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 lru_page_max = pages / 2;
98 /* initialize ll_cache data */
99 cfs_atomic_set(&sbi->ll_cache.ccc_users, 0);
100 sbi->ll_cache.ccc_lru_max = lru_page_max;
101 cfs_atomic_set(&sbi->ll_cache.ccc_lru_left, lru_page_max);
102 spin_lock_init(&sbi->ll_cache.ccc_lru_lock);
103 CFS_INIT_LIST_HEAD(&sbi->ll_cache.ccc_lru);
105 cfs_atomic_set(&sbi->ll_cache.ccc_unstable_nr, 0);
106 init_waitqueue_head(&sbi->ll_cache.ccc_unstable_waitq);
108 sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
109 SBI_DEFAULT_READAHEAD_MAX);
110 sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
111 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
112 SBI_DEFAULT_READAHEAD_WHOLE_MAX;
113 CFS_INIT_LIST_HEAD(&sbi->ll_conn_chain);
114 CFS_INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
116 ll_generate_random_uuid(uuid);
117 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
118 CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
120 spin_lock(&ll_sb_lock);
121 cfs_list_add_tail(&sbi->ll_list, &ll_super_blocks);
122 spin_unlock(&ll_sb_lock);
124 sbi->ll_flags |= LL_SBI_VERBOSE;
125 #ifdef ENABLE_CHECKSUM
126 sbi->ll_flags |= LL_SBI_CHECKSUM;
129 #ifdef HAVE_LRU_RESIZE_SUPPORT
130 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
133 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
134 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
136 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
140 /* metadata statahead is enabled by default */
141 sbi->ll_sa_max = LL_SA_RPC_DEF;
142 cfs_atomic_set(&sbi->ll_sa_total, 0);
143 cfs_atomic_set(&sbi->ll_sa_wrong, 0);
144 cfs_atomic_set(&sbi->ll_agl_total, 0);
145 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
150 void ll_free_sbi(struct super_block *sb)
152 struct ll_sb_info *sbi = ll_s2sbi(sb);
156 spin_lock(&ll_sb_lock);
157 cfs_list_del(&sbi->ll_list);
158 spin_unlock(&ll_sb_lock);
159 OBD_FREE(sbi, sizeof(*sbi));
164 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
165 struct vfsmount *mnt)
167 struct inode *root = 0;
168 struct ll_sb_info *sbi = ll_s2sbi(sb);
169 struct obd_device *obd;
170 struct obd_capa *oc = NULL;
171 struct obd_statfs *osfs = NULL;
172 struct ptlrpc_request *request = NULL;
173 struct obd_connect_data *data = NULL;
174 struct obd_uuid *uuid;
175 struct md_op_data *op_data;
176 struct lustre_md lmd;
178 int size, err, checksum;
181 obd = class_name2obd(md);
183 CERROR("MD %s: not setup or attached\n", md);
197 /* indicate the features supported by this client */
198 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
199 OBD_CONNECT_ATTRFID |
200 OBD_CONNECT_VERSION | OBD_CONNECT_BRW_SIZE |
201 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
202 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
203 OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 |
204 OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR |
205 OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH|
206 OBD_CONNECT_EINPROGRESS |
207 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
208 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS |
209 OBD_CONNECT_MAX_EASIZE |
210 OBD_CONNECT_FLOCK_DEAD |
211 OBD_CONNECT_DISP_STRIPE;
213 if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
214 data->ocd_connect_flags |= OBD_CONNECT_SOM;
216 #ifdef HAVE_LRU_RESIZE_SUPPORT
217 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
218 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
220 #ifdef CONFIG_FS_POSIX_ACL
221 data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK;
224 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
225 /* flag mdc connection as lightweight, only used for test
226 * purpose, use with care */
227 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
229 data->ocd_ibits_known = MDS_INODELOCK_FULL;
230 data->ocd_version = LUSTRE_VERSION_CODE;
232 if (sb->s_flags & MS_RDONLY)
233 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
234 if (sbi->ll_flags & LL_SBI_USER_XATTR)
235 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
237 #ifdef HAVE_MS_FLOCK_LOCK
238 /* force vfs to use lustre handler for flock() calls - bug 10743 */
239 sb->s_flags |= MS_FLOCK_LOCK;
241 #ifdef MS_HAS_NEW_AOPS
242 sb->s_flags |= MS_HAS_NEW_AOPS;
245 if (sbi->ll_flags & LL_SBI_FLOCK)
246 sbi->ll_fop = &ll_file_operations_flock;
247 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
248 sbi->ll_fop = &ll_file_operations;
250 sbi->ll_fop = &ll_file_operations_noflock;
253 data->ocd_connect_flags |= OBD_CONNECT_REAL;
254 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
255 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
257 data->ocd_brw_size = MD_MAX_BRW_SIZE;
259 err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
261 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
262 "recovery, of which this client is not a "
263 "part. Please wait for recovery to complete,"
264 " abort, or time out.\n", md);
267 CERROR("cannot connect to %s: rc = %d\n", md, err);
271 sbi->ll_md_exp->exp_connect_data = *data;
273 err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
274 LUSTRE_SEQ_METADATA);
276 CERROR("%s: Can't init metadata layer FID infrastructure, "
277 "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
281 /* For mount, we only need fs info from MDT0, and also in DNE, it
282 * can make sure the client can be mounted as long as MDT0 is
284 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
285 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
286 OBD_STATFS_FOR_MDT0);
288 GOTO(out_md_fid, err);
290 /* This needs to be after statfs to ensure connect has finished.
291 * Note that "data" does NOT contain the valid connect reply.
292 * If connecting to a 1.8 server there will be no LMV device, so
293 * we can access the MDC export directly and exp_connect_flags will
294 * be non-zero, but if accessing an upgraded 2.1 server it will
295 * have the correct flags filled in.
