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 DEFINE_SPINLOCK(ll_sb_lock);
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 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK) {
376 LCONSOLE_INFO("Layout lock feature supported.\n");
377 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
380 obd = class_name2obd(dt);
382 CERROR("DT %s: not setup or attached\n", dt);
383 GOTO(out_md_fid, err = -ENODEV);
386 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
387 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
388 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
389 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
390 OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
391 OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
392 OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
393 OBD_CONNECT_MAXBYTES |
394 OBD_CONNECT_EINPROGRESS |
395 OBD_CONNECT_JOBSTATS;
397 if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
398 data->ocd_connect_flags |= OBD_CONNECT_SOM;
400 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
401 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
402 * disabled by default, because it can still be enabled on the
403 * fly via /proc. As a consequence, we still need to come to an
404 * agreement on the supported algorithms at connect time */
405 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
407 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
408 data->ocd_cksum_types = OBD_CKSUM_ADLER;
410 data->ocd_cksum_types = cksum_types_supported_client();
413 #ifdef HAVE_LRU_RESIZE_SUPPORT
414 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
416 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
417 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
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 = cl_ocd_update;
426 data->ocd_brw_size = PTLRPC_MAX_BRW_SIZE;
428 err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data, NULL);
430 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
431 "recovery, of which this client is not a "
432 "part. Please wait for recovery to "
433 "complete, abort, or time out.\n", dt);
434 GOTO(out_md_fid, err);
436 CERROR("Cannot connect to %s: rc = %d\n", dt, err);
437 GOTO(out_md_fid, err);
440 err = obd_fid_init(sbi->ll_dt_exp);
442 CERROR("Can't init data layer FID infrastructure, "
447 cfs_mutex_lock(&sbi->ll_lco.lco_lock);
448 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
449 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
450 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
451 cfs_mutex_unlock(&sbi->ll_lco.lco_lock);
453 fid_zero(&sbi->ll_root_fid);
454 err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
456 CERROR("cannot mds_connect: rc = %d\n", err);
457 GOTO(out_lock_cn_cb, err);
459 if (!fid_is_sane(&sbi->ll_root_fid)) {
460 CERROR("Invalid root fid during mount\n");
461 GOTO(out_lock_cn_cb, err = -EINVAL);
463 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
465 sb->s_op = &lustre_super_operations;
466 #if THREAD_SIZE >= 8192 /*b=17630*/ && !defined(HAVE_FSTYPE_MOUNT) /*LU-812*/
467 sb->s_export_op = &lustre_export_operations;
471 * XXX: move this to after cbd setup? */
472 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
473 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
474 valid |= OBD_MD_FLRMTPERM;
475 else if (sbi->ll_flags & LL_SBI_ACL)
476 valid |= OBD_MD_FLACL;
478 OBD_ALLOC_PTR(op_data);
480 GOTO(out_lock_cn_cb, err = -ENOMEM);
482 op_data->op_fid1 = sbi->ll_root_fid;
483 op_data->op_mode = 0;
484 op_data->op_capa1 = oc;
485 op_data->op_valid = valid;
487 err = md_getattr(sbi->ll_md_exp, op_data, &request);
490 OBD_FREE_PTR(op_data);
492 CERROR("md_getattr failed for root: rc = %d\n", err);
493 GOTO(out_lock_cn_cb, err);
495 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
496 sbi->ll_md_exp, &lmd);
498 CERROR("failed to understand root inode md: rc = %d\n", err);
499 ptlrpc_req_finished (request);
500 GOTO(out_lock_cn_cb, err);
503 LASSERT(fid_is_sane(&sbi->ll_root_fid));
504 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid, 0), &lmd);
505 md_free_lustre_md(sbi->ll_md_exp, &lmd);
506 ptlrpc_req_finished(request);
508 if (root == NULL || IS_ERR(root)) {
510 obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
511 #ifdef CONFIG_FS_POSIX_ACL
513 posix_acl_release(lmd.posix_acl);
514 lmd.posix_acl = NULL;
517 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
519 CERROR("lustre_lite: bad iget4 for root\n");
523 err = ll_close_thread_start(&sbi->ll_lcq);
525 CERROR("cannot start close thread: rc %d\n", err);
529 #ifdef CONFIG_FS_POSIX_ACL
530 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
531 rct_init(&sbi->ll_rct);
532 et_init(&sbi->ll_et);
536 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
537 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
538 KEY_CHECKSUM, sizeof(checksum), &checksum,
542 sb->s_root = d_alloc_root(root);
543 #ifdef HAVE_DCACHE_LOCK
544 sb->s_root->d_op = &ll_d_root_ops;
546 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
547 d_set_d_op(sb->s_root, &ll_d_root_ops);
548 sb->s_d_op = &ll_d_ops;
551 sbi->ll_sdev_orig = sb->s_dev;
553 /* We set sb->s_dev equal on all lustre clients in order to support
554 * NFS export clustering. NFSD requires that the FSID be the same
556 /* s_dev is also used in lt_compare() to compare two fs, but that is
557 * only a node-local comparison. */
558 uuid = obd_get_uuid(sbi->ll_md_exp);
560 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
573 obd_fid_fini(sbi->ll_dt_exp);
575 obd_disconnect(sbi->ll_dt_exp);
576 sbi->ll_dt_exp = NULL;
578 obd_fid_fini(sbi->ll_md_exp);
580 obd_disconnect(sbi->ll_md_exp);
581 sbi->ll_md_exp = NULL;
587 lprocfs_unregister_mountpoint(sbi);
591 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
595 *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
597 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
598 KEY_MAX_EASIZE, &size, lmmsize, NULL);
600 CERROR("Get max mdsize error rc %d \n", rc);
605 void ll_dump_inode(struct inode *inode)
607 struct list_head *tmp;
608 int dentry_count = 0;
610 LASSERT(inode != NULL);
612 list_for_each(tmp, &inode->i_dentry)
615 CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
616 inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
617 inode->i_mode, atomic_read(&inode->i_count), dentry_count);
620 void lustre_dump_dentry(struct dentry *dentry, int recur)
622 struct list_head *tmp;
625 LASSERT(dentry != NULL);
627 list_for_each(tmp, &dentry->d_subdirs)
630 CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
631 " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
632 dentry->d_name.len, dentry->d_name.name,
