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 HAVE_LRU_RESIZE_SUPPORT
202 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
204 #ifdef CONFIG_FS_POSIX_ACL
205 data->ocd_connect_flags |= OBD_CONNECT_ACL;
207 data->ocd_ibits_known = MDS_INODELOCK_FULL;
208 data->ocd_version = LUSTRE_VERSION_CODE;
210 if (sb->s_flags & MS_RDONLY)
211 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
212 if (sbi->ll_flags & LL_SBI_USER_XATTR)
213 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
215 #ifdef HAVE_MS_FLOCK_LOCK
216 /* force vfs to use lustre handler for flock() calls - bug 10743 */
217 sb->s_flags |= MS_FLOCK_LOCK;
220 if (sbi->ll_flags & LL_SBI_FLOCK)
221 sbi->ll_fop = &ll_file_operations_flock;
222 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
223 sbi->ll_fop = &ll_file_operations;
225 sbi->ll_fop = &ll_file_operations_noflock;
228 data->ocd_connect_flags |= OBD_CONNECT_REAL;
229 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
230 data->ocd_connect_flags &= ~OBD_CONNECT_LCL_CLIENT;
231 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT;
233 data->ocd_connect_flags &= ~OBD_CONNECT_RMT_CLIENT;
234 data->ocd_connect_flags |= OBD_CONNECT_LCL_CLIENT;
236 data->ocd_nllu = nllu;
237 data->ocd_nllg = nllg;
239 err = obd_connect(NULL, &md_conn, obd, &sbi->ll_sb_uuid, data);
241 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
242 "recovery, of which this client is not a "
243 "part. Please wait for recovery to complete,"
244 " abort, or time out.\n", md);
247 CERROR("cannot connect to %s: rc = %d\n", md, err);
250 sbi->ll_md_exp = class_conn2export(&md_conn);
252 err = obd_fid_init(sbi->ll_md_exp);
254 CERROR("Can't init metadata layer FID infrastructure, "
259 err = obd_statfs(obd, &osfs, cfs_time_current_64() - HZ);
261 GOTO(out_md_fid, err);
263 size = sizeof(*data);
264 err = obd_get_info(sbi->ll_md_exp, strlen(KEY_CONN_DATA),
265 KEY_CONN_DATA, &size, data);
267 CERROR("Get connect data failed: %d \n", err);
271 LASSERT(osfs.os_bsize);
272 sb->s_blocksize = osfs.os_bsize;
273 sb->s_blocksize_bits = log2(osfs.os_bsize);
274 sb->s_magic = LL_SUPER_MAGIC;
276 /* for bug 11559. in $LINUX/fs/read_write.c, function do_sendfile():
277 * retval = in_file->f_op->sendfile(...);
279 * retval = -EOVERFLOW;
281 * it will check if *ppos is greater than max. However, max equals to
282 * s_maxbytes, which is a negative integer in a x86_64 box since loff_t
283 * has been defined as a signed long long ineger in linux kernel. */
284 #if BITS_PER_LONG == 64
285 sb->s_maxbytes = PAGE_CACHE_MAXBYTES >> 1;
287 sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
289 sbi->ll_namelen = osfs.os_namelen;
290 sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
292 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
293 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
294 LCONSOLE_INFO("Disabling user_xattr feature because "
295 "it is not supported on the server\n");
296 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
299 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
301 sb->s_flags |= MS_POSIXACL;
303 sbi->ll_flags |= LL_SBI_ACL;
305 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
307 sb->s_flags &= ~MS_POSIXACL;
309 sbi->ll_flags &= ~LL_SBI_ACL;
312 if (data->ocd_connect_flags & OBD_CONNECT_JOIN)
313 sbi->ll_flags |= LL_SBI_JOIN;
315 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
316 if (!(data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT)) {
317 /* sometimes local client claims to be remote, but mdt
318 * will disagree when client gss not applied. */
319 LCONSOLE_INFO("client claims to be remote, but server "
320 "rejected, forced to be local.\n");
321 sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
324 if (!(data->ocd_connect_flags & OBD_CONNECT_LCL_CLIENT)) {
325 /* with gss applied, remote client can not claim to be
326 * local, so mdt maybe force client to be remote. */
327 LCONSOLE_INFO("client claims to be local, but server "
328 "rejected, forced to be remote.\n");
329 sbi->ll_flags |= LL_SBI_RMT_CLIENT;
333 if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
334 LCONSOLE_INFO("client enabled MDS capability!\n");
335 sbi->ll_flags |= LL_SBI_MDS_CAPA;
338 if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
339 LCONSOLE_INFO("client enabled OSS capability!\n");
340 sbi->ll_flags |= LL_SBI_OSS_CAPA;
343 sbi->ll_sdev_orig = sb->s_dev;
344 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0))
345 /* We set sb->s_dev equal on all lustre clients in order to support
346 * NFS export clustering. NFSD requires that the FSID be the same
348 /* s_dev is also used in lt_compare() to compare two fs, but that is
349 * only a node-local comparison. */
351 /* XXX: this will not work with LMV */
352 sb->s_dev = get_uuid2int(sbi2mdc(sbi)->cl_target_uuid.uuid,
353 strlen(sbi2mdc(sbi)->cl_target_uuid.uuid));
356 obd = class_name2obd(dt);
358 CERROR("DT %s: not setup or attached\n", dt);
359 GOTO(out_md_fid, err = -ENODEV);
362 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
363 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
364 OBD_CONNECT_CANCELSET;
365 if (sbi->ll_flags & LL_SBI_OSS_CAPA)
366 data->ocd_connect_flags |= OBD_CONNECT_OSS_CAPA;
368 #ifdef HAVE_LRU_RESIZE_SUPPORT
369 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
371 CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
372 "ocd_grant: %d\n", data->ocd_connect_flags,
373 data->ocd_version, data->ocd_grant);
375 obd->obd_upcall.onu_owner = &sbi->ll_lco;
376 obd->obd_upcall.onu_upcall = ll_ocd_update;
377 data->ocd_brw_size = PTLRPC_MAX_BRW_PAGES << CFS_PAGE_SHIFT;
379 err = obd_connect(NULL, &dt_conn, obd, &sbi->ll_sb_uuid, data);
381 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
382 "recovery, of which this client is not a "
383 "part. Please wait for recovery to "
384 "complete, abort, or time out.\n", dt);
385 GOTO(out_md_fid, err);
387 CERROR("Cannot connect to %s: rc = %d\n", dt, err);
388 GOTO(out_md_fid, err);
391 sbi->ll_dt_exp = class_conn2export(&dt_conn);
393 err = obd_fid_init(sbi->ll_dt_exp);
395 CERROR("Can't init data layer FID infrastructure, "
400 spin_lock(&sbi->ll_lco.lco_lock);
401 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
402 spin_unlock(&sbi->ll_lco.lco_lock);
404 ll_init_ea_size(sbi->ll_md_exp, sbi->ll_dt_exp);
406 err = obd_prep_async_page(sbi->ll_dt_exp, NULL, NULL, NULL,
407 0, NULL, NULL, NULL);
409 LCONSOLE_ERROR_MSG(0x151, "There are no OST's in this "
410 "filesystem. There must be at least one "
411 "active OST for a client to start.\n");
412 GOTO(out_dt_fid, err);
415 if (!ll_async_page_slab) {
416 ll_async_page_slab_size =
417 size_round(sizeof(struct ll_async_page)) + err;
418 ll_async_page_slab = cfs_mem_cache_create("ll_async_page",
419 ll_async_page_slab_size,
421 if (!ll_async_page_slab)
422 GOTO(out_dt_fid, err = -ENOMEM);
