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_ioctl.h>
49 #include <lustre_lite.h>
50 #include <lustre_ha.h>
51 #include <lustre_dlm.h>
52 #include <lprocfs_status.h>
53 #include <lustre_disk.h>
54 #include <lustre_param.h>
55 #include <lustre_log.h>
56 #include <cl_object.h>
57 #include <obd_cksum.h>
58 #include "llite_internal.h"
60 struct kmem_cache *ll_file_data_slab;
62 static struct list_head ll_super_blocks = LIST_HEAD_INIT(ll_super_blocks);
63 static DEFINE_SPINLOCK(ll_sb_lock);
66 #define log2(n) ffz(~(n))
69 static struct ll_sb_info *ll_init_sbi(void)
71 struct ll_sb_info *sbi = NULL;
73 unsigned long lru_page_max;
83 spin_lock_init(&sbi->ll_lock);
84 mutex_init(&sbi->ll_lco.lco_lock);
85 spin_lock_init(&sbi->ll_pp_extent_lock);
86 spin_lock_init(&sbi->ll_process_lock);
87 sbi->ll_rw_stats_on = 0;
90 pages = si.totalram - si.totalhigh;
91 lru_page_max = pages / 2;
93 /* initialize ll_cache data */
94 atomic_set(&sbi->ll_cache.ccc_users, 0);
95 sbi->ll_cache.ccc_lru_max = lru_page_max;
96 atomic_set(&sbi->ll_cache.ccc_lru_left, lru_page_max);
97 spin_lock_init(&sbi->ll_cache.ccc_lru_lock);
98 INIT_LIST_HEAD(&sbi->ll_cache.ccc_lru);
100 atomic_set(&sbi->ll_cache.ccc_unstable_nr, 0);
101 init_waitqueue_head(&sbi->ll_cache.ccc_unstable_waitq);
103 sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
104 SBI_DEFAULT_READAHEAD_MAX);
105 sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
106 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
107 SBI_DEFAULT_READAHEAD_WHOLE_MAX;
108 INIT_LIST_HEAD(&sbi->ll_conn_chain);
109 INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
111 ll_generate_random_uuid(uuid);
112 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
113 CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
115 spin_lock(&ll_sb_lock);
116 list_add_tail(&sbi->ll_list, &ll_super_blocks);
117 spin_unlock(&ll_sb_lock);
119 sbi->ll_flags |= LL_SBI_VERBOSE;
120 #ifdef ENABLE_CHECKSUM
121 sbi->ll_flags |= LL_SBI_CHECKSUM;
124 #ifdef HAVE_LRU_RESIZE_SUPPORT
125 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
128 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
129 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
131 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
135 /* metadata statahead is enabled by default */
136 sbi->ll_sa_max = LL_SA_RPC_DEF;
137 atomic_set(&sbi->ll_sa_total, 0);
138 atomic_set(&sbi->ll_sa_wrong, 0);
139 atomic_set(&sbi->ll_agl_total, 0);
140 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
143 sbi->ll_squash.rsi_uid = 0;
144 sbi->ll_squash.rsi_gid = 0;
145 INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
146 init_rwsem(&sbi->ll_squash.rsi_sem);
151 static void ll_free_sbi(struct super_block *sb)
153 struct ll_sb_info *sbi = ll_s2sbi(sb);
157 spin_lock(&ll_sb_lock);
158 list_del(&sbi->ll_list);
159 spin_unlock(&ll_sb_lock);
160 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
161 cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
162 OBD_FREE(sbi, sizeof(*sbi));
167 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
168 struct vfsmount *mnt)
170 struct inode *root = 0;
171 struct ll_sb_info *sbi = ll_s2sbi(sb);
172 struct obd_device *obd;
173 struct obd_capa *oc = NULL;
174 struct obd_statfs *osfs = NULL;
175 struct ptlrpc_request *request = NULL;
176 struct obd_connect_data *data = NULL;
177 struct obd_uuid *uuid;
178 struct md_op_data *op_data;
179 struct lustre_md lmd;
181 int size, err, checksum;
184 obd = class_name2obd(md);
186 CERROR("MD %s: not setup or attached\n", md);
200 /* indicate the features supported by this client */
201 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
202 OBD_CONNECT_ATTRFID |
203 OBD_CONNECT_VERSION | OBD_CONNECT_BRW_SIZE |
204 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
205 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
206 OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 |
207 OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR |
208 OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH|
209 OBD_CONNECT_EINPROGRESS |
210 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
211 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS |
212 OBD_CONNECT_MAX_EASIZE |
213 OBD_CONNECT_FLOCK_DEAD |
214 OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK;
216 if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
217 data->ocd_connect_flags |= OBD_CONNECT_SOM;
219 #ifdef HAVE_LRU_RESIZE_SUPPORT
220 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
221 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
223 #ifdef CONFIG_FS_POSIX_ACL
224 data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK;
227 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
228 /* flag mdc connection as lightweight, only used for test
229 * purpose, use with care */
230 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
232 data->ocd_ibits_known = MDS_INODELOCK_FULL;
233 data->ocd_version = LUSTRE_VERSION_CODE;
235 if (sb->s_flags & MS_RDONLY)
236 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
237 if (sbi->ll_flags & LL_SBI_USER_XATTR)
238 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
240 #ifdef HAVE_MS_FLOCK_LOCK
241 /* force vfs to use lustre handler for flock() calls - bug 10743 */
242 sb->s_flags |= MS_FLOCK_LOCK;
244 #ifdef MS_HAS_NEW_AOPS
245 sb->s_flags |= MS_HAS_NEW_AOPS;
248 if (sbi->ll_flags & LL_SBI_FLOCK)
249 sbi->ll_fop = &ll_file_operations_flock;
250 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
251 sbi->ll_fop = &ll_file_operations;
253 sbi->ll_fop = &ll_file_operations_noflock;
256 data->ocd_connect_flags |= OBD_CONNECT_REAL;
257 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
258 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
260 data->ocd_brw_size = MD_MAX_BRW_SIZE;
262 err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
264 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
265 "recovery, of which this client is not a "
266 "part. Please wait for recovery to complete,"
267 " abort, or time out.\n", md);
270 CERROR("cannot connect to %s: rc = %d\n", md, err);
274 sbi->ll_md_exp->exp_connect_data = *data;
276 err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
277 LUSTRE_SEQ_METADATA);
279 CERROR("%s: Can't init metadata layer FID infrastructure, "
280 "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
284 /* For mount, we only need fs info from MDT0, and also in DNE, it
285 * can make sure the client can be mounted as long as MDT0 is
287 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
288 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
289 OBD_STATFS_FOR_MDT0);
291 GOTO(out_md_fid, err);
293 /* This needs to be after statfs to ensure connect has finished.
294 * Note that "data" does NOT contain the valid connect reply.
295 * If connecting to a 1.8 server there will be no LMV device, so
296 * we can access the MDC export directly and exp_connect_flags will
297 * be non-zero, but if accessing an upgraded 2.1 server it will
298 * have the correct flags filled in.
