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 static LIST_HEAD(ll_super_blocks);
62 static DEFINE_SPINLOCK(ll_sb_lock);
65 #define log2(n) ffz(~(n))
68 static struct ll_sb_info *ll_init_sbi(void)
70 struct ll_sb_info *sbi = NULL;
72 unsigned long lru_page_max;
78 OBD_ALLOC(sbi, sizeof(*sbi));
82 spin_lock_init(&sbi->ll_lock);
83 mutex_init(&sbi->ll_lco.lco_lock);
84 spin_lock_init(&sbi->ll_pp_extent_lock);
85 spin_lock_init(&sbi->ll_process_lock);
86 sbi->ll_rw_stats_on = 0;
89 pages = si.totalram - si.totalhigh;
90 lru_page_max = pages / 2;
92 /* initialize ll_cache data */
93 atomic_set(&sbi->ll_cache.ccc_users, 0);
94 sbi->ll_cache.ccc_lru_max = lru_page_max;
95 atomic_set(&sbi->ll_cache.ccc_lru_left, lru_page_max);
96 spin_lock_init(&sbi->ll_cache.ccc_lru_lock);
97 CFS_INIT_LIST_HEAD(&sbi->ll_cache.ccc_lru);
99 atomic_set(&sbi->ll_cache.ccc_unstable_nr, 0);
100 init_waitqueue_head(&sbi->ll_cache.ccc_unstable_waitq);
102 sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
103 SBI_DEFAULT_READAHEAD_MAX);
104 sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
105 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
106 SBI_DEFAULT_READAHEAD_WHOLE_MAX;
107 CFS_INIT_LIST_HEAD(&sbi->ll_conn_chain);
108 CFS_INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
110 ll_generate_random_uuid(uuid);
111 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
112 CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
114 spin_lock(&ll_sb_lock);
115 cfs_list_add_tail(&sbi->ll_list, &ll_super_blocks);
116 spin_unlock(&ll_sb_lock);
118 sbi->ll_flags |= LL_SBI_VERBOSE;
119 #ifdef ENABLE_CHECKSUM
120 sbi->ll_flags |= LL_SBI_CHECKSUM;
123 #ifdef HAVE_LRU_RESIZE_SUPPORT
124 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
127 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
128 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
130 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
134 /* metadata statahead is enabled by default */
135 sbi->ll_sa_max = LL_SA_RPC_DEF;
136 atomic_set(&sbi->ll_sa_total, 0);
137 atomic_set(&sbi->ll_sa_wrong, 0);
138 atomic_set(&sbi->ll_agl_total, 0);
139 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
144 static void ll_free_sbi(struct super_block *sb)
146 struct ll_sb_info *sbi = ll_s2sbi(sb);
150 spin_lock(&ll_sb_lock);
151 cfs_list_del(&sbi->ll_list);
152 spin_unlock(&ll_sb_lock);
153 OBD_FREE(sbi, sizeof(*sbi));
158 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
159 struct vfsmount *mnt)
161 struct inode *root = 0;
162 struct ll_sb_info *sbi = ll_s2sbi(sb);
163 struct obd_device *obd;
164 struct obd_capa *oc = NULL;
165 struct obd_statfs *osfs = NULL;
166 struct ptlrpc_request *request = NULL;
167 struct obd_connect_data *data = NULL;
168 struct obd_uuid *uuid;
169 struct md_op_data *op_data;
170 struct lustre_md lmd;
172 int size, err, checksum;
175 obd = class_name2obd(md);
177 CERROR("MD %s: not setup or attached\n", md);
191 /* indicate the features supported by this client */
192 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
193 OBD_CONNECT_ATTRFID |
194 OBD_CONNECT_VERSION | OBD_CONNECT_BRW_SIZE |
195 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
196 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
197 OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 |
198 OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR |
199 OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH|
200 OBD_CONNECT_EINPROGRESS |
201 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
202 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS |
203 OBD_CONNECT_MAX_EASIZE |
204 OBD_CONNECT_FLOCK_DEAD |
205 OBD_CONNECT_DISP_STRIPE;
207 if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
208 data->ocd_connect_flags |= OBD_CONNECT_SOM;
210 #ifdef HAVE_LRU_RESIZE_SUPPORT
211 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
212 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
214 #ifdef CONFIG_FS_POSIX_ACL
215 data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK;
218 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
219 /* flag mdc connection as lightweight, only used for test
220 * purpose, use with care */
221 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
223 data->ocd_ibits_known = MDS_INODELOCK_FULL;
224 data->ocd_version = LUSTRE_VERSION_CODE;
226 if (sb->s_flags & MS_RDONLY)
227 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
228 if (sbi->ll_flags & LL_SBI_USER_XATTR)
229 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
231 #ifdef HAVE_MS_FLOCK_LOCK
232 /* force vfs to use lustre handler for flock() calls - bug 10743 */
233 sb->s_flags |= MS_FLOCK_LOCK;
235 #ifdef MS_HAS_NEW_AOPS
236 sb->s_flags |= MS_HAS_NEW_AOPS;
239 if (sbi->ll_flags & LL_SBI_FLOCK)
240 sbi->ll_fop = &ll_file_operations_flock;
241 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
242 sbi->ll_fop = &ll_file_operations;
244 sbi->ll_fop = &ll_file_operations_noflock;
247 data->ocd_connect_flags |= OBD_CONNECT_REAL;
248 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
249 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
251 data->ocd_brw_size = MD_MAX_BRW_SIZE;
253 err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
255 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
256 "recovery, of which this client is not a "
257 "part. Please wait for recovery to complete,"
258 " abort, or time out.\n", md);
261 CERROR("cannot connect to %s: rc = %d\n", md, err);
265 sbi->ll_md_exp->exp_connect_data = *data;
267 err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
268 LUSTRE_SEQ_METADATA);
270 CERROR("%s: Can't init metadata layer FID infrastructure, "
271 "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
275 /* For mount, we only need fs info from MDT0, and also in DNE, it
276 * can make sure the client can be mounted as long as MDT0 is
278 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
279 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
280 OBD_STATFS_FOR_MDT0);
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 = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
292 if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
293 valid != CLIENT_CONNECT_MDT_REQD) {
296 OBD_ALLOC_WAIT(buf, PAGE_CACHE_SIZE);
297 obd_connect_flags2str(buf, PAGE_CACHE_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, PAGE_CACHE_SIZE);
305 GOTO(out_md_fid, 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("%s: Get connect data failed: rc = %d\n",
313 sbi->ll_md_exp->exp_obd->obd_name, err);
314 GOTO(out_md_fid, err);
317 LASSERT(osfs->os_bsize);
318 sb->s_blocksize = osfs->os_bsize;
319 sb->s_blocksize_bits = log2(osfs->os_bsize);
320 sb->s_magic = LL_SUPER_MAGIC;
321 sb->s_maxbytes = MAX_LFS_FILESIZE;
322 sbi->ll_namelen = osfs->os_namelen;
323 sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
325 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
326 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
327 LCONSOLE_INFO("Disabling user_xattr feature because "
328 "it is not supported on the server\n");
329 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
332 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
334 sb->s_flags |= MS_POSIXACL;
336 sbi->ll_flags |= LL_SBI_ACL;
338 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
340 sb->s_flags &= ~MS_POSIXACL;
342 sbi->ll_flags &= ~LL_SBI_ACL;
345 if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
346 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
347 sbi->ll_flags |= LL_SBI_RMT_CLIENT;
348 LCONSOLE_INFO("client is set as remote by default.\n");
351 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
352 sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
353 LCONSOLE_INFO("client claims to be remote, but server "
354 "rejected, forced to be local.\n");
358 if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
359 LCONSOLE_INFO("client enabled MDS capability!\n");
360 sbi->ll_flags |= LL_SBI_MDS_CAPA;
363 if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
364 LCONSOLE_INFO("client enabled OSS capability!\n");
365 sbi->ll_flags |= LL_SBI_OSS_CAPA;
368 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
369 sbi->ll_flags |= LL_SBI_64BIT_HASH;
371 if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
372 sbi->ll_md_brw_size = data->ocd_brw_size;
374 sbi->ll_md_brw_size = PAGE_CACHE_SIZE;
376 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
377 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
379 if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
380 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
381 LCONSOLE_INFO("%s: disabling xattr cache due to "
382 "unknown maximum xattr size.\n", dt);
384 sbi->ll_flags |= LL_SBI_XATTR_CACHE;
385 sbi->ll_xattr_cache_enabled = 1;
389 obd = class_name2obd(dt);
391 CERROR("DT %s: not setup or attached\n", dt);
392 GOTO(out_md_fid, err = -ENODEV);
