4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2016, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lustre/llite/llite_lib.c
34 * Lustre Light Super operations
37 #define DEBUG_SUBSYSTEM S_LLITE
39 #include <linux/module.h>
40 #include <linux/statfs.h>
41 #include <linux/time.h>
42 #include <linux/types.h>
43 #include <linux/version.h>
45 #include <linux/user_namespace.h>
46 #ifdef HAVE_UIDGID_HEADER
47 # include <linux/uidgid.h>
49 #include <linux/security.h>
51 #include <uapi/linux/lustre/lustre_ioctl.h>
52 #include <lustre_ha.h>
53 #include <lustre_dlm.h>
54 #include <lprocfs_status.h>
55 #include <lustre_disk.h>
56 #include <uapi/linux/lustre/lustre_param.h>
57 #include <lustre_log.h>
58 #include <cl_object.h>
59 #include <obd_cksum.h>
60 #include "llite_internal.h"
62 struct kmem_cache *ll_file_data_slab;
65 #define log2(n) ffz(~(n))
68 static struct ll_sb_info *ll_init_sbi(void)
70 struct ll_sb_info *sbi = NULL;
72 unsigned long lru_page_max;
82 spin_lock_init(&sbi->ll_lock);
83 mutex_init(&sbi->ll_lco.lco_lock);
84 spin_lock_init(&sbi->ll_pp_extent_lock);
85 spin_lock_init(&sbi->ll_process_lock);
86 sbi->ll_rw_stats_on = 0;
89 pages = si.totalram - si.totalhigh;
90 lru_page_max = pages / 2;
92 /* initialize ll_cache data */
93 sbi->ll_cache = cl_cache_init(lru_page_max);
94 if (sbi->ll_cache == NULL) {
95 OBD_FREE(sbi, sizeof(*sbi));
99 sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
100 SBI_DEFAULT_READAHEAD_MAX);
101 sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
102 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
103 SBI_DEFAULT_READAHEAD_WHOLE_MAX;
105 ll_generate_random_uuid(uuid);
106 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
107 CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
109 sbi->ll_flags |= LL_SBI_VERBOSE;
110 #ifdef ENABLE_CHECKSUM
111 sbi->ll_flags |= LL_SBI_CHECKSUM;
114 #ifdef HAVE_LRU_RESIZE_SUPPORT
115 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
117 sbi->ll_flags |= LL_SBI_LAZYSTATFS;
119 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
120 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
122 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
126 /* metadata statahead is enabled by default */
127 sbi->ll_sa_max = LL_SA_RPC_DEF;
128 atomic_set(&sbi->ll_sa_total, 0);
129 atomic_set(&sbi->ll_sa_wrong, 0);
130 atomic_set(&sbi->ll_sa_running, 0);
131 atomic_set(&sbi->ll_agl_total, 0);
132 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
133 sbi->ll_flags |= LL_SBI_FAST_READ;
136 sbi->ll_squash.rsi_uid = 0;
137 sbi->ll_squash.rsi_gid = 0;
138 INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
139 init_rwsem(&sbi->ll_squash.rsi_sem);
144 static void ll_free_sbi(struct super_block *sb)
146 struct ll_sb_info *sbi = ll_s2sbi(sb);
150 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
151 cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
152 if (sbi->ll_cache != NULL) {
153 cl_cache_decref(sbi->ll_cache);
154 sbi->ll_cache = NULL;
156 OBD_FREE(sbi, sizeof(*sbi));
161 static inline int obd_connect_has_secctx(struct obd_connect_data *data)
163 return data->ocd_connect_flags & OBD_CONNECT_FLAGS2 &&
164 data->ocd_connect_flags2 & OBD_CONNECT2_FILE_SECCTX;
167 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
168 struct vfsmount *mnt)
170 struct inode *root = NULL;
171 struct ll_sb_info *sbi = ll_s2sbi(sb);
172 struct obd_device *obd;
173 struct obd_statfs *osfs = NULL;
174 struct ptlrpc_request *request = NULL;
175 struct obd_connect_data *data = NULL;
176 struct obd_uuid *uuid;
177 struct md_op_data *op_data;
178 struct lustre_md lmd;
180 int size, err, checksum;
183 obd = class_name2obd(md);
185 CERROR("MD %s: not setup or attached\n", md);
199 /* indicate MDT features supported by this client */
200 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
201 OBD_CONNECT_ATTRFID | OBD_CONNECT_GRANT |
202 OBD_CONNECT_VERSION | OBD_CONNECT_BRW_SIZE |
203 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
204 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
205 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
206 OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 |
207 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
208 OBD_CONNECT_64BITHASH |
209 OBD_CONNECT_EINPROGRESS |
210 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
211 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS|
212 OBD_CONNECT_MAX_EASIZE |
213 OBD_CONNECT_FLOCK_DEAD |
214 OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
215 OBD_CONNECT_OPEN_BY_FID |
216 OBD_CONNECT_DIR_STRIPE |
217 OBD_CONNECT_BULK_MBITS | OBD_CONNECT_CKSUM |
218 OBD_CONNECT_SUBTREE |
219 OBD_CONNECT_FLAGS2 | OBD_CONNECT_MULTIMODRPCS;
221 data->ocd_connect_flags2 = 0;
223 #ifdef HAVE_LRU_RESIZE_SUPPORT
224 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
225 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
227 #ifdef CONFIG_FS_POSIX_ACL
228 data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK |
229 OBD_CONNECT_LARGE_ACL;
232 data->ocd_cksum_types = cksum_types_supported_client();
234 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
235 /* flag mdc connection as lightweight, only used for test
236 * purpose, use with care */
237 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
239 data->ocd_ibits_known = MDS_INODELOCK_FULL;
240 data->ocd_version = LUSTRE_VERSION_CODE;
242 if (sb->s_flags & MS_RDONLY)
243 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
244 if (sbi->ll_flags & LL_SBI_USER_XATTR)
245 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
248 /* Setting this indicates we correctly support S_NOSEC (See kernel
249 * commit 9e1f1de02c2275d7172e18dc4e7c2065777611bf)
251 sb->s_flags |= MS_NOSEC;
254 if (sbi->ll_flags & LL_SBI_FLOCK)
255 sbi->ll_fop = &ll_file_operations_flock;
256 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
257 sbi->ll_fop = &ll_file_operations;
259 sbi->ll_fop = &ll_file_operations_noflock;
261 /* always ping even if server suppress_pings */
262 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
263 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
265 #ifdef HAVE_SECURITY_DENTRY_INIT_SECURITY
266 data->ocd_connect_flags2 |= OBD_CONNECT2_FILE_SECCTX;
267 #endif /* HAVE_SECURITY_DENTRY_INIT_SECURITY */
269 data->ocd_brw_size = MD_MAX_BRW_SIZE;
271 err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
273 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
274 "recovery, of which this client is not a "
275 "part. Please wait for recovery to complete,"
276 " abort, or time out.\n", md);
279 CERROR("cannot connect to %s: rc = %d\n", md, err);
283 sbi->ll_md_exp->exp_connect_data = *data;
285 err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
286 LUSTRE_SEQ_METADATA);
288 CERROR("%s: Can't init metadata layer FID infrastructure, "
289 "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
293 /* For mount, we only need fs info from MDT0, and also in DNE, it
294 * can make sure the client can be mounted as long as MDT0 is
296 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
297 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
298 OBD_STATFS_FOR_MDT0);
300 GOTO(out_md_fid, err);
302 /* This needs to be after statfs to ensure connect has finished.
303 * Note that "data" does NOT contain the valid connect reply.
304 * If connecting to a 1.8 server there will be no LMV device, so
305 * we can access the MDC export directly and exp_connect_flags will
306 * be non-zero, but if accessing an upgraded 2.1 server it will
307 * have the correct flags filled in.
