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 int client_common_fill_super(struct super_block *sb, char *md, char *dt,
166 struct vfsmount *mnt)
168 struct inode *root = 0;
169 struct ll_sb_info *sbi = ll_s2sbi(sb);
170 struct obd_device *obd;
171 struct obd_capa *oc = NULL;
172 struct obd_statfs *osfs = NULL;
173 struct ptlrpc_request *request = NULL;
174 struct obd_connect_data *data = NULL;
175 struct obd_uuid *uuid;
176 struct md_op_data *op_data;
177 struct lustre_md lmd;
179 int size, err, checksum;
182 obd = class_name2obd(md);
184 CERROR("MD %s: not setup or attached\n", md);
198 if (proc_lustre_fs_root) {
199 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
202 CERROR("could not register mount in /proc/fs/lustre\n");
205 /* indicate the features supported by this client */
206 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
207 OBD_CONNECT_ATTRFID |
208 OBD_CONNECT_VERSION | OBD_CONNECT_BRW_SIZE |
209 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
210 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
211 OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 |
212 OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR |
213 OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH|
214 OBD_CONNECT_EINPROGRESS |
215 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
216 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS |
217 OBD_CONNECT_MAX_EASIZE |
218 OBD_CONNECT_FLOCK_DEAD;
220 if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
221 data->ocd_connect_flags |= OBD_CONNECT_SOM;
223 #ifdef HAVE_LRU_RESIZE_SUPPORT
224 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
225 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
227 #ifdef CONFIG_FS_POSIX_ACL
228 data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK;
231 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
232 /* flag mdc connection as lightweight, only used for test
233 * purpose, use with care */
234 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
236 data->ocd_ibits_known = MDS_INODELOCK_FULL;
237 data->ocd_version = LUSTRE_VERSION_CODE;
239 if (sb->s_flags & MS_RDONLY)
240 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
241 if (sbi->ll_flags & LL_SBI_USER_XATTR)
242 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
244 #ifdef HAVE_MS_FLOCK_LOCK
245 /* force vfs to use lustre handler for flock() calls - bug 10743 */
246 sb->s_flags |= MS_FLOCK_LOCK;
248 #ifdef MS_HAS_NEW_AOPS
249 sb->s_flags |= MS_HAS_NEW_AOPS;
252 if (sbi->ll_flags & LL_SBI_FLOCK)
253 sbi->ll_fop = &ll_file_operations_flock;
254 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
255 sbi->ll_fop = &ll_file_operations;
257 sbi->ll_fop = &ll_file_operations_noflock;
260 data->ocd_connect_flags |= OBD_CONNECT_REAL;
261 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
262 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
264 data->ocd_brw_size = MD_MAX_BRW_SIZE;
266 err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
268 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
269 "recovery, of which this client is not a "
270 "part. Please wait for recovery to complete,"
271 " abort, or time out.\n", md);
274 CERROR("cannot connect to %s: rc = %d\n", md, err);
278 sbi->ll_md_exp->exp_connect_data = *data;
280 err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
281 LUSTRE_SEQ_METADATA);
283 CERROR("%s: Can't init metadata layer FID infrastructure, "
284 "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
288 /* For mount, we only need fs info from MDT0, and also in DNE, it
289 * can make sure the client can be mounted as long as MDT0 is
291 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
292 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
293 OBD_STATFS_FOR_MDT0);
295 GOTO(out_md_fid, err);
297 /* This needs to be after statfs to ensure connect has finished.
298 * Note that "data" does NOT contain the valid connect reply.
299 * If connecting to a 1.8 server there will be no LMV device, so
300 * we can access the MDC export directly and exp_connect_flags will
301 * be non-zero, but if accessing an upgraded 2.1 server it will
302 * have the correct flags filled in.
303 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
304 valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
305 if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
306 valid != CLIENT_CONNECT_MDT_REQD) {
309 OBD_ALLOC_WAIT(buf, PAGE_CACHE_SIZE);
310 obd_connect_flags2str(buf, PAGE_CACHE_SIZE,
311 valid ^ CLIENT_CONNECT_MDT_REQD, ",");
312 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
313 "feature(s) needed for correct operation "
314 "of this client (%s). Please upgrade "
315 "server or downgrade client.\n",
316 sbi->ll_md_exp->exp_obd->obd_name, buf);
317 OBD_FREE(buf, PAGE_CACHE_SIZE);
318 GOTO(out_md_fid, err = -EPROTO);
321 size = sizeof(*data);
322 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
323 KEY_CONN_DATA, &size, data, NULL);
325 CERROR("%s: Get connect data failed: rc = %d\n",
326 sbi->ll_md_exp->exp_obd->obd_name, err);
327 GOTO(out_md_fid, err);
330 LASSERT(osfs->os_bsize);
331 sb->s_blocksize = osfs->os_bsize;
332 sb->s_blocksize_bits = log2(osfs->os_bsize);
333 sb->s_magic = LL_SUPER_MAGIC;
334 sb->s_maxbytes = MAX_LFS_FILESIZE;
335 sbi->ll_namelen = osfs->os_namelen;
336 sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
338 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
339 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
340 LCONSOLE_INFO("Disabling user_xattr feature because "
341 "it is not supported on the server\n");
342 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
345 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
347 sb->s_flags |= MS_POSIXACL;
349 sbi->ll_flags |= LL_SBI_ACL;
351 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
353 sb->s_flags &= ~MS_POSIXACL;
355 sbi->ll_flags &= ~LL_SBI_ACL;
358 if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
359 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
360 sbi->ll_flags |= LL_SBI_RMT_CLIENT;
361 LCONSOLE_INFO("client is set as remote by default.\n");
364 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
365 sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
366 LCONSOLE_INFO("client claims to be remote, but server "
367 "rejected, forced to be local.\n");
371 if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
372 LCONSOLE_INFO("client enabled MDS capability!\n");
373 sbi->ll_flags |= LL_SBI_MDS_CAPA;
376 if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
377 LCONSOLE_INFO("client enabled OSS capability!\n");
378 sbi->ll_flags |= LL_SBI_OSS_CAPA;
381 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
382 sbi->ll_flags |= LL_SBI_64BIT_HASH;
384 if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
385 sbi->ll_md_brw_size = data->ocd_brw_size;
387 sbi->ll_md_brw_size = PAGE_CACHE_SIZE;
389 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK) {
390 LCONSOLE_INFO("Layout lock feature supported.\n");
391 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
394 if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
395 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
396 LCONSOLE_INFO("%s: disabling xattr cache due to "
397 "unknown maximum xattr size.\n", dt);
399 sbi->ll_flags |= LL_SBI_XATTR_CACHE;
400 sbi->ll_xattr_cache_enabled = 1;
404 obd = class_name2obd(dt);
406 CERROR("DT %s: not setup or attached\n", dt);
407 GOTO(out_md_fid, err = -ENODEV);
410 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
411 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
412 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
413 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
414 OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
