4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2013, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/llite/llite_lib.c
38 * Lustre Light Super operations
41 #define DEBUG_SUBSYSTEM S_LLITE
43 #include <linux/module.h>
44 #include <linux/types.h>
45 #include <linux/version.h>
48 #include <lustre_lite.h>
49 #include <lustre_ha.h>
50 #include <lustre_dlm.h>
51 #include <lprocfs_status.h>
52 #include <lustre_disk.h>
53 #include <lustre_param.h>
54 #include <lustre_log.h>
55 #include <cl_object.h>
56 #include <obd_cksum.h>
57 #include "llite_internal.h"
59 struct kmem_cache *ll_file_data_slab;
61 CFS_LIST_HEAD(ll_super_blocks);
62 DEFINE_SPINLOCK(ll_sb_lock);
64 #ifndef MS_HAS_NEW_AOPS
65 extern struct address_space_operations ll_aops;
67 extern struct address_space_operations_ext ll_aops;
71 #define log2(n) ffz(~(n))
74 static struct ll_sb_info *ll_init_sbi(void)
76 struct ll_sb_info *sbi = NULL;
78 unsigned long lru_page_max;
84 OBD_ALLOC(sbi, sizeof(*sbi));
88 spin_lock_init(&sbi->ll_lock);
89 mutex_init(&sbi->ll_lco.lco_lock);
90 spin_lock_init(&sbi->ll_pp_extent_lock);
91 spin_lock_init(&sbi->ll_process_lock);
92 sbi->ll_rw_stats_on = 0;
95 pages = si.totalram - si.totalhigh;
96 if (pages >> (20 - PAGE_CACHE_SHIFT) < 512) {
97 lru_page_max = pages / 2;
99 lru_page_max = (pages / 4) * 3;
102 /* initialize lru data */
103 cfs_atomic_set(&sbi->ll_cache.ccc_users, 0);
104 sbi->ll_cache.ccc_lru_max = lru_page_max;
105 cfs_atomic_set(&sbi->ll_cache.ccc_lru_left, lru_page_max);
106 spin_lock_init(&sbi->ll_cache.ccc_lru_lock);
107 CFS_INIT_LIST_HEAD(&sbi->ll_cache.ccc_lru);
109 sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
110 SBI_DEFAULT_READAHEAD_MAX);
111 sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
112 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
113 SBI_DEFAULT_READAHEAD_WHOLE_MAX;
114 CFS_INIT_LIST_HEAD(&sbi->ll_conn_chain);
115 CFS_INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
117 ll_generate_random_uuid(uuid);
118 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
119 CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
121 spin_lock(&ll_sb_lock);
122 cfs_list_add_tail(&sbi->ll_list, &ll_super_blocks);
123 spin_unlock(&ll_sb_lock);
125 sbi->ll_flags |= LL_SBI_VERBOSE;
126 #ifdef ENABLE_CHECKSUM
127 sbi->ll_flags |= LL_SBI_CHECKSUM;
130 #ifdef HAVE_LRU_RESIZE_SUPPORT
131 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
134 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
135 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
137 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
141 /* metadata statahead is enabled by default */
142 sbi->ll_sa_max = LL_SA_RPC_DEF;
143 cfs_atomic_set(&sbi->ll_sa_total, 0);
144 cfs_atomic_set(&sbi->ll_sa_wrong, 0);
145 cfs_atomic_set(&sbi->ll_agl_total, 0);
146 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
151 void ll_free_sbi(struct super_block *sb)
153 struct ll_sb_info *sbi = ll_s2sbi(sb);
157 spin_lock(&ll_sb_lock);
158 cfs_list_del(&sbi->ll_list);
159 spin_unlock(&ll_sb_lock);
160 OBD_FREE(sbi, sizeof(*sbi));
165 static struct dentry_operations ll_d_root_ops = {
166 .d_compare = ll_dcompare,
167 .d_revalidate = ll_revalidate_nd,
170 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
171 struct vfsmount *mnt)
173 struct inode *root = 0;
174 struct ll_sb_info *sbi = ll_s2sbi(sb);
175 struct obd_device *obd;
176 struct obd_capa *oc = NULL;
177 struct obd_statfs *osfs = NULL;
178 struct ptlrpc_request *request = NULL;
179 struct obd_connect_data *data = NULL;
180 struct obd_uuid *uuid;
181 struct md_op_data *op_data;
182 struct lustre_md lmd;
184 int size, err, checksum;
187 obd = class_name2obd(md);
189 CERROR("MD %s: not setup or attached\n", md);
203 if (proc_lustre_fs_root) {
204 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
207 CERROR("could not register mount in /proc/fs/lustre\n");
210 /* indicate the features supported by this client */
211 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
212 OBD_CONNECT_ATTRFID |
213 OBD_CONNECT_VERSION | OBD_CONNECT_BRW_SIZE |
214 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
215 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
216 OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 |
217 OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR |
218 OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH|
219 OBD_CONNECT_EINPROGRESS |
220 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
221 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS;
223 if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
224 data->ocd_connect_flags |= OBD_CONNECT_SOM;
226 #ifdef HAVE_LRU_RESIZE_SUPPORT
227 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
228 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
230 #ifdef CONFIG_FS_POSIX_ACL
231 data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK;
234 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
235 /* flag mdc connection as lightweight, only used for test
236 * purpose, use with care */
237 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
239 data->ocd_ibits_known = MDS_INODELOCK_FULL;
240 data->ocd_version = LUSTRE_VERSION_CODE;
242 if (sb->s_flags & MS_RDONLY)
243 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
244 if (sbi->ll_flags & LL_SBI_USER_XATTR)
245 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
247 #ifdef HAVE_MS_FLOCK_LOCK
248 /* force vfs to use lustre handler for flock() calls - bug 10743 */
249 sb->s_flags |= MS_FLOCK_LOCK;
251 #ifdef MS_HAS_NEW_AOPS
252 sb->s_flags |= MS_HAS_NEW_AOPS;
255 if (sbi->ll_flags & LL_SBI_FLOCK)
256 sbi->ll_fop = &ll_file_operations_flock;
257 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
258 sbi->ll_fop = &ll_file_operations;
260 sbi->ll_fop = &ll_file_operations_noflock;
263 data->ocd_connect_flags |= OBD_CONNECT_REAL;
264 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
265 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
267 data->ocd_brw_size = MD_MAX_BRW_SIZE;
269 err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
271 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
272 "recovery, of which this client is not a "
273 "part. Please wait for recovery to complete,"
274 " abort, or time out.\n", md);
277 CERROR("cannot connect to %s: rc = %d\n", md, err);
281 sbi->ll_md_exp->exp_connect_data = *data;
283 err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
284 LUSTRE_SEQ_METADATA);
286 CERROR("%s: Can't init metadata layer FID infrastructure, "
287 "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
291 /* For mount, we only need fs info from MDT0, and also in DNE, it
292 * can make sure the client can be mounted as long as MDT0 is
294 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
295 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
296 OBD_STATFS_FOR_MDT0);
298 GOTO(out_md_fid, err);
300 /* This needs to be after statfs to ensure connect has finished.
301 * Note that "data" does NOT contain the valid connect reply.
302 * If connecting to a 1.8 server there will be no LMV device, so
303 * we can access the MDC export directly and exp_connect_flags will
304 * be non-zero, but if accessing an upgraded 2.1 server it will
305 * have the correct flags filled in.
