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 #include <linux/delay.h>
49 #ifdef HAVE_UIDGID_HEADER
50 # include <linux/uidgid.h>
52 #include <linux/security.h>
54 #include <uapi/linux/lustre/lustre_ioctl.h>
55 #ifdef HAVE_UAPI_LINUX_MOUNT_H
56 #include <uapi/linux/mount.h>
59 #include <lustre_ha.h>
60 #include <lustre_dlm.h>
61 #include <lprocfs_status.h>
62 #include <lustre_disk.h>
63 #include <uapi/linux/lustre/lustre_param.h>
64 #include <lustre_log.h>
65 #include <cl_object.h>
66 #include <obd_cksum.h>
67 #include "llite_internal.h"
69 struct kmem_cache *ll_file_data_slab;
72 #define log2(n) ffz(~(n))
75 static struct ll_sb_info *ll_init_sbi(void)
77 struct ll_sb_info *sbi = NULL;
79 unsigned long lru_page_max;
88 RETURN(ERR_PTR(-ENOMEM));
90 rc = pcc_super_init(&sbi->ll_pcc_super);
94 spin_lock_init(&sbi->ll_lock);
95 mutex_init(&sbi->ll_lco.lco_lock);
96 spin_lock_init(&sbi->ll_pp_extent_lock);
97 spin_lock_init(&sbi->ll_process_lock);
98 sbi->ll_rw_stats_on = 0;
99 sbi->ll_statfs_max_age = OBD_STATFS_CACHE_SECONDS;
102 pages = si.totalram - si.totalhigh;
103 lru_page_max = pages / 2;
105 sbi->ll_ra_info.ra_async_max_active = 0;
106 sbi->ll_ra_info.ll_readahead_wq =
107 alloc_workqueue("ll-readahead-wq", WQ_UNBOUND,
108 sbi->ll_ra_info.ra_async_max_active);
109 if (!sbi->ll_ra_info.ll_readahead_wq)
110 GOTO(out_pcc, rc = -ENOMEM);
112 /* initialize ll_cache data */
113 sbi->ll_cache = cl_cache_init(lru_page_max);
114 if (sbi->ll_cache == NULL)
115 GOTO(out_destroy_ra, rc = -ENOMEM);
117 sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
118 SBI_DEFAULT_READAHEAD_MAX);
119 sbi->ll_ra_info.ra_async_pages_per_file_threshold =
120 sbi->ll_ra_info.ra_max_pages_per_file;
121 sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
122 sbi->ll_ra_info.ra_max_read_ahead_whole_pages = -1;
124 sbi->ll_flags |= LL_SBI_VERBOSE;
125 #ifdef ENABLE_CHECKSUM
126 sbi->ll_flags |= LL_SBI_CHECKSUM;
129 sbi->ll_flags |= LL_SBI_FLOCK;
132 #ifdef HAVE_LRU_RESIZE_SUPPORT
133 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
135 sbi->ll_flags |= LL_SBI_LAZYSTATFS;
137 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
138 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
140 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
144 /* metadata statahead is enabled by default */
145 sbi->ll_sa_running_max = LL_SA_RUNNING_DEF;
146 sbi->ll_sa_max = LL_SA_RPC_DEF;
147 atomic_set(&sbi->ll_sa_total, 0);
148 atomic_set(&sbi->ll_sa_wrong, 0);
149 atomic_set(&sbi->ll_sa_running, 0);
150 atomic_set(&sbi->ll_agl_total, 0);
151 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
152 sbi->ll_flags |= LL_SBI_FAST_READ;
153 sbi->ll_flags |= LL_SBI_TINY_WRITE;
156 sbi->ll_squash.rsi_uid = 0;
157 sbi->ll_squash.rsi_gid = 0;
158 INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
159 spin_lock_init(&sbi->ll_squash.rsi_lock);
161 /* Per-filesystem file heat */
162 sbi->ll_heat_decay_weight = SBI_DEFAULT_HEAT_DECAY_WEIGHT;
163 sbi->ll_heat_period_second = SBI_DEFAULT_HEAT_PERIOD_SECOND;
166 destroy_workqueue(sbi->ll_ra_info.ll_readahead_wq);
168 pcc_super_fini(&sbi->ll_pcc_super);
174 static void ll_free_sbi(struct super_block *sb)
176 struct ll_sb_info *sbi = ll_s2sbi(sb);
180 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
181 cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
182 if (sbi->ll_ra_info.ll_readahead_wq)
183 destroy_workqueue(sbi->ll_ra_info.ll_readahead_wq);
184 if (sbi->ll_cache != NULL) {
185 cl_cache_decref(sbi->ll_cache);
186 sbi->ll_cache = NULL;
188 pcc_super_fini(&sbi->ll_pcc_super);
189 OBD_FREE(sbi, sizeof(*sbi));
194 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
195 struct vfsmount *mnt)
197 struct inode *root = NULL;
198 struct ll_sb_info *sbi = ll_s2sbi(sb);
199 struct obd_statfs *osfs = NULL;
200 struct ptlrpc_request *request = NULL;
201 struct obd_connect_data *data = NULL;
202 struct obd_uuid *uuid;
203 struct md_op_data *op_data;
204 struct lustre_md lmd;
206 int size, err, checksum;
209 sbi->ll_md_obd = class_name2obd(md);
210 if (!sbi->ll_md_obd) {
211 CERROR("MD %s: not setup or attached\n", md);
225 /* pass client page size via ocd_grant_blkbits, the server should report
226 * back its backend blocksize for grant calculation purpose */
227 data->ocd_grant_blkbits = PAGE_SHIFT;
229 /* indicate MDT features supported by this client */
230 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
231 OBD_CONNECT_ATTRFID | OBD_CONNECT_GRANT |
232 OBD_CONNECT_VERSION | OBD_CONNECT_BRW_SIZE |
233 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
234 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
235 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
236 OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 |
237 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
238 OBD_CONNECT_64BITHASH |
239 OBD_CONNECT_EINPROGRESS |
240 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
241 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS|
242 OBD_CONNECT_MAX_EASIZE |
243 OBD_CONNECT_FLOCK_DEAD |
244 OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
245 OBD_CONNECT_OPEN_BY_FID |
246 OBD_CONNECT_DIR_STRIPE |
247 OBD_CONNECT_BULK_MBITS | OBD_CONNECT_CKSUM |
248 OBD_CONNECT_SUBTREE |
249 OBD_CONNECT_MULTIMODRPCS |
250 OBD_CONNECT_GRANT_PARAM |
251 OBD_CONNECT_SHORTIO | OBD_CONNECT_FLAGS2;
253 data->ocd_connect_flags2 = OBD_CONNECT2_DIR_MIGRATE |
254 OBD_CONNECT2_SUM_STATFS |
255 OBD_CONNECT2_OVERSTRIPING |
257 OBD_CONNECT2_LOCK_CONVERT |
258 OBD_CONNECT2_ARCHIVE_ID_ARRAY |
259 OBD_CONNECT2_INC_XID |
261 OBD_CONNECT2_ASYNC_DISCARD |
264 #ifdef HAVE_LRU_RESIZE_SUPPORT
265 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
266 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
268 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
269 data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK |
270 OBD_CONNECT_LARGE_ACL;
273 data->ocd_cksum_types = obd_cksum_types_supported_client();
275 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
276 /* flag mdc connection as lightweight, only used for test
277 * purpose, use with care */
278 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
280 data->ocd_ibits_known = MDS_INODELOCK_FULL;
281 data->ocd_version = LUSTRE_VERSION_CODE;
283 if (sb->s_flags & SB_RDONLY)
284 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
285 if (sbi->ll_flags & LL_SBI_USER_XATTR)
286 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
289 /* Setting this indicates we correctly support S_NOSEC (See kernel
290 * commit 9e1f1de02c2275d7172e18dc4e7c2065777611bf)
292 sb->s_flags |= SB_NOSEC;
295 if (sbi->ll_flags & LL_SBI_FLOCK)
296 sbi->ll_fop = &ll_file_operations_flock;
297 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
298 sbi->ll_fop = &ll_file_operations;
300 sbi->ll_fop = &ll_file_operations_noflock;
302 /* always ping even if server suppress_pings */
303 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
304 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
306 obd_connect_set_secctx(data);
308 #if defined(CONFIG_SECURITY)
309 data->ocd_connect_flags2 |= OBD_CONNECT2_SELINUX_POLICY;
312 data->ocd_brw_size = MD_MAX_BRW_SIZE;
314 err = obd_connect(NULL, &sbi->ll_md_exp, sbi->ll_md_obd,
315 &sbi->ll_sb_uuid, data, sbi->ll_cache);
317 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
318 "recovery, of which this client is not a "
319 "part. Please wait for recovery to complete,"
320 " abort, or time out.\n", md);
323 CERROR("cannot connect to %s: rc = %d\n", md, err);
327 sbi->ll_md_exp->exp_connect_data = *data;
329 err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
330 LUSTRE_SEQ_METADATA);
332 CERROR("%s: Can't init metadata layer FID infrastructure, "
333 "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
337 /* For mount, we only need fs info from MDT0, and also in DNE, it
338 * can make sure the client can be mounted as long as MDT0 is
340 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
341 ktime_get_seconds() - sbi->ll_statfs_max_age,
342 OBD_STATFS_FOR_MDT0);
344 GOTO(out_md_fid, err);
346 /* This needs to be after statfs to ensure connect has finished.
347 * Note that "data" does NOT contain the valid connect reply.
348 * If connecting to a 1.8 server there will be no LMV device, so
349 * we can access the MDC export directly and exp_connect_flags will
350 * be non-zero, but if accessing an upgraded 2.1 server it will
351 * have the correct flags filled in.
