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 dentry *llite_root;
61 struct kset *llite_kset;
64 #define log2(n) ffz(~(n))
67 static struct ll_sb_info *ll_init_sbi(struct super_block *sb)
69 struct ll_sb_info *sbi = NULL;
71 unsigned long lru_page_max;
76 sbi = kzalloc(sizeof(*sbi), GFP_NOFS);
80 spin_lock_init(&sbi->ll_lock);
81 mutex_init(&sbi->ll_lco.lco_lock);
82 spin_lock_init(&sbi->ll_pp_extent_lock);
83 spin_lock_init(&sbi->ll_process_lock);
84 sbi->ll_rw_stats_on = 0;
87 pages = si.totalram - si.totalhigh;
88 if (pages >> (20 - PAGE_CACHE_SHIFT) < 512)
89 lru_page_max = pages / 2;
91 lru_page_max = (pages / 4) * 3;
93 /* initialize lru data */
94 atomic_set(&sbi->ll_cache.ccc_users, 0);
95 sbi->ll_cache.ccc_lru_max = lru_page_max;
96 atomic_set(&sbi->ll_cache.ccc_lru_left, lru_page_max);
97 spin_lock_init(&sbi->ll_cache.ccc_lru_lock);
98 INIT_LIST_HEAD(&sbi->ll_cache.ccc_lru);
100 sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
101 SBI_DEFAULT_READAHEAD_MAX);
102 sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
103 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
104 SBI_DEFAULT_READAHEAD_WHOLE_MAX;
105 INIT_LIST_HEAD(&sbi->ll_conn_chain);
106 INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
108 ll_generate_random_uuid(uuid);
109 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
110 CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
112 sbi->ll_flags |= LL_SBI_VERBOSE;
113 sbi->ll_flags |= LL_SBI_CHECKSUM;
115 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
117 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
118 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
120 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
124 /* metadata statahead is enabled by default */
125 sbi->ll_sa_max = LL_SA_RPC_DEF;
126 atomic_set(&sbi->ll_sa_total, 0);
127 atomic_set(&sbi->ll_sa_wrong, 0);
128 atomic_set(&sbi->ll_agl_total, 0);
129 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
136 static void ll_free_sbi(struct super_block *sb)
138 struct ll_sb_info *sbi = ll_s2sbi(sb);
143 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
144 struct vfsmount *mnt)
146 struct inode *root = NULL;
147 struct ll_sb_info *sbi = ll_s2sbi(sb);
148 struct obd_device *obd;
149 struct obd_statfs *osfs = NULL;
150 struct ptlrpc_request *request = NULL;
151 struct obd_connect_data *data = NULL;
152 struct obd_uuid *uuid;
153 struct md_op_data *op_data;
154 struct lustre_md lmd;
156 int size, err, checksum;
158 obd = class_name2obd(md);
160 CERROR("MD %s: not setup or attached\n", md);
164 data = kzalloc(sizeof(*data), GFP_NOFS);
168 osfs = kzalloc(sizeof(*osfs), GFP_NOFS);
174 if (llite_root != NULL) {
175 err = ldebugfs_register_mountpoint(llite_root, sb, dt, md);
177 CERROR("could not register mount in <debugfs>/lustre/llite\n");
180 /* indicate the features supported by this client */
181 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
182 OBD_CONNECT_ATTRFID |
183 OBD_CONNECT_VERSION | OBD_CONNECT_BRW_SIZE |
184 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
185 OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 |
186 OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR |
187 OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH|
188 OBD_CONNECT_EINPROGRESS |
189 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
190 OBD_CONNECT_LAYOUTLOCK |
191 OBD_CONNECT_PINGLESS |
192 OBD_CONNECT_MAX_EASIZE |
193 OBD_CONNECT_FLOCK_DEAD |
194 OBD_CONNECT_DISP_STRIPE;
196 if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
197 data->ocd_connect_flags |= OBD_CONNECT_SOM;
199 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
200 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
201 #ifdef CONFIG_FS_POSIX_ACL
202 data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK;
205 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
206 /* flag mdc connection as lightweight, only used for test
207 * purpose, use with care */
208 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
210 data->ocd_ibits_known = MDS_INODELOCK_FULL;
211 data->ocd_version = LUSTRE_VERSION_CODE;
213 if (sb->s_flags & MS_RDONLY)
214 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
215 if (sbi->ll_flags & LL_SBI_USER_XATTR)
216 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
218 if (sbi->ll_flags & LL_SBI_FLOCK)
219 sbi->ll_fop = &ll_file_operations_flock;
220 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
221 sbi->ll_fop = &ll_file_operations;
223 sbi->ll_fop = &ll_file_operations_noflock;
226 data->ocd_connect_flags |= OBD_CONNECT_REAL;
227 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
228 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
230 data->ocd_brw_size = MD_MAX_BRW_SIZE;
232 err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid,
235 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",
239 CERROR("cannot connect to %s: rc = %d\n", md, err);
243 sbi->ll_md_exp->exp_connect_data = *data;
245 err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
246 LUSTRE_SEQ_METADATA);
248 CERROR("%s: Can't init metadata layer FID infrastructure, rc = %d\n",
249 sbi->ll_md_exp->exp_obd->obd_name, err);
253 /* For mount, we only need fs info from MDT0, and also in DNE, it
254 * can make sure the client can be mounted as long as MDT0 is
256 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
257 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
258 OBD_STATFS_FOR_MDT0);
262 /* This needs to be after statfs to ensure connect has finished.
263 * Note that "data" does NOT contain the valid connect reply.
264 * If connecting to a 1.8 server there will be no LMV device, so
265 * we can access the MDC export directly and exp_connect_flags will
266 * be non-zero, but if accessing an upgraded 2.1 server it will
267 * have the correct flags filled in.
