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 HAVE_EXPORT___IGET
91 INIT_LIST_HEAD(&sbi->ll_deathrow);
92 spin_lock_init(&sbi->ll_deathrow_lock);
94 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
95 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].pp_r_hist.oh_lock);
96 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].pp_w_hist.oh_lock);
102 void ll_free_sbi(struct super_block *sb)
104 struct ll_sb_info *sbi = ll_s2sbi(sb);
108 spin_lock(&ll_sb_lock);
109 list_del(&sbi->ll_list);
110 spin_unlock(&ll_sb_lock);
111 OBD_FREE(sbi, sizeof(*sbi));
116 static struct dentry_operations ll_d_root_ops = {
117 #ifdef LUSTRE_KERNEL_VERSION
118 .d_compare = ll_dcompare,
122 /* Initialize the default and maximum LOV EA and cookie sizes. This allows
123 * us to make MDS RPCs with large enough reply buffers to hold the
124 * maximum-sized (= maximum striped) EA and cookie without having to
125 * calculate this (via a call into the LOV + OSCs) each time we make an RPC. */
126 static int ll_init_ea_size(struct obd_export *md_exp, struct obd_export *dt_exp)
128 struct lov_stripe_md lsm = { .lsm_magic = LOV_MAGIC };
129 __u32 valsize = sizeof(struct lov_desc);
130 int rc, easize, def_easize, cookiesize;
131 struct lov_desc desc;
135 rc = obd_get_info(dt_exp, strlen(KEY_LOVDESC) + 1, KEY_LOVDESC,
140 stripes = min(desc.ld_tgt_count, (__u32)LOV_MAX_STRIPE_COUNT);
141 lsm.lsm_stripe_count = stripes;
142 easize = obd_size_diskmd(dt_exp, &lsm);
144 lsm.lsm_stripe_count = desc.ld_default_stripe_count;
145 def_easize = obd_size_diskmd(dt_exp, &lsm);
147 cookiesize = stripes * sizeof(struct llog_cookie);
149 CDEBUG(D_HA, "updating max_mdsize/max_cookiesize: %d/%d\n",
152 rc = md_init_ea_size(md_exp, easize, def_easize, cookiesize);
156 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
157 uid_t nllu, gid_t nllg)
159 struct inode *root = 0;
160 struct ll_sb_info *sbi = ll_s2sbi(sb);
161 struct obd_device *obd;
162 struct lu_fid rootfid;
163 struct obd_capa *oc = NULL;
164 struct obd_statfs osfs;
165 struct ptlrpc_request *request = NULL;
166 struct lustre_handle dt_conn = {0, };
167 struct lustre_handle md_conn = {0, };
168 struct obd_connect_data *data = NULL;
169 struct lustre_md lmd;
174 obd = class_name2obd(md);
176 CERROR("MD %s: not setup or attached\n", md);
184 if (proc_lustre_fs_root) {
185 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
188 CERROR("could not register mount in /proc/lustre");
191 /* indicate the features supported by this client */
192 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
194 OBD_CONNECT_ATTRFID | OBD_CONNECT_VERSION |
195 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
196 OBD_CONNECT_CANCELSET;
197 #ifdef CONFIG_FS_POSIX_ACL
198 data->ocd_connect_flags |= OBD_CONNECT_ACL;
200 data->ocd_ibits_known = MDS_INODELOCK_FULL;
201 data->ocd_version = LUSTRE_VERSION_CODE;
203 if (sb->s_flags & MS_RDONLY)
204 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
205 if (sbi->ll_flags & LL_SBI_USER_XATTR)
206 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
208 #ifdef HAVE_MS_FLOCK_LOCK
209 /* force vfs to use lustre handler for flock() calls - bug 10743 */
210 sb->s_flags |= MS_FLOCK_LOCK;
213 if (sbi->ll_flags & LL_SBI_FLOCK)
214 sbi->ll_fop = &ll_file_operations_flock;
215 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
216 sbi->ll_fop = &ll_file_operations;
218 sbi->ll_fop = &ll_file_operations_noflock;
221 data->ocd_connect_flags |= OBD_CONNECT_REAL;
222 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
223 data->ocd_connect_flags &= ~OBD_CONNECT_LCL_CLIENT;
224 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT;
226 data->ocd_connect_flags &= ~OBD_CONNECT_RMT_CLIENT;
227 data->ocd_connect_flags |= OBD_CONNECT_LCL_CLIENT;
229 data->ocd_nllu = nllu;
230 data->ocd_nllg = nllg;
232 err = obd_connect(NULL, &md_conn, obd, &sbi->ll_sb_uuid, data);
234 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
235 "recovery, of which this client is not a "
236 "part. Please wait for recovery to complete,"
237 " abort, or time out.\n", md);
240 CERROR("cannot connect to %s: rc = %d\n", md, err);
243 sbi->ll_md_exp = class_conn2export(&md_conn);
245 err = obd_fid_init(sbi->ll_md_exp);
247 CERROR("Can't init metadata layer FID infrastructure, "
252 err = obd_statfs(obd, &osfs, cfs_time_current_64() - HZ);
254 GOTO(out_md_fid, err);
256 size = sizeof(*data);
257 err = obd_get_info(sbi->ll_md_exp, strlen(KEY_CONN_DATA),
258 KEY_CONN_DATA, &size, data);
260 CERROR("Get connect data failed: %d \n", err);
264 LASSERT(osfs.os_bsize);
265 sb->s_blocksize = osfs.os_bsize;
266 sb->s_blocksize_bits = log2(osfs.os_bsize);
267 sb->s_magic = LL_SUPER_MAGIC;
269 /* for bug 11559. in $LINUX/fs/read_write.c, function do_sendfile():
270 * retval = in_file->f_op->sendfile(...);
272 * retval = -EOVERFLOW;
274 * it will check if *ppos is greater than max. However, max equals to
275 * s_maxbytes, which is a negative integer in a x86_64 box since loff_t
276 * has been defined as a signed long long ineger in linux kernel. */
277 #if BITS_PER_LONG == 64
278 sb->s_maxbytes = PAGE_CACHE_MAXBYTES >> 1;
280 sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
282 sbi->ll_namelen = osfs.os_namelen;
283 sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
285 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
286 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
287 LCONSOLE_INFO("Disabling user_xattr feature because "
288 "it is not supported on the server\n");
289 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
292 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
294 sb->s_flags |= MS_POSIXACL;
296 sbi->ll_flags |= LL_SBI_ACL;
298 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
300 sb->s_flags &= ~MS_POSIXACL;
302 sbi->ll_flags &= ~LL_SBI_ACL;
305 if (data->ocd_connect_flags & OBD_CONNECT_JOIN)
306 sbi->ll_flags |= LL_SBI_JOIN;
308 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
309 if (!(data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT)) {
310 /* sometimes local client claims to be remote, but mdt
311 * will disagree when client gss not applied. */
312 LCONSOLE_INFO("client claims to be remote, but server "
313 "rejected, forced to be local.\n");
314 sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
317 if (!(data->ocd_connect_flags & OBD_CONNECT_LCL_CLIENT)) {
318 /* with gss applied, remote client can not claim to be
319 * local, so mdt maybe force client to be remote. */
320 LCONSOLE_INFO("client claims to be local, but server "
321 "rejected, forced to be remote.\n");
322 sbi->ll_flags |= LL_SBI_RMT_CLIENT;
326 if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
327 LCONSOLE_INFO("client enabled MDS capability!\n");
328 sbi->ll_flags |= LL_SBI_MDS_CAPA;
331 if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
332 LCONSOLE_INFO("client enabled OSS capability!\n");
333 sbi->ll_flags |= LL_SBI_OSS_CAPA;
336 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0))
337 /* We set sb->s_dev equal on all lustre clients in order to support
338 * NFS export clustering. NFSD requires that the FSID be the same
340 /* s_dev is also used in lt_compare() to compare two fs, but that is
341 * only a node-local comparison. */
343 /* XXX: this will not work with LMV */
344 sb->s_dev = get_uuid2int(sbi2mdc(sbi)->cl_target_uuid.uuid,
345 strlen(sbi2mdc(sbi)->cl_target_uuid.uuid));
348 obd = class_name2obd(dt);
350 CERROR("DT %s: not setup or attached\n", dt);
