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) 2003, 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.
36 * lustre/llite/llite_lib.c
38 * Lustre Light Super operations
41 #define DEBUG_SUBSYSTEM S_LLITE
43 #include <linux/module.h>
44 #include <linux/statfs.h>
45 #include <linux/types.h>
46 #include <linux/version.h>
48 #include <linux/user_namespace.h>
49 #ifdef HAVE_UIDGID_HEADER
50 # include <linux/uidgid.h>
53 #include <lustre_ioctl.h>
54 #include <lustre_ha.h>
55 #include <lustre_dlm.h>
56 #include <lprocfs_status.h>
57 #include <lustre_disk.h>
58 #include <lustre_param.h>
59 #include <lustre_log.h>
60 #include <cl_object.h>
61 #include <obd_cksum.h>
62 #include "llite_internal.h"
64 struct kmem_cache *ll_file_data_slab;
66 static struct list_head ll_super_blocks = LIST_HEAD_INIT(ll_super_blocks);
67 static DEFINE_SPINLOCK(ll_sb_lock);
70 #define log2(n) ffz(~(n))
73 static struct ll_sb_info *ll_init_sbi(void)
75 struct ll_sb_info *sbi = NULL;
77 unsigned long lru_page_max;
87 spin_lock_init(&sbi->ll_lock);
88 mutex_init(&sbi->ll_lco.lco_lock);
89 spin_lock_init(&sbi->ll_pp_extent_lock);
90 spin_lock_init(&sbi->ll_process_lock);
91 sbi->ll_rw_stats_on = 0;
94 pages = si.totalram - si.totalhigh;
95 lru_page_max = pages / 2;
97 /* initialize ll_cache data */
98 sbi->ll_cache = cl_cache_init(lru_page_max);
99 if (sbi->ll_cache == NULL) {
100 OBD_FREE(sbi, sizeof(*sbi));
104 sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
105 SBI_DEFAULT_READAHEAD_MAX);
106 sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
107 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
108 SBI_DEFAULT_READAHEAD_WHOLE_MAX;
109 INIT_LIST_HEAD(&sbi->ll_conn_chain);
110 INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
112 ll_generate_random_uuid(uuid);
113 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
114 CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
116 spin_lock(&ll_sb_lock);
117 list_add_tail(&sbi->ll_list, &ll_super_blocks);
118 spin_unlock(&ll_sb_lock);
120 sbi->ll_flags |= LL_SBI_VERBOSE;
121 #ifdef ENABLE_CHECKSUM
122 sbi->ll_flags |= LL_SBI_CHECKSUM;
125 #ifdef HAVE_LRU_RESIZE_SUPPORT
126 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
129 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
130 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
132 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
136 /* metadata statahead is enabled by default */
137 sbi->ll_sa_max = LL_SA_RPC_DEF;
138 atomic_set(&sbi->ll_sa_total, 0);
139 atomic_set(&sbi->ll_sa_wrong, 0);
140 atomic_set(&sbi->ll_sa_running, 0);
141 atomic_set(&sbi->ll_agl_total, 0);
142 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
145 sbi->ll_squash.rsi_uid = 0;
146 sbi->ll_squash.rsi_gid = 0;
147 INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
148 init_rwsem(&sbi->ll_squash.rsi_sem);
153 static void ll_free_sbi(struct super_block *sb)
155 struct ll_sb_info *sbi = ll_s2sbi(sb);
159 spin_lock(&ll_sb_lock);
160 list_del(&sbi->ll_list);
161 spin_unlock(&ll_sb_lock);
162 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
163 cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
164 if (sbi->ll_cache != NULL) {
165 cl_cache_decref(sbi->ll_cache);
166 sbi->ll_cache = NULL;
168 OBD_FREE(sbi, sizeof(*sbi));
173 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
174 struct vfsmount *mnt)
176 struct inode *root = NULL;
177 struct ll_sb_info *sbi = ll_s2sbi(sb);
178 struct obd_device *obd;
179 struct obd_capa *oc = NULL;
180 struct obd_statfs *osfs = NULL;
181 struct ptlrpc_request *request = NULL;
182 struct obd_connect_data *data = NULL;
183 struct obd_uuid *uuid;
184 struct md_op_data *op_data;
185 struct lustre_md lmd;
187 int size, err, checksum;
190 obd = class_name2obd(md);
192 CERROR("MD %s: not setup or attached\n", md);
206 /* indicate the features supported by this client */
207 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
208 OBD_CONNECT_ATTRFID |
209 OBD_CONNECT_VERSION | OBD_CONNECT_BRW_SIZE |
210 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
211 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
212 OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 |
213 OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR |
214 OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH|
215 OBD_CONNECT_EINPROGRESS |
216 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
217 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS |
218 OBD_CONNECT_MAX_EASIZE |
219 OBD_CONNECT_FLOCK_DEAD |
220 OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
221 OBD_CONNECT_OPEN_BY_FID |
222 OBD_CONNECT_DIR_STRIPE;
224 #ifdef HAVE_LRU_RESIZE_SUPPORT
225 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
226 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
228 #ifdef CONFIG_FS_POSIX_ACL
229 data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK;
232 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
233 /* flag mdc connection as lightweight, only used for test
234 * purpose, use with care */
235 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
237 data->ocd_ibits_known = MDS_INODELOCK_FULL;
238 data->ocd_version = LUSTRE_VERSION_CODE;
240 if (sb->s_flags & MS_RDONLY)
241 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
242 if (sbi->ll_flags & LL_SBI_USER_XATTR)
243 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
245 #ifdef HAVE_MS_FLOCK_LOCK
246 /* force vfs to use lustre handler for flock() calls - bug 10743 */
247 sb->s_flags |= MS_FLOCK_LOCK;
249 #ifdef MS_HAS_NEW_AOPS
250 sb->s_flags |= MS_HAS_NEW_AOPS;
253 if (sbi->ll_flags & LL_SBI_FLOCK)
254 sbi->ll_fop = &ll_file_operations_flock;
255 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
256 sbi->ll_fop = &ll_file_operations;
258 sbi->ll_fop = &ll_file_operations_noflock;
261 data->ocd_connect_flags |= OBD_CONNECT_REAL;
262 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
263 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
265 data->ocd_brw_size = MD_MAX_BRW_SIZE;
267 err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
269 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
270 "recovery, of which this client is not a "
271 "part. Please wait for recovery to complete,"
272 " abort, or time out.\n", md);
275 CERROR("cannot connect to %s: rc = %d\n", md, err);
279 sbi->ll_md_exp->exp_connect_data = *data;
281 err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
282 LUSTRE_SEQ_METADATA);
284 CERROR("%s: Can't init metadata layer FID infrastructure, "
285 "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
289 /* For mount, we only need fs info from MDT0, and also in DNE, it
290 * can make sure the client can be mounted as long as MDT0 is
292 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
293 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
294 OBD_STATFS_FOR_MDT0);
296 GOTO(out_md_fid, err);
298 /* This needs to be after statfs to ensure connect has finished.
299 * Note that "data" does NOT contain the valid connect reply.
300 * If connecting to a 1.8 server there will be no LMV device, so
301 * we can access the MDC export directly and exp_connect_flags will
302 * be non-zero, but if accessing an upgraded 2.1 server it will
303 * have the correct flags filled in.
