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, 2015, 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/time.h>
46 #include <linux/types.h>
47 #include <linux/version.h>
49 #include <linux/user_namespace.h>
50 #ifdef HAVE_UIDGID_HEADER
51 # include <linux/uidgid.h>
54 #include <lustre_ioctl.h>
55 #include <lustre_ha.h>
56 #include <lustre_dlm.h>
57 #include <lprocfs_status.h>
58 #include <lustre_disk.h>
59 #include <lustre_param.h>
60 #include <lustre_log.h>
61 #include <cl_object.h>
62 #include <obd_cksum.h>
63 #include "llite_internal.h"
65 struct kmem_cache *ll_file_data_slab;
68 #define log2(n) ffz(~(n))
71 static struct ll_sb_info *ll_init_sbi(void)
73 struct ll_sb_info *sbi = NULL;
75 unsigned long lru_page_max;
85 spin_lock_init(&sbi->ll_lock);
86 mutex_init(&sbi->ll_lco.lco_lock);
87 spin_lock_init(&sbi->ll_pp_extent_lock);
88 spin_lock_init(&sbi->ll_process_lock);
89 sbi->ll_rw_stats_on = 0;
92 pages = si.totalram - si.totalhigh;
93 lru_page_max = pages / 2;
95 /* initialize ll_cache data */
96 sbi->ll_cache = cl_cache_init(lru_page_max);
97 if (sbi->ll_cache == NULL) {
98 OBD_FREE(sbi, sizeof(*sbi));
102 sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
103 SBI_DEFAULT_READAHEAD_MAX);
104 sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
105 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
106 SBI_DEFAULT_READAHEAD_WHOLE_MAX;
107 INIT_LIST_HEAD(&sbi->ll_conn_chain);
108 INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
110 ll_generate_random_uuid(uuid);
111 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
112 CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
114 sbi->ll_flags |= LL_SBI_VERBOSE;
115 #ifdef ENABLE_CHECKSUM
116 sbi->ll_flags |= LL_SBI_CHECKSUM;
119 #ifdef HAVE_LRU_RESIZE_SUPPORT
120 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
123 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
124 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
126 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
130 /* metadata statahead is enabled by default */
131 sbi->ll_sa_max = LL_SA_RPC_DEF;
132 atomic_set(&sbi->ll_sa_total, 0);
133 atomic_set(&sbi->ll_sa_wrong, 0);
134 atomic_set(&sbi->ll_sa_running, 0);
135 atomic_set(&sbi->ll_agl_total, 0);
136 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
137 sbi->ll_flags |= LL_SBI_FAST_READ;
140 sbi->ll_squash.rsi_uid = 0;
141 sbi->ll_squash.rsi_gid = 0;
142 INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
143 init_rwsem(&sbi->ll_squash.rsi_sem);
148 static void ll_free_sbi(struct super_block *sb)
150 struct ll_sb_info *sbi = ll_s2sbi(sb);
154 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
155 cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
156 if (sbi->ll_cache != NULL) {
157 cl_cache_decref(sbi->ll_cache);
158 sbi->ll_cache = NULL;
160 OBD_FREE(sbi, sizeof(*sbi));
165 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
166 struct vfsmount *mnt)
168 struct inode *root = NULL;
169 struct ll_sb_info *sbi = ll_s2sbi(sb);
170 struct obd_device *obd;
171 struct obd_statfs *osfs = NULL;
172 struct ptlrpc_request *request = NULL;
173 struct obd_connect_data *data = NULL;
174 struct obd_uuid *uuid;
175 struct md_op_data *op_data;
176 struct lustre_md lmd;
178 int size, err, checksum;
181 obd = class_name2obd(md);
183 CERROR("MD %s: not setup or attached\n", md);
197 /* indicate the features supported by this client */
198 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
199 OBD_CONNECT_ATTRFID |
200 OBD_CONNECT_VERSION | OBD_CONNECT_BRW_SIZE |
201 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
202 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
203 OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 |
204 OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR |
205 OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH|
206 OBD_CONNECT_EINPROGRESS |
207 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
208 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS |
209 OBD_CONNECT_MAX_EASIZE |
210 OBD_CONNECT_FLOCK_DEAD |
211 OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
212 OBD_CONNECT_OPEN_BY_FID |
213 OBD_CONNECT_DIR_STRIPE |
214 OBD_CONNECT_BULK_MBITS |
217 #ifdef HAVE_LRU_RESIZE_SUPPORT
218 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
219 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
221 #ifdef CONFIG_FS_POSIX_ACL
222 data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK;
225 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
226 /* flag mdc connection as lightweight, only used for test
227 * purpose, use with care */
228 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
230 data->ocd_ibits_known = MDS_INODELOCK_FULL;
231 data->ocd_version = LUSTRE_VERSION_CODE;
233 if (sb->s_flags & MS_RDONLY)
234 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
235 if (sbi->ll_flags & LL_SBI_USER_XATTR)
236 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
238 if (sbi->ll_flags & LL_SBI_FLOCK)
239 sbi->ll_fop = &ll_file_operations_flock;
240 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
241 sbi->ll_fop = &ll_file_operations;
243 sbi->ll_fop = &ll_file_operations_noflock;
246 data->ocd_connect_flags |= OBD_CONNECT_REAL;
247 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
248 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
250 /* always ping even if server suppress_pings */
251 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
252 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
254 data->ocd_brw_size = MD_MAX_BRW_SIZE;
256 err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
258 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
259 "recovery, of which this client is not a "
260 "part. Please wait for recovery to complete,"
261 " abort, or time out.\n", md);
264 CERROR("cannot connect to %s: rc = %d\n", md, err);
268 sbi->ll_md_exp->exp_connect_data = *data;
270 err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
271 LUSTRE_SEQ_METADATA);
273 CERROR("%s: Can't init metadata layer FID infrastructure, "
274 "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
278 /* For mount, we only need fs info from MDT0, and also in DNE, it
279 * can make sure the client can be mounted as long as MDT0 is
281 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
282 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
283 OBD_STATFS_FOR_MDT0);
285 GOTO(out_md_fid, err);
287 /* This needs to be after statfs to ensure connect has finished.
288 * Note that "data" does NOT contain the valid connect reply.
289 * If connecting to a 1.8 server there will be no LMV device, so
290 * we can access the MDC export directly and exp_connect_flags will
291 * be non-zero, but if accessing an upgraded 2.1 server it will
292 * have the correct flags filled in.