633 dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
634 dentry->d_parent, dentry->d_inode, d_refcount(dentry),
635 dentry->d_flags, dentry->d_fsdata, subdirs);
636 if (dentry->d_inode != NULL)
637 ll_dump_inode(dentry->d_inode);
642 list_for_each(tmp, &dentry->d_subdirs) {
643 struct dentry *d = list_entry(tmp, struct dentry, d_u.d_child);
644 lustre_dump_dentry(d, recur - 1);
648 void client_common_put_super(struct super_block *sb)
650 struct ll_sb_info *sbi = ll_s2sbi(sb);
653 #ifdef CONFIG_FS_POSIX_ACL
654 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
655 et_fini(&sbi->ll_et);
656 rct_fini(&sbi->ll_rct);
660 obd_cancel_unused(sbi->ll_dt_exp, NULL, 0, NULL);
662 ll_close_thread_shutdown(sbi->ll_lcq);
666 cfs_list_del(&sbi->ll_conn_chain);
668 obd_fid_fini(sbi->ll_dt_exp);
669 obd_disconnect(sbi->ll_dt_exp);
670 sbi->ll_dt_exp = NULL;
672 lprocfs_unregister_mountpoint(sbi);
674 obd_fid_fini(sbi->ll_md_exp);
675 obd_disconnect(sbi->ll_md_exp);
676 sbi->ll_md_exp = NULL;
681 void ll_kill_super(struct super_block *sb)
683 struct ll_sb_info *sbi;
688 if (!(sb->s_flags & MS_ACTIVE))
692 /* we need restore s_dev from changed for clustred NFS before put_super
693 * because new kernels have cached s_dev and change sb->s_dev in
694 * put_super not affected real removing devices */
696 sb->s_dev = sbi->ll_sdev_orig;
700 char *ll_read_opt(const char *opt, char *data)
706 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
707 if (strncmp(opt, data, strlen(opt)))
709 if ((value = strchr(data, '=')) == NULL)
713 OBD_ALLOC(retval, strlen(value) + 1);
715 CERROR("out of memory!\n");
719 memcpy(retval, value, strlen(value)+1);
720 CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
724 static inline int ll_set_opt(const char *opt, char *data, int fl)
726 if (strncmp(opt, data, strlen(opt)) != 0)
732 /* non-client-specific mount options are parsed in lmd_parse */
733 static int ll_options(char *options, int *flags)
736 char *s1 = options, *s2;
742 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
745 CDEBUG(D_SUPER, "next opt=%s\n", s1);
746 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
751 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
756 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
761 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
766 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
771 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
776 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 5, 50, 0)
777 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
779 /* Ignore deprecated mount option. The client will
780 * always try to mount with ACL support, whether this
781 * is used depends on whether server supports it. */
782 LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
783 "mount option 'acl'.\n");
786 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
788 LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
789 "mount option 'noacl'.\n");
793 #warning "{no}acl options have been deprecated since 1.8, please remove them"
795 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
801 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
806 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
811 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
816 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
821 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
826 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
831 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
836 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
841 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
846 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
851 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
857 s2 = strchr(s1, ',');
865 void ll_lli_init(struct ll_inode_info *lli)
867 lli->lli_inode_magic = LLI_INODE_MAGIC;
869 lli->lli_ioepoch = 0;
870 lli->lli_maxbytes = MAX_LFS_FILESIZE;
871 cfs_spin_lock_init(&lli->lli_lock);
872 lli->lli_posix_acl = NULL;
873 lli->lli_remote_perms = NULL;
874 cfs_mutex_init(&lli->lli_rmtperm_mutex);
875 /* Do not set lli_fid, it has been initialized already. */
876 fid_zero(&lli->lli_pfid);
877 CFS_INIT_LIST_HEAD(&lli->lli_close_list);
878 CFS_INIT_LIST_HEAD(&lli->lli_oss_capas);
879 cfs_atomic_set(&lli->lli_open_count, 0);
880 lli->lli_mds_capa = NULL;
881 lli->lli_rmtperm_time = 0;
882 lli->lli_pending_och = NULL;
883 lli->lli_mds_read_och = NULL;
884 lli->lli_mds_write_och = NULL;
885 lli->lli_mds_exec_och = NULL;
886 lli->lli_open_fd_read_count = 0;
887 lli->lli_open_fd_write_count = 0;
888 lli->lli_open_fd_exec_count = 0;
889 cfs_mutex_init(&lli->lli_och_mutex);
890 cfs_spin_lock_init(&lli->lli_agl_lock);
891 lli->lli_has_smd = false;
892 lli->lli_clob = NULL;
894 LASSERT(lli->lli_vfs_inode.i_mode != 0);
895 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
896 cfs_mutex_init(&lli->lli_readdir_mutex);
897 lli->lli_opendir_key = NULL;
900 lli->lli_def_acl = NULL;
901 cfs_spin_lock_init(&lli->lli_sa_lock);
902 lli->lli_opendir_pid = 0;
904 cfs_sema_init(&lli->lli_size_sem, 1);
905 lli->lli_size_sem_owner = NULL;
906 lli->lli_symlink_name = NULL;
907 cfs_init_rwsem(&lli->lli_trunc_sem);
908 cfs_mutex_init(&lli->lli_write_mutex);
909 cfs_init_rwsem(&lli->lli_glimpse_sem);
910 lli->lli_glimpse_time = 0;
911 CFS_INIT_LIST_HEAD(&lli->lli_agl_list);
912 lli->lli_agl_index = 0;
913 lli->lli_async_rc = 0;
915 cfs_mutex_init(&lli->lli_layout_mutex);
918 static inline int ll_bdi_register(struct backing_dev_info *bdi)
920 #ifdef HAVE_BDI_REGISTER
921 static atomic_t ll_bdi_num = ATOMIC_INIT(0);
924 bdi->name = "lustre";
926 return bdi_register(bdi, NULL, "lustre-%d",
927 atomic_inc_return(&ll_bdi_num));
933 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
935 struct lustre_profile *lprof = NULL;
936 struct lustre_sb_info *lsi = s2lsi(sb);
937 struct ll_sb_info *sbi;
938 char *dt = NULL, *md = NULL;
939 char *profilenm = get_profile_name(sb);
940 struct config_llog_instance *cfg;
941 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
942 const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
946 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
954 /* client additional sb info */
955 lsi->lsi_llsbi = sbi = ll_init_sbi();
957 cfs_module_put(THIS_MODULE);
962 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
966 err = ll_bdi_init(&lsi->lsi_bdi);
969 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
970 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
971 err = ll_bdi_register(&lsi->lsi_bdi);
976 sb->s_bdi = &lsi->lsi_bdi;
979 /* Generate a string unique to this super, in case some joker tries
980 to mount the same fs at two mount points.
