1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see [sun.com URL with a
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright 2008 Sun Microsystems, Inc. All rights reserved
30 * Use is subject to license terms.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/llite/llite_lib.c
38 * Lustre Light Super operations
41 #define DEBUG_SUBSYSTEM S_LLITE
43 #include <linux/module.h>
44 #include <linux/types.h>
45 #include <linux/random.h>
46 #include <linux/version.h>
49 #include <lustre_lite.h>
50 #include <lustre_ha.h>
51 #include <lustre_dlm.h>
52 #include <lprocfs_status.h>
53 #include <lustre_disk.h>
54 #include <lustre_param.h>
55 #include <lustre_log.h>
56 #include <obd_cksum.h>
57 #include <lustre_cache.h>
58 #include "llite_internal.h"
60 cfs_mem_cache_t *ll_file_data_slab;
62 LIST_HEAD(ll_super_blocks);
63 spinlock_t ll_sb_lock = SPIN_LOCK_UNLOCKED;
65 extern struct address_space_operations ll_aops;
66 extern struct address_space_operations ll_dir_aops;
69 #define log2(n) ffz(~(n))
73 static struct ll_sb_info *ll_init_sbi(void)
75 struct ll_sb_info *sbi = NULL;
82 OBD_ALLOC(sbi, sizeof(*sbi));
86 spin_lock_init(&sbi->ll_lock);
87 spin_lock_init(&sbi->ll_lco.lco_lock);
88 spin_lock_init(&sbi->ll_pp_extent_lock);
89 spin_lock_init(&sbi->ll_process_lock);
90 sbi->ll_rw_stats_on = 0;
91 INIT_LIST_HEAD(&sbi->ll_pglist);
94 pages = si.totalram - si.totalhigh;
95 if (pages >> (20 - CFS_PAGE_SHIFT) < 512)
96 sbi->ll_async_page_max = pages / 2;
98 sbi->ll_async_page_max = (pages / 4) * 3;
99 sbi->ll_ra_info.ra_max_pages = min(num_physpages / 8,
100 SBI_DEFAULT_READAHEAD_MAX);
101 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
102 SBI_DEFAULT_READAHEAD_WHOLE_MAX;
103 sbi->ll_contention_time = SBI_DEFAULT_CONTENTION_SECONDS;
104 sbi->ll_lockless_truncate_enable = SBI_DEFAULT_LOCKLESS_TRUNCATE_ENABLE;
105 INIT_LIST_HEAD(&sbi->ll_conn_chain);
106 INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
108 ll_generate_random_uuid(uuid);
109 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
110 CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
112 spin_lock(&ll_sb_lock);
113 list_add_tail(&sbi->ll_list, &ll_super_blocks);
114 spin_unlock(&ll_sb_lock);
116 #ifdef ENABLE_LLITE_CHECKSUM
117 sbi->ll_flags |= LL_SBI_CHECKSUM;
120 #ifdef HAVE_LRU_RESIZE_SUPPORT
121 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
124 #ifdef HAVE_EXPORT___IGET
125 INIT_LIST_HEAD(&sbi->ll_deathrow);
126 spin_lock_init(&sbi->ll_deathrow_lock);
128 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
129 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].pp_r_hist.oh_lock);
130 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].pp_w_hist.oh_lock);
133 /* metadata statahead is enabled by default */
134 sbi->ll_sa_max = LL_SA_RPC_DEF;
139 void ll_free_sbi(struct super_block *sb)
141 struct ll_sb_info *sbi = ll_s2sbi(sb);
145 spin_lock(&ll_sb_lock);
146 list_del(&sbi->ll_list);
147 spin_unlock(&ll_sb_lock);
148 OBD_FREE(sbi, sizeof(*sbi));
153 static struct dentry_operations ll_d_root_ops = {
154 #ifdef DCACHE_LUSTRE_INVALID
155 .d_compare = ll_dcompare,
159 /* Initialize the default and maximum LOV EA and cookie sizes. This allows
160 * us to make MDS RPCs with large enough reply buffers to hold the
161 * maximum-sized (= maximum striped) EA and cookie without having to
162 * calculate this (via a call into the LOV + OSCs) each time we make an RPC. */
163 static int ll_init_ea_size(struct obd_export *md_exp, struct obd_export *dt_exp)
165 struct lov_stripe_md lsm = { .lsm_magic = LOV_MAGIC };
166 __u32 valsize = sizeof(struct lov_desc);
167 int rc, easize, def_easize, cookiesize;
168 struct lov_desc desc;
172 rc = obd_get_info(dt_exp, sizeof(KEY_LOVDESC), KEY_LOVDESC,
177 stripes = min(desc.ld_tgt_count, (__u32)LOV_MAX_STRIPE_COUNT);
178 lsm.lsm_stripe_count = stripes;
179 easize = obd_size_diskmd(dt_exp, &lsm);
181 lsm.lsm_stripe_count = desc.ld_default_stripe_count;
182 def_easize = obd_size_diskmd(dt_exp, &lsm);
184 cookiesize = stripes * sizeof(struct llog_cookie);
186 CDEBUG(D_HA, "updating max_mdsize/max_cookiesize: %d/%d\n",
189 rc = md_init_ea_size(md_exp, easize, def_easize, cookiesize);
193 static int client_common_fill_super(struct super_block *sb, char *md, char *dt)
195 struct inode *root = 0;
196 struct ll_sb_info *sbi = ll_s2sbi(sb);
197 struct obd_device *obd;
198 struct lu_fid rootfid;
199 struct obd_capa *oc = NULL;
200 struct obd_statfs osfs;
201 struct ptlrpc_request *request = NULL;
202 struct lustre_handle dt_conn = {0, };
203 struct lustre_handle md_conn = {0, };
204 struct obd_connect_data *data = NULL;
205 struct lustre_md lmd;
207 int size, err, checksum;
210 obd = class_name2obd(md);
212 CERROR("MD %s: not setup or attached\n", md);
220 if (proc_lustre_fs_root) {
221 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
224 CERROR("could not register mount in /proc/fs/lustre\n");
227 /* indicate the features supported by this client */
228 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
229 OBD_CONNECT_JOIN | OBD_CONNECT_ATTRFID |
230 OBD_CONNECT_VERSION | OBD_CONNECT_MDS_CAPA |
231 OBD_CONNECT_OSS_CAPA | OBD_CONNECT_CANCELSET|
232 OBD_CONNECT_FID | OBD_CONNECT_AT;
234 #ifdef HAVE_LRU_RESIZE_SUPPORT
235 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
236 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
238 #ifdef CONFIG_FS_POSIX_ACL
239 data->ocd_connect_flags |= OBD_CONNECT_ACL;
241 data->ocd_ibits_known = MDS_INODELOCK_FULL;
242 data->ocd_version = LUSTRE_VERSION_CODE;
244 if (sb->s_flags & MS_RDONLY)
245 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
246 if (sbi->ll_flags & LL_SBI_USER_XATTR)
247 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
249 #ifdef HAVE_MS_FLOCK_LOCK
250 /* force vfs to use lustre handler for flock() calls - bug 10743 */
251 sb->s_flags |= MS_FLOCK_LOCK;
254 if (sbi->ll_flags & LL_SBI_FLOCK)
255 sbi->ll_fop = &ll_file_operations_flock;
256 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
257 sbi->ll_fop = &ll_file_operations;
259 sbi->ll_fop = &ll_file_operations_noflock;
262 data->ocd_connect_flags |= OBD_CONNECT_REAL;
263 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
264 data->ocd_connect_flags &= ~OBD_CONNECT_LCL_CLIENT;
265 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT;
267 data->ocd_connect_flags &= ~OBD_CONNECT_RMT_CLIENT;
268 data->ocd_connect_flags |= OBD_CONNECT_LCL_CLIENT;
271 err = obd_connect(NULL, &md_conn, obd, &sbi->ll_sb_uuid, data, NULL);
273 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
274 "recovery, of which this client is not a "
275 "part. Please wait for recovery to complete,"
276 " abort, or time out.\n", md);
279 CERROR("cannot connect to %s: rc = %d\n", md, err);
282 sbi->ll_md_exp = class_conn2export(&md_conn);
284 err = obd_fid_init(sbi->ll_md_exp);
286 CERROR("Can't init metadata layer FID infrastructure, "
291 err = obd_statfs(obd, &osfs, cfs_time_current_64() - HZ, 0);
293 GOTO(out_md_fid, err);
295 size = sizeof(*data);
296 err = obd_get_info(sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
297 KEY_CONN_DATA, &size, data);
299 CERROR("Get connect data failed: %d \n", err);
303 LASSERT(osfs.os_bsize);
304 sb->s_blocksize = osfs.os_bsize;
305 sb->s_blocksize_bits = log2(osfs.os_bsize);
306 sb->s_magic = LL_SUPER_MAGIC;
308 /* for bug 11559. in $LINUX/fs/read_write.c, function do_sendfile():
309 * retval = in_file->f_op->sendfile(...);
311 * retval = -EOVERFLOW;
313 * it will check if *ppos is greater than max. However, max equals to
314 * s_maxbytes, which is a negative integer in a x86_64 box since loff_t
315 * has been defined as a signed long long ineger in linux kernel. */
