4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2012, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/llite/llite_lib.c
38 * Lustre Light Super operations
41 #define DEBUG_SUBSYSTEM S_LLITE
43 #include <linux/module.h>
44 #include <linux/types.h>
47 #include <lustre_lite.h>
48 #include <lustre_ha.h>
49 #include <lustre_dlm.h>
50 #include <lprocfs_status.h>
51 #include <lustre_disk.h>
52 #include <lustre_param.h>
53 #include <lustre_log.h>
54 #include <cl_object.h>
55 #include <obd_cksum.h>
56 #include "llite_internal.h"
58 struct kmem_cache *ll_file_data_slab;
60 LIST_HEAD(ll_super_blocks);
61 DEFINE_SPINLOCK(ll_sb_lock);
63 #ifndef MS_HAS_NEW_AOPS
64 extern struct address_space_operations ll_aops;
66 extern struct address_space_operations_ext ll_aops;
70 #define log2(n) ffz(~(n))
73 static struct ll_sb_info *ll_init_sbi(void)
75 struct ll_sb_info *sbi = NULL;
77 unsigned long lru_page_max;
82 OBD_ALLOC(sbi, sizeof(*sbi));
86 spin_lock_init(&sbi->ll_lock);
87 mutex_init(&sbi->ll_lco.lco_lock);
88 spin_lock_init(&sbi->ll_pp_extent_lock);
89 spin_lock_init(&sbi->ll_process_lock);
90 sbi->ll_rw_stats_on = 0;
93 pages = si.totalram - si.totalhigh;
94 if (pages >> (20 - PAGE_CACHE_SHIFT) < 512) {
95 lru_page_max = pages / 2;
97 lru_page_max = (pages / 4) * 3;
100 /* initialize lru data */
101 atomic_set(&sbi->ll_cache.ccc_users, 0);
102 sbi->ll_cache.ccc_lru_max = lru_page_max;
103 atomic_set(&sbi->ll_cache.ccc_lru_left, lru_page_max);
104 spin_lock_init(&sbi->ll_cache.ccc_lru_lock);
105 INIT_LIST_HEAD(&sbi->ll_cache.ccc_lru);
107 sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
108 SBI_DEFAULT_READAHEAD_MAX);
109 sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
110 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
111 SBI_DEFAULT_READAHEAD_WHOLE_MAX;
112 INIT_LIST_HEAD(&sbi->ll_conn_chain);
113 INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
115 ll_generate_random_uuid(uuid);
116 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
117 CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
119 spin_lock(&ll_sb_lock);
120 list_add_tail(&sbi->ll_list, &ll_super_blocks);
121 spin_unlock(&ll_sb_lock);
123 sbi->ll_flags |= LL_SBI_VERBOSE;
124 sbi->ll_flags |= LL_SBI_CHECKSUM;
126 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
128 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
129 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
131 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
135 /* metadata statahead is enabled by default */
136 sbi->ll_sa_max = LL_SA_RPC_DEF;
137 atomic_set(&sbi->ll_sa_total, 0);
138 atomic_set(&sbi->ll_sa_wrong, 0);
139 atomic_set(&sbi->ll_agl_total, 0);
140 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
145 void ll_free_sbi(struct super_block *sb)
147 struct ll_sb_info *sbi = ll_s2sbi(sb);
150 spin_lock(&ll_sb_lock);
151 list_del(&sbi->ll_list);
152 spin_unlock(&ll_sb_lock);
153 OBD_FREE(sbi, sizeof(*sbi));
157 static struct dentry_operations ll_d_root_ops = {
158 .d_compare = ll_dcompare,
159 .d_revalidate = ll_revalidate_nd,
162 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
163 struct vfsmount *mnt)
165 struct inode *root = 0;
166 struct ll_sb_info *sbi = ll_s2sbi(sb);
167 struct obd_device *obd;
168 struct obd_capa *oc = NULL;
169 struct obd_statfs *osfs = NULL;
170 struct ptlrpc_request *request = NULL;
171 struct obd_connect_data *data = NULL;
172 struct obd_uuid *uuid;
173 struct md_op_data *op_data;
174 struct lustre_md lmd;
176 int size, err, checksum;
178 obd = class_name2obd(md);
180 CERROR("MD %s: not setup or attached\n", md);
194 if (proc_lustre_fs_root) {
195 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
198 CERROR("could not register mount in /proc/fs/lustre\n");
201 /* indicate the features supported by this client */
202 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
203 OBD_CONNECT_ATTRFID |
204 OBD_CONNECT_VERSION | OBD_CONNECT_BRW_SIZE |
205 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
206 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
207 OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 |
208 OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR |
209 OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH|
210 OBD_CONNECT_EINPROGRESS |
211 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
212 OBD_CONNECT_LAYOUTLOCK |
213 OBD_CONNECT_PINGLESS | OBD_CONNECT_MAX_EASIZE;
215 if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
216 data->ocd_connect_flags |= OBD_CONNECT_SOM;
218 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
219 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 LCONSOLE_INFO("Layout lock feature supported.\n");
384 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
387 if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
388 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
390 "%s: disabling xattr cache due to unknown maximum xattr size.\n",
393 sbi->ll_flags |= LL_SBI_XATTR_CACHE;
394 sbi->ll_xattr_cache_enabled = 1;
398 obd = class_name2obd(dt);
400 CERROR("DT %s: not setup or attached\n", dt);
401 GOTO(out_md_fid, err = -ENODEV);
404 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
405 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
406 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
407 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
408 OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
409 OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
410 OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
411 OBD_CONNECT_MAXBYTES |
412 OBD_CONNECT_EINPROGRESS |
413 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
414 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS;
416 if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
417 data->ocd_connect_flags |= OBD_CONNECT_SOM;
419 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
420 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
421 * disabled by default, because it can still be enabled on the
422 * fly via /proc. As a consequence, we still need to come to an
423 * agreement on the supported algorithms at connect time */
424 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
426 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
427 data->ocd_cksum_types = OBD_CKSUM_ADLER;
429 data->ocd_cksum_types = cksum_types_supported_client();
432 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
433 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
434 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
436 CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
