#include "mdt_internal.h"
static int mdt_server_data_update(const struct lu_env *env,
- struct mdt_device *mdt,
- int need_sync);
+ struct mdt_device *mdt);
struct lu_buf *mdt_buf(const struct lu_env *env, void *area, ssize_t len)
{
return buf;
}
-int mdt_record_read(const struct lu_env *env,
- struct dt_object *dt, struct lu_buf *buf, loff_t *pos)
-{
- int rc;
-
- LASSERTF(dt != NULL, "dt is NULL when we want to read record\n");
-
- rc = dt->do_body_ops->dbo_read(env, dt, buf, pos, BYPASS_CAPA);
-
- if (rc == buf->lb_len)
- rc = 0;
- else if (rc >= 0)
- rc = -EFAULT;
- return rc;
-}
-
-int mdt_record_write(const struct lu_env *env,
- struct dt_object *dt, const struct lu_buf *buf,
- loff_t *pos, struct thandle *th)
-{
- int rc;
-
- LASSERTF(dt != NULL, "dt is NULL when we want to write record\n");
- LASSERT(th != NULL);
- rc = dt->do_body_ops->dbo_write(env, dt, buf, pos, th, BYPASS_CAPA, 1);
- if (rc == buf->lb_len)
- rc = 0;
- else if (rc >= 0)
- rc = -EFAULT;
- return rc;
-}
-
static inline int mdt_trans_credit_get(const struct lu_env *env,
struct mdt_device *mdt,
enum mdt_txn_op op)
mdt->mdt_bottom->dd_ops->dt_trans_stop(env, th);
}
-/* last_rcvd handling */
-static inline void lsd_le_to_cpu(struct lr_server_data *buf,
- struct lr_server_data *lsd)
-{
- memcpy(lsd->lsd_uuid, buf->lsd_uuid, sizeof (lsd->lsd_uuid));
- lsd->lsd_last_transno = le64_to_cpu(buf->lsd_last_transno);
- lsd->lsd_mount_count = le64_to_cpu(buf->lsd_mount_count);
- lsd->lsd_feature_compat = le32_to_cpu(buf->lsd_feature_compat);
- lsd->lsd_feature_rocompat = le32_to_cpu(buf->lsd_feature_rocompat);
- lsd->lsd_feature_incompat = le32_to_cpu(buf->lsd_feature_incompat);
- lsd->lsd_server_size = le32_to_cpu(buf->lsd_server_size);
- lsd->lsd_client_start = le32_to_cpu(buf->lsd_client_start);
- lsd->lsd_client_size = le16_to_cpu(buf->lsd_client_size);
-}
-
-static inline void lsd_cpu_to_le(struct lr_server_data *lsd,
- struct lr_server_data *buf)
-{
- memcpy(buf->lsd_uuid, lsd->lsd_uuid, sizeof (lsd->lsd_uuid));
- buf->lsd_last_transno = cpu_to_le64(lsd->lsd_last_transno);
- buf->lsd_mount_count = cpu_to_le64(lsd->lsd_mount_count);
- buf->lsd_feature_compat = cpu_to_le32(lsd->lsd_feature_compat);
- buf->lsd_feature_rocompat = cpu_to_le32(lsd->lsd_feature_rocompat);
- buf->lsd_feature_incompat = cpu_to_le32(lsd->lsd_feature_incompat);
- buf->lsd_server_size = cpu_to_le32(lsd->lsd_server_size);
- buf->lsd_client_start = cpu_to_le32(lsd->lsd_client_start);
- buf->lsd_client_size = cpu_to_le16(lsd->lsd_client_size);
-}
-
-static inline void lcd_le_to_cpu(struct lsd_client_data *buf,
- struct lsd_client_data *lcd)
-{
- memcpy(lcd->lcd_uuid, buf->lcd_uuid, sizeof (lcd->lcd_uuid));
- lcd->lcd_last_transno = le64_to_cpu(buf->lcd_last_transno);
- lcd->lcd_last_xid = le64_to_cpu(buf->lcd_last_xid);
- lcd->lcd_last_result = le32_to_cpu(buf->lcd_last_result);
- lcd->lcd_last_data = le32_to_cpu(buf->lcd_last_data);
- lcd->lcd_last_close_transno = le64_to_cpu(buf->lcd_last_close_transno);
- lcd->lcd_last_close_xid = le64_to_cpu(buf->lcd_last_close_xid);
- lcd->lcd_last_close_result = le32_to_cpu(buf->lcd_last_close_result);
-}
-
-static inline void lcd_cpu_to_le(struct lsd_client_data *lcd,
- struct lsd_client_data *buf)
-{
- memcpy(buf->lcd_uuid, lcd->lcd_uuid, sizeof (lcd->lcd_uuid));
- buf->lcd_last_transno = cpu_to_le64(lcd->lcd_last_transno);
- buf->lcd_last_xid = cpu_to_le64(lcd->lcd_last_xid);
- buf->lcd_last_result = cpu_to_le32(lcd->lcd_last_result);
- buf->lcd_last_data = cpu_to_le32(lcd->lcd_last_data);
- buf->lcd_last_close_transno = cpu_to_le64(lcd->lcd_last_close_transno);
- buf->lcd_last_close_xid = cpu_to_le64(lcd->lcd_last_close_xid);
- buf->lcd_last_close_result = cpu_to_le32(lcd->lcd_last_close_result);
-}
-
static inline int mdt_last_rcvd_header_read(const struct lu_env *env,
struct mdt_device *mdt)
{
mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
mti->mti_off = 0;
- rc = mdt_record_read(env, mdt->mdt_last_rcvd,
- mdt_buf(env, &mti->mti_lsd, sizeof(mti->mti_lsd)),
- &mti->mti_off);
+ rc = dt_record_read(env, mdt->mdt_last_rcvd,
