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/types.h>
47 #include "../include/lustre_lite.h"
48 #include "../include/lustre_ha.h"
49 #include "../include/lustre_dlm.h"
50 #include "../include/lprocfs_status.h"
51 #include "../include/lustre_disk.h"
52 #include "../include/lustre_param.h"
53 #include "../include/lustre_log.h"
54 #include "../include/cl_object.h"
55 #include "../include/obd_cksum.h"
56 #include "llite_internal.h"
58 struct kmem_cache *ll_file_data_slab;
59 struct proc_dir_entry *proc_lustre_fs_root;
61 static LIST_HEAD(ll_super_blocks);
62 static DEFINE_SPINLOCK(ll_sb_lock);
65 #define log2(n) ffz(~(n))
68 static struct ll_sb_info *ll_init_sbi(void)
70 struct ll_sb_info *sbi = NULL;
72 unsigned long lru_page_max;
77 OBD_ALLOC(sbi, sizeof(*sbi));
81 spin_lock_init(&sbi->ll_lock);
82 mutex_init(&sbi->ll_lco.lco_lock);
83 spin_lock_init(&sbi->ll_pp_extent_lock);
84 spin_lock_init(&sbi->ll_process_lock);
85 sbi->ll_rw_stats_on = 0;
88 pages = si.totalram - si.totalhigh;
89 if (pages >> (20 - PAGE_CACHE_SHIFT) < 512) {
90 lru_page_max = pages / 2;
92 lru_page_max = (pages / 4) * 3;
95 /* initialize lru data */
96 atomic_set(&sbi->ll_cache.ccc_users, 0);
97 sbi->ll_cache.ccc_lru_max = lru_page_max;
98 atomic_set(&sbi->ll_cache.ccc_lru_left, lru_page_max);
99 spin_lock_init(&sbi->ll_cache.ccc_lru_lock);
100 INIT_LIST_HEAD(&sbi->ll_cache.ccc_lru);
102 sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
103 SBI_DEFAULT_READAHEAD_MAX);
104 sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
105 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
106 SBI_DEFAULT_READAHEAD_WHOLE_MAX;
107 INIT_LIST_HEAD(&sbi->ll_conn_chain);
108 INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
110 ll_generate_random_uuid(uuid);
111 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
112 CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
114 spin_lock(&ll_sb_lock);
115 list_add_tail(&sbi->ll_list, &ll_super_blocks);
116 spin_unlock(&ll_sb_lock);
118 sbi->ll_flags |= LL_SBI_VERBOSE;
119 sbi->ll_flags |= LL_SBI_CHECKSUM;
121 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
123 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
124 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
126 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
130 /* metadata statahead is enabled by default */
131 sbi->ll_sa_max = LL_SA_RPC_DEF;
132 atomic_set(&sbi->ll_sa_total, 0);
133 atomic_set(&sbi->ll_sa_wrong, 0);
134 atomic_set(&sbi->ll_agl_total, 0);
135 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
140 static void ll_free_sbi(struct super_block *sb)
142 struct ll_sb_info *sbi = ll_s2sbi(sb);
145 spin_lock(&ll_sb_lock);
146 list_del(&sbi->ll_list);
147 spin_unlock(&ll_sb_lock);
148 OBD_FREE(sbi, sizeof(*sbi));
152 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
153 struct vfsmount *mnt)
155 struct inode *root = NULL;
156 struct ll_sb_info *sbi = ll_s2sbi(sb);
157 struct obd_device *obd;
158 struct obd_capa *oc = NULL;
159 struct obd_statfs *osfs = NULL;
160 struct ptlrpc_request *request = NULL;
161 struct obd_connect_data *data = NULL;
162 struct obd_uuid *uuid;
163 struct md_op_data *op_data;
164 struct lustre_md lmd;
166 int size, err, checksum;
168 obd = class_name2obd(md);
170 CERROR("MD %s: not setup or attached\n", md);
184 if (proc_lustre_fs_root) {
185 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
188 CERROR("could not register mount in /proc/fs/lustre\n");
191 /* indicate the features supported by this client */
192 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
193 OBD_CONNECT_ATTRFID |
194 OBD_CONNECT_VERSION | OBD_CONNECT_BRW_SIZE |
195 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
196 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
197 OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 |
198 OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR |
199 OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH|
200 OBD_CONNECT_EINPROGRESS |
201 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
202 OBD_CONNECT_LAYOUTLOCK |
203 OBD_CONNECT_PINGLESS |
204 OBD_CONNECT_MAX_EASIZE |
205 OBD_CONNECT_FLOCK_DEAD |
206 OBD_CONNECT_DISP_STRIPE;
208 if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
209 data->ocd_connect_flags |= OBD_CONNECT_SOM;
211 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
212 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
213 #ifdef CONFIG_FS_POSIX_ACL
214 data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK;
217 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
218 /* flag mdc connection as lightweight, only used for test
219 * purpose, use with care */
220 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
222 data->ocd_ibits_known = MDS_INODELOCK_FULL;
223 data->ocd_version = LUSTRE_VERSION_CODE;
225 if (sb->s_flags & MS_RDONLY)
226 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
227 if (sbi->ll_flags & LL_SBI_USER_XATTR)
228 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
230 #ifdef HAVE_MS_FLOCK_LOCK
231 /* force vfs to use lustre handler for flock() calls - bug 10743 */
232 sb->s_flags |= MS_FLOCK_LOCK;
234 #ifdef MS_HAS_NEW_AOPS
235 sb->s_flags |= MS_HAS_NEW_AOPS;
238 if (sbi->ll_flags & LL_SBI_FLOCK)
239 sbi->ll_fop = &ll_file_operations_flock;
240 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
241 sbi->ll_fop = &ll_file_operations;
243 sbi->ll_fop = &ll_file_operations_noflock;
246 data->ocd_connect_flags |= OBD_CONNECT_REAL;
247 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
248 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
250 data->ocd_brw_size = MD_MAX_BRW_SIZE;
252 err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
254 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
255 "recovery, of which this client is not a "
256 "part. Please wait for recovery to complete,"
257 " abort, or time out.\n", md);
260 CERROR("cannot connect to %s: rc = %d\n", md, err);
264 sbi->ll_md_exp->exp_connect_data = *data;
266 err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
267 LUSTRE_SEQ_METADATA);
269 CERROR("%s: Can't init metadata layer FID infrastructure, "
270 "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
274 /* For mount, we only need fs info from MDT0, and also in DNE, it
275 * can make sure the client can be mounted as long as MDT0 is
277 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
278 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
279 OBD_STATFS_FOR_MDT0);
281 GOTO(out_md_fid, err);
283 /* This needs to be after statfs to ensure connect has finished.
284 * Note that "data" does NOT contain the valid connect reply.
285 * If connecting to a 1.8 server there will be no LMV device, so
286 * we can access the MDC export directly and exp_connect_flags will
287 * be non-zero, but if accessing an upgraded 2.1 server it will
288 * have the correct flags filled in.
