#include <asm/uaccess.h>
#include <linux/slab.h>
#include <asm/segment.h>
+#include <linux/random.h>
#include <linux/obd_support.h>
#include <linux/lustre_lib.h>
+#include <linux/lustre_sec.h>
+#include <linux/lustre_ucache.h>
+#include <linux/lustre_gs.h>
+#include <linux/lustre_fsfilt.h>
+
#include "mds_internal.h"
#if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,4)
RETURN(rc);
}
+void mds_inode2id(struct obd_device *obd, struct lustre_id *id,
+ struct inode *inode, __u64 fid)
+{
+ struct mds_obd *mds = &obd->u.mds;
+ ENTRY;
+
+ LASSERT(inode != NULL);
+ LASSERT(id != NULL);
+ LASSERT(fid != 0);
+
+ id_fid(id) = fid;
+ id_ino(id) = inode->i_ino;
+ id_group(id) = mds->mds_num;
+ id_gen(id) = inode->i_generation;
+ id_type(id) = (S_IFMT & inode->i_mode);
+
+ EXIT;
+}
+
+int mds_pack_gskey(struct obd_device *obd, struct lustre_msg *repmsg,
+ int *offset, struct mds_body *body, struct inode *inode)
+{
+ struct crypto_key_md *md_key;
+ struct crypto_key *ckey;
+ __u32 buflen, *sizep;
+ void *buf;
+ int size, rc = 0;
+ ENTRY;
+
+ sizep = lustre_msg_buf(repmsg, (*offset)++, 4);
+ if (!sizep) {
+ CERROR("can't locate returned ckey size buf\n");
+ RETURN(-EPROTO);
+ }
+ *sizep = cpu_to_le32(sizeof(*ckey));
+
+ OBD_ALLOC(md_key, sizeof(*md_key));
+ if (!md_key)
+ RETURN(-ENOMEM);
+
+ buflen = repmsg->buflens[*offset];
+ buf = lustre_msg_buf(repmsg, (*offset)++, buflen);
+
+ size = fsfilt_get_md(obd, inode, md_key, sizeof(*md_key),
+ EA_KEY);
+ if (size <= 0) {
+ if (size < 0)
+ CERROR("Can not get gskey from MDS ino %lu rc %d\n",
+ inode->i_ino, size);
+ GOTO(out, rc = size);
+ }
+ if (le32_to_cpu(md_key->md_magic) != MD_KEY_MAGIC) {
+ CDEBUG(D_INFO, "given match %x != magic %x\n",
+ md_key->md_magic, MD_KEY_MAGIC);
+ GOTO(out, rc = 0);
+ }
+
+ CDEBUG(D_INFO, "get key %s mac %s for ino %lu size %d \n",
+ md_key->md_ck.ck_key, md_key->md_ck.ck_mac, inode->i_ino, size);
+ ckey=(struct crypto_key*)buf;
+
+ memcpy(ckey, &md_key->md_ck, sizeof(*ckey));
+ body->valid |= OBD_MD_FLKEY;
+out:
+ OBD_FREE(md_key, sizeof(*md_key));
+ RETURN(rc);
+}
+
+static int mds_get_gskey(struct inode *inode, struct crypto_key *ckey)
+{
+ LASSERT(ckey);
+ /*tmp create gs key here*/
+ LASSERT(ckey->ck_type == MKS_TYPE);
+ get_random_bytes(ckey->ck_key, KEY_SIZE);
+ RETURN(0);
+}
+
+int mds_set_gskey(struct obd_device *obd, void *handle,
+ struct inode *inode, void *key, int key_len,
+ int valid)
+{
+ struct crypto_key_md *md_key = NULL;
+ struct crypto_key *ckey = (struct crypto_key *)key;
+ int rc = 0;
+ ENTRY;
+
+ if (!ckey)
+ RETURN(0);
+
+ LASSERT(ckey->ck_type == MKS_TYPE || ckey->ck_type == GKS_TYPE);
+
+ OBD_ALLOC(md_key, sizeof(*md_key));
+ if (ckey->ck_type == MKS_TYPE)
+ mds_get_gskey(inode, ckey);
+
+ rc = fsfilt_get_md(obd, inode, md_key, sizeof(*md_key),
+ EA_KEY);
+ if (rc < 0)
+ GOTO(free, rc);
+ LASSERT(le32_to_cpu(md_key->md_magic) == MD_KEY_MAGIC ||
+ md_key->md_magic == 0);
+
+ if (le32_to_cpu(md_key->md_magic) == MD_KEY_MAGIC) {
+ CDEBUG(D_INFO, "reset key %s mac %s", md_key->md_ck.ck_mac,
+ md_key->md_ck.ck_key);
+ }
+
+ md_key->md_magic = cpu_to_le32(MD_KEY_MAGIC);
+ /*get key and mac from request buffer*/
+ if (valid & ATTR_MAC) {
+ memcpy(md_key->md_ck.ck_mac, ckey->ck_mac, MAC_SIZE);
+ CDEBUG(D_INFO, "set mac %s for ino %lu \n",
+ md_key->md_ck.ck_mac, inode->i_ino);
+ }
+ if (valid & ATTR_KEY) {
+ memcpy(md_key->md_ck.ck_key, ckey->ck_key, KEY_SIZE);
+ CDEBUG(D_INFO, "set key %s for ino %lu \n",
+ md_key->md_ck.ck_key, inode->i_ino);
+ }
+ rc = fsfilt_set_md(obd, inode, handle, md_key, sizeof(*md_key), EA_KEY);
+free:
+ if (md_key)
+ OBD_FREE(md_key, sizeof(*md_key));
+ RETURN(rc);
+}
+
+int mds_set_crypto_type(struct obd_device *obd, void *val, __u32 vallen)
