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>
32 #include <lustre_lite.h>
33 #include <lustre_ha.h>
34 #include <lustre_dlm.h>
35 #include <lprocfs_status.h>
36 #include <lustre_disk.h>
37 #include <lustre_param.h>
38 #include <lustre_log.h>
39 #include <obd_cksum.h>
40 #include <lustre_cache.h>
41 #include "llite_internal.h"
43 cfs_mem_cache_t *ll_file_data_slab;
45 LIST_HEAD(ll_super_blocks);
46 spinlock_t ll_sb_lock = SPIN_LOCK_UNLOCKED;
48 extern struct address_space_operations ll_aops;
49 extern struct address_space_operations ll_dir_aops;
52 #define log2(n) ffz(~(n))
56 static struct ll_sb_info *ll_init_sbi(void)
58 struct ll_sb_info *sbi = NULL;
65 OBD_ALLOC(sbi, sizeof(*sbi));
69 spin_lock_init(&sbi->ll_lock);
70 spin_lock_init(&sbi->ll_lco.lco_lock);
71 spin_lock_init(&sbi->ll_pp_extent_lock);
72 spin_lock_init(&sbi->ll_process_lock);
73 sbi->ll_rw_stats_on = 0;
74 INIT_LIST_HEAD(&sbi->ll_pglist);
77 pages = si.totalram - si.totalhigh;
78 if (pages >> (20 - CFS_PAGE_SHIFT) < 512)
79 sbi->ll_async_page_max = pages / 2;
81 sbi->ll_async_page_max = (pages / 4) * 3;
82 sbi->ll_ra_info.ra_max_pages = min(num_physpages / 8,
83 SBI_DEFAULT_READAHEAD_MAX);
84 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
85 SBI_DEFAULT_READAHEAD_WHOLE_MAX;
86 sbi->ll_contention_time = SBI_DEFAULT_CONTENTION_SECONDS;
87 sbi->ll_lockless_truncate_enable = SBI_DEFAULT_LOCKLESS_TRUNCATE_ENABLE;
88 INIT_LIST_HEAD(&sbi->ll_conn_chain);
89 INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
91 ll_generate_random_uuid(uuid);
92 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
93 CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
95 spin_lock(&ll_sb_lock);
96 list_add_tail(&sbi->ll_list, &ll_super_blocks);
97 spin_unlock(&ll_sb_lock);
99 #ifdef ENABLE_LLITE_CHECKSUM
100 sbi->ll_flags |= LL_SBI_CHECKSUM;
103 #ifdef HAVE_LRU_RESIZE_SUPPORT
104 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
107 #ifdef HAVE_EXPORT___IGET
108 INIT_LIST_HEAD(&sbi->ll_deathrow);
109 spin_lock_init(&sbi->ll_deathrow_lock);
111 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
112 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].pp_r_hist.oh_lock);
113 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].pp_w_hist.oh_lock);
116 /* metadata statahead is enabled by default */
117 sbi->ll_sa_max = LL_SA_RPC_DEF;
122 void ll_free_sbi(struct super_block *sb)
124 struct ll_sb_info *sbi = ll_s2sbi(sb);
128 spin_lock(&ll_sb_lock);
129 list_del(&sbi->ll_list);
130 spin_unlock(&ll_sb_lock);
131 OBD_FREE(sbi, sizeof(*sbi));
136 static struct dentry_operations ll_d_root_ops = {
137 #ifdef DCACHE_LUSTRE_INVALID
138 .d_compare = ll_dcompare,
142 /* Initialize the default and maximum LOV EA and cookie sizes. This allows
143 * us to make MDS RPCs with large enough reply buffers to hold the
144 * maximum-sized (= maximum striped) EA and cookie without having to
145 * calculate this (via a call into the LOV + OSCs) each time we make an RPC. */
146 static int ll_init_ea_size(struct obd_export *md_exp, struct obd_export *dt_exp)
148 struct lov_stripe_md lsm = { .lsm_magic = LOV_MAGIC };
149 __u32 valsize = sizeof(struct lov_desc);
150 int rc, easize, def_easize, cookiesize;
151 struct lov_desc desc;
155 rc = obd_get_info(dt_exp, sizeof(KEY_LOVDESC), KEY_LOVDESC,
160 stripes = min(desc.ld_tgt_count, (__u32)LOV_MAX_STRIPE_COUNT);
161 lsm.lsm_stripe_count = stripes;
162 easize = obd_size_diskmd(dt_exp, &lsm);
164 lsm.lsm_stripe_count = desc.ld_default_stripe_count;
165 def_easize = obd_size_diskmd(dt_exp, &lsm);
167 cookiesize = stripes * sizeof(struct llog_cookie);
169 CDEBUG(D_HA, "updating max_mdsize/max_cookiesize: %d/%d\n",
172 rc = md_init_ea_size(md_exp, easize, def_easize, cookiesize);
176 static int client_common_fill_super(struct super_block *sb, char *md, char *dt)
178 struct inode *root = 0;
179 struct ll_sb_info *sbi = ll_s2sbi(sb);
180 struct obd_device *obd;
181 struct lu_fid rootfid;
182 struct obd_capa *oc = NULL;
183 struct obd_statfs osfs;
184 struct ptlrpc_request *request = NULL;
185 struct lustre_handle dt_conn = {0, };
186 struct lustre_handle md_conn = {0, };
187 struct obd_connect_data *data = NULL;
188 struct lustre_md lmd;
190 int size, err, checksum;
193 obd = class_name2obd(md);
195 CERROR("MD %s: not setup or attached\n", md);
203 if (proc_lustre_fs_root) {
204 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
207 CERROR("could not register mount in /proc/fs/lustre\n");
210 /* indicate the features supported by this client */
211 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
212 OBD_CONNECT_JOIN | OBD_CONNECT_ATTRFID |
213 OBD_CONNECT_VERSION | OBD_CONNECT_MDS_CAPA |
214 OBD_CONNECT_OSS_CAPA | OBD_CONNECT_CANCELSET|
215 OBD_CONNECT_FID | OBD_CONNECT_AT;
217 #ifdef HAVE_LRU_RESIZE_SUPPORT
218 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
219 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
221 #ifdef CONFIG_FS_POSIX_ACL
222 data->ocd_connect_flags |= OBD_CONNECT_ACL;
224 data->ocd_ibits_known = MDS_INODELOCK_FULL;
225 data->ocd_version = LUSTRE_VERSION_CODE;
227 if (sb->s_flags & MS_RDONLY)
228 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
229 if (sbi->ll_flags & LL_SBI_USER_XATTR)
230 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
232 #ifdef HAVE_MS_FLOCK_LOCK
233 /* force vfs to use lustre handler for flock() calls - bug 10743 */
234 sb->s_flags |= MS_FLOCK_LOCK;
237 if (sbi->ll_flags & LL_SBI_FLOCK)
238 sbi->ll_fop = &ll_file_operations_flock;
239 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
240 sbi->ll_fop = &ll_file_operations;
242 sbi->ll_fop = &ll_file_operations_noflock;
245 data->ocd_connect_flags |= OBD_CONNECT_REAL;
246 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
247 data->ocd_connect_flags &= ~OBD_CONNECT_LCL_CLIENT;
248 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT;
250 data->ocd_connect_flags &= ~OBD_CONNECT_RMT_CLIENT;
251 data->ocd_connect_flags |= OBD_CONNECT_LCL_CLIENT;
254 err = obd_connect(NULL, &md_conn, obd, &sbi->ll_sb_uuid, data, NULL);
256 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
257 "recovery, of which this client is not a "
258 "part. Please wait for recovery to complete,"
259 " abort, or time out.\n", md);
262 CERROR("cannot connect to %s: rc = %d\n", md, err);
265 sbi->ll_md_exp = class_conn2export(&md_conn);
267 err = obd_fid_init(sbi->ll_md_exp);
269 CERROR("Can't init metadata layer FID infrastructure, "
274 err = obd_statfs(obd, &osfs, cfs_time_current_64() - HZ, 0);
276 GOTO(out_md_fid, err);
278 size = sizeof(*data);
279 err = obd_get_info(sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
280 KEY_CONN_DATA, &size, data);
282 CERROR("Get connect data failed: %d \n", err);
286 LASSERT(osfs.os_bsize);
287 sb->s_blocksize = osfs.os_bsize;
288 sb->s_blocksize_bits = log2(osfs.os_bsize);
289 sb->s_magic = LL_SUPER_MAGIC;
291 /* for bug 11559. in $LINUX/fs/read_write.c, function do_sendfile():
292 * retval = in_file->f_op->sendfile(...);
294 * retval = -EOVERFLOW;
296 * it will check if *ppos is greater than max. However, max equals to
297 * s_maxbytes, which is a negative integer in a x86_64 box since loff_t
298 * has been defined as a signed long long ineger in linux kernel. */
299 #if BITS_PER_LONG == 64
300 sb->s_maxbytes = PAGE_CACHE_MAXBYTES >> 1;
302 sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
304 sbi->ll_namelen = osfs.os_namelen;
305 sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
307 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