289 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
290 valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
291 if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
292 valid != CLIENT_CONNECT_MDT_REQD) {
295 OBD_ALLOC_WAIT(buf, PAGE_CACHE_SIZE);
296 obd_connect_flags2str(buf, PAGE_CACHE_SIZE,
297 valid ^ CLIENT_CONNECT_MDT_REQD, ",");
298 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
299 "feature(s) needed for correct operation "
300 "of this client (%s). Please upgrade "
301 "server or downgrade client.\n",
302 sbi->ll_md_exp->exp_obd->obd_name, buf);
303 OBD_FREE(buf, PAGE_CACHE_SIZE);
304 GOTO(out_md_fid, err = -EPROTO);
307 size = sizeof(*data);
308 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
309 KEY_CONN_DATA, &size, data, NULL);
311 CERROR("%s: Get connect data failed: rc = %d\n",
312 sbi->ll_md_exp->exp_obd->obd_name, err);
313 GOTO(out_md_fid, err);
316 LASSERT(osfs->os_bsize);
317 sb->s_blocksize = osfs->os_bsize;
318 sb->s_blocksize_bits = log2(osfs->os_bsize);
319 sb->s_magic = LL_SUPER_MAGIC;
320 sb->s_maxbytes = MAX_LFS_FILESIZE;
321 sbi->ll_namelen = osfs->os_namelen;
322 sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
324 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
325 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
326 LCONSOLE_INFO("Disabling user_xattr feature because "
327 "it is not supported on the server\n");
328 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
331 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
333 sb->s_flags |= MS_POSIXACL;
335 sbi->ll_flags |= LL_SBI_ACL;
337 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
339 sb->s_flags &= ~MS_POSIXACL;
341 sbi->ll_flags &= ~LL_SBI_ACL;
344 if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
345 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
346 sbi->ll_flags |= LL_SBI_RMT_CLIENT;
347 LCONSOLE_INFO("client is set as remote by default.\n");
350 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
351 sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
352 LCONSOLE_INFO("client claims to be remote, but server "
353 "rejected, forced to be local.\n");
357 if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
358 LCONSOLE_INFO("client enabled MDS capability!\n");
359 sbi->ll_flags |= LL_SBI_MDS_CAPA;
362 if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
363 LCONSOLE_INFO("client enabled OSS capability!\n");
364 sbi->ll_flags |= LL_SBI_OSS_CAPA;
367 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
368 sbi->ll_flags |= LL_SBI_64BIT_HASH;
370 if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
371 sbi->ll_md_brw_size = data->ocd_brw_size;
373 sbi->ll_md_brw_size = PAGE_CACHE_SIZE;
375 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK) {
376 LCONSOLE_INFO("Layout lock feature supported.\n");
377 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
380 if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
381 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
383 "%s: disabling xattr cache due to unknown maximum xattr size.\n",
386 sbi->ll_flags |= LL_SBI_XATTR_CACHE;
387 sbi->ll_xattr_cache_enabled = 1;
391 obd = class_name2obd(dt);
393 CERROR("DT %s: not setup or attached\n", dt);
394 GOTO(out_md_fid, err = -ENODEV);
397 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
398 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
399 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
400 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
401 OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
402 OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
403 OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
404 OBD_CONNECT_MAXBYTES |
405 OBD_CONNECT_EINPROGRESS |
406 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
407 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS;
409 if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
410 data->ocd_connect_flags |= OBD_CONNECT_SOM;
412 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
413 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
414 * disabled by default, because it can still be enabled on the
415 * fly via /proc. As a consequence, we still need to come to an
416 * agreement on the supported algorithms at connect time */
417 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
419 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
420 data->ocd_cksum_types = OBD_CKSUM_ADLER;
422 data->ocd_cksum_types = cksum_types_supported_client();
425 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
426 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
427 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
429 CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
430 "ocd_grant: %d\n", data->ocd_connect_flags,
431 data->ocd_version, data->ocd_grant);
433 obd->obd_upcall.onu_owner = &sbi->ll_lco;
434 obd->obd_upcall.onu_upcall = cl_ocd_update;
436 data->ocd_brw_size = DT_MAX_BRW_SIZE;
438 err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
441 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
442 "recovery, of which this client is not a "
443 "part. Please wait for recovery to "
444 "complete, abort, or time out.\n", dt);
447 CERROR("%s: Cannot connect to %s: rc = %d\n",
448 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
452 sbi->ll_dt_exp->exp_connect_data = *data;
454 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
455 LUSTRE_SEQ_METADATA);
457 CERROR("%s: Can't init data layer FID infrastructure, "
458 "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
462 mutex_lock(&sbi->ll_lco.lco_lock);
463 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
464 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
465 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
466 mutex_unlock(&sbi->ll_lco.lco_lock);
468 fid_zero(&sbi->ll_root_fid);
469 err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
471 CERROR("cannot mds_connect: rc = %d\n", err);
472 GOTO(out_lock_cn_cb, err);
474 if (!fid_is_sane(&sbi->ll_root_fid)) {
475 CERROR("%s: Invalid root fid "DFID" during mount\n",
476 sbi->ll_md_exp->exp_obd->obd_name,
477 PFID(&sbi->ll_root_fid));
478 GOTO(out_lock_cn_cb, err = -EINVAL);
480 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
482 sb->s_op = &lustre_super_operations;
483 #if THREAD_SIZE >= 8192 /*b=17630*/
484 sb->s_export_op = &lustre_export_operations;
488 * XXX: move this to after cbd setup? */
489 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
490 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
491 valid |= OBD_MD_FLRMTPERM;
492 else if (sbi->ll_flags & LL_SBI_ACL)
493 valid |= OBD_MD_FLACL;
495 OBD_ALLOC_PTR(op_data);
497 GOTO(out_lock_cn_cb, err = -ENOMEM);
499 op_data->op_fid1 = sbi->ll_root_fid;
500 op_data->op_mode = 0;
501 op_data->op_capa1 = oc;
502 op_data->op_valid = valid;
504 err = md_getattr(sbi->ll_md_exp, op_data, &request);
507 OBD_FREE_PTR(op_data);
509 CERROR("%s: md_getattr failed for root: rc = %d\n",
510 sbi->ll_md_exp->exp_obd->obd_name, err);
511 GOTO(out_lock_cn_cb, err);
514 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
515 sbi->ll_md_exp, &lmd);
517 CERROR("failed to understand root inode md: rc = %d\n", err);
518 ptlrpc_req_finished(request);
519 GOTO(out_lock_cn_cb, err);
522 LASSERT(fid_is_sane(&sbi->ll_root_fid));
523 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
524 sbi->ll_flags & LL_SBI_32BIT_API),
526 md_free_lustre_md(sbi->ll_md_exp, &lmd);
527 ptlrpc_req_finished(request);
529 if (root == NULL || IS_ERR(root)) {
531 obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
532 #ifdef CONFIG_FS_POSIX_ACL
534 posix_acl_release(lmd.posix_acl);
535 lmd.posix_acl = NULL;
538 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
540 CERROR("lustre_lite: bad iget4 for root\n");
544 err = ll_close_thread_start(&sbi->ll_lcq);
546 CERROR("cannot start close thread: rc %d\n", err);
550 #ifdef CONFIG_FS_POSIX_ACL
551 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
552 rct_init(&sbi->ll_rct);
553 et_init(&sbi->ll_et);
557 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
558 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
559 KEY_CHECKSUM, sizeof(checksum), &checksum,
563 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
564 KEY_CACHE_SET, sizeof(sbi->ll_cache),
565 &sbi->ll_cache, NULL);
567 sb->s_root = d_make_root(root);
568 if (sb->s_root == NULL) {
569 CERROR("%s: can't make root dentry\n",
570 ll_get_fsname(sb, NULL, 0));
571 GOTO(out_lock_cn_cb, err = -ENOMEM);
574 sbi->ll_sdev_orig = sb->s_dev;
576 /* We set sb->s_dev equal on all lustre clients in order to support
577 * NFS export clustering. NFSD requires that the FSID be the same
579 /* s_dev is also used in lt_compare() to compare two fs, but that is
580 * only a node-local comparison. */
581 uuid = obd_get_uuid(sbi->ll_md_exp);
583 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
584 get_uuid2fsid(uuid->uuid, strlen(uuid->uuid), &sbi->ll_fsid);
596 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
598 obd_disconnect(sbi->ll_dt_exp);
599 sbi->ll_dt_exp = NULL;
600 /* Make sure all OScs are gone, since cl_cache is accessing sbi. */
601 obd_zombie_barrier();
603 obd_fid_fini(sbi->ll_md_exp->exp_obd);
605 obd_disconnect(sbi->ll_md_exp);
606 sbi->ll_md_exp = NULL;
612 lprocfs_unregister_mountpoint(sbi);
616 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
620 *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
622 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
623 KEY_MAX_EASIZE, &size, lmmsize, NULL);
625 CERROR("Get max mdsize error rc %d \n", rc);
630 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
635 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
636 KEY_DEFAULT_EASIZE, &size, lmmsize, NULL);
638 CERROR("Get default mdsize error rc %d\n", rc);
643 int ll_get_max_cookiesize(struct ll_sb_info *sbi, int *lmmsize)
648 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_COOKIESIZE),
649 KEY_MAX_COOKIESIZE, &size, lmmsize, NULL);
651 CERROR("Get max cookiesize error rc %d\n", rc);
656 int ll_get_default_cookiesize(struct ll_sb_info *sbi, int *lmmsize)
661 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_COOKIESIZE),
662 KEY_DEFAULT_COOKIESIZE, &size, lmmsize, NULL);
664 CERROR("Get default cookiesize error rc %d\n", rc);
669 static void ll_dump_inode(struct inode *inode)
671 struct ll_d_hlist_node *tmp;
672 int dentry_count = 0;
674 LASSERT(inode != NULL);
676 ll_d_hlist_for_each(tmp, &inode->i_dentry)
679 CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
680 inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
681 inode->i_mode, atomic_read(&inode->i_count), dentry_count);
684 void lustre_dump_dentry(struct dentry *dentry, int recur)
686 struct list_head *tmp;
689 LASSERT(dentry != NULL);
691 list_for_each(tmp, &dentry->d_subdirs)
694 CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
695 " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
696 dentry->d_name.len, dentry->d_name.name,
697 dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
698 dentry->d_parent, dentry->d_inode, d_count(dentry),
699 dentry->d_flags, dentry->d_fsdata, subdirs);
700 if (dentry->d_inode != NULL)
701 ll_dump_inode(dentry->d_inode);
706 list_for_each(tmp, &dentry->d_subdirs) {
707 struct dentry *d = list_entry(tmp, struct dentry, d_u.d_child);
708 lustre_dump_dentry(d, recur - 1);
712 static void client_common_put_super(struct super_block *sb)
714 struct ll_sb_info *sbi = ll_s2sbi(sb);
716 #ifdef CONFIG_FS_POSIX_ACL
717 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
718 et_fini(&sbi->ll_et);
719 rct_fini(&sbi->ll_rct);
723 ll_close_thread_shutdown(sbi->ll_lcq);
727 list_del(&sbi->ll_conn_chain);
729 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
730 obd_disconnect(sbi->ll_dt_exp);
731 sbi->ll_dt_exp = NULL;
732 /* wait till all OSCs are gone, since cl_cache is accessing sbi.
