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_HA, "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_EXPORT___IGET
95 INIT_LIST_HEAD(&sbi->ll_deathrow);
96 spin_lock_init(&sbi->ll_deathrow_lock);
98 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
99 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].pp_r_hist.oh_lock);
100 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].pp_w_hist.oh_lock);
106 void ll_free_sbi(struct super_block *sb)
108 struct ll_sb_info *sbi = ll_s2sbi(sb);
112 spin_lock(&ll_sb_lock);
113 list_del(&sbi->ll_list);
114 spin_unlock(&ll_sb_lock);
115 OBD_FREE(sbi, sizeof(*sbi));
120 static struct dentry_operations ll_d_root_ops = {
121 #ifdef LUSTRE_KERNEL_VERSION
122 .d_compare = ll_dcompare,
126 /* Initialize the default and maximum LOV EA and cookie sizes. This allows
127 * us to make MDS RPCs with large enough reply buffers to hold the
128 * maximum-sized (= maximum striped) EA and cookie without having to
129 * calculate this (via a call into the LOV + OSCs) each time we make an RPC. */
130 static int ll_init_ea_size(struct obd_export *md_exp, struct obd_export *dt_exp)
132 struct lov_stripe_md lsm = { .lsm_magic = LOV_MAGIC };
133 __u32 valsize = sizeof(struct lov_desc);
134 int rc, easize, def_easize, cookiesize;
135 struct lov_desc desc;
139 rc = obd_get_info(dt_exp, strlen(KEY_LOVDESC) + 1, KEY_LOVDESC,
144 stripes = min(desc.ld_tgt_count, (__u32)LOV_MAX_STRIPE_COUNT);
145 lsm.lsm_stripe_count = stripes;
146 easize = obd_size_diskmd(dt_exp, &lsm);
148 lsm.lsm_stripe_count = desc.ld_default_stripe_count;
149 def_easize = obd_size_diskmd(dt_exp, &lsm);
151 cookiesize = stripes * sizeof(struct llog_cookie);
153 CDEBUG(D_HA, "updating max_mdsize/max_cookiesize: %d/%d\n",
156 rc = md_init_ea_size(md_exp, easize, def_easize, cookiesize);
160 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
161 uid_t nllu, gid_t nllg)
163 struct inode *root = 0;
164 struct ll_sb_info *sbi = ll_s2sbi(sb);
165 struct obd_device *obd;
166 struct lu_fid rootfid;
167 struct obd_capa *oc = NULL;
168 struct obd_statfs osfs;
169 struct ptlrpc_request *request = NULL;
170 struct lustre_handle dt_conn = {0, };
171 struct lustre_handle md_conn = {0, };
172 struct obd_connect_data *data = NULL;
173 struct lustre_md lmd;
175 int size, err, checksum;
178 obd = class_name2obd(md);
180 CERROR("MD %s: not setup or attached\n", md);
188 if (proc_lustre_fs_root) {
189 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
192 CERROR("could not register mount in /proc/lustre");
195 /* indicate the features supported by this client */
196 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
198 OBD_CONNECT_ATTRFID | OBD_CONNECT_VERSION |
199 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
200 OBD_CONNECT_CANCELSET;
201 #ifdef CONFIG_FS_POSIX_ACL
202 data->ocd_connect_flags |= OBD_CONNECT_ACL;
204 data->ocd_ibits_known = MDS_INODELOCK_FULL;
205 data->ocd_version = LUSTRE_VERSION_CODE;
207 if (sb->s_flags & MS_RDONLY)
208 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
209 if (sbi->ll_flags & LL_SBI_USER_XATTR)
210 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
212 #ifdef HAVE_MS_FLOCK_LOCK
213 /* force vfs to use lustre handler for flock() calls - bug 10743 */
214 sb->s_flags |= MS_FLOCK_LOCK;
217 if (sbi->ll_flags & LL_SBI_FLOCK)
218 sbi->ll_fop = &ll_file_operations_flock;
219 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
220 sbi->ll_fop = &ll_file_operations;
222 sbi->ll_fop = &ll_file_operations_noflock;
225 data->ocd_connect_flags |= OBD_CONNECT_REAL;
226 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
227 data->ocd_connect_flags &= ~OBD_CONNECT_LCL_CLIENT;
228 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT;
230 data->ocd_connect_flags &= ~OBD_CONNECT_RMT_CLIENT;
231 data->ocd_connect_flags |= OBD_CONNECT_LCL_CLIENT;
233 data->ocd_nllu = nllu;
234 data->ocd_nllg = nllg;
236 err = obd_connect(NULL, &md_conn, obd, &sbi->ll_sb_uuid, data);
238 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
239 "recovery, of which this client is not a "
240 "part. Please wait for recovery to complete,"
241 " abort, or time out.\n", md);
244 CERROR("cannot connect to %s: rc = %d\n", md, err);
247 sbi->ll_md_exp = class_conn2export(&md_conn);
249 err = obd_fid_init(sbi->ll_md_exp);
251 CERROR("Can't init metadata layer FID infrastructure, "
256 err = obd_statfs(obd, &osfs, cfs_time_current_64() - HZ);
258 GOTO(out_md_fid, err);
260 size = sizeof(*data);
261 err = obd_get_info(sbi->ll_md_exp, strlen(KEY_CONN_DATA),
262 KEY_CONN_DATA, &size, data);
264 CERROR("Get connect data failed: %d \n", err);
268 LASSERT(osfs.os_bsize);
269 sb->s_blocksize = osfs.os_bsize;
270 sb->s_blocksize_bits = log2(osfs.os_bsize);
271 sb->s_magic = LL_SUPER_MAGIC;
273 /* for bug 11559. in $LINUX/fs/read_write.c, function do_sendfile():
274 * retval = in_file->f_op->sendfile(...);
276 * retval = -EOVERFLOW;
278 * it will check if *ppos is greater than max. However, max equals to
279 * s_maxbytes, which is a negative integer in a x86_64 box since loff_t
280 * has been defined as a signed long long ineger in linux kernel. */
281 #if BITS_PER_LONG == 64
282 sb->s_maxbytes = PAGE_CACHE_MAXBYTES >> 1;
284 sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
286 sbi->ll_namelen = osfs.os_namelen;
287 sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
289 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
290 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
291 LCONSOLE_INFO("Disabling user_xattr feature because "
292 "it is not supported on the server\n");
293 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
296 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
298 sb->s_flags |= MS_POSIXACL;
300 sbi->ll_flags |= LL_SBI_ACL;
302 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
304 sb->s_flags &= ~MS_POSIXACL;
306 sbi->ll_flags &= ~LL_SBI_ACL;
309 if (data->ocd_connect_flags & OBD_CONNECT_JOIN)
310 sbi->ll_flags |= LL_SBI_JOIN;
312 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
313 if (!(data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT)) {
314 /* sometimes local client claims to be remote, but mdt
315 * will disagree when client gss not applied. */
316 LCONSOLE_INFO("client claims to be remote, but server "
317 "rejected, forced to be local.\n");
318 sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
321 if (!(data->ocd_connect_flags & OBD_CONNECT_LCL_CLIENT)) {
322 /* with gss applied, remote client can not claim to be
