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;
96 /* initialize lru data */
97 atomic_set(&sbi->ll_cache.ccc_users, 0);
98 sbi->ll_cache.ccc_lru_max = lru_page_max;
99 atomic_set(&sbi->ll_cache.ccc_lru_left, lru_page_max);
100 spin_lock_init(&sbi->ll_cache.ccc_lru_lock);
101 INIT_LIST_HEAD(&sbi->ll_cache.ccc_lru);
103 sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
104 SBI_DEFAULT_READAHEAD_MAX);
105 sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
106 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
107 SBI_DEFAULT_READAHEAD_WHOLE_MAX;
108 INIT_LIST_HEAD(&sbi->ll_conn_chain);
109 INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
111 ll_generate_random_uuid(uuid);
112 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
113 CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
115 spin_lock(&ll_sb_lock);
116 list_add_tail(&sbi->ll_list, &ll_super_blocks);
117 spin_unlock(&ll_sb_lock);
119 sbi->ll_flags |= LL_SBI_VERBOSE;
120 sbi->ll_flags |= LL_SBI_CHECKSUM;
122 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
124 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
125 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
127 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
131 /* metadata statahead is enabled by default */
132 sbi->ll_sa_max = LL_SA_RPC_DEF;
133 atomic_set(&sbi->ll_sa_total, 0);
134 atomic_set(&sbi->ll_sa_wrong, 0);
135 atomic_set(&sbi->ll_agl_total, 0);
136 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
141 static void ll_free_sbi(struct super_block *sb)
143 struct ll_sb_info *sbi = ll_s2sbi(sb);
146 spin_lock(&ll_sb_lock);
147 list_del(&sbi->ll_list);
148 spin_unlock(&ll_sb_lock);
149 OBD_FREE(sbi, sizeof(*sbi));
153 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
154 struct vfsmount *mnt)
156 struct inode *root = NULL;
157 struct ll_sb_info *sbi = ll_s2sbi(sb);
158 struct obd_device *obd;
159 struct obd_capa *oc = NULL;
160 struct obd_statfs *osfs = NULL;
161 struct ptlrpc_request *request = NULL;
162 struct obd_connect_data *data = NULL;
163 struct obd_uuid *uuid;
164 struct md_op_data *op_data;
165 struct lustre_md lmd;
167 int size, err, checksum;
169 obd = class_name2obd(md);
171 CERROR("MD %s: not setup or attached\n", md);
175 data = kzalloc(sizeof(*data), GFP_NOFS);
179 osfs = kzalloc(sizeof(*osfs), GFP_NOFS);
185 if (proc_lustre_fs_root) {
186 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
189 CERROR("could not register mount in /proc/fs/lustre\n");
192 /* indicate the features supported by this client */
193 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
194 OBD_CONNECT_ATTRFID |
195 OBD_CONNECT_VERSION | OBD_CONNECT_BRW_SIZE |
196 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
197 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
198 OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 |
199 OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR |
200 OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH|
201 OBD_CONNECT_EINPROGRESS |
202 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
203 OBD_CONNECT_LAYOUTLOCK |
204 OBD_CONNECT_PINGLESS |
205 OBD_CONNECT_MAX_EASIZE |
206 OBD_CONNECT_FLOCK_DEAD |
207 OBD_CONNECT_DISP_STRIPE;
209 if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
210 data->ocd_connect_flags |= OBD_CONNECT_SOM;
212 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
213 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
214 #ifdef CONFIG_FS_POSIX_ACL
215 data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK;
218 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
219 /* flag mdc connection as lightweight, only used for test
220 * purpose, use with care */
221 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
223 data->ocd_ibits_known = MDS_INODELOCK_FULL;
224 data->ocd_version = LUSTRE_VERSION_CODE;
226 if (sb->s_flags & MS_RDONLY)
227 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
228 if (sbi->ll_flags & LL_SBI_USER_XATTR)
229 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
231 #ifdef HAVE_MS_FLOCK_LOCK
232 /* force vfs to use lustre handler for flock() calls - bug 10743 */
233 sb->s_flags |= MS_FLOCK_LOCK;
235 #ifdef MS_HAS_NEW_AOPS
236 sb->s_flags |= MS_HAS_NEW_AOPS;
239 if (sbi->ll_flags & LL_SBI_FLOCK)
240 sbi->ll_fop = &ll_file_operations_flock;
241 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
242 sbi->ll_fop = &ll_file_operations;
244 sbi->ll_fop = &ll_file_operations_noflock;
247 data->ocd_connect_flags |= OBD_CONNECT_REAL;
248 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
249 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
251 data->ocd_brw_size = MD_MAX_BRW_SIZE;
253 err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid,
256 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",
260 CERROR("cannot connect to %s: rc = %d\n", md, err);
264 sbi->ll_md_exp->exp_connect_data = *data;
266 err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
267 LUSTRE_SEQ_METADATA);
269 CERROR("%s: Can't init metadata layer FID infrastructure, rc = %d\n",
270 sbi->ll_md_exp->exp_obd->obd_name, err);
274 /* For mount, we only need fs info from MDT0, and also in DNE, it
275 * can make sure the client can be mounted as long as MDT0 is
277 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
278 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
279 OBD_STATFS_FOR_MDT0);
283 /* This needs to be after statfs to ensure connect has finished.
284 * Note that "data" does NOT contain the valid connect reply.
285 * If connecting to a 1.8 server there will be no LMV device, so
286 * we can access the MDC export directly and exp_connect_flags will
287 * be non-zero, but if accessing an upgraded 2.1 server it will
288 * have the correct flags filled in.
