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);
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 if (sbi->ll_flags & LL_SBI_FLOCK)
231 sbi->ll_fop = &ll_file_operations_flock;
232 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
233 sbi->ll_fop = &ll_file_operations;
235 sbi->ll_fop = &ll_file_operations_noflock;
238 data->ocd_connect_flags |= OBD_CONNECT_REAL;
239 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
240 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
242 data->ocd_brw_size = MD_MAX_BRW_SIZE;
244 err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid,
247 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",
251 CERROR("cannot connect to %s: rc = %d\n", md, err);
255 sbi->ll_md_exp->exp_connect_data = *data;
257 err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
258 LUSTRE_SEQ_METADATA);
260 CERROR("%s: Can't init metadata layer FID infrastructure, rc = %d\n",
261 sbi->ll_md_exp->exp_obd->obd_name, err);
265 /* For mount, we only need fs info from MDT0, and also in DNE, it
266 * can make sure the client can be mounted as long as MDT0 is
268 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
269 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
270 OBD_STATFS_FOR_MDT0);
274 /* This needs to be after statfs to ensure connect has finished.
275 * Note that "data" does NOT contain the valid connect reply.
276 * If connecting to a 1.8 server there will be no LMV device, so
277 * we can access the MDC export directly and exp_connect_flags will
278 * be non-zero, but if accessing an upgraded 2.1 server it will
279 * have the correct flags filled in.
280 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
281 valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
282 if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
283 valid != CLIENT_CONNECT_MDT_REQD) {
286 buf = kzalloc(PAGE_CACHE_SIZE, GFP_KERNEL);
287 obd_connect_flags2str(buf, PAGE_CACHE_SIZE,
288 valid ^ CLIENT_CONNECT_MDT_REQD, ",");
289 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",
290 sbi->ll_md_exp->exp_obd->obd_name, buf);
296 size = sizeof(*data);
297 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
298 KEY_CONN_DATA, &size, data, NULL);
300 CERROR("%s: Get connect data failed: rc = %d\n",
301 sbi->ll_md_exp->exp_obd->obd_name, err);
305 LASSERT(osfs->os_bsize);
306 sb->s_blocksize = osfs->os_bsize;
307 sb->s_blocksize_bits = log2(osfs->os_bsize);
308 sb->s_magic = LL_SUPER_MAGIC;
309 sb->s_maxbytes = MAX_LFS_FILESIZE;
310 sbi->ll_namelen = osfs->os_namelen;
311 sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
313 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
314 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
315 LCONSOLE_INFO("Disabling user_xattr feature because it is not supported on the server\n");
316 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
319 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
321 sb->s_flags |= MS_POSIXACL;
323 sbi->ll_flags |= LL_SBI_ACL;
325 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
327 sb->s_flags &= ~MS_POSIXACL;
329 sbi->ll_flags &= ~LL_SBI_ACL;
332 if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
333 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
334 sbi->ll_flags |= LL_SBI_RMT_CLIENT;
335 LCONSOLE_INFO("client is set as remote by default.\n");
338 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
339 sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
340 LCONSOLE_INFO("client claims to be remote, but server rejected, forced to be local.\n");
344 if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
345 LCONSOLE_INFO("client enabled MDS capability!\n");
346 sbi->ll_flags |= LL_SBI_MDS_CAPA;
349 if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
350 LCONSOLE_INFO("client enabled OSS capability!\n");
351 sbi->ll_flags |= LL_SBI_OSS_CAPA;
354 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
355 sbi->ll_flags |= LL_SBI_64BIT_HASH;
357 if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
358 sbi->ll_md_brw_size = data->ocd_brw_size;
360 sbi->ll_md_brw_size = PAGE_CACHE_SIZE;
362 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK) {
363 LCONSOLE_INFO("Layout lock feature supported.\n");
364 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
367 if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
368 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
370 "%s: disabling xattr cache due to unknown maximum xattr size.\n",
373 sbi->ll_flags |= LL_SBI_XATTR_CACHE;
374 sbi->ll_xattr_cache_enabled = 1;
378 obd = class_name2obd(dt);
380 CERROR("DT %s: not setup or attached\n", dt);
385 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
386 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
387 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
388 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
389 OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
390 OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
391 OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
392 OBD_CONNECT_MAXBYTES |
393 OBD_CONNECT_EINPROGRESS |
394 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
395 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS;
397 if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
398 data->ocd_connect_flags |= OBD_CONNECT_SOM;
400 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
401 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
402 * disabled by default, because it can still be enabled on the
403 * fly via /proc. As a consequence, we still need to come to an
404 * agreement on the supported algorithms at connect time */
405 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
407 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
408 data->ocd_cksum_types = OBD_CKSUM_ADLER;
410 data->ocd_cksum_types = cksum_types_supported_client();
413 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
414 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
415 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
417 CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d ocd_grant: %d\n",
418 data->ocd_connect_flags,
419 data->ocd_version, data->ocd_grant);
421 obd->obd_upcall.onu_owner = &sbi->ll_lco;
422 obd->obd_upcall.onu_upcall = cl_ocd_update;
424 data->ocd_brw_size = DT_MAX_BRW_SIZE;
426 err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
429 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",
433 CERROR("%s: Cannot connect to %s: rc = %d\n",
434 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
438 sbi->ll_dt_exp->exp_connect_data = *data;
440 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
441 LUSTRE_SEQ_METADATA);
443 CERROR("%s: Can't init data layer FID infrastructure, rc = %d\n",
444 sbi->ll_dt_exp->exp_obd->obd_name, err);
448 mutex_lock(&sbi->ll_lco.lco_lock);
449 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
450 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
451 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
452 mutex_unlock(&sbi->ll_lco.lco_lock);
454 fid_zero(&sbi->ll_root_fid);
455 err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
457 CERROR("cannot mds_connect: rc = %d\n", err);
460 if (!fid_is_sane(&sbi->ll_root_fid)) {
461 CERROR("%s: Invalid root fid "DFID" during mount\n",
462 sbi->ll_md_exp->exp_obd->obd_name,
463 PFID(&sbi->ll_root_fid));
467 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
469 sb->s_op = &lustre_super_operations;
470 #if THREAD_SIZE >= 8192 /*b=17630*/
471 sb->s_export_op = &lustre_export_operations;
475 * XXX: move this to after cbd setup? */
476 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
477 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
478 valid |= OBD_MD_FLRMTPERM;
479 else if (sbi->ll_flags & LL_SBI_ACL)
480 valid |= OBD_MD_FLACL;
482 op_data = kzalloc(sizeof(*op_data), GFP_NOFS);
488 op_data->op_fid1 = sbi->ll_root_fid;
489 op_data->op_mode = 0;
490 op_data->op_capa1 = oc;
491 op_data->op_valid = valid;
493 err = md_getattr(sbi->ll_md_exp, op_data, &request);
498 CERROR("%s: md_getattr failed for root: rc = %d\n",
499 sbi->ll_md_exp->exp_obd->obd_name, err);
503 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
504 sbi->ll_md_exp, &lmd);
506 CERROR("failed to understand root inode md: rc = %d\n", err);
507 ptlrpc_req_finished(request);
511 LASSERT(fid_is_sane(&sbi->ll_root_fid));
512 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
513 sbi->ll_flags & LL_SBI_32BIT_API),
