X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fobdecho%2Fecho.c;h=389167bcb3c8dd3c4a07187b0f137900cdb285bc;hp=b12fe4b09582a95a82b31424030de58f54a6f907;hb=6fc066434a3b8d4cfee0e038eb5da1cf4c8b5c74;hpb=da9fe9a9db237f8996cfe0c5bab5c11b9fe8057f diff --git a/lustre/obdecho/echo.c b/lustre/obdecho/echo.c index b12fe4b..389167b 100644 --- a/lustre/obdecho/echo.c +++ b/lustre/obdecho/echo.c @@ -1,302 +1,462 @@ -/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*- - * vim:expandtab:shiftwidth=8:tabstop=8: +/* + * GPL HEADER START * - * linux/fs/obdecho/echo.c + * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * - * Copyright (C) 2001, 2002 Cluster File Systems, Inc. + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 only, + * as published by the Free Software Foundation. * - * This code is issued under the GNU General Public License. - * See the file COPYING in this distribution + * This program is distributed in the hope that it will be useful, but + * WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * General Public License version 2 for more details (a copy is included + * in the LICENSE file that accompanied this code). * - * by Peter Braam - * and Andreas Dilger + * You should have received a copy of the GNU General Public License + * version 2 along with this program; If not, see + * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf + * + * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, + * CA 95054 USA or visit www.sun.com if you need additional information or + * have any questions. + * + * GPL HEADER END + */ +/* + * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved. + * Use is subject to license terms. + * + * Copyright (c) 2010, 2013, Intel Corporation. + */ +/* + * This file is part of Lustre, http://www.lustre.org/ + * Lustre is a trademark of Sun Microsystems, Inc. + * + * lustre/obdecho/echo.c + * + * Author: Peter Braam + * Author: Andreas Dilger */ - -static char rcsid[] __attribute ((unused)) = "$Id: echo.c,v 1.39 2002/10/15 23:25:10 shaver Exp $"; -#define OBDECHO_VERSION "$Revision: 1.39 $" - -#define EXPORT_SYMTAB - -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include #define DEBUG_SUBSYSTEM S_ECHO -#include -#include -#include -#include -#include - -static atomic_t echo_page_rws; -static atomic_t echo_getattrs; - -#define ECHO_PROC_STAT "sys/obdecho" -#define ECHO_INIT_OBJID 0x1000000000000000ULL +#include +#include +#include +#include +#include -int echo_proc_read(char *page, char **start, off_t off, int count, int *eof, - void *data) -{ - long long attrs = atomic_read(&echo_getattrs); - long long pages = atomic_read(&echo_page_rws); - int len; - - *eof = 1; - if (off != 0) - return (0); +#include "echo_internal.h" - len = sprintf(page, "%Ld %Ld\n", attrs, pages); - - *start = page; - return (len); -} +/* The echo objid needs to be below 2^32, because regular FID numbers are + * limited to 2^32 objects in f_oid for the FID_SEQ_ECHO range. b=23335 */ +#define ECHO_INIT_OID 0x10000000ULL +#define ECHO_HANDLE_MAGIC 0xabcd0123fedc9876ULL -int echo_proc_write(struct file *file, const char *ubuffer, - unsigned long count, void *data) -{ - /* Ignore what we've been asked to write, and just zero the counters */ - atomic_set (&echo_page_rws, 0); - atomic_set (&echo_getattrs, 0); +#define ECHO_PERSISTENT_PAGES (ECHO_PERSISTENT_SIZE >> PAGE_CACHE_SHIFT) +static struct page *echo_persistent_pages[ECHO_PERSISTENT_PAGES]; - return (count); -} +enum { + LPROC_ECHO_READ_BYTES = 1, + LPROC_ECHO_WRITE_BYTES = 2, + LPROC_ECHO_LAST = LPROC_ECHO_WRITE_BYTES +1 +}; -void echo_proc_init(void) +static int echo_connect(const struct lu_env *env, + struct obd_export **exp, struct obd_device *obd, + struct obd_uuid *cluuid, struct obd_connect_data *data, + void *localdata) { - struct proc_dir_entry *entry; - - entry = create_proc_entry(ECHO_PROC_STAT, S_IFREG|S_IRUGO|S_IWUSR,NULL); + struct lustre_handle conn = { 0 }; + int rc; - if (entry == NULL) { - CERROR("couldn't create proc entry %s\n", ECHO_PROC_STAT); - return; + data->ocd_connect_flags &= ECHO_CONNECT_SUPPORTED; + rc = class_connect(&conn, obd, cluuid); + if (rc) { + CERROR("can't connect %d\n", rc); + return rc; } + *exp = class_conn2export(&conn); - entry->data = NULL; - entry->read_proc = echo_proc_read; - entry->write_proc = echo_proc_write; + return 0; } -void echo_proc_fini(void) +static int echo_disconnect(struct obd_export *exp) { - remove_proc_entry(ECHO_PROC_STAT, 0); + LASSERT (exp != NULL); + + return server_disconnect_export(exp); } -static int echo_connect(struct lustre_handle *conn, struct obd_device *obd, - obd_uuid_t cluuid, struct recovd_obd *recovd, - ptlrpc_recovery_cb_t recover) +static int echo_init_export(struct obd_export *exp) { - int rc; - - MOD_INC_USE_COUNT; - rc = class_connect(conn, obd, cluuid); - - if (rc) - MOD_DEC_USE_COUNT; - - return rc; + return ldlm_init_export(exp); } -static int echo_disconnect(struct lustre_handle *conn) +static int echo_destroy_export(struct obd_export *exp) { - int rc; + ENTRY; - rc = class_disconnect(conn); - if (!rc) - MOD_DEC_USE_COUNT; + target_destroy_export(exp); + ldlm_destroy_export(exp); - return rc; + RETURN(0); } -static __u64 echo_next_id(struct obd_device *obddev) + static __u64 echo_next_id(struct obd_device *obddev) { - obd_id id; + obd_id id; - spin_lock(&obddev->u.echo.eo_lock); - id = ++obddev->u.echo.eo_lastino; - spin_unlock(&obddev->u.echo.eo_lock); + spin_lock(&obddev->u.echo.eo_lock); + id = ++obddev->u.echo.eo_lastino; + spin_unlock(&obddev->u.echo.eo_lock); - return id; + return id; } -int echo_create(struct lustre_handle *conn, struct obdo *oa, - struct lov_stripe_md **ea) +static int echo_create(const struct lu_env *env, struct obd_export *exp, + struct obdo *oa, struct lov_stripe_md **ea, + struct obd_trans_info *oti) { - struct obd_device *obd = class_conn2obd(conn); + struct obd_device *obd = class_exp2obd(exp); if (!obd) { - CERROR("invalid client %Lx\n", conn->addr); + CERROR("invalid client cookie "LPX64"\n", + exp->exp_handle.h_cookie); return -EINVAL; } if (!(oa->o_mode && S_IFMT)) { - CERROR("filter obd: no type!\n"); + CERROR("echo obd: no type!\n"); return -ENOENT; } if (!