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, 2013, 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>
45 #include <linux/version.h>
48 #include <lustre_lite.h>
49 #include <lustre_ha.h>
50 #include <lustre_dlm.h>
51 #include <lprocfs_status.h>
52 #include <lustre_disk.h>
53 #include <lustre_param.h>
54 #include <lustre_log.h>
55 #include <cl_object.h>
56 #include <obd_cksum.h>
57 #include "llite_internal.h"
59 struct kmem_cache *ll_file_data_slab;
61 CFS_LIST_HEAD(ll_super_blocks);
62 DEFINE_SPINLOCK(ll_sb_lock);
64 #ifndef MS_HAS_NEW_AOPS
65 extern struct address_space_operations ll_aops;
67 extern struct address_space_operations_ext ll_aops;
71 #define log2(n) ffz(~(n))
74 static struct ll_sb_info *ll_init_sbi(void)
76 struct ll_sb_info *sbi = NULL;
78 unsigned long lru_page_max;
84 OBD_ALLOC(sbi, sizeof(*sbi));
88 spin_lock_init(&sbi->ll_lock);
89 mutex_init(&sbi->ll_lco.lco_lock);
90 spin_lock_init(&sbi->ll_pp_extent_lock);
91 spin_lock_init(&sbi->ll_process_lock);
92 sbi->ll_rw_stats_on = 0;
95 pages = si.totalram - si.totalhigh;
96 if (pages >> (20 - PAGE_CACHE_SHIFT) < 512) {
97 lru_page_max = pages / 2;
99 lru_page_max = (pages / 4) * 3;
102 /* initialize lru data */
103 cfs_atomic_set(&sbi->ll_cache.ccc_users, 0);
104 sbi->ll_cache.ccc_lru_max = lru_page_max;
105 cfs_atomic_set(&sbi->ll_cache.ccc_lru_left, lru_page_max);
106 spin_lock_init(&sbi->ll_cache.ccc_lru_lock);
107 CFS_INIT_LIST_HEAD(&sbi->ll_cache.ccc_lru);
109 sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
110 SBI_DEFAULT_READAHEAD_MAX);
111 sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
112 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
113 SBI_DEFAULT_READAHEAD_WHOLE_MAX;
114 CFS_INIT_LIST_HEAD(&sbi->ll_conn_chain);
115 CFS_INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
117 ll_generate_random_uuid(uuid);
118 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
119 CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
121 spin_lock(&ll_sb_lock);
122 cfs_list_add_tail(&sbi->ll_list, &ll_super_blocks);
123 spin_unlock(&ll_sb_lock);
125 sbi->ll_flags |= LL_SBI_VERBOSE;
126 #ifdef ENABLE_CHECKSUM
127 sbi->ll_flags |= LL_SBI_CHECKSUM;
130 #ifdef HAVE_LRU_RESIZE_SUPPORT
131 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
134 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
135 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
137 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
141 /* metadata statahead is enabled by default */
142 sbi->ll_sa_max = LL_SA_RPC_DEF;
143 cfs_atomic_set(&sbi->ll_sa_total, 0);
144 cfs_atomic_set(&sbi->ll_sa_wrong, 0);
145 cfs_atomic_set(&sbi->ll_agl_total, 0);
146 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
151 void ll_free_sbi(struct super_block *sb)
153 struct ll_sb_info *sbi = ll_s2sbi(sb);
157 spin_lock(&ll_sb_lock);
158 cfs_list_del(&sbi->ll_list);
159 spin_unlock(&ll_sb_lock);
160 OBD_FREE(sbi, sizeof(*sbi));
165 static struct dentry_operations ll_d_root_ops = {
166 .d_compare = ll_dcompare,
167 .d_revalidate = ll_revalidate_nd,
170 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
171 struct vfsmount *mnt)
173 struct inode *root = 0;
174 struct ll_sb_info *sbi = ll_s2sbi(sb);
175 struct obd_device *obd;
176 struct obd_capa *oc = NULL;
177 struct obd_statfs *osfs = NULL;
178 struct ptlrpc_request *request = NULL;
179 struct obd_connect_data *data = NULL;
180 struct obd_uuid *uuid;
181 struct md_op_data *op_data;
182 struct lustre_md lmd;
184 int size, err, checksum;
187 obd = class_name2obd(md);
189 CERROR("MD %s: not setup or attached\n", md);
203 if (proc_lustre_fs_root) {
204 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
207 CERROR("could not register mount in /proc/fs/lustre\n");
210 /* indicate the features supported by this client */
211 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
212 OBD_CONNECT_ATTRFID |
213 OBD_CONNECT_VERSION | OBD_CONNECT_BRW_SIZE |
214 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
215 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
216 OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 |
217 OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR |
218 OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH|
219 OBD_CONNECT_EINPROGRESS |
220 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
221 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS |
222 OBD_CONNECT_MAX_EASIZE;
224 if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
225 data->ocd_connect_flags |= OBD_CONNECT_SOM;
227 #ifdef HAVE_LRU_RESIZE_SUPPORT
228 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
229 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
231 #ifdef CONFIG_FS_POSIX_ACL
232 data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK;
235 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
236 /* flag mdc connection as lightweight, only used for test
237 * purpose, use with care */
238 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
240 data->ocd_ibits_known = MDS_INODELOCK_FULL;
241 data->ocd_version = LUSTRE_VERSION_CODE;
243 if (sb->s_flags & MS_RDONLY)
244 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
245 if (sbi->ll_flags & LL_SBI_USER_XATTR)
246 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
248 #ifdef HAVE_MS_FLOCK_LOCK
249 /* force vfs to use lustre handler for flock() calls - bug 10743 */
250 sb->s_flags |= MS_FLOCK_LOCK;
252 #ifdef MS_HAS_NEW_AOPS
253 sb->s_flags |= MS_HAS_NEW_AOPS;
256 if (sbi->ll_flags & LL_SBI_FLOCK)
257 sbi->ll_fop = &ll_file_operations_flock;
258 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
259 sbi->ll_fop = &ll_file_operations;
261 sbi->ll_fop = &ll_file_operations_noflock;
264 data->ocd_connect_flags |= OBD_CONNECT_REAL;
265 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
266 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
268 data->ocd_brw_size = MD_MAX_BRW_SIZE;
270 err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
272 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
273 "recovery, of which this client is not a "
274 "part. Please wait for recovery to complete,"
275 " abort, or time out.\n", md);
278 CERROR("cannot connect to %s: rc = %d\n", md, err);
282 sbi->ll_md_exp->exp_connect_data = *data;
284 err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
285 LUSTRE_SEQ_METADATA);
287 CERROR("%s: Can't init metadata layer FID infrastructure, "
288 "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
292 /* For mount, we only need fs info from MDT0, and also in DNE, it
293 * can make sure the client can be mounted as long as MDT0 is
295 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
296 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
297 OBD_STATFS_FOR_MDT0);
299 GOTO(out_md_fid, err);
301 /* This needs to be after statfs to ensure connect has finished.
302 * Note that "data" does NOT contain the valid connect reply.
303 * If connecting to a 1.8 server there will be no LMV device, so
304 * we can access the MDC export directly and exp_connect_flags will
305 * be non-zero, but if accessing an upgraded 2.1 server it will
306 * have the correct flags filled in.