316 #if BITS_PER_LONG == 64
317 sb->s_maxbytes = PAGE_CACHE_MAXBYTES >> 1;
319 sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
321 sbi->ll_namelen = osfs.os_namelen;
322 sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
324 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
325 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
326 LCONSOLE_INFO("Disabling user_xattr feature because "
327 "it is not supported on the server\n");
328 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
331 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
333 sb->s_flags |= MS_POSIXACL;
335 sbi->ll_flags |= LL_SBI_ACL;
337 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
339 sb->s_flags &= ~MS_POSIXACL;
341 sbi->ll_flags &= ~LL_SBI_ACL;
344 if (data->ocd_connect_flags & OBD_CONNECT_JOIN)
345 sbi->ll_flags |= LL_SBI_JOIN;
347 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
348 if (!(data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT)) {
349 /* sometimes local client claims to be remote, but mdt
350 * will disagree when client gss not applied. */
351 LCONSOLE_INFO("client claims to be remote, but server "
352 "rejected, forced to be local.\n");
353 sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
356 if (!(data->ocd_connect_flags & OBD_CONNECT_LCL_CLIENT)) {
357 /* with gss applied, remote client can not claim to be
358 * local, so mdt maybe force client to be remote. */
359 LCONSOLE_INFO("client claims to be local, but server "
360 "rejected, forced to be remote.\n");
361 sbi->ll_flags |= LL_SBI_RMT_CLIENT;
365 if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
366 LCONSOLE_INFO("client enabled MDS capability!\n");
367 sbi->ll_flags |= LL_SBI_MDS_CAPA;
370 if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
371 LCONSOLE_INFO("client enabled OSS capability!\n");
372 sbi->ll_flags |= LL_SBI_OSS_CAPA;
375 sbi->ll_sdev_orig = sb->s_dev;
376 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0))
377 /* We set sb->s_dev equal on all lustre clients in order to support
378 * NFS export clustering. NFSD requires that the FSID be the same
380 /* s_dev is also used in lt_compare() to compare two fs, but that is
381 * only a node-local comparison. */
383 /* XXX: this will not work with LMV */
384 sb->s_dev = get_uuid2int(sbi2mdc(sbi)->cl_target_uuid.uuid,
385 strlen(sbi2mdc(sbi)->cl_target_uuid.uuid));
388 obd = class_name2obd(dt);
390 CERROR("DT %s: not setup or attached\n", dt);
391 GOTO(out_md_fid, err = -ENODEV);
394 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
395 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
396 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
397 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
399 if (sbi->ll_flags & LL_SBI_OSS_CAPA)
400 data->ocd_connect_flags |= OBD_CONNECT_OSS_CAPA;
402 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
403 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
404 * disabled by default, because it can still be enabled on the
405 * fly via /proc. As a consequence, we still need to come to an
406 * agreement on the supported algorithms at connect time */
407 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
409 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
410 data->ocd_cksum_types = OBD_CKSUM_ADLER;
412 /* send the list of supported checksum types */
413 data->ocd_cksum_types = OBD_CKSUM_ALL;
416 #ifdef HAVE_LRU_RESIZE_SUPPORT
417 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
419 CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
420 "ocd_grant: %d\n", data->ocd_connect_flags,
421 data->ocd_version, data->ocd_grant);
423 obd->obd_upcall.onu_owner = &sbi->ll_lco;
424 obd->obd_upcall.onu_upcall = ll_ocd_update;
425 data->ocd_brw_size = PTLRPC_MAX_BRW_PAGES << CFS_PAGE_SHIFT;
427 err = obd_connect(NULL, &dt_conn, obd, &sbi->ll_sb_uuid, data, NULL);
429 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
430 "recovery, of which this client is not a "
431 "part. Please wait for recovery to "
432 "complete, abort, or time out.\n", dt);
433 GOTO(out_md_fid, err);
435 CERROR("Cannot connect to %s: rc = %d\n", dt, err);
436 GOTO(out_md_fid, err);
439 sbi->ll_dt_exp = class_conn2export(&dt_conn);
441 err = obd_fid_init(sbi->ll_dt_exp);
443 CERROR("Can't init data layer FID infrastructure, "
448 spin_lock(&sbi->ll_lco.lco_lock);
449 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
450 spin_unlock(&sbi->ll_lco.lco_lock);
452 err = obd_register_page_removal_cb(sbi->ll_dt_exp,
456 CERROR("cannot register page removal callback: rc = %d\n",err);
459 err = obd_register_lock_cancel_cb(sbi->ll_dt_exp,
460 ll_extent_lock_cancel_cb);
462 CERROR("cannot register lock cancel callback: rc = %d\n", err);
463 GOTO(out_page_rm_cb, err);
466 err = ll_init_ea_size(sbi->ll_md_exp, sbi->ll_dt_exp);;
468 CERROR("cannot set max EA and cookie sizes: rc = %d\n", err);
469 GOTO(out_lock_cn_cb, err);
472 err = obd_prep_async_page(sbi->ll_dt_exp, NULL, NULL, NULL,
473 0, NULL, NULL, NULL, 0, NULL);
475 LCONSOLE_ERROR_MSG(0x151, "There are no OST's in this "
476 "filesystem. There must be at least one "
477 "active OST for a client to start.\n");
478 GOTO(out_lock_cn_cb, err);
481 if (!ll_async_page_slab) {
482 ll_async_page_slab_size =
483 size_round(sizeof(struct ll_async_page)) + err;
484 ll_async_page_slab = cfs_mem_cache_create("ll_async_page",
485 ll_async_page_slab_size,
487 if (!ll_async_page_slab)
488 GOTO(out_lock_cn_cb, err = -ENOMEM);
491 err = md_getstatus(sbi->ll_md_exp, &rootfid, &oc);
493 CERROR("cannot mds_connect: rc = %d\n", err);
494 GOTO(out_lock_cn_cb, err);
496 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&rootfid));
497 sbi->ll_root_fid = rootfid;
499 sb->s_op = &lustre_super_operations;
500 sb->s_export_op = &lustre_export_operations;
503 * XXX: move this to after cbd setup? */
504 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
505 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
506 valid |= OBD_MD_FLRMTPERM;
507 else if (sbi->ll_flags & LL_SBI_ACL)
508 valid |= OBD_MD_FLACL;
510 err = md_getattr(sbi->ll_md_exp, &rootfid, oc, valid, 0, &request);
514 CERROR("md_getattr failed for root: rc = %d\n", err);
515 GOTO(out_lock_cn_cb, err);
517 memset(&lmd, 0, sizeof(lmd));
518 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
519 sbi->ll_md_exp, &lmd);
521 CERROR("failed to understand root inode md: rc = %d\n", err);
522 ptlrpc_req_finished (request);
523 GOTO(out_lock_cn_cb, err);
526 LASSERT(fid_is_sane(&sbi->ll_root_fid));
527 root = ll_iget(sb, ll_fid_build_ino(sbi, &sbi->ll_root_fid), &lmd);
528 md_free_lustre_md(sbi->ll_md_exp, &lmd);
529 ptlrpc_req_finished(request);
531 if (root == NULL || is_bad_inode(root)) {
533 obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
534 #ifdef CONFIG_FS_POSIX_ACL
536 posix_acl_release(lmd.posix_acl);
537 lmd.posix_acl = NULL;
540 CERROR("lustre_lite: bad iget4 for root\n");
541 GOTO(out_root, err = -EBADF);
544 err = ll_close_thread_start(&sbi->ll_lcq);
546 CERROR("cannot start close thread: rc %d\n", err);
550 #ifdef CONFIG_FS_POSIX_ACL
551 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
552 rct_init(&sbi->ll_rct);
553 et_init(&sbi->ll_et);
557 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
558 err = obd_set_info_async(sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
559 KEY_CHECKSUM, sizeof(checksum), &checksum,
562 sb->s_root = d_alloc_root(root);
564 OBD_FREE(data, sizeof(*data));
565 sb->s_root->d_op = &ll_d_root_ops;
571 obd_unregister_lock_cancel_cb(sbi->ll_dt_exp,
572 ll_extent_lock_cancel_cb);
574 obd_unregister_page_removal_cb(sbi->ll_dt_exp,
576 obd_fid_fini(sbi->ll_dt_exp);
578 obd_disconnect(sbi->ll_dt_exp);
