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, 2017, 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/random.h>
41 #include <linux/statfs.h>
42 #include <linux/time.h>
43 #include <linux/types.h>
44 #include <libcfs/linux/linux-uuid.h>
45 #include <linux/version.h>
47 #include <linux/user_namespace.h>
48 #ifdef HAVE_UIDGID_HEADER
49 # include <linux/uidgid.h>
51 #include <linux/security.h>
53 #include <uapi/linux/lustre/lustre_ioctl.h>
54 #ifdef HAVE_UAPI_LINUX_MOUNT_H
55 #include <uapi/linux/mount.h>
58 #include <lustre_ha.h>
59 #include <lustre_dlm.h>
60 #include <lprocfs_status.h>
61 #include <lustre_disk.h>
62 #include <uapi/linux/lustre/lustre_param.h>
63 #include <lustre_log.h>
64 #include <cl_object.h>
65 #include <obd_cksum.h>
66 #include "llite_internal.h"
68 struct kmem_cache *ll_file_data_slab;
71 #define log2(n) ffz(~(n))
74 static struct ll_sb_info *ll_init_sbi(void)
76 struct ll_sb_info *sbi = NULL;
78 unsigned long lru_page_max;
87 RETURN(ERR_PTR(-ENOMEM));
89 rc = pcc_super_init(&sbi->ll_pcc_super);
93 spin_lock_init(&sbi->ll_lock);
94 mutex_init(&sbi->ll_lco.lco_lock);
95 spin_lock_init(&sbi->ll_pp_extent_lock);
96 spin_lock_init(&sbi->ll_process_lock);
97 sbi->ll_rw_stats_on = 0;
98 sbi->ll_statfs_max_age = OBD_STATFS_CACHE_SECONDS;
101 pages = si.totalram - si.totalhigh;
102 lru_page_max = pages / 2;
104 sbi->ll_ra_info.ra_async_max_active = 0;
105 sbi->ll_ra_info.ll_readahead_wq =
106 alloc_workqueue("ll-readahead-wq", WQ_UNBOUND,
107 sbi->ll_ra_info.ra_async_max_active);
108 if (!sbi->ll_ra_info.ll_readahead_wq)
109 GOTO(out_pcc, rc = -ENOMEM);
111 /* initialize ll_cache data */
112 sbi->ll_cache = cl_cache_init(lru_page_max);
113 if (sbi->ll_cache == NULL)
114 GOTO(out_destroy_ra, rc = -ENOMEM);
116 sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
117 SBI_DEFAULT_READAHEAD_MAX);
118 sbi->ll_ra_info.ra_async_pages_per_file_threshold =
119 sbi->ll_ra_info.ra_max_pages_per_file;
120 sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
121 sbi->ll_ra_info.ra_max_read_ahead_whole_pages = -1;
123 sbi->ll_flags |= LL_SBI_VERBOSE;
124 #ifdef ENABLE_CHECKSUM
125 sbi->ll_flags |= LL_SBI_CHECKSUM;
128 sbi->ll_flags |= LL_SBI_FLOCK;
131 #ifdef HAVE_LRU_RESIZE_SUPPORT
132 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
134 sbi->ll_flags |= LL_SBI_LAZYSTATFS;
136 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
137 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
139 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
143 /* metadata statahead is enabled by default */
144 sbi->ll_sa_running_max = LL_SA_RUNNING_DEF;
145 sbi->ll_sa_max = LL_SA_RPC_DEF;
146 atomic_set(&sbi->ll_sa_total, 0);
147 atomic_set(&sbi->ll_sa_wrong, 0);
148 atomic_set(&sbi->ll_sa_running, 0);
149 atomic_set(&sbi->ll_agl_total, 0);
150 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
151 sbi->ll_flags |= LL_SBI_FAST_READ;
152 sbi->ll_flags |= LL_SBI_TINY_WRITE;
155 sbi->ll_squash.rsi_uid = 0;
156 sbi->ll_squash.rsi_gid = 0;
157 INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
158 spin_lock_init(&sbi->ll_squash.rsi_lock);
160 /* Per-filesystem file heat */
161 sbi->ll_heat_decay_weight = SBI_DEFAULT_HEAT_DECAY_WEIGHT;
162 sbi->ll_heat_period_second = SBI_DEFAULT_HEAT_PERIOD_SECOND;
165 destroy_workqueue(sbi->ll_ra_info.ll_readahead_wq);
167 pcc_super_fini(&sbi->ll_pcc_super);
173 static void ll_free_sbi(struct super_block *sb)
175 struct ll_sb_info *sbi = ll_s2sbi(sb);
179 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
180 cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
181 if (sbi->ll_ra_info.ll_readahead_wq)
182 destroy_workqueue(sbi->ll_ra_info.ll_readahead_wq);
183 if (sbi->ll_cache != NULL) {
184 cl_cache_decref(sbi->ll_cache);
185 sbi->ll_cache = NULL;
187 pcc_super_fini(&sbi->ll_pcc_super);
188 OBD_FREE(sbi, sizeof(*sbi));
193 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
194 struct vfsmount *mnt)
196 struct inode *root = NULL;
197 struct ll_sb_info *sbi = ll_s2sbi(sb);
198 struct obd_statfs *osfs = NULL;
199 struct ptlrpc_request *request = NULL;
200 struct obd_connect_data *data = NULL;
201 struct obd_uuid *uuid;
202 struct md_op_data *op_data;
203 struct lustre_md lmd;
205 int size, err, checksum;
208 sbi->ll_md_obd = class_name2obd(md);
209 if (!sbi->ll_md_obd) {
210 CERROR("MD %s: not setup or attached\n", md);
224 /* pass client page size via ocd_grant_blkbits, the server should report
225 * back its backend blocksize for grant calculation purpose */
226 data->ocd_grant_blkbits = PAGE_SHIFT;
228 /* indicate MDT features supported by this client */
229 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
230 OBD_CONNECT_ATTRFID | OBD_CONNECT_GRANT |
231 OBD_CONNECT_VERSION | OBD_CONNECT_BRW_SIZE |
232 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
233 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
234 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
235 OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 |
236 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
237 OBD_CONNECT_64BITHASH |
238 OBD_CONNECT_EINPROGRESS |
239 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
240 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS|
241 OBD_CONNECT_MAX_EASIZE |
242 OBD_CONNECT_FLOCK_DEAD |
243 OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
244 OBD_CONNECT_OPEN_BY_FID |
245 OBD_CONNECT_DIR_STRIPE |
246 OBD_CONNECT_BULK_MBITS | OBD_CONNECT_CKSUM |
247 OBD_CONNECT_SUBTREE |
248 OBD_CONNECT_MULTIMODRPCS |
249 OBD_CONNECT_GRANT_PARAM |
250 OBD_CONNECT_SHORTIO | OBD_CONNECT_FLAGS2;
252 data->ocd_connect_flags2 = OBD_CONNECT2_DIR_MIGRATE |
253 OBD_CONNECT2_SUM_STATFS |
254 OBD_CONNECT2_OVERSTRIPING |
256 OBD_CONNECT2_LOCK_CONVERT |
257 OBD_CONNECT2_ARCHIVE_ID_ARRAY |
258 OBD_CONNECT2_INC_XID |
260 OBD_CONNECT2_ASYNC_DISCARD |
263 #ifdef HAVE_LRU_RESIZE_SUPPORT
264 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
265 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
267 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
268 data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK |
269 OBD_CONNECT_LARGE_ACL;
272 data->ocd_cksum_types = obd_cksum_types_supported_client();
274 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
275 /* flag mdc connection as lightweight, only used for test
276 * purpose, use with care */
277 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
279 data->ocd_ibits_known = MDS_INODELOCK_FULL;
280 data->ocd_version = LUSTRE_VERSION_CODE;
282 if (sb->s_flags & SB_RDONLY)
283 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
284 if (sbi->ll_flags & LL_SBI_USER_XATTR)
285 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
288 /* Setting this indicates we correctly support S_NOSEC (See kernel
289 * commit 9e1f1de02c2275d7172e18dc4e7c2065777611bf)
291 sb->s_flags |= SB_NOSEC;
294 if (sbi->ll_flags & LL_SBI_FLOCK)
295 sbi->ll_fop = &ll_file_operations_flock;
296 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
297 sbi->ll_fop = &ll_file_operations;
299 sbi->ll_fop = &ll_file_operations_noflock;
301 /* always ping even if server suppress_pings */
302 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
303 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
305 obd_connect_set_secctx(data);
307 #if defined(CONFIG_SECURITY)
308 data->ocd_connect_flags2 |= OBD_CONNECT2_SELINUX_POLICY;
311 data->ocd_brw_size = MD_MAX_BRW_SIZE;
313 err = obd_connect(NULL, &sbi->ll_md_exp, sbi->ll_md_obd,
314 &sbi->ll_sb_uuid, data, sbi->ll_cache);
316 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
317 "recovery, of which this client is not a "
318 "part. Please wait for recovery to complete,"
319 " abort, or time out.\n", md);
322 CERROR("cannot connect to %s: rc = %d\n", md, err);
326 sbi->ll_md_exp->exp_connect_data = *data;
328 err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
329 LUSTRE_SEQ_METADATA);
331 CERROR("%s: Can't init metadata layer FID infrastructure, "
332 "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
336 /* For mount, we only need fs info from MDT0, and also in DNE, it
337 * can make sure the client can be mounted as long as MDT0 is
339 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
340 ktime_get_seconds() - sbi->ll_statfs_max_age,
341 OBD_STATFS_FOR_MDT0);
343 GOTO(out_md_fid, err);
345 /* This needs to be after statfs to ensure connect has finished.
346 * Note that "data" does NOT contain the valid connect reply.
347 * If connecting to a 1.8 server there will be no LMV device, so
348 * we can access the MDC export directly and exp_connect_flags will
349 * be non-zero, but if accessing an upgraded 2.1 server it will
350 * have the correct flags filled in.
