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;
226 #ifdef HAVE_LRU_RESIZE_SUPPORT
227 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
228 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
230 #ifdef CONFIG_FS_POSIX_ACL
231 data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK |
232 OBD_CONNECT_LARGE_ACL;
235 data->ocd_cksum_types = obd_cksum_types_supported_client();
237 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
238 /* flag mdc connection as lightweight, only used for test
239 * purpose, use with care */
240 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
242 data->ocd_ibits_known = MDS_INODELOCK_FULL;
243 data->ocd_version = LUSTRE_VERSION_CODE;
245 if (sb->s_flags & MS_RDONLY)
246 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
247 if (sbi->ll_flags & LL_SBI_USER_XATTR)
248 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
251 /* Setting this indicates we correctly support S_NOSEC (See kernel
252 * commit 9e1f1de02c2275d7172e18dc4e7c2065777611bf)
254 sb->s_flags |= MS_NOSEC;
257 if (sbi->ll_flags & LL_SBI_FLOCK)
258 sbi->ll_fop = &ll_file_operations_flock;
259 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
260 sbi->ll_fop = &ll_file_operations;
262 sbi->ll_fop = &ll_file_operations_noflock;
264 /* always ping even if server suppress_pings */
265 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
266 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
268 obd_connect_set_secctx(data);
270 data->ocd_brw_size = MD_MAX_BRW_SIZE;
272 err = obd_connect(NULL, &sbi->ll_md_exp, sbi->ll_md_obd,
273 &sbi->ll_sb_uuid, data, NULL);
275 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
276 "recovery, of which this client is not a "
277 "part. Please wait for recovery to complete,"
278 " abort, or time out.\n", md);
281 CERROR("cannot connect to %s: rc = %d\n", md, err);
285 sbi->ll_md_exp->exp_connect_data = *data;
287 err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
288 LUSTRE_SEQ_METADATA);
290 CERROR("%s: Can't init metadata layer FID infrastructure, "
291 "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
295 /* For mount, we only need fs info from MDT0, and also in DNE, it
296 * can make sure the client can be mounted as long as MDT0 is
298 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
299 ktime_get_seconds() -OBD_STATFS_CACHE_SECONDS,
300 OBD_STATFS_FOR_MDT0);
302 GOTO(out_md_fid, err);
304 /* This needs to be after statfs to ensure connect has finished.
305 * Note that "data" does NOT contain the valid connect reply.
306 * If connecting to a 1.8 server there will be no LMV device, so
307 * we can access the MDC export directly and exp_connect_flags will
308 * be non-zero, but if accessing an upgraded 2.1 server it will
309 * have the correct flags filled in.
310 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
311 valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
312 if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
313 valid != CLIENT_CONNECT_MDT_REQD) {
316 OBD_ALLOC_WAIT(buf, PAGE_SIZE);
317 obd_connect_flags2str(buf, PAGE_SIZE,
318 valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
319 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
320 "feature(s) needed for correct operation "
321 "of this client (%s). Please upgrade "
322 "server or downgrade client.\n",
323 sbi->ll_md_exp->exp_obd->obd_name, buf);
324 OBD_FREE(buf, PAGE_SIZE);
325 GOTO(out_md_fid, err = -EPROTO);
328 size = sizeof(*data);
329 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
330 KEY_CONN_DATA, &size, data);
332 CERROR("%s: Get connect data failed: rc = %d\n",
333 sbi->ll_md_exp->exp_obd->obd_name, err);
334 GOTO(out_md_fid, err);
337 LASSERT(osfs->os_bsize);
338 sb->s_blocksize = osfs->os_bsize;
339 sb->s_blocksize_bits = log2(osfs->os_bsize);
340 sb->s_magic = LL_SUPER_MAGIC;
341 sb->s_maxbytes = MAX_LFS_FILESIZE;
342 sbi->ll_namelen = osfs->os_namelen;
343 sbi->ll_mnt.mnt = current->fs->root.mnt;
345 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
346 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
347 LCONSOLE_INFO("Disabling user_xattr feature because "
348 "it is not supported on the server\n");
349 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
352 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
354 sb->s_flags |= MS_POSIXACL;
356 sbi->ll_flags |= LL_SBI_ACL;
358 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
360 sb->s_flags &= ~MS_POSIXACL;
362 sbi->ll_flags &= ~LL_SBI_ACL;
365 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
366 sbi->ll_flags |= LL_SBI_64BIT_HASH;
368 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
369 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
371 if (obd_connect_has_secctx(data))
372 sbi->ll_flags |= LL_SBI_FILE_SECCTX;
374 if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
375 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
376 LCONSOLE_INFO("%s: disabling xattr cache due to "
377 "unknown maximum xattr size.\n", dt);
378 } else if (!sbi->ll_xattr_cache_set) {
379 /* If xattr_cache is already set (no matter 0 or 1)
380 * during processing llog, it won't be enabled here. */
381 sbi->ll_flags |= LL_SBI_XATTR_CACHE;
382 sbi->ll_xattr_cache_enabled = 1;
386 sbi->ll_dt_obd = class_name2obd(dt);
387 if (!sbi->ll_dt_obd) {
388 CERROR("DT %s: not setup or attached\n", dt);
389 GOTO(out_md_fid, err = -ENODEV);
392 /* pass client page size via ocd_grant_blkbits, the server should report
393 * back its backend blocksize for grant calculation purpose */
394 data->ocd_grant_blkbits = PAGE_SHIFT;
396 /* indicate OST features supported by this client */
397 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
398 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
399 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
400 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
401 OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
402 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
403 OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
404 OBD_CONNECT_EINPROGRESS |
405 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
406 OBD_CONNECT_LAYOUTLOCK |
407 OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
408 OBD_CONNECT_BULK_MBITS | OBD_CONNECT_SHORTIO |
409 OBD_CONNECT_FLAGS2 | OBD_CONNECT_GRANT_SHRINK;
411 /* The client currently advertises support for OBD_CONNECT_LOCKAHEAD_OLD so it
412 * can interoperate with an older version of lockahead which was released prior
413 * to landing in master. This support will be dropped when 2.13 development
414 * starts. At the point, we should not just drop the connect flag (below), we
415 * should also remove the support in the code.
417 * Removing it means a few things:
418 * 1. Remove this section here
419 * 2. Remove CEF_NONBLOCK in ll_file_lockahead()
420 * 3. Remove function exp_connect_lockahead_old
421 * 4. Remove LDLM_FL_LOCKAHEAD_OLD_RESERVED in lustre_dlm_flags.h
423 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 12, 50, 0)
424 data->ocd_connect_flags |= OBD_CONNECT_LOCKAHEAD_OLD;
427 data->ocd_connect_flags2 = OBD_CONNECT2_LOCKAHEAD;
429 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
430 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
432 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
433 * disabled by default, because it can still be enabled on the
434 * fly via /proc. As a consequence, we still need to come to an
435 * agreement on the supported algorithms at connect time */
436 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
438 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
439 data->ocd_cksum_types = OBD_CKSUM_ADLER;
441 data->ocd_cksum_types = obd_cksum_types_supported_client();
443 #ifdef HAVE_LRU_RESIZE_SUPPORT
444 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
446 /* always ping even if server suppress_pings */
447 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
448 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
450 CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
451 "ocd_grant: %d\n", data->ocd_connect_flags,
452 data->ocd_version, data->ocd_grant);
454 sbi->ll_dt_obd->obd_upcall.onu_owner = &sbi->ll_lco;
455 sbi->ll_dt_obd->obd_upcall.onu_upcall = cl_ocd_update;
457 data->ocd_brw_size = DT_MAX_BRW_SIZE;
459 err = obd_connect(NULL, &sbi->ll_dt_exp, sbi->ll_dt_obd,
460 &sbi->ll_sb_uuid, data, NULL);
462 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
463 "recovery, of which this client is not a "
464 "part. Please wait for recovery to "
465 "complete, abort, or time out.\n", dt);
468 CERROR("%s: Cannot connect to %s: rc = %d\n",
469 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
473 sbi->ll_dt_exp->exp_connect_data = *data;
475 /* Don't change value if it was specified in the config log */
476 if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages == -1)
477 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
478 max_t(unsigned long, SBI_DEFAULT_READAHEAD_WHOLE_MAX,
479 (data->ocd_brw_size >> PAGE_SHIFT));
481 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
482 LUSTRE_SEQ_METADATA);
484 CERROR("%s: Can't init data layer FID infrastructure, "
485 "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
489 mutex_lock(&sbi->ll_lco.lco_lock);
490 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
491 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
492 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
493 mutex_unlock(&sbi->ll_lco.lco_lock);
495 fid_zero(&sbi->ll_root_fid);
496 err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
499 CERROR("cannot mds_connect: rc = %d\n", err);
500 GOTO(out_lock_cn_cb, err);
