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/statfs.h>
41 #include <linux/time.h>
42 #include <linux/types.h>
43 #include <linux/version.h>
45 #include <linux/user_namespace.h>
46 #ifdef HAVE_UIDGID_HEADER
47 # include <linux/uidgid.h>
49 #include <linux/security.h>
51 #include <uapi/linux/lustre/lustre_ioctl.h>
52 #include <lustre_ha.h>
53 #include <lustre_dlm.h>
54 #include <lprocfs_status.h>
55 #include <lustre_disk.h>
56 #include <uapi/linux/lustre/lustre_param.h>
57 #include <lustre_log.h>
58 #include <cl_object.h>
59 #include <obd_cksum.h>
60 #include "llite_internal.h"
62 struct kmem_cache *ll_file_data_slab;
65 #define log2(n) ffz(~(n))
68 static struct ll_sb_info *ll_init_sbi(void)
70 struct ll_sb_info *sbi = NULL;
72 unsigned long lru_page_max;
82 spin_lock_init(&sbi->ll_lock);
83 mutex_init(&sbi->ll_lco.lco_lock);
84 spin_lock_init(&sbi->ll_pp_extent_lock);
85 spin_lock_init(&sbi->ll_process_lock);
86 sbi->ll_rw_stats_on = 0;
89 pages = si.totalram - si.totalhigh;
90 lru_page_max = pages / 2;
92 /* initialize ll_cache data */
93 sbi->ll_cache = cl_cache_init(lru_page_max);
94 if (sbi->ll_cache == NULL) {
95 OBD_FREE(sbi, sizeof(*sbi));
99 sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
100 SBI_DEFAULT_READAHEAD_MAX);
101 sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
102 sbi->ll_ra_info.ra_max_read_ahead_whole_pages = -1;
104 ll_generate_random_uuid(uuid);
105 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
106 CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
108 sbi->ll_flags |= LL_SBI_VERBOSE;
109 #ifdef ENABLE_CHECKSUM
110 sbi->ll_flags |= LL_SBI_CHECKSUM;
113 #ifdef HAVE_LRU_RESIZE_SUPPORT
114 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
116 sbi->ll_flags |= LL_SBI_LAZYSTATFS;
118 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
119 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
121 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
125 /* metadata statahead is enabled by default */
126 sbi->ll_sa_running_max = LL_SA_RUNNING_DEF;
127 sbi->ll_sa_max = LL_SA_RPC_DEF;
128 atomic_set(&sbi->ll_sa_total, 0);
129 atomic_set(&sbi->ll_sa_wrong, 0);
130 atomic_set(&sbi->ll_sa_running, 0);
131 atomic_set(&sbi->ll_agl_total, 0);
132 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
133 sbi->ll_flags |= LL_SBI_FAST_READ;
134 sbi->ll_flags |= LL_SBI_TINY_WRITE;
137 sbi->ll_squash.rsi_uid = 0;
138 sbi->ll_squash.rsi_gid = 0;
139 INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
140 init_rwsem(&sbi->ll_squash.rsi_sem);
145 static void ll_free_sbi(struct super_block *sb)
147 struct ll_sb_info *sbi = ll_s2sbi(sb);
151 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
152 cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
153 if (sbi->ll_cache != NULL) {
154 cl_cache_decref(sbi->ll_cache);
155 sbi->ll_cache = NULL;
157 OBD_FREE(sbi, sizeof(*sbi));
162 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
163 struct vfsmount *mnt)
165 struct inode *root = NULL;
166 struct ll_sb_info *sbi = ll_s2sbi(sb);
167 struct obd_statfs *osfs = NULL;
168 struct ptlrpc_request *request = NULL;
169 struct obd_connect_data *data = NULL;
170 struct obd_uuid *uuid;
171 struct md_op_data *op_data;
172 struct lustre_md lmd;
174 int size, err, checksum;
177 sbi->ll_md_obd = class_name2obd(md);
178 if (!sbi->ll_md_obd) {
179 CERROR("MD %s: not setup or attached\n", md);
193 /* pass client page size via ocd_grant_blkbits, the server should report
194 * back its backend blocksize for grant calculation purpose */
195 data->ocd_grant_blkbits = PAGE_SHIFT;
197 /* indicate MDT features supported by this client */
198 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
199 OBD_CONNECT_ATTRFID | OBD_CONNECT_GRANT |
200 OBD_CONNECT_VERSION | OBD_CONNECT_BRW_SIZE |
201 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
202 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
203 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
204 OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 |
205 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
206 OBD_CONNECT_64BITHASH |
207 OBD_CONNECT_EINPROGRESS |
208 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
209 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS|
210 OBD_CONNECT_MAX_EASIZE |
211 OBD_CONNECT_FLOCK_DEAD |
212 OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
213 OBD_CONNECT_OPEN_BY_FID |
214 OBD_CONNECT_DIR_STRIPE |
215 OBD_CONNECT_BULK_MBITS | OBD_CONNECT_CKSUM |
216 OBD_CONNECT_SUBTREE |
217 OBD_CONNECT_MULTIMODRPCS |
218 OBD_CONNECT_GRANT_PARAM |
219 OBD_CONNECT_SHORTIO | OBD_CONNECT_FLAGS2;
221 data->ocd_connect_flags2 = OBD_CONNECT2_FLR |
222 OBD_CONNECT2_LOCK_CONVERT |
223 OBD_CONNECT2_DIR_MIGRATE |
224 OBD_CONNECT2_SUM_STATFS |
225 OBD_CONNECT2_ARCHIVE_ID_ARRAY;
227 #ifdef HAVE_LRU_RESIZE_SUPPORT
228 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
229 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
231 #ifdef CONFIG_FS_POSIX_ACL
232 data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK |
233 OBD_CONNECT_LARGE_ACL;
236 data->ocd_cksum_types = obd_cksum_types_supported_client();
238 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
239 /* flag mdc connection as lightweight, only used for test
240 * purpose, use with care */
241 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
243 data->ocd_ibits_known = MDS_INODELOCK_FULL;
244 data->ocd_version = LUSTRE_VERSION_CODE;
246 if (sb->s_flags & MS_RDONLY)
247 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
248 if (sbi->ll_flags & LL_SBI_USER_XATTR)
249 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
252 /* Setting this indicates we correctly support S_NOSEC (See kernel
253 * commit 9e1f1de02c2275d7172e18dc4e7c2065777611bf)
255 sb->s_flags |= MS_NOSEC;
258 if (sbi->ll_flags & LL_SBI_FLOCK)
259 sbi->ll_fop = &ll_file_operations_flock;
260 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
261 sbi->ll_fop = &ll_file_operations;
263 sbi->ll_fop = &ll_file_operations_noflock;
265 /* always ping even if server suppress_pings */
266 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
267 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
269 obd_connect_set_secctx(data);
271 #if defined(CONFIG_SECURITY)
272 data->ocd_connect_flags2 |= OBD_CONNECT2_SELINUX_POLICY;
275 data->ocd_brw_size = MD_MAX_BRW_SIZE;
277 err = obd_connect(NULL, &sbi->ll_md_exp, sbi->ll_md_obd,
278 &sbi->ll_sb_uuid, data, NULL);
280 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
281 "recovery, of which this client is not a "
282 "part. Please wait for recovery to complete,"
283 " abort, or time out.\n", md);
286 CERROR("cannot connect to %s: rc = %d\n", md, err);
290 sbi->ll_md_exp->exp_connect_data = *data;
292 err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
293 LUSTRE_SEQ_METADATA);
295 CERROR("%s: Can't init metadata layer FID infrastructure, "
296 "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
300 /* For mount, we only need fs info from MDT0, and also in DNE, it
301 * can make sure the client can be mounted as long as MDT0 is
303 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
304 ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
305 OBD_STATFS_FOR_MDT0);
307 GOTO(out_md_fid, err);
309 /* This needs to be after statfs to ensure connect has finished.
310 * Note that "data" does NOT contain the valid connect reply.
311 * If connecting to a 1.8 server there will be no LMV device, so
312 * we can access the MDC export directly and exp_connect_flags will
313 * be non-zero, but if accessing an upgraded 2.1 server it will
314 * have the correct flags filled in.
