1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright 2008 Sun Microsystems, Inc. All rights reserved
30 * Use is subject to license terms.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
37 #define DEBUG_SUBSYSTEM S_MDS
39 #ifndef AUTOCONF_INCLUDED
40 #include <linux/config.h>
42 #include <linux/module.h>
43 #include <linux/kernel.h>
45 #include <linux/string.h>
46 #include <linux/stat.h>
47 #include <linux/errno.h>
48 #include <linux/version.h>
49 #include <linux/buffer_head.h> // for wait_on_buffer
50 #include <linux/unistd.h>
52 #include <asm/system.h>
53 #include <asm/uaccess.h>
56 #include <linux/stat.h>
57 #include <asm/uaccess.h>
58 #include <linux/slab.h>
60 #include <obd_support.h>
61 #include <lustre_lib.h>
62 #include "mds_internal.h"
64 static void mds_pack_inode2fid(struct ll_fid *fid, struct inode *inode)
66 fid->id = inode->i_ino;
67 fid->generation = inode->i_generation;
68 fid->f_type = (S_IFMT & inode->i_mode);
71 /* Note that we can copy all of the fields, just some will not be "valid" */
72 void mds_pack_inode2body(struct mds_body *b, struct inode *inode)
74 b->valid |= OBD_MD_FLID | OBD_MD_FLCTIME | OBD_MD_FLUID |
75 OBD_MD_FLGID | OBD_MD_FLFLAGS | OBD_MD_FLTYPE |
76 OBD_MD_FLMODE | OBD_MD_FLNLINK | OBD_MD_FLGENER |
77 OBD_MD_FLATIME | OBD_MD_FLMTIME; /* bug 2020 */
79 if (!S_ISREG(inode->i_mode))
80 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLATIME |
81 OBD_MD_FLMTIME | OBD_MD_FLRDEV;
83 mds_pack_inode2fid(&b->fid1, inode);
84 b->ino = inode->i_ino;
85 b->atime = LTIME_S(inode->i_atime);
86 b->mtime = LTIME_S(inode->i_mtime);
87 b->ctime = LTIME_S(inode->i_ctime);
88 b->mode = inode->i_mode;
89 b->size = i_size_read(inode);
90 b->blocks = inode->i_blocks;
91 b->uid = inode->i_uid;
92 b->gid = inode->i_gid;
93 b->flags = ll_inode_to_ext_flags(inode->i_flags, MDS_BFLAG_EXT_FLAGS);
94 b->rdev = inode->i_rdev;
95 /* Return the correct link count for orphan inodes */
96 b->nlink = mds_inode_is_orphan(inode) ? 0 : inode->i_nlink;
97 b->generation = inode->i_generation;
101 static inline unsigned int attr_unpack(__u64 sa_valid) {
102 unsigned int ia_valid = 0;
104 if (sa_valid & MDS_ATTR_MODE)
105 ia_valid |= ATTR_MODE;
106 if (sa_valid & MDS_ATTR_UID)
107 ia_valid |= ATTR_UID;
108 if (sa_valid & MDS_ATTR_GID)
109 ia_valid |= ATTR_GID;
110 if (sa_valid & MDS_ATTR_SIZE)
111 ia_valid |= ATTR_SIZE;
112 if (sa_valid & MDS_ATTR_ATIME)
113 ia_valid |= ATTR_ATIME;
114 if (sa_valid & MDS_ATTR_MTIME)
115 ia_valid |= ATTR_MTIME;
116 if (sa_valid & MDS_ATTR_CTIME)
117 ia_valid |= ATTR_CTIME;
118 if (sa_valid & MDS_ATTR_ATIME_SET)
119 ia_valid |= ATTR_ATIME_SET;
120 if (sa_valid & MDS_ATTR_MTIME_SET)
121 ia_valid |= ATTR_MTIME_SET;
122 if (sa_valid & MDS_ATTR_FORCE)
123 ia_valid |= ATTR_FORCE;
124 if (sa_valid & MDS_ATTR_ATTR_FLAG)
125 ia_valid |= ATTR_ATTR_FLAG;
126 if (sa_valid & MDS_ATTR_KILL_SUID)
127 ia_valid |= ATTR_KILL_SUID;
128 if (sa_valid & MDS_ATTR_KILL_SGID)
129 ia_valid |= ATTR_KILL_SGID;
130 if (sa_valid & MDS_ATTR_CTIME_SET)
131 ia_valid |= ATTR_CTIME_SET;
132 if (sa_valid & MDS_ATTR_FROM_OPEN)
133 ia_valid |= ATTR_FROM_OPEN;
134 if (sa_valid & MDS_OPEN_OWNEROVERRIDE)
135 ia_valid |= MDS_OPEN_OWNEROVERRIDE;