425 err = md_getstatus(sbi->ll_md_exp, &rootfid, &oc);
427 CERROR("cannot mds_connect: rc = %d\n", err);
428 GOTO(out_dt_fid, err);
430 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&rootfid));
431 sbi->ll_root_fid = rootfid;
433 sb->s_op = &lustre_super_operations;
434 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
435 sb->s_export_op = &lustre_export_operations;
439 * XXX: move this to after cbd setup? */
440 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
441 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
442 valid |= OBD_MD_FLRMTPERM;
443 else if (sbi->ll_flags & LL_SBI_ACL)
444 valid |= OBD_MD_FLACL;
446 err = md_getattr(sbi->ll_md_exp, &rootfid, oc, valid, 0, &request);
450 CERROR("md_getattr failed for root: rc = %d\n", err);
451 GOTO(out_dt_fid, err);
453 memset(&lmd, 0, sizeof(lmd));
454 err = md_get_lustre_md(sbi->ll_md_exp, request,
455 REPLY_REC_OFF, sbi->ll_dt_exp, sbi->ll_md_exp,
458 CERROR("failed to understand root inode md: rc = %d\n", err);
459 ptlrpc_req_finished (request);
460 GOTO(out_dt_fid, err);
463 LASSERT(fid_is_sane(&sbi->ll_root_fid));
464 root = ll_iget(sb, ll_fid_build_ino(sbi, &sbi->ll_root_fid), &lmd);
465 md_free_lustre_md(sbi->ll_md_exp, &lmd);
466 ptlrpc_req_finished(request);
468 if (root == NULL || is_bad_inode(root)) {
470 obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
471 #ifdef CONFIG_FS_POSIX_ACL
473 posix_acl_release(lmd.posix_acl);
474 lmd.posix_acl = NULL;
477 CERROR("lustre_lite: bad iget4 for root\n");
478 GOTO(out_root, err = -EBADF);
481 err = ll_close_thread_start(&sbi->ll_lcq);
483 CERROR("cannot start close thread: rc %d\n", err);
487 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
488 err = obd_set_info_async(sbi->ll_dt_exp, strlen("checksum"),"checksum",
489 sizeof(checksum), &checksum, NULL);
491 /* making vm readahead 0 for 2.4.x. In the case of 2.6.x,
492 backing dev info assigned to inode mapping is used for
493 determining maximal readahead. */
494 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0)) && \
495 !defined(KERNEL_HAS_AS_MAX_READAHEAD)
496 /* bug 2805 - set VM readahead to zero */
497 vm_max_readahead = vm_min_readahead = 0;
500 sb->s_root = d_alloc_root(root);
502 OBD_FREE(data, sizeof(*data));
503 sb->s_root->d_op = &ll_d_root_ops;
509 obd_fid_fini(sbi->ll_dt_exp);
511 obd_disconnect(sbi->ll_dt_exp);
512 sbi->ll_dt_exp = NULL;
514 obd_fid_fini(sbi->ll_md_exp);
516 obd_disconnect(sbi->ll_md_exp);
517 sbi->ll_md_exp = NULL;
521 lprocfs_unregister_mountpoint(sbi);
525 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
529 *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
531 rc = obd_get_info(sbi->ll_md_exp, strlen("max_easize"), "max_easize",
534 CERROR("Get max mdsize error rc %d \n", rc);
539 void ll_dump_inode(struct inode *inode)
541 struct list_head *tmp;
542 int dentry_count = 0;
544 LASSERT(inode != NULL);
546 list_for_each(tmp, &inode->i_dentry)
549 CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
550 inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
551 inode->i_mode, atomic_read(&inode->i_count), dentry_count);
554 void lustre_dump_dentry(struct dentry *dentry, int recur)
556 struct list_head *tmp;
559 LASSERT(dentry != NULL);
561 list_for_each(tmp, &dentry->d_subdirs)
564 CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
565 " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
566 dentry->d_name.len, dentry->d_name.name,
567 dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
568 dentry->d_parent, dentry->d_inode, atomic_read(&dentry->d_count),
569 dentry->d_flags, dentry->d_fsdata, subdirs);
570 if (dentry->d_inode != NULL)
571 ll_dump_inode(dentry->d_inode);
576 list_for_each(tmp, &dentry->d_subdirs) {
577 struct dentry *d = list_entry(tmp, struct dentry, d_child);
578 lustre_dump_dentry(d, recur - 1);
582 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
583 void lustre_throw_orphan_dentries(struct super_block *sb)
585 struct dentry *dentry, *next;
586 struct ll_sb_info *sbi = ll_s2sbi(sb);
588 /* Do this to get rid of orphaned dentries. That is not really trw. */
589 list_for_each_entry_safe(dentry, next, &sbi->ll_orphan_dentry_list,
591 CWARN("found orphan dentry %.*s (%p->%p) at unmount, dumping "
592 "before and after shrink_dcache_parent\n",
593 dentry->d_name.len, dentry->d_name.name, dentry, next);
594 lustre_dump_dentry(dentry, 1);
595 shrink_dcache_parent(dentry);
596 lustre_dump_dentry(dentry, 1);
600 #define lustre_throw_orphan_dentries(sb)
603 #ifdef HAVE_EXPORT___IGET
604 static void prune_dir_dentries(struct inode *inode)
606 struct dentry *dentry, *prev = NULL;
608 /* due to lustre specific logic, a directory
609 * can have few dentries - a bug from VFS POV */
611 spin_lock(&dcache_lock);
612 if (!list_empty(&inode->i_dentry)) {
613 dentry = list_entry(inode->i_dentry.prev,
614 struct dentry, d_alias);
615 /* in order to prevent infinite loops we
616 * break if previous dentry is busy */
617 if (dentry != prev) {
620 spin_unlock(&dcache_lock);
622 /* try to kill all child dentries */
624 shrink_dcache_parent(dentry);
625 unlock_dentry(dentry);
628 /* now try to get rid of current dentry */
629 d_prune_aliases(inode);
633 spin_unlock(&dcache_lock);
636 static void prune_deathrow_one(struct ll_inode_info *lli)
638 struct inode *inode = ll_info2i(lli);
640 /* first, try to drop any dentries - they hold a ref on the inode */
641 if (S_ISDIR(inode->i_mode))
642 prune_dir_dentries(inode);
644 d_prune_aliases(inode);
647 /* if somebody still uses it, leave it */
648 LASSERT(atomic_read(&inode->i_count) > 0);
649 if (atomic_read(&inode->i_count) > 1)
652 CDEBUG(D_INODE, "inode %lu/%u(%d) looks a good candidate for prune\n",
653 inode->i_ino,inode->i_generation, atomic_read(&inode->i_count));
655 /* seems nobody uses it anymore */
663 static void prune_deathrow(struct ll_sb_info *sbi, int try)
665 struct ll_inode_info *lli;
669 if (need_resched() && try)
673 if (!spin_trylock(&sbi->ll_deathrow_lock))
676 spin_lock(&sbi->ll_deathrow_lock);
681 if (!list_empty(&sbi->ll_deathrow)) {
682 lli = list_entry(sbi->ll_deathrow.next,
683 struct ll_inode_info,
685 list_del_init(&lli->lli_dead_list);
686 if (!list_empty(&sbi->ll_deathrow))
689 spin_unlock(&sbi->ll_deathrow_lock);
692 prune_deathrow_one(lli);
694 } while (empty == 0);
696 #else /* !HAVE_EXPORT___IGET */
697 #define prune_deathrow(sbi, try) do {} while (0)
698 #endif /* HAVE_EXPORT___IGET */
700 void client_common_put_super(struct super_block *sb)
702 struct ll_sb_info *sbi = ll_s2sbi(sb);
705 obd_cancel_unused(sbi->ll_dt_exp, NULL, 0, NULL);
707 ll_close_thread_shutdown(sbi->ll_lcq);
709 /* destroy inodes in deathrow */
710 prune_deathrow(sbi, 0);
712 list_del(&sbi->ll_conn_chain);
714 obd_fid_fini(sbi->ll_dt_exp);
715 obd_disconnect(sbi->ll_dt_exp);
716 sbi->ll_dt_exp = NULL;
718 lprocfs_unregister_mountpoint(sbi);
720 obd_fid_fini(sbi->ll_md_exp);
