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.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2016, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lustre/llite/llite_lib.c
34 * Lustre Light Super operations
37 #define DEBUG_SUBSYSTEM S_LLITE
39 #include <linux/module.h>
40 #include <linux/statfs.h>
41 #include <linux/time.h>
42 #include <linux/types.h>
43 #include <linux/version.h>
45 #include <linux/user_namespace.h>
46 #ifdef HAVE_UIDGID_HEADER
47 # include <linux/uidgid.h>
49 #include <linux/security.h>
51 #include <uapi/linux/lustre_ioctl.h>
52 #include <lustre_ha.h>
53 #include <lustre_dlm.h>
54 #include <lprocfs_status.h>
55 #include <lustre_disk.h>
56 #include <uapi/linux/lustre_param.h>
57 #include <lustre_log.h>
58 #include <cl_object.h>
59 #include <obd_cksum.h>
60 #include "llite_internal.h"
62 struct kmem_cache *ll_file_data_slab;
65 #define log2(n) ffz(~(n))
68 static struct ll_sb_info *ll_init_sbi(void)
70 struct ll_sb_info *sbi = NULL;
72 unsigned long lru_page_max;
82 spin_lock_init(&sbi->ll_lock);
83 mutex_init(&sbi->ll_lco.lco_lock);
84 spin_lock_init(&sbi->ll_pp_extent_lock);
85 spin_lock_init(&sbi->ll_process_lock);
86 sbi->ll_rw_stats_on = 0;
89 pages = si.totalram - si.totalhigh;
90 lru_page_max = pages / 2;
92 /* initialize ll_cache data */
93 sbi->ll_cache = cl_cache_init(lru_page_max);
94 if (sbi->ll_cache == NULL) {
95 OBD_FREE(sbi, sizeof(*sbi));
99 sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
100 SBI_DEFAULT_READAHEAD_MAX);
101 sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
102 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
103 SBI_DEFAULT_READAHEAD_WHOLE_MAX;
105 ll_generate_random_uuid(uuid);
106 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
107 CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
109 sbi->ll_flags |= LL_SBI_VERBOSE;
110 #ifdef ENABLE_CHECKSUM
111 sbi->ll_flags |= LL_SBI_CHECKSUM;
114 #ifdef HAVE_LRU_RESIZE_SUPPORT
115 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
117 sbi->ll_flags |= LL_SBI_LAZYSTATFS;
119 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
120 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
122 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
126 /* metadata statahead is enabled by default */
127 sbi->ll_sa_max = LL_SA_RPC_DEF;
128 atomic_set(&sbi->ll_sa_total, 0);
129 atomic_set(&sbi->ll_sa_wrong, 0);
130 atomic_set(&sbi->ll_sa_running, 0);
131 atomic_set(&sbi->ll_agl_total, 0);
132 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
133 sbi->ll_flags |= LL_SBI_FAST_READ;
136 sbi->ll_squash.rsi_uid = 0;
137 sbi->ll_squash.rsi_gid = 0;
138 INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
139 init_rwsem(&sbi->ll_squash.rsi_sem);
144 static void ll_free_sbi(struct super_block *sb)
146 struct ll_sb_info *sbi = ll_s2sbi(sb);
150 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
151 cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
152 if (sbi->ll_cache != NULL) {
153 cl_cache_decref(sbi->ll_cache);
154 sbi->ll_cache = NULL;
156 OBD_FREE(sbi, sizeof(*sbi));
161 static inline int obd_connect_has_secctx(struct obd_connect_data *data)
163 return data->ocd_connect_flags & OBD_CONNECT_FLAGS2 &&
164 data->ocd_connect_flags2 & OBD_CONNECT2_FILE_SECCTX;
167 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
168 struct vfsmount *mnt)
170 struct inode *root = NULL;
171 struct ll_sb_info *sbi = ll_s2sbi(sb);
172 struct obd_device *obd;
173 struct obd_statfs *osfs = NULL;
174 struct ptlrpc_request *request = NULL;
175 struct obd_connect_data *data = NULL;
176 struct obd_uuid *uuid;
177 struct md_op_data *op_data;
178 struct lustre_md lmd;
180 int size, err, checksum;
183 obd = class_name2obd(md);
185 CERROR("MD %s: not setup or attached\n", md);
199 /* indicate the features supported by this client */
200 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
201 OBD_CONNECT_ATTRFID |
202 OBD_CONNECT_VERSION | OBD_CONNECT_BRW_SIZE |
203 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
204 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
205 OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 |
206 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
207 OBD_CONNECT_64BITHASH |
208 OBD_CONNECT_EINPROGRESS |
209 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
210 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS|
211 OBD_CONNECT_MAX_EASIZE |
212 OBD_CONNECT_FLOCK_DEAD |
213 OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
214 OBD_CONNECT_OPEN_BY_FID |
215 OBD_CONNECT_DIR_STRIPE |
216 OBD_CONNECT_BULK_MBITS |
217 OBD_CONNECT_SUBTREE |
218 OBD_CONNECT_FLAGS2 | OBD_CONNECT_MULTIMODRPCS;
220 data->ocd_connect_flags2 = 0;
222 #ifdef HAVE_LRU_RESIZE_SUPPORT
223 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
224 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
226 #ifdef CONFIG_FS_POSIX_ACL
227 data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK |
228 OBD_CONNECT_LARGE_ACL;
231 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
232 /* flag mdc connection as lightweight, only used for test
233 * purpose, use with care */
234 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
236 data->ocd_ibits_known = MDS_INODELOCK_FULL;
237 data->ocd_version = LUSTRE_VERSION_CODE;
239 if (sb->s_flags & MS_RDONLY)
240 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
241 if (sbi->ll_flags & LL_SBI_USER_XATTR)
242 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
245 /* Setting this indicates we correctly support S_NOSEC (See kernel
246 * commit 9e1f1de02c2275d7172e18dc4e7c2065777611bf)
248 sb->s_flags |= MS_NOSEC;
251 if (sbi->ll_flags & LL_SBI_FLOCK)
252 sbi->ll_fop = &ll_file_operations_flock;
253 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
254 sbi->ll_fop = &ll_file_operations;
256 sbi->ll_fop = &ll_file_operations_noflock;
258 /* always ping even if server suppress_pings */
259 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
260 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
262 #ifdef HAVE_SECURITY_DENTRY_INIT_SECURITY
263 data->ocd_connect_flags2 |= OBD_CONNECT2_FILE_SECCTX;
264 #endif /* HAVE_SECURITY_DENTRY_INIT_SECURITY */
266 data->ocd_brw_size = MD_MAX_BRW_SIZE;
268 err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
270 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
271 "recovery, of which this client is not a "
272 "part. Please wait for recovery to complete,"
273 " abort, or time out.\n", md);
276 CERROR("cannot connect to %s: rc = %d\n", md, err);
280 sbi->ll_md_exp->exp_connect_data = *data;
282 err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
283 LUSTRE_SEQ_METADATA);
285 CERROR("%s: Can't init metadata layer FID infrastructure, "
286 "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
290 /* For mount, we only need fs info from MDT0, and also in DNE, it
291 * can make sure the client can be mounted as long as MDT0 is
293 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
294 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
295 OBD_STATFS_FOR_MDT0);
297 GOTO(out_md_fid, err);
299 /* This needs to be after statfs to ensure connect has finished.
300 * Note that "data" does NOT contain the valid connect reply.
301 * If connecting to a 1.8 server there will be no LMV device, so
302 * we can access the MDC export directly and exp_connect_flags will
303 * be non-zero, but if accessing an upgraded 2.1 server it will
304 * have the correct flags filled in.
