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 /sys. As a consequence, we still need to come to an
435 * agreement on the supported algorithms at connect time
437 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
439 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
440 data->ocd_cksum_types = OBD_CKSUM_ADLER;
442 data->ocd_cksum_types = obd_cksum_types_supported_client();
444 #ifdef HAVE_LRU_RESIZE_SUPPORT
445 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
447 /* always ping even if server suppress_pings */
448 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
449 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
451 CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
452 "ocd_grant: %d\n", data->ocd_connect_flags,
453 data->ocd_version, data->ocd_grant);
455 sbi->ll_dt_obd->obd_upcall.onu_owner = &sbi->ll_lco;
456 sbi->ll_dt_obd->obd_upcall.onu_upcall = cl_ocd_update;
458 data->ocd_brw_size = DT_MAX_BRW_SIZE;
460 err = obd_connect(NULL, &sbi->ll_dt_exp, sbi->ll_dt_obd,
461 &sbi->ll_sb_uuid, data, NULL);
463 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
464 "recovery, of which this client is not a "
465 "part. Please wait for recovery to "
466 "complete, abort, or time out.\n", dt);
469 CERROR("%s: Cannot connect to %s: rc = %d\n",
470 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
474 sbi->ll_dt_exp->exp_connect_data = *data;
476 /* Don't change value if it was specified in the config log */
477 if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages == -1)
478 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
479 max_t(unsigned long, SBI_DEFAULT_READAHEAD_WHOLE_MAX,
480 (data->ocd_brw_size >> PAGE_SHIFT));
482 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
483 LUSTRE_SEQ_METADATA);
485 CERROR("%s: Can't init data layer FID infrastructure, "
486 "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
490 mutex_lock(&sbi->ll_lco.lco_lock);
491 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
492 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
493 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
494 mutex_unlock(&sbi->ll_lco.lco_lock);
496 fid_zero(&sbi->ll_root_fid);
497 err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
500 CERROR("cannot mds_connect: rc = %d\n", err);
501 GOTO(out_lock_cn_cb, err);
503 if (!fid_is_sane(&sbi->ll_root_fid)) {
504 CERROR("%s: Invalid root fid "DFID" during mount\n",
505 sbi->ll_md_exp->exp_obd->obd_name,
506 PFID(&sbi->ll_root_fid));
507 GOTO(out_lock_cn_cb, err = -EINVAL);
509 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
511 sb->s_op = &lustre_super_operations;
512 #ifdef HAVE_XATTR_HANDLER_FLAGS
513 sb->s_xattr = ll_xattr_handlers;
515 #if THREAD_SIZE >= 8192 /*b=17630*/
516 sb->s_export_op = &lustre_export_operations;
520 * XXX: move this to after cbd setup? */
521 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
522 if (sbi->ll_flags & LL_SBI_ACL)
523 valid |= OBD_MD_FLACL;
525 OBD_ALLOC_PTR(op_data);
527 GOTO(out_lock_cn_cb, err = -ENOMEM);
529 op_data->op_fid1 = sbi->ll_root_fid;
530 op_data->op_mode = 0;
531 op_data->op_valid = valid;
533 err = md_getattr(sbi->ll_md_exp, op_data, &request);
535 OBD_FREE_PTR(op_data);
537 CERROR("%s: md_getattr failed for root: rc = %d\n",
538 sbi->ll_md_exp->exp_obd->obd_name, err);
539 GOTO(out_lock_cn_cb, err);
542 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
543 sbi->ll_md_exp, &lmd);
545 CERROR("failed to understand root inode md: rc = %d\n", err);
546 ptlrpc_req_finished(request);
547 GOTO(out_lock_cn_cb, err);
550 LASSERT(fid_is_sane(&sbi->ll_root_fid));
551 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
552 sbi->ll_flags & LL_SBI_32BIT_API),
554 md_free_lustre_md(sbi->ll_md_exp, &lmd);
555 ptlrpc_req_finished(request);
558 #ifdef CONFIG_FS_POSIX_ACL
560 posix_acl_release(lmd.posix_acl);
561 lmd.posix_acl = NULL;
564 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
566 CERROR("lustre_lite: bad iget4 for root\n");
570 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
571 if (sbi->ll_checksum_set) {
572 err = obd_set_info_async(NULL, sbi->ll_dt_exp,
573 sizeof(KEY_CHECKSUM), KEY_CHECKSUM,
574 sizeof(checksum), &checksum, NULL);
576 CERROR("%s: Set checksum failed: rc = %d\n",
577 sbi->ll_dt_exp->exp_obd->obd_name, err);
583 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
584 KEY_CACHE_SET, sizeof(*sbi->ll_cache),
585 sbi->ll_cache, NULL);
587 CERROR("%s: Set cache_set failed: rc = %d\n",
588 sbi->ll_dt_exp->exp_obd->obd_name, err);
592 sb->s_root = d_make_root(root);
593 if (sb->s_root == NULL) {
594 CERROR("%s: can't make root dentry\n",
595 ll_get_fsname(sb, NULL, 0));
596 GOTO(out_root, err = -ENOMEM);
598 #ifdef HAVE_DCACHE_LOCK
599 sb->s_root->d_op = &ll_d_ops;
602 sbi->ll_sdev_orig = sb->s_dev;
604 /* We set sb->s_dev equal on all lustre clients in order to support
605 * NFS export clustering. NFSD requires that the FSID be the same
607 /* s_dev is also used in lt_compare() to compare two fs, but that is
608 * only a node-local comparison. */
609 uuid = obd_get_uuid(sbi->ll_md_exp);
611 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
618 if (sbi->ll_dt_obd) {
619 err = sysfs_create_link(&sbi->ll_kset.kobj,
620 &sbi->ll_dt_obd->obd_kset.kobj,
621 sbi->ll_dt_obd->obd_type->typ_name);
623 CERROR("%s: could not register %s in llite: rc = %d\n",
624 dt, ll_get_fsname(sb, NULL, 0), err);
629 if (sbi->ll_md_obd) {
630 err = sysfs_create_link(&sbi->ll_kset.kobj,
631 &sbi->ll_md_obd->obd_kset.kobj,
632 sbi->ll_md_obd->obd_type->typ_name);
634 CERROR("%s: could not register %s in llite: rc = %d\n",
635 md, ll_get_fsname(sb, NULL, 0), err);
645 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
647 obd_disconnect(sbi->ll_dt_exp);
648 sbi->ll_dt_exp = NULL;
649 sbi->ll_dt_obd = NULL;
651 obd_fid_fini(sbi->ll_md_exp->exp_obd);
653 obd_disconnect(sbi->ll_md_exp);
654 sbi->ll_md_exp = NULL;
655 sbi->ll_md_obd = NULL;
664 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
668 size = sizeof(*lmmsize);
669 rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
670 KEY_MAX_EASIZE, &size, lmmsize);
672 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
673 sbi->ll_dt_exp->exp_obd->obd_name, rc);
678 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
679 KEY_MAX_EASIZE, &size, lmmsize);
681 CERROR("Get max mdsize error rc %d\n", rc);
687 * Get the value of the default_easize parameter.
689 * \see client_obd::cl_default_mds_easize
691 * \param[in] sbi superblock info for this filesystem
692 * \param[out] lmmsize pointer to storage location for value
694 * \retval 0 on success
695 * \retval negative negated errno on failure
697 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
702 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
703 KEY_DEFAULT_EASIZE, &size, lmmsize);
705 CERROR("Get default mdsize error rc %d\n", rc);
711 * Set the default_easize parameter to the given value.
