4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2014, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
38 #include <linux/sched.h>
40 #include <linux/xattr.h>
41 #include <linux/selinux.h>
43 #define DEBUG_SUBSYSTEM S_LLITE
45 #include <obd_support.h>
46 #include <lustre_dlm.h>
47 #include <lustre_ver.h>
48 #include <lustre_eacl.h>
50 #include "llite_internal.h"
52 /* xattr related to IMA(Integrity Measurement Architecture) */
53 #ifndef XATTR_NAME_IMA
54 #define XATTR_NAME_IMA "security.ima"
56 #ifndef XATTR_NAME_EVM
57 #define XATTR_NAME_EVM "security.evm"
60 #define XATTR_USER_T (1)
61 #define XATTR_TRUSTED_T (2)
62 #define XATTR_SECURITY_T (3)
63 #define XATTR_ACL_ACCESS_T (4)
64 #define XATTR_ACL_DEFAULT_T (5)
65 #define XATTR_LUSTRE_T (6)
66 #define XATTR_OTHER_T (7)
69 int get_xattr_type(const char *name)
71 if (!strcmp(name, POSIX_ACL_XATTR_ACCESS))
72 return XATTR_ACL_ACCESS_T;
74 if (!strcmp(name, POSIX_ACL_XATTR_DEFAULT))
75 return XATTR_ACL_DEFAULT_T;
77 if (!strncmp(name, XATTR_USER_PREFIX,
78 sizeof(XATTR_USER_PREFIX) - 1))
81 if (!strncmp(name, XATTR_TRUSTED_PREFIX,
82 sizeof(XATTR_TRUSTED_PREFIX) - 1))
83 return XATTR_TRUSTED_T;
85 if (!strncmp(name, XATTR_SECURITY_PREFIX,
86 sizeof(XATTR_SECURITY_PREFIX) - 1))
87 return XATTR_SECURITY_T;
89 if (!strncmp(name, XATTR_LUSTRE_PREFIX,
90 sizeof(XATTR_LUSTRE_PREFIX) - 1))
91 return XATTR_LUSTRE_T;
97 int xattr_type_filter(struct ll_sb_info *sbi, int xattr_type)
99 if ((xattr_type == XATTR_ACL_ACCESS_T ||
100 xattr_type == XATTR_ACL_DEFAULT_T) &&
101 !(sbi->ll_flags & LL_SBI_ACL))
104 if (xattr_type == XATTR_USER_T && !(sbi->ll_flags & LL_SBI_USER_XATTR))
106 if (xattr_type == XATTR_TRUSTED_T && !cfs_capable(CFS_CAP_SYS_ADMIN))
108 if (xattr_type == XATTR_OTHER_T)
115 int ll_setxattr_common(struct inode *inode, const char *name,
116 const void *value, size_t size,
117 int flags, __u64 valid)
119 struct ll_sb_info *sbi = ll_i2sbi(inode);
120 struct ptlrpc_request *req = NULL;
122 posix_acl_xattr_header *new_value = NULL;
123 struct rmtacl_ctl_entry *rce = NULL;
124 ext_acl_xattr_header *acl = NULL;
125 const char *pv = value;
128 /*FIXME: enable IMA when the conditions are ready */
129 if (strncmp(name, XATTR_NAME_IMA,
130 sizeof(XATTR_NAME_IMA)) == 0 ||
131 strncmp(name, XATTR_NAME_EVM,
132 sizeof(XATTR_NAME_EVM)) == 0)
135 xattr_type = get_xattr_type(name);
136 rc = xattr_type_filter(sbi, xattr_type);
140 if ((xattr_type == XATTR_ACL_ACCESS_T ||
141 xattr_type == XATTR_ACL_DEFAULT_T) &&
142 #ifdef HAVE_INODE_OWNER_OR_CAPABLE
143 !inode_owner_or_capable(inode))
145 !is_owner_or_cap(inode))
149 /* b10667: ignore lustre special xattr for now */
150 if (strcmp(name, XATTR_NAME_HSM) == 0 ||
151 (xattr_type == XATTR_TRUSTED_T &&
152 strcmp(name, XATTR_NAME_LOV) == 0) ||
