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.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2016, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lustre/llite/llite_lib.c
34 * Lustre Light Super operations
37 #define DEBUG_SUBSYSTEM S_LLITE
39 #include <linux/module.h>
40 #include <linux/statfs.h>
41 #include <linux/time.h>
42 #include <linux/types.h>
43 #include <linux/version.h>
45 #include <linux/user_namespace.h>
46 #ifdef HAVE_UIDGID_HEADER
47 # include <linux/uidgid.h>
49 #include <linux/security.h>
51 #include <uapi/linux/lustre_ioctl.h>
52 #include <lustre_ha.h>
53 #include <lustre_dlm.h>
54 #include <lprocfs_status.h>
55 #include <lustre_disk.h>
56 #include <uapi/linux/lustre_param.h>
57 #include <lustre_log.h>
58 #include <cl_object.h>
59 #include <obd_cksum.h>
60 #include "llite_internal.h"
62 struct kmem_cache *ll_file_data_slab;
65 #define log2(n) ffz(~(n))
68 static struct ll_sb_info *ll_init_sbi(void)
70 struct ll_sb_info *sbi = NULL;
72 unsigned long lru_page_max;
82 spin_lock_init(&sbi->ll_lock);
83 mutex_init(&sbi->ll_lco.lco_lock);
84 spin_lock_init(&sbi->ll_pp_extent_lock);
85 spin_lock_init(&sbi->ll_process_lock);
86 sbi->ll_rw_stats_on = 0;
89 pages = si.totalram - si.totalhigh;
90 lru_page_max = pages / 2;
92 /* initialize ll_cache data */
93 sbi->ll_cache = cl_cache_init(lru_page_max);
94 if (sbi->ll_cache == NULL) {
95 OBD_FREE(sbi, sizeof(*sbi));
99 sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
100 SBI_DEFAULT_READAHEAD_MAX);
101 sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
102 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
103 SBI_DEFAULT_READAHEAD_WHOLE_MAX;
105 ll_generate_random_uuid(uuid);
106 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
107 CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
109 sbi->ll_flags |= LL_SBI_VERBOSE;
110 #ifdef ENABLE_CHECKSUM
111 sbi->ll_flags |= LL_SBI_CHECKSUM;
114 #ifdef HAVE_LRU_RESIZE_SUPPORT
115 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
117 sbi->ll_flags |= LL_SBI_LAZYSTATFS;
119 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
120 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
122 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
126 /* metadata statahead is enabled by default */
127 sbi->ll_sa_max = LL_SA_RPC_DEF;
128 atomic_set(&sbi->ll_sa_total, 0);
129 atomic_set(&sbi->ll_sa_wrong, 0);
130 atomic_set(&sbi->ll_sa_running, 0);
131 atomic_set(&sbi->ll_agl_total, 0);
132 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
133 sbi->ll_flags |= LL_SBI_FAST_READ;
136 sbi->ll_squash.rsi_uid = 0;
137 sbi->ll_squash.rsi_gid = 0;
138 INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
139 init_rwsem(&sbi->ll_squash.rsi_sem);
144 static void ll_free_sbi(struct super_block *sb)
146 struct ll_sb_info *sbi = ll_s2sbi(sb);
150 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
151 cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
152 if (sbi->ll_cache != NULL) {
153 cl_cache_decref(sbi->ll_cache);
154 sbi->ll_cache = NULL;
156 OBD_FREE(sbi, sizeof(*sbi));
161 static inline int obd_connect_has_secctx(struct obd_connect_data *data)
163 return data->ocd_connect_flags & OBD_CONNECT_FLAGS2 &&
164 data->ocd_connect_flags2 & OBD_CONNECT2_FILE_SECCTX;
167 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
168 struct vfsmount *mnt)
170 struct inode *root = NULL;
171 struct ll_sb_info *sbi = ll_s2sbi(sb);
172 struct obd_device *obd;
173 struct obd_statfs *osfs = NULL;
174 struct ptlrpc_request *request = NULL;
175 struct obd_connect_data *data = NULL;
176 struct obd_uuid *uuid;
177 struct md_op_data *op_data;
178 struct lustre_md lmd;
180 int size, err, checksum;
183 obd = class_name2obd(md);
185 CERROR("MD %s: not setup or attached\n", md);
199 /* indicate the features supported by this client */
200 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
201 OBD_CONNECT_ATTRFID |
202 OBD_CONNECT_VERSION | OBD_CONNECT_BRW_SIZE |
203 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
204 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
205 OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 |
206 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
207 OBD_CONNECT_64BITHASH |
208 OBD_CONNECT_EINPROGRESS |
209 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
210 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS|
211 OBD_CONNECT_MAX_EASIZE |
212 OBD_CONNECT_FLOCK_DEAD |
213 OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
214 OBD_CONNECT_OPEN_BY_FID |
215 OBD_CONNECT_DIR_STRIPE |
216 OBD_CONNECT_BULK_MBITS |
217 OBD_CONNECT_SUBTREE |
218 OBD_CONNECT_FLAGS2 | OBD_CONNECT_MULTIMODRPCS;
220 data->ocd_connect_flags2 = 0;
222 #ifdef HAVE_LRU_RESIZE_SUPPORT
223 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
224 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
226 #ifdef CONFIG_FS_POSIX_ACL
227 data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK |
228 OBD_CONNECT_LARGE_ACL;
231 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
232 /* flag mdc connection as lightweight, only used for test
233 * purpose, use with care */
234 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
236 data->ocd_ibits_known = MDS_INODELOCK_FULL;
237 data->ocd_version = LUSTRE_VERSION_CODE;
239 if (sb->s_flags & MS_RDONLY)
240 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
241 if (sbi->ll_flags & LL_SBI_USER_XATTR)
242 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
245 /* Setting this indicates we correctly support S_NOSEC (See kernel
246 * commit 9e1f1de02c2275d7172e18dc4e7c2065777611bf)
248 sb->s_flags |= MS_NOSEC;
251 if (sbi->ll_flags & LL_SBI_FLOCK)
252 sbi->ll_fop = &ll_file_operations_flock;
253 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
254 sbi->ll_fop = &ll_file_operations;
256 sbi->ll_fop = &ll_file_operations_noflock;
258 /* always ping even if server suppress_pings */
259 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
260 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
262 #ifdef HAVE_SECURITY_DENTRY_INIT_SECURITY
263 data->ocd_connect_flags2 |= OBD_CONNECT2_FILE_SECCTX;
264 #endif /* HAVE_SECURITY_DENTRY_INIT_SECURITY */
266 data->ocd_brw_size = MD_MAX_BRW_SIZE;
268 err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
270 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
271 "recovery, of which this client is not a "
272 "part. Please wait for recovery to complete,"
273 " abort, or time out.\n", md);
276 CERROR("cannot connect to %s: rc = %d\n", md, err);
280 sbi->ll_md_exp->exp_connect_data = *data;
282 err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
283 LUSTRE_SEQ_METADATA);
285 CERROR("%s: Can't init metadata layer FID infrastructure, "
286 "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
290 /* For mount, we only need fs info from MDT0, and also in DNE, it
291 * can make sure the client can be mounted as long as MDT0 is
293 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
294 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
295 OBD_STATFS_FOR_MDT0);
297 GOTO(out_md_fid, err);
299 /* This needs to be after statfs to ensure connect has finished.
300 * Note that "data" does NOT contain the valid connect reply.
301 * If connecting to a 1.8 server there will be no LMV device, so
302 * we can access the MDC export directly and exp_connect_flags will
303 * be non-zero, but if accessing an upgraded 2.1 server it will
304 * have the correct flags filled in.
