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, 2013, Intel Corporation.
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 struct kmem_cache *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;
67 extern struct address_space_operations_ext ll_aops;
71 #define log2(n) ffz(~(n))
74 static struct ll_sb_info *ll_init_sbi(void)
76 struct ll_sb_info *sbi = NULL;
78 unsigned long lru_page_max;
84 OBD_ALLOC(sbi, sizeof(*sbi));
88 spin_lock_init(&sbi->ll_lock);
89 mutex_init(&sbi->ll_lco.lco_lock);
90 spin_lock_init(&sbi->ll_pp_extent_lock);
91 spin_lock_init(&sbi->ll_process_lock);
92 sbi->ll_rw_stats_on = 0;
95 pages = si.totalram - si.totalhigh;
96 lru_page_max = pages / 2;
98 /* initialize ll_cache data */
99 cfs_atomic_set(&sbi->ll_cache.ccc_users, 0);
100 sbi->ll_cache.ccc_lru_max = lru_page_max;
101 cfs_atomic_set(&sbi->ll_cache.ccc_lru_left, lru_page_max);
102 spin_lock_init(&sbi->ll_cache.ccc_lru_lock);
103 CFS_INIT_LIST_HEAD(&sbi->ll_cache.ccc_lru);
105 cfs_atomic_set(&sbi->ll_cache.ccc_unstable_nr, 0);
106 init_waitqueue_head(&sbi->ll_cache.ccc_unstable_waitq);
108 sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
109 SBI_DEFAULT_READAHEAD_MAX);
110 sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
111 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
112 SBI_DEFAULT_READAHEAD_WHOLE_MAX;
113 CFS_INIT_LIST_HEAD(&sbi->ll_conn_chain);
114 CFS_INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
116 ll_generate_random_uuid(uuid);
117 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
118 CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
120 spin_lock(&ll_sb_lock);
121 cfs_list_add_tail(&sbi->ll_list, &ll_super_blocks);
122 spin_unlock(&ll_sb_lock);
124 sbi->ll_flags |= LL_SBI_VERBOSE;
125 #ifdef ENABLE_CHECKSUM
126 sbi->ll_flags |= LL_SBI_CHECKSUM;
129 #ifdef HAVE_LRU_RESIZE_SUPPORT
130 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
133 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
134 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
136 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
140 /* metadata statahead is enabled by default */
141 sbi->ll_sa_max = LL_SA_RPC_DEF;
142 cfs_atomic_set(&sbi->ll_sa_total, 0);
143 cfs_atomic_set(&sbi->ll_sa_wrong, 0);
144 cfs_atomic_set(&sbi->ll_agl_total, 0);
145 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
150 void ll_free_sbi(struct super_block *sb)
152 struct ll_sb_info *sbi = ll_s2sbi(sb);
156 spin_lock(&ll_sb_lock);
157 cfs_list_del(&sbi->ll_list);
158 spin_unlock(&ll_sb_lock);
159 OBD_FREE(sbi, sizeof(*sbi));
164 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
165 struct vfsmount *mnt)
167 struct inode *root = 0;
168 struct ll_sb_info *sbi = ll_s2sbi(sb);
169 struct obd_device *obd;
170 struct obd_capa *oc = NULL;
171 struct obd_statfs *osfs = NULL;
172 struct ptlrpc_request *request = NULL;
173 struct obd_connect_data *data = NULL;
174 struct obd_uuid *uuid;
175 struct md_op_data *op_data;
176 struct lustre_md lmd;
178 int size, err, checksum;
181 obd = class_name2obd(md);
183 CERROR("MD %s: not setup or attached\n", md);
197 /* indicate the features supported by this client */
198 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
199 OBD_CONNECT_ATTRFID |
200 OBD_CONNECT_VERSION | OBD_CONNECT_BRW_SIZE |
201 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
202 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
203 OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 |
204 OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR |
205 OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH|
206 OBD_CONNECT_EINPROGRESS |
207 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
208 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS |
209 OBD_CONNECT_MAX_EASIZE |
210 OBD_CONNECT_FLOCK_DEAD |
211 OBD_CONNECT_DISP_STRIPE;
213 if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
214 data->ocd_connect_flags |= OBD_CONNECT_SOM;
216 #ifdef HAVE_LRU_RESIZE_SUPPORT
217 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
218 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
220 #ifdef CONFIG_FS_POSIX_ACL
221 data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK;
224 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
225 /* flag mdc connection as lightweight, only used for test
226 * purpose, use with care */
227 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
229 data->ocd_ibits_known = MDS_INODELOCK_FULL;
230 data->ocd_version = LUSTRE_VERSION_CODE;
232 if (sb->s_flags & MS_RDONLY)
233 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
234 if (sbi->ll_flags & LL_SBI_USER_XATTR)
235 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
237 #ifdef HAVE_MS_FLOCK_LOCK
238 /* force vfs to use lustre handler for flock() calls - bug 10743 */
239 sb->s_flags |= MS_FLOCK_LOCK;
241 #ifdef MS_HAS_NEW_AOPS
242 sb->s_flags |= MS_HAS_NEW_AOPS;
245 if (sbi->ll_flags & LL_SBI_FLOCK)
246 sbi->ll_fop = &ll_file_operations_flock;
247 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
248 sbi->ll_fop = &ll_file_operations;
250 sbi->ll_fop = &ll_file_operations_noflock;
253 data->ocd_connect_flags |= OBD_CONNECT_REAL;
254 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
255 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
257 data->ocd_brw_size = MD_MAX_BRW_SIZE;
259 err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
261 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
262 "recovery, of which this client is not a "
263 "part. Please wait for recovery to complete,"
264 " abort, or time out.\n", md);
267 CERROR("cannot connect to %s: rc = %d\n", md, err);
271 sbi->ll_md_exp->exp_connect_data = *data;
273 err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
274 LUSTRE_SEQ_METADATA);
276 CERROR("%s: Can't init metadata layer FID infrastructure, "
277 "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
281 /* For mount, we only need fs info from MDT0, and also in DNE, it
282 * can make sure the client can be mounted as long as MDT0 is
284 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
285 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
286 OBD_STATFS_FOR_MDT0);
288 GOTO(out_md_fid, err);
290 /* This needs to be after statfs to ensure connect has finished.
291 * Note that "data" does NOT contain the valid connect reply.
292 * If connecting to a 1.8 server there will be no LMV device, so
293 * we can access the MDC export directly and exp_connect_flags will
294 * be non-zero, but if accessing an upgraded 2.1 server it will
295 * have the correct flags filled in.
296 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
297 valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
298 if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
299 valid != CLIENT_CONNECT_MDT_REQD) {
302 OBD_ALLOC_WAIT(buf, PAGE_CACHE_SIZE);
303 obd_connect_flags2str(buf, PAGE_CACHE_SIZE,
304 valid ^ CLIENT_CONNECT_MDT_REQD, ",");
305 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
306 "feature(s) needed for correct operation "
307 "of this client (%s). Please upgrade "
308 "server or downgrade client.\n",
309 sbi->ll_md_exp->exp_obd->obd_name, buf);
310 OBD_FREE(buf, PAGE_CACHE_SIZE);
311 GOTO(out_md_fid, err = -EPROTO);
314 size = sizeof(*data);
315 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
316 KEY_CONN_DATA, &size, data, NULL);
318 CERROR("%s: Get connect data failed: rc = %d\n",
319 sbi->ll_md_exp->exp_obd->obd_name, err);
320 GOTO(out_md_fid, err);
323 LASSERT(osfs->os_bsize);
324 sb->s_blocksize = osfs->os_bsize;
325 sb->s_blocksize_bits = log2(osfs->os_bsize);
326 sb->s_magic = LL_SUPER_MAGIC;
327 sb->s_maxbytes = MAX_LFS_FILESIZE;
328 sbi->ll_namelen = osfs->os_namelen;
329 sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
331 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
332 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
333 LCONSOLE_INFO("Disabling user_xattr feature because "
334 "it is not supported on the server\n");
335 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
338 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
340 sb->s_flags |= MS_POSIXACL;
342 sbi->ll_flags |= LL_SBI_ACL;
344 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
346 sb->s_flags &= ~MS_POSIXACL;
348 sbi->ll_flags &= ~LL_SBI_ACL;
351 if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
352 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
353 sbi->ll_flags |= LL_SBI_RMT_CLIENT;
354 LCONSOLE_INFO("client is set as remote by default.\n");
357 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
358 sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
359 LCONSOLE_INFO("client claims to be remote, but server "
360 "rejected, forced to be local.\n");
364 if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
365 LCONSOLE_INFO("client enabled MDS capability!\n");
366 sbi->ll_flags |= LL_SBI_MDS_CAPA;
369 if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
370 LCONSOLE_INFO("client enabled OSS capability!\n");
371 sbi->ll_flags |= LL_SBI_OSS_CAPA;
374 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
375 sbi->ll_flags |= LL_SBI_64BIT_HASH;
377 if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
378 sbi->ll_md_brw_size = data->ocd_brw_size;
380 sbi->ll_md_brw_size = PAGE_CACHE_SIZE;
382 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
