X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Flov%2Flov_pack.c;h=d992e54893f007b40c1579311563b467751e16fa;hb=0e660eab787c3b2857e4295f1ec554e016393885;hp=3d4b4b890c8ae2c32c9227f14b11f910ef3e3470;hpb=ccb42f2458669aaac84a661091b05a59bb781197;p=fs%2Flustre-release.git diff --git a/lustre/lov/lov_pack.c b/lustre/lov/lov_pack.c index 3d4b4b8..d992e54 100644 --- a/lustre/lov/lov_pack.c +++ b/lustre/lov/lov_pack.c @@ -1,58 +1,143 @@ /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*- * vim:expandtab:shiftwidth=8:tabstop=8: * - * Copyright (C) 2002 Cluster File Systems, Inc. + * GPL HEADER START * - * This file is part of Lustre, http://www.lustre.org. + * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * - * Lustre is free software; you can redistribute it and/or - * modify it under the terms of version 2 of the GNU General Public - * License as published by the Free Software Foundation. + * 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. * - * Lustre 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 for more details. + * 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). * - * You should have received a copy of the GNU General Public License - * along with Lustre; if not, write to the Free Software - * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. + * 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 2008 Sun Microsystems, Inc. All rights reserved + * Use is subject to license terms. + */ +/* + * This file is part of Lustre, http://www.lustre.org/ + * Lustre is a trademark of Sun Microsystems, Inc. + * + * lustre/lov/lov_pack.c * * (Un)packing of OST/MDS requests * + * Author: Andreas Dilger */ -#define DEBUG_SUBSYSTEM S_LLITE +#define DEBUG_SUBSYSTEM S_LOV +#ifndef __KERNEL__ +#include +#endif + +#include +#include +#include +#include +#include +#include + +#include "lov_internal.h" + +void lov_dump_lmm_v1(int level, struct lov_mds_md_v1 *lmm) +{ + struct lov_ost_data_v1 *lod; + int i; + + CDEBUG(level, "objid "LPX64", magic 0x%08x, pattern %#x\n", + le64_to_cpu(lmm->lmm_object_id), le32_to_cpu(lmm->lmm_magic), + le32_to_cpu(lmm->lmm_pattern)); + CDEBUG(level,"stripe_size %u, stripe_count %u\n", + le32_to_cpu(lmm->lmm_stripe_size), + le32_to_cpu(lmm->lmm_stripe_count)); + + if (le32_to_cpu(lmm->lmm_stripe_count) <= LOV_V1_INSANE_STRIPE_COUNT) { + for (i = 0, lod = lmm->lmm_objects; + i < (int)le32_to_cpu(lmm->lmm_stripe_count); i++, lod++) + CDEBUG(level, + "stripe %u idx %u subobj "LPX64"/"LPX64"\n", + i, le32_to_cpu(lod->l_ost_idx), + le64_to_cpu(lod->l_object_gr), + le64_to_cpu(lod->l_object_id)); + } else { + CDEBUG(level, "bad stripe_count %u > max_stripe_count %u\n", + le32_to_cpu(lmm->lmm_stripe_count), + LOV_V1_INSANE_STRIPE_COUNT); + } +} + +void lov_dump_lmm_join(int level, struct lov_mds_md_join *lmmj) +{ -#include -#include -#include -#include -#include + CDEBUG(level, "objid "LPX64", magic 0x%08X, pattern %#X\n", + le64_to_cpu(lmmj->lmmj_md.lmm_object_id), + le32_to_cpu(lmmj->lmmj_md.lmm_magic), + le32_to_cpu(lmmj->lmmj_md.lmm_pattern)); + CDEBUG(level,"stripe_size %u, stripe_count %u extent_count %u \n", + le32_to_cpu(lmmj->lmmj_md.lmm_stripe_size), + le32_to_cpu(lmmj->lmmj_md.lmm_stripe_count), + le32_to_cpu(lmmj->lmmj_extent_count)); +} -/* lov_packdesc() is