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_ioctl.h>
49 #include <lustre_lite.h>
50 #include <lustre_ha.h>
51 #include <lustre_dlm.h>
52 #include <lprocfs_status.h>
53 #include <lustre_disk.h>
54 #include <lustre_param.h>
55 #include <lustre_log.h>
56 #include <cl_object.h>
57 #include <obd_cksum.h>
58 #include "llite_internal.h"
60 struct kmem_cache *ll_file_data_slab;
62 static struct list_head ll_super_blocks = LIST_HEAD_INIT(ll_super_blocks);
63 static DEFINE_SPINLOCK(ll_sb_lock);
66 #define log2(n) ffz(~(n))
69 static struct ll_sb_info *ll_init_sbi(void)
71 struct ll_sb_info *sbi = NULL;
73 unsigned long lru_page_max;
79 OBD_ALLOC(sbi, sizeof(*sbi));
83 spin_lock_init(&sbi->ll_lock);
84 mutex_init(&sbi->ll_lco.lco_lock);
85 spin_lock_init(&sbi->ll_pp_extent_lock);
86 spin_lock_init(&sbi->ll_process_lock);
87 sbi->ll_rw_stats_on = 0;
90 pages = si.totalram - si.totalhigh;
91 lru_page_max = pages / 2;
93 /* initialize ll_cache data */
94 atomic_set(&sbi->ll_cache.ccc_users, 0);
95 sbi->ll_cache.ccc_lru_max = lru_page_max;
96 atomic_set(&sbi->ll_cache.ccc_lru_left, lru_page_max);
97 spin_lock_init(&sbi->ll_cache.ccc_lru_lock);
98 INIT_LIST_HEAD(&sbi->ll_cache.ccc_lru);
100 atomic_set(&sbi->ll_cache.ccc_unstable_nr, 0);
101 init_waitqueue_head(&sbi->ll_cache.ccc_unstable_waitq);
103 sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
104 SBI_DEFAULT_READAHEAD_MAX);
105 sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
106 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
107 SBI_DEFAULT_READAHEAD_WHOLE_MAX;
108 INIT_LIST_HEAD(&sbi->ll_conn_chain);
109 INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
111 ll_generate_random_uuid(uuid);
112 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
113 CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
115 spin_lock(&ll_sb_lock);
116 list_add_tail(&sbi->ll_list, &ll_super_blocks);
117 spin_unlock(&ll_sb_lock);
119 sbi->ll_flags |= LL_SBI_VERBOSE;
120 #ifdef ENABLE_CHECKSUM
121 sbi->ll_flags |= LL_SBI_CHECKSUM;
124 #ifdef HAVE_LRU_RESIZE_SUPPORT
125 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 static void ll_free_sbi(struct super_block *sb)
147 struct ll_sb_info *sbi = ll_s2sbi(sb);
151 spin_lock(&ll_sb_lock);
152 list_del(&sbi->ll_list);
153 spin_unlock(&ll_sb_lock);
154 OBD_FREE(sbi, sizeof(*sbi));
159 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
160 struct vfsmount *mnt)
162 struct inode *root = 0;
163 struct ll_sb_info *sbi = ll_s2sbi(sb);
164 struct obd_device *obd;
165 struct obd_capa *oc = NULL;
166 struct obd_statfs *osfs = NULL;
167 struct ptlrpc_request *request = NULL;
168 struct obd_connect_data *data = NULL;
169 struct obd_uuid *uuid;
170 struct md_op_data *op_data;
171 struct lustre_md lmd;
173 int size, err, checksum;
176 obd = class_name2obd(md);
178 CERROR("MD %s: not setup or attached\n", md);
192 /* indicate the features supported by this client */
193 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
194 OBD_CONNECT_ATTRFID |
195 OBD_CONNECT_VERSION | OBD_CONNECT_BRW_SIZE |
196 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
197 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
198 OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 |
199 OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR |
200 OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH|
201 OBD_CONNECT_EINPROGRESS |
202 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
203 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS |
204 OBD_CONNECT_MAX_EASIZE |
205 OBD_CONNECT_FLOCK_DEAD |
206 OBD_CONNECT_DISP_STRIPE;
208 if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
209 data->ocd_connect_flags |= OBD_CONNECT_SOM;
211 #ifdef HAVE_LRU_RESIZE_SUPPORT
212 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
213 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
215 #ifdef CONFIG_FS_POSIX_ACL
216 data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK;
219 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
220 /* flag mdc connection as lightweight, only used for test
221 * purpose, use with care */
222 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
224 data->ocd_ibits_known = MDS_INODELOCK_FULL;
225 data->ocd_version = LUSTRE_VERSION_CODE;
227 if (sb->s_flags & MS_RDONLY)
228 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
229 if (sbi->ll_flags & LL_SBI_USER_XATTR)
230 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
232 #ifdef HAVE_MS_FLOCK_LOCK
233 /* force vfs to use lustre handler for flock() calls - bug 10743 */
234 sb->s_flags |= MS_FLOCK_LOCK;
236 #ifdef MS_HAS_NEW_AOPS
237 sb->s_flags |= MS_HAS_NEW_AOPS;
240 if (sbi->ll_flags & LL_SBI_FLOCK)
241 sbi->ll_fop = &ll_file_operations_flock;
242 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
243 sbi->ll_fop = &ll_file_operations;
245 sbi->ll_fop = &ll_file_operations_noflock;
248 data->ocd_connect_flags |= OBD_CONNECT_REAL;
249 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
250 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
252 data->ocd_brw_size = MD_MAX_BRW_SIZE;
254 err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
256 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
257 "recovery, of which this client is not a "
258 "part. Please wait for recovery to complete,"
259 " abort, or time out.\n", md);
262 CERROR("cannot connect to %s: rc = %d\n", md, err);
266 sbi->ll_md_exp->exp_connect_data = *data;
268 err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
269 LUSTRE_SEQ_METADATA);
271 CERROR("%s: Can't init metadata layer FID infrastructure, "
272 "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
276 /* For mount, we only need fs info from MDT0, and also in DNE, it
277 * can make sure the client can be mounted as long as MDT0 is
279 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
280 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
281 OBD_STATFS_FOR_MDT0);
283 GOTO(out_md_fid, err);
285 /* This needs to be after statfs to ensure connect has finished.
286 * Note that "data" does NOT contain the valid connect reply.
287 * If connecting to a 1.8 server there will be no LMV device, so
288 * we can access the MDC export directly and exp_connect_flags will
289 * be non-zero, but if accessing an upgraded 2.1 server it will
290 * have the correct flags filled in.
