1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Copyright (c) 2003 Cluster File Systems, Inc.
6 * This file is part of Lustre, http://www.lustre.org.
8 * Lustre is free software; you can redistribute it and/or
9 * modify it under the terms of version 2 of the GNU General Public
10 * License as published by the Free Software Foundation.
12 * Lustre is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
17 * You should have received a copy of the GNU General Public License
18 * along with Lustre; if not, write to the Free Software
19 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
22 #define DEBUG_SUBSYSTEM S_MDS
24 #include <linux/config.h>
25 #include <linux/module.h>
26 #include <linux/kernel.h>
28 #include <linux/string.h>
29 #include <linux/stat.h>
30 #include <linux/errno.h>
31 #include <linux/version.h>
32 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
33 # include <linux/locks.h> // for wait_on_buffer
35 # include <linux/buffer_head.h> // for wait_on_buffer
37 #include <linux/unistd.h>
39 #include <asm/system.h>
40 #include <asm/uaccess.h>
43 #include <linux/stat.h>
44 #include <asm/uaccess.h>
45 #include <linux/slab.h>
46 #include <asm/segment.h>
47 #include <linux/random.h>
49 #include <linux/obd_support.h>
50 #include <linux/lustre_lib.h>
51 #include <linux/lustre_sec.h>
52 #include <linux/lustre_ucache.h>
53 #include <linux/lustre_gs.h>
54 #include <linux/lustre_fsfilt.h>
55 #include "mds_internal.h"
57 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,4)
58 struct group_info *groups_alloc(int ngroups)
60 struct group_info *ginfo;
62 LASSERT(ngroups <= NGROUPS_SMALL);
64 OBD_ALLOC(ginfo, sizeof(*ginfo) + 1 * sizeof(gid_t *));
67 ginfo->ngroups = ngroups;
69 ginfo->blocks[0] = ginfo->small_block;
70 atomic_set(&ginfo->usage, 1);
75 void groups_free(struct group_info *ginfo)
77 LASSERT(ginfo->ngroups <= NGROUPS_SMALL);
78 LASSERT(ginfo->nblocks == 1);
79 LASSERT(ginfo->blocks[0] == ginfo->small_block);
81 OBD_FREE(ginfo, sizeof(*ginfo) + 1 * sizeof(gid_t *));
84 /* for 2.4 the group number is small, so simply search the
87 int groups_search(struct group_info *ginfo, gid_t grp)
94 for (i = 0; i < ginfo->ngroups; i++)
95 if (GROUP_AT(ginfo, i) == grp)
102 void groups_sort(struct group_info *ginfo)
104 int base, max, stride;
105 int gidsetsize = ginfo->ngroups;
107 for (stride = 1; stride < gidsetsize; stride = 3 * stride + 1)
112 max = gidsetsize - stride;
113 for (base = 0; base < max; base++) {
115 int right = left + stride;
116 gid_t tmp = GROUP_AT(ginfo, right);
118 while (left >= 0 && GROUP_AT(ginfo, left) > tmp) {
119 GROUP_AT(ginfo, right) =
120 GROUP_AT(ginfo, left);
124 GROUP_AT(ginfo, right) = tmp;
130 int groups_search(struct group_info *ginfo, gid_t grp)
138 right = ginfo->ngroups;
139 while (left < right) {
140 int mid = (left + right) / 2;
141 int cmp = grp - GROUP_AT(ginfo, mid);
153 void groups_from_buffer(struct group_info *ginfo, __u32 *gids)
155 int i, ngroups = ginfo->ngroups;
157 for (i = 0; i < ginfo->nblocks; i++) {
158 int count = min(NGROUPS_PER_BLOCK, ngroups);
160 memcpy(ginfo->blocks[i], gids, count * sizeof(__u32));
161 gids += NGROUPS_PER_BLOCK;
166 void mds_pack_dentry2id(struct obd_device *obd,
167 struct lustre_id *id,
168 struct dentry *dentry,
171 id_ino(id) = dentry->d_inum;
172 id_gen(id) = dentry->d_generation;
175 id_fid(id) = dentry->d_fid;
176 id_group(id) = dentry->d_mdsnum;
180 void mds_pack_dentry2body(struct obd_device *obd,
182 struct dentry *dentry,
185 b->valid |= OBD_MD_FLID | OBD_MD_FLGENER |
189 b->valid |= OBD_MD_FID;
191 mds_pack_dentry2id(obd, &b->id1, dentry, fid);
194 int mds_pack_inode2id(struct obd_device *obd,
195 struct lustre_id *id,