323 * local, so mdt maybe force client to be remote. */
324 LCONSOLE_INFO("client claims to be local, but server "
325 "rejected, forced to be remote.\n");
326 sbi->ll_flags |= LL_SBI_RMT_CLIENT;
330 if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
331 LCONSOLE_INFO("client enabled MDS capability!\n");
332 sbi->ll_flags |= LL_SBI_MDS_CAPA;
335 if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
336 LCONSOLE_INFO("client enabled OSS capability!\n");
337 sbi->ll_flags |= LL_SBI_OSS_CAPA;
340 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0))
341 /* We set sb->s_dev equal on all lustre clients in order to support
342 * NFS export clustering. NFSD requires that the FSID be the same
344 /* s_dev is also used in lt_compare() to compare two fs, but that is
345 * only a node-local comparison. */
347 /* XXX: this will not work with LMV */
348 sb->s_dev = get_uuid2int(sbi2mdc(sbi)->cl_target_uuid.uuid,
349 strlen(sbi2mdc(sbi)->cl_target_uuid.uuid));
352 obd = class_name2obd(dt);
354 CERROR("DT %s: not setup or attached\n", dt);
355 GOTO(out_md_fid, err = -ENODEV);
358 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
359 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
360 OBD_CONNECT_CANCELSET;
361 if (sbi->ll_flags & LL_SBI_OSS_CAPA)
362 data->ocd_connect_flags |= OBD_CONNECT_OSS_CAPA;
364 CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
365 "ocd_grant: %d\n", data->ocd_connect_flags,
366 data->ocd_version, data->ocd_grant);
368 obd->obd_upcall.onu_owner = &sbi->ll_lco;
369 obd->obd_upcall.onu_upcall = ll_ocd_update;
370 data->ocd_brw_size = PTLRPC_MAX_BRW_PAGES << CFS_PAGE_SHIFT;
372 err = obd_connect(NULL, &dt_conn, obd, &sbi->ll_sb_uuid, data);
374 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
375 "recovery, of which this client is not a "
376 "part. Please wait for recovery to "
377 "complete, abort, or time out.\n", dt);
378 GOTO(out_md_fid, err);
380 CERROR("Cannot connect to %s: rc = %d\n", dt, err);
381 GOTO(out_md_fid, err);
384 sbi->ll_dt_exp = class_conn2export(&dt_conn);
386 err = obd_fid_init(sbi->ll_dt_exp);
388 CERROR("Can't init data layer FID infrastructure, "
393 spin_lock(&sbi->ll_lco.lco_lock);
394 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
395 spin_unlock(&sbi->ll_lco.lco_lock);
397 ll_init_ea_size(sbi->ll_md_exp, sbi->ll_dt_exp);
399 err = obd_prep_async_page(sbi->ll_dt_exp, NULL, NULL, NULL,
400 0, NULL, NULL, NULL);
402 LCONSOLE_ERROR_MSG(0x151, "There are no OST's in this "
403 "filesystem. There must be at least one "
404 "active OST for a client to start.\n");
405 GOTO(out_dt_fid, err);
408 if (!ll_async_page_slab) {
409 ll_async_page_slab_size =
410 size_round(sizeof(struct ll_async_page)) + err;
411 ll_async_page_slab = cfs_mem_cache_create("ll_async_page",
412 ll_async_page_slab_size,
414 if (!ll_async_page_slab)
415 GOTO(out_dt_fid, err = -ENOMEM);
418 err = md_getstatus(sbi->ll_md_exp, &rootfid, &oc);
420 CERROR("cannot mds_connect: rc = %d\n", err);
421 GOTO(out_dt_fid, err);
423 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&rootfid));
424 sbi->ll_root_fid = rootfid;
426 sb->s_op = &lustre_super_operations;
427 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
428 sb->s_export_op = &lustre_export_operations;
432 * XXX: move this to after cbd setup? */
433 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
434 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
435 valid |= OBD_MD_FLRMTPERM;
436 else if (sbi->ll_flags & LL_SBI_ACL)
437 valid |= OBD_MD_FLACL;
439 err = md_getattr(sbi->ll_md_exp, &rootfid, oc, valid, 0, &request);
443 CERROR("md_getattr failed for root: rc = %d\n", err);
444 GOTO(out_dt_fid, err);
446 memset(&lmd, 0, sizeof(lmd));
447 err = md_get_lustre_md(sbi->ll_md_exp, request,
448 REPLY_REC_OFF, sbi->ll_dt_exp, sbi->ll_md_exp,
451 CERROR("failed to understand root inode md: rc = %d\n", err);
452 ptlrpc_req_finished (request);
453 GOTO(out_dt_fid, err);
456 LASSERT(fid_is_sane(&sbi->ll_root_fid));
457 root = ll_iget(sb, ll_fid_build_ino(sbi, &sbi->ll_root_fid), &lmd);
458 md_free_lustre_md(sbi->ll_md_exp, &lmd);
459 ptlrpc_req_finished(request);
461 if (root == NULL || is_bad_inode(root)) {
463 obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
464 #ifdef CONFIG_FS_POSIX_ACL
466 posix_acl_release(lmd.posix_acl);
467 lmd.posix_acl = NULL;
470 CERROR("lustre_lite: bad iget4 for root\n");
471 GOTO(out_root, err = -EBADF);
474 err = ll_close_thread_start(&sbi->ll_lcq);
476 CERROR("cannot start close thread: rc %d\n", err);
480 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
481 err = obd_set_info_async(sbi->ll_dt_exp, strlen("checksum"),"checksum",
482 sizeof(checksum), &checksum, NULL);
484 /* making vm readahead 0 for 2.4.x. In the case of 2.6.x,
485 backing dev info assigned to inode mapping is used for
486 determining maximal readahead. */
487 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0)) && \
488 !defined(KERNEL_HAS_AS_MAX_READAHEAD)
489 /* bug 2805 - set VM readahead to zero */
490 vm_max_readahead = vm_min_readahead = 0;
493 sb->s_root = d_alloc_root(root);
495 OBD_FREE(data, sizeof(*data));
496 sb->s_root->d_op = &ll_d_root_ops;
502 obd_fid_fini(sbi->ll_dt_exp);
504 obd_disconnect(sbi->ll_dt_exp);
505 sbi->ll_dt_exp = NULL;
507 obd_fid_fini(sbi->ll_md_exp);
509 obd_disconnect(sbi->ll_md_exp);
510 sbi->ll_md_exp = NULL;
514 lprocfs_unregister_mountpoint(sbi);
518 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
522 *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
524 rc = obd_get_info(sbi->ll_md_exp, strlen("max_easize"), "max_easize",
527 CERROR("Get max mdsize error rc %d \n", rc);
532 void ll_dump_inode(struct inode *inode)
534 struct list_head *tmp;
535 int dentry_count = 0;
537 LASSERT(inode != NULL);
539 list_for_each(tmp, &inode->i_dentry)
542 CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
543 inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
544 inode->i_mode, atomic_read(&inode->i_count), dentry_count);
547 void lustre_dump_dentry(struct dentry *dentry, int recur)
549 struct list_head *tmp;
552 LASSERT(dentry != NULL);
554 list_for_each(tmp, &dentry->d_subdirs)
557 CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
558 " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
559 dentry->d_name.len, dentry->d_name.name,
560 dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
561 dentry->d_parent, dentry->d_inode, atomic_read(&dentry->d_count),
562 dentry->d_flags, dentry->d_fsdata, subdirs);