308 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
309 valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
310 if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
311 valid != CLIENT_CONNECT_MDT_REQD) {
314 OBD_ALLOC_WAIT(buf, PAGE_SIZE);
315 obd_connect_flags2str(buf, PAGE_SIZE,
316 valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
317 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
318 "feature(s) needed for correct operation "
319 "of this client (%s). Please upgrade "
320 "server or downgrade client.\n",
321 sbi->ll_md_exp->exp_obd->obd_name, buf);
322 OBD_FREE(buf, PAGE_SIZE);
323 GOTO(out_md_fid, err = -EPROTO);
326 size = sizeof(*data);
327 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
328 KEY_CONN_DATA, &size, data);
330 CERROR("%s: Get connect data failed: rc = %d\n",
331 sbi->ll_md_exp->exp_obd->obd_name, err);
332 GOTO(out_md_fid, err);
335 LASSERT(osfs->os_bsize);
336 sb->s_blocksize = osfs->os_bsize;
337 sb->s_blocksize_bits = log2(osfs->os_bsize);
338 sb->s_magic = LL_SUPER_MAGIC;
339 sb->s_maxbytes = MAX_LFS_FILESIZE;
340 sbi->ll_namelen = osfs->os_namelen;
341 sbi->ll_mnt.mnt = current->fs->root.mnt;
343 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
344 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
345 LCONSOLE_INFO("Disabling user_xattr feature because "
346 "it is not supported on the server\n");
347 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
350 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
352 sb->s_flags |= MS_POSIXACL;
354 sbi->ll_flags |= LL_SBI_ACL;
356 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
358 sb->s_flags &= ~MS_POSIXACL;
360 sbi->ll_flags &= ~LL_SBI_ACL;
363 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
364 sbi->ll_flags |= LL_SBI_64BIT_HASH;
366 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
367 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
369 if (obd_connect_has_secctx(data))
370 sbi->ll_flags |= LL_SBI_FILE_SECCTX;
372 if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
373 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
374 LCONSOLE_INFO("%s: disabling xattr cache due to "
375 "unknown maximum xattr size.\n", dt);
376 } else if (!sbi->ll_xattr_cache_set) {
377 /* If xattr_cache is already set (no matter 0 or 1)
378 * during processing llog, it won't be enabled here. */
379 sbi->ll_flags |= LL_SBI_XATTR_CACHE;
380 sbi->ll_xattr_cache_enabled = 1;
384 obd = class_name2obd(dt);
386 CERROR("DT %s: not setup or attached\n", dt);
387 GOTO(out_md_fid, err = -ENODEV);
390 /* pass client page size via ocd_grant_blkbits, the server should report
391 * back its backend blocksize for grant calculation purpose */
392 data->ocd_grant_blkbits = PAGE_SHIFT;
394 /* indicate OST features supported by this client */
395 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
396 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
397 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
398 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
399 OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
400 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
401 OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
402 OBD_CONNECT_EINPROGRESS |
403 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
404 OBD_CONNECT_LAYOUTLOCK |
405 OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
406 OBD_CONNECT_BULK_MBITS |
409 /* The client currently advertises support for OBD_CONNECT_LOCKAHEAD_OLD so it
410 * can interoperate with an older version of lockahead which was released prior
411 * to landing in master. This support will be dropped when 2.13 development
412 * starts. At the point, we should not just drop the connect flag (below), we
413 * should also remove the support in the code.
415 * Removing it means a few things:
416 * 1. Remove this section here
417 * 2. Remove CEF_NONBLOCK in ll_file_lockahead()
418 * 3. Remove function exp_connect_lockahead_old
419 * 4. Remove LDLM_FL_LOCKAHEAD_OLD_RESERVED in lustre_dlm_flags.h
421 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 12, 50, 0)
422 data->ocd_connect_flags |= OBD_CONNECT_LOCKAHEAD_OLD;
425 data->ocd_connect_flags2 = OBD_CONNECT2_LOCKAHEAD;
427 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
428 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
430 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
431 * disabled by default, because it can still be enabled on the
432 * fly via /proc. As a consequence, we still need to come to an
433 * agreement on the supported algorithms at connect time */
434 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
436 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
437 data->ocd_cksum_types = OBD_CKSUM_ADLER;
439 data->ocd_cksum_types = cksum_types_supported_client();
441 #ifdef HAVE_LRU_RESIZE_SUPPORT
442 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
444 /* always ping even if server suppress_pings */
445 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
446 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
448 CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
449 "ocd_grant: %d\n", data->ocd_connect_flags,
450 data->ocd_version, data->ocd_grant);
452 obd->obd_upcall.onu_owner = &sbi->ll_lco;
453 obd->obd_upcall.onu_upcall = cl_ocd_update;
455 data->ocd_brw_size = DT_MAX_BRW_SIZE;
457 err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
460 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
461 "recovery, of which this client is not a "
462 "part. Please wait for recovery to "
463 "complete, abort, or time out.\n", dt);
466 CERROR("%s: Cannot connect to %s: rc = %d\n",
467 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
471 sbi->ll_dt_exp->exp_connect_data = *data;
473 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
474 LUSTRE_SEQ_METADATA);
476 CERROR("%s: Can't init data layer FID infrastructure, "
477 "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
481 mutex_lock(&sbi->ll_lco.lco_lock);
482 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
483 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
484 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
485 mutex_unlock(&sbi->ll_lco.lco_lock);
487 fid_zero(&sbi->ll_root_fid);
488 err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
491 CERROR("cannot mds_connect: rc = %d\n", err);
492 GOTO(out_lock_cn_cb, err);
494 if (!fid_is_sane(&sbi->ll_root_fid)) {
495 CERROR("%s: Invalid root fid "DFID" during mount\n",
496 sbi->ll_md_exp->exp_obd->obd_name,
497 PFID(&sbi->ll_root_fid));
498 GOTO(out_lock_cn_cb, err = -EINVAL);
500 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
502 sb->s_op = &lustre_super_operations;
503 #ifdef HAVE_XATTR_HANDLER_FLAGS
504 sb->s_xattr = ll_xattr_handlers;
506 #if THREAD_SIZE >= 8192 /*b=17630*/
507 sb->s_export_op = &lustre_export_operations;
511 * XXX: move this to after cbd setup? */
512 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
513 if (sbi->ll_flags & LL_SBI_ACL)
514 valid |= OBD_MD_FLACL;
516 OBD_ALLOC_PTR(op_data);
518 GOTO(out_lock_cn_cb, err = -ENOMEM);
520 op_data->op_fid1 = sbi->ll_root_fid;
521 op_data->op_mode = 0;
522 op_data->op_valid = valid;
524 err = md_getattr(sbi->ll_md_exp, op_data, &request);
526 OBD_FREE_PTR(op_data);
528 CERROR("%s: md_getattr failed for root: rc = %d\n",
529 sbi->ll_md_exp->exp_obd->obd_name, err);
530 GOTO(out_lock_cn_cb, err);
533 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
534 sbi->ll_md_exp, &lmd);
536 CERROR("failed to understand root inode md: rc = %d\n", err);
537 ptlrpc_req_finished(request);
538 GOTO(out_lock_cn_cb, err);
541 LASSERT(fid_is_sane(&sbi->ll_root_fid));
542 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
543 sbi->ll_flags & LL_SBI_32BIT_API),
545 md_free_lustre_md(sbi->ll_md_exp, &lmd);
546 ptlrpc_req_finished(request);
549 #ifdef CONFIG_FS_POSIX_ACL
551 posix_acl_release(lmd.posix_acl);
552 lmd.posix_acl = NULL;
555 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
557 CERROR("lustre_lite: bad iget4 for root\n");
561 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
562 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
563 KEY_CHECKSUM, sizeof(checksum), &checksum,
566 CERROR("%s: Set checksum failed: rc = %d\n",
567 sbi->ll_dt_exp->exp_obd->obd_name, err);
572 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
573 KEY_CACHE_SET, sizeof(*sbi->ll_cache),
574 sbi->ll_cache, NULL);
576 CERROR("%s: Set cache_set failed: rc = %d\n",
577 sbi->ll_dt_exp->exp_obd->obd_name, err);
581 sb->s_root = d_make_root(root);
582 if (sb->s_root == NULL) {
583 CERROR("%s: can't make root dentry\n",
584 ll_get_fsname(sb, NULL, 0));
585 GOTO(out_root, err = -ENOMEM);
587 #ifdef HAVE_DCACHE_LOCK
588 sb->s_root->d_op = &ll_d_ops;
591 sbi->ll_sdev_orig = sb->s_dev;
593 /* We set sb->s_dev equal on all lustre clients in order to support
594 * NFS export clustering. NFSD requires that the FSID be the same
596 /* s_dev is also used in lt_compare() to compare two fs, but that is
597 * only a node-local comparison. */
598 uuid = obd_get_uuid(sbi->ll_md_exp);
600 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
607 if (sbi->ll_proc_root != NULL) {
608 err = lprocfs_ll_register_obd(sb, dt);
610 CERROR("%s: could not register %s in llite: rc = %d\n",
611 dt, ll_get_fsname(sb, NULL, 0), err);
614 err = lprocfs_ll_register_obd(sb, md);
616 CERROR("%s: could not register %s in llite: rc = %d\n",
617 md, ll_get_fsname(sb, NULL, 0), err);
627 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
629 obd_disconnect(sbi->ll_dt_exp);
630 sbi->ll_dt_exp = NULL;
632 obd_fid_fini(sbi->ll_md_exp->exp_obd);
634 obd_disconnect(sbi->ll_md_exp);
635 sbi->ll_md_exp = NULL;
644 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
648 size = sizeof(*lmmsize);
649 rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
650 KEY_MAX_EASIZE, &size, lmmsize);
652 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
653 sbi->ll_dt_exp->exp_obd->obd_name, rc);
658 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
659 KEY_MAX_EASIZE, &size, lmmsize);
661 CERROR("Get max mdsize error rc %d\n", rc);
667 * Get the value of the default_easize parameter.
669 * \see client_obd::cl_default_mds_easize
671 * \param[in] sbi superblock info for this filesystem
672 * \param[out] lmmsize pointer to storage location for value
674 * \retval 0 on success
675 * \retval negative negated errno on failure
677 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
682 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
683 KEY_DEFAULT_EASIZE, &size, lmmsize);
685 CERROR("Get default mdsize error rc %d\n", rc);
691 * Set the default_easize parameter to the given value.