306 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
307 valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
308 if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
309 valid != CLIENT_CONNECT_MDT_REQD) {
312 OBD_ALLOC_WAIT(buf, PAGE_CACHE_SIZE);
313 obd_connect_flags2str(buf, PAGE_CACHE_SIZE,
314 valid ^ CLIENT_CONNECT_MDT_REQD, ",");
315 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
316 "feature(s) needed for correct operation "
317 "of this client (%s). Please upgrade "
318 "server or downgrade client.\n",
319 sbi->ll_md_exp->exp_obd->obd_name, buf);
320 OBD_FREE(buf, PAGE_CACHE_SIZE);
321 GOTO(out_md_fid, err = -EPROTO);
324 size = sizeof(*data);
325 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
326 KEY_CONN_DATA, &size, data, NULL);
328 CERROR("%s: Get connect data failed: rc = %d\n",
329 sbi->ll_md_exp->exp_obd->obd_name, err);
330 GOTO(out_md_fid, err);
333 LASSERT(osfs->os_bsize);
334 sb->s_blocksize = osfs->os_bsize;
335 sb->s_blocksize_bits = log2(osfs->os_bsize);
336 sb->s_magic = LL_SUPER_MAGIC;
337 sb->s_maxbytes = MAX_LFS_FILESIZE;
338 sbi->ll_namelen = osfs->os_namelen;
339 sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
341 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
342 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
343 LCONSOLE_INFO("Disabling user_xattr feature because "
344 "it is not supported on the server\n");
345 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
348 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
350 sb->s_flags |= MS_POSIXACL;
352 sbi->ll_flags |= LL_SBI_ACL;
354 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
356 sb->s_flags &= ~MS_POSIXACL;
358 sbi->ll_flags &= ~LL_SBI_ACL;
361 if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
362 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
363 sbi->ll_flags |= LL_SBI_RMT_CLIENT;
364 LCONSOLE_INFO("client is set as remote by default.\n");
367 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
368 sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
369 LCONSOLE_INFO("client claims to be remote, but server "
370 "rejected, forced to be local.\n");
374 if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
375 LCONSOLE_INFO("client enabled MDS capability!\n");
376 sbi->ll_flags |= LL_SBI_MDS_CAPA;
379 if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
380 LCONSOLE_INFO("client enabled OSS capability!\n");
381 sbi->ll_flags |= LL_SBI_OSS_CAPA;
384 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
385 sbi->ll_flags |= LL_SBI_64BIT_HASH;
387 if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
388 sbi->ll_md_brw_size = data->ocd_brw_size;
390 sbi->ll_md_brw_size = PAGE_CACHE_SIZE;
392 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK) {
393 LCONSOLE_INFO("Layout lock feature supported.\n");
394 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
397 obd = class_name2obd(dt);
399 CERROR("DT %s: not setup or attached\n", dt);
400 GOTO(out_md_fid, err = -ENODEV);
403 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
404 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
405 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
406 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
407 OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
408 OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
409 OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
410 OBD_CONNECT_MAXBYTES |
411 OBD_CONNECT_EINPROGRESS |
412 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
413 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS;
415 if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
416 data->ocd_connect_flags |= OBD_CONNECT_SOM;
418 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
419 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
420 * disabled by default, because it can still be enabled on the
421 * fly via /proc. As a consequence, we still need to come to an
422 * agreement on the supported algorithms at connect time */
423 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
425 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
426 data->ocd_cksum_types = OBD_CKSUM_ADLER;
428 data->ocd_cksum_types = cksum_types_supported_client();
431 #ifdef HAVE_LRU_RESIZE_SUPPORT
432 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
434 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
435 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
437 CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
438 "ocd_grant: %d\n", data->ocd_connect_flags,
439 data->ocd_version, data->ocd_grant);
441 obd->obd_upcall.onu_owner = &sbi->ll_lco;
442 obd->obd_upcall.onu_upcall = cl_ocd_update;
444 data->ocd_brw_size = DT_MAX_BRW_SIZE;
446 err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
449 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
450 "recovery, of which this client is not a "
451 "part. Please wait for recovery to "
452 "complete, abort, or time out.\n", dt);
455 CERROR("%s: Cannot connect to %s: rc = %d\n",
456 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
460 sbi->ll_dt_exp->exp_connect_data = *data;
462 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
463 LUSTRE_SEQ_METADATA);
465 CERROR("%s: Can't init data layer FID infrastructure, "
466 "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
470 mutex_lock(&sbi->ll_lco.lco_lock);
471 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
472 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
473 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
474 mutex_unlock(&sbi->ll_lco.lco_lock);
476 fid_zero(&sbi->ll_root_fid);
477 err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
479 CERROR("cannot mds_connect: rc = %d\n", err);
480 GOTO(out_lock_cn_cb, err);
482 if (!fid_is_sane(&sbi->ll_root_fid)) {
483 CERROR("%s: Invalid root fid "DFID" during mount\n",
484 sbi->ll_md_exp->exp_obd->obd_name,
485 PFID(&sbi->ll_root_fid));
486 GOTO(out_lock_cn_cb, err = -EINVAL);
488 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
490 sb->s_op = &lustre_super_operations;
491 #if THREAD_SIZE >= 8192 /*b=17630*/
492 sb->s_export_op = &lustre_export_operations;
496 * XXX: move this to after cbd setup? */
497 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
498 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
499 valid |= OBD_MD_FLRMTPERM;
500 else if (sbi->ll_flags & LL_SBI_ACL)
501 valid |= OBD_MD_FLACL;
503 OBD_ALLOC_PTR(op_data);
505 GOTO(out_lock_cn_cb, err = -ENOMEM);
507 op_data->op_fid1 = sbi->ll_root_fid;
508 op_data->op_mode = 0;
509 op_data->op_capa1 = oc;
510 op_data->op_valid = valid;
512 err = md_getattr(sbi->ll_md_exp, op_data, &request);
515 OBD_FREE_PTR(op_data);
517 CERROR("%s: md_getattr failed for root: rc = %d\n",
518 sbi->ll_md_exp->exp_obd->obd_name, err);
519 GOTO(out_lock_cn_cb, err);
522 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
523 sbi->ll_md_exp, &lmd);
525 CERROR("failed to understand root inode md: rc = %d\n", err);
526 ptlrpc_req_finished(request);
527 GOTO(out_lock_cn_cb, err);
530 LASSERT(fid_is_sane(&sbi->ll_root_fid));
531 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
532 sbi->ll_flags & LL_SBI_32BIT_API),
534 md_free_lustre_md(sbi->ll_md_exp, &lmd);
535 ptlrpc_req_finished(request);
537 if (root == NULL || IS_ERR(root)) {
539 obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
540 #ifdef CONFIG_FS_POSIX_ACL
542 posix_acl_release(lmd.posix_acl);
543 lmd.posix_acl = NULL;
546 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
548 CERROR("lustre_lite: bad iget4 for root\n");
552 err = ll_close_thread_start(&sbi->ll_lcq);
554 CERROR("cannot start close thread: rc %d\n", err);
558 #ifdef CONFIG_FS_POSIX_ACL
559 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
560 rct_init(&sbi->ll_rct);
561 et_init(&sbi->ll_et);
565 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
566 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
567 KEY_CHECKSUM, sizeof(checksum), &checksum,
571 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
572 KEY_CACHE_SET, sizeof(sbi->ll_cache),
573 &sbi->ll_cache, NULL);
575 sb->s_root = d_make_root(root);
576 if (sb->s_root == NULL) {
577 CERROR("%s: can't make root dentry\n",
578 ll_get_fsname(sb, NULL, 0));
579 GOTO(out_root, err = -ENOMEM);
582 #ifdef HAVE_DCACHE_LOCK
583 sb->s_root->d_op = &ll_d_root_ops;
585 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
586 d_set_d_op(sb->s_root, &ll_d_root_ops);
587 sb->s_d_op = &ll_d_ops;
590 sbi->ll_sdev_orig = sb->s_dev;
592 /* We set sb->s_dev equal on all lustre clients in order to support
593 * NFS export clustering. NFSD requires that the FSID be the same
595 /* s_dev is also used in lt_compare() to compare two fs, but that is
596 * only a node-local comparison. */
597 uuid = obd_get_uuid(sbi->ll_md_exp);
599 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
600 get_uuid2fsid(uuid->uuid, strlen(uuid->uuid), &sbi->ll_fsid);
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;
629 lprocfs_unregister_mountpoint(sbi);
633 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
637 *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
639 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
640 KEY_MAX_EASIZE, &size, lmmsize, NULL);
642 CERROR("Get max mdsize error rc %d \n", rc);
647 void ll_dump_inode(struct inode *inode)
649 struct ll_d_hlist_node *tmp;
650 int dentry_count = 0;
652 LASSERT(inode != NULL);
654 ll_d_hlist_for_each(tmp, &inode->i_dentry)
657 CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
658 inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
659 inode->i_mode, atomic_read(&inode->i_count), dentry_count);
662 void lustre_dump_dentry(struct dentry *dentry, int recur)
664 struct list_head *tmp;
667 LASSERT(dentry != NULL);
669 list_for_each(tmp, &dentry->d_subdirs)
672 CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
673 " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
674 dentry->d_name.len, dentry->d_name.name,
675 dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
676 dentry->d_parent, dentry->d_inode, d_refcount(dentry),
677 dentry->d_flags, dentry->d_fsdata, subdirs);
678 if (dentry->d_inode != NULL)
679 ll_dump_inode(dentry->d_inode);
684 list_for_each(tmp, &dentry->d_subdirs) {
685 struct dentry *d = list_entry(tmp, struct dentry, d_u.d_child);
686 lustre_dump_dentry(d, recur - 1);
690 void client_common_put_super(struct super_block *sb)
692 struct ll_sb_info *sbi = ll_s2sbi(sb);
695 #ifdef CONFIG_FS_POSIX_ACL
696 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
697 et_fini(&sbi->ll_et);
698 rct_fini(&sbi->ll_rct);
702 ll_close_thread_shutdown(sbi->ll_lcq);
706 cfs_list_del(&sbi->ll_conn_chain);
708 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
709 obd_disconnect(sbi->ll_dt_exp);
710 sbi->ll_dt_exp = NULL;
711 /* wait till all OSCs are gone, since cl_cache is accessing sbi.
