/*
* TODO:
* - magic in mea struct
- * - error handling is totally missed
*/
-
-int mds_lmv_connect(struct obd_device *obd, char * lmv_name)
+int mds_md_connect(struct obd_device *obd, char *md_name)
{
struct mds_obd *mds = &obd->u.mds;
struct lustre_handle conn = {0};
int rc, valsize, value;
ENTRY;
- if (IS_ERR(mds->mds_lmv_obd))
- RETURN(PTR_ERR(mds->mds_lmv_obd));
+ if (IS_ERR(mds->mds_md_obd))
+ RETURN(PTR_ERR(mds->mds_md_obd));
- if (mds->mds_lmv_connected)
+ if (mds->mds_md_connected)
RETURN(0);
- down(&mds->mds_lmv_sem);
- if (mds->mds_lmv_connected) {
- up(&mds->mds_lmv_sem);
+ down(&mds->mds_md_sem);
+ if (mds->mds_md_connected) {
+ up(&mds->mds_md_sem);
RETURN(0);
}
- mds->mds_lmv_obd = class_name2obd(lmv_name);
- if (!mds->mds_lmv_obd) {
- CERROR("MDS cannot locate LMV %s\n",
- lmv_name);
- mds->mds_lmv_obd = ERR_PTR(-ENOTCONN);
+ mds->mds_md_obd = class_name2obd(md_name);
+ if (!mds->mds_md_obd) {
+ CERROR("MDS cannot locate MD(LMV) %s\n",
+ md_name);
+ mds->mds_md_obd = ERR_PTR(-ENOTCONN);
GOTO(err_last, rc = -ENOTCONN);
}
- rc = obd_connect(&conn, mds->mds_lmv_obd, &obd->obd_uuid, OBD_OPT_MDS_CONNECTION);
+ rc = obd_connect(&conn, mds->mds_md_obd,
+ &obd->obd_uuid, OBD_OPT_MDS_CONNECTION);
if (rc) {
- CERROR("MDS cannot connect to LMV %s (%d)\n",
- lmv_name, rc);
- mds->mds_lmv_obd = ERR_PTR(rc);
+ CERROR("MDS cannot connect to MD(LMV) %s (%d)\n",
+ md_name, rc);
+ mds->mds_md_obd = ERR_PTR(rc);
GOTO(err_last, rc);
}
- mds->mds_lmv_exp = class_conn2export(&conn);
- if (mds->mds_lmv_exp == NULL)
+ mds->mds_md_exp = class_conn2export(&conn);
+ if (mds->mds_md_exp == NULL)
CERROR("can't get export!\n");
- rc = obd_register_observer(mds->mds_lmv_obd, obd);
+ rc = obd_register_observer(mds->mds_md_obd, obd);
if (rc) {
- CERROR("MDS cannot register as observer of LMV %s, "
- "rc = %d\n", lmv_name, rc);
+ CERROR("MDS cannot register as observer of MD(LMV) %s, "
+ "rc = %d\n", md_name, rc);
GOTO(err_discon, rc);
}
/* retrieve size of EA */
- rc = obd_get_info(mds->mds_lmv_exp, strlen("mdsize"), "mdsize",
- &valsize, &value);
+ rc = obd_get_info(mds->mds_md_exp, strlen("mdsize"),
+ "mdsize", &valsize, &value);
if (rc)
GOTO(err_reg, rc);
mds->mds_max_mdsize = value;
/* find our number in LMV cluster */
- rc = obd_get_info(mds->mds_lmv_exp, strlen("mdsnum"), "mdsnum",
- &valsize, &value);
+ rc = obd_get_info(mds->mds_md_exp, strlen("mdsnum"),
+ "mdsnum", &valsize, &value);
if (rc)
GOTO(err_reg, rc);
mds->mds_num = value;
- rc = obd_set_info(mds->mds_lmv_exp, strlen("inter_mds"),
+ rc = obd_set_info(mds->mds_md_exp, strlen("inter_mds"),
"inter_mds", 0, NULL);
if (rc)
GOTO(err_reg, rc);