502 if (!fid_is_sane(&sbi->ll_root_fid)) {
503 CERROR("%s: Invalid root fid "DFID" during mount\n",
504 sbi->ll_md_exp->exp_obd->obd_name,
505 PFID(&sbi->ll_root_fid));
506 GOTO(out_lock_cn_cb, err = -EINVAL);
508 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
510 sb->s_op = &lustre_super_operations;
511 #ifdef HAVE_XATTR_HANDLER_FLAGS
512 sb->s_xattr = ll_xattr_handlers;
514 #if THREAD_SIZE >= 8192 /*b=17630*/
515 sb->s_export_op = &lustre_export_operations;
519 * XXX: move this to after cbd setup? */
520 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
521 if (sbi->ll_flags & LL_SBI_ACL)
522 valid |= OBD_MD_FLACL;
524 OBD_ALLOC_PTR(op_data);
526 GOTO(out_lock_cn_cb, err = -ENOMEM);
528 op_data->op_fid1 = sbi->ll_root_fid;
529 op_data->op_mode = 0;
530 op_data->op_valid = valid;
532 err = md_getattr(sbi->ll_md_exp, op_data, &request);
534 OBD_FREE_PTR(op_data);
536 CERROR("%s: md_getattr failed for root: rc = %d\n",
537 sbi->ll_md_exp->exp_obd->obd_name, err);
538 GOTO(out_lock_cn_cb, err);
541 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
542 sbi->ll_md_exp, &lmd);
544 CERROR("failed to understand root inode md: rc = %d\n", err);
545 ptlrpc_req_finished(request);
546 GOTO(out_lock_cn_cb, err);
549 LASSERT(fid_is_sane(&sbi->ll_root_fid));
550 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
551 sbi->ll_flags & LL_SBI_32BIT_API),
553 md_free_lustre_md(sbi->ll_md_exp, &lmd);
554 ptlrpc_req_finished(request);
557 #ifdef CONFIG_FS_POSIX_ACL
559 posix_acl_release(lmd.posix_acl);
560 lmd.posix_acl = NULL;
563 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
565 CERROR("lustre_lite: bad iget4 for root\n");
569 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
570 if (sbi->ll_checksum_set) {
571 err = obd_set_info_async(NULL, sbi->ll_dt_exp,
572 sizeof(KEY_CHECKSUM), KEY_CHECKSUM,
573 sizeof(checksum), &checksum, NULL);
575 CERROR("%s: Set checksum failed: rc = %d\n",
576 sbi->ll_dt_exp->exp_obd->obd_name, err);
582 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
583 KEY_CACHE_SET, sizeof(*sbi->ll_cache),
584 sbi->ll_cache, NULL);
586 CERROR("%s: Set cache_set failed: rc = %d\n",
587 sbi->ll_dt_exp->exp_obd->obd_name, err);
591 sb->s_root = d_make_root(root);
592 if (sb->s_root == NULL) {
593 CERROR("%s: can't make root dentry\n",
594 ll_get_fsname(sb, NULL, 0));
595 GOTO(out_root, err = -ENOMEM);
597 #ifdef HAVE_DCACHE_LOCK
598 sb->s_root->d_op = &ll_d_ops;
601 sbi->ll_sdev_orig = sb->s_dev;
603 /* We set sb->s_dev equal on all lustre clients in order to support
604 * NFS export clustering. NFSD requires that the FSID be the same
606 /* s_dev is also used in lt_compare() to compare two fs, but that is
607 * only a node-local comparison. */
608 uuid = obd_get_uuid(sbi->ll_md_exp);
610 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
617 if (sbi->ll_dt_obd) {
618 err = sysfs_create_link(&sbi->ll_kset.kobj,
619 &sbi->ll_dt_obd->obd_kset.kobj,
620 sbi->ll_dt_obd->obd_type->typ_name);
622 CERROR("%s: could not register %s in llite: rc = %d\n",
623 dt, ll_get_fsname(sb, NULL, 0), err);
628 if (sbi->ll_md_obd) {
629 err = sysfs_create_link(&sbi->ll_kset.kobj,
630 &sbi->ll_md_obd->obd_kset.kobj,
631 sbi->ll_md_obd->obd_type->typ_name);
633 CERROR("%s: could not register %s in llite: rc = %d\n",
634 md, ll_get_fsname(sb, NULL, 0), err);
644 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
646 obd_disconnect(sbi->ll_dt_exp);
647 sbi->ll_dt_exp = NULL;
648 sbi->ll_dt_obd = NULL;
650 obd_fid_fini(sbi->ll_md_exp->exp_obd);
652 obd_disconnect(sbi->ll_md_exp);
653 sbi->ll_md_exp = NULL;
654 sbi->ll_md_obd = NULL;
663 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
667 size = sizeof(*lmmsize);
668 rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
669 KEY_MAX_EASIZE, &size, lmmsize);
671 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
672 sbi->ll_dt_exp->exp_obd->obd_name, rc);
677 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
678 KEY_MAX_EASIZE, &size, lmmsize);
680 CERROR("Get max mdsize error rc %d\n", rc);
686 * Get the value of the default_easize parameter.
688 * \see client_obd::cl_default_mds_easize
690 * \param[in] sbi superblock info for this filesystem
691 * \param[out] lmmsize pointer to storage location for value
693 * \retval 0 on success
694 * \retval negative negated errno on failure
696 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
701 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
702 KEY_DEFAULT_EASIZE, &size, lmmsize);
704 CERROR("Get default mdsize error rc %d\n", rc);
710 * Set the default_easize parameter to the given value.
712 * \see client_obd::cl_default_mds_easize
714 * \param[in] sbi superblock info for this filesystem
715 * \param[in] lmmsize the size to set
717 * \retval 0 on success
718 * \retval negative negated errno on failure
720 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
724 if (lmmsize < sizeof(struct lov_mds_md) ||
725 lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
728 rc = obd_set_info_async(NULL, sbi->ll_md_exp,
729 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
730 sizeof(int), &lmmsize, NULL);
735 static void client_common_put_super(struct super_block *sb)
737 struct ll_sb_info *sbi = ll_s2sbi(sb);
742 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
743 obd_disconnect(sbi->ll_dt_exp);
744 sbi->ll_dt_exp = NULL;
746 ll_debugfs_unregister_super(sb);
748 obd_fid_fini(sbi->ll_md_exp->exp_obd);
749 obd_disconnect(sbi->ll_md_exp);
750 sbi->ll_md_exp = NULL;
755 void ll_kill_super(struct super_block *sb)
757 struct ll_sb_info *sbi;
761 if (!(sb->s_flags & MS_ACTIVE))
765 /* we need restore s_dev from changed for clustred NFS before put_super
766 * because new kernels have cached s_dev and change sb->s_dev in
767 * put_super not affected real removing devices */
769 sb->s_dev = sbi->ll_sdev_orig;
770 sbi->ll_umounting = 1;
772 /* wait running statahead threads to quit */
773 while (atomic_read(&sbi->ll_sa_running) > 0) {
774 set_current_state(TASK_UNINTERRUPTIBLE);
775 schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
782 static inline int ll_set_opt(const char *opt, char *data, int fl)
784 if (strncmp(opt, data, strlen(opt)) != 0)
790 /* non-client-specific mount options are parsed in lmd_parse */
791 static int ll_options(char *options, struct ll_sb_info *sbi)
794 char *s1 = options, *s2;
795 int *flags = &sbi->ll_flags;
801 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
804 CDEBUG(D_SUPER, "next opt=%s\n", s1);
805 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
810 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
815 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
820 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
825 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
830 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
835 tmp = ll_set_opt("context", s1, 1);
838 tmp = ll_set_opt("fscontext", s1, 1);
841 tmp = ll_set_opt("defcontext", s1, 1);
844 tmp = ll_set_opt("rootcontext", s1, 1);
847 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
852 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
858 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
861 sbi->ll_checksum_set = 1;
864 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
867 sbi->ll_checksum_set = 1;
870 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
875 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
880 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
885 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
890 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
895 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
900 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
905 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
910 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
916 s2 = strchr(s1, ',');
924 void ll_lli_init(struct ll_inode_info *lli)
926 lli->lli_inode_magic = LLI_INODE_MAGIC;
928 spin_lock_init(&lli->lli_lock);
929 lli->lli_posix_acl = NULL;
930 /* Do not set lli_fid, it has been initialized already. */
931 fid_zero(&lli->lli_pfid);
932 lli->lli_mds_read_och = NULL;
933 lli->lli_mds_write_och = NULL;
934 lli->lli_mds_exec_och = NULL;
935 lli->lli_open_fd_read_count = 0;
936 lli->lli_open_fd_write_count = 0;
937 lli->lli_open_fd_exec_count = 0;
938 mutex_init(&lli->lli_och_mutex);
939 spin_lock_init(&lli->lli_agl_lock);
940 spin_lock_init(&lli->lli_layout_lock);
941 ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
942 lli->lli_clob = NULL;
944 init_rwsem(&lli->lli_xattrs_list_rwsem);
945 mutex_init(&lli->lli_xattrs_enq_lock);
947 LASSERT(lli->lli_vfs_inode.i_mode != 0);
948 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
949 mutex_init(&lli->lli_readdir_mutex);
950 lli->lli_opendir_key = NULL;
952 spin_lock_init(&lli->lli_sa_lock);
953 lli->lli_opendir_pid = 0;
954 lli->lli_sa_enabled = 0;
955 lli->lli_def_stripe_offset = -1;
956 init_rwsem(&lli->lli_lsm_sem);
958 mutex_init(&lli->lli_size_mutex);
959 lli->lli_symlink_name = NULL;
960 init_rwsem(&lli->lli_trunc_sem);
961 range_lock_tree_init(&lli->lli_write_tree);
962 init_rwsem(&lli->lli_glimpse_sem);
963 lli->lli_glimpse_time = ktime_set(0, 0);
964 INIT_LIST_HEAD(&lli->lli_agl_list);
965 lli->lli_agl_index = 0;
966 lli->lli_async_rc = 0;
968 mutex_init(&lli->lli_layout_mutex);
969 memset(lli->lli_jobid, 0, sizeof(lli->lli_jobid));
972 #define MAX_STRING_SIZE 128
974 #ifndef HAVE_SUPER_SETUP_BDI_NAME
976 #define LSI_BDI_INITIALIZED 0x00400000
978 #ifndef HAVE_BDI_CAP_MAP_COPY
979 # define BDI_CAP_MAP_COPY 0
982 static int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
