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 CFS_LIST_HEAD(ll_super_blocks);
63 cfs_spinlock_t ll_sb_lock = CFS_SPIN_LOCK_UNLOCKED;
65 extern struct address_space_operations ll_aops;
66 extern struct address_space_operations ll_dir_aops;
69 #define log2(n) cfs_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 cfs_spin_lock_init(&sbi->ll_lock);
86 cfs_init_mutex(&sbi->ll_lco.lco_lock);
87 cfs_spin_lock_init(&sbi->ll_pp_extent_lock);
88 cfs_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_per_file = min(pages / 32,
104 SBI_DEFAULT_READAHEAD_MAX);
105 sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
106 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
107 SBI_DEFAULT_READAHEAD_WHOLE_MAX;
108 CFS_INIT_LIST_HEAD(&sbi->ll_conn_chain);
109 CFS_INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
111 ll_generate_random_uuid(uuid);
112 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
113 CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
115 cfs_spin_lock(&ll_sb_lock);
116 cfs_list_add_tail(&sbi->ll_list, &ll_super_blocks);
117 cfs_spin_unlock(&ll_sb_lock);
119 #ifdef ENABLE_LLITE_CHECKSUM
120 sbi->ll_flags |= LL_SBI_CHECKSUM;
123 #ifdef HAVE_LRU_RESIZE_SUPPORT
124 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
127 #ifdef HAVE_EXPORT___IGET
128 CFS_INIT_LIST_HEAD(&sbi->ll_deathrow);
129 cfs_spin_lock_init(&sbi->ll_deathrow_lock);
131 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
132 cfs_spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i]. \
134 cfs_spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i]. \
138 /* metadata statahead is enabled by default */
139 sbi->ll_sa_max = LL_SA_RPC_DEF;
140 atomic_set(&sbi->ll_sa_total, 0);
141 atomic_set(&sbi->ll_sa_wrong, 0);
146 void ll_free_sbi(struct super_block *sb)
148 struct ll_sb_info *sbi = ll_s2sbi(sb);
152 cfs_spin_lock(&ll_sb_lock);
153 cfs_list_del(&sbi->ll_list);
154 cfs_spin_unlock(&ll_sb_lock);
155 OBD_FREE(sbi, sizeof(*sbi));
160 static struct dentry_operations ll_d_root_ops = {
161 .d_compare = ll_dcompare,
162 .d_revalidate = ll_revalidate_nd,
165 static int client_common_fill_super(struct super_block *sb, char *md, char *dt)
167 struct inode *root = 0;
168 struct ll_sb_info *sbi = ll_s2sbi(sb);
169 struct obd_device *obd;
170 struct obd_capa *oc = NULL;
171 struct obd_statfs osfs;
172 struct ptlrpc_request *request = NULL;
173 struct obd_connect_data *data = NULL;
174 struct obd_uuid *uuid;
175 struct md_op_data *op_data;
176 struct lustre_md lmd;
178 int size, err, checksum;
181 obd = class_name2obd(md);
183 CERROR("MD %s: not setup or attached\n", md);
191 if (proc_lustre_fs_root) {
192 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
195 CERROR("could not register mount in /proc/fs/lustre\n");
198 /* indicate the features supported by this client */
199 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
200 OBD_CONNECT_JOIN | OBD_CONNECT_ATTRFID |
201 OBD_CONNECT_VERSION | OBD_CONNECT_MDS_CAPA |
202 OBD_CONNECT_OSS_CAPA | OBD_CONNECT_CANCELSET|
203 OBD_CONNECT_FID | OBD_CONNECT_AT |
204 OBD_CONNECT_LOV_V3 | OBD_CONNECT_RMT_CLIENT |
205 OBD_CONNECT_VBR | OBD_CONNECT_SOM;
207 #ifdef HAVE_LRU_RESIZE_SUPPORT
208 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
209 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
211 #ifdef CONFIG_FS_POSIX_ACL
212 data->ocd_connect_flags |= OBD_CONNECT_ACL;
214 data->ocd_ibits_known = MDS_INODELOCK_FULL;
215 data->ocd_version = LUSTRE_VERSION_CODE;
217 if (sb->s_flags & MS_RDONLY)
218 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
219 if (sbi->ll_flags & LL_SBI_USER_XATTR)
220 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
222 #ifdef HAVE_MS_FLOCK_LOCK
223 /* force vfs to use lustre handler for flock() calls - bug 10743 */
224 sb->s_flags |= MS_FLOCK_LOCK;
227 if (sbi->ll_flags & LL_SBI_FLOCK)
228 sbi->ll_fop = &ll_file_operations_flock;
229 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
230 sbi->ll_fop = &ll_file_operations;
232 sbi->ll_fop = &ll_file_operations_noflock;
235 data->ocd_connect_flags |= OBD_CONNECT_REAL;
236 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
237 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
239 err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
241 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
242 "recovery, of which this client is not a "
243 "part. Please wait for recovery to complete,"
244 " abort, or time out.\n", md);
247 CERROR("cannot connect to %s: rc = %d\n", md, err);
251 err = obd_fid_init(sbi->ll_md_exp);
253 CERROR("Can't init metadata layer FID infrastructure, "
258 err = obd_statfs(obd, &osfs, cfs_time_current_64() - CFS_HZ, 0);
260 GOTO(out_md_fid, err);
262 size = sizeof(*data);
263 err = obd_get_info(sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
264 KEY_CONN_DATA, &size, data, NULL);
266 CERROR("Get connect data failed: %d \n", err);
270 LASSERT(osfs.os_bsize);
271 sb->s_blocksize = osfs.os_bsize;
272 sb->s_blocksize_bits = log2(osfs.os_bsize);
273 sb->s_magic = LL_SUPER_MAGIC;
275 /* for bug 11559. in $LINUX/fs/read_write.c, function do_sendfile():
276 * retval = in_file->f_op->sendfile(...);
278 * retval = -EOVERFLOW;
280 * it will check if *ppos is greater than max. However, max equals to
281 * s_maxbytes, which is a negative integer in a x86_64 box since loff_t
282 * has been defined as a signed long long integer in linux kernel. */
283 #if BITS_PER_LONG == 64
284 sb->s_maxbytes = PAGE_CACHE_MAXBYTES >> 1;
286 sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
288 sbi->ll_namelen = osfs.os_namelen;
289 sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
291 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
292 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
293 LCONSOLE_INFO("Disabling user_xattr feature because "
294 "it is not supported on the server\n");
295 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
298 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
300 sb->s_flags |= MS_POSIXACL;
302 sbi->ll_flags |= LL_SBI_ACL;
304 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
306 sb->s_flags &= ~MS_POSIXACL;
308 sbi->ll_flags &= ~LL_SBI_ACL;
311 if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
312 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
313 sbi->ll_flags |= LL_SBI_RMT_CLIENT;
314 LCONSOLE_INFO("client is set as remote by default.\n");
317 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
318 sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
319 LCONSOLE_INFO("client claims to be remote, but server "
320 "rejected, forced to be local.\n");