315 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
316 valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
317 if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
318 valid != CLIENT_CONNECT_MDT_REQD) {
321 OBD_ALLOC_WAIT(buf, PAGE_SIZE);
322 obd_connect_flags2str(buf, PAGE_SIZE,
323 valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
324 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
325 "feature(s) needed for correct operation "
326 "of this client (%s). Please upgrade "
327 "server or downgrade client.\n",
328 sbi->ll_md_exp->exp_obd->obd_name, buf);
329 OBD_FREE(buf, PAGE_SIZE);
330 GOTO(out_md_fid, err = -EPROTO);
333 size = sizeof(*data);
334 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
335 KEY_CONN_DATA, &size, data);
337 CERROR("%s: Get connect data failed: rc = %d\n",
338 sbi->ll_md_exp->exp_obd->obd_name, err);
339 GOTO(out_md_fid, err);
342 LASSERT(osfs->os_bsize);
343 sb->s_blocksize = osfs->os_bsize;
344 sb->s_blocksize_bits = log2(osfs->os_bsize);
345 sb->s_magic = LL_SUPER_MAGIC;
346 sb->s_maxbytes = MAX_LFS_FILESIZE;
347 sbi->ll_namelen = osfs->os_namelen;
348 sbi->ll_mnt.mnt = current->fs->root.mnt;
350 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
351 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
352 LCONSOLE_INFO("Disabling user_xattr feature because "
353 "it is not supported on the server\n");
354 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
357 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
359 sb->s_flags |= MS_POSIXACL;
361 sbi->ll_flags |= LL_SBI_ACL;
363 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
365 sb->s_flags &= ~MS_POSIXACL;
367 sbi->ll_flags &= ~LL_SBI_ACL;
370 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
371 sbi->ll_flags |= LL_SBI_64BIT_HASH;
373 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
374 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
376 if (obd_connect_has_secctx(data))
377 sbi->ll_flags |= LL_SBI_FILE_SECCTX;
379 if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
380 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
381 LCONSOLE_INFO("%s: disabling xattr cache due to "
382 "unknown maximum xattr size.\n", dt);
383 } else if (!sbi->ll_xattr_cache_set) {
384 /* If xattr_cache is already set (no matter 0 or 1)
385 * during processing llog, it won't be enabled here. */
386 sbi->ll_flags |= LL_SBI_XATTR_CACHE;
387 sbi->ll_xattr_cache_enabled = 1;
391 sbi->ll_dt_obd = class_name2obd(dt);
392 if (!sbi->ll_dt_obd) {
393 CERROR("DT %s: not setup or attached\n", dt);
394 GOTO(out_md_fid, err = -ENODEV);
397 /* pass client page size via ocd_grant_blkbits, the server should report
398 * back its backend blocksize for grant calculation purpose */
399 data->ocd_grant_blkbits = PAGE_SHIFT;
401 /* indicate OST features supported by this client */
402 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
403 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
404 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
405 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
406 OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
407 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
408 OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
409 OBD_CONNECT_EINPROGRESS |
410 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
411 OBD_CONNECT_LAYOUTLOCK |
412 OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
413 OBD_CONNECT_BULK_MBITS | OBD_CONNECT_SHORTIO |
414 OBD_CONNECT_FLAGS2 | OBD_CONNECT_GRANT_SHRINK;
416 /* The client currently advertises support for OBD_CONNECT_LOCKAHEAD_OLD so it
417 * can interoperate with an older version of lockahead which was released prior
418 * to landing in master. This support will be dropped when 2.13 development
419 * starts. At the point, we should not just drop the connect flag (below), we
420 * should also remove the support in the code.
422 * Removing it means a few things:
423 * 1. Remove this section here
424 * 2. Remove CEF_NONBLOCK in ll_file_lockahead()
425 * 3. Remove function exp_connect_lockahead_old
426 * 4. Remove LDLM_FL_LOCKAHEAD_OLD_RESERVED in lustre_dlm_flags.h
428 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 12, 50, 0)
429 data->ocd_connect_flags |= OBD_CONNECT_LOCKAHEAD_OLD;
432 data->ocd_connect_flags2 = OBD_CONNECT2_LOCKAHEAD;
434 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
435 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
437 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
438 * disabled by default, because it can still be enabled on the
439 * fly via /sys. As a consequence, we still need to come to an
440 * agreement on the supported algorithms at connect time
442 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
444 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
445 data->ocd_cksum_types = OBD_CKSUM_ADLER;
447 data->ocd_cksum_types = obd_cksum_types_supported_client();
449 #ifdef HAVE_LRU_RESIZE_SUPPORT
450 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
452 /* always ping even if server suppress_pings */
453 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
454 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
456 CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
457 "ocd_grant: %d\n", data->ocd_connect_flags,
458 data->ocd_version, data->ocd_grant);
460 sbi->ll_dt_obd->obd_upcall.onu_owner = &sbi->ll_lco;
461 sbi->ll_dt_obd->obd_upcall.onu_upcall = cl_ocd_update;
463 data->ocd_brw_size = DT_MAX_BRW_SIZE;
465 err = obd_connect(NULL, &sbi->ll_dt_exp, sbi->ll_dt_obd,
466 &sbi->ll_sb_uuid, data, NULL);
468 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
469 "recovery, of which this client is not a "
470 "part. Please wait for recovery to "
471 "complete, abort, or time out.\n", dt);
474 CERROR("%s: Cannot connect to %s: rc = %d\n",
475 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
479 sbi->ll_dt_exp->exp_connect_data = *data;
481 /* Don't change value if it was specified in the config log */
482 if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages == -1)
483 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
484 max_t(unsigned long, SBI_DEFAULT_READAHEAD_WHOLE_MAX,
485 (data->ocd_brw_size >> PAGE_SHIFT));
487 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
488 LUSTRE_SEQ_METADATA);
490 CERROR("%s: Can't init data layer FID infrastructure, "
491 "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
495 mutex_lock(&sbi->ll_lco.lco_lock);
496 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
497 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
498 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
499 mutex_unlock(&sbi->ll_lco.lco_lock);
501 fid_zero(&sbi->ll_root_fid);
502 err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
505 CERROR("cannot mds_connect: rc = %d\n", err);
506 GOTO(out_lock_cn_cb, err);
508 if (!fid_is_sane(&sbi->ll_root_fid)) {
509 CERROR("%s: Invalid root fid "DFID" during mount\n",
510 sbi->ll_md_exp->exp_obd->obd_name,
511 PFID(&sbi->ll_root_fid));
512 GOTO(out_lock_cn_cb, err = -EINVAL);
514 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
516 sb->s_op = &lustre_super_operations;
517 #ifdef HAVE_XATTR_HANDLER_FLAGS
518 sb->s_xattr = ll_xattr_handlers;
520 #if THREAD_SIZE >= 8192 /*b=17630*/
521 sb->s_export_op = &lustre_export_operations;
525 * XXX: move this to after cbd setup? */
526 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
527 if (sbi->ll_flags & LL_SBI_ACL)
528 valid |= OBD_MD_FLACL;
530 OBD_ALLOC_PTR(op_data);
532 GOTO(out_lock_cn_cb, err = -ENOMEM);
534 op_data->op_fid1 = sbi->ll_root_fid;
535 op_data->op_mode = 0;
536 op_data->op_valid = valid;
538 err = md_getattr(sbi->ll_md_exp, op_data, &request);
540 OBD_FREE_PTR(op_data);
542 CERROR("%s: md_getattr failed for root: rc = %d\n",
543 sbi->ll_md_exp->exp_obd->obd_name, err);
544 GOTO(out_lock_cn_cb, err);
547 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
548 sbi->ll_md_exp, &lmd);
550 CERROR("failed to understand root inode md: rc = %d\n", err);
551 ptlrpc_req_finished(request);
552 GOTO(out_lock_cn_cb, err);
555 LASSERT(fid_is_sane(&sbi->ll_root_fid));
556 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
557 sbi->ll_flags & LL_SBI_32BIT_API),
559 md_free_lustre_md(sbi->ll_md_exp, &lmd);
560 ptlrpc_req_finished(request);
563 #ifdef CONFIG_FS_POSIX_ACL
565 posix_acl_release(lmd.posix_acl);
566 lmd.posix_acl = NULL;
569 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
571 CERROR("lustre_lite: bad iget4 for root\n");
575 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
576 if (sbi->ll_checksum_set) {
577 err = obd_set_info_async(NULL, sbi->ll_dt_exp,
578 sizeof(KEY_CHECKSUM), KEY_CHECKSUM,
579 sizeof(checksum), &checksum, NULL);
581 CERROR("%s: Set checksum failed: rc = %d\n",
582 sbi->ll_dt_exp->exp_obd->obd_name, err);
588 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
589 KEY_CACHE_SET, sizeof(*sbi->ll_cache),
590 sbi->ll_cache, NULL);
592 CERROR("%s: Set cache_set failed: rc = %d\n",
593 sbi->ll_dt_exp->exp_obd->obd_name, err);
597 sb->s_root = d_make_root(root);
598 if (sb->s_root == NULL) {
599 CERROR("%s: can't make root dentry\n",
600 ll_get_fsname(sb, NULL, 0));
601 GOTO(out_root, err = -ENOMEM);
603 #ifdef HAVE_DCACHE_LOCK
604 sb->s_root->d_op = &ll_d_ops;
607 sbi->ll_sdev_orig = sb->s_dev;
609 /* We set sb->s_dev equal on all lustre clients in order to support
610 * NFS export clustering. NFSD requires that the FSID be the same
612 /* s_dev is also used in lt_compare() to compare two fs, but that is
613 * only a node-local comparison. */
614 uuid = obd_get_uuid(sbi->ll_md_exp);
616 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
623 if (sbi->ll_dt_obd) {
624 err = sysfs_create_link(&sbi->ll_kset.kobj,
625 &sbi->ll_dt_obd->obd_kset.kobj,
626 sbi->ll_dt_obd->obd_type->typ_name);
628 CERROR("%s: could not register %s in llite: rc = %d\n",
629 dt, ll_get_fsname(sb, NULL, 0), err);
634 if (sbi->ll_md_obd) {
635 err = sysfs_create_link(&sbi->ll_kset.kobj,
636 &sbi->ll_md_obd->obd_kset.kobj,
637 sbi->ll_md_obd->obd_type->typ_name);
639 CERROR("%s: could not register %s in llite: rc = %d\n",
640 md, ll_get_fsname(sb, NULL, 0), err);
650 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
652 obd_disconnect(sbi->ll_dt_exp);
653 sbi->ll_dt_exp = NULL;
654 sbi->ll_dt_obd = NULL;
656 obd_fid_fini(sbi->ll_md_exp->exp_obd);
658 obd_disconnect(sbi->ll_md_exp);
659 sbi->ll_md_exp = NULL;
660 sbi->ll_md_obd = NULL;
669 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
673 size = sizeof(*lmmsize);
674 rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
675 KEY_MAX_EASIZE, &size, lmmsize);
677 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
678 sbi->ll_dt_exp->exp_obd->obd_name, rc);
683 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
684 KEY_MAX_EASIZE, &size, lmmsize);
686 CERROR("Get max mdsize error rc %d\n", rc);
692 * Get the value of the default_easize parameter.
