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);
387 static int mds_setattr_unpack(struct ptlrpc_request *req, int offset,
388 struct mds_update_record *r)
390 struct iattr *attr = &r->ur_iattr;
391 struct mds_rec_setattr *rec;
394 rec = lustre_swab_reqbuf(req, offset, sizeof(*rec),
395 lustre_swab_mds_rec_setattr);
399 r->ur_id1 = &rec->sa_id;
400 attr->ia_valid = rec->sa_valid;
401 attr->ia_mode = rec->sa_mode;
402 attr->ia_uid = rec->sa_uid;
403 attr->ia_gid = rec->sa_gid;
404 attr->ia_size = rec->sa_size;
405 LTIME_S(attr->ia_atime) = rec->sa_atime;
406 LTIME_S(attr->ia_mtime) = rec->sa_mtime;
407 LTIME_S(attr->ia_ctime) = rec->sa_ctime;
408 attr->ia_attr_flags = rec->sa_attr_flags;
410 LASSERT_REQSWAB (req, offset + 1);
411 if (req->rq_reqmsg->bufcount > offset + 1) {
412 r->ur_eadata = lustre_msg_buf (req->rq_reqmsg,
414 if (r->ur_eadata == NULL)
416 r->ur_eadatalen = req->rq_reqmsg->buflens[offset + 1];
419 if (req->rq_reqmsg->bufcount > offset + 2) {
420 r->ur_ea2data = lustre_msg_buf(req->rq_reqmsg, offset + 2, 0);
421 if (r->ur_ea2data == NULL)
424 r->ur_ea2datalen = req->rq_reqmsg->buflens[offset + 2];
427 if (req->rq_reqmsg->bufcount > offset + 3) {
428 r->ur_ea3data = lustre_msg_buf(req->rq_reqmsg, offset + 3, 0);
429 if (r->ur_ea3data == NULL)
432 r->ur_ea3datalen = req->rq_reqmsg->buflens[offset + 3];
438 static int mds_create_unpack(struct ptlrpc_request *req, int offset,
439 struct mds_update_record *r)
441 struct mds_rec_create *rec;
444 rec = lustre_swab_reqbuf (req, offset, sizeof (*rec),
445 lustre_swab_mds_rec_create);
449 r->ur_id1 = &rec->cr_id;
450 r->ur_id2 = &rec->cr_replayid;
451 r->ur_mode = rec->cr_mode;
452 r->ur_rdev = rec->cr_rdev;
453 r->ur_time = rec->cr_time;
454 r->ur_flags = rec->cr_flags;
456 LASSERT_REQSWAB (req, offset + 1);
457 r->ur_name = lustre_msg_string (req->rq_reqmsg, offset + 1, 0);
458 if (r->ur_name == NULL)
460 r->ur_namelen = req->rq_reqmsg->buflens[offset + 1];
462 LASSERT_REQSWAB (req, offset + 2);
463 if (req->rq_reqmsg->bufcount > offset + 2) {
464 if (S_ISLNK(r->ur_mode)) {
465 r->ur_tgt = lustre_msg_string(req->rq_reqmsg,
467 if (r->ur_tgt == NULL)
469 r->ur_tgtlen = req->rq_reqmsg->buflens[offset + 2];
470 } else if (S_ISDIR(r->ur_mode) ) {
471 /* Stripe info for mkdir - just a 16bit integer */
472 if (req->rq_reqmsg->buflens[offset + 2] != 2) {
473 CERROR("mkdir stripe info does not match "
474 "expected size %d vs 2\n",
475 req->rq_reqmsg->buflens[offset + 2]);
478 r->ur_eadata = lustre_swab_buf (req->rq_reqmsg,
479 offset + 2, 2, __swab16s);
480 r->ur_eadatalen = req->rq_reqmsg->buflens[offset + 2];
481 } else if (S_ISREG(r->ur_mode)){
482 r->ur_eadata = lustre_msg_buf (req->rq_reqmsg,
484 r->ur_eadatalen = req->rq_reqmsg->buflens[offset + 2];
486 /* Hm, no other users so far? */
493 static int mds_link_unpack(struct ptlrpc_request *req, int offset,
494 struct mds_update_record *r)
496 struct mds_rec_link *rec;
499 rec = lustre_swab_reqbuf (req, offset, sizeof (*rec),
500 lustre_swab_mds_rec_link);
504 r->ur_id1 = &rec->lk_id1;
505 r->ur_id2 = &rec->lk_id2;
506 r->ur_time = rec->lk_time;
508 LASSERT_REQSWAB (req, offset + 1);
509 r->ur_name = lustre_msg_string (req->rq_reqmsg, offset + 1, 0);
510 if (r->ur_name == NULL)
512 r->ur_namelen = req->rq_reqmsg->buflens[offset + 1];
516 static int mds_unlink_unpack(struct ptlrpc_request *req, int offset,
