1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Lustre Light Super operations
6 * Copyright (c) 2002-2005 Cluster File Systems, Inc.
8 * This file is part of Lustre, http://www.lustre.org.
10 * Lustre is free software; you can redistribute it and/or
11 * modify it under the terms of version 2 of the GNU General Public
12 * License as published by the Free Software Foundation.
14 * Lustre is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
17 * GNU General Public License for more details.
19 * You should have received a copy of the GNU General Public License
20 * along with Lustre; if not, write to the Free Software
21 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
24 #define DEBUG_SUBSYSTEM S_LLITE
26 #include <linux/module.h>
27 #include <linux/types.h>
28 #include <linux/random.h>
29 #include <linux/version.h>
31 #include <lustre_lite.h>
32 #include <lustre_ha.h>
33 #include <lustre_dlm.h>
34 #include <lprocfs_status.h>
35 #include <lustre_disk.h>
36 #include <lustre_param.h>
37 #include <lustre_log.h>
38 #include "llite_internal.h"
40 cfs_mem_cache_t *ll_file_data_slab;
42 LIST_HEAD(ll_super_blocks);
43 spinlock_t ll_sb_lock = SPIN_LOCK_UNLOCKED;
45 extern struct address_space_operations ll_aops;
46 extern struct address_space_operations ll_dir_aops;
49 #define log2(n) ffz(~(n))
53 static struct ll_sb_info *ll_init_sbi(void)
55 struct ll_sb_info *sbi = NULL;
60 OBD_ALLOC(sbi, sizeof(*sbi));
64 spin_lock_init(&sbi->ll_lock);
65 spin_lock_init(&sbi->ll_lco.lco_lock);
66 spin_lock_init(&sbi->ll_pp_extent_lock);
67 spin_lock_init(&sbi->ll_process_lock);
68 sbi->ll_rw_stats_on = 0;
69 INIT_LIST_HEAD(&sbi->ll_pglist);
70 if (num_physpages >> (20 - CFS_PAGE_SHIFT) < 512)
71 sbi->ll_async_page_max = num_physpages / 2;
73 sbi->ll_async_page_max = (num_physpages / 4) * 3;
74 sbi->ll_ra_info.ra_max_pages = min(num_physpages / 8,
75 SBI_DEFAULT_READAHEAD_MAX);
76 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
77 SBI_DEFAULT_READAHEAD_WHOLE_MAX;
79 INIT_LIST_HEAD(&sbi->ll_conn_chain);
80 INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
82 ll_generate_random_uuid(uuid);
83 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
84 CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
86 spin_lock(&ll_sb_lock);
87 list_add_tail(&sbi->ll_list, &ll_super_blocks);
88 spin_unlock(&ll_sb_lock);
90 #ifdef ENABLE_LLITE_CHECKSUM
91 sbi->ll_flags |= LL_SBI_CHECKSUM;
94 #ifdef HAVE_LRU_RESIZE_SUPPORT
95 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
98 #ifdef HAVE_EXPORT___IGET
99 INIT_LIST_HEAD(&sbi->ll_deathrow);
100 spin_lock_init(&sbi->ll_deathrow_lock);
102 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
103 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].pp_r_hist.oh_lock);
104 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].pp_w_hist.oh_lock);
110 void ll_free_sbi(struct super_block *sb)
112 struct ll_sb_info *sbi = ll_s2sbi(sb);
116 spin_lock(&ll_sb_lock);
117 list_del(&sbi->ll_list);
118 spin_unlock(&ll_sb_lock);
119 OBD_FREE(sbi, sizeof(*sbi));
124 static struct dentry_operations ll_d_root_ops = {
125 #ifdef DCACHE_LUSTRE_INVALID
126 .d_compare = ll_dcompare,
130 /* Initialize the default and maximum LOV EA and cookie sizes. This allows
131 * us to make MDS RPCs with large enough reply buffers to hold the
132 * maximum-sized (= maximum striped) EA and cookie without having to
133 * calculate this (via a call into the LOV + OSCs) each time we make an RPC. */
134 static int ll_init_ea_size(struct obd_export *md_exp, struct obd_export *dt_exp)
136 struct lov_stripe_md lsm = { .lsm_magic = LOV_MAGIC };
137 __u32 valsize = sizeof(struct lov_desc);
138 int rc, easize, def_easize, cookiesize;
139 struct lov_desc desc;
143 rc = obd_get_info(dt_exp, strlen(KEY_LOVDESC) + 1, KEY_LOVDESC,
148 stripes = min(desc.ld_tgt_count, (__u32)LOV_MAX_STRIPE_COUNT);
149 lsm.lsm_stripe_count = stripes;
150 easize = obd_size_diskmd(dt_exp, &lsm);
152 lsm.lsm_stripe_count = desc.ld_default_stripe_count;
153 def_easize = obd_size_diskmd(dt_exp, &lsm);
155 cookiesize = stripes * sizeof(struct llog_cookie);
157 CDEBUG(D_HA, "updating max_mdsize/max_cookiesize: %d/%d\n",
160 rc = md_init_ea_size(md_exp, easize, def_easize, cookiesize);
164 static int client_common_fill_super(struct super_block *sb, char *md, char *dt)
166 struct inode *root = 0;
167 struct ll_sb_info *sbi = ll_s2sbi(sb);
168 struct obd_device *obd;
169 struct lu_fid rootfid;
170 struct obd_capa *oc = NULL;
171 struct obd_statfs osfs;
172 struct ptlrpc_request *request = NULL;
173 struct lustre_handle dt_conn = {0, };
174 struct lustre_handle md_conn = {0, };
175 struct obd_connect_data *data = NULL;
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/lustre");
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|
205 #ifdef HAVE_LRU_RESIZE_SUPPORT
206 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
207 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
209 #ifdef CONFIG_FS_POSIX_ACL
210 data->ocd_connect_flags |= OBD_CONNECT_ACL;
212 data->ocd_ibits_known = MDS_INODELOCK_FULL;
213 data->ocd_version = LUSTRE_VERSION_CODE;
215 if (sb->s_flags & MS_RDONLY)
216 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
217 if (sbi->ll_flags & LL_SBI_USER_XATTR)
218 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
220 #ifdef HAVE_MS_FLOCK_LOCK
221 /* force vfs to use lustre handler for flock() calls - bug 10743 */
222 sb->s_flags |= MS_FLOCK_LOCK;
225 if (sbi->ll_flags & LL_SBI_FLOCK)
226 sbi->ll_fop = &ll_file_operations_flock;
227 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
228 sbi->ll_fop = &ll_file_operations;
230 sbi->ll_fop = &ll_file_operations_noflock;
233 data->ocd_connect_flags |= OBD_CONNECT_REAL;
234 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
235 data->ocd_connect_flags &= ~OBD_CONNECT_LCL_CLIENT;
236 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT;
238 data->ocd_connect_flags &= ~OBD_CONNECT_RMT_CLIENT;
239 data->ocd_connect_flags |= OBD_CONNECT_LCL_CLIENT;
242 err = obd_connect(NULL, &md_conn, obd, &sbi->ll_sb_uuid, data);
244 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
245 "recovery, of which this client is not a "
246 "part. Please wait for recovery to complete,"
247 " abort, or time out.\n", md);
250 CERROR("cannot connect to %s: rc = %d\n", md, err);
253 sbi->ll_md_exp = class_conn2export(&md_conn);
255 err = obd_fid_init(sbi->ll_md_exp);
257 CERROR("Can't init metadata layer FID infrastructure, "
262 err = obd_statfs(obd, &osfs, cfs_time_current_64() - HZ, 0);
264 GOTO(out_md_fid, err);
266 size = sizeof(*data);
267 err = obd_get_info(sbi->ll_md_exp, strlen(KEY_CONN_DATA),
268 KEY_CONN_DATA, &size, data);
270 CERROR("Get connect data failed: %d \n", err);
274 LASSERT(osfs.os_bsize);
275 sb->s_blocksize = osfs.os_bsize;
276 sb->s_blocksize_bits = log2(osfs.os_bsize);
277 sb->s_magic = LL_SUPER_MAGIC;
279 /* for bug 11559. in $LINUX/fs/read_write.c, function do_sendfile():
280 * retval = in_file->f_op->sendfile(...);
282 * retval = -EOVERFLOW;
284 * it will check if *ppos is greater than max. However, max equals to
285 * s_maxbytes, which is a negative integer in a x86_64 box since loff_t
286 * has been defined as a signed long long ineger in linux kernel. */
287 #if BITS_PER_LONG == 64
288 sb->s_maxbytes = PAGE_CACHE_MAXBYTES >> 1;
