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_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
204 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
205 OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 |
206 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
207 OBD_CONNECT_64BITHASH |
208 OBD_CONNECT_EINPROGRESS |
209 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
210 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS|
211 OBD_CONNECT_MAX_EASIZE |
212 OBD_CONNECT_FLOCK_DEAD |
213 OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
214 OBD_CONNECT_OPEN_BY_FID |
215 OBD_CONNECT_DIR_STRIPE |
216 OBD_CONNECT_BULK_MBITS | OBD_CONNECT_CKSUM |
217 OBD_CONNECT_SUBTREE |
218 OBD_CONNECT_FLAGS2 | OBD_CONNECT_MULTIMODRPCS;
220 data->ocd_connect_flags2 = 0;
222 #ifdef HAVE_LRU_RESIZE_SUPPORT
223 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
224 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
226 #ifdef CONFIG_FS_POSIX_ACL
227 data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK |
228 OBD_CONNECT_LARGE_ACL;
231 data->ocd_cksum_types = cksum_types_supported_client();
233 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
234 /* flag mdc connection as lightweight, only used for test
235 * purpose, use with care */
236 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
238 data->ocd_ibits_known = MDS_INODELOCK_FULL;
239 data->ocd_version = LUSTRE_VERSION_CODE;
241 if (sb->s_flags & MS_RDONLY)
242 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
243 if (sbi->ll_flags & LL_SBI_USER_XATTR)
244 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
247 /* Setting this indicates we correctly support S_NOSEC (See kernel
248 * commit 9e1f1de02c2275d7172e18dc4e7c2065777611bf)
250 sb->s_flags |= MS_NOSEC;
253 if (sbi->ll_flags & LL_SBI_FLOCK)
254 sbi->ll_fop = &ll_file_operations_flock;
255 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
256 sbi->ll_fop = &ll_file_operations;
258 sbi->ll_fop = &ll_file_operations_noflock;
260 /* always ping even if server suppress_pings */
261 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
262 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
264 #ifdef HAVE_SECURITY_DENTRY_INIT_SECURITY
265 data->ocd_connect_flags2 |= OBD_CONNECT2_FILE_SECCTX;
266 #endif /* HAVE_SECURITY_DENTRY_INIT_SECURITY */
268 data->ocd_brw_size = MD_MAX_BRW_SIZE;
270 err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
272 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
273 "recovery, of which this client is not a "
274 "part. Please wait for recovery to complete,"
275 " abort, or time out.\n", md);
278 CERROR("cannot connect to %s: rc = %d\n", md, err);
282 sbi->ll_md_exp->exp_connect_data = *data;
284 err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
285 LUSTRE_SEQ_METADATA);
287 CERROR("%s: Can't init metadata layer FID infrastructure, "
288 "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
292 /* For mount, we only need fs info from MDT0, and also in DNE, it
293 * can make sure the client can be mounted as long as MDT0 is
295 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
296 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
297 OBD_STATFS_FOR_MDT0);
299 GOTO(out_md_fid, err);
301 /* This needs to be after statfs to ensure connect has finished.
302 * Note that "data" does NOT contain the valid connect reply.
303 * If connecting to a 1.8 server there will be no LMV device, so
304 * we can access the MDC export directly and exp_connect_flags will
305 * be non-zero, but if accessing an upgraded 2.1 server it will
306 * have the correct flags filled in.
307 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
308 valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
309 if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
310 valid != CLIENT_CONNECT_MDT_REQD) {
313 OBD_ALLOC_WAIT(buf, PAGE_SIZE);
314 obd_connect_flags2str(buf, PAGE_SIZE,
315 valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
316 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
317 "feature(s) needed for correct operation "
318 "of this client (%s). Please upgrade "
319 "server or downgrade client.\n",
320 sbi->ll_md_exp->exp_obd->obd_name, buf);
321 OBD_FREE(buf, PAGE_SIZE);
322 GOTO(out_md_fid, err = -EPROTO);
325 size = sizeof(*data);
326 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
327 KEY_CONN_DATA, &size, data);
329 CERROR("%s: Get connect data failed: rc = %d\n",
330 sbi->ll_md_exp->exp_obd->obd_name, err);
331 GOTO(out_md_fid, err);
334 LASSERT(osfs->os_bsize);
335 sb->s_blocksize = osfs->os_bsize;
336 sb->s_blocksize_bits = log2(osfs->os_bsize);
337 sb->s_magic = LL_SUPER_MAGIC;
338 sb->s_maxbytes = MAX_LFS_FILESIZE;
339 sbi->ll_namelen = osfs->os_namelen;
340 sbi->ll_mnt.mnt = current->fs->root.mnt;
342 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
343 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
344 LCONSOLE_INFO("Disabling user_xattr feature because "
345 "it is not supported on the server\n");
346 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
349 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
351 sb->s_flags |= MS_POSIXACL;
353 sbi->ll_flags |= LL_SBI_ACL;
355 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
357 sb->s_flags &= ~MS_POSIXACL;
359 sbi->ll_flags &= ~LL_SBI_ACL;
362 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
363 sbi->ll_flags |= LL_SBI_64BIT_HASH;
365 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
366 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
368 if (obd_connect_has_secctx(data))
369 sbi->ll_flags |= LL_SBI_FILE_SECCTX;
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);
375 } else if (!sbi->ll_xattr_cache_set) {
376 /* If xattr_cache is already set (no matter 0 or 1)
377 * during processing llog, it won't be enabled here. */
378 sbi->ll_flags |= LL_SBI_XATTR_CACHE;
379 sbi->ll_xattr_cache_enabled = 1;
383 obd = class_name2obd(dt);
385 CERROR("DT %s: not setup or attached\n", dt);
386 GOTO(out_md_fid, err = -ENODEV);
389 /* pass client page size via ocd_grant_blkbits, the server should report
390 * back its backend blocksize for grant calculation purpose */
391 data->ocd_grant_blkbits = PAGE_SHIFT;
393 /* indicate OST features supported by this client */
394 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
395 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
396 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
397 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
398 OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
399 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
400 OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
401 OBD_CONNECT_EINPROGRESS |
402 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
403 OBD_CONNECT_LAYOUTLOCK |
404 OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
405 OBD_CONNECT_BULK_MBITS |
408 /* The client currently advertises support for OBD_CONNECT_LOCKAHEAD_OLD so it
409 * can interoperate with an older version of lockahead which was released prior
410 * to landing in master. This support will be dropped when 2.13 development
411 * starts. At the point, we should not just drop the connect flag (below), we
412 * should also remove the support in the code.