415 OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
416 OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
417 OBD_CONNECT_MAXBYTES |
418 OBD_CONNECT_EINPROGRESS |
419 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
420 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS;
422 if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
423 data->ocd_connect_flags |= OBD_CONNECT_SOM;
425 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
426 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
427 * disabled by default, because it can still be enabled on the
428 * fly via /proc. As a consequence, we still need to come to an
429 * agreement on the supported algorithms at connect time */
430 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
432 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
433 data->ocd_cksum_types = OBD_CKSUM_ADLER;
435 data->ocd_cksum_types = cksum_types_supported_client();
438 #ifdef HAVE_LRU_RESIZE_SUPPORT
439 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
441 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
442 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
444 CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
445 "ocd_grant: %d\n", data->ocd_connect_flags,
446 data->ocd_version, data->ocd_grant);
448 obd->obd_upcall.onu_owner = &sbi->ll_lco;
449 obd->obd_upcall.onu_upcall = cl_ocd_update;
451 data->ocd_brw_size = DT_MAX_BRW_SIZE;
453 err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
456 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
457 "recovery, of which this client is not a "
458 "part. Please wait for recovery to "
459 "complete, abort, or time out.\n", dt);
462 CERROR("%s: Cannot connect to %s: rc = %d\n",
463 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
467 sbi->ll_dt_exp->exp_connect_data = *data;
469 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
470 LUSTRE_SEQ_METADATA);
472 CERROR("%s: Can't init data layer FID infrastructure, "
473 "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
477 mutex_lock(&sbi->ll_lco.lco_lock);
478 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
479 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
480 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
481 mutex_unlock(&sbi->ll_lco.lco_lock);
483 fid_zero(&sbi->ll_root_fid);
484 err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
486 CERROR("cannot mds_connect: rc = %d\n", err);
487 GOTO(out_lock_cn_cb, err);
489 if (!fid_is_sane(&sbi->ll_root_fid)) {
490 CERROR("%s: Invalid root fid "DFID" during mount\n",
491 sbi->ll_md_exp->exp_obd->obd_name,
492 PFID(&sbi->ll_root_fid));
493 GOTO(out_lock_cn_cb, err = -EINVAL);
495 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
497 sb->s_op = &lustre_super_operations;
498 #if THREAD_SIZE >= 8192 /*b=17630*/
499 sb->s_export_op = &lustre_export_operations;
503 * XXX: move this to after cbd setup? */
504 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
505 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
506 valid |= OBD_MD_FLRMTPERM;
507 else if (sbi->ll_flags & LL_SBI_ACL)
508 valid |= OBD_MD_FLACL;
510 OBD_ALLOC_PTR(op_data);
512 GOTO(out_lock_cn_cb, err = -ENOMEM);
514 op_data->op_fid1 = sbi->ll_root_fid;
515 op_data->op_mode = 0;
516 op_data->op_capa1 = oc;
517 op_data->op_valid = valid;
519 err = md_getattr(sbi->ll_md_exp, op_data, &request);
522 OBD_FREE_PTR(op_data);
524 CERROR("%s: md_getattr failed for root: rc = %d\n",
525 sbi->ll_md_exp->exp_obd->obd_name, err);
526 GOTO(out_lock_cn_cb, err);
529 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
530 sbi->ll_md_exp, &lmd);
532 CERROR("failed to understand root inode md: rc = %d\n", err);
533 ptlrpc_req_finished(request);
534 GOTO(out_lock_cn_cb, err);
537 LASSERT(fid_is_sane(&sbi->ll_root_fid));
538 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
539 sbi->ll_flags & LL_SBI_32BIT_API),
541 md_free_lustre_md(sbi->ll_md_exp, &lmd);
542 ptlrpc_req_finished(request);
544 if (root == NULL || IS_ERR(root)) {
546 obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
547 #ifdef CONFIG_FS_POSIX_ACL
549 posix_acl_release(lmd.posix_acl);
550 lmd.posix_acl = NULL;
553 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
555 CERROR("lustre_lite: bad iget4 for root\n");
559 err = ll_close_thread_start(&sbi->ll_lcq);
561 CERROR("cannot start close thread: rc %d\n", err);
565 #ifdef CONFIG_FS_POSIX_ACL
566 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
567 rct_init(&sbi->ll_rct);
568 et_init(&sbi->ll_et);
572 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
573 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
574 KEY_CHECKSUM, sizeof(checksum), &checksum,
578 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
579 KEY_CACHE_SET, sizeof(sbi->ll_cache),
580 &sbi->ll_cache, NULL);
582 sb->s_root = d_make_root(root);
583 if (sb->s_root == NULL) {
584 CERROR("%s: can't make root dentry\n",
585 ll_get_fsname(sb, NULL, 0));
586 GOTO(out_root, err = -ENOMEM);
588 #ifdef HAVE_DCACHE_LOCK
589 sb->s_root->d_op = &ll_d_ops;
592 sbi->ll_sdev_orig = sb->s_dev;
594 /* We set sb->s_dev equal on all lustre clients in order to support
595 * NFS export clustering. NFSD requires that the FSID be the same
597 /* s_dev is also used in lt_compare() to compare two fs, but that is
598 * only a node-local comparison. */
599 uuid = obd_get_uuid(sbi->ll_md_exp);
601 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
602 get_uuid2fsid(uuid->uuid, strlen(uuid->uuid), &sbi->ll_fsid);
615 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
617 obd_disconnect(sbi->ll_dt_exp);
618 sbi->ll_dt_exp = NULL;
619 /* Make sure all OScs are gone, since cl_cache is accessing sbi. */
620 obd_zombie_barrier();
622 obd_fid_fini(sbi->ll_md_exp->exp_obd);
624 obd_disconnect(sbi->ll_md_exp);
625 sbi->ll_md_exp = NULL;
631 lprocfs_unregister_mountpoint(sbi);
635 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
639 *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
641 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
642 KEY_MAX_EASIZE, &size, lmmsize, NULL);
644 CERROR("Get max mdsize error rc %d \n", rc);
649 void ll_dump_inode(struct inode *inode)
651 struct ll_d_hlist_node *tmp;
652 int dentry_count = 0;
654 LASSERT(inode != NULL);
656 ll_d_hlist_for_each(tmp, &inode->i_dentry)
659 CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
660 inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
661 inode->i_mode, atomic_read(&inode->i_count), dentry_count);
664 void lustre_dump_dentry(struct dentry *dentry, int recur)
666 struct list_head *tmp;
669 LASSERT(dentry != NULL);
671 list_for_each(tmp, &dentry->d_subdirs)
674 CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
675 " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
676 dentry->d_name.len, dentry->d_name.name,
677 dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
678 dentry->d_parent, dentry->d_inode, d_refcount(dentry),
679 dentry->d_flags, dentry->d_fsdata, subdirs);
680 if (dentry->d_inode != NULL)
681 ll_dump_inode(dentry->d_inode);
686 list_for_each(tmp, &dentry->d_subdirs) {
687 struct dentry *d = list_entry(tmp, struct dentry, d_u.d_child);
688 lustre_dump_dentry(d, recur - 1);
692 void client_common_put_super(struct super_block *sb)
694 struct ll_sb_info *sbi = ll_s2sbi(sb);
697 #ifdef CONFIG_FS_POSIX_ACL
698 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
699 et_fini(&sbi->ll_et);
700 rct_fini(&sbi->ll_rct);
704 ll_close_thread_shutdown(sbi->ll_lcq);
708 cfs_list_del(&sbi->ll_conn_chain);
710 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
711 obd_disconnect(sbi->ll_dt_exp);
712 sbi->ll_dt_exp = NULL;
713 /* wait till all OSCs are gone, since cl_cache is accessing sbi.
