4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2012, Whamcloud, Inc.
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/version.h>
48 #include <lustre_lite.h>
49 #include <lustre_ha.h>
50 #include <lustre_dlm.h>
51 #include <lprocfs_status.h>
52 #include <lustre_disk.h>
53 #include <lustre_param.h>
54 #include <lustre_log.h>
55 #include <cl_object.h>
56 #include <obd_cksum.h>
57 #include "llite_internal.h"
59 cfs_mem_cache_t *ll_file_data_slab;
61 CFS_LIST_HEAD(ll_super_blocks);
62 cfs_spinlock_t ll_sb_lock = CFS_SPIN_LOCK_UNLOCKED;
64 #ifndef MS_HAS_NEW_AOPS
65 extern struct address_space_operations ll_aops;
66 extern struct address_space_operations ll_dir_aops;
68 extern struct address_space_operations_ext ll_aops;
69 extern struct address_space_operations_ext ll_dir_aops;
73 #define log2(n) cfs_ffz(~(n))
76 static struct ll_sb_info *ll_init_sbi(void)
78 struct ll_sb_info *sbi = NULL;
85 OBD_ALLOC(sbi, sizeof(*sbi));
89 cfs_spin_lock_init(&sbi->ll_lock);
90 cfs_mutex_init(&sbi->ll_lco.lco_lock);
91 cfs_spin_lock_init(&sbi->ll_pp_extent_lock);
92 cfs_spin_lock_init(&sbi->ll_process_lock);
93 sbi->ll_rw_stats_on = 0;
96 pages = si.totalram - si.totalhigh;
97 if (pages >> (20 - CFS_PAGE_SHIFT) < 512) {
98 #ifdef HAVE_BGL_SUPPORT
99 sbi->ll_async_page_max = pages / 4;
101 sbi->ll_async_page_max = pages / 2;
104 sbi->ll_async_page_max = (pages / 4) * 3;
107 sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
108 SBI_DEFAULT_READAHEAD_MAX);
109 sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
110 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
111 SBI_DEFAULT_READAHEAD_WHOLE_MAX;
112 CFS_INIT_LIST_HEAD(&sbi->ll_conn_chain);
113 CFS_INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
115 ll_generate_random_uuid(uuid);
116 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
117 CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
119 cfs_spin_lock(&ll_sb_lock);
120 cfs_list_add_tail(&sbi->ll_list, &ll_super_blocks);
121 cfs_spin_unlock(&ll_sb_lock);
123 sbi->ll_flags |= LL_SBI_VERBOSE;
124 #ifdef ENABLE_CHECKSUM
125 sbi->ll_flags |= LL_SBI_CHECKSUM;
128 #ifdef HAVE_LRU_RESIZE_SUPPORT
129 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
132 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
133 cfs_spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i]. \
135 cfs_spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i]. \
139 /* metadata statahead is enabled by default */
140 sbi->ll_sa_max = LL_SA_RPC_DEF;
141 cfs_atomic_set(&sbi->ll_sa_total, 0);
142 cfs_atomic_set(&sbi->ll_sa_wrong, 0);
143 cfs_atomic_set(&sbi->ll_agl_total, 0);
144 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
149 void ll_free_sbi(struct super_block *sb)
151 struct ll_sb_info *sbi = ll_s2sbi(sb);
155 cfs_spin_lock(&ll_sb_lock);
156 cfs_list_del(&sbi->ll_list);
157 cfs_spin_unlock(&ll_sb_lock);
158 OBD_FREE(sbi, sizeof(*sbi));
163 static struct dentry_operations ll_d_root_ops = {
164 .d_compare = ll_dcompare,
165 .d_revalidate = ll_revalidate_nd,
168 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
169 struct vfsmount *mnt)
171 struct inode *root = 0;
172 struct ll_sb_info *sbi = ll_s2sbi(sb);
173 struct obd_device *obd;
174 struct obd_capa *oc = NULL;
175 struct obd_statfs *osfs = NULL;
176 struct ptlrpc_request *request = NULL;
177 struct obd_connect_data *data = NULL;
178 struct obd_uuid *uuid;
179 struct md_op_data *op_data;
180 struct lustre_md lmd;
182 int size, err, checksum;
185 obd = class_name2obd(md);
187 CERROR("MD %s: not setup or attached\n", md);
201 if (proc_lustre_fs_root) {
202 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
205 CERROR("could not register mount in /proc/fs/lustre\n");
208 /* indicate the features supported by this client */
209 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
210 OBD_CONNECT_ATTRFID |
211 OBD_CONNECT_VERSION | OBD_CONNECT_BRW_SIZE |
212 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
213 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
214 OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 |
215 OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR |
216 OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH|
217 OBD_CONNECT_EINPROGRESS |
218 OBD_CONNECT_JOBSTATS;
220 if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
221 data->ocd_connect_flags |= OBD_CONNECT_SOM;
223 #ifdef HAVE_LRU_RESIZE_SUPPORT
224 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
225 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
227 #ifdef CONFIG_FS_POSIX_ACL
228 data->ocd_connect_flags |= OBD_CONNECT_ACL;
230 data->ocd_ibits_known = MDS_INODELOCK_FULL;
231 data->ocd_version = LUSTRE_VERSION_CODE;
233 if (sb->s_flags & MS_RDONLY)
234 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
235 if (sbi->ll_flags & LL_SBI_USER_XATTR)
236 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
238 #ifdef HAVE_MS_FLOCK_LOCK
239 /* force vfs to use lustre handler for flock() calls - bug 10743 */
240 sb->s_flags |= MS_FLOCK_LOCK;
242 #ifdef MS_HAS_NEW_AOPS
243 sb->s_flags |= MS_HAS_NEW_AOPS;
246 if (sbi->ll_flags & LL_SBI_FLOCK)
247 sbi->ll_fop = &ll_file_operations_flock;
248 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
249 sbi->ll_fop = &ll_file_operations;
251 sbi->ll_fop = &ll_file_operations_noflock;
254 data->ocd_connect_flags |= OBD_CONNECT_REAL;
255 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
256 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
258 data->ocd_brw_size = PTLRPC_MAX_BRW_SIZE;
260 err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
262 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
263 "recovery, of which this client is not a "
264 "part. Please wait for recovery to complete,"
265 " abort, or time out.\n", md);
268 CERROR("cannot connect to %s: rc = %d\n", md, err);
272 err = obd_fid_init(sbi->ll_md_exp);
274 CERROR("Can't init metadata layer FID infrastructure, "
279 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
280 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS), 0);
282 GOTO(out_md_fid, err);
284 /* This needs to be after statfs to ensure connect has finished.
285 * Note that "data" does NOT contain the valid connect reply.
286 * If connecting to a 1.8 server there will be no LMV device, so
287 * we can access the MDC export directly and exp_connect_flags will
288 * be non-zero, but if accessing an upgraded 2.1 server it will
289 * have the correct flags filled in.