351 GOTO(out_md_fid, err = -ENODEV);
354 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
355 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
356 OBD_CONNECT_CANCELSET;
357 if (sbi->ll_flags & LL_SBI_OSS_CAPA)
358 data->ocd_connect_flags |= OBD_CONNECT_OSS_CAPA;
360 CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
361 "ocd_grant: %d\n", data->ocd_connect_flags,
362 data->ocd_version, data->ocd_grant);
364 obd->obd_upcall.onu_owner = &sbi->ll_lco;
365 obd->obd_upcall.onu_upcall = ll_ocd_update;
366 data->ocd_brw_size = PTLRPC_MAX_BRW_PAGES << CFS_PAGE_SHIFT;
368 err = obd_connect(NULL, &dt_conn, obd, &sbi->ll_sb_uuid, data);
370 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
371 "recovery, of which this client is not a "
372 "part. Please wait for recovery to "
373 "complete, abort, or time out.\n", dt);
374 GOTO(out_md_fid, err);
376 CERROR("Cannot connect to %s: rc = %d\n", dt, err);
377 GOTO(out_md_fid, err);
380 sbi->ll_dt_exp = class_conn2export(&dt_conn);
382 err = obd_fid_init(sbi->ll_dt_exp);
384 CERROR("Can't init data layer FID infrastructure, "
389 spin_lock(&sbi->ll_lco.lco_lock);
390 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
391 spin_unlock(&sbi->ll_lco.lco_lock);
393 ll_init_ea_size(sbi->ll_md_exp, sbi->ll_dt_exp);
395 err = obd_prep_async_page(sbi->ll_dt_exp, NULL, NULL, NULL,
396 0, NULL, NULL, NULL);
398 LCONSOLE_ERROR_MSG(0x151, "There are no OST's in this "
399 "filesystem. There must be at least one "
400 "active OST for a client to start.\n");
401 GOTO(out_dt_fid, err);
404 if (!ll_async_page_slab) {
405 ll_async_page_slab_size =
406 size_round(sizeof(struct ll_async_page)) + err;
407 ll_async_page_slab = cfs_mem_cache_create("ll_async_page",
408 ll_async_page_slab_size,
410 if (!ll_async_page_slab)
411 GOTO(out_dt_fid, err = -ENOMEM);
414 err = md_getstatus(sbi->ll_md_exp, &rootfid, &oc);
416 CERROR("cannot mds_connect: rc = %d\n", err);
417 GOTO(out_dt_fid, err);
419 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&rootfid));
420 sbi->ll_root_fid = rootfid;
422 sb->s_op = &lustre_super_operations;
423 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
424 sb->s_export_op = &lustre_export_operations;
428 * XXX: move this to after cbd setup? */
429 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
430 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
431 valid |= OBD_MD_FLRMTPERM;
432 else if (sbi->ll_flags & LL_SBI_ACL)
433 valid |= OBD_MD_FLACL;
435 err = md_getattr(sbi->ll_md_exp, &rootfid, oc, valid, 0, &request);
439 CERROR("md_getattr failed for root: rc = %d\n", err);
440 GOTO(out_dt_fid, err);
442 memset(&lmd, 0, sizeof(lmd));
443 err = md_get_lustre_md(sbi->ll_md_exp, request,
444 REPLY_REC_OFF, sbi->ll_dt_exp, sbi->ll_md_exp,
447 CERROR("failed to understand root inode md: rc = %d\n", err);
448 ptlrpc_req_finished (request);
449 GOTO(out_dt_fid, err);
452 LASSERT(fid_is_sane(&sbi->ll_root_fid));
453 root = ll_iget(sb, ll_fid_build_ino(sbi, &sbi->ll_root_fid), &lmd);
454 md_free_lustre_md(sbi->ll_md_exp, &lmd);
455 ptlrpc_req_finished(request);
457 if (root == NULL || is_bad_inode(root)) {
459 obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
460 #ifdef CONFIG_FS_POSIX_ACL
462 posix_acl_release(lmd.posix_acl);
463 lmd.posix_acl = NULL;
466 CERROR("lustre_lite: bad iget4 for root\n");
467 GOTO(out_root, err = -EBADF);
470 err = ll_close_thread_start(&sbi->ll_lcq);
472 CERROR("cannot start close thread: rc %d\n", err);
476 /* making vm readahead 0 for 2.4.x. In the case of 2.6.x,
477 backing dev info assigned to inode mapping is used for
478 determining maximal readahead. */
479 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0)) && \
480 !defined(KERNEL_HAS_AS_MAX_READAHEAD)
481 /* bug 2805 - set VM readahead to zero */
482 vm_max_readahead = vm_min_readahead = 0;
485 sb->s_root = d_alloc_root(root);
487 OBD_FREE(data, sizeof(*data));
488 sb->s_root->d_op = &ll_d_root_ops;
494 obd_fid_fini(sbi->ll_dt_exp);
496 obd_disconnect(sbi->ll_dt_exp);
497 sbi->ll_dt_exp = NULL;
499 obd_fid_fini(sbi->ll_md_exp);
501 obd_disconnect(sbi->ll_md_exp);
502 sbi->ll_md_exp = NULL;
506 lprocfs_unregister_mountpoint(sbi);
510 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
514 *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
516 rc = obd_get_info(sbi->ll_md_exp, strlen("max_easize"), "max_easize",
519 CERROR("Get max mdsize error rc %d \n", rc);
524 void ll_dump_inode(struct inode *inode)
526 struct list_head *tmp;
527 int dentry_count = 0;
529 LASSERT(inode != NULL);
531 list_for_each(tmp, &inode->i_dentry)
534 CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
535 inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
536 inode->i_mode, atomic_read(&inode->i_count), dentry_count);
539 void lustre_dump_dentry(struct dentry *dentry, int recur)
541 struct list_head *tmp;
544 LASSERT(dentry != NULL);
546 list_for_each(tmp, &dentry->d_subdirs)
549 CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
550 " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
551 dentry->d_name.len, dentry->d_name.name,
552 dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
553 dentry->d_parent, dentry->d_inode, atomic_read(&dentry->d_count),
554 dentry->d_flags, dentry->d_fsdata, subdirs);
555 if (dentry->d_inode != NULL)
556 ll_dump_inode(dentry->d_inode);
561 list_for_each(tmp, &dentry->d_subdirs) {
562 struct dentry *d = list_entry(tmp, struct dentry, d_child);
563 lustre_dump_dentry(d, recur - 1);
567 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
568 void lustre_throw_orphan_dentries(struct super_block *sb)
570 struct dentry *dentry, *next;
571 struct ll_sb_info *sbi = ll_s2sbi(sb);
573 /* Do this to get rid of orphaned dentries. That is not really trw. */
574 list_for_each_entry_safe(dentry, next, &sbi->ll_orphan_dentry_list,
576 CWARN("found orphan dentry %.*s (%p->%p) at unmount, dumping "
577 "before and after shrink_dcache_parent\n",
578 dentry->d_name.len, dentry->d_name.name, dentry, next);
579 lustre_dump_dentry(dentry, 1);
580 shrink_dcache_parent(dentry);
581 lustre_dump_dentry(dentry, 1);
585 #define lustre_throw_orphan_dentries(sb)
588 #ifdef HAVE_EXPORT___IGET
589 static void prune_dir_dentries(struct inode *inode)
591 struct dentry *dentry, *prev = NULL;
593 /* due to lustre specific logic, a directory
594 * can have few dentries - a bug from VFS POV */
596 spin_lock(&dcache_lock);
597 if (!list_empty(&inode->i_dentry)) {
598 dentry = list_entry(inode->i_dentry.prev,
599 struct dentry, d_alias);
600 /* in order to prevent infinite loops we
601 * break if previous dentry is busy */
602 if (dentry != prev) {
605 spin_unlock(&dcache_lock);
607 /* try to kill all child dentries */
609 shrink_dcache_parent(dentry);
610 unlock_dentry(dentry);
613 /* now try to get rid of current dentry */
614 d_prune_aliases(inode);
618 spin_unlock(&dcache_lock);
621 static void prune_deathrow_one(struct ll_inode_info *lli)
623 struct inode *inode = ll_info2i(lli);
625 /* first, try to drop any dentries - they hold a ref on the inode */
626 if (S_ISDIR(inode->i_mode))
627 prune_dir_dentries(inode);
629 d_prune_aliases(inode);
632 /* if somebody still uses it, leave it */
633 LASSERT(atomic_read(&inode->i_count) > 0);
634 if (atomic_read(&inode->i_count) > 1)
637 CDEBUG(D_INODE, "inode %lu/%u(%d) looks a good candidate for prune\n",
638 inode->i_ino,inode->i_generation, atomic_read(&inode->i_count));
640 /* seems nobody uses it anymore */