563 if (dentry->d_inode != NULL)
564 ll_dump_inode(dentry->d_inode);
569 list_for_each(tmp, &dentry->d_subdirs) {
570 struct dentry *d = list_entry(tmp, struct dentry, d_child);
571 lustre_dump_dentry(d, recur - 1);
575 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
576 void lustre_throw_orphan_dentries(struct super_block *sb)
578 struct dentry *dentry, *next;
579 struct ll_sb_info *sbi = ll_s2sbi(sb);
581 /* Do this to get rid of orphaned dentries. That is not really trw. */
582 list_for_each_entry_safe(dentry, next, &sbi->ll_orphan_dentry_list,
584 CWARN("found orphan dentry %.*s (%p->%p) at unmount, dumping "
585 "before and after shrink_dcache_parent\n",
586 dentry->d_name.len, dentry->d_name.name, dentry, next);
587 lustre_dump_dentry(dentry, 1);
588 shrink_dcache_parent(dentry);
589 lustre_dump_dentry(dentry, 1);
593 #define lustre_throw_orphan_dentries(sb)
596 #ifdef HAVE_EXPORT___IGET
597 static void prune_dir_dentries(struct inode *inode)
599 struct dentry *dentry, *prev = NULL;
601 /* due to lustre specific logic, a directory
602 * can have few dentries - a bug from VFS POV */
604 spin_lock(&dcache_lock);
605 if (!list_empty(&inode->i_dentry)) {
606 dentry = list_entry(inode->i_dentry.prev,
607 struct dentry, d_alias);
608 /* in order to prevent infinite loops we
609 * break if previous dentry is busy */
610 if (dentry != prev) {
613 spin_unlock(&dcache_lock);
615 /* try to kill all child dentries */
617 shrink_dcache_parent(dentry);
618 unlock_dentry(dentry);
621 /* now try to get rid of current dentry */
622 d_prune_aliases(inode);
626 spin_unlock(&dcache_lock);
629 static void prune_deathrow_one(struct ll_inode_info *lli)
631 struct inode *inode = ll_info2i(lli);
633 /* first, try to drop any dentries - they hold a ref on the inode */
634 if (S_ISDIR(inode->i_mode))
635 prune_dir_dentries(inode);
637 d_prune_aliases(inode);
640 /* if somebody still uses it, leave it */
641 LASSERT(atomic_read(&inode->i_count) > 0);
642 if (atomic_read(&inode->i_count) > 1)
645 CDEBUG(D_INODE, "inode %lu/%u(%d) looks a good candidate for prune\n",
646 inode->i_ino,inode->i_generation, atomic_read(&inode->i_count));
648 /* seems nobody uses it anymore */
656 static void prune_deathrow(struct ll_sb_info *sbi, int try)
658 struct ll_inode_info *lli;
662 if (need_resched() && try)
666 if (!spin_trylock(&sbi->ll_deathrow_lock))
669 spin_lock(&sbi->ll_deathrow_lock);
674 if (!list_empty(&sbi->ll_deathrow)) {
675 lli = list_entry(sbi->ll_deathrow.next,
676 struct ll_inode_info,
678 list_del_init(&lli->lli_dead_list);
679 if (!list_empty(&sbi->ll_deathrow))
682 spin_unlock(&sbi->ll_deathrow_lock);
685 prune_deathrow_one(lli);
687 } while (empty == 0);
689 #else /* !HAVE_EXPORT___IGET */
690 #define prune_deathrow(sbi, try) do {} while (0)
691 #endif /* HAVE_EXPORT___IGET */
693 void client_common_put_super(struct super_block *sb)
695 struct ll_sb_info *sbi = ll_s2sbi(sb);
698 obd_cancel_unused(sbi->ll_dt_exp, NULL, 0, NULL);
700 ll_close_thread_shutdown(sbi->ll_lcq);
702 /* destroy inodes in deathrow */
703 prune_deathrow(sbi, 0);
705 list_del(&sbi->ll_conn_chain);
707 obd_fid_fini(sbi->ll_dt_exp);
708 obd_disconnect(sbi->ll_dt_exp);
709 sbi->ll_dt_exp = NULL;
711 lprocfs_unregister_mountpoint(sbi);
713 obd_fid_fini(sbi->ll_md_exp);
714 obd_disconnect(sbi->ll_md_exp);
715 sbi->ll_md_exp = NULL;
717 lustre_throw_orphan_dentries(sb);
721 char *ll_read_opt(const char *opt, char *data)
727 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
728 if (strncmp(opt, data, strlen(opt)))
730 if ((value = strchr(data, '=')) == NULL)
734 OBD_ALLOC(retval, strlen(value) + 1);
736 CERROR("out of memory!\n");
740 memcpy(retval, value, strlen(value)+1);
741 CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
745 static inline int ll_set_opt(const char *opt, char *data, int fl)
747 if (strncmp(opt, data, strlen(opt)) != 0)
753 /* non-client-specific mount options are parsed in lmd_parse */
754 static int ll_options(char *options, int *flags)
757 char *s1 = options, *s2;
763 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
766 CDEBUG(D_SUPER, "next opt=%s\n", s1);
767 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
772 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
777 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
782 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
787 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
792 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
797 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
799 /* Ignore deprecated mount option. The client will
800 * always try to mount with ACL support, whether this
801 * is used depends on whether server supports it. */
804 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
808 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
814 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
819 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
825 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
831 s2 = strchr(s1, ',');
839 void ll_lli_init(struct ll_inode_info *lli)
841 lli->lli_inode_magic = LLI_INODE_MAGIC;
842 sema_init(&lli->lli_size_sem, 1);
843 sema_init(&lli->lli_write_sem, 1);
845 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
846 spin_lock_init(&lli->lli_lock);
847 INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
848 INIT_LIST_HEAD(&lli->lli_close_list);
849 lli->lli_inode_magic = LLI_INODE_MAGIC;
850 sema_init(&lli->lli_och_sem, 1);
851 lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
852 lli->lli_mds_exec_och = NULL;
853 lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
854 lli->lli_open_fd_exec_count = 0;
855 INIT_LIST_HEAD(&lli->lli_dead_list);
856 lli->lli_remote_perms = NULL;
857 lli->lli_rmtperm_utime = 0;
858 sema_init(&lli->lli_rmtperm_sem, 1);
859 INIT_LIST_HEAD(&lli->lli_oss_capas);
862 int ll_fill_super(struct super_block *sb)
864 struct lustre_profile *lprof;
865 struct lustre_sb_info *lsi = s2lsi(sb);
866 struct ll_sb_info *sbi;
867 char *dt = NULL, *md = NULL;
868 char *profilenm = get_profile_name(sb);
869 struct config_llog_instance cfg = {0, };
870 char ll_instance[sizeof(sb) * 2 + 1];
874 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
878 sb->s_type->fs_flags |= FS_ODD_RENAME;
879 /* client additional sb info */
880 lsi->lsi_llsbi = sbi = ll_init_sbi();
886 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
890 /* Generate a string unique to this super, in case some joker tries
891 to mount the same fs at two mount points.