352 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
353 valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
354 if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
355 valid != CLIENT_CONNECT_MDT_REQD) {
358 OBD_ALLOC_WAIT(buf, PAGE_SIZE);
359 obd_connect_flags2str(buf, PAGE_SIZE,
360 valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
361 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
362 "feature(s) needed for correct operation "
363 "of this client (%s). Please upgrade "
364 "server or downgrade client.\n",
365 sbi->ll_md_exp->exp_obd->obd_name, buf);
366 OBD_FREE(buf, PAGE_SIZE);
367 GOTO(out_md_fid, err = -EPROTO);
370 size = sizeof(*data);
371 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
372 KEY_CONN_DATA, &size, data);
374 CERROR("%s: Get connect data failed: rc = %d\n",
375 sbi->ll_md_exp->exp_obd->obd_name, err);
376 GOTO(out_md_fid, err);
379 LASSERT(osfs->os_bsize);
380 sb->s_blocksize = osfs->os_bsize;
381 sb->s_blocksize_bits = log2(osfs->os_bsize);
382 sb->s_magic = LL_SUPER_MAGIC;
383 sb->s_maxbytes = MAX_LFS_FILESIZE;
384 sbi->ll_namelen = osfs->os_namelen;
385 sbi->ll_mnt.mnt = current->fs->root.mnt;
387 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
388 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
389 LCONSOLE_INFO("Disabling user_xattr feature because "
390 "it is not supported on the server\n");
391 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
394 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
396 sb->s_flags |= SB_POSIXACL;
398 sbi->ll_flags |= LL_SBI_ACL;
400 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
402 sb->s_flags &= ~SB_POSIXACL;
404 sbi->ll_flags &= ~LL_SBI_ACL;
407 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
408 sbi->ll_flags |= LL_SBI_64BIT_HASH;
410 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
411 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
413 if (obd_connect_has_secctx(data))
414 sbi->ll_flags |= LL_SBI_FILE_SECCTX;
416 if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
417 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
418 LCONSOLE_INFO("%s: disabling xattr cache due to "
419 "unknown maximum xattr size.\n", dt);
420 } else if (!sbi->ll_xattr_cache_set) {
421 /* If xattr_cache is already set (no matter 0 or 1)
422 * during processing llog, it won't be enabled here. */
423 sbi->ll_flags |= LL_SBI_XATTR_CACHE;
424 sbi->ll_xattr_cache_enabled = 1;
428 sbi->ll_dt_obd = class_name2obd(dt);
429 if (!sbi->ll_dt_obd) {
430 CERROR("DT %s: not setup or attached\n", dt);
431 GOTO(out_md_fid, err = -ENODEV);
434 /* pass client page size via ocd_grant_blkbits, the server should report
435 * back its backend blocksize for grant calculation purpose */
436 data->ocd_grant_blkbits = PAGE_SHIFT;
438 /* indicate OST features supported by this client */
439 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
440 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
441 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
442 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
443 OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
444 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
445 OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
446 OBD_CONNECT_EINPROGRESS |
447 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
448 OBD_CONNECT_LAYOUTLOCK |
449 OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
450 OBD_CONNECT_BULK_MBITS | OBD_CONNECT_SHORTIO |
451 OBD_CONNECT_FLAGS2 | OBD_CONNECT_GRANT_SHRINK;
453 /* The client currently advertises support for OBD_CONNECT_LOCKAHEAD_OLD so it
454 * can interoperate with an older version of lockahead which was released prior
455 * to landing in master. This support will be dropped when 2.13 development
456 * starts. At the point, we should not just drop the connect flag (below), we
457 * should also remove the support in the code.
459 * Removing it means a few things:
460 * 1. Remove this section here
461 * 2. Remove CEF_NONBLOCK in ll_file_lockahead()
462 * 3. Remove function exp_connect_lockahead_old
463 * 4. Remove LDLM_FL_LOCKAHEAD_OLD_RESERVED in lustre_dlm_flags.h
465 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 12, 50, 0)
466 data->ocd_connect_flags |= OBD_CONNECT_LOCKAHEAD_OLD;
469 data->ocd_connect_flags2 = OBD_CONNECT2_LOCKAHEAD |
470 OBD_CONNECT2_INC_XID;
472 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
473 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
475 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
476 * disabled by default, because it can still be enabled on the
477 * fly via /sys. As a consequence, we still need to come to an
478 * agreement on the supported algorithms at connect time
480 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
482 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
483 data->ocd_cksum_types = OBD_CKSUM_ADLER;
485 data->ocd_cksum_types = obd_cksum_types_supported_client();
487 #ifdef HAVE_LRU_RESIZE_SUPPORT
488 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
490 /* always ping even if server suppress_pings */
491 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
492 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
494 CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
495 "ocd_grant: %d\n", data->ocd_connect_flags,
496 data->ocd_version, data->ocd_grant);
498 sbi->ll_dt_obd->obd_upcall.onu_owner = &sbi->ll_lco;
499 sbi->ll_dt_obd->obd_upcall.onu_upcall = cl_ocd_update;
501 data->ocd_brw_size = DT_MAX_BRW_SIZE;
503 err = obd_connect(NULL, &sbi->ll_dt_exp, sbi->ll_dt_obd,
504 &sbi->ll_sb_uuid, data, sbi->ll_cache);
506 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
507 "recovery, of which this client is not a "
508 "part. Please wait for recovery to "
509 "complete, abort, or time out.\n", dt);
512 CERROR("%s: Cannot connect to %s: rc = %d\n",
513 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
517 sbi->ll_dt_exp->exp_connect_data = *data;
519 /* Don't change value if it was specified in the config log */
520 if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages == -1) {
521 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
522 max_t(unsigned long, SBI_DEFAULT_READAHEAD_WHOLE_MAX,
523 (data->ocd_brw_size >> PAGE_SHIFT));
524 if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages >
525 sbi->ll_ra_info.ra_max_pages_per_file)
526 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
527 sbi->ll_ra_info.ra_max_pages_per_file;
530 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
531 LUSTRE_SEQ_METADATA);
533 CERROR("%s: Can't init data layer FID infrastructure, "
534 "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
538 mutex_lock(&sbi->ll_lco.lco_lock);
539 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
540 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
541 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
542 mutex_unlock(&sbi->ll_lco.lco_lock);
544 fid_zero(&sbi->ll_root_fid);
545 err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
548 CERROR("cannot mds_connect: rc = %d\n", err);
549 GOTO(out_lock_cn_cb, err);
551 if (!fid_is_sane(&sbi->ll_root_fid)) {
552 CERROR("%s: Invalid root fid "DFID" during mount\n",
553 sbi->ll_md_exp->exp_obd->obd_name,
554 PFID(&sbi->ll_root_fid));
555 GOTO(out_lock_cn_cb, err = -EINVAL);
557 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
559 sb->s_op = &lustre_super_operations;
560 sb->s_xattr = ll_xattr_handlers;
561 #if THREAD_SIZE >= 8192 /*b=17630*/
562 sb->s_export_op = &lustre_export_operations;
566 * XXX: move this to after cbd setup? */
567 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
568 if (sbi->ll_flags & LL_SBI_ACL)
569 valid |= OBD_MD_FLACL;
571 OBD_ALLOC_PTR(op_data);
573 GOTO(out_lock_cn_cb, err = -ENOMEM);
575 op_data->op_fid1 = sbi->ll_root_fid;
576 op_data->op_mode = 0;
577 op_data->op_valid = valid;
579 err = md_getattr(sbi->ll_md_exp, op_data, &request);
581 OBD_FREE_PTR(op_data);
583 CERROR("%s: md_getattr failed for root: rc = %d\n",
584 sbi->ll_md_exp->exp_obd->obd_name, err);
585 GOTO(out_lock_cn_cb, err);
588 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
589 sbi->ll_md_exp, &lmd);
591 CERROR("failed to understand root inode md: rc = %d\n", err);
592 ptlrpc_req_finished(request);
593 GOTO(out_lock_cn_cb, err);
596 LASSERT(fid_is_sane(&sbi->ll_root_fid));
597 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
598 sbi->ll_flags & LL_SBI_32BIT_API),
600 md_free_lustre_md(sbi->ll_md_exp, &lmd);
601 ptlrpc_req_finished(request);
604 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
606 posix_acl_release(lmd.posix_acl);
607 lmd.posix_acl = NULL;
610 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
612 CERROR("%s: bad ll_iget() for root: rc = %d\n",
613 sbi->ll_fsname, err);
617 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
618 if (sbi->ll_checksum_set) {
619 err = obd_set_info_async(NULL, sbi->ll_dt_exp,
620 sizeof(KEY_CHECKSUM), KEY_CHECKSUM,
621 sizeof(checksum), &checksum, NULL);
623 CERROR("%s: Set checksum failed: rc = %d\n",
624 sbi->ll_dt_exp->exp_obd->obd_name, err);
630 sb->s_root = d_make_root(root);
631 if (sb->s_root == NULL) {
633 CERROR("%s: can't make root dentry: rc = %d\n",
634 sbi->ll_fsname, err);
638 sbi->ll_sdev_orig = sb->s_dev;
640 /* We set sb->s_dev equal on all lustre clients in order to support
641 * NFS export clustering. NFSD requires that the FSID be the same
643 /* s_dev is also used in lt_compare() to compare two fs, but that is
644 * only a node-local comparison. */
645 uuid = obd_get_uuid(sbi->ll_md_exp);
647 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
654 if (sbi->ll_dt_obd) {
655 err = sysfs_create_link(&sbi->ll_kset.kobj,
656 &sbi->ll_dt_obd->obd_kset.kobj,
657 sbi->ll_dt_obd->obd_type->typ_name);
659 CERROR("%s: could not register %s in llite: rc = %d\n",
660 dt, sbi->ll_fsname, err);
665 if (sbi->ll_md_obd) {
666 err = sysfs_create_link(&sbi->ll_kset.kobj,
667 &sbi->ll_md_obd->obd_kset.kobj,
668 sbi->ll_md_obd->obd_type->typ_name);
670 CERROR("%s: could not register %s in llite: rc = %d\n",
671 md, sbi->ll_fsname, err);
681 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
683 obd_disconnect(sbi->ll_dt_exp);
684 sbi->ll_dt_exp = NULL;
685 sbi->ll_dt_obd = NULL;
687 obd_fid_fini(sbi->ll_md_exp->exp_obd);
689 obd_disconnect(sbi->ll_md_exp);
690 sbi->ll_md_exp = NULL;
691 sbi->ll_md_obd = NULL;
700 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
704 size = sizeof(*lmmsize);
705 rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
706 KEY_MAX_EASIZE, &size, lmmsize);
708 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
709 sbi->ll_dt_exp->exp_obd->obd_name, rc);
713 CDEBUG(D_INFO, "max LOV ea size: %d\n", *lmmsize);
716 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
717 KEY_MAX_EASIZE, &size, lmmsize);
719 CERROR("Get max mdsize error rc %d\n", rc);
721 CDEBUG(D_INFO, "max LMV ea size: %d\n", *lmmsize);
727 * Get the value of the default_easize parameter.
729 * \see client_obd::cl_default_mds_easize
731 * \param[in] sbi superblock info for this filesystem
732 * \param[out] lmmsize pointer to storage location for value
734 * \retval 0 on success
735 * \retval negative negated errno on failure
737 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
742 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
743 KEY_DEFAULT_EASIZE, &size, lmmsize);
745 CERROR("Get default mdsize error rc %d\n", rc);
751 * Set the default_easize parameter to the given value.
