4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lustre/llite/llite_lib.c
34 * Lustre Light Super operations
37 #define DEBUG_SUBSYSTEM S_LLITE
39 #include <linux/module.h>
40 #include <linux/statfs.h>
41 #include <linux/time.h>
42 #include <linux/types.h>
43 #include <linux/version.h>
45 #include <linux/user_namespace.h>
46 #ifdef HAVE_UIDGID_HEADER
47 # include <linux/uidgid.h>
49 #include <linux/security.h>
51 #include <uapi/linux/lustre/lustre_ioctl.h>
52 #include <lustre_ha.h>
53 #include <lustre_dlm.h>
54 #include <lprocfs_status.h>
55 #include <lustre_disk.h>
56 #include <uapi/linux/lustre/lustre_param.h>
57 #include <lustre_log.h>
58 #include <cl_object.h>
59 #include <obd_cksum.h>
60 #include "llite_internal.h"
62 struct kmem_cache *ll_file_data_slab;
65 #define log2(n) ffz(~(n))
68 static struct ll_sb_info *ll_init_sbi(void)
70 struct ll_sb_info *sbi = NULL;
72 unsigned long lru_page_max;
82 spin_lock_init(&sbi->ll_lock);
83 mutex_init(&sbi->ll_lco.lco_lock);
84 spin_lock_init(&sbi->ll_pp_extent_lock);
85 spin_lock_init(&sbi->ll_process_lock);
86 sbi->ll_rw_stats_on = 0;
89 pages = si.totalram - si.totalhigh;
90 lru_page_max = pages / 2;
92 /* initialize ll_cache data */
93 sbi->ll_cache = cl_cache_init(lru_page_max);
94 if (sbi->ll_cache == NULL) {
95 OBD_FREE(sbi, sizeof(*sbi));
99 sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
100 SBI_DEFAULT_READAHEAD_MAX);
101 sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
102 sbi->ll_ra_info.ra_max_read_ahead_whole_pages = -1;
104 ll_generate_random_uuid(uuid);
105 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
106 CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
108 sbi->ll_flags |= LL_SBI_VERBOSE;
109 #ifdef ENABLE_CHECKSUM
110 sbi->ll_flags |= LL_SBI_CHECKSUM;
113 #ifdef HAVE_LRU_RESIZE_SUPPORT
114 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
116 sbi->ll_flags |= LL_SBI_LAZYSTATFS;
118 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
119 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
121 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
125 /* metadata statahead is enabled by default */
126 sbi->ll_sa_running_max = LL_SA_RUNNING_DEF;
127 sbi->ll_sa_max = LL_SA_RPC_DEF;
128 atomic_set(&sbi->ll_sa_total, 0);
129 atomic_set(&sbi->ll_sa_wrong, 0);
130 atomic_set(&sbi->ll_sa_running, 0);
131 atomic_set(&sbi->ll_agl_total, 0);
132 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
133 sbi->ll_flags |= LL_SBI_FAST_READ;
134 sbi->ll_flags |= LL_SBI_TINY_WRITE;
137 sbi->ll_squash.rsi_uid = 0;
138 sbi->ll_squash.rsi_gid = 0;
139 INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
140 init_rwsem(&sbi->ll_squash.rsi_sem);
145 static void ll_free_sbi(struct super_block *sb)
147 struct ll_sb_info *sbi = ll_s2sbi(sb);
151 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
152 cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
153 if (sbi->ll_cache != NULL) {
154 cl_cache_decref(sbi->ll_cache);
155 sbi->ll_cache = NULL;
157 OBD_FREE(sbi, sizeof(*sbi));
162 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
163 struct vfsmount *mnt)
165 struct inode *root = NULL;
166 struct ll_sb_info *sbi = ll_s2sbi(sb);
167 struct obd_statfs *osfs = NULL;
168 struct ptlrpc_request *request = NULL;
169 struct obd_connect_data *data = NULL;
170 struct obd_uuid *uuid;
171 struct md_op_data *op_data;
172 struct lustre_md lmd;
174 int size, err, checksum;
177 sbi->ll_md_obd = class_name2obd(md);
178 if (!sbi->ll_md_obd) {
179 CERROR("MD %s: not setup or attached\n", md);
193 /* pass client page size via ocd_grant_blkbits, the server should report
194 * back its backend blocksize for grant calculation purpose */
195 data->ocd_grant_blkbits = PAGE_SHIFT;
197 /* indicate MDT features supported by this client */
198 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
199 OBD_CONNECT_ATTRFID | OBD_CONNECT_GRANT |
200 OBD_CONNECT_VERSION | OBD_CONNECT_BRW_SIZE |
201 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
202 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
203 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
204 OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 |
205 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
206 OBD_CONNECT_64BITHASH |
207 OBD_CONNECT_EINPROGRESS |
208 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
209 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS|
210 OBD_CONNECT_MAX_EASIZE |
211 OBD_CONNECT_FLOCK_DEAD |
212 OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
213 OBD_CONNECT_OPEN_BY_FID |
214 OBD_CONNECT_DIR_STRIPE |
215 OBD_CONNECT_BULK_MBITS | OBD_CONNECT_CKSUM |
216 OBD_CONNECT_SUBTREE |
217 OBD_CONNECT_MULTIMODRPCS |
218 OBD_CONNECT_GRANT_PARAM |
219 OBD_CONNECT_SHORTIO | OBD_CONNECT_FLAGS2;
221 data->ocd_connect_flags2 = OBD_CONNECT2_FLR |
222 OBD_CONNECT2_LOCK_CONVERT |
223 OBD_CONNECT2_DIR_MIGRATE |
224 OBD_CONNECT2_SUM_STATFS |
225 OBD_CONNECT2_ARCHIVE_ID_ARRAY |
228 #ifdef HAVE_LRU_RESIZE_SUPPORT
229 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
230 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
232 #ifdef CONFIG_FS_POSIX_ACL
233 data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK |
234 OBD_CONNECT_LARGE_ACL;
237 data->ocd_cksum_types = obd_cksum_types_supported_client();
239 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
240 /* flag mdc connection as lightweight, only used for test
241 * purpose, use with care */
242 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
244 data->ocd_ibits_known = MDS_INODELOCK_FULL;
245 data->ocd_version = LUSTRE_VERSION_CODE;
247 if (sb->s_flags & MS_RDONLY)
248 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
249 if (sbi->ll_flags & LL_SBI_USER_XATTR)
250 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
253 /* Setting this indicates we correctly support S_NOSEC (See kernel
254 * commit 9e1f1de02c2275d7172e18dc4e7c2065777611bf)
256 sb->s_flags |= MS_NOSEC;
259 if (sbi->ll_flags & LL_SBI_FLOCK)
260 sbi->ll_fop = &ll_file_operations_flock;
261 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
262 sbi->ll_fop = &ll_file_operations;
264 sbi->ll_fop = &ll_file_operations_noflock;
266 /* always ping even if server suppress_pings */
267 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
268 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
270 obd_connect_set_secctx(data);
272 #if defined(CONFIG_SECURITY)
273 data->ocd_connect_flags2 |= OBD_CONNECT2_SELINUX_POLICY;
276 data->ocd_brw_size = MD_MAX_BRW_SIZE;
278 err = obd_connect(NULL, &sbi->ll_md_exp, sbi->ll_md_obd,
279 &sbi->ll_sb_uuid, data, NULL);
281 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
282 "recovery, of which this client is not a "
283 "part. Please wait for recovery to complete,"
284 " abort, or time out.\n", md);
287 CERROR("cannot connect to %s: rc = %d\n", md, err);
291 sbi->ll_md_exp->exp_connect_data = *data;
293 err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
294 LUSTRE_SEQ_METADATA);
296 CERROR("%s: Can't init metadata layer FID infrastructure, "
297 "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
301 /* For mount, we only need fs info from MDT0, and also in DNE, it
302 * can make sure the client can be mounted as long as MDT0 is
304 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
305 ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
306 OBD_STATFS_FOR_MDT0);
308 GOTO(out_md_fid, err);
310 /* This needs to be after statfs to ensure connect has finished.
311 * Note that "data" does NOT contain the valid connect reply.
312 * If connecting to a 1.8 server there will be no LMV device, so
313 * we can access the MDC export directly and exp_connect_flags will
314 * be non-zero, but if accessing an upgraded 2.1 server it will
315 * have the correct flags filled in.