715 obd_zombie_barrier();
717 lprocfs_unregister_mountpoint(sbi);
719 obd_fid_fini(sbi->ll_md_exp->exp_obd);
720 obd_disconnect(sbi->ll_md_exp);
721 sbi->ll_md_exp = NULL;
726 void ll_kill_super(struct super_block *sb)
728 struct ll_sb_info *sbi;
733 if (!(sb->s_flags & MS_ACTIVE))
737 /* we need restore s_dev from changed for clustred NFS before put_super
738 * because new kernels have cached s_dev and change sb->s_dev in
739 * put_super not affected real removing devices */
741 sb->s_dev = sbi->ll_sdev_orig;
742 sbi->ll_umounting = 1;
747 char *ll_read_opt(const char *opt, char *data)
753 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
754 if (strncmp(opt, data, strlen(opt)))
756 if ((value = strchr(data, '=')) == NULL)
760 OBD_ALLOC(retval, strlen(value) + 1);
762 CERROR("out of memory!\n");
766 memcpy(retval, value, strlen(value)+1);
767 CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
771 static inline int ll_set_opt(const char *opt, char *data, int fl)
773 if (strncmp(opt, data, strlen(opt)) != 0)
779 /* non-client-specific mount options are parsed in lmd_parse */
780 static int ll_options(char *options, int *flags)
783 char *s1 = options, *s2;
789 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
792 CDEBUG(D_SUPER, "next opt=%s\n", s1);
793 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
798 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
803 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
808 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
813 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
818 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
823 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 5, 50, 0)
824 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
826 /* Ignore deprecated mount option. The client will
827 * always try to mount with ACL support, whether this
828 * is used depends on whether server supports it. */
829 LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
830 "mount option 'acl'.\n");
833 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
835 LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
836 "mount option 'noacl'.\n");
840 #warning "{no}acl options have been deprecated since 1.8, please remove them"
842 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
847 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
852 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
858 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
863 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
868 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
873 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
878 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
883 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
888 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
893 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
898 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
903 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
908 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
914 s2 = strchr(s1, ',');
922 void ll_lli_init(struct ll_inode_info *lli)
924 lli->lli_inode_magic = LLI_INODE_MAGIC;
926 lli->lli_ioepoch = 0;
927 lli->lli_maxbytes = MAX_LFS_FILESIZE;
928 spin_lock_init(&lli->lli_lock);
929 lli->lli_posix_acl = NULL;
930 lli->lli_remote_perms = NULL;
931 mutex_init(&lli->lli_rmtperm_mutex);
932 /* Do not set lli_fid, it has been initialized already. */
933 fid_zero(&lli->lli_pfid);
934 CFS_INIT_LIST_HEAD(&lli->lli_close_list);
935 CFS_INIT_LIST_HEAD(&lli->lli_oss_capas);
936 cfs_atomic_set(&lli->lli_open_count, 0);
937 lli->lli_mds_capa = NULL;
938 lli->lli_rmtperm_time = 0;
939 lli->lli_pending_och = NULL;
940 lli->lli_mds_read_och = NULL;
941 lli->lli_mds_write_och = NULL;
942 lli->lli_mds_exec_och = NULL;
943 lli->lli_open_fd_read_count = 0;
944 lli->lli_open_fd_write_count = 0;
945 lli->lli_open_fd_exec_count = 0;
946 mutex_init(&lli->lli_och_mutex);
947 spin_lock_init(&lli->lli_agl_lock);
948 lli->lli_has_smd = false;
949 lli->lli_layout_gen = LL_LAYOUT_GEN_NONE;
950 lli->lli_clob = NULL;
952 init_rwsem(&lli->lli_xattrs_list_rwsem);
953 mutex_init(&lli->lli_xattrs_enq_lock);
955 LASSERT(lli->lli_vfs_inode.i_mode != 0);
956 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
957 mutex_init(&lli->lli_readdir_mutex);
958 lli->lli_opendir_key = NULL;
960 lli->lli_def_acl = NULL;
961 spin_lock_init(&lli->lli_sa_lock);
962 lli->lli_opendir_pid = 0;
964 sema_init(&lli->lli_size_sem, 1);
965 lli->lli_size_sem_owner = NULL;
966 lli->lli_symlink_name = NULL;
967 init_rwsem(&lli->lli_trunc_sem);
968 mutex_init(&lli->lli_write_mutex);
969 init_rwsem(&lli->lli_glimpse_sem);
970 lli->lli_glimpse_time = 0;
971 CFS_INIT_LIST_HEAD(&lli->lli_agl_list);
972 lli->lli_agl_index = 0;
973 lli->lli_async_rc = 0;
974 lli->lli_volatile = false;
976 mutex_init(&lli->lli_layout_mutex);
979 static inline int ll_bdi_register(struct backing_dev_info *bdi)
981 static atomic_t ll_bdi_num = ATOMIC_INIT(0);
983 bdi->name = "lustre";
984 return bdi_register(bdi, NULL, "lustre-%d",
985 atomic_inc_return(&ll_bdi_num));
988 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
990 struct lustre_profile *lprof = NULL;
991 struct lustre_sb_info *lsi = s2lsi(sb);
992 struct ll_sb_info *sbi;
993 char *dt = NULL, *md = NULL;
994 char *profilenm = get_profile_name(sb);
995 struct config_llog_instance *cfg;
996 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
997 const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
1001 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1009 /* client additional sb info */
1010 lsi->lsi_llsbi = sbi = ll_init_sbi();
1012 cfs_module_put(THIS_MODULE);
1017 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1019 GOTO(out_free, err);
1021 err = bdi_init(&lsi->lsi_bdi);
1023 GOTO(out_free, err);
1024 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1025 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1026 err = ll_bdi_register(&lsi->lsi_bdi);
1028 GOTO(out_free, err);
1030 sb->s_bdi = &lsi->lsi_bdi;
1031 #ifndef HAVE_DCACHE_LOCK
1032 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1033 sb->s_d_op = &ll_d_ops;
1036 /* Generate a string unique to this super, in case some joker tries
1037 to mount the same fs at two mount points.
1038 Use the address of the super itself.*/
1039 cfg->cfg_instance = sb;
1040 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1041 cfg->cfg_callback = class_config_llog_handler;
1042 /* set up client obds */
1043 err = lustre_process_log(sb, profilenm, cfg);
1045 CERROR("Unable to process log: %d\n", err);
1046 GOTO(out_free, err);
1049 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1050 lprof = class_get_profile(profilenm);
1051 if (lprof == NULL) {
1052 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1053 " read from the MGS. Does that filesystem "
1054 "exist?\n", profilenm);
1055 GOTO(out_free, err = -EINVAL);
1057 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1058 lprof->lp_md, lprof->lp_dt);
1060 OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1062 GOTO(out_free, err = -ENOMEM);
1063 sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1065 OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1067 GOTO(out_free, err = -ENOMEM);
1068 sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1070 /* connections, registrations, sb setup */
1071 err = client_common_fill_super(sb, md, dt, mnt);
1075 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1077 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1080 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1081 LCONSOLE_WARN("Mounted %s\n", profilenm);
1085 } /* ll_fill_super */
1088 void lu_context_keys_dump(void);
1090 void ll_put_super(struct super_block *sb)
1092 struct config_llog_instance cfg, params_cfg;
1093 struct obd_device *obd;
1094 struct lustre_sb_info *lsi = s2lsi(sb);
1095 struct ll_sb_info *sbi = ll_s2sbi(sb);