517 struct mds_update_record *r)
519 struct mds_rec_unlink *rec;
522 rec = lustre_swab_reqbuf (req, offset, sizeof (*rec),
523 lustre_swab_mds_rec_unlink);
527 r->ur_mode = rec->ul_mode;
528 r->ur_id1 = &rec->ul_id1;
529 r->ur_id2 = &rec->ul_id2;
530 r->ur_time = rec->ul_time;
532 LASSERT_REQSWAB (req, offset + 1);
533 r->ur_name = lustre_msg_string(req->rq_reqmsg, offset + 1, 0);
534 if (r->ur_name == NULL)
536 r->ur_namelen = req->rq_reqmsg->buflens[offset + 1];
540 static int mds_rename_unpack(struct ptlrpc_request *req, int offset,
541 struct mds_update_record *r)
543 struct mds_rec_rename *rec;
546 rec = lustre_swab_reqbuf (req, offset, sizeof (*rec),
547 lustre_swab_mds_rec_rename);
551 r->ur_id1 = &rec->rn_id1;
552 r->ur_id2 = &rec->rn_id2;
553 r->ur_time = rec->rn_time;
555 LASSERT_REQSWAB (req, offset + 1);
556 r->ur_name = lustre_msg_string(req->rq_reqmsg, offset + 1, 0);
557 if (r->ur_name == NULL)
559 r->ur_namelen = req->rq_reqmsg->buflens[offset + 1];
561 LASSERT_REQSWAB (req, offset + 2);
562 r->ur_tgt = lustre_msg_string(req->rq_reqmsg, offset + 2, 0);
563 if (r->ur_tgt == NULL)
565 r->ur_tgtlen = req->rq_reqmsg->buflens[offset + 2];
569 static int mds_open_unpack(struct ptlrpc_request *req, int offset,
570 struct mds_update_record *r)
572 struct mds_rec_create *rec;
575 rec = lustre_swab_reqbuf (req, offset, sizeof (*rec),
576 lustre_swab_mds_rec_create);
580 r->ur_id1 = &rec->cr_id;
581 r->ur_id2 = &rec->cr_replayid;
582 r->ur_mode = rec->cr_mode;
583 r->ur_rdev = rec->cr_rdev;
584 r->ur_time = rec->cr_time;
585 r->ur_flags = rec->cr_flags;
587 LASSERT_REQSWAB (req, offset + 1);
588 r->ur_name = lustre_msg_string (req->rq_reqmsg, offset + 1, 0);
589 if (r->ur_name == NULL)
591 r->ur_namelen = req->rq_reqmsg->buflens[offset + 1];
593 LASSERT_REQSWAB (req, offset + 2);
595 if (req->rq_reqmsg->bufcount > offset + 2) {
596 r->ur_eadata = lustre_msg_buf(req->rq_reqmsg, offset + 2, 0);
597 if (r->ur_eadata == NULL)
599 r->ur_eadatalen = req->rq_reqmsg->buflens[offset + 2];
602 if (rec->cr_flags & MDS_OPEN_HAS_KEY) {
603 LASSERT(req->rq_reqmsg->bufcount > offset + 3);
604 r->ur_ea2data = lustre_msg_buf(req->rq_reqmsg, offset + 3, 0);
605 r->ur_ea2datalen = req->rq_reqmsg->buflens[offset + 3];
610 typedef int (*update_unpacker)(struct ptlrpc_request *req, int offset,
611 struct mds_update_record *r);
613 static update_unpacker mds_unpackers[REINT_MAX + 1] = {
614 [REINT_SETATTR] mds_setattr_unpack,
615 [REINT_CREATE] mds_create_unpack,
616 [REINT_LINK] mds_link_unpack,
617 [REINT_UNLINK] mds_unlink_unpack,
618 [REINT_RENAME] mds_rename_unpack,
619 [REINT_OPEN] mds_open_unpack,
622 int mds_update_unpack(struct ptlrpc_request *req, int offset,
623 struct mds_update_record *rec)
631 * NB don't lustre_swab_reqbuf() here. We're just taking a peek and we
632 * want to leave it to the specific unpacker once we've identified the
635 opcodep = lustre_msg_buf (req->rq_reqmsg, offset, sizeof(*opcodep));
640 if (lustre_msg_swabbed (req->rq_reqmsg))
643 if (opcode > REINT_MAX ||
644 mds_unpackers[opcode] == NULL) {
645 CERROR ("Unexpected opcode %d\n", opcode);
651 rec->ur_opcode = opcode;
653 rc = mds_unpackers[opcode](req, offset, rec);
656 rec->ur_fsuid = req->rq_uid;
662 * here we take simple rule: once uid/fsuid is root, we also squash
663 * the gid/fsgid, don't care setuid/setgid attributes.