713 obd_zombie_barrier();
715 lprocfs_unregister_mountpoint(sbi);
717 obd_fid_fini(sbi->ll_md_exp->exp_obd);
718 obd_disconnect(sbi->ll_md_exp);
719 sbi->ll_md_exp = NULL;
724 void ll_kill_super(struct super_block *sb)
726 struct ll_sb_info *sbi;
731 if (!(sb->s_flags & MS_ACTIVE))
735 /* we need restore s_dev from changed for clustred NFS before put_super
736 * because new kernels have cached s_dev and change sb->s_dev in
737 * put_super not affected real removing devices */
739 sb->s_dev = sbi->ll_sdev_orig;
740 sbi->ll_umounting = 1;
745 char *ll_read_opt(const char *opt, char *data)
751 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
752 if (strncmp(opt, data, strlen(opt)))
754 if ((value = strchr(data, '=')) == NULL)
758 OBD_ALLOC(retval, strlen(value) + 1);
760 CERROR("out of memory!\n");
764 memcpy(retval, value, strlen(value)+1);
765 CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
769 static inline int ll_set_opt(const char *opt, char *data, int fl)
771 if (strncmp(opt, data, strlen(opt)) != 0)
777 /* non-client-specific mount options are parsed in lmd_parse */
778 static int ll_options(char *options, int *flags)
781 char *s1 = options, *s2;
787 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
790 CDEBUG(D_SUPER, "next opt=%s\n", s1);
791 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
796 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
801 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
806 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
811 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
816 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
821 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 5, 50, 0)
822 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
824 /* Ignore deprecated mount option. The client will
825 * always try to mount with ACL support, whether this
826 * is used depends on whether server supports it. */
827 LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
828 "mount option 'acl'.\n");
831 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
833 LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
834 "mount option 'noacl'.\n");
838 #warning "{no}acl options have been deprecated since 1.8, please remove them"
840 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
845 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
850 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
856 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
861 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
866 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
871 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
876 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
881 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
886 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
891 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
896 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
901 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
906 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
912 s2 = strchr(s1, ',');
920 void ll_lli_init(struct ll_inode_info *lli)
922 lli->lli_inode_magic = LLI_INODE_MAGIC;
924 lli->lli_ioepoch = 0;
925 lli->lli_maxbytes = MAX_LFS_FILESIZE;
926 spin_lock_init(&lli->lli_lock);
927 lli->lli_posix_acl = NULL;
928 lli->lli_remote_perms = NULL;
929 mutex_init(&lli->lli_rmtperm_mutex);
930 /* Do not set lli_fid, it has been initialized already. */
931 fid_zero(&lli->lli_pfid);
932 CFS_INIT_LIST_HEAD(&lli->lli_close_list);
933 CFS_INIT_LIST_HEAD(&lli->lli_oss_capas);
934 cfs_atomic_set(&lli->lli_open_count, 0);
935 lli->lli_mds_capa = NULL;
936 lli->lli_rmtperm_time = 0;
937 lli->lli_pending_och = NULL;
938 lli->lli_mds_read_och = NULL;
939 lli->lli_mds_write_och = NULL;
940 lli->lli_mds_exec_och = NULL;
941 lli->lli_open_fd_read_count = 0;
942 lli->lli_open_fd_write_count = 0;
943 lli->lli_open_fd_exec_count = 0;
944 mutex_init(&lli->lli_och_mutex);
945 spin_lock_init(&lli->lli_agl_lock);
946 lli->lli_has_smd = false;
947 lli->lli_layout_gen = LL_LAYOUT_GEN_NONE;
948 lli->lli_clob = NULL;
950 LASSERT(lli->lli_vfs_inode.i_mode != 0);
951 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
952 mutex_init(&lli->lli_readdir_mutex);
953 lli->lli_opendir_key = NULL;
955 lli->lli_def_acl = NULL;
956 spin_lock_init(&lli->lli_sa_lock);
957 lli->lli_opendir_pid = 0;
959 sema_init(&lli->lli_size_sem, 1);
960 lli->lli_size_sem_owner = NULL;
961 lli->lli_symlink_name = NULL;
962 init_rwsem(&lli->lli_trunc_sem);
963 mutex_init(&lli->lli_write_mutex);
964 init_rwsem(&lli->lli_glimpse_sem);
965 lli->lli_glimpse_time = 0;
966 CFS_INIT_LIST_HEAD(&lli->lli_agl_list);
967 lli->lli_agl_index = 0;
968 lli->lli_async_rc = 0;
969 lli->lli_volatile = false;
971 mutex_init(&lli->lli_layout_mutex);
974 static inline int ll_bdi_register(struct backing_dev_info *bdi)
976 static atomic_t ll_bdi_num = ATOMIC_INIT(0);
978 bdi->name = "lustre";
979 return bdi_register(bdi, NULL, "lustre-%d",
980 atomic_inc_return(&ll_bdi_num));
983 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
985 struct lustre_profile *lprof = NULL;
986 struct lustre_sb_info *lsi = s2lsi(sb);
987 struct ll_sb_info *sbi;
988 char *dt = NULL, *md = NULL;
989 char *profilenm = get_profile_name(sb);
990 struct config_llog_instance *cfg;
991 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
992 const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
996 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1004 /* client additional sb info */
1005 lsi->lsi_llsbi = sbi = ll_init_sbi();
1007 cfs_module_put(THIS_MODULE);
1012 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1014 GOTO(out_free, err);
1016 err = bdi_init(&lsi->lsi_bdi);
1018 GOTO(out_free, err);
1019 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1020 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1021 err = ll_bdi_register(&lsi->lsi_bdi);
1023 GOTO(out_free, err);
1025 sb->s_bdi = &lsi->lsi_bdi;
1027 /* Generate a string unique to this super, in case some joker tries
1028 to mount the same fs at two mount points.