(oa->o_valid & OBD_MD_FLTYPE)) { - CERROR("invalid o_valid %08x\n", oa->o_valid); + CERROR("invalid o_valid "LPX64"\n", oa->o_valid); return -EINVAL; } - oa->o_id = echo_next_id(obd); - oa->o_valid = OBD_MD_FLID; - atomic_inc(&obd->u.echo.eo_create); + ostid_set_seq_echo(&oa->o_oi); + ostid_set_id(&oa->o_oi, echo_next_id(obd)); + oa->o_valid = OBD_MD_FLID; - return 0; + return 0; } -int echo_destroy(struct lustre_handle *conn, struct obdo *oa, - struct lov_stripe_md *ea) +static int echo_destroy(const struct lu_env *env, struct obd_export *exp, + struct obdo *oa, struct lov_stripe_md *ea, + struct obd_trans_info *oti, struct obd_export *md_exp, + void *capa) { - struct obd_device *obd = class_conn2obd(conn); + struct obd_device *obd = class_exp2obd(exp); + ENTRY; if (!obd) { - CERROR("invalid client "LPX64"\n", conn->addr); + CERROR("invalid client cookie "LPX64"\n", + exp->exp_handle.h_cookie); RETURN(-EINVAL); } if (!(oa->o_valid & OBD_MD_FLID)) { - CERROR("obdo missing FLID valid flag: %08x\n", oa->o_valid); - RETURN(-EINVAL); - } - - if (oa->o_id > obd->u.echo.eo_lastino || oa->o_id < ECHO_INIT_OBJID) { - CERROR("bad destroy objid: "LPX64"\n", oa->o_id); + CERROR("obdo missing FLID valid flag: "LPX64"\n", oa->o_valid); RETURN(-EINVAL); } - atomic_inc(&obd->u.echo.eo_destroy); - - return 0; -} - -static int echo_open(struct lustre_handle *conn, struct obdo *oa, - struct lov_stripe_md *md) -{ - return 0; -} + if (ostid_id(&oa->o_oi) > obd->u.echo.eo_lastino || + ostid_id(&oa->o_oi) < ECHO_INIT_OID) { + CERROR("bad destroy objid: "DOSTID"\n", POSTID(&oa->o_oi)); + RETURN(-EINVAL); + } -static int echo_close(struct lustre_handle *conn, struct obdo *oa, - struct lov_stripe_md *md) -{ - return 0; + RETURN(0); } -static int echo_getattr(struct lustre_handle *conn, struct obdo *oa, - struct lov_stripe_md *md) +static int echo_getattr(const struct lu_env *env, struct obd_export *exp, + struct obd_info *oinfo) { - struct obd_device *obd = class_conn2obd(conn); - obd_id id = oa->o_id; + struct obd_device *obd = class_exp2obd(exp); + obd_id id = ostid_id(&oinfo->oi_oa->o_oi); + ENTRY; if (!obd) { - CERROR("invalid client "LPX64"\n", conn->addr); + CERROR("invalid client cookie "LPX64"\n", + exp->exp_handle.h_cookie); RETURN(-EINVAL); } - if (!(oa->o_valid & OBD_MD_FLID)) { - CERROR("obdo missing FLID valid flag: %08x\n", oa->o_valid); + if (!(oinfo->oi_oa->o_valid & OBD_MD_FLID)) { + CERROR("obdo missing FLID valid flag: "LPX64"\n", + oinfo->oi_oa->o_valid); RETURN(-EINVAL); } - memcpy(oa, &obd->u.echo.oa, sizeof(*oa)); - oa->o_id = id; - oa->o_valid |= OBD_MD_FLID; + obdo_cpy_md(oinfo->oi_oa, &obd->u.echo.eo_oa, oinfo->oi_oa->o_valid); + ostid_set_seq_echo(&oinfo->oi_oa->o_oi); + ostid_set_id(&oinfo->oi_oa->o_oi, id); - atomic_inc(&echo_getattrs); - - return 0; + RETURN(0); } -static int echo_setattr(struct lustre_handle *conn, struct obdo *oa, - struct lov_stripe_md *md) +static int echo_setattr(const struct lu_env *env, struct obd_export *exp, + struct obd_info *oinfo, struct obd_trans_info *oti) { - struct obd_device *obd = class_conn2obd(conn); + struct obd_device *obd = class_exp2obd(exp); + ENTRY; if (!obd) { - CERROR("invalid client "LPX64"\n", conn->addr); + CERROR("invalid client cookie "LPX64"\n", + exp->exp_handle.h_cookie); RETURN(-EINVAL); } - if (!