713 * \see client_obd::cl_default_mds_easize
715 * \param[in] sbi superblock info for this filesystem
716 * \param[in] lmmsize the size to set
718 * \retval 0 on success
719 * \retval negative negated errno on failure
721 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
725 if (lmmsize < sizeof(struct lov_mds_md) ||
726 lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
729 rc = obd_set_info_async(NULL, sbi->ll_md_exp,
730 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
731 sizeof(int), &lmmsize, NULL);
736 static void client_common_put_super(struct super_block *sb)
738 struct ll_sb_info *sbi = ll_s2sbi(sb);
743 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
744 obd_disconnect(sbi->ll_dt_exp);
745 sbi->ll_dt_exp = NULL;
747 ll_debugfs_unregister_super(sb);
749 obd_fid_fini(sbi->ll_md_exp->exp_obd);
750 obd_disconnect(sbi->ll_md_exp);
751 sbi->ll_md_exp = NULL;
756 void ll_kill_super(struct super_block *sb)
758 struct ll_sb_info *sbi;
762 if (!(sb->s_flags & MS_ACTIVE))
766 /* we need restore s_dev from changed for clustred NFS before put_super
767 * because new kernels have cached s_dev and change sb->s_dev in
768 * put_super not affected real removing devices */
770 sb->s_dev = sbi->ll_sdev_orig;
771 sbi->ll_umounting = 1;
773 /* wait running statahead threads to quit */
774 while (atomic_read(&sbi->ll_sa_running) > 0) {
775 set_current_state(TASK_UNINTERRUPTIBLE);
776 schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
783 static inline int ll_set_opt(const char *opt, char *data, int fl)
785 if (strncmp(opt, data, strlen(opt)) != 0)
791 /* non-client-specific mount options are parsed in lmd_parse */
792 static int ll_options(char *options, struct ll_sb_info *sbi)
795 char *s1 = options, *s2;
796 int *flags = &sbi->ll_flags;
802 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
805 CDEBUG(D_SUPER, "next opt=%s\n", s1);
806 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
811 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
816 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
821 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
826 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
831 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
836 tmp = ll_set_opt("context", s1, 1);
839 tmp = ll_set_opt("fscontext", s1, 1);
842 tmp = ll_set_opt("defcontext", s1, 1);
845 tmp = ll_set_opt("rootcontext", s1, 1);
848 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
853 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
859 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
862 sbi->ll_checksum_set = 1;
865 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
868 sbi->ll_checksum_set = 1;
871 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
876 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
881 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
886 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
891 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
896 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
901 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
906 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
911 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
917 s2 = strchr(s1, ',');
925 void ll_lli_init(struct ll_inode_info *lli)
927 lli->lli_inode_magic = LLI_INODE_MAGIC;
929 spin_lock_init(&lli->lli_lock);
930 lli->lli_posix_acl = NULL;
931 /* Do not set lli_fid, it has been initialized already. */
932 fid_zero(&lli->lli_pfid);
933 lli->lli_mds_read_och = NULL;
934 lli->lli_mds_write_och = NULL;
935 lli->lli_mds_exec_och = NULL;
936 lli->lli_open_fd_read_count = 0;
937 lli->lli_open_fd_write_count = 0;
938 lli->lli_open_fd_exec_count = 0;
939 mutex_init(&lli->lli_och_mutex);
940 spin_lock_init(&lli->lli_agl_lock);
941 spin_lock_init(&lli->lli_layout_lock);
942 ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
943 lli->lli_clob = NULL;
945 init_rwsem(&lli->lli_xattrs_list_rwsem);
946 mutex_init(&lli->lli_xattrs_enq_lock);
948 LASSERT(lli->lli_vfs_inode.i_mode != 0);
949 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
950 mutex_init(&lli->lli_readdir_mutex);
951 lli->lli_opendir_key = NULL;
953 spin_lock_init(&lli->lli_sa_lock);
954 lli->lli_opendir_pid = 0;
955 lli->lli_sa_enabled = 0;
956 lli->lli_def_stripe_offset = -1;
957 init_rwsem(&lli->lli_lsm_sem);
959 mutex_init(&lli->lli_size_mutex);
960 lli->lli_symlink_name = NULL;
961 init_rwsem(&lli->lli_trunc_sem);
962 range_lock_tree_init(&lli->lli_write_tree);
963 init_rwsem(&lli->lli_glimpse_sem);
964 lli->lli_glimpse_time = ktime_set(0, 0);
965 INIT_LIST_HEAD(&lli->lli_agl_list);
966 lli->lli_agl_index = 0;
967 lli->lli_async_rc = 0;
969 mutex_init(&lli->lli_layout_mutex);
970 memset(lli->lli_jobid, 0, sizeof(lli->lli_jobid));
973 #define MAX_STRING_SIZE 128
975 #ifndef HAVE_SUPER_SETUP_BDI_NAME
977 #define LSI_BDI_INITIALIZED 0x00400000
979 #ifndef HAVE_BDI_CAP_MAP_COPY
980 # define BDI_CAP_MAP_COPY 0
983 static int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
985 struct lustre_sb_info *lsi = s2lsi(sb);
986 char buf[MAX_STRING_SIZE];
990 err = bdi_init(&lsi->lsi_bdi);
994 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
995 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
996 lsi->lsi_bdi.name = "lustre";
998 vsnprintf(buf, MAX_STRING_SIZE, fmt, args);
1000 err = bdi_register(&lsi->lsi_bdi, NULL, "%s", buf);
1003 sb->s_bdi = &lsi->lsi_bdi;
1007 #endif /* !HAVE_SUPER_SETUP_BDI_NAME */
1009 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
1011 struct lustre_profile *lprof = NULL;
1012 struct lustre_sb_info *lsi = s2lsi(sb);
1013 struct ll_sb_info *sbi = NULL;
1014 char *dt = NULL, *md = NULL;
1015 char *profilenm = get_profile_name(sb);
1016 struct config_llog_instance *cfg;
1017 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1018 const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
1019 char name[MAX_STRING_SIZE];
1027 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1029 try_module_get(THIS_MODULE);
1033 GOTO(out_free_cfg, err = -ENOMEM);
1035 /* client additional sb info */
1036 lsi->lsi_llsbi = sbi = ll_init_sbi();
1038 GOTO(out_free_cfg, err = -ENOMEM);
1040 err = ll_options(lsi->lsi_lmd->lmd_opts, sbi);
1042 GOTO(out_free_cfg, err);
1044 err = super_setup_bdi_name(sb, "lustre-%p", sb);
1046 GOTO(out_free_cfg, err);
1048 #ifndef HAVE_DCACHE_LOCK
1049 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1050 sb->s_d_op = &ll_d_ops;
1053 len = strlen(lsi->lsi_lmd->lmd_profile);
1054 ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
1055 if (ptr && (strcmp(ptr, "-client") == 0))
1059 snprintf(name, MAX_STRING_SIZE, "%.*s-%p", len,
1060 lsi->lsi_lmd->lmd_profile, sb);
1062 /* Call ll_debugfs_register_super() before lustre_process_log()
1063 * so that "llite.*.*" params can be processed correctly.
1065 err = ll_debugfs_register_super(sb, name);
1067 CERROR("%s: could not register mountpoint in llite: rc = %d\n",
1068 ll_get_fsname(sb, NULL, 0), err);
1072 /* Generate a string unique to this super, in case some joker tries
1073 * to mount the same fs at two mount points.
1074 * Use the address of the super itself.