1096 char *profilenm = get_profile_name(sb);
1097 int force = 1, next;
1100 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1102 ll_print_capa_stat(sbi);
1104 cfg.cfg_instance = sb;
1105 lustre_end_log(sb, profilenm, &cfg);
1107 params_cfg.cfg_instance = sb;
1108 lustre_end_log(sb, PARAMS_FILENAME, ¶ms_cfg);
1110 if (sbi->ll_md_exp) {
1111 obd = class_exp2obd(sbi->ll_md_exp);
1113 force = obd->obd_force;
1116 /* We need to set force before the lov_disconnect in
1117 lustre_common_put_super, since l_d cleans up osc's as well. */
1120 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1122 obd->obd_force = force;
1127 /* Only if client_common_fill_super succeeded */
1128 client_common_put_super(sb);
1132 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1133 class_manual_cleanup(obd);
1136 if (sbi->ll_flags & LL_SBI_VERBOSE)
1137 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1140 class_del_profile(profilenm);
1142 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1143 bdi_destroy(&lsi->lsi_bdi);
1144 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1148 lsi->lsi_llsbi = NULL;
1150 lustre_common_put_super(sb);
1152 cfs_module_put(THIS_MODULE);
1155 } /* client_put_super */
1157 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1159 struct inode *inode = NULL;
1161 /* NOTE: we depend on atomic igrab() -bzzz */
1162 lock_res_and_lock(lock);
1163 if (lock->l_resource->lr_lvb_inode) {
1164 struct ll_inode_info * lli;
1165 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1166 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1167 inode = igrab(lock->l_resource->lr_lvb_inode);
1169 inode = lock->l_resource->lr_lvb_inode;
1170 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1171 D_WARNING, lock, "lr_lvb_inode %p is "
1172 "bogus: magic %08x",
1173 lock->l_resource->lr_lvb_inode,
1174 lli->lli_inode_magic);
1178 unlock_res_and_lock(lock);
1182 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1184 struct inode *inode = NULL;
1185 /* NOTE: we depend on atomic igrab() -bzzz */
1186 lock_res_and_lock(lock);
1187 if (lock->l_ast_data) {
1188 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1189 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1190 inode = igrab(lock->l_ast_data);
1192 inode = lock->l_ast_data;
1193 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1194 D_WARNING, lock, "l_ast_data %p is "
1195 "bogus: magic %08x", lock->l_ast_data,
1196 lli->lli_inode_magic);
1200 unlock_res_and_lock(lock);
1204 void ll_clear_inode(struct inode *inode)
1206 struct ll_inode_info *lli = ll_i2info(inode);
1207 struct ll_sb_info *sbi = ll_i2sbi(inode);
1210 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1211 inode->i_generation, inode);
1213 if (S_ISDIR(inode->i_mode)) {
1214 /* these should have been cleared in ll_file_release */
1215 LASSERT(lli->lli_opendir_key == NULL);
1216 LASSERT(lli->lli_sai == NULL);
1217 LASSERT(lli->lli_opendir_pid == 0);
1220 spin_lock(&lli->lli_lock);
1221 ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1222 spin_unlock(&lli->lli_lock);
1223 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1225 LASSERT(!lli->lli_open_fd_write_count);
1226 LASSERT(!lli->lli_open_fd_read_count);
1227 LASSERT(!lli->lli_open_fd_exec_count);
1229 if (lli->lli_mds_write_och)
1230 ll_md_real_close(inode, FMODE_WRITE);
1231 if (lli->lli_mds_exec_och)
1232 ll_md_real_close(inode, FMODE_EXEC);
1233 if (lli->lli_mds_read_och)
1234 ll_md_real_close(inode, FMODE_READ);
1236 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1237 OBD_FREE(lli->lli_symlink_name,
1238 strlen(lli->lli_symlink_name) + 1);
1239 lli->lli_symlink_name = NULL;
1242 ll_xattr_cache_destroy(inode);
1244 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1245 LASSERT(lli->lli_posix_acl == NULL);
1246 if (lli->lli_remote_perms) {
1247 free_rmtperm_hash(lli->lli_remote_perms);
1248 lli->lli_remote_perms = NULL;
1251 #ifdef CONFIG_FS_POSIX_ACL
1252 else if (lli->lli_posix_acl) {
1253 LASSERT(cfs_atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1254 LASSERT(lli->lli_remote_perms == NULL);
1255 posix_acl_release(lli->lli_posix_acl);
1256 lli->lli_posix_acl = NULL;
1259 lli->lli_inode_magic = LLI_INODE_DEAD;
1261 ll_clear_inode_capas(inode);
1262 if (!S_ISDIR(inode->i_mode))
1263 LASSERT(cfs_list_empty(&lli->lli_agl_list));
1266 * XXX This has to be done before lsm is freed below, because
1267 * cl_object still uses inode lsm.
1269 cl_inode_fini(inode);
1270 lli->lli_has_smd = false;
1275 int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data,
1276 struct md_open_data **mod)
1278 struct lustre_md md;
1279 struct inode *inode = dentry->d_inode;
1280 struct ll_sb_info *sbi = ll_i2sbi(inode);
1281 struct ptlrpc_request *request = NULL;
1285 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1286 LUSTRE_OPC_ANY, NULL);
1287 if (IS_ERR(op_data))
1288 RETURN(PTR_ERR(op_data));
1290 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1293 ptlrpc_req_finished(request);
1294 if (rc == -ENOENT) {
1296 /* Unlinked special device node? Or just a race?
1297 * Pretend we done everything. */
1298 if (!S_ISREG(inode->i_mode) &&
1299 !S_ISDIR(inode->i_mode)) {
1300 ia_valid = op_data->op_attr.ia_valid;
1301 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1302 rc = simple_setattr(dentry, &op_data->op_attr);
1303 op_data->op_attr.ia_valid = ia_valid;
1305 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1306 CERROR("md_setattr fails: rc = %d\n", rc);
1311 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1312 sbi->ll_md_exp, &md);
1314 ptlrpc_req_finished(request);
1318 ia_valid = op_data->op_attr.ia_valid;
1319 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1320 * cache is not cleared yet. */
1321 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1322 rc = simple_setattr(dentry, &op_data->op_attr);
1323 op_data->op_attr.ia_valid = ia_valid;
1325 /* Extract epoch data if obtained. */
1326 op_data->op_handle = md.body->handle;
1327 op_data->op_ioepoch = md.body->ioepoch;
1329 ll_update_inode(inode, &md);
1330 ptlrpc_req_finished(request);
1335 /* Close IO epoch and send Size-on-MDS attribute update. */
1336 static int ll_setattr_done_writing(struct inode *inode,
1337 struct md_op_data *op_data,
1338 struct md_open_data *mod)
1340 struct ll_inode_info *lli = ll_i2info(inode);
1344 LASSERT(op_data != NULL);
1345 if (!S_ISREG(inode->i_mode))
1348 CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1349 op_data->op_ioepoch, PFID(&lli->lli_fid));
1351 op_data->op_flags = MF_EPOCH_CLOSE;
1352 ll_done_writing_attr(inode, op_data);
1353 ll_pack_inode2opdata(inode, op_data, NULL);
1355 rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1356 if (rc == -EAGAIN) {
1357 /* MDS has instructed us to obtain Size-on-MDS attribute
1358 * from OSTs and send setattr to back to MDS. */
1359 rc = ll_som_update(inode, op_data);
1361 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1367 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1369 struct obd_capa *capa;
1372 if (attr->ia_valid & ATTR_SIZE)
1373 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1375 capa = ll_mdscapa_get(inode);
1377 rc = cl_setattr_ost(inode, attr, capa);
1379 if (attr->ia_valid & ATTR_SIZE)
1380 ll_truncate_free_capa(capa);
1387 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1388 * object(s) determine the file size and mtime. Otherwise, the MDS will
1389 * keep these values until such a time that objects are allocated for it.
1390 * We do the MDS operations first, as it is checking permissions for us.
1391 * We don't to the MDS RPC if there is nothing that we want to store there,
1392 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1393 * going to do an RPC anyways.
1395 * If we are doing a truncate, we will send the mtime and ctime updates
1396 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1397 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1400 * In case of HSMimport, we only set attr on MDS.