324 if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
325 LCONSOLE_INFO("client enabled MDS capability!\n");
326 sbi->ll_flags |= LL_SBI_MDS_CAPA;
329 if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
330 LCONSOLE_INFO("client enabled OSS capability!\n");
331 sbi->ll_flags |= LL_SBI_OSS_CAPA;
334 obd = class_name2obd(dt);
336 CERROR("DT %s: not setup or attached\n", dt);
337 GOTO(out_md_fid, err = -ENODEV);
340 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
341 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
342 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
343 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
344 OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
345 OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
348 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
349 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
350 * disabled by default, because it can still be enabled on the
351 * fly via /proc. As a consequence, we still need to come to an
352 * agreement on the supported algorithms at connect time */
353 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
355 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
356 data->ocd_cksum_types = OBD_CKSUM_ADLER;
358 /* send the list of supported checksum types */
359 data->ocd_cksum_types = OBD_CKSUM_ALL;
362 #ifdef HAVE_LRU_RESIZE_SUPPORT
363 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
365 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
366 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
368 CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
369 "ocd_grant: %d\n", data->ocd_connect_flags,
370 data->ocd_version, data->ocd_grant);
372 obd->obd_upcall.onu_owner = &sbi->ll_lco;
373 obd->obd_upcall.onu_upcall = cl_ocd_update;
374 data->ocd_brw_size = PTLRPC_MAX_BRW_PAGES << CFS_PAGE_SHIFT;
376 err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data, NULL);
378 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
379 "recovery, of which this client is not a "
380 "part. Please wait for recovery to "
381 "complete, abort, or time out.\n", dt);
382 GOTO(out_md_fid, err);
384 CERROR("Cannot connect to %s: rc = %d\n", dt, err);
385 GOTO(out_md_fid, err);
388 err = obd_fid_init(sbi->ll_dt_exp);
390 CERROR("Can't init data layer FID infrastructure, "
395 cfs_mutex_down(&sbi->ll_lco.lco_lock);
396 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
397 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
398 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
399 cfs_mutex_up(&sbi->ll_lco.lco_lock);
401 fid_zero(&sbi->ll_root_fid);
402 err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
404 CERROR("cannot mds_connect: rc = %d\n", err);
405 GOTO(out_lock_cn_cb, err);
407 if (!fid_is_sane(&sbi->ll_root_fid)) {
408 CERROR("Invalid root fid during mount\n");
409 GOTO(out_lock_cn_cb, err = -EINVAL);
411 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
413 sb->s_op = &lustre_super_operations;
414 sb->s_export_op = &lustre_export_operations;
417 * XXX: move this to after cbd setup? */
418 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
419 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
420 valid |= OBD_MD_FLRMTPERM;
421 else if (sbi->ll_flags & LL_SBI_ACL)
422 valid |= OBD_MD_FLACL;
424 OBD_ALLOC_PTR(op_data);
426 GOTO(out_lock_cn_cb, err = -ENOMEM);
428 op_data->op_fid1 = sbi->ll_root_fid;
429 op_data->op_mode = 0;
430 op_data->op_capa1 = oc;
431 op_data->op_valid = valid;
433 err = md_getattr(sbi->ll_md_exp, op_data, &request);
436 OBD_FREE_PTR(op_data);
438 CERROR("md_getattr failed for root: rc = %d\n", err);
439 GOTO(out_lock_cn_cb, err);
441 memset(&lmd, 0, sizeof(lmd));
442 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
443 sbi->ll_md_exp, &lmd);
445 CERROR("failed to understand root inode md: rc = %d\n", err);
446 ptlrpc_req_finished (request);
447 GOTO(out_lock_cn_cb, err);
450 LASSERT(fid_is_sane(&sbi->ll_root_fid));
451 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid), &lmd);
452 md_free_lustre_md(sbi->ll_md_exp, &lmd);
453 ptlrpc_req_finished(request);
455 if (root == NULL || IS_ERR(root)) {
457 obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
458 #ifdef CONFIG_FS_POSIX_ACL
460 posix_acl_release(lmd.posix_acl);
461 lmd.posix_acl = NULL;
464 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
466 CERROR("lustre_lite: bad iget4 for root\n");
470 err = ll_close_thread_start(&sbi->ll_lcq);
472 CERROR("cannot start close thread: rc %d\n", err);
476 #ifdef CONFIG_FS_POSIX_ACL
477 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
478 rct_init(&sbi->ll_rct);
479 et_init(&sbi->ll_et);
483 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
484 err = obd_set_info_async(sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
485 KEY_CHECKSUM, sizeof(checksum), &checksum,
489 sb->s_root = d_alloc_root(root);
491 OBD_FREE(data, sizeof(*data));
493 sb->s_root->d_op = &ll_d_root_ops;
495 sbi->ll_sdev_orig = sb->s_dev;
497 /* We set sb->s_dev equal on all lustre clients in order to support
498 * NFS export clustering. NFSD requires that the FSID be the same
500 /* s_dev is also used in lt_compare() to compare two fs, but that is
501 * only a node-local comparison. */
502 uuid = obd_get_uuid(sbi->ll_md_exp);
504 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
511 obd_fid_fini(sbi->ll_dt_exp);
513 obd_disconnect(sbi->ll_dt_exp);
514 sbi->ll_dt_exp = NULL;
516 obd_fid_fini(sbi->ll_md_exp);
518 obd_disconnect(sbi->ll_md_exp);
519 sbi->ll_md_exp = NULL;
523 lprocfs_unregister_mountpoint(sbi);
527 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
531 *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
533 rc = obd_get_info(sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
534 KEY_MAX_EASIZE, &size, lmmsize, NULL);
536 CERROR("Get max mdsize error rc %d \n", rc);
541 void ll_dump_inode(struct inode *inode)
543 struct list_head *tmp;
544 int dentry_count = 0;
546 LASSERT(inode != NULL);
548 list_for_each(tmp, &inode->i_dentry)
551 CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
552 inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
553 inode->i_mode, atomic_read(&inode->i_count), dentry_count);
556 void lustre_dump_dentry(struct dentry *dentry, int recur)
558 struct list_head *tmp;
561 LASSERT(dentry != NULL);
563 list_for_each(tmp, &dentry->d_subdirs)
566 CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
567 " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
568 dentry->d_name.len, dentry->d_name.name,
569 dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
570 dentry->d_parent, dentry->d_inode, atomic_read(&dentry->d_count),
