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 /* Initialize the default and maximum LOV EA and cookie sizes. This allows
162 * us to make MDS RPCs with large enough reply buffers to hold the
163 * maximum-sized (= maximum striped) EA and cookie without having to
164 * calculate this (via a call into the LOV + OSCs) each time we make an RPC. */
165 static int ll_init_ea_size(struct obd_export *md_exp, struct obd_export *dt_exp)
167 struct lov_stripe_md lsm = { .lsm_magic = LOV_MAGIC_V3 };
168 __u32 valsize = sizeof(struct lov_desc);
169 int rc, easize, def_easize, cookiesize;
170 struct lov_desc desc;
174 rc = obd_get_info(dt_exp, sizeof(KEY_LOVDESC), KEY_LOVDESC,
175 &valsize, &desc, NULL);
179 stripes = min(desc.ld_tgt_count, (__u32)LOV_MAX_STRIPE_COUNT);
180 lsm.lsm_stripe_count = stripes;
181 easize = obd_size_diskmd(dt_exp, &lsm);
183 lsm.lsm_stripe_count = desc.ld_default_stripe_count;
184 def_easize = obd_size_diskmd(dt_exp, &lsm);
186 cookiesize = stripes * sizeof(struct llog_cookie);
188 CDEBUG(D_HA, "updating max_mdsize/max_cookiesize: %d/%d\n",
191 rc = md_init_ea_size(md_exp, easize, def_easize, cookiesize);
195 static int client_common_fill_super(struct super_block *sb, char *md, char *dt)
197 struct inode *root = 0;
198 struct ll_sb_info *sbi = ll_s2sbi(sb);
199 struct obd_device *obd;
200 struct obd_capa *oc = NULL;
201 struct obd_statfs osfs;
202 struct ptlrpc_request *request = NULL;
203 struct lustre_handle dt_conn = {0, };
204 struct lustre_handle md_conn = {0, };
205 struct obd_connect_data *data = NULL;
206 struct obd_uuid *uuid;
207 struct lustre_md lmd;
209 int size, err, checksum;
212 obd = class_name2obd(md);
214 CERROR("MD %s: not setup or attached\n", md);
222 if (proc_lustre_fs_root) {
223 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
226 CERROR("could not register mount in /proc/fs/lustre\n");
229 /* indicate the features supported by this client */
230 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
231 OBD_CONNECT_JOIN | OBD_CONNECT_ATTRFID |
232 OBD_CONNECT_VERSION | OBD_CONNECT_MDS_CAPA |
233 OBD_CONNECT_OSS_CAPA | OBD_CONNECT_CANCELSET|
234 OBD_CONNECT_FID | OBD_CONNECT_AT |
235 OBD_CONNECT_LOV_V3 | OBD_CONNECT_RMT_CLIENT;
237 #ifdef HAVE_LRU_RESIZE_SUPPORT
238 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
239 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
241 #ifdef CONFIG_FS_POSIX_ACL
242 data->ocd_connect_flags |= OBD_CONNECT_ACL;
244 data->ocd_ibits_known = MDS_INODELOCK_FULL;
245 data->ocd_version = LUSTRE_VERSION_CODE;
247 if (sb->s_flags & MS_RDONLY)
248 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
249 if (sbi->ll_flags & LL_SBI_USER_XATTR)
250 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
252 #ifdef HAVE_MS_FLOCK_LOCK
253 /* force vfs to use lustre handler for flock() calls - bug 10743 */
254 sb->s_flags |= MS_FLOCK_LOCK;
257 if (sbi->ll_flags & LL_SBI_FLOCK)
258 sbi->ll_fop = &ll_file_operations_flock;
259 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
260 sbi->ll_fop = &ll_file_operations;
262 sbi->ll_fop = &ll_file_operations_noflock;
265 data->ocd_connect_flags |= OBD_CONNECT_REAL;
266 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
267 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
269 err = obd_connect(NULL, &md_conn, obd, &sbi->ll_sb_uuid, data, NULL);
271 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
272 "recovery, of which this client is not a "
273 "part. Please wait for recovery to complete,"
274 " abort, or time out.\n", md);
277 CERROR("cannot connect to %s: rc = %d\n", md, err);
280 sbi->ll_md_exp = class_conn2export(&md_conn);
282 err = obd_fid_init(sbi->ll_md_exp);
284 CERROR("Can't init metadata layer FID infrastructure, "
289 err = obd_statfs(obd, &osfs, cfs_time_current_64() - HZ, 0);
291 GOTO(out_md_fid, err);
293 size = sizeof(*data);
294 err = obd_get_info(sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
295 KEY_CONN_DATA, &size, data, NULL);
297 CERROR("Get connect data failed: %d \n", err);
301 LASSERT(osfs.os_bsize);
302 sb->s_blocksize = osfs.os_bsize;
303 sb->s_blocksize_bits = log2(osfs.os_bsize);
304 sb->s_magic = LL_SUPER_MAGIC;
306 /* for bug 11559. in $LINUX/fs/read_write.c, function do_sendfile():
307 * retval = in_file->f_op->sendfile(...);
309 * retval = -EOVERFLOW;
311 * it will check if *ppos is greater than max. However, max equals to
312 * s_maxbytes, which is a negative integer in a x86_64 box since loff_t
313 * has been defined as a signed long long ineger in linux kernel. */
314 #if BITS_PER_LONG == 64
315 sb->s_maxbytes = PAGE_CACHE_MAXBYTES >> 1;
317 sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
319 sbi->ll_namelen = osfs.os_namelen;
320 sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
322 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
323 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
324 LCONSOLE_INFO("Disabling user_xattr feature because "
325 "it is not supported on the server\n");
326 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
329 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
331 sb->s_flags |= MS_POSIXACL;
333 sbi->ll_flags |= LL_SBI_ACL;
335 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
337 sb->s_flags &= ~MS_POSIXACL;
339 sbi->ll_flags &= ~LL_SBI_ACL;
342 if (data->ocd_connect_flags & OBD_CONNECT_JOIN)
343 sbi->ll_flags |= LL_SBI_JOIN;
345 if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
346 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
347 sbi->ll_flags |= LL_SBI_RMT_CLIENT;
348 LCONSOLE_INFO("client is set as remote by default.\n");
351 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
352 sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
353 LCONSOLE_INFO("client claims to be remote, but server "
354 "rejected, forced to be local.\n");
358 if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
359 LCONSOLE_INFO("client enabled MDS capability!\n");
360 sbi->ll_flags |= LL_SBI_MDS_CAPA;
363 if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
364 LCONSOLE_INFO("client enabled OSS capability!\n");
365 sbi->ll_flags |= LL_SBI_OSS_CAPA;
368 obd = class_name2obd(dt);
370 CERROR("DT %s: not setup or attached\n", dt);
371 GOTO(out_md_fid, err = -ENODEV);