291 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
292 valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
293 if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
294 valid != CLIENT_CONNECT_MDT_REQD) {
297 OBD_ALLOC_WAIT(buf, PAGE_CACHE_SIZE);
298 obd_connect_flags2str(buf, PAGE_CACHE_SIZE,
299 valid ^ CLIENT_CONNECT_MDT_REQD, ",");
300 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
301 "feature(s) needed for correct operation "
302 "of this client (%s). Please upgrade "
303 "server or downgrade client.\n",
304 sbi->ll_md_exp->exp_obd->obd_name, buf);
305 OBD_FREE(buf, PAGE_CACHE_SIZE);
306 GOTO(out_md_fid, err = -EPROTO);
309 size = sizeof(*data);
310 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
311 KEY_CONN_DATA, &size, data, NULL);
313 CERROR("%s: Get connect data failed: rc = %d\n",
314 sbi->ll_md_exp->exp_obd->obd_name, err);
315 GOTO(out_md_fid, err);
318 LASSERT(osfs->os_bsize);
319 sb->s_blocksize = osfs->os_bsize;
320 sb->s_blocksize_bits = log2(osfs->os_bsize);
321 sb->s_magic = LL_SUPER_MAGIC;
322 sb->s_maxbytes = MAX_LFS_FILESIZE;
323 sbi->ll_namelen = osfs->os_namelen;
324 sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
326 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
327 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
328 LCONSOLE_INFO("Disabling user_xattr feature because "
329 "it is not supported on the server\n");
330 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
333 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
335 sb->s_flags |= MS_POSIXACL;
337 sbi->ll_flags |= LL_SBI_ACL;
339 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
341 sb->s_flags &= ~MS_POSIXACL;
343 sbi->ll_flags &= ~LL_SBI_ACL;
346 if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
347 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
348 sbi->ll_flags |= LL_SBI_RMT_CLIENT;
349 LCONSOLE_INFO("client is set as remote by default.\n");
352 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
353 sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
354 LCONSOLE_INFO("client claims to be remote, but server "
355 "rejected, forced to be local.\n");
359 if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
360 LCONSOLE_INFO("client enabled MDS capability!\n");
361 sbi->ll_flags |= LL_SBI_MDS_CAPA;
364 if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
365 LCONSOLE_INFO("client enabled OSS capability!\n");
366 sbi->ll_flags |= LL_SBI_OSS_CAPA;
369 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
370 sbi->ll_flags |= LL_SBI_64BIT_HASH;
372 if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
373 sbi->ll_md_brw_size = data->ocd_brw_size;
375 sbi->ll_md_brw_size = PAGE_CACHE_SIZE;
377 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
378 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
380 if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
381 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
382 LCONSOLE_INFO("%s: disabling xattr cache due to "
383 "unknown maximum xattr size.\n", dt);
385 sbi->ll_flags |= LL_SBI_XATTR_CACHE;
386 sbi->ll_xattr_cache_enabled = 1;
390 obd = class_name2obd(dt);
392 CERROR("DT %s: not setup or attached\n", dt);
393 GOTO(out_md_fid, err = -ENODEV);
396 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
397 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
398 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
399 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
400 OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
401 OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
402 OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
403 OBD_CONNECT_MAXBYTES |
404 OBD_CONNECT_EINPROGRESS |
405 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
406 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS;
408 if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
409 data->ocd_connect_flags |= OBD_CONNECT_SOM;
411 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
412 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
413 * disabled by default, because it can still be enabled on the
414 * fly via /proc. As a consequence, we still need to come to an
415 * agreement on the supported algorithms at connect time */
416 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
418 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
419 data->ocd_cksum_types = OBD_CKSUM_ADLER;
421 data->ocd_cksum_types = cksum_types_supported_client();
424 #ifdef HAVE_LRU_RESIZE_SUPPORT
425 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
427 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
428 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
430 CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
431 "ocd_grant: %d\n", data->ocd_connect_flags,
432 data->ocd_version, data->ocd_grant);
434 obd->obd_upcall.onu_owner = &sbi->ll_lco;
435 obd->obd_upcall.onu_upcall = cl_ocd_update;
437 data->ocd_brw_size = DT_MAX_BRW_SIZE;
439 err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
442 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
443 "recovery, of which this client is not a "
444 "part. Please wait for recovery to "
445 "complete, abort, or time out.\n", dt);
448 CERROR("%s: Cannot connect to %s: rc = %d\n",
449 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
453 sbi->ll_dt_exp->exp_connect_data = *data;
455 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
456 LUSTRE_SEQ_METADATA);
458 CERROR("%s: Can't init data layer FID infrastructure, "
459 "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
463 mutex_lock(&sbi->ll_lco.lco_lock);
464 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
465 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
466 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
467 mutex_unlock(&sbi->ll_lco.lco_lock);
469 fid_zero(&sbi->ll_root_fid);
470 err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
472 CERROR("cannot mds_connect: rc = %d\n", err);
473 GOTO(out_lock_cn_cb, err);
475 if (!fid_is_sane(&sbi->ll_root_fid)) {
476 CERROR("%s: Invalid root fid "DFID" during mount\n",
477 sbi->ll_md_exp->exp_obd->obd_name,
478 PFID(&sbi->ll_root_fid));
479 GOTO(out_lock_cn_cb, err = -EINVAL);
481 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
483 sb->s_op = &lustre_super_operations;
484 #if THREAD_SIZE >= 8192 /*b=17630*/
485 sb->s_export_op = &lustre_export_operations;
489 * XXX: move this to after cbd setup? */
490 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA |
492 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
493 valid |= OBD_MD_FLRMTPERM;
494 else if (sbi->ll_flags & LL_SBI_ACL)
495 valid |= OBD_MD_FLACL;
497 OBD_ALLOC_PTR(op_data);
499 GOTO(out_lock_cn_cb, err = -ENOMEM);
501 op_data->op_fid1 = sbi->ll_root_fid;
502 op_data->op_mode = 0;
503 op_data->op_capa1 = oc;
504 op_data->op_valid = valid;
506 err = md_getattr(sbi->ll_md_exp, op_data, &request);
509 OBD_FREE_PTR(op_data);
511 CERROR("%s: md_getattr failed for root: rc = %d\n",
512 sbi->ll_md_exp->exp_obd->obd_name, err);
513 GOTO(out_lock_cn_cb, err);
516 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
517 sbi->ll_md_exp, &lmd);
519 CERROR("failed to understand root inode md: rc = %d\n", err);
520 ptlrpc_req_finished(request);
521 GOTO(out_lock_cn_cb, err);
524 LASSERT(fid_is_sane(&sbi->ll_root_fid));
525 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
526 sbi->ll_flags & LL_SBI_32BIT_API),
528 md_free_lustre_md(sbi->ll_md_exp, &lmd);
529 ptlrpc_req_finished(request);
533 obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
534 #ifdef CONFIG_FS_POSIX_ACL
536 posix_acl_release(lmd.posix_acl);
537 lmd.posix_acl = NULL;
540 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
542 CERROR("lustre_lite: bad iget4 for root\n");
546 err = ll_close_thread_start(&sbi->ll_lcq);
548 CERROR("cannot start close thread: rc %d\n", err);
552 #ifdef CONFIG_FS_POSIX_ACL
553 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
554 rct_init(&sbi->ll_rct);
555 et_init(&sbi->ll_et);
559 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
560 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
561 KEY_CHECKSUM, sizeof(checksum), &checksum,
565 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
566 KEY_CACHE_SET, sizeof(sbi->ll_cache),
567 &sbi->ll_cache, NULL);
569 sb->s_root = d_make_root(root);
570 if (sb->s_root == NULL) {
571 CERROR("%s: can't make root dentry\n",
572 ll_get_fsname(sb, NULL, 0));
573 GOTO(out_root, err = -ENOMEM);
575 #ifdef HAVE_DCACHE_LOCK
576 sb->s_root->d_op = &ll_d_ops;
579 sbi->ll_sdev_orig = sb->s_dev;
581 /* We set sb->s_dev equal on all lustre clients in order to support
582 * NFS export clustering. NFSD requires that the FSID be the same
584 /* s_dev is also used in lt_compare() to compare two fs, but that is
585 * only a node-local comparison. */
586 uuid = obd_get_uuid(sbi->ll_md_exp);
588 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
594 if (proc_lustre_fs_root != NULL) {
595 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
598 CERROR("%s: could not register mount in lprocfs: "
599 "rc = %d\n", ll_get_fsname(sb, NULL, 0), err);
609 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
611 obd_disconnect(sbi->ll_dt_exp);
612 sbi->ll_dt_exp = NULL;
613 /* Make sure all OScs are gone, since cl_cache is accessing sbi. */
614 obd_zombie_barrier();
616 obd_fid_fini(sbi->ll_md_exp->exp_obd);
618 obd_disconnect(sbi->ll_md_exp);
619 sbi->ll_md_exp = NULL;
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 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
647 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
648 KEY_DEFAULT_EASIZE, &size, lmmsize, NULL);
650 CERROR("Get default mdsize error rc %d\n", rc);
655 int ll_get_max_cookiesize(struct ll_sb_info *sbi, int *lmmsize)
660 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_COOKIESIZE),
661 KEY_MAX_COOKIESIZE, &size, lmmsize, NULL);
663 CERROR("Get max cookiesize error rc %d\n", rc);
668 int ll_get_default_cookiesize(struct ll_sb_info *sbi, int *lmmsize)
673 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_COOKIESIZE),
674 KEY_DEFAULT_COOKIESIZE, &size, lmmsize, NULL);
676 CERROR("Get default cookiesize error rc %d\n", rc);
681 static void ll_dump_inode(struct inode *inode)
683 struct ll_d_hlist_node *tmp;
684 int dentry_count = 0;
686 LASSERT(inode != NULL);
688 ll_d_hlist_for_each(tmp, &inode->i_dentry)
691 CERROR("%s: inode %p dump: dev=%s fid="DFID
692 " mode=%o count=%u, %d dentries\n",
693 ll_get_fsname(inode->i_sb, NULL, 0), inode,
694 ll_i2mdexp(inode)->exp_obd->obd_name, PFID(ll_inode2fid(inode)),
695 inode->i_mode, atomic_read(&inode->i_count), dentry_count);
698 void lustre_dump_dentry(struct dentry *dentry, int recur)
700 struct list_head *tmp;
703 LASSERT(dentry != NULL);
705 list_for_each(tmp, &dentry->d_subdirs)
708 CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
709 " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
710 dentry->d_name.len, dentry->d_name.name,
711 dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
712 dentry->d_parent, dentry->d_inode, d_count(dentry),
713 dentry->d_flags, dentry->d_fsdata, subdirs);
714 if (dentry->d_inode != NULL)
715 ll_dump_inode(dentry->d_inode);
720 list_for_each(tmp, &dentry->d_subdirs) {
721 struct dentry *d = list_entry(tmp, struct dentry, d_u.d_child);
722 lustre_dump_dentry(d, recur - 1);
726 static void client_common_put_super(struct super_block *sb)
728 struct ll_sb_info *sbi = ll_s2sbi(sb);
731 #ifdef CONFIG_FS_POSIX_ACL
732 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
733 et_fini(&sbi->ll_et);
734 rct_fini(&sbi->ll_rct);
738 ll_close_thread_shutdown(sbi->ll_lcq);
742 list_del(&sbi->ll_conn_chain);
744 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
745 obd_disconnect(sbi->ll_dt_exp);
746 sbi->ll_dt_exp = NULL;
747 /* wait till all OSCs are gone, since cl_cache is accessing sbi.