299 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
300 valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
301 if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
302 valid != CLIENT_CONNECT_MDT_REQD) {
305 OBD_ALLOC_WAIT(buf, PAGE_CACHE_SIZE);
306 obd_connect_flags2str(buf, PAGE_CACHE_SIZE,
307 valid ^ CLIENT_CONNECT_MDT_REQD, ",");
308 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
309 "feature(s) needed for correct operation "
310 "of this client (%s). Please upgrade "
311 "server or downgrade client.\n",
312 sbi->ll_md_exp->exp_obd->obd_name, buf);
313 OBD_FREE(buf, PAGE_CACHE_SIZE);
314 GOTO(out_md_fid, err = -EPROTO);
317 size = sizeof(*data);
318 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
319 KEY_CONN_DATA, &size, data, NULL);
321 CERROR("%s: Get connect data failed: rc = %d\n",
322 sbi->ll_md_exp->exp_obd->obd_name, err);
323 GOTO(out_md_fid, err);
326 LASSERT(osfs->os_bsize);
327 sb->s_blocksize = osfs->os_bsize;
328 sb->s_blocksize_bits = log2(osfs->os_bsize);
329 sb->s_magic = LL_SUPER_MAGIC;
330 sb->s_maxbytes = MAX_LFS_FILESIZE;
331 sbi->ll_namelen = osfs->os_namelen;
332 sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
334 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
335 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
336 LCONSOLE_INFO("Disabling user_xattr feature because "
337 "it is not supported on the server\n");
338 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
341 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
343 sb->s_flags |= MS_POSIXACL;
345 sbi->ll_flags |= LL_SBI_ACL;
347 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
349 sb->s_flags &= ~MS_POSIXACL;
351 sbi->ll_flags &= ~LL_SBI_ACL;
354 if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
355 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
356 sbi->ll_flags |= LL_SBI_RMT_CLIENT;
357 LCONSOLE_INFO("client is set as remote by default.\n");
360 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
361 sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
362 LCONSOLE_INFO("client claims to be remote, but server "
363 "rejected, forced to be local.\n");
367 if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
368 LCONSOLE_INFO("client enabled MDS capability!\n");
369 sbi->ll_flags |= LL_SBI_MDS_CAPA;
372 if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
373 LCONSOLE_INFO("client enabled OSS capability!\n");
374 sbi->ll_flags |= LL_SBI_OSS_CAPA;
377 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
378 sbi->ll_flags |= LL_SBI_64BIT_HASH;
380 if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
381 sbi->ll_md_brw_pages = data->ocd_brw_size >> PAGE_CACHE_SHIFT;
383 sbi->ll_md_brw_pages = 1;
385 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
386 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
388 if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
389 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
390 LCONSOLE_INFO("%s: disabling xattr cache due to "
391 "unknown maximum xattr size.\n", dt);
393 sbi->ll_flags |= LL_SBI_XATTR_CACHE;
394 sbi->ll_xattr_cache_enabled = 1;
398 obd = class_name2obd(dt);
400 CERROR("DT %s: not setup or attached\n", dt);
401 GOTO(out_md_fid, err = -ENODEV);
404 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
405 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
406 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
407 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
408 OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
409 OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
410 OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
411 OBD_CONNECT_MAXBYTES |
412 OBD_CONNECT_EINPROGRESS |
413 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
414 OBD_CONNECT_LAYOUTLOCK |
415 OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK;
417 if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
418 data->ocd_connect_flags |= OBD_CONNECT_SOM;
420 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
421 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
422 * disabled by default, because it can still be enabled on the
423 * fly via /proc. As a consequence, we still need to come to an
424 * agreement on the supported algorithms at connect time */
425 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
427 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
428 data->ocd_cksum_types = OBD_CKSUM_ADLER;
430 data->ocd_cksum_types = cksum_types_supported_client();
433 #ifdef HAVE_LRU_RESIZE_SUPPORT
434 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
436 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
437 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
439 CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
440 "ocd_grant: %d\n", data->ocd_connect_flags,
441 data->ocd_version, data->ocd_grant);
443 obd->obd_upcall.onu_owner = &sbi->ll_lco;
444 obd->obd_upcall.onu_upcall = cl_ocd_update;
446 data->ocd_brw_size = DT_MAX_BRW_SIZE;
448 err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
451 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
452 "recovery, of which this client is not a "
453 "part. Please wait for recovery to "
454 "complete, abort, or time out.\n", dt);
457 CERROR("%s: Cannot connect to %s: rc = %d\n",
458 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
462 sbi->ll_dt_exp->exp_connect_data = *data;
464 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
465 LUSTRE_SEQ_METADATA);
467 CERROR("%s: Can't init data layer FID infrastructure, "
468 "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
472 mutex_lock(&sbi->ll_lco.lco_lock);
473 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
474 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
475 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
476 mutex_unlock(&sbi->ll_lco.lco_lock);
478 fid_zero(&sbi->ll_root_fid);
479 err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
481 CERROR("cannot mds_connect: rc = %d\n", err);
482 GOTO(out_lock_cn_cb, err);
484 if (!fid_is_sane(&sbi->ll_root_fid)) {
485 CERROR("%s: Invalid root fid "DFID" during mount\n",
486 sbi->ll_md_exp->exp_obd->obd_name,
487 PFID(&sbi->ll_root_fid));
488 GOTO(out_lock_cn_cb, err = -EINVAL);
490 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
492 sb->s_op = &lustre_super_operations;
493 #if THREAD_SIZE >= 8192 /*b=17630*/
494 sb->s_export_op = &lustre_export_operations;
498 * XXX: move this to after cbd setup? */
499 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA |
501 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
502 valid |= OBD_MD_FLRMTPERM;
503 else if (sbi->ll_flags & LL_SBI_ACL)
504 valid |= OBD_MD_FLACL;
506 OBD_ALLOC_PTR(op_data);
508 GOTO(out_lock_cn_cb, err = -ENOMEM);
510 op_data->op_fid1 = sbi->ll_root_fid;
511 op_data->op_mode = 0;
512 op_data->op_capa1 = oc;
513 op_data->op_valid = valid;
515 err = md_getattr(sbi->ll_md_exp, op_data, &request);
518 OBD_FREE_PTR(op_data);
520 CERROR("%s: md_getattr failed for root: rc = %d\n",
521 sbi->ll_md_exp->exp_obd->obd_name, err);
522 GOTO(out_lock_cn_cb, err);
525 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
526 sbi->ll_md_exp, &lmd);
528 CERROR("failed to understand root inode md: rc = %d\n", err);
529 ptlrpc_req_finished(request);
530 GOTO(out_lock_cn_cb, err);
533 LASSERT(fid_is_sane(&sbi->ll_root_fid));
534 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
535 sbi->ll_flags & LL_SBI_32BIT_API),
537 md_free_lustre_md(sbi->ll_md_exp, &lmd);
538 ptlrpc_req_finished(request);
542 obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
543 #ifdef CONFIG_FS_POSIX_ACL
545 posix_acl_release(lmd.posix_acl);
546 lmd.posix_acl = NULL;
549 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
551 CERROR("lustre_lite: bad iget4 for root\n");
555 err = ll_close_thread_start(&sbi->ll_lcq);
557 CERROR("cannot start close thread: rc %d\n", err);
561 #ifdef CONFIG_FS_POSIX_ACL
562 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
563 rct_init(&sbi->ll_rct);
564 et_init(&sbi->ll_et);
568 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
569 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
570 KEY_CHECKSUM, sizeof(checksum), &checksum,
574 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
575 KEY_CACHE_SET, sizeof(sbi->ll_cache),
576 &sbi->ll_cache, NULL);
578 sb->s_root = d_make_root(root);
579 if (sb->s_root == NULL) {
580 CERROR("%s: can't make root dentry\n",
581 ll_get_fsname(sb, NULL, 0));
582 GOTO(out_root, err = -ENOMEM);
584 #ifdef HAVE_DCACHE_LOCK
585 sb->s_root->d_op = &ll_d_ops;
588 sbi->ll_sdev_orig = sb->s_dev;
590 /* We set sb->s_dev equal on all lustre clients in order to support
591 * NFS export clustering. NFSD requires that the FSID be the same
593 /* s_dev is also used in lt_compare() to compare two fs, but that is
594 * only a node-local comparison. */
595 uuid = obd_get_uuid(sbi->ll_md_exp);
597 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
603 if (proc_lustre_fs_root != NULL) {
604 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
607 CERROR("%s: could not register mount in lprocfs: "
608 "rc = %d\n", ll_get_fsname(sb, NULL, 0), err);
618 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
620 obd_disconnect(sbi->ll_dt_exp);
621 sbi->ll_dt_exp = NULL;
622 /* Make sure all OScs are gone, since cl_cache is accessing sbi. */
623 obd_zombie_barrier();
625 obd_fid_fini(sbi->ll_md_exp->exp_obd);
627 obd_disconnect(sbi->ll_md_exp);
628 sbi->ll_md_exp = NULL;
637 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
641 *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
643 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
644 KEY_MAX_EASIZE, &size, lmmsize, NULL);
646 CERROR("Get max mdsize error rc %d\n", rc);
651 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
656 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
657 KEY_DEFAULT_EASIZE, &size, lmmsize, NULL);
659 CERROR("Get default mdsize error rc %d\n", rc);
664 int ll_get_max_cookiesize(struct ll_sb_info *sbi, int *lmmsize)
669 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_COOKIESIZE),
670 KEY_MAX_COOKIESIZE, &size, lmmsize, NULL);
672 CERROR("Get max cookiesize error rc %d\n", rc);
677 int ll_get_default_cookiesize(struct ll_sb_info *sbi, int *lmmsize)
682 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_COOKIESIZE),
683 KEY_DEFAULT_COOKIESIZE, &size, lmmsize, NULL);
685 CERROR("Get default cookiesize error rc %d\n", rc);
690 static void ll_dump_inode(struct inode *inode)
692 struct ll_d_hlist_node *tmp;
693 int dentry_count = 0;
695 LASSERT(inode != NULL);
697 ll_d_hlist_for_each(tmp, &inode->i_dentry)
700 CERROR("%s: inode %p dump: dev=%s fid="DFID
701 " mode=%o count=%u, %d dentries\n",
702 ll_get_fsname(inode->i_sb, NULL, 0), inode,
703 ll_i2mdexp(inode)->exp_obd->obd_name, PFID(ll_inode2fid(inode)),
704 inode->i_mode, atomic_read(&inode->i_count), dentry_count);
707 void lustre_dump_dentry(struct dentry *dentry, int recur)
709 struct list_head *tmp;
712 LASSERT(dentry != NULL);
714 list_for_each(tmp, &dentry->d_subdirs)
717 CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
718 " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
719 dentry->d_name.len, dentry->d_name.name,
720 dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
721 dentry->d_parent, dentry->d_inode, d_count(dentry),
722 dentry->d_flags, dentry->d_fsdata, subdirs);
723 if (dentry->d_inode != NULL)
724 ll_dump_inode(dentry->d_inode);
729 list_for_each(tmp, &dentry->d_subdirs) {
730 struct dentry *d = list_entry(tmp, struct dentry, d_u.d_child);
731 lustre_dump_dentry(d, recur - 1);
735 static void client_common_put_super(struct super_block *sb)
737 struct ll_sb_info *sbi = ll_s2sbi(sb);
740 #ifdef CONFIG_FS_POSIX_ACL
741 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
742 et_fini(&sbi->ll_et);
743 rct_fini(&sbi->ll_rct);
747 ll_close_thread_shutdown(sbi->ll_lcq);
751 list_del(&sbi->ll_conn_chain);
753 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
754 obd_disconnect(sbi->ll_dt_exp);
755 sbi->ll_dt_exp = NULL;
756 /* wait till all OSCs are gone, since cl_cache is accessing sbi.