693 * \see client_obd::cl_default_mds_easize
695 * \param[in] sbi superblock info for this filesystem
696 * \param[in] lmmsize the size to set
698 * \retval 0 on success
699 * \retval negative negated errno on failure
701 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
705 if (lmmsize < sizeof(struct lov_mds_md) ||
706 lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
709 rc = obd_set_info_async(NULL, sbi->ll_md_exp,
710 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
711 sizeof(int), &lmmsize, NULL);
716 static void client_common_put_super(struct super_block *sb)
718 struct ll_sb_info *sbi = ll_s2sbi(sb);
723 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
724 obd_disconnect(sbi->ll_dt_exp);
725 sbi->ll_dt_exp = NULL;
727 lprocfs_ll_unregister_mountpoint(sbi);
729 obd_fid_fini(sbi->ll_md_exp->exp_obd);
730 obd_disconnect(sbi->ll_md_exp);
731 sbi->ll_md_exp = NULL;
736 void ll_kill_super(struct super_block *sb)
738 struct ll_sb_info *sbi;
742 if (!(sb->s_flags & MS_ACTIVE))
746 /* we need restore s_dev from changed for clustred NFS before put_super
747 * because new kernels have cached s_dev and change sb->s_dev in
748 * put_super not affected real removing devices */
750 sb->s_dev = sbi->ll_sdev_orig;
751 sbi->ll_umounting = 1;
753 /* wait running statahead threads to quit */
754 while (atomic_read(&sbi->ll_sa_running) > 0) {
755 set_current_state(TASK_UNINTERRUPTIBLE);
756 schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
763 static inline int ll_set_opt(const char *opt, char *data, int fl)
765 if (strncmp(opt, data, strlen(opt)) != 0)
771 /* non-client-specific mount options are parsed in lmd_parse */
772 static int ll_options(char *options, int *flags)
775 char *s1 = options, *s2;
781 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
784 CDEBUG(D_SUPER, "next opt=%s\n", s1);
785 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
790 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
795 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
800 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
805 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
810 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
815 tmp = ll_set_opt("context", s1, 1);
818 tmp = ll_set_opt("fscontext", s1, 1);
821 tmp = ll_set_opt("defcontext", s1, 1);
824 tmp = ll_set_opt("rootcontext", s1, 1);
827 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
832 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
838 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
843 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
848 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
853 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
858 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
863 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
868 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
873 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
878 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
883 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
888 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
894 s2 = strchr(s1, ',');
902 void ll_lli_init(struct ll_inode_info *lli)
904 lli->lli_inode_magic = LLI_INODE_MAGIC;
906 spin_lock_init(&lli->lli_lock);
907 lli->lli_posix_acl = NULL;
908 /* Do not set lli_fid, it has been initialized already. */
909 fid_zero(&lli->lli_pfid);
910 lli->lli_mds_read_och = NULL;
911 lli->lli_mds_write_och = NULL;
912 lli->lli_mds_exec_och = NULL;
913 lli->lli_open_fd_read_count = 0;
914 lli->lli_open_fd_write_count = 0;
915 lli->lli_open_fd_exec_count = 0;
916 mutex_init(&lli->lli_och_mutex);
917 spin_lock_init(&lli->lli_agl_lock);
918 spin_lock_init(&lli->lli_layout_lock);
919 ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
920 lli->lli_clob = NULL;
922 init_rwsem(&lli->lli_xattrs_list_rwsem);
923 mutex_init(&lli->lli_xattrs_enq_lock);
925 LASSERT(lli->lli_vfs_inode.i_mode != 0);
926 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
927 mutex_init(&lli->lli_readdir_mutex);
928 lli->lli_opendir_key = NULL;
930 spin_lock_init(&lli->lli_sa_lock);
931 lli->lli_opendir_pid = 0;
932 lli->lli_sa_enabled = 0;
933 lli->lli_def_stripe_offset = -1;
935 mutex_init(&lli->lli_size_mutex);
936 lli->lli_symlink_name = NULL;
937 init_rwsem(&lli->lli_trunc_sem);
938 range_lock_tree_init(&lli->lli_write_tree);
939 init_rwsem(&lli->lli_glimpse_sem);
940 lli->lli_glimpse_time = 0;
941 INIT_LIST_HEAD(&lli->lli_agl_list);
942 lli->lli_agl_index = 0;
943 lli->lli_async_rc = 0;
945 mutex_init(&lli->lli_layout_mutex);
946 memset(lli->lli_jobid, 0, LUSTRE_JOBID_SIZE);
949 static inline int ll_bdi_register(struct backing_dev_info *bdi)
951 static atomic_t ll_bdi_num = ATOMIC_INIT(0);
953 bdi->name = "lustre";
954 return bdi_register(bdi, NULL, "lustre-%d",
955 atomic_inc_return(&ll_bdi_num));
958 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
960 struct lustre_profile *lprof = NULL;
961 struct lustre_sb_info *lsi = s2lsi(sb);
962 struct ll_sb_info *sbi;
963 char *dt = NULL, *md = NULL;
964 char *profilenm = get_profile_name(sb);
965 struct config_llog_instance *cfg;
966 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
967 const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
973 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
979 try_module_get(THIS_MODULE);
981 /* client additional sb info */
982 lsi->lsi_llsbi = sbi = ll_init_sbi();
984 module_put(THIS_MODULE);
989 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
993 err = bdi_init(&lsi->lsi_bdi);
996 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
997 #ifdef HAVE_BDI_CAP_MAP_COPY
998 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1000 lsi->lsi_bdi.capabilities = 0;
1002 err = ll_bdi_register(&lsi->lsi_bdi);
1004 GOTO(out_free, err);
1006 sb->s_bdi = &lsi->lsi_bdi;
1007 #ifndef HAVE_DCACHE_LOCK
1008 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1009 sb->s_d_op = &ll_d_ops;
1012 /* Call lprocfs_ll_register_mountpoint() before lustre_process_log()
1013 * so that "llite.*.*" params can be processed correctly. */
1014 if (proc_lustre_fs_root != NULL) {
1015 err = lprocfs_ll_register_mountpoint(proc_lustre_fs_root, sb);
1017 CERROR("%s: could not register mountpoint in llite: "
1018 "rc = %d\n", ll_get_fsname(sb, NULL, 0), err);
1023 /* Generate a string unique to this super, in case some joker tries
1024 to mount the same fs at two mount points.
1025 Use the address of the super itself.*/
1026 cfg->cfg_instance = sb;
1027 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1028 cfg->cfg_callback = class_config_llog_handler;
1029 cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1030 /* set up client obds */
1031 err = lustre_process_log(sb, profilenm, cfg);
1033 GOTO(out_proc, err);
1035 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1036 lprof = class_get_profile(profilenm);
1037 if (lprof == NULL) {
1038 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1039 " read from the MGS. Does that filesystem "
1040 "exist?\n", profilenm);
1041 GOTO(out_proc, err = -EINVAL);
1043 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1044 lprof->lp_md, lprof->lp_dt);
1046 dt_len = strlen(lprof->lp_dt) + instlen + 2;
1047 OBD_ALLOC(dt, dt_len);
1049 GOTO(out_proc, err = -ENOMEM);
1050 snprintf(dt, dt_len - 1, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1052 md_len = strlen(lprof->lp_md) + instlen + 2;
1053 OBD_ALLOC(md, md_len);
1055 GOTO(out_proc, err = -ENOMEM);
1056 snprintf(md, md_len - 1, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1058 /* connections, registrations, sb setup */
1059 err = client_common_fill_super(sb, md, dt, mnt);
1061 GOTO(out_proc, err);
1063 sbi->ll_client_common_fill_super_succeeded = 1;
1067 lprocfs_ll_unregister_mountpoint(sbi);
1070 OBD_FREE(md, md_len);
1072 OBD_FREE(dt, dt_len);
1074 class_put_profile(lprof);
1077 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1078 LCONSOLE_WARN("Mounted %s\n", profilenm);
1082 } /* ll_fill_super */
1084 void ll_put_super(struct super_block *sb)
1086 struct config_llog_instance cfg, params_cfg;
1087 struct obd_device *obd;
1088 struct lustre_sb_info *lsi = s2lsi(sb);
1089 struct ll_sb_info *sbi = ll_s2sbi(sb);
1090 char *profilenm = get_profile_name(sb);
1092 int next, force = 1, rc = 0;
1095 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1097 cfg.cfg_instance = sb;
1098 lustre_end_log(sb, profilenm, &cfg);
1100 params_cfg.cfg_instance = sb;
1101 lustre_end_log(sb, PARAMS_FILENAME, ¶ms_cfg);
1103 if (sbi->ll_md_exp) {
1104 obd = class_exp2obd(sbi->ll_md_exp);
1106 force = obd->obd_force;
1109 /* Wait for unstable pages to be committed to stable storage */
1111 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1112 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1113 atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1117 ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1118 if (force == 0 && rc != -EINTR)
1119 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1122 /* We need to set force before the lov_disconnect in
1123 lustre_common_put_super, since l_d cleans up osc's as well. */
1126 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1128 obd->obd_force = force;