892 Use the address of the super itself.*/
893 sprintf(ll_instance, "%p", sb);
894 cfg.cfg_instance = ll_instance;
895 cfg.cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
897 /* set up client obds */
898 err = lustre_process_log(sb, profilenm, &cfg);
900 CERROR("Unable to process log: %d\n", err);
904 lprof = class_get_profile(profilenm);
906 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
907 " read from the MGS. Does that filesystem "
908 "exist?\n", profilenm);
909 GOTO(out_free, err = -EINVAL);
911 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
912 lprof->lp_md, lprof->lp_dt);
914 OBD_ALLOC(dt, strlen(lprof->lp_dt) +
915 strlen(ll_instance) + 2);
917 GOTO(out_free, err = -ENOMEM);
918 sprintf(dt, "%s-%s", lprof->lp_dt, ll_instance);
920 OBD_ALLOC(md, strlen(lprof->lp_md) +
921 strlen(ll_instance) + 2);
923 GOTO(out_free, err = -ENOMEM);
924 sprintf(md, "%s-%s", lprof->lp_md, ll_instance);
926 /* connections, registrations, sb setup */
927 err = client_common_fill_super(sb, md, dt,
928 lsi->lsi_lmd->lmd_nllu,
929 lsi->lsi_lmd->lmd_nllg);
933 OBD_FREE(md, strlen(md) + 1);
935 OBD_FREE(dt, strlen(dt) + 1);
939 LCONSOLE_WARN("Client %s has started\n", profilenm);
942 } /* ll_fill_super */
945 void ll_put_super(struct super_block *sb)
947 struct config_llog_instance cfg;
948 char ll_instance[sizeof(sb) * 2 + 1];
949 struct obd_device *obd;
950 struct lustre_sb_info *lsi = s2lsi(sb);
951 struct ll_sb_info *sbi = ll_s2sbi(sb);
952 char *profilenm = get_profile_name(sb);
956 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
958 ll_print_capa_stat(sbi);
960 sprintf(ll_instance, "%p", sb);
961 cfg.cfg_instance = ll_instance;
962 lustre_end_log(sb, NULL, &cfg);
964 if (sbi->ll_md_exp) {
965 obd = class_exp2obd(sbi->ll_md_exp);
967 force = obd->obd_force;
970 /* We need to set force before the lov_disconnect in
971 lustre_common_put_super, since l_d cleans up osc's as well. */
974 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
976 obd->obd_force = force;
981 /* Only if client_common_fill_super succeeded */
982 client_common_put_super(sb);
985 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
986 class_manual_cleanup(obd);
990 class_del_profile(profilenm);
993 lsi->lsi_llsbi = NULL;
995 lustre_common_put_super(sb);
997 LCONSOLE_WARN("client %s umount complete\n", ll_instance);
1002 } /* client_put_super */
1004 #ifdef HAVE_REGISTER_CACHE
1005 #include <linux/cache_def.h>
1006 #ifdef HAVE_CACHE_RETURN_INT
1011 ll_shrink_cache(int priority, unsigned int gfp_mask)
1013 struct ll_sb_info *sbi;
1016 list_for_each_entry(sbi, &ll_super_blocks, ll_list)
1017 count += llap_shrink_cache(sbi, priority);
1019 #ifdef HAVE_CACHE_RETURN_INT
1024 struct cache_definition ll_cache_definition = {
1025 .name = "llap_cache",
1026 .shrink = ll_shrink_cache
1028 #endif /* HAVE_REGISTER_CACHE */
1030 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1032 struct inode *inode = NULL;
1033 /* NOTE: we depend on atomic igrab() -bzzz */
1034 lock_res_and_lock(lock);
1035 if (lock->l_ast_data) {
1036 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1037 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1038 inode = igrab(lock->l_ast_data);
1040 inode = lock->l_ast_data;
1041 ldlm_lock_debug(NULL, inode->i_state & I_FREEING ?
1043 lock, __FILE__, __func__, __LINE__,
1044 "l_ast_data %p is bogus: magic %08x",
1045 lock->l_ast_data, lli->lli_inode_magic);
1049 unlock_res_and_lock(lock);
1053 static int null_if_equal(struct ldlm_lock *lock, void *data)
1055 if (data == lock->l_ast_data) {
1056 lock->l_ast_data = NULL;
1058 if (lock->l_req_mode != lock->l_granted_mode)
1059 LDLM_ERROR(lock,"clearing inode with ungranted lock");
1062 return LDLM_ITER_CONTINUE;
1065 void ll_clear_inode(struct inode *inode)
1067 struct ll_inode_info *lli = ll_i2info(inode);
1068 struct ll_sb_info *sbi = ll_i2sbi(inode);
1071 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1072 inode->i_generation, inode);
1074 ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1075 md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1076 null_if_equal, inode);
1078 LASSERT(!lli->lli_open_fd_write_count);
1079 LASSERT(!lli->lli_open_fd_read_count);
1080 LASSERT(!lli->lli_open_fd_exec_count);
1082 if (lli->lli_mds_write_och)
1083 ll_md_real_close(inode, FMODE_WRITE);
1084 if (lli->lli_mds_exec_och)
1085 ll_md_real_close(inode, FMODE_EXEC);
1086 if (lli->lli_mds_read_och)
1087 ll_md_real_close(inode, FMODE_READ);
1090 obd_change_cbdata(sbi->ll_dt_exp, lli->lli_smd,
1091 null_if_equal, inode);
1093 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1094 lli->lli_smd = NULL;
1097 if (lli->lli_symlink_name) {
1098 OBD_FREE(lli->lli_symlink_name,
1099 strlen(lli->lli_symlink_name) + 1);
1100 lli->lli_symlink_name = NULL;
1103 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1104 LASSERT(lli->lli_posix_acl == NULL);
1105 if (lli->lli_remote_perms) {
1106 free_rmtperm_hash(lli->lli_remote_perms);
1107 lli->lli_remote_perms = NULL;
1110 #ifdef CONFIG_FS_POSIX_ACL
1111 else if (lli->lli_posix_acl) {
1112 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1113 LASSERT(lli->lli_remote_perms == NULL);
1114 posix_acl_release(lli->lli_posix_acl);
1115 lli->lli_posix_acl = NULL;
1118 lli->lli_inode_magic = LLI_INODE_DEAD;
1120 #ifdef HAVE_EXPORT___IGET
1121 spin_lock(&sbi->ll_deathrow_lock);
1122 list_del_init(&lli->lli_dead_list);
1123 spin_unlock(&sbi->ll_deathrow_lock);
1125 ll_clear_inode_capas(inode);
1130 int ll_md_setattr(struct inode *inode, struct md_op_data *op_data,
1131 struct md_open_data **mod)
1133 struct lustre_md md;
1134 struct ll_sb_info *sbi = ll_i2sbi(inode);
1135 struct ptlrpc_request *request = NULL;
1139 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1140 LUSTRE_OPC_ANY, NULL);
1141 if (IS_ERR(op_data))
1142 RETURN(PTR_ERR(op_data));
1144 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1147 ptlrpc_req_finished(request);
1148 if (rc == -ENOENT) {
1150 /* Unlinked special device node? Or just a race?
1151 * Pretend we done everything. */
1152 if (!S_ISREG(inode->i_mode) &&
1153 !S_ISDIR(inode->i_mode))
1154 rc = inode_setattr(inode, &op_data->op_attr);
1155 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1156 CERROR("md_setattr fails: rc = %d\n", rc);
1161 rc = md_get_lustre_md(sbi->ll_md_exp, request, REPLY_REC_OFF,
1162 sbi->ll_dt_exp, sbi->ll_md_exp, &md);
1164 ptlrpc_req_finished(request);
1168 /* We call inode_setattr to adjust timestamps.
1169 * If there is at least some data in file, we cleared ATTR_SIZE
1170 * above to avoid invoking vmtruncate, otherwise it is important
1171 * to call vmtruncate in inode_setattr to update inode->i_size
1173 rc = inode_setattr(inode, &op_data->op_attr);
1175 /* Extract epoch data if obtained. */
1176 op_data->op_handle = md.body->handle;
1177 op_data->op_ioepoch = md.body->ioepoch;
1179 ll_update_inode(inode, &md);
1180 ptlrpc_req_finished(request);
1185 /* Close IO epoch and send Size-on-MDS attribute update. */
1186 static int ll_setattr_done_writing(struct inode *inode,
1187 struct md_op_data *op_data,
1188 struct md_open_data *mod)
1190 struct ll_inode_info *lli = ll_i2info(inode);
1194 LASSERT(op_data != NULL);
1195 if (!S_ISREG(inode->i_mode))
1198 CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1199 op_data->op_ioepoch, PFID(&lli->lli_fid));
1201 op_data->op_flags = MF_EPOCH_CLOSE | MF_SOM_CHANGE;
1202 rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1203 if (rc == -EAGAIN) {
1204 /* MDS has instructed us to obtain Size-on-MDS attribute
1205 * from OSTs and send setattr to back to MDS. */
1206 rc = ll_sizeonmds_update(inode, mod, &op_data->op_handle,
1207 op_data->op_ioepoch);
1209 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1215 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1216 * object(s) determine the file size and mtime. Otherwise, the MDS will
1217 * keep these values until such a time that objects are allocated for it.
1218 * We do the MDS operations first, as it is checking permissions for us.
1219 * We don't to the MDS RPC if there is nothing that we want to store there,
1220 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1221 * going to do an RPC anyways.