758 obd_zombie_barrier();
760 lprocfs_unregister_mountpoint(sbi);
762 obd_fid_fini(sbi->ll_md_exp->exp_obd);
763 obd_disconnect(sbi->ll_md_exp);
764 sbi->ll_md_exp = NULL;
769 void ll_kill_super(struct super_block *sb)
771 struct ll_sb_info *sbi;
776 if (!(sb->s_flags & MS_ACTIVE))
780 /* we need restore s_dev from changed for clustred NFS before put_super
781 * because new kernels have cached s_dev and change sb->s_dev in
782 * put_super not affected real removing devices */
784 sb->s_dev = sbi->ll_sdev_orig;
785 sbi->ll_umounting = 1;
790 static inline int ll_set_opt(const char *opt, char *data, int fl)
792 if (strncmp(opt, data, strlen(opt)) != 0)
798 /* non-client-specific mount options are parsed in lmd_parse */
799 static int ll_options(char *options, int *flags)
802 char *s1 = options, *s2;
808 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
811 CDEBUG(D_SUPER, "next opt=%s\n", s1);
812 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
817 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
822 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
827 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
832 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
837 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
842 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 6, 51, 0)
843 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
845 /* Ignore deprecated mount option. The client will
846 * always try to mount with ACL support, whether this
847 * is used depends on whether server supports it. */
848 LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
849 "mount option 'acl'.\n");
852 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
854 LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
855 "mount option 'noacl'.\n");
859 #warning "{no}acl options have been deprecated since 1.8, please remove them"
861 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
866 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
871 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
877 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
882 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
887 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
892 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
897 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
902 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
907 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
912 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
917 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
922 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
927 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
933 s2 = strchr(s1, ',');
941 void ll_lli_init(struct ll_inode_info *lli)
943 lli->lli_inode_magic = LLI_INODE_MAGIC;
945 lli->lli_ioepoch = 0;
946 lli->lli_maxbytes = MAX_LFS_FILESIZE;
947 spin_lock_init(&lli->lli_lock);
948 lli->lli_posix_acl = NULL;
949 lli->lli_remote_perms = NULL;
950 mutex_init(&lli->lli_rmtperm_mutex);
951 /* Do not set lli_fid, it has been initialized already. */
952 fid_zero(&lli->lli_pfid);
953 INIT_LIST_HEAD(&lli->lli_close_list);
954 INIT_LIST_HEAD(&lli->lli_oss_capas);
955 atomic_set(&lli->lli_open_count, 0);
956 lli->lli_mds_capa = NULL;
957 lli->lli_rmtperm_time = 0;
958 lli->lli_pending_och = NULL;
959 lli->lli_mds_read_och = NULL;
960 lli->lli_mds_write_och = NULL;
961 lli->lli_mds_exec_och = NULL;
962 lli->lli_open_fd_read_count = 0;
963 lli->lli_open_fd_write_count = 0;
964 lli->lli_open_fd_exec_count = 0;
965 mutex_init(&lli->lli_och_mutex);
966 spin_lock_init(&lli->lli_agl_lock);
967 lli->lli_has_smd = false;
968 spin_lock_init(&lli->lli_layout_lock);
969 ll_layout_version_set(lli, LL_LAYOUT_GEN_NONE);
970 lli->lli_clob = NULL;
972 init_rwsem(&lli->lli_xattrs_list_rwsem);
973 mutex_init(&lli->lli_xattrs_enq_lock);
975 LASSERT(lli->lli_vfs_inode.i_mode != 0);
976 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
977 mutex_init(&lli->lli_readdir_mutex);
978 lli->lli_opendir_key = NULL;
980 spin_lock_init(&lli->lli_sa_lock);
981 lli->lli_opendir_pid = 0;
983 mutex_init(&lli->lli_size_mutex);
984 lli->lli_symlink_name = NULL;
985 init_rwsem(&lli->lli_trunc_sem);
986 mutex_init(&lli->lli_write_mutex);
987 init_rwsem(&lli->lli_glimpse_sem);
988 lli->lli_glimpse_time = 0;
989 INIT_LIST_HEAD(&lli->lli_agl_list);
990 lli->lli_agl_index = 0;
991 lli->lli_async_rc = 0;
993 mutex_init(&lli->lli_layout_mutex);
996 static inline int ll_bdi_register(struct backing_dev_info *bdi)
998 static atomic_t ll_bdi_num = ATOMIC_INIT(0);
1000 bdi->name = "lustre";
1001 return bdi_register(bdi, NULL, "lustre-%d",
1002 atomic_inc_return(&ll_bdi_num));
1005 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
1007 struct lustre_profile *lprof = NULL;
1008 struct lustre_sb_info *lsi = s2lsi(sb);
1009 struct ll_sb_info *sbi;
1010 char *dt = NULL, *md = NULL;
1011 char *profilenm = get_profile_name(sb);
1012 struct config_llog_instance *cfg;
1013 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1014 const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
1018 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1024 try_module_get(THIS_MODULE);
1026 /* client additional sb info */
1027 lsi->lsi_llsbi = sbi = ll_init_sbi();
1029 module_put(THIS_MODULE);
1034 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1036 GOTO(out_free, err);
1038 err = bdi_init(&lsi->lsi_bdi);
1040 GOTO(out_free, err);
1041 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1042 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1043 err = ll_bdi_register(&lsi->lsi_bdi);
1045 GOTO(out_free, err);
1047 sb->s_bdi = &lsi->lsi_bdi;
1048 #ifndef HAVE_DCACHE_LOCK
1049 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1050 sb->s_d_op = &ll_d_ops;
1053 /* Generate a string unique to this super, in case some joker tries
1054 to mount the same fs at two mount points.
1055 Use the address of the super itself.*/
1056 cfg->cfg_instance = sb;
1057 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1058 cfg->cfg_callback = class_config_llog_handler;
1059 /* set up client obds */
1060 err = lustre_process_log(sb, profilenm, cfg);
1062 GOTO(out_free, err);
1064 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1065 lprof = class_get_profile(profilenm);
1066 if (lprof == NULL) {
1067 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1068 " read from the MGS. Does that filesystem "
1069 "exist?\n", profilenm);
1070 GOTO(out_free, err = -EINVAL);
1072 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1073 lprof->lp_md, lprof->lp_dt);
1075 OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1077 GOTO(out_free, err = -ENOMEM);
1078 sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1080 OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1082 GOTO(out_free, err = -ENOMEM);
1083 sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1085 /* connections, registrations, sb setup */
1086 err = client_common_fill_super(sb, md, dt, mnt);
1090 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1092 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1095 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1096 LCONSOLE_WARN("Mounted %s\n", profilenm);
1100 } /* ll_fill_super */
1102 void ll_put_super(struct super_block *sb)
1104 struct config_llog_instance cfg, params_cfg;
1105 struct obd_device *obd;
1106 struct lustre_sb_info *lsi = s2lsi(sb);
1107 struct ll_sb_info *sbi = ll_s2sbi(sb);
1108 char *profilenm = get_profile_name(sb);
1109 int ccc_count, next, force = 1, rc = 0;
1112 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1114 ll_print_capa_stat(sbi);
1116 cfg.cfg_instance = sb;
1117 lustre_end_log(sb, profilenm, &cfg);
1119 params_cfg.cfg_instance = sb;
1120 lustre_end_log(sb, PARAMS_FILENAME, ¶ms_cfg);
1122 if (sbi->ll_md_exp) {
1123 obd = class_exp2obd(sbi->ll_md_exp);
1125 force = obd->obd_force;
1128 /* Wait for unstable pages to be committed to stable storage */
1130 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1131 rc = l_wait_event(sbi->ll_cache.ccc_unstable_waitq,
1132 atomic_read(&sbi->ll_cache.ccc_unstable_nr) == 0,
1136 ccc_count = atomic_read(&sbi->ll_cache.ccc_unstable_nr);
1137 if (force == 0 && rc != -EINTR)
1138 LASSERTF(ccc_count == 0, "count: %i\n", ccc_count);
1141 /* We need to set force before the lov_disconnect in
1142 lustre_common_put_super, since l_d cleans up osc's as well. */
1145 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1147 obd->obd_force = force;
1152 /* Only if client_common_fill_super succeeded */
1153 client_common_put_super(sb);
1157 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1158 class_manual_cleanup(obd);
1161 if (sbi->ll_flags & LL_SBI_VERBOSE)
1162 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1165 class_del_profile(profilenm);
1167 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1168 bdi_destroy(&lsi->lsi_bdi);