+ mdt_buf(env, &mti->mti_lsd, sizeof(mti->mti_lsd)),
+ &mti->mti_off);
if (rc == 0)
lsd_le_to_cpu(&mti->mti_lsd, &mdt->mdt_lsd);
- CDEBUG(D_INFO, "read last_rcvd header rc = %d:\n"
- "uuid = %s\n"
- "last_transno = "LPU64"\n",
- rc, mdt->mdt_lsd.lsd_uuid,
- mdt->mdt_lsd.lsd_last_transno);
+ CDEBUG(D_INFO, "read last_rcvd header rc = %d, uuid = %s, "
+ "last_transno = "LPU64"\n", rc, mdt->mdt_lsd.lsd_uuid,
+ mdt->mdt_lsd.lsd_last_transno);
return rc;
}
-static void mdt_client_cb(const struct mdt_device *mdt, __u64 transno,
- void *data, int err)
-{
- struct obd_device *obd = mdt2obd_dev(mdt);
- target_client_add_cb(obd, transno, data, err);
-}
-
-static inline int mdt_last_rcvd_header_write(const struct lu_env *env,
- struct mdt_device *mdt,
- int need_sync)
+static int mdt_last_rcvd_header_write(const struct lu_env *env,
+ struct mdt_device *mdt,
+ struct thandle *th)
{
struct mdt_thread_info *mti;
- struct thandle *th;
int rc;
ENTRY;
mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
- if (mti->mti_exp) {
- spin_lock(&mti->mti_exp->exp_lock);
- mti->mti_exp->exp_need_sync = need_sync;
- spin_unlock(&mti->mti_exp->exp_lock);
- }
- mdt_trans_credit_init(env, mdt, MDT_TXN_LAST_RCVD_WRITE_OP);
- th = mdt_trans_start(env, mdt);
- if (IS_ERR(th))
- RETURN(PTR_ERR(th));
-
mti->mti_off = 0;
lsd_cpu_to_le(&mdt->mdt_lsd, &mti->mti_lsd);
- if (need_sync && mti->mti_exp)
- mdt_trans_add_cb(th, mdt_client_cb, mti->mti_exp);
+ rc = dt_record_write(env, mdt->mdt_last_rcvd,
+ mdt_buf_const(env, &mti->mti_lsd,
+ sizeof(mti->mti_lsd)),
+ &mti->mti_off, th);
- rc = mdt_record_write(env, mdt->mdt_last_rcvd,
- mdt_buf_const(env, &mti->mti_lsd,
- sizeof(mti->mti_lsd)),
- &mti->mti_off, th);
-
- mdt_trans_stop(env, mdt, th);
-
- CDEBUG(D_INFO, "write last_rcvd header rc = %d:\n"
- "uuid = %s\nlast_transno = "LPU64"\n",
- rc, mdt->mdt_lsd.lsd_uuid, mdt->mdt_lsd.lsd_last_transno);
+ CDEBUG(D_INFO, "write last_rcvd header rc = %d, uuid = %s, "
+ "last_transno = "LPU64"\n", rc, mdt->mdt_lsd.lsd_uuid,
+ mdt->mdt_lsd.lsd_last_transno);
RETURN(rc);
}
-static int mdt_last_rcvd_read(const struct lu_env *env,
- struct mdt_device *mdt,
- struct lsd_client_data *lcd, loff_t *off)
+static int mdt_last_rcvd_read(const struct lu_env *env, struct mdt_device *mdt,
+ struct lsd_client_data *lcd, loff_t *off,
+ int index)
{
struct mdt_thread_info *mti;
struct lsd_client_data *tmp;
mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
tmp = &mti->mti_lcd;
- rc = mdt_record_read(env, mdt->mdt_last_rcvd,
- mdt_buf(env, tmp, sizeof(*tmp)), off);
- if (rc == 0)
+ rc = dt_record_read(env, mdt->mdt_last_rcvd,
+ mdt_buf(env, tmp, sizeof(*tmp)), off);
+ if (rc == 0) {
+ check_lcd(mdt2obd_dev(mdt)->obd_name, index, tmp);
lcd_le_to_cpu(tmp, lcd);
+ }
- CDEBUG(D_INFO, "read lcd @%d rc = %d:\n"
- "uuid = %s\n"
- "last_transno = "LPU64"\n"
- "last_xid = "LPU64"\n"
- "last_result = %u\n"
- "last_data = %u\n"
- "last_close_transno = "LPU64"\n"
- "last_close_xid = "LPU64"\n"
- "last_close_result = %u\n",
- (int)(*off - sizeof(*tmp)),
- rc,
- lcd->lcd_uuid,
- lcd->lcd_last_transno,
- lcd->lcd_last_xid,
- lcd->lcd_last_result,
- lcd->lcd_last_data,
- lcd->lcd_last_close_transno,
- lcd->lcd_last_close_xid,
- lcd->lcd_last_close_result);
+ CDEBUG(D_INFO, "read lcd @%d rc = %d, uuid = %s, last_transno = "LPU64
+ ", last_xid = "LPU64", last_result = %u, last_data = %u, "
+ "last_close_transno = "LPU64", last_close_xid = "LPU64", "
+ "last_close_result = %u\n", (int)(*off - sizeof(*tmp)),
+ rc, lcd->lcd_uuid, lcd->lcd_last_transno, lcd->lcd_last_xid,
+ lcd->lcd_last_result, lcd->lcd_last_data,
+ lcd->lcd_last_close_transno, lcd->lcd_last_close_xid,
+ lcd->lcd_last_close_result);
return rc;
}
lcd_cpu_to_le(lcd, tmp);
- rc = mdt_record_write(env, mdt->mdt_last_rcvd,
- mdt_buf_const(env, tmp, sizeof(*tmp)), off, th);
-
- CDEBUG(D_INFO, "write lcd @%d rc = %d:\n"
- "uuid = %s\n"
- "last_transno = "LPU64"\n"
- "last_xid = "LPU64"\n"
- "last_result = %u\n"
- "last_data = %u\n"
- "last_close_transno = "LPU64"\n"
- "last_close_xid = "LPU64"\n"
- "last_close_result = %u\n",