293 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
294 valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
295 if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
296 valid != CLIENT_CONNECT_MDT_REQD) {
299 OBD_ALLOC_WAIT(buf, PAGE_CACHE_SIZE);
300 obd_connect_flags2str(buf, PAGE_CACHE_SIZE,
301 valid ^ CLIENT_CONNECT_MDT_REQD, ",");
302 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
303 "feature(s) needed for correct operation "
304 "of this client (%s). Please upgrade "
305 "server or downgrade client.\n",
306 sbi->ll_md_exp->exp_obd->obd_name, buf);
307 OBD_FREE(buf, PAGE_CACHE_SIZE);
308 GOTO(out_md_fid, err = -EPROTO);
311 size = sizeof(*data);
312 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
313 KEY_CONN_DATA, &size, data);
315 CERROR("%s: Get connect data failed: rc = %d\n",
316 sbi->ll_md_exp->exp_obd->obd_name, err);
317 GOTO(out_md_fid, err);
320 LASSERT(osfs->os_bsize);
321 sb->s_blocksize = osfs->os_bsize;
322 sb->s_blocksize_bits = log2(osfs->os_bsize);
323 sb->s_magic = LL_SUPER_MAGIC;
324 sb->s_maxbytes = MAX_LFS_FILESIZE;
325 sbi->ll_namelen = osfs->os_namelen;
327 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
328 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
329 LCONSOLE_INFO("Disabling user_xattr feature because "
330 "it is not supported on the server\n");
331 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
334 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
336 sb->s_flags |= MS_POSIXACL;
338 sbi->ll_flags |= LL_SBI_ACL;
340 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
342 sb->s_flags &= ~MS_POSIXACL;
344 sbi->ll_flags &= ~LL_SBI_ACL;
347 if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
348 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
349 sbi->ll_flags |= LL_SBI_RMT_CLIENT;
350 LCONSOLE_INFO("client is set as remote by default.\n");
353 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
354 sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
355 LCONSOLE_INFO("client claims to be remote, but server "
356 "rejected, forced to be local.\n");
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_BRW_SIZE)
364 sbi->ll_md_brw_pages = data->ocd_brw_size >> PAGE_CACHE_SHIFT;
366 sbi->ll_md_brw_pages = 1;
368 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
369 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
371 if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
372 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
373 LCONSOLE_INFO("%s: disabling xattr cache due to "
374 "unknown maximum xattr size.\n", dt);
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_RMT_CLIENT |
396 OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
397 OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
398 OBD_CONNECT_MAXBYTES |
399 OBD_CONNECT_EINPROGRESS |
400 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
401 OBD_CONNECT_LAYOUTLOCK |
402 OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
403 OBD_CONNECT_BULK_MBITS;
405 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
406 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
408 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
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();
421 #ifdef HAVE_LRU_RESIZE_SUPPORT
422 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
424 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
425 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
427 /* always ping even if server suppress_pings */
428 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
429 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
431 CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
432 "ocd_grant: %d\n", data->ocd_connect_flags,
433 data->ocd_version, data->ocd_grant);
435 obd->obd_upcall.onu_owner = &sbi->ll_lco;
436 obd->obd_upcall.onu_upcall = cl_ocd_update;
438 data->ocd_brw_size = DT_MAX_BRW_SIZE;
440 err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
443 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
444 "recovery, of which this client is not a "
445 "part. Please wait for recovery to "
446 "complete, abort, or time out.\n", dt);
449 CERROR("%s: Cannot connect to %s: rc = %d\n",
450 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
454 sbi->ll_dt_exp->exp_connect_data = *data;
456 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
457 LUSTRE_SEQ_METADATA);
459 CERROR("%s: Can't init data layer FID infrastructure, "
460 "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
464 mutex_lock(&sbi->ll_lco.lco_lock);
465 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
466 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
467 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
468 mutex_unlock(&sbi->ll_lco.lco_lock);
470 fid_zero(&sbi->ll_root_fid);
471 err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
474 CERROR("cannot mds_connect: rc = %d\n", err);
475 GOTO(out_lock_cn_cb, err);
477 if (!fid_is_sane(&sbi->ll_root_fid)) {
478 CERROR("%s: Invalid root fid "DFID" during mount\n",
479 sbi->ll_md_exp->exp_obd->obd_name,
480 PFID(&sbi->ll_root_fid));
481 GOTO(out_lock_cn_cb, err = -EINVAL);
483 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
485 sb->s_op = &lustre_super_operations;
486 #if THREAD_SIZE >= 8192 /*b=17630*/
487 sb->s_export_op = &lustre_export_operations;
491 * XXX: move this to after cbd setup? */
492 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
493 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
494 valid |= OBD_MD_FLRMTPERM;
495 else if (sbi->ll_flags & LL_SBI_ACL)
496 valid |= OBD_MD_FLACL;
498 OBD_ALLOC_PTR(op_data);
500 GOTO(out_lock_cn_cb, err = -ENOMEM);
502 op_data->op_fid1 = sbi->ll_root_fid;
503 op_data->op_mode = 0;
504 op_data->op_valid = valid;
506 err = md_getattr(sbi->ll_md_exp, op_data, &request);
508 OBD_FREE_PTR(op_data);
510 CERROR("%s: md_getattr failed for root: rc = %d\n",
511 sbi->ll_md_exp->exp_obd->obd_name, err);
512 GOTO(out_lock_cn_cb, err);
515 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
516 sbi->ll_md_exp, &lmd);
518 CERROR("failed to understand root inode md: rc = %d\n", err);
519 ptlrpc_req_finished(request);
520 GOTO(out_lock_cn_cb, err);
523 LASSERT(fid_is_sane(&sbi->ll_root_fid));
524 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
525 sbi->ll_flags & LL_SBI_32BIT_API),
527 md_free_lustre_md(sbi->ll_md_exp, &lmd);
528 ptlrpc_req_finished(request);
531 #ifdef CONFIG_FS_POSIX_ACL
533 posix_acl_release(lmd.posix_acl);
534 lmd.posix_acl = NULL;
537 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
539 CERROR("lustre_lite: bad iget4 for root\n");
543 #ifdef CONFIG_FS_POSIX_ACL
544 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
545 rct_init(&sbi->ll_rct);
546 et_init(&sbi->ll_et);
550 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
551 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
552 KEY_CHECKSUM, sizeof(checksum), &checksum,
556 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
557 KEY_CACHE_SET, sizeof(*sbi->ll_cache),
558 sbi->ll_cache, NULL);
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));
585 if (proc_lustre_fs_root != NULL) {
586 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
589 CERROR("%s: could not register mount in lprocfs: "
590 "rc = %d\n", ll_get_fsname(sb, NULL, 0), err);
600 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
602 obd_disconnect(sbi->ll_dt_exp);
603 sbi->ll_dt_exp = NULL;
605 obd_fid_fini(sbi->ll_md_exp->exp_obd);
607 obd_disconnect(sbi->ll_md_exp);
608 sbi->ll_md_exp = NULL;
617 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
621 size = sizeof(*lmmsize);
622 rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
623 KEY_MAX_EASIZE, &size, lmmsize);
625 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
626 sbi->ll_dt_exp->exp_obd->obd_name, rc);
631 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
632 KEY_MAX_EASIZE, &size, lmmsize);
634 CERROR("Get max mdsize error rc %d\n", rc);
640 * Get the value of the default_easize parameter.
642 * \see client_obd::cl_default_mds_easize
644 * \param[in] sbi superblock info for this filesystem
645 * \param[out] lmmsize pointer to storage location for value
647 * \retval 0 on success
648 * \retval negative negated errno on failure
650 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
655 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
656 KEY_DEFAULT_EASIZE, &size, lmmsize);
658 CERROR("Get default mdsize error rc %d\n", rc);
664 * Set the default_easize parameter to the given value.
666 * \see client_obd::cl_default_mds_easize
668 * \param[in] sbi superblock info for this filesystem
669 * \param[in] lmmsize the size to set
671 * \retval 0 on success
672 * \retval negative negated errno on failure
674 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
678 if (lmmsize < sizeof(struct lov_mds_md) ||
679 lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
682 rc = obd_set_info_async(NULL, sbi->ll_md_exp,
683 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
684 sizeof(int), &lmmsize, NULL);
689 static void ll_dump_inode(struct inode *inode)
691 struct ll_d_hlist_node *tmp;
692 int dentry_count = 0;
694 LASSERT(inode != NULL);
696 ll_d_hlist_for_each(tmp, &inode->i_dentry)
699 CERROR("%s: inode %p dump: dev=%s fid="DFID
700 " mode=%o count=%u, %d dentries\n",
701 ll_get_fsname(inode->i_sb, NULL, 0), inode,
702 ll_i2mdexp(inode)->exp_obd->obd_name, PFID(ll_inode2fid(inode)),
703 inode->i_mode, atomic_read(&inode->i_count), dentry_count);
706 void lustre_dump_dentry(struct dentry *dentry, int recur)
708 struct list_head *tmp;
711 LASSERT(dentry != NULL);