308 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
309 LCONSOLE_INFO("Disabling user_xattr feature because "
310 "it is not supported on the server\n");
311 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
314 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
316 sb->s_flags |= MS_POSIXACL;
318 sbi->ll_flags |= LL_SBI_ACL;
320 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
322 sb->s_flags &= ~MS_POSIXACL;
324 sbi->ll_flags &= ~LL_SBI_ACL;
327 if (data->ocd_connect_flags & OBD_CONNECT_JOIN)
328 sbi->ll_flags |= LL_SBI_JOIN;
330 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
331 if (!(data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT)) {
332 /* sometimes local client claims to be remote, but mdt
333 * will disagree when client gss not applied. */
334 LCONSOLE_INFO("client claims to be remote, but server "
335 "rejected, forced to be local.\n");
336 sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
339 if (!(data->ocd_connect_flags & OBD_CONNECT_LCL_CLIENT)) {
340 /* with gss applied, remote client can not claim to be
341 * local, so mdt maybe force client to be remote. */
342 LCONSOLE_INFO("client claims to be local, but server "
343 "rejected, forced to be remote.\n");
344 sbi->ll_flags |= LL_SBI_RMT_CLIENT;
348 if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
349 LCONSOLE_INFO("client enabled MDS capability!\n");
350 sbi->ll_flags |= LL_SBI_MDS_CAPA;
353 if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
354 LCONSOLE_INFO("client enabled OSS capability!\n");
355 sbi->ll_flags |= LL_SBI_OSS_CAPA;
358 sbi->ll_sdev_orig = sb->s_dev;
359 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0))
360 /* We set sb->s_dev equal on all lustre clients in order to support
361 * NFS export clustering. NFSD requires that the FSID be the same
363 /* s_dev is also used in lt_compare() to compare two fs, but that is
364 * only a node-local comparison. */
366 /* XXX: this will not work with LMV */
367 sb->s_dev = get_uuid2int(sbi2mdc(sbi)->cl_target_uuid.uuid,
368 strlen(sbi2mdc(sbi)->cl_target_uuid.uuid));
371 obd = class_name2obd(dt);
373 CERROR("DT %s: not setup or attached\n", dt);
374 GOTO(out_md_fid, err = -ENODEV);
377 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
378 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
379 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
380 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
382 if (sbi->ll_flags & LL_SBI_OSS_CAPA)
383 data->ocd_connect_flags |= OBD_CONNECT_OSS_CAPA;
385 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
386 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
387 * disabled by default, because it can still be enabled on the
388 * fly via /proc. As a consequence, we still need to come to an
389 * agreement on the supported algorithms at connect time */
390 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
392 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
393 data->ocd_cksum_types = OBD_CKSUM_ADLER;
395 /* send the list of supported checksum types */
396 data->ocd_cksum_types = OBD_CKSUM_ALL;
399 #ifdef HAVE_LRU_RESIZE_SUPPORT
400 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
402 CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
403 "ocd_grant: %d\n", data->ocd_connect_flags,
404 data->ocd_version, data->ocd_grant);
406 obd->obd_upcall.onu_owner = &sbi->ll_lco;
407 obd->obd_upcall.onu_upcall = ll_ocd_update;
408 data->ocd_brw_size = PTLRPC_MAX_BRW_PAGES << CFS_PAGE_SHIFT;
410 err = obd_connect(NULL, &dt_conn, obd, &sbi->ll_sb_uuid, data, NULL);
412 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
413 "recovery, of which this client is not a "
414 "part. Please wait for recovery to "
415 "complete, abort, or time out.\n", dt);
416 GOTO(out_md_fid, err);
418 CERROR("Cannot connect to %s: rc = %d\n", dt, err);
419 GOTO(out_md_fid, err);
422 sbi->ll_dt_exp = class_conn2export(&dt_conn);
424 err = obd_fid_init(sbi->ll_dt_exp);
426 CERROR("Can't init data layer FID infrastructure, "
431 spin_lock(&sbi->ll_lco.lco_lock);
432 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
433 spin_unlock(&sbi->ll_lco.lco_lock);
435 err = obd_register_page_removal_cb(sbi->ll_dt_exp,
439 CERROR("cannot register page removal callback: rc = %d\n",err);
442 err = obd_register_lock_cancel_cb(sbi->ll_dt_exp,
443 ll_extent_lock_cancel_cb);
445 CERROR("cannot register lock cancel callback: rc = %d\n", err);
446 GOTO(out_page_rm_cb, err);
449 err = ll_init_ea_size(sbi->ll_md_exp, sbi->ll_dt_exp);;
451 CERROR("cannot set max EA and cookie sizes: rc = %d\n", err);
452 GOTO(out_lock_cn_cb, err);
455 err = obd_prep_async_page(sbi->ll_dt_exp, NULL, NULL, NULL,
456 0, NULL, NULL, NULL, 0, NULL);
458 LCONSOLE_ERROR_MSG(0x151, "There are no OST's in this "
459 "filesystem. There must be at least one "
460 "active OST for a client to start.\n");
461 GOTO(out_lock_cn_cb, err);
464 if (!ll_async_page_slab) {
465 ll_async_page_slab_size =
466 size_round(sizeof(struct ll_async_page)) + err;
467 ll_async_page_slab = cfs_mem_cache_create("ll_async_page",
468 ll_async_page_slab_size,
470 if (!ll_async_page_slab)
471 GOTO(out_lock_cn_cb, err = -ENOMEM);
474 err = md_getstatus(sbi->ll_md_exp, &rootfid, &oc);
476 CERROR("cannot mds_connect: rc = %d\n", err);
477 GOTO(out_lock_cn_cb, err);
479 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&rootfid));
480 sbi->ll_root_fid = rootfid;
482 sb->s_op = &lustre_super_operations;
483 sb->s_export_op = &lustre_export_operations;
486 * XXX: move this to after cbd setup? */
487 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
488 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
489 valid |= OBD_MD_FLRMTPERM;
490 else if (sbi->ll_flags & LL_SBI_ACL)
491 valid |= OBD_MD_FLACL;
493 err = md_getattr(sbi->ll_md_exp, &rootfid, oc, valid, 0, &request);
497 CERROR("md_getattr failed for root: rc = %d\n", err);
498 GOTO(out_lock_cn_cb, err);
500 memset(&lmd, 0, sizeof(lmd));
501 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
502 sbi->ll_md_exp, &lmd);
504 CERROR("failed to understand root inode md: rc = %d\n", err);
505 ptlrpc_req_finished (request);
506 GOTO(out_lock_cn_cb, err);
509 LASSERT(fid_is_sane(&sbi->ll_root_fid));
510 root = ll_iget(sb, ll_fid_build_ino(sbi, &sbi->ll_root_fid), &lmd);
511 md_free_lustre_md(sbi->ll_md_exp, &lmd);
512 ptlrpc_req_finished(request);
514 if (root == NULL || is_bad_inode(root)) {
516 obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
517 #ifdef CONFIG_FS_POSIX_ACL
519 posix_acl_release(lmd.posix_acl);
520 lmd.posix_acl = NULL;
523 CERROR("lustre_lite: bad iget4 for root\n");
524 GOTO(out_root, err = -EBADF);
527 err = ll_close_thread_start(&sbi->ll_lcq);
529 CERROR("cannot start close thread: rc %d\n", err);
533 #ifdef CONFIG_FS_POSIX_ACL
534 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
535 rct_init(&sbi->ll_rct);
536 et_init(&sbi->ll_et);
540 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
541 err = obd_set_info_async(sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
542 KEY_CHECKSUM, sizeof(checksum), &checksum,
545 sb->s_root = d_alloc_root(root);
547 OBD_FREE(data, sizeof(*data));
548 sb->s_root->d_op = &ll_d_root_ops;
554 obd_unregister_lock_cancel_cb(sbi->ll_dt_exp,
555 ll_extent_lock_cancel_cb);
557 obd_unregister_page_removal_cb(sbi->ll_dt_exp,
559 obd_fid_fini(sbi->ll_dt_exp);
561 obd_disconnect(sbi->ll_dt_exp);
562 sbi->ll_dt_exp = NULL;
564 obd_fid_fini(sbi->ll_md_exp);
566 obd_disconnect(sbi->ll_md_exp);
567 sbi->ll_md_exp = NULL;