1169 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1173 lsi->lsi_llsbi = NULL;
1175 lustre_common_put_super(sb);
1177 cl_env_cache_purge(~0);
1179 module_put(THIS_MODULE);
1182 } /* client_put_super */
1184 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1186 struct inode *inode = NULL;
1188 /* NOTE: we depend on atomic igrab() -bzzz */
1189 lock_res_and_lock(lock);
1190 if (lock->l_resource->lr_lvb_inode) {
1191 struct ll_inode_info * lli;
1192 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1193 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1194 inode = igrab(lock->l_resource->lr_lvb_inode);
1196 inode = lock->l_resource->lr_lvb_inode;
1197 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1198 D_WARNING, lock, "lr_lvb_inode %p is "
1199 "bogus: magic %08x",
1200 lock->l_resource->lr_lvb_inode,
1201 lli->lli_inode_magic);
1205 unlock_res_and_lock(lock);
1209 static void ll_dir_clear_lsm_md(struct inode *inode)
1211 struct ll_inode_info *lli = ll_i2info(inode);
1213 LASSERT(S_ISDIR(inode->i_mode));
1215 if (lli->lli_lsm_md != NULL) {
1216 lmv_free_memmd(lli->lli_lsm_md);
1217 lli->lli_lsm_md = NULL;
1221 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1222 const struct lu_fid *fid,
1223 struct lustre_md *md)
1225 struct ll_sb_info *sbi = ll_s2sbi(sb);
1226 struct mdt_body *body = md->body;
1227 struct inode *inode;
1231 ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1232 inode = iget_locked(sb, ino);
1233 if (inode == NULL) {
1234 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1235 ll_get_fsname(sb, NULL, 0), PFID(fid));
1236 RETURN(ERR_PTR(-ENOENT));
1239 if (inode->i_state & I_NEW) {
1240 struct ll_inode_info *lli = ll_i2info(inode);
1241 struct lmv_stripe_md *lsm = md->lmv;
1243 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1244 (body->mbo_mode & S_IFMT);
1245 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1248 LTIME_S(inode->i_mtime) = 0;
1249 LTIME_S(inode->i_atime) = 0;
1250 LTIME_S(inode->i_ctime) = 0;
1253 /* initializing backing dev info. */
1254 inode->i_mapping->backing_dev_info =
1255 &s2lsi(inode->i_sb)->lsi_bdi;
1256 inode->i_op = &ll_dir_inode_operations;
1257 inode->i_fop = &ll_dir_operations;
1258 lli->lli_fid = *fid;
1261 LASSERT(lsm != NULL);
1262 /* master object FID */
1263 lli->lli_pfid = body->mbo_fid1;
1264 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1265 lli, PFID(fid), PFID(&lli->lli_pfid));
1266 unlock_new_inode(inode);
1272 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1275 struct lmv_stripe_md *lsm = md->lmv;
1278 LASSERT(lsm != NULL);
1279 /* XXX sigh, this lsm_root initialization should be in
1280 * LMV layer, but it needs ll_iget right now, so we
1281 * put this here right now. */
1282 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1283 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1284 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1285 /* Unfortunately ll_iget will call ll_update_inode,
1286 * where the initialization of slave inode is slightly
1287 * different, so it reset lsm_md to NULL to avoid
1288 * initializing lsm for slave inode. */
1289 /* For migrating inode, master stripe and master object will
1290 * be same, so we only need assign this inode */
1291 if (lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION && i == 0)
1292 lsm->lsm_md_oinfo[i].lmo_root = inode;
1294 lsm->lsm_md_oinfo[i].lmo_root =
1295 ll_iget_anon_dir(inode->i_sb, fid, md);
1297 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1298 int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1300 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1305 /* Here is where the lsm is being initialized(fill lmo_info) after
1306 * client retrieve MD stripe information from MDT. */
1307 return md_update_lsm_md(ll_i2mdexp(inode), lsm, md->body,
1308 ll_md_blocking_ast);
1311 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1312 const struct lmv_stripe_md *lsm_md2)
1314 return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1315 lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1316 lsm_md1->lsm_md_master_mdt_index ==
1317 lsm_md2->lsm_md_master_mdt_index &&
1318 lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1319 lsm_md1->lsm_md_layout_version ==
1320 lsm_md2->lsm_md_layout_version &&
1321 strcmp(lsm_md1->lsm_md_pool_name,
1322 lsm_md2->lsm_md_pool_name) == 0;
1325 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1327 struct ll_inode_info *lli = ll_i2info(inode);
1328 struct lmv_stripe_md *lsm = md->lmv;
1332 LASSERT(S_ISDIR(inode->i_mode));
1333 CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1334 PFID(ll_inode2fid(inode)));
1336 /* no striped information from request. */
1338 if (lli->lli_lsm_md == NULL) {
1340 } else if (lli->lli_lsm_md->lsm_md_hash_type &
1341 LMV_HASH_FLAG_MIGRATION) {
1342 /* migration is done, the temporay MIGRATE layout has
1344 CDEBUG(D_INODE, DFID" finish migration.\n",
1345 PFID(ll_inode2fid(inode)));
1346 lmv_free_memmd(lli->lli_lsm_md);
1347 lli->lli_lsm_md = NULL;
1350 /* The lustre_md from req does not include stripeEA,
1351 * see ll_md_setattr */
1356 /* set the directory layout */
1357 if (lli->lli_lsm_md == NULL) {
1359 rc = ll_init_lsm_md(inode, md);
1363 lli->lli_lsm_md = lsm;
1364 /* set lsm_md to NULL, so the following free lustre_md
1365 * will not free this lsm */
1367 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1368 lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1372 /* Compare the old and new stripe information */
1373 if (!lsm_md_eq(lli->lli_lsm_md, lsm)) {
1374 struct lmv_stripe_md *old_lsm = lli->lli_lsm_md;
1377 CERROR("%s: lmv layout mismatch "DFID"(%p)/"DFID"(%p)"
1378 "magic:0x%x/0x%x stripe count: %d/%d master_mdt: %d/%d"
1379 "hash_type:0x%x/0x%x layout: 0x%x/0x%x pool:%s/%s\n",
1380 ll_get_fsname(inode->i_sb, NULL, 0),
1381 PFID(&lsm->lsm_md_master_fid), lsm,
1382 PFID(&old_lsm->lsm_md_master_fid), old_lsm,
1383 lsm->lsm_md_magic, old_lsm->lsm_md_magic,
1384 lsm->lsm_md_stripe_count,
1385 old_lsm->lsm_md_stripe_count,
1386 lsm->lsm_md_master_mdt_index,
1387 old_lsm->lsm_md_master_mdt_index,
1388 lsm->lsm_md_hash_type, old_lsm->lsm_md_hash_type,
1389 lsm->lsm_md_layout_version,
1390 old_lsm->lsm_md_layout_version,
1391 lsm->lsm_md_pool_name,
1392 old_lsm->lsm_md_pool_name);
1394 for (idx = 0; idx < old_lsm->lsm_md_stripe_count; idx++) {
1395 CERROR("%s: sub FIDs in old lsm idx %d, old: "DFID"\n",
1396 ll_get_fsname(inode->i_sb, NULL, 0), idx,
1397 PFID(&old_lsm->lsm_md_oinfo[idx].lmo_fid));
1400 for (idx = 0; idx < lsm->lsm_md_stripe_count; idx++) {
1401 CERROR("%s: sub FIDs in new lsm idx %d, new: "DFID"\n",
1402 ll_get_fsname(inode->i_sb, NULL, 0), idx,
1403 PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1409 rc = md_update_lsm_md(ll_i2mdexp(inode), ll_i2info(inode)->lli_lsm_md,
1410 md->body, ll_md_blocking_ast);
1415 void ll_clear_inode(struct inode *inode)
1417 struct ll_inode_info *lli = ll_i2info(inode);
1418 struct ll_sb_info *sbi = ll_i2sbi(inode);
1421 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1422 PFID(ll_inode2fid(inode)), inode);
1424 if (S_ISDIR(inode->i_mode)) {
1425 /* these should have been cleared in ll_file_release */
1426 LASSERT(lli->lli_opendir_key == NULL);
1427 LASSERT(lli->lli_sai == NULL);
1428 LASSERT(lli->lli_opendir_pid == 0);
1431 spin_lock(&lli->lli_lock);
1432 ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1433 spin_unlock(&lli->lli_lock);
1434 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1436 LASSERT(!lli->lli_open_fd_write_count);
1437 LASSERT(!lli->lli_open_fd_read_count);
1438 LASSERT(!lli->lli_open_fd_exec_count);
1440 if (lli->lli_mds_write_och)
1441 ll_md_real_close(inode, FMODE_WRITE);
1442 if (lli->lli_mds_exec_och)
1443 ll_md_real_close(inode, FMODE_EXEC);
1444 if (lli->lli_mds_read_och)
1445 ll_md_real_close(inode, FMODE_READ);
1447 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1448 OBD_FREE(lli->lli_symlink_name,
1449 strlen(lli->lli_symlink_name) + 1);
1450 lli->lli_symlink_name = NULL;
1453 ll_xattr_cache_destroy(inode);
1455 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1456 LASSERT(lli->lli_posix_acl == NULL);
1457 if (lli->lli_remote_perms) {
1458 free_rmtperm_hash(lli->lli_remote_perms);
1459 lli->lli_remote_perms = NULL;
1462 #ifdef CONFIG_FS_POSIX_ACL
1463 else if (lli->lli_posix_acl) {
1464 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1465 LASSERT(lli->lli_remote_perms == NULL);
1466 posix_acl_release(lli->lli_posix_acl);
1467 lli->lli_posix_acl = NULL;
1470 lli->lli_inode_magic = LLI_INODE_DEAD;
1472 ll_clear_inode_capas(inode);
1473 if (S_ISDIR(inode->i_mode))
1474 ll_dir_clear_lsm_md(inode);
1475 else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1476 LASSERT(list_empty(&lli->lli_agl_list));
1479 * XXX This has to be done before lsm is freed below, because
1480 * cl_object still uses inode lsm.
