESTROY 30 /** Enable thandle usage statistics */ #define OSD_THANDLE_STATS (0) @@ -90,6 +86,10 @@ struct inode; #define ADMIN_USR "admin_quotafile_v2.usr" #define ADMIN_GRP "admin_quotafile_v2.grp" +/* Statfs space reservation for fragmentation and local objects */ +#define OSD_STATFS_RESERVED (1ULL << 23) /* 8MB */ +#define OSD_STATFS_RESERVED_SHIFT (7) /* reserve 0.78% of all space */ + struct osd_directory { struct iam_container od_container; struct iam_descr od_descr; @@ -127,6 +127,11 @@ struct osd_object { struct osd_directory *oo_dir; /** protects inode attributes. */ spinlock_t oo_guard; + + __u32 oo_destroyed:1; + + /* the i_flags in LMA */ + __u32 oo_lma_flags; /** * Following two members are used to indicate the presence of dot and * dotdot in the given directory. This is required for interop mode @@ -139,6 +144,8 @@ struct osd_object { #ifdef CONFIG_LOCKDEP struct lockdep_map oo_dep_map; #endif + + struct list_head oo_xattr_list; }; struct osd_obj_seq { @@ -146,30 +153,30 @@ struct osd_obj_seq { 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 */ + u64 oos_seq; /* seq number */ + struct list_head oos_seq_list; /* list to seq_list */ }; 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 */ + struct list_head om_seq_list; /* list head for seq */ int om_subdir_count; - struct semaphore om_dir_init_sem; + struct mutex om_dir_init_mutex; }; struct osd_mdobj { struct dentry *om_root; /* AGENT/ */ - obd_seq om_index; /* mdt index */ - cfs_list_t om_list; /* list to omm_list */ + u64 om_index; /* mdt index */ + struct list_head om_list; /* list to omm_list */ }; struct osd_mdobj_map { struct dentry *omm_remote_parent; }; - -#define osd_ldiskfs_add_entry(handle, child, cinode, hlock) \ - ldiskfs_add_entry(handle, child, cinode, hlock) +int osd_ldiskfs_add_entry(struct osd_thread_info *info, struct osd_device *osd, + handle_t *handle, struct dentry *child, + struct inode *inode, struct htree_lock *hlock); #define OSD_OTABLE_IT_CACHE_SIZE 64 #define OSD_OTABLE_IT_CACHE_MASK (~(OSD_OTABLE_IT_CACHE_SIZE - 1)) @@ -177,7 +184,7 @@ struct osd_mdobj_map { struct osd_inconsistent_item { /* link into osd_scrub::os_inconsistent_items, * protected by osd_scrub::os_lock. */ - cfs_list_t oii_list; + struct list_head oii_list; /* The right FID <=> ino#/gen mapping. */ struct osd_idmap_cache oii_cache; @@ -217,6 +224,12 @@ struct osd_otable_it { ooi_waiting:1; /* it::next is waiting. */ }; +struct osd_obj_orphan { + struct list_head oor_list; + struct lu_env *oor_env; /* to identify "own" records */ + __u32 oor_ino; +}; + /* * osd device. */ @@ -235,24 +248,16 @@ struct osd_device { 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; + od_index_in_idif:1; - unsigned long od_capa_timeout; - __u32 od_capa_alg; - struct lustre_capa_key *od_capa_keys; - cfs_hlist_head_t *od_capa_hash; + __u32 od_dirent_journal; + int od_index; + struct proc_dir_entry *od_proc_entry; + struct lprocfs_stats *od_stats; - cfs_proc_dir_entry_t *od_proc_entry; - struct lprocfs_stats *od_stats; - /* - * statfs optimization: we cache a bit. - */ - cfs_time_t od_osfs_age; - struct obd_statfs od_statfs; spinlock_t od_osfs_lock; int od_connects; @@ -261,18 +266,18 @@ struct osd_device { 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; + 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 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; - cfs_list_t od_ios_list; + struct list_head od_ios_list; /* service name associated with the osd device */ char od_svname[MAX_OBD_NAME]; @@ -280,8 +285,37 @@ struct osd_device { /* quota slave instance */ struct qsd_instance *od_quota_slave; + + /* osd seq instance */ + struct lu_client_seq *od_cl_seq; + /* If the ratio of "the total OI mappings count" vs + * "the bad OI mappings count" is lower than the + * osd_device::od_full_scrub_ratio, then trigger + * OI scrub to scan the whole the device. */ + __u64 od_full_scrub_ratio; + /* If the speed of found bad OI mappings (per minute) + * exceeds the osd_device::od_full_scrub_threshold_rate, + * then trigger OI scrub to scan the whole device. */ + __u64 od_full_scrub_threshold_rate; + + /* a list of orphaned agent inodes, protected with od_osfs_lock */ + struct list_head od_orphan_list; +}; + +enum osd_full_scrub_ratio { + /* Trigger OI scrub to scan the whole device directly. */ + OFSR_DIRECTLY = 0, + + /* Because the bad OI mappings count cannot be larger than + * the total OI mappints count, then setting OFSR_NEVER means + * that the whole device scanning cannot be triggered by auto + * detected bad OI mappings during the RPC services. */ + OFSR_NEVER = 1, + OFSR_DEFAULT = 10000, }; +#define FULL_SCRUB_THRESHOLD_RATE_DEFAULT 60 + /* There are at most 10 uid/gids are affected in a transaction, and * that's rename case: * - 2 for source parent uid & gid; @@ -295,7 +329,7 @@ struct osd_device { */ #define OSD_MAX_UGID_CNT 10 -enum { +enum osd_op_type { OSD_OT_ATTR_SET = 0, OSD_OT_PUNCH = 1, OSD_OT_XATTR_SET = 2, @@ -306,29 +340,30 @@ enum { OSD_OT_WRITE = 7, OSD_OT_INSERT = 8, OSD_OT_DELETE = 9, - OSD_OT_UPDATE = 10, - OSD_OT_QUOTA = 11, - OSD_OT_MAX = 12 + OSD_OT_QUOTA = 10, + OSD_OT_MAX = 11 }; struct osd_thandle { struct thandle ot_super; handle_t *ot_handle; struct ldiskfs_journal_cb_entry ot_jcb; - cfs_list_t ot_dcb_list; + struct list_head ot_commit_dcb_list; + struct list_head ot_stop_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]; + unsigned int ot_credits; + unsigned short ot_id_cnt; + unsigned short ot_id_type; + unsigned int ot_remove_agents:1; + 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; + ktime_t oth_alloced; /** time when this thanle was started */ - cfs_time_t oth_started; + ktime_t oth_started; #endif }; @@ -343,7 +378,6 @@ enum dt_txn_op { 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, @@ -355,7 +389,7 @@ enum dt_txn_op { * osd dev stats */ -#ifdef LPROCFS +#ifdef CONFIG_PROC_FS enum { LPROC_OSD_READ_BYTES = 0, LPROC_OSD_WRITE_BYTES = 1, @@ -400,24 +434,25 @@ struct osd_it_ea_dirent { * there would be one ext3 readdir for every mdd readdir page. */ -#define OSD_IT_EA_BUFSIZE (PAGE_CACHE_SIZE + PAGE_CACHE_SIZE/4) +#define OSD_IT_EA_BUFSIZE (PAGE_SIZE + PAGE_SIZE/4) /** * This is iterator's in-memory data structure in interoperability * mode (i.e. iterator over ldiskfs style directory) */ struct osd_it_ea { - struct osd_object *oie_obj; - /** used in ldiskfs iterator, to stored file pointer */ - struct file oie_file; - /** how many entries have been read-cached from storage */ - int oie_rd_dirent; - /** current entry is being iterated by caller */ - int oie_it_dirent; - /** current processing entry */ - struct osd_it_ea_dirent *oie_dirent; - /** buffer to hold entries, size == OSD_IT_EA_BUFSIZE */ - void *oie_buf; + struct osd_object *oie_obj; + /** used in ldiskfs iterator, to stored file pointer */ + struct file oie_file; + /** how many entries have been read-cached from storage */ + int oie_rd_dirent; + /** current entry is being iterated by caller */ + int oie_it_dirent; + /** current processing entry */ + struct osd_it_ea_dirent *oie_dirent; + /** buffer to hold entries, size == OSD_IT_EA_BUFSIZE */ + void *oie_buf; + struct dentry oie_dentry; }; /** @@ -430,7 +465,7 @@ struct osd_it_iam { }; struct osd_quota_leaf { - cfs_list_t oql_link; + struct list_head oql_link; uint oql_blk; }; @@ -448,14 +483,14 @@ struct osd_it_quota { /** 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; + struct list_head oiq_list; }; -#define MAX_BLOCKS_PER_PAGE (PAGE_CACHE_SIZE / 512) +#define MAX_BLOCKS_PER_PAGE (PAGE_SIZE / 512) struct osd_iobuf { wait_queue_head_t dr_wait; - cfs_atomic_t dr_numreqs; /* number of reqs being processed */ + atomic_t dr_numreqs; /* number of reqs being processed */ int dr_max_pages; int dr_npages; int dr_error; @@ -466,26 +501,36 @@ struct osd_iobuf { struct lu_buf dr_pg_buf; struct page **dr_pages; struct lu_buf dr_bl_buf; - unsigned long *dr_blocks; + sector_t *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 */ }; +#define OSD_INS_CACHE_SIZE 8 + struct osd_thread_info { - const struct lu_env *oti_env; - /** - * used for index operations. - */ - struct dentry oti_obj_dentry; - struct dentry oti_child_dentry; + const struct lu_env *oti_env; + /** + * used for index operations. + */ + struct dentry oti_obj_dentry; + struct dentry oti_child_dentry; + + /** dentry for Iterator context. */ + struct dentry oti_it_dentry; - /** dentry for Iterator context. */ - struct dentry oti_it_dentry; - struct htree_lock *oti_hlock; + union { + /* fake struct file for osd_object_sync */ + struct file oti_file; + /* osd_statfs() */ + struct kstatfs oti_ksfs; + }; - struct lu_fid oti_fid; + struct htree_lock *oti_hlock; + + struct lu_fid oti_fid; struct lu_fid oti_fid2; struct lu_fid oti_fid3; struct osd_inode_id oti_id; @@ -497,19 +542,6 @@ struct osd_thread_info { * XXX temporary: for ->i_op calls. */ struct timespec oti_time; - /* - * XXX temporary: fake struct file for osd_object_sync - */ - struct file oti_file; - /* - * XXX temporary: for capa operations. - */ - struct lustre_capa_key oti_capa_key; - struct lustre_capa oti_capa; - - /** osd_device reference, initialized in osd_trans_start() and - used in osd_trans_stop() */ - struct osd_device *oti_dev; /** * following ipd and it structures are used for osd_index_iam_lookup() @@ -517,26 +549,16 @@ struct osd_thread_info { * in open iterator session. */ - /** 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; - 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; + unsigned int oti_it_ea_buf_used:1; - struct kstatfs oti_ksfs; - - /** IAM iterator for index operation. */ - struct iam_iterator oti_idx_it; + /* IAM iterator for index operation. */ + struct iam_iterator oti_idx_it; /** union to guarantee that ->oti_ipd[] has proper alignment. */ union { + char oti_name[48]; char oti_it_ipd[DX_IPD_MAX_SIZE]; long long oti_alignment_lieutenant; }; @@ -548,27 +570,38 @@ struct osd_thread_info { struct osd_idmap_cache oti_cache; + /* dedicated OI cache for insert (which needs inum) */ + struct osd_idmap_cache *oti_ins_cache; + int oti_ins_cache_size; + int oti_ins_cache_used; + 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 */ union { struct lustre_mdt_attrs oti_mdt_attrs; /* old LMA for compatibility */ char oti_mdt_attrs_old[LMA_OLD_SIZE]; + struct filter_fid_old oti_ff; + struct filter_fid oti_ff_new; }; /** 0-copy IO */ struct osd_iobuf oti_iobuf; - struct inode oti_inode; + /* used to access objects in /O */ + struct inode *oti_inode; #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 { -#ifdef HAVE_DQUOT_FS_DISK_QUOTA +#if defined(HAVE_DQUOT_QC_DQBLK) + struct qc_dqblk oti_qdq; +#elif