749 obd_zombie_barrier();
751 lprocfs_unregister_mountpoint(sbi);
753 obd_fid_fini(sbi->ll_md_exp->exp_obd);
754 obd_disconnect(sbi->ll_md_exp);
755 sbi->ll_md_exp = NULL;
760 void ll_kill_super(struct super_block *sb)
762 struct ll_sb_info *sbi;
767 if (!(sb->s_flags & MS_ACTIVE))
771 /* we need restore s_dev from changed for clustred NFS before put_super
772 * because new kernels have cached s_dev and change sb->s_dev in
773 * put_super not affected real removing devices */
775 sb->s_dev = sbi->ll_sdev_orig;
776 sbi->ll_umounting = 1;
781 static inline int ll_set_opt(const char *opt, char *data, int fl)
783 if (strncmp(opt, data, strlen(opt)) != 0)
789 /* non-client-specific mount options are parsed in lmd_parse */
790 static int ll_options(char *options, int *flags)
793 char *s1 = options, *s2;
799 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
802 CDEBUG(D_SUPER, "next opt=%s\n", s1);
803 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
808 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
813 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
818 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
823 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
828 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
833 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 6, 51, 0)
834 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
836 /* Ignore deprecated mount option. The client will
837 * always try to mount with ACL support, whether this
838 * is used depends on whether server supports it. */
839 LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
840 "mount option 'acl'.\n");
843 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
845 LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
846 "mount option 'noacl'.\n");
850 #warning "{no}acl options have been deprecated since 1.8, please remove them"
852 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
857 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
862 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
868 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
873 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
878 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
883 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
888 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
893 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
898 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
903 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
908 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
913 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
918 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
924 s2 = strchr(s1, ',');
932 void ll_lli_init(struct ll_inode_info *lli)
934 lli->lli_inode_magic = LLI_INODE_MAGIC;
936 lli->lli_ioepoch = 0;
937 lli->lli_maxbytes = MAX_LFS_FILESIZE;
938 spin_lock_init(&lli->lli_lock);
939 lli->lli_posix_acl = NULL;
940 lli->lli_remote_perms = NULL;
941 mutex_init(&lli->lli_rmtperm_mutex);
942 /* Do not set lli_fid, it has been initialized already. */
943 fid_zero(&lli->lli_pfid);
944 INIT_LIST_HEAD(&lli->lli_close_list);
945 INIT_LIST_HEAD(&lli->lli_oss_capas);
946 atomic_set(&lli->lli_open_count, 0);
947 lli->lli_mds_capa = NULL;
948 lli->lli_rmtperm_time = 0;
949 lli->lli_pending_och = NULL;
950 lli->lli_mds_read_och = NULL;
951 lli->lli_mds_write_och = NULL;
952 lli->lli_mds_exec_och = NULL;
953 lli->lli_open_fd_read_count = 0;
954 lli->lli_open_fd_write_count = 0;
955 lli->lli_open_fd_exec_count = 0;
956 mutex_init(&lli->lli_och_mutex);
957 spin_lock_init(&lli->lli_agl_lock);
958 lli->lli_has_smd = false;
959 spin_lock_init(&lli->lli_layout_lock);
960 ll_layout_version_set(lli, LL_LAYOUT_GEN_NONE);
961 lli->lli_clob = NULL;
963 init_rwsem(&lli->lli_xattrs_list_rwsem);
964 mutex_init(&lli->lli_xattrs_enq_lock);
966 LASSERT(lli->lli_vfs_inode.i_mode != 0);
967 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
968 mutex_init(&lli->lli_readdir_mutex);
969 lli->lli_opendir_key = NULL;
971 spin_lock_init(&lli->lli_sa_lock);
972 lli->lli_opendir_pid = 0;
974 mutex_init(&lli->lli_size_mutex);
975 lli->lli_symlink_name = NULL;
976 init_rwsem(&lli->lli_trunc_sem);
977 mutex_init(&lli->lli_write_mutex);
978 init_rwsem(&lli->lli_glimpse_sem);
979 lli->lli_glimpse_time = 0;
980 INIT_LIST_HEAD(&lli->lli_agl_list);
981 lli->lli_agl_index = 0;
982 lli->lli_async_rc = 0;
984 mutex_init(&lli->lli_layout_mutex);
987 static inline int ll_bdi_register(struct backing_dev_info *bdi)
989 static atomic_t ll_bdi_num = ATOMIC_INIT(0);
991 bdi->name = "lustre";
992 return bdi_register(bdi, NULL, "lustre-%d",
993 atomic_inc_return(&ll_bdi_num));
996 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
998 struct lustre_profile *lprof = NULL;
999 struct lustre_sb_info *lsi = s2lsi(sb);
1000 struct ll_sb_info *sbi;
1001 char *dt = NULL, *md = NULL;
1002 char *profilenm = get_profile_name(sb);
1003 struct config_llog_instance *cfg;
1004 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1005 const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
1009 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1015 try_module_get(THIS_MODULE);
1017 /* client additional sb info */
1018 lsi->lsi_llsbi = sbi = ll_init_sbi();
1020 module_put(THIS_MODULE);
1025 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1027 GOTO(out_free, err);
1029 err = bdi_init(&lsi->lsi_bdi);
1031 GOTO(out_free, err);
1032 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1033 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1034 err = ll_bdi_register(&lsi->lsi_bdi);
1036 GOTO(out_free, err);
1038 sb->s_bdi = &lsi->lsi_bdi;
1039 #ifndef HAVE_DCACHE_LOCK
1040 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1041 sb->s_d_op = &ll_d_ops;
1044 /* Generate a string unique to this super, in case some joker tries
1045 to mount the same fs at two mount points.
1046 Use the address of the super itself.*/
1047 cfg->cfg_instance = sb;
1048 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1049 cfg->cfg_callback = class_config_llog_handler;
1050 /* set up client obds */
1051 err = lustre_process_log(sb, profilenm, cfg);
1053 GOTO(out_free, err);
1055 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1056 lprof = class_get_profile(profilenm);
1057 if (lprof == NULL) {
1058 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1059 " read from the MGS. Does that filesystem "
1060 "exist?\n", profilenm);
1061 GOTO(out_free, err = -EINVAL);
1063 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1064 lprof->lp_md, lprof->lp_dt);
1066 OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1068 GOTO(out_free, err = -ENOMEM);
1069 sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1071 OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1073 GOTO(out_free, err = -ENOMEM);
1074 sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1076 /* connections, registrations, sb setup */
1077 err = client_common_fill_super(sb, md, dt, mnt);
1081 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1083 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1086 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1087 LCONSOLE_WARN("Mounted %s\n", profilenm);
1091 } /* ll_fill_super */
1093 void ll_put_super(struct super_block *sb)
1095 struct config_llog_instance cfg, params_cfg;
1096 struct obd_device *obd;
1097 struct lustre_sb_info *lsi = s2lsi(sb);
1098 struct ll_sb_info *sbi = ll_s2sbi(sb);
1099 char *profilenm = get_profile_name(sb);
1100 int ccc_count, next, force = 1, rc = 0;
1103 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1105 ll_print_capa_stat(sbi);
1107 cfg.cfg_instance = sb;
1108 lustre_end_log(sb, profilenm, &cfg);
1110 params_cfg.cfg_instance = sb;
1111 lustre_end_log(sb, PARAMS_FILENAME, ¶ms_cfg);
1113 if (sbi->ll_md_exp) {
1114 obd = class_exp2obd(sbi->ll_md_exp);
1116 force = obd->obd_force;
1119 /* Wait for unstable pages to be committed to stable storage */
1121 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1122 rc = l_wait_event(sbi->ll_cache.ccc_unstable_waitq,
1123 atomic_read(&sbi->ll_cache.ccc_unstable_nr) == 0,
1127 ccc_count = atomic_read(&sbi->ll_cache.ccc_unstable_nr);
1128 if (force == 0 && rc != -EINTR)
1129 LASSERTF(ccc_count == 0, "count: %i\n", ccc_count);
1132 /* We need to set force before the lov_disconnect in
1133 lustre_common_put_super, since l_d cleans up osc's as well. */
1136 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1138 obd->obd_force = force;
1143 /* Only if client_common_fill_super succeeded */
1144 client_common_put_super(sb);
1148 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1149 class_manual_cleanup(obd);
1152 if (sbi->ll_flags & LL_SBI_VERBOSE)
1153 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1156 class_del_profile(profilenm);
1158 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1159 bdi_destroy(&lsi->lsi_bdi);