666 int mds_squash_root(struct mds_obd *mds, struct mds_req_sec_desc *rsd,
669 if (!mds->mds_squash_uid || *peernid == mds->mds_nosquash_nid)
672 if (rsd->rsd_uid && rsd->rsd_fsuid)
675 CDEBUG(D_SEC, "squash req from "LPX64":"
676 "(%u:%u-%u:%u/%x)=>(%u:%u-%u:%u/%x)\n", *peernid,
677 rsd->rsd_uid, rsd->rsd_gid,
678 rsd->rsd_fsuid, rsd->rsd_fsgid, rsd->rsd_cap,
679 rsd->rsd_uid ? rsd->rsd_uid : mds->mds_squash_uid,
680 rsd->rsd_uid ? rsd->rsd_gid : mds->mds_squash_gid,
681 rsd->rsd_fsuid ? rsd->rsd_fsuid : mds->mds_squash_uid,
682 rsd->rsd_fsuid ? rsd->rsd_fsgid : mds->mds_squash_gid,
683 rsd->rsd_cap & ~CAP_FS_MASK);
685 if (rsd->rsd_uid == 0) {
686 rsd->rsd_uid = mds->mds_squash_uid;
687 rsd->rsd_gid = mds->mds_squash_gid;
689 if (rsd->rsd_fsuid == 0) {
690 rsd->rsd_fsuid = mds->mds_squash_uid;
691 rsd->rsd_fsgid = mds->mds_squash_gid;
693 rsd->rsd_cap &= ~CAP_FS_MASK;
698 /********************************
699 * MDS uid/gid mapping handling *
700 ********************************/
703 struct mds_idmap_entry* idmap_alloc_entry(__u32 rmt_id, __u32 lcl_id)
705 struct mds_idmap_entry *e;
707 OBD_ALLOC(e, sizeof(*e));
711 INIT_LIST_HEAD(&e->rmt_hash);
712 INIT_LIST_HEAD(&e->lcl_hash);
713 atomic_set(&e->refcount, 1);
720 void idmap_free_entry(struct mds_idmap_entry *e)
722 if (!list_empty(&e->rmt_hash))
723 list_del(&e->rmt_hash);
724 if (!list_empty(&e->lcl_hash))
725 list_del(&e->lcl_hash);
726 OBD_FREE(e, sizeof(*e));
730 int idmap_insert_entry(struct list_head *rmt_hash, struct list_head *lcl_hash,
731 struct mds_idmap_entry *new, const char *warn_msg)
733 struct list_head *rmt_head = &rmt_hash[MDS_IDMAP_HASHFUNC(new->rmt_id)];
734 struct list_head *lcl_head = &lcl_hash[MDS_IDMAP_HASHFUNC(new->lcl_id)];
735 struct mds_idmap_entry *e;
737 list_for_each_entry(e, rmt_head, rmt_hash) {
738 if (e->rmt_id == new->rmt_id &&
739 e->lcl_id == new->lcl_id) {
740 atomic_inc(&e->refcount);
743 if (e->rmt_id == new->rmt_id && warn_msg)
744 CWARN("%s: rmt id %u already map to %u (new %u)\n",
745 warn_msg, e->rmt_id, e->lcl_id, new->lcl_id);
746 if (e->lcl_id == new->lcl_id && warn_msg)
747 CWARN("%s: lcl id %u already be mapped from %u "
748 "(new %u)\n", warn_msg,
749 e->lcl_id, e->rmt_id, new->rmt_id);
752 list_add_tail(rmt_head, &new->rmt_hash);
753 list_add_tail(lcl_head, &new->lcl_hash);
758 int idmap_remove_entry(struct list_head *rmt_hash, struct list_head *lcl_hash,
759 __u32 rmt_id, __u32 lcl_id)
761 struct list_head *rmt_head = &rmt_hash[MDS_IDMAP_HASHFUNC(rmt_id)];
762 struct mds_idmap_entry *e;
764 list_for_each_entry(e, rmt_head, rmt_hash) {
765 if (e->rmt_id == rmt_id && e->lcl_id == lcl_id) {
766 if (atomic_dec_and_test(&e->refcount)) {