153 (xattr_type == XATTR_LUSTRE_T &&
154 strcmp(name, "lustre.lov") == 0))
157 /* b15587: ignore security.capability xattr for now */
158 if ((xattr_type == XATTR_SECURITY_T &&
159 strcmp(name, "security.capability") == 0))
162 /* LU-549: Disable security.selinux when selinux is disabled */
163 if (xattr_type == XATTR_SECURITY_T && !selinux_is_enabled() &&
164 strcmp(name, "security.selinux") == 0)
167 #ifdef CONFIG_FS_POSIX_ACL
168 if (sbi->ll_flags & LL_SBI_RMT_CLIENT &&
169 (xattr_type == XATTR_ACL_ACCESS_T ||
170 xattr_type == XATTR_ACL_DEFAULT_T)) {
171 rce = rct_search(&sbi->ll_rct, current_pid());
173 (rce->rce_ops != RMT_LSETFACL &&
174 rce->rce_ops != RMT_RSETFACL))
177 if (rce->rce_ops == RMT_LSETFACL) {
178 struct eacl_entry *ee;
180 ee = et_search_del(&sbi->ll_et, current_pid(),
181 ll_inode2fid(inode), xattr_type);
183 if (valid & OBD_MD_FLXATTR) {
184 acl = lustre_acl_xattr_merge2ext(
185 (posix_acl_xattr_header *)value,
189 RETURN(PTR_ERR(acl));
191 size = CFS_ACL_XATTR_SIZE(\
192 le32_to_cpu(acl->a_count), \
194 pv = (const char *)acl;
197 } else if (rce->rce_ops == RMT_RSETFACL) {
198 int acl_size = lustre_posix_acl_xattr_filter(
199 (posix_acl_xattr_header *)value,
201 if (unlikely(acl_size < 0))
205 pv = (const char *)new_value;
209 valid |= rce_ops2valid(rce->rce_ops);
212 rc = md_setxattr(sbi->ll_md_exp, ll_inode2fid(inode), valid, name, pv,
213 size, 0, flags, ll_i2suppgid(inode), &req);
215 #ifdef CONFIG_FS_POSIX_ACL
216 if (new_value != NULL)
217 lustre_posix_acl_xattr_free(new_value, size);
219 lustre_ext_acl_xattr_free(acl);
222 if (rc == -EOPNOTSUPP && xattr_type == XATTR_USER_T) {
223 LCONSOLE_INFO("Disabling user_xattr feature because "
224 "it is not supported on the server\n");
225 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
230 ptlrpc_req_finished(req);
234 static int get_hsm_state(struct inode *inode, __u32 *hus_states)
236 struct md_op_data *op_data;
237 struct hsm_user_state *hus;
244 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
245 LUSTRE_OPC_ANY, hus);
246 if (!IS_ERR(op_data)) {
247 rc = obd_iocontrol(LL_IOC_HSM_STATE_GET, ll_i2mdexp(inode),
248 sizeof(*op_data), op_data, NULL);
250 *hus_states = hus->hus_states;
252 CDEBUG(D_VFSTRACE, "obd_iocontrol failed. rc = %d\n",
255 ll_finish_md_op_data(op_data);
257 rc = PTR_ERR(op_data);
258 CDEBUG(D_VFSTRACE, "Could not prepare the opdata. rc = %d\n",
265 int ll_setxattr(struct dentry *dentry, const char *name,
266 const void *value, size_t size, int flags)
268 struct inode *inode = dentry->d_inode;
273 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), xattr %s\n",
274 PFID(ll_inode2fid(inode)), inode, name);
276 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_SETXATTR, 1);
278 if ((strncmp(name, XATTR_TRUSTED_PREFIX,