579 sbi->ll_dt_exp = NULL;
581 obd_fid_fini(sbi->ll_md_exp);
583 obd_disconnect(sbi->ll_md_exp);
584 sbi->ll_md_exp = NULL;
588 lprocfs_unregister_mountpoint(sbi);
592 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
596 *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
598 rc = obd_get_info(sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
599 KEY_MAX_EASIZE, &size, lmmsize);
601 CERROR("Get max mdsize error rc %d \n", rc);
606 void ll_dump_inode(struct inode *inode)
608 struct list_head *tmp;
609 int dentry_count = 0;
611 LASSERT(inode != NULL);
613 list_for_each(tmp, &inode->i_dentry)
616 CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
617 inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
618 inode->i_mode, atomic_read(&inode->i_count), dentry_count);
621 void lustre_dump_dentry(struct dentry *dentry, int recur)
623 struct list_head *tmp;
626 LASSERT(dentry != NULL);
628 list_for_each(tmp, &dentry->d_subdirs)
631 CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
632 " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
633 dentry->d_name.len, dentry->d_name.name,
634 dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
635 dentry->d_parent, dentry->d_inode, atomic_read(&dentry->d_count),
636 dentry->d_flags, dentry->d_fsdata, subdirs);
637 if (dentry->d_inode != NULL)
638 ll_dump_inode(dentry->d_inode);
643 list_for_each(tmp, &dentry->d_subdirs) {
644 struct dentry *d = list_entry(tmp, struct dentry, d_child);
645 lustre_dump_dentry(d, recur - 1);
649 #ifdef HAVE_EXPORT___IGET
650 static void prune_dir_dentries(struct inode *inode)
652 struct dentry *dentry, *prev = NULL;
654 /* due to lustre specific logic, a directory
655 * can have few dentries - a bug from VFS POV */
657 spin_lock(&dcache_lock);
658 if (!list_empty(&inode->i_dentry)) {
659 dentry = list_entry(inode->i_dentry.prev,
660 struct dentry, d_alias);
661 /* in order to prevent infinite loops we
662 * break if previous dentry is busy */
663 if (dentry != prev) {
666 spin_unlock(&dcache_lock);
668 /* try to kill all child dentries */
670 shrink_dcache_parent(dentry);
671 unlock_dentry(dentry);
674 /* now try to get rid of current dentry */
675 d_prune_aliases(inode);
679 spin_unlock(&dcache_lock);
682 static void prune_deathrow_one(struct ll_inode_info *lli)
684 struct inode *inode = ll_info2i(lli);
686 /* first, try to drop any dentries - they hold a ref on the inode */
687 if (S_ISDIR(inode->i_mode))
688 prune_dir_dentries(inode);
690 d_prune_aliases(inode);
693 /* if somebody still uses it, leave it */
694 LASSERT(atomic_read(&inode->i_count) > 0);
695 if (atomic_read(&inode->i_count) > 1)
698 CDEBUG(D_INODE, "inode %lu/%u(%d) looks a good candidate for prune\n",
699 inode->i_ino,inode->i_generation, atomic_read(&inode->i_count));
701 /* seems nobody uses it anymore */
709 static void prune_deathrow(struct ll_sb_info *sbi, int try)
711 struct ll_inode_info *lli;
715 if (need_resched() && try)
719 if (!spin_trylock(&sbi->ll_deathrow_lock))
722 spin_lock(&sbi->ll_deathrow_lock);
727 if (!list_empty(&sbi->ll_deathrow)) {
728 lli = list_entry(sbi->ll_deathrow.next,
729 struct ll_inode_info,
731 list_del_init(&lli->lli_dead_list);
732 if (!list_empty(&sbi->ll_deathrow))
735 spin_unlock(&sbi->ll_deathrow_lock);
738 prune_deathrow_one(lli);
740 } while (empty == 0);
742 #else /* !HAVE_EXPORT___IGET */
743 #define prune_deathrow(sbi, try) do {} while (0)
744 #endif /* HAVE_EXPORT___IGET */
746 void client_common_put_super(struct super_block *sb)
748 struct ll_sb_info *sbi = ll_s2sbi(sb);
751 #ifdef CONFIG_FS_POSIX_ACL
752 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
753 et_fini(&sbi->ll_et);
754 rct_fini(&sbi->ll_rct);
758 obd_cancel_unused(sbi->ll_dt_exp, NULL, 0, NULL);
760 ll_close_thread_shutdown(sbi->ll_lcq);
762 /* destroy inodes in deathrow */
763 prune_deathrow(sbi, 0);
765 list_del(&sbi->ll_conn_chain);
767 obd_fid_fini(sbi->ll_dt_exp);
768 obd_disconnect(sbi->ll_dt_exp);
769 sbi->ll_dt_exp = NULL;
771 lprocfs_unregister_mountpoint(sbi);
773 obd_fid_fini(sbi->ll_md_exp);
774 obd_disconnect(sbi->ll_md_exp);
775 sbi->ll_md_exp = NULL;
780 void ll_kill_super(struct super_block *sb)
782 struct ll_sb_info *sbi;
787 if (!(sb->s_flags & MS_ACTIVE))
791 /* we need restore s_dev from changed for clustred NFS before put_super
792 * because new kernels have cached s_dev and change sb->s_dev in
793 * put_super not affected real removing devices */
795 sb->s_dev = sbi->ll_sdev_orig;
799 char *ll_read_opt(const char *opt, char *data)
805 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
806 if (strncmp(opt, data, strlen(opt)))
808 if ((value = strchr(data, '=')) == NULL)
812 OBD_ALLOC(retval, strlen(value) + 1);
814 CERROR("out of memory!\n");
818 memcpy(retval, value, strlen(value)+1);
819 CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
823 static inline int ll_set_opt(const char *opt, char *data, int fl)
825 if (strncmp(opt, data, strlen(opt)) != 0)
831 /* non-client-specific mount options are parsed in lmd_parse */
832 static int ll_options(char *options, int *flags)
835 char *s1 = options, *s2;
841 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
844 CDEBUG(D_SUPER, "next opt=%s\n", s1);
845 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
850 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
855 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
860 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
865 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
870 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
875 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
877 /* Ignore deprecated mount option. The client will
878 * always try to mount with ACL support, whether this
879 * is used depends on whether server supports it. */
882 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
886 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
892 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
897 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
902 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
907 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
913 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
919 s2 = strchr(s1, ',');
927 void ll_lli_init(struct ll_inode_info *lli)
929 lli->lli_inode_magic = LLI_INODE_MAGIC;
930 sema_init(&lli->lli_size_sem, 1);
931 sema_init(&lli->lli_write_sem, 1);
933 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
934 spin_lock_init(&lli->lli_lock);
935 INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
936 INIT_LIST_HEAD(&lli->lli_close_list);
937 lli->lli_inode_magic = LLI_INODE_MAGIC;
938 sema_init(&lli->lli_och_sem, 1);
939 lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
940 lli->lli_mds_exec_och = NULL;
941 lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
942 lli->lli_open_fd_exec_count = 0;
943 INIT_LIST_HEAD(&lli->lli_dead_list);
944 lli->lli_remote_perms = NULL;
945 lli->lli_rmtperm_utime = 0;
946 sema_init(&lli->lli_rmtperm_sem, 1);
947 INIT_LIST_HEAD(&lli->lli_oss_capas);
950 int ll_fill_super(struct super_block *sb)
952 struct lustre_profile *lprof;
953 struct lustre_sb_info *lsi = s2lsi(sb);
954 struct ll_sb_info *sbi;
955 char *dt = NULL, *md = NULL;
956 char *profilenm = get_profile_name(sb);
957 struct config_llog_instance cfg = {0, };
958 char ll_instance[sizeof(sb) * 2 + 1];
962 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
966 /* client additional sb info */
967 lsi->lsi_llsbi = sbi = ll_init_sbi();
973 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
977 /* Generate a string unique to this super, in case some joker tries
978 to mount the same fs at two mount points.