383 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
385 if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
386 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
387 LCONSOLE_INFO("%s: disabling xattr cache due to "
388 "unknown maximum xattr size.\n", dt);
390 sbi->ll_flags |= LL_SBI_XATTR_CACHE;
391 sbi->ll_xattr_cache_enabled = 1;
395 obd = class_name2obd(dt);
397 CERROR("DT %s: not setup or attached\n", dt);
398 GOTO(out_md_fid, err = -ENODEV);
401 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
402 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
403 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
404 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
405 OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
406 OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
407 OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
408 OBD_CONNECT_MAXBYTES |
409 OBD_CONNECT_EINPROGRESS |
410 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
411 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS;
413 if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
414 data->ocd_connect_flags |= OBD_CONNECT_SOM;
416 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
417 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
418 * disabled by default, because it can still be enabled on the
419 * fly via /proc. As a consequence, we still need to come to an
420 * agreement on the supported algorithms at connect time */
421 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
423 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
424 data->ocd_cksum_types = OBD_CKSUM_ADLER;
426 data->ocd_cksum_types = cksum_types_supported_client();
429 #ifdef HAVE_LRU_RESIZE_SUPPORT
430 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
432 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
433 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
435 CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
436 "ocd_grant: %d\n", data->ocd_connect_flags,
437 data->ocd_version, data->ocd_grant);
439 obd->obd_upcall.onu_owner = &sbi->ll_lco;
440 obd->obd_upcall.onu_upcall = cl_ocd_update;
442 data->ocd_brw_size = DT_MAX_BRW_SIZE;
444 err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
447 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
448 "recovery, of which this client is not a "
449 "part. Please wait for recovery to "
450 "complete, abort, or time out.\n", dt);
453 CERROR("%s: Cannot connect to %s: rc = %d\n",
454 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
458 sbi->ll_dt_exp->exp_connect_data = *data;
460 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
461 LUSTRE_SEQ_METADATA);
463 CERROR("%s: Can't init data layer FID infrastructure, "
464 "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
468 mutex_lock(&sbi->ll_lco.lco_lock);
469 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
470 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
471 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
472 mutex_unlock(&sbi->ll_lco.lco_lock);
474 fid_zero(&sbi->ll_root_fid);
475 err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
477 CERROR("cannot mds_connect: rc = %d\n", err);
478 GOTO(out_lock_cn_cb, err);
480 if (!fid_is_sane(&sbi->ll_root_fid)) {
481 CERROR("%s: Invalid root fid "DFID" during mount\n",
482 sbi->ll_md_exp->exp_obd->obd_name,
483 PFID(&sbi->ll_root_fid));
484 GOTO(out_lock_cn_cb, err = -EINVAL);
486 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
488 sb->s_op = &lustre_super_operations;
489 #if THREAD_SIZE >= 8192 /*b=17630*/
490 sb->s_export_op = &lustre_export_operations;
494 * XXX: move this to after cbd setup? */
495 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA |
497 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
498 valid |= OBD_MD_FLRMTPERM;
499 else if (sbi->ll_flags & LL_SBI_ACL)
500 valid |= OBD_MD_FLACL;
502 OBD_ALLOC_PTR(op_data);
504 GOTO(out_lock_cn_cb, err = -ENOMEM);
506 op_data->op_fid1 = sbi->ll_root_fid;
507 op_data->op_mode = 0;
508 op_data->op_capa1 = oc;
509 op_data->op_valid = valid;
511 err = md_getattr(sbi->ll_md_exp, op_data, &request);
514 OBD_FREE_PTR(op_data);
516 CERROR("%s: md_getattr failed for root: rc = %d\n",
517 sbi->ll_md_exp->exp_obd->obd_name, err);
518 GOTO(out_lock_cn_cb, err);
521 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
522 sbi->ll_md_exp, &lmd);
524 CERROR("failed to understand root inode md: rc = %d\n", err);
525 ptlrpc_req_finished(request);
526 GOTO(out_lock_cn_cb, err);
529 LASSERT(fid_is_sane(&sbi->ll_root_fid));
530 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
531 sbi->ll_flags & LL_SBI_32BIT_API),
533 md_free_lustre_md(sbi->ll_md_exp, &lmd);
534 ptlrpc_req_finished(request);
536 if (root == NULL || IS_ERR(root)) {
538 obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
539 #ifdef CONFIG_FS_POSIX_ACL
541 posix_acl_release(lmd.posix_acl);
542 lmd.posix_acl = NULL;
545 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
547 CERROR("lustre_lite: bad iget4 for root\n");
551 err = ll_close_thread_start(&sbi->ll_lcq);
553 CERROR("cannot start close thread: rc %d\n", err);
557 #ifdef CONFIG_FS_POSIX_ACL
558 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
559 rct_init(&sbi->ll_rct);
560 et_init(&sbi->ll_et);
564 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
565 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
566 KEY_CHECKSUM, sizeof(checksum), &checksum,
570 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
571 KEY_CACHE_SET, sizeof(sbi->ll_cache),
572 &sbi->ll_cache, NULL);
574 sb->s_root = d_make_root(root);
575 if (sb->s_root == NULL) {
576 CERROR("%s: can't make root dentry\n",
577 ll_get_fsname(sb, NULL, 0));
578 GOTO(out_root, err = -ENOMEM);
580 #ifdef HAVE_DCACHE_LOCK
581 sb->s_root->d_op = &ll_d_ops;
584 sbi->ll_sdev_orig = sb->s_dev;
586 /* We set sb->s_dev equal on all lustre clients in order to support
587 * NFS export clustering. NFSD requires that the FSID be the same
589 /* s_dev is also used in lt_compare() to compare two fs, but that is
590 * only a node-local comparison. */
591 uuid = obd_get_uuid(sbi->ll_md_exp);
593 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
599 if (proc_lustre_fs_root != NULL) {
600 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
603 CERROR("%s: could not register mount in lprocfs: "
604 "rc = %d\n", ll_get_fsname(sb, NULL, 0), err);
614 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
616 obd_disconnect(sbi->ll_dt_exp);
617 sbi->ll_dt_exp = NULL;
618 /* Make sure all OScs are gone, since cl_cache is accessing sbi. */
619 obd_zombie_barrier();
621 obd_fid_fini(sbi->ll_md_exp->exp_obd);
623 obd_disconnect(sbi->ll_md_exp);
624 sbi->ll_md_exp = NULL;
633 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
637 *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
639 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
640 KEY_MAX_EASIZE, &size, lmmsize, NULL);
642 CERROR("Get max mdsize error rc %d\n", rc);
647 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
652 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
653 KEY_DEFAULT_EASIZE, &size, lmmsize, NULL);
655 CERROR("Get default mdsize error rc %d\n", rc);
660 int ll_get_max_cookiesize(struct ll_sb_info *sbi, int *lmmsize)
665 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_COOKIESIZE),
666 KEY_MAX_COOKIESIZE, &size, lmmsize, NULL);
668 CERROR("Get max cookiesize error rc %d\n", rc);
673 int ll_get_default_cookiesize(struct ll_sb_info *sbi, int *lmmsize)
678 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_COOKIESIZE),
679 KEY_DEFAULT_COOKIESIZE, &size, lmmsize, NULL);
681 CERROR("Get default cookiesize error rc %d\n", rc);
686 void ll_dump_inode(struct inode *inode)
688 struct ll_d_hlist_node *tmp;
689 int dentry_count = 0;
691 LASSERT(inode != NULL);
693 ll_d_hlist_for_each(tmp, &inode->i_dentry)
696 CERROR("%s: inode %p dump: dev=%s fid="DFID
697 " mode=%o count=%u, %d dentries\n",
698 ll_get_fsname(inode->i_sb, NULL, 0), inode,
699 ll_i2mdexp(inode)->exp_obd->obd_name, PFID(ll_inode2fid(inode)),
700 inode->i_mode, atomic_read(&inode->i_count), dentry_count);
703 void lustre_dump_dentry(struct dentry *dentry, int recur)
705 struct list_head *tmp;
708 LASSERT(dentry != NULL);
710 list_for_each(tmp, &dentry->d_subdirs)
713 CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
714 " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
715 dentry->d_name.len, dentry->d_name.name,
716 dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
717 dentry->d_parent, dentry->d_inode, d_count(dentry),
718 dentry->d_flags, dentry->d_fsdata, subdirs);
719 if (dentry->d_inode != NULL)
720 ll_dump_inode(dentry->d_inode);
725 list_for_each(tmp, &dentry->d_subdirs) {
726 struct dentry *d = list_entry(tmp, struct dentry, d_u.d_child);
727 lustre_dump_dentry(d, recur - 1);
731 void client_common_put_super(struct super_block *sb)
733 struct ll_sb_info *sbi = ll_s2sbi(sb);
736 #ifdef CONFIG_FS_POSIX_ACL
737 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
738 et_fini(&sbi->ll_et);
739 rct_fini(&sbi->ll_rct);
743 ll_close_thread_shutdown(sbi->ll_lcq);
747 cfs_list_del(&sbi->ll_conn_chain);
749 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
750 obd_disconnect(sbi->ll_dt_exp);
751 sbi->ll_dt_exp = NULL;
752 /* wait till all OSCs are gone, since cl_cache is accessing sbi.