351 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
352 valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
353 if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
354 valid != CLIENT_CONNECT_MDT_REQD) {
357 OBD_ALLOC_WAIT(buf, PAGE_SIZE);
358 obd_connect_flags2str(buf, PAGE_SIZE,
359 valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
360 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
361 "feature(s) needed for correct operation "
362 "of this client (%s). Please upgrade "
363 "server or downgrade client.\n",
364 sbi->ll_md_exp->exp_obd->obd_name, buf);
365 OBD_FREE(buf, PAGE_SIZE);
366 GOTO(out_md_fid, err = -EPROTO);
369 size = sizeof(*data);
370 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
371 KEY_CONN_DATA, &size, data);
373 CERROR("%s: Get connect data failed: rc = %d\n",
374 sbi->ll_md_exp->exp_obd->obd_name, err);
375 GOTO(out_md_fid, err);
378 LASSERT(osfs->os_bsize);
379 sb->s_blocksize = osfs->os_bsize;
380 sb->s_blocksize_bits = log2(osfs->os_bsize);
381 sb->s_magic = LL_SUPER_MAGIC;
382 sb->s_maxbytes = MAX_LFS_FILESIZE;
383 sbi->ll_namelen = osfs->os_namelen;
384 sbi->ll_mnt.mnt = current->fs->root.mnt;
386 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
387 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
388 LCONSOLE_INFO("Disabling user_xattr feature because "
389 "it is not supported on the server\n");
390 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
393 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
395 sb->s_flags |= SB_POSIXACL;
397 sbi->ll_flags |= LL_SBI_ACL;
399 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
401 sb->s_flags &= ~SB_POSIXACL;
403 sbi->ll_flags &= ~LL_SBI_ACL;
406 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
407 sbi->ll_flags |= LL_SBI_64BIT_HASH;
409 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
410 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
412 if (obd_connect_has_secctx(data))
413 sbi->ll_flags |= LL_SBI_FILE_SECCTX;
415 if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
416 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
417 LCONSOLE_INFO("%s: disabling xattr cache due to "
418 "unknown maximum xattr size.\n", dt);
419 } else if (!sbi->ll_xattr_cache_set) {
420 /* If xattr_cache is already set (no matter 0 or 1)
421 * during processing llog, it won't be enabled here. */
422 sbi->ll_flags |= LL_SBI_XATTR_CACHE;
423 sbi->ll_xattr_cache_enabled = 1;
427 sbi->ll_dt_obd = class_name2obd(dt);
428 if (!sbi->ll_dt_obd) {
429 CERROR("DT %s: not setup or attached\n", dt);
430 GOTO(out_md_fid, err = -ENODEV);
433 /* pass client page size via ocd_grant_blkbits, the server should report
434 * back its backend blocksize for grant calculation purpose */
435 data->ocd_grant_blkbits = PAGE_SHIFT;
437 /* indicate OST features supported by this client */
438 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
439 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
440 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
441 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
442 OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
443 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
444 OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
445 OBD_CONNECT_EINPROGRESS |
446 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
447 OBD_CONNECT_LAYOUTLOCK |
448 OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
449 OBD_CONNECT_BULK_MBITS | OBD_CONNECT_SHORTIO |
450 OBD_CONNECT_FLAGS2 | OBD_CONNECT_GRANT_SHRINK;
452 /* The client currently advertises support for OBD_CONNECT_LOCKAHEAD_OLD so it
453 * can interoperate with an older version of lockahead which was released prior
454 * to landing in master. This support will be dropped when 2.13 development
455 * starts. At the point, we should not just drop the connect flag (below), we
456 * should also remove the support in the code.
458 * Removing it means a few things:
459 * 1. Remove this section here
460 * 2. Remove CEF_NONBLOCK in ll_file_lockahead()
461 * 3. Remove function exp_connect_lockahead_old
462 * 4. Remove LDLM_FL_LOCKAHEAD_OLD_RESERVED in lustre_dlm_flags.h
464 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 12, 50, 0)
465 data->ocd_connect_flags |= OBD_CONNECT_LOCKAHEAD_OLD;
468 data->ocd_connect_flags2 = OBD_CONNECT2_LOCKAHEAD |
469 OBD_CONNECT2_INC_XID;
471 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
472 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
474 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
475 * disabled by default, because it can still be enabled on the
476 * fly via /sys. As a consequence, we still need to come to an
477 * agreement on the supported algorithms at connect time
479 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
481 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
482 data->ocd_cksum_types = OBD_CKSUM_ADLER;
484 data->ocd_cksum_types = obd_cksum_types_supported_client();
486 #ifdef HAVE_LRU_RESIZE_SUPPORT
487 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
489 /* always ping even if server suppress_pings */
490 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
491 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
493 CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
494 "ocd_grant: %d\n", data->ocd_connect_flags,
495 data->ocd_version, data->ocd_grant);
497 sbi->ll_dt_obd->obd_upcall.onu_owner = &sbi->ll_lco;
498 sbi->ll_dt_obd->obd_upcall.onu_upcall = cl_ocd_update;
500 data->ocd_brw_size = DT_MAX_BRW_SIZE;
502 err = obd_connect(NULL, &sbi->ll_dt_exp, sbi->ll_dt_obd,
503 &sbi->ll_sb_uuid, data, sbi->ll_cache);
505 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
506 "recovery, of which this client is not a "
507 "part. Please wait for recovery to "
508 "complete, abort, or time out.\n", dt);
511 CERROR("%s: Cannot connect to %s: rc = %d\n",
512 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
516 sbi->ll_dt_exp->exp_connect_data = *data;
518 /* Don't change value if it was specified in the config log */
519 if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages == -1) {
520 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
521 max_t(unsigned long, SBI_DEFAULT_READAHEAD_WHOLE_MAX,
522 (data->ocd_brw_size >> PAGE_SHIFT));
523 if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages >
524 sbi->ll_ra_info.ra_max_pages_per_file)
525 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
526 sbi->ll_ra_info.ra_max_pages_per_file;
529 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
530 LUSTRE_SEQ_METADATA);
532 CERROR("%s: Can't init data layer FID infrastructure, "
533 "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
537 mutex_lock(&sbi->ll_lco.lco_lock);
538 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
539 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
540 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
541 mutex_unlock(&sbi->ll_lco.lco_lock);
543 fid_zero(&sbi->ll_root_fid);
544 err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
547 CERROR("cannot mds_connect: rc = %d\n", err);
548 GOTO(out_lock_cn_cb, err);
550 if (!fid_is_sane(&sbi->ll_root_fid)) {
551 CERROR("%s: Invalid root fid "DFID" during mount\n",
552 sbi->ll_md_exp->exp_obd->obd_name,
553 PFID(&sbi->ll_root_fid));
554 GOTO(out_lock_cn_cb, err = -EINVAL);
556 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
558 sb->s_op = &lustre_super_operations;
559 sb->s_xattr = ll_xattr_handlers;
560 #if THREAD_SIZE >= 8192 /*b=17630*/
561 sb->s_export_op = &lustre_export_operations;
565 * XXX: move this to after cbd setup? */
566 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
567 if (sbi->ll_flags & LL_SBI_ACL)
568 valid |= OBD_MD_FLACL;
570 OBD_ALLOC_PTR(op_data);
572 GOTO(out_lock_cn_cb, err = -ENOMEM);
574 op_data->op_fid1 = sbi->ll_root_fid;
575 op_data->op_mode = 0;
576 op_data->op_valid = valid;
578 err = md_getattr(sbi->ll_md_exp, op_data, &request);
580 OBD_FREE_PTR(op_data);
582 CERROR("%s: md_getattr failed for root: rc = %d\n",
583 sbi->ll_md_exp->exp_obd->obd_name, err);
584 GOTO(out_lock_cn_cb, err);
587 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
588 sbi->ll_md_exp, &lmd);
590 CERROR("failed to understand root inode md: rc = %d\n", err);
591 ptlrpc_req_finished(request);
592 GOTO(out_lock_cn_cb, err);
595 LASSERT(fid_is_sane(&sbi->ll_root_fid));
596 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
597 sbi->ll_flags & LL_SBI_32BIT_API),
599 md_free_lustre_md(sbi->ll_md_exp, &lmd);
600 ptlrpc_req_finished(request);
603 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
605 posix_acl_release(lmd.posix_acl);
606 lmd.posix_acl = NULL;
609 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
611 CERROR("%s: bad ll_iget() for root: rc = %d\n",
612 sbi->ll_fsname, err);
616 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
617 if (sbi->ll_checksum_set) {
618 err = obd_set_info_async(NULL, sbi->ll_dt_exp,
619 sizeof(KEY_CHECKSUM), KEY_CHECKSUM,
620 sizeof(checksum), &checksum, NULL);
622 CERROR("%s: Set checksum failed: rc = %d\n",
623 sbi->ll_dt_exp->exp_obd->obd_name, err);
629 sb->s_root = d_make_root(root);
630 if (sb->s_root == NULL) {
632 CERROR("%s: can't make root dentry: rc = %d\n",
633 sbi->ll_fsname, err);
637 sbi->ll_sdev_orig = sb->s_dev;
639 /* We set sb->s_dev equal on all lustre clients in order to support
640 * NFS export clustering. NFSD requires that the FSID be the same
642 /* s_dev is also used in lt_compare() to compare two fs, but that is
643 * only a node-local comparison. */
644 uuid = obd_get_uuid(sbi->ll_md_exp);
646 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
653 if (sbi->ll_dt_obd) {
654 err = sysfs_create_link(&sbi->ll_kset.kobj,
655 &sbi->ll_dt_obd->obd_kset.kobj,
656 sbi->ll_dt_obd->obd_type->typ_name);
658 CERROR("%s: could not register %s in llite: rc = %d\n",
659 dt, sbi->ll_fsname, err);
664 if (sbi->ll_md_obd) {
665 err = sysfs_create_link(&sbi->ll_kset.kobj,
666 &sbi->ll_md_obd->obd_kset.kobj,
667 sbi->ll_md_obd->obd_type->typ_name);
669 CERROR("%s: could not register %s in llite: rc = %d\n",
670 md, sbi->ll_fsname, err);
680 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
682 obd_disconnect(sbi->ll_dt_exp);
683 sbi->ll_dt_exp = NULL;
684 sbi->ll_dt_obd = NULL;
686 obd_fid_fini(sbi->ll_md_exp->exp_obd);
688 obd_disconnect(sbi->ll_md_exp);
689 sbi->ll_md_exp = NULL;
690 sbi->ll_md_obd = NULL;
699 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
703 size = sizeof(*lmmsize);
704 rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
705 KEY_MAX_EASIZE, &size, lmmsize);
707 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
708 sbi->ll_dt_exp->exp_obd->obd_name, rc);
712 CDEBUG(D_INFO, "max LOV ea size: %d\n", *lmmsize);
715 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
716 KEY_MAX_EASIZE, &size, lmmsize);
718 CERROR("Get max mdsize error rc %d\n", rc);
720 CDEBUG(D_INFO, "max LMV ea size: %d\n", *lmmsize);
726 * Get the value of the default_easize parameter.
