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>
48 #include <linux/obd_support.h>
49 #include <linux/lustre_lib.h>
50 #include "mds_internal.h"
52 #include "mds_internal.h"
54 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,4)
55 struct group_info *groups_alloc(int ngroups)
57 struct group_info *ginfo;
59 LASSERT(ngroups <= NGROUPS_SMALL);
61 OBD_ALLOC(ginfo, sizeof(*ginfo) + 1 * sizeof(gid_t *));
64 ginfo->ngroups = ngroups;
66 ginfo->blocks[0] = ginfo->small_block;
67 atomic_set(&ginfo->usage, 1);
72 void groups_free(struct group_info *ginfo)
74 LASSERT(ginfo->ngroups <= NGROUPS_SMALL);
75 LASSERT(ginfo->nblocks == 1);
76 LASSERT(ginfo->blocks[0] == ginfo->small_block);
78 OBD_FREE(ginfo, sizeof(*ginfo) + 1 * sizeof(gid_t *));
81 /* for 2.4 the group number is small, so simply search the
84 int groups_search(struct group_info *ginfo, gid_t grp)
91 for (i = 0; i < ginfo->ngroups; i++)
92 if (GROUP_AT(ginfo, i) == grp)
99 void groups_sort(struct group_info *ginfo)
101 int base, max, stride;
102 int gidsetsize = ginfo->ngroups;
104 for (stride = 1; stride < gidsetsize; stride = 3 * stride + 1)
109 max = gidsetsize - stride;
110 for (base = 0; base < max; base++) {
112 int right = left + stride;
113 gid_t tmp = GROUP_AT(ginfo, right);
115 while (left >= 0 && GROUP_AT(ginfo, left) > tmp) {
116 GROUP_AT(ginfo, right) =
117 GROUP_AT(ginfo, left);
121 GROUP_AT(ginfo, right) = tmp;
127 int groups_search(struct group_info *ginfo, gid_t grp)
135 right = ginfo->ngroups;
136 while (left < right) {
137 int mid = (left + right) / 2;
138 int cmp = grp - GROUP_AT(ginfo, mid);
151 void groups_from_buffer(struct group_info *ginfo, __u32 *gids)
153 int i, ngroups = ginfo->ngroups;
155 for (i = 0; i < ginfo->nblocks; i++) {
156 int count = min(NGROUPS_PER_BLOCK, ngroups);
158 memcpy(ginfo->blocks[i], gids, count * sizeof(__u32));
159 gids += NGROUPS_PER_BLOCK;
164 void mds_pack_dentry2fid(struct ll_fid *fid, struct dentry *dentry)
166 fid->id = dentry->d_inum;
167 fid->generation = dentry->d_generation;
168 fid->mds = dentry->d_mdsnum;
171 void mds_pack_dentry2body(struct mds_body *b, struct dentry *dentry)
173 b->valid |= OBD_MD_FLID | OBD_MD_FLGENER;
174 b->ino = dentry->d_inum;
175 b->generation = dentry->d_generation;
176 b->mds = dentry->d_mdsnum;
179 void mds_pack_inode2fid(struct obd_device *obd, struct ll_fid *fid,
182 if (!obd || !fid || !inode) {
183 printk("obd %p, fid %p, inode %p\n", obd, fid, inode);
186 fid->id = inode->i_ino;
187 fid->generation = inode->i_generation;
188 fid->f_type = (S_IFMT & inode->i_mode);
189 fid->mds = obd->u.mds.mds_num;
192 /* Note that we can copy all of the fields, just some will not be "valid" */
193 void mds_pack_inode2body(struct obd_device *obd, struct mds_body *b,
196 b->valid |= OBD_MD_FLID | OBD_MD_FLCTIME | OBD_MD_FLUID |
197 OBD_MD_FLGID | OBD_MD_FLFLAGS | OBD_MD_FLTYPE |
198 OBD_MD_FLMODE | OBD_MD_FLNLINK | OBD_MD_FLGENER |
199 OBD_MD_FLATIME | OBD_MD_FLMTIME; /* bug 2020 */
201 if (!S_ISREG(inode->i_mode))