316 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
317 valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
318 if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
319 valid != CLIENT_CONNECT_MDT_REQD) {
322 OBD_ALLOC_WAIT(buf, PAGE_SIZE);
323 obd_connect_flags2str(buf, PAGE_SIZE,
324 valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
325 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
326 "feature(s) needed for correct operation "
327 "of this client (%s). Please upgrade "
328 "server or downgrade client.\n",
329 sbi->ll_md_exp->exp_obd->obd_name, buf);
330 OBD_FREE(buf, PAGE_SIZE);
331 GOTO(out_md_fid, err = -EPROTO);
334 size = sizeof(*data);
335 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
336 KEY_CONN_DATA, &size, data);
338 CERROR("%s: Get connect data failed: rc = %d\n",
339 sbi->ll_md_exp->exp_obd->obd_name, err);
340 GOTO(out_md_fid, err);
343 LASSERT(osfs->os_bsize);
344 sb->s_blocksize = osfs->os_bsize;
345 sb->s_blocksize_bits = log2(osfs->os_bsize);
346 sb->s_magic = LL_SUPER_MAGIC;
347 sb->s_maxbytes = MAX_LFS_FILESIZE;
348 sbi->ll_namelen = osfs->os_namelen;
349 sbi->ll_mnt.mnt = current->fs->root.mnt;
351 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
352 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
353 LCONSOLE_INFO("Disabling user_xattr feature because "
354 "it is not supported on the server\n");
355 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
358 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
360 sb->s_flags |= MS_POSIXACL;
362 sbi->ll_flags |= LL_SBI_ACL;
364 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
366 sb->s_flags &= ~MS_POSIXACL;
368 sbi->ll_flags &= ~LL_SBI_ACL;
371 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
372 sbi->ll_flags |= LL_SBI_64BIT_HASH;
374 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
375 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
377 if (obd_connect_has_secctx(data))
378 sbi->ll_flags |= LL_SBI_FILE_SECCTX;
380 if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
381 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
382 LCONSOLE_INFO("%s: disabling xattr cache due to "
383 "unknown maximum xattr size.\n", dt);
384 } else if (!sbi->ll_xattr_cache_set) {
385 /* If xattr_cache is already set (no matter 0 or 1)
386 * during processing llog, it won't be enabled here. */
387 sbi->ll_flags |= LL_SBI_XATTR_CACHE;
388 sbi->ll_xattr_cache_enabled = 1;
392 sbi->ll_dt_obd = class_name2obd(dt);
393 if (!sbi->ll_dt_obd) {
394 CERROR("DT %s: not setup or attached\n", dt);
395 GOTO(out_md_fid, err = -ENODEV);
398 /* pass client page size via ocd_grant_blkbits, the server should report
399 * back its backend blocksize for grant calculation purpose */
400 data->ocd_grant_blkbits = PAGE_SHIFT;
402 /* indicate OST features supported by this client */
403 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
404 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
405 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
406 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
407 OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
408 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
409 OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
410 OBD_CONNECT_EINPROGRESS |
411 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
412 OBD_CONNECT_LAYOUTLOCK |
413 OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
414 OBD_CONNECT_BULK_MBITS | OBD_CONNECT_SHORTIO |
415 OBD_CONNECT_FLAGS2 | OBD_CONNECT_GRANT_SHRINK;
417 /* The client currently advertises support for OBD_CONNECT_LOCKAHEAD_OLD so it
418 * can interoperate with an older version of lockahead which was released prior
419 * to landing in master. This support will be dropped when 2.13 development
420 * starts. At the point, we should not just drop the connect flag (below), we
421 * should also remove the support in the code.
423 * Removing it means a few things:
424 * 1. Remove this section here
425 * 2. Remove CEF_NONBLOCK in ll_file_lockahead()
426 * 3. Remove function exp_connect_lockahead_old
427 * 4. Remove LDLM_FL_LOCKAHEAD_OLD_RESERVED in lustre_dlm_flags.h
429 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 12, 50, 0)
430 data->ocd_connect_flags |= OBD_CONNECT_LOCKAHEAD_OLD;
433 data->ocd_connect_flags2 = OBD_CONNECT2_LOCKAHEAD;
435 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
436 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
438 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
439 * disabled by default, because it can still be enabled on the
440 * fly via /sys. As a consequence, we still need to come to an
441 * agreement on the supported algorithms at connect time
443 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
445 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
446 data->ocd_cksum_types = OBD_CKSUM_ADLER;
448 data->ocd_cksum_types = obd_cksum_types_supported_client();
450 #ifdef HAVE_LRU_RESIZE_SUPPORT
451 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
453 /* always ping even if server suppress_pings */
454 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
455 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
457 CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
458 "ocd_grant: %d\n", data->ocd_connect_flags,
459 data->ocd_version, data->ocd_grant);
461 sbi->ll_dt_obd->obd_upcall.onu_owner = &sbi->ll_lco;
462 sbi->ll_dt_obd->obd_upcall.onu_upcall = cl_ocd_update;
464 data->ocd_brw_size = DT_MAX_BRW_SIZE;
466 err = obd_connect(NULL, &sbi->ll_dt_exp, sbi->ll_dt_obd,
467 &sbi->ll_sb_uuid, data, NULL);
469 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
470 "recovery, of which this client is not a "
471 "part. Please wait for recovery to "
472 "complete, abort, or time out.\n", dt);
475 CERROR("%s: Cannot connect to %s: rc = %d\n",
476 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
480 sbi->ll_dt_exp->exp_connect_data = *data;
482 /* Don't change value if it was specified in the config log */
483 if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages == -1)
484 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
485 max_t(unsigned long, SBI_DEFAULT_READAHEAD_WHOLE_MAX,
486 (data->ocd_brw_size >> PAGE_SHIFT));
488 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
489 LUSTRE_SEQ_METADATA);
491 CERROR("%s: Can't init data layer FID infrastructure, "
492 "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
496 mutex_lock(&sbi->ll_lco.lco_lock);
497 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
498 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
499 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
500 mutex_unlock(&sbi->ll_lco.lco_lock);
502 fid_zero(&sbi->ll_root_fid);
503 err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
506 CERROR("cannot mds_connect: rc = %d\n", err);
507 GOTO(out_lock_cn_cb, err);
509 if (!fid_is_sane(&sbi->ll_root_fid)) {
510 CERROR("%s: Invalid root fid "DFID" during mount\n",
511 sbi->ll_md_exp->exp_obd->obd_name,
512 PFID(&sbi->ll_root_fid));
513 GOTO(out_lock_cn_cb, err = -EINVAL);
515 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
517 sb->s_op = &lustre_super_operations;
518 #ifdef HAVE_XATTR_HANDLER_FLAGS
519 sb->s_xattr = ll_xattr_handlers;
521 #if THREAD_SIZE >= 8192 /*b=17630*/
522 sb->s_export_op = &lustre_export_operations;
526 * XXX: move this to after cbd setup? */
527 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
528 if (sbi->ll_flags & LL_SBI_ACL)
529 valid |= OBD_MD_FLACL;
531 OBD_ALLOC_PTR(op_data);
533 GOTO(out_lock_cn_cb, err = -ENOMEM);
535 op_data->op_fid1 = sbi->ll_root_fid;
536 op_data->op_mode = 0;
537 op_data->op_valid = valid;
539 err = md_getattr(sbi->ll_md_exp, op_data, &request);
541 OBD_FREE_PTR(op_data);
543 CERROR("%s: md_getattr failed for root: rc = %d\n",
544 sbi->ll_md_exp->exp_obd->obd_name, err);
545 GOTO(out_lock_cn_cb, err);
548 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
549 sbi->ll_md_exp, &lmd);
551 CERROR("failed to understand root inode md: rc = %d\n", err);
552 ptlrpc_req_finished(request);
553 GOTO(out_lock_cn_cb, err);
556 LASSERT(fid_is_sane(&sbi->ll_root_fid));
557 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
558 sbi->ll_flags & LL_SBI_32BIT_API),
560 md_free_lustre_md(sbi->ll_md_exp, &lmd);
561 ptlrpc_req_finished(request);
564 #ifdef CONFIG_FS_POSIX_ACL
566 posix_acl_release(lmd.posix_acl);
567 lmd.posix_acl = NULL;
570 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
572 CERROR("lustre_lite: bad iget4 for root\n");
576 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
577 if (sbi->ll_checksum_set) {
578 err = obd_set_info_async(NULL, sbi->ll_dt_exp,
579 sizeof(KEY_CHECKSUM), KEY_CHECKSUM,
580 sizeof(checksum), &checksum, NULL);
582 CERROR("%s: Set checksum failed: rc = %d\n",
583 sbi->ll_dt_exp->exp_obd->obd_name, err);
589 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
590 KEY_CACHE_SET, sizeof(*sbi->ll_cache),
591 sbi->ll_cache, NULL);
593 CERROR("%s: Set cache_set failed: rc = %d\n",
594 sbi->ll_dt_exp->exp_obd->obd_name, err);
598 sb->s_root = d_make_root(root);
599 if (sb->s_root == NULL) {
600 CERROR("%s: can't make root dentry\n",
601 ll_get_fsname(sb, NULL, 0));
602 GOTO(out_root, err = -ENOMEM);
604 #ifdef HAVE_DCACHE_LOCK
605 sb->s_root->d_op = &ll_d_ops;
608 sbi->ll_sdev_orig = sb->s_dev;
610 /* We set sb->s_dev equal on all lustre clients in order to support
611 * NFS export clustering. NFSD requires that the FSID be the same
613 /* s_dev is also used in lt_compare() to compare two fs, but that is
614 * only a node-local comparison. */
615 uuid = obd_get_uuid(sbi->ll_md_exp);
617 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
624 if (sbi->ll_dt_obd) {
625 err = sysfs_create_link(&sbi->ll_kset.kobj,
626 &sbi->ll_dt_obd->obd_kset.kobj,
627 sbi->ll_dt_obd->obd_type->typ_name);
629 CERROR("%s: could not register %s in llite: rc = %d\n",
630 dt, ll_get_fsname(sb, NULL, 0), err);
635 if (sbi->ll_md_obd) {
636 err = sysfs_create_link(&sbi->ll_kset.kobj,
637 &sbi->ll_md_obd->obd_kset.kobj,
638 sbi->ll_md_obd->obd_type->typ_name);
640 CERROR("%s: could not register %s in llite: rc = %d\n",
641 md, ll_get_fsname(sb, NULL, 0), err);
651 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
653 obd_disconnect(sbi->ll_dt_exp);
654 sbi->ll_dt_exp = NULL;
655 sbi->ll_dt_obd = NULL;
657 obd_fid_fini(sbi->ll_md_exp->exp_obd);
659 obd_disconnect(sbi->ll_md_exp);
660 sbi->ll_md_exp = NULL;
661 sbi->ll_md_obd = NULL;
670 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
674 size = sizeof(*lmmsize);
675 rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
676 KEY_MAX_EASIZE, &size, lmmsize);
678 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
679 sbi->ll_dt_exp->exp_obd->obd_name, rc);
684 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
685 KEY_MAX_EASIZE, &size, lmmsize);
687 CERROR("Get max mdsize error rc %d\n", rc);
693 * Get the value of the default_easize parameter.