515 md_free_lustre_md(sbi->ll_md_exp, &lmd);
516 ptlrpc_req_finished(request);
518 if (root == NULL || IS_ERR(root)) {
520 obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
521 #ifdef CONFIG_FS_POSIX_ACL
523 posix_acl_release(lmd.posix_acl);
524 lmd.posix_acl = NULL;
527 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
529 CERROR("lustre_lite: bad iget4 for root\n");
533 err = ll_close_thread_start(&sbi->ll_lcq);
535 CERROR("cannot start close thread: rc %d\n", err);
539 #ifdef CONFIG_FS_POSIX_ACL
540 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
541 rct_init(&sbi->ll_rct);
542 et_init(&sbi->ll_et);
546 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
547 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
548 KEY_CHECKSUM, sizeof(checksum), &checksum,
552 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
553 KEY_CACHE_SET, sizeof(sbi->ll_cache),
554 &sbi->ll_cache, NULL);
556 sb->s_root = d_make_root(root);
557 if (sb->s_root == NULL) {
558 CERROR("%s: can't make root dentry\n",
559 ll_get_fsname(sb, NULL, 0));
564 sbi->ll_sdev_orig = sb->s_dev;
566 /* We set sb->s_dev equal on all lustre clients in order to support
567 * NFS export clustering. NFSD requires that the FSID be the same
569 /* s_dev is also used in lt_compare() to compare two fs, but that is
570 * only a node-local comparison. */
571 uuid = obd_get_uuid(sbi->ll_md_exp);
573 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
574 get_uuid2fsid(uuid->uuid, strlen(uuid->uuid), &sbi->ll_fsid);
584 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
586 obd_disconnect(sbi->ll_dt_exp);
587 sbi->ll_dt_exp = NULL;
588 /* Make sure all OScs are gone, since cl_cache is accessing sbi. */
589 obd_zombie_barrier();
591 obd_fid_fini(sbi->ll_md_exp->exp_obd);
593 obd_disconnect(sbi->ll_md_exp);
594 sbi->ll_md_exp = NULL;
598 lprocfs_unregister_mountpoint(sbi);
602 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
606 *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
608 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
609 KEY_MAX_EASIZE, &size, lmmsize, NULL);
611 CERROR("Get max mdsize error rc %d\n", rc);
616 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
621 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
622 KEY_DEFAULT_EASIZE, &size, lmmsize, NULL);
624 CERROR("Get default mdsize error rc %d\n", rc);
629 int ll_get_max_cookiesize(struct ll_sb_info *sbi, int *lmmsize)
634 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_COOKIESIZE),
635 KEY_MAX_COOKIESIZE, &size, lmmsize, NULL);
637 CERROR("Get max cookiesize error rc %d\n", rc);
642 int ll_get_default_cookiesize(struct ll_sb_info *sbi, int *lmmsize)
647 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_COOKIESIZE),
648 KEY_DEFAULT_COOKIESIZE, &size, lmmsize, NULL);
650 CERROR("Get default cookiesize error rc %d\n", rc);
655 static void client_common_put_super(struct super_block *sb)
657 struct ll_sb_info *sbi = ll_s2sbi(sb);
659 #ifdef CONFIG_FS_POSIX_ACL
660 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
661 et_fini(&sbi->ll_et);
662 rct_fini(&sbi->ll_rct);
666 ll_close_thread_shutdown(sbi->ll_lcq);
670 list_del(&sbi->ll_conn_chain);
672 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
673 obd_disconnect(sbi->ll_dt_exp);
674 sbi->ll_dt_exp = NULL;
675 /* wait till all OSCs are gone, since cl_cache is accessing sbi.
677 obd_zombie_barrier();
679 lprocfs_unregister_mountpoint(sbi);
681 obd_fid_fini(sbi->ll_md_exp->exp_obd);
682 obd_disconnect(sbi->ll_md_exp);
683 sbi->ll_md_exp = NULL;
686 void ll_kill_super(struct super_block *sb)
688 struct ll_sb_info *sbi;
691 if (!(sb->s_flags & MS_ACTIVE))
695 /* we need to restore s_dev from changed for clustered NFS before
696 * put_super because new kernels have cached s_dev and change sb->s_dev
697 * in put_super not affected real removing devices */
699 sb->s_dev = sbi->ll_sdev_orig;
700 sbi->ll_umounting = 1;
704 static inline int ll_set_opt(const char *opt, char *data, int fl)
706 if (strncmp(opt, data, strlen(opt)) != 0)
712 /* non-client-specific mount options are parsed in lmd_parse */
713 static int ll_options(char *options, int *flags)
716 char *s1 = options, *s2;
721 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
724 CDEBUG(D_SUPER, "next opt=%s\n", s1);
725 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
730 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
735 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
740 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
745 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
750 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
755 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
760 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
765 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
771 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
776 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
781 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
786 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
791 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
796 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
801 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
806 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
811 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
816 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
821 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
827 s2 = strchr(s1, ',');
835 void ll_lli_init(struct ll_inode_info *lli)
837 lli->lli_inode_magic = LLI_INODE_MAGIC;
839 lli->lli_ioepoch = 0;
840 lli->lli_maxbytes = MAX_LFS_FILESIZE;
841 spin_lock_init(&lli->lli_lock);
842 lli->lli_posix_acl = NULL;
843 lli->lli_remote_perms = NULL;
844 mutex_init(&lli->lli_rmtperm_mutex);
845 /* Do not set lli_fid, it has been initialized already. */
846 fid_zero(&lli->lli_pfid);
847 INIT_LIST_HEAD(&lli->lli_close_list);
848 INIT_LIST_HEAD(&lli->lli_oss_capas);
849 atomic_set(&lli->lli_open_count, 0);
850 lli->lli_mds_capa = NULL;
851 lli->lli_rmtperm_time = 0;
852 lli->lli_pending_och = NULL;
853 lli->lli_mds_read_och = NULL;
854 lli->lli_mds_write_och = NULL;
855 lli->lli_mds_exec_och = NULL;
856 lli->lli_open_fd_read_count = 0;
857 lli->lli_open_fd_write_count = 0;
858 lli->lli_open_fd_exec_count = 0;
859 mutex_init(&lli->lli_och_mutex);
860 spin_lock_init(&lli->lli_agl_lock);
861 lli->lli_has_smd = false;
862 spin_lock_init(&lli->lli_layout_lock);
863 ll_layout_version_set(lli, LL_LAYOUT_GEN_NONE);
864 lli->lli_clob = NULL;
866 init_rwsem(&lli->lli_xattrs_list_rwsem);
867 mutex_init(&lli->lli_xattrs_enq_lock);
869 LASSERT(lli->lli_vfs_inode.i_mode != 0);
870 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
871 mutex_init(&lli->lli_readdir_mutex);
872 lli->lli_opendir_key = NULL;
874 spin_lock_init(&lli->lli_sa_lock);
875 lli->lli_opendir_pid = 0;
877 mutex_init(&lli->lli_size_mutex);
878 lli->lli_symlink_name = NULL;
879 init_rwsem(&lli->lli_trunc_sem);
880 mutex_init(&lli->lli_write_mutex);
881 init_rwsem(&lli->lli_glimpse_sem);
882 lli->lli_glimpse_time = 0;
883 INIT_LIST_HEAD(&lli->lli_agl_list);
884 lli->lli_agl_index = 0;
885 lli->lli_async_rc = 0;
887 mutex_init(&lli->lli_layout_mutex);
890 static inline int ll_bdi_register(struct backing_dev_info *bdi)
892 static atomic_t ll_bdi_num = ATOMIC_INIT(0);
894 bdi->name = "lustre";
895 return bdi_register(bdi, NULL, "lustre-%d",
896 atomic_inc_return(&ll_bdi_num));
899 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
901 struct lustre_profile *lprof = NULL;
902 struct lustre_sb_info *lsi = s2lsi(sb);
903 struct ll_sb_info *sbi;
904 char *dt = NULL, *md = NULL;
905 char *profilenm = get_profile_name(sb);
906 struct config_llog_instance *cfg;
909 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
911 cfg = kzalloc(sizeof(*cfg), GFP_NOFS);
915 try_module_get(THIS_MODULE);
917 /* client additional sb info */
918 lsi->lsi_llsbi = sbi = ll_init_sbi();
920 module_put(THIS_MODULE);
925 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
929 err = bdi_init(&lsi->lsi_bdi);
932 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
933 lsi->lsi_bdi.capabilities = 0;
934 err = ll_bdi_register(&lsi->lsi_bdi);
938 sb->s_bdi = &lsi->lsi_bdi;
939 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
940 sb->s_d_op = &ll_d_ops;
942 /* Generate a string unique to this super, in case some joker tries
943 to mount the same fs at two mount points.