268 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
269 valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
270 if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
271 valid != CLIENT_CONNECT_MDT_REQD) {
274 buf = kzalloc(PAGE_CACHE_SIZE, GFP_KERNEL);
279 obd_connect_flags2str(buf, PAGE_CACHE_SIZE,
280 valid ^ CLIENT_CONNECT_MDT_REQD, ",");
281 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",
282 sbi->ll_md_exp->exp_obd->obd_name, buf);
288 size = sizeof(*data);
289 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
290 KEY_CONN_DATA, &size, data, NULL);
292 CERROR("%s: Get connect data failed: rc = %d\n",
293 sbi->ll_md_exp->exp_obd->obd_name, err);
297 LASSERT(osfs->os_bsize);
298 sb->s_blocksize = osfs->os_bsize;
299 sb->s_blocksize_bits = log2(osfs->os_bsize);
300 sb->s_magic = LL_SUPER_MAGIC;
301 sb->s_maxbytes = MAX_LFS_FILESIZE;
302 sbi->ll_namelen = osfs->os_namelen;
304 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
305 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
306 LCONSOLE_INFO("Disabling user_xattr feature because it is not supported on the server\n");
307 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
310 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
312 sb->s_flags |= MS_POSIXACL;
314 sbi->ll_flags |= LL_SBI_ACL;
316 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
318 sb->s_flags &= ~MS_POSIXACL;
320 sbi->ll_flags &= ~LL_SBI_ACL;
323 if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
324 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
325 sbi->ll_flags |= LL_SBI_RMT_CLIENT;
326 LCONSOLE_INFO("client is set as remote by default.\n");
329 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
330 sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
331 LCONSOLE_INFO("client claims to be remote, but server rejected, forced to be local.\n");
335 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
336 sbi->ll_flags |= LL_SBI_64BIT_HASH;
338 if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
339 sbi->ll_md_brw_size = data->ocd_brw_size;
341 sbi->ll_md_brw_size = PAGE_CACHE_SIZE;
343 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK) {
344 LCONSOLE_INFO("Layout lock feature supported.\n");
345 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
348 if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
349 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
351 "%s: disabling xattr cache due to unknown maximum xattr size.\n",
354 sbi->ll_flags |= LL_SBI_XATTR_CACHE;
355 sbi->ll_xattr_cache_enabled = 1;
359 obd = class_name2obd(dt);
361 CERROR("DT %s: not setup or attached\n", dt);
366 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
367 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
368 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
369 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
370 OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
371 OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
372 OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
373 OBD_CONNECT_MAXBYTES |
374 OBD_CONNECT_EINPROGRESS |
375 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
376 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS;
378 if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
379 data->ocd_connect_flags |= OBD_CONNECT_SOM;
381 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
382 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
383 * disabled by default, because it can still be enabled on the
384 * fly via /sys. As a consequence, we still need to come to an
385 * agreement on the supported algorithms at connect time */
386 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
388 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
389 data->ocd_cksum_types = OBD_CKSUM_ADLER;
391 data->ocd_cksum_types = cksum_types_supported_client();
394 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
395 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
396 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
398 CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d ocd_grant: %d\n",
399 data->ocd_connect_flags,
400 data->ocd_version, data->ocd_grant);
402 obd->obd_upcall.onu_owner = &sbi->ll_lco;
403 obd->obd_upcall.onu_upcall = cl_ocd_update;
405 data->ocd_brw_size = DT_MAX_BRW_SIZE;
407 err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
410 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",
414 CERROR("%s: Cannot connect to %s: rc = %d\n",
415 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
419 sbi->ll_dt_exp->exp_connect_data = *data;
421 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
422 LUSTRE_SEQ_METADATA);
424 CERROR("%s: Can't init data layer FID infrastructure, rc = %d\n",
425 sbi->ll_dt_exp->exp_obd->obd_name, err);
429 mutex_lock(&sbi->ll_lco.lco_lock);
430 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
431 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
432 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
433 mutex_unlock(&sbi->ll_lco.lco_lock);
435 fid_zero(&sbi->ll_root_fid);
436 err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid);
438 CERROR("cannot mds_connect: rc = %d\n", err);
441 if (!fid_is_sane(&sbi->ll_root_fid)) {
442 CERROR("%s: Invalid root fid "DFID" during mount\n",
443 sbi->ll_md_exp->exp_obd->obd_name,
444 PFID(&sbi->ll_root_fid));
448 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
450 sb->s_op = &lustre_super_operations;
451 #if THREAD_SIZE >= 8192 /*b=17630*/
452 sb->s_export_op = &lustre_export_operations;
456 * XXX: move this to after cbd setup? */
457 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS;
458 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
459 valid |= OBD_MD_FLRMTPERM;
460 else if (sbi->ll_flags & LL_SBI_ACL)
461 valid |= OBD_MD_FLACL;
463 op_data = kzalloc(sizeof(*op_data), GFP_NOFS);
469 op_data->op_fid1 = sbi->ll_root_fid;
470 op_data->op_mode = 0;
471 op_data->op_valid = valid;
473 err = md_getattr(sbi->ll_md_exp, op_data, &request);
476 CERROR("%s: md_getattr failed for root: rc = %d\n",
477 sbi->ll_md_exp->exp_obd->obd_name, err);
481 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
482 sbi->ll_md_exp, &lmd);
484 CERROR("failed to understand root inode md: rc = %d\n", err);
485 ptlrpc_req_finished(request);
489 LASSERT(fid_is_sane(&sbi->ll_root_fid));
490 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
491 sbi->ll_flags & LL_SBI_32BIT_API),
493 md_free_lustre_md(sbi->ll_md_exp, &lmd);
494 ptlrpc_req_finished(request);
496 if (root == NULL || IS_ERR(root)) {
498 obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
499 #ifdef CONFIG_FS_POSIX_ACL
501 posix_acl_release(lmd.posix_acl);
502 lmd.posix_acl = NULL;
505 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
507 CERROR("lustre_lite: bad iget4 for root\n");
511 err = ll_close_thread_start(&sbi->ll_lcq);
513 CERROR("cannot start close thread: rc %d\n", err);
517 #ifdef CONFIG_FS_POSIX_ACL
518 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
519 rct_init(&sbi->ll_rct);
520 et_init(&sbi->ll_et);
524 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
525 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
526 KEY_CHECKSUM, sizeof(checksum), &checksum,
530 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
531 KEY_CACHE_SET, sizeof(sbi->ll_cache),
532 &sbi->ll_cache, NULL);
534 sb->s_root = d_make_root(root);
535 if (sb->s_root == NULL) {
536 CERROR("%s: can't make root dentry\n",
537 ll_get_fsname(sb, NULL, 0));
542 sbi->ll_sdev_orig = sb->s_dev;
544 /* We set sb->s_dev equal on all lustre clients in order to support
545 * NFS export clustering. NFSD requires that the FSID be the same
547 /* s_dev is also used in lt_compare() to compare two fs, but that is
548 * only a node-local comparison. */
549 uuid = obd_get_uuid(sbi->ll_md_exp);
551 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
552 get_uuid2fsid(uuid->uuid, strlen(uuid->uuid), &sbi->ll_fsid);
562 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
564 obd_disconnect(sbi->ll_dt_exp);
565 sbi->ll_dt_exp = NULL;
566 /* Make sure all OScs are gone, since cl_cache is accessing sbi. */
567 obd_zombie_barrier();
569 obd_fid_fini(sbi->ll_md_exp->exp_obd);
571 obd_disconnect(sbi->ll_md_exp);
572 sbi->ll_md_exp = NULL;
576 ldebugfs_unregister_mountpoint(sbi);
580 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
584 *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
586 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
587 KEY_MAX_EASIZE, &size, lmmsize, NULL);
589 CERROR("Get max mdsize error rc %d\n", rc);
594 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
599 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
600 KEY_DEFAULT_EASIZE, &size, lmmsize, NULL);
602 CERROR("Get default mdsize error rc %d\n", rc);
607 static void client_common_put_super(struct super_block *sb)
609 struct ll_sb_info *sbi = ll_s2sbi(sb);
611 #ifdef CONFIG_FS_POSIX_ACL
612 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
613 et_fini(&sbi->ll_et);
614 rct_fini(&sbi->ll_rct);
618 ll_close_thread_shutdown(sbi->ll_lcq);
622 list_del(&sbi->ll_conn_chain);
624 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
625 obd_disconnect(sbi->ll_dt_exp);
626 sbi->ll_dt_exp = NULL;
627 /* wait till all OSCs are gone, since cl_cache is accessing sbi.