289 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
290 valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
291 if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
292 valid != CLIENT_CONNECT_MDT_REQD) {
295 buf = kzalloc(PAGE_CACHE_SIZE, GFP_KERNEL);
296 obd_connect_flags2str(buf, PAGE_CACHE_SIZE,
297 valid ^ CLIENT_CONNECT_MDT_REQD, ",");
298 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",
299 sbi->ll_md_exp->exp_obd->obd_name, buf);
300 OBD_FREE(buf, PAGE_CACHE_SIZE);
305 size = sizeof(*data);
306 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
307 KEY_CONN_DATA, &size, data, NULL);
309 CERROR("%s: Get connect data failed: rc = %d\n",
310 sbi->ll_md_exp->exp_obd->obd_name, err);
314 LASSERT(osfs->os_bsize);
315 sb->s_blocksize = osfs->os_bsize;
316 sb->s_blocksize_bits = log2(osfs->os_bsize);
317 sb->s_magic = LL_SUPER_MAGIC;
318 sb->s_maxbytes = MAX_LFS_FILESIZE;
319 sbi->ll_namelen = osfs->os_namelen;
320 sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
322 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
323 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
324 LCONSOLE_INFO("Disabling user_xattr feature because it is not supported on the server\n");
325 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
328 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
330 sb->s_flags |= MS_POSIXACL;
332 sbi->ll_flags |= LL_SBI_ACL;
334 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
336 sb->s_flags &= ~MS_POSIXACL;
338 sbi->ll_flags &= ~LL_SBI_ACL;
341 if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
342 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
343 sbi->ll_flags |= LL_SBI_RMT_CLIENT;
344 LCONSOLE_INFO("client is set as remote by default.\n");
347 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
348 sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
349 LCONSOLE_INFO("client claims to be remote, but server rejected, forced to be local.\n");
353 if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
354 LCONSOLE_INFO("client enabled MDS capability!\n");
355 sbi->ll_flags |= LL_SBI_MDS_CAPA;
358 if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
359 LCONSOLE_INFO("client enabled OSS capability!\n");
360 sbi->ll_flags |= LL_SBI_OSS_CAPA;
363 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
364 sbi->ll_flags |= LL_SBI_64BIT_HASH;
366 if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
367 sbi->ll_md_brw_size = data->ocd_brw_size;
369 sbi->ll_md_brw_size = PAGE_CACHE_SIZE;
371 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK) {
372 LCONSOLE_INFO("Layout lock feature supported.\n");
373 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
376 if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
377 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
379 "%s: disabling xattr cache due to unknown maximum xattr size.\n",
382 sbi->ll_flags |= LL_SBI_XATTR_CACHE;
383 sbi->ll_xattr_cache_enabled = 1;
387 obd = class_name2obd(dt);
389 CERROR("DT %s: not setup or attached\n", dt);
394 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
395 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
396 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
397 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
398 OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
399 OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
400 OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
401 OBD_CONNECT_MAXBYTES |
402 OBD_CONNECT_EINPROGRESS |
403 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
404 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS;
406 if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
407 data->ocd_connect_flags |= OBD_CONNECT_SOM;
409 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
410 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
411 * disabled by default, because it can still be enabled on the
412 * fly via /proc. As a consequence, we still need to come to an
413 * agreement on the supported algorithms at connect time */
414 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
416 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
417 data->ocd_cksum_types = OBD_CKSUM_ADLER;
419 data->ocd_cksum_types = cksum_types_supported_client();
422 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
423 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
424 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
426 CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d ocd_grant: %d\n",
427 data->ocd_connect_flags,
428 data->ocd_version, data->ocd_grant);
430 obd->obd_upcall.onu_owner = &sbi->ll_lco;
431 obd->obd_upcall.onu_upcall = cl_ocd_update;
433 data->ocd_brw_size = DT_MAX_BRW_SIZE;
435 err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
438 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",
442 CERROR("%s: Cannot connect to %s: rc = %d\n",
443 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
447 sbi->ll_dt_exp->exp_connect_data = *data;
449 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
450 LUSTRE_SEQ_METADATA);
452 CERROR("%s: Can't init data layer FID infrastructure, rc = %d\n",
453 sbi->ll_dt_exp->exp_obd->obd_name, err);
457 mutex_lock(&sbi->ll_lco.lco_lock);
458 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
459 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
460 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
461 mutex_unlock(&sbi->ll_lco.lco_lock);
463 fid_zero(&sbi->ll_root_fid);
464 err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
466 CERROR("cannot mds_connect: rc = %d\n", err);
469 if (!fid_is_sane(&sbi->ll_root_fid)) {
470 CERROR("%s: Invalid root fid "DFID" during mount\n",
471 sbi->ll_md_exp->exp_obd->obd_name,
472 PFID(&sbi->ll_root_fid));
476 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
478 sb->s_op = &lustre_super_operations;
479 #if THREAD_SIZE >= 8192 /*b=17630*/
480 sb->s_export_op = &lustre_export_operations;
484 * XXX: move this to after cbd setup? */
485 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
486 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
487 valid |= OBD_MD_FLRMTPERM;
488 else if (sbi->ll_flags & LL_SBI_ACL)
489 valid |= OBD_MD_FLACL;
491 op_data = kzalloc(sizeof(*op_data), GFP_NOFS);
497 op_data->op_fid1 = sbi->ll_root_fid;
498 op_data->op_mode = 0;
499 op_data->op_capa1 = oc;
500 op_data->op_valid = valid;
502 err = md_getattr(sbi->ll_md_exp, op_data, &request);
505 OBD_FREE_PTR(op_data);
507 CERROR("%s: md_getattr failed for root: rc = %d\n",
508 sbi->ll_md_exp->exp_obd->obd_name, err);
512 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
513 sbi->ll_md_exp, &lmd);
515 CERROR("failed to understand root inode md: rc = %d\n", err);
516 ptlrpc_req_finished(request);
520 LASSERT(fid_is_sane(&sbi->ll_root_fid));
521 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
522 sbi->ll_flags & LL_SBI_32BIT_API),
524 md_free_lustre_md(sbi->ll_md_exp, &lmd);
525 ptlrpc_req_finished(request);
527 if (root == NULL || IS_ERR(root)) {
529 obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
530 #ifdef CONFIG_FS_POSIX_ACL
532 posix_acl_release(lmd.posix_acl);
533 lmd.posix_acl = NULL;
536 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
538 CERROR("lustre_lite: bad iget4 for root\n");
542 err = ll_close_thread_start(&sbi->ll_lcq);
544 CERROR("cannot start close thread: rc %d\n", err);
548 #ifdef CONFIG_FS_POSIX_ACL
549 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
550 rct_init(&sbi->ll_rct);
551 et_init(&sbi->ll_et);
555 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
556 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
557 KEY_CHECKSUM, sizeof(checksum), &checksum,
561 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
562 KEY_CACHE_SET, sizeof(sbi->ll_cache),
563 &sbi->ll_cache, NULL);
565 sb->s_root = d_make_root(root);
566 if (sb->s_root == NULL) {
567 CERROR("%s: can't make root dentry\n",
568 ll_get_fsname(sb, NULL, 0));
573 sbi->ll_sdev_orig = sb->s_dev;
575 /* We set sb->s_dev equal on all lustre clients in order to support
576 * NFS export clustering. NFSD requires that the FSID be the same
578 /* s_dev is also used in lt_compare() to compare two fs, but that is
579 * only a node-local comparison. */
580 uuid = obd_get_uuid(sbi->ll_md_exp);
582 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
583 get_uuid2fsid(uuid->uuid, strlen(uuid->uuid), &sbi->ll_fsid);
595 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
597 obd_disconnect(sbi->ll_dt_exp);
598 sbi->ll_dt_exp = NULL;
599 /* Make sure all OScs are gone, since cl_cache is accessing sbi. */
600 obd_zombie_barrier();
602 obd_fid_fini(sbi->ll_md_exp->exp_obd);
604 obd_disconnect(sbi->ll_md_exp);
605 sbi->ll_md_exp = NULL;
611 lprocfs_unregister_mountpoint(sbi);
615 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
619 *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
621 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
622 KEY_MAX_EASIZE, &size, lmmsize, NULL);
624 CERROR("Get max mdsize error rc %d \n", rc);
629 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
634 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
635 KEY_DEFAULT_EASIZE, &size, lmmsize, NULL);
637 CERROR("Get default mdsize error rc %d\n", rc);
642 int ll_get_max_cookiesize(struct ll_sb_info *sbi, int *lmmsize)
647 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_COOKIESIZE),
648 KEY_MAX_COOKIESIZE, &size, lmmsize, NULL);
650 CERROR("Get max cookiesize error rc %d\n", rc);
655 int ll_get_default_cookiesize(struct ll_sb_info *sbi, int *lmmsize)
660 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_COOKIESIZE),
661 KEY_DEFAULT_COOKIESIZE, &size, lmmsize, NULL);
663 CERROR("Get default cookiesize error rc %d\n", rc);
668 static void ll_dump_inode(struct inode *inode)
670 struct ll_d_hlist_node *tmp;
671 int dentry_count = 0;
673 LASSERT(inode != NULL);
675 ll_d_hlist_for_each(tmp, &inode->i_dentry)
678 CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
679 inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
680 inode->i_mode, atomic_read(&inode->i_count), dentry_count);
683 void lustre_dump_dentry(struct dentry *dentry, int recur)
685 struct list_head *tmp;
688 LASSERT(dentry != NULL);
690 list_for_each(tmp, &dentry->d_subdirs)
693 CERROR("dentry %p dump: name=%pd parent=%pd (%p), inode=%p, count=%u, flags=0x%x, fsdata=%p, %d subdirs\n",
694 dentry, dentry, dentry->d_parent, dentry->d_parent,
695 dentry->d_inode, d_count(dentry),
696 dentry->d_flags, dentry->d_fsdata, subdirs);
697 if (dentry->d_inode != NULL)
698 ll_dump_inode(dentry->d_inode);
703 list_for_each(tmp, &dentry->d_subdirs) {
704 struct dentry *d = list_entry(tmp, struct dentry, d_child);
706 lustre_dump_dentry(d, recur - 1);
710 static void client_common_put_super(struct super_block *sb)
712 struct ll_sb_info *sbi = ll_s2sbi(sb);
714 #ifdef CONFIG_FS_POSIX_ACL
715 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
716 et_fini(&sbi->ll_et);
717 rct_fini(&sbi->ll_rct);
721 ll_close_thread_shutdown(sbi->ll_lcq);
725 list_del(&sbi->ll_conn_chain);
727 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
728 obd_disconnect(sbi->ll_dt_exp);
729 sbi->ll_dt_exp = NULL;
730 /* wait till all OSCs are gone, since cl_cache is accessing sbi.