296 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
297 valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
298 if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
299 valid != CLIENT_CONNECT_MDT_REQD) {
302 OBD_ALLOC_WAIT(buf, PAGE_CACHE_SIZE);
303 obd_connect_flags2str(buf, PAGE_CACHE_SIZE,
304 valid ^ CLIENT_CONNECT_MDT_REQD, ",");
305 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
306 "feature(s) needed for correct operation "
307 "of this client (%s). Please upgrade "
308 "server or downgrade client.\n",
309 sbi->ll_md_exp->exp_obd->obd_name, buf);
310 OBD_FREE(buf, PAGE_CACHE_SIZE);
311 GOTO(out_md_fid, err = -EPROTO);
314 size = sizeof(*data);
315 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
316 KEY_CONN_DATA, &size, data, NULL);
318 CERROR("%s: Get connect data failed: rc = %d\n",
319 sbi->ll_md_exp->exp_obd->obd_name, err);
320 GOTO(out_md_fid, err);
323 LASSERT(osfs->os_bsize);
324 sb->s_blocksize = osfs->os_bsize;
325 sb->s_blocksize_bits = log2(osfs->os_bsize);
326 sb->s_magic = LL_SUPER_MAGIC;
327 sb->s_maxbytes = MAX_LFS_FILESIZE;
328 sbi->ll_namelen = osfs->os_namelen;
329 sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
331 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
332 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
333 LCONSOLE_INFO("Disabling user_xattr feature because "
334 "it is not supported on the server\n");
335 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
338 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
340 sb->s_flags |= MS_POSIXACL;
342 sbi->ll_flags |= LL_SBI_ACL;
344 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
346 sb->s_flags &= ~MS_POSIXACL;
348 sbi->ll_flags &= ~LL_SBI_ACL;
351 if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
352 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
353 sbi->ll_flags |= LL_SBI_RMT_CLIENT;
354 LCONSOLE_INFO("client is set as remote by default.\n");
357 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
358 sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
359 LCONSOLE_INFO("client claims to be remote, but server "
360 "rejected, forced to be local.\n");
364 if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
365 LCONSOLE_INFO("client enabled MDS capability!\n");
366 sbi->ll_flags |= LL_SBI_MDS_CAPA;
369 if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
370 LCONSOLE_INFO("client enabled OSS capability!\n");
371 sbi->ll_flags |= LL_SBI_OSS_CAPA;
374 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
375 sbi->ll_flags |= LL_SBI_64BIT_HASH;
377 if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
378 sbi->ll_md_brw_size = data->ocd_brw_size;
380 sbi->ll_md_brw_size = PAGE_CACHE_SIZE;
382 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
383 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
385 if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
386 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
387 LCONSOLE_INFO("%s: disabling xattr cache due to "
388 "unknown maximum xattr size.\n", dt);
390 sbi->ll_flags |= LL_SBI_XATTR_CACHE;
391 sbi->ll_xattr_cache_enabled = 1;
395 obd = class_name2obd(dt);
397 CERROR("DT %s: not setup or attached\n", dt);
398 GOTO(out_md_fid, err = -ENODEV);
401 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
402 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
403 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
404 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
405 OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
406 OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
407 OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
408 OBD_CONNECT_MAXBYTES |
409 OBD_CONNECT_EINPROGRESS |
410 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
411 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS;
413 if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
414 data->ocd_connect_flags |= OBD_CONNECT_SOM;
416 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
417 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
418 * disabled by default, because it can still be enabled on the
419 * fly via /proc. As a consequence, we still need to come to an
420 * agreement on the supported algorithms at connect time */
421 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
423 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
424 data->ocd_cksum_types = OBD_CKSUM_ADLER;
426 data->ocd_cksum_types = cksum_types_supported_client();
429 #ifdef HAVE_LRU_RESIZE_SUPPORT
430 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
432 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
433 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
435 CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
436 "ocd_grant: %d\n", data->ocd_connect_flags,
437 data->ocd_version, data->ocd_grant);
439 obd->obd_upcall.onu_owner = &sbi->ll_lco;
440 obd->obd_upcall.onu_upcall = cl_ocd_update;
442 data->ocd_brw_size = DT_MAX_BRW_SIZE;
444 err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
447 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
448 "recovery, of which this client is not a "
449 "part. Please wait for recovery to "
450 "complete, abort, or time out.\n", dt);
453 CERROR("%s: Cannot connect to %s: rc = %d\n",
454 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
458 sbi->ll_dt_exp->exp_connect_data = *data;
460 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
461 LUSTRE_SEQ_METADATA);
463 CERROR("%s: Can't init data layer FID infrastructure, "
464 "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
468 mutex_lock(&sbi->ll_lco.lco_lock);
469 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
470 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
471 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
472 mutex_unlock(&sbi->ll_lco.lco_lock);
474 fid_zero(&sbi->ll_root_fid);
475 err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
477 CERROR("cannot mds_connect: rc = %d\n", err);
478 GOTO(out_lock_cn_cb, err);
480 if (!fid_is_sane(&sbi->ll_root_fid)) {
481 CERROR("%s: Invalid root fid "DFID" during mount\n",
482 sbi->ll_md_exp->exp_obd->obd_name,
483 PFID(&sbi->ll_root_fid));
484 GOTO(out_lock_cn_cb, err = -EINVAL);
486 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
488 sb->s_op = &lustre_super_operations;
489 #if THREAD_SIZE >= 8192 /*b=17630*/
490 sb->s_export_op = &lustre_export_operations;
494 * XXX: move this to after cbd setup? */
495 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
496 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
497 valid |= OBD_MD_FLRMTPERM;
498 else if (sbi->ll_flags & LL_SBI_ACL)
499 valid |= OBD_MD_FLACL;
501 OBD_ALLOC_PTR(op_data);
503 GOTO(out_lock_cn_cb, err = -ENOMEM);
505 op_data->op_fid1 = sbi->ll_root_fid;
506 op_data->op_mode = 0;
507 op_data->op_capa1 = oc;
508 op_data->op_valid = valid;
510 err = md_getattr(sbi->ll_md_exp, op_data, &request);
513 OBD_FREE_PTR(op_data);
515 CERROR("%s: md_getattr failed for root: rc = %d\n",
516 sbi->ll_md_exp->exp_obd->obd_name, err);
517 GOTO(out_lock_cn_cb, err);
520 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
521 sbi->ll_md_exp, &lmd);
523 CERROR("failed to understand root inode md: rc = %d\n", err);
524 ptlrpc_req_finished(request);
525 GOTO(out_lock_cn_cb, err);
528 LASSERT(fid_is_sane(&sbi->ll_root_fid));
529 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
530 sbi->ll_flags & LL_SBI_32BIT_API),
532 md_free_lustre_md(sbi->ll_md_exp, &lmd);
533 ptlrpc_req_finished(request);
535 if (root == NULL || IS_ERR(root)) {
537 obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
538 #ifdef CONFIG_FS_POSIX_ACL
540 posix_acl_release(lmd.posix_acl);
541 lmd.posix_acl = NULL;
544 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
546 CERROR("lustre_lite: bad iget4 for root\n");
550 err = ll_close_thread_start(&sbi->ll_lcq);
552 CERROR("cannot start close thread: rc %d\n", err);
556 #ifdef CONFIG_FS_POSIX_ACL
557 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
558 rct_init(&sbi->ll_rct);
559 et_init(&sbi->ll_et);
563 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
564 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
565 KEY_CHECKSUM, sizeof(checksum), &checksum,
569 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
570 KEY_CACHE_SET, sizeof(sbi->ll_cache),
571 &sbi->ll_cache, NULL);
573 sb->s_root = d_make_root(root);
574 if (sb->s_root == NULL) {
575 CERROR("%s: can't make root dentry\n",
576 ll_get_fsname(sb, NULL, 0));
577 GOTO(out_root, err = -ENOMEM);
579 #ifdef HAVE_DCACHE_LOCK
580 sb->s_root->d_op = &ll_d_ops;
583 sbi->ll_sdev_orig = sb->s_dev;
585 /* We set sb->s_dev equal on all lustre clients in order to support
586 * NFS export clustering. NFSD requires that the FSID be the same
588 /* s_dev is also used in lt_compare() to compare two fs, but that is
589 * only a node-local comparison. */
590 uuid = obd_get_uuid(sbi->ll_md_exp);
592 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
598 if (proc_lustre_fs_root != NULL) {
599 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
602 CERROR("%s: could not register mount in lprocfs: "
603 "rc = %d\n", ll_get_fsname(sb, NULL, 0), err);
613 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
615 obd_disconnect(sbi->ll_dt_exp);
616 sbi->ll_dt_exp = NULL;
617 /* Make sure all OScs are gone, since cl_cache is accessing sbi. */
618 obd_zombie_barrier();
620 obd_fid_fini(sbi->ll_md_exp->exp_obd);
622 obd_disconnect(sbi->ll_md_exp);
623 sbi->ll_md_exp = NULL;
632 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
636 *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
638 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
639 KEY_MAX_EASIZE, &size, lmmsize, NULL);
641 CERROR("Get max mdsize error rc %d \n", rc);
646 void ll_dump_inode(struct inode *inode)
648 struct ll_d_hlist_node *tmp;
649 int dentry_count = 0;
651 LASSERT(inode != NULL);
653 ll_d_hlist_for_each(tmp, &inode->i_dentry)
656 CERROR("%s: inode %p dump: dev=%s fid="DFID
657 " mode=%o count=%u, %d dentries\n",
658 ll_get_fsname(inode->i_sb, NULL, 0), inode,
659 ll_i2mdexp(inode)->exp_obd->obd_name, PFID(ll_inode2fid(inode)),
660 inode->i_mode, atomic_read(&inode->i_count), dentry_count);
663 void lustre_dump_dentry(struct dentry *dentry, int recur)
665 struct list_head *tmp;
668 LASSERT(dentry != NULL);
670 list_for_each(tmp, &dentry->d_subdirs)
673 CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
674 " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
675 dentry->d_name.len, dentry->d_name.name,
676 dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
677 dentry->d_parent, dentry->d_inode, d_count(dentry),
678 dentry->d_flags, dentry->d_fsdata, subdirs);
679 if (dentry->d_inode != NULL)
680 ll_dump_inode(dentry->d_inode);
685 list_for_each(tmp, &dentry->d_subdirs) {
686 struct dentry *d = list_entry(tmp, struct dentry, d_u.d_child);
687 lustre_dump_dentry(d, recur - 1);
691 void client_common_put_super(struct super_block *sb)
693 struct ll_sb_info *sbi = ll_s2sbi(sb);
696 #ifdef CONFIG_FS_POSIX_ACL
697 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
698 et_fini(&sbi->ll_et);
699 rct_fini(&sbi->ll_rct);
703 ll_close_thread_shutdown(sbi->ll_lcq);
707 cfs_list_del(&sbi->ll_conn_chain);
709 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
710 obd_disconnect(sbi->ll_dt_exp);
711 sbi->ll_dt_exp = NULL;
712 /* wait till all OSCs are gone, since cl_cache is accessing sbi.