984 struct lustre_sb_info *lsi = s2lsi(sb);
985 char buf[MAX_STRING_SIZE];
989 err = bdi_init(&lsi->lsi_bdi);
993 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
994 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
995 lsi->lsi_bdi.name = "lustre";
997 vsnprintf(buf, MAX_STRING_SIZE, fmt, args);
999 err = bdi_register(&lsi->lsi_bdi, NULL, "%s", buf);
1002 sb->s_bdi = &lsi->lsi_bdi;
1006 #endif /* !HAVE_SUPER_SETUP_BDI_NAME */
1008 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
1010 struct lustre_profile *lprof = NULL;
1011 struct lustre_sb_info *lsi = s2lsi(sb);
1012 struct ll_sb_info *sbi = NULL;
1013 char *dt = NULL, *md = NULL;
1014 char *profilenm = get_profile_name(sb);
1015 struct config_llog_instance *cfg;
1016 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1017 const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
1018 char name[MAX_STRING_SIZE];
1026 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1028 try_module_get(THIS_MODULE);
1032 GOTO(out_free, err = -ENOMEM);
1034 /* client additional sb info */
1035 lsi->lsi_llsbi = sbi = ll_init_sbi();
1037 GOTO(out_free, err = -ENOMEM);
1039 err = ll_options(lsi->lsi_lmd->lmd_opts, sbi);
1041 GOTO(out_free, err);
1043 err = super_setup_bdi_name(sb, "lustre-%p", sb);
1045 GOTO(out_free, err);
1047 #ifndef HAVE_DCACHE_LOCK
1048 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1049 sb->s_d_op = &ll_d_ops;
1052 len = strlen(lsi->lsi_lmd->lmd_profile);
1053 ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
1054 if (ptr && (strcmp(ptr, "-client") == 0))
1058 snprintf(name, MAX_STRING_SIZE, "%.*s-%p", len,
1059 lsi->lsi_lmd->lmd_profile, sb);
1061 /* Call ll_debugfs_register_super() before lustre_process_log()
1062 * so that "llite.*.*" params can be processed correctly.
1064 err = ll_debugfs_register_super(sb, name);
1066 CERROR("%s: could not register mountpoint in llite: rc = %d\n",
1067 ll_get_fsname(sb, NULL, 0), err);
1071 /* Generate a string unique to this super, in case some joker tries
1072 to mount the same fs at two mount points.
1073 Use the address of the super itself.*/
1074 cfg->cfg_instance = sb;
1075 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1076 cfg->cfg_callback = class_config_llog_handler;
1077 cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1078 /* set up client obds */
1079 err = lustre_process_log(sb, profilenm, cfg);
1081 GOTO(out_proc, err);
1083 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1084 lprof = class_get_profile(profilenm);
1085 if (lprof == NULL) {
1086 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1087 " read from the MGS. Does that filesystem "
1088 "exist?\n", profilenm);
1089 GOTO(out_proc, err = -EINVAL);
1091 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1092 lprof->lp_md, lprof->lp_dt);
1094 dt_len = strlen(lprof->lp_dt) + instlen + 2;
1095 OBD_ALLOC(dt, dt_len);
1097 GOTO(out_proc, err = -ENOMEM);
1098 snprintf(dt, dt_len - 1, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1100 md_len = strlen(lprof->lp_md) + instlen + 2;
1101 OBD_ALLOC(md, md_len);
1103 GOTO(out_proc, err = -ENOMEM);
1104 snprintf(md, md_len - 1, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1106 /* connections, registrations, sb setup */
1107 err = client_common_fill_super(sb, md, dt, mnt);
1109 GOTO(out_proc, err);
1111 sbi->ll_client_common_fill_super_succeeded = 1;
1115 ll_debugfs_unregister_super(sb);
1118 OBD_FREE(md, md_len);
1120 OBD_FREE(dt, dt_len);
1122 class_put_profile(lprof);
1127 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1128 LCONSOLE_WARN("Mounted %s\n", profilenm);
1130 } /* ll_fill_super */
1132 void ll_put_super(struct super_block *sb)
1134 struct config_llog_instance cfg, params_cfg;
1135 struct obd_device *obd;
1136 struct lustre_sb_info *lsi = s2lsi(sb);
1137 struct ll_sb_info *sbi = ll_s2sbi(sb);
1138 char *profilenm = get_profile_name(sb);
1140 int next, force = 1, rc = 0;
1144 GOTO(out_no_sbi, 0);
1146 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1148 cfg.cfg_instance = sb;
1149 lustre_end_log(sb, profilenm, &cfg);
1151 params_cfg.cfg_instance = sb;
1152 lustre_end_log(sb, PARAMS_FILENAME, ¶ms_cfg);
1154 if (sbi->ll_md_exp) {
1155 obd = class_exp2obd(sbi->ll_md_exp);
1157 force = obd->obd_force;
1160 /* Wait for unstable pages to be committed to stable storage */
1162 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1163 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1164 atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1168 ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1169 if (force == 0 && rc != -EINTR)
1170 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1173 /* We need to set force before the lov_disconnect in
1174 lustre_common_put_super, since l_d cleans up osc's as well. */
1177 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1179 obd->obd_force = force;
1183 if (sbi->ll_client_common_fill_super_succeeded) {
1184 /* Only if client_common_fill_super succeeded */
1185 client_common_put_super(sb);
1189 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1190 class_manual_cleanup(obd);
1193 if (sbi->ll_flags & LL_SBI_VERBOSE)
1194 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1197 class_del_profile(profilenm);
1199 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1200 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1201 bdi_destroy(&lsi->lsi_bdi);
1202 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1207 lsi->lsi_llsbi = NULL;
1209 lustre_common_put_super(sb);
1211 cl_env_cache_purge(~0);
1213 module_put(THIS_MODULE);
1216 } /* client_put_super */
1218 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1220 struct inode *inode = NULL;
1222 /* NOTE: we depend on atomic igrab() -bzzz */
1223 lock_res_and_lock(lock);
1224 if (lock->l_resource->lr_lvb_inode) {
1225 struct ll_inode_info * lli;
1226 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1227 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1228 inode = igrab(lock->l_resource->lr_lvb_inode);
1230 inode = lock->l_resource->lr_lvb_inode;
1231 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1232 D_WARNING, lock, "lr_lvb_inode %p is "
1233 "bogus: magic %08x",
1234 lock->l_resource->lr_lvb_inode,
1235 lli->lli_inode_magic);
1239 unlock_res_and_lock(lock);
1243 void ll_dir_clear_lsm_md(struct inode *inode)
1245 struct ll_inode_info *lli = ll_i2info(inode);
1247 LASSERT(S_ISDIR(inode->i_mode));
1249 if (lli->lli_lsm_md != NULL) {
1250 lmv_free_memmd(lli->lli_lsm_md);
1251 lli->lli_lsm_md = NULL;
1255 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1256 const struct lu_fid *fid,
1257 struct lustre_md *md)
1259 struct ll_sb_info *sbi = ll_s2sbi(sb);
1260 struct mdt_body *body = md->body;
1261 struct inode *inode;
1265 ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1266 inode = iget_locked(sb, ino);
1267 if (inode == NULL) {
1268 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1269 ll_get_fsname(sb, NULL, 0), PFID(fid));
1270 RETURN(ERR_PTR(-ENOENT));
1273 if (inode->i_state & I_NEW) {
1274 struct ll_inode_info *lli = ll_i2info(inode);
1275 struct lmv_stripe_md *lsm = md->lmv;
1277 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1278 (body->mbo_mode & S_IFMT);
1279 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1282 LTIME_S(inode->i_mtime) = 0;
1283 LTIME_S(inode->i_atime) = 0;
1284 LTIME_S(inode->i_ctime) = 0;
1287 #ifdef HAVE_BACKING_DEV_INFO
1288 /* initializing backing dev info. */
1289 inode->i_mapping->backing_dev_info =
1290 &s2lsi(inode->i_sb)->lsi_bdi;
1292 inode->i_op = &ll_dir_inode_operations;
1293 inode->i_fop = &ll_dir_operations;
1294 lli->lli_fid = *fid;
1297 LASSERT(lsm != NULL);
1298 /* master object FID */
1299 lli->lli_pfid = body->mbo_fid1;
1300 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1301 lli, PFID(fid), PFID(&lli->lli_pfid));
1302 unlock_new_inode(inode);
1308 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1311 struct lmv_stripe_md *lsm = md->lmv;
1312 struct ll_inode_info *lli = ll_i2info(inode);
1315 LASSERT(lsm != NULL);
1317 CDEBUG(D_INODE, "%s: "DFID" set dir layout:\n",
1318 ll_get_fsname(inode->i_sb, NULL, 0),
1319 PFID(&lli->lli_fid));
1320 lsm_md_dump(D_INODE, lsm);
1322 /* XXX sigh, this lsm_root initialization should be in
1323 * LMV layer, but it needs ll_iget right now, so we
1324 * put this here right now. */
1325 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1326 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1327 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1328 /* Unfortunately ll_iget will call ll_update_inode,
1329 * where the initialization of slave inode is slightly
1330 * different, so it reset lsm_md to NULL to avoid
1331 * initializing lsm for slave inode. */
1332 lsm->lsm_md_oinfo[i].lmo_root =
1333 ll_iget_anon_dir(inode->i_sb, fid, md);
1334 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1335 int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1337 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1339 iput(lsm->lsm_md_oinfo[i].lmo_root);
1340 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1346 lli->lli_lsm_md = lsm;
1351 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1353 struct ll_inode_info *lli = ll_i2info(inode);
1354 struct lmv_stripe_md *lsm = md->lmv;
1359 LASSERT(S_ISDIR(inode->i_mode));
1360 CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1361 PFID(ll_inode2fid(inode)));
1364 * no striped information from request, lustre_md from req does not
1365 * include stripeEA, see ll_md_setattr()
1371 * normally dir layout doesn't change, only take read lock to check
1372 * that to avoid blocking other MD operations.