732 obd_zombie_barrier();
734 lprocfs_unregister_mountpoint(sbi);
736 obd_fid_fini(sbi->ll_md_exp->exp_obd);
737 obd_disconnect(sbi->ll_md_exp);
738 sbi->ll_md_exp = NULL;
741 void ll_kill_super(struct super_block *sb)
743 struct ll_sb_info *sbi;
746 if (!(sb->s_flags & MS_ACTIVE))
750 /* we need to restore s_dev from changed for clustered NFS before
751 * put_super because new kernels have cached s_dev and change sb->s_dev
752 * in put_super not affected real removing devices */
754 sb->s_dev = sbi->ll_sdev_orig;
755 sbi->ll_umounting = 1;
759 static inline int ll_set_opt(const char *opt, char *data, int fl)
761 if (strncmp(opt, data, strlen(opt)) != 0)
767 /* non-client-specific mount options are parsed in lmd_parse */
768 static int ll_options(char *options, int *flags)
771 char *s1 = options, *s2;
776 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
779 CDEBUG(D_SUPER, "next opt=%s\n", s1);
780 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
785 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
790 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
795 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
800 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
805 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
810 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
815 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
820 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
826 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
831 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
836 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
841 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
846 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
851 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
856 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
861 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
866 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
871 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
876 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
882 s2 = strchr(s1, ',');
890 void ll_lli_init(struct ll_inode_info *lli)
892 lli->lli_inode_magic = LLI_INODE_MAGIC;
894 lli->lli_ioepoch = 0;
895 lli->lli_maxbytes = MAX_LFS_FILESIZE;
896 spin_lock_init(&lli->lli_lock);
897 lli->lli_posix_acl = NULL;
898 lli->lli_remote_perms = NULL;
899 mutex_init(&lli->lli_rmtperm_mutex);
900 /* Do not set lli_fid, it has been initialized already. */
901 fid_zero(&lli->lli_pfid);
902 INIT_LIST_HEAD(&lli->lli_close_list);
903 INIT_LIST_HEAD(&lli->lli_oss_capas);
904 atomic_set(&lli->lli_open_count, 0);
905 lli->lli_mds_capa = NULL;
906 lli->lli_rmtperm_time = 0;
907 lli->lli_pending_och = NULL;
908 lli->lli_mds_read_och = NULL;
909 lli->lli_mds_write_och = NULL;
910 lli->lli_mds_exec_och = NULL;
911 lli->lli_open_fd_read_count = 0;
912 lli->lli_open_fd_write_count = 0;
913 lli->lli_open_fd_exec_count = 0;
914 mutex_init(&lli->lli_och_mutex);
915 spin_lock_init(&lli->lli_agl_lock);
916 lli->lli_has_smd = false;
917 spin_lock_init(&lli->lli_layout_lock);
918 ll_layout_version_set(lli, LL_LAYOUT_GEN_NONE);
919 lli->lli_clob = NULL;
921 init_rwsem(&lli->lli_xattrs_list_rwsem);
922 mutex_init(&lli->lli_xattrs_enq_lock);
924 LASSERT(lli->lli_vfs_inode.i_mode != 0);
925 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
926 mutex_init(&lli->lli_readdir_mutex);
927 lli->lli_opendir_key = NULL;
929 spin_lock_init(&lli->lli_sa_lock);
930 lli->lli_opendir_pid = 0;
932 mutex_init(&lli->lli_size_mutex);
933 lli->lli_symlink_name = NULL;
934 init_rwsem(&lli->lli_trunc_sem);
935 mutex_init(&lli->lli_write_mutex);
936 init_rwsem(&lli->lli_glimpse_sem);
937 lli->lli_glimpse_time = 0;
938 INIT_LIST_HEAD(&lli->lli_agl_list);
939 lli->lli_agl_index = 0;
940 lli->lli_async_rc = 0;
942 mutex_init(&lli->lli_layout_mutex);
945 static inline int ll_bdi_register(struct backing_dev_info *bdi)
947 static atomic_t ll_bdi_num = ATOMIC_INIT(0);
949 bdi->name = "lustre";
950 return bdi_register(bdi, NULL, "lustre-%d",
951 atomic_inc_return(&ll_bdi_num));
954 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
956 struct lustre_profile *lprof = NULL;
957 struct lustre_sb_info *lsi = s2lsi(sb);
958 struct ll_sb_info *sbi;
959 char *dt = NULL, *md = NULL;
960 char *profilenm = get_profile_name(sb);
961 struct config_llog_instance *cfg;
962 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
963 const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
966 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
968 cfg = kzalloc(sizeof(*cfg), GFP_NOFS);
972 try_module_get(THIS_MODULE);
974 /* client additional sb info */
975 lsi->lsi_llsbi = sbi = ll_init_sbi();
977 module_put(THIS_MODULE);
982 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
986 err = bdi_init(&lsi->lsi_bdi);
989 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
990 lsi->lsi_bdi.capabilities = 0;
991 err = ll_bdi_register(&lsi->lsi_bdi);
995 sb->s_bdi = &lsi->lsi_bdi;
996 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
997 sb->s_d_op = &ll_d_ops;
999 /* Generate a string unique to this super, in case some joker tries
1000 to mount the same fs at two mount points.