571 dentry->d_flags, dentry->d_fsdata, subdirs);
572 if (dentry->d_inode != NULL)
573 ll_dump_inode(dentry->d_inode);
578 list_for_each(tmp, &dentry->d_subdirs) {
579 struct dentry *d = list_entry(tmp, struct dentry, d_child);
580 lustre_dump_dentry(d, recur - 1);
584 #ifdef HAVE_EXPORT___IGET
585 static void prune_dir_dentries(struct inode *inode)
587 struct dentry *dentry, *prev = NULL;
589 /* due to lustre specific logic, a directory
590 * can have few dentries - a bug from VFS POV */
592 spin_lock(&dcache_lock);
593 if (!list_empty(&inode->i_dentry)) {
594 dentry = list_entry(inode->i_dentry.prev,
595 struct dentry, d_alias);
596 /* in order to prevent infinite loops we
597 * break if previous dentry is busy */
598 if (dentry != prev) {
601 spin_unlock(&dcache_lock);
603 /* try to kill all child dentries */
605 shrink_dcache_parent(dentry);
606 unlock_dentry(dentry);
609 /* now try to get rid of current dentry */
610 d_prune_aliases(inode);
614 spin_unlock(&dcache_lock);
617 static void prune_deathrow_one(struct ll_inode_info *lli)
619 struct inode *inode = ll_info2i(lli);
621 /* first, try to drop any dentries - they hold a ref on the inode */
622 if (S_ISDIR(inode->i_mode))
623 prune_dir_dentries(inode);
625 d_prune_aliases(inode);
628 /* if somebody still uses it, leave it */
629 LASSERT(atomic_read(&inode->i_count) > 0);
630 if (atomic_read(&inode->i_count) > 1)
633 CDEBUG(D_INODE, "inode %lu/%u(%d) looks a good candidate for prune\n",
634 inode->i_ino,inode->i_generation,
635 atomic_read(&inode->i_count));
637 /* seems nobody uses it anymore */
645 static void prune_deathrow(struct ll_sb_info *sbi, int try)
647 struct ll_inode_info *lli;
651 if (need_resched() && try)
655 if (!cfs_spin_trylock(&sbi->ll_deathrow_lock))
658 cfs_spin_lock(&sbi->ll_deathrow_lock);
663 if (!cfs_list_empty(&sbi->ll_deathrow)) {
664 lli = cfs_list_entry(sbi->ll_deathrow.next,
665 struct ll_inode_info,
667 cfs_list_del_init(&lli->lli_dead_list);
668 if (!cfs_list_empty(&sbi->ll_deathrow))
671 cfs_spin_unlock(&sbi->ll_deathrow_lock);
674 prune_deathrow_one(lli);
676 } while (empty == 0);
678 #else /* !HAVE_EXPORT___IGET */
679 #define prune_deathrow(sbi, try) do {} while (0)
680 #endif /* HAVE_EXPORT___IGET */
682 void client_common_put_super(struct super_block *sb)
684 struct ll_sb_info *sbi = ll_s2sbi(sb);
687 #ifdef CONFIG_FS_POSIX_ACL
688 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
689 et_fini(&sbi->ll_et);
690 rct_fini(&sbi->ll_rct);
694 obd_cancel_unused(sbi->ll_dt_exp, NULL, 0, NULL);
696 ll_close_thread_shutdown(sbi->ll_lcq);
700 /* destroy inodes in deathrow */
701 prune_deathrow(sbi, 0);
703 cfs_list_del(&sbi->ll_conn_chain);
705 obd_fid_fini(sbi->ll_dt_exp);
706 obd_disconnect(sbi->ll_dt_exp);
707 sbi->ll_dt_exp = NULL;
709 lprocfs_unregister_mountpoint(sbi);
711 obd_fid_fini(sbi->ll_md_exp);
712 obd_disconnect(sbi->ll_md_exp);
713 sbi->ll_md_exp = NULL;
718 void ll_kill_super(struct super_block *sb)
720 struct ll_sb_info *sbi;
725 if (!(sb->s_flags & MS_ACTIVE))
729 /* we need restore s_dev from changed for clustred NFS before put_super
730 * because new kernels have cached s_dev and change sb->s_dev in
731 * put_super not affected real removing devices */
733 sb->s_dev = sbi->ll_sdev_orig;
737 char *ll_read_opt(const char *opt, char *data)
743 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
744 if (strncmp(opt, data, strlen(opt)))
746 if ((value = strchr(data, '=')) == NULL)
750 OBD_ALLOC(retval, strlen(value) + 1);
752 CERROR("out of memory!\n");
756 memcpy(retval, value, strlen(value)+1);
757 CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
761 static inline int ll_set_opt(const char *opt, char *data, int fl)
763 if (strncmp(opt, data, strlen(opt)) != 0)
769 /* non-client-specific mount options are parsed in lmd_parse */
770 static int ll_options(char *options, int *flags)
773 char *s1 = options, *s2;
779 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
782 CDEBUG(D_SUPER, "next opt=%s\n", s1);
783 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
788 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
793 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
798 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
803 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
808 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
813 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
815 /* Ignore deprecated mount option. The client will
816 * always try to mount with ACL support, whether this
817 * is used depends on whether server supports it. */
820 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
824 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
830 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
835 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
840 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
845 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
850 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
855 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
861 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
867 s2 = strchr(s1, ',');
875 void ll_lli_init(struct ll_inode_info *lli)
877 lli->lli_inode_magic = LLI_INODE_MAGIC;
878 cfs_sema_init(&lli->lli_size_sem, 1);
879 cfs_sema_init(&lli->lli_write_sem, 1);
880 cfs_sema_init(&lli->lli_trunc_sem, 1);
882 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
883 cfs_spin_lock_init(&lli->lli_lock);
884 CFS_INIT_LIST_HEAD(&lli->lli_close_list);
885 lli->lli_inode_magic = LLI_INODE_MAGIC;
886 cfs_sema_init(&lli->lli_och_sem, 1);
887 lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
888 lli->lli_mds_exec_och = NULL;
889 lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
890 lli->lli_open_fd_exec_count = 0;
891 CFS_INIT_LIST_HEAD(&lli->lli_dead_list);
892 lli->lli_remote_perms = NULL;
893 lli->lli_rmtperm_utime = 0;
894 cfs_sema_init(&lli->lli_rmtperm_sem, 1);
895 CFS_INIT_LIST_HEAD(&lli->lli_oss_capas);
898 int ll_fill_super(struct super_block *sb)
900 struct lustre_profile *lprof;
901 struct lustre_sb_info *lsi = s2lsi(sb);
902 struct ll_sb_info *sbi;
903 char *dt = NULL, *md = NULL;
904 char *profilenm = get_profile_name(sb);
905 struct config_llog_instance cfg = {0, };
906 char ll_instance[sizeof(sb) * 2 + 1];
910 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
914 /* client additional sb info */
915 lsi->lsi_llsbi = sbi = ll_init_sbi();
917 cfs_module_put(THIS_MODULE);
921 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
925 /* Generate a string unique to this super, in case some joker tries
926 to mount the same fs at two mount points.