753 * \see client_obd::cl_default_mds_easize
755 * \param[in] sbi superblock info for this filesystem
756 * \param[in] lmmsize the size to set
758 * \retval 0 on success
759 * \retval negative negated errno on failure
761 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
765 if (lmmsize < sizeof(struct lov_mds_md) ||
766 lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
769 rc = obd_set_info_async(NULL, sbi->ll_md_exp,
770 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
771 sizeof(int), &lmmsize, NULL);
776 static void client_common_put_super(struct super_block *sb)
778 struct ll_sb_info *sbi = ll_s2sbi(sb);
783 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
784 obd_disconnect(sbi->ll_dt_exp);
785 sbi->ll_dt_exp = NULL;
787 ll_debugfs_unregister_super(sb);
789 obd_fid_fini(sbi->ll_md_exp->exp_obd);
790 obd_disconnect(sbi->ll_md_exp);
791 sbi->ll_md_exp = NULL;
796 void ll_kill_super(struct super_block *sb)
798 struct ll_sb_info *sbi;
802 if (!(sb->s_flags & SB_ACTIVE))
806 /* we need restore s_dev from changed for clustred NFS before put_super
807 * because new kernels have cached s_dev and change sb->s_dev in
808 * put_super not affected real removing devices */
810 sb->s_dev = sbi->ll_sdev_orig;
812 /* wait running statahead threads to quit */
813 while (atomic_read(&sbi->ll_sa_running) > 0) {
814 set_current_state(TASK_UNINTERRUPTIBLE);
815 schedule_timeout(cfs_time_seconds(1) >> 3);
822 static inline int ll_set_opt(const char *opt, char *data, int fl)
824 if (strncmp(opt, data, strlen(opt)) != 0)
830 /* non-client-specific mount options are parsed in lmd_parse */
831 static int ll_options(char *options, struct ll_sb_info *sbi)
834 char *s1 = options, *s2;
835 int *flags = &sbi->ll_flags;
841 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
844 CDEBUG(D_SUPER, "next opt=%s\n", s1);
845 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
850 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
852 *flags = (*flags & ~LL_SBI_LOCALFLOCK) | tmp;
855 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
857 *flags = (*flags & ~LL_SBI_FLOCK) | tmp;
860 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
865 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
870 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
875 tmp = ll_set_opt("context", s1, 1);
878 tmp = ll_set_opt("fscontext", s1, 1);
881 tmp = ll_set_opt("defcontext", s1, 1);
884 tmp = ll_set_opt("rootcontext", s1, 1);
887 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
892 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
898 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
901 sbi->ll_checksum_set = 1;
904 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
907 sbi->ll_checksum_set = 1;
910 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
915 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
920 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
925 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
930 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
935 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
940 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
945 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
950 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
956 s2 = strchr(s1, ',');
964 void ll_lli_init(struct ll_inode_info *lli)
966 lli->lli_inode_magic = LLI_INODE_MAGIC;
968 spin_lock_init(&lli->lli_lock);
969 lli->lli_posix_acl = NULL;
970 /* Do not set lli_fid, it has been initialized already. */
971 fid_zero(&lli->lli_pfid);
972 lli->lli_mds_read_och = NULL;
973 lli->lli_mds_write_och = NULL;
974 lli->lli_mds_exec_och = NULL;
975 lli->lli_open_fd_read_count = 0;
976 lli->lli_open_fd_write_count = 0;
977 lli->lli_open_fd_exec_count = 0;
978 mutex_init(&lli->lli_och_mutex);
979 spin_lock_init(&lli->lli_agl_lock);
980 spin_lock_init(&lli->lli_layout_lock);
981 ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
982 lli->lli_clob = NULL;
984 init_rwsem(&lli->lli_xattrs_list_rwsem);
985 mutex_init(&lli->lli_xattrs_enq_lock);
987 LASSERT(lli->lli_vfs_inode.i_mode != 0);
988 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
989 lli->lli_opendir_key = NULL;
991 spin_lock_init(&lli->lli_sa_lock);
992 lli->lli_opendir_pid = 0;
993 lli->lli_sa_enabled = 0;
994 init_rwsem(&lli->lli_lsm_sem);
996 mutex_init(&lli->lli_size_mutex);
997 lli->lli_symlink_name = NULL;
998 init_rwsem(&lli->lli_trunc_sem);
999 range_lock_tree_init(&lli->lli_write_tree);
1000 init_rwsem(&lli->lli_glimpse_sem);
1001 lli->lli_glimpse_time = ktime_set(0, 0);
1002 INIT_LIST_HEAD(&lli->lli_agl_list);
1003 lli->lli_agl_index = 0;
1004 lli->lli_async_rc = 0;
1005 spin_lock_init(&lli->lli_heat_lock);
1006 obd_heat_clear(lli->lli_heat_instances, OBD_HEAT_COUNT);
1007 lli->lli_heat_flags = 0;
1008 mutex_init(&lli->lli_pcc_lock);
1009 lli->lli_pcc_state = PCC_STATE_FL_NONE;
1010 lli->lli_pcc_inode = NULL;
1011 lli->lli_pcc_dsflags = PCC_DATASET_INVALID;
1012 lli->lli_pcc_generation = 0;
1013 mutex_init(&lli->lli_group_mutex);
1014 lli->lli_group_users = 0;
1015 lli->lli_group_gid = 0;
1017 mutex_init(&lli->lli_layout_mutex);
1018 memset(lli->lli_jobid, 0, sizeof(lli->lli_jobid));
1021 #define MAX_STRING_SIZE 128
1023 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1025 #define LSI_BDI_INITIALIZED 0x00400000
1027 #ifndef HAVE_BDI_CAP_MAP_COPY
1028 # define BDI_CAP_MAP_COPY 0
1031 static int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
1033 struct lustre_sb_info *lsi = s2lsi(sb);
1034 char buf[MAX_STRING_SIZE];
1038 err = bdi_init(&lsi->lsi_bdi);
1042 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1043 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1044 lsi->lsi_bdi.name = "lustre";
1045 va_start(args, fmt);
1046 vsnprintf(buf, MAX_STRING_SIZE, fmt, args);
1048 err = bdi_register(&lsi->lsi_bdi, NULL, "%s", buf);
1051 sb->s_bdi = &lsi->lsi_bdi;
1055 #endif /* !HAVE_SUPER_SETUP_BDI_NAME */
1057 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
1059 struct lustre_profile *lprof = NULL;
1060 struct lustre_sb_info *lsi = s2lsi(sb);
1061 struct ll_sb_info *sbi = NULL;
1062 char *dt = NULL, *md = NULL;
1063 char *profilenm = get_profile_name(sb);
1064 struct config_llog_instance *cfg;
1065 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1066 const int instlen = 16 + 2;
1067 unsigned long cfg_instance = ll_get_cfg_instance(sb);
1068 char name[MAX_STRING_SIZE];
1077 /* for ASLR, to map between cfg_instance and hashed ptr */
1078 CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1079 profilenm, cfg_instance, sb);
1081 try_module_get(THIS_MODULE);
1085 GOTO(out_free_cfg, err = -ENOMEM);
1087 /* client additional sb info */
1088 lsi->lsi_llsbi = sbi = ll_init_sbi();
1090 GOTO(out_free_cfg, err = PTR_ERR(sbi));
1092 err = ll_options(lsi->lsi_lmd->lmd_opts, sbi);
1094 GOTO(out_free_cfg, err);
1096 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1097 sb->s_d_op = &ll_d_ops;
1100 generate_random_uuid(uuid.b);
1101 snprintf(sbi->ll_sb_uuid.uuid, UUID_SIZE, "%pU", uuid.b);
1103 CDEBUG(D_CONFIG, "llite sb uuid: %s\n", sbi->ll_sb_uuid.uuid);
1106 len = strlen(profilenm);
1107 ptr = strrchr(profilenm, '-');
1108 if (ptr && (strcmp(ptr, "-client") == 0))
1111 if (len > LUSTRE_MAXFSNAME) {
1112 if (unlikely(len >= MAX_STRING_SIZE))
1113 len = MAX_STRING_SIZE - 1;
1114 strncpy(name, profilenm, len);
1116 err = -ENAMETOOLONG;
1117 CERROR("%s: fsname longer than %u characters: rc = %d\n",
1118 name, LUSTRE_MAXFSNAME, err);
1119 GOTO(out_free_cfg, err);
1121 strncpy(sbi->ll_fsname, profilenm, len);
1122 sbi->ll_fsname[len] = '\0';
1125 snprintf(name, sizeof(name), "%.*s-%016lx", len,
1126 profilenm, cfg_instance);
1128 err = super_setup_bdi_name(sb, "%s", name);
1130 GOTO(out_free_cfg, err);
1132 /* Call ll_debugfs_register_super() before lustre_process_log()
1133 * so that "llite.*.*" params can be processed correctly.
1135 err = ll_debugfs_register_super(sb, name);
1137 CERROR("%s: could not register mountpoint in llite: rc = %d\n",
1138 sbi->ll_fsname, err);
1142 /* The cfg_instance is a value unique to this super, in case some
1143 * joker tries to mount the same fs at two mount points.