290 sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
292 sbi->ll_namelen = osfs.os_namelen;
293 sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
295 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
296 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
297 LCONSOLE_INFO("Disabling user_xattr feature because "
298 "it is not supported on the server\n");
299 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
302 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
304 sb->s_flags |= MS_POSIXACL;
306 sbi->ll_flags |= LL_SBI_ACL;
308 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
310 sb->s_flags &= ~MS_POSIXACL;
312 sbi->ll_flags &= ~LL_SBI_ACL;
315 if (data->ocd_connect_flags & OBD_CONNECT_JOIN)
316 sbi->ll_flags |= LL_SBI_JOIN;
318 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
319 if (!(data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT)) {
320 /* sometimes local client claims to be remote, but mdt
321 * will disagree when client gss not applied. */
322 LCONSOLE_INFO("client claims to be remote, but server "
323 "rejected, forced to be local.\n");
324 sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
327 if (!(data->ocd_connect_flags & OBD_CONNECT_LCL_CLIENT)) {
328 /* with gss applied, remote client can not claim to be
329 * local, so mdt maybe force client to be remote. */
330 LCONSOLE_INFO("client claims to be local, but server "
331 "rejected, forced to be remote.\n");
332 sbi->ll_flags |= LL_SBI_RMT_CLIENT;
336 if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
337 LCONSOLE_INFO("client enabled MDS capability!\n");
338 sbi->ll_flags |= LL_SBI_MDS_CAPA;
341 if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
342 LCONSOLE_INFO("client enabled OSS capability!\n");
343 sbi->ll_flags |= LL_SBI_OSS_CAPA;
346 sbi->ll_sdev_orig = sb->s_dev;
347 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0))
348 /* We set sb->s_dev equal on all lustre clients in order to support
349 * NFS export clustering. NFSD requires that the FSID be the same
351 /* s_dev is also used in lt_compare() to compare two fs, but that is
352 * only a node-local comparison. */
354 /* XXX: this will not work with LMV */
355 sb->s_dev = get_uuid2int(sbi2mdc(sbi)->cl_target_uuid.uuid,
356 strlen(sbi2mdc(sbi)->cl_target_uuid.uuid));
359 obd = class_name2obd(dt);
361 CERROR("DT %s: not setup or attached\n", dt);
362 GOTO(out_md_fid, err = -ENODEV);
365 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
366 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
367 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID;
368 if (sbi->ll_flags & LL_SBI_OSS_CAPA)
369 data->ocd_connect_flags |= OBD_CONNECT_OSS_CAPA;
371 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
372 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
373 * disabled by default, because it can still be enabled on the
374 * fly via /proc. As a consequence, we still need to come to an
375 * agreement on the supported algorithms at connect time */
376 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
378 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
379 data->ocd_cksum_types = OBD_CKSUM_ADLER;
381 /* send the list of supported checksum types */
382 data->ocd_cksum_types = OBD_CKSUM_ALL;
385 #ifdef HAVE_LRU_RESIZE_SUPPORT
386 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
388 CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
389 "ocd_grant: %d\n", data->ocd_connect_flags,
390 data->ocd_version, data->ocd_grant);
392 obd->obd_upcall.onu_owner = &sbi->ll_lco;
393 obd->obd_upcall.onu_upcall = ll_ocd_update;
394 data->ocd_brw_size = PTLRPC_MAX_BRW_PAGES << CFS_PAGE_SHIFT;
396 err = obd_connect(NULL, &dt_conn, obd, &sbi->ll_sb_uuid, data);
398 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
399 "recovery, of which this client is not a "
400 "part. Please wait for recovery to "
401 "complete, abort, or time out.\n", dt);
402 GOTO(out_md_fid, err);
404 CERROR("Cannot connect to %s: rc = %d\n", dt, err);
405 GOTO(out_md_fid, err);
408 sbi->ll_dt_exp = class_conn2export(&dt_conn);
410 err = obd_fid_init(sbi->ll_dt_exp);
412 CERROR("Can't init data layer FID infrastructure, "
417 spin_lock(&sbi->ll_lco.lco_lock);
418 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
419 spin_unlock(&sbi->ll_lco.lco_lock);
421 ll_init_ea_size(sbi->ll_md_exp, sbi->ll_dt_exp);
423 err = obd_prep_async_page(sbi->ll_dt_exp, NULL, NULL, NULL,
424 0, NULL, NULL, NULL);
426 LCONSOLE_ERROR_MSG(0x151, "There are no OST's in this "
427 "filesystem. There must be at least one "
428 "active OST for a client to start.\n");
429 GOTO(out_dt_fid, err);
432 if (!ll_async_page_slab) {
433 ll_async_page_slab_size =
434 size_round(sizeof(struct ll_async_page)) + err;
435 ll_async_page_slab = cfs_mem_cache_create("ll_async_page",
436 ll_async_page_slab_size,
438 if (!ll_async_page_slab)
439 GOTO(out_dt_fid, err = -ENOMEM);
442 err = md_getstatus(sbi->ll_md_exp, &rootfid, &oc);
444 CERROR("cannot mds_connect: rc = %d\n", err);
445 GOTO(out_dt_fid, err);
447 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&rootfid));
448 sbi->ll_root_fid = rootfid;
450 sb->s_op = &lustre_super_operations;
451 sb->s_export_op = &lustre_export_operations;
454 * XXX: move this to after cbd setup? */
455 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
456 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
457 valid |= OBD_MD_FLRMTPERM;
458 else if (sbi->ll_flags & LL_SBI_ACL)
459 valid |= OBD_MD_FLACL;
461 err = md_getattr(sbi->ll_md_exp, &rootfid, oc, valid, 0, &request);
465 CERROR("md_getattr failed for root: rc = %d\n", err);
466 GOTO(out_dt_fid, err);
468 memset(&lmd, 0, sizeof(lmd));
469 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
470 sbi->ll_md_exp, &lmd);
472 CERROR("failed to understand root inode md: rc = %d\n", err);
473 ptlrpc_req_finished (request);
474 GOTO(out_dt_fid, err);
477 LASSERT(fid_is_sane(&sbi->ll_root_fid));
478 root = ll_iget(sb, ll_fid_build_ino(sbi, &sbi->ll_root_fid), &lmd);
479 md_free_lustre_md(sbi->ll_md_exp, &lmd);
480 ptlrpc_req_finished(request);
482 if (root == NULL || is_bad_inode(root)) {
484 obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
485 #ifdef CONFIG_FS_POSIX_ACL
487 posix_acl_release(lmd.posix_acl);
488 lmd.posix_acl = NULL;
491 CERROR("lustre_lite: bad iget4 for root\n");
492 GOTO(out_root, err = -EBADF);
495 err = ll_close_thread_start(&sbi->ll_lcq);
497 CERROR("cannot start close thread: rc %d\n", err);
501 #ifdef CONFIG_FS_POSIX_ACL
502 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
503 rct_init(&sbi->ll_rct);
504 et_init(&sbi->ll_et);
508 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
509 err = obd_set_info_async(sbi->ll_dt_exp, strlen("checksum"),"checksum",
510 sizeof(checksum), &checksum, NULL);
512 sb->s_root = d_alloc_root(root);
514 OBD_FREE(data, sizeof(*data));
515 sb->s_root->d_op = &ll_d_root_ops;
521 obd_fid_fini(sbi->ll_dt_exp);
523 obd_disconnect(sbi->ll_dt_exp);
524 sbi->ll_dt_exp = NULL;
526 obd_fid_fini(sbi->ll_md_exp);
528 obd_disconnect(sbi->ll_md_exp);
529 sbi->ll_md_exp = NULL;
533 lprocfs_unregister_mountpoint(sbi);
537 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
541 *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
543 rc = obd_get_info(sbi->ll_md_exp, strlen("max_easize"), "max_easize",