305 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
306 valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
307 if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
308 valid != CLIENT_CONNECT_MDT_REQD) {
311 OBD_ALLOC_WAIT(buf, PAGE_SIZE);
312 obd_connect_flags2str(buf, PAGE_SIZE,
313 valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
314 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
315 "feature(s) needed for correct operation "
316 "of this client (%s). Please upgrade "
317 "server or downgrade client.\n",
318 sbi->ll_md_exp->exp_obd->obd_name, buf);
319 OBD_FREE(buf, PAGE_SIZE);
320 GOTO(out_md_fid, err = -EPROTO);
323 size = sizeof(*data);
324 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
325 KEY_CONN_DATA, &size, data);
327 CERROR("%s: Get connect data failed: rc = %d\n",
328 sbi->ll_md_exp->exp_obd->obd_name, err);
329 GOTO(out_md_fid, err);
332 LASSERT(osfs->os_bsize);
333 sb->s_blocksize = osfs->os_bsize;
334 sb->s_blocksize_bits = log2(osfs->os_bsize);
335 sb->s_magic = LL_SUPER_MAGIC;
336 sb->s_maxbytes = MAX_LFS_FILESIZE;
337 sbi->ll_namelen = osfs->os_namelen;
338 sbi->ll_mnt.mnt = current->fs->root.mnt;
340 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
341 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
342 LCONSOLE_INFO("Disabling user_xattr feature because "
343 "it is not supported on the server\n");
344 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
347 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
349 sb->s_flags |= MS_POSIXACL;
351 sbi->ll_flags |= LL_SBI_ACL;
353 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
355 sb->s_flags &= ~MS_POSIXACL;
357 sbi->ll_flags &= ~LL_SBI_ACL;
360 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
361 sbi->ll_flags |= LL_SBI_64BIT_HASH;
363 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
364 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
366 if (obd_connect_has_secctx(data))
367 sbi->ll_flags |= LL_SBI_FILE_SECCTX;
369 if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
370 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
371 LCONSOLE_INFO("%s: disabling xattr cache due to "
372 "unknown maximum xattr size.\n", dt);
373 } else if (!sbi->ll_xattr_cache_set) {
374 /* If xattr_cache is already set (no matter 0 or 1)
375 * during processing llog, it won't be enabled here. */
376 sbi->ll_flags |= LL_SBI_XATTR_CACHE;
377 sbi->ll_xattr_cache_enabled = 1;
381 obd = class_name2obd(dt);
383 CERROR("DT %s: not setup or attached\n", dt);
384 GOTO(out_md_fid, err = -ENODEV);
387 /* pass client page size via ocd_grant_blkbits, the server should report
388 * back its backend blocksize for grant calculation purpose */
389 data->ocd_grant_blkbits = PAGE_SHIFT;
391 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
392 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
393 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
394 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
395 OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
396 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
397 OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
398 OBD_CONNECT_EINPROGRESS |
399 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
400 OBD_CONNECT_LAYOUTLOCK |
401 OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
402 OBD_CONNECT_BULK_MBITS;
404 data->ocd_connect_flags2 = 0;
406 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
407 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
409 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
410 * disabled by default, because it can still be enabled on the
411 * fly via /proc. As a consequence, we still need to come to an
412 * agreement on the supported algorithms at connect time */
413 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
415 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
416 data->ocd_cksum_types = OBD_CKSUM_ADLER;
418 data->ocd_cksum_types = cksum_types_supported_client();
420 #ifdef HAVE_LRU_RESIZE_SUPPORT
421 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
423 /* always ping even if server suppress_pings */
424 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
425 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
427 CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
428 "ocd_grant: %d\n", data->ocd_connect_flags,
429 data->ocd_version, data->ocd_grant);
431 obd->obd_upcall.onu_owner = &sbi->ll_lco;
432 obd->obd_upcall.onu_upcall = cl_ocd_update;
434 data->ocd_brw_size = DT_MAX_BRW_SIZE;
436 err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
439 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
440 "recovery, of which this client is not a "
441 "part. Please wait for recovery to "
442 "complete, abort, or time out.\n", dt);
445 CERROR("%s: Cannot connect to %s: rc = %d\n",
446 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
450 sbi->ll_dt_exp->exp_connect_data = *data;
452 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
453 LUSTRE_SEQ_METADATA);
455 CERROR("%s: Can't init data layer FID infrastructure, "
456 "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
460 mutex_lock(&sbi->ll_lco.lco_lock);
461 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
462 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
463 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
464 mutex_unlock(&sbi->ll_lco.lco_lock);
466 fid_zero(&sbi->ll_root_fid);
467 err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
470 CERROR("cannot mds_connect: rc = %d\n", err);
471 GOTO(out_lock_cn_cb, err);
473 if (!fid_is_sane(&sbi->ll_root_fid)) {
474 CERROR("%s: Invalid root fid "DFID" during mount\n",
475 sbi->ll_md_exp->exp_obd->obd_name,
476 PFID(&sbi->ll_root_fid));
477 GOTO(out_lock_cn_cb, err = -EINVAL);
479 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
481 sb->s_op = &lustre_super_operations;
482 #ifdef HAVE_XATTR_HANDLER_FLAGS
483 sb->s_xattr = ll_xattr_handlers;
485 #if THREAD_SIZE >= 8192 /*b=17630*/
486 sb->s_export_op = &lustre_export_operations;
490 * XXX: move this to after cbd setup? */
491 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
492 if (sbi->ll_flags & LL_SBI_ACL)
493 valid |= OBD_MD_FLACL;
495 OBD_ALLOC_PTR(op_data);
497 GOTO(out_lock_cn_cb, err = -ENOMEM);
499 op_data->op_fid1 = sbi->ll_root_fid;
500 op_data->op_mode = 0;
501 op_data->op_valid = valid;
503 err = md_getattr(sbi->ll_md_exp, op_data, &request);
505 OBD_FREE_PTR(op_data);
507 CERROR("%s: md_getattr failed for root: rc = %d\n",
508 sbi->ll_md_exp->exp_obd->obd_name, err);
509 GOTO(out_lock_cn_cb, err);
512 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
513 sbi->ll_md_exp, &lmd);
515 CERROR("failed to understand root inode md: rc = %d\n", err);
516 ptlrpc_req_finished(request);
517 GOTO(out_lock_cn_cb, err);
520 LASSERT(fid_is_sane(&sbi->ll_root_fid));
521 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
522 sbi->ll_flags & LL_SBI_32BIT_API),
524 md_free_lustre_md(sbi->ll_md_exp, &lmd);
525 ptlrpc_req_finished(request);
528 #ifdef CONFIG_FS_POSIX_ACL
530 posix_acl_release(lmd.posix_acl);
531 lmd.posix_acl = NULL;
534 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
536 CERROR("lustre_lite: bad iget4 for root\n");
540 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
541 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
542 KEY_CHECKSUM, sizeof(checksum), &checksum,
545 CERROR("%s: Set checksum failed: rc = %d\n",
546 sbi->ll_dt_exp->exp_obd->obd_name, err);
551 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
552 KEY_CACHE_SET, sizeof(*sbi->ll_cache),
553 sbi->ll_cache, NULL);
555 CERROR("%s: Set cache_set failed: rc = %d\n",
556 sbi->ll_dt_exp->exp_obd->obd_name, err);
560 sb->s_root = d_make_root(root);
561 if (sb->s_root == NULL) {
562 CERROR("%s: can't make root dentry\n",
563 ll_get_fsname(sb, NULL, 0));
564 GOTO(out_root, err = -ENOMEM);
566 #ifdef HAVE_DCACHE_LOCK
567 sb->s_root->d_op = &ll_d_ops;
570 sbi->ll_sdev_orig = sb->s_dev;
572 /* We set sb->s_dev equal on all lustre clients in order to support
573 * NFS export clustering. NFSD requires that the FSID be the same
575 /* s_dev is also used in lt_compare() to compare two fs, but that is
576 * only a node-local comparison. */
577 uuid = obd_get_uuid(sbi->ll_md_exp);
579 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
586 if (sbi->ll_proc_root != NULL) {
587 err = lprocfs_ll_register_obd(sb, dt);
589 CERROR("%s: could not register %s in llite: rc = %d\n",
590 dt, ll_get_fsname(sb, NULL, 0), err);
593 err = lprocfs_ll_register_obd(sb, md);
595 CERROR("%s: could not register %s in llite: rc = %d\n",
596 md, ll_get_fsname(sb, NULL, 0), err);
606 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
608 obd_disconnect(sbi->ll_dt_exp);
609 sbi->ll_dt_exp = NULL;
611 obd_fid_fini(sbi->ll_md_exp->exp_obd);
613 obd_disconnect(sbi->ll_md_exp);
614 sbi->ll_md_exp = NULL;
623 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
627 size = sizeof(*lmmsize);
628 rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
629 KEY_MAX_EASIZE, &size, lmmsize);
631 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
632 sbi->ll_dt_exp->exp_obd->obd_name, rc);
637 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
638 KEY_MAX_EASIZE, &size, lmmsize);
640 CERROR("Get max mdsize error rc %d\n", rc);
646 * Get the value of the default_easize parameter.
648 * \see client_obd::cl_default_mds_easize
650 * \param[in] sbi superblock info for this filesystem
651 * \param[out] lmmsize pointer to storage location for value
653 * \retval 0 on success
654 * \retval negative negated errno on failure
656 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
661 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
662 KEY_DEFAULT_EASIZE, &size, lmmsize);
664 CERROR("Get default mdsize error rc %d\n", rc);
670 * Set the default_easize parameter to the given value.