290 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
291 valid = sbi->ll_md_exp->exp_connect_flags & CLIENT_CONNECT_MDT_REQD;
292 if (sbi->ll_md_exp->exp_connect_flags != 0 &&
293 valid != CLIENT_CONNECT_MDT_REQD) {
296 OBD_ALLOC_WAIT(buf, CFS_PAGE_SIZE);
297 obd_connect_flags2str(buf, CFS_PAGE_SIZE,
298 valid ^ CLIENT_CONNECT_MDT_REQD, ",");
299 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
300 "feature(s) needed for correct operation "
301 "of this client (%s). Please upgrade "
302 "server or downgrade client.\n",
303 sbi->ll_md_exp->exp_obd->obd_name, buf);
304 OBD_FREE(buf, CFS_PAGE_SIZE);
305 GOTO(out_md, err = -EPROTO);
308 size = sizeof(*data);
309 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
310 KEY_CONN_DATA, &size, data, NULL);
312 CERROR("Get connect data failed: %d \n", err);
316 LASSERT(osfs->os_bsize);
317 sb->s_blocksize = osfs->os_bsize;
318 sb->s_blocksize_bits = log2(osfs->os_bsize);
319 sb->s_magic = LL_SUPER_MAGIC;
320 sb->s_maxbytes = MAX_LFS_FILESIZE;
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_RMT_CLIENT) {
345 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
346 sbi->ll_flags |= LL_SBI_RMT_CLIENT;
347 LCONSOLE_INFO("client is set as remote by default.\n");
350 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
351 sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
352 LCONSOLE_INFO("client claims to be remote, but server "
353 "rejected, forced to be local.\n");
357 if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
358 LCONSOLE_INFO("client enabled MDS capability!\n");
359 sbi->ll_flags |= LL_SBI_MDS_CAPA;
362 if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
363 LCONSOLE_INFO("client enabled OSS capability!\n");
364 sbi->ll_flags |= LL_SBI_OSS_CAPA;
367 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
368 sbi->ll_flags |= LL_SBI_64BIT_HASH;
370 if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
371 sbi->ll_md_brw_size = data->ocd_brw_size;
373 sbi->ll_md_brw_size = CFS_PAGE_SIZE;
375 obd = class_name2obd(dt);
377 CERROR("DT %s: not setup or attached\n", dt);
378 GOTO(out_md_fid, err = -ENODEV);
381 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
382 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
383 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
384 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
385 OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
386 OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
387 OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
388 OBD_CONNECT_MAXBYTES |
389 OBD_CONNECT_EINPROGRESS |
390 OBD_CONNECT_JOBSTATS;
392 if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
393 data->ocd_connect_flags |= OBD_CONNECT_SOM;
395 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
396 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
397 * disabled by default, because it can still be enabled on the
398 * fly via /proc. As a consequence, we still need to come to an
399 * agreement on the supported algorithms at connect time */
400 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
402 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
403 data->ocd_cksum_types = OBD_CKSUM_ADLER;
405 data->ocd_cksum_types = cksum_types_supported();
408 #ifdef HAVE_LRU_RESIZE_SUPPORT
409 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
411 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
412 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
414 CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
415 "ocd_grant: %d\n", data->ocd_connect_flags,
416 data->ocd_version, data->ocd_grant);
418 obd->obd_upcall.onu_owner = &sbi->ll_lco;
419 obd->obd_upcall.onu_upcall = cl_ocd_update;
421 data->ocd_brw_size = PTLRPC_MAX_BRW_SIZE;
423 err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data, NULL);
425 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
426 "recovery, of which this client is not a "
427 "part. Please wait for recovery to "
428 "complete, abort, or time out.\n", dt);
429 GOTO(out_md_fid, err);
431 CERROR("Cannot connect to %s: rc = %d\n", dt, err);
432 GOTO(out_md_fid, err);
435 err = obd_fid_init(sbi->ll_dt_exp);
437 CERROR("Can't init data layer FID infrastructure, "
442 cfs_mutex_lock(&sbi->ll_lco.lco_lock);
443 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
444 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
445 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
446 cfs_mutex_unlock(&sbi->ll_lco.lco_lock);
448 fid_zero(&sbi->ll_root_fid);
449 err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
451 CERROR("cannot mds_connect: rc = %d\n", err);
452 GOTO(out_lock_cn_cb, err);
454 if (!fid_is_sane(&sbi->ll_root_fid)) {
455 CERROR("Invalid root fid during mount\n");
456 GOTO(out_lock_cn_cb, err = -EINVAL);
458 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
460 sb->s_op = &lustre_super_operations;
461 #if THREAD_SIZE >= 8192
462 sb->s_export_op = &lustre_export_operations;
466 * XXX: move this to after cbd setup? */
467 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
468 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
469 valid |= OBD_MD_FLRMTPERM;
470 else if (sbi->ll_flags & LL_SBI_ACL)
471 valid |= OBD_MD_FLACL;
473 OBD_ALLOC_PTR(op_data);
475 GOTO(out_lock_cn_cb, err = -ENOMEM);
477 op_data->op_fid1 = sbi->ll_root_fid;
478 op_data->op_mode = 0;
479 op_data->op_capa1 = oc;
480 op_data->op_valid = valid;
482 err = md_getattr(sbi->ll_md_exp, op_data, &request);
485 OBD_FREE_PTR(op_data);
487 CERROR("md_getattr failed for root: rc = %d\n", err);
488 GOTO(out_lock_cn_cb, err);
490 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
491 sbi->ll_md_exp, &lmd);
493 CERROR("failed to understand root inode md: rc = %d\n", err);
494 ptlrpc_req_finished (request);
495 GOTO(out_lock_cn_cb, err);
498 LASSERT(fid_is_sane(&sbi->ll_root_fid));
499 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid, 0), &lmd);
500 md_free_lustre_md(sbi->ll_md_exp, &lmd);
501 ptlrpc_req_finished(request);
503 if (root == NULL || IS_ERR(root)) {
505 obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
506 #ifdef CONFIG_FS_POSIX_ACL
508 posix_acl_release(lmd.posix_acl);
509 lmd.posix_acl = NULL;
512 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
514 CERROR("lustre_lite: bad iget4 for root\n");
518 err = ll_close_thread_start(&sbi->ll_lcq);
520 CERROR("cannot start close thread: rc %d\n", err);
524 #ifdef CONFIG_FS_POSIX_ACL
525 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
526 rct_init(&sbi->ll_rct);
527 et_init(&sbi->ll_et);
531 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
532 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
533 KEY_CHECKSUM, sizeof(checksum), &checksum,
537 sb->s_root = d_alloc_root(root);
538 #ifdef DCACHE_OP_HASH
539 d_set_d_op(sb->s_root, &ll_d_root_ops);
540 sb->s_d_op = &ll_d_ops;
542 sb->s_root->d_op = &ll_d_root_ops;
545 sbi->ll_sdev_orig = sb->s_dev;
547 /* We set sb->s_dev equal on all lustre clients in order to support
548 * NFS export clustering. NFSD requires that the FSID be the same
550 /* s_dev is also used in lt_compare() to compare two fs, but that is
551 * only a node-local comparison. */
552 uuid = obd_get_uuid(sbi->ll_md_exp);
554 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
567 obd_fid_fini(sbi->ll_dt_exp);
569 obd_disconnect(sbi->ll_dt_exp);
570 sbi->ll_dt_exp = NULL;
572 obd_fid_fini(sbi->ll_md_exp);
574 obd_disconnect(sbi->ll_md_exp);
575 sbi->ll_md_exp = NULL;
581 lprocfs_unregister_mountpoint(sbi);
585 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