927 Use the address of the super itself.*/
928 sprintf(ll_instance, "%p", sb);
929 cfg.cfg_instance = ll_instance;
930 cfg.cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
932 /* set up client obds */
933 err = lustre_process_log(sb, profilenm, &cfg);
935 CERROR("Unable to process log: %d\n", err);
939 lprof = class_get_profile(profilenm);
941 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
942 " read from the MGS. Does that filesystem "
943 "exist?\n", profilenm);
944 GOTO(out_free, err = -EINVAL);
946 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
947 lprof->lp_md, lprof->lp_dt);
949 OBD_ALLOC(dt, strlen(lprof->lp_dt) +
950 strlen(ll_instance) + 2);
952 GOTO(out_free, err = -ENOMEM);
953 sprintf(dt, "%s-%s", lprof->lp_dt, ll_instance);
955 OBD_ALLOC(md, strlen(lprof->lp_md) +
956 strlen(ll_instance) + 2);
958 GOTO(out_free, err = -ENOMEM);
959 sprintf(md, "%s-%s", lprof->lp_md, ll_instance);
961 /* connections, registrations, sb setup */
962 err = client_common_fill_super(sb, md, dt);
966 OBD_FREE(md, strlen(md) + 1);
968 OBD_FREE(dt, strlen(dt) + 1);
972 LCONSOLE_WARN("Client %s has started\n", profilenm);
975 } /* ll_fill_super */
978 void lu_context_keys_dump(void);
980 void ll_put_super(struct super_block *sb)
982 struct config_llog_instance cfg;
983 char ll_instance[sizeof(sb) * 2 + 1];
984 struct obd_device *obd;
985 struct lustre_sb_info *lsi = s2lsi(sb);
986 struct ll_sb_info *sbi = ll_s2sbi(sb);
987 char *profilenm = get_profile_name(sb);
991 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
993 ll_print_capa_stat(sbi);
995 sprintf(ll_instance, "%p", sb);
996 cfg.cfg_instance = ll_instance;
997 lustre_end_log(sb, NULL, &cfg);
999 if (sbi->ll_md_exp) {
1000 obd = class_exp2obd(sbi->ll_md_exp);
1002 force = obd->obd_force;
1005 /* We need to set force before the lov_disconnect in
1006 lustre_common_put_super, since l_d cleans up osc's as well. */
1009 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1011 obd->obd_force = force;
1016 /* Only if client_common_fill_super succeeded */
1017 client_common_put_super(sb);
1021 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1022 class_manual_cleanup(obd);
1026 class_del_profile(profilenm);
1029 lsi->lsi_llsbi = NULL;
1031 lustre_common_put_super(sb);
1033 cl_env_cache_purge(~0);
1035 LCONSOLE_WARN("client %s umount complete\n", ll_instance);
1037 cfs_module_put(THIS_MODULE);
1040 } /* client_put_super */
1042 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1044 struct inode *inode = NULL;
1045 /* NOTE: we depend on atomic igrab() -bzzz */
1046 lock_res_and_lock(lock);
1047 if (lock->l_ast_data) {
1048 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1049 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1050 inode = igrab(lock->l_ast_data);
1052 inode = lock->l_ast_data;
1053 ldlm_lock_debug(NULL, inode->i_state & I_FREEING ?
1055 lock, __FILE__, __func__, __LINE__,
1056 "l_ast_data %p is bogus: magic %08x",
1057 lock->l_ast_data, lli->lli_inode_magic);
1061 unlock_res_and_lock(lock);
1065 static int null_if_equal(struct ldlm_lock *lock, void *data)
1067 if (data == lock->l_ast_data) {
1068 lock->l_ast_data = NULL;
1070 if (lock->l_req_mode != lock->l_granted_mode)
1071 LDLM_ERROR(lock,"clearing inode with ungranted lock");
1074 return LDLM_ITER_CONTINUE;
1077 void ll_clear_inode(struct inode *inode)
1079 struct ll_inode_info *lli = ll_i2info(inode);
1080 struct ll_sb_info *sbi = ll_i2sbi(inode);
1083 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1084 inode->i_generation, inode);
1086 if (S_ISDIR(inode->i_mode)) {
1087 /* these should have been cleared in ll_file_release */
1088 LASSERT(lli->lli_sai == NULL);
1089 LASSERT(lli->lli_opendir_key == NULL);
1090 LASSERT(lli->lli_opendir_pid == 0);
1093 ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1094 md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1095 null_if_equal, inode);
1097 LASSERT(!lli->lli_open_fd_write_count);
1098 LASSERT(!lli->lli_open_fd_read_count);
1099 LASSERT(!lli->lli_open_fd_exec_count);
1101 if (lli->lli_mds_write_och)
1102 ll_md_real_close(inode, FMODE_WRITE);
1103 if (lli->lli_mds_exec_och)
1104 ll_md_real_close(inode, FMODE_EXEC);
1105 if (lli->lli_mds_read_och)
1106 ll_md_real_close(inode, FMODE_READ);
1108 if (lli->lli_symlink_name) {
1109 OBD_FREE(lli->lli_symlink_name,
1110 strlen(lli->lli_symlink_name) + 1);
1111 lli->lli_symlink_name = NULL;
1114 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1115 LASSERT(lli->lli_posix_acl == NULL);
1116 if (lli->lli_remote_perms) {
1117 free_rmtperm_hash(lli->lli_remote_perms);
1118 lli->lli_remote_perms = NULL;
1121 #ifdef CONFIG_FS_POSIX_ACL
1122 else if (lli->lli_posix_acl) {
1123 LASSERT(cfs_atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1124 LASSERT(lli->lli_remote_perms == NULL);
1125 posix_acl_release(lli->lli_posix_acl);
1126 lli->lli_posix_acl = NULL;
1129 lli->lli_inode_magic = LLI_INODE_DEAD;
1131 #ifdef HAVE_EXPORT___IGET
1132 cfs_spin_lock(&sbi->ll_deathrow_lock);
1133 cfs_list_del_init(&lli->lli_dead_list);
1134 cfs_spin_unlock(&sbi->ll_deathrow_lock);
1136 ll_clear_inode_capas(inode);
1138 * XXX This has to be done before lsm is freed below, because
1139 * cl_object still uses inode lsm.
1141 cl_inode_fini(inode);
1144 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1145 lli->lli_smd = NULL;
1152 int ll_md_setattr(struct inode *inode, struct md_op_data *op_data,
1153 struct md_open_data **mod)
1155 struct lustre_md md;
1156 struct ll_sb_info *sbi = ll_i2sbi(inode);
1157 struct ptlrpc_request *request = NULL;
1161 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1162 LUSTRE_OPC_ANY, NULL);
1163 if (IS_ERR(op_data))
1164 RETURN(PTR_ERR(op_data));
1166 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1169 ptlrpc_req_finished(request);
1170 if (rc == -ENOENT) {
1172 /* Unlinked special device node? Or just a race?
1173 * Pretend we done everything. */
1174 if (!S_ISREG(inode->i_mode) &&
1175 !S_ISDIR(inode->i_mode))
1176 rc = inode_setattr(inode, &op_data->op_attr);
1177 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1178 CERROR("md_setattr fails: rc = %d\n", rc);
1183 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1184 sbi->ll_md_exp, &md);
1186 ptlrpc_req_finished(request);
1190 /* We call inode_setattr to adjust timestamps.