571 lprocfs_unregister_mountpoint(sbi);
575 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
579 *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
581 rc = obd_get_info(sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
582 KEY_MAX_EASIZE, &size, lmmsize);
584 CERROR("Get max mdsize error rc %d \n", rc);
589 void ll_dump_inode(struct inode *inode)
591 struct list_head *tmp;
592 int dentry_count = 0;
594 LASSERT(inode != NULL);
596 list_for_each(tmp, &inode->i_dentry)
599 CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
600 inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
601 inode->i_mode, atomic_read(&inode->i_count), dentry_count);
604 void lustre_dump_dentry(struct dentry *dentry, int recur)
606 struct list_head *tmp;
609 LASSERT(dentry != NULL);
611 list_for_each(tmp, &dentry->d_subdirs)
614 CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
615 " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
616 dentry->d_name.len, dentry->d_name.name,
617 dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
618 dentry->d_parent, dentry->d_inode, atomic_read(&dentry->d_count),
619 dentry->d_flags, dentry->d_fsdata, subdirs);
620 if (dentry->d_inode != NULL)
621 ll_dump_inode(dentry->d_inode);
626 list_for_each(tmp, &dentry->d_subdirs) {
627 struct dentry *d = list_entry(tmp, struct dentry, d_child);
628 lustre_dump_dentry(d, recur - 1);
632 #ifdef HAVE_EXPORT___IGET
633 static void prune_dir_dentries(struct inode *inode)
635 struct dentry *dentry, *prev = NULL;
637 /* due to lustre specific logic, a directory
638 * can have few dentries - a bug from VFS POV */
640 spin_lock(&dcache_lock);
641 if (!list_empty(&inode->i_dentry)) {
642 dentry = list_entry(inode->i_dentry.prev,
643 struct dentry, d_alias);
644 /* in order to prevent infinite loops we
645 * break if previous dentry is busy */
646 if (dentry != prev) {
649 spin_unlock(&dcache_lock);
651 /* try to kill all child dentries */
653 shrink_dcache_parent(dentry);
654 unlock_dentry(dentry);
657 /* now try to get rid of current dentry */
658 d_prune_aliases(inode);
662 spin_unlock(&dcache_lock);
665 static void prune_deathrow_one(struct ll_inode_info *lli)
667 struct inode *inode = ll_info2i(lli);
669 /* first, try to drop any dentries - they hold a ref on the inode */
670 if (S_ISDIR(inode->i_mode))
671 prune_dir_dentries(inode);
673 d_prune_aliases(inode);
676 /* if somebody still uses it, leave it */
677 LASSERT(atomic_read(&inode->i_count) > 0);
678 if (atomic_read(&inode->i_count) > 1)
681 CDEBUG(D_INODE, "inode %lu/%u(%d) looks a good candidate for prune\n",
682 inode->i_ino,inode->i_generation, atomic_read(&inode->i_count));
684 /* seems nobody uses it anymore */
692 static void prune_deathrow(struct ll_sb_info *sbi, int try)
694 struct ll_inode_info *lli;
698 if (need_resched() && try)
702 if (!spin_trylock(&sbi->ll_deathrow_lock))
705 spin_lock(&sbi->ll_deathrow_lock);
710 if (!list_empty(&sbi->ll_deathrow)) {
711 lli = list_entry(sbi->ll_deathrow.next,
712 struct ll_inode_info,
714 list_del_init(&lli->lli_dead_list);
715 if (!list_empty(&sbi->ll_deathrow))
718 spin_unlock(&sbi->ll_deathrow_lock);
721 prune_deathrow_one(lli);
723 } while (empty == 0);
725 #else /* !HAVE_EXPORT___IGET */
726 #define prune_deathrow(sbi, try) do {} while (0)
727 #endif /* HAVE_EXPORT___IGET */
729 void client_common_put_super(struct super_block *sb)
731 struct ll_sb_info *sbi = ll_s2sbi(sb);
734 #ifdef CONFIG_FS_POSIX_ACL
735 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
736 et_fini(&sbi->ll_et);
737 rct_fini(&sbi->ll_rct);
741 obd_cancel_unused(sbi->ll_dt_exp, NULL, 0, NULL);
743 ll_close_thread_shutdown(sbi->ll_lcq);
745 /* destroy inodes in deathrow */
746 prune_deathrow(sbi, 0);
748 list_del(&sbi->ll_conn_chain);
750 obd_fid_fini(sbi->ll_dt_exp);
751 obd_disconnect(sbi->ll_dt_exp);
752 sbi->ll_dt_exp = NULL;
754 lprocfs_unregister_mountpoint(sbi);
756 obd_fid_fini(sbi->ll_md_exp);
757 obd_disconnect(sbi->ll_md_exp);
758 sbi->ll_md_exp = NULL;
763 void ll_kill_super(struct super_block *sb)
765 struct ll_sb_info *sbi;
770 if (!(sb->s_flags & MS_ACTIVE))
774 /* we need restore s_dev from changed for clustred NFS before put_super
775 * because new kernels have cached s_dev and change sb->s_dev in
776 * put_super not affected real removing devices */
778 sb->s_dev = sbi->ll_sdev_orig;
782 char *ll_read_opt(const char *opt, char *data)
788 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
789 if (strncmp(opt, data, strlen(opt)))
791 if ((value = strchr(data, '=')) == NULL)
795 OBD_ALLOC(retval, strlen(value) + 1);
797 CERROR("out of memory!\n");
801 memcpy(retval, value, strlen(value)+1);
802 CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
806 static inline int ll_set_opt(const char *opt, char *data, int fl)
808 if (strncmp(opt, data, strlen(opt)) != 0)
814 /* non-client-specific mount options are parsed in lmd_parse */
815 static int ll_options(char *options, int *flags)
818 char *s1 = options, *s2;
824 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
827 CDEBUG(D_SUPER, "next opt=%s\n", s1);
828 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
833 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
838 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
843 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
848 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
853 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
858 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
860 /* Ignore deprecated mount option. The client will
861 * always try to mount with ACL support, whether this
862 * is used depends on whether server supports it. */
865 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
869 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
875 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
880 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
885 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
890 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
896 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
902 s2 = strchr(s1, ',');
910 void ll_lli_init(struct ll_inode_info *lli)
912 lli->lli_inode_magic = LLI_INODE_MAGIC;
913 sema_init(&lli->lli_size_sem, 1);
914 sema_init(&lli->lli_write_sem, 1);
916 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
917 spin_lock_init(&lli->lli_lock);
918 INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
919 INIT_LIST_HEAD(&lli->lli_close_list);
920 lli->lli_inode_magic = LLI_INODE_MAGIC;
921 sema_init(&lli->lli_och_sem, 1);
922 lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
923 lli->lli_mds_exec_och = NULL;
924 lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
925 lli->lli_open_fd_exec_count = 0;
926 INIT_LIST_HEAD(&lli->lli_dead_list);
927 lli->lli_remote_perms = NULL;
928 lli->lli_rmtperm_utime = 0;
929 sema_init(&lli->lli_rmtperm_sem, 1);
930 INIT_LIST_HEAD(&lli->lli_oss_capas);
933 int ll_fill_super(struct super_block *sb)
935 struct lustre_profile *lprof;
936 struct lustre_sb_info *lsi = s2lsi(sb);
937 struct ll_sb_info *sbi;
938 char *dt = NULL, *md = NULL;
939 char *profilenm = get_profile_name(sb);
940 struct config_llog_instance cfg = {0, };
941 char ll_instance[sizeof(sb) * 2 + 1];
945 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
949 /* client additional sb info */
950 lsi->lsi_llsbi = sbi = ll_init_sbi();
956 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
960 /* Generate a string unique to this super, in case some joker tries
961 to mount the same fs at two mount points.