1402 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1404 struct inode *inode = dentry->d_inode;
1405 struct ll_inode_info *lli = ll_i2info(inode);
1406 struct md_op_data *op_data = NULL;
1407 struct md_open_data *mod = NULL;
1408 bool file_is_released = false;
1409 int rc = 0, rc1 = 0;
1412 CDEBUG(D_VFSTRACE, "%s: setattr inode %p/fid:"DFID" from %llu to %llu, "
1413 "valid %x, hsm_import %d\n",
1414 ll_get_fsname(inode->i_sb, NULL, 0), inode,
1415 PFID(&lli->lli_fid), i_size_read(inode), attr->ia_size,
1416 attr->ia_valid, hsm_import);
1418 if (attr->ia_valid & ATTR_SIZE) {
1419 /* Check new size against VFS/VM file size limit and rlimit */
1420 rc = inode_newsize_ok(inode, attr->ia_size);
1424 /* The maximum Lustre file size is variable, based on the
1425 * OST maximum object size and number of stripes. This
1426 * needs another check in addition to the VFS check above. */
1427 if (attr->ia_size > ll_file_maxbytes(inode)) {
1428 CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1429 PFID(&lli->lli_fid), attr->ia_size,
1430 ll_file_maxbytes(inode));
1434 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1437 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1438 if (attr->ia_valid & TIMES_SET_FLAGS) {
1439 if (cfs_curproc_fsuid() != inode->i_uid &&
1440 !cfs_capable(CFS_CAP_FOWNER))
1444 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1445 if (attr->ia_valid & ATTR_CTIME) {
1446 attr->ia_ctime = CFS_CURRENT_TIME;
1447 attr->ia_valid |= ATTR_CTIME_SET;
1449 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1450 (attr->ia_valid & ATTR_ATIME)) {
1451 attr->ia_atime = CFS_CURRENT_TIME;
1452 attr->ia_valid |= ATTR_ATIME_SET;
1454 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1455 (attr->ia_valid & ATTR_MTIME)) {
1456 attr->ia_mtime = CFS_CURRENT_TIME;
1457 attr->ia_valid |= ATTR_MTIME_SET;
1460 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1461 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1462 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1463 cfs_time_current_sec());
1465 /* If we are changing file size, file content is modified, flag it. */
1466 if (attr->ia_valid & ATTR_SIZE) {
1467 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1468 spin_lock(&lli->lli_lock);
1469 lli->lli_flags |= LLIF_DATA_MODIFIED;
1470 spin_unlock(&lli->lli_lock);
1473 /* We always do an MDS RPC, even if we're only changing the size;
1474 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1476 OBD_ALLOC_PTR(op_data);
1477 if (op_data == NULL)
1480 if (!S_ISDIR(inode->i_mode)) {
1481 if (attr->ia_valid & ATTR_SIZE)
1482 inode_dio_write_done(inode);
1483 mutex_unlock(&inode->i_mutex);
1484 down_write(&lli->lli_trunc_sem);
1487 memcpy(&op_data->op_attr, attr, sizeof(*attr));
1489 /* Open epoch for truncate. */
1490 if (exp_connect_som(ll_i2mdexp(inode)) &&
1491 (attr->ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1492 op_data->op_flags = MF_EPOCH_OPEN;
1494 /* truncate on a released file must failed with -ENODATA,
1495 * so size must not be set on MDS for released file
1496 * but other attributes must be set
1498 if (S_ISREG(inode->i_mode)) {
1499 struct lov_stripe_md *lsm;
1502 ll_layout_refresh(inode, &gen);
1503 lsm = ccc_inode_lsm_get(inode);
1504 if (lsm && lsm->lsm_pattern & LOV_PATTERN_F_RELEASED)
1505 file_is_released = true;
1506 ccc_inode_lsm_put(inode, lsm);
1509 /* if not in HSM import mode, clear size attr for released file
1510 * we clear the attribute send to MDT in op_data, not the original
1511 * received from caller in attr which is used later to
1512 * decide return code */
1513 if (file_is_released && (attr->ia_valid & ATTR_SIZE) && !hsm_import)
1514 op_data->op_attr.ia_valid &= ~ATTR_SIZE;
1516 rc = ll_md_setattr(dentry, op_data, &mod);
1520 /* truncate failed (only when non HSM import), others succeed */
1521 if (file_is_released) {
1522 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1523 GOTO(out, rc = -ENODATA);
1528 /* RPC to MDT is sent, cancel data modification flag */
1529 if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) {
1530 spin_lock(&lli->lli_lock);
1531 lli->lli_flags &= ~LLIF_DATA_MODIFIED;
1532 spin_unlock(&lli->lli_lock);
1535 ll_ioepoch_open(lli, op_data->op_ioepoch);
1536 if (!S_ISREG(inode->i_mode))
1539 if (attr->ia_valid & (ATTR_SIZE |
1540 ATTR_ATIME | ATTR_ATIME_SET |
1541 ATTR_MTIME | ATTR_MTIME_SET)) {
1542 /* For truncate and utimes sending attributes to OSTs, setting
1543 * mtime/atime to the past will be performed under PW [0:EOF]
1544 * extent lock (new_size:EOF for truncate). It may seem
1545 * excessive to send mtime/atime updates to OSTs when not
1546 * setting times to past, but it is necessary due to possible
1547 * time de-synchronization between MDT inode and OST objects */
1548 rc = ll_setattr_ost(inode, attr);
1553 if (op_data->op_ioepoch) {
1554 rc1 = ll_setattr_done_writing(inode, op_data, mod);
1558 ll_finish_md_op_data(op_data);
1560 if (!S_ISDIR(inode->i_mode)) {
1561 up_write(&lli->lli_trunc_sem);
1562 mutex_lock(&inode->i_mutex);
1563 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1564 inode_dio_wait(inode);
1567 ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1568 LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1573 int ll_setattr(struct dentry *de, struct iattr *attr)
1575 int mode = de->d_inode->i_mode;
1577 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1578 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1579 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1581 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1582 (ATTR_SIZE|ATTR_MODE)) &&
1583 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1584 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1585 !(attr->ia_mode & S_ISGID))))
1586 attr->ia_valid |= ATTR_FORCE;
1588 if ((mode & S_ISUID) &&
1589 !(attr->ia_mode & S_ISUID) &&
1590 !(attr->ia_valid & ATTR_KILL_SUID))
1591 attr->ia_valid |= ATTR_KILL_SUID;
1593 if (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1594 !(attr->ia_mode & S_ISGID) &&
1595 !(attr->ia_valid & ATTR_KILL_SGID))
1596 attr->ia_valid |= ATTR_KILL_SGID;
1598 return ll_setattr_raw(de, attr, false);
1601 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1602 __u64 max_age, __u32 flags)
1604 struct ll_sb_info *sbi = ll_s2sbi(sb);
1605 struct obd_statfs obd_osfs;
1609 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1611 CERROR("md_statfs fails: rc = %d\n", rc);
1615 osfs->os_type = sb->s_magic;
1617 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1618 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1620 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1621 flags |= OBD_STATFS_NODELAY;
1623 rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1625 CERROR("obd_statfs fails: rc = %d\n", rc);
1629 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1630 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1633 osfs->os_bsize = obd_osfs.os_bsize;
1634 osfs->os_blocks = obd_osfs.os_blocks;
1635 osfs->os_bfree = obd_osfs.os_bfree;
1636 osfs->os_bavail = obd_osfs.os_bavail;
1638 /* If we don't have as many objects free on the OST as inodes
1639 * on the MDS, we reduce the total number of inodes to
1640 * compensate, so that the "inodes in use" number is correct.
1642 if (obd_osfs.os_ffree < osfs->os_ffree) {
1643 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1645 osfs->os_ffree = obd_osfs.os_ffree;
1650 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1652 struct super_block *sb = de->d_sb;
1653 struct obd_statfs osfs;
1656 CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1657 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1659 /* Some amount of caching on the client is allowed */
1660 rc = ll_statfs_internal(sb, &osfs,
1661 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1666 statfs_unpack(sfs, &osfs);
1668 /* We need to downshift for all 32-bit kernels, because we can't
1669 * tell if the kernel is being called via sys_statfs64() or not.
