4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2012, 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/statfs.h>
45 #include <linux/types.h>
48 #include "../include/lustre_lite.h"
49 #include "../include/lustre_ha.h"
50 #include "../include/lustre_dlm.h"
51 #include "../include/lprocfs_status.h"
52 #include "../include/lustre_disk.h"
53 #include "../include/lustre_param.h"
54 #include "../include/lustre_log.h"
55 #include "../include/cl_object.h"
56 #include "../include/obd_cksum.h"
57 #include "llite_internal.h"
59 struct kmem_cache *ll_file_data_slab;
60 struct proc_dir_entry *proc_lustre_fs_root;
62 static LIST_HEAD(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;
78 sbi = kzalloc(sizeof(*sbi), GFP_NOFS);
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 if (pages >> (20 - PAGE_CACHE_SHIFT) < 512)
91 lru_page_max = pages / 2;
93 lru_page_max = (pages / 4) * 3;
95 /* initialize lru data */
96 atomic_set(&sbi->ll_cache.ccc_users, 0);
97 sbi->ll_cache.ccc_lru_max = lru_page_max;
98 atomic_set(&sbi->ll_cache.ccc_lru_left, lru_page_max);
99 spin_lock_init(&sbi->ll_cache.ccc_lru_lock);
100 INIT_LIST_HEAD(&sbi->ll_cache.ccc_lru);
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 INIT_LIST_HEAD(&sbi->ll_conn_chain);
108 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 list_add_tail(&sbi->ll_list, &ll_super_blocks);
116 spin_unlock(&ll_sb_lock);
118 sbi->ll_flags |= LL_SBI_VERBOSE;
119 sbi->ll_flags |= LL_SBI_CHECKSUM;
121 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
123 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
124 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
126 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
130 /* metadata statahead is enabled by default */
131 sbi->ll_sa_max = LL_SA_RPC_DEF;
132 atomic_set(&sbi->ll_sa_total, 0);
133 atomic_set(&sbi->ll_sa_wrong, 0);
134 atomic_set(&sbi->ll_agl_total, 0);
135 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
140 static void ll_free_sbi(struct super_block *sb)
142 struct ll_sb_info *sbi = ll_s2sbi(sb);
145 spin_lock(&ll_sb_lock);
146 list_del(&sbi->ll_list);
147 spin_unlock(&ll_sb_lock);
148 OBD_FREE(sbi, sizeof(*sbi));
152 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
153 struct vfsmount *mnt)
155 struct inode *root = NULL;
156 struct ll_sb_info *sbi = ll_s2sbi(sb);
157 struct obd_device *obd;
158 struct obd_capa *oc = NULL;
159 struct obd_statfs *osfs = NULL;
160 struct ptlrpc_request *request = NULL;
161 struct obd_connect_data *data = NULL;
162 struct obd_uuid *uuid;
163 struct md_op_data *op_data;
164 struct lustre_md lmd;
166 int size, err, checksum;
168 obd = class_name2obd(md);
170 CERROR("MD %s: not setup or attached\n", md);
174 data = kzalloc(sizeof(*data), GFP_NOFS);
178 osfs = kzalloc(sizeof(*osfs), GFP_NOFS);
184 if (proc_lustre_fs_root) {
185 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
188 CERROR("could not register mount in /proc/fs/lustre\n");
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 |
203 OBD_CONNECT_PINGLESS |
204 OBD_CONNECT_MAX_EASIZE |
205 OBD_CONNECT_FLOCK_DEAD |
206 OBD_CONNECT_DISP_STRIPE;
208 if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
209 data->ocd_connect_flags |= OBD_CONNECT_SOM;
211 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
212 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
213 #ifdef CONFIG_FS_POSIX_ACL
214 data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK;
217 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
218 /* flag mdc connection as lightweight, only used for test
219 * purpose, use with care */
220 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
222 data->ocd_ibits_known = MDS_INODELOCK_FULL;
223 data->ocd_version = LUSTRE_VERSION_CODE;
225 if (sb->s_flags & MS_RDONLY)
226 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
227 if (sbi->ll_flags & LL_SBI_USER_XATTR)
228 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
230 #ifdef HAVE_MS_FLOCK_LOCK
231 /* force vfs to use lustre handler for flock() calls - bug 10743 */
232 sb->s_flags |= MS_FLOCK_LOCK;
234 #ifdef MS_HAS_NEW_AOPS
235 sb->s_flags |= MS_HAS_NEW_AOPS;
238 if (sbi->ll_flags & LL_SBI_FLOCK)
239 sbi->ll_fop = &ll_file_operations_flock;
240 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
241 sbi->ll_fop = &ll_file_operations;
243 sbi->ll_fop = &ll_file_operations_noflock;
246 data->ocd_connect_flags |= OBD_CONNECT_REAL;
247 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
248 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
250 data->ocd_brw_size = MD_MAX_BRW_SIZE;
252 err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid,
255 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing recovery, of which this client is not a part. Please wait for recovery to complete, abort, or time out.\n",
259 CERROR("cannot connect to %s: rc = %d\n", md, err);
263 sbi->ll_md_exp->exp_connect_data = *data;
265 err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
266 LUSTRE_SEQ_METADATA);
268 CERROR("%s: Can't init metadata layer FID infrastructure, rc = %d\n",
269 sbi->ll_md_exp->exp_obd->obd_name, err);
273 /* For mount, we only need fs info from MDT0, and also in DNE, it
274 * can make sure the client can be mounted as long as MDT0 is
276 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
277 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
278 OBD_STATFS_FOR_MDT0);
282 /* This needs to be after statfs to ensure connect has finished.
283 * Note that "data" does NOT contain the valid connect reply.
284 * If connecting to a 1.8 server there will be no LMV device, so
285 * we can access the MDC export directly and exp_connect_flags will
286 * be non-zero, but if accessing an upgraded 2.1 server it will
287 * have the correct flags filled in.