944 Use the address of the super itself.*/
945 cfg->cfg_instance = sb;
946 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
947 cfg->cfg_callback = class_config_llog_handler;
948 /* set up client obds */
949 err = lustre_process_log(sb, profilenm, cfg);
951 CERROR("Unable to process log: %d\n", err);
955 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
956 lprof = class_get_profile(profilenm);
958 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be read from the MGS. Does that filesystem exist?\n",
963 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
964 lprof->lp_md, lprof->lp_dt);
966 dt = kasprintf(GFP_NOFS, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
972 md = kasprintf(GFP_NOFS, "%s-%p", lprof->lp_md, cfg->cfg_instance);
978 /* connections, registrations, sb setup */
979 err = client_common_fill_super(sb, md, dt, mnt);
986 else if (sbi->ll_flags & LL_SBI_VERBOSE)
987 LCONSOLE_WARN("Mounted %s\n", profilenm);
991 } /* ll_fill_super */
993 void ll_put_super(struct super_block *sb)
995 struct config_llog_instance cfg, params_cfg;
996 struct obd_device *obd;
997 struct lustre_sb_info *lsi = s2lsi(sb);
998 struct ll_sb_info *sbi = ll_s2sbi(sb);
999 char *profilenm = get_profile_name(sb);
1000 int next, force = 1;
1002 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1004 ll_print_capa_stat(sbi);
1006 cfg.cfg_instance = sb;
1007 lustre_end_log(sb, profilenm, &cfg);
1009 params_cfg.cfg_instance = sb;
1010 lustre_end_log(sb, PARAMS_FILENAME, ¶ms_cfg);
1012 if (sbi->ll_md_exp) {
1013 obd = class_exp2obd(sbi->ll_md_exp);
1015 force = obd->obd_force;
1018 /* We need to set force before the lov_disconnect in
1019 lustre_common_put_super, since l_d cleans up osc's as well. */
1022 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1024 obd->obd_force = force;
1029 /* Only if client_common_fill_super succeeded */
1030 client_common_put_super(sb);
1034 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)))
1035 class_manual_cleanup(obd);
1037 if (sbi->ll_flags & LL_SBI_VERBOSE)
1038 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1041 class_del_profile(profilenm);
1043 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1044 bdi_destroy(&lsi->lsi_bdi);
1045 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1049 lsi->lsi_llsbi = NULL;
1051 lustre_common_put_super(sb);
1053 module_put(THIS_MODULE);
1054 } /* client_put_super */
1056 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1058 struct inode *inode = NULL;
1060 /* NOTE: we depend on atomic igrab() -bzzz */
1061 lock_res_and_lock(lock);
1062 if (lock->l_resource->lr_lvb_inode) {
1063 struct ll_inode_info *lli;
1065 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1066 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1067 inode = igrab(lock->l_resource->lr_lvb_inode);
1069 inode = lock->l_resource->lr_lvb_inode;
1070 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1071 D_WARNING, lock, "lr_lvb_inode %p is bogus: magic %08x",
1072 lock->l_resource->lr_lvb_inode,
1073 lli->lli_inode_magic);
1077 unlock_res_and_lock(lock);
1081 void ll_clear_inode(struct inode *inode)
1083 struct ll_inode_info *lli = ll_i2info(inode);
1084 struct ll_sb_info *sbi = ll_i2sbi(inode);
1086 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1087 inode->i_generation, inode);
1089 if (S_ISDIR(inode->i_mode)) {
1090 /* these should have been cleared in ll_file_release */
1091 LASSERT(lli->lli_opendir_key == NULL);
1092 LASSERT(lli->lli_sai == NULL);
1093 LASSERT(lli->lli_opendir_pid == 0);
1096 spin_lock(&lli->lli_lock);
1097 ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1098 spin_unlock(&lli->lli_lock);
1099 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1101 LASSERT(!lli->lli_open_fd_write_count);
1102 LASSERT(!lli->lli_open_fd_read_count);
1103 LASSERT(!lli->lli_open_fd_exec_count);
1105 if (lli->lli_mds_write_och)
1106 ll_md_real_close(inode, FMODE_WRITE);
1107 if (lli->lli_mds_exec_och)
1108 ll_md_real_close(inode, FMODE_EXEC);
1109 if (lli->lli_mds_read_och)
1110 ll_md_real_close(inode, FMODE_READ);
1112 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1113 kfree(lli->lli_symlink_name);
1114 lli->lli_symlink_name = NULL;
1117 ll_xattr_cache_destroy(inode);
1119 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1120 LASSERT(lli->lli_posix_acl == NULL);
1121 if (lli->lli_remote_perms) {
1122 free_rmtperm_hash(lli->lli_remote_perms);
1123 lli->lli_remote_perms = NULL;
1126 #ifdef CONFIG_FS_POSIX_ACL
1127 else if (lli->lli_posix_acl) {
1128 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1129 LASSERT(lli->lli_remote_perms == NULL);
1130 posix_acl_release(lli->lli_posix_acl);
1131 lli->lli_posix_acl = NULL;
1134 lli->lli_inode_magic = LLI_INODE_DEAD;
1136 ll_clear_inode_capas(inode);
1137 if (!S_ISDIR(inode->i_mode))
1138 LASSERT(list_empty(&lli->lli_agl_list));
1141 * XXX This has to be done before lsm is freed below, because
1142 * cl_object still uses inode lsm.
