/**
* 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 ofd_attr_get(const struct lu_env *env, struct ofd_object *fo,
struct lu_attr *la)
{
int rc = 0;
ENTRY;
if (ofd_object_exists(fo)) {
rc = dt_attr_get(env, ofd_object_child(fo), la,
ofd_object_capa(env, fo));
#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 50, 0)
/* Try to correct for a bug in 2.1.0 (LU-221) that caused
* negative timestamps to appear to be in the far future,
* due old timestamp being stored on disk as an unsigned value.
* This fixes up any bad values stored on disk before
* returning them to the client, and ensures any timestamp
* updates are correct. LU-1042 */
if (unlikely(la->la_atime == LU221_BAD_TIME))
la->la_atime = 0;
if (unlikely(la->la_mtime == LU221_BAD_TIME))
la->la_mtime = 0;
if (unlikely(la->la_ctime == LU221_BAD_TIME))
la->la_ctime = 0;
#else
#warning "remove old LU-221/LU-1042 workaround code"
#endif
} else {
rc = -ENOENT;
}
RETURN(rc);
}
/**
* Prepare buffers for read request processing.
*
* This function converts remote buffers from client to local buffers
* and prepares the latter.
*
* \param[in] env execution environment
* \param[in] exp OBD export of client
* \param[in] ofd OFD device
* \param[in] fid FID of object
* \param[in] la object attributes
* \param[in] oa OBDO structure from client
* \param[in] niocount number of remote buffers
* \param[in] rnb remote buffers
* \param[in] nr_local number of local buffers
* \param[in] lnb local buffers
* \param[in] jobid job ID name
*
* \retval 0 on successful prepare
* \retval negative value on error
*/
static int ofd_preprw_read(const struct lu_env *env, struct obd_export *exp,
struct ofd_device *ofd, const struct lu_fid *fid,
struct lu_attr *la, struct obdo *oa, int niocount,
struct niobuf_remote *rnb, int *nr_local,
struct niobuf_local *lnb, char *jobid)
{
struct ofd_object *fo;
int i, j, rc, tot_bytes = 0;
ENTRY;
LASSERT(env != NULL);
fo = ofd_object_find(env, ofd, fid);
if (IS_ERR(fo))
RETURN(PTR_ERR(fo));
LASSERT(fo != NULL);
ofd_read_lock(env, fo);
if (!ofd_object_exists(fo))
GOTO(unlock, rc = -ENOENT);
if (ofd->ofd_lfsck_verify_pfid && oa->o_valid & OBD_MD_FLFID) {
rc = ofd_verify_ff(env, fo, oa);
if (rc != 0)
GOTO(unlock, rc);
}
*nr_local = 0;
for (i = 0, j = 0; i < niocount; i++) {
rc = dt_bufs_get(env, ofd_object_child(fo), rnb + i,
lnb + j, 0);
if (unlikely(rc < 0))
GOTO(buf_put, rc);
LASSERT(rc <= PTLRPC_MAX_BRW_PAGES);
/* correct index for local buffers to continue with */
j += rc;
*nr_local += rc;
LASSERT(j <= PTLRPC_MAX_BRW_PAGES);
tot_bytes += rnb[i].rnb_len;
}
LASSERT(*nr_local > 0 && *nr_local <= PTLRPC_MAX_BRW_PAGES);
rc = dt_attr_get(env, ofd_object_child(fo), la);
if (unlikely(rc))
GOTO(buf_put, rc);
rc = dt_read_prep(env, ofd_object_child(fo), lnb, *nr_local);
if (unlikely(rc))
GOTO(buf_put, rc);
ofd_counter_incr(exp, LPROC_OFD_STATS_READ, jobid, tot_bytes);
RETURN(0);
buf_put:
dt_bufs_put(env, ofd_object_child(fo), lnb, *nr_local);
unlock:
ofd_read_unlock(env, fo);
ofd_object_put(env, fo);
return rc;
}
static int llog_truncate(const struct lu_env *env, struct dt_object *o)
{
struct lu_attr la;
struct thandle *th;
struct dt_device *d;
int rc;
ENTRY;
LASSERT(o);
d = lu2dt_dev(o->do_lu.lo_dev);
LASSERT(d);
rc = dt_attr_get(env, o, &la);
if (rc)
RETURN(rc);
CDEBUG(D_OTHER, "original size "LPU64"\n", la.la_size);
rc = sizeof(struct llog_log_hdr) + sizeof(struct llog_mini_rec);
if (la.la_size < rc) {
CERROR("too small llog: "LPU64"\n", la.la_size);
RETURN(0);
}
/* drop 2 records */
la.la_size = la.la_size - (sizeof(struct llog_mini_rec) * 2);
la.la_valid = LA_SIZE;
th = dt_trans_create(env, d);
if (IS_ERR(th))
RETURN(PTR_ERR(th));
rc = dt_declare_attr_set(env, o, &la, th);
if (rc)
GOTO(stop, rc);
rc = dt_declare_punch(env, o, la.la_size, OBD_OBJECT_EOF, th);
rc = dt_trans_start_local(env, d, th);
if (rc)
GOTO(stop, rc);
rc = dt_punch(env, o, la.la_size, OBD_OBJECT_EOF, th);
if (rc)
GOTO(stop, rc);
rc = dt_attr_set(env, o, &la, th);
if (rc)
GOTO(stop, rc);
stop:
dt_trans_stop(env, d, th);
RETURN(rc);
}
static int out_attr_get(struct tgt_session_info *tsi)
{
const struct lu_env *env = tsi->tsi_env;
struct tgt_thread_info *tti = tgt_th_info(env);
struct object_update *update = tti->tti_u.update.tti_update;
struct obdo *obdo = &tti->tti_u.update.tti_obdo;
struct lu_attr *la = &tti->tti_attr;
struct dt_object *obj = tti->tti_u.update.tti_dt_object;
int idx = tti->tti_u.update.tti_update_reply_index;
int rc;
ENTRY;
if (unlikely(update->ou_result_size < sizeof(*obdo)))
return -EPROTO;
if (!lu_object_exists(&obj->do_lu)) {
/* Usually, this will be called when the master MDT try
* to init a remote object(see osp_object_init), so if
* the object does not exist on slave, we need set BANSHEE flag,
* so the object can be removed from the cache immediately */
set_bit(LU_OBJECT_HEARD_BANSHEE,
&obj->do_lu.lo_header->loh_flags);
RETURN(-ENOENT);