1191 * If there is at least some data in file, we cleared ATTR_SIZE
1192 * above to avoid invoking vmtruncate, otherwise it is important
1193 * to call vmtruncate in inode_setattr to update inode->i_size
1195 rc = inode_setattr(inode, &op_data->op_attr);
1197 /* Extract epoch data if obtained. */
1198 op_data->op_handle = md.body->handle;
1199 op_data->op_ioepoch = md.body->ioepoch;
1201 ll_update_inode(inode, &md);
1202 ptlrpc_req_finished(request);
1207 /* Close IO epoch and send Size-on-MDS attribute update. */
1208 static int ll_setattr_done_writing(struct inode *inode,
1209 struct md_op_data *op_data,
1210 struct md_open_data *mod)
1212 struct ll_inode_info *lli = ll_i2info(inode);
1216 LASSERT(op_data != NULL);
1217 if (!S_ISREG(inode->i_mode))
1220 CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1221 op_data->op_ioepoch, PFID(&lli->lli_fid));
1223 op_data->op_flags = MF_EPOCH_CLOSE;
1224 ll_done_writing_attr(inode, op_data);
1225 ll_pack_inode2opdata(inode, op_data, NULL);
1227 rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1228 if (rc == -EAGAIN) {
1229 /* MDS has instructed us to obtain Size-on-MDS attribute
1230 * from OSTs and send setattr to back to MDS. */
1231 rc = ll_som_update(inode, op_data);
1233 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1239 static int ll_setattr_do_truncate(struct inode *inode, loff_t size)
1241 struct obd_capa *capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1244 rc = cl_setattr_do_truncate(inode, size, capa);
1245 ll_truncate_free_capa(capa);
1249 static int ll_setattr_ost(struct inode *inode)
1251 struct obd_capa *capa = ll_mdscapa_get(inode);
1254 rc = cl_setattr_ost(inode, capa);
1260 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1261 * object(s) determine the file size and mtime. Otherwise, the MDS will
1262 * keep these values until such a time that objects are allocated for it.
1263 * We do the MDS operations first, as it is checking permissions for us.
1264 * We don't to the MDS RPC if there is nothing that we want to store there,
1265 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1266 * going to do an RPC anyways.
1268 * If we are doing a truncate, we will send the mtime and ctime updates
1269 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1270 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1273 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1275 struct ll_inode_info *lli = ll_i2info(inode);
1276 struct lov_stripe_md *lsm = lli->lli_smd;
1277 struct md_op_data *op_data = NULL;
1278 struct md_open_data *mod = NULL;
1279 int ia_valid = attr->ia_valid;
1280 int rc = 0, rc1 = 0;
1283 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1285 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_SETATTR, 1);
1287 if (ia_valid & ATTR_SIZE) {
1288 if (attr->ia_size > ll_file_maxbytes(inode)) {
1289 CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1290 attr->ia_size, ll_file_maxbytes(inode));
1294 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1297 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1298 if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1299 if (cfs_curproc_fsuid() != inode->i_uid &&
1300 !cfs_capable(CFS_CAP_FOWNER))
1304 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1305 if (attr->ia_valid & ATTR_CTIME) {
1306 attr->ia_ctime = CFS_CURRENT_TIME;
1307 attr->ia_valid |= ATTR_CTIME_SET;
1309 if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1310 attr->ia_atime = CFS_CURRENT_TIME;
1311 attr->ia_valid |= ATTR_ATIME_SET;
1313 if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1314 attr->ia_mtime = CFS_CURRENT_TIME;
1315 attr->ia_valid |= ATTR_MTIME_SET;
1317 if ((attr->ia_valid & ATTR_CTIME) && !(attr->ia_valid & ATTR_MTIME)) {
1318 /* To avoid stale mtime on mds, obtain it from ost and send
1320 rc = cl_glimpse_size(inode);
1324 attr->ia_valid |= ATTR_MTIME_SET | ATTR_MTIME;
1325 attr->ia_mtime = inode->i_mtime;
1328 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1329 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1330 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1331 cfs_time_current_sec());
1333 /* NB: ATTR_SIZE will only be set after this point if the size
1334 * resides on the MDS, ie, this file has no objects. */
1336 attr->ia_valid &= ~ATTR_SIZE;
1338 /* We always do an MDS RPC, even if we're only changing the size;
1339 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1341 OBD_ALLOC_PTR(op_data);
1342 if (op_data == NULL)
1345 UNLOCK_INODE_MUTEX(inode);
1346 if (ia_valid & ATTR_SIZE)
1347 UP_WRITE_I_ALLOC_SEM(inode);
1348 cfs_down(&lli->lli_trunc_sem);
1349 LOCK_INODE_MUTEX(inode);
1350 if (ia_valid & ATTR_SIZE)
1351 DOWN_WRITE_I_ALLOC_SEM(inode);
1353 memcpy(&op_data->op_attr, attr, sizeof(*attr));
1355 /* Open epoch for truncate. */
1356 if (exp_connect_som(ll_i2mdexp(inode)) && (ia_valid & ATTR_SIZE))
1357 op_data->op_flags = MF_EPOCH_OPEN;
1359 rc = ll_md_setattr(inode, op_data, &mod);
1363 ll_ioepoch_open(lli, op_data->op_ioepoch);
1364 if (!lsm || !S_ISREG(inode->i_mode)) {
1365 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1369 if (ia_valid & ATTR_SIZE)
1370 rc = ll_setattr_do_truncate(inode, attr->ia_size);
1371 else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
1372 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1373 inode->i_ino, LTIME_S(attr->ia_mtime));
1374 rc = ll_setattr_ost(inode);
1379 if (op_data->op_ioepoch)
1380 rc1 = ll_setattr_done_writing(inode, op_data, mod);
1381 ll_finish_md_op_data(op_data);
1383 cfs_up(&lli->lli_trunc_sem);
1384 return rc ? rc : rc1;
1387 int ll_setattr(struct dentry *de, struct iattr *attr)
1389 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1390 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1391 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1393 if ((de->d_inode->i_mode & S_ISUID) &&
1394 !(attr->ia_mode & S_ISUID) &&
1395 !(attr->ia_valid & ATTR_KILL_SUID))
1396 attr->ia_valid |= ATTR_KILL_SUID;
1398 if (((de->d_inode->i_mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1399 !(attr->ia_mode & S_ISGID) &&
1400 !(attr->ia_valid & ATTR_KILL_SGID))
1401 attr->ia_valid |= ATTR_KILL_SGID;
1403 return ll_setattr_raw(de->d_inode, attr);
1406 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1407 __u64 max_age, __u32 flags)
1409 struct ll_sb_info *sbi = ll_s2sbi(sb);
1410 struct obd_statfs obd_osfs;
1414 rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age, flags);
1416 CERROR("md_statfs fails: rc = %d\n", rc);
1420 osfs->os_type = sb->s_magic;
1422 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1423 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1425 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1426 flags |= OBD_STATFS_NODELAY;
1428 rc = obd_statfs_rqset(class_exp2obd(sbi->ll_dt_exp),
1429 &obd_osfs, max_age, flags);
1431 CERROR("obd_statfs fails: rc = %d\n", rc);
1435 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1436 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1439 osfs->os_bsize = obd_osfs.os_bsize;
1440 osfs->os_blocks = obd_osfs.os_blocks;
1441 osfs->os_bfree = obd_osfs.os_bfree;
1442 osfs->os_bavail = obd_osfs.os_bavail;
1444 /* If we don't have as many objects free on the OST as inodes
1445 * on the MDS, we reduce the total number of inodes to
1446 * compensate, so that the "inodes in use" number is correct.