288 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
289 valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
290 if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
291 valid != CLIENT_CONNECT_MDT_REQD) {
294 buf = kzalloc(PAGE_CACHE_SIZE, GFP_KERNEL);
295 obd_connect_flags2str(buf, PAGE_CACHE_SIZE,
296 valid ^ CLIENT_CONNECT_MDT_REQD, ",");
297 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support feature(s) needed for correct operation of this client (%s). Please upgrade server or downgrade client.\n",
298 sbi->ll_md_exp->exp_obd->obd_name, buf);
299 OBD_FREE(buf, PAGE_CACHE_SIZE);
304 size = sizeof(*data);
305 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
306 KEY_CONN_DATA, &size, data, NULL);
308 CERROR("%s: Get connect data failed: rc = %d\n",
309 sbi->ll_md_exp->exp_obd->obd_name, err);
313 LASSERT(osfs->os_bsize);
314 sb->s_blocksize = osfs->os_bsize;
315 sb->s_blocksize_bits = log2(osfs->os_bsize);
316 sb->s_magic = LL_SUPER_MAGIC;
317 sb->s_maxbytes = MAX_LFS_FILESIZE;
318 sbi->ll_namelen = osfs->os_namelen;
319 sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
321 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
322 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
323 LCONSOLE_INFO("Disabling user_xattr feature because it is not supported on the server\n");
324 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
327 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
329 sb->s_flags |= MS_POSIXACL;
331 sbi->ll_flags |= LL_SBI_ACL;
333 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
335 sb->s_flags &= ~MS_POSIXACL;
337 sbi->ll_flags &= ~LL_SBI_ACL;
340 if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
341 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
342 sbi->ll_flags |= LL_SBI_RMT_CLIENT;
343 LCONSOLE_INFO("client is set as remote by default.\n");
346 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
347 sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
348 LCONSOLE_INFO("client claims to be remote, but server rejected, forced to be local.\n");
352 if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
353 LCONSOLE_INFO("client enabled MDS capability!\n");
354 sbi->ll_flags |= LL_SBI_MDS_CAPA;
357 if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
358 LCONSOLE_INFO("client enabled OSS capability!\n");
359 sbi->ll_flags |= LL_SBI_OSS_CAPA;
362 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
363 sbi->ll_flags |= LL_SBI_64BIT_HASH;
365 if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
366 sbi->ll_md_brw_size = data->ocd_brw_size;
368 sbi->ll_md_brw_size = PAGE_CACHE_SIZE;
370 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK) {
371 LCONSOLE_INFO("Layout lock feature supported.\n");
372 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
375 if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
376 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
378 "%s: disabling xattr cache due to unknown maximum xattr size.\n",
381 sbi->ll_flags |= LL_SBI_XATTR_CACHE;
382 sbi->ll_xattr_cache_enabled = 1;
386 obd = class_name2obd(dt);
388 CERROR("DT %s: not setup or attached\n", dt);
393 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
394 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
395 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
396 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
397 OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
398 OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
399 OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
400 OBD_CONNECT_MAXBYTES |
401 OBD_CONNECT_EINPROGRESS |
402 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
403 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS;
405 if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
406 data->ocd_connect_flags |= OBD_CONNECT_SOM;
408 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
409 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
410 * disabled by default, because it can still be enabled on the
411 * fly via /proc. As a consequence, we still need to come to an
412 * agreement on the supported algorithms at connect time */
413 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
415 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
416 data->ocd_cksum_types = OBD_CKSUM_ADLER;
418 data->ocd_cksum_types = cksum_types_supported_client();
421 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
422 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
423 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
425 CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d ocd_grant: %d\n",
426 data->ocd_connect_flags,
427 data->ocd_version, data->ocd_grant);
429 obd->obd_upcall.onu_owner = &sbi->ll_lco;
430 obd->obd_upcall.onu_upcall = cl_ocd_update;
432 data->ocd_brw_size = DT_MAX_BRW_SIZE;
434 err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
437 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing recovery, of which this client is not a part. Please wait for recovery to complete, abort, or time out.\n",
441 CERROR("%s: Cannot connect to %s: rc = %d\n",
442 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
446 sbi->ll_dt_exp->exp_connect_data = *data;
448 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
449 LUSTRE_SEQ_METADATA);
451 CERROR("%s: Can't init data layer FID infrastructure, rc = %d\n",
452 sbi->ll_dt_exp->exp_obd->obd_name, err);
456 mutex_lock(&sbi->ll_lco.lco_lock);
457 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
458 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
459 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
460 mutex_unlock(&sbi->ll_lco.lco_lock);
462 fid_zero(&sbi->ll_root_fid);
463 err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
465 CERROR("cannot mds_connect: rc = %d\n", err);
468 if (!fid_is_sane(&sbi->ll_root_fid)) {
469 CERROR("%s: Invalid root fid "DFID" during mount\n",
470 sbi->ll_md_exp->exp_obd->obd_name,
471 PFID(&sbi->ll_root_fid));
475 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
477 sb->s_op = &lustre_super_operations;
478 #if THREAD_SIZE >= 8192 /*b=17630*/
479 sb->s_export_op = &lustre_export_operations;
483 * XXX: move this to after cbd setup? */
484 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
485 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
486 valid |= OBD_MD_FLRMTPERM;
487 else if (sbi->ll_flags & LL_SBI_ACL)
488 valid |= OBD_MD_FLACL;
490 op_data = kzalloc(sizeof(*op_data), GFP_NOFS);
496 op_data->op_fid1 = sbi->ll_root_fid;
497 op_data->op_mode = 0;
498 op_data->op_capa1 = oc;
499 op_data->op_valid = valid;
501 err = md_getattr(sbi->ll_md_exp, op_data, &request);
504 OBD_FREE_PTR(op_data);
506 CERROR("%s: md_getattr failed for root: rc = %d\n",
507 sbi->ll_md_exp->exp_obd->obd_name, err);
511 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
512 sbi->ll_md_exp, &lmd);
514 CERROR("failed to understand root inode md: rc = %d\n", err);
515 ptlrpc_req_finished(request);
519 LASSERT(fid_is_sane(&sbi->ll_root_fid));
520 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
521 sbi->ll_flags & LL_SBI_32BIT_API),
523 md_free_lustre_md(sbi->ll_md_exp, &lmd);
524 ptlrpc_req_finished(request);
526 if (root == NULL || IS_ERR(root)) {
528 obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
529 #ifdef CONFIG_FS_POSIX_ACL
531 posix_acl_release(lmd.posix_acl);
532 lmd.posix_acl = NULL;
535 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
537 CERROR("lustre_lite: bad iget4 for root\n");
541 err = ll_close_thread_start(&sbi->ll_lcq);
543 CERROR("cannot start close thread: rc %d\n", err);
547 #ifdef CONFIG_FS_POSIX_ACL
548 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
549 rct_init(&sbi->ll_rct);
550 et_init(&sbi->ll_et);
554 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
555 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
556 KEY_CHECKSUM, sizeof(checksum), &checksum,
560 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
561 KEY_CACHE_SET, sizeof(sbi->ll_cache),
562 &sbi->ll_cache, NULL);
564 sb->s_root = d_make_root(root);
565 if (sb->s_root == NULL) {
566 CERROR("%s: can't make root dentry\n",
567 ll_get_fsname(sb, NULL, 0));
572 sbi->ll_sdev_orig = sb->s_dev;
574 /* We set sb->s_dev equal on all lustre clients in order to support
575 * NFS export clustering. NFSD requires that the FSID be the same
577 /* s_dev is also used in lt_compare() to compare two fs, but that is
578 * only a node-local comparison. */
579 uuid = obd_get_uuid(sbi->ll_md_exp);
581 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
582 get_uuid2fsid(uuid->uuid, strlen(uuid->uuid), &sbi->ll_fsid);
594 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
596 obd_disconnect(sbi->ll_dt_exp);
597 sbi->ll_dt_exp = NULL;
598 /* Make sure all OScs are gone, since cl_cache is accessing sbi. */
599 obd_zombie_barrier();
601 obd_fid_fini(sbi->ll_md_exp->exp_obd);
603 obd_disconnect(sbi->ll_md_exp);
604 sbi->ll_md_exp = NULL;
610 lprocfs_unregister_mountpoint(sbi);
614 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
618 *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
620 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
621 KEY_MAX_EASIZE, &size, lmmsize, NULL);
623 CERROR("Get max mdsize error rc %d\n", rc);
628 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
633 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
634 KEY_DEFAULT_EASIZE, &size, lmmsize, NULL);
636 CERROR("Get default mdsize error rc %d\n", rc);
641 int ll_get_max_cookiesize(struct ll_sb_info *sbi, int *lmmsize)
646 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_COOKIESIZE),
647 KEY_MAX_COOKIESIZE, &size, lmmsize, NULL);
649 CERROR("Get max cookiesize error rc %d\n", rc);
654 int ll_get_default_cookiesize(struct ll_sb_info *sbi, int *lmmsize)
659 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_COOKIESIZE),
660 KEY_DEFAULT_COOKIESIZE, &size, lmmsize, NULL);
662 CERROR("Get default cookiesize error rc %d\n", rc);
667 static void client_common_put_super(struct super_block *sb)
669 struct ll_sb_info *sbi = ll_s2sbi(sb);
671 #ifdef CONFIG_FS_POSIX_ACL
672 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
673 et_fini(&sbi->ll_et);
674 rct_fini(&sbi->ll_rct);
678 ll_close_thread_shutdown(sbi->ll_lcq);
682 list_del(&sbi->ll_conn_chain);
684 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
685 obd_disconnect(sbi->ll_dt_exp);
686 sbi->ll_dt_exp = NULL;
687 /* wait till all OSCs are gone, since cl_cache is accessing sbi.