754 obd_zombie_barrier();
756 lprocfs_unregister_mountpoint(sbi);
758 obd_fid_fini(sbi->ll_md_exp->exp_obd);
759 obd_disconnect(sbi->ll_md_exp);
760 sbi->ll_md_exp = NULL;
765 void ll_kill_super(struct super_block *sb)
767 struct ll_sb_info *sbi;
772 if (!(sb->s_flags & MS_ACTIVE))
776 /* we need restore s_dev from changed for clustred NFS before put_super
777 * because new kernels have cached s_dev and change sb->s_dev in
778 * put_super not affected real removing devices */
780 sb->s_dev = sbi->ll_sdev_orig;
781 sbi->ll_umounting = 1;
786 char *ll_read_opt(const char *opt, char *data)
792 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
793 if (strncmp(opt, data, strlen(opt)))
795 if ((value = strchr(data, '=')) == NULL)
799 OBD_ALLOC(retval, strlen(value) + 1);
801 CERROR("out of memory!\n");
805 memcpy(retval, value, strlen(value)+1);
806 CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
810 static inline int ll_set_opt(const char *opt, char *data, int fl)
812 if (strncmp(opt, data, strlen(opt)) != 0)
818 /* non-client-specific mount options are parsed in lmd_parse */
819 static int ll_options(char *options, int *flags)
822 char *s1 = options, *s2;
828 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
831 CDEBUG(D_SUPER, "next opt=%s\n", s1);
832 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
837 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
842 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
847 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
852 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
857 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
862 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 6, 51, 0)
863 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
865 /* Ignore deprecated mount option. The client will
866 * always try to mount with ACL support, whether this
867 * is used depends on whether server supports it. */
868 LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
869 "mount option 'acl'.\n");
872 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
874 LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
875 "mount option 'noacl'.\n");
879 #warning "{no}acl options have been deprecated since 1.8, please remove them"
881 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
886 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
891 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
897 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
902 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
907 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
912 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
917 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
922 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
927 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
932 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
937 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
942 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
947 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
953 s2 = strchr(s1, ',');
961 void ll_lli_init(struct ll_inode_info *lli)
963 lli->lli_inode_magic = LLI_INODE_MAGIC;
965 lli->lli_ioepoch = 0;
966 lli->lli_maxbytes = MAX_LFS_FILESIZE;
967 spin_lock_init(&lli->lli_lock);
968 lli->lli_posix_acl = NULL;
969 lli->lli_remote_perms = NULL;
970 mutex_init(&lli->lli_rmtperm_mutex);
971 /* Do not set lli_fid, it has been initialized already. */
972 fid_zero(&lli->lli_pfid);
973 CFS_INIT_LIST_HEAD(&lli->lli_close_list);
974 CFS_INIT_LIST_HEAD(&lli->lli_oss_capas);
975 cfs_atomic_set(&lli->lli_open_count, 0);
976 lli->lli_mds_capa = NULL;
977 lli->lli_rmtperm_time = 0;
978 lli->lli_pending_och = NULL;
979 lli->lli_mds_read_och = NULL;
980 lli->lli_mds_write_och = NULL;
981 lli->lli_mds_exec_och = NULL;
982 lli->lli_open_fd_read_count = 0;
983 lli->lli_open_fd_write_count = 0;
984 lli->lli_open_fd_exec_count = 0;
985 mutex_init(&lli->lli_och_mutex);
986 spin_lock_init(&lli->lli_agl_lock);
987 lli->lli_has_smd = false;
988 spin_lock_init(&lli->lli_layout_lock);
989 ll_layout_version_set(lli, LL_LAYOUT_GEN_NONE);
990 lli->lli_clob = NULL;
992 init_rwsem(&lli->lli_xattrs_list_rwsem);
993 mutex_init(&lli->lli_xattrs_enq_lock);
995 LASSERT(lli->lli_vfs_inode.i_mode != 0);
996 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
997 mutex_init(&lli->lli_readdir_mutex);
998 lli->lli_opendir_key = NULL;
1000 lli->lli_def_acl = NULL;
1001 spin_lock_init(&lli->lli_sa_lock);
1002 lli->lli_opendir_pid = 0;
1004 sema_init(&lli->lli_size_sem, 1);
1005 lli->lli_size_sem_owner = NULL;
1006 lli->lli_symlink_name = NULL;
1007 init_rwsem(&lli->lli_trunc_sem);
1008 mutex_init(&lli->lli_write_mutex);
1009 init_rwsem(&lli->lli_glimpse_sem);
1010 lli->lli_glimpse_time = 0;
1011 CFS_INIT_LIST_HEAD(&lli->lli_agl_list);
1012 lli->lli_agl_index = 0;
1013 lli->lli_async_rc = 0;
1014 lli->lli_volatile = false;
1016 mutex_init(&lli->lli_layout_mutex);
1019 static inline int ll_bdi_register(struct backing_dev_info *bdi)
1021 static atomic_t ll_bdi_num = ATOMIC_INIT(0);
1023 bdi->name = "lustre";
1024 return bdi_register(bdi, NULL, "lustre-%d",
1025 atomic_inc_return(&ll_bdi_num));
1028 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
1030 struct lustre_profile *lprof = NULL;
1031 struct lustre_sb_info *lsi = s2lsi(sb);
1032 struct ll_sb_info *sbi;
1033 char *dt = NULL, *md = NULL;
1034 char *profilenm = get_profile_name(sb);
1035 struct config_llog_instance *cfg;
1036 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1037 const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
1041 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1047 try_module_get(THIS_MODULE);
1049 /* client additional sb info */
1050 lsi->lsi_llsbi = sbi = ll_init_sbi();
1052 module_put(THIS_MODULE);
1057 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1059 GOTO(out_free, err);
1061 err = bdi_init(&lsi->lsi_bdi);
1063 GOTO(out_free, err);
1064 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1065 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1066 err = ll_bdi_register(&lsi->lsi_bdi);
1068 GOTO(out_free, err);
1070 sb->s_bdi = &lsi->lsi_bdi;
1071 #ifndef HAVE_DCACHE_LOCK
1072 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1073 sb->s_d_op = &ll_d_ops;
1076 /* Generate a string unique to this super, in case some joker tries
1077 to mount the same fs at two mount points.
1078 Use the address of the super itself.*/
1079 cfg->cfg_instance = sb;
1080 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1081 cfg->cfg_callback = class_config_llog_handler;
1082 /* set up client obds */
1083 err = lustre_process_log(sb, profilenm, cfg);
1085 GOTO(out_free, err);
1087 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1088 lprof = class_get_profile(profilenm);
1089 if (lprof == NULL) {
1090 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1091 " read from the MGS. Does that filesystem "
1092 "exist?\n", profilenm);
1093 GOTO(out_free, err = -EINVAL);
1095 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1096 lprof->lp_md, lprof->lp_dt);
1098 OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1100 GOTO(out_free, err = -ENOMEM);
1101 sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1103 OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1105 GOTO(out_free, err = -ENOMEM);
1106 sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1108 /* connections, registrations, sb setup */
1109 err = client_common_fill_super(sb, md, dt, mnt);
1113 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1115 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1118 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1119 LCONSOLE_WARN("Mounted %s\n", profilenm);
1123 } /* ll_fill_super */
1126 void lu_context_keys_dump(void);
1128 void ll_put_super(struct super_block *sb)
1130 struct config_llog_instance cfg, params_cfg;
1131 struct obd_device *obd;
1132 struct lustre_sb_info *lsi = s2lsi(sb);
1133 struct ll_sb_info *sbi = ll_s2sbi(sb);
1134 char *profilenm = get_profile_name(sb);
1135 int ccc_count, next, force = 1, rc = 0;
1138 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1140 ll_print_capa_stat(sbi);
1142 cfg.cfg_instance = sb;
1143 lustre_end_log(sb, profilenm, &cfg);
1145 params_cfg.cfg_instance = sb;
1146 lustre_end_log(sb, PARAMS_FILENAME, ¶ms_cfg);
1148 if (sbi->ll_md_exp) {
1149 obd = class_exp2obd(sbi->ll_md_exp);
1151 force = obd->obd_force;
1154 /* Wait for unstable pages to be committed to stable storage */
1156 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1157 rc = l_wait_event(sbi->ll_cache.ccc_unstable_waitq,
1158 cfs_atomic_read(&sbi->ll_cache.ccc_unstable_nr) == 0,
1162 ccc_count = cfs_atomic_read(&sbi->ll_cache.ccc_unstable_nr);
1163 if (force == 0 && rc != -EINTR)
1164 LASSERTF(ccc_count == 0, "count: %i\n", ccc_count);
1167 /* We need to set force before the lov_disconnect in
1168 lustre_common_put_super, since l_d cleans up osc's as well. */
1171 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1173 obd->obd_force = force;
1178 /* Only if client_common_fill_super succeeded */
1179 client_common_put_super(sb);
1183 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1184 class_manual_cleanup(obd);
1187 if (sbi->ll_flags & LL_SBI_VERBOSE)