1145 cfg->cfg_instance = cfg_instance;
1146 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1147 cfg->cfg_callback = class_config_llog_handler;
1148 cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1149 /* set up client obds */
1150 err = lustre_process_log(sb, profilenm, cfg);
1152 GOTO(out_debugfs, err);
1154 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1155 lprof = class_get_profile(profilenm);
1156 if (lprof == NULL) {
1157 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1158 " read from the MGS. Does that filesystem "
1159 "exist?\n", profilenm);
1160 GOTO(out_debugfs, err = -EINVAL);
1162 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1163 lprof->lp_md, lprof->lp_dt);
1165 dt_len = strlen(lprof->lp_dt) + instlen + 2;
1166 OBD_ALLOC(dt, dt_len);
1168 GOTO(out_profile, err = -ENOMEM);
1169 snprintf(dt, dt_len - 1, "%s-%016lx", lprof->lp_dt, cfg_instance);
1171 md_len = strlen(lprof->lp_md) + instlen + 2;
1172 OBD_ALLOC(md, md_len);
1174 GOTO(out_free_dt, err = -ENOMEM);
1175 snprintf(md, md_len - 1, "%s-%016lx", lprof->lp_md, cfg_instance);
1177 /* connections, registrations, sb setup */
1178 err = client_common_fill_super(sb, md, dt, mnt);
1180 GOTO(out_free_md, err);
1182 sbi->ll_client_common_fill_super_succeeded = 1;
1186 OBD_FREE(md, md_len);
1189 OBD_FREE(dt, dt_len);
1192 class_put_profile(lprof);
1195 ll_debugfs_unregister_super(sb);
1202 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1203 LCONSOLE_WARN("Mounted %s\n", profilenm);
1205 } /* ll_fill_super */
1207 void ll_put_super(struct super_block *sb)
1209 struct config_llog_instance cfg, params_cfg;
1210 struct obd_device *obd;
1211 struct lustre_sb_info *lsi = s2lsi(sb);
1212 struct ll_sb_info *sbi = ll_s2sbi(sb);
1213 char *profilenm = get_profile_name(sb);
1214 unsigned long cfg_instance = ll_get_cfg_instance(sb);
1216 int next, force = 1, rc = 0;
1220 GOTO(out_no_sbi, 0);
1222 /* Should replace instance_id with something better for ASLR */
1223 CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1224 profilenm, cfg_instance, sb);
1226 cfg.cfg_instance = cfg_instance;
1227 lustre_end_log(sb, profilenm, &cfg);
1229 params_cfg.cfg_instance = cfg_instance;
1230 lustre_end_log(sb, PARAMS_FILENAME, ¶ms_cfg);
1232 if (sbi->ll_md_exp) {
1233 obd = class_exp2obd(sbi->ll_md_exp);
1235 force = obd->obd_force;
1238 /* Wait for unstable pages to be committed to stable storage */
1240 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1241 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1242 atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1246 ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1247 if (force == 0 && rc != -EINTR)
1248 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1250 /* We need to set force before the lov_disconnect in
1251 * lustre_common_put_super, since l_d cleans up osc's as well.
1255 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1257 obd->obd_force = force;
1261 if (sbi->ll_client_common_fill_super_succeeded) {
1262 /* Only if client_common_fill_super succeeded */
1263 client_common_put_super(sb);
1267 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)))
1268 class_manual_cleanup(obd);
1270 if (sbi->ll_flags & LL_SBI_VERBOSE)
1271 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1274 class_del_profile(profilenm);
1276 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1277 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1278 bdi_destroy(&lsi->lsi_bdi);
1279 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1284 lsi->lsi_llsbi = NULL;
1286 lustre_common_put_super(sb);
1288 cl_env_cache_purge(~0);
1290 module_put(THIS_MODULE);
1293 } /* client_put_super */
1295 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1297 struct inode *inode = NULL;
1299 /* NOTE: we depend on atomic igrab() -bzzz */
1300 lock_res_and_lock(lock);
1301 if (lock->l_resource->lr_lvb_inode) {
1302 struct ll_inode_info * lli;
1303 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1304 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1305 inode = igrab(lock->l_resource->lr_lvb_inode);
1307 inode = lock->l_resource->lr_lvb_inode;
1308 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1309 D_WARNING, lock, "lr_lvb_inode %p is "
1310 "bogus: magic %08x",
1311 lock->l_resource->lr_lvb_inode,
1312 lli->lli_inode_magic);
1316 unlock_res_and_lock(lock);
1320 void ll_dir_clear_lsm_md(struct inode *inode)
1322 struct ll_inode_info *lli = ll_i2info(inode);
1324 LASSERT(S_ISDIR(inode->i_mode));
1326 if (lli->lli_lsm_md) {
1327 lmv_free_memmd(lli->lli_lsm_md);
1328 lli->lli_lsm_md = NULL;
1331 if (lli->lli_default_lsm_md) {
1332 lmv_free_memmd(lli->lli_default_lsm_md);
1333 lli->lli_default_lsm_md = NULL;
1337 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1338 const struct lu_fid *fid,
1339 struct lustre_md *md)
1341 struct ll_sb_info *sbi = ll_s2sbi(sb);
1342 struct mdt_body *body = md->body;
1343 struct inode *inode;
1347 ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1348 inode = iget_locked(sb, ino);
1349 if (inode == NULL) {
1350 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1351 sbi->ll_fsname, PFID(fid));
1352 RETURN(ERR_PTR(-ENOENT));
1355 if (inode->i_state & I_NEW) {
1356 struct ll_inode_info *lli = ll_i2info(inode);
1357 struct lmv_stripe_md *lsm = md->lmv;
1359 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1360 (body->mbo_mode & S_IFMT);
1361 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1364 inode->i_mtime.tv_sec = 0;
1365 inode->i_atime.tv_sec = 0;
1366 inode->i_ctime.tv_sec = 0;
1369 #ifdef HAVE_BACKING_DEV_INFO
1370 /* initializing backing dev info. */
1371 inode->i_mapping->backing_dev_info =
1372 &s2lsi(inode->i_sb)->lsi_bdi;
1374 inode->i_op = &ll_dir_inode_operations;
1375 inode->i_fop = &ll_dir_operations;
1376 lli->lli_fid = *fid;
1379 LASSERT(lsm != NULL);
1380 /* master object FID */
1381 lli->lli_pfid = body->mbo_fid1;
1382 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1383 lli, PFID(fid), PFID(&lli->lli_pfid));
1384 unlock_new_inode(inode);
1390 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1393 struct lmv_stripe_md *lsm = md->lmv;
1394 struct ll_inode_info *lli = ll_i2info(inode);
1397 LASSERT(lsm != NULL);
1399 CDEBUG(D_INODE, "%s: "DFID" set dir layout:\n",
1400 ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid));
1401 lsm_md_dump(D_INODE, lsm);
1403 if (!lmv_dir_striped(lsm))
1406 /* XXX sigh, this lsm_root initialization should be in
1407 * LMV layer, but it needs ll_iget right now, so we
1408 * put this here right now. */
1409 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1410 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1411 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1413 if (!fid_is_sane(fid))
1416 /* Unfortunately ll_iget will call ll_update_inode,
1417 * where the initialization of slave inode is slightly
1418 * different, so it reset lsm_md to NULL to avoid
1419 * initializing lsm for slave inode. */
1420 lsm->lsm_md_oinfo[i].lmo_root =
1421 ll_iget_anon_dir(inode->i_sb, fid, md);
1422 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1423 int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1425 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1427 iput(lsm->lsm_md_oinfo[i].lmo_root);
1428 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1434 lli->lli_lsm_md = lsm;
1439 static void ll_update_default_lsm_md(struct inode *inode, struct lustre_md *md)
1441 struct ll_inode_info *lli = ll_i2info(inode);
1443 if (!md->default_lmv) {
1444 /* clear default lsm */
1445 if (lli->lli_default_lsm_md) {
1446 down_write(&lli->lli_lsm_sem);
1447 if (lli->lli_default_lsm_md) {
1448 lmv_free_memmd(lli->lli_default_lsm_md);
1449 lli->lli_default_lsm_md = NULL;
1451 up_write(&lli->lli_lsm_sem);
1453 } else if (lli->lli_default_lsm_md) {
1454 /* update default lsm if it changes */
1455 down_read(&lli->lli_lsm_sem);
1456 if (lli->lli_default_lsm_md &&
1457 !lsm_md_eq(lli->lli_default_lsm_md, md->default_lmv)) {
1458 up_read(&lli->lli_lsm_sem);
1459 down_write(&lli->lli_lsm_sem);
1460 if (lli->lli_default_lsm_md)
1461 lmv_free_memmd(lli->lli_default_lsm_md);
1462 lli->lli_default_lsm_md = md->default_lmv;
1463 lsm_md_dump(D_INODE, md->default_lmv);
1464 md->default_lmv = NULL;
1465 up_write(&lli->lli_lsm_sem);
1467 up_read(&lli->lli_lsm_sem);
1470 /* init default lsm */
1471 down_write(&lli->lli_lsm_sem);
1472 lli->lli_default_lsm_md = md->default_lmv;
1473 lsm_md_dump(D_INODE, md->default_lmv);
1474 md->default_lmv = NULL;
1475 up_write(&lli->lli_lsm_sem);
1479 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1481 struct ll_inode_info *lli = ll_i2info(inode);
1482 struct lmv_stripe_md *lsm = md->lmv;
1487 LASSERT(S_ISDIR(inode->i_mode));
1488 CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1489 PFID(ll_inode2fid(inode)));
1491 /* update default LMV */
1492 if (md->default_lmv)
1493 ll_update_default_lsm_md(inode, md);
1496 * no striped information from request, lustre_md from req does not
1497 * include stripeEA, see ll_md_setattr()
1503 * normally dir layout doesn't change, only take read lock to check
1504 * that to avoid blocking other MD operations.
1506 if (lli->lli_lsm_md)
1507 down_read(&lli->lli_lsm_sem);
1509 down_write(&lli->lli_lsm_sem);
1512 * if dir layout mismatch, check whether version is increased, which
1513 * means layout is changed, this happens in dir migration and lfsck.
1515 * foreign LMV should not change.
1517 if (lli->lli_lsm_md && !lsm_md_eq(lli->lli_lsm_md, lsm)) {
1518 if (lmv_dir_striped(lli->lli_lsm_md) &&
1519 lsm->lsm_md_layout_version <=
1520 lli->lli_lsm_md->lsm_md_layout_version) {
1521 CERROR("%s: "DFID" dir layout mismatch:\n",
1522 ll_i2sbi(inode)->ll_fsname,
1523 PFID(&lli->lli_fid));
1524 lsm_md_dump(D_ERROR, lli->lli_lsm_md);
1525 lsm_md_dump(D_ERROR, lsm);
1526 GOTO(unlock, rc = -EINVAL);
1529 /* layout changed, switch to write lock */
1530 up_read(&lli->lli_lsm_sem);
1531 down_write(&lli->lli_lsm_sem);
1532 ll_dir_clear_lsm_md(inode);
1535 /* set directory layout */
1536 if (!lli->lli_lsm_md) {
1537 struct cl_attr *attr;
1539 rc = ll_init_lsm_md(inode, md);
1540 up_write(&lli->lli_lsm_sem);
1544 /* set md->lmv to NULL, so the following free lustre_md
1545 * will not free this lsm */
1549 * md_merge_attr() may take long, since lsm is already set,
1550 * switch to read lock.