203 /* we have to avoid deadlock. */
204 if (!down_trylock(&inode->i_sem)) {
205 rc = mds_read_inode_sid(obd, inode, id);
208 rc = mds_read_inode_sid(obd, inode, id);
213 id_ino(id) = inode->i_ino;
214 id_gen(id) = inode->i_generation;
215 id_type(id) = (S_IFMT & inode->i_mode);
220 int mds_pack_gskey(struct obd_device *obd, struct lustre_msg *repmsg,
221 int *offset, struct mds_body *body, struct inode *inode)
223 struct mds_obd *mds = &obd->u.mds;
224 struct crypto_key_md *md_key;
225 struct crypto_key *ckey;
226 __u32 buflen, *sizep;
231 if ((mds->mds_crypto_type != MKS_TYPE &&
232 mds->mds_crypto_type != GKS_TYPE))
235 sizep = lustre_msg_buf(repmsg, (*offset)++, 4);
237 CERROR("can't locate returned ckey size buf\n");
240 *sizep = cpu_to_le32(sizeof(*ckey));
242 OBD_ALLOC(md_key, sizeof(*md_key));
244 buflen = repmsg->buflens[*offset];
245 buf = lustre_msg_buf(repmsg, (*offset)++, buflen);
247 size = fsfilt_get_md(obd, inode, md_key, sizeof(*md_key),
250 CERROR("Can not get gskey from MDS ino %lu rc %d\n",
252 GOTO(out, rc = size);
254 if (le32_to_cpu(md_key->md_magic) != MD_KEY_MAGIC) {
255 CDEBUG(D_INFO, "given match %x != magic %x\n",
256 md_key->md_magic, MD_KEY_MAGIC);
260 CDEBUG(D_INFO, "get key %s mac %s for ino %lu size %d \n",
261 md_key->md_ck.ck_key, md_key->md_ck.ck_mac, inode->i_ino, size);
262 ckey=(struct crypto_key*)buf;
264 memcpy(ckey, &md_key->md_ck, sizeof(*ckey));
265 body->valid |= OBD_MD_FLKEY;
267 OBD_FREE(md_key, sizeof(*md_key));
271 static int mds_get_gskey(struct inode *inode, struct crypto_key *ckey)
274 /*tmp create gs key here*/
275 get_random_bytes(ckey->ck_key, KEY_SIZE);
276 ckey->ck_type = MKS_TYPE;
280 int mds_set_gskey(struct obd_device *obd, void *handle,
281 struct inode *inode, void *key, int key_len,
284 struct crypto_key_md *md_key = NULL;
285 struct crypto_key *ckey = (struct crypto_key *)key;
286 struct mds_obd *mds = &obd->u.mds;
290 if ((mds->mds_crypto_type != MKS_TYPE &&
291 mds->mds_crypto_type != GKS_TYPE) || key_len == 0)
294 OBD_ALLOC(md_key, sizeof(*md_key));
295 LASSERT(ckey != NULL);
296 if (mds->mds_crypto_type == MKS_TYPE) {
297 mds_get_gskey(inode, ckey);
299 rc = fsfilt_get_md(obd, inode, md_key, sizeof(*md_key),
303 LASSERT(le32_to_cpu(md_key->md_magic) == MD_KEY_MAGIC ||
304 md_key->md_magic == 0);
305 if (le32_to_cpu(md_key->md_magic) == MD_KEY_MAGIC) {
306 CDEBUG(D_INFO, "reset key %s mac %s", md_key->md_ck.ck_mac,
307 md_key->md_ck.ck_key);
310 md_key->md_magic = cpu_to_le32(MD_KEY_MAGIC);
311 if (valid & ATTR_MAC) {
312 memcpy(md_key->md_ck.ck_mac, ckey->ck_mac, MAC_SIZE);
313 CDEBUG(D_INFO, "set mac %s for ino %lu \n",
314 md_key->md_ck.ck_mac, inode->i_ino);
316 if (valid & ATTR_KEY) {
317 memcpy(md_key->md_ck.ck_key, ckey->ck_key, KEY_SIZE);
318 CDEBUG(D_INFO, "set key %s for ino %lu \n",
319 md_key->md_ck.ck_key, inode->i_ino);
321 rc = fsfilt_set_md(obd, inode, handle, md_key,
322 sizeof(*md_key), EA_KEY);
325 OBD_FREE(md_key, sizeof(*md_key));
329 int mds_set_crypto_type(struct obd_device *obd, void *val, __u32 vallen)
331 struct mds_obd *mds = &obd->u.mds;
334 if (vallen >= strlen("mks") &&
335 memcmp(val, "mks", vallen) == 0) {
336 mds->mds_crypto_type = MKS_TYPE;
338 if (vallen >= strlen("gks") &&
339 memcmp(val, "gks", vallen) == 0) {
340 mds->mds_crypto_type = GKS_TYPE;
343 CDEBUG(D_IOCTL, "invalid key\n");
346 /* Note that we can copy all of the fields, just some will not be "valid" */
347 void mds_pack_inode2body(struct obd_device *obd, struct mds_body *b,
348 struct inode *inode, int fid)
350 b->valid |= OBD_MD_FLID | OBD_MD_FLCTIME | OBD_MD_FLUID |
351 OBD_MD_FLGID | OBD_MD_FLFLAGS | OBD_MD_FLTYPE |
352 OBD_MD_FLMODE | OBD_MD_FLNLINK | OBD_MD_FLGENER |