1482 cl_inode_fini(inode);
1483 lli->lli_has_smd = false;
1488 int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data,
1489 struct md_open_data **mod)
1491 struct lustre_md md;
1492 struct inode *inode = dentry->d_inode;
1493 struct ll_sb_info *sbi = ll_i2sbi(inode);
1494 struct ptlrpc_request *request = NULL;
1498 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1499 LUSTRE_OPC_ANY, NULL);
1500 if (IS_ERR(op_data))
1501 RETURN(PTR_ERR(op_data));
1503 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1506 ptlrpc_req_finished(request);
1507 if (rc == -ENOENT) {
1509 /* Unlinked special device node? Or just a race?
1510 * Pretend we done everything. */
1511 if (!S_ISREG(inode->i_mode) &&
1512 !S_ISDIR(inode->i_mode)) {
1513 ia_valid = op_data->op_attr.ia_valid;
1514 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1515 rc = simple_setattr(dentry, &op_data->op_attr);
1516 op_data->op_attr.ia_valid = ia_valid;
1518 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1519 CERROR("md_setattr fails: rc = %d\n", rc);
1524 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1525 sbi->ll_md_exp, &md);
1527 ptlrpc_req_finished(request);
1531 ia_valid = op_data->op_attr.ia_valid;
1532 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1533 * cache is not cleared yet. */
1534 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1535 rc = simple_setattr(dentry, &op_data->op_attr);
1536 op_data->op_attr.ia_valid = ia_valid;
1538 /* Extract epoch data if obtained. */
1539 op_data->op_handle = md.body->mbo_handle;
1540 op_data->op_ioepoch = md.body->mbo_ioepoch;
1542 rc = ll_update_inode(inode, &md);
1543 ptlrpc_req_finished(request);
1548 /* Close IO epoch and send Size-on-MDS attribute update. */
1549 static int ll_setattr_done_writing(struct inode *inode,
1550 struct md_op_data *op_data,
1551 struct md_open_data *mod)
1553 struct ll_inode_info *lli = ll_i2info(inode);
1557 LASSERT(op_data != NULL);
1558 if (!S_ISREG(inode->i_mode))
1561 CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1562 op_data->op_ioepoch, PFID(&lli->lli_fid));
1564 op_data->op_flags = MF_EPOCH_CLOSE;
1565 ll_done_writing_attr(inode, op_data);
1566 ll_pack_inode2opdata(inode, op_data, NULL);
1568 rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1569 if (rc == -EAGAIN) {
1570 /* MDS has instructed us to obtain Size-on-MDS attribute
1571 * from OSTs and send setattr to back to MDS. */
1572 rc = ll_som_update(inode, op_data);
1574 CERROR("%s: inode "DFID" mdc truncate failed: rc = %d\n",
1575 ll_i2sbi(inode)->ll_md_exp->exp_obd->obd_name,
1576 PFID(ll_inode2fid(inode)), rc);
1581 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1583 struct obd_capa *capa;
1586 if (attr->ia_valid & ATTR_SIZE)
1587 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1589 capa = ll_mdscapa_get(inode);
1591 rc = cl_setattr_ost(inode, attr, capa);
1593 if (attr->ia_valid & ATTR_SIZE)
1594 ll_truncate_free_capa(capa);
1601 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1602 * object(s) determine the file size and mtime. Otherwise, the MDS will
1603 * keep these values until such a time that objects are allocated for it.
1604 * We do the MDS operations first, as it is checking permissions for us.
1605 * We don't to the MDS RPC if there is nothing that we want to store there,
1606 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1607 * going to do an RPC anyways.
1609 * If we are doing a truncate, we will send the mtime and ctime updates
1610 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1611 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1614 * In case of HSMimport, we only set attr on MDS.
1616 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1618 struct inode *inode = dentry->d_inode;
1619 struct ll_inode_info *lli = ll_i2info(inode);
1620 struct md_op_data *op_data = NULL;
1621 struct md_open_data *mod = NULL;
1622 bool file_is_released = false;
1623 int rc = 0, rc1 = 0;
1626 CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1627 "valid %x, hsm_import %d\n",
1628 ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1629 inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1632 if (attr->ia_valid & ATTR_SIZE) {
1633 /* Check new size against VFS/VM file size limit and rlimit */
1634 rc = inode_newsize_ok(inode, attr->ia_size);
1638 /* The maximum Lustre file size is variable, based on the
1639 * OST maximum object size and number of stripes. This
1640 * needs another check in addition to the VFS check above. */
1641 if (attr->ia_size > ll_file_maxbytes(inode)) {
1642 CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1643 PFID(&lli->lli_fid), attr->ia_size,
1644 ll_file_maxbytes(inode));
1648 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1651 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1652 if (attr->ia_valid & TIMES_SET_FLAGS) {
1653 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1654 !cfs_capable(CFS_CAP_FOWNER))
1658 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1659 if (attr->ia_valid & ATTR_CTIME) {
1660 attr->ia_ctime = CFS_CURRENT_TIME;
1661 attr->ia_valid |= ATTR_CTIME_SET;
1663 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1664 (attr->ia_valid & ATTR_ATIME)) {
1665 attr->ia_atime = CFS_CURRENT_TIME;
1666 attr->ia_valid |= ATTR_ATIME_SET;
1668 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1669 (attr->ia_valid & ATTR_MTIME)) {
1670 attr->ia_mtime = CFS_CURRENT_TIME;
1671 attr->ia_valid |= ATTR_MTIME_SET;
1674 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1675 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1676 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1677 cfs_time_current_sec());
1679 /* We always do an MDS RPC, even if we're only changing the size;
1680 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1682 OBD_ALLOC_PTR(op_data);
1683 if (op_data == NULL)
1686 if (!S_ISDIR(inode->i_mode)) {
1687 if (attr->ia_valid & ATTR_SIZE)
1688 inode_dio_write_done(inode);
1689 mutex_unlock(&inode->i_mutex);
1692 /* truncate on a released file must failed with -ENODATA,
1693 * so size must not be set on MDS for released file
1694 * but other attributes must be set
1696 if (S_ISREG(inode->i_mode)) {
1697 struct lov_stripe_md *lsm;
1700 ll_layout_refresh(inode, &gen);
1701 lsm = ccc_inode_lsm_get(inode);
1702 if (lsm && lsm->lsm_pattern & LOV_PATTERN_F_RELEASED)
1703 file_is_released = true;
1704 ccc_inode_lsm_put(inode, lsm);
1706 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1707 if (file_is_released) {
1708 rc = ll_layout_restore(inode, 0, attr->ia_size);
1712 file_is_released = false;
1713 ll_layout_refresh(inode, &gen);
1716 /* If we are changing file size, file content is
1717 * modified, flag it. */
1718 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1719 spin_lock(&lli->lli_lock);
1720 lli->lli_flags |= LLIF_DATA_MODIFIED;
1721 spin_unlock(&lli->lli_lock);
1722 op_data->op_bias |= MDS_DATA_MODIFIED;
1726 memcpy(&op_data->op_attr, attr, sizeof(*attr));
1728 /* Open epoch for truncate. */
1729 if (exp_connect_som(ll_i2mdexp(inode)) && !hsm_import &&
1730 (attr->ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1731 op_data->op_flags = MF_EPOCH_OPEN;
1733 rc = ll_md_setattr(dentry, op_data, &mod);
1737 /* RPC to MDT is sent, cancel data modification flag */
1738 if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) {
1739 spin_lock(&lli->lli_lock);
1740 lli->lli_flags &= ~LLIF_DATA_MODIFIED;
1741 spin_unlock(&lli->lli_lock);
1744 ll_ioepoch_open(lli, op_data->op_ioepoch);
1745 if (!S_ISREG(inode->i_mode) || file_is_released)
1748 if (attr->ia_valid & (ATTR_SIZE |
1749 ATTR_ATIME | ATTR_ATIME_SET |
1750 ATTR_MTIME | ATTR_MTIME_SET)) {
1751 /* For truncate and utimes sending attributes to OSTs, setting
1752 * mtime/atime to the past will be performed under PW [0:EOF]
1753 * extent lock (new_size:EOF for truncate). It may seem
1754 * excessive to send mtime/atime updates to OSTs when not
1755 * setting times to past, but it is necessary due to possible
1756 * time de-synchronization between MDT inode and OST objects */
1757 if (attr->ia_valid & ATTR_SIZE)
1758 down_write(&lli->lli_trunc_sem);
1759 rc = ll_setattr_ost(inode, attr);
1760 if (attr->ia_valid & ATTR_SIZE)
1761 up_write(&lli->lli_trunc_sem);
1766 if (op_data->op_ioepoch) {
1767 rc1 = ll_setattr_done_writing(inode, op_data, mod);
1771 ll_finish_md_op_data(op_data);
1773 if (!S_ISDIR(inode->i_mode)) {
1774 mutex_lock(&inode->i_mutex);
1775 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1776 inode_dio_wait(inode);
1779 ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1780 LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1785 int ll_setattr(struct dentry *de, struct iattr *attr)
1787 int mode = de->d_inode->i_mode;
1789 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1790 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1791 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1793 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1794 (ATTR_SIZE|ATTR_MODE)) &&
1795 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1796 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1797 !(attr->ia_mode & S_ISGID))))
1798 attr->ia_valid |= ATTR_FORCE;
1800 if ((attr->ia_valid & ATTR_MODE) &&
1802 !(attr->ia_mode & S_ISUID) &&
1803 !(attr->ia_valid & ATTR_KILL_SUID))
1804 attr->ia_valid |= ATTR_KILL_SUID;
1806 if ((attr->ia_valid & ATTR_MODE) &&
1807 ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1808 !(attr->ia_mode & S_ISGID) &&
1809 !(attr->ia_valid & ATTR_KILL_SGID))
1810 attr->ia_valid |= ATTR_KILL_SGID;
1812 return ll_setattr_raw(de, attr, false);
1815 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1816 __u64 max_age, __u32 flags)
1818 struct ll_sb_info *sbi = ll_s2sbi(sb);
1819 struct obd_statfs obd_osfs;
1823 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1825 CERROR("md_statfs fails: rc = %d\n", rc);
1829 osfs->os_type = sb->s_magic;
1831 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1832 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1834 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1835 flags |= OBD_STATFS_NODELAY;
1837 rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1839 CERROR("obd_statfs fails: rc = %d\n", rc);
1843 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1844 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1847 osfs->os_bsize = obd_osfs.os_bsize;
1848 osfs->os_blocks = obd_osfs.os_blocks;
1849 osfs->os_bfree = obd_osfs.os_bfree;
1850 osfs->os_bavail = obd_osfs.os_bavail;
1852 /* If we don't have as many objects free on the OST as inodes
1853 * on the MDS, we reduce the total number of inodes to
1854 * compensate, so that the "inodes in use" number is correct.