304 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
305 valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
306 if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
307 valid != CLIENT_CONNECT_MDT_REQD) {
310 OBD_ALLOC_WAIT(buf, PAGE_CACHE_SIZE);
311 obd_connect_flags2str(buf, PAGE_CACHE_SIZE,
312 valid ^ CLIENT_CONNECT_MDT_REQD, ",");
313 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
314 "feature(s) needed for correct operation "
315 "of this client (%s). Please upgrade "
316 "server or downgrade client.\n",
317 sbi->ll_md_exp->exp_obd->obd_name, buf);
318 OBD_FREE(buf, PAGE_CACHE_SIZE);
319 GOTO(out_md_fid, err = -EPROTO);
322 size = sizeof(*data);
323 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
324 KEY_CONN_DATA, &size, data);
326 CERROR("%s: Get connect data failed: rc = %d\n",
327 sbi->ll_md_exp->exp_obd->obd_name, err);
328 GOTO(out_md_fid, err);
331 LASSERT(osfs->os_bsize);
332 sb->s_blocksize = osfs->os_bsize;
333 sb->s_blocksize_bits = log2(osfs->os_bsize);
334 sb->s_magic = LL_SUPER_MAGIC;
335 sb->s_maxbytes = MAX_LFS_FILESIZE;
336 sbi->ll_namelen = osfs->os_namelen;
338 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
339 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
340 LCONSOLE_INFO("Disabling user_xattr feature because "
341 "it is not supported on the server\n");
342 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
345 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
347 sb->s_flags |= MS_POSIXACL;
349 sbi->ll_flags |= LL_SBI_ACL;
351 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
353 sb->s_flags &= ~MS_POSIXACL;
355 sbi->ll_flags &= ~LL_SBI_ACL;
358 if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
359 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
360 sbi->ll_flags |= LL_SBI_RMT_CLIENT;
361 LCONSOLE_INFO("client is set as remote by default.\n");
364 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
365 sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
366 LCONSOLE_INFO("client claims to be remote, but server "
367 "rejected, forced to be local.\n");
371 if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
372 LCONSOLE_INFO("client enabled MDS capability!\n");
373 sbi->ll_flags |= LL_SBI_MDS_CAPA;
376 if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
377 LCONSOLE_INFO("client enabled OSS capability!\n");
378 sbi->ll_flags |= LL_SBI_OSS_CAPA;
381 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
382 sbi->ll_flags |= LL_SBI_64BIT_HASH;
384 if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
385 sbi->ll_md_brw_pages = data->ocd_brw_size >> PAGE_CACHE_SHIFT;
387 sbi->ll_md_brw_pages = 1;
389 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
390 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
392 if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
393 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
394 LCONSOLE_INFO("%s: disabling xattr cache due to "
395 "unknown maximum xattr size.\n", dt);
397 sbi->ll_flags |= LL_SBI_XATTR_CACHE;
398 sbi->ll_xattr_cache_enabled = 1;
402 obd = class_name2obd(dt);
404 CERROR("DT %s: not setup or attached\n", dt);
405 GOTO(out_md_fid, err = -ENODEV);
408 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
409 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
410 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
411 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
412 OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
413 OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
414 OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
415 OBD_CONNECT_MAXBYTES |
416 OBD_CONNECT_EINPROGRESS |
417 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
418 OBD_CONNECT_LAYOUTLOCK |
419 OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK;
421 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
422 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
423 * disabled by default, because it can still be enabled on the
424 * fly via /proc. As a consequence, we still need to come to an
425 * agreement on the supported algorithms at connect time */
426 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
428 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
429 data->ocd_cksum_types = OBD_CKSUM_ADLER;
431 data->ocd_cksum_types = cksum_types_supported_client();
434 #ifdef HAVE_LRU_RESIZE_SUPPORT
435 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
437 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
438 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
440 CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
441 "ocd_grant: %d\n", data->ocd_connect_flags,
442 data->ocd_version, data->ocd_grant);
444 obd->obd_upcall.onu_owner = &sbi->ll_lco;
445 obd->obd_upcall.onu_upcall = cl_ocd_update;
447 data->ocd_brw_size = DT_MAX_BRW_SIZE;
449 err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
452 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
453 "recovery, of which this client is not a "
454 "part. Please wait for recovery to "
455 "complete, abort, or time out.\n", dt);
458 CERROR("%s: Cannot connect to %s: rc = %d\n",
459 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
463 sbi->ll_dt_exp->exp_connect_data = *data;
465 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
466 LUSTRE_SEQ_METADATA);
468 CERROR("%s: Can't init data layer FID infrastructure, "
469 "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
473 mutex_lock(&sbi->ll_lco.lco_lock);
474 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
475 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
476 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
477 mutex_unlock(&sbi->ll_lco.lco_lock);
479 fid_zero(&sbi->ll_root_fid);
480 err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
482 CERROR("cannot mds_connect: rc = %d\n", err);
483 GOTO(out_lock_cn_cb, err);
485 if (!fid_is_sane(&sbi->ll_root_fid)) {
486 CERROR("%s: Invalid root fid "DFID" during mount\n",
487 sbi->ll_md_exp->exp_obd->obd_name,
488 PFID(&sbi->ll_root_fid));
489 GOTO(out_lock_cn_cb, err = -EINVAL);
491 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
493 sb->s_op = &lustre_super_operations;
494 #if THREAD_SIZE >= 8192 /*b=17630*/
495 sb->s_export_op = &lustre_export_operations;
499 * XXX: move this to after cbd setup? */
500 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA |
502 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
503 valid |= OBD_MD_FLRMTPERM;
504 else if (sbi->ll_flags & LL_SBI_ACL)
505 valid |= OBD_MD_FLACL;
507 OBD_ALLOC_PTR(op_data);
509 GOTO(out_lock_cn_cb, err = -ENOMEM);
511 op_data->op_fid1 = sbi->ll_root_fid;
512 op_data->op_mode = 0;
513 op_data->op_capa1 = oc;
514 op_data->op_valid = valid;
516 err = md_getattr(sbi->ll_md_exp, op_data, &request);
519 OBD_FREE_PTR(op_data);
521 CERROR("%s: md_getattr failed for root: rc = %d\n",
522 sbi->ll_md_exp->exp_obd->obd_name, err);
523 GOTO(out_lock_cn_cb, err);
526 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
527 sbi->ll_md_exp, &lmd);
529 CERROR("failed to understand root inode md: rc = %d\n", err);
530 ptlrpc_req_finished(request);
531 GOTO(out_lock_cn_cb, err);
534 LASSERT(fid_is_sane(&sbi->ll_root_fid));
535 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
536 sbi->ll_flags & LL_SBI_32BIT_API),
538 md_free_lustre_md(sbi->ll_md_exp, &lmd);
539 ptlrpc_req_finished(request);
543 obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
544 #ifdef CONFIG_FS_POSIX_ACL
546 posix_acl_release(lmd.posix_acl);
547 lmd.posix_acl = NULL;
550 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
552 CERROR("lustre_lite: bad iget4 for root\n");
556 #ifdef CONFIG_FS_POSIX_ACL
557 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
558 rct_init(&sbi->ll_rct);
559 et_init(&sbi->ll_et);
563 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
564 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
565 KEY_CHECKSUM, sizeof(checksum), &checksum,
569 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
570 KEY_CACHE_SET, sizeof(*sbi->ll_cache),
571 sbi->ll_cache, NULL);
573 sb->s_root = d_make_root(root);
574 if (sb->s_root == NULL) {
575 CERROR("%s: can't make root dentry\n",
576 ll_get_fsname(sb, NULL, 0));
577 GOTO(out_root, err = -ENOMEM);
579 #ifdef HAVE_DCACHE_LOCK
580 sb->s_root->d_op = &ll_d_ops;
583 sbi->ll_sdev_orig = sb->s_dev;
585 /* We set sb->s_dev equal on all lustre clients in order to support
586 * NFS export clustering. NFSD requires that the FSID be the same
588 /* s_dev is also used in lt_compare() to compare two fs, but that is
589 * only a node-local comparison. */
590 uuid = obd_get_uuid(sbi->ll_md_exp);
592 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
598 if (proc_lustre_fs_root != NULL) {
599 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
602 CERROR("%s: could not register mount in lprocfs: "
603 "rc = %d\n", ll_get_fsname(sb, NULL, 0), err);
613 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
615 obd_disconnect(sbi->ll_dt_exp);
616 sbi->ll_dt_exp = NULL;
618 obd_fid_fini(sbi->ll_md_exp->exp_obd);
620 obd_disconnect(sbi->ll_md_exp);
621 sbi->ll_md_exp = NULL;
630 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
634 size = sizeof(*lmmsize);
635 rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
636 KEY_MAX_EASIZE, &size, lmmsize);
638 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
639 sbi->ll_dt_exp->exp_obd->obd_name, rc);
644 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
645 KEY_MAX_EASIZE, &size, lmmsize);
647 CERROR("Get max mdsize error rc %d\n", rc);
653 * Get the value of the default_easize parameter.
655 * \see client_obd::cl_default_mds_easize
657 * \param[in] sbi superblock info for this filesystem
658 * \param[out] lmmsize pointer to storage location for value
660 * \retval 0 on success
661 * \retval negative negated errno on failure
663 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
668 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
669 KEY_DEFAULT_EASIZE, &size, lmmsize);
671 CERROR("Get default mdsize error rc %d\n", rc);
677 * Set the default_easize parameter to the given value.