- mds->mds_lmv_connected = 1;
- up(&mds->mds_lmv_sem);
+ mds->mds_md_connected = 1;
+ up(&mds->mds_md_sem);
RETURN(0);
err_reg:
- obd_register_observer(mds->mds_lmv_obd, NULL);
+ obd_register_observer(mds->mds_md_obd, NULL);
err_discon:
- obd_disconnect(mds->mds_lmv_exp, 0);
- mds->mds_lmv_exp = NULL;
- mds->mds_lmv_obd = ERR_PTR(rc);
+ obd_disconnect(mds->mds_md_exp, 0);
+ mds->mds_md_exp = NULL;
+ mds->mds_md_obd = ERR_PTR(rc);
err_last:
- up(&mds->mds_lmv_sem);
- RETURN(rc);
+ up(&mds->mds_md_sem);
+ return rc;
}
-int mds_lmv_postsetup(struct obd_device *obd)
+int mds_md_postsetup(struct obd_device *obd)
{
struct mds_obd *mds = &obd->u.mds;
int rc = 0;
ENTRY;
- if (mds->mds_lmv_exp)
- rc = obd_init_ea_size(mds->mds_lmv_exp, mds->mds_max_mdsize,
+ if (mds->mds_md_exp) {
+ rc = obd_init_ea_size(mds->mds_md_exp,
+ mds->mds_max_mdsize,
mds->mds_max_cookiesize);
+ }
RETURN(rc);
}
-int mds_lmv_disconnect(struct obd_device *obd, int flags)
+int mds_md_disconnect(struct obd_device *obd, int flags)
{
struct mds_obd *mds = &obd->u.mds;
int rc = 0;
ENTRY;
- down(&mds->mds_lmv_sem);
- if (!IS_ERR(mds->mds_lmv_obd) && mds->mds_lmv_exp != NULL) {
- LASSERT(mds->mds_lmv_connected != 0);
- mds->mds_lmv_connected = 0;
- obd_register_observer(mds->mds_lmv_obd, NULL);
+ if (!mds->mds_md_connected)
+ RETURN(0);
- /* if obd_disconnect fails (probably because the export was
- * disconnected by class_disconnect_exports) then we just need
- * to drop our ref. */
- rc = obd_disconnect(mds->mds_lmv_exp, flags);
- if (rc)
- class_export_put(mds->mds_lmv_exp);
+ down(&mds->mds_md_sem);
+ if (!IS_ERR(mds->mds_md_obd) && mds->mds_md_exp != NULL) {
+ LASSERT(mds->mds_md_connected);
- mds->mds_lmv_exp = NULL;
- mds->mds_lmv_obd = NULL;
- }
- up(&mds->mds_lmv_sem);
+ obd_register_observer(mds->mds_md_obd, NULL);
+
+ if (flags & OBD_OPT_FORCE) {
+ struct obd_device *lmv_obd;
+ struct obd_ioctl_data ioc_data = { 0 };
+
+ lmv_obd = class_exp2obd(mds->mds_md_exp);
+ if (lmv_obd == NULL)
+ GOTO(out, rc = 0);
+
+ /*
+ * making disconnecting lmv stuff do not send anything
+ * to all remote MDSs from LMV. This is needed to
+ * prevent possible hanging with endless recovery, when
+ * MDS sends disconnect to already disconnected
+ * target. Probably this is wrong, but client does the
+ * same in --force mode and I do not see why can't we do
+ * it here. --umka.
+ */
+ lmv_obd->obd_no_recov = 1;
+ obd_iocontrol(IOC_OSC_SET_ACTIVE, mds->mds_md_exp,
+ sizeof(ioc_data), &ioc_data, NULL);
+ }
+
+ /*
+ * if obd_disconnect() fails (probably because the export was
+ * disconnected by class_disconnect_exports()) then we just need
+ * to drop our ref.