- (int)(*off - sizeof(*tmp)),
- rc,
- lcd->lcd_uuid,
- lcd->lcd_last_transno,
- lcd->lcd_last_xid,
- lcd->lcd_last_result,
- lcd->lcd_last_data,
- lcd->lcd_last_close_transno,
- lcd->lcd_last_close_xid,
- lcd->lcd_last_close_result);
+ rc = dt_record_write(env, mdt->mdt_last_rcvd,
+ mdt_buf_const(env, tmp, sizeof(*tmp)), off, th);
+
+ CDEBUG(D_INFO, "write lcd @%d rc = %d, uuid = %s, last_transno = "LPU64
+ ", last_xid = "LPU64", last_result = %u, last_data = %u, "
+ "last_close_transno = "LPU64", last_close_xid = "LPU64" ,"
+ "last_close_result = %u\n", (int)(*off - sizeof(*tmp)),
+ rc, lcd->lcd_uuid, lcd->lcd_last_transno, lcd->lcd_last_xid,
+ lcd->lcd_last_result, lcd->lcd_last_data,
+ lcd->lcd_last_close_transno, lcd->lcd_last_close_xid,
+ lcd->lcd_last_close_result);
return rc;
}
/* When we do a clean MDS shutdown, we save the last_transno into
* the header. If we find clients with higher last_transno values
* then those clients may need recovery done. */
- LASSERT(atomic_read(&obd->obd_req_replay_clients) == 0);
+ LASSERT(cfs_atomic_read(&obd->obd_req_replay_clients) == 0);
for (cl_idx = 0, off = lsd->lsd_client_start;
off < last_size; cl_idx++) {
__u64 last_transno;
off = lsd->lsd_client_start +
cl_idx * lsd->lsd_client_size;
- rc = mdt_last_rcvd_read(env, mdt, lcd, &off);
+ rc = mdt_last_rcvd_read(env, mdt, lcd, &off, cl_idx);
if (rc) {
CERROR("error reading MDS %s idx %d, off %llu: rc %d\n",
LAST_RCVD, cl_idx, off, rc);
if (PTR_ERR(exp) == -EALREADY) {
/* export already exists, zero out this one */
lcd->lcd_uuid[0] = '\0';
- } else
+ } else {
GOTO(err_client, rc = PTR_ERR(exp));
+ }
} else {
struct mdt_thread_info *mti;
mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
rc = mdt_client_add(env, mdt, cl_idx);
/* can't fail existing */
LASSERTF(rc == 0, "rc = %d\n", rc);
+ /* VBR: set export last committed version */
+ exp->exp_last_committed = last_transno;
lcd = NULL;
- spin_lock(&exp->exp_lock);
+ cfs_spin_lock(&exp->exp_lock);
exp->exp_connecting = 0;
exp->exp_in_recovery = 0;
- spin_unlock(&exp->exp_lock);
+ cfs_spin_unlock(&exp->exp_lock);
obd->obd_max_recoverable_clients++;
class_export_put(exp);
}
CDEBUG(D_OTHER, "client at idx %d has last_transno="LPU64"\n",
cl_idx, last_transno);
/* protect __u64 value update */
- spin_lock(&mdt->mdt_transno_lock);
+ cfs_spin_lock(&mdt->mdt_transno_lock);
mdt->mdt_last_transno = max(last_transno,
mdt->mdt_last_transno);
- spin_unlock(&mdt->mdt_transno_lock);
+ cfs_spin_unlock(&mdt->mdt_transno_lock);
}
err_client:
lsd->lsd_server_size = LR_SERVER_SIZE;
lsd->lsd_client_start = LR_CLIENT_START;
lsd->lsd_client_size = LR_CLIENT_SIZE;
+ lsd->lsd_feature_compat = OBD_COMPAT_MDT;
lsd->lsd_feature_rocompat = OBD_ROCOMPAT_LOVOBJID;
lsd->lsd_feature_incompat = OBD_INCOMPAT_MDT |
- OBD_INCOMPAT_COMMON_LR;
+ OBD_INCOMPAT_COMMON_LR;
} else {
LCONSOLE_WARN("%s: used disk, loading\n", obd->obd_name);
rc = mdt_last_rcvd_header_read(env, mdt);
obd->obd_uuid.uuid, lsd->lsd_uuid);
GOTO(out, rc = -EINVAL);
}
+ lsd->lsd_feature_compat |= OBD_COMPAT_MDT;
+ lsd->lsd_feature_incompat |= OBD_INCOMPAT_MDT |
+ OBD_INCOMPAT_COMMON_LR;
}
mount_count = lsd->lsd_mount_count;
ldd = lsi->lsi_ldd;
+ if (lsd->lsd_feature_incompat & ~MDT_INCOMPAT_SUPP) {
+ CERROR("%s: unsupported incompat filesystem feature(s) %x\n",
+ obd->obd_name,
+ lsd->lsd_feature_incompat & ~MDT_INCOMPAT_SUPP);
+ GOTO(out, rc = -EINVAL);
+ }
+ if (lsd->lsd_feature_rocompat & ~MDT_ROCOMPAT_SUPP) {
+ CERROR("%s: unsupported read-only filesystem feature(s) %x\n",
+ obd->obd_name,
+ lsd->lsd_feature_rocompat & ~MDT_ROCOMPAT_SUPP);
+ /* XXX: Do something like remount filesystem read-only */
+ GOTO(out, rc = -EINVAL);
+ }
+ /** Interop: evict all clients at first boot with 1.8 last_rcvd */
+ if (!(lsd->lsd_feature_compat & OBD_COMPAT_20)) {
+ LCONSOLE_WARN("Mounting %s at first time on 1.8 FS, remove all"
+ " clients for interop needs\n", obd->obd_name);
+ simple_truncate(lsi->lsi_srv_mnt->mnt_sb->s_root,
+ lsi->lsi_srv_mnt, LAST_RCVD,
+ lsd->lsd_client_start);
+ last_rcvd_size = lsd->lsd_client_start;