679 * \see client_obd::cl_default_mds_easize
681 * \param[in] sbi superblock info for this filesystem
682 * \param[in] lmmsize the size to set
684 * \retval 0 on success
685 * \retval negative negated errno on failure
687 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
691 if (lmmsize < sizeof(struct lov_mds_md) ||
692 lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
695 rc = obd_set_info_async(NULL, sbi->ll_md_exp,
696 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
697 sizeof(int), &lmmsize, NULL);
702 static void ll_dump_inode(struct inode *inode)
704 struct ll_d_hlist_node *tmp;
705 int dentry_count = 0;
707 LASSERT(inode != NULL);
709 ll_d_hlist_for_each(tmp, &inode->i_dentry)
712 CERROR("%s: inode %p dump: dev=%s fid="DFID
713 " mode=%o count=%u, %d dentries\n",
714 ll_get_fsname(inode->i_sb, NULL, 0), inode,
715 ll_i2mdexp(inode)->exp_obd->obd_name, PFID(ll_inode2fid(inode)),
716 inode->i_mode, atomic_read(&inode->i_count), dentry_count);
719 void lustre_dump_dentry(struct dentry *dentry, int recur)
721 struct list_head *tmp;
724 LASSERT(dentry != NULL);
726 list_for_each(tmp, &dentry->d_subdirs)
729 CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
730 " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
731 dentry->d_name.len, dentry->d_name.name,
732 dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
733 dentry->d_parent, dentry->d_inode, ll_d_count(dentry),
734 dentry->d_flags, dentry->d_fsdata, subdirs);
735 if (dentry->d_inode != NULL)
736 ll_dump_inode(dentry->d_inode);
741 list_for_each(tmp, &dentry->d_subdirs) {
742 struct dentry *d = list_entry(tmp, struct dentry, d_child);
743 lustre_dump_dentry(d, recur - 1);
747 static void client_common_put_super(struct super_block *sb)
749 struct ll_sb_info *sbi = ll_s2sbi(sb);
752 #ifdef CONFIG_FS_POSIX_ACL
753 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
754 et_fini(&sbi->ll_et);
755 rct_fini(&sbi->ll_rct);
761 list_del(&sbi->ll_conn_chain);
763 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
764 obd_disconnect(sbi->ll_dt_exp);
765 sbi->ll_dt_exp = NULL;
767 lprocfs_unregister_mountpoint(sbi);
769 obd_fid_fini(sbi->ll_md_exp->exp_obd);
770 obd_disconnect(sbi->ll_md_exp);
771 sbi->ll_md_exp = NULL;
776 void ll_kill_super(struct super_block *sb)
778 struct ll_sb_info *sbi;
782 if (!(sb->s_flags & MS_ACTIVE))
786 /* we need restore s_dev from changed for clustred NFS before put_super
787 * because new kernels have cached s_dev and change sb->s_dev in
788 * put_super not affected real removing devices */
790 sb->s_dev = sbi->ll_sdev_orig;
791 sbi->ll_umounting = 1;
793 /* wait running statahead threads to quit */
794 while (atomic_read(&sbi->ll_sa_running) > 0) {
795 set_current_state(TASK_UNINTERRUPTIBLE);
796 schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
803 static inline int ll_set_opt(const char *opt, char *data, int fl)
805 if (strncmp(opt, data, strlen(opt)) != 0)
811 /* non-client-specific mount options are parsed in lmd_parse */
812 static int ll_options(char *options, int *flags)
815 char *s1 = options, *s2;
821 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
824 CDEBUG(D_SUPER, "next opt=%s\n", s1);
825 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
830 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
835 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
840 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
845 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
850 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
855 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
860 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
865 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
871 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
876 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
881 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
886 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
891 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
896 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
901 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
906 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
911 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
916 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
922 s2 = strchr(s1, ',');
930 void ll_lli_init(struct ll_inode_info *lli)
932 lli->lli_inode_magic = LLI_INODE_MAGIC;
934 spin_lock_init(&lli->lli_lock);
935 lli->lli_posix_acl = NULL;
936 lli->lli_remote_perms = NULL;
937 mutex_init(&lli->lli_rmtperm_mutex);
938 /* Do not set lli_fid, it has been initialized already. */
939 fid_zero(&lli->lli_pfid);
940 INIT_LIST_HEAD(&lli->lli_oss_capas);
941 atomic_set(&lli->lli_open_count, 0);
942 lli->lli_mds_capa = NULL;
943 lli->lli_rmtperm_time = 0;
944 lli->lli_mds_read_och = NULL;
945 lli->lli_mds_write_och = NULL;
946 lli->lli_mds_exec_och = NULL;
947 lli->lli_open_fd_read_count = 0;
948 lli->lli_open_fd_write_count = 0;
949 lli->lli_open_fd_exec_count = 0;
950 mutex_init(&lli->lli_och_mutex);
951 spin_lock_init(&lli->lli_agl_lock);
952 spin_lock_init(&lli->lli_layout_lock);
953 ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
954 lli->lli_clob = NULL;
956 init_rwsem(&lli->lli_xattrs_list_rwsem);
957 mutex_init(&lli->lli_xattrs_enq_lock);
959 LASSERT(lli->lli_vfs_inode.i_mode != 0);
960 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
961 mutex_init(&lli->lli_readdir_mutex);
962 lli->lli_opendir_key = NULL;
964 spin_lock_init(&lli->lli_sa_lock);
965 lli->lli_opendir_pid = 0;
966 lli->lli_sa_enabled = 0;
967 lli->lli_def_stripe_offset = -1;
969 mutex_init(&lli->lli_size_mutex);
970 lli->lli_symlink_name = NULL;
971 init_rwsem(&lli->lli_trunc_sem);
972 range_lock_tree_init(&lli->lli_write_tree);
973 init_rwsem(&lli->lli_glimpse_sem);
974 lli->lli_glimpse_time = 0;
975 INIT_LIST_HEAD(&lli->lli_agl_list);
976 lli->lli_agl_index = 0;
977 lli->lli_async_rc = 0;
979 mutex_init(&lli->lli_layout_mutex);
982 static inline int ll_bdi_register(struct backing_dev_info *bdi)
984 static atomic_t ll_bdi_num = ATOMIC_INIT(0);
986 bdi->name = "lustre";
987 return bdi_register(bdi, NULL, "lustre-%d",
988 atomic_inc_return(&ll_bdi_num));
991 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
993 struct lustre_profile *lprof = NULL;
994 struct lustre_sb_info *lsi = s2lsi(sb);
995 struct ll_sb_info *sbi;
996 char *dt = NULL, *md = NULL;
997 char *profilenm = get_profile_name(sb);
998 struct config_llog_instance *cfg;
999 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1000 const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
1004 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1010 try_module_get(THIS_MODULE);
1012 /* client additional sb info */
1013 lsi->lsi_llsbi = sbi = ll_init_sbi();
1015 module_put(THIS_MODULE);
1020 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1022 GOTO(out_free, err);
1024 err = bdi_init(&lsi->lsi_bdi);
1026 GOTO(out_free, err);
1027 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1028 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1029 err = ll_bdi_register(&lsi->lsi_bdi);
1031 GOTO(out_free, err);
1033 sb->s_bdi = &lsi->lsi_bdi;
1034 #ifndef HAVE_DCACHE_LOCK
1035 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1036 sb->s_d_op = &ll_d_ops;
1039 /* Generate a string unique to this super, in case some joker tries
1040 to mount the same fs at two mount points.
1041 Use the address of the super itself.*/
1042 cfg->cfg_instance = sb;
1043 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1044 cfg->cfg_callback = class_config_llog_handler;
1045 /* set up client obds */
1046 err = lustre_process_log(sb, profilenm, cfg);
1048 GOTO(out_free, err);
1050 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1051 lprof = class_get_profile(profilenm);
1052 if (lprof == NULL) {
1053 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1054 " read from the MGS. Does that filesystem "
1055 "exist?\n", profilenm);
1056 GOTO(out_free, err = -EINVAL);
1058 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1059 lprof->lp_md, lprof->lp_dt);
1061 OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1063 GOTO(out_free, err = -ENOMEM);
1064 sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1066 OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1068 GOTO(out_free, err = -ENOMEM);
1069 sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1071 /* connections, registrations, sb setup */
1072 err = client_common_fill_super(sb, md, dt, mnt);
1074 GOTO(out_free, err);
1076 sbi->ll_client_common_fill_super_succeeded = 1;
1080 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1082 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1085 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1086 LCONSOLE_WARN("Mounted %s\n", profilenm);
1090 } /* ll_fill_super */
1092 void ll_put_super(struct super_block *sb)
1094 struct config_llog_instance cfg, params_cfg;
1095 struct obd_device *obd;
1096 struct lustre_sb_info *lsi = s2lsi(sb);
1097 struct ll_sb_info *sbi = ll_s2sbi(sb);
1098 char *profilenm = get_profile_name(sb);
1100 int next, force = 1, rc = 0;
1103 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1105 ll_print_capa_stat(sbi);
1107 cfg.cfg_instance = sb;
1108 lustre_end_log(sb, profilenm, &cfg);
1110 params_cfg.cfg_instance = sb;
1111 lustre_end_log(sb, PARAMS_FILENAME, ¶ms_cfg);
1113 if (sbi->ll_md_exp) {
1114 obd = class_exp2obd(sbi->ll_md_exp);
1116 force = obd->obd_force;