395 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
396 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
397 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
398 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
399 OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
400 OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
401 OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
402 OBD_CONNECT_MAXBYTES |
403 OBD_CONNECT_EINPROGRESS |
404 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
405 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS;
407 if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
408 data->ocd_connect_flags |= OBD_CONNECT_SOM;
410 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
411 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
412 * disabled by default, because it can still be enabled on the
413 * fly via /proc. As a consequence, we still need to come to an
414 * agreement on the supported algorithms at connect time */
415 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
417 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
418 data->ocd_cksum_types = OBD_CKSUM_ADLER;
420 data->ocd_cksum_types = cksum_types_supported_client();
423 #ifdef HAVE_LRU_RESIZE_SUPPORT
424 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
426 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
427 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
429 CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
430 "ocd_grant: %d\n", data->ocd_connect_flags,
431 data->ocd_version, data->ocd_grant);
433 obd->obd_upcall.onu_owner = &sbi->ll_lco;
434 obd->obd_upcall.onu_upcall = cl_ocd_update;
436 data->ocd_brw_size = DT_MAX_BRW_SIZE;
438 err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
441 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
442 "recovery, of which this client is not a "
443 "part. Please wait for recovery to "
444 "complete, abort, or time out.\n", dt);
447 CERROR("%s: Cannot connect to %s: rc = %d\n",
448 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
452 sbi->ll_dt_exp->exp_connect_data = *data;
454 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
455 LUSTRE_SEQ_METADATA);
457 CERROR("%s: Can't init data layer FID infrastructure, "
458 "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
462 mutex_lock(&sbi->ll_lco.lco_lock);
463 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
464 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
465 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
466 mutex_unlock(&sbi->ll_lco.lco_lock);
468 fid_zero(&sbi->ll_root_fid);
469 err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
471 CERROR("cannot mds_connect: rc = %d\n", err);
472 GOTO(out_lock_cn_cb, err);
474 if (!fid_is_sane(&sbi->ll_root_fid)) {
475 CERROR("%s: Invalid root fid "DFID" during mount\n",
476 sbi->ll_md_exp->exp_obd->obd_name,
477 PFID(&sbi->ll_root_fid));
478 GOTO(out_lock_cn_cb, err = -EINVAL);
480 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
482 sb->s_op = &lustre_super_operations;
483 #if THREAD_SIZE >= 8192 /*b=17630*/
484 sb->s_export_op = &lustre_export_operations;
488 * XXX: move this to after cbd setup? */
489 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA |
491 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
492 valid |= OBD_MD_FLRMTPERM;
493 else if (sbi->ll_flags & LL_SBI_ACL)
494 valid |= OBD_MD_FLACL;
496 OBD_ALLOC_PTR(op_data);
498 GOTO(out_lock_cn_cb, err = -ENOMEM);
500 op_data->op_fid1 = sbi->ll_root_fid;
501 op_data->op_mode = 0;
502 op_data->op_capa1 = oc;
503 op_data->op_valid = valid;
505 err = md_getattr(sbi->ll_md_exp, op_data, &request);
508 OBD_FREE_PTR(op_data);
510 CERROR("%s: md_getattr failed for root: rc = %d\n",
511 sbi->ll_md_exp->exp_obd->obd_name, err);
512 GOTO(out_lock_cn_cb, err);
515 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
516 sbi->ll_md_exp, &lmd);
518 CERROR("failed to understand root inode md: rc = %d\n", err);
519 ptlrpc_req_finished(request);
520 GOTO(out_lock_cn_cb, err);
523 LASSERT(fid_is_sane(&sbi->ll_root_fid));
524 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
525 sbi->ll_flags & LL_SBI_32BIT_API),
527 md_free_lustre_md(sbi->ll_md_exp, &lmd);
528 ptlrpc_req_finished(request);
530 if (root == NULL || IS_ERR(root)) {
532 obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
533 #ifdef CONFIG_FS_POSIX_ACL
535 posix_acl_release(lmd.posix_acl);
536 lmd.posix_acl = NULL;
539 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
541 CERROR("lustre_lite: bad iget4 for root\n");
545 err = ll_close_thread_start(&sbi->ll_lcq);
547 CERROR("cannot start close thread: rc %d\n", err);
551 #ifdef CONFIG_FS_POSIX_ACL
552 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
553 rct_init(&sbi->ll_rct);
554 et_init(&sbi->ll_et);
558 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
559 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
560 KEY_CHECKSUM, sizeof(checksum), &checksum,
564 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
565 KEY_CACHE_SET, sizeof(sbi->ll_cache),
566 &sbi->ll_cache, NULL);
568 sb->s_root = d_make_root(root);
569 if (sb->s_root == NULL) {
570 CERROR("%s: can't make root dentry\n",
571 ll_get_fsname(sb, NULL, 0));
572 GOTO(out_root, err = -ENOMEM);
574 #ifdef HAVE_DCACHE_LOCK
575 sb->s_root->d_op = &ll_d_ops;
578 sbi->ll_sdev_orig = sb->s_dev;
580 /* We set sb->s_dev equal on all lustre clients in order to support
581 * NFS export clustering. NFSD requires that the FSID be the same
583 /* s_dev is also used in lt_compare() to compare two fs, but that is
584 * only a node-local comparison. */
585 uuid = obd_get_uuid(sbi->ll_md_exp);
587 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
593 if (proc_lustre_fs_root != NULL) {
594 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
597 CERROR("%s: could not register mount in lprocfs: "
598 "rc = %d\n", ll_get_fsname(sb, NULL, 0), err);
608 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
610 obd_disconnect(sbi->ll_dt_exp);
611 sbi->ll_dt_exp = NULL;
612 /* Make sure all OScs are gone, since cl_cache is accessing sbi. */
613 obd_zombie_barrier();
615 obd_fid_fini(sbi->ll_md_exp->exp_obd);
617 obd_disconnect(sbi->ll_md_exp);
618 sbi->ll_md_exp = NULL;
627 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
631 *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
633 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
634 KEY_MAX_EASIZE, &size, lmmsize, NULL);
636 CERROR("Get max mdsize error rc %d\n", rc);
641 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
646 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
647 KEY_DEFAULT_EASIZE, &size, lmmsize, NULL);
649 CERROR("Get default mdsize error rc %d\n", rc);
654 int ll_get_max_cookiesize(struct ll_sb_info *sbi, int *lmmsize)
659 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_COOKIESIZE),
660 KEY_MAX_COOKIESIZE, &size, lmmsize, NULL);
662 CERROR("Get max cookiesize error rc %d\n", rc);
667 int ll_get_default_cookiesize(struct ll_sb_info *sbi, int *lmmsize)
672 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_COOKIESIZE),
673 KEY_DEFAULT_COOKIESIZE, &size, lmmsize, NULL);
675 CERROR("Get default cookiesize error rc %d\n", rc);
680 static void ll_dump_inode(struct inode *inode)
682 struct ll_d_hlist_node *tmp;
683 int dentry_count = 0;
685 LASSERT(inode != NULL);
687 ll_d_hlist_for_each(tmp, &inode->i_dentry)
690 CERROR("%s: inode %p dump: dev=%s fid="DFID
691 " mode=%o count=%u, %d dentries\n",
692 ll_get_fsname(inode->i_sb, NULL, 0), inode,
693 ll_i2mdexp(inode)->exp_obd->obd_name, PFID(ll_inode2fid(inode)),
694 inode->i_mode, atomic_read(&inode->i_count), dentry_count);
697 void lustre_dump_dentry(struct dentry *dentry, int recur)
699 struct list_head *tmp;
702 LASSERT(dentry != NULL);
704 list_for_each(tmp, &dentry->d_subdirs)
707 CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
708 " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
709 dentry->d_name.len, dentry->d_name.name,
710 dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
711 dentry->d_parent, dentry->d_inode, d_count(dentry),
712 dentry->d_flags, dentry->d_fsdata, subdirs);
713 if (dentry->d_inode != NULL)
714 ll_dump_inode(dentry->d_inode);
719 list_for_each(tmp, &dentry->d_subdirs) {
720 struct dentry *d = list_entry(tmp, struct dentry, d_u.d_child);
721 lustre_dump_dentry(d, recur - 1);
725 static void client_common_put_super(struct super_block *sb)
727 struct ll_sb_info *sbi = ll_s2sbi(sb);
730 #ifdef CONFIG_FS_POSIX_ACL
731 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
732 et_fini(&sbi->ll_et);
733 rct_fini(&sbi->ll_rct);
737 ll_close_thread_shutdown(sbi->ll_lcq);
741 cfs_list_del(&sbi->ll_conn_chain);
743 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
744 obd_disconnect(sbi->ll_dt_exp);
745 sbi->ll_dt_exp = NULL;
746 /* wait till all OSCs are gone, since cl_cache is accessing sbi.