202 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLATIME |
203 OBD_MD_FLMTIME | OBD_MD_FLRDEV;
205 b->ino = inode->i_ino;
206 b->atime = LTIME_S(inode->i_atime);
207 b->mtime = LTIME_S(inode->i_mtime);
208 b->ctime = LTIME_S(inode->i_ctime);
209 b->mode = inode->i_mode;
210 b->size = inode->i_size;
211 b->blocks = inode->i_blocks;
212 b->uid = inode->i_uid;
213 b->gid = inode->i_gid;
214 b->flags = inode->i_flags;
215 b->rdev = inode->i_rdev;
216 /* Return the correct link count for orphan inodes */
217 if (mds_inode_is_orphan(inode)) {
219 } else if (S_ISDIR(inode->i_mode)) {
222 b->nlink = inode->i_nlink;
224 b->generation = inode->i_generation;
225 b->mds = obd->u.mds.mds_num;
229 static int mds_setattr_unpack(struct ptlrpc_request *req, int offset,
230 struct mds_update_record *r)
232 struct iattr *attr = &r->ur_iattr;
233 struct mds_rec_setattr *rec;
236 rec = lustre_swab_reqbuf(req, offset, sizeof(*rec),
237 lustre_swab_mds_rec_setattr);
241 r->ur_fid1 = &rec->sa_fid;
242 attr->ia_valid = rec->sa_valid;
243 attr->ia_mode = rec->sa_mode;
244 attr->ia_uid = rec->sa_uid;
245 attr->ia_gid = rec->sa_gid;
246 attr->ia_size = rec->sa_size;
247 LTIME_S(attr->ia_atime) = rec->sa_atime;
248 LTIME_S(attr->ia_mtime) = rec->sa_mtime;
249 LTIME_S(attr->ia_ctime) = rec->sa_ctime;
250 attr->ia_attr_flags = rec->sa_attr_flags;
252 LASSERT_REQSWAB (req, offset + 1);
253 if (req->rq_reqmsg->bufcount > offset + 1) {
254 r->ur_eadata = lustre_msg_buf (req->rq_reqmsg,
256 if (r->ur_eadata == NULL)
258 r->ur_eadatalen = req->rq_reqmsg->buflens[offset + 1];
261 if (req->rq_reqmsg->bufcount > offset + 2) {
262 r->ur_logcookies = lustre_msg_buf(req->rq_reqmsg, offset + 2,0);
263 if (r->ur_eadata == NULL)
266 r->ur_cookielen = req->rq_reqmsg->buflens[offset + 2];
272 static int mds_create_unpack(struct ptlrpc_request *req, int offset,
273 struct mds_update_record *r)
275 struct mds_rec_create *rec;
278 rec = lustre_swab_reqbuf (req, offset, sizeof (*rec),
279 lustre_swab_mds_rec_create);
283 r->ur_fid1 = &rec->cr_fid;
284 r->ur_fid2 = &rec->cr_replayfid;
285 r->ur_mode = rec->cr_mode;
286 r->ur_rdev = rec->cr_rdev;
287 r->ur_time = rec->cr_time;
288 r->ur_flags = rec->cr_flags;
290 LASSERT_REQSWAB (req, offset + 1);
291 r->ur_name = lustre_msg_string (req->rq_reqmsg, offset + 1, 0);
292 if (r->ur_name == NULL)
294 r->ur_namelen = req->rq_reqmsg->buflens[offset + 1];
296 LASSERT_REQSWAB (req, offset + 2);
297 if (req->rq_reqmsg->bufcount > offset + 2) {
298 if (S_ISLNK(r->ur_mode)) {
299 r->ur_tgt = lustre_msg_string(req->rq_reqmsg,
301 if (r->ur_tgt == NULL)
303 r->ur_tgtlen = req->rq_reqmsg->buflens[offset + 2];
304 } else if (S_ISDIR(r->ur_mode)) {
305 /* Stripe info for mkdir - just a 16bit integer */
306 if (req->rq_reqmsg->buflens[offset + 2] != 2) {
307 CERROR("mkdir stripe info does not match "
308 "expected size %d vs 2\n",
309 req->rq_reqmsg->buflens[offset + 2]);
312 r->ur_eadata = lustre_swab_buf (req->rq_reqmsg,
313 offset + 2, 2, __swab16s);
314 r->ur_eadatalen = req->rq_reqmsg->buflens[offset + 2];