734 obd_zombie_barrier();
736 lprocfs_unregister_mountpoint(sbi);
738 obd_fid_fini(sbi->ll_md_exp->exp_obd);
739 obd_disconnect(sbi->ll_md_exp);
740 sbi->ll_md_exp = NULL;
743 void ll_kill_super(struct super_block *sb)
745 struct ll_sb_info *sbi;
748 if (!(sb->s_flags & MS_ACTIVE))
752 /* we need to restore s_dev from changed for clustered NFS before put_super
753 * because new kernels have cached s_dev and change sb->s_dev in
754 * put_super not affected real removing devices */
756 sb->s_dev = sbi->ll_sdev_orig;
757 sbi->ll_umounting = 1;
761 static inline int ll_set_opt(const char *opt, char *data, int fl)
763 if (strncmp(opt, data, strlen(opt)) != 0)
769 /* non-client-specific mount options are parsed in lmd_parse */
770 static int ll_options(char *options, int *flags)
773 char *s1 = options, *s2;
778 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
781 CDEBUG(D_SUPER, "next opt=%s\n", s1);
782 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
787 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
792 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
797 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
802 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
807 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
812 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 5, 50, 0)
813 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
815 /* Ignore deprecated mount option. The client will
816 * always try to mount with ACL support, whether this
817 * is used depends on whether server supports it. */
818 LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
819 "mount option 'acl'.\n");
822 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
824 LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
825 "mount option 'noacl'.\n");
829 #warning "{no}acl options have been deprecated since 1.8, please remove them"
831 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
836 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
841 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
847 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
852 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
857 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
862 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
867 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
872 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
877 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
882 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
887 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
892 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
897 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
903 s2 = strchr(s1, ',');
911 void ll_lli_init(struct ll_inode_info *lli)
913 lli->lli_inode_magic = LLI_INODE_MAGIC;
915 lli->lli_ioepoch = 0;
916 lli->lli_maxbytes = MAX_LFS_FILESIZE;
917 spin_lock_init(&lli->lli_lock);
918 lli->lli_posix_acl = NULL;
919 lli->lli_remote_perms = NULL;
920 mutex_init(&lli->lli_rmtperm_mutex);
921 /* Do not set lli_fid, it has been initialized already. */
922 fid_zero(&lli->lli_pfid);
923 INIT_LIST_HEAD(&lli->lli_close_list);
924 INIT_LIST_HEAD(&lli->lli_oss_capas);
925 atomic_set(&lli->lli_open_count, 0);
926 lli->lli_mds_capa = NULL;
927 lli->lli_rmtperm_time = 0;
928 lli->lli_pending_och = NULL;
929 lli->lli_mds_read_och = NULL;
930 lli->lli_mds_write_och = NULL;
931 lli->lli_mds_exec_och = NULL;
932 lli->lli_open_fd_read_count = 0;
933 lli->lli_open_fd_write_count = 0;
934 lli->lli_open_fd_exec_count = 0;
935 mutex_init(&lli->lli_och_mutex);
936 spin_lock_init(&lli->lli_agl_lock);
937 lli->lli_has_smd = false;
938 spin_lock_init(&lli->lli_layout_lock);
939 ll_layout_version_set(lli, LL_LAYOUT_GEN_NONE);
940 lli->lli_clob = NULL;
942 init_rwsem(&lli->lli_xattrs_list_rwsem);
943 mutex_init(&lli->lli_xattrs_enq_lock);
945 LASSERT(lli->lli_vfs_inode.i_mode != 0);
946 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
947 mutex_init(&lli->lli_readdir_mutex);
948 lli->lli_opendir_key = NULL;
950 spin_lock_init(&lli->lli_sa_lock);
951 lli->lli_opendir_pid = 0;
953 mutex_init(&lli->lli_size_mutex);
954 lli->lli_symlink_name = NULL;
955 init_rwsem(&lli->lli_trunc_sem);
956 mutex_init(&lli->lli_write_mutex);
957 init_rwsem(&lli->lli_glimpse_sem);
958 lli->lli_glimpse_time = 0;
959 INIT_LIST_HEAD(&lli->lli_agl_list);
960 lli->lli_agl_index = 0;
961 lli->lli_async_rc = 0;
963 mutex_init(&lli->lli_layout_mutex);
966 static inline int ll_bdi_register(struct backing_dev_info *bdi)
968 static atomic_t ll_bdi_num = ATOMIC_INIT(0);
970 bdi->name = "lustre";
971 return bdi_register(bdi, NULL, "lustre-%d",
972 atomic_inc_return(&ll_bdi_num));
975 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
977 struct lustre_profile *lprof = NULL;
978 struct lustre_sb_info *lsi = s2lsi(sb);
979 struct ll_sb_info *sbi;
980 char *dt = NULL, *md = NULL;
981 char *profilenm = get_profile_name(sb);
982 struct config_llog_instance *cfg;
983 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
984 const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
987 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
993 try_module_get(THIS_MODULE);
995 /* client additional sb info */
996 lsi->lsi_llsbi = sbi = ll_init_sbi();
998 module_put(THIS_MODULE);
1003 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1005 GOTO(out_free, err);
1007 err = bdi_init(&lsi->lsi_bdi);
1009 GOTO(out_free, err);
1010 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1011 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1012 err = ll_bdi_register(&lsi->lsi_bdi);
1014 GOTO(out_free, err);
1016 sb->s_bdi = &lsi->lsi_bdi;
1017 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1018 sb->s_d_op = &ll_d_ops;
1020 /* Generate a string unique to this super, in case some joker tries
1021 to mount the same fs at two mount points.