695 * \see client_obd::cl_default_mds_easize
697 * \param[in] sbi superblock info for this filesystem
698 * \param[out] lmmsize pointer to storage location for value
700 * \retval 0 on success
701 * \retval negative negated errno on failure
703 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
708 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
709 KEY_DEFAULT_EASIZE, &size, lmmsize);
711 CERROR("Get default mdsize error rc %d\n", rc);
717 * Set the default_easize parameter to the given value.
719 * \see client_obd::cl_default_mds_easize
721 * \param[in] sbi superblock info for this filesystem
722 * \param[in] lmmsize the size to set
724 * \retval 0 on success
725 * \retval negative negated errno on failure
727 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
731 if (lmmsize < sizeof(struct lov_mds_md) ||
732 lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
735 rc = obd_set_info_async(NULL, sbi->ll_md_exp,
736 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
737 sizeof(int), &lmmsize, NULL);
742 static void client_common_put_super(struct super_block *sb)
744 struct ll_sb_info *sbi = ll_s2sbi(sb);
749 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
750 obd_disconnect(sbi->ll_dt_exp);
751 sbi->ll_dt_exp = NULL;
753 ll_debugfs_unregister_super(sb);
755 obd_fid_fini(sbi->ll_md_exp->exp_obd);
756 obd_disconnect(sbi->ll_md_exp);
757 sbi->ll_md_exp = NULL;
762 void ll_kill_super(struct super_block *sb)
764 struct ll_sb_info *sbi;
768 if (!(sb->s_flags & MS_ACTIVE))
772 /* we need restore s_dev from changed for clustred NFS before put_super
773 * because new kernels have cached s_dev and change sb->s_dev in
774 * put_super not affected real removing devices */
776 sb->s_dev = sbi->ll_sdev_orig;
777 sbi->ll_umounting = 1;
779 /* wait running statahead threads to quit */
780 while (atomic_read(&sbi->ll_sa_running) > 0) {
781 set_current_state(TASK_UNINTERRUPTIBLE);
782 schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
789 static inline int ll_set_opt(const char *opt, char *data, int fl)
791 if (strncmp(opt, data, strlen(opt)) != 0)
797 /* non-client-specific mount options are parsed in lmd_parse */
798 static int ll_options(char *options, struct ll_sb_info *sbi)
801 char *s1 = options, *s2;
802 int *flags = &sbi->ll_flags;
808 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
811 CDEBUG(D_SUPER, "next opt=%s\n", s1);
812 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
817 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
822 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
827 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
832 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
837 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
842 tmp = ll_set_opt("context", s1, 1);
845 tmp = ll_set_opt("fscontext", s1, 1);
848 tmp = ll_set_opt("defcontext", s1, 1);
851 tmp = ll_set_opt("rootcontext", s1, 1);
854 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
859 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
865 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
868 sbi->ll_checksum_set = 1;
871 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
874 sbi->ll_checksum_set = 1;
877 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
882 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
887 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
892 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
897 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
902 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
907 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
912 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
917 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
923 s2 = strchr(s1, ',');
931 void ll_lli_init(struct ll_inode_info *lli)
933 lli->lli_inode_magic = LLI_INODE_MAGIC;
935 spin_lock_init(&lli->lli_lock);
936 lli->lli_posix_acl = NULL;
937 /* Do not set lli_fid, it has been initialized already. */
938 fid_zero(&lli->lli_pfid);
939 lli->lli_mds_read_och = NULL;
940 lli->lli_mds_write_och = NULL;
941 lli->lli_mds_exec_och = NULL;
942 lli->lli_open_fd_read_count = 0;
943 lli->lli_open_fd_write_count = 0;
944 lli->lli_open_fd_exec_count = 0;
945 mutex_init(&lli->lli_och_mutex);
946 spin_lock_init(&lli->lli_agl_lock);
947 spin_lock_init(&lli->lli_layout_lock);
948 ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
949 lli->lli_clob = NULL;
951 init_rwsem(&lli->lli_xattrs_list_rwsem);
952 mutex_init(&lli->lli_xattrs_enq_lock);
954 LASSERT(lli->lli_vfs_inode.i_mode != 0);
955 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
956 mutex_init(&lli->lli_readdir_mutex);
957 lli->lli_opendir_key = NULL;
959 spin_lock_init(&lli->lli_sa_lock);
960 lli->lli_opendir_pid = 0;
961 lli->lli_sa_enabled = 0;
962 lli->lli_def_stripe_offset = -1;
963 init_rwsem(&lli->lli_lsm_sem);
965 mutex_init(&lli->lli_size_mutex);
966 lli->lli_symlink_name = NULL;
967 init_rwsem(&lli->lli_trunc_sem);
968 range_lock_tree_init(&lli->lli_write_tree);
969 init_rwsem(&lli->lli_glimpse_sem);
970 lli->lli_glimpse_time = ktime_set(0, 0);
971 INIT_LIST_HEAD(&lli->lli_agl_list);
972 lli->lli_agl_index = 0;
973 lli->lli_async_rc = 0;
975 mutex_init(&lli->lli_layout_mutex);
976 memset(lli->lli_jobid, 0, sizeof(lli->lli_jobid));
979 #define MAX_STRING_SIZE 128
981 #ifndef HAVE_SUPER_SETUP_BDI_NAME
983 #define LSI_BDI_INITIALIZED 0x00400000
985 #ifndef HAVE_BDI_CAP_MAP_COPY
986 # define BDI_CAP_MAP_COPY 0
989 static int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
991 struct lustre_sb_info *lsi = s2lsi(sb);
992 char buf[MAX_STRING_SIZE];
996 err = bdi_init(&lsi->lsi_bdi);
1000 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1001 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1002 lsi->lsi_bdi.name = "lustre";
1003 va_start(args, fmt);
1004 vsnprintf(buf, MAX_STRING_SIZE, fmt, args);
1006 err = bdi_register(&lsi->lsi_bdi, NULL, "%s", buf);
1009 sb->s_bdi = &lsi->lsi_bdi;
1013 #endif /* !HAVE_SUPER_SETUP_BDI_NAME */
1015 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
1017 struct lustre_profile *lprof = NULL;
1018 struct lustre_sb_info *lsi = s2lsi(sb);
1019 struct ll_sb_info *sbi = NULL;
1020 char *dt = NULL, *md = NULL;
1021 char *profilenm = get_profile_name(sb);
1022 struct config_llog_instance *cfg;
1023 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1024 const int instlen = 16 + 2;
1025 unsigned long cfg_instance = ll_get_cfg_instance(sb);
1026 char name[MAX_STRING_SIZE];
1034 /* for ASLR, to map between cfg_instance and hashed ptr */
1035 CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1036 profilenm, cfg_instance, sb);
1038 try_module_get(THIS_MODULE);
1042 GOTO(out_free_cfg, err = -ENOMEM);
1044 /* client additional sb info */
1045 lsi->lsi_llsbi = sbi = ll_init_sbi();
1047 GOTO(out_free_cfg, err = -ENOMEM);
1049 err = ll_options(lsi->lsi_lmd->lmd_opts, sbi);
1051 GOTO(out_free_cfg, err);
1053 err = super_setup_bdi_name(sb, "lustre-%016lx", cfg_instance);
1055 GOTO(out_free_cfg, err);
1057 #ifndef HAVE_DCACHE_LOCK
1058 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1059 sb->s_d_op = &ll_d_ops;
1062 len = strlen(profilenm);
1063 ptr = strrchr(profilenm, '-');
1064 if (ptr && (strcmp(ptr, "-client") == 0))
1068 snprintf(name, MAX_STRING_SIZE, "%.*s-%016lx", len,
1069 profilenm, cfg_instance);
1071 /* Call ll_debugfs_register_super() before lustre_process_log()
1072 * so that "llite.*.*" params can be processed correctly.
1074 err = ll_debugfs_register_super(sb, name);
1076 CERROR("%s: could not register mountpoint in llite: rc = %d\n",
1077 ll_get_fsname(sb, NULL, 0), err);
1081 /* The cfg_instance is a value unique to this super, in case some
1082 * joker tries to mount the same fs at two mount points.