728 * \see client_obd::cl_default_mds_easize
730 * \param[in] sbi superblock info for this filesystem
731 * \param[out] lmmsize pointer to storage location for value
733 * \retval 0 on success
734 * \retval negative negated errno on failure
736 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
741 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
742 KEY_DEFAULT_EASIZE, &size, lmmsize);
744 CERROR("Get default mdsize error rc %d\n", rc);
750 * Set the default_easize parameter to the given value.
752 * \see client_obd::cl_default_mds_easize
754 * \param[in] sbi superblock info for this filesystem
755 * \param[in] lmmsize the size to set
757 * \retval 0 on success
758 * \retval negative negated errno on failure
760 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
764 if (lmmsize < sizeof(struct lov_mds_md) ||
765 lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
768 rc = obd_set_info_async(NULL, sbi->ll_md_exp,
769 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
770 sizeof(int), &lmmsize, NULL);
775 static void client_common_put_super(struct super_block *sb)
777 struct ll_sb_info *sbi = ll_s2sbi(sb);
782 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
783 obd_disconnect(sbi->ll_dt_exp);
784 sbi->ll_dt_exp = NULL;
786 ll_debugfs_unregister_super(sb);
788 obd_fid_fini(sbi->ll_md_exp->exp_obd);
789 obd_disconnect(sbi->ll_md_exp);
790 sbi->ll_md_exp = NULL;
795 void ll_kill_super(struct super_block *sb)
797 struct ll_sb_info *sbi;
801 if (!(sb->s_flags & SB_ACTIVE))
805 /* we need restore s_dev from changed for clustred NFS before put_super
806 * because new kernels have cached s_dev and change sb->s_dev in
807 * put_super not affected real removing devices */
809 sb->s_dev = sbi->ll_sdev_orig;
811 /* wait running statahead threads to quit */
812 while (atomic_read(&sbi->ll_sa_running) > 0) {
813 set_current_state(TASK_UNINTERRUPTIBLE);
814 schedule_timeout(cfs_time_seconds(1) >> 3);
821 static inline int ll_set_opt(const char *opt, char *data, int fl)
823 if (strncmp(opt, data, strlen(opt)) != 0)
829 /* non-client-specific mount options are parsed in lmd_parse */
830 static int ll_options(char *options, struct ll_sb_info *sbi)
833 char *s1 = options, *s2;
834 int *flags = &sbi->ll_flags;
840 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
843 CDEBUG(D_SUPER, "next opt=%s\n", s1);
844 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
849 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
854 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
859 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
864 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
869 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
874 tmp = ll_set_opt("context", s1, 1);
877 tmp = ll_set_opt("fscontext", s1, 1);
880 tmp = ll_set_opt("defcontext", s1, 1);
883 tmp = ll_set_opt("rootcontext", s1, 1);
886 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
891 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
897 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
900 sbi->ll_checksum_set = 1;
903 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
906 sbi->ll_checksum_set = 1;
909 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
914 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
919 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
924 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
929 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
934 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
939 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
944 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
949 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
955 s2 = strchr(s1, ',');
963 void ll_lli_init(struct ll_inode_info *lli)
965 lli->lli_inode_magic = LLI_INODE_MAGIC;
967 spin_lock_init(&lli->lli_lock);
968 lli->lli_posix_acl = NULL;
969 /* Do not set lli_fid, it has been initialized already. */
970 fid_zero(&lli->lli_pfid);
971 lli->lli_mds_read_och = NULL;
972 lli->lli_mds_write_och = NULL;
973 lli->lli_mds_exec_och = NULL;
974 lli->lli_open_fd_read_count = 0;
975 lli->lli_open_fd_write_count = 0;
976 lli->lli_open_fd_exec_count = 0;
977 mutex_init(&lli->lli_och_mutex);
978 spin_lock_init(&lli->lli_agl_lock);
979 spin_lock_init(&lli->lli_layout_lock);
980 ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
981 lli->lli_clob = NULL;
983 init_rwsem(&lli->lli_xattrs_list_rwsem);
984 mutex_init(&lli->lli_xattrs_enq_lock);
986 LASSERT(lli->lli_vfs_inode.i_mode != 0);
987 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
988 lli->lli_opendir_key = NULL;
990 spin_lock_init(&lli->lli_sa_lock);
991 lli->lli_opendir_pid = 0;
992 lli->lli_sa_enabled = 0;
993 init_rwsem(&lli->lli_lsm_sem);
995 mutex_init(&lli->lli_size_mutex);
996 lli->lli_symlink_name = NULL;
997 init_rwsem(&lli->lli_trunc_sem);
998 range_lock_tree_init(&lli->lli_write_tree);
999 init_rwsem(&lli->lli_glimpse_sem);
1000 lli->lli_glimpse_time = ktime_set(0, 0);
1001 INIT_LIST_HEAD(&lli->lli_agl_list);
1002 lli->lli_agl_index = 0;
1003 lli->lli_async_rc = 0;
1004 spin_lock_init(&lli->lli_heat_lock);
1005 obd_heat_clear(lli->lli_heat_instances, OBD_HEAT_COUNT);
1006 lli->lli_heat_flags = 0;
1007 mutex_init(&lli->lli_pcc_lock);
1008 lli->lli_pcc_state = PCC_STATE_FL_NONE;
1009 lli->lli_pcc_inode = NULL;
1010 mutex_init(&lli->lli_group_mutex);
1011 lli->lli_group_users = 0;
1012 lli->lli_group_gid = 0;
1014 mutex_init(&lli->lli_layout_mutex);
1015 memset(lli->lli_jobid, 0, sizeof(lli->lli_jobid));
1018 #define MAX_STRING_SIZE 128
1020 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1022 #define LSI_BDI_INITIALIZED 0x00400000
1024 #ifndef HAVE_BDI_CAP_MAP_COPY
1025 # define BDI_CAP_MAP_COPY 0
1028 static int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
1030 struct lustre_sb_info *lsi = s2lsi(sb);
1031 char buf[MAX_STRING_SIZE];
1035 err = bdi_init(&lsi->lsi_bdi);
1039 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1040 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1041 lsi->lsi_bdi.name = "lustre";
1042 va_start(args, fmt);
1043 vsnprintf(buf, MAX_STRING_SIZE, fmt, args);
1045 err = bdi_register(&lsi->lsi_bdi, NULL, "%s", buf);
1048 sb->s_bdi = &lsi->lsi_bdi;
1052 #endif /* !HAVE_SUPER_SETUP_BDI_NAME */
1054 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
1056 struct lustre_profile *lprof = NULL;
1057 struct lustre_sb_info *lsi = s2lsi(sb);
1058 struct ll_sb_info *sbi = NULL;
1059 char *dt = NULL, *md = NULL;
1060 char *profilenm = get_profile_name(sb);
1061 struct config_llog_instance *cfg;
1062 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1063 const int instlen = 16 + 2;
1064 unsigned long cfg_instance = ll_get_cfg_instance(sb);
1065 char name[MAX_STRING_SIZE];
1074 /* for ASLR, to map between cfg_instance and hashed ptr */
1075 CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1076 profilenm, cfg_instance, sb);
1078 try_module_get(THIS_MODULE);
1082 GOTO(out_free_cfg, err = -ENOMEM);
1084 /* client additional sb info */
1085 lsi->lsi_llsbi = sbi = ll_init_sbi();
1087 GOTO(out_free_cfg, err = PTR_ERR(sbi));
1089 err = ll_options(lsi->lsi_lmd->lmd_opts, sbi);
1091 GOTO(out_free_cfg, err);
1093 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1094 sb->s_d_op = &ll_d_ops;
1097 generate_random_uuid(uuid.b);
1098 snprintf(sbi->ll_sb_uuid.uuid, UUID_SIZE, "%pU", uuid.b);
1100 CDEBUG(D_CONFIG, "llite sb uuid: %s\n", sbi->ll_sb_uuid.uuid);
1103 len = strlen(profilenm);
1104 ptr = strrchr(profilenm, '-');
1105 if (ptr && (strcmp(ptr, "-client") == 0))
1108 if (len > LUSTRE_MAXFSNAME) {
1109 if (unlikely(len >= MAX_STRING_SIZE))
1110 len = MAX_STRING_SIZE - 1;
1111 strncpy(name, profilenm, len);
1113 err = -ENAMETOOLONG;
1114 CERROR("%s: fsname longer than %u characters: rc = %d\n",
1115 name, LUSTRE_MAXFSNAME, err);
1116 GOTO(out_free_cfg, err);
1118 strncpy(sbi->ll_fsname, profilenm, len);
1119 sbi->ll_fsname[len] = '\0';
1122 snprintf(name, sizeof(name), "%.*s-%016lx", len,
1123 profilenm, cfg_instance);
1125 err = super_setup_bdi_name(sb, "%s", name);
1127 GOTO(out_free_cfg, err);
1129 /* Call ll_debugfs_register_super() before lustre_process_log()
1130 * so that "llite.*.*" params can be processed correctly.
1132 err = ll_debugfs_register_super(sb, name);
1134 CERROR("%s: could not register mountpoint in llite: rc = %d\n",
1135 sbi->ll_fsname, err);
1139 /* The cfg_instance is a value unique to this super, in case some
1140 * joker tries to mount the same fs at two mount points.