1022 Use the address of the super itself.*/
1023 cfg->cfg_instance = sb;
1024 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1025 cfg->cfg_callback = class_config_llog_handler;
1026 /* set up client obds */
1027 err = lustre_process_log(sb, profilenm, cfg);
1029 CERROR("Unable to process log: %d\n", err);
1030 GOTO(out_free, err);
1033 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1034 lprof = class_get_profile(profilenm);
1035 if (lprof == NULL) {
1036 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1037 " read from the MGS. Does that filesystem "
1038 "exist?\n", profilenm);
1039 GOTO(out_free, err = -EINVAL);
1041 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1042 lprof->lp_md, lprof->lp_dt);
1044 OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1046 GOTO(out_free, err = -ENOMEM);
1047 sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1049 OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1051 GOTO(out_free, err = -ENOMEM);
1052 sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1054 /* connections, registrations, sb setup */
1055 err = client_common_fill_super(sb, md, dt, mnt);
1059 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1061 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1064 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1065 LCONSOLE_WARN("Mounted %s\n", profilenm);
1069 } /* ll_fill_super */
1071 void ll_put_super(struct super_block *sb)
1073 struct config_llog_instance cfg, params_cfg;
1074 struct obd_device *obd;
1075 struct lustre_sb_info *lsi = s2lsi(sb);
1076 struct ll_sb_info *sbi = ll_s2sbi(sb);
1077 char *profilenm = get_profile_name(sb);
1078 int next, force = 1;
1080 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1082 ll_print_capa_stat(sbi);
1084 cfg.cfg_instance = sb;
1085 lustre_end_log(sb, profilenm, &cfg);
1087 params_cfg.cfg_instance = sb;
1088 lustre_end_log(sb, PARAMS_FILENAME, ¶ms_cfg);
1090 if (sbi->ll_md_exp) {
1091 obd = class_exp2obd(sbi->ll_md_exp);
1093 force = obd->obd_force;
1096 /* We need to set force before the lov_disconnect in
1097 lustre_common_put_super, since l_d cleans up osc's as well. */
1100 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1102 obd->obd_force = force;
1107 /* Only if client_common_fill_super succeeded */
1108 client_common_put_super(sb);
1112 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1113 class_manual_cleanup(obd);
1116 if (sbi->ll_flags & LL_SBI_VERBOSE)
1117 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1120 class_del_profile(profilenm);
1122 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1123 bdi_destroy(&lsi->lsi_bdi);
1124 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1128 lsi->lsi_llsbi = NULL;
1130 lustre_common_put_super(sb);
1132 module_put(THIS_MODULE);
1133 } /* client_put_super */
1135 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1137 struct inode *inode = NULL;
1139 /* NOTE: we depend on atomic igrab() -bzzz */
1140 lock_res_and_lock(lock);
1141 if (lock->l_resource->lr_lvb_inode) {
1142 struct ll_inode_info * lli;
1143 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1144 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1145 inode = igrab(lock->l_resource->lr_lvb_inode);
1147 inode = lock->l_resource->lr_lvb_inode;
1148 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1149 D_WARNING, lock, "lr_lvb_inode %p is "
1150 "bogus: magic %08x",
1151 lock->l_resource->lr_lvb_inode,
1152 lli->lli_inode_magic);
1156 unlock_res_and_lock(lock);
1160 void ll_clear_inode(struct inode *inode)
1162 struct ll_inode_info *lli = ll_i2info(inode);
1163 struct ll_sb_info *sbi = ll_i2sbi(inode);
1165 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1166 inode->i_generation, inode);
1168 if (S_ISDIR(inode->i_mode)) {
1169 /* these should have been cleared in ll_file_release */
1170 LASSERT(lli->lli_opendir_key == NULL);
1171 LASSERT(lli->lli_sai == NULL);
1172 LASSERT(lli->lli_opendir_pid == 0);
1175 spin_lock(&lli->lli_lock);
1176 ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1177 spin_unlock(&lli->lli_lock);
1178 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1180 LASSERT(!lli->lli_open_fd_write_count);
1181 LASSERT(!lli->lli_open_fd_read_count);
1182 LASSERT(!lli->lli_open_fd_exec_count);
1184 if (lli->lli_mds_write_och)
1185 ll_md_real_close(inode, FMODE_WRITE);
1186 if (lli->lli_mds_exec_och)
1187 ll_md_real_close(inode, FMODE_EXEC);
1188 if (lli->lli_mds_read_och)
1189 ll_md_real_close(inode, FMODE_READ);
1191 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1192 OBD_FREE(lli->lli_symlink_name,
1193 strlen(lli->lli_symlink_name) + 1);
1194 lli->lli_symlink_name = NULL;
1197 ll_xattr_cache_destroy(inode);
1199 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1200 LASSERT(lli->lli_posix_acl == NULL);
1201 if (lli->lli_remote_perms) {
1202 free_rmtperm_hash(lli->lli_remote_perms);
1203 lli->lli_remote_perms = NULL;
1206 #ifdef CONFIG_FS_POSIX_ACL
1207 else if (lli->lli_posix_acl) {
1208 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1209 LASSERT(lli->lli_remote_perms == NULL);
1210 posix_acl_release(lli->lli_posix_acl);
1211 lli->lli_posix_acl = NULL;
1214 lli->lli_inode_magic = LLI_INODE_DEAD;
1216 ll_clear_inode_capas(inode);
1217 if (!S_ISDIR(inode->i_mode))
1218 LASSERT(list_empty(&lli->lli_agl_list));
1221 * XXX This has to be done before lsm is freed below, because
1222 * cl_object still uses inode lsm.
1224 cl_inode_fini(inode);
1225 lli->lli_has_smd = false;
1228 int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data,
1229 struct md_open_data **mod)
1231 struct lustre_md md;
1232 struct inode *inode = dentry->d_inode;
1233 struct ll_sb_info *sbi = ll_i2sbi(inode);
1234 struct ptlrpc_request *request = NULL;
1237 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1238 LUSTRE_OPC_ANY, NULL);
1239 if (IS_ERR(op_data))
1240 return PTR_ERR(op_data);
1242 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1245 ptlrpc_req_finished(request);
1246 if (rc == -ENOENT) {
1248 /* Unlinked special device node? Or just a race?
1249 * Pretend we done everything. */
1250 if (!S_ISREG(inode->i_mode) &&
1251 !S_ISDIR(inode->i_mode)) {
1252 ia_valid = op_data->op_attr.ia_valid;
1253 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1254 rc = simple_setattr(dentry, &op_data->op_attr);
1255 op_data->op_attr.ia_valid = ia_valid;
1257 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1258 CERROR("md_setattr fails: rc = %d\n", rc);
1263 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1264 sbi->ll_md_exp, &md);
1266 ptlrpc_req_finished(request);
1270 ia_valid = op_data->op_attr.ia_valid;
1271 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1272 * cache is not cleared yet. */
1273 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1274 rc = simple_setattr(dentry, &op_data->op_attr);
1275 op_data->op_attr.ia_valid = ia_valid;
1277 /* Extract epoch data if obtained. */
1278 op_data->op_handle = md.body->handle;
1279 op_data->op_ioepoch = md.body->ioepoch;
1281 ll_update_inode(inode, &md);
1282 ptlrpc_req_finished(request);
1287 /* Close IO epoch and send Size-on-MDS attribute update. */
1288 static int ll_setattr_done_writing(struct inode *inode,
1289 struct md_op_data *op_data,
1290 struct md_open_data *mod)
1292 struct ll_inode_info *lli = ll_i2info(inode);
1295 LASSERT(op_data != NULL);
1296 if (!S_ISREG(inode->i_mode))
1299 CDEBUG(D_INODE, "Epoch %llu closed on "DFID" for truncate\n",
1300 op_data->op_ioepoch, PFID(&lli->lli_fid));
1302 op_data->op_flags = MF_EPOCH_CLOSE;
1303 ll_done_writing_attr(inode, op_data);
1304 ll_pack_inode2opdata(inode, op_data, NULL);
1306 rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1307 if (rc == -EAGAIN) {
1308 /* MDS has instructed us to obtain Size-on-MDS attribute
1309 * from OSTs and send setattr to back to MDS. */
1310 rc = ll_som_update(inode, op_data);
1312 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1318 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1320 struct obd_capa *capa;
1323 if (attr->ia_valid & ATTR_SIZE)
1324 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1326 capa = ll_mdscapa_get(inode);
1328 rc = cl_setattr_ost(inode, attr, capa);
1330 if (attr->ia_valid & ATTR_SIZE)
1331 ll_truncate_free_capa(capa);
1339 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1340 * object(s) determine the file size and mtime. Otherwise, the MDS will
1341 * keep these values until such a time that objects are allocated for it.