}
dt_read_lock(env, obj, MOR_TGT_CHILD);
rc = dt_attr_get(env, obj, la);
if (rc)
GOTO(out_unlock, rc);
obdo->o_valid = 0;
obdo_from_la(obdo, la, la->la_valid);
lustre_set_wire_obdo(NULL, obdo, obdo);
out_unlock:
dt_read_unlock(env, obj);
CDEBUG(D_INFO, "%s: insert attr get reply %p index %d: rc = %d\n",
tgt_name(tsi->tsi_tgt), tti->tti_u.update.tti_update_reply,
0, rc);
object_update_result_insert(tti->tti_u.update.tti_update_reply, obdo,
sizeof(*obdo), idx, rc);
RETURN(rc);
}
static int llog_osd_declare_write_rec(const struct lu_env *env,
struct llog_handle *loghandle,
struct llog_rec_hdr *rec,
int idx, struct thandle *th)
{
struct llog_thread_info *lgi = llog_info(env);
struct dt_object *o;
int rc;
ENTRY;
LASSERT(env);
LASSERT(th);
LASSERT(loghandle);
o = loghandle->lgh_obj;
LASSERT(o);
/* each time we update header */
rc = dt_declare_record_write(env, o, sizeof(struct llog_log_hdr), 0,
th);
if (rc || idx == 0) /* if error or just header */
RETURN(rc);
if (dt_object_exists(o)) {
rc = dt_attr_get(env, o, &lgi->lgi_attr, BYPASS_CAPA);
lgi->lgi_off = lgi->lgi_attr.la_size;
LASSERT(ergo(rc == 0, lgi->lgi_attr.la_valid & LA_SIZE));
if (rc)
RETURN(rc);
rc = dt_declare_punch(env, o, lgi->lgi_off, OBD_OBJECT_EOF, th);
if (rc)
RETURN(rc);
} else {
lgi->lgi_off = 0;
}
/* XXX: implement declared window or multi-chunks approach */
rc = dt_declare_record_write(env, o, 32 * 1024, lgi->lgi_off, th);
RETURN(rc);
}
/*
* If the object still has SUID+SGID bits set (see ofd_precreate_object()) then
* we will accept the UID+GID if sent by the client for initializing the
* ownership of this object. We only allow this to happen once (so clear these
* bits) and later only allow setattr.
*/
int ofd_attr_handle_ugid(const struct lu_env *env, struct ofd_object *fo,
struct lu_attr *la, int is_setattr)
{
struct ofd_thread_info *info = ofd_info(env);
struct lu_attr *ln = &info->fti_attr2;
__u32 mask = 0;
int rc;
ENTRY;
if (!(la->la_valid & LA_UID) && !(la->la_valid & LA_GID))
RETURN(0);
rc = dt_attr_get(env, ofd_object_child(fo), ln, BYPASS_CAPA);
if (rc != 0)
RETURN(rc);
LASSERT(ln->la_valid & LA_MODE);
if (!is_setattr) {
if (!(ln->la_mode & S_ISUID))
la->la_valid &= ~LA_UID;
if (!(ln->la_mode & S_ISGID))
la->la_valid &= ~LA_GID;
}
if ((la->la_valid & LA_UID) && (ln->la_mode & S_ISUID))
mask |= S_ISUID;
if ((la->la_valid & LA_GID) && (ln->la_mode & S_ISGID))
mask |= S_ISGID;
if (mask != 0) {
if (!(la->la_valid & LA_MODE) || !is_setattr) {
la->la_mode = ln->la_mode;
la->la_valid |= LA_MODE;
}
la->la_mode &= ~mask;
}
RETURN(0);
}
/**
* Implementation of the llog_operations::lop_read_header
*
* This function reads the current llog header from the bottom storage
* device.
*
* \param[in] env execution environment
* \param[in] handle llog handle of the current llog
*
* \retval 0 on successful header read
* \retval negative error if read failed
*/
static int llog_osd_read_header(const struct lu_env *env,
struct llog_handle *handle)
{
struct llog_rec_hdr *llh_hdr;
struct dt_object *o;
struct llog_thread_info *lgi;
int rc;
ENTRY;
LASSERT(sizeof(*handle->lgh_hdr) == LLOG_CHUNK_SIZE);
o = handle->lgh_obj;
LASSERT(o);
lgi = llog_info(env);
rc = dt_attr_get(env, o, &lgi->lgi_attr, NULL);
if (rc)
RETURN(rc);
LASSERT(lgi->lgi_attr.la_valid & LA_SIZE);
if (lgi->lgi_attr.la_size == 0) {
CDEBUG(D_HA, "not reading header from 0-byte log\n");
RETURN(LLOG_EEMPTY);
}
lgi->lgi_off = 0;
lgi->lgi_buf.lb_buf = handle->lgh_hdr;
lgi->lgi_buf.lb_len = LLOG_CHUNK_SIZE;
rc = dt_record_read(env, o, &lgi->lgi_buf, &lgi->lgi_off);
if (rc) {
CERROR("%s: error reading log header from "DFID": rc = %d\n",
o->do_lu.lo_dev->ld_obd->obd_name,
PFID(lu_object_fid(&o->do_lu)), rc);
RETURN(rc);
}
llh_hdr = &handle->lgh_hdr->llh_hdr;
if (LLOG_REC_HDR_NEEDS_SWABBING(llh_hdr))
lustre_swab_llog_hdr(handle->lgh_hdr);
if (llh_hdr->lrh_type != LLOG_HDR_MAGIC) {
CERROR("%s: bad log %s "DFID" header magic: %#x "
"(expected %#x)\n", o->do_lu.lo_dev->ld_obd->obd_name,
handle->lgh_name ? handle->lgh_name : "",
PFID(lu_object_fid(&o->do_lu)),
llh_hdr->lrh_type, LLOG_HDR_MAGIC);
RETURN(-EIO);
} else if (llh_hdr->lrh_len != LLOG_CHUNK_SIZE) {
CERROR("%s: incorrectly sized log %s "DFID" header: "
"%#x (expected %#x)\n"
"you may need to re-run lconf --write_conf.\n",
o->do_lu.lo_dev->ld_obd->obd_name,
handle->lgh_name ? handle->lgh_name : "",
PFID(lu_object_fid(&o->do_lu)),
llh_hdr->lrh_len, LLOG_CHUNK_SIZE);
RETURN(-EIO);
}
handle->lgh_last_idx = handle->lgh_hdr->llh_tail.lrt_index;
RETURN(0);
}
/**
* Read the special file which contains the list of llog catalogs IDs
*
* This function reads the CATALOGS file which contains the array of llog
* catalogs IDs. The main purpose of this file is 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 read
* \param[out] idarray the buffer for the data. If it is NULL then
* function returns just number of catalog IDs
* in the file.