748 obd_zombie_barrier();
750 lprocfs_unregister_mountpoint(sbi);
752 obd_fid_fini(sbi->ll_md_exp->exp_obd);
753 obd_disconnect(sbi->ll_md_exp);
754 sbi->ll_md_exp = NULL;
759 void ll_kill_super(struct super_block *sb)
761 struct ll_sb_info *sbi;
766 if (!(sb->s_flags & MS_ACTIVE))
770 /* we need restore s_dev from changed for clustred NFS before put_super
771 * because new kernels have cached s_dev and change sb->s_dev in
772 * put_super not affected real removing devices */
774 sb->s_dev = sbi->ll_sdev_orig;
775 sbi->ll_umounting = 1;
780 static inline int ll_set_opt(const char *opt, char *data, int fl)
782 if (strncmp(opt, data, strlen(opt)) != 0)
788 /* non-client-specific mount options are parsed in lmd_parse */
789 static int ll_options(char *options, int *flags)
792 char *s1 = options, *s2;
798 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
801 CDEBUG(D_SUPER, "next opt=%s\n", s1);
802 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
807 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
812 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
817 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
822 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
827 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
832 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 6, 51, 0)
833 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
835 /* Ignore deprecated mount option. The client will
836 * always try to mount with ACL support, whether this
837 * is used depends on whether server supports it. */
838 LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
839 "mount option 'acl'.\n");
842 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
844 LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
845 "mount option 'noacl'.\n");
849 #warning "{no}acl options have been deprecated since 1.8, please remove them"
851 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
856 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
861 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
867 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
872 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
877 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
882 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
887 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
892 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
897 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
902 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
907 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
912 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
917 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
923 s2 = strchr(s1, ',');
931 void ll_lli_init(struct ll_inode_info *lli)
933 lli->lli_inode_magic = LLI_INODE_MAGIC;
935 lli->lli_ioepoch = 0;
936 lli->lli_maxbytes = MAX_LFS_FILESIZE;
937 spin_lock_init(&lli->lli_lock);
938 lli->lli_posix_acl = NULL;
939 lli->lli_remote_perms = NULL;
940 mutex_init(&lli->lli_rmtperm_mutex);
941 /* Do not set lli_fid, it has been initialized already. */
942 fid_zero(&lli->lli_pfid);
943 INIT_LIST_HEAD(&lli->lli_close_list);
944 INIT_LIST_HEAD(&lli->lli_oss_capas);
945 atomic_set(&lli->lli_open_count, 0);
946 lli->lli_mds_capa = NULL;
947 lli->lli_rmtperm_time = 0;
948 lli->lli_pending_och = NULL;
949 lli->lli_mds_read_och = NULL;
950 lli->lli_mds_write_och = NULL;
951 lli->lli_mds_exec_och = NULL;
952 lli->lli_open_fd_read_count = 0;
953 lli->lli_open_fd_write_count = 0;
954 lli->lli_open_fd_exec_count = 0;
955 mutex_init(&lli->lli_och_mutex);
956 spin_lock_init(&lli->lli_agl_lock);
957 lli->lli_has_smd = false;
958 spin_lock_init(&lli->lli_layout_lock);
959 ll_layout_version_set(lli, LL_LAYOUT_GEN_NONE);
960 lli->lli_clob = NULL;
962 init_rwsem(&lli->lli_xattrs_list_rwsem);
963 mutex_init(&lli->lli_xattrs_enq_lock);
965 LASSERT(lli->lli_vfs_inode.i_mode != 0);
966 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
967 mutex_init(&lli->lli_readdir_mutex);
968 lli->lli_opendir_key = NULL;
970 spin_lock_init(&lli->lli_sa_lock);
971 lli->lli_opendir_pid = 0;
973 mutex_init(&lli->lli_size_mutex);
974 lli->lli_symlink_name = NULL;
975 init_rwsem(&lli->lli_trunc_sem);
976 mutex_init(&lli->lli_write_mutex);
977 init_rwsem(&lli->lli_glimpse_sem);
978 lli->lli_glimpse_time = 0;
979 CFS_INIT_LIST_HEAD(&lli->lli_agl_list);
980 lli->lli_agl_index = 0;
981 lli->lli_async_rc = 0;
983 mutex_init(&lli->lli_layout_mutex);
986 static inline int ll_bdi_register(struct backing_dev_info *bdi)
988 static atomic_t ll_bdi_num = ATOMIC_INIT(0);
990 bdi->name = "lustre";
991 return bdi_register(bdi, NULL, "lustre-%d",
992 atomic_inc_return(&ll_bdi_num));
995 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
997 struct lustre_profile *lprof = NULL;
998 struct lustre_sb_info *lsi = s2lsi(sb);
999 struct ll_sb_info *sbi;
1000 char *dt = NULL, *md = NULL;
1001 char *profilenm = get_profile_name(sb);
1002 struct config_llog_instance *cfg;
1003 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1004 const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
1008 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1014 try_module_get(THIS_MODULE);
1016 /* client additional sb info */
1017 lsi->lsi_llsbi = sbi = ll_init_sbi();
1019 module_put(THIS_MODULE);
1024 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1026 GOTO(out_free, err);
1028 err = bdi_init(&lsi->lsi_bdi);
1030 GOTO(out_free, err);
1031 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1032 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1033 err = ll_bdi_register(&lsi->lsi_bdi);
1035 GOTO(out_free, err);
1037 sb->s_bdi = &lsi->lsi_bdi;
1038 #ifndef HAVE_DCACHE_LOCK
1039 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1040 sb->s_d_op = &ll_d_ops;
1043 /* Generate a string unique to this super, in case some joker tries
1044 to mount the same fs at two mount points.
1045 Use the address of the super itself.*/
1046 cfg->cfg_instance = sb;
1047 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1048 cfg->cfg_callback = class_config_llog_handler;
1049 /* set up client obds */
1050 err = lustre_process_log(sb, profilenm, cfg);
1052 GOTO(out_free, err);
1054 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1055 lprof = class_get_profile(profilenm);
1056 if (lprof == NULL) {
1057 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1058 " read from the MGS. Does that filesystem "
1059 "exist?\n", profilenm);
1060 GOTO(out_free, err = -EINVAL);
1062 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1063 lprof->lp_md, lprof->lp_dt);
1065 OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1067 GOTO(out_free, err = -ENOMEM);
1068 sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1070 OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1072 GOTO(out_free, err = -ENOMEM);
1073 sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1075 /* connections, registrations, sb setup */
1076 err = client_common_fill_super(sb, md, dt, mnt);
1080 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1082 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1085 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1086 LCONSOLE_WARN("Mounted %s\n", profilenm);
1090 } /* ll_fill_super */
1092 void ll_put_super(struct super_block *sb)
1094 struct config_llog_instance cfg, params_cfg;
1095 struct obd_device *obd;
1096 struct lustre_sb_info *lsi = s2lsi(sb);
1097 struct ll_sb_info *sbi = ll_s2sbi(sb);
1098 char *profilenm = get_profile_name(sb);
1099 int ccc_count, next, force = 1, rc = 0;
1102 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1104 ll_print_capa_stat(sbi);
1106 cfg.cfg_instance = sb;
1107 lustre_end_log(sb, profilenm, &cfg);
1109 params_cfg.cfg_instance = sb;
1110 lustre_end_log(sb, PARAMS_FILENAME, ¶ms_cfg);
1112 if (sbi->ll_md_exp) {
1113 obd = class_exp2obd(sbi->ll_md_exp);
1115 force = obd->obd_force;
1118 /* Wait for unstable pages to be committed to stable storage */
1120 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1121 rc = l_wait_event(sbi->ll_cache.ccc_unstable_waitq,
1122 atomic_read(&sbi->ll_cache.ccc_unstable_nr) == 0,
1126 ccc_count = atomic_read(&sbi->ll_cache.ccc_unstable_nr);
1127 if (force == 0 && rc != -EINTR)
1128 LASSERTF(ccc_count == 0, "count: %i\n", ccc_count);
1131 /* We need to set force before the lov_disconnect in
1132 lustre_common_put_super, since l_d cleans up osc's as well. */
1135 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1137 obd->obd_force = force;
1142 /* Only if client_common_fill_super succeeded */
1143 client_common_put_super(sb);
1147 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1148 class_manual_cleanup(obd);
1151 if (sbi->ll_flags & LL_SBI_VERBOSE)