1029 Use the address of the super itself.*/
1030 cfg->cfg_instance = sb;
1031 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1032 cfg->cfg_callback = class_config_llog_handler;
1033 /* set up client obds */
1034 err = lustre_process_log(sb, profilenm, cfg);
1036 CERROR("Unable to process log: %d\n", err);
1037 GOTO(out_free, err);
1040 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1041 lprof = class_get_profile(profilenm);
1042 if (lprof == NULL) {
1043 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1044 " read from the MGS. Does that filesystem "
1045 "exist?\n", profilenm);
1046 GOTO(out_free, err = -EINVAL);
1048 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1049 lprof->lp_md, lprof->lp_dt);
1051 OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1053 GOTO(out_free, err = -ENOMEM);
1054 sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1056 OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1058 GOTO(out_free, err = -ENOMEM);
1059 sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1061 /* connections, registrations, sb setup */
1062 err = client_common_fill_super(sb, md, dt, mnt);
1066 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1068 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1071 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1072 LCONSOLE_WARN("Mounted %s\n", profilenm);
1076 } /* ll_fill_super */
1079 void lu_context_keys_dump(void);
1081 void ll_put_super(struct super_block *sb)
1083 struct config_llog_instance cfg;
1084 struct obd_device *obd;
1085 struct lustre_sb_info *lsi = s2lsi(sb);
1086 struct ll_sb_info *sbi = ll_s2sbi(sb);
1087 char *profilenm = get_profile_name(sb);
1088 int force = 1, next;
1091 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1093 ll_print_capa_stat(sbi);
1095 cfg.cfg_instance = sb;
1096 lustre_end_log(sb, profilenm, &cfg);
1098 if (sbi->ll_md_exp) {
1099 obd = class_exp2obd(sbi->ll_md_exp);
1101 force = obd->obd_force;
1104 /* We need to set force before the lov_disconnect in
1105 lustre_common_put_super, since l_d cleans up osc's as well. */
1108 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1110 obd->obd_force = force;
1115 /* Only if client_common_fill_super succeeded */
1116 client_common_put_super(sb);
1120 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1121 class_manual_cleanup(obd);
1124 if (sbi->ll_flags & LL_SBI_VERBOSE)
1125 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1128 class_del_profile(profilenm);
1130 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1131 bdi_destroy(&lsi->lsi_bdi);
1132 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1136 lsi->lsi_llsbi = NULL;
1138 lustre_common_put_super(sb);
1140 cfs_module_put(THIS_MODULE);
1143 } /* client_put_super */
1145 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1147 struct inode *inode = NULL;
1149 /* NOTE: we depend on atomic igrab() -bzzz */
1150 lock_res_and_lock(lock);
1151 if (lock->l_resource->lr_lvb_inode) {
1152 struct ll_inode_info * lli;
1153 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1154 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1155 inode = igrab(lock->l_resource->lr_lvb_inode);
1157 inode = lock->l_resource->lr_lvb_inode;
1158 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1159 D_WARNING, lock, "lr_lvb_inode %p is "
1160 "bogus: magic %08x",
1161 lock->l_resource->lr_lvb_inode,
1162 lli->lli_inode_magic);
1166 unlock_res_and_lock(lock);
1170 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1172 struct inode *inode = NULL;
1173 /* NOTE: we depend on atomic igrab() -bzzz */
1174 lock_res_and_lock(lock);
1175 if (lock->l_ast_data) {
1176 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1177 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1178 inode = igrab(lock->l_ast_data);
1180 inode = lock->l_ast_data;
1181 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1182 D_WARNING, lock, "l_ast_data %p is "
1183 "bogus: magic %08x", lock->l_ast_data,
1184 lli->lli_inode_magic);
1188 unlock_res_and_lock(lock);
1192 void ll_clear_inode(struct inode *inode)
1194 struct ll_inode_info *lli = ll_i2info(inode);
1195 struct ll_sb_info *sbi = ll_i2sbi(inode);
1198 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1199 inode->i_generation, inode);
1201 if (S_ISDIR(inode->i_mode)) {
1202 /* these should have been cleared in ll_file_release */
1203 LASSERT(lli->lli_opendir_key == NULL);
1204 LASSERT(lli->lli_sai == NULL);
1205 LASSERT(lli->lli_opendir_pid == 0);
1208 spin_lock(&lli->lli_lock);
1209 ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1210 spin_unlock(&lli->lli_lock);
1211 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1213 LASSERT(!lli->lli_open_fd_write_count);
1214 LASSERT(!lli->lli_open_fd_read_count);
1215 LASSERT(!lli->lli_open_fd_exec_count);
1217 if (lli->lli_mds_write_och)
1218 ll_md_real_close(inode, FMODE_WRITE);
1219 if (lli->lli_mds_exec_och)
1220 ll_md_real_close(inode, FMODE_EXEC);
1221 if (lli->lli_mds_read_och)
1222 ll_md_real_close(inode, FMODE_READ);
1224 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1225 OBD_FREE(lli->lli_symlink_name,
1226 strlen(lli->lli_symlink_name) + 1);
1227 lli->lli_symlink_name = NULL;
1230 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1231 LASSERT(lli->lli_posix_acl == NULL);
1232 if (lli->lli_remote_perms) {
1233 free_rmtperm_hash(lli->lli_remote_perms);
1234 lli->lli_remote_perms = NULL;
1237 #ifdef CONFIG_FS_POSIX_ACL
1238 else if (lli->lli_posix_acl) {
1239 LASSERT(cfs_atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1240 LASSERT(lli->lli_remote_perms == NULL);
1241 posix_acl_release(lli->lli_posix_acl);
1242 lli->lli_posix_acl = NULL;
1245 lli->lli_inode_magic = LLI_INODE_DEAD;
1247 ll_clear_inode_capas(inode);
1248 if (!S_ISDIR(inode->i_mode))
1249 LASSERT(cfs_list_empty(&lli->lli_agl_list));
1252 * XXX This has to be done before lsm is freed below, because
1253 * cl_object still uses inode lsm.
1255 cl_inode_fini(inode);
1256 lli->lli_has_smd = false;
1261 int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data,
1262 struct md_open_data **mod)
1264 struct lustre_md md;
1265 struct inode *inode = dentry->d_inode;
1266 struct ll_sb_info *sbi = ll_i2sbi(inode);
1267 struct ptlrpc_request *request = NULL;
1271 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1272 LUSTRE_OPC_ANY, NULL);
1273 if (IS_ERR(op_data))
1274 RETURN(PTR_ERR(op_data));
1276 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1279 ptlrpc_req_finished(request);
1280 if (rc == -ENOENT) {
1282 /* Unlinked special device node? Or just a race?
1283 * Pretend we done everything. */
1284 if (!S_ISREG(inode->i_mode) &&
1285 !S_ISDIR(inode->i_mode)) {
1286 ia_valid = op_data->op_attr.ia_valid;
1287 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1288 rc = simple_setattr(dentry, &op_data->op_attr);
1289 op_data->op_attr.ia_valid = ia_valid;
1291 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1292 CERROR("md_setattr fails: rc = %d\n", rc);
1297 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1298 sbi->ll_md_exp, &md);
1300 ptlrpc_req_finished(request);
1304 ia_valid = op_data->op_attr.ia_valid;
1305 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1306 * cache is not cleared yet. */
1307 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1308 rc = simple_setattr(dentry, &op_data->op_attr);
1309 op_data->op_attr.ia_valid = ia_valid;
1311 /* Extract epoch data if obtained. */
1312 op_data->op_handle = md.body->handle;
1313 op_data->op_ioepoch = md.body->ioepoch;
1315 ll_update_inode(inode, &md);
1316 ptlrpc_req_finished(request);
1321 /* Close IO epoch and send Size-on-MDS attribute update. */
1322 static int ll_setattr_done_writing(struct inode *inode,
1323 struct md_op_data *op_data,
1324 struct md_open_data *mod)
1326 struct ll_inode_info *lli = ll_i2info(inode);
1330 LASSERT(op_data != NULL);
1331 if (!S_ISREG(inode->i_mode))
1334 CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1335 op_data->op_ioepoch, PFID(&lli->lli_fid));
1337 op_data->op_flags = MF_EPOCH_CLOSE;
1338 ll_done_writing_attr(inode, op_data);
1339 ll_pack_inode2opdata(inode, op_data, NULL);
1341 rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1342 if (rc == -EAGAIN) {
1343 /* MDS has instructed us to obtain Size-on-MDS attribute
1344 * from OSTs and send setattr to back to MDS. */
1345 rc = ll_som_update(inode, op_data);
1347 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1353 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1355 struct obd_capa *capa;
1358 if (attr->ia_valid & ATTR_SIZE)
1359 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1361 capa = ll_mdscapa_get(inode);
1363 rc = cl_setattr_ost(inode, attr, capa);
1365 if (attr->ia_valid & ATTR_SIZE)
1366 ll_truncate_free_capa(capa);
1373 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1374 * object(s) determine the file size and mtime. Otherwise, the MDS will
1375 * keep these values until such a time that objects are allocated for it.