1076 cfg->cfg_instance = sb;
1077 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1078 cfg->cfg_callback = class_config_llog_handler;
1079 cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1080 /* set up client obds */
1081 err = lustre_process_log(sb, profilenm, cfg);
1083 GOTO(out_debugfs, err);
1085 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1086 lprof = class_get_profile(profilenm);
1087 if (lprof == NULL) {
1088 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1089 " read from the MGS. Does that filesystem "
1090 "exist?\n", profilenm);
1091 GOTO(out_debugfs, err = -EINVAL);
1093 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1094 lprof->lp_md, lprof->lp_dt);
1096 dt_len = strlen(lprof->lp_dt) + instlen + 2;
1097 OBD_ALLOC(dt, dt_len);
1099 GOTO(out_profile, err = -ENOMEM);
1100 snprintf(dt, dt_len - 1, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1102 md_len = strlen(lprof->lp_md) + instlen + 2;
1103 OBD_ALLOC(md, md_len);
1105 GOTO(out_free_dt, err = -ENOMEM);
1106 snprintf(md, md_len - 1, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1108 /* connections, registrations, sb setup */
1109 err = client_common_fill_super(sb, md, dt, mnt);
1111 GOTO(out_free_md, err);
1113 sbi->ll_client_common_fill_super_succeeded = 1;
1117 OBD_FREE(md, md_len);
1120 OBD_FREE(dt, dt_len);
1123 class_put_profile(lprof);
1126 ll_debugfs_unregister_super(sb);
1133 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1134 LCONSOLE_WARN("Mounted %s\n", profilenm);
1136 } /* ll_fill_super */
1138 void ll_put_super(struct super_block *sb)
1140 struct config_llog_instance cfg, params_cfg;
1141 struct obd_device *obd;
1142 struct lustre_sb_info *lsi = s2lsi(sb);
1143 struct ll_sb_info *sbi = ll_s2sbi(sb);
1144 char *profilenm = get_profile_name(sb);
1146 int next, force = 1, rc = 0;
1150 GOTO(out_no_sbi, 0);
1152 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1154 cfg.cfg_instance = sb;
1155 lustre_end_log(sb, profilenm, &cfg);
1157 params_cfg.cfg_instance = sb;
1158 lustre_end_log(sb, PARAMS_FILENAME, ¶ms_cfg);
1160 if (sbi->ll_md_exp) {
1161 obd = class_exp2obd(sbi->ll_md_exp);
1163 force = obd->obd_force;
1166 /* Wait for unstable pages to be committed to stable storage */
1168 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1169 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1170 atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1174 ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1175 if (force == 0 && rc != -EINTR)
1176 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1179 /* We need to set force before the lov_disconnect in
1180 lustre_common_put_super, since l_d cleans up osc's as well. */
1183 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1185 obd->obd_force = force;
1189 if (sbi->ll_client_common_fill_super_succeeded) {
1190 /* Only if client_common_fill_super succeeded */
1191 client_common_put_super(sb);
1195 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1196 class_manual_cleanup(obd);
1199 if (sbi->ll_flags & LL_SBI_VERBOSE)
1200 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1203 class_del_profile(profilenm);
1205 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1206 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1207 bdi_destroy(&lsi->lsi_bdi);
1208 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1213 lsi->lsi_llsbi = NULL;
1215 lustre_common_put_super(sb);
1217 cl_env_cache_purge(~0);
1219 module_put(THIS_MODULE);
1222 } /* client_put_super */
1224 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1226 struct inode *inode = NULL;
1228 /* NOTE: we depend on atomic igrab() -bzzz */
1229 lock_res_and_lock(lock);
1230 if (lock->l_resource->lr_lvb_inode) {
1231 struct ll_inode_info * lli;
1232 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1233 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1234 inode = igrab(lock->l_resource->lr_lvb_inode);
1236 inode = lock->l_resource->lr_lvb_inode;
1237 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1238 D_WARNING, lock, "lr_lvb_inode %p is "
1239 "bogus: magic %08x",
1240 lock->l_resource->lr_lvb_inode,
1241 lli->lli_inode_magic);
1245 unlock_res_and_lock(lock);
1249 void ll_dir_clear_lsm_md(struct inode *inode)
1251 struct ll_inode_info *lli = ll_i2info(inode);
1253 LASSERT(S_ISDIR(inode->i_mode));
1255 if (lli->lli_lsm_md != NULL) {
1256 lmv_free_memmd(lli->lli_lsm_md);
1257 lli->lli_lsm_md = NULL;
1261 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1262 const struct lu_fid *fid,
1263 struct lustre_md *md)
1265 struct ll_sb_info *sbi = ll_s2sbi(sb);
1266 struct mdt_body *body = md->body;
1267 struct inode *inode;
1271 ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1272 inode = iget_locked(sb, ino);
1273 if (inode == NULL) {
1274 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1275 ll_get_fsname(sb, NULL, 0), PFID(fid));
1276 RETURN(ERR_PTR(-ENOENT));
1279 if (inode->i_state & I_NEW) {
1280 struct ll_inode_info *lli = ll_i2info(inode);
1281 struct lmv_stripe_md *lsm = md->lmv;
1283 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1284 (body->mbo_mode & S_IFMT);
1285 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1288 LTIME_S(inode->i_mtime) = 0;
1289 LTIME_S(inode->i_atime) = 0;
1290 LTIME_S(inode->i_ctime) = 0;
1293 #ifdef HAVE_BACKING_DEV_INFO
1294 /* initializing backing dev info. */
1295 inode->i_mapping->backing_dev_info =
1296 &s2lsi(inode->i_sb)->lsi_bdi;
1298 inode->i_op = &ll_dir_inode_operations;
1299 inode->i_fop = &ll_dir_operations;
1300 lli->lli_fid = *fid;
1303 LASSERT(lsm != NULL);
1304 /* master object FID */
1305 lli->lli_pfid = body->mbo_fid1;
1306 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1307 lli, PFID(fid), PFID(&lli->lli_pfid));
1308 unlock_new_inode(inode);
1314 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1317 struct lmv_stripe_md *lsm = md->lmv;
1318 struct ll_inode_info *lli = ll_i2info(inode);
1321 LASSERT(lsm != NULL);
1323 CDEBUG(D_INODE, "%s: "DFID" set dir layout:\n",
1324 ll_get_fsname(inode->i_sb, NULL, 0),
1325 PFID(&lli->lli_fid));
1326 lsm_md_dump(D_INODE, lsm);
1328 /* XXX sigh, this lsm_root initialization should be in
1329 * LMV layer, but it needs ll_iget right now, so we
1330 * put this here right now. */
1331 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1332 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1333 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1334 /* Unfortunately ll_iget will call ll_update_inode,
1335 * where the initialization of slave inode is slightly
1336 * different, so it reset lsm_md to NULL to avoid
1337 * initializing lsm for slave inode. */
1338 lsm->lsm_md_oinfo[i].lmo_root =
1339 ll_iget_anon_dir(inode->i_sb, fid, md);
1340 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1341 int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1343 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1345 iput(lsm->lsm_md_oinfo[i].lmo_root);
1346 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1352 lli->lli_lsm_md = lsm;
1357 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1359 struct ll_inode_info *lli = ll_i2info(inode);
1360 struct lmv_stripe_md *lsm = md->lmv;
1365 LASSERT(S_ISDIR(inode->i_mode));
1366 CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1367 PFID(ll_inode2fid(inode)));
1370 * no striped information from request, lustre_md from req does not
1371 * include stripeEA, see ll_md_setattr()
1377 * normally dir layout doesn't change, only take read lock to check
1378 * that to avoid blocking other MD operations.
1380 if (lli->lli_lsm_md)
1381 down_read(&lli->lli_lsm_sem);
1383 down_write(&lli->lli_lsm_sem);
1386 * if dir layout mismatch, check whether version is increased, which
1387 * means layout is changed, this happens in dir migration and lfsck.
1389 if (lli->lli_lsm_md && !lsm_md_eq(lli->lli_lsm_md, lsm)) {
1390 if (lsm->lsm_md_layout_version <=
1391 lli->lli_lsm_md->lsm_md_layout_version) {
1392 CERROR("%s: "DFID" dir layout mismatch:\n",
1393 ll_get_fsname(inode->i_sb, NULL, 0),
1394 PFID(&lli->lli_fid));
1395 lsm_md_dump(D_ERROR, lli->lli_lsm_md);
1396 lsm_md_dump(D_ERROR, lsm);
1397 GOTO(unlock, rc = -EINVAL);
1400 /* layout changed, switch to write lock */
1401 up_read(&lli->lli_lsm_sem);
1402 down_write(&lli->lli_lsm_sem);
1403 ll_dir_clear_lsm_md(inode);
1406 /* set directory layout */
1407 if (!lli->lli_lsm_md) {
1408 struct cl_attr *attr;
1410 rc = ll_init_lsm_md(inode, md);
1411 up_write(&lli->lli_lsm_sem);
1415 /* set md->lmv to NULL, so the following free lustre_md
1416 * will not free this lsm */
1420 * md_merge_attr() may take long, since lsm is already set,
1421 * switch to read lock.