1670 * Stop before overflowing f_bsize - in which case it is better
1671 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1672 if (sizeof(long) < 8) {
1673 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1676 osfs.os_blocks >>= 1;
1677 osfs.os_bfree >>= 1;
1678 osfs.os_bavail >>= 1;
1682 sfs->f_blocks = osfs.os_blocks;
1683 sfs->f_bfree = osfs.os_bfree;
1684 sfs->f_bavail = osfs.os_bavail;
1685 sfs->f_fsid = ll_s2sbi(sb)->ll_fsid;
1689 void ll_inode_size_lock(struct inode *inode)
1691 struct ll_inode_info *lli;
1693 LASSERT(!S_ISDIR(inode->i_mode));
1695 lli = ll_i2info(inode);
1696 LASSERT(lli->lli_size_sem_owner != current);
1697 down(&lli->lli_size_sem);
1698 LASSERT(lli->lli_size_sem_owner == NULL);
1699 lli->lli_size_sem_owner = current;
1702 void ll_inode_size_unlock(struct inode *inode)
1704 struct ll_inode_info *lli;
1706 lli = ll_i2info(inode);
1707 LASSERT(lli->lli_size_sem_owner == current);
1708 lli->lli_size_sem_owner = NULL;
1709 up(&lli->lli_size_sem);
1712 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1714 struct ll_inode_info *lli = ll_i2info(inode);
1715 struct mdt_body *body = md->body;
1716 struct lov_stripe_md *lsm = md->lsm;
1717 struct ll_sb_info *sbi = ll_i2sbi(inode);
1719 LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1721 if (!lli->lli_has_smd &&
1722 !(sbi->ll_flags & LL_SBI_LAYOUT_LOCK))
1723 cl_file_inode_init(inode, md);
1725 lli->lli_maxbytes = lsm->lsm_maxbytes;
1726 if (lli->lli_maxbytes > MAX_LFS_FILESIZE)
1727 lli->lli_maxbytes = MAX_LFS_FILESIZE;
1730 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1731 if (body->valid & OBD_MD_FLRMTPERM)
1732 ll_update_remote_perm(inode, md->remote_perm);
1734 #ifdef CONFIG_FS_POSIX_ACL
1735 else if (body->valid & OBD_MD_FLACL) {
1736 spin_lock(&lli->lli_lock);
1737 if (lli->lli_posix_acl)
1738 posix_acl_release(lli->lli_posix_acl);
1739 lli->lli_posix_acl = md->posix_acl;
1740 spin_unlock(&lli->lli_lock);
1743 inode->i_ino = cl_fid_build_ino(&body->fid1,
1744 sbi->ll_flags & LL_SBI_32BIT_API);
1745 inode->i_generation = cl_fid_build_gen(&body->fid1);
1747 if (body->valid & OBD_MD_FLATIME) {
1748 if (body->atime > LTIME_S(inode->i_atime))
1749 LTIME_S(inode->i_atime) = body->atime;
1750 lli->lli_lvb.lvb_atime = body->atime;
1752 if (body->valid & OBD_MD_FLMTIME) {
1753 if (body->mtime > LTIME_S(inode->i_mtime)) {
1754 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1755 "to "LPU64"\n", inode->i_ino,
1756 LTIME_S(inode->i_mtime), body->mtime);
1757 LTIME_S(inode->i_mtime) = body->mtime;
1759 lli->lli_lvb.lvb_mtime = body->mtime;
1761 if (body->valid & OBD_MD_FLCTIME) {
1762 if (body->ctime > LTIME_S(inode->i_ctime))
1763 LTIME_S(inode->i_ctime) = body->ctime;
1764 lli->lli_lvb.lvb_ctime = body->ctime;
1766 if (body->valid & OBD_MD_FLMODE)
1767 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1768 if (body->valid & OBD_MD_FLTYPE)
1769 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1770 LASSERT(inode->i_mode != 0);
1771 if (S_ISREG(inode->i_mode)) {
1772 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1774 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1776 if (body->valid & OBD_MD_FLUID)
1777 inode->i_uid = body->uid;
1778 if (body->valid & OBD_MD_FLGID)
1779 inode->i_gid = body->gid;
1780 if (body->valid & OBD_MD_FLFLAGS)
1781 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1782 if (body->valid & OBD_MD_FLNLINK)
1783 set_nlink(inode, body->nlink);
1784 if (body->valid & OBD_MD_FLRDEV)
1785 inode->i_rdev = old_decode_dev(body->rdev);
1787 if (body->valid & OBD_MD_FLID) {
1788 /* FID shouldn't be changed! */
1789 if (fid_is_sane(&lli->lli_fid)) {
1790 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1791 "Trying to change FID "DFID
1792 " to the "DFID", inode %lu/%u(%p)\n",
1793 PFID(&lli->lli_fid), PFID(&body->fid1),
1794 inode->i_ino, inode->i_generation, inode);
1796 lli->lli_fid = body->fid1;
1799 LASSERT(fid_seq(&lli->lli_fid) != 0);
1801 if (body->valid & OBD_MD_FLSIZE) {
1802 if (exp_connect_som(ll_i2mdexp(inode)) &&
1803 S_ISREG(inode->i_mode)) {
1804 struct lustre_handle lockh;
1807 /* As it is possible a blocking ast has been processed
1808 * by this time, we need to check there is an UPDATE
1809 * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1811 mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1812 &lockh, LDLM_FL_CBPENDING,
1816 if (lli->lli_flags & (LLIF_DONE_WRITING |
1817 LLIF_EPOCH_PENDING |
1819 CERROR("ino %lu flags %u still has "
1820 "size authority! do not trust "
1821 "the size got from MDS\n",
1822 inode->i_ino, lli->lli_flags);
1824 /* Use old size assignment to avoid
1825 * deadlock bz14138 & bz14326 */
1826 i_size_write(inode, body->size);
1827 spin_lock(&lli->lli_lock);
1828 lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1829 spin_unlock(&lli->lli_lock);
1831 ldlm_lock_decref(&lockh, mode);
1834 /* Use old size assignment to avoid
1835 * deadlock bz14138 & bz14326 */
1836 i_size_write(inode, body->size);
1838 CDEBUG(D_VFSTRACE, "inode=%lu, updating i_size %llu\n",
1839 inode->i_ino, (unsigned long long)body->size);
1842 if (body->valid & OBD_MD_FLBLOCKS)
1843 inode->i_blocks = body->blocks;
1846 if (body->valid & OBD_MD_FLMDSCAPA) {
1847 LASSERT(md->mds_capa);
1848 ll_add_capa(inode, md->mds_capa);
1850 if (body->valid & OBD_MD_FLOSSCAPA) {
1851 LASSERT(md->oss_capa);
1852 ll_add_capa(inode, md->oss_capa);
1855 if (body->valid & OBD_MD_TSTATE) {
1856 if (body->t_state & MS_RESTORE)
1857 lli->lli_flags |= LLIF_FILE_RESTORING;
1861 void ll_read_inode2(struct inode *inode, void *opaque)
1863 struct lustre_md *md = opaque;
1864 struct ll_inode_info *lli = ll_i2info(inode);
1867 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1868 PFID(&lli->lli_fid), inode);
1870 LASSERT(!lli->lli_has_smd);
1872 /* Core attributes from the MDS first. This is a new inode, and
1873 * the VFS doesn't zero times in the core inode so we have to do
1874 * it ourselves. They will be overwritten by either MDS or OST
1875 * attributes - we just need to make sure they aren't newer. */
1876 LTIME_S(inode->i_mtime) = 0;
1877 LTIME_S(inode->i_atime) = 0;
1878 LTIME_S(inode->i_ctime) = 0;
1880 ll_update_inode(inode, md);
1882 /* OIDEBUG(inode); */
1884 /* initializing backing dev info. */
1885 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1888 if (S_ISREG(inode->i_mode)) {
1889 struct ll_sb_info *sbi = ll_i2sbi(inode);
1890 inode->i_op = &ll_file_inode_operations;
1891 inode->i_fop = sbi->ll_fop;
