X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Ftarget%2Fout_handler.c;h=e91c33824a1b397c4bdc4939a319845324d49949;hb=c2a6f1a2afee6c416a3e7659b863bcc6246761f0;hp=002cda25a360dddb043f08460e0068a97a157251;hpb=dcaf8a6724ef36f2124428a755b5bb39bcd0c5fd;p=fs%2Flustre-release.git diff --git a/lustre/target/out_handler.c b/lustre/target/out_handler.c index 002cda2..e91c338 100644 --- a/lustre/target/out_handler.c +++ b/lustre/target/out_handler.c @@ -20,7 +20,7 @@ * GPL HEADER END */ /* - * Copyright (c) 2013, 2014, Intel Corporation. + * Copyright (c) 2013, 2017, Intel Corporation. * * lustre/target/out_handler.c * @@ -31,10 +31,21 @@ #define DEBUG_SUBSYSTEM S_CLASS -#include +#include +#include +#include +#include #include +#include #include "tgt_internal.h" -#include + +static inline void orr_cpu_to_le(struct out_read_reply *orr_dst, + const struct out_read_reply *orr_src) +{ + orr_dst->orr_size = cpu_to_le32(orr_src->orr_size); + orr_dst->orr_padding = cpu_to_le32(orr_src->orr_padding); + orr_dst->orr_offset = cpu_to_le64(orr_dst->orr_offset); +} static void out_reconstruct(const struct lu_env *env, struct dt_device *dt, struct dt_object *obj, @@ -45,7 +56,6 @@ static void out_reconstruct(const struct lu_env *env, struct dt_device *dt, dt_obd_name(dt), reply, index, 0); object_update_result_insert(reply, NULL, 0, index, 0); - return; } typedef void (*out_reconstruct_t)(const struct lu_env *env, @@ -79,13 +89,12 @@ static inline int out_check_resent(const struct lu_env *env, 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); + DEBUG_REQ(D_RPCTRACE, req, "restoring resent RPC"); reconstruct(env, dt, obj, reply, index); return 1; } - DEBUG_REQ(D_HA, req, "no reply for RESENT req (have "LPD64")", + DEBUG_REQ(D_HA, req, "no reply for RESENT req (have %lld)", req->rq_export->exp_target_data.ted_lcd->lcd_last_xid); return 0; } @@ -106,10 +115,10 @@ static int out_create(struct tgt_session_info *tsi) ENTRY; wobdo = object_update_param_get(update, 0, &size); - if (wobdo == NULL || IS_ERR(wobdo) || size != sizeof(*wobdo)) { - CERROR("%s: obdo is NULL, invalid RPC: rc = %d\n", - tgt_name(tsi->tsi_tgt), -EPROTO); - RETURN(err_serious(-EPROTO)); + if (IS_ERR(wobdo) || size != sizeof(*wobdo)) { + CERROR("%s: obdo is NULL, invalid RPC: rc = %ld\n", + tgt_name(tsi->tsi_tgt), PTR_ERR(wobdo)); + RETURN(PTR_ERR(wobdo)); } if (ptlrpc_req_need_swab(tsi->tsi_pill->rc_req)) @@ -120,17 +129,17 @@ static int out_create(struct tgt_session_info *tsi) dof->dof_type = dt_mode_to_dft(attr->la_mode); if (update->ou_params_count > 1) { fid = object_update_param_get(update, 1, &size); - if (fid == NULL || IS_ERR(fid) || size != sizeof(*fid)) { - CERROR("%s: invalid fid: rc = %d\n", - tgt_name(tsi->tsi_tgt), -EPROTO); - RETURN(err_serious(-EPROTO)); + if (IS_ERR(fid) || size != sizeof(*fid)) { + CERROR("%s: invalid fid: rc = %ld\n", + tgt_name(tsi->tsi_tgt), PTR_ERR(fid)); + RETURN(PTR_ERR(fid)); } if (ptlrpc_req_need_swab(tsi->tsi_pill->rc_req)) lustre_swab_lu_fid(fid); if (!fid_is_sane(fid)) { CERROR("%s: invalid fid "DFID": rc = %d\n", tgt_name(tsi->tsi_tgt), PFID(fid), -EPROTO); - RETURN(err_serious(-EPROTO)); + RETURN(-EPROTO); } } @@ -159,10 +168,10 @@ static int out_attr_set(struct tgt_session_info *tsi) ENTRY; wobdo = object_update_param_get(update, 0, &size); - if (wobdo == NULL || IS_ERR(wobdo) || size != sizeof(*wobdo)) { - CERROR("%s: empty obdo in the update: rc = %d\n", - tgt_name(tsi->tsi_tgt), -EPROTO); - RETURN(err_serious(-EPROTO)); + if (IS_ERR(wobdo) || size != sizeof(*wobdo)) { + CERROR("%s: empty obdo in the update: rc = %ld\n", + tgt_name(tsi->tsi_tgt), PTR_ERR(wobdo)); + RETURN(PTR_ERR(wobdo)); } attr->la_valid = 0; @@ -185,6 +194,7 @@ static int out_attr_get(struct tgt_session_info *tsi) { const struct lu_env *env = tsi->tsi_env; struct tgt_thread_info *tti = tgt_th_info(env); + struct object_update *update = tti->tti_u.update.tti_update; struct obdo *obdo = &tti->tti_u.update.tti_obdo; struct lu_attr *la = &tti->tti_attr; struct dt_object *obj = tti->tti_u.update.tti_dt_object; @@ -193,6 +203,9 @@ static int out_attr_get(struct tgt_session_info *tsi) ENTRY; + if (unlikely(update->ou_result_size < sizeof(*obdo))) + return -EPROTO; + if (!lu_object_exists(&obj->do_lu)) { /* Usually, this will be called when the master MDT try * to init a remote object(see osp_object_init), so if @@ -203,14 +216,13 @@ static int out_attr_get(struct tgt_session_info *tsi) RETURN(-ENOENT); } - dt_read_lock(env, obj, MOR_TGT_CHILD); + dt_read_lock(env, obj, DT_TGT_CHILD); rc = dt_attr_get(env, obj, la); if (rc) GOTO(out_unlock, rc); obdo->o_valid = 0; obdo_from_la(obdo, la, la->la_valid); - lustre_set_wire_obdo(NULL, obdo, obdo); out_unlock: dt_read_unlock(env, obj); @@ -247,41 +259,97 @@ static int out_xattr_get(struct tgt_session_info *tsi) } name = object_update_param_get(update, 0, NULL); - if (name == NULL || IS_ERR(name)) { - CERROR("%s: empty name for xattr get: rc = %d\n", - tgt_name(tsi->tsi_tgt), -EPROTO); - RETURN(err_serious(-EPROTO)); + if (IS_ERR(name)) { + CERROR("%s: empty name for xattr get: rc = %ld\n", + tgt_name(tsi->tsi_tgt), PTR_ERR(name)); + RETURN(PTR_ERR(name)); } - update_result = object_update_result_get(reply, 0, NULL); + update_result = object_update_result_get(reply, idx, NULL); if (update_result == NULL) { CERROR("%s: empty name for xattr get: rc = %d\n", tgt_name(tsi->tsi_tgt), -EPROTO); - RETURN(err_serious(-EPROTO)); + RETURN(-EPROTO); } - lbuf->lb_buf = update_result->our_data; - lbuf->lb_len = OUT_UPDATE_REPLY_SIZE - - cfs_size_round((unsigned long)update_result->our_data - - (unsigned long)update_result); - dt_read_lock(env, obj, MOR_TGT_CHILD); + lbuf->lb_len = (int)tti->tti_u.update.tti_update->ou_result_size; + if (lbuf->lb_len == 0) + lbuf->lb_buf = NULL; + else + lbuf->lb_buf = update_result->our_data; + + dt_read_lock(env, obj, DT_TGT_CHILD); rc = dt_xattr_get(env, obj, lbuf, name); dt_read_unlock(env, obj); - if (rc < 0) { + if (rc <= 0) { lbuf->lb_len = 0; - GOTO(out, rc); + if (unlikely(!rc)) + rc = -ENODATA; + } else if (lbuf->lb_buf) { + lbuf->lb_len = rc; } - lbuf->lb_len = rc; - rc = 0; CDEBUG(D_INFO, "%s: "DFID" get xattr %s len %d\n", tgt_name(tsi->tsi_tgt), PFID(lu_object_fid(&obj->do_lu)), - name, (int)lbuf->lb_len); + name, rc); GOTO(out, rc); out: object_update_result_insert(reply, lbuf->lb_buf, lbuf->lb_len, idx, rc); - RETURN(rc); + RETURN(0); +} + +static int out_xattr_list(struct tgt_session_info *tsi) +{ + const struct lu_env *env = tsi->tsi_env; + struct tgt_thread_info *tti = tgt_th_info(env); + struct lu_buf *lbuf = &tti->tti_buf; + struct object_update_reply *reply = tti->tti_u.update.tti_update_reply; + struct dt_object *obj = tti->tti_u.update.tti_dt_object; + struct object_update_result *update_result; + int idx = tti->tti_u.update.tti_update_reply_index; + int rc; + + ENTRY; + + if (!lu_object_exists(&obj->do_lu)) { + set_bit(LU_OBJECT_HEARD_BANSHEE, + &obj->do_lu.lo_header->loh_flags); + RETURN(-ENOENT); + } + + update_result = object_update_result_get(reply, 0, NULL); + if (!update_result) { + rc = -EPROTO; + CERROR("%s: empty buf for xattr list: rc = %d\n", + tgt_name(tsi->tsi_tgt), rc); + RETURN(rc); + } + + lbuf->lb_len = (int)tti->tti_u.update.tti_update->ou_result_size; + lbuf->lb_buf = update_result->our_data; + if (lbuf->lb_len == 0) + lbuf->lb_buf = 0; + + dt_read_lock(env, obj, DT_TGT_CHILD); + rc = dt_xattr_list(env, obj, lbuf); + dt_read_unlock(env, obj); + if (rc <= 0) { + lbuf->lb_len = 0; + if (unlikely(!rc)) + rc = -ENODATA; + } else if (lbuf->lb_buf) { + lbuf->lb_len = rc; + } + + CDEBUG(D_INFO, "%s: "DFID" list xattr len %d\n", + tgt_name(tsi->tsi_tgt), PFID(lu_object_fid(&obj->do_lu)), rc); + + /* Since we directly use update_result->our_data as the lbuf->lb_buf, + * then use NULL for result_insert to avoid unnecessary memory copy. */ + object_update_result_insert(reply, NULL, lbuf->lb_len, idx, rc); + + RETURN(0); } static int out_index_lookup(struct tgt_session_info *tsi) @@ -295,17 +363,20 @@ static int out_index_lookup(struct tgt_session_info *tsi) ENTRY; + if (unlikely(update->ou_result_size < sizeof(tti->tti_fid1))) + return -EPROTO; + if (!lu_object_exists(&obj->do_lu)) RETURN(-ENOENT); name = object_update_param_get(update, 0, NULL); - if (name == NULL || IS_ERR(name)) { - CERROR("%s: empty name for lookup: rc = %d\n", - tgt_name(tsi->tsi_tgt), -EPROTO); - RETURN(err_serious(-EPROTO)); + if (IS_ERR(name)) { + CERROR("%s: empty name for lookup: rc = %ld\n", + tgt_name(tsi->tsi_tgt), PTR_ERR(name)); + RETURN(PTR_ERR(name)); } - dt_read_lock(env, obj, MOR_TGT_CHILD); + dt_read_lock(env, obj, DT_TGT_CHILD); if (!dt_try_as_dir(env, obj)) GOTO(out_unlock, rc = -ENOTDIR); @@ -351,24 +422,25 @@ static int out_xattr_set(struct tgt_session_info *tsi) ENTRY; name = object_update_param_get(update, 0, NULL); - if (name == NULL || IS_ERR(name)) { - CERROR("%s: empty name for xattr set: rc = %d\n", - tgt_name(tsi->tsi_tgt), -EPROTO); - RETURN(err_serious(-EPROTO)); + if (IS_ERR(name)) { + CERROR("%s: empty name for xattr set: rc = %ld\n", + tgt_name(tsi->tsi_tgt), PTR_ERR(name)); + RETURN(PTR_ERR(name)); } + /* If buffer == NULL (-ENODATA), then it might mean delete xattr */ buf = object_update_param_get(update, 1, &buf_len); - if (IS_ERR(buf)) - RETURN(err_serious(-EPROTO)); + if (IS_ERR(buf) && PTR_ERR(buf) != -ENODATA) + RETURN(PTR_ERR(buf)); lbuf->lb_buf = buf; lbuf->lb_len = buf_len; tmp = object_update_param_get(update, 2, &size); - if (tmp == NULL || IS_ERR(tmp) || size != sizeof(*tmp)) { - CERROR("%s: emptry or wrong size %zu flag: rc = %d\n", - tgt_name(tsi->tsi_tgt), size, -EPROTO); - RETURN(err_serious(-EPROTO)); + if (IS_ERR(tmp) || size != sizeof(*tmp)) { + CERROR("%s: emptry or wrong size %zu flag: rc = %ld\n", + tgt_name(tsi->tsi_tgt), size, PTR_ERR(tmp)); + RETURN(PTR_ERR(tmp)); } if (ptlrpc_req_need_swab(tsi->tsi_pill->rc_req)) @@ -392,10 +464,10 @@ static int out_xattr_del(struct tgt_session_info *tsi) ENTRY; name = object_update_param_get(update, 0, NULL); - if (name == NULL || IS_ERR(name)) { - CERROR("%s: empty name for xattr set: rc = %d\n", - tgt_name(tsi->tsi_tgt), -EPROTO); - RETURN(err_serious(-EPROTO)); + if (IS_ERR(name)) { + CERROR("%s: empty name for xattr set: rc = %ld\n", + tgt_name(tsi->tsi_tgt), PTR_ERR(name)); + RETURN(PTR_ERR(name)); } rc = out_tx_xattr_del(tsi->tsi_env, obj, name, &tti->tti_tea, @@ -455,17 +527,17 @@ static int out_index_insert(struct tgt_session_info *tsi) ENTRY; name = object_update_param_get(update, 0, NULL); - if (name == NULL || IS_ERR(name)) { - CERROR("%s: empty name for index insert: rc = %d\n", - tgt_name(tsi->tsi_tgt), -EPROTO); - RETURN(err_serious(-EPROTO)); + if (IS_ERR(name)) { + CERROR("%s: empty name for index insert: rc = %ld\n", + tgt_name(tsi->tsi_tgt), PTR_ERR(name)); + RETURN(PTR_ERR(name)); } fid = object_update_param_get(update, 1, &size); - if (fid == NULL || IS_ERR(fid) || size != sizeof(*fid)) { - CERROR("%s: invalid fid: rc = %d\n", - tgt_name(tsi->tsi_tgt), -EPROTO); - RETURN(err_serious(-EPROTO)); + if (IS_ERR(fid) || size != sizeof(*fid)) { + CERROR("%s: invalid fid: rc = %ld\n", + tgt_name(tsi->tsi_tgt), PTR_ERR(fid)); + RETURN(PTR_ERR(fid)); } if (ptlrpc_req_need_swab(tsi->tsi_pill->rc_req)) @@ -474,14 +546,14 @@ static int out_index_insert(struct tgt_session_info *tsi) if (!fid_is_sane(fid)) { CERROR("%s: invalid FID "DFID": rc = %d\n", tgt_name(tsi->tsi_tgt), PFID(fid), -EPROTO); - RETURN(err_serious(-EPROTO)); + RETURN(-EPROTO); } ptype = object_update_param_get(update, 2, &size); - if (ptype == NULL || IS_ERR(ptype) || size != sizeof(*ptype)) { - CERROR("%s: invalid type for index insert: rc = %d\n", - tgt_name(tsi->tsi_tgt), -EPROTO); - RETURN(err_serious(-EPROTO)); + if (IS_ERR(ptype) || size != sizeof(*ptype)) { + CERROR("%s: invalid type for index insert: rc = %ld\n", + tgt_name(tsi->tsi_tgt), PTR_ERR(ptype)); + RETURN(PTR_ERR(ptype)); } if (ptlrpc_req_need_swab(tsi->tsi_pill->rc_req)) @@ -495,6 +567,11 @@ static int out_index_insert(struct tgt_session_info *tsi) tti->tti_tea.ta_handle, tti->tti_u.update.tti_update_reply, tti->tti_u.update.tti_update_reply_index); + + CDEBUG(D_INFO, "%s: "DFID" index insert %s: rc = %d\n", + tgt_name(tsi->tsi_tgt), PFID(lu_object_fid(&obj->do_lu)), + name, rc); + RETURN(rc); } @@ -510,10 +587,10 @@ static int out_index_delete(struct tgt_session_info *tsi) RETURN(-ENOENT); name = object_update_param_get(update, 0, NULL); - if (name == NULL || IS_ERR(name)) { - CERROR("%s: empty name for index delete: rc = %d\n", - tgt_name(tsi->tsi_tgt), -EPROTO); - RETURN(err_serious(-EPROTO)); + if (IS_ERR(name)) { + CERROR("%s: empty name for index delete: rc = %ld\n", + tgt_name(tsi->tsi_tgt), PTR_ERR(name)); + RETURN(PTR_ERR(name)); } rc = out_tx_index_delete(tsi->tsi_env, obj, (const struct dt_key *)name, @@ -536,7 +613,7 @@ static int out_destroy(struct tgt_session_info *tsi) if (!fid_is_sane(fid)) { CERROR("%s: invalid FID "DFID": rc = %d\n", tgt_name(tsi->tsi_tgt), PFID(fid), -EPROTO); - RETURN(err_serious(-EPROTO)); + RETURN(-EPROTO); } if (!lu_object_exists(&obj->do_lu)) @@ -565,19 +642,19 @@ static int out_write(struct tgt_session_info *tsi) ENTRY; buf = object_update_param_get(update, 0, &buf_len); - if (buf == NULL || IS_ERR(buf) || buf_len == 0) { - CERROR("%s: empty buf for xattr set: rc = %d\n", - tgt_name(tsi->tsi_tgt), -EPROTO); - RETURN(err_serious(-EPROTO)); + if (IS_ERR(buf) || buf_len == 0) { + CERROR("%s: empty buf for xattr set: rc = %ld\n", + tgt_name(tsi->tsi_tgt), PTR_ERR(buf)); + RETURN(PTR_ERR(buf)); } lbuf->lb_buf = buf; lbuf->lb_len = buf_len; tmp = object_update_param_get(update, 1, &size); - if (tmp == NULL || IS_ERR(tmp) || size != sizeof(*tmp)) { - CERROR("%s: empty or wrong size %zu pos: rc = %d\n", - tgt_name(tsi->tsi_tgt), size, -EPROTO); - RETURN(err_serious(-EPROTO)); + if (IS_ERR(tmp) || size != sizeof(*tmp)) { + CERROR("%s: empty or wrong size %zu pos: rc = %ld\n", + tgt_name(tsi->tsi_tgt), size, PTR_ERR(tmp)); + RETURN(PTR_ERR(tmp)); } if (ptlrpc_req_need_swab(tsi->tsi_pill->rc_req)) @@ -598,14 +675,17 @@ static int out_read(struct tgt_session_info *tsi) struct object_update *update = tti->tti_u.update.tti_update; struct dt_object *obj = tti->tti_u.update.tti_dt_object; struct object_update_reply *reply = tti->tti_u.update.tti_update_reply; - int index = tti->tti_u.update.tti_update_reply_index; + int index = tti->tti_u.update.tti_update_reply_index; + struct lu_rdbuf *rdbuf; struct object_update_result *update_result; - struct lu_buf *lbuf = &tti->tti_buf; struct out_read_reply *orr; - void *tmp; - size_t size; - __u64 pos; - int rc; + void *tmp; + size_t size; + size_t total_size = 0; + __u64 pos; + unsigned int i; + unsigned int nbufs; + int rc = 