979 Use the address of the super itself.*/
980 sprintf(ll_instance, "%p", sb);
981 cfg.cfg_instance = ll_instance;
982 cfg.cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
984 /* set up client obds */
985 err = lustre_process_log(sb, profilenm, &cfg);
987 CERROR("Unable to process log: %d\n", err);
991 lprof = class_get_profile(profilenm);
993 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
994 " read from the MGS. Does that filesystem "
995 "exist?\n", profilenm);
996 GOTO(out_free, err = -EINVAL);
998 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
999 lprof->lp_md, lprof->lp_dt);
1001 OBD_ALLOC(dt, strlen(lprof->lp_dt) +
1002 strlen(ll_instance) + 2);
1004 GOTO(out_free, err = -ENOMEM);
1005 sprintf(dt, "%s-%s", lprof->lp_dt, ll_instance);
1007 OBD_ALLOC(md, strlen(lprof->lp_md) +
1008 strlen(ll_instance) + 2);
1010 GOTO(out_free, err = -ENOMEM);
1011 sprintf(md, "%s-%s", lprof->lp_md, ll_instance);
1013 /* connections, registrations, sb setup */
1014 err = client_common_fill_super(sb, md, dt);
1018 OBD_FREE(md, strlen(md) + 1);
1020 OBD_FREE(dt, strlen(dt) + 1);
1024 LCONSOLE_WARN("Client %s has started\n", profilenm);
1027 } /* ll_fill_super */
1030 void ll_put_super(struct super_block *sb)
1032 struct config_llog_instance cfg;
1033 char ll_instance[sizeof(sb) * 2 + 1];
1034 struct obd_device *obd;
1035 struct lustre_sb_info *lsi = s2lsi(sb);
1036 struct ll_sb_info *sbi = ll_s2sbi(sb);
1037 char *profilenm = get_profile_name(sb);
1038 int force = 1, next;
1041 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1043 ll_print_capa_stat(sbi);
1045 sprintf(ll_instance, "%p", sb);
1046 cfg.cfg_instance = ll_instance;
1047 lustre_end_log(sb, NULL, &cfg);
1049 if (sbi->ll_md_exp) {
1050 obd = class_exp2obd(sbi->ll_md_exp);
1052 force = obd->obd_force;
1055 /* We need to set force before the lov_disconnect in
1056 lustre_common_put_super, since l_d cleans up osc's as well. */
1059 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1061 obd->obd_force = force;
1066 /* Only if client_common_fill_super succeeded */
1067 client_common_put_super(sb);
1070 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1071 class_manual_cleanup(obd);
1075 class_del_profile(profilenm);
1078 lsi->lsi_llsbi = NULL;
1080 lustre_common_put_super(sb);
1082 LCONSOLE_WARN("client %s umount complete\n", ll_instance);
1087 } /* client_put_super */
1089 #ifdef HAVE_REGISTER_CACHE
1090 #include <linux/cache_def.h>
1091 #ifdef HAVE_CACHE_RETURN_INT
1096 ll_shrink_cache(int priority, unsigned int gfp_mask)
1098 struct ll_sb_info *sbi;
1101 list_for_each_entry(sbi, &ll_super_blocks, ll_list)
1102 count += llap_shrink_cache(sbi, priority);
1104 #ifdef HAVE_CACHE_RETURN_INT
1109 struct cache_definition ll_cache_definition = {
1110 .name = "llap_cache",
1111 .shrink = ll_shrink_cache
1113 #endif /* HAVE_REGISTER_CACHE */
1115 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1117 struct inode *inode = NULL;
1118 /* NOTE: we depend on atomic igrab() -bzzz */
1119 lock_res_and_lock(lock);
1120 if (lock->l_ast_data) {
1121 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1122 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1123 inode = igrab(lock->l_ast_data);
1125 inode = lock->l_ast_data;
1126 ldlm_lock_debug(NULL, inode->i_state & I_FREEING ?
1128 lock, __FILE__, __func__, __LINE__,
1129 "l_ast_data %p is bogus: magic %08x",
1130 lock->l_ast_data, lli->lli_inode_magic);
1134 unlock_res_and_lock(lock);
1138 static int null_if_equal(struct ldlm_lock *lock, void *data)
1140 if (data == lock->l_ast_data) {
1141 lock->l_ast_data = NULL;
1143 if (lock->l_req_mode != lock->l_granted_mode)
1144 LDLM_ERROR(lock,"clearing inode with ungranted lock");
1147 return LDLM_ITER_CONTINUE;
1150 void ll_clear_inode(struct inode *inode)
1152 struct ll_inode_info *lli = ll_i2info(inode);
1153 struct ll_sb_info *sbi = ll_i2sbi(inode);
1156 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1157 inode->i_generation, inode);
1159 if (S_ISDIR(inode->i_mode)) {
1160 /* these should have been cleared in ll_file_release */
1161 LASSERT(lli->lli_sai == NULL);
1162 LASSERT(lli->lli_opendir_key == NULL);
1163 LASSERT(lli->lli_opendir_pid == 0);
1166 ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1167 md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1168 null_if_equal, inode);
1170 LASSERT(!lli->lli_open_fd_write_count);
1171 LASSERT(!lli->lli_open_fd_read_count);
1172 LASSERT(!lli->lli_open_fd_exec_count);
1174 if (lli->lli_mds_write_och)
1175 ll_md_real_close(inode, FMODE_WRITE);
1176 if (lli->lli_mds_exec_och)
1177 ll_md_real_close(inode, FMODE_EXEC);
1178 if (lli->lli_mds_read_och)
1179 ll_md_real_close(inode, FMODE_READ);
1182 obd_change_cbdata(sbi->ll_dt_exp, lli->lli_smd,
1183 null_if_equal, inode);
1185 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1186 lli->lli_smd = NULL;
1189 if (lli->lli_symlink_name) {
1190 OBD_FREE(lli->lli_symlink_name,
1191 strlen(lli->lli_symlink_name) + 1);
1192 lli->lli_symlink_name = NULL;
1195 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1196 LASSERT(lli->lli_posix_acl == NULL);
1197 if (lli->lli_remote_perms) {
1198 free_rmtperm_hash(lli->lli_remote_perms);
1199 lli->lli_remote_perms = NULL;
1202 #ifdef CONFIG_FS_POSIX_ACL
1203 else if (lli->lli_posix_acl) {
1204 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1205 LASSERT(lli->lli_remote_perms == NULL);
1206 posix_acl_release(lli->lli_posix_acl);
1207 lli->lli_posix_acl = NULL;
1210 lli->lli_inode_magic = LLI_INODE_DEAD;
1212 #ifdef HAVE_EXPORT___IGET
1213 spin_lock(&sbi->ll_deathrow_lock);
1214 list_del_init(&lli->lli_dead_list);
1215 spin_unlock(&sbi->ll_deathrow_lock);
1217 ll_clear_inode_capas(inode);
1222 int ll_md_setattr(struct inode *inode, struct md_op_data *op_data,
1223 struct md_open_data **mod)
1225 struct lustre_md md;
1226 struct ll_sb_info *sbi = ll_i2sbi(inode);
1227 struct ptlrpc_request *request = NULL;
1231 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1232 LUSTRE_OPC_ANY, NULL);
1233 if (IS_ERR(op_data))
1234 RETURN(PTR_ERR(op_data));
1236 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1239 ptlrpc_req_finished(request);
1240 if (rc == -ENOENT) {
1242 /* Unlinked special device node? Or just a race?