648 static void prune_deathrow(struct ll_sb_info *sbi, int try)
650 struct ll_inode_info *lli;
654 if (need_resched() && try)
658 if (!spin_trylock(&sbi->ll_deathrow_lock))
661 spin_lock(&sbi->ll_deathrow_lock);
666 if (!list_empty(&sbi->ll_deathrow)) {
667 lli = list_entry(sbi->ll_deathrow.next,
668 struct ll_inode_info,
670 list_del_init(&lli->lli_dead_list);
671 if (!list_empty(&sbi->ll_deathrow))
674 spin_unlock(&sbi->ll_deathrow_lock);
677 prune_deathrow_one(lli);
679 } while (empty == 0);
681 #else /* !HAVE_EXPORT___IGET */
682 #define prune_deathrow(sbi, try) do {} while (0)
683 #endif /* HAVE_EXPORT___IGET */
685 void client_common_put_super(struct super_block *sb)
687 struct ll_sb_info *sbi = ll_s2sbi(sb);
690 obd_cancel_unused(sbi->ll_dt_exp, NULL, 0, NULL);
692 ll_close_thread_shutdown(sbi->ll_lcq);
694 /* destroy inodes in deathrow */
695 prune_deathrow(sbi, 0);
697 list_del(&sbi->ll_conn_chain);
699 obd_fid_fini(sbi->ll_dt_exp);
700 obd_disconnect(sbi->ll_dt_exp);
701 sbi->ll_dt_exp = NULL;
703 lprocfs_unregister_mountpoint(sbi);
705 obd_fid_fini(sbi->ll_md_exp);
706 obd_disconnect(sbi->ll_md_exp);
707 sbi->ll_md_exp = NULL;
709 lustre_throw_orphan_dentries(sb);
713 char *ll_read_opt(const char *opt, char *data)
719 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
720 if (strncmp(opt, data, strlen(opt)))
722 if ((value = strchr(data, '=')) == NULL)
726 OBD_ALLOC(retval, strlen(value) + 1);
728 CERROR("out of memory!\n");
732 memcpy(retval, value, strlen(value)+1);
733 CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
737 static inline int ll_set_opt(const char *opt, char *data, int fl)
739 if (strncmp(opt, data, strlen(opt)) != 0)
745 /* non-client-specific mount options are parsed in lmd_parse */
746 static int ll_options(char *options, int *flags)
749 char *s1 = options, *s2;
755 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
758 CDEBUG(D_SUPER, "next opt=%s\n", s1);
759 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
764 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
769 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
774 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
779 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
784 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
789 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
791 /* Ignore deprecated mount option. The client will
792 * always try to mount with ACL support, whether this
793 * is used depends on whether server supports it. */
796 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
800 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
806 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
812 s2 = strchr(s1, ',');
820 void ll_lli_init(struct ll_inode_info *lli)
822 lli->lli_inode_magic = LLI_INODE_MAGIC;
823 sema_init(&lli->lli_size_sem, 1);
824 sema_init(&lli->lli_write_sem, 1);
826 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
827 spin_lock_init(&lli->lli_lock);
828 INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
829 INIT_LIST_HEAD(&lli->lli_close_list);
830 lli->lli_inode_magic = LLI_INODE_MAGIC;
831 sema_init(&lli->lli_och_sem, 1);
832 lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
833 lli->lli_mds_exec_och = NULL;
834 lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
835 lli->lli_open_fd_exec_count = 0;
836 INIT_LIST_HEAD(&lli->lli_dead_list);
837 lli->lli_remote_perms = NULL;
838 lli->lli_rmtperm_utime = 0;
839 sema_init(&lli->lli_rmtperm_sem, 1);
840 INIT_LIST_HEAD(&lli->lli_oss_capas);
843 int ll_fill_super(struct super_block *sb)
845 struct lustre_profile *lprof;
846 struct lustre_sb_info *lsi = s2lsi(sb);
847 struct ll_sb_info *sbi;
848 char *dt = NULL, *md = NULL;
849 char *profilenm = get_profile_name(sb);
850 struct config_llog_instance cfg = {0, };
851 char ll_instance[sizeof(sb) * 2 + 1];
855 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
859 sb->s_type->fs_flags |= FS_ODD_RENAME;
860 /* client additional sb info */
861 lsi->lsi_llsbi = sbi = ll_init_sbi();
867 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
871 /* Generate a string unique to this super, in case some joker tries
872 to mount the same fs at two mount points.
873 Use the address of the super itself.*/
874 sprintf(ll_instance, "%p", sb);
875 cfg.cfg_instance = ll_instance;
876 cfg.cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
878 /* set up client obds */
879 err = lustre_process_log(sb, profilenm, &cfg);
881 CERROR("Unable to process log: %d\n", err);
885 lprof = class_get_profile(profilenm);
887 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
888 " read from the MGS. Does that filesystem "
889 "exist?\n", profilenm);
890 GOTO(out_free, err = -EINVAL);
892 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
893 lprof->lp_md, lprof->lp_dt);
895 OBD_ALLOC(dt, strlen(lprof->lp_dt) +
896 strlen(ll_instance) + 2);
898 GOTO(out_free, err = -ENOMEM);
899 sprintf(dt, "%s-%s", lprof->lp_dt, ll_instance);
901 OBD_ALLOC(md, strlen(lprof->lp_md) +
902 strlen(ll_instance) + 2);
904 GOTO(out_free, err = -ENOMEM);
905 sprintf(md, "%s-%s", lprof->lp_md, ll_instance);
907 /* connections, registrations, sb setup */
908 err = client_common_fill_super(sb, md, dt,
909 lsi->lsi_lmd->lmd_nllu,
910 lsi->lsi_lmd->lmd_nllg);
914 OBD_FREE(md, strlen(md) + 1);
916 OBD_FREE(dt, strlen(dt) + 1);
920 LCONSOLE_WARN("Client %s has started\n", profilenm);
923 } /* ll_fill_super */
926 void ll_put_super(struct super_block *sb)
928 struct config_llog_instance cfg;
929 char ll_instance[sizeof(sb) * 2 + 1];
930 struct obd_device *obd;
931 struct lustre_sb_info *lsi = s2lsi(sb);
932 struct ll_sb_info *sbi = ll_s2sbi(sb);
933 char *profilenm = get_profile_name(sb);
937 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
939 ll_print_capa_stat(sbi);
941 sprintf(ll_instance, "%p", sb);
942 cfg.cfg_instance = ll_instance;
943 lustre_end_log(sb, NULL, &cfg);
945 if (sbi->ll_md_exp) {
946 obd = class_exp2obd(sbi->ll_md_exp);
948 force = obd->obd_force;
951 /* We need to set force before the lov_disconnect in
952 lustre_common_put_super, since l_d cleans up osc's as well. */
955 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
957 obd->obd_force = force;
962 /* Only if client_common_fill_super succeeded */
963 client_common_put_super(sb);
966 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
967 class_manual_cleanup(obd);
971 class_del_profile(profilenm);
974 lsi->lsi_llsbi = NULL;
976 lustre_common_put_super(sb);
978 LCONSOLE_WARN("client %s umount complete\n", ll_instance);
983 } /* client_put_super */
985 #ifdef HAVE_REGISTER_CACHE
986 #include <linux/cache_def.h>
987 #ifdef HAVE_CACHE_RETURN_INT
992 ll_shrink_cache(int priority, unsigned int gfp_mask)
994 struct ll_sb_info *sbi;
997 list_for_each_entry(sbi, &ll_super_blocks, ll_list)
998 count += llap_shrink_cache(sbi, priority);
1000 #ifdef HAVE_CACHE_RETURN_INT
1005 struct cache_definition ll_cache_definition = {
1006 .name = "llap_cache",
1007 .shrink = ll_shrink_cache
1009 #endif /* HAVE_REGISTER_CACHE */
1011 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1013 struct inode *inode = NULL;
1014 /* NOTE: we depend on atomic igrab() -bzzz */
1015 lock_res_and_lock(lock);
1016 if (lock->l_ast_data) {
1017 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1018 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1019 inode = igrab(lock->l_ast_data);
1021 inode = lock->l_ast_data;
1022 ldlm_lock_debug(NULL, inode->i_state & I_FREEING ?