1423 down_read(&lli->lli_lsm_sem);
1425 OBD_ALLOC_PTR(attr);
1427 GOTO(unlock, rc = -ENOMEM);
1429 /* validate the lsm */
1430 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1431 ll_md_blocking_ast);
1437 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1438 md->body->mbo_nlink = attr->cat_nlink;
1439 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1440 md->body->mbo_size = attr->cat_size;
1441 if (md->body->mbo_valid & OBD_MD_FLATIME)
1442 md->body->mbo_atime = attr->cat_atime;
1443 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1444 md->body->mbo_ctime = attr->cat_ctime;
1445 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1446 md->body->mbo_mtime = attr->cat_mtime;
1451 up_read(&lli->lli_lsm_sem);
1456 void ll_clear_inode(struct inode *inode)
1458 struct ll_inode_info *lli = ll_i2info(inode);
1459 struct ll_sb_info *sbi = ll_i2sbi(inode);
1462 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1463 PFID(ll_inode2fid(inode)), inode);
1465 if (S_ISDIR(inode->i_mode)) {
1466 /* these should have been cleared in ll_file_release */
1467 LASSERT(lli->lli_opendir_key == NULL);
1468 LASSERT(lli->lli_sai == NULL);
1469 LASSERT(lli->lli_opendir_pid == 0);
1472 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1474 LASSERT(!lli->lli_open_fd_write_count);
1475 LASSERT(!lli->lli_open_fd_read_count);
1476 LASSERT(!lli->lli_open_fd_exec_count);
1478 if (lli->lli_mds_write_och)
1479 ll_md_real_close(inode, FMODE_WRITE);
1480 if (lli->lli_mds_exec_och)
1481 ll_md_real_close(inode, FMODE_EXEC);
1482 if (lli->lli_mds_read_och)
1483 ll_md_real_close(inode, FMODE_READ);
1485 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1486 OBD_FREE(lli->lli_symlink_name,
1487 strlen(lli->lli_symlink_name) + 1);
1488 lli->lli_symlink_name = NULL;
1491 ll_xattr_cache_destroy(inode);
1493 #ifdef CONFIG_FS_POSIX_ACL
1494 forget_all_cached_acls(inode);
1495 if (lli->lli_posix_acl) {
1496 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1497 posix_acl_release(lli->lli_posix_acl);
1498 lli->lli_posix_acl = NULL;
1501 lli->lli_inode_magic = LLI_INODE_DEAD;
1503 if (S_ISDIR(inode->i_mode))
1504 ll_dir_clear_lsm_md(inode);
1505 else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1506 LASSERT(list_empty(&lli->lli_agl_list));
1509 * XXX This has to be done before lsm is freed below, because
1510 * cl_object still uses inode lsm.
1512 cl_inode_fini(inode);
1517 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1519 struct lustre_md md;
1520 struct inode *inode = dentry->d_inode;
1521 struct ll_sb_info *sbi = ll_i2sbi(inode);
1522 struct ptlrpc_request *request = NULL;
1526 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1527 LUSTRE_OPC_ANY, NULL);
1528 if (IS_ERR(op_data))
1529 RETURN(PTR_ERR(op_data));
1531 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1533 ptlrpc_req_finished(request);
1534 if (rc == -ENOENT) {
1536 /* Unlinked special device node? Or just a race?
1537 * Pretend we done everything. */
1538 if (!S_ISREG(inode->i_mode) &&
1539 !S_ISDIR(inode->i_mode)) {
1540 ia_valid = op_data->op_attr.ia_valid;
1541 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1542 rc = simple_setattr(dentry, &op_data->op_attr);
1543 op_data->op_attr.ia_valid = ia_valid;
1545 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1546 CERROR("md_setattr fails: rc = %d\n", rc);
1551 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1552 sbi->ll_md_exp, &md);
1554 ptlrpc_req_finished(request);
1558 ia_valid = op_data->op_attr.ia_valid;
1559 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1560 * cache is not cleared yet. */
1561 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1562 if (S_ISREG(inode->i_mode))
1564 rc = simple_setattr(dentry, &op_data->op_attr);
1565 if (S_ISREG(inode->i_mode))
1566 inode_unlock(inode);
1567 op_data->op_attr.ia_valid = ia_valid;
1569 rc = ll_update_inode(inode, &md);
1570 ptlrpc_req_finished(request);
1575 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1576 * object(s) determine the file size and mtime. Otherwise, the MDS will
1577 * keep these values until such a time that objects are allocated for it.
1578 * We do the MDS operations first, as it is checking permissions for us.
1579 * We don't to the MDS RPC if there is nothing that we want to store there,
1580 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1581 * going to do an RPC anyways.
1583 * If we are doing a truncate, we will send the mtime and ctime updates
1584 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1585 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1588 * In case of HSMimport, we only set attr on MDS.
1590 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr,
1591 enum op_xvalid xvalid, bool hsm_import)
1593 struct inode *inode = dentry->d_inode;
1594 struct ll_inode_info *lli = ll_i2info(inode);
1595 struct md_op_data *op_data = NULL;
1599 CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1600 "valid %x, hsm_import %d\n",
1601 ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1602 inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1605 if (attr->ia_valid & ATTR_SIZE) {
1606 /* Check new size against VFS/VM file size limit and rlimit */
1607 rc = inode_newsize_ok(inode, attr->ia_size);
1611 /* The maximum Lustre file size is variable, based on the
1612 * OST maximum object size and number of stripes. This
1613 * needs another check in addition to the VFS check above. */
1614 if (attr->ia_size > ll_file_maxbytes(inode)) {
1615 CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
1616 PFID(&lli->lli_fid), attr->ia_size,
1617 ll_file_maxbytes(inode));
1621 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1624 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1625 if (attr->ia_valid & TIMES_SET_FLAGS) {
1626 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1627 !cfs_capable(CFS_CAP_FOWNER))
1631 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1632 if (!(xvalid & OP_XVALID_CTIME_SET) &&
1633 (attr->ia_valid & ATTR_CTIME)) {
1634 attr->ia_ctime = current_time(inode);
1635 xvalid |= OP_XVALID_CTIME_SET;
1637 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1638 (attr->ia_valid & ATTR_ATIME)) {
1639 attr->ia_atime = current_time(inode);
1640 attr->ia_valid |= ATTR_ATIME_SET;
1642 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1643 (attr->ia_valid & ATTR_MTIME)) {
1644 attr->ia_mtime = current_time(inode);
1645 attr->ia_valid |= ATTR_MTIME_SET;
1648 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1649 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %llu\n",
1650 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1651 (s64)ktime_get_real_seconds());
1653 if (S_ISREG(inode->i_mode)) {
1654 if (attr->ia_valid & ATTR_SIZE)
1655 inode_dio_write_done(inode);
1656 inode_unlock(inode);
1659 /* We always do an MDS RPC, even if we're only changing the size;
1660 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1662 OBD_ALLOC_PTR(op_data);
1663 if (op_data == NULL)
1664 GOTO(out, rc = -ENOMEM);
1666 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1667 /* If we are changing file size, file content is
1668 * modified, flag it.
1670 xvalid |= OP_XVALID_OWNEROVERRIDE;
1671 op_data->op_bias |= MDS_DATA_MODIFIED;
1672 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1675 if (attr->ia_valid & ATTR_FILE) {
1676 struct ll_file_data *fd = LUSTRE_FPRIVATE(attr->ia_file);
1678 if (fd->fd_lease_och)
1679 op_data->op_bias |= MDS_TRUNC_KEEP_LEASE;
1682 op_data->op_attr = *attr;
1683 op_data->op_xvalid = xvalid;
1685 rc = ll_md_setattr(dentry, op_data);
1689 if (!S_ISREG(inode->i_mode) || hsm_import)
1692 if (attr->ia_valid & (ATTR_SIZE | ATTR_ATIME | ATTR_ATIME_SET |
1693 ATTR_MTIME | ATTR_MTIME_SET | ATTR_CTIME) ||
1694 xvalid & OP_XVALID_CTIME_SET) {
1695 /* For truncate and utimes sending attributes to OSTs, setting
1696 * mtime/atime to the past will be performed under PW [0:EOF]
1697 * extent lock (new_size:EOF for truncate). It may seem
1698 * excessive to send mtime/atime updates to OSTs when not
1699 * setting times to past, but it is necessary due to possible
1700 * time de-synchronization between MDT inode and OST objects
1702 rc = cl_setattr_ost(lli->lli_clob, attr, xvalid, 0);
1705 /* If the file was restored, it needs to set dirty flag.