414 * Removing it means a few things:
415 * 1. Remove this section here
416 * 2. Remove CEF_NONBLOCK in ll_file_lockahead()
417 * 3. Remove function exp_connect_lockahead_old
418 * 4. Remove LDLM_FL_LOCKAHEAD_OLD_RESERVED in lustre_dlm_flags.h
420 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 12, 50, 0)
421 data->ocd_connect_flags |= OBD_CONNECT_LOCKAHEAD_OLD;
424 data->ocd_connect_flags2 = OBD_CONNECT2_LOCKAHEAD;
426 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
427 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
429 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
430 * disabled by default, because it can still be enabled on the
431 * fly via /proc. As a consequence, we still need to come to an
432 * agreement on the supported algorithms at connect time */
433 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
435 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
436 data->ocd_cksum_types = OBD_CKSUM_ADLER;
438 data->ocd_cksum_types = cksum_types_supported_client();
440 #ifdef HAVE_LRU_RESIZE_SUPPORT
441 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
443 /* always ping even if server suppress_pings */
444 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
445 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
447 CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
448 "ocd_grant: %d\n", data->ocd_connect_flags,
449 data->ocd_version, data->ocd_grant);
451 obd->obd_upcall.onu_owner = &sbi->ll_lco;
452 obd->obd_upcall.onu_upcall = cl_ocd_update;
454 data->ocd_brw_size = DT_MAX_BRW_SIZE;
456 err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
459 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
460 "recovery, of which this client is not a "
461 "part. Please wait for recovery to "
462 "complete, abort, or time out.\n", dt);
465 CERROR("%s: Cannot connect to %s: rc = %d\n",
466 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
470 sbi->ll_dt_exp->exp_connect_data = *data;
472 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
473 LUSTRE_SEQ_METADATA);
475 CERROR("%s: Can't init data layer FID infrastructure, "
476 "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
480 mutex_lock(&sbi->ll_lco.lco_lock);
481 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
482 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
483 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
484 mutex_unlock(&sbi->ll_lco.lco_lock);
486 fid_zero(&sbi->ll_root_fid);
487 err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
490 CERROR("cannot mds_connect: rc = %d\n", err);
491 GOTO(out_lock_cn_cb, err);
493 if (!fid_is_sane(&sbi->ll_root_fid)) {
494 CERROR("%s: Invalid root fid "DFID" during mount\n",
495 sbi->ll_md_exp->exp_obd->obd_name,
496 PFID(&sbi->ll_root_fid));
497 GOTO(out_lock_cn_cb, err = -EINVAL);
499 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
501 sb->s_op = &lustre_super_operations;
502 #ifdef HAVE_XATTR_HANDLER_FLAGS
503 sb->s_xattr = ll_xattr_handlers;
505 #if THREAD_SIZE >= 8192 /*b=17630*/
506 sb->s_export_op = &lustre_export_operations;
510 * XXX: move this to after cbd setup? */
511 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
512 if (sbi->ll_flags & LL_SBI_ACL)
513 valid |= OBD_MD_FLACL;
515 OBD_ALLOC_PTR(op_data);
517 GOTO(out_lock_cn_cb, err = -ENOMEM);
519 op_data->op_fid1 = sbi->ll_root_fid;
520 op_data->op_mode = 0;
521 op_data->op_valid = valid;
523 err = md_getattr(sbi->ll_md_exp, op_data, &request);
525 OBD_FREE_PTR(op_data);
527 CERROR("%s: md_getattr failed for root: rc = %d\n",
528 sbi->ll_md_exp->exp_obd->obd_name, err);
529 GOTO(out_lock_cn_cb, err);
532 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
533 sbi->ll_md_exp, &lmd);
535 CERROR("failed to understand root inode md: rc = %d\n", err);
536 ptlrpc_req_finished(request);
537 GOTO(out_lock_cn_cb, err);
540 LASSERT(fid_is_sane(&sbi->ll_root_fid));
541 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
542 sbi->ll_flags & LL_SBI_32BIT_API),
544 md_free_lustre_md(sbi->ll_md_exp, &lmd);
545 ptlrpc_req_finished(request);
548 #ifdef CONFIG_FS_POSIX_ACL
550 posix_acl_release(lmd.posix_acl);
551 lmd.posix_acl = NULL;
554 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
556 CERROR("lustre_lite: bad iget4 for root\n");
560 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
561 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
562 KEY_CHECKSUM, sizeof(checksum), &checksum,
565 CERROR("%s: Set checksum failed: rc = %d\n",
566 sbi->ll_dt_exp->exp_obd->obd_name, err);
571 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
572 KEY_CACHE_SET, sizeof(*sbi->ll_cache),
573 sbi->ll_cache, NULL);
575 CERROR("%s: Set cache_set failed: rc = %d\n",
576 sbi->ll_dt_exp->exp_obd->obd_name, err);
580 sb->s_root = d_make_root(root);
581 if (sb->s_root == NULL) {
582 CERROR("%s: can't make root dentry\n",
583 ll_get_fsname(sb, NULL, 0));
584 GOTO(out_root, err = -ENOMEM);
586 #ifdef HAVE_DCACHE_LOCK
587 sb->s_root->d_op = &ll_d_ops;
590 sbi->ll_sdev_orig = sb->s_dev;
592 /* We set sb->s_dev equal on all lustre clients in order to support
593 * NFS export clustering. NFSD requires that the FSID be the same
595 /* s_dev is also used in lt_compare() to compare two fs, but that is
596 * only a node-local comparison. */
597 uuid = obd_get_uuid(sbi->ll_md_exp);
599 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
606 if (sbi->ll_proc_root != NULL) {
607 err = lprocfs_ll_register_obd(sb, dt);
609 CERROR("%s: could not register %s in llite: rc = %d\n",
610 dt, ll_get_fsname(sb, NULL, 0), err);
613 err = lprocfs_ll_register_obd(sb, md);
615 CERROR("%s: could not register %s in llite: rc = %d\n",
616 md, ll_get_fsname(sb, NULL, 0), err);
626 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
628 obd_disconnect(sbi->ll_dt_exp);
629 sbi->ll_dt_exp = NULL;
631 obd_fid_fini(sbi->ll_md_exp->exp_obd);
633 obd_disconnect(sbi->ll_md_exp);
634 sbi->ll_md_exp = NULL;
643 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
647 size = sizeof(*lmmsize);
648 rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
649 KEY_MAX_EASIZE, &size, lmmsize);
651 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
652 sbi->ll_dt_exp->exp_obd->obd_name, rc);
657 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
658 KEY_MAX_EASIZE, &size, lmmsize);
660 CERROR("Get max mdsize error rc %d\n", rc);
666 * Get the value of the default_easize parameter.
668 * \see client_obd::cl_default_mds_easize
670 * \param[in] sbi superblock info for this filesystem
671 * \param[out] lmmsize pointer to storage location for value
673 * \retval 0 on success
674 * \retval negative negated errno on failure
676 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
681 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
682 KEY_DEFAULT_EASIZE, &size, lmmsize);
684 CERROR("Get default mdsize error rc %d\n", rc);
690 * Set the default_easize parameter to the given value.
692 * \see client_obd::cl_default_mds_easize
694 * \param[in] sbi superblock info for this filesystem
695 * \param[in] lmmsize the size to set
697 * \retval 0 on success
698 * \retval negative negated errno on failure
700 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
704 if (lmmsize < sizeof(struct lov_mds_md) ||
705 lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
708 rc = obd_set_info_async(NULL, sbi->ll_md_exp,
709 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
710 sizeof(int), &lmmsize, NULL);
715 static void client_common_put_super(struct super_block *sb)
717 struct ll_sb_info *sbi = ll_s2sbi(sb);
722 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
723 obd_disconnect(sbi->ll_dt_exp);
724 sbi->ll_dt_exp = NULL;
726 lprocfs_ll_unregister_mountpoint(sbi);
728 obd_fid_fini(sbi->ll_md_exp->exp_obd);
729 obd_disconnect(sbi->ll_md_exp);
730 sbi->ll_md_exp = NULL;
735 void ll_kill_super(struct super_block *sb)
737 struct ll_sb_info *sbi;
741 if (!(sb->s_flags & MS_ACTIVE))
745 /* we need restore s_dev from changed for clustred NFS before put_super
746 * because new kernels have cached s_dev and change sb->s_dev in
747 * put_super not affected real removing devices */
749 sb->s_dev = sbi->ll_sdev_orig;
750 sbi->ll_umounting = 1;
752 /* wait running statahead threads to quit */
753 while (atomic_read(&sbi->ll_sa_running) > 0) {
754 set_current_state(TASK_UNINTERRUPTIBLE);
755 schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
762 static inline int ll_set_opt(const char *opt, char *data, int fl)
764 if (strncmp(opt, data, strlen(opt)) != 0)
770 /* non-client-specific mount options are parsed in lmd_parse */
771 static int ll_options(char *options, int *flags)
774 char *s1 = options, *s2;
780 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
783 CDEBUG(D_SUPER, "next opt=%s\n", s1);
784 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
789 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
794 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
799 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
804 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
809 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
814 tmp = ll_set_opt("context", s1, 1);
817 tmp = ll_set_opt("fscontext", s1, 1);
820 tmp = ll_set_opt("defcontext", s1, 1);
823 tmp = ll_set_opt("rootcontext", s1, 1);
826 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
831 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
837 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
842 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
847 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
852 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
857 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
862 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
867 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
872 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
877 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
882 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
887 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
893 s2 = strchr(s1, ',');
901 void ll_lli_init(struct ll_inode_info *lli)
903 lli->lli_inode_magic = LLI_INODE_MAGIC;
905 spin_lock_init(&lli->lli_lock);
906 lli->lli_posix_acl = NULL;
907 /* Do not set lli_fid, it has been initialized already. */
908 fid_zero(&lli->lli_pfid);
909 lli->lli_mds_read_och = NULL;
910 lli->lli_mds_write_och = NULL;
911 lli->lli_mds_exec_och = NULL;
912 lli->lli_open_fd_read_count = 0;
913 lli->lli_open_fd_write_count = 0;
914 lli->lli_open_fd_exec_count = 0;
915 mutex_init(&lli->lli_och_mutex);
916 spin_lock_init(&lli->lli_agl_lock);
917 spin_lock_init(&lli->lli_layout_lock);
918 ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
919 lli->lli_clob = NULL;
921 init_rwsem(&lli->lli_xattrs_list_rwsem);
922 mutex_init(&lli->lli_xattrs_enq_lock);
924 LASSERT(lli->lli_vfs_inode.i_mode != 0);
925 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
926 mutex_init(&lli->lli_readdir_mutex);
927 lli->lli_opendir_key = NULL;
929 spin_lock_init(&lli->lli_sa_lock);
930 lli->lli_opendir_pid = 0;
931 lli->lli_sa_enabled = 0;
932 lli->lli_def_stripe_offset = -1;
934 mutex_init(&lli->lli_size_mutex);
935 lli->lli_symlink_name = NULL;
936 init_rwsem(&lli->lli_trunc_sem);
937 range_lock_tree_init(&lli->lli_write_tree);
938 init_rwsem(&lli->lli_glimpse_sem);
939 lli->lli_glimpse_time = 0;
940 INIT_LIST_HEAD(&lli->lli_agl_list);
941 lli->lli_agl_index = 0;
942 lli->lli_async_rc = 0;
944 mutex_init(&lli->lli_layout_mutex);
945 memset(lli->lli_jobid, 0, LUSTRE_JOBID_SIZE);
948 static inline int ll_bdi_register(struct backing_dev_info *bdi)
950 static atomic_t ll_bdi_num = ATOMIC_INIT(0);
952 bdi->name = "lustre";
953 return bdi_register(bdi, NULL, "lustre-%d",
954 atomic_inc_return(&ll_bdi_num));
957 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
959 struct lustre_profile *lprof = NULL;
960 struct lustre_sb_info *lsi = s2lsi(sb);
961 struct ll_sb_info *sbi;
962 char *dt = NULL, *md = NULL;
963 char *profilenm = get_profile_name(sb);
964 struct config_llog_instance *cfg;
965 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
966 const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
972 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
978 try_module_get(THIS_MODULE);
980 /* client additional sb info */
981 lsi->lsi_llsbi = sbi = ll_init_sbi();
983 module_put(THIS_MODULE);
988 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
992 err = bdi_init(&lsi->lsi_bdi);
995 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
996 #ifdef HAVE_BDI_CAP_MAP_COPY
997 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
999 lsi->lsi_bdi.capabilities = 0;
1001 err = ll_bdi_register(&lsi->lsi_bdi);
1003 GOTO(out_free, err);
1005 sb->s_bdi = &lsi->lsi_bdi;
1006 #ifndef HAVE_DCACHE_LOCK
1007 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1008 sb->s_d_op = &ll_d_ops;
1011 /* Call lprocfs_ll_register_mountpoint() before lustre_process_log()
1012 * so that "llite.*.*" params can be processed correctly. */
1013 if (proc_lustre_fs_root != NULL) {
1014 err = lprocfs_ll_register_mountpoint(proc_lustre_fs_root, sb);
1016 CERROR("%s: could not register mountpoint in llite: "
1017 "rc = %d\n", ll_get_fsname(sb, NULL, 0), err);
1022 /* Generate a string unique to this super, in case some joker tries
1023 to mount the same fs at two mount points.