1024 lock, __FILE__, __func__, __LINE__,
1025 "l_ast_data %p is bogus: magic %08x",
1026 lock->l_ast_data, lli->lli_inode_magic);
1030 unlock_res_and_lock(lock);
1034 static int null_if_equal(struct ldlm_lock *lock, void *data)
1036 if (data == lock->l_ast_data) {
1037 lock->l_ast_data = NULL;
1039 if (lock->l_req_mode != lock->l_granted_mode)
1040 LDLM_ERROR(lock,"clearing inode with ungranted lock");
1043 return LDLM_ITER_CONTINUE;
1046 void ll_clear_inode(struct inode *inode)
1048 struct ll_inode_info *lli = ll_i2info(inode);
1049 struct ll_sb_info *sbi = ll_i2sbi(inode);
1052 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1053 inode->i_generation, inode);
1055 ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1056 md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1057 null_if_equal, inode);
1059 LASSERT(!lli->lli_open_fd_write_count);
1060 LASSERT(!lli->lli_open_fd_read_count);
1061 LASSERT(!lli->lli_open_fd_exec_count);
1063 if (lli->lli_mds_write_och)
1064 ll_md_real_close(inode, FMODE_WRITE);
1065 if (lli->lli_mds_exec_och)
1066 ll_md_real_close(inode, FMODE_EXEC);
1067 if (lli->lli_mds_read_och)
1068 ll_md_real_close(inode, FMODE_READ);
1071 obd_change_cbdata(sbi->ll_dt_exp, lli->lli_smd,
1072 null_if_equal, inode);
1074 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1075 lli->lli_smd = NULL;
1078 if (lli->lli_symlink_name) {
1079 OBD_FREE(lli->lli_symlink_name,
1080 strlen(lli->lli_symlink_name) + 1);
1081 lli->lli_symlink_name = NULL;
1084 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1085 LASSERT(lli->lli_posix_acl == NULL);
1086 if (lli->lli_remote_perms) {
1087 free_rmtperm_hash(lli->lli_remote_perms);
1088 lli->lli_remote_perms = NULL;
1091 #ifdef CONFIG_FS_POSIX_ACL
1092 else if (lli->lli_posix_acl) {
1093 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1094 LASSERT(lli->lli_remote_perms == NULL);
1095 posix_acl_release(lli->lli_posix_acl);
1096 lli->lli_posix_acl = NULL;
1099 lli->lli_inode_magic = LLI_INODE_DEAD;
1101 #ifdef HAVE_EXPORT___IGET
1102 spin_lock(&sbi->ll_deathrow_lock);
1103 list_del_init(&lli->lli_dead_list);
1104 spin_unlock(&sbi->ll_deathrow_lock);
1106 ll_clear_inode_capas(inode);
1111 int ll_md_setattr(struct inode *inode, struct md_op_data *op_data)
1113 struct lustre_md md;
1114 struct ll_sb_info *sbi = ll_i2sbi(inode);
1115 struct ptlrpc_request *request = NULL;
1119 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1120 LUSTRE_OPC_ANY, NULL);
1121 if (IS_ERR(op_data))
1122 RETURN(PTR_ERR(op_data));
1124 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1127 ptlrpc_req_finished(request);
1128 if (rc == -ENOENT) {
1130 /* Unlinked special device node? Or just a race?
1131 * Pretend we done everything. */
1132 if (!S_ISREG(inode->i_mode) &&
1133 !S_ISDIR(inode->i_mode))
1134 rc = inode_setattr(inode, &op_data->op_attr);
1135 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1136 CERROR("md_setattr fails: rc = %d\n", rc);
1141 rc = md_get_lustre_md(sbi->ll_md_exp, request, REPLY_REC_OFF,
1142 sbi->ll_dt_exp, sbi->ll_md_exp, &md);
1144 ptlrpc_req_finished(request);
1148 /* We call inode_setattr to adjust timestamps.
1149 * If there is at least some data in file, we cleared ATTR_SIZE
1150 * above to avoid invoking vmtruncate, otherwise it is important
1151 * to call vmtruncate in inode_setattr to update inode->i_size
1153 rc = inode_setattr(inode, &op_data->op_attr);
1155 /* Extract epoch data if obtained. */
1156 memcpy(&op_data->op_handle, &md.body->handle, sizeof(op_data->op_handle));
1157 op_data->op_ioepoch = md.body->ioepoch;
1159 ll_update_inode(inode, &md);
1160 ptlrpc_req_finished(request);
1165 /* Close IO epoch and send Size-on-MDS attribute update. */
1166 static int ll_setattr_done_writing(struct inode *inode,
1167 struct md_op_data *op_data)
1169 struct ll_inode_info *lli = ll_i2info(inode);
1173 LASSERT(op_data != NULL);
1174 if (!S_ISREG(inode->i_mode))
1177 CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1178 op_data->op_ioepoch, PFID(&lli->lli_fid));
1180 op_data->op_flags = MF_EPOCH_CLOSE | MF_SOM_CHANGE;
1181 /* XXX: pass och here for the recovery purpose. */
1182 rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, NULL);
1183 if (rc == -EAGAIN) {
1184 /* MDS has instructed us to obtain Size-on-MDS attribute
1185 * from OSTs and send setattr to back to MDS. */
1186 rc = ll_sizeonmds_update(inode, &op_data->op_handle,
1187 op_data->op_ioepoch);
1189 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1195 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1196 * object(s) determine the file size and mtime. Otherwise, the MDS will
1197 * keep these values until such a time that objects are allocated for it.
1198 * We do the MDS operations first, as it is checking permissions for us.
1199 * We don't to the MDS RPC if there is nothing that we want to store there,
1200 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1201 * going to do an RPC anyways.