* \param[in] fid LLOG_CATALOGS_OID for CATALOG object
*
* \retval 0 on successful read of catalog IDs
* \retval negative value on error
* \retval positive value which is number of records in
* the file if \a idarray is NULL
*/
int llog_osd_get_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;
ENTRY;
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)) {
th = dt_trans_create(env, d);
if (IS_ERR(th))
GOTO(out, rc = PTR_ERR(th));
lgi->lgi_attr.la_valid = LA_MODE;
lgi->lgi_attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
lgi->lgi_dof.dof_type = dt_mode_to_dft(S_IFREG);
rc = dt_declare_create(env, o, &lgi->lgi_attr, NULL,
&lgi->lgi_dof, th);
if (rc)
GOTO(out_trans, rc);
rc = dt_trans_start_local(env, d, th);
if (rc)
GOTO(out_trans, rc);
dt_write_lock(env, o, 0);
if (!dt_object_exists(o))
rc = dt_create(env, o, &lgi->lgi_attr, NULL,
&lgi->lgi_dof, th);
dt_write_unlock(env, o);
out_trans:
dt_trans_stop(env, d, th);
if (rc)
GOTO(out, rc);
}
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);
}
CDEBUG(D_CONFIG, "cat list: disk size=%d, read=%d\n",
(int)lgi->lgi_attr.la_size, size);
/* return just number of llogs */
if (idarray == NULL) {
rc = lgi->lgi_attr.la_size / sizeof(*idarray);
GOTO(out, rc);
}
/* read for new ost index or for empty file */
memset(idarray, 0, size);
if (lgi->lgi_attr.la_size <= lgi->lgi_off)
GOTO(out, rc = 0);
if (lgi->lgi_attr.la_size < lgi->lgi_off + size)
size = lgi->lgi_attr.la_size - lgi->lgi_off;
lgi->lgi_buf.lb_buf = idarray;
lgi->lgi_buf.lb_len = size;
rc = dt_record_read(env, o, &lgi->lgi_buf, &lgi->lgi_off);
if (rc) {
CERROR("%s: error reading CATALOGS: rc = %d\n",
o->do_lu.lo_dev->ld_obd->obd_name, rc);
GOTO(out, rc);
}
EXIT;
out:
lu_object_put(env, &o->do_lu);
RETURN(rc);
}
static int llog_osd_destroy(const struct lu_env *env,
struct llog_handle *loghandle)
{
struct llog_thread_info *lgi = llog_info(env);
struct llog_ctxt *ctxt;
struct dt_object *o, *llog_dir = NULL;
struct dt_device *d;
struct thandle *th;
char *name = NULL;
int rc;
ENTRY;
ctxt = loghandle->lgh_ctxt;
LASSERT(ctxt);
o = loghandle->lgh_obj;
LASSERT(o);
d = lu2dt_dev(o->do_lu.lo_dev);
LASSERT(d);
LASSERT(d == ctxt->loc_exp->exp_obd->obd_lvfs_ctxt.dt);
th = dt_trans_create(env, d);
if (IS_ERR(th))
RETURN(PTR_ERR(th));
if (loghandle->lgh_name) {
llog_dir = llog_osd_dir_get(env, ctxt);
if (IS_ERR(llog_dir))
GOTO(out_trans, rc = PTR_ERR(llog_dir));
dt_declare_ref_del(env, o, th);
name = loghandle->lgh_name;
rc = dt_declare_delete(env, llog_dir,
(struct dt_key *)name, th);
if (rc)
GOTO(out_trans, rc);
}
dt_declare_ref_del(env, o, th);
rc = dt_declare_destroy(env, o, th);
if (rc)
GOTO(out_trans, rc);
rc = dt_trans_start_local(env, d, th);
if (rc)
GOTO(out_trans, rc);
dt_write_lock(env, o, 0);
if (dt_object_exists(o)) {
if (name) {
dt_ref_del(env, o, th);
dt_read_lock(env, llog_dir, 0);
rc = dt_delete(env, llog_dir,
(struct dt_key *) name,
th, BYPASS_CAPA);
dt_read_unlock(env, llog_dir);
if (rc) {
CERROR("%s: can't remove llog %s: rc = %d\n",
o->do_lu.lo_dev->ld_obd->obd_name,
name, rc);
GOTO(out_unlock, rc);
}
}
/*
* XXX: compatibility bits
* on old filesystems llogs are referenced by the name
* on the new ones they are referenced by OI and by
* the name
*/
rc = dt_attr_get(env, o, &lgi->lgi_attr, NULL);
if (rc)
GOTO(out_unlock, rc);
LASSERT(lgi->lgi_attr.la_nlink < 2);
if (lgi->lgi_attr.la_nlink == 1)
dt_ref_del(env, o, th);
rc = dt_destroy(env, o, th);
if (rc)
GOTO(out_unlock, rc);
}
out_unlock:
dt_write_unlock(env, o);
out_trans:
dt_trans_stop(env, d, th);
if (llog_dir != NULL)
lu_object_put(env, &llog_dir->do_lu);
RETURN(rc);
}
/* sets:
* - cur_offset to the furthest point read in the log file
* - cur_idx to the log index preceeding cur_offset
* returns -EIO/-EINVAL on error
*/
static int llog_osd_next_block(const struct lu_env *env,
struct llog_handle *loghandle, int *cur_idx,
int next_idx, __u64 *cur_offset, void *buf,
int len)
{
struct llog_thread_info *lgi = llog_info(env);
struct dt_object *o;
struct dt_device *dt;
int rc;
ENTRY;
LASSERT(env);
LASSERT(lgi);
if (len == 0 || len & (LLOG_CHUNK_SIZE - 1))
RETURN(-EINVAL);
CDEBUG(D_OTHER, "looking for log index %u (cur idx %u off "LPU64")\n",
next_idx, *cur_idx, *cur_offset);
LASSERT(loghandle);
LASSERT(loghandle->lgh_ctxt);
o = loghandle->lgh_obj;
LASSERT(o);
LASSERT(dt_object_exists(o));
dt = lu2dt_dev(o->do_lu.lo_dev);
LASSERT(dt);
rc = dt_attr_get(env, o, &lgi->lgi_attr, BYPASS_CAPA);
if (rc)
GOTO(out, rc);
while (*cur_offset < lgi->lgi_attr.la_size) {
struct llog_rec_hdr *rec, *last_rec;
struct llog_rec_tail *tail;
llog_skip_over(cur_offset, *cur_idx, next_idx);
/* read up to next LLOG_CHUNK_SIZE block */
lgi->lgi_buf.lb_len = LLOG_CHUNK_SIZE -
(*cur_offset & (LLOG_CHUNK_SIZE - 1));
lgi->lgi_buf.lb_buf = buf;
/* Note: read lock is not needed around la_size get above at
* the time of dt_attr_get(). There are only two cases that
* matter. Either la_size == cur_offset, in which case the
* entire read is skipped, or la_size > cur_offset and the loop
* is entered and this thread is blocked at dt_read_lock()
* until the write is completed. When the write completes, then
* the dt_read() will be done with the full length, and will
* get the full data.