981 Use the address of the super itself.*/
982 cfg->cfg_instance = sb;
983 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
985 /* set up client obds */
986 err = lustre_process_log(sb, profilenm, cfg);
988 CERROR("Unable to process log: %d\n", err);
992 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
993 lprof = class_get_profile(profilenm);
995 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
996 " read from the MGS. Does that filesystem "
997 "exist?\n", profilenm);
998 GOTO(out_free, err = -EINVAL);
1000 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1001 lprof->lp_md, lprof->lp_dt);
1003 OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1005 GOTO(out_free, err = -ENOMEM);
1006 sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1008 OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1010 GOTO(out_free, err = -ENOMEM);
1011 sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1013 /* connections, registrations, sb setup */
1014 err = client_common_fill_super(sb, md, dt, mnt);
1018 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1020 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1023 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1024 LCONSOLE_WARN("Mounted %s\n", profilenm);
1028 } /* ll_fill_super */
1031 void lu_context_keys_dump(void);
1033 void ll_put_super(struct super_block *sb)
1035 struct config_llog_instance cfg;
1036 struct obd_device *obd;
1037 struct lustre_sb_info *lsi = s2lsi(sb);
1038 struct ll_sb_info *sbi = ll_s2sbi(sb);
1039 char *profilenm = get_profile_name(sb);
1040 int force = 1, next;
1043 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1045 ll_print_capa_stat(sbi);
1047 cfg.cfg_instance = sb;
1048 lustre_end_log(sb, profilenm, &cfg);
1050 if (sbi->ll_md_exp) {
1051 obd = class_exp2obd(sbi->ll_md_exp);
1053 force = obd->obd_force;
1056 /* We need to set force before the lov_disconnect in
1057 lustre_common_put_super, since l_d cleans up osc's as well. */
1060 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1062 obd->obd_force = force;
1067 /* Only if client_common_fill_super succeeded */
1068 client_common_put_super(sb);
1072 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1073 class_manual_cleanup(obd);
1076 if (sbi->ll_flags & LL_SBI_VERBOSE)
1077 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1080 class_del_profile(profilenm);
1082 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1083 ll_bdi_destroy(&lsi->lsi_bdi);
1084 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1088 lsi->lsi_llsbi = NULL;
1090 lustre_common_put_super(sb);
1092 cl_env_cache_purge(~0);
1094 cfs_module_put(THIS_MODULE);
1097 } /* client_put_super */
1099 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1101 struct inode *inode = NULL;
1102 /* NOTE: we depend on atomic igrab() -bzzz */
1103 lock_res_and_lock(lock);
1104 if (lock->l_ast_data) {
1105 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1106 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1107 inode = igrab(lock->l_ast_data);
1109 inode = lock->l_ast_data;
1110 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1111 D_WARNING, lock, "l_ast_data %p is "
1112 "bogus: magic %08x", lock->l_ast_data,
1113 lli->lli_inode_magic);
1117 unlock_res_and_lock(lock);
1121 static int null_if_equal(struct ldlm_lock *lock, void *data)
1123 if (data == lock->l_ast_data) {
1124 lock->l_ast_data = NULL;
1126 if (lock->l_req_mode != lock->l_granted_mode)
1127 LDLM_ERROR(lock,"clearing inode with ungranted lock");
1130 return LDLM_ITER_CONTINUE;
1133 void ll_clear_inode(struct inode *inode)
1135 struct ll_inode_info *lli = ll_i2info(inode);
1136 struct ll_sb_info *sbi = ll_i2sbi(inode);
1139 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1140 inode->i_generation, inode);
1142 if (S_ISDIR(inode->i_mode)) {
1143 /* these should have been cleared in ll_file_release */
1144 LASSERT(lli->lli_opendir_key == NULL);
1145 LASSERT(lli->lli_sai == NULL);
1146 LASSERT(lli->lli_opendir_pid == 0);
1149 ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1150 md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1151 null_if_equal, inode);
1153 LASSERT(!lli->lli_open_fd_write_count);
1154 LASSERT(!lli->lli_open_fd_read_count);
1155 LASSERT(!lli->lli_open_fd_exec_count);
1157 if (lli->lli_mds_write_och)
1158 ll_md_real_close(inode, FMODE_WRITE);
1159 if (lli->lli_mds_exec_och)
1160 ll_md_real_close(inode, FMODE_EXEC);
1161 if (lli->lli_mds_read_och)
1162 ll_md_real_close(inode, FMODE_READ);
1164 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1165 OBD_FREE(lli->lli_symlink_name,
1166 strlen(lli->lli_symlink_name) + 1);
1167 lli->lli_symlink_name = NULL;
1170 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1171 LASSERT(lli->lli_posix_acl == NULL);
1172 if (lli->lli_remote_perms) {
1173 free_rmtperm_hash(lli->lli_remote_perms);
1174 lli->lli_remote_perms = NULL;
1177 #ifdef CONFIG_FS_POSIX_ACL
1178 else if (lli->lli_posix_acl) {
1179 LASSERT(cfs_atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1180 LASSERT(lli->lli_remote_perms == NULL);
1181 posix_acl_release(lli->lli_posix_acl);
1182 lli->lli_posix_acl = NULL;
1185 lli->lli_inode_magic = LLI_INODE_DEAD;
1187 ll_clear_inode_capas(inode);
1188 if (!S_ISDIR(inode->i_mode))
1189 LASSERT(cfs_list_empty(&lli->lli_agl_list));
1192 * XXX This has to be done before lsm is freed below, because
1193 * cl_object still uses inode lsm.
1195 cl_inode_fini(inode);
1196 lli->lli_has_smd = false;
1201 int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data,
1202 struct md_open_data **mod)
1204 struct lustre_md md;
1205 struct inode *inode = dentry->d_inode;
1206 struct ll_sb_info *sbi = ll_i2sbi(inode);
1207 struct ptlrpc_request *request = NULL;
1211 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1212 LUSTRE_OPC_ANY, NULL);
1213 if (IS_ERR(op_data))
1214 RETURN(PTR_ERR(op_data));
1216 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1219 ptlrpc_req_finished(request);
1220 if (rc == -ENOENT) {
1222 /* Unlinked special device node? Or just a race?
1223 * Pretend we done everything. */
1224 if (!S_ISREG(inode->i_mode) &&
1225 !S_ISDIR(inode->i_mode)) {
1226 ia_valid = op_data->op_attr.ia_valid;
1227 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1228 rc = simple_setattr(dentry, &op_data->op_attr);
1229 op_data->op_attr.ia_valid = ia_valid;
1231 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1232 CERROR("md_setattr fails: rc = %d\n", rc);
1237 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1238 sbi->ll_md_exp, &md);
1240 ptlrpc_req_finished(request);
1244 /* We want to adjust timestamps.