1203 * If we are doing a truncate, we will send the mtime and ctime updates
1204 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1205 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1208 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1210 struct ll_inode_info *lli = ll_i2info(inode);
1211 struct lov_stripe_md *lsm = lli->lli_smd;
1212 struct ll_sb_info *sbi = ll_i2sbi(inode);
1213 struct md_op_data *op_data = NULL;
1214 int ia_valid = attr->ia_valid;
1215 int rc = 0, rc1 = 0;
1218 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1220 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_SETATTR, 1);
1222 if (ia_valid & ATTR_SIZE) {
1223 if (attr->ia_size > ll_file_maxbytes(inode)) {
1224 CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1225 attr->ia_size, ll_file_maxbytes(inode));
1229 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1232 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1233 if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1234 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1238 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1239 if (attr->ia_valid & ATTR_CTIME) {
1240 attr->ia_ctime = CURRENT_TIME;
1241 attr->ia_valid |= ATTR_CTIME_SET;
1243 if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1244 attr->ia_atime = CURRENT_TIME;
1245 attr->ia_valid |= ATTR_ATIME_SET;
1247 if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1248 attr->ia_mtime = CURRENT_TIME;
1249 attr->ia_valid |= ATTR_MTIME_SET;
1251 if ((attr->ia_valid & ATTR_CTIME) && !(attr->ia_valid & ATTR_MTIME)) {
1252 /* To avoid stale mtime on mds, obtain it from ost and send
1254 rc = ll_glimpse_size(inode, 0);
1258 attr->ia_valid |= ATTR_MTIME_SET | ATTR_MTIME;
1259 attr->ia_mtime = inode->i_mtime;
1262 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1263 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1264 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1267 /* NB: ATTR_SIZE will only be set after this point if the size
1268 * resides on the MDS, ie, this file has no objects. */
1270 attr->ia_valid &= ~ATTR_SIZE;
1272 /* We always do an MDS RPC, even if we're only changing the size;
1273 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1275 OBD_ALLOC_PTR(op_data);
1276 if (op_data == NULL)
1279 memcpy(&op_data->op_attr, attr, sizeof(*attr));
1281 /* Open epoch for truncate. */
1282 if (ia_valid & ATTR_SIZE)
1283 op_data->op_flags = MF_EPOCH_OPEN;
1285 rc = ll_md_setattr(inode, op_data);
1289 if (op_data->op_ioepoch)
1290 CDEBUG(D_INODE, "Epoch "LPU64" opened on "DFID" for "
1291 "truncate\n", op_data->op_ioepoch, PFID(&lli->lli_fid));
1293 if (!lsm || !S_ISREG(inode->i_mode)) {
1294 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1298 /* We really need to get our PW lock before we change inode->i_size.
1299 * If we don't we can race with other i_size updaters on our node, like
1300 * ll_file_read. We can also race with i_size propogation to other
1301 * nodes through dirtying and writeback of final cached pages. This
1302 * last one is especially bad for racing o_append users on other
1304 if (ia_valid & ATTR_SIZE) {
1305 ldlm_policy_data_t policy = { .l_extent = {attr->ia_size,
1307 struct lustre_handle lockh = { 0 };
1308 int err, ast_flags = 0;
1309 /* XXX when we fix the AST intents to pass the discard-range
1310 * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
1312 if (attr->ia_size == 0)
1313 ast_flags = LDLM_AST_DISCARD_DATA;
1315 UNLOCK_INODE_MUTEX(inode);
1316 UP_WRITE_I_ALLOC_SEM(inode);
1317 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy, &lockh,
1319 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1320 DOWN_WRITE_I_ALLOC_SEM(inode);
1321 LOCK_INODE_MUTEX(inode);
1323 LOCK_INODE_MUTEX(inode);
1324 DOWN_WRITE_I_ALLOC_SEM(inode);
1329 /* Only ll_inode_size_lock is taken at this level.
1330 * lov_stripe_lock() is grabbed by ll_truncate() only over
1331 * call to obd_adjust_kms(). If vmtruncate returns 0, then
1332 * ll_truncate dropped ll_inode_size_lock() */
1333 ll_inode_size_lock(inode, 0);
1334 rc = vmtruncate(inode, attr->ia_size);
1336 LASSERT(atomic_read(&lli->lli_size_sem.count) <= 0);
1337 ll_inode_size_unlock(inode, 0);
1340 err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
1342 CERROR("ll_extent_unlock failed: %d\n", err);
1346 } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
1348 struct obd_info oinfo = { { { 0 } } };
1351 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1352 inode->i_ino, LTIME_S(attr->ia_mtime));
1356 oa->o_id = lsm->lsm_object_id;
1357 oa->o_gr = lsm->lsm_object_gr;
1358 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1360 flags = OBD_MD_FLTYPE | OBD_MD_FLATIME |
1361 OBD_MD_FLMTIME | OBD_MD_FLCTIME |
1362 OBD_MD_FLFID | OBD_MD_FLGENER |
1365 obdo_from_inode(oa, inode, flags);
1369 oinfo.oi_capa = ll_mdscapa_get(inode);
1371 /* XXX: this looks unnecessary now. */
1372 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1373 capa_put(oinfo.oi_capa);
1375 CERROR("obd_setattr_async fails: rc=%d\n", rc);
1384 if (op_data->op_ioepoch) {
1385 rc1 = ll_setattr_done_writing(inode, op_data);
1387 ll_finish_md_op_data(op_data);
1389 return rc ? rc : rc1;
1392 int ll_setattr(struct dentry *de, struct iattr *attr)
1394 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1395 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1396 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1398 return ll_setattr_raw(de->d_inode, attr);
1401 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1404 struct ll_sb_info *sbi = ll_s2sbi(sb);
1405 struct obd_statfs obd_osfs;
1409 rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age);
1411 CERROR("md_statfs fails: rc = %d\n", rc);
1415 osfs->os_type = sb->s_magic;
1417 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1418 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1420 rc = obd_statfs_rqset(class_exp2obd(sbi->ll_dt_exp),
1421 &obd_osfs, max_age);
1423 CERROR("obd_statfs fails: rc = %d\n", rc);
1427 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1428 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1431 osfs->os_bsize = obd_osfs.os_bsize;
1432 osfs->os_blocks = obd_osfs.os_blocks;
1433 osfs->os_bfree = obd_osfs.os_bfree;
1434 osfs->os_bavail = obd_osfs.os_bavail;
1436 /* If we don't have as many objects free on the OST as inodes
1437 * on the MDS, we reduce the total number of inodes to
1438 * compensate, so that the "inodes in use" number is correct.
1440 if (obd_osfs.os_ffree < osfs->os_ffree) {
1441 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1443 osfs->os_ffree = obd_osfs.os_ffree;
1448 #ifndef HAVE_STATFS_DENTRY_PARAM
1449 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1452 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1454 struct super_block *sb = de->d_sb;
1456 struct obd_statfs osfs;
1459 CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1460 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1462 /* For now we will always get up-to-date statfs values, but in the
1463 * future we may allow some amount of caching on the client (e.g.
1464 * from QOS or lprocfs updates). */
1465 rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - 1);
1469 statfs_unpack(sfs, &osfs);
1471 /* We need to downshift for all 32-bit kernels, because we can't
1472 * tell if the kernel is being called via sys_statfs64() or not.