767 list_del(&e->rmt_hash);
768 list_del(&e->lcl_hash);
769 OBD_FREE(e, sizeof(*e));
778 int mds_idmap_add(struct mds_idmap_table *tbl,
779 uid_t rmt_uid, uid_t lcl_uid,
780 gid_t rmt_gid, gid_t lcl_gid)
782 struct mds_idmap_entry *ue, *ge;
788 ue = idmap_alloc_entry(rmt_uid, lcl_uid);
791 ge = idmap_alloc_entry(rmt_gid, lcl_gid);
793 idmap_free_entry(ue);
797 spin_lock(&tbl->mit_lock);
799 if (idmap_insert_entry(tbl->mit_idmaps[MDS_RMT_UIDMAP_IDX],
800 tbl->mit_idmaps[MDS_LCL_UIDMAP_IDX],
801 ue, "UID mapping")) {
802 idmap_free_entry(ue);
805 if (idmap_insert_entry(tbl->mit_idmaps[MDS_RMT_GIDMAP_IDX],
806 tbl->mit_idmaps[MDS_LCL_GIDMAP_IDX],
807 ge, "GID mapping")) {
808 idmap_free_entry(ge);
811 spin_unlock(&tbl->mit_lock);
815 int mds_idmap_del(struct mds_idmap_table *tbl,
816 uid_t rmt_uid, uid_t lcl_uid,
817 gid_t rmt_gid, gid_t lcl_gid)
824 spin_lock(&tbl->mit_lock);
825 idmap_remove_entry(tbl->mit_idmaps[MDS_RMT_UIDMAP_IDX],
826 tbl->mit_idmaps[MDS_LCL_UIDMAP_IDX],
828 idmap_remove_entry(tbl->mit_idmaps[MDS_RMT_GIDMAP_IDX],
829 tbl->mit_idmaps[MDS_LCL_GIDMAP_IDX],
831 spin_unlock(&tbl->mit_lock);
836 __u32 idmap_lookup_id(struct list_head *hash, int reverse, __u32 id)
838 struct list_head *head = &hash[MDS_IDMAP_HASHFUNC(id)];
839 struct mds_idmap_entry *e;
842 list_for_each_entry(e, head, rmt_hash) {
846 return MDS_IDMAP_NOTFOUND;
848 list_for_each_entry(e, head, lcl_hash) {
852 return MDS_IDMAP_NOTFOUND;
856 int mds_idmap_lookup_uid(struct mds_idmap_table *tbl, int reverse, uid_t uid)
858 struct list_head *hash;
861 return MDS_IDMAP_NOTFOUND;
864 hash = tbl->mit_idmaps[MDS_RMT_UIDMAP_IDX];
866 hash = tbl->mit_idmaps[MDS_LCL_UIDMAP_IDX];
868 spin_lock(&tbl->mit_lock);
869 uid = idmap_lookup_id(hash, reverse, uid);
870 spin_unlock(&tbl->mit_lock);
875 int mds_idmap_lookup_gid(struct mds_idmap_table *tbl, int reverse, gid_t gid)
877 struct list_head *hash;
880 return MDS_IDMAP_NOTFOUND;
883 hash = tbl->mit_idmaps[MDS_RMT_GIDMAP_IDX];
885 hash = tbl->mit_idmaps[MDS_LCL_GIDMAP_IDX];
887 spin_lock(&tbl->mit_lock);
888 gid = idmap_lookup_id(hash, reverse, gid);
889 spin_unlock(&tbl->mit_lock);
894 struct mds_idmap_table *mds_idmap_alloc()
896 struct mds_idmap_table *tbl;
899 OBD_ALLOC(tbl, sizeof(*tbl));
903 spin_lock_init(&tbl->mit_lock);
904 for (i = 0; i < MDS_IDMAP_N_HASHES; i++)
905 for (j = 0; j < MDS_IDMAP_HASHSIZE; j++)
906 INIT_LIST_HEAD(&tbl->mit_idmaps[i][j]);
911 static void idmap_clear_rmt_hash(struct list_head *list)
913 struct mds_idmap_entry *e;
916 for (i = 0; i < MDS_IDMAP_HASHSIZE; i++) {
917 while (!list_empty(&list[i])) {
918 e = list_entry(list[i].next, struct mds_idmap_entry,
925 void mds_idmap_free(struct mds_idmap_table *tbl)
