* Copyright 2009 Sun Microsystems, Inc. All rights reserved
* Use is subject to license terms.
*
- * Copyright (c) 2012, 2016, Intel Corporation.
+ * Copyright (c) 2012, 2017, Intel Corporation.
*/
/*
* lustre/lod/lod_object.c
#define DEBUG_SUBSYSTEM S_MDS
+#include <linux/random.h>
+
#include <obd.h>
#include <obd_class.h>
#include <obd_support.h>
return dt_attr_get(env, dt_object_child(dt), attr);
}
+static inline void lod_adjust_stripe_info(struct lod_layout_component *comp,
+ struct lov_desc *desc)
+{
+ if (comp->llc_pattern != LOV_PATTERN_MDT) {
+ if (!comp->llc_stripe_count)
+ comp->llc_stripe_count =
+ desc->ld_default_stripe_count;
+ }
+ if (comp->llc_stripe_size <= 0)
+ comp->llc_stripe_size = desc->ld_default_stripe_size;
+}
+
int lod_obj_for_each_stripe(const struct lu_env *env, struct lod_object *lo,
- struct thandle *th, lod_obj_stripe_cb_t cb,
+ struct thandle *th,
struct lod_obj_stripe_cb_data *data)
{
struct lod_layout_component *lod_comp;
if (lod_comp->llc_stripe == NULL)
continue;
+ /* has stripe but not inited yet, this component has been
+ * declared to be created, but hasn't created yet.
+ */
+ if (!lod_comp_inited(lod_comp))
+ continue;
+
+ if (data->locd_comp_skip_cb &&
+ data->locd_comp_skip_cb(env, lo, i, data))
+ continue;
+
LASSERT(lod_comp->llc_stripe_count > 0);
for (j = 0; j < lod_comp->llc_stripe_count; j++) {
struct dt_object *dt = lod_comp->llc_stripe[j];
if (dt == NULL)
continue;
- rc = cb(env, lo, dt, th, j, data);
+ rc = data->locd_stripe_cb(env, lo, dt, th, i, j, data);
if (rc != 0)
RETURN(rc);
}
RETURN(0);
}
+static bool lod_obj_attr_set_comp_skip_cb(const struct lu_env *env,
+ struct lod_object *lo, int comp_idx,
+ struct lod_obj_stripe_cb_data *data)
+{
+ struct lod_layout_component *lod_comp = &lo->ldo_comp_entries[comp_idx];
+ bool skipped = false;
+
+ if (!(data->locd_attr->la_valid & LA_LAYOUT_VERSION))
+ return skipped;
+
+ switch (lo->ldo_flr_state) {
+ case LCM_FL_WRITE_PENDING: {
+ int i;
+
+ /* skip stale components */
+ if (lod_comp->llc_flags & LCME_FL_STALE) {
+ skipped = true;
+ break;
+ }
+
+ /* skip valid and overlapping components, therefore any
+ * attempts to write overlapped components will never succeed
+ * because client will get EINPROGRESS. */
+ for (i = 0; i < lo->ldo_comp_cnt; i++) {
+ if (i == comp_idx)
+ continue;
+
+ if (lo->ldo_comp_entries[i].llc_flags & LCME_FL_STALE)
+ continue;
+
+ if (lu_extent_is_overlapped(&lod_comp->llc_extent,
+ &lo->ldo_comp_entries[i].llc_extent)) {
+ skipped = true;
+ break;
+ }
+ }
+ break;
+ }
+ default:
+ LASSERTF(0, "impossible: %d\n", lo->ldo_flr_state);
+ case LCM_FL_SYNC_PENDING:
+ break;
+ }
+
+ CDEBUG(D_LAYOUT, DFID": %s to set component %x to version: %u\n",
+ PFID(lu_object_fid(&lo->ldo_obj.do_lu)),
+ skipped ? "skipped" : "chose", lod_comp->llc_id,
+ data->locd_attr->la_layout_version);
+
+ return skipped;
+}
+
static inline int
lod_obj_stripe_attr_set_cb(const struct lu_env *env, struct lod_object *lo,
struct dt_object *dt, struct thandle *th,
- int stripe_idx, struct lod_obj_stripe_cb_data *data)
+ int comp_idx, int stripe_idx,
+ struct lod_obj_stripe_cb_data *data)
{
if (data->locd_declare)
return lod_sub_declare_attr_set(env, dt, data->locd_attr, th);
+ if (data->locd_attr->la_valid & LA_LAYOUT_VERSION) {
+ CDEBUG(D_LAYOUT, DFID": set layout version: %u, comp_idx: %d\n",
+ PFID(lu_object_fid(&dt->do_lu)),
+ data->locd_attr->la_layout_version, comp_idx);
+ }
+
return lod_sub_attr_set(env, dt, data->locd_attr, th);
}
* speed up rename().
*/
if (!S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
- if (!(attr->la_valid & (LA_UID | LA_GID | LA_PROJID)))
+ if (!(attr->la_valid & LA_REMOTE_ATTR_SET))
RETURN(rc);
if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_OWNER))
RETURN(rc);
}
} else {
- struct lod_obj_stripe_cb_data data;
+ struct lod_obj_stripe_cb_data data = { { 0 } };
data.locd_attr = attr;
data.locd_declare = true;
- rc = lod_obj_for_each_stripe(env, lo, th,
- lod_obj_stripe_attr_set_cb, &data);
+ data.locd_stripe_cb = lod_obj_stripe_attr_set_cb;
+ rc = lod_obj_for_each_stripe(env, lo, th, &data);
}
if (rc)
RETURN(rc);
if (!S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
- if (!(attr->la_valid & (LA_UID | LA_GID | LA_PROJID)))
+ if (!(attr->la_valid & LA_REMOTE_ATTR_SET))
RETURN(rc);
if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_OWNER))
RETURN(rc);
}
+ /* FIXME: a tricky case in the code path of mdd_layout_change():
+ * the in-memory striping information has been freed in lod_xattr_set()
+ * due to layout change. It has to load stripe here again. It only
+ * changes flags of layout so declare_attr_set() is still accurate */
+ rc = lod_load_striping_locked(env, lo);
+ if (rc)
+ RETURN(rc);
+
if (!lod_obj_is_striped(dt))
RETURN(0);
break;
}
} else {
- struct lod_obj_stripe_cb_data data;
+ struct lod_obj_stripe_cb_data data = { { 0 } };
data.locd_attr = attr;
data.locd_declare = false;
- rc = lod_obj_for_each_stripe(env, lo, th,
- lod_obj_stripe_attr_set_cb, &data);
+ data.locd_comp_skip_cb = lod_obj_attr_set_comp_skip_cb;
+ data.locd_stripe_cb = lod_obj_stripe_attr_set_cb;
+ rc = lod_obj_for_each_stripe(env, lo, th, &data);
}
if (rc)
int rc = 0;
__u32 i;
__u32 j;
+ bool is_specific = false;
ENTRY;
/* The lum has been verifed in lod_verify_md_striping */
- LASSERT(le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC);
+ LASSERT(le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC ||
+ le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC_SPECIFIC);
LASSERT(le32_to_cpu(lum->lum_stripe_count) > 0);
stripe_count = le32_to_cpu(lum->lum_stripe_count);
/* Start index must be the master MDT */
master_index = lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id;
idx_array[0] = master_index;
+ if (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC_SPECIFIC) {
+ is_specific = true;
+ for (i = 1; i < stripe_count; i++)
+ idx_array[i] = le32_to_cpu(lum->lum_objects[i].lum_mds);
+ }
+
for (i = 0; i < stripe_count; i++) {
struct lod_tgt_desc *tgt = NULL;
struct dt_object *dto;
CDEBUG(D_INFO, "try idx %d, mdt cnt %u, allocated %u\n",
idx, lod->lod_remote_mdt_count + 1, i);
- if (likely(!OBD_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE))) {
+ if (likely(!is_specific &&
+ !OBD_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE))) {
/* check whether the idx already exists
* in current allocated array */
for (k = 0; k < i; k++) {
idx, i, PFID(&fid));
idx_array[i] = idx;
/* Set the start index for next stripe allocation */
- if (i < stripe_count - 1)
+ if (!is_specific && i < stripe_count - 1)
idx_array[i + 1] = (idx + 1) %
(lod->lod_remote_mdt_count + 1);
/* tgt_dt and fid must be ready after search avaible OSP
struct thandle *th)
{
struct lod_object *lo = lod_dt_obj(dt);
- struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
- struct lmv_user_md_v1 *lum;
+ struct lmv_user_md_v1 *lum = lum_buf->lb_buf;
int rc;
ENTRY;
- lum = lum_buf->lb_buf;
LASSERT(lum != NULL);
CDEBUG(D_INFO, "lum magic = %x count = %u offset = %d\n",
le32_to_cpu(lum->lum_magic), le32_to_cpu(lum->lum_stripe_count),
(int)le32_to_cpu(lum->lum_stripe_offset));
- if (le32_to_cpu(lum->lum_stripe_count) == 0)
+ if (lo->ldo_dir_stripe_count == 0)
GOTO(out, rc = 0);
- rc = lod_verify_md_striping(lod, lum);
- if (rc != 0)
- GOTO(out, rc);
-
/* prepare dir striped objects */
rc = lod_prep_md_striped_create(env, dt, attr, lum, dof, th);
if (rc != 0) {
if (rc != 0)
RETURN(rc);
} else if (strcmp(name, XATTR_NAME_LOV) == 0) {
- rc = lod_verify_striping(d, buf, false, 0);
+ rc = lod_verify_striping(d, lo, buf, false);
if (rc != 0)
RETURN(rc);
}