140 static int mds_setattr_unpack(struct ptlrpc_request *req, int offset,
141 struct mds_update_record *r)
143 struct iattr *attr = &r->ur_iattr;
144 struct mds_rec_setattr *rec;
147 rec = lustre_swab_reqbuf(req, offset, sizeof(*rec),
148 lustre_swab_mds_rec_setattr);
152 r->ur_uc.luc_fsuid = rec->sa_fsuid;
153 r->ur_uc.luc_fsgid = rec->sa_fsgid;
154 cfs_kernel_cap_unpack(&r->ur_uc.luc_cap, rec->sa_cap);
155 r->ur_uc.luc_suppgid1 = rec->sa_suppgid;
156 r->ur_uc.luc_suppgid2 = -1;
157 r->ur_fid1 = &rec->sa_fid;
158 attr->ia_valid = attr_unpack(rec->sa_valid);
159 attr->ia_mode = rec->sa_mode;
160 attr->ia_uid = rec->sa_uid;
161 attr->ia_gid = rec->sa_gid;
162 attr->ia_size = rec->sa_size;
163 LTIME_S(attr->ia_atime) = rec->sa_atime;
164 LTIME_S(attr->ia_mtime) = rec->sa_mtime;
165 LTIME_S(attr->ia_ctime) = rec->sa_ctime;
166 r->ur_flags = rec->sa_attr_flags;
168 lustre_set_req_swabbed(req, offset + 1);
169 r->ur_eadatalen = lustre_msg_buflen(req->rq_reqmsg, offset + 1);
170 if (r->ur_eadatalen) {
171 r->ur_eadata = lustre_msg_buf(req->rq_reqmsg, offset + 1, 0);
172 if (r->ur_eadata == NULL)
175 r->ur_cookielen = lustre_msg_buflen(req->rq_reqmsg, offset + 2);
176 if (r->ur_cookielen) {
177 r->ur_logcookies = lustre_msg_buf(req->rq_reqmsg, offset + 2,0);
178 if (r->ur_eadata == NULL)
181 if (lustre_msg_buflen(req->rq_reqmsg, offset + 3)) {
182 r->ur_dlm = lustre_swab_reqbuf(req, offset + 3,
184 lustre_swab_ldlm_request);
185 if (r->ur_dlm == NULL)
191 static int mds_create_unpack(struct ptlrpc_request *req, int offset,
192 struct mds_update_record *r)
194 struct mds_rec_create *rec;
197 rec = lustre_swab_reqbuf(req, offset, sizeof (*rec),
198 lustre_swab_mds_rec_create);
202 r->ur_uc.luc_fsuid = rec->cr_fsuid;
203 r->ur_uc.luc_fsgid = rec->cr_fsgid;
204 cfs_kernel_cap_unpack(&r->ur_uc.luc_cap, rec->cr_cap);
205 r->ur_uc.luc_suppgid1 = rec->cr_suppgid;
206 r->ur_uc.luc_suppgid2 = -1;
207 r->ur_fid1 = &rec->cr_fid;
208 r->ur_fid2 = &rec->cr_replayfid;
209 r->ur_mode = rec->cr_mode;
210 r->ur_rdev = rec->cr_rdev;
211 r->ur_time = rec->cr_time;
212 r->ur_flags = rec->cr_flags;
214 lustre_set_req_swabbed(req, offset + 1);
215 r->ur_name = lustre_msg_string(req->rq_reqmsg, offset + 1, 0);
216 if (r->ur_name == NULL)
218 r->ur_namelen = lustre_msg_buflen(req->rq_reqmsg, offset + 1);
220 lustre_set_req_swabbed(req, offset + 2);
221 r->ur_tgtlen = lustre_msg_buflen(req->rq_reqmsg, offset + 2);
223 /* NB for now, we only seem to pass NULL terminated symlink
224 * target strings here. If this ever changes, we'll have
225 * to stop checking for a buffer filled completely with a
226 * NULL terminated string here, and make the callers check
227 * depending on what they expect. We should probably stash
228 * it in r->ur_eadata in that case, so it's obvious... -eeb
230 r->ur_tgt = lustre_msg_string(req->rq_reqmsg, offset + 2, 0);
231 if (r->ur_tgt == NULL)
234 if (lustre_msg_buflen(req->rq_reqmsg, offset + 3)) {
235 r->ur_dlm = lustre_swab_reqbuf(req, offset + 3,
237 lustre_swab_ldlm_request);
238 if (r->ur_dlm == NULL)
244 static int mds_link_unpack(struct ptlrpc_request *req, int offset,
245 struct mds_update_record *r)
247 struct mds_rec_link *rec;