1024 Use the address of the super itself.*/
1025 cfg->cfg_instance = sb;
1026 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1027 cfg->cfg_callback = class_config_llog_handler;
1028 cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1029 /* set up client obds */
1030 err = lustre_process_log(sb, profilenm, cfg);
1032 GOTO(out_proc, err);
1034 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1035 lprof = class_get_profile(profilenm);
1036 if (lprof == NULL) {
1037 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1038 " read from the MGS. Does that filesystem "
1039 "exist?\n", profilenm);
1040 GOTO(out_proc, err = -EINVAL);
1042 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1043 lprof->lp_md, lprof->lp_dt);
1045 dt_len = strlen(lprof->lp_dt) + instlen + 2;
1046 OBD_ALLOC(dt, dt_len);
1048 GOTO(out_proc, err = -ENOMEM);
1049 snprintf(dt, dt_len - 1, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1051 md_len = strlen(lprof->lp_md) + instlen + 2;
1052 OBD_ALLOC(md, md_len);
1054 GOTO(out_proc, err = -ENOMEM);
1055 snprintf(md, md_len - 1, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1057 /* connections, registrations, sb setup */
1058 err = client_common_fill_super(sb, md, dt, mnt);
1060 GOTO(out_proc, err);
1062 sbi->ll_client_common_fill_super_succeeded = 1;
1066 lprocfs_ll_unregister_mountpoint(sbi);
1069 OBD_FREE(md, md_len);
1071 OBD_FREE(dt, dt_len);
1073 class_put_profile(lprof);
1076 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1077 LCONSOLE_WARN("Mounted %s\n", profilenm);
1081 } /* ll_fill_super */
1083 void ll_put_super(struct super_block *sb)
1085 struct config_llog_instance cfg, params_cfg;
1086 struct obd_device *obd;
1087 struct lustre_sb_info *lsi = s2lsi(sb);
1088 struct ll_sb_info *sbi = ll_s2sbi(sb);
1089 char *profilenm = get_profile_name(sb);
1091 int next, force = 1, rc = 0;
1094 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1096 cfg.cfg_instance = sb;
1097 lustre_end_log(sb, profilenm, &cfg);
1099 params_cfg.cfg_instance = sb;
1100 lustre_end_log(sb, PARAMS_FILENAME, ¶ms_cfg);
1102 if (sbi->ll_md_exp) {
1103 obd = class_exp2obd(sbi->ll_md_exp);
1105 force = obd->obd_force;
1108 /* Wait for unstable pages to be committed to stable storage */
1110 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1111 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1112 atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1116 ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1117 if (force == 0 && rc != -EINTR)
1118 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1121 /* We need to set force before the lov_disconnect in
1122 lustre_common_put_super, since l_d cleans up osc's as well. */
1125 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1127 obd->obd_force = force;
1131 if (sbi->ll_client_common_fill_super_succeeded) {
1132 /* Only if client_common_fill_super succeeded */
1133 client_common_put_super(sb);
1137 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1138 class_manual_cleanup(obd);
1141 if (sbi->ll_flags & LL_SBI_VERBOSE)
1142 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1145 class_del_profile(profilenm);
1147 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1148 bdi_destroy(&lsi->lsi_bdi);
1149 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1153 lsi->lsi_llsbi = NULL;
1155 lustre_common_put_super(sb);
1157 cl_env_cache_purge(~0);
1159 module_put(THIS_MODULE);
1162 } /* client_put_super */
1164 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1166 struct inode *inode = NULL;
1168 /* NOTE: we depend on atomic igrab() -bzzz */
1169 lock_res_and_lock(lock);
1170 if (lock->l_resource->lr_lvb_inode) {
1171 struct ll_inode_info * lli;
1172 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1173 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1174 inode = igrab(lock->l_resource->lr_lvb_inode);
1176 inode = lock->l_resource->lr_lvb_inode;
1177 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1178 D_WARNING, lock, "lr_lvb_inode %p is "
1179 "bogus: magic %08x",
1180 lock->l_resource->lr_lvb_inode,
1181 lli->lli_inode_magic);
1185 unlock_res_and_lock(lock);
1189 void ll_dir_clear_lsm_md(struct inode *inode)
1191 struct ll_inode_info *lli = ll_i2info(inode);
1193 LASSERT(S_ISDIR(inode->i_mode));
1195 if (lli->lli_lsm_md != NULL) {
1196 lmv_free_memmd(lli->lli_lsm_md);
1197 lli->lli_lsm_md = NULL;
1201 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1202 const struct lu_fid *fid,
1203 struct lustre_md *md)
1205 struct ll_sb_info *sbi = ll_s2sbi(sb);
1206 struct mdt_body *body = md->body;
1207 struct inode *inode;
1211 ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1212 inode = iget_locked(sb, ino);
1213 if (inode == NULL) {
1214 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1215 ll_get_fsname(sb, NULL, 0), PFID(fid));
1216 RETURN(ERR_PTR(-ENOENT));
1219 if (inode->i_state & I_NEW) {
1220 struct ll_inode_info *lli = ll_i2info(inode);
1221 struct lmv_stripe_md *lsm = md->lmv;
1223 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1224 (body->mbo_mode & S_IFMT);
1225 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1228 LTIME_S(inode->i_mtime) = 0;
1229 LTIME_S(inode->i_atime) = 0;
1230 LTIME_S(inode->i_ctime) = 0;
1233 #ifdef HAVE_BACKING_DEV_INFO
1234 /* initializing backing dev info. */
1235 inode->i_mapping->backing_dev_info =
1236 &s2lsi(inode->i_sb)->lsi_bdi;
1238 inode->i_op = &ll_dir_inode_operations;
1239 inode->i_fop = &ll_dir_operations;
1240 lli->lli_fid = *fid;
1243 LASSERT(lsm != NULL);
1244 /* master object FID */
1245 lli->lli_pfid = body->mbo_fid1;
1246 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1247 lli, PFID(fid), PFID(&lli->lli_pfid));
1248 unlock_new_inode(inode);
1254 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1257 struct lmv_stripe_md *lsm = md->lmv;
1260 LASSERT(lsm != NULL);
1261 /* XXX sigh, this lsm_root initialization should be in
1262 * LMV layer, but it needs ll_iget right now, so we
1263 * put this here right now. */
1264 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1265 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1266 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1267 /* Unfortunately ll_iget will call ll_update_inode,
1268 * where the initialization of slave inode is slightly
1269 * different, so it reset lsm_md to NULL to avoid
1270 * initializing lsm for slave inode. */
1271 /* For migrating inode, master stripe and master object will
1272 * be same, so we only need assign this inode */
1273 if (lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION && i == 0)
1274 lsm->lsm_md_oinfo[i].lmo_root = inode;
1276 lsm->lsm_md_oinfo[i].lmo_root =
1277 ll_iget_anon_dir(inode->i_sb, fid, md);
1279 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1280 int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1282 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1290 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1291 const struct lmv_stripe_md *lsm_md2)