lod_obj_stripe_replace_parent_fid_cb(const struct lu_env *env,
struct lod_object *lo,
struct dt_object *dt, struct thandle *th,
- int stripe_idx,
+ int comp_idx, int stripe_idx,
struct lod_obj_stripe_cb_data *data)
{
struct lod_thread_info *info = lod_env_info(env);
+ struct lod_layout_component *comp = &lo->ldo_comp_entries[comp_idx];
struct filter_fid *ff = &info->lti_ff;
struct lu_buf *buf = &info->lti_buf;
int rc;
buf->lb_buf = ff;
buf->lb_len = sizeof(*ff);
rc = dt_xattr_get(env, dt, buf, XATTR_NAME_FID);
- if (rc == -ENODATA)
- return 0;
-
- if (rc < 0)
+ if (rc < 0) {
+ if (rc == -ENODATA)
+ return 0;
return rc;
+ }
+ filter_fid_le_to_cpu(ff, ff, sizeof(*ff));
+ if (lu_fid_eq(lu_object_fid(&lo->ldo_obj.do_lu), &ff->ff_parent) &&
+ ff->ff_layout.ol_comp_id == comp->llc_id)
+ return 0;
+
+ /* rewrite filter_fid */
+ memset(ff, 0, sizeof(*ff));
ff->ff_parent = *lu_object_fid(&lo->ldo_obj.do_lu);
ff->ff_parent.f_ver = stripe_idx;
- fid_cpu_to_le(&ff->ff_parent, &ff->ff_parent);
+ ff->ff_layout.ol_stripe_size = comp->llc_stripe_size;
+ ff->ff_layout.ol_stripe_count = comp->llc_stripe_count;
+ ff->ff_layout.ol_comp_id = comp->llc_id;
+ ff->ff_layout.ol_comp_start = comp->llc_extent.e_start;
+ ff->ff_layout.ol_comp_end = comp->llc_extent.e_end;
+ filter_fid_cpu_to_le(ff, ff, sizeof(*ff));
+
if (data->locd_declare)
rc = lod_sub_declare_xattr_set(env, dt, buf, XATTR_NAME_FID,
LU_XATTR_REPLACE, th);
struct lod_thread_info *info = lod_env_info(env);
struct lu_buf *buf = &info->lti_buf;
struct filter_fid *ff;
- struct lod_obj_stripe_cb_data data;
+ struct lod_obj_stripe_cb_data data = { { 0 } };
int rc;
ENTRY;
buf->lb_len = info->lti_ea_store_size;
data.locd_declare = declare;
- rc = lod_obj_for_each_stripe(env, lo, th,
- lod_obj_stripe_replace_parent_fid_cb,
- &data);
+ data.locd_stripe_cb = lod_obj_stripe_replace_parent_fid_cb;
+ rc = lod_obj_for_each_stripe(env, lo, th, &data);
RETURN(rc);
}
struct thandle *th)
{
struct lod_thread_info *info = lod_env_info(env);
- struct lod_layout_component *comp_array, *lod_comp;
+ struct lod_layout_component *comp_array, *lod_comp, *old_array;
struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
struct dt_object *next = dt_object_child(dt);
struct lov_desc *desc = &d->lod_desc;
struct lov_user_md_v3 *v3;
struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
__u32 magic;
- __u64 prev_end;
- int i, rc, array_cnt;
+ int i, rc, array_cnt, old_array_cnt;
ENTRY;
LASSERT(lo->ldo_is_composite);
- prev_end = lo->ldo_comp_entries[lo->ldo_comp_cnt - 1].llc_extent.e_end;
- rc = lod_verify_striping(d, buf, false, prev_end);
+ if (lo->ldo_flr_state != LCM_FL_NONE)
+ RETURN(-EBUSY);
+
+ rc = lod_verify_striping(d, lo, buf, false);
if (rc != 0)
RETURN(rc);
lod_comp->llc_extent.e_start = ext->e_start;
lod_comp->llc_extent.e_end = ext->e_end;
lod_comp->llc_stripe_offset = v1->lmm_stripe_offset;
+ lod_comp->llc_flags = comp_v1->lcm_entries[i].lcme_flags;
lod_comp->llc_stripe_count = v1->lmm_stripe_count;
- if (!lod_comp->llc_stripe_count ||
- lod_comp->llc_stripe_count == (__u16)-1)
- lod_comp->llc_stripe_count =
- desc->ld_default_stripe_count;
lod_comp->llc_stripe_size = v1->lmm_stripe_size;
- if (!lod_comp->llc_stripe_size)
- lod_comp->llc_stripe_size =
- desc->ld_default_stripe_size;
+ lod_adjust_stripe_info(lod_comp, desc);
if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
v3 = (struct lov_user_md_v3 *) v1;
}
}
- OBD_FREE(lo->ldo_comp_entries, sizeof(*lod_comp) * lo->ldo_comp_cnt);
+ old_array = lo->ldo_comp_entries;
+ old_array_cnt = lo->ldo_comp_cnt;
+
lo->ldo_comp_entries = comp_array;
lo->ldo_comp_cnt = array_cnt;
+
/* No need to increase layout generation here, it will be increased
* later when generating component ID for the new components */
info->lti_buf.lb_len = lod_comp_md_size(lo, false);
rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
XATTR_NAME_LOV, 0, th);
- if (rc)
+ if (rc) {
+ lo->ldo_comp_entries = old_array;
+ lo->ldo_comp_cnt = old_array_cnt;
GOTO(error, rc);
+ }
+
+ OBD_FREE(old_array, sizeof(*lod_comp) * old_array_cnt);
+
+ LASSERT(lo->ldo_mirror_count == 1);
+ lo->ldo_mirrors[0].lme_end = array_cnt - 1;
RETURN(0);
struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
struct lod_object *lo = lod_dt_obj(dt);
struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
- __u32 magic, id;
+ __u32 magic;
int i, j, rc;
bool changed = false;
ENTRY;
}
for (i = 0; i < comp_v1->lcm_entry_count; i++) {
- id = comp_v1->lcm_entries[i].lcme_id;
+ __u32 id = comp_v1->lcm_entries[i].lcme_id;
+ __u32 flags = comp_v1->lcm_entries[i].lcme_flags;
+
+ if (flags & LCME_FL_INIT) {
+ if (changed)
+ lod_object_free_striping(env, lo);
+ RETURN(-EINVAL);
+ }
for (j = 0; j < lo->ldo_comp_cnt; j++) {
lod_comp = &lo->ldo_comp_entries[j];
- if (id == lod_comp->llc_id || id == LCME_ID_ALL) {
- lod_comp->llc_flags =
- comp_v1->lcm_entries[i].lcme_flags;
- changed = true;
+ if (id != lod_comp->llc_id)
+ continue;
+
+ if (flags & LCME_FL_NEG) {
+ flags &= ~LCME_FL_NEG;
+ lod_comp->llc_flags &= ~flags;
+ } else {
+ lod_comp->llc_flags |= flags;
}
+ changed = true;
}
}
lod_obj_inc_layout_gen(lo);
info->lti_buf.lb_len = lod_comp_md_size(lo, false);
- rc = lod_sub_declare_xattr_set(env, dt, &info->lti_buf,
- XATTR_NAME_LOV, 0, th);
+ rc = lod_sub_declare_xattr_set(env, dt_object_child(dt), &info->lti_buf,
+ XATTR_NAME_LOV, LU_XATTR_REPLACE, th);
RETURN(rc);
}
LASSERT(lo->ldo_is_composite);
- rc = lod_verify_striping(d, buf, false, 0);
- if (rc != 0)
- RETURN(rc);
+ if (lo->ldo_flr_state != LCM_FL_NONE)
+ RETURN(-EBUSY);
magic = comp_v1->lcm_magic;
if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
RETURN(-EINVAL);
}
+ if (id == LCME_ID_INVAL && !flags) {
+ CDEBUG(D_LAYOUT, "%s: no id or flags specified.\n",
+ lod2obd(d)->obd_name);
+ RETURN(-EINVAL);
+ }
+
if (flags & LCME_FL_NEG) {
neg_flags = flags & ~LCME_FL_NEG;
flags = 0;
}
/**
+ * Convert a plain file lov_mds_md to a composite layout.
+ *
+ * \param[in,out] info the thread info::lti_ea_store buffer contains little
+ * endian plain file layout
+ *
+ * \retval 0 on success, <0 on failure
+ */
+static int lod_layout_convert(struct lod_thread_info *info)
+{
+ struct lov_mds_md *lmm = info->lti_ea_store;
+ struct lov_mds_md *lmm_save;
+ struct lov_comp_md_v1 *lcm;
+ struct lov_comp_md_entry_v1 *lcme;
+ size_t size;
+ __u32 blob_size;
+ int rc = 0;
+ ENTRY;
+
+ /* realloc buffer to a composite layout which contains one component */
+ blob_size = lov_mds_md_size(le16_to_cpu(lmm->lmm_stripe_count),
+ le32_to_cpu(lmm->lmm_magic));
+ size = sizeof(*lcm) + sizeof(*lcme) + blob_size;
+
+ OBD_ALLOC_LARGE(lmm_save, blob_size);
+ if (!lmm_save)
+ GOTO(out, rc = -ENOMEM);
+
+ memcpy(lmm_save, lmm, blob_size);
+
+ if (info->lti_ea_store_size < size) {
+ rc = lod_ea_store_resize(info, size);
+ if (rc)
+ GOTO(out, rc);
+ }
+
+ lcm = info->lti_ea_store;
+ lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_COMP_V1);
+ lcm->lcm_size = cpu_to_le32(size);
+ lcm->lcm_layout_gen = cpu_to_le32(le16_to_cpu(
+ lmm_save->lmm_layout_gen));
+ lcm->lcm_flags = cpu_to_le16(LCM_FL_NONE);
+ lcm->lcm_entry_count = cpu_to_le16(1);
+ lcm->lcm_mirror_count = 0;
+
+ lcme = &lcm->lcm_entries[0];
+ lcme->lcme_flags = cpu_to_le32(LCME_FL_INIT);
+ lcme->lcme_extent.e_start = 0;
+ lcme->lcme_extent.e_end = cpu_to_le64(OBD_OBJECT_EOF);
+ lcme->lcme_offset = cpu_to_le32(sizeof(*lcm) + sizeof(*lcme));
+ lcme->lcme_size = cpu_to_le32(blob_size);
+
+ memcpy((char *)lcm + lcme->lcme_offset, (char *)lmm_save, blob_size);
+
+ EXIT;
+out:
+ if (lmm_save)
+ OBD_FREE_LARGE(lmm_save, blob_size);
+ return rc;
+}
+
+/**
+ * Merge layouts to form a mirrored file.