353 OBD_MD_FLATIME | OBD_MD_FLMTIME; /* bug 2020 */
355 if (!S_ISREG(inode->i_mode)) {
356 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS |
357 OBD_MD_FLATIME | OBD_MD_FLMTIME |
360 b->atime = LTIME_S(inode->i_atime);
361 b->mtime = LTIME_S(inode->i_mtime);
362 b->ctime = LTIME_S(inode->i_ctime);
363 b->mode = inode->i_mode;
364 b->size = inode->i_size;
365 b->blocks = inode->i_blocks;
366 b->uid = inode->i_uid;
367 b->gid = inode->i_gid;
368 b->flags = inode->i_flags;
369 b->rdev = inode->i_rdev;
371 /* Return the correct link count for orphan inodes */
372 if (mds_inode_is_orphan(inode)) {
374 } else if (S_ISDIR(inode->i_mode)) {
377 b->nlink = inode->i_nlink;
381 b->valid |= OBD_MD_FID;
383 mds_pack_inode2id(obd, &b->id1, inode, fid);
388 static int mds_setattr_unpack(struct ptlrpc_request *req, int offset,
389 struct mds_update_record *r)
391 struct iattr *attr = &r->ur_iattr;
392 struct mds_rec_setattr *rec;
395 rec = lustre_swab_reqbuf(req, offset, sizeof(*rec),
396 lustre_swab_mds_rec_setattr);
400 r->ur_id1 = &rec->sa_id;
401 attr->ia_valid = rec->sa_valid;
402 attr->ia_mode = rec->sa_mode;
403 attr->ia_uid = rec->sa_uid;
404 attr->ia_gid = rec->sa_gid;
405 attr->ia_size = rec->sa_size;
406 LTIME_S(attr->ia_atime) = rec->sa_atime;
407 LTIME_S(attr->ia_mtime) = rec->sa_mtime;
408 LTIME_S(attr->ia_ctime) = rec->sa_ctime;
409 attr->ia_attr_flags = rec->sa_attr_flags;
411 LASSERT_REQSWAB (req, offset + 1);
412 if (req->rq_reqmsg->bufcount > offset + 1) {
413 r->ur_eadata = lustre_msg_buf (req->rq_reqmsg,
415 if (r->ur_eadata == NULL)
417 r->ur_eadatalen = req->rq_reqmsg->buflens[offset + 1];
420 if (req->rq_reqmsg->bufcount > offset + 2) {
421 r->ur_ea2data = lustre_msg_buf(req->rq_reqmsg, offset + 2, 0);
422 if (r->ur_ea2data == NULL)
425 r->ur_ea2datalen = req->rq_reqmsg->buflens[offset + 2];
428 if (req->rq_reqmsg->bufcount > offset + 3) {
429 r->ur_ea3data = lustre_msg_buf(req->rq_reqmsg, offset + 3, 0);
430 if (r->ur_ea3data == NULL)
433 r->ur_ea3datalen = req->rq_reqmsg->buflens[offset + 3];
439 static int mds_create_unpack(struct ptlrpc_request *req, int offset,
440 struct mds_update_record *r)
442 struct mds_rec_create *rec;
445 rec = lustre_swab_reqbuf (req, offset, sizeof (*rec),
446 lustre_swab_mds_rec_create);
450 r->ur_id1 = &rec->cr_id;
451 r->ur_id2 = &rec->cr_replayid;
452 r->ur_mode = rec->cr_mode;
453 r->ur_rdev = rec->cr_rdev;
454 r->ur_time = rec->cr_time;
455 r->ur_flags = rec->cr_flags;
457 LASSERT_REQSWAB (req, offset + 1);
458 r->ur_name = lustre_msg_string (req->rq_reqmsg, offset + 1, 0);
459 if (r->ur_name == NULL)
461 r->ur_namelen = req->rq_reqmsg->buflens[offset + 1];
463 LASSERT_REQSWAB (req, offset + 2);
464 if (req->rq_reqmsg->bufcount > offset + 2) {
465 if (S_ISLNK(r->ur_mode)) {
466 r->ur_tgt = lustre_msg_string(req->rq_reqmsg,
468 if (r->ur_tgt == NULL)
470 r->ur_tgtlen = req->rq_reqmsg->buflens[offset + 2];
471 } else if (S_ISDIR(r->ur_mode) ) {
472 /* Stripe info for mkdir - just a 16bit integer */
473 if (req->rq_reqmsg->buflens[offset + 2] != 2) {
474 CERROR("mkdir stripe info does not match "
475 "expected size %d vs 2\n",
476 req->rq_reqmsg->buflens[offset + 2]);
479 r->ur_eadata = lustre_swab_buf (req->rq_reqmsg,
480 offset + 2, 2, __swab16s);
481 r->ur_eadatalen = req->rq_reqmsg->buflens[offset + 2];
482 } else if (S_ISREG(r->ur_mode)){
483 r->ur_eadata = lustre_msg_buf (req->rq_reqmsg,
485 r->ur_eadatalen = req->rq_reqmsg->buflens[offset + 2];
487 /* Hm, no other users so far? */
494 static int mds_link_unpack(struct ptlrpc_request *req, int offset,
495 struct mds_update_record *r)
497 struct mds_rec_link *rec;
500 rec = lustre_swab_reqbuf (req, offset, sizeof (*rec),
501 lustre_swab_mds_rec_link);