589 *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
591 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
592 KEY_MAX_EASIZE, &size, lmmsize, NULL);
594 CERROR("Get max mdsize error rc %d \n", rc);
599 void ll_dump_inode(struct inode *inode)
601 struct list_head *tmp;
602 int dentry_count = 0;
604 LASSERT(inode != NULL);
606 list_for_each(tmp, &inode->i_dentry)
609 CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
610 inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
611 inode->i_mode, atomic_read(&inode->i_count), dentry_count);
614 void lustre_dump_dentry(struct dentry *dentry, int recur)
616 struct list_head *tmp;
619 LASSERT(dentry != NULL);
621 list_for_each(tmp, &dentry->d_subdirs)
624 CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
625 " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
626 dentry->d_name.len, dentry->d_name.name,
627 dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
628 dentry->d_parent, dentry->d_inode, d_refcount(dentry),
629 dentry->d_flags, dentry->d_fsdata, subdirs);
630 if (dentry->d_inode != NULL)
631 ll_dump_inode(dentry->d_inode);
636 list_for_each(tmp, &dentry->d_subdirs) {
637 struct dentry *d = list_entry(tmp, struct dentry, d_child);
638 lustre_dump_dentry(d, recur - 1);
642 void client_common_put_super(struct super_block *sb)
644 struct ll_sb_info *sbi = ll_s2sbi(sb);
647 #ifdef CONFIG_FS_POSIX_ACL
648 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
649 et_fini(&sbi->ll_et);
650 rct_fini(&sbi->ll_rct);
654 obd_cancel_unused(sbi->ll_dt_exp, NULL, 0, NULL);
656 ll_close_thread_shutdown(sbi->ll_lcq);
660 cfs_list_del(&sbi->ll_conn_chain);
662 obd_fid_fini(sbi->ll_dt_exp);
663 obd_disconnect(sbi->ll_dt_exp);
664 sbi->ll_dt_exp = NULL;
666 lprocfs_unregister_mountpoint(sbi);
668 obd_fid_fini(sbi->ll_md_exp);
669 obd_disconnect(sbi->ll_md_exp);
670 sbi->ll_md_exp = NULL;
675 void ll_kill_super(struct super_block *sb)
677 struct ll_sb_info *sbi;
682 if (!(sb->s_flags & MS_ACTIVE))
686 /* we need restore s_dev from changed for clustred NFS before put_super
687 * because new kernels have cached s_dev and change sb->s_dev in
688 * put_super not affected real removing devices */
690 sb->s_dev = sbi->ll_sdev_orig;
694 char *ll_read_opt(const char *opt, char *data)
700 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
701 if (strncmp(opt, data, strlen(opt)))
703 if ((value = strchr(data, '=')) == NULL)
707 OBD_ALLOC(retval, strlen(value) + 1);
709 CERROR("out of memory!\n");
713 memcpy(retval, value, strlen(value)+1);
714 CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
718 static inline int ll_set_opt(const char *opt, char *data, int fl)
720 if (strncmp(opt, data, strlen(opt)) != 0)
726 /* non-client-specific mount options are parsed in lmd_parse */
727 static int ll_options(char *options, int *flags)
730 char *s1 = options, *s2;
736 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
739 CDEBUG(D_SUPER, "next opt=%s\n", s1);
740 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
745 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
750 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
755 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
760 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
765 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
770 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2,5,50,0)
771 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
773 /* Ignore deprecated mount option. The client will
774 * always try to mount with ACL support, whether this
775 * is used depends on whether server supports it. */
776 LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
777 "mount option 'acl'.\n");
780 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
782 LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
783 "mount option 'noacl'.\n");
787 #warning "{no}acl options have been deprecated since 1.8, please remove them"
789 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
795 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
800 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
805 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
810 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
815 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
820 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
825 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
830 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
835 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
840 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
845 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
851 s2 = strchr(s1, ',');
859 void ll_lli_init(struct ll_inode_info *lli)
861 lli->lli_inode_magic = LLI_INODE_MAGIC;
863 lli->lli_ioepoch = 0;
864 lli->lli_maxbytes = MAX_LFS_FILESIZE;
865 cfs_spin_lock_init(&lli->lli_lock);
866 lli->lli_posix_acl = NULL;
867 lli->lli_remote_perms = NULL;
868 cfs_mutex_init(&lli->lli_rmtperm_mutex);
869 /* Do not set lli_fid, it has been initialized already. */
870 fid_zero(&lli->lli_pfid);
871 CFS_INIT_LIST_HEAD(&lli->lli_close_list);
872 CFS_INIT_LIST_HEAD(&lli->lli_oss_capas);
873 cfs_atomic_set(&lli->lli_open_count, 0);
874 lli->lli_mds_capa = NULL;
875 lli->lli_rmtperm_time = 0;
876 lli->lli_pending_och = NULL;
877 lli->lli_mds_read_och = NULL;
878 lli->lli_mds_write_och = NULL;
879 lli->lli_mds_exec_och = NULL;
880 lli->lli_open_fd_read_count = 0;
881 lli->lli_open_fd_write_count = 0;
882 lli->lli_open_fd_exec_count = 0;
883 cfs_mutex_init(&lli->lli_och_mutex);
884 cfs_spin_lock_init(&lli->lli_agl_lock);
886 lli->lli_clob = NULL;
888 LASSERT(lli->lli_vfs_inode.i_mode != 0);
889 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
890 cfs_mutex_init(&lli->lli_readdir_mutex);
891 lli->lli_opendir_key = NULL;
894 lli->lli_def_acl = NULL;
895 cfs_spin_lock_init(&lli->lli_sa_lock);
896 lli->lli_opendir_pid = 0;
898 cfs_sema_init(&lli->lli_size_sem, 1);
899 lli->lli_size_sem_owner = NULL;
900 lli->lli_symlink_name = NULL;
901 cfs_init_rwsem(&lli->lli_trunc_sem);
902 cfs_mutex_init(&lli->lli_write_mutex);
903 lli->lli_async_rc = 0;
904 lli->lli_write_rc = 0;
905 cfs_init_rwsem(&lli->lli_glimpse_sem);
906 lli->lli_glimpse_time = 0;
907 CFS_INIT_LIST_HEAD(&lli->lli_agl_list);
908 lli->lli_agl_index = 0;
912 static inline int ll_bdi_register(struct backing_dev_info *bdi)
914 #ifdef HAVE_BDI_REGISTER
915 static atomic_t ll_bdi_num = ATOMIC_INIT(0);
918 bdi->name = "lustre";
920 return bdi_register(bdi, NULL, "lustre-%d",
921 atomic_inc_return(&ll_bdi_num));
927 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
929 struct lustre_profile *lprof = NULL;
930 struct lustre_sb_info *lsi = s2lsi(sb);
931 struct ll_sb_info *sbi;
932 char *dt = NULL, *md = NULL;
933 char *profilenm = get_profile_name(sb);
934 struct config_llog_instance *cfg;
935 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
936 const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
940 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
948 /* client additional sb info */
949 lsi->lsi_llsbi = sbi = ll_init_sbi();
951 cfs_module_put(THIS_MODULE);
956 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
960 err = ll_bdi_init(&lsi->lsi_bdi);
963 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
964 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
965 err = ll_bdi_register(&lsi->lsi_bdi);
970 sb->s_bdi = &lsi->lsi_bdi;
973 /* Generate a string unique to this super, in case some joker tries
974 to mount the same fs at two mount points.
