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/selinux.h>
42 #define DEBUG_SUBSYSTEM S_LLITE
44 #include <obd_support.h>
45 #include <lustre_dlm.h>
46 #include <lustre_ver.h>
47 #include <lustre_eacl.h>
49 #include "llite_internal.h"
51 #define XATTR_USER_T (1)
52 #define XATTR_TRUSTED_T (2)
53 #define XATTR_SECURITY_T (3)
54 #define XATTR_ACL_ACCESS_T (4)
55 #define XATTR_ACL_DEFAULT_T (5)
56 #define XATTR_LUSTRE_T (6)
57 #define XATTR_OTHER_T (7)
60 int get_xattr_type(const char *name)
62 if (!strcmp(name, POSIX_ACL_XATTR_ACCESS))
63 return XATTR_ACL_ACCESS_T;
65 if (!strcmp(name, POSIX_ACL_XATTR_DEFAULT))
66 return XATTR_ACL_DEFAULT_T;
68 if (!strncmp(name, XATTR_USER_PREFIX,
69 sizeof(XATTR_USER_PREFIX) - 1))
72 if (!strncmp(name, XATTR_TRUSTED_PREFIX,
73 sizeof(XATTR_TRUSTED_PREFIX) - 1))
74 return XATTR_TRUSTED_T;
76 if (!strncmp(name, XATTR_SECURITY_PREFIX,
77 sizeof(XATTR_SECURITY_PREFIX) - 1))
78 return XATTR_SECURITY_T;
80 if (!strncmp(name, XATTR_LUSTRE_PREFIX,
81 sizeof(XATTR_LUSTRE_PREFIX) - 1))
82 return XATTR_LUSTRE_T;
88 int xattr_type_filter(struct ll_sb_info *sbi, int xattr_type)
90 if ((xattr_type == XATTR_ACL_ACCESS_T ||
91 xattr_type == XATTR_ACL_DEFAULT_T) &&
92 !(sbi->ll_flags & LL_SBI_ACL))
95 if (xattr_type == XATTR_USER_T && !(sbi->ll_flags & LL_SBI_USER_XATTR))
97 if (xattr_type == XATTR_TRUSTED_T && !cfs_capable(CFS_CAP_SYS_ADMIN))
99 if (xattr_type == XATTR_OTHER_T)
105 /* xattr related to IMA(Integrity Measurement Architecture) */
106 #ifndef XATTR_NAME_IMA
107 #define XATTR_NAME_IMA "security.ima"
108 #define XATTR_NAME_EVM "security.evm"
112 int ll_setxattr_common(struct inode *inode, const char *name,
113 const void *value, size_t size,
114 int flags, __u64 valid)
116 struct ll_sb_info *sbi = ll_i2sbi(inode);
117 struct ptlrpc_request *req = NULL;
119 posix_acl_xattr_header *new_value = NULL;
120 struct rmtacl_ctl_entry *rce = NULL;
121 ext_acl_xattr_header *acl = NULL;
122 const char *pv = value;
125 /*FIXME: enable IMA when the conditions are ready */
126 if (strncmp(name, XATTR_NAME_IMA,
127 sizeof(XATTR_NAME_IMA)) == 0 ||
128 strncmp(name, XATTR_NAME_EVM,
129 sizeof(XATTR_NAME_EVM)) == 0)
132 xattr_type = get_xattr_type(name);
133 rc = xattr_type_filter(sbi, xattr_type);
137 if ((xattr_type == XATTR_ACL_ACCESS_T ||
138 xattr_type == XATTR_ACL_DEFAULT_T) &&
139 #ifdef HAVE_INODE_OWNER_OR_CAPABLE
140 !inode_owner_or_capable(inode))
142 !is_owner_or_cap(inode))
146 /* b10667: ignore lustre special xattr for now */
147 if (strcmp(name, XATTR_NAME_HSM) == 0 ||
148 (xattr_type == XATTR_TRUSTED_T &&
149 strcmp(name, XATTR_NAME_LOV) == 0) ||
150 (xattr_type == XATTR_LUSTRE_T &&
151 strcmp(name, "lustre.lov") == 0))
154 /* b15587: ignore security.capability xattr for now */
155 if ((xattr_type == XATTR_SECURITY_T &&
156 strcmp(name, "security.capability") == 0))
159 /* LU-549: Disable security.selinux when selinux is disabled */
160 if (xattr_type == XATTR_SECURITY_T && !selinux_is_enabled() &&
161 strcmp(name, "security.selinux") == 0)
164 #ifdef CONFIG_FS_POSIX_ACL
165 if (sbi->ll_flags & LL_SBI_RMT_CLIENT &&
166 (xattr_type == XATTR_ACL_ACCESS_T ||
167 xattr_type == XATTR_ACL_DEFAULT_T)) {
168 rce = rct_search(&sbi->ll_rct, current_pid());
170 (rce->rce_ops != RMT_LSETFACL &&
171 rce->rce_ops != RMT_RSETFACL))
174 if (rce->rce_ops == RMT_LSETFACL) {
175 struct eacl_entry *ee;
177 ee = et_search_del(&sbi->ll_et, current_pid(),
178 ll_inode2fid(inode), xattr_type);
180 if (valid & OBD_MD_FLXATTR) {
181 acl = lustre_acl_xattr_merge2ext(
182 (posix_acl_xattr_header *)value,