1856 if (obd_osfs.os_ffree < osfs->os_ffree) {
1857 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1859 osfs->os_ffree = obd_osfs.os_ffree;
1864 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1866 struct super_block *sb = de->d_sb;
1867 struct obd_statfs osfs;
1868 __u64 fsid = huge_encode_dev(sb->s_dev);
1871 CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1872 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1874 /* Some amount of caching on the client is allowed */
1875 rc = ll_statfs_internal(sb, &osfs,
1876 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1881 statfs_unpack(sfs, &osfs);
1883 /* We need to downshift for all 32-bit kernels, because we can't
1884 * tell if the kernel is being called via sys_statfs64() or not.
1885 * Stop before overflowing f_bsize - in which case it is better
1886 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1887 if (sizeof(long) < 8) {
1888 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1891 osfs.os_blocks >>= 1;
1892 osfs.os_bfree >>= 1;
1893 osfs.os_bavail >>= 1;
1897 sfs->f_blocks = osfs.os_blocks;
1898 sfs->f_bfree = osfs.os_bfree;
1899 sfs->f_bavail = osfs.os_bavail;
1900 sfs->f_fsid.val[0] = (__u32)fsid;
1901 sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1905 void ll_inode_size_lock(struct inode *inode)
1907 struct ll_inode_info *lli;
1909 LASSERT(!S_ISDIR(inode->i_mode));
1911 lli = ll_i2info(inode);
1912 mutex_lock(&lli->lli_size_mutex);
1915 void ll_inode_size_unlock(struct inode *inode)
1917 struct ll_inode_info *lli;
1919 lli = ll_i2info(inode);
1920 mutex_unlock(&lli->lli_size_mutex);
1923 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1925 struct ll_inode_info *lli = ll_i2info(inode);
1926 struct mdt_body *body = md->body;
1927 struct lov_stripe_md *lsm = md->lsm;
1928 struct ll_sb_info *sbi = ll_i2sbi(inode);
1930 LASSERT((lsm != NULL) == ((body->mbo_valid & OBD_MD_FLEASIZE) != 0));
1932 if (!lli->lli_has_smd &&
1933 !(sbi->ll_flags & LL_SBI_LAYOUT_LOCK))
1934 cl_file_inode_init(inode, md);
1936 lli->lli_maxbytes = lsm->lsm_maxbytes;
1937 if (lli->lli_maxbytes > MAX_LFS_FILESIZE)
1938 lli->lli_maxbytes = MAX_LFS_FILESIZE;
1941 if (S_ISDIR(inode->i_mode)) {
1944 rc = ll_update_lsm_md(inode, md);
1949 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1950 if (body->mbo_valid & OBD_MD_FLRMTPERM)
1951 ll_update_remote_perm(inode, md->remote_perm);
1953 #ifdef CONFIG_FS_POSIX_ACL
1954 else if (body->mbo_valid & OBD_MD_FLACL) {
1955 spin_lock(&lli->lli_lock);
1956 if (lli->lli_posix_acl)
1957 posix_acl_release(lli->lli_posix_acl);
1958 lli->lli_posix_acl = md->posix_acl;
1959 spin_unlock(&lli->lli_lock);
1962 inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1963 sbi->ll_flags & LL_SBI_32BIT_API);
1964 inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1966 if (body->mbo_valid & OBD_MD_FLATIME) {
1967 if (body->mbo_atime > LTIME_S(inode->i_atime))
1968 LTIME_S(inode->i_atime) = body->mbo_atime;
1969 lli->lli_lvb.lvb_atime = body->mbo_atime;
1972 if (body->mbo_valid & OBD_MD_FLMTIME) {
1973 if (body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1974 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1975 "to "LPU64"\n", inode->i_ino,
1976 LTIME_S(inode->i_mtime), body->mbo_mtime);
1977 LTIME_S(inode->i_mtime) = body->mbo_mtime;
1979 lli->lli_lvb.lvb_mtime = body->mbo_mtime;
1982 if (body->mbo_valid & OBD_MD_FLCTIME) {
1983 if (body->mbo_ctime > LTIME_S(inode->i_ctime))
1984 LTIME_S(inode->i_ctime) = body->mbo_ctime;
1985 lli->lli_lvb.lvb_ctime = body->mbo_ctime;
1988 if (body->mbo_valid & OBD_MD_FLMODE)
1989 inode->i_mode = (inode->i_mode & S_IFMT) |
1990 (body->mbo_mode & ~S_IFMT);
1992 if (body->mbo_valid & OBD_MD_FLTYPE)
1993 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1994 (body->mbo_mode & S_IFMT);
1996 LASSERT(inode->i_mode != 0);
1997 if (S_ISREG(inode->i_mode))
1998 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1999 LL_MAX_BLKSIZE_BITS);
2001 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
2003 if (body->mbo_valid & OBD_MD_FLUID)
2004 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
2005 if (body->mbo_valid & OBD_MD_FLGID)
2006 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
2007 if (body->mbo_valid & OBD_MD_FLFLAGS)
2008 inode->i_flags = ll_ext_to_inode_flags(body->mbo_flags);
2009 if (body->mbo_valid & OBD_MD_FLNLINK)
2010 set_nlink(inode, body->mbo_nlink);
2011 if (body->mbo_valid & OBD_MD_FLRDEV)
2012 inode->i_rdev = old_decode_dev(body->mbo_rdev);
2014 if (body->mbo_valid & OBD_MD_FLID) {
2015 /* FID shouldn't be changed! */
2016 if (fid_is_sane(&lli->lli_fid)) {
2017 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
2018 "Trying to change FID "DFID
2019 " to the "DFID", inode "DFID"(%p)\n",
2020 PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
2021 PFID(ll_inode2fid(inode)), inode);
2023 lli->lli_fid = body->mbo_fid1;
2027 LASSERT(fid_seq(&lli->lli_fid) != 0);
2029 if (body->mbo_valid & OBD_MD_FLSIZE) {
2030 if (exp_connect_som(ll_i2mdexp(inode)) &&
2031 S_ISREG(inode->i_mode)) {
2032 struct lustre_handle lockh;
2035 /* As it is possible a blocking ast has been processed
2036 * by this time, we need to check there is an UPDATE
2037 * lock on the client and set LLIF_MDS_SIZE_LOCK holding
2039 mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
2040 &lockh, LDLM_FL_CBPENDING,
2044 if (lli->lli_flags & (LLIF_DONE_WRITING |
2045 LLIF_EPOCH_PENDING |
2047 CERROR("%s: inode "DFID" flags %u still"
2048 " has size authority! do not "
2049 "trust the size from MDS\n",
2050 sbi->ll_md_exp->exp_obd->obd_name,
2051 PFID(ll_inode2fid(inode)),
2054 /* Use old size assignment to avoid
2055 * deadlock bz14138 & bz14326 */
2056 i_size_write(inode, body->mbo_size);
2057 spin_lock(&lli->lli_lock);
2058 lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
2059 spin_unlock(&lli->lli_lock);
2061 ldlm_lock_decref(&lockh, mode);
2064 /* Use old size assignment to avoid
2065 * deadlock bz14138 & bz14326 */
2066 i_size_write(inode, body->mbo_size);
2069 "inode="DFID", updating i_size %llu\n",
2070 PFID(ll_inode2fid(inode)),
2071 (unsigned long long)body->mbo_size);
2074 if (body->mbo_valid & OBD_MD_FLBLOCKS)
2075 inode->i_blocks = body->mbo_blocks;
2078 if (body->mbo_valid & OBD_MD_FLMDSCAPA) {
2079 LASSERT(md->mds_capa);
2080 ll_add_capa(inode, md->mds_capa);
2083 if (body->mbo_valid & OBD_MD_FLOSSCAPA) {
2084 LASSERT(md->oss_capa);
2085 ll_add_capa(inode, md->oss_capa);
2088 if (body->mbo_valid & OBD_MD_TSTATE) {
2089 if (body->mbo_t_state & MS_RESTORE)
2090 lli->lli_flags |= LLIF_FILE_RESTORING;
2096 int ll_read_inode2(struct inode *inode, void *opaque)
2098 struct lustre_md *md = opaque;
2099 struct ll_inode_info *lli = ll_i2info(inode);
2103 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2104 PFID(&lli->lli_fid), inode);
2106 LASSERT(!lli->lli_has_smd);
2108 /* Core attributes from the MDS first. This is a new inode, and
2109 * the VFS doesn't zero times in the core inode so we have to do
2110 * it ourselves. They will be overwritten by either MDS or OST