1001 Use the address of the super itself.*/
1002 cfg->cfg_instance = sb;
1003 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1004 cfg->cfg_callback = class_config_llog_handler;
1005 /* set up client obds */
1006 err = lustre_process_log(sb, profilenm, cfg);
1008 CERROR("Unable to process log: %d\n", err);
1012 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1013 lprof = class_get_profile(profilenm);
1014 if (lprof == NULL) {
1015 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be read from the MGS. Does that filesystem exist?\n",
1020 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1021 lprof->lp_md, lprof->lp_dt);
1023 dt = kzalloc(strlen(lprof->lp_dt) + instlen + 2, GFP_NOFS);
1028 sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1030 md = kzalloc(strlen(lprof->lp_md) + instlen + 2, GFP_NOFS);
1035 sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1037 /* connections, registrations, sb setup */
1038 err = client_common_fill_super(sb, md, dt, mnt);
1042 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1044 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1047 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1048 LCONSOLE_WARN("Mounted %s\n", profilenm);
1052 } /* ll_fill_super */
1054 void ll_put_super(struct super_block *sb)
1056 struct config_llog_instance cfg, params_cfg;
1057 struct obd_device *obd;
1058 struct lustre_sb_info *lsi = s2lsi(sb);
1059 struct ll_sb_info *sbi = ll_s2sbi(sb);
1060 char *profilenm = get_profile_name(sb);
1061 int next, force = 1;
1063 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1065 ll_print_capa_stat(sbi);
1067 cfg.cfg_instance = sb;
1068 lustre_end_log(sb, profilenm, &cfg);
1070 params_cfg.cfg_instance = sb;
1071 lustre_end_log(sb, PARAMS_FILENAME, ¶ms_cfg);
1073 if (sbi->ll_md_exp) {
1074 obd = class_exp2obd(sbi->ll_md_exp);
1076 force = obd->obd_force;
1079 /* We need to set force before the lov_disconnect in
1080 lustre_common_put_super, since l_d cleans up osc's as well. */
1083 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1085 obd->obd_force = force;
1090 /* Only if client_common_fill_super succeeded */
1091 client_common_put_super(sb);
1095 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)))
1096 class_manual_cleanup(obd);
1098 if (sbi->ll_flags & LL_SBI_VERBOSE)
1099 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1102 class_del_profile(profilenm);
1104 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1105 bdi_destroy(&lsi->lsi_bdi);
1106 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1110 lsi->lsi_llsbi = NULL;
1112 lustre_common_put_super(sb);
1114 module_put(THIS_MODULE);
1115 } /* client_put_super */
1117 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1119 struct inode *inode = NULL;
1121 /* NOTE: we depend on atomic igrab() -bzzz */
1122 lock_res_and_lock(lock);
1123 if (lock->l_resource->lr_lvb_inode) {
1124 struct ll_inode_info *lli;
1126 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1127 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1128 inode = igrab(lock->l_resource->lr_lvb_inode);
1130 inode = lock->l_resource->lr_lvb_inode;
1131 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1132 D_WARNING, lock, "lr_lvb_inode %p is bogus: magic %08x",
1133 lock->l_resource->lr_lvb_inode,
1134 lli->lli_inode_magic);
1138 unlock_res_and_lock(lock);
1142 void ll_clear_inode(struct inode *inode)
1144 struct ll_inode_info *lli = ll_i2info(inode);
1145 struct ll_sb_info *sbi = ll_i2sbi(inode);
1147 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1148 inode->i_generation, inode);
1150 if (S_ISDIR(inode->i_mode)) {
1151 /* these should have been cleared in ll_file_release */
1152 LASSERT(lli->lli_opendir_key == NULL);
1153 LASSERT(lli->lli_sai == NULL);
1154 LASSERT(lli->lli_opendir_pid == 0);
1157 spin_lock(&lli->lli_lock);
1158 ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1159 spin_unlock(&lli->lli_lock);
1160 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1162 LASSERT(!lli->lli_open_fd_write_count);
1163 LASSERT(!lli->lli_open_fd_read_count);
1164 LASSERT(!lli->lli_open_fd_exec_count);
1166 if (lli->lli_mds_write_och)
1167 ll_md_real_close(inode, FMODE_WRITE);
1168 if (lli->lli_mds_exec_och)
1169 ll_md_real_close(inode, FMODE_EXEC);
1170 if (lli->lli_mds_read_och)
1171 ll_md_real_close(inode, FMODE_READ);
1173 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1174 OBD_FREE(lli->lli_symlink_name,
1175 strlen(lli->lli_symlink_name) + 1);
1176 lli->lli_symlink_name = NULL;
1179 ll_xattr_cache_destroy(inode);
1181 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1182 LASSERT(lli->lli_posix_acl == NULL);
1183 if (lli->lli_remote_perms) {
1184 free_rmtperm_hash(lli->lli_remote_perms);
1185 lli->lli_remote_perms = NULL;
1188 #ifdef CONFIG_FS_POSIX_ACL
1189 else if (lli->lli_posix_acl) {
1190 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1191 LASSERT(lli->lli_remote_perms == NULL);
1192 posix_acl_release(lli->lli_posix_acl);
1193 lli->lli_posix_acl = NULL;
1196 lli->lli_inode_magic = LLI_INODE_DEAD;
1198 ll_clear_inode_capas(inode);
1199 if (!S_ISDIR(inode->i_mode))
1200 LASSERT(list_empty(&lli->lli_agl_list));
1203 * XXX This has to be done before lsm is freed below, because
1204 * cl_object still uses inode lsm.
1206 cl_inode_fini(inode);
1207 lli->lli_has_smd = false;
1210 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data,
1211 struct md_open_data **mod)
1213 struct lustre_md md;
1214 struct inode *inode = dentry->d_inode;
1215 struct ll_sb_info *sbi = ll_i2sbi(inode);
1216 struct ptlrpc_request *request = NULL;
1219 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1220 LUSTRE_OPC_ANY, NULL);
1221 if (IS_ERR(op_data))
1222 return PTR_ERR(op_data);
1224 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1227 ptlrpc_req_finished(request);
1228 if (rc == -ENOENT) {
1230 /* Unlinked special device node? Or just a race?
1231 * Pretend we done everything. */
1232 if (!S_ISREG(inode->i_mode) &&
1233 !S_ISDIR(inode->i_mode)) {
1234 ia_valid = op_data->op_attr.ia_valid;
1235 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1236 rc = simple_setattr(dentry, &op_data->op_attr);
1237 op_data->op_attr.ia_valid = ia_valid;
1239 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1240 CERROR("md_setattr fails: rc = %d\n", rc);
1245 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1246 sbi->ll_md_exp, &md);
1248 ptlrpc_req_finished(request);
1252 ia_valid = op_data->op_attr.ia_valid;
1253 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1254 * cache is not cleared yet. */
1255 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1256 rc = simple_setattr(dentry, &op_data->op_attr);
1257 op_data->op_attr.ia_valid = ia_valid;
1259 /* Extract epoch data if obtained. */
1260 op_data->op_handle = md.body->handle;
1261 op_data->op_ioepoch = md.body->ioepoch;
1263 ll_update_inode(inode, &md);
1264 ptlrpc_req_finished(request);
1269 /* Close IO epoch and send Size-on-MDS attribute update. */
1270 static int ll_setattr_done_writing(struct inode *inode,
1271 struct md_op_data *op_data,
1272 struct md_open_data *mod)
1274 struct ll_inode_info *lli = ll_i2info(inode);
1277 LASSERT(op_data != NULL);
1278 if (!S_ISREG(inode->i_mode))
1281 CDEBUG(D_INODE, "Epoch %llu closed on "DFID" for truncate\n",
1282 op_data->op_ioepoch, PFID(&lli->lli_fid));
1284 op_data->op_flags = MF_EPOCH_CLOSE;
1285 ll_done_writing_attr(inode, op_data);
1286 ll_pack_inode2opdata(inode, op_data, NULL);
1288 rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1289 if (rc == -EAGAIN) {
1290 /* MDS has instructed us to obtain Size-on-MDS attribute
1291 * from OSTs and send setattr to back to MDS. */
1292 rc = ll_som_update(inode, op_data);
1294 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1300 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1302 struct obd_capa *capa;
1305 if (attr->ia_valid & ATTR_SIZE)
1306 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1308 capa = ll_mdscapa_get(inode);
1310 rc = cl_setattr_ost(inode, attr, capa);
1312 if (attr->ia_valid & ATTR_SIZE)
1313 ll_truncate_free_capa(capa);
1321 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1322 * object(s) determine the file size and mtime. Otherwise, the MDS will
1323 * keep these values until such a time that objects are allocated for it.
1324 * We do the MDS operations first, as it is checking permissions for us.
1325 * We don't to the MDS RPC if there is nothing that we want to store there,
1326 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1327 * going to do an RPC anyways.
1329 * If we are doing a truncate, we will send the mtime and ctime updates
1330 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1331 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1334 * In case of HSMimport, we only set attr on MDS.