437 "ocd_grant: %d\n", data->ocd_connect_flags,
438 data->ocd_version, data->ocd_grant);
440 obd->obd_upcall.onu_owner = &sbi->ll_lco;
441 obd->obd_upcall.onu_upcall = cl_ocd_update;
443 data->ocd_brw_size = DT_MAX_BRW_SIZE;
445 err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
448 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
449 "recovery, of which this client is not a "
450 "part. Please wait for recovery to "
451 "complete, abort, or time out.\n", dt);
454 CERROR("%s: Cannot connect to %s: rc = %d\n",
455 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
459 sbi->ll_dt_exp->exp_connect_data = *data;
461 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
462 LUSTRE_SEQ_METADATA);
464 CERROR("%s: Can't init data layer FID infrastructure, "
465 "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
469 mutex_lock(&sbi->ll_lco.lco_lock);
470 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
471 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
472 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
473 mutex_unlock(&sbi->ll_lco.lco_lock);
475 fid_zero(&sbi->ll_root_fid);
476 err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
478 CERROR("cannot mds_connect: rc = %d\n", err);
479 GOTO(out_lock_cn_cb, err);
481 if (!fid_is_sane(&sbi->ll_root_fid)) {
482 CERROR("%s: Invalid root fid "DFID" during mount\n",
483 sbi->ll_md_exp->exp_obd->obd_name,
484 PFID(&sbi->ll_root_fid));
485 GOTO(out_lock_cn_cb, err = -EINVAL);
487 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
489 sb->s_op = &lustre_super_operations;
490 #if THREAD_SIZE >= 8192 /*b=17630*/
491 sb->s_export_op = &lustre_export_operations;
495 * XXX: move this to after cbd setup? */
496 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
497 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
498 valid |= OBD_MD_FLRMTPERM;
499 else if (sbi->ll_flags & LL_SBI_ACL)
500 valid |= OBD_MD_FLACL;
502 OBD_ALLOC_PTR(op_data);
504 GOTO(out_lock_cn_cb, err = -ENOMEM);
506 op_data->op_fid1 = sbi->ll_root_fid;
507 op_data->op_mode = 0;
508 op_data->op_capa1 = oc;
509 op_data->op_valid = valid;
511 err = md_getattr(sbi->ll_md_exp, op_data, &request);
514 OBD_FREE_PTR(op_data);
516 CERROR("%s: md_getattr failed for root: rc = %d\n",
517 sbi->ll_md_exp->exp_obd->obd_name, err);
518 GOTO(out_lock_cn_cb, err);
521 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
522 sbi->ll_md_exp, &lmd);
524 CERROR("failed to understand root inode md: rc = %d\n", err);
525 ptlrpc_req_finished(request);
526 GOTO(out_lock_cn_cb, err);
529 LASSERT(fid_is_sane(&sbi->ll_root_fid));
530 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
531 sbi->ll_flags & LL_SBI_32BIT_API),
533 md_free_lustre_md(sbi->ll_md_exp, &lmd);
534 ptlrpc_req_finished(request);
536 if (root == NULL || IS_ERR(root)) {
538 obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
539 #ifdef CONFIG_FS_POSIX_ACL
541 posix_acl_release(lmd.posix_acl);
542 lmd.posix_acl = NULL;
545 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
547 CERROR("lustre_lite: bad iget4 for root\n");
551 err = ll_close_thread_start(&sbi->ll_lcq);
553 CERROR("cannot start close thread: rc %d\n", err);
557 #ifdef CONFIG_FS_POSIX_ACL
558 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
559 rct_init(&sbi->ll_rct);
560 et_init(&sbi->ll_et);
564 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
565 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
566 KEY_CHECKSUM, sizeof(checksum), &checksum,
570 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
571 KEY_CACHE_SET, sizeof(sbi->ll_cache),
572 &sbi->ll_cache, NULL);
574 sb->s_root = d_make_root(root);
575 if (sb->s_root == NULL) {
576 CERROR("%s: can't make root dentry\n",
577 ll_get_fsname(sb, NULL, 0));
578 GOTO(out_root, err = -ENOMEM);
581 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
582 d_set_d_op(sb->s_root, &ll_d_root_ops);
583 sb->s_d_op = &ll_d_ops;
585 sbi->ll_sdev_orig = sb->s_dev;
587 /* We set sb->s_dev equal on all lustre clients in order to support
588 * NFS export clustering. NFSD requires that the FSID be the same
590 /* s_dev is also used in lt_compare() to compare two fs, but that is
591 * only a node-local comparison. */
592 uuid = obd_get_uuid(sbi->ll_md_exp);
594 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
595 get_uuid2fsid(uuid->uuid, strlen(uuid->uuid), &sbi->ll_fsid);
608 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
610 obd_disconnect(sbi->ll_dt_exp);
611 sbi->ll_dt_exp = NULL;
612 /* Make sure all OScs are gone, since cl_cache is accessing sbi. */
613 obd_zombie_barrier();
615 obd_fid_fini(sbi->ll_md_exp->exp_obd);
617 obd_disconnect(sbi->ll_md_exp);
618 sbi->ll_md_exp = NULL;
624 lprocfs_unregister_mountpoint(sbi);
628 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
632 *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
634 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
635 KEY_MAX_EASIZE, &size, lmmsize, NULL);
637 CERROR("Get max mdsize error rc %d \n", rc);
642 void ll_dump_inode(struct inode *inode)
644 struct ll_d_hlist_node *tmp;
645 int dentry_count = 0;
647 LASSERT(inode != NULL);
649 ll_d_hlist_for_each(tmp, &inode->i_dentry)
652 CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
653 inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
654 inode->i_mode, atomic_read(&inode->i_count), dentry_count);
657 void lustre_dump_dentry(struct dentry *dentry, int recur)
659 struct list_head *tmp;
662 LASSERT(dentry != NULL);
664 list_for_each(tmp, &dentry->d_subdirs)
667 CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
668 " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
669 dentry->d_name.len, dentry->d_name.name,
670 dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
671 dentry->d_parent, dentry->d_inode, d_count(dentry),
672 dentry->d_flags, dentry->d_fsdata, subdirs);
673 if (dentry->d_inode != NULL)
674 ll_dump_inode(dentry->d_inode);
679 list_for_each(tmp, &dentry->d_subdirs) {
680 struct dentry *d = list_entry(tmp, struct dentry, d_u.d_child);
681 lustre_dump_dentry(d, recur - 1);
685 void client_common_put_super(struct super_block *sb)
687 struct ll_sb_info *sbi = ll_s2sbi(sb);
689 #ifdef CONFIG_FS_POSIX_ACL
690 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
691 et_fini(&sbi->ll_et);
692 rct_fini(&sbi->ll_rct);
696 ll_close_thread_shutdown(sbi->ll_lcq);
700 list_del(&sbi->ll_conn_chain);
702 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
703 obd_disconnect(sbi->ll_dt_exp);
704 sbi->ll_dt_exp = NULL;
705 /* wait till all OSCs are gone, since cl_cache is accessing sbi.