721 obd_disconnect(sbi->ll_md_exp);
722 sbi->ll_md_exp = NULL;
724 lustre_throw_orphan_dentries(sb);
729 void ll_kill_super(struct super_block *sb)
731 struct ll_sb_info *sbi;
736 if (!(sb->s_flags & MS_ACTIVE))
740 /* we need restore s_dev from changed for clustred NFS before put_super
741 * because new kernels have cached s_dev and change sb->s_dev in
742 * put_super not affected real removing devices */
744 sb->s_dev = sbi->ll_sdev_orig;
748 char *ll_read_opt(const char *opt, char *data)
754 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
755 if (strncmp(opt, data, strlen(opt)))
757 if ((value = strchr(data, '=')) == NULL)
761 OBD_ALLOC(retval, strlen(value) + 1);
763 CERROR("out of memory!\n");
767 memcpy(retval, value, strlen(value)+1);
768 CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
772 static inline int ll_set_opt(const char *opt, char *data, int fl)
774 if (strncmp(opt, data, strlen(opt)) != 0)
780 /* non-client-specific mount options are parsed in lmd_parse */
781 static int ll_options(char *options, int *flags)
784 char *s1 = options, *s2;
790 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
793 CDEBUG(D_SUPER, "next opt=%s\n", s1);
794 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
799 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
804 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
809 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
814 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
819 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
824 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
826 /* Ignore deprecated mount option. The client will
827 * always try to mount with ACL support, whether this
828 * is used depends on whether server supports it. */
831 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
835 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
841 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
846 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
852 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
858 s2 = strchr(s1, ',');
866 void ll_lli_init(struct ll_inode_info *lli)
868 lli->lli_inode_magic = LLI_INODE_MAGIC;
869 sema_init(&lli->lli_size_sem, 1);
870 sema_init(&lli->lli_write_sem, 1);
872 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
873 spin_lock_init(&lli->lli_lock);
874 INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
875 INIT_LIST_HEAD(&lli->lli_close_list);
876 lli->lli_inode_magic = LLI_INODE_MAGIC;
877 sema_init(&lli->lli_och_sem, 1);
878 lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
879 lli->lli_mds_exec_och = NULL;
880 lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
881 lli->lli_open_fd_exec_count = 0;
882 INIT_LIST_HEAD(&lli->lli_dead_list);
883 lli->lli_remote_perms = NULL;
884 lli->lli_rmtperm_utime = 0;
885 sema_init(&lli->lli_rmtperm_sem, 1);
886 INIT_LIST_HEAD(&lli->lli_oss_capas);
889 int ll_fill_super(struct super_block *sb)
891 struct lustre_profile *lprof;
892 struct lustre_sb_info *lsi = s2lsi(sb);
893 struct ll_sb_info *sbi;
894 char *dt = NULL, *md = NULL;
895 char *profilenm = get_profile_name(sb);
896 struct config_llog_instance cfg = {0, };
897 char ll_instance[sizeof(sb) * 2 + 1];
901 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
905 sb->s_type->fs_flags |= FS_ODD_RENAME;
906 /* client additional sb info */
907 lsi->lsi_llsbi = sbi = ll_init_sbi();
913 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
917 /* Generate a string unique to this super, in case some joker tries
918 to mount the same fs at two mount points.
919 Use the address of the super itself.*/
920 sprintf(ll_instance, "%p", sb);
921 cfg.cfg_instance = ll_instance;
922 cfg.cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
924 /* set up client obds */
925 err = lustre_process_log(sb, profilenm, &cfg);
927 CERROR("Unable to process log: %d\n", err);
931 lprof = class_get_profile(profilenm);
933 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
934 " read from the MGS. Does that filesystem "
935 "exist?\n", profilenm);
936 GOTO(out_free, err = -EINVAL);
938 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
939 lprof->lp_md, lprof->lp_dt);
941 OBD_ALLOC(dt, strlen(lprof->lp_dt) +
942 strlen(ll_instance) + 2);
944 GOTO(out_free, err = -ENOMEM);
945 sprintf(dt, "%s-%s", lprof->lp_dt, ll_instance);
947 OBD_ALLOC(md, strlen(lprof->lp_md) +
948 strlen(ll_instance) + 2);
950 GOTO(out_free, err = -ENOMEM);
951 sprintf(md, "%s-%s", lprof->lp_md, ll_instance);
953 /* connections, registrations, sb setup */
954 err = client_common_fill_super(sb, md, dt,
955 lsi->lsi_lmd->lmd_nllu,
956 lsi->lsi_lmd->lmd_nllg);
960 OBD_FREE(md, strlen(md) + 1);
962 OBD_FREE(dt, strlen(dt) + 1);
966 LCONSOLE_WARN("Client %s has started\n", profilenm);
969 } /* ll_fill_super */
972 void ll_put_super(struct super_block *sb)
974 struct config_llog_instance cfg;
975 char ll_instance[sizeof(sb) * 2 + 1];
976 struct obd_device *obd;
977 struct lustre_sb_info *lsi = s2lsi(sb);
978 struct ll_sb_info *sbi = ll_s2sbi(sb);
979 char *profilenm = get_profile_name(sb);
983 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
985 ll_print_capa_stat(sbi);
987 sprintf(ll_instance, "%p", sb);
988 cfg.cfg_instance = ll_instance;
989 lustre_end_log(sb, NULL, &cfg);
991 if (sbi->ll_md_exp) {
992 obd = class_exp2obd(sbi->ll_md_exp);
994 force = obd->obd_force;
997 /* We need to set force before the lov_disconnect in
998 lustre_common_put_super, since l_d cleans up osc's as well. */
1001 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1003 obd->obd_force = force;
1008 /* Only if client_common_fill_super succeeded */
1009 client_common_put_super(sb);
1012 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1013 class_manual_cleanup(obd);
1017 class_del_profile(profilenm);
1020 lsi->lsi_llsbi = NULL;
1022 lustre_common_put_super(sb);
1024 LCONSOLE_WARN("client %s umount complete\n", ll_instance);
1029 } /* client_put_super */
1031 #ifdef HAVE_REGISTER_CACHE
1032 #include <linux/cache_def.h>
1033 #ifdef HAVE_CACHE_RETURN_INT
1038 ll_shrink_cache(int priority, unsigned int gfp_mask)
1040 struct ll_sb_info *sbi;
1043 list_for_each_entry(sbi, &ll_super_blocks, ll_list)
1044 count += llap_shrink_cache(sbi, priority);
1046 #ifdef HAVE_CACHE_RETURN_INT
1051 struct cache_definition ll_cache_definition = {
1052 .name = "llap_cache",
1053 .shrink = ll_shrink_cache
1055 #endif /* HAVE_REGISTER_CACHE */
1057 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1059 struct inode *inode = NULL;
1060 /* NOTE: we depend on atomic igrab() -bzzz */
1061 lock_res_and_lock(lock);
1062 if (lock->l_ast_data) {
1063 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1064 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1065 inode = igrab(lock->l_ast_data);
1067 inode = lock->l_ast_data;
1068 ldlm_lock_debug(NULL, inode->i_state & I_FREEING ?