546 CERROR("Get max mdsize error rc %d \n", rc);
551 void ll_dump_inode(struct inode *inode)
553 struct list_head *tmp;
554 int dentry_count = 0;
556 LASSERT(inode != NULL);
558 list_for_each(tmp, &inode->i_dentry)
561 CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
562 inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
563 inode->i_mode, atomic_read(&inode->i_count), dentry_count);
566 void lustre_dump_dentry(struct dentry *dentry, int recur)
568 struct list_head *tmp;
571 LASSERT(dentry != NULL);
573 list_for_each(tmp, &dentry->d_subdirs)
576 CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
577 " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
578 dentry->d_name.len, dentry->d_name.name,
579 dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
580 dentry->d_parent, dentry->d_inode, atomic_read(&dentry->d_count),
581 dentry->d_flags, dentry->d_fsdata, subdirs);
582 if (dentry->d_inode != NULL)
583 ll_dump_inode(dentry->d_inode);
588 list_for_each(tmp, &dentry->d_subdirs) {
589 struct dentry *d = list_entry(tmp, struct dentry, d_child);
590 lustre_dump_dentry(d, recur - 1);
594 #ifdef HAVE_EXPORT___IGET
595 static void prune_dir_dentries(struct inode *inode)
597 struct dentry *dentry, *prev = NULL;
599 /* due to lustre specific logic, a directory
600 * can have few dentries - a bug from VFS POV */
602 spin_lock(&dcache_lock);
603 if (!list_empty(&inode->i_dentry)) {
604 dentry = list_entry(inode->i_dentry.prev,
605 struct dentry, d_alias);
606 /* in order to prevent infinite loops we
607 * break if previous dentry is busy */
608 if (dentry != prev) {
611 spin_unlock(&dcache_lock);
613 /* try to kill all child dentries */
615 shrink_dcache_parent(dentry);
616 unlock_dentry(dentry);
619 /* now try to get rid of current dentry */
620 d_prune_aliases(inode);
624 spin_unlock(&dcache_lock);
627 static void prune_deathrow_one(struct ll_inode_info *lli)
629 struct inode *inode = ll_info2i(lli);
631 /* first, try to drop any dentries - they hold a ref on the inode */
632 if (S_ISDIR(inode->i_mode))
633 prune_dir_dentries(inode);
635 d_prune_aliases(inode);
638 /* if somebody still uses it, leave it */
639 LASSERT(atomic_read(&inode->i_count) > 0);
640 if (atomic_read(&inode->i_count) > 1)
643 CDEBUG(D_INODE, "inode %lu/%u(%d) looks a good candidate for prune\n",
644 inode->i_ino,inode->i_generation, atomic_read(&inode->i_count));
646 /* seems nobody uses it anymore */
654 static void prune_deathrow(struct ll_sb_info *sbi, int try)
656 struct ll_inode_info *lli;
660 if (need_resched() && try)
664 if (!spin_trylock(&sbi->ll_deathrow_lock))
667 spin_lock(&sbi->ll_deathrow_lock);
672 if (!list_empty(&sbi->ll_deathrow)) {
673 lli = list_entry(sbi->ll_deathrow.next,
674 struct ll_inode_info,
676 list_del_init(&lli->lli_dead_list);
677 if (!list_empty(&sbi->ll_deathrow))
680 spin_unlock(&sbi->ll_deathrow_lock);
683 prune_deathrow_one(lli);
685 } while (empty == 0);
687 #else /* !HAVE_EXPORT___IGET */
688 #define prune_deathrow(sbi, try) do {} while (0)
689 #endif /* HAVE_EXPORT___IGET */
691 void client_common_put_super(struct super_block *sb)
693 struct ll_sb_info *sbi = ll_s2sbi(sb);
696 #ifdef CONFIG_FS_POSIX_ACL
697 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
698 et_fini(&sbi->ll_et);
699 rct_fini(&sbi->ll_rct);
703 obd_cancel_unused(sbi->ll_dt_exp, NULL, 0, NULL);
705 ll_close_thread_shutdown(sbi->ll_lcq);
707 /* destroy inodes in deathrow */
708 prune_deathrow(sbi, 0);
710 list_del(&sbi->ll_conn_chain);
712 obd_fid_fini(sbi->ll_dt_exp);
713 obd_disconnect(sbi->ll_dt_exp);
714 sbi->ll_dt_exp = NULL;
716 lprocfs_unregister_mountpoint(sbi);
718 obd_fid_fini(sbi->ll_md_exp);
719 obd_disconnect(sbi->ll_md_exp);
720 sbi->ll_md_exp = NULL;
725 void ll_kill_super(struct super_block *sb)
727 struct ll_sb_info *sbi;
732 if (!(sb->s_flags & MS_ACTIVE))
736 /* we need restore s_dev from changed for clustred NFS before put_super
737 * because new kernels have cached s_dev and change sb->s_dev in
738 * put_super not affected real removing devices */
740 sb->s_dev = sbi->ll_sdev_orig;
744 char *ll_read_opt(const char *opt, char *data)
750 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
751 if (strncmp(opt, data, strlen(opt)))
753 if ((value = strchr(data, '=')) == NULL)
757 OBD_ALLOC(retval, strlen(value) + 1);
759 CERROR("out of memory!\n");
763 memcpy(retval, value, strlen(value)+1);
764 CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
768 static inline int ll_set_opt(const char *opt, char *data, int fl)
770 if (strncmp(opt, data, strlen(opt)) != 0)
776 /* non-client-specific mount options are parsed in lmd_parse */
777 static int ll_options(char *options, int *flags)
780 char *s1 = options, *s2;
786 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
789 CDEBUG(D_SUPER, "next opt=%s\n", s1);
790 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
795 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
800 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
805 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
810 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
815 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
820 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
822 /* Ignore deprecated mount option. The client will
823 * always try to mount with ACL support, whether this
824 * is used depends on whether server supports it. */
827 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
831 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
837 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
842 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
847 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
852 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
858 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
864 s2 = strchr(s1, ',');
872 void ll_lli_init(struct ll_inode_info *lli)
874 lli->lli_inode_magic = LLI_INODE_MAGIC;
875 sema_init(&lli->lli_size_sem, 1);
876 sema_init(&lli->lli_write_sem, 1);
878 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
879 spin_lock_init(&lli->lli_lock);
880 INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
881 INIT_LIST_HEAD(&lli->lli_close_list);
882 lli->lli_inode_magic = LLI_INODE_MAGIC;
883 sema_init(&lli->lli_och_sem, 1);
884 lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
885 lli->lli_mds_exec_och = NULL;
886 lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
887 lli->lli_open_fd_exec_count = 0;
888 INIT_LIST_HEAD(&lli->lli_dead_list);
889 lli->lli_remote_perms = NULL;
890 lli->lli_rmtperm_utime = 0;
891 sema_init(&lli->lli_rmtperm_sem, 1);
892 INIT_LIST_HEAD(&lli->lli_oss_capas);
895 int ll_fill_super(struct super_block *sb)
897 struct lustre_profile *lprof;
898 struct lustre_sb_info *lsi = s2lsi(sb);
899 struct ll_sb_info *sbi;
900 char *dt = NULL, *md = NULL;
901 char *profilenm = get_profile_name(sb);
902 struct config_llog_instance cfg = {0, };
903 char ll_instance[sizeof(sb) * 2 + 1];
907 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
911 sb->s_type->fs_flags |= FS_RENAME_DOES_D_MOVE;
912 /* client additional sb info */
913 lsi->lsi_llsbi = sbi = ll_init_sbi();
919 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
923 /* Generate a string unique to this super, in case some joker tries
924 to mount the same fs at two mount points.