713 list_for_each(tmp, &dentry->d_subdirs)
716 CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
717 " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
718 dentry->d_name.len, dentry->d_name.name,
719 dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
720 dentry->d_parent, dentry->d_inode, ll_d_count(dentry),
721 dentry->d_flags, dentry->d_fsdata, subdirs);
722 if (dentry->d_inode != NULL)
723 ll_dump_inode(dentry->d_inode);
728 list_for_each(tmp, &dentry->d_subdirs) {
729 struct dentry *d = list_entry(tmp, struct dentry, d_child);
730 lustre_dump_dentry(d, recur - 1);
734 static void client_common_put_super(struct super_block *sb)
736 struct ll_sb_info *sbi = ll_s2sbi(sb);
739 #ifdef CONFIG_FS_POSIX_ACL
740 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
741 et_fini(&sbi->ll_et);
742 rct_fini(&sbi->ll_rct);
748 list_del(&sbi->ll_conn_chain);
750 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
751 obd_disconnect(sbi->ll_dt_exp);
752 sbi->ll_dt_exp = NULL;
754 lprocfs_unregister_mountpoint(sbi);
756 obd_fid_fini(sbi->ll_md_exp->exp_obd);
757 obd_disconnect(sbi->ll_md_exp);
758 sbi->ll_md_exp = NULL;
763 void ll_kill_super(struct super_block *sb)
765 struct ll_sb_info *sbi;
769 if (!(sb->s_flags & MS_ACTIVE))
773 /* we need restore s_dev from changed for clustred NFS before put_super
774 * because new kernels have cached s_dev and change sb->s_dev in
775 * put_super not affected real removing devices */
777 sb->s_dev = sbi->ll_sdev_orig;
778 sbi->ll_umounting = 1;
780 /* wait running statahead threads to quit */
781 while (atomic_read(&sbi->ll_sa_running) > 0) {
782 set_current_state(TASK_UNINTERRUPTIBLE);
783 schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
790 static inline int ll_set_opt(const char *opt, char *data, int fl)
792 if (strncmp(opt, data, strlen(opt)) != 0)
798 /* non-client-specific mount options are parsed in lmd_parse */
799 static int ll_options(char *options, int *flags)
802 char *s1 = options, *s2;
808 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
811 CDEBUG(D_SUPER, "next opt=%s\n", s1);
812 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
817 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
822 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
827 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
832 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
837 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
842 tmp = ll_set_opt("context", s1, 1);
845 tmp = ll_set_opt("fscontext", s1, 1);
848 tmp = ll_set_opt("defcontext", s1, 1);
851 tmp = ll_set_opt("rootcontext", s1, 1);
854 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
859 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
864 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
870 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
875 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
880 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
885 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
890 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
895 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
900 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
905 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
910 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
915 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
920 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
926 s2 = strchr(s1, ',');
934 void ll_lli_init(struct ll_inode_info *lli)
936 lli->lli_inode_magic = LLI_INODE_MAGIC;
938 spin_lock_init(&lli->lli_lock);
939 lli->lli_posix_acl = NULL;
940 lli->lli_remote_perms = NULL;
941 mutex_init(&lli->lli_rmtperm_mutex);
942 /* Do not set lli_fid, it has been initialized already. */
943 fid_zero(&lli->lli_pfid);
944 atomic_set(&lli->lli_open_count, 0);
945 lli->lli_rmtperm_time = 0;
946 lli->lli_mds_read_och = NULL;
947 lli->lli_mds_write_och = NULL;
948 lli->lli_mds_exec_och = NULL;
949 lli->lli_open_fd_read_count = 0;
950 lli->lli_open_fd_write_count = 0;
951 lli->lli_open_fd_exec_count = 0;
952 mutex_init(&lli->lli_och_mutex);
953 spin_lock_init(&lli->lli_agl_lock);
954 spin_lock_init(&lli->lli_layout_lock);
955 ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
956 lli->lli_clob = NULL;
958 init_rwsem(&lli->lli_xattrs_list_rwsem);
959 mutex_init(&lli->lli_xattrs_enq_lock);
961 LASSERT(lli->lli_vfs_inode.i_mode != 0);
962 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
963 mutex_init(&lli->lli_readdir_mutex);
964 lli->lli_opendir_key = NULL;
966 spin_lock_init(&lli->lli_sa_lock);
967 lli->lli_opendir_pid = 0;
968 lli->lli_sa_enabled = 0;
969 lli->lli_def_stripe_offset = -1;
971 mutex_init(&lli->lli_size_mutex);
972 lli->lli_symlink_name = NULL;
973 init_rwsem(&lli->lli_trunc_sem);
974 range_lock_tree_init(&lli->lli_write_tree);
975 init_rwsem(&lli->lli_glimpse_sem);
976 lli->lli_glimpse_time = 0;
977 INIT_LIST_HEAD(&lli->lli_agl_list);
978 lli->lli_agl_index = 0;
979 lli->lli_async_rc = 0;
981 mutex_init(&lli->lli_layout_mutex);
984 static inline int ll_bdi_register(struct backing_dev_info *bdi)
986 static atomic_t ll_bdi_num = ATOMIC_INIT(0);
988 bdi->name = "lustre";
989 return bdi_register(bdi, NULL, "lustre-%d",
990 atomic_inc_return(&ll_bdi_num));
993 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
995 struct lustre_profile *lprof = NULL;
996 struct lustre_sb_info *lsi = s2lsi(sb);
997 struct ll_sb_info *sbi;
998 char *dt = NULL, *md = NULL;
999 char *profilenm = get_profile_name(sb);
1000 struct config_llog_instance *cfg;
1001 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1002 const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
1006 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1012 try_module_get(THIS_MODULE);
1014 /* client additional sb info */
1015 lsi->lsi_llsbi = sbi = ll_init_sbi();
1017 module_put(THIS_MODULE);
1022 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1024 GOTO(out_free, err);
1026 err = bdi_init(&lsi->lsi_bdi);
1028 GOTO(out_free, err);
1029 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1030 #ifdef HAVE_BDI_CAP_MAP_COPY
1031 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1033 lsi->lsi_bdi.capabilities = 0;
1035 err = ll_bdi_register(&lsi->lsi_bdi);
1037 GOTO(out_free, err);
1039 sb->s_bdi = &lsi->lsi_bdi;
1040 #ifndef HAVE_DCACHE_LOCK
1041 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1042 sb->s_d_op = &ll_d_ops;
1045 /* Generate a string unique to this super, in case some joker tries
1046 to mount the same fs at two mount points.
1047 Use the address of the super itself.*/
1048 cfg->cfg_instance = sb;
1049 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1050 cfg->cfg_callback = class_config_llog_handler;
1051 /* set up client obds */
1052 err = lustre_process_log(sb, profilenm, cfg);
1054 GOTO(out_free, err);
1056 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1057 lprof = class_get_profile(profilenm);
1058 if (lprof == NULL) {
1059 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1060 " read from the MGS. Does that filesystem "
1061 "exist?\n", profilenm);
1062 GOTO(out_free, err = -EINVAL);
1064 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1065 lprof->lp_md, lprof->lp_dt);
1067 OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1069 GOTO(out_free, err = -ENOMEM);
1070 sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1072 OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1074 GOTO(out_free, err = -ENOMEM);
1075 sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1077 /* connections, registrations, sb setup */
1078 err = client_common_fill_super(sb, md, dt, mnt);
1080 GOTO(out_free, err);
1082 sbi->ll_client_common_fill_super_succeeded = 1;
1086 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1088 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1090 class_put_profile(lprof);
1093 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1094 LCONSOLE_WARN("Mounted %s\n", profilenm);
1098 } /* ll_fill_super */
1100 void ll_put_super(struct super_block *sb)
1102 struct config_llog_instance cfg, params_cfg;
1103 struct obd_device *obd;
1104 struct lustre_sb_info *lsi = s2lsi(sb);
1105 struct ll_sb_info *sbi = ll_s2sbi(sb);
1106 char *profilenm = get_profile_name(sb);
1108 int next, force = 1, rc = 0;
1111 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1113 cfg.cfg_instance = sb;
1114 lustre_end_log(sb, profilenm, &cfg);
1116 params_cfg.cfg_instance = sb;
1117 lustre_end_log(sb, PARAMS_FILENAME, ¶ms_cfg);
1119 if (sbi->ll_md_exp) {
1120 obd = class_exp2obd(sbi->ll_md_exp);
1122 force = obd->obd_force;
1125 /* Wait for unstable pages to be committed to stable storage */
1127 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1128 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1129 atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1133 ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1134 if (force == 0 && rc != -EINTR)
1135 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1138 /* We need to set force before the lov_disconnect in
1139 lustre_common_put_super, since l_d cleans up osc's as well. */
1142 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1144 obd->obd_force = force;
1148 if (sbi->ll_client_common_fill_super_succeeded) {
1149 /* Only if client_common_fill_super succeeded */
1150 client_common_put_super(sb);
1154 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1155 class_manual_cleanup(obd);
1158 if (sbi->ll_flags & LL_SBI_VERBOSE)
1159 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1162 class_del_profile(profilenm);
1164 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1165 bdi_destroy(&lsi->lsi_bdi);