929 spin_lock(&tbl->mit_lock);
930 idmap_clear_rmt_hash(tbl->mit_idmaps[MDS_RMT_UIDMAP_IDX]);
931 idmap_clear_rmt_hash(tbl->mit_idmaps[MDS_RMT_GIDMAP_IDX]);
933 /* paranoid checking */
934 for (i = 0; i < MDS_IDMAP_HASHSIZE; i++) {
935 LASSERT(list_empty(&tbl->mit_idmaps[MDS_LCL_UIDMAP_IDX][i]));
936 LASSERT(list_empty(&tbl->mit_idmaps[MDS_LCL_GIDMAP_IDX][i]));
938 spin_unlock(&tbl->mit_lock);
940 OBD_FREE(tbl, sizeof(*tbl));
943 /*********************************
944 * helpers doing mapping for MDS *
945 *********************************/
948 * we allow remote setuid/setgid to an "authencated" one,
949 * this policy probably change later.
952 int mds_req_secdesc_do_map(struct mds_export_data *med,
953 struct mds_req_sec_desc *rsd)
955 struct mds_idmap_table *idmap = med->med_idmap;
959 uid = mds_idmap_lookup_uid(idmap, 0, rsd->rsd_uid);
960 if (uid == MDS_IDMAP_NOTFOUND) {
961 CERROR("can't find map for uid %u\n", rsd->rsd_uid);
965 if (rsd->rsd_uid == rsd->rsd_fsuid)
968 fsuid = mds_idmap_lookup_uid(idmap, 0, rsd->rsd_fsuid);
969 if (fsuid == MDS_IDMAP_NOTFOUND) {
970 CERROR("can't find map for fsuid %u\n", rsd->rsd_fsuid);
975 gid = mds_idmap_lookup_gid(idmap, 0, rsd->rsd_gid);
976 if (gid == MDS_IDMAP_NOTFOUND) {
977 CERROR("can't find map for gid %u\n", rsd->rsd_gid);
981 if (rsd->rsd_gid == rsd->rsd_fsgid)
984 fsgid = mds_idmap_lookup_gid(idmap, 0, rsd->rsd_fsgid);
985 if (fsgid == MDS_IDMAP_NOTFOUND) {
986 CERROR("can't find map for fsgid %u\n", rsd->rsd_fsgid);
993 rsd->rsd_fsuid = fsuid;
994 rsd->rsd_fsgid = fsgid;
999 void mds_body_do_reverse_map(struct mds_export_data *med,
1000 struct mds_body *body)
1005 if (!med->med_remote)
1009 if (body->valid & OBD_MD_FLUID) {
1010 uid = mds_idmap_lookup_uid(med->med_idmap, 1, body->uid);
1011 if (uid == MDS_IDMAP_NOTFOUND) {
1012 uid = med->med_nllu;
1013 if (body->valid & OBD_MD_FLMODE) {
1014 body->mode = (body->mode & ~S_IRWXU) |
1015 ((body->mode & S_IRWXO) << 6);
1020 if (body->valid & OBD_MD_FLGID) {
1021 gid = mds_idmap_lookup_gid(med->med_idmap, 1, body->gid);
1022 if (gid == MDS_IDMAP_NOTFOUND) {
1023 gid = med->med_nllg;
1024 if (body->valid & OBD_MD_FLMODE) {
1025 body->mode = (body->mode & ~S_IRWXG) |
1026 ((body->mode & S_IRWXO) << 3);
1035 /**********************
1036 * MDS ucred handling *
1037 **********************/
1039 static inline void drop_ucred_ginfo(struct lvfs_ucred *ucred)
1041 if (ucred->luc_ginfo) {
1042 put_group_info(ucred->luc_ginfo);
1043 ucred->luc_ginfo = NULL;
1047 static inline void drop_ucred_lsd(struct lvfs_ucred *ucred)
1049 if (ucred->luc_lsd) {
1050 mds_put_lsd(ucred->luc_lsd);
1051 ucred->luc_lsd = NULL;
1056 * the heart of the uid/gid handling and security checking.