714 obd_zombie_barrier();
716 lprocfs_unregister_mountpoint(sbi);
718 obd_fid_fini(sbi->ll_md_exp->exp_obd);
719 obd_disconnect(sbi->ll_md_exp);
720 sbi->ll_md_exp = NULL;
725 void ll_kill_super(struct super_block *sb)
727 struct ll_sb_info *sbi;
732 if (!(sb->s_flags & MS_ACTIVE))
736 /* we need restore s_dev from changed for clustred NFS before put_super
737 * because new kernels have cached s_dev and change sb->s_dev in
738 * put_super not affected real removing devices */
740 sb->s_dev = sbi->ll_sdev_orig;
741 sbi->ll_umounting = 1;
746 char *ll_read_opt(const char *opt, char *data)
752 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
753 if (strncmp(opt, data, strlen(opt)))
755 if ((value = strchr(data, '=')) == NULL)
759 OBD_ALLOC(retval, strlen(value) + 1);
761 CERROR("out of memory!\n");
765 memcpy(retval, value, strlen(value)+1);
766 CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
770 static inline int ll_set_opt(const char *opt, char *data, int fl)
772 if (strncmp(opt, data, strlen(opt)) != 0)
778 /* non-client-specific mount options are parsed in lmd_parse */
779 static int ll_options(char *options, int *flags)
782 char *s1 = options, *s2;
788 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
791 CDEBUG(D_SUPER, "next opt=%s\n", s1);
792 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
797 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
802 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
807 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
812 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
817 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
822 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 5, 53, 0)
823 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
825 /* Ignore deprecated mount option. The client will
826 * always try to mount with ACL support, whether this
827 * is used depends on whether server supports it. */
828 LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
829 "mount option 'acl'.\n");
832 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
834 LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
835 "mount option 'noacl'.\n");
839 #warning "{no}acl options have been deprecated since 1.8, please remove them"
841 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
846 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
851 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
857 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
862 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
867 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
872 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
877 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
882 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
887 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
892 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
897 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
902 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
907 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
913 s2 = strchr(s1, ',');
921 void ll_lli_init(struct ll_inode_info *lli)
923 lli->lli_inode_magic = LLI_INODE_MAGIC;
925 lli->lli_ioepoch = 0;
926 lli->lli_maxbytes = MAX_LFS_FILESIZE;
927 spin_lock_init(&lli->lli_lock);
928 lli->lli_posix_acl = NULL;
929 lli->lli_remote_perms = NULL;
930 mutex_init(&lli->lli_rmtperm_mutex);
931 /* Do not set lli_fid, it has been initialized already. */
932 fid_zero(&lli->lli_pfid);
933 CFS_INIT_LIST_HEAD(&lli->lli_close_list);
934 CFS_INIT_LIST_HEAD(&lli->lli_oss_capas);
935 cfs_atomic_set(&lli->lli_open_count, 0);
936 lli->lli_mds_capa = NULL;
937 lli->lli_rmtperm_time = 0;
938 lli->lli_pending_och = NULL;
939 lli->lli_mds_read_och = NULL;
940 lli->lli_mds_write_och = NULL;
941 lli->lli_mds_exec_och = NULL;
942 lli->lli_open_fd_read_count = 0;
943 lli->lli_open_fd_write_count = 0;
944 lli->lli_open_fd_exec_count = 0;
945 mutex_init(&lli->lli_och_mutex);
946 spin_lock_init(&lli->lli_agl_lock);
947 lli->lli_has_smd = false;
948 lli->lli_layout_gen = LL_LAYOUT_GEN_NONE;
949 lli->lli_clob = NULL;
951 init_rwsem(&lli->lli_xattrs_list_rwsem);
952 mutex_init(&lli->lli_xattrs_enq_lock);
954 LASSERT(lli->lli_vfs_inode.i_mode != 0);
955 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
956 mutex_init(&lli->lli_readdir_mutex);
957 lli->lli_opendir_key = NULL;
959 lli->lli_def_acl = NULL;
960 spin_lock_init(&lli->lli_sa_lock);
961 lli->lli_opendir_pid = 0;
963 sema_init(&lli->lli_size_sem, 1);
964 lli->lli_size_sem_owner = NULL;
965 lli->lli_symlink_name = NULL;
966 init_rwsem(&lli->lli_trunc_sem);
967 mutex_init(&lli->lli_write_mutex);
968 init_rwsem(&lli->lli_glimpse_sem);
969 lli->lli_glimpse_time = 0;
970 CFS_INIT_LIST_HEAD(&lli->lli_agl_list);
971 lli->lli_agl_index = 0;
972 lli->lli_async_rc = 0;
973 lli->lli_volatile = false;
975 mutex_init(&lli->lli_layout_mutex);
978 static inline int ll_bdi_register(struct backing_dev_info *bdi)
980 static atomic_t ll_bdi_num = ATOMIC_INIT(0);
982 bdi->name = "lustre";
983 return bdi_register(bdi, NULL, "lustre-%d",
984 atomic_inc_return(&ll_bdi_num));
987 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
989 struct lustre_profile *lprof = NULL;
990 struct lustre_sb_info *lsi = s2lsi(sb);
991 struct ll_sb_info *sbi;
992 char *dt = NULL, *md = NULL;
993 char *profilenm = get_profile_name(sb);
994 struct config_llog_instance *cfg;
995 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
996 const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
1000 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1006 try_module_get(THIS_MODULE);
1008 /* client additional sb info */
1009 lsi->lsi_llsbi = sbi = ll_init_sbi();
1011 module_put(THIS_MODULE);
1016 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1018 GOTO(out_free, err);
1020 err = bdi_init(&lsi->lsi_bdi);
1022 GOTO(out_free, err);
1023 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1024 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1025 err = ll_bdi_register(&lsi->lsi_bdi);
1027 GOTO(out_free, err);
1029 sb->s_bdi = &lsi->lsi_bdi;
1030 #ifndef HAVE_DCACHE_LOCK
1031 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1032 sb->s_d_op = &ll_d_ops;
1035 /* Generate a string unique to this super, in case some joker tries
1036 to mount the same fs at two mount points.