1152 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1155 class_del_profile(profilenm);
1157 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1158 bdi_destroy(&lsi->lsi_bdi);
1159 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1163 lsi->lsi_llsbi = NULL;
1165 lustre_common_put_super(sb);
1167 cl_env_cache_purge(~0);
1169 module_put(THIS_MODULE);
1172 } /* client_put_super */
1174 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1176 struct inode *inode = NULL;
1178 /* NOTE: we depend on atomic igrab() -bzzz */
1179 lock_res_and_lock(lock);
1180 if (lock->l_resource->lr_lvb_inode) {
1181 struct ll_inode_info * lli;
1182 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1183 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1184 inode = igrab(lock->l_resource->lr_lvb_inode);
1186 inode = lock->l_resource->lr_lvb_inode;
1187 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1188 D_WARNING, lock, "lr_lvb_inode %p is "
1189 "bogus: magic %08x",
1190 lock->l_resource->lr_lvb_inode,
1191 lli->lli_inode_magic);
1195 unlock_res_and_lock(lock);
1199 static void ll_dir_clear_lsm_md(struct inode *inode)
1201 struct ll_inode_info *lli = ll_i2info(inode);
1203 LASSERT(S_ISDIR(inode->i_mode));
1205 if (lli->lli_lsm_md != NULL) {
1206 lmv_free_memmd(lli->lli_lsm_md);
1207 lli->lli_lsm_md = NULL;
1211 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1212 const struct lu_fid *fid,
1213 struct lustre_md *md)
1215 struct ll_sb_info *sbi = ll_s2sbi(sb);
1216 struct mdt_body *body = md->body;
1217 struct inode *inode;
1221 ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1222 inode = iget_locked(sb, ino);
1223 if (inode == NULL) {
1224 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1225 ll_get_fsname(sb, NULL, 0), PFID(fid));
1226 RETURN(ERR_PTR(-ENOENT));
1229 if (inode->i_state & I_NEW) {
1230 struct ll_inode_info *lli = ll_i2info(inode);
1231 struct lmv_stripe_md *lsm = md->lmv;
1233 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1234 (body->mode & S_IFMT);
1235 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1238 LTIME_S(inode->i_mtime) = 0;
1239 LTIME_S(inode->i_atime) = 0;
1240 LTIME_S(inode->i_ctime) = 0;
1243 /* initializing backing dev info. */
1244 inode->i_mapping->backing_dev_info =
1245 &s2lsi(inode->i_sb)->lsi_bdi;
1246 inode->i_op = &ll_dir_inode_operations;
1247 inode->i_fop = &ll_dir_operations;
1248 lli->lli_fid = *fid;
1251 LASSERT(lsm != NULL);
1252 /* master stripe FID */
1253 lli->lli_pfid = lsm->lsm_md_oinfo[0].lmo_fid;
1254 CDEBUG(D_INODE, "lli %p master "DFID" slave "DFID"\n",
1255 lli, PFID(fid), PFID(&lli->lli_pfid));
1256 unlock_new_inode(inode);
1262 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1265 struct lmv_stripe_md *lsm = md->lmv;
1268 LASSERT(lsm != NULL);
1269 /* XXX sigh, this lsm_root initialization should be in
1270 * LMV layer, but it needs ll_iget right now, so we
1271 * put this here right now. */
1272 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1273 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1274 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1276 lsm->lsm_md_oinfo[i].lmo_root = inode;
1278 /* Unfortunately ll_iget will call ll_update_inode,
1279 * where the initialization of slave inode is slightly
1280 * different, so it reset lsm_md to NULL to avoid
1281 * initializing lsm for slave inode. */
1282 lsm->lsm_md_oinfo[i].lmo_root =
1283 ll_iget_anon_dir(inode->i_sb, fid, md);
1284 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1285 int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1287 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1293 /* Here is where the lsm is being initialized(fill lmo_info) after
1294 * client retrieve MD stripe information from MDT. */
1295 return md_update_lsm_md(ll_i2mdexp(inode), lsm, md->body,
1296 ll_md_blocking_ast);
1299 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1300 const struct lmv_stripe_md *lsm_md2)
1302 return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1303 lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1304 lsm_md1->lsm_md_master_mdt_index ==
1305 lsm_md2->lsm_md_master_mdt_index &&
1306 lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1307 lsm_md1->lsm_md_layout_version ==
1308 lsm_md2->lsm_md_layout_version &&
1309 strcmp(lsm_md1->lsm_md_pool_name,
1310 lsm_md2->lsm_md_pool_name) == 0;
1313 static void ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1315 struct ll_inode_info *lli = ll_i2info(inode);
1316 struct lmv_stripe_md *lsm = md->lmv;
1320 LASSERT(S_ISDIR(inode->i_mode));
1321 CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1322 PFID(ll_inode2fid(inode)));
1324 /* no striped information from request. */
1326 if (lli->lli_lsm_md == NULL) {
1328 } else if (lli->lli_lsm_md->lsm_md_magic == LMV_MAGIC_MIGRATE) {
1329 /* migration is done, the temporay MIGRATE layout has
1331 CDEBUG(D_INODE, DFID" finish migration.\n",
1332 PFID(ll_inode2fid(inode)));
1333 lmv_free_memmd(lli->lli_lsm_md);
1334 lli->lli_lsm_md = NULL;
1337 /* The lustre_md from req does not include stripeEA,
1338 * see ll_md_setattr */
1343 /* set the directory layout */
1344 if (lli->lli_lsm_md == NULL) {
1347 rc = ll_init_lsm_md(inode, md);
1349 CERROR("%s: init "DFID" failed: rc = %d\n",
1350 ll_get_fsname(inode->i_sb, NULL, 0),
1351 PFID(&lli->lli_fid), rc);
1354 lli->lli_lsm_md = lsm;
1355 /* set lsm_md to NULL, so the following free lustre_md
1356 * will not free this lsm */
1358 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1359 lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1363 /* Compare the old and new stripe information */
1364 if (!lli_lsm_md_eq(lli->lli_lsm_md, lsm)) {
1365 CERROR("inode %p %lu mismatch\n"
1366 " new(%p) vs lli_lsm_md(%p):\n"
1370 " hash_type: %x %x\n"
1373 inode, inode->i_ino, lsm, lli->lli_lsm_md,
1374 lsm->lsm_md_magic, lli->lli_lsm_md->lsm_md_magic,
1375 lsm->lsm_md_stripe_count,
1376 lli->lli_lsm_md->lsm_md_stripe_count,
1377 lsm->lsm_md_master_mdt_index,
1378 lli->lli_lsm_md->lsm_md_master_mdt_index,
1379 lsm->lsm_md_hash_type, lli->lli_lsm_md->lsm_md_hash_type,
1380 lsm->lsm_md_layout_version,
1381 lli->lli_lsm_md->lsm_md_layout_version,
1382 lsm->lsm_md_pool_name,
1383 lli->lli_lsm_md->lsm_md_pool_name);
1387 for (idx = 0; idx < lli->lli_lsm_md->lsm_md_stripe_count; idx++) {
1388 if (!lu_fid_eq(&lli->lli_lsm_md->lsm_md_oinfo[idx].lmo_fid,
1389 &lsm->lsm_md_oinfo[idx].lmo_fid)) {
1390 CERROR("%s: FID in lsm mismatch idx %d, old: "DFID
1392 ll_get_fsname(inode->i_sb, NULL, 0), idx,
1393 PFID(&lli->lli_lsm_md->lsm_md_oinfo[idx].lmo_fid),
1394 PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1399 md_update_lsm_md(ll_i2mdexp(inode), ll_i2info(inode)->lli_lsm_md,
1400 md->body, ll_md_blocking_ast);
1405 void ll_clear_inode(struct inode *inode)
1407 struct ll_inode_info *lli = ll_i2info(inode);
1408 struct ll_sb_info *sbi = ll_i2sbi(inode);
1411 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1412 PFID(ll_inode2fid(inode)), inode);
1414 if (S_ISDIR(inode->i_mode)) {
1415 /* these should have been cleared in ll_file_release */
1416 LASSERT(lli->lli_opendir_key == NULL);
1417 LASSERT(lli->lli_sai == NULL);
1418 LASSERT(lli->lli_opendir_pid == 0);
1421 spin_lock(&lli->lli_lock);
1422 ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1423 spin_unlock(&lli->lli_lock);
1424 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1426 LASSERT(!lli->lli_open_fd_write_count);
1427 LASSERT(!lli->lli_open_fd_read_count);
1428 LASSERT(!lli->lli_open_fd_exec_count);
1430 if (lli->lli_mds_write_och)
1431 ll_md_real_close(inode, FMODE_WRITE);
1432 if (lli->lli_mds_exec_och)
1433 ll_md_real_close(inode, FMODE_EXEC);
1434 if (lli->lli_mds_read_och)
1435 ll_md_real_close(inode, FMODE_READ);
1437 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1438 OBD_FREE(lli->lli_symlink_name,
1439 strlen(lli->lli_symlink_name) + 1);
1440 lli->lli_symlink_name = NULL;
1443 ll_xattr_cache_destroy(inode);
1445 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1446 LASSERT(lli->lli_posix_acl == NULL);
1447 if (lli->lli_remote_perms) {
1448 free_rmtperm_hash(lli->lli_remote_perms);
1449 lli->lli_remote_perms = NULL;
1452 #ifdef CONFIG_FS_POSIX_ACL
1453 else if (lli->lli_posix_acl) {
1454 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1455 LASSERT(lli->lli_remote_perms == NULL);
1456 posix_acl_release(lli->lli_posix_acl);
1457 lli->lli_posix_acl = NULL;
1460 lli->lli_inode_magic = LLI_INODE_DEAD;
1462 ll_clear_inode_capas(inode);
1463 if (S_ISDIR(inode->i_mode))
1464 ll_dir_clear_lsm_md(inode);
1466 LASSERT(list_empty(&lli->lli_agl_list));
1469 * XXX This has to be done before lsm is freed below, because
1470 * cl_object still uses inode lsm.