1058 * root could set any group_info if we allowed setgroups, while
1059 * normal user only could 'reduce' their group members -- which
1060 * is somewhat expensive.
1062 * authenticated as mds user (using mds service credential) could
1063 * bypass all checkings.
1065 int mds_init_ucred(struct lvfs_ucred *ucred,
1066 struct ptlrpc_request *req,
1067 struct mds_req_sec_desc *rsd)
1069 struct mds_obd *mds = &req->rq_export->exp_obd->u.mds;
1070 struct mds_export_data *med = &req->rq_export->u.eu_mds_data;
1071 struct lustre_sec_desc *lsd;
1072 ptl_nid_t peernid = req->rq_peer.peer_id.nid;
1073 struct group_info *gnew;
1074 unsigned int setuid, setgid, strong_sec, root_squashed;
1080 LASSERT(rsd->rsd_ngroups <= LUSTRE_MAX_GROUPS);
1082 if (SEC_FLAVOR_MAJOR(req->rq_req_secflvr) == PTLRPCS_FLVR_MAJOR_GSS &&
1083 (SEC_FLAVOR_SVC(req->rq_req_secflvr) == PTLRPCS_SVC_AUTH ||
1084 SEC_FLAVOR_SVC(req->rq_req_secflvr) == PTLRPCS_SVC_PRIV))
1089 LASSERT(!(req->rq_remote_realm && !strong_sec));
1091 if (strong_sec && req->rq_auth_uid == -1) {
1092 CWARN("user not authenticated, deny access\n");
1096 /* sanity check: if we use strong authentication, we expect the
1097 * uid which client claimed is true.
1098 * not apply to special mds user .
1100 if (!req->rq_auth_usr_mds && strong_sec) {
1101 if (!med->med_remote) {
1102 if (req->rq_auth_uid != rsd->rsd_uid) {
1103 CERROR("local client "LPU64": auth uid %u "
1104 "while client claim %u:%u/%u:%u\n",
1105 peernid, req->rq_auth_uid,
1106 rsd->rsd_uid, rsd->rsd_gid,
1107 rsd->rsd_fsuid, rsd->rsd_fsgid);
1111 if (req->rq_mapped_uid == MDS_IDMAP_NOTFOUND) {
1112 CWARN("no mapping found, deny\n");
1116 if (mds_req_secdesc_do_map(med, rsd))
1119 if (req->rq_mapped_uid != rsd->rsd_uid) {
1120 CERROR("remote client "LPU64": auth uid %u "
1121 "while client claim %u:%u/%u:%u\n",
1122 peernid, req->rq_auth_uid,
1123 rsd->rsd_uid, rsd->rsd_gid,
1124 rsd->rsd_fsuid, rsd->rsd_fsgid);
1130 /* now LSD come into play */
1131 ucred->luc_ginfo = NULL;
1132 ucred->luc_lsd = lsd = mds_get_lsd(rsd->rsd_uid);
1135 CERROR("Deny access without LSD: uid %d\n", rsd->rsd_uid);
1139 lsd_perms = mds_lsd_get_perms(lsd, med->med_remote, 0, peernid);
1141 /* check setuid/setgid permissions.