305 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
306 valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
307 if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
308 valid != CLIENT_CONNECT_MDT_REQD) {
311 OBD_ALLOC_WAIT(buf, PAGE_SIZE);
312 obd_connect_flags2str(buf, PAGE_SIZE,
313 valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
314 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
315 "feature(s) needed for correct operation "
316 "of this client (%s). Please upgrade "
317 "server or downgrade client.\n",
318 sbi->ll_md_exp->exp_obd->obd_name, buf);
319 OBD_FREE(buf, PAGE_SIZE);
320 GOTO(out_md_fid, err = -EPROTO);
323 size = sizeof(*data);
324 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
325 KEY_CONN_DATA, &size, data);
327 CERROR("%s: Get connect data failed: rc = %d\n",
328 sbi->ll_md_exp->exp_obd->obd_name, err);
329 GOTO(out_md_fid, err);
332 LASSERT(osfs->os_bsize);
333 sb->s_blocksize = osfs->os_bsize;
334 sb->s_blocksize_bits = log2(osfs->os_bsize);
335 sb->s_magic = LL_SUPER_MAGIC;
336 sb->s_maxbytes = MAX_LFS_FILESIZE;
337 sbi->ll_namelen = osfs->os_namelen;
338 sbi->ll_mnt.mnt = current->fs->root.mnt;
340 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
341 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
342 LCONSOLE_INFO("Disabling user_xattr feature because "
343 "it is not supported on the server\n");
344 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
347 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
349 sb->s_flags |= MS_POSIXACL;
351 sbi->ll_flags |= LL_SBI_ACL;
353 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
355 sb->s_flags &= ~MS_POSIXACL;
357 sbi->ll_flags &= ~LL_SBI_ACL;
360 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
361 sbi->ll_flags |= LL_SBI_64BIT_HASH;
363 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
364 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
366 if (obd_connect_has_secctx(data))
367 sbi->ll_flags |= LL_SBI_FILE_SECCTX;
369 if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
370 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
371 LCONSOLE_INFO("%s: disabling xattr cache due to "
372 "unknown maximum xattr size.\n", dt);
373 } else if (!sbi->ll_xattr_cache_set) {
374 /* If xattr_cache is already set (no matter 0 or 1)
375 * during processing llog, it won't be enabled here. */
376 sbi->ll_flags |= LL_SBI_XATTR_CACHE;
377 sbi->ll_xattr_cache_enabled = 1;
381 obd = class_name2obd(dt);
383 CERROR("DT %s: not setup or attached\n", dt);
384 GOTO(out_md_fid, err = -ENODEV);
387 /* pass client page size via ocd_grant_blkbits, the server should report
388 * back its backend blocksize for grant calculation purpose */
389 data->ocd_grant_blkbits = PAGE_SHIFT;
391 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
392 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
393 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
394 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
395 OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
396 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
397 OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
398 OBD_CONNECT_EINPROGRESS |
399 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
400 OBD_CONNECT_LAYOUTLOCK |
401 OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
402 OBD_CONNECT_BULK_MBITS;
404 data->ocd_connect_flags2 = 0;
406 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
407 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
409 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
410 * disabled by default, because it can still be enabled on the
411 * fly via /proc. As a consequence, we still need to come to an
412 * agreement on the supported algorithms at connect time */
413 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
415 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
416 data->ocd_cksum_types = OBD_CKSUM_ADLER;
418 data->ocd_cksum_types = cksum_types_supported_client();
420 #ifdef HAVE_LRU_RESIZE_SUPPORT
421 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
423 /* always ping even if server suppress_pings */
424 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
425 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
427 CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
428 "ocd_grant: %d\n", data->ocd_connect_flags,
429 data->ocd_version, data->ocd_grant);
431 obd->obd_upcall.onu_owner = &sbi->ll_lco;
432 obd->obd_upcall.onu_upcall = cl_ocd_update;
434 data->ocd_brw_size = DT_MAX_BRW_SIZE;
436 err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
439 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
440 "recovery, of which this client is not a "
441 "part. Please wait for recovery to "
442 "complete, abort, or time out.\n", dt);
445 CERROR("%s: Cannot connect to %s: rc = %d\n",
446 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
450 sbi->ll_dt_exp->exp_connect_data = *data;
452 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
453 LUSTRE_SEQ_METADATA);
455 CERROR("%s: Can't init data layer FID infrastructure, "
456 "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
460 mutex_lock(&sbi->ll_lco.lco_lock);
461 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
462 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
463 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
464 mutex_unlock(&sbi->ll_lco.lco_lock);
466 fid_zero(&sbi->ll_root_fid);
467 err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
470 CERROR("cannot mds_connect: rc = %d\n", err);
471 GOTO(out_lock_cn_cb, err);
473 if (!fid_is_sane(&sbi->ll_root_fid)) {
474 CERROR("%s: Invalid root fid "DFID" during mount\n",
475 sbi->ll_md_exp->exp_obd->obd_name,
476 PFID(&sbi->ll_root_fid));
477 GOTO(out_lock_cn_cb, err = -EINVAL);
479 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
481 sb->s_op = &lustre_super_operations;
482 #if THREAD_SIZE >= 8192 /*b=17630*/
483 sb->s_export_op = &lustre_export_operations;
487 * XXX: move this to after cbd setup? */
488 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
489 if (sbi->ll_flags & LL_SBI_ACL)
490 valid |= OBD_MD_FLACL;
492 OBD_ALLOC_PTR(op_data);
494 GOTO(out_lock_cn_cb, err = -ENOMEM);
496 op_data->op_fid1 = sbi->ll_root_fid;
497 op_data->op_mode = 0;
498 op_data->op_valid = valid;
500 err = md_getattr(sbi->ll_md_exp, op_data, &request);
502 OBD_FREE_PTR(op_data);
504 CERROR("%s: md_getattr failed for root: rc = %d\n",
505 sbi->ll_md_exp->exp_obd->obd_name, err);
506 GOTO(out_lock_cn_cb, err);
509 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
510 sbi->ll_md_exp, &lmd);
512 CERROR("failed to understand root inode md: rc = %d\n", err);
513 ptlrpc_req_finished(request);
514 GOTO(out_lock_cn_cb, err);
517 LASSERT(fid_is_sane(&sbi->ll_root_fid));
518 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
519 sbi->ll_flags & LL_SBI_32BIT_API),
521 md_free_lustre_md(sbi->ll_md_exp, &lmd);
522 ptlrpc_req_finished(request);
525 #ifdef CONFIG_FS_POSIX_ACL
527 posix_acl_release(lmd.posix_acl);
528 lmd.posix_acl = NULL;
531 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
533 CERROR("lustre_lite: bad iget4 for root\n");
537 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
538 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
539 KEY_CHECKSUM, sizeof(checksum), &checksum,
542 CERROR("%s: Set checksum failed: rc = %d\n",
543 sbi->ll_dt_exp->exp_obd->obd_name, err);
548 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
549 KEY_CACHE_SET, sizeof(*sbi->ll_cache),
550 sbi->ll_cache, NULL);
552 CERROR("%s: Set cache_set failed: rc = %d\n",
553 sbi->ll_dt_exp->exp_obd->obd_name, err);
557 sb->s_root = d_make_root(root);
558 if (sb->s_root == NULL) {
559 CERROR("%s: can't make root dentry\n",
560 ll_get_fsname(sb, NULL, 0));
561 GOTO(out_root, err = -ENOMEM);
563 #ifdef HAVE_DCACHE_LOCK
564 sb->s_root->d_op = &ll_d_ops;
567 sbi->ll_sdev_orig = sb->s_dev;
569 /* We set sb->s_dev equal on all lustre clients in order to support
570 * NFS export clustering. NFSD requires that the FSID be the same
572 /* s_dev is also used in lt_compare() to compare two fs, but that is
573 * only a node-local comparison. */
574 uuid = obd_get_uuid(sbi->ll_md_exp);
576 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
583 if (sbi->ll_proc_root != NULL) {
584 err = lprocfs_ll_register_obd(sb, dt);
586 CERROR("%s: could not register %s in llite: rc = %d\n",
587 dt, ll_get_fsname(sb, NULL, 0), err);
590 err = lprocfs_ll_register_obd(sb, md);
592 CERROR("%s: could not register %s in llite: rc = %d\n",
593 md, ll_get_fsname(sb, NULL, 0), err);
603 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
605 obd_disconnect(sbi->ll_dt_exp);
606 sbi->ll_dt_exp = NULL;
608 obd_fid_fini(sbi->ll_md_exp->exp_obd);
610 obd_disconnect(sbi->ll_md_exp);
611 sbi->ll_md_exp = NULL;
620 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
624 size = sizeof(*lmmsize);
625 rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
626 KEY_MAX_EASIZE, &size, lmmsize);
628 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
629 sbi->ll_dt_exp->exp_obd->obd_name, rc);
634 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
635 KEY_MAX_EASIZE, &size, lmmsize);
637 CERROR("Get max mdsize error rc %d\n", rc);
643 * Get the value of the default_easize parameter.
645 * \see client_obd::cl_default_mds_easize
647 * \param[in] sbi superblock info for this filesystem
648 * \param[out] lmmsize pointer to storage location for value
650 * \retval 0 on success
651 * \retval negative negated errno on failure
653 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
658 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
659 KEY_DEFAULT_EASIZE, &size, lmmsize);
661 CERROR("Get default mdsize error rc %d\n", rc);
667 * Set the default_easize parameter to the given value.