1374 if (lli->lli_lsm_md)
1375 down_read(&lli->lli_lsm_sem);
1377 down_write(&lli->lli_lsm_sem);
1380 * if dir layout mismatch, check whether version is increased, which
1381 * means layout is changed, this happens in dir migration and lfsck.
1383 if (lli->lli_lsm_md && !lsm_md_eq(lli->lli_lsm_md, lsm)) {
1384 if (lsm->lsm_md_layout_version <=
1385 lli->lli_lsm_md->lsm_md_layout_version) {
1386 CERROR("%s: "DFID" dir layout mismatch:\n",
1387 ll_get_fsname(inode->i_sb, NULL, 0),
1388 PFID(&lli->lli_fid));
1389 lsm_md_dump(D_ERROR, lli->lli_lsm_md);
1390 lsm_md_dump(D_ERROR, lsm);
1391 GOTO(unlock, rc = -EINVAL);
1394 /* layout changed, switch to write lock */
1395 up_read(&lli->lli_lsm_sem);
1396 down_write(&lli->lli_lsm_sem);
1397 ll_dir_clear_lsm_md(inode);
1400 /* set directory layout */
1401 if (!lli->lli_lsm_md) {
1402 struct cl_attr *attr;
1404 rc = ll_init_lsm_md(inode, md);
1405 up_write(&lli->lli_lsm_sem);
1409 /* set md->lmv to NULL, so the following free lustre_md
1410 * will not free this lsm */
1414 * md_merge_attr() may take long, since lsm is already set,
1415 * switch to read lock.
1417 down_read(&lli->lli_lsm_sem);
1419 OBD_ALLOC_PTR(attr);
1421 GOTO(unlock, rc = -ENOMEM);
1423 /* validate the lsm */
1424 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1425 ll_md_blocking_ast);
1431 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1432 md->body->mbo_nlink = attr->cat_nlink;
1433 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1434 md->body->mbo_size = attr->cat_size;
1435 if (md->body->mbo_valid & OBD_MD_FLATIME)
1436 md->body->mbo_atime = attr->cat_atime;
1437 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1438 md->body->mbo_ctime = attr->cat_ctime;
1439 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1440 md->body->mbo_mtime = attr->cat_mtime;
1445 up_read(&lli->lli_lsm_sem);
1450 void ll_clear_inode(struct inode *inode)
1452 struct ll_inode_info *lli = ll_i2info(inode);
1453 struct ll_sb_info *sbi = ll_i2sbi(inode);
1456 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1457 PFID(ll_inode2fid(inode)), inode);
1459 if (S_ISDIR(inode->i_mode)) {
1460 /* these should have been cleared in ll_file_release */
1461 LASSERT(lli->lli_opendir_key == NULL);
1462 LASSERT(lli->lli_sai == NULL);
1463 LASSERT(lli->lli_opendir_pid == 0);
1466 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1468 LASSERT(!lli->lli_open_fd_write_count);
1469 LASSERT(!lli->lli_open_fd_read_count);
1470 LASSERT(!lli->lli_open_fd_exec_count);
1472 if (lli->lli_mds_write_och)
1473 ll_md_real_close(inode, FMODE_WRITE);
1474 if (lli->lli_mds_exec_och)
1475 ll_md_real_close(inode, FMODE_EXEC);
1476 if (lli->lli_mds_read_och)
1477 ll_md_real_close(inode, FMODE_READ);
1479 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1480 OBD_FREE(lli->lli_symlink_name,
1481 strlen(lli->lli_symlink_name) + 1);
1482 lli->lli_symlink_name = NULL;
1485 ll_xattr_cache_destroy(inode);
1487 #ifdef CONFIG_FS_POSIX_ACL
1488 forget_all_cached_acls(inode);
1489 if (lli->lli_posix_acl) {
1490 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1491 posix_acl_release(lli->lli_posix_acl);
1492 lli->lli_posix_acl = NULL;
1495 lli->lli_inode_magic = LLI_INODE_DEAD;
1497 if (S_ISDIR(inode->i_mode))
1498 ll_dir_clear_lsm_md(inode);
1499 else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1500 LASSERT(list_empty(&lli->lli_agl_list));
1503 * XXX This has to be done before lsm is freed below, because
1504 * cl_object still uses inode lsm.
1506 cl_inode_fini(inode);
1511 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1513 struct lustre_md md;
1514 struct inode *inode = dentry->d_inode;
1515 struct ll_sb_info *sbi = ll_i2sbi(inode);
1516 struct ptlrpc_request *request = NULL;
1520 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1521 LUSTRE_OPC_ANY, NULL);
1522 if (IS_ERR(op_data))
1523 RETURN(PTR_ERR(op_data));
1525 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1527 ptlrpc_req_finished(request);
1528 if (rc == -ENOENT) {
1530 /* Unlinked special device node? Or just a race?
1531 * Pretend we done everything. */
1532 if (!S_ISREG(inode->i_mode) &&
1533 !S_ISDIR(inode->i_mode)) {
1534 ia_valid = op_data->op_attr.ia_valid;
1535 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1536 rc = simple_setattr(dentry, &op_data->op_attr);
1537 op_data->op_attr.ia_valid = ia_valid;
1539 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1540 CERROR("md_setattr fails: rc = %d\n", rc);
1545 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1546 sbi->ll_md_exp, &md);
1548 ptlrpc_req_finished(request);
1552 ia_valid = op_data->op_attr.ia_valid;
1553 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1554 * cache is not cleared yet. */
1555 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1556 if (S_ISREG(inode->i_mode))
1558 rc = simple_setattr(dentry, &op_data->op_attr);
1559 if (S_ISREG(inode->i_mode))
1560 inode_unlock(inode);
1561 op_data->op_attr.ia_valid = ia_valid;
1563 rc = ll_update_inode(inode, &md);
1564 ptlrpc_req_finished(request);
1569 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1570 * object(s) determine the file size and mtime. Otherwise, the MDS will
1571 * keep these values until such a time that objects are allocated for it.
1572 * We do the MDS operations first, as it is checking permissions for us.
1573 * We don't to the MDS RPC if there is nothing that we want to store there,
1574 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1575 * going to do an RPC anyways.
1577 * If we are doing a truncate, we will send the mtime and ctime updates
1578 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1579 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1582 * In case of HSMimport, we only set attr on MDS.