672 * \see client_obd::cl_default_mds_easize
674 * \param[in] sbi superblock info for this filesystem
675 * \param[in] lmmsize the size to set
677 * \retval 0 on success
678 * \retval negative negated errno on failure
680 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
684 if (lmmsize < sizeof(struct lov_mds_md) ||
685 lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
688 rc = obd_set_info_async(NULL, sbi->ll_md_exp,
689 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
690 sizeof(int), &lmmsize, NULL);
695 static void client_common_put_super(struct super_block *sb)
697 struct ll_sb_info *sbi = ll_s2sbi(sb);
702 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
703 obd_disconnect(sbi->ll_dt_exp);
704 sbi->ll_dt_exp = NULL;
706 lprocfs_ll_unregister_mountpoint(sbi);
708 obd_fid_fini(sbi->ll_md_exp->exp_obd);
709 obd_disconnect(sbi->ll_md_exp);
710 sbi->ll_md_exp = NULL;
715 void ll_kill_super(struct super_block *sb)
717 struct ll_sb_info *sbi;
721 if (!(sb->s_flags & MS_ACTIVE))
725 /* we need restore s_dev from changed for clustred NFS before put_super
726 * because new kernels have cached s_dev and change sb->s_dev in
727 * put_super not affected real removing devices */
729 sb->s_dev = sbi->ll_sdev_orig;
730 sbi->ll_umounting = 1;
732 /* wait running statahead threads to quit */
733 while (atomic_read(&sbi->ll_sa_running) > 0) {
734 set_current_state(TASK_UNINTERRUPTIBLE);
735 schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
742 static inline int ll_set_opt(const char *opt, char *data, int fl)
744 if (strncmp(opt, data, strlen(opt)) != 0)
750 /* non-client-specific mount options are parsed in lmd_parse */
751 static int ll_options(char *options, int *flags)
754 char *s1 = options, *s2;
760 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
763 CDEBUG(D_SUPER, "next opt=%s\n", s1);
764 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
769 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
774 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
779 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
784 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
789 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
794 tmp = ll_set_opt("context", s1, 1);
797 tmp = ll_set_opt("fscontext", s1, 1);
800 tmp = ll_set_opt("defcontext", s1, 1);
803 tmp = ll_set_opt("rootcontext", s1, 1);
806 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
811 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
817 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
822 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
827 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
832 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
837 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
842 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
847 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
852 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
857 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
862 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
867 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
873 s2 = strchr(s1, ',');
881 void ll_lli_init(struct ll_inode_info *lli)
883 lli->lli_inode_magic = LLI_INODE_MAGIC;
885 spin_lock_init(&lli->lli_lock);
886 lli->lli_posix_acl = NULL;
887 /* Do not set lli_fid, it has been initialized already. */
888 fid_zero(&lli->lli_pfid);
889 lli->lli_mds_read_och = NULL;
890 lli->lli_mds_write_och = NULL;
891 lli->lli_mds_exec_och = NULL;
892 lli->lli_open_fd_read_count = 0;
893 lli->lli_open_fd_write_count = 0;
894 lli->lli_open_fd_exec_count = 0;
895 mutex_init(&lli->lli_och_mutex);
896 spin_lock_init(&lli->lli_agl_lock);
897 spin_lock_init(&lli->lli_layout_lock);
898 ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
899 lli->lli_clob = NULL;
901 init_rwsem(&lli->lli_xattrs_list_rwsem);
902 mutex_init(&lli->lli_xattrs_enq_lock);
904 LASSERT(lli->lli_vfs_inode.i_mode != 0);
905 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
906 mutex_init(&lli->lli_readdir_mutex);
907 lli->lli_opendir_key = NULL;
909 spin_lock_init(&lli->lli_sa_lock);
910 lli->lli_opendir_pid = 0;
911 lli->lli_sa_enabled = 0;
912 lli->lli_def_stripe_offset = -1;
914 mutex_init(&lli->lli_size_mutex);
915 lli->lli_symlink_name = NULL;
916 init_rwsem(&lli->lli_trunc_sem);
917 range_lock_tree_init(&lli->lli_write_tree);
918 init_rwsem(&lli->lli_glimpse_sem);
919 lli->lli_glimpse_time = 0;
920 INIT_LIST_HEAD(&lli->lli_agl_list);
921 lli->lli_agl_index = 0;
922 lli->lli_async_rc = 0;
924 mutex_init(&lli->lli_layout_mutex);
925 memset(lli->lli_jobid, 0, LUSTRE_JOBID_SIZE);
928 static inline int ll_bdi_register(struct backing_dev_info *bdi)
930 static atomic_t ll_bdi_num = ATOMIC_INIT(0);
932 bdi->name = "lustre";
933 return bdi_register(bdi, NULL, "lustre-%d",
934 atomic_inc_return(&ll_bdi_num));
937 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
939 struct lustre_profile *lprof = NULL;
940 struct lustre_sb_info *lsi = s2lsi(sb);
941 struct ll_sb_info *sbi;
942 char *dt = NULL, *md = NULL;
943 char *profilenm = get_profile_name(sb);
944 struct config_llog_instance *cfg;
945 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
946 const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
952 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
958 try_module_get(THIS_MODULE);
960 /* client additional sb info */
961 lsi->lsi_llsbi = sbi = ll_init_sbi();
963 module_put(THIS_MODULE);
968 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
972 err = bdi_init(&lsi->lsi_bdi);
975 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
976 #ifdef HAVE_BDI_CAP_MAP_COPY
977 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
979 lsi->lsi_bdi.capabilities = 0;
981 err = ll_bdi_register(&lsi->lsi_bdi);
985 sb->s_bdi = &lsi->lsi_bdi;
986 #ifndef HAVE_DCACHE_LOCK
987 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
988 sb->s_d_op = &ll_d_ops;
991 /* Call lprocfs_ll_register_mountpoint() before lustre_process_log()
992 * so that "llite.*.*" params can be processed correctly. */
993 if (proc_lustre_fs_root != NULL) {
994 err = lprocfs_ll_register_mountpoint(proc_lustre_fs_root, sb);
996 CERROR("%s: could not register mountpoint in llite: "
997 "rc = %d\n", ll_get_fsname(sb, NULL, 0), err);
1002 /* Generate a string unique to this super, in case some joker tries
1003 to mount the same fs at two mount points.
1004 Use the address of the super itself.*/
1005 cfg->cfg_instance = sb;
1006 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1007 cfg->cfg_callback = class_config_llog_handler;
1008 cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1009 /* set up client obds */
1010 err = lustre_process_log(sb, profilenm, cfg);
1012 GOTO(out_proc, err);
1014 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1015 lprof = class_get_profile(profilenm);
1016 if (lprof == NULL) {
1017 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1018 " read from the MGS. Does that filesystem "
1019 "exist?\n", profilenm);
1020 GOTO(out_proc, err = -EINVAL);
1022 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1023 lprof->lp_md, lprof->lp_dt);
1025 dt_len = strlen(lprof->lp_dt) + instlen + 2;
1026 OBD_ALLOC(dt, dt_len);
1028 GOTO(out_proc, err = -ENOMEM);
1029 snprintf(dt, dt_len - 1, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1031 md_len = strlen(lprof->lp_md) + instlen + 2;
1032 OBD_ALLOC(md, md_len);
1034 GOTO(out_proc, err = -ENOMEM);
1035 snprintf(md, md_len - 1, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1037 /* connections, registrations, sb setup */
1038 err = client_common_fill_super(sb, md, dt, mnt);
1040 GOTO(out_proc, err);
1042 sbi->ll_client_common_fill_super_succeeded = 1;
1046 lprocfs_ll_unregister_mountpoint(sbi);
1049 OBD_FREE(md, md_len);
1051 OBD_FREE(dt, dt_len);
1053 class_put_profile(lprof);
1056 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1057 LCONSOLE_WARN("Mounted %s\n", profilenm);
1061 } /* ll_fill_super */
1063 void ll_put_super(struct super_block *sb)
1065 struct config_llog_instance cfg, params_cfg;
1066 struct obd_device *obd;
1067 struct lustre_sb_info *lsi = s2lsi(sb);
1068 struct ll_sb_info *sbi = ll_s2sbi(sb);
1069 char *profilenm = get_profile_name(sb);
1071 int next, force = 1, rc = 0;
1074 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1076 cfg.cfg_instance = sb;
1077 lustre_end_log(sb, profilenm, &cfg);
1079 params_cfg.cfg_instance = sb;
1080 lustre_end_log(sb, PARAMS_FILENAME, ¶ms_cfg);
1082 if (sbi->ll_md_exp) {
1083 obd = class_exp2obd(sbi->ll_md_exp);
1085 force = obd->obd_force;
1088 /* Wait for unstable pages to be committed to stable storage */
1090 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1091 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1092 atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1096 ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1097 if (force == 0 && rc != -EINTR)
1098 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1101 /* We need to set force before the lov_disconnect in
1102 lustre_common_put_super, since l_d cleans up osc's as well. */
1105 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1107 obd->obd_force = force;
1111 if (sbi->ll_client_common_fill_super_succeeded) {