1293 return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1294 lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1295 lsm_md1->lsm_md_master_mdt_index ==
1296 lsm_md2->lsm_md_master_mdt_index &&
1297 lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1298 lsm_md1->lsm_md_layout_version ==
1299 lsm_md2->lsm_md_layout_version &&
1300 strcmp(lsm_md1->lsm_md_pool_name,
1301 lsm_md2->lsm_md_pool_name) == 0;
1304 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1306 struct ll_inode_info *lli = ll_i2info(inode);
1307 struct lmv_stripe_md *lsm = md->lmv;
1311 LASSERT(S_ISDIR(inode->i_mode));
1312 CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1313 PFID(ll_inode2fid(inode)));
1315 /* no striped information from request. */
1317 if (lli->lli_lsm_md == NULL) {
1319 } else if (lli->lli_lsm_md->lsm_md_hash_type &
1320 LMV_HASH_FLAG_MIGRATION) {
1321 /* migration is done, the temporay MIGRATE layout has
1323 CDEBUG(D_INODE, DFID" finish migration.\n",
1324 PFID(ll_inode2fid(inode)));
1325 lmv_free_memmd(lli->lli_lsm_md);
1326 lli->lli_lsm_md = NULL;
1329 /* The lustre_md from req does not include stripeEA,
1330 * see ll_md_setattr */
1335 /* set the directory layout */
1336 if (lli->lli_lsm_md == NULL) {
1337 struct cl_attr *attr;
1339 rc = ll_init_lsm_md(inode, md);
1343 /* set md->lmv to NULL, so the following free lustre_md
1344 * will not free this lsm */
1346 lli->lli_lsm_md = lsm;
1348 OBD_ALLOC_PTR(attr);
1352 /* validate the lsm */
1353 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1354 ll_md_blocking_ast);
1360 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1361 md->body->mbo_nlink = attr->cat_nlink;
1362 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1363 md->body->mbo_size = attr->cat_size;
1364 if (md->body->mbo_valid & OBD_MD_FLATIME)
1365 md->body->mbo_atime = attr->cat_atime;
1366 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1367 md->body->mbo_ctime = attr->cat_ctime;
1368 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1369 md->body->mbo_mtime = attr->cat_mtime;
1373 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1374 lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1378 /* Compare the old and new stripe information */
1379 if (!lsm_md_eq(lli->lli_lsm_md, lsm)) {
1380 struct lmv_stripe_md *old_lsm = lli->lli_lsm_md;
1383 CERROR("%s: inode "DFID"(%p)'s lmv layout mismatch (%p)/(%p)"
1384 "magic:0x%x/0x%x stripe count: %d/%d master_mdt: %d/%d"
1385 "hash_type:0x%x/0x%x layout: 0x%x/0x%x pool:%s/%s\n",
1386 ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1387 inode, lsm, old_lsm,
1388 lsm->lsm_md_magic, old_lsm->lsm_md_magic,
1389 lsm->lsm_md_stripe_count,
1390 old_lsm->lsm_md_stripe_count,
1391 lsm->lsm_md_master_mdt_index,
1392 old_lsm->lsm_md_master_mdt_index,
1393 lsm->lsm_md_hash_type, old_lsm->lsm_md_hash_type,
1394 lsm->lsm_md_layout_version,
1395 old_lsm->lsm_md_layout_version,
1396 lsm->lsm_md_pool_name,
1397 old_lsm->lsm_md_pool_name);
1399 for (idx = 0; idx < old_lsm->lsm_md_stripe_count; idx++) {
1400 CERROR("%s: sub FIDs in old lsm idx %d, old: "DFID"\n",
1401 ll_get_fsname(inode->i_sb, NULL, 0), idx,
1402 PFID(&old_lsm->lsm_md_oinfo[idx].lmo_fid));
1405 for (idx = 0; idx < lsm->lsm_md_stripe_count; idx++) {
1406 CERROR("%s: sub FIDs in new lsm idx %d, new: "DFID"\n",
1407 ll_get_fsname(inode->i_sb, NULL, 0), idx,
1408 PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1417 void ll_clear_inode(struct inode *inode)
1419 struct ll_inode_info *lli = ll_i2info(inode);
1420 struct ll_sb_info *sbi = ll_i2sbi(inode);
1423 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1424 PFID(ll_inode2fid(inode)), inode);
1426 if (S_ISDIR(inode->i_mode)) {
1427 /* these should have been cleared in ll_file_release */
1428 LASSERT(lli->lli_opendir_key == NULL);
1429 LASSERT(lli->lli_sai == NULL);
1430 LASSERT(lli->lli_opendir_pid == 0);
1433 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1435 LASSERT(!lli->lli_open_fd_write_count);
1436 LASSERT(!lli->lli_open_fd_read_count);
1437 LASSERT(!lli->lli_open_fd_exec_count);
1439 if (lli->lli_mds_write_och)
1440 ll_md_real_close(inode, FMODE_WRITE);
1441 if (lli->lli_mds_exec_och)
1442 ll_md_real_close(inode, FMODE_EXEC);
1443 if (lli->lli_mds_read_och)
1444 ll_md_real_close(inode, FMODE_READ);
1446 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1447 OBD_FREE(lli->lli_symlink_name,
1448 strlen(lli->lli_symlink_name) + 1);
1449 lli->lli_symlink_name = NULL;
1452 ll_xattr_cache_destroy(inode);
1454 #ifdef CONFIG_FS_POSIX_ACL
1455 forget_all_cached_acls(inode);
1456 if (lli->lli_posix_acl) {
1457 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1458 posix_acl_release(lli->lli_posix_acl);
1459 lli->lli_posix_acl = NULL;
1462 lli->lli_inode_magic = LLI_INODE_DEAD;
1464 if (S_ISDIR(inode->i_mode))
1465 ll_dir_clear_lsm_md(inode);
1466 else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1467 LASSERT(list_empty(&lli->lli_agl_list));
1470 * XXX This has to be done before lsm is freed below, because
1471 * cl_object still uses inode lsm.
1473 cl_inode_fini(inode);
1478 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1480 struct lustre_md md;
1481 struct inode *inode = dentry->d_inode;
1482 struct ll_sb_info *sbi = ll_i2sbi(inode);
1483 struct ptlrpc_request *request = NULL;
1487 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1488 LUSTRE_OPC_ANY, NULL);
1489 if (IS_ERR(op_data))
1490 RETURN(PTR_ERR(op_data));
1492 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1494 ptlrpc_req_finished(request);
1495 if (rc == -ENOENT) {
1497 /* Unlinked special device node? Or just a race?
1498 * Pretend we done everything. */
1499 if (!S_ISREG(inode->i_mode) &&
1500 !S_ISDIR(inode->i_mode)) {
1501 ia_valid = op_data->op_attr.ia_valid;
1502 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1503 rc = simple_setattr(dentry, &op_data->op_attr);
1504 op_data->op_attr.ia_valid = ia_valid;
1506 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1507 CERROR("md_setattr fails: rc = %d\n", rc);
1512 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1513 sbi->ll_md_exp, &md);
1515 ptlrpc_req_finished(request);
1519 ia_valid = op_data->op_attr.ia_valid;
1520 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1521 * cache is not cleared yet. */
1522 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1523 if (S_ISREG(inode->i_mode))
1525 rc = simple_setattr(dentry, &op_data->op_attr);
1526 if (S_ISREG(inode->i_mode))
1527 inode_unlock(inode);
1528 op_data->op_attr.ia_valid = ia_valid;
1530 rc = ll_update_inode(inode, &md);
1531 ptlrpc_req_finished(request);
1536 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1537 * object(s) determine the file size and mtime. Otherwise, the MDS will
1538 * keep these values until such a time that objects are allocated for it.
1539 * We do the MDS operations first, as it is checking permissions for us.
1540 * We don't to the MDS RPC if there is nothing that we want to store there,
1541 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1542 * going to do an RPC anyways.