669 * \see client_obd::cl_default_mds_easize
671 * \param[in] sbi superblock info for this filesystem
672 * \param[in] lmmsize the size to set
674 * \retval 0 on success
675 * \retval negative negated errno on failure
677 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
681 if (lmmsize < sizeof(struct lov_mds_md) ||
682 lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
685 rc = obd_set_info_async(NULL, sbi->ll_md_exp,
686 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
687 sizeof(int), &lmmsize, NULL);
692 static void client_common_put_super(struct super_block *sb)
694 struct ll_sb_info *sbi = ll_s2sbi(sb);
699 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
700 obd_disconnect(sbi->ll_dt_exp);
701 sbi->ll_dt_exp = NULL;
703 lprocfs_ll_unregister_mountpoint(sbi);
705 obd_fid_fini(sbi->ll_md_exp->exp_obd);
706 obd_disconnect(sbi->ll_md_exp);
707 sbi->ll_md_exp = NULL;
712 void ll_kill_super(struct super_block *sb)
714 struct ll_sb_info *sbi;
718 if (!(sb->s_flags & MS_ACTIVE))
722 /* we need restore s_dev from changed for clustred NFS before put_super
723 * because new kernels have cached s_dev and change sb->s_dev in
724 * put_super not affected real removing devices */
726 sb->s_dev = sbi->ll_sdev_orig;
727 sbi->ll_umounting = 1;
729 /* wait running statahead threads to quit */
730 while (atomic_read(&sbi->ll_sa_running) > 0) {
731 set_current_state(TASK_UNINTERRUPTIBLE);
732 schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
739 static inline int ll_set_opt(const char *opt, char *data, int fl)
741 if (strncmp(opt, data, strlen(opt)) != 0)
747 /* non-client-specific mount options are parsed in lmd_parse */
748 static int ll_options(char *options, int *flags)
751 char *s1 = options, *s2;
757 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
760 CDEBUG(D_SUPER, "next opt=%s\n", s1);
761 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
766 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
771 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
776 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
781 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
786 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
791 tmp = ll_set_opt("context", s1, 1);
794 tmp = ll_set_opt("fscontext", s1, 1);
797 tmp = ll_set_opt("defcontext", s1, 1);
800 tmp = ll_set_opt("rootcontext", s1, 1);
803 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
808 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
814 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
819 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
824 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
829 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
834 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
839 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
844 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
849 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
854 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
859 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
864 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
870 s2 = strchr(s1, ',');
878 void ll_lli_init(struct ll_inode_info *lli)
880 lli->lli_inode_magic = LLI_INODE_MAGIC;
882 spin_lock_init(&lli->lli_lock);
883 lli->lli_posix_acl = NULL;
884 /* Do not set lli_fid, it has been initialized already. */
885 fid_zero(&lli->lli_pfid);
886 lli->lli_mds_read_och = NULL;
887 lli->lli_mds_write_och = NULL;
888 lli->lli_mds_exec_och = NULL;
889 lli->lli_open_fd_read_count = 0;
890 lli->lli_open_fd_write_count = 0;
891 lli->lli_open_fd_exec_count = 0;
892 mutex_init(&lli->lli_och_mutex);
893 spin_lock_init(&lli->lli_agl_lock);
894 spin_lock_init(&lli->lli_layout_lock);
895 ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
896 lli->lli_clob = NULL;
898 init_rwsem(&lli->lli_xattrs_list_rwsem);
899 mutex_init(&lli->lli_xattrs_enq_lock);
901 LASSERT(lli->lli_vfs_inode.i_mode != 0);
902 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
903 mutex_init(&lli->lli_readdir_mutex);
904 lli->lli_opendir_key = NULL;
906 spin_lock_init(&lli->lli_sa_lock);
907 lli->lli_opendir_pid = 0;
908 lli->lli_sa_enabled = 0;
909 lli->lli_def_stripe_offset = -1;
911 mutex_init(&lli->lli_size_mutex);
912 lli->lli_symlink_name = NULL;
913 init_rwsem(&lli->lli_trunc_sem);
914 range_lock_tree_init(&lli->lli_write_tree);
915 init_rwsem(&lli->lli_glimpse_sem);
916 lli->lli_glimpse_time = 0;
917 INIT_LIST_HEAD(&lli->lli_agl_list);
918 lli->lli_agl_index = 0;
919 lli->lli_async_rc = 0;
921 mutex_init(&lli->lli_layout_mutex);
922 memset(lli->lli_jobid, 0, LUSTRE_JOBID_SIZE);
925 static inline int ll_bdi_register(struct backing_dev_info *bdi)
927 static atomic_t ll_bdi_num = ATOMIC_INIT(0);
929 bdi->name = "lustre";
930 return bdi_register(bdi, NULL, "lustre-%d",
931 atomic_inc_return(&ll_bdi_num));
934 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
936 struct lustre_profile *lprof = NULL;
937 struct lustre_sb_info *lsi = s2lsi(sb);
938 struct ll_sb_info *sbi;
939 char *dt = NULL, *md = NULL;
940 char *profilenm = get_profile_name(sb);
941 struct config_llog_instance *cfg;
942 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
943 const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
949 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
955 try_module_get(THIS_MODULE);
957 /* client additional sb info */
958 lsi->lsi_llsbi = sbi = ll_init_sbi();
960 module_put(THIS_MODULE);
965 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
969 err = bdi_init(&lsi->lsi_bdi);
972 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
973 #ifdef HAVE_BDI_CAP_MAP_COPY
974 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
976 lsi->lsi_bdi.capabilities = 0;
978 err = ll_bdi_register(&lsi->lsi_bdi);
982 sb->s_bdi = &lsi->lsi_bdi;
983 #ifndef HAVE_DCACHE_LOCK
984 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
985 sb->s_d_op = &ll_d_ops;
988 /* Call lprocfs_ll_register_mountpoint() before lustre_process_log()
989 * so that "llite.*.*" params can be processed correctly. */
990 if (proc_lustre_fs_root != NULL) {
991 err = lprocfs_ll_register_mountpoint(proc_lustre_fs_root, sb);
993 CERROR("%s: could not register mountpoint in llite: "
994 "rc = %d\n", ll_get_fsname(sb, NULL, 0), err);
999 /* Generate a string unique to this super, in case some joker tries
1000 to mount the same fs at two mount points.
1001 Use the address of the super itself.*/
1002 cfg->cfg_instance = sb;
1003 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1004 cfg->cfg_callback = class_config_llog_handler;
1005 cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1006 /* set up client obds */
1007 err = lustre_process_log(sb, profilenm, cfg);
1009 GOTO(out_proc, err);
1011 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1012 lprof = class_get_profile(profilenm);
1013 if (lprof == NULL) {
1014 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1015 " read from the MGS. Does that filesystem "
1016 "exist?\n", profilenm);
1017 GOTO(out_proc, err = -EINVAL);
1019 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1020 lprof->lp_md, lprof->lp_dt);
1022 dt_len = strlen(lprof->lp_dt) + instlen + 2;
1023 OBD_ALLOC(dt, dt_len);
1025 GOTO(out_proc, err = -ENOMEM);
1026 snprintf(dt, dt_len - 1, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1028 md_len = strlen(lprof->lp_md) + instlen + 2;
1029 OBD_ALLOC(md, md_len);
1031 GOTO(out_proc, err = -ENOMEM);
1032 snprintf(md, md_len - 1, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1034 /* connections, registrations, sb setup */
1035 err = client_common_fill_super(sb, md, dt, mnt);
1037 GOTO(out_proc, err);
1039 sbi->ll_client_common_fill_super_succeeded = 1;
1043 lprocfs_ll_unregister_mountpoint(sbi);
1046 OBD_FREE(md, md_len);
1048 OBD_FREE(dt, dt_len);
1050 class_put_profile(lprof);
1053 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1054 LCONSOLE_WARN("Mounted %s\n", profilenm);
1058 } /* ll_fill_super */
1060 void ll_put_super(struct super_block *sb)
1062 struct config_llog_instance cfg, params_cfg;
1063 struct obd_device *obd;
1064 struct lustre_sb_info *lsi = s2lsi(sb);
1065 struct ll_sb_info *sbi = ll_s2sbi(sb);
1066 char *profilenm = get_profile_name(sb);
1068 int next, force = 1, rc = 0;
1071 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1073 cfg.cfg_instance = sb;
1074 lustre_end_log(sb, profilenm, &cfg);
1076 params_cfg.cfg_instance = sb;
1077 lustre_end_log(sb, PARAMS_FILENAME, ¶ms_cfg);
1079 if (sbi->ll_md_exp) {
1080 obd = class_exp2obd(sbi->ll_md_exp);
1082 force = obd->obd_force;
1085 /* Wait for unstable pages to be committed to stable storage */
1087 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1088 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1089 atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1093 ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1094 if (force == 0 && rc != -EINTR)
1095 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1098 /* We need to set force before the lov_disconnect in
1099 lustre_common_put_super, since l_d cleans up osc's as well. */
1102 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1104 obd->obd_force = force;