1707 * We've already sent MDS_DATA_MODIFIED flag in
1708 * ll_md_setattr() for truncate. However, the MDT refuses to
1709 * set the HS_DIRTY flag on released files, so we have to set
1710 * it again if the file has been restored. Please check how
1711 * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1713 * Please notice that if the file is not released, the previous
1714 * MDS_DATA_MODIFIED has taken effect and usually
1715 * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1716 * This way we can save an RPC for common open + trunc
1718 if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1719 struct hsm_state_set hss = {
1720 .hss_valid = HSS_SETMASK,
1721 .hss_setmask = HS_DIRTY,
1725 rc2 = ll_hsm_state_set(inode, &hss);
1726 /* truncate and write can happen at the same time, so that
1727 * the file can be set modified even though the file is not
1728 * restored from released state, and ll_hsm_state_set() is
1729 * not applicable for the file, and rc2 < 0 is normal in this
1732 CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1733 PFID(ll_inode2fid(inode)), rc2);
1738 if (op_data != NULL)
1739 ll_finish_md_op_data(op_data);
1741 if (S_ISREG(inode->i_mode)) {
1743 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1744 inode_dio_wait(inode);
1745 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
1746 * flag. ll_update_inode (called from ll_md_setattr), clears
1747 * inode flags, so there is a gap where S_NOSEC is not set.
1748 * This can cause a writer to take the i_mutex unnecessarily,
1749 * but this is safe to do and should be rare. */
1750 inode_has_no_xattr(inode);
1753 ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1754 LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1759 int ll_setattr(struct dentry *de, struct iattr *attr)
1761 int mode = de->d_inode->i_mode;
1762 enum op_xvalid xvalid = 0;
1764 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1765 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1766 xvalid |= OP_XVALID_OWNEROVERRIDE;
1768 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1769 (ATTR_SIZE|ATTR_MODE)) &&
1770 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1771 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1772 !(attr->ia_mode & S_ISGID))))
1773 attr->ia_valid |= ATTR_FORCE;
1775 if ((attr->ia_valid & ATTR_MODE) &&
1777 !(attr->ia_mode & S_ISUID) &&
1778 !(attr->ia_valid & ATTR_KILL_SUID))
1779 attr->ia_valid |= ATTR_KILL_SUID;
1781 if ((attr->ia_valid & ATTR_MODE) &&
1782 ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1783 !(attr->ia_mode & S_ISGID) &&
1784 !(attr->ia_valid & ATTR_KILL_SGID))
1785 attr->ia_valid |= ATTR_KILL_SGID;
1787 return ll_setattr_raw(de, attr, xvalid, false);
1790 int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
1793 struct obd_statfs obd_osfs;
1798 max_age = ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS;
1800 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1802 CERROR("md_statfs fails: rc = %d\n", rc);
1806 osfs->os_type = LL_SUPER_MAGIC;
1808 CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1809 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1811 if (osfs->os_state & OS_STATE_SUM)
1814 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1815 flags |= OBD_STATFS_NODELAY;
1817 rc = obd_statfs(NULL, sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1819 CERROR("obd_statfs fails: rc = %d\n", rc);
1823 CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1824 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1827 osfs->os_bsize = obd_osfs.os_bsize;
1828 osfs->os_blocks = obd_osfs.os_blocks;
1829 osfs->os_bfree = obd_osfs.os_bfree;
1830 osfs->os_bavail = obd_osfs.os_bavail;
1832 /* If we don't have as many objects free on the OST as inodes
1833 * on the MDS, we reduce the total number of inodes to
1834 * compensate, so that the "inodes in use" number is correct.
1836 if (obd_osfs.os_ffree < osfs->os_ffree) {
1837 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1839 osfs->os_ffree = obd_osfs.os_ffree;
1845 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1847 struct super_block *sb = de->d_sb;
1848 struct obd_statfs osfs;
1849 __u64 fsid = huge_encode_dev(sb->s_dev);
1852 CDEBUG(D_VFSTRACE, "VFS Op: at %llu jiffies\n", get_jiffies_64());
1853 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1855 /* Some amount of caching on the client is allowed */
1856 rc = ll_statfs_internal(ll_s2sbi(sb), &osfs, OBD_STATFS_SUM);
1860 statfs_unpack(sfs, &osfs);
1862 /* We need to downshift for all 32-bit kernels, because we can't
1863 * tell if the kernel is being called via sys_statfs64() or not.
1864 * Stop before overflowing f_bsize - in which case it is better
1865 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1866 if (sizeof(long) < 8) {
1867 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1870 osfs.os_blocks >>= 1;
1871 osfs.os_bfree >>= 1;
1872 osfs.os_bavail >>= 1;
1876 sfs->f_blocks = osfs.os_blocks;
1877 sfs->f_bfree = osfs.os_bfree;
1878 sfs->f_bavail = osfs.os_bavail;
1879 sfs->f_fsid.val[0] = (__u32)fsid;
1880 sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1884 void ll_inode_size_lock(struct inode *inode)
1886 struct ll_inode_info *lli;
1888 LASSERT(!S_ISDIR(inode->i_mode));
1890 lli = ll_i2info(inode);
1891 mutex_lock(&lli->lli_size_mutex);
1894 void ll_inode_size_unlock(struct inode *inode)
1896 struct ll_inode_info *lli;
1898 lli = ll_i2info(inode);
1899 mutex_unlock(&lli->lli_size_mutex);
1902 void ll_update_inode_flags(struct inode *inode, int ext_flags)
1904 inode->i_flags = ll_ext_to_inode_flags(ext_flags);
1905 if (ext_flags & LUSTRE_PROJINHERIT_FL)
1906 ll_file_set_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
1908 ll_file_clear_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
1911 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1913 struct ll_inode_info *lli = ll_i2info(inode);
1914 struct mdt_body *body = md->body;
1915 struct ll_sb_info *sbi = ll_i2sbi(inode);
1918 if (body->mbo_valid & OBD_MD_FLEASIZE) {
1919 rc = cl_file_inode_init(inode, md);
1924 if (S_ISDIR(inode->i_mode)) {
1925 rc = ll_update_lsm_md(inode, md);
1930 #ifdef CONFIG_FS_POSIX_ACL
1931 if (body->mbo_valid & OBD_MD_FLACL) {
1932 spin_lock(&lli->lli_lock);
1933 if (lli->lli_posix_acl)
1934 posix_acl_release(lli->lli_posix_acl);
1935 lli->lli_posix_acl = md->posix_acl;
1936 spin_unlock(&lli->lli_lock);
1939 inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1940 sbi->ll_flags & LL_SBI_32BIT_API);
1941 inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1943 if (body->mbo_valid & OBD_MD_FLATIME) {
1944 if (body->mbo_atime > LTIME_S(inode->i_atime))
1945 LTIME_S(inode->i_atime) = body->mbo_atime;
1946 lli->lli_atime = body->mbo_atime;
1949 if (body->mbo_valid & OBD_MD_FLMTIME) {
1950 if (body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1951 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1952 "to %llu\n", inode->i_ino,
1953 LTIME_S(inode->i_mtime), body->mbo_mtime);
1954 LTIME_S(inode->i_mtime) = body->mbo_mtime;
1956 lli->lli_mtime = body->mbo_mtime;
1959 if (body->mbo_valid & OBD_MD_FLCTIME) {
1960 if (body->mbo_ctime > LTIME_S(inode->i_ctime))
1961 LTIME_S(inode->i_ctime) = body->mbo_ctime;
1962 lli->lli_ctime = body->mbo_ctime;
1965 /* Clear i_flags to remove S_NOSEC before permissions are updated */