+ */
+static int lod_declare_layout_merge(const struct lu_env *env,
+ struct dt_object *dt, const struct lu_buf *mbuf,
+ struct thandle *th)
+{
+ struct lod_thread_info *info = lod_env_info(env);
+ struct lu_buf *buf = &info->lti_buf;
+ struct lod_object *lo = lod_dt_obj(dt);
+ struct lov_comp_md_v1 *lcm;
+ struct lov_comp_md_v1 *cur_lcm;
+ struct lov_comp_md_v1 *merge_lcm;
+ struct lov_comp_md_entry_v1 *lcme;
+ size_t size = 0;
+ size_t offset;
+ __u16 cur_entry_count;
+ __u16 merge_entry_count;
+ __u32 id = 0;
+ __u16 mirror_id = 0;
+ __u32 mirror_count;
+ int rc, i;
+ ENTRY;
+
+ merge_lcm = mbuf->lb_buf;
+ if (mbuf->lb_len < sizeof(*merge_lcm))
+ RETURN(-EINVAL);
+
+ /* must be an existing layout from disk */
+ if (le32_to_cpu(merge_lcm->lcm_magic) != LOV_MAGIC_COMP_V1)
+ RETURN(-EINVAL);
+
+ merge_entry_count = le16_to_cpu(merge_lcm->lcm_entry_count);
+
+ /* do not allow to merge two mirrored files */
+ if (le16_to_cpu(merge_lcm->lcm_mirror_count))
+ RETURN(-EBUSY);
+
+ /* verify the target buffer */
+ rc = lod_get_lov_ea(env, lo);
+ if (rc <= 0)
+ RETURN(rc ? : -ENODATA);
+
+ cur_lcm = info->lti_ea_store;
+ switch (le32_to_cpu(cur_lcm->lcm_magic)) {
+ case LOV_MAGIC_V1:
+ case LOV_MAGIC_V3:
+ rc = lod_layout_convert(info);
+ break;
+ case LOV_MAGIC_COMP_V1:
+ rc = 0;
+ break;
+ default:
+ rc = -EINVAL;
+ }
+ if (rc)
+ RETURN(rc);
+
+ /* info->lti_ea_store could be reallocated in lod_layout_convert() */
+ cur_lcm = info->lti_ea_store;
+ cur_entry_count = le16_to_cpu(cur_lcm->lcm_entry_count);
+
+ /* 'lcm_mirror_count + 1' is the current # of mirrors the file has */
+ mirror_count = le16_to_cpu(cur_lcm->lcm_mirror_count) + 1;
+ if (mirror_count + 1 > LUSTRE_MIRROR_COUNT_MAX)
+ RETURN(-ERANGE);
+
+ /* size of new layout */
+ size = le32_to_cpu(cur_lcm->lcm_size) +
+ le32_to_cpu(merge_lcm->lcm_size) - sizeof(*cur_lcm);
+
+ memset(buf, 0, sizeof(*buf));
+ lu_buf_alloc(buf, size);
+ if (buf->lb_buf == NULL)
+ RETURN(-ENOMEM);
+
+ lcm = buf->lb_buf;
+ memcpy(lcm, cur_lcm, sizeof(*lcm) + cur_entry_count * sizeof(*lcme));
+
+ offset = sizeof(*lcm) +
+ sizeof(*lcme) * (cur_entry_count + merge_entry_count);
+ for (i = 0; i < cur_entry_count; i++) {
+ struct lov_comp_md_entry_v1 *cur_lcme;
+
+ lcme = &lcm->lcm_entries[i];
+ cur_lcme = &cur_lcm->lcm_entries[i];
+
+ lcme->lcme_offset = cpu_to_le32(offset);
+ memcpy((char *)lcm + offset,
+ (char *)cur_lcm + le32_to_cpu(cur_lcme->lcme_offset),
+ le32_to_cpu(lcme->lcme_size));
+
+ offset += le32_to_cpu(lcme->lcme_size);
+
+ if (mirror_count == 1) {
+ /* new mirrored file, create new mirror ID */
+ id = pflr_id(1, i + 1);
+ lcme->lcme_id = cpu_to_le32(id);
+ }
+
+ id = MAX(le32_to_cpu(lcme->lcme_id), id);
+ }
+
+ mirror_id = mirror_id_of(id) + 1;
+ for (i = 0; i < merge_entry_count; i++) {
+ struct lov_comp_md_entry_v1 *merge_lcme;
+
+ merge_lcme = &merge_lcm->lcm_entries[i];
+ lcme = &lcm->lcm_entries[cur_entry_count + i];
+
+ *lcme = *merge_lcme;
+ lcme->lcme_offset = cpu_to_le32(offset);
+
+ id = pflr_id(mirror_id, i + 1);
+ lcme->lcme_id = cpu_to_le32(id);
+
+ memcpy((char *)lcm + offset,
+ (char *)merge_lcm + le32_to_cpu(merge_lcme->lcme_offset),
+ le32_to_cpu(lcme->lcme_size));
+
+ offset += le32_to_cpu(lcme->lcme_size);
+ }
+
+ /* fixup layout information */
+ lod_obj_inc_layout_gen(lo);
+ lcm->lcm_layout_gen = cpu_to_le32(lo->ldo_layout_gen);
+ lcm->lcm_size = cpu_to_le32(size);
+ lcm->lcm_entry_count = cpu_to_le16(cur_entry_count + merge_entry_count);
+ lcm->lcm_mirror_count = cpu_to_le16(mirror_count);
+ if ((le16_to_cpu(lcm->lcm_flags) & LCM_FL_FLR_MASK) == LCM_FL_NONE)
+ lcm->lcm_flags = cpu_to_le32(LCM_FL_RDONLY);
+
+ LASSERT(dt_write_locked(env, dt_object_child(dt)));
+ lod_object_free_striping(env, lo);
+ rc = lod_parse_striping(env, lo, buf);
+ if (rc)
+ GOTO(out, rc);
+
+ rc = lod_sub_declare_xattr_set(env, dt_object_child(dt), buf,
+ XATTR_NAME_LOV, LU_XATTR_REPLACE, th);
+
+out:
+ lu_buf_free(buf);
+ RETURN(rc);
+}
+
+/**
+ * Split layouts, just set the LOVEA with the layout from mbuf.
+ */
+static int lod_declare_layout_split(const struct lu_env *env,
+ struct dt_object *dt, const struct lu_buf *mbuf,
+ struct thandle *th)
+{
+ struct lod_object *lo = lod_dt_obj(dt);
+ struct lov_comp_md_v1 *lcm = mbuf->lb_buf;
+ int rc;
+ ENTRY;
+
+ lod_obj_inc_layout_gen(lo);
+ lcm->lcm_layout_gen = cpu_to_le32(lo->ldo_layout_gen);
+
+ lod_object_free_striping(env, lo);
+ rc = lod_parse_striping(env, lo, mbuf);
+ if (rc)
+ RETURN(rc);
+
+ rc = lod_sub_declare_xattr_set(env, dt_object_child(dt), mbuf,
+ XATTR_NAME_LOV, LU_XATTR_REPLACE, th);
+ RETURN(rc);
+}
+
+/**
* Implementation of dt_object_operations::do_declare_xattr_set.