689 obd_zombie_barrier();
691 lprocfs_unregister_mountpoint(sbi);
693 obd_fid_fini(sbi->ll_md_exp->exp_obd);
694 obd_disconnect(sbi->ll_md_exp);
695 sbi->ll_md_exp = NULL;
698 void ll_kill_super(struct super_block *sb)
700 struct ll_sb_info *sbi;
703 if (!(sb->s_flags & MS_ACTIVE))
707 /* we need to restore s_dev from changed for clustered NFS before
708 * put_super because new kernels have cached s_dev and change sb->s_dev
709 * in put_super not affected real removing devices */
711 sb->s_dev = sbi->ll_sdev_orig;
712 sbi->ll_umounting = 1;
716 static inline int ll_set_opt(const char *opt, char *data, int fl)
718 if (strncmp(opt, data, strlen(opt)) != 0)
724 /* non-client-specific mount options are parsed in lmd_parse */
725 static int ll_options(char *options, int *flags)
728 char *s1 = options, *s2;
733 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
736 CDEBUG(D_SUPER, "next opt=%s\n", s1);
737 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
742 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
747 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
752 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
757 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
762 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
767 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
772 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
777 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
783 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
788 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
793 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
798 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
803 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
808 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
813 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
818 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
823 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
828 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
833 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
839 s2 = strchr(s1, ',');
847 void ll_lli_init(struct ll_inode_info *lli)
849 lli->lli_inode_magic = LLI_INODE_MAGIC;
851 lli->lli_ioepoch = 0;
852 lli->lli_maxbytes = MAX_LFS_FILESIZE;
853 spin_lock_init(&lli->lli_lock);
854 lli->lli_posix_acl = NULL;
855 lli->lli_remote_perms = NULL;
856 mutex_init(&lli->lli_rmtperm_mutex);
857 /* Do not set lli_fid, it has been initialized already. */
858 fid_zero(&lli->lli_pfid);
859 INIT_LIST_HEAD(&lli->lli_close_list);
860 INIT_LIST_HEAD(&lli->lli_oss_capas);
861 atomic_set(&lli->lli_open_count, 0);
862 lli->lli_mds_capa = NULL;
863 lli->lli_rmtperm_time = 0;
864 lli->lli_pending_och = NULL;
865 lli->lli_mds_read_och = NULL;
866 lli->lli_mds_write_och = NULL;
867 lli->lli_mds_exec_och = NULL;
868 lli->lli_open_fd_read_count = 0;
869 lli->lli_open_fd_write_count = 0;
870 lli->lli_open_fd_exec_count = 0;
871 mutex_init(&lli->lli_och_mutex);
872 spin_lock_init(&lli->lli_agl_lock);
873 lli->lli_has_smd = false;
874 spin_lock_init(&lli->lli_layout_lock);
875 ll_layout_version_set(lli, LL_LAYOUT_GEN_NONE);
876 lli->lli_clob = NULL;
878 init_rwsem(&lli->lli_xattrs_list_rwsem);
879 mutex_init(&lli->lli_xattrs_enq_lock);
881 LASSERT(lli->lli_vfs_inode.i_mode != 0);
882 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
883 mutex_init(&lli->lli_readdir_mutex);
884 lli->lli_opendir_key = NULL;
886 spin_lock_init(&lli->lli_sa_lock);
887 lli->lli_opendir_pid = 0;
889 mutex_init(&lli->lli_size_mutex);
890 lli->lli_symlink_name = NULL;
891 init_rwsem(&lli->lli_trunc_sem);
892 mutex_init(&lli->lli_write_mutex);
893 init_rwsem(&lli->lli_glimpse_sem);
894 lli->lli_glimpse_time = 0;
895 INIT_LIST_HEAD(&lli->lli_agl_list);
896 lli->lli_agl_index = 0;
897 lli->lli_async_rc = 0;
899 mutex_init(&lli->lli_layout_mutex);
902 static inline int ll_bdi_register(struct backing_dev_info *bdi)
904 static atomic_t ll_bdi_num = ATOMIC_INIT(0);
906 bdi->name = "lustre";
907 return bdi_register(bdi, NULL, "lustre-%d",
908 atomic_inc_return(&ll_bdi_num));
911 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
913 struct lustre_profile *lprof = NULL;
914 struct lustre_sb_info *lsi = s2lsi(sb);
915 struct ll_sb_info *sbi;
916 char *dt = NULL, *md = NULL;
917 char *profilenm = get_profile_name(sb);
918 struct config_llog_instance *cfg;
919 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
920 const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
923 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
925 cfg = kzalloc(sizeof(*cfg), GFP_NOFS);
929 try_module_get(THIS_MODULE);
931 /* client additional sb info */
932 lsi->lsi_llsbi = sbi = ll_init_sbi();
934 module_put(THIS_MODULE);
939 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
943 err = bdi_init(&lsi->lsi_bdi);
946 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
947 lsi->lsi_bdi.capabilities = 0;
948 err = ll_bdi_register(&lsi->lsi_bdi);
952 sb->s_bdi = &lsi->lsi_bdi;
953 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
954 sb->s_d_op = &ll_d_ops;
956 /* Generate a string unique to this super, in case some joker tries
957 to mount the same fs at two mount points.