1132 if (sbi->ll_client_common_fill_super_succeeded) {
1133 /* Only if client_common_fill_super succeeded */
1134 client_common_put_super(sb);
1138 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1139 class_manual_cleanup(obd);
1142 if (sbi->ll_flags & LL_SBI_VERBOSE)
1143 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1146 class_del_profile(profilenm);
1148 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1149 bdi_destroy(&lsi->lsi_bdi);
1150 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1154 lsi->lsi_llsbi = NULL;
1156 lustre_common_put_super(sb);
1158 cl_env_cache_purge(~0);
1160 module_put(THIS_MODULE);
1163 } /* client_put_super */
1165 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1167 struct inode *inode = NULL;
1169 /* NOTE: we depend on atomic igrab() -bzzz */
1170 lock_res_and_lock(lock);
1171 if (lock->l_resource->lr_lvb_inode) {
1172 struct ll_inode_info * lli;
1173 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1174 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1175 inode = igrab(lock->l_resource->lr_lvb_inode);
1177 inode = lock->l_resource->lr_lvb_inode;
1178 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1179 D_WARNING, lock, "lr_lvb_inode %p is "
1180 "bogus: magic %08x",
1181 lock->l_resource->lr_lvb_inode,
1182 lli->lli_inode_magic);
1186 unlock_res_and_lock(lock);
1190 void ll_dir_clear_lsm_md(struct inode *inode)
1192 struct ll_inode_info *lli = ll_i2info(inode);
1194 LASSERT(S_ISDIR(inode->i_mode));
1196 if (lli->lli_lsm_md != NULL) {
1197 lmv_free_memmd(lli->lli_lsm_md);
1198 lli->lli_lsm_md = NULL;
1202 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1203 const struct lu_fid *fid,
1204 struct lustre_md *md)
1206 struct ll_sb_info *sbi = ll_s2sbi(sb);
1207 struct mdt_body *body = md->body;
1208 struct inode *inode;
1212 ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1213 inode = iget_locked(sb, ino);
1214 if (inode == NULL) {
1215 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1216 ll_get_fsname(sb, NULL, 0), PFID(fid));
1217 RETURN(ERR_PTR(-ENOENT));
1220 if (inode->i_state & I_NEW) {
1221 struct ll_inode_info *lli = ll_i2info(inode);
1222 struct lmv_stripe_md *lsm = md->lmv;
1224 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1225 (body->mbo_mode & S_IFMT);
1226 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1229 LTIME_S(inode->i_mtime) = 0;
1230 LTIME_S(inode->i_atime) = 0;
1231 LTIME_S(inode->i_ctime) = 0;
1234 #ifdef HAVE_BACKING_DEV_INFO
1235 /* initializing backing dev info. */
1236 inode->i_mapping->backing_dev_info =
1237 &s2lsi(inode->i_sb)->lsi_bdi;
1239 inode->i_op = &ll_dir_inode_operations;
1240 inode->i_fop = &ll_dir_operations;
1241 lli->lli_fid = *fid;
1244 LASSERT(lsm != NULL);
1245 /* master object FID */
1246 lli->lli_pfid = body->mbo_fid1;
1247 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1248 lli, PFID(fid), PFID(&lli->lli_pfid));
1249 unlock_new_inode(inode);
1255 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1258 struct lmv_stripe_md *lsm = md->lmv;
1261 LASSERT(lsm != NULL);
1262 /* XXX sigh, this lsm_root initialization should be in
1263 * LMV layer, but it needs ll_iget right now, so we
1264 * put this here right now. */
1265 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1266 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1267 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1268 /* Unfortunately ll_iget will call ll_update_inode,
1269 * where the initialization of slave inode is slightly
1270 * different, so it reset lsm_md to NULL to avoid
1271 * initializing lsm for slave inode. */
1272 /* For migrating inode, master stripe and master object will
1273 * be same, so we only need assign this inode */
1274 if (lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION && i == 0)
1275 lsm->lsm_md_oinfo[i].lmo_root = inode;
1277 lsm->lsm_md_oinfo[i].lmo_root =
1278 ll_iget_anon_dir(inode->i_sb, fid, md);
1280 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1281 int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1283 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1291 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1292 const struct lmv_stripe_md *lsm_md2)
1294 return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1295 lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1296 lsm_md1->lsm_md_master_mdt_index ==
1297 lsm_md2->lsm_md_master_mdt_index &&
1298 lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1299 lsm_md1->lsm_md_layout_version ==
1300 lsm_md2->lsm_md_layout_version &&
1301 strcmp(lsm_md1->lsm_md_pool_name,
1302 lsm_md2->lsm_md_pool_name) == 0;
1305 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1307 struct ll_inode_info *lli = ll_i2info(inode);
1308 struct lmv_stripe_md *lsm = md->lmv;
1312 LASSERT(S_ISDIR(inode->i_mode));
1313 CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1314 PFID(ll_inode2fid(inode)));
1316 /* no striped information from request. */
1318 if (lli->lli_lsm_md == NULL) {
1320 } else if (lli->lli_lsm_md->lsm_md_hash_type &
1321 LMV_HASH_FLAG_MIGRATION) {
1322 /* migration is done, the temporay MIGRATE layout has
1324 CDEBUG(D_INODE, DFID" finish migration.\n",
1325 PFID(ll_inode2fid(inode)));
1326 lmv_free_memmd(lli->lli_lsm_md);
1327 lli->lli_lsm_md = NULL;
1330 /* The lustre_md from req does not include stripeEA,
1331 * see ll_md_setattr */
1336 /* set the directory layout */
1337 if (lli->lli_lsm_md == NULL) {
1338 struct cl_attr *attr;
1340 rc = ll_init_lsm_md(inode, md);
1344 /* set md->lmv to NULL, so the following free lustre_md
1345 * will not free this lsm */
1347 lli->lli_lsm_md = lsm;
1349 OBD_ALLOC_PTR(attr);
1353 /* validate the lsm */
1354 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1355 ll_md_blocking_ast);
1361 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1362 md->body->mbo_nlink = attr->cat_nlink;
1363 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1364 md->body->mbo_size = attr->cat_size;
1365 if (md->body->mbo_valid & OBD_MD_FLATIME)
1366 md->body->mbo_atime = attr->cat_atime;
1367 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1368 md->body->mbo_ctime = attr->cat_ctime;
1369 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1370 md->body->mbo_mtime = attr->cat_mtime;
1374 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1375 lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1379 /* Compare the old and new stripe information */
1380 if (!lsm_md_eq(lli->lli_lsm_md, lsm)) {
1381 struct lmv_stripe_md *old_lsm = lli->lli_lsm_md;
1384 CERROR("%s: inode "DFID"(%p)'s lmv layout mismatch (%p)/(%p)"
1385 "magic:0x%x/0x%x stripe count: %d/%d master_mdt: %d/%d"
1386 "hash_type:0x%x/0x%x layout: 0x%x/0x%x pool:%s/%s\n",
1387 ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1388 inode, lsm, old_lsm,
1389 lsm->lsm_md_magic, old_lsm->lsm_md_magic,
1390 lsm->lsm_md_stripe_count,
1391 old_lsm->lsm_md_stripe_count,
1392 lsm->lsm_md_master_mdt_index,
1393 old_lsm->lsm_md_master_mdt_index,
1394 lsm->lsm_md_hash_type, old_lsm->lsm_md_hash_type,
1395 lsm->lsm_md_layout_version,
1396 old_lsm->lsm_md_layout_version,
1397 lsm->lsm_md_pool_name,
1398 old_lsm->lsm_md_pool_name);
1400 for (idx = 0; idx < old_lsm->lsm_md_stripe_count; idx++) {
1401 CERROR("%s: sub FIDs in old lsm idx %d, old: "DFID"\n",
1402 ll_get_fsname(inode->i_sb, NULL, 0), idx,
1403 PFID(&old_lsm->lsm_md_oinfo[idx].lmo_fid));
1406 for (idx = 0; idx < lsm->lsm_md_stripe_count; idx++) {
1407 CERROR("%s: sub FIDs in new lsm idx %d, new: "DFID"\n",
1408 ll_get_fsname(inode->i_sb, NULL, 0), idx,
1409 PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1418 void ll_clear_inode(struct inode *inode)
1420 struct ll_inode_info *lli = ll_i2info(inode);
1421 struct ll_sb_info *sbi = ll_i2sbi(inode);
1424 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1425 PFID(ll_inode2fid(inode)), inode);
1427 if (S_ISDIR(inode->i_mode)) {
1428 /* these should have been cleared in ll_file_release */
1429 LASSERT(lli->lli_opendir_key == NULL);
1430 LASSERT(lli->lli_sai == NULL);
1431 LASSERT(lli->lli_opendir_pid == 0);
1434 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1436 LASSERT(!lli->lli_open_fd_write_count);
1437 LASSERT(!lli->lli_open_fd_read_count);
1438 LASSERT(!lli->lli_open_fd_exec_count);
1440 if (lli->lli_mds_write_och)
1441 ll_md_real_close(inode, FMODE_WRITE);
1442 if (lli->lli_mds_exec_och)
1443 ll_md_real_close(inode, FMODE_EXEC);
1444 if (lli->lli_mds_read_och)
1445 ll_md_real_close(inode, FMODE_READ);
1447 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1448 OBD_FREE(lli->lli_symlink_name,
1449 strlen(lli->lli_symlink_name) + 1);
1450 lli->lli_symlink_name = NULL;
1453 ll_xattr_cache_destroy(inode);
1455 #ifdef CONFIG_FS_POSIX_ACL
1456 forget_all_cached_acls(inode);
1457 if (lli->lli_posix_acl) {
1458 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1459 posix_acl_release(lli->lli_posix_acl);
1460 lli->lli_posix_acl = NULL;
1463 lli->lli_inode_magic = LLI_INODE_DEAD;
1465 if (S_ISDIR(inode->i_mode))
1466 ll_dir_clear_lsm_md(inode);
1467 else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1468 LASSERT(list_empty(&lli->lli_agl_list));
1471 * XXX This has to be done before lsm is freed below, because
1472 * cl_object still uses inode lsm.