+{
+ struct mds_obd *mds = &obd->u.mds;
+ ENTRY;
+ if (vallen >= strlen("mks") &&
+ memcmp(val, "mks", vallen) == 0) {
+ mds->mds_crypto_type = MKS_TYPE;
+ CDEBUG(D_IOCTL, "mks type\n");
+ }
+ if (vallen >= strlen("gks") &&
+ memcmp(val, "gks", vallen) == 0) {
+ mds->mds_crypto_type = GKS_TYPE;
+ CDEBUG(D_IOCTL, "gks type \n");
+ }
+ RETURN(0);
+}
+
/* Note that we can copy all of the fields, just some will not be "valid" */
void mds_pack_inode2body(struct obd_device *obd, struct mds_body *b,
struct inode *inode, int fid)
RETURN (-EFAULT);
r->ur_id1 = &rec->sa_id;
+ r->ur_flags = rec->sa_flags;
attr->ia_valid = rec->sa_valid;
attr->ia_mode = rec->sa_mode;
attr->ia_uid = rec->sa_uid;
LTIME_S(attr->ia_ctime) = rec->sa_ctime;
attr->ia_attr_flags = rec->sa_attr_flags;
- LASSERT_REQSWAB (req, offset + 1);
+ LASSERT_REQSWAB(req, offset + 1);
if (req->rq_reqmsg->bufcount > offset + 1) {
- r->ur_eadata = lustre_msg_buf (req->rq_reqmsg,
- offset + 1, 0);
+ r->ur_eadata = lustre_msg_buf(req->rq_reqmsg,
+ offset + 1, 0);
if (r->ur_eadata == NULL)
RETURN (-EFAULT);
r->ur_eadatalen = req->rq_reqmsg->buflens[offset + 1];
r->ur_ea2datalen = req->rq_reqmsg->buflens[offset + 2];
}
+ if (req->rq_reqmsg->bufcount > offset + 3) {
+ r->ur_ea3data = lustre_msg_buf(req->rq_reqmsg, offset + 3, 0);
+ if (r->ur_ea3data == NULL)
+ RETURN (-EFAULT);
+
+ r->ur_ea3datalen = req->rq_reqmsg->buflens[offset + 3];
+ }
+
RETURN(0);
}
struct mds_rec_create *rec;
ENTRY;
- rec = lustre_swab_reqbuf (req, offset, sizeof (*rec),
- lustre_swab_mds_rec_create);
+ rec = lustre_swab_reqbuf(req, offset, sizeof(*rec),
+ lustre_swab_mds_rec_create);
if (rec == NULL)
RETURN (-EFAULT);
r->ur_time = rec->cr_time;
r->ur_flags = rec->cr_flags;
- LASSERT_REQSWAB (req, offset + 1);
- r->ur_name = lustre_msg_string (req->rq_reqmsg, offset + 1, 0);
+ LASSERT_REQSWAB(req, offset + 1);
+ r->ur_name = lustre_msg_string(req->rq_reqmsg, offset + 1, 0);
if (r->ur_name == NULL)
- RETURN (-EFAULT);
+ RETURN(-EFAULT);
r->ur_namelen = req->rq_reqmsg->buflens[offset + 1];
- LASSERT_REQSWAB (req, offset + 2);
+ LASSERT_REQSWAB(req, offset + 2);
if (req->rq_reqmsg->bufcount > offset + 2) {
if (S_ISLNK(r->ur_mode)) {
r->ur_tgt = lustre_msg_string(req->rq_reqmsg,
offset + 2, 0);
if (r->ur_tgt == NULL)
- RETURN (-EFAULT);
+ RETURN(-EFAULT);
r->ur_tgtlen = req->rq_reqmsg->buflens[offset + 2];
- } else if (S_ISDIR(r->ur_mode)) {
+ } else if (S_ISDIR(r->ur_mode) ) {
/* Stripe info for mkdir - just a 16bit integer */
if (req->rq_reqmsg->buflens[offset + 2] != 2) {
CERROR("mkdir stripe info does not match "
"expected size %d vs 2\n",
req->rq_reqmsg->buflens[offset + 2]);
- RETURN (-EINVAL);
+ RETURN(-EINVAL);
}
- r->ur_eadata = lustre_swab_buf (req->rq_reqmsg,
- offset + 2, 2, __swab16s);
+ r->ur_eadata = lustre_swab_buf(req->rq_reqmsg,
+ offset + 2, 2,
+ __swab16s);
+ r->ur_eadatalen = req->rq_reqmsg->buflens[offset + 2];
+ } else if (S_ISREG(r->ur_mode)){
+ r->ur_eadata = lustre_msg_buf(req->rq_reqmsg,
+ offset + 2, 0);
r->ur_eadatalen = req->rq_reqmsg->buflens[offset + 2];
} else {
/* Hm, no other users so far? */
struct mds_rec_link *rec;
ENTRY;
- rec = lustre_swab_reqbuf (req, offset, sizeof (*rec),
- lustre_swab_mds_rec_link);
+ rec = lustre_swab_reqbuf(req, offset, sizeof(*rec),
+ lustre_swab_mds_rec_link);
if (rec == NULL)
- RETURN (-EFAULT);
+ RETURN(-EFAULT);
r->ur_id1 = &rec->lk_id1;
r->ur_id2 = &rec->lk_id2;
r->ur_time = rec->lk_time;
+ r->ur_flags = rec->lk_flags;
- LASSERT_REQSWAB (req, offset + 1);
- r->ur_name = lustre_msg_string (req->rq_reqmsg, offset + 1, 0);
+ LASSERT_REQSWAB(req, offset + 1);
+ r->ur_name = lustre_msg_string(req->rq_reqmsg, offset + 1, 0);
if (r->ur_name == NULL)
- RETURN (-EFAULT);