975 Use the address of the super itself.*/
976 cfg->cfg_instance = sb;
977 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
979 /* set up client obds */
980 err = lustre_process_log(sb, profilenm, cfg);
982 CERROR("Unable to process log: %d\n", err);
986 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
987 lprof = class_get_profile(profilenm);
989 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
990 " read from the MGS. Does that filesystem "
991 "exist?\n", profilenm);
992 GOTO(out_free, err = -EINVAL);
994 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
995 lprof->lp_md, lprof->lp_dt);
997 OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
999 GOTO(out_free, err = -ENOMEM);
1000 sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1002 OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1004 GOTO(out_free, err = -ENOMEM);
1005 sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1007 /* connections, registrations, sb setup */
1008 err = client_common_fill_super(sb, md, dt, mnt);
1012 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1014 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1017 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1018 LCONSOLE_WARN("Mounted %s\n", profilenm);
1022 } /* ll_fill_super */
1025 void lu_context_keys_dump(void);
1027 void ll_put_super(struct super_block *sb)
1029 struct config_llog_instance cfg;
1030 struct obd_device *obd;
1031 struct lustre_sb_info *lsi = s2lsi(sb);
1032 struct ll_sb_info *sbi = ll_s2sbi(sb);
1033 char *profilenm = get_profile_name(sb);
1034 int force = 1, next;
1037 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1039 ll_print_capa_stat(sbi);
1041 cfg.cfg_instance = sb;
1042 lustre_end_log(sb, profilenm, &cfg);
1044 if (sbi->ll_md_exp) {
1045 obd = class_exp2obd(sbi->ll_md_exp);
1047 force = obd->obd_force;
1050 /* We need to set force before the lov_disconnect in
1051 lustre_common_put_super, since l_d cleans up osc's as well. */
1054 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1056 obd->obd_force = force;
1061 /* Only if client_common_fill_super succeeded */
1062 client_common_put_super(sb);
1066 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1067 class_manual_cleanup(obd);
1070 if (sbi->ll_flags & LL_SBI_VERBOSE)
1071 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1074 class_del_profile(profilenm);
1076 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1077 ll_bdi_destroy(&lsi->lsi_bdi);
1078 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1082 lsi->lsi_llsbi = NULL;
1084 lustre_common_put_super(sb);
1086 cl_env_cache_purge(~0);
1088 cfs_module_put(THIS_MODULE);
1091 } /* client_put_super */
1093 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1095 struct inode *inode = NULL;
1096 /* NOTE: we depend on atomic igrab() -bzzz */
1097 lock_res_and_lock(lock);
1098 if (lock->l_ast_data) {
1099 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1100 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1101 inode = igrab(lock->l_ast_data);
1103 inode = lock->l_ast_data;
1104 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1105 D_WARNING, lock, "l_ast_data %p is "
1106 "bogus: magic %08x", lock->l_ast_data,
1107 lli->lli_inode_magic);
1111 unlock_res_and_lock(lock);
1115 static int null_if_equal(struct ldlm_lock *lock, void *data)
1117 if (data == lock->l_ast_data) {
1118 lock->l_ast_data = NULL;
1120 if (lock->l_req_mode != lock->l_granted_mode)
1121 LDLM_ERROR(lock,"clearing inode with ungranted lock");
1124 return LDLM_ITER_CONTINUE;
1127 void ll_clear_inode(struct inode *inode)
1129 struct ll_inode_info *lli = ll_i2info(inode);
1130 struct ll_sb_info *sbi = ll_i2sbi(inode);
1133 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1134 inode->i_generation, inode);
1136 if (S_ISDIR(inode->i_mode)) {
1137 /* these should have been cleared in ll_file_release */
1138 LASSERT(lli->lli_opendir_key == NULL);
1139 LASSERT(lli->lli_sai == NULL);
1140 LASSERT(lli->lli_opendir_pid == 0);
1143 ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1144 md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1145 null_if_equal, inode);
1147 LASSERT(!lli->lli_open_fd_write_count);
1148 LASSERT(!lli->lli_open_fd_read_count);
1149 LASSERT(!lli->lli_open_fd_exec_count);
1151 if (lli->lli_mds_write_och)
1152 ll_md_real_close(inode, FMODE_WRITE);
1153 if (lli->lli_mds_exec_och)
1154 ll_md_real_close(inode, FMODE_EXEC);
1155 if (lli->lli_mds_read_och)
1156 ll_md_real_close(inode, FMODE_READ);
1158 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1159 OBD_FREE(lli->lli_symlink_name,
1160 strlen(lli->lli_symlink_name) + 1);
1161 lli->lli_symlink_name = NULL;
1164 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1165 LASSERT(lli->lli_posix_acl == NULL);
1166 if (lli->lli_remote_perms) {
1167 free_rmtperm_hash(lli->lli_remote_perms);
1168 lli->lli_remote_perms = NULL;
1171 #ifdef CONFIG_FS_POSIX_ACL
1172 else if (lli->lli_posix_acl) {
1173 LASSERT(cfs_atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1174 LASSERT(lli->lli_remote_perms == NULL);
1175 posix_acl_release(lli->lli_posix_acl);
1176 lli->lli_posix_acl = NULL;
1179 lli->lli_inode_magic = LLI_INODE_DEAD;
1181 ll_clear_inode_capas(inode);
1182 if (!S_ISDIR(inode->i_mode))
1183 LASSERT(cfs_list_empty(&lli->lli_agl_list));
1186 * XXX This has to be done before lsm is freed below, because
1187 * cl_object still uses inode lsm.
1189 cl_inode_fini(inode);
1192 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1193 lli->lli_smd = NULL;
1200 int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data,
1201 struct md_open_data **mod)
1203 struct lustre_md md;
1204 struct inode *inode = dentry->d_inode;
1205 struct ll_sb_info *sbi = ll_i2sbi(inode);
1206 struct ptlrpc_request *request = NULL;
1210 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1211 LUSTRE_OPC_ANY, NULL);
1212 if (IS_ERR(op_data))
1213 RETURN(PTR_ERR(op_data));
1215 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1218 ptlrpc_req_finished(request);
1219 if (rc == -ENOENT) {
1221 /* Unlinked special device node? Or just a race?
1222 * Pretend we done everything. */
1223 if (!S_ISREG(inode->i_mode) &&
1224 !S_ISDIR(inode->i_mode)) {
1225 ia_valid = op_data->op_attr.ia_valid;
1226 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1227 rc = simple_setattr(dentry, &op_data->op_attr);
1228 op_data->op_attr.ia_valid = ia_valid;
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 want to adjust timestamps.
1244 * If there is at least some data in file, we cleared ATTR_SIZE
1245 * to avoid update size, otherwise it is important to do.(SOM case)
1247 ia_valid = op_data->op_attr.ia_valid;
1248 /* Since we set ATTR_*_SET flags above, and already done permission
1249 * check, So don't let inode_change_ok() check it again. */
1250 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1251 rc = simple_setattr(dentry, &op_data->op_attr);
1252 op_data->op_attr.ia_valid = ia_valid;
1254 /* Extract epoch data if obtained. */
1255 op_data->op_handle = md.body->handle;
1256 op_data->op_ioepoch = md.body->ioepoch;
1258 ll_update_inode(inode, &md);
1259 ptlrpc_req_finished(request);
1264 /* Close IO epoch and send Size-on-MDS attribute update. */
1265 static int ll_setattr_done_writing(struct inode *inode,
1266 struct md_op_data *op_data,
1267 struct md_open_data *mod)
1269 struct ll_inode_info *lli = ll_i2info(inode);
1273 LASSERT(op_data != NULL);
1274 if (!S_ISREG(inode->i_mode))
1277 CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1278 op_data->op_ioepoch, PFID(&lli->lli_fid));
1280 op_data->op_flags = MF_EPOCH_CLOSE;
1281 ll_done_writing_attr(inode, op_data);
1282 ll_pack_inode2opdata(inode, op_data, NULL);
1284 rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1285 if (rc == -EAGAIN) {
1286 /* MDS has instructed us to obtain Size-on-MDS attribute
1287 * from OSTs and send setattr to back to MDS. */
1288 rc = ll_som_update(inode, op_data);
1290 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1296 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1298 struct obd_capa *capa;
1301 if (attr->ia_valid & ATTR_SIZE)
1302 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1304 capa = ll_mdscapa_get(inode);
1306 rc = cl_setattr_ost(inode, attr, capa);
1308 if (attr->ia_valid & ATTR_SIZE)
1309 ll_truncate_free_capa(capa);
1316 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1317 * object(s) determine the file size and mtime. Otherwise, the MDS will
1318 * keep these values until such a time that objects are allocated for it.