+ /** set 2.0 flag to upgrade/downgrade between 1.8 and 2.0 */
+ lsd->lsd_feature_compat |= OBD_COMPAT_20;
+ }
+
if (ldd->ldd_flags & LDD_F_IAM_DIR)
lsd->lsd_feature_incompat |= OBD_INCOMPAT_IAM_DIR;
- lsd->lsd_feature_compat = OBD_COMPAT_MDT;
lsd->lsd_feature_incompat |= OBD_INCOMPAT_FID;
- spin_lock(&mdt->mdt_transno_lock);
+ cfs_spin_lock(&mdt->mdt_transno_lock);
mdt->mdt_last_transno = lsd->lsd_last_transno;
- spin_unlock(&mdt->mdt_transno_lock);
+ cfs_spin_unlock(&mdt->mdt_transno_lock);
CDEBUG(D_INODE, "========BEGIN DUMPING LAST_RCVD========\n");
CDEBUG(D_INODE, "%s: server last_transno: "LPU64"\n",
if (rc)
GOTO(err_client, rc);
- spin_lock(&mdt->mdt_transno_lock);
+ cfs_spin_lock(&mdt->mdt_transno_lock);
/* obd_last_committed is used for compatibility
* with other lustre recovery code */
obd->obd_last_committed = mdt->mdt_last_transno;
- spin_unlock(&mdt->mdt_transno_lock);
+ cfs_spin_unlock(&mdt->mdt_transno_lock);
- mdt->mdt_mount_count++;
+ mdt->mdt_mount_count = mount_count + 1;
lsd->lsd_mount_count = mdt->mdt_mount_count;
/* save it, so mount count and last_transno is current */
- rc = mdt_server_data_update(env, mdt, (mti->mti_exp &&
- mti->mti_exp->exp_need_sync));
+ rc = mdt_server_data_update(env, mdt);
if (rc)
GOTO(err_client, rc);
RETURN(0);
err_client:
- target_recovery_fini(obd);
+ class_disconnect_exports(obd);
out:
return rc;
}
static int mdt_server_data_update(const struct lu_env *env,
- struct mdt_device *mdt,
- int need_sync)
+ struct mdt_device *mdt)
{
- int rc = 0;
- ENTRY;
+ struct mdt_thread_info *mti;
+ struct thandle *th;
+ int rc;
+
+ mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
+
+ mdt_trans_credit_init(env, mdt, MDT_TXN_LAST_RCVD_WRITE_OP);
+ th = mdt_trans_start(env, mdt);
+ if (IS_ERR(th))
+ RETURN(PTR_ERR(th));
CDEBUG(D_SUPER, "MDS mount_count is "LPU64", last_transno is "LPU64"\n",
mdt->mdt_mount_count, mdt->mdt_last_transno);
- spin_lock(&mdt->mdt_transno_lock);
+ cfs_spin_lock(&mdt->mdt_transno_lock);
mdt->mdt_lsd.lsd_last_transno = mdt->mdt_last_transno;
- spin_unlock(&mdt->mdt_transno_lock);
+ cfs_spin_unlock(&mdt->mdt_transno_lock);
- /*
- * This may be called from difficult reply handler and
- * mdt->mdt_last_rcvd may be NULL that time.
- */
- if (mdt->mdt_last_rcvd != NULL)
- rc = mdt_last_rcvd_header_write(env, mdt, need_sync);
- RETURN(rc);
+ rc = mdt_last_rcvd_header_write(env, mdt, th);
+ mdt_trans_stop(env, mdt, th);
+ return rc;
}
+
int mdt_client_new(const struct lu_env *env, struct mdt_device *mdt)
{
unsigned long *bitmap = mdt->mdt_client_bitmap;
/* the bitmap operations can handle cl_idx > sizeof(long) * 8, so
* there's no need for extra complication here
*/
- spin_lock(&mdt->mdt_client_bitmap_lock);
- cl_idx = find_first_zero_bit(bitmap, LR_MAX_CLIENTS);
+ cfs_spin_lock(&mdt->mdt_client_bitmap_lock);
+ cl_idx = cfs_find_first_zero_bit(bitmap, LR_MAX_CLIENTS);
if (cl_idx >= LR_MAX_CLIENTS ||
OBD_FAIL_CHECK(OBD_FAIL_MDS_CLIENT_ADD)) {
CERROR("no room for %u clients - fix LR_MAX_CLIENTS\n",
cl_idx);
- spin_unlock(&mdt->mdt_client_bitmap_lock);
+ cfs_spin_unlock(&mdt->mdt_client_bitmap_lock);
RETURN(-EOVERFLOW);
}
- set_bit(cl_idx, bitmap);
- spin_unlock(&mdt->mdt_client_bitmap_lock);
+ cfs_set_bit(cl_idx, bitmap);
+ cfs_spin_unlock(&mdt->mdt_client_bitmap_lock);
CDEBUG(D_INFO, "client at idx %d with UUID '%s' added\n",
cl_idx, med->med_lcd->lcd_uuid);
med->med_lr_idx = cl_idx;
med->med_lr_off = lsd->lsd_client_start +
(cl_idx * lsd->lsd_client_size);
- init_mutex(&med->med_lcd_lock);
+ cfs_init_mutex(&med->med_lcd_lock);
LASSERTF(med->med_lr_off > 0, "med_lr_off = %llu\n", med->med_lr_off);
/* Write new client data. */
off = med->med_lr_off;
+
+ if (OBD_FAIL_CHECK(OBD_FAIL_TGT_CLIENT_ADD))
+ RETURN(-ENOSPC);
+
mdt_trans_credit_init(env, mdt, MDT_TXN_LAST_RCVD_WRITE_OP);
th = mdt_trans_start(env, mdt);
if (IS_ERR(th))
RETURN(PTR_ERR(th));
- /*
+ /*
* Until this operations will be committed the sync is needed
* for this export. This should be done _after_ starting the
* transaction so that many connecting clients will not bring
- * server down with lots of sync writes.
