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, data, NULL);
255 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
256 "recovery, of which this client is not a "
257 "part. Please wait for recovery to complete,"
258 " abort, or time out.\n", md);
261 CERROR("cannot connect to %s: rc = %d\n", md, err);
265 sbi->ll_md_exp->exp_connect_data = *data;
267 err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
268 LUSTRE_SEQ_METADATA);
270 CERROR("%s: Can't init metadata layer FID infrastructure, "
271 "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
275 /* For mount, we only need fs info from MDT0, and also in DNE, it
276 * can make sure the client can be mounted as long as MDT0 is
278 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
279 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
280 OBD_STATFS_FOR_MDT0);
284 /* This needs to be after statfs to ensure connect has finished.
285 * Note that "data" does NOT contain the valid connect reply.
286 * If connecting to a 1.8 server there will be no LMV device, so
287 * we can access the MDC export directly and exp_connect_flags will
288 * be non-zero, but if accessing an upgraded 2.1 server it will
289 * have the correct flags filled in.
290 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
291 valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
292 if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
293 valid != CLIENT_CONNECT_MDT_REQD) {
296 buf = kzalloc(PAGE_CACHE_SIZE, GFP_KERNEL);
297 obd_connect_flags2str(buf, PAGE_CACHE_SIZE,
298 valid ^ CLIENT_CONNECT_MDT_REQD, ",");
299 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
300 "feature(s) needed for correct operation "
301 "of this client (%s). Please upgrade "
302 "server or downgrade client.\n",
303 sbi->ll_md_exp->exp_obd->obd_name, buf);
304 OBD_FREE(buf, PAGE_CACHE_SIZE);
309 size = sizeof(*data);
310 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
311 KEY_CONN_DATA, &size, data, NULL);
313 CERROR("%s: Get connect data failed: rc = %d\n",
314 sbi->ll_md_exp->exp_obd->obd_name, err);
318 LASSERT(osfs->os_bsize);
319 sb->s_blocksize = osfs->os_bsize;
320 sb->s_blocksize_bits = log2(osfs->os_bsize);
321 sb->s_magic = LL_SUPER_MAGIC;
322 sb->s_maxbytes = MAX_LFS_FILESIZE;
323 sbi->ll_namelen = osfs->os_namelen;
324 sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
326 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
327 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
328 LCONSOLE_INFO("Disabling user_xattr feature because "
329 "it is not supported on the server\n");
330 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
333 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
335 sb->s_flags |= MS_POSIXACL;
337 sbi->ll_flags |= LL_SBI_ACL;
339 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
341 sb->s_flags &= ~MS_POSIXACL;
343 sbi->ll_flags &= ~LL_SBI_ACL;
346 if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
347 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
348 sbi->ll_flags |= LL_SBI_RMT_CLIENT;
349 LCONSOLE_INFO("client is set as remote by default.\n");
352 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
353 sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
354 LCONSOLE_INFO("client claims to be remote, but server "
355 "rejected, forced to be local.\n");
359 if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
360 LCONSOLE_INFO("client enabled MDS capability!\n");
361 sbi->ll_flags |= LL_SBI_MDS_CAPA;
364 if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
365 LCONSOLE_INFO("client enabled OSS capability!\n");
366 sbi->ll_flags |= LL_SBI_OSS_CAPA;
369 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
370 sbi->ll_flags |= LL_SBI_64BIT_HASH;
372 if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
373 sbi->ll_md_brw_size = data->ocd_brw_size;
375 sbi->ll_md_brw_size = PAGE_CACHE_SIZE;
377 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK) {
378 LCONSOLE_INFO("Layout lock feature supported.\n");
379 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
382 if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
383 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
385 "%s: disabling xattr cache due to unknown maximum xattr size.\n",
388 sbi->ll_flags |= LL_SBI_XATTR_CACHE;
389 sbi->ll_xattr_cache_enabled = 1;
393 obd = class_name2obd(dt);
395 CERROR("DT %s: not setup or attached\n", dt);
400 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
401 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
402 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
403 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
404 OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
405 OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
406 OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
407 OBD_CONNECT_MAXBYTES |
408 OBD_CONNECT_EINPROGRESS |
409 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
410 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS;
412 if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
413 data->ocd_connect_flags |= OBD_CONNECT_SOM;
415 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
416 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
417 * disabled by default, because it can still be enabled on the
418 * fly via /proc. As a consequence, we still need to come to an
419 * agreement on the supported algorithms at connect time */
420 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
422 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
423 data->ocd_cksum_types = OBD_CKSUM_ADLER;
425 data->ocd_cksum_types = cksum_types_supported_client();
428 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
429 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
430 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
432 CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
433 "ocd_grant: %d\n", data->ocd_connect_flags,
434 data->ocd_version, data->ocd_grant);
436 obd->obd_upcall.onu_owner = &sbi->ll_lco;
437 obd->obd_upcall.onu_upcall = cl_ocd_update;
439 data->ocd_brw_size = DT_MAX_BRW_SIZE;
441 err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
444 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
445 "recovery, of which this client is not a "
446 "part. Please wait for recovery to "
447 "complete, abort, or time out.\n", dt);
450 CERROR("%s: Cannot connect to %s: rc = %d\n",
451 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
455 sbi->ll_dt_exp->exp_connect_data = *data;
457 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
458 LUSTRE_SEQ_METADATA);
460 CERROR("%s: Can't init data layer FID infrastructure, "
461 "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
465 mutex_lock(&sbi->ll_lco.lco_lock);
466 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
467 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
468 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
469 mutex_unlock(&sbi->ll_lco.lco_lock);
471 fid_zero(&sbi->ll_root_fid);
472 err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
474 CERROR("cannot mds_connect: rc = %d\n", err);
477 if (!fid_is_sane(&sbi->ll_root_fid)) {
478 CERROR("%s: Invalid root fid "DFID" during mount\n",
479 sbi->ll_md_exp->exp_obd->obd_name,
480 PFID(&sbi->ll_root_fid));
484 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
486 sb->s_op = &lustre_super_operations;
487 #if THREAD_SIZE >= 8192 /*b=17630*/
488 sb->s_export_op = &lustre_export_operations;
492 * XXX: move this to after cbd setup? */
493 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
494 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
495 valid |= OBD_MD_FLRMTPERM;
496 else if (sbi->ll_flags & LL_SBI_ACL)
497 valid |= OBD_MD_FLACL;
499 op_data = kzalloc(sizeof(*op_data), GFP_NOFS);
505 op_data->op_fid1 = sbi->ll_root_fid;
506 op_data->op_mode = 0;
507 op_data->op_capa1 = oc;
508 op_data->op_valid = valid;
510 err = md_getattr(sbi->ll_md_exp, op_data, &request);
513 OBD_FREE_PTR(op_data);