1892 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1894 } else if (S_ISDIR(inode->i_mode)) {
1895 inode->i_op = &ll_dir_inode_operations;
1896 inode->i_fop = &ll_dir_operations;
1898 } else if (S_ISLNK(inode->i_mode)) {
1899 inode->i_op = &ll_fast_symlink_inode_operations;
1902 inode->i_op = &ll_special_inode_operations;
1904 init_special_inode(inode, inode->i_mode,
1911 void ll_delete_inode(struct inode *inode)
1913 struct cl_inode_info *lli = cl_i2info(inode);
1916 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
1917 /* discard all dirty pages before truncating them, required by
1918 * osc_extent implementation at LU-1030. */
1919 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF,
1920 CL_FSYNC_DISCARD, 1);
1922 truncate_inode_pages(&inode->i_data, 0);
1924 /* Workaround for LU-118 */
1925 if (inode->i_data.nrpages) {
1926 TREE_READ_LOCK_IRQ(&inode->i_data);
1927 TREE_READ_UNLOCK_IRQ(&inode->i_data);
1928 LASSERTF(inode->i_data.nrpages == 0,
1929 "inode=%lu/%u(%p) nrpages=%lu, see "
1930 "http://jira.whamcloud.com/browse/LU-118\n",
1931 inode->i_ino, inode->i_generation, inode,
1932 inode->i_data.nrpages);
1934 /* Workaround end */
1936 #ifdef HAVE_SBOPS_EVICT_INODE
1937 ll_clear_inode(inode);
1944 int ll_iocontrol(struct inode *inode, struct file *file,
1945 unsigned int cmd, unsigned long arg)
1947 struct ll_sb_info *sbi = ll_i2sbi(inode);
1948 struct ptlrpc_request *req = NULL;
1953 case FSFILT_IOC_GETFLAGS: {
1954 struct mdt_body *body;
1955 struct md_op_data *op_data;
1957 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
1958 0, 0, LUSTRE_OPC_ANY,
1960 if (IS_ERR(op_data))
1961 RETURN(PTR_ERR(op_data));
1963 op_data->op_valid = OBD_MD_FLFLAGS;
1964 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
1965 ll_finish_md_op_data(op_data);
1967 CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1971 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1973 flags = body->flags;
1975 ptlrpc_req_finished(req);
1977 RETURN(put_user(flags, (int *)arg));
1979 case FSFILT_IOC_SETFLAGS: {
1980 struct lov_stripe_md *lsm;
1981 struct obd_info oinfo = { { { 0 } } };
1982 struct md_op_data *op_data;
1984 if (get_user(flags, (int *)arg))
1987 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1988 LUSTRE_OPC_ANY, NULL);
1989 if (IS_ERR(op_data))
1990 RETURN(PTR_ERR(op_data));
1992 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1993 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1994 rc = md_setattr(sbi->ll_md_exp, op_data,
1995 NULL, 0, NULL, 0, &req, NULL);
1996 ll_finish_md_op_data(op_data);
1997 ptlrpc_req_finished(req);
2001 inode->i_flags = ll_ext_to_inode_flags(flags);
2003 lsm = ccc_inode_lsm_get(inode);
2004 if (!lsm_has_objects(lsm)) {
2005 ccc_inode_lsm_put(inode, lsm);
2009 OBDO_ALLOC(oinfo.oi_oa);
2011 ccc_inode_lsm_put(inode, lsm);
2015 oinfo.oi_oa->o_oi = lsm->lsm_oi;
2016 oinfo.oi_oa->o_flags = flags;
2017 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
2019 oinfo.oi_capa = ll_mdscapa_get(inode);
2020 obdo_set_parent_fid(oinfo.oi_oa, &ll_i2info(inode)->lli_fid);
2021 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
2022 capa_put(oinfo.oi_capa);
2023 OBDO_FREE(oinfo.oi_oa);
2024 ccc_inode_lsm_put(inode, lsm);
2026 if (rc && rc != -EPERM && rc != -EACCES)
2027 CERROR("osc_setattr_async fails: rc = %d\n", rc);
2038 int ll_flush_ctx(struct inode *inode)
2040 struct ll_sb_info *sbi = ll_i2sbi(inode);
2042 CDEBUG(D_SEC, "flush context for user %d\n", cfs_curproc_uid());
2044 obd_set_info_async(NULL, sbi->ll_md_exp,
2045 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2047 obd_set_info_async(NULL, sbi->ll_dt_exp,
2048 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2053 /* umount -f client means force down, don't save state */
2054 void ll_umount_begin(struct super_block *sb)
2056 struct ll_sb_info *sbi = ll_s2sbi(sb);
2057 struct obd_device *obd;
2058 struct obd_ioctl_data *ioc_data;
2061 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2062 sb->s_count, atomic_read(&sb->s_active));
2064 obd = class_exp2obd(sbi->ll_md_exp);
2066 CERROR("Invalid MDC connection handle "LPX64"\n",
2067 sbi->ll_md_exp->exp_handle.h_cookie);
2073 obd = class_exp2obd(sbi->ll_dt_exp);
2075 CERROR("Invalid LOV connection handle "LPX64"\n",
2076 sbi->ll_dt_exp->exp_handle.h_cookie);
2082 OBD_ALLOC_PTR(ioc_data);
2084 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2085 sizeof *ioc_data, ioc_data, NULL);
2087 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2088 sizeof *ioc_data, ioc_data, NULL);
2090 OBD_FREE_PTR(ioc_data);
2093 /* Really, we'd like to wait until there are no requests outstanding,
2094 * and then continue. For now, we just invalidate the requests,
2095 * schedule() and sleep one second if needed, and hope.
2101 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2103 struct ll_sb_info *sbi = ll_s2sbi(sb);
2104 char *profilenm = get_profile_name(sb);
2108 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2109 read_only = *flags & MS_RDONLY;
2110 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2111 sizeof(KEY_READ_ONLY),
2112 KEY_READ_ONLY, sizeof(read_only),
2115 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2116 profilenm, read_only ?
2117 "read-only" : "read-write", err);
2122 sb->s_flags |= MS_RDONLY;
2124 sb->s_flags &= ~MS_RDONLY;
2126 if (sbi->ll_flags & LL_SBI_VERBOSE)
2127 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2128 read_only ? "read-only" : "read-write");
2133 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2134 struct super_block *sb, struct lookup_intent *it)
2136 struct ll_sb_info *sbi = NULL;
2137 struct lustre_md md;
2141 LASSERT(*inode || sb);
2142 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2143 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2144 sbi->ll_md_exp, &md);
2149 ll_update_inode(*inode, &md);
2151 LASSERT(sb != NULL);
2154 * At this point server returns to client's same fid as client
2155 * generated for creating. So using ->fid1 is okay here.
2157 LASSERT(fid_is_sane(&md.body->fid1));
2159 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1,
2160 sbi->ll_flags & LL_SBI_32BIT_API),
2162 if (*inode == NULL || IS_ERR(*inode)) {
2163 #ifdef CONFIG_FS_POSIX_ACL
2165 posix_acl_release(md.posix_acl);
2166 md.posix_acl = NULL;
2169 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2171 CERROR("new_inode -fatal: rc %d\n", rc);
2176 /* Handling piggyback layout lock.
2177 * Layout lock can be piggybacked by getattr and open request.