279 sizeof(XATTR_TRUSTED_PREFIX) - 1) == 0 &&
280 strcmp(name + sizeof(XATTR_TRUSTED_PREFIX) - 1, "lov") == 0) ||
281 (strncmp(name, XATTR_LUSTRE_PREFIX,
282 sizeof(XATTR_LUSTRE_PREFIX) - 1) == 0 &&
283 strcmp(name + sizeof(XATTR_LUSTRE_PREFIX) - 1, "lov") == 0)) {
284 struct lov_user_md *lump = (struct lov_user_md *)value;
287 /* Attributes that are saved via getxattr will always have
288 * the stripe_offset as 0. Instead, the MDS should be
289 * allowed to pick the starting OST index. b=17846 */
290 if (lump != NULL && lump->lmm_stripe_offset == 0)
291 lump->lmm_stripe_offset = -1;
292 /* Avoid anyone directly setting the RELEASED flag. */
294 (lump->lmm_pattern & LOV_PATTERN_F_RELEASED)) {
295 /* Only if we have a released flag check if the file
296 * was indeed archived. */
297 __u32 state = HS_NONE;
298 rc = get_hsm_state(inode, &state);
301 if (!(state & HS_ARCHIVED)) {
303 "hus_states state = %x, pattern = %x\n",
304 state, lump->lmm_pattern);
305 /* Here the state is: real file is not
306 * archived but user is requesting to set
307 * the RELEASED flag so we mask off the
308 * released flag from the request */
309 lump->lmm_pattern ^= LOV_PATTERN_F_RELEASED;
313 if (lump != NULL && S_ISREG(inode->i_mode)) {
315 __u64 it_flags = FMODE_WRITE;
318 lum_size = ll_lov_user_md_size(lump);
319 if (lum_size < 0 || size < lum_size)
320 return 0; /* b=10667: ignore error */
322 memset(&f, 0, sizeof(f)); /* f.f_flags is used below */
323 f.f_path.dentry = dentry;
324 rc = ll_lov_setstripe_ea_info(inode, &f, it_flags, lump,
326 /* b=10667: rc always be 0 here for now */
328 } else if (S_ISDIR(inode->i_mode)) {
329 rc = ll_dir_setstripe(inode, lump, 0);
334 } else if (strcmp(name, XATTR_NAME_LMA) == 0 ||
335 strcmp(name, XATTR_NAME_LINK) == 0)
338 return ll_setxattr_common(inode, name, value, size, flags,
342 int ll_removexattr(struct dentry *dentry, const char *name)
344 struct inode *inode = dentry->d_inode;
349 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), xattr %s\n",
350 PFID(ll_inode2fid(inode)), inode, name);
352 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_REMOVEXATTR, 1);
353 return ll_setxattr_common(inode, name, NULL, 0, 0,
357 int ll_getxattr_common(struct inode *inode, const char *name,
358 void *buffer, size_t size, __u64 valid)
360 struct ll_sb_info *sbi = ll_i2sbi(inode);
361 struct ptlrpc_request *req = NULL;
362 struct mdt_body *body;
365 struct rmtacl_ctl_entry *rce = NULL;
366 struct ll_inode_info *lli = ll_i2info(inode);
369 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
370 PFID(ll_inode2fid(inode)), inode);
372 /* listxattr have slightly different behavior from of ext3:
373 * without 'user_xattr' ext3 will list all xattr names but
374 * filtered out "^user..*"; we list them all for simplicity.