in mds/mds_lov.c */ -void lov_unpackdesc(struct lov_desc *ld) +void lov_dump_lmm_v3(int level, struct lov_mds_md_v3 *lmm) { - ld->ld_tgt_count = NTOH__u32(ld->ld_tgt_count); - ld->ld_default_stripe_count = HTON__u32(ld->ld_default_stripe_count); - ld->ld_default_stripe_size = HTON__u32(ld->ld_default_stripe_size); - ld->ld_pattern = HTON__u32(ld->ld_pattern); + struct lov_ost_data_v1 *lod; + int i; + + CDEBUG(level, "objid "LPX64", magic 0x%08x, pattern %#x\n", + le64_to_cpu(lmm->lmm_object_id), le32_to_cpu(lmm->lmm_magic), + le32_to_cpu(lmm->lmm_pattern)); + CDEBUG(level,"stripe_size %u, stripe_count %u\n", + le32_to_cpu(lmm->lmm_stripe_size), + le32_to_cpu(lmm->lmm_stripe_count)); + CDEBUG(level,"pool_name "LOV_POOLNAMEF"\n", lmm->lmm_pool_name); + + if (le32_to_cpu(lmm->lmm_stripe_count) <= LOV_V1_INSANE_STRIPE_COUNT) { + for (i = 0, lod = lmm->lmm_objects; + i < (int)le32_to_cpu(lmm->lmm_stripe_count); i++, lod++) + CDEBUG(level, + "stripe %u idx %u subobj "LPX64"/"LPX64"\n", + i, le32_to_cpu(lod->l_ost_idx), + le64_to_cpu(lod->l_object_gr), + le64_to_cpu(lod->l_object_id)); + } else { + CDEBUG(level, "bad stripe_count %u > max_stripe_count %u\n", + le32_to_cpu(lmm->lmm_stripe_count), + LOV_V1_INSANE_STRIPE_COUNT); + } } -void lov_dump_lmm(int level, struct lov_mds_md *lmm) +void lov_dump_lmm(int level, void *lmm) { - struct lov_object_id *loi; - int idx; - - CDEBUG(level, "objid "LPX64", magic %#08x, ost_count %u\n", - lmm->lmm_object_id, lmm->lmm_magic, lmm->lmm_ost_count); - CDEBUG(level,"stripe_size %u, stripe_count %u, stripe_offset %u\n", - lmm->lmm_stripe_size, lmm->lmm_stripe_count, - lmm->lmm_stripe_offset); - for (idx = 0, loi = lmm->lmm_objects; idx < lmm->lmm_ost_count; - idx++, loi++) - CDEBUG(level, "ost idx %u subobj "LPX64"\n", idx, - loi->l_object_id); + int magic; + + magic = ((struct lov_mds_md_v1 *)(lmm))->lmm_magic; + switch (magic) { + case LOV_MAGIC_V1: + return lov_dump_lmm_v1(level, (struct lov_mds_md_v1 *)(lmm)); + case LOV_MAGIC_JOIN: + return lov_dump_lmm_join(level, (struct lov_mds_md_join *)(lmm)); + case LOV_MAGIC_V3: + return lov_dump_lmm_v3(level, (struct lov_mds_md_v3 *)(lmm)); + default: + CERROR("Cannot recognize lmm_magic %x", magic); + } + return; } #define LMM_ASSERT(test) \ @@ -61,52 +146,72 @@ do { \ LASSERT(test); /* so we know what assertion failed */ \ } while(0) -/* Pack LOV object metadata for shipment to the MDS. +/* Pack LOV object metadata for disk storage. It is packed in LE byte + * order and is opaque to the networking layer. * * XXX In the future, this will be enhanced to get the EA size from the * underlying OSC device(s) to get their EA sizes so we can stack * LOVs properly. For now lov_mds_md_size() just assumes one obd_id * per stripe. */ -int lov_packmd(struct lustre_handle *conn, struct lov_mds_md **lmmp, +int lov_packmd(struct obd_export *exp, struct lov_mds_md **lmmp, struct lov_stripe_md *lsm) { - struct obd_device *obd = class_conn2obd(conn); + struct obd_device *obd = class_exp2obd(exp); struct lov_obd *lov = &obd->u.lov; - struct lov_oinfo *loi; - struct lov_mds_md *lmm; - int ost_count = lov->desc.ld_tgt_count; - int stripe_count = ost_count; - int lmm_size; + struct lov_mds_md_v1 *lmmv1; + struct lov_mds_md_v3 *lmmv3; + int stripe_count; + struct lov_ost_data_v1 *lmm_objects; + int lmm_size, lmm_magic; int i; ENTRY; if (lsm) { - int i, max = 0; - if (lsm->lsm_magic != LOV_MAGIC) { - CERROR("bad mem LOV MAGIC: %#010x != %#010x\n", - lsm->lsm_magic, LOV_MAGIC); - RETURN(-EINVAL); + lmm_magic = lsm->lsm_magic; + + /* If we are just sizing the EA, limit the stripe count + * to the actual number of OSTs in this filesystem. */ + if (!lmmp) { + stripe_count = lov_get_stripecnt(lov, + lsm->lsm_stripe_count); + lsm->lsm_stripe_count = stripe_count; + } else { + stripe_count = lsm->lsm_stripe_count; } - stripe_count = lsm->lsm_stripe_count; + } else { + /* No needs to allocated more than LOV_MAX_STRIPE_COUNT. + * Anyway, this is pretty inaccurate since ld_tgt_count now + * represents max index and we should rely on the actual number + * of OSTs instead */ + stripe_count = min((__u32)LOV_MAX_STRIPE_COUNT, + lov->desc.ld_tgt_count); + + if (lmmp && *lmmp) + lmm_magic = le32_to_cpu((*lmmp)->lmm_magic); + else + /* lsm == NULL and lmmp == NULL */ + lmm_magic = LOV_MAGIC; + } + + if ((lmm_magic != LOV_MAGIC_V1) && + (lmm_magic != LOV_MAGIC_V3)) { + CERROR("bad mem LOV MAGIC: 0x%08X != 0x%08X nor 0x%08X\n", + lmm_magic, LOV_MAGIC_V1, LOV_MAGIC_V3); + RETURN(-EINVAL); - for (i = 0,loi = lsm->lsm_oinfo; i < stripe_count; i++,loi++) { - if (loi->loi_ost_idx > max) - max = loi->loi_ost_idx; - } - ost_count = max + 1; } /* XXX LOV STACKING call into osc for sizes */ - lmm_size = lov_mds_md_size(ost_count); + lmm_size = lov_mds_md_size(stripe_count, lmm_magic); if (!lmmp) RETURN(lmm_size); if (*lmmp && !lsm) { - /* endianness */ - ost_count = ((*lmmp)->lmm_ost_count); - OBD_FREE(*lmmp, lov_mds_md_size(ost_count)); + stripe_count = le32_to_cpu((*lmmp)->lmm_stripe_count); + lmm_size = lov_mds_md_size(stripe_count, lmm_magic); + OBD_FREE(*lmmp, lmm_size); *lmmp = NULL; RETURN(0); } @@ -117,229 +222,430 @@ int lov_packmd(struct lustre_handle *conn, struct lov_mds_md **lmmp, RETURN(-ENOMEM); } - lmm = *lmmp; + CDEBUG(D_INFO, "lov_packmd: LOV_MAGIC 0x%08X, lmm_size = %d \n", + lmm_magic, lmm_size); + + lmmv1 = *lmmp; + lmmv3 = (struct lov_mds_md_v3 *)*lmmp; + if (lmm_magic == LOV_MAGIC_V3) + lmmv3->lmm_magic = cpu_to_le32(LOV_MAGIC_V3); + else + lmmv1->lmm_magic = cpu_to_le32(LOV_MAGIC_V1); - lmm->lmm_stripe_count = (stripe_count); if (!lsm) RETURN(lmm_size); - /* XXX endianness */ - lmm->lmm_magic = (lsm->lsm_magic); - lmm->lmm_object_id = (lsm->lsm_object_id); - LASSERT(lsm->lsm_object_id); - lmm->lmm_stripe_size = (lsm->lsm_stripe_size); - lmm->lmm_stripe_offset = (lsm->lsm_stripe_offset); - lmm->lmm_ost_count = (ost_count); - - /* Only fill in the object ids which we are actually using. - * Assumes lmm_objects is otherwise zero-filled. */ - for (i = 0, loi = lsm->lsm_oinfo; i < stripe_count; i++, loi++) { - /* XXX call down to osc_packmd() to do the packing */ - LASSERT(loi->loi_id); - lmm->lmm_objects[loi->loi_ost_idx].l_object_id = (loi->loi_id); + /* lmmv1 and lmmv3 point to the same struct and have the + * same first fields + */ + lmmv1->lmm_object_id = cpu_to_le64(lsm->lsm_object_id); + lmmv1->lmm_object_gr = cpu_to_le64(lsm->lsm_object_gr); + lmmv1->lmm_stripe_size = cpu_to_le32(lsm->lsm_stripe_size); + lmmv1->lmm_stripe_count = cpu_to_le32(stripe_count); + lmmv1->lmm_pattern = cpu_to_le32(lsm->lsm_pattern); + if (lsm->lsm_magic == LOV_MAGIC_V3) { + strncpy(lmmv3->lmm_pool_name, lsm->lsm_pool_name, + LOV_MAXPOOLNAME); + lmm_objects = lmmv3->lmm_objects; + } else { + lmm_objects = lmmv1->lmm_objects; + } + + for (i = 0; i < stripe_count; i++) { + struct lov_oinfo *loi = lsm->lsm_oinfo[i]; + + /* XXX LOV STACKING call down to osc_packmd() to do packing */ + LASSERTF(loi->loi_id, "lmm_oid "LPU64" stripe %u/%u idx %u\n", + lmmv1->lmm_object_id, i, stripe_count, loi->loi_ost_idx); + lmm_objects[i].l_object_id = cpu_to_le64(loi->loi_id); + lmm_objects[i].l_object_gr = cpu_to_le64(loi->loi_gr); + lmm_objects[i].l_ost_gen = cpu_to_le32(loi->loi_ost_gen); + lmm_objects[i].l_ost_idx = cpu_to_le32(loi->loi_ost_idx); } RETURN(lmm_size); } -static int lov_get_stripecnt(struct lov_obd *lov, int stripe_count) +/* Find the max stripecount we should use */ +int lov_get_stripecnt(struct lov_obd *lov, __u32 stripe_count) { if (!stripe_count) stripe_count = lov->desc.ld_default_stripe_count; - if (!stripe_count || stripe_count > lov->desc.ld_active_tgt_count) + if (stripe_count > lov->desc.ld_active_tgt_count) stripe_count = lov->desc.ld_active_tgt_count; + if (!stripe_count) + stripe_count = 1; + /* for now, we limit the stripe count directly, when bug 4424 is + * fixed this needs to be somewhat dynamic based on whether ext3 + * can handle larger EA sizes. */ + if (stripe_count > LOV_MAX_STRIPE_COUNT) + stripe_count = LOV_MAX_STRIPE_COUNT; return stripe_count; } -int lov_unpackmd(struct lustre_handle *conn, struct lov_stripe_md **lsmp, - struct lov_mds_md *lmm) + +static int lov_verify_lmm(void *lmm, int lmm_bytes, int *stripe_count) { - struct obd_device *obd = class_conn2obd(conn); - struct lov_obd *lov = &obd->u.lov; - struct lov_stripe_md *lsm; - struct lov_oinfo *loi; - int ost_count; - int ost_offset = 0; - int stripe_count; - int lsm_size; - int i; - ENTRY; + int rc; - if (lmm) { - /* endianness */ - if (lmm->lmm_magic != LOV_MAGIC) { - CERROR("bad wire LOV MAGIC: %#08x != %#08x\n", - lmm->lmm_magic, LOV_MAGIC); - RETURN(-EINVAL); + if (lsm_op_find(le32_to_cpu(*(__u32 *)lmm)) == NULL) { + char *buffer; + int sz; + + CERROR("bad disk LOV MAGIC: 0x%08X; dumping LMM (size=%d):\n", + le32_to_cpu(*(__u32 *)lmm), lmm_bytes); + sz = lmm_bytes * 2 + 1; + OBD_ALLOC(buffer, sz); + if (buffer != NULL) { + int i; + + for (i = 0; i < lmm_bytes; i++) + sprintf(buffer+2*i, "%.2X", ((char *)lmm)[i]); + buffer[sz] = '\0'; + CERROR("%s\n", buffer); + OBD_FREE(buffer, sz); } - stripe_count = (lmm->lmm_stripe_count); - LASSERT(stripe_count); - } else - stripe_count = lov_get_stripecnt(lov, 0); - - /* XXX LOV STACKING call into osc for sizes */ - lsm_size = lov_stripe_md_size(stripe_count); + return -EINVAL; + } + rc = lsm_op_find(le32_to_cpu(*(__u32 *)lmm))->lsm_lmm_verify(lmm, + lmm_bytes, stripe_count); + return rc; +} - if (!lsmp) - RETURN(lsm_size); +int lov_alloc_memmd(struct lov_stripe_md **lsmp, int stripe_count, + int pattern, int magic) +{ + int i, lsm_size; + ENTRY; - if (*lsmp && !lmm) { - stripe_count = (*lsmp)->lsm_stripe_count; - OBD_FREE(*lsmp, lov_stripe_md_size(stripe_count)); - *lsmp = NULL; - RETURN(0); - } + CDEBUG(D_INFO, "alloc lsm, stripe_count %d\n", stripe_count); + *lsmp = lsm_alloc_plain(stripe_count, &lsm_size); if (!