374 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
375 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
376 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
377 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
378 OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
379 OBD_CONNECT_OSS_CAPA;
381 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
382 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
383 * disabled by default, because it can still be enabled on the
384 * fly via /proc. As a consequence, we still need to come to an
385 * agreement on the supported algorithms at connect time */
386 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
388 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
389 data->ocd_cksum_types = OBD_CKSUM_ADLER;
391 /* send the list of supported checksum types */
392 data->ocd_cksum_types = OBD_CKSUM_ALL;
395 #ifdef HAVE_LRU_RESIZE_SUPPORT
396 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
398 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
399 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
401 CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
402 "ocd_grant: %d\n", data->ocd_connect_flags,
403 data->ocd_version, data->ocd_grant);
405 obd->obd_upcall.onu_owner = &sbi->ll_lco;
406 obd->obd_upcall.onu_upcall = ll_ocd_update;
407 data->ocd_brw_size = PTLRPC_MAX_BRW_PAGES << CFS_PAGE_SHIFT;
409 err = obd_connect(NULL, &dt_conn, obd, &sbi->ll_sb_uuid, data, NULL);
411 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
412 "recovery, of which this client is not a "
413 "part. Please wait for recovery to "
414 "complete, abort, or time out.\n", dt);
415 GOTO(out_md_fid, err);
417 CERROR("Cannot connect to %s: rc = %d\n", dt, err);
418 GOTO(out_md_fid, err);
421 sbi->ll_dt_exp = class_conn2export(&dt_conn);
423 err = obd_fid_init(sbi->ll_dt_exp);
425 CERROR("Can't init data layer FID infrastructure, "
430 spin_lock(&sbi->ll_lco.lco_lock);
431 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
432 spin_unlock(&sbi->ll_lco.lco_lock);
434 err = ll_init_ea_size(sbi->ll_md_exp, sbi->ll_dt_exp);;
436 CERROR("cannot set max EA and cookie sizes: rc = %d\n", err);
437 GOTO(out_lock_cn_cb, err);
440 fid_zero(&sbi->ll_root_fid);
441 err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
443 CERROR("cannot mds_connect: rc = %d\n", err);
444 GOTO(out_lock_cn_cb, err);
446 if (!fid_is_sane(&sbi->ll_root_fid)) {
447 CERROR("Invalid root fid during mount\n");
448 GOTO(out_lock_cn_cb, err = -EINVAL);
450 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
452 sb->s_op = &lustre_super_operations;
453 sb->s_export_op = &lustre_export_operations;
456 * XXX: move this to after cbd setup? */
457 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
458 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
459 valid |= OBD_MD_FLRMTPERM;
460 else if (sbi->ll_flags & LL_SBI_ACL)
461 valid |= OBD_MD_FLACL;
463 err = md_getattr(sbi->ll_md_exp, &sbi->ll_root_fid, oc, valid, 0,
468 CERROR("md_getattr failed for root: rc = %d\n", err);
469 GOTO(out_lock_cn_cb, err);
471 memset(&lmd, 0, sizeof(lmd));
472 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
473 sbi->ll_md_exp, &lmd);
475 CERROR("failed to understand root inode md: rc = %d\n", err);
476 ptlrpc_req_finished (request);
477 GOTO(out_lock_cn_cb, err);
480 LASSERT(fid_is_sane(&sbi->ll_root_fid));
481 root = ll_iget(sb, ll_fid_build_ino(sbi, &sbi->ll_root_fid), &lmd);
482 md_free_lustre_md(sbi->ll_md_exp, &lmd);
483 ptlrpc_req_finished(request);
485 if (root == NULL || IS_ERR(root)) {
487 obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
488 #ifdef CONFIG_FS_POSIX_ACL
490 posix_acl_release(lmd.posix_acl);
491 lmd.posix_acl = NULL;
494 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
496 CERROR("lustre_lite: bad iget4 for root\n");
500 err = ll_close_thread_start(&sbi->ll_lcq);
502 CERROR("cannot start close thread: rc %d\n", err);
506 #ifdef CONFIG_FS_POSIX_ACL
507 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
508 rct_init(&sbi->ll_rct);
509 et_init(&sbi->ll_et);
513 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
514 err = obd_set_info_async(sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
515 KEY_CHECKSUM, sizeof(checksum), &checksum,
519 sb->s_root = d_alloc_root(root);
521 OBD_FREE(data, sizeof(*data));
523 sb->s_root->d_op = &ll_d_root_ops;
525 sbi->ll_sdev_orig = sb->s_dev;
527 /* We set sb->s_dev equal on all lustre clients in order to support
528 * NFS export clustering. NFSD requires that the FSID be the same
530 /* s_dev is also used in lt_compare() to compare two fs, but that is
531 * only a node-local comparison. */
532 uuid = obd_get_uuid(sbi->ll_md_exp);
534 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
541 obd_fid_fini(sbi->ll_dt_exp);
543 obd_disconnect(sbi->ll_dt_exp);
544 sbi->ll_dt_exp = NULL;
546 obd_fid_fini(sbi->ll_md_exp);
548 obd_disconnect(sbi->ll_md_exp);
549 sbi->ll_md_exp = NULL;
553 lprocfs_unregister_mountpoint(sbi);
557 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
561 *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
563 rc = obd_get_info(sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
564 KEY_MAX_EASIZE, &size, lmmsize, NULL);
566 CERROR("Get max mdsize error rc %d \n", rc);
571 void ll_dump_inode(struct inode *inode)
573 struct list_head *tmp;
574 int dentry_count = 0;
576 LASSERT(inode != NULL);
578 list_for_each(tmp, &inode->i_dentry)
581 CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
582 inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
583 inode->i_mode, atomic_read(&inode->i_count), dentry_count);
586 void lustre_dump_dentry(struct dentry *dentry, int recur)
588 struct list_head *tmp;
591 LASSERT(dentry != NULL);
593 list_for_each(tmp, &dentry->d_subdirs)
596 CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
597 " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
598 dentry->d_name.len, dentry->d_name.name,
599 dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
600 dentry->d_parent, dentry->d_inode, atomic_read(&dentry->d_count),
601 dentry->d_flags, dentry->d_fsdata, subdirs);
602 if (dentry->d_inode != NULL)
603 ll_dump_inode(dentry->d_inode);
608 list_for_each(tmp, &dentry->d_subdirs) {