377 xattr_type = XATTR_OTHER_T;
381 xattr_type = get_xattr_type(name);
382 rc = xattr_type_filter(sbi, xattr_type);
386 /* b15587: ignore security.capability xattr for now */
387 if ((xattr_type == XATTR_SECURITY_T &&
388 strcmp(name, "security.capability") == 0))
391 /* LU-549: Disable security.selinux when selinux is disabled */
392 if (xattr_type == XATTR_SECURITY_T && !selinux_is_enabled() &&
393 strcmp(name, "security.selinux") == 0)
396 #ifdef CONFIG_FS_POSIX_ACL
397 if (sbi->ll_flags & LL_SBI_RMT_CLIENT &&
398 (xattr_type == XATTR_ACL_ACCESS_T ||
399 xattr_type == XATTR_ACL_DEFAULT_T)) {
400 rce = rct_search(&sbi->ll_rct, current_pid());
402 (rce->rce_ops != RMT_LSETFACL &&
403 rce->rce_ops != RMT_LGETFACL &&
404 rce->rce_ops != RMT_RSETFACL &&
405 rce->rce_ops != RMT_RGETFACL))
409 /* posix acl is under protection of LOOKUP lock. when calling to this,
410 * we just have path resolution to the target inode, so we have great
411 * chance that cached ACL is uptodate.
413 if (xattr_type == XATTR_ACL_ACCESS_T &&
414 !(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
416 struct posix_acl *acl;
418 spin_lock(&lli->lli_lock);
419 acl = posix_acl_dup(lli->lli_posix_acl);
420 spin_unlock(&lli->lli_lock);
425 rc = posix_acl_to_xattr(&init_user_ns, acl, buffer, size);
426 posix_acl_release(acl);
429 if (xattr_type == XATTR_ACL_DEFAULT_T && !S_ISDIR(inode->i_mode))
434 if (sbi->ll_xattr_cache_enabled &&
435 xattr_type != XATTR_ACL_ACCESS_T &&
436 (xattr_type != XATTR_SECURITY_T ||
437 strcmp(name, "security.selinux") != 0)) {
438 rc = ll_xattr_cache_get(inode, name, buffer, size, valid);
440 goto getxattr_nocache;
444 /* Add "system.posix_acl_access" to the list */
445 if (lli->lli_posix_acl != NULL && valid & OBD_MD_FLXATTRLS) {
447 rc += sizeof(XATTR_NAME_ACL_ACCESS);
448 } else if (size - rc >= sizeof(XATTR_NAME_ACL_ACCESS)) {
449 memcpy(buffer + rc, XATTR_NAME_ACL_ACCESS,
450 sizeof(XATTR_NAME_ACL_ACCESS));
451 rc += sizeof(XATTR_NAME_ACL_ACCESS);
453 GOTO(out_xattr, rc = -ERANGE);
458 rc = md_getxattr(sbi->ll_md_exp, ll_inode2fid(inode),
459 valid | (rce ? rce_ops2valid(rce->rce_ops) : 0),
460 name, NULL, 0, size, 0, &req);
465 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
468 /* only detect the xattr size */
470 GOTO(out, rc = body->mbo_eadatasize);
472 if (size < body->mbo_eadatasize) {
473 CERROR("server bug: replied size %u > %u\n",
474 body->mbo_eadatasize, (int)size);
475 GOTO(out, rc = -ERANGE);
478 if (body->mbo_eadatasize == 0)
479 GOTO(out, rc = -ENODATA);
481 /* do not need swab xattr data */
482 xdata = req_capsule_server_sized_get(&req->rq_pill, &RMF_EADATA,
483 body->mbo_eadatasize);
485 GOTO(out, rc = -EFAULT);
487 memcpy(buffer, xdata, body->mbo_eadatasize);
488 rc = body->mbo_eadatasize;
491 #ifdef CONFIG_FS_POSIX_ACL
492 if (rce != NULL && rce->rce_ops == RMT_LSETFACL) {
493 ext_acl_xattr_header *acl;
495 acl = lustre_posix_acl_xattr_2ext(buffer, rc);
497 GOTO(out, rc = PTR_ERR(acl));
499 rc = ee_add(&sbi->ll_et, current_pid(), ll_inode2fid(inode),