1112 /* Only if client_common_fill_super succeeded */
1113 client_common_put_super(sb);
1117 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1118 class_manual_cleanup(obd);
1121 if (sbi->ll_flags & LL_SBI_VERBOSE)
1122 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1125 class_del_profile(profilenm);
1127 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1128 bdi_destroy(&lsi->lsi_bdi);
1129 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1133 lsi->lsi_llsbi = NULL;
1135 lustre_common_put_super(sb);
1137 cl_env_cache_purge(~0);
1139 module_put(THIS_MODULE);
1142 } /* client_put_super */
1144 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1146 struct inode *inode = NULL;
1148 /* NOTE: we depend on atomic igrab() -bzzz */
1149 lock_res_and_lock(lock);
1150 if (lock->l_resource->lr_lvb_inode) {
1151 struct ll_inode_info * lli;
1152 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1153 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1154 inode = igrab(lock->l_resource->lr_lvb_inode);
1156 inode = lock->l_resource->lr_lvb_inode;
1157 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1158 D_WARNING, lock, "lr_lvb_inode %p is "
1159 "bogus: magic %08x",
1160 lock->l_resource->lr_lvb_inode,
1161 lli->lli_inode_magic);
1165 unlock_res_and_lock(lock);
1169 void ll_dir_clear_lsm_md(struct inode *inode)
1171 struct ll_inode_info *lli = ll_i2info(inode);
1173 LASSERT(S_ISDIR(inode->i_mode));
1175 if (lli->lli_lsm_md != NULL) {
1176 lmv_free_memmd(lli->lli_lsm_md);
1177 lli->lli_lsm_md = NULL;
1181 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1182 const struct lu_fid *fid,
1183 struct lustre_md *md)
1185 struct ll_sb_info *sbi = ll_s2sbi(sb);
1186 struct mdt_body *body = md->body;
1187 struct inode *inode;
1191 ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1192 inode = iget_locked(sb, ino);
1193 if (inode == NULL) {
1194 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1195 ll_get_fsname(sb, NULL, 0), PFID(fid));
1196 RETURN(ERR_PTR(-ENOENT));
1199 if (inode->i_state & I_NEW) {
1200 struct ll_inode_info *lli = ll_i2info(inode);
1201 struct lmv_stripe_md *lsm = md->lmv;
1203 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1204 (body->mbo_mode & S_IFMT);
1205 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1208 LTIME_S(inode->i_mtime) = 0;
1209 LTIME_S(inode->i_atime) = 0;
1210 LTIME_S(inode->i_ctime) = 0;
1213 #ifdef HAVE_BACKING_DEV_INFO
1214 /* initializing backing dev info. */
1215 inode->i_mapping->backing_dev_info =
1216 &s2lsi(inode->i_sb)->lsi_bdi;
1218 inode->i_op = &ll_dir_inode_operations;
1219 inode->i_fop = &ll_dir_operations;
1220 lli->lli_fid = *fid;
1223 LASSERT(lsm != NULL);
1224 /* master object FID */
1225 lli->lli_pfid = body->mbo_fid1;
1226 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1227 lli, PFID(fid), PFID(&lli->lli_pfid));
1228 unlock_new_inode(inode);
1234 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1237 struct lmv_stripe_md *lsm = md->lmv;
1240 LASSERT(lsm != NULL);
1241 /* XXX sigh, this lsm_root initialization should be in
1242 * LMV layer, but it needs ll_iget right now, so we
1243 * put this here right now. */
1244 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1245 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1246 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1247 /* Unfortunately ll_iget will call ll_update_inode,
1248 * where the initialization of slave inode is slightly
1249 * different, so it reset lsm_md to NULL to avoid
1250 * initializing lsm for slave inode. */
1251 /* For migrating inode, master stripe and master object will
1252 * be same, so we only need assign this inode */
1253 if (lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION && i == 0)
1254 lsm->lsm_md_oinfo[i].lmo_root = inode;
1256 lsm->lsm_md_oinfo[i].lmo_root =
1257 ll_iget_anon_dir(inode->i_sb, fid, md);
1259 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1260 int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1262 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1270 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1271 const struct lmv_stripe_md *lsm_md2)
1273 return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1274 lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1275 lsm_md1->lsm_md_master_mdt_index ==
1276 lsm_md2->lsm_md_master_mdt_index &&
1277 lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1278 lsm_md1->lsm_md_layout_version ==
1279 lsm_md2->lsm_md_layout_version &&
1280 strcmp(lsm_md1->lsm_md_pool_name,
1281 lsm_md2->lsm_md_pool_name) == 0;
1284 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1286 struct ll_inode_info *lli = ll_i2info(inode);
1287 struct lmv_stripe_md *lsm = md->lmv;
1291 LASSERT(S_ISDIR(inode->i_mode));
1292 CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1293 PFID(ll_inode2fid(inode)));
1295 /* no striped information from request. */
1297 if (lli->lli_lsm_md == NULL) {
1299 } else if (lli->lli_lsm_md->lsm_md_hash_type &
1300 LMV_HASH_FLAG_MIGRATION) {
1301 /* migration is done, the temporay MIGRATE layout has
1303 CDEBUG(D_INODE, DFID" finish migration.\n",
1304 PFID(ll_inode2fid(inode)));
1305 lmv_free_memmd(lli->lli_lsm_md);
1306 lli->lli_lsm_md = NULL;
1309 /* The lustre_md from req does not include stripeEA,
1310 * see ll_md_setattr */
1315 /* set the directory layout */
1316 if (lli->lli_lsm_md == NULL) {
1317 struct cl_attr *attr;
1319 rc = ll_init_lsm_md(inode, md);
1323 /* set md->lmv to NULL, so the following free lustre_md
1324 * will not free this lsm */
1326 lli->lli_lsm_md = lsm;
1328 OBD_ALLOC_PTR(attr);
1332 /* validate the lsm */
1333 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1334 ll_md_blocking_ast);
1340 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1341 md->body->mbo_nlink = attr->cat_nlink;
1342 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1343 md->body->mbo_size = attr->cat_size;
1344 if (md->body->mbo_valid & OBD_MD_FLATIME)
1345 md->body->mbo_atime = attr->cat_atime;
1346 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1347 md->body->mbo_ctime = attr->cat_ctime;
1348 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1349 md->body->mbo_mtime = attr->cat_mtime;
1353 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1354 lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1358 /* Compare the old and new stripe information */
1359 if (!lsm_md_eq(lli->lli_lsm_md, lsm)) {
1360 struct lmv_stripe_md *old_lsm = lli->lli_lsm_md;
1363 CERROR("%s: inode "DFID"(%p)'s lmv layout mismatch (%p)/(%p)"
1364 "magic:0x%x/0x%x stripe count: %d/%d master_mdt: %d/%d"
1365 "hash_type:0x%x/0x%x layout: 0x%x/0x%x pool:%s/%s\n",
1366 ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1367 inode, lsm, old_lsm,
1368 lsm->lsm_md_magic, old_lsm->lsm_md_magic,
1369 lsm->lsm_md_stripe_count,
1370 old_lsm->lsm_md_stripe_count,
1371 lsm->lsm_md_master_mdt_index,
1372 old_lsm->lsm_md_master_mdt_index,
1373 lsm->lsm_md_hash_type, old_lsm->lsm_md_hash_type,
1374 lsm->lsm_md_layout_version,
1375 old_lsm->lsm_md_layout_version,
1376 lsm->lsm_md_pool_name,
1377 old_lsm->lsm_md_pool_name);
1379 for (idx = 0; idx < old_lsm->lsm_md_stripe_count; idx++) {
1380 CERROR("%s: sub FIDs in old lsm idx %d, old: "DFID"\n",
1381 ll_get_fsname(inode->i_sb, NULL, 0), idx,
1382 PFID(&old_lsm->lsm_md_oinfo[idx].lmo_fid));
1385 for (idx = 0; idx < lsm->lsm_md_stripe_count; idx++) {
1386 CERROR("%s: sub FIDs in new lsm idx %d, new: "DFID"\n",
1387 ll_get_fsname(inode->i_sb, NULL, 0), idx,
1388 PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1397 void ll_clear_inode(struct inode *inode)
1399 struct ll_inode_info *lli = ll_i2info(inode);
1400 struct ll_sb_info *sbi = ll_i2sbi(inode);
1403 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1404 PFID(ll_inode2fid(inode)), inode);
1406 if (S_ISDIR(inode->i_mode)) {
1407 /* these should have been cleared in ll_file_release */
1408 LASSERT(lli->lli_opendir_key == NULL);
1409 LASSERT(lli->lli_sai == NULL);
1410 LASSERT(lli->lli_opendir_pid == 0);
1413 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1415 LASSERT(!lli->lli_open_fd_write_count);
1416 LASSERT(!lli->lli_open_fd_read_count);
1417 LASSERT(!lli->lli_open_fd_exec_count);
1419 if (lli->lli_mds_write_och)
1420 ll_md_real_close(inode, FMODE_WRITE);
1421 if (lli->lli_mds_exec_och)
1422 ll_md_real_close(inode, FMODE_EXEC);
1423 if (lli->lli_mds_read_och)
1424 ll_md_real_close(inode, FMODE_READ);
1426 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1427 OBD_FREE(lli->lli_symlink_name,
1428 strlen(lli->lli_symlink_name) + 1);
1429 lli->lli_symlink_name = NULL;
1432 ll_xattr_cache_destroy(inode);
1434 #ifdef CONFIG_FS_POSIX_ACL
1435 forget_all_cached_acls(inode);
1436 if (lli->lli_posix_acl) {
1437 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1438 posix_acl_release(lli->lli_posix_acl);
1439 lli->lli_posix_acl = NULL;
1442 lli->lli_inode_magic = LLI_INODE_DEAD;
1444 if (S_ISDIR(inode->i_mode))
1445 ll_dir_clear_lsm_md(inode);
1446 else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1447 LASSERT(list_empty(&lli->lli_agl_list));
1450 * XXX This has to be done before lsm is freed below, because
1451 * cl_object still uses inode lsm.