*
* \see dt_object_operations::do_declare_xattr_set() in the API description
ENTRY;
mode = dt->do_lu.lo_header->loh_attr & S_IFMT;
- if ((S_ISREG(mode) || mode == 0) && !(fl & LU_XATTR_REPLACE) &&
+ if ((S_ISREG(mode) || mode == 0) &&
+ !(fl & (LU_XATTR_REPLACE | LU_XATTR_MERGE | LU_XATTR_SPLIT)) &&
(strcmp(name, XATTR_NAME_LOV) == 0 ||
strcmp(name, XATTR_LUSTRE_LOV) == 0)) {
/*
attr->la_mode = S_IFREG;
}
rc = lod_declare_striped_create(env, dt, attr, buf, th);
+ } else if (fl & LU_XATTR_MERGE) {
+ LASSERT(strcmp(name, XATTR_NAME_LOV) == 0 ||
+ strcmp(name, XATTR_LUSTRE_LOV) == 0);
+ rc = lod_declare_layout_merge(env, dt, buf, th);
+ } else if (fl & LU_XATTR_SPLIT) {
+ LASSERT(strcmp(name, XATTR_NAME_LOV) == 0 ||
+ strcmp(name, XATTR_LUSTRE_LOV) == 0);
+ rc = lod_declare_layout_split(env, dt, buf, th);
} else if (S_ISREG(mode) &&
strlen(name) > strlen(XATTR_LUSTRE_LOV) + 1 &&
strncmp(name, XATTR_LUSTRE_LOV,
* \param[in] env execution environment
* \param[in] dt object
* \param[in] attr attributes the stripes will be created with
+ * \param[in] lmu lmv_user_md if MDT indices are specified
* \param[in] dof format of stripes (see OSD API description)
* \param[in] th transaction handle
* \param[in] declare where to call "declare" or "execute" methods
static int lod_dir_striping_create_internal(const struct lu_env *env,
struct dt_object *dt,
struct lu_attr *attr,
+ const struct lu_buf *lmu,
struct dt_object_format *dof,
struct thandle *th,
bool declare)
if (!LMVEA_DELETE_VALUES(lo->ldo_dir_stripe_count,
lo->ldo_dir_stripe_offset)) {
- struct lmv_user_md_v1 *v1 = info->lti_ea_store;
- int stripe_count = lo->ldo_dir_stripe_count;
+ if (!lmu) {
+ struct lmv_user_md_v1 *v1 = info->lti_ea_store;
+ int stripe_count = lo->ldo_dir_stripe_count;
- if (info->lti_ea_store_size < sizeof(*v1)) {
- rc = lod_ea_store_resize(info, sizeof(*v1));
- if (rc != 0)
- RETURN(rc);
- v1 = info->lti_ea_store;
- }
+ if (info->lti_ea_store_size < sizeof(*v1)) {
+ rc = lod_ea_store_resize(info, sizeof(*v1));
+ if (rc != 0)
+ RETURN(rc);
+ v1 = info->lti_ea_store;
+ }
- memset(v1, 0, sizeof(*v1));
- v1->lum_magic = cpu_to_le32(LMV_USER_MAGIC);
- v1->lum_stripe_count = cpu_to_le32(stripe_count);
- v1->lum_stripe_offset =
- cpu_to_le32(lo->ldo_dir_stripe_offset);
+ memset(v1, 0, sizeof(*v1));
+ v1->lum_magic = cpu_to_le32(LMV_USER_MAGIC);
+ v1->lum_stripe_count = cpu_to_le32(stripe_count);
+ v1->lum_stripe_offset =
+ cpu_to_le32(lo->ldo_dir_stripe_offset);
- info->lti_buf.lb_buf = v1;
- info->lti_buf.lb_len = sizeof(*v1);
+ info->lti_buf.lb_buf = v1;
+ info->lti_buf.lb_len = sizeof(*v1);
+ lmu = &info->lti_buf;
+ }
if (declare)
- rc = lod_declare_xattr_set_lmv(env, dt, attr,
- &info->lti_buf, dof, th);
+ rc = lod_declare_xattr_set_lmv(env, dt, attr, lmu, dof,
+ th);
else
- rc = lod_xattr_set_lmv(env, dt, &info->lti_buf,
- XATTR_NAME_LMV, 0, th);
+ rc = lod_xattr_set_lmv(env, dt, lmu, XATTR_NAME_LMV, 0,
+ th);
if (rc != 0)
RETURN(rc);
}
static int lod_declare_dir_striping_create(const struct lu_env *env,
struct dt_object *dt,
struct lu_attr *attr,
+ struct lu_buf *lmu,
struct dt_object_format *dof,
struct thandle *th)
{
- return lod_dir_striping_create_internal(env, dt, attr, dof, th, true);
+ return lod_dir_striping_create_internal(env, dt, attr, lmu, dof, th,
+ true);
}
static int lod_dir_striping_create(const struct lu_env *env,
struct dt_object_format *dof,
struct thandle *th)
{
- return lod_dir_striping_create_internal(env, dt, attr, dof, th, false);
+ return lod_dir_striping_create_internal(env, dt, attr, NULL, dof, th,
+ false);
}
/**
sizeof(*comp_array) * lo->ldo_comp_cnt);
lo->ldo_comp_entries = comp_array;
lo->ldo_comp_cnt = left;
+
+ LASSERT(lo->ldo_mirror_count == 1);
+ lo->ldo_mirrors[0].lme_end = left - 1;
lod_obj_inc_layout_gen(lo);
} else {
lod_free_comp_entries(lo);
return rc;
}
+
+static int lod_get_default_lov_striping(const struct lu_env *env,
+ struct lod_object *lo,
+ struct lod_default_striping *lds);
/**
* Implementation of dt_object_operations::do_xattr_set.
*
if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
strcmp(name, XATTR_NAME_LOV) == 0) {
- /* default LOVEA */
- rc = lod_xattr_set_lov_on_dir(env, dt, buf, name, fl, th);
+ struct lod_thread_info *info = lod_env_info(env);
+ struct lod_default_striping *lds = &info->lti_def_striping;
+ struct lov_user_md_v1 *v1 = buf->lb_buf;
+ char pool[LOV_MAXPOOLNAME + 1];
+ bool is_del;
+
+ /* get existing striping config */
+ rc = lod_get_default_lov_striping(env, lod_dt_obj(dt), lds);
+ if (rc)
+ RETURN(rc);
+
+ memset(pool, 0, sizeof(pool));
+ if (lds->lds_def_striping_set == 1)
+ lod_layout_get_pool(lds->lds_def_comp_entries,
+ lds->lds_def_comp_cnt, pool,
+ sizeof(pool));
+
+ is_del = LOVEA_DELETE_VALUES(v1->lmm_stripe_size,
+ v1->lmm_stripe_count,
+ v1->lmm_stripe_offset,
+ NULL);
+
+ /* Retain the pool name if it is not given */
+ if (v1->lmm_magic == LOV_USER_MAGIC_V1 && pool[0] != '\0' &&
+ !is_del) {
+ struct lod_thread_info *info = lod_env_info(env);
+ struct lov_user_md_v3 *v3 = info->lti_ea_store;
+
+ memset(v3, 0, sizeof(*v3));
+ v3->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V3);
+ v3->lmm_pattern = cpu_to_le32(v1->lmm_pattern);
+ v3->lmm_stripe_count =
+ cpu_to_le32(v1->lmm_stripe_count);
+ v3->lmm_stripe_offset =
+ cpu_to_le32(v1->lmm_stripe_offset);
+ v3->lmm_stripe_size = cpu_to_le32(v1->lmm_stripe_size);
+
+ strlcpy(v3->lmm_pool_name, pool,
+ sizeof(v3->lmm_pool_name));
+
+ info->lti_buf.lb_buf = v3;
+ info->lti_buf.lb_len = sizeof(*v3);
+ rc = lod_xattr_set_lov_on_dir(env, dt, &info->lti_buf,
+ name, fl, th);
+ } else {
+ rc = lod_xattr_set_lov_on_dir(env, dt, buf, name,
+ fl, th);
+ }
+
+ if (lds->lds_def_striping_set == 1 &&
+ lds->lds_def_comp_entries != NULL)
+ lod_free_def_comp_entries(lds);
+
RETURN(rc);
} else if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
struct lov_user_md_v3 *v3 = NULL;
struct lov_comp_md_v1 *comp_v1 = NULL;
__u16 comp_cnt;
+ __u16 mirror_cnt;
bool composite;
int rc, i;
ENTRY;
comp_cnt = comp_v1->lcm_entry_count;
if (comp_cnt == 0)
RETURN(-EINVAL);
+ mirror_cnt = comp_v1->lcm_mirror_count + 1;
composite = true;
} else {
comp_cnt = 1;
+ mirror_cnt = 0;
composite = false;
}
RETURN(rc);
lds->lds_def_comp_cnt = comp_cnt;
- lds->lds_def_striping_is_composite = composite ? 1 : 0;
+ lds->lds_def_striping_is_composite = composite;
+ lds->lds_def_mirror_cnt = mirror_cnt;
for (i = 0; i < comp_cnt; i++) {
struct lod_layout_component *lod_comp;
int i, rc;
if (lds->lds_def_striping_set && S_ISREG(mode)) {
- rc = lod_alloc_comp_entries(lo, lds->lds_def_comp_cnt);
+ rc = lod_alloc_comp_entries(lo, lds->lds_def_mirror_cnt,
+ lds->lds_def_comp_cnt);
if (rc != 0)
return;
lo->ldo_is_composite = lds->lds_def_striping_is_composite;
+ if (lds->lds_def_mirror_cnt > 1)
+ lo->ldo_flr_state = LCM_FL_RDONLY;
for (i = 0; i < lo->ldo_comp_cnt; i++) {
struct lod_layout_component *obj_comp =
if (!lo->ldo_is_composite)
continue;
- if (obj_comp->llc_stripe_count <= 0 &&
- obj_comp->llc_pattern != LOV_PATTERN_MDT)
- obj_comp->llc_stripe_count =
- desc->ld_default_stripe_count;
- if (obj_comp->llc_stripe_size <= 0)
- obj_comp->llc_stripe_size =
- desc->ld_default_stripe_size;
+ lod_adjust_stripe_info(obj_comp, desc);
}
} else if (lds->lds_dir_def_striping_set && S_ISDIR(mode)) {
if (lo->ldo_dir_stripe_count == 0)
nextc->do_ops->do_ah_init(env, ah, nextp, nextc, child_mode);
if (S_ISDIR(child_mode)) {
+ const struct lmv_user_md_v1 *lum1 = ah->dah_eadata;
+
/* other default values are 0 */
lc->ldo_dir_stripe_offset = -1;
lc->ldo_def_striping = lds;
/* It should always honour the specified stripes */
+ /* Note: old client (< 2.7)might also do lfs mkdir, whose EA
+ * will have old magic. In this case, we should ignore the
+ * stripe count and try to create dir by default stripe.
+ */
if (ah->dah_eadata != NULL && ah->dah_eadata_len != 0 &&
- lod_verify_md_striping(d, ah->dah_eadata) == 0) {
- const struct lmv_user_md_v1 *lum1 = ah->dah_eadata;
-
+ (le32_to_cpu(lum1->lum_magic) == LMV_USER_MAGIC ||
+ le32_to_cpu(lum1->lum_magic) == LMV_USER_MAGIC_SPECIFIC)) {
lc->ldo_dir_stripe_count =
le32_to_cpu(lum1->lum_stripe_count);
lc->ldo_dir_stripe_offset =
* in config log, use them.