307 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
308 valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
309 if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
310 valid != CLIENT_CONNECT_MDT_REQD) {
313 OBD_ALLOC_WAIT(buf, PAGE_CACHE_SIZE);
314 obd_connect_flags2str(buf, PAGE_CACHE_SIZE,
315 valid ^ CLIENT_CONNECT_MDT_REQD, ",");
316 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
317 "feature(s) needed for correct operation "
318 "of this client (%s). Please upgrade "
319 "server or downgrade client.\n",
320 sbi->ll_md_exp->exp_obd->obd_name, buf);
321 OBD_FREE(buf, PAGE_CACHE_SIZE);
322 GOTO(out_md_fid, err = -EPROTO);
325 size = sizeof(*data);
326 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
327 KEY_CONN_DATA, &size, data, NULL);
329 CERROR("%s: Get connect data failed: rc = %d\n",
330 sbi->ll_md_exp->exp_obd->obd_name, err);
331 GOTO(out_md_fid, err);
334 LASSERT(osfs->os_bsize);
335 sb->s_blocksize = osfs->os_bsize;
336 sb->s_blocksize_bits = log2(osfs->os_bsize);
337 sb->s_magic = LL_SUPER_MAGIC;
338 sb->s_maxbytes = MAX_LFS_FILESIZE;
339 sbi->ll_namelen = osfs->os_namelen;
340 sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
342 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
343 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
344 LCONSOLE_INFO("Disabling user_xattr feature because "
345 "it is not supported on the server\n");
346 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
349 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
351 sb->s_flags |= MS_POSIXACL;
353 sbi->ll_flags |= LL_SBI_ACL;
355 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
357 sb->s_flags &= ~MS_POSIXACL;
359 sbi->ll_flags &= ~LL_SBI_ACL;
362 if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
363 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
364 sbi->ll_flags |= LL_SBI_RMT_CLIENT;
365 LCONSOLE_INFO("client is set as remote by default.\n");
368 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
369 sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
370 LCONSOLE_INFO("client claims to be remote, but server "
371 "rejected, forced to be local.\n");
375 if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
376 LCONSOLE_INFO("client enabled MDS capability!\n");
377 sbi->ll_flags |= LL_SBI_MDS_CAPA;
380 if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
381 LCONSOLE_INFO("client enabled OSS capability!\n");
382 sbi->ll_flags |= LL_SBI_OSS_CAPA;
385 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
386 sbi->ll_flags |= LL_SBI_64BIT_HASH;
388 if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
389 sbi->ll_md_brw_size = data->ocd_brw_size;
391 sbi->ll_md_brw_size = PAGE_CACHE_SIZE;
393 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK) {
394 LCONSOLE_INFO("Layout lock feature supported.\n");
395 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
398 if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
399 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
400 LCONSOLE_INFO("%s: disabling xattr cache due to "
401 "unknown maximum xattr size.\n", dt);
403 sbi->ll_flags |= LL_SBI_XATTR_CACHE;
404 sbi->ll_xattr_cache_enabled = 1;
408 obd = class_name2obd(dt);
410 CERROR("DT %s: not setup or attached\n", dt);
411 GOTO(out_md_fid, err = -ENODEV);
414 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
415 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
416 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
417 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
418 OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
419 OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
420 OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
421 OBD_CONNECT_MAXBYTES |
422 OBD_CONNECT_EINPROGRESS |
423 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
424 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS;
426 if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
427 data->ocd_connect_flags |= OBD_CONNECT_SOM;
429 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
430 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
431 * disabled by default, because it can still be enabled on the
432 * fly via /proc. As a consequence, we still need to come to an
433 * agreement on the supported algorithms at connect time */
434 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
436 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
437 data->ocd_cksum_types = OBD_CKSUM_ADLER;
439 data->ocd_cksum_types = cksum_types_supported_client();
442 #ifdef HAVE_LRU_RESIZE_SUPPORT
443 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
445 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
446 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
448 CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
449 "ocd_grant: %d\n", data->ocd_connect_flags,
450 data->ocd_version, data->ocd_grant);
452 obd->obd_upcall.onu_owner = &sbi->ll_lco;
453 obd->obd_upcall.onu_upcall = cl_ocd_update;
455 data->ocd_brw_size = DT_MAX_BRW_SIZE;
457 err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
460 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
461 "recovery, of which this client is not a "
462 "part. Please wait for recovery to "
463 "complete, abort, or time out.\n", dt);
466 CERROR("%s: Cannot connect to %s: rc = %d\n",
467 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
471 sbi->ll_dt_exp->exp_connect_data = *data;
473 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
474 LUSTRE_SEQ_METADATA);
476 CERROR("%s: Can't init data layer FID infrastructure, "
477 "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
481 mutex_lock(&sbi->ll_lco.lco_lock);
482 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
483 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
484 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
485 mutex_unlock(&sbi->ll_lco.lco_lock);
487 fid_zero(&sbi->ll_root_fid);
488 err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
490 CERROR("cannot mds_connect: rc = %d\n", err);
491 GOTO(out_lock_cn_cb, err);
493 if (!fid_is_sane(&sbi->ll_root_fid)) {
494 CERROR("%s: Invalid root fid "DFID" during mount\n",
495 sbi->ll_md_exp->exp_obd->obd_name,
496 PFID(&sbi->ll_root_fid));
497 GOTO(out_lock_cn_cb, err = -EINVAL);
499 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
501 sb->s_op = &lustre_super_operations;
502 #if THREAD_SIZE >= 8192 /*b=17630*/
503 sb->s_export_op = &lustre_export_operations;
507 * XXX: move this to after cbd setup? */
508 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
509 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
510 valid |= OBD_MD_FLRMTPERM;
511 else if (sbi->ll_flags & LL_SBI_ACL)
512 valid |= OBD_MD_FLACL;
514 OBD_ALLOC_PTR(op_data);
516 GOTO(out_lock_cn_cb, err = -ENOMEM);
518 op_data->op_fid1 = sbi->ll_root_fid;
519 op_data->op_mode = 0;
520 op_data->op_capa1 = oc;
521 op_data->op_valid = valid;
523 err = md_getattr(sbi->ll_md_exp, op_data, &request);
526 OBD_FREE_PTR(op_data);
528 CERROR("%s: md_getattr failed for root: rc = %d\n",
529 sbi->ll_md_exp->exp_obd->obd_name, err);
530 GOTO(out_lock_cn_cb, err);
533 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
534 sbi->ll_md_exp, &lmd);
536 CERROR("failed to understand root inode md: rc = %d\n", err);
537 ptlrpc_req_finished(request);
538 GOTO(out_lock_cn_cb, err);
541 LASSERT(fid_is_sane(&sbi->ll_root_fid));
542 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
543 sbi->ll_flags & LL_SBI_32BIT_API),
545 md_free_lustre_md(sbi->ll_md_exp, &lmd);
546 ptlrpc_req_finished(request);
548 if (root == NULL || IS_ERR(root)) {
550 obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
551 #ifdef CONFIG_FS_POSIX_ACL
553 posix_acl_release(lmd.posix_acl);
554 lmd.posix_acl = NULL;
557 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
559 CERROR("lustre_lite: bad iget4 for root\n");
563 err = ll_close_thread_start(&sbi->ll_lcq);
565 CERROR("cannot start close thread: rc %d\n", err);
569 #ifdef CONFIG_FS_POSIX_ACL
570 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
571 rct_init(&sbi->ll_rct);
572 et_init(&sbi->ll_et);
576 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
577 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
578 KEY_CHECKSUM, sizeof(checksum), &checksum,
582 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
583 KEY_CACHE_SET, sizeof(sbi->ll_cache),
584 &sbi->ll_cache, NULL);
586 sb->s_root = d_make_root(root);
587 if (sb->s_root == NULL) {
588 CERROR("%s: can't make root dentry\n",
589 ll_get_fsname(sb, NULL, 0));
590 GOTO(out_root, err = -ENOMEM);
593 #ifdef HAVE_DCACHE_LOCK
594 sb->s_root->d_op = &ll_d_root_ops;
596 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
597 d_set_d_op(sb->s_root, &ll_d_root_ops);
598 sb->s_d_op = &ll_d_ops;
601 sbi->ll_sdev_orig = sb->s_dev;
603 /* We set sb->s_dev equal on all lustre clients in order to support
604 * NFS export clustering. NFSD requires that the FSID be the same
606 /* s_dev is also used in lt_compare() to compare two fs, but that is
607 * only a node-local comparison. */
608 uuid = obd_get_uuid(sbi->ll_md_exp);
610 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
611 get_uuid2fsid(uuid->uuid, strlen(uuid->uuid), &sbi->ll_fsid);
624 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
626 obd_disconnect(sbi->ll_dt_exp);
627 sbi->ll_dt_exp = NULL;
628 /* Make sure all OScs are gone, since cl_cache is accessing sbi. */
629 obd_zombie_barrier();
631 obd_fid_fini(sbi->ll_md_exp->exp_obd);
633 obd_disconnect(sbi->ll_md_exp);
634 sbi->ll_md_exp = NULL;
640 lprocfs_unregister_mountpoint(sbi);
644 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
648 *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
650 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
651 KEY_MAX_EASIZE, &size, lmmsize, NULL);
653 CERROR("Get max mdsize error rc %d \n", rc);
658 void ll_dump_inode(struct inode *inode)
660 struct ll_d_hlist_node *tmp;
661 int dentry_count = 0;
663 LASSERT(inode != NULL);
665 ll_d_hlist_for_each(tmp, &inode->i_dentry)
668 CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
669 inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
670 inode->i_mode, atomic_read(&inode->i_count), dentry_count);
673 void lustre_dump_dentry(struct dentry *dentry, int recur)
675 struct list_head *tmp;
678 LASSERT(dentry != NULL);
680 list_for_each(tmp, &dentry->d_subdirs)
683 CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
684 " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
685 dentry->d_name.len, dentry->d_name.name,
686 dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
687 dentry->d_parent, dentry->d_inode, d_refcount(dentry),
688 dentry->d_flags, dentry->d_fsdata, subdirs);
689 if (dentry->d_inode != NULL)
690 ll_dump_inode(dentry->d_inode);
695 list_for_each(tmp, &dentry->d_subdirs) {
696 struct dentry *d = list_entry(tmp, struct dentry, d_u.d_child);
697 lustre_dump_dentry(d, recur - 1);
701 void client_common_put_super(struct super_block *sb)
703 struct ll_sb_info *sbi = ll_s2sbi(sb);
706 #ifdef CONFIG_FS_POSIX_ACL
707 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
708 et_fini(&sbi->ll_et);
709 rct_fini(&sbi->ll_rct);
713 ll_close_thread_shutdown(sbi->ll_lcq);
717 cfs_list_del(&sbi->ll_conn_chain);
719 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
720 obd_disconnect(sbi->ll_dt_exp);
721 sbi->ll_dt_exp = NULL;
722 /* wait till all OSCs are gone, since cl_cache is accessing sbi.