925 Use the address of the super itself.*/
926 sprintf(ll_instance, "%p", sb);
927 cfg.cfg_instance = ll_instance;
928 cfg.cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
930 /* set up client obds */
931 err = lustre_process_log(sb, profilenm, &cfg);
933 CERROR("Unable to process log: %d\n", err);
937 lprof = class_get_profile(profilenm);
939 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
940 " read from the MGS. Does that filesystem "
941 "exist?\n", profilenm);
942 GOTO(out_free, err = -EINVAL);
944 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
945 lprof->lp_md, lprof->lp_dt);
947 OBD_ALLOC(dt, strlen(lprof->lp_dt) +
948 strlen(ll_instance) + 2);
950 GOTO(out_free, err = -ENOMEM);
951 sprintf(dt, "%s-%s", lprof->lp_dt, ll_instance);
953 OBD_ALLOC(md, strlen(lprof->lp_md) +
954 strlen(ll_instance) + 2);
956 GOTO(out_free, err = -ENOMEM);
957 sprintf(md, "%s-%s", lprof->lp_md, ll_instance);
959 /* connections, registrations, sb setup */
960 err = client_common_fill_super(sb, md, dt);
964 OBD_FREE(md, strlen(md) + 1);
966 OBD_FREE(dt, strlen(dt) + 1);
970 LCONSOLE_WARN("Client %s has started\n", profilenm);
973 } /* ll_fill_super */
976 void ll_put_super(struct super_block *sb)
978 struct config_llog_instance cfg;
979 char ll_instance[sizeof(sb) * 2 + 1];
980 struct obd_device *obd;
981 struct lustre_sb_info *lsi = s2lsi(sb);
982 struct ll_sb_info *sbi = ll_s2sbi(sb);
983 char *profilenm = get_profile_name(sb);
987 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
989 ll_print_capa_stat(sbi);
991 sprintf(ll_instance, "%p", sb);
992 cfg.cfg_instance = ll_instance;
993 lustre_end_log(sb, NULL, &cfg);
995 if (sbi->ll_md_exp) {
996 obd = class_exp2obd(sbi->ll_md_exp);
998 force = obd->obd_force;
1001 /* We need to set force before the lov_disconnect in
1002 lustre_common_put_super, since l_d cleans up osc's as well. */
1005 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1007 obd->obd_force = force;
1012 /* Only if client_common_fill_super succeeded */
1013 client_common_put_super(sb);
1016 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1017 class_manual_cleanup(obd);
1021 class_del_profile(profilenm);
1024 lsi->lsi_llsbi = NULL;
1026 lustre_common_put_super(sb);
1028 LCONSOLE_WARN("client %s umount complete\n", ll_instance);
1033 } /* client_put_super */
1035 #ifdef HAVE_REGISTER_CACHE
1036 #include <linux/cache_def.h>
1037 #ifdef HAVE_CACHE_RETURN_INT
1042 ll_shrink_cache(int priority, unsigned int gfp_mask)
1044 struct ll_sb_info *sbi;
1047 list_for_each_entry(sbi, &ll_super_blocks, ll_list)
1048 count += llap_shrink_cache(sbi, priority);
1050 #ifdef HAVE_CACHE_RETURN_INT
1055 struct cache_definition ll_cache_definition = {
1056 .name = "llap_cache",
1057 .shrink = ll_shrink_cache
1059 #endif /* HAVE_REGISTER_CACHE */
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 ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1106 md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1107 null_if_equal, inode);
1109 LASSERT(!lli->lli_open_fd_write_count);
1110 LASSERT(!lli->lli_open_fd_read_count);
1111 LASSERT(!lli->lli_open_fd_exec_count);
1113 if (lli->lli_mds_write_och)
1114 ll_md_real_close(inode, FMODE_WRITE);
1115 if (lli->lli_mds_exec_och)
1116 ll_md_real_close(inode, FMODE_EXEC);
1117 if (lli->lli_mds_read_och)
1118 ll_md_real_close(inode, FMODE_READ);
1121 obd_change_cbdata(sbi->ll_dt_exp, lli->lli_smd,
1122 null_if_equal, inode);
1124 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1125 lli->lli_smd = NULL;
1128 if (lli->lli_symlink_name) {
1129 OBD_FREE(lli->lli_symlink_name,
1130 strlen(lli->lli_symlink_name) + 1);
1131 lli->lli_symlink_name = NULL;
1134 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1135 LASSERT(lli->lli_posix_acl == NULL);
1136 if (lli->lli_remote_perms) {
1137 free_rmtperm_hash(lli->lli_remote_perms);
1138 lli->lli_remote_perms = NULL;
1141 #ifdef CONFIG_FS_POSIX_ACL
1142 else if (lli->lli_posix_acl) {
1143 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1144 LASSERT(lli->lli_remote_perms == NULL);
1145 posix_acl_release(lli->lli_posix_acl);
1146 lli->lli_posix_acl = NULL;
1149 lli->lli_inode_magic = LLI_INODE_DEAD;
1151 #ifdef HAVE_EXPORT___IGET
1152 spin_lock(&sbi->ll_deathrow_lock);
1153 list_del_init(&lli->lli_dead_list);
1154 spin_unlock(&sbi->ll_deathrow_lock);
1156 ll_clear_inode_capas(inode);
1161 int ll_md_setattr(struct inode *inode, struct md_op_data *op_data,
1162 struct md_open_data **mod)
1164 struct lustre_md md;
1165 struct ll_sb_info *sbi = ll_i2sbi(inode);
1166 struct ptlrpc_request *request = NULL;
1170 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1171 LUSTRE_OPC_ANY, NULL);
1172 if (IS_ERR(op_data))
1173 RETURN(PTR_ERR(op_data));
1175 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1178 ptlrpc_req_finished(request);
1179 if (rc == -ENOENT) {
1181 /* Unlinked special device node? Or just a race?