609 struct dentry *d = list_entry(tmp, struct dentry, d_child);
610 lustre_dump_dentry(d, recur - 1);
614 #ifdef HAVE_EXPORT___IGET
615 static void prune_dir_dentries(struct inode *inode)
617 struct dentry *dentry, *prev = NULL;
619 /* due to lustre specific logic, a directory
620 * can have few dentries - a bug from VFS POV */
622 spin_lock(&dcache_lock);
623 if (!list_empty(&inode->i_dentry)) {
624 dentry = list_entry(inode->i_dentry.prev,
625 struct dentry, d_alias);
626 /* in order to prevent infinite loops we
627 * break if previous dentry is busy */
628 if (dentry != prev) {
631 spin_unlock(&dcache_lock);
633 /* try to kill all child dentries */
635 shrink_dcache_parent(dentry);
636 unlock_dentry(dentry);
639 /* now try to get rid of current dentry */
640 d_prune_aliases(inode);
644 spin_unlock(&dcache_lock);
647 static void prune_deathrow_one(struct ll_inode_info *lli)
649 struct inode *inode = ll_info2i(lli);
651 /* first, try to drop any dentries - they hold a ref on the inode */
652 if (S_ISDIR(inode->i_mode))
653 prune_dir_dentries(inode);
655 d_prune_aliases(inode);
658 /* if somebody still uses it, leave it */
659 LASSERT(atomic_read(&inode->i_count) > 0);
660 if (atomic_read(&inode->i_count) > 1)
663 CDEBUG(D_INODE, "inode %lu/%u(%d) looks a good candidate for prune\n",
664 inode->i_ino,inode->i_generation, atomic_read(&inode->i_count));
666 /* seems nobody uses it anymore */
674 static void prune_deathrow(struct ll_sb_info *sbi, int try)
676 struct ll_inode_info *lli;
680 if (need_resched() && try)
684 if (!spin_trylock(&sbi->ll_deathrow_lock))
687 spin_lock(&sbi->ll_deathrow_lock);
692 if (!list_empty(&sbi->ll_deathrow)) {
693 lli = list_entry(sbi->ll_deathrow.next,
694 struct ll_inode_info,
696 list_del_init(&lli->lli_dead_list);
697 if (!list_empty(&sbi->ll_deathrow))
700 spin_unlock(&sbi->ll_deathrow_lock);
703 prune_deathrow_one(lli);
705 } while (empty == 0);
707 #else /* !HAVE_EXPORT___IGET */
708 #define prune_deathrow(sbi, try) do {} while (0)
709 #endif /* HAVE_EXPORT___IGET */
711 void client_common_put_super(struct super_block *sb)
713 struct ll_sb_info *sbi = ll_s2sbi(sb);
716 #ifdef CONFIG_FS_POSIX_ACL
717 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
718 et_fini(&sbi->ll_et);
719 rct_fini(&sbi->ll_rct);
723 obd_cancel_unused(sbi->ll_dt_exp, NULL, 0, NULL);
725 ll_close_thread_shutdown(sbi->ll_lcq);
727 /* destroy inodes in deathrow */
728 prune_deathrow(sbi, 0);
730 list_del(&sbi->ll_conn_chain);
732 obd_fid_fini(sbi->ll_dt_exp);
733 obd_disconnect(sbi->ll_dt_exp);
734 sbi->ll_dt_exp = NULL;
736 lprocfs_unregister_mountpoint(sbi);
738 obd_fid_fini(sbi->ll_md_exp);
739 obd_disconnect(sbi->ll_md_exp);
740 sbi->ll_md_exp = NULL;
745 void ll_kill_super(struct super_block *sb)
747 struct ll_sb_info *sbi;
752 if (!(sb->s_flags & MS_ACTIVE))
756 /* we need restore s_dev from changed for clustred NFS before put_super
757 * because new kernels have cached s_dev and change sb->s_dev in
758 * put_super not affected real removing devices */
760 sb->s_dev = sbi->ll_sdev_orig;
764 char *ll_read_opt(const char *opt, char *data)
770 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
771 if (strncmp(opt, data, strlen(opt)))
773 if ((value = strchr(data, '=')) == NULL)
777 OBD_ALLOC(retval, strlen(value) + 1);
779 CERROR("out of memory!\n");
783 memcpy(retval, value, strlen(value)+1);
784 CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
788 static inline int ll_set_opt(const char *opt, char *data, int fl)
790 if (strncmp(opt, data, strlen(opt)) != 0)
796 /* non-client-specific mount options are parsed in lmd_parse */
797 static int ll_options(char *options, int *flags)
800 char *s1 = options, *s2;
806 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
809 CDEBUG(D_SUPER, "next opt=%s\n", s1);
810 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
815 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
820 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
825 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
830 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
835 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
840 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
842 /* Ignore deprecated mount option. The client will
843 * always try to mount with ACL support, whether this
844 * is used depends on whether server supports it. */
847 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
851 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
857 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
862 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
867 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
872 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
878 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
884 s2 = strchr(s1, ',');
892 void ll_lli_init(struct ll_inode_info *lli)
894 lli->lli_inode_magic = LLI_INODE_MAGIC;
895 sema_init(&lli->lli_size_sem, 1);
896 sema_init(&lli->lli_write_sem, 1);
898 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
899 spin_lock_init(&lli->lli_lock);
900 INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
901 INIT_LIST_HEAD(&lli->lli_close_list);
902 lli->lli_inode_magic = LLI_INODE_MAGIC;
903 sema_init(&lli->lli_och_sem, 1);
904 lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
905 lli->lli_mds_exec_och = NULL;
906 lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
907 lli->lli_open_fd_exec_count = 0;
908 INIT_LIST_HEAD(&lli->lli_dead_list);
909 lli->lli_remote_perms = NULL;
910 lli->lli_rmtperm_utime = 0;
911 sema_init(&lli->lli_rmtperm_sem, 1);
912 INIT_LIST_HEAD(&lli->lli_oss_capas);
915 int ll_fill_super(struct super_block *sb)
917 struct lustre_profile *lprof;
918 struct lustre_sb_info *lsi = s2lsi(sb);
919 struct ll_sb_info *sbi;
920 char *dt = NULL, *md = NULL;
921 char *profilenm = get_profile_name(sb);
922 struct config_llog_instance cfg = {0, };
923 char ll_instance[sizeof(sb) * 2 + 1];
927 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
931 /* client additional sb info */
932 lsi->lsi_llsbi = sbi = ll_init_sbi();
938 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
942 /* Generate a string unique to this super, in case some joker tries
943 to mount the same fs at two mount points.