+ * server down with lots of sync writes.
*/
- mdt_trans_add_cb(th, mdt_client_cb, mti->mti_exp);
- spin_lock(&mti->mti_exp->exp_lock);
+ mdt_trans_add_cb(th, lut_cb_client, class_export_cb_get(mti->mti_exp));
+ cfs_spin_lock(&mti->mti_exp->exp_lock);
mti->mti_exp->exp_need_sync = 1;
- spin_unlock(&mti->mti_exp->exp_lock);
+ cfs_spin_unlock(&mti->mti_exp->exp_lock);
rc = mdt_last_rcvd_write(env, mdt, lcd, &off, th);
- CDEBUG(D_INFO, "wrote client lcd at idx %u off %llu (len "LPSZ")\n",
- cl_idx, med->med_lr_off, sizeof(*lcd));
+ CDEBUG(D_INFO, "wrote client lcd at idx %u off %llu (len %u)\n",
+ cl_idx, med->med_lr_off, (int)sizeof(*lcd));
mdt_trans_stop(env, mdt, th);
RETURN(rc);
if (!strcmp(med->med_lcd->lcd_uuid, obd->obd_uuid.uuid))
RETURN(0);
- spin_lock(&mdt->mdt_client_bitmap_lock);
- if (test_and_set_bit(cl_idx, bitmap)) {
+ cfs_spin_lock(&mdt->mdt_client_bitmap_lock);
+ if (cfs_test_and_set_bit(cl_idx, bitmap)) {
CERROR("MDS client %d: bit already set in bitmap!!\n",
cl_idx);
LBUG();
}
- spin_unlock(&mdt->mdt_client_bitmap_lock);
+ cfs_spin_unlock(&mdt->mdt_client_bitmap_lock);
CDEBUG(D_INFO, "client at idx %d with UUID '%s' added\n",
cl_idx, med->med_lcd->lcd_uuid);
med->med_lr_idx = cl_idx;
med->med_lr_off = lsd->lsd_client_start +
(cl_idx * lsd->lsd_client_size);
- init_mutex(&med->med_lcd_lock);
+ cfs_init_mutex(&med->med_lcd_lock);
LASSERTF(med->med_lr_off > 0, "med_lr_off = %llu\n", med->med_lr_off);
struct obd_device *obd = mdt2obd_dev(mdt);
struct obd_export *exp;
struct thandle *th;
- int need_sync;
loff_t off;
int rc = 0;
ENTRY;
* Clear the bit _after_ zeroing out the client so we don't race with
* mdt_client_add and zero out new clients.
*/
- if (!test_bit(med->med_lr_idx, mdt->mdt_client_bitmap)) {
+ if (!cfs_test_bit(med->med_lr_idx, mdt->mdt_client_bitmap)) {
CERROR("MDT client %u: bit already clear in bitmap!!\n",
med->med_lr_idx);
LBUG();
}
- /* Don't force sync on disconnect if aborting recovery,
- * or it does num_clients * num_osts. b=17194 */
- need_sync = (!exp->exp_libclient || exp->exp_need_sync) &&
- !(exp->exp_flags & OBD_OPT_ABORT_RECOV);
+ /* Make sure the server's last_transno is up to date.
+ * This should be done before zeroing client slot so last_transno will
+ * be in server data or in client data in case of failure */
+ mdt_server_data_update(env, mdt);
- /*
- * This may be called from difficult reply handler path and
- * mdt->mdt_last_rcvd may be NULL that time.
- */
- if (mdt->mdt_last_rcvd != NULL) {
- mdt_trans_credit_init(env, mdt, MDT_TXN_LAST_RCVD_WRITE_OP);
-
- spin_lock(&exp->exp_lock);
- exp->exp_need_sync = need_sync;
- spin_unlock(&exp->exp_lock);
-
- th = mdt_trans_start(env, mdt);
- if (IS_ERR(th))
- GOTO(free, rc = PTR_ERR(th));
-
- if (need_sync) {
- /*
- * Until this operations will be committed the sync
- * is needed for this export.