1144 cl_inode_fini(inode);
1145 lli->lli_has_smd = false;
1148 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data,
1149 struct md_open_data **mod)
1151 struct lustre_md md;
1152 struct inode *inode = d_inode(dentry);
1153 struct ll_sb_info *sbi = ll_i2sbi(inode);
1154 struct ptlrpc_request *request = NULL;
1157 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1158 LUSTRE_OPC_ANY, NULL);
1159 if (IS_ERR(op_data))
1160 return PTR_ERR(op_data);
1162 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1165 ptlrpc_req_finished(request);
1166 if (rc == -ENOENT) {
1168 /* Unlinked special device node? Or just a race?
1169 * Pretend we done everything. */
1170 if (!S_ISREG(inode->i_mode) &&
1171 !S_ISDIR(inode->i_mode)) {
1172 ia_valid = op_data->op_attr.ia_valid;
1173 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1174 rc = simple_setattr(dentry, &op_data->op_attr);
1175 op_data->op_attr.ia_valid = ia_valid;
1177 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1178 CERROR("md_setattr fails: rc = %d\n", rc);
1183 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1184 sbi->ll_md_exp, &md);
1186 ptlrpc_req_finished(request);
1190 ia_valid = op_data->op_attr.ia_valid;
1191 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1192 * cache is not cleared yet. */
1193 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1194 rc = simple_setattr(dentry, &op_data->op_attr);
1195 op_data->op_attr.ia_valid = ia_valid;
1197 /* Extract epoch data if obtained. */
1198 op_data->op_handle = md.body->handle;
1199 op_data->op_ioepoch = md.body->ioepoch;
1201 ll_update_inode(inode, &md);
1202 ptlrpc_req_finished(request);
1207 /* Close IO epoch and send Size-on-MDS attribute update. */
1208 static int ll_setattr_done_writing(struct inode *inode,
1209 struct md_op_data *op_data,
1210 struct md_open_data *mod)
1212 struct ll_inode_info *lli = ll_i2info(inode);
1215 LASSERT(op_data != NULL);
1216 if (!S_ISREG(inode->i_mode))
1219 CDEBUG(D_INODE, "Epoch %llu closed on "DFID" for truncate\n",
1220 op_data->op_ioepoch, PFID(&lli->lli_fid));
1222 op_data->op_flags = MF_EPOCH_CLOSE;
1223 ll_done_writing_attr(inode, op_data);
1224 ll_pack_inode2opdata(inode, op_data, NULL);
1226 rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1227 if (rc == -EAGAIN) {
1228 /* MDS has instructed us to obtain Size-on-MDS attribute
1229 * from OSTs and send setattr to back to MDS. */
1230 rc = ll_som_update(inode, op_data);
1232 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1238 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1240 struct obd_capa *capa;
1243 if (attr->ia_valid & ATTR_SIZE)
1244 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1246 capa = ll_mdscapa_get(inode);
1248 rc = cl_setattr_ost(inode, attr, capa);
1250 if (attr->ia_valid & ATTR_SIZE)
1251 ll_truncate_free_capa(capa);
1259 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1260 * object(s) determine the file size and mtime. Otherwise, the MDS will
1261 * keep these values until such a time that objects are allocated for it.
1262 * We do the MDS operations first, as it is checking permissions for us.
1263 * We don't to the MDS RPC if there is nothing that we want to store there,
1264 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1265 * going to do an RPC anyways.
1267 * If we are doing a truncate, we will send the mtime and ctime updates
1268 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1269 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1272 * In case of HSMimport, we only set attr on MDS.
1274 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1276 struct inode *inode = d_inode(dentry);
1277 struct ll_inode_info *lli = ll_i2info(inode);
1278 struct md_op_data *op_data = NULL;
1279 struct md_open_data *mod = NULL;
1280 bool file_is_released = false;
1281 int rc = 0, rc1 = 0;
1284 "%s: setattr inode %p/fid:"DFID
1285 " from %llu to %llu, valid %x, hsm_import %d\n",
1286 ll_get_fsname(inode->i_sb, NULL, 0), inode,
1287 PFID(&lli->lli_fid), i_size_read(inode), attr->ia_size,
1288 attr->ia_valid, hsm_import);
1290 if (attr->ia_valid & ATTR_SIZE) {
1291 /* Check new size against VFS/VM file size limit and rlimit */
1292 rc = inode_newsize_ok(inode, attr->ia_size);
1296 /* The maximum Lustre file size is variable, based on the
1297 * OST maximum object size and number of stripes. This
1298 * needs another check in addition to the VFS check above. */
1299 if (attr->ia_size > ll_file_maxbytes(inode)) {
1300 CDEBUG(D_INODE, "file "DFID" too large %llu > %llu\n",
1301 PFID(&lli->lli_fid), attr->ia_size,
1302 ll_file_maxbytes(inode));
1306 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1309 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1310 if (attr->ia_valid & TIMES_SET_FLAGS) {
1311 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1312 !capable(CFS_CAP_FOWNER))
1316 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1317 if (attr->ia_valid & ATTR_CTIME) {
1318 attr->ia_ctime = CURRENT_TIME;
1319 attr->ia_valid |= ATTR_CTIME_SET;
1321 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1322 (attr->ia_valid & ATTR_ATIME)) {
1323 attr->ia_atime = CURRENT_TIME;
1324 attr->ia_valid |= ATTR_ATIME_SET;
1326 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1327 (attr->ia_valid & ATTR_MTIME)) {
1328 attr->ia_mtime = CURRENT_TIME;
1329 attr->ia_valid |= ATTR_MTIME_SET;
1332 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1333 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1334 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1337 /* If we are changing file size, file content is modified, flag it. */
1338 if (attr->ia_valid & ATTR_SIZE) {
1339 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1340 spin_lock(&lli->lli_lock);
1341 lli->lli_flags |= LLIF_DATA_MODIFIED;