944 Use the address of the super itself.*/
945 sprintf(ll_instance, "%p", sb);
946 cfg.cfg_instance = ll_instance;
947 cfg.cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
949 /* set up client obds */
950 err = lustre_process_log(sb, profilenm, &cfg);
952 CERROR("Unable to process log: %d\n", err);
956 lprof = class_get_profile(profilenm);
958 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
959 " read from the MGS. Does that filesystem "
960 "exist?\n", profilenm);
961 GOTO(out_free, err = -EINVAL);
963 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
964 lprof->lp_md, lprof->lp_dt);
966 OBD_ALLOC(dt, strlen(lprof->lp_dt) +
967 strlen(ll_instance) + 2);
969 GOTO(out_free, err = -ENOMEM);
970 sprintf(dt, "%s-%s", lprof->lp_dt, ll_instance);
972 OBD_ALLOC(md, strlen(lprof->lp_md) +
973 strlen(ll_instance) + 2);
975 GOTO(out_free, err = -ENOMEM);
976 sprintf(md, "%s-%s", lprof->lp_md, ll_instance);
978 /* connections, registrations, sb setup */
979 err = client_common_fill_super(sb, md, dt);
983 OBD_FREE(md, strlen(md) + 1);
985 OBD_FREE(dt, strlen(dt) + 1);
989 LCONSOLE_WARN("Client %s has started\n", profilenm);
992 } /* ll_fill_super */
995 void lu_context_keys_dump(void);
997 void ll_put_super(struct super_block *sb)
999 struct config_llog_instance cfg;
1000 char ll_instance[sizeof(sb) * 2 + 1];
1001 struct obd_device *obd;
1002 struct lustre_sb_info *lsi = s2lsi(sb);
1003 struct ll_sb_info *sbi = ll_s2sbi(sb);
1004 char *profilenm = get_profile_name(sb);
1005 int force = 1, next;
1008 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1010 ll_print_capa_stat(sbi);
1012 sprintf(ll_instance, "%p", sb);
1013 cfg.cfg_instance = ll_instance;
1014 lustre_end_log(sb, NULL, &cfg);
1016 if (sbi->ll_md_exp) {
1017 obd = class_exp2obd(sbi->ll_md_exp);
1019 force = obd->obd_force;
1022 /* We need to set force before the lov_disconnect in
1023 lustre_common_put_super, since l_d cleans up osc's as well. */
1026 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1028 obd->obd_force = force;
1035 /* Only if client_common_fill_super succeeded */
1036 client_common_put_super(sb);
1039 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1040 class_manual_cleanup(obd);
1044 class_del_profile(profilenm);
1047 lsi->lsi_llsbi = NULL;
1049 lustre_common_put_super(sb);
1051 cl_env_cache_purge(~0);
1052 lu_context_keys_dump();
1054 LCONSOLE_WARN("client %s umount complete\n", ll_instance);
1059 } /* client_put_super */
1061 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1063 struct inode *inode = NULL;
1064 /* NOTE: we depend on atomic igrab() -bzzz */
1065 lock_res_and_lock(lock);
1066 if (lock->l_ast_data) {
1067 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1068 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1069 inode = igrab(lock->l_ast_data);
1071 inode = lock->l_ast_data;
1072 ldlm_lock_debug(NULL, inode->i_state & I_FREEING ?
1074 lock, __FILE__, __func__, __LINE__,
1075 "l_ast_data %p is bogus: magic %08x",
1076 lock->l_ast_data, lli->lli_inode_magic);
1080 unlock_res_and_lock(lock);
1084 static int null_if_equal(struct ldlm_lock *lock, void *data)
1086 if (data == lock->l_ast_data) {
1087 lock->l_ast_data = NULL;
1089 if (lock->l_req_mode != lock->l_granted_mode)
1090 LDLM_ERROR(lock,"clearing inode with ungranted lock");
1093 return LDLM_ITER_CONTINUE;
1096 void ll_clear_inode(struct inode *inode)
1098 struct ll_inode_info *lli = ll_i2info(inode);
1099 struct ll_sb_info *sbi = ll_i2sbi(inode);
1102 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1103 inode->i_generation, inode);
1105 if (S_ISDIR(inode->i_mode)) {
1106 /* these should have been cleared in ll_file_release */
1107 LASSERT(lli->lli_sai == NULL);
1108 LASSERT(lli->lli_opendir_key == NULL);
1109 LASSERT(lli->lli_opendir_pid == 0);
1112 ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1113 md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1114 null_if_equal, inode);
1116 LASSERT(!lli->lli_open_fd_write_count);
1117 LASSERT(!lli->lli_open_fd_read_count);
1118 LASSERT(!lli->lli_open_fd_exec_count);
1120 if (lli->lli_mds_write_och)
1121 ll_md_real_close(inode, FMODE_WRITE);
1122 if (lli->lli_mds_exec_och)
1123 ll_md_real_close(inode, FMODE_EXEC);
1124 if (lli->lli_mds_read_och)
1125 ll_md_real_close(inode, FMODE_READ);
1127 if (lli->lli_symlink_name) {
1128 OBD_FREE(lli->lli_symlink_name,
1129 strlen(lli->lli_symlink_name) + 1);
1130 lli->lli_symlink_name = NULL;
1133 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1134 LASSERT(lli->lli_posix_acl == NULL);
1135 if (lli->lli_remote_perms) {
1136 free_rmtperm_hash(lli->lli_remote_perms);
1137 lli->lli_remote_perms = NULL;
1140 #ifdef CONFIG_FS_POSIX_ACL
1141 else if (lli->lli_posix_acl) {
1142 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1143 LASSERT(lli->lli_remote_perms == NULL);
1144 posix_acl_release(lli->lli_posix_acl);
1145 lli->lli_posix_acl = NULL;
1148 lli->lli_inode_magic = LLI_INODE_DEAD;
1150 #ifdef HAVE_EXPORT___IGET
1151 spin_lock(&sbi->ll_deathrow_lock);
1152 list_del_init(&lli->lli_dead_list);
1153 spin_unlock(&sbi->ll_deathrow_lock);
1155 ll_clear_inode_capas(inode);
1157 * XXX This has to be done before lsm is freed below, because
1158 * cl_object still uses inode lsm.
1160 cl_inode_fini(inode);
1163 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1164 lli->lli_smd = NULL;
1171 int ll_md_setattr(struct inode *inode, struct md_op_data *op_data,
1172 struct md_open_data **mod)
1174 struct lustre_md md;
1175 struct ll_sb_info *sbi = ll_i2sbi(inode);
1176 struct ptlrpc_request *request = NULL;
1180 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1181 LUSTRE_OPC_ANY, NULL);
1182 if (IS_ERR(op_data))
1183 RETURN(PTR_ERR(op_data));
1185 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1188 ptlrpc_req_finished(request);
1189 if (rc == -ENOENT) {
1191 /* Unlinked special device node? Or just a race?
1192 * Pretend we done everything. */
1193 if (!S_ISREG(inode->i_mode) &&
1194 !S_ISDIR(inode->i_mode))
1195 rc = inode_setattr(inode, &op_data->op_attr);
1196 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1197 CERROR("md_setattr fails: rc = %d\n", rc);
1202 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1203 sbi->ll_md_exp, &md);
1205 ptlrpc_req_finished(request);
1209 /* We call inode_setattr to adjust timestamps.
1210 * If there is at least some data in file, we cleared ATTR_SIZE
1211 * above to avoid invoking vmtruncate, otherwise it is important
1212 * to call vmtruncate in inode_setattr to update inode->i_size
1214 rc = inode_setattr(inode, &op_data->op_attr);
1216 /* Extract epoch data if obtained. */
1217 op_data->op_handle = md.body->handle;
1218 op_data->op_ioepoch = md.body->ioepoch;
1220 ll_update_inode(inode, &md);
1221 ptlrpc_req_finished(request);
1226 /* Close IO epoch and send Size-on-MDS attribute update. */
1227 static int ll_setattr_done_writing(struct inode *inode,
1228 struct md_op_data *op_data,
1229 struct md_open_data *mod)
1231 struct ll_inode_info *lli = ll_i2info(inode);
1235 LASSERT(op_data != NULL);
1236 if (!S_ISREG(inode->i_mode))
1239 CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1240 op_data->op_ioepoch, PFID(&lli->lli_fid));
1242 op_data->op_flags = MF_EPOCH_CLOSE | MF_SOM_CHANGE;
1243 rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1244 if (rc == -EAGAIN) {
1245 /* MDS has instructed us to obtain Size-on-MDS attribute
1246 * from OSTs and send setattr to back to MDS. */
1247 rc = ll_sizeonmds_update(inode, mod, &op_data->op_handle,
1248 op_data->op_ioepoch);
1250 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1256 static int ll_setattr_do_truncate(struct inode *inode, loff_t size)
1258 struct obd_capa *capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1261 rc = cl_setattr_do_truncate(inode, size, capa);
1262 ll_truncate_free_capa(capa);
1266 static int ll_setattr_ost(struct inode *inode)
1268 struct obd_capa *capa = ll_mdscapa_get(inode);
1271 rc = cl_setattr_ost(inode, capa);
1277 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1278 * object(s) determine the file size and mtime. Otherwise, the MDS will
1279 * keep these values until such a time that objects are allocated for it.
