1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright 2008 Sun Microsystems, Inc. All rights reserved
30 * Use is subject to license terms.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/llite/llite_lib.c
38 * Lustre Light Super operations
41 #define DEBUG_SUBSYSTEM S_LLITE
43 #include <linux/module.h>
44 #include <linux/types.h>
45 #include <linux/random.h>
46 #include <linux/version.h>
49 #include <lustre_lite.h>
50 #include <lustre_ha.h>
51 #include <lustre_dlm.h>
52 #include <lprocfs_status.h>
53 #include <lustre_disk.h>
54 #include <lustre_param.h>
55 #include <lustre_log.h>
56 #include <cl_object.h>
57 #include <obd_cksum.h>
58 #include "llite_internal.h"
60 cfs_mem_cache_t *ll_file_data_slab;
62 LIST_HEAD(ll_super_blocks);
63 spinlock_t ll_sb_lock = SPIN_LOCK_UNLOCKED;
65 extern struct address_space_operations ll_aops;
66 extern struct address_space_operations ll_dir_aops;
69 #define log2(n) ffz(~(n))
72 static struct ll_sb_info *ll_init_sbi(void)
74 struct ll_sb_info *sbi = NULL;
81 OBD_ALLOC(sbi, sizeof(*sbi));
85 spin_lock_init(&sbi->ll_lock);
86 spin_lock_init(&sbi->ll_lco.lco_lock);
87 spin_lock_init(&sbi->ll_pp_extent_lock);
88 spin_lock_init(&sbi->ll_process_lock);
89 sbi->ll_rw_stats_on = 0;
92 pages = si.totalram - si.totalhigh;
93 if (pages >> (20 - CFS_PAGE_SHIFT) < 512) {
94 #ifdef HAVE_BGL_SUPPORT
95 sbi->ll_async_page_max = pages / 4;
97 sbi->ll_async_page_max = pages / 2;
100 sbi->ll_async_page_max = (pages / 4) * 3;
103 sbi->ll_ra_info.ra_max_pages = min(pages / 32,
104 SBI_DEFAULT_READAHEAD_MAX);
105 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
106 SBI_DEFAULT_READAHEAD_WHOLE_MAX;
107 INIT_LIST_HEAD(&sbi->ll_conn_chain);
108 INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
110 ll_generate_random_uuid(uuid);
111 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
112 CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
114 spin_lock(&ll_sb_lock);
115 list_add_tail(&sbi->ll_list, &ll_super_blocks);
116 spin_unlock(&ll_sb_lock);
118 #ifdef ENABLE_LLITE_CHECKSUM
119 sbi->ll_flags |= LL_SBI_CHECKSUM;
122 #ifdef HAVE_LRU_RESIZE_SUPPORT
123 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
126 #ifdef HAVE_EXPORT___IGET
127 INIT_LIST_HEAD(&sbi->ll_deathrow);
128 spin_lock_init(&sbi->ll_deathrow_lock);
130 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
131 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].pp_r_hist.oh_lock);
132 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].pp_w_hist.oh_lock);
135 /* metadata statahead is enabled by default */
136 sbi->ll_sa_max = LL_SA_RPC_DEF;
141 void ll_free_sbi(struct super_block *sb)
143 struct ll_sb_info *sbi = ll_s2sbi(sb);
147 spin_lock(&ll_sb_lock);
148 list_del(&sbi->ll_list);
149 spin_unlock(&ll_sb_lock);
150 OBD_FREE(sbi, sizeof(*sbi));
155 static struct dentry_operations ll_d_root_ops = {
156 #ifdef DCACHE_LUSTRE_INVALID
157 .d_compare = ll_dcompare,
161 static int client_common_fill_super(struct super_block *sb, char *md, char *dt)
163 struct inode *root = 0;
164 struct ll_sb_info *sbi = ll_s2sbi(sb);
165 struct obd_device *obd;
166 struct obd_capa *oc = NULL;
167 struct obd_statfs osfs;
168 struct ptlrpc_request *request = NULL;
169 struct obd_connect_data *data = NULL;
170 struct obd_uuid *uuid;
171 struct lustre_md lmd;
173 int size, err, checksum;
176 obd = class_name2obd(md);
178 CERROR("MD %s: not setup or attached\n", md);
186 if (proc_lustre_fs_root) {
187 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
190 CERROR("could not register mount in /proc/fs/lustre\n");
193 /* indicate the features supported by this client */
194 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
195 OBD_CONNECT_JOIN | OBD_CONNECT_ATTRFID |
196 OBD_CONNECT_VERSION | OBD_CONNECT_MDS_CAPA |
197 OBD_CONNECT_OSS_CAPA | OBD_CONNECT_CANCELSET|
198 OBD_CONNECT_FID | OBD_CONNECT_AT |
199 OBD_CONNECT_LOV_V3 | OBD_CONNECT_RMT_CLIENT |
202 #ifdef HAVE_LRU_RESIZE_SUPPORT
203 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
204 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
206 #ifdef CONFIG_FS_POSIX_ACL
207 data->ocd_connect_flags |= OBD_CONNECT_ACL;
209 data->ocd_ibits_known = MDS_INODELOCK_FULL;
210 data->ocd_version = LUSTRE_VERSION_CODE;
212 if (sb->s_flags & MS_RDONLY)
213 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
214 if (sbi->ll_flags & LL_SBI_USER_XATTR)
215 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
217 #ifdef HAVE_MS_FLOCK_LOCK
218 /* force vfs to use lustre handler for flock() calls - bug 10743 */
219 sb->s_flags |= MS_FLOCK_LOCK;
222 if (sbi->ll_flags & LL_SBI_FLOCK)
223 sbi->ll_fop = &ll_file_operations_flock;
224 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
225 sbi->ll_fop = &ll_file_operations;
227 sbi->ll_fop = &ll_file_operations_noflock;
230 data->ocd_connect_flags |= OBD_CONNECT_REAL;
231 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
232 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
234 err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
236 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
237 "recovery, of which this client is not a "
238 "part. Please wait for recovery to complete,"
239 " abort, or time out.\n", md);
242 CERROR("cannot connect to %s: rc = %d\n", md, err);
246 err = obd_fid_init(sbi->ll_md_exp);
248 CERROR("Can't init metadata layer FID infrastructure, "
253 err = obd_statfs(obd, &osfs, cfs_time_current_64() - HZ, 0);
255 GOTO(out_md_fid, err);
257 size = sizeof(*data);
258 err = obd_get_info(sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
259 KEY_CONN_DATA, &size, data, NULL);
261 CERROR("Get connect data failed: %d \n", err);
265 LASSERT(osfs.os_bsize);
266 sb->s_blocksize = osfs.os_bsize;
267 sb->s_blocksize_bits = log2(osfs.os_bsize);
268 sb->s_magic = LL_SUPER_MAGIC;
270 /* for bug 11559. in $LINUX/fs/read_write.c, function do_sendfile():
271 * retval = in_file->f_op->sendfile(...);
273 * retval = -EOVERFLOW;
275 * it will check if *ppos is greater than max. However, max equals to
276 * s_maxbytes, which is a negative integer in a x86_64 box since loff_t
277 * has been defined as a signed long long ineger in linux kernel. */
278 #if BITS_PER_LONG == 64
279 sb->s_maxbytes = PAGE_CACHE_MAXBYTES >> 1;
281 sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
283 sbi->ll_namelen = osfs.os_namelen;
284 sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
286 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
287 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
288 LCONSOLE_INFO("Disabling user_xattr feature because "
289 "it is not supported on the server\n");
290 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
293 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
295 sb->s_flags |= MS_POSIXACL;
297 sbi->ll_flags |= LL_SBI_ACL;