1160 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1164 lsi->lsi_llsbi = NULL;
1166 lustre_common_put_super(sb);
1168 cl_env_cache_purge(~0);
1170 module_put(THIS_MODULE);
1173 } /* client_put_super */
1175 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1177 struct inode *inode = NULL;
1179 /* NOTE: we depend on atomic igrab() -bzzz */
1180 lock_res_and_lock(lock);
1181 if (lock->l_resource->lr_lvb_inode) {
1182 struct ll_inode_info * lli;
1183 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1184 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1185 inode = igrab(lock->l_resource->lr_lvb_inode);
1187 inode = lock->l_resource->lr_lvb_inode;
1188 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1189 D_WARNING, lock, "lr_lvb_inode %p is "
1190 "bogus: magic %08x",
1191 lock->l_resource->lr_lvb_inode,
1192 lli->lli_inode_magic);
1196 unlock_res_and_lock(lock);
1200 static void ll_dir_clear_lsm_md(struct inode *inode)
1202 struct ll_inode_info *lli = ll_i2info(inode);
1204 LASSERT(S_ISDIR(inode->i_mode));
1206 if (lli->lli_lsm_md != NULL) {
1207 lmv_free_memmd(lli->lli_lsm_md);
1208 lli->lli_lsm_md = NULL;
1212 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1213 const struct lu_fid *fid,
1214 struct lustre_md *md)
1216 struct ll_sb_info *sbi = ll_s2sbi(sb);
1217 struct mdt_body *body = md->body;
1218 struct inode *inode;
1222 ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1223 inode = iget_locked(sb, ino);
1224 if (inode == NULL) {
1225 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1226 ll_get_fsname(sb, NULL, 0), PFID(fid));
1227 RETURN(ERR_PTR(-ENOENT));
1230 if (inode->i_state & I_NEW) {
1231 struct ll_inode_info *lli = ll_i2info(inode);
1232 struct lmv_stripe_md *lsm = md->lmv;
1234 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1235 (body->mode & S_IFMT);
1236 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1239 LTIME_S(inode->i_mtime) = 0;
1240 LTIME_S(inode->i_atime) = 0;
1241 LTIME_S(inode->i_ctime) = 0;
1244 /* initializing backing dev info. */
1245 inode->i_mapping->backing_dev_info =
1246 &s2lsi(inode->i_sb)->lsi_bdi;
1247 inode->i_op = &ll_dir_inode_operations;
1248 inode->i_fop = &ll_dir_operations;
1249 lli->lli_fid = *fid;
1252 LASSERT(lsm != NULL);
1253 /* master object FID */
1254 lli->lli_pfid = body->fid1;
1255 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1256 lli, PFID(fid), PFID(&lli->lli_pfid));
1257 unlock_new_inode(inode);
1263 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1266 struct lmv_stripe_md *lsm = md->lmv;
1269 LASSERT(lsm != NULL);
1270 /* XXX sigh, this lsm_root initialization should be in
1271 * LMV layer, but it needs ll_iget right now, so we
1272 * put this here right now. */
1273 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1274 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1275 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1276 /* Unfortunately ll_iget will call ll_update_inode,
1277 * where the initialization of slave inode is slightly
1278 * different, so it reset lsm_md to NULL to avoid
1279 * initializing lsm for slave inode. */
1280 /* For migrating inode, master stripe and master object will
1281 * be same, so we only need assign this inode */
1282 if (lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION && i == 0)
1283 lsm->lsm_md_oinfo[i].lmo_root = inode;
1285 lsm->lsm_md_oinfo[i].lmo_root =
1286 ll_iget_anon_dir(inode->i_sb, fid, md);
1288 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1289 int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1291 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1296 /* Here is where the lsm is being initialized(fill lmo_info) after
1297 * client retrieve MD stripe information from MDT. */
1298 return md_update_lsm_md(ll_i2mdexp(inode), lsm, md->body,
1299 ll_md_blocking_ast);
1302 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1303 const struct lmv_stripe_md *lsm_md2)
1305 return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1306 lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1307 lsm_md1->lsm_md_master_mdt_index ==
1308 lsm_md2->lsm_md_master_mdt_index &&
1309 lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1310 lsm_md1->lsm_md_layout_version ==
1311 lsm_md2->lsm_md_layout_version &&
1312 strcmp(lsm_md1->lsm_md_pool_name,
1313 lsm_md2->lsm_md_pool_name) == 0;
1316 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1318 struct ll_inode_info *lli = ll_i2info(inode);
1319 struct lmv_stripe_md *lsm = md->lmv;
1323 LASSERT(S_ISDIR(inode->i_mode));
1324 CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1325 PFID(ll_inode2fid(inode)));
1327 /* no striped information from request. */
1329 if (lli->lli_lsm_md == NULL) {
1331 } else if (lli->lli_lsm_md->lsm_md_hash_type &
1332 LMV_HASH_FLAG_MIGRATION) {
1333 /* migration is done, the temporay MIGRATE layout has
1335 CDEBUG(D_INODE, DFID" finish migration.\n",
1336 PFID(ll_inode2fid(inode)));
1337 lmv_free_memmd(lli->lli_lsm_md);
1338 lli->lli_lsm_md = NULL;
1341 /* The lustre_md from req does not include stripeEA,
1342 * see ll_md_setattr */
1347 /* set the directory layout */
1348 if (lli->lli_lsm_md == NULL) {
1350 rc = ll_init_lsm_md(inode, md);
1354 lli->lli_lsm_md = lsm;
1355 /* set lsm_md to NULL, so the following free lustre_md
1356 * will not free this lsm */
1358 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1359 lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1363 /* Compare the old and new stripe information */
1364 if (!lsm_md_eq(lli->lli_lsm_md, lsm)) {
1365 struct lmv_stripe_md *old_lsm = lli->lli_lsm_md;
1368 CERROR("%s: lmv layout mismatch "DFID"(%p)/"DFID"(%p)"
1369 "magic:0x%x/0x%x stripe count: %d/%d master_mdt: %d/%d"
1370 "hash_type:0x%x/0x%x layout: 0x%x/0x%x pool:%s/%s\n",
1371 ll_get_fsname(inode->i_sb, NULL, 0),
1372 PFID(&lsm->lsm_md_master_fid), lsm,
1373 PFID(&old_lsm->lsm_md_master_fid), old_lsm,
1374 lsm->lsm_md_magic, old_lsm->lsm_md_magic,
1375 lsm->lsm_md_stripe_count,
1376 old_lsm->lsm_md_stripe_count,
1377 lsm->lsm_md_master_mdt_index,
1378 old_lsm->lsm_md_master_mdt_index,
1379 lsm->lsm_md_hash_type, old_lsm->lsm_md_hash_type,
1380 lsm->lsm_md_layout_version,
1381 old_lsm->lsm_md_layout_version,
1382 lsm->lsm_md_pool_name,
1383 old_lsm->lsm_md_pool_name);
1385 for (idx = 0; idx < old_lsm->lsm_md_stripe_count; idx++) {
1386 CERROR("%s: sub FIDs in old lsm idx %d, old: "DFID"\n",
1387 ll_get_fsname(inode->i_sb, NULL, 0), idx,
1388 PFID(&old_lsm->lsm_md_oinfo[idx].lmo_fid));
1391 for (idx = 0; idx < lsm->lsm_md_stripe_count; idx++) {
1392 CERROR("%s: sub FIDs in new lsm idx %d, new: "DFID"\n",
1393 ll_get_fsname(inode->i_sb, NULL, 0), idx,
1394 PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1400 rc = md_update_lsm_md(ll_i2mdexp(inode), ll_i2info(inode)->lli_lsm_md,
1401 md->body, ll_md_blocking_ast);
1406 void ll_clear_inode(struct inode *inode)
1408 struct ll_inode_info *lli = ll_i2info(inode);
1409 struct ll_sb_info *sbi = ll_i2sbi(inode);
1412 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1413 PFID(ll_inode2fid(inode)), inode);
1415 if (S_ISDIR(inode->i_mode)) {
1416 /* these should have been cleared in ll_file_release */
1417 LASSERT(lli->lli_opendir_key == NULL);
1418 LASSERT(lli->lli_sai == NULL);
1419 LASSERT(lli->lli_opendir_pid == 0);
1422 spin_lock(&lli->lli_lock);
1423 ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1424 spin_unlock(&lli->lli_lock);
1425 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1427 LASSERT(!lli->lli_open_fd_write_count);
1428 LASSERT(!lli->lli_open_fd_read_count);
1429 LASSERT(!lli->lli_open_fd_exec_count);
1431 if (lli->lli_mds_write_och)
1432 ll_md_real_close(inode, FMODE_WRITE);
1433 if (lli->lli_mds_exec_och)
1434 ll_md_real_close(inode, FMODE_EXEC);
1435 if (lli->lli_mds_read_och)
1436 ll_md_real_close(inode, FMODE_READ);
1438 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1439 OBD_FREE(lli->lli_symlink_name,
1440 strlen(lli->lli_symlink_name) + 1);
1441 lli->lli_symlink_name = NULL;
1444 ll_xattr_cache_destroy(inode);
1446 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1447 LASSERT(lli->lli_posix_acl == NULL);
1448 if (lli->lli_remote_perms) {
1449 free_rmtperm_hash(lli->lli_remote_perms);
1450 lli->lli_remote_perms = NULL;
1453 #ifdef CONFIG_FS_POSIX_ACL
1454 else if (lli->lli_posix_acl) {
1455 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1456 LASSERT(lli->lli_remote_perms == NULL);
1457 posix_acl_release(lli->lli_posix_acl);
1458 lli->lli_posix_acl = NULL;
1461 lli->lli_inode_magic = LLI_INODE_DEAD;
1463 ll_clear_inode_capas(inode);
1464 if (S_ISDIR(inode->i_mode))
1465 ll_dir_clear_lsm_md(inode);
1466 else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1467 LASSERT(list_empty(&lli->lli_agl_list));
1470 * XXX This has to be done before lsm is freed below, because
1471 * cl_object still uses inode lsm.