1188 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1191 class_del_profile(profilenm);
1193 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1194 bdi_destroy(&lsi->lsi_bdi);
1195 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1199 lsi->lsi_llsbi = NULL;
1201 lustre_common_put_super(sb);
1203 cl_env_cache_purge(~0);
1205 module_put(THIS_MODULE);
1208 } /* client_put_super */
1210 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1212 struct inode *inode = NULL;
1214 /* NOTE: we depend on atomic igrab() -bzzz */
1215 lock_res_and_lock(lock);
1216 if (lock->l_resource->lr_lvb_inode) {
1217 struct ll_inode_info * lli;
1218 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1219 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1220 inode = igrab(lock->l_resource->lr_lvb_inode);
1222 inode = lock->l_resource->lr_lvb_inode;
1223 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1224 D_WARNING, lock, "lr_lvb_inode %p is "
1225 "bogus: magic %08x",
1226 lock->l_resource->lr_lvb_inode,
1227 lli->lli_inode_magic);
1231 unlock_res_and_lock(lock);
1235 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1237 struct inode *inode = NULL;
1238 /* NOTE: we depend on atomic igrab() -bzzz */
1239 lock_res_and_lock(lock);
1240 if (lock->l_ast_data) {
1241 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1242 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1243 inode = igrab(lock->l_ast_data);
1245 inode = lock->l_ast_data;
1246 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1247 D_WARNING, lock, "l_ast_data %p is "
1248 "bogus: magic %08x", lock->l_ast_data,
1249 lli->lli_inode_magic);
1253 unlock_res_and_lock(lock);
1257 static void ll_dir_clear_lsm_md(struct inode *inode)
1259 struct ll_inode_info *lli = ll_i2info(inode);
1261 LASSERT(S_ISDIR(inode->i_mode));
1263 if (lli->lli_lsm_md != NULL) {
1264 lmv_free_memmd(lli->lli_lsm_md);
1265 lli->lli_lsm_md = NULL;
1269 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1270 const struct lu_fid *fid,
1271 struct lustre_md *md)
1273 struct ll_sb_info *sbi = ll_s2sbi(sb);
1274 struct mdt_body *body = md->body;
1275 struct inode *inode;
1279 ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1280 inode = iget_locked(sb, ino);
1281 if (inode == NULL) {
1282 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1283 ll_get_fsname(sb, NULL, 0), PFID(fid));
1284 RETURN(ERR_PTR(-ENOENT));
1287 if (inode->i_state & I_NEW) {
1288 struct ll_inode_info *lli = ll_i2info(inode);
1289 struct lmv_stripe_md *lsm = md->lmv;
1291 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1292 (body->mode & S_IFMT);
1293 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1296 LTIME_S(inode->i_mtime) = 0;
1297 LTIME_S(inode->i_atime) = 0;
1298 LTIME_S(inode->i_ctime) = 0;
1301 /* initializing backing dev info. */
1302 inode->i_mapping->backing_dev_info =
1303 &s2lsi(inode->i_sb)->lsi_bdi;
1304 inode->i_op = &ll_dir_inode_operations;
1305 inode->i_fop = &ll_dir_operations;
1306 lli->lli_fid = *fid;
1309 LASSERT(lsm != NULL);
1310 /* master stripe FID */
1311 lli->lli_pfid = lsm->lsm_md_oinfo[0].lmo_fid;
1312 CDEBUG(D_INODE, "lli %p master "DFID" slave "DFID"\n",
1313 lli, PFID(fid), PFID(&lli->lli_pfid));
1314 unlock_new_inode(inode);
1320 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1323 struct lmv_stripe_md *lsm = md->lmv;
1326 LASSERT(lsm != NULL);
1327 /* XXX sigh, this lsm_root initialization should be in
1328 * LMV layer, but it needs ll_iget right now, so we
1329 * put this here right now. */
1330 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1331 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1332 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1334 lsm->lsm_md_oinfo[i].lmo_root = inode;
1336 /* Unfortunately ll_iget will call ll_update_inode,
1337 * where the initialization of slave inode is slightly
1338 * different, so it reset lsm_md to NULL to avoid
1339 * initializing lsm for slave inode. */
1340 lsm->lsm_md_oinfo[i].lmo_root =
1341 ll_iget_anon_dir(inode->i_sb, fid, md);
1342 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1343 int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1345 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1351 /* Here is where the lsm is being initialized(fill lmo_info) after
1352 * client retrieve MD stripe information from MDT. */
1353 return md_update_lsm_md(ll_i2mdexp(inode), lsm, md->body,
1354 ll_md_blocking_ast);
1357 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1358 const struct lmv_stripe_md *lsm_md2)
1360 return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1361 lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1362 lsm_md1->lsm_md_master_mdt_index ==
1363 lsm_md2->lsm_md_master_mdt_index &&
1364 lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1365 lsm_md1->lsm_md_layout_version ==
1366 lsm_md2->lsm_md_layout_version &&
1367 strcmp(lsm_md1->lsm_md_pool_name,
1368 lsm_md2->lsm_md_pool_name) == 0;
1371 static void ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1373 struct ll_inode_info *lli = ll_i2info(inode);
1374 struct lmv_stripe_md *lsm = md->lmv;
1378 LASSERT(S_ISDIR(inode->i_mode));
1379 CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1380 PFID(ll_inode2fid(inode)));
1382 /* no striped information from request. */
1384 if (lli->lli_lsm_md == NULL) {
1386 } else if (lli->lli_lsm_md->lsm_md_magic == LMV_MAGIC_MIGRATE) {
1387 /* migration is done, the temporay MIGRATE layout has
1389 CDEBUG(D_INODE, DFID" finish migration.\n",
1390 PFID(ll_inode2fid(inode)));
1391 lmv_free_memmd(lli->lli_lsm_md);
1392 lli->lli_lsm_md = NULL;
1395 /* The lustre_md from req does not include stripeEA,
1396 * see ll_md_setattr */
1401 /* set the directory layout */
1402 if (lli->lli_lsm_md == NULL) {
1405 rc = ll_init_lsm_md(inode, md);
1407 CERROR("%s: init "DFID" failed: rc = %d\n",
1408 ll_get_fsname(inode->i_sb, NULL, 0),
1409 PFID(&lli->lli_fid), rc);
1412 lli->lli_lsm_md = lsm;
1413 /* set lsm_md to NULL, so the following free lustre_md
1414 * will not free this lsm */
1416 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1417 lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1421 /* Compare the old and new stripe information */
1422 if (!lli_lsm_md_eq(lli->lli_lsm_md, lsm)) {
1423 CERROR("inode %p %lu mismatch\n"
1424 " new(%p) vs lli_lsm_md(%p):\n"
1428 " hash_type: %x %x\n"
1431 inode, inode->i_ino, lsm, lli->lli_lsm_md,
1432 lsm->lsm_md_magic, lli->lli_lsm_md->lsm_md_magic,
1433 lsm->lsm_md_stripe_count,
1434 lli->lli_lsm_md->lsm_md_stripe_count,
1435 lsm->lsm_md_master_mdt_index,
1436 lli->lli_lsm_md->lsm_md_master_mdt_index,
1437 lsm->lsm_md_hash_type, lli->lli_lsm_md->lsm_md_hash_type,
1438 lsm->lsm_md_layout_version,
1439 lli->lli_lsm_md->lsm_md_layout_version,
1440 lsm->lsm_md_pool_name,
1441 lli->lli_lsm_md->lsm_md_pool_name);
1445 for (idx = 0; idx < lli->lli_lsm_md->lsm_md_stripe_count; idx++) {
1446 if (!lu_fid_eq(&lli->lli_lsm_md->lsm_md_oinfo[idx].lmo_fid,
1447 &lsm->lsm_md_oinfo[idx].lmo_fid)) {
1448 CERROR("%s: FID in lsm mismatch idx %d, old: "DFID
1450 ll_get_fsname(inode->i_sb, NULL, 0), idx,
1451 PFID(&lli->lli_lsm_md->lsm_md_oinfo[idx].lmo_fid),
1452 PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1457 md_update_lsm_md(ll_i2mdexp(inode), ll_i2info(inode)->lli_lsm_md,
1458 md->body, ll_md_blocking_ast);
1463 void ll_clear_inode(struct inode *inode)
1465 struct ll_inode_info *lli = ll_i2info(inode);
1466 struct ll_sb_info *sbi = ll_i2sbi(inode);
1469 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1470 PFID(ll_inode2fid(inode)), inode);
1472 if (S_ISDIR(inode->i_mode)) {
1473 /* these should have been cleared in ll_file_release */
1474 LASSERT(lli->lli_opendir_key == NULL);
1475 LASSERT(lli->lli_sai == NULL);
1476 LASSERT(lli->lli_opendir_pid == 0);
1479 spin_lock(&lli->lli_lock);
1480 ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1481 spin_unlock(&lli->lli_lock);
1482 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1484 LASSERT(!lli->lli_open_fd_write_count);
1485 LASSERT(!lli->lli_open_fd_read_count);
1486 LASSERT(!lli->lli_open_fd_exec_count);
1488 if (lli->lli_mds_write_och)
1489 ll_md_real_close(inode, FMODE_WRITE);
1490 if (lli->lli_mds_exec_och)
1491 ll_md_real_close(inode, FMODE_EXEC);
1492 if (lli->lli_mds_read_och)
1493 ll_md_real_close(inode, FMODE_READ);
1495 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1496 OBD_FREE(lli->lli_symlink_name,
1497 strlen(lli->lli_symlink_name) + 1);
1498 lli->lli_symlink_name = NULL;
1501 ll_xattr_cache_destroy(inode);
1503 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1504 LASSERT(lli->lli_posix_acl == NULL);
1505 if (lli->lli_remote_perms) {
1506 free_rmtperm_hash(lli->lli_remote_perms);
1507 lli->lli_remote_perms = NULL;
1510 #ifdef CONFIG_FS_POSIX_ACL
1511 else if (lli->lli_posix_acl) {
1512 LASSERT(cfs_atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1513 LASSERT(lli->lli_remote_perms == NULL);
1514 posix_acl_release(lli->lli_posix_acl);
1515 lli->lli_posix_acl = NULL;
1518 lli->lli_inode_magic = LLI_INODE_DEAD;
1520 ll_clear_inode_capas(inode);
1521 if (S_ISDIR(inode->i_mode))
1522 ll_dir_clear_lsm_md(inode);
1524 LASSERT(list_empty(&lli->lli_agl_list));
1527 * XXX This has to be done before lsm is freed below, because
1528 * cl_object still uses inode lsm.