1474 cl_inode_fini(inode);
1479 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1481 struct lustre_md md;
1482 struct inode *inode = dentry->d_inode;
1483 struct ll_sb_info *sbi = ll_i2sbi(inode);
1484 struct ptlrpc_request *request = NULL;
1488 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1489 LUSTRE_OPC_ANY, NULL);
1490 if (IS_ERR(op_data))
1491 RETURN(PTR_ERR(op_data));
1493 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1495 ptlrpc_req_finished(request);
1496 if (rc == -ENOENT) {
1498 /* Unlinked special device node? Or just a race?
1499 * Pretend we done everything. */
1500 if (!S_ISREG(inode->i_mode) &&
1501 !S_ISDIR(inode->i_mode)) {
1502 ia_valid = op_data->op_attr.ia_valid;
1503 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1504 rc = simple_setattr(dentry, &op_data->op_attr);
1505 op_data->op_attr.ia_valid = ia_valid;
1507 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1508 CERROR("md_setattr fails: rc = %d\n", rc);
1513 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1514 sbi->ll_md_exp, &md);
1516 ptlrpc_req_finished(request);
1520 ia_valid = op_data->op_attr.ia_valid;
1521 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1522 * cache is not cleared yet. */
1523 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1524 if (S_ISREG(inode->i_mode))
1526 rc = simple_setattr(dentry, &op_data->op_attr);
1527 if (S_ISREG(inode->i_mode))
1528 inode_unlock(inode);
1529 op_data->op_attr.ia_valid = ia_valid;
1531 rc = ll_update_inode(inode, &md);
1532 ptlrpc_req_finished(request);
1537 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1538 * object(s) determine the file size and mtime. Otherwise, the MDS will
1539 * keep these values until such a time that objects are allocated for it.
1540 * We do the MDS operations first, as it is checking permissions for us.
1541 * We don't to the MDS RPC if there is nothing that we want to store there,
1542 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1543 * going to do an RPC anyways.
1545 * If we are doing a truncate, we will send the mtime and ctime updates
1546 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1547 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1550 * In case of HSMimport, we only set attr on MDS.
1552 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1554 struct inode *inode = dentry->d_inode;
1555 struct ll_inode_info *lli = ll_i2info(inode);
1556 struct md_op_data *op_data = NULL;
1560 CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1561 "valid %x, hsm_import %d\n",
1562 ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1563 inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1566 if (attr->ia_valid & ATTR_SIZE) {
1567 /* Check new size against VFS/VM file size limit and rlimit */
1568 rc = inode_newsize_ok(inode, attr->ia_size);
1572 /* The maximum Lustre file size is variable, based on the
1573 * OST maximum object size and number of stripes. This
1574 * needs another check in addition to the VFS check above. */
1575 if (attr->ia_size > ll_file_maxbytes(inode)) {
1576 CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
1577 PFID(&lli->lli_fid), attr->ia_size,
1578 ll_file_maxbytes(inode));
1582 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1585 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1586 if (attr->ia_valid & TIMES_SET_FLAGS) {
1587 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1588 !cfs_capable(CFS_CAP_FOWNER))
1592 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1593 if (!(attr->ia_valid & ATTR_CTIME_SET) &&
1594 (attr->ia_valid & ATTR_CTIME)) {
1595 attr->ia_ctime = current_time(inode);
1596 attr->ia_valid |= ATTR_CTIME_SET;
1598 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1599 (attr->ia_valid & ATTR_ATIME)) {
1600 attr->ia_atime = current_time(inode);
1601 attr->ia_valid |= ATTR_ATIME_SET;
1603 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1604 (attr->ia_valid & ATTR_MTIME)) {
1605 attr->ia_mtime = current_time(inode);
1606 attr->ia_valid |= ATTR_MTIME_SET;
1609 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1610 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %llu\n",
1611 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1612 (s64)ktime_get_real_seconds());
1614 if (S_ISREG(inode->i_mode)) {
1615 if (attr->ia_valid & ATTR_SIZE)
1616 inode_dio_write_done(inode);
1617 inode_unlock(inode);
1620 /* We always do an MDS RPC, even if we're only changing the size;
1621 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1623 OBD_ALLOC_PTR(op_data);
1624 if (op_data == NULL)
1625 GOTO(out, rc = -ENOMEM);
1627 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1628 /* If we are changing file size, file content is
1629 * modified, flag it. */
1630 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1631 op_data->op_bias |= MDS_DATA_MODIFIED;
1632 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1635 op_data->op_attr = *attr;
1637 rc = ll_md_setattr(dentry, op_data);
1641 if (!S_ISREG(inode->i_mode) || hsm_import)
1644 if (attr->ia_valid & (ATTR_SIZE |
1645 ATTR_ATIME | ATTR_ATIME_SET |
1646 ATTR_MTIME | ATTR_MTIME_SET |
1647 ATTR_CTIME | ATTR_CTIME_SET)) {
1648 /* For truncate and utimes sending attributes to OSTs, setting
1649 * mtime/atime to the past will be performed under PW [0:EOF]
1650 * extent lock (new_size:EOF for truncate). It may seem
1651 * excessive to send mtime/atime updates to OSTs when not
1652 * setting times to past, but it is necessary due to possible
1653 * time de-synchronization between MDT inode and OST objects */
1654 rc = cl_setattr_ost(lli->lli_clob, attr, 0);
1657 /* If the file was restored, it needs to set dirty flag.
1659 * We've already sent MDS_DATA_MODIFIED flag in
1660 * ll_md_setattr() for truncate. However, the MDT refuses to
1661 * set the HS_DIRTY flag on released files, so we have to set
1662 * it again if the file has been restored. Please check how
1663 * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1665 * Please notice that if the file is not released, the previous
1666 * MDS_DATA_MODIFIED has taken effect and usually
1667 * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1668 * This way we can save an RPC for common open + trunc
1670 if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1671 struct hsm_state_set hss = {
1672 .hss_valid = HSS_SETMASK,
1673 .hss_setmask = HS_DIRTY,
1677 rc2 = ll_hsm_state_set(inode, &hss);
1678 /* truncate and write can happen at the same time, so that
1679 * the file can be set modified even though the file is not
1680 * restored from released state, and ll_hsm_state_set() is
1681 * not applicable for the file, and rc2 < 0 is normal in this
1684 CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1685 PFID(ll_inode2fid(inode)), rc2);
1690 if (op_data != NULL)
1691 ll_finish_md_op_data(op_data);
1693 if (S_ISREG(inode->i_mode)) {
1695 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1696 inode_dio_wait(inode);
1697 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
1698 * flag. ll_update_inode (called from ll_md_setattr), clears
1699 * inode flags, so there is a gap where S_NOSEC is not set.
1700 * This can cause a writer to take the i_mutex unnecessarily,
1701 * but this is safe to do and should be rare. */
1702 inode_has_no_xattr(inode);
1705 ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1706 LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1711 int ll_setattr(struct dentry *de, struct iattr *attr)
1713 int mode = de->d_inode->i_mode;
1715 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1716 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1717 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1719 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1720 (ATTR_SIZE|ATTR_MODE)) &&
1721 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1722 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1723 !(attr->ia_mode & S_ISGID))))
1724 attr->ia_valid |= ATTR_FORCE;
1726 if ((attr->ia_valid & ATTR_MODE) &&
1728 !(attr->ia_mode & S_ISUID) &&
1729 !(attr->ia_valid & ATTR_KILL_SUID))
1730 attr->ia_valid |= ATTR_KILL_SUID;
1732 if ((attr->ia_valid & ATTR_MODE) &&
1733 ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1734 !(attr->ia_mode & S_ISGID) &&
1735 !(attr->ia_valid & ATTR_KILL_SGID))
1736 attr->ia_valid |= ATTR_KILL_SGID;
1738 /* avoid polluted from ATTR_TIMES_SET,
1739 * projid is not expected to be set here */
1740 attr->ia_valid &= ~MDS_ATTR_PROJID;
1742 return ll_setattr_raw(de, attr, false);
1745 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1746 __u64 max_age, __u32 flags)
1748 struct ll_sb_info *sbi = ll_s2sbi(sb);
1749 struct obd_statfs obd_osfs;
1753 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1755 CERROR("md_statfs fails: rc = %d\n", rc);
1759 osfs->os_type = sb->s_magic;
1761 CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1762 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1764 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1765 flags |= OBD_STATFS_NODELAY;
1767 rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1769 CERROR("obd_statfs fails: rc = %d\n", rc);
1773 CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1774 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1777 osfs->os_bsize = obd_osfs.os_bsize;
1778 osfs->os_blocks = obd_osfs.os_blocks;
1779 osfs->os_bfree = obd_osfs.os_bfree;
1780 osfs->os_bavail = obd_osfs.os_bavail;
1782 /* If we don't have as many objects free on the OST as inodes
1783 * on the MDS, we reduce the total number of inodes to
1784 * compensate, so that the "inodes in use" number is correct.