629 obd_zombie_barrier();
631 ldebugfs_unregister_mountpoint(sbi);
633 obd_fid_fini(sbi->ll_md_exp->exp_obd);
634 obd_disconnect(sbi->ll_md_exp);
635 sbi->ll_md_exp = NULL;
638 void ll_kill_super(struct super_block *sb)
640 struct ll_sb_info *sbi;
643 if (!(sb->s_flags & MS_ACTIVE))
647 /* we need to restore s_dev from changed for clustered NFS before
648 * put_super because new kernels have cached s_dev and change sb->s_dev
649 * in put_super not affected real removing devices */
651 sb->s_dev = sbi->ll_sdev_orig;
652 sbi->ll_umounting = 1;
656 static inline int ll_set_opt(const char *opt, char *data, int fl)
658 if (strncmp(opt, data, strlen(opt)) != 0)
664 /* non-client-specific mount options are parsed in lmd_parse */
665 static int ll_options(char *options, int *flags)
668 char *s1 = options, *s2;
673 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
676 CDEBUG(D_SUPER, "next opt=%s\n", s1);
677 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
682 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
687 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
692 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
697 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
702 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
707 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
712 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
717 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
723 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
728 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
733 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
738 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
743 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
748 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
753 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
758 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
763 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
768 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
773 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
779 s2 = strchr(s1, ',');
787 void ll_lli_init(struct ll_inode_info *lli)
789 lli->lli_inode_magic = LLI_INODE_MAGIC;
791 lli->lli_ioepoch = 0;
792 lli->lli_maxbytes = MAX_LFS_FILESIZE;
793 spin_lock_init(&lli->lli_lock);
794 lli->lli_posix_acl = NULL;
795 lli->lli_remote_perms = NULL;
796 mutex_init(&lli->lli_rmtperm_mutex);
797 /* Do not set lli_fid, it has been initialized already. */
798 fid_zero(&lli->lli_pfid);
799 INIT_LIST_HEAD(&lli->lli_close_list);
800 atomic_set(&lli->lli_open_count, 0);
801 lli->lli_rmtperm_time = 0;
802 lli->lli_pending_och = NULL;
803 lli->lli_mds_read_och = NULL;
804 lli->lli_mds_write_och = NULL;
805 lli->lli_mds_exec_och = NULL;
806 lli->lli_open_fd_read_count = 0;
807 lli->lli_open_fd_write_count = 0;
808 lli->lli_open_fd_exec_count = 0;
809 mutex_init(&lli->lli_och_mutex);
810 spin_lock_init(&lli->lli_agl_lock);
811 lli->lli_has_smd = false;
812 spin_lock_init(&lli->lli_layout_lock);
813 ll_layout_version_set(lli, LL_LAYOUT_GEN_NONE);
814 lli->lli_clob = NULL;
816 init_rwsem(&lli->lli_xattrs_list_rwsem);
817 mutex_init(&lli->lli_xattrs_enq_lock);
819 LASSERT(lli->lli_vfs_inode.i_mode != 0);
820 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
821 mutex_init(&lli->lli_readdir_mutex);
822 lli->lli_opendir_key = NULL;
824 spin_lock_init(&lli->lli_sa_lock);
825 lli->lli_opendir_pid = 0;
827 mutex_init(&lli->lli_size_mutex);
828 lli->lli_symlink_name = NULL;
829 init_rwsem(&lli->lli_trunc_sem);
830 mutex_init(&lli->lli_write_mutex);
831 init_rwsem(&lli->lli_glimpse_sem);
832 lli->lli_glimpse_time = 0;
833 INIT_LIST_HEAD(&lli->lli_agl_list);
834 lli->lli_agl_index = 0;
835 lli->lli_async_rc = 0;
837 mutex_init(&lli->lli_layout_mutex);
840 static inline int ll_bdi_register(struct backing_dev_info *bdi)
842 static atomic_t ll_bdi_num = ATOMIC_INIT(0);
844 bdi->name = "lustre";
845 return bdi_register(bdi, NULL, "lustre-%d",
846 atomic_inc_return(&ll_bdi_num));
849 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
851 struct lustre_profile *lprof = NULL;
852 struct lustre_sb_info *lsi = s2lsi(sb);
853 struct ll_sb_info *sbi;
854 char *dt = NULL, *md = NULL;
855 char *profilenm = get_profile_name(sb);
856 struct config_llog_instance *cfg;
859 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
861 cfg = kzalloc(sizeof(*cfg), GFP_NOFS);
865 try_module_get(THIS_MODULE);
867 /* client additional sb info */
868 lsi->lsi_llsbi = sbi = ll_init_sbi(sb);
870 module_put(THIS_MODULE);
875 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
879 err = bdi_init(&lsi->lsi_bdi);
882 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
883 lsi->lsi_bdi.capabilities = 0;
884 err = ll_bdi_register(&lsi->lsi_bdi);
888 sb->s_bdi = &lsi->lsi_bdi;
889 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
890 sb->s_d_op = &ll_d_ops;
892 /* Generate a string unique to this super, in case some joker tries
893 to mount the same fs at two mount points.
894 Use the address of the super itself.*/
895 cfg->cfg_instance = sb;
896 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
897 cfg->cfg_callback = class_config_llog_handler;
898 /* set up client obds */
899 err = lustre_process_log(sb, profilenm, cfg);
901 CERROR("Unable to process log: %d\n", err);
905 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
906 lprof = class_get_profile(profilenm);
908 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be read from the MGS. Does that filesystem exist?\n",
913 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
914 lprof->lp_md, lprof->lp_dt);
916 dt = kasprintf(GFP_NOFS, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
922 md = kasprintf(GFP_NOFS, "%s-%p", lprof->lp_md, cfg->cfg_instance);
928 /* connections, registrations, sb setup */
929 err = client_common_fill_super(sb, md, dt, mnt);
936 else if (sbi->ll_flags & LL_SBI_VERBOSE)
937 LCONSOLE_WARN("Mounted %s\n", profilenm);
941 } /* ll_fill_super */
943 void ll_put_super(struct super_block *sb)
945 struct config_llog_instance cfg, params_cfg;
946 struct obd_device *obd;
947 struct lustre_sb_info *lsi = s2lsi(sb);
948 struct ll_sb_info *sbi = ll_s2sbi(sb);
949 char *profilenm = get_profile_name(sb);
952 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
954 cfg.cfg_instance = sb;
955 lustre_end_log(sb, profilenm, &cfg);
957 params_cfg.cfg_instance = sb;
958 lustre_end_log(sb, PARAMS_FILENAME, ¶ms_cfg);
960 if (sbi->ll_md_exp) {
961 obd = class_exp2obd(sbi->ll_md_exp);
963 force = obd->obd_force;
966 /* We need to set force before the lov_disconnect in
967 lustre_common_put_super, since l_d cleans up osc's as well. */
970 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
972 obd->obd_force = force;
977 /* Only if client_common_fill_super succeeded */
978 client_common_put_super(sb);
982 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)))
983 class_manual_cleanup(obd);
985 if (sbi->ll_flags & LL_SBI_VERBOSE)
986 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
989 class_del_profile(profilenm);
991 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
992 bdi_destroy(&lsi->lsi_bdi);
993 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
997 lsi->lsi_llsbi = NULL;
999 lustre_common_put_super(sb);
1001 module_put(THIS_MODULE);
1002 } /* client_put_super */
1004 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1006 struct inode *inode = NULL;
1008 /* NOTE: we depend on atomic igrab() -bzzz */
1009 lock_res_and_lock(lock);
1010 if (lock->l_resource->lr_lvb_inode) {
1011 struct ll_inode_info *lli;
1013 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1014 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1015 inode = igrab(lock->l_resource->lr_lvb_inode);
1017 inode = lock->l_resource->lr_lvb_inode;
1018 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1019 D_WARNING, lock, "lr_lvb_inode %p is bogus: magic %08x",
1020 lock->l_resource->lr_lvb_inode,
1021 lli->lli_inode_magic);
1025 unlock_res_and_lock(lock);
1029 void ll_clear_inode(struct inode *inode)
1031 struct ll_inode_info *lli = ll_i2info(inode);
1032 struct ll_sb_info *sbi = ll_i2sbi(inode);
1034 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1035 inode->i_generation, inode);
1037 if (S_ISDIR(inode->i_mode)) {
1038 /* these should have been cleared in ll_file_release */
1039 LASSERT(lli->lli_opendir_key == NULL);
1040 LASSERT(lli->lli_sai == NULL);
1041 LASSERT(lli->lli_opendir_pid == 0);
1044 spin_lock(&lli->lli_lock);
1045 ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1046 spin_unlock(&lli->lli_lock);
1047 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1049 LASSERT(!lli->lli_open_fd_write_count);
1050 LASSERT(!lli->lli_open_fd_read_count);
1051 LASSERT(!lli->lli_open_fd_exec_count);
1053 if (lli->lli_mds_write_och)
1054 ll_md_real_close(inode, FMODE_WRITE);
1055 if (lli->lli_mds_exec_och)
1056 ll_md_real_close(inode, FMODE_EXEC);
1057 if (lli->lli_mds_read_och)
1058 ll_md_real_close(inode, FMODE_READ);
1060 if (S_ISLNK(inode->i_mode)) {
1061 kfree(lli->lli_symlink_name);
1062 lli->lli_symlink_name = NULL;
1065 ll_xattr_cache_destroy(inode);
1067 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1068 LASSERT(lli->lli_posix_acl == NULL);
1069 if (lli->lli_remote_perms) {
1070 free_rmtperm_hash(lli->lli_remote_perms);
1071 lli->lli_remote_perms = NULL;
1074 #ifdef CONFIG_FS_POSIX_ACL
1075 else if (lli->lli_posix_acl) {
1076 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1077 LASSERT(lli->lli_remote_perms == NULL);
1078 posix_acl_release(lli->lli_posix_acl);
1079 lli->lli_posix_acl = NULL;
1082 lli->lli_inode_magic = LLI_INODE_DEAD;
1084 if (!S_ISDIR(inode->i_mode))
1085 LASSERT(list_empty(&lli->lli_agl_list));
1088 * XXX This has to be done before lsm is freed below, because
1089 * cl_object still uses inode lsm.