1342 * We do the MDS operations first, as it is checking permissions for us.
1343 * We don't to the MDS RPC if there is nothing that we want to store there,
1344 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1345 * going to do an RPC anyways.
1347 * If we are doing a truncate, we will send the mtime and ctime updates
1348 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1349 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1352 * In case of HSMimport, we only set attr on MDS.
1354 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1356 struct inode *inode = dentry->d_inode;
1357 struct ll_inode_info *lli = ll_i2info(inode);
1358 struct md_op_data *op_data = NULL;
1359 struct md_open_data *mod = NULL;
1360 bool file_is_released = false;
1361 int rc = 0, rc1 = 0;
1364 "%s: setattr inode %p/fid:"DFID
1365 " from %llu to %llu, valid %x, hsm_import %d\n",
1366 ll_get_fsname(inode->i_sb, NULL, 0), inode,
1367 PFID(&lli->lli_fid), i_size_read(inode), attr->ia_size,
1368 attr->ia_valid, hsm_import);
1370 if (attr->ia_valid & ATTR_SIZE) {
1371 /* Check new size against VFS/VM file size limit and rlimit */
1372 rc = inode_newsize_ok(inode, attr->ia_size);
1376 /* The maximum Lustre file size is variable, based on the
1377 * OST maximum object size and number of stripes. This
1378 * needs another check in addition to the VFS check above. */
1379 if (attr->ia_size > ll_file_maxbytes(inode)) {
1380 CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
1381 PFID(&lli->lli_fid), attr->ia_size,
1382 ll_file_maxbytes(inode));
1386 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1389 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1390 if (attr->ia_valid & TIMES_SET_FLAGS) {
1391 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1392 !capable(CFS_CAP_FOWNER))
1396 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1397 if (attr->ia_valid & ATTR_CTIME) {
1398 attr->ia_ctime = CURRENT_TIME;
1399 attr->ia_valid |= ATTR_CTIME_SET;
1401 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1402 (attr->ia_valid & ATTR_ATIME)) {
1403 attr->ia_atime = CURRENT_TIME;
1404 attr->ia_valid |= ATTR_ATIME_SET;
1406 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1407 (attr->ia_valid & ATTR_MTIME)) {
1408 attr->ia_mtime = CURRENT_TIME;
1409 attr->ia_valid |= ATTR_MTIME_SET;
1412 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1413 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1414 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1417 /* If we are changing file size, file content is modified, flag it. */
1418 if (attr->ia_valid & ATTR_SIZE) {
1419 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1420 spin_lock(&lli->lli_lock);
1421 lli->lli_flags |= LLIF_DATA_MODIFIED;
1422 spin_unlock(&lli->lli_lock);
1425 /* We always do an MDS RPC, even if we're only changing the size;
1426 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1428 OBD_ALLOC_PTR(op_data);
1429 if (op_data == NULL)
1432 if (!S_ISDIR(inode->i_mode)) {
1433 if (attr->ia_valid & ATTR_SIZE)
1434 inode_dio_write_done(inode);
1435 mutex_unlock(&inode->i_mutex);
1438 memcpy(&op_data->op_attr, attr, sizeof(*attr));
1440 /* Open epoch for truncate. */
1441 if (exp_connect_som(ll_i2mdexp(inode)) &&
1442 (attr->ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1443 op_data->op_flags = MF_EPOCH_OPEN;
1445 /* truncate on a released file must failed with -ENODATA,
1446 * so size must not be set on MDS for released file
1447 * but other attributes must be set
1449 if (S_ISREG(inode->i_mode)) {
1450 struct lov_stripe_md *lsm;
1453 ll_layout_refresh(inode, &gen);
1454 lsm = ccc_inode_lsm_get(inode);
1455 if (lsm && lsm->lsm_pattern & LOV_PATTERN_F_RELEASED)
1456 file_is_released = true;
1457 ccc_inode_lsm_put(inode, lsm);
1460 /* if not in HSM import mode, clear size attr for released file
1461 * we clear the attribute send to MDT in op_data, not the original
1462 * received from caller in attr which is used later to
1463 * decide return code */
1464 if (file_is_released && (attr->ia_valid & ATTR_SIZE) && !hsm_import)
1465 op_data->op_attr.ia_valid &= ~ATTR_SIZE;
1467 rc = ll_md_setattr(dentry, op_data, &mod);
1471 /* truncate failed (only when non HSM import), others succeed */
1472 if (file_is_released) {
1473 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1474 GOTO(out, rc = -ENODATA);
1479 /* RPC to MDT is sent, cancel data modification flag */
1480 if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) {
1481 spin_lock(&lli->lli_lock);
1482 lli->lli_flags &= ~LLIF_DATA_MODIFIED;
1483 spin_unlock(&lli->lli_lock);
1486 ll_ioepoch_open(lli, op_data->op_ioepoch);
1487 if (!S_ISREG(inode->i_mode))
1490 if (attr->ia_valid & (ATTR_SIZE |
1491 ATTR_ATIME | ATTR_ATIME_SET |
1492 ATTR_MTIME | ATTR_MTIME_SET))
1493 /* For truncate and utimes sending attributes to OSTs, setting
1494 * mtime/atime to the past will be performed under PW [0:EOF]
1495 * extent lock (new_size:EOF for truncate). It may seem
1496 * excessive to send mtime/atime updates to OSTs when not
1497 * setting times to past, but it is necessary due to possible
1498 * time de-synchronization between MDT inode and OST objects */
1499 if (attr->ia_valid & ATTR_SIZE)
1500 down_write(&lli->lli_trunc_sem);
1501 rc = ll_setattr_ost(inode, attr);
1502 if (attr->ia_valid & ATTR_SIZE)
1503 up_write(&lli->lli_trunc_sem);
1506 if (op_data->op_ioepoch) {
1507 rc1 = ll_setattr_done_writing(inode, op_data, mod);
1511 ll_finish_md_op_data(op_data);
1513 if (!S_ISDIR(inode->i_mode)) {
1514 mutex_lock(&inode->i_mutex);
1515 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1516 inode_dio_wait(inode);
1519 ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1520 LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1525 int ll_setattr(struct dentry *de, struct iattr *attr)
1527 int mode = de->d_inode->i_mode;
1529 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1530 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1531 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1533 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1534 (ATTR_SIZE|ATTR_MODE)) &&
1535 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1536 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1537 !(attr->ia_mode & S_ISGID))))
1538 attr->ia_valid |= ATTR_FORCE;
1540 if ((attr->ia_valid & ATTR_MODE) &&
1542 !(attr->ia_mode & S_ISUID) &&
1543 !(attr->ia_valid & ATTR_KILL_SUID))
1544 attr->ia_valid |= ATTR_KILL_SUID;
1546 if ((attr->ia_valid & ATTR_MODE) &&
1547 ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1548 !(attr->ia_mode & S_ISGID) &&
1549 !(attr->ia_valid & ATTR_KILL_SGID))
1550 attr->ia_valid |= ATTR_KILL_SGID;
1552 return ll_setattr_raw(de, attr, false);
1555 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1556 __u64 max_age, __u32 flags)
1558 struct ll_sb_info *sbi = ll_s2sbi(sb);
1559 struct obd_statfs obd_osfs;
1562 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1564 CERROR("md_statfs fails: rc = %d\n", rc);
1568 osfs->os_type = sb->s_magic;
1570 CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1571 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1573 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1574 flags |= OBD_STATFS_NODELAY;
1576 rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1578 CERROR("obd_statfs fails: rc = %d\n", rc);
1582 CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1583 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1586 osfs->os_bsize = obd_osfs.os_bsize;
1587 osfs->os_blocks = obd_osfs.os_blocks;
1588 osfs->os_bfree = obd_osfs.os_bfree;
1589 osfs->os_bavail = obd_osfs.os_bavail;
1591 /* If we don't have as many objects free on the OST as inodes
1592 * on the MDS, we reduce the total number of inodes to
1593 * compensate, so that the "inodes in use" number is correct.