1473 cl_inode_fini(inode);
1474 lli->lli_has_smd = false;
1479 int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data,
1480 struct md_open_data **mod)
1482 struct lustre_md md;
1483 struct inode *inode = dentry->d_inode;
1484 struct ll_sb_info *sbi = ll_i2sbi(inode);
1485 struct ptlrpc_request *request = NULL;
1489 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1490 LUSTRE_OPC_ANY, NULL);
1491 if (IS_ERR(op_data))
1492 RETURN(PTR_ERR(op_data));
1494 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1497 ptlrpc_req_finished(request);
1498 if (rc == -ENOENT) {
1500 /* Unlinked special device node? Or just a race?
1501 * Pretend we done everything. */
1502 if (!S_ISREG(inode->i_mode) &&
1503 !S_ISDIR(inode->i_mode)) {
1504 ia_valid = op_data->op_attr.ia_valid;
1505 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1506 rc = simple_setattr(dentry, &op_data->op_attr);
1507 op_data->op_attr.ia_valid = ia_valid;
1509 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1510 CERROR("md_setattr fails: rc = %d\n", rc);
1515 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1516 sbi->ll_md_exp, &md);
1518 ptlrpc_req_finished(request);
1522 ia_valid = op_data->op_attr.ia_valid;
1523 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1524 * cache is not cleared yet. */
1525 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1526 rc = simple_setattr(dentry, &op_data->op_attr);
1527 op_data->op_attr.ia_valid = ia_valid;
1529 /* Extract epoch data if obtained. */
1530 op_data->op_handle = md.body->handle;
1531 op_data->op_ioepoch = md.body->ioepoch;
1533 rc = ll_update_inode(inode, &md);
1534 ptlrpc_req_finished(request);
1539 /* Close IO epoch and send Size-on-MDS attribute update. */
1540 static int ll_setattr_done_writing(struct inode *inode,
1541 struct md_op_data *op_data,
1542 struct md_open_data *mod)
1544 struct ll_inode_info *lli = ll_i2info(inode);
1548 LASSERT(op_data != NULL);
1549 if (!S_ISREG(inode->i_mode))
1552 CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1553 op_data->op_ioepoch, PFID(&lli->lli_fid));
1555 op_data->op_flags = MF_EPOCH_CLOSE;
1556 ll_done_writing_attr(inode, op_data);
1557 ll_pack_inode2opdata(inode, op_data, NULL);
1559 rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1560 if (rc == -EAGAIN) {
1561 /* MDS has instructed us to obtain Size-on-MDS attribute
1562 * from OSTs and send setattr to back to MDS. */
1563 rc = ll_som_update(inode, op_data);
1565 CERROR("%s: inode "DFID" mdc truncate failed: rc = %d\n",
1566 ll_i2sbi(inode)->ll_md_exp->exp_obd->obd_name,
1567 PFID(ll_inode2fid(inode)), rc);
1572 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1574 struct obd_capa *capa;
1577 if (attr->ia_valid & ATTR_SIZE)
1578 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1580 capa = ll_mdscapa_get(inode);
1582 rc = cl_setattr_ost(inode, attr, capa);
1584 if (attr->ia_valid & ATTR_SIZE)
1585 ll_truncate_free_capa(capa);
1592 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1593 * object(s) determine the file size and mtime. Otherwise, the MDS will
1594 * keep these values until such a time that objects are allocated for it.
1595 * We do the MDS operations first, as it is checking permissions for us.
1596 * We don't to the MDS RPC if there is nothing that we want to store there,
1597 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1598 * going to do an RPC anyways.
1600 * If we are doing a truncate, we will send the mtime and ctime updates
1601 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1602 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1605 * In case of HSMimport, we only set attr on MDS.
1607 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1609 struct inode *inode = dentry->d_inode;
1610 struct ll_inode_info *lli = ll_i2info(inode);
1611 struct md_op_data *op_data = NULL;
1612 struct md_open_data *mod = NULL;
1613 bool file_is_released = false;
1614 int rc = 0, rc1 = 0;
1617 CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1618 "valid %x, hsm_import %d\n",
1619 ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1620 inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1623 if (attr->ia_valid & ATTR_SIZE) {
1624 /* Check new size against VFS/VM file size limit and rlimit */
1625 rc = inode_newsize_ok(inode, attr->ia_size);
1629 /* The maximum Lustre file size is variable, based on the
1630 * OST maximum object size and number of stripes. This
1631 * needs another check in addition to the VFS check above. */
1632 if (attr->ia_size > ll_file_maxbytes(inode)) {
1633 CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1634 PFID(&lli->lli_fid), attr->ia_size,
1635 ll_file_maxbytes(inode));
1639 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1642 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1643 if (attr->ia_valid & TIMES_SET_FLAGS) {
1644 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1645 !cfs_capable(CFS_CAP_FOWNER))
1649 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1650 if (attr->ia_valid & ATTR_CTIME) {
1651 attr->ia_ctime = CFS_CURRENT_TIME;
1652 attr->ia_valid |= ATTR_CTIME_SET;
1654 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1655 (attr->ia_valid & ATTR_ATIME)) {
1656 attr->ia_atime = CFS_CURRENT_TIME;
1657 attr->ia_valid |= ATTR_ATIME_SET;
1659 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1660 (attr->ia_valid & ATTR_MTIME)) {
1661 attr->ia_mtime = CFS_CURRENT_TIME;
1662 attr->ia_valid |= ATTR_MTIME_SET;
1665 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1666 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1667 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1668 cfs_time_current_sec());
1670 /* We always do an MDS RPC, even if we're only changing the size;
1671 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1673 OBD_ALLOC_PTR(op_data);
1674 if (op_data == NULL)
1677 if (!S_ISDIR(inode->i_mode)) {
1678 if (attr->ia_valid & ATTR_SIZE)
1679 inode_dio_write_done(inode);
1680 mutex_unlock(&inode->i_mutex);
1683 /* truncate on a released file must failed with -ENODATA,
1684 * so size must not be set on MDS for released file
1685 * but other attributes must be set
1687 if (S_ISREG(inode->i_mode)) {
1688 struct lov_stripe_md *lsm;
1691 ll_layout_refresh(inode, &gen);
1692 lsm = ccc_inode_lsm_get(inode);
1693 if (lsm && lsm->lsm_pattern & LOV_PATTERN_F_RELEASED)
1694 file_is_released = true;
1695 ccc_inode_lsm_put(inode, lsm);
1697 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1698 if (file_is_released) {
1699 rc = ll_layout_restore(inode, 0, attr->ia_size);
1703 file_is_released = false;
1704 ll_layout_refresh(inode, &gen);
1707 /* If we are changing file size, file content is
1708 * modified, flag it. */
1709 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1710 spin_lock(&lli->lli_lock);
1711 lli->lli_flags |= LLIF_DATA_MODIFIED;
1712 spin_unlock(&lli->lli_lock);
1713 op_data->op_bias |= MDS_DATA_MODIFIED;
1717 memcpy(&op_data->op_attr, attr, sizeof(*attr));
1719 /* Open epoch for truncate. */
1720 if (exp_connect_som(ll_i2mdexp(inode)) && !hsm_import &&
1721 (attr->ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1722 op_data->op_flags = MF_EPOCH_OPEN;
1724 rc = ll_md_setattr(dentry, op_data, &mod);
1728 /* RPC to MDT is sent, cancel data modification flag */
1729 if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) {
1730 spin_lock(&lli->lli_lock);
1731 lli->lli_flags &= ~LLIF_DATA_MODIFIED;
1732 spin_unlock(&lli->lli_lock);
1735 ll_ioepoch_open(lli, op_data->op_ioepoch);
1736 if (!S_ISREG(inode->i_mode) || file_is_released)
1739 if (attr->ia_valid & (ATTR_SIZE |
1740 ATTR_ATIME | ATTR_ATIME_SET |
1741 ATTR_MTIME | ATTR_MTIME_SET)) {
1742 /* For truncate and utimes sending attributes to OSTs, setting
1743 * mtime/atime to the past will be performed under PW [0:EOF]
1744 * extent lock (new_size:EOF for truncate). It may seem
1745 * excessive to send mtime/atime updates to OSTs when not
1746 * setting times to past, but it is necessary due to possible
1747 * time de-synchronization between MDT inode and OST objects */
1748 if (attr->ia_valid & ATTR_SIZE)
1749 down_write(&lli->lli_trunc_sem);
1750 rc = ll_setattr_ost(inode, attr);
1751 if (attr->ia_valid & ATTR_SIZE)
1752 up_write(&lli->lli_trunc_sem);
1757 if (op_data->op_ioepoch) {
1758 rc1 = ll_setattr_done_writing(inode, op_data, mod);
1762 ll_finish_md_op_data(op_data);
1764 if (!S_ISDIR(inode->i_mode)) {
1765 mutex_lock(&inode->i_mutex);
1766 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1767 inode_dio_wait(inode);
1770 ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1771 LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1776 int ll_setattr(struct dentry *de, struct iattr *attr)
1778 int mode = de->d_inode->i_mode;
1780 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1781 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1782 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1784 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1785 (ATTR_SIZE|ATTR_MODE)) &&
1786 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1787 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1788 !(attr->ia_mode & S_ISGID))))
1789 attr->ia_valid |= ATTR_FORCE;
1791 if ((mode & S_ISUID) &&
1792 !(attr->ia_mode & S_ISUID) &&
1793 !(attr->ia_valid & ATTR_KILL_SUID))
1794 attr->ia_valid |= ATTR_KILL_SUID;
1796 if (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1797 !(attr->ia_mode & S_ISGID) &&
1798 !(attr->ia_valid & ATTR_KILL_SGID))
1799 attr->ia_valid |= ATTR_KILL_SGID;
1801 return ll_setattr_raw(de, attr, false);
1804 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1805 __u64 max_age, __u32 flags)
1807 struct ll_sb_info *sbi = ll_s2sbi(sb);
1808 struct obd_statfs obd_osfs;
1812 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1814 CERROR("md_statfs fails: rc = %d\n", rc);
1818 osfs->os_type = sb->s_magic;
1820 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1821 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1823 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1824 flags |= OBD_STATFS_NODELAY;
1826 rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1828 CERROR("obd_statfs fails: rc = %d\n", rc);
1832 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1833 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1836 osfs->os_bsize = obd_osfs.os_bsize;
1837 osfs->os_blocks = obd_osfs.os_blocks;
1838 osfs->os_bfree = obd_osfs.os_bfree;
1839 osfs->os_bavail = obd_osfs.os_bavail;
1841 /* If we don't have as many objects free on the OST as inodes
1842 * on the MDS, we reduce the total number of inodes to
1843 * compensate, so that the "inodes in use" number is correct.