1182 * Pretend we done everything. */
1183 if (!S_ISREG(inode->i_mode) &&
1184 !S_ISDIR(inode->i_mode))
1185 rc = inode_setattr(inode, &op_data->op_attr);
1186 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1187 CERROR("md_setattr fails: rc = %d\n", rc);
1192 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1193 sbi->ll_md_exp, &md);
1195 ptlrpc_req_finished(request);
1199 /* We call inode_setattr to adjust timestamps.
1200 * If there is at least some data in file, we cleared ATTR_SIZE
1201 * above to avoid invoking vmtruncate, otherwise it is important
1202 * to call vmtruncate in inode_setattr to update inode->i_size
1204 rc = inode_setattr(inode, &op_data->op_attr);
1206 /* Extract epoch data if obtained. */
1207 op_data->op_handle = md.body->handle;
1208 op_data->op_ioepoch = md.body->ioepoch;
1210 ll_update_inode(inode, &md);
1211 ptlrpc_req_finished(request);
1216 /* Close IO epoch and send Size-on-MDS attribute update. */
1217 static int ll_setattr_done_writing(struct inode *inode,
1218 struct md_op_data *op_data,
1219 struct md_open_data *mod)
1221 struct ll_inode_info *lli = ll_i2info(inode);
1225 LASSERT(op_data != NULL);
1226 if (!S_ISREG(inode->i_mode))
1229 CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1230 op_data->op_ioepoch, PFID(&lli->lli_fid));
1232 op_data->op_flags = MF_EPOCH_CLOSE | MF_SOM_CHANGE;
1233 rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1234 if (rc == -EAGAIN) {
1235 /* MDS has instructed us to obtain Size-on-MDS attribute
1236 * from OSTs and send setattr to back to MDS. */
1237 rc = ll_sizeonmds_update(inode, mod, &op_data->op_handle,
1238 op_data->op_ioepoch);
1240 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1246 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1247 * object(s) determine the file size and mtime. Otherwise, the MDS will
1248 * keep these values until such a time that objects are allocated for it.
1249 * We do the MDS operations first, as it is checking permissions for us.
1250 * We don't to the MDS RPC if there is nothing that we want to store there,
1251 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1252 * going to do an RPC anyways.
1254 * If we are doing a truncate, we will send the mtime and ctime updates
1255 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1256 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1259 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1261 struct ll_inode_info *lli = ll_i2info(inode);
1262 struct lov_stripe_md *lsm = lli->lli_smd;
1263 struct ll_sb_info *sbi = ll_i2sbi(inode);
1264 struct md_op_data *op_data = NULL;
1265 struct md_open_data *mod = NULL;
1266 int ia_valid = attr->ia_valid;
1267 int rc = 0, rc1 = 0;
1270 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1272 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_SETATTR, 1);
1274 if (ia_valid & ATTR_SIZE) {
1275 if (attr->ia_size > ll_file_maxbytes(inode)) {
1276 CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1277 attr->ia_size, ll_file_maxbytes(inode));
1281 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1284 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1285 if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1286 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1290 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1291 if (attr->ia_valid & ATTR_CTIME) {
1292 attr->ia_ctime = CURRENT_TIME;
1293 attr->ia_valid |= ATTR_CTIME_SET;
1295 if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1296 attr->ia_atime = CURRENT_TIME;
1297 attr->ia_valid |= ATTR_ATIME_SET;
1299 if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1300 attr->ia_mtime = CURRENT_TIME;
1301 attr->ia_valid |= ATTR_MTIME_SET;
1303 if ((attr->ia_valid & ATTR_CTIME) && !(attr->ia_valid & ATTR_MTIME)) {
1304 /* To avoid stale mtime on mds, obtain it from ost and send
1306 rc = ll_glimpse_size(inode, 0);
1310 attr->ia_valid |= ATTR_MTIME_SET | ATTR_MTIME;
1311 attr->ia_mtime = inode->i_mtime;
1314 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1315 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1316 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1319 /* NB: ATTR_SIZE will only be set after this point if the size
1320 * resides on the MDS, ie, this file has no objects. */
1322 attr->ia_valid &= ~ATTR_SIZE;
1324 /* We always do an MDS RPC, even if we're only changing the size;
1325 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1327 OBD_ALLOC_PTR(op_data);
1328 if (op_data == NULL)
1331 memcpy(&op_data->op_attr, attr, sizeof(*attr));
1333 /* Open epoch for truncate. */
1334 if ((ll_i2mdexp(inode)->exp_connect_flags & OBD_CONNECT_SOM) &&
1335 (ia_valid & ATTR_SIZE))
1336 op_data->op_flags = MF_EPOCH_OPEN;
1338 rc = ll_md_setattr(inode, op_data, &mod);
1342 if (op_data->op_ioepoch)
1343 CDEBUG(D_INODE, "Epoch "LPU64" opened on "DFID" for "
1344 "truncate\n", op_data->op_ioepoch, PFID(&lli->lli_fid));
1346 if (!lsm || !S_ISREG(inode->i_mode)) {
1347 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1351 /* We really need to get our PW lock before we change inode->i_size.
1352 * If we don't we can race with other i_size updaters on our node, like
1353 * ll_file_read. We can also race with i_size propogation to other
1354 * nodes through dirtying and writeback of final cached pages. This
1355 * last one is especially bad for racing o_append users on other
1357 if (ia_valid & ATTR_SIZE) {
1358 ldlm_policy_data_t policy = { .l_extent = {attr->ia_size,
1360 struct lustre_handle lockh = { 0 };
1361 int err, ast_flags = 0;
1362 /* XXX when we fix the AST intents to pass the discard-range
1363 * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
1365 if (attr->ia_size == 0)
1366 ast_flags = LDLM_AST_DISCARD_DATA;
1368 UNLOCK_INODE_MUTEX(inode);
1369 UP_WRITE_I_ALLOC_SEM(inode);
1370 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy, &lockh,
1372 LOCK_INODE_MUTEX(inode);
1373 DOWN_WRITE_I_ALLOC_SEM(inode);
1378 /* Only ll_inode_size_lock is taken at this level.
1379 * lov_stripe_lock() is grabbed by ll_truncate() only over
1380 * call to obd_adjust_kms(). If vmtruncate returns 0, then
1381 * ll_truncate dropped ll_inode_size_lock() */
1382 ll_inode_size_lock(inode, 0);
1383 rc = vmtruncate(inode, attr->ia_size);
1385 LASSERT(atomic_read(&lli->lli_size_sem.count) <= 0);
1386 ll_inode_size_unlock(inode, 0);
1389 err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
1391 CERROR("ll_extent_unlock failed: %d\n", err);
1395 } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
1397 struct obd_info oinfo = { { { 0 } } };
1400 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1401 inode->i_ino, LTIME_S(attr->ia_mtime));
1405 oa->o_id = lsm->lsm_object_id;
1406 oa->o_gr = lsm->lsm_object_gr;
1407 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1409 flags = OBD_MD_FLTYPE | OBD_MD_FLATIME |
1410 OBD_MD_FLMTIME | OBD_MD_FLCTIME |
1411 OBD_MD_FLFID | OBD_MD_FLGENER |
1414 obdo_from_inode(oa, inode, flags);
1418 oinfo.oi_capa = ll_mdscapa_get(inode);
1420 /* XXX: this looks unnecessary now. */
1421 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1422 capa_put(oinfo.oi_capa);
1424 CERROR("obd_setattr_async fails: rc=%d\n", rc);
1433 if (op_data->op_ioepoch)
1434 rc1 = ll_setattr_done_writing(inode, op_data, mod);
1435 ll_finish_md_op_data(op_data);
1437 return rc ? rc : rc1;
1440 int ll_setattr(struct dentry *de, struct iattr *attr)
1442 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1443 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1444 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1446 return ll_setattr_raw(de->d_inode, attr);
1449 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1450 __u64 max_age, __u32 flags)
1452 struct ll_sb_info *sbi = ll_s2sbi(sb);
1453 struct obd_statfs obd_osfs;
1457 rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age, flags);
1459 CERROR("md_statfs fails: rc = %d\n", rc);
1463 osfs->os_type = sb->s_magic;
1465 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1466 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1468 rc = obd_statfs_rqset(class_exp2obd(sbi->ll_dt_exp),
1469 &obd_osfs, max_age, flags);
1471 CERROR("obd_statfs fails: rc = %d\n", rc);
1475 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1476 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1479 osfs->os_bsize = obd_osfs.os_bsize;
1480 osfs->os_blocks = obd_osfs.os_blocks;
1481 osfs->os_bfree = obd_osfs.os_bfree;
1482 osfs->os_bavail = obd_osfs.os_bavail;
1484 /* If we don't have as many objects free on the OST as inodes
1485 * on the MDS, we reduce the total number of inodes to
1486 * compensate, so that the "inodes in use" number is correct.
