X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fobdecho%2Fecho_client.c;h=bbb5f52ee78618306ab52d898518b68146578e80;hp=3f54f2ed7161a6de9ba68929301d294d50eb8384;hb=3e2d4d2d7477dd338a87de103027d37865285e44;hpb=7877f9ebf7cde785319f92e4a1fd033def11f23a diff --git a/lustre/obdecho/echo_client.c b/lustre/obdecho/echo_client.c index 3f54f2e..bbb5f52 100644 --- a/lustre/obdecho/echo_client.c +++ b/lustre/obdecho/echo_client.c @@ -1,6 +1,4 @@ -/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*- - * vim:expandtab:shiftwidth=8:tabstop=8: - * +/* * GPL HEADER START * * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. @@ -26,8 +24,10 @@ * GPL HEADER END */ /* - * Copyright 2008 Sun Microsystems, Inc. All rights reserved + * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. + * + * Copyright (c) 2011, 2013, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -47,6 +47,10 @@ #include #include #include +#include +#include +#include +#include #include "echo_internal.h" @@ -62,6 +66,8 @@ struct echo_device { struct cl_site *ed_site; struct lu_device *ed_next; int ed_next_islov; + int ed_next_ismd; + struct lu_client_seq *ed_cl_seq; }; struct echo_object { @@ -69,9 +75,9 @@ struct echo_object { struct cl_object_header eo_hdr; struct echo_device *eo_dev; - struct list_head eo_obj_chain; + cfs_list_t eo_obj_chain; struct lov_stripe_md *eo_lsm; - atomic_t eo_npages; + cfs_atomic_t eo_npages; int eo_deleted; }; @@ -82,15 +88,16 @@ struct echo_object_conf { struct echo_page { struct cl_page_slice ep_cl; + struct mutex ep_lock; cfs_page_t *ep_vmpage; }; struct echo_lock { struct cl_lock_slice el_cl; - struct list_head el_chain; + cfs_list_t el_chain; struct echo_object *el_object; __u64 el_cookie; - atomic_t el_refcount; + cfs_atomic_t el_refcount; }; struct echo_io { @@ -103,7 +110,8 @@ struct echo_req { }; #endif -static int echo_client_setup(struct obd_device *obddev, +static int echo_client_setup(const struct lu_env *env, + struct obd_device *obddev, struct lustre_cfg *lcfg); static int echo_client_cleanup(struct obd_device *obddev); @@ -166,12 +174,6 @@ struct echo_object_conf *cl2echo_conf(const struct cl_object_conf *c) return container_of(c, struct echo_object_conf, eoc_cl); } -static inline void lsm2fid(struct lov_stripe_md *lsm, struct lu_fid *fid) -{ - fid_zero(fid); - fid->f_seq = lsm->lsm_object_gr << 16 | lsm->lsm_object_id >> 32; - fid->f_oid = lsm->lsm_object_id; -} /** @} echo_helpers */ static struct echo_object *cl_echo_object_find(struct echo_device *d, @@ -193,6 +195,18 @@ struct echo_thread_info { struct cl_io eti_io; struct cl_lock_descr eti_descr; struct lu_fid eti_fid; + struct lu_fid eti_fid2; + struct md_op_spec eti_spec; + struct lov_mds_md_v3 eti_lmm; + struct lov_user_md_v3 eti_lum; + struct md_attr eti_ma; + struct lu_name eti_lname; + /* per-thread values, can be re-used */ + void *eti_big_lmm; + int eti_big_lmmsize; + char eti_name[20]; + struct lu_buf eti_buf; + char eti_xattr_buf[LUSTRE_POSIX_ACL_MAX_SIZE]; }; /* No session used right now */ @@ -200,7 +214,6 @@ struct echo_session_info { unsigned long dummy; }; -static cfs_mem_cache_t *echo_page_kmem; static cfs_mem_cache_t *echo_lock_kmem; static cfs_mem_cache_t *echo_object_kmem; static cfs_mem_cache_t *echo_thread_kmem; @@ -209,11 +222,6 @@ static cfs_mem_cache_t *echo_session_kmem; static struct lu_kmem_descr echo_caches[] = { { - .ckd_cache = &echo_page_kmem, - .ckd_name = "echo_page_kmem", - .ckd_size = sizeof (struct echo_page) - }, - { .ckd_cache = &echo_lock_kmem, .ckd_name = "echo_lock_kmem", .ckd_size = sizeof (struct echo_lock) @@ -251,12 +259,35 @@ static struct lu_kmem_descr echo_caches[] = { * * @{ */ -cfs_page_t *echo_page_vmpage(const struct lu_env *env, - const struct cl_page_slice *slice) +static cfs_page_t *echo_page_vmpage(const struct lu_env *env, + const struct cl_page_slice *slice) { return cl2echo_page(slice)->ep_vmpage; } +static int echo_page_own(const struct lu_env *env, + const struct cl_page_slice *slice, + struct cl_io *io, int nonblock) +{ + struct echo_page *ep = cl2echo_page(slice); + + if (!nonblock) + mutex_lock(&ep->ep_lock); + else if (!mutex_trylock(&ep->ep_lock)) + return -EAGAIN; + return 0; +} + +static void echo_page_disown(const struct lu_env *env, + const struct cl_page_slice *slice, + struct cl_io *io) +{ + struct echo_page *ep = cl2echo_page(slice); + + LASSERT(mutex_is_locked(&ep->ep_lock)); + mutex_unlock(&ep->ep_lock); +} + static void echo_page_discard(const struct lu_env *env, const struct cl_page_slice *slice, struct cl_io *unused) @@ -267,7 +298,9 @@ static void echo_page_discard(const struct lu_env *env, static int echo_page_is_vmlocked(const struct lu_env *env, const struct cl_page_slice *slice) { - return 1; + if (mutex_is_locked(&cl2echo_page(slice)->ep_lock)) + return -EBUSY; + return -ENODATA; } static void echo_page_completion(const struct lu_env *env, @@ -285,9 +318,8 @@ static void echo_page_fini(const struct lu_env *env, cfs_page_t *vmpage = ep->ep_vmpage; ENTRY; - atomic_dec(&eco->eo_npages); + cfs_atomic_dec(&eco->eo_npages); page_cache_release(vmpage); - OBD_SLAB_FREE_PTR(ep, echo_page_kmem); EXIT; } @@ -304,12 +336,14 @@ static int echo_page_print(const struct lu_env *env, { struct echo_page *ep = cl2echo_page(slice); - (*printer)(env, cookie, LUSTRE_ECHO_CLIENT_NAME"-page@%p vm@%p\n", - ep, ep->ep_vmpage); + (*printer)(env, cookie, LUSTRE_ECHO_CLIENT_NAME"-page@%p %d vm@%p\n", + ep, mutex_is_locked(&ep->ep_lock), ep->ep_vmpage); return 0; } static const struct cl_page_operations echo_page_ops = { + .cpo_own = echo_page_own, + .cpo_disown = echo_page_disown, .cpo_discard = echo_page_discard, .cpo_vmpage = echo_page_vmpage, .cpo_fini = echo_page_fini, @@ -339,7 +373,7 @@ static void echo_lock_fini(const struct lu_env *env, { struct echo_lock *ecl = cl2echo_lock(slice); - LASSERT(list_empty(&ecl->el_chain)); + LASSERT(cfs_list_empty(&ecl->el_chain)); OBD_SLAB_FREE_PTR(ecl, echo_lock_kmem); } @@ -348,7 +382,7 @@ static void echo_lock_delete(const struct lu_env *env, { struct echo_lock *ecl = cl2echo_lock(slice); - LASSERT(list_empty(&ecl->el_chain)); + LASSERT(cfs_list_empty(&ecl->el_chain)); } static int echo_lock_fits_into(const struct lu_env *env, @@ -373,22 +407,19 @@ static struct cl_lock_operations echo_lock_ops = { * * @{ */ -static struct cl_page *echo_page_init(const struct lu_env *env, - struct cl_object *obj, - struct cl_page *page, cfs_page_t *vmpage) +static int echo_page_init(const struct lu_env *env, struct cl_object *obj, + struct cl_page *page, cfs_page_t *vmpage) { - struct echo_page *ep; + struct echo_page *ep = cl_object_page_slice(obj, page); + struct echo_object *eco = cl2echo_obj(obj); ENTRY; - OBD_SLAB_ALLOC_PTR_GFP(ep, echo_page_kmem, CFS_ALLOC_IO); - if (ep != NULL) { - struct echo_object *eco = cl2echo_obj(obj); - ep->ep_vmpage = vmpage; - page_cache_get(vmpage); - cl_page_slice_add(page, &ep->ep_cl, obj, &echo_page_ops); - atomic_inc(&eco->eo_npages); - } - RETURN(ERR_PTR(ep ? 0 : -ENOMEM)); + ep->ep_vmpage = vmpage; + page_cache_get(vmpage); + mutex_init(&ep->ep_lock); + cl_page_slice_add(page, &ep->ep_cl, obj, &echo_page_ops); + cfs_atomic_inc(&eco->eo_npages); + RETURN(0); } static int echo_io_init(const struct lu_env *env, struct cl_object *obj, @@ -409,7 +440,7 @@ static int echo_lock_init(const struct lu_env *env, cl_lock_slice_add(lock, &el->el_cl, obj, &echo_lock_ops); el->el_object = cl2echo_obj(obj); CFS_INIT_LIST_HEAD(&el->el_chain); - atomic_set(&el->el_refcount, 0); + cfs_atomic_set(&el->el_refcount, 0); } RETURN(el == NULL ? -ENOMEM : 0); } @@ -437,8 +468,6 @@ static const struct cl_object_operations echo_cl_obj_ops = { static int echo_object_init(const struct lu_env *env, struct lu_object *obj, const struct lu_object_conf *conf) { - const struct cl_object_conf *cconf = lu2cl_conf(conf); - struct echo_object_conf *econf = cl2echo_conf(cconf); struct echo_device *ed = cl2echo_dev(lu2cl_dev(obj->lo_dev)); struct echo_client_obd *ec = ed->ed_ec; struct echo_object *eco = cl2echo_obj(lu2cl(obj)); @@ -456,41 +485,99 @@ static int echo_object_init(const struct lu_env *env, struct lu_object *obj, lu_object_add(obj, below); } - LASSERT(econf->eoc_md); - eco->eo_lsm = *econf->eoc_md; + if (!ed->ed_next_ismd) { + const struct cl_object_conf *cconf = lu2cl_conf(conf); + struct echo_object_conf *econf = cl2echo_conf(cconf); + + LASSERT(econf->eoc_md); + eco->eo_lsm = *econf->eoc_md; + /* clear the lsm pointer so that it won't get freed. */ + *econf->eoc_md = NULL; + } else { + eco->eo_lsm = NULL; + } + eco->eo_dev = ed; - atomic_set(&eco->eo_npages, 0); + cfs_atomic_set(&eco->eo_npages, 0); + cl_object_page_init(lu2cl(obj), sizeof(struct echo_page)); - /* clear the lsm pointer so that it won't get freed. */ - *econf->eoc_md = NULL; + spin_lock(&ec->ec_lock); + cfs_list_add_tail(&eco->eo_obj_chain, &ec->ec_objects); + spin_unlock(&ec->ec_lock); - spin_lock(&ec->ec_lock); - list_add_tail(&eco->eo_obj_chain, &ec->ec_objects); - spin_unlock(&ec->ec_lock); + RETURN(0); +} - RETURN(0); +/* taken from osc_unpackmd() */ +static int echo_alloc_memmd(struct echo_device *ed, + struct lov_stripe_md **lsmp) +{ + int lsm_size; + + ENTRY; + + /* If export is lov/osc then use their obd method */ + if (ed->ed_next != NULL) + return obd_alloc_memmd(ed->ed_ec->ec_exp, lsmp); + /* OFD has no unpackmd method, do everything here */ + lsm_size = lov_stripe_md_size(1); + + LASSERT(*lsmp == NULL); + OBD_ALLOC(*lsmp, lsm_size); + if (*lsmp == NULL) + RETURN(-ENOMEM); + + OBD_ALLOC((*lsmp)->lsm_oinfo[0], sizeof(struct lov_oinfo)); + if ((*lsmp)->lsm_oinfo[0] == NULL) { + OBD_FREE(*lsmp, lsm_size); + RETURN(-ENOMEM); + } + + loi_init((*lsmp)->lsm_oinfo[0]); + (*lsmp)->lsm_maxbytes = LUSTRE_STRIPE_MAXBYTES; + ostid_set_seq_echo(&(*lsmp)->lsm_oi); + + RETURN(lsm_size); +} + +static int echo_free_memmd(struct echo_device *ed, struct lov_stripe_md **lsmp) +{ + int lsm_size; + + ENTRY; + + /* If export is lov/osc then use their obd method */ + if (ed->ed_next != NULL) + return obd_free_memmd(ed->ed_ec->ec_exp, lsmp); + /* OFD has no unpackmd method, do everything here */ + lsm_size = lov_stripe_md_size(1); + + LASSERT(*lsmp != NULL); + OBD_FREE((*lsmp)->lsm_oinfo[0], sizeof(struct lov_oinfo)); + OBD_FREE(*lsmp, lsm_size); + *lsmp = NULL; + RETURN(0); } static void echo_object_free(const struct lu_env *env, struct lu_object *obj) { struct echo_object *eco = cl2echo_obj(lu2cl(obj)); struct echo_client_obd *ec = eco->eo_dev->ed_ec; - struct lov_stripe_md *lsm = eco->eo_lsm; ENTRY; - LASSERT(atomic_read(&eco->eo_npages) == 0); + LASSERT(cfs_atomic_read(&eco->eo_npages) == 0); - spin_lock(&ec->ec_lock); - list_del_init(&eco->eo_obj_chain); - spin_unlock(&ec->ec_lock); + spin_lock(&ec->ec_lock); + cfs_list_del_init(&eco->eo_obj_chain); + spin_unlock(&ec->ec_lock); lu_object_fini(obj); lu_object_header_fini(obj->lo_header); - if (lsm) - obd_free_memmd(ec->ec_exp, &lsm); - OBD_SLAB_FREE_PTR(eco, echo_object_kmem); - EXIT; + if (eco->eo_lsm) + echo_free_memmd(eco->eo_dev, &eco->eo_lsm); + OBD_SLAB_FREE_PTR(eco, echo_object_kmem); + EXIT; } static int echo_object_print(const struct lu_env *env, void *cookie, @@ -501,7 +588,6 @@ static int echo_object_print(const struct lu_env *env, void *cookie, return (*p)(env, cookie, "echoclient-object@%p", obj); } - static const struct lu_object_operations echo_lu_obj_ops = { .loo_object_init = echo_object_init, .loo_object_delete = NULL, @@ -546,6 +632,7 @@ static struct lu_object *echo_object_alloc(const struct lu_env *env, static struct lu_device_operations echo_device_lu_ops = { .ldo_object_alloc = echo_object_alloc, }; + /** @} echo_lu_dev_ops */ static struct cl_device_operations echo_device_cl_ops = { @@ -580,7 +667,8 @@ static int echo_site_init(const struct lu_env *env, struct echo_device *ed) static void echo_site_fini(const struct lu_env *env, struct echo_device *ed) { if (ed->ed_site) { - cl_site_fini(ed->ed_site); + if (!ed->ed_next_ismd) + cl_site_fini(ed->ed_site); ed->ed_site = NULL; } } @@ -647,6 +735,55 @@ static struct lu_context_key echo_session_key = { LU_TYPE_INIT_FINI(echo, &echo_thread_key, &echo_session_key); +#define ECHO_SEQ_WIDTH 0xffffffff +static int echo_fid_init(struct echo_device *ed, char *obd_name, + struct seq_server_site *ss) +{ + char *prefix; + int rc; + ENTRY; + + OBD_ALLOC_PTR(ed->ed_cl_seq); + if (ed->ed_cl_seq == NULL) + RETURN(-ENOMEM); + + OBD_ALLOC(prefix, MAX_OBD_NAME + 5); + if (prefix == NULL) + GOTO(out_free_seq, rc = -ENOMEM); + + snprintf(prefix, MAX_OBD_NAME + 5, "srv-%s", obd_name); + + /* Init client side sequence-manager */ + rc = seq_client_init(ed->ed_cl_seq, NULL, + LUSTRE_SEQ_METADATA, + prefix, ss->ss_server_seq); + ed->ed_cl_seq->lcs_width = ECHO_SEQ_WIDTH; + OBD_FREE(prefix, MAX_OBD_NAME + 5); + if (rc) + GOTO(out_free_seq, rc); + + RETURN(0); + +out_free_seq: + OBD_FREE_PTR(ed->ed_cl_seq); + ed->ed_cl_seq = NULL; + RETURN(rc); +} + +static int echo_fid_fini(struct obd_device *obddev) +{ + struct echo_device *ed = obd2echo_dev(obddev); + ENTRY; + + if (ed->ed_cl_seq != NULL) { + seq_client_fini(ed->ed_cl_seq); + OBD_FREE_PTR(ed->ed_cl_seq); + ed->ed_cl_seq = NULL; + } + + RETURN(0); +} + static struct lu_device *echo_device_alloc(const struct lu_env *env, struct lu_device_type *t, struct lustre_cfg *cfg) @@ -675,55 +812,115 @@ static struct lu_device *echo_device_alloc(const struct lu_env *env, cd->cd_ops = &echo_device_cl_ops; cleanup = 2; - rc = echo_site_init(env, ed); - if (rc) - GOTO(out, rc); - - cleanup = 3; obd = class_name2obd(lustre_cfg_string(cfg, 0)); LASSERT(obd != NULL); - rc = echo_client_setup(obd, cfg); - if (rc) - GOTO(out, rc); - ed->ed_ec = &obd->u.echo_client; + LASSERT(env != NULL); - cleanup = 4; tgt = class_name2obd(lustre_cfg_string(cfg, 1)); - LASSERT(tgt != NULL); - next = tgt->obd_lu_dev; - if (next != NULL && !lu_device_is_cl(next)) - next = NULL; + if (tgt == NULL) { + CERROR("Can not find tgt device %s\n", + lustre_cfg_string(cfg, 1)); + GOTO(out, rc = -ENODEV); + } - /* - * if echo client is to be stacked upon ost device, the next is NULL - * since ost is not a clio device so far - */ - tgt_type_name = tgt->obd_type->typ_name; - if (next != NULL) { - LASSERT(next != NULL); - if (next->ld_site != NULL) - GOTO(out, rc = -EBUSY); - - next->ld_site = &ed->ed_site->cs_lu; - rc = next->ld_type->ldt_ops->ldto_device_init(env, next, - next->ld_type->ldt_name, NULL); + next = tgt->obd_lu_dev; + if (!strcmp(tgt->obd_type->typ_name, LUSTRE_MDT_NAME)) { + ed->ed_next_ismd = 1; + } else { + ed->ed_next_ismd = 0; + rc = echo_site_init(env, ed); if (rc) GOTO(out, rc); + } + cleanup = 3; - /* Trikcy case, I have to determine the obd type since clio - * uses the different parameters to initialize objects for - * lov & osc. - */ - if (strcmp(tgt_type_name, LUSTRE_LOV_NAME) == 0) - ed->ed_next_islov = 1; - else - LASSERT(strcmp(tgt_type_name, LUSTRE_OSC_NAME) == 0); - } else - LASSERT(strcmp(tgt_type_name, LUSTRE_OST_NAME) == 0); + rc = echo_client_setup(env, obd, cfg); + if (rc) + GOTO(out, rc); + + ed->ed_ec = &obd->u.echo_client; + cleanup = 4; + + if (ed->ed_next_ismd) { + /* Suppose to connect to some Metadata layer */ + struct lu_site *ls; + struct lu_device *ld; + int found = 0; + + if (next == NULL) { + CERROR("%s is not lu device type!\n", + lustre_cfg_string(cfg, 1)); + GOTO(out, rc = -EINVAL); + } + + tgt_type_name = lustre_cfg_string(cfg, 2); + if (!tgt_type_name) { + CERROR("%s no type name for echo %s setup\n", + lustre_cfg_string(cfg, 1), + tgt->obd_type->typ_name); + GOTO(out, rc = -EINVAL); + } + + ls = next->ld_site; + + spin_lock(&ls->ls_ld_lock); + cfs_list_for_each_entry(ld, &ls->ls_ld_linkage, ld_linkage) { + if (strcmp(ld->ld_type->ldt_name, tgt_type_name) == 0) { + found = 1; + break; + } + } + spin_unlock(&ls->ls_ld_lock); + + if (found == 0) { + CERROR("%s is not lu device type!\n", + lustre_cfg_string(cfg, 1)); + GOTO(out, rc = -EINVAL); + } + + next = ld; + /* For MD echo client, it will use the site in MDS stack */ + ed->ed_site_myself.cs_lu = *ls; + ed->ed_site = &ed->ed_site_myself; + ed->ed_cl.cd_lu_dev.ld_site = &ed->ed_site_myself.cs_lu; + rc = echo_fid_init(ed, obd->obd_name, lu_site2seq(ls)); + if (rc) { + CERROR("echo fid init error %d\n", rc); + GOTO(out, rc); + } + } else { + /* if echo client is to be stacked upon ost device, the next is + * NULL since ost is not a clio device so far */ + if (next != NULL && !lu_device_is_cl(next)) + next = NULL; + + tgt_type_name = tgt->obd_type->typ_name; + if (next != NULL) { + LASSERT(next != NULL); + if (next->ld_site != NULL) + GOTO(out, rc = -EBUSY); + + next->ld_site = &ed->ed_site->cs_lu; + rc = next->ld_type->ldt_ops->ldto_device_init(env, next, + next->ld_type->ldt_name, + NULL); + if (rc) + GOTO(out, rc); + + /* Tricky case, I have to determine the obd type since + * CLIO uses the different parameters to initialize + * objects for lov & osc. */ + if (strcmp(tgt_type_name, LUSTRE_LOV_NAME) == 0) + ed->ed_next_islov = 1; + else + LASSERT(strcmp(tgt_type_name, + LUSTRE_OSC_NAME) == 0); + } else + LASSERT(strcmp(tgt_type_name, LUSTRE_OST_NAME) == 0); + } ed->ed_next = next; RETURN(&cd->cd_lu_dev); - out: switch(cleanup) { case 4: { @@ -760,7 +957,7 @@ static struct lu_device *echo_device_fini(const struct lu_env *env, struct echo_device *ed = cl2echo_dev(lu2cl_dev(d)); struct lu_device *next = ed->ed_next; - while (next) + while (next && !ed->ed_next_ismd) next = next->ld_type->ldt_ops->ldto_device_fini(env, next); return NULL; } @@ -791,30 +988,9 @@ static struct lu_device *echo_device_free(const struct lu_env *env, struct echo_object *eco; struct lu_device *next = ed->ed_next; - CDEBUG(D_INFO, "echo device:%p is going to be freed, next = %p\n", ed, next); - - /* destroy locks */ - spin_lock(&ec->ec_lock); - while (!list_empty(&ec->ec_locks)) { - struct echo_lock *ecl = list_entry(ec->ec_locks.next, - struct echo_lock, el_chain); - int still_used = 0; - - if (atomic_dec_and_test(&ecl->el_refcount)) - list_del_init(&ecl->el_chain); - else - still_used = 1; - spin_unlock(&ec->ec_lock); - - CERROR("echo client: pending lock %p refs %d\n", - ecl, atomic_read(&ecl->el_refcount)); - - echo_lock_release(env, ecl, still_used); - spin_lock(&ec->ec_lock); - } - spin_unlock(&ec->ec_lock); + CDEBUG(D_INFO, "echo device:%p is going to be freed, next = %p\n", + ed, next); - LASSERT(ed->ed_site); lu_site_purge(env, &ed->ed_site->cs_lu, -1); /* check if there are objects still alive. @@ -822,32 +998,37 @@ static struct lu_device *echo_device_free(const struct lu_env *env, * all of cached objects. Anyway, probably the echo device is being * parallelly accessed. */ - spin_lock(&ec->ec_lock); - list_for_each_entry(eco, &ec->ec_objects, eo_obj_chain) - eco->eo_deleted = 1; - spin_unlock(&ec->ec_lock); - - /* purge again */ - lu_site_purge(env, &ed->ed_site->cs_lu, -1); - - CDEBUG(D_INFO, "Waiting for the reference of echo object to be dropped\n"); - - /* Wait for the last reference to be dropped. */ - spin_lock(&ec->ec_lock); - while (!list_empty(&ec->ec_objects)) { - spin_unlock(&ec->ec_lock); - CERROR("echo_client still has objects at cleanup time, " - "wait for 1 second\n"); - cfs_schedule_timeout(CFS_TASK_UNINT, cfs_time_seconds(1)); - spin_lock(&ec->ec_lock); - } - spin_unlock(&ec->ec_lock); + spin_lock(&ec->ec_lock); + cfs_list_for_each_entry(eco, &ec->ec_objects, eo_obj_chain) + eco->eo_deleted = 1; + spin_unlock(&ec->ec_lock); + + /* purge again */ + lu_site_purge(env, &ed->ed_site->cs_lu, -1); + + CDEBUG(D_INFO, + "Waiting for the reference of echo object to be dropped\n"); + + /* Wait for the last reference to be dropped. */ + spin_lock(&ec->ec_lock); + while (!cfs_list_empty(&ec->ec_objects)) { + spin_unlock(&ec->ec_lock); + CERROR("echo_client still has objects at cleanup time, " + "wait for 1 second\n"); + cfs_schedule_timeout_and_set_state(CFS_TASK_UNINT, + cfs_time_seconds(1)); + lu_site_purge(env, &ed->ed_site->cs_lu, -1); + spin_lock(&ec->ec_lock); + } + spin_unlock(&ec->ec_lock); + + LASSERT(cfs_list_empty(&ec->ec_locks)); CDEBUG(D_INFO, "No object exists, exiting...\n"); echo_client_cleanup(d->ld_obd); - - while (next) + echo_fid_fini(d->ld_obd); + while (next && !ed->ed_next_ismd) next = next->ld_type->ldt_ops->ldto_device_free(env, next); LASSERT(ed->ed_site == lu2cl_site(d->ld_site)); @@ -875,7 +1056,7 @@ static struct lu_device_type echo_device_type = { .ldt_tags = LU_DEVICE_CL, .ldt_name = LUSTRE_ECHO_CLIENT_NAME, .ldt_ops = &echo_device_type_ops, - .ldt_ctx_tags = LCT_CL_THREAD + .ldt_ctx_tags = LCT_CL_THREAD | LCT_MD_THREAD | LCT_DT_THREAD, }; /** @} echo_init */ @@ -898,12 +1079,15 @@ static struct echo_object *cl_echo_object_find(struct echo_device *d, struct cl_object *obj; struct lu_fid *fid; int refcheck; - ENTRY; + int rc; + ENTRY; - LASSERT(lsmp); - lsm = *lsmp; - LASSERT(lsm); - LASSERT(lsm->lsm_object_id); + LASSERT(lsmp); + lsm = *lsmp; + LASSERT(lsm); + LASSERTF(ostid_id(&lsm->lsm_oi) != 0, DOSTID"\n", POSTID(&lsm->lsm_oi)); + LASSERTF(ostid_seq(&lsm->lsm_oi) == FID_SEQ_ECHO, DOSTID"\n", + POSTID(&lsm->lsm_oi)); /* Never return an object if the obd is to be freed. */ if (echo_dev2cl(d)->cd_lu_dev.ld_obd->obd_stopping) @@ -919,8 +1103,7 @@ static struct echo_object *cl_echo_object_find(struct echo_device *d, if (!d->ed_next_islov) { struct lov_oinfo *oinfo = lsm->lsm_oinfo[0]; LASSERT(oinfo != NULL); - oinfo->loi_id = lsm->lsm_object_id; - oinfo->loi_gr = lsm->lsm_object_gr; + oinfo->loi_oi = lsm->lsm_oi; conf->eoc_cl.u.coc_oinfo = oinfo; } else { struct lustre_md *md; @@ -932,9 +1115,14 @@ static struct echo_object *cl_echo_object_find(struct echo_device *d, } conf->eoc_md = lsmp; - fid = &info->eti_fid; - lsm2fid(lsm, fid); + fid = &info->eti_fid; + rc = ostid_to_fid(fid, &lsm->lsm_oi, 0); + if (rc != 0) + GOTO(out, eco = ERR_PTR(rc)); + /* In the function below, .hs_keycmp resolves to + * lu_obj_hop_keycmp() */ + /* coverity[overrun-buffer-val] */ obj = cl_object_find(env, echo_dev2cl(d), fid, &conf->eoc_cl); if (IS_ERR(obj)) GOTO(out, eco = (void*)obj); @@ -965,8 +1153,7 @@ static int cl_echo_object_put(struct echo_object *eco) if (eco->eo_deleted) { struct lu_object_header *loh = obj->co_lu.lo_header; LASSERT(&eco->eo_hdr == luh2coh(loh)); - set_bit(LU_OBJECT_HEARD_BANSHEE, &loh->loh_flags); - cl_object_prune(env, obj); + set_bit(LU_OBJECT_HEARD_BANSHEE, &loh->loh_flags); } cl_object_put(env, obj); @@ -995,7 +1182,7 @@ static int cl_echo_enqueue0(struct lu_env *env, struct echo_object *eco, descr->cld_start = cl_index(obj, start); descr->cld_end = cl_index(obj, end); descr->cld_mode = mode == LCK_PW ? CLM_WRITE : CLM_READ; - descr->cld_enq_flags = CEF_ASYNC | enqflags; + descr->cld_enq_flags = enqflags; io->ci_obj = obj; lck = cl_lock_request(env, io, descr, "ec enqueue", eco); @@ -1006,18 +1193,19 @@ static int cl_echo_enqueue0(struct lu_env *env, struct echo_object *eco, rc = cl_wait(env, lck); if (rc == 0) { el = cl2echo_lock(cl_lock_at(lck, &echo_device_type)); - spin_lock(&ec->ec_lock); - if (list_empty(&el->el_chain)) { - list_add(&el->el_chain, &ec->ec_locks); - el->el_cookie = ++ec->ec_unique; - } - atomic_inc(&el->el_refcount); - *cookie = el->el_cookie; - spin_unlock(&ec->ec_lock); - } else - cl_lock_release(env, lck, "ec enqueue", cfs_current()); - } - RETURN(rc); + spin_lock(&ec->ec_lock); + if (cfs_list_empty(&el->el_chain)) { + cfs_list_add(&el->el_chain, &ec->ec_locks); + el->el_cookie = ++ec->ec_unique; + } + cfs_atomic_inc(&el->el_refcount); + *cookie = el->el_cookie; + spin_unlock(&ec->ec_lock); + } else { + cl_lock_release(env, lck, "ec enqueue", cfs_current()); + } + } + RETURN(rc); } static int cl_echo_enqueue(struct echo_object *eco, obd_off start, obd_off end, @@ -1037,6 +1225,7 @@ static int cl_echo_enqueue(struct echo_object *eco, obd_off start, obd_off end, info = echo_env_info(env); io = &info->eti_io; + io->ci_ignore_layout = 1; result = cl_io_init(env, io, CIT_MISC, echo_obj2cl(eco)); if (result < 0) GOTO(out, result); @@ -1056,25 +1245,25 @@ static int cl_echo_cancel0(struct lu_env *env, struct echo_device *ed, { struct echo_client_obd *ec = ed->ed_ec; struct echo_lock *ecl = NULL; - struct list_head *el; + cfs_list_t *el; int found = 0, still_used = 0; ENTRY; LASSERT(ec != NULL); - spin_lock (&ec->ec_lock); - list_for_each (el, &ec->ec_locks) { - ecl = list_entry (el, struct echo_lock, el_chain); - CDEBUG(D_INFO, "ecl: %p, cookie: %llx\n", ecl, ecl->el_cookie); + spin_lock(&ec->ec_lock); + cfs_list_for_each (el, &ec->ec_locks) { + ecl = cfs_list_entry (el, struct echo_lock, el_chain); + CDEBUG(D_INFO, "ecl: %p, cookie: "LPX64"\n", ecl, ecl->el_cookie); found = (ecl->el_cookie == cookie); if (found) { - if (atomic_dec_and_test(&ecl->el_refcount)) - list_del_init(&ecl->el_chain); + if (cfs_atomic_dec_and_test(&ecl->el_refcount)) + cfs_list_del_init(&ecl->el_chain); else still_used = 1; break; } } - spin_unlock (&ec->ec_lock); + spin_unlock(&ec->ec_lock); if (!found) RETURN(-ENOENT); @@ -1146,6 +1335,8 @@ static int cl_echo_object_brw(struct echo_object *eco, int rw, obd_off offset, queue = &info->eti_queue; cl_2queue_init(queue); + + io->ci_ignore_layout = 1; rc = cl_io_init(env, io, CIT_MISC, obj); if (rc < 0) GOTO(out, rc); @@ -1154,8 +1345,8 @@ static int cl_echo_object_brw(struct echo_object *eco, int rw, obd_off offset, rc = cl_echo_enqueue0(env, eco, offset, offset + npages * CFS_PAGE_SIZE - 1, - rw == READ ? LCK_PW : LCK_PW, &lh.cookie, - CILR_NEVER); + rw == READ ? LCK_PR : LCK_PW, &lh.cookie, + CEF_NEVER); if (rc < 0) GOTO(error_lock, rc); @@ -1193,8 +1384,7 @@ static int cl_echo_object_brw(struct echo_object *eco, int rw, obd_off offset, if (async) rc = cl_echo_async_brw(env, io, typ, queue); else - rc = cl_io_submit_sync(env, io, typ, queue, - CRP_NORMAL, 0); + rc = cl_io_submit_sync(env, io, typ, queue, 0); CDEBUG(D_INFO, "echo_client %s write returns %d\n", async ? "async" : "sync", rc); } @@ -1225,12 +1415,12 @@ echo_copyout_lsm (struct lov_stripe_md *lsm, void *_ulsm, int ulsm_nob) if (nob > ulsm_nob) return (-EINVAL); - if (copy_to_user (ulsm, lsm, sizeof(ulsm))) + if (cfs_copy_to_user (ulsm, lsm, sizeof(ulsm))) return (-EFAULT); for (i = 0; i < lsm->lsm_stripe_count; i++) { - if (copy_to_user (ulsm->lsm_oinfo[i], lsm->lsm_oinfo[i], - sizeof(lsm->lsm_oinfo[0]))) + if (cfs_copy_to_user (ulsm->lsm_oinfo[i], lsm->lsm_oinfo[i], + sizeof(lsm->lsm_oinfo[0]))) return (-EFAULT); } return 0; @@ -1246,7 +1436,7 @@ echo_copyin_lsm (struct echo_device *ed, struct lov_stripe_md *lsm, if (ulsm_nob < sizeof (*lsm)) return (-EINVAL); - if (copy_from_user (lsm, ulsm, sizeof (*lsm))) + if (cfs_copy_from_user (lsm, ulsm, sizeof (*lsm))) return (-EFAULT); if (lsm->lsm_stripe_count > ec->ec_nstripes || @@ -1257,17 +1447,800 @@ echo_copyin_lsm (struct echo_device *ed, struct lov_stripe_md *lsm, for (i = 0; i < lsm->lsm_stripe_count; i++) { - if (copy_from_user(lsm->lsm_oinfo[i], - ((struct lov_stripe_md *)ulsm)->lsm_oinfo[i], - sizeof(lsm->lsm_oinfo[0]))) + if (cfs_copy_from_user(lsm->lsm_oinfo[i], + ((struct lov_stripe_md *)ulsm)-> \ + lsm_oinfo[i], + sizeof(lsm->lsm_oinfo[0]))) return (-EFAULT); } return (0); } -static int echo_create_object(struct echo_device *ed, int on_target, - struct obdo *oa, void *ulsm, int ulsm_nob, - struct obd_trans_info *oti) +static inline void echo_md_build_name(struct lu_name *lname, char *name, + __u64 id) +{ + sprintf(name, LPU64, id); + lname->ln_name = name; + lname->ln_namelen = strlen(name); +} + +/* similar to mdt_attr_get_complex */ +static int echo_big_lmm_get(const struct lu_env *env, struct md_object *o, + struct md_attr *ma) +{ + struct echo_thread_info *info = echo_env_info(env); + int rc; + + ENTRY; + + LASSERT(ma->ma_lmm_size > 0); + + rc = mo_xattr_get(env, o, &LU_BUF_NULL, XATTR_NAME_LOV); + if (rc < 0) + RETURN(rc); + + /* big_lmm may need to be grown */ + if (info->eti_big_lmmsize < rc) { + int size = size_roundup_power2(rc); + + if (info->eti_big_lmmsize > 0) { + /* free old buffer */ + LASSERT(info->eti_big_lmm); + OBD_FREE_LARGE(info->eti_big_lmm, + info->eti_big_lmmsize); + info->eti_big_lmm = NULL; + info->eti_big_lmmsize = 0; + } + + OBD_ALLOC_LARGE(info->eti_big_lmm, size); + if (info->eti_big_lmm == NULL) + RETURN(-ENOMEM); + info->eti_big_lmmsize = size; + } + LASSERT(info->eti_big_lmmsize >= rc); + + info->eti_buf.lb_buf = info->eti_big_lmm; + info->eti_buf.lb_len = info->eti_big_lmmsize; + rc = mo_xattr_get(env, o, &info->eti_buf, XATTR_NAME_LOV); + if (rc < 0) + RETURN(rc); + + ma->ma_valid |= MA_LOV; + ma->ma_lmm = info->eti_big_lmm; + ma->ma_lmm_size = rc; + + RETURN(0); +} + +int echo_attr_get_complex(const struct lu_env *env, struct md_object *next, + struct md_attr *ma) +{ + struct echo_thread_info *info = echo_env_info(env); + struct lu_buf *buf = &info->eti_buf; + cfs_umode_t mode = lu_object_attr(&next->mo_lu); + int need = ma->ma_need; + int rc = 0, rc2; + + ENTRY; + + ma->ma_valid = 0; + + if (need & MA_INODE) { + ma->ma_need = MA_INODE; + rc = mo_attr_get(env, next, ma); + if (rc) + GOTO(out, rc); + ma->ma_valid |= MA_INODE; + } + + if (need & MA_LOV) { + if (S_ISREG(mode) || S_ISDIR(mode)) { + LASSERT(ma->ma_lmm_size > 0); + buf->lb_buf = ma->ma_lmm; + buf->lb_len = ma->ma_lmm_size; + rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_LOV); + if (rc2 > 0) { + ma->ma_lmm_size = rc2; + ma->ma_valid |= MA_LOV; + } else if (rc2 == -ENODATA) { + /* no LOV EA */ + ma->ma_lmm_size = 0; + } else if (rc2 == -ERANGE) { + rc2 = echo_big_lmm_get(env, next, ma); + if (rc2 < 0) + GOTO(out, rc = rc2); + } else { + GOTO(out, rc = rc2); + } + } + } + +#ifdef CONFIG_FS_POSIX_ACL + if (need & MA_ACL_DEF && S_ISDIR(mode)) { + buf->lb_buf = ma->ma_acl; + buf->lb_len = ma->ma_acl_size; + rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT); + if (rc2 > 0) { + ma->ma_acl_size = rc2; + ma->ma_valid |= MA_ACL_DEF; + } else if (rc2 == -ENODATA) { + /* no ACLs */ + ma->ma_acl_size = 0; + } else { + GOTO(out, rc = rc2); + } + } +#endif +out: + ma->ma_need = need; + CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = "LPX64" ma_lmm=%p\n", + rc, ma->ma_valid, ma->ma_lmm); + RETURN(rc); +} + +static int +echo_md_create_internal(const struct lu_env *env, struct echo_device *ed, + struct md_object *parent, struct lu_fid *fid, + struct lu_name *lname, struct md_op_spec *spec, + struct md_attr *ma) +{ + struct lu_object *ec_child, *child; + struct lu_device *ld = ed->ed_next; + struct echo_thread_info *info = echo_env_info(env); + struct lu_fid *fid2 = &info->eti_fid2; + struct lu_object_conf conf = { .loc_flags = LOC_F_NEW }; + int rc; + + ENTRY; + + rc = mdo_lookup(env, parent, lname, fid2, spec); + if (rc == 0) + return -EEXIST; + else if (rc != -ENOENT) + return rc; + + ec_child = lu_object_find_at(env, &ed->ed_cl.cd_lu_dev, + fid, &conf); + if (IS_ERR(ec_child)) { + CERROR("Can not find the child "DFID": rc = %ld\n", PFID(fid), + PTR_ERR(ec_child)); + RETURN(PTR_ERR(ec_child)); + } + + child = lu_object_locate(ec_child->lo_header, ld->ld_type); + if (child == NULL) { + CERROR("Can not locate the child "DFID"\n", PFID(fid)); + GOTO(out_put, rc = -EINVAL); + } + + CDEBUG(D_RPCTRACE, "Start creating object "DFID" %s %p\n", + PFID(lu_object_fid(&parent->mo_lu)), lname->ln_name, parent); + + /* + * Do not perform lookup sanity check. We know that name does not exist. + */ + spec->sp_cr_lookup = 0; + rc = mdo_create(env, parent, lname, lu2md(child), spec, ma); + if (rc) { + CERROR("Can not create child "DFID": rc = %d\n", PFID(fid), rc); + GOTO(out_put, rc); + } + CDEBUG(D_RPCTRACE, "End creating object "DFID" %s %p rc = %d\n", + PFID(lu_object_fid(&parent->mo_lu)), lname->ln_name, parent, rc); + EXIT; +out_put: + lu_object_put(env, ec_child); + return rc; +} + +static int echo_set_lmm_size(const struct lu_env *env, struct lu_device *ld, + struct md_attr *ma) +{ + struct echo_thread_info *info = echo_env_info(env); + + if (strcmp(ld->ld_type->ldt_name, LUSTRE_MDD_NAME)) { + ma->ma_lmm = (void *)&info->eti_lmm; + ma->ma_lmm_size = sizeof(info->eti_lmm); + } else { + LASSERT(info->eti_big_lmmsize); + ma->ma_lmm = info->eti_big_lmm; + ma->ma_lmm_size = info->eti_big_lmmsize; + } + + return 0; +} + +static int echo_create_md_object(const struct lu_env *env, + struct echo_device *ed, + struct lu_object *ec_parent, + struct lu_fid *fid, + char *name, int namelen, + __u64 id, __u32 mode, int count, + int stripe_count, int stripe_offset) +{ + struct lu_object *parent; + struct echo_thread_info *info = echo_env_info(env); + struct lu_name *lname = &info->eti_lname; + struct md_op_spec *spec = &info->eti_spec; + struct md_attr *ma = &info->eti_ma; + struct lu_device *ld = ed->ed_next; + int rc = 0; + int i; + + ENTRY; + + if (ec_parent == NULL) + return -1; + parent = lu_object_locate(ec_parent->lo_header, ld->ld_type); + if (parent == NULL) + RETURN(-ENXIO); + + memset(ma, 0, sizeof(*ma)); + memset(spec, 0, sizeof(*spec)); + if (stripe_count != 0) { + spec->sp_cr_flags |= FMODE_WRITE; + echo_set_lmm_size(env, ld, ma); + if (stripe_count != -1) { + struct lov_user_md_v3 *lum = &info->eti_lum; + + lum->lmm_magic = LOV_USER_MAGIC_V3; + lum->lmm_stripe_count = stripe_count; + lum->lmm_stripe_offset = stripe_offset; + lum->lmm_pattern = 0; + spec->u.sp_ea.eadata = lum; + spec->u.sp_ea.eadatalen = sizeof(*lum); + spec->sp_cr_flags |= MDS_OPEN_HAS_EA; + } + } + + ma->ma_attr.la_mode = mode; + ma->ma_attr.la_valid = LA_CTIME | LA_MODE; + ma->ma_attr.la_ctime = cfs_time_current_64(); + + if (name != NULL) { + lname->ln_name = name; + lname->ln_namelen = namelen; + /* If name is specified, only create one object by name */ + rc = echo_md_create_internal(env, ed, lu2md(parent), fid, lname, + spec, ma); + RETURN(rc); + } + + /* Create multiple object sequenced by id */ + for (i = 0; i < count; i++) { + char *tmp_name = info->eti_name; + + echo_md_build_name(lname, tmp_name, id); + + rc = echo_md_create_internal(env, ed, lu2md(parent), fid, lname, + spec, ma); + if (rc) { + CERROR("Can not create child %s: rc = %d\n", tmp_name, + rc); + break; + } + id++; + fid->f_oid++; + } + + RETURN(rc); +} + +static struct lu_object *echo_md_lookup(const struct lu_env *env, + struct echo_device *ed, + struct md_object *parent, + struct lu_name *lname) +{ + struct echo_thread_info *info = echo_env_info(env); + struct lu_fid *fid = &info->eti_fid; + struct lu_object *child; + int rc; + ENTRY; + + CDEBUG(D_INFO, "lookup %s in parent "DFID" %p\n", lname->ln_name, + PFID(fid), parent); + rc = mdo_lookup(env, parent, lname, fid, NULL); + if (rc) { + CERROR("lookup %s: rc = %d\n", lname->ln_name, rc); + RETURN(ERR_PTR(rc)); + } + + /* In the function below, .hs_keycmp resolves to + * lu_obj_hop_keycmp() */ + /* coverity[overrun-buffer-val] */ + child = lu_object_find_at(env, &ed->ed_cl.cd_lu_dev, fid, NULL); + + RETURN(child); +} + +static int echo_setattr_object(const struct lu_env *env, + struct echo_device *ed, + struct lu_object *ec_parent, + __u64 id, int count) +{ + struct lu_object *parent; + struct echo_thread_info *info = echo_env_info(env); + struct lu_name *lname = &info->eti_lname; + char *name = info->eti_name; + struct