1552 down_read(&lli->lli_lsm_sem);
1554 if (!lmv_dir_striped(lli->lli_lsm_md))
1557 OBD_ALLOC_PTR(attr);
1559 GOTO(unlock, rc = -ENOMEM);
1561 /* validate the lsm */
1562 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1563 ll_md_blocking_ast);
1569 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1570 md->body->mbo_nlink = attr->cat_nlink;
1571 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1572 md->body->mbo_size = attr->cat_size;
1573 if (md->body->mbo_valid & OBD_MD_FLATIME)
1574 md->body->mbo_atime = attr->cat_atime;
1575 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1576 md->body->mbo_ctime = attr->cat_ctime;
1577 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1578 md->body->mbo_mtime = attr->cat_mtime;
1583 up_read(&lli->lli_lsm_sem);
1588 void ll_clear_inode(struct inode *inode)
1590 struct ll_inode_info *lli = ll_i2info(inode);
1591 struct ll_sb_info *sbi = ll_i2sbi(inode);
1595 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1596 PFID(ll_inode2fid(inode)), inode);
1598 if (S_ISDIR(inode->i_mode)) {
1599 /* these should have been cleared in ll_file_release */
1600 LASSERT(lli->lli_opendir_key == NULL);
1601 LASSERT(lli->lli_sai == NULL);
1602 LASSERT(lli->lli_opendir_pid == 0);
1604 pcc_inode_free(inode);
1607 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1609 LASSERT(!lli->lli_open_fd_write_count);
1610 LASSERT(!lli->lli_open_fd_read_count);
1611 LASSERT(!lli->lli_open_fd_exec_count);
1613 if (lli->lli_mds_write_och)
1614 ll_md_real_close(inode, FMODE_WRITE);
1615 if (lli->lli_mds_exec_och)
1616 ll_md_real_close(inode, FMODE_EXEC);
1617 if (lli->lli_mds_read_och)
1618 ll_md_real_close(inode, FMODE_READ);
1620 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1621 OBD_FREE(lli->lli_symlink_name,
1622 strlen(lli->lli_symlink_name) + 1);
1623 lli->lli_symlink_name = NULL;
1626 ll_xattr_cache_destroy(inode);
1628 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
1629 forget_all_cached_acls(inode);
1630 if (lli->lli_posix_acl) {
1631 posix_acl_release(lli->lli_posix_acl);
1632 lli->lli_posix_acl = NULL;
1635 lli->lli_inode_magic = LLI_INODE_DEAD;
1637 if (S_ISDIR(inode->i_mode))
1638 ll_dir_clear_lsm_md(inode);
1639 else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1640 LASSERT(list_empty(&lli->lli_agl_list));
1643 * XXX This has to be done before lsm is freed below, because
1644 * cl_object still uses inode lsm.
1646 cl_inode_fini(inode);
1651 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1653 struct lustre_md md;
1654 struct inode *inode = dentry->d_inode;
1655 struct ll_sb_info *sbi = ll_i2sbi(inode);
1656 struct ptlrpc_request *request = NULL;
1660 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1661 LUSTRE_OPC_ANY, NULL);
1662 if (IS_ERR(op_data))
1663 RETURN(PTR_ERR(op_data));
1665 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1667 ptlrpc_req_finished(request);
1668 if (rc == -ENOENT) {
1670 /* Unlinked special device node? Or just a race?
1671 * Pretend we done everything. */
1672 if (!S_ISREG(inode->i_mode) &&
1673 !S_ISDIR(inode->i_mode)) {
1674 ia_valid = op_data->op_attr.ia_valid;
1675 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1676 rc = simple_setattr(dentry, &op_data->op_attr);
1677 op_data->op_attr.ia_valid = ia_valid;
1679 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1680 CERROR("md_setattr fails: rc = %d\n", rc);
1685 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1686 sbi->ll_md_exp, &md);
1688 ptlrpc_req_finished(request);
1692 ia_valid = op_data->op_attr.ia_valid;
1693 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1694 * cache is not cleared yet. */
1695 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1696 if (S_ISREG(inode->i_mode))
1698 rc = simple_setattr(dentry, &op_data->op_attr);
1699 if (S_ISREG(inode->i_mode))
1700 inode_unlock(inode);
1701 op_data->op_attr.ia_valid = ia_valid;
1703 rc = ll_update_inode(inode, &md);
1704 ptlrpc_req_finished(request);
1709 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1710 * object(s) determine the file size and mtime. Otherwise, the MDS will
1711 * keep these values until such a time that objects are allocated for it.
1712 * We do the MDS operations first, as it is checking permissions for us.
1713 * We don't to the MDS RPC if there is nothing that we want to store there,
1714 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1715 * going to do an RPC anyways.
1717 * If we are doing a truncate, we will send the mtime and ctime updates
1718 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1719 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1722 * In case of HSMimport, we only set attr on MDS.
1724 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr,
1725 enum op_xvalid xvalid, bool hsm_import)
1727 struct inode *inode = dentry->d_inode;
1728 struct ll_inode_info *lli = ll_i2info(inode);
1729 struct md_op_data *op_data = NULL;
1730 ktime_t kstart = ktime_get();
1735 CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1736 "valid %x, hsm_import %d\n",
1737 ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid),
1738 inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1741 if (attr->ia_valid & ATTR_SIZE) {
1742 /* Check new size against VFS/VM file size limit and rlimit */
1743 rc = inode_newsize_ok(inode, attr->ia_size);
1747 /* The maximum Lustre file size is variable, based on the
1748 * OST maximum object size and number of stripes. This
1749 * needs another check in addition to the VFS check above. */
1750 if (attr->ia_size > ll_file_maxbytes(inode)) {
1751 CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
1752 PFID(&lli->lli_fid), attr->ia_size,
1753 ll_file_maxbytes(inode));
1757 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1760 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1761 if (attr->ia_valid & TIMES_SET_FLAGS) {
1762 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1763 !cfs_capable(CFS_CAP_FOWNER))
1767 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1768 if (!(xvalid & OP_XVALID_CTIME_SET) &&
1769 (attr->ia_valid & ATTR_CTIME)) {
1770 attr->ia_ctime = current_time(inode);
1771 xvalid |= OP_XVALID_CTIME_SET;
1773 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1774 (attr->ia_valid & ATTR_ATIME)) {
1775 attr->ia_atime = current_time(inode);
1776 attr->ia_valid |= ATTR_ATIME_SET;
1778 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1779 (attr->ia_valid & ATTR_MTIME)) {
1780 attr->ia_mtime = current_time(inode);
1781 attr->ia_valid |= ATTR_MTIME_SET;
1784 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1785 CDEBUG(D_INODE, "setting mtime %lld, ctime %lld, now = %lld\n",
1786 (s64)attr->ia_mtime.tv_sec, (s64)attr->ia_ctime.tv_sec,
1787 ktime_get_real_seconds());
1789 if (S_ISREG(inode->i_mode)) {
1790 if (attr->ia_valid & ATTR_SIZE)
1791 inode_dio_write_done(inode);
1792 inode_unlock(inode);
1795 /* We always do an MDS RPC, even if we're only changing the size;
1796 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1798 OBD_ALLOC_PTR(op_data);
1799 if (op_data == NULL)
1800 GOTO(out, rc = -ENOMEM);
1802 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1803 /* If we are changing file size, file content is
1804 * modified, flag it.
1806 xvalid |= OP_XVALID_OWNEROVERRIDE;
1807 op_data->op_bias |= MDS_DATA_MODIFIED;
1808 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1811 if (attr->ia_valid & ATTR_FILE) {
1812 struct ll_file_data *fd = LUSTRE_FPRIVATE(attr->ia_file);
1814 if (fd->fd_lease_och)
1815 op_data->op_bias |= MDS_TRUNC_KEEP_LEASE;
1818 op_data->op_attr = *attr;
1819 op_data->op_xvalid = xvalid;
1821 rc = ll_md_setattr(dentry, op_data);
1825 if (!S_ISREG(inode->i_mode) || hsm_import)
1828 if (attr->ia_valid & (ATTR_SIZE | ATTR_ATIME | ATTR_ATIME_SET |
1829 ATTR_MTIME | ATTR_MTIME_SET | ATTR_CTIME) ||
1830 xvalid & OP_XVALID_CTIME_SET) {
1831 bool cached = false;
1833 rc = pcc_inode_setattr(inode, attr, &cached);
1836 CERROR("%s: PCC inode "DFID" setattr failed: "
1838 ll_i2sbi(inode)->ll_fsname,
1839 PFID(&lli->lli_fid), rc);
1843 /* For truncate and utimes sending attributes to OSTs,
1844 * setting mtime/atime to the past will be performed
1845 * under PW [0:EOF] extent lock (new_size:EOF for
1846 * truncate). It may seem excessive to send mtime/atime
1847 * updates to OSTs when not setting times to past, but
1848 * it is necessary due to possible time
1849 * de-synchronization between MDT inode and OST objects
1851 rc = cl_setattr_ost(lli->lli_clob, attr, xvalid, 0);
1855 /* If the file was restored, it needs to set dirty flag.
1857 * We've already sent MDS_DATA_MODIFIED flag in
1858 * ll_md_setattr() for truncate. However, the MDT refuses to
1859 * set the HS_DIRTY flag on released files, so we have to set
1860 * it again if the file has been restored. Please check how
1861 * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1863 * Please notice that if the file is not released, the previous
1864 * MDS_DATA_MODIFIED has taken effect and usually
1865 * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1866 * This way we can save an RPC for common open + trunc
1868 if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1869 struct hsm_state_set hss = {
1870 .hss_valid = HSS_SETMASK,
1871 .hss_setmask = HS_DIRTY,
1875 rc2 = ll_hsm_state_set(inode, &hss);
1876 /* truncate and write can happen at the same time, so that
1877 * the file can be set modified even though the file is not
1878 * restored from released state, and ll_hsm_state_set() is
1879 * not applicable for the file, and rc2 < 0 is normal in this
1882 CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1883 PFID(ll_inode2fid(inode)), rc2);
1888 if (op_data != NULL)
1889 ll_finish_md_op_data(op_data);
1891 if (S_ISREG(inode->i_mode)) {
1893 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1894 inode_dio_wait(inode);
1895 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
1896 * flag. ll_update_inode (called from ll_md_setattr), clears
1897 * inode flags, so there is a gap where S_NOSEC is not set.
1898 * This can cause a writer to take the i_mutex unnecessarily,
1899 * but this is safe to do and should be rare. */
1900 inode_has_no_xattr(inode);
1904 ll_stats_ops_tally(ll_i2sbi(inode), attr->ia_valid & ATTR_SIZE ?