694 * \see client_obd::cl_default_mds_easize
696 * \param[in] sbi superblock info for this filesystem
697 * \param[out] lmmsize pointer to storage location for value
699 * \retval 0 on success
700 * \retval negative negated errno on failure
702 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
707 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
708 KEY_DEFAULT_EASIZE, &size, lmmsize);
710 CERROR("Get default mdsize error rc %d\n", rc);
716 * Set the default_easize parameter to the given value.
718 * \see client_obd::cl_default_mds_easize
720 * \param[in] sbi superblock info for this filesystem
721 * \param[in] lmmsize the size to set
723 * \retval 0 on success
724 * \retval negative negated errno on failure
726 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
730 if (lmmsize < sizeof(struct lov_mds_md) ||
731 lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
734 rc = obd_set_info_async(NULL, sbi->ll_md_exp,
735 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
736 sizeof(int), &lmmsize, NULL);
741 static void client_common_put_super(struct super_block *sb)
743 struct ll_sb_info *sbi = ll_s2sbi(sb);
748 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
749 obd_disconnect(sbi->ll_dt_exp);
750 sbi->ll_dt_exp = NULL;
752 ll_debugfs_unregister_super(sb);
754 obd_fid_fini(sbi->ll_md_exp->exp_obd);
755 obd_disconnect(sbi->ll_md_exp);
756 sbi->ll_md_exp = NULL;
761 void ll_kill_super(struct super_block *sb)
763 struct ll_sb_info *sbi;
767 if (!(sb->s_flags & MS_ACTIVE))
771 /* we need restore s_dev from changed for clustred NFS before put_super
772 * because new kernels have cached s_dev and change sb->s_dev in
773 * put_super not affected real removing devices */
775 sb->s_dev = sbi->ll_sdev_orig;
776 sbi->ll_umounting = 1;
778 /* wait running statahead threads to quit */
779 while (atomic_read(&sbi->ll_sa_running) > 0) {
780 set_current_state(TASK_UNINTERRUPTIBLE);
781 schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
788 static inline int ll_set_opt(const char *opt, char *data, int fl)
790 if (strncmp(opt, data, strlen(opt)) != 0)
796 /* non-client-specific mount options are parsed in lmd_parse */
797 static int ll_options(char *options, struct ll_sb_info *sbi)
800 char *s1 = options, *s2;
801 int *flags = &sbi->ll_flags;
807 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
810 CDEBUG(D_SUPER, "next opt=%s\n", s1);
811 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
816 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
821 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
826 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
831 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
836 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
841 tmp = ll_set_opt("context", s1, 1);
844 tmp = ll_set_opt("fscontext", s1, 1);
847 tmp = ll_set_opt("defcontext", s1, 1);
850 tmp = ll_set_opt("rootcontext", s1, 1);
853 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
858 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
864 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
867 sbi->ll_checksum_set = 1;
870 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
873 sbi->ll_checksum_set = 1;
876 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
881 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
886 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
891 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
896 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
901 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
906 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
911 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
916 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
922 s2 = strchr(s1, ',');
930 void ll_lli_init(struct ll_inode_info *lli)
932 lli->lli_inode_magic = LLI_INODE_MAGIC;
934 spin_lock_init(&lli->lli_lock);
935 lli->lli_posix_acl = NULL;
936 /* Do not set lli_fid, it has been initialized already. */
937 fid_zero(&lli->lli_pfid);
938 lli->lli_mds_read_och = NULL;
939 lli->lli_mds_write_och = NULL;
940 lli->lli_mds_exec_och = NULL;
941 lli->lli_open_fd_read_count = 0;
942 lli->lli_open_fd_write_count = 0;
943 lli->lli_open_fd_exec_count = 0;
944 mutex_init(&lli->lli_och_mutex);
945 spin_lock_init(&lli->lli_agl_lock);
946 spin_lock_init(&lli->lli_layout_lock);
947 ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
948 lli->lli_clob = NULL;
950 init_rwsem(&lli->lli_xattrs_list_rwsem);
951 mutex_init(&lli->lli_xattrs_enq_lock);
953 LASSERT(lli->lli_vfs_inode.i_mode != 0);
954 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
955 mutex_init(&lli->lli_readdir_mutex);
956 lli->lli_opendir_key = NULL;
958 spin_lock_init(&lli->lli_sa_lock);
959 lli->lli_opendir_pid = 0;
960 lli->lli_sa_enabled = 0;
961 lli->lli_def_stripe_offset = -1;
962 init_rwsem(&lli->lli_lsm_sem);
964 mutex_init(&lli->lli_size_mutex);
965 lli->lli_symlink_name = NULL;
966 init_rwsem(&lli->lli_trunc_sem);
967 range_lock_tree_init(&lli->lli_write_tree);
968 init_rwsem(&lli->lli_glimpse_sem);
969 lli->lli_glimpse_time = ktime_set(0, 0);
970 INIT_LIST_HEAD(&lli->lli_agl_list);
971 lli->lli_agl_index = 0;
972 lli->lli_async_rc = 0;
974 mutex_init(&lli->lli_layout_mutex);
975 memset(lli->lli_jobid, 0, sizeof(lli->lli_jobid));
978 #define MAX_STRING_SIZE 128
980 #ifndef HAVE_SUPER_SETUP_BDI_NAME
982 #define LSI_BDI_INITIALIZED 0x00400000
984 #ifndef HAVE_BDI_CAP_MAP_COPY
985 # define BDI_CAP_MAP_COPY 0
988 static int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
990 struct lustre_sb_info *lsi = s2lsi(sb);
991 char buf[MAX_STRING_SIZE];
995 err = bdi_init(&lsi->lsi_bdi);
999 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1000 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1001 lsi->lsi_bdi.name = "lustre";
1002 va_start(args, fmt);
1003 vsnprintf(buf, MAX_STRING_SIZE, fmt, args);
1005 err = bdi_register(&lsi->lsi_bdi, NULL, "%s", buf);
1008 sb->s_bdi = &lsi->lsi_bdi;
1012 #endif /* !HAVE_SUPER_SETUP_BDI_NAME */
1014 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
1016 struct lustre_profile *lprof = NULL;
1017 struct lustre_sb_info *lsi = s2lsi(sb);
1018 struct ll_sb_info *sbi = NULL;
1019 char *dt = NULL, *md = NULL;
1020 char *profilenm = get_profile_name(sb);
1021 struct config_llog_instance *cfg;
1022 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1023 const int instlen = 16 + 2;
1024 unsigned long cfg_instance = ll_get_cfg_instance(sb);
1025 char name[MAX_STRING_SIZE];
1033 /* for ASLR, to map between cfg_instance and hashed ptr */
1034 CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1035 profilenm, cfg_instance, sb);
1037 try_module_get(THIS_MODULE);
1041 GOTO(out_free_cfg, err = -ENOMEM);
1043 /* client additional sb info */
1044 lsi->lsi_llsbi = sbi = ll_init_sbi();
1046 GOTO(out_free_cfg, err = -ENOMEM);
1048 err = ll_options(lsi->lsi_lmd->lmd_opts, sbi);
1050 GOTO(out_free_cfg, err);
1052 err = super_setup_bdi_name(sb, "lustre-%016lx", cfg_instance);
1054 GOTO(out_free_cfg, err);
1056 #ifndef HAVE_DCACHE_LOCK
1057 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1058 sb->s_d_op = &ll_d_ops;
1061 len = strlen(profilenm);
1062 ptr = strrchr(profilenm, '-');
1063 if (ptr && (strcmp(ptr, "-client") == 0))
1067 snprintf(name, MAX_STRING_SIZE, "%.*s-%016lx", len,
1068 profilenm, cfg_instance);
1070 /* Call ll_debugfs_register_super() before lustre_process_log()
1071 * so that "llite.*.*" params can be processed correctly.
1073 err = ll_debugfs_register_super(sb, name);
1075 CERROR("%s: could not register mountpoint in llite: rc = %d\n",
1076 ll_get_fsname(sb, NULL, 0), err);
1080 /* The cfg_instance is a value unique to this super, in case some
1081 * joker tries to mount the same fs at two mount points.