1472 cl_inode_fini(inode);
1473 lli->lli_has_smd = false;
1478 int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data,
1479 struct md_open_data **mod)
1481 struct lustre_md md;
1482 struct inode *inode = dentry->d_inode;
1483 struct ll_sb_info *sbi = ll_i2sbi(inode);
1484 struct ptlrpc_request *request = NULL;
1488 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1489 LUSTRE_OPC_ANY, NULL);
1490 if (IS_ERR(op_data))
1491 RETURN(PTR_ERR(op_data));
1493 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1496 ptlrpc_req_finished(request);
1497 if (rc == -ENOENT) {
1499 /* Unlinked special device node? Or just a race?
1500 * Pretend we done everything. */
1501 if (!S_ISREG(inode->i_mode) &&
1502 !S_ISDIR(inode->i_mode)) {
1503 ia_valid = op_data->op_attr.ia_valid;
1504 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1505 rc = simple_setattr(dentry, &op_data->op_attr);
1506 op_data->op_attr.ia_valid = ia_valid;
1508 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1509 CERROR("md_setattr fails: rc = %d\n", rc);
1514 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1515 sbi->ll_md_exp, &md);
1517 ptlrpc_req_finished(request);
1521 ia_valid = op_data->op_attr.ia_valid;
1522 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1523 * cache is not cleared yet. */
1524 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1525 rc = simple_setattr(dentry, &op_data->op_attr);
1526 op_data->op_attr.ia_valid = ia_valid;
1528 /* Extract epoch data if obtained. */
1529 op_data->op_handle = md.body->handle;
1530 op_data->op_ioepoch = md.body->ioepoch;
1532 ll_update_inode(inode, &md);
1533 ptlrpc_req_finished(request);
1538 /* Close IO epoch and send Size-on-MDS attribute update. */
1539 static int ll_setattr_done_writing(struct inode *inode,
1540 struct md_op_data *op_data,
1541 struct md_open_data *mod)
1543 struct ll_inode_info *lli = ll_i2info(inode);
1547 LASSERT(op_data != NULL);
1548 if (!S_ISREG(inode->i_mode))
1551 CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1552 op_data->op_ioepoch, PFID(&lli->lli_fid));
1554 op_data->op_flags = MF_EPOCH_CLOSE;
1555 ll_done_writing_attr(inode, op_data);
1556 ll_pack_inode2opdata(inode, op_data, NULL);
1558 rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1559 if (rc == -EAGAIN) {
1560 /* MDS has instructed us to obtain Size-on-MDS attribute
1561 * from OSTs and send setattr to back to MDS. */
1562 rc = ll_som_update(inode, op_data);
1564 CERROR("%s: inode "DFID" mdc truncate failed: rc = %d\n",
1565 ll_i2sbi(inode)->ll_md_exp->exp_obd->obd_name,
1566 PFID(ll_inode2fid(inode)), rc);
1571 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1573 struct obd_capa *capa;
1576 if (attr->ia_valid & ATTR_SIZE)
1577 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1579 capa = ll_mdscapa_get(inode);
1581 rc = cl_setattr_ost(inode, attr, capa);
1583 if (attr->ia_valid & ATTR_SIZE)
1584 ll_truncate_free_capa(capa);
1591 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1592 * object(s) determine the file size and mtime. Otherwise, the MDS will
1593 * keep these values until such a time that objects are allocated for it.
1594 * We do the MDS operations first, as it is checking permissions for us.
1595 * We don't to the MDS RPC if there is nothing that we want to store there,
1596 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1597 * going to do an RPC anyways.
1599 * If we are doing a truncate, we will send the mtime and ctime updates
1600 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1601 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1604 * In case of HSMimport, we only set attr on MDS.
1606 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1608 struct inode *inode = dentry->d_inode;
1609 struct ll_inode_info *lli = ll_i2info(inode);
1610 struct md_op_data *op_data = NULL;
1611 struct md_open_data *mod = NULL;
1612 bool file_is_released = false;
1613 int rc = 0, rc1 = 0;
1616 CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1617 "valid %x, hsm_import %d\n",
1618 ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1619 inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1622 if (attr->ia_valid & ATTR_SIZE) {
1623 /* Check new size against VFS/VM file size limit and rlimit */
1624 rc = inode_newsize_ok(inode, attr->ia_size);
1628 /* The maximum Lustre file size is variable, based on the
1629 * OST maximum object size and number of stripes. This
1630 * needs another check in addition to the VFS check above. */
1631 if (attr->ia_size > ll_file_maxbytes(inode)) {
1632 CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1633 PFID(&lli->lli_fid), attr->ia_size,
1634 ll_file_maxbytes(inode));
1638 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1641 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1642 if (attr->ia_valid & TIMES_SET_FLAGS) {
1643 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1644 !cfs_capable(CFS_CAP_FOWNER))
1648 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1649 if (attr->ia_valid & ATTR_CTIME) {
1650 attr->ia_ctime = CFS_CURRENT_TIME;
1651 attr->ia_valid |= ATTR_CTIME_SET;
1653 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1654 (attr->ia_valid & ATTR_ATIME)) {
1655 attr->ia_atime = CFS_CURRENT_TIME;
1656 attr->ia_valid |= ATTR_ATIME_SET;
1658 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1659 (attr->ia_valid & ATTR_MTIME)) {
1660 attr->ia_mtime = CFS_CURRENT_TIME;
1661 attr->ia_valid |= ATTR_MTIME_SET;
1664 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1665 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1666 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1667 cfs_time_current_sec());
1669 /* We always do an MDS RPC, even if we're only changing the size;
1670 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1672 OBD_ALLOC_PTR(op_data);
1673 if (op_data == NULL)
1676 if (!S_ISDIR(inode->i_mode)) {
1677 if (attr->ia_valid & ATTR_SIZE)
1678 inode_dio_write_done(inode);
1679 mutex_unlock(&inode->i_mutex);
1682 /* truncate on a released file must failed with -ENODATA,
1683 * so size must not be set on MDS for released file
1684 * but other attributes must be set
1686 if (S_ISREG(inode->i_mode)) {
1687 struct lov_stripe_md *lsm;
1690 ll_layout_refresh(inode, &gen);
1691 lsm = ccc_inode_lsm_get(inode);
1692 if (lsm && lsm->lsm_pattern & LOV_PATTERN_F_RELEASED)
1693 file_is_released = true;
1694 ccc_inode_lsm_put(inode, lsm);
1696 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1697 if (file_is_released) {
1698 rc = ll_layout_restore(inode, 0, attr->ia_size);
1702 file_is_released = false;
1703 ll_layout_refresh(inode, &gen);
1706 /* If we are changing file size, file content is
1707 * modified, flag it. */
1708 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1709 spin_lock(&lli->lli_lock);
1710 lli->lli_flags |= LLIF_DATA_MODIFIED;
1711 spin_unlock(&lli->lli_lock);
1712 op_data->op_bias |= MDS_DATA_MODIFIED;
1716 memcpy(&op_data->op_attr, attr, sizeof(*attr));
1718 /* Open epoch for truncate. */
1719 if (exp_connect_som(ll_i2mdexp(inode)) && !hsm_import &&
1720 (attr->ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1721 op_data->op_flags = MF_EPOCH_OPEN;
1723 rc = ll_md_setattr(dentry, op_data, &mod);
1727 /* RPC to MDT is sent, cancel data modification flag */
1728 if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) {
1729 spin_lock(&lli->lli_lock);
1730 lli->lli_flags &= ~LLIF_DATA_MODIFIED;
1731 spin_unlock(&lli->lli_lock);
1734 ll_ioepoch_open(lli, op_data->op_ioepoch);
1735 if (!S_ISREG(inode->i_mode) || file_is_released)
1738 if (attr->ia_valid & (ATTR_SIZE |
1739 ATTR_ATIME | ATTR_ATIME_SET |
1740 ATTR_MTIME | ATTR_MTIME_SET)) {
1741 /* For truncate and utimes sending attributes to OSTs, setting
1742 * mtime/atime to the past will be performed under PW [0:EOF]
1743 * extent lock (new_size:EOF for truncate). It may seem
1744 * excessive to send mtime/atime updates to OSTs when not
1745 * setting times to past, but it is necessary due to possible
1746 * time de-synchronization between MDT inode and OST objects */
1747 if (attr->ia_valid & ATTR_SIZE)
1748 down_write(&lli->lli_trunc_sem);
1749 rc = ll_setattr_ost(inode, attr);
1750 if (attr->ia_valid & ATTR_SIZE)
1751 up_write(&lli->lli_trunc_sem);
1756 if (op_data->op_ioepoch) {
1757 rc1 = ll_setattr_done_writing(inode, op_data, mod);
1761 ll_finish_md_op_data(op_data);
1763 if (!S_ISDIR(inode->i_mode)) {
1764 mutex_lock(&inode->i_mutex);
1765 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1766 inode_dio_wait(inode);
1769 ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1770 LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1775 int ll_setattr(struct dentry *de, struct iattr *attr)
1777 int mode = de->d_inode->i_mode;
1779 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1780 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1781 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1783 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1784 (ATTR_SIZE|ATTR_MODE)) &&
1785 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1786 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1787 !(attr->ia_mode & S_ISGID))))
1788 attr->ia_valid |= ATTR_FORCE;
1790 if ((mode & S_ISUID) &&
1791 !(attr->ia_mode & S_ISUID) &&
1792 !(attr->ia_valid & ATTR_KILL_SUID))
1793 attr->ia_valid |= ATTR_KILL_SUID;
1795 if (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1796 !(attr->ia_mode & S_ISGID) &&
1797 !(attr->ia_valid & ATTR_KILL_SGID))
1798 attr->ia_valid |= ATTR_KILL_SGID;
1800 return ll_setattr_raw(de, attr, false);
1803 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1804 __u64 max_age, __u32 flags)
1806 struct ll_sb_info *sbi = ll_s2sbi(sb);
1807 struct obd_statfs obd_osfs;
1811 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1813 CERROR("md_statfs fails: rc = %d\n", rc);
1817 osfs->os_type = sb->s_magic;
1819 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1820 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1822 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1823 flags |= OBD_STATFS_NODELAY;
1825 rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1827 CERROR("obd_statfs fails: rc = %d\n", rc);
1831 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1832 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1835 osfs->os_bsize = obd_osfs.os_bsize;
1836 osfs->os_blocks = obd_osfs.os_blocks;
1837 osfs->os_bfree = obd_osfs.os_bfree;
1838 osfs->os_bavail = obd_osfs.os_bavail;
1840 /* If we don't have as many objects free on the OST as inodes
1841 * on the MDS, we reduce the total number of inodes to
1842 * compensate, so that the "inodes in use" number is correct.