+ RETURN(-EFAULT);
r->ur_namelen = req->rq_reqmsg->buflens[offset + 1];
RETURN(0);
}
struct mds_rec_unlink *rec;
ENTRY;
- rec = lustre_swab_reqbuf (req, offset, sizeof (*rec),
- lustre_swab_mds_rec_unlink);
+ rec = lustre_swab_reqbuf(req, offset, sizeof (*rec),
+ lustre_swab_mds_rec_unlink);
if (rec == NULL)
RETURN(-EFAULT);
r->ur_id1 = &rec->ul_id1;
r->ur_id2 = &rec->ul_id2;
r->ur_time = rec->ul_time;
+ r->ur_flags = rec->ul_flags;
- LASSERT_REQSWAB (req, offset + 1);
+ LASSERT_REQSWAB(req, offset + 1);
r->ur_name = lustre_msg_string(req->rq_reqmsg, offset + 1, 0);
if (r->ur_name == NULL)
RETURN(-EFAULT);
struct mds_rec_rename *rec;
ENTRY;
- rec = lustre_swab_reqbuf (req, offset, sizeof (*rec),
- lustre_swab_mds_rec_rename);
+ rec = lustre_swab_reqbuf(req, offset, sizeof (*rec),
+ lustre_swab_mds_rec_rename);
if (rec == NULL)
RETURN(-EFAULT);
r->ur_id1 = &rec->rn_id1;
r->ur_id2 = &rec->rn_id2;
r->ur_time = rec->rn_time;
+ r->ur_flags = rec->rn_flags;
- LASSERT_REQSWAB (req, offset + 1);
+ LASSERT_REQSWAB(req, offset + 1);
r->ur_name = lustre_msg_string(req->rq_reqmsg, offset + 1, 0);
if (r->ur_name == NULL)
RETURN(-EFAULT);
r->ur_namelen = req->rq_reqmsg->buflens[offset + 1];
- LASSERT_REQSWAB (req, offset + 2);
+ LASSERT_REQSWAB(req, offset + 2);
r->ur_tgt = lustre_msg_string(req->rq_reqmsg, offset + 2, 0);
if (r->ur_tgt == NULL)
RETURN(-EFAULT);
struct mds_rec_create *rec;
ENTRY;
- rec = lustre_swab_reqbuf (req, offset, sizeof (*rec),
- lustre_swab_mds_rec_create);
+ rec = lustre_swab_reqbuf(req, offset, sizeof (*rec),
+ lustre_swab_mds_rec_create);
if (rec == NULL)
- RETURN (-EFAULT);
+ RETURN(-EFAULT);
r->ur_id1 = &rec->cr_id;
r->ur_id2 = &rec->cr_replayid;
r->ur_rdev = rec->cr_rdev;
r->ur_time = rec->cr_time;
r->ur_flags = rec->cr_flags;
+ r->ur_ioepoch = rec->cr_ioepoch;
- LASSERT_REQSWAB (req, offset + 1);
- r->ur_name = lustre_msg_string (req->rq_reqmsg, offset + 1, 0);
+ LASSERT_REQSWAB(req, offset + 1);
+ r->ur_name = lustre_msg_string(req->rq_reqmsg, offset + 1, 0);
+
if (r->ur_name == NULL)
- RETURN (-EFAULT);
+ RETURN(-EFAULT);
r->ur_namelen = req->rq_reqmsg->buflens[offset + 1];
- LASSERT_REQSWAB (req, offset + 2);
+ LASSERT_REQSWAB(req, offset + 2);
if (req->rq_reqmsg->bufcount > offset + 2) {
r->ur_eadata = lustre_msg_buf(req->rq_reqmsg, offset + 2, 0);
if (r->ur_eadata == NULL)
- RETURN (-EFAULT);
+ RETURN(-EFAULT);
r->ur_eadatalen = req->rq_reqmsg->buflens[offset + 2];
}
+
+ if (rec->cr_flags & MDS_OPEN_HAS_KEY) {
+ LASSERT(req->rq_reqmsg->bufcount > offset + 3);
+ r->ur_ea2data = lustre_msg_buf(req->rq_reqmsg, offset + 3, 0);
+ r->ur_ea2datalen = req->rq_reqmsg->buflens[offset + 3];
+ }
RETURN(0);
}
int rc;
ENTRY;
- /*
- * NB don't lustre_swab_reqbuf() here. We're just taking a peek and we
+ /* NB don't lustre_swab_reqbuf() here. We're just taking a peek and we
* want to leave it to the specific unpacker once we've identified the
- * message type.
- */
- opcodep = lustre_msg_buf (req->rq_reqmsg, offset, sizeof(*opcodep));
+ * message type. */
+ opcodep = lustre_msg_buf(req->rq_reqmsg, offset, sizeof(*opcodep));
if (opcodep == NULL)
RETURN(-EFAULT);
opcode = *opcodep;
- if (lustre_msg_swabbed (req->rq_reqmsg))
- __swab32s (&opcode);
+ if (lustre_msg_swabbed(req->rq_reqmsg))
+ __swab32s(&opcode);
if (opcode > REINT_MAX ||
mds_unpackers[opcode] == NULL) {
- CERROR ("Unexpected opcode %d\n", opcode);
+ CERROR("Unexpected opcode %d\n", opcode);
RETURN(-EFAULT);
}
rec->ur_opcode = opcode;
rc = mds_unpackers[opcode](req, offset, rec);
+
+#if CRAY_PORTALS
+ rec->ur_fsuid = req->rq_uid;
+#endif
RETURN(rc);
}
+/*
+ * here we take simple rule: once uid/fsuid is root, we also squash
+ * the gid/fsgid, don't care setuid/setgid attributes.