1037 Use the address of the super itself.*/
1038 cfg->cfg_instance = sb;
1039 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1040 cfg->cfg_callback = class_config_llog_handler;
1041 /* set up client obds */
1042 err = lustre_process_log(sb, profilenm, cfg);
1044 CERROR("Unable to process log: %d\n", err);
1045 GOTO(out_free, err);
1048 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1049 lprof = class_get_profile(profilenm);
1050 if (lprof == NULL) {
1051 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1052 " read from the MGS. Does that filesystem "
1053 "exist?\n", profilenm);
1054 GOTO(out_free, err = -EINVAL);
1056 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1057 lprof->lp_md, lprof->lp_dt);
1059 OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1061 GOTO(out_free, err = -ENOMEM);
1062 sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1064 OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1066 GOTO(out_free, err = -ENOMEM);
1067 sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1069 /* connections, registrations, sb setup */
1070 err = client_common_fill_super(sb, md, dt, mnt);
1074 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1076 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1079 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1080 LCONSOLE_WARN("Mounted %s\n", profilenm);
1084 } /* ll_fill_super */
1087 void lu_context_keys_dump(void);
1089 void ll_put_super(struct super_block *sb)
1091 struct config_llog_instance cfg, params_cfg;
1092 struct obd_device *obd;
1093 struct lustre_sb_info *lsi = s2lsi(sb);
1094 struct ll_sb_info *sbi = ll_s2sbi(sb);
1095 char *profilenm = get_profile_name(sb);
1096 int ccc_count, next, force = 1, rc = 0;
1099 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1101 ll_print_capa_stat(sbi);
1103 cfg.cfg_instance = sb;
1104 lustre_end_log(sb, profilenm, &cfg);
1106 params_cfg.cfg_instance = sb;
1107 lustre_end_log(sb, PARAMS_FILENAME, ¶ms_cfg);
1109 if (sbi->ll_md_exp) {
1110 obd = class_exp2obd(sbi->ll_md_exp);
1112 force = obd->obd_force;
1115 /* Wait for unstable pages to be committed to stable storage */
1117 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1118 rc = l_wait_event(sbi->ll_cache.ccc_unstable_waitq,
1119 cfs_atomic_read(&sbi->ll_cache.ccc_unstable_nr) == 0,
1123 ccc_count = cfs_atomic_read(&sbi->ll_cache.ccc_unstable_nr);
1124 if (force == 0 && rc != -EINTR)
1125 LASSERTF(ccc_count == 0, "count: %i\n", ccc_count);
1128 /* We need to set force before the lov_disconnect in
1129 lustre_common_put_super, since l_d cleans up osc's as well. */
1132 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1134 obd->obd_force = force;
1139 /* Only if client_common_fill_super succeeded */
1140 client_common_put_super(sb);
1144 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1145 class_manual_cleanup(obd);
1148 if (sbi->ll_flags & LL_SBI_VERBOSE)
1149 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1152 class_del_profile(profilenm);
1154 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1155 bdi_destroy(&lsi->lsi_bdi);
1156 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1160 lsi->lsi_llsbi = NULL;
1162 lustre_common_put_super(sb);
1164 cl_env_cache_purge(~0);
1166 module_put(THIS_MODULE);
1169 } /* client_put_super */
1171 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1173 struct inode *inode = NULL;
1175 /* NOTE: we depend on atomic igrab() -bzzz */
1176 lock_res_and_lock(lock);
1177 if (lock->l_resource->lr_lvb_inode) {
1178 struct ll_inode_info * lli;
1179 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1180 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1181 inode = igrab(lock->l_resource->lr_lvb_inode);
1183 inode = lock->l_resource->lr_lvb_inode;
1184 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1185 D_WARNING, lock, "lr_lvb_inode %p is "
1186 "bogus: magic %08x",
1187 lock->l_resource->lr_lvb_inode,
1188 lli->lli_inode_magic);
1192 unlock_res_and_lock(lock);
1196 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1198 struct inode *inode = NULL;
1199 /* NOTE: we depend on atomic igrab() -bzzz */
1200 lock_res_and_lock(lock);
1201 if (lock->l_ast_data) {
1202 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1203 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1204 inode = igrab(lock->l_ast_data);
1206 inode = lock->l_ast_data;
1207 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1208 D_WARNING, lock, "l_ast_data %p is "
1209 "bogus: magic %08x", lock->l_ast_data,
1210 lli->lli_inode_magic);
1214 unlock_res_and_lock(lock);
1218 void ll_clear_inode(struct inode *inode)
1220 struct ll_inode_info *lli = ll_i2info(inode);
1221 struct ll_sb_info *sbi = ll_i2sbi(inode);
1224 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1225 PFID(ll_inode2fid(inode)), inode);
1227 if (S_ISDIR(inode->i_mode)) {
1228 /* these should have been cleared in ll_file_release */
1229 LASSERT(lli->lli_opendir_key == NULL);
1230 LASSERT(lli->lli_sai == NULL);
1231 LASSERT(lli->lli_opendir_pid == 0);
1234 spin_lock(&lli->lli_lock);
1235 ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1236 spin_unlock(&lli->lli_lock);
1237 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1239 LASSERT(!lli->lli_open_fd_write_count);
1240 LASSERT(!lli->lli_open_fd_read_count);
1241 LASSERT(!lli->lli_open_fd_exec_count);
1243 if (lli->lli_mds_write_och)
1244 ll_md_real_close(inode, FMODE_WRITE);
1245 if (lli->lli_mds_exec_och)
1246 ll_md_real_close(inode, FMODE_EXEC);
1247 if (lli->lli_mds_read_och)
1248 ll_md_real_close(inode, FMODE_READ);
1250 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1251 OBD_FREE(lli->lli_symlink_name,
1252 strlen(lli->lli_symlink_name) + 1);
1253 lli->lli_symlink_name = NULL;
1256 ll_xattr_cache_destroy(inode);
1258 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1259 LASSERT(lli->lli_posix_acl == NULL);
1260 if (lli->lli_remote_perms) {
1261 free_rmtperm_hash(lli->lli_remote_perms);
1262 lli->lli_remote_perms = NULL;
1265 #ifdef CONFIG_FS_POSIX_ACL
1266 else if (lli->lli_posix_acl) {
1267 LASSERT(cfs_atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1268 LASSERT(lli->lli_remote_perms == NULL);
1269 posix_acl_release(lli->lli_posix_acl);
1270 lli->lli_posix_acl = NULL;
1273 lli->lli_inode_magic = LLI_INODE_DEAD;
1275 ll_clear_inode_capas(inode);
1276 if (!S_ISDIR(inode->i_mode))
1277 LASSERT(cfs_list_empty(&lli->lli_agl_list));
1280 * XXX This has to be done before lsm is freed below, because
1281 * cl_object still uses inode lsm.
1283 cl_inode_fini(inode);
1284 lli->lli_has_smd = false;
1289 int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data,
1290 struct md_open_data **mod)
1292 struct lustre_md md;
1293 struct inode *inode = dentry->d_inode;
1294 struct ll_sb_info *sbi = ll_i2sbi(inode);
1295 struct ptlrpc_request *request = NULL;
1299 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1300 LUSTRE_OPC_ANY, NULL);
1301 if (IS_ERR(op_data))
1302 RETURN(PTR_ERR(op_data));
1304 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1307 ptlrpc_req_finished(request);
1308 if (rc == -ENOENT) {
1310 /* Unlinked special device node? Or just a race?
1311 * Pretend we done everything. */
1312 if (!S_ISREG(inode->i_mode) &&
1313 !S_ISDIR(inode->i_mode)) {
1314 ia_valid = op_data->op_attr.ia_valid;
1315 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1316 rc = simple_setattr(dentry, &op_data->op_attr);
1317 op_data->op_attr.ia_valid = ia_valid;
1319 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1320 CERROR("md_setattr fails: rc = %d\n", rc);
1325 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1326 sbi->ll_md_exp, &md);
1328 ptlrpc_req_finished(request);
1332 ia_valid = op_data->op_attr.ia_valid;
1333 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1334 * cache is not cleared yet. */
1335 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1336 rc = simple_setattr(dentry, &op_data->op_attr);
1337 op_data->op_attr.ia_valid = ia_valid;
1339 /* Extract epoch data if obtained. */
1340 op_data->op_handle = md.body->handle;
1341 op_data->op_ioepoch = md.body->ioepoch;
1343 ll_update_inode(inode, &md);
1344 ptlrpc_req_finished(request);
1349 /* Close IO epoch and send Size-on-MDS attribute update. */
1350 static int ll_setattr_done_writing(struct inode *inode,
1351 struct md_op_data *op_data,
1352 struct md_open_data *mod)
1354 struct ll_inode_info *lli = ll_i2info(inode);
1358 LASSERT(op_data != NULL);
1359 if (!S_ISREG(inode->i_mode))
1362 CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1363 op_data->op_ioepoch, PFID(&lli->lli_fid));
1365 op_data->op_flags = MF_EPOCH_CLOSE;
1366 ll_done_writing_attr(inode, op_data);
1367 ll_pack_inode2opdata(inode, op_data, NULL);
1369 rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1370 if (rc == -EAGAIN) {
1371 /* MDS has instructed us to obtain Size-on-MDS attribute
1372 * from OSTs and send setattr to back to MDS. */
1373 rc = ll_som_update(inode, op_data);
1375 CERROR("%s: inode "DFID" mdc truncate failed: rc = %d\n",
1376 ll_i2sbi(inode)->ll_md_exp->exp_obd->obd_name,
1377 PFID(ll_inode2fid(inode)), rc);
1382 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1384 struct obd_capa *capa;
1387 if (attr->ia_valid & ATTR_SIZE)
1388 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1390 capa = ll_mdscapa_get(inode);
1392 rc = cl_setattr_ost(inode, attr, capa);
1394 if (attr->ia_valid & ATTR_SIZE)
1395 ll_truncate_free_capa(capa);
1402 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1403 * object(s) determine the file size and mtime. Otherwise, the MDS will
1404 * keep these values until such a time that objects are allocated for it.