316 /* Hm, no other users so far? */
323 static int mds_link_unpack(struct ptlrpc_request *req, int offset,
324 struct mds_update_record *r)
326 struct mds_rec_link *rec;
329 rec = lustre_swab_reqbuf (req, offset, sizeof (*rec),
330 lustre_swab_mds_rec_link);
334 r->ur_fid1 = &rec->lk_fid1;
335 r->ur_fid2 = &rec->lk_fid2;
336 r->ur_time = rec->lk_time;
338 LASSERT_REQSWAB (req, offset + 1);
339 r->ur_name = lustre_msg_string (req->rq_reqmsg, offset + 1, 0);
340 if (r->ur_name == NULL)
342 r->ur_namelen = req->rq_reqmsg->buflens[offset + 1];
346 static int mds_unlink_unpack(struct ptlrpc_request *req, int offset,
347 struct mds_update_record *r)
349 struct mds_rec_unlink *rec;
352 rec = lustre_swab_reqbuf (req, offset, sizeof (*rec),
353 lustre_swab_mds_rec_unlink);
357 r->ur_mode = rec->ul_mode;
358 r->ur_fid1 = &rec->ul_fid1;
359 r->ur_fid2 = &rec->ul_fid2;
360 r->ur_time = rec->ul_time;
362 LASSERT_REQSWAB (req, offset + 1);
363 r->ur_name = lustre_msg_string(req->rq_reqmsg, offset + 1, 0);
364 if (r->ur_name == NULL)
366 r->ur_namelen = req->rq_reqmsg->buflens[offset + 1];
370 static int mds_rename_unpack(struct ptlrpc_request *req, int offset,
371 struct mds_update_record *r)
373 struct mds_rec_rename *rec;
376 rec = lustre_swab_reqbuf (req, offset, sizeof (*rec),
377 lustre_swab_mds_rec_rename);
381 r->ur_fid1 = &rec->rn_fid1;
382 r->ur_fid2 = &rec->rn_fid2;
383 r->ur_time = rec->rn_time;
385 LASSERT_REQSWAB (req, offset + 1);
386 r->ur_name = lustre_msg_string(req->rq_reqmsg, offset + 1, 0);
387 if (r->ur_name == NULL)
389 r->ur_namelen = req->rq_reqmsg->buflens[offset + 1];
391 LASSERT_REQSWAB (req, offset + 2);
392 r->ur_tgt = lustre_msg_string(req->rq_reqmsg, offset + 2, 0);
393 if (r->ur_tgt == NULL)
395 r->ur_tgtlen = req->rq_reqmsg->buflens[offset + 2];
399 static int mds_open_unpack(struct ptlrpc_request *req, int offset,
400 struct mds_update_record *r)
402 struct mds_rec_create *rec;
405 rec = lustre_swab_reqbuf (req, offset, sizeof (*rec),
406 lustre_swab_mds_rec_create);
410 r->ur_fid1 = &rec->cr_fid;
411 r->ur_fid2 = &rec->cr_replayfid;
412 r->ur_mode = rec->cr_mode;
413 r->ur_rdev = rec->cr_rdev;
414 r->ur_time = rec->cr_time;
415 r->ur_flags = rec->cr_flags;
417 LASSERT_REQSWAB (req, offset + 1);
418 r->ur_name = lustre_msg_string (req->rq_reqmsg, offset + 1, 0);
419 if (r->ur_name == NULL)
421 r->ur_namelen = req->rq_reqmsg->buflens[offset + 1];
423 LASSERT_REQSWAB (req, offset + 2);
424 if (req->rq_reqmsg->bufcount > offset + 2) {
425 r->ur_eadata = lustre_msg_buf(req->rq_reqmsg, offset + 2, 0);
426 if (r->ur_eadata == NULL)
428 r->ur_eadatalen = req->rq_reqmsg->buflens[offset + 2];
433 typedef int (*update_unpacker)(struct ptlrpc_request *req, int offset,
434 struct mds_update_record *r);
436 static update_unpacker mds_unpackers[REINT_MAX + 1] = {
437 [REINT_SETATTR] mds_setattr_unpack,
438 [REINT_CREATE] mds_create_unpack,
439 [REINT_LINK] mds_link_unpack,
440 [REINT_UNLINK] mds_unlink_unpack,
441 [REINT_RENAME] mds_rename_unpack,