1166 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1170 lsi->lsi_llsbi = NULL;
1172 lustre_common_put_super(sb);
1174 cl_env_cache_purge(~0);
1176 module_put(THIS_MODULE);
1179 } /* client_put_super */
1181 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1183 struct inode *inode = NULL;
1185 /* NOTE: we depend on atomic igrab() -bzzz */
1186 lock_res_and_lock(lock);
1187 if (lock->l_resource->lr_lvb_inode) {
1188 struct ll_inode_info * lli;
1189 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1190 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1191 inode = igrab(lock->l_resource->lr_lvb_inode);
1193 inode = lock->l_resource->lr_lvb_inode;
1194 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1195 D_WARNING, lock, "lr_lvb_inode %p is "
1196 "bogus: magic %08x",
1197 lock->l_resource->lr_lvb_inode,
1198 lli->lli_inode_magic);
1202 unlock_res_and_lock(lock);
1206 void ll_dir_clear_lsm_md(struct inode *inode)
1208 struct ll_inode_info *lli = ll_i2info(inode);
1210 LASSERT(S_ISDIR(inode->i_mode));
1212 if (lli->lli_lsm_md != NULL) {
1213 lmv_free_memmd(lli->lli_lsm_md);
1214 lli->lli_lsm_md = NULL;
1218 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1219 const struct lu_fid *fid,
1220 struct lustre_md *md)
1222 struct ll_sb_info *sbi = ll_s2sbi(sb);
1223 struct mdt_body *body = md->body;
1224 struct inode *inode;
1228 ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1229 inode = iget_locked(sb, ino);
1230 if (inode == NULL) {
1231 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1232 ll_get_fsname(sb, NULL, 0), PFID(fid));
1233 RETURN(ERR_PTR(-ENOENT));
1236 if (inode->i_state & I_NEW) {
1237 struct ll_inode_info *lli = ll_i2info(inode);
1238 struct lmv_stripe_md *lsm = md->lmv;
1240 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1241 (body->mbo_mode & S_IFMT);
1242 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1245 LTIME_S(inode->i_mtime) = 0;
1246 LTIME_S(inode->i_atime) = 0;
1247 LTIME_S(inode->i_ctime) = 0;
1250 #ifdef HAVE_BACKING_DEV_INFO
1251 /* initializing backing dev info. */
1252 inode->i_mapping->backing_dev_info =
1253 &s2lsi(inode->i_sb)->lsi_bdi;
1255 inode->i_op = &ll_dir_inode_operations;
1256 inode->i_fop = &ll_dir_operations;
1257 lli->lli_fid = *fid;
1260 LASSERT(lsm != NULL);
1261 /* master object FID */
1262 lli->lli_pfid = body->mbo_fid1;
1263 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1264 lli, PFID(fid), PFID(&lli->lli_pfid));
1265 unlock_new_inode(inode);
1271 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1274 struct lmv_stripe_md *lsm = md->lmv;
1277 LASSERT(lsm != NULL);
1278 /* XXX sigh, this lsm_root initialization should be in
1279 * LMV layer, but it needs ll_iget right now, so we
1280 * put this here right now. */
1281 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1282 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1283 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1284 /* Unfortunately ll_iget will call ll_update_inode,
1285 * where the initialization of slave inode is slightly
1286 * different, so it reset lsm_md to NULL to avoid
1287 * initializing lsm for slave inode. */
1288 /* For migrating inode, master stripe and master object will
1289 * be same, so we only need assign this inode */
1290 if (lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION && i == 0)
1291 lsm->lsm_md_oinfo[i].lmo_root = inode;
1293 lsm->lsm_md_oinfo[i].lmo_root =
1294 ll_iget_anon_dir(inode->i_sb, fid, md);
1296 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1297 int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1299 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1307 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1308 const struct lmv_stripe_md *lsm_md2)
1310 return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1311 lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1312 lsm_md1->lsm_md_master_mdt_index ==
1313 lsm_md2->lsm_md_master_mdt_index &&
1314 lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1315 lsm_md1->lsm_md_layout_version ==
1316 lsm_md2->lsm_md_layout_version &&
1317 strcmp(lsm_md1->lsm_md_pool_name,
1318 lsm_md2->lsm_md_pool_name) == 0;
1321 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1323 struct ll_inode_info *lli = ll_i2info(inode);
1324 struct lmv_stripe_md *lsm = md->lmv;
1328 LASSERT(S_ISDIR(inode->i_mode));
1329 CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1330 PFID(ll_inode2fid(inode)));
1332 /* no striped information from request. */
1334 if (lli->lli_lsm_md == NULL) {
1336 } else if (lli->lli_lsm_md->lsm_md_hash_type &
1337 LMV_HASH_FLAG_MIGRATION) {
1338 /* migration is done, the temporay MIGRATE layout has
1340 CDEBUG(D_INODE, DFID" finish migration.\n",
1341 PFID(ll_inode2fid(inode)));
1342 lmv_free_memmd(lli->lli_lsm_md);
1343 lli->lli_lsm_md = NULL;
1346 /* The lustre_md from req does not include stripeEA,
1347 * see ll_md_setattr */
1352 /* set the directory layout */
1353 if (lli->lli_lsm_md == NULL) {
1354 struct cl_attr *attr;
1356 rc = ll_init_lsm_md(inode, md);
1360 /* set md->lmv to NULL, so the following free lustre_md
1361 * will not free this lsm */
1363 lli->lli_lsm_md = lsm;
1365 OBD_ALLOC_PTR(attr);
1369 /* validate the lsm */
1370 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1371 ll_md_blocking_ast);
1377 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1378 md->body->mbo_nlink = attr->cat_nlink;
1379 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1380 md->body->mbo_size = attr->cat_size;
1381 if (md->body->mbo_valid & OBD_MD_FLATIME)
1382 md->body->mbo_atime = attr->cat_atime;
1383 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1384 md->body->mbo_ctime = attr->cat_ctime;
1385 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1386 md->body->mbo_mtime = attr->cat_mtime;
1390 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1391 lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1395 /* Compare the old and new stripe information */
1396 if (!lsm_md_eq(lli->lli_lsm_md, lsm)) {
1397 struct lmv_stripe_md *old_lsm = lli->lli_lsm_md;
1400 CERROR("%s: inode "DFID"(%p)'s lmv layout mismatch (%p)/(%p)"
1401 "magic:0x%x/0x%x stripe count: %d/%d master_mdt: %d/%d"
1402 "hash_type:0x%x/0x%x layout: 0x%x/0x%x pool:%s/%s\n",
1403 ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1404 inode, lsm, old_lsm,
1405 lsm->lsm_md_magic, old_lsm->lsm_md_magic,
1406 lsm->lsm_md_stripe_count,
1407 old_lsm->lsm_md_stripe_count,
1408 lsm->lsm_md_master_mdt_index,
1409 old_lsm->lsm_md_master_mdt_index,
1410 lsm->lsm_md_hash_type, old_lsm->lsm_md_hash_type,
1411 lsm->lsm_md_layout_version,
1412 old_lsm->lsm_md_layout_version,
1413 lsm->lsm_md_pool_name,
1414 old_lsm->lsm_md_pool_name);
1416 for (idx = 0; idx < old_lsm->lsm_md_stripe_count; idx++) {
1417 CERROR("%s: sub FIDs in old lsm idx %d, old: "DFID"\n",
1418 ll_get_fsname(inode->i_sb, NULL, 0), idx,
1419 PFID(&old_lsm->lsm_md_oinfo[idx].lmo_fid));
1422 for (idx = 0; idx < lsm->lsm_md_stripe_count; idx++) {
1423 CERROR("%s: sub FIDs in new lsm idx %d, new: "DFID"\n",
1424 ll_get_fsname(inode->i_sb, NULL, 0), idx,
1425 PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1434 void ll_clear_inode(struct inode *inode)
1436 struct ll_inode_info *lli = ll_i2info(inode);
1437 struct ll_sb_info *sbi = ll_i2sbi(inode);
1440 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1441 PFID(ll_inode2fid(inode)), inode);
1443 if (S_ISDIR(inode->i_mode)) {
1444 /* these should have been cleared in ll_file_release */
1445 LASSERT(lli->lli_opendir_key == NULL);
1446 LASSERT(lli->lli_sai == NULL);
1447 LASSERT(lli->lli_opendir_pid == 0);
1450 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1452 LASSERT(!lli->lli_open_fd_write_count);
1453 LASSERT(!lli->lli_open_fd_read_count);
1454 LASSERT(!lli->lli_open_fd_exec_count);
1456 if (lli->lli_mds_write_och)
1457 ll_md_real_close(inode, FMODE_WRITE);
1458 if (lli->lli_mds_exec_och)
1459 ll_md_real_close(inode, FMODE_EXEC);
1460 if (lli->lli_mds_read_och)
1461 ll_md_real_close(inode, FMODE_READ);
1463 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1464 OBD_FREE(lli->lli_symlink_name,
1465 strlen(lli->lli_symlink_name) + 1);
1466 lli->lli_symlink_name = NULL;
1469 ll_xattr_cache_destroy(inode);
1471 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1472 LASSERT(lli->lli_posix_acl == NULL);
1473 if (lli->lli_remote_perms) {
1474 free_rmtperm_hash(lli->lli_remote_perms);
1475 lli->lli_remote_perms = NULL;
1478 #ifdef CONFIG_FS_POSIX_ACL
1479 else if (lli->lli_posix_acl) {
1480 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1481 LASSERT(lli->lli_remote_perms == NULL);
1482 posix_acl_release(lli->lli_posix_acl);
1483 lli->lli_posix_acl = NULL;
1486 lli->lli_inode_magic = LLI_INODE_DEAD;
1488 if (S_ISDIR(inode->i_mode))
1489 ll_dir_clear_lsm_md(inode);
1490 else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1491 LASSERT(list_empty(&lli->lli_agl_list));
1494 * XXX This has to be done before lsm is freed below, because
1495 * cl_object still uses inode lsm.
1497 cl_inode_fini(inode);
1502 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1504 struct lustre_md md;
1505 struct inode *inode = dentry->d_inode;
1506 struct ll_sb_info *sbi = ll_i2sbi(inode);
1507 struct ptlrpc_request *request = NULL;
1511 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1512 LUSTRE_OPC_ANY, NULL);
1513 if (IS_ERR(op_data))
1514 RETURN(PTR_ERR(op_data));
1516 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1518 ptlrpc_req_finished(request);
1519 if (rc == -ENOENT) {
1521 /* Unlinked special device node? Or just a race?