515 CERROR("%s: md_getattr failed for root: rc = %d\n",
516 sbi->ll_md_exp->exp_obd->obd_name, err);
520 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
521 sbi->ll_md_exp, &lmd);
523 CERROR("failed to understand root inode md: rc = %d\n", err);
524 ptlrpc_req_finished(request);
528 LASSERT(fid_is_sane(&sbi->ll_root_fid));
529 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
530 sbi->ll_flags & LL_SBI_32BIT_API),
532 md_free_lustre_md(sbi->ll_md_exp, &lmd);
533 ptlrpc_req_finished(request);
535 if (root == NULL || IS_ERR(root)) {
537 obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
538 #ifdef CONFIG_FS_POSIX_ACL
540 posix_acl_release(lmd.posix_acl);
541 lmd.posix_acl = NULL;
544 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
546 CERROR("lustre_lite: bad iget4 for root\n");
550 err = ll_close_thread_start(&sbi->ll_lcq);
552 CERROR("cannot start close thread: rc %d\n", err);
556 #ifdef CONFIG_FS_POSIX_ACL
557 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
558 rct_init(&sbi->ll_rct);
559 et_init(&sbi->ll_et);
563 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
564 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
565 KEY_CHECKSUM, sizeof(checksum), &checksum,
569 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
570 KEY_CACHE_SET, sizeof(sbi->ll_cache),
571 &sbi->ll_cache, NULL);
573 sb->s_root = d_make_root(root);
574 if (sb->s_root == NULL) {
575 CERROR("%s: can't make root dentry\n",
576 ll_get_fsname(sb, NULL, 0));
581 sbi->ll_sdev_orig = sb->s_dev;
583 /* We set sb->s_dev equal on all lustre clients in order to support
584 * NFS export clustering. NFSD requires that the FSID be the same
586 /* s_dev is also used in lt_compare() to compare two fs, but that is
587 * only a node-local comparison. */
588 uuid = obd_get_uuid(sbi->ll_md_exp);
590 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
591 get_uuid2fsid(uuid->uuid, strlen(uuid->uuid), &sbi->ll_fsid);
603 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
605 obd_disconnect(sbi->ll_dt_exp);
606 sbi->ll_dt_exp = NULL;
607 /* Make sure all OScs are gone, since cl_cache is accessing sbi. */
608 obd_zombie_barrier();
610 obd_fid_fini(sbi->ll_md_exp->exp_obd);
612 obd_disconnect(sbi->ll_md_exp);
613 sbi->ll_md_exp = NULL;
619 lprocfs_unregister_mountpoint(sbi);
623 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
627 *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
629 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
630 KEY_MAX_EASIZE, &size, lmmsize, NULL);
632 CERROR("Get max mdsize error rc %d \n", rc);
637 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
642 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
643 KEY_DEFAULT_EASIZE, &size, lmmsize, NULL);
645 CERROR("Get default mdsize error rc %d\n", rc);
650 int ll_get_max_cookiesize(struct ll_sb_info *sbi, int *lmmsize)
655 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_COOKIESIZE),
656 KEY_MAX_COOKIESIZE, &size, lmmsize, NULL);
658 CERROR("Get max cookiesize error rc %d\n", rc);
663 int ll_get_default_cookiesize(struct ll_sb_info *sbi, int *lmmsize)
668 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_COOKIESIZE),
669 KEY_DEFAULT_COOKIESIZE, &size, lmmsize, NULL);
671 CERROR("Get default cookiesize error rc %d\n", rc);
676 static void ll_dump_inode(struct inode *inode)
678 struct ll_d_hlist_node *tmp;
679 int dentry_count = 0;
681 LASSERT(inode != NULL);
683 ll_d_hlist_for_each(tmp, &inode->i_dentry)
686 CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
687 inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
688 inode->i_mode, atomic_read(&inode->i_count), dentry_count);
691 void lustre_dump_dentry(struct dentry *dentry, int recur)
693 struct list_head *tmp;
696 LASSERT(dentry != NULL);
698 list_for_each(tmp, &dentry->d_subdirs)
701 CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
702 " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
703 dentry->d_name.len, dentry->d_name.name,
704 dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
705 dentry->d_parent, dentry->d_inode, d_count(dentry),
706 dentry->d_flags, dentry->d_fsdata, subdirs);
707 if (dentry->d_inode != NULL)
708 ll_dump_inode(dentry->d_inode);
713 list_for_each(tmp, &dentry->d_subdirs) {
714 struct dentry *d = list_entry(tmp, struct dentry, d_u.d_child);
715 lustre_dump_dentry(d, recur - 1);
719 static void client_common_put_super(struct super_block *sb)
721 struct ll_sb_info *sbi = ll_s2sbi(sb);
723 #ifdef CONFIG_FS_POSIX_ACL
724 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
725 et_fini(&sbi->ll_et);
726 rct_fini(&sbi->ll_rct);
730 ll_close_thread_shutdown(sbi->ll_lcq);
734 list_del(&sbi->ll_conn_chain);
736 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
737 obd_disconnect(sbi->ll_dt_exp);
738 sbi->ll_dt_exp = NULL;
739 /* wait till all OSCs are gone, since cl_cache is accessing sbi.
741 obd_zombie_barrier();
743 lprocfs_unregister_mountpoint(sbi);
745 obd_fid_fini(sbi->ll_md_exp->exp_obd);
746 obd_disconnect(sbi->ll_md_exp);
747 sbi->ll_md_exp = NULL;
750 void ll_kill_super(struct super_block *sb)
752 struct ll_sb_info *sbi;
755 if (!(sb->s_flags & MS_ACTIVE))
759 /* we need to restore s_dev from changed for clustered NFS before put_super
760 * because new kernels have cached s_dev and change sb->s_dev in
761 * put_super not affected real removing devices */
763 sb->s_dev = sbi->ll_sdev_orig;
764 sbi->ll_umounting = 1;
768 static inline int ll_set_opt(const char *opt, char *data, int fl)
770 if (strncmp(opt, data, strlen(opt)) != 0)
776 /* non-client-specific mount options are parsed in lmd_parse */
777 static int ll_options(char *options, int *flags)
780 char *s1 = options, *s2;
785 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
788 CDEBUG(D_SUPER, "next opt=%s\n", s1);
789 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
794 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
799 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
804 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
809 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
814 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
819 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 5, 50, 0)
820 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
822 /* Ignore deprecated mount option. The client will
823 * always try to mount with ACL support, whether this
824 * is used depends on whether server supports it. */
825 LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
826 "mount option 'acl'.\n");
829 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
831 LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
832 "mount option 'noacl'.\n");
836 #warning "{no}acl options have been deprecated since 1.8, please remove them"
838 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
843 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
848 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
854 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
859 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
864 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
869 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
874 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
879 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
884 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
889 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
894 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
899 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
904 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
910 s2 = strchr(s1, ',');
918 void ll_lli_init(struct ll_inode_info *lli)
920 lli->lli_inode_magic = LLI_INODE_MAGIC;
922 lli->lli_ioepoch = 0;
923 lli->lli_maxbytes = MAX_LFS_FILESIZE;
924 spin_lock_init(&lli->lli_lock);
925 lli->lli_posix_acl = NULL;
926 lli->lli_remote_perms = NULL;
927 mutex_init(&lli->lli_rmtperm_mutex);
928 /* Do not set lli_fid, it has been initialized already. */
929 fid_zero(&lli->lli_pfid);
930 INIT_LIST_HEAD(&lli->lli_close_list);