962 Use the address of the super itself.*/
963 sprintf(ll_instance, "%p", sb);
964 cfg.cfg_instance = ll_instance;
965 cfg.cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
967 /* set up client obds */
968 err = lustre_process_log(sb, profilenm, &cfg);
970 CERROR("Unable to process log: %d\n", err);
974 lprof = class_get_profile(profilenm);
976 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
977 " read from the MGS. Does that filesystem "
978 "exist?\n", profilenm);
979 GOTO(out_free, err = -EINVAL);
981 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
982 lprof->lp_md, lprof->lp_dt);
984 OBD_ALLOC(dt, strlen(lprof->lp_dt) +
985 strlen(ll_instance) + 2);
987 GOTO(out_free, err = -ENOMEM);
988 sprintf(dt, "%s-%s", lprof->lp_dt, ll_instance);
990 OBD_ALLOC(md, strlen(lprof->lp_md) +
991 strlen(ll_instance) + 2);
993 GOTO(out_free, err = -ENOMEM);
994 sprintf(md, "%s-%s", lprof->lp_md, ll_instance);
996 /* connections, registrations, sb setup */
997 err = client_common_fill_super(sb, md, dt);
1001 OBD_FREE(md, strlen(md) + 1);
1003 OBD_FREE(dt, strlen(dt) + 1);
1007 LCONSOLE_WARN("Client %s has started\n", profilenm);
1010 } /* ll_fill_super */
1013 void ll_put_super(struct super_block *sb)
1015 struct config_llog_instance cfg;
1016 char ll_instance[sizeof(sb) * 2 + 1];
1017 struct obd_device *obd;
1018 struct lustre_sb_info *lsi = s2lsi(sb);
1019 struct ll_sb_info *sbi = ll_s2sbi(sb);
1020 char *profilenm = get_profile_name(sb);
1021 int force = 1, next;
1024 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1026 ll_print_capa_stat(sbi);
1028 sprintf(ll_instance, "%p", sb);
1029 cfg.cfg_instance = ll_instance;
1030 lustre_end_log(sb, NULL, &cfg);
1032 if (sbi->ll_md_exp) {
1033 obd = class_exp2obd(sbi->ll_md_exp);
1035 force = obd->obd_force;
1038 /* We need to set force before the lov_disconnect in
1039 lustre_common_put_super, since l_d cleans up osc's as well. */
1042 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1044 obd->obd_force = force;
1049 /* Only if client_common_fill_super succeeded */
1050 client_common_put_super(sb);
1053 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1054 class_manual_cleanup(obd);
1058 class_del_profile(profilenm);
1061 lsi->lsi_llsbi = NULL;
1063 lustre_common_put_super(sb);
1065 LCONSOLE_WARN("client %s umount complete\n", ll_instance);
1070 } /* client_put_super */
1072 #ifdef HAVE_REGISTER_CACHE
1073 #include <linux/cache_def.h>
1074 #ifdef HAVE_CACHE_RETURN_INT
1079 ll_shrink_cache(int priority, unsigned int gfp_mask)
1081 struct ll_sb_info *sbi;
1084 list_for_each_entry(sbi, &ll_super_blocks, ll_list)
1085 count += llap_shrink_cache(sbi, priority);
1087 #ifdef HAVE_CACHE_RETURN_INT
1092 struct cache_definition ll_cache_definition = {
1093 .name = "llap_cache",
1094 .shrink = ll_shrink_cache
1096 #endif /* HAVE_REGISTER_CACHE */
1098 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1100 struct inode *inode = NULL;
1101 /* NOTE: we depend on atomic igrab() -bzzz */
1102 lock_res_and_lock(lock);
1103 if (lock->l_ast_data) {
1104 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1105 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1106 inode = igrab(lock->l_ast_data);
1108 inode = lock->l_ast_data;
1109 ldlm_lock_debug(NULL, inode->i_state & I_FREEING ?
1111 lock, __FILE__, __func__, __LINE__,
1112 "l_ast_data %p is bogus: magic %08x",
1113 lock->l_ast_data, lli->lli_inode_magic);
1117 unlock_res_and_lock(lock);
1121 static int null_if_equal(struct ldlm_lock *lock, void *data)
1123 if (data == lock->l_ast_data) {
1124 lock->l_ast_data = NULL;
1126 if (lock->l_req_mode != lock->l_granted_mode)
1127 LDLM_ERROR(lock,"clearing inode with ungranted lock");
1130 return LDLM_ITER_CONTINUE;
1133 void ll_clear_inode(struct inode *inode)
1135 struct ll_inode_info *lli = ll_i2info(inode);
1136 struct ll_sb_info *sbi = ll_i2sbi(inode);
1139 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1140 inode->i_generation, inode);
1142 if (S_ISDIR(inode->i_mode)) {
1143 /* these should have been cleared in ll_file_release */
1144 LASSERT(lli->lli_sai == NULL);
1145 LASSERT(lli->lli_opendir_key == NULL);
1146 LASSERT(lli->lli_opendir_pid == 0);
1149 ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1150 md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1151 null_if_equal, inode);
1153 LASSERT(!lli->lli_open_fd_write_count);
1154 LASSERT(!lli->lli_open_fd_read_count);
1155 LASSERT(!lli->lli_open_fd_exec_count);
1157 if (lli->lli_mds_write_och)
1158 ll_md_real_close(inode, FMODE_WRITE);
1159 if (lli->lli_mds_exec_och)
1160 ll_md_real_close(inode, FMODE_EXEC);
1161 if (lli->lli_mds_read_och)
1162 ll_md_real_close(inode, FMODE_READ);
1165 obd_change_cbdata(sbi->ll_dt_exp, lli->lli_smd,
1166 null_if_equal, inode);
1168 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1169 lli->lli_smd = NULL;
1172 if (lli->lli_symlink_name) {
1173 OBD_FREE(lli->lli_symlink_name,
1174 strlen(lli->lli_symlink_name) + 1);
1175 lli->lli_symlink_name = NULL;
1178 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1179 LASSERT(lli->lli_posix_acl == NULL);
1180 if (lli->lli_remote_perms) {
1181 free_rmtperm_hash(lli->lli_remote_perms);
1182 lli->lli_remote_perms = NULL;
1185 #ifdef CONFIG_FS_POSIX_ACL
1186 else if (lli->lli_posix_acl) {
1187 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1188 LASSERT(lli->lli_remote_perms == NULL);
1189 posix_acl_release(lli->lli_posix_acl);
1190 lli->lli_posix_acl = NULL;
1193 lli->lli_inode_magic = LLI_INODE_DEAD;
1195 #ifdef HAVE_EXPORT___IGET
1196 spin_lock(&sbi->ll_deathrow_lock);
1197 list_del_init(&lli->lli_dead_list);
1198 spin_unlock(&sbi->ll_deathrow_lock);
1200 ll_clear_inode_capas(inode);
1205 int ll_md_setattr(struct inode *inode, struct md_op_data *op_data,
1206 struct md_open_data **mod)
1208 struct lustre_md md;
1209 struct ll_sb_info *sbi = ll_i2sbi(inode);
1210 struct ptlrpc_request *request = NULL;
1214 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1215 LUSTRE_OPC_ANY, NULL);
1216 if (IS_ERR(op_data))
1217 RETURN(PTR_ERR(op_data));
1219 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1222 ptlrpc_req_finished(request);
1223 if (rc == -ENOENT) {
1225 /* Unlinked special device node? Or just a race?