1405 * We do the MDS operations first, as it is checking permissions for us.
1406 * We don't to the MDS RPC if there is nothing that we want to store there,
1407 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1408 * going to do an RPC anyways.
1410 * If we are doing a truncate, we will send the mtime and ctime updates
1411 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1412 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1415 * In case of HSMimport, we only set attr on MDS.
1417 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1419 struct inode *inode = dentry->d_inode;
1420 struct ll_inode_info *lli = ll_i2info(inode);
1421 struct md_op_data *op_data = NULL;
1422 struct md_open_data *mod = NULL;
1423 bool file_is_released = false;
1424 int rc = 0, rc1 = 0;
1427 CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1428 "valid %x, hsm_import %d\n",
1429 ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1430 inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1433 if (attr->ia_valid & ATTR_SIZE) {
1434 /* Check new size against VFS/VM file size limit and rlimit */
1435 rc = inode_newsize_ok(inode, attr->ia_size);
1439 /* The maximum Lustre file size is variable, based on the
1440 * OST maximum object size and number of stripes. This
1441 * needs another check in addition to the VFS check above. */
1442 if (attr->ia_size > ll_file_maxbytes(inode)) {
1443 CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1444 PFID(&lli->lli_fid), attr->ia_size,
1445 ll_file_maxbytes(inode));
1449 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1452 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1453 if (attr->ia_valid & TIMES_SET_FLAGS) {
1454 if (current_fsuid() != inode->i_uid &&
1455 !cfs_capable(CFS_CAP_FOWNER))
1459 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1460 if (attr->ia_valid & ATTR_CTIME) {
1461 attr->ia_ctime = CFS_CURRENT_TIME;
1462 attr->ia_valid |= ATTR_CTIME_SET;
1464 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1465 (attr->ia_valid & ATTR_ATIME)) {
1466 attr->ia_atime = CFS_CURRENT_TIME;
1467 attr->ia_valid |= ATTR_ATIME_SET;
1469 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1470 (attr->ia_valid & ATTR_MTIME)) {
1471 attr->ia_mtime = CFS_CURRENT_TIME;
1472 attr->ia_valid |= ATTR_MTIME_SET;
1475 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1476 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1477 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1478 cfs_time_current_sec());
1480 /* We always do an MDS RPC, even if we're only changing the size;
1481 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1483 OBD_ALLOC_PTR(op_data);
1484 if (op_data == NULL)
1487 if (!S_ISDIR(inode->i_mode)) {
1488 if (attr->ia_valid & ATTR_SIZE)
1489 inode_dio_write_done(inode);
1490 mutex_unlock(&inode->i_mutex);
1491 down_write(&lli->lli_trunc_sem);
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);
1508 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1509 if (file_is_released) {
1510 rc = ll_layout_restore(inode, 0, attr->ia_size);
1514 file_is_released = false;
1515 ll_layout_refresh(inode, &gen);
1518 /* If we are changing file size, file content is
1519 * modified, flag it. */
1520 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1521 spin_lock(&lli->lli_lock);
1522 lli->lli_flags |= LLIF_DATA_MODIFIED;
1523 spin_unlock(&lli->lli_lock);
1524 op_data->op_bias |= MDS_DATA_MODIFIED;
1528 memcpy(&op_data->op_attr, attr, sizeof(*attr));
1530 /* Open epoch for truncate. */
1531 if (exp_connect_som(ll_i2mdexp(inode)) && !hsm_import &&
1532 (attr->ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1533 op_data->op_flags = MF_EPOCH_OPEN;
1535 rc = ll_md_setattr(dentry, op_data, &mod);
1539 /* RPC to MDT is sent, cancel data modification flag */
1540 if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) {
1541 spin_lock(&lli->lli_lock);
1542 lli->lli_flags &= ~LLIF_DATA_MODIFIED;
1543 spin_unlock(&lli->lli_lock);
1546 ll_ioepoch_open(lli, op_data->op_ioepoch);
1547 if (!S_ISREG(inode->i_mode) || file_is_released)
1550 if (attr->ia_valid & (ATTR_SIZE |
1551 ATTR_ATIME | ATTR_ATIME_SET |
1552 ATTR_MTIME | ATTR_MTIME_SET)) {
1553 /* For truncate and utimes sending attributes to OSTs, setting
1554 * mtime/atime to the past will be performed under PW [0:EOF]
1555 * extent lock (new_size:EOF for truncate). It may seem
1556 * excessive to send mtime/atime updates to OSTs when not
1557 * setting times to past, but it is necessary due to possible
1558 * time de-synchronization between MDT inode and OST objects */
1559 rc = ll_setattr_ost(inode, attr);
1564 if (op_data->op_ioepoch) {
1565 rc1 = ll_setattr_done_writing(inode, op_data, mod);
1569 ll_finish_md_op_data(op_data);
1571 if (!S_ISDIR(inode->i_mode)) {
1572 up_write(&lli->lli_trunc_sem);
1573 mutex_lock(&inode->i_mutex);
1574 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1575 inode_dio_wait(inode);
1578 ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1579 LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1584 int ll_setattr(struct dentry *de, struct iattr *attr)
1586 int mode = de->d_inode->i_mode;
1588 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1589 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1590 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1592 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1593 (ATTR_SIZE|ATTR_MODE)) &&
1594 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1595 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1596 !(attr->ia_mode & S_ISGID))))
1597 attr->ia_valid |= ATTR_FORCE;
1599 if ((mode & S_ISUID) &&
1600 !(attr->ia_mode & S_ISUID) &&
1601 !(attr->ia_valid & ATTR_KILL_SUID))
1602 attr->ia_valid |= ATTR_KILL_SUID;
1604 if (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1605 !(attr->ia_mode & S_ISGID) &&
1606 !(attr->ia_valid & ATTR_KILL_SGID))
1607 attr->ia_valid |= ATTR_KILL_SGID;
1609 return ll_setattr_raw(de, attr, false);
1612 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1613 __u64 max_age, __u32 flags)
1615 struct ll_sb_info *sbi = ll_s2sbi(sb);
1616 struct obd_statfs obd_osfs;
1620 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1622 CERROR("md_statfs fails: rc = %d\n", rc);
1626 osfs->os_type = sb->s_magic;
1628 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1629 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1631 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1632 flags |= OBD_STATFS_NODELAY;
1634 rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1636 CERROR("obd_statfs fails: rc = %d\n", rc);
1640 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1641 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1644 osfs->os_bsize = obd_osfs.os_bsize;
1645 osfs->os_blocks = obd_osfs.os_blocks;
1646 osfs->os_bfree = obd_osfs.os_bfree;
1647 osfs->os_bavail = obd_osfs.os_bavail;
1649 /* If we don't have as many objects free on the OST as inodes
1650 * on the MDS, we reduce the total number of inodes to
1651 * compensate, so that the "inodes in use" number is correct.
1653 if (obd_osfs.os_ffree < osfs->os_ffree) {
1654 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1656 osfs->os_ffree = obd_osfs.os_ffree;
1661 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1663 struct super_block *sb = de->d_sb;
1664 struct obd_statfs osfs;
1665 __u64 fsid = huge_encode_dev(sb->s_dev);
1668 CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1669 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1671 /* Some amount of caching on the client is allowed */
1672 rc = ll_statfs_internal(sb, &osfs,
1673 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1678 statfs_unpack(sfs, &osfs);
1680 /* We need to downshift for all 32-bit kernels, because we can't
1681 * tell if the kernel is being called via sys_statfs64() or not.
