*
*/
-#define DEBUG_SUBSYSTEM S_LLITE
+#define DEBUG_SUBSYSTEM S_LOV
#ifndef __KERNEL__
#include <liblustre.h>
#endif
-#include <linux/lustre_net.h>
-#include <linux/obd.h>
-#include <linux/obd_lov.h>
-#include <linux/obd_class.h>
-#include <linux/obd_support.h>
+#include <lustre_net.h>
+#include <obd.h>
+#include <obd_lov.h>
+#include <obd_class.h>
+#include <obd_support.h>
+#include <lustre/lustre_user.h>
-/* lov_packdesc() is in mds/mds_lov.c */
-void lov_unpackdesc(struct lov_desc *ld)
+#include "lov_internal.h"
+
+void lov_dump_lmm_v1(int level, struct lov_mds_md_v1 *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));
+ for (i = 0, lod = lmm->lmm_objects;
+ i < 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));
}
-void lov_dump_lmm(int level, struct lov_mds_md *lmm)
+void lov_dump_lmm_join(int level, struct lov_mds_md_join *lmmj)
{
- 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);
+
+ 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));
}
#define LMM_ASSERT(test) \
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 stripe_count = lov->desc.ld_tgt_count;
int lmm_size;
int i;
ENTRY;
if (lsm) {
- int i, max = 0;
if (lsm->lsm_magic != LOV_MAGIC) {
- CERROR("bad mem LOV MAGIC: %#010x != %#010x\n",
+ CERROR("bad mem LOV MAGIC: 0x%08X != 0x%08X\n",
lsm->lsm_magic, LOV_MAGIC);
RETURN(-EINVAL);
}
stripe_count = lsm->lsm_stripe_count;
-
- 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);
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);
+ OBD_FREE(*lmmp, lov_mds_md_size(stripe_count));
*lmmp = NULL;
RETURN(0);
}
}
lmm = *lmmp;
+ lmm->lmm_magic = cpu_to_le32(LOV_MAGIC); /* only write new format */
- 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);
+ lmm->lmm_object_id = cpu_to_le64(lsm->lsm_object_id);
+ lmm->lmm_object_gr = cpu_to_le64(lsm->lsm_object_gr);
+ lmm->lmm_stripe_size = cpu_to_le32(lsm->lsm_stripe_size);
+ lmm->lmm_stripe_count = cpu_to_le32(stripe_count);
+ lmm->lmm_pattern = cpu_to_le32(lsm->lsm_pattern);
+
+ 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",
+ lmm->lmm_object_id, i, stripe_count, loi->loi_ost_idx);
+ lmm->lmm_objects[i].l_object_id = cpu_to_le64(loi->loi_id);
+ lmm->lmm_objects[i].l_object_gr = cpu_to_le64(loi->loi_gr);
+ lmm->lmm_objects[i].l_ost_gen = cpu_to_le32(loi->loi_ost_gen);
+ lmm->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, int 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)
-{
- 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;
- if (lmm) {
- /* endianness */
- if (lmm->lmm_magic != LOV_MAGIC) {
- CERROR("bad wire LOV MAGIC: %#08x != %#08x\n",
- lmm->lmm_magic, LOV_MAGIC);
- RETURN(-EINVAL);
- }
- stripe_count = (lmm->lmm_stripe_count);
- LASSERT(stripe_count);
- } else
- stripe_count = lov_get_stripecnt(lov, 0);
+static int lov_verify_lmm(void *lmm, int lmm_bytes, int *stripe_count)
+{
+ int rc;
- /* XXX LOV STACKING call into osc for sizes */
- lsm_size = lov_stripe_md_size(stripe_count);
+ if (lsm_op_find(le32_to_cpu(*(__u32 *)lmm)) == NULL) {
+ CERROR("bad disk LOV MAGIC: 0x%08X; dumping V1 LMM:\n",
+ le32_to_cpu(*(__u32 *)lmm));
+ lov_dump_lmm_v1(D_WARNING, lmm);
+ 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_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]);
- /* 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);
+ RETURN(lsm_size);
+}
- ost_count = (lmm->lmm_ost_count);
+void lov_free_memmd(struct lov_stripe_md **lsmp)
+{
+ struct lov_stripe_md *lsm = *lsmp;