1342 spin_unlock(&lli->lli_lock);
1345 /* We always do an MDS RPC, even if we're only changing the size;
1346 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1348 op_data = kzalloc(sizeof(*op_data), GFP_NOFS);
1352 if (!S_ISDIR(inode->i_mode)) {
1353 if (attr->ia_valid & ATTR_SIZE)
1354 inode_dio_write_done(inode);
1355 mutex_unlock(&inode->i_mutex);
1358 memcpy(&op_data->op_attr, attr, sizeof(*attr));
1360 /* Open epoch for truncate. */
1361 if (exp_connect_som(ll_i2mdexp(inode)) &&
1362 (attr->ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1363 op_data->op_flags = MF_EPOCH_OPEN;
1365 /* truncate on a released file must failed with -ENODATA,
1366 * so size must not be set on MDS for released file
1367 * but other attributes must be set
1369 if (S_ISREG(inode->i_mode)) {
1370 struct lov_stripe_md *lsm;
1373 ll_layout_refresh(inode, &gen);
1374 lsm = ccc_inode_lsm_get(inode);
1375 if (lsm && lsm->lsm_pattern & LOV_PATTERN_F_RELEASED)
1376 file_is_released = true;
1377 ccc_inode_lsm_put(inode, lsm);
1380 /* if not in HSM import mode, clear size attr for released file
1381 * we clear the attribute send to MDT in op_data, not the original
1382 * received from caller in attr which is used later to
1383 * decide return code */
1384 if (file_is_released && (attr->ia_valid & ATTR_SIZE) && !hsm_import)
1385 op_data->op_attr.ia_valid &= ~ATTR_SIZE;
1387 rc = ll_md_setattr(dentry, op_data, &mod);
1391 /* truncate failed (only when non HSM import), others succeed */
1392 if (file_is_released) {
1393 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1400 /* RPC to MDT is sent, cancel data modification flag */
1401 if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) {
1402 spin_lock(&lli->lli_lock);
1403 lli->lli_flags &= ~LLIF_DATA_MODIFIED;
1404 spin_unlock(&lli->lli_lock);
1407 ll_ioepoch_open(lli, op_data->op_ioepoch);
1408 if (!S_ISREG(inode->i_mode)) {
1413 if (attr->ia_valid & (ATTR_SIZE |
1414 ATTR_ATIME | ATTR_ATIME_SET |
1415 ATTR_MTIME | ATTR_MTIME_SET)) {
1416 /* For truncate and utimes sending attributes to OSTs, setting
1417 * mtime/atime to the past will be performed under PW [0:EOF]
1418 * extent lock (new_size:EOF for truncate). It may seem
1419 * excessive to send mtime/atime updates to OSTs when not
1420 * setting times to past, but it is necessary due to possible
1421 * time de-synchronization between MDT inode and OST objects */
1422 if (attr->ia_valid & ATTR_SIZE)
1423 down_write(&lli->lli_trunc_sem);
1424 rc = ll_setattr_ost(inode, attr);
1425 if (attr->ia_valid & ATTR_SIZE)
1426 up_write(&lli->lli_trunc_sem);
1430 if (op_data->op_ioepoch) {
1431 rc1 = ll_setattr_done_writing(inode, op_data, mod);
1435 ll_finish_md_op_data(op_data);
1437 if (!S_ISDIR(inode->i_mode)) {
1438 mutex_lock(&inode->i_mutex);
1439 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1440 inode_dio_wait(inode);
1443 ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1444 LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1449 int ll_setattr(struct dentry *de, struct iattr *attr)
1451 int mode = d_inode(de)->i_mode;
1453 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1454 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1455 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1457 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1458 (ATTR_SIZE|ATTR_MODE)) &&
1459 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1460 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1461 !(attr->ia_mode & S_ISGID))))
1462 attr->ia_valid |= ATTR_FORCE;
1464 if ((attr->ia_valid & ATTR_MODE) &&
1466 !(attr->ia_mode & S_ISUID) &&
1467 !(attr->ia_valid & ATTR_KILL_SUID))
1468 attr->ia_valid |= ATTR_KILL_SUID;
1470 if ((attr->ia_valid & ATTR_MODE) &&
1471 ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1472 !(attr->ia_mode & S_ISGID) &&
1473 !(attr->ia_valid & ATTR_KILL_SGID))
1474 attr->ia_valid |= ATTR_KILL_SGID;
1476 return ll_setattr_raw(de, attr, false);
1479 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1480 __u64 max_age, __u32 flags)
1482 struct ll_sb_info *sbi = ll_s2sbi(sb);
1483 struct obd_statfs obd_osfs;
1486 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1488 CERROR("md_statfs fails: rc = %d\n", rc);
1492 osfs->os_type = sb->s_magic;
1494 CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1495 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,
1498 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1499 flags |= OBD_STATFS_NODELAY;
1501 rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1503 CERROR("obd_statfs fails: rc = %d\n", rc);
1507 CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1508 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1511 osfs->os_bsize = obd_osfs.os_bsize;
1512 osfs->os_blocks = obd_osfs.os_blocks;
1513 osfs->os_bfree = obd_osfs.os_bfree;
1514 osfs->os_bavail = obd_osfs.os_bavail;
1516 /* If we don't have as many objects free on the OST as inodes
1517 * on the MDS, we reduce the total number of inodes to
1518 * compensate, so that the "inodes in use" number is correct.
1520 if (obd_osfs.os_ffree < osfs->os_ffree) {
1521 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1523 osfs->os_ffree = obd_osfs.os_ffree;
1528 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1530 struct super_block *sb = de->d_sb;
1531 struct obd_statfs osfs;
1534 CDEBUG(D_VFSTRACE, "VFS Op: at %llu jiffies\n", get_jiffies_64());
1535 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1537 /* Some amount of caching on the client is allowed */
1538 rc = ll_statfs_internal(sb, &osfs,
1539 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1544 statfs_unpack(sfs, &osfs);
1546 /* We need to downshift for all 32-bit kernels, because we can't
1547 * tell if the kernel is being called via sys_statfs64() or not.