1584 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr,
1585 enum op_xvalid xvalid, bool hsm_import)
1587 struct inode *inode = dentry->d_inode;
1588 struct ll_inode_info *lli = ll_i2info(inode);
1589 struct md_op_data *op_data = NULL;
1593 CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1594 "valid %x, hsm_import %d\n",
1595 ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1596 inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1599 if (attr->ia_valid & ATTR_SIZE) {
1600 /* Check new size against VFS/VM file size limit and rlimit */
1601 rc = inode_newsize_ok(inode, attr->ia_size);
1605 /* The maximum Lustre file size is variable, based on the
1606 * OST maximum object size and number of stripes. This
1607 * needs another check in addition to the VFS check above. */
1608 if (attr->ia_size > ll_file_maxbytes(inode)) {
1609 CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
1610 PFID(&lli->lli_fid), attr->ia_size,
1611 ll_file_maxbytes(inode));
1615 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1618 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1619 if (attr->ia_valid & TIMES_SET_FLAGS) {
1620 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1621 !cfs_capable(CFS_CAP_FOWNER))
1625 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1626 if (!(xvalid & OP_XVALID_CTIME_SET) &&
1627 (attr->ia_valid & ATTR_CTIME)) {
1628 attr->ia_ctime = current_time(inode);
1629 xvalid |= OP_XVALID_CTIME_SET;
1631 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1632 (attr->ia_valid & ATTR_ATIME)) {
1633 attr->ia_atime = current_time(inode);
1634 attr->ia_valid |= ATTR_ATIME_SET;
1636 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1637 (attr->ia_valid & ATTR_MTIME)) {
1638 attr->ia_mtime = current_time(inode);
1639 attr->ia_valid |= ATTR_MTIME_SET;
1642 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1643 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %llu\n",
1644 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1645 (s64)ktime_get_real_seconds());
1647 if (S_ISREG(inode->i_mode)) {
1648 if (attr->ia_valid & ATTR_SIZE)
1649 inode_dio_write_done(inode);
1650 inode_unlock(inode);
1653 /* We always do an MDS RPC, even if we're only changing the size;
1654 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1656 OBD_ALLOC_PTR(op_data);
1657 if (op_data == NULL)
1658 GOTO(out, rc = -ENOMEM);
1660 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1661 /* If we are changing file size, file content is
1662 * modified, flag it.
1664 xvalid |= OP_XVALID_OWNEROVERRIDE;
1665 op_data->op_bias |= MDS_DATA_MODIFIED;
1666 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1669 if (attr->ia_valid & ATTR_FILE) {
1670 struct ll_file_data *fd = LUSTRE_FPRIVATE(attr->ia_file);
1672 if (fd->fd_lease_och)
1673 op_data->op_bias |= MDS_TRUNC_KEEP_LEASE;
1676 op_data->op_attr = *attr;
1677 op_data->op_xvalid = xvalid;
1679 rc = ll_md_setattr(dentry, op_data);
1683 if (!S_ISREG(inode->i_mode) || hsm_import)
1686 if (attr->ia_valid & (ATTR_SIZE | ATTR_ATIME | ATTR_ATIME_SET |
1687 ATTR_MTIME | ATTR_MTIME_SET | ATTR_CTIME) ||
1688 xvalid & OP_XVALID_CTIME_SET) {
1689 /* For truncate and utimes sending attributes to OSTs, setting
1690 * mtime/atime to the past will be performed under PW [0:EOF]
1691 * extent lock (new_size:EOF for truncate). It may seem
1692 * excessive to send mtime/atime updates to OSTs when not
1693 * setting times to past, but it is necessary due to possible
1694 * time de-synchronization between MDT inode and OST objects
1696 rc = cl_setattr_ost(lli->lli_clob, attr, xvalid, 0);
1699 /* If the file was restored, it needs to set dirty flag.
1701 * We've already sent MDS_DATA_MODIFIED flag in
1702 * ll_md_setattr() for truncate. However, the MDT refuses to
1703 * set the HS_DIRTY flag on released files, so we have to set
1704 * it again if the file has been restored. Please check how
1705 * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1707 * Please notice that if the file is not released, the previous
1708 * MDS_DATA_MODIFIED has taken effect and usually
1709 * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1710 * This way we can save an RPC for common open + trunc
1712 if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1713 struct hsm_state_set hss = {
1714 .hss_valid = HSS_SETMASK,
1715 .hss_setmask = HS_DIRTY,
1719 rc2 = ll_hsm_state_set(inode, &hss);
1720 /* truncate and write can happen at the same time, so that
1721 * the file can be set modified even though the file is not
1722 * restored from released state, and ll_hsm_state_set() is
1723 * not applicable for the file, and rc2 < 0 is normal in this
1726 CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1727 PFID(ll_inode2fid(inode)), rc2);
1732 if (op_data != NULL)
1733 ll_finish_md_op_data(op_data);
1735 if (S_ISREG(inode->i_mode)) {
1737 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1738 inode_dio_wait(inode);
1739 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
1740 * flag. ll_update_inode (called from ll_md_setattr), clears
1741 * inode flags, so there is a gap where S_NOSEC is not set.
1742 * This can cause a writer to take the i_mutex unnecessarily,
1743 * but this is safe to do and should be rare. */
1744 inode_has_no_xattr(inode);
1747 ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1748 LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1753 int ll_setattr(struct dentry *de, struct iattr *attr)
1755 int mode = de->d_inode->i_mode;
1756 enum op_xvalid xvalid = 0;
1758 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1759 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1760 xvalid |= OP_XVALID_OWNEROVERRIDE;
1762 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1763 (ATTR_SIZE|ATTR_MODE)) &&
1764 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1765 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1766 !(attr->ia_mode & S_ISGID))))
1767 attr->ia_valid |= ATTR_FORCE;
1769 if ((attr->ia_valid & ATTR_MODE) &&
1771 !(attr->ia_mode & S_ISUID) &&
1772 !(attr->ia_valid & ATTR_KILL_SUID))
1773 attr->ia_valid |= ATTR_KILL_SUID;
1775 if ((attr->ia_valid & ATTR_MODE) &&
1776 ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1777 !(attr->ia_mode & S_ISGID) &&
1778 !(attr->ia_valid & ATTR_KILL_SGID))
1779 attr->ia_valid |= ATTR_KILL_SGID;
1781 return ll_setattr_raw(de, attr, xvalid, false);
1784 int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
1787 struct obd_statfs obd_osfs;
1792 max_age = ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS;
1794 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1796 CERROR("md_statfs fails: rc = %d\n", rc);
1800 osfs->os_type = LL_SUPER_MAGIC;
1802 CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1803 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1805 if (osfs->os_state & OS_STATE_SUM)
1808 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1809 flags |= OBD_STATFS_NODELAY;
1811 rc = obd_statfs(NULL, sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1813 CERROR("obd_statfs fails: rc = %d\n", rc);
1817 CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1818 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1821 osfs->os_bsize = obd_osfs.os_bsize;
1822 osfs->os_blocks = obd_osfs.os_blocks;
1823 osfs->os_bfree = obd_osfs.os_bfree;
1824 osfs->os_bavail = obd_osfs.os_bavail;
1826 /* If we don't have as many objects free on the OST as inodes
1827 * on the MDS, we reduce the total number of inodes to
1828 * compensate, so that the "inodes in use" number is correct.
1830 if (obd_osfs.os_ffree < osfs->os_ffree) {
1831 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1833 osfs->os_ffree = obd_osfs.os_ffree;
1839 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1841 struct super_block *sb = de->d_sb;
1842 struct obd_statfs osfs;
1843 __u64 fsid = huge_encode_dev(sb->s_dev);
1846 CDEBUG(D_VFSTRACE, "VFS Op: at %llu jiffies\n", get_jiffies_64());
1847 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1849 /* Some amount of caching on the client is allowed */
1850 rc = ll_statfs_internal(ll_s2sbi(sb), &osfs, OBD_STATFS_SUM);
1854 statfs_unpack(sfs, &osfs);
1856 /* We need to downshift for all 32-bit kernels, because we can't
1857 * tell if the kernel is being called via sys_statfs64() or not.