1453 cl_inode_fini(inode);
1458 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1460 struct lustre_md md;
1461 struct inode *inode = dentry->d_inode;
1462 struct ll_sb_info *sbi = ll_i2sbi(inode);
1463 struct ptlrpc_request *request = NULL;
1467 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1468 LUSTRE_OPC_ANY, NULL);
1469 if (IS_ERR(op_data))
1470 RETURN(PTR_ERR(op_data));
1472 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1474 ptlrpc_req_finished(request);
1475 if (rc == -ENOENT) {
1477 /* Unlinked special device node? Or just a race?
1478 * Pretend we done everything. */
1479 if (!S_ISREG(inode->i_mode) &&
1480 !S_ISDIR(inode->i_mode)) {
1481 ia_valid = op_data->op_attr.ia_valid;
1482 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1483 rc = simple_setattr(dentry, &op_data->op_attr);
1484 op_data->op_attr.ia_valid = ia_valid;
1486 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1487 CERROR("md_setattr fails: rc = %d\n", rc);
1492 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1493 sbi->ll_md_exp, &md);
1495 ptlrpc_req_finished(request);
1499 ia_valid = op_data->op_attr.ia_valid;
1500 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1501 * cache is not cleared yet. */
1502 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1503 if (S_ISREG(inode->i_mode))
1505 rc = simple_setattr(dentry, &op_data->op_attr);
1506 if (S_ISREG(inode->i_mode))
1507 inode_unlock(inode);
1508 op_data->op_attr.ia_valid = ia_valid;
1510 rc = ll_update_inode(inode, &md);
1511 ptlrpc_req_finished(request);
1516 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1517 * object(s) determine the file size and mtime. Otherwise, the MDS will
1518 * keep these values until such a time that objects are allocated for it.
1519 * We do the MDS operations first, as it is checking permissions for us.
1520 * We don't to the MDS RPC if there is nothing that we want to store there,
1521 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1522 * going to do an RPC anyways.
1524 * If we are doing a truncate, we will send the mtime and ctime updates
1525 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1526 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1529 * In case of HSMimport, we only set attr on MDS.
1531 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1533 struct inode *inode = dentry->d_inode;
1534 struct ll_inode_info *lli = ll_i2info(inode);
1535 struct md_op_data *op_data = NULL;
1539 CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1540 "valid %x, hsm_import %d\n",
1541 ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1542 inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1545 if (attr->ia_valid & ATTR_SIZE) {
1546 /* Check new size against VFS/VM file size limit and rlimit */
1547 rc = inode_newsize_ok(inode, attr->ia_size);
1551 /* The maximum Lustre file size is variable, based on the
1552 * OST maximum object size and number of stripes. This
1553 * needs another check in addition to the VFS check above. */
1554 if (attr->ia_size > ll_file_maxbytes(inode)) {
1555 CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
1556 PFID(&lli->lli_fid), attr->ia_size,
1557 ll_file_maxbytes(inode));
1561 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1564 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1565 if (attr->ia_valid & TIMES_SET_FLAGS) {
1566 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1567 !cfs_capable(CFS_CAP_FOWNER))
1571 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1572 if (!(attr->ia_valid & ATTR_CTIME_SET) &&
1573 (attr->ia_valid & ATTR_CTIME)) {
1574 attr->ia_ctime = CURRENT_TIME;
1575 attr->ia_valid |= ATTR_CTIME_SET;
1577 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1578 (attr->ia_valid & ATTR_ATIME)) {
1579 attr->ia_atime = CURRENT_TIME;
1580 attr->ia_valid |= ATTR_ATIME_SET;
1582 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1583 (attr->ia_valid & ATTR_MTIME)) {
1584 attr->ia_mtime = CURRENT_TIME;
1585 attr->ia_valid |= ATTR_MTIME_SET;
1588 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1589 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %llu\n",
1590 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1591 (s64)ktime_get_real_seconds());
1593 if (S_ISREG(inode->i_mode)) {
1594 if (attr->ia_valid & ATTR_SIZE)
1595 inode_dio_write_done(inode);
1596 inode_unlock(inode);
1599 /* We always do an MDS RPC, even if we're only changing the size;
1600 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1602 OBD_ALLOC_PTR(op_data);
1603 if (op_data == NULL)
1604 GOTO(out, rc = -ENOMEM);
1606 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1607 /* If we are changing file size, file content is
1608 * modified, flag it. */
1609 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1610 op_data->op_bias |= MDS_DATA_MODIFIED;
1611 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1614 op_data->op_attr = *attr;
1616 rc = ll_md_setattr(dentry, op_data);
1620 if (!S_ISREG(inode->i_mode) || hsm_import)
1623 if (attr->ia_valid & (ATTR_SIZE |
1624 ATTR_ATIME | ATTR_ATIME_SET |
1625 ATTR_MTIME | ATTR_MTIME_SET |
1626 ATTR_CTIME | ATTR_CTIME_SET)) {
1627 /* For truncate and utimes sending attributes to OSTs, setting
1628 * mtime/atime to the past will be performed under PW [0:EOF]
1629 * extent lock (new_size:EOF for truncate). It may seem
1630 * excessive to send mtime/atime updates to OSTs when not
1631 * setting times to past, but it is necessary due to possible
1632 * time de-synchronization between MDT inode and OST objects */
1633 rc = cl_setattr_ost(lli->lli_clob, attr, 0);
1636 /* If the file was restored, it needs to set dirty flag.
1638 * We've already sent MDS_DATA_MODIFIED flag in
1639 * ll_md_setattr() for truncate. However, the MDT refuses to
1640 * set the HS_DIRTY flag on released files, so we have to set
1641 * it again if the file has been restored. Please check how
1642 * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1644 * Please notice that if the file is not released, the previous
1645 * MDS_DATA_MODIFIED has taken effect and usually
1646 * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1647 * This way we can save an RPC for common open + trunc
1649 if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1650 struct hsm_state_set hss = {
1651 .hss_valid = HSS_SETMASK,
1652 .hss_setmask = HS_DIRTY,
1656 rc2 = ll_hsm_state_set(inode, &hss);
1657 /* truncate and write can happen at the same time, so that
1658 * the file can be set modified even though the file is not
1659 * restored from released state, and ll_hsm_state_set() is
1660 * not applicable for the file, and rc2 < 0 is normal in this
1663 CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1664 PFID(ll_inode2fid(inode)), rc2);
1669 if (op_data != NULL)
1670 ll_finish_md_op_data(op_data);
1672 if (S_ISREG(inode->i_mode)) {
1674 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1675 inode_dio_wait(inode);
1676 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
1677 * flag. ll_update_inode (called from ll_md_setattr), clears
1678 * inode flags, so there is a gap where S_NOSEC is not set.
1679 * This can cause a writer to take the i_mutex unnecessarily,
1680 * but this is safe to do and should be rare. */
1681 inode_has_no_xattr(inode);
1684 ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1685 LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1690 int ll_setattr(struct dentry *de, struct iattr *attr)
1692 int mode = de->d_inode->i_mode;
1694 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1695 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1696 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1698 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1699 (ATTR_SIZE|ATTR_MODE)) &&
1700 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1701 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1702 !(attr->ia_mode & S_ISGID))))
1703 attr->ia_valid |= ATTR_FORCE;
1705 if ((attr->ia_valid & ATTR_MODE) &&
1707 !(attr->ia_mode & S_ISUID) &&
1708 !(attr->ia_valid & ATTR_KILL_SUID))
1709 attr->ia_valid |= ATTR_KILL_SUID;
1711 if ((attr->ia_valid & ATTR_MODE) &&
1712 ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1713 !(attr->ia_mode & S_ISGID) &&
1714 !(attr->ia_valid & ATTR_KILL_SGID))
1715 attr->ia_valid |= ATTR_KILL_SGID;
1717 /* avoid polluted from ATTR_TIMES_SET,
1718 * projid is not expected to be set here */
1719 attr->ia_valid &= ~MDS_ATTR_PROJID;
1721 return ll_setattr_raw(de, attr, false);
1724 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1725 __u64 max_age, __u32 flags)
1727 struct ll_sb_info *sbi = ll_s2sbi(sb);
1728 struct obd_statfs obd_osfs;
1732 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1734 CERROR("md_statfs fails: rc = %d\n", rc);
1738 osfs->os_type = sb->s_magic;
1740 CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1741 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1743 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1744 flags |= OBD_STATFS_NODELAY;
1746 rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1748 CERROR("obd_statfs fails: rc = %d\n", rc);
1752 CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1753 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1756 osfs->os_bsize = obd_osfs.os_bsize;
1757 osfs->os_blocks = obd_osfs.os_blocks;
1758 osfs->os_bfree = obd_osfs.os_bfree;
1759 osfs->os_bavail = obd_osfs.os_bavail;
1761 /* If we don't have as many objects free on the OST as inodes
1762 * on the MDS, we reduce the total number of inodes to
1763 * compensate, so that the "inodes in use" number is correct.