186 RETURN(PTR_ERR(acl));
188 size = CFS_ACL_XATTR_SIZE(\
189 le32_to_cpu(acl->a_count), \
191 pv = (const char *)acl;
194 } else if (rce->rce_ops == RMT_RSETFACL) {
195 int acl_size = lustre_posix_acl_xattr_filter(
196 (posix_acl_xattr_header *)value,
198 if (unlikely(acl_size < 0))
202 pv = (const char *)new_value;
206 valid |= rce_ops2valid(rce->rce_ops);
209 rc = md_setxattr(sbi->ll_md_exp, ll_inode2fid(inode), valid, name, pv,
210 size, 0, flags, ll_i2suppgid(inode), &req);
212 #ifdef CONFIG_FS_POSIX_ACL
213 if (new_value != NULL)
214 lustre_posix_acl_xattr_free(new_value, size);
216 lustre_ext_acl_xattr_free(acl);
219 if (rc == -EOPNOTSUPP && xattr_type == XATTR_USER_T) {
220 LCONSOLE_INFO("Disabling user_xattr feature because "
221 "it is not supported on the server\n");
222 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
227 ptlrpc_req_finished(req);
231 static int get_hsm_state(struct inode *inode, __u32 *hus_states)
233 struct md_op_data *op_data;
234 struct hsm_user_state *hus;
241 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
242 LUSTRE_OPC_ANY, hus);
243 if (!IS_ERR(op_data)) {
244 rc = obd_iocontrol(LL_IOC_HSM_STATE_GET, ll_i2mdexp(inode),
245 sizeof(*op_data), op_data, NULL);
247 *hus_states = hus->hus_states;
249 CDEBUG(D_VFSTRACE, "obd_iocontrol failed. rc = %d\n",
252 ll_finish_md_op_data(op_data);
254 rc = PTR_ERR(op_data);
255 CDEBUG(D_VFSTRACE, "Could not prepare the opdata. rc = %d\n",
262 int ll_setxattr(struct dentry *dentry, const char *name,
263 const void *value, size_t size, int flags)
265 struct inode *inode = dentry->d_inode;
270 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), xattr %s\n",
271 PFID(ll_inode2fid(inode)), inode, name);
273 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_SETXATTR, 1);
275 if ((strncmp(name, XATTR_TRUSTED_PREFIX,
276 sizeof(XATTR_TRUSTED_PREFIX) - 1) == 0 &&
277 strcmp(name + sizeof(XATTR_TRUSTED_PREFIX) - 1, "lov") == 0) ||
278 (strncmp(name, XATTR_LUSTRE_PREFIX,
279 sizeof(XATTR_LUSTRE_PREFIX) - 1) == 0 &&
280 strcmp(name + sizeof(XATTR_LUSTRE_PREFIX) - 1, "lov") == 0)) {
281 struct lov_user_md *lump = (struct lov_user_md *)value;
284 /* Attributes that are saved via getxattr will always have
285 * the stripe_offset as 0. Instead, the MDS should be
286 * allowed to pick the starting OST index. b=17846 */
287 if (lump != NULL && lump->lmm_stripe_offset == 0)
288 lump->lmm_stripe_offset = -1;
289 /* Avoid anyone directly setting the RELEASED flag. */
291 (lump->lmm_pattern & LOV_PATTERN_F_RELEASED)) {
292 /* Only if we have a released flag check if the file
293 * was indeed archived. */
294 __u32 state = HS_NONE;
295 rc = get_hsm_state(inode, &state);
298 if (!(state & HS_ARCHIVED)) {
300 "hus_states state = %x, pattern = %x\n",
301 state, lump->lmm_pattern);
302 /* Here the state is: real file is not
303 * archived but user is requesting to set
304 * the RELEASED flag so we mask off the
305 * released flag from the request */
306 lump->lmm_pattern ^= LOV_PATTERN_F_RELEASED;
310 if (lump != NULL && S_ISREG(inode->i_mode)) {
312 __u64 it_flags = FMODE_WRITE;
315 lum_size = ll_lov_user_md_size(lump);
316 if (lum_size < 0 || size < lum_size)
317 return 0; /* b=10667: ignore error */
319 memset(&f, 0, sizeof(f)); /* f.f_flags is used below */
320 f.f_path.dentry = dentry;
321 rc = ll_lov_setstripe_ea_info(inode, &f, it_flags, lump,
323 /* b=10667: rc always be 0 here for now */
325 } else if (S_ISDIR(inode->i_mode)) {
326 rc = ll_dir_setstripe(inode, lump, 0);
331 } else if (strcmp(name, XATTR_NAME_LMA) == 0 ||