1376 * We do the MDS operations first, as it is checking permissions for us.
1377 * We don't to the MDS RPC if there is nothing that we want to store there,
1378 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1379 * going to do an RPC anyways.
1381 * If we are doing a truncate, we will send the mtime and ctime updates
1382 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1383 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1386 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr)
1388 struct inode *inode = dentry->d_inode;
1389 struct ll_inode_info *lli = ll_i2info(inode);
1390 struct md_op_data *op_data = NULL;
1391 struct md_open_data *mod = NULL;
1392 int rc = 0, rc1 = 0;
1395 CDEBUG(D_VFSTRACE, "%s: setattr inode %p/fid:"DFID" from %llu to %llu, "
1396 "valid %x\n", ll_get_fsname(inode->i_sb, NULL, 0), inode,
1397 PFID(&lli->lli_fid), i_size_read(inode), attr->ia_size,
1400 if (attr->ia_valid & ATTR_SIZE) {
1401 /* Check new size against VFS/VM file size limit and rlimit */
1402 rc = inode_newsize_ok(inode, attr->ia_size);
1406 /* The maximum Lustre file size is variable, based on the
1407 * OST maximum object size and number of stripes. This
1408 * needs another check in addition to the VFS check above. */
1409 if (attr->ia_size > ll_file_maxbytes(inode)) {
1410 CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1411 PFID(&lli->lli_fid), attr->ia_size,
1412 ll_file_maxbytes(inode));
1416 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1419 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1420 if (attr->ia_valid & TIMES_SET_FLAGS) {
1421 if (cfs_curproc_fsuid() != inode->i_uid &&
1422 !cfs_capable(CFS_CAP_FOWNER))
1426 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1427 if (attr->ia_valid & ATTR_CTIME) {
1428 attr->ia_ctime = CFS_CURRENT_TIME;
1429 attr->ia_valid |= ATTR_CTIME_SET;
1431 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1432 (attr->ia_valid & ATTR_ATIME)) {
1433 attr->ia_atime = CFS_CURRENT_TIME;
1434 attr->ia_valid |= ATTR_ATIME_SET;
1436 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1437 (attr->ia_valid & ATTR_MTIME)) {
1438 attr->ia_mtime = CFS_CURRENT_TIME;
1439 attr->ia_valid |= ATTR_MTIME_SET;
1442 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1443 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1444 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1445 cfs_time_current_sec());
1447 /* If we are changing file size, file content is modified, flag it. */
1448 if (attr->ia_valid & ATTR_SIZE) {
1449 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1450 spin_lock(&lli->lli_lock);
1451 lli->lli_flags |= LLIF_DATA_MODIFIED;
1452 spin_unlock(&lli->lli_lock);
1455 /* We always do an MDS RPC, even if we're only changing the size;
1456 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1458 OBD_ALLOC_PTR(op_data);
1459 if (op_data == NULL)
1462 if (!S_ISDIR(inode->i_mode)) {
1463 if (attr->ia_valid & ATTR_SIZE)
1464 inode_dio_write_done(inode);
1465 mutex_unlock(&inode->i_mutex);
1466 down_write(&lli->lli_trunc_sem);
1469 memcpy(&op_data->op_attr, attr, sizeof(*attr));
1471 /* Open epoch for truncate. */
1472 if (exp_connect_som(ll_i2mdexp(inode)) &&
1473 (attr->ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1474 op_data->op_flags = MF_EPOCH_OPEN;
1476 rc = ll_md_setattr(dentry, op_data, &mod);
1480 /* RPC to MDT is sent, cancel data modification flag */
1481 if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) {
1482 spin_lock(&lli->lli_lock);
1483 lli->lli_flags &= ~LLIF_DATA_MODIFIED;
1484 spin_unlock(&lli->lli_lock);
1487 ll_ioepoch_open(lli, op_data->op_ioepoch);
1488 if (!S_ISREG(inode->i_mode))
1491 if (attr->ia_valid & (ATTR_SIZE |
1492 ATTR_ATIME | ATTR_ATIME_SET |
1493 ATTR_MTIME | ATTR_MTIME_SET))
1494 /* For truncate and utimes sending attributes to OSTs, setting
1495 * mtime/atime to the past will be performed under PW [0:EOF]
1496 * extent lock (new_size:EOF for truncate). It may seem
1497 * excessive to send mtime/atime updates to OSTs when not
1498 * setting times to past, but it is necessary due to possible
1499 * time de-synchronization between MDT inode and OST objects */
1500 rc = ll_setattr_ost(inode, attr);
1504 if (op_data->op_ioepoch) {
1505 rc1 = ll_setattr_done_writing(inode, op_data, mod);
1509 ll_finish_md_op_data(op_data);
1511 if (!S_ISDIR(inode->i_mode)) {
1512 up_write(&lli->lli_trunc_sem);
1513 mutex_lock(&inode->i_mutex);
1514 if (attr->ia_valid & ATTR_SIZE)
1515 inode_dio_wait(inode);
1518 ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1519 LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1524 int ll_setattr(struct dentry *de, struct iattr *attr)
1526 int mode = de->d_inode->i_mode;
1528 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1529 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1530 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1532 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1533 (ATTR_SIZE|ATTR_MODE)) &&
1534 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1535 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1536 !(attr->ia_mode & S_ISGID))))
1537 attr->ia_valid |= ATTR_FORCE;
1539 if ((mode & S_ISUID) &&
1540 !(attr->ia_mode & S_ISUID) &&
1541 !(attr->ia_valid & ATTR_KILL_SUID))
1542 attr->ia_valid |= ATTR_KILL_SUID;
1544 if (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1545 !(attr->ia_mode & S_ISGID) &&
1546 !(attr->ia_valid & ATTR_KILL_SGID))
1547 attr->ia_valid |= ATTR_KILL_SGID;
1549 return ll_setattr_raw(de, attr);
1552 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1553 __u64 max_age, __u32 flags)
1555 struct ll_sb_info *sbi = ll_s2sbi(sb);
1556 struct obd_statfs obd_osfs;
1560 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1562 CERROR("md_statfs fails: rc = %d\n", rc);
1566 osfs->os_type = sb->s_magic;
1568 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1569 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1571 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1572 flags |= OBD_STATFS_NODELAY;
1574 rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1576 CERROR("obd_statfs fails: rc = %d\n", rc);
1580 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1581 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1584 osfs->os_bsize = obd_osfs.os_bsize;
1585 osfs->os_blocks = obd_osfs.os_blocks;
1586 osfs->os_bfree = obd_osfs.os_bfree;
1587 osfs->os_bavail = obd_osfs.os_bavail;
1589 /* If we don't have as many objects free on the OST as inodes
1590 * on the MDS, we reduce the total number of inodes to
1591 * compensate, so that the "inodes in use" number is correct.
1593 if (obd_osfs.os_ffree < osfs->os_ffree) {
1594 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1596 osfs->os_ffree = obd_osfs.os_ffree;
1601 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1603 struct super_block *sb = de->d_sb;
1604 struct obd_statfs osfs;
1607 CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1608 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1610 /* Some amount of caching on the client is allowed */
1611 rc = ll_statfs_internal(sb, &osfs,
1612 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1617 statfs_unpack(sfs, &osfs);
1619 /* We need to downshift for all 32-bit kernels, because we can't
1620 * tell if the kernel is being called via sys_statfs64() or not.