1530 cl_inode_fini(inode);
1531 lli->lli_has_smd = false;
1536 int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data,
1537 struct md_open_data **mod)
1539 struct lustre_md md;
1540 struct inode *inode = dentry->d_inode;
1541 struct ll_sb_info *sbi = ll_i2sbi(inode);
1542 struct ptlrpc_request *request = NULL;
1546 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1547 LUSTRE_OPC_ANY, NULL);
1548 if (IS_ERR(op_data))
1549 RETURN(PTR_ERR(op_data));
1551 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1554 ptlrpc_req_finished(request);
1555 if (rc == -ENOENT) {
1557 /* Unlinked special device node? Or just a race?
1558 * Pretend we done everything. */
1559 if (!S_ISREG(inode->i_mode) &&
1560 !S_ISDIR(inode->i_mode)) {
1561 ia_valid = op_data->op_attr.ia_valid;
1562 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1563 rc = simple_setattr(dentry, &op_data->op_attr);
1564 op_data->op_attr.ia_valid = ia_valid;
1566 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1567 CERROR("md_setattr fails: rc = %d\n", rc);
1572 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1573 sbi->ll_md_exp, &md);
1575 ptlrpc_req_finished(request);
1579 ia_valid = op_data->op_attr.ia_valid;
1580 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1581 * cache is not cleared yet. */
1582 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1583 rc = simple_setattr(dentry, &op_data->op_attr);
1584 op_data->op_attr.ia_valid = ia_valid;
1586 /* Extract epoch data if obtained. */
1587 op_data->op_handle = md.body->handle;
1588 op_data->op_ioepoch = md.body->ioepoch;
1590 ll_update_inode(inode, &md);
1591 ptlrpc_req_finished(request);
1596 /* Close IO epoch and send Size-on-MDS attribute update. */
1597 static int ll_setattr_done_writing(struct inode *inode,
1598 struct md_op_data *op_data,
1599 struct md_open_data *mod)
1601 struct ll_inode_info *lli = ll_i2info(inode);
1605 LASSERT(op_data != NULL);
1606 if (!S_ISREG(inode->i_mode))
1609 CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1610 op_data->op_ioepoch, PFID(&lli->lli_fid));
1612 op_data->op_flags = MF_EPOCH_CLOSE;
1613 ll_done_writing_attr(inode, op_data);
1614 ll_pack_inode2opdata(inode, op_data, NULL);
1616 rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1617 if (rc == -EAGAIN) {
1618 /* MDS has instructed us to obtain Size-on-MDS attribute
1619 * from OSTs and send setattr to back to MDS. */
1620 rc = ll_som_update(inode, op_data);
1622 CERROR("%s: inode "DFID" mdc truncate failed: rc = %d\n",
1623 ll_i2sbi(inode)->ll_md_exp->exp_obd->obd_name,
1624 PFID(ll_inode2fid(inode)), rc);
1629 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1631 struct obd_capa *capa;
1634 if (attr->ia_valid & ATTR_SIZE)
1635 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1637 capa = ll_mdscapa_get(inode);
1639 rc = cl_setattr_ost(inode, attr, capa);
1641 if (attr->ia_valid & ATTR_SIZE)
1642 ll_truncate_free_capa(capa);
1649 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1650 * object(s) determine the file size and mtime. Otherwise, the MDS will
1651 * keep these values until such a time that objects are allocated for it.
1652 * We do the MDS operations first, as it is checking permissions for us.
1653 * We don't to the MDS RPC if there is nothing that we want to store there,
1654 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1655 * going to do an RPC anyways.
1657 * If we are doing a truncate, we will send the mtime and ctime updates
1658 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1659 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1662 * In case of HSMimport, we only set attr on MDS.
1664 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1666 struct inode *inode = dentry->d_inode;
1667 struct ll_inode_info *lli = ll_i2info(inode);
1668 struct md_op_data *op_data = NULL;
1669 struct md_open_data *mod = NULL;
1670 bool file_is_released = false;
1671 int rc = 0, rc1 = 0;
1674 CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1675 "valid %x, hsm_import %d\n",
1676 ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1677 inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1680 if (attr->ia_valid & ATTR_SIZE) {
1681 /* Check new size against VFS/VM file size limit and rlimit */
1682 rc = inode_newsize_ok(inode, attr->ia_size);
1686 /* The maximum Lustre file size is variable, based on the
1687 * OST maximum object size and number of stripes. This
1688 * needs another check in addition to the VFS check above. */
1689 if (attr->ia_size > ll_file_maxbytes(inode)) {
1690 CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1691 PFID(&lli->lli_fid), attr->ia_size,
1692 ll_file_maxbytes(inode));
1696 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1699 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1700 if (attr->ia_valid & TIMES_SET_FLAGS) {
1701 if (current_fsuid() != inode->i_uid &&
1702 !cfs_capable(CFS_CAP_FOWNER))
1706 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1707 if (attr->ia_valid & ATTR_CTIME) {
1708 attr->ia_ctime = CFS_CURRENT_TIME;
1709 attr->ia_valid |= ATTR_CTIME_SET;
1711 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1712 (attr->ia_valid & ATTR_ATIME)) {
1713 attr->ia_atime = CFS_CURRENT_TIME;
1714 attr->ia_valid |= ATTR_ATIME_SET;
1716 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1717 (attr->ia_valid & ATTR_MTIME)) {
1718 attr->ia_mtime = CFS_CURRENT_TIME;
1719 attr->ia_valid |= ATTR_MTIME_SET;
1722 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1723 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1724 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1725 cfs_time_current_sec());
1727 /* We always do an MDS RPC, even if we're only changing the size;
1728 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1730 OBD_ALLOC_PTR(op_data);
1731 if (op_data == NULL)
1734 if (!S_ISDIR(inode->i_mode)) {
1735 if (attr->ia_valid & ATTR_SIZE)
1736 inode_dio_write_done(inode);
1737 mutex_unlock(&inode->i_mutex);
1740 /* truncate on a released file must failed with -ENODATA,
1741 * so size must not be set on MDS for released file
1742 * but other attributes must be set
1744 if (S_ISREG(inode->i_mode)) {
1745 struct lov_stripe_md *lsm;
1748 ll_layout_refresh(inode, &gen);
1749 lsm = ccc_inode_lsm_get(inode);
1750 if (lsm && lsm->lsm_pattern & LOV_PATTERN_F_RELEASED)
1751 file_is_released = true;
1752 ccc_inode_lsm_put(inode, lsm);
1754 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1755 if (file_is_released) {
1756 rc = ll_layout_restore(inode, 0, attr->ia_size);
1760 file_is_released = false;
1761 ll_layout_refresh(inode, &gen);
1764 /* If we are changing file size, file content is
1765 * modified, flag it. */
1766 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1767 spin_lock(&lli->lli_lock);
1768 lli->lli_flags |= LLIF_DATA_MODIFIED;
1769 spin_unlock(&lli->lli_lock);
1770 op_data->op_bias |= MDS_DATA_MODIFIED;
1774 memcpy(&op_data->op_attr, attr, sizeof(*attr));
1776 /* Open epoch for truncate. */
1777 if (exp_connect_som(ll_i2mdexp(inode)) && !hsm_import &&
1778 (attr->ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1779 op_data->op_flags = MF_EPOCH_OPEN;
1781 rc = ll_md_setattr(dentry, op_data, &mod);
1785 /* RPC to MDT is sent, cancel data modification flag */
1786 if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) {
1787 spin_lock(&lli->lli_lock);
1788 lli->lli_flags &= ~LLIF_DATA_MODIFIED;
1789 spin_unlock(&lli->lli_lock);
1792 ll_ioepoch_open(lli, op_data->op_ioepoch);
1793 if (!S_ISREG(inode->i_mode) || file_is_released)
1796 if (attr->ia_valid & (ATTR_SIZE |
1797 ATTR_ATIME | ATTR_ATIME_SET |
1798 ATTR_MTIME | ATTR_MTIME_SET)) {
1799 /* For truncate and utimes sending attributes to OSTs, setting
1800 * mtime/atime to the past will be performed under PW [0:EOF]
1801 * extent lock (new_size:EOF for truncate). It may seem
1802 * excessive to send mtime/atime updates to OSTs when not
1803 * setting times to past, but it is necessary due to possible
1804 * time de-synchronization between MDT inode and OST objects */
1805 if (attr->ia_valid & ATTR_SIZE)
1806 down_write(&lli->lli_trunc_sem);
1807 rc = ll_setattr_ost(inode, attr);
1808 if (attr->ia_valid & ATTR_SIZE)
1809 up_write(&lli->lli_trunc_sem);
1814 if (op_data->op_ioepoch) {
1815 rc1 = ll_setattr_done_writing(inode, op_data, mod);
1819 ll_finish_md_op_data(op_data);
1821 if (!S_ISDIR(inode->i_mode)) {
1822 mutex_lock(&inode->i_mutex);
1823 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1824 inode_dio_wait(inode);
1827 ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1828 LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1833 int ll_setattr(struct dentry *de, struct iattr *attr)
1835 int mode = de->d_inode->i_mode;
1837 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1838 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1839 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1841 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1842 (ATTR_SIZE|ATTR_MODE)) &&
1843 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1844 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1845 !(attr->ia_mode & S_ISGID))))
1846 attr->ia_valid |= ATTR_FORCE;
1848 if ((mode & S_ISUID) &&
1849 !(attr->ia_mode & S_ISUID) &&
1850 !(attr->ia_valid & ATTR_KILL_SUID))
1851 attr->ia_valid |= ATTR_KILL_SUID;
1853 if (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1854 !(attr->ia_mode & S_ISGID) &&
1855 !(attr->ia_valid & ATTR_KILL_SGID))
1856 attr->ia_valid |= ATTR_KILL_SGID;
1858 return ll_setattr_raw(de, attr, false);
1861 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1862 __u64 max_age, __u32 flags)
1864 struct ll_sb_info *sbi = ll_s2sbi(sb);
1865 struct obd_statfs obd_osfs;
1869 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1871 CERROR("md_statfs fails: rc = %d\n", rc);
1875 osfs->os_type = sb->s_magic;
1877 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1878 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1880 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1881 flags |= OBD_STATFS_NODELAY;
1883 rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1885 CERROR("obd_statfs fails: rc = %d\n", rc);
1889 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1890 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1893 osfs->os_bsize = obd_osfs.os_bsize;
1894 osfs->os_blocks = obd_osfs.os_blocks;
1895 osfs->os_bfree = obd_osfs.os_bfree;
1896 osfs->os_bavail = obd_osfs.os_bavail;
1898 /* If we don't have as many objects free on the OST as inodes
1899 * on the MDS, we reduce the total number of inodes to
1900 * compensate, so that the "inodes in use" number is correct.