1091 cl_inode_fini(inode);
1092 lli->lli_has_smd = false;
1095 #define TIMES_SET_FLAGS (ATTR_MTIME_SET | ATTR_ATIME_SET | ATTR_TIMES_SET)
1097 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data,
1098 struct md_open_data **mod)
1100 struct lustre_md md;
1101 struct inode *inode = d_inode(dentry);
1102 struct ll_sb_info *sbi = ll_i2sbi(inode);
1103 struct ptlrpc_request *request = NULL;
1106 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1107 LUSTRE_OPC_ANY, NULL);
1108 if (IS_ERR(op_data))
1109 return PTR_ERR(op_data);
1111 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1114 ptlrpc_req_finished(request);
1115 if (rc == -ENOENT) {
1117 /* Unlinked special device node? Or just a race?
1118 * Pretend we done everything. */
1119 if (!S_ISREG(inode->i_mode) &&
1120 !S_ISDIR(inode->i_mode)) {
1121 ia_valid = op_data->op_attr.ia_valid;
1122 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1123 rc = simple_setattr(dentry, &op_data->op_attr);
1124 op_data->op_attr.ia_valid = ia_valid;
1126 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1127 CERROR("md_setattr fails: rc = %d\n", rc);
1132 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1133 sbi->ll_md_exp, &md);
1135 ptlrpc_req_finished(request);
1139 ia_valid = op_data->op_attr.ia_valid;
1140 /* inode size will be in cl_setattr_ost, can't do it now since dirty
1141 * cache is not cleared yet. */
1142 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1143 rc = simple_setattr(dentry, &op_data->op_attr);
1144 op_data->op_attr.ia_valid = ia_valid;
1146 /* Extract epoch data if obtained. */
1147 op_data->op_handle = md.body->handle;
1148 op_data->op_ioepoch = md.body->ioepoch;
1150 ll_update_inode(inode, &md);
1151 ptlrpc_req_finished(request);
1156 /* Close IO epoch and send Size-on-MDS attribute update. */
1157 static int ll_setattr_done_writing(struct inode *inode,
1158 struct md_op_data *op_data,
1159 struct md_open_data *mod)
1161 struct ll_inode_info *lli = ll_i2info(inode);
1164 LASSERT(op_data != NULL);
1165 if (!S_ISREG(inode->i_mode))
1168 CDEBUG(D_INODE, "Epoch %llu closed on "DFID" for truncate\n",
1169 op_data->op_ioepoch, PFID(&lli->lli_fid));
1171 op_data->op_flags = MF_EPOCH_CLOSE;
1172 ll_done_writing_attr(inode, op_data);
1173 ll_pack_inode2opdata(inode, op_data, NULL);
1175 rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1177 /* MDS has instructed us to obtain Size-on-MDS attribute
1178 * from OSTs and send setattr to back to MDS. */
1179 rc = ll_som_update(inode, op_data);
1181 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1186 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1187 * object(s) determine the file size and mtime. Otherwise, the MDS will
1188 * keep these values until such a time that objects are allocated for it.
1189 * We do the MDS operations first, as it is checking permissions for us.
1190 * We don't to the MDS RPC if there is nothing that we want to store there,
1191 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1192 * going to do an RPC anyways.
1194 * If we are doing a truncate, we will send the mtime and ctime updates
1195 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1196 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1199 * In case of HSMimport, we only set attr on MDS.
1201 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1203 struct inode *inode = d_inode(dentry);
1204 struct ll_inode_info *lli = ll_i2info(inode);
1205 struct md_op_data *op_data = NULL;
1206 struct md_open_data *mod = NULL;
1207 bool file_is_released = false;
1208 int rc = 0, rc1 = 0;
1211 "%s: setattr inode %p/fid:"DFID
1212 " from %llu to %llu, valid %x, hsm_import %d\n",
1213 ll_get_fsname(inode->i_sb, NULL, 0), inode,
1214 PFID(&lli->lli_fid), i_size_read(inode), attr->ia_size,
1215 attr->ia_valid, hsm_import);
1217 if (attr->ia_valid & ATTR_SIZE) {
1218 /* Check new size against VFS/VM file size limit and rlimit */
1219 rc = inode_newsize_ok(inode, attr->ia_size);
1223 /* The maximum Lustre file size is variable, based on the
1224 * OST maximum object size and number of stripes. This
1225 * needs another check in addition to the VFS check above. */
1226 if (attr->ia_size > ll_file_maxbytes(inode)) {
1227 CDEBUG(D_INODE, "file "DFID" too large %llu > %llu\n",
1228 PFID(&lli->lli_fid), attr->ia_size,
1229 ll_file_maxbytes(inode));
1233 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1236 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1237 if (attr->ia_valid & TIMES_SET_FLAGS) {
1238 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1239 !capable(CFS_CAP_FOWNER))
1243 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1244 if (attr->ia_valid & ATTR_CTIME) {
1245 attr->ia_ctime = CURRENT_TIME;
1246 attr->ia_valid |= ATTR_CTIME_SET;
1248 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1249 (attr->ia_valid & ATTR_ATIME)) {
1250 attr->ia_atime = CURRENT_TIME;
1251 attr->ia_valid |= ATTR_ATIME_SET;
1253 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1254 (attr->ia_valid & ATTR_MTIME)) {
1255 attr->ia_mtime = CURRENT_TIME;
1256 attr->ia_valid |= ATTR_MTIME_SET;
1259 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1260 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %llu\n",
1261 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1262 (s64)ktime_get_real_seconds());
1264 /* If we are changing file size, file content is modified, flag it. */
1265 if (attr->ia_valid & ATTR_SIZE) {
1266 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1267 spin_lock(&lli->lli_lock);
1268 lli->lli_flags |= LLIF_DATA_MODIFIED;
1269 spin_unlock(&lli->lli_lock);