(oa->o_valid & OBD_MD_FLID)) { - CERROR("obdo missing FLID valid flag: %08x\n", oa->o_valid); + if (!(oinfo->oi_oa->o_valid & OBD_MD_FLID)) { + CERROR("obdo missing FLID valid flag: "LPX64"\n", + oinfo->oi_oa->o_valid); RETURN(-EINVAL); } - memcpy(&obd->u.echo.oa, oa, sizeof(*oa)); + memcpy(&obd->u.echo.eo_oa, oinfo->oi_oa, sizeof(*oinfo->oi_oa)); - atomic_inc(&obd->u.echo.eo_setattr); + if (ostid_id(&oinfo->oi_oa->o_oi) & 4) { + /* Save lock to force ACKed reply */ + ldlm_lock_addref (&obd->u.echo.eo_nl_lock, LCK_NL); + oti->oti_ack_locks[0].mode = LCK_NL; + oti->oti_ack_locks[0].lock = obd->u.echo.eo_nl_lock; + } - return 0; + RETURN(0); +} + +static void +echo_page_debug_setup(struct page *page, int rw, obd_id id, + __u64 offset, int len) +{ + int page_offset = offset & ~CFS_PAGE_MASK; + char *addr = ((char *)kmap(page)) + page_offset; + + if (len % OBD_ECHO_BLOCK_SIZE != 0) + CERROR("Unexpected block size %d\n", len); + + while (len > 0) { + if (rw & OBD_BRW_READ) + block_debug_setup(addr, OBD_ECHO_BLOCK_SIZE, + offset, id); + else + block_debug_setup(addr, OBD_ECHO_BLOCK_SIZE, + 0xecc0ecc0ecc0ecc0ULL, + 0xecc0ecc0ecc0ecc0ULL); + + addr += OBD_ECHO_BLOCK_SIZE; + offset += OBD_ECHO_BLOCK_SIZE; + len -= OBD_ECHO_BLOCK_SIZE; + } + + kunmap(page); +} + +static int +echo_page_debug_check(struct page *page, obd_id id, + __u64 offset, int len) +{ + int page_offset = offset & ~CFS_PAGE_MASK; + char *addr = ((char *)kmap(page)) + page_offset; + int rc = 0; + int rc2; + + if (len % OBD_ECHO_BLOCK_SIZE != 0) + CERROR("Unexpected block size %d\n", len); + + while (len > 0) { + rc2 = block_debug_check("echo", addr, OBD_ECHO_BLOCK_SIZE, + offset, id); + + if (rc2 != 0 && rc == 0) + rc = rc2; + + addr += OBD_ECHO_BLOCK_SIZE; + offset += OBD_ECHO_BLOCK_SIZE; + len -= OBD_ECHO_BLOCK_SIZE; + } + + kunmap(page); + + return rc; } /* This allows us to verify that desc_private is passed unmolested */ #define DESC_PRIV 0x10293847 -int echo_preprw(int cmd, struct lustre_handle *conn, int objcount, - struct obd_ioobj *obj, int niocount, struct niobuf_remote *nb, - struct niobuf_local *res, void **desc_private) +static int echo_map_nb_to_lb(struct obdo *oa, struct obd_ioobj *obj, + struct niobuf_remote *nb, int *pages, + struct niobuf_local *lb, int cmd, int *left) +{ + int gfp_mask = (ostid_id(&obj->ioo_oid) & 1) ? + GFP_HIGHUSER : GFP_IOFS; + int ispersistent = ostid_id(&obj->ioo_oid) == ECHO_PERSISTENT_OBJID; + int debug_setup = (!ispersistent && + (oa->o_valid & OBD_MD_FLFLAGS) != 0 && + (oa->o_flags & OBD_FL_DEBUG_CHECK) != 0); + struct niobuf_local *res = lb; + obd_off offset = nb->offset; + int len = nb->len; + + while (len > 0) { + int plen = PAGE_CACHE_SIZE - (offset & (PAGE_CACHE_SIZE-1)); + if (len < plen) + plen = len; + + /* check for local buf overflow */ + if (*left == 0) + return -EINVAL; + + res->lnb_file_offset = offset; + res->lnb_len = plen; + LASSERT((res->lnb_file_offset & ~CFS_PAGE_MASK) + + res->lnb_len <= PAGE_CACHE_SIZE); + + if (ispersistent && + ((res->lnb_file_offset >> PAGE_CACHE_SHIFT) < + ECHO_PERSISTENT_PAGES)) { + res->lnb_page = + echo_persistent_pages[res->lnb_file_offset >> + PAGE_CACHE_SHIFT]; + /* Take extra ref so __free_pages() can be called OK */ + get_page(res->lnb_page); + } else { + OBD_PAGE_ALLOC(res->lnb_page, gfp_mask); + if (res->lnb_page == NULL) { + CERROR("can't get page for id " DOSTID"\n", + POSTID(&obj->ioo_oid)); + return -ENOMEM; + } + } + + CDEBUG(D_PAGE, "$$$$ get page %p @ "LPU64" for %d\n", + res->lnb_page, res->lnb_file_offset, res->lnb_len); + + if (cmd & OBD_BRW_READ) + res->lnb_rc = res->lnb_len; + + if (debug_setup) + echo_page_debug_setup(res->lnb_page, cmd, + ostid_id(&obj->ioo_oid), + res->lnb_file_offset, + res->lnb_len); + + offset += plen; + len -= plen; + res++; + + (*left)--; + (*pages)++; + } + + return 0; +} + +static int echo_finalize_lb(struct obdo *oa, struct obd_ioobj *obj, + struct niobuf_remote *rb, int *pgs, + struct niobuf_local *lb, int verify) +{ + struct niobuf_local *res = lb; + obd_off start = rb->offset >> PAGE_CACHE_SHIFT; + obd_off end = (rb->offset + rb->len + PAGE_CACHE_SIZE - 1) >> + PAGE_CACHE_SHIFT; + int count = (int)(end - start); + int rc = 0; + int i; + + for (i = 0; i < count; i++, (*pgs) ++, res++) { + struct page *page = res->lnb_page; + void *addr; + + if (page == NULL) { + CERROR("null page objid "LPU64":%p, buf %d/%d\n", + ostid_id(&obj->ioo_oid), page, i, + obj->ioo_bufcnt); + return -EFAULT; + } + + addr = kmap(page); + + CDEBUG(D_PAGE, "$$$$ use page %p, addr %p@"LPU64"\n", + res->lnb_page, addr, res->lnb_file_offset); + + if (verify) { + int vrc = echo_page_debug_check(page, + ostid_id(&obj->ioo_oid), + res->lnb_file_offset, + res->lnb_len); + /* check all the pages always */ + if (vrc != 0 && rc == 0) + rc = vrc; + } + + kunmap(page); + /* NB see comment above regarding persistent pages */ + OBD_PAGE_FREE(page); + } + + return rc; +} + +static int echo_preprw(const struct lu_env *env, int cmd, + struct obd_export *export, struct obdo *oa, + int objcount, struct obd_ioobj *obj, + struct niobuf_remote *nb, int *pages, + struct niobuf_local *res, struct obd_trans_info *oti, + struct lustre_capa *unused) { struct obd_device *obd; - struct niobuf_local *r = res; + int tot_bytes = 0; int rc = 0; - int i; - + int i, left; ENTRY; - obd = class_conn2obd(conn); - if (!obd) { - CERROR("invalid client "LPX64"\n", conn->addr); + obd = export->exp_obd; + if (obd == NULL) RETURN(-EINVAL); - } - memset(res, 0, sizeof(*res) * niocount); + /* Temp fix to stop falling foul of osc_announce_cached() */ + oa->o_valid &= ~(OBD_MD_FLBLOCKS | OBD_MD_FLGRANT); + + memset(res, 0, sizeof(*res) * *pages); CDEBUG(D_PAGE, "%s %d obdos with %d IOs\n", - cmd == OBD_BRW_READ ? "reading" : "writing", objcount, niocount); + cmd == OBD_BRW_READ ? "reading" : "writing", objcount, *pages); - *desc_private = (void *)DESC_PRIV; + if (oti) + oti->oti_handle = (void *)DESC_PRIV; + + left = *pages; + *pages = 0; for (i = 0; i < objcount; i++, obj++) { - int gfp_mask = (obj->ioo_id & 1) ? GFP_HIGHUSER : GFP_KERNEL; - int verify = obj->ioo_id != 0; int j; - for (j = 0 ; j < obj->ioo_bufcnt ; j++, nb++, r++) { - r->page = alloc_pages(gfp_mask, 0); - if (!r->page) { - CERROR("can't get page %d/%d for id "LPU64"\n", - j, obj->ioo_bufcnt, obj->ioo_id); - GOTO(preprw_cleanup, rc = -ENOMEM); - } - atomic_inc(&obd->u.echo.eo_prep); - - r->offset = nb->offset; - r->addr = kmap(r->page); - r->len = nb->len; - - CDEBUG(D_PAGE, "$$$$ get page %p, addr %p@"LPU64"\n", - r->page, r->addr, r->offset); - - if (verify && cmd == OBD_BRW_READ) - page_debug_setup(r->addr, r->len, r->offset, - obj->ioo_id); - else if (verify) - page_debug_setup(r->addr, r->len, - 0xecc0ecc0ecc0ecc0, - 0xecc0ecc0ecc0ecc0); + for (j = 0 ; j < obj->ioo_bufcnt ; j++, nb++) { + + rc = echo_map_nb_to_lb(oa, obj, nb, pages, + res + *pages, cmd, &left); + if (rc) + GOTO(preprw_cleanup, rc); + + tot_bytes += nb->len; } } + + atomic_add(*pages, &obd->u.echo.eo_prep); + + if (cmd & OBD_BRW_READ) + lprocfs_counter_add(obd->obd_stats, LPROC_ECHO_READ_BYTES, + tot_bytes); + else + lprocfs_counter_add(obd->obd_stats, LPROC_ECHO_WRITE_BYTES, + tot_bytes); + CDEBUG(D_PAGE, "%d pages allocated after prep\n", - atomic_read(&obd->u.echo.eo_prep)); + atomic_read(&obd->u.echo.eo_prep)); RETURN(0); @@ -306,32 +466,37 @@ preprw_cleanup: * all down again. I believe that this is what the in-kernel * prep/commit operations do. */ - CERROR("cleaning up %ld pages (%d obdos)\n", (long)(r - res), objcount); - while (r-- > res) { - kunmap(r->page); - __free_pages(r->page, 0); - atomic_dec(&obd->u.echo.eo_prep); - } - memset(res, 0, sizeof(*res) * niocount); - - return rc; + CERROR("cleaning up %u pages (%d obdos)\n", *pages, objcount); + for (i = 0; i < *pages; i++) { + kunmap(res[i].lnb_page); + /* NB if this is a persistent page, __free_pages will just + * lose the extra ref gained above */ + OBD_PAGE_FREE(res[i].lnb_page); + res[i].lnb_page = NULL; + atomic_dec(&obd->u.echo.eo_prep); + } + + return rc; } -int echo_commitrw(int cmd, struct lustre_handle *conn, int objcount, - struct obd_ioobj *obj, int niocount, struct niobuf_local *res, - void *desc_private) +static int echo_commitrw(const struct lu_env *env, int cmd, + struct obd_export *export, struct obdo *oa, + int objcount, struct obd_ioobj *obj, + struct niobuf_remote *rb, int niocount, + struct niobuf_local *res, struct obd_trans_info *oti, + int rc) { struct obd_device *obd; - struct niobuf_local *r = res; - int rc = 0; + int pgs = 0; int i; ENTRY; - obd = class_conn2obd(conn); - if (!obd) { - CERROR("invalid client "LPX64"\n", conn->addr); + obd = export->exp_obd; + if (obd == NULL) RETURN(-EINVAL); - } + + if (rc) + GOTO(commitrw_cleanup, rc); if ((cmd & OBD_BRW_RWMASK) == OBD_BRW_READ) { CDEBUG(D_PAGE, "reading %d obdos with %d IOs\n", @@ -341,122 +506,182 @@ int echo_commitrw(int cmd, struct lustre_handle *conn, int objcount, objcount, niocount); } - if (niocount && !r) { + if (niocount && res == NULL) { CERROR("NULL res niobuf with niocount %d\n", niocount); RETURN(-EINVAL); } - LASSERT(desc_private == (void *)DESC_PRIV); + LASSERT(oti == NULL || oti->oti_handle == (void *)DESC_PRIV); - for (i = 0; i < objcount; i++, obj++) { - int verify = obj->ioo_id != 0; - int j; + for (i = 0; i < objcount; i++, obj++) { + int verify = (rc == 0 && + ostid_id(&obj->ioo_oid) != ECHO_PERSISTENT_OBJID && + (oa->o_valid & OBD_MD_FLFLAGS) != 0 && + (oa->o_flags & OBD_FL_DEBUG_CHECK) != 0); + int j; - for (j = 0 ; j < obj->ioo_bufcnt ; j++, r++) { - struct page *page = r->page; - void *addr; + for (j = 0 ; j < obj->ioo_bufcnt ; j++, rb++) { + int vrc = echo_finalize_lb(oa, obj, rb, &pgs, &res[pgs], + verify); + if (vrc == 0) + continue; - if (!page || !