505 r->ur_id1 = &rec->lk_id1;
506 r->ur_id2 = &rec->lk_id2;
507 r->ur_time = rec->lk_time;
509 LASSERT_REQSWAB (req, offset + 1);
510 r->ur_name = lustre_msg_string (req->rq_reqmsg, offset + 1, 0);
511 if (r->ur_name == NULL)
513 r->ur_namelen = req->rq_reqmsg->buflens[offset + 1];
517 static int mds_unlink_unpack(struct ptlrpc_request *req, int offset,
518 struct mds_update_record *r)
520 struct mds_rec_unlink *rec;
523 rec = lustre_swab_reqbuf (req, offset, sizeof (*rec),
524 lustre_swab_mds_rec_unlink);
528 r->ur_mode = rec->ul_mode;
529 r->ur_id1 = &rec->ul_id1;
530 r->ur_id2 = &rec->ul_id2;
531 r->ur_time = rec->ul_time;
533 LASSERT_REQSWAB (req, offset + 1);
534 r->ur_name = lustre_msg_string(req->rq_reqmsg, offset + 1, 0);
535 if (r->ur_name == NULL)
537 r->ur_namelen = req->rq_reqmsg->buflens[offset + 1];
541 static int mds_rename_unpack(struct ptlrpc_request *req, int offset,
542 struct mds_update_record *r)
544 struct mds_rec_rename *rec;
547 rec = lustre_swab_reqbuf (req, offset, sizeof (*rec),
548 lustre_swab_mds_rec_rename);
552 r->ur_id1 = &rec->rn_id1;
553 r->ur_id2 = &rec->rn_id2;
554 r->ur_time = rec->rn_time;
556 LASSERT_REQSWAB (req, offset + 1);
557 r->ur_name = lustre_msg_string(req->rq_reqmsg, offset + 1, 0);
558 if (r->ur_name == NULL)
560 r->ur_namelen = req->rq_reqmsg->buflens[offset + 1];
562 LASSERT_REQSWAB (req, offset + 2);
563 r->ur_tgt = lustre_msg_string(req->rq_reqmsg, offset + 2, 0);
564 if (r->ur_tgt == NULL)
566 r->ur_tgtlen = req->rq_reqmsg->buflens[offset + 2];
570 static int mds_open_unpack(struct ptlrpc_request *req, int offset,
571 struct mds_update_record *r)
573 struct mds_rec_create *rec;
576 rec = lustre_swab_reqbuf (req, offset, sizeof (*rec),
577 lustre_swab_mds_rec_create);
581 r->ur_id1 = &rec->cr_id;
582 r->ur_id2 = &rec->cr_replayid;
583 r->ur_mode = rec->cr_mode;
584 r->ur_rdev = rec->cr_rdev;
585 r->ur_time = rec->cr_time;
586 r->ur_flags = rec->cr_flags;
588 LASSERT_REQSWAB (req, offset + 1);
589 r->ur_name = lustre_msg_string (req->rq_reqmsg, offset + 1, 0);
590 if (r->ur_name == NULL)
592 r->ur_namelen = req->rq_reqmsg->buflens[offset + 1];
594 LASSERT_REQSWAB (req, offset + 2);
596 if (req->rq_reqmsg->bufcount > offset + 2) {
597 r->ur_eadata = lustre_msg_buf(req->rq_reqmsg, offset + 2, 0);
598 if (r->ur_eadata == NULL)
600 r->ur_eadatalen = req->rq_reqmsg->buflens[offset + 2];
603 if (rec->cr_flags & MDS_OPEN_HAS_KEY) {
604 LASSERT(req->rq_reqmsg->bufcount > offset + 3);
605 r->ur_ea2data = lustre_msg_buf(req->rq_reqmsg, offset + 3, 0);
606 r->ur_ea2datalen = req->rq_reqmsg->buflens[offset + 3];
611 typedef int (*update_unpacker)(struct ptlrpc_request *req, int offset,
612 struct mds_update_record *r);
614 static update_unpacker mds_unpackers[REINT_MAX + 1] = {
615 [REINT_SETATTR] mds_setattr_unpack,
616 [REINT_CREATE] mds_create_unpack,
617 [REINT_LINK] mds_link_unpack,
618 [REINT_UNLINK] mds_unlink_unpack,
619 [REINT_RENAME] mds_rename_unpack,
620 [REINT_OPEN] mds_open_unpack,
623 int mds_update_unpack(struct ptlrpc_request *req, int offset,
624 struct mds_update_record *rec)
632 * NB don't lustre_swab_reqbuf() here. We're just taking a peek and we
633 * want to leave it to the specific unpacker once we've identified the
636 opcodep = lustre_msg_buf (req->rq_reqmsg, offset, sizeof(*opcodep));
641 if (lustre_msg_swabbed (req->rq_reqmsg))
644 if (opcode > REINT_MAX ||
645 mds_unpackers[opcode] == NULL) {
646 CERROR ("Unexpected opcode %d\n", opcode);
652 rec->ur_opcode = opcode;
654 rc = mds_unpackers[opcode](req, offset, rec);
657 rec->ur_fsuid = req->rq_uid;
663 * here we take simple rule: once uid/fsuid is root, we also squash
664 * the gid/fsgid, don't care setuid/setgid attributes.