1243 * Pretend we done everything. */
1244 if (!S_ISREG(inode->i_mode) &&
1245 !S_ISDIR(inode->i_mode))
1246 rc = inode_setattr(inode, &op_data->op_attr);
1247 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1248 CERROR("md_setattr fails: rc = %d\n", rc);
1253 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1254 sbi->ll_md_exp, &md);
1256 ptlrpc_req_finished(request);
1260 /* We call inode_setattr to adjust timestamps.
1261 * If there is at least some data in file, we cleared ATTR_SIZE
1262 * above to avoid invoking vmtruncate, otherwise it is important
1263 * to call vmtruncate in inode_setattr to update inode->i_size
1265 rc = inode_setattr(inode, &op_data->op_attr);
1267 /* Extract epoch data if obtained. */
1268 op_data->op_handle = md.body->handle;
1269 op_data->op_ioepoch = md.body->ioepoch;
1271 ll_update_inode(inode, &md);
1272 ptlrpc_req_finished(request);
1277 /* Close IO epoch and send Size-on-MDS attribute update. */
1278 static int ll_setattr_done_writing(struct inode *inode,
1279 struct md_op_data *op_data,
1280 struct md_open_data *mod)
1282 struct ll_inode_info *lli = ll_i2info(inode);
1286 LASSERT(op_data != NULL);
1287 if (!S_ISREG(inode->i_mode))
1290 CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1291 op_data->op_ioepoch, PFID(&lli->lli_fid));
1293 op_data->op_flags = MF_EPOCH_CLOSE | MF_SOM_CHANGE;
1294 rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1295 if (rc == -EAGAIN) {
1296 /* MDS has instructed us to obtain Size-on-MDS attribute
1297 * from OSTs and send setattr to back to MDS. */
1298 rc = ll_sizeonmds_update(inode, mod, &op_data->op_handle,
1299 op_data->op_ioepoch);
1301 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1307 static int ll_setattr_do_truncate(struct inode *inode, loff_t new_size)
1309 struct ll_sb_info *sbi = ll_i2sbi(inode);
1310 struct ll_inode_info *lli = ll_i2info(inode);
1311 struct lov_stripe_md *lsm = lli->lli_smd;
1313 ldlm_policy_data_t policy = { .l_extent = {new_size,
1315 struct lustre_handle lockh = { 0 };
1316 int local_lock = 0; /* 0 - no local lock;
1317 * 1 - lock taken by lock_extent;
1318 * 2 - by obd_match*/
1323 UNLOCK_INODE_MUTEX(inode);
1324 UP_WRITE_I_ALLOC_SEM(inode);
1326 if (sbi->ll_lockless_truncate_enable &&
1327 (sbi->ll_lco.lco_flags & OBD_CONNECT_TRUNCLOCK)) {
1328 ast_flags = LDLM_FL_BLOCK_GRANTED;
1329 rc = obd_match(sbi->ll_dt_exp, lsm, LDLM_EXTENT,
1330 &policy, LCK_PW, &ast_flags, inode, &lockh);
1334 } else if (rc == 0) {
1335 rc = ll_file_punch(inode, new_size, 1);
1338 /* XXX when we fix the AST intents to pass the discard-range
1339 * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
1341 ast_flags = (new_size == 0) ? LDLM_AST_DISCARD_DATA : 0;
1342 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy,
1344 if (likely(rc == 0))
1348 LOCK_INODE_MUTEX(inode);
1349 DOWN_WRITE_I_ALLOC_SEM(inode);
1351 if (likely(rc == 0)) {
1352 /* Only ll_inode_size_lock is taken at this level.
1353 * lov_stripe_lock() is grabbed by ll_truncate() only over
1354 * call to obd_adjust_kms(). If vmtruncate returns 0, then
1355 * ll_truncate dropped ll_inode_size_lock() */
1356 ll_inode_size_lock(inode, 0);
1358 spin_lock(&lli->lli_lock);
1359 lli->lli_flags |= LLIF_SRVLOCK;
1360 spin_unlock(&lli->lli_lock);
1362 rc = vmtruncate(inode, new_size);
1364 spin_lock(&lli->lli_lock);
1365 lli->lli_flags &= ~LLIF_SRVLOCK;
1366 spin_unlock(&lli->lli_lock);
1369 LASSERT(atomic_read(&lli->lli_size_sem.count) <= 0);
1370 ll_inode_size_unlock(inode, 0);
1375 if (local_lock == 2)
1376 err = obd_cancel(sbi->ll_dt_exp, lsm, LCK_PW, &lockh);
1378 err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
1379 if (unlikely(err != 0)){
1380 CERROR("extent unlock failed: err=%d,"
1381 " unlock method =%d\n", err, local_lock);
1389 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1390 * object(s) determine the file size and mtime. Otherwise, the MDS will
1391 * keep these values until such a time that objects are allocated for it.
1392 * We do the MDS operations first, as it is checking permissions for us.
1393 * We don't to the MDS RPC if there is nothing that we want to store there,
1394 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1395 * going to do an RPC anyways.
1397 * If we are doing a truncate, we will send the mtime and ctime updates
1398 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1399 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1402 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1404 struct ll_inode_info *lli = ll_i2info(inode);
1405 struct lov_stripe_md *lsm = lli->lli_smd;
1406 struct ll_sb_info *sbi = ll_i2sbi(inode);
1407 struct md_op_data *op_data = NULL;
1408 struct md_open_data *mod = NULL;
1409 int ia_valid = attr->ia_valid;
1410 int rc = 0, rc1 = 0;
1413 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1415 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_SETATTR, 1);
1417 if (ia_valid & ATTR_SIZE) {
1418 if (attr->ia_size > ll_file_maxbytes(inode)) {
1419 CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1420 attr->ia_size, ll_file_maxbytes(inode));
1424 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1427 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1428 if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1429 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1433 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1434 if (attr->ia_valid & ATTR_CTIME) {
1435 attr->ia_ctime = CURRENT_TIME;
1436 attr->ia_valid |= ATTR_CTIME_SET;
1438 if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1439 attr->ia_atime = CURRENT_TIME;
1440 attr->ia_valid |= ATTR_ATIME_SET;
1442 if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1443 attr->ia_mtime = CURRENT_TIME;
1444 attr->ia_valid |= ATTR_MTIME_SET;
1446 if ((attr->ia_valid & ATTR_CTIME) && !(attr->ia_valid & ATTR_MTIME)) {
1447 /* To avoid stale mtime on mds, obtain it from ost and send
1449 rc = ll_glimpse_size(inode, 0);
1453 attr->ia_valid |= ATTR_MTIME_SET | ATTR_MTIME;
1454 attr->ia_mtime = inode->i_mtime;
1457 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1458 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1459 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1460 cfs_time_current_sec());
1462 /* NB: ATTR_SIZE will only be set after this point if the size
1463 * resides on the MDS, ie, this file has no objects. */
1466 printk("LSM mode, unset the ATTR_SIZE\n" );
1467 attr->ia_valid &= ~ATTR_SIZE;
1470 /* We always do an MDS RPC, even if we're only changing the size;
1471 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1473 OBD_ALLOC_PTR(op_data);
1474 if (op_data == NULL)
1477 memcpy(&op_data->op_attr, attr, sizeof(*attr));
1479 /* Open epoch for truncate. */
1480 if ((ll_i2mdexp(inode)->exp_connect_flags & OBD_CONNECT_SOM) &&
1481 (ia_valid & ATTR_SIZE))
1482 op_data->op_flags = MF_EPOCH_OPEN;
1484 rc = ll_md_setattr(inode, op_data, &mod);
1488 if (op_data->op_ioepoch)
1489 CDEBUG(D_INODE, "Epoch "LPU64" opened on "DFID" for "
1490 "truncate\n", op_data->op_ioepoch, PFID(&lli->lli_fid));
1492 if (!lsm || !S_ISREG(inode->i_mode)) {
1493 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1497 /* We really need to get our PW lock before we change inode->i_size.