1336 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1338 struct inode *inode = dentry->d_inode;
1339 struct ll_inode_info *lli = ll_i2info(inode);
1340 struct md_op_data *op_data = NULL;
1341 struct md_open_data *mod = NULL;
1342 bool file_is_released = false;
1343 int rc = 0, rc1 = 0;
1346 "%s: setattr inode %p/fid:"DFID
1347 " from %llu to %llu, valid %x, hsm_import %d\n",
1348 ll_get_fsname(inode->i_sb, NULL, 0), inode,
1349 PFID(&lli->lli_fid), i_size_read(inode), attr->ia_size,
1350 attr->ia_valid, hsm_import);
1352 if (attr->ia_valid & ATTR_SIZE) {
1353 /* Check new size against VFS/VM file size limit and rlimit */
1354 rc = inode_newsize_ok(inode, attr->ia_size);
1358 /* The maximum Lustre file size is variable, based on the
1359 * OST maximum object size and number of stripes. This
1360 * needs another check in addition to the VFS check above. */
1361 if (attr->ia_size > ll_file_maxbytes(inode)) {
1362 CDEBUG(D_INODE, "file "DFID" too large %llu > %llu\n",
1363 PFID(&lli->lli_fid), attr->ia_size,
1364 ll_file_maxbytes(inode));
1368 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1371 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1372 if (attr->ia_valid & TIMES_SET_FLAGS) {
1373 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1374 !capable(CFS_CAP_FOWNER))
1378 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1379 if (attr->ia_valid & ATTR_CTIME) {
1380 attr->ia_ctime = CURRENT_TIME;
1381 attr->ia_valid |= ATTR_CTIME_SET;
1383 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1384 (attr->ia_valid & ATTR_ATIME)) {
1385 attr->ia_atime = CURRENT_TIME;
1386 attr->ia_valid |= ATTR_ATIME_SET;
1388 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1389 (attr->ia_valid & ATTR_MTIME)) {
1390 attr->ia_mtime = CURRENT_TIME;
1391 attr->ia_valid |= ATTR_MTIME_SET;
1394 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1395 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1396 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1399 /* If we are changing file size, file content is modified, flag it. */
1400 if (attr->ia_valid & ATTR_SIZE) {
1401 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1402 spin_lock(&lli->lli_lock);
1403 lli->lli_flags |= LLIF_DATA_MODIFIED;
1404 spin_unlock(&lli->lli_lock);
1407 /* We always do an MDS RPC, even if we're only changing the size;
1408 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1410 op_data = kzalloc(sizeof(*op_data), GFP_NOFS);
1414 if (!S_ISDIR(inode->i_mode)) {
1415 if (attr->ia_valid & ATTR_SIZE)
1416 inode_dio_write_done(inode);
1417 mutex_unlock(&inode->i_mutex);
1420 memcpy(&op_data->op_attr, attr, sizeof(*attr));
1422 /* Open epoch for truncate. */
1423 if (exp_connect_som(ll_i2mdexp(inode)) &&
1424 (attr->ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1425 op_data->op_flags = MF_EPOCH_OPEN;
1427 /* truncate on a released file must failed with -ENODATA,
1428 * so size must not be set on MDS for released file
1429 * but other attributes must be set
1431 if (S_ISREG(inode->i_mode)) {
1432 struct lov_stripe_md *lsm;
1435 ll_layout_refresh(inode, &gen);
1436 lsm = ccc_inode_lsm_get(inode);
1437 if (lsm && lsm->lsm_pattern & LOV_PATTERN_F_RELEASED)
1438 file_is_released = true;
1439 ccc_inode_lsm_put(inode, lsm);
1442 /* if not in HSM import mode, clear size attr for released file
1443 * we clear the attribute send to MDT in op_data, not the original
1444 * received from caller in attr which is used later to
1445 * decide return code */
1446 if (file_is_released && (attr->ia_valid & ATTR_SIZE) && !hsm_import)
1447 op_data->op_attr.ia_valid &= ~ATTR_SIZE;
1449 rc = ll_md_setattr(dentry, op_data, &mod);
1453 /* truncate failed (only when non HSM import), others succeed */
1454 if (file_is_released) {
1455 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1462 /* RPC to MDT is sent, cancel data modification flag */
1463 if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) {
1464 spin_lock(&lli->lli_lock);
1465 lli->lli_flags &= ~LLIF_DATA_MODIFIED;
1466 spin_unlock(&lli->lli_lock);
1469 ll_ioepoch_open(lli, op_data->op_ioepoch);
1470 if (!S_ISREG(inode->i_mode)) {
1475 if (attr->ia_valid & (ATTR_SIZE |
1476 ATTR_ATIME | ATTR_ATIME_SET |
1477 ATTR_MTIME | ATTR_MTIME_SET))
1478 /* For truncate and utimes sending attributes to OSTs, setting
1479 * mtime/atime to the past will be performed under PW [0:EOF]
1480 * extent lock (new_size:EOF for truncate). It may seem
1481 * excessive to send mtime/atime updates to OSTs when not
1482 * setting times to past, but it is necessary due to possible
1483 * time de-synchronization between MDT inode and OST objects */
1484 if (attr->ia_valid & ATTR_SIZE)
1485 down_write(&lli->lli_trunc_sem);
1486 rc = ll_setattr_ost(inode, attr);
1487 if (attr->ia_valid & ATTR_SIZE)
1488 up_write(&lli->lli_trunc_sem);
1491 if (op_data->op_ioepoch) {
1492 rc1 = ll_setattr_done_writing(inode, op_data, mod);
1496 ll_finish_md_op_data(op_data);
1498 if (!S_ISDIR(inode->i_mode)) {
1499 mutex_lock(&inode->i_mutex);
1500 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1501 inode_dio_wait(inode);
1504 ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1505 LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1510 int ll_setattr(struct dentry *de, struct iattr *attr)
1512 int mode = de->d_inode->i_mode;
1514 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1515 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1516 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1518 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1519 (ATTR_SIZE|ATTR_MODE)) &&
1520 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1521 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1522 !(attr->ia_mode & S_ISGID))))
1523 attr->ia_valid |= ATTR_FORCE;
1525 if ((attr->ia_valid & ATTR_MODE) &&
1527 !(attr->ia_mode & S_ISUID) &&
1528 !(attr->ia_valid & ATTR_KILL_SUID))
1529 attr->ia_valid |= ATTR_KILL_SUID;
1531 if ((attr->ia_valid & ATTR_MODE) &&
1532 ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1533 !(attr->ia_mode & S_ISGID) &&
1534 !(attr->ia_valid & ATTR_KILL_SGID))
1535 attr->ia_valid |= ATTR_KILL_SGID;
1537 return ll_setattr_raw(de, attr, false);
1540 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1541 __u64 max_age, __u32 flags)
1543 struct ll_sb_info *sbi = ll_s2sbi(sb);
1544 struct obd_statfs obd_osfs;
1547 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1549 CERROR("md_statfs fails: rc = %d\n", rc);
1553 osfs->os_type = sb->s_magic;
1555 CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1556 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,
1559 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1560 flags |= OBD_STATFS_NODELAY;
1562 rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1564 CERROR("obd_statfs fails: rc = %d\n", rc);
1568 CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1569 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1572 osfs->os_bsize = obd_osfs.os_bsize;
1573 osfs->os_blocks = obd_osfs.os_blocks;
1574 osfs->os_bfree = obd_osfs.os_bfree;
1575 osfs->os_bavail = obd_osfs.os_bavail;
1577 /* If we don't have as many objects free on the OST as inodes
1578 * on the MDS, we reduce the total number of inodes to
1579 * compensate, so that the "inodes in use" number is correct.
1581 if (obd_osfs.os_ffree < osfs->os_ffree) {
1582 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1584 osfs->os_ffree = obd_osfs.os_ffree;
1589 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1591 struct super_block *sb = de->d_sb;
1592 struct obd_statfs osfs;
1595 CDEBUG(D_VFSTRACE, "VFS Op: at %llu jiffies\n", get_jiffies_64());
1596 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1598 /* Some amount of caching on the client is allowed */
1599 rc = ll_statfs_internal(sb, &osfs,
1600 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1605 statfs_unpack(sfs, &osfs);
1607 /* We need to downshift for all 32-bit kernels, because we can't
1608 * tell if the kernel is being called via sys_statfs64() or not.