lu_device *ld = ed->ed_next; + struct lu_buf *buf = &info->eti_buf; + int rc = 0; + int i; + + ENTRY; + + if (ec_parent == NULL) + return -1; + parent = lu_object_locate(ec_parent->lo_header, ld->ld_type); + if (parent == NULL) + RETURN(-ENXIO); + + for (i = 0; i < count; i++) { + struct lu_object *ec_child, *child; + + echo_md_build_name(lname, name, id); + + ec_child = echo_md_lookup(env, ed, lu2md(parent), lname); + if (IS_ERR(ec_child)) { + CERROR("Can't find child %s: rc = %ld\n", + lname->ln_name, PTR_ERR(ec_child)); + RETURN(PTR_ERR(ec_child)); + } + + child = lu_object_locate(ec_child->lo_header, ld->ld_type); + if (child == NULL) { + CERROR("Can not locate the child %s\n", lname->ln_name); + lu_object_put(env, ec_child); + rc = -EINVAL; + break; + } + + CDEBUG(D_RPCTRACE, "Start setattr object "DFID"\n", + PFID(lu_object_fid(child))); + + buf->lb_buf = info->eti_xattr_buf; + buf->lb_len = sizeof(info->eti_xattr_buf); + + sprintf(name, "%s.test1", XATTR_USER_PREFIX); + rc = mo_xattr_set(env, lu2md(child), buf, name, + LU_XATTR_CREATE); + if (rc < 0) { + CERROR("Can not setattr child "DFID": rc = %d\n", + PFID(lu_object_fid(child)), rc); + lu_object_put(env, ec_child); + break; + } + CDEBUG(D_RPCTRACE, "End setattr object "DFID"\n", + PFID(lu_object_fid(child))); + id++; + lu_object_put(env, ec_child); + } + RETURN(rc); +} + +static int echo_getattr_object(const struct lu_env *env, + struct echo_device *ed, + struct lu_object *ec_parent, + __u64 id, int count) +{ + struct lu_object *parent; + struct echo_thread_info *info = echo_env_info(env); + struct lu_name *lname = &info->eti_lname; + char *name = info->eti_name; + struct md_attr *ma = &info->eti_ma; + struct lu_device *ld = ed->ed_next; + int rc = 0; + int i; + + ENTRY; + + if (ec_parent == NULL) + return -1; + parent = lu_object_locate(ec_parent->lo_header, ld->ld_type); + if (parent == NULL) + RETURN(-ENXIO); + + memset(ma, 0, sizeof(*ma)); + ma->ma_need |= MA_INODE | MA_LOV | MA_PFID | MA_HSM | MA_ACL_DEF; + ma->ma_acl = info->eti_xattr_buf; + ma->ma_acl_size = sizeof(info->eti_xattr_buf); + + for (i = 0; i < count; i++) { + struct lu_object *ec_child, *child; + + ma->ma_valid = 0; + echo_md_build_name(lname, name, id); + echo_set_lmm_size(env, ld, ma); + + ec_child = echo_md_lookup(env, ed, lu2md(parent), lname); + if (IS_ERR(ec_child)) { + CERROR("Can't find child %s: rc = %ld\n", + lname->ln_name, PTR_ERR(ec_child)); + RETURN(PTR_ERR(ec_child)); + } + + child = lu_object_locate(ec_child->lo_header, ld->ld_type); + if (child == NULL) { + CERROR("Can not locate the child %s\n", lname->ln_name); + lu_object_put(env, ec_child); + RETURN(-EINVAL); + } + + CDEBUG(D_RPCTRACE, "Start getattr object "DFID"\n", + PFID(lu_object_fid(child))); + rc = echo_attr_get_complex(env, lu2md(child), ma); + if (rc) { + CERROR("Can not getattr child "DFID": rc = %d\n", + PFID(lu_object_fid(child)), rc); + lu_object_put(env, ec_child); + break; + } + CDEBUG(D_RPCTRACE, "End getattr object "DFID"\n", + PFID(lu_object_fid(child))); + id++; + lu_object_put(env, ec_child); + } + + RETURN(rc); +} + +static int echo_lookup_object(const struct lu_env *env, + struct echo_device *ed, + struct lu_object *ec_parent, + __u64 id, int count) +{ + struct lu_object *parent; + struct echo_thread_info *info = echo_env_info(env); + struct lu_name *lname = &info->eti_lname; + char *name = info->eti_name; + struct lu_fid *fid = &info->eti_fid; + struct lu_device *ld = ed->ed_next; + int rc = 0; + int i; + + if (ec_parent == NULL) + return -1; + parent = lu_object_locate(ec_parent->lo_header, ld->ld_type); + if (parent == NULL) + return -ENXIO; + + /*prepare the requests*/ + for (i = 0; i < count; i++) { + echo_md_build_name(lname, name, id); + + CDEBUG(D_RPCTRACE, "Start lookup object "DFID" %s %p\n", + PFID(lu_object_fid(parent)), lname->ln_name, parent); + + rc = mdo_lookup(env, lu2md(parent), lname, fid, NULL); + if (rc) { + CERROR("Can not lookup child %s: rc = %d\n", name, rc); + break; + } + CDEBUG(D_RPCTRACE, "End lookup object "DFID" %s %p\n", + PFID(lu_object_fid(parent)), lname->ln_name, parent); + + id++; + } + return rc; +} + +static int echo_md_destroy_internal(const struct lu_env *env, + struct echo_device *ed, + struct md_object *parent, + struct lu_name *lname, + struct md_attr *ma) +{ + struct lu_device *ld = ed->ed_next; + struct lu_object *ec_child; + struct lu_object *child; + int rc; + + ENTRY; + + ec_child = echo_md_lookup(env, ed, parent, lname); + if (IS_ERR(ec_child)) { + CERROR("Can't find child %s: rc = %ld\n", lname->ln_name, + PTR_ERR(ec_child)); + RETURN(PTR_ERR(ec_child)); + } + + child = lu_object_locate(ec_child->lo_header, ld->ld_type); + if (child == NULL) { + CERROR("Can not locate the child %s\n", lname->ln_name); + GOTO(out_put, rc = -EINVAL); + } + + CDEBUG(D_RPCTRACE, "Start destroy object "DFID" %s %p\n", + PFID(lu_object_fid(&parent->mo_lu)), lname->ln_name, parent); + + rc = mdo_unlink(env, parent, lu2md(child), lname, ma, 0); + if (rc) { + CERROR("Can not unlink child %s: rc = %d\n", + lname->ln_name, rc); + GOTO(out_put, rc); + } + CDEBUG(D_RPCTRACE, "End destroy object "DFID" %s %p\n", + PFID(lu_object_fid(&parent->mo_lu)), lname->ln_name, parent); +out_put: + lu_object_put(env, ec_child); + return rc; +} + +static int echo_destroy_object(const struct lu_env *env, + struct echo_device *ed, + struct lu_object *ec_parent, + char *name, int namelen, + __u64 id, __u32 mode, + int count) +{ + struct echo_thread_info *info = echo_env_info(env); + struct lu_name *lname = &info->eti_lname; + struct md_attr *ma = &info->eti_ma; + struct lu_device *ld = ed->ed_next; + struct lu_object *parent; + int rc = 0; + int i; + ENTRY; + + parent = lu_object_locate(ec_parent->lo_header, ld->ld_type); + if (parent == NULL) + RETURN(-EINVAL); + + memset(ma, 0, sizeof(*ma)); + ma->ma_attr.la_mode = mode; + ma->ma_attr.la_valid = LA_CTIME; + ma->ma_attr.la_ctime = cfs_time_current_64(); + ma->ma_need = MA_INODE; + ma->ma_valid = 0; + + if (name != NULL) { + lname->ln_name = name; + lname->ln_namelen = namelen; + rc = echo_md_destroy_internal(env, ed, lu2md(parent), lname, + ma); + RETURN(rc); + } + + /*prepare the requests*/ + for (i = 0; i < count; i++) { + char *tmp_name = info->eti_name; + + ma->ma_valid = 0; + echo_md_build_name(lname, tmp_name, id); + + rc = echo_md_destroy_internal(env, ed, lu2md(parent), lname, + ma); + if (rc) { + CERROR("Can not unlink child %s: rc = %d\n", name, rc); + break; + } + id++; + } + + RETURN(rc); +} + +static struct lu_object *echo_resolve_path(const struct lu_env *env, + struct echo_device *ed, char *path, + int path_len) +{ + struct lu_device *ld = ed->ed_next; + struct md_device *md = lu2md_dev(ld); + struct echo_thread_info *info = echo_env_info(env); + struct lu_fid *fid = &info->eti_fid; + struct lu_name *lname = &info->eti_lname; + struct lu_object *parent = NULL; + struct lu_object *child = NULL; + int rc = 0; + ENTRY; + + /*Only support MDD layer right now*/ + rc = md->md_ops->mdo_root_get(env, md, fid); + if (rc) { + CERROR("get root error: rc = %d\n", rc); + RETURN(ERR_PTR(rc)); + } + + /* In the function below, .hs_keycmp resolves to + * lu_obj_hop_keycmp() */ + /* coverity[overrun-buffer-val] */ + parent = lu_object_find_at(env, &ed->ed_cl.cd_lu_dev, fid, NULL); + if (IS_ERR(parent)) { + CERROR("Can not find the parent "DFID": rc = %ld\n", + PFID(fid), PTR_ERR(parent)); + RETURN(parent); + } + + while (1) { + struct lu_object *ld_parent; + char *e; + + e = strsep(&path, "/"); + if (e == NULL) + break; + + if (e[0] == 0) { + if (!path || path[0] == '\0') + break; + continue; + } + + lname->ln_name = e; + lname->ln_namelen = strlen(e); + + ld_parent = lu_object_locate(parent->lo_header, ld->ld_type); + if (ld_parent == NULL) { + lu_object_put(env, parent); + rc = -EINVAL; + break; + } + + child = echo_md_lookup(env, ed, lu2md(ld_parent), lname); + lu_object_put(env, parent); + if (IS_ERR(child)) { + rc = (int)PTR_ERR(child); + CERROR("lookup %s under parent "DFID": rc = %d\n", + lname->ln_name, PFID(lu_object_fid(ld_parent)), + rc); + break; + } + parent = child; + } + if (rc) + RETURN(ERR_PTR(rc)); + + RETURN(parent); +} + +static void echo_ucred_init(struct lu_env *env) +{ + struct lu_ucred *ucred = lu_ucred(env); + + ucred->uc_valid = UCRED_INVALID; + + ucred->uc_suppgids[0] = -1; + ucred->uc_suppgids[1] = -1; + + ucred->uc_uid = ucred->uc_o_uid = cfs_curproc_uid(); + ucred->uc_gid = ucred->uc_o_gid = cfs_curproc_gid(); + ucred->uc_fsuid = ucred->uc_o_fsuid = cfs_curproc_fsuid(); + ucred->uc_fsgid = ucred->uc_o_fsgid = cfs_curproc_fsgid(); + ucred->uc_cap = cfs_curproc_cap_pack(); + + /* remove fs privilege for non-root user. */ + if (ucred->uc_fsuid) + ucred->uc_cap &= ~CFS_CAP_FS_MASK; + ucred->uc_valid = UCRED_NEW; +} + +static void echo_ucred_fini(struct lu_env *env) +{ + struct lu_ucred *ucred = lu_ucred(env); + ucred->uc_valid = UCRED_INIT; +} + +#define ECHO_MD_CTX_TAG (LCT_REMEMBER | LCT_MD_THREAD) +#define ECHO_MD_SES_TAG (LCT_REMEMBER | LCT_SESSION) +static int echo_md_handler(struct echo_device *ed, int command, + char *path, int path_len, __u64 id, int count, + struct obd_ioctl_data *data) +{ + struct echo_thread_info *info; + struct lu_device *ld = ed->ed_next; + struct lu_env *env; + int refcheck; + struct lu_object *parent; + char *name = NULL; + int namelen = data->ioc_plen2; + int rc = 0; + ENTRY; + + if (ld == NULL) { + CERROR("MD echo client is not being initialized properly\n"); + RETURN(-EINVAL); + } + + if (strcmp(ld->ld_type->ldt_name, LUSTRE_MDD_NAME)) { + CERROR("Only support MDD layer right now!