1966 if (body->mbo_valid & OBD_MD_FLFLAGS)
1967 ll_update_inode_flags(inode, body->mbo_flags);
1968 if (body->mbo_valid & OBD_MD_FLMODE)
1969 inode->i_mode = (inode->i_mode & S_IFMT) |
1970 (body->mbo_mode & ~S_IFMT);
1972 if (body->mbo_valid & OBD_MD_FLTYPE)
1973 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1974 (body->mbo_mode & S_IFMT);
1976 LASSERT(inode->i_mode != 0);
1977 if (S_ISREG(inode->i_mode))
1978 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1979 LL_MAX_BLKSIZE_BITS);
1981 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1983 if (body->mbo_valid & OBD_MD_FLUID)
1984 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
1985 if (body->mbo_valid & OBD_MD_FLGID)
1986 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
1987 if (body->mbo_valid & OBD_MD_FLPROJID)
1988 lli->lli_projid = body->mbo_projid;
1989 if (body->mbo_valid & OBD_MD_FLNLINK)
1990 set_nlink(inode, body->mbo_nlink);
1991 if (body->mbo_valid & OBD_MD_FLRDEV)
1992 inode->i_rdev = old_decode_dev(body->mbo_rdev);
1994 if (body->mbo_valid & OBD_MD_FLID) {
1995 /* FID shouldn't be changed! */
1996 if (fid_is_sane(&lli->lli_fid)) {
1997 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
1998 "Trying to change FID "DFID
1999 " to the "DFID", inode "DFID"(%p)\n",
2000 PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
2001 PFID(ll_inode2fid(inode)), inode);
2003 lli->lli_fid = body->mbo_fid1;
2007 LASSERT(fid_seq(&lli->lli_fid) != 0);
2009 if (body->mbo_valid & OBD_MD_FLSIZE) {
2010 i_size_write(inode, body->mbo_size);
2012 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
2013 PFID(ll_inode2fid(inode)),
2014 (unsigned long long)body->mbo_size);
2016 if (body->mbo_valid & OBD_MD_FLBLOCKS)
2017 inode->i_blocks = body->mbo_blocks;
2020 if (body->mbo_valid & OBD_MD_TSTATE) {
2021 /* Set LLIF_FILE_RESTORING if restore ongoing and
2022 * clear it when done to ensure to start again
2023 * glimpsing updated attrs
2025 if (body->mbo_t_state & MS_RESTORE)
2026 ll_file_set_flag(lli, LLIF_FILE_RESTORING);
2028 ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
2034 int ll_read_inode2(struct inode *inode, void *opaque)
2036 struct lustre_md *md = opaque;
2037 struct ll_inode_info *lli = ll_i2info(inode);
2041 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2042 PFID(&lli->lli_fid), inode);
2044 /* Core attributes from the MDS first. This is a new inode, and
2045 * the VFS doesn't zero times in the core inode so we have to do
2046 * it ourselves. They will be overwritten by either MDS or OST
2047 * attributes - we just need to make sure they aren't newer. */
2048 LTIME_S(inode->i_mtime) = 0;
2049 LTIME_S(inode->i_atime) = 0;
2050 LTIME_S(inode->i_ctime) = 0;
2052 rc = ll_update_inode(inode, md);
2056 /* OIDEBUG(inode); */
2058 #ifdef HAVE_BACKING_DEV_INFO
2059 /* initializing backing dev info. */
2060 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2062 if (S_ISREG(inode->i_mode)) {
2063 struct ll_sb_info *sbi = ll_i2sbi(inode);
2064 inode->i_op = &ll_file_inode_operations;
2065 inode->i_fop = sbi->ll_fop;
2066 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2068 } else if (S_ISDIR(inode->i_mode)) {
2069 inode->i_op = &ll_dir_inode_operations;
2070 inode->i_fop = &ll_dir_operations;
2072 } else if (S_ISLNK(inode->i_mode)) {
2073 inode->i_op = &ll_fast_symlink_inode_operations;
2076 inode->i_op = &ll_special_inode_operations;
2078 init_special_inode(inode, inode->i_mode,
2087 void ll_delete_inode(struct inode *inode)
2089 struct ll_inode_info *lli = ll_i2info(inode);
2092 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2093 /* It is last chance to write out dirty pages,
2094 * otherwise we may lose data while umount */
2095 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
2097 truncate_inode_pages_final(&inode->i_data);
2099 LASSERTF(inode->i_data.nrpages == 0, "inode="DFID"(%p) nrpages=%lu, "
2100 "see https://jira.whamcloud.com/browse/LU-118\n",
2101 PFID(ll_inode2fid(inode)), inode, inode->i_data.nrpages);
2103 #ifdef HAVE_SBOPS_EVICT_INODE
2104 ll_clear_inode(inode);
2111 int ll_iocontrol(struct inode *inode, struct file *file,
2112 unsigned int cmd, unsigned long arg)
2114 struct ll_sb_info *sbi = ll_i2sbi(inode);
2115 struct ptlrpc_request *req = NULL;
2120 case FS_IOC_GETFLAGS: {
2121 struct mdt_body *body;
2122 struct md_op_data *op_data;
2124 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2125 0, 0, LUSTRE_OPC_ANY,
2127 if (IS_ERR(op_data))
2128 RETURN(PTR_ERR(op_data));
2130 op_data->op_valid = OBD_MD_FLFLAGS;
2131 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2132 ll_finish_md_op_data(op_data);
2134 CERROR("%s: failure inode "DFID": rc = %d\n",
2135 sbi->ll_md_exp->exp_obd->obd_name,
2136 PFID(ll_inode2fid(inode)), rc);
2140 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2142 flags = body->mbo_flags;
2144 ptlrpc_req_finished(req);
2146 RETURN(put_user(flags, (int __user *)arg));
2148 case FS_IOC_SETFLAGS: {
2150 struct md_op_data *op_data;
2151 struct cl_object *obj;
2152 struct fsxattr fa = { 0 };
2154 if (get_user(flags, (int __user *)arg))
2157 fa.fsx_projid = ll_i2info(inode)->lli_projid;
2158 if (flags & LUSTRE_PROJINHERIT_FL)
2159 fa.fsx_xflags = FS_XFLAG_PROJINHERIT;
2161 rc = ll_ioctl_check_project(inode, &fa);
2165 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2166 LUSTRE_OPC_ANY, NULL);
2167 if (IS_ERR(op_data))
2168 RETURN(PTR_ERR(op_data));
2170 op_data->op_attr_flags = flags;
2171 op_data->op_xvalid |= OP_XVALID_FLAGS;
2172 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2173 ll_finish_md_op_data(op_data);
2174 ptlrpc_req_finished(req);
2178 ll_update_inode_flags(inode, flags);
2180 obj = ll_i2info(inode)->lli_clob;
2184 OBD_ALLOC_PTR(attr);
2188 rc = cl_setattr_ost(obj, attr, OP_XVALID_FLAGS, flags);
2200 int ll_flush_ctx(struct inode *inode)
2202 struct ll_sb_info *sbi = ll_i2sbi(inode);
2204 CDEBUG(D_SEC, "flush context for user %d\n",
2205 from_kuid(&init_user_ns, current_uid()));
2207 obd_set_info_async(NULL, sbi->ll_md_exp,
2208 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2210 obd_set_info_async(NULL, sbi->ll_dt_exp,
2211 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2216 /* umount -f client means force down, don't save state */
2217 void ll_umount_begin(struct super_block *sb)
2219 struct ll_sb_info *sbi = ll_s2sbi(sb);
2220 struct obd_device *obd;
2221 struct obd_ioctl_data *ioc_data;
2222 struct l_wait_info lwi;
2223 wait_queue_head_t waitq;
2226 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2227 sb->s_count, atomic_read(&sb->s_active));
2229 obd = class_exp2obd(sbi->ll_md_exp);
2231 CERROR("Invalid MDC connection handle %#llx\n",
2232 sbi->ll_md_exp->exp_handle.h_cookie);
2238 obd = class_exp2obd(sbi->ll_dt_exp);
2240 CERROR("Invalid LOV connection handle %#llx\n",
2241 sbi->ll_dt_exp->exp_handle.h_cookie);
2247 OBD_ALLOC_PTR(ioc_data);
2249 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2250 sizeof *ioc_data, ioc_data, NULL);
2252 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2253 sizeof *ioc_data, ioc_data, NULL);
2255 OBD_FREE_PTR(ioc_data);
2258 /* Really, we'd like to wait until there are no requests outstanding,
2259 * and then continue. For now, we just periodically checking for vfs
2260 * to decrement mnt_cnt and hope to finish it within 10sec.
