static int llog_osd_pad(const struct lu_env *env, struct dt_object *o,
loff_t *off, int len, int index, struct thandle *th)
{
struct llog_thread_info *lgi = llog_info(env);
int rc;
LASSERT(th);
LASSERT(off);
LASSERT(len >= LLOG_MIN_REC_SIZE && (len & 0x7) == 0);
lgi->lgi_tail.lrt_len = lgi->lgi_lrh.lrh_len = len;
lgi->lgi_tail.lrt_index = lgi->lgi_lrh.lrh_index = index;
lgi->lgi_lrh.lrh_type = LLOG_PAD_MAGIC;
lgi->lgi_buf.lb_buf = &lgi->lgi_lrh;
lgi->lgi_buf.lb_len = sizeof(lgi->lgi_lrh);
dt_write_lock(env, o, 0);
rc = dt_record_write(env, o, &lgi->lgi_buf, off, th);
if (rc) {
CERROR("%s: error writing padding record: rc = %d\n",
o->do_lu.lo_dev->ld_obd->obd_name, rc);
GOTO(out, rc);
}
lgi->lgi_buf.lb_buf = &lgi->lgi_tail;
lgi->lgi_buf.lb_len = sizeof(lgi->lgi_tail);
*off += len - sizeof(lgi->lgi_lrh) - sizeof(lgi->lgi_tail);
rc = dt_record_write(env, o, &lgi->lgi_buf, off, th);
if (rc)
CERROR("%s: error writing padding record: rc = %d\n",
o->do_lu.lo_dev->ld_obd->obd_name, rc);
out:
dt_write_unlock(env, o);
return rc;
}
/**
* Write the special file which contains the list of llog catalogs IDs
*
* This function writes the CATALOG file which contains the array of llog
* catalogs IDs. It is used mostly to store OSP llogs indexed by OST/MDT
* number.
*
* \param[in] env execution environment
* \param[in] d corresponding storage device
* \param[in] idx position to start from, usually OST/MDT index
* \param[in] count how many catalog IDs to write
* \param[out] idarray the buffer with the data to write.
* \param[in] fid LLOG_CATALOGS_OID for CATALOG object
*
* \retval 0 on successful write of catalog IDs
* \retval negative value on error
*/
int llog_osd_put_cat_list(const struct lu_env *env, struct dt_device *d,
int idx, int count, struct llog_catid *idarray,
const struct lu_fid *fid)
{
struct llog_thread_info *lgi = llog_info(env);
struct dt_object *o = NULL;
struct thandle *th;
int rc, size;
if (count == 0)
RETURN(0);
LASSERT(d);
size = sizeof(*idarray) * count;
lgi->lgi_off = idx * sizeof(*idarray);
lgi->lgi_fid = *fid;
o = dt_locate(env, d, &lgi->lgi_fid);
if (IS_ERR(o))
RETURN(PTR_ERR(o));
if (!dt_object_exists(o))
GOTO(out, rc = -ENOENT);
rc = dt_attr_get(env, o, &lgi->lgi_attr, BYPASS_CAPA);
if (rc)
GOTO(out, rc);
if (!S_ISREG(lgi->lgi_attr.la_mode)) {
CERROR("%s: CATALOGS is not a regular file!: mode = %o\n",
o->do_lu.lo_dev->ld_obd->obd_name,
lgi->lgi_attr.la_mode);
GOTO(out, rc = -ENOENT);
}
th = dt_trans_create(env, d);
if (IS_ERR(th))
GOTO(out, rc = PTR_ERR(th));
lgi->lgi_buf.lb_len = size;
lgi->lgi_buf.lb_buf = idarray;
rc = dt_declare_record_write(env, o, &lgi->lgi_buf, lgi->lgi_off, th);
if (rc)
GOTO(out, rc);
rc = dt_trans_start_local(env, d, th);
if (rc)
GOTO(out_trans, rc);
rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
if (rc)
CDEBUG(D_INODE, "can't write CATALOGS at index %d: rc = %d\n",
idx, rc);
out_trans:
dt_trans_stop(env, d, th);
out:
lu_object_put(env, &o->do_lu);
RETURN(rc);
}
int lfsck_bookmark_store(const struct lu_env *env, struct lfsck_instance *lfsck)
{
struct thandle *handle;
struct dt_object *obj = lfsck->li_bookmark_obj;
loff_t pos = 0;
int len = sizeof(struct lfsck_bookmark);
int rc;
ENTRY;
lfsck_bookmark_cpu_to_le(&lfsck->li_bookmark_disk,
&lfsck->li_bookmark_ram);
handle = dt_trans_create(env, lfsck->li_bottom);
if (IS_ERR(handle)) {
rc = PTR_ERR(handle);
CERROR("%s: fail to create trans for storing lfsck_bookmark: "
"rc = %d\n", lfsck_lfsck2name(lfsck), rc);
RETURN(rc);
}
rc = dt_declare_record_write(env, obj,
lfsck_buf_get(env,
&lfsck->li_bookmark_disk, len),
0, handle);
if (rc != 0) {
CERROR("%s: fail to declare trans for storing lfsck_bookmark: "
"rc = %d\n", lfsck_lfsck2name(lfsck), rc);
GOTO(out, rc);
}
rc = dt_trans_start_local(env, lfsck->li_bottom, handle);
if (rc != 0) {
CERROR("%s: fail to start trans for storing lfsck_bookmark: "
"rc = %d\n", lfsck_lfsck2name(lfsck), rc);
GOTO(out, rc);
}
rc = dt_record_write(env, obj,
lfsck_buf_get(env, &lfsck->li_bookmark_disk, len),
&pos, handle);
if (rc != 0)
CERROR("%s: fail to store lfsck_bookmark: expected = %d, "