*lsmp) { - OBD_ALLOC(*lsmp, lsm_size); - if (!*lsmp) - RETURN(-ENOMEM); + CERROR("can't allocate lsmp stripe_count %d\n", stripe_count); + RETURN(-ENOMEM); } - lsm = *lsmp; + spin_lock_init(&(*lsmp)->lsm_lock); + (*lsmp)->lsm_magic = magic; + (*lsmp)->lsm_stripe_count = stripe_count; + (*lsmp)->lsm_maxbytes = LUSTRE_STRIPE_MAXBYTES * stripe_count; + (*lsmp)->lsm_pattern = pattern; + (*lsmp)->lsm_pool_name[0] = '\0'; + (*lsmp)->lsm_oinfo[0]->loi_ost_idx = ~0; - lsm->lsm_stripe_count = stripe_count; - if (!lmm) - RETURN(lsm_size); + for (i = 0; i < stripe_count; i++) + loi_init((*lsmp)->lsm_oinfo[i]); + + RETURN(lsm_size); +} + +void lov_free_memmd(struct lov_stripe_md **lsmp) +{ + struct lov_stripe_md *lsm = *lsmp; + + LASSERT(lsm_op_find(lsm->lsm_magic) != NULL); + lsm_op_find(lsm->lsm_magic)->lsm_free(lsm); - /* XXX endianness */ - ost_offset = lsm->lsm_stripe_offset = (lmm->lmm_stripe_offset); - lsm->lsm_magic = (lmm->lmm_magic); - lsm->lsm_object_id = (lmm->lmm_object_id); - lsm->lsm_stripe_size = (lmm->lmm_stripe_size); + *lsmp = NULL; +} + + +/* Unpack LOV object metadata from disk storage. It is packed in LE byte + * order and is opaque to the networking layer. + */ +int lov_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp, + struct lov_mds_md *lmm, int lmm_bytes) +{ + struct obd_device *obd = class_exp2obd(exp); + struct lov_obd *lov = &obd->u.lov; + int rc = 0, stripe_count, lsm_size; + __u32 magic; + ENTRY; - ost_count = (lmm->lmm_ost_count); + /* If passed an MDS struct use values from there, otherwise defaults */ + if (lmm) { + rc = lov_verify_lmm(lmm, lmm_bytes, &stripe_count); + if (rc) + RETURN(rc); + magic = le32_to_cpu(lmm->lmm_magic); + } else { + stripe_count = lov_get_stripecnt(lov, 0); + magic = LOV_MAGIC; + } - LMM_ASSERT(lsm->lsm_object_id); - LMM_ASSERT(ost_count); + /* If we aren't passed an lsmp struct, we just want the size */ + if (!lsmp) { + /* XXX LOV STACKING call into osc for sizes */ + LBUG(); + RETURN(lov_stripe_md_size(stripe_count)); + } + /* If we are passed an allocated struct but nothing to unpack, free */ + if (*lsmp && !lmm) { + lov_free_memmd(lsmp); + RETURN(0); + } - for (i = 0, loi = lsm->lsm_oinfo; i < ost_count; i++, ost_offset++) { - ost_offset %= ost_count; + lsm_size = lov_alloc_memmd(lsmp, stripe_count, LOV_PATTERN_RAID0, + magic); + if (lsm_size < 0) + RETURN(lsm_size); - if (!lmm->lmm_objects[ost_offset].l_object_id) - continue; + /* If we are passed a pointer but nothing to unpack, we only alloc */ + if (!lmm) + RETURN(lsm_size); - LMM_ASSERT(loi - lsm->lsm_oinfo < stripe_count); - /* XXX LOV STACKING call down to osc_unpackmd() */ - loi->loi_id = (lmm->lmm_objects[ost_offset].l_object_id); - loi->loi_ost_idx = ost_offset; - loi++; + LASSERT(lsm_op_find(magic) != NULL); + rc = lsm_op_find(magic)->lsm_unpackmd(lov, *lsmp, lmm); + if (rc) { + lov_free_memmd(lsmp); + RETURN(rc); } - LMM_ASSERT(loi - lsm->lsm_oinfo > 0); - LMM_ASSERT(loi - lsm->lsm_oinfo == stripe_count); RETURN(lsm_size); } -/* Configure object striping information on a new file. - * - * @lmmu