+ */
static
-struct mds_idmap_table *__get_idmap_table(struct mds_export_data *med,
- int create)
+int mds_squash_root(struct mds_obd *mds, struct mds_req_sec_desc *rsd,
+ ptl_nid_t *peernid)
{
- struct mds_idmap_table *new;
- int i;
+ if (!mds->mds_squash_uid || *peernid == mds->mds_nosquash_nid)
+ return 0;
- if (!create || med->med_idmap)
- return med->med_idmap;
+ if (rsd->rsd_uid && rsd->rsd_fsuid)
+ return 0;
- spin_unlock(&med->med_idmap_lock);
- OBD_ALLOC(new, sizeof(*new));
- spin_lock(&med->med_idmap_lock);
+ CDEBUG(D_SEC, "squash req from "LPX64":"
+ "(%u:%u-%u:%u/%x)=>(%u:%u-%u:%u/%x)\n", *peernid,
+ rsd->rsd_uid, rsd->rsd_gid,
+ rsd->rsd_fsuid, rsd->rsd_fsgid, rsd->rsd_cap,
+ rsd->rsd_uid ? rsd->rsd_uid : mds->mds_squash_uid,
+ rsd->rsd_uid ? rsd->rsd_gid : mds->mds_squash_gid,
+ rsd->rsd_fsuid ? rsd->rsd_fsuid : mds->mds_squash_uid,
+ rsd->rsd_fsuid ? rsd->rsd_fsgid : mds->mds_squash_gid,
+ rsd->rsd_cap & ~CAP_FS_MASK);
+
+ if (rsd->rsd_uid == 0) {
+ rsd->rsd_uid = mds->mds_squash_uid;
+ rsd->rsd_gid = mds->mds_squash_gid;
+ }
+ if (rsd->rsd_fsuid == 0) {
+ rsd->rsd_fsuid = mds->mds_squash_uid;
+ rsd->rsd_fsgid = mds->mds_squash_gid;
+ }
+ rsd->rsd_cap &= ~CAP_FS_MASK;
+
+ return 1;
+}
+
+/********************************
+ * MDS uid/gid mapping handling *
+ ********************************/
+
+static
+struct mds_idmap_entry* idmap_alloc_entry(__u32 rmt_id, __u32 lcl_id)
+{
+ struct mds_idmap_entry *e;
- if (!new) {
- CERROR("fail to alloc %d\n", sizeof(*new));
+ OBD_ALLOC(e, sizeof(*e));
+ if (!e)
return NULL;
- }
- if (med->med_idmap) {
- OBD_FREE(new, sizeof(*new));
- return med->med_idmap;
- }
+ INIT_LIST_HEAD(&e->rmt_hash);
+ INIT_LIST_HEAD(&e->lcl_hash);
+ atomic_set(&e->refcount, 1);
+ e->rmt_id = rmt_id;
+ e->lcl_id = lcl_id;
- for (i = 0; i < MDS_IDMAP_HASHSIZE; i++) {
- INIT_LIST_HEAD(&new->uidmap[i]);
- INIT_LIST_HEAD(&new->gidmap[i]);
- }
+ return e;
+}
- CDEBUG(D_SEC, "allocate idmap table for med %p\n", med);
- med->med_idmap = new;
- return new;
+void idmap_free_entry(struct mds_idmap_entry *e)
+{
+ if (!list_empty(&e->rmt_hash))
+ list_del(&e->rmt_hash);
+ if (!list_empty(&e->lcl_hash))
+ list_del(&e->lcl_hash);
+ OBD_FREE(e, sizeof(*e));
}
-static void __flush_mapping_table(struct list_head *table)
+static
+int idmap_insert_entry(struct list_head *rmt_hash, struct list_head *lcl_hash,
+ struct mds_idmap_entry *new, const char *warn_msg)
{
- struct mds_idmap_item *item;
- int i;
+ struct list_head *rmt_head = &rmt_hash[MDS_IDMAP_HASHFUNC(new->rmt_id)];
+ struct list_head *lcl_head = &lcl_hash[MDS_IDMAP_HASHFUNC(new->lcl_id)];
+ struct mds_idmap_entry *e;
+
+ list_for_each_entry(e, rmt_head, rmt_hash) {
+ if (e->rmt_id == new->rmt_id &&
+ e->lcl_id == new->lcl_id) {
+ atomic_inc(&e->refcount);
+ return 1;
+ }
+ if (e->rmt_id == new->rmt_id && warn_msg)
+ CWARN("%s: rmt id %u already map to %u (new %u)\n",
+ warn_msg, e->rmt_id, e->lcl_id, new->lcl_id);
+ if (e->lcl_id == new->lcl_id && warn_msg)
+ CWARN("%s: lcl id %u already be mapped from %u "
+ "(new %u)\n", warn_msg,
+ e->lcl_id, e->rmt_id, new->rmt_id);
+ }
- for (i = 0; i < MDS_IDMAP_HASHSIZE; i++) {
- while (!list_empty(&table[i])) {
- item = list_entry(table[i].next, struct mds_idmap_item,
- hash);
- list_del(&item->hash);
- OBD_FREE(item, sizeof(*item));
+ list_add_tail(rmt_head, &new->rmt_hash);
+ list_add_tail(lcl_head, &new->lcl_hash);
+ return 0;
+}
+
+static
+int idmap_remove_entry(struct list_head *rmt_hash, struct list_head *lcl_hash,
+ __u32 rmt_id, __u32 lcl_id)
+{
+ struct list_head *rmt_head = &rmt_hash[MDS_IDMAP_HASHFUNC(rmt_id)];
+ struct mds_idmap_entry *e;
+
+ list_for_each_entry(e, rmt_head, rmt_hash) {
+ if (e->rmt_id == rmt_id && e->lcl_id == lcl_id) {
+ if (atomic_dec_and_test(&e->refcount)) {
+ list_del(&e->rmt_hash);
+ list_del(&e->lcl_hash);
+ OBD_FREE(e, sizeof(*e));
+ return 0;
+ } else
+ return 1;
}
}
+ return -ENOENT;
}
-void mds_idmap_cleanup(struct mds_export_data *med)
+int mds_idmap_add(struct mds_idmap_table *tbl,
+ uid_t rmt_uid, uid_t lcl_uid,
+ gid_t rmt_gid, gid_t lcl_gid)
{
+ struct mds_idmap_entry *ue, *ge;
ENTRY;