1488 if (obd_osfs.os_ffree < osfs->os_ffree) {
1489 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1491 osfs->os_ffree = obd_osfs.os_ffree;
1496 #ifndef HAVE_STATFS_DENTRY_PARAM
1497 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1500 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1502 struct super_block *sb = de->d_sb;
1504 struct obd_statfs osfs;
1507 CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1508 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1510 /* For now we will always get up-to-date statfs values, but in the
1511 * future we may allow some amount of caching on the client (e.g.
1512 * from QOS or lprocfs updates). */
1513 rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - 1, 0);
1517 statfs_unpack(sfs, &osfs);
1519 /* We need to downshift for all 32-bit kernels, because we can't
1520 * tell if the kernel is being called via sys_statfs64() or not.
1521 * Stop before overflowing f_bsize - in which case it is better
1522 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1523 if (sizeof(long) < 8) {
1524 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1527 osfs.os_blocks >>= 1;
1528 osfs.os_bfree >>= 1;
1529 osfs.os_bavail >>= 1;
1533 sfs->f_blocks = osfs.os_blocks;
1534 sfs->f_bfree = osfs.os_bfree;
1535 sfs->f_bavail = osfs.os_bavail;
1540 void ll_inode_size_lock(struct inode *inode, int lock_lsm)
1542 struct ll_inode_info *lli;
1543 struct lov_stripe_md *lsm;
1545 lli = ll_i2info(inode);
1546 LASSERT(lli->lli_size_sem_owner != current);
1547 down(&lli->lli_size_sem);
1548 LASSERT(lli->lli_size_sem_owner == NULL);
1549 lli->lli_size_sem_owner = current;
1551 LASSERTF(lsm != NULL || lock_lsm == 0, "lsm %p, lock_lsm %d\n",
1554 lov_stripe_lock(lsm);
1557 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm)
1559 struct ll_inode_info *lli;
1560 struct lov_stripe_md *lsm;
1562 lli = ll_i2info(inode);
1564 LASSERTF(lsm != NULL || unlock_lsm == 0, "lsm %p, lock_lsm %d\n",
1567 lov_stripe_unlock(lsm);
1568 LASSERT(lli->lli_size_sem_owner == current);
1569 lli->lli_size_sem_owner = NULL;
1570 up(&lli->lli_size_sem);
1573 static void ll_replace_lsm(struct inode *inode, struct lov_stripe_md *lsm)
1575 struct ll_inode_info *lli = ll_i2info(inode);
1577 dump_lsm(D_INODE, lsm);
1578 dump_lsm(D_INODE, lli->lli_smd);
1579 LASSERTF(lsm->lsm_magic == LOV_MAGIC_JOIN,
1580 "lsm must be joined lsm %p\n", lsm);
1581 obd_free_memmd(ll_i2dtexp(inode), &lli->lli_smd);
1582 CDEBUG(D_INODE, "replace lsm %p to lli_smd %p for inode %lu%u(%p)\n",
1583 lsm, lli->lli_smd, inode->i_ino, inode->i_generation, inode);
1585 lli->lli_maxbytes = lsm->lsm_maxbytes;
1586 if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1587 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1590 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1592 struct ll_inode_info *lli = ll_i2info(inode);
1593 struct mdt_body *body = md->body;
1594 struct lov_stripe_md *lsm = md->lsm;
1595 struct ll_sb_info *sbi = ll_i2sbi(inode);
1597 LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1599 if (lli->lli_smd == NULL) {
1600 if (lsm->lsm_magic != LOV_MAGIC &&
1601 lsm->lsm_magic != LOV_MAGIC_JOIN) {
1602 dump_lsm(D_ERROR, lsm);
1605 CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1606 lsm, inode->i_ino, inode->i_generation, inode);
1607 /* ll_inode_size_lock() requires it is only called
1608 * with lli_smd != NULL or lock_lsm == 0 or we can
1609 * race between lock/unlock. bug 9547 */
1611 lli->lli_maxbytes = lsm->lsm_maxbytes;
1612 if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1613 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1615 if (lli->lli_smd->lsm_magic == lsm->lsm_magic &&
1616 lli->lli_smd->lsm_stripe_count ==
1617 lsm->lsm_stripe_count) {
1618 if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
1619 CERROR("lsm mismatch for inode %ld\n",
1621 CERROR("lli_smd:\n");
1622 dump_lsm(D_ERROR, lli->lli_smd);
1624 dump_lsm(D_ERROR, lsm);
1628 ll_replace_lsm(inode, lsm);
1630 if (lli->lli_smd != lsm)
1631 obd_free_memmd(ll_i2dtexp(inode), &lsm);
1634 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1635 if (body->valid & OBD_MD_FLRMTPERM)
1636 ll_update_remote_perm(inode, md->remote_perm);
1638 #ifdef CONFIG_FS_POSIX_ACL
1639 else if (body->valid & OBD_MD_FLACL) {
1640 spin_lock(&lli->lli_lock);
1641 if (lli->lli_posix_acl)
1642 posix_acl_release(lli->lli_posix_acl);
1643 lli->lli_posix_acl = md->posix_acl;
1644 spin_unlock(&lli->lli_lock);
1647 if (body->valid & OBD_MD_FLATIME &&
1648 body->atime > LTIME_S(inode->i_atime))
1649 LTIME_S(inode->i_atime) = body->atime;
1651 /* mtime is always updated with ctime, but can be set in past.