1280 * We do the MDS operations first, as it is checking permissions for us.
1281 * We don't to the MDS RPC if there is nothing that we want to store there,
1282 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1283 * going to do an RPC anyways.
1285 * If we are doing a truncate, we will send the mtime and ctime updates
1286 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1287 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1290 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1292 struct ll_inode_info *lli = ll_i2info(inode);
1293 struct lov_stripe_md *lsm = lli->lli_smd;
1294 struct md_op_data *op_data = NULL;
1295 struct md_open_data *mod = NULL;
1296 int ia_valid = attr->ia_valid;
1297 int rc = 0, rc1 = 0;
1300 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1302 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_SETATTR, 1);
1304 if (ia_valid & ATTR_SIZE) {
1305 if (attr->ia_size > ll_file_maxbytes(inode)) {
1306 CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1307 attr->ia_size, ll_file_maxbytes(inode));
1311 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1314 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1315 if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1316 if (current->fsuid != inode->i_uid &&
1317 !cfs_capable(CFS_CAP_FOWNER))
1321 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1322 if (attr->ia_valid & ATTR_CTIME) {
1323 attr->ia_ctime = CURRENT_TIME;
1324 attr->ia_valid |= ATTR_CTIME_SET;
1326 if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1327 attr->ia_atime = CURRENT_TIME;
1328 attr->ia_valid |= ATTR_ATIME_SET;
1330 if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1331 attr->ia_mtime = CURRENT_TIME;
1332 attr->ia_valid |= ATTR_MTIME_SET;
1334 if ((attr->ia_valid & ATTR_CTIME) && !(attr->ia_valid & ATTR_MTIME)) {
1335 /* To avoid stale mtime on mds, obtain it from ost and send
1337 rc = cl_glimpse_size(inode);
1341 attr->ia_valid |= ATTR_MTIME_SET | ATTR_MTIME;
1342 attr->ia_mtime = inode->i_mtime;
1345 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1346 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1347 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1348 cfs_time_current_sec());
1350 /* NB: ATTR_SIZE will only be set after this point if the size
1351 * resides on the MDS, ie, this file has no objects. */
1353 attr->ia_valid &= ~ATTR_SIZE;
1355 /* We always do an MDS RPC, even if we're only changing the size;
1356 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1358 OBD_ALLOC_PTR(op_data);
1359 if (op_data == NULL)
1362 memcpy(&op_data->op_attr, attr, sizeof(*attr));
1364 /* Open epoch for truncate. */
1365 if ((ll_i2mdexp(inode)->exp_connect_flags & OBD_CONNECT_SOM) &&
1366 (ia_valid & ATTR_SIZE))
1367 op_data->op_flags = MF_EPOCH_OPEN;
1369 rc = ll_md_setattr(inode, op_data, &mod);
1373 if (op_data->op_ioepoch)
1374 CDEBUG(D_INODE, "Epoch "LPU64" opened on "DFID" for "
1375 "truncate\n", op_data->op_ioepoch, PFID(&lli->lli_fid));
1377 if (!lsm || !S_ISREG(inode->i_mode)) {
1378 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1382 if (ia_valid & ATTR_SIZE)
1383 rc = ll_setattr_do_truncate(inode, attr->ia_size);
1384 else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
1385 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1386 inode->i_ino, LTIME_S(attr->ia_mtime));
1387 rc = ll_setattr_ost(inode);
1392 if (op_data->op_ioepoch)
1393 rc1 = ll_setattr_done_writing(inode, op_data, mod);
1394 ll_finish_md_op_data(op_data);
1396 return rc ? rc : rc1;
1399 int ll_setattr(struct dentry *de, struct iattr *attr)
1401 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1402 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1403 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1405 if ((de->d_inode->i_mode & S_ISUID) &&
1406 !(attr->ia_mode & S_ISUID) &&
1407 !(attr->ia_valid & ATTR_KILL_SUID))
1408 attr->ia_valid |= ATTR_KILL_SUID;
1410 if (((de->d_inode->i_mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1411 !(attr->ia_mode & S_ISGID) &&
1412 !(attr->ia_valid & ATTR_KILL_SGID))
1413 attr->ia_valid |= ATTR_KILL_SGID;
1415 return ll_setattr_raw(de->d_inode, attr);
1418 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1419 __u64 max_age, __u32 flags)
1421 struct ll_sb_info *sbi = ll_s2sbi(sb);
1422 struct obd_statfs obd_osfs;
1426 rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age, flags);
1428 CERROR("md_statfs fails: rc = %d\n", rc);
1432 osfs->os_type = sb->s_magic;
1434 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1435 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1437 rc = obd_statfs_rqset(class_exp2obd(sbi->ll_dt_exp),
1438 &obd_osfs, max_age, flags);
1440 CERROR("obd_statfs fails: rc = %d\n", rc);
1444 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1445 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1448 osfs->os_bsize = obd_osfs.os_bsize;
1449 osfs->os_blocks = obd_osfs.os_blocks;
1450 osfs->os_bfree = obd_osfs.os_bfree;
1451 osfs->os_bavail = obd_osfs.os_bavail;
1453 /* If we don't have as many objects free on the OST as inodes
1454 * on the MDS, we reduce the total number of inodes to
1455 * compensate, so that the "inodes in use" number is correct.
1457 if (obd_osfs.os_ffree < osfs->os_ffree) {
1458 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1460 osfs->os_ffree = obd_osfs.os_ffree;
1465 #ifndef HAVE_STATFS_DENTRY_PARAM
1466 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1469 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1471 struct super_block *sb = de->d_sb;
1473 struct obd_statfs osfs;
1476 CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1477 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1479 /* For now we will always get up-to-date statfs values, but in the
1480 * future we may allow some amount of caching on the client (e.g.
1481 * from QOS or lprocfs updates). */
1482 rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - 1, 0);
1486 statfs_unpack(sfs, &osfs);
1488 /* We need to downshift for all 32-bit kernels, because we can't
1489 * tell if the kernel is being called via sys_statfs64() or not.