667 int mds_squash_root(struct mds_obd *mds, struct mds_req_sec_desc *rsd,
670 if (!mds->mds_squash_uid || *peernid == mds->mds_nosquash_nid)
673 if (rsd->rsd_uid && rsd->rsd_fsuid)
676 CDEBUG(D_SEC, "squash req from "LPX64":"
677 "(%u:%u-%u:%u/%x)=>(%u:%u-%u:%u/%x)\n", *peernid,
678 rsd->rsd_uid, rsd->rsd_gid,
679 rsd->rsd_fsuid, rsd->rsd_fsgid, rsd->rsd_cap,
680 rsd->rsd_uid ? rsd->rsd_uid : mds->mds_squash_uid,
681 rsd->rsd_uid ? rsd->rsd_gid : mds->mds_squash_gid,
682 rsd->rsd_fsuid ? rsd->rsd_fsuid : mds->mds_squash_uid,
683 rsd->rsd_fsuid ? rsd->rsd_fsgid : mds->mds_squash_gid,
684 rsd->rsd_cap & ~CAP_FS_MASK);
686 if (rsd->rsd_uid == 0) {
687 rsd->rsd_uid = mds->mds_squash_uid;
688 rsd->rsd_gid = mds->mds_squash_gid;
690 if (rsd->rsd_fsuid == 0) {
691 rsd->rsd_fsuid = mds->mds_squash_uid;
692 rsd->rsd_fsgid = mds->mds_squash_gid;
694 rsd->rsd_cap &= ~CAP_FS_MASK;
699 /********************************
700 * MDS uid/gid mapping handling *
701 ********************************/
704 struct mds_idmap_entry* idmap_alloc_entry(__u32 rmt_id, __u32 lcl_id)
706 struct mds_idmap_entry *e;
708 OBD_ALLOC(e, sizeof(*e));
712 INIT_LIST_HEAD(&e->rmt_hash);
713 INIT_LIST_HEAD(&e->lcl_hash);
714 atomic_set(&e->refcount, 1);
721 void idmap_free_entry(struct mds_idmap_entry *e)
723 if (!list_empty(&e->rmt_hash))
724 list_del(&e->rmt_hash);
725 if (!list_empty(&e->lcl_hash))
726 list_del(&e->lcl_hash);
727 OBD_FREE(e, sizeof(*e));
731 int idmap_insert_entry(struct list_head *rmt_hash, struct list_head *lcl_hash,
732 struct mds_idmap_entry *new, const char *warn_msg)
734 struct list_head *rmt_head = &rmt_hash[MDS_IDMAP_HASHFUNC(new->rmt_id)];
735 struct list_head *lcl_head = &lcl_hash[MDS_IDMAP_HASHFUNC(new->lcl_id)];
736 struct mds_idmap_entry *e;
738 list_for_each_entry(e, rmt_head, rmt_hash) {
739 if (e->rmt_id == new->rmt_id &&
740 e->lcl_id == new->lcl_id) {
741 atomic_inc(&e->refcount);
744 if (e->rmt_id == new->rmt_id && warn_msg)
745 CWARN("%s: rmt id %u already map to %u (new %u)\n",
746 warn_msg, e->rmt_id, e->lcl_id, new->lcl_id);
747 if (e->lcl_id == new->lcl_id && warn_msg)
748 CWARN("%s: lcl id %u already be mapped from %u "
749 "(new %u)\n", warn_msg,
750 e->lcl_id, e->rmt_id, new->rmt_id);
753 list_add_tail(rmt_head, &new->rmt_hash);
754 list_add_tail(lcl_head, &new->lcl_hash);
759 int idmap_remove_entry(struct list_head *rmt_hash, struct list_head *lcl_hash,
760 __u32 rmt_id, __u32 lcl_id)
762 struct list_head *rmt_head = &rmt_hash[MDS_IDMAP_HASHFUNC(rmt_id)];
763 struct mds_idmap_entry *e;
765 list_for_each_entry(e, rmt_head, rmt_hash) {
766 if (e->rmt_id == rmt_id && e->lcl_id == lcl_id) {
767 if (atomic_dec_and_test(&e->refcount)) {
768 list_del(&e->rmt_hash);
769 list_del(&e->lcl_hash);
770 OBD_FREE(e, sizeof(*e));
779 int mds_idmap_add(struct mds_idmap_table *tbl,
780 uid_t rmt_uid, uid_t lcl_uid,
781 gid_t rmt_gid, gid_t lcl_gid)
783 struct mds_idmap_entry *ue, *ge;
789 ue = idmap_alloc_entry(rmt_uid, lcl_uid);
792 ge = idmap_alloc_entry(rmt_gid, lcl_gid);