724 obd_zombie_barrier();
726 lprocfs_unregister_mountpoint(sbi);
728 obd_fid_fini(sbi->ll_md_exp->exp_obd);
729 obd_disconnect(sbi->ll_md_exp);
730 sbi->ll_md_exp = NULL;
735 void ll_kill_super(struct super_block *sb)
737 struct ll_sb_info *sbi;
742 if (!(sb->s_flags & MS_ACTIVE))
746 /* we need restore s_dev from changed for clustred NFS before put_super
747 * because new kernels have cached s_dev and change sb->s_dev in
748 * put_super not affected real removing devices */
750 sb->s_dev = sbi->ll_sdev_orig;
751 sbi->ll_umounting = 1;
756 char *ll_read_opt(const char *opt, char *data)
762 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
763 if (strncmp(opt, data, strlen(opt)))
765 if ((value = strchr(data, '=')) == NULL)
769 OBD_ALLOC(retval, strlen(value) + 1);
771 CERROR("out of memory!\n");
775 memcpy(retval, value, strlen(value)+1);
776 CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
780 static inline int ll_set_opt(const char *opt, char *data, int fl)
782 if (strncmp(opt, data, strlen(opt)) != 0)
788 /* non-client-specific mount options are parsed in lmd_parse */
789 static int ll_options(char *options, int *flags)
792 char *s1 = options, *s2;
798 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
801 CDEBUG(D_SUPER, "next opt=%s\n", s1);
802 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
807 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
812 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
817 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
822 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
827 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
832 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 5, 50, 0)
833 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
835 /* Ignore deprecated mount option. The client will
836 * always try to mount with ACL support, whether this
837 * is used depends on whether server supports it. */
838 LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
839 "mount option 'acl'.\n");
842 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
844 LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
845 "mount option 'noacl'.\n");
849 #warning "{no}acl options have been deprecated since 1.8, please remove them"
851 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
856 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
861 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
867 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
872 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
877 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
882 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
887 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
892 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
897 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
902 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
907 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
912 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
917 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
923 s2 = strchr(s1, ',');
931 void ll_lli_init(struct ll_inode_info *lli)
933 lli->lli_inode_magic = LLI_INODE_MAGIC;
935 lli->lli_ioepoch = 0;
936 lli->lli_maxbytes = MAX_LFS_FILESIZE;
937 spin_lock_init(&lli->lli_lock);
938 lli->lli_posix_acl = NULL;
939 lli->lli_remote_perms = NULL;
940 mutex_init(&lli->lli_rmtperm_mutex);
941 /* Do not set lli_fid, it has been initialized already. */
942 fid_zero(&lli->lli_pfid);
943 CFS_INIT_LIST_HEAD(&lli->lli_close_list);
944 CFS_INIT_LIST_HEAD(&lli->lli_oss_capas);
945 cfs_atomic_set(&lli->lli_open_count, 0);
946 lli->lli_mds_capa = NULL;
947 lli->lli_rmtperm_time = 0;
948 lli->lli_pending_och = NULL;
949 lli->lli_mds_read_och = NULL;
950 lli->lli_mds_write_och = NULL;
951 lli->lli_mds_exec_och = NULL;
952 lli->lli_open_fd_read_count = 0;
953 lli->lli_open_fd_write_count = 0;
954 lli->lli_open_fd_exec_count = 0;
955 mutex_init(&lli->lli_och_mutex);
956 spin_lock_init(&lli->lli_agl_lock);
957 lli->lli_has_smd = false;
958 lli->lli_layout_gen = LL_LAYOUT_GEN_NONE;
959 lli->lli_clob = NULL;
961 init_rwsem(&lli->lli_xattrs_list_rwsem);
962 mutex_init(&lli->lli_xattrs_enq_lock);
964 LASSERT(lli->lli_vfs_inode.i_mode != 0);
965 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
966 mutex_init(&lli->lli_readdir_mutex);
967 lli->lli_opendir_key = NULL;
969 lli->lli_def_acl = NULL;
970 spin_lock_init(&lli->lli_sa_lock);
971 lli->lli_opendir_pid = 0;
973 sema_init(&lli->lli_size_sem, 1);
974 lli->lli_size_sem_owner = NULL;
975 lli->lli_symlink_name = NULL;
976 init_rwsem(&lli->lli_trunc_sem);
977 mutex_init(&lli->lli_write_mutex);
978 init_rwsem(&lli->lli_glimpse_sem);
979 lli->lli_glimpse_time = 0;
980 CFS_INIT_LIST_HEAD(&lli->lli_agl_list);
981 lli->lli_agl_index = 0;
982 lli->lli_async_rc = 0;
983 lli->lli_volatile = false;
985 mutex_init(&lli->lli_layout_mutex);
988 static inline int ll_bdi_register(struct backing_dev_info *bdi)
990 static atomic_t ll_bdi_num = ATOMIC_INIT(0);
992 bdi->name = "lustre";
993 return bdi_register(bdi, NULL, "lustre-%d",
994 atomic_inc_return(&ll_bdi_num));
997 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
999 struct lustre_profile *lprof = NULL;
1000 struct lustre_sb_info *lsi = s2lsi(sb);
1001 struct ll_sb_info *sbi;
1002 char *dt = NULL, *md = NULL;
1003 char *profilenm = get_profile_name(sb);
1004 struct config_llog_instance *cfg;
1005 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1006 const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
1010 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1018 /* client additional sb info */
1019 lsi->lsi_llsbi = sbi = ll_init_sbi();
1021 cfs_module_put(THIS_MODULE);
1026 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1028 GOTO(out_free, err);
1030 err = bdi_init(&lsi->lsi_bdi);
1032 GOTO(out_free, err);
1033 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1034 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1035 err = ll_bdi_register(&lsi->lsi_bdi);
1037 GOTO(out_free, err);
1039 sb->s_bdi = &lsi->lsi_bdi;
1041 /* Generate a string unique to this super, in case some joker tries
1042 to mount the same fs at two mount points.