1544 * If we are doing a truncate, we will send the mtime and ctime updates
1545 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1546 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1549 * In case of HSMimport, we only set attr on MDS.
1551 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1553 struct inode *inode = dentry->d_inode;
1554 struct ll_inode_info *lli = ll_i2info(inode);
1555 struct md_op_data *op_data = NULL;
1559 CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1560 "valid %x, hsm_import %d\n",
1561 ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1562 inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1565 if (attr->ia_valid & ATTR_SIZE) {
1566 /* Check new size against VFS/VM file size limit and rlimit */
1567 rc = inode_newsize_ok(inode, attr->ia_size);
1571 /* The maximum Lustre file size is variable, based on the
1572 * OST maximum object size and number of stripes. This
1573 * needs another check in addition to the VFS check above. */
1574 if (attr->ia_size > ll_file_maxbytes(inode)) {
1575 CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
1576 PFID(&lli->lli_fid), attr->ia_size,
1577 ll_file_maxbytes(inode));
1581 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1584 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1585 if (attr->ia_valid & TIMES_SET_FLAGS) {
1586 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1587 !cfs_capable(CFS_CAP_FOWNER))
1591 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1592 if (!(attr->ia_valid & ATTR_CTIME_SET) &&
1593 (attr->ia_valid & ATTR_CTIME)) {
1594 attr->ia_ctime = current_time(inode);
1595 attr->ia_valid |= ATTR_CTIME_SET;
1597 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1598 (attr->ia_valid & ATTR_ATIME)) {
1599 attr->ia_atime = current_time(inode);
1600 attr->ia_valid |= ATTR_ATIME_SET;
1602 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1603 (attr->ia_valid & ATTR_MTIME)) {
1604 attr->ia_mtime = current_time(inode);
1605 attr->ia_valid |= ATTR_MTIME_SET;
1608 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1609 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %llu\n",
1610 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1611 (s64)ktime_get_real_seconds());
1613 if (S_ISREG(inode->i_mode)) {
1614 if (attr->ia_valid & ATTR_SIZE)
1615 inode_dio_write_done(inode);
1616 inode_unlock(inode);
1619 /* We always do an MDS RPC, even if we're only changing the size;
1620 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1622 OBD_ALLOC_PTR(op_data);
1623 if (op_data == NULL)
1624 GOTO(out, rc = -ENOMEM);
1626 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1627 /* If we are changing file size, file content is
1628 * modified, flag it. */
1629 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1630 op_data->op_bias |= MDS_DATA_MODIFIED;
1631 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1634 op_data->op_attr = *attr;
1636 rc = ll_md_setattr(dentry, op_data);
1640 if (!S_ISREG(inode->i_mode) || hsm_import)
1643 if (attr->ia_valid & (ATTR_SIZE |
1644 ATTR_ATIME | ATTR_ATIME_SET |
1645 ATTR_MTIME | ATTR_MTIME_SET |
1646 ATTR_CTIME | ATTR_CTIME_SET)) {
1647 /* For truncate and utimes sending attributes to OSTs, setting
1648 * mtime/atime to the past will be performed under PW [0:EOF]
1649 * extent lock (new_size:EOF for truncate). It may seem
1650 * excessive to send mtime/atime updates to OSTs when not
1651 * setting times to past, but it is necessary due to possible
1652 * time de-synchronization between MDT inode and OST objects */
1653 rc = cl_setattr_ost(lli->lli_clob, attr, 0);
1656 /* If the file was restored, it needs to set dirty flag.
1658 * We've already sent MDS_DATA_MODIFIED flag in
1659 * ll_md_setattr() for truncate. However, the MDT refuses to
1660 * set the HS_DIRTY flag on released files, so we have to set
1661 * it again if the file has been restored. Please check how
1662 * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1664 * Please notice that if the file is not released, the previous
1665 * MDS_DATA_MODIFIED has taken effect and usually
1666 * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1667 * This way we can save an RPC for common open + trunc
1669 if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1670 struct hsm_state_set hss = {
1671 .hss_valid = HSS_SETMASK,
1672 .hss_setmask = HS_DIRTY,
1676 rc2 = ll_hsm_state_set(inode, &hss);
1677 /* truncate and write can happen at the same time, so that
1678 * the file can be set modified even though the file is not
1679 * restored from released state, and ll_hsm_state_set() is
1680 * not applicable for the file, and rc2 < 0 is normal in this
1683 CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1684 PFID(ll_inode2fid(inode)), rc2);
1689 if (op_data != NULL)
1690 ll_finish_md_op_data(op_data);
1692 if (S_ISREG(inode->i_mode)) {
1694 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1695 inode_dio_wait(inode);
1696 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
1697 * flag. ll_update_inode (called from ll_md_setattr), clears
1698 * inode flags, so there is a gap where S_NOSEC is not set.
1699 * This can cause a writer to take the i_mutex unnecessarily,
1700 * but this is safe to do and should be rare. */
1701 inode_has_no_xattr(inode);
1704 ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1705 LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1710 int ll_setattr(struct dentry *de, struct iattr *attr)
1712 int mode = de->d_inode->i_mode;
1714 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1715 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1716 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1718 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1719 (ATTR_SIZE|ATTR_MODE)) &&
1720 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1721 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1722 !(attr->ia_mode & S_ISGID))))
1723 attr->ia_valid |= ATTR_FORCE;
1725 if ((attr->ia_valid & ATTR_MODE) &&
1727 !(attr->ia_mode & S_ISUID) &&
1728 !(attr->ia_valid & ATTR_KILL_SUID))
1729 attr->ia_valid |= ATTR_KILL_SUID;
1731 if ((attr->ia_valid & ATTR_MODE) &&
1732 ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1733 !(attr->ia_mode & S_ISGID) &&
1734 !(attr->ia_valid & ATTR_KILL_SGID))
1735 attr->ia_valid |= ATTR_KILL_SGID;
1737 /* avoid polluted from ATTR_TIMES_SET,
1738 * projid is not expected to be set here */
1739 attr->ia_valid &= ~MDS_ATTR_PROJID;
1741 return ll_setattr_raw(de, attr, false);
1744 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1745 __u64 max_age, __u32 flags)
1747 struct ll_sb_info *sbi = ll_s2sbi(sb);
1748 struct obd_statfs obd_osfs;
1752 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1754 CERROR("md_statfs fails: rc = %d\n", rc);
1758 osfs->os_type = sb->s_magic;
1760 CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1761 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1763 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1764 flags |= OBD_STATFS_NODELAY;
1766 rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1768 CERROR("obd_statfs fails: rc = %d\n", rc);
1772 CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1773 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1776 osfs->os_bsize = obd_osfs.os_bsize;
1777 osfs->os_blocks = obd_osfs.os_blocks;
1778 osfs->os_bfree = obd_osfs.os_bfree;
1779 osfs->os_bavail = obd_osfs.os_bavail;
1781 /* If we don't have as many objects free on the OST as inodes
1782 * on the MDS, we reduce the total number of inodes to
1783 * compensate, so that the "inodes in use" number is correct.
1785 if (obd_osfs.os_ffree < osfs->os_ffree) {
1786 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1788 osfs->os_ffree = obd_osfs.os_ffree;
1793 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1795 struct super_block *sb = de->d_sb;
1796 struct obd_statfs osfs;
1797 __u64 fsid = huge_encode_dev(sb->s_dev);
1800 CDEBUG(D_VFSTRACE, "VFS Op: at %llu jiffies\n", get_jiffies_64());
1801 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1803 /* Some amount of caching on the client is allowed */
1804 rc = ll_statfs_internal(sb, &osfs,
1805 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1810 statfs_unpack(sfs, &osfs);
1812 /* We need to downshift for all 32-bit kernels, because we can't
1813 * tell if the kernel is being called via sys_statfs64() or not.