1070 lock, __FILE__, __func__, __LINE__,
1071 "l_ast_data %p is bogus: magic %08x",
1072 lock->l_ast_data, lli->lli_inode_magic);
1076 unlock_res_and_lock(lock);
1080 static int null_if_equal(struct ldlm_lock *lock, void *data)
1082 if (data == lock->l_ast_data) {
1083 lock->l_ast_data = NULL;
1085 if (lock->l_req_mode != lock->l_granted_mode)
1086 LDLM_ERROR(lock,"clearing inode with ungranted lock");
1089 return LDLM_ITER_CONTINUE;
1092 void ll_clear_inode(struct inode *inode)
1094 struct ll_inode_info *lli = ll_i2info(inode);
1095 struct ll_sb_info *sbi = ll_i2sbi(inode);
1098 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1099 inode->i_generation, inode);
1101 ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1102 md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1103 null_if_equal, inode);
1105 LASSERT(!lli->lli_open_fd_write_count);
1106 LASSERT(!lli->lli_open_fd_read_count);
1107 LASSERT(!lli->lli_open_fd_exec_count);
1109 if (lli->lli_mds_write_och)
1110 ll_md_real_close(inode, FMODE_WRITE);
1111 if (lli->lli_mds_exec_och)
1112 ll_md_real_close(inode, FMODE_EXEC);
1113 if (lli->lli_mds_read_och)
1114 ll_md_real_close(inode, FMODE_READ);
1117 obd_change_cbdata(sbi->ll_dt_exp, lli->lli_smd,
1118 null_if_equal, inode);
1120 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1121 lli->lli_smd = NULL;
1124 if (lli->lli_symlink_name) {
1125 OBD_FREE(lli->lli_symlink_name,
1126 strlen(lli->lli_symlink_name) + 1);
1127 lli->lli_symlink_name = NULL;
1130 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1131 LASSERT(lli->lli_posix_acl == NULL);
1132 if (lli->lli_remote_perms) {
1133 free_rmtperm_hash(lli->lli_remote_perms);
1134 lli->lli_remote_perms = NULL;
1137 #ifdef CONFIG_FS_POSIX_ACL
1138 else if (lli->lli_posix_acl) {
1139 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1140 LASSERT(lli->lli_remote_perms == NULL);
1141 posix_acl_release(lli->lli_posix_acl);
1142 lli->lli_posix_acl = NULL;
1145 lli->lli_inode_magic = LLI_INODE_DEAD;
1147 #ifdef HAVE_EXPORT___IGET
1148 spin_lock(&sbi->ll_deathrow_lock);
1149 list_del_init(&lli->lli_dead_list);
1150 spin_unlock(&sbi->ll_deathrow_lock);
1152 ll_clear_inode_capas(inode);
1157 int ll_md_setattr(struct inode *inode, struct md_op_data *op_data,
1158 struct md_open_data **mod)
1160 struct lustre_md md;
1161 struct ll_sb_info *sbi = ll_i2sbi(inode);
1162 struct ptlrpc_request *request = NULL;
1166 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1167 LUSTRE_OPC_ANY, NULL);
1168 if (IS_ERR(op_data))
1169 RETURN(PTR_ERR(op_data));
1171 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1174 ptlrpc_req_finished(request);
1175 if (rc == -ENOENT) {
1177 /* Unlinked special device node? Or just a race?
1178 * Pretend we done everything. */
1179 if (!S_ISREG(inode->i_mode) &&
1180 !S_ISDIR(inode->i_mode))
1181 rc = inode_setattr(inode, &op_data->op_attr);
1182 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1183 CERROR("md_setattr fails: rc = %d\n", rc);
1188 rc = md_get_lustre_md(sbi->ll_md_exp, request, REPLY_REC_OFF,
1189 sbi->ll_dt_exp, sbi->ll_md_exp, &md);
1191 ptlrpc_req_finished(request);
1195 /* We call inode_setattr to adjust timestamps.
1196 * If there is at least some data in file, we cleared ATTR_SIZE
1197 * above to avoid invoking vmtruncate, otherwise it is important
1198 * to call vmtruncate in inode_setattr to update inode->i_size
1200 rc = inode_setattr(inode, &op_data->op_attr);
1202 /* Extract epoch data if obtained. */
1203 op_data->op_handle = md.body->handle;
1204 op_data->op_ioepoch = md.body->ioepoch;
1206 ll_update_inode(inode, &md);
1207 ptlrpc_req_finished(request);
1212 /* Close IO epoch and send Size-on-MDS attribute update. */
1213 static int ll_setattr_done_writing(struct inode *inode,
1214 struct md_op_data *op_data,
1215 struct md_open_data *mod)
1217 struct ll_inode_info *lli = ll_i2info(inode);
1221 LASSERT(op_data != NULL);
1222 if (!S_ISREG(inode->i_mode))
1225 CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1226 op_data->op_ioepoch, PFID(&lli->lli_fid));
1228 op_data->op_flags = MF_EPOCH_CLOSE | MF_SOM_CHANGE;
1229 rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1230 if (rc == -EAGAIN) {
1231 /* MDS has instructed us to obtain Size-on-MDS attribute
1232 * from OSTs and send setattr to back to MDS. */
1233 rc = ll_sizeonmds_update(inode, mod, &op_data->op_handle,
1234 op_data->op_ioepoch);
1236 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1242 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1243 * object(s) determine the file size and mtime. Otherwise, the MDS will
1244 * keep these values until such a time that objects are allocated for it.
1245 * We do the MDS operations first, as it is checking permissions for us.
1246 * We don't to the MDS RPC if there is nothing that we want to store there,
1247 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1248 * going to do an RPC anyways.
1250 * If we are doing a truncate, we will send the mtime and ctime updates
1251 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1252 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1255 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1257 struct ll_inode_info *lli = ll_i2info(inode);
1258 struct lov_stripe_md *lsm = lli->lli_smd;
1259 struct ll_sb_info *sbi = ll_i2sbi(inode);
1260 struct md_op_data *op_data = NULL;
1261 struct md_open_data *mod = NULL;
1262 int ia_valid = attr->ia_valid;
1263 int rc = 0, rc1 = 0;
1266 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1268 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_SETATTR, 1);
1270 if (ia_valid & ATTR_SIZE) {
1271 if (attr->ia_size > ll_file_maxbytes(inode)) {
1272 CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1273 attr->ia_size, ll_file_maxbytes(inode));
1277 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1280 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1281 if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1282 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1286 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1287 if (attr->ia_valid & ATTR_CTIME) {
1288 attr->ia_ctime = CURRENT_TIME;
1289 attr->ia_valid |= ATTR_CTIME_SET;
1291 if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1292 attr->ia_atime = CURRENT_TIME;
1293 attr->ia_valid |= ATTR_ATIME_SET;
1295 if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1296 attr->ia_mtime = CURRENT_TIME;
1297 attr->ia_valid |= ATTR_MTIME_SET;
1299 if ((attr->ia_valid & ATTR_CTIME) && !(attr->ia_valid & ATTR_MTIME)) {
1300 /* To avoid stale mtime on mds, obtain it from ost and send
1302 rc = ll_glimpse_size(inode, 0);
1306 attr->ia_valid |= ATTR_MTIME_SET | ATTR_MTIME;
1307 attr->ia_mtime = inode->i_mtime;
1310 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1311 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1312 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1315 /* NB: ATTR_SIZE will only be set after this point if the size
1316 * resides on the MDS, ie, this file has no objects. */
1318 attr->ia_valid &= ~ATTR_SIZE;
1320 /* We always do an MDS RPC, even if we're only changing the size;
1321 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1323 OBD_ALLOC_PTR(op_data);
1324 if (op_data == NULL)
1327 memcpy(&op_data->op_attr, attr, sizeof(*attr));
1329 /* Open epoch for truncate. */
1330 if (ia_valid & ATTR_SIZE)
1331 op_data->op_flags = MF_EPOCH_OPEN;
1333 rc = ll_md_setattr(inode, op_data, &mod);
1337 if (op_data->op_ioepoch)
1338 CDEBUG(D_INODE, "Epoch "LPU64" opened on "DFID" for "
1339 "truncate\n", op_data->op_ioepoch, PFID(&lli->lli_fid));
1341 if (!lsm || !S_ISREG(inode->i_mode)) {
1342 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1346 /* We really need to get our PW lock before we change inode->i_size.