1083 cfg->cfg_instance = cfg_instance;
1084 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1085 cfg->cfg_callback = class_config_llog_handler;
1086 cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1087 /* set up client obds */
1088 err = lustre_process_log(sb, profilenm, cfg);
1090 GOTO(out_debugfs, err);
1092 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1093 lprof = class_get_profile(profilenm);
1094 if (lprof == NULL) {
1095 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1096 " read from the MGS. Does that filesystem "
1097 "exist?\n", profilenm);
1098 GOTO(out_debugfs, err = -EINVAL);
1100 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1101 lprof->lp_md, lprof->lp_dt);
1103 dt_len = strlen(lprof->lp_dt) + instlen + 2;
1104 OBD_ALLOC(dt, dt_len);
1106 GOTO(out_profile, err = -ENOMEM);
1107 snprintf(dt, dt_len - 1, "%s-%016lx", lprof->lp_dt, cfg_instance);
1109 md_len = strlen(lprof->lp_md) + instlen + 2;
1110 OBD_ALLOC(md, md_len);
1112 GOTO(out_free_dt, err = -ENOMEM);
1113 snprintf(md, md_len - 1, "%s-%016lx", lprof->lp_md, cfg_instance);
1115 /* connections, registrations, sb setup */
1116 err = client_common_fill_super(sb, md, dt, mnt);
1118 GOTO(out_free_md, err);
1120 sbi->ll_client_common_fill_super_succeeded = 1;
1124 OBD_FREE(md, md_len);
1127 OBD_FREE(dt, dt_len);
1130 class_put_profile(lprof);
1133 ll_debugfs_unregister_super(sb);
1140 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1141 LCONSOLE_WARN("Mounted %s\n", profilenm);
1143 } /* ll_fill_super */
1145 void ll_put_super(struct super_block *sb)
1147 struct config_llog_instance cfg, params_cfg;
1148 struct obd_device *obd;
1149 struct lustre_sb_info *lsi = s2lsi(sb);
1150 struct ll_sb_info *sbi = ll_s2sbi(sb);
1151 char *profilenm = get_profile_name(sb);
1152 unsigned long cfg_instance = ll_get_cfg_instance(sb);
1154 int next, force = 1, rc = 0;
1158 GOTO(out_no_sbi, 0);
1160 /* Should replace instance_id with something better for ASLR */
1161 CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1162 profilenm, cfg_instance, sb);
1164 cfg.cfg_instance = cfg_instance;
1165 lustre_end_log(sb, profilenm, &cfg);
1167 params_cfg.cfg_instance = cfg_instance;
1168 lustre_end_log(sb, PARAMS_FILENAME, ¶ms_cfg);
1170 if (sbi->ll_md_exp) {
1171 obd = class_exp2obd(sbi->ll_md_exp);
1173 force = obd->obd_force;
1176 /* Wait for unstable pages to be committed to stable storage */
1178 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1179 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1180 atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1184 ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1185 if (force == 0 && rc != -EINTR)
1186 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1188 /* We need to set force before the lov_disconnect in
1189 lustre_common_put_super, since l_d cleans up osc's as well. */
1192 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1194 obd->obd_force = force;
1198 if (sbi->ll_client_common_fill_super_succeeded) {
1199 /* Only if client_common_fill_super succeeded */
1200 client_common_put_super(sb);
1204 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1205 class_manual_cleanup(obd);
1208 if (sbi->ll_flags & LL_SBI_VERBOSE)
1209 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1212 class_del_profile(profilenm);
1214 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1215 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1216 bdi_destroy(&lsi->lsi_bdi);
1217 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1222 lsi->lsi_llsbi = NULL;
1224 lustre_common_put_super(sb);
1226 cl_env_cache_purge(~0);
1228 module_put(THIS_MODULE);
1231 } /* client_put_super */
1233 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1235 struct inode *inode = NULL;
1237 /* NOTE: we depend on atomic igrab() -bzzz */
1238 lock_res_and_lock(lock);
1239 if (lock->l_resource->lr_lvb_inode) {
1240 struct ll_inode_info * lli;
1241 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1242 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1243 inode = igrab(lock->l_resource->lr_lvb_inode);
1245 inode = lock->l_resource->lr_lvb_inode;
1246 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1247 D_WARNING, lock, "lr_lvb_inode %p is "
1248 "bogus: magic %08x",
1249 lock->l_resource->lr_lvb_inode,
1250 lli->lli_inode_magic);
1254 unlock_res_and_lock(lock);
1258 void ll_dir_clear_lsm_md(struct inode *inode)
1260 struct ll_inode_info *lli = ll_i2info(inode);
1262 LASSERT(S_ISDIR(inode->i_mode));
1264 if (lli->lli_lsm_md != NULL) {
1265 lmv_free_memmd(lli->lli_lsm_md);
1266 lli->lli_lsm_md = NULL;
1270 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1271 const struct lu_fid *fid,
1272 struct lustre_md *md)
1274 struct ll_sb_info *sbi = ll_s2sbi(sb);
1275 struct mdt_body *body = md->body;
1276 struct inode *inode;
1280 ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1281 inode = iget_locked(sb, ino);
1282 if (inode == NULL) {
1283 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1284 ll_get_fsname(sb, NULL, 0), PFID(fid));
1285 RETURN(ERR_PTR(-ENOENT));
1288 if (inode->i_state & I_NEW) {
1289 struct ll_inode_info *lli = ll_i2info(inode);
1290 struct lmv_stripe_md *lsm = md->lmv;
1292 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1293 (body->mbo_mode & S_IFMT);
1294 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1297 LTIME_S(inode->i_mtime) = 0;
1298 LTIME_S(inode->i_atime) = 0;
1299 LTIME_S(inode->i_ctime) = 0;
1302 #ifdef HAVE_BACKING_DEV_INFO
1303 /* initializing backing dev info. */
1304 inode->i_mapping->backing_dev_info =
1305 &s2lsi(inode->i_sb)->lsi_bdi;
1307 inode->i_op = &ll_dir_inode_operations;
1308 inode->i_fop = &ll_dir_operations;
1309 lli->lli_fid = *fid;
1312 LASSERT(lsm != NULL);
1313 /* master object FID */
1314 lli->lli_pfid = body->mbo_fid1;
1315 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1316 lli, PFID(fid), PFID(&lli->lli_pfid));
1317 unlock_new_inode(inode);
1323 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1326 struct lmv_stripe_md *lsm = md->lmv;
1327 struct ll_inode_info *lli = ll_i2info(inode);
1330 LASSERT(lsm != NULL);
1332 CDEBUG(D_INODE, "%s: "DFID" set dir layout:\n",
1333 ll_get_fsname(inode->i_sb, NULL, 0),
1334 PFID(&lli->lli_fid));
1335 lsm_md_dump(D_INODE, lsm);
1337 /* XXX sigh, this lsm_root initialization should be in
1338 * LMV layer, but it needs ll_iget right now, so we
1339 * put this here right now. */
1340 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1341 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1342 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1343 /* Unfortunately ll_iget will call ll_update_inode,
1344 * where the initialization of slave inode is slightly
1345 * different, so it reset lsm_md to NULL to avoid
1346 * initializing lsm for slave inode. */
1347 lsm->lsm_md_oinfo[i].lmo_root =
1348 ll_iget_anon_dir(inode->i_sb, fid, md);
1349 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1350 int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1352 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1354 iput(lsm->lsm_md_oinfo[i].lmo_root);
1355 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1361 lli->lli_lsm_md = lsm;
1366 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1368 struct ll_inode_info *lli = ll_i2info(inode);
1369 struct lmv_stripe_md *lsm = md->lmv;
1374 LASSERT(S_ISDIR(inode->i_mode));
1375 CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1376 PFID(ll_inode2fid(inode)));
1379 * no striped information from request, lustre_md from req does not
1380 * include stripeEA, see ll_md_setattr()
1386 * normally dir layout doesn't change, only take read lock to check
1387 * that to avoid blocking other MD operations.
1389 if (lli->lli_lsm_md)
1390 down_read(&lli->lli_lsm_sem);
1392 down_write(&lli->lli_lsm_sem);
1395 * if dir layout mismatch, check whether version is increased, which
1396 * means layout is changed, this happens in dir migration and lfsck.
1398 if (lli->lli_lsm_md && !lsm_md_eq(lli->lli_lsm_md, lsm)) {
1399 if (lsm->lsm_md_layout_version <=
1400 lli->lli_lsm_md->lsm_md_layout_version) {
1401 CERROR("%s: "DFID" dir layout mismatch:\n",
1402 ll_get_fsname(inode->i_sb, NULL, 0),
1403 PFID(&lli->lli_fid));
1404 lsm_md_dump(D_ERROR, lli->lli_lsm_md);
1405 lsm_md_dump(D_ERROR, lsm);
1406 GOTO(unlock, rc = -EINVAL);
1409 /* layout changed, switch to write lock */
1410 up_read(&lli->lli_lsm_sem);
1411 down_write(&lli->lli_lsm_sem);
1412 ll_dir_clear_lsm_md(inode);
1415 /* set directory layout */
1416 if (!lli->lli_lsm_md) {
1417 struct cl_attr *attr;
1419 rc = ll_init_lsm_md(inode, md);
1420 up_write(&lli->lli_lsm_sem);
1424 /* set md->lmv to NULL, so the following free lustre_md
1425 * will not free this lsm */
1429 * md_merge_attr() may take long, since lsm is already set,
1430 * switch to read lock.
1432 down_read(&lli->lli_lsm_sem);
1434 OBD_ALLOC_PTR(attr);
1436 GOTO(unlock, rc = -ENOMEM);
1438 /* validate the lsm */
1439 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1440 ll_md_blocking_ast);
1446 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1447 md->body->mbo_nlink = attr->cat_nlink;
1448 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1449 md->body->mbo_size = attr->cat_size;
1450 if (md->body->mbo_valid & OBD_MD_FLATIME)
1451 md->body->mbo_atime = attr->cat_atime;
1452 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1453 md->body->mbo_ctime = attr->cat_ctime;
1454 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1455 md->body->mbo_mtime = attr->cat_mtime;
1460 up_read(&lli->lli_lsm_sem);
1465 void ll_clear_inode(struct inode *inode)
1467 struct ll_inode_info *lli = ll_i2info(inode);
1468 struct ll_sb_info *sbi = ll_i2sbi(inode);
1471 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1472 PFID(ll_inode2fid(inode)), inode);
1474 if (S_ISDIR(inode->i_mode)) {
1475 /* these should have been cleared in ll_file_release */
1476 LASSERT(lli->lli_opendir_key == NULL);
1477 LASSERT(lli->lli_sai == NULL);
1478 LASSERT(lli->lli_opendir_pid == 0);
1481 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1483 LASSERT(!lli->lli_open_fd_write_count);
1484 LASSERT(!lli->lli_open_fd_read_count);
1485 LASSERT(!lli->lli_open_fd_exec_count);
1487 if (lli->lli_mds_write_och)
1488 ll_md_real_close(inode, FMODE_WRITE);
1489 if (lli->lli_mds_exec_och)
1490 ll_md_real_close(inode, FMODE_EXEC);
1491 if (lli->lli_mds_read_och)
1492 ll_md_real_close(inode, FMODE_READ);
1494 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1495 OBD_FREE(lli->lli_symlink_name,
1496 strlen(lli->lli_symlink_name) + 1);
1497 lli->lli_symlink_name = NULL;
1500 ll_xattr_cache_destroy(inode);
1502 #ifdef CONFIG_FS_POSIX_ACL
1503 forget_all_cached_acls(inode);
1504 if (lli->lli_posix_acl) {
1505 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1506 posix_acl_release(lli->lli_posix_acl);
1507 lli->lli_posix_acl = NULL;
1510 lli->lli_inode_magic = LLI_INODE_DEAD;
1512 if (S_ISDIR(inode->i_mode))
1513 ll_dir_clear_lsm_md(inode);
1514 else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1515 LASSERT(list_empty(&lli->lli_agl_list));
1518 * XXX This has to be done before lsm is freed below, because
1519 * cl_object still uses inode lsm.