1621 * Stop before overflowing f_bsize - in which case it is better
1622 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1623 if (sizeof(long) < 8) {
1624 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1627 osfs.os_blocks >>= 1;
1628 osfs.os_bfree >>= 1;
1629 osfs.os_bavail >>= 1;
1633 sfs->f_blocks = osfs.os_blocks;
1634 sfs->f_bfree = osfs.os_bfree;
1635 sfs->f_bavail = osfs.os_bavail;
1636 sfs->f_fsid = ll_s2sbi(sb)->ll_fsid;
1640 void ll_inode_size_lock(struct inode *inode)
1642 struct ll_inode_info *lli;
1644 LASSERT(!S_ISDIR(inode->i_mode));
1646 lli = ll_i2info(inode);
1647 LASSERT(lli->lli_size_sem_owner != current);
1648 down(&lli->lli_size_sem);
1649 LASSERT(lli->lli_size_sem_owner == NULL);
1650 lli->lli_size_sem_owner = current;
1653 void ll_inode_size_unlock(struct inode *inode)
1655 struct ll_inode_info *lli;
1657 lli = ll_i2info(inode);
1658 LASSERT(lli->lli_size_sem_owner == current);
1659 lli->lli_size_sem_owner = NULL;
1660 up(&lli->lli_size_sem);
1663 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1665 struct ll_inode_info *lli = ll_i2info(inode);
1666 struct mdt_body *body = md->body;
1667 struct lov_stripe_md *lsm = md->lsm;
1668 struct ll_sb_info *sbi = ll_i2sbi(inode);
1670 LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1672 if (!lli->lli_has_smd &&
1673 !(sbi->ll_flags & LL_SBI_LAYOUT_LOCK))
1674 cl_file_inode_init(inode, md);
1676 lli->lli_maxbytes = lsm->lsm_maxbytes;
1677 if (lli->lli_maxbytes > MAX_LFS_FILESIZE)
1678 lli->lli_maxbytes = MAX_LFS_FILESIZE;
1681 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1682 if (body->valid & OBD_MD_FLRMTPERM)
1683 ll_update_remote_perm(inode, md->remote_perm);
1685 #ifdef CONFIG_FS_POSIX_ACL
1686 else if (body->valid & OBD_MD_FLACL) {
1687 spin_lock(&lli->lli_lock);
1688 if (lli->lli_posix_acl)
1689 posix_acl_release(lli->lli_posix_acl);
1690 lli->lli_posix_acl = md->posix_acl;
1691 spin_unlock(&lli->lli_lock);
1694 inode->i_ino = cl_fid_build_ino(&body->fid1,
1695 sbi->ll_flags & LL_SBI_32BIT_API);
1696 inode->i_generation = cl_fid_build_gen(&body->fid1);
1698 if (body->valid & OBD_MD_FLATIME) {
1699 if (body->atime > LTIME_S(inode->i_atime))
1700 LTIME_S(inode->i_atime) = body->atime;
1701 lli->lli_lvb.lvb_atime = body->atime;
1703 if (body->valid & OBD_MD_FLMTIME) {
1704 if (body->mtime > LTIME_S(inode->i_mtime)) {
1705 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1706 "to "LPU64"\n", inode->i_ino,
1707 LTIME_S(inode->i_mtime), body->mtime);
1708 LTIME_S(inode->i_mtime) = body->mtime;
1710 lli->lli_lvb.lvb_mtime = body->mtime;
1712 if (body->valid & OBD_MD_FLCTIME) {
1713 if (body->ctime > LTIME_S(inode->i_ctime))
1714 LTIME_S(inode->i_ctime) = body->ctime;
1715 lli->lli_lvb.lvb_ctime = body->ctime;
1717 if (body->valid & OBD_MD_FLMODE)
1718 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1719 if (body->valid & OBD_MD_FLTYPE)
1720 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1721 LASSERT(inode->i_mode != 0);
1722 if (S_ISREG(inode->i_mode)) {
1723 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1725 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1727 if (body->valid & OBD_MD_FLUID)
1728 inode->i_uid = body->uid;
1729 if (body->valid & OBD_MD_FLGID)
1730 inode->i_gid = body->gid;
1731 if (body->valid & OBD_MD_FLFLAGS)
1732 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1733 if (body->valid & OBD_MD_FLNLINK)
1734 set_nlink(inode, body->nlink);
1735 if (body->valid & OBD_MD_FLRDEV)
1736 inode->i_rdev = old_decode_dev(body->rdev);
1738 if (body->valid & OBD_MD_FLID) {
1739 /* FID shouldn't be changed! */
1740 if (fid_is_sane(&lli->lli_fid)) {
1741 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1742 "Trying to change FID "DFID
1743 " to the "DFID", inode %lu/%u(%p)\n",
1744 PFID(&lli->lli_fid), PFID(&body->fid1),
1745 inode->i_ino, inode->i_generation, inode);
1747 lli->lli_fid = body->fid1;
1750 LASSERT(fid_seq(&lli->lli_fid) != 0);
1752 if (body->valid & OBD_MD_FLSIZE) {
1753 if (exp_connect_som(ll_i2mdexp(inode)) &&
1754 S_ISREG(inode->i_mode)) {
1755 struct lustre_handle lockh;
1758 /* As it is possible a blocking ast has been processed
1759 * by this time, we need to check there is an UPDATE
1760 * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1762 mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1763 &lockh, LDLM_FL_CBPENDING);
1765 if (lli->lli_flags & (LLIF_DONE_WRITING |
1766 LLIF_EPOCH_PENDING |
1768 CERROR("ino %lu flags %u still has "
1769 "size authority! do not trust "
1770 "the size got from MDS\n",
1771 inode->i_ino, lli->lli_flags);
1773 /* Use old size assignment to avoid
1774 * deadlock bz14138 & bz14326 */
1775 i_size_write(inode, body->size);
1776 spin_lock(&lli->lli_lock);
1777 lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1778 spin_unlock(&lli->lli_lock);
1780 ldlm_lock_decref(&lockh, mode);
1783 /* Use old size assignment to avoid
1784 * deadlock bz14138 & bz14326 */
1785 i_size_write(inode, body->size);
1787 CDEBUG(D_VFSTRACE, "inode=%lu, updating i_size %llu\n",
1788 inode->i_ino, (unsigned long long)body->size);
1791 if (body->valid & OBD_MD_FLBLOCKS)
1792 inode->i_blocks = body->blocks;
1795 if (body->valid & OBD_MD_FLMDSCAPA) {
1796 LASSERT(md->mds_capa);
1797 ll_add_capa(inode, md->mds_capa);
1799 if (body->valid & OBD_MD_FLOSSCAPA) {
1800 LASSERT(md->oss_capa);
1801 ll_add_capa(inode, md->oss_capa);
1805 void ll_read_inode2(struct inode *inode, void *opaque)
1807 struct lustre_md *md = opaque;
1808 struct ll_inode_info *lli = ll_i2info(inode);
1811 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1812 PFID(&lli->lli_fid), inode);
1814 LASSERT(!lli->lli_has_smd);
1816 /* Core attributes from the MDS first. This is a new inode, and
1817 * the VFS doesn't zero times in the core inode so we have to do
1818 * it ourselves. They will be overwritten by either MDS or OST
1819 * attributes - we just need to make sure they aren't newer. */
1820 LTIME_S(inode->i_mtime) = 0;
1821 LTIME_S(inode->i_atime) = 0;
1822 LTIME_S(inode->i_ctime) = 0;
1824 ll_update_inode(inode, md);
1826 /* OIDEBUG(inode); */
1828 /* initializing backing dev info. */
1829 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1832 if (S_ISREG(inode->i_mode)) {
1833 struct ll_sb_info *sbi = ll_i2sbi(inode);
1834 inode->i_op = &ll_file_inode_operations;
1835 inode->i_fop = sbi->ll_fop;
1836 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1838 } else if (S_ISDIR(inode->i_mode)) {