1548 * Stop before overflowing f_bsize - in which case it is better
1549 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1550 if (sizeof(long) < 8) {
1551 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1554 osfs.os_blocks >>= 1;
1555 osfs.os_bfree >>= 1;
1556 osfs.os_bavail >>= 1;
1560 sfs->f_blocks = osfs.os_blocks;
1561 sfs->f_bfree = osfs.os_bfree;
1562 sfs->f_bavail = osfs.os_bavail;
1563 sfs->f_fsid = ll_s2sbi(sb)->ll_fsid;
1567 void ll_inode_size_lock(struct inode *inode)
1569 struct ll_inode_info *lli;
1571 LASSERT(!S_ISDIR(inode->i_mode));
1573 lli = ll_i2info(inode);
1574 mutex_lock(&lli->lli_size_mutex);
1577 void ll_inode_size_unlock(struct inode *inode)
1579 struct ll_inode_info *lli;
1581 lli = ll_i2info(inode);
1582 mutex_unlock(&lli->lli_size_mutex);
1585 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1587 struct ll_inode_info *lli = ll_i2info(inode);
1588 struct mdt_body *body = md->body;
1589 struct lov_stripe_md *lsm = md->lsm;
1590 struct ll_sb_info *sbi = ll_i2sbi(inode);
1592 LASSERT((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1594 if (!lli->lli_has_smd &&
1595 !(sbi->ll_flags & LL_SBI_LAYOUT_LOCK))
1596 cl_file_inode_init(inode, md);
1598 lli->lli_maxbytes = lsm->lsm_maxbytes;
1599 if (lli->lli_maxbytes > MAX_LFS_FILESIZE)
1600 lli->lli_maxbytes = MAX_LFS_FILESIZE;
1603 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1604 if (body->valid & OBD_MD_FLRMTPERM)
1605 ll_update_remote_perm(inode, md->remote_perm);
1607 #ifdef CONFIG_FS_POSIX_ACL
1608 else if (body->valid & OBD_MD_FLACL) {
1609 spin_lock(&lli->lli_lock);
1610 if (lli->lli_posix_acl)
1611 posix_acl_release(lli->lli_posix_acl);
1612 lli->lli_posix_acl = md->posix_acl;
1613 spin_unlock(&lli->lli_lock);
1616 inode->i_ino = cl_fid_build_ino(&body->fid1,
1617 sbi->ll_flags & LL_SBI_32BIT_API);
1618 inode->i_generation = cl_fid_build_gen(&body->fid1);
1620 if (body->valid & OBD_MD_FLATIME) {
1621 if (body->atime > LTIME_S(inode->i_atime))
1622 LTIME_S(inode->i_atime) = body->atime;
1623 lli->lli_lvb.lvb_atime = body->atime;
1625 if (body->valid & OBD_MD_FLMTIME) {
1626 if (body->mtime > LTIME_S(inode->i_mtime)) {
1627 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to %llu\n",
1628 inode->i_ino, LTIME_S(inode->i_mtime),
1630 LTIME_S(inode->i_mtime) = body->mtime;
1632 lli->lli_lvb.lvb_mtime = body->mtime;
1634 if (body->valid & OBD_MD_FLCTIME) {
1635 if (body->ctime > LTIME_S(inode->i_ctime))
1636 LTIME_S(inode->i_ctime) = body->ctime;
1637 lli->lli_lvb.lvb_ctime = body->ctime;
1639 if (body->valid & OBD_MD_FLMODE)
1640 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1641 if (body->valid & OBD_MD_FLTYPE)
1642 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1643 LASSERT(inode->i_mode != 0);
1644 if (S_ISREG(inode->i_mode))
1645 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1646 LL_MAX_BLKSIZE_BITS);
1648 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1649 if (body->valid & OBD_MD_FLUID)
1650 inode->i_uid = make_kuid(&init_user_ns, body->uid);
1651 if (body->valid & OBD_MD_FLGID)
1652 inode->i_gid = make_kgid(&init_user_ns, body->gid);
1653 if (body->valid & OBD_MD_FLFLAGS)
1654 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1655 if (body->valid & OBD_MD_FLNLINK)
1656 set_nlink(inode, body->nlink);
1657 if (body->valid & OBD_MD_FLRDEV)
1658 inode->i_rdev = old_decode_dev(body->rdev);
1660 if (body->valid & OBD_MD_FLID) {
1661 /* FID shouldn't be changed! */
1662 if (fid_is_sane(&lli->lli_fid)) {
1663 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1664 "Trying to change FID "DFID
1665 " to the "DFID", inode %lu/%u(%p)\n",
1666 PFID(&lli->lli_fid), PFID(&body->fid1),
1667 inode->i_ino, inode->i_generation, inode);
1669 lli->lli_fid = body->fid1;
1672 LASSERT(fid_seq(&lli->lli_fid) != 0);
1674 if (body->valid & OBD_MD_FLSIZE) {
1675 if (exp_connect_som(ll_i2mdexp(inode)) &&
1676 S_ISREG(inode->i_mode)) {
1677 struct lustre_handle lockh;
1680 /* As it is possible a blocking ast has been processed
1681 * by this time, we need to check there is an UPDATE
1682 * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1684 mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1685 &lockh, LDLM_FL_CBPENDING,
1689 if (lli->lli_flags & (LLIF_DONE_WRITING |
1690 LLIF_EPOCH_PENDING |
1692 CERROR("ino %lu flags %u still has size authority! do not trust the size got from MDS\n",
1693 inode->i_ino, lli->lli_flags);
1695 /* Use old size assignment to avoid
1696 * deadlock bz14138 & bz14326 */
1697 i_size_write(inode, body->size);
1698 spin_lock(&lli->lli_lock);
1699 lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1700 spin_unlock(&lli->lli_lock);
1702 ldlm_lock_decref(&lockh, mode);
1705 /* Use old size assignment to avoid
1706 * deadlock bz14138 & bz14326 */
1707 i_size_write(inode, body->size);
1709 CDEBUG(D_VFSTRACE, "inode=%lu, updating i_size %llu\n",
1710 inode->i_ino, (unsigned long long)body->size);
1713 if (body->valid & OBD_MD_FLBLOCKS)
1714 inode->i_blocks = body->blocks;
1717 if (body->valid & OBD_MD_FLMDSCAPA) {
1718 LASSERT(md->mds_capa);
1719 ll_add_capa(inode, md->mds_capa);
1721 if (body->valid & OBD_MD_FLOSSCAPA) {
1722 LASSERT(md->oss_capa);
1723 ll_add_capa(inode, md->oss_capa);
1726 if (body->valid & OBD_MD_TSTATE) {
1727 if (body->t_state & MS_RESTORE)
1728 lli->lli_flags |= LLIF_FILE_RESTORING;
1732 void ll_read_inode2(struct inode *inode, void *opaque)
1734 struct lustre_md *md = opaque;
1735 struct ll_inode_info *lli = ll_i2info(inode);
1737 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1738 PFID(&lli->lli_fid), inode);
1740 LASSERT(!lli->lli_has_smd);
1742 /* Core attributes from the MDS first. This is a new inode, and
1743 * the VFS doesn't zero times in the core inode so we have to do
1744 * it ourselves. They will be overwritten by either MDS or OST
1745 * attributes - we just need to make sure they aren't newer. */
1746 LTIME_S(inode->i_mtime) = 0;
1747 LTIME_S(inode->i_atime) = 0;
1748 LTIME_S(inode->i_ctime) = 0;
1750 ll_update_inode(inode, md);
1752 /* OIDEBUG(inode); */
1754 if (S_ISREG(inode->i_mode)) {
1755 struct ll_sb_info *sbi = ll_i2sbi(inode);
1757 inode->i_op = &ll_file_inode_operations;
1758 inode->i_fop = sbi->ll_fop;
1759 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1760 } else if (S_ISDIR(inode->i_mode)) {
1761 inode->i_op = &ll_dir_inode_operations;