1142 cfg->cfg_instance = cfg_instance;
1143 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1144 cfg->cfg_callback = class_config_llog_handler;
1145 cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1146 /* set up client obds */
1147 err = lustre_process_log(sb, profilenm, cfg);
1149 GOTO(out_debugfs, err);
1151 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1152 lprof = class_get_profile(profilenm);
1153 if (lprof == NULL) {
1154 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1155 " read from the MGS. Does that filesystem "
1156 "exist?\n", profilenm);
1157 GOTO(out_debugfs, err = -EINVAL);
1159 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1160 lprof->lp_md, lprof->lp_dt);
1162 dt_len = strlen(lprof->lp_dt) + instlen + 2;
1163 OBD_ALLOC(dt, dt_len);
1165 GOTO(out_profile, err = -ENOMEM);
1166 snprintf(dt, dt_len - 1, "%s-%016lx", lprof->lp_dt, cfg_instance);
1168 md_len = strlen(lprof->lp_md) + instlen + 2;
1169 OBD_ALLOC(md, md_len);
1171 GOTO(out_free_dt, err = -ENOMEM);
1172 snprintf(md, md_len - 1, "%s-%016lx", lprof->lp_md, cfg_instance);
1174 /* connections, registrations, sb setup */
1175 err = client_common_fill_super(sb, md, dt, mnt);
1177 GOTO(out_free_md, err);
1179 sbi->ll_client_common_fill_super_succeeded = 1;
1183 OBD_FREE(md, md_len);
1186 OBD_FREE(dt, dt_len);
1189 class_put_profile(lprof);
1192 ll_debugfs_unregister_super(sb);
1199 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1200 LCONSOLE_WARN("Mounted %s\n", profilenm);
1202 } /* ll_fill_super */
1204 void ll_put_super(struct super_block *sb)
1206 struct config_llog_instance cfg, params_cfg;
1207 struct obd_device *obd;
1208 struct lustre_sb_info *lsi = s2lsi(sb);
1209 struct ll_sb_info *sbi = ll_s2sbi(sb);
1210 char *profilenm = get_profile_name(sb);
1211 unsigned long cfg_instance = ll_get_cfg_instance(sb);
1213 int next, force = 1, rc = 0;
1217 GOTO(out_no_sbi, 0);
1219 /* Should replace instance_id with something better for ASLR */
1220 CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1221 profilenm, cfg_instance, sb);
1223 cfg.cfg_instance = cfg_instance;
1224 lustre_end_log(sb, profilenm, &cfg);
1226 params_cfg.cfg_instance = cfg_instance;
1227 lustre_end_log(sb, PARAMS_FILENAME, ¶ms_cfg);
1229 if (sbi->ll_md_exp) {
1230 obd = class_exp2obd(sbi->ll_md_exp);
1232 force = obd->obd_force;
1235 /* Wait for unstable pages to be committed to stable storage */
1237 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1238 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1239 atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1243 ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1244 if (force == 0 && rc != -EINTR)
1245 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1247 /* We need to set force before the lov_disconnect in
1248 lustre_common_put_super, since l_d cleans up osc's as well. */
1251 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1253 obd->obd_force = force;
1257 if (sbi->ll_client_common_fill_super_succeeded) {
1258 /* Only if client_common_fill_super succeeded */
1259 client_common_put_super(sb);
1263 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1264 class_manual_cleanup(obd);
1267 if (sbi->ll_flags & LL_SBI_VERBOSE)
1268 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1271 class_del_profile(profilenm);
1273 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1274 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1275 bdi_destroy(&lsi->lsi_bdi);
1276 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1281 lsi->lsi_llsbi = NULL;
1283 lustre_common_put_super(sb);
1285 cl_env_cache_purge(~0);
1287 module_put(THIS_MODULE);
1290 } /* client_put_super */
1292 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1294 struct inode *inode = NULL;
1296 /* NOTE: we depend on atomic igrab() -bzzz */
1297 lock_res_and_lock(lock);
1298 if (lock->l_resource->lr_lvb_inode) {
1299 struct ll_inode_info * lli;
1300 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1301 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1302 inode = igrab(lock->l_resource->lr_lvb_inode);
1304 inode = lock->l_resource->lr_lvb_inode;
1305 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1306 D_WARNING, lock, "lr_lvb_inode %p is "
1307 "bogus: magic %08x",
1308 lock->l_resource->lr_lvb_inode,
1309 lli->lli_inode_magic);
1313 unlock_res_and_lock(lock);
1317 void ll_dir_clear_lsm_md(struct inode *inode)
1319 struct ll_inode_info *lli = ll_i2info(inode);
1321 LASSERT(S_ISDIR(inode->i_mode));
1323 if (lli->lli_lsm_md) {
1324 lmv_free_memmd(lli->lli_lsm_md);
1325 lli->lli_lsm_md = NULL;
1328 if (lli->lli_default_lsm_md) {
1329 lmv_free_memmd(lli->lli_default_lsm_md);
1330 lli->lli_default_lsm_md = NULL;
1334 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1335 const struct lu_fid *fid,
1336 struct lustre_md *md)
1338 struct ll_sb_info *sbi = ll_s2sbi(sb);
1339 struct mdt_body *body = md->body;
1340 struct inode *inode;
1344 ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1345 inode = iget_locked(sb, ino);
1346 if (inode == NULL) {
1347 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1348 sbi->ll_fsname, PFID(fid));
1349 RETURN(ERR_PTR(-ENOENT));
1352 if (inode->i_state & I_NEW) {
1353 struct ll_inode_info *lli = ll_i2info(inode);
1354 struct lmv_stripe_md *lsm = md->lmv;
1356 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1357 (body->mbo_mode & S_IFMT);
1358 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1361 inode->i_mtime.tv_sec = 0;
1362 inode->i_atime.tv_sec = 0;
1363 inode->i_ctime.tv_sec = 0;
1366 #ifdef HAVE_BACKING_DEV_INFO
1367 /* initializing backing dev info. */
1368 inode->i_mapping->backing_dev_info =
1369 &s2lsi(inode->i_sb)->lsi_bdi;
1371 inode->i_op = &ll_dir_inode_operations;
1372 inode->i_fop = &ll_dir_operations;
1373 lli->lli_fid = *fid;
1376 LASSERT(lsm != NULL);
1377 /* master object FID */
1378 lli->lli_pfid = body->mbo_fid1;
1379 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1380 lli, PFID(fid), PFID(&lli->lli_pfid));
1381 unlock_new_inode(inode);
1387 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1390 struct lmv_stripe_md *lsm = md->lmv;
1391 struct ll_inode_info *lli = ll_i2info(inode);
1394 LASSERT(lsm != NULL);
1396 CDEBUG(D_INODE, "%s: "DFID" set dir layout:\n",
1397 ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid));
1398 lsm_md_dump(D_INODE, lsm);
1400 if (!lmv_dir_striped(lsm))
1403 /* XXX sigh, this lsm_root initialization should be in
1404 * LMV layer, but it needs ll_iget right now, so we
1405 * put this here right now. */
1406 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1407 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1408 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1410 if (!fid_is_sane(fid))
1413 /* Unfortunately ll_iget will call ll_update_inode,
1414 * where the initialization of slave inode is slightly
1415 * different, so it reset lsm_md to NULL to avoid
1416 * initializing lsm for slave inode. */
1417 lsm->lsm_md_oinfo[i].lmo_root =
1418 ll_iget_anon_dir(inode->i_sb, fid, md);
1419 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1420 int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1422 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1424 iput(lsm->lsm_md_oinfo[i].lmo_root);
1425 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1431 lli->lli_lsm_md = lsm;
1436 static void ll_update_default_lsm_md(struct inode *inode, struct lustre_md *md)
1438 struct ll_inode_info *lli = ll_i2info(inode);
1440 if (!md->default_lmv) {
1441 /* clear default lsm */
1442 if (lli->lli_default_lsm_md) {
1443 down_write(&lli->lli_lsm_sem);
1444 if (lli->lli_default_lsm_md) {
1445 lmv_free_memmd(lli->lli_default_lsm_md);
1446 lli->lli_default_lsm_md = NULL;
1448 up_write(&lli->lli_lsm_sem);
1450 } else if (lli->lli_default_lsm_md) {
1451 /* update default lsm if it changes */
1452 down_read(&lli->lli_lsm_sem);
1453 if (lli->lli_default_lsm_md &&
1454 !lsm_md_eq(lli->lli_default_lsm_md, md->default_lmv)) {
1455 up_read(&lli->lli_lsm_sem);
1456 down_write(&lli->lli_lsm_sem);
1457 if (lli->lli_default_lsm_md)
1458 lmv_free_memmd(lli->lli_default_lsm_md);
1459 lli->lli_default_lsm_md = md->default_lmv;
1460 lsm_md_dump(D_INODE, md->default_lmv);
1461 md->default_lmv = NULL;
1462 up_write(&lli->lli_lsm_sem);
1464 up_read(&lli->lli_lsm_sem);
1467 /* init default lsm */
1468 down_write(&lli->lli_lsm_sem);
1469 lli->lli_default_lsm_md = md->default_lmv;
1470 lsm_md_dump(D_INODE, md->default_lmv);
1471 md->default_lmv = NULL;
1472 up_write(&lli->lli_lsm_sem);
1476 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1478 struct ll_inode_info *lli = ll_i2info(inode);
1479 struct lmv_stripe_md *lsm = md->lmv;
1484 LASSERT(S_ISDIR(inode->i_mode));
1485 CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1486 PFID(ll_inode2fid(inode)));
1488 /* update default LMV */
1489 if (md->default_lmv)
1490 ll_update_default_lsm_md(inode, md);
1493 * no striped information from request, lustre_md from req does not
1494 * include stripeEA, see ll_md_setattr()
1500 * normally dir layout doesn't change, only take read lock to check
1501 * that to avoid blocking other MD operations.
1503 if (lli->lli_lsm_md)
1504 down_read(&lli->lli_lsm_sem);
1506 down_write(&lli->lli_lsm_sem);
1509 * if dir layout mismatch, check whether version is increased, which
1510 * means layout is changed, this happens in dir migration and lfsck.
1512 * foreign LMV should not change.
1514 if (lli->lli_lsm_md && !lsm_md_eq(lli->lli_lsm_md, lsm)) {
1515 if (lmv_dir_striped(lli->lli_lsm_md) &&
1516 lsm->lsm_md_layout_version <=
1517 lli->lli_lsm_md->lsm_md_layout_version) {
1518 CERROR("%s: "DFID" dir layout mismatch:\n",
1519 ll_i2sbi(inode)->ll_fsname,
1520 PFID(&lli->lli_fid));
1521 lsm_md_dump(D_ERROR, lli->lli_lsm_md);
1522 lsm_md_dump(D_ERROR, lsm);
1523 GOTO(unlock, rc = -EINVAL);
1526 /* layout changed, switch to write lock */
1527 up_read(&lli->lli_lsm_sem);
1528 down_write(&lli->lli_lsm_sem);
1529 ll_dir_clear_lsm_md(inode);
1532 /* set directory layout */
1533 if (!lli->lli_lsm_md) {
1534 struct cl_attr *attr;
1536 rc = ll_init_lsm_md(inode, md);
1537 up_write(&lli->lli_lsm_sem);
1541 /* set md->lmv to NULL, so the following free lustre_md
1542 * will not free this lsm */
1546 * md_merge_attr() may take long, since lsm is already set,
1547 * switch to read lock.