1473 * Stop before overflowing f_bsize - in which case it is better
1474 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1475 if (sizeof(long) < 8) {
1476 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1479 osfs.os_blocks >>= 1;
1480 osfs.os_bfree >>= 1;
1481 osfs.os_bavail >>= 1;
1485 sfs->f_blocks = osfs.os_blocks;
1486 sfs->f_bfree = osfs.os_bfree;
1487 sfs->f_bavail = osfs.os_bavail;
1492 void ll_inode_size_lock(struct inode *inode, int lock_lsm)
1494 struct ll_inode_info *lli;
1495 struct lov_stripe_md *lsm;
1497 lli = ll_i2info(inode);
1498 LASSERT(lli->lli_size_sem_owner != current);
1499 down(&lli->lli_size_sem);
1500 LASSERT(lli->lli_size_sem_owner == NULL);
1501 lli->lli_size_sem_owner = current;
1503 LASSERTF(lsm != NULL || lock_lsm == 0, "lsm %p, lock_lsm %d\n",
1506 lov_stripe_lock(lsm);
1509 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm)
1511 struct ll_inode_info *lli;
1512 struct lov_stripe_md *lsm;
1514 lli = ll_i2info(inode);
1516 LASSERTF(lsm != NULL || unlock_lsm == 0, "lsm %p, lock_lsm %d\n",
1519 lov_stripe_unlock(lsm);
1520 LASSERT(lli->lli_size_sem_owner == current);
1521 lli->lli_size_sem_owner = NULL;
1522 up(&lli->lli_size_sem);
1525 static void ll_replace_lsm(struct inode *inode, struct lov_stripe_md *lsm)
1527 struct ll_inode_info *lli = ll_i2info(inode);
1529 dump_lsm(D_INODE, lsm);
1530 dump_lsm(D_INODE, lli->lli_smd);
1531 LASSERTF(lsm->lsm_magic == LOV_MAGIC_JOIN,
1532 "lsm must be joined lsm %p\n", lsm);
1533 obd_free_memmd(ll_i2dtexp(inode), &lli->lli_smd);
1534 CDEBUG(D_INODE, "replace lsm %p to lli_smd %p for inode %lu%u(%p)\n",
1535 lsm, lli->lli_smd, inode->i_ino, inode->i_generation, inode);
1537 lli->lli_maxbytes = lsm->lsm_maxbytes;
1538 if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1539 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1542 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1544 struct ll_inode_info *lli = ll_i2info(inode);
1545 struct mdt_body *body = md->body;
1546 struct lov_stripe_md *lsm = md->lsm;
1547 struct ll_sb_info *sbi = ll_i2sbi(inode);
1549 LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1551 if (lli->lli_smd == NULL) {
1552 if (lsm->lsm_magic != LOV_MAGIC &&
1553 lsm->lsm_magic != LOV_MAGIC_JOIN) {
1554 dump_lsm(D_ERROR, lsm);
1557 CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1558 lsm, inode->i_ino, inode->i_generation, inode);
1559 /* ll_inode_size_lock() requires it is only called
1560 * with lli_smd != NULL or lock_lsm == 0 or we can
1561 * race between lock/unlock. bug 9547 */
1563 lli->lli_maxbytes = lsm->lsm_maxbytes;
1564 if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1565 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1567 if (lli->lli_smd->lsm_magic == lsm->lsm_magic &&
1568 lli->lli_smd->lsm_stripe_count ==
1569 lsm->lsm_stripe_count) {
1570 if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
1571 CERROR("lsm mismatch for inode %ld\n",
1573 CERROR("lli_smd:\n");
1574 dump_lsm(D_ERROR, lli->lli_smd);
1576 dump_lsm(D_ERROR, lsm);
1580 ll_replace_lsm(inode, lsm);
1582 if (lli->lli_smd != lsm)
1583 obd_free_memmd(ll_i2dtexp(inode), &lsm);
1586 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1587 if (body->valid & OBD_MD_FLRMTPERM)
1588 ll_update_remote_perm(inode, md->remote_perm);
1590 #ifdef CONFIG_FS_POSIX_ACL
1591 else if (body->valid & OBD_MD_FLACL) {
1592 spin_lock(&lli->lli_lock);
1593 if (lli->lli_posix_acl)
1594 posix_acl_release(lli->lli_posix_acl);
1595 lli->lli_posix_acl = md->posix_acl;
1596 spin_unlock(&lli->lli_lock);
1599 if (body->valid & OBD_MD_FLATIME &&
1600 body->atime > LTIME_S(inode->i_atime))
1601 LTIME_S(inode->i_atime) = body->atime;
1603 /* mtime is always updated with ctime, but can be set in past.
1604 As write and utime(2) may happen within 1 second, and utime's
1605 mtime has a priority over write's one, so take mtime from mds
1606 for the same ctimes. */
1607 if (body->valid & OBD_MD_FLCTIME &&
1608 body->ctime >= LTIME_S(inode->i_ctime)) {
1609 LTIME_S(inode->i_ctime) = body->ctime;
1610 if (body->valid & OBD_MD_FLMTIME) {
1611 CDEBUG(D_INODE, "setting ino %lu mtime "
1612 "from %lu to "LPU64"\n", inode->i_ino,
1613 LTIME_S(inode->i_mtime), body->mtime);
1614 LTIME_S(inode->i_mtime) = body->mtime;
1617 if (body->valid & OBD_MD_FLMODE)
1618 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1619 if (body->valid & OBD_MD_FLTYPE)
1620 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1621 if (S_ISREG(inode->i_mode)) {
1622 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1624 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1626 #ifdef HAVE_INODE_BLKSIZE
1627 inode->i_blksize = 1<<inode->i_blkbits;
1629 if (body->valid & OBD_MD_FLUID)
1630 inode->i_uid = body->uid;
1631 if (body->valid & OBD_MD_FLGID)
1632 inode->i_gid = body->gid;
1633 if (body->valid & OBD_MD_FLFLAGS)
1634 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1635 if (body->valid & OBD_MD_FLNLINK)
1636 inode->i_nlink = body->nlink;
1637 if (body->valid & OBD_MD_FLRDEV)
1638 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1639 inode->i_rdev = body->rdev;
1641 inode->i_rdev = old_decode_dev(body->rdev);
1643 if (body->valid & OBD_MD_FLSIZE) {
1644 if (ll_i2mdexp(inode)->exp_connect_flags & OBD_CONNECT_SOM) {
1645 if (lli->lli_flags & (LLIF_DONE_WRITING |
1646 LLIF_EPOCH_PENDING |
1648 CWARN("ino %lu flags %lu still has size authority!"
1649 "do not trust the size got from MDS\n",
1650 inode->i_ino, lli->lli_flags);
1652 inode->i_size = body->size;
1653 lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1656 inode->i_size = body->size;
1659 if (body->valid & OBD_MD_FLBLOCKS)
1660 inode->i_blocks = body->blocks;
1663 if (body->valid & OBD_MD_FLID) {
1664 /* FID shouldn't be changed! */
1665 if (fid_is_sane(&lli->lli_fid)) {
1666 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1667 "Trying to change FID "DFID
1668 " to the "DFID", inode %lu/%u(%p)\n",
1669 PFID(&lli->lli_fid), PFID(&body->fid1),
1670 inode->i_ino, inode->i_generation, inode);
1672 lli->lli_fid = body->fid1;
1675 LASSERT(fid_seq(&lli->lli_fid) != 0);
1677 if (body->valid & OBD_MD_FLMDSCAPA) {
1678 LASSERT(md->mds_capa);
1679 ll_add_capa(inode, md->mds_capa);
1681 if (body->valid & OBD_MD_FLOSSCAPA) {
1682 LASSERT(md->oss_capa);
1683 ll_add_capa(inode, md->oss_capa);
1687 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
1688 static struct backing_dev_info ll_backing_dev_info = {
1689 .ra_pages = 0, /* No readahead */
1690 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,12))
1691 .capabilities = 0, /* Does contribute to dirty memory */