1319 * We do the MDS operations first, as it is checking permissions for us.
1320 * We don't to the MDS RPC if there is nothing that we want to store there,
1321 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1322 * going to do an RPC anyways.
1324 * If we are doing a truncate, we will send the mtime and ctime updates
1325 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1326 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1329 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr)
1331 struct inode *inode = dentry->d_inode;
1332 struct ll_inode_info *lli = ll_i2info(inode);
1333 struct lov_stripe_md *lsm = lli->lli_smd;
1334 struct md_op_data *op_data = NULL;
1335 struct md_open_data *mod = NULL;
1336 int ia_valid = attr->ia_valid;
1337 int rc = 0, rc1 = 0;
1340 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1343 if (ia_valid & ATTR_SIZE) {
1344 /* Check new size against VFS/VM file size limit and rlimit */
1345 rc = inode_newsize_ok(inode, attr->ia_size);
1349 /* The maximum Lustre file size is variable, based on the
1350 * OST maximum object size and number of stripes. This
1351 * needs another check in addition to the VFS check above. */
1352 if (attr->ia_size > ll_file_maxbytes(inode)) {
1353 CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1354 PFID(&lli->lli_fid), attr->ia_size,
1355 ll_file_maxbytes(inode));
1359 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1362 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1363 if (ia_valid & TIMES_SET_FLAGS) {
1364 if (cfs_curproc_fsuid() != inode->i_uid &&
1365 !cfs_capable(CFS_CAP_FOWNER))
1369 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1370 if (attr->ia_valid & ATTR_CTIME) {
1371 attr->ia_ctime = CFS_CURRENT_TIME;
1372 attr->ia_valid |= ATTR_CTIME_SET;
1374 if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1375 attr->ia_atime = CFS_CURRENT_TIME;
1376 attr->ia_valid |= ATTR_ATIME_SET;
1378 if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1379 attr->ia_mtime = CFS_CURRENT_TIME;
1380 attr->ia_valid |= ATTR_MTIME_SET;
1383 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1384 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1385 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1386 cfs_time_current_sec());
1388 /* NB: ATTR_SIZE will only be set after this point if the size
1389 * resides on the MDS, ie, this file has no objects. */
1391 attr->ia_valid &= ~ATTR_SIZE;
1393 /* We always do an MDS RPC, even if we're only changing the size;
1394 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1396 OBD_ALLOC_PTR(op_data);
1397 if (op_data == NULL)
1400 if (!S_ISDIR(inode->i_mode)) {
1401 if (ia_valid & ATTR_SIZE)
1402 UP_WRITE_I_ALLOC_SEM(inode);
1403 UNLOCK_INODE_MUTEX(inode);
1404 cfs_down_write(&lli->lli_trunc_sem);
1405 LOCK_INODE_MUTEX(inode);
1406 if (ia_valid & ATTR_SIZE)
1407 DOWN_WRITE_I_ALLOC_SEM(inode);
1410 memcpy(&op_data->op_attr, attr, sizeof(*attr));
1412 /* Open epoch for truncate. */
1413 if (exp_connect_som(ll_i2mdexp(inode)) &&
1414 (ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1415 op_data->op_flags = MF_EPOCH_OPEN;
1417 rc = ll_md_setattr(dentry, op_data, &mod);
1421 ll_ioepoch_open(lli, op_data->op_ioepoch);
1422 if (!lsm || !S_ISREG(inode->i_mode)) {
1423 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1427 if (ia_valid & ATTR_SIZE)
1428 attr->ia_valid |= ATTR_SIZE;
1429 if (ia_valid & (ATTR_SIZE |
1430 ATTR_ATIME | ATTR_ATIME_SET |
1431 ATTR_MTIME | ATTR_MTIME_SET))
1432 /* For truncate and utimes sending attributes to OSTs, setting
1433 * mtime/atime to the past will be performed under PW [0:EOF]
1434 * extent lock (new_size:EOF for truncate). It may seem
1435 * excessive to send mtime/atime updates to OSTs when not
1436 * setting times to past, but it is necessary due to possible
1437 * time de-synchronization between MDT inode and OST objects */
1438 rc = ll_setattr_ost(inode, attr);
1442 if (op_data->op_ioepoch) {
1443 rc1 = ll_setattr_done_writing(inode, op_data, mod);
1447 ll_finish_md_op_data(op_data);
1449 if (!S_ISDIR(inode->i_mode))
1450 cfs_up_write(&lli->lli_trunc_sem);
1452 ll_stats_ops_tally(ll_i2sbi(inode), (ia_valid & ATTR_SIZE) ?
1453 LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1458 int ll_setattr(struct dentry *de, struct iattr *attr)
1460 int mode = de->d_inode->i_mode;
1462 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1463 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1464 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1466 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1467 (ATTR_SIZE|ATTR_MODE)) &&
1468 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1469 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1470 !(attr->ia_mode & S_ISGID))))
1471 attr->ia_valid |= ATTR_FORCE;
1473 if ((mode & S_ISUID) &&
1474 !(attr->ia_mode & S_ISUID) &&
1475 !(attr->ia_valid & ATTR_KILL_SUID))
1476 attr->ia_valid |= ATTR_KILL_SUID;
1478 if (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1479 !(attr->ia_mode & S_ISGID) &&
1480 !(attr->ia_valid & ATTR_KILL_SGID))
1481 attr->ia_valid |= ATTR_KILL_SGID;
1483 return ll_setattr_raw(de, attr);
1486 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1487 __u64 max_age, __u32 flags)
1489 struct ll_sb_info *sbi = ll_s2sbi(sb);
1490 struct obd_statfs obd_osfs;
1494 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1496 CERROR("md_statfs fails: rc = %d\n", rc);
1500 osfs->os_type = sb->s_magic;
1502 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1503 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1505 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1506 flags |= OBD_STATFS_NODELAY;
1508 rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1510 CERROR("obd_statfs fails: rc = %d\n", rc);
1514 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1515 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1518 osfs->os_bsize = obd_osfs.os_bsize;
1519 osfs->os_blocks = obd_osfs.os_blocks;
1520 osfs->os_bfree = obd_osfs.os_bfree;
1521 osfs->os_bavail = obd_osfs.os_bavail;
1523 /* If we don't have as many objects free on the OST as inodes
1524 * on the MDS, we reduce the total number of inodes to
1525 * compensate, so that the "inodes in use" number is correct.
1527 if (obd_osfs.os_ffree < osfs->os_ffree) {
1528 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1530 osfs->os_ffree = obd_osfs.os_ffree;
1535 #ifndef HAVE_STATFS_DENTRY_PARAM
1536 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1539 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1541 struct super_block *sb = de->d_sb;
1543 struct obd_statfs osfs;
1546 CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1547 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1549 /* Some amount of caching on the client is allowed */
1550 rc = ll_statfs_internal(sb, &osfs,
1551 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1556 statfs_unpack(sfs, &osfs);
1558 /* We need to downshift for all 32-bit kernels, because we can't
1559 * tell if the kernel is being called via sys_statfs64() or not.