1084 cfg->cfg_instance = cfg_instance;
1085 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1086 cfg->cfg_callback = class_config_llog_handler;
1087 cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1088 /* set up client obds */
1089 err = lustre_process_log(sb, profilenm, cfg);
1091 GOTO(out_debugfs, err);
1093 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1094 lprof = class_get_profile(profilenm);
1095 if (lprof == NULL) {
1096 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1097 " read from the MGS. Does that filesystem "
1098 "exist?\n", profilenm);
1099 GOTO(out_debugfs, err = -EINVAL);
1101 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1102 lprof->lp_md, lprof->lp_dt);
1104 dt_len = strlen(lprof->lp_dt) + instlen + 2;
1105 OBD_ALLOC(dt, dt_len);
1107 GOTO(out_profile, err = -ENOMEM);
1108 snprintf(dt, dt_len - 1, "%s-%016lx", lprof->lp_dt, cfg_instance);
1110 md_len = strlen(lprof->lp_md) + instlen + 2;
1111 OBD_ALLOC(md, md_len);
1113 GOTO(out_free_dt, err = -ENOMEM);
1114 snprintf(md, md_len - 1, "%s-%016lx", lprof->lp_md, cfg_instance);
1116 /* connections, registrations, sb setup */
1117 err = client_common_fill_super(sb, md, dt, mnt);
1119 GOTO(out_free_md, err);
1121 sbi->ll_client_common_fill_super_succeeded = 1;
1125 OBD_FREE(md, md_len);
1128 OBD_FREE(dt, dt_len);
1131 class_put_profile(lprof);
1134 ll_debugfs_unregister_super(sb);
1141 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1142 LCONSOLE_WARN("Mounted %s\n", profilenm);
1144 } /* ll_fill_super */
1146 void ll_put_super(struct super_block *sb)
1148 struct config_llog_instance cfg, params_cfg;
1149 struct obd_device *obd;
1150 struct lustre_sb_info *lsi = s2lsi(sb);
1151 struct ll_sb_info *sbi = ll_s2sbi(sb);
1152 char *profilenm = get_profile_name(sb);
1153 unsigned long cfg_instance = ll_get_cfg_instance(sb);
1155 int next, force = 1, rc = 0;
1159 GOTO(out_no_sbi, 0);
1161 /* Should replace instance_id with something better for ASLR */
1162 CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1163 profilenm, cfg_instance, sb);
1165 cfg.cfg_instance = cfg_instance;
1166 lustre_end_log(sb, profilenm, &cfg);
1168 params_cfg.cfg_instance = cfg_instance;
1169 lustre_end_log(sb, PARAMS_FILENAME, ¶ms_cfg);
1171 if (sbi->ll_md_exp) {
1172 obd = class_exp2obd(sbi->ll_md_exp);
1174 force = obd->obd_force;
1177 /* Wait for unstable pages to be committed to stable storage */
1179 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1180 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1181 atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1185 ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1186 if (force == 0 && rc != -EINTR)
1187 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1189 /* We need to set force before the lov_disconnect in
1190 lustre_common_put_super, since l_d cleans up osc's as well. */
1193 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1195 obd->obd_force = force;
1199 if (sbi->ll_client_common_fill_super_succeeded) {
1200 /* Only if client_common_fill_super succeeded */
1201 client_common_put_super(sb);
1205 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1206 class_manual_cleanup(obd);
1209 if (sbi->ll_flags & LL_SBI_VERBOSE)
1210 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1213 class_del_profile(profilenm);
1215 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1216 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1217 bdi_destroy(&lsi->lsi_bdi);
1218 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1223 lsi->lsi_llsbi = NULL;
1225 lustre_common_put_super(sb);
1227 cl_env_cache_purge(~0);
1229 module_put(THIS_MODULE);
1232 } /* client_put_super */
1234 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1236 struct inode *inode = NULL;
1238 /* NOTE: we depend on atomic igrab() -bzzz */
1239 lock_res_and_lock(lock);
1240 if (lock->l_resource->lr_lvb_inode) {
1241 struct ll_inode_info * lli;
1242 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1243 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1244 inode = igrab(lock->l_resource->lr_lvb_inode);
1246 inode = lock->l_resource->lr_lvb_inode;
1247 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1248 D_WARNING, lock, "lr_lvb_inode %p is "
1249 "bogus: magic %08x",
1250 lock->l_resource->lr_lvb_inode,
1251 lli->lli_inode_magic);
1255 unlock_res_and_lock(lock);
1259 void ll_dir_clear_lsm_md(struct inode *inode)
1261 struct ll_inode_info *lli = ll_i2info(inode);
1263 LASSERT(S_ISDIR(inode->i_mode));
1265 if (lli->lli_lsm_md != NULL) {
1266 lmv_free_memmd(lli->lli_lsm_md);
1267 lli->lli_lsm_md = NULL;
1271 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1272 const struct lu_fid *fid,
1273 struct lustre_md *md)
1275 struct ll_sb_info *sbi = ll_s2sbi(sb);
1276 struct mdt_body *body = md->body;
1277 struct inode *inode;
1281 ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1282 inode = iget_locked(sb, ino);
1283 if (inode == NULL) {
1284 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1285 ll_get_fsname(sb, NULL, 0), PFID(fid));
1286 RETURN(ERR_PTR(-ENOENT));
1289 if (inode->i_state & I_NEW) {
1290 struct ll_inode_info *lli = ll_i2info(inode);
1291 struct lmv_stripe_md *lsm = md->lmv;
1293 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1294 (body->mbo_mode & S_IFMT);
1295 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1298 LTIME_S(inode->i_mtime) = 0;
1299 LTIME_S(inode->i_atime) = 0;
1300 LTIME_S(inode->i_ctime) = 0;
1303 #ifdef HAVE_BACKING_DEV_INFO
1304 /* initializing backing dev info. */
1305 inode->i_mapping->backing_dev_info =
1306 &s2lsi(inode->i_sb)->lsi_bdi;
1308 inode->i_op = &ll_dir_inode_operations;
1309 inode->i_fop = &ll_dir_operations;
1310 lli->lli_fid = *fid;
1313 LASSERT(lsm != NULL);
1314 /* master object FID */
1315 lli->lli_pfid = body->mbo_fid1;
1316 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1317 lli, PFID(fid), PFID(&lli->lli_pfid));
1318 unlock_new_inode(inode);
1324 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1327 struct lmv_stripe_md *lsm = md->lmv;
1328 struct ll_inode_info *lli = ll_i2info(inode);
1331 LASSERT(lsm != NULL);
1333 CDEBUG(D_INODE, "%s: "DFID" set dir layout:\n",
1334 ll_get_fsname(inode->i_sb, NULL, 0),
1335 PFID(&lli->lli_fid));
1336 lsm_md_dump(D_INODE, lsm);
1338 /* XXX sigh, this lsm_root initialization should be in
1339 * LMV layer, but it needs ll_iget right now, so we
1340 * put this here right now. */
1341 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1342 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1343 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1344 /* Unfortunately ll_iget will call ll_update_inode,
1345 * where the initialization of slave inode is slightly
1346 * different, so it reset lsm_md to NULL to avoid
1347 * initializing lsm for slave inode. */
1348 lsm->lsm_md_oinfo[i].lmo_root =
1349 ll_iget_anon_dir(inode->i_sb, fid, md);
1350 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1351 int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1353 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1355 iput(lsm->lsm_md_oinfo[i].lmo_root);
1356 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1362 lli->lli_lsm_md = lsm;
1367 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1369 struct ll_inode_info *lli = ll_i2info(inode);
1370 struct lmv_stripe_md *lsm = md->lmv;
1375 LASSERT(S_ISDIR(inode->i_mode));
1376 CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1377 PFID(ll_inode2fid(inode)));
1380 * no striped information from request, lustre_md from req does not
1381 * include stripeEA, see ll_md_setattr()
1387 * normally dir layout doesn't change, only take read lock to check
1388 * that to avoid blocking other MD operations.
1390 if (lli->lli_lsm_md)
1391 down_read(&lli->lli_lsm_sem);
1393 down_write(&lli->lli_lsm_sem);
1396 * if dir layout mismatch, check whether version is increased, which
1397 * means layout is changed, this happens in dir migration and lfsck.
1399 if (lli->lli_lsm_md && !lsm_md_eq(lli->lli_lsm_md, lsm)) {
1400 if (lsm->lsm_md_layout_version <=
1401 lli->lli_lsm_md->lsm_md_layout_version) {
1402 CERROR("%s: "DFID" dir layout mismatch:\n",
1403 ll_get_fsname(inode->i_sb, NULL, 0),
1404 PFID(&lli->lli_fid));
1405 lsm_md_dump(D_ERROR, lli->lli_lsm_md);
1406 lsm_md_dump(D_ERROR, lsm);
1407 GOTO(unlock, rc = -EINVAL);
1410 /* layout changed, switch to write lock */
1411 up_read(&lli->lli_lsm_sem);
1412 down_write(&lli->lli_lsm_sem);
1413 ll_dir_clear_lsm_md(inode);
1416 /* set directory layout */
1417 if (!lli->lli_lsm_md) {
1418 struct cl_attr *attr;
1420 rc = ll_init_lsm_md(inode, md);
1421 up_write(&lli->lli_lsm_sem);
1425 /* set md->lmv to NULL, so the following free lustre_md
1426 * will not free this lsm */
1430 * md_merge_attr() may take long, since lsm is already set,
1431 * switch to read lock.
1433 down_read(&lli->lli_lsm_sem);
1435 OBD_ALLOC_PTR(attr);
1437 GOTO(unlock, rc = -ENOMEM);
1439 /* validate the lsm */
1440 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1441 ll_md_blocking_ast);
1447 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1448 md->body->mbo_nlink = attr->cat_nlink;
1449 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1450 md->body->mbo_size = attr->cat_size;
1451 if (md->body->mbo_valid & OBD_MD_FLATIME)
1452 md->body->mbo_atime = attr->cat_atime;
1453 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1454 md->body->mbo_ctime = attr->cat_ctime;
1455 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1456 md->body->mbo_mtime = attr->cat_mtime;
1461 up_read(&lli->lli_lsm_sem);
1466 void ll_clear_inode(struct inode *inode)
1468 struct ll_inode_info *lli = ll_i2info(inode);
1469 struct ll_sb_info *sbi = ll_i2sbi(inode);
1472 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1473 PFID(ll_inode2fid(inode)), inode);
1475 if (S_ISDIR(inode->i_mode)) {
1476 /* these should have been cleared in ll_file_release */
1477 LASSERT(lli->lli_opendir_key == NULL);
1478 LASSERT(lli->lli_sai == NULL);
1479 LASSERT(lli->lli_opendir_pid == 0);
1482 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1484 LASSERT(!lli->lli_open_fd_write_count);
1485 LASSERT(!lli->lli_open_fd_read_count);
1486 LASSERT(!lli->lli_open_fd_exec_count);
1488 if (lli->lli_mds_write_och)
1489 ll_md_real_close(inode, FMODE_WRITE);
1490 if (lli->lli_mds_exec_och)
1491 ll_md_real_close(inode, FMODE_EXEC);
1492 if (lli->lli_mds_read_och)
1493 ll_md_real_close(inode, FMODE_READ);
1495 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1496 OBD_FREE(lli->lli_symlink_name,
1497 strlen(lli->lli_symlink_name) + 1);
1498 lli->lli_symlink_name = NULL;
1501 ll_xattr_cache_destroy(inode);
1503 #ifdef CONFIG_FS_POSIX_ACL
1504 forget_all_cached_acls(inode);
1505 if (lli->lli_posix_acl) {
1506 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1507 posix_acl_release(lli->lli_posix_acl);
1508 lli->lli_posix_acl = NULL;
1511 lli->lli_inode_magic = LLI_INODE_DEAD;
1513 if (S_ISDIR(inode->i_mode))
1514 ll_dir_clear_lsm_md(inode);
1515 else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1516 LASSERT(list_empty(&lli->lli_agl_list));
1519 * XXX This has to be done before lsm is freed below, because
1520 * cl_object still uses inode lsm.