299 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
301 sb->s_flags &= ~MS_POSIXACL;
303 sbi->ll_flags &= ~LL_SBI_ACL;
306 if (data->ocd_connect_flags & OBD_CONNECT_JOIN)
307 sbi->ll_flags |= LL_SBI_JOIN;
309 if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
310 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
311 sbi->ll_flags |= LL_SBI_RMT_CLIENT;
312 LCONSOLE_INFO("client is set as remote by default.\n");
315 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
316 sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
317 LCONSOLE_INFO("client claims to be remote, but server "
318 "rejected, forced to be local.\n");
322 if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
323 LCONSOLE_INFO("client enabled MDS capability!\n");
324 sbi->ll_flags |= LL_SBI_MDS_CAPA;
327 if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
328 LCONSOLE_INFO("client enabled OSS capability!\n");
329 sbi->ll_flags |= LL_SBI_OSS_CAPA;
332 obd = class_name2obd(dt);
334 CERROR("DT %s: not setup or attached\n", dt);
335 GOTO(out_md_fid, err = -ENODEV);
338 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
339 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
340 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
341 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
342 OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
343 OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
344 OBD_CONNECT_GRANT_SHRINK;
346 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
347 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
348 * disabled by default, because it can still be enabled on the
349 * fly via /proc. As a consequence, we still need to come to an
350 * agreement on the supported algorithms at connect time */
351 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
353 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
354 data->ocd_cksum_types = OBD_CKSUM_ADLER;
356 /* send the list of supported checksum types */
357 data->ocd_cksum_types = OBD_CKSUM_ALL;
360 #ifdef HAVE_LRU_RESIZE_SUPPORT
361 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
363 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
364 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
366 CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
367 "ocd_grant: %d\n", data->ocd_connect_flags,
368 data->ocd_version, data->ocd_grant);
370 obd->obd_upcall.onu_owner = &sbi->ll_lco;
371 obd->obd_upcall.onu_upcall = cl_ocd_update;
372 data->ocd_brw_size = PTLRPC_MAX_BRW_PAGES << CFS_PAGE_SHIFT;
374 err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data, NULL);
376 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
377 "recovery, of which this client is not a "
378 "part. Please wait for recovery to "
379 "complete, abort, or time out.\n", dt);
380 GOTO(out_md_fid, err);
382 CERROR("Cannot connect to %s: rc = %d\n", dt, err);
383 GOTO(out_md_fid, err);
386 err = obd_fid_init(sbi->ll_dt_exp);
388 CERROR("Can't init data layer FID infrastructure, "
393 spin_lock(&sbi->ll_lco.lco_lock);
394 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
395 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
396 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
397 spin_unlock(&sbi->ll_lco.lco_lock);
399 fid_zero(&sbi->ll_root_fid);
400 err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
402 CERROR("cannot mds_connect: rc = %d\n", err);
403 GOTO(out_lock_cn_cb, err);
405 if (!fid_is_sane(&sbi->ll_root_fid)) {
406 CERROR("Invalid root fid during mount\n");
407 GOTO(out_lock_cn_cb, err = -EINVAL);
409 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
411 sb->s_op = &lustre_super_operations;
412 sb->s_export_op = &lustre_export_operations;
415 * XXX: move this to after cbd setup? */
416 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
417 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
418 valid |= OBD_MD_FLRMTPERM;
419 else if (sbi->ll_flags & LL_SBI_ACL)
420 valid |= OBD_MD_FLACL;
422 err = md_getattr(sbi->ll_md_exp, &sbi->ll_root_fid, oc, valid, 0,
427 CERROR("md_getattr failed for root: rc = %d\n", err);
428 GOTO(out_lock_cn_cb, err);
430 memset(&lmd, 0, sizeof(lmd));
431 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
432 sbi->ll_md_exp, &lmd);
434 CERROR("failed to understand root inode md: rc = %d\n", err);
435 ptlrpc_req_finished (request);
436 GOTO(out_lock_cn_cb, err);
439 LASSERT(fid_is_sane(&sbi->ll_root_fid));
440 root = ll_iget(sb, ll_fid_build_ino(sbi, &sbi->ll_root_fid), &lmd);
441 md_free_lustre_md(sbi->ll_md_exp, &lmd);
442 ptlrpc_req_finished(request);
444 if (root == NULL || IS_ERR(root)) {
446 obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
447 #ifdef CONFIG_FS_POSIX_ACL
449 posix_acl_release(lmd.posix_acl);
450 lmd.posix_acl = NULL;
453 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
455 CERROR("lustre_lite: bad iget4 for root\n");
459 err = ll_close_thread_start(&sbi->ll_lcq);
461 CERROR("cannot start close thread: rc %d\n", err);
465 #ifdef CONFIG_FS_POSIX_ACL
466 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
467 rct_init(&sbi->ll_rct);
468 et_init(&sbi->ll_et);
472 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
473 err = obd_set_info_async(sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
474 KEY_CHECKSUM, sizeof(checksum), &checksum,
478 sb->s_root = d_alloc_root(root);
480 OBD_FREE(data, sizeof(*data));
482 sb->s_root->d_op = &ll_d_root_ops;
484 sbi->ll_sdev_orig = sb->s_dev;
486 /* We set sb->s_dev equal on all lustre clients in order to support
487 * NFS export clustering. NFSD requires that the FSID be the same
489 /* s_dev is also used in lt_compare() to compare two fs, but that is
490 * only a node-local comparison. */
491 uuid = obd_get_uuid(sbi->ll_md_exp);
493 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
500 obd_fid_fini(sbi->ll_dt_exp);
502 obd_disconnect(sbi->ll_dt_exp);
503 sbi->ll_dt_exp = NULL;
505 obd_fid_fini(sbi->ll_md_exp);
507 obd_disconnect(sbi->ll_md_exp);
508 sbi->ll_md_exp = NULL;
512 lprocfs_unregister_mountpoint(sbi);
516 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
520 *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
522 rc = obd_get_info(sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
523 KEY_MAX_EASIZE, &size, lmmsize, NULL);
525 CERROR("Get max mdsize error rc %d \n", rc);
530 void ll_dump_inode(struct inode *inode)
532 struct list_head *tmp;
533 int dentry_count = 0;
535 LASSERT(inode != NULL);
537 list_for_each(tmp, &inode->i_dentry)
540 CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
541 inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
542 inode->i_mode, atomic_read(&inode->i_count), dentry_count);
545 void lustre_dump_dentry(struct dentry *dentry, int recur)
547 struct list_head *tmp;
550 LASSERT(dentry != NULL);
552 list_for_each(tmp, &dentry->d_subdirs)