1043 Use the address of the super itself.*/
1044 cfg->cfg_instance = sb;
1045 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1046 cfg->cfg_callback = class_config_llog_handler;
1047 /* set up client obds */
1048 err = lustre_process_log(sb, profilenm, cfg);
1050 CERROR("Unable to process log: %d\n", err);
1051 GOTO(out_free, err);
1054 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1055 lprof = class_get_profile(profilenm);
1056 if (lprof == NULL) {
1057 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1058 " read from the MGS. Does that filesystem "
1059 "exist?\n", profilenm);
1060 GOTO(out_free, err = -EINVAL);
1062 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1063 lprof->lp_md, lprof->lp_dt);
1065 OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1067 GOTO(out_free, err = -ENOMEM);
1068 sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1070 OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1072 GOTO(out_free, err = -ENOMEM);
1073 sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1075 /* connections, registrations, sb setup */
1076 err = client_common_fill_super(sb, md, dt, mnt);
1080 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1082 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1085 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1086 LCONSOLE_WARN("Mounted %s\n", profilenm);
1090 } /* ll_fill_super */
1093 void lu_context_keys_dump(void);
1095 void ll_put_super(struct super_block *sb)
1097 struct config_llog_instance cfg;
1098 struct obd_device *obd;
1099 struct lustre_sb_info *lsi = s2lsi(sb);
1100 struct ll_sb_info *sbi = ll_s2sbi(sb);
1101 char *profilenm = get_profile_name(sb);
1102 int force = 1, next;
1105 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1107 ll_print_capa_stat(sbi);
1109 cfg.cfg_instance = sb;
1110 lustre_end_log(sb, profilenm, &cfg);
1112 if (sbi->ll_md_exp) {
1113 obd = class_exp2obd(sbi->ll_md_exp);
1115 force = obd->obd_force;
1118 /* We need to set force before the lov_disconnect in
1119 lustre_common_put_super, since l_d cleans up osc's as well. */
1122 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1124 obd->obd_force = force;
1129 /* Only if client_common_fill_super succeeded */
1130 client_common_put_super(sb);
1134 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1135 class_manual_cleanup(obd);
1138 if (sbi->ll_flags & LL_SBI_VERBOSE)
1139 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1142 class_del_profile(profilenm);
1144 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1145 bdi_destroy(&lsi->lsi_bdi);
1146 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1150 lsi->lsi_llsbi = NULL;
1152 lustre_common_put_super(sb);
1154 cfs_module_put(THIS_MODULE);
1157 } /* client_put_super */
1159 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1161 struct inode *inode = NULL;
1163 /* NOTE: we depend on atomic igrab() -bzzz */
1164 lock_res_and_lock(lock);
1165 if (lock->l_resource->lr_lvb_inode) {
1166 struct ll_inode_info * lli;
1167 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1168 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1169 inode = igrab(lock->l_resource->lr_lvb_inode);
1171 inode = lock->l_resource->lr_lvb_inode;
1172 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1173 D_WARNING, lock, "lr_lvb_inode %p is "
1174 "bogus: magic %08x",
1175 lock->l_resource->lr_lvb_inode,
1176 lli->lli_inode_magic);
1180 unlock_res_and_lock(lock);
1184 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1186 struct inode *inode = NULL;
1187 /* NOTE: we depend on atomic igrab() -bzzz */
1188 lock_res_and_lock(lock);
1189 if (lock->l_ast_data) {
1190 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1191 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1192 inode = igrab(lock->l_ast_data);
1194 inode = lock->l_ast_data;
1195 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1196 D_WARNING, lock, "l_ast_data %p is "
1197 "bogus: magic %08x", lock->l_ast_data,
1198 lli->lli_inode_magic);
1202 unlock_res_and_lock(lock);
1206 void ll_clear_inode(struct inode *inode)
1208 struct ll_inode_info *lli = ll_i2info(inode);
1209 struct ll_sb_info *sbi = ll_i2sbi(inode);
1212 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1213 inode->i_generation, inode);
1215 if (S_ISDIR(inode->i_mode)) {
1216 /* these should have been cleared in ll_file_release */
1217 LASSERT(lli->lli_opendir_key == NULL);
1218 LASSERT(lli->lli_sai == NULL);
1219 LASSERT(lli->lli_opendir_pid == 0);
1222 spin_lock(&lli->lli_lock);
1223 ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1224 spin_unlock(&lli->lli_lock);
1225 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1227 LASSERT(!lli->lli_open_fd_write_count);
1228 LASSERT(!lli->lli_open_fd_read_count);
1229 LASSERT(!lli->lli_open_fd_exec_count);
1231 if (lli->lli_mds_write_och)
1232 ll_md_real_close(inode, FMODE_WRITE);
1233 if (lli->lli_mds_exec_och)
1234 ll_md_real_close(inode, FMODE_EXEC);
1235 if (lli->lli_mds_read_och)
1236 ll_md_real_close(inode, FMODE_READ);
1238 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1239 OBD_FREE(lli->lli_symlink_name,
1240 strlen(lli->lli_symlink_name) + 1);
1241 lli->lli_symlink_name = NULL;
1244 ll_xattr_cache_destroy(inode);
1246 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1247 LASSERT(lli->lli_posix_acl == NULL);
1248 if (lli->lli_remote_perms) {
1249 free_rmtperm_hash(lli->lli_remote_perms);
1250 lli->lli_remote_perms = NULL;
1253 #ifdef CONFIG_FS_POSIX_ACL
1254 else if (lli->lli_posix_acl) {
1255 LASSERT(cfs_atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1256 LASSERT(lli->lli_remote_perms == NULL);
1257 posix_acl_release(lli->lli_posix_acl);
1258 lli->lli_posix_acl = NULL;
1261 lli->lli_inode_magic = LLI_INODE_DEAD;
1263 ll_clear_inode_capas(inode);
1264 if (!S_ISDIR(inode->i_mode))
1265 LASSERT(cfs_list_empty(&lli->lli_agl_list));
1268 * XXX This has to be done before lsm is freed below, because
1269 * cl_object still uses inode lsm.
1271 cl_inode_fini(inode);
1272 lli->lli_has_smd = false;
1277 int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data,
1278 struct md_open_data **mod)
1280 struct lustre_md md;
1281 struct inode *inode = dentry->d_inode;
1282 struct ll_sb_info *sbi = ll_i2sbi(inode);
1283 struct ptlrpc_request *request = NULL;
1287 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1288 LUSTRE_OPC_ANY, NULL);
1289 if (IS_ERR(op_data))
1290 RETURN(PTR_ERR(op_data));
1292 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1295 ptlrpc_req_finished(request);
1296 if (rc == -ENOENT) {
1298 /* Unlinked special device node? Or just a race?
1299 * Pretend we done everything. */
1300 if (!S_ISREG(inode->i_mode) &&
1301 !S_ISDIR(inode->i_mode)) {
1302 ia_valid = op_data->op_attr.ia_valid;
1303 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1304 rc = simple_setattr(dentry, &op_data->op_attr);
1305 op_data->op_attr.ia_valid = ia_valid;
1307 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1308 CERROR("md_setattr fails: rc = %d\n", rc);
1313 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1314 sbi->ll_md_exp, &md);
1316 ptlrpc_req_finished(request);
1320 ia_valid = op_data->op_attr.ia_valid;
1321 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1322 * cache is not cleared yet. */
1323 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1324 rc = simple_setattr(dentry, &op_data->op_attr);
1325 op_data->op_attr.ia_valid = ia_valid;
1327 /* Extract epoch data if obtained. */
1328 op_data->op_handle = md.body->handle;
1329 op_data->op_ioepoch = md.body->ioepoch;
1331 ll_update_inode(inode, &md);
1332 ptlrpc_req_finished(request);
1337 /* Close IO epoch and send Size-on-MDS attribute update. */
1338 static int ll_setattr_done_writing(struct inode *inode,
1339 struct md_op_data *op_data,
1340 struct md_open_data *mod)
1342 struct ll_inode_info *lli = ll_i2info(inode);
1346 LASSERT(op_data != NULL);
1347 if (!S_ISREG(inode->i_mode))
1350 CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1351 op_data->op_ioepoch, PFID(&lli->lli_fid));
1353 op_data->op_flags = MF_EPOCH_CLOSE;
1354 ll_done_writing_attr(inode, op_data);
1355 ll_pack_inode2opdata(inode, op_data, NULL);
1357 rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1358 if (rc == -EAGAIN) {
1359 /* MDS has instructed us to obtain Size-on-MDS attribute
1360 * from OSTs and send setattr to back to MDS. */
1361 rc = ll_som_update(inode, op_data);
1363 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1369 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1371 struct obd_capa *capa;
1374 if (attr->ia_valid & ATTR_SIZE)
1375 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1377 capa = ll_mdscapa_get(inode);
1379 rc = cl_setattr_ost(inode, attr, capa);
1381 if (attr->ia_valid & ATTR_SIZE)
1382 ll_truncate_free_capa(capa);
1389 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1390 * object(s) determine the file size and mtime. Otherwise, the MDS will
1391 * keep these values until such a time that objects are allocated for it.