707 obd_zombie_barrier();
709 lprocfs_unregister_mountpoint(sbi);
711 obd_fid_fini(sbi->ll_md_exp->exp_obd);
712 obd_disconnect(sbi->ll_md_exp);
713 sbi->ll_md_exp = NULL;
716 void ll_kill_super(struct super_block *sb)
718 struct ll_sb_info *sbi;
721 if (!(sb->s_flags & MS_ACTIVE))
725 /* we need restore s_dev from changed for clustred NFS before put_super
726 * because new kernels have cached s_dev and change sb->s_dev in
727 * put_super not affected real removing devices */
729 sb->s_dev = sbi->ll_sdev_orig;
730 sbi->ll_umounting = 1;
734 char *ll_read_opt(const char *opt, char *data)
739 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
740 if (strncmp(opt, data, strlen(opt)))
742 if ((value = strchr(data, '=')) == NULL)
746 OBD_ALLOC(retval, strlen(value) + 1);
748 CERROR("out of memory!\n");
752 memcpy(retval, value, strlen(value)+1);
753 CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
757 static inline int ll_set_opt(const char *opt, char *data, int fl)
759 if (strncmp(opt, data, strlen(opt)) != 0)
765 /* non-client-specific mount options are parsed in lmd_parse */
766 static int ll_options(char *options, int *flags)
769 char *s1 = options, *s2;
774 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
777 CDEBUG(D_SUPER, "next opt=%s\n", s1);
778 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
783 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
788 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
793 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
798 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
803 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
808 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 5, 50, 0)
809 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
811 /* Ignore deprecated mount option. The client will
812 * always try to mount with ACL support, whether this
813 * is used depends on whether server supports it. */
814 LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
815 "mount option 'acl'.\n");
818 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
820 LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
821 "mount option 'noacl'.\n");
825 #warning "{no}acl options have been deprecated since 1.8, please remove them"
827 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
832 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
837 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
843 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
848 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
853 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
858 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
863 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
868 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
873 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
878 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
883 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
888 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
893 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
899 s2 = strchr(s1, ',');
907 void ll_lli_init(struct ll_inode_info *lli)
909 lli->lli_inode_magic = LLI_INODE_MAGIC;
911 lli->lli_ioepoch = 0;
912 lli->lli_maxbytes = MAX_LFS_FILESIZE;
913 spin_lock_init(&lli->lli_lock);
914 lli->lli_posix_acl = NULL;
915 lli->lli_remote_perms = NULL;
916 mutex_init(&lli->lli_rmtperm_mutex);
917 /* Do not set lli_fid, it has been initialized already. */
918 fid_zero(&lli->lli_pfid);
919 INIT_LIST_HEAD(&lli->lli_close_list);
920 INIT_LIST_HEAD(&lli->lli_oss_capas);
921 atomic_set(&lli->lli_open_count, 0);
922 lli->lli_mds_capa = NULL;
923 lli->lli_rmtperm_time = 0;
924 lli->lli_pending_och = NULL;
925 lli->lli_mds_read_och = NULL;
926 lli->lli_mds_write_och = NULL;
927 lli->lli_mds_exec_och = NULL;
928 lli->lli_open_fd_read_count = 0;
929 lli->lli_open_fd_write_count = 0;
930 lli->lli_open_fd_exec_count = 0;
931 mutex_init(&lli->lli_och_mutex);
932 spin_lock_init(&lli->lli_agl_lock);
933 lli->lli_has_smd = false;
934 lli->lli_layout_gen = LL_LAYOUT_GEN_NONE;
935 lli->lli_clob = NULL;
937 init_rwsem(&lli->lli_xattrs_list_rwsem);
938 mutex_init(&lli->lli_xattrs_enq_lock);
940 LASSERT(lli->lli_vfs_inode.i_mode != 0);
941 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
942 mutex_init(&lli->lli_readdir_mutex);
943 lli->lli_opendir_key = NULL;
945 lli->lli_def_acl = NULL;
946 spin_lock_init(&lli->lli_sa_lock);
947 lli->lli_opendir_pid = 0;
949 sema_init(&lli->lli_size_sem, 1);
950 lli->lli_size_sem_owner = NULL;
951 lli->lli_symlink_name = NULL;
952 init_rwsem(&lli->lli_trunc_sem);
953 mutex_init(&lli->lli_write_mutex);
954 init_rwsem(&lli->lli_glimpse_sem);
955 lli->lli_glimpse_time = 0;
956 INIT_LIST_HEAD(&lli->lli_agl_list);
957 lli->lli_agl_index = 0;
958 lli->lli_async_rc = 0;
959 lli->lli_volatile = false;
961 mutex_init(&lli->lli_layout_mutex);
964 static inline int ll_bdi_register(struct backing_dev_info *bdi)
966 static atomic_t ll_bdi_num = ATOMIC_INIT(0);
968 bdi->name = "lustre";
969 return bdi_register(bdi, NULL, "lustre-%d",
970 atomic_inc_return(&ll_bdi_num));
973 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
975 struct lustre_profile *lprof = NULL;
976 struct lustre_sb_info *lsi = s2lsi(sb);
977 struct ll_sb_info *sbi;
978 char *dt = NULL, *md = NULL;
979 char *profilenm = get_profile_name(sb);
980 struct config_llog_instance *cfg;
981 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
982 const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
985 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
991 try_module_get(THIS_MODULE);
993 /* client additional sb info */
994 lsi->lsi_llsbi = sbi = ll_init_sbi();
996 module_put(THIS_MODULE);
1001 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1003 GOTO(out_free, err);
1005 err = bdi_init(&lsi->lsi_bdi);
1007 GOTO(out_free, err);
1008 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1009 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1010 err = ll_bdi_register(&lsi->lsi_bdi);
1012 GOTO(out_free, err);
1014 sb->s_bdi = &lsi->lsi_bdi;
1016 /* Generate a string unique to this super, in case some joker tries
1017 to mount the same fs at two mount points.