958 Use the address of the super itself.*/
959 cfg->cfg_instance = sb;
960 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
961 cfg->cfg_callback = class_config_llog_handler;
962 /* set up client obds */
963 err = lustre_process_log(sb, profilenm, cfg);
965 CERROR("Unable to process log: %d\n", err);
969 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
970 lprof = class_get_profile(profilenm);
972 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be read from the MGS. Does that filesystem exist?\n",
977 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
978 lprof->lp_md, lprof->lp_dt);
980 dt = kasprintf(GFP_NOFS, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
986 md = kasprintf(GFP_NOFS, "%s-%p", lprof->lp_md, cfg->cfg_instance);
992 /* connections, registrations, sb setup */
993 err = client_common_fill_super(sb, md, dt, mnt);
997 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
999 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1002 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1003 LCONSOLE_WARN("Mounted %s\n", profilenm);
1007 } /* ll_fill_super */
1009 void ll_put_super(struct super_block *sb)
1011 struct config_llog_instance cfg, params_cfg;
1012 struct obd_device *obd;
1013 struct lustre_sb_info *lsi = s2lsi(sb);
1014 struct ll_sb_info *sbi = ll_s2sbi(sb);
1015 char *profilenm = get_profile_name(sb);
1016 int next, force = 1;
1018 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1020 ll_print_capa_stat(sbi);
1022 cfg.cfg_instance = sb;
1023 lustre_end_log(sb, profilenm, &cfg);
1025 params_cfg.cfg_instance = sb;
1026 lustre_end_log(sb, PARAMS_FILENAME, ¶ms_cfg);
1028 if (sbi->ll_md_exp) {
1029 obd = class_exp2obd(sbi->ll_md_exp);
1031 force = obd->obd_force;
1034 /* We need to set force before the lov_disconnect in
1035 lustre_common_put_super, since l_d cleans up osc's as well. */
1038 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1040 obd->obd_force = force;
1045 /* Only if client_common_fill_super succeeded */
1046 client_common_put_super(sb);
1050 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)))
1051 class_manual_cleanup(obd);
1053 if (sbi->ll_flags & LL_SBI_VERBOSE)
1054 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1057 class_del_profile(profilenm);
1059 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1060 bdi_destroy(&lsi->lsi_bdi);
1061 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1065 lsi->lsi_llsbi = NULL;
1067 lustre_common_put_super(sb);
1069 module_put(THIS_MODULE);
1070 } /* client_put_super */
1072 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1074 struct inode *inode = NULL;
1076 /* NOTE: we depend on atomic igrab() -bzzz */
1077 lock_res_and_lock(lock);
1078 if (lock->l_resource->lr_lvb_inode) {
1079 struct ll_inode_info *lli;
1081 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1082 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1083 inode = igrab(lock->l_resource->lr_lvb_inode);
1085 inode = lock->l_resource->lr_lvb_inode;
1086 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1087 D_WARNING, lock, "lr_lvb_inode %p is bogus: magic %08x",
1088 lock->l_resource->lr_lvb_inode,
1089 lli->lli_inode_magic);
1093 unlock_res_and_lock(lock);
1097 void ll_clear_inode(struct inode *inode)
1099 struct ll_inode_info *lli = ll_i2info(inode);
1100 struct ll_sb_info *sbi = ll_i2sbi(inode);
1102 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1103 inode->i_generation, inode);
1105 if (S_ISDIR(inode->i_mode)) {
1106 /* these should have been cleared in ll_file_release */
1107 LASSERT(lli->lli_opendir_key == NULL);
1108 LASSERT(lli->lli_sai == NULL);
1109 LASSERT(lli->lli_opendir_pid == 0);
1112 spin_lock(&lli->lli_lock);
1113 ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1114 spin_unlock(&lli->lli_lock);
1115 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1117 LASSERT(!lli->lli_open_fd_write_count);
1118 LASSERT(!lli->lli_open_fd_read_count);
1119 LASSERT(!lli->lli_open_fd_exec_count);
1121 if (lli->lli_mds_write_och)
1122 ll_md_real_close(inode, FMODE_WRITE);
1123 if (lli->lli_mds_exec_och)
1124 ll_md_real_close(inode, FMODE_EXEC);
1125 if (lli->lli_mds_read_och)
1126 ll_md_real_close(inode, FMODE_READ);
1128 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1129 OBD_FREE(lli->lli_symlink_name,
1130 strlen(lli->lli_symlink_name) + 1);
1131 lli->lli_symlink_name = NULL;
1134 ll_xattr_cache_destroy(inode);
1136 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1137 LASSERT(lli->lli_posix_acl == NULL);
1138 if (lli->lli_remote_perms) {
1139 free_rmtperm_hash(lli->lli_remote_perms);
1140 lli->lli_remote_perms = NULL;
1143 #ifdef CONFIG_FS_POSIX_ACL
1144 else if (lli->lli_posix_acl) {
1145 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1146 LASSERT(lli->lli_remote_perms == NULL);
1147 posix_acl_release(lli->lli_posix_acl);
1148 lli->lli_posix_acl = NULL;
1151 lli->lli_inode_magic = LLI_INODE_DEAD;
1153 ll_clear_inode_capas(inode);
1154 if (!S_ISDIR(inode->i_mode))
1155 LASSERT(list_empty(&lli->lli_agl_list));
1158 * XXX This has to be done before lsm is freed below, because
1159 * cl_object still uses inode lsm.
1161 cl_inode_fini(inode);
1162 lli->lli_has_smd = false;
1165 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data,
1166 struct md_open_data **mod)
1168 struct lustre_md md;
1169 struct inode *inode = d_inode(dentry);
1170 struct ll_sb_info *sbi = ll_i2sbi(inode);
1171 struct ptlrpc_request *request = NULL;
1174 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1175 LUSTRE_OPC_ANY, NULL);
1176 if (IS_ERR(op_data))
1177 return PTR_ERR(op_data);
1179 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1182 ptlrpc_req_finished(request);
1183 if (rc == -ENOENT) {
1185 /* Unlinked special device node? Or just a race?
1186 * Pretend we done everything. */
1187 if (!S_ISREG(inode->i_mode) &&
1188 !S_ISDIR(inode->i_mode)) {
1189 ia_valid = op_data->op_attr.ia_valid;
1190 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1191 rc = simple_setattr(dentry, &op_data->op_attr);
1192 op_data->op_attr.ia_valid = ia_valid;
1194 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1195 CERROR("md_setattr fails: rc = %d\n", rc);
1200 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1201 sbi->ll_md_exp, &md);
1203 ptlrpc_req_finished(request);
1207 ia_valid = op_data->op_attr.ia_valid;
1208 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1209 * cache is not cleared yet. */
1210 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1211 rc = simple_setattr(dentry, &op_data->op_attr);
1212 op_data->op_attr.ia_valid = ia_valid;
1214 /* Extract epoch data if obtained. */
1215 op_data->op_handle = md.body->handle;
1216 op_data->op_ioepoch = md.body->ioepoch;
1218 ll_update_inode(inode, &md);
1219 ptlrpc_req_finished(request);
1224 /* Close IO epoch and send Size-on-MDS attribute update. */
1225 static int ll_setattr_done_writing(struct inode *inode,
1226 struct md_op_data *op_data,
1227 struct md_open_data *mod)
1229 struct ll_inode_info *lli = ll_i2info(inode);
1232 LASSERT(op_data != NULL);
1233 if (!S_ISREG(inode->i_mode))
1236 CDEBUG(D_INODE, "Epoch %llu closed on "DFID" for truncate\n",
1237 op_data->op_ioepoch, PFID(&lli->lli_fid));
1239 op_data->op_flags = MF_EPOCH_CLOSE;
1240 ll_done_writing_attr(inode, op_data);
1241 ll_pack_inode2opdata(inode, op_data, NULL);
1243 rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1244 if (rc == -EAGAIN) {
1245 /* MDS has instructed us to obtain Size-on-MDS attribute
1246 * from OSTs and send setattr to back to MDS. */
1247 rc = ll_som_update(inode, op_data);
1249 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1255 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1257 struct obd_capa *capa;
1260 if (attr->ia_valid & ATTR_SIZE)
1261 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1263 capa = ll_mdscapa_get(inode);
1265 rc = cl_setattr_ost(inode, attr, capa);
1267 if (attr->ia_valid & ATTR_SIZE)
1268 ll_truncate_free_capa(capa);
1276 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1277 * object(s) determine the file size and mtime. Otherwise, the MDS will
1278 * keep these values until such a time that objects are allocated for it.
1279 * We do the MDS operations first, as it is checking permissions for us.
1280 * We don't to the MDS RPC if there is nothing that we want to store there,
1281 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1282 * going to do an RPC anyways.
1284 * If we are doing a truncate, we will send the mtime and ctime updates
1285 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1286 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1289 * In case of HSMimport, we only set attr on MDS.