1902 if (obd_osfs.os_ffree < osfs->os_ffree) {
1903 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1905 osfs->os_ffree = obd_osfs.os_ffree;
1910 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1912 struct super_block *sb = de->d_sb;
1913 struct obd_statfs osfs;
1914 __u64 fsid = huge_encode_dev(sb->s_dev);
1917 CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1918 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1920 /* Some amount of caching on the client is allowed */
1921 rc = ll_statfs_internal(sb, &osfs,
1922 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1927 statfs_unpack(sfs, &osfs);
1929 /* We need to downshift for all 32-bit kernels, because we can't
1930 * tell if the kernel is being called via sys_statfs64() or not.
1931 * Stop before overflowing f_bsize - in which case it is better
1932 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1933 if (sizeof(long) < 8) {
1934 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1937 osfs.os_blocks >>= 1;
1938 osfs.os_bfree >>= 1;
1939 osfs.os_bavail >>= 1;
1943 sfs->f_blocks = osfs.os_blocks;
1944 sfs->f_bfree = osfs.os_bfree;
1945 sfs->f_bavail = osfs.os_bavail;
1946 sfs->f_fsid.val[0] = (__u32)fsid;
1947 sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1951 void ll_inode_size_lock(struct inode *inode)
1953 struct ll_inode_info *lli;
1955 LASSERT(!S_ISDIR(inode->i_mode));
1957 lli = ll_i2info(inode);
1958 LASSERT(lli->lli_size_sem_owner != current);
1959 down(&lli->lli_size_sem);
1960 LASSERT(lli->lli_size_sem_owner == NULL);
1961 lli->lli_size_sem_owner = current;
1964 void ll_inode_size_unlock(struct inode *inode)
1966 struct ll_inode_info *lli;
1968 lli = ll_i2info(inode);
1969 LASSERT(lli->lli_size_sem_owner == current);
1970 lli->lli_size_sem_owner = NULL;
1971 up(&lli->lli_size_sem);
1974 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1976 struct ll_inode_info *lli = ll_i2info(inode);
1977 struct mdt_body *body = md->body;
1978 struct lov_stripe_md *lsm = md->lsm;
1979 struct ll_sb_info *sbi = ll_i2sbi(inode);
1981 LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1983 if (!lli->lli_has_smd &&
1984 !(sbi->ll_flags & LL_SBI_LAYOUT_LOCK))
1985 cl_file_inode_init(inode, md);
1987 lli->lli_maxbytes = lsm->lsm_maxbytes;
1988 if (lli->lli_maxbytes > MAX_LFS_FILESIZE)
1989 lli->lli_maxbytes = MAX_LFS_FILESIZE;
1992 if (S_ISDIR(inode->i_mode))
1993 ll_update_lsm_md(inode, md);
1995 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1996 if (body->valid & OBD_MD_FLRMTPERM)
1997 ll_update_remote_perm(inode, md->remote_perm);
1999 #ifdef CONFIG_FS_POSIX_ACL
2000 else if (body->valid & OBD_MD_FLACL) {
2001 spin_lock(&lli->lli_lock);
2002 if (lli->lli_posix_acl)
2003 posix_acl_release(lli->lli_posix_acl);
2004 lli->lli_posix_acl = md->posix_acl;
2005 spin_unlock(&lli->lli_lock);
2008 inode->i_ino = cl_fid_build_ino(&body->fid1,
2009 sbi->ll_flags & LL_SBI_32BIT_API);
2010 inode->i_generation = cl_fid_build_gen(&body->fid1);
2012 if (body->valid & OBD_MD_FLATIME) {
2013 if (body->atime > LTIME_S(inode->i_atime))
2014 LTIME_S(inode->i_atime) = body->atime;
2015 lli->lli_lvb.lvb_atime = body->atime;
2017 if (body->valid & OBD_MD_FLMTIME) {
2018 if (body->mtime > LTIME_S(inode->i_mtime)) {
2019 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
2020 "to "LPU64"\n", inode->i_ino,
2021 LTIME_S(inode->i_mtime), body->mtime);
2022 LTIME_S(inode->i_mtime) = body->mtime;
2024 lli->lli_lvb.lvb_mtime = body->mtime;
2026 if (body->valid & OBD_MD_FLCTIME) {
2027 if (body->ctime > LTIME_S(inode->i_ctime))
2028 LTIME_S(inode->i_ctime) = body->ctime;
2029 lli->lli_lvb.lvb_ctime = body->ctime;
2031 if (body->valid & OBD_MD_FLMODE)
2032 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
2033 if (body->valid & OBD_MD_FLTYPE)
2034 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
2035 LASSERT(inode->i_mode != 0);
2036 if (S_ISREG(inode->i_mode)) {
2037 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
2039 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
2041 if (body->valid & OBD_MD_FLUID)
2042 inode->i_uid = body->uid;
2043 if (body->valid & OBD_MD_FLGID)
2044 inode->i_gid = body->gid;
2045 if (body->valid & OBD_MD_FLFLAGS)
2046 inode->i_flags = ll_ext_to_inode_flags(body->flags);
2047 if (body->valid & OBD_MD_FLNLINK)
2048 set_nlink(inode, body->nlink);
2049 if (body->valid & OBD_MD_FLRDEV)
2050 inode->i_rdev = old_decode_dev(body->rdev);
2052 if (body->valid & OBD_MD_FLID) {
2053 /* FID shouldn't be changed! */
2054 if (fid_is_sane(&lli->lli_fid)) {
2055 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
2056 "Trying to change FID "DFID
2057 " to the "DFID", inode "DFID"(%p)\n",
2058 PFID(&lli->lli_fid), PFID(&body->fid1),
2059 PFID(ll_inode2fid(inode)), inode);
2061 lli->lli_fid = body->fid1;
2065 LASSERT(fid_seq(&lli->lli_fid) != 0);
2067 if (body->valid & OBD_MD_FLSIZE) {
2068 if (exp_connect_som(ll_i2mdexp(inode)) &&
2069 S_ISREG(inode->i_mode)) {
2070 struct lustre_handle lockh;
2073 /* As it is possible a blocking ast has been processed
2074 * by this time, we need to check there is an UPDATE
2075 * lock on the client and set LLIF_MDS_SIZE_LOCK holding
2077 mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
2078 &lockh, LDLM_FL_CBPENDING,
2082 if (lli->lli_flags & (LLIF_DONE_WRITING |
2083 LLIF_EPOCH_PENDING |
2085 CERROR("%s: inode "DFID" flags %u still"
2086 " has size authority! do not "
2087 "trust the size from MDS\n",
2088 sbi->ll_md_exp->exp_obd->obd_name,
2089 PFID(ll_inode2fid(inode)),
2092 /* Use old size assignment to avoid
2093 * deadlock bz14138 & bz14326 */
2094 i_size_write(inode, body->size);
2095 spin_lock(&lli->lli_lock);
2096 lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
2097 spin_unlock(&lli->lli_lock);
2099 ldlm_lock_decref(&lockh, mode);
2102 /* Use old size assignment to avoid
2103 * deadlock bz14138 & bz14326 */
2104 i_size_write(inode, body->size);
2107 "inode="DFID", updating i_size %llu\n",
2108 PFID(ll_inode2fid(inode)),
2109 (unsigned long long)body->size);
2112 if (body->valid & OBD_MD_FLBLOCKS)
2113 inode->i_blocks = body->blocks;
2116 if (body->valid & OBD_MD_FLMDSCAPA) {
2117 LASSERT(md->mds_capa);
2118 ll_add_capa(inode, md->mds_capa);
2120 if (body->valid & OBD_MD_FLOSSCAPA) {
2121 LASSERT(md->oss_capa);
2122 ll_add_capa(inode, md->oss_capa);
2125 if (body->valid & OBD_MD_TSTATE) {
2126 if (body->t_state & MS_RESTORE)
2127 lli->lli_flags |= LLIF_FILE_RESTORING;
2131 void ll_read_inode2(struct inode *inode, void *opaque)
2133 struct lustre_md *md = opaque;
2134 struct ll_inode_info *lli = ll_i2info(inode);
2137 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2138 PFID(&lli->lli_fid), inode);
2140 LASSERT(!lli->lli_has_smd);
2142 /* Core attributes from the MDS first. This is a new inode, and
2143 * the VFS doesn't zero times in the core inode so we have to do
2144 * it ourselves. They will be overwritten by either MDS or OST
2145 * attributes - we just need to make sure they aren't newer. */