1272 /* We always do an MDS RPC, even if we're only changing the size;
1273 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1275 op_data = kzalloc(sizeof(*op_data), GFP_NOFS);
1279 if (!S_ISDIR(inode->i_mode))
1280 mutex_unlock(&inode->i_mutex);
1282 memcpy(&op_data->op_attr, attr, sizeof(*attr));
1284 /* Open epoch for truncate. */
1285 if (exp_connect_som(ll_i2mdexp(inode)) &&
1286 (attr->ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1287 op_data->op_flags = MF_EPOCH_OPEN;
1289 /* truncate on a released file must failed with -ENODATA,
1290 * so size must not be set on MDS for released file
1291 * but other attributes must be set
1293 if (S_ISREG(inode->i_mode)) {
1294 struct lov_stripe_md *lsm;
1297 ll_layout_refresh(inode, &gen);
1298 lsm = ccc_inode_lsm_get(inode);
1299 if (lsm && lsm->lsm_pattern & LOV_PATTERN_F_RELEASED)
1300 file_is_released = true;
1301 ccc_inode_lsm_put(inode, lsm);
1304 /* if not in HSM import mode, clear size attr for released file
1305 * we clear the attribute send to MDT in op_data, not the original
1306 * received from caller in attr which is used later to
1307 * decide return code */
1308 if (file_is_released && (attr->ia_valid & ATTR_SIZE) && !hsm_import)
1309 op_data->op_attr.ia_valid &= ~ATTR_SIZE;
1311 rc = ll_md_setattr(dentry, op_data, &mod);
1315 /* truncate failed (only when non HSM import), others succeed */
1316 if (file_is_released) {
1317 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1324 /* RPC to MDT is sent, cancel data modification flag */
1325 if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) {
1326 spin_lock(&lli->lli_lock);
1327 lli->lli_flags &= ~LLIF_DATA_MODIFIED;
1328 spin_unlock(&lli->lli_lock);
1331 ll_ioepoch_open(lli, op_data->op_ioepoch);
1332 if (!S_ISREG(inode->i_mode)) {
1337 if (attr->ia_valid & (ATTR_SIZE |
1338 ATTR_ATIME | ATTR_ATIME_SET |
1339 ATTR_MTIME | ATTR_MTIME_SET)) {
1340 /* For truncate and utimes sending attributes to OSTs, setting
1341 * mtime/atime to the past will be performed under PW [0:EOF]
1342 * extent lock (new_size:EOF for truncate). It may seem
1343 * excessive to send mtime/atime updates to OSTs when not
1344 * setting times to past, but it is necessary due to possible
1345 * time de-synchronization between MDT inode and OST objects */
1346 if (attr->ia_valid & ATTR_SIZE)
1347 down_write(&lli->lli_trunc_sem);
1348 rc = cl_setattr_ost(inode, attr);
1349 if (attr->ia_valid & ATTR_SIZE)
1350 up_write(&lli->lli_trunc_sem);
1353 if (op_data->op_ioepoch) {
1354 rc1 = ll_setattr_done_writing(inode, op_data, mod);
1358 ll_finish_md_op_data(op_data);
1360 if (!S_ISDIR(inode->i_mode)) {
1361 mutex_lock(&inode->i_mutex);
1362 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1363 inode_dio_wait(inode);
1366 ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1367 LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1372 int ll_setattr(struct dentry *de, struct iattr *attr)
1374 int mode = d_inode(de)->i_mode;
1376 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1377 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1378 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1380 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1381 (ATTR_SIZE|ATTR_MODE)) &&
1382 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1383 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1384 !(attr->ia_mode & S_ISGID))))
1385 attr->ia_valid |= ATTR_FORCE;
1387 if ((attr->ia_valid & ATTR_MODE) &&
1389 !(attr->ia_mode & S_ISUID) &&
1390 !(attr->ia_valid & ATTR_KILL_SUID))
1391 attr->ia_valid |= ATTR_KILL_SUID;
1393 if ((attr->ia_valid & ATTR_MODE) &&
1394 ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1395 !(attr->ia_mode & S_ISGID) &&
1396 !(attr->ia_valid & ATTR_KILL_SGID))
1397 attr->ia_valid |= ATTR_KILL_SGID;
1399 return ll_setattr_raw(de, attr, false);
1402 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1403 __u64 max_age, __u32 flags)
1405 struct ll_sb_info *sbi = ll_s2sbi(sb);
1406 struct obd_statfs obd_osfs;
1409 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1411 CERROR("md_statfs fails: rc = %d\n", rc);
1415 osfs->os_type = sb->s_magic;
1417 CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1418 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,
1421 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1422 flags |= OBD_STATFS_NODELAY;
1424 rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1426 CERROR("obd_statfs fails: rc = %d\n", rc);
1430 CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1431 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1434 osfs->os_bsize = obd_osfs.os_bsize;
1435 osfs->os_blocks = obd_osfs.os_blocks;
1436 osfs->os_bfree = obd_osfs.os_bfree;
1437 osfs->os_bavail = obd_osfs.os_bavail;
1439 /* If we don't have as many objects free on the OST as inodes
1440 * on the MDS, we reduce the total number of inodes to
1441 * compensate, so that the "inodes in use" number is correct.
1443 if (obd_osfs.os_ffree < osfs->os_ffree) {
1444 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1446 osfs->os_ffree = obd_osfs.os_ffree;
1452 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1454 struct super_block *sb = de->d_sb;
1455 struct obd_statfs osfs;
1458 CDEBUG(D_VFSTRACE, "VFS Op: at %llu jiffies\n", get_jiffies_64());
1459 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1461 /* Some amount of caching on the client is allowed */
1462 rc = ll_statfs_internal(sb, &osfs,
1463 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1468 statfs_unpack(sfs, &osfs);
1470 /* We need to downshift for all 32-bit kernels, because we can't
1471 * tell if the kernel is being called via sys_statfs64() or not.