+ */
+ mds->mds_md_connected = 0;
+ rc = obd_disconnect(mds->mds_md_exp, flags);
+ if (rc)
+ class_export_put(mds->mds_md_exp);
+ out:
+ mds->mds_md_exp = NULL;
+ mds->mds_md_obd = NULL;
+ }
+ up(&mds->mds_md_sem);
RETURN(rc);
}
-int mds_get_lmv_attr(struct obd_device *obd, struct inode *inode,
- struct mea **mea, int *mea_size)
+int mds_md_get_attr(struct obd_device *obd, struct inode *inode,
+ struct mea **mea, int *mea_size)
{
struct mds_obd *mds = &obd->u.mds;
int rc;
ENTRY;
- if (!mds->mds_lmv_obd)
+ if (!mds->mds_md_obd)
RETURN(0);
if (!S_ISDIR(inode->i_mode))
RETURN(0);
/* first calculate mea size */
- *mea_size = obd_alloc_diskmd(mds->mds_lmv_exp,
+ *mea_size = obd_alloc_diskmd(mds->mds_md_exp,
(struct lov_mds_md **)mea);
if (*mea_size < 0 || *mea == NULL)
return *mea_size < 0 ? *mea_size : -EINVAL;
__u16 mds;
__u32 ino;
__u32 generation;
+ __u32 fid;
char name[0];
};
#define DIR_PAD 4
#define DIR_ROUND (DIR_PAD - 1)
-#define DIR_REC_LEN(name_len) (((name_len) + 12 + DIR_ROUND) & ~DIR_ROUND)
+#define DIR_REC_LEN(name_len) (((name_len) + 16 + DIR_ROUND) & ~DIR_ROUND)
/* this struct holds dir entries for particular MDS to be flushed */
struct dir_cache {
page = alloc_page(GFP_KERNEL);
if (page == NULL)
return -ENOMEM;
- list_add_tail(&page->list, &dirc->list);
+ list_add_tail(&page->lru, &dirc->list);
dirc->cur = page_address(page);
dirc->free = PAGE_SIZE;
return 0;
struct mds_obd *mds = &dc->obd->u.mds;
struct dir_entry *de;
struct dentry *dentry;
- char * end;
+ char *end;
end = buf + PAGE_SIZE;
de = (struct dir_entry *) buf;
while ((char *) de < end && de->namelen) {
/* lookup an inode */
LASSERT(de->namelen <= 255);
- dentry = ll_lookup_one_len(de->name, dc->dentry, de->namelen);
+ dentry = ll_lookup_one_len(de->name, dc->dentry,
+ de->namelen);
if (IS_ERR(dentry)) {
CERROR("can't lookup %*s: %d\n", de->namelen,
de->name, (int) PTR_ERR(dentry));
goto next;
}
if (dentry->d_inode != NULL) {
+ int rc;
+ struct lustre_id sid;
+
+ down(&dentry->d_inode->i_sem);
+ rc = mds_read_inode_sid(dc->obd,
+ dentry->d_inode, &sid);
+ up(&dentry->d_inode->i_sem);
+ if (rc) {
+ CERROR("Can't read inode self id, "
+ "inode %lu, rc %d\n",
+ dentry->d_inode->i_ino, rc);
+ goto next;
+ }
+
+ de->fid = id_fid(&sid);
de->mds = mds->mds_num;
de->ino = dentry->d_inode->i_ino;
de->generation = dentry->d_inode->i_generation;
} else if (dentry->d_flags & DCACHE_CROSS_REF) {
- de->mds = dentry->d_mdsnum;
+ de->fid = dentry->d_fid;
de->ino = dentry->d_inum;
+ de->mds = dentry->d_mdsnum;
de->generation = dentry->d_generation;
} else {
CERROR("can't lookup %*s\n", de->namelen, de->name);
list_for_each_safe(cur, tmp, &ca->list) {
struct page *page;
- page = list_entry(cur, struct page, list);
+ page = list_entry(cur, struct page, lru);
LASSERT(page != NULL);
retrieve_generation_numbers(dc, page_address(page));
ca->brwc.pg = page;
- ca->brwc.off = 0;
+ ca->brwc.disk_offset = ca->brwc.page_offset = 0;
ca->brwc.count = PAGE_SIZE;
ca->brwc.flag = 0;
ca->oa.o_mds = mdsnum;
- rc = obd_brw(OBD_BRW_WRITE, mds->mds_lmv_exp, &ca->oa,
+ rc = obd_brw(OBD_BRW_WRITE, mds->mds_md_exp, &ca->oa,
(struct lov_stripe_md *) dc->mea,
1, &ca->brwc, NULL);
if (rc)
ca = dc->cache + mdsnum;
list_for_each_safe(cur, tmp, &ca->list) {
- page = list_entry(cur, struct page, list);
+ page = list_entry(cur, struct page, lru);
buf = page_address(page);
end = buf + PAGE_SIZE;
/* lookup an inode */
LASSERT(de->namelen <= 255);
- dentry = ll_lookup_one_len(de->name, dc->dentry,
+ dentry = ll_lookup_one_len(de->name, dc->dentry,
de->namelen);
if (IS_ERR(dentry)) {
CERROR("can't lookup %*s: %d\n", de->namelen,
- de->name, (int) PTR_ERR(dentry));
+ de->name, (int) PTR_ERR(dentry));
goto next;
}
rc = fsfilt_del_dir_entry(dc->obd, dentry);
RETURN(0);
}
-static int filldir(void * __buf, const char * name, int namlen, loff_t offset,
- ino_t ino, unsigned int d_type)