1347 * If we don't we can race with other i_size updaters on our node, like
1348 * ll_file_read. We can also race with i_size propogation to other
1349 * nodes through dirtying and writeback of final cached pages. This
1350 * last one is especially bad for racing o_append users on other
1352 if (ia_valid & ATTR_SIZE) {
1353 ldlm_policy_data_t policy = { .l_extent = {attr->ia_size,
1355 struct lustre_handle lockh = { 0 };
1356 int err, ast_flags = 0;
1357 /* XXX when we fix the AST intents to pass the discard-range
1358 * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
1360 if (attr->ia_size == 0)
1361 ast_flags = LDLM_AST_DISCARD_DATA;
1363 UNLOCK_INODE_MUTEX(inode);
1364 UP_WRITE_I_ALLOC_SEM(inode);
1365 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy, &lockh,
1367 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1368 DOWN_WRITE_I_ALLOC_SEM(inode);
1369 LOCK_INODE_MUTEX(inode);
1371 LOCK_INODE_MUTEX(inode);
1372 DOWN_WRITE_I_ALLOC_SEM(inode);
1377 /* Only ll_inode_size_lock is taken at this level.
1378 * lov_stripe_lock() is grabbed by ll_truncate() only over
1379 * call to obd_adjust_kms(). If vmtruncate returns 0, then
1380 * ll_truncate dropped ll_inode_size_lock() */
1381 ll_inode_size_lock(inode, 0);
1382 rc = vmtruncate(inode, attr->ia_size);
1384 LASSERT(atomic_read(&lli->lli_size_sem.count) <= 0);
1385 ll_inode_size_unlock(inode, 0);
1388 err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
1390 CERROR("ll_extent_unlock failed: %d\n", err);
1394 } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
1396 struct obd_info oinfo = { { { 0 } } };
1399 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1400 inode->i_ino, LTIME_S(attr->ia_mtime));
1404 oa->o_id = lsm->lsm_object_id;
1405 oa->o_gr = lsm->lsm_object_gr;
1406 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1408 flags = OBD_MD_FLTYPE | OBD_MD_FLATIME |
1409 OBD_MD_FLMTIME | OBD_MD_FLCTIME |
1410 OBD_MD_FLFID | OBD_MD_FLGENER |
1413 obdo_from_inode(oa, inode, flags);
1417 oinfo.oi_capa = ll_mdscapa_get(inode);
1419 /* XXX: this looks unnecessary now. */
1420 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1421 capa_put(oinfo.oi_capa);
1423 CERROR("obd_setattr_async fails: rc=%d\n", rc);
1432 if (op_data->op_ioepoch)
1433 rc1 = ll_setattr_done_writing(inode, op_data, mod);
1434 ll_finish_md_op_data(op_data);
1436 return rc ? rc : rc1;
1439 int ll_setattr(struct dentry *de, struct iattr *attr)
1441 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1442 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1443 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1445 return ll_setattr_raw(de->d_inode, attr);
1448 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1451 struct ll_sb_info *sbi = ll_s2sbi(sb);
1452 struct obd_statfs obd_osfs;
1456 rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age);
1458 CERROR("md_statfs fails: rc = %d\n", rc);
1462 osfs->os_type = sb->s_magic;
1464 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1465 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1467 rc = obd_statfs_rqset(class_exp2obd(sbi->ll_dt_exp),
1468 &obd_osfs, max_age);
1470 CERROR("obd_statfs fails: rc = %d\n", rc);
1474 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1475 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1478 osfs->os_bsize = obd_osfs.os_bsize;
1479 osfs->os_blocks = obd_osfs.os_blocks;
1480 osfs->os_bfree = obd_osfs.os_bfree;
1481 osfs->os_bavail = obd_osfs.os_bavail;
1483 /* If we don't have as many objects free on the OST as inodes
1484 * on the MDS, we reduce the total number of inodes to
1485 * compensate, so that the "inodes in use" number is correct.
1487 if (obd_osfs.os_ffree < osfs->os_ffree) {
1488 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1490 osfs->os_ffree = obd_osfs.os_ffree;
1495 #ifndef HAVE_STATFS_DENTRY_PARAM
1496 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1499 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1501 struct super_block *sb = de->d_sb;
1503 struct obd_statfs osfs;
1506 CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1507 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1509 /* For now we will always get up-to-date statfs values, but in the
1510 * future we may allow some amount of caching on the client (e.g.
1511 * from QOS or lprocfs updates). */
1512 rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - 1);
1516 statfs_unpack(sfs, &osfs);
1518 /* We need to downshift for all 32-bit kernels, because we can't
1519 * tell if the kernel is being called via sys_statfs64() or not.