1108 if (sbi->ll_client_common_fill_super_succeeded) {
1109 /* Only if client_common_fill_super succeeded */
1110 client_common_put_super(sb);
1114 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1115 class_manual_cleanup(obd);
1118 if (sbi->ll_flags & LL_SBI_VERBOSE)
1119 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1122 class_del_profile(profilenm);
1124 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1125 bdi_destroy(&lsi->lsi_bdi);
1126 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1130 lsi->lsi_llsbi = NULL;
1132 lustre_common_put_super(sb);
1134 cl_env_cache_purge(~0);
1136 module_put(THIS_MODULE);
1139 } /* client_put_super */
1141 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1143 struct inode *inode = NULL;
1145 /* NOTE: we depend on atomic igrab() -bzzz */
1146 lock_res_and_lock(lock);
1147 if (lock->l_resource->lr_lvb_inode) {
1148 struct ll_inode_info * lli;
1149 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1150 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1151 inode = igrab(lock->l_resource->lr_lvb_inode);
1153 inode = lock->l_resource->lr_lvb_inode;
1154 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1155 D_WARNING, lock, "lr_lvb_inode %p is "
1156 "bogus: magic %08x",
1157 lock->l_resource->lr_lvb_inode,
1158 lli->lli_inode_magic);
1162 unlock_res_and_lock(lock);
1166 void ll_dir_clear_lsm_md(struct inode *inode)
1168 struct ll_inode_info *lli = ll_i2info(inode);
1170 LASSERT(S_ISDIR(inode->i_mode));
1172 if (lli->lli_lsm_md != NULL) {
1173 lmv_free_memmd(lli->lli_lsm_md);
1174 lli->lli_lsm_md = NULL;
1178 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1179 const struct lu_fid *fid,
1180 struct lustre_md *md)
1182 struct ll_sb_info *sbi = ll_s2sbi(sb);
1183 struct mdt_body *body = md->body;
1184 struct inode *inode;
1188 ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1189 inode = iget_locked(sb, ino);
1190 if (inode == NULL) {
1191 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1192 ll_get_fsname(sb, NULL, 0), PFID(fid));
1193 RETURN(ERR_PTR(-ENOENT));
1196 if (inode->i_state & I_NEW) {
1197 struct ll_inode_info *lli = ll_i2info(inode);
1198 struct lmv_stripe_md *lsm = md->lmv;
1200 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1201 (body->mbo_mode & S_IFMT);
1202 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1205 LTIME_S(inode->i_mtime) = 0;
1206 LTIME_S(inode->i_atime) = 0;
1207 LTIME_S(inode->i_ctime) = 0;
1210 #ifdef HAVE_BACKING_DEV_INFO
1211 /* initializing backing dev info. */
1212 inode->i_mapping->backing_dev_info =
1213 &s2lsi(inode->i_sb)->lsi_bdi;
1215 inode->i_op = &ll_dir_inode_operations;
1216 inode->i_fop = &ll_dir_operations;
1217 lli->lli_fid = *fid;
1220 LASSERT(lsm != NULL);
1221 /* master object FID */
1222 lli->lli_pfid = body->mbo_fid1;
1223 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1224 lli, PFID(fid), PFID(&lli->lli_pfid));
1225 unlock_new_inode(inode);
1231 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1234 struct lmv_stripe_md *lsm = md->lmv;
1237 LASSERT(lsm != NULL);
1238 /* XXX sigh, this lsm_root initialization should be in
1239 * LMV layer, but it needs ll_iget right now, so we
1240 * put this here right now. */
1241 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1242 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1243 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1244 /* Unfortunately ll_iget will call ll_update_inode,
1245 * where the initialization of slave inode is slightly
1246 * different, so it reset lsm_md to NULL to avoid
1247 * initializing lsm for slave inode. */
1248 /* For migrating inode, master stripe and master object will
1249 * be same, so we only need assign this inode */
1250 if (lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION && i == 0)
1251 lsm->lsm_md_oinfo[i].lmo_root = inode;
1253 lsm->lsm_md_oinfo[i].lmo_root =
1254 ll_iget_anon_dir(inode->i_sb, fid, md);
1256 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1257 int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1259 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1267 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1268 const struct lmv_stripe_md *lsm_md2)
1270 return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1271 lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1272 lsm_md1->lsm_md_master_mdt_index ==
1273 lsm_md2->lsm_md_master_mdt_index &&
1274 lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1275 lsm_md1->lsm_md_layout_version ==
1276 lsm_md2->lsm_md_layout_version &&
1277 strcmp(lsm_md1->lsm_md_pool_name,
1278 lsm_md2->lsm_md_pool_name) == 0;
1281 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1283 struct ll_inode_info *lli = ll_i2info(inode);
1284 struct lmv_stripe_md *lsm = md->lmv;
1288 LASSERT(S_ISDIR(inode->i_mode));
1289 CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1290 PFID(ll_inode2fid(inode)));
1292 /* no striped information from request. */
1294 if (lli->lli_lsm_md == NULL) {
1296 } else if (lli->lli_lsm_md->lsm_md_hash_type &
1297 LMV_HASH_FLAG_MIGRATION) {
1298 /* migration is done, the temporay MIGRATE layout has
1300 CDEBUG(D_INODE, DFID" finish migration.\n",
1301 PFID(ll_inode2fid(inode)));
1302 lmv_free_memmd(lli->lli_lsm_md);
1303 lli->lli_lsm_md = NULL;
1306 /* The lustre_md from req does not include stripeEA,
1307 * see ll_md_setattr */
1312 /* set the directory layout */
1313 if (lli->lli_lsm_md == NULL) {
1314 struct cl_attr *attr;
1316 rc = ll_init_lsm_md(inode, md);
1320 /* set md->lmv to NULL, so the following free lustre_md
1321 * will not free this lsm */
1323 lli->lli_lsm_md = lsm;
1325 OBD_ALLOC_PTR(attr);
1329 /* validate the lsm */
1330 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1331 ll_md_blocking_ast);
1337 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1338 md->body->mbo_nlink = attr->cat_nlink;
1339 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1340 md->body->mbo_size = attr->cat_size;
1341 if (md->body->mbo_valid & OBD_MD_FLATIME)
1342 md->body->mbo_atime = attr->cat_atime;
1343 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1344 md->body->mbo_ctime = attr->cat_ctime;
1345 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1346 md->body->mbo_mtime = attr->cat_mtime;
1350 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1351 lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1355 /* Compare the old and new stripe information */
1356 if (!lsm_md_eq(lli->lli_lsm_md, lsm)) {
1357 struct lmv_stripe_md *old_lsm = lli->lli_lsm_md;
1360 CERROR("%s: inode "DFID"(%p)'s lmv layout mismatch (%p)/(%p)"
1361 "magic:0x%x/0x%x stripe count: %d/%d master_mdt: %d/%d"
1362 "hash_type:0x%x/0x%x layout: 0x%x/0x%x pool:%s/%s\n",
1363 ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1364 inode, lsm, old_lsm,
1365 lsm->lsm_md_magic, old_lsm->lsm_md_magic,
1366 lsm->lsm_md_stripe_count,
1367 old_lsm->lsm_md_stripe_count,
1368 lsm->lsm_md_master_mdt_index,
1369 old_lsm->lsm_md_master_mdt_index,
1370 lsm->lsm_md_hash_type, old_lsm->lsm_md_hash_type,
1371 lsm->lsm_md_layout_version,
1372 old_lsm->lsm_md_layout_version,
1373 lsm->lsm_md_pool_name,
1374 old_lsm->lsm_md_pool_name);
1376 for (idx = 0; idx < old_lsm->lsm_md_stripe_count; idx++) {
1377 CERROR("%s: sub FIDs in old lsm idx %d, old: "DFID"\n",
1378 ll_get_fsname(inode->i_sb, NULL, 0), idx,
1379 PFID(&old_lsm->lsm_md_oinfo[idx].lmo_fid));
1382 for (idx = 0; idx < lsm->lsm_md_stripe_count; idx++) {
1383 CERROR("%s: sub FIDs in new lsm idx %d, new: "DFID"\n",
1384 ll_get_fsname(inode->i_sb, NULL, 0), idx,
1385 PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1394 void ll_clear_inode(struct inode *inode)
1396 struct ll_inode_info *lli = ll_i2info(inode);
1397 struct ll_sb_info *sbi = ll_i2sbi(inode);
1400 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1401 PFID(ll_inode2fid(inode)), inode);
1403 if (S_ISDIR(inode->i_mode)) {
1404 /* these should have been cleared in ll_file_release */
1405 LASSERT(lli->lli_opendir_key == NULL);
1406 LASSERT(lli->lli_sai == NULL);
1407 LASSERT(lli->lli_opendir_pid == 0);
1410 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1412 LASSERT(!lli->lli_open_fd_write_count);
1413 LASSERT(!lli->lli_open_fd_read_count);
1414 LASSERT(!lli->lli_open_fd_exec_count);
1416 if (lli->lli_mds_write_och)
1417 ll_md_real_close(inode, FMODE_WRITE);
1418 if (lli->lli_mds_exec_och)
1419 ll_md_real_close(inode, FMODE_EXEC);
1420 if (lli->lli_mds_read_och)
1421 ll_md_real_close(inode, FMODE_READ);
1423 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1424 OBD_FREE(lli->lli_symlink_name,
1425 strlen(lli->lli_symlink_name) + 1);
1426 lli->lli_symlink_name = NULL;
1429 ll_xattr_cache_destroy(inode);
1431 #ifdef CONFIG_FS_POSIX_ACL
1432 if (lli->lli_posix_acl) {
1433 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1434 posix_acl_release(lli->lli_posix_acl);
1435 lli->lli_posix_acl = NULL;
1438 lli->lli_inode_magic = LLI_INODE_DEAD;
1440 if (S_ISDIR(inode->i_mode))
1441 ll_dir_clear_lsm_md(inode);
1442 else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1443 LASSERT(list_empty(&lli->lli_agl_list));
1446 * XXX This has to be done before lsm is freed below, because
1447 * cl_object still uses inode lsm.