+static int filldir(void * __buf, const char * name, int namlen,
+ loff_t offset, ino_t ino, unsigned int d_type)
{
struct dirsplit_control *dc = __buf;
struct mds_obd *mds = &dc->obd->u.mds;
char *n;
ENTRY;
- if (name[0] == '.' && (namlen == 1 ||
- (namlen == 2 && name[1] == '.'))) {
+ if (name[0] == '.' &&
+ (namlen == 1 || (namlen == 2 && name[1] == '.'))) {
/* skip special entries */
RETURN(0);
}
}
int scan_and_distribute(struct obd_device *obd, struct dentry *dentry,
- struct mea *mea)
+ struct mea *mea)
{
struct inode *dir = dentry->d_inode;
struct dirsplit_control dc;
OBD_ALLOC(file_name, nlen);
if (!file_name)
RETURN(-ENOMEM);
- i = sprintf(file_name, "__iopen__/0x%lx", dentry->d_inode->i_ino);
+
+ i = sprintf(file_name, "__iopen__/0x%lx",
+ dentry->d_inode->i_ino);
file = filp_open(file_name, O_RDONLY, 0);
if (IS_ERR(file)) {
GOTO(cleanup, err);
}
+ EXIT;
cleanup:
for (i = 0; i < mea->mea_count; i++) {
struct list_head *cur, *tmp;
continue;
list_for_each_safe(cur, tmp, &dc.cache[i].list) {
struct page *page;
- page = list_entry(cur, struct page, list);
- list_del(&page->list);
+ page = list_entry(cur, struct page, lru);
+ list_del(&page->lru);
__free_page(page);
}
}
OBD_FREE(dc.cache, sizeof(struct dir_cache) * mea->mea_count);
OBD_FREE(file_name, nlen);
- RETURN(err);
+ return err;
}
-#define MAX_DIR_SIZE (64 * 1024)
-
-#define I_NON_SPLITTABLE 256
+#define MAX_DIR_SIZE (64 * 1024)
+#define I_NON_SPLITTABLE (256)
int mds_splitting_expected(struct obd_device *obd, struct dentry *dentry)
{
int rc, size;
/* clustered MD ? */
- if (!mds->mds_lmv_obd)
- RETURN(MDS_NO_SPLITTABLE);
+ if (!mds->mds_md_obd)
+ return MDS_NO_SPLITTABLE;
/* inode exist? */
if (dentry->d_inode == NULL)
return MDS_NO_SPLITTABLE;
/* don't split root directory */
- if (dentry->d_inode->i_ino == mds->mds_rootfid.id)
+ if (dentry->d_inode->i_ino == id_ino(&mds->mds_rootid))
return MDS_NO_SPLITTABLE;
/* large enough to be splitted? */
if (dentry->d_inode->i_size < MAX_DIR_SIZE)
return MDS_NO_SPLIT_EXPECTED;
- mds_get_lmv_attr(obd, dentry->d_inode, &mea, &size);
+ mds_md_get_attr(obd, dentry->d_inode, &mea, &size);
if (mea) {
/* already splitted or slave object: shouldn't be splitted */
rc = MDS_NO_SPLITTABLE;
/* mark to skip subsequent checks */
dentry->d_inode->i_flags |= I_NON_SPLITTABLE;
+ OBD_FREE(mea, size);
} else {
/* may be splitted */
rc = MDS_EXPECT_SPLIT;
}
- if (mea)
- OBD_FREE(mea, size);
- RETURN(rc);
+ return rc;
}
/*
struct mea *tmea = NULL;
struct obdo *oa = NULL;
int rc, mea_size = 0;
+ struct lustre_id id;
void *handle;
ENTRY;
if (update_mode != LCK_EX)
return 0;
+
/* TODO: optimization possible - we already may have mea here */
rc = mds_splitting_expected(obd, dentry);
if (rc == MDS_NO_SPLITTABLE)
if (mea == NULL)
mea = &tmea;
- mea_size = obd_size_diskmd(mds->mds_lmv_exp, NULL);
+ mea_size = obd_size_diskmd(mds->mds_md_exp, NULL);
/* FIXME: Actually we may only want to allocate enough space for
* necessary amount of stripes, but on the other hand with this
* approach of allocating maximal possible amount of MDS slots,
* it would be easier to split the dir over more MDSes */
- rc = obd_alloc_diskmd(mds->mds_lmv_exp, (void *)mea);
+ rc = obd_alloc_diskmd(mds->mds_md_exp, (void *)mea);
if (rc < 0) {
CERROR("obd_alloc_diskmd() failed, error %d.\n", rc);
RETURN(rc);
}
if (*mea == NULL)
- RETURN(-EINVAL);
+ RETURN(-ENOMEM);
(*mea)->mea_count = nstripes;
if (!oa)
RETURN(-ENOMEM);
- oa->o_id = dir->i_ino;
oa->o_generation = dir->i_generation;
obdo_from_inode(oa, dir, OBD_MD_FLTYPE | OBD_MD_FLATIME |
oa->o_valid |= OBD_MD_FLID | OBD_MD_FLFLAGS | OBD_MD_FLGROUP;
oa->o_mode = dir->i_mode;
+ /*
+ * until lmv_obd_create() properly rewritten, it is important to have
+ * here oa->o_id = dir->i_ino, as otherwise master object will have
+ * invalid store cookie (zero inode num), what will lead to -ESTALE in
+ * mds_open() or somewhere else.