1245 * If there is at least some data in file, we cleared ATTR_SIZE
1246 * to avoid update size, otherwise it is important to do.(SOM case)
1248 ia_valid = op_data->op_attr.ia_valid;
1249 /* Since we set ATTR_*_SET flags above, and already done permission
1250 * check, So don't let inode_change_ok() check it again. */
1251 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1252 rc = simple_setattr(dentry, &op_data->op_attr);
1253 op_data->op_attr.ia_valid = ia_valid;
1255 /* Extract epoch data if obtained. */
1256 op_data->op_handle = md.body->handle;
1257 op_data->op_ioepoch = md.body->ioepoch;
1259 ll_update_inode(inode, &md);
1260 ptlrpc_req_finished(request);
1265 /* Close IO epoch and send Size-on-MDS attribute update. */
1266 static int ll_setattr_done_writing(struct inode *inode,
1267 struct md_op_data *op_data,
1268 struct md_open_data *mod)
1270 struct ll_inode_info *lli = ll_i2info(inode);
1274 LASSERT(op_data != NULL);
1275 if (!S_ISREG(inode->i_mode))
1278 CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1279 op_data->op_ioepoch, PFID(&lli->lli_fid));
1281 op_data->op_flags = MF_EPOCH_CLOSE;
1282 ll_done_writing_attr(inode, op_data);
1283 ll_pack_inode2opdata(inode, op_data, NULL);
1285 rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1286 if (rc == -EAGAIN) {
1287 /* MDS has instructed us to obtain Size-on-MDS attribute
1288 * from OSTs and send setattr to back to MDS. */
1289 rc = ll_som_update(inode, op_data);
1291 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1297 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1299 struct obd_capa *capa;
1302 if (attr->ia_valid & ATTR_SIZE)
1303 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1305 capa = ll_mdscapa_get(inode);
1307 rc = cl_setattr_ost(inode, attr, capa);
1309 if (attr->ia_valid & ATTR_SIZE)
1310 ll_truncate_free_capa(capa);
1317 #ifndef HAVE_VFS_INODE_NEWSIZE_OK
1319 * inode_newsize_ok - may this inode be truncated to a given size
1320 * @inode: the inode to be truncated
1321 * @offset: the new size to assign to the inode
1322 * @Returns: 0 on success, -ve errno on failure
1324 * inode_newsize_ok will check filesystem limits and ulimits to check that the
1325 * new inode size is within limits. inode_newsize_ok will also send SIGXFSZ
1326 * when necessary. Caller must not proceed with inode size change if failure is
1327 * returned. @inode must be a file (not directory), with appropriate
1328 * permissions to allow truncate (inode_newsize_ok does NOT check these
1331 * inode_newsize_ok must be called with i_mutex held.
1333 int inode_newsize_ok(const struct inode *inode, loff_t offset)
1335 if (inode->i_size < offset) {
1336 unsigned long limit;
1338 limit = rlimit(RLIMIT_FSIZE);
1339 if (limit != RLIM_INFINITY && offset > limit)
1341 if (offset > inode->i_sb->s_maxbytes)
1345 * truncation of in-use swapfiles is disallowed - it would
1346 * cause subsequent swapout to scribble on the now-freed
1349 if (IS_SWAPFILE(inode))
1355 send_sig(SIGXFSZ, current, 0);
1361 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1362 * object(s) determine the file size and mtime. Otherwise, the MDS will
1363 * keep these values until such a time that objects are allocated for it.
1364 * We do the MDS operations first, as it is checking permissions for us.
1365 * We don't to the MDS RPC if there is nothing that we want to store there,
1366 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1367 * going to do an RPC anyways.
1369 * If we are doing a truncate, we will send the mtime and ctime updates
1370 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1371 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1374 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr)
1376 struct lov_stripe_md *lsm;
1377 struct inode *inode = dentry->d_inode;
1378 struct ll_inode_info *lli = ll_i2info(inode);
1379 struct md_op_data *op_data = NULL;
1380 struct md_open_data *mod = NULL;
1381 int ia_valid = attr->ia_valid;
1382 int rc = 0, rc1 = 0;
1385 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1388 if (ia_valid & ATTR_SIZE) {
1389 /* Check new size against VFS/VM file size limit and rlimit */
1390 rc = inode_newsize_ok(inode, attr->ia_size);
1394 /* The maximum Lustre file size is variable, based on the
1395 * OST maximum object size and number of stripes. This
1396 * needs another check in addition to the VFS check above. */
1397 if (attr->ia_size > ll_file_maxbytes(inode)) {
1398 CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1399 PFID(&lli->lli_fid), attr->ia_size,
1400 ll_file_maxbytes(inode));
1404 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1407 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1408 if (ia_valid & TIMES_SET_FLAGS) {
1409 if (cfs_curproc_fsuid() != inode->i_uid &&
1410 !cfs_capable(CFS_CAP_FOWNER))
1414 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1415 if (attr->ia_valid & ATTR_CTIME) {
1416 attr->ia_ctime = CFS_CURRENT_TIME;
1417 attr->ia_valid |= ATTR_CTIME_SET;
1419 if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1420 attr->ia_atime = CFS_CURRENT_TIME;
1421 attr->ia_valid |= ATTR_ATIME_SET;
1423 if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1424 attr->ia_mtime = CFS_CURRENT_TIME;
1425 attr->ia_valid |= ATTR_MTIME_SET;
1428 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1429 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1430 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1431 cfs_time_current_sec());
1433 /* We always do an MDS RPC, even if we're only changing the size;
1434 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1436 OBD_ALLOC_PTR(op_data);
1437 if (op_data == NULL)
1440 if (!S_ISDIR(inode->i_mode)) {
1441 if (ia_valid & ATTR_SIZE)
1442 UP_WRITE_I_ALLOC_SEM(inode);
1443 mutex_unlock(&inode->i_mutex);
1444 cfs_down_write(&lli->lli_trunc_sem);
1445 mutex_lock(&inode->i_mutex);
1446 if (ia_valid & ATTR_SIZE)
1447 DOWN_WRITE_I_ALLOC_SEM(inode);
1450 /* We need a steady stripe configuration for setattr to avoid
1452 lsm = ccc_inode_lsm_get(inode);
1454 /* NB: ATTR_SIZE will only be set after this point if the size
1455 * resides on the MDS, ie, this file has no objects. */
1457 attr->ia_valid &= ~ATTR_SIZE;
1459 memcpy(&op_data->op_attr, attr, sizeof(*attr));
1461 /* Open epoch for truncate. */
1462 if (exp_connect_som(ll_i2mdexp(inode)) &&
1463 (ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1464 op_data->op_flags = MF_EPOCH_OPEN;
1466 rc = ll_md_setattr(dentry, op_data, &mod);
1470 ll_ioepoch_open(lli, op_data->op_ioepoch);
1471 if (lsm == NULL || !S_ISREG(inode->i_mode)) {
1472 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1476 if (ia_valid & ATTR_SIZE)
1477 attr->ia_valid |= ATTR_SIZE;
1478 if (ia_valid & (ATTR_SIZE |
1479 ATTR_ATIME | ATTR_ATIME_SET |
1480 ATTR_MTIME | ATTR_MTIME_SET))
1481 /* For truncate and utimes sending attributes to OSTs, setting
1482 * mtime/atime to the past will be performed under PW [0:EOF]
1483 * extent lock (new_size:EOF for truncate). It may seem
1484 * excessive to send mtime/atime updates to OSTs when not
1485 * setting times to past, but it is necessary due to possible
1486 * time de-synchronization between MDT inode and OST objects */
1487 rc = ll_setattr_ost(inode, attr);
1490 ccc_inode_lsm_put(inode, lsm);
1492 if (op_data->op_ioepoch) {
1493 rc1 = ll_setattr_done_writing(inode, op_data, mod);
1497 ll_finish_md_op_data(op_data);
1499 if (!S_ISDIR(inode->i_mode))
1500 cfs_up_write(&lli->lli_trunc_sem);
1502 ll_stats_ops_tally(ll_i2sbi(inode), (ia_valid & ATTR_SIZE) ?