2111 * attributes - we just need to make sure they aren't newer. */
2112 LTIME_S(inode->i_mtime) = 0;
2113 LTIME_S(inode->i_atime) = 0;
2114 LTIME_S(inode->i_ctime) = 0;
2116 rc = ll_update_inode(inode, md);
2120 /* OIDEBUG(inode); */
2122 /* initializing backing dev info. */
2123 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2126 if (S_ISREG(inode->i_mode)) {
2127 struct ll_sb_info *sbi = ll_i2sbi(inode);
2128 inode->i_op = &ll_file_inode_operations;
2129 inode->i_fop = sbi->ll_fop;
2130 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2132 } else if (S_ISDIR(inode->i_mode)) {
2133 inode->i_op = &ll_dir_inode_operations;
2134 inode->i_fop = &ll_dir_operations;
2136 } else if (S_ISLNK(inode->i_mode)) {
2137 inode->i_op = &ll_fast_symlink_inode_operations;
2140 inode->i_op = &ll_special_inode_operations;
2142 init_special_inode(inode, inode->i_mode,
2151 void ll_delete_inode(struct inode *inode)
2153 struct cl_inode_info *lli = cl_i2info(inode);
2156 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2157 /* discard all dirty pages before truncating them, required by
2158 * osc_extent implementation at LU-1030. */
2159 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF,
2160 CL_FSYNC_DISCARD, 1);
2162 truncate_inode_pages(&inode->i_data, 0);
2164 /* Workaround for LU-118 */
2165 if (inode->i_data.nrpages) {
2166 spin_lock_irq(&inode->i_data.tree_lock);
2167 spin_unlock_irq(&inode->i_data.tree_lock);
2168 LASSERTF(inode->i_data.nrpages == 0,
2169 "inode="DFID"(%p) nrpages=%lu, see "
2170 "http://jira.whamcloud.com/browse/LU-118\n",
2171 PFID(ll_inode2fid(inode)), inode,
2172 inode->i_data.nrpages);
2174 /* Workaround end */
2176 #ifdef HAVE_SBOPS_EVICT_INODE
2177 ll_clear_inode(inode);
2184 int ll_iocontrol(struct inode *inode, struct file *file,
2185 unsigned int cmd, unsigned long arg)
2187 struct ll_sb_info *sbi = ll_i2sbi(inode);
2188 struct ptlrpc_request *req = NULL;
2193 case FSFILT_IOC_GETFLAGS: {
2194 struct mdt_body *body;
2195 struct md_op_data *op_data;
2197 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2198 0, 0, LUSTRE_OPC_ANY,
2200 if (IS_ERR(op_data))
2201 RETURN(PTR_ERR(op_data));
2203 op_data->op_valid = OBD_MD_FLFLAGS;
2204 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2205 ll_finish_md_op_data(op_data);
2207 CERROR("%s: failure inode "DFID": rc = %d\n",
2208 sbi->ll_md_exp->exp_obd->obd_name,
2209 PFID(ll_inode2fid(inode)), rc);
2213 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2215 flags = body->mbo_flags;
2217 ptlrpc_req_finished(req);
2219 RETURN(put_user(flags, (int *)arg));
2221 case FSFILT_IOC_SETFLAGS: {
2222 struct lov_stripe_md *lsm;
2223 struct obd_info oinfo = { { { 0 } } };
2224 struct md_op_data *op_data;
2226 if (get_user(flags, (int *)arg))
2229 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2230 LUSTRE_OPC_ANY, NULL);
2231 if (IS_ERR(op_data))
2232 RETURN(PTR_ERR(op_data));
2234 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
2235 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2236 rc = md_setattr(sbi->ll_md_exp, op_data,
2237 NULL, 0, NULL, 0, &req, NULL);
2238 ll_finish_md_op_data(op_data);
2239 ptlrpc_req_finished(req);
2243 inode->i_flags = ll_ext_to_inode_flags(flags);
2245 lsm = ccc_inode_lsm_get(inode);
2246 if (!lsm_has_objects(lsm)) {
2247 ccc_inode_lsm_put(inode, lsm);
2251 OBDO_ALLOC(oinfo.oi_oa);
2253 ccc_inode_lsm_put(inode, lsm);
2257 oinfo.oi_oa->o_oi = lsm->lsm_oi;
2258 oinfo.oi_oa->o_flags = flags;
2259 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
2261 oinfo.oi_capa = ll_mdscapa_get(inode);
2262 obdo_set_parent_fid(oinfo.oi_oa, &ll_i2info(inode)->lli_fid);
2263 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
2264 capa_put(oinfo.oi_capa);
2265 OBDO_FREE(oinfo.oi_oa);
2266 ccc_inode_lsm_put(inode, lsm);
2268 if (rc && rc != -EPERM && rc != -EACCES)
2269 CERROR("osc_setattr_async fails: rc = %d\n", rc);
2280 int ll_flush_ctx(struct inode *inode)
2282 struct ll_sb_info *sbi = ll_i2sbi(inode);
2284 CDEBUG(D_SEC, "flush context for user %d\n",
2285 from_kuid(&init_user_ns, current_uid()));
2287 obd_set_info_async(NULL, sbi->ll_md_exp,
2288 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2290 obd_set_info_async(NULL, sbi->ll_dt_exp,
2291 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2296 /* umount -f client means force down, don't save state */
2297 void ll_umount_begin(struct super_block *sb)
2299 struct ll_sb_info *sbi = ll_s2sbi(sb);
2300 struct obd_device *obd;
2301 struct obd_ioctl_data *ioc_data;
2304 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2305 sb->s_count, atomic_read(&sb->s_active));
2307 obd = class_exp2obd(sbi->ll_md_exp);
2309 CERROR("Invalid MDC connection handle "LPX64"\n",
2310 sbi->ll_md_exp->exp_handle.h_cookie);
2316 obd = class_exp2obd(sbi->ll_dt_exp);
2318 CERROR("Invalid LOV connection handle "LPX64"\n",
2319 sbi->ll_dt_exp->exp_handle.h_cookie);
2325 OBD_ALLOC_PTR(ioc_data);
2327 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2328 sizeof *ioc_data, ioc_data, NULL);
2330 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2331 sizeof *ioc_data, ioc_data, NULL);
2333 OBD_FREE_PTR(ioc_data);
2336 /* Really, we'd like to wait until there are no requests outstanding,
2337 * and then continue. For now, we just invalidate the requests,
2338 * schedule() and sleep one second if needed, and hope.
2344 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2346 struct ll_sb_info *sbi = ll_s2sbi(sb);
2347 char *profilenm = get_profile_name(sb);
2351 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2352 read_only = *flags & MS_RDONLY;
2353 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2354 sizeof(KEY_READ_ONLY),
2355 KEY_READ_ONLY, sizeof(read_only),
2358 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2359 profilenm, read_only ?
2360 "read-only" : "read-write", err);
2365 sb->s_flags |= MS_RDONLY;
2367 sb->s_flags &= ~MS_RDONLY;
2369 if (sbi->ll_flags & LL_SBI_VERBOSE)
2370 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2371 read_only ? "read-only" : "read-write");
2376 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2377 struct super_block *sb, struct lookup_intent *it)
2379 struct ll_sb_info *sbi = NULL;
2380 struct lustre_md md = { 0 };
2384 LASSERT(*inode || sb);
2385 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2386 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2387 sbi->ll_md_exp, &md);
2392 rc = ll_update_inode(*inode, &md);
2396 LASSERT(sb != NULL);
2399 * At this point server returns to client's same fid as client
2400 * generated for creating. So using ->fid1 is okay here.
2402 LASSERT(fid_is_sane(&md.body->mbo_fid1));
2404 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2405 sbi->ll_flags & LL_SBI_32BIT_API),
2407 if (IS_ERR(*inode)) {
2408 #ifdef CONFIG_FS_POSIX_ACL
2410 posix_acl_release(md.posix_acl);
2411 md.posix_acl = NULL;
2414 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2416 CERROR("new_inode -fatal: rc %d\n", rc);
2421 /* Handling piggyback layout lock.
2422 * Layout lock can be piggybacked by getattr and open request.