- */
- mdt_trans_add_cb(th, mdt_client_cb, exp);
- }
-
- mutex_down(&med->med_lcd_lock);
- memset(lcd, 0, sizeof *lcd);
+ mdt_trans_credit_init(env, mdt, MDT_TXN_LAST_RCVD_WRITE_OP);
+ th = mdt_trans_start(env, mdt);
+ if (IS_ERR(th))
+ GOTO(free, rc = PTR_ERR(th));
- rc = mdt_last_rcvd_write(env, mdt, lcd, &off, th);
- mutex_up(&med->med_lcd_lock);
- mdt_trans_stop(env, mdt, th);
- }
+ cfs_mutex_down(&med->med_lcd_lock);
+ memset(lcd->lcd_uuid, 0, sizeof lcd->lcd_uuid);
+ rc = mdt_last_rcvd_write(env, mdt, lcd, &off, th);
+ cfs_mutex_up(&med->med_lcd_lock);
+ mdt_trans_stop(env, mdt, th);
CDEBUG(rc == 0 ? D_INFO : D_ERROR, "Zeroing out client idx %u in "
- "%s %ssync rc %d\n", med->med_lr_idx, LAST_RCVD,
- need_sync ? "" : "a", rc);
-
- spin_lock(&mdt->mdt_client_bitmap_lock);
- clear_bit(med->med_lr_idx, mdt->mdt_client_bitmap);
- spin_unlock(&mdt->mdt_client_bitmap_lock);
-
- /*
- * Make sure the server's last_transno is up to date. Do this
- * after the client is freed so we know all the client's
- * transactions have been committed.
- */
- mdt_server_data_update(env, mdt, need_sync);
-
- EXIT;
+ "%s, rc %d\n", med->med_lr_idx, LAST_RCVD, rc);
+ RETURN(0);
free:
- OBD_FREE_PTR(lcd);
- med->med_lcd = NULL;
return 0;
}
loff_t off;
int err;
__s32 rc = th->th_result;
- __u64 *transno_p;
ENTRY;
LASSERT(req);
LASSERT(mdt);
med = &req->rq_export->exp_mdt_data;
LASSERT(med);
+
+ cfs_mutex_down(&med->med_lcd_lock);
lcd = med->med_lcd;
- /* if the export has already been failed, we have no last_rcvd slot */
- if (req->rq_export->exp_failed) {
+ /* if the export has already been disconnected, we have no last_rcvd slot,
+ * update server data with latest transno then */
+ if (lcd == NULL) {
+ cfs_mutex_up(&med->med_lcd_lock);
CWARN("commit transaction for disconnected client %s: rc %d\n",
req->rq_export->exp_client_uuid.uuid, rc);
- if (rc == 0)
- rc = -ENOTCONN;
- RETURN(rc);
+ err = mdt_last_rcvd_header_write(mti->mti_env, mdt, th);
+ RETURN(err);
}
off = med->med_lr_off;
- mutex_down(&med->med_lcd_lock);
+ LASSERT(ergo(mti->mti_transno == 0, rc != 0));
if (lustre_msg_get_opc(req->rq_reqmsg) == MDS_CLOSE ||
lustre_msg_get_opc(req->rq_reqmsg) == MDS_DONE_WRITING) {
- transno_p = &lcd->lcd_last_close_transno;
+ if (mti->mti_transno != 0)
+ lcd->lcd_last_close_transno = mti->mti_transno;
lcd->lcd_last_close_xid = req->rq_xid;
lcd->lcd_last_close_result = rc;
} else {
- transno_p = &lcd->lcd_last_transno;
+ /* VBR: save versions in last_rcvd for reconstruct. */
+ __u64 *pre_versions = lustre_msg_get_versions(req->rq_repmsg);
+ if (pre_versions) {
+ lcd->lcd_pre_versions[0] = pre_versions[0];
+ lcd->lcd_pre_versions[1] = pre_versions[1];
+ lcd->lcd_pre_versions[2] = pre_versions[2];
+ lcd->lcd_pre_versions[3] = pre_versions[3];
+ }
+ if (mti->mti_transno != 0)
+ lcd->lcd_last_transno = mti->mti_transno;
lcd->lcd_last_xid = req->rq_xid;
lcd->lcd_last_result = rc;
/*XXX: save intent_disposition in mdt_thread_info?
lcd->lcd_last_data = mti->mti_opdata;
}
- /*
- * When we store zero transno in lcd we can lost last transno value
- * because lcd contains 0, but lsd is not yet written
- * The server data should be updated also if the latest
- * transno is rewritten by zero. See the bug 11125 for details.