1018 Use the address of the super itself.*/
1019 cfg->cfg_instance = sb;
1020 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1021 cfg->cfg_callback = class_config_llog_handler;
1022 /* set up client obds */
1023 err = lustre_process_log(sb, profilenm, cfg);
1025 CERROR("Unable to process log: %d\n", err);
1026 GOTO(out_free, err);
1029 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1030 lprof = class_get_profile(profilenm);
1031 if (lprof == NULL) {
1032 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1033 " read from the MGS. Does that filesystem "
1034 "exist?\n", profilenm);
1035 GOTO(out_free, err = -EINVAL);
1037 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1038 lprof->lp_md, lprof->lp_dt);
1040 OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1042 GOTO(out_free, err = -ENOMEM);
1043 sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1045 OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1047 GOTO(out_free, err = -ENOMEM);
1048 sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1050 /* connections, registrations, sb setup */
1051 err = client_common_fill_super(sb, md, dt, mnt);
1055 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1057 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1060 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1061 LCONSOLE_WARN("Mounted %s\n", profilenm);
1065 } /* ll_fill_super */
1067 void ll_put_super(struct super_block *sb)
1069 struct config_llog_instance cfg;
1070 struct obd_device *obd;
1071 struct lustre_sb_info *lsi = s2lsi(sb);
1072 struct ll_sb_info *sbi = ll_s2sbi(sb);
1073 char *profilenm = get_profile_name(sb);
1074 int next, force = 1;
1076 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1078 ll_print_capa_stat(sbi);
1080 cfg.cfg_instance = sb;
1081 lustre_end_log(sb, profilenm, &cfg);
1083 if (sbi->ll_md_exp) {
1084 obd = class_exp2obd(sbi->ll_md_exp);
1086 force = obd->obd_force;
1089 /* We need to set force before the lov_disconnect in
1090 lustre_common_put_super, since l_d cleans up osc's as well. */
1093 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1095 obd->obd_force = force;
1100 /* Only if client_common_fill_super succeeded */
1101 client_common_put_super(sb);
1105 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1106 class_manual_cleanup(obd);
1109 if (sbi->ll_flags & LL_SBI_VERBOSE)
1110 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1113 class_del_profile(profilenm);
1115 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1116 bdi_destroy(&lsi->lsi_bdi);
1117 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1121 lsi->lsi_llsbi = NULL;
1123 lustre_common_put_super(sb);
1125 module_put(THIS_MODULE);
1126 } /* client_put_super */
1128 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1130 struct inode *inode = NULL;
1132 /* NOTE: we depend on atomic igrab() -bzzz */
1133 lock_res_and_lock(lock);
1134 if (lock->l_resource->lr_lvb_inode) {
1135 struct ll_inode_info * lli;
1136 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1137 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1138 inode = igrab(lock->l_resource->lr_lvb_inode);
1140 inode = lock->l_resource->lr_lvb_inode;
1141 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1142 D_WARNING, lock, "lr_lvb_inode %p is "
1143 "bogus: magic %08x",
1144 lock->l_resource->lr_lvb_inode,
1145 lli->lli_inode_magic);
1149 unlock_res_and_lock(lock);
1153 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1155 struct inode *inode = NULL;
1156 /* NOTE: we depend on atomic igrab() -bzzz */
1157 lock_res_and_lock(lock);
1158 if (lock->l_ast_data) {
1159 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1160 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1161 inode = igrab(lock->l_ast_data);
1163 inode = lock->l_ast_data;
1164 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1165 D_WARNING, lock, "l_ast_data %p is "
1166 "bogus: magic %08x", lock->l_ast_data,
1167 lli->lli_inode_magic);
1171 unlock_res_and_lock(lock);
1175 void ll_clear_inode(struct inode *inode)
1177 struct ll_inode_info *lli = ll_i2info(inode);
1178 struct ll_sb_info *sbi = ll_i2sbi(inode);
1180 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1181 inode->i_generation, inode);
1183 if (S_ISDIR(inode->i_mode)) {
1184 /* these should have been cleared in ll_file_release */
1185 LASSERT(lli->lli_opendir_key == NULL);
1186 LASSERT(lli->lli_sai == NULL);
1187 LASSERT(lli->lli_opendir_pid == 0);
1190 spin_lock(&lli->lli_lock);
1191 ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1192 spin_unlock(&lli->lli_lock);
1193 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1195 LASSERT(!lli->lli_open_fd_write_count);
1196 LASSERT(!lli->lli_open_fd_read_count);
1197 LASSERT(!lli->lli_open_fd_exec_count);
1199 if (lli->lli_mds_write_och)
1200 ll_md_real_close(inode, FMODE_WRITE);
1201 if (lli->lli_mds_exec_och)
1202 ll_md_real_close(inode, FMODE_EXEC);
1203 if (lli->lli_mds_read_och)
1204 ll_md_real_close(inode, FMODE_READ);
1206 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1207 OBD_FREE(lli->lli_symlink_name,
1208 strlen(lli->lli_symlink_name) + 1);
1209 lli->lli_symlink_name = NULL;
1212 ll_xattr_cache_destroy(inode);
1214 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1215 LASSERT(lli->lli_posix_acl == NULL);
1216 if (lli->lli_remote_perms) {
1217 free_rmtperm_hash(lli->lli_remote_perms);
1218 lli->lli_remote_perms = NULL;
1221 #ifdef CONFIG_FS_POSIX_ACL
1222 else if (lli->lli_posix_acl) {
1223 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1224 LASSERT(lli->lli_remote_perms == NULL);
1225 posix_acl_release(lli->lli_posix_acl);
1226 lli->lli_posix_acl = NULL;
1229 lli->lli_inode_magic = LLI_INODE_DEAD;
1231 ll_clear_inode_capas(inode);
1232 if (!S_ISDIR(inode->i_mode))
1233 LASSERT(list_empty(&lli->lli_agl_list));
1236 * XXX This has to be done before lsm is freed below, because
1237 * cl_object still uses inode lsm.
1239 cl_inode_fini(inode);
1240 lli->lli_has_smd = false;
1243 int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data,
1244 struct md_open_data **mod)
1246 struct lustre_md md;
1247 struct inode *inode = dentry->d_inode;
1248 struct ll_sb_info *sbi = ll_i2sbi(inode);
1249 struct ptlrpc_request *request = NULL;
1252 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1253 LUSTRE_OPC_ANY, NULL);
1254 if (IS_ERR(op_data))
1255 return PTR_ERR(op_data);
1257 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1260 ptlrpc_req_finished(request);
1261 if (rc == -ENOENT) {
1263 /* Unlinked special device node? Or just a race?
1264 * Pretend we done everything. */
1265 if (!S_ISREG(inode->i_mode) &&
1266 !S_ISDIR(inode->i_mode)) {
1267 ia_valid = op_data->op_attr.ia_valid;
1268 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1269 rc = simple_setattr(dentry, &op_data->op_attr);
1270 op_data->op_attr.ia_valid = ia_valid;
1272 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1273 CERROR("md_setattr fails: rc = %d\n", rc);
1278 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1279 sbi->ll_md_exp, &md);
1281 ptlrpc_req_finished(request);
1285 ia_valid = op_data->op_attr.ia_valid;
1286 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1287 * cache is not cleared yet. */
1288 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1289 rc = simple_setattr(dentry, &op_data->op_attr);
1290 op_data->op_attr.ia_valid = ia_valid;
1292 /* Extract epoch data if obtained. */
1293 op_data->op_handle = md.body->handle;
1294 op_data->op_ioepoch = md.body->ioepoch;
1296 ll_update_inode(inode, &md);
1297 ptlrpc_req_finished(request);
1302 /* Close IO epoch and send Size-on-MDS attribute update. */
1303 static int ll_setattr_done_writing(struct inode *inode,
1304 struct md_op_data *op_data,
1305 struct md_open_data *mod)
1307 struct ll_inode_info *lli = ll_i2info(inode);
1310 LASSERT(op_data != NULL);
1311 if (!S_ISREG(inode->i_mode))
1314 CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1315 op_data->op_ioepoch, PFID(&lli->lli_fid));
1317 op_data->op_flags = MF_EPOCH_CLOSE;
1318 ll_done_writing_attr(inode, op_data);
1319 ll_pack_inode2opdata(inode, op_data, NULL);
1321 rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1322 if (rc == -EAGAIN) {
1323 /* MDS has instructed us to obtain Size-on-MDS attribute
1324 * from OSTs and send setattr to back to MDS. */
1325 rc = ll_som_update(inode, op_data);
1327 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1333 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1335 struct obd_capa *capa;
1338 if (attr->ia_valid & ATTR_SIZE)
1339 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1341 capa = ll_mdscapa_get(inode);
1343 rc = cl_setattr_ost(inode, attr, capa);
1345 if (attr->ia_valid & ATTR_SIZE)
1346 ll_truncate_free_capa(capa);
1354 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1355 * object(s) determine the file size and mtime. Otherwise, the MDS will
1356 * keep these values until such a time that objects are allocated for it.