"rc = %d\n", lfsck_lfsck2name(lfsck), len, rc);
GOTO(out, rc);
out:
dt_trans_stop(env, lfsck->li_bottom, handle);
return rc;
}
static int out_tx_write_exec(const struct lu_env *env, struct thandle *th,
struct tx_arg *arg)
{
struct dt_object *dt_obj = arg->object;
int rc;
dt_write_lock(env, dt_obj, MOR_TGT_CHILD);
rc = dt_record_write(env, dt_obj, &arg->u.write.buf,
&arg->u.write.pos, th);
dt_write_unlock(env, dt_obj);
if (rc == 0)
rc = arg->u.write.buf.lb_len;
object_update_result_insert(arg->reply, NULL, 0, arg->index, rc);
return rc > 0 ? 0 : rc;
}
static int write_capa_keys(const struct lu_env *env,
struct mdt_device *mdt,
struct lustre_capa_key *keys)
{
struct mdt_thread_info *mti;
struct lustre_capa_key *tmp;
struct thandle *th;
loff_t off = 0;
int i, rc;
mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
th = dt_trans_create(env, mdt->mdt_bottom);
if (IS_ERR(th))
RETURN(PTR_ERR(th));
rc = dt_declare_record_write(env, mdt->mdt_ck_obj,
mdt_buf_const(env, NULL,
sizeof(*tmp) * 3), 0, th);
if (rc)
goto stop;
rc = dt_trans_start_local(env, mdt->mdt_bottom, th);
if (rc)
goto stop;
tmp = &mti->mti_capa_key;
for (i = 0; i < 2; i++) {
lck_cpu_to_le(tmp, &keys[i]);
rc = dt_record_write(env, mdt->mdt_ck_obj,
mdt_buf_const(env, tmp, sizeof(*tmp)),
&off, th);
if (rc)
break;
}
stop:
dt_trans_stop(env, mdt->mdt_bottom, th);
CDEBUG(D_INFO, "write capability keys rc = %d:\n", rc);
return rc;
}
int lfsck_bookmark_store(const struct lu_env *env, struct lfsck_instance *lfsck)
{
struct thandle *handle;
struct dt_object *obj = lfsck->li_bookmark_obj;
struct dt_device *dev = lfsck_obj2dev(obj);
loff_t pos = 0;
int len = sizeof(struct lfsck_bookmark);
int rc;
ENTRY;
lfsck_bookmark_cpu_to_le(&lfsck->li_bookmark_disk,
&lfsck->li_bookmark_ram);
handle = dt_trans_create(env, dev);
if (IS_ERR(handle))
GOTO(log, rc = PTR_ERR(handle));
rc = dt_declare_record_write(env, obj,
lfsck_buf_get(env,
&lfsck->li_bookmark_disk, len),
0, handle);
if (rc != 0)
GOTO(out, rc);
rc = dt_trans_start_local(env, dev, handle);
if (rc != 0)
GOTO(out, rc);
rc = dt_record_write(env, obj,
lfsck_buf_get(env, &lfsck->li_bookmark_disk, len),
&pos, handle);
GOTO(out, rc);
out:
dt_trans_stop(env, dev, handle);
log:
if (rc != 0)
CDEBUG(D_LFSCK, "%s: fail to store lfsck_bookmark: rc = %d\n",
lfsck_lfsck2name(lfsck), rc);
return rc;
}
/**
* Write a padding record to the llog
*
* This function writes a padding record to the end of llog. That may
* be needed if llog contains records of variable size, e.g. config logs
* or changelogs.
* The padding record just aligns llog to the LLOG_CHUNK_SIZE boundary if
* the current record doesn't fit in the remaining space.
*
* It allocates full length to avoid two separate writes for header and tail.
* Such 2-steps scheme needs extra protection and complex error handling.
*
* \param[in] env execution environment
* \param[in] o dt_object to create
* \param[in,out] off pointer to the padding start offset
* \param[in] len padding length
* \param[in] index index of the padding record in a llog
* \param[in] th current transaction handle
*
* \retval 0 on successful padding write
* \retval negative error if write failed
*/
static int llog_osd_pad(const struct lu_env *env, struct dt_object *o,
loff_t *off, int len, int index, struct thandle *th)
{
struct llog_thread_info *lgi = llog_info(env);
struct llog_rec_hdr *rec;
struct llog_rec_tail *tail;
int rc;
ENTRY;
LASSERT(th);
LASSERT(off);
LASSERT(len >= LLOG_MIN_REC_SIZE && (len & 0x7) == 0);
OBD_ALLOC(rec, len);
if (rec == NULL)
RETURN(-ENOMEM);
rec->lrh_len = len;
rec->lrh_index = index;
rec->lrh_type = LLOG_PAD_MAGIC;
tail = rec_tail(rec);
tail->lrt_len = len;
tail->lrt_index = index;
lgi->lgi_buf.lb_buf = rec;
lgi->lgi_buf.lb_len = len;
rc = dt_record_write(env, o, &lgi->lgi_buf, off, th);
if (rc)
CERROR("%s: error writing padding record: rc = %d\n",
o->do_lu.lo_dev->ld_obd->obd_name, rc);
OBD_FREE(rec, len);
RETURN(rc);
}
/**
* Implementation of the llog_operations::lop_write
*
* This function writes the new record in the llog or modify the existed one.