1814 * Stop before overflowing f_bsize - in which case it is better
1815 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1816 if (sizeof(long) < 8) {
1817 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1820 osfs.os_blocks >>= 1;
1821 osfs.os_bfree >>= 1;
1822 osfs.os_bavail >>= 1;
1826 sfs->f_blocks = osfs.os_blocks;
1827 sfs->f_bfree = osfs.os_bfree;
1828 sfs->f_bavail = osfs.os_bavail;
1829 sfs->f_fsid.val[0] = (__u32)fsid;
1830 sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1834 void ll_inode_size_lock(struct inode *inode)
1836 struct ll_inode_info *lli;
1838 LASSERT(!S_ISDIR(inode->i_mode));
1840 lli = ll_i2info(inode);
1841 mutex_lock(&lli->lli_size_mutex);
1844 void ll_inode_size_unlock(struct inode *inode)
1846 struct ll_inode_info *lli;
1848 lli = ll_i2info(inode);
1849 mutex_unlock(&lli->lli_size_mutex);
1852 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1854 struct ll_inode_info *lli = ll_i2info(inode);
1855 struct mdt_body *body = md->body;
1856 struct ll_sb_info *sbi = ll_i2sbi(inode);
1859 if (body->mbo_valid & OBD_MD_FLEASIZE) {
1860 rc = cl_file_inode_init(inode, md);
1865 if (S_ISDIR(inode->i_mode)) {
1866 rc = ll_update_lsm_md(inode, md);
1871 #ifdef CONFIG_FS_POSIX_ACL
1872 if (body->mbo_valid & OBD_MD_FLACL) {
1873 spin_lock(&lli->lli_lock);
1874 if (lli->lli_posix_acl)
1875 posix_acl_release(lli->lli_posix_acl);
1876 lli->lli_posix_acl = md->posix_acl;
1877 spin_unlock(&lli->lli_lock);
1880 inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1881 sbi->ll_flags & LL_SBI_32BIT_API);
1882 inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1884 if (body->mbo_valid & OBD_MD_FLATIME) {
1885 if (body->mbo_atime > LTIME_S(inode->i_atime))
1886 LTIME_S(inode->i_atime) = body->mbo_atime;
1887 lli->lli_atime = body->mbo_atime;
1890 if (body->mbo_valid & OBD_MD_FLMTIME) {
1891 if (body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1892 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1893 "to %llu\n", inode->i_ino,
1894 LTIME_S(inode->i_mtime), body->mbo_mtime);
1895 LTIME_S(inode->i_mtime) = body->mbo_mtime;
1897 lli->lli_mtime = body->mbo_mtime;
1900 if (body->mbo_valid & OBD_MD_FLCTIME) {
1901 if (body->mbo_ctime > LTIME_S(inode->i_ctime))
1902 LTIME_S(inode->i_ctime) = body->mbo_ctime;
1903 lli->lli_ctime = body->mbo_ctime;
1906 /* Clear i_flags to remove S_NOSEC before permissions are updated */
1907 if (body->mbo_valid & OBD_MD_FLFLAGS)
1908 inode->i_flags = ll_ext_to_inode_flags(body->mbo_flags);
1909 if (body->mbo_valid & OBD_MD_FLMODE)
1910 inode->i_mode = (inode->i_mode & S_IFMT) |
1911 (body->mbo_mode & ~S_IFMT);
1913 if (body->mbo_valid & OBD_MD_FLTYPE)
1914 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1915 (body->mbo_mode & S_IFMT);
1917 LASSERT(inode->i_mode != 0);
1918 if (S_ISREG(inode->i_mode))
1919 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1920 LL_MAX_BLKSIZE_BITS);
1922 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1924 if (body->mbo_valid & OBD_MD_FLUID)
1925 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
1926 if (body->mbo_valid & OBD_MD_FLGID)
1927 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
1928 if (body->mbo_valid & OBD_MD_FLPROJID)
1929 lli->lli_projid = body->mbo_projid;
1930 if (body->mbo_valid & OBD_MD_FLNLINK)
1931 set_nlink(inode, body->mbo_nlink);
1932 if (body->mbo_valid & OBD_MD_FLRDEV)
1933 inode->i_rdev = old_decode_dev(body->mbo_rdev);
1935 if (body->mbo_valid & OBD_MD_FLID) {
1936 /* FID shouldn't be changed! */
1937 if (fid_is_sane(&lli->lli_fid)) {
1938 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
1939 "Trying to change FID "DFID
1940 " to the "DFID", inode "DFID"(%p)\n",
1941 PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
1942 PFID(ll_inode2fid(inode)), inode);
1944 lli->lli_fid = body->mbo_fid1;
1948 LASSERT(fid_seq(&lli->lli_fid) != 0);
1950 if (body->mbo_valid & OBD_MD_FLSIZE) {
1951 i_size_write(inode, body->mbo_size);
1953 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
1954 PFID(ll_inode2fid(inode)),
1955 (unsigned long long)body->mbo_size);
1957 if (body->mbo_valid & OBD_MD_FLBLOCKS)
1958 inode->i_blocks = body->mbo_blocks;
1961 if (body->mbo_valid & OBD_MD_TSTATE) {
1962 /* Set LLIF_FILE_RESTORING if restore ongoing and
1963 * clear it when done to ensure to start again
1964 * glimpsing updated attrs
1966 if (body->mbo_t_state & MS_RESTORE)
1967 ll_file_set_flag(lli, LLIF_FILE_RESTORING);
1969 ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
1975 int ll_read_inode2(struct inode *inode, void *opaque)
1977 struct lustre_md *md = opaque;
1978 struct ll_inode_info *lli = ll_i2info(inode);
1982 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1983 PFID(&lli->lli_fid), inode);
1985 /* Core attributes from the MDS first. This is a new inode, and
1986 * the VFS doesn't zero times in the core inode so we have to do
1987 * it ourselves. They will be overwritten by either MDS or OST
1988 * attributes - we just need to make sure they aren't newer. */
1989 LTIME_S(inode->i_mtime) = 0;
1990 LTIME_S(inode->i_atime) = 0;
1991 LTIME_S(inode->i_ctime) = 0;
1993 rc = ll_update_inode(inode, md);
1997 /* OIDEBUG(inode); */
1999 #ifdef HAVE_BACKING_DEV_INFO
2000 /* initializing backing dev info. */
2001 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2003 if (S_ISREG(inode->i_mode)) {
2004 struct ll_sb_info *sbi = ll_i2sbi(inode);
2005 inode->i_op = &ll_file_inode_operations;
2006 inode->i_fop = sbi->ll_fop;
2007 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2009 } else if (S_ISDIR(inode->i_mode)) {
2010 inode->i_op = &ll_dir_inode_operations;
2011 inode->i_fop = &ll_dir_operations;
2013 } else if (S_ISLNK(inode->i_mode)) {
2014 inode->i_op = &ll_fast_symlink_inode_operations;
2017 inode->i_op = &ll_special_inode_operations;
2019 init_special_inode(inode, inode->i_mode,
2028 void ll_delete_inode(struct inode *inode)
2030 struct ll_inode_info *lli = ll_i2info(inode);
2033 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2034 /* It is last chance to write out dirty pages,
2035 * otherwise we may lose data while umount */
2036 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
2038 truncate_inode_pages_final(&inode->i_data);
2040 LASSERTF(inode->i_data.nrpages == 0, "inode="DFID"(%p) nrpages=%lu, "
2041 "see https://jira.hpdd.intel.com/browse/LU-118\n",
2042 PFID(ll_inode2fid(inode)), inode, inode->i_data.nrpages);
2044 #ifdef HAVE_SBOPS_EVICT_INODE
2045 ll_clear_inode(inode);
2052 int ll_iocontrol(struct inode *inode, struct file *file,
2053 unsigned int cmd, unsigned long arg)
2055 struct ll_sb_info *sbi = ll_i2sbi(inode);
2056 struct ptlrpc_request *req = NULL;
2061 case FSFILT_IOC_GETFLAGS: {
2062 struct mdt_body *body;
2063 struct md_op_data *op_data;
2065 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2066 0, 0, LUSTRE_OPC_ANY,
2068 if (IS_ERR(op_data))
2069 RETURN(PTR_ERR(op_data));
2071 op_data->op_valid = OBD_MD_FLFLAGS;
2072 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2073 ll_finish_md_op_data(op_data);
2075 CERROR("%s: failure inode "DFID": rc = %d\n",
2076 sbi->ll_md_exp->exp_obd->obd_name,
2077 PFID(ll_inode2fid(inode)), rc);
2081 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2083 flags = body->mbo_flags;
2085 ptlrpc_req_finished(req);
2087 RETURN(put_user(flags, (int __user *)arg));
2089 case FSFILT_IOC_SETFLAGS: {
2091 struct md_op_data *op_data;
2092 struct cl_object *obj;
2094 if (get_user(flags, (int __user *)arg))
2097 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2098 LUSTRE_OPC_ANY, NULL);
2099 if (IS_ERR(op_data))
2100 RETURN(PTR_ERR(op_data));
2102 op_data->op_attr_flags = flags;
2103 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2104 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2105 ll_finish_md_op_data(op_data);
2106 ptlrpc_req_finished(req);
2110 inode->i_flags = ll_ext_to_inode_flags(flags);
2112 obj = ll_i2info(inode)->lli_clob;
2116 OBD_ALLOC_PTR(attr);
2120 attr->ia_valid = ATTR_ATTR_FLAG;
2121 rc = cl_setattr_ost(obj, attr, flags);
2133 int ll_flush_ctx(struct inode *inode)
2135 struct ll_sb_info *sbi = ll_i2sbi(inode);
2137 CDEBUG(D_SEC, "flush context for user %d\n",
2138 from_kuid(&init_user_ns, current_uid()));
2140 obd_set_info_async(NULL, sbi->ll_md_exp,
2141 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2143 obd_set_info_async(NULL, sbi->ll_dt_exp,
2144 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2149 /* umount -f client means force down, don't save state */
2150 void ll_umount_begin(struct super_block *sb)
2152 struct ll_sb_info *sbi = ll_s2sbi(sb);
2153 struct obd_device *obd;
2154 struct obd_ioctl_data *ioc_data;
2155 struct l_wait_info lwi;
2156 wait_queue_head_t waitq;
2159 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2160 sb->s_count, atomic_read(&sb->s_active));
2162 obd = class_exp2obd(sbi->ll_md_exp);
2164 CERROR("Invalid MDC connection handle %#llx\n",
2165 sbi->ll_md_exp->exp_handle.h_cookie);
2171 obd = class_exp2obd(sbi->ll_dt_exp);
2173 CERROR("Invalid LOV connection handle %#llx\n",
2174 sbi->ll_dt_exp->exp_handle.h_cookie);
2180 OBD_ALLOC_PTR(ioc_data);
2182 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2183 sizeof *ioc_data, ioc_data, NULL);
2185 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2186 sizeof *ioc_data, ioc_data, NULL);
2188 OBD_FREE_PTR(ioc_data);
2191 /* Really, we'd like to wait until there are no requests outstanding,
2192 * and then continue. For now, we just periodically checking for vfs
2193 * to decrement mnt_cnt and hope to finish it within 10sec.