1682 * Stop before overflowing f_bsize - in which case it is better
1683 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1684 if (sizeof(long) < 8) {
1685 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1688 osfs.os_blocks >>= 1;
1689 osfs.os_bfree >>= 1;
1690 osfs.os_bavail >>= 1;
1694 sfs->f_blocks = osfs.os_blocks;
1695 sfs->f_bfree = osfs.os_bfree;
1696 sfs->f_bavail = osfs.os_bavail;
1697 sfs->f_fsid.val[0] = (__u32)fsid;
1698 sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1702 void ll_inode_size_lock(struct inode *inode)
1704 struct ll_inode_info *lli;
1706 LASSERT(!S_ISDIR(inode->i_mode));
1708 lli = ll_i2info(inode);
1709 LASSERT(lli->lli_size_sem_owner != current);
1710 down(&lli->lli_size_sem);
1711 LASSERT(lli->lli_size_sem_owner == NULL);
1712 lli->lli_size_sem_owner = current;
1715 void ll_inode_size_unlock(struct inode *inode)
1717 struct ll_inode_info *lli;
1719 lli = ll_i2info(inode);
1720 LASSERT(lli->lli_size_sem_owner == current);
1721 lli->lli_size_sem_owner = NULL;
1722 up(&lli->lli_size_sem);
1725 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1727 struct ll_inode_info *lli = ll_i2info(inode);
1728 struct mdt_body *body = md->body;
1729 struct lov_stripe_md *lsm = md->lsm;
1730 struct ll_sb_info *sbi = ll_i2sbi(inode);
1732 LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1734 if (!lli->lli_has_smd &&
1735 !(sbi->ll_flags & LL_SBI_LAYOUT_LOCK))
1736 cl_file_inode_init(inode, md);
1738 lli->lli_maxbytes = lsm->lsm_maxbytes;
1739 if (lli->lli_maxbytes > MAX_LFS_FILESIZE)
1740 lli->lli_maxbytes = MAX_LFS_FILESIZE;
1743 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1744 if (body->valid & OBD_MD_FLRMTPERM)
1745 ll_update_remote_perm(inode, md->remote_perm);
1747 #ifdef CONFIG_FS_POSIX_ACL
1748 else if (body->valid & OBD_MD_FLACL) {
1749 spin_lock(&lli->lli_lock);
1750 if (lli->lli_posix_acl)
1751 posix_acl_release(lli->lli_posix_acl);
1752 lli->lli_posix_acl = md->posix_acl;
1753 spin_unlock(&lli->lli_lock);
1756 inode->i_ino = cl_fid_build_ino(&body->fid1,
1757 sbi->ll_flags & LL_SBI_32BIT_API);
1758 inode->i_generation = cl_fid_build_gen(&body->fid1);
1760 if (body->valid & OBD_MD_FLATIME) {
1761 if (body->atime > LTIME_S(inode->i_atime))
1762 LTIME_S(inode->i_atime) = body->atime;
1763 lli->lli_lvb.lvb_atime = body->atime;
1765 if (body->valid & OBD_MD_FLMTIME) {
1766 if (body->mtime > LTIME_S(inode->i_mtime)) {
1767 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1768 "to "LPU64"\n", inode->i_ino,
1769 LTIME_S(inode->i_mtime), body->mtime);
1770 LTIME_S(inode->i_mtime) = body->mtime;
1772 lli->lli_lvb.lvb_mtime = body->mtime;
1774 if (body->valid & OBD_MD_FLCTIME) {
1775 if (body->ctime > LTIME_S(inode->i_ctime))
1776 LTIME_S(inode->i_ctime) = body->ctime;
1777 lli->lli_lvb.lvb_ctime = body->ctime;
1779 if (body->valid & OBD_MD_FLMODE)
1780 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1781 if (body->valid & OBD_MD_FLTYPE)
1782 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1783 LASSERT(inode->i_mode != 0);
1784 if (S_ISREG(inode->i_mode)) {
1785 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1787 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1789 if (body->valid & OBD_MD_FLUID)
1790 inode->i_uid = body->uid;
1791 if (body->valid & OBD_MD_FLGID)
1792 inode->i_gid = body->gid;
1793 if (body->valid & OBD_MD_FLFLAGS)
1794 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1795 if (body->valid & OBD_MD_FLNLINK)
1796 set_nlink(inode, body->nlink);
1797 if (body->valid & OBD_MD_FLRDEV)
1798 inode->i_rdev = old_decode_dev(body->rdev);
1800 if (body->valid & OBD_MD_FLID) {
1801 /* FID shouldn't be changed! */
1802 if (fid_is_sane(&lli->lli_fid)) {
1803 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1804 "Trying to change FID "DFID
1805 " to the "DFID", inode "DFID"(%p)\n",
1806 PFID(&lli->lli_fid), PFID(&body->fid1),
1807 PFID(ll_inode2fid(inode)), inode);
1809 lli->lli_fid = body->fid1;
1812 LASSERT(fid_seq(&lli->lli_fid) != 0);
1814 if (body->valid & OBD_MD_FLSIZE) {
1815 if (exp_connect_som(ll_i2mdexp(inode)) &&
1816 S_ISREG(inode->i_mode)) {
1817 struct lustre_handle lockh;
1820 /* As it is possible a blocking ast has been processed
1821 * by this time, we need to check there is an UPDATE
1822 * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1824 mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1825 &lockh, LDLM_FL_CBPENDING,
1829 if (lli->lli_flags & (LLIF_DONE_WRITING |
1830 LLIF_EPOCH_PENDING |
1832 CERROR("%s: inode "DFID" flags %u still"
1833 " has size authority! do not "
1834 "trust the size from MDS\n",
1835 sbi->ll_md_exp->exp_obd->obd_name,
1836 PFID(ll_inode2fid(inode)),
1839 /* Use old size assignment to avoid
1840 * deadlock bz14138 & bz14326 */
1841 i_size_write(inode, body->size);
1842 spin_lock(&lli->lli_lock);
1843 lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1844 spin_unlock(&lli->lli_lock);
1846 ldlm_lock_decref(&lockh, mode);
1849 /* Use old size assignment to avoid
1850 * deadlock bz14138 & bz14326 */
1851 i_size_write(inode, body->size);
1854 "inode="DFID", updating i_size %llu\n",
1855 PFID(ll_inode2fid(inode)),
1856 (unsigned long long)body->size);
1859 if (body->valid & OBD_MD_FLBLOCKS)
1860 inode->i_blocks = body->blocks;
1863 if (body->valid & OBD_MD_FLMDSCAPA) {
1864 LASSERT(md->mds_capa);
1865 ll_add_capa(inode, md->mds_capa);
1867 if (body->valid & OBD_MD_FLOSSCAPA) {
1868 LASSERT(md->oss_capa);
1869 ll_add_capa(inode, md->oss_capa);
1872 if (body->valid & OBD_MD_TSTATE) {
1873 if (body->t_state & MS_RESTORE)
1874 lli->lli_flags |= LLIF_FILE_RESTORING;
1878 void ll_read_inode2(struct inode *inode, void *opaque)
1880 struct lustre_md *md = opaque;
1881 struct ll_inode_info *lli = ll_i2info(inode);
1884 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1885 PFID(&lli->lli_fid), inode);
1887 LASSERT(!lli->lli_has_smd);
1889 /* Core attributes from the MDS first. This is a new inode, and
1890 * the VFS doesn't zero times in the core inode so we have to do
1891 * it ourselves. They will be overwritten by either MDS or OST
1892 * attributes - we just need to make sure they aren't newer. */
1893 LTIME_S(inode->i_mtime) = 0;
1894 LTIME_S(inode->i_atime) = 0;
1895 LTIME_S(inode->i_ctime) = 0;
1897 ll_update_inode(inode, md);
1899 /* OIDEBUG(inode); */
1901 /* initializing backing dev info. */
1902 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1905 if (S_ISREG(inode->i_mode)) {
1906 struct ll_sb_info *sbi = ll_i2sbi(inode);
1907 inode->i_op = &ll_file_inode_operations;
1908 inode->i_fop = sbi->ll_fop;