*
* \param[in] env execution environment
* \param[in] loghandle llog handle of the current llog
* \param[in] rec llog record header. This is a real header of
* the full llog record to write. This is
* the beginning of buffer to write, the length
* of buffer is stored in \a rec::lrh_len
* \param[out] reccookie pointer to the cookie to return back if needed.
* It is used for further cancel of this llog
* record.
* \param[in] idx index of the llog record. If \a idx == -1 then
* this is append case, otherwise \a idx is
* the index of record to modify
* \param[in] th current transaction handle
*
* \retval 0 on successful write && \a reccookie == NULL
* 1 on successful write && \a reccookie != NULL
* \retval negative error if write failed
*/
static int llog_osd_write_rec(const struct lu_env *env,
struct llog_handle *loghandle,
struct llog_rec_hdr *rec,
struct llog_cookie *reccookie,
int idx, struct thandle *th)
{
struct llog_thread_info *lgi = llog_info(env);
struct llog_log_hdr *llh;
int reclen = rec->lrh_len;
int index, rc;
struct llog_rec_tail *lrt;
struct dt_object *o;
size_t left;
bool header_is_updated = false;
ENTRY;
LASSERT(env);
llh = loghandle->lgh_hdr;
LASSERT(llh);
o = loghandle->lgh_obj;
LASSERT(o);
LASSERT(th);
CDEBUG(D_OTHER, "new record %x to "DFID"\n",
rec->lrh_type, PFID(lu_object_fid(&o->do_lu)));
/* record length should not bigger than LLOG_CHUNK_SIZE */
if (reclen > LLOG_CHUNK_SIZE)
RETURN(-E2BIG);
rc = dt_attr_get(env, o, &lgi->lgi_attr, NULL);
if (rc)
RETURN(rc);
/**
* The modification case.
* If idx set then the record with that index must be modified.
* There are three cases possible:
* 1) the common case is the llog header update (idx == 0)
* 2) the llog record modification during llog process.
* This is indicated by the \a loghandle::lgh_cur_idx > 0.
* In that case the \a loghandle::lgh_cur_offset
* 3) otherwise this is assumed that llog consist of records of
* fixed size, i.e. catalog. The llog header must has llh_size
* field equal to record size. The record offset is calculated
* just by /a idx value
*
* During modification we don't need extra header update because
* the bitmap and record count are not changed. The record header
* and tail remains the same too.
*/
if (idx != LLOG_NEXT_IDX) {
/* llog can be empty only when first record is being written */
LASSERT(ergo(idx > 0, lgi->lgi_attr.la_size > 0));
if (!ext2_test_bit(idx, llh->llh_bitmap)) {
CERROR("%s: modify unset record %u\n",
o->do_lu.lo_dev->ld_obd->obd_name, idx);
RETURN(-ENOENT);
}
if (idx != rec->lrh_index) {
CERROR("%s: modify index mismatch %d %u\n",
o->do_lu.lo_dev->ld_obd->obd_name, idx,
rec->lrh_index);
RETURN(-EFAULT);
}
if (idx == LLOG_HEADER_IDX) {
/* llog header update */
LASSERT(reclen == sizeof(struct llog_log_hdr));
LASSERT(rec == &llh->llh_hdr);
lgi->lgi_off = 0;
lgi->lgi_buf.lb_len = reclen;
lgi->lgi_buf.lb_buf = rec;
rc = dt_record_write(env, o, &lgi->lgi_buf,
&lgi->lgi_off, th);
RETURN(rc);
} else if (loghandle->lgh_cur_idx > 0) {
/**
* The lgh_cur_offset can be used only if index is
* the same.
*/
if (idx != loghandle->lgh_cur_idx) {
CERROR("%s: modify index mismatch %d %d\n",
o->do_lu.lo_dev->ld_obd->obd_name, idx,
loghandle->lgh_cur_idx);
RETURN(-EFAULT);
}
lgi->lgi_off = loghandle->lgh_cur_offset;
CDEBUG(D_OTHER, "modify record "DOSTID": idx:%d, "
"len:%u offset %llu\n",
POSTID(&loghandle->lgh_id.lgl_oi), idx,
rec->lrh_len, (long long)lgi->lgi_off);
} else if (llh->llh_size > 0) {
if (llh->llh_size != rec->lrh_len) {
CERROR("%s: wrong record size, llh_size is %u"
" but record size is %u\n",
o->do_lu.lo_dev->ld_obd->obd_name,
llh->llh_size, rec->lrh_len);
RETURN(-EINVAL);
}
lgi->lgi_off = sizeof(*llh) + (idx - 1) * reclen;
} else {
/* This can be result of lgh_cur_idx is not set during
* llog processing or llh_size is not set to proper
* record size for fixed records llog. Therefore it is
* impossible to get record offset. */
CERROR("%s: can't get record offset, idx:%d, "
"len:%u.\n", o->do_lu.lo_dev->ld_obd->obd_name,
idx, rec->lrh_len);
RETURN(-EFAULT);
}
/* update only data, header and tail remain the same */
lgi->lgi_off += sizeof(struct llog_rec_hdr);
lgi->lgi_buf.lb_len = REC_DATA_LEN(rec);
lgi->lgi_buf.lb_buf = REC_DATA(rec);
rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
if (rc == 0 && reccookie) {
reccookie->lgc_lgl = loghandle->lgh_id;
reccookie->lgc_index = idx;
rc = 1;
}
RETURN(rc);
}
/**
* The append case.