1448 if (obd_osfs.os_ffree < osfs->os_ffree) {
1449 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1451 osfs->os_ffree = obd_osfs.os_ffree;
1456 #ifndef HAVE_STATFS_DENTRY_PARAM
1457 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1460 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1462 struct super_block *sb = de->d_sb;
1464 struct obd_statfs osfs;
1467 CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1468 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1470 /* For now we will always get up-to-date statfs values, but in the
1471 * future we may allow some amount of caching on the client (e.g.
1472 * from QOS or lprocfs updates). */
1473 rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - 1, 0);
1477 statfs_unpack(sfs, &osfs);
1479 /* We need to downshift for all 32-bit kernels, because we can't
1480 * tell if the kernel is being called via sys_statfs64() or not.
1481 * Stop before overflowing f_bsize - in which case it is better
1482 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1483 if (sizeof(long) < 8) {
1484 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1487 osfs.os_blocks >>= 1;
1488 osfs.os_bfree >>= 1;
1489 osfs.os_bavail >>= 1;
1493 sfs->f_blocks = osfs.os_blocks;
1494 sfs->f_bfree = osfs.os_bfree;
1495 sfs->f_bavail = osfs.os_bavail;
1500 void ll_inode_size_lock(struct inode *inode, int lock_lsm)
1502 struct ll_inode_info *lli;
1503 struct lov_stripe_md *lsm;
1505 lli = ll_i2info(inode);
1506 LASSERT(lli->lli_size_sem_owner != current);
1507 cfs_down(&lli->lli_size_sem);
1508 LASSERT(lli->lli_size_sem_owner == NULL);
1509 lli->lli_size_sem_owner = current;
1511 LASSERTF(lsm != NULL || lock_lsm == 0, "lsm %p, lock_lsm %d\n",
1514 lov_stripe_lock(lsm);
1517 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm)
1519 struct ll_inode_info *lli;
1520 struct lov_stripe_md *lsm;
1522 lli = ll_i2info(inode);
1524 LASSERTF(lsm != NULL || unlock_lsm == 0, "lsm %p, lock_lsm %d\n",
1527 lov_stripe_unlock(lsm);
1528 LASSERT(lli->lli_size_sem_owner == current);
1529 lli->lli_size_sem_owner = NULL;
1530 cfs_up(&lli->lli_size_sem);
1533 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1535 struct ll_inode_info *lli = ll_i2info(inode);
1536 struct mdt_body *body = md->body;
1537 struct lov_stripe_md *lsm = md->lsm;
1538 struct ll_sb_info *sbi = ll_i2sbi(inode);
1540 LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1542 if (lli->lli_smd == NULL) {
1543 if (lsm->lsm_magic != LOV_MAGIC_V1 &&
1544 lsm->lsm_magic != LOV_MAGIC_V3) {
1545 dump_lsm(D_ERROR, lsm);
1548 CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1549 lsm, inode->i_ino, inode->i_generation, inode);
1550 cl_inode_init(inode, md);
1551 /* ll_inode_size_lock() requires it is only
1552 * called with lli_smd != NULL or lock_lsm == 0
1553 * or we can race between lock/unlock.
1556 lli->lli_maxbytes = lsm->lsm_maxbytes;
1557 if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1558 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1560 LASSERT(lli->lli_smd->lsm_magic == lsm->lsm_magic &&
1561 lli->lli_smd->lsm_stripe_count ==
1562 lsm->lsm_stripe_count);
1563 if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
1564 CERROR("lsm mismatch for inode %ld\n",
1566 CERROR("lli_smd:\n");
1567 dump_lsm(D_ERROR, lli->lli_smd);
1569 dump_lsm(D_ERROR, lsm);
1573 if (lli->lli_smd != lsm)
1574 obd_free_memmd(ll_i2dtexp(inode), &lsm);
1577 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1578 if (body->valid & OBD_MD_FLRMTPERM)
1579 ll_update_remote_perm(inode, md->remote_perm);
1581 #ifdef CONFIG_FS_POSIX_ACL
1582 else if (body->valid & OBD_MD_FLACL) {
1583 cfs_spin_lock(&lli->lli_lock);
1584 if (lli->lli_posix_acl)
1585 posix_acl_release(lli->lli_posix_acl);
1586 lli->lli_posix_acl = md->posix_acl;
1587 cfs_spin_unlock(&lli->lli_lock);
1590 inode->i_ino = cl_fid_build_ino(&body->fid1);
1591 inode->i_generation = cl_fid_build_gen(&body->fid1);
1593 if (body->valid & OBD_MD_FLATIME &&
1594 body->atime > LTIME_S(inode->i_atime))
1595 LTIME_S(inode->i_atime) = body->atime;
1597 /* mtime is always updated with ctime, but can be set in past.
1598 As write and utime(2) may happen within 1 second, and utime's
1599 mtime has a priority over write's one, so take mtime from mds
1600 for the same ctimes. */
1601 if (body->valid & OBD_MD_FLCTIME &&
1602 body->ctime >= LTIME_S(inode->i_ctime)) {
1603 LTIME_S(inode->i_ctime) = body->ctime;
1604 if (body->valid & OBD_MD_FLMTIME) {
1605 CDEBUG(D_INODE, "setting ino %lu mtime "
1606 "from %lu to "LPU64"\n", inode->i_ino,
1607 LTIME_S(inode->i_mtime), body->mtime);
1608 LTIME_S(inode->i_mtime) = body->mtime;
1611 if (body->valid & OBD_MD_FLMODE)
1612 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1613 if (body->valid & OBD_MD_FLTYPE)
1614 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1615 LASSERT(inode->i_mode != 0);
1616 if (S_ISREG(inode->i_mode)) {
1617 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1619 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1621 #ifdef HAVE_INODE_BLKSIZE
1622 inode->i_blksize = 1<<inode->i_blkbits;
1624 if (body->valid & OBD_MD_FLUID)
1625 inode->i_uid = body->uid;
1626 if (body->valid & OBD_MD_FLGID)
1627 inode->i_gid = body->gid;
1628 if (body->valid & OBD_MD_FLFLAGS)
1629 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1630 if (body->valid & OBD_MD_FLNLINK)
1631 inode->i_nlink = body->nlink;
1632 if (body->valid & OBD_MD_FLRDEV)
1633 inode->i_rdev = old_decode_dev(body->rdev);
1635 if (body->valid & OBD_MD_FLID) {
1636 /* FID shouldn't be changed! */
1637 if (fid_is_sane(&lli->lli_fid)) {
1638 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1639 "Trying to change FID "DFID
1640 " to the "DFID", inode %lu/%u(%p)\n",
1641 PFID(&lli->lli_fid), PFID(&body->fid1),
1642 inode->i_ino, inode->i_generation, inode);
1644 lli->lli_fid = body->fid1;
1647 LASSERT(fid_seq(&lli->lli_fid) != 0);
1649 if (body->valid & OBD_MD_FLSIZE) {
1650 if (exp_connect_som(ll_i2mdexp(inode)) &&
1651 S_ISREG(inode->i_mode) && lli->lli_smd) {
1652 struct lustre_handle lockh;
1655 /* As it is possible a blocking ast has been processed
1656 * by this time, we need to check there is an UPDATE
1657 * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1659 mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1662 if (lli->lli_flags & (LLIF_DONE_WRITING |
1663 LLIF_EPOCH_PENDING |
1665 CERROR("ino %lu flags %lu still has "
1666 "size authority! do not trust "
1667 "the size got from MDS\n",