1844 if (obd_osfs.os_ffree < osfs->os_ffree) {
1845 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1847 osfs->os_ffree = obd_osfs.os_ffree;
1852 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1854 struct super_block *sb = de->d_sb;
1855 struct obd_statfs osfs;
1856 __u64 fsid = huge_encode_dev(sb->s_dev);
1859 CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1860 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1862 /* Some amount of caching on the client is allowed */
1863 rc = ll_statfs_internal(sb, &osfs,
1864 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1869 statfs_unpack(sfs, &osfs);
1871 /* We need to downshift for all 32-bit kernels, because we can't
1872 * tell if the kernel is being called via sys_statfs64() or not.
1873 * Stop before overflowing f_bsize - in which case it is better
1874 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1875 if (sizeof(long) < 8) {
1876 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1879 osfs.os_blocks >>= 1;
1880 osfs.os_bfree >>= 1;
1881 osfs.os_bavail >>= 1;
1885 sfs->f_blocks = osfs.os_blocks;
1886 sfs->f_bfree = osfs.os_bfree;
1887 sfs->f_bavail = osfs.os_bavail;
1888 sfs->f_fsid.val[0] = (__u32)fsid;
1889 sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1893 void ll_inode_size_lock(struct inode *inode)
1895 struct ll_inode_info *lli;
1897 LASSERT(!S_ISDIR(inode->i_mode));
1899 lli = ll_i2info(inode);
1900 mutex_lock(&lli->lli_size_mutex);
1903 void ll_inode_size_unlock(struct inode *inode)
1905 struct ll_inode_info *lli;
1907 lli = ll_i2info(inode);
1908 mutex_unlock(&lli->lli_size_mutex);
1911 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1913 struct ll_inode_info *lli = ll_i2info(inode);
1914 struct mdt_body *body = md->body;
1915 struct lov_stripe_md *lsm = md->lsm;
1916 struct ll_sb_info *sbi = ll_i2sbi(inode);
1918 LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1920 if (!lli->lli_has_smd &&
1921 !(sbi->ll_flags & LL_SBI_LAYOUT_LOCK))
1922 cl_file_inode_init(inode, md);
1924 lli->lli_maxbytes = lsm->lsm_maxbytes;
1925 if (lli->lli_maxbytes > MAX_LFS_FILESIZE)
1926 lli->lli_maxbytes = MAX_LFS_FILESIZE;
1929 if (S_ISDIR(inode->i_mode))
1930 ll_update_lsm_md(inode, md);
1932 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1933 if (body->valid & OBD_MD_FLRMTPERM)
1934 ll_update_remote_perm(inode, md->remote_perm);
1936 #ifdef CONFIG_FS_POSIX_ACL
1937 else if (body->valid & OBD_MD_FLACL) {
1938 spin_lock(&lli->lli_lock);
1939 if (lli->lli_posix_acl)
1940 posix_acl_release(lli->lli_posix_acl);
1941 lli->lli_posix_acl = md->posix_acl;
1942 spin_unlock(&lli->lli_lock);
1945 inode->i_ino = cl_fid_build_ino(&body->fid1,
1946 sbi->ll_flags & LL_SBI_32BIT_API);
1947 inode->i_generation = cl_fid_build_gen(&body->fid1);
1949 if (body->valid & OBD_MD_FLATIME) {
1950 if (body->atime > LTIME_S(inode->i_atime))
1951 LTIME_S(inode->i_atime) = body->atime;
1952 lli->lli_lvb.lvb_atime = body->atime;
1954 if (body->valid & OBD_MD_FLMTIME) {
1955 if (body->mtime > LTIME_S(inode->i_mtime)) {
1956 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1957 "to "LPU64"\n", inode->i_ino,
1958 LTIME_S(inode->i_mtime), body->mtime);
1959 LTIME_S(inode->i_mtime) = body->mtime;
1961 lli->lli_lvb.lvb_mtime = body->mtime;
1963 if (body->valid & OBD_MD_FLCTIME) {
1964 if (body->ctime > LTIME_S(inode->i_ctime))
1965 LTIME_S(inode->i_ctime) = body->ctime;
1966 lli->lli_lvb.lvb_ctime = body->ctime;
1968 if (body->valid & OBD_MD_FLMODE)
1969 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1970 if (body->valid & OBD_MD_FLTYPE)
1971 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1972 LASSERT(inode->i_mode != 0);
1973 if (S_ISREG(inode->i_mode)) {
1974 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1976 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1978 if (body->valid & OBD_MD_FLUID)
1979 inode->i_uid = make_kuid(&init_user_ns, body->uid);
1980 if (body->valid & OBD_MD_FLGID)
1981 inode->i_gid = make_kgid(&init_user_ns, body->gid);
1982 if (body->valid & OBD_MD_FLFLAGS)
1983 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1984 if (body->valid & OBD_MD_FLNLINK)
1985 set_nlink(inode, body->nlink);
1986 if (body->valid & OBD_MD_FLRDEV)
1987 inode->i_rdev = old_decode_dev(body->rdev);
1989 if (body->valid & OBD_MD_FLID) {
1990 /* FID shouldn't be changed! */
1991 if (fid_is_sane(&lli->lli_fid)) {
1992 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1993 "Trying to change FID "DFID
1994 " to the "DFID", inode "DFID"(%p)\n",
1995 PFID(&lli->lli_fid), PFID(&body->fid1),
1996 PFID(ll_inode2fid(inode)), inode);
1998 lli->lli_fid = body->fid1;
2002 LASSERT(fid_seq(&lli->lli_fid) != 0);
2004 if (body->valid & OBD_MD_FLSIZE) {
2005 if (exp_connect_som(ll_i2mdexp(inode)) &&
2006 S_ISREG(inode->i_mode)) {
2007 struct lustre_handle lockh;
2010 /* As it is possible a blocking ast has been processed
2011 * by this time, we need to check there is an UPDATE
2012 * lock on the client and set LLIF_MDS_SIZE_LOCK holding
2014 mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
2015 &lockh, LDLM_FL_CBPENDING,
2019 if (lli->lli_flags & (LLIF_DONE_WRITING |
2020 LLIF_EPOCH_PENDING |
2022 CERROR("%s: inode "DFID" flags %u still"
2023 " has size authority! do not "
2024 "trust the size from MDS\n",
2025 sbi->ll_md_exp->exp_obd->obd_name,
2026 PFID(ll_inode2fid(inode)),
2029 /* Use old size assignment to avoid
2030 * deadlock bz14138 & bz14326 */
2031 i_size_write(inode, body->size);
2032 spin_lock(&lli->lli_lock);
2033 lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
2034 spin_unlock(&lli->lli_lock);
2036 ldlm_lock_decref(&lockh, mode);
2039 /* Use old size assignment to avoid
2040 * deadlock bz14138 & bz14326 */
2041 i_size_write(inode, body->size);
2044 "inode="DFID", updating i_size %llu\n",
2045 PFID(ll_inode2fid(inode)),
2046 (unsigned long long)body->size);
2049 if (body->valid & OBD_MD_FLBLOCKS)
2050 inode->i_blocks = body->blocks;
2053 if (body->valid & OBD_MD_FLMDSCAPA) {
2054 LASSERT(md->mds_capa);
2055 ll_add_capa(inode, md->mds_capa);
2057 if (body->valid & OBD_MD_FLOSSCAPA) {
2058 LASSERT(md->oss_capa);
2059 ll_add_capa(inode, md->oss_capa);
2062 if (body->valid & OBD_MD_TSTATE) {
2063 if (body->t_state & MS_RESTORE)
2064 lli->lli_flags |= LLIF_FILE_RESTORING;
2068 void ll_read_inode2(struct inode *inode, void *opaque)
2070 struct lustre_md *md = opaque;
2071 struct ll_inode_info *lli = ll_i2info(inode);
2074 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2075 PFID(&lli->lli_fid), inode);
2077 LASSERT(!lli->lli_has_smd);
2079 /* Core attributes from the MDS first. This is a new inode, and
2080 * the VFS doesn't zero times in the core inode so we have to do
2081 * it ourselves. They will be overwritten by either MDS or OST
2082 * attributes - we just need to make sure they aren't newer. */
2083 LTIME_S(inode->i_mtime) = 0;
2084 LTIME_S(inode->i_atime) = 0;
2085 LTIME_S(inode->i_ctime) = 0;