1119 /* Wait for unstable pages to be committed to stable storage */
1121 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1122 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1123 atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1127 ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1128 if (force == 0 && rc != -EINTR)
1129 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1132 /* We need to set force before the lov_disconnect in
1133 lustre_common_put_super, since l_d cleans up osc's as well. */
1136 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1138 obd->obd_force = force;
1142 if (sbi->ll_client_common_fill_super_succeeded) {
1143 /* Only if client_common_fill_super succeeded */
1144 client_common_put_super(sb);
1148 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1149 class_manual_cleanup(obd);
1152 if (sbi->ll_flags & LL_SBI_VERBOSE)
1153 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1156 class_del_profile(profilenm);
1158 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1159 bdi_destroy(&lsi->lsi_bdi);
1160 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1164 lsi->lsi_llsbi = NULL;
1166 lustre_common_put_super(sb);
1168 cl_env_cache_purge(~0);
1170 module_put(THIS_MODULE);
1173 } /* client_put_super */
1175 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1177 struct inode *inode = NULL;
1179 /* NOTE: we depend on atomic igrab() -bzzz */
1180 lock_res_and_lock(lock);
1181 if (lock->l_resource->lr_lvb_inode) {
1182 struct ll_inode_info * lli;
1183 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1184 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1185 inode = igrab(lock->l_resource->lr_lvb_inode);
1187 inode = lock->l_resource->lr_lvb_inode;
1188 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1189 D_WARNING, lock, "lr_lvb_inode %p is "
1190 "bogus: magic %08x",
1191 lock->l_resource->lr_lvb_inode,
1192 lli->lli_inode_magic);
1196 unlock_res_and_lock(lock);
1200 static void ll_dir_clear_lsm_md(struct inode *inode)
1202 struct ll_inode_info *lli = ll_i2info(inode);
1204 LASSERT(S_ISDIR(inode->i_mode));
1206 if (lli->lli_lsm_md != NULL) {
1207 lmv_free_memmd(lli->lli_lsm_md);
1208 lli->lli_lsm_md = NULL;
1212 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1213 const struct lu_fid *fid,
1214 struct lustre_md *md)
1216 struct ll_sb_info *sbi = ll_s2sbi(sb);
1217 struct mdt_body *body = md->body;
1218 struct inode *inode;
1222 ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1223 inode = iget_locked(sb, ino);
1224 if (inode == NULL) {
1225 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1226 ll_get_fsname(sb, NULL, 0), PFID(fid));
1227 RETURN(ERR_PTR(-ENOENT));
1230 if (inode->i_state & I_NEW) {
1231 struct ll_inode_info *lli = ll_i2info(inode);
1232 struct lmv_stripe_md *lsm = md->lmv;
1234 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1235 (body->mbo_mode & S_IFMT);
1236 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1239 LTIME_S(inode->i_mtime) = 0;
1240 LTIME_S(inode->i_atime) = 0;
1241 LTIME_S(inode->i_ctime) = 0;
1244 /* initializing backing dev info. */
1245 inode->i_mapping->backing_dev_info =
1246 &s2lsi(inode->i_sb)->lsi_bdi;
1247 inode->i_op = &ll_dir_inode_operations;
1248 inode->i_fop = &ll_dir_operations;
1249 lli->lli_fid = *fid;
1252 LASSERT(lsm != NULL);
1253 /* master object FID */
1254 lli->lli_pfid = body->mbo_fid1;
1255 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1256 lli, PFID(fid), PFID(&lli->lli_pfid));
1257 unlock_new_inode(inode);
1263 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1266 struct lmv_stripe_md *lsm = md->lmv;
1269 LASSERT(lsm != NULL);
1270 /* XXX sigh, this lsm_root initialization should be in
1271 * LMV layer, but it needs ll_iget right now, so we
1272 * put this here right now. */
1273 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1274 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1275 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1276 /* Unfortunately ll_iget will call ll_update_inode,
1277 * where the initialization of slave inode is slightly
1278 * different, so it reset lsm_md to NULL to avoid
1279 * initializing lsm for slave inode. */
1280 /* For migrating inode, master stripe and master object will
1281 * be same, so we only need assign this inode */
1282 if (lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION && i == 0)
1283 lsm->lsm_md_oinfo[i].lmo_root = inode;
1285 lsm->lsm_md_oinfo[i].lmo_root =
1286 ll_iget_anon_dir(inode->i_sb, fid, md);
1288 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1289 int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1291 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1299 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1300 const struct lmv_stripe_md *lsm_md2)
1302 return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1303 lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1304 lsm_md1->lsm_md_master_mdt_index ==
1305 lsm_md2->lsm_md_master_mdt_index &&
1306 lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1307 lsm_md1->lsm_md_layout_version ==
1308 lsm_md2->lsm_md_layout_version &&
1309 strcmp(lsm_md1->lsm_md_pool_name,
1310 lsm_md2->lsm_md_pool_name) == 0;
1313 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1315 struct ll_inode_info *lli = ll_i2info(inode);
1316 struct lmv_stripe_md *lsm = md->lmv;
1320 LASSERT(S_ISDIR(inode->i_mode));
1321 CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1322 PFID(ll_inode2fid(inode)));
1324 /* no striped information from request. */
1326 if (lli->lli_lsm_md == NULL) {
1328 } else if (lli->lli_lsm_md->lsm_md_hash_type &
1329 LMV_HASH_FLAG_MIGRATION) {
1330 /* migration is done, the temporay MIGRATE layout has
1332 CDEBUG(D_INODE, DFID" finish migration.\n",
1333 PFID(ll_inode2fid(inode)));
1334 lmv_free_memmd(lli->lli_lsm_md);
1335 lli->lli_lsm_md = NULL;
1338 /* The lustre_md from req does not include stripeEA,
1339 * see ll_md_setattr */
1344 /* set the directory layout */
1345 if (lli->lli_lsm_md == NULL) {
1347 rc = ll_init_lsm_md(inode, md);
1351 lli->lli_lsm_md = lsm;
1352 /* set lsm_md to NULL, so the following free lustre_md
1353 * will not free this lsm */
1355 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1356 lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1360 /* Compare the old and new stripe information */
1361 if (!lsm_md_eq(lli->lli_lsm_md, lsm)) {
1362 struct lmv_stripe_md *old_lsm = lli->lli_lsm_md;
1365 CERROR("%s: inode "DFID"(%p)'s lmv layout mismatch (%p)/(%p)"
1366 "magic:0x%x/0x%x stripe count: %d/%d master_mdt: %d/%d"
1367 "hash_type:0x%x/0x%x layout: 0x%x/0x%x pool:%s/%s\n",
1368 ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1369 inode, lsm, old_lsm,
1370 lsm->lsm_md_magic, old_lsm->lsm_md_magic,
1371 lsm->lsm_md_stripe_count,
1372 old_lsm->lsm_md_stripe_count,
1373 lsm->lsm_md_master_mdt_index,
1374 old_lsm->lsm_md_master_mdt_index,
1375 lsm->lsm_md_hash_type, old_lsm->lsm_md_hash_type,
1376 lsm->lsm_md_layout_version,
1377 old_lsm->lsm_md_layout_version,
1378 lsm->lsm_md_pool_name,
1379 old_lsm->lsm_md_pool_name);
1381 for (idx = 0; idx < old_lsm->lsm_md_stripe_count; idx++) {
1382 CERROR("%s: sub FIDs in old lsm idx %d, old: "DFID"\n",
1383 ll_get_fsname(inode->i_sb, NULL, 0), idx,
1384 PFID(&old_lsm->lsm_md_oinfo[idx].lmo_fid));
1387 for (idx = 0; idx < lsm->lsm_md_stripe_count; idx++) {
1388 CERROR("%s: sub FIDs in new lsm idx %d, new: "DFID"\n",
1389 ll_get_fsname(inode->i_sb, NULL, 0), idx,
1390 PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1399 void ll_clear_inode(struct inode *inode)
1401 struct ll_inode_info *lli = ll_i2info(inode);
1402 struct ll_sb_info *sbi = ll_i2sbi(inode);
1405 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1406 PFID(ll_inode2fid(inode)), inode);
1408 if (S_ISDIR(inode->i_mode)) {
1409 /* these should have been cleared in ll_file_release */
1410 LASSERT(lli->lli_opendir_key == NULL);
1411 LASSERT(lli->lli_sai == NULL);
1412 LASSERT(lli->lli_opendir_pid == 0);
1415 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1417 LASSERT(!lli->lli_open_fd_write_count);
1418 LASSERT(!lli->lli_open_fd_read_count);
1419 LASSERT(!lli->lli_open_fd_exec_count);
1421 if (lli->lli_mds_write_och)
1422 ll_md_real_close(inode, FMODE_WRITE);
1423 if (lli->lli_mds_exec_och)
1424 ll_md_real_close(inode, FMODE_EXEC);
1425 if (lli->lli_mds_read_och)
1426 ll_md_real_close(inode, FMODE_READ);
1428 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1429 OBD_FREE(lli->lli_symlink_name,
1430 strlen(lli->lli_symlink_name) + 1);
1431 lli->lli_symlink_name = NULL;
1434 ll_xattr_cache_destroy(inode);
1436 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1437 LASSERT(lli->lli_posix_acl == NULL);
1438 if (lli->lli_remote_perms) {
1439 free_rmtperm_hash(lli->lli_remote_perms);
1440 lli->lli_remote_perms = NULL;
1443 #ifdef CONFIG_FS_POSIX_ACL
1444 else if (lli->lli_posix_acl) {
1445 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1446 LASSERT(lli->lli_remote_perms == NULL);
1447 posix_acl_release(lli->lli_posix_acl);
1448 lli->lli_posix_acl = NULL;
1451 lli->lli_inode_magic = LLI_INODE_DEAD;
1453 ll_clear_inode_capas(inode);
1454 if (S_ISDIR(inode->i_mode))
1455 ll_dir_clear_lsm_md(inode);
1456 else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1457 LASSERT(list_empty(&lli->lli_agl_list));
1460 * XXX This has to be done before lsm is freed below, because
1461 * cl_object still uses inode lsm.