1490 * Stop before overflowing f_bsize - in which case it is better
1491 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1492 if (sizeof(long) < 8) {
1493 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1496 osfs.os_blocks >>= 1;
1497 osfs.os_bfree >>= 1;
1498 osfs.os_bavail >>= 1;
1502 sfs->f_blocks = osfs.os_blocks;
1503 sfs->f_bfree = osfs.os_bfree;
1504 sfs->f_bavail = osfs.os_bavail;
1509 void ll_inode_size_lock(struct inode *inode, int lock_lsm)
1511 struct ll_inode_info *lli;
1512 struct lov_stripe_md *lsm;
1514 lli = ll_i2info(inode);
1515 LASSERT(lli->lli_size_sem_owner != current);
1516 down(&lli->lli_size_sem);
1517 LASSERT(lli->lli_size_sem_owner == NULL);
1518 lli->lli_size_sem_owner = current;
1520 LASSERTF(lsm != NULL || lock_lsm == 0, "lsm %p, lock_lsm %d\n",
1523 lov_stripe_lock(lsm);
1526 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm)
1528 struct ll_inode_info *lli;
1529 struct lov_stripe_md *lsm;
1531 lli = ll_i2info(inode);
1533 LASSERTF(lsm != NULL || unlock_lsm == 0, "lsm %p, lock_lsm %d\n",
1536 lov_stripe_unlock(lsm);
1537 LASSERT(lli->lli_size_sem_owner == current);
1538 lli->lli_size_sem_owner = NULL;
1539 up(&lli->lli_size_sem);
1542 static void ll_replace_lsm(struct inode *inode, struct lov_stripe_md *lsm)
1544 struct ll_inode_info *lli = ll_i2info(inode);
1546 dump_lsm(D_INODE, lsm);
1547 dump_lsm(D_INODE, lli->lli_smd);
1548 LASSERTF(lsm->lsm_magic == LOV_MAGIC_JOIN,
1549 "lsm must be joined lsm %p\n", lsm);
1550 obd_free_memmd(ll_i2dtexp(inode), &lli->lli_smd);
1551 CDEBUG(D_INODE, "replace lsm %p to lli_smd %p for inode %lu%u(%p)\n",
1552 lsm, lli->lli_smd, inode->i_ino, inode->i_generation, inode);
1554 lli->lli_maxbytes = lsm->lsm_maxbytes;
1555 if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1556 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1559 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1561 struct ll_inode_info *lli = ll_i2info(inode);
1562 struct mdt_body *body = md->body;
1563 struct lov_stripe_md *lsm = md->lsm;
1564 struct ll_sb_info *sbi = ll_i2sbi(inode);
1566 LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1568 if (lli->lli_smd == NULL) {
1569 if (lsm->lsm_magic != LOV_MAGIC_V1 &&
1570 lsm->lsm_magic != LOV_MAGIC_V3 &&
1571 lsm->lsm_magic != LOV_MAGIC_JOIN) {
1572 dump_lsm(D_ERROR, lsm);
1575 CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1576 lsm, inode->i_ino, inode->i_generation, inode);
1577 cl_inode_init(inode, md);
1578 /* ll_inode_size_lock() requires it is only
1579 * called with lli_smd != NULL or lock_lsm == 0
1580 * or we can race between lock/unlock.
1583 lli->lli_maxbytes = lsm->lsm_maxbytes;
1584 if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1585 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1587 if (lli->lli_smd->lsm_magic == lsm->lsm_magic &&
1588 lli->lli_smd->lsm_stripe_count ==
1589 lsm->lsm_stripe_count) {
1590 if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
1591 CERROR("lsm mismatch for inode %ld\n",
1593 CERROR("lli_smd:\n");
1594 dump_lsm(D_ERROR, lli->lli_smd);
1596 dump_lsm(D_ERROR, lsm);
1600 cl_inode_init(inode, md);
1601 ll_replace_lsm(inode, lsm);
1604 if (lli->lli_smd != lsm)
1605 obd_free_memmd(ll_i2dtexp(inode), &lsm);
1608 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1609 if (body->valid & OBD_MD_FLRMTPERM)
1610 ll_update_remote_perm(inode, md->remote_perm);
1612 #ifdef CONFIG_FS_POSIX_ACL
1613 else if (body->valid & OBD_MD_FLACL) {
1614 spin_lock(&lli->lli_lock);
1615 if (lli->lli_posix_acl)
1616 posix_acl_release(lli->lli_posix_acl);
1617 lli->lli_posix_acl = md->posix_acl;
1618 spin_unlock(&lli->lli_lock);
1621 inode->i_ino = ll_fid_build_ino(sbi, &body->fid1);
1622 inode->i_generation = ll_fid_build_gen(sbi, &body->fid1);
1624 if (body->valid & OBD_MD_FLATIME &&
1625 body->atime > LTIME_S(inode->i_atime))
1626 LTIME_S(inode->i_atime) = body->atime;
1628 /* mtime is always updated with ctime, but can be set in past.
1629 As write and utime(2) may happen within 1 second, and utime's
1630 mtime has a priority over write's one, so take mtime from mds
1631 for the same ctimes. */
1632 if (body->valid & OBD_MD_FLCTIME &&
1633 body->ctime >= LTIME_S(inode->i_ctime)) {
1634 LTIME_S(inode->i_ctime) = body->ctime;
1635 if (body->valid & OBD_MD_FLMTIME) {
1636 CDEBUG(D_INODE, "setting ino %lu mtime "
1637 "from %lu to "LPU64"\n", inode->i_ino,
1638 LTIME_S(inode->i_mtime), body->mtime);
1639 LTIME_S(inode->i_mtime) = body->mtime;
1642 if (body->valid & OBD_MD_FLMODE)
1643 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1644 if (body->valid & OBD_MD_FLTYPE)
1645 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1646 LASSERT(inode->i_mode != 0);
1647 if (S_ISREG(inode->i_mode)) {
1648 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1650 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1652 #ifdef HAVE_INODE_BLKSIZE
1653 inode->i_blksize = 1<<inode->i_blkbits;
1655 if (body->valid & OBD_MD_FLUID)
1656 inode->i_uid = body->uid;
1657 if (body->valid & OBD_MD_FLGID)
1658 inode->i_gid = body->gid;
1659 if (body->valid & OBD_MD_FLFLAGS)
1660 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1661 if (body->valid & OBD_MD_FLNLINK)
1662 inode->i_nlink = body->nlink;
1663 if (body->valid & OBD_MD_FLRDEV)
1664 inode->i_rdev = old_decode_dev(body->rdev);
1666 if (body->valid & OBD_MD_FLID) {
1667 /* FID shouldn't be changed! */
1668 if (fid_is_sane(&lli->lli_fid)) {
1669 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1670 "Trying to change FID "DFID
1671 " to the "DFID", inode %lu/%u(%p)\n",
1672 PFID(&lli->lli_fid), PFID(&body->fid1),
1673 inode->i_ino, inode->i_generation, inode);
1675 lli->lli_fid = body->fid1;
1678 LASSERT(fid_seq(&lli->lli_fid) != 0);
1680 if (body->valid & OBD_MD_FLSIZE) {
1681 if ((ll_i2mdexp(inode)->exp_connect_flags & OBD_CONNECT_SOM) &&
1682 S_ISREG(inode->i_mode) && lli->lli_smd) {
1683 struct lustre_handle lockh;
1686 /* As it is possible a blocking ast has been processed
1687 * by this time, we need to check there is an UPDATE
1688 * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1690 mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1693 if (lli->lli_flags & (LLIF_DONE_WRITING |
1694 LLIF_EPOCH_PENDING |
1696 CERROR("ino %lu flags %lu still has "