1845 if (obd_osfs.os_ffree < osfs->os_ffree) {
1846 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1848 osfs->os_ffree = obd_osfs.os_ffree;
1853 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1855 struct super_block *sb = de->d_sb;
1856 struct obd_statfs osfs;
1857 __u64 fsid = huge_encode_dev(sb->s_dev);
1860 CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1861 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1863 /* Some amount of caching on the client is allowed */
1864 rc = ll_statfs_internal(sb, &osfs,
1865 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1870 statfs_unpack(sfs, &osfs);
1872 /* We need to downshift for all 32-bit kernels, because we can't
1873 * tell if the kernel is being called via sys_statfs64() or not.
1874 * Stop before overflowing f_bsize - in which case it is better
1875 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1876 if (sizeof(long) < 8) {
1877 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1880 osfs.os_blocks >>= 1;
1881 osfs.os_bfree >>= 1;
1882 osfs.os_bavail >>= 1;
1886 sfs->f_blocks = osfs.os_blocks;
1887 sfs->f_bfree = osfs.os_bfree;
1888 sfs->f_bavail = osfs.os_bavail;
1889 sfs->f_fsid.val[0] = (__u32)fsid;
1890 sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1894 void ll_inode_size_lock(struct inode *inode)
1896 struct ll_inode_info *lli;
1898 LASSERT(!S_ISDIR(inode->i_mode));
1900 lli = ll_i2info(inode);
1901 mutex_lock(&lli->lli_size_mutex);
1904 void ll_inode_size_unlock(struct inode *inode)
1906 struct ll_inode_info *lli;
1908 lli = ll_i2info(inode);
1909 mutex_unlock(&lli->lli_size_mutex);
1912 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1914 struct ll_inode_info *lli = ll_i2info(inode);
1915 struct mdt_body *body = md->body;
1916 struct lov_stripe_md *lsm = md->lsm;
1917 struct ll_sb_info *sbi = ll_i2sbi(inode);
1919 LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1921 if (!lli->lli_has_smd &&
1922 !(sbi->ll_flags & LL_SBI_LAYOUT_LOCK))
1923 cl_file_inode_init(inode, md);
1925 lli->lli_maxbytes = lsm->lsm_maxbytes;
1926 if (lli->lli_maxbytes > MAX_LFS_FILESIZE)
1927 lli->lli_maxbytes = MAX_LFS_FILESIZE;
1930 if (S_ISDIR(inode->i_mode)) {
1933 rc = ll_update_lsm_md(inode, md);
1938 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1939 if (body->valid & OBD_MD_FLRMTPERM)
1940 ll_update_remote_perm(inode, md->remote_perm);
1942 #ifdef CONFIG_FS_POSIX_ACL
1943 else if (body->valid & OBD_MD_FLACL) {
1944 spin_lock(&lli->lli_lock);
1945 if (lli->lli_posix_acl)
1946 posix_acl_release(lli->lli_posix_acl);
1947 lli->lli_posix_acl = md->posix_acl;
1948 spin_unlock(&lli->lli_lock);
1951 inode->i_ino = cl_fid_build_ino(&body->fid1,
1952 sbi->ll_flags & LL_SBI_32BIT_API);
1953 inode->i_generation = cl_fid_build_gen(&body->fid1);
1955 if (body->valid & OBD_MD_FLATIME) {
1956 if (body->atime > LTIME_S(inode->i_atime))
1957 LTIME_S(inode->i_atime) = body->atime;
1958 lli->lli_lvb.lvb_atime = body->atime;
1960 if (body->valid & OBD_MD_FLMTIME) {
1961 if (body->mtime > LTIME_S(inode->i_mtime)) {
1962 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1963 "to "LPU64"\n", inode->i_ino,
1964 LTIME_S(inode->i_mtime), body->mtime);
1965 LTIME_S(inode->i_mtime) = body->mtime;
1967 lli->lli_lvb.lvb_mtime = body->mtime;
1969 if (body->valid & OBD_MD_FLCTIME) {
1970 if (body->ctime > LTIME_S(inode->i_ctime))
1971 LTIME_S(inode->i_ctime) = body->ctime;
1972 lli->lli_lvb.lvb_ctime = body->ctime;
1974 if (body->valid & OBD_MD_FLMODE)
1975 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1976 if (body->valid & OBD_MD_FLTYPE)
1977 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1978 LASSERT(inode->i_mode != 0);
1979 if (S_ISREG(inode->i_mode)) {
1980 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1982 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1984 if (body->valid & OBD_MD_FLUID)
1985 inode->i_uid = make_kuid(&init_user_ns, body->uid);
1986 if (body->valid & OBD_MD_FLGID)
1987 inode->i_gid = make_kgid(&init_user_ns, body->gid);
1988 if (body->valid & OBD_MD_FLFLAGS)
1989 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1990 if (body->valid & OBD_MD_FLNLINK)
1991 set_nlink(inode, body->nlink);
1992 if (body->valid & OBD_MD_FLRDEV)
1993 inode->i_rdev = old_decode_dev(body->rdev);
1995 if (body->valid & OBD_MD_FLID) {
1996 /* FID shouldn't be changed! */
1997 if (fid_is_sane(&lli->lli_fid)) {
1998 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1999 "Trying to change FID "DFID
2000 " to the "DFID", inode "DFID"(%p)\n",
2001 PFID(&lli->lli_fid), PFID(&body->fid1),
2002 PFID(ll_inode2fid(inode)), inode);
2004 lli->lli_fid = body->fid1;
2008 LASSERT(fid_seq(&lli->lli_fid) != 0);
2010 if (body->valid & OBD_MD_FLSIZE) {
2011 if (exp_connect_som(ll_i2mdexp(inode)) &&
2012 S_ISREG(inode->i_mode)) {
2013 struct lustre_handle lockh;
2016 /* As it is possible a blocking ast has been processed
2017 * by this time, we need to check there is an UPDATE
2018 * lock on the client and set LLIF_MDS_SIZE_LOCK holding
2020 mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
2021 &lockh, LDLM_FL_CBPENDING,
2025 if (lli->lli_flags & (LLIF_DONE_WRITING |
2026 LLIF_EPOCH_PENDING |
2028 CERROR("%s: inode "DFID" flags %u still"
2029 " has size authority! do not "
2030 "trust the size from MDS\n",
2031 sbi->ll_md_exp->exp_obd->obd_name,
2032 PFID(ll_inode2fid(inode)),
2035 /* Use old size assignment to avoid
2036 * deadlock bz14138 & bz14326 */
2037 i_size_write(inode, body->size);
2038 spin_lock(&lli->lli_lock);
2039 lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
2040 spin_unlock(&lli->lli_lock);
2042 ldlm_lock_decref(&lockh, mode);
2045 /* Use old size assignment to avoid
2046 * deadlock bz14138 & bz14326 */
2047 i_size_write(inode, body->size);
2050 "inode="DFID", updating i_size %llu\n",
2051 PFID(ll_inode2fid(inode)),
2052 (unsigned long long)body->size);
2055 if (body->valid & OBD_MD_FLBLOCKS)
2056 inode->i_blocks = body->blocks;
2059 if (body->valid & OBD_MD_FLMDSCAPA) {
2060 LASSERT(md->mds_capa);
2061 ll_add_capa(inode, md->mds_capa);
2063 if (body->valid & OBD_MD_FLOSSCAPA) {
2064 LASSERT(md->oss_capa);
2065 ll_add_capa(inode, md->oss_capa);
2068 if (body->valid & OBD_MD_TSTATE) {
2069 if (body->t_state & MS_RESTORE)
2070 lli->lli_flags |= LLIF_FILE_RESTORING;
2076 int ll_read_inode2(struct inode *inode, void *opaque)
2078 struct lustre_md *md = opaque;
2079 struct ll_inode_info *lli = ll_i2info(inode);
2083 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2084 PFID(&lli->lli_fid), inode);
2086 LASSERT(!lli->lli_has_smd);
2088 /* Core attributes from the MDS first. This is a new inode, and
2089 * the VFS doesn't zero times in the core inode so we have to do
2090 * it ourselves. They will be overwritten by either MDS or OST
2091 * attributes - we just need to make sure they aren't newer. */
2092 LTIME_S(inode->i_mtime) = 0;
2093 LTIME_S(inode->i_atime) = 0;
2094 LTIME_S(inode->i_ctime) = 0;