1560 * Stop before overflowing f_bsize - in which case it is better
1561 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1562 if (sizeof(long) < 8) {
1563 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1566 osfs.os_blocks >>= 1;
1567 osfs.os_bfree >>= 1;
1568 osfs.os_bavail >>= 1;
1572 sfs->f_blocks = osfs.os_blocks;
1573 sfs->f_bfree = osfs.os_bfree;
1574 sfs->f_bavail = osfs.os_bavail;
1579 void ll_inode_size_lock(struct inode *inode, int lock_lsm)
1581 struct ll_inode_info *lli;
1582 struct lov_stripe_md *lsm;
1584 LASSERT(!S_ISDIR(inode->i_mode));
1586 lli = ll_i2info(inode);
1587 LASSERT(lli->lli_size_sem_owner != current);
1588 cfs_down(&lli->lli_size_sem);
1589 LASSERT(lli->lli_size_sem_owner == NULL);
1590 lli->lli_size_sem_owner = current;
1592 LASSERTF(lsm != NULL || lock_lsm == 0, "lsm %p, lock_lsm %d\n",
1595 lov_stripe_lock(lsm);
1598 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm)
1600 struct ll_inode_info *lli;
1601 struct lov_stripe_md *lsm;
1603 lli = ll_i2info(inode);
1605 LASSERTF(lsm != NULL || unlock_lsm == 0, "lsm %p, lock_lsm %d\n",
1608 lov_stripe_unlock(lsm);
1609 LASSERT(lli->lli_size_sem_owner == current);
1610 lli->lli_size_sem_owner = NULL;
1611 cfs_up(&lli->lli_size_sem);
1614 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1616 struct ll_inode_info *lli = ll_i2info(inode);
1617 struct mdt_body *body = md->body;
1618 struct lov_stripe_md *lsm = md->lsm;
1619 struct ll_sb_info *sbi = ll_i2sbi(inode);
1621 LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1623 LASSERT(S_ISREG(inode->i_mode));
1625 cfs_mutex_lock(&lli->lli_och_mutex);
1626 if (lli->lli_smd == NULL) {
1627 if (lsm->lsm_magic != LOV_MAGIC_V1 &&
1628 lsm->lsm_magic != LOV_MAGIC_V3) {
1629 dump_lsm(D_ERROR, lsm);
1632 CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1633 lsm, inode->i_ino, inode->i_generation, inode);
1634 /* cl_file_inode_init must go before lli_smd or a race
1635 * is possible where client thinks the file has stripes,
1636 * but lov raid0 is not setup yet and parallel e.g.
1637 * glimpse would try to use uninitialized lov */
1638 cl_file_inode_init(inode, md);
1639 cfs_spin_lock(&lli->lli_lock);
1641 cfs_spin_unlock(&lli->lli_lock);
1642 cfs_mutex_unlock(&lli->lli_och_mutex);
1643 lli->lli_maxbytes = lsm->lsm_maxbytes;
1644 if (lli->lli_maxbytes > MAX_LFS_FILESIZE)
1645 lli->lli_maxbytes = MAX_LFS_FILESIZE;
1647 cfs_mutex_unlock(&lli->lli_och_mutex);
1648 LASSERT(lli->lli_smd->lsm_magic == lsm->lsm_magic &&
1649 lli->lli_smd->lsm_stripe_count ==
1650 lsm->lsm_stripe_count);
1651 if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
1652 CERROR("lsm mismatch for inode %ld\n",
1654 CERROR("lli_smd:\n");
1655 dump_lsm(D_ERROR, lli->lli_smd);
1657 dump_lsm(D_ERROR, lsm);
1661 if (lli->lli_smd != lsm)
1662 obd_free_memmd(ll_i2dtexp(inode), &lsm);
1665 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1666 if (body->valid & OBD_MD_FLRMTPERM)
1667 ll_update_remote_perm(inode, md->remote_perm);
1669 #ifdef CONFIG_FS_POSIX_ACL
1670 else if (body->valid & OBD_MD_FLACL) {
1671 cfs_spin_lock(&lli->lli_lock);
1672 if (lli->lli_posix_acl)
1673 posix_acl_release(lli->lli_posix_acl);
1674 lli->lli_posix_acl = md->posix_acl;
1675 cfs_spin_unlock(&lli->lli_lock);
1678 inode->i_ino = cl_fid_build_ino(&body->fid1, 0);
1679 inode->i_generation = cl_fid_build_gen(&body->fid1);
1681 if (body->valid & OBD_MD_FLATIME) {
1682 if (body->atime > LTIME_S(inode->i_atime))
1683 LTIME_S(inode->i_atime) = body->atime;
1684 lli->lli_lvb.lvb_atime = body->atime;
1686 if (body->valid & OBD_MD_FLMTIME) {
1687 if (body->mtime > LTIME_S(inode->i_mtime)) {
1688 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1689 "to "LPU64"\n", inode->i_ino,
1690 LTIME_S(inode->i_mtime), body->mtime);
1691 LTIME_S(inode->i_mtime) = body->mtime;
1693 lli->lli_lvb.lvb_mtime = body->mtime;
1695 if (body->valid & OBD_MD_FLCTIME) {
1696 if (body->ctime > LTIME_S(inode->i_ctime))
1697 LTIME_S(inode->i_ctime) = body->ctime;
1698 lli->lli_lvb.lvb_ctime = body->ctime;
1700 if (body->valid & OBD_MD_FLMODE)
1701 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1702 if (body->valid & OBD_MD_FLTYPE)
1703 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1704 LASSERT(inode->i_mode != 0);
1705 if (S_ISREG(inode->i_mode)) {
1706 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1708 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1710 #ifdef HAVE_INODE_BLKSIZE
1711 inode->i_blksize = 1<<inode->i_blkbits;
1713 if (body->valid & OBD_MD_FLUID)
1714 inode->i_uid = body->uid;
1715 if (body->valid & OBD_MD_FLGID)
1716 inode->i_gid = body->gid;
1717 if (body->valid & OBD_MD_FLFLAGS)
1718 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1719 if (body->valid & OBD_MD_FLNLINK)
1720 inode->i_nlink = body->nlink;
1721 if (body->valid & OBD_MD_FLRDEV)
1722 inode->i_rdev = old_decode_dev(body->rdev);
1724 if (body->valid & OBD_MD_FLID) {
1725 /* FID shouldn't be changed! */
1726 if (fid_is_sane(&lli->lli_fid)) {
1727 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1728 "Trying to change FID "DFID
1729 " to the "DFID", inode %lu/%u(%p)\n",
1730 PFID(&lli->lli_fid), PFID(&body->fid1),
1731 inode->i_ino, inode->i_generation, inode);
1733 lli->lli_fid = body->fid1;
1736 LASSERT(fid_seq(&lli->lli_fid) != 0);
1738 if (body->valid & OBD_MD_FLSIZE) {
1739 if (exp_connect_som(ll_i2mdexp(inode)) &&
1740 S_ISREG(inode->i_mode) && lli->lli_smd) {
1741 struct lustre_handle lockh;
1744 /* As it is possible a blocking ast has been processed
1745 * by this time, we need to check there is an UPDATE
1746 * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1748 mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1751 if (lli->lli_flags & (LLIF_DONE_WRITING |
1752 LLIF_EPOCH_PENDING |
1754 CERROR("ino %lu flags %u still has "
1755 "size authority! do not trust "
1756 "the size got from MDS\n",
1757 inode->i_ino, lli->lli_flags);
1759 /* Use old size assignment to avoid
1760 * deadlock bz14138 & bz14326 */
1761 i_size_write(inode, body->size);
1762 lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1764 ldlm_lock_decref(&lockh, mode);
1767 /* Use old size assignment to avoid
1768 * deadlock bz14138 & bz14326 */
1769 i_size_write(inode, body->size);
1771 CDEBUG(D_VFSTRACE, "inode=%lu, updating i_size %llu\n",
1772 inode->i_ino, (unsigned long long)body->size);
1775 if (body->valid & OBD_MD_FLBLOCKS)
1776 inode->i_blocks = body->blocks;
1779 if (body->valid & OBD_MD_FLMDSCAPA) {
1780 LASSERT(md->mds_capa);