1839 inode->i_op = &ll_dir_inode_operations;
1840 inode->i_fop = &ll_dir_operations;
1842 } else if (S_ISLNK(inode->i_mode)) {
1843 inode->i_op = &ll_fast_symlink_inode_operations;
1846 inode->i_op = &ll_special_inode_operations;
1848 init_special_inode(inode, inode->i_mode,
1855 void ll_delete_inode(struct inode *inode)
1857 struct cl_inode_info *lli = cl_i2info(inode);
1860 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
1861 /* discard all dirty pages before truncating them, required by
1862 * osc_extent implementation at LU-1030. */
1863 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF,
1864 CL_FSYNC_DISCARD, 1);
1866 truncate_inode_pages(&inode->i_data, 0);
1868 /* Workaround for LU-118 */
1869 if (inode->i_data.nrpages) {
1870 TREE_READ_LOCK_IRQ(&inode->i_data);
1871 TREE_READ_UNLOCK_IRQ(&inode->i_data);
1872 LASSERTF(inode->i_data.nrpages == 0,
1873 "inode=%lu/%u(%p) nrpages=%lu, see "
1874 "http://jira.whamcloud.com/browse/LU-118\n",
1875 inode->i_ino, inode->i_generation, inode,
1876 inode->i_data.nrpages);
1878 /* Workaround end */
1880 #ifdef HAVE_SBOPS_EVICT_INODE
1881 ll_clear_inode(inode);
1888 int ll_iocontrol(struct inode *inode, struct file *file,
1889 unsigned int cmd, unsigned long arg)
1891 struct ll_sb_info *sbi = ll_i2sbi(inode);
1892 struct ptlrpc_request *req = NULL;
1897 case FSFILT_IOC_GETFLAGS: {
1898 struct mdt_body *body;
1899 struct md_op_data *op_data;
1901 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
1902 0, 0, LUSTRE_OPC_ANY,
1904 if (IS_ERR(op_data))
1905 RETURN(PTR_ERR(op_data));
1907 op_data->op_valid = OBD_MD_FLFLAGS;
1908 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
1909 ll_finish_md_op_data(op_data);
1911 CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1915 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1917 flags = body->flags;
1919 ptlrpc_req_finished(req);
1921 RETURN(put_user(flags, (int *)arg));
1923 case FSFILT_IOC_SETFLAGS: {
1924 struct lov_stripe_md *lsm;
1925 struct obd_info oinfo = { { { 0 } } };
1926 struct md_op_data *op_data;
1928 if (get_user(flags, (int *)arg))
1931 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1932 LUSTRE_OPC_ANY, NULL);
1933 if (IS_ERR(op_data))
1934 RETURN(PTR_ERR(op_data));
1936 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1937 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1938 rc = md_setattr(sbi->ll_md_exp, op_data,
1939 NULL, 0, NULL, 0, &req, NULL);
1940 ll_finish_md_op_data(op_data);
1941 ptlrpc_req_finished(req);
1945 inode->i_flags = ll_ext_to_inode_flags(flags);
1947 lsm = ccc_inode_lsm_get(inode);
1948 if (!lsm_has_objects(lsm)) {
1949 ccc_inode_lsm_put(inode, lsm);
1953 OBDO_ALLOC(oinfo.oi_oa);
1955 ccc_inode_lsm_put(inode, lsm);
1959 oinfo.oi_oa->o_oi = lsm->lsm_oi;
1960 oinfo.oi_oa->o_flags = flags;
1961 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1963 oinfo.oi_capa = ll_mdscapa_get(inode);
1964 obdo_set_parent_fid(oinfo.oi_oa, &ll_i2info(inode)->lli_fid);
1965 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1966 capa_put(oinfo.oi_capa);
1967 OBDO_FREE(oinfo.oi_oa);
1968 ccc_inode_lsm_put(inode, lsm);
1970 if (rc && rc != -EPERM && rc != -EACCES)
1971 CERROR("osc_setattr_async fails: rc = %d\n", rc);
1982 int ll_flush_ctx(struct inode *inode)
1984 struct ll_sb_info *sbi = ll_i2sbi(inode);
1986 CDEBUG(D_SEC, "flush context for user %d\n", cfs_curproc_uid());
1988 obd_set_info_async(NULL, sbi->ll_md_exp,
1989 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1991 obd_set_info_async(NULL, sbi->ll_dt_exp,
1992 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1997 /* umount -f client means force down, don't save state */
1998 void ll_umount_begin(struct super_block *sb)
2000 struct ll_sb_info *sbi = ll_s2sbi(sb);
2001 struct obd_device *obd;
2002 struct obd_ioctl_data *ioc_data;
2005 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2006 sb->s_count, atomic_read(&sb->s_active));
2008 obd = class_exp2obd(sbi->ll_md_exp);
2010 CERROR("Invalid MDC connection handle "LPX64"\n",
2011 sbi->ll_md_exp->exp_handle.h_cookie);
2017 obd = class_exp2obd(sbi->ll_dt_exp);
2019 CERROR("Invalid LOV connection handle "LPX64"\n",
2020 sbi->ll_dt_exp->exp_handle.h_cookie);
2026 OBD_ALLOC_PTR(ioc_data);
2028 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2029 sizeof *ioc_data, ioc_data, NULL);
2031 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2032 sizeof *ioc_data, ioc_data, NULL);
2034 OBD_FREE_PTR(ioc_data);
2037 /* Really, we'd like to wait until there are no requests outstanding,
2038 * and then continue. For now, we just invalidate the requests,
2039 * schedule() and sleep one second if needed, and hope.
2045 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2047 struct ll_sb_info *sbi = ll_s2sbi(sb);
2048 char *profilenm = get_profile_name(sb);
2052 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2053 read_only = *flags & MS_RDONLY;
2054 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2055 sizeof(KEY_READ_ONLY),
2056 KEY_READ_ONLY, sizeof(read_only),
2059 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2060 profilenm, read_only ?
2061 "read-only" : "read-write", err);
2066 sb->s_flags |= MS_RDONLY;
2068 sb->s_flags &= ~MS_RDONLY;
2070 if (sbi->ll_flags & LL_SBI_VERBOSE)
2071 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2072 read_only ? "read-only" : "read-write");
2077 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2078 struct super_block *sb, struct lookup_intent *it)
2080 struct ll_sb_info *sbi = NULL;
2081 struct lustre_md md;
2085 LASSERT(*inode || sb);
2086 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2087 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2088 sbi->ll_md_exp, &md);
2093 ll_update_inode(*inode, &md);
2095 LASSERT(sb != NULL);
2098 * At this point server returns to client's same fid as client
2099 * generated for creating. So using ->fid1 is okay here.
2101 LASSERT(fid_is_sane(&md.body->fid1));
2103 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1,
2104 sbi->ll_flags & LL_SBI_32BIT_API),
2106 if (*inode == NULL || IS_ERR(*inode)) {
2107 #ifdef CONFIG_FS_POSIX_ACL
2109 posix_acl_release(md.posix_acl);
2110 md.posix_acl = NULL;
2113 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2115 CERROR("new_inode -fatal: rc %d\n", rc);
2120 /* Handling piggyback layout lock.
2121 * Layout lock can be piggybacked by getattr and open request.