- */
- if (mti->mti_transno == 0 &&
- *transno_p == mdt->mdt_last_transno)
- mdt_server_data_update(mti->mti_env, mdt,
- (mti->mti_exp &&
- mti->mti_exp->exp_need_sync));
-
- *transno_p = mti->mti_transno;
-
if (off <= 0) {
CERROR("client idx %d has offset %lld\n", med->med_lr_idx, off);
err = -EINVAL;
} else {
err = mdt_last_rcvd_write(mti->mti_env, mdt, lcd, &off, th);
}
- mutex_up(&med->med_lcd_lock);
+ cfs_mutex_up(&med->med_lcd_lock);
RETURN(err);
}
return 0;
}
+/* Set new object versions */
+static void mdt_versions_set(struct mdt_thread_info *info)
+{
+ int i;
+ for (i = 0; i < PTLRPC_NUM_VERSIONS; i++)
+ if (info->mti_mos[i] != NULL)
+ mo_version_set(info->mti_env,
+ mdt_object_child(info->mti_mos[i]),
+ info->mti_transno);
+}
+
/* Update last_rcvd records with latests transaction data */
static int mdt_txn_stop_cb(const struct lu_env *env,
struct thandle *txn, void *cookie)
}
mti->mti_has_trans = 1;
- spin_lock(&mdt->mdt_transno_lock);
+ cfs_spin_lock(&mdt->mdt_transno_lock);
if (txn->th_result != 0) {
if (mti->mti_transno != 0) {
CERROR("Replay transno "LPU64" failed: rc %i\n",
mti->mti_transno, txn->th_result);
- mti->mti_transno = 0;
}
} else if (mti->mti_transno == 0) {
mti->mti_transno = ++ mdt->mdt_last_transno;
if (mti->mti_transno > mdt->mdt_last_transno)
mdt->mdt_last_transno = mti->mti_transno;
}
-
+ cfs_spin_unlock(&mdt->mdt_transno_lock);
/* sometimes the reply message has not been successfully packed */
LASSERT(req != NULL && req->rq_repmsg != NULL);
+ /** VBR: set new versions */
+ if (txn->th_result == 0)
+ mdt_versions_set(mti);
+
/* filling reply data */
CDEBUG(D_INODE, "transno = %llu, last_committed = %llu\n",
mti->mti_transno, req->rq_export->exp_obd->obd_last_committed);
req->rq_transno = mti->mti_transno;
lustre_msg_set_transno(req->rq_repmsg, mti->mti_transno);
- lustre_msg_set_last_xid(req->rq_repmsg,
- lcd_last_xid(req->rq_export->exp_mdt_data.med_lcd));
/* save transno for the commit callback */
txi->txi_transno = mti->mti_transno;
- spin_unlock(&mdt->mdt_transno_lock);
+
+ /* add separate commit callback for transaction handling because we need
+ * export as parameter */
+ mdt_trans_add_cb(txn, lut_cb_last_committed,
+ class_export_cb_get(mti->mti_exp));
return mdt_last_rcvd_update(mti, txn);
}
struct thandle *txn, void *cookie)
{
struct mdt_device *mdt = cookie;
- struct obd_device *obd = mdt2obd_dev(mdt);
struct mdt_txn_info *txi;
int i;
txi = lu_context_key_get(&txn->th_ctx, &mdt_txn_key);
- /* copy of obd_transno_commit_cb() but with locking */
- spin_lock(&mdt->mdt_transno_lock);
- if (txi->txi_transno > obd->obd_last_committed) {
- obd->obd_last_committed = txi->txi_transno;
- spin_unlock(&mdt->mdt_transno_lock);
- ptlrpc_commit_replies(obd);
- } else
- spin_unlock(&mdt->mdt_transno_lock);
-
- if (txi->txi_transno)
- CDEBUG(D_HA, "%s: transno "LPD64" is committed\n",
- obd->obd_name, txi->txi_transno);
-
/* iterate through all additional callbacks */
for (i = 0; i < txi->txi_cb_count; i++) {
- txi->txi_cb[i].mdt_cb_func(mdt, txi->txi_transno,
- txi->txi_cb[i].mdt_cb_data, 0);
+ txi->txi_cb[i].lut_cb_func(&mdt->mdt_lut, txi->txi_transno,
+ txi->txi_cb[i].lut_cb_data, 0);
}
return 0;
}
mdt->mdt_txn_cb.dtc_txn_stop = mdt_txn_stop_cb;
mdt->mdt_txn_cb.dtc_txn_commit = mdt_txn_commit_cb;
mdt->mdt_txn_cb.dtc_cookie = mdt;
+ mdt->mdt_txn_cb.dtc_tag = LCT_MD_THREAD;
CFS_INIT_LIST_HEAD(&mdt->mdt_txn_cb.dtc_linkage);
dt_txn_callback_add(mdt->mdt_bottom, &mdt->mdt_txn_cb);
- o = dt_store_open(env, mdt->mdt_bottom, "", LAST_RCVD, &fid);
- if (!IS_ERR(o)) {
- mdt->mdt_last_rcvd = o;
- rc = mdt_server_data_init(env, mdt, lsi);
- if (rc)
- GOTO(put_last_rcvd, rc);
- } else {
- rc = PTR_ERR(o);
- CERROR("cannot open %s: rc = %d\n", LAST_RCVD, rc);
+ rc = mdt_server_data_init(env, mdt, lsi);
+ if (rc)
RETURN(rc);
- }
o = dt_store_open(env, mdt->mdt_bottom, "", CAPA_KEYS, &fid);
if (!IS_ERR(o)) {
} else {
rc = PTR_ERR(o);
CERROR("cannot open %s: rc = %d\n", CAPA_KEYS, rc);
- GOTO(put_last_rcvd, rc);
+ GOTO(disconnect_exports, rc);
}
RETURN(0);
put_ck_object:
lu_object_put(env, &o->do_lu);
mdt->mdt_ck_obj = NULL;
-put_last_rcvd:
- lu_object_put(env, &mdt->mdt_last_rcvd->do_lu);
- mdt->mdt_last_rcvd = NULL;
+disconnect_exports:
+ class_disconnect_exports(obd);
return rc;
}
/* Remove transaction callback */