931 INIT_LIST_HEAD(&lli->lli_oss_capas);
932 atomic_set(&lli->lli_open_count, 0);
933 lli->lli_mds_capa = NULL;
934 lli->lli_rmtperm_time = 0;
935 lli->lli_pending_och = NULL;
936 lli->lli_mds_read_och = NULL;
937 lli->lli_mds_write_och = NULL;
938 lli->lli_mds_exec_och = NULL;
939 lli->lli_open_fd_read_count = 0;
940 lli->lli_open_fd_write_count = 0;
941 lli->lli_open_fd_exec_count = 0;
942 mutex_init(&lli->lli_och_mutex);
943 spin_lock_init(&lli->lli_agl_lock);
944 lli->lli_has_smd = false;
945 spin_lock_init(&lli->lli_layout_lock);
946 ll_layout_version_set(lli, LL_LAYOUT_GEN_NONE);
947 lli->lli_clob = NULL;
949 init_rwsem(&lli->lli_xattrs_list_rwsem);
950 mutex_init(&lli->lli_xattrs_enq_lock);
952 LASSERT(lli->lli_vfs_inode.i_mode != 0);
953 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
954 mutex_init(&lli->lli_readdir_mutex);
955 lli->lli_opendir_key = NULL;
957 spin_lock_init(&lli->lli_sa_lock);
958 lli->lli_opendir_pid = 0;
960 mutex_init(&lli->lli_size_mutex);
961 lli->lli_symlink_name = NULL;
962 init_rwsem(&lli->lli_trunc_sem);
963 mutex_init(&lli->lli_write_mutex);
964 init_rwsem(&lli->lli_glimpse_sem);
965 lli->lli_glimpse_time = 0;
966 INIT_LIST_HEAD(&lli->lli_agl_list);
967 lli->lli_agl_index = 0;
968 lli->lli_async_rc = 0;
970 mutex_init(&lli->lli_layout_mutex);
973 static inline int ll_bdi_register(struct backing_dev_info *bdi)
975 static atomic_t ll_bdi_num = ATOMIC_INIT(0);
977 bdi->name = "lustre";
978 return bdi_register(bdi, NULL, "lustre-%d",
979 atomic_inc_return(&ll_bdi_num));
982 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
984 struct lustre_profile *lprof = NULL;
985 struct lustre_sb_info *lsi = s2lsi(sb);
986 struct ll_sb_info *sbi;
987 char *dt = NULL, *md = NULL;
988 char *profilenm = get_profile_name(sb);
989 struct config_llog_instance *cfg;
990 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
991 const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
994 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
996 cfg = kzalloc(sizeof(*cfg), GFP_NOFS);
1000 try_module_get(THIS_MODULE);
1002 /* client additional sb info */
1003 lsi->lsi_llsbi = sbi = ll_init_sbi();
1005 module_put(THIS_MODULE);
1010 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1014 err = bdi_init(&lsi->lsi_bdi);
1017 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1018 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1019 err = ll_bdi_register(&lsi->lsi_bdi);
1023 sb->s_bdi = &lsi->lsi_bdi;
1024 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1025 sb->s_d_op = &ll_d_ops;
1027 /* Generate a string unique to this super, in case some joker tries
1028 to mount the same fs at two mount points.
1029 Use the address of the super itself.*/
1030 cfg->cfg_instance = sb;
1031 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1032 cfg->cfg_callback = class_config_llog_handler;
1033 /* set up client obds */
1034 err = lustre_process_log(sb, profilenm, cfg);
1036 CERROR("Unable to process log: %d\n", err);
1040 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1041 lprof = class_get_profile(profilenm);
1042 if (lprof == NULL) {
1043 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1044 " read from the MGS. Does that filesystem "
1045 "exist?\n", profilenm);
1049 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1050 lprof->lp_md, lprof->lp_dt);
1052 dt = kzalloc(strlen(lprof->lp_dt) + instlen + 2, GFP_NOFS);
1057 sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1059 md = kzalloc(strlen(lprof->lp_md) + instlen + 2, GFP_NOFS);
1064 sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1066 /* connections, registrations, sb setup */
1067 err = client_common_fill_super(sb, md, dt, mnt);
1071 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1073 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1076 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1077 LCONSOLE_WARN("Mounted %s\n", profilenm);
1081 } /* ll_fill_super */
1083 void ll_put_super(struct super_block *sb)
1085 struct config_llog_instance cfg, params_cfg;
1086 struct obd_device *obd;
1087 struct lustre_sb_info *lsi = s2lsi(sb);
1088 struct ll_sb_info *sbi = ll_s2sbi(sb);
1089 char *profilenm = get_profile_name(sb);
1090 int next, force = 1;
1092 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1094 ll_print_capa_stat(sbi);
1096 cfg.cfg_instance = sb;
1097 lustre_end_log(sb, profilenm, &cfg);
1099 params_cfg.cfg_instance = sb;
1100 lustre_end_log(sb, PARAMS_FILENAME, ¶ms_cfg);
1102 if (sbi->ll_md_exp) {
1103 obd = class_exp2obd(sbi->ll_md_exp);
1105 force = obd->obd_force;
1108 /* We need to set force before the lov_disconnect in
1109 lustre_common_put_super, since l_d cleans up osc's as well. */
1112 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1114 obd->obd_force = force;
1119 /* Only if client_common_fill_super succeeded */
1120 client_common_put_super(sb);
1124 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1125 class_manual_cleanup(obd);
1128 if (sbi->ll_flags & LL_SBI_VERBOSE)
1129 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1132 class_del_profile(profilenm);
1134 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1135 bdi_destroy(&lsi->lsi_bdi);
1136 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1140 lsi->lsi_llsbi = NULL;
1142 lustre_common_put_super(sb);
1144 module_put(THIS_MODULE);
1145 } /* client_put_super */
1147 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1149 struct inode *inode = NULL;
1151 /* NOTE: we depend on atomic igrab() -bzzz */
1152 lock_res_and_lock(lock);
1153 if (lock->l_resource->lr_lvb_inode) {
1154 struct ll_inode_info *lli;
1155 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1156 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1157 inode = igrab(lock->l_resource->lr_lvb_inode);
1159 inode = lock->l_resource->lr_lvb_inode;
1160 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1161 D_WARNING, lock, "lr_lvb_inode %p is "
1162 "bogus: magic %08x",
1163 lock->l_resource->lr_lvb_inode,
1164 lli->lli_inode_magic);
1168 unlock_res_and_lock(lock);
1172 void ll_clear_inode(struct inode *inode)
1174 struct ll_inode_info *lli = ll_i2info(inode);
1175 struct ll_sb_info *sbi = ll_i2sbi(inode);
1177 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1178 inode->i_generation, inode);
1180 if (S_ISDIR(inode->i_mode)) {
1181 /* these should have been cleared in ll_file_release */
1182 LASSERT(lli->lli_opendir_key == NULL);
1183 LASSERT(lli->lli_sai == NULL);
1184 LASSERT(lli->lli_opendir_pid == 0);
1187 spin_lock(&lli->lli_lock);
1188 ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1189 spin_unlock(&lli->lli_lock);
1190 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1192 LASSERT(!lli->lli_open_fd_write_count);
1193 LASSERT(!lli->lli_open_fd_read_count);
1194 LASSERT(!lli->lli_open_fd_exec_count);
1196 if (lli->lli_mds_write_och)
1197 ll_md_real_close(inode, FMODE_WRITE);
1198 if (lli->lli_mds_exec_och)
1199 ll_md_real_close(inode, FMODE_EXEC);
1200 if (lli->lli_mds_read_och)
1201 ll_md_real_close(inode, FMODE_READ);
1203 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1204 OBD_FREE(lli->lli_symlink_name,
1205 strlen(lli->lli_symlink_name) + 1);
1206 lli->lli_symlink_name = NULL;
1209 ll_xattr_cache_destroy(inode);
1211 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1212 LASSERT(lli->lli_posix_acl == NULL);
1213 if (lli->lli_remote_perms) {
1214 free_rmtperm_hash(lli->lli_remote_perms);
1215 lli->lli_remote_perms = NULL;
1218 #ifdef CONFIG_FS_POSIX_ACL
1219 else if (lli->lli_posix_acl) {
1220 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1221 LASSERT(lli->lli_remote_perms == NULL);
1222 posix_acl_release(lli->lli_posix_acl);
1223 lli->lli_posix_acl = NULL;
1226 lli->lli_inode_magic = LLI_INODE_DEAD;
1228 ll_clear_inode_capas(inode);
1229 if (!S_ISDIR(inode->i_mode))
1230 LASSERT(list_empty(&lli->lli_agl_list));
1233 * XXX This has to be done before lsm is freed below, because
1234 * cl_object still uses inode lsm.