1521 cl_inode_fini(inode);
1526 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1528 struct lustre_md md;
1529 struct inode *inode = dentry->d_inode;
1530 struct ll_sb_info *sbi = ll_i2sbi(inode);
1531 struct ptlrpc_request *request = NULL;
1535 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1536 LUSTRE_OPC_ANY, NULL);
1537 if (IS_ERR(op_data))
1538 RETURN(PTR_ERR(op_data));
1540 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1542 ptlrpc_req_finished(request);
1543 if (rc == -ENOENT) {
1545 /* Unlinked special device node? Or just a race?
1546 * Pretend we done everything. */
1547 if (!S_ISREG(inode->i_mode) &&
1548 !S_ISDIR(inode->i_mode)) {
1549 ia_valid = op_data->op_attr.ia_valid;
1550 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1551 rc = simple_setattr(dentry, &op_data->op_attr);
1552 op_data->op_attr.ia_valid = ia_valid;
1554 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1555 CERROR("md_setattr fails: rc = %d\n", rc);
1560 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1561 sbi->ll_md_exp, &md);
1563 ptlrpc_req_finished(request);
1567 ia_valid = op_data->op_attr.ia_valid;
1568 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1569 * cache is not cleared yet. */
1570 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1571 if (S_ISREG(inode->i_mode))
1573 rc = simple_setattr(dentry, &op_data->op_attr);
1574 if (S_ISREG(inode->i_mode))
1575 inode_unlock(inode);
1576 op_data->op_attr.ia_valid = ia_valid;
1578 rc = ll_update_inode(inode, &md);
1579 ptlrpc_req_finished(request);
1584 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1585 * object(s) determine the file size and mtime. Otherwise, the MDS will
1586 * keep these values until such a time that objects are allocated for it.
1587 * We do the MDS operations first, as it is checking permissions for us.
1588 * We don't to the MDS RPC if there is nothing that we want to store there,
1589 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1590 * going to do an RPC anyways.
1592 * If we are doing a truncate, we will send the mtime and ctime updates
1593 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1594 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1597 * In case of HSMimport, we only set attr on MDS.
1599 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr,
1600 enum op_xvalid xvalid, bool hsm_import)
1602 struct inode *inode = dentry->d_inode;
1603 struct ll_inode_info *lli = ll_i2info(inode);
1604 struct md_op_data *op_data = NULL;
1608 CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1609 "valid %x, hsm_import %d\n",
1610 ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1611 inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1614 if (attr->ia_valid & ATTR_SIZE) {
1615 /* Check new size against VFS/VM file size limit and rlimit */
1616 rc = inode_newsize_ok(inode, attr->ia_size);
1620 /* The maximum Lustre file size is variable, based on the
1621 * OST maximum object size and number of stripes. This
1622 * needs another check in addition to the VFS check above. */
1623 if (attr->ia_size > ll_file_maxbytes(inode)) {
1624 CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
1625 PFID(&lli->lli_fid), attr->ia_size,
1626 ll_file_maxbytes(inode));
1630 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1633 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1634 if (attr->ia_valid & TIMES_SET_FLAGS) {
1635 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1636 !cfs_capable(CFS_CAP_FOWNER))
1640 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1641 if (!(xvalid & OP_XVALID_CTIME_SET) &&
1642 (attr->ia_valid & ATTR_CTIME)) {
1643 attr->ia_ctime = current_time(inode);
1644 xvalid |= OP_XVALID_CTIME_SET;
1646 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1647 (attr->ia_valid & ATTR_ATIME)) {
1648 attr->ia_atime = current_time(inode);
1649 attr->ia_valid |= ATTR_ATIME_SET;
1651 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1652 (attr->ia_valid & ATTR_MTIME)) {
1653 attr->ia_mtime = current_time(inode);
1654 attr->ia_valid |= ATTR_MTIME_SET;
1657 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1658 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %llu\n",
1659 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1660 (s64)ktime_get_real_seconds());
1662 if (S_ISREG(inode->i_mode)) {
1663 if (attr->ia_valid & ATTR_SIZE)
1664 inode_dio_write_done(inode);
1665 inode_unlock(inode);
1668 /* We always do an MDS RPC, even if we're only changing the size;
1669 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1671 OBD_ALLOC_PTR(op_data);
1672 if (op_data == NULL)
1673 GOTO(out, rc = -ENOMEM);
1675 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1676 /* If we are changing file size, file content is
1677 * modified, flag it.
1679 xvalid |= OP_XVALID_OWNEROVERRIDE;
1680 op_data->op_bias |= MDS_DATA_MODIFIED;
1681 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1684 if (attr->ia_valid & ATTR_FILE) {
1685 struct ll_file_data *fd = LUSTRE_FPRIVATE(attr->ia_file);
1687 if (fd->fd_lease_och)
1688 op_data->op_bias |= MDS_TRUNC_KEEP_LEASE;
1691 op_data->op_attr = *attr;
1692 op_data->op_xvalid = xvalid;
1694 rc = ll_md_setattr(dentry, op_data);
1698 if (!S_ISREG(inode->i_mode) || hsm_import)
1701 if (attr->ia_valid & (ATTR_SIZE | ATTR_ATIME | ATTR_ATIME_SET |
1702 ATTR_MTIME | ATTR_MTIME_SET | ATTR_CTIME) ||
1703 xvalid & OP_XVALID_CTIME_SET) {
1704 /* For truncate and utimes sending attributes to OSTs, setting
1705 * mtime/atime to the past will be performed under PW [0:EOF]
1706 * extent lock (new_size:EOF for truncate). It may seem
1707 * excessive to send mtime/atime updates to OSTs when not
1708 * setting times to past, but it is necessary due to possible
1709 * time de-synchronization between MDT inode and OST objects
1711 rc = cl_setattr_ost(lli->lli_clob, attr, xvalid, 0);
1714 /* If the file was restored, it needs to set dirty flag.
1716 * We've already sent MDS_DATA_MODIFIED flag in
1717 * ll_md_setattr() for truncate. However, the MDT refuses to
1718 * set the HS_DIRTY flag on released files, so we have to set
1719 * it again if the file has been restored. Please check how
1720 * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1722 * Please notice that if the file is not released, the previous
1723 * MDS_DATA_MODIFIED has taken effect and usually
1724 * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1725 * This way we can save an RPC for common open + trunc
1727 if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1728 struct hsm_state_set hss = {
1729 .hss_valid = HSS_SETMASK,
1730 .hss_setmask = HS_DIRTY,
1734 rc2 = ll_hsm_state_set(inode, &hss);
1735 /* truncate and write can happen at the same time, so that
1736 * the file can be set modified even though the file is not
1737 * restored from released state, and ll_hsm_state_set() is
1738 * not applicable for the file, and rc2 < 0 is normal in this
1741 CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1742 PFID(ll_inode2fid(inode)), rc2);
1747 if (op_data != NULL)
1748 ll_finish_md_op_data(op_data);
1750 if (S_ISREG(inode->i_mode)) {
1752 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1753 inode_dio_wait(inode);
1754 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
1755 * flag. ll_update_inode (called from ll_md_setattr), clears
1756 * inode flags, so there is a gap where S_NOSEC is not set.
1757 * This can cause a writer to take the i_mutex unnecessarily,
1758 * but this is safe to do and should be rare. */
1759 inode_has_no_xattr(inode);
1762 ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1763 LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1768 int ll_setattr(struct dentry *de, struct iattr *attr)
1770 int mode = de->d_inode->i_mode;
1771 enum op_xvalid xvalid = 0;
1773 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1774 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1775 xvalid |= OP_XVALID_OWNEROVERRIDE;
1777 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1778 (ATTR_SIZE|ATTR_MODE)) &&
1779 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1780 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1781 !(attr->ia_mode & S_ISGID))))
1782 attr->ia_valid |= ATTR_FORCE;
1784 if ((attr->ia_valid & ATTR_MODE) &&
1786 !(attr->ia_mode & S_ISUID) &&
1787 !(attr->ia_valid & ATTR_KILL_SUID))
1788 attr->ia_valid |= ATTR_KILL_SUID;
1790 if ((attr->ia_valid & ATTR_MODE) &&
1791 ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1792 !(attr->ia_mode & S_ISGID) &&
1793 !(attr->ia_valid & ATTR_KILL_SGID))
1794 attr->ia_valid |= ATTR_KILL_SGID;
1796 return ll_setattr_raw(de, attr, xvalid, false);
1799 int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
1802 struct obd_statfs obd_osfs;
1807 max_age = ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS;
1809 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1811 CERROR("md_statfs fails: rc = %d\n", rc);
1815 osfs->os_type = LL_SUPER_MAGIC;
1817 CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1818 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1820 if (osfs->os_state & OS_STATE_SUM)
1823 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1824 flags |= OBD_STATFS_NODELAY;
1826 rc = obd_statfs(NULL, sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1828 CERROR("obd_statfs fails: rc = %d\n", rc);
1832 CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1833 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1836 osfs->os_bsize = obd_osfs.os_bsize;
1837 osfs->os_blocks = obd_osfs.os_blocks;
1838 osfs->os_bfree = obd_osfs.os_bfree;
1839 osfs->os_bavail = obd_osfs.os_bavail;
1841 /* If we don't have as many objects free on the OST as inodes
1842 * on the MDS, we reduce the total number of inodes to
1843 * compensate, so that the "inodes in use" number is correct.