* The most common case of using llog. The new index is assigned to
* the new record, new bit is set in llog bitmap and llog count is
* incremented.
*
* Make sure that records don't cross a chunk boundary, so we can
* process them page-at-a-time if needed. If it will cross a chunk
* boundary, write in a fake (but referenced) entry to pad the chunk.
*/
LASSERT(lgi->lgi_attr.la_valid & LA_SIZE);
lgi->lgi_off = lgi->lgi_attr.la_size;
left = LLOG_CHUNK_SIZE - (lgi->lgi_off & (LLOG_CHUNK_SIZE - 1));
/* NOTE: padding is a record, but no bit is set */
if (left != 0 && left != reclen &&
left < (reclen + LLOG_MIN_REC_SIZE)) {
index = loghandle->lgh_last_idx + 1;
rc = llog_osd_pad(env, o, &lgi->lgi_off, left, index, th);
if (rc)
RETURN(rc);
loghandle->lgh_last_idx++; /* for pad rec */
}
/* if it's the last idx in log file, then return -ENOSPC */
if (loghandle->lgh_last_idx >= LLOG_BITMAP_SIZE(llh) - 1)
RETURN(-ENOSPC);
/* increment the last_idx along with llh_tail index, they should
* be equal for a llog lifetime */
loghandle->lgh_last_idx++;
index = loghandle->lgh_last_idx;
llh->llh_tail.lrt_index = index;
/**
* NB: the caller should make sure only 1 process access
* the lgh_last_idx, e.g. append should be exclusive.
* Otherwise it might hit the assert.
*/
LASSERT(index < LLOG_BITMAP_SIZE(llh));
rec->lrh_index = index;
lrt = rec_tail(rec);
lrt->lrt_len = rec->lrh_len;
lrt->lrt_index = rec->lrh_index;
/* the lgh_hdr_lock protects llog header data from concurrent
* update/cancel, the llh_count and llh_bitmap are protected */
spin_lock(&loghandle->lgh_hdr_lock);
if (ext2_set_bit(index, llh->llh_bitmap)) {
CERROR("%s: index %u already set in log bitmap\n",
o->do_lu.lo_dev->ld_obd->obd_name, index);
spin_unlock(&loghandle->lgh_hdr_lock);
LBUG(); /* should never happen */
}
llh->llh_count++;
spin_unlock(&loghandle->lgh_hdr_lock);
lgi->lgi_off = 0;
lgi->lgi_buf.lb_len = llh->llh_hdr.lrh_len;
lgi->lgi_buf.lb_buf = &llh->llh_hdr;
rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
if (rc)
GOTO(out, rc);
header_is_updated = true;
rc = dt_attr_get(env, o, &lgi->lgi_attr, NULL);
if (rc)
GOTO(out, rc);
LASSERT(lgi->lgi_attr.la_valid & LA_SIZE);
lgi->lgi_off = lgi->lgi_attr.la_size;
lgi->lgi_buf.lb_len = reclen;
lgi->lgi_buf.lb_buf = rec;
rc = dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
if (rc < 0)
GOTO(out, rc);
CDEBUG(D_OTHER, "added record "DOSTID": idx: %u, %u\n",
POSTID(&loghandle->lgh_id.lgl_oi), index, rec->lrh_len);
if (reccookie != NULL) {
reccookie->lgc_lgl = loghandle->lgh_id;
reccookie->lgc_index = index;
if ((rec->lrh_type == MDS_UNLINK_REC) ||
(rec->lrh_type == MDS_SETATTR64_REC))
reccookie->lgc_subsys = LLOG_MDS_OST_ORIG_CTXT;
else if (rec->lrh_type == OST_SZ_REC)
reccookie->lgc_subsys = LLOG_SIZE_ORIG_CTXT;
else
reccookie->lgc_subsys = -1;
rc = 1;
}
RETURN(rc);
out:
/* cleanup llog for error case */
spin_lock(&loghandle->lgh_hdr_lock);
ext2_clear_bit(index, llh->llh_bitmap);
llh->llh_count--;
spin_unlock(&loghandle->lgh_hdr_lock);
/* restore llog last_idx */
loghandle->lgh_last_idx--;
llh->llh_tail.lrt_index = loghandle->lgh_last_idx;