defined(HAVE_DQUOT_FS_DISK_QUOTA) struct fs_disk_quota oti_fdq; #else struct if_dqblk oti_dqblk; @@ -584,16 +617,10 @@ struct osd_thread_info { /* 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; - }; + unsigned int oti_credits_before; + unsigned int oti_declare_ops[OSD_OT_MAX]; + unsigned int oti_declare_ops_cred[OSD_OT_MAX]; + unsigned int oti_declare_ops_used[OSD_OT_MAX]; }; extern int ldiskfs_pdo; @@ -621,27 +648,28 @@ static inline int __osd_xattr_set(struct osd_thread_info *info, return inode->i_op->setxattr(dentry, name, buf, buflen, fl); } -#ifdef LPROCFS +#ifdef CONFIG_PROC_FS /* osd_lproc.c */ extern struct lprocfs_vars lprocfs_osd_obd_vars[]; extern struct lprocfs_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_brw_stats_update(struct osd_device *osd, struct osd_iobuf *iobuf); +#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0) +int osd_register_proc_index_in_idif(struct osd_device *osd); +#endif #endif int osd_statfs(const struct lu_env *env, struct dt_device *dev, 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); +void 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); @@ -672,17 +700,17 @@ int osd_obj_spec_update(struct osd_thread_info *info, struct osd_device *osd, 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_start(struct osd_device *dev, __u32 flags); 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_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); + u64 seq, struct lu_seq_range *range); int osd_delete_from_remote_parent(const struct lu_env *env, struct osd_device *osd, @@ -710,25 +738,95 @@ loff_t find_tree_dqentry(const struct lu_env *env, 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); + 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, - bool is_blk, bool allocated, int *flags, bool force); + 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); +int osd_quota_migration(const struct lu_env *env, struct dt_object *dt); + +#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 + +#ifdef HAVE_LDISKFS_INFO_JINODE +# define osd_attach_jinode(inode) ldiskfs_inode_attach_jinode(inode) +#else /* HAVE_LDISKFS_INFO_JINODE */ +# define osd_attach_jinode(inode) 0 +#endif /* HAVE_LDISKFS_INFO_JINODE */ + +#ifdef LDISKFS_HT_MISC +# define osd_journal_start_sb(sb, type, nblock) \ + ldiskfs_journal_start_sb(sb, type, nblock) +static inline struct buffer_head *osd_ldiskfs_append(handle_t *handle, + struct inode *inode, + ldiskfs_lblk_t *nblock) +{ + int rc; + + rc = osd_attach_jinode(inode); + if (rc) + return ERR_PTR(rc); + return ldiskfs_append(handle, inode, nblock); +} +# define osd_ldiskfs_find_entry(dir, name, de, inlined, lock) \ + (__ldiskfs_find_entry(dir, name, de, inlined, lock) ?: \ + ERR_PTR(-ENOENT)) +# 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) -static inline bool is_quota_glb_feat(const struct dt_index_features *feat) +static inline struct buffer_head *osd_ldiskfs_append(handle_t *handle, + struct inode *inode, + ldiskfs_lblk_t *nblock) { - return (feat == &dt_quota_iusr_features || - feat == &dt_quota_busr_features || - feat == &dt_quota_igrp_features || - feat == &dt_quota_bgrp_features) ? true : false; + struct buffer_head *bh; + int err = 0; + + bh = ldiskfs_append(handle, inode, nblock, &err); + if (bh == NULL) + bh = ERR_PTR(err); + + return bh; } +# define osd_ldiskfs_find_entry(dir, name, de, inlined, lock) \ + (__ldiskfs_find_entry(dir, name, de, lock) ?: \ + ERR_PTR(-ENOENT)) +# 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 + /* * Invariants, assertions. */ @@ -839,7 +937,12 @@ static inline struct seq_server_site *osd_seq_site(struct osd_device *osd) static inline char *osd_name(struct osd_device *osd) { - return osd->od_dt_dev.dd_lu_dev.ld_obd->obd_name; + return osd->od_svname; +} + +static inline bool osd_is_ea_inode(struct inode *inode) +{ + return !!