1226 * Pretend we done everything. */
1227 if (!S_ISREG(inode->i_mode) &&
1228 !S_ISDIR(inode->i_mode))
1229 rc = inode_setattr(inode, &op_data->op_attr);
1230 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1231 CERROR("md_setattr fails: rc = %d\n", rc);
1236 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1237 sbi->ll_md_exp, &md);
1239 ptlrpc_req_finished(request);
1243 /* We call inode_setattr to adjust timestamps.
1244 * If there is at least some data in file, we cleared ATTR_SIZE
1245 * above to avoid invoking vmtruncate, otherwise it is important
1246 * to call vmtruncate in inode_setattr to update inode->i_size
1248 rc = inode_setattr(inode, &op_data->op_attr);
1250 /* Extract epoch data if obtained. */
1251 op_data->op_handle = md.body->handle;
1252 op_data->op_ioepoch = md.body->ioepoch;
1254 ll_update_inode(inode, &md);
1255 ptlrpc_req_finished(request);
1260 /* Close IO epoch and send Size-on-MDS attribute update. */
1261 static int ll_setattr_done_writing(struct inode *inode,
1262 struct md_op_data *op_data,
1263 struct md_open_data *mod)
1265 struct ll_inode_info *lli = ll_i2info(inode);
1269 LASSERT(op_data != NULL);
1270 if (!S_ISREG(inode->i_mode))
1273 CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1274 op_data->op_ioepoch, PFID(&lli->lli_fid));
1276 op_data->op_flags = MF_EPOCH_CLOSE | MF_SOM_CHANGE;
1277 rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1278 if (rc == -EAGAIN) {
1279 /* MDS has instructed us to obtain Size-on-MDS attribute
1280 * from OSTs and send setattr to back to MDS. */
1281 rc = ll_sizeonmds_update(inode, mod, &op_data->op_handle,
1282 op_data->op_ioepoch);
1284 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1290 static int ll_setattr_do_truncate(struct inode *inode, loff_t new_size)
1292 struct ll_sb_info *sbi = ll_i2sbi(inode);
1293 struct ll_inode_info *lli = ll_i2info(inode);
1294 struct lov_stripe_md *lsm = lli->lli_smd;
1296 ldlm_policy_data_t policy = { .l_extent = {new_size,
1298 struct lustre_handle lockh = { 0 };
1299 int local_lock = 0; /* 0 - no local lock;
1300 * 1 - lock taken by lock_extent;
1301 * 2 - by obd_match*/
1306 UNLOCK_INODE_MUTEX(inode);
1307 UP_WRITE_I_ALLOC_SEM(inode);
1309 if (sbi->ll_lockless_truncate_enable &&
1310 (sbi->ll_lco.lco_flags & OBD_CONNECT_TRUNCLOCK)) {
1311 ast_flags = LDLM_FL_BLOCK_GRANTED;
1312 rc = obd_match(sbi->ll_dt_exp, lsm, LDLM_EXTENT,
1313 &policy, LCK_PW, &ast_flags, inode, &lockh);
1317 } else if (rc == 0) {
1318 rc = ll_file_punch(inode, new_size, 1);
1321 /* XXX when we fix the AST intents to pass the discard-range
1322 * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
1324 ast_flags = (new_size == 0) ? LDLM_AST_DISCARD_DATA : 0;
1325 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy,
1327 if (likely(rc == 0))
1331 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1332 DOWN_WRITE_I_ALLOC_SEM(inode);
1333 LOCK_INODE_MUTEX(inode);
1335 LOCK_INODE_MUTEX(inode);
1336 DOWN_WRITE_I_ALLOC_SEM(inode);
1338 if (likely(rc == 0)) {
1339 /* Only ll_inode_size_lock is taken at this level.
1340 * lov_stripe_lock() is grabbed by ll_truncate() only over
1341 * call to obd_adjust_kms(). If vmtruncate returns 0, then
1342 * ll_truncate dropped ll_inode_size_lock() */
1343 ll_inode_size_lock(inode, 0);
1345 spin_lock(&lli->lli_lock);
1346 lli->lli_flags |= LLIF_SRVLOCK;
1347 spin_unlock(&lli->lli_lock);
1349 rc = vmtruncate(inode, new_size);
1351 spin_lock(&lli->lli_lock);
1352 lli->lli_flags &= ~LLIF_SRVLOCK;
1353 spin_unlock(&lli->lli_lock);
1356 LASSERT(atomic_read(&lli->lli_size_sem.count) <= 0);
1357 ll_inode_size_unlock(inode, 0);
1362 if (local_lock == 2)
1363 err = obd_cancel(sbi->ll_dt_exp, lsm, LCK_PW, &lockh);
1365 err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
1366 if (unlikely(err != 0)){
1367 CERROR("extent unlock failed: err=%d,"
1368 " unlock method =%d\n", err, local_lock);
1376 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1377 * object(s) determine the file size and mtime. Otherwise, the MDS will
1378 * keep these values until such a time that objects are allocated for it.
1379 * We do the MDS operations first, as it is checking permissions for us.
1380 * We don't to the MDS RPC if there is nothing that we want to store there,
1381 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1382 * going to do an RPC anyways.
1384 * If we are doing a truncate, we will send the mtime and ctime updates
1385 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1386 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1389 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1391 struct ll_inode_info *lli = ll_i2info(inode);
1392 struct lov_stripe_md *lsm = lli->lli_smd;
1393 struct ll_sb_info *sbi = ll_i2sbi(inode);
1394 struct md_op_data *op_data = NULL;
1395 struct md_open_data *mod = NULL;
1396 int ia_valid = attr->ia_valid;
1397 int rc = 0, rc1 = 0;
1400 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1402 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_SETATTR, 1);
1404 if (ia_valid & ATTR_SIZE) {
1405 if (attr->ia_size > ll_file_maxbytes(inode)) {
1406 CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1407 attr->ia_size, ll_file_maxbytes(inode));
1411 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1414 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1415 if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1416 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1420 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1421 if (attr->ia_valid & ATTR_CTIME) {
1422 attr->ia_ctime = CURRENT_TIME;
1423 attr->ia_valid |= ATTR_CTIME_SET;
1425 if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1426 attr->ia_atime = CURRENT_TIME;
1427 attr->ia_valid |= ATTR_ATIME_SET;
1429 if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1430 attr->ia_mtime = CURRENT_TIME;
1431 attr->ia_valid |= ATTR_MTIME_SET;
1433 if ((attr->ia_valid & ATTR_CTIME) && !(attr->ia_valid & ATTR_MTIME)) {
1434 /* To avoid stale mtime on mds, obtain it from ost and send
1436 rc = ll_glimpse_size(inode, 0);
1440 attr->ia_valid |= ATTR_MTIME_SET | ATTR_MTIME;
1441 attr->ia_mtime = inode->i_mtime;
1444 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1445 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1446 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1447 cfs_time_current_sec());
1449 /* NB: ATTR_SIZE will only be set after this point if the size
1450 * resides on the MDS, ie, this file has no objects. */
1452 attr->ia_valid &= ~ATTR_SIZE;
1454 /* We always do an MDS RPC, even if we're only changing the size;
1455 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1457 OBD_ALLOC_PTR(op_data);
1458 if (op_data == NULL)
1461 memcpy(&op_data->op_attr, attr, sizeof(*attr));
1463 /* Open epoch for truncate. */
1464 if ((ll_i2mdexp(inode)->exp_connect_flags & OBD_CONNECT_SOM) &&
1465 (ia_valid & ATTR_SIZE))
1466 op_data->op_flags = MF_EPOCH_OPEN;
1468 rc = ll_md_setattr(inode, op_data, &mod);
1472 if (op_data->op_ioepoch)
1473 CDEBUG(D_INODE, "Epoch "LPU64" opened on "DFID" for "
1474 "truncate\n", op_data->op_ioepoch, PFID(&lli->lli_fid));
1476 if (!lsm || !S_ISREG(inode->i_mode)) {
1477 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1481 /* We really need to get our PW lock before we change inode->i_size.