2262 init_waitqueue_head(&waitq);
2263 lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(10),
2264 cfs_time_seconds(1), NULL, NULL);
2265 l_wait_event(waitq, may_umount(sbi->ll_mnt.mnt), &lwi);
2270 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2272 struct ll_sb_info *sbi = ll_s2sbi(sb);
2273 char *profilenm = get_profile_name(sb);
2277 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2278 read_only = *flags & MS_RDONLY;
2279 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2280 sizeof(KEY_READ_ONLY),
2281 KEY_READ_ONLY, sizeof(read_only),
2284 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2285 profilenm, read_only ?
2286 "read-only" : "read-write", err);
2291 sb->s_flags |= MS_RDONLY;
2293 sb->s_flags &= ~MS_RDONLY;
2295 if (sbi->ll_flags & LL_SBI_VERBOSE)
2296 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2297 read_only ? "read-only" : "read-write");
2303 * Cleanup the open handle that is cached on MDT-side.
2305 * For open case, the client side open handling thread may hit error
2306 * after the MDT grant the open. Under such case, the client should
2307 * send close RPC to the MDT as cleanup; otherwise, the open handle
2308 * on the MDT will be leaked there until the client umount or evicted.
2310 * In further, if someone unlinked the file, because the open handle
2311 * holds the reference on such file/object, then it will block the
2312 * subsequent threads that want to locate such object via FID.
2314 * \param[in] sb super block for this file-system
2315 * \param[in] open_req pointer to the original open request
2317 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2319 struct mdt_body *body;
2320 struct md_op_data *op_data;
2321 struct ptlrpc_request *close_req = NULL;
2322 struct obd_export *exp = ll_s2sbi(sb)->ll_md_exp;
2325 body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2326 OBD_ALLOC_PTR(op_data);
2327 if (op_data == NULL) {
2328 CWARN("%s: cannot allocate op_data to release open handle for "
2330 ll_get_fsname(sb, NULL, 0), PFID(&body->mbo_fid1));
2335 op_data->op_fid1 = body->mbo_fid1;
2336 op_data->op_open_handle = body->mbo_open_handle;
2337 op_data->op_mod_time = ktime_get_real_seconds();
2338 md_close(exp, op_data, NULL, &close_req);
2339 ptlrpc_req_finished(close_req);
2340 ll_finish_md_op_data(op_data);
2345 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2346 struct super_block *sb, struct lookup_intent *it)
2348 struct ll_sb_info *sbi = NULL;
2349 struct lustre_md md = { NULL };
2353 LASSERT(*inode || sb);
2354 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2355 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2356 sbi->ll_md_exp, &md);
2361 rc = ll_update_inode(*inode, &md);
2365 LASSERT(sb != NULL);
2368 * At this point server returns to client's same fid as client
2369 * generated for creating. So using ->fid1 is okay here.
2371 if (!fid_is_sane(&md.body->mbo_fid1)) {
2372 CERROR("%s: Fid is insane "DFID"\n",
2373 ll_get_fsname(sb, NULL, 0),
2374 PFID(&md.body->mbo_fid1));
2375 GOTO(out, rc = -EINVAL);
2378 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2379 sbi->ll_flags & LL_SBI_32BIT_API),
2381 if (IS_ERR(*inode)) {
2382 #ifdef CONFIG_FS_POSIX_ACL
2384 posix_acl_release(md.posix_acl);
2385 md.posix_acl = NULL;
2388 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2390 CERROR("new_inode -fatal: rc %d\n", rc);
2395 /* Handling piggyback layout lock.
2396 * Layout lock can be piggybacked by getattr and open request.
2397 * The lsm can be applied to inode only if it comes with a layout lock
2398 * otherwise correct layout may be overwritten, for example:
2399 * 1. proc1: mdt returns a lsm but not granting layout
2400 * 2. layout was changed by another client
2401 * 3. proc2: refresh layout and layout lock granted
2402 * 4. proc1: to apply a stale layout */
2403 if (it != NULL && it->it_lock_mode != 0) {
2404 struct lustre_handle lockh;
2405 struct ldlm_lock *lock;
2407 lockh.cookie = it->it_lock_handle;
2408 lock = ldlm_handle2lock(&lockh);
2409 LASSERT(lock != NULL);
2410 if (ldlm_has_layout(lock)) {
2411 struct cl_object_conf conf;
2413 memset(&conf, 0, sizeof(conf));
2414 conf.coc_opc = OBJECT_CONF_SET;
2415 conf.coc_inode = *inode;
2416 conf.coc_lock = lock;
2417 conf.u.coc_layout = md.layout;
2418 (void)ll_layout_conf(*inode, &conf);
2420 LDLM_LOCK_PUT(lock);
2426 md_free_lustre_md(sbi->ll_md_exp, &md);
2429 if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2430 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2435 int ll_obd_statfs(struct inode *inode, void __user *arg)
2437 struct ll_sb_info *sbi = NULL;
2438 struct obd_export *exp;
2440 struct obd_ioctl_data *data = NULL;
2444 if (!inode || !(sbi = ll_i2sbi(inode)))
2445 GOTO(out_statfs, rc = -EINVAL);
2447 rc = obd_ioctl_getdata(&buf, &len, arg);
2449 GOTO(out_statfs, rc);
2452 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2453 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2454 GOTO(out_statfs, rc = -EINVAL);
2456 if (data->ioc_inllen1 != sizeof(__u32) ||
2457 data->ioc_inllen2 != sizeof(__u32) ||
2458 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2459 data->ioc_plen2 != sizeof(struct obd_uuid))
2460 GOTO(out_statfs, rc = -EINVAL);
2462 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2463 if (type & LL_STATFS_LMV)
2464 exp = sbi->ll_md_exp;
2465 else if (type & LL_STATFS_LOV)
2466 exp = sbi->ll_dt_exp;
2468 GOTO(out_statfs, rc = -ENODEV);
2470 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2472 GOTO(out_statfs, rc);
2474 OBD_FREE_LARGE(buf, len);
2479 * this is normally called in ll_fini_md_op_data(), but sometimes it needs to
2480 * be called early to avoid deadlock.