1449 cl_inode_fini(inode);
1454 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1456 struct lustre_md md;
1457 struct inode *inode = dentry->d_inode;
1458 struct ll_sb_info *sbi = ll_i2sbi(inode);
1459 struct ptlrpc_request *request = NULL;
1463 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1464 LUSTRE_OPC_ANY, NULL);
1465 if (IS_ERR(op_data))
1466 RETURN(PTR_ERR(op_data));
1468 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1470 ptlrpc_req_finished(request);
1471 if (rc == -ENOENT) {
1473 /* Unlinked special device node? Or just a race?
1474 * Pretend we done everything. */
1475 if (!S_ISREG(inode->i_mode) &&
1476 !S_ISDIR(inode->i_mode)) {
1477 ia_valid = op_data->op_attr.ia_valid;
1478 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1479 rc = simple_setattr(dentry, &op_data->op_attr);
1480 op_data->op_attr.ia_valid = ia_valid;
1482 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1483 CERROR("md_setattr fails: rc = %d\n", rc);
1488 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1489 sbi->ll_md_exp, &md);
1491 ptlrpc_req_finished(request);
1495 ia_valid = op_data->op_attr.ia_valid;
1496 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1497 * cache is not cleared yet. */
1498 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1499 if (S_ISREG(inode->i_mode))
1501 rc = simple_setattr(dentry, &op_data->op_attr);
1502 if (S_ISREG(inode->i_mode))
1503 inode_unlock(inode);
1504 op_data->op_attr.ia_valid = ia_valid;
1506 rc = ll_update_inode(inode, &md);
1507 ptlrpc_req_finished(request);
1512 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1513 * object(s) determine the file size and mtime. Otherwise, the MDS will
1514 * keep these values until such a time that objects are allocated for it.
1515 * We do the MDS operations first, as it is checking permissions for us.
1516 * We don't to the MDS RPC if there is nothing that we want to store there,
1517 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1518 * going to do an RPC anyways.
1520 * If we are doing a truncate, we will send the mtime and ctime updates
1521 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1522 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1525 * In case of HSMimport, we only set attr on MDS.
1527 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1529 struct inode *inode = dentry->d_inode;
1530 struct ll_inode_info *lli = ll_i2info(inode);
1531 struct md_op_data *op_data = NULL;
1535 CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1536 "valid %x, hsm_import %d\n",
1537 ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1538 inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1541 if (attr->ia_valid & ATTR_SIZE) {
1542 /* Check new size against VFS/VM file size limit and rlimit */
1543 rc = inode_newsize_ok(inode, attr->ia_size);
1547 /* The maximum Lustre file size is variable, based on the
1548 * OST maximum object size and number of stripes. This
1549 * needs another check in addition to the VFS check above. */
1550 if (attr->ia_size > ll_file_maxbytes(inode)) {
1551 CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
1552 PFID(&lli->lli_fid), attr->ia_size,
1553 ll_file_maxbytes(inode));
1557 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1560 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1561 if (attr->ia_valid & TIMES_SET_FLAGS) {
1562 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1563 !cfs_capable(CFS_CAP_FOWNER))
1567 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1568 if (!(attr->ia_valid & ATTR_CTIME_SET) &&
1569 (attr->ia_valid & ATTR_CTIME)) {
1570 attr->ia_ctime = CURRENT_TIME;
1571 attr->ia_valid |= ATTR_CTIME_SET;
1573 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1574 (attr->ia_valid & ATTR_ATIME)) {
1575 attr->ia_atime = CURRENT_TIME;
1576 attr->ia_valid |= ATTR_ATIME_SET;
1578 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1579 (attr->ia_valid & ATTR_MTIME)) {
1580 attr->ia_mtime = CURRENT_TIME;
1581 attr->ia_valid |= ATTR_MTIME_SET;
1584 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1585 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %llu\n",
1586 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1587 (s64)ktime_get_real_seconds());
1589 if (S_ISREG(inode->i_mode)) {
1590 if (attr->ia_valid & ATTR_SIZE)
1591 inode_dio_write_done(inode);
1592 inode_unlock(inode);
1595 /* We always do an MDS RPC, even if we're only changing the size;
1596 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1598 OBD_ALLOC_PTR(op_data);
1599 if (op_data == NULL)
1600 GOTO(out, rc = -ENOMEM);
1602 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1603 /* If we are changing file size, file content is
1604 * modified, flag it. */
1605 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1606 op_data->op_bias |= MDS_DATA_MODIFIED;
1607 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1610 op_data->op_attr = *attr;
1612 rc = ll_md_setattr(dentry, op_data);
1616 if (!S_ISREG(inode->i_mode) || hsm_import)
1619 if (attr->ia_valid & (ATTR_SIZE |
1620 ATTR_ATIME | ATTR_ATIME_SET |
1621 ATTR_MTIME | ATTR_MTIME_SET |
1622 ATTR_CTIME | ATTR_CTIME_SET)) {
1623 /* For truncate and utimes sending attributes to OSTs, setting
1624 * mtime/atime to the past will be performed under PW [0:EOF]
1625 * extent lock (new_size:EOF for truncate). It may seem
1626 * excessive to send mtime/atime updates to OSTs when not
1627 * setting times to past, but it is necessary due to possible
1628 * time de-synchronization between MDT inode and OST objects */
1629 rc = cl_setattr_ost(lli->lli_clob, attr, 0);
1632 /* If the file was restored, it needs to set dirty flag.
1634 * We've already sent MDS_DATA_MODIFIED flag in
1635 * ll_md_setattr() for truncate. However, the MDT refuses to
1636 * set the HS_DIRTY flag on released files, so we have to set
1637 * it again if the file has been restored. Please check how
1638 * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1640 * Please notice that if the file is not released, the previous
1641 * MDS_DATA_MODIFIED has taken effect and usually
1642 * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1643 * This way we can save an RPC for common open + trunc
1645 if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1646 struct hsm_state_set hss = {
1647 .hss_valid = HSS_SETMASK,
1648 .hss_setmask = HS_DIRTY,
1652 rc2 = ll_hsm_state_set(inode, &hss);
1653 /* truncate and write can happen at the same time, so that
1654 * the file can be set modified even though the file is not
1655 * restored from released state, and ll_hsm_state_set() is
1656 * not applicable for the file, and rc2 < 0 is normal in this
1659 CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1660 PFID(ll_inode2fid(inode)), rc2);
1665 if (op_data != NULL)
1666 ll_finish_md_op_data(op_data);
1668 if (S_ISREG(inode->i_mode)) {
1670 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1671 inode_dio_wait(inode);
1672 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
1673 * flag. ll_update_inode (called from ll_md_setattr), clears
1674 * inode flags, so there is a gap where S_NOSEC is not set.
1675 * This can cause a writer to take the i_mutex unnecessarily,
1676 * but this is safe to do and should be rare. */
1677 inode_has_no_xattr(inode);
1680 ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1681 LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1686 int ll_setattr(struct dentry *de, struct iattr *attr)
1688 int mode = de->d_inode->i_mode;
1690 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1691 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1692 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1694 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1695 (ATTR_SIZE|ATTR_MODE)) &&
1696 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1697 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1698 !(attr->ia_mode & S_ISGID))))
1699 attr->ia_valid |= ATTR_FORCE;
1701 if ((attr->ia_valid & ATTR_MODE) &&
1703 !(attr->ia_mode & S_ISUID) &&
1704 !(attr->ia_valid & ATTR_KILL_SUID))
1705 attr->ia_valid |= ATTR_KILL_SUID;
1707 if ((attr->ia_valid & ATTR_MODE) &&
1708 ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1709 !(attr->ia_mode & S_ISGID) &&
1710 !(attr->ia_valid & ATTR_KILL_SGID))
1711 attr->ia_valid |= ATTR_KILL_SGID;
1713 /* avoid polluted from ATTR_TIMES_SET,
1714 * projid is not expected to be set here */
1715 attr->ia_valid &= ~MDS_ATTR_PROJID;
1717 return ll_setattr_raw(de, attr, false);
1720 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1721 __u64 max_age, __u32 flags)
1723 struct ll_sb_info *sbi = ll_s2sbi(sb);
1724 struct obd_statfs obd_osfs;
1728 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1730 CERROR("md_statfs fails: rc = %d\n", rc);
1734 osfs->os_type = sb->s_magic;
1736 CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1737 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1739 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1740 flags |= OBD_STATFS_NODELAY;
1742 rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1744 CERROR("obd_statfs fails: rc = %d\n", rc);
1748 CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1749 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1752 osfs->os_bsize = obd_osfs.os_bsize;
1753 osfs->os_blocks = obd_osfs.os_blocks;
1754 osfs->os_bfree = obd_osfs.os_bfree;
1755 osfs->os_bavail = obd_osfs.os_bavail;
1757 /* If we don't have as many objects free on the OST as inodes
1758 * on the MDS, we reduce the total number of inodes to
1759 * compensate, so that the "inodes in use" number is correct.