1845 if (obd_osfs.os_ffree < osfs->os_ffree) {
1846 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1848 osfs->os_ffree = obd_osfs.os_ffree;
1854 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1856 struct super_block *sb = de->d_sb;
1857 struct obd_statfs osfs;
1858 __u64 fsid = huge_encode_dev(sb->s_dev);
1861 CDEBUG(D_VFSTRACE, "VFS Op: at %llu jiffies\n", get_jiffies_64());
1862 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1864 /* Some amount of caching on the client is allowed */
1865 rc = ll_statfs_internal(ll_s2sbi(sb), &osfs, OBD_STATFS_SUM);
1869 statfs_unpack(sfs, &osfs);
1871 /* We need to downshift for all 32-bit kernels, because we can't
1872 * tell if the kernel is being called via sys_statfs64() or not.
1873 * Stop before overflowing f_bsize - in which case it is better
1874 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1875 if (sizeof(long) < 8) {
1876 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1879 osfs.os_blocks >>= 1;
1880 osfs.os_bfree >>= 1;
1881 osfs.os_bavail >>= 1;
1885 sfs->f_blocks = osfs.os_blocks;
1886 sfs->f_bfree = osfs.os_bfree;
1887 sfs->f_bavail = osfs.os_bavail;
1888 sfs->f_fsid.val[0] = (__u32)fsid;
1889 sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1893 void ll_inode_size_lock(struct inode *inode)
1895 struct ll_inode_info *lli;
1897 LASSERT(!S_ISDIR(inode->i_mode));
1899 lli = ll_i2info(inode);
1900 mutex_lock(&lli->lli_size_mutex);
1903 void ll_inode_size_unlock(struct inode *inode)
1905 struct ll_inode_info *lli;
1907 lli = ll_i2info(inode);
1908 mutex_unlock(&lli->lli_size_mutex);
1911 void ll_update_inode_flags(struct inode *inode, int ext_flags)
1913 inode->i_flags = ll_ext_to_inode_flags(ext_flags);
1914 if (ext_flags & LUSTRE_PROJINHERIT_FL)
1915 ll_file_set_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
1917 ll_file_clear_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
1920 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1922 struct ll_inode_info *lli = ll_i2info(inode);
1923 struct mdt_body *body = md->body;
1924 struct ll_sb_info *sbi = ll_i2sbi(inode);
1927 if (body->mbo_valid & OBD_MD_FLEASIZE) {
1928 rc = cl_file_inode_init(inode, md);
1933 if (S_ISDIR(inode->i_mode)) {
1934 rc = ll_update_lsm_md(inode, md);
1939 #ifdef CONFIG_FS_POSIX_ACL
1940 if (body->mbo_valid & OBD_MD_FLACL) {
1941 spin_lock(&lli->lli_lock);
1942 if (lli->lli_posix_acl)
1943 posix_acl_release(lli->lli_posix_acl);
1944 lli->lli_posix_acl = md->posix_acl;
1945 spin_unlock(&lli->lli_lock);
1948 inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1949 sbi->ll_flags & LL_SBI_32BIT_API);
1950 inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1952 if (body->mbo_valid & OBD_MD_FLATIME) {
1953 if (body->mbo_atime > LTIME_S(inode->i_atime))
1954 LTIME_S(inode->i_atime) = body->mbo_atime;
1955 lli->lli_atime = body->mbo_atime;
1958 if (body->mbo_valid & OBD_MD_FLMTIME) {
1959 if (body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1960 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1961 "to %llu\n", inode->i_ino,
1962 LTIME_S(inode->i_mtime), body->mbo_mtime);
1963 LTIME_S(inode->i_mtime) = body->mbo_mtime;
1965 lli->lli_mtime = body->mbo_mtime;
1968 if (body->mbo_valid & OBD_MD_FLCTIME) {
1969 if (body->mbo_ctime > LTIME_S(inode->i_ctime))
1970 LTIME_S(inode->i_ctime) = body->mbo_ctime;
1971 lli->lli_ctime = body->mbo_ctime;
1974 /* Clear i_flags to remove S_NOSEC before permissions are updated */
1975 if (body->mbo_valid & OBD_MD_FLFLAGS)
1976 ll_update_inode_flags(inode, body->mbo_flags);
1977 if (body->mbo_valid & OBD_MD_FLMODE)
1978 inode->i_mode = (inode->i_mode & S_IFMT) |
1979 (body->mbo_mode & ~S_IFMT);
1981 if (body->mbo_valid & OBD_MD_FLTYPE)
1982 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1983 (body->mbo_mode & S_IFMT);
1985 LASSERT(inode->i_mode != 0);
1986 if (S_ISREG(inode->i_mode))
1987 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1988 LL_MAX_BLKSIZE_BITS);
1990 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1992 if (body->mbo_valid & OBD_MD_FLUID)
1993 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
1994 if (body->mbo_valid & OBD_MD_FLGID)
1995 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
1996 if (body->mbo_valid & OBD_MD_FLPROJID)
1997 lli->lli_projid = body->mbo_projid;
1998 if (body->mbo_valid & OBD_MD_FLNLINK)
1999 set_nlink(inode, body->mbo_nlink);
2000 if (body->mbo_valid & OBD_MD_FLRDEV)
2001 inode->i_rdev = old_decode_dev(body->mbo_rdev);
2003 if (body->mbo_valid & OBD_MD_FLID) {
2004 /* FID shouldn't be changed! */
2005 if (fid_is_sane(&lli->lli_fid)) {
2006 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
2007 "Trying to change FID "DFID
2008 " to the "DFID", inode "DFID"(%p)\n",
2009 PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
2010 PFID(ll_inode2fid(inode)), inode);
2012 lli->lli_fid = body->mbo_fid1;
2016 LASSERT(fid_seq(&lli->lli_fid) != 0);
2018 if (body->mbo_valid & OBD_MD_FLSIZE) {
2019 i_size_write(inode, body->mbo_size);
2021 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
2022 PFID(ll_inode2fid(inode)),
2023 (unsigned long long)body->mbo_size);
2025 if (body->mbo_valid & OBD_MD_FLBLOCKS)
2026 inode->i_blocks = body->mbo_blocks;
2029 if (body->mbo_valid & OBD_MD_TSTATE) {
2030 /* Set LLIF_FILE_RESTORING if restore ongoing and
2031 * clear it when done to ensure to start again
2032 * glimpsing updated attrs
2034 if (body->mbo_t_state & MS_RESTORE)
2035 ll_file_set_flag(lli, LLIF_FILE_RESTORING);
2037 ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
2043 int ll_read_inode2(struct inode *inode, void *opaque)
2045 struct lustre_md *md = opaque;
2046 struct ll_inode_info *lli = ll_i2info(inode);
2050 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2051 PFID(&lli->lli_fid), inode);
2053 /* Core attributes from the MDS first. This is a new inode, and
2054 * the VFS doesn't zero times in the core inode so we have to do
2055 * it ourselves. They will be overwritten by either MDS or OST
2056 * attributes - we just need to make sure they aren't newer. */
2057 LTIME_S(inode->i_mtime) = 0;
2058 LTIME_S(inode->i_atime) = 0;
2059 LTIME_S(inode->i_ctime) = 0;
2061 rc = ll_update_inode(inode, md);
2065 /* OIDEBUG(inode); */
2067 #ifdef HAVE_BACKING_DEV_INFO
2068 /* initializing backing dev info. */
2069 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2071 if (S_ISREG(inode->i_mode)) {
2072 struct ll_sb_info *sbi = ll_i2sbi(inode);
2073 inode->i_op = &ll_file_inode_operations;
2074 inode->i_fop = sbi->ll_fop;
2075 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2077 } else if (S_ISDIR(inode->i_mode)) {
2078 inode->i_op = &ll_dir_inode_operations;
2079 inode->i_fop = &ll_dir_operations;
2081 } else if (S_ISLNK(inode->i_mode)) {
2082 inode->i_op = &ll_fast_symlink_inode_operations;
2085 inode->i_op = &ll_special_inode_operations;
2087 init_special_inode(inode, inode->i_mode,
2096 void ll_delete_inode(struct inode *inode)
2098 struct ll_inode_info *lli = ll_i2info(inode);
2099 struct address_space *mapping = &inode->i_data;
2100 unsigned long nrpages;
2103 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2104 /* It is last chance to write out dirty pages,
2105 * otherwise we may lose data while umount */
2106 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
2108 truncate_inode_pages_final(mapping);
2110 /* Workaround for LU-118: Note nrpages may not be totally updated when
2111 * truncate_inode_pages() returns, as there can be a page in the process
2112 * of deletion (inside __delete_from_page_cache()) in the specified
2113 * range. Thus mapping->nrpages can be non-zero when this function
2114 * returns even after truncation of the whole mapping. Only do this if
2115 * npages isn't already zero.