2146 LTIME_S(inode->i_mtime) = 0;
2147 LTIME_S(inode->i_atime) = 0;
2148 LTIME_S(inode->i_ctime) = 0;
2150 ll_update_inode(inode, md);
2152 /* OIDEBUG(inode); */
2154 /* initializing backing dev info. */
2155 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2158 if (S_ISREG(inode->i_mode)) {
2159 struct ll_sb_info *sbi = ll_i2sbi(inode);
2160 inode->i_op = &ll_file_inode_operations;
2161 inode->i_fop = sbi->ll_fop;
2162 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2164 } else if (S_ISDIR(inode->i_mode)) {
2165 inode->i_op = &ll_dir_inode_operations;
2166 inode->i_fop = &ll_dir_operations;
2168 } else if (S_ISLNK(inode->i_mode)) {
2169 inode->i_op = &ll_fast_symlink_inode_operations;
2172 inode->i_op = &ll_special_inode_operations;
2174 init_special_inode(inode, inode->i_mode,
2181 void ll_delete_inode(struct inode *inode)
2183 struct cl_inode_info *lli = cl_i2info(inode);
2186 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2187 /* discard all dirty pages before truncating them, required by
2188 * osc_extent implementation at LU-1030. */
2189 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF,
2190 CL_FSYNC_DISCARD, 1);
2192 truncate_inode_pages(&inode->i_data, 0);
2194 /* Workaround for LU-118 */
2195 if (inode->i_data.nrpages) {
2196 spin_lock_irq(&inode->i_data.tree_lock);
2197 spin_unlock_irq(&inode->i_data.tree_lock);
2198 LASSERTF(inode->i_data.nrpages == 0,
2199 "inode="DFID"(%p) nrpages=%lu, see "
2200 "http://jira.whamcloud.com/browse/LU-118\n",
2201 PFID(ll_inode2fid(inode)), inode,
2202 inode->i_data.nrpages);
2204 /* Workaround end */
2206 #ifdef HAVE_SBOPS_EVICT_INODE
2207 ll_clear_inode(inode);
2214 int ll_iocontrol(struct inode *inode, struct file *file,
2215 unsigned int cmd, unsigned long arg)
2217 struct ll_sb_info *sbi = ll_i2sbi(inode);
2218 struct ptlrpc_request *req = NULL;
2223 case FSFILT_IOC_GETFLAGS: {
2224 struct mdt_body *body;
2225 struct md_op_data *op_data;
2227 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2228 0, 0, LUSTRE_OPC_ANY,
2230 if (IS_ERR(op_data))
2231 RETURN(PTR_ERR(op_data));
2233 op_data->op_valid = OBD_MD_FLFLAGS;
2234 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2235 ll_finish_md_op_data(op_data);
2237 CERROR("%s: failure inode "DFID": rc = %d\n",
2238 sbi->ll_md_exp->exp_obd->obd_name,
2239 PFID(ll_inode2fid(inode)), rc);
2243 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2245 flags = body->flags;
2247 ptlrpc_req_finished(req);
2249 RETURN(put_user(flags, (int *)arg));
2251 case FSFILT_IOC_SETFLAGS: {
2252 struct lov_stripe_md *lsm;
2253 struct obd_info oinfo = { { { 0 } } };
2254 struct md_op_data *op_data;
2256 if (get_user(flags, (int *)arg))
2259 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2260 LUSTRE_OPC_ANY, NULL);
2261 if (IS_ERR(op_data))
2262 RETURN(PTR_ERR(op_data));
2264 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
2265 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2266 rc = md_setattr(sbi->ll_md_exp, op_data,
2267 NULL, 0, NULL, 0, &req, NULL);
2268 ll_finish_md_op_data(op_data);
2269 ptlrpc_req_finished(req);
2273 inode->i_flags = ll_ext_to_inode_flags(flags);
2275 lsm = ccc_inode_lsm_get(inode);
2276 if (!lsm_has_objects(lsm)) {
2277 ccc_inode_lsm_put(inode, lsm);
2281 OBDO_ALLOC(oinfo.oi_oa);
2283 ccc_inode_lsm_put(inode, lsm);
2287 oinfo.oi_oa->o_oi = lsm->lsm_oi;
2288 oinfo.oi_oa->o_flags = flags;
2289 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
2291 oinfo.oi_capa = ll_mdscapa_get(inode);
2292 obdo_set_parent_fid(oinfo.oi_oa, &ll_i2info(inode)->lli_fid);
2293 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
2294 capa_put(oinfo.oi_capa);
2295 OBDO_FREE(oinfo.oi_oa);
2296 ccc_inode_lsm_put(inode, lsm);
2298 if (rc && rc != -EPERM && rc != -EACCES)
2299 CERROR("osc_setattr_async fails: rc = %d\n", rc);
2310 int ll_flush_ctx(struct inode *inode)
2312 struct ll_sb_info *sbi = ll_i2sbi(inode);
2314 CDEBUG(D_SEC, "flush context for user %d\n", current_uid());
2316 obd_set_info_async(NULL, sbi->ll_md_exp,
2317 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2319 obd_set_info_async(NULL, sbi->ll_dt_exp,
2320 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2325 /* umount -f client means force down, don't save state */
2326 void ll_umount_begin(struct super_block *sb)
2328 struct ll_sb_info *sbi = ll_s2sbi(sb);
2329 struct obd_device *obd;
2330 struct obd_ioctl_data *ioc_data;
2333 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2334 sb->s_count, atomic_read(&sb->s_active));
2336 obd = class_exp2obd(sbi->ll_md_exp);
2338 CERROR("Invalid MDC connection handle "LPX64"\n",
2339 sbi->ll_md_exp->exp_handle.h_cookie);
2345 obd = class_exp2obd(sbi->ll_dt_exp);
2347 CERROR("Invalid LOV connection handle "LPX64"\n",
2348 sbi->ll_dt_exp->exp_handle.h_cookie);
2354 OBD_ALLOC_PTR(ioc_data);
2356 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2357 sizeof *ioc_data, ioc_data, NULL);
2359 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2360 sizeof *ioc_data, ioc_data, NULL);
2362 OBD_FREE_PTR(ioc_data);
2365 /* Really, we'd like to wait until there are no requests outstanding,
2366 * and then continue. For now, we just invalidate the requests,
2367 * schedule() and sleep one second if needed, and hope.
2373 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2375 struct ll_sb_info *sbi = ll_s2sbi(sb);
2376 char *profilenm = get_profile_name(sb);
2380 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2381 read_only = *flags & MS_RDONLY;
2382 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2383 sizeof(KEY_READ_ONLY),
2384 KEY_READ_ONLY, sizeof(read_only),
2387 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2388 profilenm, read_only ?
2389 "read-only" : "read-write", err);
2394 sb->s_flags |= MS_RDONLY;
2396 sb->s_flags &= ~MS_RDONLY;
2398 if (sbi->ll_flags & LL_SBI_VERBOSE)
2399 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2400 read_only ? "read-only" : "read-write");
2405 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2406 struct super_block *sb, struct lookup_intent *it)
2408 struct ll_sb_info *sbi = NULL;
2409 struct lustre_md md = { 0 };
2413 LASSERT(*inode || sb);
2414 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2415 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2416 sbi->ll_md_exp, &md);
2421 ll_update_inode(*inode, &md);
2423 LASSERT(sb != NULL);
2426 * At this point server returns to client's same fid as client
2427 * generated for creating. So using ->fid1 is okay here.
2429 LASSERT(fid_is_sane(&md.body->fid1));
2431 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1,
2432 sbi->ll_flags & LL_SBI_32BIT_API),
2434 if (*inode == NULL || IS_ERR(*inode)) {
2435 #ifdef CONFIG_FS_POSIX_ACL
2437 posix_acl_release(md.posix_acl);
2438 md.posix_acl = NULL;
2441 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2443 CERROR("new_inode -fatal: rc %d\n", rc);
2448 /* Handling piggyback layout lock.
2449 * Layout lock can be piggybacked by getattr and open request.