1291 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1293 struct inode *inode = d_inode(dentry);
1294 struct ll_inode_info *lli = ll_i2info(inode);
1295 struct md_op_data *op_data = NULL;
1296 struct md_open_data *mod = NULL;
1297 bool file_is_released = false;
1298 int rc = 0, rc1 = 0;
1301 "%s: setattr inode %p/fid:"DFID
1302 " from %llu to %llu, valid %x, hsm_import %d\n",
1303 ll_get_fsname(inode->i_sb, NULL, 0), inode,
1304 PFID(&lli->lli_fid), i_size_read(inode), attr->ia_size,
1305 attr->ia_valid, hsm_import);
1307 if (attr->ia_valid & ATTR_SIZE) {
1308 /* Check new size against VFS/VM file size limit and rlimit */
1309 rc = inode_newsize_ok(inode, attr->ia_size);
1313 /* The maximum Lustre file size is variable, based on the
1314 * OST maximum object size and number of stripes. This
1315 * needs another check in addition to the VFS check above. */
1316 if (attr->ia_size > ll_file_maxbytes(inode)) {
1317 CDEBUG(D_INODE, "file "DFID" too large %llu > %llu\n",
1318 PFID(&lli->lli_fid), attr->ia_size,
1319 ll_file_maxbytes(inode));
1323 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1326 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1327 if (attr->ia_valid & TIMES_SET_FLAGS) {
1328 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1329 !capable(CFS_CAP_FOWNER))
1333 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1334 if (attr->ia_valid & ATTR_CTIME) {
1335 attr->ia_ctime = CURRENT_TIME;
1336 attr->ia_valid |= ATTR_CTIME_SET;
1338 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1339 (attr->ia_valid & ATTR_ATIME)) {
1340 attr->ia_atime = CURRENT_TIME;
1341 attr->ia_valid |= ATTR_ATIME_SET;
1343 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1344 (attr->ia_valid & ATTR_MTIME)) {
1345 attr->ia_mtime = CURRENT_TIME;
1346 attr->ia_valid |= ATTR_MTIME_SET;
1349 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1350 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1351 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1354 /* If we are changing file size, file content is modified, flag it. */
1355 if (attr->ia_valid & ATTR_SIZE) {
1356 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1357 spin_lock(&lli->lli_lock);
1358 lli->lli_flags |= LLIF_DATA_MODIFIED;
1359 spin_unlock(&lli->lli_lock);
1362 /* We always do an MDS RPC, even if we're only changing the size;
1363 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1365 op_data = kzalloc(sizeof(*op_data), GFP_NOFS);
1369 if (!S_ISDIR(inode->i_mode)) {
1370 if (attr->ia_valid & ATTR_SIZE)
1371 inode_dio_write_done(inode);
1372 mutex_unlock(&inode->i_mutex);
1375 memcpy(&op_data->op_attr, attr, sizeof(*attr));
1377 /* Open epoch for truncate. */
1378 if (exp_connect_som(ll_i2mdexp(inode)) &&
1379 (attr->ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1380 op_data->op_flags = MF_EPOCH_OPEN;
1382 /* truncate on a released file must failed with -ENODATA,
1383 * so size must not be set on MDS for released file
1384 * but other attributes must be set
1386 if (S_ISREG(inode->i_mode)) {
1387 struct lov_stripe_md *lsm;
1390 ll_layout_refresh(inode, &gen);
1391 lsm = ccc_inode_lsm_get(inode);
1392 if (lsm && lsm->lsm_pattern & LOV_PATTERN_F_RELEASED)
1393 file_is_released = true;
1394 ccc_inode_lsm_put(inode, lsm);
1397 /* if not in HSM import mode, clear size attr for released file
1398 * we clear the attribute send to MDT in op_data, not the original
1399 * received from caller in attr which is used later to
1400 * decide return code */
1401 if (file_is_released && (attr->ia_valid & ATTR_SIZE) && !hsm_import)
1402 op_data->op_attr.ia_valid &= ~ATTR_SIZE;
1404 rc = ll_md_setattr(dentry, op_data, &mod);
1408 /* truncate failed (only when non HSM import), others succeed */
1409 if (file_is_released) {
1410 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1417 /* RPC to MDT is sent, cancel data modification flag */
1418 if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) {
1419 spin_lock(&lli->lli_lock);
1420 lli->lli_flags &= ~LLIF_DATA_MODIFIED;
1421 spin_unlock(&lli->lli_lock);
1424 ll_ioepoch_open(lli, op_data->op_ioepoch);
1425 if (!S_ISREG(inode->i_mode)) {
1430 if (attr->ia_valid & (ATTR_SIZE |
1431 ATTR_ATIME | ATTR_ATIME_SET |
1432 ATTR_MTIME | ATTR_MTIME_SET)) {
1433 /* For truncate and utimes sending attributes to OSTs, setting
1434 * mtime/atime to the past will be performed under PW [0:EOF]
1435 * extent lock (new_size:EOF for truncate). It may seem
1436 * excessive to send mtime/atime updates to OSTs when not
1437 * setting times to past, but it is necessary due to possible
1438 * time de-synchronization between MDT inode and OST objects */
1439 if (attr->ia_valid & ATTR_SIZE)
1440 down_write(&lli->lli_trunc_sem);
1441 rc = ll_setattr_ost(inode, attr);
1442 if (attr->ia_valid & ATTR_SIZE)
1443 up_write(&lli->lli_trunc_sem);
1447 if (op_data->op_ioepoch) {
1448 rc1 = ll_setattr_done_writing(inode, op_data, mod);
1452 ll_finish_md_op_data(op_data);
1454 if (!S_ISDIR(inode->i_mode)) {
1455 mutex_lock(&inode->i_mutex);
1456 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1457 inode_dio_wait(inode);
1460 ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1461 LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1466 int ll_setattr(struct dentry *de, struct iattr *attr)
1468 int mode = d_inode(de)->i_mode;
1470 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1471 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1472 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1474 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1475 (ATTR_SIZE|ATTR_MODE)) &&
1476 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1477 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1478 !(attr->ia_mode & S_ISGID))))
1479 attr->ia_valid |= ATTR_FORCE;
1481 if ((attr->ia_valid & ATTR_MODE) &&
1483 !(attr->ia_mode & S_ISUID) &&
1484 !(attr->ia_valid & ATTR_KILL_SUID))
1485 attr->ia_valid |= ATTR_KILL_SUID;
1487 if ((attr->ia_valid & ATTR_MODE) &&
1488 ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1489 !(attr->ia_mode & S_ISGID) &&
1490 !(attr->ia_valid & ATTR_KILL_SGID))
1491 attr->ia_valid |= ATTR_KILL_SGID;
1493 return ll_setattr_raw(de, attr, false);
1496 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1497 __u64 max_age, __u32 flags)
1499 struct ll_sb_info *sbi = ll_s2sbi(sb);
1500 struct obd_statfs obd_osfs;
1503 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1505 CERROR("md_statfs fails: rc = %d\n", rc);
1509 osfs->os_type = sb->s_magic;
1511 CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1512 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,
1515 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1516 flags |= OBD_STATFS_NODELAY;
1518 rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1520 CERROR("obd_statfs fails: rc = %d\n", rc);
1524 CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1525 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1528 osfs->os_bsize = obd_osfs.os_bsize;
1529 osfs->os_blocks = obd_osfs.os_blocks;
1530 osfs->os_bfree = obd_osfs.os_bfree;
1531 osfs->os_bavail = obd_osfs.os_bavail;
1533 /* If we don't have as many objects free on the OST as inodes
1534 * on the MDS, we reduce the total number of inodes to
1535 * compensate, so that the "inodes in use" number is correct.
1537 if (obd_osfs.os_ffree < osfs->os_ffree) {
1538 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1540 osfs->os_ffree = obd_osfs.os_ffree;
1545 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1547 struct super_block *sb = de->d_sb;
1548 struct obd_statfs osfs;
1551 CDEBUG(D_VFSTRACE, "VFS Op: at %llu jiffies\n", get_jiffies_64());
1552 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1554 /* Some amount of caching on the client is allowed */
1555 rc = ll_statfs_internal(sb, &osfs,
1556 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1561 statfs_unpack(sfs, &osfs);
1563 /* We need to downshift for all 32-bit kernels, because we can't
1564 * tell if the kernel is being called via sys_statfs64() or not.