2117 nrpages = mapping->nrpages;
2119 spin_lock_irq(&mapping->tree_lock);
2120 nrpages = mapping->nrpages;
2121 spin_unlock_irq(&mapping->tree_lock);
2122 } /* Workaround end */
2124 LASSERTF(nrpages == 0, "%s: inode="DFID"(%p) nrpages=%lu, "
2125 "see https://jira.whamcloud.com/browse/LU-118\n",
2126 ll_get_fsname(inode->i_sb, NULL, 0),
2127 PFID(ll_inode2fid(inode)), inode, nrpages);
2129 #ifdef HAVE_SBOPS_EVICT_INODE
2130 ll_clear_inode(inode);
2137 int ll_iocontrol(struct inode *inode, struct file *file,
2138 unsigned int cmd, unsigned long arg)
2140 struct ll_sb_info *sbi = ll_i2sbi(inode);
2141 struct ptlrpc_request *req = NULL;
2146 case FS_IOC_GETFLAGS: {
2147 struct mdt_body *body;
2148 struct md_op_data *op_data;
2150 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2151 0, 0, LUSTRE_OPC_ANY,
2153 if (IS_ERR(op_data))
2154 RETURN(PTR_ERR(op_data));
2156 op_data->op_valid = OBD_MD_FLFLAGS;
2157 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2158 ll_finish_md_op_data(op_data);
2160 CERROR("%s: failure inode "DFID": rc = %d\n",
2161 sbi->ll_md_exp->exp_obd->obd_name,
2162 PFID(ll_inode2fid(inode)), rc);
2166 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2168 flags = body->mbo_flags;
2170 ptlrpc_req_finished(req);
2172 RETURN(put_user(flags, (int __user *)arg));
2174 case FS_IOC_SETFLAGS: {
2176 struct md_op_data *op_data;
2177 struct cl_object *obj;
2178 struct fsxattr fa = { 0 };
2180 if (get_user(flags, (int __user *)arg))
2183 fa.fsx_projid = ll_i2info(inode)->lli_projid;
2184 if (flags & LUSTRE_PROJINHERIT_FL)
2185 fa.fsx_xflags = FS_XFLAG_PROJINHERIT;
2187 rc = ll_ioctl_check_project(inode, &fa);
2191 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2192 LUSTRE_OPC_ANY, NULL);
2193 if (IS_ERR(op_data))
2194 RETURN(PTR_ERR(op_data));
2196 op_data->op_attr_flags = flags;
2197 op_data->op_xvalid |= OP_XVALID_FLAGS;
2198 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2199 ll_finish_md_op_data(op_data);
2200 ptlrpc_req_finished(req);
2204 ll_update_inode_flags(inode, flags);
2206 obj = ll_i2info(inode)->lli_clob;
2210 OBD_ALLOC_PTR(attr);
2214 rc = cl_setattr_ost(obj, attr, OP_XVALID_FLAGS, flags);
2226 int ll_flush_ctx(struct inode *inode)
2228 struct ll_sb_info *sbi = ll_i2sbi(inode);
2230 CDEBUG(D_SEC, "flush context for user %d\n",
2231 from_kuid(&init_user_ns, current_uid()));
2233 obd_set_info_async(NULL, sbi->ll_md_exp,
2234 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2236 obd_set_info_async(NULL, sbi->ll_dt_exp,
2237 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2242 /* umount -f client means force down, don't save state */
2243 void ll_umount_begin(struct super_block *sb)
2245 struct ll_sb_info *sbi = ll_s2sbi(sb);
2246 struct obd_device *obd;
2247 struct obd_ioctl_data *ioc_data;
2248 struct l_wait_info lwi;
2249 wait_queue_head_t waitq;
2252 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2253 sb->s_count, atomic_read(&sb->s_active));
2255 obd = class_exp2obd(sbi->ll_md_exp);
2257 CERROR("Invalid MDC connection handle %#llx\n",
2258 sbi->ll_md_exp->exp_handle.h_cookie);
2264 obd = class_exp2obd(sbi->ll_dt_exp);
2266 CERROR("Invalid LOV connection handle %#llx\n",
2267 sbi->ll_dt_exp->exp_handle.h_cookie);
2273 OBD_ALLOC_PTR(ioc_data);
2275 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2276 sizeof *ioc_data, ioc_data, NULL);
2278 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2279 sizeof *ioc_data, ioc_data, NULL);
2281 OBD_FREE_PTR(ioc_data);
2284 /* Really, we'd like to wait until there are no requests outstanding,
2285 * and then continue. For now, we just periodically checking for vfs
2286 * to decrement mnt_cnt and hope to finish it within 10sec.
2288 init_waitqueue_head(&waitq);
2289 lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(10),
2290 cfs_time_seconds(1), NULL, NULL);
2291 l_wait_event(waitq, may_umount(sbi->ll_mnt.mnt), &lwi);
2296 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2298 struct ll_sb_info *sbi = ll_s2sbi(sb);
2299 char *profilenm = get_profile_name(sb);
2303 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2304 read_only = *flags & MS_RDONLY;
2305 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2306 sizeof(KEY_READ_ONLY),
2307 KEY_READ_ONLY, sizeof(read_only),
2310 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2311 profilenm, read_only ?
2312 "read-only" : "read-write", err);
2317 sb->s_flags |= MS_RDONLY;
2319 sb->s_flags &= ~MS_RDONLY;
2321 if (sbi->ll_flags & LL_SBI_VERBOSE)
2322 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2323 read_only ? "read-only" : "read-write");
2329 * Cleanup the open handle that is cached on MDT-side.
2331 * For open case, the client side open handling thread may hit error
2332 * after the MDT grant the open. Under such case, the client should
2333 * send close RPC to the MDT as cleanup; otherwise, the open handle
2334 * on the MDT will be leaked there until the client umount or evicted.
2336 * In further, if someone unlinked the file, because the open handle
2337 * holds the reference on such file/object, then it will block the
2338 * subsequent threads that want to locate such object via FID.
2340 * \param[in] sb super block for this file-system
2341 * \param[in] open_req pointer to the original open request
2343 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2345 struct mdt_body *body;
2346 struct md_op_data *op_data;
2347 struct ptlrpc_request *close_req = NULL;
2348 struct obd_export *exp = ll_s2sbi(sb)->ll_md_exp;
2351 body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2352 OBD_ALLOC_PTR(op_data);
2353 if (op_data == NULL) {
2354 CWARN("%s: cannot allocate op_data to release open handle for "
2356 ll_get_fsname(sb, NULL, 0), PFID(&body->mbo_fid1));
2361 op_data->op_fid1 = body->mbo_fid1;
2362 op_data->op_open_handle = body->mbo_open_handle;
2363 op_data->op_mod_time = ktime_get_real_seconds();
2364 md_close(exp, op_data, NULL, &close_req);
2365 ptlrpc_req_finished(close_req);
2366 ll_finish_md_op_data(op_data);
2371 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2372 struct super_block *sb, struct lookup_intent *it)
2374 struct ll_sb_info *sbi = NULL;
2375 struct lustre_md md = { NULL };
2379 LASSERT(*inode || sb);
2380 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2381 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2382 sbi->ll_md_exp, &md);
2387 rc = ll_update_inode(*inode, &md);
2391 LASSERT(sb != NULL);
2394 * At this point server returns to client's same fid as client
2395 * generated for creating. So using ->fid1 is okay here.
2397 if (!fid_is_sane(&md.body->mbo_fid1)) {
2398 CERROR("%s: Fid is insane "DFID"\n",
2399 ll_get_fsname(sb, NULL, 0),
2400 PFID(&md.body->mbo_fid1));
2401 GOTO(out, rc = -EINVAL);
2404 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2405 sbi->ll_flags & LL_SBI_32BIT_API),
2407 if (IS_ERR(*inode)) {
2408 #ifdef CONFIG_FS_POSIX_ACL
2410 posix_acl_release(md.posix_acl);
2411 md.posix_acl = NULL;
2414 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2416 CERROR("new_inode -fatal: rc %d\n", rc);
2421 /* Handling piggyback layout lock.
2422 * Layout lock can be piggybacked by getattr and open request.
2423 * The lsm can be applied to inode only if it comes with a layout lock
2424 * otherwise correct layout may be overwritten, for example:
2425 * 1. proc1: mdt returns a lsm but not granting layout
2426 * 2. layout was changed by another client
2427 * 3. proc2: refresh layout and layout lock granted
2428 * 4. proc1: to apply a stale layout */
2429 if (it != NULL && it->it_lock_mode != 0) {
2430 struct lustre_handle lockh;
2431 struct ldlm_lock *lock;
2433 lockh.cookie = it->it_lock_handle;
2434 lock = ldlm_handle2lock(&lockh);
2435 LASSERT(lock != NULL);
2436 if (ldlm_has_layout(lock)) {
2437 struct cl_object_conf conf;
2439 memset(&conf, 0, sizeof(conf));
2440 conf.coc_opc = OBJECT_CONF_SET;
2441 conf.coc_inode = *inode;
2442 conf.coc_lock = lock;
2443 conf.u.coc_layout = md.layout;
2444 (void)ll_layout_conf(*inode, &conf);
2446 LDLM_LOCK_PUT(lock);
2452 md_free_lustre_md(sbi->ll_md_exp, &md);
2455 if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2456 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2461 int ll_obd_statfs(struct inode *inode, void __user *arg)
2463 struct ll_sb_info *sbi = NULL;
2464 struct obd_export *exp;
2466 struct obd_ioctl_data *data = NULL;
2470 if (!inode || !(sbi = ll_i2sbi(inode)))
2471 GOTO(out_statfs, rc = -EINVAL);
2473 rc = obd_ioctl_getdata(&buf, &len, arg);
2475 GOTO(out_statfs, rc);
2478 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2479 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2480 GOTO(out_statfs, rc = -EINVAL);
2482 if (data->ioc_inllen1 != sizeof(__u32) ||
2483 data->ioc_inllen2 != sizeof(__u32) ||
2484 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2485 data->ioc_plen2 != sizeof(struct obd_uuid))
2486 GOTO(out_statfs, rc = -EINVAL);
2488 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2489 if (type & LL_STATFS_LMV)
2490 exp = sbi->ll_md_exp;
2491 else if (type & LL_STATFS_LOV)
2492 exp = sbi->ll_dt_exp;
2494 GOTO(out_statfs, rc = -ENODEV);
2496 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2498 GOTO(out_statfs, rc);
2500 OBD_FREE_LARGE(buf, len);
2505 * this is normally called in ll_fini_md_op_data(), but sometimes it needs to
2506 * be called early to avoid deadlock.