250 rec = lustre_swab_reqbuf(req, offset, sizeof (*rec),
251 lustre_swab_mds_rec_link);
255 r->ur_uc.luc_fsuid = rec->lk_fsuid;
256 r->ur_uc.luc_fsgid = rec->lk_fsgid;
257 cfs_kernel_cap_unpack(&r->ur_uc.luc_cap, rec->lk_cap);
258 r->ur_uc.luc_suppgid1 = rec->lk_suppgid1;
259 r->ur_uc.luc_suppgid2 = rec->lk_suppgid2;
260 r->ur_fid1 = &rec->lk_fid1;
261 r->ur_fid2 = &rec->lk_fid2;
262 r->ur_time = rec->lk_time;
264 lustre_set_req_swabbed(req, offset + 1);
265 r->ur_name = lustre_msg_string(req->rq_reqmsg, offset + 1, 0);
266 if (r->ur_name == NULL)
268 r->ur_namelen = lustre_msg_buflen(req->rq_reqmsg, offset + 1);
269 if (lustre_msg_buflen(req->rq_reqmsg, offset + 2)) {
270 r->ur_dlm = lustre_swab_reqbuf(req, offset + 2,
272 lustre_swab_ldlm_request);
273 if (r->ur_dlm == NULL)
279 static int mds_unlink_unpack(struct ptlrpc_request *req, int offset,
280 struct mds_update_record *r)
282 struct mds_rec_unlink *rec;
285 rec = lustre_swab_reqbuf(req, offset, sizeof (*rec),
286 lustre_swab_mds_rec_unlink);
290 r->ur_uc.luc_fsuid = rec->ul_fsuid;
291 r->ur_uc.luc_fsgid = rec->ul_fsgid;
292 cfs_kernel_cap_unpack(&r->ur_uc.luc_cap, rec->ul_cap);
293 r->ur_uc.luc_suppgid1 = rec->ul_suppgid;
294 r->ur_uc.luc_suppgid2 = -1;
295 r->ur_mode = rec->ul_mode;
296 r->ur_fid1 = &rec->ul_fid1;
297 r->ur_fid2 = &rec->ul_fid2;
298 r->ur_time = rec->ul_time;
300 lustre_set_req_swabbed(req, offset + 1);
301 r->ur_name = lustre_msg_string(req->rq_reqmsg, offset + 1, 0);
302 if (r->ur_name == NULL)
304 r->ur_namelen = lustre_msg_buflen(req->rq_reqmsg, offset + 1);
305 if (lustre_msg_buflen(req->rq_reqmsg, offset + 2)) {
306 r->ur_dlm = lustre_swab_reqbuf(req, offset + 2,
308 lustre_swab_ldlm_request);
309 if (r->ur_dlm == NULL)
315 static int mds_rename_unpack(struct ptlrpc_request *req, int offset,
316 struct mds_update_record *r)
318 struct mds_rec_rename *rec;
321 rec = lustre_swab_reqbuf(req, offset, sizeof (*rec),
322 lustre_swab_mds_rec_rename);
326 r->ur_uc.luc_fsuid = rec->rn_fsuid;
327 r->ur_uc.luc_fsgid = rec->rn_fsgid;
328 cfs_kernel_cap_unpack(&r->ur_uc.luc_cap, rec->rn_cap);
329 r->ur_uc.luc_suppgid1 = rec->rn_suppgid1;
330 r->ur_uc.luc_suppgid2 = rec->rn_suppgid2;
331 r->ur_fid1 = &rec->rn_fid1;
332 r->ur_fid2 = &rec->rn_fid2;
333 r->ur_time = rec->rn_time;
335 lustre_set_req_swabbed(req, offset + 1);
336 r->ur_name = lustre_msg_string(req->rq_reqmsg, offset + 1, 0);
337 if (r->ur_name == NULL)
339 r->ur_namelen = lustre_msg_buflen(req->rq_reqmsg, offset + 1);
341 lustre_set_req_swabbed(req, offset + 2);
342 r->ur_tgt = lustre_msg_string(req->rq_reqmsg, offset + 2, 0);
343 if (r->ur_tgt == NULL)
345 r->ur_tgtlen = lustre_msg_buflen(req->rq_reqmsg, offset + 2);
346 if (lustre_msg_buflen(req->rq_reqmsg, offset + 3)) {
347 r->ur_dlm = lustre_swab_reqbuf(req, offset + 3,
349 lustre_swab_ldlm_request);
350 if (r->ur_dlm == NULL)
356 static int mds_open_unpack(struct ptlrpc_request *req, int offset,
357 struct mds_update_record *r)
359 struct mds_rec_create *rec;
362 rec = lustre_swab_reqbuf(req, offset, sizeof(*rec),
363 lustre_swab_mds_rec_create);
367 r->ur_uc.luc_fsuid = rec->cr_fsuid;
368 r->ur_uc.luc_fsgid = rec->cr_fsgid;
369 cfs_kernel_cap_unpack(&r->ur_uc.luc_cap, rec->cr_cap);