1565 * Stop before overflowing f_bsize - in which case it is better
1566 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1567 if (sizeof(long) < 8) {
1568 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1571 osfs.os_blocks >>= 1;
1572 osfs.os_bfree >>= 1;
1573 osfs.os_bavail >>= 1;
1577 sfs->f_blocks = osfs.os_blocks;
1578 sfs->f_bfree = osfs.os_bfree;
1579 sfs->f_bavail = osfs.os_bavail;
1580 sfs->f_fsid = ll_s2sbi(sb)->ll_fsid;
1584 void ll_inode_size_lock(struct inode *inode)
1586 struct ll_inode_info *lli;
1588 LASSERT(!S_ISDIR(inode->i_mode));
1590 lli = ll_i2info(inode);
1591 mutex_lock(&lli->lli_size_mutex);
1594 void ll_inode_size_unlock(struct inode *inode)
1596 struct ll_inode_info *lli;
1598 lli = ll_i2info(inode);
1599 mutex_unlock(&lli->lli_size_mutex);
1602 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1604 struct ll_inode_info *lli = ll_i2info(inode);
1605 struct mdt_body *body = md->body;
1606 struct lov_stripe_md *lsm = md->lsm;
1607 struct ll_sb_info *sbi = ll_i2sbi(inode);
1609 LASSERT((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1611 if (!lli->lli_has_smd &&
1612 !(sbi->ll_flags & LL_SBI_LAYOUT_LOCK))
1613 cl_file_inode_init(inode, md);
1615 lli->lli_maxbytes = lsm->lsm_maxbytes;
1616 if (lli->lli_maxbytes > MAX_LFS_FILESIZE)
1617 lli->lli_maxbytes = MAX_LFS_FILESIZE;
1620 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1621 if (body->valid & OBD_MD_FLRMTPERM)
1622 ll_update_remote_perm(inode, md->remote_perm);
1624 #ifdef CONFIG_FS_POSIX_ACL
1625 else if (body->valid & OBD_MD_FLACL) {
1626 spin_lock(&lli->lli_lock);
1627 if (lli->lli_posix_acl)
1628 posix_acl_release(lli->lli_posix_acl);
1629 lli->lli_posix_acl = md->posix_acl;
1630 spin_unlock(&lli->lli_lock);
1633 inode->i_ino = cl_fid_build_ino(&body->fid1,
1634 sbi->ll_flags & LL_SBI_32BIT_API);
1635 inode->i_generation = cl_fid_build_gen(&body->fid1);
1637 if (body->valid & OBD_MD_FLATIME) {
1638 if (body->atime > LTIME_S(inode->i_atime))
1639 LTIME_S(inode->i_atime) = body->atime;
1640 lli->lli_lvb.lvb_atime = body->atime;
1642 if (body->valid & OBD_MD_FLMTIME) {
1643 if (body->mtime > LTIME_S(inode->i_mtime)) {
1644 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to %llu\n",
1645 inode->i_ino, LTIME_S(inode->i_mtime),
1647 LTIME_S(inode->i_mtime) = body->mtime;
1649 lli->lli_lvb.lvb_mtime = body->mtime;
1651 if (body->valid & OBD_MD_FLCTIME) {
1652 if (body->ctime > LTIME_S(inode->i_ctime))
1653 LTIME_S(inode->i_ctime) = body->ctime;
1654 lli->lli_lvb.lvb_ctime = body->ctime;
1656 if (body->valid & OBD_MD_FLMODE)
1657 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1658 if (body->valid & OBD_MD_FLTYPE)
1659 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1660 LASSERT(inode->i_mode != 0);
1661 if (S_ISREG(inode->i_mode))
1662 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1663 LL_MAX_BLKSIZE_BITS);
1665 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1666 if (body->valid & OBD_MD_FLUID)
1667 inode->i_uid = make_kuid(&init_user_ns, body->uid);
1668 if (body->valid & OBD_MD_FLGID)
1669 inode->i_gid = make_kgid(&init_user_ns, body->gid);
1670 if (body->valid & OBD_MD_FLFLAGS)
1671 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1672 if (body->valid & OBD_MD_FLNLINK)
1673 set_nlink(inode, body->nlink);
1674 if (body->valid & OBD_MD_FLRDEV)
1675 inode->i_rdev = old_decode_dev(body->rdev);
1677 if (body->valid & OBD_MD_FLID) {
1678 /* FID shouldn't be changed! */
1679 if (fid_is_sane(&lli->lli_fid)) {
1680 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1681 "Trying to change FID "DFID
1682 " to the "DFID", inode %lu/%u(%p)\n",
1683 PFID(&lli->lli_fid), PFID(&body->fid1),
1684 inode->i_ino, inode->i_generation, inode);
1686 lli->lli_fid = body->fid1;
1689 LASSERT(fid_seq(&lli->lli_fid) != 0);
1691 if (body->valid & OBD_MD_FLSIZE) {
1692 if (exp_connect_som(ll_i2mdexp(inode)) &&
1693 S_ISREG(inode->i_mode)) {
1694 struct lustre_handle lockh;
1697 /* As it is possible a blocking ast has been processed
1698 * by this time, we need to check there is an UPDATE
1699 * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1701 mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1702 &lockh, LDLM_FL_CBPENDING,
1706 if (lli->lli_flags & (LLIF_DONE_WRITING |
1707 LLIF_EPOCH_PENDING |
1709 CERROR("ino %lu flags %u still has size authority! do not trust the size got from MDS\n",
1710 inode->i_ino, lli->lli_flags);
1712 /* Use old size assignment to avoid
1713 * deadlock bz14138 & bz14326 */
1714 i_size_write(inode, body->size);
1715 spin_lock(&lli->lli_lock);
1716 lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1717 spin_unlock(&lli->lli_lock);
1719 ldlm_lock_decref(&lockh, mode);
1722 /* Use old size assignment to avoid
1723 * deadlock bz14138 & bz14326 */
1724 i_size_write(inode, body->size);
1726 CDEBUG(D_VFSTRACE, "inode=%lu, updating i_size %llu\n",
1727 inode->i_ino, (unsigned long long)body->size);
1730 if (body->valid & OBD_MD_FLBLOCKS)
1731 inode->i_blocks = body->blocks;
1734 if (body->valid & OBD_MD_FLMDSCAPA) {
1735 LASSERT(md->mds_capa);
1736 ll_add_capa(inode, md->mds_capa);
1738 if (body->valid & OBD_MD_FLOSSCAPA) {
1739 LASSERT(md->oss_capa);
1740 ll_add_capa(inode, md->oss_capa);
1743 if (body->valid & OBD_MD_TSTATE) {
1744 if (body->t_state & MS_RESTORE)
1745 lli->lli_flags |= LLIF_FILE_RESTORING;
1749 void ll_read_inode2(struct inode *inode, void *opaque)
1751 struct lustre_md *md = opaque;
1752 struct ll_inode_info *lli = ll_i2info(inode);
1754 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1755 PFID(&lli->lli_fid), inode);
1757 LASSERT(!lli->lli_has_smd);
1759 /* Core attributes from the MDS first. This is a new inode, and
1760 * the VFS doesn't zero times in the core inode so we have to do
1761 * it ourselves. They will be overwritten by either MDS or OST
1762 * attributes - we just need to make sure they aren't newer. */
1763 LTIME_S(inode->i_mtime) = 0;
1764 LTIME_S(inode->i_atime) = 0;
1765 LTIME_S(inode->i_ctime) = 0;
1767 ll_update_inode(inode, md);
1769 /* OIDEBUG(inode); */
1771 if (S_ISREG(inode->i_mode)) {
1772 struct ll_sb_info *sbi = ll_i2sbi(inode);
1774 inode->i_op = &ll_file_inode_operations;
1775 inode->i_fop = sbi->ll_fop;
1776 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1777 } else if (S_ISDIR(inode->i_mode)) {
1778 inode->i_op = &ll_dir_inode_operations;