2482 void ll_unlock_md_op_lsm(struct md_op_data *op_data)
2484 if (op_data->op_mea2_sem) {
2485 up_read(op_data->op_mea2_sem);
2486 op_data->op_mea2_sem = NULL;
2489 if (op_data->op_mea1_sem) {
2490 up_read(op_data->op_mea1_sem);
2491 op_data->op_mea1_sem = NULL;
2495 /* this function prepares md_op_data hint for passing it down to MD stack. */
2496 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2497 struct inode *i1, struct inode *i2,
2498 const char *name, size_t namelen,
2499 __u32 mode, __u32 opc, void *data)
2501 LASSERT(i1 != NULL);
2504 /* Do not reuse namelen for something else. */
2506 return ERR_PTR(-EINVAL);
2508 if (namelen > ll_i2sbi(i1)->ll_namelen)
2509 return ERR_PTR(-ENAMETOOLONG);
2511 if (!lu_name_is_valid_2(name, namelen))
2512 return ERR_PTR(-EINVAL);
2515 if (op_data == NULL)
2516 OBD_ALLOC_PTR(op_data);
2518 if (op_data == NULL)
2519 return ERR_PTR(-ENOMEM);
2521 ll_i2gids(op_data->op_suppgids, i1, i2);
2522 op_data->op_fid1 = *ll_inode2fid(i1);
2523 op_data->op_default_stripe_offset = -1;
2525 if (S_ISDIR(i1->i_mode)) {
2526 down_read(&ll_i2info(i1)->lli_lsm_sem);
2527 op_data->op_mea1_sem = &ll_i2info(i1)->lli_lsm_sem;
2528 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2529 if (opc == LUSTRE_OPC_MKDIR)
2530 op_data->op_default_stripe_offset =
2531 ll_i2info(i1)->lli_def_stripe_offset;
2535 op_data->op_fid2 = *ll_inode2fid(i2);
2536 if (S_ISDIR(i2->i_mode)) {
2538 down_read(&ll_i2info(i2)->lli_lsm_sem);
2539 op_data->op_mea2_sem =
2540 &ll_i2info(i2)->lli_lsm_sem;
2542 op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2545 fid_zero(&op_data->op_fid2);
2548 if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2549 op_data->op_cli_flags |= CLI_HASH64;
2551 if (ll_need_32bit_api(ll_i2sbi(i1)))
2552 op_data->op_cli_flags |= CLI_API32;
2554 op_data->op_name = name;
2555 op_data->op_namelen = namelen;
2556 op_data->op_mode = mode;
2557 op_data->op_mod_time = ktime_get_real_seconds();
2558 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2559 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2560 op_data->op_cap = cfs_curproc_cap_pack();
2561 op_data->op_mds = 0;
2562 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2563 filename_is_volatile(name, namelen, &op_data->op_mds)) {
2564 op_data->op_bias |= MDS_CREATE_VOLATILE;
2566 op_data->op_data = data;
2571 void ll_finish_md_op_data(struct md_op_data *op_data)
2573 ll_unlock_md_op_lsm(op_data);
2574 security_release_secctx(op_data->op_file_secctx,
2575 op_data->op_file_secctx_size);
2576 OBD_FREE_PTR(op_data);
2579 #ifdef HAVE_SUPEROPS_USE_DENTRY
2580 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2582 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2585 struct ll_sb_info *sbi;
2587 #ifdef HAVE_SUPEROPS_USE_DENTRY
2588 LASSERT((seq != NULL) && (dentry != NULL));
2589 sbi = ll_s2sbi(dentry->d_sb);
2591 LASSERT((seq != NULL) && (vfs != NULL));
2592 sbi = ll_s2sbi(vfs->mnt_sb);
2595 if (sbi->ll_flags & LL_SBI_NOLCK)
2596 seq_puts(seq, ",nolock");
2598 if (sbi->ll_flags & LL_SBI_FLOCK)
2599 seq_puts(seq, ",flock");
2601 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2602 seq_puts(seq, ",localflock");
2604 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2605 seq_puts(seq, ",user_xattr");
2607 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2608 seq_puts(seq, ",lazystatfs");
2610 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2611 seq_puts(seq, ",user_fid2path");
2613 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2614 seq_puts(seq, ",always_ping");
2620 * Get obd name by cmd, and copy out to user space
2622 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2624 struct ll_sb_info *sbi = ll_i2sbi(inode);
2625 struct obd_device *obd;
2628 if (cmd == OBD_IOC_GETDTNAME)
2629 obd = class_exp2obd(sbi->ll_dt_exp);
2630 else if (cmd == OBD_IOC_GETMDNAME)
2631 obd = class_exp2obd(sbi->ll_md_exp);
2638 if (copy_to_user((void __user *)arg, obd->obd_name,
2639 strlen(obd->obd_name) + 1))
2646 * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2647 * fsname will be returned in this buffer; otherwise, a static buffer will be
2648 * used to store the fsname and returned to caller.
2650 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2652 static char fsname_static[MTI_NAME_MAXLEN];
2653 struct lustre_sb_info *lsi = s2lsi(sb);
2658 /* this means the caller wants to use static buffer
2659 * and it doesn't care about race. Usually this is
2660 * in error reporting path */
2661 buf = fsname_static;
2662 buflen = sizeof(fsname_static);
2665 len = strlen(lsi->lsi_lmd->lmd_profile);
2666 ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2667 if (ptr && (strcmp(ptr, "-client") == 0))
2670 if (unlikely(len >= buflen))
2672 strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2678 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2685 p.mnt = current->fs->root.mnt;
2687 path = d_path(&p, buf, bufsize);
2692 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2694 char *buf, *path = NULL;
2695 struct dentry *dentry = NULL;
2696 struct inode *inode = page->mapping->host;
2698 /* this can be called inside spin lock so use GFP_ATOMIC. */
2699 buf = (char *)__get_free_page(GFP_ATOMIC);
2701 dentry = d_find_alias(page->mapping->host);
2703 path = ll_d_path(dentry, buf, PAGE_SIZE);
2707 "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2708 "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2709 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2710 PFID(ll_inode2fid(inode)),
2711 (path && !IS_ERR(path)) ? path : "", ioret);
2717 free_page((unsigned long)buf);
2720 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2721 struct lov_user_md **kbuf)
2723 struct lov_user_md lum;
2727 if (copy_from_user(&lum, md, sizeof(lum)))
2730 lum_size = ll_lov_user_md_size(&lum);
2734 OBD_ALLOC(*kbuf, lum_size);
2738 if (copy_from_user(*kbuf, md, lum_size) != 0) {
2739 OBD_FREE(*kbuf, lum_size);
2747 * Compute llite root squash state after a change of root squash
2748 * configuration setting or add/remove of a lnet nid
2750 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2752 struct root_squash_info *squash = &sbi->ll_squash;
2755 struct lnet_process_id id;
2757 /* Update norootsquash flag */
2758 down_write(&squash->rsi_sem);
2759 if (list_empty(&squash->rsi_nosquash_nids))
2760 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2762 /* Do not apply root squash as soon as one of our NIDs is
2763 * in the nosquash_nids list */
2766 while (LNetGetId(i++, &id) != -ENOENT) {
2767 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2769 if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2775 sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2777 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2779 up_write(&squash->rsi_sem);
2783 * Parse linkea content to extract information about a given hardlink
2785 * \param[in] ldata - Initialized linkea data
2786 * \param[in] linkno - Link identifier
2787 * \param[out] parent_fid - The entry's parent FID
2788 * \param[out] ln - Entry name destination buffer
2790 * \retval 0 on success
2791 * \retval Appropriate negative error code on failure
2793 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2794 struct lu_fid *parent_fid, struct lu_name *ln)
2800 rc = linkea_init_with_rec(ldata);
2804 if (linkno >= ldata->ld_leh->leh_reccount)
2805 /* beyond last link */
2808 linkea_first_entry(ldata);
2809 for (idx = 0; ldata->ld_lee != NULL; idx++) {
2810 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2815 linkea_next_entry(ldata);
2825 * Get parent FID and name of an identified link. Operation is performed for
2826 * a given link number, letting the caller iterate over linkno to list one or
2827 * all links of an entry.
2829 * \param[in] file - File descriptor against which to perform the operation
2830 * \param[in,out] arg - User-filled structure containing the linkno to operate
2831 * on and the available size. It is eventually filled with
2832 * the requested information or left untouched on error
2834 * \retval - 0 on success
2835 * \retval - Appropriate negative error code on failure
2837 int ll_getparent(struct file *file, struct getparent __user *arg)
2839 struct inode *inode = file_inode(file);
2840 struct linkea_data *ldata;
2841 struct lu_buf buf = LU_BUF_NULL;
2843 struct lu_fid parent_fid;
2850 if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2851 !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2854 if (get_user(name_size, &arg->gp_name_size))
2857 if (get_user(linkno, &arg->gp_linkno))
2860 if (name_size > PATH_MAX)
2863 OBD_ALLOC(ldata, sizeof(*ldata));
2867 rc = linkea_data_new(ldata, &buf);
2869 GOTO(ldata_free, rc);
2871 #ifdef HAVE_XATTR_HANDLER_FLAGS
2872 rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
2873 buf.lb_len, OBD_MD_FLXATTR);
2875 rc = ll_getxattr(file_dentry(file), XATTR_NAME_LINK, buf.lb_buf,
2877 #endif /* HAVE_XATTR_HANDLER_FLAGS */
2881 rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2885 if (ln.ln_namelen >= name_size)
2886 GOTO(lb_free, rc = -EOVERFLOW);
2888 if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2889 GOTO(lb_free, rc = -EFAULT);
2891 if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2892 GOTO(lb_free, rc = -EFAULT);
2894 if (put_user('\0', arg->gp_name + ln.ln_namelen))
2895 GOTO(lb_free, rc = -EFAULT);
2900 OBD_FREE(ldata, sizeof(*ldata));