+ */
+ oa->o_id = dir->i_ino;
+
+ down(&dir->i_sem);
+ rc = mds_read_inode_sid(obd, dir, &id);
+ up(&dir->i_sem);
+ if (rc) {
+ CERROR("Can't read inode self id, inode %lu, "
+ "rc %d.\n", dir->i_ino, rc);
+ GOTO(err_oa, rc);
+ }
+ oa->o_fid = id_fid(&id);
+ oa->o_mds = mds->mds_num;
+ LASSERT(oa->o_fid != 0);
+
CDEBUG(D_OTHER, "%s: create subdirs with mode %o, uid %u, gid %u\n",
- obd->obd_name, dir->i_mode, dir->i_uid, dir->i_gid);
+ obd->obd_name, dir->i_mode, dir->i_uid, dir->i_gid);
- rc = obd_create(mds->mds_lmv_exp, oa,
+ rc = obd_create(mds->mds_md_exp, oa,
(struct lov_stripe_md **)mea, NULL);
- if (rc)
+ if (rc) {
+ CERROR("Can't create remote inode, rc = %d\n", rc);
GOTO(err_oa, rc);
+ }
+ LASSERT(id_fid(&(*mea)->mea_ids[0]));
CDEBUG(D_OTHER, "%d dirobjects created\n", (int)(*mea)->mea_count);
/* 2) update dir attribute */
CERROR("fsfilt_start() failed: %d\n", (int) PTR_ERR(handle));
GOTO(err_oa, rc = PTR_ERR(handle));
}
-
+
rc = fsfilt_set_md(obd, dir, handle, *mea, mea_size);
if (rc) {
up(&dir->i_sem);
/* 3) read through the dir and distribute it over objects */
rc = scan_and_distribute(obd, dentry, *mea);
if (mea == &tmea)
- obd_free_diskmd(mds->mds_lmv_exp, (struct lov_mds_md **)mea);
+ obd_free_diskmd(mds->mds_md_exp, (struct lov_mds_md **)mea);
if (rc) {
CERROR("scan_and_distribute() failed, error %d.\n", rc);
RETURN(rc);
err_oa:
obdo_free(oa);
- RETURN(rc);
+ return rc;
}
static int filter_start_page_write(struct inode *inode,
struct page *page = alloc_pages(GFP_HIGHUSER, 0);
if (page == NULL) {
CERROR("no memory for a temp page\n");
- RETURN(lnb->rc = -ENOMEM);
+ return lnb->rc = -ENOMEM;
}
POISON_PAGE(page, 0xf1);
page->index = lnb->offset >> PAGE_SHIFT;
lnb->page = page;
-
return 0;
}
-struct dentry *filter_fid2dentry(struct obd_device *obd,
- struct dentry *dir_dentry,
- obd_gr group, obd_id id);
+struct dentry *filter_id2dentry(struct obd_device *obd,
+ struct dentry *dir_dentry,
+ obd_gr group, obd_id id);
int mds_preprw(int cmd, struct obd_export *exp, struct obdo *oa,
int objcount, struct obd_ioobj *obj,
struct niobuf_local *res,
struct obd_trans_info *oti)
{
- struct mds_obd *mds = &exp->exp_obd->u.mds;
struct niobuf_remote *rnb;
struct niobuf_local *lnb = NULL;
int rc = 0, i, tot_bytes = 0;
unsigned long now = jiffies;
struct dentry *dentry;
- struct ll_fid fid;
+ struct lustre_id id;
ENTRY;
+
LASSERT(objcount == 1);
LASSERT(obj->ioo_bufcnt > 0);
memset(res, 0, niocount * sizeof(*res));
- fid.id = obj->ioo_id;
- fid.generation = obj->ioo_gr;
- dentry = mds_fid2dentry(mds, &fid, NULL);
+ id_fid(&id) = 0;
+ id_group(&id) = 0;
+ id_ino(&id) = obj->ioo_id;
+ id_gen(&id) = obj->ioo_gr;
+
+ dentry = mds_id2dentry(exp->exp_obd, &id, NULL);
if (IS_ERR(dentry)) {
CERROR("can't get dentry for "LPU64"/%u: %d\n",
- fid.id, fid.generation, (int) PTR_ERR(dentry));
+ id.li_stc.u.e3s.l3s_ino, id.li_stc.u.e3s.l3s_gen,
+ (int)PTR_ERR(dentry));
GOTO(cleanup, rc = (int) PTR_ERR(dentry));
}
while ((char *) de < end && de->namelen) {