1498 * If we don't we can race with other i_size updaters on our node, like
1499 * ll_file_read. We can also race with i_size propogation to other
1500 * nodes through dirtying and writeback of final cached pages. This
1501 * last one is especially bad for racing o_append users on other
1503 if (ia_valid & ATTR_SIZE) {
1504 rc = ll_setattr_do_truncate(inode, attr->ia_size);
1505 } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
1507 struct obd_info oinfo = { { { 0 } } };
1510 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1511 inode->i_ino, LTIME_S(attr->ia_mtime));
1515 oa->o_id = lsm->lsm_object_id;
1516 oa->o_gr = lsm->lsm_object_gr;
1517 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1519 flags = OBD_MD_FLTYPE | OBD_MD_FLATIME |
1520 OBD_MD_FLMTIME | OBD_MD_FLCTIME |
1521 OBD_MD_FLFID | OBD_MD_FLGENER |
1524 obdo_from_inode(oa, inode, flags);
1528 oinfo.oi_capa = ll_mdscapa_get(inode);
1530 /* XXX: this looks unnecessary now. */
1531 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1532 capa_put(oinfo.oi_capa);
1534 CERROR("obd_setattr_async fails: rc=%d\n", rc);
1543 if (op_data->op_ioepoch)
1544 rc1 = ll_setattr_done_writing(inode, op_data, mod);
1545 ll_finish_md_op_data(op_data);
1547 return rc ? rc : rc1;
1550 int ll_setattr(struct dentry *de, struct iattr *attr)
1554 printk("ia_valid=%x\n", attr->ia_valid );
1555 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1556 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1557 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1559 return ll_setattr_raw(de->d_inode, attr);
1562 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1563 __u64 max_age, __u32 flags)
1565 struct ll_sb_info *sbi = ll_s2sbi(sb);
1566 struct obd_statfs obd_osfs;
1570 rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age, flags);
1572 CERROR("md_statfs fails: rc = %d\n", rc);
1576 osfs->os_type = sb->s_magic;
1578 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1579 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1581 rc = obd_statfs_rqset(class_exp2obd(sbi->ll_dt_exp),
1582 &obd_osfs, max_age, flags);
1584 CERROR("obd_statfs fails: rc = %d\n", rc);
1588 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1589 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1592 osfs->os_bsize = obd_osfs.os_bsize;
1593 osfs->os_blocks = obd_osfs.os_blocks;
1594 osfs->os_bfree = obd_osfs.os_bfree;
1595 osfs->os_bavail = obd_osfs.os_bavail;
1597 /* If we don't have as many objects free on the OST as inodes
1598 * on the MDS, we reduce the total number of inodes to
1599 * compensate, so that the "inodes in use" number is correct.
1601 if (obd_osfs.os_ffree < osfs->os_ffree) {
1602 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1604 osfs->os_ffree = obd_osfs.os_ffree;
1609 #ifndef HAVE_STATFS_DENTRY_PARAM
1610 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1613 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1615 struct super_block *sb = de->d_sb;
1617 struct obd_statfs osfs;
1620 CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1621 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1623 /* For now we will always get up-to-date statfs values, but in the
1624 * future we may allow some amount of caching on the client (e.g.
1625 * from QOS or lprocfs updates). */
1626 rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - 1, 0);
1630 statfs_unpack(sfs, &osfs);
1632 /* We need to downshift for all 32-bit kernels, because we can't
1633 * tell if the kernel is being called via sys_statfs64() or not.
1634 * Stop before overflowing f_bsize - in which case it is better
1635 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1636 if (sizeof(long) < 8) {
1637 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1640 osfs.os_blocks >>= 1;
1641 osfs.os_bfree >>= 1;
1642 osfs.os_bavail >>= 1;
1646 sfs->f_blocks = osfs.os_blocks;
1647 sfs->f_bfree = osfs.os_bfree;
1648 sfs->f_bavail = osfs.os_bavail;
1653 void ll_inode_size_lock(struct inode *inode, int lock_lsm)
1655 struct ll_inode_info *lli;
1656 struct lov_stripe_md *lsm;
1658 lli = ll_i2info(inode);
1659 LASSERT(lli->lli_size_sem_owner != current);
1660 down(&lli->lli_size_sem);
1661 LASSERT(lli->lli_size_sem_owner == NULL);
1662 lli->lli_size_sem_owner = current;
1664 LASSERTF(lsm != NULL || lock_lsm == 0, "lsm %p, lock_lsm %d\n",
1667 lov_stripe_lock(lsm);
1670 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm)
1672 struct ll_inode_info *lli;
1673 struct lov_stripe_md *lsm;
1675 lli = ll_i2info(inode);
1677 LASSERTF(lsm != NULL || unlock_lsm == 0, "lsm %p, lock_lsm %d\n",
1680 lov_stripe_unlock(lsm);
1681 LASSERT(lli->lli_size_sem_owner == current);
1682 lli->lli_size_sem_owner = NULL;
1683 up(&lli->lli_size_sem);
1686 static void ll_replace_lsm(struct inode *inode, struct lov_stripe_md *lsm)
1688 struct ll_inode_info *lli = ll_i2info(inode);
1690 dump_lsm(D_INODE, lsm);
1691 dump_lsm(D_INODE, lli->lli_smd);
1692 LASSERTF(lsm->lsm_magic == LOV_MAGIC_JOIN,
1693 "lsm must be joined lsm %p\n", lsm);
1694 obd_free_memmd(ll_i2dtexp(inode), &lli->lli_smd);
1695 CDEBUG(D_INODE, "replace lsm %p to lli_smd %p for inode %lu%u(%p)\n",
1696 lsm, lli->lli_smd, inode->i_ino, inode->i_generation, inode);
1698 lli->lli_maxbytes = lsm->lsm_maxbytes;
1699 if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1700 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1703 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1705 struct ll_inode_info *lli = ll_i2info(inode);
1706 struct mdt_body *body = md->body;
1707 struct lov_stripe_md *lsm = md->lsm;
1708 struct ll_sb_info *sbi = ll_i2sbi(inode);
1710 LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1712 if (lli->lli_smd == NULL) {
1713 if (lsm->lsm_magic != LOV_MAGIC &&
1714 lsm->lsm_magic != LOV_MAGIC_JOIN) {
1715 dump_lsm(D_ERROR, lsm);
1718 CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1719 lsm, inode->i_ino, inode->i_generation, inode);
1720 /* ll_inode_size_lock() requires it is only called
1721 * with lli_smd != NULL or lock_lsm == 0 or we can
1722 * race between lock/unlock. bug 9547 */
1724 lli->lli_maxbytes = lsm->lsm_maxbytes;
1725 if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1726 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1728 if (lli->lli_smd->lsm_magic == lsm->lsm_magic &&
1729 lli->lli_smd->lsm_stripe_count ==
1730 lsm->lsm_stripe_count) {
1731 if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
1732 CERROR("lsm mismatch for inode %ld\n",
1734 CERROR("lli_smd:\n");
1735 dump_lsm(D_ERROR, lli->lli_smd);
1737 dump_lsm(D_ERROR, lsm);
1741 ll_replace_lsm(inode, lsm);
1743 if (lli->lli_smd != lsm)
1744 obd_free_memmd(ll_i2dtexp(inode), &lsm);
1747 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1748 if (body->valid & OBD_MD_FLRMTPERM)
1749 ll_update_remote_perm(inode, md->remote_perm);
1751 #ifdef CONFIG_FS_POSIX_ACL
1752 else if (body->valid & OBD_MD_FLACL) {
1753 spin_lock(&lli->lli_lock);
1754 if (lli->lli_posix_acl)
1755 posix_acl_release(lli->lli_posix_acl);
1756 lli->lli_posix_acl = md->posix_acl;
1757 spin_unlock(&lli->lli_lock);
1760 inode->i_ino = ll_fid_build_ino(sbi, &body->fid1);
1762 if (body->valid & OBD_MD_FLATIME &&
1763 body->atime > LTIME_S(inode->i_atime))
1764 LTIME_S(inode->i_atime) = body->atime;
1766 /* mtime is always updated with ctime, but can be set in past.