1765 if (obd_osfs.os_ffree < osfs->os_ffree) {
1766 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1768 osfs->os_ffree = obd_osfs.os_ffree;
1773 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1775 struct super_block *sb = de->d_sb;
1776 struct obd_statfs osfs;
1777 __u64 fsid = huge_encode_dev(sb->s_dev);
1780 CDEBUG(D_VFSTRACE, "VFS Op: at %llu jiffies\n", get_jiffies_64());
1781 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1783 /* Some amount of caching on the client is allowed */
1784 rc = ll_statfs_internal(sb, &osfs,
1785 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1790 statfs_unpack(sfs, &osfs);
1792 /* We need to downshift for all 32-bit kernels, because we can't
1793 * tell if the kernel is being called via sys_statfs64() or not.
1794 * Stop before overflowing f_bsize - in which case it is better
1795 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1796 if (sizeof(long) < 8) {
1797 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1800 osfs.os_blocks >>= 1;
1801 osfs.os_bfree >>= 1;
1802 osfs.os_bavail >>= 1;
1806 sfs->f_blocks = osfs.os_blocks;
1807 sfs->f_bfree = osfs.os_bfree;
1808 sfs->f_bavail = osfs.os_bavail;
1809 sfs->f_fsid.val[0] = (__u32)fsid;
1810 sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1814 void ll_inode_size_lock(struct inode *inode)
1816 struct ll_inode_info *lli;
1818 LASSERT(!S_ISDIR(inode->i_mode));
1820 lli = ll_i2info(inode);
1821 mutex_lock(&lli->lli_size_mutex);
1824 void ll_inode_size_unlock(struct inode *inode)
1826 struct ll_inode_info *lli;
1828 lli = ll_i2info(inode);
1829 mutex_unlock(&lli->lli_size_mutex);
1832 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1834 struct ll_inode_info *lli = ll_i2info(inode);
1835 struct mdt_body *body = md->body;
1836 struct ll_sb_info *sbi = ll_i2sbi(inode);
1839 if (body->mbo_valid & OBD_MD_FLEASIZE) {
1840 rc = cl_file_inode_init(inode, md);
1845 if (S_ISDIR(inode->i_mode)) {
1846 rc = ll_update_lsm_md(inode, md);
1851 #ifdef CONFIG_FS_POSIX_ACL
1852 if (body->mbo_valid & OBD_MD_FLACL) {
1853 spin_lock(&lli->lli_lock);
1854 if (lli->lli_posix_acl)
1855 posix_acl_release(lli->lli_posix_acl);
1856 lli->lli_posix_acl = md->posix_acl;
1857 spin_unlock(&lli->lli_lock);
1860 inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1861 sbi->ll_flags & LL_SBI_32BIT_API);
1862 inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1864 if (body->mbo_valid & OBD_MD_FLATIME) {
1865 if (body->mbo_atime > LTIME_S(inode->i_atime))
1866 LTIME_S(inode->i_atime) = body->mbo_atime;
1867 lli->lli_atime = body->mbo_atime;
1870 if (body->mbo_valid & OBD_MD_FLMTIME) {
1871 if (body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1872 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1873 "to %llu\n", inode->i_ino,
1874 LTIME_S(inode->i_mtime), body->mbo_mtime);
1875 LTIME_S(inode->i_mtime) = body->mbo_mtime;
1877 lli->lli_mtime = body->mbo_mtime;
1880 if (body->mbo_valid & OBD_MD_FLCTIME) {
1881 if (body->mbo_ctime > LTIME_S(inode->i_ctime))
1882 LTIME_S(inode->i_ctime) = body->mbo_ctime;
1883 lli->lli_ctime = body->mbo_ctime;
1886 /* Clear i_flags to remove S_NOSEC before permissions are updated */
1887 if (body->mbo_valid & OBD_MD_FLFLAGS)
1888 inode->i_flags = ll_ext_to_inode_flags(body->mbo_flags);
1889 if (body->mbo_valid & OBD_MD_FLMODE)
1890 inode->i_mode = (inode->i_mode & S_IFMT) |
1891 (body->mbo_mode & ~S_IFMT);
1893 if (body->mbo_valid & OBD_MD_FLTYPE)
1894 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1895 (body->mbo_mode & S_IFMT);
1897 LASSERT(inode->i_mode != 0);
1898 if (S_ISREG(inode->i_mode))
1899 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1900 LL_MAX_BLKSIZE_BITS);
1902 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1904 if (body->mbo_valid & OBD_MD_FLUID)
1905 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
1906 if (body->mbo_valid & OBD_MD_FLGID)
1907 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
1908 if (body->mbo_valid & OBD_MD_FLPROJID)
1909 lli->lli_projid = body->mbo_projid;
1910 if (body->mbo_valid & OBD_MD_FLNLINK)
1911 set_nlink(inode, body->mbo_nlink);
1912 if (body->mbo_valid & OBD_MD_FLRDEV)
1913 inode->i_rdev = old_decode_dev(body->mbo_rdev);
1915 if (body->mbo_valid & OBD_MD_FLID) {
1916 /* FID shouldn't be changed! */
1917 if (fid_is_sane(&lli->lli_fid)) {
1918 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
1919 "Trying to change FID "DFID
1920 " to the "DFID", inode "DFID"(%p)\n",
1921 PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
1922 PFID(ll_inode2fid(inode)), inode);
1924 lli->lli_fid = body->mbo_fid1;
1928 LASSERT(fid_seq(&lli->lli_fid) != 0);
1930 if (body->mbo_valid & OBD_MD_FLSIZE) {
1931 i_size_write(inode, body->mbo_size);
1933 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
1934 PFID(ll_inode2fid(inode)),
1935 (unsigned long long)body->mbo_size);
1937 if (body->mbo_valid & OBD_MD_FLBLOCKS)
1938 inode->i_blocks = body->mbo_blocks;
1941 if (body->mbo_valid & OBD_MD_TSTATE) {
1942 /* Set LLIF_FILE_RESTORING if restore ongoing and
1943 * clear it when done to ensure to start again
1944 * glimpsing updated attrs
1946 if (body->mbo_t_state & MS_RESTORE)
1947 ll_file_set_flag(lli, LLIF_FILE_RESTORING);
1949 ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
1955 int ll_read_inode2(struct inode *inode, void *opaque)
1957 struct lustre_md *md = opaque;
1958 struct ll_inode_info *lli = ll_i2info(inode);
1962 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1963 PFID(&lli->lli_fid), inode);
1965 /* Core attributes from the MDS first. This is a new inode, and
1966 * the VFS doesn't zero times in the core inode so we have to do
1967 * it ourselves. They will be overwritten by either MDS or OST
1968 * attributes - we just need to make sure they aren't newer. */
1969 LTIME_S(inode->i_mtime) = 0;
1970 LTIME_S(inode->i_atime) = 0;
1971 LTIME_S(inode->i_ctime) = 0;
1973 rc = ll_update_inode(inode, md);
1977 /* OIDEBUG(inode); */
1979 #ifdef HAVE_BACKING_DEV_INFO
1980 /* initializing backing dev info. */
1981 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1983 if (S_ISREG(inode->i_mode)) {
1984 struct ll_sb_info *sbi = ll_i2sbi(inode);
1985 inode->i_op = &ll_file_inode_operations;
1986 inode->i_fop = sbi->ll_fop;
1987 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1989 } else if (S_ISDIR(inode->i_mode)) {
1990 inode->i_op = &ll_dir_inode_operations;
1991 inode->i_fop = &ll_dir_operations;
1993 } else if (S_ISLNK(inode->i_mode)) {
1994 inode->i_op = &ll_fast_symlink_inode_operations;
1997 inode->i_op = &ll_special_inode_operations;
1999 init_special_inode(inode, inode->i_mode,
2008 void ll_delete_inode(struct inode *inode)
2010 struct ll_inode_info *lli = ll_i2info(inode);
2013 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2014 /* It is last chance to write out dirty pages,
2015 * otherwise we may lose data while umount */
2016 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
2018 truncate_inode_pages_final(&inode->i_data);
2020 LASSERTF(inode->i_data.nrpages == 0, "inode="DFID"(%p) nrpages=%lu, "
2021 "see https://jira.hpdd.intel.com/browse/LU-118\n",
2022 PFID(ll_inode2fid(inode)), inode, inode->i_data.nrpages);
2024 #ifdef HAVE_SBOPS_EVICT_INODE
2025 ll_clear_inode(inode);
2032 int ll_iocontrol(struct inode *inode, struct file *file,
2033 unsigned int cmd, unsigned long arg)
2035 struct ll_sb_info *sbi = ll_i2sbi(inode);
2036 struct ptlrpc_request *req = NULL;
2041 case FSFILT_IOC_GETFLAGS: {
2042 struct mdt_body *body;
2043 struct md_op_data *op_data;
2045 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2046 0, 0, LUSTRE_OPC_ANY,
2048 if (IS_ERR(op_data))
2049 RETURN(PTR_ERR(op_data));
2051 op_data->op_valid = OBD_MD_FLFLAGS;
2052 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2053 ll_finish_md_op_data(op_data);
2055 CERROR("%s: failure inode "DFID": rc = %d\n",
2056 sbi->ll_md_exp->exp_obd->obd_name,
2057 PFID(ll_inode2fid(inode)), rc);
2061 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2063 flags = body->mbo_flags;
2065 ptlrpc_req_finished(req);
2067 RETURN(put_user(flags, (int __user *)arg));
2069 case FSFILT_IOC_SETFLAGS: {
2071 struct md_op_data *op_data;
2072 struct cl_object *obj;
2074 if (get_user(flags, (int __user *)arg))
2077 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2078 LUSTRE_OPC_ANY, NULL);
2079 if (IS_ERR(op_data))
2080 RETURN(PTR_ERR(op_data));
2082 op_data->op_attr_flags = flags;
2083 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2084 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2085 ll_finish_md_op_data(op_data);
2086 ptlrpc_req_finished(req);
2090 inode->i_flags = ll_ext_to_inode_flags(flags);
2092 obj = ll_i2info(inode)->lli_clob;
2096 OBD_ALLOC_PTR(attr);
2100 attr->ia_valid = ATTR_ATTR_FLAG;
2101 rc = cl_setattr_ost(obj, attr, flags);
2113 int ll_flush_ctx(struct inode *inode)
2115 struct ll_sb_info *sbi = ll_i2sbi(inode);
2117 CDEBUG(D_SEC, "flush context for user %d\n",
2118 from_kuid(&init_user_ns, current_uid()));
2120 obd_set_info_async(NULL, sbi->ll_md_exp,
2121 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2123 obd_set_info_async(NULL, sbi->ll_dt_exp,
2124 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2129 /* umount -f client means force down, don't save state */
2130 void ll_umount_begin(struct super_block *sb)
2132 struct ll_sb_info *sbi = ll_s2sbi(sb);
2133 struct obd_device *obd;
2134 struct obd_ioctl_data *ioc_data;
2135 struct l_wait_info lwi;
2136 wait_queue_head_t waitq;
2139 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2140 sb->s_count, atomic_read(&sb->s_active));
2142 obd = class_exp2obd(sbi->ll_md_exp);
2144 CERROR("Invalid MDC connection handle %#llx\n",
2145 sbi->ll_md_exp->exp_handle.h_cookie);
2151 obd = class_exp2obd(sbi->ll_dt_exp);
2153 CERROR("Invalid LOV connection handle %#llx\n",
2154 sbi->ll_dt_exp->exp_handle.h_cookie);
2160 OBD_ALLOC_PTR(ioc_data);
2162 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2163 sizeof *ioc_data, ioc_data, NULL);
2165 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2166 sizeof *ioc_data, ioc_data, NULL);
2168 OBD_FREE_PTR(ioc_data);
2171 /* Really, we'd like to wait until there are no requests outstanding,
2172 * and then continue. For now, we just periodically checking for vfs
2173 * to decrement mnt_cnt and hope to finish it within 10sec.