1357 * We do the MDS operations first, as it is checking permissions for us.
1358 * We don't to the MDS RPC if there is nothing that we want to store there,
1359 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1360 * going to do an RPC anyways.
1362 * If we are doing a truncate, we will send the mtime and ctime updates
1363 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1364 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1367 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr)
1369 struct inode *inode = dentry->d_inode;
1370 struct ll_inode_info *lli = ll_i2info(inode);
1371 struct md_op_data *op_data = NULL;
1372 struct md_open_data *mod = NULL;
1373 bool file_is_released = false;
1374 int rc = 0, rc1 = 0;
1376 CDEBUG(D_VFSTRACE, "%s: setattr inode %p/fid:"DFID" from %llu to %llu, "
1377 "valid %x\n", ll_get_fsname(inode->i_sb, NULL, 0), inode,
1378 PFID(&lli->lli_fid), i_size_read(inode), attr->ia_size,
1381 if (attr->ia_valid & ATTR_SIZE) {
1382 /* Check new size against VFS/VM file size limit and rlimit */
1383 rc = inode_newsize_ok(inode, attr->ia_size);
1387 /* The maximum Lustre file size is variable, based on the
1388 * OST maximum object size and number of stripes. This
1389 * needs another check in addition to the VFS check above. */
1390 if (attr->ia_size > ll_file_maxbytes(inode)) {
1391 CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1392 PFID(&lli->lli_fid), attr->ia_size,
1393 ll_file_maxbytes(inode));
1397 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1400 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1401 if (attr->ia_valid & TIMES_SET_FLAGS) {
1402 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1403 !cfs_capable(CFS_CAP_FOWNER))
1407 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1408 if (attr->ia_valid & ATTR_CTIME) {
1409 attr->ia_ctime = CURRENT_TIME;
1410 attr->ia_valid |= ATTR_CTIME_SET;
1412 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1413 (attr->ia_valid & ATTR_ATIME)) {
1414 attr->ia_atime = CURRENT_TIME;
1415 attr->ia_valid |= ATTR_ATIME_SET;
1417 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1418 (attr->ia_valid & ATTR_MTIME)) {
1419 attr->ia_mtime = CURRENT_TIME;
1420 attr->ia_valid |= ATTR_MTIME_SET;
1423 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1424 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1425 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1426 cfs_time_current_sec());
1428 /* If we are changing file size, file content is modified, flag it. */
1429 if (attr->ia_valid & ATTR_SIZE) {
1430 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1431 spin_lock(&lli->lli_lock);
1432 lli->lli_flags |= LLIF_DATA_MODIFIED;
1433 spin_unlock(&lli->lli_lock);
1436 /* We always do an MDS RPC, even if we're only changing the size;
1437 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1439 OBD_ALLOC_PTR(op_data);
1440 if (op_data == NULL)
1443 if (!S_ISDIR(inode->i_mode)) {
1444 if (attr->ia_valid & ATTR_SIZE)
1445 inode_dio_write_done(inode);
1446 mutex_unlock(&inode->i_mutex);
1447 down_write(&lli->lli_trunc_sem);
1450 memcpy(&op_data->op_attr, attr, sizeof(*attr));
1452 /* Open epoch for truncate. */
1453 if (exp_connect_som(ll_i2mdexp(inode)) &&
1454 (attr->ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1455 op_data->op_flags = MF_EPOCH_OPEN;
1457 /* truncate on a released file must failed with -ENODATA,
1458 * so size must not be set on MDS for released file
1459 * but other attributes must be set
1461 if (S_ISREG(inode->i_mode)) {
1462 struct lov_stripe_md *lsm;
1465 ll_layout_refresh(inode, &gen);
1466 lsm = ccc_inode_lsm_get(inode);
1467 if (lsm && lsm->lsm_pattern & LOV_PATTERN_F_RELEASED)
1468 file_is_released = true;
1469 ccc_inode_lsm_put(inode, lsm);
1472 /* clear size attr for released file
1473 * we clear the attribute send to MDT in op_data, not the original
1474 * received from caller in attr which is used later to
1475 * decide return code */
1476 if (file_is_released && (attr->ia_valid & ATTR_SIZE))
1477 op_data->op_attr.ia_valid &= ~ATTR_SIZE;
1479 rc = ll_md_setattr(dentry, op_data, &mod);
1483 /* truncate failed, others succeed */
1484 if (file_is_released) {
1485 if (attr->ia_valid & ATTR_SIZE)
1486 GOTO(out, rc = -ENODATA);
1491 /* RPC to MDT is sent, cancel data modification flag */
1492 if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) {
1493 spin_lock(&lli->lli_lock);
1494 lli->lli_flags &= ~LLIF_DATA_MODIFIED;
1495 spin_unlock(&lli->lli_lock);
1498 ll_ioepoch_open(lli, op_data->op_ioepoch);
1499 if (!S_ISREG(inode->i_mode))
1502 if (attr->ia_valid & (ATTR_SIZE |
1503 ATTR_ATIME | ATTR_ATIME_SET |
1504 ATTR_MTIME | ATTR_MTIME_SET))
1505 /* For truncate and utimes sending attributes to OSTs, setting
1506 * mtime/atime to the past will be performed under PW [0:EOF]
1507 * extent lock (new_size:EOF for truncate). It may seem
1508 * excessive to send mtime/atime updates to OSTs when not
1509 * setting times to past, but it is necessary due to possible
1510 * time de-synchronization between MDT inode and OST objects */
1511 rc = ll_setattr_ost(inode, attr);
1514 if (op_data->op_ioepoch) {
1515 rc1 = ll_setattr_done_writing(inode, op_data, mod);
1519 ll_finish_md_op_data(op_data);
1521 if (!S_ISDIR(inode->i_mode)) {
1522 up_write(&lli->lli_trunc_sem);
1523 mutex_lock(&inode->i_mutex);
1524 if (attr->ia_valid & ATTR_SIZE)
1525 inode_dio_wait(inode);
1528 ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1529 LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1534 int ll_setattr(struct dentry *de, struct iattr *attr)
1536 int mode = de->d_inode->i_mode;
1538 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1539 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1540 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1542 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1543 (ATTR_SIZE|ATTR_MODE)) &&