1522 cl_inode_fini(inode);
1527 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1529 struct lustre_md md;
1530 struct inode *inode = dentry->d_inode;
1531 struct ll_sb_info *sbi = ll_i2sbi(inode);
1532 struct ptlrpc_request *request = NULL;
1536 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1537 LUSTRE_OPC_ANY, NULL);
1538 if (IS_ERR(op_data))
1539 RETURN(PTR_ERR(op_data));
1541 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1543 ptlrpc_req_finished(request);
1544 if (rc == -ENOENT) {
1546 /* Unlinked special device node? Or just a race?
1547 * Pretend we done everything. */
1548 if (!S_ISREG(inode->i_mode) &&
1549 !S_ISDIR(inode->i_mode)) {
1550 ia_valid = op_data->op_attr.ia_valid;
1551 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1552 rc = simple_setattr(dentry, &op_data->op_attr);
1553 op_data->op_attr.ia_valid = ia_valid;
1555 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1556 CERROR("md_setattr fails: rc = %d\n", rc);
1561 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1562 sbi->ll_md_exp, &md);
1564 ptlrpc_req_finished(request);
1568 ia_valid = op_data->op_attr.ia_valid;
1569 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1570 * cache is not cleared yet. */
1571 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1572 if (S_ISREG(inode->i_mode))
1574 rc = simple_setattr(dentry, &op_data->op_attr);
1575 if (S_ISREG(inode->i_mode))
1576 inode_unlock(inode);
1577 op_data->op_attr.ia_valid = ia_valid;
1579 rc = ll_update_inode(inode, &md);
1580 ptlrpc_req_finished(request);
1585 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1586 * object(s) determine the file size and mtime. Otherwise, the MDS will
1587 * keep these values until such a time that objects are allocated for it.
1588 * We do the MDS operations first, as it is checking permissions for us.
1589 * We don't to the MDS RPC if there is nothing that we want to store there,
1590 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1591 * going to do an RPC anyways.
1593 * If we are doing a truncate, we will send the mtime and ctime updates
1594 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1595 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1598 * In case of HSMimport, we only set attr on MDS.
1600 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr,
1601 enum op_xvalid xvalid, bool hsm_import)
1603 struct inode *inode = dentry->d_inode;
1604 struct ll_inode_info *lli = ll_i2info(inode);
1605 struct md_op_data *op_data = NULL;
1609 CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1610 "valid %x, hsm_import %d\n",
1611 ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1612 inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1615 if (attr->ia_valid & ATTR_SIZE) {
1616 /* Check new size against VFS/VM file size limit and rlimit */
1617 rc = inode_newsize_ok(inode, attr->ia_size);
1621 /* The maximum Lustre file size is variable, based on the
1622 * OST maximum object size and number of stripes. This
1623 * needs another check in addition to the VFS check above. */
1624 if (attr->ia_size > ll_file_maxbytes(inode)) {
1625 CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
1626 PFID(&lli->lli_fid), attr->ia_size,
1627 ll_file_maxbytes(inode));
1631 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1634 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1635 if (attr->ia_valid & TIMES_SET_FLAGS) {
1636 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1637 !cfs_capable(CFS_CAP_FOWNER))
1641 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1642 if (!(xvalid & OP_XVALID_CTIME_SET) &&
1643 (attr->ia_valid & ATTR_CTIME)) {
1644 attr->ia_ctime = current_time(inode);
1645 xvalid |= OP_XVALID_CTIME_SET;
1647 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1648 (attr->ia_valid & ATTR_ATIME)) {
1649 attr->ia_atime = current_time(inode);
1650 attr->ia_valid |= ATTR_ATIME_SET;
1652 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1653 (attr->ia_valid & ATTR_MTIME)) {
1654 attr->ia_mtime = current_time(inode);
1655 attr->ia_valid |= ATTR_MTIME_SET;
1658 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1659 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %llu\n",
1660 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1661 (s64)ktime_get_real_seconds());
1663 if (S_ISREG(inode->i_mode)) {
1664 if (attr->ia_valid & ATTR_SIZE)
1665 inode_dio_write_done(inode);
1666 inode_unlock(inode);
1669 /* We always do an MDS RPC, even if we're only changing the size;
1670 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1672 OBD_ALLOC_PTR(op_data);
1673 if (op_data == NULL)
1674 GOTO(out, rc = -ENOMEM);
1676 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1677 /* If we are changing file size, file content is
1678 * modified, flag it.
1680 xvalid |= OP_XVALID_OWNEROVERRIDE;
1681 op_data->op_bias |= MDS_DATA_MODIFIED;
1682 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1685 if (attr->ia_valid & ATTR_FILE) {
1686 struct ll_file_data *fd = LUSTRE_FPRIVATE(attr->ia_file);
1688 if (fd->fd_lease_och)
1689 op_data->op_bias |= MDS_TRUNC_KEEP_LEASE;
1692 op_data->op_attr = *attr;
1693 op_data->op_xvalid = xvalid;
1695 rc = ll_md_setattr(dentry, op_data);
1699 if (!S_ISREG(inode->i_mode) || hsm_import)
1702 if (attr->ia_valid & (ATTR_SIZE | ATTR_ATIME | ATTR_ATIME_SET |
1703 ATTR_MTIME | ATTR_MTIME_SET | ATTR_CTIME) ||
1704 xvalid & OP_XVALID_CTIME_SET) {
1705 /* For truncate and utimes sending attributes to OSTs, setting
1706 * mtime/atime to the past will be performed under PW [0:EOF]
1707 * extent lock (new_size:EOF for truncate). It may seem
1708 * excessive to send mtime/atime updates to OSTs when not
1709 * setting times to past, but it is necessary due to possible
1710 * time de-synchronization between MDT inode and OST objects
1712 rc = cl_setattr_ost(lli->lli_clob, attr, xvalid, 0);
1715 /* If the file was restored, it needs to set dirty flag.
1717 * We've already sent MDS_DATA_MODIFIED flag in
1718 * ll_md_setattr() for truncate. However, the MDT refuses to
1719 * set the HS_DIRTY flag on released files, so we have to set
1720 * it again if the file has been restored. Please check how
1721 * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1723 * Please notice that if the file is not released, the previous
1724 * MDS_DATA_MODIFIED has taken effect and usually
1725 * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1726 * This way we can save an RPC for common open + trunc
1728 if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1729 struct hsm_state_set hss = {
1730 .hss_valid = HSS_SETMASK,
1731 .hss_setmask = HS_DIRTY,
1735 rc2 = ll_hsm_state_set(inode, &hss);
1736 /* truncate and write can happen at the same time, so that
1737 * the file can be set modified even though the file is not
1738 * restored from released state, and ll_hsm_state_set() is
1739 * not applicable for the file, and rc2 < 0 is normal in this
1742 CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1743 PFID(ll_inode2fid(inode)), rc2);
1748 if (op_data != NULL)
1749 ll_finish_md_op_data(op_data);
1751 if (S_ISREG(inode->i_mode)) {
1753 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1754 inode_dio_wait(inode);
1755 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
1756 * flag. ll_update_inode (called from ll_md_setattr), clears
1757 * inode flags, so there is a gap where S_NOSEC is not set.
1758 * This can cause a writer to take the i_mutex unnecessarily,
1759 * but this is safe to do and should be rare. */
1760 inode_has_no_xattr(inode);
1763 ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1764 LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1769 int ll_setattr(struct dentry *de, struct iattr *attr)
1771 int mode = de->d_inode->i_mode;
1772 enum op_xvalid xvalid = 0;
1774 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1775 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1776 xvalid |= OP_XVALID_OWNEROVERRIDE;
1778 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1779 (ATTR_SIZE|ATTR_MODE)) &&
1780 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1781 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1782 !(attr->ia_mode & S_ISGID))))
1783 attr->ia_valid |= ATTR_FORCE;
1785 if ((attr->ia_valid & ATTR_MODE) &&
1787 !(attr->ia_mode & S_ISUID) &&
1788 !(attr->ia_valid & ATTR_KILL_SUID))
1789 attr->ia_valid |= ATTR_KILL_SUID;
1791 if ((attr->ia_valid & ATTR_MODE) &&
1792 ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1793 !(attr->ia_mode & S_ISGID) &&
1794 !(attr->ia_valid & ATTR_KILL_SGID))
1795 attr->ia_valid |= ATTR_KILL_SGID;
1797 return ll_setattr_raw(de, attr, xvalid, false);
1800 int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
1803 struct obd_statfs obd_osfs;
1808 max_age = ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS;
1810 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1812 CERROR("md_statfs fails: rc = %d\n", rc);
1816 osfs->os_type = LL_SUPER_MAGIC;
1818 CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1819 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1821 if (osfs->os_state & OS_STATE_SUM)
1824 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1825 flags |= OBD_STATFS_NODELAY;
1827 rc = obd_statfs(NULL, sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1829 CERROR("obd_statfs fails: rc = %d\n", rc);
1833 CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1834 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1837 osfs->os_bsize = obd_osfs.os_bsize;
1838 osfs->os_blocks = obd_osfs.os_blocks;
1839 osfs->os_bfree = obd_osfs.os_bfree;
1840 osfs->os_bavail = obd_osfs.os_bavail;
1842 /* If we don't have as many objects free on the OST as inodes
1843 * on the MDS, we reduce the total number of inodes to
1844 * compensate, so that the "inodes in use" number is correct.