370 r->ur_uc.luc_suppgid1 = rec->cr_suppgid;
371 r->ur_uc.luc_suppgid2 = -1;
372 r->ur_fid1 = &rec->cr_fid;
373 r->ur_fid2 = &rec->cr_replayfid;
374 r->ur_mode = rec->cr_mode;
375 r->ur_rdev = rec->cr_rdev;
376 r->ur_time = rec->cr_time;
377 r->ur_flags = rec->cr_flags;
379 lustre_set_req_swabbed(req, offset + 1);
380 r->ur_name = lustre_msg_string(req->rq_reqmsg, offset + 1, 0);
381 if (r->ur_name == NULL)
383 r->ur_namelen = lustre_msg_buflen(req->rq_reqmsg, offset + 1);
385 lustre_set_req_swabbed(req, offset + 2);
386 r->ur_eadatalen = lustre_msg_buflen(req->rq_reqmsg, offset + 2);
387 if (r->ur_eadatalen) {
388 r->ur_eadata = lustre_msg_buf(req->rq_reqmsg, offset + 2, 0);
389 if (r->ur_eadata == NULL)
395 typedef int (*update_unpacker)(struct ptlrpc_request *req, int offset,
396 struct mds_update_record *r);
398 static update_unpacker mds_unpackers[REINT_MAX] = {
399 [REINT_SETATTR] mds_setattr_unpack,
400 [REINT_CREATE] mds_create_unpack,
401 [REINT_LINK] mds_link_unpack,
402 [REINT_UNLINK] mds_unlink_unpack,
403 [REINT_RENAME] mds_rename_unpack,
404 [REINT_OPEN] mds_open_unpack,
407 int mds_update_unpack(struct ptlrpc_request *req, int offset,
408 struct mds_update_record *rec)
410 mds_reint_t opcode, *opcodep;
414 /* NB don't lustre_swab_reqbuf() here. We're just taking a peek
415 * and we want to leave it to the specific unpacker once we've
416 * identified the message type */
417 opcodep = lustre_msg_buf(req->rq_reqmsg, offset, sizeof (*opcodep));
422 if (lustre_req_need_swab(req))
425 if (opcode >= REINT_MAX || mds_unpackers[opcode] == NULL) {
426 CERROR("Unexpected opcode %d\n", opcode);
430 rec->ur_opcode = opcode;
431 rc = mds_unpackers[opcode](req, offset, rec);
436 void mds_root_squash(struct mds_obd *mds, lnet_nid_t *peernid,
437 __u32 *fsuid, __u32 *fsgid, cfs_kernel_cap_t *kcap,
438 __u32 *suppgid, __u32 *suppgid2)
440 if (!mds->mds_squash_uid || *fsuid)
443 if (*peernid == mds->mds_nosquash_nid)
446 CDEBUG(D_OTHER, "squash req from %s, (%d:%d)=>(%d:%d)\n",
447 libcfs_nid2str(*peernid), *fsuid, *fsgid,
448 mds->mds_squash_uid, mds->mds_squash_gid);
450 *fsuid = mds->mds_squash_uid;
451 *fsgid = mds->mds_squash_gid;
452 cfs_kernel_cap_unpack(kcap, 0);
458 int mds_init_ucred(struct lvfs_ucred *ucred, struct ptlrpc_request *req,
461 struct mds_body *body = lustre_msg_buf(req->rq_reqmsg, offset,
463 struct mds_obd *mds = mds_req2mds(req);
466 LASSERT(body != NULL); /* previously verified & swabbed by caller */
469 if (req->rq_uid != LNET_UID_ANY) {
470 /* Non-root local cluster client */
471 LASSERT (req->rq_uid != 0);
472 ucred->luc_fsuid = req->rq_uid;
476 cfs_kernel_cap_unpack(&ucred->luc_cap, body->capability);
477 mds_root_squash(mds, &req->rq_peer.nid, &body->fsuid,
478 &body->fsgid, &ucred->luc_cap,
479 &body->suppgid, NULL);
481 ucred->luc_fsuid = body->fsuid;
482 ucred->luc_fsgid = body->fsgid;
485 ucred->luc_uce = upcall_cache_get_entry(mds->mds_group_hash,
489 if (IS_ERR(ucred->luc_uce)) {
490 rc = PTR_ERR(ucred->luc_uce);
491 ucred->luc_uce = NULL;
497 ucred->luc_fsgid = ucred->luc_uce->ue_primary;
503 void mds_exit_ucred(struct lvfs_ucred *ucred, struct mds_obd *mds)
505 upcall_cache_put_entry(mds->mds_group_hash, ucred->luc_uce);