1761 if (obd_osfs.os_ffree < osfs->os_ffree) {
1762 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1764 osfs->os_ffree = obd_osfs.os_ffree;
1769 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1771 struct super_block *sb = de->d_sb;
1772 struct obd_statfs osfs;
1773 __u64 fsid = huge_encode_dev(sb->s_dev);
1776 CDEBUG(D_VFSTRACE, "VFS Op: at %llu jiffies\n", get_jiffies_64());
1777 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1779 /* Some amount of caching on the client is allowed */
1780 rc = ll_statfs_internal(sb, &osfs,
1781 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1786 statfs_unpack(sfs, &osfs);
1788 /* We need to downshift for all 32-bit kernels, because we can't
1789 * tell if the kernel is being called via sys_statfs64() or not.
1790 * Stop before overflowing f_bsize - in which case it is better
1791 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1792 if (sizeof(long) < 8) {
1793 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1796 osfs.os_blocks >>= 1;
1797 osfs.os_bfree >>= 1;
1798 osfs.os_bavail >>= 1;
1802 sfs->f_blocks = osfs.os_blocks;
1803 sfs->f_bfree = osfs.os_bfree;
1804 sfs->f_bavail = osfs.os_bavail;
1805 sfs->f_fsid.val[0] = (__u32)fsid;
1806 sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1810 void ll_inode_size_lock(struct inode *inode)
1812 struct ll_inode_info *lli;
1814 LASSERT(!S_ISDIR(inode->i_mode));
1816 lli = ll_i2info(inode);
1817 mutex_lock(&lli->lli_size_mutex);
1820 void ll_inode_size_unlock(struct inode *inode)
1822 struct ll_inode_info *lli;
1824 lli = ll_i2info(inode);
1825 mutex_unlock(&lli->lli_size_mutex);
1828 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1830 struct ll_inode_info *lli = ll_i2info(inode);
1831 struct mdt_body *body = md->body;
1832 struct ll_sb_info *sbi = ll_i2sbi(inode);
1834 if (body->mbo_valid & OBD_MD_FLEASIZE)
1835 cl_file_inode_init(inode, md);
1837 if (S_ISDIR(inode->i_mode)) {
1840 rc = ll_update_lsm_md(inode, md);
1845 #ifdef CONFIG_FS_POSIX_ACL
1846 if (body->mbo_valid & OBD_MD_FLACL) {
1847 spin_lock(&lli->lli_lock);
1848 if (lli->lli_posix_acl)
1849 posix_acl_release(lli->lli_posix_acl);
1850 lli->lli_posix_acl = md->posix_acl;
1851 spin_unlock(&lli->lli_lock);
1854 inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1855 sbi->ll_flags & LL_SBI_32BIT_API);
1856 inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1858 if (body->mbo_valid & OBD_MD_FLATIME) {
1859 if (body->mbo_atime > LTIME_S(inode->i_atime))
1860 LTIME_S(inode->i_atime) = body->mbo_atime;
1861 lli->lli_atime = body->mbo_atime;
1864 if (body->mbo_valid & OBD_MD_FLMTIME) {
1865 if (body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1866 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1867 "to %llu\n", inode->i_ino,
1868 LTIME_S(inode->i_mtime), body->mbo_mtime);
1869 LTIME_S(inode->i_mtime) = body->mbo_mtime;
1871 lli->lli_mtime = body->mbo_mtime;
1874 if (body->mbo_valid & OBD_MD_FLCTIME) {
1875 if (body->mbo_ctime > LTIME_S(inode->i_ctime))
1876 LTIME_S(inode->i_ctime) = body->mbo_ctime;
1877 lli->lli_ctime = body->mbo_ctime;
1880 /* Clear i_flags to remove S_NOSEC before permissions are updated */
1881 if (body->mbo_valid & OBD_MD_FLFLAGS)
1882 inode->i_flags = ll_ext_to_inode_flags(body->mbo_flags);
1883 if (body->mbo_valid & OBD_MD_FLMODE)
1884 inode->i_mode = (inode->i_mode & S_IFMT) |
1885 (body->mbo_mode & ~S_IFMT);
1887 if (body->mbo_valid & OBD_MD_FLTYPE)
1888 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1889 (body->mbo_mode & S_IFMT);
1891 LASSERT(inode->i_mode != 0);
1892 if (S_ISREG(inode->i_mode))
1893 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1894 LL_MAX_BLKSIZE_BITS);
1896 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1898 if (body->mbo_valid & OBD_MD_FLUID)
1899 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
1900 if (body->mbo_valid & OBD_MD_FLGID)
1901 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
1902 if (body->mbo_valid & OBD_MD_FLPROJID)
1903 lli->lli_projid = body->mbo_projid;
1904 if (body->mbo_valid & OBD_MD_FLNLINK)
1905 set_nlink(inode, body->mbo_nlink);
1906 if (body->mbo_valid & OBD_MD_FLRDEV)
1907 inode->i_rdev = old_decode_dev(body->mbo_rdev);
1909 if (body->mbo_valid & OBD_MD_FLID) {
1910 /* FID shouldn't be changed! */
1911 if (fid_is_sane(&lli->lli_fid)) {
1912 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
1913 "Trying to change FID "DFID
1914 " to the "DFID", inode "DFID"(%p)\n",
1915 PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
1916 PFID(ll_inode2fid(inode)), inode);
1918 lli->lli_fid = body->mbo_fid1;
1922 LASSERT(fid_seq(&lli->lli_fid) != 0);
1924 if (body->mbo_valid & OBD_MD_FLSIZE) {
1925 i_size_write(inode, body->mbo_size);
1927 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
1928 PFID(ll_inode2fid(inode)),
1929 (unsigned long long)body->mbo_size);
1931 if (body->mbo_valid & OBD_MD_FLBLOCKS)
1932 inode->i_blocks = body->mbo_blocks;
1935 if (body->mbo_valid & OBD_MD_TSTATE) {
1936 /* Set LLIF_FILE_RESTORING if restore ongoing and
1937 * clear it when done to ensure to start again
1938 * glimpsing updated attrs
1940 if (body->mbo_t_state & MS_RESTORE)
1941 ll_file_set_flag(lli, LLIF_FILE_RESTORING);
1943 ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
1949 int ll_read_inode2(struct inode *inode, void *opaque)
1951 struct lustre_md *md = opaque;
1952 struct ll_inode_info *lli = ll_i2info(inode);
1956 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1957 PFID(&lli->lli_fid), inode);
1959 /* Core attributes from the MDS first. This is a new inode, and
1960 * the VFS doesn't zero times in the core inode so we have to do
1961 * it ourselves. They will be overwritten by either MDS or OST
1962 * attributes - we just need to make sure they aren't newer. */
1963 LTIME_S(inode->i_mtime) = 0;
1964 LTIME_S(inode->i_atime) = 0;
1965 LTIME_S(inode->i_ctime) = 0;
1967 rc = ll_update_inode(inode, md);
1971 /* OIDEBUG(inode); */
1973 #ifdef HAVE_BACKING_DEV_INFO
1974 /* initializing backing dev info. */
1975 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1977 if (S_ISREG(inode->i_mode)) {
1978 struct ll_sb_info *sbi = ll_i2sbi(inode);
1979 inode->i_op = &ll_file_inode_operations;
1980 inode->i_fop = sbi->ll_fop;
1981 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1983 } else if (S_ISDIR(inode->i_mode)) {
1984 inode->i_op = &ll_dir_inode_operations;
1985 inode->i_fop = &ll_dir_operations;
1987 } else if (S_ISLNK(inode->i_mode)) {
1988 inode->i_op = &ll_fast_symlink_inode_operations;
1991 inode->i_op = &ll_special_inode_operations;
1993 init_special_inode(inode, inode->i_mode,
2002 void ll_delete_inode(struct inode *inode)
2004 struct ll_inode_info *lli = ll_i2info(inode);
2007 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2008 /* It is last chance to write out dirty pages,
2009 * otherwise we may lose data while umount */
2010 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
2012 truncate_inode_pages_final(&inode->i_data);
2014 LASSERTF(inode->i_data.nrpages == 0, "inode="DFID"(%p) nrpages=%lu, "
2015 "see https://jira.hpdd.intel.com/browse/LU-118\n",
2016 PFID(ll_inode2fid(inode)), inode, inode->i_data.nrpages);
2018 #ifdef HAVE_SBOPS_EVICT_INODE
2019 ll_clear_inode(inode);
2026 int ll_iocontrol(struct inode *inode, struct file *file,
2027 unsigned int cmd, unsigned long arg)
2029 struct ll_sb_info *sbi = ll_i2sbi(inode);
2030 struct ptlrpc_request *req = NULL;
2035 case FSFILT_IOC_GETFLAGS: {
2036 struct mdt_body *body;
2037 struct md_op_data *op_data;
2039 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2040 0, 0, LUSTRE_OPC_ANY,
2042 if (IS_ERR(op_data))
2043 RETURN(PTR_ERR(op_data));
2045 op_data->op_valid = OBD_MD_FLFLAGS;
2046 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2047 ll_finish_md_op_data(op_data);
2049 CERROR("%s: failure inode "DFID": rc = %d\n",
2050 sbi->ll_md_exp->exp_obd->obd_name,
2051 PFID(ll_inode2fid(inode)), rc);
2055 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2057 flags = body->mbo_flags;
2059 ptlrpc_req_finished(req);
2061 RETURN(put_user(flags, (int __user *)arg));
2063 case FSFILT_IOC_SETFLAGS: {
2065 struct md_op_data *op_data;
2066 struct cl_object *obj;
2068 if (get_user(flags, (int __user *)arg))
2071 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2072 LUSTRE_OPC_ANY, NULL);
2073 if (IS_ERR(op_data))
2074 RETURN(PTR_ERR(op_data));
2076 op_data->op_attr_flags = flags;
2077 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2078 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2079 ll_finish_md_op_data(op_data);
2080 ptlrpc_req_finished(req);
2084 inode->i_flags = ll_ext_to_inode_flags(flags);
2086 obj = ll_i2info(inode)->lli_clob;
2090 OBD_ALLOC_PTR(attr);
2094 attr->ia_valid = ATTR_ATTR_FLAG;
2095 rc = cl_setattr_ost(obj, attr, flags);
2107 int ll_flush_ctx(struct inode *inode)
2109 struct ll_sb_info *sbi = ll_i2sbi(inode);
2111 CDEBUG(D_SEC, "flush context for user %d\n",
2112 from_kuid(&init_user_ns, current_uid()));
2114 obd_set_info_async(NULL, sbi->ll_md_exp,
2115 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2117 obd_set_info_async(NULL, sbi->ll_dt_exp,
2118 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2123 /* umount -f client means force down, don't save state */
2124 void ll_umount_begin(struct super_block *sb)
2126 struct ll_sb_info *sbi = ll_s2sbi(sb);
2127 struct obd_device *obd;
2128 struct obd_ioctl_data *ioc_data;
2129 struct l_wait_info lwi;
2130 wait_queue_head_t waitq;
2133 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2134 sb->s_count, atomic_read(&sb->s_active));
2136 obd = class_exp2obd(sbi->ll_md_exp);
2138 CERROR("Invalid MDC connection handle %#llx\n",
2139 sbi->ll_md_exp->exp_handle.h_cookie);
2145 obd = class_exp2obd(sbi->ll_dt_exp);
2147 CERROR("Invalid LOV connection handle %#llx\n",
2148 sbi->ll_dt_exp->exp_handle.h_cookie);
2154 OBD_ALLOC_PTR(ioc_data);
2156 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2157 sizeof *ioc_data, ioc_data, NULL);
2159 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2160 sizeof *ioc_data, ioc_data, NULL);
2162 OBD_FREE_PTR(ioc_data);
2165 /* Really, we'd like to wait until there are no requests outstanding,
2166 * and then continue. For now, we just periodically checking for vfs
2167 * to decrement mnt_cnt and hope to finish it within 10sec.