2087 ll_update_inode(inode, md);
2089 /* OIDEBUG(inode); */
2091 /* initializing backing dev info. */
2092 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2095 if (S_ISREG(inode->i_mode)) {
2096 struct ll_sb_info *sbi = ll_i2sbi(inode);
2097 inode->i_op = &ll_file_inode_operations;
2098 inode->i_fop = sbi->ll_fop;
2099 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2101 } else if (S_ISDIR(inode->i_mode)) {
2102 inode->i_op = &ll_dir_inode_operations;
2103 inode->i_fop = &ll_dir_operations;
2105 } else if (S_ISLNK(inode->i_mode)) {
2106 inode->i_op = &ll_fast_symlink_inode_operations;
2109 inode->i_op = &ll_special_inode_operations;
2111 init_special_inode(inode, inode->i_mode,
2118 void ll_delete_inode(struct inode *inode)
2120 struct cl_inode_info *lli = cl_i2info(inode);
2123 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2124 /* discard all dirty pages before truncating them, required by
2125 * osc_extent implementation at LU-1030. */
2126 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF,
2127 CL_FSYNC_DISCARD, 1);
2129 truncate_inode_pages(&inode->i_data, 0);
2131 /* Workaround for LU-118 */
2132 if (inode->i_data.nrpages) {
2133 spin_lock_irq(&inode->i_data.tree_lock);
2134 spin_unlock_irq(&inode->i_data.tree_lock);
2135 LASSERTF(inode->i_data.nrpages == 0,
2136 "inode="DFID"(%p) nrpages=%lu, see "
2137 "http://jira.whamcloud.com/browse/LU-118\n",
2138 PFID(ll_inode2fid(inode)), inode,
2139 inode->i_data.nrpages);
2141 /* Workaround end */
2143 #ifdef HAVE_SBOPS_EVICT_INODE
2144 ll_clear_inode(inode);
2151 int ll_iocontrol(struct inode *inode, struct file *file,
2152 unsigned int cmd, unsigned long arg)
2154 struct ll_sb_info *sbi = ll_i2sbi(inode);
2155 struct ptlrpc_request *req = NULL;
2160 case FSFILT_IOC_GETFLAGS: {
2161 struct mdt_body *body;
2162 struct md_op_data *op_data;
2164 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2165 0, 0, LUSTRE_OPC_ANY,
2167 if (IS_ERR(op_data))
2168 RETURN(PTR_ERR(op_data));
2170 op_data->op_valid = OBD_MD_FLFLAGS;
2171 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2172 ll_finish_md_op_data(op_data);
2174 CERROR("%s: failure inode "DFID": rc = %d\n",
2175 sbi->ll_md_exp->exp_obd->obd_name,
2176 PFID(ll_inode2fid(inode)), rc);
2180 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2182 flags = body->flags;
2184 ptlrpc_req_finished(req);
2186 RETURN(put_user(flags, (int *)arg));
2188 case FSFILT_IOC_SETFLAGS: {
2189 struct lov_stripe_md *lsm;
2190 struct obd_info oinfo = { { { 0 } } };
2191 struct md_op_data *op_data;
2193 if (get_user(flags, (int *)arg))
2196 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2197 LUSTRE_OPC_ANY, NULL);
2198 if (IS_ERR(op_data))
2199 RETURN(PTR_ERR(op_data));
2201 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
2202 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2203 rc = md_setattr(sbi->ll_md_exp, op_data,
2204 NULL, 0, NULL, 0, &req, NULL);
2205 ll_finish_md_op_data(op_data);
2206 ptlrpc_req_finished(req);
2210 inode->i_flags = ll_ext_to_inode_flags(flags);
2212 lsm = ccc_inode_lsm_get(inode);
2213 if (!lsm_has_objects(lsm)) {
2214 ccc_inode_lsm_put(inode, lsm);
2218 OBDO_ALLOC(oinfo.oi_oa);
2220 ccc_inode_lsm_put(inode, lsm);
2224 oinfo.oi_oa->o_oi = lsm->lsm_oi;
2225 oinfo.oi_oa->o_flags = flags;
2226 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
2228 oinfo.oi_capa = ll_mdscapa_get(inode);
2229 obdo_set_parent_fid(oinfo.oi_oa, &ll_i2info(inode)->lli_fid);
2230 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
2231 capa_put(oinfo.oi_capa);
2232 OBDO_FREE(oinfo.oi_oa);
2233 ccc_inode_lsm_put(inode, lsm);
2235 if (rc && rc != -EPERM && rc != -EACCES)
2236 CERROR("osc_setattr_async fails: rc = %d\n", rc);
2247 int ll_flush_ctx(struct inode *inode)
2249 struct ll_sb_info *sbi = ll_i2sbi(inode);
2251 CDEBUG(D_SEC, "flush context for user %d\n",
2252 from_kuid(&init_user_ns, current_uid()));
2254 obd_set_info_async(NULL, sbi->ll_md_exp,
2255 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2257 obd_set_info_async(NULL, sbi->ll_dt_exp,
2258 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2263 /* umount -f client means force down, don't save state */
2264 void ll_umount_begin(struct super_block *sb)
2266 struct ll_sb_info *sbi = ll_s2sbi(sb);
2267 struct obd_device *obd;
2268 struct obd_ioctl_data *ioc_data;
2271 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2272 sb->s_count, atomic_read(&sb->s_active));
2274 obd = class_exp2obd(sbi->ll_md_exp);
2276 CERROR("Invalid MDC connection handle "LPX64"\n",
2277 sbi->ll_md_exp->exp_handle.h_cookie);
2283 obd = class_exp2obd(sbi->ll_dt_exp);
2285 CERROR("Invalid LOV connection handle "LPX64"\n",
2286 sbi->ll_dt_exp->exp_handle.h_cookie);
2292 OBD_ALLOC_PTR(ioc_data);
2294 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2295 sizeof *ioc_data, ioc_data, NULL);
2297 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2298 sizeof *ioc_data, ioc_data, NULL);
2300 OBD_FREE_PTR(ioc_data);
2303 /* Really, we'd like to wait until there are no requests outstanding,
2304 * and then continue. For now, we just invalidate the requests,
2305 * schedule() and sleep one second if needed, and hope.
2311 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2313 struct ll_sb_info *sbi = ll_s2sbi(sb);
2314 char *profilenm = get_profile_name(sb);
2318 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2319 read_only = *flags & MS_RDONLY;
2320 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2321 sizeof(KEY_READ_ONLY),
2322 KEY_READ_ONLY, sizeof(read_only),
2325 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2326 profilenm, read_only ?
2327 "read-only" : "read-write", err);
2332 sb->s_flags |= MS_RDONLY;
2334 sb->s_flags &= ~MS_RDONLY;
2336 if (sbi->ll_flags & LL_SBI_VERBOSE)
2337 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2338 read_only ? "read-only" : "read-write");
2343 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2344 struct super_block *sb, struct lookup_intent *it)
2346 struct ll_sb_info *sbi = NULL;
2347 struct lustre_md md = { 0 };
2351 LASSERT(*inode || sb);
2352 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2353 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2354 sbi->ll_md_exp, &md);
2359 ll_update_inode(*inode, &md);
2361 LASSERT(sb != NULL);
2364 * At this point server returns to client's same fid as client
2365 * generated for creating. So using ->fid1 is okay here.
2367 LASSERT(fid_is_sane(&md.body->fid1));
2369 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1,
2370 sbi->ll_flags & LL_SBI_32BIT_API),
2372 if (*inode == NULL || IS_ERR(*inode)) {
2373 #ifdef CONFIG_FS_POSIX_ACL
2375 posix_acl_release(md.posix_acl);
2376 md.posix_acl = NULL;
2379 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2381 CERROR("new_inode -fatal: rc %d\n", rc);
2386 /* Handling piggyback layout lock.
2387 * Layout lock can be piggybacked by getattr and open request.