1595 if (obd_osfs.os_ffree < osfs->os_ffree) {
1596 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1598 osfs->os_ffree = obd_osfs.os_ffree;
1603 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1605 struct super_block *sb = de->d_sb;
1606 struct obd_statfs osfs;
1609 CDEBUG(D_VFSTRACE, "VFS Op: at %llu jiffies\n", get_jiffies_64());
1610 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1612 /* Some amount of caching on the client is allowed */
1613 rc = ll_statfs_internal(sb, &osfs,
1614 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1619 statfs_unpack(sfs, &osfs);
1621 /* We need to downshift for all 32-bit kernels, because we can't
1622 * tell if the kernel is being called via sys_statfs64() or not.
1623 * Stop before overflowing f_bsize - in which case it is better
1624 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1625 if (sizeof(long) < 8) {
1626 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1629 osfs.os_blocks >>= 1;
1630 osfs.os_bfree >>= 1;
1631 osfs.os_bavail >>= 1;
1635 sfs->f_blocks = osfs.os_blocks;
1636 sfs->f_bfree = osfs.os_bfree;
1637 sfs->f_bavail = osfs.os_bavail;
1638 sfs->f_fsid = ll_s2sbi(sb)->ll_fsid;
1642 void ll_inode_size_lock(struct inode *inode)
1644 struct ll_inode_info *lli;
1646 LASSERT(!S_ISDIR(inode->i_mode));
1648 lli = ll_i2info(inode);
1649 mutex_lock(&lli->lli_size_mutex);
1652 void ll_inode_size_unlock(struct inode *inode)
1654 struct ll_inode_info *lli;
1656 lli = ll_i2info(inode);
1657 mutex_unlock(&lli->lli_size_mutex);
1660 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1662 struct ll_inode_info *lli = ll_i2info(inode);
1663 struct mdt_body *body = md->body;
1664 struct lov_stripe_md *lsm = md->lsm;
1665 struct ll_sb_info *sbi = ll_i2sbi(inode);
1667 LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1669 if (!lli->lli_has_smd &&
1670 !(sbi->ll_flags & LL_SBI_LAYOUT_LOCK))
1671 cl_file_inode_init(inode, md);
1673 lli->lli_maxbytes = lsm->lsm_maxbytes;
1674 if (lli->lli_maxbytes > MAX_LFS_FILESIZE)
1675 lli->lli_maxbytes = MAX_LFS_FILESIZE;
1678 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1679 if (body->valid & OBD_MD_FLRMTPERM)
1680 ll_update_remote_perm(inode, md->remote_perm);
1682 #ifdef CONFIG_FS_POSIX_ACL
1683 else if (body->valid & OBD_MD_FLACL) {
1684 spin_lock(&lli->lli_lock);
1685 if (lli->lli_posix_acl)
1686 posix_acl_release(lli->lli_posix_acl);
1687 lli->lli_posix_acl = md->posix_acl;
1688 spin_unlock(&lli->lli_lock);
1691 inode->i_ino = cl_fid_build_ino(&body->fid1,
1692 sbi->ll_flags & LL_SBI_32BIT_API);
1693 inode->i_generation = cl_fid_build_gen(&body->fid1);
1695 if (body->valid & OBD_MD_FLATIME) {
1696 if (body->atime > LTIME_S(inode->i_atime))
1697 LTIME_S(inode->i_atime) = body->atime;
1698 lli->lli_lvb.lvb_atime = body->atime;
1700 if (body->valid & OBD_MD_FLMTIME) {
1701 if (body->mtime > LTIME_S(inode->i_mtime)) {
1702 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to %llu\n",
1703 inode->i_ino, LTIME_S(inode->i_mtime),
1705 LTIME_S(inode->i_mtime) = body->mtime;
1707 lli->lli_lvb.lvb_mtime = body->mtime;
1709 if (body->valid & OBD_MD_FLCTIME) {
1710 if (body->ctime > LTIME_S(inode->i_ctime))
1711 LTIME_S(inode->i_ctime) = body->ctime;
1712 lli->lli_lvb.lvb_ctime = body->ctime;
1714 if (body->valid & OBD_MD_FLMODE)
1715 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1716 if (body->valid & OBD_MD_FLTYPE)
1717 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1718 LASSERT(inode->i_mode != 0);
1719 if (S_ISREG(inode->i_mode)) {
1720 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1722 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1724 if (body->valid & OBD_MD_FLUID)
1725 inode->i_uid = make_kuid(&init_user_ns, body->uid);
1726 if (body->valid & OBD_MD_FLGID)
1727 inode->i_gid = make_kgid(&init_user_ns, body->gid);
1728 if (body->valid & OBD_MD_FLFLAGS)
1729 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1730 if (body->valid & OBD_MD_FLNLINK)
1731 set_nlink(inode, body->nlink);
1732 if (body->valid & OBD_MD_FLRDEV)
1733 inode->i_rdev = old_decode_dev(body->rdev);
1735 if (body->valid & OBD_MD_FLID) {
1736 /* FID shouldn't be changed! */
1737 if (fid_is_sane(&lli->lli_fid)) {
1738 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1739 "Trying to change FID "DFID
1740 " to the "DFID", inode %lu/%u(%p)\n",
1741 PFID(&lli->lli_fid), PFID(&body->fid1),
1742 inode->i_ino, inode->i_generation, inode);
1744 lli->lli_fid = body->fid1;
1747 LASSERT(fid_seq(&lli->lli_fid) != 0);
1749 if (body->valid & OBD_MD_FLSIZE) {
1750 if (exp_connect_som(ll_i2mdexp(inode)) &&
1751 S_ISREG(inode->i_mode)) {
1752 struct lustre_handle lockh;
1755 /* As it is possible a blocking ast has been processed
1756 * by this time, we need to check there is an UPDATE
1757 * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1759 mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1760 &lockh, LDLM_FL_CBPENDING,
1764 if (lli->lli_flags & (LLIF_DONE_WRITING |
1765 LLIF_EPOCH_PENDING |
1767 CERROR("ino %lu flags %u still has "
1768 "size authority! do not trust "
1769 "the size got from MDS\n",
1770 inode->i_ino, lli->lli_flags);
1772 /* Use old size assignment to avoid
1773 * deadlock bz14138 & bz14326 */
1774 i_size_write(inode, body->size);
1775 spin_lock(&lli->lli_lock);
1776 lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1777 spin_unlock(&lli->lli_lock);
1779 ldlm_lock_decref(&lockh, mode);
1782 /* Use old size assignment to avoid
1783 * deadlock bz14138 & bz14326 */
1784 i_size_write(inode, body->size);
1786 CDEBUG(D_VFSTRACE, "inode=%lu, updating i_size %llu\n",
1787 inode->i_ino, (unsigned long long)body->size);
1790 if (body->valid & OBD_MD_FLBLOCKS)
1791 inode->i_blocks = body->blocks;
1794 if (body->valid & OBD_MD_FLMDSCAPA) {
1795 LASSERT(md->mds_capa);
1796 ll_add_capa(inode, md->mds_capa);
1798 if (body->valid & OBD_MD_FLOSSCAPA) {
1799 LASSERT(md->oss_capa);
1800 ll_add_capa(inode, md->oss_capa);
1803 if (body->valid & OBD_MD_TSTATE) {
1804 if (body->t_state & MS_RESTORE)
1805 lli->lli_flags |= LLIF_FILE_RESTORING;
1809 void ll_read_inode2(struct inode *inode, void *opaque)
1811 struct lustre_md *md = opaque;
1812 struct ll_inode_info *lli = ll_i2info(inode);
1814 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1815 PFID(&lli->lli_fid), inode);
1817 LASSERT(!lli->lli_has_smd);
1819 /* Core attributes from the MDS first. This is a new inode, and
1820 * the VFS doesn't zero times in the core inode so we have to do
1821 * it ourselves. They will be overwritten by either MDS or OST
1822 * attributes - we just need to make sure they aren't newer. */
1823 LTIME_S(inode->i_mtime) = 0;
1824 LTIME_S(inode->i_atime) = 0;
1825 LTIME_S(inode->i_ctime) = 0;
1827 ll_update_inode(inode, md);