1846 if (obd_osfs.os_ffree < osfs->os_ffree) {
1847 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1849 osfs->os_ffree = obd_osfs.os_ffree;
1855 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1857 struct super_block *sb = de->d_sb;
1858 struct obd_statfs osfs;
1859 __u64 fsid = huge_encode_dev(sb->s_dev);
1862 CDEBUG(D_VFSTRACE, "VFS Op: at %llu jiffies\n", get_jiffies_64());
1863 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1865 /* Some amount of caching on the client is allowed */
1866 rc = ll_statfs_internal(ll_s2sbi(sb), &osfs, OBD_STATFS_SUM);
1870 statfs_unpack(sfs, &osfs);
1872 /* We need to downshift for all 32-bit kernels, because we can't
1873 * tell if the kernel is being called via sys_statfs64() or not.
1874 * Stop before overflowing f_bsize - in which case it is better
1875 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1876 if (sizeof(long) < 8) {
1877 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1880 osfs.os_blocks >>= 1;
1881 osfs.os_bfree >>= 1;
1882 osfs.os_bavail >>= 1;
1886 sfs->f_blocks = osfs.os_blocks;
1887 sfs->f_bfree = osfs.os_bfree;
1888 sfs->f_bavail = osfs.os_bavail;
1889 sfs->f_fsid.val[0] = (__u32)fsid;
1890 sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1894 void ll_inode_size_lock(struct inode *inode)
1896 struct ll_inode_info *lli;
1898 LASSERT(!S_ISDIR(inode->i_mode));
1900 lli = ll_i2info(inode);
1901 mutex_lock(&lli->lli_size_mutex);
1904 void ll_inode_size_unlock(struct inode *inode)
1906 struct ll_inode_info *lli;
1908 lli = ll_i2info(inode);
1909 mutex_unlock(&lli->lli_size_mutex);
1912 void ll_update_inode_flags(struct inode *inode, int ext_flags)
1914 inode->i_flags = ll_ext_to_inode_flags(ext_flags);
1915 if (ext_flags & LUSTRE_PROJINHERIT_FL)
1916 ll_file_set_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
1918 ll_file_clear_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
1921 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1923 struct ll_inode_info *lli = ll_i2info(inode);
1924 struct mdt_body *body = md->body;
1925 struct ll_sb_info *sbi = ll_i2sbi(inode);
1928 if (body->mbo_valid & OBD_MD_FLEASIZE) {
1929 rc = cl_file_inode_init(inode, md);
1934 if (S_ISDIR(inode->i_mode)) {
1935 rc = ll_update_lsm_md(inode, md);
1940 #ifdef CONFIG_FS_POSIX_ACL
1941 if (body->mbo_valid & OBD_MD_FLACL) {
1942 spin_lock(&lli->lli_lock);
1943 if (lli->lli_posix_acl)
1944 posix_acl_release(lli->lli_posix_acl);
1945 lli->lli_posix_acl = md->posix_acl;
1946 spin_unlock(&lli->lli_lock);
1949 inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1950 sbi->ll_flags & LL_SBI_32BIT_API);
1951 inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1953 if (body->mbo_valid & OBD_MD_FLATIME) {
1954 if (body->mbo_atime > LTIME_S(inode->i_atime))
1955 LTIME_S(inode->i_atime) = body->mbo_atime;
1956 lli->lli_atime = body->mbo_atime;
1959 if (body->mbo_valid & OBD_MD_FLMTIME) {
1960 if (body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1961 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1962 "to %llu\n", inode->i_ino,
1963 LTIME_S(inode->i_mtime), body->mbo_mtime);
1964 LTIME_S(inode->i_mtime) = body->mbo_mtime;
1966 lli->lli_mtime = body->mbo_mtime;
1969 if (body->mbo_valid & OBD_MD_FLCTIME) {
1970 if (body->mbo_ctime > LTIME_S(inode->i_ctime))
1971 LTIME_S(inode->i_ctime) = body->mbo_ctime;
1972 lli->lli_ctime = body->mbo_ctime;
1975 /* Clear i_flags to remove S_NOSEC before permissions are updated */
1976 if (body->mbo_valid & OBD_MD_FLFLAGS)
1977 ll_update_inode_flags(inode, body->mbo_flags);
1978 if (body->mbo_valid & OBD_MD_FLMODE)
1979 inode->i_mode = (inode->i_mode & S_IFMT) |
1980 (body->mbo_mode & ~S_IFMT);
1982 if (body->mbo_valid & OBD_MD_FLTYPE)
1983 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1984 (body->mbo_mode & S_IFMT);
1986 LASSERT(inode->i_mode != 0);
1987 if (S_ISREG(inode->i_mode))
1988 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1989 LL_MAX_BLKSIZE_BITS);
1991 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1993 if (body->mbo_valid & OBD_MD_FLUID)
1994 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
1995 if (body->mbo_valid & OBD_MD_FLGID)
1996 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
1997 if (body->mbo_valid & OBD_MD_FLPROJID)
1998 lli->lli_projid = body->mbo_projid;
1999 if (body->mbo_valid & OBD_MD_FLNLINK)
2000 set_nlink(inode, body->mbo_nlink);
2001 if (body->mbo_valid & OBD_MD_FLRDEV)
2002 inode->i_rdev = old_decode_dev(body->mbo_rdev);
2004 if (body->mbo_valid & OBD_MD_FLID) {
2005 /* FID shouldn't be changed! */
2006 if (fid_is_sane(&lli->lli_fid)) {
2007 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
2008 "Trying to change FID "DFID
2009 " to the "DFID", inode "DFID"(%p)\n",
2010 PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
2011 PFID(ll_inode2fid(inode)), inode);
2013 lli->lli_fid = body->mbo_fid1;
2017 LASSERT(fid_seq(&lli->lli_fid) != 0);
2019 if (body->mbo_valid & OBD_MD_FLSIZE) {
2020 i_size_write(inode, body->mbo_size);
2022 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
2023 PFID(ll_inode2fid(inode)),
2024 (unsigned long long)body->mbo_size);
2026 if (body->mbo_valid & OBD_MD_FLBLOCKS)
2027 inode->i_blocks = body->mbo_blocks;
2030 if (body->mbo_valid & OBD_MD_TSTATE) {
2031 /* Set LLIF_FILE_RESTORING if restore ongoing and
2032 * clear it when done to ensure to start again
2033 * glimpsing updated attrs
2035 if (body->mbo_t_state & MS_RESTORE)
2036 ll_file_set_flag(lli, LLIF_FILE_RESTORING);
2038 ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
2044 int ll_read_inode2(struct inode *inode, void *opaque)
2046 struct lustre_md *md = opaque;
2047 struct ll_inode_info *lli = ll_i2info(inode);
2051 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2052 PFID(&lli->lli_fid), inode);
2054 /* Core attributes from the MDS first. This is a new inode, and
2055 * the VFS doesn't zero times in the core inode so we have to do
2056 * it ourselves. They will be overwritten by either MDS or OST
2057 * attributes - we just need to make sure they aren't newer. */
2058 LTIME_S(inode->i_mtime) = 0;
2059 LTIME_S(inode->i_atime) = 0;
2060 LTIME_S(inode->i_ctime) = 0;
2062 rc = ll_update_inode(inode, md);
2066 /* OIDEBUG(inode); */
2068 #ifdef HAVE_BACKING_DEV_INFO
2069 /* initializing backing dev info. */
2070 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2072 if (S_ISREG(inode->i_mode)) {
2073 struct ll_sb_info *sbi = ll_i2sbi(inode);
2074 inode->i_op = &ll_file_inode_operations;
2075 inode->i_fop = sbi->ll_fop;
2076 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2078 } else if (S_ISDIR(inode->i_mode)) {
2079 inode->i_op = &ll_dir_inode_operations;
2080 inode->i_fop = &ll_dir_operations;
2082 } else if (S_ISLNK(inode->i_mode)) {
2083 inode->i_op = &ll_fast_symlink_inode_operations;
2086 inode->i_op = &ll_special_inode_operations;
2088 init_special_inode(inode, inode->i_mode,
2097 void ll_delete_inode(struct inode *inode)
2099 struct ll_inode_info *lli = ll_i2info(inode);
2100 struct address_space *mapping = &inode->i_data;
2101 unsigned long nrpages;
2104 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2105 /* It is last chance to write out dirty pages,
2106 * otherwise we may lose data while umount */
2107 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
2109 truncate_inode_pages_final(mapping);
2111 /* Workaround for LU-118: Note nrpages may not be totally updated when
2112 * truncate_inode_pages() returns, as there can be a page in the process
2113 * of deletion (inside __delete_from_page_cache()) in the specified
2114 * range. Thus mapping->nrpages can be non-zero when this function
2115 * returns even after truncation of the whole mapping. Only do this if
2116 * npages isn't already zero.