2169 init_waitqueue_head(&waitq);
2170 lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(10),
2171 cfs_time_seconds(1), NULL, NULL);
2172 l_wait_event(waitq, may_umount(sbi->ll_mnt.mnt), &lwi);
2177 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2179 struct ll_sb_info *sbi = ll_s2sbi(sb);
2180 char *profilenm = get_profile_name(sb);
2184 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2185 read_only = *flags & MS_RDONLY;
2186 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2187 sizeof(KEY_READ_ONLY),
2188 KEY_READ_ONLY, sizeof(read_only),
2191 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2192 profilenm, read_only ?
2193 "read-only" : "read-write", err);
2198 sb->s_flags |= MS_RDONLY;
2200 sb->s_flags &= ~MS_RDONLY;
2202 if (sbi->ll_flags & LL_SBI_VERBOSE)
2203 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2204 read_only ? "read-only" : "read-write");
2210 * Cleanup the open handle that is cached on MDT-side.
2212 * For open case, the client side open handling thread may hit error
2213 * after the MDT grant the open. Under such case, the client should
2214 * send close RPC to the MDT as cleanup; otherwise, the open handle
2215 * on the MDT will be leaked there until the client umount or evicted.
2217 * In further, if someone unlinked the file, because the open handle
2218 * holds the reference on such file/object, then it will block the
2219 * subsequent threads that want to locate such object via FID.
2221 * \param[in] sb super block for this file-system
2222 * \param[in] open_req pointer to the original open request
2224 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2226 struct mdt_body *body;
2227 struct md_op_data *op_data;
2228 struct ptlrpc_request *close_req = NULL;
2229 struct obd_export *exp = ll_s2sbi(sb)->ll_md_exp;
2232 body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2233 OBD_ALLOC_PTR(op_data);
2234 if (op_data == NULL) {
2235 CWARN("%s: cannot allocate op_data to release open handle for "
2237 ll_get_fsname(sb, NULL, 0), PFID(&body->mbo_fid1));
2242 op_data->op_fid1 = body->mbo_fid1;
2243 op_data->op_handle = body->mbo_handle;
2244 op_data->op_mod_time = ktime_get_real_seconds();
2245 md_close(exp, op_data, NULL, &close_req);
2246 ptlrpc_req_finished(close_req);
2247 ll_finish_md_op_data(op_data);
2252 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2253 struct super_block *sb, struct lookup_intent *it)
2255 struct ll_sb_info *sbi = NULL;
2256 struct lustre_md md = { NULL };
2260 LASSERT(*inode || sb);
2261 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2262 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2263 sbi->ll_md_exp, &md);
2268 rc = ll_update_inode(*inode, &md);
2272 LASSERT(sb != NULL);
2275 * At this point server returns to client's same fid as client
2276 * generated for creating. So using ->fid1 is okay here.
2278 if (!fid_is_sane(&md.body->mbo_fid1)) {
2279 CERROR("%s: Fid is insane "DFID"\n",
2280 ll_get_fsname(sb, NULL, 0),
2281 PFID(&md.body->mbo_fid1));
2282 GOTO(out, rc = -EINVAL);
2285 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2286 sbi->ll_flags & LL_SBI_32BIT_API),
2288 if (IS_ERR(*inode)) {
2289 #ifdef CONFIG_FS_POSIX_ACL
2291 posix_acl_release(md.posix_acl);
2292 md.posix_acl = NULL;
2295 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2297 CERROR("new_inode -fatal: rc %d\n", rc);
2302 /* Handling piggyback layout lock.
2303 * Layout lock can be piggybacked by getattr and open request.
2304 * The lsm can be applied to inode only if it comes with a layout lock
2305 * otherwise correct layout may be overwritten, for example:
2306 * 1. proc1: mdt returns a lsm but not granting layout
2307 * 2. layout was changed by another client
2308 * 3. proc2: refresh layout and layout lock granted
2309 * 4. proc1: to apply a stale layout */
2310 if (it != NULL && it->it_lock_mode != 0) {
2311 struct lustre_handle lockh;
2312 struct ldlm_lock *lock;
2314 lockh.cookie = it->it_lock_handle;
2315 lock = ldlm_handle2lock(&lockh);
2316 LASSERT(lock != NULL);
2317 if (ldlm_has_layout(lock)) {
2318 struct cl_object_conf conf;
2320 memset(&conf, 0, sizeof(conf));
2321 conf.coc_opc = OBJECT_CONF_SET;
2322 conf.coc_inode = *inode;
2323 conf.coc_lock = lock;
2324 conf.u.coc_layout = md.layout;
2325 (void)ll_layout_conf(*inode, &conf);
2327 LDLM_LOCK_PUT(lock);
2333 md_free_lustre_md(sbi->ll_md_exp, &md);
2336 if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2337 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2342 int ll_obd_statfs(struct inode *inode, void __user *arg)
2344 struct ll_sb_info *sbi = NULL;
2345 struct obd_export *exp;
2347 struct obd_ioctl_data *data = NULL;
2351 if (!inode || !(sbi = ll_i2sbi(inode)))
2352 GOTO(out_statfs, rc = -EINVAL);
2354 rc = obd_ioctl_getdata(&buf, &len, arg);
2356 GOTO(out_statfs, rc);
2359 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2360 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2361 GOTO(out_statfs, rc = -EINVAL);
2363 if (data->ioc_inllen1 != sizeof(__u32) ||
2364 data->ioc_inllen2 != sizeof(__u32) ||
2365 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2366 data->ioc_plen2 != sizeof(struct obd_uuid))
2367 GOTO(out_statfs, rc = -EINVAL);
2369 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2370 if (type & LL_STATFS_LMV)
2371 exp = sbi->ll_md_exp;
2372 else if (type & LL_STATFS_LOV)
2373 exp = sbi->ll_dt_exp;
2375 GOTO(out_statfs, rc = -ENODEV);
2377 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2379 GOTO(out_statfs, rc);
2381 OBD_FREE_LARGE(buf, len);
2385 int ll_process_config(struct lustre_cfg *lcfg)
2387 struct super_block *sb;
2392 /* The instance name contains the sb: lustre-client-aacfe000 */
2393 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2394 if (!ptr || !*(++ptr))
2396 if (sscanf(ptr, "%lx", &x) != 1)
2398 sb = (struct super_block *)x;
2399 /* This better be a real Lustre superblock! */
2400 LASSERT(s2lsi(sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2402 /* Note we have not called client_common_fill_super yet, so
2403 proc fns must be able to handle that! */
2404 rc = class_process_proc_param(PARAM_LLITE, lprocfs_llite_obd_vars,
2411 /* this function prepares md_op_data hint for passing it down to MD stack. */
2412 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2413 struct inode *i1, struct inode *i2,
2414 const char *name, size_t namelen,
2415 __u32 mode, __u32 opc, void *data)
2417 LASSERT(i1 != NULL);
2420 /* Do not reuse namelen for something else. */
2422 return ERR_PTR(-EINVAL);
2424 if (namelen > ll_i2sbi(i1)->ll_namelen)
2425 return ERR_PTR(-ENAMETOOLONG);
2427 if (!lu_name_is_valid_2(name, namelen))
2428 return ERR_PTR(-EINVAL);
2431 if (op_data == NULL)
2432 OBD_ALLOC_PTR(op_data);
2434 if (op_data == NULL)
2435 return ERR_PTR(-ENOMEM);
2437 ll_i2gids(op_data->op_suppgids, i1, i2);
2438 op_data->op_fid1 = *ll_inode2fid(i1);
2439 op_data->op_default_stripe_offset = -1;
2440 if (S_ISDIR(i1->i_mode)) {
2441 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2442 if (opc == LUSTRE_OPC_MKDIR)
2443 op_data->op_default_stripe_offset =
2444 ll_i2info(i1)->lli_def_stripe_offset;
2448 op_data->op_fid2 = *ll_inode2fid(i2);
2449 if (S_ISDIR(i2->i_mode))
2450 op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2452 fid_zero(&op_data->op_fid2);
2455 if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2456 op_data->op_cli_flags |= CLI_HASH64;
2458 if (ll_need_32bit_api(ll_i2sbi(i1)))