794 idmap_free_entry(ue);
798 spin_lock(&tbl->mit_lock);
800 if (idmap_insert_entry(tbl->mit_idmaps[MDS_RMT_UIDMAP_IDX],
801 tbl->mit_idmaps[MDS_LCL_UIDMAP_IDX],
802 ue, "UID mapping")) {
803 idmap_free_entry(ue);
806 if (idmap_insert_entry(tbl->mit_idmaps[MDS_RMT_GIDMAP_IDX],
807 tbl->mit_idmaps[MDS_LCL_GIDMAP_IDX],
808 ge, "GID mapping")) {
809 idmap_free_entry(ge);
812 spin_unlock(&tbl->mit_lock);
816 int mds_idmap_del(struct mds_idmap_table *tbl,
817 uid_t rmt_uid, uid_t lcl_uid,
818 gid_t rmt_gid, gid_t lcl_gid)
825 spin_lock(&tbl->mit_lock);
826 idmap_remove_entry(tbl->mit_idmaps[MDS_RMT_UIDMAP_IDX],
827 tbl->mit_idmaps[MDS_LCL_UIDMAP_IDX],
829 idmap_remove_entry(tbl->mit_idmaps[MDS_RMT_GIDMAP_IDX],
830 tbl->mit_idmaps[MDS_LCL_GIDMAP_IDX],
832 spin_unlock(&tbl->mit_lock);
837 __u32 idmap_lookup_id(struct list_head *hash, int reverse, __u32 id)
839 struct list_head *head = &hash[MDS_IDMAP_HASHFUNC(id)];
840 struct mds_idmap_entry *e;
843 list_for_each_entry(e, head, rmt_hash) {
847 return MDS_IDMAP_NOTFOUND;
849 list_for_each_entry(e, head, lcl_hash) {
853 return MDS_IDMAP_NOTFOUND;
857 int mds_idmap_lookup_uid(struct mds_idmap_table *tbl, int reverse, uid_t uid)
859 struct list_head *hash;
862 return MDS_IDMAP_NOTFOUND;
865 hash = tbl->mit_idmaps[MDS_RMT_UIDMAP_IDX];
867 hash = tbl->mit_idmaps[MDS_LCL_UIDMAP_IDX];
869 spin_lock(&tbl->mit_lock);
870 uid = idmap_lookup_id(hash, reverse, uid);
871 spin_unlock(&tbl->mit_lock);
876 int mds_idmap_lookup_gid(struct mds_idmap_table *tbl, int reverse, gid_t gid)
878 struct list_head *hash;
881 return MDS_IDMAP_NOTFOUND;
884 hash = tbl->mit_idmaps[MDS_RMT_GIDMAP_IDX];
886 hash = tbl->mit_idmaps[MDS_LCL_GIDMAP_IDX];
888 spin_lock(&tbl->mit_lock);
889 gid = idmap_lookup_id(hash, reverse, gid);
890 spin_unlock(&tbl->mit_lock);
895 struct mds_idmap_table *mds_idmap_alloc()
897 struct mds_idmap_table *tbl;
900 OBD_ALLOC(tbl, sizeof(*tbl));
904 spin_lock_init(&tbl->mit_lock);
905 for (i = 0; i < MDS_IDMAP_N_HASHES; i++)
906 for (j = 0; j < MDS_IDMAP_HASHSIZE; j++)
907 INIT_LIST_HEAD(&tbl->mit_idmaps[i][j]);
912 static void idmap_clear_rmt_hash(struct list_head *list)
914 struct mds_idmap_entry *e;
917 for (i = 0; i < MDS_IDMAP_HASHSIZE; i++) {
918 while (!list_empty(&list[i])) {
919 e = list_entry(list[i].next, struct mds_idmap_entry,
926 void mds_idmap_free(struct mds_idmap_table *tbl)
930 spin_lock(&tbl->mit_lock);
931 idmap_clear_rmt_hash(tbl->mit_idmaps[MDS_RMT_UIDMAP_IDX]);
932 idmap_clear_rmt_hash(tbl->mit_idmaps[MDS_RMT_GIDMAP_IDX]);
934 /* paranoid checking */
935 for (i = 0; i < MDS_IDMAP_HASHSIZE; i++) {
936 LASSERT(list_empty(&tbl->mit_idmaps[MDS_LCL_UIDMAP_IDX][i]));
937 LASSERT(list_empty(&tbl->mit_idmaps[MDS_LCL_GIDMAP_IDX][i]));
939 spin_unlock(&tbl->mit_lock);
941 OBD_FREE(tbl, sizeof(*tbl));
944 /*********************************
945 * helpers doing mapping for MDS *
946 *********************************/
949 * we allow remote setuid/setgid to an "authencated" one,
950 * this policy probably change later.