1503 LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1508 int ll_setattr(struct dentry *de, struct iattr *attr)
1510 int mode = de->d_inode->i_mode;
1512 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1513 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1514 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1516 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1517 (ATTR_SIZE|ATTR_MODE)) &&
1518 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1519 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1520 !(attr->ia_mode & S_ISGID))))
1521 attr->ia_valid |= ATTR_FORCE;
1523 if ((mode & S_ISUID) &&
1524 !(attr->ia_mode & S_ISUID) &&
1525 !(attr->ia_valid & ATTR_KILL_SUID))
1526 attr->ia_valid |= ATTR_KILL_SUID;
1528 if (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1529 !(attr->ia_mode & S_ISGID) &&
1530 !(attr->ia_valid & ATTR_KILL_SGID))
1531 attr->ia_valid |= ATTR_KILL_SGID;
1533 return ll_setattr_raw(de, attr);
1536 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1537 __u64 max_age, __u32 flags)
1539 struct ll_sb_info *sbi = ll_s2sbi(sb);
1540 struct obd_statfs obd_osfs;
1544 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1546 CERROR("md_statfs fails: rc = %d\n", rc);
1550 osfs->os_type = sb->s_magic;
1552 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1553 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1555 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1556 flags |= OBD_STATFS_NODELAY;
1558 rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1560 CERROR("obd_statfs fails: rc = %d\n", rc);
1564 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1565 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1568 osfs->os_bsize = obd_osfs.os_bsize;
1569 osfs->os_blocks = obd_osfs.os_blocks;
1570 osfs->os_bfree = obd_osfs.os_bfree;
1571 osfs->os_bavail = obd_osfs.os_bavail;
1573 /* If we don't have as many objects free on the OST as inodes
1574 * on the MDS, we reduce the total number of inodes to
1575 * compensate, so that the "inodes in use" number is correct.
1577 if (obd_osfs.os_ffree < osfs->os_ffree) {
1578 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1580 osfs->os_ffree = obd_osfs.os_ffree;
1585 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1587 struct super_block *sb = de->d_sb;
1588 struct obd_statfs osfs;
1591 CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1592 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1594 /* Some amount of caching on the client is allowed */
1595 rc = ll_statfs_internal(sb, &osfs,
1596 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1601 statfs_unpack(sfs, &osfs);
1603 /* We need to downshift for all 32-bit kernels, because we can't
1604 * tell if the kernel is being called via sys_statfs64() or not.
1605 * Stop before overflowing f_bsize - in which case it is better
1606 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1607 if (sizeof(long) < 8) {
1608 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1611 osfs.os_blocks >>= 1;
1612 osfs.os_bfree >>= 1;
1613 osfs.os_bavail >>= 1;
1617 sfs->f_blocks = osfs.os_blocks;
1618 sfs->f_bfree = osfs.os_bfree;
1619 sfs->f_bavail = osfs.os_bavail;
1624 void ll_inode_size_lock(struct inode *inode)
1626 struct ll_inode_info *lli;
1628 LASSERT(!S_ISDIR(inode->i_mode));
1630 lli = ll_i2info(inode);
1631 LASSERT(lli->lli_size_sem_owner != current);
1632 cfs_down(&lli->lli_size_sem);
1633 LASSERT(lli->lli_size_sem_owner == NULL);
1634 lli->lli_size_sem_owner = current;
1637 void ll_inode_size_unlock(struct inode *inode)
1639 struct ll_inode_info *lli;
1641 lli = ll_i2info(inode);
1642 LASSERT(lli->lli_size_sem_owner == current);
1643 lli->lli_size_sem_owner = NULL;
1644 cfs_up(&lli->lli_size_sem);
1647 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1649 struct ll_inode_info *lli = ll_i2info(inode);
1650 struct mdt_body *body = md->body;
1651 struct lov_stripe_md *lsm = md->lsm;
1652 struct ll_sb_info *sbi = ll_i2sbi(inode);
1654 LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1656 LASSERT(S_ISREG(inode->i_mode));
1657 cfs_mutex_lock(&lli->lli_och_mutex);
1658 CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1659 lsm, inode->i_ino, inode->i_generation, inode);
1660 /* cl_file_inode_init must go before lli_has_smd or a race
1661 * is possible where client thinks the file has stripes,
1662 * but lov raid0 is not setup yet and parallel e.g.