2459 op_data->op_cli_flags |= CLI_API32;
2461 op_data->op_name = name;
2462 op_data->op_namelen = namelen;
2463 op_data->op_mode = mode;
2464 op_data->op_mod_time = cfs_time_current_sec();
2465 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2466 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2467 op_data->op_cap = cfs_curproc_cap_pack();
2468 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2469 filename_is_volatile(name, namelen, &op_data->op_mds)) {
2470 op_data->op_bias |= MDS_CREATE_VOLATILE;
2472 op_data->op_mds = 0;
2474 op_data->op_data = data;
2479 void ll_finish_md_op_data(struct md_op_data *op_data)
2481 security_release_secctx(op_data->op_file_secctx,
2482 op_data->op_file_secctx_size);
2483 OBD_FREE_PTR(op_data);
2486 #ifdef HAVE_SUPEROPS_USE_DENTRY
2487 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2489 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2492 struct ll_sb_info *sbi;
2494 #ifdef HAVE_SUPEROPS_USE_DENTRY
2495 LASSERT((seq != NULL) && (dentry != NULL));
2496 sbi = ll_s2sbi(dentry->d_sb);
2498 LASSERT((seq != NULL) && (vfs != NULL));
2499 sbi = ll_s2sbi(vfs->mnt_sb);
2502 if (sbi->ll_flags & LL_SBI_NOLCK)
2503 seq_puts(seq, ",nolock");
2505 if (sbi->ll_flags & LL_SBI_FLOCK)
2506 seq_puts(seq, ",flock");
2508 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2509 seq_puts(seq, ",localflock");
2511 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2512 seq_puts(seq, ",user_xattr");
2514 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2515 seq_puts(seq, ",lazystatfs");
2517 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2518 seq_puts(seq, ",user_fid2path");
2520 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2521 seq_puts(seq, ",always_ping");
2527 * Get obd name by cmd, and copy out to user space
2529 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2531 struct ll_sb_info *sbi = ll_i2sbi(inode);
2532 struct obd_device *obd;
2535 if (cmd == OBD_IOC_GETDTNAME)
2536 obd = class_exp2obd(sbi->ll_dt_exp);
2537 else if (cmd == OBD_IOC_GETMDNAME)
2538 obd = class_exp2obd(sbi->ll_md_exp);
2545 if (copy_to_user((void __user *)arg, obd->obd_name,
2546 strlen(obd->obd_name) + 1))
2553 * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2554 * fsname will be returned in this buffer; otherwise, a static buffer will be
2555 * used to store the fsname and returned to caller.
2557 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2559 static char fsname_static[MTI_NAME_MAXLEN];
2560 struct lustre_sb_info *lsi = s2lsi(sb);
2565 /* this means the caller wants to use static buffer
2566 * and it doesn't care about race. Usually this is
2567 * in error reporting path */
2568 buf = fsname_static;
2569 buflen = sizeof(fsname_static);
2572 len = strlen(lsi->lsi_lmd->lmd_profile);
2573 ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2574 if (ptr && (strcmp(ptr, "-client") == 0))
2577 if (unlikely(len >= buflen))
2579 strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2585 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2592 p.mnt = current->fs->root.mnt;
2594 path = d_path(&p, buf, bufsize);
2599 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2601 char *buf, *path = NULL;
2602 struct dentry *dentry = NULL;
2603 struct inode *inode = page->mapping->host;
2605 /* this can be called inside spin lock so use GFP_ATOMIC. */
2606 buf = (char *)__get_free_page(GFP_ATOMIC);
2608 dentry = d_find_alias(page->mapping->host);
2610 path = ll_d_path(dentry, buf, PAGE_SIZE);
2614 "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2615 "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2616 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2617 PFID(ll_inode2fid(inode)),
2618 (path && !IS_ERR(path)) ? path : "", ioret);
2624 free_page((unsigned long)buf);
2627 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2628 struct lov_user_md **kbuf)
2630 struct lov_user_md lum;
2634 if (copy_from_user(&lum, md, sizeof(lum)))
2637 lum_size = ll_lov_user_md_size(&lum);
2641 OBD_ALLOC(*kbuf, lum_size);
2645 if (copy_from_user(*kbuf, md, lum_size) != 0) {
2646 OBD_FREE(*kbuf, lum_size);
2654 * Compute llite root squash state after a change of root squash
2655 * configuration setting or add/remove of a lnet nid
2657 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2659 struct root_squash_info *squash = &sbi->ll_squash;
2662 struct lnet_process_id id;
2664 /* Update norootsquash flag */
2665 down_write(&squash->rsi_sem);
2666 if (list_empty(&squash->rsi_nosquash_nids))
2667 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2669 /* Do not apply root squash as soon as one of our NIDs is
2670 * in the nosquash_nids list */
2673 while (LNetGetId(i++, &id) != -ENOENT) {
2674 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2676 if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2682 sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2684 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2686 up_write(&squash->rsi_sem);
2690 * Parse linkea content to extract information about a given hardlink
2692 * \param[in] ldata - Initialized linkea data
2693 * \param[in] linkno - Link identifier
2694 * \param[out] parent_fid - The entry's parent FID
2695 * \param[out] ln - Entry name destination buffer
2697 * \retval 0 on success
2698 * \retval Appropriate negative error code on failure
2700 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2701 struct lu_fid *parent_fid, struct lu_name *ln)
2707 rc = linkea_init_with_rec(ldata);
2711 if (linkno >= ldata->ld_leh->leh_reccount)
2712 /* beyond last link */
2715 linkea_first_entry(ldata);
2716 for (idx = 0; ldata->ld_lee != NULL; idx++) {
2717 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2722 linkea_next_entry(ldata);
2732 * Get parent FID and name of an identified link. Operation is performed for
2733 * a given link number, letting the caller iterate over linkno to list one or
2734 * all links of an entry.
2736 * \param[in] file - File descriptor against which to perform the operation
2737 * \param[in,out] arg - User-filled structure containing the linkno to operate
2738 * on and the available size. It is eventually filled with
2739 * the requested information or left untouched on error
2741 * \retval - 0 on success
2742 * \retval - Appropriate negative error code on failure
2744 int ll_getparent(struct file *file, struct getparent __user *arg)
2746 struct dentry *dentry = file_dentry(file);
2747 struct inode *inode = file_inode(file);
2748 struct linkea_data *ldata;
2749 struct lu_buf buf = LU_BUF_NULL;
2751 struct lu_fid parent_fid;
2758 if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2759 !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2762 if (get_user(name_size, &arg->gp_name_size))
2765 if (get_user(linkno, &arg->gp_linkno))
2768 if (name_size > PATH_MAX)
2771 OBD_ALLOC(ldata, sizeof(*ldata));
2775 rc = linkea_data_new(ldata, &buf);
2777 GOTO(ldata_free, rc);
2779 rc = ll_getxattr(dentry, XATTR_NAME_LINK, buf.lb_buf, buf.lb_len);
2783 rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2787 if (ln.ln_namelen >= name_size)
2788 GOTO(lb_free, rc = -EOVERFLOW);
2790 if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2791 GOTO(lb_free, rc = -EFAULT);
2793 if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2794 GOTO(lb_free, rc = -EFAULT);
2796 if (put_user('\0', arg->gp_name + ln.ln_namelen))
2797 GOTO(lb_free, rc = -EFAULT);
2802 OBD_FREE(ldata, sizeof(*ldata));