1652 As write and utime(2) may happen within 1 second, and utime's
1653 mtime has a priority over write's one, so take mtime from mds
1654 for the same ctimes. */
1655 if (body->valid & OBD_MD_FLCTIME &&
1656 body->ctime >= LTIME_S(inode->i_ctime)) {
1657 LTIME_S(inode->i_ctime) = body->ctime;
1658 if (body->valid & OBD_MD_FLMTIME) {
1659 CDEBUG(D_INODE, "setting ino %lu mtime "
1660 "from %lu to "LPU64"\n", inode->i_ino,
1661 LTIME_S(inode->i_mtime), body->mtime);
1662 LTIME_S(inode->i_mtime) = body->mtime;
1665 if (body->valid & OBD_MD_FLMODE)
1666 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1667 if (body->valid & OBD_MD_FLTYPE)
1668 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1669 if (S_ISREG(inode->i_mode)) {
1670 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1672 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1674 #ifdef HAVE_INODE_BLKSIZE
1675 inode->i_blksize = 1<<inode->i_blkbits;
1677 if (body->valid & OBD_MD_FLUID)
1678 inode->i_uid = body->uid;
1679 if (body->valid & OBD_MD_FLGID)
1680 inode->i_gid = body->gid;
1681 if (body->valid & OBD_MD_FLFLAGS)
1682 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1683 if (body->valid & OBD_MD_FLNLINK)
1684 inode->i_nlink = body->nlink;
1685 if (body->valid & OBD_MD_FLRDEV)
1686 inode->i_rdev = old_decode_dev(body->rdev);
1688 if (body->valid & OBD_MD_FLID) {
1689 /* FID shouldn't be changed! */
1690 if (fid_is_sane(&lli->lli_fid)) {
1691 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1692 "Trying to change FID "DFID
1693 " to the "DFID", inode %lu/%u(%p)\n",
1694 PFID(&lli->lli_fid), PFID(&body->fid1),
1695 inode->i_ino, inode->i_generation, inode);
1697 lli->lli_fid = body->fid1;
1700 LASSERT(fid_seq(&lli->lli_fid) != 0);
1702 if (body->valid & OBD_MD_FLSIZE) {
1703 if ((ll_i2mdexp(inode)->exp_connect_flags & OBD_CONNECT_SOM) &&
1704 S_ISREG(inode->i_mode) && lli->lli_smd) {
1705 struct lustre_handle lockh;
1708 /* As it is possible a blocking ast has been processed
1709 * by this time, we need to check there is an UPDATE
1710 * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1712 mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1715 if (lli->lli_flags & (LLIF_DONE_WRITING |
1716 LLIF_EPOCH_PENDING |
1718 CERROR("ino %lu flags %lu still has "
1719 "size authority! do not trust "
1720 "the size got from MDS\n",
1721 inode->i_ino, lli->lli_flags);
1723 /* Use old size assignment to avoid
1724 * deadlock bz14138 & bz14326 */
1725 inode->i_size = body->size;
1726 lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1728 ldlm_lock_decref(&lockh, mode);
1731 /* Use old size assignment to avoid
1732 * deadlock bz14138 & bz14326 */
1733 inode->i_size = body->size;
1736 if (body->valid & OBD_MD_FLBLOCKS)
1737 inode->i_blocks = body->blocks;
1740 if (body->valid & OBD_MD_FLMDSCAPA) {
1741 LASSERT(md->mds_capa);
1742 ll_add_capa(inode, md->mds_capa);
1744 if (body->valid & OBD_MD_FLOSSCAPA) {
1745 LASSERT(md->oss_capa);
1746 ll_add_capa(inode, md->oss_capa);
1750 static struct backing_dev_info ll_backing_dev_info = {
1751 .ra_pages = 0, /* No readahead */
1752 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,12))
1753 .capabilities = 0, /* Does contribute to dirty memory */
1755 .memory_backed = 0, /* Does contribute to dirty memory */
1759 void ll_read_inode2(struct inode *inode, void *opaque)
1761 struct lustre_md *md = opaque;
1762 struct ll_inode_info *lli = ll_i2info(inode);
1765 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n",
1766 inode->i_ino, inode->i_generation, inode);
1770 LASSERT(!lli->lli_smd);
1772 /* Core attributes from the MDS first. This is a new inode, and
1773 * the VFS doesn't zero times in the core inode so we have to do
1774 * it ourselves. They will be overwritten by either MDS or OST
1775 * attributes - we just need to make sure they aren't newer. */
1776 LTIME_S(inode->i_mtime) = 0;
1777 LTIME_S(inode->i_atime) = 0;
1778 LTIME_S(inode->i_ctime) = 0;
1780 ll_update_inode(inode, md);
1782 /* OIDEBUG(inode); */
1784 if (S_ISREG(inode->i_mode)) {
1785 struct ll_sb_info *sbi = ll_i2sbi(inode);
1786 inode->i_op = &ll_file_inode_operations;
1787 inode->i_fop = sbi->ll_fop;
1788 inode->i_mapping->a_ops = &ll_aops;
1790 } else if (S_ISDIR(inode->i_mode)) {
1791 inode->i_op = &ll_dir_inode_operations;
1792 inode->i_fop = &ll_dir_operations;
1793 inode->i_mapping->a_ops = &ll_dir_aops;
1795 } else if (S_ISLNK(inode->i_mode)) {
1796 inode->i_op = &ll_fast_symlink_inode_operations;
1799 inode->i_op = &ll_special_inode_operations;
1801 init_special_inode(inode, inode->i_mode,
1802 kdev_t_to_nr(inode->i_rdev));
1804 /* initializing backing dev info. */
1805 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
1811 void ll_delete_inode(struct inode *inode)
1813 struct ll_sb_info *sbi = ll_i2sbi(inode);
1817 rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
1819 CERROR("fid_delete() failed, rc %d\n", rc);
1826 int ll_iocontrol(struct inode *inode, struct file *file,
1827 unsigned int cmd, unsigned long arg)
1829 struct ll_sb_info *sbi = ll_i2sbi(inode);
1830 struct ptlrpc_request *req = NULL;
1835 case EXT3_IOC_GETFLAGS: {
1836 struct mdt_body *body;
1837 struct obd_capa *oc;
1839 oc = ll_mdscapa_get(inode);
1840 rc = md_getattr(sbi->ll_md_exp, ll_inode2fid(inode), oc,
1841 OBD_MD_FLFLAGS, 0, &req);
1844 CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1848 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1850 flags = body->flags;
1852 ptlrpc_req_finished(req);
1854 RETURN(put_user(flags, (int *)arg));
1856 case EXT3_IOC_SETFLAGS: {
1857 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1858 struct obd_info oinfo = { { { 0 } } };
1859 struct md_op_data *op_data;
1861 if (get_user(flags, (int *)arg))
1865 OBDO_ALLOC(oinfo.oi_oa);
1869 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1870 LUSTRE_OPC_ANY, NULL);
1871 if (IS_ERR(op_data))
1872 RETURN(PTR_ERR(op_data));
1874 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1875 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1876 rc = md_setattr(sbi->ll_md_exp, op_data,
1877 NULL, 0, NULL, 0, &req, NULL);
1878 ll_finish_md_op_data(op_data);
1879 ptlrpc_req_finished(req);
1880 if (rc || lsm == NULL) {
1881 OBDO_FREE(oinfo.oi_oa);
1885 oinfo.oi_oa->o_id = lsm->lsm_object_id;
1886 oinfo.oi_oa->o_gr = lsm->lsm_object_gr;
1887 oinfo.oi_oa->o_flags = flags;
1888 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1890 oinfo.oi_capa = ll_mdscapa_get(inode);
1892 obdo_from_inode(oinfo.oi_oa, inode,
1893 OBD_MD_FLFID | OBD_MD_FLGENER);
1894 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1895 capa_put(oinfo.oi_capa);
1896 OBDO_FREE(oinfo.oi_oa);
1898 if (rc != -EPERM && rc != -EACCES)
1899 CERROR("md_setattr_async fails: rc = %d\n", rc);
1903 inode->i_flags = ll_ext_to_inode_flags(flags |
1904 MDS_BFLAG_EXT_FLAGS);
1914 int ll_flush_ctx(struct inode *inode)
1916 struct ll_sb_info *sbi = ll_i2sbi(inode);
1918 CDEBUG(D_SEC, "flush context for user %d\n", current->uid);
1920 obd_set_info_async(sbi->ll_md_exp,
1921 sizeof(KEY_FLUSH_CTX) - 1, KEY_FLUSH_CTX,
1923 obd_set_info_async(sbi->ll_dt_exp,
1924 sizeof(KEY_FLUSH_CTX) - 1, KEY_FLUSH_CTX,
1929 /* umount -f client means force down, don't save state */
1930 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1931 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
1933 struct super_block *sb = vfsmnt->mnt_sb;
1935 void ll_umount_begin(struct super_block *sb)
1938 struct lustre_sb_info *lsi = s2lsi(sb);
1939 struct ll_sb_info *sbi = ll_s2sbi(sb);
1940 struct obd_device *obd;
1941 struct obd_ioctl_data ioc_data = { 0 };
1944 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1945 if (!(flags & MNT_FORCE)) {
1951 /* Tell the MGC we got umount -f */
1952 lsi->lsi_flags |= LSI_UMOUNT_FORCE;
1954 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1955 sb->s_count, atomic_read(&sb->s_active));
1957 obd = class_exp2obd(sbi->ll_md_exp);
1959 CERROR("Invalid MDC connection handle "LPX64"\n",
1960 sbi->ll_md_exp->exp_handle.h_cookie);
1965 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp, sizeof ioc_data,
1968 obd = class_exp2obd(sbi->ll_dt_exp);
1970 CERROR("Invalid LOV connection handle "LPX64"\n",
1971 sbi->ll_dt_exp->exp_handle.h_cookie);
1977 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp, sizeof ioc_data,
1980 /* Really, we'd like to wait until there are no requests outstanding,
1981 * and then continue. For now, we just invalidate the requests,
1982 * schedule, and hope.