1482 * If we don't we can race with other i_size updaters on our node, like
1483 * ll_file_read. We can also race with i_size propogation to other
1484 * nodes through dirtying and writeback of final cached pages. This
1485 * last one is especially bad for racing o_append users on other
1487 if (ia_valid & ATTR_SIZE) {
1488 rc = ll_setattr_do_truncate(inode, attr->ia_size);
1489 } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
1491 struct obd_info oinfo = { { { 0 } } };
1494 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1495 inode->i_ino, LTIME_S(attr->ia_mtime));
1499 oa->o_id = lsm->lsm_object_id;
1500 oa->o_gr = lsm->lsm_object_gr;
1501 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1503 flags = OBD_MD_FLTYPE | OBD_MD_FLATIME |
1504 OBD_MD_FLMTIME | OBD_MD_FLCTIME |
1505 OBD_MD_FLFID | OBD_MD_FLGENER |
1508 obdo_from_inode(oa, inode, flags);
1512 oinfo.oi_capa = ll_mdscapa_get(inode);
1514 /* XXX: this looks unnecessary now. */
1515 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1516 capa_put(oinfo.oi_capa);
1518 CERROR("obd_setattr_async fails: rc=%d\n", rc);
1527 if (op_data->op_ioepoch)
1528 rc1 = ll_setattr_done_writing(inode, op_data, mod);
1529 ll_finish_md_op_data(op_data);
1531 return rc ? rc : rc1;
1534 int ll_setattr(struct dentry *de, struct iattr *attr)
1538 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1539 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1540 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1541 if ((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1542 (ATTR_SIZE|ATTR_MODE)) {
1543 mode = de->d_inode->i_mode;
1544 if (((mode & S_ISUID) && (!(attr->ia_mode & S_ISUID))) ||
1545 ((mode & S_ISGID) && (mode & S_IXGRP) &&
1546 (!(attr->ia_mode & S_ISGID))))
1547 attr->ia_valid |= ATTR_FORCE;
1550 return ll_setattr_raw(de->d_inode, attr);
1553 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1554 __u64 max_age, __u32 flags)
1556 struct ll_sb_info *sbi = ll_s2sbi(sb);
1557 struct obd_statfs obd_osfs;
1561 rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age, flags);
1563 CERROR("md_statfs fails: rc = %d\n", rc);
1567 osfs->os_type = sb->s_magic;
1569 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1570 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1572 rc = obd_statfs_rqset(class_exp2obd(sbi->ll_dt_exp),
1573 &obd_osfs, max_age, flags);
1575 CERROR("obd_statfs fails: rc = %d\n", rc);
1579 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1580 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1583 osfs->os_bsize = obd_osfs.os_bsize;
1584 osfs->os_blocks = obd_osfs.os_blocks;
1585 osfs->os_bfree = obd_osfs.os_bfree;
1586 osfs->os_bavail = obd_osfs.os_bavail;
1588 /* If we don't have as many objects free on the OST as inodes
1589 * on the MDS, we reduce the total number of inodes to
1590 * compensate, so that the "inodes in use" number is correct.
1592 if (obd_osfs.os_ffree < osfs->os_ffree) {
1593 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1595 osfs->os_ffree = obd_osfs.os_ffree;
1600 #ifndef HAVE_STATFS_DENTRY_PARAM
1601 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1604 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1606 struct super_block *sb = de->d_sb;
1608 struct obd_statfs osfs;
1611 CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1612 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1614 /* For now we will always get up-to-date statfs values, but in the
1615 * future we may allow some amount of caching on the client (e.g.
1616 * from QOS or lprocfs updates). */
1617 rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - 1, 0);
1621 statfs_unpack(sfs, &osfs);
1623 /* We need to downshift for all 32-bit kernels, because we can't
1624 * tell if the kernel is being called via sys_statfs64() or not.
1625 * Stop before overflowing f_bsize - in which case it is better
1626 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1627 if (sizeof(long) < 8) {
1628 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1631 osfs.os_blocks >>= 1;
1632 osfs.os_bfree >>= 1;
1633 osfs.os_bavail >>= 1;
1637 sfs->f_blocks = osfs.os_blocks;
1638 sfs->f_bfree = osfs.os_bfree;
1639 sfs->f_bavail = osfs.os_bavail;
1644 void ll_inode_size_lock(struct inode *inode, int lock_lsm)
1646 struct ll_inode_info *lli;
1647 struct lov_stripe_md *lsm;
1649 lli = ll_i2info(inode);
1650 LASSERT(lli->lli_size_sem_owner != current);
1651 down(&lli->lli_size_sem);
1652 LASSERT(lli->lli_size_sem_owner == NULL);
1653 lli->lli_size_sem_owner = current;
1655 LASSERTF(lsm != NULL || lock_lsm == 0, "lsm %p, lock_lsm %d\n",
1658 lov_stripe_lock(lsm);
1661 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm)
1663 struct ll_inode_info *lli;
1664 struct lov_stripe_md *lsm;
1666 lli = ll_i2info(inode);
1668 LASSERTF(lsm != NULL || unlock_lsm == 0, "lsm %p, lock_lsm %d\n",
1671 lov_stripe_unlock(lsm);
1672 LASSERT(lli->lli_size_sem_owner == current);
1673 lli->lli_size_sem_owner = NULL;
1674 up(&lli->lli_size_sem);
1677 static void ll_replace_lsm(struct inode *inode, struct lov_stripe_md *lsm)
1679 struct ll_inode_info *lli = ll_i2info(inode);
1681 dump_lsm(D_INODE, lsm);
1682 dump_lsm(D_INODE, lli->lli_smd);
1683 LASSERTF(lsm->lsm_magic == LOV_MAGIC_JOIN,
1684 "lsm must be joined lsm %p\n", lsm);
1685 obd_free_memmd(ll_i2dtexp(inode), &lli->lli_smd);
1686 CDEBUG(D_INODE, "replace lsm %p to lli_smd %p for inode %lu%u(%p)\n",
1687 lsm, lli->lli_smd, inode->i_ino, inode->i_generation, inode);
1689 lli->lli_maxbytes = lsm->lsm_maxbytes;
1690 if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1691 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1694 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1696 struct ll_inode_info *lli = ll_i2info(inode);
1697 struct mdt_body *body = md->body;
1698 struct lov_stripe_md *lsm = md->lsm;
1699 struct ll_sb_info *sbi = ll_i2sbi(inode);
1701 LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1703 if (lli->lli_smd == NULL) {
1704 if (lsm->lsm_magic != LOV_MAGIC &&
1705 lsm->lsm_magic != LOV_MAGIC_JOIN) {
1706 dump_lsm(D_ERROR, lsm);
1709 CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1710 lsm, inode->i_ino, inode->i_generation, inode);
1711 /* ll_inode_size_lock() requires it is only called
1712 * with lli_smd != NULL or lock_lsm == 0 or we can
1713 * race between lock/unlock. bug 9547 */
1715 lli->lli_maxbytes = lsm->lsm_maxbytes;
1716 if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1717 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1719 if (lli->lli_smd->lsm_magic == lsm->lsm_magic &&
1720 lli->lli_smd->lsm_stripe_count ==
1721 lsm->lsm_stripe_count) {
1722 if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
1723 CERROR("lsm mismatch for inode %ld\n",
1725 CERROR("lli_smd:\n");
1726 dump_lsm(D_ERROR, lli->lli_smd);
1728 dump_lsm(D_ERROR, lsm);
1732 ll_replace_lsm(inode, lsm);
1734 if (lli->lli_smd != lsm)
1735 obd_free_memmd(ll_i2dtexp(inode), &lsm);
1738 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1739 if (body->valid & OBD_MD_FLRMTPERM)
1740 ll_update_remote_perm(inode, md->remote_perm);
1742 #ifdef CONFIG_FS_POSIX_ACL
1743 else if (body->valid & OBD_MD_FLACL) {
1744 spin_lock(&lli->lli_lock);
1745 if (lli->lli_posix_acl)
1746 posix_acl_release(lli->lli_posix_acl);
1747 lli->lli_posix_acl = md->posix_acl;
1748 spin_unlock(&lli->lli_lock);
1751 inode->i_ino = ll_fid_build_ino(sbi, &body->fid1);
1753 if (body->valid & OBD_MD_FLATIME &&
1754 body->atime > LTIME_S(inode->i_atime))
1755 LTIME_S(inode->i_atime) = body->atime;
1757 /* mtime is always updated with ctime, but can be set in past.