1223 * If we are doing a truncate, we will send the mtime and ctime updates
1224 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1225 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1228 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1230 struct ll_inode_info *lli = ll_i2info(inode);
1231 struct lov_stripe_md *lsm = lli->lli_smd;
1232 struct ll_sb_info *sbi = ll_i2sbi(inode);
1233 struct md_op_data *op_data = NULL;
1234 struct md_open_data *mod = NULL;
1235 int ia_valid = attr->ia_valid;
1236 int rc = 0, rc1 = 0;
1239 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1241 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_SETATTR, 1);
1243 if (ia_valid & ATTR_SIZE) {
1244 if (attr->ia_size > ll_file_maxbytes(inode)) {
1245 CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1246 attr->ia_size, ll_file_maxbytes(inode));
1250 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1253 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1254 if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1255 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1259 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1260 if (attr->ia_valid & ATTR_CTIME) {
1261 attr->ia_ctime = CURRENT_TIME;
1262 attr->ia_valid |= ATTR_CTIME_SET;
1264 if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1265 attr->ia_atime = CURRENT_TIME;
1266 attr->ia_valid |= ATTR_ATIME_SET;
1268 if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1269 attr->ia_mtime = CURRENT_TIME;
1270 attr->ia_valid |= ATTR_MTIME_SET;
1272 if ((attr->ia_valid & ATTR_CTIME) && !(attr->ia_valid & ATTR_MTIME)) {
1273 /* To avoid stale mtime on mds, obtain it from ost and send
1275 rc = ll_glimpse_size(inode, 0);
1279 attr->ia_valid |= ATTR_MTIME_SET | ATTR_MTIME;
1280 attr->ia_mtime = inode->i_mtime;
1283 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1284 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1285 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1288 /* NB: ATTR_SIZE will only be set after this point if the size
1289 * resides on the MDS, ie, this file has no objects. */
1291 attr->ia_valid &= ~ATTR_SIZE;
1293 /* We always do an MDS RPC, even if we're only changing the size;
1294 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1296 OBD_ALLOC_PTR(op_data);
1297 if (op_data == NULL)
1300 memcpy(&op_data->op_attr, attr, sizeof(*attr));
1302 /* Open epoch for truncate. */
1303 if (ia_valid & ATTR_SIZE)
1304 op_data->op_flags = MF_EPOCH_OPEN;
1306 rc = ll_md_setattr(inode, op_data, &mod);
1310 if (op_data->op_ioepoch)
1311 CDEBUG(D_INODE, "Epoch "LPU64" opened on "DFID" for "
1312 "truncate\n", op_data->op_ioepoch, PFID(&lli->lli_fid));
1314 if (!lsm || !S_ISREG(inode->i_mode)) {
1315 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1319 /* We really need to get our PW lock before we change inode->i_size.
1320 * If we don't we can race with other i_size updaters on our node, like
1321 * ll_file_read. We can also race with i_size propogation to other
1322 * nodes through dirtying and writeback of final cached pages. This
1323 * last one is especially bad for racing o_append users on other
1325 if (ia_valid & ATTR_SIZE) {
1326 ldlm_policy_data_t policy = { .l_extent = {attr->ia_size,
1328 struct lustre_handle lockh = { 0 };
1329 int err, ast_flags = 0;
1330 /* XXX when we fix the AST intents to pass the discard-range
1331 * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
1333 if (attr->ia_size == 0)
1334 ast_flags = LDLM_AST_DISCARD_DATA;
1336 UNLOCK_INODE_MUTEX(inode);
1337 UP_WRITE_I_ALLOC_SEM(inode);
1338 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy, &lockh,
1340 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1341 DOWN_WRITE_I_ALLOC_SEM(inode);
1342 LOCK_INODE_MUTEX(inode);
1344 LOCK_INODE_MUTEX(inode);
1345 DOWN_WRITE_I_ALLOC_SEM(inode);
1350 /* Only ll_inode_size_lock is taken at this level.
1351 * lov_stripe_lock() is grabbed by ll_truncate() only over
1352 * call to obd_adjust_kms(). If vmtruncate returns 0, then
1353 * ll_truncate dropped ll_inode_size_lock() */
1354 ll_inode_size_lock(inode, 0);
1355 rc = vmtruncate(inode, attr->ia_size);
1357 LASSERT(atomic_read(&lli->lli_size_sem.count) <= 0);
1358 ll_inode_size_unlock(inode, 0);
1361 err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
1363 CERROR("ll_extent_unlock failed: %d\n", err);
1367 } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
1369 struct obd_info oinfo = { { { 0 } } };
1372 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1373 inode->i_ino, LTIME_S(attr->ia_mtime));
1377 oa->o_id = lsm->lsm_object_id;
1378 oa->o_gr = lsm->lsm_object_gr;
1379 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1381 flags = OBD_MD_FLTYPE | OBD_MD_FLATIME |
1382 OBD_MD_FLMTIME | OBD_MD_FLCTIME |
1383 OBD_MD_FLFID | OBD_MD_FLGENER |
1386 obdo_from_inode(oa, inode, flags);
1390 oinfo.oi_capa = ll_mdscapa_get(inode);
1392 /* XXX: this looks unnecessary now. */
1393 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1394 capa_put(oinfo.oi_capa);
1396 CERROR("obd_setattr_async fails: rc=%d\n", rc);
1405 if (op_data->op_ioepoch)
1406 rc1 = ll_setattr_done_writing(inode, op_data, mod);
1407 ll_finish_md_op_data(op_data);
1409 return rc ? rc : rc1;
1412 int ll_setattr(struct dentry *de, struct iattr *attr)
1414 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1415 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1416 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1418 return ll_setattr_raw(de->d_inode, attr);
1421 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1424 struct ll_sb_info *sbi = ll_s2sbi(sb);
1425 struct obd_statfs obd_osfs;
1429 rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age);
1431 CERROR("md_statfs fails: rc = %d\n", rc);
1435 osfs->os_type = sb->s_magic;
1437 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1438 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1440 rc = obd_statfs_rqset(class_exp2obd(sbi->ll_dt_exp),
1441 &obd_osfs, max_age);
1443 CERROR("obd_statfs fails: rc = %d\n", rc);
1447 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1448 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1451 osfs->os_bsize = obd_osfs.os_bsize;
1452 osfs->os_blocks = obd_osfs.os_blocks;
1453 osfs->os_bfree = obd_osfs.os_bfree;
1454 osfs->os_bavail = obd_osfs.os_bavail;
1456 /* If we don't have as many objects free on the OST as inodes
1457 * on the MDS, we reduce the total number of inodes to
1458 * compensate, so that the "inodes in use" number is correct.
1460 if (obd_osfs.os_ffree < osfs->os_ffree) {
1461 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1463 osfs->os_ffree = obd_osfs.os_ffree;
1468 #ifndef HAVE_STATFS_DENTRY_PARAM
1469 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1472 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1474 struct super_block *sb = de->d_sb;
1476 struct obd_statfs osfs;
1479 CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1480 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1482 /* For now we will always get up-to-date statfs values, but in the
1483 * future we may allow some amount of caching on the client (e.g.
1484 * from QOS or lprocfs updates). */
1485 rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - 1);
1489 statfs_unpack(sfs, &osfs);
1491 /* We need to downshift for all 32-bit kernels, because we can't
1492 * tell if the kernel is being called via sys_statfs64() or not.