1909 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1911 } else if (S_ISDIR(inode->i_mode)) {
1912 inode->i_op = &ll_dir_inode_operations;
1913 inode->i_fop = &ll_dir_operations;
1915 } else if (S_ISLNK(inode->i_mode)) {
1916 inode->i_op = &ll_fast_symlink_inode_operations;
1919 inode->i_op = &ll_special_inode_operations;
1921 init_special_inode(inode, inode->i_mode,
1928 void ll_delete_inode(struct inode *inode)
1930 struct cl_inode_info *lli = cl_i2info(inode);
1933 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
1934 /* discard all dirty pages before truncating them, required by
1935 * osc_extent implementation at LU-1030. */
1936 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF,
1937 CL_FSYNC_DISCARD, 1);
1939 truncate_inode_pages(&inode->i_data, 0);
1941 /* Workaround for LU-118 */
1942 if (inode->i_data.nrpages) {
1943 spin_lock_irq(&inode->i_data.tree_lock);
1944 spin_unlock_irq(&inode->i_data.tree_lock);
1945 LASSERTF(inode->i_data.nrpages == 0,
1946 "inode="DFID"(%p) nrpages=%lu, see "
1947 "http://jira.whamcloud.com/browse/LU-118\n",
1948 PFID(ll_inode2fid(inode)), inode,
1949 inode->i_data.nrpages);
1951 /* Workaround end */
1953 #ifdef HAVE_SBOPS_EVICT_INODE
1954 ll_clear_inode(inode);
1961 int ll_iocontrol(struct inode *inode, struct file *file,
1962 unsigned int cmd, unsigned long arg)
1964 struct ll_sb_info *sbi = ll_i2sbi(inode);
1965 struct ptlrpc_request *req = NULL;
1970 case FSFILT_IOC_GETFLAGS: {
1971 struct mdt_body *body;
1972 struct md_op_data *op_data;
1974 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
1975 0, 0, LUSTRE_OPC_ANY,
1977 if (IS_ERR(op_data))
1978 RETURN(PTR_ERR(op_data));
1980 op_data->op_valid = OBD_MD_FLFLAGS;
1981 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
1982 ll_finish_md_op_data(op_data);
1984 CERROR("%s: failure inode "DFID": rc = %d\n",
1985 sbi->ll_md_exp->exp_obd->obd_name,
1986 PFID(ll_inode2fid(inode)), rc);
1990 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1992 flags = body->flags;
1994 ptlrpc_req_finished(req);
1996 RETURN(put_user(flags, (int *)arg));
1998 case FSFILT_IOC_SETFLAGS: {
1999 struct lov_stripe_md *lsm;
2000 struct obd_info oinfo = { { { 0 } } };
2001 struct md_op_data *op_data;
2003 if (get_user(flags, (int *)arg))
2006 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2007 LUSTRE_OPC_ANY, NULL);
2008 if (IS_ERR(op_data))
2009 RETURN(PTR_ERR(op_data));
2011 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
2012 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2013 rc = md_setattr(sbi->ll_md_exp, op_data,
2014 NULL, 0, NULL, 0, &req, NULL);
2015 ll_finish_md_op_data(op_data);
2016 ptlrpc_req_finished(req);
2020 inode->i_flags = ll_ext_to_inode_flags(flags);
2022 lsm = ccc_inode_lsm_get(inode);
2023 if (!lsm_has_objects(lsm)) {
2024 ccc_inode_lsm_put(inode, lsm);
2028 OBDO_ALLOC(oinfo.oi_oa);
2030 ccc_inode_lsm_put(inode, lsm);
2034 oinfo.oi_oa->o_oi = lsm->lsm_oi;
2035 oinfo.oi_oa->o_flags = flags;
2036 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
2038 oinfo.oi_capa = ll_mdscapa_get(inode);
2039 obdo_set_parent_fid(oinfo.oi_oa, &ll_i2info(inode)->lli_fid);
2040 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
2041 capa_put(oinfo.oi_capa);
2042 OBDO_FREE(oinfo.oi_oa);
2043 ccc_inode_lsm_put(inode, lsm);
2045 if (rc && rc != -EPERM && rc != -EACCES)
2046 CERROR("osc_setattr_async fails: rc = %d\n", rc);
2057 int ll_flush_ctx(struct inode *inode)
2059 struct ll_sb_info *sbi = ll_i2sbi(inode);
2061 CDEBUG(D_SEC, "flush context for user %d\n", current_uid());
2063 obd_set_info_async(NULL, sbi->ll_md_exp,
2064 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2066 obd_set_info_async(NULL, sbi->ll_dt_exp,
2067 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2072 /* umount -f client means force down, don't save state */
2073 void ll_umount_begin(struct super_block *sb)
2075 struct ll_sb_info *sbi = ll_s2sbi(sb);
2076 struct obd_device *obd;
2077 struct obd_ioctl_data *ioc_data;
2080 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2081 sb->s_count, atomic_read(&sb->s_active));
2083 obd = class_exp2obd(sbi->ll_md_exp);
2085 CERROR("Invalid MDC connection handle "LPX64"\n",
2086 sbi->ll_md_exp->exp_handle.h_cookie);
2092 obd = class_exp2obd(sbi->ll_dt_exp);
2094 CERROR("Invalid LOV connection handle "LPX64"\n",
2095 sbi->ll_dt_exp->exp_handle.h_cookie);
2101 OBD_ALLOC_PTR(ioc_data);
2103 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2104 sizeof *ioc_data, ioc_data, NULL);
2106 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2107 sizeof *ioc_data, ioc_data, NULL);
2109 OBD_FREE_PTR(ioc_data);
2112 /* Really, we'd like to wait until there are no requests outstanding,
2113 * and then continue. For now, we just invalidate the requests,
2114 * schedule() and sleep one second if needed, and hope.
2120 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2122 struct ll_sb_info *sbi = ll_s2sbi(sb);
2123 char *profilenm = get_profile_name(sb);
2127 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2128 read_only = *flags & MS_RDONLY;
2129 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2130 sizeof(KEY_READ_ONLY),
2131 KEY_READ_ONLY, sizeof(read_only),
2134 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2135 profilenm, read_only ?
2136 "read-only" : "read-write", err);
2141 sb->s_flags |= MS_RDONLY;
2143 sb->s_flags &= ~MS_RDONLY;
2145 if (sbi->ll_flags & LL_SBI_VERBOSE)
2146 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2147 read_only ? "read-only" : "read-write");
2152 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2153 struct super_block *sb, struct lookup_intent *it)
2155 struct ll_sb_info *sbi = NULL;
2156 struct lustre_md md;
2160 LASSERT(*inode || sb);
2161 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2162 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2163 sbi->ll_md_exp, &md);
2168 ll_update_inode(*inode, &md);
2170 LASSERT(sb != NULL);
2173 * At this point server returns to client's same fid as client
2174 * generated for creating. So using ->fid1 is okay here.
2176 LASSERT(fid_is_sane(&md.body->fid1));
2178 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1,
2179 sbi->ll_flags & LL_SBI_32BIT_API),
2181 if (*inode == NULL || IS_ERR(*inode)) {
2182 #ifdef CONFIG_FS_POSIX_ACL
2184 posix_acl_release(md.posix_acl);
2185 md.posix_acl = NULL;
2188 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2190 CERROR("new_inode -fatal: rc %d\n", rc);
2195 /* Handling piggyback layout lock.
2196 * Layout lock can be piggybacked by getattr and open request.