953 int mds_req_secdesc_do_map(struct mds_export_data *med,
954 struct mds_req_sec_desc *rsd)
956 struct mds_idmap_table *idmap = med->med_idmap;
960 uid = mds_idmap_lookup_uid(idmap, 0, rsd->rsd_uid);
961 if (uid == MDS_IDMAP_NOTFOUND) {
962 CERROR("can't find map for uid %u\n", rsd->rsd_uid);
966 if (rsd->rsd_uid == rsd->rsd_fsuid)
969 fsuid = mds_idmap_lookup_uid(idmap, 0, rsd->rsd_fsuid);
970 if (fsuid == MDS_IDMAP_NOTFOUND) {
971 CERROR("can't find map for fsuid %u\n", rsd->rsd_fsuid);
976 gid = mds_idmap_lookup_gid(idmap, 0, rsd->rsd_gid);
977 if (gid == MDS_IDMAP_NOTFOUND) {
978 CERROR("can't find map for gid %u\n", rsd->rsd_gid);
982 if (rsd->rsd_gid == rsd->rsd_fsgid)
985 fsgid = mds_idmap_lookup_gid(idmap, 0, rsd->rsd_fsgid);
986 if (fsgid == MDS_IDMAP_NOTFOUND) {
987 CERROR("can't find map for fsgid %u\n", rsd->rsd_fsgid);
994 rsd->rsd_fsuid = fsuid;
995 rsd->rsd_fsgid = fsgid;
1000 void mds_body_do_reverse_map(struct mds_export_data *med,
1001 struct mds_body *body)
1006 if (!med->med_remote)
1010 if (body->valid & OBD_MD_FLUID) {
1011 uid = mds_idmap_lookup_uid(med->med_idmap, 1, body->uid);
1012 if (uid == MDS_IDMAP_NOTFOUND) {
1013 uid = med->med_nllu;
1014 if (body->valid & OBD_MD_FLMODE) {
1015 body->mode = (body->mode & ~S_IRWXU) |
1016 ((body->mode & S_IRWXO) << 6);
1021 if (body->valid & OBD_MD_FLGID) {
1022 gid = mds_idmap_lookup_gid(med->med_idmap, 1, body->gid);
1023 if (gid == MDS_IDMAP_NOTFOUND) {
1024 gid = med->med_nllg;
1025 if (body->valid & OBD_MD_FLMODE) {
1026 body->mode = (body->mode & ~S_IRWXG) |
1027 ((body->mode & S_IRWXO) << 3);
1036 /**********************
1037 * MDS ucred handling *
1038 **********************/
1040 static inline void drop_ucred_ginfo(struct lvfs_ucred *ucred)
1042 if (ucred->luc_ginfo) {
1043 put_group_info(ucred->luc_ginfo);
1044 ucred->luc_ginfo = NULL;
1048 static inline void drop_ucred_lsd(struct lvfs_ucred *ucred)
1050 if (ucred->luc_lsd) {
1051 mds_put_lsd(ucred->luc_lsd);
1052 ucred->luc_lsd = NULL;
1057 * the heart of the uid/gid handling and security checking.
1059 * root could set any group_info if we allowed setgroups, while
1060 * normal user only could 'reduce' their group members -- which
1061 * is somewhat expensive.
1063 * authenticated as mds user (using mds service credential) could
1064 * bypass all checkings.
1066 int mds_init_ucred(struct lvfs_ucred *ucred,
1067 struct ptlrpc_request *req,
1068 struct mds_req_sec_desc *rsd)
1070 struct mds_obd *mds = &req->rq_export->exp_obd->u.mds;
1071 struct mds_export_data *med = &req->rq_export->u.eu_mds_data;
1072 struct lustre_sec_desc *lsd;
1073 ptl_nid_t peernid = req->rq_peer.peer_id.nid;
1074 struct group_info *gnew;
1075 unsigned int setuid, setgid, strong_sec, root_squashed;
1081 LASSERT(rsd->rsd_ngroups <= LUSTRE_MAX_GROUPS);
1083 if (SEC_FLAVOR_MAJOR(req->rq_req_secflvr) == PTLRPCS_FLVR_MAJOR_GSS &&
1084 (SEC_FLAVOR_SVC(req->rq_req_secflvr) == PTLRPCS_SVC_AUTH ||
1085 SEC_FLAVOR_SVC(req->rq_req_secflvr) == PTLRPCS_SVC_PRIV))
1090 LASSERT(!(req->rq_remote_realm && !strong_sec));
1092 if (strong_sec && req->rq_auth_uid == -1) {
1093 CWARN("user not authenticated, deny access\n");
1097 /* sanity check: if we use strong authentication, we expect the
1098 * uid which client claimed is true.
1099 * not apply to special mds user .