1392 * We do the MDS operations first, as it is checking permissions for us.
1393 * We don't to the MDS RPC if there is nothing that we want to store there,
1394 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1395 * going to do an RPC anyways.
1397 * If we are doing a truncate, we will send the mtime and ctime updates
1398 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1399 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1402 * In case of HSMimport, we only set attr on MDS.
1404 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1406 struct inode *inode = dentry->d_inode;
1407 struct ll_inode_info *lli = ll_i2info(inode);
1408 struct md_op_data *op_data = NULL;
1409 struct md_open_data *mod = NULL;
1410 bool file_is_released = false;
1411 int rc = 0, rc1 = 0;
1414 CDEBUG(D_VFSTRACE, "%s: setattr inode %p/fid:"DFID" from %llu to %llu, "
1415 "valid %x, hsm_import %d\n",
1416 ll_get_fsname(inode->i_sb, NULL, 0), inode,
1417 PFID(&lli->lli_fid), i_size_read(inode), attr->ia_size,
1418 attr->ia_valid, hsm_import);
1420 if (attr->ia_valid & ATTR_SIZE) {
1421 /* Check new size against VFS/VM file size limit and rlimit */
1422 rc = inode_newsize_ok(inode, attr->ia_size);
1426 /* The maximum Lustre file size is variable, based on the
1427 * OST maximum object size and number of stripes. This
1428 * needs another check in addition to the VFS check above. */
1429 if (attr->ia_size > ll_file_maxbytes(inode)) {
1430 CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1431 PFID(&lli->lli_fid), attr->ia_size,
1432 ll_file_maxbytes(inode));
1436 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1439 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1440 if (attr->ia_valid & TIMES_SET_FLAGS) {
1441 if (cfs_curproc_fsuid() != inode->i_uid &&
1442 !cfs_capable(CFS_CAP_FOWNER))
1446 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1447 if (attr->ia_valid & ATTR_CTIME) {
1448 attr->ia_ctime = CFS_CURRENT_TIME;
1449 attr->ia_valid |= ATTR_CTIME_SET;
1451 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1452 (attr->ia_valid & ATTR_ATIME)) {
1453 attr->ia_atime = CFS_CURRENT_TIME;
1454 attr->ia_valid |= ATTR_ATIME_SET;
1456 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1457 (attr->ia_valid & ATTR_MTIME)) {
1458 attr->ia_mtime = CFS_CURRENT_TIME;
1459 attr->ia_valid |= ATTR_MTIME_SET;
1462 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1463 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1464 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1465 cfs_time_current_sec());
1467 /* If we are changing file size, file content is modified, flag it. */
1468 if (attr->ia_valid & ATTR_SIZE) {
1469 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1470 spin_lock(&lli->lli_lock);
1471 lli->lli_flags |= LLIF_DATA_MODIFIED;
1472 spin_unlock(&lli->lli_lock);
1475 /* We always do an MDS RPC, even if we're only changing the size;
1476 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1478 OBD_ALLOC_PTR(op_data);
1479 if (op_data == NULL)
1482 if (!S_ISDIR(inode->i_mode)) {
1483 if (attr->ia_valid & ATTR_SIZE)
1484 inode_dio_write_done(inode);
1485 mutex_unlock(&inode->i_mutex);
1486 down_write(&lli->lli_trunc_sem);
1489 memcpy(&op_data->op_attr, attr, sizeof(*attr));
1491 /* Open epoch for truncate. */
1492 if (exp_connect_som(ll_i2mdexp(inode)) &&
1493 (attr->ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1494 op_data->op_flags = MF_EPOCH_OPEN;
1496 /* truncate on a released file must failed with -ENODATA,
1497 * so size must not be set on MDS for released file
1498 * but other attributes must be set
1500 if (S_ISREG(inode->i_mode)) {
1501 struct lov_stripe_md *lsm;
1504 ll_layout_refresh(inode, &gen);
1505 lsm = ccc_inode_lsm_get(inode);
1506 if (lsm && lsm->lsm_pattern & LOV_PATTERN_F_RELEASED)
1507 file_is_released = true;
1508 ccc_inode_lsm_put(inode, lsm);
1511 /* if not in HSM import mode, clear size attr for released file
1512 * we clear the attribute send to MDT in op_data, not the original
1513 * received from caller in attr which is used later to
1514 * decide return code */
1515 if (file_is_released && (attr->ia_valid & ATTR_SIZE) && !hsm_import)
1516 op_data->op_attr.ia_valid &= ~ATTR_SIZE;
1518 rc = ll_md_setattr(dentry, op_data, &mod);
1522 /* truncate failed (only when non HSM import), others succeed */
1523 if (file_is_released) {
1524 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1525 GOTO(out, rc = -ENODATA);
1530 /* RPC to MDT is sent, cancel data modification flag */
1531 if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) {
1532 spin_lock(&lli->lli_lock);
1533 lli->lli_flags &= ~LLIF_DATA_MODIFIED;
1534 spin_unlock(&lli->lli_lock);
1537 ll_ioepoch_open(lli, op_data->op_ioepoch);
1538 if (!S_ISREG(inode->i_mode))
1541 if (attr->ia_valid & (ATTR_SIZE |
1542 ATTR_ATIME | ATTR_ATIME_SET |
1543 ATTR_MTIME | ATTR_MTIME_SET)) {
1544 /* For truncate and utimes sending attributes to OSTs, setting
1545 * mtime/atime to the past will be performed under PW [0:EOF]
1546 * extent lock (new_size:EOF for truncate). It may seem
1547 * excessive to send mtime/atime updates to OSTs when not
1548 * setting times to past, but it is necessary due to possible
1549 * time de-synchronization between MDT inode and OST objects */
1550 rc = ll_setattr_ost(inode, attr);
1555 if (op_data->op_ioepoch) {
1556 rc1 = ll_setattr_done_writing(inode, op_data, mod);
1560 ll_finish_md_op_data(op_data);
1562 if (!S_ISDIR(inode->i_mode)) {
1563 up_write(&lli->lli_trunc_sem);
1564 mutex_lock(&inode->i_mutex);
1565 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1566 inode_dio_wait(inode);
1569 ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1570 LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1575 int ll_setattr(struct dentry *de, struct iattr *attr)
1577 int mode = de->d_inode->i_mode;
1579 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1580 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1581 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1583 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1584 (ATTR_SIZE|ATTR_MODE)) &&
1585 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1586 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1587 !(attr->ia_mode & S_ISGID))))
1588 attr->ia_valid |= ATTR_FORCE;
1590 if ((mode & S_ISUID) &&
1591 !(attr->ia_mode & S_ISUID) &&
1592 !(attr->ia_valid & ATTR_KILL_SUID))
1593 attr->ia_valid |= ATTR_KILL_SUID;
1595 if (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1596 !(attr->ia_mode & S_ISGID) &&
1597 !(attr->ia_valid & ATTR_KILL_SGID))
1598 attr->ia_valid |= ATTR_KILL_SGID;
1600 return ll_setattr_raw(de, attr, false);
1603 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1604 __u64 max_age, __u32 flags)
1606 struct ll_sb_info *sbi = ll_s2sbi(sb);
1607 struct obd_statfs obd_osfs;
1611 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1613 CERROR("md_statfs fails: rc = %d\n", rc);
1617 osfs->os_type = sb->s_magic;
1619 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1620 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1622 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1623 flags |= OBD_STATFS_NODELAY;
1625 rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1627 CERROR("obd_statfs fails: rc = %d\n", rc);
1631 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1632 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1635 osfs->os_bsize = obd_osfs.os_bsize;
1636 osfs->os_blocks = obd_osfs.os_blocks;
1637 osfs->os_bfree = obd_osfs.os_bfree;
1638 osfs->os_bavail = obd_osfs.os_bavail;
1640 /* If we don't have as many objects free on the OST as inodes
1641 * on the MDS, we reduce the total number of inodes to
1642 * compensate, so that the "inodes in use" number is correct.