err = fsfilt_add_dir_entry(obd, res->dentry, de->name,
de->namelen, de->ino,
- de->generation, de->mds);
+ de->generation, de->mds,
+ de->fid);
if (err) {
CERROR("can't add dir entry %*s->%u/%u/%u"
- " to %lu/%u: %d\n",
- de->namelen, de->name,
- de->mds, (unsigned) de->ino,
+ " to %lu/%u: %d\n", de->namelen,
+ de->name, de->mds, (unsigned)de->ino,
(unsigned) de->generation,
res->dentry->d_inode->i_ino,
res->dentry->d_inode->i_generation,
if (flags & REC_REINT_CREATE) {
i = mds->mds_num;
- } else if (mds->mds_lmv_exp) {
- lmv = &mds->mds_lmv_exp->exp_obd->u.lmv;
+ } else if (mds->mds_md_exp) {
+ lmv = &mds->mds_md_exp->exp_obd->u.lmv;
i = raw_name2idx(MEA_MAGIC_LAST_CHAR, lmv->desc.ld_tgt_count, name, len);
}
RETURN(i);
struct lustre_handle **rlockh)
{
struct mds_obd *mds = &obd->u.mds;
- struct mdc_op_data op_data;
+ struct mdc_op_data *op_data;
struct lookup_intent it;
struct mea *mea = NULL;
int mea_size, rc;
+ int handle_size;
+ ENTRY;
LASSERT(rlockh != NULL);
LASSERT(dentry != NULL);
LASSERT(dentry->d_inode != NULL);
/* clustered MD ? */
- if (!mds->mds_lmv_obd)
- return 0;
+ if (!mds->mds_md_obd)
+ RETURN(0);
/* a dir can be splitted only */
if (!S_ISDIR(dentry->d_inode->i_mode))
- return 0;
+ RETURN(0);
- rc = mds_get_lmv_attr(obd, dentry->d_inode, &mea, &mea_size);
+ rc = mds_md_get_attr(obd, dentry->d_inode,
+ &mea, &mea_size);
if (rc)
- return rc;
+ RETURN(rc);
if (mea == NULL)
- return 0;
- if (mea->mea_count == 0) {
+ RETURN(0);
+
+ if (mea->mea_count == 0)
/* this is slave object */
GOTO(cleanup, rc = 0);
- }
- CDEBUG(D_OTHER, "%s: lock slaves for %lu/%lu\n", obd->obd_name,
- (unsigned long) dentry->d_inode->i_ino,
- (unsigned long) dentry->d_inode->i_generation);
+ CDEBUG(D_OTHER, "%s: lock slaves for %lu/%lu\n",
+ obd->obd_name, (unsigned long)dentry->d_inode->i_ino,
+ (unsigned long)dentry->d_inode->i_generation);
- OBD_ALLOC(*rlockh, sizeof(struct lustre_handle) * mea->mea_count);
+ handle_size = sizeof(struct lustre_handle) *
+ mea->mea_count;
+
+ OBD_ALLOC(*rlockh, handle_size);
if (*rlockh == NULL)
GOTO(cleanup, rc = -ENOMEM);
- memset(*rlockh, 0, sizeof(struct lustre_handle) * mea->mea_count);
- memset(&op_data, 0, sizeof(op_data));
- op_data.mea1 = mea;
+ memset(*rlockh, 0, handle_size);
+ OBD_ALLOC(op_data, sizeof(*op_data));
+ if (op_data == NULL) {
+ OBD_FREE(*rlockh, handle_size);
+ RETURN(-ENOMEM);
+ }
+ memset(op_data, 0, sizeof(*op_data));
+
+ op_data->mea1 = mea;
it.it_op = IT_UNLINK;
- rc = md_enqueue(mds->mds_lmv_exp, LDLM_IBITS, &it, LCK_EX, &op_data,
- *rlockh, NULL, 0, ldlm_completion_ast, mds_blocking_ast,
- NULL);
+
+ rc = md_enqueue(mds->mds_md_exp, LDLM_IBITS, &it, LCK_EX,
+ op_data, *rlockh, NULL, 0, ldlm_completion_ast,
+ mds_blocking_ast, NULL);
+ OBD_FREE(op_data, sizeof(*op_data));
+ EXIT;
cleanup:
OBD_FREE(mea, mea_size);
- RETURN(rc);
+ return rc;
}
void mds_unlock_slave_objs(struct obd_device *obd, struct dentry *dentry,
- struct lustre_handle *lockh)
+ struct lustre_handle *lockh)
{
struct mds_obd *mds = &obd->u.mds;