1905 LPROC_LL_TRUNC : LPROC_LL_SETATTR,
1906 ktime_us_delta(ktime_get(), kstart));
1911 int ll_setattr(struct dentry *de, struct iattr *attr)
1913 int mode = de->d_inode->i_mode;
1914 enum op_xvalid xvalid = 0;
1916 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1917 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1918 xvalid |= OP_XVALID_OWNEROVERRIDE;
1920 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1921 (ATTR_SIZE|ATTR_MODE)) &&
1922 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1923 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1924 !(attr->ia_mode & S_ISGID))))
1925 attr->ia_valid |= ATTR_FORCE;
1927 if ((attr->ia_valid & ATTR_MODE) &&
1929 !(attr->ia_mode & S_ISUID) &&
1930 !(attr->ia_valid & ATTR_KILL_SUID))
1931 attr->ia_valid |= ATTR_KILL_SUID;
1933 if ((attr->ia_valid & ATTR_MODE) &&
1934 ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1935 !(attr->ia_mode & S_ISGID) &&
1936 !(attr->ia_valid & ATTR_KILL_SGID))
1937 attr->ia_valid |= ATTR_KILL_SGID;
1939 return ll_setattr_raw(de, attr, xvalid, false);
1942 int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
1945 struct obd_statfs obd_osfs = { 0 };
1950 max_age = ktime_get_seconds() - sbi->ll_statfs_max_age;
1952 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1953 flags |= OBD_STATFS_NODELAY;
1955 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1959 osfs->os_type = LL_SUPER_MAGIC;
1961 CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1962 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree, osfs->os_files);
1964 if (osfs->os_state & OS_STATE_SUM)
1967 rc = obd_statfs(NULL, sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1968 if (rc) /* Possibly a filesystem with no OSTs. Report MDT totals. */
1971 CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1972 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1975 osfs->os_bsize = obd_osfs.os_bsize;
1976 osfs->os_blocks = obd_osfs.os_blocks;
1977 osfs->os_bfree = obd_osfs.os_bfree;
1978 osfs->os_bavail = obd_osfs.os_bavail;
1980 /* If we have _some_ OSTs, but don't have as many free objects on the
1981 * OSTs as inodes on the MDTs, reduce the reported number of inodes
1982 * to compensate, so that the "inodes in use" number is correct.
1983 * This should be kept in sync with lod_statfs() behaviour.
1985 if (obd_osfs.os_files && obd_osfs.os_ffree < osfs->os_ffree) {
1986 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1988 osfs->os_ffree = obd_osfs.os_ffree;
1995 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1997 struct super_block *sb = de->d_sb;
1998 struct obd_statfs osfs;
1999 __u64 fsid = huge_encode_dev(sb->s_dev);
2000 ktime_t kstart = ktime_get();
2003 CDEBUG(D_VFSTRACE, "VFS Op:sb=%s (%p)\n", sb->s_id, sb);
2005 /* Some amount of caching on the client is allowed */
2006 rc = ll_statfs_internal(ll_s2sbi(sb), &osfs, OBD_STATFS_SUM);
2010 statfs_unpack(sfs, &osfs);
2012 /* We need to downshift for all 32-bit kernels, because we can't
2013 * tell if the kernel is being called via sys_statfs64() or not.
2014 * Stop before overflowing f_bsize - in which case it is better
2015 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
2016 if (sizeof(long) < 8) {
2017 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
2020 osfs.os_blocks >>= 1;
2021 osfs.os_bfree >>= 1;
2022 osfs.os_bavail >>= 1;
2026 sfs->f_blocks = osfs.os_blocks;
2027 sfs->f_bfree = osfs.os_bfree;
2028 sfs->f_bavail = osfs.os_bavail;
2029 sfs->f_fsid.val[0] = (__u32)fsid;
2030 sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
2032 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STATFS,
2033 ktime_us_delta(ktime_get(), kstart));
2038 void ll_inode_size_lock(struct inode *inode)
2040 struct ll_inode_info *lli;
2042 LASSERT(!S_ISDIR(inode->i_mode));
2044 lli = ll_i2info(inode);
2045 mutex_lock(&lli->lli_size_mutex);
2048 void ll_inode_size_unlock(struct inode *inode)
2050 struct ll_inode_info *lli;
2052 lli = ll_i2info(inode);
2053 mutex_unlock(&lli->lli_size_mutex);
2056 void ll_update_inode_flags(struct inode *inode, int ext_flags)
2058 inode->i_flags = ll_ext_to_inode_flags(ext_flags);
2059 if (ext_flags & LUSTRE_PROJINHERIT_FL)
2060 ll_file_set_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
2062 ll_file_clear_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
2065 int ll_update_inode(struct inode *inode, struct lustre_md *md)
2067 struct ll_inode_info *lli = ll_i2info(inode);
2068 struct mdt_body *body = md->body;
2069 struct ll_sb_info *sbi = ll_i2sbi(inode);
2072 if (body->mbo_valid & OBD_MD_FLEASIZE) {
2073 rc = cl_file_inode_init(inode, md);
2078 if (S_ISDIR(inode->i_mode)) {
2079 rc = ll_update_lsm_md(inode, md);
2084 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
2085 if (body->mbo_valid & OBD_MD_FLACL) {
2086 spin_lock(&lli->lli_lock);
2087 if (lli->lli_posix_acl)
2088 posix_acl_release(lli->lli_posix_acl);
2089 lli->lli_posix_acl = md->posix_acl;
2090 spin_unlock(&lli->lli_lock);
2093 inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
2094 sbi->ll_flags & LL_SBI_32BIT_API);
2095 inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
2097 if (body->mbo_valid & OBD_MD_FLATIME) {
2098 if (body->mbo_atime > inode->i_atime.tv_sec)
2099 inode->i_atime.tv_sec = body->mbo_atime;
2100 lli->lli_atime = body->mbo_atime;
2103 if (body->mbo_valid & OBD_MD_FLMTIME) {
2104 if (body->mbo_mtime > inode->i_mtime.tv_sec) {
2106 "setting ino %lu mtime from %lld to %llu\n",
2107 inode->i_ino, (s64)inode->i_mtime.tv_sec,
2109 inode->i_mtime.tv_sec = body->mbo_mtime;
2111 lli->lli_mtime = body->mbo_mtime;
2114 if (body->mbo_valid & OBD_MD_FLCTIME) {
2115 if (body->mbo_ctime > inode->i_ctime.tv_sec)
2116 inode->i_ctime.tv_sec = body->mbo_ctime;
2117 lli->lli_ctime = body->mbo_ctime;
2120 /* Clear i_flags to remove S_NOSEC before permissions are updated */
2121 if (body->mbo_valid & OBD_MD_FLFLAGS)
2122 ll_update_inode_flags(inode, body->mbo_flags);
2123 if (body->mbo_valid & OBD_MD_FLMODE)
2124 inode->i_mode = (inode->i_mode & S_IFMT) |
2125 (body->mbo_mode & ~S_IFMT);
2127 if (body->mbo_valid & OBD_MD_FLTYPE)
2128 inode->i_mode = (inode->i_mode & ~S_IFMT) |
2129 (body->mbo_mode & S_IFMT);
2131 LASSERT(inode->i_mode != 0);
2132 if (S_ISREG(inode->i_mode))
2133 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
2134 LL_MAX_BLKSIZE_BITS);
2136 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
2138 if (body->mbo_valid & OBD_MD_FLUID)
2139 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
2140 if (body->mbo_valid & OBD_MD_FLGID)
2141 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
2142 if (body->mbo_valid & OBD_MD_FLPROJID)
2143 lli->lli_projid = body->mbo_projid;
2144 if (body->mbo_valid & OBD_MD_FLNLINK)
2145 set_nlink(inode, body->mbo_nlink);
2146 if (body->mbo_valid & OBD_MD_FLRDEV)
2147 inode->i_rdev = old_decode_dev(body->mbo_rdev);
2149 if (body->mbo_valid & OBD_MD_FLID) {
2150 /* FID shouldn't be changed! */
2151 if (fid_is_sane(&lli->lli_fid)) {
2152 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
2153 "Trying to change FID "DFID
2154 " to the "DFID", inode "DFID"(%p)\n",
2155 PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
2156 PFID(ll_inode2fid(inode)), inode);
2158 lli->lli_fid = body->mbo_fid1;
2162 LASSERT(fid_seq(&lli->lli_fid) != 0);
2164 if (body->mbo_valid & OBD_MD_FLSIZE) {
2165 i_size_write(inode, body->mbo_size);
2167 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
2168 PFID(ll_inode2fid(inode)),
2169 (unsigned long long)body->mbo_size);
2171 if (body->mbo_valid & OBD_MD_FLBLOCKS)
2172 inode->i_blocks = body->mbo_blocks;
2175 if (body->mbo_valid & OBD_MD_TSTATE) {
2176 /* Set LLIF_FILE_RESTORING if restore ongoing and
2177 * clear it when done to ensure to start again
2178 * glimpsing updated attrs
2180 if (body->mbo_t_state & MS_RESTORE)
2181 ll_file_set_flag(lli, LLIF_FILE_RESTORING);
2183 ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
2189 int ll_read_inode2(struct inode *inode, void *opaque)
2191 struct lustre_md *md = opaque;
2192 struct ll_inode_info *lli = ll_i2info(inode);
2196 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2197 PFID(&lli->lli_fid), inode);
2199 /* Core attributes from the MDS first. This is a new inode, and
2200 * the VFS doesn't zero times in the core inode so we have to do
2201 * it ourselves. They will be overwritten by either MDS or OST
2202 * attributes - we just need to make sure they aren't newer.
2204 inode->i_mtime.tv_sec = 0;
2205 inode->i_atime.tv_sec = 0;
2206 inode->i_ctime.tv_sec = 0;
2208 rc = ll_update_inode(inode, md);
2212 /* OIDEBUG(inode); */
2214 #ifdef HAVE_BACKING_DEV_INFO
2215 /* initializing backing dev info. */
2216 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2218 if (S_ISREG(inode->i_mode)) {
2219 struct ll_sb_info *sbi = ll_i2sbi(inode);
2220 inode->i_op = &ll_file_inode_operations;
2221 inode->i_fop = sbi->ll_fop;
2222 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2224 } else if (S_ISDIR(inode->i_mode)) {
2225 inode->i_op = &ll_dir_inode_operations;
2226 inode->i_fop = &ll_dir_operations;
2228 } else if (S_ISLNK(inode->i_mode)) {
2229 inode->i_op = &ll_fast_symlink_inode_operations;
2232 inode->i_op = &ll_special_inode_operations;
2234 init_special_inode(inode, inode->i_mode,
2243 void ll_delete_inode(struct inode *inode)
2245 struct ll_inode_info *lli = ll_i2info(inode);
2246 struct address_space *mapping = &inode->i_data;
2247 unsigned long nrpages;
2248 unsigned long flags;
2252 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL) {
2253 /* It is last chance to write out dirty pages,
2254 * otherwise we may lose data while umount.