- if (!med->med_idmap) {
- EXIT;
- return;
+ if (!tbl)
+ RETURN(-EPERM);
+
+ ue = idmap_alloc_entry(rmt_uid, lcl_uid);
+ if (!ue)
+ RETURN(-ENOMEM);
+ ge = idmap_alloc_entry(rmt_gid, lcl_gid);
+ if (!ge) {
+ idmap_free_entry(ue);
+ RETURN(-ENOMEM);
}
- spin_lock(&med->med_idmap_lock);
- __flush_mapping_table(med->med_idmap->uidmap);
- __flush_mapping_table(med->med_idmap->gidmap);
- OBD_FREE(med->med_idmap, sizeof(struct mds_idmap_table));
- spin_unlock(&med->med_idmap_lock);
+ spin_lock(&tbl->mit_lock);
+
+ if (idmap_insert_entry(tbl->mit_idmaps[MDS_RMT_UIDMAP_IDX],
+ tbl->mit_idmaps[MDS_LCL_UIDMAP_IDX],
+ ue, "UID mapping")) {
+ idmap_free_entry(ue);
+ }
+
+ if (idmap_insert_entry(tbl->mit_idmaps[MDS_RMT_GIDMAP_IDX],
+ tbl->mit_idmaps[MDS_LCL_GIDMAP_IDX],
+ ge, "GID mapping")) {
+ idmap_free_entry(ge);
+ }
+
+ spin_unlock(&tbl->mit_lock);
+ RETURN(0);
}
-static inline int idmap_hash(__u32 id)
+int mds_idmap_del(struct mds_idmap_table *tbl,
+ uid_t rmt_uid, uid_t lcl_uid,
+ gid_t rmt_gid, gid_t lcl_gid)
{
- return (id & (MDS_IDMAP_HASHSIZE - 1));
+ ENTRY;
+
+ if (!tbl)
+ RETURN(0);
+
+ spin_lock(&tbl->mit_lock);
+ idmap_remove_entry(tbl->mit_idmaps[MDS_RMT_UIDMAP_IDX],
+ tbl->mit_idmaps[MDS_LCL_UIDMAP_IDX],
+ rmt_uid, lcl_uid);
+ idmap_remove_entry(tbl->mit_idmaps[MDS_RMT_GIDMAP_IDX],
+ tbl->mit_idmaps[MDS_LCL_GIDMAP_IDX],
+ rmt_gid, lcl_gid);
+ spin_unlock(&tbl->mit_lock);
+ RETURN(0);
}
static
-int __idmap_set_item(struct mds_export_data *med,
- struct list_head *table,
- __u32 id1, __u32 id2)
-{
- struct list_head *head;
- struct mds_idmap_item *item, *new = NULL;
- int found = 0;
-
- head = table + idmap_hash(id1);
-again:
- list_for_each_entry(item, head, hash) {
- if (item->id1 == id1) {
- found = 1;
- break;
- }
- }
+__u32 idmap_lookup_id(struct list_head *hash, int reverse, __u32 id)
+{
+ struct list_head *head = &hash[MDS_IDMAP_HASHFUNC(id)];
+ struct mds_idmap_entry *e;
- if (!found) {
- if (new == NULL) {
- spin_unlock(&med->med_idmap_lock);
- OBD_ALLOC(new, sizeof(*new));
- spin_lock(&med->med_idmap_lock);
- if (!new) {
- CERROR("fail to alloc %d\n", sizeof(*new));
- return -ENOMEM;
- }
- goto again;
+ if (!reverse) {
+ list_for_each_entry(e, head, rmt_hash) {
+ if (e->rmt_id == id)
+ return e->lcl_id;
}
- new->id1 = id1;
- new->id2 = id2;
- list_add(&new->hash, head);
+ return MDS_IDMAP_NOTFOUND;
} else {
- if (new)
- OBD_FREE(new, sizeof(*new));
- if (item->id2 != id2) {
- CWARN("mapping changed: %u ==> (%u -> %u)\n",
- id1, item->id2, id2);
- item->id2 = id2;
+ list_for_each_entry(e, head, lcl_hash) {
+ if (e->lcl_id == id)
+ return e->rmt_id;
}
- list_move(&item->hash, head);
+ return MDS_IDMAP_NOTFOUND;
}
-
- return 0;
}
-int mds_idmap_set(struct mds_export_data *med, __u32 id1, __u32 id2,
- int is_uid_mapping)
+int mds_idmap_lookup_uid(struct mds_idmap_table *tbl, int reverse, uid_t uid)
{
- struct mds_idmap_table *idmap;
- int rc;
- ENTRY;
+ struct list_head *hash;
- spin_lock(&med->med_idmap_lock);
+ if (!tbl)
+ return MDS_IDMAP_NOTFOUND;
- idmap = __get_idmap_table(med, 1);
- if (!idmap)
- GOTO(out, rc = -ENOMEM);
+ if (!reverse)
+ hash = tbl->mit_idmaps[MDS_RMT_UIDMAP_IDX];
+ else
+ hash = tbl->mit_idmaps[MDS_LCL_UIDMAP_IDX];
+
+ spin_lock(&tbl->mit_lock);
+ uid = idmap_lookup_id(hash, reverse, uid);
+ spin_unlock(&tbl->mit_lock);
+
+ return uid;
+}
- if (is_uid_mapping)
- rc = __idmap_set_item(med, idmap->uidmap, id1, id2);
+int mds_idmap_lookup_gid(struct mds_idmap_table *tbl, int reverse, gid_t gid)
+{
+ struct list_head *hash;
+
+ if (!tbl)
+ return MDS_IDMAP_NOTFOUND;
+
+ if (!reverse)
+ hash = tbl->mit_idmaps[MDS_RMT_GIDMAP_IDX];
else
- rc = __idmap_set_item(med, idmap->gidmap, id1, id2);
+ hash = tbl->mit_idmaps[MDS_LCL_GIDMAP_IDX];
-out:
- spin_unlock(&med->med_idmap_lock);
- RETURN(rc);
+ spin_lock(&tbl->mit_lock);
+ gid = idmap_lookup_id(hash, reverse, gid);
+ spin_unlock(&tbl->mit_lock);
+
+ return gid;
}
-__u32 mds_idmap_get(struct mds_export_data *med, __u32 id,
- int is_uid_mapping)
+struct mds_idmap_table *mds_idmap_alloc()
{
- struct mds_idmap_table *idmap;
- struct list_head *table;