1549 down_read(&lli->lli_lsm_sem);
1551 if (!lmv_dir_striped(lli->lli_lsm_md))
1554 OBD_ALLOC_PTR(attr);
1556 GOTO(unlock, rc = -ENOMEM);
1558 /* validate the lsm */
1559 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1560 ll_md_blocking_ast);
1566 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1567 md->body->mbo_nlink = attr->cat_nlink;
1568 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1569 md->body->mbo_size = attr->cat_size;
1570 if (md->body->mbo_valid & OBD_MD_FLATIME)
1571 md->body->mbo_atime = attr->cat_atime;
1572 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1573 md->body->mbo_ctime = attr->cat_ctime;
1574 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1575 md->body->mbo_mtime = attr->cat_mtime;
1580 up_read(&lli->lli_lsm_sem);
1585 void ll_clear_inode(struct inode *inode)
1587 struct ll_inode_info *lli = ll_i2info(inode);
1588 struct ll_sb_info *sbi = ll_i2sbi(inode);
1592 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1593 PFID(ll_inode2fid(inode)), inode);
1595 if (S_ISDIR(inode->i_mode)) {
1596 /* these should have been cleared in ll_file_release */
1597 LASSERT(lli->lli_opendir_key == NULL);
1598 LASSERT(lli->lli_sai == NULL);
1599 LASSERT(lli->lli_opendir_pid == 0);
1601 pcc_inode_free(inode);
1604 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1606 LASSERT(!lli->lli_open_fd_write_count);
1607 LASSERT(!lli->lli_open_fd_read_count);
1608 LASSERT(!lli->lli_open_fd_exec_count);
1610 if (lli->lli_mds_write_och)
1611 ll_md_real_close(inode, FMODE_WRITE);
1612 if (lli->lli_mds_exec_och)
1613 ll_md_real_close(inode, FMODE_EXEC);
1614 if (lli->lli_mds_read_och)
1615 ll_md_real_close(inode, FMODE_READ);
1617 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1618 OBD_FREE(lli->lli_symlink_name,
1619 strlen(lli->lli_symlink_name) + 1);
1620 lli->lli_symlink_name = NULL;
1623 ll_xattr_cache_destroy(inode);
1625 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
1626 forget_all_cached_acls(inode);
1627 if (lli->lli_posix_acl) {
1628 posix_acl_release(lli->lli_posix_acl);
1629 lli->lli_posix_acl = NULL;
1632 lli->lli_inode_magic = LLI_INODE_DEAD;
1634 if (S_ISDIR(inode->i_mode))
1635 ll_dir_clear_lsm_md(inode);
1636 else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1637 LASSERT(list_empty(&lli->lli_agl_list));
1640 * XXX This has to be done before lsm is freed below, because
1641 * cl_object still uses inode lsm.
1643 cl_inode_fini(inode);
1648 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1650 struct lustre_md md;
1651 struct inode *inode = dentry->d_inode;
1652 struct ll_sb_info *sbi = ll_i2sbi(inode);
1653 struct ptlrpc_request *request = NULL;
1657 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1658 LUSTRE_OPC_ANY, NULL);
1659 if (IS_ERR(op_data))
1660 RETURN(PTR_ERR(op_data));
1662 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1664 ptlrpc_req_finished(request);
1665 if (rc == -ENOENT) {
1667 /* Unlinked special device node? Or just a race?
1668 * Pretend we done everything. */
1669 if (!S_ISREG(inode->i_mode) &&
1670 !S_ISDIR(inode->i_mode)) {
1671 ia_valid = op_data->op_attr.ia_valid;
1672 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1673 rc = simple_setattr(dentry, &op_data->op_attr);
1674 op_data->op_attr.ia_valid = ia_valid;
1676 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1677 CERROR("md_setattr fails: rc = %d\n", rc);
1682 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1683 sbi->ll_md_exp, &md);
1685 ptlrpc_req_finished(request);
1689 ia_valid = op_data->op_attr.ia_valid;
1690 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1691 * cache is not cleared yet. */
1692 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1693 if (S_ISREG(inode->i_mode))
1695 rc = simple_setattr(dentry, &op_data->op_attr);
1696 if (S_ISREG(inode->i_mode))
1697 inode_unlock(inode);
1698 op_data->op_attr.ia_valid = ia_valid;
1700 rc = ll_update_inode(inode, &md);
1701 ptlrpc_req_finished(request);
1706 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1707 * object(s) determine the file size and mtime. Otherwise, the MDS will
1708 * keep these values until such a time that objects are allocated for it.
1709 * We do the MDS operations first, as it is checking permissions for us.
1710 * We don't to the MDS RPC if there is nothing that we want to store there,
1711 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1712 * going to do an RPC anyways.
1714 * If we are doing a truncate, we will send the mtime and ctime updates
1715 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1716 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1719 * In case of HSMimport, we only set attr on MDS.
1721 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr,
1722 enum op_xvalid xvalid, bool hsm_import)
1724 struct inode *inode = dentry->d_inode;
1725 struct ll_inode_info *lli = ll_i2info(inode);
1726 struct md_op_data *op_data = NULL;
1731 CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1732 "valid %x, hsm_import %d\n",
1733 ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid),
1734 inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1737 if (attr->ia_valid & ATTR_SIZE) {
1738 /* Check new size against VFS/VM file size limit and rlimit */
1739 rc = inode_newsize_ok(inode, attr->ia_size);
1743 /* The maximum Lustre file size is variable, based on the
1744 * OST maximum object size and number of stripes. This
1745 * needs another check in addition to the VFS check above. */
1746 if (attr->ia_size > ll_file_maxbytes(inode)) {
1747 CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
1748 PFID(&lli->lli_fid), attr->ia_size,
1749 ll_file_maxbytes(inode));
1753 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1756 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1757 if (attr->ia_valid & TIMES_SET_FLAGS) {
1758 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1759 !cfs_capable(CFS_CAP_FOWNER))
1763 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1764 if (!(xvalid & OP_XVALID_CTIME_SET) &&
1765 (attr->ia_valid & ATTR_CTIME)) {
1766 attr->ia_ctime = current_time(inode);
1767 xvalid |= OP_XVALID_CTIME_SET;
1769 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1770 (attr->ia_valid & ATTR_ATIME)) {
1771 attr->ia_atime = current_time(inode);
1772 attr->ia_valid |= ATTR_ATIME_SET;
1774 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1775 (attr->ia_valid & ATTR_MTIME)) {
1776 attr->ia_mtime = current_time(inode);
1777 attr->ia_valid |= ATTR_MTIME_SET;
1780 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1781 CDEBUG(D_INODE, "setting mtime %lld, ctime %lld, now = %lld\n",
1782 (s64)attr->ia_mtime.tv_sec, (s64)attr->ia_ctime.tv_sec,
1783 ktime_get_real_seconds());
1785 if (S_ISREG(inode->i_mode)) {
1786 if (attr->ia_valid & ATTR_SIZE)
1787 inode_dio_write_done(inode);
1788 inode_unlock(inode);
1791 /* We always do an MDS RPC, even if we're only changing the size;
1792 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1794 OBD_ALLOC_PTR(op_data);
1795 if (op_data == NULL)
1796 GOTO(out, rc = -ENOMEM);
1798 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1799 /* If we are changing file size, file content is
1800 * modified, flag it.
1802 xvalid |= OP_XVALID_OWNEROVERRIDE;
1803 op_data->op_bias |= MDS_DATA_MODIFIED;
1804 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1807 if (attr->ia_valid & ATTR_FILE) {
1808 struct ll_file_data *fd = LUSTRE_FPRIVATE(attr->ia_file);
1810 if (fd->fd_lease_och)
1811 op_data->op_bias |= MDS_TRUNC_KEEP_LEASE;
1814 op_data->op_attr = *attr;
1815 op_data->op_xvalid = xvalid;
1817 rc = ll_md_setattr(dentry, op_data);
1821 if (!S_ISREG(inode->i_mode) || hsm_import)
1824 if (attr->ia_valid & (ATTR_SIZE | ATTR_ATIME | ATTR_ATIME_SET |
1825 ATTR_MTIME | ATTR_MTIME_SET | ATTR_CTIME) ||
1826 xvalid & OP_XVALID_CTIME_SET) {
1827 bool cached = false;
1829 rc = pcc_inode_setattr(inode, attr, &cached);
1832 CERROR("%s: PCC inode "DFID" setattr failed: "
1834 ll_i2sbi(inode)->ll_fsname,
1835 PFID(&lli->lli_fid), rc);
1839 /* For truncate and utimes sending attributes to OSTs,
1840 * setting mtime/atime to the past will be performed
1841 * under PW [0:EOF] extent lock (new_size:EOF for
1842 * truncate). It may seem excessive to send mtime/atime
1843 * updates to OSTs when not setting times to past, but
1844 * it is necessary due to possible time
1845 * de-synchronization between MDT inode and OST objects
1847 rc = cl_setattr_ost(lli->lli_clob, attr, xvalid, 0);
1851 /* If the file was restored, it needs to set dirty flag.
1853 * We've already sent MDS_DATA_MODIFIED flag in
1854 * ll_md_setattr() for truncate. However, the MDT refuses to
1855 * set the HS_DIRTY flag on released files, so we have to set
1856 * it again if the file has been restored. Please check how
1857 * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1859 * Please notice that if the file is not released, the previous
1860 * MDS_DATA_MODIFIED has taken effect and usually
1861 * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1862 * This way we can save an RPC for common open + trunc
1864 if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1865 struct hsm_state_set hss = {
1866 .hss_valid = HSS_SETMASK,
1867 .hss_setmask = HS_DIRTY,
1871 rc2 = ll_hsm_state_set(inode, &hss);
1872 /* truncate and write can happen at the same time, so that
1873 * the file can be set modified even though the file is not
1874 * restored from released state, and ll_hsm_state_set() is
1875 * not applicable for the file, and rc2 < 0 is normal in this
1878 CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1879 PFID(ll_inode2fid(inode)), rc2);
1884 if (op_data != NULL)
1885 ll_finish_md_op_data(op_data);
1887 if (S_ISREG(inode->i_mode)) {
1889 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1890 inode_dio_wait(inode);
1891 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
1892 * flag. ll_update_inode (called from ll_md_setattr), clears
1893 * inode flags, so there is a gap where S_NOSEC is not set.