/* restore the header on disk if it was written */
if (header_is_updated) {
lgi->lgi_off = 0;
lgi->lgi_buf.lb_len = llh->llh_hdr.lrh_len;
lgi->lgi_buf.lb_buf = &llh->llh_hdr;
dt_record_write(env, o, &lgi->lgi_buf, &lgi->lgi_off, th);
}
RETURN(rc);
}
static int llog_osd_write_blob(const struct lu_env *env, struct dt_object *o,
struct llog_rec_hdr *rec, void *buf,
loff_t *off, struct thandle *th)
{
struct llog_thread_info *lgi = llog_info(env);
int buflen = rec->lrh_len;
int rc;
ENTRY;
LASSERT(env);
LASSERT(o);
if (buflen == 0)
CWARN("0-length record\n");
CDEBUG(D_OTHER, "write blob with type %x, buf %p/%u at off %llu\n",
rec->lrh_type, buf, buflen, *off);
if (!buf) {
lgi->lgi_buf.lb_len = buflen;
lgi->lgi_buf.lb_buf = rec;
rc = dt_record_write(env, o, &lgi->lgi_buf, off, th);
if (rc)
CERROR("%s: error writing log record: rc = %d\n",
o->do_lu.lo_dev->ld_obd->obd_name, rc);
RETURN(rc);
}
/* the buf case */
/* protect the following 3 writes from concurrent read */
dt_write_lock(env, o, 0);
rec->lrh_len = sizeof(*rec) + buflen + sizeof(lgi->lgi_tail);
lgi->lgi_buf.lb_len = sizeof(*rec);
lgi->lgi_buf.lb_buf = rec;
rc = dt_record_write(env, o, &lgi->lgi_buf, off, th);
if (rc) {
CERROR("%s: error writing log hdr: rc = %d\n",
o->do_lu.lo_dev->ld_obd->obd_name, rc);
GOTO(out, rc);
}
lgi->lgi_buf.lb_len = buflen;
lgi->lgi_buf.lb_buf = buf;
rc = dt_record_write(env, o, &lgi->lgi_buf, off, th);
if (rc) {
CERROR("%s: error writing log buffer: rc = %d\n",
o->do_lu.lo_dev->ld_obd->obd_name, rc);
GOTO(out, rc);
}
lgi->lgi_tail.lrt_len = rec->lrh_len;
lgi->lgi_tail.lrt_index = rec->lrh_index;
lgi->lgi_buf.lb_len = sizeof(lgi->lgi_tail);
lgi->lgi_buf.lb_buf = &lgi->lgi_tail;
rc = dt_record_write(env, o, &lgi->lgi_buf, off, th);
if (rc)
CERROR("%s: error writing log tail: rc = %d\n",
o->do_lu.lo_dev->ld_obd->obd_name, rc);
out:
dt_write_unlock(env, o);
RETURN(rc);
}
int ofd_precreate_objects(const struct lu_env *env, struct ofd_device *ofd,
obd_id id, struct ofd_seq *oseq, int nr, int sync)
{
struct ofd_thread_info *info = ofd_info(env);
struct ofd_object *fo = NULL;
struct dt_object *next;
struct thandle *th;
struct ofd_object **batch;
struct lu_fid *fid = &info->fti_fid;
obd_id tmp;
int rc;
int i;
int objects = 0;
int nr_saved = nr;
ENTRY;
/* Don't create objects beyond the valid range for this SEQ */
if (unlikely(fid_seq_is_mdt0(ostid_seq(&oseq->os_oi)) &&
(id + nr) >= IDIF_MAX_OID)) {
CERROR("%s:"DOSTID" hit the IDIF_MAX_OID (1<<48)!\n",
ofd_name(ofd), id, ostid_seq(&oseq->os_oi));
RETURN(rc = -ENOSPC);
} else if (unlikely(!fid_seq_is_mdt0(ostid_seq(&oseq->os_oi)) &&
(id + nr) >= OBIF_MAX_OID)) {
CERROR("%s:"DOSTID" hit the OBIF_MAX_OID (1<<32)!\n",
ofd_name(ofd), id, ostid_seq(&oseq->os_oi));
RETURN(rc = -ENOSPC);
}
OBD_ALLOC(batch, nr_saved * sizeof(struct ofd_object *));
if (batch == NULL)
RETURN(-ENOMEM);
info->fti_attr.la_valid = LA_TYPE | LA_MODE;
/*
* We mark object SUID+SGID to flag it for accepting UID+GID from
* client on first write. Currently the permission bits on the OST are
* never used, so this is OK.