1989 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
1991 struct ll_sb_info *sbi = ll_s2sbi(sb);
1995 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
1996 read_only = *flags & MS_RDONLY;
1997 err = obd_set_info_async(sbi->ll_md_exp,
1998 sizeof(KEY_READ_ONLY) - 1,
1999 KEY_READ_ONLY, sizeof(read_only),
2002 CERROR("Failed to change the read-only flag during "
2003 "remount: %d\n", err);
2008 sb->s_flags |= MS_RDONLY;
2010 sb->s_flags &= ~MS_RDONLY;
2015 int ll_prep_inode(struct inode **inode,
2016 struct ptlrpc_request *req,
2017 struct super_block *sb)
2019 struct ll_sb_info *sbi = NULL;
2020 struct lustre_md md;
2024 LASSERT(*inode || sb);
2025 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2026 prune_deathrow(sbi, 1);
2027 memset(&md, 0, sizeof(struct lustre_md));
2029 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2030 sbi->ll_md_exp, &md);
2035 ll_update_inode(*inode, &md);
2037 LASSERT(sb != NULL);
2040 * At this point server returns to client's same fid as client
2041 * generated for creating. So using ->fid1 is okay here.
2043 LASSERT(fid_is_sane(&md.body->fid1));
2045 *inode = ll_iget(sb, ll_fid_build_ino(sbi, &md.body->fid1), &md);
2046 if (*inode == NULL || is_bad_inode(*inode)) {
2048 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2049 #ifdef CONFIG_FS_POSIX_ACL
2051 posix_acl_release(md.posix_acl);
2052 md.posix_acl = NULL;
2056 CERROR("new_inode -fatal: rc %d\n", rc);
2061 rc = obd_checkmd(sbi->ll_dt_exp, sbi->ll_md_exp,
2062 ll_i2info(*inode)->lli_smd);
2064 md_free_lustre_md(sbi->ll_md_exp, &md);
2068 char *llap_origins[] = {
2069 [LLAP_ORIGIN_UNKNOWN] = "--",
2070 [LLAP_ORIGIN_READPAGE] = "rp",
2071 [LLAP_ORIGIN_READAHEAD] = "ra",
2072 [LLAP_ORIGIN_COMMIT_WRITE] = "cw",
2073 [LLAP_ORIGIN_WRITEPAGE] = "wp",
2076 struct ll_async_page *llite_pglist_next_llap(struct ll_sb_info *sbi,
2077 struct list_head *list)
2079 struct ll_async_page *llap;
2080 struct list_head *pos;
2082 list_for_each(pos, list) {
2083 if (pos == &sbi->ll_pglist)
2085 llap = list_entry(pos, struct ll_async_page, llap_pglist_item);
2086 if (llap->llap_page == NULL)
2094 int ll_obd_statfs(struct inode *inode, void *arg)
2096 struct ll_sb_info *sbi = NULL;
2097 struct obd_export *exp;
2099 struct obd_ioctl_data *data = NULL;
2103 if (!inode || !(sbi = ll_i2sbi(inode)))
2104 GOTO(out_statfs, rc = -EINVAL);
2106 rc = obd_ioctl_getdata(&buf, &len, arg);
2108 GOTO(out_statfs, rc);
2111 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2112 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2113 GOTO(out_statfs, rc = -EINVAL);
2115 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2116 if (type == LL_STATFS_MDC)
2117 exp = sbi->ll_md_exp;
2118 else if (type == LL_STATFS_LOV)
2119 exp = sbi->ll_dt_exp;
2121 GOTO(out_statfs, rc = -ENODEV);
2123 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2125 GOTO(out_statfs, rc);
2128 obd_ioctl_freedata(buf, len);
2132 int ll_process_config(struct lustre_cfg *lcfg)
2136 struct lprocfs_static_vars lvars;
2140 lprocfs_llite_init_vars(&lvars);
2142 /* The instance name contains the sb: lustre-client-aacfe000 */
2143 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2144 if (!ptr || !*(++ptr))
2146 if (sscanf(ptr, "%lx", &x) != 1)
2149 /* This better be a real Lustre superblock! */
2150 LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2152 /* Note we have not called client_common_fill_super yet, so
2153 proc fns must be able to handle that! */
2154 rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2159 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2160 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2161 struct inode *i1, struct inode *i2,
2162 const char *name, int namelen,
2163 int mode, __u32 opc, void *data)
2165 LASSERT(i1 != NULL);
2167 if (namelen > ll_i2sbi(i1)->ll_namelen)
2168 return ERR_PTR(-ENAMETOOLONG);
2170 if (op_data == NULL)
2171 OBD_ALLOC_PTR(op_data);
2173 if (op_data == NULL)
2174 return ERR_PTR(-ENOMEM);
2176 ll_i2gids(op_data->op_suppgids, i1, i2);
2177 op_data->op_fid1 = *ll_inode2fid(i1);
2178 op_data->op_capa1 = ll_mdscapa_get(i1);
2181 op_data->op_fid2 = *ll_inode2fid(i2);
2182 op_data->op_capa2 = ll_mdscapa_get(i2);
2184 fid_zero(&op_data->op_fid2);
2187 op_data->op_name = name;
2188 op_data->op_namelen = namelen;
2189 op_data->op_mode = mode;
2190 op_data->op_mod_time = CURRENT_SECONDS;
2191 op_data->op_fsuid = current->fsuid;
2192 op_data->op_fsgid = current->fsgid;
2193 op_data->op_cap = current->cap_effective;
2194 op_data->op_bias = MDS_CHECK_SPLIT;
2195 op_data->op_opc = opc;
2196 op_data->op_mds = 0;
2197 op_data->op_data = data;
2202 void ll_finish_md_op_data(struct md_op_data *op_data)
2204 capa_put(op_data->op_capa1);
2205 capa_put(op_data->op_capa2);
2206 OBD_FREE_PTR(op_data);