1758 As write and utime(2) may happen within 1 second, and utime's
1759 mtime has a priority over write's one, so take mtime from mds
1760 for the same ctimes. */
1761 if (body->valid & OBD_MD_FLCTIME &&
1762 body->ctime >= LTIME_S(inode->i_ctime)) {
1763 LTIME_S(inode->i_ctime) = body->ctime;
1764 if (body->valid & OBD_MD_FLMTIME) {
1765 CDEBUG(D_INODE, "setting ino %lu mtime "
1766 "from %lu to "LPU64"\n", inode->i_ino,
1767 LTIME_S(inode->i_mtime), body->mtime);
1768 LTIME_S(inode->i_mtime) = body->mtime;
1771 if (body->valid & OBD_MD_FLMODE)
1772 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1773 if (body->valid & OBD_MD_FLTYPE)
1774 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1775 if (S_ISREG(inode->i_mode)) {
1776 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1778 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1780 #ifdef HAVE_INODE_BLKSIZE
1781 inode->i_blksize = 1<<inode->i_blkbits;
1783 if (body->valid & OBD_MD_FLUID)
1784 inode->i_uid = body->uid;
1785 if (body->valid & OBD_MD_FLGID)
1786 inode->i_gid = body->gid;
1787 if (body->valid & OBD_MD_FLFLAGS)
1788 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1789 if (body->valid & OBD_MD_FLNLINK)
1790 inode->i_nlink = body->nlink;
1791 if (body->valid & OBD_MD_FLRDEV)
1792 inode->i_rdev = old_decode_dev(body->rdev);
1794 if (body->valid & OBD_MD_FLID) {
1795 /* FID shouldn't be changed! */
1796 if (fid_is_sane(&lli->lli_fid)) {
1797 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1798 "Trying to change FID "DFID
1799 " to the "DFID", inode %lu/%u(%p)\n",
1800 PFID(&lli->lli_fid), PFID(&body->fid1),
1801 inode->i_ino, inode->i_generation, inode);
1803 lli->lli_fid = body->fid1;
1806 LASSERT(fid_seq(&lli->lli_fid) != 0);
1808 if (body->valid & OBD_MD_FLSIZE) {
1809 if ((ll_i2mdexp(inode)->exp_connect_flags & OBD_CONNECT_SOM) &&
1810 S_ISREG(inode->i_mode) && lli->lli_smd) {
1811 struct lustre_handle lockh;
1814 /* As it is possible a blocking ast has been processed
1815 * by this time, we need to check there is an UPDATE
1816 * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1818 mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1821 if (lli->lli_flags & (LLIF_DONE_WRITING |
1822 LLIF_EPOCH_PENDING |
1824 CERROR("ino %lu flags %lu still has "
1825 "size authority! do not trust "
1826 "the size got from MDS\n",
1827 inode->i_ino, lli->lli_flags);
1829 /* Use old size assignment to avoid
1830 * deadlock bz14138 & bz14326 */
1831 inode->i_size = body->size;
1832 lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1834 ldlm_lock_decref(&lockh, mode);
1837 /* Use old size assignment to avoid
1838 * deadlock bz14138 & bz14326 */
1839 inode->i_size = body->size;
1842 if (body->valid & OBD_MD_FLBLOCKS)
1843 inode->i_blocks = body->blocks;
1846 if (body->valid & OBD_MD_FLMDSCAPA) {
1847 LASSERT(md->mds_capa);
1848 ll_add_capa(inode, md->mds_capa);
1850 if (body->valid & OBD_MD_FLOSSCAPA) {
1851 LASSERT(md->oss_capa);
1852 ll_add_capa(inode, md->oss_capa);
1856 static struct backing_dev_info ll_backing_dev_info = {
1857 .ra_pages = 0, /* No readahead */
1858 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,12))
1859 .capabilities = 0, /* Does contribute to dirty memory */
1861 .memory_backed = 0, /* Does contribute to dirty memory */
1865 void ll_read_inode2(struct inode *inode, void *opaque)
1867 struct lustre_md *md = opaque;
1868 struct ll_inode_info *lli = ll_i2info(inode);
1871 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n",
1872 inode->i_ino, inode->i_generation, inode);
1876 LASSERT(!lli->lli_smd);
1878 /* Core attributes from the MDS first. This is a new inode, and
1879 * the VFS doesn't zero times in the core inode so we have to do
1880 * it ourselves. They will be overwritten by either MDS or OST
1881 * attributes - we just need to make sure they aren't newer. */
1882 LTIME_S(inode->i_mtime) = 0;
1883 LTIME_S(inode->i_atime) = 0;
1884 LTIME_S(inode->i_ctime) = 0;
1886 ll_update_inode(inode, md);
1888 /* OIDEBUG(inode); */
1890 if (S_ISREG(inode->i_mode)) {
1891 struct ll_sb_info *sbi = ll_i2sbi(inode);
1892 inode->i_op = &ll_file_inode_operations;
1893 inode->i_fop = sbi->ll_fop;
1894 inode->i_mapping->a_ops = &ll_aops;
1896 } else if (S_ISDIR(inode->i_mode)) {
1897 inode->i_op = &ll_dir_inode_operations;
1898 inode->i_fop = &ll_dir_operations;
1899 inode->i_mapping->a_ops = &ll_dir_aops;
1901 } else if (S_ISLNK(inode->i_mode)) {
1902 inode->i_op = &ll_fast_symlink_inode_operations;
1905 inode->i_op = &ll_special_inode_operations;
1907 init_special_inode(inode, inode->i_mode,
1908 kdev_t_to_nr(inode->i_rdev));
1910 /* initializing backing dev info. */
1911 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
1917 void ll_delete_inode(struct inode *inode)
1919 struct ll_sb_info *sbi = ll_i2sbi(inode);
1923 rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
1925 CERROR("fid_delete() failed, rc %d\n", rc);
1927 truncate_inode_pages(&inode->i_data, 0);
1933 int ll_iocontrol(struct inode *inode, struct file *file,
1934 unsigned int cmd, unsigned long arg)
1936 struct ll_sb_info *sbi = ll_i2sbi(inode);
1937 struct ptlrpc_request *req = NULL;
1942 case EXT3_IOC_GETFLAGS: {
1943 struct mdt_body *body;
1944 struct obd_capa *oc;
1946 oc = ll_mdscapa_get(inode);
1947 rc = md_getattr(sbi->ll_md_exp, ll_inode2fid(inode), oc,
1948 OBD_MD_FLFLAGS, 0, &req);
1951 CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1955 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1957 flags = body->flags;
1959 ptlrpc_req_finished(req);
1961 RETURN(put_user(flags, (int *)arg));
1963 case EXT3_IOC_SETFLAGS: {
1964 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1965 struct obd_info oinfo = { { { 0 } } };
1966 struct md_op_data *op_data;
1968 if (get_user(flags, (int *)arg))
1972 OBDO_ALLOC(oinfo.oi_oa);
1976 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1977 LUSTRE_OPC_ANY, NULL);
1978 if (IS_ERR(op_data))
1979 RETURN(PTR_ERR(op_data));
1981 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1982 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1983 rc = md_setattr(sbi->ll_md_exp, op_data,
1984 NULL, 0, NULL, 0, &req, NULL);
1985 ll_finish_md_op_data(op_data);
1986 ptlrpc_req_finished(req);
1987 if (rc || lsm == NULL) {
1988 OBDO_FREE(oinfo.oi_oa);
1992 oinfo.oi_oa->o_id = lsm->lsm_object_id;
1993 oinfo.oi_oa->o_gr = lsm->lsm_object_gr;
1994 oinfo.oi_oa->o_flags = flags;
1995 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1997 oinfo.oi_capa = ll_mdscapa_get(inode);
1999 obdo_from_inode(oinfo.oi_oa, inode,
2000 OBD_MD_FLFID | OBD_MD_FLGENER);
2001 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
2002 capa_put(oinfo.oi_capa);
2003 OBDO_FREE(oinfo.oi_oa);
2005 if (rc != -EPERM && rc != -EACCES)
2006 CERROR("md_setattr_async fails: rc = %d\n", rc);
2010 inode->i_flags = ll_ext_to_inode_flags(flags |
2011 MDS_BFLAG_EXT_FLAGS);
2021 int ll_flush_ctx(struct inode *inode)
2023 struct ll_sb_info *sbi = ll_i2sbi(inode);
2025 CDEBUG(D_SEC, "flush context for user %d\n", current->uid);
2027 obd_set_info_async(sbi->ll_md_exp,
2028 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2030 obd_set_info_async(sbi->ll_dt_exp,
2031 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2036 /* umount -f client means force down, don't save state */
2037 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2038 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
2040 struct super_block *sb = vfsmnt->mnt_sb;
2042 void ll_umount_begin(struct super_block *sb)
2045 struct lustre_sb_info *lsi = s2lsi(sb);
2046 struct ll_sb_info *sbi = ll_s2sbi(sb);
2047 struct obd_device *obd;
2048 struct obd_ioctl_data ioc_data = { 0 };
2051 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2052 if (!(flags & MNT_FORCE)) {
2058 /* Tell the MGC we got umount -f */
2059 lsi->lsi_flags |= LSI_UMOUNT_FORCE;
2061 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2062 sb->s_count, atomic_read(&sb->s_active));
2064 obd = class_exp2obd(sbi->ll_md_exp);
2066 CERROR("Invalid MDC connection handle "LPX64"\n",
2067 sbi->ll_md_exp->exp_handle.h_cookie);
2072 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp, sizeof ioc_data,
2075 obd = class_exp2obd(sbi->ll_dt_exp);
2077 CERROR("Invalid LOV connection handle "LPX64"\n",
2078 sbi->ll_dt_exp->exp_handle.h_cookie);
2084 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp, sizeof ioc_data,
2087 /* Really, we'd like to wait until there are no requests outstanding,
2088 * and then continue. For now, we just invalidate the requests,
2089 * schedule, and hope.