1472 * Stop before overflowing f_bsize - in which case it is better
1473 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1474 if (sizeof(long) < 8) {
1475 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1478 osfs.os_blocks >>= 1;
1479 osfs.os_bfree >>= 1;
1480 osfs.os_bavail >>= 1;
1484 sfs->f_blocks = osfs.os_blocks;
1485 sfs->f_bfree = osfs.os_bfree;
1486 sfs->f_bavail = osfs.os_bavail;
1487 sfs->f_fsid = ll_s2sbi(sb)->ll_fsid;
1491 void ll_inode_size_lock(struct inode *inode)
1493 struct ll_inode_info *lli;
1495 LASSERT(!S_ISDIR(inode->i_mode));
1497 lli = ll_i2info(inode);
1498 mutex_lock(&lli->lli_size_mutex);
1501 void ll_inode_size_unlock(struct inode *inode)
1503 struct ll_inode_info *lli;
1505 lli = ll_i2info(inode);
1506 mutex_unlock(&lli->lli_size_mutex);
1509 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1511 struct ll_inode_info *lli = ll_i2info(inode);
1512 struct mdt_body *body = md->body;
1513 struct lov_stripe_md *lsm = md->lsm;
1514 struct ll_sb_info *sbi = ll_i2sbi(inode);
1516 LASSERT((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1518 if (!lli->lli_has_smd &&
1519 !(sbi->ll_flags & LL_SBI_LAYOUT_LOCK))
1520 cl_file_inode_init(inode, md);
1522 lli->lli_maxbytes = lsm->lsm_maxbytes;
1523 if (lli->lli_maxbytes > MAX_LFS_FILESIZE)
1524 lli->lli_maxbytes = MAX_LFS_FILESIZE;
1527 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1528 if (body->valid & OBD_MD_FLRMTPERM)
1529 ll_update_remote_perm(inode, md->remote_perm);
1531 #ifdef CONFIG_FS_POSIX_ACL
1532 else if (body->valid & OBD_MD_FLACL) {
1533 spin_lock(&lli->lli_lock);
1534 if (lli->lli_posix_acl)
1535 posix_acl_release(lli->lli_posix_acl);
1536 lli->lli_posix_acl = md->posix_acl;
1537 spin_unlock(&lli->lli_lock);
1540 inode->i_ino = cl_fid_build_ino(&body->fid1,
1541 sbi->ll_flags & LL_SBI_32BIT_API);
1542 inode->i_generation = cl_fid_build_gen(&body->fid1);
1544 if (body->valid & OBD_MD_FLATIME) {
1545 if (body->atime > LTIME_S(inode->i_atime))
1546 LTIME_S(inode->i_atime) = body->atime;
1547 lli->lli_lvb.lvb_atime = body->atime;
1549 if (body->valid & OBD_MD_FLMTIME) {
1550 if (body->mtime > LTIME_S(inode->i_mtime)) {
1551 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to %llu\n",
1552 inode->i_ino, LTIME_S(inode->i_mtime),
1554 LTIME_S(inode->i_mtime) = body->mtime;
1556 lli->lli_lvb.lvb_mtime = body->mtime;
1558 if (body->valid & OBD_MD_FLCTIME) {
1559 if (body->ctime > LTIME_S(inode->i_ctime))
1560 LTIME_S(inode->i_ctime) = body->ctime;
1561 lli->lli_lvb.lvb_ctime = body->ctime;
1563 if (body->valid & OBD_MD_FLMODE)
1564 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1565 if (body->valid & OBD_MD_FLTYPE)
1566 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1567 LASSERT(inode->i_mode != 0);
1568 if (S_ISREG(inode->i_mode))
1569 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1570 LL_MAX_BLKSIZE_BITS);
1572 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1573 if (body->valid & OBD_MD_FLUID)
1574 inode->i_uid = make_kuid(&init_user_ns, body->uid);
1575 if (body->valid & OBD_MD_FLGID)
1576 inode->i_gid = make_kgid(&init_user_ns, body->gid);
1577 if (body->valid & OBD_MD_FLFLAGS)
1578 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1579 if (body->valid & OBD_MD_FLNLINK)
1580 set_nlink(inode, body->nlink);
1581 if (body->valid & OBD_MD_FLRDEV)
1582 inode->i_rdev = old_decode_dev(body->rdev);
1584 if (body->valid & OBD_MD_FLID) {
1585 /* FID shouldn't be changed! */
1586 if (fid_is_sane(&lli->lli_fid)) {
1587 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1588 "Trying to change FID "DFID
1589 " to the "DFID", inode %lu/%u(%p)\n",
1590 PFID(&lli->lli_fid), PFID(&body->fid1),
1591 inode->i_ino, inode->i_generation, inode);
1593 lli->lli_fid = body->fid1;
1596 LASSERT(fid_seq(&lli->lli_fid) != 0);
1598 if (body->valid & OBD_MD_FLSIZE) {
1599 if (exp_connect_som(ll_i2mdexp(inode)) &&
1600 S_ISREG(inode->i_mode)) {
1601 struct lustre_handle lockh;
1604 /* As it is possible a blocking ast has been processed
1605 * by this time, we need to check there is an UPDATE
1606 * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1608 mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1609 &lockh, LDLM_FL_CBPENDING,
1613 if (lli->lli_flags & (LLIF_DONE_WRITING |
1614 LLIF_EPOCH_PENDING |
1616 CERROR("ino %lu flags %u still has size authority! do not trust the size got from MDS\n",
1617 inode->i_ino, lli->lli_flags);
1619 /* Use old size assignment to avoid
1620 * deadlock bz14138 & bz14326 */
1621 i_size_write(inode, body->size);
1622 spin_lock(&lli->lli_lock);
1623 lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1624 spin_unlock(&lli->lli_lock);
1626 ldlm_lock_decref(&lockh, mode);
1629 /* Use old size assignment to avoid
1630 * deadlock bz14138 & bz14326 */
1631 i_size_write(inode, body->size);
1633 CDEBUG(D_VFSTRACE, "inode=%lu, updating i_size %llu\n",
1634 inode->i_ino, (unsigned long long)body->size);
1637 if (body->valid & OBD_MD_FLBLOCKS)
1638 inode->i_blocks = body->blocks;
1641 if (body->valid & OBD_MD_TSTATE) {
1642 if (body->t_state & MS_RESTORE)
1643 lli->lli_flags |= LLIF_FILE_RESTORING;
1647 void ll_read_inode2(struct inode *inode, void *opaque)
1649 struct lustre_md *md = opaque;
1650 struct ll_inode_info *lli = ll_i2info(inode);
1652 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1653 PFID(&lli->lli_fid), inode);
1655 LASSERT(!lli->lli_has_smd);
1657 /* Core attributes from the MDS first. This is a new inode, and
1658 * the VFS doesn't zero times in the core inode so we have to do
1659 * it ourselves. They will be overwritten by either MDS or OST
1660 * attributes - we just need to make sure they aren't newer. */
1661 LTIME_S(inode->i_mtime) = 0;
1662 LTIME_S(inode->i_atime) = 0;
1663 LTIME_S(inode->i_ctime) = 0;
1665 ll_update_inode(inode, md);
1667 /* OIDEBUG(inode); */
1669 if (S_ISREG(inode->i_mode)) {
1670 struct ll_sb_info *sbi = ll_i2sbi(inode);
1672 inode->i_op = &ll_file_inode_operations;
1673 inode->i_fop = sbi->ll_fop;
1674 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1675 } else if (S_ISDIR(inode->i_mode)) {
1676 inode->i_op = &ll_dir_inode_operations;
1677 inode->i_fop = &ll_dir_operations;
1678 } else if (S_ISLNK(inode->i_mode)) {
1679 inode->i_op = &ll_fast_symlink_inode_operations;
1681 inode->i_op = &ll_special_inode_operations;
1683 init_special_inode(inode, inode->i_mode,
1688 void ll_delete_inode(struct inode *inode)
1690 struct cl_inode_info *lli = cl_i2info(inode);
1692 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
1693 /* discard all dirty pages before truncating them, required by
1694 * osc_extent implementation at LU-1030. */
1695 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF,
1696 CL_FSYNC_DISCARD, 1);
1698 truncate_inode_pages_final(&inode->i_data);
1700 /* Workaround for LU-118 */
1701 if (inode->i_data.nrpages) {
1702 spin_lock_irq(&inode->i_data.tree_lock);
1703 spin_unlock_irq(&inode->i_data.tree_lock);