2096 rc = ll_update_inode(inode, md);
2100 /* OIDEBUG(inode); */
2102 /* initializing backing dev info. */
2103 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2106 if (S_ISREG(inode->i_mode)) {
2107 struct ll_sb_info *sbi = ll_i2sbi(inode);
2108 inode->i_op = &ll_file_inode_operations;
2109 inode->i_fop = sbi->ll_fop;
2110 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2112 } else if (S_ISDIR(inode->i_mode)) {
2113 inode->i_op = &ll_dir_inode_operations;
2114 inode->i_fop = &ll_dir_operations;
2116 } else if (S_ISLNK(inode->i_mode)) {
2117 inode->i_op = &ll_fast_symlink_inode_operations;
2120 inode->i_op = &ll_special_inode_operations;
2122 init_special_inode(inode, inode->i_mode,
2131 void ll_delete_inode(struct inode *inode)
2133 struct cl_inode_info *lli = cl_i2info(inode);
2136 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2137 /* discard all dirty pages before truncating them, required by
2138 * osc_extent implementation at LU-1030. */
2139 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF,
2140 CL_FSYNC_DISCARD, 1);
2142 truncate_inode_pages(&inode->i_data, 0);
2144 /* Workaround for LU-118 */
2145 if (inode->i_data.nrpages) {
2146 spin_lock_irq(&inode->i_data.tree_lock);
2147 spin_unlock_irq(&inode->i_data.tree_lock);
2148 LASSERTF(inode->i_data.nrpages == 0,
2149 "inode="DFID"(%p) nrpages=%lu, see "
2150 "http://jira.whamcloud.com/browse/LU-118\n",
2151 PFID(ll_inode2fid(inode)), inode,
2152 inode->i_data.nrpages);
2154 /* Workaround end */
2156 #ifdef HAVE_SBOPS_EVICT_INODE
2157 ll_clear_inode(inode);
2164 int ll_iocontrol(struct inode *inode, struct file *file,
2165 unsigned int cmd, unsigned long arg)
2167 struct ll_sb_info *sbi = ll_i2sbi(inode);
2168 struct ptlrpc_request *req = NULL;
2173 case FSFILT_IOC_GETFLAGS: {
2174 struct mdt_body *body;
2175 struct md_op_data *op_data;
2177 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2178 0, 0, LUSTRE_OPC_ANY,
2180 if (IS_ERR(op_data))
2181 RETURN(PTR_ERR(op_data));
2183 op_data->op_valid = OBD_MD_FLFLAGS;
2184 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2185 ll_finish_md_op_data(op_data);
2187 CERROR("%s: failure inode "DFID": rc = %d\n",
2188 sbi->ll_md_exp->exp_obd->obd_name,
2189 PFID(ll_inode2fid(inode)), rc);
2193 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2195 flags = body->flags;
2197 ptlrpc_req_finished(req);
2199 RETURN(put_user(flags, (int *)arg));
2201 case FSFILT_IOC_SETFLAGS: {
2202 struct lov_stripe_md *lsm;
2203 struct obd_info oinfo = { { { 0 } } };
2204 struct md_op_data *op_data;
2206 if (get_user(flags, (int *)arg))
2209 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2210 LUSTRE_OPC_ANY, NULL);
2211 if (IS_ERR(op_data))
2212 RETURN(PTR_ERR(op_data));
2214 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
2215 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2216 rc = md_setattr(sbi->ll_md_exp, op_data,
2217 NULL, 0, NULL, 0, &req, NULL);
2218 ll_finish_md_op_data(op_data);
2219 ptlrpc_req_finished(req);
2223 inode->i_flags = ll_ext_to_inode_flags(flags);
2225 lsm = ccc_inode_lsm_get(inode);
2226 if (!lsm_has_objects(lsm)) {
2227 ccc_inode_lsm_put(inode, lsm);
2231 OBDO_ALLOC(oinfo.oi_oa);
2233 ccc_inode_lsm_put(inode, lsm);
2237 oinfo.oi_oa->o_oi = lsm->lsm_oi;
2238 oinfo.oi_oa->o_flags = flags;
2239 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
2241 oinfo.oi_capa = ll_mdscapa_get(inode);
2242 obdo_set_parent_fid(oinfo.oi_oa, &ll_i2info(inode)->lli_fid);
2243 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
2244 capa_put(oinfo.oi_capa);
2245 OBDO_FREE(oinfo.oi_oa);
2246 ccc_inode_lsm_put(inode, lsm);
2248 if (rc && rc != -EPERM && rc != -EACCES)
2249 CERROR("osc_setattr_async fails: rc = %d\n", rc);
2260 int ll_flush_ctx(struct inode *inode)
2262 struct ll_sb_info *sbi = ll_i2sbi(inode);
2264 CDEBUG(D_SEC, "flush context for user %d\n",
2265 from_kuid(&init_user_ns, current_uid()));
2267 obd_set_info_async(NULL, sbi->ll_md_exp,
2268 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2270 obd_set_info_async(NULL, sbi->ll_dt_exp,
2271 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2276 /* umount -f client means force down, don't save state */
2277 void ll_umount_begin(struct super_block *sb)
2279 struct ll_sb_info *sbi = ll_s2sbi(sb);
2280 struct obd_device *obd;
2281 struct obd_ioctl_data *ioc_data;
2284 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2285 sb->s_count, atomic_read(&sb->s_active));
2287 obd = class_exp2obd(sbi->ll_md_exp);
2289 CERROR("Invalid MDC connection handle "LPX64"\n",
2290 sbi->ll_md_exp->exp_handle.h_cookie);
2296 obd = class_exp2obd(sbi->ll_dt_exp);
2298 CERROR("Invalid LOV connection handle "LPX64"\n",
2299 sbi->ll_dt_exp->exp_handle.h_cookie);
2305 OBD_ALLOC_PTR(ioc_data);
2307 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2308 sizeof *ioc_data, ioc_data, NULL);
2310 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2311 sizeof *ioc_data, ioc_data, NULL);
2313 OBD_FREE_PTR(ioc_data);
2316 /* Really, we'd like to wait until there are no requests outstanding,
2317 * and then continue. For now, we just invalidate the requests,
2318 * schedule() and sleep one second if needed, and hope.
2324 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2326 struct ll_sb_info *sbi = ll_s2sbi(sb);
2327 char *profilenm = get_profile_name(sb);
2331 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2332 read_only = *flags & MS_RDONLY;
2333 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2334 sizeof(KEY_READ_ONLY),
2335 KEY_READ_ONLY, sizeof(read_only),
2338 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2339 profilenm, read_only ?
2340 "read-only" : "read-write", err);
2345 sb->s_flags |= MS_RDONLY;
2347 sb->s_flags &= ~MS_RDONLY;
2349 if (sbi->ll_flags & LL_SBI_VERBOSE)
2350 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2351 read_only ? "read-only" : "read-write");
2356 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2357 struct super_block *sb, struct lookup_intent *it)
2359 struct ll_sb_info *sbi = NULL;
2360 struct lustre_md md = { 0 };
2364 LASSERT(*inode || sb);
2365 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2366 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2367 sbi->ll_md_exp, &md);
2372 rc = ll_update_inode(*inode, &md);
2376 LASSERT(sb != NULL);
2379 * At this point server returns to client's same fid as client
2380 * generated for creating. So using ->fid1 is okay here.
2382 LASSERT(fid_is_sane(&md.body->fid1));
2384 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1,
2385 sbi->ll_flags & LL_SBI_32BIT_API),
2387 if (IS_ERR(*inode)) {
2388 #ifdef CONFIG_FS_POSIX_ACL
2390 posix_acl_release(md.posix_acl);
2391 md.posix_acl = NULL;
2394 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2396 CERROR("new_inode -fatal: rc %d\n", rc);
2401 /* Handling piggyback layout lock.
2402 * Layout lock can be piggybacked by getattr and open request.