2178 * The lsm can be applied to inode only if it comes with a layout lock
2179 * otherwise correct layout may be overwritten, for example:
2180 * 1. proc1: mdt returns a lsm but not granting layout
2181 * 2. layout was changed by another client
2182 * 3. proc2: refresh layout and layout lock granted
2183 * 4. proc1: to apply a stale layout */
2184 if (it != NULL && it->d.lustre.it_lock_mode != 0) {
2185 struct lustre_handle lockh;
2186 struct ldlm_lock *lock;
2188 lockh.cookie = it->d.lustre.it_lock_handle;
2189 lock = ldlm_handle2lock(&lockh);
2190 LASSERT(lock != NULL);
2191 if (ldlm_has_layout(lock)) {
2192 struct cl_object_conf conf;
2194 memset(&conf, 0, sizeof(conf));
2195 conf.coc_opc = OBJECT_CONF_SET;
2196 conf.coc_inode = *inode;
2197 conf.coc_lock = lock;
2198 conf.u.coc_md = &md;
2199 (void)ll_layout_conf(*inode, &conf);
2201 LDLM_LOCK_PUT(lock);
2206 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2207 md_free_lustre_md(sbi->ll_md_exp, &md);
2211 int ll_obd_statfs(struct inode *inode, void *arg)
2213 struct ll_sb_info *sbi = NULL;
2214 struct obd_export *exp;
2216 struct obd_ioctl_data *data = NULL;
2221 if (!inode || !(sbi = ll_i2sbi(inode)))
2222 GOTO(out_statfs, rc = -EINVAL);
2224 rc = obd_ioctl_getdata(&buf, &len, arg);
2226 GOTO(out_statfs, rc);
2229 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2230 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2231 GOTO(out_statfs, rc = -EINVAL);
2233 if (data->ioc_inllen1 != sizeof(__u32) ||
2234 data->ioc_inllen2 != sizeof(__u32) ||
2235 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2236 data->ioc_plen2 != sizeof(struct obd_uuid))
2237 GOTO(out_statfs, rc = -EINVAL);
2239 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2240 if (type & LL_STATFS_LMV)
2241 exp = sbi->ll_md_exp;
2242 else if (type & LL_STATFS_LOV)
2243 exp = sbi->ll_dt_exp;
2245 GOTO(out_statfs, rc = -ENODEV);
2247 flags = (type & LL_STATFS_NODELAY) ? OBD_STATFS_NODELAY : 0;
2248 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, &flags);
2250 GOTO(out_statfs, rc);
2253 obd_ioctl_freedata(buf, len);
2257 int ll_process_config(struct lustre_cfg *lcfg)
2261 struct lprocfs_static_vars lvars;
2265 lprocfs_llite_init_vars(&lvars);
2267 /* The instance name contains the sb: lustre-client-aacfe000 */
2268 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2269 if (!ptr || !*(++ptr))
2271 if (sscanf(ptr, "%lx", &x) != 1)
2274 /* This better be a real Lustre superblock! */
2275 LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2277 /* Note we have not called client_common_fill_super yet, so
2278 proc fns must be able to handle that! */
2279 rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2286 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2287 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2288 struct inode *i1, struct inode *i2,
2289 const char *name, int namelen,
2290 int mode, __u32 opc, void *data)
2292 LASSERT(i1 != NULL);
2294 if (namelen > ll_i2sbi(i1)->ll_namelen)
2295 return ERR_PTR(-ENAMETOOLONG);
2297 if (op_data == NULL)
2298 OBD_ALLOC_PTR(op_data);
2300 if (op_data == NULL)
2301 return ERR_PTR(-ENOMEM);
2303 ll_i2gids(op_data->op_suppgids, i1, i2);
2304 op_data->op_fid1 = *ll_inode2fid(i1);
2305 op_data->op_capa1 = ll_mdscapa_get(i1);
2308 op_data->op_fid2 = *ll_inode2fid(i2);
2309 op_data->op_capa2 = ll_mdscapa_get(i2);
2311 fid_zero(&op_data->op_fid2);
2312 op_data->op_capa2 = NULL;
2315 op_data->op_name = name;
2316 op_data->op_namelen = namelen;
2317 op_data->op_mode = mode;
2318 op_data->op_mod_time = cfs_time_current_sec();
2319 op_data->op_fsuid = cfs_curproc_fsuid();
2320 op_data->op_fsgid = cfs_curproc_fsgid();
2321 op_data->op_cap = cfs_curproc_cap_pack();
2322 op_data->op_bias = 0;
2323 op_data->op_cli_flags = 0;
2324 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2325 filename_is_volatile(name, namelen, NULL))
2326 op_data->op_bias |= MDS_CREATE_VOLATILE;
2327 op_data->op_opc = opc;
2328 op_data->op_mds = 0;
2329 op_data->op_data = data;
2331 /* If the file is being opened after mknod() (normally due to NFS)
2332 * try to use the default stripe data from parent directory for
2333 * allocating OST objects. Try to pass the parent FID to MDS. */
2334 if (opc == LUSTRE_OPC_CREATE && i1 == i2 && S_ISREG(i2->i_mode) &&
2335 !ll_i2info(i2)->lli_has_smd) {
2336 struct ll_inode_info *lli = ll_i2info(i2);
2338 spin_lock(&lli->lli_lock);
2339 if (likely(!lli->lli_has_smd && !fid_is_zero(&lli->lli_pfid)))
2340 op_data->op_fid1 = lli->lli_pfid;
2341 spin_unlock(&lli->lli_lock);
2342 /** We ignore parent's capability temporary. */
2345 /* When called by ll_setattr_raw, file is i1. */
2346 if (LLIF_DATA_MODIFIED & ll_i2info(i1)->lli_flags)
2347 op_data->op_bias |= MDS_DATA_MODIFIED;
2352 void ll_finish_md_op_data(struct md_op_data *op_data)
2354 capa_put(op_data->op_capa1);
2355 capa_put(op_data->op_capa2);
2356 OBD_FREE_PTR(op_data);
2359 #ifdef HAVE_SUPEROPS_USE_DENTRY
2360 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2362 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2365 struct ll_sb_info *sbi;
2367 #ifdef HAVE_SUPEROPS_USE_DENTRY
2368 LASSERT((seq != NULL) && (dentry != NULL));
2369 sbi = ll_s2sbi(dentry->d_sb);
2371 LASSERT((seq != NULL) && (vfs != NULL));
2372 sbi = ll_s2sbi(vfs->mnt_sb);
2375 if (sbi->ll_flags & LL_SBI_NOLCK)
2376 seq_puts(seq, ",nolock");
2378 if (sbi->ll_flags & LL_SBI_FLOCK)
2379 seq_puts(seq, ",flock");
2381 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2382 seq_puts(seq, ",localflock");
2384 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2385 seq_puts(seq, ",user_xattr");
2387 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2388 seq_puts(seq, ",lazystatfs");
2390 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2391 seq_puts(seq, ",user_fid2path");
2397 * Get obd name by cmd, and copy out to user space
2399 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2401 struct ll_sb_info *sbi = ll_i2sbi(inode);
2402 struct obd_device *obd;
2405 if (cmd == OBD_IOC_GETDTNAME)
2406 obd = class_exp2obd(sbi->ll_dt_exp);
2407 else if (cmd == OBD_IOC_GETMDNAME)
2408 obd = class_exp2obd(sbi->ll_md_exp);
2415 if (copy_to_user((void *)arg, obd->obd_name,
2416 strlen(obd->obd_name) + 1))
2423 * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2424 * fsname will be returned in this buffer; otherwise, a static buffer will be
2425 * used to store the fsname and returned to caller.
2427 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2429 static char fsname_static[MTI_NAME_MAXLEN];
2430 struct lustre_sb_info *lsi = s2lsi(sb);
2435 /* this means the caller wants to use static buffer
2436 * and it doesn't care about race. Usually this is
2437 * in error reporting path */
2438 buf = fsname_static;
2439 buflen = sizeof(fsname_static);
2442 len = strlen(lsi->lsi_lmd->lmd_profile);
2443 ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2444 if (ptr && (strcmp(ptr, "-client") == 0))
2447 if (unlikely(len >= buflen))
2449 strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2455 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2459 #ifdef HAVE_FS_STRUCT_USE_PATH
2463 p.mnt = current->fs->root.mnt;
2465 path = d_path(&p, buf, bufsize);
2468 path = d_path(dentry, current->fs->rootmnt, buf, bufsize);
2474 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2476 char *buf, *path = NULL;
2477 struct dentry *dentry = NULL;
2478 struct ccc_object *obj = cl_inode2ccc(page->mapping->host);
2480 /* this can be called inside spin lock so use GFP_ATOMIC. */
2481 buf = (char *)__get_free_page(GFP_ATOMIC);
2483 dentry = d_find_alias(page->mapping->host);
2485 path = ll_d_path(dentry, buf, PAGE_SIZE);
2488 CWARN("%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2489 "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2490 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2491 PFID(&obj->cob_header.coh_lu.loh_fid),
2492 (path && !IS_ERR(path)) ? path : "", ioret);
2498 free_page((unsigned long)buf);