1236 cl_inode_fini(inode);
1237 lli->lli_has_smd = false;
1240 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data,
1241 struct md_open_data **mod)
1243 struct lustre_md md;
1244 struct inode *inode = dentry->d_inode;
1245 struct ll_sb_info *sbi = ll_i2sbi(inode);
1246 struct ptlrpc_request *request = NULL;
1249 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1250 LUSTRE_OPC_ANY, NULL);
1251 if (IS_ERR(op_data))
1252 return PTR_ERR(op_data);
1254 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1257 ptlrpc_req_finished(request);
1258 if (rc == -ENOENT) {
1260 /* Unlinked special device node? Or just a race?
1261 * Pretend we done everything. */
1262 if (!S_ISREG(inode->i_mode) &&
1263 !S_ISDIR(inode->i_mode)) {
1264 ia_valid = op_data->op_attr.ia_valid;
1265 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1266 rc = simple_setattr(dentry, &op_data->op_attr);
1267 op_data->op_attr.ia_valid = ia_valid;
1269 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1270 CERROR("md_setattr fails: rc = %d\n", rc);
1275 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1276 sbi->ll_md_exp, &md);
1278 ptlrpc_req_finished(request);
1282 ia_valid = op_data->op_attr.ia_valid;
1283 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1284 * cache is not cleared yet. */
1285 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1286 rc = simple_setattr(dentry, &op_data->op_attr);
1287 op_data->op_attr.ia_valid = ia_valid;
1289 /* Extract epoch data if obtained. */
1290 op_data->op_handle = md.body->handle;
1291 op_data->op_ioepoch = md.body->ioepoch;
1293 ll_update_inode(inode, &md);
1294 ptlrpc_req_finished(request);
1299 /* Close IO epoch and send Size-on-MDS attribute update. */
1300 static int ll_setattr_done_writing(struct inode *inode,
1301 struct md_op_data *op_data,
1302 struct md_open_data *mod)
1304 struct ll_inode_info *lli = ll_i2info(inode);
1307 LASSERT(op_data != NULL);
1308 if (!S_ISREG(inode->i_mode))
1311 CDEBUG(D_INODE, "Epoch %llu closed on "DFID" for truncate\n",
1312 op_data->op_ioepoch, PFID(&lli->lli_fid));
1314 op_data->op_flags = MF_EPOCH_CLOSE;
1315 ll_done_writing_attr(inode, op_data);
1316 ll_pack_inode2opdata(inode, op_data, NULL);
1318 rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1319 if (rc == -EAGAIN) {
1320 /* MDS has instructed us to obtain Size-on-MDS attribute
1321 * from OSTs and send setattr to back to MDS. */
1322 rc = ll_som_update(inode, op_data);
1324 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1330 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1332 struct obd_capa *capa;
1335 if (attr->ia_valid & ATTR_SIZE)
1336 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1338 capa = ll_mdscapa_get(inode);
1340 rc = cl_setattr_ost(inode, attr, capa);
1342 if (attr->ia_valid & ATTR_SIZE)
1343 ll_truncate_free_capa(capa);
1351 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1352 * object(s) determine the file size and mtime. Otherwise, the MDS will
1353 * keep these values until such a time that objects are allocated for it.
1354 * We do the MDS operations first, as it is checking permissions for us.
1355 * We don't to the MDS RPC if there is nothing that we want to store there,
1356 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1357 * going to do an RPC anyways.
1359 * If we are doing a truncate, we will send the mtime and ctime updates
1360 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1361 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1364 * In case of HSMimport, we only set attr on MDS.
1366 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1368 struct inode *inode = dentry->d_inode;
1369 struct ll_inode_info *lli = ll_i2info(inode);
1370 struct md_op_data *op_data = NULL;
1371 struct md_open_data *mod = NULL;
1372 bool file_is_released = false;
1373 int rc = 0, rc1 = 0;
1376 "%s: setattr inode %p/fid:"DFID
1377 " from %llu to %llu, valid %x, hsm_import %d\n",
1378 ll_get_fsname(inode->i_sb, NULL, 0), inode,
1379 PFID(&lli->lli_fid), i_size_read(inode), attr->ia_size,
1380 attr->ia_valid, hsm_import);
1382 if (attr->ia_valid & ATTR_SIZE) {
1383 /* Check new size against VFS/VM file size limit and rlimit */
1384 rc = inode_newsize_ok(inode, attr->ia_size);
1388 /* The maximum Lustre file size is variable, based on the
1389 * OST maximum object size and number of stripes. This
1390 * needs another check in addition to the VFS check above. */
1391 if (attr->ia_size > ll_file_maxbytes(inode)) {
1392 CDEBUG(D_INODE, "file "DFID" too large %llu > %llu\n",
1393 PFID(&lli->lli_fid), attr->ia_size,
1394 ll_file_maxbytes(inode));
1398 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1401 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1402 if (attr->ia_valid & TIMES_SET_FLAGS) {
1403 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1404 !capable(CFS_CAP_FOWNER))
1408 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1409 if (attr->ia_valid & ATTR_CTIME) {
1410 attr->ia_ctime = CURRENT_TIME;
1411 attr->ia_valid |= ATTR_CTIME_SET;
1413 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1414 (attr->ia_valid & ATTR_ATIME)) {
1415 attr->ia_atime = CURRENT_TIME;
1416 attr->ia_valid |= ATTR_ATIME_SET;
1418 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1419 (attr->ia_valid & ATTR_MTIME)) {
1420 attr->ia_mtime = CURRENT_TIME;
1421 attr->ia_valid |= ATTR_MTIME_SET;
1424 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1425 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1426 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1429 /* If we are changing file size, file content is modified, flag it. */
1430 if (attr->ia_valid & ATTR_SIZE) {
1431 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1432 spin_lock(&lli->lli_lock);
1433 lli->lli_flags |= LLIF_DATA_MODIFIED;