*/
dt_read_lock(env, o, 0);
rc = dt_read(env, o, &lgi->lgi_buf, cur_offset);
dt_read_unlock(env, o);
if (rc < 0) {
CERROR("%s: can't read llog block from log "DFID
" offset "LPU64": rc = %d\n",
o->do_lu.lo_dev->ld_obd->obd_name,
PFID(lu_object_fid(&o->do_lu)), *cur_offset,
rc);
GOTO(out, rc);
}
if (rc < len) {
/* signal the end of the valid buffer to
* llog_process */
memset(buf + rc, 0, len - rc);
}
if (rc == 0) /* end of file, nothing to do */
GOTO(out, rc);
if (rc < sizeof(*tail)) {
CERROR("%s: invalid llog block at log id "LPU64"/%u "
"offset "LPU64"\n",
o->do_lu.lo_dev->ld_obd->obd_name,
loghandle->lgh_id.lgl_oid,
loghandle->lgh_id.lgl_ogen, *cur_offset);
GOTO(out, rc = -EINVAL);
}
rec = buf;
if (LLOG_REC_HDR_NEEDS_SWABBING(rec))
lustre_swab_llog_rec(rec);
tail = (struct llog_rec_tail *)((char *)buf + rc -
sizeof(struct llog_rec_tail));
/* get the last record in block */
last_rec = (struct llog_rec_hdr *)((char *)buf + rc -
le32_to_cpu(tail->lrt_len));
if (LLOG_REC_HDR_NEEDS_SWABBING(last_rec))
lustre_swab_llog_rec(last_rec);
LASSERT(last_rec->lrh_index == tail->lrt_index);
*cur_idx = tail->lrt_index;
/* this shouldn't happen */
if (tail->lrt_index == 0) {
CERROR("%s: invalid llog tail at log id "LPU64"/%u "
"offset "LPU64"\n",
o->do_lu.lo_dev->ld_obd->obd_name,
loghandle->lgh_id.lgl_oid,
loghandle->lgh_id.lgl_ogen, *cur_offset);
GOTO(out, rc = -EINVAL);
}
if (tail->lrt_index < next_idx)
continue;
/* sanity check that the start of the new buffer is no farther
* than the record that we wanted. This shouldn't happen. */
if (rec->lrh_index > next_idx) {
CERROR("%s: missed desired record? %u > %u\n",
o->do_lu.lo_dev->ld_obd->obd_name,
rec->lrh_index, next_idx);
GOTO(out, rc = -ENOENT);
}
GOTO(out, rc = 0);
}
GOTO(out, rc = -EIO);
out:
return rc;
}
/* appends if idx == -1, otherwise overwrites record idx. */
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 cookiecount,
void *buf, 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;
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 (buf)
rc = (reclen > LLOG_CHUNK_SIZE - sizeof(struct llog_rec_hdr) -
sizeof(struct llog_rec_tail)) ? -E2BIG : 0;
else
rc = (reclen > LLOG_CHUNK_SIZE) ? -E2BIG : 0;
if (rc)
RETURN(rc);
rc = dt_attr_get(env, o, &lgi->lgi_attr, NULL);
if (rc)
RETURN(rc);
if (buf)
/* write_blob adds header and tail to lrh_len. */
reclen = sizeof(*rec) + rec->lrh_len +
sizeof(struct llog_rec_tail);
if (idx != -1) {
/* no header: only allowed to insert record 1 */
if (idx != 1 && lgi->lgi_attr.la_size == 0)
LBUG();
if (idx && llh->llh_size && llh->llh_size != rec->lrh_len)
RETURN(-EINVAL);
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);
if (idx != rec->lrh_index)
CERROR("%s: index mismatch %d %u\n",
o->do_lu.lo_dev->ld_obd->obd_name, idx,
rec->lrh_index);
lgi->lgi_off = 0;
rc = llog_osd_write_blob(env, o, &llh->llh_hdr, NULL,
&lgi->lgi_off, th);
/* we are done if we only write the header or on error */
if (rc || idx == 0)
RETURN(rc);
if (buf) {
/* We assume that caller has set lgh_cur_* */
lgi->lgi_off = loghandle->lgh_cur_offset;
CDEBUG(D_OTHER,
"modify record "LPX64": idx:%d/%u/%d, len:%u "
"offset %llu\n",
loghandle->lgh_id.lgl_oid, idx, rec->lrh_index,
loghandle->lgh_cur_idx, rec->lrh_len,
(long long)(lgi->lgi_off - sizeof(*llh)));
if (rec->lrh_index != loghandle->lgh_cur_idx) {
CERROR("%s: modify idx mismatch %u/%d\n",
o->do_lu.lo_dev->ld_obd->obd_name, idx,
loghandle->lgh_cur_idx);
RETURN(-EFAULT);
}
} else {
/* Assumes constant lrh_len */
lgi->lgi_off = sizeof(*llh) + (idx - 1) * reclen;
}
rc = llog_osd_write_blob(env, o, rec, buf, &lgi->lgi_off, th);
if (rc == 0 && reccookie) {
reccookie->lgc_lgl = loghandle->lgh_id;
reccookie->lgc_index = idx;
rc = 1;
}
RETURN(rc);
}
/* 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.
*
* We know that llog_current_log() will return a loghandle that is
* big enough to hold reclen, so all we care about is padding here.