1663 * glimpse would try to use uninitialized lov */
1664 if (cl_file_inode_init(inode, md) == 0)
1665 lli->lli_has_smd = true;
1666 cfs_mutex_unlock(&lli->lli_och_mutex);
1668 lli->lli_maxbytes = lsm->lsm_maxbytes;
1669 if (lli->lli_maxbytes > MAX_LFS_FILESIZE)
1670 lli->lli_maxbytes = MAX_LFS_FILESIZE;
1671 if (md->lsm != NULL)
1672 obd_free_memmd(ll_i2dtexp(inode), &md->lsm);
1675 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1676 if (body->valid & OBD_MD_FLRMTPERM)
1677 ll_update_remote_perm(inode, md->remote_perm);
1679 #ifdef CONFIG_FS_POSIX_ACL
1680 else if (body->valid & OBD_MD_FLACL) {
1681 cfs_spin_lock(&lli->lli_lock);
1682 if (lli->lli_posix_acl)
1683 posix_acl_release(lli->lli_posix_acl);
1684 lli->lli_posix_acl = md->posix_acl;
1685 cfs_spin_unlock(&lli->lli_lock);
1688 inode->i_ino = cl_fid_build_ino(&body->fid1, 0);
1689 inode->i_generation = cl_fid_build_gen(&body->fid1);
1691 if (body->valid & OBD_MD_FLATIME) {
1692 if (body->atime > LTIME_S(inode->i_atime))
1693 LTIME_S(inode->i_atime) = body->atime;
1694 lli->lli_lvb.lvb_atime = body->atime;
1696 if (body->valid & OBD_MD_FLMTIME) {
1697 if (body->mtime > LTIME_S(inode->i_mtime)) {
1698 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1699 "to "LPU64"\n", inode->i_ino,
1700 LTIME_S(inode->i_mtime), body->mtime);
1701 LTIME_S(inode->i_mtime) = body->mtime;
1703 lli->lli_lvb.lvb_mtime = body->mtime;
1705 if (body->valid & OBD_MD_FLCTIME) {
1706 if (body->ctime > LTIME_S(inode->i_ctime))
1707 LTIME_S(inode->i_ctime) = body->ctime;
1708 lli->lli_lvb.lvb_ctime = body->ctime;
1710 if (body->valid & OBD_MD_FLMODE)
1711 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1712 if (body->valid & OBD_MD_FLTYPE)
1713 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1714 LASSERT(inode->i_mode != 0);
1715 if (S_ISREG(inode->i_mode)) {
1716 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1718 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1720 if (body->valid & OBD_MD_FLUID)
1721 inode->i_uid = body->uid;
1722 if (body->valid & OBD_MD_FLGID)
1723 inode->i_gid = body->gid;
1724 if (body->valid & OBD_MD_FLFLAGS)
1725 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1726 if (body->valid & OBD_MD_FLNLINK)
1727 inode->i_nlink = body->nlink;
1728 if (body->valid & OBD_MD_FLRDEV)
1729 inode->i_rdev = old_decode_dev(body->rdev);
1731 if (body->valid & OBD_MD_FLID) {
1732 /* FID shouldn't be changed! */
1733 if (fid_is_sane(&lli->lli_fid)) {
1734 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1735 "Trying to change FID "DFID
1736 " to the "DFID", inode %lu/%u(%p)\n",
1737 PFID(&lli->lli_fid), PFID(&body->fid1),
1738 inode->i_ino, inode->i_generation, inode);
1740 lli->lli_fid = body->fid1;
1743 LASSERT(fid_seq(&lli->lli_fid) != 0);
1745 if (body->valid & OBD_MD_FLSIZE) {
1746 if (exp_connect_som(ll_i2mdexp(inode)) &&
1747 S_ISREG(inode->i_mode) && lli->lli_has_smd) {
1748 struct lustre_handle lockh;
1751 /* As it is possible a blocking ast has been processed
1752 * by this time, we need to check there is an UPDATE
1753 * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1755 mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1758 if (lli->lli_flags & (LLIF_DONE_WRITING |
1759 LLIF_EPOCH_PENDING |
1761 CERROR("ino %lu flags %u still has "
1762 "size authority! do not trust "
1763 "the size got from MDS\n",
1764 inode->i_ino, lli->lli_flags);
1766 /* Use old size assignment to avoid
1767 * deadlock bz14138 & bz14326 */
1768 i_size_write(inode, body->size);
1769 lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1771 ldlm_lock_decref(&lockh, mode);
1774 /* Use old size assignment to avoid
1775 * deadlock bz14138 & bz14326 */
1776 i_size_write(inode, body->size);
1778 CDEBUG(D_VFSTRACE, "inode=%lu, updating i_size %llu\n",
1779 inode->i_ino, (unsigned long long)body->size);
1782 if (body->valid & OBD_MD_FLBLOCKS)
1783 inode->i_blocks = body->blocks;
1786 if (body->valid & OBD_MD_FLMDSCAPA) {
1787 LASSERT(md->mds_capa);
1788 ll_add_capa(inode, md->mds_capa);
1790 if (body->valid & OBD_MD_FLOSSCAPA) {
1791 LASSERT(md->oss_capa);
1792 ll_add_capa(inode, md->oss_capa);
1796 void ll_read_inode2(struct inode *inode, void *opaque)
1798 struct lustre_md *md = opaque;
1799 struct ll_inode_info *lli = ll_i2info(inode);
1802 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1803 PFID(&lli->lli_fid), inode);
1805 LASSERT(!lli->lli_has_smd);
1807 /* Core attributes from the MDS first. This is a new inode, and
1808 * the VFS doesn't zero times in the core inode so we have to do
1809 * it ourselves. They will be overwritten by either MDS or OST
1810 * attributes - we just need to make sure they aren't newer. */
1811 LTIME_S(inode->i_mtime) = 0;
1812 LTIME_S(inode->i_atime) = 0;
1813 LTIME_S(inode->i_ctime) = 0;
1815 ll_update_inode(inode, md);
1817 /* OIDEBUG(inode); */
1819 /* initializing backing dev info. */
1820 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1823 if (S_ISREG(inode->i_mode)) {
1824 struct ll_sb_info *sbi = ll_i2sbi(inode);
1825 inode->i_op = &ll_file_inode_operations;
1826 inode->i_fop = sbi->ll_fop;
1827 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1829 } else if (S_ISDIR(inode->i_mode)) {
1830 inode->i_op = &ll_dir_inode_operations;
1831 inode->i_fop = &ll_dir_operations;
1832 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_dir_aops;
1834 } else if (S_ISLNK(inode->i_mode)) {
1835 inode->i_op = &ll_fast_symlink_inode_operations;
1838 inode->i_op = &ll_special_inode_operations;
1840 init_special_inode(inode, inode->i_mode,
1841 kdev_t_to_nr(inode->i_rdev));