*/
if (lod_need_inherit_more(lc, false)) {
-
if (lc->ldo_comp_cnt == 0) {
- rc = lod_alloc_comp_entries(lc, 1);
+ rc = lod_alloc_comp_entries(lc, 0, 1);
if (rc)
/* fail to allocate memory, will create a
* non-striped file. */
LASSERT(!lc->ldo_is_composite);
lod_comp = &lc->ldo_comp_entries[0];
desc = &d->lod_desc;
- if (lod_comp->llc_stripe_count <= 0)
- lod_comp->llc_stripe_count =
- desc->ld_default_stripe_count;
- if (lod_comp->llc_stripe_size <= 0)
- lod_comp->llc_stripe_size =
- desc->ld_default_stripe_size;
+ lod_adjust_stripe_info(lod_comp, desc);
}
EXIT;
struct lu_attr *attr = &lod_env_info(env)->lti_attr;
uint64_t size, offs;
int i, rc, stripe, stripe_count = 0, stripe_size = 0;
+ struct lu_extent size_ext;
ENTRY;
if (!lod_obj_is_striped(dt))
if (size == 0)
RETURN(0);
+ size_ext = (typeof(size_ext)){ .e_start = size - 1, .e_end = size };
for (i = 0; i < lo->ldo_comp_cnt; i++) {
struct lod_layout_component *lod_comp;
struct lu_extent *extent;
continue;
extent = &lod_comp->llc_extent;
- CDEBUG(D_INFO, "%lld [%lld, %lld)\n",
- size, extent->e_start, extent->e_end);
+ CDEBUG(D_INFO, "%lld "DEXT"\n", size, PEXT(extent));
if (!lo->ldo_is_composite ||
- (size >= extent->e_start && size < extent->e_end)) {
+ lu_extent_is_overlapped(extent, &size_ext)) {
objects = lod_comp->llc_stripe;
stripe_count = lod_comp->llc_stripe_count;
stripe_size = lod_comp->llc_stripe_size;
- break;
- }
- }
-
- if (stripe_count == 0)
- RETURN(0);
- LASSERT(objects != NULL && stripe_size != 0);
+ /* next mirror */
+ if (stripe_count == 0)
+ continue;
- /* ll_do_div64(a, b) returns a % b, and a = a / b */
- ll_do_div64(size, (__u64)stripe_size);
- stripe = ll_do_div64(size, (__u64)stripe_count);
- LASSERT(objects[stripe] != NULL);
+ LASSERT(objects != NULL && stripe_size != 0);
+ /* ll_do_div64(a, b) returns a % b, and a = a / b */
+ ll_do_div64(size, (__u64)stripe_size);
+ stripe = ll_do_div64(size, (__u64)stripe_count);
+ LASSERT(objects[stripe] != NULL);
- size = size * stripe_size;
- offs = attr->la_size;
- size += ll_do_div64(offs, stripe_size);
+ size = size * stripe_size;
+ offs = attr->la_size;
+ size += ll_do_div64(offs, stripe_size);
- attr->la_valid = LA_SIZE;
- attr->la_size = size;
+ attr->la_valid = LA_SIZE;
+ attr->la_size = size;
- rc = lod_sub_declare_attr_set(env, objects[stripe], attr, th);
+ rc = lod_sub_declare_attr_set(env, objects[stripe],
+ attr, th);
+ }
+ }
RETURN(rc);
}
NULL, th);
} else if (dof->dof_type == DFT_DIR) {
struct seq_server_site *ss;
+ struct lu_buf buf = { NULL };
+ struct lu_buf *lmu = NULL;
ss = lu_site2seq(dt->do_lu.lo_dev->ld_site);
else
GOTO(out, rc = -EINVAL);
}
+ } else if (hint && hint->dah_eadata) {
+ lmu = &buf;
+ lmu->lb_buf = (void *)hint->dah_eadata;
+ lmu->lb_len = hint->dah_eadata_len;
}
- rc = lod_declare_dir_striping_create(env, dt, attr, dof, th);
+ rc = lod_declare_dir_striping_create(env, dt, attr, lmu, dof,
+ th);
}
out:
/* failed to create striping or to set initial size, let's reset
}
/**
+ * Generate component ID for new created component.
+ *
+ * \param[in] lo LOD object
+ * \param[in] comp_idx index of ldo_comp_entries
+ *
+ * \retval component ID on success
+ * \retval LCME_ID_INVAL on failure
+ */
+static __u32 lod_gen_component_id(struct lod_object *lo,
+ int mirror_id, int comp_idx)
+{
+ struct lod_layout_component *lod_comp;
+ __u32 id, start, end;
+ int i;
+
+ LASSERT(lo->ldo_comp_entries[comp_idx].llc_id == LCME_ID_INVAL);
+
+ lod_obj_inc_layout_gen(lo);
+ id = lo->ldo_layout_gen;
+ if (likely(id <= SEQ_ID_MAX))
+ RETURN(pflr_id(mirror_id, id & SEQ_ID_MASK));
+
+ /* Layout generation wraps, need to check collisions. */
+ start = id & SEQ_ID_MASK;
+ end = SEQ_ID_MAX;
+again:
+ for (id = start; id <= end; id++) {
+ for (i = 0; i < lo->ldo_comp_cnt; i++) {
+ lod_comp = &lo->ldo_comp_entries[i];
+ if (pflr_id(mirror_id, id) == lod_comp->llc_id)
+ break;
+ }
+ /* Found the ununsed ID */
+ if (i == lo->ldo_comp_cnt)
+ RETURN(pflr_id(mirror_id, id));
+ }
+ if (end == LCME_ID_MAX) {
+ start = 1;
+ end = min(lo->ldo_layout_gen & LCME_ID_MASK,
+ (__u32)(LCME_ID_MAX - 1));
+ goto again;
+ }
+
+ RETURN(LCME_ID_INVAL);
+}
+
+/**
* Creation of a striped regular object.
*
* The function is called to create the stripe objects for a regular
{
struct lod_layout_component *lod_comp;
struct lod_object *lo = lod_dt_obj(dt);
+ __u16 mirror_id;
int rc = 0, i, j;
ENTRY;
LASSERT(lo->ldo_comp_cnt != 0 && lo->ldo_comp_entries != NULL);
+ mirror_id = lo->ldo_mirror_count > 1 ? 1 : 0;
+
/* create all underlying objects */
for (i = 0; i < lo->ldo_comp_cnt; i++) {
lod_comp = &lo->ldo_comp_entries[i];
+ if (lod_comp->llc_extent.e_start == 0 && i > 0) /* new mirror */
+ ++mirror_id;
+
+ if (lod_comp->llc_id == LCME_ID_INVAL) {
+ lod_comp->llc_id = lod_gen_component_id(lo,
+ mirror_id, i);
+ if (lod_comp->llc_id == LCME_ID_INVAL)
+ GOTO(out, rc = -ERANGE);
+ }
+
if (lod_comp_inited(lod_comp))
continue;
LASSERT(object != NULL);
rc = lod_sub_create(env, object, attr, NULL, dof, th);
if (rc)
- break;
+ GOTO(out, rc);
}
lod_comp_set_init(lod_comp);
}
- if (rc == 0)
- rc = lod_generate_and_set_lovea(env, lo, th);
-
- if (rc == 0)
- lo->ldo_comp_cached = 1;
- else
- lod_object_free_striping(env, lo);
+ rc = lod_fill_mirrors(lo);
+ if (rc)
+ GOTO(out, rc);
+ rc = lod_generate_and_set_lovea(env, lo, th);
+ if (rc)
+ GOTO(out, rc);
+
+ lo->ldo_comp_cached = 1;
+ RETURN(0);
+
+out:
+ lod_object_free_striping(env, lo);
RETURN(rc);
}
static inline int
lod_obj_stripe_destroy_cb(const struct lu_env *env, struct lod_object *lo,
struct dt_object *dt, struct thandle *th,
- int stripe_idx, struct lod_obj_stripe_cb_data *data)
+ int comp_idx, int stripe_idx,
+ struct lod_obj_stripe_cb_data *data)
{
if (data->locd_declare)
return lod_sub_declare_destroy(env, dt, th);
break;
}
} else {
- struct lod_obj_stripe_cb_data data;
+ struct lod_obj_stripe_cb_data data = { { 0 } };
data.locd_declare = true;
- rc = lod_obj_for_each_stripe(env, lo, th,
- lod_obj_stripe_destroy_cb, &data);
+ data.locd_stripe_cb = lod_obj_stripe_destroy_cb;
+ rc = lod_obj_for_each_stripe(env, lo, th, &data);
}
RETURN(rc);
}
}
} else {
- struct lod_obj_stripe_cb_data data;
+ struct lod_obj_stripe_cb_data data = { { 0 } };
data.locd_declare = false;
- rc = lod_obj_for_each_stripe(env, lo, th,
- lod_obj_stripe_destroy_cb, &data);
+ data.locd_stripe_cb = lod_obj_stripe_destroy_cb;
+ rc = lod_obj_for_each_stripe(env, lo, th, &data);
}
RETURN(rc);
return dt_invalidate(env, dt_object_child(dt));
}
-static int lod_declare_layout_change(const struct lu_env *env,
- struct dt_object *dt,
- struct layout_intent *layout,
- const struct lu_buf *buf,
- struct thandle *th)
+static int lod_layout_data_init(struct lod_thread_info *info, __u32 comp_cnt)
{
- struct lod_thread_info *info = lod_env_info(env);
- struct lod_object *lo = lod_dt_obj(dt);
- struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
- struct dt_object *next = dt_object_child(dt);
+ ENTRY;
+
+ /* clear memory region that will be used for layout change */
+ memset(&info->lti_layout_attr, 0, sizeof(struct lu_attr));
+ info->lti_count = 0;
+
+ if (info->lti_comp_size >= comp_cnt)
+ RETURN(0);
+
+ if (info->lti_comp_size > 0) {
+ OBD_FREE(info->lti_comp_idx,
+ info->lti_comp_size * sizeof(__u32));
+ info->lti_comp_size = 0;
+ }
+
+ OBD_ALLOC(info->lti_comp_idx, comp_cnt * sizeof(__u32));
+ if (!info->lti_comp_idx)
+ RETURN(-ENOMEM);
+
+ info->lti_comp_size = comp_cnt;
+ RETURN(0);
+}
+
+static int lod_declare_instantiate_components(const struct lu_env *env,
+ struct lod_object *lo, struct thandle *th)
+{
+ struct lod_thread_info *info = lod_env_info(env);
struct ost_pool *inuse = &info->lti_inuse_osts;
+ int i;
+ int rc = 0;
+ ENTRY;
+