1522 * Pretend we done everything. */
1523 if (!S_ISREG(inode->i_mode) &&
1524 !S_ISDIR(inode->i_mode)) {
1525 ia_valid = op_data->op_attr.ia_valid;
1526 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1527 rc = simple_setattr(dentry, &op_data->op_attr);
1528 op_data->op_attr.ia_valid = ia_valid;
1530 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1531 CERROR("md_setattr fails: rc = %d\n", rc);
1536 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1537 sbi->ll_md_exp, &md);
1539 ptlrpc_req_finished(request);
1543 ia_valid = op_data->op_attr.ia_valid;
1544 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1545 * cache is not cleared yet. */
1546 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1547 if (S_ISREG(inode->i_mode))
1548 mutex_lock(&inode->i_mutex);
1549 rc = simple_setattr(dentry, &op_data->op_attr);
1550 if (S_ISREG(inode->i_mode))
1551 mutex_unlock(&inode->i_mutex);
1552 op_data->op_attr.ia_valid = ia_valid;
1554 rc = ll_update_inode(inode, &md);
1555 ptlrpc_req_finished(request);
1560 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1561 * object(s) determine the file size and mtime. Otherwise, the MDS will
1562 * keep these values until such a time that objects are allocated for it.
1563 * We do the MDS operations first, as it is checking permissions for us.
1564 * We don't to the MDS RPC if there is nothing that we want to store there,
1565 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1566 * going to do an RPC anyways.
1568 * If we are doing a truncate, we will send the mtime and ctime updates
1569 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1570 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1573 * In case of HSMimport, we only set attr on MDS.
1575 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1577 struct inode *inode = dentry->d_inode;
1578 struct ll_inode_info *lli = ll_i2info(inode);
1579 struct md_op_data *op_data = NULL;
1583 CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1584 "valid %x, hsm_import %d\n",
1585 ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1586 inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1589 if (attr->ia_valid & ATTR_SIZE) {
1590 /* Check new size against VFS/VM file size limit and rlimit */
1591 rc = inode_newsize_ok(inode, attr->ia_size);
1595 /* The maximum Lustre file size is variable, based on the
1596 * OST maximum object size and number of stripes. This
1597 * needs another check in addition to the VFS check above. */
1598 if (attr->ia_size > ll_file_maxbytes(inode)) {
1599 CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1600 PFID(&lli->lli_fid), attr->ia_size,
1601 ll_file_maxbytes(inode));
1605 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1608 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1609 if (attr->ia_valid & TIMES_SET_FLAGS) {
1610 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1611 !cfs_capable(CFS_CAP_FOWNER))
1615 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1616 if (!(attr->ia_valid & ATTR_CTIME_SET) &&
1617 (attr->ia_valid & ATTR_CTIME)) {
1618 attr->ia_ctime = CURRENT_TIME;
1619 attr->ia_valid |= ATTR_CTIME_SET;
1621 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1622 (attr->ia_valid & ATTR_ATIME)) {
1623 attr->ia_atime = CURRENT_TIME;
1624 attr->ia_valid |= ATTR_ATIME_SET;
1626 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1627 (attr->ia_valid & ATTR_MTIME)) {
1628 attr->ia_mtime = CURRENT_TIME;
1629 attr->ia_valid |= ATTR_MTIME_SET;
1632 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1633 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1634 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1635 cfs_time_current_sec());
1637 if (S_ISREG(inode->i_mode)) {
1638 if (attr->ia_valid & ATTR_SIZE)
1639 inode_dio_write_done(inode);
1640 mutex_unlock(&inode->i_mutex);
1643 /* We always do an MDS RPC, even if we're only changing the size;
1644 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1646 OBD_ALLOC_PTR(op_data);
1647 if (op_data == NULL)
1648 GOTO(out, rc = -ENOMEM);
1650 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1651 /* If we are changing file size, file content is
1652 * modified, flag it. */
1653 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1654 op_data->op_bias |= MDS_DATA_MODIFIED;
1655 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1658 op_data->op_attr = *attr;
1660 rc = ll_md_setattr(dentry, op_data);
1664 if (!S_ISREG(inode->i_mode) || hsm_import)
1667 if (attr->ia_valid & (ATTR_SIZE |
1668 ATTR_ATIME | ATTR_ATIME_SET |
1669 ATTR_MTIME | ATTR_MTIME_SET |
1670 ATTR_CTIME | ATTR_CTIME_SET)) {
1671 /* For truncate and utimes sending attributes to OSTs, setting
1672 * mtime/atime to the past will be performed under PW [0:EOF]
1673 * extent lock (new_size:EOF for truncate). It may seem
1674 * excessive to send mtime/atime updates to OSTs when not
1675 * setting times to past, but it is necessary due to possible
1676 * time de-synchronization between MDT inode and OST objects */
1677 rc = cl_setattr_ost(lli->lli_clob, attr, 0);
1680 /* If the file was restored, it needs to set dirty flag.
1682 * We've already sent MDS_DATA_MODIFIED flag in
1683 * ll_md_setattr() for truncate. However, the MDT refuses to
1684 * set the HS_DIRTY flag on released files, so we have to set
1685 * it again if the file has been restored. Please check how
1686 * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1688 * Please notice that if the file is not released, the previous
1689 * MDS_DATA_MODIFIED has taken effect and usually
1690 * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1691 * This way we can save an RPC for common open + trunc
1693 if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1694 struct hsm_state_set hss = {
1695 .hss_valid = HSS_SETMASK,
1696 .hss_setmask = HS_DIRTY,
1700 rc2 = ll_hsm_state_set(inode, &hss);
1701 /* truncate and write can happen at the same time, so that
1702 * the file can be set modified even though the file is not
1703 * restored from released state, and ll_hsm_state_set() is
1704 * not applicable for the file, and rc2 < 0 is normal in this
1707 CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1708 PFID(ll_inode2fid(inode)), rc2);
1713 if (op_data != NULL)
1714 ll_finish_md_op_data(op_data);
1716 if (S_ISREG(inode->i_mode)) {
1717 mutex_lock(&inode->i_mutex);
1718 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1719 inode_dio_wait(inode);
1722 ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1723 LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1728 int ll_setattr(struct dentry *de, struct iattr *attr)
1730 int mode = de->d_inode->i_mode;
1732 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1733 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1734 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1736 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1737 (ATTR_SIZE|ATTR_MODE)) &&
1738 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1739 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1740 !(attr->ia_mode & S_ISGID))))
1741 attr->ia_valid |= ATTR_FORCE;
1743 if ((attr->ia_valid & ATTR_MODE) &&
1745 !(attr->ia_mode & S_ISUID) &&
1746 !(attr->ia_valid & ATTR_KILL_SUID))
1747 attr->ia_valid |= ATTR_KILL_SUID;
1749 if ((attr->ia_valid & ATTR_MODE) &&
1750 ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1751 !(attr->ia_mode & S_ISGID) &&
1752 !(attr->ia_valid & ATTR_KILL_SGID))
1753 attr->ia_valid |= ATTR_KILL_SGID;
1755 return ll_setattr_raw(de, attr, false);
1758 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1759 __u64 max_age, __u32 flags)
1761 struct ll_sb_info *sbi = ll_s2sbi(sb);
1762 struct obd_statfs obd_osfs;
1766 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1768 CERROR("md_statfs fails: rc = %d\n", rc);
1772 osfs->os_type = sb->s_magic;
1774 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1775 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1777 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1778 flags |= OBD_STATFS_NODELAY;
1780 rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1782 CERROR("obd_statfs fails: rc = %d\n", rc);
1786 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1787 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1790 osfs->os_bsize = obd_osfs.os_bsize;
1791 osfs->os_blocks = obd_osfs.os_blocks;
1792 osfs->os_bfree = obd_osfs.os_bfree;
1793 osfs->os_bavail = obd_osfs.os_bavail;
1795 /* If we don't have as many objects free on the OST as inodes
1796 * on the MDS, we reduce the total number of inodes to
1797 * compensate, so that the "inodes in use" number is correct.
1799 if (obd_osfs.os_ffree < osfs->os_ffree) {
1800 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1802 osfs->os_ffree = obd_osfs.os_ffree;
1807 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1809 struct super_block *sb = de->d_sb;
1810 struct obd_statfs osfs;
1811 __u64 fsid = huge_encode_dev(sb->s_dev);
1814 CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1815 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1817 /* Some amount of caching on the client is allowed */
1818 rc = ll_statfs_internal(sb, &osfs,
1819 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1824 statfs_unpack(sfs, &osfs);
1826 /* We need to downshift for all 32-bit kernels, because we can't
1827 * tell if the kernel is being called via sys_statfs64() or not.