1544 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1545 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1546 !(attr->ia_mode & S_ISGID))))
1547 attr->ia_valid |= ATTR_FORCE;
1549 if ((mode & S_ISUID) &&
1550 !(attr->ia_mode & S_ISUID) &&
1551 !(attr->ia_valid & ATTR_KILL_SUID))
1552 attr->ia_valid |= ATTR_KILL_SUID;
1554 if (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1555 !(attr->ia_mode & S_ISGID) &&
1556 !(attr->ia_valid & ATTR_KILL_SGID))
1557 attr->ia_valid |= ATTR_KILL_SGID;
1559 return ll_setattr_raw(de, attr);
1562 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1563 __u64 max_age, __u32 flags)
1565 struct ll_sb_info *sbi = ll_s2sbi(sb);
1566 struct obd_statfs obd_osfs;
1569 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1571 CERROR("md_statfs fails: rc = %d\n", rc);
1575 osfs->os_type = sb->s_magic;
1577 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1578 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1580 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1581 flags |= OBD_STATFS_NODELAY;
1583 rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1585 CERROR("obd_statfs fails: rc = %d\n", rc);
1589 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1590 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1593 osfs->os_bsize = obd_osfs.os_bsize;
1594 osfs->os_blocks = obd_osfs.os_blocks;
1595 osfs->os_bfree = obd_osfs.os_bfree;
1596 osfs->os_bavail = obd_osfs.os_bavail;
1598 /* If we don't have as many objects free on the OST as inodes
1599 * on the MDS, we reduce the total number of inodes to
1600 * compensate, so that the "inodes in use" number is correct.
1602 if (obd_osfs.os_ffree < osfs->os_ffree) {
1603 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1605 osfs->os_ffree = obd_osfs.os_ffree;
1610 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1612 struct super_block *sb = de->d_sb;
1613 struct obd_statfs osfs;
1616 CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1617 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1619 /* Some amount of caching on the client is allowed */
1620 rc = ll_statfs_internal(sb, &osfs,
1621 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1626 statfs_unpack(sfs, &osfs);
1628 /* We need to downshift for all 32-bit kernels, because we can't
1629 * tell if the kernel is being called via sys_statfs64() or not.
1630 * Stop before overflowing f_bsize - in which case it is better
1631 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1632 if (sizeof(long) < 8) {
1633 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1636 osfs.os_blocks >>= 1;
1637 osfs.os_bfree >>= 1;
1638 osfs.os_bavail >>= 1;
1642 sfs->f_blocks = osfs.os_blocks;
1643 sfs->f_bfree = osfs.os_bfree;
1644 sfs->f_bavail = osfs.os_bavail;
1645 sfs->f_fsid = ll_s2sbi(sb)->ll_fsid;
1649 void ll_inode_size_lock(struct inode *inode)
1651 struct ll_inode_info *lli;
1653 LASSERT(!S_ISDIR(inode->i_mode));
1655 lli = ll_i2info(inode);
1656 LASSERT(lli->lli_size_sem_owner != current);
1657 down(&lli->lli_size_sem);
1658 LASSERT(lli->lli_size_sem_owner == NULL);
1659 lli->lli_size_sem_owner = current;
1662 void ll_inode_size_unlock(struct inode *inode)
1664 struct ll_inode_info *lli;
1666 lli = ll_i2info(inode);
1667 LASSERT(lli->lli_size_sem_owner == current);
1668 lli->lli_size_sem_owner = NULL;
1669 up(&lli->lli_size_sem);
1672 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1674 struct ll_inode_info *lli = ll_i2info(inode);
1675 struct mdt_body *body = md->body;
1676 struct lov_stripe_md *lsm = md->lsm;
1677 struct ll_sb_info *sbi = ll_i2sbi(inode);
1679 LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1681 if (!lli->lli_has_smd &&
1682 !(sbi->ll_flags & LL_SBI_LAYOUT_LOCK))
1683 cl_file_inode_init(inode, md);
1685 lli->lli_maxbytes = lsm->lsm_maxbytes;
1686 if (lli->lli_maxbytes > MAX_LFS_FILESIZE)
1687 lli->lli_maxbytes = MAX_LFS_FILESIZE;
1690 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1691 if (body->valid & OBD_MD_FLRMTPERM)
1692 ll_update_remote_perm(inode, md->remote_perm);
1694 #ifdef CONFIG_FS_POSIX_ACL
1695 else if (body->valid & OBD_MD_FLACL) {
1696 spin_lock(&lli->lli_lock);
1697 if (lli->lli_posix_acl)
1698 posix_acl_release(lli->lli_posix_acl);
1699 lli->lli_posix_acl = md->posix_acl;
1700 spin_unlock(&lli->lli_lock);
1703 inode->i_ino = cl_fid_build_ino(&body->fid1,
1704 sbi->ll_flags & LL_SBI_32BIT_API);
1705 inode->i_generation = cl_fid_build_gen(&body->fid1);
1707 if (body->valid & OBD_MD_FLATIME) {
1708 if (body->atime > LTIME_S(inode->i_atime))
1709 LTIME_S(inode->i_atime) = body->atime;
1710 lli->lli_lvb.lvb_atime = body->atime;
1712 if (body->valid & OBD_MD_FLMTIME) {
1713 if (body->mtime > LTIME_S(inode->i_mtime)) {
1714 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1715 "to "LPU64"\n", inode->i_ino,
1716 LTIME_S(inode->i_mtime), body->mtime);
1717 LTIME_S(inode->i_mtime) = body->mtime;
1719 lli->lli_lvb.lvb_mtime = body->mtime;
1721 if (body->valid & OBD_MD_FLCTIME) {
1722 if (body->ctime > LTIME_S(inode->i_ctime))
1723 LTIME_S(inode->i_ctime) = body->ctime;
1724 lli->lli_lvb.lvb_ctime = body->ctime;
1726 if (body->valid & OBD_MD_FLMODE)
1727 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1728 if (body->valid & OBD_MD_FLTYPE)
1729 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1730 LASSERT(inode->i_mode != 0);
1731 if (S_ISREG(inode->i_mode)) {
1732 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1734 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1736 if (body->valid & OBD_MD_FLUID)
1737 inode->i_uid = make_kuid(&init_user_ns, body->uid);
1738 if (body->valid & OBD_MD_FLGID)
1739 inode->i_gid = make_kgid(&init_user_ns, body->gid);
1740 if (body->valid & OBD_MD_FLFLAGS)
1741 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1742 if (body->valid & OBD_MD_FLNLINK)