- struct list_head *head;
- struct mds_idmap_item *item;
- int found = 0;
- __u32 res;
+ struct mds_idmap_table *tbl;
+ int i, j;
+
+ OBD_ALLOC(tbl, sizeof(*tbl));
+ if (!tbl)
+ return NULL;
- spin_lock(&med->med_idmap_lock);
- idmap = __get_idmap_table(med, 0);
- if (!idmap)
- goto nllu;
+ spin_lock_init(&tbl->mit_lock);
+ for (i = 0; i < MDS_IDMAP_N_HASHES; i++)
+ for (j = 0; j < MDS_IDMAP_HASHSIZE; j++)
+ INIT_LIST_HEAD(&tbl->mit_idmaps[i][j]);
- table = is_uid_mapping ? idmap->uidmap : idmap->gidmap;
- head = table + idmap_hash(id);
+ return tbl;
+}
- list_for_each_entry(item, head, hash) {
- if (item->id1 == id) {
- found = 1;
- break;
+static void idmap_clear_rmt_hash(struct list_head *list)
+{
+ struct mds_idmap_entry *e;
+ int i;
+
+ for (i = 0; i < MDS_IDMAP_HASHSIZE; i++) {
+ while (!list_empty(&list[i])) {
+ e = list_entry(list[i].next, struct mds_idmap_entry,
+ rmt_hash);
+ idmap_free_entry(e);
}
}
- if (!found)
- goto nllu;
+}
- res = item->id2;
-out:
- spin_unlock(&med->med_idmap_lock);
- return res;
-nllu:
- res = is_uid_mapping ? med->med_nllu : med->med_nllg;
- goto out;
+void mds_idmap_free(struct mds_idmap_table *tbl)
+{
+ int i;
+
+ spin_lock(&tbl->mit_lock);
+ idmap_clear_rmt_hash(tbl->mit_idmaps[MDS_RMT_UIDMAP_IDX]);
+ idmap_clear_rmt_hash(tbl->mit_idmaps[MDS_RMT_GIDMAP_IDX]);
+
+ /* paranoid checking */
+ for (i = 0; i < MDS_IDMAP_HASHSIZE; i++) {
+ LASSERT(list_empty(&tbl->mit_idmaps[MDS_LCL_UIDMAP_IDX][i]));
+ LASSERT(list_empty(&tbl->mit_idmaps[MDS_LCL_GIDMAP_IDX][i]));
+ }
+ spin_unlock(&tbl->mit_lock);
+
+ OBD_FREE(tbl, sizeof(*tbl));
}
-void mds_reverse_map_ugid(struct ptlrpc_request *req,
- struct mds_body *body)
+/*********************************
+ * helpers doing mapping for MDS *
+ *********************************/
+
+/*
+ * we allow remote setuid/setgid to an "authencated" one,
+ * this policy probably change later.
+ */
+static
+int mds_req_secdesc_do_map(struct mds_export_data *med,
+ struct mds_req_sec_desc *rsd)
{
- struct mds_export_data *med = &req->rq_export->u.eu_mds_data;
+ struct mds_idmap_table *idmap = med->med_idmap;
+ uid_t uid, fsuid;
+ gid_t gid, fsgid;
+
+ uid = mds_idmap_lookup_uid(idmap, 0, rsd->rsd_uid);
+ if (uid == MDS_IDMAP_NOTFOUND) {
+ CERROR("can't find map for uid %u\n", rsd->rsd_uid);
+ return -EPERM;
+ }
+
+ if (rsd->rsd_uid == rsd->rsd_fsuid)
+ fsuid = uid;
+ else {
+ fsuid = mds_idmap_lookup_uid(idmap, 0, rsd->rsd_fsuid);
+ if (fsuid == MDS_IDMAP_NOTFOUND) {
+ CERROR("can't find map for fsuid %u\n", rsd->rsd_fsuid);
+ return -EPERM;
+ }
+ }
- LASSERT(req->rq_remote);
+ gid = mds_idmap_lookup_gid(idmap, 0, rsd->rsd_gid);
+ if (gid == MDS_IDMAP_NOTFOUND) {
+ CERROR("can't find map for gid %u\n", rsd->rsd_gid);
+ return -EPERM;
+ }
+
+ if (rsd->rsd_gid == rsd->rsd_fsgid)
+ fsgid = gid;
+ else {
+ fsgid = mds_idmap_lookup_gid(idmap, 0, rsd->rsd_fsgid);
+ if (fsgid == MDS_IDMAP_NOTFOUND) {
+ CERROR("can't find map for fsgid %u\n", rsd->rsd_fsgid);
+ return -EPERM;
+ }
+ }
- if (body->valid & OBD_MD_FLUID)
- body->uid = mds_idmap_get(med, body->uid, 1);
+ rsd->rsd_uid = uid;
+ rsd->rsd_gid = gid;
+ rsd->rsd_fsuid = fsuid;
+ rsd->rsd_fsgid = fsgid;
- if (body->valid & OBD_MD_FLGID)
- body->gid = mds_idmap_get(med, body->gid, 0);
+ return 0;
}
+void mds_body_do_reverse_map(struct mds_export_data *med,
+ struct mds_body *body)
+{
+ uid_t uid;
+ gid_t gid;
+
+ if (!med->med_remote)
+ return;
+
+ ENTRY;
+ if (body->valid & OBD_MD_FLUID) {
+ uid = mds_idmap_lookup_uid(med->med_idmap, 1, body->uid);
+ if (uid == MDS_IDMAP_NOTFOUND) {
+ uid = med->med_nllu;
+ if (body->valid & OBD_MD_FLMODE) {
+ body->mode = (body->mode & ~S_IRWXU) |
+ ((body->mode & S_IRWXO) << 6);
+ }
+ }
+ body->uid = uid;
+ }
+ if (body->valid & OBD_MD_FLGID) {
+ gid = mds_idmap_lookup_gid(med->med_idmap, 1, body->gid);
+ if (gid == MDS_IDMAP_NOTFOUND) {
+ gid = med->med_nllg;
+ if (body->valid & OBD_MD_FLMODE) {
+ body->mode = (body->mode & ~S_IRWXG) |
+ ((body->mode & S_IRWXO) << 3);
+ }
+ }
+ body->gid = gid;
+ }
+
+ EXIT;
+}
+
+/*
+ * return error if can't find mapping, it's a error so should not
+ * fall into nllu/nllg.