\n"); + RETURN(-EINVAL); + } + + env = cl_env_get(&refcheck); + if (IS_ERR(env)) + RETURN(PTR_ERR(env)); + + rc = lu_env_refill_by_tags(env, ECHO_MD_CTX_TAG, ECHO_MD_SES_TAG); + if (rc != 0) + GOTO(out_env, rc); + + /* init big_lmm buffer */ + info = echo_env_info(env); + LASSERT(info->eti_big_lmm == NULL); + OBD_ALLOC_LARGE(info->eti_big_lmm, MIN_MD_SIZE); + if (info->eti_big_lmm == NULL) + GOTO(out_env, rc = -ENOMEM); + info->eti_big_lmmsize = MIN_MD_SIZE; + + parent = echo_resolve_path(env, ed, path, path_len); + if (IS_ERR(parent)) { + CERROR("Can not resolve the path %s: rc = %ld\n", path, + PTR_ERR(parent)); + GOTO(out_free, rc = PTR_ERR(parent)); + } + + if (namelen > 0) { + OBD_ALLOC(name, namelen + 1); + if (name == NULL) + GOTO(out_put, rc = -ENOMEM); + if (cfs_copy_from_user(name, data->ioc_pbuf2, namelen)) + GOTO(out_name, rc = -EFAULT); + } + + echo_ucred_init(env); + + switch (command) { + case ECHO_MD_CREATE: + case ECHO_MD_MKDIR: { + struct echo_thread_info *info = echo_env_info(env); + __u32 mode = data->ioc_obdo2.o_mode; + struct lu_fid *fid = &info->eti_fid; + int stripe_count = (int)data->ioc_obdo2.o_misc; + int stripe_index = (int)data->ioc_obdo2.o_stripe_idx; + + rc = ostid_to_fid(fid, &data->ioc_obdo1.o_oi, 0); + if (rc != 0) + break; + + /* In the function below, .hs_keycmp resolves to + * lu_obj_hop_keycmp() */ + /* coverity[overrun-buffer-val] */ + rc = echo_create_md_object(env, ed, parent, fid, name, namelen, + id, mode, count, stripe_count, + stripe_index); + break; + } + case ECHO_MD_DESTROY: + case ECHO_MD_RMDIR: { + __u32 mode = data->ioc_obdo2.o_mode; + + rc = echo_destroy_object(env, ed, parent, name, namelen, + id, mode, count); + break; + } + case ECHO_MD_LOOKUP: + rc = echo_lookup_object(env, ed, parent, id, count); + break; + case ECHO_MD_GETATTR: + rc = echo_getattr_object(env, ed, parent, id, count); + break; + case ECHO_MD_SETATTR: + rc = echo_setattr_object(env, ed, parent, id, count); + break; + default: + CERROR("unknown command %d\n", command); + rc = -EINVAL; + break; + } + echo_ucred_fini(env); + +out_name: + if (name != NULL) + OBD_FREE(name, namelen + 1); +out_put: + lu_object_put(env, parent); +out_free: + LASSERT(info->eti_big_lmm); + OBD_FREE_LARGE(info->eti_big_lmm, info->eti_big_lmmsize); + info->eti_big_lmm = NULL; + info->eti_big_lmmsize = 0; +out_env: + cl_env_put(env, &refcheck); + return rc; +} + +static int echo_create_object(const struct lu_env *env, struct echo_device *ed, + int on_target, struct obdo *oa, void *ulsm, + int ulsm_nob, struct obd_trans_info *oti) { struct echo_object *eco; struct echo_client_obd *ec = ed->ed_ec; @@ -1283,9 +2256,9 @@ static int echo_create_object(struct echo_device *ed, int on_target, RETURN(-EINVAL); } - rc = obd_alloc_memmd(ec->ec_exp, &lsm); + rc = echo_alloc_memmd(ed, &lsm); if (rc < 0) { - CERROR("Cannot allocate md, rc = %d\n", rc); + CERROR("Cannot allocate md: rc = %d\n", rc); GOTO(failed, rc); } @@ -1302,40 +2275,42 @@ static int echo_create_object(struct echo_device *ed, int on_target, if (lsm->lsm_stripe_size == 0) lsm->lsm_stripe_size = CFS_PAGE_SIZE; - idx = ll_rand(); + idx = cfs_rand(); - /* setup stripes: indices + default ids if required */ - for (i = 0; i < lsm->lsm_stripe_count; i++) { - if (lsm->lsm_oinfo[i]->loi_id == 0) - lsm->lsm_oinfo[i]->loi_id = lsm->lsm_object_id; + /* setup stripes: indices + default ids if required */ + for (i = 0; i < lsm->lsm_stripe_count; i++) { + if (ostid_id(&lsm->lsm_oinfo[i]->loi_oi) == 0) + lsm->lsm_oinfo[i]->loi_oi = lsm->lsm_oi; - lsm->lsm_oinfo[i]->loi_ost_idx = - (idx + i) % ec->ec_nstripes; - } + lsm->lsm_oinfo[i]->loi_ost_idx = + (idx + i) % ec->ec_nstripes; + } } - /* setup object ID here for !on_target and LOV hint */ - if (oa->o_valid & OBD_MD_FLID) - lsm->lsm_object_id = oa->o_id; + /* setup object ID here for !on_target and LOV hint */ + if (oa->o_valid & OBD_MD_FLID) { + LASSERT(oa->o_valid & OBD_MD_FLGROUP); + lsm->lsm_oi = oa->o_oi; + } - if (lsm->lsm_object_id == 0) - lsm->lsm_object_id = ++last_object_id; + if (ostid_id(&lsm->lsm_oi) == 0) + ostid_set_id(&lsm->lsm_oi, ++last_object_id); rc = 0; - if (on_target) { - /* Only echo objects are allowed to be created */ - LASSERT((oa->o_valid & OBD_MD_FLGROUP) && - (oa->o_gr == FILTER_GROUP_ECHO)); - rc = obd_create(ec->ec_exp, oa, &lsm, oti); - if (rc != 0) { - CERROR("Cannot create objects, rc = %d\n", rc); - GOTO(failed, rc); - } - created = 1; - } + if (on_target) { + /* Only echo objects are allowed to be created */ + LASSERT((oa->o_valid & OBD_MD_FLGROUP) && + (ostid_seq(&oa->o_oi) == FID_SEQ_ECHO)); + rc = obd_create(env, ec->ec_exp, oa, &lsm, oti); + if (rc != 0) { + CERROR("Cannot create objects: rc = %d\n", rc); + GOTO(failed, rc); + } + created = 1; + } /* See what object ID we were given */ - oa->o_id = lsm->lsm_object_id; + oa->o_oi = lsm->lsm_oi; oa->o_valid |= OBD_MD_FLID; eco = cl_echo_object_find(ed, &lsm); @@ -1343,44 +2318,40 @@ static int echo_create_object(struct echo_device *ed, int on_target, GOTO(failed, rc = PTR_ERR(eco)); cl_echo_object_put(eco); - CDEBUG(D_INFO, "oa->o_id = %lx\n", (long)oa->o_id); + CDEBUG(D_INFO, "oa oid "DOSTID"\n", POSTID(&oa->o_oi)); EXIT; failed: if (created && rc) - obd_destroy(ec->ec_exp, oa, lsm, oti, NULL, NULL); + obd_destroy(env, ec->ec_exp, oa, lsm, oti, NULL, NULL); if (lsm) - obd_free_memmd(ec->ec_exp, &lsm); + echo_free_memmd(ed, &lsm); if (rc) - CERROR("create object failed with rc = %d\n", rc); + CERROR("create object failed with: rc = %d\n", rc); return (rc); } static int echo_get_object(struct echo_object **ecop, struct echo_device *ed, struct obdo *oa) { - struct echo_client_obd *ec = ed->ed_ec; struct lov_stripe_md *lsm = NULL; struct echo_object *eco; int rc; ENTRY; - if ((oa->o_valid & OBD_MD_FLID) == 0 || - oa->o_id == 0) /* disallow use of object id 0 */ - { + if ((oa->o_valid & OBD_MD_FLID) == 0 || ostid_id(&oa->o_oi) == 0) { + /* disallow use of object id 0 */ CERROR ("No valid oid\n"); RETURN(-EINVAL); } - rc = obd_alloc_memmd(ec->ec_exp, &lsm); + rc = echo_alloc_memmd(ed, &lsm); if (rc < 0) RETURN(rc); - lsm->lsm_object_id = oa->o_id; - if (oa->o_valid & OBD_MD_FLGROUP) - lsm->lsm_object_gr = oa->o_gr; - else - lsm->lsm_object_gr = FILTER_GROUP_ECHO; + lsm->lsm_oi = oa->o_oi; + if (!(oa->o_valid & OBD_MD_FLGROUP)) + ostid_set_seq_echo(&lsm->lsm_oi); rc = 0; eco = cl_echo_object_find(ed, &lsm); @@ -1389,7 +2360,7 @@ static int echo_get_object(struct echo_object **ecop, struct echo_device *ed, else rc = PTR_ERR(eco); if (lsm) - obd_free_memmd(ec->ec_exp, &lsm); + echo_free_memmd(ed, &lsm); RETURN(rc); } @@ -1424,8 +2395,8 @@ echo_get_stripe_off_id (struct lov_stripe_md *lsm, obd_off *offp, obd_id *idp) stripe_index = woffset / stripe_size; - *idp = lsm->lsm_oinfo[stripe_index]->loi_id; - *offp = offset * stripe_size + woffset % stripe_size; + *idp = ostid_id(&lsm->lsm_oinfo[stripe_index]->loi_oi); + *offp = offset * stripe_size + woffset % stripe_size; } static void @@ -1498,7 +2469,6 @@ static int echo_client_kbrw(struct echo_device *ed, int rw, struct obdo *oa, obd_size count, int async, struct obd_trans_info *oti) { - struct echo_client_obd *ec = ed->ed_ec; struct lov_stripe_md *lsm = eco->eo_lsm; obd_count npages; struct brw_page *pga; @@ -1509,17 +2479,18 @@ static int echo_client_kbrw(struct echo_device *ed, int rw, struct obdo *oa, int rc; int verify; int gfp_mask; + int brw_flags = 0; ENTRY; - verify = ((oa->o_id) != ECHO_PERSISTENT_OBJID && + verify = (ostid_id(&oa->o_oi) != ECHO_PERSISTENT_OBJID && (oa->o_valid & OBD_MD_FLFLAGS) != 0 && (oa->o_flags & OBD_FL_DEBUG_CHECK) != 0); - gfp_mask = ((oa->o_id & 2) == 0) ? CFS_ALLOC_STD : CFS_ALLOC_HIGHUSER; + gfp_mask = ((ostid_id(&oa->o_oi) & 2) == 0) ? CFS_ALLOC_STD : CFS_ALLOC_HIGHUSER; - LASSERT(rw == OBD_BRW_WRITE || rw == OBD_BRW_READ); - LASSERT(lsm != NULL); - LASSERT(lsm->lsm_object_id == oa->o_id); + LASSERT(rw == OBD_BRW_WRITE || rw == OBD_BRW_READ); + LASSERT(lsm != NULL); + LASSERT(ostid_id(&lsm->lsm_oi) == ostid_id(&oa->o_oi)); if (count <= 0 || (count & (~CFS_PAGE_MASK)) != 0) @@ -1528,6 +2499,9 @@ static int echo_client_kbrw(struct echo_device *ed, int rw, struct obdo *oa, /* XXX think again with misaligned I/O */ npages = count >> CFS_PAGE_SHIFT; + if (rw == OBD_BRW_WRITE) + brw_flags = OBD_BRW_ASYNC; + OBD_ALLOC(pga, npages * sizeof(*pga)); if (pga == NULL) RETURN(-ENOMEM); @@ -1552,20 +2526,17 @@ static int echo_client_kbrw(struct echo_device *ed, int rw, struct obdo *oa, pages[i] = pgp->pg; pgp->count = CFS_PAGE_SIZE; pgp->off = off; - pgp->flag = 0; + pgp->flag = brw_flags; - if (verify) - echo_client_page_debug_setup(lsm, pgp->pg, rw, - oa->o_id, off, pgp->count); + if (verify) + echo_client_page_debug_setup(lsm, pgp->pg, rw, + ostid_id(&oa->o_oi), off, + pgp->count); } - if (ed->ed_next == NULL) { - struct obd_info oinfo = { { { 0 } } }; - oinfo.oi_oa = oa; - oinfo.oi_md = lsm; - rc = obd_brw(rw, ec->ec_exp, &oinfo, npages, pga, oti); - } else - rc = cl_echo_object_brw(eco, rw, offset, pages, npages, async); + /* brw mode can only be used at client */ + LASSERT(ed->ed_next != NULL); + rc = cl_echo_object_brw(eco, rw, offset, pages, npages, async); out: if (rc != 0 || rw != OBD_BRW_READ) @@ -1575,24 +2546,27 @@ static int echo_client_kbrw(struct echo_device *ed, int rw, struct obdo *oa, if (pgp->pg == NULL) continue; - if (verify) { - int vrc; - vrc = echo_client_page_debug_check(lsm, pgp->pg, oa->o_id, - pgp->off, pgp->count); - if (vrc != 0 && rc == 0) - rc = vrc; - } - OBD_PAGE_FREE(pgp->pg); + if (verify) { + int vrc; + vrc = echo_client_page_debug_check(lsm, pgp->pg, + ostid_id(&oa->o_oi), + pgp->off, pgp->count); + if (vrc != 0 && rc == 0) + rc = vrc; + } + OBD_PAGE_FREE(pgp->pg); } OBD_FREE(pga, npages * sizeof(*pga)); OBD_FREE(pages, npages * sizeof(*pages)); RETURN(rc); } -static int echo_client_prep_commit(struct obd_export *exp, int rw, - struct obdo *oa, struct echo_object *eco, - obd_off offset, obd_size count, - obd_size batch, struct obd_trans_info *oti) +static int echo_client_prep_commit(const struct lu_env *env, + struct obd_export *exp, int rw, + struct obdo *oa, struct echo_object *eco, + obd_off offset, obd_size count, + obd_size batch, struct obd_trans_info *oti, + int async) { struct lov_stripe_md *lsm = eco->eo_lsm; struct obd_ioobj ioo; @@ -1600,12 +2574,13 @@ static int echo_client_prep_commit(struct obd_export *exp, int rw, struct niobuf_remote *rnb; obd_off off; obd_size npages, tot_pages; - int i, ret = 0; + int i, ret = 0, brw_flags = 0; + ENTRY; - if (count <= 0 || (count & (~CFS_PAGE_MASK)) != 0 || - (lsm != NULL && lsm->lsm_object_id != oa->o_id)) - RETURN(-EINVAL); + if (count <= 0 || (count & (~CFS_PAGE_MASK)) != 0 || + (lsm != NULL && ostid_id(&lsm->lsm_oi) != ostid_id(&oa->o_oi))) + RETURN(-EINVAL); npages = batch >> CFS_PAGE_SHIFT; tot_pages = count >> CFS_PAGE_SHIFT; @@ -1616,6 +2591,9 @@ static int echo_client_prep_commit(struct obd_export *exp, int rw, if (lnb == NULL || rnb == NULL) GOTO(out, ret = -ENOMEM); + if (rw == OBD_BRW_WRITE && async) + brw_flags |= OBD_BRW_ASYNC; + obdo_to_ioobj(oa, &ioo); off = offset; @@ -1629,14 +2607,15 @@ static int echo_client_prep_commit(struct obd_export *exp, int rw, for (i = 0; i < npages; i++, off += CFS_PAGE_SIZE) { rnb[i].offset = off; rnb[i].len = CFS_PAGE_SIZE; + rnb[i].flags = brw_flags; } ioo.ioo_bufcnt = npages; oti->oti_transno = 0; lpages = npages; - ret = obd_preprw(rw, exp, oa, 1, &ioo, rnb, &lpages, lnb, oti, - NULL); + ret = obd_preprw(env, rw, exp, oa, 1, &ioo, rnb, &lpages, + lnb, oti, NULL); if (ret != 0) GOTO(out, ret); LASSERT(lpages == npages); @@ -1648,24 +2627,28 @@ static int echo_client_prep_commit(struct obd_export *exp, int rw, if (page == NULL && lnb[i].rc == 0) continue; - if (oa->o_id == ECHO_PERSISTENT_OBJID || - (oa->o_valid & OBD_MD_FLFLAGS) == 0 || - (oa->o_flags & OBD_FL_DEBUG_CHECK) == 0) - continue; - - if (rw == OBD_BRW_WRITE) - echo_client_page_debug_setup(lsm, page, rw, - oa->o_id, - rnb[i].offset, - rnb[i].len); - else - echo_client_page_debug_check(lsm, page, - oa->o_id, - rnb[i].offset, - rnb[i].len); - } - - ret = obd_commitrw(rw, exp, oa, 1,&ioo,rnb,npages,lnb,oti,ret); + if (async) + lnb[i].flags |= OBD_BRW_ASYNC; + + if (ostid_id(&oa->o_oi) == ECHO_PERSISTENT_OBJID || + (oa->o_valid & OBD_MD_FLFLAGS) == 0 || + (oa->o_flags & OBD_FL_DEBUG_CHECK) == 0) + continue; + + if (rw == OBD_BRW_WRITE) + echo_client_page_debug_setup(lsm, page, rw, + ostid_id(&oa->o_oi), + rnb[i].offset, + rnb[i].len); + else + echo_client_page_debug_check(lsm, page, + ostid_id(&oa->o_oi), + rnb[i].offset, + rnb[i].len); + } + + ret = obd_commitrw(env, rw, exp, oa, 1, &ioo, + rnb, npages, lnb, oti, ret); if (ret != 0) GOTO(out, ret); @@ -1681,17 +2664,19 @@ out: RETURN(ret); } -static int echo_client_brw_ioctl(int rw, struct obd_export *exp, - struct obd_ioctl_data *data) +static int echo_client_brw_ioctl(const struct lu_env *env, int rw, + struct obd_export *exp, + struct obd_ioctl_data *data, + struct obd_trans_info *dummy_oti) { struct obd_device *obd = class_exp2obd(exp); struct echo_device *ed = obd2echo_dev(obd); struct echo_client_obd *ec = ed->ed_ec; - struct obd_trans_info dummy_oti = { .oti_thread = NULL }; struct obdo *oa = &data->ioc_obdo1; struct echo_object *eco; int rc; int async = 1; + long test_mode; ENTRY; LASSERT(oa->o_valid & OBD_MD_FLGROUP); @@ -1702,20 +2687,33 @@ static int echo_client_brw_ioctl(int rw, struct obd_export *exp, oa->o_valid &= ~OBD_MD_FLHANDLE; - switch((long)data->ioc_pbuf1) { - case 1: + /* OFD/obdfilter works only via prep/commit */ + test_mode = (long)data->ioc_pbuf1; + if (test_mode == 1) async = 0; + + if (ed->ed_next == NULL && test_mode != 3) { + test_mode = 3; + data->ioc_plen1 = data->ioc_count; + } + + /* Truncate batch size to maximum */ + if (data->ioc_plen1 > PTLRPC_MAX_BRW_SIZE) + data->ioc_plen1 = PTLRPC_MAX_BRW_SIZE; + + switch (test_mode) { + case 1: /* fall through */ case 2: rc = echo_client_kbrw(ed, rw, oa, eco, data->ioc_offset, - data->ioc_count, async, &dummy_oti); + data->ioc_count, async, dummy_oti); break; case 3: - rc = echo_client_prep_commit(ec->ec_exp, rw, oa, - eco, data->ioc_offset, - data->ioc_count, data->ioc_plen1, - &dummy_oti); + rc = echo_client_prep_commit(env, ec->ec_exp, rw, oa, + eco, data->ioc_offset, + data->ioc_count, data->ioc_plen1, + dummy_oti, async); break; default: rc = -EINVAL; @@ -1753,7 +2751,7 @@ echo_client_enqueue(struct obd_export *exp, struct obdo *oa, rc = cl_echo_enqueue(eco, offset, end, mode, &ulh->cookie); if (rc == 0) { oa->o_valid |= OBD_MD_FLHANDLE; - CDEBUG(D_INFO, "Cookie is %llx\n", ulh->cookie); + CDEBUG(D_INFO, "Cookie is "LPX64"\n", ulh->cookie); } echo_put_object(eco); RETURN(rc); @@ -1768,13 +2766,13 @@ echo_client_cancel(struct obd_export *exp, struct obdo *oa) if ((oa->o_valid & OBD_MD_FLHANDLE) == 0) return -EINVAL; - CDEBUG(D_INFO, "Cookie is %llx\n", cookie); + CDEBUG(D_INFO, "Cookie is "LPX64"\n", cookie); return cl_echo_cancel(ed, cookie); } static int -echo_client_iocontrol(unsigned int cmd, struct obd_export *exp, - int len, void *karg, void *uarg) +echo_client_iocontrol(unsigned int cmd, struct obd_export *exp, int len, + void *karg, void *uarg) { struct obd_device *obd = exp->exp_obd; struct echo_device *ed = obd2echo_dev(obd); @@ -1782,42 +2780,118 @@ echo_client_iocontrol(unsigned int cmd, struct obd_export *exp, struct echo_object *eco; struct obd_ioctl_data *data = karg; struct obd_trans_info dummy_oti; + struct lu_env *env; struct oti_req_ack_lock *ack_lock; struct obdo *oa; + struct lu_fid fid; int rw = OBD_BRW_READ; int rc = 0; int i; ENTRY; - unlock_kernel(); - memset(&dummy_oti, 0, sizeof(dummy_oti)); - oa = &data->ioc_obdo1; - if (!(oa->o_valid & OBD_MD_FLGROUP)) { - oa->o_valid |= OBD_MD_FLGROUP; - oa->o_gr = FILTER_GROUP_ECHO; - } - /* assume we can touch filter native objects with echo device. */ - /* LASSERT(oa->o_gr == FILTER_GROUP_ECHO); */ + oa = &data->ioc_obdo1; + if (!