1894 * This can cause a writer to take the i_mutex unnecessarily,
1895 * but this is safe to do and should be rare. */
1896 inode_has_no_xattr(inode);
1899 ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1900 LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1905 int ll_setattr(struct dentry *de, struct iattr *attr)
1907 int mode = de->d_inode->i_mode;
1908 enum op_xvalid xvalid = 0;
1910 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1911 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1912 xvalid |= OP_XVALID_OWNEROVERRIDE;
1914 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1915 (ATTR_SIZE|ATTR_MODE)) &&
1916 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1917 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1918 !(attr->ia_mode & S_ISGID))))
1919 attr->ia_valid |= ATTR_FORCE;
1921 if ((attr->ia_valid & ATTR_MODE) &&
1923 !(attr->ia_mode & S_ISUID) &&
1924 !(attr->ia_valid & ATTR_KILL_SUID))
1925 attr->ia_valid |= ATTR_KILL_SUID;
1927 if ((attr->ia_valid & ATTR_MODE) &&
1928 ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1929 !(attr->ia_mode & S_ISGID) &&
1930 !(attr->ia_valid & ATTR_KILL_SGID))
1931 attr->ia_valid |= ATTR_KILL_SGID;
1933 return ll_setattr_raw(de, attr, xvalid, false);
1936 int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
1939 struct obd_statfs obd_osfs = { 0 };
1944 max_age = ktime_get_seconds() - sbi->ll_statfs_max_age;
1946 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1950 osfs->os_type = LL_SUPER_MAGIC;
1952 CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1953 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree, osfs->os_files);
1955 if (osfs->os_state & OS_STATE_SUM)
1958 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1959 flags |= OBD_STATFS_NODELAY;
1961 rc = obd_statfs(NULL, sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1962 if (rc) /* Possibly a filesystem with no OSTs. Report MDT totals. */
1965 CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1966 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1969 osfs->os_bsize = obd_osfs.os_bsize;
1970 osfs->os_blocks = obd_osfs.os_blocks;
1971 osfs->os_bfree = obd_osfs.os_bfree;
1972 osfs->os_bavail = obd_osfs.os_bavail;
1974 /* If we have _some_ OSTs, but don't have as many free objects on the
1975 * OSTs as inodes on the MDTs, reduce the reported number of inodes
1976 * to compensate, so that the "inodes in use" number is correct.
1977 * This should be kept in sync with lod_statfs() behaviour.
1979 if (obd_osfs.os_files && obd_osfs.os_ffree < osfs->os_ffree) {
1980 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1982 osfs->os_ffree = obd_osfs.os_ffree;
1988 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1990 struct super_block *sb = de->d_sb;
1991 struct obd_statfs osfs;
1992 __u64 fsid = huge_encode_dev(sb->s_dev);
1995 CDEBUG(D_VFSTRACE, "VFS Op: at %llu jiffies\n", get_jiffies_64());
1996 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STATFS, 1);
1998 /* Some amount of caching on the client is allowed */
1999 rc = ll_statfs_internal(ll_s2sbi(sb), &osfs, OBD_STATFS_SUM);
2003 statfs_unpack(sfs, &osfs);
2005 /* We need to downshift for all 32-bit kernels, because we can't
2006 * tell if the kernel is being called via sys_statfs64() or not.
2007 * Stop before overflowing f_bsize - in which case it is better
2008 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
2009 if (sizeof(long) < 8) {
2010 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
2013 osfs.os_blocks >>= 1;
2014 osfs.os_bfree >>= 1;
2015 osfs.os_bavail >>= 1;
2019 sfs->f_blocks = osfs.os_blocks;
2020 sfs->f_bfree = osfs.os_bfree;
2021 sfs->f_bavail = osfs.os_bavail;
2022 sfs->f_fsid.val[0] = (__u32)fsid;
2023 sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
2027 void ll_inode_size_lock(struct inode *inode)
2029 struct ll_inode_info *lli;
2031 LASSERT(!S_ISDIR(inode->i_mode));
2033 lli = ll_i2info(inode);
2034 mutex_lock(&lli->lli_size_mutex);
2037 void ll_inode_size_unlock(struct inode *inode)
2039 struct ll_inode_info *lli;
2041 lli = ll_i2info(inode);
2042 mutex_unlock(&lli->lli_size_mutex);
2045 void ll_update_inode_flags(struct inode *inode, int ext_flags)
2047 inode->i_flags = ll_ext_to_inode_flags(ext_flags);
2048 if (ext_flags & LUSTRE_PROJINHERIT_FL)
2049 ll_file_set_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
2051 ll_file_clear_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
2054 int ll_update_inode(struct inode *inode, struct lustre_md *md)
2056 struct ll_inode_info *lli = ll_i2info(inode);
2057 struct mdt_body *body = md->body;
2058 struct ll_sb_info *sbi = ll_i2sbi(inode);
2061 if (body->mbo_valid & OBD_MD_FLEASIZE) {
2062 rc = cl_file_inode_init(inode, md);
2067 if (S_ISDIR(inode->i_mode)) {
2068 rc = ll_update_lsm_md(inode, md);
2073 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
2074 if (body->mbo_valid & OBD_MD_FLACL) {
2075 spin_lock(&lli->lli_lock);
2076 if (lli->lli_posix_acl)
2077 posix_acl_release(lli->lli_posix_acl);
2078 lli->lli_posix_acl = md->posix_acl;
2079 spin_unlock(&lli->lli_lock);
2082 inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
2083 sbi->ll_flags & LL_SBI_32BIT_API);
2084 inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
2086 if (body->mbo_valid & OBD_MD_FLATIME) {
2087 if (body->mbo_atime > inode->i_atime.tv_sec)
2088 inode->i_atime.tv_sec = body->mbo_atime;
2089 lli->lli_atime = body->mbo_atime;
2092 if (body->mbo_valid & OBD_MD_FLMTIME) {
2093 if (body->mbo_mtime > inode->i_mtime.tv_sec) {
2095 "setting ino %lu mtime from %lld to %llu\n",
2096 inode->i_ino, (s64)inode->i_mtime.tv_sec,
2098 inode->i_mtime.tv_sec = body->mbo_mtime;
2100 lli->lli_mtime = body->mbo_mtime;
2103 if (body->mbo_valid & OBD_MD_FLCTIME) {
2104 if (body->mbo_ctime > inode->i_ctime.tv_sec)
2105 inode->i_ctime.tv_sec = body->mbo_ctime;
2106 lli->lli_ctime = body->mbo_ctime;
2109 /* Clear i_flags to remove S_NOSEC before permissions are updated */
2110 if (body->mbo_valid & OBD_MD_FLFLAGS)
2111 ll_update_inode_flags(inode, body->mbo_flags);
2112 if (body->mbo_valid & OBD_MD_FLMODE)
2113 inode->i_mode = (inode->i_mode & S_IFMT) |
2114 (body->mbo_mode & ~S_IFMT);
2116 if (body->mbo_valid & OBD_MD_FLTYPE)
2117 inode->i_mode = (inode->i_mode & ~S_IFMT) |
2118 (body->mbo_mode & S_IFMT);
2120 LASSERT(inode->i_mode != 0);
2121 if (S_ISREG(inode->i_mode))
2122 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
2123 LL_MAX_BLKSIZE_BITS);
2125 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
2127 if (body->mbo_valid & OBD_MD_FLUID)
2128 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
2129 if (body->mbo_valid & OBD_MD_FLGID)
2130 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
2131 if (body->mbo_valid & OBD_MD_FLPROJID)
2132 lli->lli_projid = body->mbo_projid;
2133 if (body->mbo_valid & OBD_MD_FLNLINK)
2134 set_nlink(inode, body->mbo_nlink);
2135 if (body->mbo_valid & OBD_MD_FLRDEV)
2136 inode->i_rdev = old_decode_dev(body->mbo_rdev);
2138 if (body->mbo_valid & OBD_MD_FLID) {
2139 /* FID shouldn't be changed! */
2140 if (fid_is_sane(&lli->lli_fid)) {
2141 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
2142 "Trying to change FID "DFID
2143 " to the "DFID", inode "DFID"(%p)\n",
2144 PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
2145 PFID(ll_inode2fid(inode)), inode);
2147 lli->lli_fid = body->mbo_fid1;
2151 LASSERT(fid_seq(&lli->lli_fid) != 0);
2153 if (body->mbo_valid & OBD_MD_FLSIZE) {
2154 i_size_write(inode, body->mbo_size);
2156 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
2157 PFID(ll_inode2fid(inode)),
2158 (unsigned long long)body->mbo_size);
2160 if (body->mbo_valid & OBD_MD_FLBLOCKS)
2161 inode->i_blocks = body->mbo_blocks;
2164 if (body->mbo_valid & OBD_MD_TSTATE) {
2165 /* Set LLIF_FILE_RESTORING if restore ongoing and
2166 * clear it when done to ensure to start again
2167 * glimpsing updated attrs
2169 if (body->mbo_t_state & MS_RESTORE)
2170 ll_file_set_flag(lli, LLIF_FILE_RESTORING);
2172 ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
2178 int ll_read_inode2(struct inode *inode, void *opaque)
2180 struct lustre_md *md = opaque;
2181 struct ll_inode_info *lli = ll_i2info(inode);
2185 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2186 PFID(&lli->lli_fid), inode);
2188 /* Core attributes from the MDS first. This is a new inode, and
2189 * the VFS doesn't zero times in the core inode so we have to do
2190 * it ourselves. They will be overwritten by either MDS or OST
2191 * attributes - we just need to make sure they aren't newer.
2193 inode->i_mtime.tv_sec = 0;
2194 inode->i_atime.tv_sec = 0;
2195 inode->i_ctime.tv_sec = 0;
2197 rc = ll_update_inode(inode, md);
2201 /* OIDEBUG(inode); */
2203 #ifdef HAVE_BACKING_DEV_INFO
2204 /* initializing backing dev info. */
2205 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2207 if (S_ISREG(inode->i_mode)) {
2208 struct ll_sb_info *sbi = ll_i2sbi(inode);
2209 inode->i_op = &ll_file_inode_operations;
2210 inode->i_fop = sbi->ll_fop;
2211 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2213 } else if (S_ISDIR(inode->i_mode)) {
2214 inode->i_op = &ll_dir_inode_operations;
2215 inode->i_fop = &ll_dir_operations;
2217 } else if (S_ISLNK(inode->i_mode)) {
2218 inode->i_op = &ll_fast_symlink_inode_operations;
2221 inode->i_op = &ll_special_inode_operations;
2223 init_special_inode(inode, inode->i_mode,
2232 void ll_delete_inode(struct inode *inode)
2234 struct ll_inode_info *lli = ll_i2info(inode);
2235 struct address_space *mapping = &inode->i_data;
2236 unsigned long nrpages;
2237 unsigned long flags;
2241 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL) {
2242 /* It is last chance to write out dirty pages,
2243 * otherwise we may lose data while umount.