+ LASSERT(info->lti_count < lo->ldo_comp_cnt);
+ if (info->lti_count > 0) {
+ /* Prepare inuse array for composite file */
+ rc = lod_prepare_inuse(env, lo);
+ if (rc)
+ RETURN(rc);
+ }
+
+ for (i = 0; i < info->lti_count; i++) {
+ rc = lod_qos_prep_create(env, lo, NULL, th,
+ info->lti_comp_idx[i], inuse);
+ if (rc)
+ break;
+ }
+
+ if (!rc) {
+ info->lti_buf.lb_len = lod_comp_md_size(lo, false);
+ rc = lod_sub_declare_xattr_set(env, lod_object_child(lo),
+ &info->lti_buf, XATTR_NAME_LOV, 0, th);
+ }
+
+ RETURN(rc);
+}
+
+static int lod_declare_update_plain(const struct lu_env *env,
+ struct lod_object *lo, struct layout_intent *layout,
+ const struct lu_buf *buf, struct thandle *th)
+{
+ struct lod_thread_info *info = lod_env_info(env);
+ struct lod_device *d = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
struct lod_layout_component *lod_comp;
struct lov_comp_md_v1 *comp_v1 = NULL;
bool replay = false;
- bool need_create = false;
int i, rc;
ENTRY;
- if (!S_ISREG(dt->do_lu.lo_header->loh_attr) || !dt_object_exists(dt) ||
- dt_object_remote(next))
- RETURN(-EINVAL);
+ LASSERT(lo->ldo_flr_state == LCM_FL_NONE);
- dt_write_lock(env, next, 0);
/*
* In case the client is passing lovea, which only happens during
* the replay of layout intent write RPC for now, we may need to
if (buf && buf->lb_len) {
struct lov_user_md_v1 *v1 = buf->lb_buf;
- if (v1->lmm_magic != (LOV_MAGIC_DEF | LOV_MAGIC_COMP_V1) &&
- v1->lmm_magic !=
- __swab32(LOV_MAGIC_DEF | LOV_MAGIC_COMP_V1)) {
+ if (v1->lmm_magic != (LOV_MAGIC_DEFINED | LOV_MAGIC_COMP_V1) &&
+ v1->lmm_magic != __swab32(LOV_MAGIC_DEFINED |
+ LOV_MAGIC_COMP_V1)) {
CERROR("%s: the replay buffer of layout extend "
"(magic %#x) does not contain expected "
"composite layout.\n",
if (rc <= 0)
GOTO(out, rc);
/* old on-disk EA is stored in info->lti_buf */
- comp_v1 = (struct lov_comp_md_v1 *)&info->lti_buf.lb_buf;
+ comp_v1 = (struct lov_comp_md_v1 *)info->lti_buf.lb_buf;
replay = true;
} else {
/* non replay path */
rc = lod_load_striping_locked(env, lo);
if (rc)
GOTO(out, rc);
-
- /* Prepare inuse array for composite file */
- rc = lod_prepare_inuse(env, lo);
- if (rc)
- GOTO(out, rc);
}
/* Make sure defined layout covers the requested write range. */
lod_comp = &lo->ldo_comp_entries[lo->ldo_comp_cnt - 1];
if (lo->ldo_comp_cnt > 1 &&
lod_comp->llc_extent.e_end != OBD_OBJECT_EOF &&
- lod_comp->llc_extent.e_end < layout->li_end) {
+ lod_comp->llc_extent.e_end < layout->li_extent.e_end) {
CDEBUG(replay ? D_ERROR : D_LAYOUT,
"%s: the defined layout [0, %#llx) does not covers "
- "the write range [%#llx, %#llx).\n",
+ "the write range "DEXT"\n",
lod2obd(d)->obd_name, lod_comp->llc_extent.e_end,
- layout->li_start, layout->li_end);
+ PEXT(&layout->li_extent));
GOTO(out, rc = -EINVAL);
}
+ CDEBUG(D_LAYOUT, "%s: "DFID": instantiate components "DEXT"\n",
+ lod2obd(d)->obd_name, PFID(lod_object_fid(lo)),
+ PEXT(&layout->li_extent));
+
/*
* Iterate ld->ldo_comp_entries, find the component whose extent under
* the write range and not instantianted.
for (i = 0; i < lo->ldo_comp_cnt; i++) {
lod_comp = &lo->ldo_comp_entries[i];
- if (lod_comp->llc_extent.e_start >= layout->li_end)
+ if (lod_comp->llc_extent.e_start >= layout->li_extent.e_end)
break;
if (!replay) {
if (lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)
GOTO(out, rc = -EINVAL);
- need_create = true;
+ LASSERT(info->lti_comp_idx != NULL);
+ info->lti_comp_idx[info->lti_count++] = i;
+ }
+
+ if (info->lti_count == 0)
+ RETURN(-EALREADY);
+
+ lod_obj_inc_layout_gen(lo);
+ rc = lod_declare_instantiate_components(env, lo, th);
+out:
+ if (rc)
+ lod_object_free_striping(env, lo);
+ RETURN(rc);
+}
+
+#define lod_foreach_mirror_comp(comp, lo, mirror_idx) \
+for (comp = &lo->ldo_comp_entries[lo->ldo_mirrors[mirror_idx].lme_start]; \
+ comp <= &lo->ldo_comp_entries[lo->ldo_mirrors[mirror_idx].lme_end]; \
+ comp++)
+
+static inline int lod_comp_index(struct lod_object *lo,
+ struct lod_layout_component *lod_comp)
+{
+ LASSERT(lod_comp >= lo->ldo_comp_entries &&
+ lod_comp <= &lo->ldo_comp_entries[lo->ldo_comp_cnt - 1]);
+
+ return lod_comp - lo->ldo_comp_entries;
+}
+
+/**
+ * Stale other mirrors by writing extent.
+ */
+static void lod_stale_components(struct lod_object *lo, int primary,
+ struct lu_extent *extent)
+{
+ struct lod_layout_component *pri_comp, *lod_comp;
+ int i;
+
+ /* The writing extent decides which components in the primary
+ * are affected... */
+ CDEBUG(D_LAYOUT, "primary mirror %d, "DEXT"\n", primary, PEXT(extent));
+ lod_foreach_mirror_comp(pri_comp, lo, primary) {
+ if (!lu_extent_is_overlapped(extent, &pri_comp->llc_extent))
+ continue;
+
+ CDEBUG(D_LAYOUT, "primary comp %u "DEXT"\n",
+ lod_comp_index(lo, pri_comp),
+ PEXT(&pri_comp->llc_extent));
+
+ for (i = 0; i < lo->ldo_mirror_count; i++) {
+ if (i == primary)
+ continue;
+
+ /* ... and then stale other components that are
+ * overlapping with primary components */
+ lod_foreach_mirror_comp(lod_comp, lo, i) {
+ if (!lu_extent_is_overlapped(
+ &pri_comp->llc_extent,
+ &lod_comp->llc_extent))
+ continue;
+
+ CDEBUG(D_LAYOUT, "stale: %u / %u\n",
+ i, lod_comp_index(lo, lod_comp));
+
+ lod_comp->llc_flags |= LCME_FL_STALE;
+ lo->ldo_mirrors[i].lme_stale = 1;
+ }
+ }
+ }
+}
+
+/**
+ * check an OST's availability
+ * \param[in] env execution environment
+ * \param[in] lo lod object
+ * \param[in] dt dt object
+ * \param[in] index mirror index
+ *
+ * \retval negative if failed
+ * \retval 1 if \a dt is available
+ * \retval 0 if \a dt is not available
+ */
+static inline int lod_check_ost_avail(const struct lu_env *env,
+ struct lod_object *lo,
+ struct dt_object *dt, int index)
+{
+ struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
+ struct lod_tgt_desc *ost;
+ __u32 idx;
+ int type = LU_SEQ_RANGE_OST;
+ int rc;
+
+ rc = lod_fld_lookup(env, lod, lu_object_fid(&dt->do_lu), &idx, &type);
+ if (rc < 0) {
+ CERROR("%s: can't locate "DFID":rc = %d\n",
+ lod2obd(lod)->obd_name, PFID(lu_object_fid(&dt->do_lu)),
+ rc);
+ return rc;
+ }
+
+ ost = OST_TGT(lod, idx);
+ if (ost->ltd_statfs.os_state &
+ (OS_STATE_READONLY | OS_STATE_ENOSPC | OS_STATE_ENOINO) ||
+ ost->ltd_active == 0) {
+ CDEBUG(D_LAYOUT, DFID ": mirror %d OST%d unavail, rc = %d\n",
+ PFID(lod_object_fid(lo)), index, idx, rc);
+ return 0;
+ }
+
+ return 1;
+}
+
+/**
+ * Pick primary mirror for write
+ * \param[in] env execution environment
+ * \param[in] lo object
+ * \param[in] extent write range
+ */
+static int lod_primary_pick(const struct lu_env *env, struct lod_object *lo,
+ struct lu_extent *extent)
+{
+ struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
+ unsigned int seq = 0;
+ struct lod_layout_component *lod_comp;
+ int i, j, rc;
+ int picked = -1, second_pick = -1, third_pick = -1;
+ ENTRY;
+
+ if (OBD_FAIL_CHECK(OBD_FAIL_FLR_RANDOM_PICK_MIRROR)) {
+ get_random_bytes(&seq, sizeof(seq));
+ seq %= lo->ldo_mirror_count;
+ }
+
+ /**
+ * Pick a mirror as the primary, and check the availability of OSTs.
+ *
+ * This algo can be revised later after knowing the topology of
+ * cluster.
+ */
+ lod_qos_statfs_update(env, lod);
+ for (i = 0; i < lo->ldo_mirror_count; i++) {
+ bool ost_avail = true;
+ int index = (i + seq) % lo->ldo_mirror_count;
+
+ if (lo->ldo_mirrors[index].lme_stale) {
+ CDEBUG(D_LAYOUT, DFID": mirror %d stale\n",
+ PFID(lod_object_fid(lo)), index);
+ continue;
+ }
+
+ /* 2nd pick is for the primary mirror containing unavail OST */
+ if (lo->ldo_mirrors[index].lme_primary && second_pick < 0)
+ second_pick = index;
+
+ /* 3rd pick is for non-primary mirror containing unavail OST */
+ if (second_pick < 0 && third_pick < 0)
+ third_pick = index;
+
+ /**
+ * we found a non-primary 1st pick, we'd like to find a
+ * potential pirmary mirror.