struct mea *mea = NULL;
int mea_size, rc, i;
+ ENTRY;
- if (lockh == NULL)
+ if (lockh == NULL) {
+ EXIT;
return;
+ }
- LASSERT(mds->mds_lmv_obd != NULL);
+ LASSERT(mds->mds_md_obd != NULL);
LASSERT(S_ISDIR(dentry->d_inode->i_mode));
- rc = mds_get_lmv_attr(obd, dentry->d_inode, &mea, &mea_size);
+ rc = mds_md_get_attr(obd, dentry->d_inode, &mea, &mea_size);
if (rc) {
CERROR("locks are leaked\n");
+ EXIT;
return;
}
LASSERT(mea_size != 0);
OBD_FREE(lockh, sizeof(struct lustre_handle) * mea->mea_count);
OBD_FREE(mea, mea_size);
- return;
+ EXIT;
}
int mds_unlink_slave_objs(struct obd_device *obd, struct dentry *dentry)
{
struct mds_obd *mds = &obd->u.mds;
struct ptlrpc_request *req = NULL;
- struct mdc_op_data op_data;
+ struct mdc_op_data *op_data;
struct mea *mea = NULL;
int mea_size, rc;
+ ENTRY;
/* clustered MD ? */
- if (!mds->mds_lmv_obd)
- return 0;
+ if (!mds->mds_md_obd)
+ RETURN(0);
/* a dir can be splitted only */
if (!S_ISDIR(dentry->d_inode->i_mode))
RETURN(0);
- rc = mds_get_lmv_attr(obd, dentry->d_inode, &mea, &mea_size);
+ rc = mds_md_get_attr(obd, dentry->d_inode, &mea, &mea_size);
if (rc)
RETURN(rc);
if (mea == NULL)
- return 0;
+ RETURN(0);
+
if (mea->mea_count == 0)
GOTO(cleanup, rc = 0);
CDEBUG(D_OTHER, "%s: unlink slaves for %lu/%lu\n", obd->obd_name,
- (unsigned long) dentry->d_inode->i_ino,
- (unsigned long) dentry->d_inode->i_generation);
+ (unsigned long)dentry->d_inode->i_ino,
+ (unsigned long)dentry->d_inode->i_generation);
- memset(&op_data, 0, sizeof(op_data));
- op_data.mea1 = mea;
- rc = md_unlink(mds->mds_lmv_exp, &op_data, &req);
+ OBD_ALLOC(op_data, sizeof(*op_data));
+ if (op_data == NULL)
+ RETURN(-ENOMEM);
+
+ memset(op_data, 0, sizeof(*op_data));
+ op_data->mea1 = mea;
+ rc = md_unlink(mds->mds_md_exp, op_data, &req);
+ OBD_FREE(op_data, sizeof(*op_data));
LASSERT(req == NULL);
+ EXIT;
cleanup:
OBD_FREE(mea, mea_size);
- RETURN(rc);
+ return rc;
}
struct ide_tracking {
return -ENOTEMPTY;
}
+/* checks if passed dentry points to empty dir. */
int mds_is_dir_empty(struct obd_device *obd, struct dentry *dentry)
{
struct ide_tracking it;
OBD_ALLOC(file_name, nlen);
if (!file_name)
RETURN(-ENOMEM);
- i = sprintf(file_name, "__iopen__/0x%lx", dentry->d_inode->i_ino);
+
+ LASSERT(dentry->d_inode != NULL);
+ i = sprintf(file_name, "__iopen__/0x%lx",
+ dentry->d_inode->i_ino);
file = filp_open(file_name, O_RDONLY, 0);
if (IS_ERR(file)) {
struct dentry *dentry = NULL;
struct lvfs_run_ctxt saved;
int cleanup_phase = 0;
+ struct mds_req_sec_desc *rsd;
struct mds_body *body;
struct lvfs_ucred uc;
int rc, update_mode;
ENTRY;
+ rsd = lustre_swab_mds_secdesc(req, MDS_REQ_SECDESC_OFF);
+ if (!rsd) {
+ CERROR("Can't unpack security desc\n");
+ GOTO(cleanup, rc = -EFAULT);
+ }
+ mds_squash_root(&obd->u.mds, rsd, &req->rq_peer.peer_id.nid);
+
body = lustre_swab_reqbuf(req, offset, sizeof(*body),
lustre_swab_mds_body);
if (body == NULL) {
CERROR("Can't swab mds_body\n");
GOTO(cleanup, rc = -EFAULT);
}