*/
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);
loghandle->lgh_last_idx++;
index = loghandle->lgh_last_idx;
LASSERT(index < LLOG_BITMAP_SIZE(llh));
rec->lrh_index = index;
if (buf == NULL) {
lrt = (struct llog_rec_tail *)((char *)rec + rec->lrh_len -
sizeof(*lrt));
lrt->lrt_len = rec->lrh_len;
lrt->lrt_index = rec->lrh_index;
}
/* The caller should make sure only 1 process access the lgh_last_idx,
* Otherwise it might hit the assert.*/
LASSERT(index < LLOG_BITMAP_SIZE(llh));
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);
llh->llh_tail.lrt_index = index;
lgi->lgi_off = 0;
rc = llog_osd_write_blob(env, o, &llh->llh_hdr, NULL, &lgi->lgi_off,
th);
if (rc)
RETURN(rc);
rc = dt_attr_get(env, o, &lgi->lgi_attr, NULL);
if (rc)
RETURN(rc);
LASSERT(lgi->lgi_attr.la_valid & LA_SIZE);
lgi->lgi_off = lgi->lgi_attr.la_size;
rc = llog_osd_write_blob(env, o, rec, buf, &lgi->lgi_off, th);
if (rc)
RETURN(rc);
CDEBUG(D_RPCTRACE, "added record "LPX64": idx: %u, %u\n",
loghandle->lgh_id.lgl_oid, index, rec->lrh_len);
if (rc == 0 && reccookie) {
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);
}
static int out_attr_get(struct tgt_session_info *tsi)
{
const struct lu_env *env = tsi->tsi_env;
struct tgt_thread_info *tti = tgt_th_info(env);
struct obdo *obdo = &tti->tti_u.update.tti_obdo;
struct lu_attr *la = &tti->tti_attr;
struct dt_object *obj = tti->tti_u.update.tti_dt_object;
int rc;
ENTRY;
if (!lu_object_exists(&obj->do_lu))
RETURN(-ENOENT);
dt_read_lock(env, obj, MOR_TGT_CHILD);
rc = dt_attr_get(env, obj, la, NULL);
if (rc)
GOTO(out_unlock, rc);
/*
* If it is a directory, we will also check whether the
* directory is empty.
* la_flags = 0 : Empty.
* = 1 : Not empty.
*/
la->la_flags = 0;
if (S_ISDIR(la->la_mode)) {
struct dt_it *it;
const struct dt_it_ops *iops;
if (!dt_try_as_dir(env, obj))
GOTO(out_unlock, rc = -ENOTDIR);
iops = &obj->do_index_ops->dio_it;
it = iops->init(env, obj, LUDA_64BITHASH, BYPASS_CAPA);
if (!IS_ERR(it)) {
int result;
result = iops->get(env, it, (const void *)"");
if (result > 0) {
int i;
for (result = 0, i = 0; result == 0 && i < 3;
++i)
result = iops->next(env, it);
if (result == 0)
la->la_flags = 1;
} else if (result == 0)
/*
* Huh? Index contains no zero key?
*/
rc = -EIO;
iops->put(env, it);
iops->fini(env, it);
}
}
obdo->o_valid = 0;
obdo_from_la(obdo, la, la->la_valid);
obdo_cpu_to_le(obdo, obdo);
lustre_set_wire_obdo(NULL, obdo, obdo);
out_unlock:
dt_read_unlock(env, obj);
CDEBUG(D_INFO, "%s: insert attr get reply %p index %d: rc = %d\n",
tgt_name(tsi->tsi_tgt), tti->tti_u.update.tti_update_reply,
0, rc);
update_insert_reply(tti->tti_u.update.tti_update_reply, obdo,
sizeof(*obdo), 0, rc);
RETURN(rc);
}
static int lfsck_namespace_exec_dir(const struct lu_env *env,
struct lfsck_component *com,
struct dt_object *obj,
struct lu_dirent *ent)
{
struct lfsck_thread_info *info = lfsck_env_info(env);
struct lu_attr *la = &info->lti_la;
struct lfsck_instance *lfsck = com->lc_lfsck;
struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
struct lfsck_namespace *ns = com->lc_file_ram;
struct linkea_data ldata = { 0 };
const struct lu_fid *pfid = lfsck_dto2fid(lfsck->li_obj_dir);
const struct lu_fid *cfid = lfsck_dto2fid(obj);
const struct lu_name *cname;
struct thandle *handle = NULL;
bool repaired = false;
bool locked = false;
bool remove;
bool newdata;
int count = 0;
int rc;
ENTRY;
cname = lfsck_name_get_const(env, ent->lde_name, ent->lde_namelen);
down_write(&com->lc_sem);
com->lc_new_checked++;
if (ent->lde_attrs & LUDA_UPGRADE) {
ns->ln_flags |= LF_UPGRADE;
repaired = true;
} else if (ent->lde_attrs & LUDA_REPAIR) {
ns->ln_flags |= LF_INCONSISTENT;
repaired = true;
}
if (ent->lde_name[0] == '.' &&
(ent->lde_namelen == 1 ||
(ent->lde_namelen == 2 && ent->lde_name[1] == '.') ||
fid_is_dot_lustre(&ent->lde_fid)))
GOTO(out, rc = 0);
if (!(bk->lb_param & LPF_DRYRUN) &&
(com->lc_journal || repaired)) {
again:
LASSERT(!locked);
com->lc_journal = 1;
handle = dt_trans_create(env, lfsck->li_next);
if (IS_ERR(handle))
GOTO(out, rc = PTR_ERR(handle));
rc = lfsck_declare_namespace_exec_dir(env, obj, handle);
if (rc != 0)
GOTO(stop, rc);
rc = dt_trans_start(env, lfsck->li_next, handle);
if (rc != 0)
GOTO(stop, rc);
dt_write_lock(env, obj, MOR_TGT_CHILD);
locked = true;
}
rc = lfsck_namespace_check_exist(env, lfsck, obj, ent->lde_name);
if (rc != 0)
GOTO(stop, rc);
rc = lfsck_links_read(env, obj, &ldata);
if (rc == 0) {
count = ldata.ld_leh->leh_reccount;
rc = linkea_links_find(&ldata, cname, pfid);
if ((rc == 0) &&
(count == 1 || !S_ISDIR(lfsck_object_type(obj))))
goto record;
ns->ln_flags |= LF_INCONSISTENT;
/* For dir, if there are more than one linkea entries, or the
* linkea entry does not match the name entry, then remove all
* and add the correct one. */
if (S_ISDIR(lfsck_object_type(obj))) {
remove = true;
newdata = true;
} else {
remove = false;
newdata = false;
}
goto nodata;
} else if (unlikely(rc == -EINVAL)) {
count = 1;
ns->ln_flags |= LF_INCONSISTENT;
/* The magic crashed, we are not sure whether there are more
* corrupt data in the linkea, so remove all linkea entries. */
remove = true;
newdata = true;
goto nodata;
} else if (rc == -ENODATA) {
count = 1;
ns->ln_flags |= LF_UPGRADE;
remove = false;
newdata = true;
nodata:
if (bk->lb_param & LPF_DRYRUN) {
repaired = true;
goto record;
}
if (!com->lc_journal)
goto again;
if (remove) {
LASSERT(newdata);
rc = dt_xattr_del(env, obj, XATTR_NAME_LINK, handle,
BYPASS_CAPA);
if (rc != 0)
GOTO(stop, rc);
}
if (newdata) {
rc = linkea_data_new(&ldata,
&lfsck_env_info(env)->lti_linkea_buf);
if (rc != 0)
GOTO(stop, rc);
}
rc = linkea_add_buf(&ldata, cname, pfid);
if (rc != 0)
GOTO(stop, rc);
rc = lfsck_links_write(env, obj, &ldata, handle);
if (rc != 0)
GOTO(stop, rc);
count = ldata.ld_leh->leh_reccount;