1743 set_nlink(inode, body->nlink);
1744 if (body->valid & OBD_MD_FLRDEV)
1745 inode->i_rdev = old_decode_dev(body->rdev);
1747 if (body->valid & OBD_MD_FLID) {
1748 /* FID shouldn't be changed! */
1749 if (fid_is_sane(&lli->lli_fid)) {
1750 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1751 "Trying to change FID "DFID
1752 " to the "DFID", inode %lu/%u(%p)\n",
1753 PFID(&lli->lli_fid), PFID(&body->fid1),
1754 inode->i_ino, inode->i_generation, inode);
1756 lli->lli_fid = body->fid1;
1759 LASSERT(fid_seq(&lli->lli_fid) != 0);
1761 if (body->valid & OBD_MD_FLSIZE) {
1762 if (exp_connect_som(ll_i2mdexp(inode)) &&
1763 S_ISREG(inode->i_mode)) {
1764 struct lustre_handle lockh;
1767 /* As it is possible a blocking ast has been processed
1768 * by this time, we need to check there is an UPDATE
1769 * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1771 mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1772 &lockh, LDLM_FL_CBPENDING,
1776 if (lli->lli_flags & (LLIF_DONE_WRITING |
1777 LLIF_EPOCH_PENDING |
1779 CERROR("ino %lu flags %u still has "
1780 "size authority! do not trust "
1781 "the size got from MDS\n",
1782 inode->i_ino, lli->lli_flags);
1784 /* Use old size assignment to avoid
1785 * deadlock bz14138 & bz14326 */
1786 i_size_write(inode, body->size);
1787 spin_lock(&lli->lli_lock);
1788 lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1789 spin_unlock(&lli->lli_lock);
1791 ldlm_lock_decref(&lockh, mode);
1794 /* Use old size assignment to avoid
1795 * deadlock bz14138 & bz14326 */
1796 i_size_write(inode, body->size);
1798 CDEBUG(D_VFSTRACE, "inode=%lu, updating i_size %llu\n",
1799 inode->i_ino, (unsigned long long)body->size);
1802 if (body->valid & OBD_MD_FLBLOCKS)
1803 inode->i_blocks = body->blocks;
1806 if (body->valid & OBD_MD_FLMDSCAPA) {
1807 LASSERT(md->mds_capa);
1808 ll_add_capa(inode, md->mds_capa);
1810 if (body->valid & OBD_MD_FLOSSCAPA) {
1811 LASSERT(md->oss_capa);
1812 ll_add_capa(inode, md->oss_capa);
1815 if (body->valid & OBD_MD_TSTATE) {
1816 if (body->t_state & MS_RESTORE)
1817 lli->lli_flags |= LLIF_FILE_RESTORING;
1821 void ll_read_inode2(struct inode *inode, void *opaque)
1823 struct lustre_md *md = opaque;
1824 struct ll_inode_info *lli = ll_i2info(inode);
1826 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1827 PFID(&lli->lli_fid), inode);
1829 LASSERT(!lli->lli_has_smd);
1831 /* Core attributes from the MDS first. This is a new inode, and
1832 * the VFS doesn't zero times in the core inode so we have to do
1833 * it ourselves. They will be overwritten by either MDS or OST
1834 * attributes - we just need to make sure they aren't newer. */
1835 LTIME_S(inode->i_mtime) = 0;
1836 LTIME_S(inode->i_atime) = 0;
1837 LTIME_S(inode->i_ctime) = 0;
1839 ll_update_inode(inode, md);
1841 /* OIDEBUG(inode); */
1843 /* initializing backing dev info. */
1844 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1847 if (S_ISREG(inode->i_mode)) {
1848 struct ll_sb_info *sbi = ll_i2sbi(inode);
1849 inode->i_op = &ll_file_inode_operations;
1850 inode->i_fop = sbi->ll_fop;
1851 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1852 } else if (S_ISDIR(inode->i_mode)) {
1853 inode->i_op = &ll_dir_inode_operations;
1854 inode->i_fop = &ll_dir_operations;
1855 } else if (S_ISLNK(inode->i_mode)) {
1856 inode->i_op = &ll_fast_symlink_inode_operations;
1858 inode->i_op = &ll_special_inode_operations;
1860 init_special_inode(inode, inode->i_mode,
1865 void ll_delete_inode(struct inode *inode)
1867 struct cl_inode_info *lli = cl_i2info(inode);
1869 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
1870 /* discard all dirty pages before truncating them, required by
1871 * osc_extent implementation at LU-1030. */
1872 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF,
1873 CL_FSYNC_DISCARD, 1);
1875 truncate_inode_pages(&inode->i_data, 0);
1877 /* Workaround for LU-118 */
1878 if (inode->i_data.nrpages) {
1879 TREE_READ_LOCK_IRQ(&inode->i_data);
1880 TREE_READ_UNLOCK_IRQ(&inode->i_data);
1881 LASSERTF(inode->i_data.nrpages == 0,
1882 "inode=%lu/%u(%p) nrpages=%lu, see "
1883 "http://jira.whamcloud.com/browse/LU-118\n",
1884 inode->i_ino, inode->i_generation, inode,
1885 inode->i_data.nrpages);
1887 /* Workaround end */
1889 ll_clear_inode(inode);
1893 int ll_iocontrol(struct inode *inode, struct file *file,
1894 unsigned int cmd, unsigned long arg)
1896 struct ll_sb_info *sbi = ll_i2sbi(inode);
1897 struct ptlrpc_request *req = NULL;
1901 case FSFILT_IOC_GETFLAGS: {
1902 struct mdt_body *body;
1903 struct md_op_data *op_data;
1905 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
1906 0, 0, LUSTRE_OPC_ANY,
1908 if (IS_ERR(op_data))
1909 return PTR_ERR(op_data);
1911 op_data->op_valid = OBD_MD_FLFLAGS;
1912 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
1913 ll_finish_md_op_data(op_data);
1915 CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1919 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1921 flags = body->flags;
1923 ptlrpc_req_finished(req);
1925 return put_user(flags, (int *)arg);
1927 case FSFILT_IOC_SETFLAGS: {
1928 struct lov_stripe_md *lsm;
1929 struct obd_info oinfo = { { { 0 } } };
1930 struct md_op_data *op_data;
1932 if (get_user(flags, (int *)arg))
1935 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1936 LUSTRE_OPC_ANY, NULL);
1937 if (IS_ERR(op_data))
1938 return PTR_ERR(op_data);
1940 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1941 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1942 rc = md_setattr(sbi->ll_md_exp, op_data,
1943 NULL, 0, NULL, 0, &req, NULL);
1944 ll_finish_md_op_data(op_data);
1945 ptlrpc_req_finished(req);
1949 inode->i_flags = ll_ext_to_inode_flags(flags);
1951 lsm = ccc_inode_lsm_get(inode);
1952 if (!lsm_has_objects(lsm)) {
1953 ccc_inode_lsm_put(inode, lsm);
1957 OBDO_ALLOC(oinfo.oi_oa);