1520 * Stop before overflowing f_bsize - in which case it is better
1521 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1522 if (sizeof(long) < 8) {
1523 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1526 osfs.os_blocks >>= 1;
1527 osfs.os_bfree >>= 1;
1528 osfs.os_bavail >>= 1;
1532 sfs->f_blocks = osfs.os_blocks;
1533 sfs->f_bfree = osfs.os_bfree;
1534 sfs->f_bavail = osfs.os_bavail;
1539 void ll_inode_size_lock(struct inode *inode, int lock_lsm)
1541 struct ll_inode_info *lli;
1542 struct lov_stripe_md *lsm;
1544 lli = ll_i2info(inode);
1545 LASSERT(lli->lli_size_sem_owner != current);
1546 down(&lli->lli_size_sem);
1547 LASSERT(lli->lli_size_sem_owner == NULL);
1548 lli->lli_size_sem_owner = current;
1550 LASSERTF(lsm != NULL || lock_lsm == 0, "lsm %p, lock_lsm %d\n",
1553 lov_stripe_lock(lsm);
1556 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm)
1558 struct ll_inode_info *lli;
1559 struct lov_stripe_md *lsm;
1561 lli = ll_i2info(inode);
1563 LASSERTF(lsm != NULL || unlock_lsm == 0, "lsm %p, lock_lsm %d\n",
1566 lov_stripe_unlock(lsm);
1567 LASSERT(lli->lli_size_sem_owner == current);
1568 lli->lli_size_sem_owner = NULL;
1569 up(&lli->lli_size_sem);
1572 static void ll_replace_lsm(struct inode *inode, struct lov_stripe_md *lsm)
1574 struct ll_inode_info *lli = ll_i2info(inode);
1576 dump_lsm(D_INODE, lsm);
1577 dump_lsm(D_INODE, lli->lli_smd);
1578 LASSERTF(lsm->lsm_magic == LOV_MAGIC_JOIN,
1579 "lsm must be joined lsm %p\n", lsm);
1580 obd_free_memmd(ll_i2dtexp(inode), &lli->lli_smd);
1581 CDEBUG(D_INODE, "replace lsm %p to lli_smd %p for inode %lu%u(%p)\n",
1582 lsm, lli->lli_smd, inode->i_ino, inode->i_generation, inode);
1584 lli->lli_maxbytes = lsm->lsm_maxbytes;
1585 if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1586 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1589 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1591 struct ll_inode_info *lli = ll_i2info(inode);
1592 struct mdt_body *body = md->body;
1593 struct lov_stripe_md *lsm = md->lsm;
1594 struct ll_sb_info *sbi = ll_i2sbi(inode);
1596 LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1598 if (lli->lli_smd == NULL) {
1599 if (lsm->lsm_magic != LOV_MAGIC &&
1600 lsm->lsm_magic != LOV_MAGIC_JOIN) {
1601 dump_lsm(D_ERROR, lsm);
1604 CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1605 lsm, inode->i_ino, inode->i_generation, inode);
1606 /* ll_inode_size_lock() requires it is only called
1607 * with lli_smd != NULL or lock_lsm == 0 or we can
1608 * race between lock/unlock. bug 9547 */
1610 lli->lli_maxbytes = lsm->lsm_maxbytes;
1611 if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1612 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1614 if (lli->lli_smd->lsm_magic == lsm->lsm_magic &&
1615 lli->lli_smd->lsm_stripe_count ==
1616 lsm->lsm_stripe_count) {
1617 if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
1618 CERROR("lsm mismatch for inode %ld\n",
1620 CERROR("lli_smd:\n");
1621 dump_lsm(D_ERROR, lli->lli_smd);
1623 dump_lsm(D_ERROR, lsm);
1627 ll_replace_lsm(inode, lsm);
1629 if (lli->lli_smd != lsm)
1630 obd_free_memmd(ll_i2dtexp(inode), &lsm);
1633 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1634 if (body->valid & OBD_MD_FLRMTPERM)
1635 ll_update_remote_perm(inode, md->remote_perm);
1637 #ifdef CONFIG_FS_POSIX_ACL
1638 else if (body->valid & OBD_MD_FLACL) {
1639 spin_lock(&lli->lli_lock);
1640 if (lli->lli_posix_acl)
1641 posix_acl_release(lli->lli_posix_acl);
1642 lli->lli_posix_acl = md->posix_acl;
1643 spin_unlock(&lli->lli_lock);
1646 if (body->valid & OBD_MD_FLATIME &&
1647 body->atime > LTIME_S(inode->i_atime))
1648 LTIME_S(inode->i_atime) = body->atime;
1650 /* mtime is always updated with ctime, but can be set in past.
1651 As write and utime(2) may happen within 1 second, and utime's
1652 mtime has a priority over write's one, so take mtime from mds
1653 for the same ctimes. */
1654 if (body->valid & OBD_MD_FLCTIME &&
1655 body->ctime >= LTIME_S(inode->i_ctime)) {
1656 LTIME_S(inode->i_ctime) = body->ctime;
1657 if (body->valid & OBD_MD_FLMTIME) {
1658 CDEBUG(D_INODE, "setting ino %lu mtime "
1659 "from %lu to "LPU64"\n", inode->i_ino,
1660 LTIME_S(inode->i_mtime), body->mtime);
1661 LTIME_S(inode->i_mtime) = body->mtime;
1664 if (body->valid & OBD_MD_FLMODE)
1665 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1666 if (body->valid & OBD_MD_FLTYPE)
1667 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1668 if (S_ISREG(inode->i_mode)) {
1669 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1671 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1673 #ifdef HAVE_INODE_BLKSIZE
1674 inode->i_blksize = 1<<inode->i_blkbits;
1676 if (body->valid & OBD_MD_FLUID)
1677 inode->i_uid = body->uid;
1678 if (body->valid & OBD_MD_FLGID)
1679 inode->i_gid = body->gid;
1680 if (body->valid & OBD_MD_FLFLAGS)
1681 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1682 if (body->valid & OBD_MD_FLNLINK)
1683 inode->i_nlink = body->nlink;
1684 if (body->valid & OBD_MD_FLRDEV)
1685 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1686 inode->i_rdev = body->rdev;
1688 inode->i_rdev = old_decode_dev(body->rdev);
1690 if (body->valid & OBD_MD_FLSIZE) {
1691 if (ll_i2mdexp(inode)->exp_connect_flags & OBD_CONNECT_SOM) {
1692 if (lli->lli_flags & (LLIF_DONE_WRITING |
1693 LLIF_EPOCH_PENDING |
1695 CWARN("ino %lu flags %lu still has size authority!"
1696 "do not trust the size got from MDS\n",
1697 inode->i_ino, lli->lli_flags);
1699 i_size_write(inode, body->size);
1700 lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1703 i_size_write(inode, body->size);
1706 if (body->valid & OBD_MD_FLBLOCKS)
1707 inode->i_blocks = body->blocks;
1710 if (body->valid & OBD_MD_FLID) {
1711 /* FID shouldn't be changed! */
1712 if (fid_is_sane(&lli->lli_fid)) {
1713 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1714 "Trying to change FID "DFID
1715 " to the "DFID", inode %lu/%u(%p)\n",
1716 PFID(&lli->lli_fid), PFID(&body->fid1),
1717 inode->i_ino, inode->i_generation, inode);
1719 lli->lli_fid = body->fid1;
1722 LASSERT(fid_seq(&lli->lli_fid) != 0);
1724 if (body->valid & OBD_MD_FLMDSCAPA) {
1725 LASSERT(md->mds_capa);
1726 ll_add_capa(inode, md->mds_capa);
1728 if (body->valid & OBD_MD_FLOSSCAPA) {
1729 LASSERT(md->oss_capa);
1730 ll_add_capa(inode, md->oss_capa);
1734 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
1735 static struct backing_dev_info ll_backing_dev_info = {
1736 .ra_pages = 0, /* No readahead */
1737 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,12))