2118 nrpages = mapping->nrpages;
2120 spin_lock_irq(&mapping->tree_lock);
2121 nrpages = mapping->nrpages;
2122 spin_unlock_irq(&mapping->tree_lock);
2123 } /* Workaround end */
2125 LASSERTF(nrpages == 0, "%s: inode="DFID"(%p) nrpages=%lu, "
2126 "see https://jira.whamcloud.com/browse/LU-118\n",
2127 ll_get_fsname(inode->i_sb, NULL, 0),
2128 PFID(ll_inode2fid(inode)), inode, nrpages);
2130 #ifdef HAVE_SBOPS_EVICT_INODE
2131 ll_clear_inode(inode);
2138 int ll_iocontrol(struct inode *inode, struct file *file,
2139 unsigned int cmd, unsigned long arg)
2141 struct ll_sb_info *sbi = ll_i2sbi(inode);
2142 struct ptlrpc_request *req = NULL;
2147 case FS_IOC_GETFLAGS: {
2148 struct mdt_body *body;
2149 struct md_op_data *op_data;
2151 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2152 0, 0, LUSTRE_OPC_ANY,
2154 if (IS_ERR(op_data))
2155 RETURN(PTR_ERR(op_data));
2157 op_data->op_valid = OBD_MD_FLFLAGS;
2158 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2159 ll_finish_md_op_data(op_data);
2161 CERROR("%s: failure inode "DFID": rc = %d\n",
2162 sbi->ll_md_exp->exp_obd->obd_name,
2163 PFID(ll_inode2fid(inode)), rc);
2167 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2169 flags = body->mbo_flags;
2171 ptlrpc_req_finished(req);
2173 RETURN(put_user(flags, (int __user *)arg));
2175 case FS_IOC_SETFLAGS: {
2177 struct md_op_data *op_data;
2178 struct cl_object *obj;
2179 struct fsxattr fa = { 0 };
2181 if (get_user(flags, (int __user *)arg))
2184 fa.fsx_projid = ll_i2info(inode)->lli_projid;
2185 if (flags & LUSTRE_PROJINHERIT_FL)
2186 fa.fsx_xflags = FS_XFLAG_PROJINHERIT;
2188 rc = ll_ioctl_check_project(inode, &fa);
2192 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2193 LUSTRE_OPC_ANY, NULL);
2194 if (IS_ERR(op_data))
2195 RETURN(PTR_ERR(op_data));
2197 op_data->op_attr_flags = flags;
2198 op_data->op_xvalid |= OP_XVALID_FLAGS;
2199 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2200 ll_finish_md_op_data(op_data);
2201 ptlrpc_req_finished(req);
2205 ll_update_inode_flags(inode, flags);
2207 obj = ll_i2info(inode)->lli_clob;
2211 OBD_ALLOC_PTR(attr);
2215 rc = cl_setattr_ost(obj, attr, OP_XVALID_FLAGS, flags);
2227 int ll_flush_ctx(struct inode *inode)
2229 struct ll_sb_info *sbi = ll_i2sbi(inode);
2231 CDEBUG(D_SEC, "flush context for user %d\n",
2232 from_kuid(&init_user_ns, current_uid()));
2234 obd_set_info_async(NULL, sbi->ll_md_exp,
2235 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2237 obd_set_info_async(NULL, sbi->ll_dt_exp,
2238 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2243 /* umount -f client means force down, don't save state */
2244 void ll_umount_begin(struct super_block *sb)
2246 struct ll_sb_info *sbi = ll_s2sbi(sb);
2247 struct obd_device *obd;
2248 struct obd_ioctl_data *ioc_data;
2249 struct l_wait_info lwi;
2250 wait_queue_head_t waitq;
2253 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2254 sb->s_count, atomic_read(&sb->s_active));
2256 obd = class_exp2obd(sbi->ll_md_exp);
2258 CERROR("Invalid MDC connection handle %#llx\n",
2259 sbi->ll_md_exp->exp_handle.h_cookie);
2265 obd = class_exp2obd(sbi->ll_dt_exp);
2267 CERROR("Invalid LOV connection handle %#llx\n",
2268 sbi->ll_dt_exp->exp_handle.h_cookie);
2274 OBD_ALLOC_PTR(ioc_data);
2276 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2277 sizeof *ioc_data, ioc_data, NULL);
2279 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2280 sizeof *ioc_data, ioc_data, NULL);
2282 OBD_FREE_PTR(ioc_data);
2285 /* Really, we'd like to wait until there are no requests outstanding,
2286 * and then continue. For now, we just periodically checking for vfs
2287 * to decrement mnt_cnt and hope to finish it within 10sec.
2289 init_waitqueue_head(&waitq);
2290 lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(10),
2291 cfs_time_seconds(1), NULL, NULL);
2292 l_wait_event(waitq, may_umount(sbi->ll_mnt.mnt), &lwi);
2297 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2299 struct ll_sb_info *sbi = ll_s2sbi(sb);
2300 char *profilenm = get_profile_name(sb);
2304 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2305 read_only = *flags & MS_RDONLY;
2306 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2307 sizeof(KEY_READ_ONLY),
2308 KEY_READ_ONLY, sizeof(read_only),
2311 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2312 profilenm, read_only ?
2313 "read-only" : "read-write", err);
2318 sb->s_flags |= MS_RDONLY;
2320 sb->s_flags &= ~MS_RDONLY;
2322 if (sbi->ll_flags & LL_SBI_VERBOSE)
2323 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2324 read_only ? "read-only" : "read-write");
2330 * Cleanup the open handle that is cached on MDT-side.
2332 * For open case, the client side open handling thread may hit error
2333 * after the MDT grant the open. Under such case, the client should
2334 * send close RPC to the MDT as cleanup; otherwise, the open handle
2335 * on the MDT will be leaked there until the client umount or evicted.
2337 * In further, if someone unlinked the file, because the open handle
2338 * holds the reference on such file/object, then it will block the
2339 * subsequent threads that want to locate such object via FID.
2341 * \param[in] sb super block for this file-system
2342 * \param[in] open_req pointer to the original open request
2344 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2346 struct mdt_body *body;
2347 struct md_op_data *op_data;
2348 struct ptlrpc_request *close_req = NULL;
2349 struct obd_export *exp = ll_s2sbi(sb)->ll_md_exp;
2352 body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2353 OBD_ALLOC_PTR(op_data);
2354 if (op_data == NULL) {
2355 CWARN("%s: cannot allocate op_data to release open handle for "
2357 ll_get_fsname(sb, NULL, 0), PFID(&body->mbo_fid1));
2362 op_data->op_fid1 = body->mbo_fid1;
2363 op_data->op_open_handle = body->mbo_open_handle;
2364 op_data->op_mod_time = ktime_get_real_seconds();
2365 md_close(exp, op_data, NULL, &close_req);
2366 ptlrpc_req_finished(close_req);
2367 ll_finish_md_op_data(op_data);
2372 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2373 struct super_block *sb, struct lookup_intent *it)
2375 struct ll_sb_info *sbi = NULL;
2376 struct lustre_md md = { NULL };
2380 LASSERT(*inode || sb);
2381 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2382 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2383 sbi->ll_md_exp, &md);
2388 rc = ll_update_inode(*inode, &md);
2392 LASSERT(sb != NULL);
2395 * At this point server returns to client's same fid as client
2396 * generated for creating. So using ->fid1 is okay here.
2398 if (!fid_is_sane(&md.body->mbo_fid1)) {
2399 CERROR("%s: Fid is insane "DFID"\n",
2400 ll_get_fsname(sb, NULL, 0),
2401 PFID(&md.body->mbo_fid1));
2402 GOTO(out, rc = -EINVAL);
2405 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2406 sbi->ll_flags & LL_SBI_32BIT_API),
2408 if (IS_ERR(*inode)) {
2409 #ifdef CONFIG_FS_POSIX_ACL
2411 posix_acl_release(md.posix_acl);
2412 md.posix_acl = NULL;
2415 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2417 CERROR("new_inode -fatal: rc %d\n", rc);
2422 /* Handling piggyback layout lock.
2423 * Layout lock can be piggybacked by getattr and open request.
2424 * The lsm can be applied to inode only if it comes with a layout lock
2425 * otherwise correct layout may be overwritten, for example:
2426 * 1. proc1: mdt returns a lsm but not granting layout
2427 * 2. layout was changed by another client
2428 * 3. proc2: refresh layout and layout lock granted
2429 * 4. proc1: to apply a stale layout */
2430 if (it != NULL && it->it_lock_mode != 0) {
2431 struct lustre_handle lockh;
2432 struct ldlm_lock *lock;
2434 lockh.cookie = it->it_lock_handle;
2435 lock = ldlm_handle2lock(&lockh);
2436 LASSERT(lock != NULL);
2437 if (ldlm_has_layout(lock)) {
2438 struct cl_object_conf conf;
2440 memset(&conf, 0, sizeof(conf));
2441 conf.coc_opc = OBJECT_CONF_SET;
2442 conf.coc_inode = *inode;
2443 conf.coc_lock = lock;
2444 conf.u.coc_layout = md.layout;
2445 (void)ll_layout_conf(*inode, &conf);
2447 LDLM_LOCK_PUT(lock);
2453 md_free_lustre_md(sbi->ll_md_exp, &md);
2456 if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2457 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2462 int ll_obd_statfs(struct inode *inode, void __user *arg)
2464 struct ll_sb_info *sbi = NULL;
2465 struct obd_export *exp;
2467 struct obd_ioctl_data *data = NULL;
2471 if (!inode || !(sbi = ll_i2sbi(inode)))
2472 GOTO(out_statfs, rc = -EINVAL);
2474 rc = obd_ioctl_getdata(&buf, &len, arg);
2476 GOTO(out_statfs, rc);
2479 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2480 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2481 GOTO(out_statfs, rc = -EINVAL);
2483 if (data->ioc_inllen1 != sizeof(__u32) ||
2484 data->ioc_inllen2 != sizeof(__u32) ||
2485 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2486 data->ioc_plen2 != sizeof(struct obd_uuid))
2487 GOTO(out_statfs, rc = -EINVAL);
2489 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2490 if (type & LL_STATFS_LMV)
2491 exp = sbi->ll_md_exp;
2492 else if (type & LL_STATFS_LOV)
2493 exp = sbi->ll_dt_exp;
2495 GOTO(out_statfs, rc = -ENODEV);
2497 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2499 GOTO(out_statfs, rc);
2501 OBD_FREE_LARGE(buf, len);
2506 * this is normally called in ll_fini_md_op_data(), but sometimes it needs to
2507 * be called early to avoid deadlock.