1959 ccc_inode_lsm_put(inode, lsm);
1963 oinfo.oi_oa->o_oi = lsm->lsm_oi;
1964 oinfo.oi_oa->o_flags = flags;
1965 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1967 oinfo.oi_capa = ll_mdscapa_get(inode);
1968 obdo_set_parent_fid(oinfo.oi_oa, &ll_i2info(inode)->lli_fid);
1969 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1970 capa_put(oinfo.oi_capa);
1971 OBDO_FREE(oinfo.oi_oa);
1972 ccc_inode_lsm_put(inode, lsm);
1974 if (rc && rc != -EPERM && rc != -EACCES)
1975 CERROR("osc_setattr_async fails: rc = %d\n", rc);
1986 int ll_flush_ctx(struct inode *inode)
1988 struct ll_sb_info *sbi = ll_i2sbi(inode);
1990 CDEBUG(D_SEC, "flush context for user %d\n",
1991 from_kuid(&init_user_ns, current_uid()));
1993 obd_set_info_async(NULL, sbi->ll_md_exp,
1994 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1996 obd_set_info_async(NULL, sbi->ll_dt_exp,
1997 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2002 /* umount -f client means force down, don't save state */
2003 void ll_umount_begin(struct super_block *sb)
2005 struct ll_sb_info *sbi = ll_s2sbi(sb);
2006 struct obd_device *obd;
2007 struct obd_ioctl_data *ioc_data;
2009 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2010 sb->s_count, atomic_read(&sb->s_active));
2012 obd = class_exp2obd(sbi->ll_md_exp);
2014 CERROR("Invalid MDC connection handle "LPX64"\n",
2015 sbi->ll_md_exp->exp_handle.h_cookie);
2020 obd = class_exp2obd(sbi->ll_dt_exp);
2022 CERROR("Invalid LOV connection handle "LPX64"\n",
2023 sbi->ll_dt_exp->exp_handle.h_cookie);
2028 OBD_ALLOC_PTR(ioc_data);
2030 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2031 sizeof(*ioc_data), ioc_data, NULL);
2033 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2034 sizeof(*ioc_data), ioc_data, NULL);
2036 OBD_FREE_PTR(ioc_data);
2039 /* Really, we'd like to wait until there are no requests outstanding,
2040 * and then continue. For now, we just invalidate the requests,
2041 * schedule() and sleep one second if needed, and hope.
2046 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2048 struct ll_sb_info *sbi = ll_s2sbi(sb);
2049 char *profilenm = get_profile_name(sb);
2053 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2054 read_only = *flags & MS_RDONLY;
2055 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2056 sizeof(KEY_READ_ONLY),
2057 KEY_READ_ONLY, sizeof(read_only),
2060 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2061 profilenm, read_only ?
2062 "read-only" : "read-write", err);
2067 sb->s_flags |= MS_RDONLY;
2069 sb->s_flags &= ~MS_RDONLY;
2071 if (sbi->ll_flags & LL_SBI_VERBOSE)
2072 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2073 read_only ? "read-only" : "read-write");
2078 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2079 struct super_block *sb, struct lookup_intent *it)
2081 struct ll_sb_info *sbi = NULL;
2082 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 = from_kuid(&init_user_ns, current_fsuid());
2264 op_data->op_fsgid = from_kgid(&init_user_ns, current_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 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2305 struct ll_sb_info *sbi;
2307 LASSERT((seq != NULL) && (dentry != NULL));
2308 sbi = ll_s2sbi(dentry->d_sb);
2310 if (sbi->ll_flags & LL_SBI_NOLCK)
2311 seq_puts(seq, ",nolock");
2313 if (sbi->ll_flags & LL_SBI_FLOCK)
2314 seq_puts(seq, ",flock");
2316 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2317 seq_puts(seq, ",localflock");
2319 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2320 seq_puts(seq, ",user_xattr");
2322 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2323 seq_puts(seq, ",lazystatfs");
2325 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2326 seq_puts(seq, ",user_fid2path");
2332 * Get obd name by cmd, and copy out to user space
2334 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2336 struct ll_sb_info *sbi = ll_i2sbi(inode);
2337 struct obd_device *obd;
2339 if (cmd == OBD_IOC_GETDTNAME)
2340 obd = class_exp2obd(sbi->ll_dt_exp);
2341 else if (cmd == OBD_IOC_GETMDNAME)
2342 obd = class_exp2obd(sbi->ll_md_exp);
2349 if (copy_to_user((void *)arg, obd->obd_name,
2350 strlen(obd->obd_name) + 1))
2357 * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2358 * fsname will be returned in this buffer; otherwise, a static buffer will be
2359 * used to store the fsname and returned to caller.
2361 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2363 static char fsname_static[MTI_NAME_MAXLEN];
2364 struct lustre_sb_info *lsi = s2lsi(sb);
2369 /* this means the caller wants to use static buffer
2370 * and it doesn't care about race. Usually this is
2371 * in error reporting path */
2372 buf = fsname_static;
2373 buflen = sizeof(fsname_static);
2376 len = strlen(lsi->lsi_lmd->lmd_profile);
2377 ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2378 if (ptr && (strcmp(ptr, "-client") == 0))
2381 if (unlikely(len >= buflen))
2383 strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2389 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2396 p.mnt = current->fs->root.mnt;
2398 path = d_path(&p, buf, bufsize);
2404 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2406 char *buf, *path = NULL;
2407 struct dentry *dentry = NULL;
2408 struct ccc_object *obj = cl_inode2ccc(page->mapping->host);
2410 /* this can be called inside spin lock so use GFP_ATOMIC. */
2411 buf = (char *)__get_free_page(GFP_ATOMIC);
2413 dentry = d_find_alias(page->mapping->host);
2415 path = ll_d_path(dentry, buf, PAGE_SIZE);
2418 CWARN("%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2419 "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2420 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2421 PFID(&obj->cob_header.coh_lu.loh_fid),
2422 (path && !IS_ERR(path)) ? path : "", ioret);
2428 free_page((unsigned long)buf);