555 CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
556 " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
557 dentry->d_name.len, dentry->d_name.name,
558 dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
559 dentry->d_parent, dentry->d_inode, atomic_read(&dentry->d_count),
560 dentry->d_flags, dentry->d_fsdata, subdirs);
561 if (dentry->d_inode != NULL)
562 ll_dump_inode(dentry->d_inode);
567 list_for_each(tmp, &dentry->d_subdirs) {
568 struct dentry *d = list_entry(tmp, struct dentry, d_child);
569 lustre_dump_dentry(d, recur - 1);
573 #ifdef HAVE_EXPORT___IGET
574 static void prune_dir_dentries(struct inode *inode)
576 struct dentry *dentry, *prev = NULL;
578 /* due to lustre specific logic, a directory
579 * can have few dentries - a bug from VFS POV */
581 spin_lock(&dcache_lock);
582 if (!list_empty(&inode->i_dentry)) {
583 dentry = list_entry(inode->i_dentry.prev,
584 struct dentry, d_alias);
585 /* in order to prevent infinite loops we
586 * break if previous dentry is busy */
587 if (dentry != prev) {
590 spin_unlock(&dcache_lock);
592 /* try to kill all child dentries */
594 shrink_dcache_parent(dentry);
595 unlock_dentry(dentry);
598 /* now try to get rid of current dentry */
599 d_prune_aliases(inode);
603 spin_unlock(&dcache_lock);
606 static void prune_deathrow_one(struct ll_inode_info *lli)
608 struct inode *inode = ll_info2i(lli);
610 /* first, try to drop any dentries - they hold a ref on the inode */
611 if (S_ISDIR(inode->i_mode))
612 prune_dir_dentries(inode);
614 d_prune_aliases(inode);
617 /* if somebody still uses it, leave it */
618 LASSERT(atomic_read(&inode->i_count) > 0);
619 if (atomic_read(&inode->i_count) > 1)
622 CDEBUG(D_INODE, "inode %lu/%u(%d) looks a good candidate for prune\n",
623 inode->i_ino,inode->i_generation, atomic_read(&inode->i_count));
625 /* seems nobody uses it anymore */
633 static void prune_deathrow(struct ll_sb_info *sbi, int try)
635 struct ll_inode_info *lli;
639 if (need_resched() && try)
643 if (!spin_trylock(&sbi->ll_deathrow_lock))
646 spin_lock(&sbi->ll_deathrow_lock);
651 if (!list_empty(&sbi->ll_deathrow)) {
652 lli = list_entry(sbi->ll_deathrow.next,
653 struct ll_inode_info,
655 list_del_init(&lli->lli_dead_list);
656 if (!list_empty(&sbi->ll_deathrow))
659 spin_unlock(&sbi->ll_deathrow_lock);
662 prune_deathrow_one(lli);
664 } while (empty == 0);
666 #else /* !HAVE_EXPORT___IGET */
667 #define prune_deathrow(sbi, try) do {} while (0)
668 #endif /* HAVE_EXPORT___IGET */
670 void client_common_put_super(struct super_block *sb)
672 struct ll_sb_info *sbi = ll_s2sbi(sb);
675 #ifdef CONFIG_FS_POSIX_ACL
676 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
677 et_fini(&sbi->ll_et);
678 rct_fini(&sbi->ll_rct);
682 obd_cancel_unused(sbi->ll_dt_exp, NULL, 0, NULL);
684 ll_close_thread_shutdown(sbi->ll_lcq);
686 /* destroy inodes in deathrow */
687 prune_deathrow(sbi, 0);
689 list_del(&sbi->ll_conn_chain);
691 obd_fid_fini(sbi->ll_dt_exp);
692 obd_disconnect(sbi->ll_dt_exp);
693 sbi->ll_dt_exp = NULL;
695 lprocfs_unregister_mountpoint(sbi);
697 obd_fid_fini(sbi->ll_md_exp);
698 obd_disconnect(sbi->ll_md_exp);
699 sbi->ll_md_exp = NULL;
704 void ll_kill_super(struct super_block *sb)
706 struct ll_sb_info *sbi;
711 if (!(sb->s_flags & MS_ACTIVE))
715 /* we need restore s_dev from changed for clustred NFS before put_super
716 * because new kernels have cached s_dev and change sb->s_dev in
717 * put_super not affected real removing devices */
719 sb->s_dev = sbi->ll_sdev_orig;
723 char *ll_read_opt(const char *opt, char *data)
729 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
730 if (strncmp(opt, data, strlen(opt)))
732 if ((value = strchr(data, '=')) == NULL)
736 OBD_ALLOC(retval, strlen(value) + 1);
738 CERROR("out of memory!\n");
742 memcpy(retval, value, strlen(value)+1);
743 CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
747 static inline int ll_set_opt(const char *opt, char *data, int fl)
749 if (strncmp(opt, data, strlen(opt)) != 0)
755 /* non-client-specific mount options are parsed in lmd_parse */
756 static int ll_options(char *options, int *flags)
759 char *s1 = options, *s2;
765 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
768 CDEBUG(D_SUPER, "next opt=%s\n", s1);
769 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
774 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
779 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
784 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
789 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
794 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
799 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
801 /* Ignore deprecated mount option. The client will
802 * always try to mount with ACL support, whether this
803 * is used depends on whether server supports it. */
806 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
810 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
816 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
821 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
826 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
831 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
836 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
841 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
847 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
853 s2 = strchr(s1, ',');
861 void ll_lli_init(struct ll_inode_info *lli)
863 lli->lli_inode_magic = LLI_INODE_MAGIC;
864 sema_init(&lli->lli_size_sem, 1);
865 sema_init(&lli->lli_write_sem, 1);
867 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
868 spin_lock_init(&lli->lli_lock);
869 INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
870 INIT_LIST_HEAD(&lli->lli_close_list);
871 lli->lli_inode_magic = LLI_INODE_MAGIC;
872 sema_init(&lli->lli_och_sem, 1);
873 lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
874 lli->lli_mds_exec_och = NULL;
875 lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
876 lli->lli_open_fd_exec_count = 0;
877 INIT_LIST_HEAD(&lli->lli_dead_list);
878 lli->lli_remote_perms = NULL;
879 lli->lli_rmtperm_utime = 0;
880 sema_init(&lli->lli_rmtperm_sem, 1);
881 INIT_LIST_HEAD(&lli->lli_oss_capas);
884 int ll_fill_super(struct super_block *sb)
886 struct lustre_profile *lprof;
887 struct lustre_sb_info *lsi = s2lsi(sb);
888 struct ll_sb_info *sbi;
889 char *dt = NULL, *md = NULL;
890 char *profilenm = get_profile_name(sb);
891 struct config_llog_instance cfg = {0, };
892 char ll_instance[sizeof(sb) * 2 + 1];
896 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
900 /* client additional sb info */
901 lsi->lsi_llsbi = sbi = ll_init_sbi();
907 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
911 /* Generate a string unique to this super, in case some joker tries
912 to mount the same fs at two mount points.
