Whamcloud - gitweb
LU-9019 osd: migrate osd-ldiskfs thandle stats to 64 bit time
[fs/lustre-release.git] / lustre / osd-ldiskfs / osd_internal.h
index b47f653..90fa9a2 100644 (file)
  *
  * 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, 2015, Intel Corporation.
+ * Copyright (c) 2011, 2016, Intel Corporation.
  */
 /*
  * This file is part of Lustre, http://www.lustre.org/
@@ -90,6 +86,10 @@ extern struct kmem_cache *dynlock_cachep;
 #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;
@@ -129,6 +129,9 @@ struct osd_object {
        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
@@ -141,6 +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 {
@@ -169,9 +174,9 @@ struct osd_mdobj {
 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))
@@ -324,7 +329,7 @@ enum osd_full_scrub_ratio {
  */
 #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,
@@ -347,18 +352,18 @@ struct osd_thandle {
        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;
-       int                     ot_remove_agents:1;
-        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
 };
 
@@ -429,7 +434,7 @@ 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
@@ -481,7 +486,7 @@ struct osd_it_quota {
        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;
@@ -503,6 +508,8 @@ struct osd_iobuf {
        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;
        /**
@@ -536,10 +543,6 @@ struct osd_thread_info {
          */
         struct timespec        oti_time;
 
-        /** 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()
          * these are defined separately as we might do index operation
@@ -567,6 +570,11 @@ 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;
@@ -591,7 +599,9 @@ struct osd_thread_info {
 
        /* 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;
@@ -608,9 +618,9 @@ struct osd_thread_info {
         * cases where a large number of credits are being allocated for
         * single transaction. */
        unsigned int            oti_credits_before;
-       unsigned short          oti_declare_ops[OSD_OT_MAX];
-       unsigned short          oti_declare_ops_cred[OSD_OT_MAX];
-       unsigned short          oti_declare_ops_used[OSD_OT_MAX];
+       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;
@@ -762,13 +772,29 @@ static inline void i_gid_write(struct inode *inode, gid_t 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)
-# define osd_ldiskfs_append(handle, inode, nblock) \
-               ldiskfs_append(handle, inode, 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)
+               (__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) \
@@ -793,7 +819,8 @@ static inline struct buffer_head *osd_ldiskfs_append(handle_t *handle,
 }
 
 # define osd_ldiskfs_find_entry(dir, name, de, inlined, lock) \
-               __ldiskfs_find_entry(dir, name, de, 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) \
@@ -910,7 +937,7 @@ 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)
@@ -988,7 +1015,7 @@ 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);
 
@@ -1009,7 +1036,8 @@ 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,
@@ -1021,8 +1049,9 @@ static inline void osd_trans_exec_op(const struct lu_env *env,
                if (unlikely(ldiskfs_track_declares_assert))
                        LASSERT(op < OSD_OT_MAX);
                else {
-                       CWARN("%s: Invalid operation index %d\n",
-                             osd_name(osd_dt_dev(oh->ot_super.th_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;
                }
@@ -1046,11 +1075,10 @@ static inline void osd_trans_exec_op(const struct lu_env *env,
                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);
-               CERROR("%s: op = %d, rb = %d\n",
-                      osd_name(osd_dt_dev(oh->ot_super.th_dev)), op, rb);
-               if (unlikely(ldiskfs_track_declares_assert))
-                       LBUG();
+               LASSERT(!ldiskfs_track_declares_assert);
        }
 
 proceed:
@@ -1058,24 +1086,23 @@ proceed:
        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);
-               CERROR("%s: op = %d, rb = %d\n",
-                      osd_name(osd_dt_dev(oh->ot_super.th_dev)), op, rb);
                /* 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)) {
-                       if (unlikely(ldiskfs_track_declares_assert))
-                               LBUG();
-               }
+               if (unlikely(oh->ot_credits < rb))
+                       LASSERT(!ldiskfs_track_declares_assert);
        }
 }
 
 static inline void osd_trans_exec_check(const struct lu_env *env,
                                        struct thandle *th,
-                                       unsigned int op)
+                                       enum osd_op_type op)
 {
        struct osd_thread_info *oti = osd_oti_get(env);
        struct osd_thandle     *oh  = container_of(th, struct osd_thandle,
@@ -1115,8 +1142,9 @@ static inline void osd_trans_exec_check(const struct lu_env *env,
                oti->oti_declare_ops_used[OSD_OT_QUOTA] += over;
                oti->oti_declare_ops_used[op] -= over;
        } else {
-               CWARN("op %d: used %u, used now %u, reserved %u\n",
-                     op, oti->oti_declare_ops_used[op], used,
+               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))
@@ -1165,7 +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
@@ -1190,9 +1218,39 @@ 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
+
+       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);
 
 void osd_fini_iobuf(struct osd_device *d, struct osd_iobuf *iobuf);
 
+
 #endif /* _OSD_INTERNAL_H */