2175 init_waitqueue_head(&waitq);
2176 lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(10),
2177 cfs_time_seconds(1), NULL, NULL);
2178 l_wait_event(waitq, may_umount(sbi->ll_mnt.mnt), &lwi);
2183 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2185 struct ll_sb_info *sbi = ll_s2sbi(sb);
2186 char *profilenm = get_profile_name(sb);
2190 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2191 read_only = *flags & MS_RDONLY;
2192 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2193 sizeof(KEY_READ_ONLY),
2194 KEY_READ_ONLY, sizeof(read_only),
2197 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2198 profilenm, read_only ?
2199 "read-only" : "read-write", err);
2204 sb->s_flags |= MS_RDONLY;
2206 sb->s_flags &= ~MS_RDONLY;
2208 if (sbi->ll_flags & LL_SBI_VERBOSE)
2209 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2210 read_only ? "read-only" : "read-write");
2216 * Cleanup the open handle that is cached on MDT-side.
2218 * For open case, the client side open handling thread may hit error
2219 * after the MDT grant the open. Under such case, the client should
2220 * send close RPC to the MDT as cleanup; otherwise, the open handle
2221 * on the MDT will be leaked there until the client umount or evicted.
2223 * In further, if someone unlinked the file, because the open handle
2224 * holds the reference on such file/object, then it will block the
2225 * subsequent threads that want to locate such object via FID.
2227 * \param[in] sb super block for this file-system
2228 * \param[in] open_req pointer to the original open request
2230 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2232 struct mdt_body *body;
2233 struct md_op_data *op_data;
2234 struct ptlrpc_request *close_req = NULL;
2235 struct obd_export *exp = ll_s2sbi(sb)->ll_md_exp;
2238 body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2239 OBD_ALLOC_PTR(op_data);
2240 if (op_data == NULL) {
2241 CWARN("%s: cannot allocate op_data to release open handle for "
2243 ll_get_fsname(sb, NULL, 0), PFID(&body->mbo_fid1));
2248 op_data->op_fid1 = body->mbo_fid1;
2249 op_data->op_handle = body->mbo_handle;
2250 op_data->op_mod_time = ktime_get_real_seconds();
2251 md_close(exp, op_data, NULL, &close_req);
2252 ptlrpc_req_finished(close_req);
2253 ll_finish_md_op_data(op_data);
2258 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2259 struct super_block *sb, struct lookup_intent *it)
2261 struct ll_sb_info *sbi = NULL;
2262 struct lustre_md md = { NULL };
2266 LASSERT(*inode || sb);
2267 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2268 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2269 sbi->ll_md_exp, &md);
2274 rc = ll_update_inode(*inode, &md);
2278 LASSERT(sb != NULL);
2281 * At this point server returns to client's same fid as client
2282 * generated for creating. So using ->fid1 is okay here.
2284 if (!fid_is_sane(&md.body->mbo_fid1)) {
2285 CERROR("%s: Fid is insane "DFID"\n",
2286 ll_get_fsname(sb, NULL, 0),
2287 PFID(&md.body->mbo_fid1));
2288 GOTO(out, rc = -EINVAL);
2291 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2292 sbi->ll_flags & LL_SBI_32BIT_API),
2294 if (IS_ERR(*inode)) {
2295 #ifdef CONFIG_FS_POSIX_ACL
2297 posix_acl_release(md.posix_acl);
2298 md.posix_acl = NULL;
2301 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2303 CERROR("new_inode -fatal: rc %d\n", rc);
2308 /* Handling piggyback layout lock.
2309 * Layout lock can be piggybacked by getattr and open request.
2310 * The lsm can be applied to inode only if it comes with a layout lock
2311 * otherwise correct layout may be overwritten, for example:
2312 * 1. proc1: mdt returns a lsm but not granting layout
2313 * 2. layout was changed by another client
2314 * 3. proc2: refresh layout and layout lock granted
2315 * 4. proc1: to apply a stale layout */
2316 if (it != NULL && it->it_lock_mode != 0) {
2317 struct lustre_handle lockh;
2318 struct ldlm_lock *lock;
2320 lockh.cookie = it->it_lock_handle;
2321 lock = ldlm_handle2lock(&lockh);
2322 LASSERT(lock != NULL);
2323 if (ldlm_has_layout(lock)) {
2324 struct cl_object_conf conf;
2326 memset(&conf, 0, sizeof(conf));
2327 conf.coc_opc = OBJECT_CONF_SET;
2328 conf.coc_inode = *inode;
2329 conf.coc_lock = lock;
2330 conf.u.coc_layout = md.layout;
2331 (void)ll_layout_conf(*inode, &conf);
2333 LDLM_LOCK_PUT(lock);
2339 md_free_lustre_md(sbi->ll_md_exp, &md);
2342 if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2343 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2348 int ll_obd_statfs(struct inode *inode, void __user *arg)
2350 struct ll_sb_info *sbi = NULL;
2351 struct obd_export *exp;
2353 struct obd_ioctl_data *data = NULL;
2357 if (!inode || !(sbi = ll_i2sbi(inode)))
2358 GOTO(out_statfs, rc = -EINVAL);
2360 rc = obd_ioctl_getdata(&buf, &len, arg);
2362 GOTO(out_statfs, rc);
2365 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2366 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2367 GOTO(out_statfs, rc = -EINVAL);
2369 if (data->ioc_inllen1 != sizeof(__u32) ||
2370 data->ioc_inllen2 != sizeof(__u32) ||
2371 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2372 data->ioc_plen2 != sizeof(struct obd_uuid))
2373 GOTO(out_statfs, rc = -EINVAL);
2375 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2376 if (type & LL_STATFS_LMV)
2377 exp = sbi->ll_md_exp;
2378 else if (type & LL_STATFS_LOV)
2379 exp = sbi->ll_dt_exp;
2381 GOTO(out_statfs, rc = -ENODEV);
2383 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2385 GOTO(out_statfs, rc);
2387 OBD_FREE_LARGE(buf, len);
2391 int ll_process_config(struct lustre_cfg *lcfg)
2393 struct super_block *sb;
2398 /* The instance name contains the sb: lustre-client-aacfe000 */
2399 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2400 if (!ptr || !*(++ptr))
2402 if (sscanf(ptr, "%lx", &x) != 1)
2404 sb = (struct super_block *)x;
2405 /* This better be a real Lustre superblock! */
2406 LASSERT(s2lsi(sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2408 /* Note we have not called client_common_fill_super yet, so
2409 proc fns must be able to handle that! */
2410 rc = class_process_proc_param(PARAM_LLITE, lprocfs_llite_obd_vars,
2417 /* this function prepares md_op_data hint for passing it down to MD stack. */
2418 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2419 struct inode *i1, struct inode *i2,
2420 const char *name, size_t namelen,
2421 __u32 mode, __u32 opc, void *data)
2423 LASSERT(i1 != NULL);
2426 /* Do not reuse namelen for something else. */
2428 return ERR_PTR(-EINVAL);
2430 if (namelen > ll_i2sbi(i1)->ll_namelen)
2431 return ERR_PTR(-ENAMETOOLONG);
2433 if (!lu_name_is_valid_2(name, namelen))
2434 return ERR_PTR(-EINVAL);
2437 if (op_data == NULL)
2438 OBD_ALLOC_PTR(op_data);
2440 if (op_data == NULL)
2441 return ERR_PTR(-ENOMEM);
2443 ll_i2gids(op_data->op_suppgids, i1, i2);
2444 op_data->op_fid1 = *ll_inode2fid(i1);
2445 op_data->op_default_stripe_offset = -1;
2446 if (S_ISDIR(i1->i_mode)) {
2447 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2448 if (opc == LUSTRE_OPC_MKDIR)
2449 op_data->op_default_stripe_offset =
2450 ll_i2info(i1)->lli_def_stripe_offset;
2454 op_data->op_fid2 = *ll_inode2fid(i2);
2455 if (S_ISDIR(i2->i_mode))
2456 op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2458 fid_zero(&op_data->op_fid2);
2461 if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2462 op_data->op_cli_flags |= CLI_HASH64;
2464 if (ll_need_32bit_api(ll_i2sbi(i1)))
2465 op_data->op_cli_flags |= CLI_API32;
2467 op_data->op_name = name;