2256 * If i_nlink is 0 then just discard data. This is safe because
2257 * local inode gets i_nlink 0 from server only for the last
2258 * unlink, so that file is not opened somewhere else
2260 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, inode->i_nlink ?
2261 CL_FSYNC_LOCAL : CL_FSYNC_DISCARD, 1);
2263 truncate_inode_pages_final(mapping);
2265 /* Workaround for LU-118: Note nrpages may not be totally updated when
2266 * truncate_inode_pages() returns, as there can be a page in the process
2267 * of deletion (inside __delete_from_page_cache()) in the specified
2268 * range. Thus mapping->nrpages can be non-zero when this function
2269 * returns even after truncation of the whole mapping. Only do this if
2270 * npages isn't already zero.
2272 nrpages = mapping->nrpages;
2274 xa_lock_irqsave(&mapping->i_pages, flags);
2275 nrpages = mapping->nrpages;
2276 xa_unlock_irqrestore(&mapping->i_pages, flags);
2277 } /* Workaround end */
2279 LASSERTF(nrpages == 0, "%s: inode="DFID"(%p) nrpages=%lu, "
2280 "see https://jira.whamcloud.com/browse/LU-118\n",
2281 ll_i2sbi(inode)->ll_fsname,
2282 PFID(ll_inode2fid(inode)), inode, nrpages);
2284 ll_clear_inode(inode);
2290 int ll_iocontrol(struct inode *inode, struct file *file,
2291 unsigned int cmd, unsigned long arg)
2293 struct ll_sb_info *sbi = ll_i2sbi(inode);
2294 struct ptlrpc_request *req = NULL;
2299 case FS_IOC_GETFLAGS: {
2300 struct mdt_body *body;
2301 struct md_op_data *op_data;
2303 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2304 0, 0, LUSTRE_OPC_ANY,
2306 if (IS_ERR(op_data))
2307 RETURN(PTR_ERR(op_data));
2309 op_data->op_valid = OBD_MD_FLFLAGS;
2310 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2311 ll_finish_md_op_data(op_data);
2313 CERROR("%s: failure inode "DFID": rc = %d\n",
2314 sbi->ll_md_exp->exp_obd->obd_name,
2315 PFID(ll_inode2fid(inode)), rc);
2319 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2321 flags = body->mbo_flags;
2323 ptlrpc_req_finished(req);
2325 RETURN(put_user(flags, (int __user *)arg));
2327 case FS_IOC_SETFLAGS: {
2329 struct md_op_data *op_data;
2330 struct cl_object *obj;
2331 struct fsxattr fa = { 0 };
2333 if (get_user(flags, (int __user *)arg))
2336 fa.fsx_projid = ll_i2info(inode)->lli_projid;
2337 if (flags & LUSTRE_PROJINHERIT_FL)
2338 fa.fsx_xflags = FS_XFLAG_PROJINHERIT;
2340 rc = ll_ioctl_check_project(inode, &fa);
2344 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2345 LUSTRE_OPC_ANY, NULL);
2346 if (IS_ERR(op_data))
2347 RETURN(PTR_ERR(op_data));
2349 op_data->op_attr_flags = flags;
2350 op_data->op_xvalid |= OP_XVALID_FLAGS;
2351 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2352 ll_finish_md_op_data(op_data);
2353 ptlrpc_req_finished(req);
2357 ll_update_inode_flags(inode, flags);
2359 obj = ll_i2info(inode)->lli_clob;
2363 OBD_ALLOC_PTR(attr);
2367 rc = cl_setattr_ost(obj, attr, OP_XVALID_FLAGS, flags);
2379 int ll_flush_ctx(struct inode *inode)
2381 struct ll_sb_info *sbi = ll_i2sbi(inode);
2383 CDEBUG(D_SEC, "flush context for user %d\n",
2384 from_kuid(&init_user_ns, current_uid()));
2386 obd_set_info_async(NULL, sbi->ll_md_exp,
2387 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2389 obd_set_info_async(NULL, sbi->ll_dt_exp,
2390 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2395 /* umount -f client means force down, don't save state */
2396 void ll_umount_begin(struct super_block *sb)
2398 struct ll_sb_info *sbi = ll_s2sbi(sb);
2399 struct obd_device *obd;
2400 struct obd_ioctl_data *ioc_data;
2404 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2405 sb->s_count, atomic_read(&sb->s_active));
2407 obd = class_exp2obd(sbi->ll_md_exp);
2409 CERROR("Invalid MDC connection handle %#llx\n",
2410 sbi->ll_md_exp->exp_handle.h_cookie);
2416 obd = class_exp2obd(sbi->ll_dt_exp);
2418 CERROR("Invalid LOV connection handle %#llx\n",
2419 sbi->ll_dt_exp->exp_handle.h_cookie);
2425 OBD_ALLOC_PTR(ioc_data);
2427 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2428 sizeof *ioc_data, ioc_data, NULL);
2430 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2431 sizeof *ioc_data, ioc_data, NULL);
2433 OBD_FREE_PTR(ioc_data);
2436 /* Really, we'd like to wait until there are no requests outstanding,
2437 * and then continue. For now, we just periodically checking for vfs
2438 * to decrement mnt_cnt and hope to finish it within 10sec.
2442 !may_umount(sbi->ll_mnt.mnt)) {
2450 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2452 struct ll_sb_info *sbi = ll_s2sbi(sb);
2453 char *profilenm = get_profile_name(sb);
2457 if ((*flags & MS_RDONLY) != (sb->s_flags & SB_RDONLY)) {
2458 read_only = *flags & MS_RDONLY;
2459 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2460 sizeof(KEY_READ_ONLY),
2461 KEY_READ_ONLY, sizeof(read_only),
2464 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2465 profilenm, read_only ?
2466 "read-only" : "read-write", err);
2471 sb->s_flags |= SB_RDONLY;
2473 sb->s_flags &= ~SB_RDONLY;
2475 if (sbi->ll_flags & LL_SBI_VERBOSE)
2476 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2477 read_only ? "read-only" : "read-write");
2483 * Cleanup the open handle that is cached on MDT-side.
2485 * For open case, the client side open handling thread may hit error
2486 * after the MDT grant the open. Under such case, the client should
2487 * send close RPC to the MDT as cleanup; otherwise, the open handle
2488 * on the MDT will be leaked there until the client umount or evicted.
2490 * In further, if someone unlinked the file, because the open handle
2491 * holds the reference on such file/object, then it will block the
2492 * subsequent threads that want to locate such object via FID.
2494 * \param[in] sb super block for this file-system
2495 * \param[in] open_req pointer to the original open request
2497 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2499 struct mdt_body *body;
2500 struct md_op_data *op_data;
2501 struct ptlrpc_request *close_req = NULL;
2502 struct obd_export *exp = ll_s2sbi(sb)->ll_md_exp;
2505 body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2506 OBD_ALLOC_PTR(op_data);
2507 if (op_data == NULL) {
2508 CWARN("%s: cannot allocate op_data to release open handle for "
2509 DFID"\n", ll_s2sbi(sb)->ll_fsname, PFID(&body->mbo_fid1));
2514 op_data->op_fid1 = body->mbo_fid1;
2515 op_data->op_open_handle = body->mbo_open_handle;
2516 op_data->op_mod_time = ktime_get_real_seconds();
2517 md_close(exp, op_data, NULL, &close_req);
2518 ptlrpc_req_finished(close_req);
2519 ll_finish_md_op_data(op_data);
2524 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2525 struct super_block *sb, struct lookup_intent *it)
2527 struct ll_sb_info *sbi = NULL;
2528 struct lustre_md md = { NULL };
2529 bool default_lmv_deleted = false;
2534 LASSERT(*inode || sb);
2535 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2536 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2537 sbi->ll_md_exp, &md);
2542 * clear default_lmv only if intent_getattr reply doesn't contain it.
2543 * but it needs to be done after iget, check this early because
2544 * ll_update_lsm_md() may change md.
2546 if (it && (it->it_op & (IT_LOOKUP | IT_GETATTR)) &&
2547 S_ISDIR(md.body->mbo_mode) && !md.default_lmv)
2548 default_lmv_deleted = true;
2551 rc = ll_update_inode(*inode, &md);
2555 LASSERT(sb != NULL);
2558 * At this point server returns to client's same fid as client
2559 * generated for creating. So using ->fid1 is okay here.
2561 if (!fid_is_sane(&md.body->mbo_fid1)) {
2562 CERROR("%s: Fid is insane "DFID"\n",
2564 PFID(&md.body->mbo_fid1));
2565 GOTO(out, rc = -EINVAL);
2568 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2569 sbi->ll_flags & LL_SBI_32BIT_API),
2571 if (IS_ERR(*inode)) {
2572 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
2574 posix_acl_release(md.posix_acl);
2575 md.posix_acl = NULL;
2578 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2580 CERROR("new_inode -fatal: rc %d\n", rc);
2585 /* Handling piggyback layout lock.
2586 * Layout lock can be piggybacked by getattr and open request.
2587 * The lsm can be applied to inode only if it comes with a layout lock
2588 * otherwise correct layout may be overwritten, for example:
2589 * 1. proc1: mdt returns a lsm but not granting layout
2590 * 2. layout was changed by another client
2591 * 3. proc2: refresh layout and layout lock granted
2592 * 4. proc1: to apply a stale layout */
2593 if (it != NULL && it->it_lock_mode != 0) {
2594 struct lustre_handle lockh;
2595 struct ldlm_lock *lock;
2597 lockh.cookie = it->it_lock_handle;
2598 lock = ldlm_handle2lock(&lockh);
2599 LASSERT(lock != NULL);
2600 if (ldlm_has_layout(lock)) {
2601 struct cl_object_conf conf;
2603 memset(&conf, 0, sizeof(conf));
2604 conf.coc_opc = OBJECT_CONF_SET;
2605 conf.coc_inode = *inode;
2606 conf.coc_lock = lock;
2607 conf.u.coc_layout = md.layout;
2608 (void)ll_layout_conf(*inode, &conf);
2610 LDLM_LOCK_PUT(lock);
2613 if (default_lmv_deleted)
2614 ll_update_default_lsm_md(*inode, &md);
2619 /* cleanup will be done if necessary */
2620 md_free_lustre_md(sbi->ll_md_exp, &md);
2622 if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2623 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2628 int ll_obd_statfs(struct inode *inode, void __user *arg)
2630 struct ll_sb_info *sbi = NULL;
2631 struct obd_export *exp;
2633 struct obd_ioctl_data *data = NULL;
2637 if (!inode || !(sbi = ll_i2sbi(inode)))
2638 GOTO(out_statfs, rc = -EINVAL);
2640 rc = obd_ioctl_getdata(&buf, &len, arg);
2642 GOTO(out_statfs, rc);
2645 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2646 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2647 GOTO(out_statfs, rc = -EINVAL);
2649 if (data->ioc_inllen1 != sizeof(__u32) ||
2650 data->ioc_inllen2 != sizeof(__u32) ||
2651 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2652 data->ioc_plen2 != sizeof(struct obd_uuid))
2653 GOTO(out_statfs, rc = -EINVAL);
2655 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2656 if (type & LL_STATFS_LMV)
2657 exp = sbi->ll_md_exp;
2658 else if (type & LL_STATFS_LOV)
2659 exp = sbi->ll_dt_exp;
2661 GOTO(out_statfs, rc = -ENODEV);
2663 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2665 GOTO(out_statfs, rc);
2667 OBD_FREE_LARGE(buf, len);
2672 * this is normally called in ll_fini_md_op_data(), but sometimes it needs to
2673 * be called early to avoid deadlock.