1101 if (!req->rq_auth_usr_mds && strong_sec) {
1102 if (!med->med_remote) {
1103 if (req->rq_auth_uid != rsd->rsd_uid) {
1104 CERROR("local client "LPU64": auth uid %u "
1105 "while client claim %u:%u/%u:%u\n",
1106 peernid, req->rq_auth_uid,
1107 rsd->rsd_uid, rsd->rsd_gid,
1108 rsd->rsd_fsuid, rsd->rsd_fsgid);
1112 if (req->rq_mapped_uid == MDS_IDMAP_NOTFOUND) {
1113 CWARN("no mapping found, deny\n");
1117 if (mds_req_secdesc_do_map(med, rsd))
1120 if (req->rq_mapped_uid != rsd->rsd_uid) {
1121 CERROR("remote client "LPU64": auth uid %u "
1122 "while client claim %u:%u/%u:%u\n",
1123 peernid, req->rq_auth_uid,
1124 rsd->rsd_uid, rsd->rsd_gid,
1125 rsd->rsd_fsuid, rsd->rsd_fsgid);
1131 /* now LSD come into play */
1132 ucred->luc_ginfo = NULL;
1133 ucred->luc_lsd = lsd = mds_get_lsd(rsd->rsd_uid);
1136 CERROR("Deny access without LSD: uid %d\n", rsd->rsd_uid);
1140 lsd_perms = mds_lsd_get_perms(lsd, med->med_remote, 0, peernid);
1142 /* check setuid/setgid permissions.
1143 * again not apply to special mds user.
1145 if (!req->rq_auth_usr_mds) {
1146 /* find out the setuid/setgid attempt */
1147 setuid = (rsd->rsd_uid != rsd->rsd_fsuid);
1148 setgid = (rsd->rsd_gid != rsd->rsd_fsgid ||
1149 rsd->rsd_gid != lsd->lsd_gid);
1151 /* check permission of setuid */
1152 if (setuid && !(lsd_perms & LSD_PERM_SETUID)) {
1153 CWARN("mds blocked setuid attempt (%u -> %u) "
1154 "from "LPU64"\n", rsd->rsd_uid, rsd->rsd_fsuid,
1159 /* check permission of setgid */
1160 if (setgid && !(lsd_perms & LSD_PERM_SETGID)) {
1161 CWARN("mds blocked setgid attempt (%u:%u/%u:%u -> %u) "
1162 "from "LPU64"\n", rsd->rsd_uid, rsd->rsd_gid,
1163 rsd->rsd_fsuid, rsd->rsd_fsgid, lsd->lsd_gid,
1169 root_squashed = mds_squash_root(mds, rsd, &peernid);
1171 /* remove privilege for non-root user */
1173 rsd->rsd_cap &= ~CAP_FS_MASK;
1175 /* by now every fields other than groups in rsd have been granted */
1176 ucred->luc_uid = rsd->rsd_uid;
1177 ucred->luc_gid = rsd->rsd_gid;
1178 ucred->luc_fsuid = rsd->rsd_fsuid;
1179 ucred->luc_fsgid = rsd->rsd_fsgid;
1180 ucred->luc_cap = rsd->rsd_cap;
1182 /* don't use any supplementary group if we squashed root.
1183 * XXX The exact behavior of root_squash is not defined, we just
1184 * keep the reminder here */
1188 /* install groups from LSD */
1189 if (lsd->lsd_ginfo) {
1190 ucred->luc_ginfo = lsd->lsd_ginfo;
1191 get_group_info(ucred->luc_ginfo);
1194 /* everything is done if we don't allow setgroups, or it is
1195 * from remote client (which implies forced to be no-setgroups).
1197 * Note: remote user's supplementary groups sent along the request
1198 * (if any) are all ignored, but we make the mapped local user's
1199 * supplementary groups take effect.
1201 if (med->med_remote || !(lsd_perms & LSD_PERM_SETGRP))
1204 /* root could set any groups as he want (if allowed), normal
1205 * users only could reduce his group array.
1207 if (ucred->luc_uid == 0) {
1208 drop_ucred_ginfo(ucred);
1210 if (rsd->rsd_ngroups == 0)
1213 gnew = groups_alloc(rsd->rsd_ngroups);
1215 CERROR("out of memory\n");
1216 drop_ucred_lsd(ucred);
1219 groups_from_buffer(gnew, rsd->rsd_groups);
1220 groups_sort(gnew); /* don't rely on client doing this */
1222 ucred->luc_ginfo = gnew;
1224 __u32 set = 0, cur = 0;
1225 struct group_info *ginfo = ucred->luc_ginfo;
1230 /* Note: freeing a group_info count on 'nblocks' instead of
1231 * 'ngroups', thus we can safely alloc enough buffer and reduce
1232 * and ngroups number later.
1234 gnew = groups_alloc(rsd->rsd_ngroups);
1236 CERROR("out of memory\n");
1237 drop_ucred_ginfo(ucred);
1238 drop_ucred_lsd(ucred);
1242 while (cur < rsd->rsd_ngroups) {
1243 if (groups_search(ginfo, rsd->rsd_groups[cur])) {
1244 GROUP_AT(gnew, set) = rsd->rsd_groups[cur];
1249 gnew->ngroups = set;
1251 put_group_info(ucred->luc_ginfo);
1252 ucred->luc_ginfo = gnew;
1257 void mds_exit_ucred(struct lvfs_ucred *ucred)
1260 drop_ucred_ginfo(ucred);
1261 drop_ucred_lsd(ucred);