1697 "size authority! do not trust "
1698 "the size got from MDS\n",
1699 inode->i_ino, lli->lli_flags);
1701 /* Use old size assignment to avoid
1702 * deadlock bz14138 & bz14326 */
1703 inode->i_size = body->size;
1704 lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1706 ldlm_lock_decref(&lockh, mode);
1709 /* Use old size assignment to avoid
1710 * deadlock bz14138 & bz14326 */
1711 inode->i_size = body->size;
1714 if (body->valid & OBD_MD_FLBLOCKS)
1715 inode->i_blocks = body->blocks;
1718 if (body->valid & OBD_MD_FLMDSCAPA) {
1719 LASSERT(md->mds_capa);
1720 ll_add_capa(inode, md->mds_capa);
1722 if (body->valid & OBD_MD_FLOSSCAPA) {
1723 LASSERT(md->oss_capa);
1724 ll_add_capa(inode, md->oss_capa);
1728 static struct backing_dev_info ll_backing_dev_info = {
1729 .ra_pages = 0, /* No readahead */
1730 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,12))
1731 .capabilities = 0, /* Does contribute to dirty memory */
1733 .memory_backed = 0, /* Does contribute to dirty memory */
1737 void ll_read_inode2(struct inode *inode, void *opaque)
1739 struct lustre_md *md = opaque;
1740 struct ll_inode_info *lli = ll_i2info(inode);
1743 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n",
1744 inode->i_ino, inode->i_generation, inode);
1748 LASSERT(!lli->lli_smd);
1750 /* Core attributes from the MDS first. This is a new inode, and
1751 * the VFS doesn't zero times in the core inode so we have to do
1752 * it ourselves. They will be overwritten by either MDS or OST
1753 * attributes - we just need to make sure they aren't newer. */
1754 LTIME_S(inode->i_mtime) = 0;
1755 LTIME_S(inode->i_atime) = 0;
1756 LTIME_S(inode->i_ctime) = 0;
1758 ll_update_inode(inode, md);
1760 /* OIDEBUG(inode); */
1762 if (S_ISREG(inode->i_mode)) {
1763 struct ll_sb_info *sbi = ll_i2sbi(inode);
1764 inode->i_op = &ll_file_inode_operations;
1765 inode->i_fop = sbi->ll_fop;
1766 inode->i_mapping->a_ops = &ll_aops;
1768 } else if (S_ISDIR(inode->i_mode)) {
1769 inode->i_op = &ll_dir_inode_operations;
1770 inode->i_fop = &ll_dir_operations;
1771 inode->i_mapping->a_ops = &ll_dir_aops;
1773 } else if (S_ISLNK(inode->i_mode)) {
1774 inode->i_op = &ll_fast_symlink_inode_operations;
1777 inode->i_op = &ll_special_inode_operations;
1779 init_special_inode(inode, inode->i_mode,
1780 kdev_t_to_nr(inode->i_rdev));
1782 /* initializing backing dev info. */
1783 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
1789 void ll_delete_inode(struct inode *inode)
1791 struct ll_sb_info *sbi = ll_i2sbi(inode);
1795 rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
1797 CERROR("fid_delete() failed, rc %d\n", rc);
1799 truncate_inode_pages(&inode->i_data, 0);
1805 int ll_iocontrol(struct inode *inode, struct file *file,
1806 unsigned int cmd, unsigned long arg)
1808 struct ll_sb_info *sbi = ll_i2sbi(inode);
1809 struct ptlrpc_request *req = NULL;
1814 case EXT3_IOC_GETFLAGS: {
1815 struct mdt_body *body;
1816 struct obd_capa *oc;
1818 oc = ll_mdscapa_get(inode);
1819 rc = md_getattr(sbi->ll_md_exp, ll_inode2fid(inode), oc,
1820 OBD_MD_FLFLAGS, 0, &req);
1823 CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1827 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1829 flags = body->flags;
1831 ptlrpc_req_finished(req);
1833 RETURN(put_user(flags, (int *)arg));
1835 case EXT3_IOC_SETFLAGS: {
1836 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1837 struct obd_info oinfo = { { { 0 } } };
1838 struct md_op_data *op_data;
1840 if (get_user(flags, (int *)arg))
1844 OBDO_ALLOC(oinfo.oi_oa);
1848 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1849 LUSTRE_OPC_ANY, NULL);
1850 if (IS_ERR(op_data))
1851 RETURN(PTR_ERR(op_data));
1853 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1854 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1855 rc = md_setattr(sbi->ll_md_exp, op_data,
1856 NULL, 0, NULL, 0, &req, NULL);
1857 ll_finish_md_op_data(op_data);
1858 ptlrpc_req_finished(req);
1859 if (rc || lsm == NULL) {
1860 OBDO_FREE(oinfo.oi_oa);
1864 oinfo.oi_oa->o_id = lsm->lsm_object_id;
1865 oinfo.oi_oa->o_gr = lsm->lsm_object_gr;
1866 oinfo.oi_oa->o_flags = flags;
1867 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1869 oinfo.oi_capa = ll_mdscapa_get(inode);
1871 obdo_from_inode(oinfo.oi_oa, inode,
1872 OBD_MD_FLFID | OBD_MD_FLGENER);
1873 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1874 capa_put(oinfo.oi_capa);
1875 OBDO_FREE(oinfo.oi_oa);
1877 if (rc != -EPERM && rc != -EACCES)
1878 CERROR("md_setattr_async fails: rc = %d\n", rc);
1882 inode->i_flags = ll_ext_to_inode_flags(flags |
1883 MDS_BFLAG_EXT_FLAGS);
1893 int ll_flush_ctx(struct inode *inode)
1895 struct ll_sb_info *sbi = ll_i2sbi(inode);
1897 CDEBUG(D_SEC, "flush context for user %d\n", current->uid);
1899 obd_set_info_async(sbi->ll_md_exp,
1900 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1902 obd_set_info_async(sbi->ll_dt_exp,
1903 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1908 /* umount -f client means force down, don't save state */
1909 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1910 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
1912 struct super_block *sb = vfsmnt->mnt_sb;
1914 void ll_umount_begin(struct super_block *sb)
1917 struct lustre_sb_info *lsi = s2lsi(sb);
1918 struct ll_sb_info *sbi = ll_s2sbi(sb);
1919 struct obd_device *obd;
1920 struct obd_ioctl_data ioc_data = { 0 };
1923 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1924 if (!(flags & MNT_FORCE)) {
1930 /* Tell the MGC we got umount -f */
1931 lsi->lsi_flags |= LSI_UMOUNT_FORCE;
1933 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1934 sb->s_count, atomic_read(&sb->s_active));
1936 obd = class_exp2obd(sbi->ll_md_exp);
1938 CERROR("Invalid MDC connection handle "LPX64"\n",
1939 sbi->ll_md_exp->exp_handle.h_cookie);
1944 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp, sizeof ioc_data,
1947 obd = class_exp2obd(sbi->ll_dt_exp);
1949 CERROR("Invalid LOV connection handle "LPX64"\n",
1950 sbi->ll_dt_exp->exp_handle.h_cookie);
1956 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp, sizeof ioc_data,
1959 /* Really, we'd like to wait until there are no requests outstanding,
1960 * and then continue. For now, we just invalidate the requests,
1961 * schedule() and sleep one second if needed, and hope.