+ */
+int mds_remote_perm_do_reverse_map(struct mds_export_data *med,
+ struct mds_remote_perm *perm)
+{
+ uid_t uid;
+ gid_t gid;
+
+ LASSERT(med->med_remote);
+
+ uid = mds_idmap_lookup_uid(med->med_idmap, 1, perm->mrp_auth_uid);
+ if (uid == MDS_IDMAP_NOTFOUND) {
+ CERROR("no map for uid %u\n", perm->mrp_auth_uid);
+ return -EPERM;
+ }
+ gid = mds_idmap_lookup_gid(med->med_idmap, 1, perm->mrp_auth_gid);
+ if (gid == MDS_IDMAP_NOTFOUND) {
+ CERROR("no map for uid %u\n", perm->mrp_auth_uid);
+ return -EPERM;
+ }
+
+ perm->mrp_auth_uid = uid;
+ perm->mrp_auth_gid = gid;
+ return 0;
+}
+
+/**********************
+ * MDS ucred handling *
+ **********************/
+
static inline void drop_ucred_ginfo(struct lvfs_ucred *ucred)
{
if (ucred->luc_ginfo) {
* root could set any group_info if we allowed setgroups, while
* normal user only could 'reduce' their group members -- which
* is somewhat expensive.
+ *
+ * authenticated as mds user (using mds service credential) could
+ * bypass all checkings.
*/
int mds_init_ucred(struct lvfs_ucred *ucred,
struct ptlrpc_request *req,
struct lustre_sec_desc *lsd;
ptl_nid_t peernid = req->rq_peer.peer_id.nid;
struct group_info *gnew;
- unsigned int setuid, setgid, strong_sec;
+ unsigned int setuid, setgid, strong_sec, root_squashed;
+ __u32 lsd_perms;
ENTRY;
LASSERT(ucred);
LASSERT(rsd);
LASSERT(rsd->rsd_ngroups <= LUSTRE_MAX_GROUPS);
- strong_sec = (req->rq_auth_uid != -1);
- LASSERT(!(req->rq_remote && !strong_sec));
-
- /* sanity check & set local/remote flag */
- if (req->rq_remote) {
- if (med->med_local) {
- CWARN("exp %p: client on nid "LPX64" was local, "
- "set to remote\n", req->rq_export, peernid);
- med->med_local = 0;
- }
- } else {
- if (!med->med_local) {
- CWARN("exp %p: client on nid "LPX64" was remote, "
- "set to local\n", req->rq_export, peernid);
- med->med_local = 1;
- }
- }
+ if (SEC_FLAVOR_MAJOR(req->rq_req_secflvr) == PTLRPCS_FLVR_MAJOR_GSS &&
+ (SEC_FLAVOR_SVC(req->rq_req_secflvr) == PTLRPCS_SVC_AUTH ||
+ SEC_FLAVOR_SVC(req->rq_req_secflvr) == PTLRPCS_SVC_PRIV))
+ strong_sec = 1;
+ else
+ strong_sec = 0;
- setuid = (rsd->rsd_fsuid != rsd->rsd_uid);
- setgid = (rsd->rsd_fsgid != rsd->rsd_gid);
+ LASSERT(!(req->rq_remote_realm && !strong_sec));
- /* deny setuid/setgid for remote client */
- if ((setuid || setgid) && !med->med_local) {
- CWARN("deny setxid (%u/%u) from remote client "LPX64"\n",
- setuid, setgid, peernid);
+ if (strong_sec && req->rq_auth_uid == -1) {
+ CWARN("user not authenticated, deny access\n");
RETURN(-EPERM);
}
- /* take care of uid/gid mapping for client in remote realm */
- if (req->rq_remote) {
- /* record the uid mapping here */
- mds_idmap_set(med, req->rq_auth_uid, rsd->rsd_uid, 1);
-
- /* now we act as the authenticated user */
- rsd->rsd_uid = rsd->rsd_fsuid = req->rq_auth_uid;
- } else if (strong_sec && req->rq_auth_uid != rsd->rsd_uid) {
- /* if we use strong authentication on this request, we
- * expect the uid which client claimed is true.
- *
- * FIXME root's machine_credential in krb5 will be interpret
- * as "nobody", which is not good for mds-mds and mds-ost
- * connection.
- */
- CWARN("nid "LPX64": UID %u was authenticated while client "
- "claimed %u, set %u by force\n",
- peernid, req->rq_auth_uid, rsd->rsd_uid,
- req->rq_auth_uid);
- rsd->rsd_uid = req->rq_auth_uid;
+ /* sanity check: if we use strong authentication, we expect the
+ * uid which client claimed is true.
+ * not apply to special mds user .
+ */
+ if (!req->rq_auth_usr_mds && strong_sec) {
+ if (!med->med_remote) {
+ if (req->rq_auth_uid != rsd->rsd_uid) {
+ CERROR("local client "LPU64": auth uid %u "
+ "while client claim %u:%u/%u:%u\n",
+ peernid, req->rq_auth_uid,
+ rsd->rsd_uid, rsd->rsd_gid,
+ rsd->rsd_fsuid, rsd->rsd_fsgid);
+ RETURN(-EPERM);
+ }
+ } else {
+ if (req->rq_mapped_uid == MDS_IDMAP_NOTFOUND) {
+ CWARN("no mapping found, deny\n");
+ RETURN(-EPERM);
+ }
+
+ if (mds_req_secdesc_do_map(med, rsd))
+ RETURN(-EPERM);
+
+ if (req->rq_mapped_uid != rsd->rsd_uid) {
+ CERROR("remote client "LPU64": auth uid %u "
+ "while client claim %u:%u/%u:%u\n",
+ peernid, req->rq_auth_uid,
+ rsd->rsd_uid, rsd->rsd_gid,
+ rsd->rsd_fsuid, rsd->rsd_fsgid);
+ RETURN(-EPERM);
+ }
+ }
}
- /* now lsd come into play */
+ /* now LSD come into play */
ucred->luc_ginfo = NULL;
ucred->luc_lsd = lsd = mds_get_lsd(rsd->rsd_uid);
- if (lsd) {
- if (req->rq_remote) {
- /* record the gid mapping here */
- mds_idmap_set(med, lsd->lsd_gid, rsd->rsd_gid, 0);
- /* now we act as the authenticated group */
- rsd->rsd_gid = rsd->rsd_fsgid = lsd->lsd_gid;
- } else if (rsd->rsd_gid != lsd->lsd_gid) {
- /* verify gid which client declared is true */
- CWARN("GID: %u while client declare %u, "
- "set %u by force\n",
- lsd->lsd_gid, rsd->rsd_gid,
- lsd->lsd_gid);
- rsd->rsd_gid = lsd->lsd_gid;
- }
+ if (!lsd) {
+ CERROR("Deny access without LSD: uid %d\n", rsd->rsd_uid);
+ RETURN(-EPERM);
+ }
- if (lsd->lsd_ginfo) {
- ucred->luc_ginfo = lsd->lsd_ginfo;
- get_group_info(ucred->luc_ginfo);
- }
+ lsd_perms = mds_lsd_get_perms(lsd, med->med_remote, 0, peernid);
- /* check permission of setuid */
- if (setuid) {
- if (!lsd->lsd_allow_setuid) {
- CWARN("mds blocked setuid attempt: %u -> %u\n",
- rsd->rsd_uid, rsd->rsd_fsuid);
- RETURN(-EPERM);
- }
- }
+ /* check setuid/setgid permissions.