2423 * The lsm can be applied to inode only if it comes with a layout lock
2424 * otherwise correct layout may be overwritten, for example:
2425 * 1. proc1: mdt returns a lsm but not granting layout
2426 * 2. layout was changed by another client
2427 * 3. proc2: refresh layout and layout lock granted
2428 * 4. proc1: to apply a stale layout */
2429 if (it != NULL && it->d.lustre.it_lock_mode != 0) {
2430 struct lustre_handle lockh;
2431 struct ldlm_lock *lock;
2433 lockh.cookie = it->d.lustre.it_lock_handle;
2434 lock = ldlm_handle2lock(&lockh);
2435 LASSERT(lock != NULL);
2436 if (ldlm_has_layout(lock)) {
2437 struct cl_object_conf conf;
2439 memset(&conf, 0, sizeof(conf));
2440 conf.coc_opc = OBJECT_CONF_SET;
2441 conf.coc_inode = *inode;
2442 conf.coc_lock = lock;
2443 conf.u.coc_md = &md;
2444 (void)ll_layout_conf(*inode, &conf);
2446 LDLM_LOCK_PUT(lock);
2451 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2452 md_free_lustre_md(sbi->ll_md_exp, &md);
2456 int ll_obd_statfs(struct inode *inode, void *arg)
2458 struct ll_sb_info *sbi = NULL;
2459 struct obd_export *exp;
2461 struct obd_ioctl_data *data = NULL;
2466 if (!inode || !(sbi = ll_i2sbi(inode)))
2467 GOTO(out_statfs, rc = -EINVAL);
2469 rc = obd_ioctl_getdata(&buf, &len, arg);
2471 GOTO(out_statfs, rc);
2474 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2475 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2476 GOTO(out_statfs, rc = -EINVAL);
2478 if (data->ioc_inllen1 != sizeof(__u32) ||
2479 data->ioc_inllen2 != sizeof(__u32) ||
2480 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2481 data->ioc_plen2 != sizeof(struct obd_uuid))
2482 GOTO(out_statfs, rc = -EINVAL);
2484 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2485 if (type & LL_STATFS_LMV)
2486 exp = sbi->ll_md_exp;
2487 else if (type & LL_STATFS_LOV)
2488 exp = sbi->ll_dt_exp;
2490 GOTO(out_statfs, rc = -ENODEV);
2492 flags = (type & LL_STATFS_NODELAY) ? OBD_STATFS_NODELAY : 0;
2493 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, &flags);
2495 GOTO(out_statfs, rc);
2498 obd_ioctl_freedata(buf, len);
2502 int ll_process_config(struct lustre_cfg *lcfg)
2504 struct super_block *sb;
2509 /* The instance name contains the sb: lustre-client-aacfe000 */
2510 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2511 if (!ptr || !*(++ptr))
2513 if (sscanf(ptr, "%lx", &x) != 1)
2515 sb = (struct super_block *)x;
2516 /* This better be a real Lustre superblock! */
2517 LASSERT(s2lsi(sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2519 /* Note we have not called client_common_fill_super yet, so
2520 proc fns must be able to handle that! */
2521 rc = class_process_proc_seq_param(PARAM_LLITE, lprocfs_llite_obd_vars,
2528 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2529 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2530 struct inode *i1, struct inode *i2,
2531 const char *name, int namelen,
2532 int mode, __u32 opc, void *data)
2534 LASSERT(i1 != NULL);
2537 /* Do not reuse namelen for something else. */
2539 return ERR_PTR(-EINVAL);
2541 if (namelen > ll_i2sbi(i1)->ll_namelen)
2542 return ERR_PTR(-ENAMETOOLONG);
2544 if (!lu_name_is_valid_2(name, namelen))
2545 return ERR_PTR(-EINVAL);
2548 if (op_data == NULL)
2549 OBD_ALLOC_PTR(op_data);
2551 if (op_data == NULL)
2552 return ERR_PTR(-ENOMEM);
2554 ll_i2gids(op_data->op_suppgids, i1, i2);
2555 op_data->op_fid1 = *ll_inode2fid(i1);
2556 op_data->op_capa1 = ll_mdscapa_get(i1);
2557 if (S_ISDIR(i1->i_mode))
2558 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2561 op_data->op_fid2 = *ll_inode2fid(i2);
2562 op_data->op_capa2 = ll_mdscapa_get(i2);
2563 if (S_ISDIR(i2->i_mode))
2564 op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2566 fid_zero(&op_data->op_fid2);
2567 op_data->op_capa2 = NULL;
2570 if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2571 op_data->op_cli_flags |= CLI_HASH64;
2573 if (ll_need_32bit_api(ll_i2sbi(i1)))
2574 op_data->op_cli_flags |= CLI_API32;
2576 op_data->op_name = name;
2577 op_data->op_namelen = namelen;
2578 op_data->op_mode = mode;
2579 op_data->op_mod_time = cfs_time_current_sec();
2580 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2581 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2582 op_data->op_cap = cfs_curproc_cap_pack();
2583 op_data->op_bias = 0;
2584 op_data->op_cli_flags = 0;
2585 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2586 filename_is_volatile(name, namelen, NULL))
2587 op_data->op_bias |= MDS_CREATE_VOLATILE;
2588 op_data->op_mds = 0;
2589 op_data->op_data = data;
2591 /* If the file is being opened after mknod() (normally due to NFS)
2592 * try to use the default stripe data from parent directory for
2593 * allocating OST objects. Try to pass the parent FID to MDS. */
2594 if (opc == LUSTRE_OPC_CREATE && i1 == i2 && S_ISREG(i2->i_mode) &&
2595 !ll_i2info(i2)->lli_has_smd) {
2596 struct ll_inode_info *lli = ll_i2info(i2);
2598 spin_lock(&lli->lli_lock);
2599 if (likely(!lli->lli_has_smd && !fid_is_zero(&lli->lli_pfid)))
2600 op_data->op_fid1 = lli->lli_pfid;
2601 spin_unlock(&lli->lli_lock);
2602 /** We ignore parent's capability temporary. */
2605 /* When called by ll_setattr_raw, file is i1. */
2606 if (LLIF_DATA_MODIFIED & ll_i2info(i1)->lli_flags)
2607 op_data->op_bias |= MDS_DATA_MODIFIED;
2612 void ll_finish_md_op_data(struct md_op_data *op_data)
2614 capa_put(op_data->op_capa1);
2615 capa_put(op_data->op_capa2);
2616 OBD_FREE_PTR(op_data);
2619 #ifdef HAVE_SUPEROPS_USE_DENTRY
2620 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2622 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2625 struct ll_sb_info *sbi;
2627 #ifdef HAVE_SUPEROPS_USE_DENTRY
2628 LASSERT((seq != NULL) && (dentry != NULL));
2629 sbi = ll_s2sbi(dentry->d_sb);
2631 LASSERT((seq != NULL) && (vfs != NULL));
2632 sbi = ll_s2sbi(vfs->mnt_sb);
2635 if (sbi->ll_flags & LL_SBI_NOLCK)
2636 seq_puts(seq, ",nolock");
2638 if (sbi->ll_flags & LL_SBI_FLOCK)
2639 seq_puts(seq, ",flock");
2641 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2642 seq_puts(seq, ",localflock");
2644 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2645 seq_puts(seq, ",user_xattr");
2647 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2648 seq_puts(seq, ",lazystatfs");
2650 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2651 seq_puts(seq, ",user_fid2path");
2657 * Get obd name by cmd, and copy out to user space
2659 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2661 struct ll_sb_info *sbi = ll_i2sbi(inode);
2662 struct obd_device *obd;
2665 if (cmd == OBD_IOC_GETDTNAME)
2666 obd = class_exp2obd(sbi->ll_dt_exp);
2667 else if (cmd == OBD_IOC_GETMDNAME)
2668 obd = class_exp2obd(sbi->ll_md_exp);
2675 if (copy_to_user((void *)arg, obd->obd_name,
2676 strlen(obd->obd_name) + 1))
2683 * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2684 * fsname will be returned in this buffer; otherwise, a static buffer will be
2685 * used to store the fsname and returned to caller.
2687 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2689 static char fsname_static[MTI_NAME_MAXLEN];
2690 struct lustre_sb_info *lsi = s2lsi(sb);
2695 /* this means the caller wants to use static buffer
2696 * and it doesn't care about race. Usually this is
2697 * in error reporting path */
2698 buf = fsname_static;
2699 buflen = sizeof(fsname_static);
2702 len = strlen(lsi->lsi_lmd->lmd_profile);
2703 ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2704 if (ptr && (strcmp(ptr, "-client") == 0))
2707 if (unlikely(len >= buflen))
2709 strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2715 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2722 p.mnt = current->fs->root.mnt;
2724 path = d_path(&p, buf, bufsize);
2729 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2731 char *buf, *path = NULL;
2732 struct dentry *dentry = NULL;
2733 struct ccc_object *obj = cl_inode2ccc(page->mapping->host);
2735 /* this can be called inside spin lock so use GFP_ATOMIC. */
2736 buf = (char *)__get_free_page(GFP_ATOMIC);
2738 dentry = d_find_alias(page->mapping->host);
2740 path = ll_d_path(dentry, buf, PAGE_SIZE);
2744 "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2745 "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2746 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2747 PFID(&obj->cob_header.coh_lu.loh_fid),
2748 (path && !IS_ERR(path)) ? path : "", ioret);
2754 free_page((unsigned long)buf);
2758 * Compute llite root squash state after a change of root squash
2759 * configuration setting or add/remove of a lnet nid
2761 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2763 struct root_squash_info *squash = &sbi->ll_squash;
2766 lnet_process_id_t id;
2768 /* Update norootsquash flag */
2769 down_write(&squash->rsi_sem);
2770 if (list_empty(&squash->rsi_nosquash_nids))
2771 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2773 /* Do not apply root squash as soon as one of our NIDs is
2774 * in the nosquash_nids list */
2777 while (LNetGetId(i++, &id) != -ENOENT) {
2778 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2780 if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2786 sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2788 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2790 up_write(&squash->rsi_sem);