913 Use the address of the super itself.*/
914 sprintf(ll_instance, "%p", sb);
915 cfg.cfg_instance = ll_instance;
916 cfg.cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
918 /* set up client obds */
919 err = lustre_process_log(sb, profilenm, &cfg);
921 CERROR("Unable to process log: %d\n", err);
925 lprof = class_get_profile(profilenm);
927 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
928 " read from the MGS. Does that filesystem "
929 "exist?\n", profilenm);
930 GOTO(out_free, err = -EINVAL);
932 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
933 lprof->lp_md, lprof->lp_dt);
935 OBD_ALLOC(dt, strlen(lprof->lp_dt) +
936 strlen(ll_instance) + 2);
938 GOTO(out_free, err = -ENOMEM);
939 sprintf(dt, "%s-%s", lprof->lp_dt, ll_instance);
941 OBD_ALLOC(md, strlen(lprof->lp_md) +
942 strlen(ll_instance) + 2);
944 GOTO(out_free, err = -ENOMEM);
945 sprintf(md, "%s-%s", lprof->lp_md, ll_instance);
947 /* connections, registrations, sb setup */
948 err = client_common_fill_super(sb, md, dt);
952 OBD_FREE(md, strlen(md) + 1);
954 OBD_FREE(dt, strlen(dt) + 1);
958 LCONSOLE_WARN("Client %s has started\n", profilenm);
961 } /* ll_fill_super */
964 void lu_context_keys_dump(void);
966 void ll_put_super(struct super_block *sb)
968 struct config_llog_instance cfg;
969 char ll_instance[sizeof(sb) * 2 + 1];
970 struct obd_device *obd;
971 struct lustre_sb_info *lsi = s2lsi(sb);
972 struct ll_sb_info *sbi = ll_s2sbi(sb);
973 char *profilenm = get_profile_name(sb);
977 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
979 ll_print_capa_stat(sbi);
981 sprintf(ll_instance, "%p", sb);
982 cfg.cfg_instance = ll_instance;
983 lustre_end_log(sb, NULL, &cfg);
985 if (sbi->ll_md_exp) {
986 obd = class_exp2obd(sbi->ll_md_exp);
988 force = obd->obd_force;
991 /* We need to set force before the lov_disconnect in
992 lustre_common_put_super, since l_d cleans up osc's as well. */
995 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
997 obd->obd_force = force;
1004 /* Only if client_common_fill_super succeeded */
1005 client_common_put_super(sb);
1008 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1009 class_manual_cleanup(obd);
1013 class_del_profile(profilenm);
1016 lsi->lsi_llsbi = NULL;
1018 lustre_common_put_super(sb);
1020 cl_env_cache_purge(~0);
1022 LCONSOLE_WARN("client %s umount complete\n", ll_instance);
1027 } /* client_put_super */
1029 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1031 struct inode *inode = NULL;
1032 /* NOTE: we depend on atomic igrab() -bzzz */
1033 lock_res_and_lock(lock);
1034 if (lock->l_ast_data) {
1035 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1036 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1037 inode = igrab(lock->l_ast_data);
1039 inode = lock->l_ast_data;
1040 ldlm_lock_debug(NULL, inode->i_state & I_FREEING ?
1042 lock, __FILE__, __func__, __LINE__,
1043 "l_ast_data %p is bogus: magic %08x",
1044 lock->l_ast_data, lli->lli_inode_magic);
1048 unlock_res_and_lock(lock);
1052 static int null_if_equal(struct ldlm_lock *lock, void *data)
1054 if (data == lock->l_ast_data) {
1055 lock->l_ast_data = NULL;
1057 if (lock->l_req_mode != lock->l_granted_mode)
1058 LDLM_ERROR(lock,"clearing inode with ungranted lock");
1061 return LDLM_ITER_CONTINUE;
1064 void ll_clear_inode(struct inode *inode)
1066 struct ll_inode_info *lli = ll_i2info(inode);
1067 struct ll_sb_info *sbi = ll_i2sbi(inode);
1070 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1071 inode->i_generation, inode);
1073 if (S_ISDIR(inode->i_mode)) {
1074 /* these should have been cleared in ll_file_release */
1075 LASSERT(lli->lli_sai == NULL);
1076 LASSERT(lli->lli_opendir_key == NULL);
1077 LASSERT(lli->lli_opendir_pid == 0);
1080 ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1081 md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1082 null_if_equal, inode);
1084 LASSERT(!lli->lli_open_fd_write_count);
1085 LASSERT(!lli->lli_open_fd_read_count);
1086 LASSERT(!lli->lli_open_fd_exec_count);
1088 if (lli->lli_mds_write_och)
1089 ll_md_real_close(inode, FMODE_WRITE);
1090 if (lli->lli_mds_exec_och)
1091 ll_md_real_close(inode, FMODE_EXEC);
1092 if (lli->lli_mds_read_och)
1093 ll_md_real_close(inode, FMODE_READ);
1095 if (lli->lli_symlink_name) {
1096 OBD_FREE(lli->lli_symlink_name,
1097 strlen(lli->lli_symlink_name) + 1);
1098 lli->lli_symlink_name = NULL;
1101 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1102 LASSERT(lli->lli_posix_acl == NULL);
1103 if (lli->lli_remote_perms) {
1104 free_rmtperm_hash(lli->lli_remote_perms);
1105 lli->lli_remote_perms = NULL;
1108 #ifdef CONFIG_FS_POSIX_ACL
1109 else if (lli->lli_posix_acl) {
1110 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1111 LASSERT(lli->lli_remote_perms == NULL);
1112 posix_acl_release(lli->lli_posix_acl);
1113 lli->lli_posix_acl = NULL;
1116 lli->lli_inode_magic = LLI_INODE_DEAD;
1118 #ifdef HAVE_EXPORT___IGET
1119 spin_lock(&sbi->ll_deathrow_lock);
1120 list_del_init(&lli->lli_dead_list);
1121 spin_unlock(&sbi->ll_deathrow_lock);
1123 ll_clear_inode_capas(inode);
1125 * XXX This has to be done before lsm is freed below, because
1126 * cl_object still uses inode lsm.
1128 cl_inode_fini(inode);
1131 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1132 lli->lli_smd = NULL;
1139 int ll_md_setattr(struct inode *inode, struct md_op_data *op_data,
1140 struct md_open_data **mod)
1142 struct lustre_md md;
1143 struct ll_sb_info *sbi = ll_i2sbi(inode);
1144 struct ptlrpc_request *request = NULL;
1148 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1149 LUSTRE_OPC_ANY, NULL);
1150 if (IS_ERR(op_data))
1151 RETURN(PTR_ERR(op_data));
1153 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1156 ptlrpc_req_finished(request);
1157 if (rc == -ENOENT) {
1159 /* Unlinked special device node? Or just a race?
1160 * Pretend we done everything. */
1161 if (!S_ISREG(inode->i_mode) &&
1162 !S_ISDIR(inode->i_mode))
1163 rc = inode_setattr(inode, &op_data->op_attr);
1164 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1165 CERROR("md_setattr fails: rc = %d\n", rc);
1170 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1171 sbi->ll_md_exp, &md);
1173 ptlrpc_req_finished(request);
1177 /* We call inode_setattr to adjust timestamps.