2468 op_data->op_namelen = namelen;
2469 op_data->op_mode = mode;
2470 op_data->op_mod_time = cfs_time_current_sec();
2471 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2472 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2473 op_data->op_cap = cfs_curproc_cap_pack();
2474 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2475 filename_is_volatile(name, namelen, &op_data->op_mds)) {
2476 op_data->op_bias |= MDS_CREATE_VOLATILE;
2478 op_data->op_mds = 0;
2480 op_data->op_data = data;
2485 void ll_finish_md_op_data(struct md_op_data *op_data)
2487 security_release_secctx(op_data->op_file_secctx,
2488 op_data->op_file_secctx_size);
2489 OBD_FREE_PTR(op_data);
2492 #ifdef HAVE_SUPEROPS_USE_DENTRY
2493 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2495 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2498 struct ll_sb_info *sbi;
2500 #ifdef HAVE_SUPEROPS_USE_DENTRY
2501 LASSERT((seq != NULL) && (dentry != NULL));
2502 sbi = ll_s2sbi(dentry->d_sb);
2504 LASSERT((seq != NULL) && (vfs != NULL));
2505 sbi = ll_s2sbi(vfs->mnt_sb);
2508 if (sbi->ll_flags & LL_SBI_NOLCK)
2509 seq_puts(seq, ",nolock");
2511 if (sbi->ll_flags & LL_SBI_FLOCK)
2512 seq_puts(seq, ",flock");
2514 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2515 seq_puts(seq, ",localflock");
2517 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2518 seq_puts(seq, ",user_xattr");
2520 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2521 seq_puts(seq, ",lazystatfs");
2523 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2524 seq_puts(seq, ",user_fid2path");
2526 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2527 seq_puts(seq, ",always_ping");
2533 * Get obd name by cmd, and copy out to user space
2535 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2537 struct ll_sb_info *sbi = ll_i2sbi(inode);
2538 struct obd_device *obd;
2541 if (cmd == OBD_IOC_GETDTNAME)
2542 obd = class_exp2obd(sbi->ll_dt_exp);
2543 else if (cmd == OBD_IOC_GETMDNAME)
2544 obd = class_exp2obd(sbi->ll_md_exp);
2551 if (copy_to_user((void __user *)arg, obd->obd_name,
2552 strlen(obd->obd_name) + 1))
2559 * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2560 * fsname will be returned in this buffer; otherwise, a static buffer will be
2561 * used to store the fsname and returned to caller.
2563 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2565 static char fsname_static[MTI_NAME_MAXLEN];
2566 struct lustre_sb_info *lsi = s2lsi(sb);
2571 /* this means the caller wants to use static buffer
2572 * and it doesn't care about race. Usually this is
2573 * in error reporting path */
2574 buf = fsname_static;
2575 buflen = sizeof(fsname_static);
2578 len = strlen(lsi->lsi_lmd->lmd_profile);
2579 ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2580 if (ptr && (strcmp(ptr, "-client") == 0))
2583 if (unlikely(len >= buflen))
2585 strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2591 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2598 p.mnt = current->fs->root.mnt;
2600 path = d_path(&p, buf, bufsize);
2605 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2607 char *buf, *path = NULL;
2608 struct dentry *dentry = NULL;
2609 struct inode *inode = page->mapping->host;
2611 /* this can be called inside spin lock so use GFP_ATOMIC. */
2612 buf = (char *)__get_free_page(GFP_ATOMIC);
2614 dentry = d_find_alias(page->mapping->host);
2616 path = ll_d_path(dentry, buf, PAGE_SIZE);
2620 "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2621 "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2622 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2623 PFID(ll_inode2fid(inode)),
2624 (path && !IS_ERR(path)) ? path : "", ioret);
2630 free_page((unsigned long)buf);
2633 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2634 struct lov_user_md **kbuf)
2636 struct lov_user_md lum;
2640 if (copy_from_user(&lum, md, sizeof(lum)))
2643 lum_size = ll_lov_user_md_size(&lum);
2647 OBD_ALLOC(*kbuf, lum_size);
2651 if (copy_from_user(*kbuf, md, lum_size) != 0) {
2652 OBD_FREE(*kbuf, lum_size);
2660 * Compute llite root squash state after a change of root squash
2661 * configuration setting or add/remove of a lnet nid
2663 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2665 struct root_squash_info *squash = &sbi->ll_squash;
2668 struct lnet_process_id id;
2670 /* Update norootsquash flag */
2671 down_write(&squash->rsi_sem);
2672 if (list_empty(&squash->rsi_nosquash_nids))
2673 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2675 /* Do not apply root squash as soon as one of our NIDs is
2676 * in the nosquash_nids list */
2679 while (LNetGetId(i++, &id) != -ENOENT) {
2680 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2682 if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2688 sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2690 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2692 up_write(&squash->rsi_sem);
2696 * Parse linkea content to extract information about a given hardlink
2698 * \param[in] ldata - Initialized linkea data
2699 * \param[in] linkno - Link identifier
2700 * \param[out] parent_fid - The entry's parent FID
2701 * \param[out] ln - Entry name destination buffer
2703 * \retval 0 on success
2704 * \retval Appropriate negative error code on failure
2706 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2707 struct lu_fid *parent_fid, struct lu_name *ln)
2713 rc = linkea_init_with_rec(ldata);
2717 if (linkno >= ldata->ld_leh->leh_reccount)
2718 /* beyond last link */
2721 linkea_first_entry(ldata);
2722 for (idx = 0; ldata->ld_lee != NULL; idx++) {
2723 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2728 linkea_next_entry(ldata);
2738 * Get parent FID and name of an identified link. Operation is performed for
2739 * a given link number, letting the caller iterate over linkno to list one or
2740 * all links of an entry.
2742 * \param[in] file - File descriptor against which to perform the operation
2743 * \param[in,out] arg - User-filled structure containing the linkno to operate
2744 * on and the available size. It is eventually filled with
2745 * the requested information or left untouched on error
2747 * \retval - 0 on success
2748 * \retval - Appropriate negative error code on failure
2750 int ll_getparent(struct file *file, struct getparent __user *arg)
2752 struct inode *inode = file_inode(file);
2753 struct linkea_data *ldata;
2754 struct lu_buf buf = LU_BUF_NULL;
2756 struct lu_fid parent_fid;
2763 if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2764 !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2767 if (get_user(name_size, &arg->gp_name_size))
2770 if (get_user(linkno, &arg->gp_linkno))
2773 if (name_size > PATH_MAX)
2776 OBD_ALLOC(ldata, sizeof(*ldata));
2780 rc = linkea_data_new(ldata, &buf);
2782 GOTO(ldata_free, rc);
2784 #ifdef HAVE_XATTR_HANDLER_FLAGS
2785 rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
2786 buf.lb_len, OBD_MD_FLXATTR);
2788 rc = ll_getxattr(file_dentry(file), XATTR_NAME_LINK, buf.lb_buf,
2790 #endif /* HAVE_XATTR_HANDLER_FLAGS */
2794 rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2798 if (ln.ln_namelen >= name_size)
2799 GOTO(lb_free, rc = -EOVERFLOW);
2801 if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2802 GOTO(lb_free, rc = -EFAULT);
2804 if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2805 GOTO(lb_free, rc = -EFAULT);
2807 if (put_user('\0', arg->gp_name + ln.ln_namelen))
2808 GOTO(lb_free, rc = -EFAULT);
2813 OBD_FREE(ldata, sizeof(*ldata));