1704 LASSERTF(inode->i_data.nrpages == 0,
1705 "inode=%lu/%u(%p) nrpages=%lu, see http://jira.whamcloud.com/browse/LU-118\n",
1706 inode->i_ino, inode->i_generation, inode,
1707 inode->i_data.nrpages);
1709 /* Workaround end */
1711 ll_clear_inode(inode);
1715 int ll_iocontrol(struct inode *inode, struct file *file,
1716 unsigned int cmd, unsigned long arg)
1718 struct ll_sb_info *sbi = ll_i2sbi(inode);
1719 struct ptlrpc_request *req = NULL;
1723 case FSFILT_IOC_GETFLAGS: {
1724 struct mdt_body *body;
1725 struct md_op_data *op_data;
1727 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
1728 0, 0, LUSTRE_OPC_ANY,
1730 if (IS_ERR(op_data))
1731 return PTR_ERR(op_data);
1733 op_data->op_valid = OBD_MD_FLFLAGS;
1734 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
1735 ll_finish_md_op_data(op_data);
1737 CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1741 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1743 flags = body->flags;
1745 ptlrpc_req_finished(req);
1747 return put_user(flags, (int *)arg);
1749 case FSFILT_IOC_SETFLAGS: {
1750 struct lov_stripe_md *lsm;
1751 struct obd_info oinfo = { };
1752 struct md_op_data *op_data;
1754 if (get_user(flags, (int *)arg))
1757 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1758 LUSTRE_OPC_ANY, NULL);
1759 if (IS_ERR(op_data))
1760 return PTR_ERR(op_data);
1762 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1763 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1764 rc = md_setattr(sbi->ll_md_exp, op_data,
1765 NULL, 0, NULL, 0, &req, NULL);
1766 ll_finish_md_op_data(op_data);
1767 ptlrpc_req_finished(req);
1771 inode->i_flags = ll_ext_to_inode_flags(flags);
1773 lsm = ccc_inode_lsm_get(inode);
1774 if (!lsm_has_objects(lsm)) {
1775 ccc_inode_lsm_put(inode, lsm);
1779 oinfo.oi_oa = kmem_cache_alloc(obdo_cachep,
1780 GFP_NOFS | __GFP_ZERO);
1782 ccc_inode_lsm_put(inode, lsm);
1786 oinfo.oi_oa->o_oi = lsm->lsm_oi;
1787 oinfo.oi_oa->o_flags = flags;
1788 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1790 obdo_set_parent_fid(oinfo.oi_oa, &ll_i2info(inode)->lli_fid);
1791 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1792 kmem_cache_free(obdo_cachep, oinfo.oi_oa);
1793 ccc_inode_lsm_put(inode, lsm);
1795 if (rc && rc != -EPERM && rc != -EACCES)
1796 CERROR("osc_setattr_async fails: rc = %d\n", rc);
1807 int ll_flush_ctx(struct inode *inode)
1809 struct ll_sb_info *sbi = ll_i2sbi(inode);
1811 CDEBUG(D_SEC, "flush context for user %d\n",
1812 from_kuid(&init_user_ns, current_uid()));
1814 obd_set_info_async(NULL, sbi->ll_md_exp,
1815 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1817 obd_set_info_async(NULL, sbi->ll_dt_exp,
1818 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1823 /* umount -f client means force down, don't save state */
1824 void ll_umount_begin(struct super_block *sb)
1826 struct ll_sb_info *sbi = ll_s2sbi(sb);
1827 struct obd_device *obd;
1828 struct obd_ioctl_data *ioc_data;
1830 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1831 sb->s_count, atomic_read(&sb->s_active));
1833 obd = class_exp2obd(sbi->ll_md_exp);
1835 CERROR("Invalid MDC connection handle %#llx\n",
1836 sbi->ll_md_exp->exp_handle.h_cookie);
1841 obd = class_exp2obd(sbi->ll_dt_exp);
1843 CERROR("Invalid LOV connection handle %#llx\n",
1844 sbi->ll_dt_exp->exp_handle.h_cookie);
1849 ioc_data = kzalloc(sizeof(*ioc_data), GFP_NOFS);
1851 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
1852 sizeof(*ioc_data), ioc_data, NULL);
1854 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
1855 sizeof(*ioc_data), ioc_data, NULL);
1860 /* Really, we'd like to wait until there are no requests outstanding,
1861 * and then continue. For now, we just invalidate the requests,
1862 * schedule() and sleep one second if needed, and hope.
1867 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
1869 struct ll_sb_info *sbi = ll_s2sbi(sb);
1870 char *profilenm = get_profile_name(sb);
1874 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
1875 read_only = *flags & MS_RDONLY;
1876 err = obd_set_info_async(NULL, sbi->ll_md_exp,
1877 sizeof(KEY_READ_ONLY),
1878 KEY_READ_ONLY, sizeof(read_only),
1881 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
1882 profilenm, read_only ?
1883 "read-only" : "read-write", err);
1888 sb->s_flags |= MS_RDONLY;
1890 sb->s_flags &= ~MS_RDONLY;
1892 if (sbi->ll_flags & LL_SBI_VERBOSE)
1893 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
1894 read_only ? "read-only" : "read-write");
1900 * Cleanup the open handle that is cached on MDT-side.
1902 * For open case, the client side open handling thread may hit error
1903 * after the MDT grant the open. Under such case, the client should
1904 * send close RPC to the MDT as cleanup; otherwise, the open handle
1905 * on the MDT will be leaked there until the client umount or evicted.
1907 * In further, if someone unlinked the file, because the open handle
1908 * holds the reference on such file/object, then it will block the
1909 * subsequent threads that want to locate such object via FID.
1911 * \param[in] sb super block for this file-system
1912 * \param[in] open_req pointer to the original open request
1914 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
1916 struct mdt_body *body;
1917 struct md_op_data *op_data;
1918 struct ptlrpc_request *close_req = NULL;
1919 struct obd_export *exp = ll_s2sbi(sb)->ll_md_exp;
1921 body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
1922 op_data = kzalloc(sizeof(*op_data), GFP_NOFS);
1924 CWARN("%s: cannot allocate op_data to release open handle for "
1926 ll_get_fsname(sb, NULL, 0), PFID(&body->fid1));
1931 op_data->op_fid1 = body->fid1;
1932 op_data->op_ioepoch = body->ioepoch;
1933 op_data->op_handle = body->handle;
1934 op_data->op_mod_time = get_seconds();
1935 md_close(exp, op_data, NULL, &close_req);
1936 ptlrpc_req_finished(close_req);
1937 ll_finish_md_op_data(op_data);
1940 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
1941 struct super_block *sb, struct lookup_intent *it)
1943 struct ll_sb_info *sbi = NULL;
1944 struct lustre_md md;
1947 LASSERT(*inode || sb);
1948 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
1949 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
1950 sbi->ll_md_exp, &md);
1955 ll_update_inode(*inode, &md);
1957 LASSERT(sb != NULL);
1960 * At this point server returns to client's same fid as client
1961 * generated for creating. So using ->fid1 is okay here.
1963 LASSERT(fid_is_sane(&md.body->fid1));
1965 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1,
1966 sbi->ll_flags & LL_SBI_32BIT_API),
1968 if (*inode == NULL || IS_ERR(*inode)) {
1969 #ifdef CONFIG_FS_POSIX_ACL
1971 posix_acl_release(md.posix_acl);
1972 md.posix_acl = NULL;
1975 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
1977 CERROR("new_inode -fatal: rc %d\n", rc);
1982 /* Handling piggyback layout lock.
1983 * Layout lock can be piggybacked by getattr and open request.