1434 spin_unlock(&lli->lli_lock);
1437 /* We always do an MDS RPC, even if we're only changing the size;
1438 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1440 op_data = kzalloc(sizeof(*op_data), GFP_NOFS);
1444 if (!S_ISDIR(inode->i_mode)) {
1445 if (attr->ia_valid & ATTR_SIZE)
1446 inode_dio_write_done(inode);
1447 mutex_unlock(&inode->i_mutex);
1450 memcpy(&op_data->op_attr, attr, sizeof(*attr));
1452 /* Open epoch for truncate. */
1453 if (exp_connect_som(ll_i2mdexp(inode)) &&
1454 (attr->ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1455 op_data->op_flags = MF_EPOCH_OPEN;
1457 /* truncate on a released file must failed with -ENODATA,
1458 * so size must not be set on MDS for released file
1459 * but other attributes must be set
1461 if (S_ISREG(inode->i_mode)) {
1462 struct lov_stripe_md *lsm;
1465 ll_layout_refresh(inode, &gen);
1466 lsm = ccc_inode_lsm_get(inode);
1467 if (lsm && lsm->lsm_pattern & LOV_PATTERN_F_RELEASED)
1468 file_is_released = true;
1469 ccc_inode_lsm_put(inode, lsm);
1472 /* if not in HSM import mode, clear size attr for released file
1473 * we clear the attribute send to MDT in op_data, not the original
1474 * received from caller in attr which is used later to
1475 * decide return code */
1476 if (file_is_released && (attr->ia_valid & ATTR_SIZE) && !hsm_import)
1477 op_data->op_attr.ia_valid &= ~ATTR_SIZE;
1479 rc = ll_md_setattr(dentry, op_data, &mod);
1483 /* truncate failed (only when non HSM import), others succeed */
1484 if (file_is_released) {
1485 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1492 /* RPC to MDT is sent, cancel data modification flag */
1493 if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) {
1494 spin_lock(&lli->lli_lock);
1495 lli->lli_flags &= ~LLIF_DATA_MODIFIED;
1496 spin_unlock(&lli->lli_lock);
1499 ll_ioepoch_open(lli, op_data->op_ioepoch);
1500 if (!S_ISREG(inode->i_mode)) {
1505 if (attr->ia_valid & (ATTR_SIZE |
1506 ATTR_ATIME | ATTR_ATIME_SET |
1507 ATTR_MTIME | ATTR_MTIME_SET))
1508 /* For truncate and utimes sending attributes to OSTs, setting
1509 * mtime/atime to the past will be performed under PW [0:EOF]
1510 * extent lock (new_size:EOF for truncate). It may seem
1511 * excessive to send mtime/atime updates to OSTs when not
1512 * setting times to past, but it is necessary due to possible
1513 * time de-synchronization between MDT inode and OST objects */
1514 if (attr->ia_valid & ATTR_SIZE)
1515 down_write(&lli->lli_trunc_sem);
1516 rc = ll_setattr_ost(inode, attr);
1517 if (attr->ia_valid & ATTR_SIZE)
1518 up_write(&lli->lli_trunc_sem);
1521 if (op_data->op_ioepoch) {
1522 rc1 = ll_setattr_done_writing(inode, op_data, mod);
1526 ll_finish_md_op_data(op_data);
1528 if (!S_ISDIR(inode->i_mode)) {
1529 mutex_lock(&inode->i_mutex);
1530 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1531 inode_dio_wait(inode);
1534 ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1535 LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1540 int ll_setattr(struct dentry *de, struct iattr *attr)
1542 int mode = de->d_inode->i_mode;
1544 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1545 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1546 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1548 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1549 (ATTR_SIZE|ATTR_MODE)) &&
1550 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1551 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1552 !(attr->ia_mode & S_ISGID))))
1553 attr->ia_valid |= ATTR_FORCE;
1555 if ((attr->ia_valid & ATTR_MODE) &&
1557 !(attr->ia_mode & S_ISUID) &&
1558 !(attr->ia_valid & ATTR_KILL_SUID))
1559 attr->ia_valid |= ATTR_KILL_SUID;
1561 if ((attr->ia_valid & ATTR_MODE) &&
1562 ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1563 !(attr->ia_mode & S_ISGID) &&
1564 !(attr->ia_valid & ATTR_KILL_SGID))
1565 attr->ia_valid |= ATTR_KILL_SGID;
1567 return ll_setattr_raw(de, attr, false);
1570 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1571 __u64 max_age, __u32 flags)
1573 struct ll_sb_info *sbi = ll_s2sbi(sb);
1574 struct obd_statfs obd_osfs;
1577 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1579 CERROR("md_statfs fails: rc = %d\n", rc);
1583 osfs->os_type = sb->s_magic;
1585 CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1586 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,
1589 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1590 flags |= OBD_STATFS_NODELAY;
1592 rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1594 CERROR("obd_statfs fails: rc = %d\n", rc);
1598 CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1599 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1602 osfs->os_bsize = obd_osfs.os_bsize;
1603 osfs->os_blocks = obd_osfs.os_blocks;
1604 osfs->os_bfree = obd_osfs.os_bfree;
1605 osfs->os_bavail = obd_osfs.os_bavail;
1607 /* If we don't have as many objects free on the OST as inodes
1608 * on the MDS, we reduce the total number of inodes to
1609 * compensate, so that the "inodes in use" number is correct.
1611 if (obd_osfs.os_ffree < osfs->os_ffree) {
1612 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1614 osfs->os_ffree = obd_osfs.os_ffree;
1619 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1621 struct super_block *sb = de->d_sb;
1622 struct obd_statfs osfs;
1625 CDEBUG(D_VFSTRACE, "VFS Op: at %llu jiffies\n", get_jiffies_64());
1626 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1628 /* Some amount of caching on the client is allowed */
1629 rc = ll_statfs_internal(sb, &osfs,
1630 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1635 statfs_unpack(sfs, &osfs);
1637 /* We need to downshift for all 32-bit kernels, because we can't
1638 * tell if the kernel is being called via sys_statfs64() or not.