+ * again not apply to special mds user.
+ */
+ if (!req->rq_auth_usr_mds) {
+ /* find out the setuid/setgid attempt */
+ setuid = (rsd->rsd_uid != rsd->rsd_fsuid);
+ setgid = (rsd->rsd_gid != rsd->rsd_fsgid ||
+ rsd->rsd_gid != lsd->lsd_gid);
- /* check permission of setgid */
- if (setgid) {
- if (!lsd->lsd_allow_setgid) {
- CWARN("mds blocked setgid attempt: %u -> %u\n",
- rsd->rsd_gid, rsd->rsd_fsgid);
- RETURN(-EPERM);
- }
- }
- } else {
- /* failed to get lsd, right now we simply deny any access
- * if strong authentication is used,
- */
- if (strong_sec) {
- CWARN("mds deny access without LSD\n");
+ /* check permission of setuid */
+ if (setuid && !(lsd_perms & LSD_PERM_SETUID)) {
+ CWARN("mds blocked setuid attempt (%u -> %u) "
+ "from "LPU64"\n", rsd->rsd_uid, rsd->rsd_fsuid,
+ peernid);
RETURN(-EPERM);
}
- /* and otherwise deny setuid/setgid attempt */
- if (setuid || setgid) {
- CWARN("mds deny setuid/setgid without LSD\n");
+ /* check permission of setgid */
+ if (setgid && !(lsd_perms & LSD_PERM_SETGID)) {
+ CWARN("mds blocked setgid attempt (%u:%u/%u:%u -> %u) "
+ "from "LPU64"\n", rsd->rsd_uid, rsd->rsd_gid,
+ rsd->rsd_fsuid, rsd->rsd_fsgid, lsd->lsd_gid,
+ peernid);
RETURN(-EPERM);
}
}
- /* NOTE: we have already obtained supplementary groups,
- * it will be retained across root_squash. will it be a
- * security problem??
- */
- mds_squash_root(mds, rsd, &peernid);
+ root_squashed = mds_squash_root(mds, rsd, &peernid);
/* remove privilege for non-root user */
if (rsd->rsd_fsuid)
rsd->rsd_cap &= ~CAP_FS_MASK;
- /* by now every fields in rsd have been granted */
+ /* by now every fields other than groups in rsd have been granted */
+ ucred->luc_nid = peernid;
+ ucred->luc_uid = rsd->rsd_uid;
+ ucred->luc_gid = rsd->rsd_gid;
ucred->luc_fsuid = rsd->rsd_fsuid;
ucred->luc_fsgid = rsd->rsd_fsgid;
ucred->luc_cap = rsd->rsd_cap;
- ucred->luc_uid = rsd->rsd_uid;
- /* everything is done if we don't allow setgroups */
- if (!lsd || !lsd->lsd_allow_setgrp)
+ /* don't use any supplementary group if we squashed root.
+ * XXX The exact behavior of root_squash is not defined, we just
+ * keep the reminder here */
+ if (root_squashed)
RETURN(0);
+ /* install groups from LSD */
+ if (lsd->lsd_ginfo) {
+ ucred->luc_ginfo = lsd->lsd_ginfo;
+ get_group_info(ucred->luc_ginfo);
+ }
+
+ /* everything is done if we don't allow setgroups, or it is
+ * from remote client (which implies forced to be no-setgroups).
+ *
+ * Note: remote user's supplementary groups sent along the request
+ * (if any) are all ignored, but we make the mapped local user's
+ * supplementary groups take effect.
+ */
+ if (med->med_remote || !(lsd_perms & LSD_PERM_SETGRP))
+ RETURN(0);
+
+ /* root could set any groups as he want (if allowed), normal
+ * users only could reduce his group array.
+ */
if (ucred->luc_uid == 0) {
- if (rsd->rsd_ngroups == 0) {
- drop_ucred_ginfo(ucred);
+ drop_ucred_ginfo(ucred);
+
+ if (rsd->rsd_ngroups == 0)
RETURN(0);
- }
gnew = groups_alloc(rsd->rsd_ngroups);
if (!gnew) {
CERROR("out of memory\n");
- drop_ucred_ginfo(ucred);
drop_ucred_lsd(ucred);
RETURN(-ENOMEM);
}
groups_from_buffer(gnew, rsd->rsd_groups);
- groups_sort(gnew); /* can't rely on client */
+ groups_sort(gnew); /* don't rely on client doing this */
- drop_ucred_ginfo(ucred);
ucred->luc_ginfo = gnew;
} else {
__u32 set = 0, cur = 0;
- struct group_info *ginfo;
+ struct group_info *ginfo = ucred->luc_ginfo;
- /* if no group info in hash, we don't
- * bother createing new
- */
- if (!ucred->luc_ginfo)
+ if (!ginfo)
RETURN(0);
/* Note: freeing a group_info count on 'nblocks' instead of
RETURN(-ENOMEM);
}
- ginfo = ucred->luc_ginfo;
while (cur < rsd->rsd_ngroups) {
if (groups_search(ginfo, rsd->rsd_groups[cur])) {
GROUP_AT(gnew, set) = rsd->rsd_groups[cur];
drop_ucred_lsd(ucred);
EXIT;
}
+
git://git.whamcloud.com / fs / lustre-release.git / blobdiff