1767 As write and utime(2) may happen within 1 second, and utime's
1768 mtime has a priority over write's one, so take mtime from mds
1769 for the same ctimes. */
1770 if (body->valid & OBD_MD_FLCTIME &&
1771 body->ctime >= LTIME_S(inode->i_ctime)) {
1772 LTIME_S(inode->i_ctime) = body->ctime;
1773 if (body->valid & OBD_MD_FLMTIME) {
1774 CDEBUG(D_INODE, "setting ino %lu mtime "
1775 "from %lu to "LPU64"\n", inode->i_ino,
1776 LTIME_S(inode->i_mtime), body->mtime);
1777 LTIME_S(inode->i_mtime) = body->mtime;
1780 if (body->valid & OBD_MD_FLMODE)
1781 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1782 if (body->valid & OBD_MD_FLTYPE)
1783 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1784 if (S_ISREG(inode->i_mode)) {
1785 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1787 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1789 #ifdef HAVE_INODE_BLKSIZE
1790 inode->i_blksize = 1<<inode->i_blkbits;
1792 if (body->valid & OBD_MD_FLUID)
1793 inode->i_uid = body->uid;
1794 if (body->valid & OBD_MD_FLGID)
1795 inode->i_gid = body->gid;
1796 if (body->valid & OBD_MD_FLFLAGS)
1797 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1798 if (body->valid & OBD_MD_FLNLINK)
1799 inode->i_nlink = body->nlink;
1800 if (body->valid & OBD_MD_FLRDEV)
1801 inode->i_rdev = old_decode_dev(body->rdev);
1803 if (body->valid & OBD_MD_FLID) {
1804 /* FID shouldn't be changed! */
1805 if (fid_is_sane(&lli->lli_fid)) {
1806 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1807 "Trying to change FID "DFID
1808 " to the "DFID", inode %lu/%u(%p)\n",
1809 PFID(&lli->lli_fid), PFID(&body->fid1),
1810 inode->i_ino, inode->i_generation, inode);
1812 lli->lli_fid = body->fid1;
1815 LASSERT(fid_seq(&lli->lli_fid) != 0);
1817 if (body->valid & OBD_MD_FLSIZE) {
1818 if ((ll_i2mdexp(inode)->exp_connect_flags & OBD_CONNECT_SOM) &&
1819 S_ISREG(inode->i_mode) && lli->lli_smd) {
1820 struct lustre_handle lockh;
1823 /* As it is possible a blocking ast has been processed
1824 * by this time, we need to check there is an UPDATE
1825 * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1827 mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1830 if (lli->lli_flags & (LLIF_DONE_WRITING |
1831 LLIF_EPOCH_PENDING |
1833 CERROR("ino %lu flags %lu still has "
1834 "size authority! do not trust "
1835 "the size got from MDS\n",
1836 inode->i_ino, lli->lli_flags);
1838 /* Use old size assignment to avoid
1839 * deadlock bz14138 & bz14326 */
1840 inode->i_size = body->size;
1841 lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1843 ldlm_lock_decref(&lockh, mode);
1846 /* Use old size assignment to avoid
1847 * deadlock bz14138 & bz14326 */
1848 inode->i_size = body->size;
1851 if (body->valid & OBD_MD_FLBLOCKS)
1852 inode->i_blocks = body->blocks;
1855 if (body->valid & OBD_MD_FLMDSCAPA) {
1856 LASSERT(md->mds_capa);
1857 ll_add_capa(inode, md->mds_capa);
1859 if (body->valid & OBD_MD_FLOSSCAPA) {
1860 LASSERT(md->oss_capa);
1861 ll_add_capa(inode, md->oss_capa);
1865 static struct backing_dev_info ll_backing_dev_info = {
1866 .ra_pages = 0, /* No readahead */
1867 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,12))
1868 .capabilities = 0, /* Does contribute to dirty memory */
1870 .memory_backed = 0, /* Does contribute to dirty memory */
1874 void ll_read_inode2(struct inode *inode, void *opaque)
1876 struct lustre_md *md = opaque;
1877 struct ll_inode_info *lli = ll_i2info(inode);
1880 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n",
1881 inode->i_ino, inode->i_generation, inode);
1885 LASSERT(!lli->lli_smd);
1887 /* Core attributes from the MDS first. This is a new inode, and
1888 * the VFS doesn't zero times in the core inode so we have to do
1889 * it ourselves. They will be overwritten by either MDS or OST
1890 * attributes - we just need to make sure they aren't newer. */
1891 LTIME_S(inode->i_mtime) = 0;
1892 LTIME_S(inode->i_atime) = 0;
1893 LTIME_S(inode->i_ctime) = 0;
1895 ll_update_inode(inode, md);
1897 /* OIDEBUG(inode); */
1899 if (S_ISREG(inode->i_mode)) {
1900 struct ll_sb_info *sbi = ll_i2sbi(inode);
1901 inode->i_op = &ll_file_inode_operations;
1902 inode->i_fop = sbi->ll_fop;
1903 inode->i_mapping->a_ops = &ll_aops;
1905 } else if (S_ISDIR(inode->i_mode)) {
1906 inode->i_op = &ll_dir_inode_operations;
1907 inode->i_fop = &ll_dir_operations;
1908 inode->i_mapping->a_ops = &ll_dir_aops;
1910 } else if (S_ISLNK(inode->i_mode)) {
1911 inode->i_op = &ll_fast_symlink_inode_operations;
1914 inode->i_op = &ll_special_inode_operations;
1916 init_special_inode(inode, inode->i_mode,
1917 kdev_t_to_nr(inode->i_rdev));
1919 /* initializing backing dev info. */
1920 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
1926 void ll_delete_inode(struct inode *inode)
1928 struct ll_sb_info *sbi = ll_i2sbi(inode);
1932 rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
1934 CERROR("fid_delete() failed, rc %d\n", rc);
1936 truncate_inode_pages(&inode->i_data, 0);
1942 int ll_iocontrol(struct inode *inode, struct file *file,
1943 unsigned int cmd, unsigned long arg)
1945 struct ll_sb_info *sbi = ll_i2sbi(inode);
1946 struct ptlrpc_request *req = NULL;
1951 case EXT3_IOC_GETFLAGS: {
1952 struct mdt_body *body;
1953 struct obd_capa *oc;
1955 oc = ll_mdscapa_get(inode);
1956 rc = md_getattr(sbi->ll_md_exp, ll_inode2fid(inode), oc,
1957 OBD_MD_FLFLAGS, 0, &req);
1960 CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1964 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1966 flags = body->flags;
1968 ptlrpc_req_finished(req);
1970 RETURN(put_user(flags, (int *)arg));
1972 case EXT3_IOC_SETFLAGS: {
1973 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1974 struct obd_info oinfo = { { { 0 } } };
1975 struct md_op_data *op_data;
1977 if (get_user(flags, (int *)arg))
1981 OBDO_ALLOC(oinfo.oi_oa);
1985 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1986 LUSTRE_OPC_ANY, NULL);
1987 if (IS_ERR(op_data))
1988 RETURN(PTR_ERR(op_data));
1990 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1991 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1992 rc = md_setattr(sbi->ll_md_exp, op_data,
1993 NULL, 0, NULL, 0, &req, NULL);
1994 ll_finish_md_op_data(op_data);
1995 ptlrpc_req_finished(req);
1996 if (rc || lsm == NULL) {
1997 OBDO_FREE(oinfo.oi_oa);
2001 oinfo.oi_oa->o_id = lsm->lsm_object_id;
2002 oinfo.oi_oa->o_gr = lsm->lsm_object_gr;
2003 oinfo.oi_oa->o_flags = flags;
2004 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
2006 oinfo.oi_capa = ll_mdscapa_get(inode);
2008 obdo_from_inode(oinfo.oi_oa, inode,
2009 OBD_MD_FLFID | OBD_MD_FLGENER);
2010 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
2011 capa_put(oinfo.oi_capa);
2012 OBDO_FREE(oinfo.oi_oa);
2014 if (rc != -EPERM && rc != -EACCES)
2015 CERROR("md_setattr_async fails: rc = %d\n", rc);
2019 inode->i_flags = ll_ext_to_inode_flags(flags |
2020 MDS_BFLAG_EXT_FLAGS);
2030 int ll_flush_ctx(struct inode *inode)
2032 struct ll_sb_info *sbi = ll_i2sbi(inode);
2034 CDEBUG(D_SEC, "flush context for user %d\n", current->uid);
2036 obd_set_info_async(sbi->ll_md_exp,
2037 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2039 obd_set_info_async(sbi->ll_dt_exp,
2040 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2045 /* umount -f client means force down, don't save state */
2046 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2047 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
2049 struct super_block *sb = vfsmnt->mnt_sb;
2051 void ll_umount_begin(struct super_block *sb)
2054 struct lustre_sb_info *lsi = s2lsi(sb);
2055 struct ll_sb_info *sbi = ll_s2sbi(sb);
2056 struct obd_device *obd;
2057 struct obd_ioctl_data ioc_data = { 0 };
2060 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2061 if (!(flags & MNT_FORCE)) {
2067 /* Tell the MGC we got umount -f */
2068 lsi->lsi_flags |= LSI_UMOUNT_FORCE;
2070 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2071 sb->s_count, atomic_read(&sb->s_active));
2073 obd = class_exp2obd(sbi->ll_md_exp);
2075 CERROR("Invalid MDC connection handle "LPX64"\n",
2076 sbi->ll_md_exp->exp_handle.h_cookie);
2081 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp, sizeof ioc_data,
2084 obd = class_exp2obd(sbi->ll_dt_exp);
2086 CERROR("Invalid LOV connection handle "LPX64"\n",
2087 sbi->ll_dt_exp->exp_handle.h_cookie);
2093 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp, sizeof ioc_data,
2096 /* Really, we'd like to wait until there are no requests outstanding,
2097 * and then continue. For now, we just invalidate the requests,
2098 * schedule, and hope.