1178 * If there is at least some data in file, we cleared ATTR_SIZE
1179 * above to avoid invoking vmtruncate, otherwise it is important
1180 * to call vmtruncate in inode_setattr to update inode->i_size
1182 rc = inode_setattr(inode, &op_data->op_attr);
1184 /* Extract epoch data if obtained. */
1185 op_data->op_handle = md.body->handle;
1186 op_data->op_ioepoch = md.body->ioepoch;
1188 ll_update_inode(inode, &md);
1189 ptlrpc_req_finished(request);
1194 /* Close IO epoch and send Size-on-MDS attribute update. */
1195 static int ll_setattr_done_writing(struct inode *inode,
1196 struct md_op_data *op_data,
1197 struct md_open_data *mod)
1199 struct ll_inode_info *lli = ll_i2info(inode);
1203 LASSERT(op_data != NULL);
1204 if (!S_ISREG(inode->i_mode))
1207 CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1208 op_data->op_ioepoch, PFID(&lli->lli_fid));
1210 op_data->op_flags = MF_EPOCH_CLOSE | MF_SOM_CHANGE;
1211 rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1212 if (rc == -EAGAIN) {
1213 /* MDS has instructed us to obtain Size-on-MDS attribute
1214 * from OSTs and send setattr to back to MDS. */
1215 rc = ll_sizeonmds_update(inode, &op_data->op_handle,
1216 op_data->op_ioepoch);
1218 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1224 static int ll_setattr_do_truncate(struct inode *inode, loff_t size)
1226 struct obd_capa *capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1229 rc = cl_setattr_do_truncate(inode, size, capa);
1230 ll_truncate_free_capa(capa);
1234 static int ll_setattr_ost(struct inode *inode)
1236 struct obd_capa *capa = ll_mdscapa_get(inode);
1239 rc = cl_setattr_ost(inode, capa);
1245 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1246 * object(s) determine the file size and mtime. Otherwise, the MDS will
1247 * keep these values until such a time that objects are allocated for it.
1248 * We do the MDS operations first, as it is checking permissions for us.
1249 * We don't to the MDS RPC if there is nothing that we want to store there,
1250 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1251 * going to do an RPC anyways.
1253 * If we are doing a truncate, we will send the mtime and ctime updates
1254 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1255 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1258 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1260 struct ll_inode_info *lli = ll_i2info(inode);
1261 struct lov_stripe_md *lsm = lli->lli_smd;
1262 struct md_op_data *op_data = NULL;
1263 struct md_open_data *mod = NULL;
1264 int ia_valid = attr->ia_valid;
1265 int rc = 0, rc1 = 0;
1268 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1270 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_SETATTR, 1);
1272 if (ia_valid & ATTR_SIZE) {
1273 if (attr->ia_size > ll_file_maxbytes(inode)) {
1274 CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1275 attr->ia_size, ll_file_maxbytes(inode));
1279 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1282 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1283 if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1284 if (current->fsuid != inode->i_uid &&
1285 !cfs_capable(CFS_CAP_FOWNER))
1289 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1290 if (attr->ia_valid & ATTR_CTIME) {
1291 attr->ia_ctime = CURRENT_TIME;
1292 attr->ia_valid |= ATTR_CTIME_SET;
1294 if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1295 attr->ia_atime = CURRENT_TIME;
1296 attr->ia_valid |= ATTR_ATIME_SET;
1298 if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1299 attr->ia_mtime = CURRENT_TIME;
1300 attr->ia_valid |= ATTR_MTIME_SET;
1302 if ((attr->ia_valid & ATTR_CTIME) && !(attr->ia_valid & ATTR_MTIME)) {
1303 /* To avoid stale mtime on mds, obtain it from ost and send
1305 rc = cl_glimpse_size(inode);
1309 attr->ia_valid |= ATTR_MTIME_SET | ATTR_MTIME;
1310 attr->ia_mtime = inode->i_mtime;
1313 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1314 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1315 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1316 cfs_time_current_sec());
1318 /* NB: ATTR_SIZE will only be set after this point if the size
1319 * resides on the MDS, ie, this file has no objects. */
1321 attr->ia_valid &= ~ATTR_SIZE;
1323 /* We always do an MDS RPC, even if we're only changing the size;
1324 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1326 OBD_ALLOC_PTR(op_data);
1327 if (op_data == NULL)
1330 memcpy(&op_data->op_attr, attr, sizeof(*attr));
1332 /* Open epoch for truncate. */
1333 if ((ll_i2mdexp(inode)->exp_connect_flags & OBD_CONNECT_SOM) &&
1334 (ia_valid & ATTR_SIZE))
1335 op_data->op_flags = MF_EPOCH_OPEN;
1337 rc = ll_md_setattr(inode, op_data, &mod);
1341 if (op_data->op_ioepoch)
1342 CDEBUG(D_INODE, "Epoch "LPU64" opened on "DFID" for "
1343 "truncate\n", op_data->op_ioepoch, PFID(&lli->lli_fid));
1345 if (!lsm || !S_ISREG(inode->i_mode)) {
1346 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1350 if (ia_valid & ATTR_SIZE)
1351 rc = ll_setattr_do_truncate(inode, attr->ia_size);
1352 else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
1353 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1354 inode->i_ino, LTIME_S(attr->ia_mtime));
1355 rc = ll_setattr_ost(inode);
1360 if (op_data->op_ioepoch)
1361 rc1 = ll_setattr_done_writing(inode, op_data, mod);
1362 ll_finish_md_op_data(op_data);
1364 return rc ? rc : rc1;
1367 int ll_setattr(struct dentry *de, struct iattr *attr)
1369 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1370 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1371 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1373 if ((de->d_inode->i_mode & S_ISUID) &&
1374 !(attr->ia_mode & S_ISUID) &&
1375 !(attr->ia_valid & ATTR_KILL_SUID))
1376 attr->ia_valid |= ATTR_KILL_SUID;
1378 if (((de->d_inode->i_mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1379 !(attr->ia_mode & S_ISGID) &&
1380 !(attr->ia_valid & ATTR_KILL_SGID))
1381 attr->ia_valid |= ATTR_KILL_SGID;
1383 return ll_setattr_raw(de->d_inode, attr);
1386 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1387 __u64 max_age, __u32 flags)
1389 struct ll_sb_info *sbi = ll_s2sbi(sb);
1390 struct obd_statfs obd_osfs;
1394 rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age, flags);
1396 CERROR("md_statfs fails: rc = %d\n", rc);
1400 osfs->os_type = sb->s_magic;
1402 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1403 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1405 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1406 flags |= OBD_STATFS_NODELAY;
1408 rc = obd_statfs_rqset(class_exp2obd(sbi->ll_dt_exp),
1409 &obd_osfs, max_age, flags);
1411 CERROR("obd_statfs fails: rc = %d\n", rc);
1415 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1416 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1419 osfs->os_bsize = obd_osfs.os_bsize;
1420 osfs->os_blocks = obd_osfs.os_blocks;
1421 osfs->os_bfree = obd_osfs.os_bfree;
1422 osfs->os_bavail = obd_osfs.os_bavail;
1424 /* If we don't have as many objects free on the OST as inodes
1425 * on the MDS, we reduce the total number of inodes to
1426 * compensate, so that the "inodes in use" number is correct.
1428 if (obd_osfs.os_ffree < osfs->os_ffree) {
1429 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1431 osfs->os_ffree = obd_osfs.os_ffree;
1436 #ifndef HAVE_STATFS_DENTRY_PARAM
1437 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1440 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1442 struct super_block *sb = de->d_sb;
1444 struct obd_statfs osfs;
1447 CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1448 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1450 /* For now we will always get up-to-date statfs values, but in the
1451 * future we may allow some amount of caching on the client (e.g.