2675 void ll_unlock_md_op_lsm(struct md_op_data *op_data)
2677 if (op_data->op_mea2_sem) {
2678 up_read(op_data->op_mea2_sem);
2679 op_data->op_mea2_sem = NULL;
2682 if (op_data->op_mea1_sem) {
2683 up_read(op_data->op_mea1_sem);
2684 op_data->op_mea1_sem = NULL;
2688 /* this function prepares md_op_data hint for passing it down to MD stack. */
2689 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2690 struct inode *i1, struct inode *i2,
2691 const char *name, size_t namelen,
2692 __u32 mode, enum md_op_code opc,
2695 LASSERT(i1 != NULL);
2698 /* Do not reuse namelen for something else. */
2700 return ERR_PTR(-EINVAL);
2702 if (namelen > ll_i2sbi(i1)->ll_namelen)
2703 return ERR_PTR(-ENAMETOOLONG);
2705 if (!lu_name_is_valid_2(name, namelen))
2706 return ERR_PTR(-EINVAL);
2709 if (op_data == NULL)
2710 OBD_ALLOC_PTR(op_data);
2712 if (op_data == NULL)
2713 return ERR_PTR(-ENOMEM);
2715 ll_i2gids(op_data->op_suppgids, i1, i2);
2716 op_data->op_fid1 = *ll_inode2fid(i1);
2717 op_data->op_code = opc;
2719 if (S_ISDIR(i1->i_mode)) {
2720 down_read(&ll_i2info(i1)->lli_lsm_sem);
2721 op_data->op_mea1_sem = &ll_i2info(i1)->lli_lsm_sem;
2722 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2723 op_data->op_default_mea1 = ll_i2info(i1)->lli_default_lsm_md;
2727 op_data->op_fid2 = *ll_inode2fid(i2);
2728 if (S_ISDIR(i2->i_mode)) {
2730 down_read(&ll_i2info(i2)->lli_lsm_sem);
2731 op_data->op_mea2_sem =
2732 &ll_i2info(i2)->lli_lsm_sem;
2734 op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2737 fid_zero(&op_data->op_fid2);
2740 if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2741 op_data->op_cli_flags |= CLI_HASH64;
2743 if (ll_need_32bit_api(ll_i2sbi(i1)))
2744 op_data->op_cli_flags |= CLI_API32;
2746 op_data->op_name = name;
2747 op_data->op_namelen = namelen;
2748 op_data->op_mode = mode;
2749 op_data->op_mod_time = ktime_get_real_seconds();
2750 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2751 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2752 op_data->op_cap = cfs_curproc_cap_pack();
2753 op_data->op_mds = 0;
2754 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2755 filename_is_volatile(name, namelen, &op_data->op_mds)) {
2756 op_data->op_bias |= MDS_CREATE_VOLATILE;
2758 op_data->op_data = data;
2763 void ll_finish_md_op_data(struct md_op_data *op_data)
2765 ll_unlock_md_op_lsm(op_data);
2766 security_release_secctx(op_data->op_file_secctx,
2767 op_data->op_file_secctx_size);
2768 OBD_FREE_PTR(op_data);
2771 #ifdef HAVE_SUPEROPS_USE_DENTRY
2772 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2774 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2777 struct ll_sb_info *sbi;
2779 #ifdef HAVE_SUPEROPS_USE_DENTRY
2780 LASSERT((seq != NULL) && (dentry != NULL));
2781 sbi = ll_s2sbi(dentry->d_sb);
2783 LASSERT((seq != NULL) && (vfs != NULL));
2784 sbi = ll_s2sbi(vfs->mnt_sb);
2787 if (sbi->ll_flags & LL_SBI_NOLCK)
2788 seq_puts(seq, ",nolock");
2790 /* "flock" is the default since 2.13, but it wasn't for many years,
2791 * so it is still useful to print this to show it is enabled.
2792 * Start to print "noflock" so it is now clear when flock is disabled.
2794 if (sbi->ll_flags & LL_SBI_FLOCK)
2795 seq_puts(seq, ",flock");
2796 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2797 seq_puts(seq, ",localflock");
2799 seq_puts(seq, ",noflock");
2801 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2802 seq_puts(seq, ",user_xattr");
2804 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2805 seq_puts(seq, ",lazystatfs");
2807 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2808 seq_puts(seq, ",user_fid2path");
2810 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2811 seq_puts(seq, ",always_ping");
2817 * Get obd name by cmd, and copy out to user space
2819 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2821 struct ll_sb_info *sbi = ll_i2sbi(inode);
2822 struct obd_device *obd;
2825 if (cmd == OBD_IOC_GETDTNAME)
2826 obd = class_exp2obd(sbi->ll_dt_exp);
2827 else if (cmd == OBD_IOC_GETMDNAME)
2828 obd = class_exp2obd(sbi->ll_md_exp);
2835 if (copy_to_user((void __user *)arg, obd->obd_name,
2836 strlen(obd->obd_name) + 1))
2842 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2849 p.mnt = current->fs->root.mnt;
2851 path = d_path(&p, buf, bufsize);
2856 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2858 char *buf, *path = NULL;
2859 struct dentry *dentry = NULL;
2860 struct inode *inode = page->mapping->host;
2862 /* this can be called inside spin lock so use GFP_ATOMIC. */
2863 buf = (char *)__get_free_page(GFP_ATOMIC);
2865 dentry = d_find_alias(page->mapping->host);
2867 path = ll_d_path(dentry, buf, PAGE_SIZE);
2870 /* The below message is checked in recovery-small.sh test_24b */
2872 "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2873 "(rc %d)\n", ll_i2sbi(inode)->ll_fsname,
2874 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2875 PFID(ll_inode2fid(inode)),
2876 (path && !IS_ERR(path)) ? path : "", ioret);
2882 free_page((unsigned long)buf);
2885 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2886 struct lov_user_md **kbuf)
2888 struct lov_user_md lum;
2892 if (copy_from_user(&lum, md, sizeof(lum)))
2895 lum_size = ll_lov_user_md_size(&lum);
2899 OBD_ALLOC_LARGE(*kbuf, lum_size);
2903 if (copy_from_user(*kbuf, md, lum_size) != 0) {
2904 OBD_FREE_LARGE(*kbuf, lum_size);
2912 * Compute llite root squash state after a change of root squash
2913 * configuration setting or add/remove of a lnet nid
2915 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2917 struct root_squash_info *squash = &sbi->ll_squash;
2920 struct lnet_process_id id;
2922 /* Update norootsquash flag */
2923 spin_lock(&squash->rsi_lock);
2924 if (list_empty(&squash->rsi_nosquash_nids))
2925 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2927 /* Do not apply root squash as soon as one of our NIDs is
2928 * in the nosquash_nids list */
2931 while (LNetGetId(i++, &id) != -ENOENT) {
2932 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2934 if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2940 sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2942 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2944 spin_unlock(&squash->rsi_lock);
2948 * Parse linkea content to extract information about a given hardlink
2950 * \param[in] ldata - Initialized linkea data
2951 * \param[in] linkno - Link identifier
2952 * \param[out] parent_fid - The entry's parent FID
2953 * \param[out] ln - Entry name destination buffer
2955 * \retval 0 on success
2956 * \retval Appropriate negative error code on failure
2958 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2959 struct lu_fid *parent_fid, struct lu_name *ln)
2965 rc = linkea_init_with_rec(ldata);
2969 if (linkno >= ldata->ld_leh->leh_reccount)
2970 /* beyond last link */
2973 linkea_first_entry(ldata);
2974 for (idx = 0; ldata->ld_lee != NULL; idx++) {
2975 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2980 linkea_next_entry(ldata);
2990 * Get parent FID and name of an identified link. Operation is performed for
2991 * a given link number, letting the caller iterate over linkno to list one or
2992 * all links of an entry.
2994 * \param[in] file - File descriptor against which to perform the operation
2995 * \param[in,out] arg - User-filled structure containing the linkno to operate
2996 * on and the available size. It is eventually filled with
2997 * the requested information or left untouched on error
2999 * \retval - 0 on success
3000 * \retval - Appropriate negative error code on failure
3002 int ll_getparent(struct file *file, struct getparent __user *arg)
3004 struct inode *inode = file_inode(file);
3005 struct linkea_data *ldata;
3006 struct lu_buf buf = LU_BUF_NULL;
3008 struct lu_fid parent_fid;
3015 if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
3016 !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
3019 if (get_user(name_size, &arg->gp_name_size))
3022 if (get_user(linkno, &arg->gp_linkno))
3025 if (name_size > PATH_MAX)
3028 OBD_ALLOC(ldata, sizeof(*ldata));
3032 rc = linkea_data_new(ldata, &buf);
3034 GOTO(ldata_free, rc);
3036 rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
3037 buf.lb_len, OBD_MD_FLXATTR);
3041 rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
3045 if (ln.ln_namelen >= name_size)
3046 GOTO(lb_free, rc = -EOVERFLOW);
3048 if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
3049 GOTO(lb_free, rc = -EFAULT);
3051 if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
3052 GOTO(lb_free, rc = -EFAULT);
3054 if (put_user('\0', arg->gp_name + ln.ln_namelen))
3055 GOTO(lb_free, rc = -EFAULT);
3060 OBD_FREE(ldata, sizeof(*ldata));