*/
info->fti_attr.la_mode = S_IFREG | S_ISUID | S_ISGID | 0666;
info->fti_dof.dof_type = dt_mode_to_dft(S_IFREG);
/* Initialize a/c/m time so any client timestamp will always
* be newer and update the inode. ctime = 0 is also handled
* specially in osd_inode_setattr(). See LU-221, LU-1042 */
info->fti_attr.la_valid |= LA_ATIME | LA_MTIME | LA_CTIME;
info->fti_attr.la_atime = 0;
info->fti_attr.la_mtime = 0;
info->fti_attr.la_ctime = 0;
LASSERT(id != 0);
/* prepare objects */
*fid = *lu_object_fid(&oseq->os_lastid_obj->do_lu);
for (i = 0; i < nr; i++) {
rc = fid_set_id(fid, id + i);
if (rc != 0) {
if (i == 0)
GOTO(out, rc);
nr = i;
break;
}
fo = ofd_object_find(env, ofd, fid);
if (IS_ERR(fo)) {
if (i == 0)
GOTO(out, rc = PTR_ERR(fo));
nr = i;
break;
}
ofd_write_lock(env, fo);
batch[i] = fo;
}
info->fti_buf.lb_buf = &tmp;
info->fti_buf.lb_len = sizeof(tmp);
info->fti_off = 0;
th = ofd_trans_create(env, ofd);
if (IS_ERR(th))
GOTO(out, rc = PTR_ERR(th));
th->th_sync |= sync;
rc = dt_declare_record_write(env, oseq->os_lastid_obj, &info->fti_buf,
info->fti_off, th);
if (rc)
GOTO(trans_stop, rc);
for (i = 0; i < nr; i++) {
fo = batch[i];
LASSERT(fo);
if (unlikely(ofd_object_exists(fo))) {
/* object may exist being re-created by write replay */
CDEBUG(D_INODE, "object "LPX64"/"LPX64" exists: "
DFID"\n", ostid_seq(&oseq->os_oi), id,
PFID(lu_object_fid(&fo->ofo_obj.do_lu)));
continue;
}
next = ofd_object_child(fo);
LASSERT(next != NULL);
rc = dt_declare_create(env, next, &info->fti_attr, NULL,
&info->fti_dof, th);
if (rc) {
nr = i;
break;
}
}
rc = dt_trans_start_local(env, ofd->ofd_osd, th);
if (rc)
GOTO(trans_stop, rc);
CDEBUG(D_OTHER, "%s: create new object "DFID" nr %d\n",
ofd_name(ofd), PFID(fid), nr);
LASSERT(nr > 0);
/* When the LFSCK scanning the whole device to verify the LAST_ID file
* consistency, it will load the last_id into RAM firstly, and compare
* the last_id with each OST-object's ID. If the later one is larger,
* then it will regard the LAST_ID file crashed. But during the LFSCK
* scanning, the OFD may continue to create new OST-objects. Those new
* created OST-objects will have larger IDs than the LFSCK known ones.
* So from the LFSCK view, it needs to re-load the last_id from disk
* file, and if the latest last_id is still smaller than the object's
* ID, then the LAST_ID file is real crashed.
*
* To make above mechanism to work, before OFD pre-create OST-objects,
* it needs to update the LAST_ID file firstly, otherwise, the LFSCK
* may cannot get latest last_id although new OST-object created. */
if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_SKIP_LASTID)) {
tmp = cpu_to_le64(id + nr - 1);
dt_write_lock(env, oseq->os_lastid_obj, 0);
rc = dt_record_write(env, oseq->os_lastid_obj,
&info->fti_buf, &info->fti_off, th);
dt_write_unlock(env, oseq->os_lastid_obj);
if (rc != 0)
GOTO(trans_stop, rc);
}
for (i = 0; i < nr; i++) {
fo = batch[i];
LASSERT(fo);
/* Only the new created objects need to be recorded. */
if (ofd->ofd_osd->dd_record_fid_accessed) {
lfsck_pack_rfa(&ofd_info(env)->fti_lr,
lu_object_fid(&fo->ofo_obj.do_lu));
lfsck_in_notify(env, ofd->ofd_osd,
&ofd_info(env)->fti_lr);
}
if (likely(!ofd_object_exists(fo) &&
!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING))) {
next = ofd_object_child(fo);
LASSERT(next != NULL);
rc = dt_create(env, next, &info->fti_attr, NULL,
&info->fti_dof, th);
if (rc)
break;
LASSERT(ofd_object_exists(fo));
}
ofd_seq_last_oid_set(oseq, id + i);
}
objects = i;
/* NOT all the wanted objects have been created,
* set the LAST_ID as the real created. */
if (unlikely(objects < nr)) {
int rc1;
info->fti_off = 0;
tmp = cpu_to_le64(ofd_seq_last_oid(oseq));
dt_write_lock(env, oseq->os_lastid_obj, 0);
rc1 = dt_record_write(env, oseq->os_lastid_obj,
&info->fti_buf, &info->fti_off, th);
dt_write_unlock(env, oseq->os_lastid_obj);
if (rc1 != 0)
CERROR("%s: fail to reset the LAST_ID for seq ("LPX64
") from "LPU64" to "LPU64"\n", ofd_name(ofd),
ostid_seq(&oseq->os_oi), id + nr - 1,
ofd_seq_last_oid(oseq));
}
trans_stop:
ofd_trans_stop(env, ofd, th, rc);
out:
for (i = 0; i < nr_saved; i++) {
fo = batch[i];
if (fo) {
ofd_write_unlock(env, fo);
ofd_object_put(env, fo);
}
}
OBD_FREE(batch, nr_saved * sizeof(struct ofd_object *));
CDEBUG((objects == 0 && rc == 0) ? D_ERROR : D_OTHER,
"created %d/%d objects: %d\n", objects, nr_saved, rc);
LASSERT(ergo(objects == 0, rc < 0));
RETURN(objects > 0 ? objects : rc);
}
int ofd_precreate_objects(const struct lu_env *env, struct ofd_device *ofd,
obd_id id, struct ofd_seq *oseq, int nr, int sync)
{
struct ofd_thread_info *info = ofd_info(env);
struct ofd_object *fo = NULL;
struct dt_object *next;
struct thandle *th;
struct ofd_object **batch;
obd_id tmp;
int rc;
int i;
int objects = 0;
int nr_saved = nr;
ENTRY;
/* Don't create objects beyond the valid range for this SEQ */
if (unlikely(fid_seq_is_mdt0(ostid_seq(&oseq->os_oi)) &&
(id + nr) >= IDIF_MAX_OID)) {
CERROR("%s:"DOSTID" hit the IDIF_MAX_OID (1<<48)!\n",
ofd_name(ofd), id, ostid_seq(&oseq->os_oi));
RETURN(rc = -ENOSPC);
} else if (unlikely(!fid_seq_is_mdt0(ostid_seq(&oseq->os_oi)) &&
(id + nr) >= OBIF_MAX_OID)) {
CERROR("%s:"DOSTID" hit the OBIF_MAX_OID (1<<32)!\n",
ofd_name(ofd), id, ostid_seq(&oseq->os_oi));
RETURN(rc = -ENOSPC);
}
OBD_ALLOC(batch, nr_saved * sizeof(struct ofd_object *));
if (batch == NULL)
RETURN(-ENOMEM);
info->fti_attr.la_valid = LA_TYPE | LA_MODE;
/*
* We mark object SUID+SGID to flag it for accepting UID+GID from
* client on first write. Currently the permission bits on the OST are
* never used, so this is OK.