- LMM_ASSERT(lsm->lsm_object_id);
- LMM_ASSERT(ost_count);
+ LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
+ lsm_op_find(lsm->lsm_magic)->lsm_free(lsm);
- for (i = 0, loi = lsm->lsm_oinfo; i < ost_count; i++, ost_offset++) {
- ost_offset %= ost_count;
+ *lsmp = NULL;
+}
- if (!lmm->lmm_objects[ost_offset].l_object_id)
- continue;
- 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++;
+/* 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;
+
+ /* 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;
+ }
+
+ /* 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);
+ }
+
+ lsm_size = lov_alloc_memmd(lsmp, stripe_count, LOV_PATTERN_RAID0,
+ magic);
+ if (lsm_size < 0)
+ RETURN(lsm_size);
+
+ /* If we are passed a pointer but nothing to unpack, we only alloc */
+ if (!lmm)
+ RETURN(lsm_size);
+
+ 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 lum;
int stripe_count;
int rc;
ENTRY;
- rc = copy_from_user(&lmm, lmmu, sizeof(lmm));
+ rc = copy_from_user(&lum, lump, sizeof(lum));
if (rc)
RETURN(-EFAULT);
- if (lmm.lmm_magic != LOV_MAGIC) {
- CERROR("bad wire LOV MAGIC: %#08x != %#08x\n",
- lmm.lmm_magic, LOV_MAGIC);
- RETURN(-EINVAL);
+ if (lum.lmm_magic != LOV_USER_MAGIC) {
+ if (lum.lmm_magic == __swab32(LOV_USER_MAGIC)) {
+ lustre_swab_lov_user_md(&lum);
+ } else {
+ CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
+ " %#08x != %#08x\n",
+ lum.lmm_magic, LOV_USER_MAGIC);
+ RETURN(-EINVAL);
+ }
}
-#if 0 /* the stripe_count/offset is "advisory", and it gets fixed later */
- if (lmm.lmm_stripe_count > lov->desc.ld_tgt_count &&
- lmm.lmm_stripe_count != 0xffffffff) {
- CERROR("stripe count %u more than OST count %d\n",
- lmm.lmm_stripe_count, lov->desc.ld_tgt_count);
- RETURN(-EINVAL);
+
+ if (lum.lmm_pattern == 0) {
+ lum.lmm_pattern = lov->desc.ld_pattern ?
+ lov->desc.ld_pattern : LOV_PATTERN_RAID0;
}
- if (lmm.lmm_stripe_offset >= lov->desc.ld_tgt_count &&
- lmm.lmm_stripe_offset != 0xffffffff) {
- CERROR("stripe offset %u more than max OST index %d\n",
- lmm.lmm_stripe_offset, lov->desc.ld_tgt_count);
+
+ if (lum.lmm_pattern != LOV_PATTERN_RAID0) {
+ CDEBUG(D_IOCTL, "bad userland stripe pattern: %#x\n",
+ lum.lmm_pattern);
RETURN(-EINVAL);
}
-#endif
- if (lmm.lmm_stripe_size & (PAGE_SIZE - 1)) {
- CERROR("stripe size %u not multiple of %lu\n",
- lmm.lmm_stripe_size, PAGE_SIZE);
+
+ /* 64kB is the largest common page size we see (ia64), and matches the
+ * check in lfs */
+ if (lum.lmm_stripe_size & (LOV_MIN_STRIPE_SIZE - 1)) {
+ CDEBUG(D_IOCTL, "stripe size %u not multiple of %u, fixing\n",
+ lum.lmm_stripe_size, LOV_MIN_STRIPE_SIZE);
+ lum.lmm_stripe_size = LOV_MIN_STRIPE_SIZE;
+ }
+
+ if ((lum.lmm_stripe_offset >= lov->desc.ld_active_tgt_count) &&
+ (lum.lmm_stripe_offset != (typeof(lum.lmm_stripe_offset))(-1))) {
+ CDEBUG(D_IOCTL, "stripe offset %u > number of active OSTs %u\n",
+ lum.lmm_stripe_offset, lov->desc.ld_active_tgt_count);
RETURN(-EINVAL);
}
- stripe_count = lov_get_stripecnt(lov, lmm.lmm_stripe_count);
+ stripe_count = lov_get_stripecnt(lov, lum.lmm_stripe_count);
- if ((__u64)lmm.lmm_stripe_size * 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 ((__u64)lum.lmm_stripe_size * stripe_count > ~0UL) {
+ CDEBUG(D_IOCTL, "stripe width %ux%i exeeds %lu bytes\n",
+ lum.lmm_stripe_size, (int)lum.lmm_stripe_count, ~0UL);
RETURN(-EINVAL);
}
- /* XXX LOV STACKING call into osc for sizes */
- OBD_ALLOC(lsm, lov_stripe_md_size(stripe_count));