is a pointer to a user struct with one or more of the fields set to - * indicate the application preference: lmm_stripe_count, lmm_stripe_size, - * lmm_stripe_offset, and lmm_stripe_pattern. lmm_magic must be LOV_MAGIC. - * @lsmp is a pointer to an in-core stripe MD that needs to be filled in. - */ -int lov_setstripe(struct lustre_handle *conn, struct lov_stripe_md **lsmp, - struct lov_mds_md *lmmu) +static int __lov_setstripe(struct obd_export *exp, struct lov_stripe_md **lsmp, + struct lov_user_md *lump) { - struct obd_device *obd = class_conn2obd(conn); + struct obd_device *obd = class_exp2obd(exp); struct lov_obd *lov = &obd->u.lov; - struct lov_mds_md lmm; - struct lov_stripe_md *lsm; + struct lov_user_md_v3 lumv3; + struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3; + int lmm_magic; int stripe_count; int rc; ENTRY; - rc = copy_from_user(&lmm, lmmu, sizeof(lmm)); - if (rc) + if (copy_from_user(&lumv3, lump, sizeof(struct lov_user_md_v1))) RETURN(-EFAULT); - if (lmm.lmm_magic != LOV_MAGIC) { - CERROR("bad wire LOV MAGIC: %#08x != %#08x\n", - lmm.lmm_magic, LOV_MAGIC); - RETURN(-EINVAL); + lmm_magic = lumv1->lmm_magic; + + if (lmm_magic == __swab32(LOV_USER_MAGIC_V1)) { + lustre_swab_lov_user_md_v1(lumv1); + lmm_magic = LOV_USER_MAGIC_V1; + } else if (lmm_magic == LOV_USER_MAGIC_V3) { + if (copy_from_user(&lumv3, lump, sizeof(lumv3))) + RETURN(-EFAULT); + } else if (lmm_magic == __swab32(LOV_USER_MAGIC_V3)) { + if (copy_from_user(&lumv3, lump, sizeof(lumv3))) + RETURN(-EFAULT); + lustre_swab_lov_user_md_v3(&lumv3); + lmm_magic = LOV_USER_MAGIC_V3; + } else if (lmm_magic != LOV_USER_MAGIC_V1) { + CDEBUG(D_IOCTL, + "bad userland LOV MAGIC: %#08x != %#08x nor %#08x\n", + lmm_magic, LOV_USER_MAGIC_V1, LOV_USER_MAGIC_V3); + RETURN(-EINVAL); } - if (lmm.lmm_stripe_count > lov->desc.ld_tgt_count) { - CERROR("stripe count %d more than OST count %d\n", - (int)lmm.lmm_stripe_count, lov->desc.ld_tgt_count); - RETURN(-EINVAL); + + /* in the rest of the tests, as *lumv1 and lumv3 have the same + * fields, we use lumv1 to avoid code duplication */ + + if (lumv1->lmm_pattern == 0) { + lumv1->lmm_pattern = lov->desc.ld_pattern ? + lov->desc.ld_pattern : LOV_PATTERN_RAID0; } - if (lmm.lmm_stripe_offset >= lov->desc.ld_tgt_count) { - CERROR("stripe offset %d more than max OST index %d\n", - (int)lmm.lmm_stripe_count, lov->desc.ld_tgt_count); + + if (lumv1->lmm_pattern != LOV_PATTERN_RAID0) { + CDEBUG(D_IOCTL, "bad userland stripe pattern: %#x\n", + lumv1->lmm_pattern); RETURN(-EINVAL); } - if (lmm.lmm_stripe_size & (PAGE_SIZE - 1)) { - CERROR("stripe size %u not multiple of %lu\n", - lmm.lmm_stripe_size, PAGE_SIZE); - RETURN(-EINVAL); + + /* 64kB is the largest common page size we see (ia64), and matches the + * check in lfs */ + if (lumv1->lmm_stripe_size & (LOV_MIN_STRIPE_SIZE - 1)) { + CDEBUG(D_IOCTL, "stripe size %u not multiple of %u, fixing\n", + lumv1->lmm_stripe_size, LOV_MIN_STRIPE_SIZE); + lumv1->lmm_stripe_size = LOV_MIN_STRIPE_SIZE; } - if (lmm.lmm_stripe_size * lmm.lmm_stripe_count > ~0UL) { - CERROR("stripe width %ux%u > %lu on 32-bit system\n", - lmm.lmm_stripe_size, (int)lmm.lmm_stripe_count, ~0UL); + + if ((lumv1->lmm_stripe_offset >= lov->desc.ld_tgt_count) && + (lumv1->lmm_stripe_offset != + (typeof(lumv1->lmm_stripe_offset))(-1))) { + CDEBUG(D_IOCTL, "stripe offset %u > number of OSTs %u\n", + lumv1->lmm_stripe_offset, lov->desc.ld_tgt_count); RETURN(-EINVAL); } + stripe_count = lov_get_stripecnt(lov, lumv1->lmm_stripe_count); - stripe_count = lov_get_stripecnt(lov, lmm.lmm_stripe_count); + if (lmm_magic == LOV_USER_MAGIC_V3) { + struct pool_desc *pool; - /* XXX LOV STACKING call into osc for sizes */ - OBD_ALLOC(lsm, lov_stripe_md_size(stripe_count)); - if (!lsm) - RETURN(-ENOMEM); + pool = lov_find_pool(lov, lumv3.lmm_pool_name); + if (pool == NULL) + RETURN(-EINVAL); + + if (lumv3.lmm_stripe_offset != + (typeof(lumv3.lmm_stripe_offset))(-1)) { + rc = lov_check_index_in_pool(lumv3.lmm_stripe_offset, + pool); + if (rc < 0) + RETURN(-EINVAL); + } - lsm->lsm_magic = LOV_MAGIC; - /* This is all validated in lov_create() */ - lsm->lsm_stripe_count = stripe_count; - lsm->lsm_stripe_offset = lmm.lmm_stripe_offset; - lsm->lsm_stripe_size = lmm.lmm_stripe_size; + if (stripe_count > pool_tgt_count(pool)) + stripe_count = pool_tgt_count(pool); + } - *lsmp = lsm; + if ((__u64)lumv1->lmm_stripe_size * stripe_count > ~0UL) { + CDEBUG(D_IOCTL, "stripe width %ux%i exeeds %lu bytes\n", + lumv1->lmm_stripe_size, (int)lumv1->lmm_stripe_count, + ~0UL); + RETURN(-EINVAL); + } + + rc = lov_alloc_memmd(lsmp, stripe_count, lumv1->lmm_pattern, lmm_magic); + + if (rc >= 0) { + (*lsmp)->lsm_oinfo[0]->loi_ost_idx = lumv1->lmm_stripe_offset; + (*lsmp)->lsm_stripe_size = lumv1->lmm_stripe_size; + if (lmm_magic == LOV_USER_MAGIC_V3) + strncpy((*lsmp)->lsm_pool_name, lumv3.lmm_pool_name, + LOV_MAXPOOLNAME); + rc = 0; + } RETURN(rc); } -/* Retrieve object striping information. +/* Configure object striping information on a new file. * - * @lmmu is a pointer to an in-core struct with lmm_ost_count indicating - * the maximum number of OST indices which will fit in the user buffer. - * lmm_magic must be LOV_MAGIC. + * @lmmu is a pointer to a user struct with one or more of the fields set to + * indicate the application preference: lmm_stripe_count, lmm_stripe_size, + * lmm_stripe_offset, and lmm_stripe_pattern. lmm_magic must be LOV_MAGIC. + * @lsmp is a pointer to an in-core stripe MD that needs to be filled in. */ -int lov_getstripe(struct lustre_handle *conn, struct lov_stripe_md *lsm, - struct lov_mds_md *lmmu) +int lov_setstripe(struct obd_export *exp, struct lov_stripe_md **lsmp, + struct lov_user_md *lump) { - struct obd_device *obd = class_conn2obd(conn); - struct lov_obd *lov = &obd->u.lov; - struct lov_mds_md lmm, *lmmk = NULL; - int ost_count, rc, lmm_size; - ENTRY; + int rc; + mm_segment_t seg; - if (!lsm) - RETURN(-ENODATA); + seg = get_fs(); + set_fs(KERNEL_DS); - rc = copy_from_user(&lmm, lmmu, sizeof(lmm)); - if (rc) - RETURN(-EFAULT); + rc = __lov_setstripe(exp, lsmp, lump); + set_fs(seg); + RETURN(rc); +} - if (lmm.lmm_magic != LOV_MAGIC) - RETURN(-EINVAL); +int lov_setea(struct obd_export *exp, struct lov_stripe_md **lsmp, + struct lov_user_md *lump) +{ + int i; + int rc; + struct obd_export *oexp; + struct lov_obd *lov = &exp->exp_obd->u.lov; + obd_id last_id = 0; + struct lov_user_ost_data_v1 *lmm_objects; - ost_count = lov->desc.ld_tgt_count; + ENTRY; - /* XXX