1644 if (obd_osfs.os_ffree < osfs->os_ffree) {
1645 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1647 osfs->os_ffree = obd_osfs.os_ffree;
1652 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1654 struct super_block *sb = de->d_sb;
1655 struct obd_statfs osfs;
1658 CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1659 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1661 /* Some amount of caching on the client is allowed */
1662 rc = ll_statfs_internal(sb, &osfs,
1663 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1668 statfs_unpack(sfs, &osfs);
1670 /* We need to downshift for all 32-bit kernels, because we can't
1671 * tell if the kernel is being called via sys_statfs64() or not.
1672 * Stop before overflowing f_bsize - in which case it is better
1673 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1674 if (sizeof(long) < 8) {
1675 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1678 osfs.os_blocks >>= 1;
1679 osfs.os_bfree >>= 1;
1680 osfs.os_bavail >>= 1;
1684 sfs->f_blocks = osfs.os_blocks;
1685 sfs->f_bfree = osfs.os_bfree;
1686 sfs->f_bavail = osfs.os_bavail;
1687 sfs->f_fsid = ll_s2sbi(sb)->ll_fsid;
1691 void ll_inode_size_lock(struct inode *inode)
1693 struct ll_inode_info *lli;
1695 LASSERT(!S_ISDIR(inode->i_mode));
1697 lli = ll_i2info(inode);
1698 LASSERT(lli->lli_size_sem_owner != current);
1699 down(&lli->lli_size_sem);
1700 LASSERT(lli->lli_size_sem_owner == NULL);
1701 lli->lli_size_sem_owner = current;
1704 void ll_inode_size_unlock(struct inode *inode)
1706 struct ll_inode_info *lli;
1708 lli = ll_i2info(inode);
1709 LASSERT(lli->lli_size_sem_owner == current);
1710 lli->lli_size_sem_owner = NULL;
1711 up(&lli->lli_size_sem);
1714 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1716 struct ll_inode_info *lli = ll_i2info(inode);
1717 struct mdt_body *body = md->body;
1718 struct lov_stripe_md *lsm = md->lsm;
1719 struct ll_sb_info *sbi = ll_i2sbi(inode);
1721 LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1723 if (!lli->lli_has_smd &&
1724 !(sbi->ll_flags & LL_SBI_LAYOUT_LOCK))
1725 cl_file_inode_init(inode, md);
1727 lli->lli_maxbytes = lsm->lsm_maxbytes;
1728 if (lli->lli_maxbytes > MAX_LFS_FILESIZE)
1729 lli->lli_maxbytes = MAX_LFS_FILESIZE;
1732 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1733 if (body->valid & OBD_MD_FLRMTPERM)
1734 ll_update_remote_perm(inode, md->remote_perm);
1736 #ifdef CONFIG_FS_POSIX_ACL
1737 else if (body->valid & OBD_MD_FLACL) {
1738 spin_lock(&lli->lli_lock);
1739 if (lli->lli_posix_acl)
1740 posix_acl_release(lli->lli_posix_acl);
1741 lli->lli_posix_acl = md->posix_acl;
1742 spin_unlock(&lli->lli_lock);
1745 inode->i_ino = cl_fid_build_ino(&body->fid1,
1746 sbi->ll_flags & LL_SBI_32BIT_API);
1747 inode->i_generation = cl_fid_build_gen(&body->fid1);
1749 if (body->valid & OBD_MD_FLATIME) {
1750 if (body->atime > LTIME_S(inode->i_atime))
1751 LTIME_S(inode->i_atime) = body->atime;
1752 lli->lli_lvb.lvb_atime = body->atime;
1754 if (body->valid & OBD_MD_FLMTIME) {
1755 if (body->mtime > LTIME_S(inode->i_mtime)) {
1756 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1757 "to "LPU64"\n", inode->i_ino,
1758 LTIME_S(inode->i_mtime), body->mtime);
1759 LTIME_S(inode->i_mtime) = body->mtime;
1761 lli->lli_lvb.lvb_mtime = body->mtime;
1763 if (body->valid & OBD_MD_FLCTIME) {
1764 if (body->ctime > LTIME_S(inode->i_ctime))
1765 LTIME_S(inode->i_ctime) = body->ctime;
1766 lli->lli_lvb.lvb_ctime = body->ctime;
1768 if (body->valid & OBD_MD_FLMODE)
1769 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1770 if (body->valid & OBD_MD_FLTYPE)
1771 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1772 LASSERT(inode->i_mode != 0);
1773 if (S_ISREG(inode->i_mode)) {
1774 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1776 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1778 if (body->valid & OBD_MD_FLUID)
1779 inode->i_uid = body->uid;
1780 if (body->valid & OBD_MD_FLGID)
1781 inode->i_gid = body->gid;
1782 if (body->valid & OBD_MD_FLFLAGS)
1783 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1784 if (body->valid & OBD_MD_FLNLINK)
1785 set_nlink(inode, body->nlink);
1786 if (body->valid & OBD_MD_FLRDEV)
1787 inode->i_rdev = old_decode_dev(body->rdev);
1789 if (body->valid & OBD_MD_FLID) {
1790 /* FID shouldn't be changed! */
1791 if (fid_is_sane(&lli->lli_fid)) {
1792 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1793 "Trying to change FID "DFID
1794 " to the "DFID", inode %lu/%u(%p)\n",
1795 PFID(&lli->lli_fid), PFID(&body->fid1),
1796 inode->i_ino, inode->i_generation, inode);
1798 lli->lli_fid = body->fid1;
1801 LASSERT(fid_seq(&lli->lli_fid) != 0);
1803 if (body->valid & OBD_MD_FLSIZE) {
1804 if (exp_connect_som(ll_i2mdexp(inode)) &&
1805 S_ISREG(inode->i_mode)) {
1806 struct lustre_handle lockh;
1809 /* As it is possible a blocking ast has been processed
1810 * by this time, we need to check there is an UPDATE
1811 * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1813 mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1814 &lockh, LDLM_FL_CBPENDING,
1818 if (lli->lli_flags & (LLIF_DONE_WRITING |
1819 LLIF_EPOCH_PENDING |
1821 CERROR("ino %lu flags %u still has "
1822 "size authority! do not trust "
1823 "the size got from MDS\n",
1824 inode->i_ino, lli->lli_flags);
1826 /* Use old size assignment to avoid
1827 * deadlock bz14138 & bz14326 */
1828 i_size_write(inode, body->size);
1829 spin_lock(&lli->lli_lock);
1830 lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1831 spin_unlock(&lli->lli_lock);
1833 ldlm_lock_decref(&lockh, mode);
1836 /* Use old size assignment to avoid
1837 * deadlock bz14138 & bz14326 */
1838 i_size_write(inode, body->size);
1840 CDEBUG(D_VFSTRACE, "inode=%lu, updating i_size %llu\n",
1841 inode->i_ino, (unsigned long long)body->size);
1844 if (body->valid & OBD_MD_FLBLOCKS)
1845 inode->i_blocks = body->blocks;
1848 if (body->valid & OBD_MD_FLMDSCAPA) {
1849 LASSERT(md->mds_capa);
1850 ll_add_capa(inode, md->mds_capa);
1852 if (body->valid & OBD_MD_FLOSSCAPA) {
1853 LASSERT(md->oss_capa);
1854 ll_add_capa(inode, md->oss_capa);
1857 if (body->valid & OBD_MD_TSTATE) {
1858 if (body->t_state & MS_RESTORE)
1859 lli->lli_flags |= LLIF_FILE_RESTORING;
1863 void ll_read_inode2(struct inode *inode, void *opaque)
1865 struct lustre_md *md = opaque;
1866 struct ll_inode_info *lli = ll_i2info(inode);
1869 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1870 PFID(&lli->lli_fid), inode);
1872 LASSERT(!lli->lli_has_smd);
1874 /* Core attributes from the MDS first. This is a new inode, and
1875 * the VFS doesn't zero times in the core inode so we have to do
1876 * it ourselves. They will be overwritten by either MDS or OST
1877 * attributes - we just need to make sure they aren't newer. */
1878 LTIME_S(inode->i_mtime) = 0;
1879 LTIME_S(inode->i_atime) = 0;