2105 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2107 struct ll_sb_info *sbi = ll_s2sbi(sb);
2111 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2112 read_only = *flags & MS_RDONLY;
2113 err = obd_set_info_async(sbi->ll_md_exp,
2114 sizeof(KEY_READ_ONLY),
2115 KEY_READ_ONLY, sizeof(read_only),
2118 CERROR("Failed to change the read-only flag during "
2119 "remount: %d\n", err);
2124 sb->s_flags |= MS_RDONLY;
2126 sb->s_flags &= ~MS_RDONLY;
2131 int ll_prep_inode(struct inode **inode,
2132 struct ptlrpc_request *req,
2133 struct super_block *sb)
2135 struct ll_sb_info *sbi = NULL;
2136 struct lustre_md md;
2140 LASSERT(*inode || sb);
2141 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2142 prune_deathrow(sbi, 1);
2143 memset(&md, 0, sizeof(struct lustre_md));
2145 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2146 sbi->ll_md_exp, &md);
2151 ll_update_inode(*inode, &md);
2153 LASSERT(sb != NULL);
2156 * At this point server returns to client's same fid as client
2157 * generated for creating. So using ->fid1 is okay here.
2159 LASSERT(fid_is_sane(&md.body->fid1));
2161 *inode = ll_iget(sb, ll_fid_build_ino(sbi, &md.body->fid1), &md);
2162 if (*inode == NULL || is_bad_inode(*inode)) {
2164 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2165 #ifdef CONFIG_FS_POSIX_ACL
2167 posix_acl_release(md.posix_acl);
2168 md.posix_acl = NULL;
2172 CERROR("new_inode -fatal: rc %d\n", rc);
2177 rc = obd_checkmd(sbi->ll_dt_exp, sbi->ll_md_exp,
2178 ll_i2info(*inode)->lli_smd);
2180 md_free_lustre_md(sbi->ll_md_exp, &md);
2184 char *llap_origins[] = {
2185 [LLAP_ORIGIN_UNKNOWN] = "--",
2186 [LLAP_ORIGIN_READPAGE] = "rp",
2187 [LLAP_ORIGIN_READAHEAD] = "ra",
2188 [LLAP_ORIGIN_COMMIT_WRITE] = "cw",
2189 [LLAP_ORIGIN_WRITEPAGE] = "wp",
2190 [LLAP_ORIGIN_LOCKLESS_IO] = "ls"
2193 struct ll_async_page *llite_pglist_next_llap(struct ll_sb_info *sbi,
2194 struct list_head *list)
2196 struct ll_async_page *llap;
2197 struct list_head *pos;
2199 list_for_each(pos, list) {
2200 if (pos == &sbi->ll_pglist)
2202 llap = list_entry(pos, struct ll_async_page, llap_pglist_item);
2203 if (llap->llap_page == NULL)
2211 int ll_obd_statfs(struct inode *inode, void *arg)
2213 struct ll_sb_info *sbi = NULL;
2214 struct obd_export *exp;
2216 struct obd_ioctl_data *data = NULL;
2220 if (!inode || !(sbi = ll_i2sbi(inode)))
2221 GOTO(out_statfs, rc = -EINVAL);
2223 rc = obd_ioctl_getdata(&buf, &len, arg);
2225 GOTO(out_statfs, rc);
2228 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2229 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2230 GOTO(out_statfs, rc = -EINVAL);
2232 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2233 if (type == LL_STATFS_MDC)
2234 exp = sbi->ll_md_exp;
2235 else if (type == LL_STATFS_LOV)
2236 exp = sbi->ll_dt_exp;
2238 GOTO(out_statfs, rc = -ENODEV);
2240 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2242 GOTO(out_statfs, rc);
2245 obd_ioctl_freedata(buf, len);
2249 int ll_process_config(struct lustre_cfg *lcfg)
2253 struct lprocfs_static_vars lvars;
2257 lprocfs_llite_init_vars(&lvars);
2259 /* The instance name contains the sb: lustre-client-aacfe000 */
2260 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2261 if (!ptr || !*(++ptr))
2263 if (sscanf(ptr, "%lx", &x) != 1)
2266 /* This better be a real Lustre superblock! */
2267 LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2269 /* Note we have not called client_common_fill_super yet, so
2270 proc fns must be able to handle that! */
2271 rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2276 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2277 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2278 struct inode *i1, struct inode *i2,
2279 const char *name, int namelen,
2280 int mode, __u32 opc, void *data)
2282 LASSERT(i1 != NULL);
2284 if (namelen > ll_i2sbi(i1)->ll_namelen)
2285 return ERR_PTR(-ENAMETOOLONG);
2287 if (op_data == NULL)
2288 OBD_ALLOC_PTR(op_data);
2290 if (op_data == NULL)
2291 return ERR_PTR(-ENOMEM);
2293 ll_i2gids(op_data->op_suppgids, i1, i2);
2294 op_data->op_fid1 = *ll_inode2fid(i1);
2295 op_data->op_capa1 = ll_mdscapa_get(i1);
2298 op_data->op_fid2 = *ll_inode2fid(i2);
2299 op_data->op_capa2 = ll_mdscapa_get(i2);
2301 fid_zero(&op_data->op_fid2);
2302 op_data->op_capa2 = NULL;
2305 op_data->op_name = name;
2306 op_data->op_namelen = namelen;
2307 op_data->op_mode = mode;
2308 op_data->op_mod_time = cfs_time_current_sec();
2309 op_data->op_fsuid = current->fsuid;
2310 op_data->op_fsgid = current->fsgid;
2311 op_data->op_cap = current->cap_effective;
2312 op_data->op_bias = MDS_CHECK_SPLIT;
2313 op_data->op_opc = opc;
2314 op_data->op_mds = 0;
2315 op_data->op_data = data;
2320 void ll_finish_md_op_data(struct md_op_data *op_data)
2322 capa_put(op_data->op_capa1);
2323 capa_put(op_data->op_capa2);
2324 OBD_FREE_PTR(op_data);