1609 * Stop before overflowing f_bsize - in which case it is better
1610 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1611 if (sizeof(long) < 8) {
1612 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1615 osfs.os_blocks >>= 1;
1616 osfs.os_bfree >>= 1;
1617 osfs.os_bavail >>= 1;
1621 sfs->f_blocks = osfs.os_blocks;
1622 sfs->f_bfree = osfs.os_bfree;
1623 sfs->f_bavail = osfs.os_bavail;
1624 sfs->f_fsid = ll_s2sbi(sb)->ll_fsid;
1628 void ll_inode_size_lock(struct inode *inode)
1630 struct ll_inode_info *lli;
1632 LASSERT(!S_ISDIR(inode->i_mode));
1634 lli = ll_i2info(inode);
1635 mutex_lock(&lli->lli_size_mutex);
1638 void ll_inode_size_unlock(struct inode *inode)
1640 struct ll_inode_info *lli;
1642 lli = ll_i2info(inode);
1643 mutex_unlock(&lli->lli_size_mutex);
1646 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1648 struct ll_inode_info *lli = ll_i2info(inode);
1649 struct mdt_body *body = md->body;
1650 struct lov_stripe_md *lsm = md->lsm;
1651 struct ll_sb_info *sbi = ll_i2sbi(inode);
1653 LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1655 if (!lli->lli_has_smd &&
1656 !(sbi->ll_flags & LL_SBI_LAYOUT_LOCK))
1657 cl_file_inode_init(inode, md);
1659 lli->lli_maxbytes = lsm->lsm_maxbytes;
1660 if (lli->lli_maxbytes > MAX_LFS_FILESIZE)
1661 lli->lli_maxbytes = MAX_LFS_FILESIZE;
1664 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1665 if (body->valid & OBD_MD_FLRMTPERM)
1666 ll_update_remote_perm(inode, md->remote_perm);
1668 #ifdef CONFIG_FS_POSIX_ACL
1669 else if (body->valid & OBD_MD_FLACL) {
1670 spin_lock(&lli->lli_lock);
1671 if (lli->lli_posix_acl)
1672 posix_acl_release(lli->lli_posix_acl);
1673 lli->lli_posix_acl = md->posix_acl;
1674 spin_unlock(&lli->lli_lock);
1677 inode->i_ino = cl_fid_build_ino(&body->fid1,
1678 sbi->ll_flags & LL_SBI_32BIT_API);
1679 inode->i_generation = cl_fid_build_gen(&body->fid1);
1681 if (body->valid & OBD_MD_FLATIME) {
1682 if (body->atime > LTIME_S(inode->i_atime))
1683 LTIME_S(inode->i_atime) = body->atime;
1684 lli->lli_lvb.lvb_atime = body->atime;
1686 if (body->valid & OBD_MD_FLMTIME) {
1687 if (body->mtime > LTIME_S(inode->i_mtime)) {
1688 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to %llu\n",
1689 inode->i_ino, LTIME_S(inode->i_mtime),
1691 LTIME_S(inode->i_mtime) = body->mtime;
1693 lli->lli_lvb.lvb_mtime = body->mtime;
1695 if (body->valid & OBD_MD_FLCTIME) {
1696 if (body->ctime > LTIME_S(inode->i_ctime))
1697 LTIME_S(inode->i_ctime) = body->ctime;
1698 lli->lli_lvb.lvb_ctime = body->ctime;
1700 if (body->valid & OBD_MD_FLMODE)
1701 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1702 if (body->valid & OBD_MD_FLTYPE)
1703 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1704 LASSERT(inode->i_mode != 0);
1705 if (S_ISREG(inode->i_mode))
1706 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1707 LL_MAX_BLKSIZE_BITS);
1709 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1710 if (body->valid & OBD_MD_FLUID)
1711 inode->i_uid = make_kuid(&init_user_ns, body->uid);
1712 if (body->valid & OBD_MD_FLGID)
1713 inode->i_gid = make_kgid(&init_user_ns, body->gid);
1714 if (body->valid & OBD_MD_FLFLAGS)
1715 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1716 if (body->valid & OBD_MD_FLNLINK)
1717 set_nlink(inode, body->nlink);
1718 if (body->valid & OBD_MD_FLRDEV)
1719 inode->i_rdev = old_decode_dev(body->rdev);
1721 if (body->valid & OBD_MD_FLID) {
1722 /* FID shouldn't be changed! */
1723 if (fid_is_sane(&lli->lli_fid)) {
1724 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1725 "Trying to change FID "DFID
1726 " to the "DFID", inode %lu/%u(%p)\n",
1727 PFID(&lli->lli_fid), PFID(&body->fid1),
1728 inode->i_ino, inode->i_generation, inode);
1730 lli->lli_fid = body->fid1;
1733 LASSERT(fid_seq(&lli->lli_fid) != 0);
1735 if (body->valid & OBD_MD_FLSIZE) {
1736 if (exp_connect_som(ll_i2mdexp(inode)) &&
1737 S_ISREG(inode->i_mode)) {
1738 struct lustre_handle lockh;
1741 /* As it is possible a blocking ast has been processed
1742 * by this time, we need to check there is an UPDATE
1743 * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1745 mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1746 &lockh, LDLM_FL_CBPENDING,
1750 if (lli->lli_flags & (LLIF_DONE_WRITING |
1751 LLIF_EPOCH_PENDING |
1753 CERROR("ino %lu flags %u still has size authority! do not trust the size got from MDS\n",
1754 inode->i_ino, lli->lli_flags);
1756 /* Use old size assignment to avoid
1757 * deadlock bz14138 & bz14326 */
1758 i_size_write(inode, body->size);
1759 spin_lock(&lli->lli_lock);
1760 lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1761 spin_unlock(&lli->lli_lock);
1763 ldlm_lock_decref(&lockh, mode);
1766 /* Use old size assignment to avoid
1767 * deadlock bz14138 & bz14326 */
1768 i_size_write(inode, body->size);
1770 CDEBUG(D_VFSTRACE, "inode=%lu, updating i_size %llu\n",
1771 inode->i_ino, (unsigned long long)body->size);
1774 if (body->valid & OBD_MD_FLBLOCKS)
1775 inode->i_blocks = body->blocks;
1778 if (body->valid & OBD_MD_FLMDSCAPA) {
1779 LASSERT(md->mds_capa);
1780 ll_add_capa(inode, md->mds_capa);
1782 if (body->valid & OBD_MD_FLOSSCAPA) {
1783 LASSERT(md->oss_capa);
1784 ll_add_capa(inode, md->oss_capa);
1787 if (body->valid & OBD_MD_TSTATE) {
1788 if (body->t_state & MS_RESTORE)
1789 lli->lli_flags |= LLIF_FILE_RESTORING;
1793 void ll_read_inode2(struct inode *inode, void *opaque)
1795 struct lustre_md *md = opaque;
1796 struct ll_inode_info *lli = ll_i2info(inode);
1798 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1799 PFID(&lli->lli_fid), inode);
1801 LASSERT(!lli->lli_has_smd);
1803 /* Core attributes from the MDS first. This is a new inode, and
1804 * the VFS doesn't zero times in the core inode so we have to do
1805 * it ourselves. They will be overwritten by either MDS or OST
1806 * attributes - we just need to make sure they aren't newer. */
1807 LTIME_S(inode->i_mtime) = 0;
1808 LTIME_S(inode->i_atime) = 0;
1809 LTIME_S(inode->i_ctime) = 0;
1811 ll_update_inode(inode, md);
1813 /* OIDEBUG(inode); */
1815 if (S_ISREG(inode->i_mode)) {
1816 struct ll_sb_info *sbi = ll_i2sbi(inode);
1818 inode->i_op = &ll_file_inode_operations;
1819 inode->i_fop = sbi->ll_fop;
1820 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1821 } else if (S_ISDIR(inode->i_mode)) {
1822 inode->i_op = &ll_dir_inode_operations;