2245 * If i_nlink is 0 then just discard data. This is safe because
2246 * local inode gets i_nlink 0 from server only for the last
2247 * unlink, so that file is not opened somewhere else
2249 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, inode->i_nlink ?
2250 CL_FSYNC_LOCAL : CL_FSYNC_DISCARD, 1);
2252 truncate_inode_pages_final(mapping);
2254 /* Workaround for LU-118: Note nrpages may not be totally updated when
2255 * truncate_inode_pages() returns, as there can be a page in the process
2256 * of deletion (inside __delete_from_page_cache()) in the specified
2257 * range. Thus mapping->nrpages can be non-zero when this function
2258 * returns even after truncation of the whole mapping. Only do this if
2259 * npages isn't already zero.
2261 nrpages = mapping->nrpages;
2263 xa_lock_irqsave(&mapping->i_pages, flags);
2264 nrpages = mapping->nrpages;
2265 xa_unlock_irqrestore(&mapping->i_pages, flags);
2266 } /* Workaround end */
2268 LASSERTF(nrpages == 0, "%s: inode="DFID"(%p) nrpages=%lu, "
2269 "see https://jira.whamcloud.com/browse/LU-118\n",
2270 ll_i2sbi(inode)->ll_fsname,
2271 PFID(ll_inode2fid(inode)), inode, nrpages);
2273 ll_clear_inode(inode);
2279 int ll_iocontrol(struct inode *inode, struct file *file,
2280 unsigned int cmd, unsigned long arg)
2282 struct ll_sb_info *sbi = ll_i2sbi(inode);
2283 struct ptlrpc_request *req = NULL;
2288 case FS_IOC_GETFLAGS: {
2289 struct mdt_body *body;
2290 struct md_op_data *op_data;
2292 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2293 0, 0, LUSTRE_OPC_ANY,
2295 if (IS_ERR(op_data))
2296 RETURN(PTR_ERR(op_data));
2298 op_data->op_valid = OBD_MD_FLFLAGS;
2299 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2300 ll_finish_md_op_data(op_data);
2302 CERROR("%s: failure inode "DFID": rc = %d\n",
2303 sbi->ll_md_exp->exp_obd->obd_name,
2304 PFID(ll_inode2fid(inode)), rc);
2308 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2310 flags = body->mbo_flags;
2312 ptlrpc_req_finished(req);
2314 RETURN(put_user(flags, (int __user *)arg));
2316 case FS_IOC_SETFLAGS: {
2318 struct md_op_data *op_data;
2319 struct cl_object *obj;
2320 struct fsxattr fa = { 0 };
2322 if (get_user(flags, (int __user *)arg))
2325 fa.fsx_projid = ll_i2info(inode)->lli_projid;
2326 if (flags & LUSTRE_PROJINHERIT_FL)
2327 fa.fsx_xflags = FS_XFLAG_PROJINHERIT;
2329 rc = ll_ioctl_check_project(inode, &fa);
2333 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2334 LUSTRE_OPC_ANY, NULL);
2335 if (IS_ERR(op_data))
2336 RETURN(PTR_ERR(op_data));
2338 op_data->op_attr_flags = flags;
2339 op_data->op_xvalid |= OP_XVALID_FLAGS;
2340 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2341 ll_finish_md_op_data(op_data);
2342 ptlrpc_req_finished(req);
2346 ll_update_inode_flags(inode, flags);
2348 obj = ll_i2info(inode)->lli_clob;
2352 OBD_ALLOC_PTR(attr);
2356 rc = cl_setattr_ost(obj, attr, OP_XVALID_FLAGS, flags);
2368 int ll_flush_ctx(struct inode *inode)
2370 struct ll_sb_info *sbi = ll_i2sbi(inode);
2372 CDEBUG(D_SEC, "flush context for user %d\n",
2373 from_kuid(&init_user_ns, current_uid()));
2375 obd_set_info_async(NULL, sbi->ll_md_exp,
2376 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2378 obd_set_info_async(NULL, sbi->ll_dt_exp,
2379 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2384 /* umount -f client means force down, don't save state */
2385 void ll_umount_begin(struct super_block *sb)
2387 struct ll_sb_info *sbi = ll_s2sbi(sb);
2388 struct obd_device *obd;
2389 struct obd_ioctl_data *ioc_data;
2390 struct l_wait_info lwi;
2391 wait_queue_head_t waitq;
2394 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2395 sb->s_count, atomic_read(&sb->s_active));
2397 obd = class_exp2obd(sbi->ll_md_exp);
2399 CERROR("Invalid MDC connection handle %#llx\n",
2400 sbi->ll_md_exp->exp_handle.h_cookie);
2406 obd = class_exp2obd(sbi->ll_dt_exp);
2408 CERROR("Invalid LOV connection handle %#llx\n",
2409 sbi->ll_dt_exp->exp_handle.h_cookie);
2415 OBD_ALLOC_PTR(ioc_data);
2417 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2418 sizeof *ioc_data, ioc_data, NULL);
2420 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2421 sizeof *ioc_data, ioc_data, NULL);
2423 OBD_FREE_PTR(ioc_data);
2426 /* Really, we'd like to wait until there are no requests outstanding,
2427 * and then continue. For now, we just periodically checking for vfs
2428 * to decrement mnt_cnt and hope to finish it within 10sec.
2430 init_waitqueue_head(&waitq);
2431 lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(10),
2432 cfs_time_seconds(1), NULL, NULL);
2433 l_wait_event(waitq, may_umount(sbi->ll_mnt.mnt), &lwi);
2438 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2440 struct ll_sb_info *sbi = ll_s2sbi(sb);
2441 char *profilenm = get_profile_name(sb);
2445 if ((*flags & MS_RDONLY) != (sb->s_flags & SB_RDONLY)) {
2446 read_only = *flags & MS_RDONLY;
2447 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2448 sizeof(KEY_READ_ONLY),
2449 KEY_READ_ONLY, sizeof(read_only),
2452 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2453 profilenm, read_only ?
2454 "read-only" : "read-write", err);
2459 sb->s_flags |= SB_RDONLY;
2461 sb->s_flags &= ~SB_RDONLY;
2463 if (sbi->ll_flags & LL_SBI_VERBOSE)
2464 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2465 read_only ? "read-only" : "read-write");
2471 * Cleanup the open handle that is cached on MDT-side.
2473 * For open case, the client side open handling thread may hit error
2474 * after the MDT grant the open. Under such case, the client should
2475 * send close RPC to the MDT as cleanup; otherwise, the open handle
2476 * on the MDT will be leaked there until the client umount or evicted.
2478 * In further, if someone unlinked the file, because the open handle
2479 * holds the reference on such file/object, then it will block the
2480 * subsequent threads that want to locate such object via FID.
2482 * \param[in] sb super block for this file-system
2483 * \param[in] open_req pointer to the original open request
2485 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2487 struct mdt_body *body;
2488 struct md_op_data *op_data;
2489 struct ptlrpc_request *close_req = NULL;
2490 struct obd_export *exp = ll_s2sbi(sb)->ll_md_exp;
2493 body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2494 OBD_ALLOC_PTR(op_data);
2495 if (op_data == NULL) {
2496 CWARN("%s: cannot allocate op_data to release open handle for "
2497 DFID"\n", ll_s2sbi(sb)->ll_fsname, PFID(&body->mbo_fid1));
2502 op_data->op_fid1 = body->mbo_fid1;
2503 op_data->op_open_handle = body->mbo_open_handle;
2504 op_data->op_mod_time = ktime_get_real_seconds();
2505 md_close(exp, op_data, NULL, &close_req);
2506 ptlrpc_req_finished(close_req);
2507 ll_finish_md_op_data(op_data);
2512 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2513 struct super_block *sb, struct lookup_intent *it)
2515 struct ll_sb_info *sbi = NULL;
2516 struct lustre_md md = { NULL };
2517 bool default_lmv_deleted = false;
2522 LASSERT(*inode || sb);
2523 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2524 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2525 sbi->ll_md_exp, &md);
2530 * clear default_lmv only if intent_getattr reply doesn't contain it.
2531 * but it needs to be done after iget, check this early because
2532 * ll_update_lsm_md() may change md.
2534 if (it && (it->it_op & (IT_LOOKUP | IT_GETATTR)) &&
2535 S_ISDIR(md.body->mbo_mode) && !md.default_lmv)
2536 default_lmv_deleted = true;
2539 rc = ll_update_inode(*inode, &md);
2543 LASSERT(sb != NULL);
2546 * At this point server returns to client's same fid as client
2547 * generated for creating. So using ->fid1 is okay here.
2549 if (!fid_is_sane(&md.body->mbo_fid1)) {
2550 CERROR("%s: Fid is insane "DFID"\n",
2552 PFID(&md.body->mbo_fid1));
2553 GOTO(out, rc = -EINVAL);
2556 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2557 sbi->ll_flags & LL_SBI_32BIT_API),
2559 if (IS_ERR(*inode)) {
2560 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
2562 posix_acl_release(md.posix_acl);
2563 md.posix_acl = NULL;
2566 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2568 CERROR("new_inode -fatal: rc %d\n", rc);
2573 /* Handling piggyback layout lock.
2574 * Layout lock can be piggybacked by getattr and open request.