1964 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1965 if (atomic_read(&vfsmnt->mnt_count) > 2) {
1966 cfs_schedule_timeout(CFS_TASK_INTERRUPTIBLE,
1967 cfs_time_seconds(1));
1968 if (atomic_read(&vfsmnt->mnt_count) > 2)
1969 LCONSOLE_WARN("Mount still busy with %d refs! You "
1970 "may try to umount it a bit later\n",
1971 atomic_read(&vfsmnt->mnt_count));
1978 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
1980 struct ll_sb_info *sbi = ll_s2sbi(sb);
1984 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
1985 read_only = *flags & MS_RDONLY;
1986 err = obd_set_info_async(sbi->ll_md_exp,
1987 sizeof(KEY_READ_ONLY),
1988 KEY_READ_ONLY, sizeof(read_only),
1991 CERROR("Failed to change the read-only flag during "
1992 "remount: %d\n", err);
1997 sb->s_flags |= MS_RDONLY;
1999 sb->s_flags &= ~MS_RDONLY;
2004 int ll_prep_inode(struct inode **inode,
2005 struct ptlrpc_request *req,
2006 struct super_block *sb)
2008 struct ll_sb_info *sbi = NULL;
2009 struct lustre_md md;
2013 LASSERT(*inode || sb);
2014 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2015 prune_deathrow(sbi, 1);
2016 memset(&md, 0, sizeof(struct lustre_md));
2018 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2019 sbi->ll_md_exp, &md);
2024 ll_update_inode(*inode, &md);
2026 LASSERT(sb != NULL);
2029 * At this point server returns to client's same fid as client
2030 * generated for creating. So using ->fid1 is okay here.
2032 LASSERT(fid_is_sane(&md.body->fid1));
2034 *inode = ll_iget(sb, ll_fid_build_ino(sbi, &md.body->fid1),&md);
2035 if (*inode == NULL || IS_ERR(*inode)) {
2037 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2038 #ifdef CONFIG_FS_POSIX_ACL
2040 posix_acl_release(md.posix_acl);
2041 md.posix_acl = NULL;
2044 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2046 CERROR("new_inode -fatal: rc %d\n", rc);
2051 rc = obd_checkmd(sbi->ll_dt_exp, sbi->ll_md_exp,
2052 ll_i2info(*inode)->lli_smd);
2054 md_free_lustre_md(sbi->ll_md_exp, &md);
2058 int ll_obd_statfs(struct inode *inode, void *arg)
2060 struct ll_sb_info *sbi = NULL;
2061 struct obd_export *exp;
2063 struct obd_ioctl_data *data = NULL;
2067 if (!inode || !(sbi = ll_i2sbi(inode)))
2068 GOTO(out_statfs, rc = -EINVAL);
2070 rc = obd_ioctl_getdata(&buf, &len, arg);
2072 GOTO(out_statfs, rc);
2075 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2076 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2077 GOTO(out_statfs, rc = -EINVAL);
2079 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2080 if (type == LL_STATFS_MDC)
2081 exp = sbi->ll_md_exp;
2082 else if (type == LL_STATFS_LOV)
2083 exp = sbi->ll_dt_exp;
2085 GOTO(out_statfs, rc = -ENODEV);
2087 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2089 GOTO(out_statfs, rc);
2092 obd_ioctl_freedata(buf, len);
2096 int ll_process_config(struct lustre_cfg *lcfg)
2100 struct lprocfs_static_vars lvars;
2104 lprocfs_llite_init_vars(&lvars);
2106 /* The instance name contains the sb: lustre-client-aacfe000 */
2107 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2108 if (!ptr || !*(++ptr))
2110 if (sscanf(ptr, "%lx", &x) != 1)
2113 /* This better be a real Lustre superblock! */
2114 LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2116 /* Note we have not called client_common_fill_super yet, so
2117 proc fns must be able to handle that! */
2118 rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2125 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2126 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2127 struct inode *i1, struct inode *i2,
2128 const char *name, int namelen,
2129 int mode, __u32 opc, void *data)
2131 LASSERT(i1 != NULL);
2133 if (namelen > ll_i2sbi(i1)->ll_namelen)
2134 return ERR_PTR(-ENAMETOOLONG);
2136 if (op_data == NULL)
2137 OBD_ALLOC_PTR(op_data);
2139 if (op_data == NULL)
2140 return ERR_PTR(-ENOMEM);
2142 ll_i2gids(op_data->op_suppgids, i1, i2);
2143 op_data->op_fid1 = *ll_inode2fid(i1);
2144 op_data->op_capa1 = ll_mdscapa_get(i1);
2147 op_data->op_fid2 = *ll_inode2fid(i2);
2148 op_data->op_capa2 = ll_mdscapa_get(i2);
2150 fid_zero(&op_data->op_fid2);
2151 op_data->op_capa2 = NULL;
2154 op_data->op_name = name;
2155 op_data->op_namelen = namelen;
2156 op_data->op_mode = mode;
2157 op_data->op_mod_time = cfs_time_current_sec();
2158 op_data->op_fsuid = current->fsuid;
2159 op_data->op_fsgid = current->fsgid;
2160 op_data->op_cap = cfs_curproc_cap_pack();
2161 op_data->op_bias = MDS_CHECK_SPLIT;
2162 op_data->op_opc = opc;
2163 op_data->op_mds = 0;
2164 op_data->op_data = data;
2169 void ll_finish_md_op_data(struct md_op_data *op_data)
2171 capa_put(op_data->op_capa1);
2172 capa_put(op_data->op_capa2);
2173 OBD_FREE_PTR(op_data);
2176 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2178 struct ll_sb_info *sbi;
2180 LASSERT((seq != NULL) && (vfs != NULL));
2181 sbi = ll_s2sbi(vfs->mnt_sb);
2183 if (sbi->ll_flags & LL_SBI_NOLCK)
2184 seq_puts(seq, ",nolock");
2186 if (sbi->ll_flags & LL_SBI_FLOCK)
2187 seq_puts(seq, ",flock");
2189 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2190 seq_puts(seq, ",localflock");
2192 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2193 seq_puts(seq, ",user_xattr");
2195 if (sbi->ll_flags & LL_SBI_ACL)
2196 seq_puts(seq, ",acl");