1693 .memory_backed = 0, /* Does contribute to dirty memory */
1698 void ll_read_inode2(struct inode *inode, void *opaque)
1700 struct lustre_md *md = opaque;
1701 struct ll_inode_info *lli = ll_i2info(inode);
1704 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n",
1705 inode->i_ino, inode->i_generation, inode);
1709 LASSERT(!lli->lli_smd);
1711 /* Core attributes from the MDS first. This is a new inode, and
1712 * the VFS doesn't zero times in the core inode so we have to do
1713 * it ourselves. They will be overwritten by either MDS or OST
1714 * attributes - we just need to make sure they aren't newer. */
1715 LTIME_S(inode->i_mtime) = 0;
1716 LTIME_S(inode->i_atime) = 0;
1717 LTIME_S(inode->i_ctime) = 0;
1719 ll_update_inode(inode, md);
1721 /* OIDEBUG(inode); */
1723 if (S_ISREG(inode->i_mode)) {
1724 struct ll_sb_info *sbi = ll_i2sbi(inode);
1725 inode->i_op = &ll_file_inode_operations;
1726 inode->i_fop = sbi->ll_fop;
1727 inode->i_mapping->a_ops = &ll_aops;
1729 } else if (S_ISDIR(inode->i_mode)) {
1730 inode->i_op = &ll_dir_inode_operations;
1731 inode->i_fop = &ll_dir_operations;
1732 inode->i_mapping->a_ops = &ll_dir_aops;
1734 } else if (S_ISLNK(inode->i_mode)) {
1735 inode->i_op = &ll_fast_symlink_inode_operations;
1738 inode->i_op = &ll_special_inode_operations;
1740 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
1741 init_special_inode(inode, inode->i_mode,
1742 kdev_t_to_nr(inode->i_rdev));
1744 /* initializing backing dev info. */
1745 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
1747 init_special_inode(inode, inode->i_mode, inode->i_rdev);
1753 void ll_delete_inode(struct inode *inode)
1755 struct ll_sb_info *sbi = ll_i2sbi(inode);
1759 rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
1761 CERROR("fid_delete() failed, rc %d\n", rc);
1768 int ll_iocontrol(struct inode *inode, struct file *file,
1769 unsigned int cmd, unsigned long arg)
1771 struct ll_sb_info *sbi = ll_i2sbi(inode);
1772 struct ptlrpc_request *req = NULL;
1777 case EXT3_IOC_GETFLAGS: {
1778 struct mdt_body *body;
1779 struct obd_capa *oc;
1781 oc = ll_mdscapa_get(inode);
1782 rc = md_getattr(sbi->ll_md_exp, ll_inode2fid(inode), oc,
1783 OBD_MD_FLFLAGS, 0, &req);
1786 CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1790 body = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
1793 /*Now the ext3 will be packed directly back to client,
1794 *no need convert here*/
1795 flags = body->flags;
1797 ptlrpc_req_finished (req);
1799 RETURN(put_user(flags, (int *)arg));
1801 case EXT3_IOC_SETFLAGS: {
1802 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1803 struct obd_info oinfo = { { { 0 } } };
1804 struct md_op_data *op_data;
1806 if (get_user(flags, (int *)arg))
1810 OBDO_ALLOC(oinfo.oi_oa);
1814 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1815 LUSTRE_OPC_ANY, NULL);
1816 if (IS_ERR(op_data))
1817 RETURN(PTR_ERR(op_data));
1819 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1820 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1821 rc = md_setattr(sbi->ll_md_exp, op_data,
1822 NULL, 0, NULL, 0, &req);
1823 ll_finish_md_op_data(op_data);
1824 ptlrpc_req_finished(req);
1825 if (rc || lsm == NULL) {
1826 OBDO_FREE(oinfo.oi_oa);
1830 oinfo.oi_oa->o_id = lsm->lsm_object_id;
1831 oinfo.oi_oa->o_gr = lsm->lsm_object_gr;
1832 oinfo.oi_oa->o_flags = flags;
1833 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1835 oinfo.oi_capa = ll_mdscapa_get(inode);
1837 obdo_from_inode(oinfo.oi_oa, inode,
1838 OBD_MD_FLFID | OBD_MD_FLGENER);
1839 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1840 capa_put(oinfo.oi_capa);
1841 OBDO_FREE(oinfo.oi_oa);
1843 if (rc != -EPERM && rc != -EACCES)
1844 CERROR("md_setattr_async fails: rc = %d\n", rc);
1848 inode->i_flags = ll_ext_to_inode_flags(flags |
1849 MDS_BFLAG_EXT_FLAGS);
1859 int ll_flush_ctx(struct inode *inode)
1861 struct ll_sb_info *sbi = ll_i2sbi(inode);
1863 CDEBUG(D_SEC, "flush context for user %d\n", current->uid);
1865 obd_set_info_async(sbi->ll_md_exp,
1866 sizeof(KEY_FLUSH_CTX) - 1, KEY_FLUSH_CTX,
1868 obd_set_info_async(sbi->ll_dt_exp,
1869 sizeof(KEY_FLUSH_CTX) - 1, KEY_FLUSH_CTX,
1874 /* umount -f client means force down, don't save state */
1875 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1876 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
1878 struct super_block *sb = vfsmnt->mnt_sb;
1880 void ll_umount_begin(struct super_block *sb)
1883 struct lustre_sb_info *lsi = s2lsi(sb);
1884 struct ll_sb_info *sbi = ll_s2sbi(sb);
1885 struct obd_device *obd;
1886 struct obd_ioctl_data ioc_data = { 0 };
1889 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1890 if (!(flags & MNT_FORCE)) {
1896 /* Tell the MGC we got umount -f */
1897 lsi->lsi_flags |= LSI_UMOUNT_FORCE;
1899 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1900 sb->s_count, atomic_read(&sb->s_active));
1902 obd = class_exp2obd(sbi->ll_md_exp);
1904 CERROR("Invalid MDC connection handle "LPX64"\n",
1905 sbi->ll_md_exp->exp_handle.h_cookie);
1910 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp, sizeof ioc_data,
1913 obd = class_exp2obd(sbi->ll_dt_exp);
1915 CERROR("Invalid LOV connection handle "LPX64"\n",
1916 sbi->ll_dt_exp->exp_handle.h_cookie);
1922 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp, sizeof ioc_data,
1925 /* Really, we'd like to wait until there are no requests outstanding,
1926 * and then continue. For now, we just invalidate the requests,
1927 * schedule, and hope.
1934 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
1936 struct ll_sb_info *sbi = ll_s2sbi(sb);
1940 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
1941 read_only = *flags & MS_RDONLY;
1942 err = obd_set_info_async(sbi->ll_md_exp,
1943 sizeof(KEY_READ_ONLY) - 1,
1944 KEY_READ_ONLY, sizeof(read_only),
1947 CERROR("Failed to change the read-only flag during "
1948 "remount: %d\n", err);
1953 sb->s_flags |= MS_RDONLY;
1955 sb->s_flags &= ~MS_RDONLY;
1960 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
1961 int offset, struct super_block *sb)
1963 struct ll_sb_info *sbi = NULL;
1964 struct lustre_md md;
1968 LASSERT(*inode || sb);
1969 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
1970 prune_deathrow(sbi, 1);
1971 memset(&md, 0, sizeof(struct lustre_md));
1973 rc = md_get_lustre_md(sbi->ll_md_exp, req, offset,
1974 sbi->ll_dt_exp, sbi->ll_md_exp, &md);
1979 ll_update_inode(*inode, &md);
1981 LASSERT(sb != NULL);
1984 * At this point server returns to client's same fid as client
1985 * generated for creating. So using ->fid1 is okay here.