332 strcmp(name, XATTR_NAME_LINK) == 0)
335 return ll_setxattr_common(inode, name, value, size, flags,
339 int ll_removexattr(struct dentry *dentry, const char *name)
341 struct inode *inode = dentry->d_inode;
346 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), xattr %s\n",
347 PFID(ll_inode2fid(inode)), inode, name);
349 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_REMOVEXATTR, 1);
350 return ll_setxattr_common(inode, name, NULL, 0, 0,
354 int ll_getxattr_common(struct inode *inode, const char *name,
355 void *buffer, size_t size, __u64 valid)
357 struct ll_sb_info *sbi = ll_i2sbi(inode);
358 struct ptlrpc_request *req = NULL;
359 struct mdt_body *body;
362 struct rmtacl_ctl_entry *rce = NULL;
363 struct ll_inode_info *lli = ll_i2info(inode);
366 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
367 PFID(ll_inode2fid(inode)), inode);
369 /* listxattr have slightly different behavior from of ext3:
370 * without 'user_xattr' ext3 will list all xattr names but
371 * filtered out "^user..*"; we list them all for simplicity.
374 xattr_type = XATTR_OTHER_T;
378 xattr_type = get_xattr_type(name);
379 rc = xattr_type_filter(sbi, xattr_type);
383 /* b15587: ignore security.capability xattr for now */
384 if ((xattr_type == XATTR_SECURITY_T &&
385 strcmp(name, "security.capability") == 0))
388 /* LU-549: Disable security.selinux when selinux is disabled */
389 if (xattr_type == XATTR_SECURITY_T && !selinux_is_enabled() &&
390 strcmp(name, "security.selinux") == 0)
393 #ifdef CONFIG_FS_POSIX_ACL
394 if (sbi->ll_flags & LL_SBI_RMT_CLIENT &&
395 (xattr_type == XATTR_ACL_ACCESS_T ||
396 xattr_type == XATTR_ACL_DEFAULT_T)) {
397 rce = rct_search(&sbi->ll_rct, current_pid());
399 (rce->rce_ops != RMT_LSETFACL &&
400 rce->rce_ops != RMT_LGETFACL &&
401 rce->rce_ops != RMT_RSETFACL &&
402 rce->rce_ops != RMT_RGETFACL))
406 /* posix acl is under protection of LOOKUP lock. when calling to this,
407 * we just have path resolution to the target inode, so we have great
408 * chance that cached ACL is uptodate.
410 if (xattr_type == XATTR_ACL_ACCESS_T &&
411 !(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
413 struct posix_acl *acl;
415 spin_lock(&lli->lli_lock);
416 acl = posix_acl_dup(lli->lli_posix_acl);
417 spin_unlock(&lli->lli_lock);
422 rc = posix_acl_to_xattr(&init_user_ns, acl, buffer, size);
423 posix_acl_release(acl);
426 if (xattr_type == XATTR_ACL_DEFAULT_T && !S_ISDIR(inode->i_mode))
431 if (sbi->ll_xattr_cache_enabled &&
432 xattr_type != XATTR_ACL_ACCESS_T &&
433 (xattr_type != XATTR_SECURITY_T ||
434 strcmp(name, "security.selinux") != 0)) {
435 rc = ll_xattr_cache_get(inode, name, buffer, size, valid);
437 goto getxattr_nocache;
441 /* Add "system.posix_acl_access" to the list */
442 if (lli->lli_posix_acl != NULL && valid & OBD_MD_FLXATTRLS) {
444 rc += sizeof(XATTR_NAME_ACL_ACCESS);
445 } else if (size - rc >= sizeof(XATTR_NAME_ACL_ACCESS)) {
446 memcpy(buffer + rc, XATTR_NAME_ACL_ACCESS,
447 sizeof(XATTR_NAME_ACL_ACCESS));
448 rc += sizeof(XATTR_NAME_ACL_ACCESS);
450 GOTO(out_xattr, rc = -ERANGE);
455 rc = md_getxattr(sbi->ll_md_exp, ll_inode2fid(inode),
456 valid | (rce ? rce_ops2valid(rce->rce_ops) : 0),
457 name, NULL, 0, size, 0, &req);
462 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
465 /* only detect the xattr size */
467 GOTO(out, rc = body->mbo_eadatasize);
469 if (size < body->mbo_eadatasize) {
470 CERROR("server bug: replied size %u > %u\n",
471 body->mbo_eadatasize, (int)size);
472 GOTO(out, rc = -ERANGE);
475 if (body->mbo_eadatasize == 0)