1639 * Stop before overflowing f_bsize - in which case it is better
1640 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1641 if (sizeof(long) < 8) {
1642 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1645 osfs.os_blocks >>= 1;
1646 osfs.os_bfree >>= 1;
1647 osfs.os_bavail >>= 1;
1651 sfs->f_blocks = osfs.os_blocks;
1652 sfs->f_bfree = osfs.os_bfree;
1653 sfs->f_bavail = osfs.os_bavail;
1654 sfs->f_fsid = ll_s2sbi(sb)->ll_fsid;
1658 void ll_inode_size_lock(struct inode *inode)
1660 struct ll_inode_info *lli;
1662 LASSERT(!S_ISDIR(inode->i_mode));
1664 lli = ll_i2info(inode);
1665 mutex_lock(&lli->lli_size_mutex);
1668 void ll_inode_size_unlock(struct inode *inode)
1670 struct ll_inode_info *lli;
1672 lli = ll_i2info(inode);
1673 mutex_unlock(&lli->lli_size_mutex);
1676 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1678 struct ll_inode_info *lli = ll_i2info(inode);
1679 struct mdt_body *body = md->body;
1680 struct lov_stripe_md *lsm = md->lsm;
1681 struct ll_sb_info *sbi = ll_i2sbi(inode);
1683 LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1685 if (!lli->lli_has_smd &&
1686 !(sbi->ll_flags & LL_SBI_LAYOUT_LOCK))
1687 cl_file_inode_init(inode, md);
1689 lli->lli_maxbytes = lsm->lsm_maxbytes;
1690 if (lli->lli_maxbytes > MAX_LFS_FILESIZE)
1691 lli->lli_maxbytes = MAX_LFS_FILESIZE;
1694 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1695 if (body->valid & OBD_MD_FLRMTPERM)
1696 ll_update_remote_perm(inode, md->remote_perm);
1698 #ifdef CONFIG_FS_POSIX_ACL
1699 else if (body->valid & OBD_MD_FLACL) {
1700 spin_lock(&lli->lli_lock);
1701 if (lli->lli_posix_acl)
1702 posix_acl_release(lli->lli_posix_acl);
1703 lli->lli_posix_acl = md->posix_acl;
1704 spin_unlock(&lli->lli_lock);
1707 inode->i_ino = cl_fid_build_ino(&body->fid1,
1708 sbi->ll_flags & LL_SBI_32BIT_API);
1709 inode->i_generation = cl_fid_build_gen(&body->fid1);
1711 if (body->valid & OBD_MD_FLATIME) {
1712 if (body->atime > LTIME_S(inode->i_atime))
1713 LTIME_S(inode->i_atime) = body->atime;
1714 lli->lli_lvb.lvb_atime = body->atime;
1716 if (body->valid & OBD_MD_FLMTIME) {
1717 if (body->mtime > LTIME_S(inode->i_mtime)) {
1718 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to %llu\n",
1719 inode->i_ino, LTIME_S(inode->i_mtime),
1721 LTIME_S(inode->i_mtime) = body->mtime;
1723 lli->lli_lvb.lvb_mtime = body->mtime;
1725 if (body->valid & OBD_MD_FLCTIME) {
1726 if (body->ctime > LTIME_S(inode->i_ctime))
1727 LTIME_S(inode->i_ctime) = body->ctime;
1728 lli->lli_lvb.lvb_ctime = body->ctime;
1730 if (body->valid & OBD_MD_FLMODE)
1731 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1732 if (body->valid & OBD_MD_FLTYPE)
1733 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1734 LASSERT(inode->i_mode != 0);
1735 if (S_ISREG(inode->i_mode)) {
1736 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1738 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1740 if (body->valid & OBD_MD_FLUID)
1741 inode->i_uid = make_kuid(&init_user_ns, body->uid);
1742 if (body->valid & OBD_MD_FLGID)
1743 inode->i_gid = make_kgid(&init_user_ns, body->gid);
1744 if (body->valid & OBD_MD_FLFLAGS)
1745 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1746 if (body->valid & OBD_MD_FLNLINK)
1747 set_nlink(inode, body->nlink);
1748 if (body->valid & OBD_MD_FLRDEV)
1749 inode->i_rdev = old_decode_dev(body->rdev);
1751 if (body->valid & OBD_MD_FLID) {
1752 /* FID shouldn't be changed! */
1753 if (fid_is_sane(&lli->lli_fid)) {
1754 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1755 "Trying to change FID "DFID
1756 " to the "DFID", inode %lu/%u(%p)\n",
1757 PFID(&lli->lli_fid), PFID(&body->fid1),
1758 inode->i_ino, inode->i_generation, inode);
1760 lli->lli_fid = body->fid1;
1763 LASSERT(fid_seq(&lli->lli_fid) != 0);
1765 if (body->valid & OBD_MD_FLSIZE) {
1766 if (exp_connect_som(ll_i2mdexp(inode)) &&
1767 S_ISREG(inode->i_mode)) {
1768 struct lustre_handle lockh;
1771 /* As it is possible a blocking ast has been processed
1772 * by this time, we need to check there is an UPDATE
1773 * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1775 mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1776 &lockh, LDLM_FL_CBPENDING,
1780 if (lli->lli_flags & (LLIF_DONE_WRITING |
1781 LLIF_EPOCH_PENDING |
1783 CERROR("ino %lu flags %u still has "
1784 "size authority! do not trust "
1785 "the size got from MDS\n",
1786 inode->i_ino, lli->lli_flags);
1788 /* Use old size assignment to avoid
1789 * deadlock bz14138 & bz14326 */
1790 i_size_write(inode, body->size);
1791 spin_lock(&lli->lli_lock);
1792 lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1793 spin_unlock(&lli->lli_lock);
1795 ldlm_lock_decref(&lockh, mode);
1798 /* Use old size assignment to avoid
1799 * deadlock bz14138 & bz14326 */
1800 i_size_write(inode, body->size);
1802 CDEBUG(D_VFSTRACE, "inode=%lu, updating i_size %llu\n",
1803 inode->i_ino, (unsigned long long)body->size);
1806 if (body->valid & OBD_MD_FLBLOCKS)
1807 inode->i_blocks = body->blocks;
1810 if (body->valid & OBD_MD_FLMDSCAPA) {
1811 LASSERT(md->mds_capa);
1812 ll_add_capa(inode, md->mds_capa);
1814 if (body->valid & OBD_MD_FLOSSCAPA) {
1815 LASSERT(md->oss_capa);
1816 ll_add_capa(inode, md->oss_capa);
1819 if (body->valid & OBD_MD_TSTATE) {
1820 if (body->t_state & MS_RESTORE)
1821 lli->lli_flags |= LLIF_FILE_RESTORING;
1825 void ll_read_inode2(struct inode *inode, void *opaque)
1827 struct lustre_md *md = opaque;
1828 struct ll_inode_info *lli = ll_i2info(inode);
1830 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1831 PFID(&lli->lli_fid), inode);
1833 LASSERT(!lli->lli_has_smd);
1835 /* Core attributes from the MDS first. This is a new inode, and
1836 * the VFS doesn't zero times in the core inode so we have to do
1837 * it ourselves. They will be overwritten by either MDS or OST
1838 * attributes - we just need to make sure they aren't newer. */
1839 LTIME_S(inode->i_mtime) = 0;
1840 LTIME_S(inode->i_atime) = 0;
1841 LTIME_S(inode->i_ctime) = 0;
1843 ll_update_inode(inode, md);
1845 /* OIDEBUG(inode); */
1847 /* initializing backing dev info. */
1848 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1851 if (S_ISREG(inode->i_mode)) {
1852 struct ll_sb_info *sbi = ll_i2sbi(inode);
1853 inode->i_op = &ll_file_inode_operations;
1854 inode->i_fop = sbi->ll_fop;
1855 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1856 } else if (S_ISDIR(inode->i_mode)) {