1858 * Stop before overflowing f_bsize - in which case it is better
1859 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1860 if (sizeof(long) < 8) {
1861 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1864 osfs.os_blocks >>= 1;
1865 osfs.os_bfree >>= 1;
1866 osfs.os_bavail >>= 1;
1870 sfs->f_blocks = osfs.os_blocks;
1871 sfs->f_bfree = osfs.os_bfree;
1872 sfs->f_bavail = osfs.os_bavail;
1873 sfs->f_fsid.val[0] = (__u32)fsid;
1874 sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1878 void ll_inode_size_lock(struct inode *inode)
1880 struct ll_inode_info *lli;
1882 LASSERT(!S_ISDIR(inode->i_mode));
1884 lli = ll_i2info(inode);
1885 mutex_lock(&lli->lli_size_mutex);
1888 void ll_inode_size_unlock(struct inode *inode)
1890 struct ll_inode_info *lli;
1892 lli = ll_i2info(inode);
1893 mutex_unlock(&lli->lli_size_mutex);
1896 void ll_update_inode_flags(struct inode *inode, int ext_flags)
1898 inode->i_flags = ll_ext_to_inode_flags(ext_flags);
1899 if (ext_flags & LUSTRE_PROJINHERIT_FL)
1900 ll_file_set_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
1902 ll_file_clear_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
1905 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1907 struct ll_inode_info *lli = ll_i2info(inode);
1908 struct mdt_body *body = md->body;
1909 struct ll_sb_info *sbi = ll_i2sbi(inode);
1912 if (body->mbo_valid & OBD_MD_FLEASIZE) {
1913 rc = cl_file_inode_init(inode, md);
1918 if (S_ISDIR(inode->i_mode)) {
1919 rc = ll_update_lsm_md(inode, md);
1924 #ifdef CONFIG_FS_POSIX_ACL
1925 if (body->mbo_valid & OBD_MD_FLACL) {
1926 spin_lock(&lli->lli_lock);
1927 if (lli->lli_posix_acl)
1928 posix_acl_release(lli->lli_posix_acl);
1929 lli->lli_posix_acl = md->posix_acl;
1930 spin_unlock(&lli->lli_lock);
1933 inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1934 sbi->ll_flags & LL_SBI_32BIT_API);
1935 inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1937 if (body->mbo_valid & OBD_MD_FLATIME) {
1938 if (body->mbo_atime > LTIME_S(inode->i_atime))
1939 LTIME_S(inode->i_atime) = body->mbo_atime;
1940 lli->lli_atime = body->mbo_atime;
1943 if (body->mbo_valid & OBD_MD_FLMTIME) {
1944 if (body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1945 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1946 "to %llu\n", inode->i_ino,
1947 LTIME_S(inode->i_mtime), body->mbo_mtime);
1948 LTIME_S(inode->i_mtime) = body->mbo_mtime;
1950 lli->lli_mtime = body->mbo_mtime;
1953 if (body->mbo_valid & OBD_MD_FLCTIME) {
1954 if (body->mbo_ctime > LTIME_S(inode->i_ctime))
1955 LTIME_S(inode->i_ctime) = body->mbo_ctime;
1956 lli->lli_ctime = body->mbo_ctime;
1959 /* Clear i_flags to remove S_NOSEC before permissions are updated */
1960 if (body->mbo_valid & OBD_MD_FLFLAGS)
1961 ll_update_inode_flags(inode, body->mbo_flags);
1962 if (body->mbo_valid & OBD_MD_FLMODE)
1963 inode->i_mode = (inode->i_mode & S_IFMT) |
1964 (body->mbo_mode & ~S_IFMT);
1966 if (body->mbo_valid & OBD_MD_FLTYPE)
1967 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1968 (body->mbo_mode & S_IFMT);
1970 LASSERT(inode->i_mode != 0);
1971 if (S_ISREG(inode->i_mode))
1972 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1973 LL_MAX_BLKSIZE_BITS);
1975 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1977 if (body->mbo_valid & OBD_MD_FLUID)
1978 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
1979 if (body->mbo_valid & OBD_MD_FLGID)
1980 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
1981 if (body->mbo_valid & OBD_MD_FLPROJID)
1982 lli->lli_projid = body->mbo_projid;
1983 if (body->mbo_valid & OBD_MD_FLNLINK)
1984 set_nlink(inode, body->mbo_nlink);
1985 if (body->mbo_valid & OBD_MD_FLRDEV)
1986 inode->i_rdev = old_decode_dev(body->mbo_rdev);
1988 if (body->mbo_valid & OBD_MD_FLID) {
1989 /* FID shouldn't be changed! */
1990 if (fid_is_sane(&lli->lli_fid)) {
1991 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
1992 "Trying to change FID "DFID
1993 " to the "DFID", inode "DFID"(%p)\n",
1994 PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
1995 PFID(ll_inode2fid(inode)), inode);
1997 lli->lli_fid = body->mbo_fid1;
2001 LASSERT(fid_seq(&lli->lli_fid) != 0);
2003 if (body->mbo_valid & OBD_MD_FLSIZE) {
2004 i_size_write(inode, body->mbo_size);
2006 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
2007 PFID(ll_inode2fid(inode)),
2008 (unsigned long long)body->mbo_size);
2010 if (body->mbo_valid & OBD_MD_FLBLOCKS)
2011 inode->i_blocks = body->mbo_blocks;
2014 if (body->mbo_valid & OBD_MD_TSTATE) {
2015 /* Set LLIF_FILE_RESTORING if restore ongoing and
2016 * clear it when done to ensure to start again
2017 * glimpsing updated attrs
2019 if (body->mbo_t_state & MS_RESTORE)
2020 ll_file_set_flag(lli, LLIF_FILE_RESTORING);
2022 ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
2028 int ll_read_inode2(struct inode *inode, void *opaque)
2030 struct lustre_md *md = opaque;
2031 struct ll_inode_info *lli = ll_i2info(inode);
2035 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2036 PFID(&lli->lli_fid), inode);
2038 /* Core attributes from the MDS first. This is a new inode, and
2039 * the VFS doesn't zero times in the core inode so we have to do
2040 * it ourselves. They will be overwritten by either MDS or OST
2041 * attributes - we just need to make sure they aren't newer. */
2042 LTIME_S(inode->i_mtime) = 0;
2043 LTIME_S(inode->i_atime) = 0;
2044 LTIME_S(inode->i_ctime) = 0;
2046 rc = ll_update_inode(inode, md);
2050 /* OIDEBUG(inode); */
2052 #ifdef HAVE_BACKING_DEV_INFO
2053 /* initializing backing dev info. */
2054 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2056 if (S_ISREG(inode->i_mode)) {
2057 struct ll_sb_info *sbi = ll_i2sbi(inode);
2058 inode->i_op = &ll_file_inode_operations;
2059 inode->i_fop = sbi->ll_fop;
2060 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2062 } else if (S_ISDIR(inode->i_mode)) {
2063 inode->i_op = &ll_dir_inode_operations;
2064 inode->i_fop = &ll_dir_operations;
2066 } else if (S_ISLNK(inode->i_mode)) {
2067 inode->i_op = &ll_fast_symlink_inode_operations;
2070 inode->i_op = &ll_special_inode_operations;
2072 init_special_inode(inode, inode->i_mode,
2081 void ll_delete_inode(struct inode *inode)
2083 struct ll_inode_info *lli = ll_i2info(inode);
2086 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2087 /* It is last chance to write out dirty pages,
2088 * otherwise we may lose data while umount */
2089 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
2091 truncate_inode_pages_final(&inode->i_data);
2093 LASSERTF(inode->i_data.nrpages == 0, "inode="DFID"(%p) nrpages=%lu, "
2094 "see https://jira.whamcloud.com/browse/LU-118\n",
2095 PFID(ll_inode2fid(inode)), inode, inode->i_data.nrpages);
2097 #ifdef HAVE_SBOPS_EVICT_INODE
2098 ll_clear_inode(inode);
2105 int ll_iocontrol(struct inode *inode, struct file *file,
2106 unsigned int cmd, unsigned long arg)
2108 struct ll_sb_info *sbi = ll_i2sbi(inode);
2109 struct ptlrpc_request *req = NULL;
2114 case FS_IOC_GETFLAGS: {
2115 struct mdt_body *body;
2116 struct md_op_data *op_data;
2118 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2119 0, 0, LUSTRE_OPC_ANY,
2121 if (IS_ERR(op_data))
2122 RETURN(PTR_ERR(op_data));
2124 op_data->op_valid = OBD_MD_FLFLAGS;
2125 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2126 ll_finish_md_op_data(op_data);
2128 CERROR("%s: failure inode "DFID": rc = %d\n",
2129 sbi->ll_md_exp->exp_obd->obd_name,
2130 PFID(ll_inode2fid(inode)), rc);
2134 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2136 flags = body->mbo_flags;
2138 ptlrpc_req_finished(req);
2140 RETURN(put_user(flags, (int __user *)arg));
2142 case FS_IOC_SETFLAGS: {
2144 struct md_op_data *op_data;
2145 struct cl_object *obj;
2146 struct fsxattr fa = { 0 };
2148 if (get_user(flags, (int __user *)arg))
2151 fa.fsx_projid = ll_i2info(inode)->lli_projid;
2152 if (flags & LUSTRE_PROJINHERIT_FL)
2153 fa.fsx_xflags = FS_XFLAG_PROJINHERIT;
2155 rc = ll_ioctl_check_project(inode, &fa);
2159 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2160 LUSTRE_OPC_ANY, NULL);
2161 if (IS_ERR(op_data))
2162 RETURN(PTR_ERR(op_data));
2164 op_data->op_attr_flags = flags;
2165 op_data->op_xvalid |= OP_XVALID_FLAGS;
2166 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2167 ll_finish_md_op_data(op_data);
2168 ptlrpc_req_finished(req);
2172 ll_update_inode_flags(inode, flags);
2174 obj = ll_i2info(inode)->lli_clob;
2178 OBD_ALLOC_PTR(attr);
2182 rc = cl_setattr_ost(obj, attr, OP_XVALID_FLAGS, flags);
2194 int ll_flush_ctx(struct inode *inode)
2196 struct ll_sb_info *sbi = ll_i2sbi(inode);
2198 CDEBUG(D_SEC, "flush context for user %d\n",
2199 from_kuid(&init_user_ns, current_uid()));
2201 obd_set_info_async(NULL, sbi->ll_md_exp,
2202 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2204 obd_set_info_async(NULL, sbi->ll_dt_exp,
2205 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2210 /* umount -f client means force down, don't save state */
2211 void ll_umount_begin(struct super_block *sb)
2213 struct ll_sb_info *sbi = ll_s2sbi(sb);
2214 struct obd_device *obd;
2215 struct obd_ioctl_data *ioc_data;
2216 struct l_wait_info lwi;
2217 wait_queue_head_t waitq;
2220 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2221 sb->s_count, atomic_read(&sb->s_active));
2223 obd = class_exp2obd(sbi->ll_md_exp);
2225 CERROR("Invalid MDC connection handle %#llx\n",
2226 sbi->ll_md_exp->exp_handle.h_cookie);
2232 obd = class_exp2obd(sbi->ll_dt_exp);
2234 CERROR("Invalid LOV connection handle %#llx\n",
2235 sbi->ll_dt_exp->exp_handle.h_cookie);
2241 OBD_ALLOC_PTR(ioc_data);
2243 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2244 sizeof *ioc_data, ioc_data, NULL);
2246 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2247 sizeof *ioc_data, ioc_data, NULL);
2249 OBD_FREE_PTR(ioc_data);
2252 /* Really, we'd like to wait until there are no requests outstanding,
2253 * and then continue. For now, we just periodically checking for vfs
2254 * to decrement mnt_cnt and hope to finish it within 10sec.