(addr = page_address(page)) || - !kern_addr_valid(addr)) { + if (vrc == -EFAULT) + GOTO(commitrw_cleanup, rc = vrc); - CERROR("bad page objid "LPU64":%p, buf %d/%d\n", - obj->ioo_id, page, j, obj->ioo_bufcnt); - GOTO(commitrw_cleanup, rc = -EFAULT); - } + if (rc == 0) + rc = vrc; + } - atomic_inc(&echo_page_rws); + } - CDEBUG(D_PAGE, "$$$$ use page %p, addr %p@"LPU64"\n", - r->page, addr, r->offset); + atomic_sub(pgs, &obd->u.echo.eo_prep); - if (verify) - page_debug_check("echo", addr, r->len, - r->offset, obj->ioo_id); - - kunmap(page); - __free_pages(page, 0); - atomic_dec(&obd->u.echo.eo_prep); - } - } CDEBUG(D_PAGE, "%d pages remain after commit\n", - atomic_read(&obd->u.echo.eo_prep)); - RETURN(0); + atomic_read(&obd->u.echo.eo_prep)); + RETURN(rc); commitrw_cleanup: - CERROR("cleaning up %ld pages (%d obdos)\n", - niocount - (long)(r - res) - 1, objcount); - while (++r < res + niocount) { - struct page *page = r->page; - - kunmap(page); - __free_pages(page, 0); - atomic_dec(&obd->u.echo.eo_prep); + atomic_sub(pgs, &obd->u.echo.eo_prep); + + CERROR("cleaning up %d pages (%d obdos)\n", + niocount - pgs - 1, objcount); + + while (pgs < niocount) { + struct page *page = res[pgs++].lnb_page; + + if (page == NULL) + continue; + + /* NB see comment above regarding persistent pages */ + OBD_PAGE_FREE(page); + atomic_dec(&obd->u.echo.eo_prep); } return rc; } -static int echo_setup(struct obd_device *obddev, obd_count len, void *buf) -{ - ENTRY; +LPROC_SEQ_FOPS_RO_TYPE(echo, uuid); +static struct lprocfs_seq_vars lprocfs_echo_obd_vars[] = { + { .name = "uuid", + .fops = &echo_uuid_fops }, + { 0 } +}; - obddev->obd_namespace = - ldlm_namespace_new("echo-tgt", LDLM_NAMESPACE_SERVER); - if (obddev->obd_namespace == NULL) { +static int echo_setup(struct obd_device *obd, struct lustre_cfg *lcfg) +{ + int rc; + __u64 lock_flags = 0; + struct ldlm_res_id res_id = {.name = {1}}; + char ns_name[48]; + ENTRY; + + obd->u.echo.eo_obt.obt_magic = OBT_MAGIC; + spin_lock_init(&obd->u.echo.eo_lock); + obd->u.echo.eo_lastino = ECHO_INIT_OID; + + sprintf(ns_name, "echotgt-%s", obd->obd_uuid.uuid); + obd->obd_namespace = ldlm_namespace_new(obd, ns_name, + LDLM_NAMESPACE_SERVER, + LDLM_NAMESPACE_MODEST, + LDLM_NS_TYPE_OST); + if (obd->obd_namespace == NULL) { LBUG(); RETURN(-ENOMEM); } - spin_lock_init(&obddev->u.echo.eo_lock); - obddev->u.echo.eo_lastino = ECHO_INIT_OBJID; + rc = ldlm_cli_enqueue_local(obd->obd_namespace, &res_id, LDLM_PLAIN, + NULL, LCK_NL, &lock_flags, NULL, + ldlm_completion_ast, NULL, NULL, 0, + LVB_T_NONE, NULL, &obd->u.echo.eo_nl_lock); + LASSERT (rc == ELDLM_OK); + + obd->obd_vars = lprocfs_echo_obd_vars; + if (lprocfs_seq_obd_setup(obd) == 0 && + lprocfs_alloc_obd_stats(obd, LPROC_ECHO_LAST) == 0) { + lprocfs_counter_init(obd->obd_stats, LPROC_ECHO_READ_BYTES, + LPROCFS_CNTR_AVGMINMAX, + "read_bytes", "bytes"); + lprocfs_counter_init(obd->obd_stats, LPROC_ECHO_WRITE_BYTES, + LPROCFS_CNTR_AVGMINMAX, + "write_bytes", "bytes"); + } + + ptlrpc_init_client (LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL, + "echo_ldlm_cb_client", &obd->obd_ldlm_client); RETURN(0); } -static int echo_cleanup(struct obd_device *obddev) +static int echo_cleanup(struct obd_device *obd) { - ENTRY; + int leaked; + ENTRY; - ldlm_namespace_free(obddev->obd_namespace); - CERROR("%d prep/commitrw pages leaked\n", - atomic_read(&obddev->u.echo.eo_prep)); + lprocfs_obd_cleanup(obd); + lprocfs_free_obd_stats(obd); - RETURN(0); + ldlm_lock_decref(&obd->u.echo.eo_nl_lock, LCK_NL); + + /* XXX Bug 3413; wait for a bit to ensure the BL callback has + * happened before calling ldlm_namespace_free() */ + schedule_timeout_and_set_state(TASK_UNINTERRUPTIBLE, cfs_time_seconds(1)); + + ldlm_namespace_free(obd->obd_namespace, NULL, obd->obd_force); + obd->obd_namespace = NULL; + + leaked = atomic_read(&obd->u.echo.eo_prep); + if (leaked != 0) + CERROR("%d prep/commitrw pages leaked\n", leaked); + + RETURN(0); } struct obd_ops echo_obd_ops = { - o_connect: echo_connect, - o_disconnect: echo_disconnect, - o_create: echo_create, - o_destroy: echo_destroy, - o_open: echo_open, - o_close: echo_close, - o_getattr: echo_getattr, - o_setattr: echo_setattr, - o_preprw: echo_preprw, - o_commitrw: echo_commitrw, - o_setup: echo_setup, - o_cleanup: echo_cleanup + .o_owner = THIS_MODULE, + .o_connect = echo_connect, + .o_disconnect = echo_disconnect, + .o_init_export = echo_init_export, + .o_destroy_export = echo_destroy_export, + .o_create = echo_create, + .o_destroy = echo_destroy, + .o_getattr = echo_getattr, + .o_setattr = echo_setattr, + .o_preprw = echo_preprw, + .o_commitrw = echo_commitrw, + .o_setup = echo_setup, + .o_cleanup = echo_cleanup }; -static int __init obdecho_init(void) +void echo_persistent_pages_fini(void) { - printk(KERN_INFO "Echo OBD driver " OBDECHO_VERSION - " info@clusterfs.com\n"); + int i; - echo_proc_init(); - - return class_register_type(&echo_obd_ops, OBD_ECHO_DEVICENAME); + for (i = 0; i < ECHO_PERSISTENT_PAGES; i++) + if (echo_persistent_pages[i] != NULL) { + OBD_PAGE_FREE(echo_persistent_pages[i]); + echo_persistent_pages[i] = NULL; + } } -static void __exit obdecho_exit(void) +int echo_persistent_pages_init(void) { - echo_proc_fini(); + struct page *pg; + int i; - class_unregister_type(OBD_ECHO_DEVICENAME); -} + for (i = 0; i < ECHO_PERSISTENT_PAGES; i++) { + int gfp_mask = (i < ECHO_PERSISTENT_PAGES/2) ? + GFP_IOFS : GFP_HIGHUSER; + + OBD_PAGE_ALLOC(pg, gfp_mask); + if (pg == NULL) { + echo_persistent_pages_fini(); + return -ENOMEM; + } -MODULE_AUTHOR("Cluster Filesystems Inc. "); -MODULE_DESCRIPTION("Lustre Testing Echo OBD driver " OBDECHO_VERSION); -MODULE_LICENSE("GPL"); + memset (kmap (pg), 0, PAGE_CACHE_SIZE); + kunmap (pg); -module_init(obdecho_init); -module_exit(obdecho_exit); + echo_persistent_pages[i] = pg; + } + + return 0; +}