1829 /* OIDEBUG(inode); */
1831 /* initializing backing dev info. */
1832 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1835 if (S_ISREG(inode->i_mode)) {
1836 struct ll_sb_info *sbi = ll_i2sbi(inode);
1837 inode->i_op = &ll_file_inode_operations;
1838 inode->i_fop = sbi->ll_fop;
1839 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1840 } else if (S_ISDIR(inode->i_mode)) {
1841 inode->i_op = &ll_dir_inode_operations;
1842 inode->i_fop = &ll_dir_operations;
1843 } else if (S_ISLNK(inode->i_mode)) {
1844 inode->i_op = &ll_fast_symlink_inode_operations;
1846 inode->i_op = &ll_special_inode_operations;
1848 init_special_inode(inode, inode->i_mode,
1853 void ll_delete_inode(struct inode *inode)
1855 struct cl_inode_info *lli = cl_i2info(inode);
1857 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
1858 /* discard all dirty pages before truncating them, required by
1859 * osc_extent implementation at LU-1030. */
1860 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF,
1861 CL_FSYNC_DISCARD, 1);
1863 truncate_inode_pages_final(&inode->i_data);
1865 /* Workaround for LU-118 */
1866 if (inode->i_data.nrpages) {
1867 TREE_READ_LOCK_IRQ(&inode->i_data);
1868 TREE_READ_UNLOCK_IRQ(&inode->i_data);
1869 LASSERTF(inode->i_data.nrpages == 0,
1870 "inode=%lu/%u(%p) nrpages=%lu, see "
1871 "http://jira.whamcloud.com/browse/LU-118\n",
1872 inode->i_ino, inode->i_generation, inode,
1873 inode->i_data.nrpages);
1875 /* Workaround end */
1877 ll_clear_inode(inode);
1881 int ll_iocontrol(struct inode *inode, struct file *file,
1882 unsigned int cmd, unsigned long arg)
1884 struct ll_sb_info *sbi = ll_i2sbi(inode);
1885 struct ptlrpc_request *req = NULL;
1889 case FSFILT_IOC_GETFLAGS: {
1890 struct mdt_body *body;
1891 struct md_op_data *op_data;
1893 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
1894 0, 0, LUSTRE_OPC_ANY,
1896 if (IS_ERR(op_data))
1897 return PTR_ERR(op_data);
1899 op_data->op_valid = OBD_MD_FLFLAGS;
1900 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
1901 ll_finish_md_op_data(op_data);
1903 CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1907 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1909 flags = body->flags;
1911 ptlrpc_req_finished(req);
1913 return put_user(flags, (int *)arg);
1915 case FSFILT_IOC_SETFLAGS: {
1916 struct lov_stripe_md *lsm;
1917 struct obd_info oinfo = { { { 0 } } };
1918 struct md_op_data *op_data;
1920 if (get_user(flags, (int *)arg))
1923 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1924 LUSTRE_OPC_ANY, NULL);
1925 if (IS_ERR(op_data))
1926 return PTR_ERR(op_data);
1928 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1929 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1930 rc = md_setattr(sbi->ll_md_exp, op_data,
1931 NULL, 0, NULL, 0, &req, NULL);
1932 ll_finish_md_op_data(op_data);
1933 ptlrpc_req_finished(req);
1937 inode->i_flags = ll_ext_to_inode_flags(flags);
1939 lsm = ccc_inode_lsm_get(inode);
1940 if (!lsm_has_objects(lsm)) {
1941 ccc_inode_lsm_put(inode, lsm);
1945 OBDO_ALLOC(oinfo.oi_oa);
1947 ccc_inode_lsm_put(inode, lsm);
1951 oinfo.oi_oa->o_oi = lsm->lsm_oi;
1952 oinfo.oi_oa->o_flags = flags;
1953 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1955 oinfo.oi_capa = ll_mdscapa_get(inode);
1956 obdo_set_parent_fid(oinfo.oi_oa, &ll_i2info(inode)->lli_fid);
1957 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1958 capa_put(oinfo.oi_capa);
1959 OBDO_FREE(oinfo.oi_oa);
1960 ccc_inode_lsm_put(inode, lsm);
1962 if (rc && rc != -EPERM && rc != -EACCES)
1963 CERROR("osc_setattr_async fails: rc = %d\n", rc);
1974 int ll_flush_ctx(struct inode *inode)
1976 struct ll_sb_info *sbi = ll_i2sbi(inode);
1978 CDEBUG(D_SEC, "flush context for user %d\n",
1979 from_kuid(&init_user_ns, current_uid()));
1981 obd_set_info_async(NULL, sbi->ll_md_exp,
1982 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1984 obd_set_info_async(NULL, sbi->ll_dt_exp,
1985 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1990 /* umount -f client means force down, don't save state */
1991 void ll_umount_begin(struct super_block *sb)
1993 struct ll_sb_info *sbi = ll_s2sbi(sb);
1994 struct obd_device *obd;
1995 struct obd_ioctl_data *ioc_data;
1997 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1998 sb->s_count, atomic_read(&sb->s_active));
2000 obd = class_exp2obd(sbi->ll_md_exp);
2002 CERROR("Invalid MDC connection handle %#llx\n",
2003 sbi->ll_md_exp->exp_handle.h_cookie);
2008 obd = class_exp2obd(sbi->ll_dt_exp);
2010 CERROR("Invalid LOV connection handle %#llx\n",
2011 sbi->ll_dt_exp->exp_handle.h_cookie);
2016 OBD_ALLOC_PTR(ioc_data);
2018 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2019 sizeof(*ioc_data), ioc_data, NULL);
2021 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2022 sizeof(*ioc_data), ioc_data, NULL);
2024 OBD_FREE_PTR(ioc_data);
2027 /* Really, we'd like to wait until there are no requests outstanding,
2028 * and then continue. For now, we just invalidate the requests,
2029 * schedule() and sleep one second if needed, and hope.
2034 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2036 struct ll_sb_info *sbi = ll_s2sbi(sb);
2037 char *profilenm = get_profile_name(sb);
2041 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2042 read_only = *flags & MS_RDONLY;
2043 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2044 sizeof(KEY_READ_ONLY),
2045 KEY_READ_ONLY, sizeof(read_only),
2048 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2049 profilenm, read_only ?
2050 "read-only" : "read-write", err);
2055 sb->s_flags |= MS_RDONLY;
2057 sb->s_flags &= ~MS_RDONLY;
2059 if (sbi->ll_flags & LL_SBI_VERBOSE)
2060 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2061 read_only ? "read-only" : "read-write");
2066 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2067 struct super_block *sb, struct lookup_intent *it)
2069 struct ll_sb_info *sbi = NULL;
2070 struct lustre_md md;
2073 LASSERT(*inode || sb);
2074 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2075 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2076 sbi->ll_md_exp, &md);
2081 ll_update_inode(*inode, &md);
2083 LASSERT(sb != NULL);
2086 * At this point server returns to client's same fid as client
2087 * generated for creating. So using ->fid1 is okay here.
2089 LASSERT(fid_is_sane(&md.body->fid1));
2091 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1,
2092 sbi->ll_flags & LL_SBI_32BIT_API),
2094 if (*inode == NULL || IS_ERR(*inode)) {
2095 #ifdef CONFIG_FS_POSIX_ACL
2097 posix_acl_release(md.posix_acl);
2098 md.posix_acl = NULL;
2101 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2103 CERROR("new_inode -fatal: rc %d\n", rc);
2108 /* Handling piggyback layout lock.
2109 * Layout lock can be piggybacked by getattr and open request.