1779 inode->i_fop = &ll_dir_operations;
1780 } else if (S_ISLNK(inode->i_mode)) {
1781 inode->i_op = &ll_fast_symlink_inode_operations;
1783 inode->i_op = &ll_special_inode_operations;
1785 init_special_inode(inode, inode->i_mode,
1790 void ll_delete_inode(struct inode *inode)
1792 struct cl_inode_info *lli = cl_i2info(inode);
1794 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
1795 /* discard all dirty pages before truncating them, required by
1796 * osc_extent implementation at LU-1030. */
1797 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF,
1798 CL_FSYNC_DISCARD, 1);
1800 truncate_inode_pages_final(&inode->i_data);
1802 /* Workaround for LU-118 */
1803 if (inode->i_data.nrpages) {
1804 spin_lock_irq(&inode->i_data.tree_lock);
1805 spin_unlock_irq(&inode->i_data.tree_lock);
1806 LASSERTF(inode->i_data.nrpages == 0,
1807 "inode=%lu/%u(%p) nrpages=%lu, see http://jira.whamcloud.com/browse/LU-118\n",
1808 inode->i_ino, inode->i_generation, inode,
1809 inode->i_data.nrpages);
1811 /* Workaround end */
1813 ll_clear_inode(inode);
1817 int ll_iocontrol(struct inode *inode, struct file *file,
1818 unsigned int cmd, unsigned long arg)
1820 struct ll_sb_info *sbi = ll_i2sbi(inode);
1821 struct ptlrpc_request *req = NULL;
1825 case FSFILT_IOC_GETFLAGS: {
1826 struct mdt_body *body;
1827 struct md_op_data *op_data;
1829 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
1830 0, 0, LUSTRE_OPC_ANY,
1832 if (IS_ERR(op_data))
1833 return PTR_ERR(op_data);
1835 op_data->op_valid = OBD_MD_FLFLAGS;
1836 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
1837 ll_finish_md_op_data(op_data);
1839 CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1843 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1845 flags = body->flags;
1847 ptlrpc_req_finished(req);
1849 return put_user(flags, (int *)arg);
1851 case FSFILT_IOC_SETFLAGS: {
1852 struct lov_stripe_md *lsm;
1853 struct obd_info oinfo = { { { 0 } } };
1854 struct md_op_data *op_data;
1856 if (get_user(flags, (int *)arg))
1859 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1860 LUSTRE_OPC_ANY, NULL);
1861 if (IS_ERR(op_data))
1862 return PTR_ERR(op_data);
1864 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1865 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1866 rc = md_setattr(sbi->ll_md_exp, op_data,
1867 NULL, 0, NULL, 0, &req, NULL);
1868 ll_finish_md_op_data(op_data);
1869 ptlrpc_req_finished(req);
1873 inode->i_flags = ll_ext_to_inode_flags(flags);
1875 lsm = ccc_inode_lsm_get(inode);
1876 if (!lsm_has_objects(lsm)) {
1877 ccc_inode_lsm_put(inode, lsm);
1881 OBDO_ALLOC(oinfo.oi_oa);
1883 ccc_inode_lsm_put(inode, lsm);
1887 oinfo.oi_oa->o_oi = lsm->lsm_oi;
1888 oinfo.oi_oa->o_flags = flags;
1889 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1891 oinfo.oi_capa = ll_mdscapa_get(inode);
1892 obdo_set_parent_fid(oinfo.oi_oa, &ll_i2info(inode)->lli_fid);
1893 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1894 capa_put(oinfo.oi_capa);
1895 OBDO_FREE(oinfo.oi_oa);
1896 ccc_inode_lsm_put(inode, lsm);
1898 if (rc && rc != -EPERM && rc != -EACCES)
1899 CERROR("osc_setattr_async fails: rc = %d\n", rc);
1910 int ll_flush_ctx(struct inode *inode)
1912 struct ll_sb_info *sbi = ll_i2sbi(inode);
1914 CDEBUG(D_SEC, "flush context for user %d\n",
1915 from_kuid(&init_user_ns, current_uid()));
1917 obd_set_info_async(NULL, sbi->ll_md_exp,
1918 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1920 obd_set_info_async(NULL, sbi->ll_dt_exp,
1921 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1926 /* umount -f client means force down, don't save state */
1927 void ll_umount_begin(struct super_block *sb)
1929 struct ll_sb_info *sbi = ll_s2sbi(sb);
1930 struct obd_device *obd;
1931 struct obd_ioctl_data *ioc_data;
1933 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1934 sb->s_count, atomic_read(&sb->s_active));
1936 obd = class_exp2obd(sbi->ll_md_exp);
1938 CERROR("Invalid MDC connection handle %#llx\n",
1939 sbi->ll_md_exp->exp_handle.h_cookie);
1944 obd = class_exp2obd(sbi->ll_dt_exp);
1946 CERROR("Invalid LOV connection handle %#llx\n",
1947 sbi->ll_dt_exp->exp_handle.h_cookie);
1952 ioc_data = kzalloc(sizeof(*ioc_data), GFP_NOFS);
1954 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
1955 sizeof(*ioc_data), ioc_data, NULL);
1957 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
1958 sizeof(*ioc_data), ioc_data, NULL);
1960 OBD_FREE_PTR(ioc_data);
1963 /* Really, we'd like to wait until there are no requests outstanding,
1964 * and then continue. For now, we just invalidate the requests,
1965 * schedule() and sleep one second if needed, and hope.
1970 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
1972 struct ll_sb_info *sbi = ll_s2sbi(sb);
1973 char *profilenm = get_profile_name(sb);
1977 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
1978 read_only = *flags & MS_RDONLY;
1979 err = obd_set_info_async(NULL, sbi->ll_md_exp,
1980 sizeof(KEY_READ_ONLY),
1981 KEY_READ_ONLY, sizeof(read_only),
1984 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
1985 profilenm, read_only ?
1986 "read-only" : "read-write", err);
1991 sb->s_flags |= MS_RDONLY;
1993 sb->s_flags &= ~MS_RDONLY;
1995 if (sbi->ll_flags & LL_SBI_VERBOSE)
1996 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
1997 read_only ? "read-only" : "read-write");
2002 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2003 struct super_block *sb, struct lookup_intent *it)
2005 struct ll_sb_info *sbi = NULL;
2006 struct lustre_md md;
2009 LASSERT(*inode || sb);
2010 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2011 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2012 sbi->ll_md_exp, &md);
2017 ll_update_inode(*inode, &md);
2019 LASSERT(sb != NULL);
2022 * At this point server returns to client's same fid as client
2023 * generated for creating. So using ->fid1 is okay here.
2025 LASSERT(fid_is_sane(&md.body->fid1));
2027 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1,
2028 sbi->ll_flags & LL_SBI_32BIT_API),
2030 if (*inode == NULL || IS_ERR(*inode)) {
2031 #ifdef CONFIG_FS_POSIX_ACL
2033 posix_acl_release(md.posix_acl);
2034 md.posix_acl = NULL;
2037 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2039 CERROR("new_inode -fatal: rc %d\n", rc);
2044 /* Handling piggyback layout lock.
2045 * Layout lock can be piggybacked by getattr and open request.