1738 .capabilities = 0, /* Does contribute to dirty memory */
1740 .memory_backed = 0, /* Does contribute to dirty memory */
1745 void ll_read_inode2(struct inode *inode, void *opaque)
1747 struct lustre_md *md = opaque;
1748 struct ll_inode_info *lli = ll_i2info(inode);
1751 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n",
1752 inode->i_ino, inode->i_generation, inode);
1756 LASSERT(!lli->lli_smd);
1758 /* Core attributes from the MDS first. This is a new inode, and
1759 * the VFS doesn't zero times in the core inode so we have to do
1760 * it ourselves. They will be overwritten by either MDS or OST
1761 * attributes - we just need to make sure they aren't newer. */
1762 LTIME_S(inode->i_mtime) = 0;
1763 LTIME_S(inode->i_atime) = 0;
1764 LTIME_S(inode->i_ctime) = 0;
1766 ll_update_inode(inode, md);
1768 /* OIDEBUG(inode); */
1770 if (S_ISREG(inode->i_mode)) {
1771 struct ll_sb_info *sbi = ll_i2sbi(inode);
1772 inode->i_op = &ll_file_inode_operations;
1773 inode->i_fop = sbi->ll_fop;
1774 inode->i_mapping->a_ops = &ll_aops;
1776 } else if (S_ISDIR(inode->i_mode)) {
1777 inode->i_op = &ll_dir_inode_operations;
1778 inode->i_fop = &ll_dir_operations;
1779 inode->i_mapping->a_ops = &ll_dir_aops;
1781 } else if (S_ISLNK(inode->i_mode)) {
1782 inode->i_op = &ll_fast_symlink_inode_operations;
1785 inode->i_op = &ll_special_inode_operations;
1787 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
1788 init_special_inode(inode, inode->i_mode,
1789 kdev_t_to_nr(inode->i_rdev));
1791 /* initializing backing dev info. */
1792 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
1794 init_special_inode(inode, inode->i_mode, inode->i_rdev);
1800 void ll_delete_inode(struct inode *inode)
1802 struct ll_sb_info *sbi = ll_i2sbi(inode);
1806 rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
1808 CERROR("fid_delete() failed, rc %d\n", rc);
1815 int ll_iocontrol(struct inode *inode, struct file *file,
1816 unsigned int cmd, unsigned long arg)
1818 struct ll_sb_info *sbi = ll_i2sbi(inode);
1819 struct ptlrpc_request *req = NULL;
1824 case EXT3_IOC_GETFLAGS: {
1825 struct mdt_body *body;
1826 struct obd_capa *oc;
1828 oc = ll_mdscapa_get(inode);
1829 rc = md_getattr(sbi->ll_md_exp, ll_inode2fid(inode), oc,
1830 OBD_MD_FLFLAGS, 0, &req);
1833 CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1837 body = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
1840 /*Now the ext3 will be packed directly back to client,
1841 *no need convert here*/
1842 flags = body->flags;
1844 ptlrpc_req_finished (req);
1846 RETURN(put_user(flags, (int *)arg));
1848 case EXT3_IOC_SETFLAGS: {
1849 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1850 struct obd_info oinfo = { { { 0 } } };
1851 struct md_op_data *op_data;
1853 if (get_user(flags, (int *)arg))
1857 OBDO_ALLOC(oinfo.oi_oa);
1861 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1862 LUSTRE_OPC_ANY, NULL);
1863 if (IS_ERR(op_data))
1864 RETURN(PTR_ERR(op_data));
1866 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1867 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1868 rc = md_setattr(sbi->ll_md_exp, op_data,
1869 NULL, 0, NULL, 0, &req, NULL);
1870 ll_finish_md_op_data(op_data);
1871 ptlrpc_req_finished(req);
1872 if (rc || lsm == NULL) {
1873 OBDO_FREE(oinfo.oi_oa);
1877 oinfo.oi_oa->o_id = lsm->lsm_object_id;
1878 oinfo.oi_oa->o_gr = lsm->lsm_object_gr;
1879 oinfo.oi_oa->o_flags = flags;
1880 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1882 oinfo.oi_capa = ll_mdscapa_get(inode);
1884 obdo_from_inode(oinfo.oi_oa, inode,
1885 OBD_MD_FLFID | OBD_MD_FLGENER);
1886 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1887 capa_put(oinfo.oi_capa);
1888 OBDO_FREE(oinfo.oi_oa);
1890 if (rc != -EPERM && rc != -EACCES)
1891 CERROR("md_setattr_async fails: rc = %d\n", rc);
1895 inode->i_flags = ll_ext_to_inode_flags(flags |
1896 MDS_BFLAG_EXT_FLAGS);
1906 int ll_flush_ctx(struct inode *inode)
1908 struct ll_sb_info *sbi = ll_i2sbi(inode);
1910 CDEBUG(D_SEC, "flush context for user %d\n", current->uid);
1912 obd_set_info_async(sbi->ll_md_exp,
1913 sizeof(KEY_FLUSH_CTX) - 1, KEY_FLUSH_CTX,
1915 obd_set_info_async(sbi->ll_dt_exp,
1916 sizeof(KEY_FLUSH_CTX) - 1, KEY_FLUSH_CTX,
1921 /* umount -f client means force down, don't save state */
1922 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1923 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
1925 struct super_block *sb = vfsmnt->mnt_sb;
1927 void ll_umount_begin(struct super_block *sb)
1930 struct lustre_sb_info *lsi = s2lsi(sb);
1931 struct ll_sb_info *sbi = ll_s2sbi(sb);
1932 struct obd_device *obd;
1933 struct obd_ioctl_data ioc_data = { 0 };
1936 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1937 if (!(flags & MNT_FORCE)) {
1943 /* Tell the MGC we got umount -f */
1944 lsi->lsi_flags |= LSI_UMOUNT_FORCE;
1946 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1947 sb->s_count, atomic_read(&sb->s_active));
1949 obd = class_exp2obd(sbi->ll_md_exp);
1951 CERROR("Invalid MDC connection handle "LPX64"\n",
1952 sbi->ll_md_exp->exp_handle.h_cookie);
1957 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp, sizeof ioc_data,
1960 obd = class_exp2obd(sbi->ll_dt_exp);
1962 CERROR("Invalid LOV connection handle "LPX64"\n",
1963 sbi->ll_dt_exp->exp_handle.h_cookie);
1969 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp, sizeof ioc_data,
1972 /* Really, we'd like to wait until there are no requests outstanding,
1973 * and then continue. For now, we just invalidate the requests,
1974 * schedule, and hope.
1981 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
1983 struct ll_sb_info *sbi = ll_s2sbi(sb);
1987 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
1988 read_only = *flags & MS_RDONLY;
1989 err = obd_set_info_async(sbi->ll_md_exp,
1990 sizeof(KEY_READ_ONLY) - 1,
1991 KEY_READ_ONLY, sizeof(read_only),
1994 CERROR("Failed to change the read-only flag during "
1995 "remount: %d\n", err);
2000 sb->s_flags |= MS_RDONLY;
2002 sb->s_flags &= ~MS_RDONLY;
2007 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2008 int offset, struct super_block *sb)
2010 struct ll_sb_info *sbi = NULL;
2011 struct lustre_md md;
2015 LASSERT(*inode || sb);
2016 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2017 prune_deathrow(sbi, 1);
2018 memset(&md, 0, sizeof(struct lustre_md));
2020 rc = md_get_lustre_md(sbi->ll_md_exp, req, offset,
2021 sbi->ll_dt_exp, sbi->ll_md_exp, &md);
2026 ll_update_inode(*inode, &md);
2028 LASSERT(sb != NULL);
2031 * At this point server returns to client's same fid as client
2032 * generated for creating. So using ->fid1 is okay here.