1880 LTIME_S(inode->i_ctime) = 0;
1882 ll_update_inode(inode, md);
1884 /* OIDEBUG(inode); */
1886 /* initializing backing dev info. */
1887 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1890 if (S_ISREG(inode->i_mode)) {
1891 struct ll_sb_info *sbi = ll_i2sbi(inode);
1892 inode->i_op = &ll_file_inode_operations;
1893 inode->i_fop = sbi->ll_fop;
1894 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1896 } else if (S_ISDIR(inode->i_mode)) {
1897 inode->i_op = &ll_dir_inode_operations;
1898 inode->i_fop = &ll_dir_operations;
1900 } else if (S_ISLNK(inode->i_mode)) {
1901 inode->i_op = &ll_fast_symlink_inode_operations;
1904 inode->i_op = &ll_special_inode_operations;
1906 init_special_inode(inode, inode->i_mode,
1913 void ll_delete_inode(struct inode *inode)
1915 struct cl_inode_info *lli = cl_i2info(inode);
1918 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
1919 /* discard all dirty pages before truncating them, required by
1920 * osc_extent implementation at LU-1030. */
1921 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF,
1922 CL_FSYNC_DISCARD, 1);
1924 truncate_inode_pages(&inode->i_data, 0);
1926 /* Workaround for LU-118 */
1927 if (inode->i_data.nrpages) {
1928 TREE_READ_LOCK_IRQ(&inode->i_data);
1929 TREE_READ_UNLOCK_IRQ(&inode->i_data);
1930 LASSERTF(inode->i_data.nrpages == 0,
1931 "inode=%lu/%u(%p) nrpages=%lu, see "
1932 "http://jira.whamcloud.com/browse/LU-118\n",
1933 inode->i_ino, inode->i_generation, inode,
1934 inode->i_data.nrpages);
1936 /* Workaround end */
1938 #ifdef HAVE_SBOPS_EVICT_INODE
1939 ll_clear_inode(inode);
1946 int ll_iocontrol(struct inode *inode, struct file *file,
1947 unsigned int cmd, unsigned long arg)
1949 struct ll_sb_info *sbi = ll_i2sbi(inode);
1950 struct ptlrpc_request *req = NULL;
1955 case FSFILT_IOC_GETFLAGS: {
1956 struct mdt_body *body;
1957 struct md_op_data *op_data;
1959 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
1960 0, 0, LUSTRE_OPC_ANY,
1962 if (IS_ERR(op_data))
1963 RETURN(PTR_ERR(op_data));
1965 op_data->op_valid = OBD_MD_FLFLAGS;
1966 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
1967 ll_finish_md_op_data(op_data);
1969 CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1973 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1975 flags = body->flags;
1977 ptlrpc_req_finished(req);
1979 RETURN(put_user(flags, (int *)arg));
1981 case FSFILT_IOC_SETFLAGS: {
1982 struct lov_stripe_md *lsm;
1983 struct obd_info oinfo = { { { 0 } } };
1984 struct md_op_data *op_data;
1986 if (get_user(flags, (int *)arg))
1989 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1990 LUSTRE_OPC_ANY, NULL);
1991 if (IS_ERR(op_data))
1992 RETURN(PTR_ERR(op_data));
1994 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1995 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1996 rc = md_setattr(sbi->ll_md_exp, op_data,
1997 NULL, 0, NULL, 0, &req, NULL);
1998 ll_finish_md_op_data(op_data);
1999 ptlrpc_req_finished(req);
2003 inode->i_flags = ll_ext_to_inode_flags(flags);
2005 lsm = ccc_inode_lsm_get(inode);
2006 if (!lsm_has_objects(lsm)) {
2007 ccc_inode_lsm_put(inode, lsm);
2011 OBDO_ALLOC(oinfo.oi_oa);
2013 ccc_inode_lsm_put(inode, lsm);
2017 oinfo.oi_oa->o_oi = lsm->lsm_oi;
2018 oinfo.oi_oa->o_flags = flags;
2019 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
2021 oinfo.oi_capa = ll_mdscapa_get(inode);
2022 obdo_set_parent_fid(oinfo.oi_oa, &ll_i2info(inode)->lli_fid);
2023 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
2024 capa_put(oinfo.oi_capa);
2025 OBDO_FREE(oinfo.oi_oa);
2026 ccc_inode_lsm_put(inode, lsm);
2028 if (rc && rc != -EPERM && rc != -EACCES)
2029 CERROR("osc_setattr_async fails: rc = %d\n", rc);
2040 int ll_flush_ctx(struct inode *inode)
2042 struct ll_sb_info *sbi = ll_i2sbi(inode);
2044 CDEBUG(D_SEC, "flush context for user %d\n", cfs_curproc_uid());
2046 obd_set_info_async(NULL, sbi->ll_md_exp,
2047 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2049 obd_set_info_async(NULL, sbi->ll_dt_exp,
2050 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2055 /* umount -f client means force down, don't save state */
2056 void ll_umount_begin(struct super_block *sb)
2058 struct ll_sb_info *sbi = ll_s2sbi(sb);
2059 struct obd_device *obd;
2060 struct obd_ioctl_data *ioc_data;
2063 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2064 sb->s_count, atomic_read(&sb->s_active));
2066 obd = class_exp2obd(sbi->ll_md_exp);
2068 CERROR("Invalid MDC connection handle "LPX64"\n",
2069 sbi->ll_md_exp->exp_handle.h_cookie);
2075 obd = class_exp2obd(sbi->ll_dt_exp);
2077 CERROR("Invalid LOV connection handle "LPX64"\n",
2078 sbi->ll_dt_exp->exp_handle.h_cookie);
2084 OBD_ALLOC_PTR(ioc_data);
2086 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2087 sizeof *ioc_data, ioc_data, NULL);
2089 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2090 sizeof *ioc_data, ioc_data, NULL);
2092 OBD_FREE_PTR(ioc_data);
2095 /* Really, we'd like to wait until there are no requests outstanding,
2096 * and then continue. For now, we just invalidate the requests,
2097 * schedule() and sleep one second if needed, and hope.
2103 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2105 struct ll_sb_info *sbi = ll_s2sbi(sb);
2106 char *profilenm = get_profile_name(sb);
2110 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2111 read_only = *flags & MS_RDONLY;
2112 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2113 sizeof(KEY_READ_ONLY),
2114 KEY_READ_ONLY, sizeof(read_only),
2117 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2118 profilenm, read_only ?
2119 "read-only" : "read-write", err);
2124 sb->s_flags |= MS_RDONLY;
2126 sb->s_flags &= ~MS_RDONLY;
2128 if (sbi->ll_flags & LL_SBI_VERBOSE)
2129 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2130 read_only ? "read-only" : "read-write");
2135 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2136 struct super_block *sb, struct lookup_intent *it)
2138 struct ll_sb_info *sbi = NULL;
2139 struct lustre_md md;
2143 LASSERT(*inode || sb);
2144 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2145 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2146 sbi->ll_md_exp, &md);
2151 ll_update_inode(*inode, &md);
2153 LASSERT(sb != NULL);
2156 * At this point server returns to client's same fid as client
2157 * generated for creating. So using ->fid1 is okay here.
2159 LASSERT(fid_is_sane(&md.body->fid1));
2161 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1,
2162 sbi->ll_flags & LL_SBI_32BIT_API),
2164 if (*inode == NULL || IS_ERR(*inode)) {
2165 #ifdef CONFIG_FS_POSIX_ACL
2167 posix_acl_release(md.posix_acl);
2168 md.posix_acl = NULL;
2171 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2173 CERROR("new_inode -fatal: rc %d\n", rc);
2178 /* Handling piggyback layout lock.
2179 * Layout lock can be piggybacked by getattr and open request.