1762 inode->i_fop = &ll_dir_operations;
1763 } else if (S_ISLNK(inode->i_mode)) {
1764 inode->i_op = &ll_fast_symlink_inode_operations;
1766 inode->i_op = &ll_special_inode_operations;
1768 init_special_inode(inode, inode->i_mode,
1773 void ll_delete_inode(struct inode *inode)
1775 struct cl_inode_info *lli = cl_i2info(inode);
1777 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
1778 /* discard all dirty pages before truncating them, required by
1779 * osc_extent implementation at LU-1030. */
1780 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF,
1781 CL_FSYNC_DISCARD, 1);
1783 truncate_inode_pages_final(&inode->i_data);
1785 /* Workaround for LU-118 */
1786 if (inode->i_data.nrpages) {
1787 spin_lock_irq(&inode->i_data.tree_lock);
1788 spin_unlock_irq(&inode->i_data.tree_lock);
1789 LASSERTF(inode->i_data.nrpages == 0,
1790 "inode=%lu/%u(%p) nrpages=%lu, see http://jira.whamcloud.com/browse/LU-118\n",
1791 inode->i_ino, inode->i_generation, inode,
1792 inode->i_data.nrpages);
1794 /* Workaround end */
1796 ll_clear_inode(inode);
1800 int ll_iocontrol(struct inode *inode, struct file *file,
1801 unsigned int cmd, unsigned long arg)
1803 struct ll_sb_info *sbi = ll_i2sbi(inode);
1804 struct ptlrpc_request *req = NULL;
1808 case FSFILT_IOC_GETFLAGS: {
1809 struct mdt_body *body;
1810 struct md_op_data *op_data;
1812 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
1813 0, 0, LUSTRE_OPC_ANY,
1815 if (IS_ERR(op_data))
1816 return PTR_ERR(op_data);
1818 op_data->op_valid = OBD_MD_FLFLAGS;
1819 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
1820 ll_finish_md_op_data(op_data);
1822 CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1826 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1828 flags = body->flags;
1830 ptlrpc_req_finished(req);
1832 return put_user(flags, (int *)arg);
1834 case FSFILT_IOC_SETFLAGS: {
1835 struct lov_stripe_md *lsm;
1836 struct obd_info oinfo = { { { 0 } } };
1837 struct md_op_data *op_data;
1839 if (get_user(flags, (int *)arg))
1842 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1843 LUSTRE_OPC_ANY, NULL);
1844 if (IS_ERR(op_data))
1845 return PTR_ERR(op_data);
1847 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1848 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1849 rc = md_setattr(sbi->ll_md_exp, op_data,
1850 NULL, 0, NULL, 0, &req, NULL);
1851 ll_finish_md_op_data(op_data);
1852 ptlrpc_req_finished(req);
1856 inode->i_flags = ll_ext_to_inode_flags(flags);
1858 lsm = ccc_inode_lsm_get(inode);
1859 if (!lsm_has_objects(lsm)) {
1860 ccc_inode_lsm_put(inode, lsm);
1864 OBDO_ALLOC(oinfo.oi_oa);
1866 ccc_inode_lsm_put(inode, lsm);
1870 oinfo.oi_oa->o_oi = lsm->lsm_oi;
1871 oinfo.oi_oa->o_flags = flags;
1872 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1874 oinfo.oi_capa = ll_mdscapa_get(inode);
1875 obdo_set_parent_fid(oinfo.oi_oa, &ll_i2info(inode)->lli_fid);
1876 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1877 capa_put(oinfo.oi_capa);
1878 OBDO_FREE(oinfo.oi_oa);
1879 ccc_inode_lsm_put(inode, lsm);
1881 if (rc && rc != -EPERM && rc != -EACCES)
1882 CERROR("osc_setattr_async fails: rc = %d\n", rc);
1893 int ll_flush_ctx(struct inode *inode)
1895 struct ll_sb_info *sbi = ll_i2sbi(inode);
1897 CDEBUG(D_SEC, "flush context for user %d\n",
1898 from_kuid(&init_user_ns, current_uid()));
1900 obd_set_info_async(NULL, sbi->ll_md_exp,
1901 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1903 obd_set_info_async(NULL, sbi->ll_dt_exp,
1904 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1909 /* umount -f client means force down, don't save state */
1910 void ll_umount_begin(struct super_block *sb)
1912 struct ll_sb_info *sbi = ll_s2sbi(sb);
1913 struct obd_device *obd;
1914 struct obd_ioctl_data *ioc_data;
1916 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1917 sb->s_count, atomic_read(&sb->s_active));
1919 obd = class_exp2obd(sbi->ll_md_exp);
1921 CERROR("Invalid MDC connection handle %#llx\n",
1922 sbi->ll_md_exp->exp_handle.h_cookie);
1927 obd = class_exp2obd(sbi->ll_dt_exp);
1929 CERROR("Invalid LOV connection handle %#llx\n",
1930 sbi->ll_dt_exp->exp_handle.h_cookie);
1935 ioc_data = kzalloc(sizeof(*ioc_data), GFP_NOFS);
1937 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
1938 sizeof(*ioc_data), ioc_data, NULL);
1940 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
1941 sizeof(*ioc_data), ioc_data, NULL);
1946 /* Really, we'd like to wait until there are no requests outstanding,
1947 * and then continue. For now, we just invalidate the requests,
1948 * schedule() and sleep one second if needed, and hope.
1953 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
1955 struct ll_sb_info *sbi = ll_s2sbi(sb);
1956 char *profilenm = get_profile_name(sb);
1960 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
1961 read_only = *flags & MS_RDONLY;
1962 err = obd_set_info_async(NULL, sbi->ll_md_exp,
1963 sizeof(KEY_READ_ONLY),
1964 KEY_READ_ONLY, sizeof(read_only),
1967 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
1968 profilenm, read_only ?
1969 "read-only" : "read-write", err);
1974 sb->s_flags |= MS_RDONLY;
1976 sb->s_flags &= ~MS_RDONLY;
1978 if (sbi->ll_flags & LL_SBI_VERBOSE)
1979 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
1980 read_only ? "read-only" : "read-write");
1985 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
1986 struct super_block *sb, struct lookup_intent *it)
1988 struct ll_sb_info *sbi = NULL;
1989 struct lustre_md md;
1992 LASSERT(*inode || sb);
1993 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
1994 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
1995 sbi->ll_md_exp, &md);
2000 ll_update_inode(*inode, &md);
2002 LASSERT(sb != NULL);
2005 * At this point server returns to client's same fid as client
2006 * generated for creating. So using ->fid1 is okay here.
2008 LASSERT(fid_is_sane(&md.body->fid1));
2010 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1,
2011 sbi->ll_flags & LL_SBI_32BIT_API),
2013 if (*inode == NULL || IS_ERR(*inode)) {
2014 #ifdef CONFIG_FS_POSIX_ACL
2016 posix_acl_release(md.posix_acl);
2017 md.posix_acl = NULL;
2020 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2022 CERROR("new_inode -fatal: rc %d\n", rc);
2027 /* Handling piggyback layout lock.
2028 * Layout lock can be piggybacked by getattr and open request.