2197 * The lsm can be applied to inode only if it comes with a layout lock
2198 * otherwise correct layout may be overwritten, for example:
2199 * 1. proc1: mdt returns a lsm but not granting layout
2200 * 2. layout was changed by another client
2201 * 3. proc2: refresh layout and layout lock granted
2202 * 4. proc1: to apply a stale layout */
2203 if (it != NULL && it->d.lustre.it_lock_mode != 0) {
2204 struct lustre_handle lockh;
2205 struct ldlm_lock *lock;
2207 lockh.cookie = it->d.lustre.it_lock_handle;
2208 lock = ldlm_handle2lock(&lockh);
2209 LASSERT(lock != NULL);
2210 if (ldlm_has_layout(lock)) {
2211 struct cl_object_conf conf;
2213 memset(&conf, 0, sizeof(conf));
2214 conf.coc_opc = OBJECT_CONF_SET;
2215 conf.coc_inode = *inode;
2216 conf.coc_lock = lock;
2217 conf.u.coc_md = &md;
2218 (void)ll_layout_conf(*inode, &conf);
2220 LDLM_LOCK_PUT(lock);
2225 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2226 md_free_lustre_md(sbi->ll_md_exp, &md);
2230 int ll_obd_statfs(struct inode *inode, void *arg)
2232 struct ll_sb_info *sbi = NULL;
2233 struct obd_export *exp;
2235 struct obd_ioctl_data *data = NULL;
2240 if (!inode || !(sbi = ll_i2sbi(inode)))
2241 GOTO(out_statfs, rc = -EINVAL);
2243 rc = obd_ioctl_getdata(&buf, &len, arg);
2245 GOTO(out_statfs, rc);
2248 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2249 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2250 GOTO(out_statfs, rc = -EINVAL);
2252 if (data->ioc_inllen1 != sizeof(__u32) ||
2253 data->ioc_inllen2 != sizeof(__u32) ||
2254 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2255 data->ioc_plen2 != sizeof(struct obd_uuid))
2256 GOTO(out_statfs, rc = -EINVAL);
2258 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2259 if (type & LL_STATFS_LMV)
2260 exp = sbi->ll_md_exp;
2261 else if (type & LL_STATFS_LOV)
2262 exp = sbi->ll_dt_exp;
2264 GOTO(out_statfs, rc = -ENODEV);
2266 flags = (type & LL_STATFS_NODELAY) ? OBD_STATFS_NODELAY : 0;
2267 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, &flags);
2269 GOTO(out_statfs, rc);
2272 obd_ioctl_freedata(buf, len);
2276 int ll_process_config(struct lustre_cfg *lcfg)
2280 struct lprocfs_static_vars lvars;
2284 lprocfs_llite_init_vars(&lvars);
2286 /* The instance name contains the sb: lustre-client-aacfe000 */
2287 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2288 if (!ptr || !*(++ptr))
2290 if (sscanf(ptr, "%lx", &x) != 1)
2293 /* This better be a real Lustre superblock! */
2294 LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2296 /* Note we have not called client_common_fill_super yet, so
2297 proc fns must be able to handle that! */
2298 rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2305 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2306 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2307 struct inode *i1, struct inode *i2,
2308 const char *name, int namelen,
2309 int mode, __u32 opc, void *data)
2311 LASSERT(i1 != NULL);
2313 if (namelen > ll_i2sbi(i1)->ll_namelen)
2314 return ERR_PTR(-ENAMETOOLONG);
2316 if (op_data == NULL)
2317 OBD_ALLOC_PTR(op_data);
2319 if (op_data == NULL)
2320 return ERR_PTR(-ENOMEM);
2322 ll_i2gids(op_data->op_suppgids, i1, i2);
2323 op_data->op_fid1 = *ll_inode2fid(i1);
2324 op_data->op_capa1 = ll_mdscapa_get(i1);
2327 op_data->op_fid2 = *ll_inode2fid(i2);
2328 op_data->op_capa2 = ll_mdscapa_get(i2);
2330 fid_zero(&op_data->op_fid2);
2331 op_data->op_capa2 = NULL;
2334 op_data->op_name = name;
2335 op_data->op_namelen = namelen;
2336 op_data->op_mode = mode;
2337 op_data->op_mod_time = cfs_time_current_sec();
2338 op_data->op_fsuid = current_fsuid();
2339 op_data->op_fsgid = current_fsgid();
2340 op_data->op_cap = cfs_curproc_cap_pack();
2341 op_data->op_bias = 0;
2342 op_data->op_cli_flags = 0;
2343 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2344 filename_is_volatile(name, namelen, NULL))
2345 op_data->op_bias |= MDS_CREATE_VOLATILE;
2346 op_data->op_opc = opc;
2347 op_data->op_mds = 0;
2348 op_data->op_data = data;
2350 /* If the file is being opened after mknod() (normally due to NFS)
2351 * try to use the default stripe data from parent directory for
2352 * allocating OST objects. Try to pass the parent FID to MDS. */
2353 if (opc == LUSTRE_OPC_CREATE && i1 == i2 && S_ISREG(i2->i_mode) &&
2354 !ll_i2info(i2)->lli_has_smd) {
2355 struct ll_inode_info *lli = ll_i2info(i2);
2357 spin_lock(&lli->lli_lock);
2358 if (likely(!lli->lli_has_smd && !fid_is_zero(&lli->lli_pfid)))
2359 op_data->op_fid1 = lli->lli_pfid;
2360 spin_unlock(&lli->lli_lock);
2361 /** We ignore parent's capability temporary. */
2364 /* When called by ll_setattr_raw, file is i1. */
2365 if (LLIF_DATA_MODIFIED & ll_i2info(i1)->lli_flags)
2366 op_data->op_bias |= MDS_DATA_MODIFIED;
2371 void ll_finish_md_op_data(struct md_op_data *op_data)
2373 capa_put(op_data->op_capa1);
2374 capa_put(op_data->op_capa2);
2375 OBD_FREE_PTR(op_data);
2378 #ifdef HAVE_SUPEROPS_USE_DENTRY
2379 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2381 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2384 struct ll_sb_info *sbi;
2386 #ifdef HAVE_SUPEROPS_USE_DENTRY
2387 LASSERT((seq != NULL) && (dentry != NULL));
2388 sbi = ll_s2sbi(dentry->d_sb);
2390 LASSERT((seq != NULL) && (vfs != NULL));
2391 sbi = ll_s2sbi(vfs->mnt_sb);
2394 if (sbi->ll_flags & LL_SBI_NOLCK)
2395 seq_puts(seq, ",nolock");
2397 if (sbi->ll_flags & LL_SBI_FLOCK)
2398 seq_puts(seq, ",flock");
2400 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2401 seq_puts(seq, ",localflock");
2403 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2404 seq_puts(seq, ",user_xattr");
2406 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2407 seq_puts(seq, ",lazystatfs");
2409 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2410 seq_puts(seq, ",user_fid2path");
2416 * Get obd name by cmd, and copy out to user space
2418 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2420 struct ll_sb_info *sbi = ll_i2sbi(inode);
2421 struct obd_device *obd;
2424 if (cmd == OBD_IOC_GETDTNAME)
2425 obd = class_exp2obd(sbi->ll_dt_exp);
2426 else if (cmd == OBD_IOC_GETMDNAME)
2427 obd = class_exp2obd(sbi->ll_md_exp);
2434 if (copy_to_user((void *)arg, obd->obd_name,
2435 strlen(obd->obd_name) + 1))
2442 * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2443 * fsname will be returned in this buffer; otherwise, a static buffer will be
2444 * used to store the fsname and returned to caller.
2446 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2448 static char fsname_static[MTI_NAME_MAXLEN];
2449 struct lustre_sb_info *lsi = s2lsi(sb);
2454 /* this means the caller wants to use static buffer
2455 * and it doesn't care about race. Usually this is
2456 * in error reporting path */
2457 buf = fsname_static;
2458 buflen = sizeof(fsname_static);
2461 len = strlen(lsi->lsi_lmd->lmd_profile);
2462 ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2463 if (ptr && (strcmp(ptr, "-client") == 0))
2466 if (unlikely(len >= buflen))
2468 strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2474 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2481 p.mnt = current->fs->root.mnt;
2483 path = d_path(&p, buf, bufsize);
2488 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2490 char *buf, *path = NULL;
2491 struct dentry *dentry = NULL;
2492 struct ccc_object *obj = cl_inode2ccc(page->mapping->host);
2494 /* this can be called inside spin lock so use GFP_ATOMIC. */
2495 buf = (char *)__get_free_page(GFP_ATOMIC);
2497 dentry = d_find_alias(page->mapping->host);
2499 path = ll_d_path(dentry, buf, PAGE_SIZE);
2502 CWARN("%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2503 "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2504 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2505 PFID(&obj->cob_header.coh_lu.loh_fid),
2506 (path && !IS_ERR(path)) ? path : "", ioret);
2512 free_page((unsigned long)buf);