repaired = true;
} else {
GOTO(stop, rc);
}
record:
LASSERT(count > 0);
rc = dt_attr_get(env, obj, la, BYPASS_CAPA);
if (rc != 0)
GOTO(stop, rc);
if ((count == 1) &&
(la->la_nlink == 1 || S_ISDIR(lfsck_object_type(obj))))
/* Usually, it is for single linked object or dir, do nothing.*/
GOTO(stop, rc);
/* Following modification will be in another transaction. */
if (handle != NULL) {
LASSERT(dt_write_locked(env, obj));
dt_write_unlock(env, obj);
locked = false;
dt_trans_stop(env, lfsck->li_next, handle);
handle = NULL;
}
ns->ln_mlinked_checked++;
rc = lfsck_namespace_update(env, com, cfid,
count != la->la_nlink ? LLF_UNMATCH_NLINKS : 0, false);
GOTO(out, rc);
stop:
if (locked)
dt_write_unlock(env, obj);
if (handle != NULL)
dt_trans_stop(env, lfsck->li_next, handle);
out:
if (rc < 0) {
ns->ln_items_failed++;
if (lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent))
lfsck_pos_fill(env, lfsck,
&ns->ln_pos_first_inconsistent, false);
if (!(bk->lb_param & LPF_FAILOUT))
rc = 0;
} else {
if (repaired) {
ns->ln_items_repaired++;
if (bk->lb_param & LPF_DRYRUN &&
lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent))
lfsck_pos_fill(env, lfsck,
&ns->ln_pos_first_inconsistent,
false);
} else {
com->lc_journal = 0;
}
rc = 0;
}
up_write(&com->lc_sem);
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);
}
/**
* Implementation of the llog_operations::lop_next_block
*
* This function finds the the next llog block to return which contains
* record with required index. It is main part of llog processing.
*
* \param[in] env execution environment
* \param[in] loghandle llog handle of the current llog
* \param[in,out] cur_idx index preceeding cur_offset
* \param[in] next_idx target index to find
* \param[in,out] cur_offset furtherst point read in the file
* \param[in] buf pointer to data buffer to fill
* \param[in] len required len to read, it is
* LLOG_CHUNK_SIZE usually.
*
* \retval 0 on successful buffer read
* \retval negative value on error
*/
static int llog_osd_next_block(const struct lu_env *env,
struct llog_handle *loghandle, int *cur_idx,
int next_idx, __u64 *cur_offset, void *buf,
int len)
{
struct llog_thread_info *lgi = llog_info(env);
struct dt_object *o;
struct dt_device *dt;
int rc;
ENTRY;
LASSERT(env);
LASSERT(lgi);
if (len == 0 || len & (LLOG_CHUNK_SIZE - 1))
RETURN(-EINVAL);
CDEBUG(D_OTHER, "looking for log index %u (cur idx %u off "LPU64")\n",
next_idx, *cur_idx, *cur_offset);
LASSERT(loghandle);
LASSERT(loghandle->lgh_ctxt);
o = loghandle->lgh_obj;
LASSERT(o);
LASSERT(dt_object_exists(o));
dt = lu2dt_dev(o->do_lu.lo_dev);
LASSERT(dt);
rc = dt_attr_get(env, o, &lgi->lgi_attr, BYPASS_CAPA);
if (rc)
GOTO(out, rc);
while (*cur_offset < lgi->lgi_attr.la_size) {
struct llog_rec_hdr *rec, *last_rec;
struct llog_rec_tail *tail;
llog_skip_over(cur_offset, *cur_idx, next_idx);
/* read up to next LLOG_CHUNK_SIZE block */
lgi->lgi_buf.lb_len = LLOG_CHUNK_SIZE -
(*cur_offset & (LLOG_CHUNK_SIZE - 1));
lgi->lgi_buf.lb_buf = buf;
rc = dt_read(env, o, &lgi->lgi_buf, cur_offset);
if (rc < 0) {
CERROR("%s: can't read llog block from log "DFID
" offset "LPU64": rc = %d\n",
o->do_lu.lo_dev->ld_obd->obd_name,
PFID(lu_object_fid(&o->do_lu)), *cur_offset,
rc);
GOTO(out, rc);
}
if (rc < len) {
/* signal the end of the valid buffer to
* llog_process */
memset(buf + rc, 0, len - rc);
}
if (rc == 0) /* end of file, nothing to do */
GOTO(out, rc);
if (rc < sizeof(*tail)) {
CERROR("%s: invalid llog block at log id "DOSTID"/%u "
"offset "LPU64"\n",
o->do_lu.lo_dev->ld_obd->obd_name,
POSTID(&loghandle->lgh_id.lgl_oi),
loghandle->lgh_id.lgl_ogen, *cur_offset);
GOTO(out, rc = -EINVAL);
}
rec = buf;
if (LLOG_REC_HDR_NEEDS_SWABBING(rec))
lustre_swab_llog_rec(rec);
tail = (struct llog_rec_tail *)((char *)buf + rc -
sizeof(struct llog_rec_tail));
/* get the last record in block */
last_rec = (struct llog_rec_hdr *)((char *)buf + rc -
tail->lrt_len);
if (LLOG_REC_HDR_NEEDS_SWABBING(last_rec))
lustre_swab_llog_rec(last_rec);
LASSERT(last_rec->lrh_index == tail->lrt_index);
*cur_idx = tail->lrt_index;
/* this shouldn't happen */
if (tail->lrt_index == 0) {
CERROR("%s: invalid llog tail at log id "DOSTID"/%u "
"offset "LPU64"\n",
o->do_lu.lo_dev->ld_obd->obd_name,
POSTID(&loghandle->lgh_id.lgl_oi),
loghandle->lgh_id.lgl_ogen, *cur_offset);
GOTO(out, rc = -EINVAL);
}
if (tail->lrt_index < next_idx)
continue;
/* sanity check that the start of the new buffer is no farther
* than the record that we wanted. This shouldn't happen. */
if (rec->lrh_index > next_idx) {
CERROR("%s: missed desired record? %u > %u\n",
o->do_lu.lo_dev->ld_obd->obd_name,
rec->lrh_index, next_idx);
GOTO(out, rc = -ENOENT);
}
/* Trim unsupported extensions for compat w/ older clients */
if (!(loghandle->lgh_hdr->llh_flags & LLOG_F_EXT_JOBID))
changelog_block_trim_ext(rec, last_rec,
CLF_VERSION | CLF_RENAME);
GOTO(out, rc = 0);
}
GOTO(out, rc = -EIO);
out:
return rc;
}
static int out_attr_get(struct tgt_session_info *tsi)
{
const struct lu_env *env = tsi->tsi_env;
struct tgt_thread_info *tti = tgt_th_info(env);
struct obdo *obdo = &tti->tti_u.update.tti_obdo;
struct lu_attr *la = &tti->tti_attr;
struct dt_object *obj = tti->tti_u.update.tti_dt_object;
int idx = tti->tti_u.update.tti_update_reply_index;
int rc;
ENTRY;
if (!lu_object_exists(&obj->do_lu)) {
/* Usually, this will be called when the master MDT try
* to init a remote object(see osp_object_init), so if
* the object does not exist on slave, we need set BANSHEE flag,
* so the object can be removed from the cache immediately */
set_bit(LU_OBJECT_HEARD_BANSHEE,
&obj->do_lu.lo_header->loh_flags);
RETURN(-ENOENT);
}
dt_read_lock(env, obj, MOR_TGT_CHILD);
rc = dt_attr_get(env, obj, la, NULL);
if (rc)
GOTO(out_unlock, rc);
/*
* If it is a directory, we will also check whether the
* directory is empty.