2256 init_waitqueue_head(&waitq);
2257 lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(10),
2258 cfs_time_seconds(1), NULL, NULL);
2259 l_wait_event(waitq, may_umount(sbi->ll_mnt.mnt), &lwi);
2264 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2266 struct ll_sb_info *sbi = ll_s2sbi(sb);
2267 char *profilenm = get_profile_name(sb);
2271 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2272 read_only = *flags & MS_RDONLY;
2273 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2274 sizeof(KEY_READ_ONLY),
2275 KEY_READ_ONLY, sizeof(read_only),
2278 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2279 profilenm, read_only ?
2280 "read-only" : "read-write", err);
2285 sb->s_flags |= MS_RDONLY;
2287 sb->s_flags &= ~MS_RDONLY;
2289 if (sbi->ll_flags & LL_SBI_VERBOSE)
2290 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2291 read_only ? "read-only" : "read-write");
2297 * Cleanup the open handle that is cached on MDT-side.
2299 * For open case, the client side open handling thread may hit error
2300 * after the MDT grant the open. Under such case, the client should
2301 * send close RPC to the MDT as cleanup; otherwise, the open handle
2302 * on the MDT will be leaked there until the client umount or evicted.
2304 * In further, if someone unlinked the file, because the open handle
2305 * holds the reference on such file/object, then it will block the
2306 * subsequent threads that want to locate such object via FID.
2308 * \param[in] sb super block for this file-system
2309 * \param[in] open_req pointer to the original open request
2311 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2313 struct mdt_body *body;
2314 struct md_op_data *op_data;
2315 struct ptlrpc_request *close_req = NULL;
2316 struct obd_export *exp = ll_s2sbi(sb)->ll_md_exp;
2319 body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2320 OBD_ALLOC_PTR(op_data);
2321 if (op_data == NULL) {
2322 CWARN("%s: cannot allocate op_data to release open handle for "
2324 ll_get_fsname(sb, NULL, 0), PFID(&body->mbo_fid1));
2329 op_data->op_fid1 = body->mbo_fid1;
2330 op_data->op_open_handle = body->mbo_open_handle;
2331 op_data->op_mod_time = ktime_get_real_seconds();
2332 md_close(exp, op_data, NULL, &close_req);
2333 ptlrpc_req_finished(close_req);
2334 ll_finish_md_op_data(op_data);
2339 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2340 struct super_block *sb, struct lookup_intent *it)
2342 struct ll_sb_info *sbi = NULL;
2343 struct lustre_md md = { NULL };
2347 LASSERT(*inode || sb);
2348 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2349 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2350 sbi->ll_md_exp, &md);
2355 rc = ll_update_inode(*inode, &md);
2359 LASSERT(sb != NULL);
2362 * At this point server returns to client's same fid as client
2363 * generated for creating. So using ->fid1 is okay here.
2365 if (!fid_is_sane(&md.body->mbo_fid1)) {
2366 CERROR("%s: Fid is insane "DFID"\n",
2367 ll_get_fsname(sb, NULL, 0),
2368 PFID(&md.body->mbo_fid1));
2369 GOTO(out, rc = -EINVAL);
2372 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2373 sbi->ll_flags & LL_SBI_32BIT_API),
2375 if (IS_ERR(*inode)) {
2376 #ifdef CONFIG_FS_POSIX_ACL
2378 posix_acl_release(md.posix_acl);
2379 md.posix_acl = NULL;
2382 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2384 CERROR("new_inode -fatal: rc %d\n", rc);
2389 /* Handling piggyback layout lock.
2390 * Layout lock can be piggybacked by getattr and open request.
2391 * The lsm can be applied to inode only if it comes with a layout lock
2392 * otherwise correct layout may be overwritten, for example:
2393 * 1. proc1: mdt returns a lsm but not granting layout
2394 * 2. layout was changed by another client
2395 * 3. proc2: refresh layout and layout lock granted
2396 * 4. proc1: to apply a stale layout */
2397 if (it != NULL && it->it_lock_mode != 0) {
2398 struct lustre_handle lockh;
2399 struct ldlm_lock *lock;
2401 lockh.cookie = it->it_lock_handle;
2402 lock = ldlm_handle2lock(&lockh);
2403 LASSERT(lock != NULL);
2404 if (ldlm_has_layout(lock)) {
2405 struct cl_object_conf conf;
2407 memset(&conf, 0, sizeof(conf));
2408 conf.coc_opc = OBJECT_CONF_SET;
2409 conf.coc_inode = *inode;
2410 conf.coc_lock = lock;
2411 conf.u.coc_layout = md.layout;
2412 (void)ll_layout_conf(*inode, &conf);
2414 LDLM_LOCK_PUT(lock);
2420 md_free_lustre_md(sbi->ll_md_exp, &md);
2423 if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2424 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2429 int ll_obd_statfs(struct inode *inode, void __user *arg)
2431 struct ll_sb_info *sbi = NULL;
2432 struct obd_export *exp;
2434 struct obd_ioctl_data *data = NULL;
2438 if (!inode || !(sbi = ll_i2sbi(inode)))
2439 GOTO(out_statfs, rc = -EINVAL);
2441 rc = obd_ioctl_getdata(&buf, &len, arg);
2443 GOTO(out_statfs, rc);
2446 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2447 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2448 GOTO(out_statfs, rc = -EINVAL);
2450 if (data->ioc_inllen1 != sizeof(__u32) ||
2451 data->ioc_inllen2 != sizeof(__u32) ||
2452 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2453 data->ioc_plen2 != sizeof(struct obd_uuid))
2454 GOTO(out_statfs, rc = -EINVAL);
2456 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2457 if (type & LL_STATFS_LMV)
2458 exp = sbi->ll_md_exp;
2459 else if (type & LL_STATFS_LOV)
2460 exp = sbi->ll_dt_exp;
2462 GOTO(out_statfs, rc = -ENODEV);
2464 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2466 GOTO(out_statfs, rc);
2468 OBD_FREE_LARGE(buf, len);
2473 * this is normally called in ll_fini_md_op_data(), but sometimes it needs to
2474 * be called early to avoid deadlock.