we _could_ check if indices > user lmm_ost_count are zero */ - if (lmm.lmm_ost_count < ost_count) - RETURN(-EOVERFLOW); + if (lump->lmm_magic == LOV_USER_MAGIC_V3) + lmm_objects = ((struct lov_user_md_v3 *)lump)->lmm_objects; + else + lmm_objects = lump->lmm_objects; + + for (i = 0; i < lump->lmm_stripe_count; i++) { + __u32 len = sizeof(last_id); + oexp = lov->lov_tgts[lmm_objects[i].l_ost_idx]->ltd_exp; + rc = obd_get_info(oexp, sizeof(KEY_LAST_ID), KEY_LAST_ID, + &len, &last_id, NULL); + if (rc) + RETURN(rc); + if (lmm_objects[i].l_object_id > last_id) { + CERROR("Setting EA for object > than last id on " + "ost idx %d "LPD64" > "LPD64" \n", + lmm_objects[i].l_ost_idx, + lmm_objects[i].l_object_id, last_id); + RETURN(-EINVAL); + } + } - rc = lov_packmd(conn, &lmmk, lsm); - if (rc < 0) + rc = lov_setstripe(exp, lsmp, lump); + if (rc) RETURN(rc); - lmm_size = rc; - rc = 0; + for (i = 0; i < lump->lmm_stripe_count; i++) { + (*lsmp)->lsm_oinfo[i]->loi_ost_idx = + lmm_objects[i].l_ost_idx; + (*lsmp)->lsm_oinfo[i]->loi_id = lmm_objects[i].l_object_id; + (*lsmp)->lsm_oinfo[i]->loi_gr = lmm_objects[i].l_object_gr; + } + RETURN(0); +} - if (lmm_size && copy_to_user(lmmu, lmmk, lmm_size)) - rc = -EFAULT; - obd_free_wiremd(conn, &lmmk); +/* Retrieve object striping information. + * + * @lump is a pointer to an in-core struct with lmm_ost_count indicating + * the maximum number of OST indices which will fit in the user buffer. + * lmm_magic must be LOV_USER_MAGIC. + */ +int lov_getstripe(struct obd_export *exp, struct lov_stripe_md *lsm, + struct lov_user_md *lump) +{ + /* + * XXX huge struct allocated on stack. + */ + /* we use lov_user_md_v3 because it is larger than lov_user_md_v1 */ + struct lov_user_md_v3 lum; + struct lov_mds_md *lmmk = NULL; + int rc, lmm_size; + int lum_size; + mm_segment_t seg; + ENTRY; + + if (!lsm) + RETURN(-ENODATA); + /* + * "Switch to kernel segment" to allow copying from kernel space by + * copy_{to,from}_user(). + */ + seg = get_fs(); + set_fs(KERNEL_DS); + + /* we only need the header part from user space to get lmm_magic and + * lmm_stripe_count, (the header part is common to v1 and v3) */ + lum_size = sizeof(struct lov_user_md_v1); + if (copy_from_user(&lum, lump, lum_size)) + rc = -EFAULT; + else if ((lum.lmm_magic != LOV_USER_MAGIC) && + (lum.lmm_magic != LOV_USER_MAGIC_V3)) + rc = -EINVAL; + else { + /* if v3 we just have to update the lum_size */ + if (lum.lmm_magic == LOV_USER_MAGIC_V3) + lum_size = sizeof(struct lov_user_md_v3); + + rc = lov_packmd(exp, &lmmk, lsm); + if (rc < 0) + RETURN(rc); + lmm_size = rc; + rc = 0; + + /* FIXME: Bug 1185 - copy fields properly when structs change */ + /* struct lov_user_md_v3 and struct lov_mds_md_v3 must be the same */ + CLASSERT(sizeof(lum) == sizeof(struct lov_mds_md_v3)); + CLASSERT(sizeof lum.lmm_objects[0] == + sizeof lmmk->lmm_objects[0]); + + /* User wasn't expecting this many OST entries */ + if (lum.lmm_stripe_count == 0) { + if (copy_to_user(lump, lmmk, lum_size)) + rc = -EFAULT; + } else if (lum.lmm_stripe_count < lmmk->lmm_stripe_count) { + rc = -EOVERFLOW; + } else if (copy_to_user(lump, lmmk, lmm_size)) + rc = -EFAULT; + + obd_free_diskmd(exp, &lmmk); + } + set_fs(seg); RETURN(rc); }