2575 * The lsm can be applied to inode only if it comes with a layout lock
2576 * otherwise correct layout may be overwritten, for example:
2577 * 1. proc1: mdt returns a lsm but not granting layout
2578 * 2. layout was changed by another client
2579 * 3. proc2: refresh layout and layout lock granted
2580 * 4. proc1: to apply a stale layout */
2581 if (it != NULL && it->it_lock_mode != 0) {
2582 struct lustre_handle lockh;
2583 struct ldlm_lock *lock;
2585 lockh.cookie = it->it_lock_handle;
2586 lock = ldlm_handle2lock(&lockh);
2587 LASSERT(lock != NULL);
2588 if (ldlm_has_layout(lock)) {
2589 struct cl_object_conf conf;
2591 memset(&conf, 0, sizeof(conf));
2592 conf.coc_opc = OBJECT_CONF_SET;
2593 conf.coc_inode = *inode;
2594 conf.coc_lock = lock;
2595 conf.u.coc_layout = md.layout;
2596 (void)ll_layout_conf(*inode, &conf);
2598 LDLM_LOCK_PUT(lock);
2601 if (default_lmv_deleted)
2602 ll_update_default_lsm_md(*inode, &md);
2607 /* cleanup will be done if necessary */
2608 md_free_lustre_md(sbi->ll_md_exp, &md);
2610 if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2611 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2616 int ll_obd_statfs(struct inode *inode, void __user *arg)
2618 struct ll_sb_info *sbi = NULL;
2619 struct obd_export *exp;
2621 struct obd_ioctl_data *data = NULL;
2625 if (!inode || !(sbi = ll_i2sbi(inode)))
2626 GOTO(out_statfs, rc = -EINVAL);
2628 rc = obd_ioctl_getdata(&buf, &len, arg);
2630 GOTO(out_statfs, rc);
2633 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2634 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2635 GOTO(out_statfs, rc = -EINVAL);
2637 if (data->ioc_inllen1 != sizeof(__u32) ||
2638 data->ioc_inllen2 != sizeof(__u32) ||
2639 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2640 data->ioc_plen2 != sizeof(struct obd_uuid))
2641 GOTO(out_statfs, rc = -EINVAL);
2643 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2644 if (type & LL_STATFS_LMV)
2645 exp = sbi->ll_md_exp;
2646 else if (type & LL_STATFS_LOV)
2647 exp = sbi->ll_dt_exp;
2649 GOTO(out_statfs, rc = -ENODEV);
2651 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2653 GOTO(out_statfs, rc);
2655 OBD_FREE_LARGE(buf, len);
2660 * this is normally called in ll_fini_md_op_data(), but sometimes it needs to
2661 * be called early to avoid deadlock.
2663 void ll_unlock_md_op_lsm(struct md_op_data *op_data)
2665 if (op_data->op_mea2_sem) {
2666 up_read(op_data->op_mea2_sem);
2667 op_data->op_mea2_sem = NULL;
2670 if (op_data->op_mea1_sem) {
2671 up_read(op_data->op_mea1_sem);
2672 op_data->op_mea1_sem = NULL;
2676 /* this function prepares md_op_data hint for passing it down to MD stack. */
2677 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2678 struct inode *i1, struct inode *i2,
2679 const char *name, size_t namelen,
2680 __u32 mode, enum md_op_code opc,
2683 LASSERT(i1 != NULL);
2686 /* Do not reuse namelen for something else. */
2688 return ERR_PTR(-EINVAL);
2690 if (namelen > ll_i2sbi(i1)->ll_namelen)
2691 return ERR_PTR(-ENAMETOOLONG);
2693 if (!lu_name_is_valid_2(name, namelen))
2694 return ERR_PTR(-EINVAL);
2697 if (op_data == NULL)
2698 OBD_ALLOC_PTR(op_data);
2700 if (op_data == NULL)
2701 return ERR_PTR(-ENOMEM);
2703 ll_i2gids(op_data->op_suppgids, i1, i2);
2704 op_data->op_fid1 = *ll_inode2fid(i1);
2705 op_data->op_code = opc;
2707 if (S_ISDIR(i1->i_mode)) {
2708 down_read(&ll_i2info(i1)->lli_lsm_sem);
2709 op_data->op_mea1_sem = &ll_i2info(i1)->lli_lsm_sem;
2710 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2711 op_data->op_default_mea1 = ll_i2info(i1)->lli_default_lsm_md;
2715 op_data->op_fid2 = *ll_inode2fid(i2);
2716 if (S_ISDIR(i2->i_mode)) {
2718 down_read(&ll_i2info(i2)->lli_lsm_sem);
2719 op_data->op_mea2_sem =
2720 &ll_i2info(i2)->lli_lsm_sem;
2722 op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2725 fid_zero(&op_data->op_fid2);
2728 if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2729 op_data->op_cli_flags |= CLI_HASH64;
2731 if (ll_need_32bit_api(ll_i2sbi(i1)))
2732 op_data->op_cli_flags |= CLI_API32;
2734 op_data->op_name = name;
2735 op_data->op_namelen = namelen;
2736 op_data->op_mode = mode;
2737 op_data->op_mod_time = ktime_get_real_seconds();
2738 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2739 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2740 op_data->op_cap = cfs_curproc_cap_pack();
2741 op_data->op_mds = 0;
2742 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2743 filename_is_volatile(name, namelen, &op_data->op_mds)) {
2744 op_data->op_bias |= MDS_CREATE_VOLATILE;
2746 op_data->op_data = data;
2751 void ll_finish_md_op_data(struct md_op_data *op_data)
2753 ll_unlock_md_op_lsm(op_data);
2754 security_release_secctx(op_data->op_file_secctx,
2755 op_data->op_file_secctx_size);
2756 OBD_FREE_PTR(op_data);
2759 #ifdef HAVE_SUPEROPS_USE_DENTRY
2760 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2762 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2765 struct ll_sb_info *sbi;
2767 #ifdef HAVE_SUPEROPS_USE_DENTRY
2768 LASSERT((seq != NULL) && (dentry != NULL));
2769 sbi = ll_s2sbi(dentry->d_sb);
2771 LASSERT((seq != NULL) && (vfs != NULL));
2772 sbi = ll_s2sbi(vfs->mnt_sb);
2775 if (sbi->ll_flags & LL_SBI_NOLCK)
2776 seq_puts(seq, ",nolock");
2778 if (sbi->ll_flags & LL_SBI_FLOCK)
2779 seq_puts(seq, ",flock");
2781 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2782 seq_puts(seq, ",localflock");
2784 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2785 seq_puts(seq, ",user_xattr");
2787 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2788 seq_puts(seq, ",lazystatfs");
2790 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2791 seq_puts(seq, ",user_fid2path");
2793 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2794 seq_puts(seq, ",always_ping");
2800 * Get obd name by cmd, and copy out to user space
2802 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2804 struct ll_sb_info *sbi = ll_i2sbi(inode);
2805 struct obd_device *obd;
2808 if (cmd == OBD_IOC_GETDTNAME)
2809 obd = class_exp2obd(sbi->ll_dt_exp);
2810 else if (cmd == OBD_IOC_GETMDNAME)
2811 obd = class_exp2obd(sbi->ll_md_exp);
2818 if (copy_to_user((void __user *)arg, obd->obd_name,
2819 strlen(obd->obd_name) + 1))
2825 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2832 p.mnt = current->fs->root.mnt;
2834 path = d_path(&p, buf, bufsize);
2839 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2841 char *buf, *path = NULL;
2842 struct dentry *dentry = NULL;
2843 struct inode *inode = page->mapping->host;
2845 /* this can be called inside spin lock so use GFP_ATOMIC. */
2846 buf = (char *)__get_free_page(GFP_ATOMIC);
2848 dentry = d_find_alias(page->mapping->host);
2850 path = ll_d_path(dentry, buf, PAGE_SIZE);
2853 /* The below message is checked in recovery-small.sh test_24b */
2855 "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2856 "(rc %d)\n", ll_i2sbi(inode)->ll_fsname,
2857 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2858 PFID(ll_inode2fid(inode)),
2859 (path && !IS_ERR(path)) ? path : "", ioret);
2865 free_page((unsigned long)buf);
2868 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2869 struct lov_user_md **kbuf)
2871 struct lov_user_md lum;
2875 if (copy_from_user(&lum, md, sizeof(lum)))
2878 lum_size = ll_lov_user_md_size(&lum);
2882 OBD_ALLOC_LARGE(*kbuf, lum_size);
2886 if (copy_from_user(*kbuf, md, lum_size) != 0) {
2887 OBD_FREE_LARGE(*kbuf, lum_size);
2895 * Compute llite root squash state after a change of root squash
2896 * configuration setting or add/remove of a lnet nid
2898 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2900 struct root_squash_info *squash = &sbi->ll_squash;
2903 struct lnet_process_id id;
2905 /* Update norootsquash flag */
2906 spin_lock(&squash->rsi_lock);
2907 if (list_empty(&squash->rsi_nosquash_nids))
2908 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2910 /* Do not apply root squash as soon as one of our NIDs is
2911 * in the nosquash_nids list */
2914 while (LNetGetId(i++, &id) != -ENOENT) {
2915 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2917 if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2923 sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2925 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2927 spin_unlock(&squash->rsi_lock);
2931 * Parse linkea content to extract information about a given hardlink
2933 * \param[in] ldata - Initialized linkea data
2934 * \param[in] linkno - Link identifier
2935 * \param[out] parent_fid - The entry's parent FID
2936 * \param[out] ln - Entry name destination buffer
2938 * \retval 0 on success
2939 * \retval Appropriate negative error code on failure
2941 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2942 struct lu_fid *parent_fid, struct lu_name *ln)
2948 rc = linkea_init_with_rec(ldata);
2952 if (linkno >= ldata->ld_leh->leh_reccount)
2953 /* beyond last link */
2956 linkea_first_entry(ldata);
2957 for (idx = 0; ldata->ld_lee != NULL; idx++) {
2958 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2963 linkea_next_entry(ldata);
2973 * Get parent FID and name of an identified link. Operation is performed for
2974 * a given link number, letting the caller iterate over linkno to list one or
2975 * all links of an entry.
2977 * \param[in] file - File descriptor against which to perform the operation
2978 * \param[in,out] arg - User-filled structure containing the linkno to operate
2979 * on and the available size. It is eventually filled with
2980 * the requested information or left untouched on error
2982 * \retval - 0 on success
2983 * \retval - Appropriate negative error code on failure
2985 int ll_getparent(struct file *file, struct getparent __user *arg)
2987 struct inode *inode = file_inode(file);
2988 struct linkea_data *ldata;
2989 struct lu_buf buf = LU_BUF_NULL;
2991 struct lu_fid parent_fid;
2998 if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2999 !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
3002 if (get_user(name_size, &arg->gp_name_size))
3005 if (get_user(linkno, &arg->gp_linkno))
3008 if (name_size > PATH_MAX)
3011 OBD_ALLOC(ldata, sizeof(*ldata));
3015 rc = linkea_data_new(ldata, &buf);
3017 GOTO(ldata_free, rc);
3019 rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
3020 buf.lb_len, OBD_MD_FLXATTR);
3024 rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
3028 if (ln.ln_namelen >= name_size)
3029 GOTO(lb_free, rc = -EOVERFLOW);
3031 if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
3032 GOTO(lb_free, rc = -EFAULT);
3034 if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
3035 GOTO(lb_free, rc = -EFAULT);
3037 if (put_user('\0', arg->gp_name + ln.ln_namelen))
3038 GOTO(lb_free, rc = -EFAULT);
3043 OBD_FREE(ldata, sizeof(*ldata));