1828 * Stop before overflowing f_bsize - in which case it is better
1829 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1830 if (sizeof(long) < 8) {
1831 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1834 osfs.os_blocks >>= 1;
1835 osfs.os_bfree >>= 1;
1836 osfs.os_bavail >>= 1;
1840 sfs->f_blocks = osfs.os_blocks;
1841 sfs->f_bfree = osfs.os_bfree;
1842 sfs->f_bavail = osfs.os_bavail;
1843 sfs->f_fsid.val[0] = (__u32)fsid;
1844 sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1848 void ll_inode_size_lock(struct inode *inode)
1850 struct ll_inode_info *lli;
1852 LASSERT(!S_ISDIR(inode->i_mode));
1854 lli = ll_i2info(inode);
1855 mutex_lock(&lli->lli_size_mutex);
1858 void ll_inode_size_unlock(struct inode *inode)
1860 struct ll_inode_info *lli;
1862 lli = ll_i2info(inode);
1863 mutex_unlock(&lli->lli_size_mutex);
1866 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1868 struct ll_inode_info *lli = ll_i2info(inode);
1869 struct mdt_body *body = md->body;
1870 struct ll_sb_info *sbi = ll_i2sbi(inode);
1872 if (body->mbo_valid & OBD_MD_FLEASIZE)
1873 cl_file_inode_init(inode, md);
1875 if (S_ISDIR(inode->i_mode)) {
1878 rc = ll_update_lsm_md(inode, md);
1883 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1884 if (body->mbo_valid & OBD_MD_FLRMTPERM)
1885 ll_update_remote_perm(inode, md->remote_perm);
1887 #ifdef CONFIG_FS_POSIX_ACL
1888 else if (body->mbo_valid & OBD_MD_FLACL) {
1889 spin_lock(&lli->lli_lock);
1890 if (lli->lli_posix_acl)
1891 posix_acl_release(lli->lli_posix_acl);
1892 lli->lli_posix_acl = md->posix_acl;
1893 spin_unlock(&lli->lli_lock);
1896 inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1897 sbi->ll_flags & LL_SBI_32BIT_API);
1898 inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1900 if (body->mbo_valid & OBD_MD_FLATIME) {
1901 if (body->mbo_atime > LTIME_S(inode->i_atime))
1902 LTIME_S(inode->i_atime) = body->mbo_atime;
1903 lli->lli_atime = body->mbo_atime;
1906 if (body->mbo_valid & OBD_MD_FLMTIME) {
1907 if (body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1908 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1909 "to "LPU64"\n", inode->i_ino,
1910 LTIME_S(inode->i_mtime), body->mbo_mtime);
1911 LTIME_S(inode->i_mtime) = body->mbo_mtime;
1913 lli->lli_mtime = body->mbo_mtime;
1916 if (body->mbo_valid & OBD_MD_FLCTIME) {
1917 if (body->mbo_ctime > LTIME_S(inode->i_ctime))
1918 LTIME_S(inode->i_ctime) = body->mbo_ctime;
1919 lli->lli_ctime = body->mbo_ctime;
1922 if (body->mbo_valid & OBD_MD_FLMODE)
1923 inode->i_mode = (inode->i_mode & S_IFMT) |
1924 (body->mbo_mode & ~S_IFMT);
1926 if (body->mbo_valid & OBD_MD_FLTYPE)
1927 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1928 (body->mbo_mode & S_IFMT);
1930 LASSERT(inode->i_mode != 0);
1931 if (S_ISREG(inode->i_mode))
1932 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1933 LL_MAX_BLKSIZE_BITS);
1935 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1937 if (body->mbo_valid & OBD_MD_FLUID)
1938 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
1939 if (body->mbo_valid & OBD_MD_FLGID)
1940 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
1941 if (body->mbo_valid & OBD_MD_FLFLAGS)
1942 inode->i_flags = ll_ext_to_inode_flags(body->mbo_flags);
1943 if (body->mbo_valid & OBD_MD_FLNLINK)
1944 set_nlink(inode, body->mbo_nlink);
1945 if (body->mbo_valid & OBD_MD_FLRDEV)
1946 inode->i_rdev = old_decode_dev(body->mbo_rdev);
1948 if (body->mbo_valid & OBD_MD_FLID) {
1949 /* FID shouldn't be changed! */
1950 if (fid_is_sane(&lli->lli_fid)) {
1951 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
1952 "Trying to change FID "DFID
1953 " to the "DFID", inode "DFID"(%p)\n",
1954 PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
1955 PFID(ll_inode2fid(inode)), inode);
1957 lli->lli_fid = body->mbo_fid1;
1961 LASSERT(fid_seq(&lli->lli_fid) != 0);
1963 if (body->mbo_valid & OBD_MD_FLSIZE) {
1964 i_size_write(inode, body->mbo_size);
1966 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
1967 PFID(ll_inode2fid(inode)),
1968 (unsigned long long)body->mbo_size);
1970 if (body->mbo_valid & OBD_MD_FLBLOCKS)
1971 inode->i_blocks = body->mbo_blocks;
1974 if (body->mbo_valid & OBD_MD_TSTATE) {
1975 /* Set LLIF_FILE_RESTORING if restore ongoing and
1976 * clear it when done to ensure to start again
1977 * glimpsing updated attrs
1979 if (body->mbo_t_state & MS_RESTORE)
1980 ll_file_set_flag(lli, LLIF_FILE_RESTORING);
1982 ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
1988 int ll_read_inode2(struct inode *inode, void *opaque)
1990 struct lustre_md *md = opaque;
1991 struct ll_inode_info *lli = ll_i2info(inode);
1995 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1996 PFID(&lli->lli_fid), inode);
1998 /* Core attributes from the MDS first. This is a new inode, and
1999 * the VFS doesn't zero times in the core inode so we have to do
2000 * it ourselves. They will be overwritten by either MDS or OST
2001 * attributes - we just need to make sure they aren't newer. */
2002 LTIME_S(inode->i_mtime) = 0;
2003 LTIME_S(inode->i_atime) = 0;
2004 LTIME_S(inode->i_ctime) = 0;
2006 rc = ll_update_inode(inode, md);
2010 /* OIDEBUG(inode); */
2012 #ifdef HAVE_BACKING_DEV_INFO
2013 /* initializing backing dev info. */
2014 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2016 if (S_ISREG(inode->i_mode)) {
2017 struct ll_sb_info *sbi = ll_i2sbi(inode);
2018 inode->i_op = &ll_file_inode_operations;
2019 inode->i_fop = sbi->ll_fop;
2020 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2022 } else if (S_ISDIR(inode->i_mode)) {
2023 inode->i_op = &ll_dir_inode_operations;
2024 inode->i_fop = &ll_dir_operations;
2026 } else if (S_ISLNK(inode->i_mode)) {
2027 inode->i_op = &ll_fast_symlink_inode_operations;
2030 inode->i_op = &ll_special_inode_operations;
2032 init_special_inode(inode, inode->i_mode,
2041 void ll_delete_inode(struct inode *inode)
2043 struct ll_inode_info *lli = ll_i2info(inode);
2046 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2047 /* It is last chance to write out dirty pages,
2048 * otherwise we may lose data while umount */
2049 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
2051 truncate_inode_pages_final(&inode->i_data);
2053 LASSERTF(inode->i_data.nrpages == 0, "inode="DFID"(%p) nrpages=%lu, "
2054 "see https://jira.hpdd.intel.com/browse/LU-118\n",
2055 PFID(ll_inode2fid(inode)), inode, inode->i_data.nrpages);
2057 #ifdef HAVE_SBOPS_EVICT_INODE
2058 ll_clear_inode(inode);
2065 int ll_iocontrol(struct inode *inode, struct file *file,
2066 unsigned int cmd, unsigned long arg)
2068 struct ll_sb_info *sbi = ll_i2sbi(inode);
2069 struct ptlrpc_request *req = NULL;
2074 case FSFILT_IOC_GETFLAGS: {
2075 struct mdt_body *body;
2076 struct md_op_data *op_data;
2078 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2079 0, 0, LUSTRE_OPC_ANY,
2081 if (IS_ERR(op_data))
2082 RETURN(PTR_ERR(op_data));
2084 op_data->op_valid = OBD_MD_FLFLAGS;
2085 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2086 ll_finish_md_op_data(op_data);
2088 CERROR("%s: failure inode "DFID": rc = %d\n",
2089 sbi->ll_md_exp->exp_obd->obd_name,
2090 PFID(ll_inode2fid(inode)), rc);
2094 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2096 flags = body->mbo_flags;
2098 ptlrpc_req_finished(req);
2100 RETURN(put_user(flags, (int __user *)arg));
2102 case FSFILT_IOC_SETFLAGS: {
2104 struct md_op_data *op_data;
2105 struct cl_object *obj;
2107 if (get_user(flags, (int __user *)arg))
2110 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2111 LUSTRE_OPC_ANY, NULL);
2112 if (IS_ERR(op_data))
2113 RETURN(PTR_ERR(op_data));
2115 op_data->op_attr_flags = flags;
2116 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2117 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2118 ll_finish_md_op_data(op_data);
2119 ptlrpc_req_finished(req);
2123 inode->i_flags = ll_ext_to_inode_flags(flags);
2125 obj = ll_i2info(inode)->lli_clob;
2129 OBD_ALLOC_PTR(attr);
2133 attr->ia_valid = ATTR_ATTR_FLAG;
2134 rc = cl_setattr_ost(obj, attr, flags);
2146 int ll_flush_ctx(struct inode *inode)
2148 struct ll_sb_info *sbi = ll_i2sbi(inode);
2150 CDEBUG(D_SEC, "flush context for user %d\n",
2151 from_kuid(&init_user_ns, current_uid()));
2153 obd_set_info_async(NULL, sbi->ll_md_exp,
2154 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2156 obd_set_info_async(NULL, sbi->ll_dt_exp,
2157 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2162 /* umount -f client means force down, don't save state */
2163 void ll_umount_begin(struct super_block *sb)
2165 struct ll_sb_info *sbi = ll_s2sbi(sb);
2166 struct obd_device *obd;
2167 struct obd_ioctl_data *ioc_data;
2170 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2171 sb->s_count, atomic_read(&sb->s_active));
2173 obd = class_exp2obd(sbi->ll_md_exp);
2175 CERROR("Invalid MDC connection handle "LPX64"\n",
2176 sbi->ll_md_exp->exp_handle.h_cookie);
2182 obd = class_exp2obd(sbi->ll_dt_exp);
2184 CERROR("Invalid LOV connection handle "LPX64"\n",
2185 sbi->ll_dt_exp->exp_handle.h_cookie);
2191 OBD_ALLOC_PTR(ioc_data);
2193 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2194 sizeof *ioc_data, ioc_data, NULL);
2196 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2197 sizeof *ioc_data, ioc_data, NULL);
2199 OBD_FREE_PTR(ioc_data);
2202 /* Really, we'd like to wait until there are no requests outstanding,
2203 * and then continue. For now, we just invalidate the requests,
2204 * schedule() and sleep one second if needed, and hope.