*/
info->fti_attr.la_mode = S_IFREG | S_ISUID | S_ISGID | 0666;
info->fti_dof.dof_type = dt_mode_to_dft(S_IFREG);
/* Initialize a/c/m time so any client timestamp will always
* be newer and update the inode. ctime = 0 is also handled
* specially in osd_inode_setattr(). See LU-221, LU-1042 */
info->fti_attr.la_valid |= LA_ATIME | LA_MTIME | LA_CTIME;
info->fti_attr.la_atime = 0;
info->fti_attr.la_mtime = 0;
info->fti_attr.la_ctime = 0;
/* prepare objects */
ostid_set_seq(&info->fti_ostid, ostid_seq(&oseq->os_oi));
for (i = 0; i < nr; i++) {
ostid_set_id(&info->fti_ostid, id + i);
rc = ostid_to_fid(&info->fti_fid, &info->fti_ostid, 0);
if (rc) {
if (i == 0)
GOTO(out, rc);
nr = i;
break;
}
fo = ofd_object_find(env, ofd, &info->fti_fid);
if (IS_ERR(fo)) {
if (i == 0)
GOTO(out, rc = PTR_ERR(fo));
nr = i;
break;
}
ofd_write_lock(env, fo);
batch[i] = fo;
}
info->fti_buf.lb_buf = &tmp;
info->fti_buf.lb_len = sizeof(tmp);
info->fti_off = 0;
th = ofd_trans_create(env, ofd);
if (IS_ERR(th))
GOTO(out, rc = PTR_ERR(th));
th->th_sync |= sync;
rc = dt_declare_record_write(env, oseq->os_lastid_obj, sizeof(tmp),
info->fti_off, th);
if (rc)
GOTO(trans_stop, rc);
for (i = 0; i < nr; i++) {
fo = batch[i];
LASSERT(fo);
if (unlikely(ofd_object_exists(fo))) {
/* object may exist being re-created by write replay */
CDEBUG(D_INODE, "object "LPX64"/"LPX64" exists: "
DFID"\n", ostid_seq(&oseq->os_oi), id,
PFID(&info->fti_fid));
continue;
}
next = ofd_object_child(fo);
LASSERT(next != NULL);
rc = dt_declare_create(env, next, &info->fti_attr, NULL,
&info->fti_dof, th);
if (rc) {
nr = i;
break;
}
}
rc = dt_trans_start_local(env, ofd->ofd_osd, th);
if (rc)
GOTO(trans_stop, rc);
CDEBUG(D_OTHER, "%s: create new object "DFID" nr %d\n",
ofd_name(ofd), PFID(&info->fti_fid), nr);
for (i = 0; i < nr; i++) {
fo = batch[i];
LASSERT(fo);
if (likely(!ofd_object_exists(fo))) {
next = ofd_object_child(fo);
LASSERT(next != NULL);
rc = dt_create(env, next, &info->fti_attr, NULL,
&info->fti_dof, th);
if (rc)
break;
LASSERT(ofd_object_exists(fo));
}
ofd_seq_last_oid_set(oseq, id + i);
}
objects = i;
if (objects > 0) {
tmp = cpu_to_le64(ofd_seq_last_oid(oseq));
rc = dt_record_write(env, oseq->os_lastid_obj,
&info->fti_buf, &info->fti_off, th);
}
trans_stop:
ofd_trans_stop(env, ofd, th, rc);
out:
for (i = 0; i < nr_saved; i++) {
fo = batch[i];
if (fo) {
ofd_write_unlock(env, fo);
ofd_object_put(env, fo);
}
}
OBD_FREE(batch, nr_saved * sizeof(struct ofd_object *));
CDEBUG((objects == 0 && rc == 0) ? D_ERROR : D_OTHER,
"created %d/%d objects: %d\n", objects, nr_saved, rc);
LASSERT(ergo(objects == 0, rc < 0));
RETURN(objects > 0 ? objects : rc);
}
static int osp_object_create(const struct lu_env *env, struct dt_object *dt,
struct lu_attr *attr,
struct dt_allocation_hint *hint,
struct dt_object_format *dof, struct thandle *th)
{
struct osp_thread_info *osi = osp_env_info(env);
struct osp_device *d = lu2osp_dev(dt->do_lu.lo_dev);
struct osp_object *o = dt2osp_obj(dt);
int rc = 0;
struct lu_fid *fid = &osi->osi_fid;
ENTRY;
if (o->opo_reserved) {
/* regular case, fid is assigned holding trunsaction open */
osp_object_assign_fid(env, d, o);
}
memcpy(fid, lu_object_fid(&dt->do_lu), sizeof(*fid));
LASSERTF(fid_is_sane(fid), "fid for osp_obj %p is insane"DFID"!\n",
osp_obj, PFID(fid));
if (!o->opo_reserved) {
/* special case, id was assigned outside of transaction
* see comments in osp_declare_attr_set */
spin_lock(&d->opd_pre_lock);
osp_update_last_fid(d, fid);
spin_unlock(&d->opd_pre_lock);
}