1463 cl_inode_fini(inode);
1468 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1470 struct lustre_md md;
1471 struct inode *inode = dentry->d_inode;
1472 struct ll_sb_info *sbi = ll_i2sbi(inode);
1473 struct ptlrpc_request *request = NULL;
1477 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1478 LUSTRE_OPC_ANY, NULL);
1479 if (IS_ERR(op_data))
1480 RETURN(PTR_ERR(op_data));
1482 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1484 ptlrpc_req_finished(request);
1485 if (rc == -ENOENT) {
1487 /* Unlinked special device node? Or just a race?
1488 * Pretend we done everything. */
1489 if (!S_ISREG(inode->i_mode) &&
1490 !S_ISDIR(inode->i_mode)) {
1491 ia_valid = op_data->op_attr.ia_valid;
1492 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1493 rc = simple_setattr(dentry, &op_data->op_attr);
1494 op_data->op_attr.ia_valid = ia_valid;
1496 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1497 CERROR("md_setattr fails: rc = %d\n", rc);
1502 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1503 sbi->ll_md_exp, &md);
1505 ptlrpc_req_finished(request);
1509 ia_valid = op_data->op_attr.ia_valid;
1510 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1511 * cache is not cleared yet. */
1512 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1513 rc = simple_setattr(dentry, &op_data->op_attr);
1514 op_data->op_attr.ia_valid = ia_valid;
1516 rc = ll_update_inode(inode, &md);
1517 ptlrpc_req_finished(request);
1522 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1523 * object(s) determine the file size and mtime. Otherwise, the MDS will
1524 * keep these values until such a time that objects are allocated for it.
1525 * We do the MDS operations first, as it is checking permissions for us.
1526 * We don't to the MDS RPC if there is nothing that we want to store there,
1527 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1528 * going to do an RPC anyways.
1530 * If we are doing a truncate, we will send the mtime and ctime updates
1531 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1532 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1535 * In case of HSMimport, we only set attr on MDS.
1537 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1539 struct inode *inode = dentry->d_inode;
1540 struct ll_inode_info *lli = ll_i2info(inode);
1541 struct md_op_data *op_data = NULL;
1542 bool file_is_released = false;
1546 CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1547 "valid %x, hsm_import %d\n",
1548 ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1549 inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1552 if (attr->ia_valid & ATTR_SIZE) {
1553 /* Check new size against VFS/VM file size limit and rlimit */
1554 rc = inode_newsize_ok(inode, attr->ia_size);
1558 /* The maximum Lustre file size is variable, based on the
1559 * OST maximum object size and number of stripes. This
1560 * needs another check in addition to the VFS check above. */
1561 if (attr->ia_size > ll_file_maxbytes(inode)) {
1562 CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1563 PFID(&lli->lli_fid), attr->ia_size,
1564 ll_file_maxbytes(inode));
1568 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1571 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1572 if (attr->ia_valid & TIMES_SET_FLAGS) {
1573 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1574 !cfs_capable(CFS_CAP_FOWNER))
1578 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1579 if (attr->ia_valid & ATTR_CTIME) {
1580 attr->ia_ctime = CFS_CURRENT_TIME;
1581 attr->ia_valid |= ATTR_CTIME_SET;
1583 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1584 (attr->ia_valid & ATTR_ATIME)) {
1585 attr->ia_atime = CFS_CURRENT_TIME;
1586 attr->ia_valid |= ATTR_ATIME_SET;
1588 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1589 (attr->ia_valid & ATTR_MTIME)) {
1590 attr->ia_mtime = CFS_CURRENT_TIME;
1591 attr->ia_valid |= ATTR_MTIME_SET;
1594 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1595 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1596 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1597 cfs_time_current_sec());
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)
1606 if (!S_ISDIR(inode->i_mode)) {
1607 if (attr->ia_valid & ATTR_SIZE)
1608 inode_dio_write_done(inode);
1609 mutex_unlock(&inode->i_mutex);
1612 /* truncate on a released file must failed with -ENODATA,
1613 * so size must not be set on MDS for released file
1614 * but other attributes must be set
1616 if (S_ISREG(inode->i_mode)) {
1617 struct cl_layout cl = {
1618 .cl_is_released = false,
1624 rc = ll_layout_refresh(inode, &gen);
1628 /* XXX: the only place we need to know the layout type,
1629 * this will be removed by a later patch. -Jinshan */
1630 env = cl_env_get(&refcheck);
1632 GOTO(out, rc = PTR_ERR(env));
1634 rc = cl_object_layout_get(env, lli->lli_clob, &cl);
1635 cl_env_put(env, &refcheck);
1639 file_is_released = cl.cl_is_released;
1641 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1642 if (file_is_released) {
1643 rc = ll_layout_restore(inode, 0, attr->ia_size);
1647 file_is_released = false;
1648 ll_layout_refresh(inode, &gen);
1651 /* If we are changing file size, file content is
1652 * modified, flag it. */
1653 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1654 spin_lock(&lli->lli_lock);
1655 lli->lli_flags |= LLIF_DATA_MODIFIED;
1656 spin_unlock(&lli->lli_lock);
1657 op_data->op_bias |= MDS_DATA_MODIFIED;
1661 memcpy(&op_data->op_attr, attr, sizeof(*attr));
1663 rc = ll_md_setattr(dentry, op_data);
1667 /* RPC to MDT is sent, cancel data modification flag */
1668 if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) {
1669 spin_lock(&lli->lli_lock);
1670 lli->lli_flags &= ~LLIF_DATA_MODIFIED;
1671 spin_unlock(&lli->lli_lock);
1674 if (!S_ISREG(inode->i_mode) || file_is_released)
1677 if (attr->ia_valid & (ATTR_SIZE |
1678 ATTR_ATIME | ATTR_ATIME_SET |
1679 ATTR_MTIME | ATTR_MTIME_SET)) {
1680 /* For truncate and utimes sending attributes to OSTs, setting
1681 * mtime/atime to the past will be performed under PW [0:EOF]
1682 * extent lock (new_size:EOF for truncate). It may seem
1683 * excessive to send mtime/atime updates to OSTs when not
1684 * setting times to past, but it is necessary due to possible
1685 * time de-synchronization between MDT inode and OST objects */
1686 rc = cl_setattr_ost(lli->lli_clob, attr, 0);
1690 if (op_data != NULL)
1691 ll_finish_md_op_data(op_data);
1693 if (!S_ISDIR(inode->i_mode)) {
1694 mutex_lock(&inode->i_mutex);
1695 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1696 inode_dio_wait(inode);
1699 ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1700 LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1705 int ll_setattr(struct dentry *de, struct iattr *attr)
1707 int mode = de->d_inode->i_mode;
1709 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1710 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1711 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1713 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1714 (ATTR_SIZE|ATTR_MODE)) &&
1715 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1716 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1717 !(attr->ia_mode & S_ISGID))))
1718 attr->ia_valid |= ATTR_FORCE;
1720 if ((attr->ia_valid & ATTR_MODE) &&
1722 !(attr->ia_mode & S_ISUID) &&
1723 !(attr->ia_valid & ATTR_KILL_SUID))
1724 attr->ia_valid |= ATTR_KILL_SUID;
1726 if ((attr->ia_valid & ATTR_MODE) &&
1727 ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1728 !(attr->ia_mode & S_ISGID) &&
1729 !(attr->ia_valid & ATTR_KILL_SGID))
1730 attr->ia_valid |= ATTR_KILL_SGID;
1732 return ll_setattr_raw(de, attr, false);
1735 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1736 __u64 max_age, __u32 flags)
1738 struct ll_sb_info *sbi = ll_s2sbi(sb);
1739 struct obd_statfs obd_osfs;
1743 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1745 CERROR("md_statfs fails: rc = %d\n", rc);
1749 osfs->os_type = sb->s_magic;
1751 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1752 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1754 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1755 flags |= OBD_STATFS_NODELAY;
1757 rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1759 CERROR("obd_statfs fails: rc = %d\n", rc);
1763 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1764 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1767 osfs->os_bsize = obd_osfs.os_bsize;
1768 osfs->os_blocks = obd_osfs.os_blocks;
1769 osfs->os_bfree = obd_osfs.os_bfree;
1770 osfs->os_bavail = obd_osfs.os_bavail;
1772 /* If we don't have as many objects free on the OST as inodes
1773 * on the MDS, we reduce the total number of inodes to
1774 * compensate, so that the "inodes in use" number is correct.
1776 if (obd_osfs.os_ffree < osfs->os_ffree) {
1777 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1779 osfs->os_ffree = obd_osfs.os_ffree;
1784 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1786 struct super_block *sb = de->d_sb;
1787 struct obd_statfs osfs;
1788 __u64 fsid = huge_encode_dev(sb->s_dev);
1791 CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1792 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1794 /* Some amount of caching on the client is allowed */
1795 rc = ll_statfs_internal(sb, &osfs,
1796 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1801 statfs_unpack(sfs, &osfs);
1803 /* We need to downshift for all 32-bit kernels, because we can't
1804 * tell if the kernel is being called via sys_statfs64() or not.