1781 ll_add_capa(inode, md->mds_capa);
1783 if (body->valid & OBD_MD_FLOSSCAPA) {
1784 LASSERT(md->oss_capa);
1785 ll_add_capa(inode, md->oss_capa);
1789 void ll_read_inode2(struct inode *inode, void *opaque)
1791 struct lustre_md *md = opaque;
1792 struct ll_inode_info *lli = ll_i2info(inode);
1795 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1796 PFID(&lli->lli_fid), inode);
1798 LASSERT(!lli->lli_smd);
1800 /* Core attributes from the MDS first. This is a new inode, and
1801 * the VFS doesn't zero times in the core inode so we have to do
1802 * it ourselves. They will be overwritten by either MDS or OST
1803 * attributes - we just need to make sure they aren't newer. */
1804 LTIME_S(inode->i_mtime) = 0;
1805 LTIME_S(inode->i_atime) = 0;
1806 LTIME_S(inode->i_ctime) = 0;
1808 ll_update_inode(inode, md);
1810 /* OIDEBUG(inode); */
1812 /* initializing backing dev info. */
1813 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1816 if (S_ISREG(inode->i_mode)) {
1817 struct ll_sb_info *sbi = ll_i2sbi(inode);
1818 inode->i_op = &ll_file_inode_operations;
1819 inode->i_fop = sbi->ll_fop;
1820 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1822 } else if (S_ISDIR(inode->i_mode)) {
1823 inode->i_op = &ll_dir_inode_operations;
1824 inode->i_fop = &ll_dir_operations;
1825 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_dir_aops;
1827 } else if (S_ISLNK(inode->i_mode)) {
1828 inode->i_op = &ll_fast_symlink_inode_operations;
1831 inode->i_op = &ll_special_inode_operations;
1833 init_special_inode(inode, inode->i_mode,
1834 kdev_t_to_nr(inode->i_rdev));
1840 void ll_delete_inode(struct inode *inode)
1842 struct cl_inode_info *lli = cl_i2info(inode);
1843 struct ll_sb_info *sbi = ll_i2sbi(inode);
1847 rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
1849 CERROR("fid_delete() failed, rc %d\n", rc);
1851 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
1852 /* discard all dirty pages before truncating them, required by
1853 * osc_extent implementation at LU-1030. */
1854 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_DISCARD);
1856 truncate_inode_pages(&inode->i_data, 0);
1858 /* Workaround for LU-118 */
1859 if (inode->i_data.nrpages) {
1860 TREE_READ_LOCK_IRQ(&inode->i_data);
1861 TREE_READ_UNLOCK_IRQ(&inode->i_data);
1862 LASSERTF(inode->i_data.nrpages == 0,
1863 "inode=%lu/%u(%p) nrpages=%lu, see "
1864 "http://jira.whamcloud.com/browse/LU-118\n",
1865 inode->i_ino, inode->i_generation, inode,
1866 inode->i_data.nrpages);
1868 /* Workaround end */
1870 #ifdef HAVE_SBOPS_EVICT_INODE
1871 ll_clear_inode(inode);
1872 end_writeback(inode);
1880 int ll_iocontrol(struct inode *inode, struct file *file,
1881 unsigned int cmd, unsigned long arg)
1883 struct ll_sb_info *sbi = ll_i2sbi(inode);
1884 struct ptlrpc_request *req = NULL;
1889 case FSFILT_IOC_GETFLAGS: {
1890 struct mdt_body *body;
1891 struct md_op_data *op_data;
1893 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
1894 0, 0, LUSTRE_OPC_ANY,
1896 if (IS_ERR(op_data))
1897 RETURN(PTR_ERR(op_data));
1899 op_data->op_valid = OBD_MD_FLFLAGS;
1900 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
1901 ll_finish_md_op_data(op_data);
1903 CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1907 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1909 flags = body->flags;
1911 ptlrpc_req_finished(req);
1913 RETURN(put_user(flags, (int *)arg));
1915 case FSFILT_IOC_SETFLAGS: {
1916 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1917 struct obd_info oinfo = { { { 0 } } };
1918 struct md_op_data *op_data;
1920 if (get_user(flags, (int *)arg))
1924 OBDO_ALLOC(oinfo.oi_oa);
1928 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1929 LUSTRE_OPC_ANY, NULL);
1930 if (IS_ERR(op_data))
1931 RETURN(PTR_ERR(op_data));
1933 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1934 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1935 rc = md_setattr(sbi->ll_md_exp, op_data,
1936 NULL, 0, NULL, 0, &req, NULL);
1937 ll_finish_md_op_data(op_data);
1938 ptlrpc_req_finished(req);
1940 OBDO_FREE(oinfo.oi_oa);
1945 OBDO_FREE(oinfo.oi_oa);
1946 GOTO(update_cache, rc);
1949 oinfo.oi_oa->o_id = lsm->lsm_object_id;
1950 oinfo.oi_oa->o_seq = lsm->lsm_object_seq;
1951 oinfo.oi_oa->o_flags = flags;
1952 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1954 oinfo.oi_capa = ll_mdscapa_get(inode);
1955 obdo_set_parent_fid(oinfo.oi_oa, &ll_i2info(inode)->lli_fid);
1956 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1957 capa_put(oinfo.oi_capa);
1958 OBDO_FREE(oinfo.oi_oa);
1960 if (rc != -EPERM && rc != -EACCES)
1961 CERROR("osc_setattr_async fails: rc = %d\n",rc);
1967 inode->i_flags = ll_ext_to_inode_flags(flags);
1977 int ll_flush_ctx(struct inode *inode)
1979 struct ll_sb_info *sbi = ll_i2sbi(inode);
1981 CDEBUG(D_SEC, "flush context for user %d\n", cfs_curproc_uid());
1983 obd_set_info_async(NULL, sbi->ll_md_exp,
1984 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1986 obd_set_info_async(NULL, sbi->ll_dt_exp,
1987 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1992 /* umount -f client means force down, don't save state */
1993 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1994 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
1996 struct super_block *sb = vfsmnt->mnt_sb;
1998 void ll_umount_begin(struct super_block *sb)
2001 struct lustre_sb_info *lsi = s2lsi(sb);
2002 struct ll_sb_info *sbi = ll_s2sbi(sb);
2003 struct obd_device *obd;
2004 struct obd_ioctl_data *ioc_data;
2007 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2008 if (!(flags & MNT_FORCE)) {
2014 /* Tell the MGC we got umount -f */
2015 lsi->lsi_flags |= LSI_UMOUNT_FORCE;
2017 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2018 sb->s_count, atomic_read(&sb->s_active));
2020 obd = class_exp2obd(sbi->ll_md_exp);
2022 CERROR("Invalid MDC connection handle "LPX64"\n",
2023 sbi->ll_md_exp->exp_handle.h_cookie);
2029 obd = class_exp2obd(sbi->ll_dt_exp);
2031 CERROR("Invalid LOV connection handle "LPX64"\n",
2032 sbi->ll_dt_exp->exp_handle.h_cookie);
2038 OBD_ALLOC_PTR(ioc_data);
2040 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2041 sizeof ioc_data, ioc_data, NULL);
2043 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2044 sizeof ioc_data, ioc_data, NULL);
2046 OBD_FREE_PTR(ioc_data);
2050 /* Really, we'd like to wait until there are no requests outstanding,
2051 * and then continue. For now, we just invalidate the requests,
2052 * schedule() and sleep one second if needed, and hope.