1984 * The lsm can be applied to inode only if it comes with a layout lock
1985 * otherwise correct layout may be overwritten, for example:
1986 * 1. proc1: mdt returns a lsm but not granting layout
1987 * 2. layout was changed by another client
1988 * 3. proc2: refresh layout and layout lock granted
1989 * 4. proc1: to apply a stale layout */
1990 if (it != NULL && it->d.lustre.it_lock_mode != 0) {
1991 struct lustre_handle lockh;
1992 struct ldlm_lock *lock;
1994 lockh.cookie = it->d.lustre.it_lock_handle;
1995 lock = ldlm_handle2lock(&lockh);
1996 LASSERT(lock != NULL);
1997 if (ldlm_has_layout(lock)) {
1998 struct cl_object_conf conf;
2000 memset(&conf, 0, sizeof(conf));
2001 conf.coc_opc = OBJECT_CONF_SET;
2002 conf.coc_inode = *inode;
2003 conf.coc_lock = lock;
2004 conf.u.coc_md = &md;
2005 (void)ll_layout_conf(*inode, &conf);
2007 LDLM_LOCK_PUT(lock);
2012 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2013 md_free_lustre_md(sbi->ll_md_exp, &md);
2016 if (rc != 0 && it && it->it_op & IT_OPEN)
2017 ll_open_cleanup(sb ? sb : (*inode)->i_sb, req);
2022 int ll_obd_statfs(struct inode *inode, void *arg)
2024 struct ll_sb_info *sbi = NULL;
2025 struct obd_export *exp;
2027 struct obd_ioctl_data *data = NULL;
2037 sbi = ll_i2sbi(inode);
2043 rc = obd_ioctl_getdata(&buf, &len, arg);
2048 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2049 !data->ioc_pbuf1 || !data->ioc_pbuf2) {
2054 if (data->ioc_inllen1 != sizeof(__u32) ||
2055 data->ioc_inllen2 != sizeof(__u32) ||
2056 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2057 data->ioc_plen2 != sizeof(struct obd_uuid)) {
2062 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2063 if (type & LL_STATFS_LMV)
2064 exp = sbi->ll_md_exp;
2065 else if (type & LL_STATFS_LOV)
2066 exp = sbi->ll_dt_exp;
2072 flags = (type & LL_STATFS_NODELAY) ? OBD_STATFS_NODELAY : 0;
2073 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, &flags);
2078 obd_ioctl_freedata(buf, len);
2082 int ll_process_config(struct lustre_cfg *lcfg)
2086 struct lprocfs_static_vars lvars;
2090 lprocfs_llite_init_vars(&lvars);
2092 /* The instance name contains the sb: lustre-client-aacfe000 */
2093 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2094 if (!ptr || !*(++ptr))
2096 rc = kstrtoul(ptr, 16, &x);
2100 /* This better be a real Lustre superblock! */
2101 LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2103 /* Note we have not called client_common_fill_super yet, so
2104 proc fns must be able to handle that! */
2105 rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2112 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2113 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2114 struct inode *i1, struct inode *i2,
2115 const char *name, int namelen,
2116 int mode, __u32 opc, void *data)
2118 LASSERT(i1 != NULL);
2120 if (namelen > ll_i2sbi(i1)->ll_namelen)
2121 return ERR_PTR(-ENAMETOOLONG);
2123 if (op_data == NULL)
2124 op_data = kzalloc(sizeof(*op_data), GFP_NOFS);
2126 if (op_data == NULL)
2127 return ERR_PTR(-ENOMEM);
2129 ll_i2gids(op_data->op_suppgids, i1, i2);
2130 op_data->op_fid1 = *ll_inode2fid(i1);
2133 op_data->op_fid2 = *ll_inode2fid(i2);
2135 fid_zero(&op_data->op_fid2);
2137 op_data->op_name = name;
2138 op_data->op_namelen = namelen;
2139 op_data->op_mode = mode;
2140 op_data->op_mod_time = ktime_get_real_seconds();
2141 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2142 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2143 op_data->op_cap = cfs_curproc_cap_pack();
2144 op_data->op_bias = 0;
2145 op_data->op_cli_flags = 0;
2146 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2147 filename_is_volatile(name, namelen, NULL))
2148 op_data->op_bias |= MDS_CREATE_VOLATILE;
2149 op_data->op_opc = opc;
2150 op_data->op_mds = 0;
2151 op_data->op_data = data;
2153 /* If the file is being opened after mknod() (normally due to NFS)
2154 * try to use the default stripe data from parent directory for
2155 * allocating OST objects. Try to pass the parent FID to MDS. */
2156 if (opc == LUSTRE_OPC_CREATE && i1 == i2 && S_ISREG(i2->i_mode) &&
2157 !ll_i2info(i2)->lli_has_smd) {
2158 struct ll_inode_info *lli = ll_i2info(i2);
2160 spin_lock(&lli->lli_lock);
2161 if (likely(!lli->lli_has_smd && !fid_is_zero(&lli->lli_pfid)))
2162 op_data->op_fid1 = lli->lli_pfid;
2163 spin_unlock(&lli->lli_lock);
2166 /* When called by ll_setattr_raw, file is i1. */
2167 if (ll_i2info(i1)->lli_flags & LLIF_DATA_MODIFIED)
2168 op_data->op_bias |= MDS_DATA_MODIFIED;
2173 void ll_finish_md_op_data(struct md_op_data *op_data)
2178 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2180 struct ll_sb_info *sbi;
2182 LASSERT((seq != NULL) && (dentry != NULL));
2183 sbi = ll_s2sbi(dentry->d_sb);
2185 if (sbi->ll_flags & LL_SBI_NOLCK)
2186 seq_puts(seq, ",nolock");
2188 if (sbi->ll_flags & LL_SBI_FLOCK)
2189 seq_puts(seq, ",flock");
2191 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2192 seq_puts(seq, ",localflock");
2194 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2195 seq_puts(seq, ",user_xattr");
2197 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2198 seq_puts(seq, ",lazystatfs");
2200 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2201 seq_puts(seq, ",user_fid2path");
2207 * Get obd name by cmd, and copy out to user space
2209 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2211 struct ll_sb_info *sbi = ll_i2sbi(inode);
2212 struct obd_device *obd;
2214 if (cmd == OBD_IOC_GETDTNAME)
2215 obd = class_exp2obd(sbi->ll_dt_exp);
2216 else if (cmd == OBD_IOC_GETMDNAME)
2217 obd = class_exp2obd(sbi->ll_md_exp);
2224 if (copy_to_user((void *)arg, obd->obd_name,
2225 strlen(obd->obd_name) + 1))
2232 * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2233 * fsname will be returned in this buffer; otherwise, a static buffer will be
2234 * used to store the fsname and returned to caller.
2236 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2238 static char fsname_static[MTI_NAME_MAXLEN];
2239 struct lustre_sb_info *lsi = s2lsi(sb);
2244 /* this means the caller wants to use static buffer
2245 * and it doesn't care about race. Usually this is
2246 * in error reporting path */
2247 buf = fsname_static;
2248 buflen = sizeof(fsname_static);
2251 len = strlen(lsi->lsi_lmd->lmd_profile);
2252 ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2253 if (ptr && (strcmp(ptr, "-client") == 0))
2256 if (unlikely(len >= buflen))
2258 strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2264 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2266 char *buf, *path = NULL;
2267 struct dentry *dentry = NULL;
2268 struct ccc_object *obj = cl_inode2ccc(page->mapping->host);
2270 /* this can be called inside spin lock so use GFP_ATOMIC. */
2271 buf = (char *)__get_free_page(GFP_ATOMIC);
2273 dentry = d_find_alias(page->mapping->host);
2275 path = dentry_path_raw(dentry, buf, PAGE_SIZE);
2279 "%s: dirty page discard: %s/fid: " DFID "/%s may get corrupted (rc %d)\n",
2280 ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2281 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2282 PFID(&obj->cob_header.coh_lu.loh_fid),
2283 (path && !IS_ERR(path)) ? path : "", ioret);
2289 free_page((unsigned long)buf);