1805 * Stop before overflowing f_bsize - in which case it is better
1806 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1807 if (sizeof(long) < 8) {
1808 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1811 osfs.os_blocks >>= 1;
1812 osfs.os_bfree >>= 1;
1813 osfs.os_bavail >>= 1;
1817 sfs->f_blocks = osfs.os_blocks;
1818 sfs->f_bfree = osfs.os_bfree;
1819 sfs->f_bavail = osfs.os_bavail;
1820 sfs->f_fsid.val[0] = (__u32)fsid;
1821 sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1825 void ll_inode_size_lock(struct inode *inode)
1827 struct ll_inode_info *lli;
1829 LASSERT(!S_ISDIR(inode->i_mode));
1831 lli = ll_i2info(inode);
1832 mutex_lock(&lli->lli_size_mutex);
1835 void ll_inode_size_unlock(struct inode *inode)
1837 struct ll_inode_info *lli;
1839 lli = ll_i2info(inode);
1840 mutex_unlock(&lli->lli_size_mutex);
1843 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1845 struct ll_inode_info *lli = ll_i2info(inode);
1846 struct mdt_body *body = md->body;
1847 struct lov_stripe_md *lsm = md->lsm;
1848 struct ll_sb_info *sbi = ll_i2sbi(inode);
1850 LASSERT((lsm != NULL) == ((body->mbo_valid & OBD_MD_FLEASIZE) != 0));
1852 cl_file_inode_init(inode, md);
1854 if (S_ISDIR(inode->i_mode)) {
1857 rc = ll_update_lsm_md(inode, md);
1862 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1863 if (body->mbo_valid & OBD_MD_FLRMTPERM)
1864 ll_update_remote_perm(inode, md->remote_perm);
1866 #ifdef CONFIG_FS_POSIX_ACL
1867 else if (body->mbo_valid & OBD_MD_FLACL) {
1868 spin_lock(&lli->lli_lock);
1869 if (lli->lli_posix_acl)
1870 posix_acl_release(lli->lli_posix_acl);
1871 lli->lli_posix_acl = md->posix_acl;
1872 spin_unlock(&lli->lli_lock);
1875 inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1876 sbi->ll_flags & LL_SBI_32BIT_API);
1877 inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1879 if (body->mbo_valid & OBD_MD_FLATIME) {
1880 if (body->mbo_atime > LTIME_S(inode->i_atime))
1881 LTIME_S(inode->i_atime) = body->mbo_atime;
1882 lli->lli_atime = body->mbo_atime;
1885 if (body->mbo_valid & OBD_MD_FLMTIME) {
1886 if (body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1887 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1888 "to "LPU64"\n", inode->i_ino,
1889 LTIME_S(inode->i_mtime), body->mbo_mtime);
1890 LTIME_S(inode->i_mtime) = body->mbo_mtime;
1892 lli->lli_mtime = body->mbo_mtime;
1895 if (body->mbo_valid & OBD_MD_FLCTIME) {
1896 if (body->mbo_ctime > LTIME_S(inode->i_ctime))
1897 LTIME_S(inode->i_ctime) = body->mbo_ctime;
1898 lli->lli_ctime = body->mbo_ctime;
1901 if (body->mbo_valid & OBD_MD_FLMODE)
1902 inode->i_mode = (inode->i_mode & S_IFMT) |
1903 (body->mbo_mode & ~S_IFMT);
1905 if (body->mbo_valid & OBD_MD_FLTYPE)
1906 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1907 (body->mbo_mode & S_IFMT);
1909 LASSERT(inode->i_mode != 0);
1910 if (S_ISREG(inode->i_mode))
1911 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1912 LL_MAX_BLKSIZE_BITS);
1914 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1916 if (body->mbo_valid & OBD_MD_FLUID)
1917 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
1918 if (body->mbo_valid & OBD_MD_FLGID)
1919 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
1920 if (body->mbo_valid & OBD_MD_FLFLAGS)
1921 inode->i_flags = ll_ext_to_inode_flags(body->mbo_flags);
1922 if (body->mbo_valid & OBD_MD_FLNLINK)
1923 set_nlink(inode, body->mbo_nlink);
1924 if (body->mbo_valid & OBD_MD_FLRDEV)
1925 inode->i_rdev = old_decode_dev(body->mbo_rdev);
1927 if (body->mbo_valid & OBD_MD_FLID) {
1928 /* FID shouldn't be changed! */
1929 if (fid_is_sane(&lli->lli_fid)) {
1930 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
1931 "Trying to change FID "DFID
1932 " to the "DFID", inode "DFID"(%p)\n",
1933 PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
1934 PFID(ll_inode2fid(inode)), inode);
1936 lli->lli_fid = body->mbo_fid1;
1940 LASSERT(fid_seq(&lli->lli_fid) != 0);
1942 if (body->mbo_valid & OBD_MD_FLSIZE) {
1943 i_size_write(inode, body->mbo_size);
1945 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
1946 PFID(ll_inode2fid(inode)),
1947 (unsigned long long)body->mbo_size);
1949 if (body->mbo_valid & OBD_MD_FLBLOCKS)
1950 inode->i_blocks = body->mbo_blocks;
1953 if (body->mbo_valid & OBD_MD_FLMDSCAPA) {
1954 LASSERT(md->mds_capa);
1955 ll_add_capa(inode, md->mds_capa);
1958 if (body->mbo_valid & OBD_MD_FLOSSCAPA) {
1959 LASSERT(md->oss_capa);
1960 ll_add_capa(inode, md->oss_capa);
1963 if (body->mbo_valid & OBD_MD_TSTATE) {
1964 if (body->mbo_t_state & MS_RESTORE)
1965 lli->lli_flags |= LLIF_FILE_RESTORING;
1971 int ll_read_inode2(struct inode *inode, void *opaque)
1973 struct lustre_md *md = opaque;
1974 struct ll_inode_info *lli = ll_i2info(inode);
1978 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1979 PFID(&lli->lli_fid), inode);
1981 /* Core attributes from the MDS first. This is a new inode, and
1982 * the VFS doesn't zero times in the core inode so we have to do
1983 * it ourselves. They will be overwritten by either MDS or OST
1984 * attributes - we just need to make sure they aren't newer. */
1985 LTIME_S(inode->i_mtime) = 0;
1986 LTIME_S(inode->i_atime) = 0;
1987 LTIME_S(inode->i_ctime) = 0;
1989 rc = ll_update_inode(inode, md);
1993 /* OIDEBUG(inode); */
1995 /* initializing backing dev info. */
1996 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1999 if (S_ISREG(inode->i_mode)) {
2000 struct ll_sb_info *sbi = ll_i2sbi(inode);
2001 inode->i_op = &ll_file_inode_operations;
2002 inode->i_fop = sbi->ll_fop;
2003 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2005 } else if (S_ISDIR(inode->i_mode)) {
2006 inode->i_op = &ll_dir_inode_operations;
2007 inode->i_fop = &ll_dir_operations;
2009 } else if (S_ISLNK(inode->i_mode)) {
2010 inode->i_op = &ll_fast_symlink_inode_operations;
2013 inode->i_op = &ll_special_inode_operations;
2015 init_special_inode(inode, inode->i_mode,
2024 void ll_delete_inode(struct inode *inode)
2026 struct ll_inode_info *lli = ll_i2info(inode);
2029 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2030 /* It is last chance to write out dirty pages,
2031 * otherwise we may lose data while umount */
2032 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
2034 truncate_inode_pages_final(&inode->i_data);
2036 LASSERTF(inode->i_data.nrpages == 0, "inode="DFID"(%p) nrpages=%lu, "
2037 "see https://jira.hpdd.intel.com/browse/LU-118\n",
2038 PFID(ll_inode2fid(inode)), inode, inode->i_data.nrpages);
2040 #ifdef HAVE_SBOPS_EVICT_INODE
2041 ll_clear_inode(inode);
2048 int ll_iocontrol(struct inode *inode, struct file *file,
2049 unsigned int cmd, unsigned long arg)
2051 struct ll_sb_info *sbi = ll_i2sbi(inode);
2052 struct ptlrpc_request *req = NULL;
2057 case FSFILT_IOC_GETFLAGS: {
2058 struct mdt_body *body;
2059 struct md_op_data *op_data;
2061 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2062 0, 0, LUSTRE_OPC_ANY,
2064 if (IS_ERR(op_data))
2065 RETURN(PTR_ERR(op_data));
2067 op_data->op_valid = OBD_MD_FLFLAGS;
2068 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2069 ll_finish_md_op_data(op_data);
2071 CERROR("%s: failure inode "DFID": rc = %d\n",
2072 sbi->ll_md_exp->exp_obd->obd_name,
2073 PFID(ll_inode2fid(inode)), rc);
2077 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2079 flags = body->mbo_flags;
2081 ptlrpc_req_finished(req);
2083 RETURN(put_user(flags, (int __user *)arg));
2085 case FSFILT_IOC_SETFLAGS: {
2087 struct md_op_data *op_data;
2088 struct cl_object *obj;
2090 if (get_user(flags, (int __user *)arg))
2093 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2094 LUSTRE_OPC_ANY, NULL);
2095 if (IS_ERR(op_data))
2096 RETURN(PTR_ERR(op_data));
2098 op_data->op_attr_flags = flags;
2099 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2100 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2101 ll_finish_md_op_data(op_data);
2102 ptlrpc_req_finished(req);
2106 inode->i_flags = ll_ext_to_inode_flags(flags);
2108 obj = ll_i2info(inode)->lli_clob;
2112 OBD_ALLOC_PTR(attr);
2116 attr->ia_valid = ATTR_ATTR_FLAG;
2117 rc = cl_setattr_ost(obj, attr, flags);
2129 int ll_flush_ctx(struct inode *inode)
2131 struct ll_sb_info *sbi = ll_i2sbi(inode);
2133 CDEBUG(D_SEC, "flush context for user %d\n",
2134 from_kuid(&init_user_ns, current_uid()));
2136 obd_set_info_async(NULL, sbi->ll_md_exp,
2137 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2139 obd_set_info_async(NULL, sbi->ll_dt_exp,
2140 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2145 /* umount -f client means force down, don't save state */
2146 void ll_umount_begin(struct super_block *sb)
2148 struct ll_sb_info *sbi = ll_s2sbi(sb);
2149 struct obd_device *obd;
2150 struct obd_ioctl_data *ioc_data;
2153 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2154 sb->s_count, atomic_read(&sb->s_active));
2156 obd = class_exp2obd(sbi->ll_md_exp);
2158 CERROR("Invalid MDC connection handle "LPX64"\n",
2159 sbi->ll_md_exp->exp_handle.h_cookie);
2165 obd = class_exp2obd(sbi->ll_dt_exp);
2167 CERROR("Invalid LOV connection handle "LPX64"\n",
2168 sbi->ll_dt_exp->exp_handle.h_cookie);
2174 OBD_ALLOC_PTR(ioc_data);
2176 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2177 sizeof *ioc_data, ioc_data, NULL);
2179 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2180 sizeof *ioc_data, ioc_data, NULL);
2182 OBD_FREE_PTR(ioc_data);
2185 /* Really, we'd like to wait until there are no requests outstanding,
2186 * and then continue. For now, we just invalidate the requests,
2187 * schedule() and sleep one second if needed, and hope.