1847 void ll_delete_inode(struct inode *inode)
1849 struct cl_inode_info *lli = cl_i2info(inode);
1850 struct ll_sb_info *sbi = ll_i2sbi(inode);
1854 rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
1856 CERROR("fid_delete() failed, rc %d\n", rc);
1858 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
1859 /* discard all dirty pages before truncating them, required by
1860 * osc_extent implementation at LU-1030. */
1861 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_DISCARD);
1863 truncate_inode_pages(&inode->i_data, 0);
1865 /* Workaround for LU-118 */
1866 if (inode->i_data.nrpages) {
1867 TREE_READ_LOCK_IRQ(&inode->i_data);
1868 TREE_READ_UNLOCK_IRQ(&inode->i_data);
1869 LASSERTF(inode->i_data.nrpages == 0,
1870 "inode=%lu/%u(%p) nrpages=%lu, see "
1871 "http://jira.whamcloud.com/browse/LU-118\n",
1872 inode->i_ino, inode->i_generation, inode,
1873 inode->i_data.nrpages);
1875 /* Workaround end */
1877 #ifdef HAVE_SBOPS_EVICT_INODE
1878 ll_clear_inode(inode);
1879 end_writeback(inode);
1887 int ll_iocontrol(struct inode *inode, struct file *file,
1888 unsigned int cmd, unsigned long arg)
1890 struct ll_sb_info *sbi = ll_i2sbi(inode);
1891 struct ptlrpc_request *req = NULL;
1896 case FSFILT_IOC_GETFLAGS: {
1897 struct mdt_body *body;
1898 struct md_op_data *op_data;
1900 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
1901 0, 0, LUSTRE_OPC_ANY,
1903 if (IS_ERR(op_data))
1904 RETURN(PTR_ERR(op_data));
1906 op_data->op_valid = OBD_MD_FLFLAGS;
1907 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
1908 ll_finish_md_op_data(op_data);
1910 CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1914 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1916 flags = body->flags;
1918 ptlrpc_req_finished(req);
1920 RETURN(put_user(flags, (int *)arg));
1922 case FSFILT_IOC_SETFLAGS: {
1923 struct lov_stripe_md *lsm;
1924 struct obd_info oinfo = { { { 0 } } };
1925 struct md_op_data *op_data;
1927 if (get_user(flags, (int *)arg))
1930 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1931 LUSTRE_OPC_ANY, NULL);
1932 if (IS_ERR(op_data))
1933 RETURN(PTR_ERR(op_data));
1935 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1936 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1937 rc = md_setattr(sbi->ll_md_exp, op_data,
1938 NULL, 0, NULL, 0, &req, NULL);
1939 ll_finish_md_op_data(op_data);
1940 ptlrpc_req_finished(req);
1944 OBDO_ALLOC(oinfo.oi_oa);
1948 lsm = ccc_inode_lsm_get(inode);
1950 inode->i_flags = ll_ext_to_inode_flags(flags);
1951 OBDO_FREE(oinfo.oi_oa);
1956 oinfo.oi_oa->o_id = lsm->lsm_object_id;
1957 oinfo.oi_oa->o_seq = lsm->lsm_object_seq;
1958 oinfo.oi_oa->o_flags = flags;
1959 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1961 oinfo.oi_capa = ll_mdscapa_get(inode);
1962 obdo_set_parent_fid(oinfo.oi_oa, &ll_i2info(inode)->lli_fid);
1963 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1964 capa_put(oinfo.oi_capa);
1965 OBDO_FREE(oinfo.oi_oa);
1966 ccc_inode_lsm_put(inode, lsm);
1968 if (rc && rc != -EPERM && rc != -EACCES)
1969 CERROR("osc_setattr_async fails: rc = %d\n", rc);
1980 int ll_flush_ctx(struct inode *inode)
1982 struct ll_sb_info *sbi = ll_i2sbi(inode);
1984 CDEBUG(D_SEC, "flush context for user %d\n", cfs_curproc_uid());
1986 obd_set_info_async(NULL, sbi->ll_md_exp,
1987 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1989 obd_set_info_async(NULL, sbi->ll_dt_exp,
1990 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1995 /* umount -f client means force down, don't save state */
1996 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1997 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
1999 struct super_block *sb = vfsmnt->mnt_sb;
2001 void ll_umount_begin(struct super_block *sb)
2004 struct ll_sb_info *sbi = ll_s2sbi(sb);
2005 struct obd_device *obd;
2006 struct obd_ioctl_data *ioc_data;
2009 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2010 if (!(flags & MNT_FORCE)) {
2016 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2017 sb->s_count, atomic_read(&sb->s_active));
2019 obd = class_exp2obd(sbi->ll_md_exp);
2021 CERROR("Invalid MDC connection handle "LPX64"\n",
2022 sbi->ll_md_exp->exp_handle.h_cookie);
2028 obd = class_exp2obd(sbi->ll_dt_exp);
2030 CERROR("Invalid LOV connection handle "LPX64"\n",
2031 sbi->ll_dt_exp->exp_handle.h_cookie);
2037 OBD_ALLOC_PTR(ioc_data);
2039 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2040 sizeof ioc_data, ioc_data, NULL);
2042 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2043 sizeof ioc_data, ioc_data, NULL);
2045 OBD_FREE_PTR(ioc_data);
2049 /* Really, we'd like to wait until there are no requests outstanding,
2050 * and then continue. For now, we just invalidate the requests,
2051 * schedule() and sleep one second if needed, and hope.
2054 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2055 if (atomic_read(&vfsmnt->mnt_count) > 2) {
2056 cfs_schedule_timeout_and_set_state(CFS_TASK_INTERRUPTIBLE,
2057 cfs_time_seconds(1));
2058 if (atomic_read(&vfsmnt->mnt_count) > 2)
2059 LCONSOLE_WARN("Mount still busy with %d refs! You "
2060 "may try to umount it a bit later\n",
2061 atomic_read(&vfsmnt->mnt_count));
2068 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2070 struct ll_sb_info *sbi = ll_s2sbi(sb);
2071 char *profilenm = get_profile_name(sb);
2075 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2076 read_only = *flags & MS_RDONLY;
2077 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2078 sizeof(KEY_READ_ONLY),
2079 KEY_READ_ONLY, sizeof(read_only),
2082 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2083 profilenm, read_only ?