2210 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2212 struct ll_sb_info *sbi = ll_s2sbi(sb);
2213 char *profilenm = get_profile_name(sb);
2217 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2218 read_only = *flags & MS_RDONLY;
2219 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2220 sizeof(KEY_READ_ONLY),
2221 KEY_READ_ONLY, sizeof(read_only),
2224 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2225 profilenm, read_only ?
2226 "read-only" : "read-write", err);
2231 sb->s_flags |= MS_RDONLY;
2233 sb->s_flags &= ~MS_RDONLY;
2235 if (sbi->ll_flags & LL_SBI_VERBOSE)
2236 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2237 read_only ? "read-only" : "read-write");
2243 * Cleanup the open handle that is cached on MDT-side.
2245 * For open case, the client side open handling thread may hit error
2246 * after the MDT grant the open. Under such case, the client should
2247 * send close RPC to the MDT as cleanup; otherwise, the open handle
2248 * on the MDT will be leaked there until the client umount or evicted.
2250 * In further, if someone unlinked the file, because the open handle
2251 * holds the reference on such file/object, then it will block the
2252 * subsequent threads that want to locate such object via FID.
2254 * \param[in] sb super block for this file-system
2255 * \param[in] open_req pointer to the original open request
2257 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2259 struct mdt_body *body;
2260 struct md_op_data *op_data;
2261 struct ptlrpc_request *close_req = NULL;
2262 struct obd_export *exp = ll_s2sbi(sb)->ll_md_exp;
2265 body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2266 OBD_ALLOC_PTR(op_data);
2267 if (op_data == NULL) {
2268 CWARN("%s: cannot allocate op_data to release open handle for "
2270 ll_get_fsname(sb, NULL, 0), PFID(&body->mbo_fid1));
2275 op_data->op_fid1 = body->mbo_fid1;
2276 op_data->op_handle = body->mbo_handle;
2277 op_data->op_mod_time = cfs_time_current_sec();
2278 md_close(exp, op_data, NULL, &close_req);
2279 ptlrpc_req_finished(close_req);
2280 ll_finish_md_op_data(op_data);
2285 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2286 struct super_block *sb, struct lookup_intent *it)
2288 struct ll_sb_info *sbi = NULL;
2289 struct lustre_md md = { NULL };
2293 LASSERT(*inode || sb);
2294 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2295 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2296 sbi->ll_md_exp, &md);
2301 rc = ll_update_inode(*inode, &md);
2305 LASSERT(sb != NULL);
2308 * At this point server returns to client's same fid as client
2309 * generated for creating. So using ->fid1 is okay here.
2311 if (!fid_is_sane(&md.body->mbo_fid1)) {
2312 CERROR("%s: Fid is insane "DFID"\n",
2313 ll_get_fsname(sb, NULL, 0),
2314 PFID(&md.body->mbo_fid1));
2315 GOTO(out, rc = -EINVAL);
2318 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2319 sbi->ll_flags & LL_SBI_32BIT_API),
2321 if (IS_ERR(*inode)) {
2322 #ifdef CONFIG_FS_POSIX_ACL
2324 posix_acl_release(md.posix_acl);
2325 md.posix_acl = NULL;
2328 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2330 CERROR("new_inode -fatal: rc %d\n", rc);
2335 /* Handling piggyback layout lock.
2336 * Layout lock can be piggybacked by getattr and open request.
2337 * The lsm can be applied to inode only if it comes with a layout lock
2338 * otherwise correct layout may be overwritten, for example:
2339 * 1. proc1: mdt returns a lsm but not granting layout
2340 * 2. layout was changed by another client
2341 * 3. proc2: refresh layout and layout lock granted
2342 * 4. proc1: to apply a stale layout */
2343 if (it != NULL && it->it_lock_mode != 0) {
2344 struct lustre_handle lockh;
2345 struct ldlm_lock *lock;
2347 lockh.cookie = it->it_lock_handle;
2348 lock = ldlm_handle2lock(&lockh);
2349 LASSERT(lock != NULL);
2350 if (ldlm_has_layout(lock)) {
2351 struct cl_object_conf conf;
2353 memset(&conf, 0, sizeof(conf));
2354 conf.coc_opc = OBJECT_CONF_SET;
2355 conf.coc_inode = *inode;
2356 conf.coc_lock = lock;
2357 conf.u.coc_layout = md.layout;
2358 (void)ll_layout_conf(*inode, &conf);
2360 LDLM_LOCK_PUT(lock);
2366 md_free_lustre_md(sbi->ll_md_exp, &md);
2369 if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2370 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2375 int ll_obd_statfs(struct inode *inode, void __user *arg)
2377 struct ll_sb_info *sbi = NULL;
2378 struct obd_export *exp;
2380 struct obd_ioctl_data *data = NULL;
2384 if (!inode || !(sbi = ll_i2sbi(inode)))
2385 GOTO(out_statfs, rc = -EINVAL);
2387 rc = obd_ioctl_getdata(&buf, &len, arg);
2389 GOTO(out_statfs, rc);
2392 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2393 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2394 GOTO(out_statfs, rc = -EINVAL);
2396 if (data->ioc_inllen1 != sizeof(__u32) ||
2397 data->ioc_inllen2 != sizeof(__u32) ||
2398 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2399 data->ioc_plen2 != sizeof(struct obd_uuid))
2400 GOTO(out_statfs, rc = -EINVAL);
2402 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2403 if (type & LL_STATFS_LMV)
2404 exp = sbi->ll_md_exp;
2405 else if (type & LL_STATFS_LOV)
2406 exp = sbi->ll_dt_exp;
2408 GOTO(out_statfs, rc = -ENODEV);
2410 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2412 GOTO(out_statfs, rc);
2415 obd_ioctl_freedata(buf, len);
2419 int ll_process_config(struct lustre_cfg *lcfg)
2421 struct super_block *sb;
2426 /* The instance name contains the sb: lustre-client-aacfe000 */
2427 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2428 if (!ptr || !*(++ptr))
2430 if (sscanf(ptr, "%lx", &x) != 1)
2432 sb = (struct super_block *)x;
2433 /* This better be a real Lustre superblock! */
2434 LASSERT(s2lsi(sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2436 /* Note we have not called client_common_fill_super yet, so
2437 proc fns must be able to handle that! */
2438 rc = class_process_proc_param(PARAM_LLITE, lprocfs_llite_obd_vars,
2445 /* this function prepares md_op_data hint for passing it down to MD stack. */
2446 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2447 struct inode *i1, struct inode *i2,
2448 const char *name, size_t namelen,
2449 __u32 mode, __u32 opc, void *data)
2451 LASSERT(i1 != NULL);
2454 /* Do not reuse namelen for something else. */
2456 return ERR_PTR(-EINVAL);
2458 if (namelen > ll_i2sbi(i1)->ll_namelen)
2459 return ERR_PTR(-ENAMETOOLONG);
2461 if (!lu_name_is_valid_2(name, namelen))
2462 return ERR_PTR(-EINVAL);
2465 if (op_data == NULL)
2466 OBD_ALLOC_PTR(op_data);
2468 if (op_data == NULL)
2469 return ERR_PTR(-ENOMEM);
2471 ll_i2gids(op_data->op_suppgids, i1, i2);
2472 op_data->op_fid1 = *ll_inode2fid(i1);
2473 op_data->op_default_stripe_offset = -1;
2474 if (S_ISDIR(i1->i_mode)) {
2475 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2476 if (opc == LUSTRE_OPC_MKDIR)
2477 op_data->op_default_stripe_offset =
2478 ll_i2info(i1)->lli_def_stripe_offset;
2482 op_data->op_fid2 = *ll_inode2fid(i2);
2483 if (S_ISDIR(i2->i_mode))
2484 op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2486 fid_zero(&op_data->op_fid2);
2489 if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2490 op_data->op_cli_flags |= CLI_HASH64;
2492 if (ll_need_32bit_api(ll_i2sbi(i1)))
2493 op_data->op_cli_flags |= CLI_API32;