1493 * Stop before overflowing f_bsize - in which case it is better
1494 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1495 if (sizeof(long) < 8) {
1496 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1499 osfs.os_blocks >>= 1;
1500 osfs.os_bfree >>= 1;
1501 osfs.os_bavail >>= 1;
1505 sfs->f_blocks = osfs.os_blocks;
1506 sfs->f_bfree = osfs.os_bfree;
1507 sfs->f_bavail = osfs.os_bavail;
1512 void ll_inode_size_lock(struct inode *inode, int lock_lsm)
1514 struct ll_inode_info *lli;
1515 struct lov_stripe_md *lsm;
1517 lli = ll_i2info(inode);
1518 LASSERT(lli->lli_size_sem_owner != current);
1519 down(&lli->lli_size_sem);
1520 LASSERT(lli->lli_size_sem_owner == NULL);
1521 lli->lli_size_sem_owner = current;
1523 LASSERTF(lsm != NULL || lock_lsm == 0, "lsm %p, lock_lsm %d\n",
1526 lov_stripe_lock(lsm);
1529 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm)
1531 struct ll_inode_info *lli;
1532 struct lov_stripe_md *lsm;
1534 lli = ll_i2info(inode);
1536 LASSERTF(lsm != NULL || unlock_lsm == 0, "lsm %p, lock_lsm %d\n",
1539 lov_stripe_unlock(lsm);
1540 LASSERT(lli->lli_size_sem_owner == current);
1541 lli->lli_size_sem_owner = NULL;
1542 up(&lli->lli_size_sem);
1545 static void ll_replace_lsm(struct inode *inode, struct lov_stripe_md *lsm)
1547 struct ll_inode_info *lli = ll_i2info(inode);
1549 dump_lsm(D_INODE, lsm);
1550 dump_lsm(D_INODE, lli->lli_smd);
1551 LASSERTF(lsm->lsm_magic == LOV_MAGIC_JOIN,
1552 "lsm must be joined lsm %p\n", lsm);
1553 obd_free_memmd(ll_i2dtexp(inode), &lli->lli_smd);
1554 CDEBUG(D_INODE, "replace lsm %p to lli_smd %p for inode %lu%u(%p)\n",
1555 lsm, lli->lli_smd, inode->i_ino, inode->i_generation, inode);
1557 lli->lli_maxbytes = lsm->lsm_maxbytes;
1558 if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1559 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1562 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1564 struct ll_inode_info *lli = ll_i2info(inode);
1565 struct mdt_body *body = md->body;
1566 struct lov_stripe_md *lsm = md->lsm;
1567 struct ll_sb_info *sbi = ll_i2sbi(inode);
1569 LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1571 if (lli->lli_smd == NULL) {
1572 if (lsm->lsm_magic != LOV_MAGIC &&
1573 lsm->lsm_magic != LOV_MAGIC_JOIN) {
1574 dump_lsm(D_ERROR, lsm);
1577 CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1578 lsm, inode->i_ino, inode->i_generation, inode);
1579 /* ll_inode_size_lock() requires it is only called
1580 * with lli_smd != NULL or lock_lsm == 0 or we can
1581 * race between lock/unlock. bug 9547 */
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 ll_replace_lsm(inode, lsm);
1602 if (lli->lli_smd != lsm)
1603 obd_free_memmd(ll_i2dtexp(inode), &lsm);
1606 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1607 if (body->valid & OBD_MD_FLRMTPERM)
1608 ll_update_remote_perm(inode, md->remote_perm);
1610 #ifdef CONFIG_FS_POSIX_ACL
1611 else if (body->valid & OBD_MD_FLACL) {
1612 spin_lock(&lli->lli_lock);
1613 if (lli->lli_posix_acl)
1614 posix_acl_release(lli->lli_posix_acl);
1615 lli->lli_posix_acl = md->posix_acl;
1616 spin_unlock(&lli->lli_lock);
1619 if (body->valid & OBD_MD_FLATIME &&
1620 body->atime > LTIME_S(inode->i_atime))
1621 LTIME_S(inode->i_atime) = body->atime;
1623 /* mtime is always updated with ctime, but can be set in past.
1624 As write and utime(2) may happen within 1 second, and utime's
1625 mtime has a priority over write's one, so take mtime from mds
1626 for the same ctimes. */
1627 if (body->valid & OBD_MD_FLCTIME &&
1628 body->ctime >= LTIME_S(inode->i_ctime)) {
1629 LTIME_S(inode->i_ctime) = body->ctime;
1630 if (body->valid & OBD_MD_FLMTIME) {
1631 CDEBUG(D_INODE, "setting ino %lu mtime "
1632 "from %lu to "LPU64"\n", inode->i_ino,
1633 LTIME_S(inode->i_mtime), body->mtime);
1634 LTIME_S(inode->i_mtime) = body->mtime;
1637 if (body->valid & OBD_MD_FLMODE)
1638 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1639 if (body->valid & OBD_MD_FLTYPE)
1640 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1641 if (S_ISREG(inode->i_mode)) {
1642 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1644 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1646 #ifdef HAVE_INODE_BLKSIZE
1647 inode->i_blksize = 1<<inode->i_blkbits;
1649 if (body->valid & OBD_MD_FLUID)
1650 inode->i_uid = body->uid;
1651 if (body->valid & OBD_MD_FLGID)
1652 inode->i_gid = body->gid;
1653 if (body->valid & OBD_MD_FLFLAGS)
1654 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1655 if (body->valid & OBD_MD_FLNLINK)
1656 inode->i_nlink = body->nlink;
1657 if (body->valid & OBD_MD_FLRDEV)
1658 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1659 inode->i_rdev = body->rdev;
1661 inode->i_rdev = old_decode_dev(body->rdev);
1663 if (body->valid & OBD_MD_FLSIZE) {
1664 if (ll_i2mdexp(inode)->exp_connect_flags & OBD_CONNECT_SOM) {
1665 if (lli->lli_flags & (LLIF_DONE_WRITING |
1666 LLIF_EPOCH_PENDING |
1668 CWARN("ino %lu flags %lu still has size authority!"
1669 "do not trust the size got from MDS\n",
1670 inode->i_ino, lli->lli_flags);
1672 i_size_write(inode, body->size);
1673 lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1676 i_size_write(inode, body->size);
1679 if (body->valid & OBD_MD_FLBLOCKS)
1680 inode->i_blocks = body->blocks;
1683 if (body->valid & OBD_MD_FLID) {
1684 /* FID shouldn't be changed! */
1685 if (fid_is_sane(&lli->lli_fid)) {
1686 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1687 "Trying to change FID "DFID
1688 " to the "DFID", inode %lu/%u(%p)\n",
1689 PFID(&lli->lli_fid), PFID(&body->fid1),
1690 inode->i_ino, inode->i_generation, inode);
1692 lli->lli_fid = body->fid1;
1695 LASSERT(fid_seq(&lli->lli_fid) != 0);
1697 if (body->valid & OBD_MD_FLMDSCAPA) {
1698 LASSERT(md->mds_capa);
1699 ll_add_capa(inode, md->mds_capa);
1701 if (body->valid & OBD_MD_FLOSSCAPA) {
1702 LASSERT(md->oss_capa);
1703 ll_add_capa(inode, md->oss_capa);
1707 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
1708 static struct backing_dev_info ll_backing_dev_info = {
1709 .ra_pages = 0, /* No readahead */
1710 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,12))
1711 .capabilities = 0, /* Does contribute to dirty memory */
1713 .memory_backed = 0, /* Does contribute to dirty memory */
1718 void ll_read_inode2(struct inode *inode, void *opaque)
1720 struct lustre_md *md = opaque;
1721 struct ll_inode_info *lli = ll_i2info(inode);