1452 * from QOS or lprocfs updates). */
1453 rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - 1, 0);
1457 statfs_unpack(sfs, &osfs);
1459 /* We need to downshift for all 32-bit kernels, because we can't
1460 * tell if the kernel is being called via sys_statfs64() or not.
1461 * Stop before overflowing f_bsize - in which case it is better
1462 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1463 if (sizeof(long) < 8) {
1464 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1467 osfs.os_blocks >>= 1;
1468 osfs.os_bfree >>= 1;
1469 osfs.os_bavail >>= 1;
1473 sfs->f_blocks = osfs.os_blocks;
1474 sfs->f_bfree = osfs.os_bfree;
1475 sfs->f_bavail = osfs.os_bavail;
1480 void ll_inode_size_lock(struct inode *inode, int lock_lsm)
1482 struct ll_inode_info *lli;
1483 struct lov_stripe_md *lsm;
1485 lli = ll_i2info(inode);
1486 LASSERT(lli->lli_size_sem_owner != current);
1487 down(&lli->lli_size_sem);
1488 LASSERT(lli->lli_size_sem_owner == NULL);
1489 lli->lli_size_sem_owner = current;
1491 LASSERTF(lsm != NULL || lock_lsm == 0, "lsm %p, lock_lsm %d\n",
1494 lov_stripe_lock(lsm);
1497 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm)
1499 struct ll_inode_info *lli;
1500 struct lov_stripe_md *lsm;
1502 lli = ll_i2info(inode);
1504 LASSERTF(lsm != NULL || unlock_lsm == 0, "lsm %p, lock_lsm %d\n",
1507 lov_stripe_unlock(lsm);
1508 LASSERT(lli->lli_size_sem_owner == current);
1509 lli->lli_size_sem_owner = NULL;
1510 up(&lli->lli_size_sem);
1513 static void ll_replace_lsm(struct inode *inode, struct lov_stripe_md *lsm)
1515 struct ll_inode_info *lli = ll_i2info(inode);
1517 dump_lsm(D_INODE, lsm);
1518 dump_lsm(D_INODE, lli->lli_smd);
1519 LASSERTF(lsm->lsm_magic == LOV_MAGIC_JOIN,
1520 "lsm must be joined lsm %p\n", lsm);
1521 obd_free_memmd(ll_i2dtexp(inode), &lli->lli_smd);
1522 CDEBUG(D_INODE, "replace lsm %p to lli_smd %p for inode %lu%u(%p)\n",
1523 lsm, lli->lli_smd, inode->i_ino, inode->i_generation, inode);
1525 lli->lli_maxbytes = lsm->lsm_maxbytes;
1526 if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1527 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1530 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1532 struct ll_inode_info *lli = ll_i2info(inode);
1533 struct mdt_body *body = md->body;
1534 struct lov_stripe_md *lsm = md->lsm;
1535 struct ll_sb_info *sbi = ll_i2sbi(inode);
1537 LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1539 if (lli->lli_smd == NULL) {
1540 if (lsm->lsm_magic != LOV_MAGIC_V1 &&
1541 lsm->lsm_magic != LOV_MAGIC_V3 &&
1542 lsm->lsm_magic != LOV_MAGIC_JOIN) {
1543 dump_lsm(D_ERROR, lsm);
1546 CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1547 lsm, inode->i_ino, inode->i_generation, inode);
1548 cl_inode_init(inode, md);
1549 /* ll_inode_size_lock() requires it is only
1550 * called with lli_smd != NULL or lock_lsm == 0
1551 * or we can race between lock/unlock.
1554 lli->lli_maxbytes = lsm->lsm_maxbytes;
1555 if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1556 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1558 if (lli->lli_smd->lsm_magic == lsm->lsm_magic &&
1559 lli->lli_smd->lsm_stripe_count ==
1560 lsm->lsm_stripe_count) {
1561 if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
1562 CERROR("lsm mismatch for inode %ld\n",
1564 CERROR("lli_smd:\n");
1565 dump_lsm(D_ERROR, lli->lli_smd);
1567 dump_lsm(D_ERROR, lsm);
1571 cl_inode_init(inode, md);
1572 ll_replace_lsm(inode, lsm);
1575 if (lli->lli_smd != lsm)
1576 obd_free_memmd(ll_i2dtexp(inode), &lsm);
1579 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1580 if (body->valid & OBD_MD_FLRMTPERM)
1581 ll_update_remote_perm(inode, md->remote_perm);
1583 #ifdef CONFIG_FS_POSIX_ACL
1584 else if (body->valid & OBD_MD_FLACL) {
1585 spin_lock(&lli->lli_lock);
1586 if (lli->lli_posix_acl)
1587 posix_acl_release(lli->lli_posix_acl);
1588 lli->lli_posix_acl = md->posix_acl;
1589 spin_unlock(&lli->lli_lock);
1592 inode->i_ino = ll_fid_build_ino(sbi, &body->fid1);
1593 inode->i_generation = ll_fid_build_gen(sbi, &body->fid1);
1595 if (body->valid & OBD_MD_FLATIME &&
1596 body->atime > LTIME_S(inode->i_atime))
1597 LTIME_S(inode->i_atime) = body->atime;
1599 /* mtime is always updated with ctime, but can be set in past.