1987 LASSERT(fid_is_sane(&md.body->fid1));
1989 *inode = ll_iget(sb, ll_fid_build_ino(sbi, &md.body->fid1), &md);
1990 if (*inode == NULL || is_bad_inode(*inode)) {
1992 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
1993 #ifdef CONFIG_FS_POSIX_ACL
1995 posix_acl_release(md.posix_acl);
1996 md.posix_acl = NULL;
2000 CERROR("new_inode -fatal: rc %d\n", rc);
2005 rc = obd_checkmd(sbi->ll_dt_exp, sbi->ll_md_exp,
2006 ll_i2info(*inode)->lli_smd);
2008 md_free_lustre_md(sbi->ll_md_exp, &md);
2012 char *llap_origins[] = {
2013 [LLAP_ORIGIN_UNKNOWN] = "--",
2014 [LLAP_ORIGIN_READPAGE] = "rp",
2015 [LLAP_ORIGIN_READAHEAD] = "ra",
2016 [LLAP_ORIGIN_COMMIT_WRITE] = "cw",
2017 [LLAP_ORIGIN_WRITEPAGE] = "wp",
2020 struct ll_async_page *llite_pglist_next_llap(struct ll_sb_info *sbi,
2021 struct list_head *list)
2023 struct ll_async_page *llap;
2024 struct list_head *pos;
2026 list_for_each(pos, list) {
2027 if (pos == &sbi->ll_pglist)
2029 llap = list_entry(pos, struct ll_async_page, llap_pglist_item);
2030 if (llap->llap_page == NULL)
2038 int ll_obd_statfs(struct inode *inode, void *arg)
2040 struct ll_sb_info *sbi = NULL;
2041 struct obd_device *client_obd = NULL, *lov_obd = NULL;
2042 struct lov_obd *lov = NULL;
2043 struct obd_statfs stat_buf = {0};
2045 struct obd_ioctl_data *data = NULL;
2049 if (!inode || !(sbi = ll_i2sbi(inode)))
2050 GOTO(out_statfs, rc = -EINVAL);
2052 rc = obd_ioctl_getdata(&buf, &len, arg);
2054 GOTO(out_statfs, rc);
2057 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2058 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2059 GOTO(out_statfs, rc = -EINVAL);
2061 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2062 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
2064 if (type == LL_STATFS_MDC) {
2066 GOTO(out_statfs, rc = -ENODEV);
2067 client_obd = class_exp2obd(sbi->ll_md_exp);
2068 } else if (type == LL_STATFS_LOV) {
2069 lov_obd = class_exp2obd(sbi->ll_dt_exp);
2070 lov = &lov_obd->u.lov;
2072 if ((index >= lov->desc.ld_tgt_count))
2073 GOTO(out_statfs, rc = -ENODEV);
2074 if (!lov->lov_tgts[index])
2075 /* Try again with the next index */
2076 GOTO(out_statfs, rc = -EAGAIN);
2078 client_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
2079 if (!lov->lov_tgts[index]->ltd_active)
2080 GOTO(out_uuid, rc = -ENODATA);
2084 GOTO(out_statfs, rc = -EINVAL);
2086 rc = obd_statfs(client_obd, &stat_buf, cfs_time_current_64() - 1);
2088 GOTO(out_statfs, rc);
2090 if (copy_to_user(data->ioc_pbuf1, &stat_buf, data->ioc_plen1))
2091 GOTO(out_statfs, rc = -EFAULT);
2094 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(client_obd),
2100 obd_ioctl_freedata(buf, len);
2104 int ll_process_config(struct lustre_cfg *lcfg)
2108 struct lprocfs_static_vars lvars;
2112 lprocfs_init_vars(llite, &lvars);
2114 /* The instance name contains the sb: lustre-client-aacfe000 */
2115 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2116 if (!ptr || !*(++ptr))
2118 if (sscanf(ptr, "%lx", &x) != 1)
2121 /* This better be a real Lustre superblock! */
2122 LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2124 /* Note we have not called client_common_fill_super yet, so
2125 proc fns must be able to handle that! */
2126 rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2131 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2132 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2133 struct inode *i1, struct inode *i2,
2134 const char *name, int namelen,
2135 int mode, __u32 opc, void *data)
2137 LASSERT(i1 != NULL);
2139 if (namelen > ll_i2sbi(i1)->ll_namelen)
2140 return ERR_PTR(-ENAMETOOLONG);
2142 if (op_data == NULL)
2143 OBD_ALLOC_PTR(op_data);
2145 if (op_data == NULL)
2146 return ERR_PTR(-ENOMEM);
2148 ll_i2gids(op_data->op_suppgids, i1, i2);
2149 op_data->op_fid1 = *ll_inode2fid(i1);
2150 op_data->op_capa1 = ll_mdscapa_get(i1);
2153 op_data->op_fid2 = *ll_inode2fid(i2);
2154 op_data->op_capa2 = ll_mdscapa_get(i2);
2156 fid_zero(&op_data->op_fid2);
2159 op_data->op_name = name;
2160 op_data->op_namelen = namelen;
2161 op_data->op_mode = mode;
2162 op_data->op_mod_time = CURRENT_SECONDS;
2163 op_data->op_fsuid = current->fsuid;
2164 op_data->op_fsgid = current->fsgid;
2165 op_data->op_cap = current->cap_effective;
2166 op_data->op_bias = MDS_CHECK_SPLIT;
2167 op_data->op_opc = opc;
2168 op_data->op_mds = 0;
2169 op_data->op_data = data;
2174 void ll_finish_md_op_data(struct md_op_data *op_data)
2176 capa_put(op_data->op_capa1);
2177 capa_put(op_data->op_capa2);
2178 OBD_FREE_PTR(op_data);
2181 int ll_ioctl_getfacl(struct inode *inode, struct rmtacl_ioctl_data *ioc)
2183 struct ll_sb_info *sbi = ll_i2sbi(inode);
2184 struct ptlrpc_request *req = NULL;
2185 struct mdt_body *body;
2187 struct obd_capa *oc;
2191 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT))
2194 LASSERT(ioc->cmd && ioc->cmd_len && ioc->res && ioc->res_len);
2196 OBD_ALLOC(cmd, ioc->cmd_len);
2199 if (copy_from_user(cmd, ioc->cmd, ioc->cmd_len))
2200 GOTO(out, rc = -EFAULT);
2202 oc = ll_mdscapa_get(inode);
2203 rc = md_getxattr(ll_i2sbi(inode)->ll_md_exp, ll_inode2fid(inode), oc,
2204 OBD_MD_FLXATTR, XATTR_NAME_LUSTRE_ACL, cmd,
2205 ioc->cmd_len, ioc->res_len, 0, &req);
2208 CERROR("mdc_getxattr %s [%s] failed: %d\n",
2209 XATTR_NAME_LUSTRE_ACL, cmd, rc);
2213 body = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF, sizeof(*body));
2216 buflen = lustre_msg_buflen(req->rq_repmsg, REPLY_REC_OFF);
2217 LASSERT(buflen <= ioc->res_len);
2218 buf = lustre_msg_string(req->rq_repmsg, REPLY_REC_OFF + 1, ioc->res_len);
2220 if (copy_to_user(ioc->res, buf, buflen))
2221 GOTO(out, rc = -EFAULT);
2225 ptlrpc_req_finished(req);
2226 OBD_FREE(cmd, ioc->cmd_len);
2230 int ll_ioctl_setfacl(struct inode *inode, struct rmtacl_ioctl_data *ioc)
2232 struct ll_sb_info *sbi = ll_i2sbi(inode);
2233 struct ptlrpc_request *req = NULL;
2235 struct obd_capa *oc;
2239 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT))
2242 if (!(sbi->ll_flags & LL_SBI_ACL))
2243 RETURN(-EOPNOTSUPP);
2245 LASSERT(ioc->cmd && ioc->cmd_len && ioc->res && ioc->res_len);
2247 OBD_ALLOC(cmd, ioc->cmd_len);
2250 if (copy_from_user(cmd, ioc->cmd, ioc->cmd_len))
2251 GOTO(out, rc = -EFAULT);
2253 oc = ll_mdscapa_get(inode);
2254 rc = md_setxattr(ll_i2sbi(inode)->ll_md_exp, ll_inode2fid(inode), oc,
2255 OBD_MD_FLXATTR, XATTR_NAME_LUSTRE_ACL, cmd,
2256 ioc->cmd_len, ioc->res_len, 0, &req);
2259 CERROR("mdc_setxattr %s [%s] failed: %d\n",
2260 XATTR_NAME_LUSTRE_ACL, cmd, rc);
2264 buflen = lustre_msg_buflen(req->rq_repmsg, REPLY_REC_OFF);
2265 LASSERT(buflen <= ioc->res_len);
2266 buf = lustre_msg_string(req->rq_repmsg, REPLY_REC_OFF, ioc->res_len);
2268 if (copy_to_user(ioc->res, buf, buflen))
2269 GOTO(out, rc = -EFAULT);
2273 ptlrpc_req_finished(req);
2274 OBD_FREE(cmd, ioc->cmd_len);