1857 inode->i_op = &ll_dir_inode_operations;
1858 inode->i_fop = &ll_dir_operations;
1859 } else if (S_ISLNK(inode->i_mode)) {
1860 inode->i_op = &ll_fast_symlink_inode_operations;
1862 inode->i_op = &ll_special_inode_operations;
1864 init_special_inode(inode, inode->i_mode,
1869 void ll_delete_inode(struct inode *inode)
1871 struct cl_inode_info *lli = cl_i2info(inode);
1873 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
1874 /* discard all dirty pages before truncating them, required by
1875 * osc_extent implementation at LU-1030. */
1876 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF,
1877 CL_FSYNC_DISCARD, 1);
1879 truncate_inode_pages_final(&inode->i_data);
1881 /* Workaround for LU-118 */
1882 if (inode->i_data.nrpages) {
1883 TREE_READ_LOCK_IRQ(&inode->i_data);
1884 TREE_READ_UNLOCK_IRQ(&inode->i_data);
1885 LASSERTF(inode->i_data.nrpages == 0,
1886 "inode=%lu/%u(%p) nrpages=%lu, see "
1887 "http://jira.whamcloud.com/browse/LU-118\n",
1888 inode->i_ino, inode->i_generation, inode,
1889 inode->i_data.nrpages);
1891 /* Workaround end */
1893 ll_clear_inode(inode);
1897 int ll_iocontrol(struct inode *inode, struct file *file,
1898 unsigned int cmd, unsigned long arg)
1900 struct ll_sb_info *sbi = ll_i2sbi(inode);
1901 struct ptlrpc_request *req = NULL;
1905 case FSFILT_IOC_GETFLAGS: {
1906 struct mdt_body *body;
1907 struct md_op_data *op_data;
1909 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
1910 0, 0, LUSTRE_OPC_ANY,
1912 if (IS_ERR(op_data))
1913 return PTR_ERR(op_data);
1915 op_data->op_valid = OBD_MD_FLFLAGS;
1916 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
1917 ll_finish_md_op_data(op_data);
1919 CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1923 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1925 flags = body->flags;
1927 ptlrpc_req_finished(req);
1929 return put_user(flags, (int *)arg);
1931 case FSFILT_IOC_SETFLAGS: {
1932 struct lov_stripe_md *lsm;
1933 struct obd_info oinfo = { { { 0 } } };
1934 struct md_op_data *op_data;
1936 if (get_user(flags, (int *)arg))
1939 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1940 LUSTRE_OPC_ANY, NULL);
1941 if (IS_ERR(op_data))
1942 return PTR_ERR(op_data);
1944 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1945 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1946 rc = md_setattr(sbi->ll_md_exp, op_data,
1947 NULL, 0, NULL, 0, &req, NULL);
1948 ll_finish_md_op_data(op_data);
1949 ptlrpc_req_finished(req);
1953 inode->i_flags = ll_ext_to_inode_flags(flags);
1955 lsm = ccc_inode_lsm_get(inode);
1956 if (!lsm_has_objects(lsm)) {
1957 ccc_inode_lsm_put(inode, lsm);
1961 OBDO_ALLOC(oinfo.oi_oa);
1963 ccc_inode_lsm_put(inode, lsm);
1967 oinfo.oi_oa->o_oi = lsm->lsm_oi;
1968 oinfo.oi_oa->o_flags = flags;
1969 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1971 oinfo.oi_capa = ll_mdscapa_get(inode);
1972 obdo_set_parent_fid(oinfo.oi_oa, &ll_i2info(inode)->lli_fid);
1973 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1974 capa_put(oinfo.oi_capa);
1975 OBDO_FREE(oinfo.oi_oa);
1976 ccc_inode_lsm_put(inode, lsm);
1978 if (rc && rc != -EPERM && rc != -EACCES)
1979 CERROR("osc_setattr_async fails: rc = %d\n", rc);
1990 int ll_flush_ctx(struct inode *inode)
1992 struct ll_sb_info *sbi = ll_i2sbi(inode);
1994 CDEBUG(D_SEC, "flush context for user %d\n",
1995 from_kuid(&init_user_ns, current_uid()));
1997 obd_set_info_async(NULL, sbi->ll_md_exp,
1998 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2000 obd_set_info_async(NULL, sbi->ll_dt_exp,
2001 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2006 /* umount -f client means force down, don't save state */
2007 void ll_umount_begin(struct super_block *sb)
2009 struct ll_sb_info *sbi = ll_s2sbi(sb);
2010 struct obd_device *obd;
2011 struct obd_ioctl_data *ioc_data;
2013 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2014 sb->s_count, atomic_read(&sb->s_active));
2016 obd = class_exp2obd(sbi->ll_md_exp);
2018 CERROR("Invalid MDC connection handle %#llx\n",
2019 sbi->ll_md_exp->exp_handle.h_cookie);
2024 obd = class_exp2obd(sbi->ll_dt_exp);
2026 CERROR("Invalid LOV connection handle %#llx\n",
2027 sbi->ll_dt_exp->exp_handle.h_cookie);
2032 ioc_data = kzalloc(sizeof(*ioc_data), GFP_NOFS);
2034 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2035 sizeof(*ioc_data), ioc_data, NULL);
2037 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2038 sizeof(*ioc_data), ioc_data, NULL);
2040 OBD_FREE_PTR(ioc_data);
2043 /* Really, we'd like to wait until there are no requests outstanding,
2044 * and then continue. For now, we just invalidate the requests,
2045 * schedule() and sleep one second if needed, and hope.
2050 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2052 struct ll_sb_info *sbi = ll_s2sbi(sb);
2053 char *profilenm = get_profile_name(sb);
2057 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2058 read_only = *flags & MS_RDONLY;
2059 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2060 sizeof(KEY_READ_ONLY),
2061 KEY_READ_ONLY, sizeof(read_only),
2064 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2065 profilenm, read_only ?
2066 "read-only" : "read-write", err);
2071 sb->s_flags |= MS_RDONLY;
2073 sb->s_flags &= ~MS_RDONLY;
2075 if (sbi->ll_flags & LL_SBI_VERBOSE)
2076 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2077 read_only ? "read-only" : "read-write");
2082 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2083 struct super_block *sb, struct lookup_intent *it)
2085 struct ll_sb_info *sbi = NULL;
2086 struct lustre_md md;
2089 LASSERT(*inode || sb);
2090 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2091 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2092 sbi->ll_md_exp, &md);
2097 ll_update_inode(*inode, &md);
2099 LASSERT(sb != NULL);
2102 * At this point server returns to client's same fid as client
2103 * generated for creating. So using ->fid1 is okay here.
2105 LASSERT(fid_is_sane(&md.body->fid1));
2107 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1,
2108 sbi->ll_flags & LL_SBI_32BIT_API),
2110 if (*inode == NULL || IS_ERR(*inode)) {
2111 #ifdef CONFIG_FS_POSIX_ACL
2113 posix_acl_release(md.posix_acl);
2114 md.posix_acl = NULL;
2117 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2119 CERROR("new_inode -fatal: rc %d\n", rc);
2124 /* Handling piggyback layout lock.
2125 * Layout lock can be piggybacked by getattr and open request.