501 if (unlikely(rc < 0)) {
502 lustre_ext_acl_xattr_free(acl);
510 if (rc == -EOPNOTSUPP && xattr_type == XATTR_USER_T) {
511 LCONSOLE_INFO("%s: disabling user_xattr feature because "
512 "it is not supported on the server: rc = %d\n",
513 ll_get_fsname(inode->i_sb, NULL, 0), rc);
514 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
517 ptlrpc_req_finished(req);
521 static ssize_t ll_getxattr_lov(struct inode *inode, void *buf, size_t buf_size)
525 if (S_ISREG(inode->i_mode)) {
526 struct cl_object *obj = ll_i2info(inode)->lli_clob;
528 struct cl_layout cl = {
529 .cl_buf.lb_buf = buf,
530 .cl_buf.lb_len = buf_size,
537 env = cl_env_get(&refcheck);
539 RETURN(PTR_ERR(env));
541 rc = cl_object_layout_get(env, obj, &cl);
546 GOTO(out_env, rc = -ENODATA);
553 LASSERT(buf != NULL && rc <= buf_size);
555 /* Do not return layout gen for getxattr() since
556 * otherwise it would confuse tar --xattr by
557 * recognizing layout gen as stripe offset when the
558 * file is restored. See LU-2809. */
559 ((struct lov_mds_md *)buf)->lmm_layout_gen = 0;
561 cl_env_put(env, &refcheck);
564 } else if (S_ISDIR(inode->i_mode)) {
565 struct lov_mds_md *lmm = NULL;
567 struct ptlrpc_request *req = NULL;
569 rc = ll_dir_getstripe(inode, (void **)&lmm, &lmm_size,
575 GOTO(out_req, rc = lmm_size);
577 if (buf_size < lmm_size)
578 GOTO(out_req, rc = -ERANGE);
580 memcpy(buf, lmm, lmm_size);
581 GOTO(out_req, rc = lmm_size);
584 ptlrpc_req_finished(req);
592 ssize_t ll_getxattr(struct dentry *dentry, const char *name, void *buf,
595 struct inode *inode = dentry->d_inode;
600 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), xattr %s\n",
601 PFID(ll_inode2fid(inode)), inode, name);
603 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_GETXATTR, 1);
605 if (strcmp(name, XATTR_LUSTRE_LOV) == 0 ||
606 strcmp(name, XATTR_NAME_LOV) == 0)
607 return ll_getxattr_lov(inode, buf, buf_size);
609 return ll_getxattr_common(inode, name, buf, buf_size,
613 ssize_t ll_listxattr(struct dentry *dentry, char *buf, size_t buf_size)
615 struct inode *inode = dentry->d_inode;
616 struct ll_sb_info *sbi = ll_i2sbi(inode);
621 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
622 PFID(ll_inode2fid(inode)), inode);
624 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_LISTXATTR, 1);
626 rc = ll_getxattr_common(inode, NULL, buf, buf_size, OBD_MD_FLXATTRLS);
630 /* If we're being called to get the size of the xattr list
631 * (buf_size == 0) then just assume that a lustre.lov xattr
634 RETURN(rc + sizeof(XATTR_LUSTRE_LOV));
640 len = strnlen(xattr_name, rem - 1) + 1;
642 if (xattr_type_filter(sbi, get_xattr_type(xattr_name)) == 0) {
643 /* Skip OK xattr type, leave it in buffer. */
648 /* Move up remaining xattrs in buffer removing the
649 * xattr that is not OK. */
650 memmove(xattr_name, xattr_name + len, rem);
654 rc2 = ll_getxattr_lov(inode, NULL, 0);
661 if (buf_size < rc + sizeof(XATTR_LUSTRE_LOV))
664 memcpy(buf + rc, XATTR_LUSTRE_LOV, sizeof(XATTR_LUSTRE_LOV));
666 RETURN(rc + sizeof(XATTR_LUSTRE_LOV));