476 GOTO(out, rc = -ENODATA);
478 /* do not need swab xattr data */
479 xdata = req_capsule_server_sized_get(&req->rq_pill, &RMF_EADATA,
480 body->mbo_eadatasize);
482 GOTO(out, rc = -EFAULT);
484 memcpy(buffer, xdata, body->mbo_eadatasize);
485 rc = body->mbo_eadatasize;
488 #ifdef CONFIG_FS_POSIX_ACL
489 if (rce != NULL && rce->rce_ops == RMT_LSETFACL) {
490 ext_acl_xattr_header *acl;
492 acl = lustre_posix_acl_xattr_2ext(buffer, rc);
494 GOTO(out, rc = PTR_ERR(acl));
496 rc = ee_add(&sbi->ll_et, current_pid(), ll_inode2fid(inode),
498 if (unlikely(rc < 0)) {
499 lustre_ext_acl_xattr_free(acl);
507 if (rc == -EOPNOTSUPP && xattr_type == XATTR_USER_T) {
508 LCONSOLE_INFO("%s: disabling user_xattr feature because "
509 "it is not supported on the server: rc = %d\n",
510 ll_get_fsname(inode->i_sb, NULL, 0), rc);
511 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
514 ptlrpc_req_finished(req);
518 static ssize_t ll_getxattr_lov(struct inode *inode, void *buf, size_t buf_size)
522 if (S_ISREG(inode->i_mode)) {
523 struct cl_object *obj = ll_i2info(inode)->lli_clob;
525 struct cl_layout cl = {
526 .cl_buf.lb_buf = buf,
527 .cl_buf.lb_len = buf_size,
534 env = cl_env_get(&refcheck);
536 RETURN(PTR_ERR(env));
538 rc = cl_object_layout_get(env, obj, &cl);
543 GOTO(out_env, rc = -ENODATA);
550 LASSERT(buf != NULL && rc <= buf_size);
552 /* Do not return layout gen for getxattr() since
553 * otherwise it would confuse tar --xattr by
554 * recognizing layout gen as stripe offset when the
555 * file is restored. See LU-2809. */
556 ((struct lov_mds_md *)buf)->lmm_layout_gen = 0;
558 cl_env_put(env, &refcheck);
561 } else if (S_ISDIR(inode->i_mode)) {
562 struct lov_mds_md *lmm = NULL;
564 struct ptlrpc_request *req = NULL;
566 rc = ll_dir_getstripe(inode, (void **)&lmm, &lmm_size,
572 GOTO(out_req, rc = lmm_size);
574 if (buf_size < lmm_size)
575 GOTO(out_req, rc = -ERANGE);
577 memcpy(buf, lmm, lmm_size);
578 GOTO(out_req, rc = lmm_size);
581 ptlrpc_req_finished(req);
589 ssize_t ll_getxattr(struct dentry *dentry, const char *name, void *buf,
592 struct inode *inode = dentry->d_inode;
597 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), xattr %s\n",
598 PFID(ll_inode2fid(inode)), inode, name);
600 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_GETXATTR, 1);
602 if (strcmp(name, XATTR_LUSTRE_LOV) == 0 ||
603 strcmp(name, XATTR_NAME_LOV) == 0)
604 return ll_getxattr_lov(inode, buf, buf_size);
606 return ll_getxattr_common(inode, name, buf, buf_size,
610 ssize_t ll_listxattr(struct dentry *dentry, char *buf, size_t buf_size)
612 struct inode *inode = dentry->d_inode;
613 struct ll_sb_info *sbi = ll_i2sbi(inode);
618 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
619 PFID(ll_inode2fid(inode)), inode);
621 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_LISTXATTR, 1);
623 rc = ll_getxattr_common(inode, NULL, buf, buf_size, OBD_MD_FLXATTRLS);
627 /* If we're being called to get the size of the xattr list
628 * (buf_size == 0) then just assume that a lustre.lov xattr
631 RETURN(rc + sizeof(XATTR_LUSTRE_LOV));
637 len = strnlen(xattr_name, rem - 1) + 1;
639 if (xattr_type_filter(sbi, get_xattr_type(xattr_name)) == 0) {
640 /* Skip OK xattr type, leave it in buffer. */
645 /* Move up remaining xattrs in buffer removing the
646 * xattr that is not OK. */
647 memmove(xattr_name, xattr_name + len, rem);
651 rc2 = ll_getxattr_lov(inode, NULL, 0);
658 if (buf_size < rc + sizeof(XATTR_LUSTRE_LOV))
661 memcpy(buf + rc, XATTR_LUSTRE_LOV, sizeof(XATTR_LUSTRE_LOV));
663 RETURN(rc + sizeof(XATTR_LUSTRE_LOV));