2034 LASSERT(fid_is_sane(&md.body->fid1));
2036 *inode = ll_iget(sb, ll_fid_build_ino(sbi, &md.body->fid1), &md);
2037 if (*inode == NULL || is_bad_inode(*inode)) {
2039 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2040 #ifdef CONFIG_FS_POSIX_ACL
2042 posix_acl_release(md.posix_acl);
2043 md.posix_acl = NULL;
2047 CERROR("new_inode -fatal: rc %d\n", rc);
2052 rc = obd_checkmd(sbi->ll_dt_exp, sbi->ll_md_exp,
2053 ll_i2info(*inode)->lli_smd);
2055 md_free_lustre_md(sbi->ll_md_exp, &md);
2059 char *llap_origins[] = {
2060 [LLAP_ORIGIN_UNKNOWN] = "--",
2061 [LLAP_ORIGIN_READPAGE] = "rp",
2062 [LLAP_ORIGIN_READAHEAD] = "ra",
2063 [LLAP_ORIGIN_COMMIT_WRITE] = "cw",
2064 [LLAP_ORIGIN_WRITEPAGE] = "wp",
2067 struct ll_async_page *llite_pglist_next_llap(struct ll_sb_info *sbi,
2068 struct list_head *list)
2070 struct ll_async_page *llap;
2071 struct list_head *pos;
2073 list_for_each(pos, list) {
2074 if (pos == &sbi->ll_pglist)
2076 llap = list_entry(pos, struct ll_async_page, llap_pglist_item);
2077 if (llap->llap_page == NULL)
2085 int ll_obd_statfs(struct inode *inode, void *arg)
2087 struct ll_sb_info *sbi = NULL;
2088 struct obd_device *client_obd = NULL, *lov_obd = NULL;
2089 struct lov_obd *lov = NULL;
2090 struct obd_statfs stat_buf = {0};
2092 struct obd_ioctl_data *data = NULL;
2096 if (!inode || !(sbi = ll_i2sbi(inode)))
2097 GOTO(out_statfs, rc = -EINVAL);
2099 rc = obd_ioctl_getdata(&buf, &len, arg);
2101 GOTO(out_statfs, rc);
2104 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2105 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2106 GOTO(out_statfs, rc = -EINVAL);
2108 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2109 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
2111 if (type == LL_STATFS_MDC) {
2113 GOTO(out_statfs, rc = -ENODEV);
2114 client_obd = class_exp2obd(sbi->ll_md_exp);
2115 } else if (type == LL_STATFS_LOV) {
2116 lov_obd = class_exp2obd(sbi->ll_dt_exp);
2117 lov = &lov_obd->u.lov;
2119 if ((index >= lov->desc.ld_tgt_count))
2120 GOTO(out_statfs, rc = -ENODEV);
2121 if (!lov->lov_tgts[index])
2122 /* Try again with the next index */
2123 GOTO(out_statfs, rc = -EAGAIN);
2125 client_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
2126 if (!lov->lov_tgts[index]->ltd_active)
2127 GOTO(out_uuid, rc = -ENODATA);
2131 GOTO(out_statfs, rc = -EINVAL);
2133 rc = obd_statfs(client_obd, &stat_buf, cfs_time_current_64() - 1);
2135 GOTO(out_statfs, rc);
2137 if (copy_to_user(data->ioc_pbuf1, &stat_buf, data->ioc_plen1))
2138 GOTO(out_statfs, rc = -EFAULT);
2141 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(client_obd),
2147 obd_ioctl_freedata(buf, len);
2151 int ll_process_config(struct lustre_cfg *lcfg)
2155 struct lprocfs_static_vars lvars;
2159 lprocfs_init_vars(llite, &lvars);
2161 /* The instance name contains the sb: lustre-client-aacfe000 */
2162 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2163 if (!ptr || !*(++ptr))
2165 if (sscanf(ptr, "%lx", &x) != 1)
2168 /* This better be a real Lustre superblock! */
2169 LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2171 /* Note we have not called client_common_fill_super yet, so
2172 proc fns must be able to handle that! */
2173 rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2178 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2179 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2180 struct inode *i1, struct inode *i2,
2181 const char *name, int namelen,
2182 int mode, __u32 opc, void *data)
2184 LASSERT(i1 != NULL);
2186 if (namelen > ll_i2sbi(i1)->ll_namelen)
2187 return ERR_PTR(-ENAMETOOLONG);
2189 if (op_data == NULL)
2190 OBD_ALLOC_PTR(op_data);
2192 if (op_data == NULL)
2193 return ERR_PTR(-ENOMEM);
2195 ll_i2gids(op_data->op_suppgids, i1, i2);
2196 op_data->op_fid1 = *ll_inode2fid(i1);
2197 op_data->op_capa1 = ll_mdscapa_get(i1);
2200 op_data->op_fid2 = *ll_inode2fid(i2);
2201 op_data->op_capa2 = ll_mdscapa_get(i2);
2203 fid_zero(&op_data->op_fid2);
2206 op_data->op_name = name;
2207 op_data->op_namelen = namelen;
2208 op_data->op_mode = mode;
2209 op_data->op_mod_time = CURRENT_SECONDS;
2210 op_data->op_fsuid = current->fsuid;
2211 op_data->op_fsgid = current->fsgid;
2212 op_data->op_cap = current->cap_effective;
2213 op_data->op_bias = MDS_CHECK_SPLIT;
2214 op_data->op_opc = opc;
2215 op_data->op_mds = 0;
2216 op_data->op_data = data;
2221 void ll_finish_md_op_data(struct md_op_data *op_data)
2223 capa_put(op_data->op_capa1);
2224 capa_put(op_data->op_capa2);
2225 OBD_FREE_PTR(op_data);
2228 int ll_ioctl_getfacl(struct inode *inode, struct rmtacl_ioctl_data *ioc)
2230 struct ll_sb_info *sbi = ll_i2sbi(inode);
2231 struct ptlrpc_request *req = NULL;
2232 struct mdt_body *body;
2234 struct obd_capa *oc;
2238 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT))
2241 LASSERT(ioc->cmd && ioc->cmd_len && ioc->res && ioc->res_len);
2243 OBD_ALLOC(cmd, ioc->cmd_len);
2246 if (copy_from_user(cmd, ioc->cmd, ioc->cmd_len))
2247 GOTO(out, rc = -EFAULT);
2249 oc = ll_mdscapa_get(inode);
2250 rc = md_getxattr(ll_i2sbi(inode)->ll_md_exp, ll_inode2fid(inode), oc,
2251 OBD_MD_FLXATTR, XATTR_NAME_LUSTRE_ACL, cmd,
2252 ioc->cmd_len, ioc->res_len, 0, &req);
2255 CERROR("mdc_getxattr %s [%s] failed: %d\n",
2256 XATTR_NAME_LUSTRE_ACL, cmd, rc);
2260 body = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF, sizeof(*body));
2263 buflen = lustre_msg_buflen(req->rq_repmsg, REPLY_REC_OFF);
2264 LASSERT(buflen <= ioc->res_len);
2265 buf = lustre_msg_string(req->rq_repmsg, REPLY_REC_OFF + 1, ioc->res_len);
2267 if (copy_to_user(ioc->res, buf, buflen))
2268 GOTO(out, rc = -EFAULT);
2272 ptlrpc_req_finished(req);
2273 OBD_FREE(cmd, ioc->cmd_len);
2277 int ll_ioctl_setfacl(struct inode *inode, struct rmtacl_ioctl_data *ioc)
2279 struct ll_sb_info *sbi = ll_i2sbi(inode);
2280 struct ptlrpc_request *req = NULL;
2282 struct obd_capa *oc;
2286 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT))
2289 if (!(sbi->ll_flags & LL_SBI_ACL))
2290 RETURN(-EOPNOTSUPP);
2292 LASSERT(ioc->cmd && ioc->cmd_len && ioc->res && ioc->res_len);
2294 OBD_ALLOC(cmd, ioc->cmd_len);
2297 if (copy_from_user(cmd, ioc->cmd, ioc->cmd_len))
2298 GOTO(out, rc = -EFAULT);
2300 oc = ll_mdscapa_get(inode);
2301 rc = md_setxattr(ll_i2sbi(inode)->ll_md_exp, ll_inode2fid(inode), oc,
2302 OBD_MD_FLXATTR, XATTR_NAME_LUSTRE_ACL, cmd,
2303 ioc->cmd_len, ioc->res_len, 0, &req);
2306 CERROR("mdc_setxattr %s [%s] failed: %d\n",
2307 XATTR_NAME_LUSTRE_ACL, cmd, rc);
2311 buflen = lustre_msg_buflen(req->rq_repmsg, REPLY_REC_OFF);
2312 LASSERT(buflen <= ioc->res_len);
2313 buf = lustre_msg_string(req->rq_repmsg, REPLY_REC_OFF, ioc->res_len);
2315 if (copy_to_user(ioc->res, buf, buflen))
2316 GOTO(out, rc = -EFAULT);
2320 ptlrpc_req_finished(req);
2321 OBD_FREE(cmd, ioc->cmd_len);