dt_txn_callback_del(mdt->mdt_bottom, &mdt->mdt_txn_cb);
- if (mdt->mdt_last_rcvd)
- lu_object_put(env, &mdt->mdt_last_rcvd->do_lu);
- mdt->mdt_last_rcvd = NULL;
if (mdt->mdt_ck_obj)
lu_object_put(env, &mdt->mdt_ck_obj->do_lu);
mdt->mdt_ck_obj = NULL;
static void mdt_steal_ack_locks(struct ptlrpc_request *req)
{
struct obd_export *exp = req->rq_export;
- struct list_head *tmp;
+ cfs_list_t *tmp;
struct ptlrpc_reply_state *oldrep;
struct ptlrpc_service *svc;
int i;
/* CAVEAT EMPTOR: spinlock order */
- spin_lock(&exp->exp_lock);
- list_for_each (tmp, &exp->exp_outstanding_replies) {
- oldrep = list_entry(tmp, struct ptlrpc_reply_state,rs_exp_list);
+ cfs_spin_lock(&exp->exp_lock);
+ cfs_list_for_each (tmp, &exp->exp_outstanding_replies) {
+ oldrep = cfs_list_entry(tmp, struct ptlrpc_reply_state,
+ rs_exp_list);
if (oldrep->rs_xid != req->rq_xid)
continue;
- if (lustre_msg_get_opc(oldrep->rs_msg) !=
- lustre_msg_get_opc(req->rq_reqmsg))
- CERROR ("Resent req xid "LPX64" has mismatched opc: "
+ if (oldrep->rs_opc != lustre_msg_get_opc(req->rq_reqmsg))
+ CERROR ("Resent req xid "LPU64" has mismatched opc: "
"new %d old %d\n", req->rq_xid,
lustre_msg_get_opc(req->rq_reqmsg),
- lustre_msg_get_opc(oldrep->rs_msg));
+ oldrep->rs_opc);
svc = oldrep->rs_service;
- spin_lock (&svc->srv_lock);
+ cfs_spin_lock (&svc->srv_lock);
- list_del_init (&oldrep->rs_exp_list);
+ cfs_list_del_init (&oldrep->rs_exp_list);
CWARN("Stealing %d locks from rs %p x"LPD64".t"LPD64
" o%d NID %s\n",
oldrep->rs_nlocks, oldrep,
- oldrep->rs_xid, oldrep->rs_transno,
- lustre_msg_get_opc(oldrep->rs_msg),
+ oldrep->rs_xid, oldrep->rs_transno, oldrep->rs_opc,
libcfs_nid2str(exp->exp_connection->c_peer.nid));
for (i = 0; i < oldrep->rs_nlocks; i++)
oldrep->rs_nlocks = 0;
DEBUG_REQ(D_HA, req, "stole locks for");
- spin_lock(&oldrep->rs_lock);
+ cfs_spin_lock(&oldrep->rs_lock);
ptlrpc_schedule_difficult_reply (oldrep);
- spin_unlock(&oldrep->rs_lock);
+ cfs_spin_unlock(&oldrep->rs_lock);
- spin_unlock (&svc->srv_lock);
+ cfs_spin_unlock (&svc->srv_lock);
break;
}
- spin_unlock(&exp->exp_lock);
+ cfs_spin_unlock(&exp->exp_lock);
+}
+
+/**
+ * VBR: restore versions
+ */
+void mdt_vbr_reconstruct(struct ptlrpc_request *req,
+ struct lsd_client_data *lcd)
+{
+ __u64 pre_versions[4] = {0};
+ pre_versions[0] = lcd->lcd_pre_versions[0];
+ pre_versions[1] = lcd->lcd_pre_versions[1];
+ pre_versions[2] = lcd->lcd_pre_versions[2];
+ pre_versions[3] = lcd->lcd_pre_versions[3];
+ lustre_msg_set_versions(req->rq_repmsg, pre_versions);
}
void mdt_req_from_lcd(struct ptlrpc_request *req,
lustre_msg_get_opc(req->rq_repmsg) == MDS_DONE_WRITING) {
req->rq_transno = lcd->lcd_last_close_transno;
req->rq_status = lcd->lcd_last_close_result;
- lustre_msg_set_transno(req->rq_repmsg, req->rq_transno);
- lustre_msg_set_status(req->rq_repmsg, req->rq_status);
} else {
req->rq_transno = lcd->lcd_last_transno;
req->rq_status = lcd->lcd_last_result;
- lustre_msg_set_transno(req->rq_repmsg, req->rq_transno);
- lustre_msg_set_status(req->rq_repmsg, req->rq_status);
+ mdt_vbr_reconstruct(req, lcd);
}
+ if (req->rq_status != 0)
+ req->rq_transno = 0;
+ lustre_msg_set_transno(req->rq_repmsg, req->rq_transno);
+ lustre_msg_set_status(req->rq_repmsg, req->rq_status);
+ DEBUG_REQ(D_RPCTRACE, req, "restoring transno "LPD64"/status %d",
+ req->rq_transno, req->rq_status);
+
mdt_steal_ack_locks(req);
}
mo_attr_get(mti->mti_env, mdt_object_child(obj), &mti->mti_attr);
mdt_pack_attr2body(mti, body, &mti->mti_attr.ma_attr,
mdt_object_fid(obj));
- if (mti->mti_epoch && (mti->mti_epoch->flags & MF_EPOCH_OPEN)) {
+ if (mti->mti_ioepoch && (mti->mti_ioepoch->flags & MF_EPOCH_OPEN)) {
struct mdt_file_data *mfd;
struct mdt_body *repbody;
repbody = req_capsule_server_get(mti->mti_pill, &RMF_MDT_BODY);
repbody->ioepoch = obj->mot_ioepoch;
- spin_lock(&med->med_open_lock);
- list_for_each_entry(mfd, &med->med_open_head, mfd_list) {
+ cfs_spin_lock(&med->med_open_lock);
+ cfs_list_for_each_entry(mfd, &med->med_open_head, mfd_list) {
if (mfd->mfd_xid == req->rq_xid)
break;
}
LASSERT(&mfd->mfd_list != &med->med_open_head);
- spin_unlock(&med->med_open_lock);
+ cfs_spin_unlock(&med->med_open_lock);
repbody->handle.cookie = mfd->mfd_handle.h_cookie;
}
git://git.whamcloud.com / fs / lustre-release.git / blobdiff