2193 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2195 struct ll_sb_info *sbi = ll_s2sbi(sb);
2196 char *profilenm = get_profile_name(sb);
2200 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2201 read_only = *flags & MS_RDONLY;
2202 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2203 sizeof(KEY_READ_ONLY),
2204 KEY_READ_ONLY, sizeof(read_only),
2207 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2208 profilenm, read_only ?
2209 "read-only" : "read-write", err);
2214 sb->s_flags |= MS_RDONLY;
2216 sb->s_flags &= ~MS_RDONLY;
2218 if (sbi->ll_flags & LL_SBI_VERBOSE)
2219 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2220 read_only ? "read-only" : "read-write");
2226 * Cleanup the open handle that is cached on MDT-side.
2228 * For open case, the client side open handling thread may hit error
2229 * after the MDT grant the open. Under such case, the client should
2230 * send close RPC to the MDT as cleanup; otherwise, the open handle
2231 * on the MDT will be leaked there until the client umount or evicted.
2233 * In further, if someone unlinked the file, because the open handle
2234 * holds the reference on such file/object, then it will block the
2235 * subsequent threads that want to locate such object via FID.
2237 * \param[in] sb super block for this file-system
2238 * \param[in] open_req pointer to the original open request
2240 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2242 struct mdt_body *body;
2243 struct md_op_data *op_data;
2244 struct ptlrpc_request *close_req = NULL;
2245 struct obd_export *exp = ll_s2sbi(sb)->ll_md_exp;
2248 body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2249 OBD_ALLOC_PTR(op_data);
2250 if (op_data == NULL) {
2251 CWARN("%s: cannot allocate op_data to release open handle for "
2253 ll_get_fsname(sb, NULL, 0), PFID(&body->mbo_fid1));
2258 op_data->op_fid1 = body->mbo_fid1;
2259 op_data->op_handle = body->mbo_handle;
2260 op_data->op_mod_time = cfs_time_current_sec();
2261 md_close(exp, op_data, NULL, &close_req);
2262 ptlrpc_req_finished(close_req);
2263 ll_finish_md_op_data(op_data);
2268 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2269 struct super_block *sb, struct lookup_intent *it)
2271 struct ll_sb_info *sbi = NULL;
2272 struct lustre_md md = { NULL };
2276 LASSERT(*inode || sb);
2277 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2278 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2279 sbi->ll_md_exp, &md);
2284 rc = ll_update_inode(*inode, &md);
2288 LASSERT(sb != NULL);
2291 * At this point server returns to client's same fid as client
2292 * generated for creating. So using ->fid1 is okay here.
2294 LASSERT(fid_is_sane(&md.body->mbo_fid1));
2296 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2297 sbi->ll_flags & LL_SBI_32BIT_API),
2299 if (IS_ERR(*inode)) {
2300 #ifdef CONFIG_FS_POSIX_ACL
2302 posix_acl_release(md.posix_acl);
2303 md.posix_acl = NULL;
2306 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2308 CERROR("new_inode -fatal: rc %d\n", rc);
2313 /* Handling piggyback layout lock.
2314 * Layout lock can be piggybacked by getattr and open request.
2315 * The lsm can be applied to inode only if it comes with a layout lock
2316 * otherwise correct layout may be overwritten, for example:
2317 * 1. proc1: mdt returns a lsm but not granting layout
2318 * 2. layout was changed by another client
2319 * 3. proc2: refresh layout and layout lock granted
2320 * 4. proc1: to apply a stale layout */
2321 if (it != NULL && it->d.lustre.it_lock_mode != 0) {
2322 struct lustre_handle lockh;
2323 struct ldlm_lock *lock;
2325 lockh.cookie = it->d.lustre.it_lock_handle;
2326 lock = ldlm_handle2lock(&lockh);
2327 LASSERT(lock != NULL);
2328 if (ldlm_has_layout(lock)) {
2329 struct cl_object_conf conf;
2331 memset(&conf, 0, sizeof(conf));
2332 conf.coc_opc = OBJECT_CONF_SET;
2333 conf.coc_inode = *inode;
2334 conf.coc_lock = lock;
2335 conf.u.coc_md = &md;
2336 (void)ll_layout_conf(*inode, &conf);
2338 LDLM_LOCK_PUT(lock);
2345 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2346 md_free_lustre_md(sbi->ll_md_exp, &md);
2349 if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2350 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2355 int ll_obd_statfs(struct inode *inode, void __user *arg)
2357 struct ll_sb_info *sbi = NULL;
2358 struct obd_export *exp;
2360 struct obd_ioctl_data *data = NULL;
2362 __u32 __user flags; /* not user, but obd_iocontrol is abused */
2365 if (!inode || !(sbi = ll_i2sbi(inode)))
2366 GOTO(out_statfs, rc = -EINVAL);
2368 rc = obd_ioctl_getdata(&buf, &len, arg);
2370 GOTO(out_statfs, rc);
2373 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2374 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2375 GOTO(out_statfs, rc = -EINVAL);
2377 if (data->ioc_inllen1 != sizeof(__u32) ||
2378 data->ioc_inllen2 != sizeof(__u32) ||
2379 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2380 data->ioc_plen2 != sizeof(struct obd_uuid))
2381 GOTO(out_statfs, rc = -EINVAL);
2383 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2384 if (type & LL_STATFS_LMV)
2385 exp = sbi->ll_md_exp;
2386 else if (type & LL_STATFS_LOV)
2387 exp = sbi->ll_dt_exp;
2389 GOTO(out_statfs, rc = -ENODEV);
2391 flags = (type & LL_STATFS_NODELAY) ? OBD_STATFS_NODELAY : 0;
2392 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, &flags);
2394 GOTO(out_statfs, rc);
2397 obd_ioctl_freedata(buf, len);
2401 int ll_process_config(struct lustre_cfg *lcfg)
2403 struct super_block *sb;
2408 /* The instance name contains the sb: lustre-client-aacfe000 */
2409 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2410 if (!ptr || !*(++ptr))
2412 if (sscanf(ptr, "%lx", &x) != 1)
2414 sb = (struct super_block *)x;
2415 /* This better be a real Lustre superblock! */
2416 LASSERT(s2lsi(sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2418 /* Note we have not called client_common_fill_super yet, so
2419 proc fns must be able to handle that! */
2420 rc = class_process_proc_param(PARAM_LLITE, lprocfs_llite_obd_vars,
2427 /* this function prepares md_op_data hint for passing it down to MD stack. */
2428 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2429 struct inode *i1, struct inode *i2,
2430 const char *name, size_t namelen,
2431 __u32 mode, __u32 opc, void *data)
2433 LASSERT(i1 != NULL);
2436 /* Do not reuse namelen for something else. */
2438 return ERR_PTR(-EINVAL);
2440 if (namelen > ll_i2sbi(i1)->ll_namelen)
2441 return ERR_PTR(-ENAMETOOLONG);
2443 if (!lu_name_is_valid_2(name, namelen))
2444 return ERR_PTR(-EINVAL);
2447 if (op_data == NULL)
2448 OBD_ALLOC_PTR(op_data);
2450 if (op_data == NULL)
2451 return ERR_PTR(-ENOMEM);
2453 ll_i2gids(op_data->op_suppgids, i1, i2);
2454 op_data->op_fid1 = *ll_inode2fid(i1);
2455 op_data->op_capa1 = ll_mdscapa_get(i1);
2456 op_data->op_default_stripe_offset = -1;
2457 if (S_ISDIR(i1->i_mode)) {
2458 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2459 if (opc == LUSTRE_OPC_MKDIR)
2460 op_data->op_default_stripe_offset =
2461 ll_i2info(i1)->lli_def_stripe_offset;
2465 op_data->op_fid2 = *ll_inode2fid(i2);
2466 op_data->op_capa2 = ll_mdscapa_get(i2);
2467 if (S_ISDIR(i2->i_mode))
2468 op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2470 fid_zero(&op_data->op_fid2);
2471 op_data->op_capa2 = NULL;
2474 if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2475 op_data->op_cli_flags |= CLI_HASH64;
2477 if (ll_need_32bit_api(ll_i2sbi(i1)))
2478 op_data->op_cli_flags |= CLI_API32;