+ */
+ if (picked >= 0 && !lo->ldo_mirrors[index].lme_primary)
+ continue;
+
+ /* check the availability of OSTs */
+ lod_foreach_mirror_comp(lod_comp, lo, index) {
+ if (!lod_comp_inited(lod_comp) || !lod_comp->llc_stripe)
+ continue;
+
+ for (j = 0; j < lod_comp->llc_stripe_count; j++) {
+ struct dt_object *dt = lod_comp->llc_stripe[j];
+
+ rc = lod_check_ost_avail(env, lo, dt, index);
+ if (rc < 0)
+ RETURN(rc);
+
+ ost_avail = !!rc;
+ if (!ost_avail)
+ break;
+ } /* for all dt object in one component */
+ if (!ost_avail)
+ break;
+ } /* for all components in a mirror */
+
+ /**
+ * the OSTs where allocated objects locates in the components
+ * of the mirror are available.
+ */
+ if (!ost_avail)
+ continue;
+
+ /* this mirror has all OSTs available */
+ picked = index;
+
+ /**
+ * primary with all OSTs are available, this is the perfect
+ * 1st pick.
+ */
+ if (lo->ldo_mirrors[index].lme_primary)
+ break;
+ } /* for all mirrors */
+
+ /* failed to pick a sound mirror, lower our expectation */
+ if (picked < 0)
+ picked = second_pick;
+ if (picked < 0)
+ picked = third_pick;
+ if (picked < 0)
+ RETURN(-ENODATA);
+
+ RETURN(picked);
+}
+
+/**
+ * figure out the components should be instantiated for resync.
+ */
+static int lod_prepare_resync(const struct lu_env *env, struct lod_object *lo,
+ struct lu_extent *extent)
+{
+ struct lod_thread_info *info = lod_env_info(env);
+ struct lod_layout_component *lod_comp;
+ unsigned int need_sync = 0;
+ int i;
+
+ CDEBUG(D_LAYOUT,
+ DFID": instantiate all stale components in "DEXT"\n",
+ PFID(lod_object_fid(lo)), PEXT(extent));
+
+ /**
+ * instantiate all components within this extent, even non-stale
+ * components.
+ */
+ for (i = 0; i < lo->ldo_mirror_count; i++) {
+ if (!lo->ldo_mirrors[i].lme_stale)
+ continue;
+
+ lod_foreach_mirror_comp(lod_comp, lo, i) {
+ if (!lu_extent_is_overlapped(extent,
+ &lod_comp->llc_extent))
+ break;
+
+ need_sync++;
+
+ if (lod_comp_inited(lod_comp))
+ continue;
+
+ CDEBUG(D_LAYOUT, "resync instantiate %d / %d\n",
+ i, lod_comp_index(lo, lod_comp));
+ info->lti_comp_idx[info->lti_count++] =
+ lod_comp_index(lo, lod_comp);
+ }
+ }
+
+ return need_sync ? 0 : -EALREADY;
+}
+
+static int lod_declare_update_rdonly(const struct lu_env *env,
+ struct lod_object *lo, struct md_layout_change *mlc,
+ struct thandle *th)
+{
+ struct lod_thread_info *info = lod_env_info(env);
+ struct lu_attr *layout_attr = &info->lti_layout_attr;
+ struct lod_layout_component *lod_comp;
+ struct lu_extent extent = { 0 };
+ int rc;
+ ENTRY;
+
+ LASSERT(lo->ldo_flr_state == LCM_FL_RDONLY);
+ LASSERT(mlc->mlc_opc == MD_LAYOUT_WRITE ||
+ mlc->mlc_opc == MD_LAYOUT_RESYNC);
+ LASSERT(lo->ldo_mirror_count > 0);
+
+ if (mlc->mlc_opc == MD_LAYOUT_WRITE) {
+ struct layout_intent *layout = mlc->mlc_intent;
+ int picked;
+
+ extent = layout->li_extent;
+ CDEBUG(D_LAYOUT, DFID": trying to write :"DEXT"\n",
+ PFID(lod_object_fid(lo)), PEXT(&extent));
+
+ picked = lod_primary_pick(env, lo, &extent);
+ if (picked < 0)
+ RETURN(picked);
+
+ CDEBUG(D_LAYOUT, DFID": picked mirror id %u as primary\n",
+ PFID(lod_object_fid(lo)),
+ lo->ldo_mirrors[picked].lme_id);
+
+ if (layout->li_opc == LAYOUT_INTENT_TRUNC) {
+ /**
+ * trunc transfers [0, size) in the intent extent, we'd
+ * stale components overlapping [size, eof).
+ */
+ extent.e_start = extent.e_end;
+ extent.e_end = OBD_OBJECT_EOF;
+ }
+
+ /* stale overlapping components from other mirrors */
+ lod_stale_components(lo, picked, &extent);
+
+ /* restore truncate intent extent */
+ if (layout->li_opc == LAYOUT_INTENT_TRUNC)
+ extent.e_end = extent.e_start;
- rc = lod_qos_prep_create(env, lo, NULL, th, i, inuse);
+ /* instantiate components for the picked mirror, start from 0 */
+ extent.e_start = 0;
+
+ lod_foreach_mirror_comp(lod_comp, lo, picked) {
+ if (!lu_extent_is_overlapped(&extent,
+ &lod_comp->llc_extent))
+ break;
+
+ if (lod_comp_inited(lod_comp))
+ continue;
+
+ info->lti_comp_idx[info->lti_count++] =
+ lod_comp_index(lo, lod_comp);
+ }
+
+ lo->ldo_flr_state = LCM_FL_WRITE_PENDING;
+ } else { /* MD_LAYOUT_RESYNC */
+ int i;
+
+ /**
+ * could contain multiple non-stale mirrors, so we need to
+ * prep uninited all components assuming any non-stale mirror
+ * could be picked as the primary mirror.
+ */
+ for (i = 0; i < lo->ldo_mirror_count; i++) {
+ if (lo->ldo_mirrors[i].lme_stale)
+ continue;
+
+ lod_foreach_mirror_comp(lod_comp, lo, i) {
+ if (!lod_comp_inited(lod_comp))
+ break;
+
+ if (extent.e_end < lod_comp->llc_extent.e_end)
+ extent.e_end =
+ lod_comp->llc_extent.e_end;
+ }
+ }
+
+ rc = lod_prepare_resync(env, lo, &extent);
if (rc)
+ GOTO(out, rc);
+ /* change the file state to SYNC_PENDING */
+ lo->ldo_flr_state = LCM_FL_SYNC_PENDING;
+ }
+
+ /* Reset the layout version once it's becoming too large.
+ * This way it can make sure that the layout version is
+ * monotonously increased in this writing era. */
+ lod_obj_inc_layout_gen(lo);
+ if (lo->ldo_layout_gen > (LCME_ID_MAX >> 1)) {
+ __u32 layout_version;
+
+ cfs_get_random_bytes(&layout_version, sizeof(layout_version));
+ lo->ldo_layout_gen = layout_version & 0xffff;
+ }
+
+ rc = lod_declare_instantiate_components(env, lo, th);
+ if (rc)
+ GOTO(out, rc);
+
+ layout_attr->la_valid = LA_LAYOUT_VERSION;
+ layout_attr->la_layout_version = 0; /* set current version */
+ if (mlc->mlc_opc == MD_LAYOUT_RESYNC)
+ layout_attr->la_layout_version = LU_LAYOUT_RESYNC;
+ rc = lod_declare_attr_set(env, &lo->ldo_obj, layout_attr, th);
+ if (rc)
+ GOTO(out, rc);
+
+out:
+ if (rc)
+ lod_object_free_striping(env, lo);
+ RETURN(rc);
+}
+
+static int lod_declare_update_write_pending(const struct lu_env *env,
+ struct lod_object *lo, struct md_layout_change *mlc,
+ struct thandle *th)
+{
+ struct lod_thread_info *info = lod_env_info(env);
+ struct lu_attr *layout_attr = &info->lti_layout_attr;
+ struct lod_layout_component *lod_comp;
+ struct lu_extent extent = { 0 };
+ int primary = -1;
+ int i;
+ int rc;
+ ENTRY;
+
+ LASSERT(lo->ldo_flr_state == LCM_FL_WRITE_PENDING);
+ LASSERT(mlc->mlc_opc == MD_LAYOUT_WRITE ||
+ mlc->mlc_opc == MD_LAYOUT_RESYNC);
+
+ /* look for the primary mirror */
+ for (i = 0; i < lo->ldo_mirror_count; i++) {
+ if (lo->ldo_mirrors[i].lme_stale)
+ continue;
+
+ LASSERTF(primary < 0, DFID " has multiple primary: %u / %u",
+ PFID(lod_object_fid(lo)),
+ lo->ldo_mirrors[i].lme_id,
+ lo->ldo_mirrors[primary].lme_id);
+
+ primary = i;
+ }
+ if (primary < 0) {
+ CERROR(DFID ": doesn't have a primary mirror\n",
+ PFID(lod_object_fid(lo)));
+ GOTO(out, rc = -ENODATA);
+ }
+
+ CDEBUG(D_LAYOUT, DFID": found primary %u\n",
+ PFID(lod_object_fid(lo)), lo->ldo_mirrors[primary].lme_id);
+
+ LASSERT(!lo->ldo_mirrors[primary].lme_stale);
+
+ /* for LAYOUT_WRITE opc, it has to do the following operations:
+ * 1. stale overlapping componets from stale mirrors;
+ * 2. instantiate components of the primary mirror;
+ * 3. transfter layout version to all objects of the primary;
+ *
+ * for LAYOUT_RESYNC opc, it will do:
+ * 1. instantiate components of all stale mirrors;
+ * 2. transfer layout version to all objects to close write era. */
+
+ if (mlc->mlc_opc == MD_LAYOUT_WRITE) {
+ LASSERT(mlc->mlc_intent != NULL);
+
+ extent = mlc->mlc_intent->li_extent;
+
+ CDEBUG(D_LAYOUT, DFID": intent to write: "DEXT"\n",
+ PFID(lod_object_fid(lo)), PEXT(&extent));
+
+ if (mlc->mlc_intent->li_opc == LAYOUT_INTENT_TRUNC) {
+ /**
+ * trunc transfers [0, size) in the intent extent, we'd
+ * stale components overlapping [size, eof).