1786 if (obd_osfs.os_ffree < osfs->os_ffree) {
1787 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1789 osfs->os_ffree = obd_osfs.os_ffree;
1794 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1796 struct super_block *sb = de->d_sb;
1797 struct obd_statfs osfs;
1798 __u64 fsid = huge_encode_dev(sb->s_dev);
1801 CDEBUG(D_VFSTRACE, "VFS Op: at %llu jiffies\n", get_jiffies_64());
1802 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1804 /* Some amount of caching on the client is allowed */
1805 rc = ll_statfs_internal(sb, &osfs,
1806 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1811 statfs_unpack(sfs, &osfs);
1813 /* We need to downshift for all 32-bit kernels, because we can't
1814 * tell if the kernel is being called via sys_statfs64() or not.
1815 * Stop before overflowing f_bsize - in which case it is better
1816 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1817 if (sizeof(long) < 8) {
1818 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1821 osfs.os_blocks >>= 1;
1822 osfs.os_bfree >>= 1;
1823 osfs.os_bavail >>= 1;
1827 sfs->f_blocks = osfs.os_blocks;
1828 sfs->f_bfree = osfs.os_bfree;
1829 sfs->f_bavail = osfs.os_bavail;
1830 sfs->f_fsid.val[0] = (__u32)fsid;
1831 sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1835 void ll_inode_size_lock(struct inode *inode)
1837 struct ll_inode_info *lli;
1839 LASSERT(!S_ISDIR(inode->i_mode));
1841 lli = ll_i2info(inode);
1842 mutex_lock(&lli->lli_size_mutex);
1845 void ll_inode_size_unlock(struct inode *inode)
1847 struct ll_inode_info *lli;
1849 lli = ll_i2info(inode);
1850 mutex_unlock(&lli->lli_size_mutex);
1853 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1855 struct ll_inode_info *lli = ll_i2info(inode);
1856 struct mdt_body *body = md->body;
1857 struct ll_sb_info *sbi = ll_i2sbi(inode);
1860 if (body->mbo_valid & OBD_MD_FLEASIZE) {
1861 rc = cl_file_inode_init(inode, md);
1866 if (S_ISDIR(inode->i_mode)) {
1867 rc = ll_update_lsm_md(inode, md);
1872 #ifdef CONFIG_FS_POSIX_ACL
1873 if (body->mbo_valid & OBD_MD_FLACL) {
1874 spin_lock(&lli->lli_lock);
1875 if (lli->lli_posix_acl)
1876 posix_acl_release(lli->lli_posix_acl);
1877 lli->lli_posix_acl = md->posix_acl;
1878 spin_unlock(&lli->lli_lock);
1881 inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1882 sbi->ll_flags & LL_SBI_32BIT_API);
1883 inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1885 if (body->mbo_valid & OBD_MD_FLATIME) {
1886 if (body->mbo_atime > LTIME_S(inode->i_atime))
1887 LTIME_S(inode->i_atime) = body->mbo_atime;
1888 lli->lli_atime = body->mbo_atime;
1891 if (body->mbo_valid & OBD_MD_FLMTIME) {
1892 if (body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1893 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1894 "to %llu\n", inode->i_ino,
1895 LTIME_S(inode->i_mtime), body->mbo_mtime);
1896 LTIME_S(inode->i_mtime) = body->mbo_mtime;
1898 lli->lli_mtime = body->mbo_mtime;
1901 if (body->mbo_valid & OBD_MD_FLCTIME) {
1902 if (body->mbo_ctime > LTIME_S(inode->i_ctime))
1903 LTIME_S(inode->i_ctime) = body->mbo_ctime;
1904 lli->lli_ctime = body->mbo_ctime;
1907 /* Clear i_flags to remove S_NOSEC before permissions are updated */
1908 if (body->mbo_valid & OBD_MD_FLFLAGS)
1909 inode->i_flags = ll_ext_to_inode_flags(body->mbo_flags);
1910 if (body->mbo_valid & OBD_MD_FLMODE)
1911 inode->i_mode = (inode->i_mode & S_IFMT) |
1912 (body->mbo_mode & ~S_IFMT);
1914 if (body->mbo_valid & OBD_MD_FLTYPE)
1915 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1916 (body->mbo_mode & S_IFMT);
1918 LASSERT(inode->i_mode != 0);
1919 if (S_ISREG(inode->i_mode))
1920 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1921 LL_MAX_BLKSIZE_BITS);
1923 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1925 if (body->mbo_valid & OBD_MD_FLUID)
1926 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
1927 if (body->mbo_valid & OBD_MD_FLGID)
1928 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
1929 if (body->mbo_valid & OBD_MD_FLPROJID)
1930 lli->lli_projid = body->mbo_projid;
1931 if (body->mbo_valid & OBD_MD_FLNLINK)
1932 set_nlink(inode, body->mbo_nlink);
1933 if (body->mbo_valid & OBD_MD_FLRDEV)
1934 inode->i_rdev = old_decode_dev(body->mbo_rdev);
1936 if (body->mbo_valid & OBD_MD_FLID) {
1937 /* FID shouldn't be changed! */
1938 if (fid_is_sane(&lli->lli_fid)) {
1939 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
1940 "Trying to change FID "DFID
1941 " to the "DFID", inode "DFID"(%p)\n",
1942 PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
1943 PFID(ll_inode2fid(inode)), inode);
1945 lli->lli_fid = body->mbo_fid1;
1949 LASSERT(fid_seq(&lli->lli_fid) != 0);
1951 if (body->mbo_valid & OBD_MD_FLSIZE) {
1952 i_size_write(inode, body->mbo_size);
1954 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
1955 PFID(ll_inode2fid(inode)),
1956 (unsigned long long)body->mbo_size);
1958 if (body->mbo_valid & OBD_MD_FLBLOCKS)
1959 inode->i_blocks = body->mbo_blocks;
1962 if (body->mbo_valid & OBD_MD_TSTATE) {
1963 /* Set LLIF_FILE_RESTORING if restore ongoing and
1964 * clear it when done to ensure to start again
1965 * glimpsing updated attrs
1967 if (body->mbo_t_state & MS_RESTORE)
1968 ll_file_set_flag(lli, LLIF_FILE_RESTORING);
1970 ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
1976 int ll_read_inode2(struct inode *inode, void *opaque)
1978 struct lustre_md *md = opaque;
1979 struct ll_inode_info *lli = ll_i2info(inode);
1983 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1984 PFID(&lli->lli_fid), inode);
1986 /* Core attributes from the MDS first. This is a new inode, and
1987 * the VFS doesn't zero times in the core inode so we have to do
1988 * it ourselves. They will be overwritten by either MDS or OST
1989 * attributes - we just need to make sure they aren't newer. */
1990 LTIME_S(inode->i_mtime) = 0;
1991 LTIME_S(inode->i_atime) = 0;
1992 LTIME_S(inode->i_ctime) = 0;
1994 rc = ll_update_inode(inode, md);
1998 /* OIDEBUG(inode); */
2000 #ifdef HAVE_BACKING_DEV_INFO
2001 /* initializing backing dev info. */
2002 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2004 if (S_ISREG(inode->i_mode)) {
2005 struct ll_sb_info *sbi = ll_i2sbi(inode);
2006 inode->i_op = &ll_file_inode_operations;
2007 inode->i_fop = sbi->ll_fop;
2008 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2010 } else if (S_ISDIR(inode->i_mode)) {
2011 inode->i_op = &ll_dir_inode_operations;
2012 inode->i_fop = &ll_dir_operations;
2014 } else if (S_ISLNK(inode->i_mode)) {
2015 inode->i_op = &ll_fast_symlink_inode_operations;
2018 inode->i_op = &ll_special_inode_operations;
2020 init_special_inode(inode, inode->i_mode,
2029 void ll_delete_inode(struct inode *inode)
2031 struct ll_inode_info *lli = ll_i2info(inode);
2034 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2035 /* It is last chance to write out dirty pages,
2036 * otherwise we may lose data while umount */