1823 inode->i_fop = &ll_dir_operations;
1824 } else if (S_ISLNK(inode->i_mode)) {
1825 inode->i_op = &ll_fast_symlink_inode_operations;
1827 inode->i_op = &ll_special_inode_operations;
1829 init_special_inode(inode, inode->i_mode,
1834 void ll_delete_inode(struct inode *inode)
1836 struct cl_inode_info *lli = cl_i2info(inode);
1838 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
1839 /* discard all dirty pages before truncating them, required by
1840 * osc_extent implementation at LU-1030. */
1841 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF,
1842 CL_FSYNC_DISCARD, 1);
1844 truncate_inode_pages_final(&inode->i_data);
1846 /* Workaround for LU-118 */
1847 if (inode->i_data.nrpages) {
1848 spin_lock_irq(&inode->i_data.tree_lock);
1849 spin_unlock_irq(&inode->i_data.tree_lock);
1850 LASSERTF(inode->i_data.nrpages == 0,
1851 "inode=%lu/%u(%p) nrpages=%lu, see http://jira.whamcloud.com/browse/LU-118\n",
1852 inode->i_ino, inode->i_generation, inode,
1853 inode->i_data.nrpages);
1855 /* Workaround end */
1857 ll_clear_inode(inode);
1861 int ll_iocontrol(struct inode *inode, struct file *file,
1862 unsigned int cmd, unsigned long arg)
1864 struct ll_sb_info *sbi = ll_i2sbi(inode);
1865 struct ptlrpc_request *req = NULL;
1869 case FSFILT_IOC_GETFLAGS: {
1870 struct mdt_body *body;
1871 struct md_op_data *op_data;
1873 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
1874 0, 0, LUSTRE_OPC_ANY,
1876 if (IS_ERR(op_data))
1877 return PTR_ERR(op_data);
1879 op_data->op_valid = OBD_MD_FLFLAGS;
1880 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
1881 ll_finish_md_op_data(op_data);
1883 CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1887 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1889 flags = body->flags;
1891 ptlrpc_req_finished(req);
1893 return put_user(flags, (int *)arg);
1895 case FSFILT_IOC_SETFLAGS: {
1896 struct lov_stripe_md *lsm;
1897 struct obd_info oinfo = { { { 0 } } };
1898 struct md_op_data *op_data;
1900 if (get_user(flags, (int *)arg))
1903 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1904 LUSTRE_OPC_ANY, NULL);
1905 if (IS_ERR(op_data))
1906 return PTR_ERR(op_data);
1908 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1909 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1910 rc = md_setattr(sbi->ll_md_exp, op_data,
1911 NULL, 0, NULL, 0, &req, NULL);
1912 ll_finish_md_op_data(op_data);
1913 ptlrpc_req_finished(req);
1917 inode->i_flags = ll_ext_to_inode_flags(flags);
1919 lsm = ccc_inode_lsm_get(inode);
1920 if (!lsm_has_objects(lsm)) {
1921 ccc_inode_lsm_put(inode, lsm);
1925 OBDO_ALLOC(oinfo.oi_oa);
1927 ccc_inode_lsm_put(inode, lsm);
1931 oinfo.oi_oa->o_oi = lsm->lsm_oi;
1932 oinfo.oi_oa->o_flags = flags;
1933 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1935 oinfo.oi_capa = ll_mdscapa_get(inode);
1936 obdo_set_parent_fid(oinfo.oi_oa, &ll_i2info(inode)->lli_fid);
1937 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1938 capa_put(oinfo.oi_capa);
1939 OBDO_FREE(oinfo.oi_oa);
1940 ccc_inode_lsm_put(inode, lsm);
1942 if (rc && rc != -EPERM && rc != -EACCES)
1943 CERROR("osc_setattr_async fails: rc = %d\n", rc);
1954 int ll_flush_ctx(struct inode *inode)
1956 struct ll_sb_info *sbi = ll_i2sbi(inode);
1958 CDEBUG(D_SEC, "flush context for user %d\n",
1959 from_kuid(&init_user_ns, current_uid()));
1961 obd_set_info_async(NULL, sbi->ll_md_exp,
1962 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1964 obd_set_info_async(NULL, sbi->ll_dt_exp,
1965 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1970 /* umount -f client means force down, don't save state */
1971 void ll_umount_begin(struct super_block *sb)
1973 struct ll_sb_info *sbi = ll_s2sbi(sb);
1974 struct obd_device *obd;
1975 struct obd_ioctl_data *ioc_data;
1977 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1978 sb->s_count, atomic_read(&sb->s_active));
1980 obd = class_exp2obd(sbi->ll_md_exp);
1982 CERROR("Invalid MDC connection handle %#llx\n",
1983 sbi->ll_md_exp->exp_handle.h_cookie);
1988 obd = class_exp2obd(sbi->ll_dt_exp);
1990 CERROR("Invalid LOV connection handle %#llx\n",
1991 sbi->ll_dt_exp->exp_handle.h_cookie);
1996 ioc_data = kzalloc(sizeof(*ioc_data), GFP_NOFS);
1998 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
1999 sizeof(*ioc_data), ioc_data, NULL);
2001 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2002 sizeof(*ioc_data), ioc_data, NULL);
2004 OBD_FREE_PTR(ioc_data);
2007 /* Really, we'd like to wait until there are no requests outstanding,
2008 * and then continue. For now, we just invalidate the requests,
2009 * schedule() and sleep one second if needed, and hope.
2014 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2016 struct ll_sb_info *sbi = ll_s2sbi(sb);
2017 char *profilenm = get_profile_name(sb);
2021 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2022 read_only = *flags & MS_RDONLY;
2023 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2024 sizeof(KEY_READ_ONLY),
2025 KEY_READ_ONLY, sizeof(read_only),
2028 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2029 profilenm, read_only ?
2030 "read-only" : "read-write", err);
2035 sb->s_flags |= MS_RDONLY;
2037 sb->s_flags &= ~MS_RDONLY;
2039 if (sbi->ll_flags & LL_SBI_VERBOSE)
2040 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2041 read_only ? "read-only" : "read-write");
2046 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2047 struct super_block *sb, struct lookup_intent *it)
2049 struct ll_sb_info *sbi = NULL;
2050 struct lustre_md md;
2053 LASSERT(*inode || sb);
2054 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2055 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2056 sbi->ll_md_exp, &md);
2061 ll_update_inode(*inode, &md);
2063 LASSERT(sb != NULL);
2066 * At this point server returns to client's same fid as client
2067 * generated for creating. So using ->fid1 is okay here.
2069 LASSERT(fid_is_sane(&md.body->fid1));
2071 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1,
2072 sbi->ll_flags & LL_SBI_32BIT_API),
2074 if (*inode == NULL || IS_ERR(*inode)) {
2075 #ifdef CONFIG_FS_POSIX_ACL
2077 posix_acl_release(md.posix_acl);
2078 md.posix_acl = NULL;
2081 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2083 CERROR("new_inode -fatal: rc %d\n", rc);
2088 /* Handling piggyback layout lock.
2089 * Layout lock can be piggybacked by getattr and open request.