2403 * The lsm can be applied to inode only if it comes with a layout lock
2404 * otherwise correct layout may be overwritten, for example:
2405 * 1. proc1: mdt returns a lsm but not granting layout
2406 * 2. layout was changed by another client
2407 * 3. proc2: refresh layout and layout lock granted
2408 * 4. proc1: to apply a stale layout */
2409 if (it != NULL && it->d.lustre.it_lock_mode != 0) {
2410 struct lustre_handle lockh;
2411 struct ldlm_lock *lock;
2413 lockh.cookie = it->d.lustre.it_lock_handle;
2414 lock = ldlm_handle2lock(&lockh);
2415 LASSERT(lock != NULL);
2416 if (ldlm_has_layout(lock)) {
2417 struct cl_object_conf conf;
2419 memset(&conf, 0, sizeof(conf));
2420 conf.coc_opc = OBJECT_CONF_SET;
2421 conf.coc_inode = *inode;
2422 conf.coc_lock = lock;
2423 conf.u.coc_md = &md;
2424 (void)ll_layout_conf(*inode, &conf);
2426 LDLM_LOCK_PUT(lock);
2431 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2432 md_free_lustre_md(sbi->ll_md_exp, &md);
2436 int ll_obd_statfs(struct inode *inode, void *arg)
2438 struct ll_sb_info *sbi = NULL;
2439 struct obd_export *exp;
2441 struct obd_ioctl_data *data = NULL;
2446 if (!inode || !(sbi = ll_i2sbi(inode)))
2447 GOTO(out_statfs, rc = -EINVAL);
2449 rc = obd_ioctl_getdata(&buf, &len, arg);
2451 GOTO(out_statfs, rc);
2454 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2455 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2456 GOTO(out_statfs, rc = -EINVAL);
2458 if (data->ioc_inllen1 != sizeof(__u32) ||
2459 data->ioc_inllen2 != sizeof(__u32) ||
2460 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2461 data->ioc_plen2 != sizeof(struct obd_uuid))
2462 GOTO(out_statfs, rc = -EINVAL);
2464 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2465 if (type & LL_STATFS_LMV)
2466 exp = sbi->ll_md_exp;
2467 else if (type & LL_STATFS_LOV)
2468 exp = sbi->ll_dt_exp;
2470 GOTO(out_statfs, rc = -ENODEV);
2472 flags = (type & LL_STATFS_NODELAY) ? OBD_STATFS_NODELAY : 0;
2473 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, &flags);
2475 GOTO(out_statfs, rc);
2478 obd_ioctl_freedata(buf, len);
2482 int ll_process_config(struct lustre_cfg *lcfg)
2484 struct super_block *sb;
2489 /* The instance name contains the sb: lustre-client-aacfe000 */
2490 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2491 if (!ptr || !*(++ptr))
2493 if (sscanf(ptr, "%lx", &x) != 1)
2495 sb = (struct super_block *)x;
2496 /* This better be a real Lustre superblock! */
2497 LASSERT(s2lsi(sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2499 /* Note we have not called client_common_fill_super yet, so
2500 proc fns must be able to handle that! */
2501 rc = class_process_proc_seq_param(PARAM_LLITE, lprocfs_llite_obd_vars,
2508 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2509 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2510 struct inode *i1, struct inode *i2,
2511 const char *name, int namelen,
2512 int mode, __u32 opc, void *data)
2514 LASSERT(i1 != NULL);
2516 if (namelen > ll_i2sbi(i1)->ll_namelen)
2517 return ERR_PTR(-ENAMETOOLONG);
2519 if (op_data == NULL)
2520 OBD_ALLOC_PTR(op_data);
2522 if (op_data == NULL)
2523 return ERR_PTR(-ENOMEM);
2525 ll_i2gids(op_data->op_suppgids, i1, i2);
2526 op_data->op_fid1 = *ll_inode2fid(i1);
2527 op_data->op_capa1 = ll_mdscapa_get(i1);
2528 if (S_ISDIR(i1->i_mode))
2529 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2532 op_data->op_fid2 = *ll_inode2fid(i2);
2533 op_data->op_capa2 = ll_mdscapa_get(i2);
2534 if (S_ISDIR(i2->i_mode))
2535 op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2537 fid_zero(&op_data->op_fid2);
2538 op_data->op_capa2 = NULL;
2541 if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2542 op_data->op_cli_flags |= CLI_HASH64;
2544 if (ll_need_32bit_api(ll_i2sbi(i1)))
2545 op_data->op_cli_flags |= CLI_API32;
2547 op_data->op_name = name;
2548 op_data->op_namelen = namelen;
2549 op_data->op_mode = mode;
2550 op_data->op_mod_time = cfs_time_current_sec();
2551 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2552 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2553 op_data->op_cap = cfs_curproc_cap_pack();
2554 op_data->op_bias = 0;
2555 op_data->op_cli_flags = 0;
2556 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2557 filename_is_volatile(name, namelen, NULL))
2558 op_data->op_bias |= MDS_CREATE_VOLATILE;
2559 op_data->op_mds = 0;
2560 op_data->op_data = data;
2562 /* If the file is being opened after mknod() (normally due to NFS)
2563 * try to use the default stripe data from parent directory for
2564 * allocating OST objects. Try to pass the parent FID to MDS. */
2565 if (opc == LUSTRE_OPC_CREATE && i1 == i2 && S_ISREG(i2->i_mode) &&
2566 !ll_i2info(i2)->lli_has_smd) {
2567 struct ll_inode_info *lli = ll_i2info(i2);
2569 spin_lock(&lli->lli_lock);
2570 if (likely(!lli->lli_has_smd && !fid_is_zero(&lli->lli_pfid)))
2571 op_data->op_fid1 = lli->lli_pfid;
2572 spin_unlock(&lli->lli_lock);
2573 /** We ignore parent's capability temporary. */
2576 /* When called by ll_setattr_raw, file is i1. */
2577 if (LLIF_DATA_MODIFIED & ll_i2info(i1)->lli_flags)
2578 op_data->op_bias |= MDS_DATA_MODIFIED;
2583 void ll_finish_md_op_data(struct md_op_data *op_data)
2585 capa_put(op_data->op_capa1);
2586 capa_put(op_data->op_capa2);
2587 OBD_FREE_PTR(op_data);
2590 #ifdef HAVE_SUPEROPS_USE_DENTRY
2591 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2593 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2596 struct ll_sb_info *sbi;
2598 #ifdef HAVE_SUPEROPS_USE_DENTRY
2599 LASSERT((seq != NULL) && (dentry != NULL));
2600 sbi = ll_s2sbi(dentry->d_sb);
2602 LASSERT((seq != NULL) && (vfs != NULL));
2603 sbi = ll_s2sbi(vfs->mnt_sb);
2606 if (sbi->ll_flags & LL_SBI_NOLCK)
2607 seq_puts(seq, ",nolock");
2609 if (sbi->ll_flags & LL_SBI_FLOCK)
2610 seq_puts(seq, ",flock");
2612 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2613 seq_puts(seq, ",localflock");
2615 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2616 seq_puts(seq, ",user_xattr");
2618 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2619 seq_puts(seq, ",lazystatfs");
2621 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2622 seq_puts(seq, ",user_fid2path");
2628 * Get obd name by cmd, and copy out to user space
2630 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2632 struct ll_sb_info *sbi = ll_i2sbi(inode);
2633 struct obd_device *obd;
2636 if (cmd == OBD_IOC_GETDTNAME)
2637 obd = class_exp2obd(sbi->ll_dt_exp);
2638 else if (cmd == OBD_IOC_GETMDNAME)
2639 obd = class_exp2obd(sbi->ll_md_exp);
2646 if (copy_to_user((void *)arg, obd->obd_name,
2647 strlen(obd->obd_name) + 1))
2654 * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2655 * fsname will be returned in this buffer; otherwise, a static buffer will be
2656 * used to store the fsname and returned to caller.
2658 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2660 static char fsname_static[MTI_NAME_MAXLEN];
2661 struct lustre_sb_info *lsi = s2lsi(sb);
2666 /* this means the caller wants to use static buffer
2667 * and it doesn't care about race. Usually this is
2668 * in error reporting path */
2669 buf = fsname_static;
2670 buflen = sizeof(fsname_static);
2673 len = strlen(lsi->lsi_lmd->lmd_profile);
2674 ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2675 if (ptr && (strcmp(ptr, "-client") == 0))
2678 if (unlikely(len >= buflen))
2680 strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2686 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2693 p.mnt = current->fs->root.mnt;
2695 path = d_path(&p, buf, bufsize);
2700 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2702 char *buf, *path = NULL;
2703 struct dentry *dentry = NULL;
2704 struct ccc_object *obj = cl_inode2ccc(page->mapping->host);
2706 /* this can be called inside spin lock so use GFP_ATOMIC. */
2707 buf = (char *)__get_free_page(GFP_ATOMIC);
2709 dentry = d_find_alias(page->mapping->host);
2711 path = ll_d_path(dentry, buf, PAGE_SIZE);
2715 "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2716 "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2717 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2718 PFID(&obj->cob_header.coh_lu.loh_fid),
2719 (path && !IS_ERR(path)) ? path : "", ioret);
2725 free_page((unsigned long)buf);