2480 op_data->op_name = name;
2481 op_data->op_namelen = namelen;
2482 op_data->op_mode = mode;
2483 op_data->op_mod_time = cfs_time_current_sec();
2484 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2485 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2486 op_data->op_cap = cfs_curproc_cap_pack();
2487 op_data->op_bias = 0;
2488 op_data->op_cli_flags = 0;
2489 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2490 filename_is_volatile(name, namelen, &op_data->op_mds)) {
2491 op_data->op_bias |= MDS_CREATE_VOLATILE;
2493 op_data->op_mds = 0;
2495 op_data->op_data = data;
2497 /* When called by ll_setattr_raw, file is i1. */
2498 if (LLIF_DATA_MODIFIED & ll_i2info(i1)->lli_flags)
2499 op_data->op_bias |= MDS_DATA_MODIFIED;
2504 void ll_finish_md_op_data(struct md_op_data *op_data)
2506 capa_put(op_data->op_capa1);
2507 capa_put(op_data->op_capa2);
2508 OBD_FREE_PTR(op_data);
2511 #ifdef HAVE_SUPEROPS_USE_DENTRY
2512 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2514 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2517 struct ll_sb_info *sbi;
2519 #ifdef HAVE_SUPEROPS_USE_DENTRY
2520 LASSERT((seq != NULL) && (dentry != NULL));
2521 sbi = ll_s2sbi(dentry->d_sb);
2523 LASSERT((seq != NULL) && (vfs != NULL));
2524 sbi = ll_s2sbi(vfs->mnt_sb);
2527 if (sbi->ll_flags & LL_SBI_NOLCK)
2528 seq_puts(seq, ",nolock");
2530 if (sbi->ll_flags & LL_SBI_FLOCK)
2531 seq_puts(seq, ",flock");
2533 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2534 seq_puts(seq, ",localflock");
2536 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2537 seq_puts(seq, ",user_xattr");
2539 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2540 seq_puts(seq, ",lazystatfs");
2542 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2543 seq_puts(seq, ",user_fid2path");
2549 * Get obd name by cmd, and copy out to user space
2551 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2553 struct ll_sb_info *sbi = ll_i2sbi(inode);
2554 struct obd_device *obd;
2557 if (cmd == OBD_IOC_GETDTNAME)
2558 obd = class_exp2obd(sbi->ll_dt_exp);
2559 else if (cmd == OBD_IOC_GETMDNAME)
2560 obd = class_exp2obd(sbi->ll_md_exp);
2567 if (copy_to_user((void __user *)arg, obd->obd_name,
2568 strlen(obd->obd_name) + 1))
2575 * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2576 * fsname will be returned in this buffer; otherwise, a static buffer will be
2577 * used to store the fsname and returned to caller.
2579 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2581 static char fsname_static[MTI_NAME_MAXLEN];
2582 struct lustre_sb_info *lsi = s2lsi(sb);
2587 /* this means the caller wants to use static buffer
2588 * and it doesn't care about race. Usually this is
2589 * in error reporting path */
2590 buf = fsname_static;
2591 buflen = sizeof(fsname_static);
2594 len = strlen(lsi->lsi_lmd->lmd_profile);
2595 ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2596 if (ptr && (strcmp(ptr, "-client") == 0))
2599 if (unlikely(len >= buflen))
2601 strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2607 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2614 p.mnt = current->fs->root.mnt;
2616 path = d_path(&p, buf, bufsize);
2621 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2623 char *buf, *path = NULL;
2624 struct dentry *dentry = NULL;
2625 struct inode *inode = page->mapping->host;
2627 /* this can be called inside spin lock so use GFP_ATOMIC. */
2628 buf = (char *)__get_free_page(GFP_ATOMIC);
2630 dentry = d_find_alias(page->mapping->host);
2632 path = ll_d_path(dentry, buf, PAGE_SIZE);
2636 "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2637 "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2638 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2639 PFID(ll_inode2fid(inode)),
2640 (path && !IS_ERR(path)) ? path : "", ioret);
2646 free_page((unsigned long)buf);
2649 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2650 struct lov_user_md **kbuf)
2652 struct lov_user_md lum;
2656 if (copy_from_user(&lum, md, sizeof(lum)))
2659 lum_size = ll_lov_user_md_size(&lum);
2663 OBD_ALLOC(*kbuf, lum_size);
2667 if (copy_from_user(*kbuf, md, lum_size) != 0) {
2668 OBD_FREE(*kbuf, lum_size);
2676 * Compute llite root squash state after a change of root squash
2677 * configuration setting or add/remove of a lnet nid
2679 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2681 struct root_squash_info *squash = &sbi->ll_squash;
2684 lnet_process_id_t id;
2686 /* Update norootsquash flag */
2687 down_write(&squash->rsi_sem);
2688 if (list_empty(&squash->rsi_nosquash_nids))
2689 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2691 /* Do not apply root squash as soon as one of our NIDs is
2692 * in the nosquash_nids list */
2695 while (LNetGetId(i++, &id) != -ENOENT) {
2696 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2698 if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2704 sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2706 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2708 up_write(&squash->rsi_sem);
2712 * Parse linkea content to extract information about a given hardlink
2714 * \param[in] ldata - Initialized linkea data
2715 * \param[in] linkno - Link identifier
2716 * \param[out] parent_fid - The entry's parent FID
2717 * \param[out] ln - Entry name destination buffer
2719 * \retval 0 on success
2720 * \retval Appropriate negative error code on failure
2722 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2723 struct lu_fid *parent_fid, struct lu_name *ln)
2729 rc = linkea_init(ldata);
2733 if (linkno >= ldata->ld_leh->leh_reccount)
2734 /* beyond last link */
2737 linkea_first_entry(ldata);
2738 for (idx = 0; ldata->ld_lee != NULL; idx++) {
2739 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2744 linkea_next_entry(ldata);
2754 * Get parent FID and name of an identified link. Operation is performed for
2755 * a given link number, letting the caller iterate over linkno to list one or
2756 * all links of an entry.
2758 * \param[in] file - File descriptor against which to perform the operation
2759 * \param[in,out] arg - User-filled structure containing the linkno to operate
2760 * on and the available size. It is eventually filled with
2761 * the requested information or left untouched on error
2763 * \retval - 0 on success
2764 * \retval - Appropriate negative error code on failure
2766 int ll_getparent(struct file *file, struct getparent __user *arg)
2768 struct dentry *dentry = file->f_path.dentry;
2769 struct inode *inode = dentry->d_inode;
2770 struct linkea_data *ldata;
2771 struct lu_buf buf = LU_BUF_NULL;
2773 struct lu_fid parent_fid;
2780 if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2781 !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2784 if (get_user(name_size, &arg->gp_name_size))
2787 if (get_user(linkno, &arg->gp_linkno))
2790 if (name_size > PATH_MAX)
2793 OBD_ALLOC(ldata, sizeof(*ldata));
2797 rc = linkea_data_new(ldata, &buf);
2799 GOTO(ldata_free, rc);
2801 rc = ll_getxattr(dentry, XATTR_NAME_LINK, buf.lb_buf, buf.lb_len);
2805 rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2809 if (ln.ln_namelen >= name_size)
2810 GOTO(lb_free, rc = -EOVERFLOW);
2812 if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2813 GOTO(lb_free, rc = -EFAULT);
2815 if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2816 GOTO(lb_free, rc = -EFAULT);
2818 if (put_user('\0', arg->gp_name + ln.ln_namelen))
2819 GOTO(lb_free, rc = -EFAULT);
2824 OBD_FREE(ldata, sizeof(*ldata));