2180 * The lsm can be applied to inode only if it comes with a layout lock
2181 * otherwise correct layout may be overwritten, for example:
2182 * 1. proc1: mdt returns a lsm but not granting layout
2183 * 2. layout was changed by another client
2184 * 3. proc2: refresh layout and layout lock granted
2185 * 4. proc1: to apply a stale layout */
2186 if (it != NULL && it->d.lustre.it_lock_mode != 0) {
2187 struct lustre_handle lockh;
2188 struct ldlm_lock *lock;
2190 lockh.cookie = it->d.lustre.it_lock_handle;
2191 lock = ldlm_handle2lock(&lockh);
2192 LASSERT(lock != NULL);
2193 if (ldlm_has_layout(lock)) {
2194 struct cl_object_conf conf;
2196 memset(&conf, 0, sizeof(conf));
2197 conf.coc_opc = OBJECT_CONF_SET;
2198 conf.coc_inode = *inode;
2199 conf.coc_lock = lock;
2200 conf.u.coc_md = &md;
2201 (void)ll_layout_conf(*inode, &conf);
2203 LDLM_LOCK_PUT(lock);
2208 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2209 md_free_lustre_md(sbi->ll_md_exp, &md);
2213 int ll_obd_statfs(struct inode *inode, void *arg)
2215 struct ll_sb_info *sbi = NULL;
2216 struct obd_export *exp;
2218 struct obd_ioctl_data *data = NULL;
2223 if (!inode || !(sbi = ll_i2sbi(inode)))
2224 GOTO(out_statfs, rc = -EINVAL);
2226 rc = obd_ioctl_getdata(&buf, &len, arg);
2228 GOTO(out_statfs, rc);
2231 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2232 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2233 GOTO(out_statfs, rc = -EINVAL);
2235 if (data->ioc_inllen1 != sizeof(__u32) ||
2236 data->ioc_inllen2 != sizeof(__u32) ||
2237 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2238 data->ioc_plen2 != sizeof(struct obd_uuid))
2239 GOTO(out_statfs, rc = -EINVAL);
2241 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2242 if (type & LL_STATFS_LMV)
2243 exp = sbi->ll_md_exp;
2244 else if (type & LL_STATFS_LOV)
2245 exp = sbi->ll_dt_exp;
2247 GOTO(out_statfs, rc = -ENODEV);
2249 flags = (type & LL_STATFS_NODELAY) ? OBD_STATFS_NODELAY : 0;
2250 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, &flags);
2252 GOTO(out_statfs, rc);
2255 obd_ioctl_freedata(buf, len);
2259 int ll_process_config(struct lustre_cfg *lcfg)
2263 struct lprocfs_static_vars lvars;
2267 lprocfs_llite_init_vars(&lvars);
2269 /* The instance name contains the sb: lustre-client-aacfe000 */
2270 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2271 if (!ptr || !*(++ptr))
2273 if (sscanf(ptr, "%lx", &x) != 1)
2276 /* This better be a real Lustre superblock! */
2277 LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2279 /* Note we have not called client_common_fill_super yet, so
2280 proc fns must be able to handle that! */
2281 rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2288 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2289 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2290 struct inode *i1, struct inode *i2,
2291 const char *name, int namelen,
2292 int mode, __u32 opc, void *data)
2294 LASSERT(i1 != NULL);
2296 if (namelen > ll_i2sbi(i1)->ll_namelen)
2297 return ERR_PTR(-ENAMETOOLONG);
2299 if (op_data == NULL)
2300 OBD_ALLOC_PTR(op_data);
2302 if (op_data == NULL)
2303 return ERR_PTR(-ENOMEM);
2305 ll_i2gids(op_data->op_suppgids, i1, i2);
2306 op_data->op_fid1 = *ll_inode2fid(i1);
2307 op_data->op_capa1 = ll_mdscapa_get(i1);
2310 op_data->op_fid2 = *ll_inode2fid(i2);
2311 op_data->op_capa2 = ll_mdscapa_get(i2);
2313 fid_zero(&op_data->op_fid2);
2314 op_data->op_capa2 = NULL;
2317 op_data->op_name = name;
2318 op_data->op_namelen = namelen;
2319 op_data->op_mode = mode;
2320 op_data->op_mod_time = cfs_time_current_sec();
2321 op_data->op_fsuid = cfs_curproc_fsuid();
2322 op_data->op_fsgid = cfs_curproc_fsgid();
2323 op_data->op_cap = cfs_curproc_cap_pack();
2324 op_data->op_bias = 0;
2325 op_data->op_cli_flags = 0;
2326 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2327 filename_is_volatile(name, namelen, NULL))
2328 op_data->op_bias |= MDS_CREATE_VOLATILE;
2329 op_data->op_opc = opc;
2330 op_data->op_mds = 0;
2331 op_data->op_data = data;
2333 /* If the file is being opened after mknod() (normally due to NFS)
2334 * try to use the default stripe data from parent directory for
2335 * allocating OST objects. Try to pass the parent FID to MDS. */
2336 if (opc == LUSTRE_OPC_CREATE && i1 == i2 && S_ISREG(i2->i_mode) &&
2337 !ll_i2info(i2)->lli_has_smd) {
2338 struct ll_inode_info *lli = ll_i2info(i2);
2340 spin_lock(&lli->lli_lock);
2341 if (likely(!lli->lli_has_smd && !fid_is_zero(&lli->lli_pfid)))
2342 op_data->op_fid1 = lli->lli_pfid;
2343 spin_unlock(&lli->lli_lock);
2344 /** We ignore parent's capability temporary. */
2347 /* When called by ll_setattr_raw, file is i1. */
2348 if (LLIF_DATA_MODIFIED & ll_i2info(i1)->lli_flags)
2349 op_data->op_bias |= MDS_DATA_MODIFIED;
2354 void ll_finish_md_op_data(struct md_op_data *op_data)
2356 capa_put(op_data->op_capa1);
2357 capa_put(op_data->op_capa2);
2358 OBD_FREE_PTR(op_data);
2361 #ifdef HAVE_SUPEROPS_USE_DENTRY
2362 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2364 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2367 struct ll_sb_info *sbi;
2369 #ifdef HAVE_SUPEROPS_USE_DENTRY
2370 LASSERT((seq != NULL) && (dentry != NULL));
2371 sbi = ll_s2sbi(dentry->d_sb);
2373 LASSERT((seq != NULL) && (vfs != NULL));
2374 sbi = ll_s2sbi(vfs->mnt_sb);
2377 if (sbi->ll_flags & LL_SBI_NOLCK)
2378 seq_puts(seq, ",nolock");
2380 if (sbi->ll_flags & LL_SBI_FLOCK)
2381 seq_puts(seq, ",flock");
2383 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2384 seq_puts(seq, ",localflock");
2386 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2387 seq_puts(seq, ",user_xattr");
2389 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2390 seq_puts(seq, ",lazystatfs");
2392 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2393 seq_puts(seq, ",user_fid2path");
2399 * Get obd name by cmd, and copy out to user space
2401 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2403 struct ll_sb_info *sbi = ll_i2sbi(inode);
2404 struct obd_device *obd;
2407 if (cmd == OBD_IOC_GETDTNAME)
2408 obd = class_exp2obd(sbi->ll_dt_exp);
2409 else if (cmd == OBD_IOC_GETMDNAME)
2410 obd = class_exp2obd(sbi->ll_md_exp);
2417 if (copy_to_user((void *)arg, obd->obd_name,
2418 strlen(obd->obd_name) + 1))
2425 * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2426 * fsname will be returned in this buffer; otherwise, a static buffer will be
2427 * used to store the fsname and returned to caller.
2429 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2431 static char fsname_static[MTI_NAME_MAXLEN];
2432 struct lustre_sb_info *lsi = s2lsi(sb);
2437 /* this means the caller wants to use static buffer
2438 * and it doesn't care about race. Usually this is
2439 * in error reporting path */
2440 buf = fsname_static;
2441 buflen = sizeof(fsname_static);
2444 len = strlen(lsi->lsi_lmd->lmd_profile);
2445 ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2446 if (ptr && (strcmp(ptr, "-client") == 0))
2449 if (unlikely(len >= buflen))
2451 strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2457 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2461 #ifdef HAVE_FS_STRUCT_USE_PATH
2465 p.mnt = current->fs->root.mnt;
2467 path = d_path(&p, buf, bufsize);
2470 path = d_path(dentry, current->fs->rootmnt, buf, bufsize);
2476 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2478 char *buf, *path = NULL;
2479 struct dentry *dentry = NULL;
2480 struct ccc_object *obj = cl_inode2ccc(page->mapping->host);
2482 /* this can be called inside spin lock so use GFP_ATOMIC. */
2483 buf = (char *)__get_free_page(GFP_ATOMIC);
2485 dentry = d_find_alias(page->mapping->host);
2487 path = ll_d_path(dentry, buf, PAGE_SIZE);
2490 CWARN("%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2491 "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2492 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2493 PFID(&obj->cob_header.coh_lu.loh_fid),
2494 (path && !IS_ERR(path)) ? path : "", ioret);
2500 free_page((unsigned long)buf);