2388 * The lsm can be applied to inode only if it comes with a layout lock
2389 * otherwise correct layout may be overwritten, for example:
2390 * 1. proc1: mdt returns a lsm but not granting layout
2391 * 2. layout was changed by another client
2392 * 3. proc2: refresh layout and layout lock granted
2393 * 4. proc1: to apply a stale layout */
2394 if (it != NULL && it->d.lustre.it_lock_mode != 0) {
2395 struct lustre_handle lockh;
2396 struct ldlm_lock *lock;
2398 lockh.cookie = it->d.lustre.it_lock_handle;
2399 lock = ldlm_handle2lock(&lockh);
2400 LASSERT(lock != NULL);
2401 if (ldlm_has_layout(lock)) {
2402 struct cl_object_conf conf;
2404 memset(&conf, 0, sizeof(conf));
2405 conf.coc_opc = OBJECT_CONF_SET;
2406 conf.coc_inode = *inode;
2407 conf.coc_lock = lock;
2408 conf.u.coc_md = &md;
2409 (void)ll_layout_conf(*inode, &conf);
2411 LDLM_LOCK_PUT(lock);
2416 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2417 md_free_lustre_md(sbi->ll_md_exp, &md);
2421 int ll_obd_statfs(struct inode *inode, void *arg)
2423 struct ll_sb_info *sbi = NULL;
2424 struct obd_export *exp;
2426 struct obd_ioctl_data *data = NULL;
2431 if (!inode || !(sbi = ll_i2sbi(inode)))
2432 GOTO(out_statfs, rc = -EINVAL);
2434 rc = obd_ioctl_getdata(&buf, &len, arg);
2436 GOTO(out_statfs, rc);
2439 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2440 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2441 GOTO(out_statfs, rc = -EINVAL);
2443 if (data->ioc_inllen1 != sizeof(__u32) ||
2444 data->ioc_inllen2 != sizeof(__u32) ||
2445 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2446 data->ioc_plen2 != sizeof(struct obd_uuid))
2447 GOTO(out_statfs, rc = -EINVAL);
2449 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2450 if (type & LL_STATFS_LMV)
2451 exp = sbi->ll_md_exp;
2452 else if (type & LL_STATFS_LOV)
2453 exp = sbi->ll_dt_exp;
2455 GOTO(out_statfs, rc = -ENODEV);
2457 flags = (type & LL_STATFS_NODELAY) ? OBD_STATFS_NODELAY : 0;
2458 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, &flags);
2460 GOTO(out_statfs, rc);
2463 obd_ioctl_freedata(buf, len);
2467 int ll_process_config(struct lustre_cfg *lcfg)
2469 struct super_block *sb;
2474 /* The instance name contains the sb: lustre-client-aacfe000 */
2475 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2476 if (!ptr || !*(++ptr))
2478 if (sscanf(ptr, "%lx", &x) != 1)
2480 sb = (struct super_block *)x;
2481 /* This better be a real Lustre superblock! */
2482 LASSERT(s2lsi(sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2484 /* Note we have not called client_common_fill_super yet, so
2485 proc fns must be able to handle that! */
2486 rc = class_process_proc_seq_param(PARAM_LLITE, lprocfs_llite_obd_vars,
2493 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2494 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2495 struct inode *i1, struct inode *i2,
2496 const char *name, int namelen,
2497 int mode, __u32 opc, void *data)
2499 LASSERT(i1 != NULL);
2501 if (namelen > ll_i2sbi(i1)->ll_namelen)
2502 return ERR_PTR(-ENAMETOOLONG);
2504 if (op_data == NULL)
2505 OBD_ALLOC_PTR(op_data);
2507 if (op_data == NULL)
2508 return ERR_PTR(-ENOMEM);
2510 ll_i2gids(op_data->op_suppgids, i1, i2);
2511 op_data->op_fid1 = *ll_inode2fid(i1);
2512 op_data->op_capa1 = ll_mdscapa_get(i1);
2513 if (S_ISDIR(i1->i_mode))
2514 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2517 op_data->op_fid2 = *ll_inode2fid(i2);
2518 op_data->op_capa2 = ll_mdscapa_get(i2);
2519 if (S_ISDIR(i2->i_mode))
2520 op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2522 fid_zero(&op_data->op_fid2);
2523 op_data->op_capa2 = NULL;
2526 if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2527 op_data->op_cli_flags |= CLI_HASH64;
2529 if (ll_need_32bit_api(ll_i2sbi(i1)))
2530 op_data->op_cli_flags |= CLI_API32;
2532 op_data->op_name = name;
2533 op_data->op_namelen = namelen;
2534 op_data->op_mode = mode;
2535 op_data->op_mod_time = cfs_time_current_sec();
2536 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2537 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2538 op_data->op_cap = cfs_curproc_cap_pack();
2539 op_data->op_bias = 0;
2540 op_data->op_cli_flags = 0;
2541 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2542 filename_is_volatile(name, namelen, NULL))
2543 op_data->op_bias |= MDS_CREATE_VOLATILE;
2544 op_data->op_mds = 0;
2545 op_data->op_data = data;
2547 /* If the file is being opened after mknod() (normally due to NFS)
2548 * try to use the default stripe data from parent directory for
2549 * allocating OST objects. Try to pass the parent FID to MDS. */
2550 if (opc == LUSTRE_OPC_CREATE && i1 == i2 && S_ISREG(i2->i_mode) &&
2551 !ll_i2info(i2)->lli_has_smd) {
2552 struct ll_inode_info *lli = ll_i2info(i2);
2554 spin_lock(&lli->lli_lock);
2555 if (likely(!lli->lli_has_smd && !fid_is_zero(&lli->lli_pfid)))
2556 op_data->op_fid1 = lli->lli_pfid;
2557 spin_unlock(&lli->lli_lock);
2558 /** We ignore parent's capability temporary. */
2561 /* When called by ll_setattr_raw, file is i1. */
2562 if (LLIF_DATA_MODIFIED & ll_i2info(i1)->lli_flags)
2563 op_data->op_bias |= MDS_DATA_MODIFIED;
2568 void ll_finish_md_op_data(struct md_op_data *op_data)
2570 capa_put(op_data->op_capa1);
2571 capa_put(op_data->op_capa2);
2572 OBD_FREE_PTR(op_data);
2575 #ifdef HAVE_SUPEROPS_USE_DENTRY
2576 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2578 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2581 struct ll_sb_info *sbi;
2583 #ifdef HAVE_SUPEROPS_USE_DENTRY
2584 LASSERT((seq != NULL) && (dentry != NULL));
2585 sbi = ll_s2sbi(dentry->d_sb);
2587 LASSERT((seq != NULL) && (vfs != NULL));
2588 sbi = ll_s2sbi(vfs->mnt_sb);
2591 if (sbi->ll_flags & LL_SBI_NOLCK)
2592 seq_puts(seq, ",nolock");
2594 if (sbi->ll_flags & LL_SBI_FLOCK)
2595 seq_puts(seq, ",flock");
2597 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2598 seq_puts(seq, ",localflock");
2600 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2601 seq_puts(seq, ",user_xattr");
2603 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2604 seq_puts(seq, ",lazystatfs");
2606 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2607 seq_puts(seq, ",user_fid2path");
2613 * Get obd name by cmd, and copy out to user space
2615 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2617 struct ll_sb_info *sbi = ll_i2sbi(inode);
2618 struct obd_device *obd;
2621 if (cmd == OBD_IOC_GETDTNAME)
2622 obd = class_exp2obd(sbi->ll_dt_exp);
2623 else if (cmd == OBD_IOC_GETMDNAME)
2624 obd = class_exp2obd(sbi->ll_md_exp);
2631 if (copy_to_user((void *)arg, obd->obd_name,
2632 strlen(obd->obd_name) + 1))
2639 * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2640 * fsname will be returned in this buffer; otherwise, a static buffer will be
2641 * used to store the fsname and returned to caller.
2643 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2645 static char fsname_static[MTI_NAME_MAXLEN];
2646 struct lustre_sb_info *lsi = s2lsi(sb);
2651 /* this means the caller wants to use static buffer
2652 * and it doesn't care about race. Usually this is
2653 * in error reporting path */
2654 buf = fsname_static;
2655 buflen = sizeof(fsname_static);
2658 len = strlen(lsi->lsi_lmd->lmd_profile);
2659 ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2660 if (ptr && (strcmp(ptr, "-client") == 0))
2663 if (unlikely(len >= buflen))
2665 strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2671 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2678 p.mnt = current->fs->root.mnt;
2680 path = d_path(&p, buf, bufsize);
2685 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2687 char *buf, *path = NULL;
2688 struct dentry *dentry = NULL;
2689 struct ccc_object *obj = cl_inode2ccc(page->mapping->host);
2691 /* this can be called inside spin lock so use GFP_ATOMIC. */
2692 buf = (char *)__get_free_page(GFP_ATOMIC);
2694 dentry = d_find_alias(page->mapping->host);
2696 path = ll_d_path(dentry, buf, PAGE_SIZE);
2700 "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2701 "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2702 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2703 PFID(&obj->cob_header.coh_lu.loh_fid),
2704 (path && !IS_ERR(path)) ? path : "", ioret);
2710 free_page((unsigned long)buf);