2126 * The lsm can be applied to inode only if it comes with a layout lock
2127 * otherwise correct layout may be overwritten, for example:
2128 * 1. proc1: mdt returns a lsm but not granting layout
2129 * 2. layout was changed by another client
2130 * 3. proc2: refresh layout and layout lock granted
2131 * 4. proc1: to apply a stale layout */
2132 if (it != NULL && it->d.lustre.it_lock_mode != 0) {
2133 struct lustre_handle lockh;
2134 struct ldlm_lock *lock;
2136 lockh.cookie = it->d.lustre.it_lock_handle;
2137 lock = ldlm_handle2lock(&lockh);
2138 LASSERT(lock != NULL);
2139 if (ldlm_has_layout(lock)) {
2140 struct cl_object_conf conf;
2142 memset(&conf, 0, sizeof(conf));
2143 conf.coc_opc = OBJECT_CONF_SET;
2144 conf.coc_inode = *inode;
2145 conf.coc_lock = lock;
2146 conf.u.coc_md = &md;
2147 (void)ll_layout_conf(*inode, &conf);
2149 LDLM_LOCK_PUT(lock);
2154 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2155 md_free_lustre_md(sbi->ll_md_exp, &md);
2159 int ll_obd_statfs(struct inode *inode, void *arg)
2161 struct ll_sb_info *sbi = NULL;
2162 struct obd_export *exp;
2164 struct obd_ioctl_data *data = NULL;
2169 if (!inode || !(sbi = ll_i2sbi(inode))) {
2174 rc = obd_ioctl_getdata(&buf, &len, arg);
2179 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2180 !data->ioc_pbuf1 || !data->ioc_pbuf2) {
2185 if (data->ioc_inllen1 != sizeof(__u32) ||
2186 data->ioc_inllen2 != sizeof(__u32) ||
2187 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2188 data->ioc_plen2 != sizeof(struct obd_uuid)) {
2193 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2194 if (type & LL_STATFS_LMV)
2195 exp = sbi->ll_md_exp;
2196 else if (type & LL_STATFS_LOV)
2197 exp = sbi->ll_dt_exp;
2203 flags = (type & LL_STATFS_NODELAY) ? OBD_STATFS_NODELAY : 0;
2204 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, &flags);
2209 obd_ioctl_freedata(buf, len);
2213 int ll_process_config(struct lustre_cfg *lcfg)
2217 struct lprocfs_static_vars lvars;
2221 lprocfs_llite_init_vars(&lvars);
2223 /* The instance name contains the sb: lustre-client-aacfe000 */
2224 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2225 if (!ptr || !*(++ptr))
2227 if (sscanf(ptr, "%lx", &x) != 1)
2230 /* This better be a real Lustre superblock! */
2231 LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2233 /* Note we have not called client_common_fill_super yet, so
2234 proc fns must be able to handle that! */
2235 rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2242 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2243 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2244 struct inode *i1, struct inode *i2,
2245 const char *name, int namelen,
2246 int mode, __u32 opc, void *data)
2248 LASSERT(i1 != NULL);
2250 if (namelen > ll_i2sbi(i1)->ll_namelen)
2251 return ERR_PTR(-ENAMETOOLONG);
2253 if (op_data == NULL)
2254 op_data = kzalloc(sizeof(*op_data), GFP_NOFS);
2256 if (op_data == NULL)
2257 return ERR_PTR(-ENOMEM);
2259 ll_i2gids(op_data->op_suppgids, i1, i2);
2260 op_data->op_fid1 = *ll_inode2fid(i1);
2261 op_data->op_capa1 = ll_mdscapa_get(i1);
2264 op_data->op_fid2 = *ll_inode2fid(i2);
2265 op_data->op_capa2 = ll_mdscapa_get(i2);
2267 fid_zero(&op_data->op_fid2);
2268 op_data->op_capa2 = NULL;
2271 op_data->op_name = name;
2272 op_data->op_namelen = namelen;
2273 op_data->op_mode = mode;
2274 op_data->op_mod_time = get_seconds();
2275 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2276 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2277 op_data->op_cap = cfs_curproc_cap_pack();
2278 op_data->op_bias = 0;
2279 op_data->op_cli_flags = 0;
2280 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2281 filename_is_volatile(name, namelen, NULL))
2282 op_data->op_bias |= MDS_CREATE_VOLATILE;
2283 op_data->op_opc = opc;
2284 op_data->op_mds = 0;
2285 op_data->op_data = data;
2287 /* If the file is being opened after mknod() (normally due to NFS)
2288 * try to use the default stripe data from parent directory for
2289 * allocating OST objects. Try to pass the parent FID to MDS. */
2290 if (opc == LUSTRE_OPC_CREATE && i1 == i2 && S_ISREG(i2->i_mode) &&
2291 !ll_i2info(i2)->lli_has_smd) {
2292 struct ll_inode_info *lli = ll_i2info(i2);
2294 spin_lock(&lli->lli_lock);
2295 if (likely(!lli->lli_has_smd && !fid_is_zero(&lli->lli_pfid)))
2296 op_data->op_fid1 = lli->lli_pfid;
2297 spin_unlock(&lli->lli_lock);
2298 /** We ignore parent's capability temporary. */
2301 /* When called by ll_setattr_raw, file is i1. */
2302 if (LLIF_DATA_MODIFIED & ll_i2info(i1)->lli_flags)
2303 op_data->op_bias |= MDS_DATA_MODIFIED;
2308 void ll_finish_md_op_data(struct md_op_data *op_data)
2310 capa_put(op_data->op_capa1);
2311 capa_put(op_data->op_capa2);
2312 OBD_FREE_PTR(op_data);
2315 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2317 struct ll_sb_info *sbi;
2319 LASSERT((seq != NULL) && (dentry != NULL));
2320 sbi = ll_s2sbi(dentry->d_sb);
2322 if (sbi->ll_flags & LL_SBI_NOLCK)
2323 seq_puts(seq, ",nolock");
2325 if (sbi->ll_flags & LL_SBI_FLOCK)
2326 seq_puts(seq, ",flock");
2328 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2329 seq_puts(seq, ",localflock");
2331 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2332 seq_puts(seq, ",user_xattr");
2334 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2335 seq_puts(seq, ",lazystatfs");
2337 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2338 seq_puts(seq, ",user_fid2path");
2344 * Get obd name by cmd, and copy out to user space
2346 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2348 struct ll_sb_info *sbi = ll_i2sbi(inode);
2349 struct obd_device *obd;
2351 if (cmd == OBD_IOC_GETDTNAME)
2352 obd = class_exp2obd(sbi->ll_dt_exp);
2353 else if (cmd == OBD_IOC_GETMDNAME)
2354 obd = class_exp2obd(sbi->ll_md_exp);
2361 if (copy_to_user((void *)arg, obd->obd_name,
2362 strlen(obd->obd_name) + 1))
2369 * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2370 * fsname will be returned in this buffer; otherwise, a static buffer will be
2371 * used to store the fsname and returned to caller.
2373 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2375 static char fsname_static[MTI_NAME_MAXLEN];
2376 struct lustre_sb_info *lsi = s2lsi(sb);
2381 /* this means the caller wants to use static buffer
2382 * and it doesn't care about race. Usually this is
2383 * in error reporting path */
2384 buf = fsname_static;
2385 buflen = sizeof(fsname_static);
2388 len = strlen(lsi->lsi_lmd->lmd_profile);
2389 ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2390 if (ptr && (strcmp(ptr, "-client") == 0))
2393 if (unlikely(len >= buflen))
2395 strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2401 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2408 p.mnt = current->fs->root.mnt;
2410 path = d_path(&p, buf, bufsize);
2416 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2418 char *buf, *path = NULL;
2419 struct dentry *dentry = NULL;
2420 struct ccc_object *obj = cl_inode2ccc(page->mapping->host);
2422 /* this can be called inside spin lock so use GFP_ATOMIC. */
2423 buf = (char *)__get_free_page(GFP_ATOMIC);
2425 dentry = d_find_alias(page->mapping->host);
2427 path = ll_d_path(dentry, buf, PAGE_SIZE);
2431 "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2432 "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2433 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2434 PFID(&obj->cob_header.coh_lu.loh_fid),
2435 (path && !IS_ERR(path)) ? path : "", ioret);
2441 free_page((unsigned long)buf);