(LDISKFS_I(inode)->i_flags & LDISKFS_EA_INODE_FL); } extern const struct dt_body_operations osd_body_ops; @@ -878,6 +981,10 @@ static inline void osd_ipd_put(const struct lu_env *env, bag->ic_descr->id_ops->id_ipd_free(ipd); } +int osd_calc_bkmap_credits(struct super_block *sb, struct inode *inode, + const loff_t size, const loff_t pos, + const int blocks); + 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); @@ -904,20 +1011,21 @@ struct dentry *osd_child_dentry_by_inode(const struct lu_env *env, extern int osd_trans_declare_op2rb[]; extern int ldiskfs_track_declares_assert; +void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th); static inline void osd_trans_declare_op(const struct lu_env *env, struct osd_thandle *oh, - unsigned int op, int credits) + enum osd_op_type 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)) + if (unlikely(ldiskfs_track_declares_assert)) { LASSERT(op < OSD_OT_MAX); - else { + } else { CWARN("%s: Invalid operation index %d\n", - osd_name(oti->oti_dev), op); + osd_name(osd_dt_dev(oh->ot_super.th_dev)), op); libcfs_debug_dumpstack(NULL); } } else { @@ -928,77 +1036,119 @@ static inline void osd_trans_declare_op(const struct lu_env *env, } static inline void osd_trans_exec_op(const struct lu_env *env, - struct thandle *th, unsigned int op) + struct thandle *th, + enum osd_op_type 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; + unsigned int rb, left; 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); + CWARN("%s: opcode %u: invalid value >= %u\n", + osd_name(osd_dt_dev(oh->ot_super.th_dev)), + op, OSD_OT_MAX); 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]--; + /* find rollback (or reverse) operation for the given one + * such an operation doesn't require additional credits + * as the same set of blocks are modified */ + rb = osd_trans_declare_op2rb[op]; + + /* check whether credits for this operation were reserved at all */ + if (unlikely(oti->oti_declare_ops_cred[op] == 0 && + oti->oti_declare_ops_cred[rb] == 0)) { + /* the API is not perfect yet: CREATE does REF_ADD internally + * while DESTROY does not. To rollback CREATE the callers + * needs to call REF_DEL+DESTROY which is hard to detect using + * a simple table of rollback operations */ + if (op == OSD_OT_REF_DEL && + oti->oti_declare_ops_cred[OSD_OT_CREATE] > 0) + goto proceed; + if (op == OSD_OT_REF_ADD && + oti->oti_declare_ops_cred[OSD_OT_DESTROY] > 0) + goto proceed; + CWARN("%s: opcode %u: credits = 0, rollback = %u\n", + osd_name(osd_dt_dev(oh->ot_super.th_dev)), op, rb); + osd_trans_dump_creds(env, th); + LASSERT(!ldiskfs_track_declares_assert); + } + +proceed: + /* remember how many credits we have unused before the operation */ + oti->oti_credits_before = oh->ot_handle->h_buffer_credits; + left = oti->oti_declare_ops_cred[op] - oti->oti_declare_ops_used[op]; + if (unlikely(oti->oti_credits_before < left)) { + CWARN("%s: opcode %u: before %u < left %u, rollback = %u\n", + osd_name(osd_dt_dev(oh->ot_super.th_dev)), op, + oti->oti_credits_before, left, rb); + osd_trans_dump_creds(env, th); + /* on a very small fs (testing?) it's possible that + * the transaction can't fit 1/4 of journal, so we + * just request less credits (see osd_trans_start()). + * ignore the same case here */ + rb = osd_transaction_size(osd_dt_dev(th->th_dev)); + if (unlikely(oh->ot_credits < rb)) + LASSERT(!ldiskfs_track_declares_assert); } } -static inline void osd_trans_declare_rb(const struct lu_env *env, - struct thandle *th, unsigned int op) +static inline void osd_trans_exec_check(const struct lu_env *env, + struct thandle *th, + enum osd_op_type op) { struct osd_thread_info *oti = osd_oti_get(env); struct osd_thandle *oh = container_of(th, struct osd_thandle, ot_super); + int used, over, quota; - 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); - } + /* how many credits have been used by the operation */ + used = oti->oti_credits_before - oh->ot_handle->h_buffer_credits; + + if (unlikely(used < 0)) { + /* if some block was allocated and released in the same + * transaction, then it won't be a part of the transaction + * and delta can be negative */ + return; + } + + if (used == 0) { + /* rollback operations (e.g. when we destroy just created + * object) should not consume any credits. there is no point + * to confuse the checks below */ + return; + } + oti->oti_declare_ops_used[op] += used; + if (oti->oti_declare_ops_used[op] <= oti->oti_declare_ops_cred[op]) + return; + + /* we account quota for a whole transaction and any operation can + * consume corresponding credits */ + over = oti->oti_declare_ops_used[op] - + oti->oti_declare_ops_cred[op]; + quota = oti->oti_declare_ops_cred[OSD_OT_QUOTA] - + oti->oti_declare_ops_used[OSD_OT_QUOTA]; + if (over <= quota) { + /* probably that credits were consumed by + * quota indirectly (in the depths of ldiskfs) */ + oti->oti_declare_ops_used[OSD_OT_QUOTA] += over; + oti->oti_declare_ops_used[op] -= over; } else { - oti->oti_declare_ops_rb[op]++; + CWARN("%s: opcode %d: used %u, used now %u, reserved %u\n", + osd_name(osd_dt_dev(oh->ot_super.th_dev)), op, + oti->oti_declare_ops_used[op], used, + oti->oti_declare_ops_cred[op]); + osd_trans_dump_creds(env, th); + if (unlikely(ldiskfs_track_declares_assert)) + LBUG(); } } @@ -1043,7 +1193,7 @@ int osd_acct_obj_lookup(struct osd_thread_info *info, struct osd_device *osd, static inline int is_32bit_api(void) { #ifdef CONFIG_COMPAT - return is_compat_task(); + return in_compat_syscall(); #else return (BITS_PER_LONG == 32); #endif @@ -1063,29 +1213,44 @@ static inline int fid_is_internal(const struct lu_fid *fid) return (!fid_is_namespace_visible(fid) && !fid_is_idif(fid)); } -#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); +static inline unsigned long osd_remote_parent_ino(struct osd_device *dev) +{ + return dev->od_mdt_map->omm_remote_parent->d_inode->i_ino; +} + +/** + * ext4_bread/ldiskfs_bread has either 5 or 4 parameters. The error + * return code has been removed and integrated into the pointer in the + * kernel 3.18. + */ +static inline struct buffer_head *__ldiskfs_bread(handle_t *handle, + struct inode *inode, + ldiskfs_lblk_t block, + int create) +{ + int rc = 0; + struct buffer_head *bh; + + if (create) { + rc = osd_attach_jinode(inode); + if (rc) + return ERR_PTR(rc); + } +#ifdef HAVE_EXT4_BREAD_4ARGS + bh = ldiskfs_bread(handle, inode, block, create); #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); + + bh = ldiskfs_bread(handle, inode, block, create, &rc); + if (bh == NULL && rc != 0) + bh = ERR_PTR(rc); #endif + return bh; +} void ldiskfs_inc_count(handle_t *handle, struct inode *inode); void ldiskfs_dec_count(handle_t *handle, struct inode *inode); -#endif /* __KERNEL__ */ +void osd_fini_iobuf(struct osd_device *d, struct osd_iobuf *iobuf); + + #endif /* _OSD_INTERNAL_H */