2084 "read-only" : "read-write", err);
2089 sb->s_flags |= MS_RDONLY;
2091 sb->s_flags &= ~MS_RDONLY;
2093 if (sbi->ll_flags & LL_SBI_VERBOSE)
2094 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2095 read_only ? "read-only" : "read-write");
2100 int ll_prep_inode(struct inode **inode,
2101 struct ptlrpc_request *req,
2102 struct super_block *sb)
2104 struct ll_sb_info *sbi = NULL;
2105 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);
2145 /* sanity check for LAYOUT lock. */
2146 ibits = MDS_INODELOCK_LAYOUT;
2147 if (S_ISREG(md.body->mode) && sbi->ll_flags & LL_SBI_LAYOUT_LOCK &&
2148 md.lsm != NULL && !ll_have_md_lock(*inode, &ibits, LCK_MINMODE)) {
2149 char *fsname = ll_get_fsname(*inode);
2150 CERROR("%s: inode "DFID" (%p) layout lock not granted.\n",
2151 fsname, PFID(ll_inode2fid(*inode)),
2154 OBD_FREE(fsname, MGS_PARAM_MAXLEN);
2158 md_free_lustre_md(sbi->ll_md_exp, &md);
2162 int ll_obd_statfs(struct inode *inode, void *arg)
2164 struct ll_sb_info *sbi = NULL;
2165 struct obd_export *exp;
2167 struct obd_ioctl_data *data = NULL;
2171 if (!inode || !(sbi = ll_i2sbi(inode)))
2172 GOTO(out_statfs, rc = -EINVAL);
2174 rc = obd_ioctl_getdata(&buf, &len, arg);
2176 GOTO(out_statfs, rc);
2179 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2180 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2181 GOTO(out_statfs, rc = -EINVAL);
2183 if (data->ioc_inllen1 != sizeof(__u32) ||
2184 data->ioc_inllen2 != sizeof(__u32) ||
2185 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2186 data->ioc_plen2 != sizeof(struct obd_uuid))
2187 GOTO(out_statfs, rc = -EINVAL);
2189 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2190 if (type == LL_STATFS_LMV)
2191 exp = sbi->ll_md_exp;
2192 else if (type == LL_STATFS_LOV)
2193 exp = sbi->ll_dt_exp;
2195 GOTO(out_statfs, rc = -ENODEV);
2197 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2199 GOTO(out_statfs, rc);
2202 obd_ioctl_freedata(buf, len);
2206 int ll_process_config(struct lustre_cfg *lcfg)
2210 struct lprocfs_static_vars lvars;
2214 lprocfs_llite_init_vars(&lvars);
2216 /* The instance name contains the sb: lustre-client-aacfe000 */
2217 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2218 if (!ptr || !*(++ptr))
2220 if (sscanf(ptr, "%lx", &x) != 1)
2223 /* This better be a real Lustre superblock! */
2224 LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2226 /* Note we have not called client_common_fill_super yet, so
2227 proc fns must be able to handle that! */
2228 rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2235 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2236 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2237 struct inode *i1, struct inode *i2,
2238 const char *name, int namelen,
2239 int mode, __u32 opc, void *data)
2241 LASSERT(i1 != NULL);
2243 if (namelen > ll_i2sbi(i1)->ll_namelen)
2244 return ERR_PTR(-ENAMETOOLONG);
2246 if (op_data == NULL)
2247 OBD_ALLOC_PTR(op_data);
2249 if (op_data == NULL)
2250 return ERR_PTR(-ENOMEM);
2252 ll_i2gids(op_data->op_suppgids, i1, i2);
2253 op_data->op_fid1 = *ll_inode2fid(i1);
2254 op_data->op_capa1 = ll_mdscapa_get(i1);
2257 op_data->op_fid2 = *ll_inode2fid(i2);
2258 op_data->op_capa2 = ll_mdscapa_get(i2);
2260 fid_zero(&op_data->op_fid2);
2261 op_data->op_capa2 = NULL;
2264 op_data->op_name = name;
2265 op_data->op_namelen = namelen;
2266 op_data->op_mode = mode;
2267 op_data->op_mod_time = cfs_time_current_sec();
2268 op_data->op_fsuid = cfs_curproc_fsuid();
2269 op_data->op_fsgid = cfs_curproc_fsgid();
2270 op_data->op_cap = cfs_curproc_cap_pack();
2271 op_data->op_bias = MDS_CHECK_SPLIT;
2272 op_data->op_opc = opc;
2273 op_data->op_mds = 0;
2274 op_data->op_data = data;
2276 /* If the file is being opened after mknod() (normally due to NFS)
2277 * try to use the default stripe data from parent directory for
2278 * allocating OST objects. Try to pass the parent FID to MDS. */
2279 if (opc == LUSTRE_OPC_CREATE && i1 == i2 && S_ISREG(i2->i_mode) &&
2280 !ll_i2info(i2)->lli_has_smd) {
2281 struct ll_inode_info *lli = ll_i2info(i2);
2283 cfs_spin_lock(&lli->lli_lock);
2284 if (likely(!lli->lli_has_smd && !fid_is_zero(&lli->lli_pfid)))
2285 op_data->op_fid1 = lli->lli_pfid;
2286 cfs_spin_unlock(&lli->lli_lock);
2287 /** We ignore parent's capability temporary. */
2293 void ll_finish_md_op_data(struct md_op_data *op_data)
2295 capa_put(op_data->op_capa1);
2296 capa_put(op_data->op_capa2);
2297 OBD_FREE_PTR(op_data);
2300 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2302 struct ll_sb_info *sbi;
2304 LASSERT((seq != NULL) && (vfs != NULL));
2305 sbi = ll_s2sbi(vfs->mnt_sb);
2307 if (sbi->ll_flags & LL_SBI_NOLCK)
2308 seq_puts(seq, ",nolock");
2310 if (sbi->ll_flags & LL_SBI_FLOCK)
2311 seq_puts(seq, ",flock");
2313 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2314 seq_puts(seq, ",localflock");
2316 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2317 seq_puts(seq, ",user_xattr");
2319 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2320 seq_puts(seq, ",lazystatfs");
2326 * Get obd name by cmd, and copy out to user space
2328 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2330 struct ll_sb_info *sbi = ll_i2sbi(inode);
2331 struct obd_device *obd;
2334 if (cmd == OBD_IOC_GETDTNAME)
2335 obd = class_exp2obd(sbi->ll_dt_exp);
2336 else if (cmd == OBD_IOC_GETMDNAME)
2337 obd = class_exp2obd(sbi->ll_md_exp);
2344 if (cfs_copy_to_user((void *)arg, obd->obd_name,
2345 strlen(obd->obd_name) + 1))