2110 * The lsm can be applied to inode only if it comes with a layout lock
2111 * otherwise correct layout may be overwritten, for example:
2112 * 1. proc1: mdt returns a lsm but not granting layout
2113 * 2. layout was changed by another client
2114 * 3. proc2: refresh layout and layout lock granted
2115 * 4. proc1: to apply a stale layout */
2116 if (it != NULL && it->d.lustre.it_lock_mode != 0) {
2117 struct lustre_handle lockh;
2118 struct ldlm_lock *lock;
2120 lockh.cookie = it->d.lustre.it_lock_handle;
2121 lock = ldlm_handle2lock(&lockh);
2122 LASSERT(lock != NULL);
2123 if (ldlm_has_layout(lock)) {
2124 struct cl_object_conf conf;
2126 memset(&conf, 0, sizeof(conf));
2127 conf.coc_opc = OBJECT_CONF_SET;
2128 conf.coc_inode = *inode;
2129 conf.coc_lock = lock;
2130 conf.u.coc_md = &md;
2131 (void)ll_layout_conf(*inode, &conf);
2133 LDLM_LOCK_PUT(lock);
2138 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2139 md_free_lustre_md(sbi->ll_md_exp, &md);
2143 int ll_obd_statfs(struct inode *inode, void *arg)
2145 struct ll_sb_info *sbi = NULL;
2146 struct obd_export *exp;
2148 struct obd_ioctl_data *data = NULL;
2153 if (!inode || !(sbi = ll_i2sbi(inode)))
2154 GOTO(out_statfs, rc = -EINVAL);
2156 rc = obd_ioctl_getdata(&buf, &len, arg);
2158 GOTO(out_statfs, rc);
2161 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2162 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2163 GOTO(out_statfs, rc = -EINVAL);
2165 if (data->ioc_inllen1 != sizeof(__u32) ||
2166 data->ioc_inllen2 != sizeof(__u32) ||
2167 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2168 data->ioc_plen2 != sizeof(struct obd_uuid))
2169 GOTO(out_statfs, rc = -EINVAL);
2171 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2172 if (type & LL_STATFS_LMV)
2173 exp = sbi->ll_md_exp;
2174 else if (type & LL_STATFS_LOV)
2175 exp = sbi->ll_dt_exp;
2177 GOTO(out_statfs, rc = -ENODEV);
2179 flags = (type & LL_STATFS_NODELAY) ? OBD_STATFS_NODELAY : 0;
2180 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, &flags);
2182 GOTO(out_statfs, rc);
2185 obd_ioctl_freedata(buf, len);
2189 int ll_process_config(struct lustre_cfg *lcfg)
2193 struct lprocfs_static_vars lvars;
2197 lprocfs_llite_init_vars(&lvars);
2199 /* The instance name contains the sb: lustre-client-aacfe000 */
2200 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2201 if (!ptr || !*(++ptr))
2203 if (sscanf(ptr, "%lx", &x) != 1)
2206 /* This better be a real Lustre superblock! */
2207 LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2209 /* Note we have not called client_common_fill_super yet, so
2210 proc fns must be able to handle that! */
2211 rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2218 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2219 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2220 struct inode *i1, struct inode *i2,
2221 const char *name, int namelen,
2222 int mode, __u32 opc, void *data)
2224 LASSERT(i1 != NULL);
2226 if (namelen > ll_i2sbi(i1)->ll_namelen)
2227 return ERR_PTR(-ENAMETOOLONG);
2229 if (op_data == NULL)
2230 OBD_ALLOC_PTR(op_data);
2232 if (op_data == NULL)
2233 return ERR_PTR(-ENOMEM);
2235 ll_i2gids(op_data->op_suppgids, i1, i2);
2236 op_data->op_fid1 = *ll_inode2fid(i1);
2237 op_data->op_capa1 = ll_mdscapa_get(i1);
2240 op_data->op_fid2 = *ll_inode2fid(i2);
2241 op_data->op_capa2 = ll_mdscapa_get(i2);
2243 fid_zero(&op_data->op_fid2);
2244 op_data->op_capa2 = NULL;
2247 op_data->op_name = name;
2248 op_data->op_namelen = namelen;
2249 op_data->op_mode = mode;
2250 op_data->op_mod_time = get_seconds();
2251 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2252 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2253 op_data->op_cap = cfs_curproc_cap_pack();
2254 op_data->op_bias = 0;
2255 op_data->op_cli_flags = 0;
2256 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2257 filename_is_volatile(name, namelen, NULL))
2258 op_data->op_bias |= MDS_CREATE_VOLATILE;
2259 op_data->op_opc = opc;
2260 op_data->op_mds = 0;
2261 op_data->op_data = data;
2263 /* If the file is being opened after mknod() (normally due to NFS)
2264 * try to use the default stripe data from parent directory for
2265 * allocating OST objects. Try to pass the parent FID to MDS. */
2266 if (opc == LUSTRE_OPC_CREATE && i1 == i2 && S_ISREG(i2->i_mode) &&
2267 !ll_i2info(i2)->lli_has_smd) {
2268 struct ll_inode_info *lli = ll_i2info(i2);
2270 spin_lock(&lli->lli_lock);
2271 if (likely(!lli->lli_has_smd && !fid_is_zero(&lli->lli_pfid)))
2272 op_data->op_fid1 = lli->lli_pfid;
2273 spin_unlock(&lli->lli_lock);
2274 /** We ignore parent's capability temporary. */
2277 /* When called by ll_setattr_raw, file is i1. */
2278 if (LLIF_DATA_MODIFIED & ll_i2info(i1)->lli_flags)
2279 op_data->op_bias |= MDS_DATA_MODIFIED;
2284 void ll_finish_md_op_data(struct md_op_data *op_data)
2286 capa_put(op_data->op_capa1);
2287 capa_put(op_data->op_capa2);
2288 OBD_FREE_PTR(op_data);
2291 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2293 struct ll_sb_info *sbi;
2295 LASSERT((seq != NULL) && (dentry != NULL));
2296 sbi = ll_s2sbi(dentry->d_sb);
2298 if (sbi->ll_flags & LL_SBI_NOLCK)
2299 seq_puts(seq, ",nolock");
2301 if (sbi->ll_flags & LL_SBI_FLOCK)
2302 seq_puts(seq, ",flock");
2304 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2305 seq_puts(seq, ",localflock");
2307 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2308 seq_puts(seq, ",user_xattr");
2310 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2311 seq_puts(seq, ",lazystatfs");
2313 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2314 seq_puts(seq, ",user_fid2path");
2320 * Get obd name by cmd, and copy out to user space
2322 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2324 struct ll_sb_info *sbi = ll_i2sbi(inode);
2325 struct obd_device *obd;
2327 if (cmd == OBD_IOC_GETDTNAME)
2328 obd = class_exp2obd(sbi->ll_dt_exp);
2329 else if (cmd == OBD_IOC_GETMDNAME)
2330 obd = class_exp2obd(sbi->ll_md_exp);
2337 if (copy_to_user((void *)arg, obd->obd_name,
2338 strlen(obd->obd_name) + 1))
2345 * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2346 * fsname will be returned in this buffer; otherwise, a static buffer will be
2347 * used to store the fsname and returned to caller.
2349 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2351 static char fsname_static[MTI_NAME_MAXLEN];
2352 struct lustre_sb_info *lsi = s2lsi(sb);
2357 /* this means the caller wants to use static buffer
2358 * and it doesn't care about race. Usually this is
2359 * in error reporting path */
2360 buf = fsname_static;
2361 buflen = sizeof(fsname_static);
2364 len = strlen(lsi->lsi_lmd->lmd_profile);
2365 ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2366 if (ptr && (strcmp(ptr, "-client") == 0))
2369 if (unlikely(len >= buflen))
2371 strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2377 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2384 p.mnt = current->fs->root.mnt;
2386 path = d_path(&p, buf, bufsize);
2392 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2394 char *buf, *path = NULL;
2395 struct dentry *dentry = NULL;
2396 struct ccc_object *obj = cl_inode2ccc(page->mapping->host);
2398 /* this can be called inside spin lock so use GFP_ATOMIC. */
2399 buf = (char *)__get_free_page(GFP_ATOMIC);
2401 dentry = d_find_alias(page->mapping->host);
2403 path = ll_d_path(dentry, buf, PAGE_SIZE);
2407 "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2408 "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2409 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2410 PFID(&obj->cob_header.coh_lu.loh_fid),
2411 (path && !IS_ERR(path)) ? path : "", ioret);
2417 free_page((unsigned long)buf);