1600 As write and utime(2) may happen within 1 second, and utime's
1601 mtime has a priority over write's one, so take mtime from mds
1602 for the same ctimes. */
1603 if (body->valid & OBD_MD_FLCTIME &&
1604 body->ctime >= LTIME_S(inode->i_ctime)) {
1605 LTIME_S(inode->i_ctime) = body->ctime;
1606 if (body->valid & OBD_MD_FLMTIME) {
1607 CDEBUG(D_INODE, "setting ino %lu mtime "
1608 "from %lu to "LPU64"\n", inode->i_ino,
1609 LTIME_S(inode->i_mtime), body->mtime);
1610 LTIME_S(inode->i_mtime) = body->mtime;
1613 if (body->valid & OBD_MD_FLMODE)
1614 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1615 if (body->valid & OBD_MD_FLTYPE)
1616 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1617 LASSERT(inode->i_mode != 0);
1618 if (S_ISREG(inode->i_mode)) {
1619 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1621 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1623 #ifdef HAVE_INODE_BLKSIZE
1624 inode->i_blksize = 1<<inode->i_blkbits;
1626 if (body->valid & OBD_MD_FLUID)
1627 inode->i_uid = body->uid;
1628 if (body->valid & OBD_MD_FLGID)
1629 inode->i_gid = body->gid;
1630 if (body->valid & OBD_MD_FLFLAGS)
1631 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1632 if (body->valid & OBD_MD_FLNLINK)
1633 inode->i_nlink = body->nlink;
1634 if (body->valid & OBD_MD_FLRDEV)
1635 inode->i_rdev = old_decode_dev(body->rdev);
1637 if (body->valid & OBD_MD_FLID) {
1638 /* FID shouldn't be changed! */
1639 if (fid_is_sane(&lli->lli_fid)) {
1640 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1641 "Trying to change FID "DFID
1642 " to the "DFID", inode %lu/%u(%p)\n",
1643 PFID(&lli->lli_fid), PFID(&body->fid1),
1644 inode->i_ino, inode->i_generation, inode);
1646 lli->lli_fid = body->fid1;
1649 LASSERT(fid_seq(&lli->lli_fid) != 0);
1651 if (body->valid & OBD_MD_FLSIZE) {
1652 if ((ll_i2mdexp(inode)->exp_connect_flags & OBD_CONNECT_SOM) &&
1653 S_ISREG(inode->i_mode) && lli->lli_smd) {
1654 struct lustre_handle lockh;
1657 /* As it is possible a blocking ast has been processed
1658 * by this time, we need to check there is an UPDATE
1659 * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1661 mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1664 if (lli->lli_flags & (LLIF_DONE_WRITING |
1665 LLIF_EPOCH_PENDING |
1667 CERROR("ino %lu flags %lu still has "
1668 "size authority! do not trust "
1669 "the size got from MDS\n",
1670 inode->i_ino, lli->lli_flags);
1672 /* Use old size assignment to avoid
1673 * deadlock bz14138 & bz14326 */
1674 inode->i_size = body->size;
1675 lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1677 ldlm_lock_decref(&lockh, mode);
1680 /* Use old size assignment to avoid
1681 * deadlock bz14138 & bz14326 */
1682 inode->i_size = body->size;
1685 if (body->valid & OBD_MD_FLBLOCKS)
1686 inode->i_blocks = body->blocks;
1689 if (body->valid & OBD_MD_FLMDSCAPA) {
1690 LASSERT(md->mds_capa);
1691 ll_add_capa(inode, md->mds_capa);
1693 if (body->valid & OBD_MD_FLOSSCAPA) {
1694 LASSERT(md->oss_capa);
1695 ll_add_capa(inode, md->oss_capa);
1699 static struct backing_dev_info ll_backing_dev_info = {
1700 .ra_pages = 0, /* No readahead */
1701 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,12))
1702 .capabilities = 0, /* Does contribute to dirty memory */
1704 .memory_backed = 0, /* Does contribute to dirty memory */
1708 void ll_read_inode2(struct inode *inode, void *opaque)
1710 struct lustre_md *md = opaque;
1711 struct ll_inode_info *lli = ll_i2info(inode);
1714 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n",
1715 inode->i_ino, inode->i_generation, inode);
1719 LASSERT(!lli->lli_smd);
1721 /* Core attributes from the MDS first. This is a new inode, and
1722 * the VFS doesn't zero times in the core inode so we have to do
1723 * it ourselves. They will be overwritten by either MDS or OST
1724 * attributes - we just need to make sure they aren't newer. */
1725 LTIME_S(inode->i_mtime) = 0;
1726 LTIME_S(inode->i_atime) = 0;
1727 LTIME_S(inode->i_ctime) = 0;
1729 ll_update_inode(inode, md);
1731 /* OIDEBUG(inode); */
1733 if (S_ISREG(inode->i_mode)) {
1734 struct ll_sb_info *sbi = ll_i2sbi(inode);
1735 inode->i_op = &ll_file_inode_operations;
1736 inode->i_fop = sbi->ll_fop;
1737 inode->i_mapping->a_ops = &ll_aops;
1739 } else if (S_ISDIR(inode->i_mode)) {
1740 inode->i_op = &ll_dir_inode_operations;
1741 inode->i_fop = &ll_dir_operations;
1742 inode->i_mapping->a_ops = &ll_dir_aops;
1744 } else if (S_ISLNK(inode->i_mode)) {
1745 inode->i_op = &ll_fast_symlink_inode_operations;
1748 inode->i_op = &ll_special_inode_operations;
1750 init_special_inode(inode, inode->i_mode,
1751 kdev_t_to_nr(inode->i_rdev));
1753 /* initializing backing dev info. */
1754 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
1760 void ll_delete_inode(struct inode *inode)
1762 struct ll_sb_info *sbi = ll_i2sbi(inode);
1766 rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
1768 CERROR("fid_delete() failed, rc %d\n", rc);
1770 truncate_inode_pages(&inode->i_data, 0);
1776 int ll_iocontrol(struct inode *inode, struct file *file,
1777 unsigned int cmd, unsigned long arg)
1779 struct ll_sb_info *sbi = ll_i2sbi(inode);
1780 struct ptlrpc_request *req = NULL;
1785 case EXT3_IOC_GETFLAGS: {
1786 struct mdt_body *body;
1787 struct obd_capa *oc;
1789 oc = ll_mdscapa_get(inode);
1790 rc = md_getattr(sbi->ll_md_exp, ll_inode2fid(inode), oc,
1791 OBD_MD_FLFLAGS, 0, &req);
1794 CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1798 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1800 flags = body->flags;
1802 ptlrpc_req_finished(req);
1804 RETURN(put_user(flags, (int *)arg));
1806 case EXT3_IOC_SETFLAGS: {
1807 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1808 struct obd_info oinfo = { { { 0 } } };
1809 struct md_op_data *op_data;
1811 if (get_user(flags, (int *)arg))
1815 OBDO_ALLOC(oinfo.oi_oa);
1819 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1820 LUSTRE_OPC_ANY, NULL);
1821 if (IS_ERR(op_data))
1822 RETURN(PTR_ERR(op_data));
1824 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1825 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1826 rc = md_setattr(sbi->ll_md_exp, op_data,
1827 NULL, 0, NULL, 0, &req, NULL);
1828 ll_finish_md_op_data(op_data);
1829 ptlrpc_req_finished(req);
1830 if (rc || lsm == NULL) {
1831 OBDO_FREE(oinfo.oi_oa);
1835 oinfo.oi_oa->o_id = lsm->lsm_object_id;
1836 oinfo.oi_oa->o_gr = lsm->lsm_object_gr;
1837 oinfo.oi_oa->o_flags = flags;
1838 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1840 oinfo.oi_capa = ll_mdscapa_get(inode);
1842 obdo_from_inode(oinfo.oi_oa, inode,
1843 OBD_MD_FLFID | OBD_MD_FLGENER);
1844 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1845 capa_put(oinfo.oi_capa);
1846 OBDO_FREE(oinfo.oi_oa);
1848 if (rc != -EPERM && rc != -EACCES)
1849 CERROR("md_setattr_async fails: rc = %d\n", rc);
1853 inode->i_flags = ll_ext_to_inode_flags(flags |
1854 MDS_BFLAG_EXT_FLAGS);
1864 int ll_flush_ctx(struct inode *inode)
1866 struct ll_sb_info *sbi = ll_i2sbi(inode);
1868 CDEBUG(D_SEC, "flush context for user %d\n", current->uid);
1870 obd_set_info_async(sbi->ll_md_exp,
1871 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1873 obd_set_info_async(sbi->ll_dt_exp,
1874 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1879 /* umount -f client means force down, don't save state */
1880 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1881 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
1883 struct super_block *sb = vfsmnt->mnt_sb;
1885 void ll_umount_begin(struct super_block *sb)
1888 struct lustre_sb_info *lsi = s2lsi(sb);
1889 struct ll_sb_info *sbi = ll_s2sbi(sb);
1890 struct obd_device *obd;
1891 struct obd_ioctl_data ioc_data = { 0 };
1894 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1895 if (!(flags & MNT_FORCE)) {
1901 /* Tell the MGC we got umount -f */
1902 lsi->lsi_flags |= LSI_UMOUNT_FORCE;
1904 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1905 sb->s_count, atomic_read(&sb->s_active));
1907 obd = class_exp2obd(sbi->ll_md_exp);
1909 CERROR("Invalid MDC connection handle "LPX64"\n",
1910 sbi->ll_md_exp->exp_handle.h_cookie);
1915 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp, sizeof ioc_data,
1918 obd = class_exp2obd(sbi->ll_dt_exp);
1920 CERROR("Invalid LOV connection handle "LPX64"\n",
1921 sbi->ll_dt_exp->exp_handle.h_cookie);
1927 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp, sizeof ioc_data,
1930 /* Really, we'd like to wait until there are no requests outstanding,
1931 * and then continue. For now, we just invalidate the requests,
1932 * schedule() and sleep one second if needed, and hope.