2509 void ll_unlock_md_op_lsm(struct md_op_data *op_data)
2511 if (op_data->op_mea2_sem) {
2512 up_read(op_data->op_mea2_sem);
2513 op_data->op_mea2_sem = NULL;
2516 if (op_data->op_mea1_sem) {
2517 up_read(op_data->op_mea1_sem);
2518 op_data->op_mea1_sem = NULL;
2522 /* this function prepares md_op_data hint for passing it down to MD stack. */
2523 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2524 struct inode *i1, struct inode *i2,
2525 const char *name, size_t namelen,
2526 __u32 mode, __u32 opc, void *data)
2528 LASSERT(i1 != NULL);
2531 /* Do not reuse namelen for something else. */
2533 return ERR_PTR(-EINVAL);
2535 if (namelen > ll_i2sbi(i1)->ll_namelen)
2536 return ERR_PTR(-ENAMETOOLONG);
2538 if (!lu_name_is_valid_2(name, namelen))
2539 return ERR_PTR(-EINVAL);
2542 if (op_data == NULL)
2543 OBD_ALLOC_PTR(op_data);
2545 if (op_data == NULL)
2546 return ERR_PTR(-ENOMEM);
2548 ll_i2gids(op_data->op_suppgids, i1, i2);
2549 op_data->op_fid1 = *ll_inode2fid(i1);
2550 op_data->op_default_stripe_offset = -1;
2552 if (S_ISDIR(i1->i_mode)) {
2553 down_read(&ll_i2info(i1)->lli_lsm_sem);
2554 op_data->op_mea1_sem = &ll_i2info(i1)->lli_lsm_sem;
2555 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2556 if (opc == LUSTRE_OPC_MKDIR)
2557 op_data->op_default_stripe_offset =
2558 ll_i2info(i1)->lli_def_stripe_offset;
2562 op_data->op_fid2 = *ll_inode2fid(i2);
2563 if (S_ISDIR(i2->i_mode)) {
2565 down_read(&ll_i2info(i2)->lli_lsm_sem);
2566 op_data->op_mea2_sem =
2567 &ll_i2info(i2)->lli_lsm_sem;
2569 op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2572 fid_zero(&op_data->op_fid2);
2575 if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2576 op_data->op_cli_flags |= CLI_HASH64;
2578 if (ll_need_32bit_api(ll_i2sbi(i1)))
2579 op_data->op_cli_flags |= CLI_API32;
2581 op_data->op_name = name;
2582 op_data->op_namelen = namelen;
2583 op_data->op_mode = mode;
2584 op_data->op_mod_time = ktime_get_real_seconds();
2585 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2586 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2587 op_data->op_cap = cfs_curproc_cap_pack();
2588 op_data->op_mds = 0;
2589 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2590 filename_is_volatile(name, namelen, &op_data->op_mds)) {
2591 op_data->op_bias |= MDS_CREATE_VOLATILE;
2593 op_data->op_data = data;
2598 void ll_finish_md_op_data(struct md_op_data *op_data)
2600 ll_unlock_md_op_lsm(op_data);
2601 security_release_secctx(op_data->op_file_secctx,
2602 op_data->op_file_secctx_size);
2603 OBD_FREE_PTR(op_data);
2606 #ifdef HAVE_SUPEROPS_USE_DENTRY
2607 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2609 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2612 struct ll_sb_info *sbi;
2614 #ifdef HAVE_SUPEROPS_USE_DENTRY
2615 LASSERT((seq != NULL) && (dentry != NULL));
2616 sbi = ll_s2sbi(dentry->d_sb);
2618 LASSERT((seq != NULL) && (vfs != NULL));
2619 sbi = ll_s2sbi(vfs->mnt_sb);
2622 if (sbi->ll_flags & LL_SBI_NOLCK)
2623 seq_puts(seq, ",nolock");
2625 if (sbi->ll_flags & LL_SBI_FLOCK)
2626 seq_puts(seq, ",flock");
2628 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2629 seq_puts(seq, ",localflock");
2631 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2632 seq_puts(seq, ",user_xattr");
2634 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2635 seq_puts(seq, ",lazystatfs");
2637 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2638 seq_puts(seq, ",user_fid2path");
2640 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2641 seq_puts(seq, ",always_ping");
2647 * Get obd name by cmd, and copy out to user space
2649 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2651 struct ll_sb_info *sbi = ll_i2sbi(inode);
2652 struct obd_device *obd;
2655 if (cmd == OBD_IOC_GETDTNAME)
2656 obd = class_exp2obd(sbi->ll_dt_exp);
2657 else if (cmd == OBD_IOC_GETMDNAME)
2658 obd = class_exp2obd(sbi->ll_md_exp);
2665 if (copy_to_user((void __user *)arg, obd->obd_name,
2666 strlen(obd->obd_name) + 1))
2673 * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2674 * fsname will be returned in this buffer; otherwise, a static buffer will be
2675 * used to store the fsname and returned to caller.
2677 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2679 static char fsname_static[MTI_NAME_MAXLEN];
2680 struct lustre_sb_info *lsi = s2lsi(sb);
2685 /* this means the caller wants to use static buffer
2686 * and it doesn't care about race. Usually this is
2687 * in error reporting path */
2688 buf = fsname_static;
2689 buflen = sizeof(fsname_static);
2692 len = strlen(lsi->lsi_lmd->lmd_profile);
2693 ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2694 if (ptr && (strcmp(ptr, "-client") == 0))
2697 if (unlikely(len >= buflen))
2699 strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2705 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2712 p.mnt = current->fs->root.mnt;
2714 path = d_path(&p, buf, bufsize);
2719 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2721 char *buf, *path = NULL;
2722 struct dentry *dentry = NULL;
2723 struct inode *inode = page->mapping->host;
2725 /* this can be called inside spin lock so use GFP_ATOMIC. */
2726 buf = (char *)__get_free_page(GFP_ATOMIC);
2728 dentry = d_find_alias(page->mapping->host);
2730 path = ll_d_path(dentry, buf, PAGE_SIZE);
2734 "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2735 "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2736 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2737 PFID(ll_inode2fid(inode)),
2738 (path && !IS_ERR(path)) ? path : "", ioret);
2744 free_page((unsigned long)buf);
2747 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2748 struct lov_user_md **kbuf)
2750 struct lov_user_md lum;
2754 if (copy_from_user(&lum, md, sizeof(lum)))
2757 lum_size = ll_lov_user_md_size(&lum);
2761 OBD_ALLOC(*kbuf, lum_size);
2765 if (copy_from_user(*kbuf, md, lum_size) != 0) {
2766 OBD_FREE(*kbuf, lum_size);
2774 * Compute llite root squash state after a change of root squash
2775 * configuration setting or add/remove of a lnet nid
2777 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2779 struct root_squash_info *squash = &sbi->ll_squash;
2782 struct lnet_process_id id;
2784 /* Update norootsquash flag */
2785 down_write(&squash->rsi_sem);
2786 if (list_empty(&squash->rsi_nosquash_nids))
2787 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2789 /* Do not apply root squash as soon as one of our NIDs is
2790 * in the nosquash_nids list */
2793 while (LNetGetId(i++, &id) != -ENOENT) {
2794 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2796 if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2802 sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2804 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2806 up_write(&squash->rsi_sem);
2810 * Parse linkea content to extract information about a given hardlink
2812 * \param[in] ldata - Initialized linkea data
2813 * \param[in] linkno - Link identifier
2814 * \param[out] parent_fid - The entry's parent FID
2815 * \param[out] ln - Entry name destination buffer
2817 * \retval 0 on success
2818 * \retval Appropriate negative error code on failure
2820 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2821 struct lu_fid *parent_fid, struct lu_name *ln)
2827 rc = linkea_init_with_rec(ldata);
2831 if (linkno >= ldata->ld_leh->leh_reccount)
2832 /* beyond last link */
2835 linkea_first_entry(ldata);
2836 for (idx = 0; ldata->ld_lee != NULL; idx++) {
2837 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2842 linkea_next_entry(ldata);
2852 * Get parent FID and name of an identified link. Operation is performed for
2853 * a given link number, letting the caller iterate over linkno to list one or
2854 * all links of an entry.
2856 * \param[in] file - File descriptor against which to perform the operation
2857 * \param[in,out] arg - User-filled structure containing the linkno to operate
2858 * on and the available size. It is eventually filled with
2859 * the requested information or left untouched on error
2861 * \retval - 0 on success
2862 * \retval - Appropriate negative error code on failure
2864 int ll_getparent(struct file *file, struct getparent __user *arg)
2866 struct inode *inode = file_inode(file);
2867 struct linkea_data *ldata;
2868 struct lu_buf buf = LU_BUF_NULL;
2870 struct lu_fid parent_fid;
2877 if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2878 !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2881 if (get_user(name_size, &arg->gp_name_size))
2884 if (get_user(linkno, &arg->gp_linkno))
2887 if (name_size > PATH_MAX)
2890 OBD_ALLOC(ldata, sizeof(*ldata));
2894 rc = linkea_data_new(ldata, &buf);
2896 GOTO(ldata_free, rc);
2898 #ifdef HAVE_XATTR_HANDLER_FLAGS
2899 rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
2900 buf.lb_len, OBD_MD_FLXATTR);
2902 rc = ll_getxattr(file_dentry(file), XATTR_NAME_LINK, buf.lb_buf,
2904 #endif /* HAVE_XATTR_HANDLER_FLAGS */
2908 rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2912 if (ln.ln_namelen >= name_size)
2913 GOTO(lb_free, rc = -EOVERFLOW);
2915 if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2916 GOTO(lb_free, rc = -EFAULT);
2918 if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2919 GOTO(lb_free, rc = -EFAULT);
2921 if (put_user('\0', arg->gp_name + ln.ln_namelen))
2922 GOTO(lb_free, rc = -EFAULT);
2927 OBD_FREE(ldata, sizeof(*ldata));