(oa->o_valid & OBD_MD_FLGROUP)) { + oa->o_valid |= OBD_MD_FLGROUP; + ostid_set_seq_echo(&oa->o_oi); + } + + /* This FID is unpacked just for validation at this point */ + rc = ostid_to_fid(&fid, &oa->o_oi, 0); + if (rc < 0) + RETURN(rc); + + OBD_ALLOC_PTR(env); + if (env == NULL) + RETURN(-ENOMEM); + + rc = lu_env_init(env, LCT_DT_THREAD); + if (rc) + GOTO(out, rc = -ENOMEM); switch (cmd) { case OBD_IOC_CREATE: /* may create echo object */ if (!cfs_capable(CFS_CAP_SYS_ADMIN)) GOTO (out, rc = -EPERM); - rc = echo_create_object (ed, 1, oa, - data->ioc_pbuf1, data->ioc_plen1, - &dummy_oti); + rc = echo_create_object(env, ed, 1, oa, data->ioc_pbuf1, + data->ioc_plen1, &dummy_oti); GOTO(out, rc); + case OBD_IOC_ECHO_MD: { + int count; + int cmd; + char *dir = NULL; + int dirlen; + __u64 id; + + if (!cfs_capable(CFS_CAP_SYS_ADMIN)) + GOTO(out, rc = -EPERM); + + count = data->ioc_count; + cmd = data->ioc_command; + + id = ostid_id(&data->ioc_obdo2.o_oi); + + dirlen = data->ioc_plen1; + OBD_ALLOC(dir, dirlen + 1); + if (dir == NULL) + GOTO(out, rc = -ENOMEM); + + if (cfs_copy_from_user(dir, data->ioc_pbuf1, dirlen)) { + OBD_FREE(dir, data->ioc_plen1 + 1); + GOTO(out, rc = -EFAULT); + } + + rc = echo_md_handler(ed, cmd, dir, dirlen, id, count, data); + OBD_FREE(dir, dirlen + 1); + GOTO(out, rc); + } + case OBD_IOC_ECHO_ALLOC_SEQ: { + struct lu_env *cl_env; + int refcheck; + __u64 seq; + int max_count; + + if (!cfs_capable(CFS_CAP_SYS_ADMIN)) + GOTO(out, rc = -EPERM); + + cl_env = cl_env_get(&refcheck); + if (IS_ERR(cl_env)) + GOTO(out, rc = PTR_ERR(cl_env)); + + rc = lu_env_refill_by_tags(cl_env, ECHO_MD_CTX_TAG, + ECHO_MD_SES_TAG); + if (rc != 0) { + cl_env_put(cl_env, &refcheck); + GOTO(out, rc); + } + + rc = seq_client_get_seq(cl_env, ed->ed_cl_seq, &seq); + cl_env_put(cl_env, &refcheck); + if (rc < 0) { + CERROR("%s: Can not alloc seq: rc = %d\n", + obd->obd_name, rc); + GOTO(out, rc); + } + + if (cfs_copy_to_user(data->ioc_pbuf1, &seq, data->ioc_plen1)) + return -EFAULT; + + max_count = LUSTRE_METADATA_SEQ_MAX_WIDTH; + if (cfs_copy_to_user(data->ioc_pbuf2, &max_count, + data->ioc_plen2)) + return -EFAULT; + GOTO(out, rc); + } case OBD_IOC_DESTROY: if (!cfs_capable(CFS_CAP_SYS_ADMIN)) GOTO (out, rc = -EPERM); - rc = echo_get_object (&eco, ed, oa); + rc = echo_get_object(&eco, ed, oa); if (rc == 0) { - rc = obd_destroy(ec->ec_exp, oa, eco->eo_lsm, + rc = obd_destroy(env, ec->ec_exp, oa, eco->eo_lsm, &dummy_oti, NULL, NULL); if (rc == 0) eco->eo_deleted = 1; @@ -1826,12 +2900,12 @@ echo_client_iocontrol(unsigned int cmd, struct obd_export *exp, GOTO(out, rc); case OBD_IOC_GETATTR: - rc = echo_get_object (&eco, ed, oa); + rc = echo_get_object(&eco, ed, oa); if (rc == 0) { struct obd_info oinfo = { { { 0 } } }; oinfo.oi_md = eco->eo_lsm; oinfo.oi_oa = oa; - rc = obd_getattr(ec->ec_exp, &oinfo); + rc = obd_getattr(env, ec->ec_exp, &oinfo); echo_put_object(eco); } GOTO(out, rc); @@ -1840,13 +2914,13 @@ echo_client_iocontrol(unsigned int cmd, struct obd_export *exp, if (!cfs_capable(CFS_CAP_SYS_ADMIN)) GOTO (out, rc = -EPERM); - rc = echo_get_object (&eco, ed, oa); + rc = echo_get_object(&eco, ed, oa); if (rc == 0) { struct obd_info oinfo = { { { 0 } } }; oinfo.oi_oa = oa; oinfo.oi_md = eco->eo_lsm; - rc = obd_setattr(ec->ec_exp, &oinfo, NULL); + rc = obd_setattr(env, ec->ec_exp, &oinfo, NULL); echo_put_object(eco); } GOTO(out, rc); @@ -1858,7 +2932,7 @@ echo_client_iocontrol(unsigned int cmd, struct obd_export *exp, rw = OBD_BRW_WRITE; /* fall through */ case OBD_IOC_BRW_READ: - rc = echo_client_brw_ioctl(rw, exp, data); + rc = echo_client_brw_ioctl(env, rw, exp, data, &dummy_oti); GOTO(out, rc); case ECHO_IOC_GET_STRIPE: @@ -1881,7 +2955,7 @@ echo_client_iocontrol(unsigned int cmd, struct obd_export *exp, echo_put_object(eco); } } else { - rc = echo_create_object(ed, 0, oa, + rc = echo_create_object(env, ed, 0, oa, data->ioc_pbuf1, data->ioc_plen1, &dummy_oti); } @@ -1907,7 +2981,9 @@ echo_client_iocontrol(unsigned int cmd, struct obd_export *exp, } EXIT; - out: +out: + lu_env_fini(env); + OBD_FREE_PTR(env); /* XXX this should be in a helper also called by target_send_reply */ for (ack_lock = dummy_oti.oti_ack_locks, i = 0; i < 4; @@ -1917,12 +2993,11 @@ echo_client_iocontrol(unsigned int cmd, struct obd_export *exp, ldlm_lock_decref(&ack_lock->lock, ack_lock->mode); } - lock_kernel(); - return rc; } -static int echo_client_setup(struct obd_device *obddev, struct lustre_cfg *lcfg) +static int echo_client_setup(const struct lu_env *env, + struct obd_device *obddev, struct lustre_cfg *lcfg) { struct echo_client_obd *ec = &obddev->u.echo_client; struct obd_device *tgt; @@ -1943,12 +3018,18 @@ static int echo_client_setup(struct obd_device *obddev, struct lustre_cfg *lcfg) RETURN(-EINVAL); } - spin_lock_init (&ec->ec_lock); + spin_lock_init(&ec->ec_lock); CFS_INIT_LIST_HEAD (&ec->ec_objects); CFS_INIT_LIST_HEAD (&ec->ec_locks); ec->ec_unique = 0; ec->ec_nstripes = 0; + if (!strcmp(tgt->obd_type->typ_name, LUSTRE_MDT_NAME)) { + lu_context_tags_update(ECHO_MD_CTX_TAG); + lu_session_tags_update(ECHO_MD_SES_TAG); + RETURN(0); + } + OBD_ALLOC(ocd, sizeof(*ocd)); if (ocd == NULL) { CERROR("Can't alloc ocd connecting to %s\n", @@ -1957,16 +3038,19 @@ static int echo_client_setup(struct obd_device *obddev, struct lustre_cfg *lcfg) } ocd->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_REQPORTAL | - OBD_CONNECT_GRANT; + OBD_CONNECT_BRW_SIZE | + OBD_CONNECT_GRANT | OBD_CONNECT_FULL20 | + OBD_CONNECT_64BITHASH | OBD_CONNECT_LVB_TYPE; + ocd->ocd_brw_size = DT_MAX_BRW_SIZE; ocd->ocd_version = LUSTRE_VERSION_CODE; - ocd->ocd_group = FILTER_GROUP_ECHO; + ocd->ocd_group = FID_SEQ_ECHO; - rc = obd_connect(NULL, &ec->ec_exp, tgt, &echo_uuid, ocd, NULL); + rc = obd_connect(env, &ec->ec_exp, tgt, &echo_uuid, ocd, NULL); if (rc == 0) { /* Turn off pinger because it connects to tgt obd directly. */ - spin_lock(&tgt->obd_dev_lock); - list_del_init(&ec->ec_exp->exp_obd_chain_timed); - spin_unlock(&tgt->obd_dev_lock); + spin_lock(&tgt->obd_dev_lock); + cfs_list_del_init(&ec->ec_exp->exp_obd_chain_timed); + spin_unlock(&tgt->obd_dev_lock); } OBD_FREE(ocd, sizeof(*ocd)); @@ -1982,16 +3066,27 @@ static int echo_client_setup(struct obd_device *obddev, struct lustre_cfg *lcfg) static int echo_client_cleanup(struct obd_device *obddev) { + struct echo_device *ed = obd2echo_dev(obddev); struct echo_client_obd *ec = &obddev->u.echo_client; int rc; ENTRY; - if (!list_empty(&obddev->obd_exports)) { + /*Do nothing for Metadata echo client*/ + if (ed == NULL ) + RETURN(0); + + if (ed->ed_next_ismd) { + lu_context_tags_clear(ECHO_MD_CTX_TAG); + lu_session_tags_clear(ECHO_MD_SES_TAG); + RETURN(0); + } + + if (!cfs_list_empty(&obddev->obd_exports)) { CERROR("still has clients!\n"); RETURN(-EBUSY); } - LASSERT(atomic_read(&ec->ec_exp->exp_refcount) > 0); + LASSERT(cfs_atomic_read(&ec->ec_exp->exp_refcount) > 0); rc = obd_disconnect(ec->ec_exp); if (rc != 0) CERROR("fail to disconnect device: %d\n", rc); @@ -2034,10 +3129,10 @@ static int echo_client_disconnect(struct obd_export *exp) ec = &obd->u.echo_client; /* no more contention on export's lock list */ - while (!list_empty (&exp->exp_ec_data.eced_locks)) { - ecl = list_entry (exp->exp_ec_data.eced_locks.next, - struct ec_lock, ecl_exp_chain); - list_del (&ecl->ecl_exp_chain); + while (!cfs_list_empty (&exp->exp_ec_data.eced_locks)) { + ecl = cfs_list_entry (exp->exp_ec_data.eced_locks.next, + struct ec_lock, ecl_exp_chain); + cfs_list_del (&ecl->ecl_exp_chain); rc = obd_cancel(ec->ec_exp, ecl->ecl_object->eco_lsm, ecl->ecl_mode, &ecl->ecl_lock_handle); @@ -2056,7 +3151,7 @@ static int echo_client_disconnect(struct obd_export *exp) return rc; } -static struct obd_ops echo_obd_ops = { +static struct obd_ops echo_client_obd_ops = { .o_owner = THIS_MODULE, #if 0 @@ -2075,10 +3170,16 @@ int echo_client_init(void) int rc; lprocfs_echo_init_vars(&lvars); - rc = class_register_type(&echo_obd_ops, NULL, lvars.module_vars, - LUSTRE_ECHO_CLIENT_NAME, &echo_device_type); - if (rc == 0) - lu_kmem_init(echo_caches); + + rc = lu_kmem_init(echo_caches); + if (rc == 0) { + rc = class_register_type(&echo_client_obd_ops, NULL, + lvars.module_vars, + LUSTRE_ECHO_CLIENT_NAME, + &echo_device_type); + if (rc) + lu_kmem_fini(echo_caches); + } return rc; } @@ -2088,4 +3189,59 @@ void echo_client_exit(void) lu_kmem_fini(echo_caches); } +#ifdef __KERNEL__ +static int __init obdecho_init(void) +{ + struct lprocfs_static_vars lvars; + int rc; + + ENTRY; + LCONSOLE_INFO("Echo OBD driver; http://www.lustre.org/\n"); + + LASSERT(CFS_PAGE_SIZE % OBD_ECHO_BLOCK_SIZE == 0); + + lprocfs_echo_init_vars(&lvars); + +# ifdef HAVE_SERVER_SUPPORT + rc = echo_persistent_pages_init(); + if (rc != 0) + goto failed_0; + + rc = class_register_type(&echo_obd_ops, NULL, lvars.module_vars, + LUSTRE_ECHO_NAME, NULL); + if (rc != 0) + goto failed_1; +# endif + + rc = echo_client_init(); + +# ifdef HAVE_SERVER_SUPPORT + if (rc == 0) + RETURN(0); + + class_unregister_type(LUSTRE_ECHO_NAME); +failed_1: + echo_persistent_pages_fini(); +failed_0: +# endif + RETURN(rc); +} + +static void /*__exit*/ obdecho_exit(void) +{ + echo_client_exit(); + +# ifdef HAVE_SERVER_SUPPORT + class_unregister_type(LUSTRE_ECHO_NAME); + echo_persistent_pages_fini(); +# endif +} + +MODULE_AUTHOR("Sun Microsystems, Inc. "); +MODULE_DESCRIPTION("Lustre Testing Echo OBD driver"); +MODULE_LICENSE("GPL"); + +cfs_module(obdecho, LUSTRE_VERSION_STRING, obdecho_init, obdecho_exit); +#endif /* __KERNEL__ */ + /** @} echo_client */