2495 op_data->op_name = name;
2496 op_data->op_namelen = namelen;
2497 op_data->op_mode = mode;
2498 op_data->op_mod_time = cfs_time_current_sec();
2499 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2500 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2501 op_data->op_cap = cfs_curproc_cap_pack();
2502 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2503 filename_is_volatile(name, namelen, &op_data->op_mds)) {
2504 op_data->op_bias |= MDS_CREATE_VOLATILE;
2506 op_data->op_mds = 0;
2508 op_data->op_data = data;
2513 void ll_finish_md_op_data(struct md_op_data *op_data)
2515 OBD_FREE_PTR(op_data);
2518 #ifdef HAVE_SUPEROPS_USE_DENTRY
2519 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2521 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2524 struct ll_sb_info *sbi;
2526 #ifdef HAVE_SUPEROPS_USE_DENTRY
2527 LASSERT((seq != NULL) && (dentry != NULL));
2528 sbi = ll_s2sbi(dentry->d_sb);
2530 LASSERT((seq != NULL) && (vfs != NULL));
2531 sbi = ll_s2sbi(vfs->mnt_sb);
2534 if (sbi->ll_flags & LL_SBI_NOLCK)
2535 seq_puts(seq, ",nolock");
2537 if (sbi->ll_flags & LL_SBI_FLOCK)
2538 seq_puts(seq, ",flock");
2540 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2541 seq_puts(seq, ",localflock");
2543 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2544 seq_puts(seq, ",user_xattr");
2546 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2547 seq_puts(seq, ",lazystatfs");
2549 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2550 seq_puts(seq, ",user_fid2path");
2552 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2553 seq_puts(seq, ",always_ping");
2559 * Get obd name by cmd, and copy out to user space
2561 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2563 struct ll_sb_info *sbi = ll_i2sbi(inode);
2564 struct obd_device *obd;
2567 if (cmd == OBD_IOC_GETDTNAME)
2568 obd = class_exp2obd(sbi->ll_dt_exp);
2569 else if (cmd == OBD_IOC_GETMDNAME)
2570 obd = class_exp2obd(sbi->ll_md_exp);
2577 if (copy_to_user((void __user *)arg, obd->obd_name,
2578 strlen(obd->obd_name) + 1))
2585 * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2586 * fsname will be returned in this buffer; otherwise, a static buffer will be
2587 * used to store the fsname and returned to caller.
2589 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2591 static char fsname_static[MTI_NAME_MAXLEN];
2592 struct lustre_sb_info *lsi = s2lsi(sb);
2597 /* this means the caller wants to use static buffer
2598 * and it doesn't care about race. Usually this is
2599 * in error reporting path */
2600 buf = fsname_static;
2601 buflen = sizeof(fsname_static);
2604 len = strlen(lsi->lsi_lmd->lmd_profile);
2605 ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2606 if (ptr && (strcmp(ptr, "-client") == 0))
2609 if (unlikely(len >= buflen))
2611 strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2617 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2624 p.mnt = current->fs->root.mnt;
2626 path = d_path(&p, buf, bufsize);
2631 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2633 char *buf, *path = NULL;
2634 struct dentry *dentry = NULL;
2635 struct inode *inode = page->mapping->host;
2637 /* this can be called inside spin lock so use GFP_ATOMIC. */
2638 buf = (char *)__get_free_page(GFP_ATOMIC);
2640 dentry = d_find_alias(page->mapping->host);
2642 path = ll_d_path(dentry, buf, PAGE_SIZE);
2646 "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2647 "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2648 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2649 PFID(ll_inode2fid(inode)),
2650 (path && !IS_ERR(path)) ? path : "", ioret);
2656 free_page((unsigned long)buf);
2659 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2660 struct lov_user_md **kbuf)
2662 struct lov_user_md lum;
2666 if (copy_from_user(&lum, md, sizeof(lum)))
2669 lum_size = ll_lov_user_md_size(&lum);
2673 OBD_ALLOC(*kbuf, lum_size);
2677 if (copy_from_user(*kbuf, md, lum_size) != 0) {
2678 OBD_FREE(*kbuf, lum_size);
2686 * Compute llite root squash state after a change of root squash
2687 * configuration setting or add/remove of a lnet nid
2689 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2691 struct root_squash_info *squash = &sbi->ll_squash;
2694 lnet_process_id_t id;
2696 /* Update norootsquash flag */
2697 down_write(&squash->rsi_sem);
2698 if (list_empty(&squash->rsi_nosquash_nids))
2699 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2701 /* Do not apply root squash as soon as one of our NIDs is
2702 * in the nosquash_nids list */
2705 while (LNetGetId(i++, &id) != -ENOENT) {
2706 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2708 if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2714 sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2716 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2718 up_write(&squash->rsi_sem);
2722 * Parse linkea content to extract information about a given hardlink
2724 * \param[in] ldata - Initialized linkea data
2725 * \param[in] linkno - Link identifier
2726 * \param[out] parent_fid - The entry's parent FID
2727 * \param[out] ln - Entry name destination buffer
2729 * \retval 0 on success
2730 * \retval Appropriate negative error code on failure
2732 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2733 struct lu_fid *parent_fid, struct lu_name *ln)
2739 rc = linkea_init(ldata);
2743 if (linkno >= ldata->ld_leh->leh_reccount)
2744 /* beyond last link */
2747 linkea_first_entry(ldata);
2748 for (idx = 0; ldata->ld_lee != NULL; idx++) {
2749 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2754 linkea_next_entry(ldata);
2764 * Get parent FID and name of an identified link. Operation is performed for
2765 * a given link number, letting the caller iterate over linkno to list one or
2766 * all links of an entry.
2768 * \param[in] file - File descriptor against which to perform the operation
2769 * \param[in,out] arg - User-filled structure containing the linkno to operate
2770 * on and the available size. It is eventually filled with
2771 * the requested information or left untouched on error
2773 * \retval - 0 on success
2774 * \retval - Appropriate negative error code on failure
2776 int ll_getparent(struct file *file, struct getparent __user *arg)
2778 struct dentry *dentry = file->f_path.dentry;
2779 struct inode *inode = dentry->d_inode;
2780 struct linkea_data *ldata;
2781 struct lu_buf buf = LU_BUF_NULL;
2783 struct lu_fid parent_fid;
2790 if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2791 !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2794 if (get_user(name_size, &arg->gp_name_size))
2797 if (get_user(linkno, &arg->gp_linkno))
2800 if (name_size > PATH_MAX)
2803 OBD_ALLOC(ldata, sizeof(*ldata));
2807 rc = linkea_data_new(ldata, &buf);
2809 GOTO(ldata_free, rc);
2811 rc = ll_getxattr(dentry, XATTR_NAME_LINK, buf.lb_buf, buf.lb_len);
2815 rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2819 if (ln.ln_namelen >= name_size)
2820 GOTO(lb_free, rc = -EOVERFLOW);
2822 if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2823 GOTO(lb_free, rc = -EFAULT);
2825 if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2826 GOTO(lb_free, rc = -EFAULT);
2828 if (put_user('\0', arg->gp_name + ln.ln_namelen))
2829 GOTO(lb_free, rc = -EFAULT);
2834 OBD_FREE(ldata, sizeof(*ldata));