2090 * The lsm can be applied to inode only if it comes with a layout lock
2091 * otherwise correct layout may be overwritten, for example:
2092 * 1. proc1: mdt returns a lsm but not granting layout
2093 * 2. layout was changed by another client
2094 * 3. proc2: refresh layout and layout lock granted
2095 * 4. proc1: to apply a stale layout */
2096 if (it != NULL && it->d.lustre.it_lock_mode != 0) {
2097 struct lustre_handle lockh;
2098 struct ldlm_lock *lock;
2100 lockh.cookie = it->d.lustre.it_lock_handle;
2101 lock = ldlm_handle2lock(&lockh);
2102 LASSERT(lock != NULL);
2103 if (ldlm_has_layout(lock)) {
2104 struct cl_object_conf conf;
2106 memset(&conf, 0, sizeof(conf));
2107 conf.coc_opc = OBJECT_CONF_SET;
2108 conf.coc_inode = *inode;
2109 conf.coc_lock = lock;
2110 conf.u.coc_md = &md;
2111 (void)ll_layout_conf(*inode, &conf);
2113 LDLM_LOCK_PUT(lock);
2118 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2119 md_free_lustre_md(sbi->ll_md_exp, &md);
2123 int ll_obd_statfs(struct inode *inode, void *arg)
2125 struct ll_sb_info *sbi = NULL;
2126 struct obd_export *exp;
2128 struct obd_ioctl_data *data = NULL;
2138 sbi = ll_i2sbi(inode);
2144 rc = obd_ioctl_getdata(&buf, &len, arg);
2149 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2150 !data->ioc_pbuf1 || !data->ioc_pbuf2) {
2155 if (data->ioc_inllen1 != sizeof(__u32) ||
2156 data->ioc_inllen2 != sizeof(__u32) ||
2157 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2158 data->ioc_plen2 != sizeof(struct obd_uuid)) {
2163 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2164 if (type & LL_STATFS_LMV)
2165 exp = sbi->ll_md_exp;
2166 else if (type & LL_STATFS_LOV)
2167 exp = sbi->ll_dt_exp;
2173 flags = (type & LL_STATFS_NODELAY) ? OBD_STATFS_NODELAY : 0;
2174 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, &flags);
2179 obd_ioctl_freedata(buf, len);
2183 int ll_process_config(struct lustre_cfg *lcfg)
2187 struct lprocfs_static_vars lvars;
2191 lprocfs_llite_init_vars(&lvars);
2193 /* The instance name contains the sb: lustre-client-aacfe000 */
2194 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2195 if (!ptr || !*(++ptr))
2197 if (sscanf(ptr, "%lx", &x) != 1)
2200 /* This better be a real Lustre superblock! */
2201 LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2203 /* Note we have not called client_common_fill_super yet, so
2204 proc fns must be able to handle that! */
2205 rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2212 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2213 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2214 struct inode *i1, struct inode *i2,
2215 const char *name, int namelen,
2216 int mode, __u32 opc, void *data)
2218 LASSERT(i1 != NULL);
2220 if (namelen > ll_i2sbi(i1)->ll_namelen)
2221 return ERR_PTR(-ENAMETOOLONG);
2223 if (op_data == NULL)
2224 op_data = kzalloc(sizeof(*op_data), GFP_NOFS);
2226 if (op_data == NULL)
2227 return ERR_PTR(-ENOMEM);
2229 ll_i2gids(op_data->op_suppgids, i1, i2);
2230 op_data->op_fid1 = *ll_inode2fid(i1);
2231 op_data->op_capa1 = ll_mdscapa_get(i1);
2234 op_data->op_fid2 = *ll_inode2fid(i2);
2235 op_data->op_capa2 = ll_mdscapa_get(i2);
2237 fid_zero(&op_data->op_fid2);
2238 op_data->op_capa2 = NULL;
2241 op_data->op_name = name;
2242 op_data->op_namelen = namelen;
2243 op_data->op_mode = mode;
2244 op_data->op_mod_time = get_seconds();
2245 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2246 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2247 op_data->op_cap = cfs_curproc_cap_pack();
2248 op_data->op_bias = 0;
2249 op_data->op_cli_flags = 0;
2250 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2251 filename_is_volatile(name, namelen, NULL))
2252 op_data->op_bias |= MDS_CREATE_VOLATILE;
2253 op_data->op_opc = opc;
2254 op_data->op_mds = 0;
2255 op_data->op_data = data;
2257 /* If the file is being opened after mknod() (normally due to NFS)
2258 * try to use the default stripe data from parent directory for
2259 * allocating OST objects. Try to pass the parent FID to MDS. */
2260 if (opc == LUSTRE_OPC_CREATE && i1 == i2 && S_ISREG(i2->i_mode) &&
2261 !ll_i2info(i2)->lli_has_smd) {
2262 struct ll_inode_info *lli = ll_i2info(i2);
2264 spin_lock(&lli->lli_lock);
2265 if (likely(!lli->lli_has_smd && !fid_is_zero(&lli->lli_pfid)))
2266 op_data->op_fid1 = lli->lli_pfid;
2267 spin_unlock(&lli->lli_lock);
2268 /** We ignore parent's capability temporary. */
2271 /* When called by ll_setattr_raw, file is i1. */
2272 if (LLIF_DATA_MODIFIED & ll_i2info(i1)->lli_flags)
2273 op_data->op_bias |= MDS_DATA_MODIFIED;
2278 void ll_finish_md_op_data(struct md_op_data *op_data)
2280 capa_put(op_data->op_capa1);
2281 capa_put(op_data->op_capa2);
2282 OBD_FREE_PTR(op_data);
2285 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2287 struct ll_sb_info *sbi;
2289 LASSERT((seq != NULL) && (dentry != NULL));
2290 sbi = ll_s2sbi(dentry->d_sb);
2292 if (sbi->ll_flags & LL_SBI_NOLCK)
2293 seq_puts(seq, ",nolock");
2295 if (sbi->ll_flags & LL_SBI_FLOCK)
2296 seq_puts(seq, ",flock");
2298 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2299 seq_puts(seq, ",localflock");
2301 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2302 seq_puts(seq, ",user_xattr");
2304 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2305 seq_puts(seq, ",lazystatfs");
2307 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2308 seq_puts(seq, ",user_fid2path");
2314 * Get obd name by cmd, and copy out to user space
2316 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2318 struct ll_sb_info *sbi = ll_i2sbi(inode);
2319 struct obd_device *obd;
2321 if (cmd == OBD_IOC_GETDTNAME)
2322 obd = class_exp2obd(sbi->ll_dt_exp);
2323 else if (cmd == OBD_IOC_GETMDNAME)
2324 obd = class_exp2obd(sbi->ll_md_exp);
2331 if (copy_to_user((void *)arg, obd->obd_name,
2332 strlen(obd->obd_name) + 1))
2339 * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2340 * fsname will be returned in this buffer; otherwise, a static buffer will be
2341 * used to store the fsname and returned to caller.
2343 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2345 static char fsname_static[MTI_NAME_MAXLEN];
2346 struct lustre_sb_info *lsi = s2lsi(sb);
2351 /* this means the caller wants to use static buffer
2352 * and it doesn't care about race. Usually this is
2353 * in error reporting path */
2354 buf = fsname_static;
2355 buflen = sizeof(fsname_static);
2358 len = strlen(lsi->lsi_lmd->lmd_profile);
2359 ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2360 if (ptr && (strcmp(ptr, "-client") == 0))
2363 if (unlikely(len >= buflen))
2365 strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2371 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2373 char *buf, *path = NULL;
2374 struct dentry *dentry = NULL;
2375 struct ccc_object *obj = cl_inode2ccc(page->mapping->host);
2377 /* this can be called inside spin lock so use GFP_ATOMIC. */
2378 buf = (char *)__get_free_page(GFP_ATOMIC);
2380 dentry = d_find_alias(page->mapping->host);
2382 path = dentry_path_raw(dentry, buf, PAGE_SIZE);
2386 "%s: dirty page discard: %s/fid: " DFID "/%s may get corrupted (rc %d)\n",
2387 ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2388 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2389 PFID(&obj->cob_header.coh_lu.loh_fid),
2390 (path && !IS_ERR(path)) ? path : "", ioret);
2396 free_page((unsigned long)buf);