* la_flags = 0 : Empty.
* = 1 : Not empty.
*/
la->la_flags = 0;
if (S_ISDIR(la->la_mode)) {
struct dt_it *it;
const struct dt_it_ops *iops;
if (!dt_try_as_dir(env, obj))
GOTO(out_unlock, rc = -ENOTDIR);
iops = &obj->do_index_ops->dio_it;
it = iops->init(env, obj, LUDA_64BITHASH, BYPASS_CAPA);
if (!IS_ERR(it)) {
int result;
result = iops->get(env, it, (const void *)"");
if (result > 0) {
int i;
for (result = 0, i = 0; result == 0 && i < 3;
++i)
result = iops->next(env, it);
if (result == 0)
la->la_flags = 1;
} else if (result == 0)
/*
* Huh? Index contains no zero key?
*/
rc = -EIO;
iops->put(env, it);
iops->fini(env, it);
}
}
obdo->o_valid = 0;
obdo_from_la(obdo, la, la->la_valid);
obdo_cpu_to_le(obdo, obdo);
lustre_set_wire_obdo(NULL, obdo, obdo);
out_unlock:
dt_read_unlock(env, obj);
CDEBUG(D_INFO, "%s: insert attr get reply %p index %d: rc = %d\n",
tgt_name(tsi->tsi_tgt), tti->tti_u.update.tti_update_reply,
0, rc);
object_update_result_insert(tti->tti_u.update.tti_update_reply, obdo,
sizeof(*obdo), idx, rc);
RETURN(rc);
}
/**
* Implementation of the llog_operations::lop_prev_block
*
* This function finds the llog block to return which contains
* record with required index but in reverse order - from end of llog
* to the beginning.
* It is main part of reverse llog processing.
*
* \param[in] env execution environment
* \param[in] loghandle llog handle of the current llog
* \param[in] prev_idx target index to find
* \param[in] buf pointer to data buffer to fill
* \param[in] len required len to read, it is LLOG_CHUNK_SIZE usually.
*
* \retval 0 on successful buffer read
* \retval negative value on error
*/
static int llog_osd_prev_block(const struct lu_env *env,
struct llog_handle *loghandle,
int prev_idx, void *buf, int len)
{
struct llog_thread_info *lgi = llog_info(env);
struct dt_object *o;
struct dt_device *dt;
loff_t cur_offset;
int rc;
ENTRY;
if (len == 0 || len & (LLOG_CHUNK_SIZE - 1))
RETURN(-EINVAL);
CDEBUG(D_OTHER, "looking for log index %u\n", prev_idx);
LASSERT(loghandle);
LASSERT(loghandle->lgh_ctxt);
o = loghandle->lgh_obj;
LASSERT(o);
LASSERT(dt_object_exists(o));
dt = lu2dt_dev(o->do_lu.lo_dev);
LASSERT(dt);
cur_offset = LLOG_CHUNK_SIZE;
llog_skip_over(&cur_offset, 0, prev_idx);
rc = dt_attr_get(env, o, &lgi->lgi_attr, BYPASS_CAPA);
if (rc)
GOTO(out, rc);
while (cur_offset < lgi->lgi_attr.la_size) {
struct llog_rec_hdr *rec, *last_rec;
struct llog_rec_tail *tail;
lgi->lgi_buf.lb_len = len;
lgi->lgi_buf.lb_buf = buf;
rc = dt_read(env, o, &lgi->lgi_buf, &cur_offset);
if (rc < 0) {
CERROR("%s: can't read llog block from log "DFID
" offset "LPU64": rc = %d\n",
o->do_lu.lo_dev->ld_obd->obd_name,
PFID(lu_object_fid(&o->do_lu)), cur_offset, rc);
GOTO(out, rc);
}
if (rc == 0) /* end of file, nothing to do */
GOTO(out, rc);
if (rc < sizeof(*tail)) {
CERROR("%s: invalid llog block at log id "DOSTID"/%u "
"offset "LPU64"\n",
o->do_lu.lo_dev->ld_obd->obd_name,
POSTID(&loghandle->lgh_id.lgl_oi),
loghandle->lgh_id.lgl_ogen, cur_offset);
GOTO(out, rc = -EINVAL);
}
rec = buf;
if (LLOG_REC_HDR_NEEDS_SWABBING(rec))
lustre_swab_llog_rec(rec);
tail = (struct llog_rec_tail *)((char *)buf + rc -
sizeof(struct llog_rec_tail));
/* get the last record in block */
last_rec = (struct llog_rec_hdr *)((char *)buf + rc -
le32_to_cpu(tail->lrt_len));
if (LLOG_REC_HDR_NEEDS_SWABBING(last_rec))
lustre_swab_llog_rec(last_rec);
LASSERT(last_rec->lrh_index == tail->lrt_index);
/* this shouldn't happen */
if (tail->lrt_index == 0) {
CERROR("%s: invalid llog tail at log id "DOSTID"/%u "
"offset "LPU64"\n",
o->do_lu.lo_dev->ld_obd->obd_name,
POSTID(&loghandle->lgh_id.lgl_oi),
loghandle->lgh_id.lgl_ogen, cur_offset);
GOTO(out, rc = -EINVAL);
}
if (tail->lrt_index < prev_idx)
continue;
/* sanity check that the start of the new buffer is no farther
* than the record that we wanted. This shouldn't happen. */
if (rec->lrh_index > prev_idx) {
CERROR("%s: missed desired record? %u > %u\n",
o->do_lu.lo_dev->ld_obd->obd_name,
rec->lrh_index, prev_idx);
GOTO(out, rc = -ENOENT);
}
GOTO(out, rc = 0);
}
GOTO(out, rc = -EIO);
out:
return rc;
}
/**
* \retval +ve repaired
* \retval 0 no need to repair
* \retval -ve error cases
*/
static int lfsck_namespace_double_scan_one(const struct lu_env *env,
struct lfsck_component *com,
struct dt_object *child, __u8 flags)
{