2450 * The lsm can be applied to inode only if it comes with a layout lock
2451 * otherwise correct layout may be overwritten, for example:
2452 * 1. proc1: mdt returns a lsm but not granting layout
2453 * 2. layout was changed by another client
2454 * 3. proc2: refresh layout and layout lock granted
2455 * 4. proc1: to apply a stale layout */
2456 if (it != NULL && it->d.lustre.it_lock_mode != 0) {
2457 struct lustre_handle lockh;
2458 struct ldlm_lock *lock;
2460 lockh.cookie = it->d.lustre.it_lock_handle;
2461 lock = ldlm_handle2lock(&lockh);
2462 LASSERT(lock != NULL);
2463 if (ldlm_has_layout(lock)) {
2464 struct cl_object_conf conf;
2466 memset(&conf, 0, sizeof(conf));
2467 conf.coc_opc = OBJECT_CONF_SET;
2468 conf.coc_inode = *inode;
2469 conf.coc_lock = lock;
2470 conf.u.coc_md = &md;
2471 (void)ll_layout_conf(*inode, &conf);
2473 LDLM_LOCK_PUT(lock);
2478 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2479 md_free_lustre_md(sbi->ll_md_exp, &md);
2483 int ll_obd_statfs(struct inode *inode, void *arg)
2485 struct ll_sb_info *sbi = NULL;
2486 struct obd_export *exp;
2488 struct obd_ioctl_data *data = NULL;
2493 if (!inode || !(sbi = ll_i2sbi(inode)))
2494 GOTO(out_statfs, rc = -EINVAL);
2496 rc = obd_ioctl_getdata(&buf, &len, arg);
2498 GOTO(out_statfs, rc);
2501 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2502 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2503 GOTO(out_statfs, rc = -EINVAL);
2505 if (data->ioc_inllen1 != sizeof(__u32) ||
2506 data->ioc_inllen2 != sizeof(__u32) ||
2507 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2508 data->ioc_plen2 != sizeof(struct obd_uuid))
2509 GOTO(out_statfs, rc = -EINVAL);
2511 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2512 if (type & LL_STATFS_LMV)
2513 exp = sbi->ll_md_exp;
2514 else if (type & LL_STATFS_LOV)
2515 exp = sbi->ll_dt_exp;
2517 GOTO(out_statfs, rc = -ENODEV);
2519 flags = (type & LL_STATFS_NODELAY) ? OBD_STATFS_NODELAY : 0;
2520 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, &flags);
2522 GOTO(out_statfs, rc);
2525 obd_ioctl_freedata(buf, len);
2529 int ll_process_config(struct lustre_cfg *lcfg)
2531 struct super_block *sb;
2536 /* The instance name contains the sb: lustre-client-aacfe000 */
2537 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2538 if (!ptr || !*(++ptr))
2540 if (sscanf(ptr, "%lx", &x) != 1)
2542 sb = (struct super_block *)x;
2543 /* This better be a real Lustre superblock! */
2544 LASSERT(s2lsi(sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2546 /* Note we have not called client_common_fill_super yet, so
2547 proc fns must be able to handle that! */
2548 rc = class_process_proc_seq_param(PARAM_LLITE, lprocfs_llite_obd_vars,
2555 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2556 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2557 struct inode *i1, struct inode *i2,
2558 const char *name, int namelen,
2559 int mode, __u32 opc, void *data)
2561 LASSERT(i1 != NULL);
2563 if (namelen > ll_i2sbi(i1)->ll_namelen)
2564 return ERR_PTR(-ENAMETOOLONG);
2566 if (op_data == NULL)
2567 OBD_ALLOC_PTR(op_data);
2569 if (op_data == NULL)
2570 return ERR_PTR(-ENOMEM);
2572 ll_i2gids(op_data->op_suppgids, i1, i2);
2573 op_data->op_fid1 = *ll_inode2fid(i1);
2574 op_data->op_capa1 = ll_mdscapa_get(i1);
2575 if (S_ISDIR(i1->i_mode))
2576 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2579 op_data->op_fid2 = *ll_inode2fid(i2);
2580 op_data->op_capa2 = ll_mdscapa_get(i2);
2581 if (S_ISDIR(i2->i_mode))
2582 op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2584 fid_zero(&op_data->op_fid2);
2585 op_data->op_capa2 = NULL;
2588 if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2589 op_data->op_cli_flags |= CLI_HASH64;
2591 if (ll_need_32bit_api(ll_i2sbi(i1)))
2592 op_data->op_cli_flags |= CLI_API32;
2594 op_data->op_name = name;
2595 op_data->op_namelen = namelen;
2596 op_data->op_mode = mode;
2597 op_data->op_mod_time = cfs_time_current_sec();
2598 op_data->op_fsuid = current_fsuid();
2599 op_data->op_fsgid = current_fsgid();
2600 op_data->op_cap = cfs_curproc_cap_pack();
2601 op_data->op_bias = 0;
2602 op_data->op_cli_flags = 0;
2603 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2604 filename_is_volatile(name, namelen, NULL))
2605 op_data->op_bias |= MDS_CREATE_VOLATILE;
2606 op_data->op_mds = 0;
2607 op_data->op_data = data;
2609 /* If the file is being opened after mknod() (normally due to NFS)
2610 * try to use the default stripe data from parent directory for
2611 * allocating OST objects. Try to pass the parent FID to MDS. */
2612 if (opc == LUSTRE_OPC_CREATE && i1 == i2 && S_ISREG(i2->i_mode) &&
2613 !ll_i2info(i2)->lli_has_smd) {
2614 struct ll_inode_info *lli = ll_i2info(i2);
2616 spin_lock(&lli->lli_lock);
2617 if (likely(!lli->lli_has_smd && !fid_is_zero(&lli->lli_pfid)))
2618 op_data->op_fid1 = lli->lli_pfid;
2619 spin_unlock(&lli->lli_lock);
2620 /** We ignore parent's capability temporary. */
2623 /* When called by ll_setattr_raw, file is i1. */
2624 if (LLIF_DATA_MODIFIED & ll_i2info(i1)->lli_flags)
2625 op_data->op_bias |= MDS_DATA_MODIFIED;
2630 void ll_finish_md_op_data(struct md_op_data *op_data)
2632 capa_put(op_data->op_capa1);
2633 capa_put(op_data->op_capa2);
2634 OBD_FREE_PTR(op_data);
2637 #ifdef HAVE_SUPEROPS_USE_DENTRY
2638 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2640 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2643 struct ll_sb_info *sbi;
2645 #ifdef HAVE_SUPEROPS_USE_DENTRY
2646 LASSERT((seq != NULL) && (dentry != NULL));
2647 sbi = ll_s2sbi(dentry->d_sb);
2649 LASSERT((seq != NULL) && (vfs != NULL));
2650 sbi = ll_s2sbi(vfs->mnt_sb);
2653 if (sbi->ll_flags & LL_SBI_NOLCK)
2654 seq_puts(seq, ",nolock");
2656 if (sbi->ll_flags & LL_SBI_FLOCK)
2657 seq_puts(seq, ",flock");
2659 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2660 seq_puts(seq, ",localflock");
2662 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2663 seq_puts(seq, ",user_xattr");
2665 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2666 seq_puts(seq, ",lazystatfs");
2668 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2669 seq_puts(seq, ",user_fid2path");
2675 * Get obd name by cmd, and copy out to user space
2677 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2679 struct ll_sb_info *sbi = ll_i2sbi(inode);
2680 struct obd_device *obd;
2683 if (cmd == OBD_IOC_GETDTNAME)
2684 obd = class_exp2obd(sbi->ll_dt_exp);
2685 else if (cmd == OBD_IOC_GETMDNAME)
2686 obd = class_exp2obd(sbi->ll_md_exp);
2693 if (copy_to_user((void *)arg, obd->obd_name,
2694 strlen(obd->obd_name) + 1))
2701 * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2702 * fsname will be returned in this buffer; otherwise, a static buffer will be
2703 * used to store the fsname and returned to caller.
2705 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2707 static char fsname_static[MTI_NAME_MAXLEN];
2708 struct lustre_sb_info *lsi = s2lsi(sb);
2713 /* this means the caller wants to use static buffer
2714 * and it doesn't care about race. Usually this is
2715 * in error reporting path */
2716 buf = fsname_static;
2717 buflen = sizeof(fsname_static);
2720 len = strlen(lsi->lsi_lmd->lmd_profile);
2721 ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2722 if (ptr && (strcmp(ptr, "-client") == 0))
2725 if (unlikely(len >= buflen))
2727 strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2733 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2740 p.mnt = current->fs->root.mnt;
2742 path = d_path(&p, buf, bufsize);
2747 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2749 char *buf, *path = NULL;
2750 struct dentry *dentry = NULL;
2751 struct ccc_object *obj = cl_inode2ccc(page->mapping->host);
2753 /* this can be called inside spin lock so use GFP_ATOMIC. */
2754 buf = (char *)__get_free_page(GFP_ATOMIC);
2756 dentry = d_find_alias(page->mapping->host);
2758 path = ll_d_path(dentry, buf, PAGE_SIZE);
2761 CWARN("%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2762 "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2763 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2764 PFID(&obj->cob_header.coh_lu.loh_fid),
2765 (path && !IS_ERR(path)) ? path : "", ioret);
2771 free_page((unsigned long)buf);