2508 void ll_unlock_md_op_lsm(struct md_op_data *op_data)
2510 if (op_data->op_mea2_sem) {
2511 up_read(op_data->op_mea2_sem);
2512 op_data->op_mea2_sem = NULL;
2515 if (op_data->op_mea1_sem) {
2516 up_read(op_data->op_mea1_sem);
2517 op_data->op_mea1_sem = NULL;
2521 /* this function prepares md_op_data hint for passing it down to MD stack. */
2522 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2523 struct inode *i1, struct inode *i2,
2524 const char *name, size_t namelen,
2525 __u32 mode, __u32 opc, void *data)
2527 LASSERT(i1 != NULL);
2530 /* Do not reuse namelen for something else. */
2532 return ERR_PTR(-EINVAL);
2534 if (namelen > ll_i2sbi(i1)->ll_namelen)
2535 return ERR_PTR(-ENAMETOOLONG);
2537 if (!lu_name_is_valid_2(name, namelen))
2538 return ERR_PTR(-EINVAL);
2541 if (op_data == NULL)
2542 OBD_ALLOC_PTR(op_data);
2544 if (op_data == NULL)
2545 return ERR_PTR(-ENOMEM);
2547 ll_i2gids(op_data->op_suppgids, i1, i2);
2548 op_data->op_fid1 = *ll_inode2fid(i1);
2549 op_data->op_default_stripe_offset = -1;
2551 if (S_ISDIR(i1->i_mode)) {
2552 down_read(&ll_i2info(i1)->lli_lsm_sem);
2553 op_data->op_mea1_sem = &ll_i2info(i1)->lli_lsm_sem;
2554 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2555 if (opc == LUSTRE_OPC_MKDIR)
2556 op_data->op_default_stripe_offset =
2557 ll_i2info(i1)->lli_def_stripe_offset;
2561 op_data->op_fid2 = *ll_inode2fid(i2);
2562 if (S_ISDIR(i2->i_mode)) {
2564 down_read(&ll_i2info(i2)->lli_lsm_sem);
2565 op_data->op_mea2_sem =
2566 &ll_i2info(i2)->lli_lsm_sem;
2568 op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2571 fid_zero(&op_data->op_fid2);
2574 if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2575 op_data->op_cli_flags |= CLI_HASH64;
2577 if (ll_need_32bit_api(ll_i2sbi(i1)))
2578 op_data->op_cli_flags |= CLI_API32;
2580 op_data->op_name = name;
2581 op_data->op_namelen = namelen;
2582 op_data->op_mode = mode;
2583 op_data->op_mod_time = ktime_get_real_seconds();
2584 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2585 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2586 op_data->op_cap = cfs_curproc_cap_pack();
2587 op_data->op_mds = 0;
2588 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2589 filename_is_volatile(name, namelen, &op_data->op_mds)) {
2590 op_data->op_bias |= MDS_CREATE_VOLATILE;
2592 op_data->op_data = data;
2597 void ll_finish_md_op_data(struct md_op_data *op_data)
2599 ll_unlock_md_op_lsm(op_data);
2600 security_release_secctx(op_data->op_file_secctx,
2601 op_data->op_file_secctx_size);
2602 OBD_FREE_PTR(op_data);
2605 #ifdef HAVE_SUPEROPS_USE_DENTRY
2606 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2608 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2611 struct ll_sb_info *sbi;
2613 #ifdef HAVE_SUPEROPS_USE_DENTRY
2614 LASSERT((seq != NULL) && (dentry != NULL));
2615 sbi = ll_s2sbi(dentry->d_sb);
2617 LASSERT((seq != NULL) && (vfs != NULL));
2618 sbi = ll_s2sbi(vfs->mnt_sb);
2621 if (sbi->ll_flags & LL_SBI_NOLCK)
2622 seq_puts(seq, ",nolock");
2624 if (sbi->ll_flags & LL_SBI_FLOCK)
2625 seq_puts(seq, ",flock");
2627 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2628 seq_puts(seq, ",localflock");
2630 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2631 seq_puts(seq, ",user_xattr");
2633 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2634 seq_puts(seq, ",lazystatfs");
2636 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2637 seq_puts(seq, ",user_fid2path");
2639 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2640 seq_puts(seq, ",always_ping");
2646 * Get obd name by cmd, and copy out to user space
2648 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2650 struct ll_sb_info *sbi = ll_i2sbi(inode);
2651 struct obd_device *obd;
2654 if (cmd == OBD_IOC_GETDTNAME)
2655 obd = class_exp2obd(sbi->ll_dt_exp);
2656 else if (cmd == OBD_IOC_GETMDNAME)
2657 obd = class_exp2obd(sbi->ll_md_exp);
2664 if (copy_to_user((void __user *)arg, obd->obd_name,
2665 strlen(obd->obd_name) + 1))
2672 * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2673 * fsname will be returned in this buffer; otherwise, a static buffer will be
2674 * used to store the fsname and returned to caller.
2676 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2678 static char fsname_static[MTI_NAME_MAXLEN];
2679 struct lustre_sb_info *lsi = s2lsi(sb);
2684 /* this means the caller wants to use static buffer
2685 * and it doesn't care about race. Usually this is
2686 * in error reporting path */
2687 buf = fsname_static;
2688 buflen = sizeof(fsname_static);
2691 len = strlen(lsi->lsi_lmd->lmd_profile);
2692 ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2693 if (ptr && (strcmp(ptr, "-client") == 0))
2696 if (unlikely(len >= buflen))
2698 strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2704 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2711 p.mnt = current->fs->root.mnt;
2713 path = d_path(&p, buf, bufsize);
2718 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2720 char *buf, *path = NULL;
2721 struct dentry *dentry = NULL;
2722 struct inode *inode = page->mapping->host;
2724 /* this can be called inside spin lock so use GFP_ATOMIC. */
2725 buf = (char *)__get_free_page(GFP_ATOMIC);
2727 dentry = d_find_alias(page->mapping->host);
2729 path = ll_d_path(dentry, buf, PAGE_SIZE);
2733 "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2734 "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2735 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2736 PFID(ll_inode2fid(inode)),
2737 (path && !IS_ERR(path)) ? path : "", ioret);
2743 free_page((unsigned long)buf);
2746 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2747 struct lov_user_md **kbuf)
2749 struct lov_user_md lum;
2753 if (copy_from_user(&lum, md, sizeof(lum)))
2756 lum_size = ll_lov_user_md_size(&lum);
2760 OBD_ALLOC(*kbuf, lum_size);
2764 if (copy_from_user(*kbuf, md, lum_size) != 0) {
2765 OBD_FREE(*kbuf, lum_size);
2773 * Compute llite root squash state after a change of root squash
2774 * configuration setting or add/remove of a lnet nid
2776 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2778 struct root_squash_info *squash = &sbi->ll_squash;
2781 struct lnet_process_id id;
2783 /* Update norootsquash flag */
2784 down_write(&squash->rsi_sem);
2785 if (list_empty(&squash->rsi_nosquash_nids))
2786 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2788 /* Do not apply root squash as soon as one of our NIDs is
2789 * in the nosquash_nids list */
2792 while (LNetGetId(i++, &id) != -ENOENT) {
2793 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2795 if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2801 sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2803 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2805 up_write(&squash->rsi_sem);
2809 * Parse linkea content to extract information about a given hardlink
2811 * \param[in] ldata - Initialized linkea data
2812 * \param[in] linkno - Link identifier
2813 * \param[out] parent_fid - The entry's parent FID
2814 * \param[out] ln - Entry name destination buffer
2816 * \retval 0 on success
2817 * \retval Appropriate negative error code on failure
2819 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2820 struct lu_fid *parent_fid, struct lu_name *ln)
2826 rc = linkea_init_with_rec(ldata);
2830 if (linkno >= ldata->ld_leh->leh_reccount)
2831 /* beyond last link */
2834 linkea_first_entry(ldata);
2835 for (idx = 0; ldata->ld_lee != NULL; idx++) {
2836 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2841 linkea_next_entry(ldata);
2851 * Get parent FID and name of an identified link. Operation is performed for
2852 * a given link number, letting the caller iterate over linkno to list one or
2853 * all links of an entry.
2855 * \param[in] file - File descriptor against which to perform the operation
2856 * \param[in,out] arg - User-filled structure containing the linkno to operate
2857 * on and the available size. It is eventually filled with
2858 * the requested information or left untouched on error
2860 * \retval - 0 on success
2861 * \retval - Appropriate negative error code on failure
2863 int ll_getparent(struct file *file, struct getparent __user *arg)
2865 struct inode *inode = file_inode(file);
2866 struct linkea_data *ldata;
2867 struct lu_buf buf = LU_BUF_NULL;
2869 struct lu_fid parent_fid;
2876 if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2877 !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2880 if (get_user(name_size, &arg->gp_name_size))
2883 if (get_user(linkno, &arg->gp_linkno))
2886 if (name_size > PATH_MAX)
2889 OBD_ALLOC(ldata, sizeof(*ldata));
2893 rc = linkea_data_new(ldata, &buf);
2895 GOTO(ldata_free, rc);
2897 #ifdef HAVE_XATTR_HANDLER_FLAGS
2898 rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
2899 buf.lb_len, OBD_MD_FLXATTR);
2901 rc = ll_getxattr(file_dentry(file), XATTR_NAME_LINK, buf.lb_buf,
2903 #endif /* HAVE_XATTR_HANDLER_FLAGS */
2907 rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2911 if (ln.ln_namelen >= name_size)
2912 GOTO(lb_free, rc = -EOVERFLOW);
2914 if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2915 GOTO(lb_free, rc = -EFAULT);
2917 if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2918 GOTO(lb_free, rc = -EFAULT);
2920 if (put_user('\0', arg->gp_name + ln.ln_namelen))
2921 GOTO(lb_free, rc = -EFAULT);
2926 OBD_FREE(ldata, sizeof(*ldata));