struct lfsck_thread_info *info = lfsck_env_info(env);
struct lu_attr *la = &info->lti_la;
struct lu_name *cname = &info->lti_name;
struct lu_fid *pfid = &info->lti_fid;
struct lu_fid *cfid = &info->lti_fid2;
struct lfsck_instance *lfsck = com->lc_lfsck;
struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
struct lfsck_namespace *ns = com->lc_file_ram;
struct linkea_data ldata = { 0 };
struct thandle *handle = NULL;
bool locked = false;
bool update = false;
int rc;
ENTRY;
if (com->lc_journal) {
again:
LASSERT(!locked);
update = false;
com->lc_journal = 1;
handle = dt_trans_create(env, lfsck->li_next);
if (IS_ERR(handle))
RETURN(rc = PTR_ERR(handle));
rc = dt_declare_xattr_set(env, child,
lfsck_buf_get_const(env, NULL, DEFAULT_LINKEA_SIZE),
XATTR_NAME_LINK, 0, handle);
if (rc != 0)
GOTO(stop, rc);
rc = dt_trans_start(env, lfsck->li_next, handle);
if (rc != 0)
GOTO(stop, rc);
dt_write_lock(env, child, MOR_TGT_CHILD);
locked = true;
}
if (unlikely(lfsck_is_dead_obj(child)))
GOTO(stop, rc = 0);
rc = dt_attr_get(env, child, la, BYPASS_CAPA);
if (rc == 0)
rc = lfsck_links_read(env, child, &ldata);
if (rc != 0) {
if ((bk->lb_param & LPF_DRYRUN) &&
(rc == -EINVAL || rc == -ENODATA))
rc = 1;
GOTO(stop, rc);
}
linkea_first_entry(&ldata);
while (ldata.ld_lee != NULL) {
struct dt_object *parent = NULL;
rc = lfsck_linkea_entry_unpack(lfsck, &ldata, cname, pfid);
if (rc > 0)
update = true;
if (!fid_is_sane(pfid))
goto shrink;
parent = lfsck_object_find(env, lfsck, pfid);
if (parent == NULL)
goto shrink;
else if (IS_ERR(parent))
GOTO(stop, rc = PTR_ERR(parent));
if (!dt_object_exists(parent))
goto shrink;
/* XXX: Currently, skip remote object, the consistency for
* remote object will be processed in LFSCK phase III. */
if (dt_object_remote(parent)) {
lfsck_object_put(env, parent);
linkea_next_entry(&ldata);
continue;
}
if (unlikely(!dt_try_as_dir(env, parent)))
goto shrink;
/* To guarantee the 'name' is terminated with '0'. */
memcpy(info->lti_key, cname->ln_name, cname->ln_namelen);
info->lti_key[cname->ln_namelen] = 0;
cname->ln_name = info->lti_key;
rc = dt_lookup(env, parent, (struct dt_rec *)cfid,
(const struct dt_key *)cname->ln_name,
BYPASS_CAPA);
if (rc != 0 && rc != -ENOENT) {
lfsck_object_put(env, parent);
GOTO(stop, rc);
}
if (rc == 0) {
if (lu_fid_eq(cfid, lfsck_dto2fid(child))) {
lfsck_object_put(env, parent);
linkea_next_entry(&ldata);
continue;
}
goto shrink;
}
/* If there is no name entry in the parent dir and the object
* link count is less than the linkea entries count, then the
* linkea entry should be removed. */
if (ldata.ld_leh->leh_reccount > la->la_nlink)
goto shrink;
/* XXX: For the case of there is a linkea entry, but without
* name entry pointing to the object and its hard links
* count is not less than the object name entries count,
* then seems we should add the 'missed' name entry back
* to namespace, but before LFSCK phase III finished, we
* do not know whether the object has some inconsistency
* on other MDTs. So now, do NOT add the name entry back
* to the namespace, but keep the linkEA entry. LU-2914 */
lfsck_object_put(env, parent);
linkea_next_entry(&ldata);
continue;
shrink:
if (parent != NULL)
lfsck_object_put(env, parent);
if (bk->lb_param & LPF_DRYRUN)
RETURN(1);
CDEBUG(D_LFSCK, "Remove linkEA: "DFID"[%.*s], "DFID"\n",
PFID(lfsck_dto2fid(child)), cname->ln_namelen, cname->ln_name,
PFID(pfid));
linkea_del_buf(&ldata, cname);
update = true;
}
if (update) {
if (!com->lc_journal) {
com->lc_journal = 1;
goto again;
}
rc = lfsck_links_write(env, child, &ldata, handle);
}
GOTO(stop, rc);
stop:
if (locked) {
/* XXX: For the case linkea entries count does not match the object hard
* links count, we cannot update the later one simply. Before LFSCK
* phase III finished, we cannot know whether there are some remote
* name entries to be repaired or not. LU-2914 */
if (rc == 0 && !lfsck_is_dead_obj(child) &&
ldata.ld_leh != NULL &&
ldata.ld_leh->leh_reccount != la->la_nlink)
CWARN("%s: the object "DFID" linkEA entry count %u "
"may not match its hardlink count %u\n",
lfsck_lfsck2name(lfsck), PFID(cfid),
ldata.ld_leh->leh_reccount, la->la_nlink);
dt_write_unlock(env, child);
}
if (handle != NULL)
dt_trans_stop(env, lfsck->li_next, handle);
if (rc == 0 && update) {
ns->ln_objs_nlink_repaired++;
rc = 1;
}
return rc;
}