- if (!lsm)
- RETURN(-ENOMEM);
+ rc = lov_alloc_memmd(lsmp, stripe_count, lum.lmm_pattern, LOV_MAGIC);
+
+ if (rc >= 0) {
+ (*lsmp)->lsm_oinfo[0]->loi_ost_idx = lum.lmm_stripe_offset;
+ (*lsmp)->lsm_stripe_size = lum.lmm_stripe_size;
+ rc = 0;
+ }
- lsm->lsm_magic = LOV_MAGIC;
- lsm->lsm_stripe_count = stripe_count;
- lsm->lsm_stripe_offset = lmm.lmm_stripe_offset;
- lsm->lsm_stripe_size = lmm.lmm_stripe_size;
+ RETURN(0);
+}
+
+/* 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 obd_export *exp, struct lov_stripe_md **lsmp,
+ struct lov_user_md *lump)
+{
+ int rc;
+ mm_segment_t seg;
- *lsmp = lsm;
+ seg = get_fs();
+ set_fs(KERNEL_DS);
+ rc = __lov_setstripe(exp, lsmp, lump);
+ set_fs(seg);
RETURN(rc);
}
+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;
+
+ ENTRY;
+ for (i = 0; i < lump->lmm_stripe_count; i++) {
+ __u32 len = sizeof(last_id);
+ oexp = lov->lov_tgts[lump->lmm_objects[i].l_ost_idx]->ltd_exp;
+ rc = obd_get_info(oexp, strlen("last_id"), "last_id",
+ &len, &last_id);
+ if (rc)
+ RETURN(rc);
+ if (lump->lmm_objects[i].l_object_id > last_id) {
+ CERROR("Setting EA for object > than last id on "
+ "ost idx %d "LPD64" > "LPD64" \n",
+ lump->lmm_objects[i].l_ost_idx,
+ lump->lmm_objects[i].l_object_id, last_id);
+ RETURN(-EINVAL);
+ }
+ }
+
+ rc = lov_setstripe(exp, lsmp, lump);
+ if (rc)
+ RETURN(rc);
+
+ for (i = 0; i < lump->lmm_stripe_count; i++) {
+ (*lsmp)->lsm_oinfo[i]->loi_ost_idx =
+ lump->lmm_objects[i].l_ost_idx;
+ (*lsmp)->lsm_oinfo[i]->loi_id = lump->lmm_objects[i].l_object_id;
+ (*lsmp)->lsm_oinfo[i]->loi_gr = lump->lmm_objects[i].l_object_gr;
+ }
+ RETURN(0);
+}
+
+
/* Retrieve object striping information.
*
- * @lmmu is a pointer to an in-core struct with lmm_ost_count indicating
+ * @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_MAGIC.
+ * lmm_magic must be LOV_USER_MAGIC.
*/
-int lov_getstripe(struct lustre_handle *conn, struct lov_stripe_md *lsm,
- struct lov_mds_md *lmmu)
+int lov_getstripe(struct obd_export *exp, struct lov_stripe_md *lsm,
+ 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;
+ /*
+ * XXX huge struct allocated on stack.
+ */
+ struct lov_user_md lum;
+ struct lov_mds_md *lmmk = NULL;
+ int rc, lmm_size;
+ mm_segment_t seg;
ENTRY;
if (!lsm)
RETURN(-ENODATA);
- rc = copy_from_user(&lmm, lmmu, sizeof(lmm));
+ /*
+ * "Switch to kernel segment" to allow copying from kernel space by
+ * copy_{to,from}_user().
+ */
+ seg = get_fs();
+ set_fs(KERNEL_DS);
+ rc = copy_from_user(&lum, lump, sizeof(lum));
if (rc)
- RETURN(-EFAULT);
-
- if (lmm.lmm_magic != LOV_MAGIC)
- RETURN(-EINVAL);
-
- ost_count = lov->desc.ld_tgt_count;
-
- /* XXX we _could_ check if indices > user lmm_ost_count are zero */
- if (lmm.lmm_ost_count < ost_count)
- RETURN(-EOVERFLOW);
-
- rc = lov_packmd(conn, &lmmk, lsm);
- if (rc < 0)
- RETURN(rc);
-
- lmm_size = rc;
- rc = 0;
-
- if (lmm_size && copy_to_user(lmmu, lmmk, lmm_size))
rc = -EFAULT;
-
- obd_free_wiremd(conn, &lmmk);
-
+ else if (lum.lmm_magic != LOV_USER_MAGIC)
+ rc = -EINVAL;
+ else {
+ 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 */
+ CLASSERT(sizeof lum == sizeof *lmmk);
+ 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, sizeof lum))
+ rc = -EFAULT;
+ } else if (lum.lmm_stripe_count < lmmk->lmm_stripe_count) {
+ rc = -EOVERFLOW;
+ } else if (copy_to_user(lump, lmmk, sizeof lum))
+ rc = -EFAULT;
+
+ obd_free_diskmd(exp, &lmmk);
+ }
+ set_fs(seg);
RETURN(rc);
}