- CDEBUG(D_OTHER, "%s: check slave %lu/%lu\n", obd->obd_name,
- (unsigned long) body->fid1.id,
- (unsigned long) body->fid1.generation);
- dentry = mds_fid2locked_dentry(obd, &body->fid1, NULL, LCK_EX, lockh,
- &update_mode, NULL, 0,
- MDS_INODELOCK_UPDATE);
+ CDEBUG(D_OTHER, "%s: check slave "DLID4"\n", obd->obd_name,
+ OLID4(&body->id1));
+
+ dentry = mds_id2locked_dentry(obd, &body->id1, NULL, LCK_EX,
+ lockh, &update_mode, NULL, 0,
+ MDS_INODELOCK_UPDATE);
if (IS_ERR(dentry)) {
CERROR("can't find inode: %d\n", (int) PTR_ERR(dentry));
GOTO(cleanup, rc = PTR_ERR(dentry));
}
cleanup_phase = 1;
- LASSERT(S_ISDIR(dentry->d_inode->i_mode));
+ /*
+ * handling the case when remote MDS checks if dir is empty before
+ * rename. But it also does it for all entries, because inode is stored
+ * here and remote MDS does not know if rename point to dir or to reg
+ * file. So we check it here.
+ */
+ if (!S_ISDIR(dentry->d_inode->i_mode))
+ GOTO(cleanup, rc = 0);
- uc.luc_fsuid = body->fsuid;
- uc.luc_fsgid = body->fsgid;
- uc.luc_cap = body->capability;
- uc.luc_suppgid1 = body->suppgid;
- uc.luc_suppgid2 = -1;
- push_ctxt(&saved, &obd->obd_lvfs_ctxt, &uc);
+ rc = mds_init_ucred(&uc, rsd);
+ if (rc) {
+ CERROR("can't init ucred\n");
+ GOTO(cleanup, rc);
+ }
- rc = 0;
- if (!mds_is_dir_empty(obd, dentry))
- rc = -ENOTEMPTY;
+ push_ctxt(&saved, &obd->obd_lvfs_ctxt, &uc);
+ rc = mds_is_dir_empty(obd, dentry) ? 0 : -ENOTEMPTY;
+ pop_ctxt(&saved, &obd->obd_lvfs_ctxt, &uc);
+ mds_exit_ucred(&uc);
+ EXIT;
cleanup:
switch(cleanup_phase) {
case 1:
if (rc)
ldlm_lock_decref(lockh, LCK_EX);
l_dput(dentry);
- pop_ctxt(&saved, &obd->obd_lvfs_ctxt, &uc);
default:
break;
}
- RETURN(rc);
+ return rc;
}
int mds_convert_mea_ea(struct obd_device *obd, struct inode *inode,
RETURN(0);
old = (struct mea_old *) lmm;
- rc = sizeof(struct ll_fid) * old->mea_count + sizeof(struct mea_old);
+
+ rc = sizeof(struct lustre_id) * old->mea_count +
+ sizeof(struct mea_old);
+
if (old->mea_count > 256 || old->mea_master > 256 || lmmsize < rc
|| old->mea_master > old->mea_count) {
CWARN("unknown MEA format, dont convert it\n");
}
CWARN("converting MEA EA on %lu/%u from V0 to V1 (%u/%u)\n",
- inode->i_ino, inode->i_generation, old->mea_count, old->mea_master);
+ inode->i_ino, inode->i_generation, old->mea_count,
+ old->mea_master);
- size = sizeof(struct ll_fid) * old->mea_count + sizeof(struct mea);
+ size = sizeof(struct lustre_id) * old->mea_count +
+ sizeof(struct mea);
+
OBD_ALLOC(new, size);
if (new == NULL)
RETURN(-ENOMEM);
new->mea_count = old->mea_count;
new->mea_master = old->mea_master;
for (i = 0; i < new->mea_count; i++)
- new->mea_fids[i] = old->mea_fids[i];
+ new->mea_ids[i] = old->mea_ids[i];
handle = fsfilt_start(obd, inode, FSFILT_OP_SETATTR, NULL);
if (IS_ERR(handle)) {
err = fsfilt_commit(obd, obd->u.mds.mds_sb, inode, handle, 0);
if (!rc)
rc = err ? err : size;
- GOTO(conv_free, rc);
+ EXIT;
conv_free:
OBD_FREE(new, size);
return rc;
git://git.whamcloud.com / fs / lustre-release.git / blobdiff