+ */
+ extent.e_start = extent.e_end;
+ extent.e_end = OBD_OBJECT_EOF;
+ }
+ /* 1. stale overlapping components */
+ lod_stale_components(lo, primary, &extent);
+
+ /* 2. find out the components need instantiating.
+ * instantiate [0, mlc->mlc_intent->e_end) */
+
+ /* restore truncate intent extent */
+ if (mlc->mlc_intent->li_opc == LAYOUT_INTENT_TRUNC)
+ extent.e_end = extent.e_start;
+ extent.e_start = 0;
+
+ lod_foreach_mirror_comp(lod_comp, lo, primary) {
+ if (!lu_extent_is_overlapped(&extent,
+ &lod_comp->llc_extent))
+ break;
+
+ if (lod_comp_inited(lod_comp))
+ continue;
+
+ CDEBUG(D_LAYOUT, "write instantiate %d / %d\n",
+ primary, lod_comp_index(lo, lod_comp));
+ info->lti_comp_idx[info->lti_count++] =
+ lod_comp_index(lo, lod_comp);
+ }
+ } else { /* MD_LAYOUT_RESYNC */
+ lod_foreach_mirror_comp(lod_comp, lo, primary) {
+ if (!lod_comp_inited(lod_comp))
+ break;
+
+ extent.e_end = lod_comp->llc_extent.e_end;
+ }
+
+ rc = lod_prepare_resync(env, lo, &extent);
+ if (rc)
+ GOTO(out, rc);
+ /* change the file state to SYNC_PENDING */
+ lo->ldo_flr_state = LCM_FL_SYNC_PENDING;
+ }
+
+ rc = lod_declare_instantiate_components(env, lo, th);
+ if (rc)
+ GOTO(out, rc);
+
+ /* 3. transfer layout version to OST objects.
+ * transfer new layout version to OST objects so that stale writes
+ * can be denied. It also ends an era of writing by setting
+ * LU_LAYOUT_RESYNC. Normal client can never use this bit to
+ * send write RPC; only resync RPCs could do it. */
+ layout_attr->la_valid = LA_LAYOUT_VERSION;
+ layout_attr->la_layout_version = 0; /* set current version */
+ if (mlc->mlc_opc == MD_LAYOUT_RESYNC)
+ layout_attr->la_layout_version = LU_LAYOUT_RESYNC;
+ rc = lod_declare_attr_set(env, &lo->ldo_obj, layout_attr, th);
+ if (rc)
+ GOTO(out, rc);
+
+ lod_obj_inc_layout_gen(lo);
+out:
+ if (rc)
+ lod_object_free_striping(env, lo);
+ RETURN(rc);
+}
+
+static int lod_declare_update_sync_pending(const struct lu_env *env,
+ struct lod_object *lo, struct md_layout_change *mlc,
+ struct thandle *th)
+{
+ struct lod_thread_info *info = lod_env_info(env);
+ unsigned sync_components = 0;
+ unsigned resync_components = 0;
+ int i;
+ int rc;
+ ENTRY;
+
+ LASSERT(lo->ldo_flr_state == LCM_FL_SYNC_PENDING);
+ LASSERT(mlc->mlc_opc == MD_LAYOUT_RESYNC_DONE ||
+ mlc->mlc_opc == MD_LAYOUT_WRITE);
+
+ CDEBUG(D_LAYOUT, DFID ": received op %d in sync pending\n",
+ PFID(lod_object_fid(lo)), mlc->mlc_opc);
+
+ if (mlc->mlc_opc == MD_LAYOUT_WRITE) {
+ CDEBUG(D_LAYOUT, DFID": cocurrent write to sync pending\n",
+ PFID(lod_object_fid(lo)));
+
+ lo->ldo_flr_state = LCM_FL_WRITE_PENDING;
+ return lod_declare_update_write_pending(env, lo, mlc, th);
+ }
+
+ /* MD_LAYOUT_RESYNC_DONE */
+
+ for (i = 0; i < lo->ldo_comp_cnt; i++) {
+ struct lod_layout_component *lod_comp;
+ int j;
+
+ lod_comp = &lo->ldo_comp_entries[i];
+
+ if (!(lod_comp->llc_flags & LCME_FL_STALE)) {
+ sync_components++;
+ continue;
+ }
+
+ for (j = 0; j < mlc->mlc_resync_count; j++) {
+ if (lod_comp->llc_id != mlc->mlc_resync_ids[j])
+ continue;
+
+ mlc->mlc_resync_ids[j] = LCME_ID_INVAL;
+ lod_comp->llc_flags &= ~LCME_FL_STALE;
+ resync_components++;
break;
+ }
}
- if (need_create)
- lod_obj_inc_layout_gen(lo);
- else
- GOTO(unlock, rc = -EALREADY);
+ /* valid check */
+ for (i = 0; i < mlc->mlc_resync_count; i++) {
+ if (mlc->mlc_resync_ids[i] == LCME_ID_INVAL)
+ continue;
- if (!rc) {
- info->lti_buf.lb_len = lod_comp_md_size(lo, false);
- rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
- XATTR_NAME_LOV, 0, th);
+ CDEBUG(D_LAYOUT, DFID": lcme id %u (%d / %zd) not exist "
+ "or already synced\n", PFID(lod_object_fid(lo)),
+ mlc->mlc_resync_ids[i], i, mlc->mlc_resync_count);
+ GOTO(out, rc = -EINVAL);
}
+
+ if (!sync_components || (mlc->mlc_resync_count && !resync_components)) {
+ CDEBUG(D_LAYOUT, DFID": no mirror in sync\n",
+ PFID(lod_object_fid(lo)));
+
+ /* tend to return an error code here to prevent
+ * the MDT from setting SoM attribute */
+ GOTO(out, rc = -EINVAL);
+ }
+
+ CDEBUG(D_LAYOUT, DFID": resynced %u/%zu components\n",
+ PFID(lod_object_fid(lo)),
+ resync_components, mlc->mlc_resync_count);
+
+ lo->ldo_flr_state = LCM_FL_RDONLY;
+ lod_obj_inc_layout_gen(lo);
+
+ info->lti_buf.lb_len = lod_comp_md_size(lo, false);
+ rc = lod_sub_declare_xattr_set(env, lod_object_child(lo),
+ &info->lti_buf, XATTR_NAME_LOV, 0, th);
+ EXIT;
+
out:
if (rc)
lod_object_free_striping(env, lo);
+ RETURN(rc);
+}
-unlock:
- dt_write_unlock(env, next);
+static int lod_declare_layout_change(const struct lu_env *env,
+ struct dt_object *dt, struct md_layout_change *mlc,
+ struct thandle *th)
+{
+ struct lod_thread_info *info = lod_env_info(env);
+ struct lod_object *lo = lod_dt_obj(dt);
+ int rc;
+ ENTRY;
+
+ if (!S_ISREG(dt->do_lu.lo_header->loh_attr) || !dt_object_exists(dt) ||
+ dt_object_remote(dt_object_child(dt)))
+ RETURN(-EINVAL);
+
+ lod_write_lock(env, dt, 0);
+ rc = lod_load_striping_locked(env, lo);
+ if (rc)
+ GOTO(out, rc);
+
+ LASSERT(lo->ldo_comp_cnt > 0);
+ rc = lod_layout_data_init(info, lo->ldo_comp_cnt);
+ if (rc)
+ GOTO(out, rc);
+
+ switch (lo->ldo_flr_state) {
+ case LCM_FL_NONE:
+ rc = lod_declare_update_plain(env, lo, mlc->mlc_intent,
+ &mlc->mlc_buf, th);
+ break;
+ case LCM_FL_RDONLY:
+ rc = lod_declare_update_rdonly(env, lo, mlc, th);
+ break;
+ case LCM_FL_WRITE_PENDING:
+ rc = lod_declare_update_write_pending(env, lo, mlc, th);
+ break;
+ case LCM_FL_SYNC_PENDING:
+ rc = lod_declare_update_sync_pending(env, lo, mlc, th);
+ break;
+ default:
+ rc = -ENOTSUPP;
+ break;
+ }
+out:
+ dt_write_unlock(env, dt);
RETURN(rc);
}
* Instantiate layout component objects which covers the intent write offset.
*/
static int lod_layout_change(const struct lu_env *env, struct dt_object *dt,
- struct layout_intent *layout,
- const struct lu_buf *buf, struct thandle *th)
+ struct md_layout_change *mlc, struct thandle *th)
{
struct lu_attr *attr = &lod_env_info(env)->lti_attr;
+ struct lu_attr *layout_attr = &lod_env_info(env)->lti_layout_attr;
+ struct lod_object *lo = lod_dt_obj(dt);
+ int rc;
- RETURN(lod_striped_create(env, dt, attr, NULL, th));
+ rc = lod_striped_create(env, dt, attr, NULL, th);
+ if (!rc && layout_attr->la_valid & LA_LAYOUT_VERSION) {
+ layout_attr->la_layout_version |= lo->ldo_layout_gen;
+ rc = lod_attr_set(env, dt, layout_attr, th);
+ }
+
+ return rc;
}
struct dt_object_operations lod_obj_ops = {
git://git.whamcloud.com / fs / lustre-release.git / blobdiff