2122 * The lsm can be applied to inode only if it comes with a layout lock
2123 * otherwise correct layout may be overwritten, for example:
2124 * 1. proc1: mdt returns a lsm but not granting layout
2125 * 2. layout was changed by another client
2126 * 3. proc2: refresh layout and layout lock granted
2127 * 4. proc1: to apply a stale layout */
2128 if (it != NULL && it->d.lustre.it_lock_mode != 0) {
2129 struct lustre_handle lockh;
2130 struct ldlm_lock *lock;
2132 lockh.cookie = it->d.lustre.it_lock_handle;
2133 lock = ldlm_handle2lock(&lockh);
2134 LASSERT(lock != NULL);
2135 if (ldlm_has_layout(lock)) {
2136 struct cl_object_conf conf;
2138 memset(&conf, 0, sizeof(conf));
2139 conf.coc_opc = OBJECT_CONF_SET;
2140 conf.coc_inode = *inode;
2141 conf.coc_lock = lock;
2142 conf.u.coc_md = &md;
2143 (void)ll_layout_conf(*inode, &conf);
2145 LDLM_LOCK_PUT(lock);
2150 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2151 md_free_lustre_md(sbi->ll_md_exp, &md);
2155 int ll_obd_statfs(struct inode *inode, void *arg)
2157 struct ll_sb_info *sbi = NULL;
2158 struct obd_export *exp;
2160 struct obd_ioctl_data *data = NULL;
2165 if (!inode || !(sbi = ll_i2sbi(inode)))
2166 GOTO(out_statfs, rc = -EINVAL);
2168 rc = obd_ioctl_getdata(&buf, &len, arg);
2170 GOTO(out_statfs, rc);
2173 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2174 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2175 GOTO(out_statfs, rc = -EINVAL);
2177 if (data->ioc_inllen1 != sizeof(__u32) ||
2178 data->ioc_inllen2 != sizeof(__u32) ||
2179 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2180 data->ioc_plen2 != sizeof(struct obd_uuid))
2181 GOTO(out_statfs, rc = -EINVAL);
2183 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2184 if (type & LL_STATFS_LMV)
2185 exp = sbi->ll_md_exp;
2186 else if (type & LL_STATFS_LOV)
2187 exp = sbi->ll_dt_exp;
2189 GOTO(out_statfs, rc = -ENODEV);
2191 flags = (type & LL_STATFS_NODELAY) ? OBD_STATFS_NODELAY : 0;
2192 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, &flags);
2194 GOTO(out_statfs, rc);
2197 obd_ioctl_freedata(buf, len);
2201 int ll_process_config(struct lustre_cfg *lcfg)
2205 struct lprocfs_static_vars lvars;
2209 lprocfs_llite_init_vars(&lvars);
2211 /* The instance name contains the sb: lustre-client-aacfe000 */
2212 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2213 if (!ptr || !*(++ptr))
2215 if (sscanf(ptr, "%lx", &x) != 1)
2218 /* This better be a real Lustre superblock! */
2219 LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2221 /* Note we have not called client_common_fill_super yet, so
2222 proc fns must be able to handle that! */
2223 rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2230 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2231 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2232 struct inode *i1, struct inode *i2,
2233 const char *name, int namelen,
2234 int mode, __u32 opc, void *data)
2236 LASSERT(i1 != NULL);
2238 if (namelen > ll_i2sbi(i1)->ll_namelen)
2239 return ERR_PTR(-ENAMETOOLONG);
2241 if (op_data == NULL)
2242 OBD_ALLOC_PTR(op_data);
2244 if (op_data == NULL)
2245 return ERR_PTR(-ENOMEM);
2247 ll_i2gids(op_data->op_suppgids, i1, i2);
2248 op_data->op_fid1 = *ll_inode2fid(i1);
2249 op_data->op_capa1 = ll_mdscapa_get(i1);
2252 op_data->op_fid2 = *ll_inode2fid(i2);
2253 op_data->op_capa2 = ll_mdscapa_get(i2);
2255 fid_zero(&op_data->op_fid2);
2256 op_data->op_capa2 = NULL;
2259 op_data->op_name = name;
2260 op_data->op_namelen = namelen;
2261 op_data->op_mode = mode;
2262 op_data->op_mod_time = cfs_time_current_sec();
2263 op_data->op_fsuid = cfs_curproc_fsuid();
2264 op_data->op_fsgid = cfs_curproc_fsgid();
2265 op_data->op_cap = cfs_curproc_cap_pack();
2266 op_data->op_bias = 0;
2267 op_data->op_cli_flags = 0;
2268 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2269 filename_is_volatile(name, namelen, NULL))
2270 op_data->op_bias |= MDS_CREATE_VOLATILE;
2271 op_data->op_opc = opc;
2272 op_data->op_mds = 0;
2273 op_data->op_data = data;
2275 /* If the file is being opened after mknod() (normally due to NFS)
2276 * try to use the default stripe data from parent directory for
2277 * allocating OST objects. Try to pass the parent FID to MDS. */
2278 if (opc == LUSTRE_OPC_CREATE && i1 == i2 && S_ISREG(i2->i_mode) &&
2279 !ll_i2info(i2)->lli_has_smd) {
2280 struct ll_inode_info *lli = ll_i2info(i2);
2282 spin_lock(&lli->lli_lock);
2283 if (likely(!lli->lli_has_smd && !fid_is_zero(&lli->lli_pfid)))
2284 op_data->op_fid1 = lli->lli_pfid;
2285 spin_unlock(&lli->lli_lock);
2286 /** We ignore parent's capability temporary. */
2289 /* When called by ll_setattr_raw, file is i1. */
2290 if (LLIF_DATA_MODIFIED & ll_i2info(i1)->lli_flags)
2291 op_data->op_bias |= MDS_DATA_MODIFIED;
2296 void ll_finish_md_op_data(struct md_op_data *op_data)
2298 capa_put(op_data->op_capa1);
2299 capa_put(op_data->op_capa2);
2300 OBD_FREE_PTR(op_data);
2303 #ifdef HAVE_SUPEROPS_USE_DENTRY
2304 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2306 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2309 struct ll_sb_info *sbi;
2311 #ifdef HAVE_SUPEROPS_USE_DENTRY
2312 LASSERT((seq != NULL) && (dentry != NULL));
2313 sbi = ll_s2sbi(dentry->d_sb);
2315 LASSERT((seq != NULL) && (vfs != NULL));
2316 sbi = ll_s2sbi(vfs->mnt_sb);
2319 if (sbi->ll_flags & LL_SBI_NOLCK)
2320 seq_puts(seq, ",nolock");
2322 if (sbi->ll_flags & LL_SBI_FLOCK)
2323 seq_puts(seq, ",flock");
2325 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2326 seq_puts(seq, ",localflock");
2328 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2329 seq_puts(seq, ",user_xattr");
2331 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2332 seq_puts(seq, ",lazystatfs");
2334 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2335 seq_puts(seq, ",user_fid2path");
2341 * Get obd name by cmd, and copy out to user space
2343 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2345 struct ll_sb_info *sbi = ll_i2sbi(inode);
2346 struct obd_device *obd;
2349 if (cmd == OBD_IOC_GETDTNAME)
2350 obd = class_exp2obd(sbi->ll_dt_exp);
2351 else if (cmd == OBD_IOC_GETMDNAME)
2352 obd = class_exp2obd(sbi->ll_md_exp);
2359 if (copy_to_user((void *)arg, obd->obd_name,
2360 strlen(obd->obd_name) + 1))
2367 * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2368 * fsname will be returned in this buffer; otherwise, a static buffer will be
2369 * used to store the fsname and returned to caller.
2371 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2373 static char fsname_static[MTI_NAME_MAXLEN];
2374 struct lustre_sb_info *lsi = s2lsi(sb);
2379 /* this means the caller wants to use static buffer
2380 * and it doesn't care about race. Usually this is
2381 * in error reporting path */
2382 buf = fsname_static;
2383 buflen = sizeof(fsname_static);
2386 len = strlen(lsi->lsi_lmd->lmd_profile);
2387 ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2388 if (ptr && (strcmp(ptr, "-client") == 0))
2391 if (unlikely(len >= buflen))
2393 strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2399 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2403 #ifdef HAVE_FS_STRUCT_USE_PATH
2407 p.mnt = current->fs->root.mnt;
2409 path = d_path(&p, buf, bufsize);
2412 path = d_path(dentry, current->fs->rootmnt, buf, bufsize);
2418 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2420 char *buf, *path = NULL;
2421 struct dentry *dentry = NULL;
2422 struct ccc_object *obj = cl_inode2ccc(page->mapping->host);
2424 /* this can be called inside spin lock so use GFP_ATOMIC. */
2425 buf = (char *)__get_free_page(GFP_ATOMIC);
2427 dentry = d_find_alias(page->mapping->host);
2429 path = ll_d_path(dentry, buf, PAGE_SIZE);
2432 CWARN("%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2433 "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2434 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2435 PFID(&obj->cob_header.coh_lu.loh_fid),
2436 (path && !IS_ERR(path)) ? path : "", ioret);
2442 free_page((unsigned long)buf);