2195 init_waitqueue_head(&waitq);
2196 lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(10),
2197 cfs_time_seconds(1), NULL, NULL);
2198 l_wait_event(waitq, may_umount(sbi->ll_mnt.mnt), &lwi);
2203 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2205 struct ll_sb_info *sbi = ll_s2sbi(sb);
2206 char *profilenm = get_profile_name(sb);
2210 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2211 read_only = *flags & MS_RDONLY;
2212 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2213 sizeof(KEY_READ_ONLY),
2214 KEY_READ_ONLY, sizeof(read_only),
2217 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2218 profilenm, read_only ?
2219 "read-only" : "read-write", err);
2224 sb->s_flags |= MS_RDONLY;
2226 sb->s_flags &= ~MS_RDONLY;
2228 if (sbi->ll_flags & LL_SBI_VERBOSE)
2229 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2230 read_only ? "read-only" : "read-write");
2236 * Cleanup the open handle that is cached on MDT-side.
2238 * For open case, the client side open handling thread may hit error
2239 * after the MDT grant the open. Under such case, the client should
2240 * send close RPC to the MDT as cleanup; otherwise, the open handle
2241 * on the MDT will be leaked there until the client umount or evicted.
2243 * In further, if someone unlinked the file, because the open handle
2244 * holds the reference on such file/object, then it will block the
2245 * subsequent threads that want to locate such object via FID.
2247 * \param[in] sb super block for this file-system
2248 * \param[in] open_req pointer to the original open request
2250 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2252 struct mdt_body *body;
2253 struct md_op_data *op_data;
2254 struct ptlrpc_request *close_req = NULL;
2255 struct obd_export *exp = ll_s2sbi(sb)->ll_md_exp;
2258 body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2259 OBD_ALLOC_PTR(op_data);
2260 if (op_data == NULL) {
2261 CWARN("%s: cannot allocate op_data to release open handle for "
2263 ll_get_fsname(sb, NULL, 0), PFID(&body->mbo_fid1));
2268 op_data->op_fid1 = body->mbo_fid1;
2269 op_data->op_handle = body->mbo_handle;
2270 op_data->op_mod_time = ktime_get_real_seconds();
2271 md_close(exp, op_data, NULL, &close_req);
2272 ptlrpc_req_finished(close_req);
2273 ll_finish_md_op_data(op_data);
2278 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2279 struct super_block *sb, struct lookup_intent *it)
2281 struct ll_sb_info *sbi = NULL;
2282 struct lustre_md md = { NULL };
2286 LASSERT(*inode || sb);
2287 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2288 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2289 sbi->ll_md_exp, &md);
2294 rc = ll_update_inode(*inode, &md);
2298 LASSERT(sb != NULL);
2301 * At this point server returns to client's same fid as client
2302 * generated for creating. So using ->fid1 is okay here.
2304 if (!fid_is_sane(&md.body->mbo_fid1)) {
2305 CERROR("%s: Fid is insane "DFID"\n",
2306 ll_get_fsname(sb, NULL, 0),
2307 PFID(&md.body->mbo_fid1));
2308 GOTO(out, rc = -EINVAL);
2311 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2312 sbi->ll_flags & LL_SBI_32BIT_API),
2314 if (IS_ERR(*inode)) {
2315 #ifdef CONFIG_FS_POSIX_ACL
2317 posix_acl_release(md.posix_acl);
2318 md.posix_acl = NULL;
2321 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2323 CERROR("new_inode -fatal: rc %d\n", rc);
2328 /* Handling piggyback layout lock.
2329 * Layout lock can be piggybacked by getattr and open request.
2330 * The lsm can be applied to inode only if it comes with a layout lock
2331 * otherwise correct layout may be overwritten, for example:
2332 * 1. proc1: mdt returns a lsm but not granting layout
2333 * 2. layout was changed by another client
2334 * 3. proc2: refresh layout and layout lock granted
2335 * 4. proc1: to apply a stale layout */
2336 if (it != NULL && it->it_lock_mode != 0) {
2337 struct lustre_handle lockh;
2338 struct ldlm_lock *lock;
2340 lockh.cookie = it->it_lock_handle;
2341 lock = ldlm_handle2lock(&lockh);
2342 LASSERT(lock != NULL);
2343 if (ldlm_has_layout(lock)) {
2344 struct cl_object_conf conf;
2346 memset(&conf, 0, sizeof(conf));
2347 conf.coc_opc = OBJECT_CONF_SET;
2348 conf.coc_inode = *inode;
2349 conf.coc_lock = lock;
2350 conf.u.coc_layout = md.layout;
2351 (void)ll_layout_conf(*inode, &conf);
2353 LDLM_LOCK_PUT(lock);
2359 md_free_lustre_md(sbi->ll_md_exp, &md);
2362 if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2363 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2368 int ll_obd_statfs(struct inode *inode, void __user *arg)
2370 struct ll_sb_info *sbi = NULL;
2371 struct obd_export *exp;
2373 struct obd_ioctl_data *data = NULL;
2377 if (!inode || !(sbi = ll_i2sbi(inode)))
2378 GOTO(out_statfs, rc = -EINVAL);
2380 rc = obd_ioctl_getdata(&buf, &len, arg);
2382 GOTO(out_statfs, rc);
2385 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2386 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2387 GOTO(out_statfs, rc = -EINVAL);
2389 if (data->ioc_inllen1 != sizeof(__u32) ||
2390 data->ioc_inllen2 != sizeof(__u32) ||
2391 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2392 data->ioc_plen2 != sizeof(struct obd_uuid))
2393 GOTO(out_statfs, rc = -EINVAL);
2395 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2396 if (type & LL_STATFS_LMV)
2397 exp = sbi->ll_md_exp;
2398 else if (type & LL_STATFS_LOV)
2399 exp = sbi->ll_dt_exp;
2401 GOTO(out_statfs, rc = -ENODEV);
2403 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2405 GOTO(out_statfs, rc);
2407 OBD_FREE_LARGE(buf, len);
2411 int ll_process_config(struct lustre_cfg *lcfg)
2413 struct super_block *sb;
2418 /* The instance name contains the sb: lustre-client-aacfe000 */
2419 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2420 if (!ptr || !*(++ptr))
2422 if (sscanf(ptr, "%lx", &x) != 1)
2424 sb = (struct super_block *)x;
2425 /* This better be a real Lustre superblock! */
2426 LASSERT(s2lsi(sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2428 /* Note we have not called client_common_fill_super yet, so
2429 proc fns must be able to handle that! */
2430 rc = class_process_proc_param(PARAM_LLITE, lprocfs_llite_obd_vars,
2437 /* this function prepares md_op_data hint for passing it down to MD stack. */
2438 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2439 struct inode *i1, struct inode *i2,
2440 const char *name, size_t namelen,
2441 __u32 mode, __u32 opc, void *data)
2443 LASSERT(i1 != NULL);
2446 /* Do not reuse namelen for something else. */
2448 return ERR_PTR(-EINVAL);
2450 if (namelen > ll_i2sbi(i1)->ll_namelen)
2451 return ERR_PTR(-ENAMETOOLONG);
2453 if (!lu_name_is_valid_2(name, namelen))
2454 return ERR_PTR(-EINVAL);
2457 if (op_data == NULL)
2458 OBD_ALLOC_PTR(op_data);
2460 if (op_data == NULL)
2461 return ERR_PTR(-ENOMEM);
2463 ll_i2gids(op_data->op_suppgids, i1, i2);
2464 op_data->op_fid1 = *ll_inode2fid(i1);
2465 op_data->op_default_stripe_offset = -1;
2466 if (S_ISDIR(i1->i_mode)) {
2467 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2468 if (opc == LUSTRE_OPC_MKDIR)
2469 op_data->op_default_stripe_offset =
2470 ll_i2info(i1)->lli_def_stripe_offset;
2474 op_data->op_fid2 = *ll_inode2fid(i2);
2475 if (S_ISDIR(i2->i_mode))
2476 op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2478 fid_zero(&op_data->op_fid2);
2481 if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2482 op_data->op_cli_flags |= CLI_HASH64;
2484 if (ll_need_32bit_api(ll_i2sbi(i1)))
2485 op_data->op_cli_flags |= CLI_API32;
2487 op_data->op_name = name;