2046 * The lsm can be applied to inode only if it comes with a layout lock
2047 * otherwise correct layout may be overwritten, for example:
2048 * 1. proc1: mdt returns a lsm but not granting layout
2049 * 2. layout was changed by another client
2050 * 3. proc2: refresh layout and layout lock granted
2051 * 4. proc1: to apply a stale layout */
2052 if (it != NULL && it->d.lustre.it_lock_mode != 0) {
2053 struct lustre_handle lockh;
2054 struct ldlm_lock *lock;
2056 lockh.cookie = it->d.lustre.it_lock_handle;
2057 lock = ldlm_handle2lock(&lockh);
2058 LASSERT(lock != NULL);
2059 if (ldlm_has_layout(lock)) {
2060 struct cl_object_conf conf;
2062 memset(&conf, 0, sizeof(conf));
2063 conf.coc_opc = OBJECT_CONF_SET;
2064 conf.coc_inode = *inode;
2065 conf.coc_lock = lock;
2066 conf.u.coc_md = &md;
2067 (void)ll_layout_conf(*inode, &conf);
2069 LDLM_LOCK_PUT(lock);
2074 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2075 md_free_lustre_md(sbi->ll_md_exp, &md);
2079 int ll_obd_statfs(struct inode *inode, void *arg)
2081 struct ll_sb_info *sbi = NULL;
2082 struct obd_export *exp;
2084 struct obd_ioctl_data *data = NULL;
2094 sbi = ll_i2sbi(inode);
2100 rc = obd_ioctl_getdata(&buf, &len, arg);
2105 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2106 !data->ioc_pbuf1 || !data->ioc_pbuf2) {
2111 if (data->ioc_inllen1 != sizeof(__u32) ||
2112 data->ioc_inllen2 != sizeof(__u32) ||
2113 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2114 data->ioc_plen2 != sizeof(struct obd_uuid)) {
2119 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2120 if (type & LL_STATFS_LMV)
2121 exp = sbi->ll_md_exp;
2122 else if (type & LL_STATFS_LOV)
2123 exp = sbi->ll_dt_exp;
2129 flags = (type & LL_STATFS_NODELAY) ? OBD_STATFS_NODELAY : 0;
2130 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, &flags);
2135 obd_ioctl_freedata(buf, len);
2139 int ll_process_config(struct lustre_cfg *lcfg)
2143 struct lprocfs_static_vars lvars;
2147 lprocfs_llite_init_vars(&lvars);
2149 /* The instance name contains the sb: lustre-client-aacfe000 */
2150 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2151 if (!ptr || !*(++ptr))
2153 rc = kstrtoul(ptr, 16, &x);
2157 /* This better be a real Lustre superblock! */
2158 LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2160 /* Note we have not called client_common_fill_super yet, so
2161 proc fns must be able to handle that! */
2162 rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2169 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2170 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2171 struct inode *i1, struct inode *i2,
2172 const char *name, int namelen,
2173 int mode, __u32 opc, void *data)
2175 LASSERT(i1 != NULL);
2177 if (namelen > ll_i2sbi(i1)->ll_namelen)
2178 return ERR_PTR(-ENAMETOOLONG);
2180 if (op_data == NULL)
2181 op_data = kzalloc(sizeof(*op_data), GFP_NOFS);
2183 if (op_data == NULL)
2184 return ERR_PTR(-ENOMEM);
2186 ll_i2gids(op_data->op_suppgids, i1, i2);
2187 op_data->op_fid1 = *ll_inode2fid(i1);
2188 op_data->op_capa1 = ll_mdscapa_get(i1);
2191 op_data->op_fid2 = *ll_inode2fid(i2);
2192 op_data->op_capa2 = ll_mdscapa_get(i2);
2194 fid_zero(&op_data->op_fid2);
2195 op_data->op_capa2 = NULL;
2198 op_data->op_name = name;
2199 op_data->op_namelen = namelen;
2200 op_data->op_mode = mode;
2201 op_data->op_mod_time = get_seconds();
2202 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2203 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2204 op_data->op_cap = cfs_curproc_cap_pack();
2205 op_data->op_bias = 0;
2206 op_data->op_cli_flags = 0;
2207 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2208 filename_is_volatile(name, namelen, NULL))
2209 op_data->op_bias |= MDS_CREATE_VOLATILE;
2210 op_data->op_opc = opc;
2211 op_data->op_mds = 0;
2212 op_data->op_data = data;
2214 /* If the file is being opened after mknod() (normally due to NFS)
2215 * try to use the default stripe data from parent directory for
2216 * allocating OST objects. Try to pass the parent FID to MDS. */
2217 if (opc == LUSTRE_OPC_CREATE && i1 == i2 && S_ISREG(i2->i_mode) &&
2218 !ll_i2info(i2)->lli_has_smd) {
2219 struct ll_inode_info *lli = ll_i2info(i2);
2221 spin_lock(&lli->lli_lock);
2222 if (likely(!lli->lli_has_smd && !fid_is_zero(&lli->lli_pfid)))
2223 op_data->op_fid1 = lli->lli_pfid;
2224 spin_unlock(&lli->lli_lock);
2225 /** We ignore parent's capability temporary. */
2228 /* When called by ll_setattr_raw, file is i1. */
2229 if (LLIF_DATA_MODIFIED & ll_i2info(i1)->lli_flags)
2230 op_data->op_bias |= MDS_DATA_MODIFIED;
2235 void ll_finish_md_op_data(struct md_op_data *op_data)
2237 capa_put(op_data->op_capa1);
2238 capa_put(op_data->op_capa2);
2239 OBD_FREE_PTR(op_data);
2242 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2244 struct ll_sb_info *sbi;
2246 LASSERT((seq != NULL) && (dentry != NULL));
2247 sbi = ll_s2sbi(dentry->d_sb);
2249 if (sbi->ll_flags & LL_SBI_NOLCK)
2250 seq_puts(seq, ",nolock");
2252 if (sbi->ll_flags & LL_SBI_FLOCK)
2253 seq_puts(seq, ",flock");
2255 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2256 seq_puts(seq, ",localflock");
2258 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2259 seq_puts(seq, ",user_xattr");
2261 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2262 seq_puts(seq, ",lazystatfs");
2264 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2265 seq_puts(seq, ",user_fid2path");
2271 * Get obd name by cmd, and copy out to user space
2273 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2275 struct ll_sb_info *sbi = ll_i2sbi(inode);
2276 struct obd_device *obd;
2278 if (cmd == OBD_IOC_GETDTNAME)
2279 obd = class_exp2obd(sbi->ll_dt_exp);
2280 else if (cmd == OBD_IOC_GETMDNAME)
2281 obd = class_exp2obd(sbi->ll_md_exp);
2288 if (copy_to_user((void *)arg, obd->obd_name,
2289 strlen(obd->obd_name) + 1))
2296 * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2297 * fsname will be returned in this buffer; otherwise, a static buffer will be
2298 * used to store the fsname and returned to caller.
2300 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2302 static char fsname_static[MTI_NAME_MAXLEN];
2303 struct lustre_sb_info *lsi = s2lsi(sb);
2308 /* this means the caller wants to use static buffer
2309 * and it doesn't care about race. Usually this is
2310 * in error reporting path */
2311 buf = fsname_static;
2312 buflen = sizeof(fsname_static);
2315 len = strlen(lsi->lsi_lmd->lmd_profile);
2316 ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2317 if (ptr && (strcmp(ptr, "-client") == 0))
2320 if (unlikely(len >= buflen))
2322 strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2328 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2330 char *buf, *path = NULL;
2331 struct dentry *dentry = NULL;
2332 struct ccc_object *obj = cl_inode2ccc(page->mapping->host);
2334 /* this can be called inside spin lock so use GFP_ATOMIC. */
2335 buf = (char *)__get_free_page(GFP_ATOMIC);
2337 dentry = d_find_alias(page->mapping->host);
2339 path = dentry_path_raw(dentry, buf, PAGE_SIZE);
2343 "%s: dirty page discard: %s/fid: " DFID "/%s may get corrupted (rc %d)\n",
2344 ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2345 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2346 PFID(&obj->cob_header.coh_lu.loh_fid),
2347 (path && !IS_ERR(path)) ? path : "", ioret);
2353 free_page((unsigned long)buf);