1935 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1936 if (atomic_read(&vfsmnt->mnt_count) > 2) {
1937 cfs_schedule_timeout(CFS_TASK_INTERRUPTIBLE,
1938 cfs_time_seconds(1));
1939 if (atomic_read(&vfsmnt->mnt_count) > 2)
1940 LCONSOLE_WARN("Mount still busy with %d refs! You "
1941 "may try to umount it a bit later\n",
1942 atomic_read(&vfsmnt->mnt_count));
1949 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
1951 struct ll_sb_info *sbi = ll_s2sbi(sb);
1955 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
1956 read_only = *flags & MS_RDONLY;
1957 err = obd_set_info_async(sbi->ll_md_exp,
1958 sizeof(KEY_READ_ONLY),
1959 KEY_READ_ONLY, sizeof(read_only),
1962 CERROR("Failed to change the read-only flag during "
1963 "remount: %d\n", err);
1968 sb->s_flags |= MS_RDONLY;
1970 sb->s_flags &= ~MS_RDONLY;
1975 int ll_prep_inode(struct inode **inode,
1976 struct ptlrpc_request *req,
1977 struct super_block *sb)
1979 struct ll_sb_info *sbi = NULL;
1980 struct lustre_md md;
1984 LASSERT(*inode || sb);
1985 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
1986 prune_deathrow(sbi, 1);
1987 memset(&md, 0, sizeof(struct lustre_md));
1989 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
1990 sbi->ll_md_exp, &md);
1995 ll_update_inode(*inode, &md);
1997 LASSERT(sb != NULL);
2000 * At this point server returns to client's same fid as client
2001 * generated for creating. So using ->fid1 is okay here.
2003 LASSERT(fid_is_sane(&md.body->fid1));
2005 *inode = ll_iget(sb, ll_fid_build_ino(sbi, &md.body->fid1),&md);
2006 if (*inode == NULL || IS_ERR(*inode)) {
2008 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2009 #ifdef CONFIG_FS_POSIX_ACL
2011 posix_acl_release(md.posix_acl);
2012 md.posix_acl = NULL;
2015 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2017 CERROR("new_inode -fatal: rc %d\n", rc);
2022 rc = obd_checkmd(sbi->ll_dt_exp, sbi->ll_md_exp,
2023 ll_i2info(*inode)->lli_smd);
2025 md_free_lustre_md(sbi->ll_md_exp, &md);
2029 int ll_obd_statfs(struct inode *inode, void *arg)
2031 struct ll_sb_info *sbi = NULL;
2032 struct obd_export *exp;
2034 struct obd_ioctl_data *data = NULL;
2038 if (!inode || !(sbi = ll_i2sbi(inode)))
2039 GOTO(out_statfs, rc = -EINVAL);
2041 rc = obd_ioctl_getdata(&buf, &len, arg);
2043 GOTO(out_statfs, rc);
2046 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2047 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2048 GOTO(out_statfs, rc = -EINVAL);
2050 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2051 if (type == LL_STATFS_MDC)
2052 exp = sbi->ll_md_exp;
2053 else if (type == LL_STATFS_LOV)
2054 exp = sbi->ll_dt_exp;
2056 GOTO(out_statfs, rc = -ENODEV);
2058 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2060 GOTO(out_statfs, rc);
2063 obd_ioctl_freedata(buf, len);
2067 int ll_process_config(struct lustre_cfg *lcfg)
2071 struct lprocfs_static_vars lvars;
2075 lprocfs_llite_init_vars(&lvars);
2077 /* The instance name contains the sb: lustre-client-aacfe000 */
2078 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2079 if (!ptr || !*(++ptr))
2081 if (sscanf(ptr, "%lx", &x) != 1)
2084 /* This better be a real Lustre superblock! */
2085 LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2087 /* Note we have not called client_common_fill_super yet, so
2088 proc fns must be able to handle that! */
2089 rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2096 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2097 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2098 struct inode *i1, struct inode *i2,
2099 const char *name, int namelen,
2100 int mode, __u32 opc, void *data)
2102 LASSERT(i1 != NULL);
2104 if (namelen > ll_i2sbi(i1)->ll_namelen)
2105 return ERR_PTR(-ENAMETOOLONG);
2107 if (op_data == NULL)
2108 OBD_ALLOC_PTR(op_data);
2110 if (op_data == NULL)
2111 return ERR_PTR(-ENOMEM);
2113 ll_i2gids(op_data->op_suppgids, i1, i2);
2114 op_data->op_fid1 = *ll_inode2fid(i1);
2115 op_data->op_capa1 = ll_mdscapa_get(i1);
2118 op_data->op_fid2 = *ll_inode2fid(i2);
2119 op_data->op_capa2 = ll_mdscapa_get(i2);
2121 fid_zero(&op_data->op_fid2);
2122 op_data->op_capa2 = NULL;
2125 op_data->op_name = name;
2126 op_data->op_namelen = namelen;
2127 op_data->op_mode = mode;
2128 op_data->op_mod_time = cfs_time_current_sec();
2129 op_data->op_fsuid = current->fsuid;
2130 op_data->op_fsgid = current->fsgid;
2131 op_data->op_cap = cfs_curproc_cap_pack();
2132 op_data->op_bias = MDS_CHECK_SPLIT;
2133 op_data->op_opc = opc;
2134 op_data->op_mds = 0;
2135 op_data->op_data = data;
2140 void ll_finish_md_op_data(struct md_op_data *op_data)
2142 capa_put(op_data->op_capa1);
2143 capa_put(op_data->op_capa2);
2144 OBD_FREE_PTR(op_data);
2147 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2149 struct ll_sb_info *sbi;
2151 LASSERT((seq != NULL) && (vfs != NULL));
2152 sbi = ll_s2sbi(vfs->mnt_sb);
2154 if (sbi->ll_flags & LL_SBI_NOLCK)
2155 seq_puts(seq, ",nolock");
2157 if (sbi->ll_flags & LL_SBI_FLOCK)
2158 seq_puts(seq, ",flock");
2160 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2161 seq_puts(seq, ",localflock");
2163 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2164 seq_puts(seq, ",user_xattr");
2166 if (sbi->ll_flags & LL_SBI_ACL)
2167 seq_puts(seq, ",acl");
2169 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2170 seq_puts(seq, ",lazystatfs");