2488 op_data->op_namelen = namelen;
2489 op_data->op_mode = mode;
2490 op_data->op_mod_time = cfs_time_current_sec();
2491 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2492 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2493 op_data->op_cap = cfs_curproc_cap_pack();
2494 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2495 filename_is_volatile(name, namelen, &op_data->op_mds)) {
2496 op_data->op_bias |= MDS_CREATE_VOLATILE;
2498 op_data->op_mds = 0;
2500 op_data->op_data = data;
2505 void ll_finish_md_op_data(struct md_op_data *op_data)
2507 security_release_secctx(op_data->op_file_secctx,
2508 op_data->op_file_secctx_size);
2509 OBD_FREE_PTR(op_data);
2512 #ifdef HAVE_SUPEROPS_USE_DENTRY
2513 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2515 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2518 struct ll_sb_info *sbi;
2520 #ifdef HAVE_SUPEROPS_USE_DENTRY
2521 LASSERT((seq != NULL) && (dentry != NULL));
2522 sbi = ll_s2sbi(dentry->d_sb);
2524 LASSERT((seq != NULL) && (vfs != NULL));
2525 sbi = ll_s2sbi(vfs->mnt_sb);
2528 if (sbi->ll_flags & LL_SBI_NOLCK)
2529 seq_puts(seq, ",nolock");
2531 if (sbi->ll_flags & LL_SBI_FLOCK)
2532 seq_puts(seq, ",flock");
2534 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2535 seq_puts(seq, ",localflock");
2537 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2538 seq_puts(seq, ",user_xattr");
2540 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2541 seq_puts(seq, ",lazystatfs");
2543 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2544 seq_puts(seq, ",user_fid2path");
2546 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2547 seq_puts(seq, ",always_ping");
2553 * Get obd name by cmd, and copy out to user space
2555 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2557 struct ll_sb_info *sbi = ll_i2sbi(inode);
2558 struct obd_device *obd;
2561 if (cmd == OBD_IOC_GETDTNAME)
2562 obd = class_exp2obd(sbi->ll_dt_exp);
2563 else if (cmd == OBD_IOC_GETMDNAME)
2564 obd = class_exp2obd(sbi->ll_md_exp);
2571 if (copy_to_user((void __user *)arg, obd->obd_name,
2572 strlen(obd->obd_name) + 1))
2579 * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2580 * fsname will be returned in this buffer; otherwise, a static buffer will be
2581 * used to store the fsname and returned to caller.
2583 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2585 static char fsname_static[MTI_NAME_MAXLEN];
2586 struct lustre_sb_info *lsi = s2lsi(sb);
2591 /* this means the caller wants to use static buffer
2592 * and it doesn't care about race. Usually this is
2593 * in error reporting path */
2594 buf = fsname_static;
2595 buflen = sizeof(fsname_static);
2598 len = strlen(lsi->lsi_lmd->lmd_profile);
2599 ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2600 if (ptr && (strcmp(ptr, "-client") == 0))
2603 if (unlikely(len >= buflen))
2605 strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2611 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2618 p.mnt = current->fs->root.mnt;
2620 path = d_path(&p, buf, bufsize);
2625 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2627 char *buf, *path = NULL;
2628 struct dentry *dentry = NULL;
2629 struct inode *inode = page->mapping->host;
2631 /* this can be called inside spin lock so use GFP_ATOMIC. */
2632 buf = (char *)__get_free_page(GFP_ATOMIC);
2634 dentry = d_find_alias(page->mapping->host);
2636 path = ll_d_path(dentry, buf, PAGE_SIZE);
2640 "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2641 "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2642 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2643 PFID(ll_inode2fid(inode)),
2644 (path && !IS_ERR(path)) ? path : "", ioret);
2650 free_page((unsigned long)buf);
2653 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2654 struct lov_user_md **kbuf)
2656 struct lov_user_md lum;
2660 if (copy_from_user(&lum, md, sizeof(lum)))
2663 lum_size = ll_lov_user_md_size(&lum);
2667 OBD_ALLOC(*kbuf, lum_size);
2671 if (copy_from_user(*kbuf, md, lum_size) != 0) {
2672 OBD_FREE(*kbuf, lum_size);
2680 * Compute llite root squash state after a change of root squash
2681 * configuration setting or add/remove of a lnet nid
2683 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2685 struct root_squash_info *squash = &sbi->ll_squash;
2688 struct lnet_process_id id;
2690 /* Update norootsquash flag */
2691 down_write(&squash->rsi_sem);
2692 if (list_empty(&squash->rsi_nosquash_nids))
2693 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2695 /* Do not apply root squash as soon as one of our NIDs is
2696 * in the nosquash_nids list */
2699 while (LNetGetId(i++, &id) != -ENOENT) {
2700 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2702 if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2708 sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2710 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2712 up_write(&squash->rsi_sem);
2716 * Parse linkea content to extract information about a given hardlink
2718 * \param[in] ldata - Initialized linkea data
2719 * \param[in] linkno - Link identifier
2720 * \param[out] parent_fid - The entry's parent FID
2721 * \param[out] ln - Entry name destination buffer
2723 * \retval 0 on success
2724 * \retval Appropriate negative error code on failure
2726 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2727 struct lu_fid *parent_fid, struct lu_name *ln)
2733 rc = linkea_init_with_rec(ldata);
2737 if (linkno >= ldata->ld_leh->leh_reccount)
2738 /* beyond last link */
2741 linkea_first_entry(ldata);
2742 for (idx = 0; ldata->ld_lee != NULL; idx++) {
2743 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2748 linkea_next_entry(ldata);
2758 * Get parent FID and name of an identified link. Operation is performed for
2759 * a given link number, letting the caller iterate over linkno to list one or
2760 * all links of an entry.
2762 * \param[in] file - File descriptor against which to perform the operation
2763 * \param[in,out] arg - User-filled structure containing the linkno to operate
2764 * on and the available size. It is eventually filled with
2765 * the requested information or left untouched on error
2767 * \retval - 0 on success
2768 * \retval - Appropriate negative error code on failure
2770 int ll_getparent(struct file *file, struct getparent __user *arg)
2772 struct inode *inode = file_inode(file);
2773 struct linkea_data *ldata;
2774 struct lu_buf buf = LU_BUF_NULL;
2776 struct lu_fid parent_fid;
2783 if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2784 !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2787 if (get_user(name_size, &arg->gp_name_size))
2790 if (get_user(linkno, &arg->gp_linkno))
2793 if (name_size > PATH_MAX)
2796 OBD_ALLOC(ldata, sizeof(*ldata));
2800 rc = linkea_data_new(ldata, &buf);
2802 GOTO(ldata_free, rc);
2804 #ifdef HAVE_XATTR_HANDLER_FLAGS
2805 rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
2806 buf.lb_len, OBD_MD_FLXATTR);
2808 rc = ll_getxattr(file_dentry(file), XATTR_NAME_LINK, buf.lb_buf,
2810 #endif /* HAVE_XATTR_HANDLER_FLAGS */
2814 rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2818 if (ln.ln_namelen >= name_size)
2819 GOTO(lb_free, rc = -EOVERFLOW);
2821 if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2822 GOTO(lb_free, rc = -EFAULT);
2824 if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2825 GOTO(lb_free, rc = -EFAULT);
2827 if (put_user('\0', arg->gp_name + ln.ln_namelen))
2828 GOTO(lb_free, rc = -EFAULT);
2833 OBD_FREE(ldata, sizeof(*ldata));