2037 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
2039 truncate_inode_pages_final(&inode->i_data);
2041 LASSERTF(inode->i_data.nrpages == 0, "inode="DFID"(%p) nrpages=%lu, "
2042 "see https://jira.hpdd.intel.com/browse/LU-118\n",
2043 PFID(ll_inode2fid(inode)), inode, inode->i_data.nrpages);
2045 #ifdef HAVE_SBOPS_EVICT_INODE
2046 ll_clear_inode(inode);
2053 int ll_iocontrol(struct inode *inode, struct file *file,
2054 unsigned int cmd, unsigned long arg)
2056 struct ll_sb_info *sbi = ll_i2sbi(inode);
2057 struct ptlrpc_request *req = NULL;
2062 case FSFILT_IOC_GETFLAGS: {
2063 struct mdt_body *body;
2064 struct md_op_data *op_data;
2066 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2067 0, 0, LUSTRE_OPC_ANY,
2069 if (IS_ERR(op_data))
2070 RETURN(PTR_ERR(op_data));
2072 op_data->op_valid = OBD_MD_FLFLAGS;
2073 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2074 ll_finish_md_op_data(op_data);
2076 CERROR("%s: failure inode "DFID": rc = %d\n",
2077 sbi->ll_md_exp->exp_obd->obd_name,
2078 PFID(ll_inode2fid(inode)), rc);
2082 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2084 flags = body->mbo_flags;
2086 ptlrpc_req_finished(req);
2088 RETURN(put_user(flags, (int __user *)arg));
2090 case FSFILT_IOC_SETFLAGS: {
2092 struct md_op_data *op_data;
2093 struct cl_object *obj;
2095 if (get_user(flags, (int __user *)arg))
2098 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2099 LUSTRE_OPC_ANY, NULL);
2100 if (IS_ERR(op_data))
2101 RETURN(PTR_ERR(op_data));
2103 op_data->op_attr_flags = flags;
2104 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2105 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2106 ll_finish_md_op_data(op_data);
2107 ptlrpc_req_finished(req);
2111 inode->i_flags = ll_ext_to_inode_flags(flags);
2113 obj = ll_i2info(inode)->lli_clob;
2117 OBD_ALLOC_PTR(attr);
2121 attr->ia_valid = ATTR_ATTR_FLAG;
2122 rc = cl_setattr_ost(obj, attr, flags);
2134 int ll_flush_ctx(struct inode *inode)
2136 struct ll_sb_info *sbi = ll_i2sbi(inode);
2138 CDEBUG(D_SEC, "flush context for user %d\n",
2139 from_kuid(&init_user_ns, current_uid()));
2141 obd_set_info_async(NULL, sbi->ll_md_exp,
2142 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2144 obd_set_info_async(NULL, sbi->ll_dt_exp,
2145 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2150 /* umount -f client means force down, don't save state */
2151 void ll_umount_begin(struct super_block *sb)
2153 struct ll_sb_info *sbi = ll_s2sbi(sb);
2154 struct obd_device *obd;
2155 struct obd_ioctl_data *ioc_data;
2156 struct l_wait_info lwi;
2157 wait_queue_head_t waitq;
2160 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2161 sb->s_count, atomic_read(&sb->s_active));
2163 obd = class_exp2obd(sbi->ll_md_exp);
2165 CERROR("Invalid MDC connection handle %#llx\n",
2166 sbi->ll_md_exp->exp_handle.h_cookie);
2172 obd = class_exp2obd(sbi->ll_dt_exp);
2174 CERROR("Invalid LOV connection handle %#llx\n",
2175 sbi->ll_dt_exp->exp_handle.h_cookie);
2181 OBD_ALLOC_PTR(ioc_data);
2183 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2184 sizeof *ioc_data, ioc_data, NULL);
2186 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2187 sizeof *ioc_data, ioc_data, NULL);
2189 OBD_FREE_PTR(ioc_data);
2192 /* Really, we'd like to wait until there are no requests outstanding,
2193 * and then continue. For now, we just periodically checking for vfs
2194 * to decrement mnt_cnt and hope to finish it within 10sec.
2196 init_waitqueue_head(&waitq);
2197 lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(10),
2198 cfs_time_seconds(1), NULL, NULL);
2199 l_wait_event(waitq, may_umount(sbi->ll_mnt.mnt), &lwi);
2204 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2206 struct ll_sb_info *sbi = ll_s2sbi(sb);
2207 char *profilenm = get_profile_name(sb);
2211 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2212 read_only = *flags & MS_RDONLY;
2213 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2214 sizeof(KEY_READ_ONLY),
2215 KEY_READ_ONLY, sizeof(read_only),
2218 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2219 profilenm, read_only ?
2220 "read-only" : "read-write", err);
2225 sb->s_flags |= MS_RDONLY;
2227 sb->s_flags &= ~MS_RDONLY;
2229 if (sbi->ll_flags & LL_SBI_VERBOSE)
2230 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2231 read_only ? "read-only" : "read-write");
2237 * Cleanup the open handle that is cached on MDT-side.
2239 * For open case, the client side open handling thread may hit error
2240 * after the MDT grant the open. Under such case, the client should
2241 * send close RPC to the MDT as cleanup; otherwise, the open handle
2242 * on the MDT will be leaked there until the client umount or evicted.
2244 * In further, if someone unlinked the file, because the open handle
2245 * holds the reference on such file/object, then it will block the
2246 * subsequent threads that want to locate such object via FID.
2248 * \param[in] sb super block for this file-system
2249 * \param[in] open_req pointer to the original open request
2251 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2253 struct mdt_body *body;
2254 struct md_op_data *op_data;
2255 struct ptlrpc_request *close_req = NULL;
2256 struct obd_export *exp = ll_s2sbi(sb)->ll_md_exp;
2259 body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2260 OBD_ALLOC_PTR(op_data);
2261 if (op_data == NULL) {
2262 CWARN("%s: cannot allocate op_data to release open handle for "
2264 ll_get_fsname(sb, NULL, 0), PFID(&body->mbo_fid1));
2269 op_data->op_fid1 = body->mbo_fid1;
2270 op_data->op_handle = body->mbo_handle;
2271 op_data->op_mod_time = ktime_get_real_seconds();
2272 md_close(exp, op_data, NULL, &close_req);
2273 ptlrpc_req_finished(close_req);
2274 ll_finish_md_op_data(op_data);
2279 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2280 struct super_block *sb, struct lookup_intent *it)
2282 struct ll_sb_info *sbi = NULL;
2283 struct lustre_md md = { NULL };
2287 LASSERT(*inode || sb);
2288 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2289 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2290 sbi->ll_md_exp, &md);
2295 rc = ll_update_inode(*inode, &md);
2299 LASSERT(sb != NULL);
2302 * At this point server returns to client's same fid as client
2303 * generated for creating. So using ->fid1 is okay here.
2305 if (!fid_is_sane(&md.body->mbo_fid1)) {
2306 CERROR("%s: Fid is insane "DFID"\n",
2307 ll_get_fsname(sb, NULL, 0),
2308 PFID(&md.body->mbo_fid1));
2309 GOTO(out, rc = -EINVAL);
2312 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2313 sbi->ll_flags & LL_SBI_32BIT_API),
2315 if (IS_ERR(*inode)) {
2316 #ifdef CONFIG_FS_POSIX_ACL
2318 posix_acl_release(md.posix_acl);
2319 md.posix_acl = NULL;
2322 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2324 CERROR("new_inode -fatal: rc %d\n", rc);
2329 /* Handling piggyback layout lock.
2330 * Layout lock can be piggybacked by getattr and open request.