CDEBUG(D_INODE, "fid for osp_obj %p is "DFID"!\n", osp_obj, PFID(fid));
/* If the precreate ends, it means it will be ready to rollover to
* the new sequence soon, all the creation should be synchronized,
* otherwise during replay, the replay fid will be inconsistent with
* last_used/create fid */
if (osp_precreate_end_seq(env, d) && osp_is_fid_client(d))
th->th_sync = 1;
/*
* it's OK if the import is inactive by this moment - id was created
* by OST earlier, we just need to maintain it consistently on the disk
* once import is reconnected, OSP will claim this and other objects
* used and OST either keep them, if they exist or recreate
*/
/* we might have lost precreated objects */
if (unlikely(d->opd_gap_count) > 0) {
spin_lock(&d->opd_pre_lock);
if (d->opd_gap_count > 0) {
int count = d->opd_gap_count;
ostid_set_id(&osi->osi_oi,
fid_oid(&d->opd_gap_start_fid));
d->opd_gap_count = 0;
spin_unlock(&d->opd_pre_lock);
CDEBUG(D_HA, "Writting gap "DFID"+%d in llog\n",
PFID(&d->opd_gap_start_fid), count);
/* real gap handling is disabled intil ORI-692 will be
* fixed, now we only report gaps */
} else {
spin_unlock(&d->opd_pre_lock);
}
}
/* new object, the very first ->attr_set()
* initializing attributes needs no logging */
o->opo_new = 1;
/* Only need update last_used oid file, seq file will only be update
* during seq rollover */
osp_objid_buf_prep(&osi->osi_lb, &osi->osi_off,
&d->opd_last_used_fid.f_oid, d->opd_index);
rc = dt_record_write(env, d->opd_last_used_oid_file, &osi->osi_lb,
&osi->osi_off, th);
CDEBUG(D_HA, "%s: Wrote last used FID: "DFID", index %d: %d\n",
d->opd_obd->obd_name, PFID(fid), d->opd_index, rc);
RETURN(rc);
}
static int llog_osd_write_blob(const struct lu_env *env, struct dt_object *o,
struct llog_rec_hdr *rec, void *buf,
loff_t *off, struct thandle *th)
{
struct llog_thread_info *lgi = llog_info(env);
int buflen = rec->lrh_len;
int rc;
LASSERT(env);
LASSERT(o);
if (buflen == 0)
CWARN("0-length record\n");
CDEBUG(D_OTHER, "write blob with type %x, buf %p/%u at off %llu\n",
rec->lrh_type, buf, buflen, *off);
lgi->lgi_attr.la_valid = LA_SIZE;
lgi->lgi_attr.la_size = *off;
if (!buf) {
lgi->lgi_buf.lb_len = buflen;
lgi->lgi_buf.lb_buf = rec;
rc = dt_record_write(env, o, &lgi->lgi_buf, off, th);
if (rc)
CERROR("%s: error writing log record: rc = %d\n",
o->do_lu.lo_dev->ld_obd->obd_name, rc);
GOTO(out, rc);
}
/* the buf case */
/* protect the following 3 writes from concurrent read */
dt_write_lock(env, o, 0);
rec->lrh_len = sizeof(*rec) + buflen + sizeof(lgi->lgi_tail);
lgi->lgi_buf.lb_len = sizeof(*rec);
lgi->lgi_buf.lb_buf = rec;
rc = dt_record_write(env, o, &lgi->lgi_buf, off, th);
if (rc) {
CERROR("%s: error writing log hdr: rc = %d\n",
o->do_lu.lo_dev->ld_obd->obd_name, rc);
GOTO(out_unlock, rc);
}
lgi->lgi_buf.lb_len = buflen;
lgi->lgi_buf.lb_buf = buf;
rc = dt_record_write(env, o, &lgi->lgi_buf, off, th);
if (rc) {
CERROR("%s: error writing log buffer: rc = %d\n",
o->do_lu.lo_dev->ld_obd->obd_name, rc);
GOTO(out_unlock, rc);
}
lgi->lgi_tail.lrt_len = rec->lrh_len;
lgi->lgi_tail.lrt_index = rec->lrh_index;
lgi->lgi_buf.lb_len = sizeof(lgi->lgi_tail);
lgi->lgi_buf.lb_buf = &lgi->lgi_tail;
rc = dt_record_write(env, o, &lgi->lgi_buf, off, th);
if (rc)
CERROR("%s: error writing log tail: rc = %d\n",
o->do_lu.lo_dev->ld_obd->obd_name, rc);
out_unlock:
dt_write_unlock(env, o);
out:
/* cleanup the content written above */
if (rc) {
dt_punch(env, o, lgi->lgi_attr.la_size, OBD_OBJECT_EOF, th,
BYPASS_CAPA);
dt_attr_set(env, o, &lgi->lgi_attr, th, BYPASS_CAPA);
}
return rc;
}