1668 inode->i_ino, lli->lli_flags);
1670 /* Use old size assignment to avoid
1671 * deadlock bz14138 & bz14326 */
1672 inode->i_size = body->size;
1673 lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1675 ldlm_lock_decref(&lockh, mode);
1678 /* Use old size assignment to avoid
1679 * deadlock bz14138 & bz14326 */
1680 inode->i_size = body->size;
1683 if (body->valid & OBD_MD_FLBLOCKS)
1684 inode->i_blocks = body->blocks;
1687 if (body->valid & OBD_MD_FLMDSCAPA) {
1688 LASSERT(md->mds_capa);
1689 ll_add_capa(inode, md->mds_capa);
1691 if (body->valid & OBD_MD_FLOSSCAPA) {
1692 LASSERT(md->oss_capa);
1693 ll_add_capa(inode, md->oss_capa);
1697 static struct backing_dev_info ll_backing_dev_info = {
1698 .ra_pages = 0, /* No readahead */
1699 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,12))
1700 .capabilities = 0, /* Does contribute to dirty memory */
1702 .memory_backed = 0, /* Does contribute to dirty memory */
1706 void ll_read_inode2(struct inode *inode, void *opaque)
1708 struct lustre_md *md = opaque;
1709 struct ll_inode_info *lli = ll_i2info(inode);
1712 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n",
1713 inode->i_ino, inode->i_generation, inode);
1717 LASSERT(!lli->lli_smd);
1719 /* Core attributes from the MDS first. This is a new inode, and
1720 * the VFS doesn't zero times in the core inode so we have to do
1721 * it ourselves. They will be overwritten by either MDS or OST
1722 * attributes - we just need to make sure they aren't newer. */
1723 LTIME_S(inode->i_mtime) = 0;
1724 LTIME_S(inode->i_atime) = 0;
1725 LTIME_S(inode->i_ctime) = 0;
1727 ll_update_inode(inode, md);
1729 /* OIDEBUG(inode); */
1731 if (S_ISREG(inode->i_mode)) {
1732 struct ll_sb_info *sbi = ll_i2sbi(inode);
1733 inode->i_op = &ll_file_inode_operations;
1734 inode->i_fop = sbi->ll_fop;
1735 inode->i_mapping->a_ops = &ll_aops;
1737 } else if (S_ISDIR(inode->i_mode)) {
1738 inode->i_op = &ll_dir_inode_operations;
1739 inode->i_fop = &ll_dir_operations;
1740 inode->i_mapping->a_ops = &ll_dir_aops;
1742 } else if (S_ISLNK(inode->i_mode)) {
1743 inode->i_op = &ll_fast_symlink_inode_operations;
1746 inode->i_op = &ll_special_inode_operations;
1748 init_special_inode(inode, inode->i_mode,
1749 kdev_t_to_nr(inode->i_rdev));
1751 /* initializing backing dev info. */
1752 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
1758 void ll_delete_inode(struct inode *inode)
1760 struct ll_sb_info *sbi = ll_i2sbi(inode);
1764 rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
1766 CERROR("fid_delete() failed, rc %d\n", rc);
1768 truncate_inode_pages(&inode->i_data, 0);
1774 int ll_iocontrol(struct inode *inode, struct file *file,
1775 unsigned int cmd, unsigned long arg)
1777 struct ll_sb_info *sbi = ll_i2sbi(inode);
1778 struct ptlrpc_request *req = NULL;
1783 case FSFILT_IOC_GETFLAGS: {
1784 struct mdt_body *body;
1785 struct md_op_data *op_data;
1787 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
1788 0, 0, LUSTRE_OPC_ANY,
1790 if (op_data == NULL)
1793 op_data->op_valid = OBD_MD_FLFLAGS;
1794 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
1795 ll_finish_md_op_data(op_data);
1797 CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1801 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1803 flags = body->flags;
1805 ptlrpc_req_finished(req);
1807 RETURN(put_user(flags, (int *)arg));
1809 case FSFILT_IOC_SETFLAGS: {
1810 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1811 struct obd_info oinfo = { { { 0 } } };
1812 struct md_op_data *op_data;
1814 if (get_user(flags, (int *)arg))
1818 OBDO_ALLOC(oinfo.oi_oa);
1822 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1823 LUSTRE_OPC_ANY, NULL);
1824 if (IS_ERR(op_data))
1825 RETURN(PTR_ERR(op_data));
1827 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1828 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1829 rc = md_setattr(sbi->ll_md_exp, op_data,
1830 NULL, 0, NULL, 0, &req, NULL);
1831 ll_finish_md_op_data(op_data);
1832 ptlrpc_req_finished(req);
1834 OBDO_FREE(oinfo.oi_oa);
1839 OBDO_FREE(oinfo.oi_oa);
1840 GOTO(update_cache, rc);
1843 oinfo.oi_oa->o_id = lsm->lsm_object_id;
1844 oinfo.oi_oa->o_gr = lsm->lsm_object_gr;
1845 oinfo.oi_oa->o_flags = flags;
1846 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1848 oinfo.oi_capa = ll_mdscapa_get(inode);
1850 obdo_from_inode(oinfo.oi_oa, inode,
1851 OBD_MD_FLFID | OBD_MD_FLGENER);
1852 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1853 capa_put(oinfo.oi_capa);
1854 OBDO_FREE(oinfo.oi_oa);
1856 if (rc != -EPERM && rc != -EACCES)
1857 CERROR("osc_setattr_async fails: rc = %d\n",rc);
1863 inode->i_flags = ll_ext_to_inode_flags(flags);
1873 int ll_flush_ctx(struct inode *inode)
1875 struct ll_sb_info *sbi = ll_i2sbi(inode);
1877 CDEBUG(D_SEC, "flush context for user %d\n", cfs_curproc_uid());
1879 obd_set_info_async(sbi->ll_md_exp,
1880 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1882 obd_set_info_async(sbi->ll_dt_exp,
1883 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1888 /* umount -f client means force down, don't save state */
1889 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1890 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
1892 struct super_block *sb = vfsmnt->mnt_sb;
1894 void ll_umount_begin(struct super_block *sb)
1897 struct lustre_sb_info *lsi = s2lsi(sb);
1898 struct ll_sb_info *sbi = ll_s2sbi(sb);
1899 struct obd_device *obd;
1900 struct obd_ioctl_data *ioc_data;
1903 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1904 if (!(flags & MNT_FORCE)) {
1910 /* Tell the MGC we got umount -f */
1911 lsi->lsi_flags |= LSI_UMOUNT_FORCE;
1913 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1914 sb->s_count, atomic_read(&sb->s_active));
1916 obd = class_exp2obd(sbi->ll_md_exp);
1918 CERROR("Invalid MDC connection handle "LPX64"\n",
1919 sbi->ll_md_exp->exp_handle.h_cookie);
1925 obd = class_exp2obd(sbi->ll_dt_exp);
1927 CERROR("Invalid LOV connection handle "LPX64"\n",
1928 sbi->ll_dt_exp->exp_handle.h_cookie);
1934 OBD_ALLOC_PTR(ioc_data);
1936 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
1937 sizeof ioc_data, ioc_data, NULL);
1939 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
1940 sizeof ioc_data, ioc_data, NULL);
1942 OBD_FREE_PTR(ioc_data);
1946 /* Really, we'd like to wait until there are no requests outstanding,
1947 * and then continue. For now, we just invalidate the requests,
1948 * schedule() and sleep one second if needed, and hope.