2331 * The lsm can be applied to inode only if it comes with a layout lock
2332 * otherwise correct layout may be overwritten, for example:
2333 * 1. proc1: mdt returns a lsm but not granting layout
2334 * 2. layout was changed by another client
2335 * 3. proc2: refresh layout and layout lock granted
2336 * 4. proc1: to apply a stale layout */
2337 if (it != NULL && it->it_lock_mode != 0) {
2338 struct lustre_handle lockh;
2339 struct ldlm_lock *lock;
2341 lockh.cookie = it->it_lock_handle;
2342 lock = ldlm_handle2lock(&lockh);
2343 LASSERT(lock != NULL);
2344 if (ldlm_has_layout(lock)) {
2345 struct cl_object_conf conf;
2347 memset(&conf, 0, sizeof(conf));
2348 conf.coc_opc = OBJECT_CONF_SET;
2349 conf.coc_inode = *inode;
2350 conf.coc_lock = lock;
2351 conf.u.coc_layout = md.layout;
2352 (void)ll_layout_conf(*inode, &conf);
2354 LDLM_LOCK_PUT(lock);
2360 md_free_lustre_md(sbi->ll_md_exp, &md);
2363 if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2364 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2369 int ll_obd_statfs(struct inode *inode, void __user *arg)
2371 struct ll_sb_info *sbi = NULL;
2372 struct obd_export *exp;
2374 struct obd_ioctl_data *data = NULL;
2378 if (!inode || !(sbi = ll_i2sbi(inode)))
2379 GOTO(out_statfs, rc = -EINVAL);
2381 rc = obd_ioctl_getdata(&buf, &len, arg);
2383 GOTO(out_statfs, rc);
2386 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2387 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2388 GOTO(out_statfs, rc = -EINVAL);
2390 if (data->ioc_inllen1 != sizeof(__u32) ||
2391 data->ioc_inllen2 != sizeof(__u32) ||
2392 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2393 data->ioc_plen2 != sizeof(struct obd_uuid))
2394 GOTO(out_statfs, rc = -EINVAL);
2396 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2397 if (type & LL_STATFS_LMV)
2398 exp = sbi->ll_md_exp;
2399 else if (type & LL_STATFS_LOV)
2400 exp = sbi->ll_dt_exp;
2402 GOTO(out_statfs, rc = -ENODEV);
2404 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2406 GOTO(out_statfs, rc);
2408 OBD_FREE_LARGE(buf, len);
2412 int ll_process_config(struct lustre_cfg *lcfg)
2414 struct super_block *sb;
2419 /* The instance name contains the sb: lustre-client-aacfe000 */
2420 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2421 if (!ptr || !*(++ptr))
2423 if (sscanf(ptr, "%lx", &x) != 1)
2425 sb = (struct super_block *)x;
2426 /* This better be a real Lustre superblock! */
2427 LASSERT(s2lsi(sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2429 /* Note we have not called client_common_fill_super yet, so
2430 proc fns must be able to handle that! */
2431 rc = class_process_proc_param(PARAM_LLITE, lprocfs_llite_obd_vars,
2438 /* this function prepares md_op_data hint for passing it down to MD stack. */
2439 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2440 struct inode *i1, struct inode *i2,
2441 const char *name, size_t namelen,
2442 __u32 mode, __u32 opc, void *data)
2444 LASSERT(i1 != NULL);
2447 /* Do not reuse namelen for something else. */
2449 return ERR_PTR(-EINVAL);
2451 if (namelen > ll_i2sbi(i1)->ll_namelen)
2452 return ERR_PTR(-ENAMETOOLONG);
2454 if (!lu_name_is_valid_2(name, namelen))
2455 return ERR_PTR(-EINVAL);
2458 if (op_data == NULL)
2459 OBD_ALLOC_PTR(op_data);
2461 if (op_data == NULL)
2462 return ERR_PTR(-ENOMEM);
2464 ll_i2gids(op_data->op_suppgids, i1, i2);
2465 op_data->op_fid1 = *ll_inode2fid(i1);
2466 op_data->op_default_stripe_offset = -1;
2467 if (S_ISDIR(i1->i_mode)) {
2468 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2469 if (opc == LUSTRE_OPC_MKDIR)
2470 op_data->op_default_stripe_offset =
2471 ll_i2info(i1)->lli_def_stripe_offset;
2475 op_data->op_fid2 = *ll_inode2fid(i2);
2476 if (S_ISDIR(i2->i_mode))
2477 op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2479 fid_zero(&op_data->op_fid2);
2482 if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2483 op_data->op_cli_flags |= CLI_HASH64;
2485 if (ll_need_32bit_api(ll_i2sbi(i1)))
2486 op_data->op_cli_flags |= CLI_API32;
2488 op_data->op_name = name;
2489 op_data->op_namelen = namelen;
2490 op_data->op_mode = mode;
2491 op_data->op_mod_time = cfs_time_current_sec();
2492 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2493 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2494 op_data->op_cap = cfs_curproc_cap_pack();
2495 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2496 filename_is_volatile(name, namelen, &op_data->op_mds)) {
2497 op_data->op_bias |= MDS_CREATE_VOLATILE;
2499 op_data->op_mds = 0;
2501 op_data->op_data = data;
2506 void ll_finish_md_op_data(struct md_op_data *op_data)
2508 security_release_secctx(op_data->op_file_secctx,
2509 op_data->op_file_secctx_size);
2510 OBD_FREE_PTR(op_data);
2513 #ifdef HAVE_SUPEROPS_USE_DENTRY
2514 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2516 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2519 struct ll_sb_info *sbi;
2521 #ifdef HAVE_SUPEROPS_USE_DENTRY
2522 LASSERT((seq != NULL) && (dentry != NULL));
2523 sbi = ll_s2sbi(dentry->d_sb);
2525 LASSERT((seq != NULL) && (vfs != NULL));
2526 sbi = ll_s2sbi(vfs->mnt_sb);
2529 if (sbi->ll_flags & LL_SBI_NOLCK)
2530 seq_puts(seq, ",nolock");
2532 if (sbi->ll_flags & LL_SBI_FLOCK)
2533 seq_puts(seq, ",flock");
2535 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2536 seq_puts(seq, ",localflock");
2538 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2539 seq_puts(seq, ",user_xattr");
2541 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2542 seq_puts(seq, ",lazystatfs");
2544 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2545 seq_puts(seq, ",user_fid2path");
2547 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2548 seq_puts(seq, ",always_ping");
2554 * Get obd name by cmd, and copy out to user space
2556 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2558 struct ll_sb_info *sbi = ll_i2sbi(inode);
2559 struct obd_device *obd;
2562 if (cmd == OBD_IOC_GETDTNAME)
2563 obd = class_exp2obd(sbi->ll_dt_exp);
2564 else if (cmd == OBD_IOC_GETMDNAME)
2565 obd = class_exp2obd(sbi->ll_md_exp);
2572 if (copy_to_user((void __user *)arg, obd->obd_name,
2573 strlen(obd->obd_name) + 1))
2580 * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2581 * fsname will be returned in this buffer; otherwise, a static buffer will be
2582 * used to store the fsname and returned to caller.
2584 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2586 static char fsname_static[MTI_NAME_MAXLEN];
2587 struct lustre_sb_info *lsi = s2lsi(sb);
2592 /* this means the caller wants to use static buffer
2593 * and it doesn't care about race. Usually this is
2594 * in error reporting path */
2595 buf = fsname_static;
2596 buflen = sizeof(fsname_static);
2599 len = strlen(lsi->lsi_lmd->lmd_profile);
2600 ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2601 if (ptr && (strcmp(ptr, "-client") == 0))
2604 if (unlikely(len >= buflen))
2606 strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2612 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2619 p.mnt = current->fs->root.mnt;
2621 path = d_path(&p, buf, bufsize);
2626 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2628 char *buf, *path = NULL;
2629 struct dentry *dentry = NULL;
2630 struct inode *inode = page->mapping->host;
2632 /* this can be called inside spin lock so use GFP_ATOMIC. */
2633 buf = (char *)__get_free_page(GFP_ATOMIC);
2635 dentry = d_find_alias(page->mapping->host);
2637 path = ll_d_path(dentry, buf, PAGE_SIZE);
2641 "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2642 "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2643 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2644 PFID(ll_inode2fid(inode)),
2645 (path && !IS_ERR(path)) ? path : "", ioret);
2651 free_page((unsigned long)buf);
2654 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2655 struct lov_user_md **kbuf)
2657 struct lov_user_md lum;
2661 if (copy_from_user(&lum, md, sizeof(lum)))
2664 lum_size = ll_lov_user_md_size(&lum);
2668 OBD_ALLOC(*kbuf, lum_size);
2672 if (copy_from_user(*kbuf, md, lum_size) != 0) {
2673 OBD_FREE(*kbuf, lum_size);
2681 * Compute llite root squash state after a change of root squash
2682 * configuration setting or add/remove of a lnet nid
2684 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2686 struct root_squash_info *squash = &sbi->ll_squash;
2689 struct lnet_process_id id;
2691 /* Update norootsquash flag */
2692 down_write(&squash->rsi_sem);
2693 if (list_empty(&squash->rsi_nosquash_nids))
2694 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2696 /* Do not apply root squash as soon as one of our NIDs is
2697 * in the nosquash_nids list */
2700 while (LNetGetId(i++, &id) != -ENOENT) {
2701 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2703 if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2709 sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2711 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2713 up_write(&squash->rsi_sem);
2717 * Parse linkea content to extract information about a given hardlink
2719 * \param[in] ldata - Initialized linkea data
2720 * \param[in] linkno - Link identifier
2721 * \param[out] parent_fid - The entry's parent FID
2722 * \param[out] ln - Entry name destination buffer
2724 * \retval 0 on success
2725 * \retval Appropriate negative error code on failure
2727 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2728 struct lu_fid *parent_fid, struct lu_name *ln)
2734 rc = linkea_init_with_rec(ldata);
2738 if (linkno >= ldata->ld_leh->leh_reccount)
2739 /* beyond last link */
2742 linkea_first_entry(ldata);
2743 for (idx = 0; ldata->ld_lee != NULL; idx++) {
2744 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2749 linkea_next_entry(ldata);
2759 * Get parent FID and name of an identified link. Operation is performed for
2760 * a given link number, letting the caller iterate over linkno to list one or
2761 * all links of an entry.
2763 * \param[in] file - File descriptor against which to perform the operation
2764 * \param[in,out] arg - User-filled structure containing the linkno to operate
2765 * on and the available size. It is eventually filled with
2766 * the requested information or left untouched on error
2768 * \retval - 0 on success
2769 * \retval - Appropriate negative error code on failure
2771 int ll_getparent(struct file *file, struct getparent __user *arg)
2773 struct inode *inode = file_inode(file);
2774 struct linkea_data *ldata;
2775 struct lu_buf buf = LU_BUF_NULL;
2777 struct lu_fid parent_fid;
2784 if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2785 !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2788 if (get_user(name_size, &arg->gp_name_size))
2791 if (get_user(linkno, &arg->gp_linkno))
2794 if (name_size > PATH_MAX)
2797 OBD_ALLOC(ldata, sizeof(*ldata));
2801 rc = linkea_data_new(ldata, &buf);
2803 GOTO(ldata_free, rc);
2805 #ifdef HAVE_XATTR_HANDLER_FLAGS
2806 rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
2807 buf.lb_len, OBD_MD_FLXATTR);
2809 rc = ll_getxattr(file_dentry(file), XATTR_NAME_LINK, buf.lb_buf,
2811 #endif /* HAVE_XATTR_HANDLER_FLAGS */
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));