0; ENTRY; update_result = object_update_result_get(reply, index, NULL); @@ -616,47 +696,72 @@ static int out_read(struct tgt_session_info *tsi) GOTO(out, rc = -ENOENT); tmp = object_update_param_get(update, 0, NULL); - if (tmp == NULL || IS_ERR(tmp)) { - CERROR("%s: empty size for read: rc = %d\n", - tgt_name(tsi->tsi_tgt), -EPROTO); - GOTO(out, rc = err_serious(-EPROTO)); + if (IS_ERR(tmp)) { + CERROR("%s: empty size for read: rc = %ld\n", + tgt_name(tsi->tsi_tgt), PTR_ERR(tmp)); + GOTO(out, rc = PTR_ERR(tmp)); } size = le64_to_cpu(*(size_t *)(tmp)); tmp = object_update_param_get(update, 1, NULL); - if (tmp == NULL || IS_ERR(tmp)) { - CERROR("%s: empty pos for read: rc = %d\n", - tgt_name(tsi->tsi_tgt), -EPROTO); - GOTO(out, rc = err_serious(-EPROTO)); + if (IS_ERR(tmp)) { + CERROR("%s: empty pos for read: rc = %ld\n", + tgt_name(tsi->tsi_tgt), PTR_ERR(tmp)); + GOTO(out, rc = PTR_ERR(tmp)); } pos = le64_to_cpu(*(__u64 *)(tmp)); - /* Check if the read buffer can hold the read_size */ - if (size > OUT_UPDATE_REPLY_SIZE - - cfs_size_round(offsetof(struct object_update_reply, - ourp_lens[1])) - - cfs_size_round(sizeof(*update_result)) - - cfs_size_round(sizeof(*orr))) { - CERROR("%s: get %zu the biggest read size is %d: rc = %d\n", - tgt_name(tsi->tsi_tgt), size, OUT_UPDATE_REPLY_SIZE, - -EPROTO); - GOTO(out, rc = err_serious(-EPROTO)); - } - /* Put the offset into the begining of the buffer in reply */ orr = (struct out_read_reply *)update_result->our_data; - lbuf->lb_buf = orr->orr_data; - lbuf->lb_len = size; + nbufs = (size + OUT_BULK_BUFFER_SIZE - 1) / OUT_BULK_BUFFER_SIZE; + OBD_ALLOC(rdbuf, sizeof(*rdbuf) + nbufs * sizeof(rdbuf->rb_bufs[0])); + if (rdbuf == NULL) + GOTO(out, rc = -ENOMEM); - dt_read_lock(env, obj, MOR_TGT_CHILD); - rc = dt_read(env, obj, lbuf, &pos); - dt_read_unlock(env, obj); - orr->orr_size = rc < 0 ? 0 : rc; + rdbuf->rb_nbufs = 0; + total_size = 0; + for (i = 0; i < nbufs; i++) { + __u32 read_size; + + read_size = size > OUT_BULK_BUFFER_SIZE ? + OUT_BULK_BUFFER_SIZE : size; + OBD_ALLOC(rdbuf->rb_bufs[i].lb_buf, read_size); + if (rdbuf->rb_bufs[i].lb_buf == NULL) + GOTO(out_free, rc = -ENOMEM); + + rdbuf->rb_bufs[i].lb_len = read_size; + dt_read_lock(env, obj, DT_TGT_CHILD); + rc = dt_read(env, obj, &rdbuf->rb_bufs[i], &pos); + dt_read_unlock(env, obj); + + total_size += rc < 0 ? 0 : rc; + if (rc <= 0) + break; + + rdbuf->rb_nbufs++; + size -= read_size; + } + + /* send pages to client */ + rc = tgt_send_buffer(tsi, rdbuf); + if (rc < 0) + GOTO(out_free, rc); + + orr->orr_size = total_size; orr->orr_offset = pos; orr_cpu_to_le(orr, orr); update_result->our_datalen += orr->orr_size; +out_free: + for (i = 0; i < nbufs; i++) { + if (rdbuf->rb_bufs[i].lb_buf != NULL) { + OBD_FREE(rdbuf->rb_bufs[i].lb_buf, + rdbuf->rb_bufs[i].lb_len); + } + } + OBD_FREE(rdbuf, sizeof(*rdbuf) + + nbufs * sizeof(rdbuf->rb_bufs[0])); out: /* Insert read buffer */ update_result->our_rc = ptlrpc_status_hton(rc); @@ -682,33 +787,35 @@ static int out_noop(struct tgt_session_info *tsi) } static struct tgt_handler out_update_ops[] = { - DEF_OUT_HNDL(OUT_CREATE, "out_create", MUTABOR | HABEO_REFERO, + DEF_OUT_HNDL(OUT_CREATE, "out_create", IS_MUTABLE | HAS_REPLY, out_create), - DEF_OUT_HNDL(OUT_DESTROY, "out_create", MUTABOR | HABEO_REFERO, + DEF_OUT_HNDL(OUT_DESTROY, "out_create", IS_MUTABLE | HAS_REPLY, out_destroy), - DEF_OUT_HNDL(OUT_REF_ADD, "out_ref_add", MUTABOR | HABEO_REFERO, + DEF_OUT_HNDL(OUT_REF_ADD, "out_ref_add", IS_MUTABLE | HAS_REPLY, out_ref_add), - DEF_OUT_HNDL(OUT_REF_DEL, "out_ref_del", MUTABOR | HABEO_REFERO, + DEF_OUT_HNDL(OUT_REF_DEL, "out_ref_del", IS_MUTABLE | HAS_REPLY, out_ref_del), - DEF_OUT_HNDL(OUT_ATTR_SET, "out_attr_set", MUTABOR | HABEO_REFERO, + DEF_OUT_HNDL(OUT_ATTR_SET, "out_attr_set", IS_MUTABLE | HAS_REPLY, out_attr_set), - DEF_OUT_HNDL(OUT_ATTR_GET, "out_attr_get", HABEO_REFERO, + DEF_OUT_HNDL(OUT_ATTR_GET, "out_attr_get", HAS_REPLY, out_attr_get), - DEF_OUT_HNDL(OUT_XATTR_SET, "out_xattr_set", MUTABOR | HABEO_REFERO, + DEF_OUT_HNDL(OUT_XATTR_SET, "out_xattr_set", IS_MUTABLE | HAS_REPLY, out_xattr_set), - DEF_OUT_HNDL(OUT_XATTR_DEL, "out_xattr_del", MUTABOR | HABEO_REFERO, + DEF_OUT_HNDL(OUT_XATTR_DEL, "out_xattr_del", IS_MUTABLE | HAS_REPLY, out_xattr_del), - DEF_OUT_HNDL(OUT_XATTR_GET, "out_xattr_get", HABEO_REFERO, + DEF_OUT_HNDL(OUT_XATTR_GET, "out_xattr_get", HAS_REPLY, out_xattr_get), - DEF_OUT_HNDL(OUT_INDEX_LOOKUP, "out_index_lookup", HABEO_REFERO, + DEF_OUT_HNDL(OUT_INDEX_LOOKUP, "out_index_lookup", HAS_REPLY, out_index_lookup), DEF_OUT_HNDL(OUT_INDEX_INSERT, "out_index_insert", - MUTABOR | HABEO_REFERO, out_index_insert), + IS_MUTABLE | HAS_REPLY, out_index_insert), DEF_OUT_HNDL(OUT_INDEX_DELETE, "out_index_delete", - MUTABOR | HABEO_REFERO, out_index_delete), - DEF_OUT_HNDL(OUT_WRITE, "out_write", MUTABOR | HABEO_REFERO, out_write), - DEF_OUT_HNDL(OUT_READ, "out_read", HABEO_REFERO, out_read), - DEF_OUT_HNDL(OUT_NOOP, "out_noop", HABEO_REFERO, out_noop), + IS_MUTABLE | HAS_REPLY, out_index_delete), + DEF_OUT_HNDL(OUT_WRITE, "out_write", IS_MUTABLE | HAS_REPLY, out_write), + DEF_OUT_HNDL(OUT_READ, "out_read", HAS_REPLY, out_read), + DEF_OUT_HNDL(OUT_NOOP, "out_noop", HAS_REPLY, out_noop), + DEF_OUT_HNDL(OUT_XATTR_LIST, "out_xattr_list", HAS_REPLY, + out_xattr_list), }; static struct tgt_handler *out_handler_find(__u32 opc) @@ -762,17 +869,7 @@ static int out_trans_stop(const struct lu_env *env, rc = dt_trans_stop(env, ta->ta_handle->th_dev, ta->ta_handle); for (i = 0; i < ta->ta_argno; i++) { if (ta->ta_args[i]->object != NULL) { - struct dt_object *obj = ta->ta_args[i]->object; - - /* If the object is being created during this - * transaction, we need to remove them from the - * cache immediately, because a few layers are - * missing in OUT handler, i.e. the object might - * not be initialized in all layers */ - if (ta->ta_args[i]->exec_fn == out_tx_create_exec) - set_bit(LU_OBJECT_HEARD_BANSHEE, - &obj->do_lu.lo_header->loh_flags); - lu_object_put(env, &ta->ta_args[i]->object->do_lu); + dt_object_put(env, ta->ta_args[i]->object); ta->ta_args[i]->object = NULL; } } @@ -867,7 +964,6 @@ int out_handle(struct tgt_session_info *tsi) struct object_update *update; struct object_update_reply *reply; struct ptlrpc_bulk_desc *desc = NULL; - struct l_wait_info lwi; void **update_bufs; int current_batchid = -1; __u32 update_buf_count; @@ -875,7 +971,8 @@ int out_handle(struct tgt_session_info *tsi) unsigned int reply_index = 0; int rc = 0; int rc1 = 0; - int ouh_size; + int ouh_size, reply_size; + int updates; ENTRY; req_capsule_set(pill, &RQF_OUT_UPDATE); @@ -899,45 +996,41 @@ int out_handle(struct tgt_session_info *tsi) if (update_buf_count == 0) RETURN(err_serious(-EPROTO)); - req_capsule_set_size(pill, &RMF_OUT_UPDATE_REPLY, RCL_SERVER, - OUT_UPDATE_REPLY_SIZE); - rc = req_capsule_server_pack(pill); - if (rc != 0) { - CERROR("%s: Can't pack response: rc = %d\n", - tgt_name(tsi->tsi_tgt), rc); - RETURN(rc); - } - OBD_ALLOC(update_bufs, sizeof(*update_bufs) * update_buf_count); if (update_bufs == NULL) - RETURN(-ENOMEM); + RETURN(err_serious(-ENOMEM)); if (ouh->ouh_inline_length > 0) { update_bufs[0] = ouh->ouh_inline_data; } else { struct out_update_buffer *tmp; + int page_count = 0; oub = req_capsule_client_get(pill, &RMF_OUT_UPDATE_BUF); if (oub == NULL) - GOTO(out_free, rc = -EPROTO); + GOTO(out_free, rc = err_serious(-EPROTO)); - desc = ptlrpc_prep_bulk_exp(pill->rc_req, update_buf_count, + for (i = 0; i < update_buf_count; i++) + /* First *and* last might be partial pages, hence +1 */ + page_count += DIV_ROUND_UP(oub[i].oub_size, + PAGE_SIZE) + 1; + + desc = ptlrpc_prep_bulk_exp(pill->rc_req, page_count, PTLRPC_BULK_OPS_COUNT, - PTLRPC_BULK_GET_SINK | - PTLRPC_BULK_BUF_KVEC, + PTLRPC_BULK_GET_SINK, MDS_BULK_PORTAL, - &ptlrpc_bulk_kvec_ops); + &ptlrpc_bulk_kiov_nopin_ops); if (desc == NULL) - GOTO(out_free, rc = -ENOMEM); + GOTO(out_free, rc = err_serious(-ENOMEM)); tmp = oub; for (i = 0; i < update_buf_count; i++, tmp++) { if (tmp->oub_size >= OUT_MAXREQSIZE) - GOTO(out_free, rc = -EPROTO); + GOTO(out_free, rc = err_serious(-EPROTO)); - OBD_ALLOC(update_bufs[i], tmp->oub_size); + OBD_ALLOC_LARGE(update_bufs[i], tmp->oub_size); if (update_bufs[i] == NULL) - GOTO(out_free, rc = -ENOMEM); + GOTO(out_free, rc = err_serious(-ENOMEM)); desc->bd_frag_ops->add_iov_frag(desc, update_bufs[i], tmp->oub_size); @@ -946,41 +1039,82 @@ int out_handle(struct tgt_session_info *tsi) pill->rc_req->rq_bulk_write = 1; rc = sptlrpc_svc_prep_bulk(pill->rc_req, desc); if (rc != 0) - GOTO(out_free, rc); + GOTO(out_free, rc = err_serious(rc)); - rc = target_bulk_io(pill->rc_req->rq_export, desc, &lwi); + rc = target_bulk_io(pill->rc_req->rq_export, desc); if (rc < 0) - GOTO(out_free, rc); + GOTO(out_free, rc = err_serious(rc)); } - /* Prepare the update reply buffer */ - reply = req_capsule_server_get(pill, &RMF_OUT_UPDATE_REPLY); - if (reply == NULL) - GOTO(out_free, rc = err_serious(-EPROTO)); - reply->ourp_magic = UPDATE_REPLY_MAGIC; - tti->tti_u.update.tti_update_reply = reply; - tti->tti_mult_trans = !req_is_replay(tgt_ses_req(tsi)); - /* validate the request and calculate the total update count and * set it to reply */ + reply_size = 0; + updates = 0; for (i = 0; i < update_buf_count; i++) { - struct object_update_request *our; - int update_count; + struct object_update_request *our; + int j; our = update_bufs[i]; if (ptlrpc_req_need_swab(pill->rc_req)) - lustre_swab_object_update_request(our); + lustre_swab_object_update_request(our, 0); if (our->ourq_magic != UPDATE_REQUEST_MAGIC) { CERROR("%s: invalid update buffer magic %x" " expect %x: rc = %d\n", tgt_name(tsi->tsi_tgt), our->ourq_magic, UPDATE_REQUEST_MAGIC, -EPROTO); - GOTO(out_free, rc = -EPROTO); + GOTO(out_free, rc = err_serious(-EPROTO)); + } + updates += our->ourq_count; + + /* need to calculate reply size */ + for (j = 0; j < our->ourq_count; j++) { + update = object_update_request_get(our, j, NULL); + if (update == NULL) + GOTO(out, rc = err_serious(-EPROTO)); + if (ptlrpc_req_need_swab(pill->rc_req)) + lustre_swab_object_update(update); + + if (!fid_is_sane(&update->ou_fid)) { + CERROR("%s: invalid FID "DFID": rc = %d\n", + tgt_name(tsi->tsi_tgt), + PFID(&update->ou_fid), -EPROTO); + GOTO(out, rc = err_serious(-EPROTO)); + } + + /* XXX: what ou_result_size can be considered safe? */ + + reply_size += sizeof(reply->ourp_lens[0]); + reply_size += sizeof(struct object_update_result); + reply_size += update->ou_result_size; } - update_count = our->ourq_count; - reply->ourp_count += update_count; } - + reply_size += sizeof(*reply); + + if (unlikely(reply_size > ouh->ouh_reply_size)) { + CERROR("%s: too small reply buf %u for %u, need %u at least\n", + tgt_name(tsi->tsi_tgt), ouh->ouh_reply_size, + updates, reply_size); + GOTO(out_free, rc = err_serious(-EPROTO)); + } + + req_capsule_set_size(pill, &RMF_OUT_UPDATE_REPLY, RCL_SERVER, + ouh->ouh_reply_size); + rc = req_capsule_server_pack(pill); + if (rc != 0) { + CERROR("%s: Can't pack response: rc = %d\n", + tgt_name(tsi->tsi_tgt), rc); + GOTO(out_free, rc = err_serious(-EPROTO)); + } + + /* Prepare the update reply buffer */ + reply = req_capsule_server_get(pill, &RMF_OUT_UPDATE_REPLY); + if (reply == NULL) + GOTO(out_free, rc = -EPROTO); + reply->ourp_magic = UPDATE_REPLY_MAGIC; + reply->ourp_count = updates; + tti->tti_u.update.tti_update_reply = reply; + tti->tti_mult_trans = !req_is_replay(tgt_ses_req(tsi)); + /* Walk through updates in the request to execute them */ for (i = 0; i < update_buf_count; i++) { struct tgt_handler *h; @@ -992,31 +1126,27 @@ int out_handle(struct tgt_session_info *tsi) our = update_bufs[i]; update_count = our->ourq_count; for (j = 0; j < update_count; j++) { - update = object_update_request_get(our, j, NULL); - if (update == NULL) - GOTO(out, rc = -EPROTO); - - if (ptlrpc_req_need_swab(pill->rc_req)) - lustre_swab_object_update(update); + struct lu_object_conf conf; - if (!fid_is_sane(&update->ou_fid)) { - CERROR("%s: invalid FID "DFID": rc = %d\n", - tgt_name(tsi->tsi_tgt), - PFID(&update->ou_fid), -EPROTO); - GOTO(out, rc = err_serious(-EPROTO)); - } + update = object_update_request_get(our, j, NULL); + if (update->ou_type == OUT_CREATE) + conf.loc_flags = LOC_F_NEW; + else + conf.loc_flags = 0; - dt_obj = dt_locate(env, dt, &update->ou_fid); + dt_obj = dt_locate_at(env, dt, &update->ou_fid, + dt->dd_lu_dev.ld_site->ls_top_dev, &conf); if (IS_ERR(dt_obj)) GOTO(out, rc = PTR_ERR(dt_obj)); if (dt->dd_record_fid_accessed) { - lfsck_pack_rfa(&tti->tti_lr, + struct lfsck_req_local *lrl = &tti->tti_lrl; + + lfsck_pack_rfa(lrl, lu_object_fid(&dt_obj->do_lu), - LE_FID_ACCESSED, + LEL_FID_ACCESSED, LFSCK_TYPE_LAYOUT); - tgt_lfsck_in_notify(env, dt, &tti->tti_lr, - NULL); + tgt_lfsck_in_notify_local(env, dt, lrl, NULL); } tti->tti_u.update.tti_dt_object = dt_obj; @@ -1031,17 +1161,20 @@ int out_handle(struct tgt_session_info *tsi) } /* Check resend case only for modifying RPC */ - if (h->th_flags & MUTABOR) { + if (h->th_flags & IS_MUTABLE) { struct ptlrpc_request *req = tgt_ses_req(tsi); if (out_check_resent(env, dt, dt_obj, req, out_reconstruct, reply, reply_index)) GOTO(next, rc = 0); + + if (dt->dd_rdonly) + GOTO(next, rc = -EROFS); } /* start transaction for modification RPC only */ - if (h->th_flags & MUTABOR && current_batchid == -1) { + if (h->th_flags & IS_MUTABLE && current_batchid == -1) { current_batchid = update->ou_batchid; rc = out_tx_start(env, dt, ta, tsi->tsi_exp); if (rc != 0) @@ -1054,7 +1187,7 @@ int out_handle(struct tgt_session_info *tsi) /* Stop the current update transaction, if the update * has different batchid, or read-only update */ if (((current_batchid != update->ou_batchid) || - !(h->th_flags & MUTABOR)) && + !(h->th_flags & IS_MUTABLE)) && ta->ta_handle != NULL) { rc = out_tx_end(env, ta, rc); current_batchid = -1; @@ -1062,7 +1195,7 @@ int out_handle(struct tgt_session_info *tsi) GOTO(next, rc); /* start a new transaction if needed */ - if (h->th_flags & MUTABOR) { + if (h->th_flags & IS_MUTABLE) { rc = out_tx_start(env, dt, ta, tsi->tsi_exp); if (rc != 0) @@ -1076,7 +1209,7 @@ int out_handle(struct tgt_session_info *tsi) rc = h->th_act(tsi); next: reply_index++; - lu_object_put(env, &dt_obj->do_lu); + dt_object_put(env, dt_obj); if (rc < 0) GOTO(out, rc); } @@ -1093,7 +1226,8 @@ out_free: if (oub != NULL) { for (i = 0; i < update_buf_count; i++, oub++) { if (update_bufs[i] != NULL) - OBD_FREE(update_bufs[i], oub->oub_size); + OBD_FREE_LARGE(update_bufs[i], + oub->oub_size); } } @@ -1107,7 +1241,7 @@ out_free: } struct tgt_handler tgt_out_handlers[] = { -TGT_UPDATE_HDL(MUTABOR, OUT_UPDATE, out_handle), +TGT_UPDATE_HDL(IS_MUTABLE, OUT_UPDATE, out_handle), }; EXPORT_SYMBOL(tgt_out_handlers);