1951 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1952 if (atomic_read(&vfsmnt->mnt_count) > 2) {
1953 cfs_schedule_timeout_and_set_state(CFS_TASK_INTERRUPTIBLE,
1954 cfs_time_seconds(1));
1955 if (atomic_read(&vfsmnt->mnt_count) > 2)
1956 LCONSOLE_WARN("Mount still busy with %d refs! You "
1957 "may try to umount it a bit later\n",
1958 atomic_read(&vfsmnt->mnt_count));
1965 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
1967 struct ll_sb_info *sbi = ll_s2sbi(sb);
1971 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
1972 read_only = *flags & MS_RDONLY;
1973 err = obd_set_info_async(sbi->ll_md_exp,
1974 sizeof(KEY_READ_ONLY),
1975 KEY_READ_ONLY, sizeof(read_only),
1978 CERROR("Failed to change the read-only flag during "
1979 "remount: %d\n", err);
1984 sb->s_flags |= MS_RDONLY;
1986 sb->s_flags &= ~MS_RDONLY;
1991 int ll_prep_inode(struct inode **inode,
1992 struct ptlrpc_request *req,
1993 struct super_block *sb)
1995 struct ll_sb_info *sbi = NULL;
1996 struct lustre_md md;
2000 LASSERT(*inode || sb);
2001 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2002 prune_deathrow(sbi, 1);
2003 memset(&md, 0, sizeof(struct lustre_md));
2005 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2006 sbi->ll_md_exp, &md);
2011 ll_update_inode(*inode, &md);
2013 LASSERT(sb != NULL);
2016 * At this point server returns to client's same fid as client
2017 * generated for creating. So using ->fid1 is okay here.
2019 LASSERT(fid_is_sane(&md.body->fid1));
2021 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1), &md);
2022 if (*inode == NULL || IS_ERR(*inode)) {
2024 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2025 #ifdef CONFIG_FS_POSIX_ACL
2027 posix_acl_release(md.posix_acl);
2028 md.posix_acl = NULL;
2031 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2033 CERROR("new_inode -fatal: rc %d\n", rc);
2039 md_free_lustre_md(sbi->ll_md_exp, &md);
2043 int ll_obd_statfs(struct inode *inode, void *arg)
2045 struct ll_sb_info *sbi = NULL;
2046 struct obd_export *exp;
2048 struct obd_ioctl_data *data = NULL;
2052 if (!inode || !(sbi = ll_i2sbi(inode)))
2053 GOTO(out_statfs, rc = -EINVAL);
2055 rc = obd_ioctl_getdata(&buf, &len, arg);
2057 GOTO(out_statfs, rc);
2060 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2061 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2062 GOTO(out_statfs, rc = -EINVAL);
2064 if (data->ioc_inllen1 != sizeof(__u32) ||
2065 data->ioc_inllen2 != sizeof(__u32) ||
2066 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2067 data->ioc_plen2 != sizeof(struct obd_uuid))
2068 GOTO(out_statfs, rc = -EINVAL);
2070 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2071 if (type == LL_STATFS_MDC)
2072 exp = sbi->ll_md_exp;
2073 else if (type == LL_STATFS_LOV)
2074 exp = sbi->ll_dt_exp;
2076 GOTO(out_statfs, rc = -ENODEV);
2078 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2080 GOTO(out_statfs, rc);
2083 obd_ioctl_freedata(buf, len);
2087 int ll_process_config(struct lustre_cfg *lcfg)
2091 struct lprocfs_static_vars lvars;
2095 lprocfs_llite_init_vars(&lvars);
2097 /* The instance name contains the sb: lustre-client-aacfe000 */
2098 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2099 if (!ptr || !*(++ptr))
2101 if (sscanf(ptr, "%lx", &x) != 1)
2104 /* This better be a real Lustre superblock! */
2105 LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2107 /* Note we have not called client_common_fill_super yet, so
2108 proc fns must be able to handle that! */
2109 rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2116 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2117 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2118 struct inode *i1, struct inode *i2,
2119 const char *name, int namelen,
2120 int mode, __u32 opc, void *data)
2122 LASSERT(i1 != NULL);
2124 if (namelen > ll_i2sbi(i1)->ll_namelen)
2125 return ERR_PTR(-ENAMETOOLONG);
2127 if (op_data == NULL)
2128 OBD_ALLOC_PTR(op_data);
2130 if (op_data == NULL)
2131 return ERR_PTR(-ENOMEM);
2133 ll_i2gids(op_data->op_suppgids, i1, i2);
2134 op_data->op_fid1 = *ll_inode2fid(i1);
2135 op_data->op_capa1 = ll_mdscapa_get(i1);
2138 op_data->op_fid2 = *ll_inode2fid(i2);
2139 op_data->op_capa2 = ll_mdscapa_get(i2);
2141 fid_zero(&op_data->op_fid2);
2142 op_data->op_capa2 = NULL;
2145 op_data->op_name = name;
2146 op_data->op_namelen = namelen;
2147 op_data->op_mode = mode;
2148 op_data->op_mod_time = cfs_time_current_sec();
2149 op_data->op_fsuid = cfs_curproc_fsuid();
2150 op_data->op_fsgid = cfs_curproc_fsgid();
2151 op_data->op_cap = cfs_curproc_cap_pack();
2152 op_data->op_bias = MDS_CHECK_SPLIT;
2153 op_data->op_opc = opc;
2154 op_data->op_mds = 0;
2155 op_data->op_data = data;
2160 void ll_finish_md_op_data(struct md_op_data *op_data)
2162 capa_put(op_data->op_capa1);
2163 capa_put(op_data->op_capa2);
2164 OBD_FREE_PTR(op_data);
2167 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2169 struct ll_sb_info *sbi;
2171 LASSERT((seq != NULL) && (vfs != NULL));
2172 sbi = ll_s2sbi(vfs->mnt_sb);
2174 if (sbi->ll_flags & LL_SBI_NOLCK)
2175 seq_puts(seq, ",nolock");
2177 if (sbi->ll_flags & LL_SBI_FLOCK)
2178 seq_puts(seq, ",flock");
2180 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2181 seq_puts(seq, ",localflock");
2183 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2184 seq_puts(seq, ",user_xattr");
2186 if (sbi->ll_flags & LL_SBI_ACL)
2187 seq_puts(seq, ",acl");
2189 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2190 seq_puts(seq, ",lazystatfs");