1724 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n",
1725 inode->i_ino, inode->i_generation, inode);
1729 LASSERT(!lli->lli_smd);
1731 /* Core attributes from the MDS first. This is a new inode, and
1732 * the VFS doesn't zero times in the core inode so we have to do
1733 * it ourselves. They will be overwritten by either MDS or OST
1734 * attributes - we just need to make sure they aren't newer. */
1735 LTIME_S(inode->i_mtime) = 0;
1736 LTIME_S(inode->i_atime) = 0;
1737 LTIME_S(inode->i_ctime) = 0;
1739 ll_update_inode(inode, md);
1741 /* OIDEBUG(inode); */
1743 if (S_ISREG(inode->i_mode)) {
1744 struct ll_sb_info *sbi = ll_i2sbi(inode);
1745 inode->i_op = &ll_file_inode_operations;
1746 inode->i_fop = sbi->ll_fop;
1747 inode->i_mapping->a_ops = &ll_aops;
1749 } else if (S_ISDIR(inode->i_mode)) {
1750 inode->i_op = &ll_dir_inode_operations;
1751 inode->i_fop = &ll_dir_operations;
1752 inode->i_mapping->a_ops = &ll_dir_aops;
1754 } else if (S_ISLNK(inode->i_mode)) {
1755 inode->i_op = &ll_fast_symlink_inode_operations;
1758 inode->i_op = &ll_special_inode_operations;
1760 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
1761 init_special_inode(inode, inode->i_mode,
1762 kdev_t_to_nr(inode->i_rdev));
1764 /* initializing backing dev info. */
1765 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
1767 init_special_inode(inode, inode->i_mode, inode->i_rdev);
1773 void ll_delete_inode(struct inode *inode)
1775 struct ll_sb_info *sbi = ll_i2sbi(inode);
1779 rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
1781 CERROR("fid_delete() failed, rc %d\n", rc);
1788 int ll_iocontrol(struct inode *inode, struct file *file,
1789 unsigned int cmd, unsigned long arg)
1791 struct ll_sb_info *sbi = ll_i2sbi(inode);
1792 struct ptlrpc_request *req = NULL;
1797 case EXT3_IOC_GETFLAGS: {
1798 struct mdt_body *body;
1799 struct obd_capa *oc;
1801 oc = ll_mdscapa_get(inode);
1802 rc = md_getattr(sbi->ll_md_exp, ll_inode2fid(inode), oc,
1803 OBD_MD_FLFLAGS, 0, &req);
1806 CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1810 body = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
1813 /*Now the ext3 will be packed directly back to client,
1814 *no need convert here*/
1815 flags = body->flags;
1817 ptlrpc_req_finished (req);
1819 RETURN(put_user(flags, (int *)arg));
1821 case EXT3_IOC_SETFLAGS: {
1822 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1823 struct obd_info oinfo = { { { 0 } } };
1824 struct md_op_data *op_data;
1826 if (get_user(flags, (int *)arg))
1830 OBDO_ALLOC(oinfo.oi_oa);
1834 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1835 LUSTRE_OPC_ANY, NULL);
1836 if (IS_ERR(op_data))
1837 RETURN(PTR_ERR(op_data));
1839 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1840 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1841 rc = md_setattr(sbi->ll_md_exp, op_data,
1842 NULL, 0, NULL, 0, &req, NULL);
1843 ll_finish_md_op_data(op_data);
1844 ptlrpc_req_finished(req);
1845 if (rc || lsm == NULL) {
1846 OBDO_FREE(oinfo.oi_oa);
1850 oinfo.oi_oa->o_id = lsm->lsm_object_id;
1851 oinfo.oi_oa->o_gr = lsm->lsm_object_gr;
1852 oinfo.oi_oa->o_flags = flags;
1853 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1855 oinfo.oi_capa = ll_mdscapa_get(inode);
1857 obdo_from_inode(oinfo.oi_oa, inode,
1858 OBD_MD_FLFID | OBD_MD_FLGENER);
1859 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1860 capa_put(oinfo.oi_capa);
1861 OBDO_FREE(oinfo.oi_oa);
1863 if (rc != -EPERM && rc != -EACCES)
1864 CERROR("md_setattr_async fails: rc = %d\n", rc);
1868 inode->i_flags = ll_ext_to_inode_flags(flags |
1869 MDS_BFLAG_EXT_FLAGS);
1879 int ll_flush_ctx(struct inode *inode)
1881 struct ll_sb_info *sbi = ll_i2sbi(inode);
1883 CDEBUG(D_SEC, "flush context for user %d\n", current->uid);
1885 obd_set_info_async(sbi->ll_md_exp,
1886 sizeof(KEY_FLUSH_CTX) - 1, KEY_FLUSH_CTX,
1888 obd_set_info_async(sbi->ll_dt_exp,
1889 sizeof(KEY_FLUSH_CTX) - 1, KEY_FLUSH_CTX,
1894 /* umount -f client means force down, don't save state */
1895 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1896 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
1898 struct super_block *sb = vfsmnt->mnt_sb;
1900 void ll_umount_begin(struct super_block *sb)
1903 struct lustre_sb_info *lsi = s2lsi(sb);
1904 struct ll_sb_info *sbi = ll_s2sbi(sb);
1905 struct obd_device *obd;
1906 struct obd_ioctl_data ioc_data = { 0 };
1909 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1910 if (!(flags & MNT_FORCE)) {
1916 /* Tell the MGC we got umount -f */
1917 lsi->lsi_flags |= LSI_UMOUNT_FORCE;
1919 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1920 sb->s_count, atomic_read(&sb->s_active));
1922 obd = class_exp2obd(sbi->ll_md_exp);
1924 CERROR("Invalid MDC connection handle "LPX64"\n",
1925 sbi->ll_md_exp->exp_handle.h_cookie);
1930 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp, sizeof ioc_data,
1933 obd = class_exp2obd(sbi->ll_dt_exp);
1935 CERROR("Invalid LOV connection handle "LPX64"\n",
1936 sbi->ll_dt_exp->exp_handle.h_cookie);
1942 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp, sizeof ioc_data,
1945 /* Really, we'd like to wait until there are no requests outstanding,
1946 * and then continue. For now, we just invalidate the requests,
1947 * schedule, and hope.
1954 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
1956 struct ll_sb_info *sbi = ll_s2sbi(sb);
1960 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
1961 read_only = *flags & MS_RDONLY;
1962 err = obd_set_info_async(sbi->ll_md_exp,
1963 sizeof(KEY_READ_ONLY) - 1,
1964 KEY_READ_ONLY, sizeof(read_only),
1967 CERROR("Failed to change the read-only flag during "
1968 "remount: %d\n", err);
1973 sb->s_flags |= MS_RDONLY;
1975 sb->s_flags &= ~MS_RDONLY;
1980 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
1981 int offset, struct super_block *sb)
1983 struct ll_sb_info *sbi = NULL;
1984 struct lustre_md md;
1988 LASSERT(*inode || sb);
1989 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
1990 prune_deathrow(sbi, 1);
1991 memset(&md, 0, sizeof(struct lustre_md));
1993 rc = md_get_lustre_md(sbi->ll_md_exp, req, offset,
1994 sbi->ll_dt_exp, sbi->ll_md_exp, &md);
1999 ll_update_inode(*inode, &md);
2001 LASSERT(sb != NULL);
2004 * At this point server returns to client's same fid as client
2005 * generated for creating. So using ->fid1 is okay here.