2055 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2056 if (atomic_read(&vfsmnt->mnt_count) > 2) {
2057 cfs_schedule_timeout_and_set_state(CFS_TASK_INTERRUPTIBLE,
2058 cfs_time_seconds(1));
2059 if (atomic_read(&vfsmnt->mnt_count) > 2)
2060 LCONSOLE_WARN("Mount still busy with %d refs! You "
2061 "may try to umount it a bit later\n",
2062 atomic_read(&vfsmnt->mnt_count));
2069 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2071 struct ll_sb_info *sbi = ll_s2sbi(sb);
2072 char *profilenm = get_profile_name(sb);
2076 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2077 read_only = *flags & MS_RDONLY;
2078 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2079 sizeof(KEY_READ_ONLY),
2080 KEY_READ_ONLY, sizeof(read_only),
2083 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2084 profilenm, read_only ?
2085 "read-only" : "read-write", err);
2090 sb->s_flags |= MS_RDONLY;
2092 sb->s_flags &= ~MS_RDONLY;
2094 if (sbi->ll_flags & LL_SBI_VERBOSE)
2095 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2096 read_only ? "read-only" : "read-write");
2101 int ll_prep_inode(struct inode **inode,
2102 struct ptlrpc_request *req,
2103 struct super_block *sb)
2105 struct ll_sb_info *sbi = NULL;
2106 struct lustre_md md;
2110 LASSERT(*inode || sb);
2111 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2112 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2113 sbi->ll_md_exp, &md);
2118 ll_update_inode(*inode, &md);
2120 LASSERT(sb != NULL);
2123 * At this point server returns to client's same fid as client
2124 * generated for creating. So using ->fid1 is okay here.
2126 LASSERT(fid_is_sane(&md.body->fid1));
2128 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1, 0), &md);
2129 if (*inode == NULL || IS_ERR(*inode)) {
2131 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2132 #ifdef CONFIG_FS_POSIX_ACL
2134 posix_acl_release(md.posix_acl);
2135 md.posix_acl = NULL;
2138 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2140 CERROR("new_inode -fatal: rc %d\n", rc);
2146 md_free_lustre_md(sbi->ll_md_exp, &md);
2150 int ll_obd_statfs(struct inode *inode, void *arg)
2152 struct ll_sb_info *sbi = NULL;
2153 struct obd_export *exp;
2155 struct obd_ioctl_data *data = NULL;
2159 if (!inode || !(sbi = ll_i2sbi(inode)))
2160 GOTO(out_statfs, rc = -EINVAL);
2162 rc = obd_ioctl_getdata(&buf, &len, arg);
2164 GOTO(out_statfs, rc);
2167 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2168 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2169 GOTO(out_statfs, rc = -EINVAL);
2171 if (data->ioc_inllen1 != sizeof(__u32) ||
2172 data->ioc_inllen2 != sizeof(__u32) ||
2173 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2174 data->ioc_plen2 != sizeof(struct obd_uuid))
2175 GOTO(out_statfs, rc = -EINVAL);
2177 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2178 if (type == LL_STATFS_LMV)
2179 exp = sbi->ll_md_exp;
2180 else if (type == LL_STATFS_LOV)
2181 exp = sbi->ll_dt_exp;
2183 GOTO(out_statfs, rc = -ENODEV);
2185 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2187 GOTO(out_statfs, rc);
2190 obd_ioctl_freedata(buf, len);
2194 int ll_process_config(struct lustre_cfg *lcfg)
2198 struct lprocfs_static_vars lvars;
2202 lprocfs_llite_init_vars(&lvars);
2204 /* The instance name contains the sb: lustre-client-aacfe000 */
2205 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2206 if (!ptr || !*(++ptr))
2208 if (sscanf(ptr, "%lx", &x) != 1)
2211 /* This better be a real Lustre superblock! */
2212 LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2214 /* Note we have not called client_common_fill_super yet, so
2215 proc fns must be able to handle that! */
2216 rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2223 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2224 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2225 struct inode *i1, struct inode *i2,
2226 const char *name, int namelen,
2227 int mode, __u32 opc, void *data)
2229 LASSERT(i1 != NULL);
2231 if (namelen > ll_i2sbi(i1)->ll_namelen)
2232 return ERR_PTR(-ENAMETOOLONG);
2234 if (op_data == NULL)
2235 OBD_ALLOC_PTR(op_data);
2237 if (op_data == NULL)
2238 return ERR_PTR(-ENOMEM);
2240 ll_i2gids(op_data->op_suppgids, i1, i2);
2241 op_data->op_fid1 = *ll_inode2fid(i1);
2242 op_data->op_capa1 = ll_mdscapa_get(i1);
2245 op_data->op_fid2 = *ll_inode2fid(i2);
2246 op_data->op_capa2 = ll_mdscapa_get(i2);
2248 fid_zero(&op_data->op_fid2);
2249 op_data->op_capa2 = NULL;
2252 op_data->op_name = name;
2253 op_data->op_namelen = namelen;
2254 op_data->op_mode = mode;
2255 op_data->op_mod_time = cfs_time_current_sec();
2256 op_data->op_fsuid = cfs_curproc_fsuid();
2257 op_data->op_fsgid = cfs_curproc_fsgid();
2258 op_data->op_cap = cfs_curproc_cap_pack();
2259 op_data->op_bias = MDS_CHECK_SPLIT;
2260 op_data->op_opc = opc;
2261 op_data->op_mds = 0;
2262 op_data->op_data = data;
2264 /* If the file is being opened after mknod() (normally due to NFS)
2265 * try to use the default stripe data from parent directory for
2266 * allocating OST objects. Try to pass the parent FID to MDS. */
2267 if (opc == LUSTRE_OPC_CREATE && i1 == i2 && S_ISREG(i2->i_mode) &&
2268 ll_i2info(i2)->lli_smd == NULL) {
2269 struct ll_inode_info *lli = ll_i2info(i2);
2271 cfs_spin_lock(&lli->lli_lock);
2272 if (likely(lli->lli_smd == NULL &&
2273 !fid_is_zero(&lli->lli_pfid)))
2274 op_data->op_fid1 = lli->lli_pfid;
2275 cfs_spin_unlock(&lli->lli_lock);
2276 /** We ignore parent's capability temporary. */
2282 void ll_finish_md_op_data(struct md_op_data *op_data)
2284 capa_put(op_data->op_capa1);
2285 capa_put(op_data->op_capa2);
2286 OBD_FREE_PTR(op_data);
2289 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2291 struct ll_sb_info *sbi;
2293 LASSERT((seq != NULL) && (vfs != NULL));
2294 sbi = ll_s2sbi(vfs->mnt_sb);
2296 if (sbi->ll_flags & LL_SBI_NOLCK)
2297 seq_puts(seq, ",nolock");
2299 if (sbi->ll_flags & LL_SBI_FLOCK)
2300 seq_puts(seq, ",flock");
2302 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2303 seq_puts(seq, ",localflock");
2305 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2306 seq_puts(seq, ",user_xattr");
2308 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2309 seq_puts(seq, ",lazystatfs");
2315 * Get obd name by cmd, and copy out to user space
2317 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2319 struct ll_sb_info *sbi = ll_i2sbi(inode);
2320 struct obd_device *obd;
2323 if (cmd == OBD_IOC_GETDTNAME)
2324 obd = class_exp2obd(sbi->ll_dt_exp);
2325 else if (cmd == OBD_IOC_GETMDNAME)
2326 obd = class_exp2obd(sbi->ll_md_exp);
2333 if (cfs_copy_to_user((void *)arg, obd->obd_name,
2334 strlen(obd->obd_name) + 1))