2029 * The lsm can be applied to inode only if it comes with a layout lock
2030 * otherwise correct layout may be overwritten, for example:
2031 * 1. proc1: mdt returns a lsm but not granting layout
2032 * 2. layout was changed by another client
2033 * 3. proc2: refresh layout and layout lock granted
2034 * 4. proc1: to apply a stale layout */
2035 if (it != NULL && it->d.lustre.it_lock_mode != 0) {
2036 struct lustre_handle lockh;
2037 struct ldlm_lock *lock;
2039 lockh.cookie = it->d.lustre.it_lock_handle;
2040 lock = ldlm_handle2lock(&lockh);
2041 LASSERT(lock != NULL);
2042 if (ldlm_has_layout(lock)) {
2043 struct cl_object_conf conf;
2045 memset(&conf, 0, sizeof(conf));
2046 conf.coc_opc = OBJECT_CONF_SET;
2047 conf.coc_inode = *inode;
2048 conf.coc_lock = lock;
2049 conf.u.coc_md = &md;
2050 (void)ll_layout_conf(*inode, &conf);
2052 LDLM_LOCK_PUT(lock);
2057 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2058 md_free_lustre_md(sbi->ll_md_exp, &md);
2062 int ll_obd_statfs(struct inode *inode, void *arg)
2064 struct ll_sb_info *sbi = NULL;
2065 struct obd_export *exp;
2067 struct obd_ioctl_data *data = NULL;
2077 sbi = ll_i2sbi(inode);
2083 rc = obd_ioctl_getdata(&buf, &len, arg);
2088 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2089 !data->ioc_pbuf1 || !data->ioc_pbuf2) {
2094 if (data->ioc_inllen1 != sizeof(__u32) ||
2095 data->ioc_inllen2 != sizeof(__u32) ||
2096 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2097 data->ioc_plen2 != sizeof(struct obd_uuid)) {
2102 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2103 if (type & LL_STATFS_LMV)
2104 exp = sbi->ll_md_exp;
2105 else if (type & LL_STATFS_LOV)
2106 exp = sbi->ll_dt_exp;
2112 flags = (type & LL_STATFS_NODELAY) ? OBD_STATFS_NODELAY : 0;
2113 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, &flags);
2118 obd_ioctl_freedata(buf, len);
2122 int ll_process_config(struct lustre_cfg *lcfg)
2126 struct lprocfs_static_vars lvars;
2130 lprocfs_llite_init_vars(&lvars);
2132 /* The instance name contains the sb: lustre-client-aacfe000 */
2133 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2134 if (!ptr || !*(++ptr))
2136 rc = kstrtoul(ptr, 16, &x);
2140 /* This better be a real Lustre superblock! */
2141 LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2143 /* Note we have not called client_common_fill_super yet, so
2144 proc fns must be able to handle that! */
2145 rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2152 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2153 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2154 struct inode *i1, struct inode *i2,
2155 const char *name, int namelen,
2156 int mode, __u32 opc, void *data)
2158 LASSERT(i1 != NULL);
2160 if (namelen > ll_i2sbi(i1)->ll_namelen)
2161 return ERR_PTR(-ENAMETOOLONG);
2163 if (op_data == NULL)
2164 op_data = kzalloc(sizeof(*op_data), GFP_NOFS);
2166 if (op_data == NULL)
2167 return ERR_PTR(-ENOMEM);
2169 ll_i2gids(op_data->op_suppgids, i1, i2);
2170 op_data->op_fid1 = *ll_inode2fid(i1);
2171 op_data->op_capa1 = ll_mdscapa_get(i1);
2174 op_data->op_fid2 = *ll_inode2fid(i2);
2175 op_data->op_capa2 = ll_mdscapa_get(i2);
2177 fid_zero(&op_data->op_fid2);
2178 op_data->op_capa2 = NULL;
2181 op_data->op_name = name;
2182 op_data->op_namelen = namelen;
2183 op_data->op_mode = mode;
2184 op_data->op_mod_time = get_seconds();
2185 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2186 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2187 op_data->op_cap = cfs_curproc_cap_pack();
2188 op_data->op_bias = 0;
2189 op_data->op_cli_flags = 0;
2190 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2191 filename_is_volatile(name, namelen, NULL))
2192 op_data->op_bias |= MDS_CREATE_VOLATILE;
2193 op_data->op_opc = opc;
2194 op_data->op_mds = 0;
2195 op_data->op_data = data;
2197 /* If the file is being opened after mknod() (normally due to NFS)
2198 * try to use the default stripe data from parent directory for
2199 * allocating OST objects. Try to pass the parent FID to MDS. */
2200 if (opc == LUSTRE_OPC_CREATE && i1 == i2 && S_ISREG(i2->i_mode) &&
2201 !ll_i2info(i2)->lli_has_smd) {
2202 struct ll_inode_info *lli = ll_i2info(i2);
2204 spin_lock(&lli->lli_lock);
2205 if (likely(!lli->lli_has_smd && !fid_is_zero(&lli->lli_pfid)))
2206 op_data->op_fid1 = lli->lli_pfid;
2207 spin_unlock(&lli->lli_lock);
2208 /** We ignore parent's capability temporary. */
2211 /* When called by ll_setattr_raw, file is i1. */
2212 if (LLIF_DATA_MODIFIED & ll_i2info(i1)->lli_flags)
2213 op_data->op_bias |= MDS_DATA_MODIFIED;
2218 void ll_finish_md_op_data(struct md_op_data *op_data)
2220 capa_put(op_data->op_capa1);
2221 capa_put(op_data->op_capa2);
2225 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2227 struct ll_sb_info *sbi;
2229 LASSERT((seq != NULL) && (dentry != NULL));
2230 sbi = ll_s2sbi(dentry->d_sb);
2232 if (sbi->ll_flags & LL_SBI_NOLCK)
2233 seq_puts(seq, ",nolock");
2235 if (sbi->ll_flags & LL_SBI_FLOCK)
2236 seq_puts(seq, ",flock");
2238 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2239 seq_puts(seq, ",localflock");
2241 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2242 seq_puts(seq, ",user_xattr");
2244 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2245 seq_puts(seq, ",lazystatfs");
2247 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2248 seq_puts(seq, ",user_fid2path");
2254 * Get obd name by cmd, and copy out to user space
2256 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2258 struct ll_sb_info *sbi = ll_i2sbi(inode);
2259 struct obd_device *obd;
2261 if (cmd == OBD_IOC_GETDTNAME)
2262 obd = class_exp2obd(sbi->ll_dt_exp);
2263 else if (cmd == OBD_IOC_GETMDNAME)
2264 obd = class_exp2obd(sbi->ll_md_exp);
2271 if (copy_to_user((void *)arg, obd->obd_name,
2272 strlen(obd->obd_name) + 1))
2279 * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2280 * fsname will be returned in this buffer; otherwise, a static buffer will be
2281 * used to store the fsname and returned to caller.
2283 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2285 static char fsname_static[MTI_NAME_MAXLEN];
2286 struct lustre_sb_info *lsi = s2lsi(sb);
2291 /* this means the caller wants to use static buffer
2292 * and it doesn't care about race. Usually this is
2293 * in error reporting path */
2294 buf = fsname_static;
2295 buflen = sizeof(fsname_static);
2298 len = strlen(lsi->lsi_lmd->lmd_profile);
2299 ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2300 if (ptr && (strcmp(ptr, "-client") == 0))
2303 if (unlikely(len >= buflen))
2305 strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2311 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2313 char *buf, *path = NULL;
2314 struct dentry *dentry = NULL;
2315 struct ccc_object *obj = cl_inode2ccc(page->mapping->host);
2317 /* this can be called inside spin lock so use GFP_ATOMIC. */
2318 buf = (char *)__get_free_page(GFP_ATOMIC);
2320 dentry = d_find_alias(page->mapping->host);
2322 path = dentry_path_raw(dentry, buf, PAGE_SIZE);
2326 "%s: dirty page discard: %s/fid: " DFID "/%s may get corrupted (rc %d)\n",
2327 ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2328 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2329 PFID(&obj->cob_header.coh_lu.loh_fid),
2330 (path && !IS_ERR(path)) ? path : "", ioret);
2336 free_page((unsigned long)buf);