2007 LASSERT(fid_is_sane(&md.body->fid1));
2009 *inode = ll_iget(sb, ll_fid_build_ino(sbi, &md.body->fid1), &md);
2010 if (*inode == NULL || is_bad_inode(*inode)) {
2012 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2013 #ifdef CONFIG_FS_POSIX_ACL
2015 posix_acl_release(md.posix_acl);
2016 md.posix_acl = NULL;
2020 CERROR("new_inode -fatal: rc %d\n", rc);
2025 rc = obd_checkmd(sbi->ll_dt_exp, sbi->ll_md_exp,
2026 ll_i2info(*inode)->lli_smd);
2028 md_free_lustre_md(sbi->ll_md_exp, &md);
2032 char *llap_origins[] = {
2033 [LLAP_ORIGIN_UNKNOWN] = "--",
2034 [LLAP_ORIGIN_READPAGE] = "rp",
2035 [LLAP_ORIGIN_READAHEAD] = "ra",
2036 [LLAP_ORIGIN_COMMIT_WRITE] = "cw",
2037 [LLAP_ORIGIN_WRITEPAGE] = "wp",
2040 struct ll_async_page *llite_pglist_next_llap(struct ll_sb_info *sbi,
2041 struct list_head *list)
2043 struct ll_async_page *llap;
2044 struct list_head *pos;
2046 list_for_each(pos, list) {
2047 if (pos == &sbi->ll_pglist)
2049 llap = list_entry(pos, struct ll_async_page, llap_pglist_item);
2050 if (llap->llap_page == NULL)
2058 int ll_obd_statfs(struct inode *inode, void *arg)
2060 struct ll_sb_info *sbi = NULL;
2061 struct obd_device *client_obd = NULL, *lov_obd = NULL;
2062 struct lov_obd *lov = NULL;
2063 struct obd_statfs stat_buf = {0};
2065 struct obd_ioctl_data *data = NULL;
2069 if (!inode || !(sbi = ll_i2sbi(inode)))
2070 GOTO(out_statfs, rc = -EINVAL);
2072 rc = obd_ioctl_getdata(&buf, &len, arg);
2074 GOTO(out_statfs, rc);
2077 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2078 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2079 GOTO(out_statfs, rc = -EINVAL);
2081 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2082 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
2084 if (type == LL_STATFS_MDC) {
2086 GOTO(out_statfs, rc = -ENODEV);
2087 client_obd = class_exp2obd(sbi->ll_md_exp);
2088 } else if (type == LL_STATFS_LOV) {
2089 lov_obd = class_exp2obd(sbi->ll_dt_exp);
2090 lov = &lov_obd->u.lov;
2092 if ((index >= lov->desc.ld_tgt_count))
2093 GOTO(out_statfs, rc = -ENODEV);
2094 if (!lov->lov_tgts[index])
2095 /* Try again with the next index */
2096 GOTO(out_statfs, rc = -EAGAIN);
2098 client_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
2099 if (!lov->lov_tgts[index]->ltd_active)
2100 GOTO(out_uuid, rc = -ENODATA);
2104 GOTO(out_statfs, rc = -EINVAL);
2106 rc = obd_statfs(client_obd, &stat_buf, cfs_time_current_64() - 1);
2108 GOTO(out_statfs, rc);
2110 if (copy_to_user(data->ioc_pbuf1, &stat_buf, data->ioc_plen1))
2111 GOTO(out_statfs, rc = -EFAULT);
2114 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(client_obd),
2120 obd_ioctl_freedata(buf, len);
2124 int ll_process_config(struct lustre_cfg *lcfg)
2128 struct lprocfs_static_vars lvars;
2132 lprocfs_init_vars(llite, &lvars);
2134 /* The instance name contains the sb: lustre-client-aacfe000 */
2135 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2136 if (!ptr || !*(++ptr))
2138 if (sscanf(ptr, "%lx", &x) != 1)
2141 /* This better be a real Lustre superblock! */
2142 LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2144 /* Note we have not called client_common_fill_super yet, so
2145 proc fns must be able to handle that! */
2146 rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2151 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2152 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2153 struct inode *i1, struct inode *i2,
2154 const char *name, int namelen,
2155 int mode, __u32 opc, void *data)
2157 LASSERT(i1 != NULL);
2159 if (namelen > ll_i2sbi(i1)->ll_namelen)
2160 return ERR_PTR(-ENAMETOOLONG);
2162 if (op_data == NULL)
2163 OBD_ALLOC_PTR(op_data);
2165 if (op_data == NULL)
2166 return ERR_PTR(-ENOMEM);
2168 ll_i2gids(op_data->op_suppgids, i1, i2);
2169 op_data->op_fid1 = *ll_inode2fid(i1);
2170 op_data->op_capa1 = ll_mdscapa_get(i1);
2173 op_data->op_fid2 = *ll_inode2fid(i2);
2174 op_data->op_capa2 = ll_mdscapa_get(i2);
2176 fid_zero(&op_data->op_fid2);
2179 op_data->op_name = name;
2180 op_data->op_namelen = namelen;
2181 op_data->op_mode = mode;
2182 op_data->op_mod_time = CURRENT_SECONDS;
2183 op_data->op_fsuid = current->fsuid;
2184 op_data->op_fsgid = current->fsgid;
2185 op_data->op_cap = current->cap_effective;
2186 op_data->op_bias = MDS_CHECK_SPLIT;
2187 op_data->op_opc = opc;
2188 op_data->op_mds = 0;
2189 op_data->op_data = data;
2194 void ll_finish_md_op_data(struct md_op_data *op_data)
2196 capa_put(op_data->op_capa1);
2197 capa_put(op_data->op_capa2);
2198 OBD_FREE_PTR(op_data);
2201 int ll_ioctl_getfacl(struct inode *inode, struct rmtacl_ioctl_data *ioc)
2203 struct ll_sb_info *sbi = ll_i2sbi(inode);
2204 struct ptlrpc_request *req = NULL;
2205 struct mdt_body *body;
2207 struct obd_capa *oc;
2211 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT))
2214 LASSERT(ioc->cmd && ioc->cmd_len && ioc->res && ioc->res_len);
2216 OBD_ALLOC(cmd, ioc->cmd_len);
2219 if (copy_from_user(cmd, ioc->cmd, ioc->cmd_len))
2220 GOTO(out, rc = -EFAULT);
2222 oc = ll_mdscapa_get(inode);
2223 rc = md_getxattr(ll_i2sbi(inode)->ll_md_exp, ll_inode2fid(inode), oc,
2224 OBD_MD_FLXATTR, XATTR_NAME_LUSTRE_ACL, cmd,
2225 ioc->cmd_len, ioc->res_len, 0, &req);
2228 CERROR("mdc_getxattr %s [%s] failed: %d\n",
2229 XATTR_NAME_LUSTRE_ACL, cmd, rc);
2233 body = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF, sizeof(*body));
2236 buflen = lustre_msg_buflen(req->rq_repmsg, REPLY_REC_OFF);
2237 LASSERT(buflen <= ioc->res_len);
2238 buf = lustre_msg_string(req->rq_repmsg, REPLY_REC_OFF + 1, ioc->res_len);
2240 if (copy_to_user(ioc->res, buf, buflen))
2241 GOTO(out, rc = -EFAULT);
2245 ptlrpc_req_finished(req);
2246 OBD_FREE(cmd, ioc->cmd_len);
2250 int ll_ioctl_setfacl(struct inode *inode, struct rmtacl_ioctl_data *ioc)
2252 struct ll_sb_info *sbi = ll_i2sbi(inode);
2253 struct ptlrpc_request *req = NULL;
2255 struct obd_capa *oc;
2259 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT))
2262 if (!(sbi->ll_flags & LL_SBI_ACL))
2263 RETURN(-EOPNOTSUPP);
2265 LASSERT(ioc->cmd && ioc->cmd_len && ioc->res && ioc->res_len);
2267 OBD_ALLOC(cmd, ioc->cmd_len);
2270 if (copy_from_user(cmd, ioc->cmd, ioc->cmd_len))
2271 GOTO(out, rc = -EFAULT);
2273 oc = ll_mdscapa_get(inode);
2274 rc = md_setxattr(ll_i2sbi(inode)->ll_md_exp, ll_inode2fid(inode), oc,
2275 OBD_MD_FLXATTR, XATTR_NAME_LUSTRE_ACL, cmd,
2276 ioc->cmd_len, ioc->res_len, 0, &req);
2279 CERROR("mdc_setxattr %s [%s] failed: %d\n",
2280 XATTR_NAME_LUSTRE_ACL, cmd, rc);
2284 buflen = lustre_msg_buflen(req->rq_repmsg, REPLY_REC_OFF);
2285 LASSERT(buflen <= ioc->res_len);
2286 buf = lustre_msg_string(req->rq_repmsg, REPLY_REC_OFF, ioc->res_len);
2288 if (copy_to_user(ioc->res, buf, buflen))
2289 GOTO(out, rc = -EFAULT);
2293 ptlrpc_req_finished(req);
2294 OBD_FREE(cmd, ioc->cmd_len);