2476 void ll_unlock_md_op_lsm(struct md_op_data *op_data)
2478 if (op_data->op_mea2_sem) {
2479 up_read(op_data->op_mea2_sem);
2480 op_data->op_mea2_sem = NULL;
2483 if (op_data->op_mea1_sem) {
2484 up_read(op_data->op_mea1_sem);
2485 op_data->op_mea1_sem = NULL;
2489 /* this function prepares md_op_data hint for passing it down to MD stack. */
2490 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2491 struct inode *i1, struct inode *i2,
2492 const char *name, size_t namelen,
2493 __u32 mode, __u32 opc, void *data)
2495 LASSERT(i1 != NULL);
2498 /* Do not reuse namelen for something else. */
2500 return ERR_PTR(-EINVAL);
2502 if (namelen > ll_i2sbi(i1)->ll_namelen)
2503 return ERR_PTR(-ENAMETOOLONG);
2505 if (!lu_name_is_valid_2(name, namelen))
2506 return ERR_PTR(-EINVAL);
2509 if (op_data == NULL)
2510 OBD_ALLOC_PTR(op_data);
2512 if (op_data == NULL)
2513 return ERR_PTR(-ENOMEM);
2515 ll_i2gids(op_data->op_suppgids, i1, i2);
2516 op_data->op_fid1 = *ll_inode2fid(i1);
2517 op_data->op_default_stripe_offset = -1;
2519 if (S_ISDIR(i1->i_mode)) {
2520 down_read(&ll_i2info(i1)->lli_lsm_sem);
2521 op_data->op_mea1_sem = &ll_i2info(i1)->lli_lsm_sem;
2522 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2523 if (opc == LUSTRE_OPC_MKDIR)
2524 op_data->op_default_stripe_offset =
2525 ll_i2info(i1)->lli_def_stripe_offset;
2529 op_data->op_fid2 = *ll_inode2fid(i2);
2530 if (S_ISDIR(i2->i_mode)) {
2532 down_read(&ll_i2info(i2)->lli_lsm_sem);
2533 op_data->op_mea2_sem =
2534 &ll_i2info(i2)->lli_lsm_sem;
2536 op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2539 fid_zero(&op_data->op_fid2);
2542 if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2543 op_data->op_cli_flags |= CLI_HASH64;
2545 if (ll_need_32bit_api(ll_i2sbi(i1)))
2546 op_data->op_cli_flags |= CLI_API32;
2548 op_data->op_name = name;
2549 op_data->op_namelen = namelen;
2550 op_data->op_mode = mode;
2551 op_data->op_mod_time = ktime_get_real_seconds();
2552 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2553 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2554 op_data->op_cap = cfs_curproc_cap_pack();
2555 op_data->op_mds = 0;
2556 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2557 filename_is_volatile(name, namelen, &op_data->op_mds)) {
2558 op_data->op_bias |= MDS_CREATE_VOLATILE;
2560 op_data->op_data = data;
2565 void ll_finish_md_op_data(struct md_op_data *op_data)
2567 ll_unlock_md_op_lsm(op_data);
2568 security_release_secctx(op_data->op_file_secctx,
2569 op_data->op_file_secctx_size);
2570 OBD_FREE_PTR(op_data);
2573 #ifdef HAVE_SUPEROPS_USE_DENTRY
2574 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2576 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2579 struct ll_sb_info *sbi;
2581 #ifdef HAVE_SUPEROPS_USE_DENTRY
2582 LASSERT((seq != NULL) && (dentry != NULL));
2583 sbi = ll_s2sbi(dentry->d_sb);
2585 LASSERT((seq != NULL) && (vfs != NULL));
2586 sbi = ll_s2sbi(vfs->mnt_sb);
2589 if (sbi->ll_flags & LL_SBI_NOLCK)
2590 seq_puts(seq, ",nolock");
2592 if (sbi->ll_flags & LL_SBI_FLOCK)
2593 seq_puts(seq, ",flock");
2595 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2596 seq_puts(seq, ",localflock");
2598 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2599 seq_puts(seq, ",user_xattr");
2601 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2602 seq_puts(seq, ",lazystatfs");
2604 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2605 seq_puts(seq, ",user_fid2path");
2607 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2608 seq_puts(seq, ",always_ping");
2614 * Get obd name by cmd, and copy out to user space
2616 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2618 struct ll_sb_info *sbi = ll_i2sbi(inode);
2619 struct obd_device *obd;
2622 if (cmd == OBD_IOC_GETDTNAME)
2623 obd = class_exp2obd(sbi->ll_dt_exp);
2624 else if (cmd == OBD_IOC_GETMDNAME)
2625 obd = class_exp2obd(sbi->ll_md_exp);
2632 if (copy_to_user((void __user *)arg, obd->obd_name,
2633 strlen(obd->obd_name) + 1))
2640 * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2641 * fsname will be returned in this buffer; otherwise, a static buffer will be
2642 * used to store the fsname and returned to caller.
2644 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2646 static char fsname_static[MTI_NAME_MAXLEN];
2647 struct lustre_sb_info *lsi = s2lsi(sb);
2652 /* this means the caller wants to use static buffer
2653 * and it doesn't care about race. Usually this is
2654 * in error reporting path */
2655 buf = fsname_static;
2656 buflen = sizeof(fsname_static);
2659 len = strlen(lsi->lsi_lmd->lmd_profile);
2660 ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2661 if (ptr && (strcmp(ptr, "-client") == 0))
2664 if (unlikely(len >= buflen))
2666 strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2672 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2679 p.mnt = current->fs->root.mnt;
2681 path = d_path(&p, buf, bufsize);
2686 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2688 char *buf, *path = NULL;
2689 struct dentry *dentry = NULL;
2690 struct inode *inode = page->mapping->host;
2692 /* this can be called inside spin lock so use GFP_ATOMIC. */
2693 buf = (char *)__get_free_page(GFP_ATOMIC);
2695 dentry = d_find_alias(page->mapping->host);
2697 path = ll_d_path(dentry, buf, PAGE_SIZE);
2701 "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2702 "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2703 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2704 PFID(ll_inode2fid(inode)),
2705 (path && !IS_ERR(path)) ? path : "", ioret);
2711 free_page((unsigned long)buf);
2714 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2715 struct lov_user_md **kbuf)
2717 struct lov_user_md lum;
2721 if (copy_from_user(&lum, md, sizeof(lum)))
2724 lum_size = ll_lov_user_md_size(&lum);
2728 OBD_ALLOC(*kbuf, lum_size);
2732 if (copy_from_user(*kbuf, md, lum_size) != 0) {
2733 OBD_FREE(*kbuf, lum_size);
2741 * Compute llite root squash state after a change of root squash
2742 * configuration setting or add/remove of a lnet nid
2744 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2746 struct root_squash_info *squash = &sbi->ll_squash;
2749 struct lnet_process_id id;
2751 /* Update norootsquash flag */
2752 down_write(&squash->rsi_sem);
2753 if (list_empty(&squash->rsi_nosquash_nids))
2754 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2756 /* Do not apply root squash as soon as one of our NIDs is
2757 * in the nosquash_nids list */
2760 while (LNetGetId(i++, &id) != -ENOENT) {
2761 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2763 if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2769 sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2771 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2773 up_write(&squash->rsi_sem);
2777 * Parse linkea content to extract information about a given hardlink
2779 * \param[in] ldata - Initialized linkea data
2780 * \param[in] linkno - Link identifier
2781 * \param[out] parent_fid - The entry's parent FID
2782 * \param[out] ln - Entry name destination buffer
2784 * \retval 0 on success
2785 * \retval Appropriate negative error code on failure
2787 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2788 struct lu_fid *parent_fid, struct lu_name *ln)
2794 rc = linkea_init_with_rec(ldata);
2798 if (linkno >= ldata->ld_leh->leh_reccount)
2799 /* beyond last link */
2802 linkea_first_entry(ldata);
2803 for (idx = 0; ldata->ld_lee != NULL; idx++) {
2804 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2809 linkea_next_entry(ldata);
2819 * Get parent FID and name of an identified link. Operation is performed for
2820 * a given link number, letting the caller iterate over linkno to list one or
2821 * all links of an entry.
2823 * \param[in] file - File descriptor against which to perform the operation
2824 * \param[in,out] arg - User-filled structure containing the linkno to operate
2825 * on and the available size. It is eventually filled with
2826 * the requested information or left untouched on error
2828 * \retval - 0 on success
2829 * \retval - Appropriate negative error code on failure
2831 int ll_getparent(struct file *file, struct getparent __user *arg)
2833 struct inode *inode = file_inode(file);
2834 struct linkea_data *ldata;
2835 struct lu_buf buf = LU_BUF_NULL;
2837 struct lu_fid parent_fid;
2844 if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2845 !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2848 if (get_user(name_size, &arg->gp_name_size))
2851 if (get_user(linkno, &arg->gp_linkno))
2854 if (name_size > PATH_MAX)
2857 OBD_ALLOC(ldata, sizeof(*ldata));
2861 rc = linkea_data_new(ldata, &buf);
2863 GOTO(ldata_free, rc);
2865 #ifdef HAVE_XATTR_HANDLER_FLAGS
2866 rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
2867 buf.lb_len, OBD_MD_FLXATTR);
2869 rc = ll_getxattr(file_dentry(file), XATTR_NAME_LINK, buf.lb_buf,
2871 #endif /* HAVE_XATTR_HANDLER_FLAGS */
2875 rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2879 if (ln.ln_namelen >= name_size)
2880 GOTO(lb_free, rc = -EOVERFLOW);
2882 if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2883 GOTO(lb_free, rc = -EFAULT);
2885 if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2886 GOTO(lb_free, rc = -EFAULT);
2888 if (put_user('\0', arg->gp_name + ln.ln_namelen))
2889 GOTO(lb_free, rc = -EFAULT);
2894 OBD_FREE(ldata, sizeof(*ldata));