2096 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2098 struct ll_sb_info *sbi = ll_s2sbi(sb);
2102 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2103 read_only = *flags & MS_RDONLY;
2104 err = obd_set_info_async(sbi->ll_md_exp,
2105 sizeof(KEY_READ_ONLY),
2106 KEY_READ_ONLY, sizeof(read_only),
2109 CERROR("Failed to change the read-only flag during "
2110 "remount: %d\n", err);
2115 sb->s_flags |= MS_RDONLY;
2117 sb->s_flags &= ~MS_RDONLY;
2122 int ll_prep_inode(struct inode **inode,
2123 struct ptlrpc_request *req,
2124 struct super_block *sb)
2126 struct ll_sb_info *sbi = NULL;
2127 struct lustre_md md;
2131 LASSERT(*inode || sb);
2132 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2133 prune_deathrow(sbi, 1);
2134 memset(&md, 0, sizeof(struct lustre_md));
2136 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2137 sbi->ll_md_exp, &md);
2142 ll_update_inode(*inode, &md);
2144 LASSERT(sb != NULL);
2147 * At this point server returns to client's same fid as client
2148 * generated for creating. So using ->fid1 is okay here.
2150 LASSERT(fid_is_sane(&md.body->fid1));
2152 *inode = ll_iget(sb, ll_fid_build_ino(sbi, &md.body->fid1), &md);
2153 if (*inode == NULL || is_bad_inode(*inode)) {
2155 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2156 #ifdef CONFIG_FS_POSIX_ACL
2158 posix_acl_release(md.posix_acl);
2159 md.posix_acl = NULL;
2163 CERROR("new_inode -fatal: rc %d\n", rc);
2168 rc = obd_checkmd(sbi->ll_dt_exp, sbi->ll_md_exp,
2169 ll_i2info(*inode)->lli_smd);
2171 md_free_lustre_md(sbi->ll_md_exp, &md);
2175 char *llap_origins[] = {
2176 [LLAP_ORIGIN_UNKNOWN] = "--",
2177 [LLAP_ORIGIN_READPAGE] = "rp",
2178 [LLAP_ORIGIN_READAHEAD] = "ra",
2179 [LLAP_ORIGIN_COMMIT_WRITE] = "cw",
2180 [LLAP_ORIGIN_WRITEPAGE] = "wp",
2181 [LLAP_ORIGIN_LOCKLESS_IO] = "ls"
2184 struct ll_async_page *llite_pglist_next_llap(struct ll_sb_info *sbi,
2185 struct list_head *list)
2187 struct ll_async_page *llap;
2188 struct list_head *pos;
2190 list_for_each(pos, list) {
2191 if (pos == &sbi->ll_pglist)
2193 llap = list_entry(pos, struct ll_async_page, llap_pglist_item);
2194 if (llap->llap_page == NULL)
2202 int ll_obd_statfs(struct inode *inode, void *arg)
2204 struct ll_sb_info *sbi = NULL;
2205 struct obd_export *exp;
2207 struct obd_ioctl_data *data = NULL;
2211 if (!inode || !(sbi = ll_i2sbi(inode)))
2212 GOTO(out_statfs, rc = -EINVAL);
2214 rc = obd_ioctl_getdata(&buf, &len, arg);
2216 GOTO(out_statfs, rc);
2219 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2220 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2221 GOTO(out_statfs, rc = -EINVAL);
2223 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2224 if (type == LL_STATFS_MDC)
2225 exp = sbi->ll_md_exp;
2226 else if (type == LL_STATFS_LOV)
2227 exp = sbi->ll_dt_exp;
2229 GOTO(out_statfs, rc = -ENODEV);
2231 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2233 GOTO(out_statfs, rc);
2236 obd_ioctl_freedata(buf, len);
2240 int ll_process_config(struct lustre_cfg *lcfg)
2244 struct lprocfs_static_vars lvars;
2248 lprocfs_llite_init_vars(&lvars);
2250 /* The instance name contains the sb: lustre-client-aacfe000 */
2251 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2252 if (!ptr || !*(++ptr))
2254 if (sscanf(ptr, "%lx", &x) != 1)
2257 /* This better be a real Lustre superblock! */
2258 LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2260 /* Note we have not called client_common_fill_super yet, so
2261 proc fns must be able to handle that! */
2262 rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2267 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2268 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2269 struct inode *i1, struct inode *i2,
2270 const char *name, int namelen,
2271 int mode, __u32 opc, void *data)
2273 LASSERT(i1 != NULL);
2275 if (namelen > ll_i2sbi(i1)->ll_namelen)
2276 return ERR_PTR(-ENAMETOOLONG);
2278 if (op_data == NULL)
2279 OBD_ALLOC_PTR(op_data);
2281 if (op_data == NULL)
2282 return ERR_PTR(-ENOMEM);
2284 ll_i2gids(op_data->op_suppgids, i1, i2);
2285 op_data->op_fid1 = *ll_inode2fid(i1);
2286 op_data->op_capa1 = ll_mdscapa_get(i1);
2289 op_data->op_fid2 = *ll_inode2fid(i2);
2290 op_data->op_capa2 = ll_mdscapa_get(i2);
2292 fid_zero(&op_data->op_fid2);
2293 op_data->op_capa2 = NULL;
2296 op_data->op_name = name;
2297 op_data->op_namelen = namelen;
2298 op_data->op_mode = mode;
2299 op_data->op_mod_time = cfs_time_current_sec();
2300 op_data->op_fsuid = current->fsuid;
2301 op_data->op_fsgid = current->fsgid;
2302 op_data->op_cap = current->cap_effective;
2303 op_data->op_bias = MDS_CHECK_SPLIT;
2304 op_data->op_opc = opc;
2305 op_data->op_mds = 0;
2306 op_data->op_data = data;
2311 void ll_finish_md_op_data(struct md_op_data *op_data)
2313 capa_put(op_data->op_capa1);
2314 capa_put(op_data->op_capa2);
2315 OBD_FREE_PTR(op_data);