1142 * again not apply to special mds user.
1144 if (!req->rq_auth_usr_mds) {
1145 /* find out the setuid/setgid attempt */
1146 setuid = (rsd->rsd_uid != rsd->rsd_fsuid);
1147 setgid = (rsd->rsd_gid != rsd->rsd_fsgid ||
1148 rsd->rsd_gid != lsd->lsd_gid);
1150 /* check permission of setuid */
1151 if (setuid && !(lsd_perms & LSD_PERM_SETUID)) {
1152 CWARN("mds blocked setuid attempt (%u -> %u) "
1153 "from "LPU64"\n", rsd->rsd_uid, rsd->rsd_fsuid,
1158 /* check permission of setgid */
1159 if (setgid && !(lsd_perms & LSD_PERM_SETGID)) {
1160 CWARN("mds blocked setgid attempt (%u:%u/%u:%u -> %u) "
1161 "from "LPU64"\n", rsd->rsd_uid, rsd->rsd_gid,
1162 rsd->rsd_fsuid, rsd->rsd_fsgid, lsd->lsd_gid,
1168 root_squashed = mds_squash_root(mds, rsd, &peernid);
1170 /* remove privilege for non-root user */
1172 rsd->rsd_cap &= ~CAP_FS_MASK;
1174 /* by now every fields other than groups in rsd have been granted */
1175 ucred->luc_uid = rsd->rsd_uid;
1176 ucred->luc_gid = rsd->rsd_gid;
1177 ucred->luc_fsuid = rsd->rsd_fsuid;
1178 ucred->luc_fsgid = rsd->rsd_fsgid;
1179 ucred->luc_cap = rsd->rsd_cap;
1181 /* don't use any supplementary group if we squashed root.
1182 * XXX The exact behavior of root_squash is not defined, we just
1183 * keep the reminder here */
1187 /* install groups from LSD */
1188 if (lsd->lsd_ginfo) {
1189 ucred->luc_ginfo = lsd->lsd_ginfo;
1190 get_group_info(ucred->luc_ginfo);
1193 /* everything is done if we don't allow setgroups, or it is
1194 * from remote client (which implies forced to be no-setgroups).
1196 * Note: remote user's supplementary groups sent along the request
1197 * (if any) are all ignored, but we make the mapped local user's
1198 * supplementary groups take effect.
1200 if (med->med_remote || !(lsd_perms & LSD_PERM_SETGRP))
1203 /* root could set any groups as he want (if allowed), normal
1204 * users only could reduce his group array.
1206 if (ucred->luc_uid == 0) {
1207 drop_ucred_ginfo(ucred);
1209 if (rsd->rsd_ngroups == 0)
1212 gnew = groups_alloc(rsd->rsd_ngroups);
1214 CERROR("out of memory\n");
1215 drop_ucred_lsd(ucred);
1218 groups_from_buffer(gnew, rsd->rsd_groups);
1219 groups_sort(gnew); /* don't rely on client doing this */
1221 ucred->luc_ginfo = gnew;
1223 __u32 set = 0, cur = 0;
1224 struct group_info *ginfo = ucred->luc_ginfo;
1229 /* Note: freeing a group_info count on 'nblocks' instead of
1230 * 'ngroups', thus we can safely alloc enough buffer and reduce
1231 * and ngroups number later.
1233 gnew = groups_alloc(rsd->rsd_ngroups);
1235 CERROR("out of memory\n");
1236 drop_ucred_ginfo(ucred);
1237 drop_ucred_lsd(ucred);
1241 while (cur < rsd->rsd_ngroups) {
1242 if (groups_search(ginfo, rsd->rsd_groups[cur])) {
1243 GROUP_AT(gnew, set) = rsd->rsd_groups[cur];
1248 gnew->ngroups = set;
1250 put_group_info(ucred->luc_ginfo);
1251 ucred->luc_ginfo = gnew;
1256 void mds_exit_ucred(struct lvfs_ucred *ucred)
1259 drop_ucred_ginfo(ucred);
1260 drop_ucred_lsd(ucred);