442 [REINT_OPEN] mds_open_unpack,
445 int mds_update_unpack(struct ptlrpc_request *req, int offset,
446 struct mds_update_record *rec)
453 /* NB don't lustre_swab_reqbuf() here. We're just taking a peek
454 * and we want to leave it to the specific unpacker once we've
455 * identified the message type */
456 opcodep = lustre_msg_buf (req->rq_reqmsg, offset, sizeof (*opcodep));
461 if (lustre_msg_swabbed (req->rq_reqmsg))
464 if (opcode > REINT_MAX ||
465 mds_unpackers[opcode] == NULL) {
466 CERROR ("Unexpected opcode %d\n", opcode);
470 rec->ur_opcode = opcode;
471 rc = mds_unpackers[opcode](req, offset, rec);
475 static inline void drop_ucred_ginfo(struct lvfs_ucred *ucred)
477 if (ucred->luc_ginfo) {
478 put_group_info(ucred->luc_ginfo);
479 ucred->luc_ginfo = NULL;
484 * root could set any group_info if we allowed setgroups, while
485 * normal user only could 'reduce' their group members -- which
486 * is somewhat expensive.
488 int mds_init_ucred(struct lvfs_ucred *ucred, struct mds_req_sec_desc *rsd)
490 struct group_info *gnew;
496 ucred->luc_fsuid = rsd->rsd_fsuid;
497 ucred->luc_fsgid = rsd->rsd_fsgid;
498 ucred->luc_cap = rsd->rsd_cap;
499 ucred->luc_uid = rsd->rsd_uid;
500 ucred->luc_ghash = mds_get_group_entry(NULL, rsd->rsd_uid);
501 ucred->luc_ginfo = NULL;
503 if (ucred->luc_ghash && ucred->luc_ghash->ge_group_info) {
504 ucred->luc_ginfo = ucred->luc_ghash->ge_group_info;
505 get_group_info(ucred->luc_ginfo);
508 /* everything is done if we don't allow setgroups */
509 if (!mds_allow_setgroups())
512 if (rsd->rsd_ngroups > LUSTRE_MAX_GROUPS) {
513 CERROR("client provide too many groups: %d\n",
515 drop_ucred_ginfo(ucred);
516 mds_put_group_entry(NULL, ucred->luc_ghash);
520 if (ucred->luc_uid == 0) {
521 if (rsd->rsd_ngroups == 0) {
522 drop_ucred_ginfo(ucred);
526 gnew = groups_alloc(rsd->rsd_ngroups);
528 CERROR("out of memory\n");
529 drop_ucred_ginfo(ucred);
530 mds_put_group_entry(NULL, ucred->luc_ghash);
533 groups_from_buffer(gnew, rsd->rsd_groups);
534 /* can't rely on client to sort them */
537 drop_ucred_ginfo(ucred);
538 ucred->luc_ginfo = gnew;
540 struct group_info *ginfo;
541 __u32 set = 0, cur = 0;
543 /* if no group info in hash, we don't
544 * bother createing new
546 if (!ucred->luc_ginfo)
549 /* Note: freeing a group_info count on 'nblocks' instead of
550 * 'ngroups', thus we can safely alloc enough buffer and reduce
551 * and ngroups number later.
553 gnew = groups_alloc(rsd->rsd_ngroups);
555 CERROR("out of memory\n");
556 drop_ucred_ginfo(ucred);
557 mds_put_group_entry(NULL, ucred->luc_ghash);
561 ginfo = ucred->luc_ginfo;
562 while (cur < rsd->rsd_ngroups) {
563 if (groups_search(ginfo, rsd->rsd_groups[cur]))
564 GROUP_AT(gnew, set++) = rsd->rsd_groups[cur];
569 put_group_info(ucred->luc_ginfo);
570 ucred->luc_ginfo = gnew;
575 void mds_exit_ucred(struct lvfs_ucred *ucred)
579 if (ucred->luc_ginfo)
580 put_group_info(ucred->luc_ginfo);
581 if (ucred->luc_ghash)
582 mds_put_group_entry(NULL, ucred->luc_ghash);