4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lustre/llite/llite_lib.c
34 * Lustre Light Super operations
37 #define DEBUG_SUBSYSTEM S_LLITE
39 #include <linux/module.h>
40 #include <linux/random.h>
41 #include <linux/statfs.h>
42 #include <linux/time.h>
43 #include <linux/types.h>
44 #include <libcfs/linux/linux-uuid.h>
45 #include <linux/version.h>
47 #include <linux/user_namespace.h>
48 #ifdef HAVE_UIDGID_HEADER
49 # include <linux/uidgid.h>
51 #include <linux/security.h>
53 #include <uapi/linux/lustre/lustre_ioctl.h>
54 #include <lustre_ha.h>
55 #include <lustre_dlm.h>
56 #include <lprocfs_status.h>
57 #include <lustre_disk.h>
58 #include <uapi/linux/lustre/lustre_param.h>
59 #include <lustre_log.h>
60 #include <cl_object.h>
61 #include <obd_cksum.h>
62 #include "llite_internal.h"
64 struct kmem_cache *ll_file_data_slab;
67 #define log2(n) ffz(~(n))
70 static struct ll_sb_info *ll_init_sbi(void)
72 struct ll_sb_info *sbi = NULL;
74 unsigned long lru_page_max;
83 spin_lock_init(&sbi->ll_lock);
84 mutex_init(&sbi->ll_lco.lco_lock);
85 spin_lock_init(&sbi->ll_pp_extent_lock);
86 spin_lock_init(&sbi->ll_process_lock);
87 sbi->ll_rw_stats_on = 0;
90 pages = si.totalram - si.totalhigh;
91 lru_page_max = pages / 2;
93 /* initialize ll_cache data */
94 sbi->ll_cache = cl_cache_init(lru_page_max);
95 if (sbi->ll_cache == NULL) {
96 OBD_FREE(sbi, sizeof(*sbi));
100 sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
101 SBI_DEFAULT_READAHEAD_MAX);
102 sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
103 sbi->ll_ra_info.ra_max_read_ahead_whole_pages = -1;
105 sbi->ll_flags |= LL_SBI_VERBOSE;
106 #ifdef ENABLE_CHECKSUM
107 sbi->ll_flags |= LL_SBI_CHECKSUM;
110 sbi->ll_flags |= LL_SBI_FLOCK;
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);
142 /* Per-filesystem file heat */
143 sbi->ll_heat_decay_weight = SBI_DEFAULT_HEAT_DECAY_WEIGHT;
144 sbi->ll_heat_period_second = SBI_DEFAULT_HEAT_PERIOD_SECOND;
148 static void ll_free_sbi(struct super_block *sb)
150 struct ll_sb_info *sbi = ll_s2sbi(sb);
154 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
155 cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
156 if (sbi->ll_cache != NULL) {
157 cl_cache_decref(sbi->ll_cache);
158 sbi->ll_cache = NULL;
160 OBD_FREE(sbi, sizeof(*sbi));
165 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
166 struct vfsmount *mnt)
168 struct inode *root = NULL;
169 struct ll_sb_info *sbi = ll_s2sbi(sb);
170 struct obd_statfs *osfs = NULL;
171 struct ptlrpc_request *request = NULL;
172 struct obd_connect_data *data = NULL;
173 struct obd_uuid *uuid;
174 struct md_op_data *op_data;
175 struct lustre_md lmd;
177 int size, err, checksum;
180 sbi->ll_md_obd = class_name2obd(md);
181 if (!sbi->ll_md_obd) {
182 CERROR("MD %s: not setup or attached\n", md);
196 /* pass client page size via ocd_grant_blkbits, the server should report
197 * back its backend blocksize for grant calculation purpose */
198 data->ocd_grant_blkbits = PAGE_SHIFT;
200 /* indicate MDT features supported by this client */
201 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
202 OBD_CONNECT_ATTRFID | OBD_CONNECT_GRANT |
203 OBD_CONNECT_VERSION | OBD_CONNECT_BRW_SIZE |
204 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
205 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
206 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
207 OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 |
208 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
209 OBD_CONNECT_64BITHASH |
210 OBD_CONNECT_EINPROGRESS |
211 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
212 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS|
213 OBD_CONNECT_MAX_EASIZE |
214 OBD_CONNECT_FLOCK_DEAD |
215 OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
216 OBD_CONNECT_OPEN_BY_FID |
217 OBD_CONNECT_DIR_STRIPE |
218 OBD_CONNECT_BULK_MBITS | OBD_CONNECT_CKSUM |
219 OBD_CONNECT_SUBTREE |
220 OBD_CONNECT_MULTIMODRPCS |
221 OBD_CONNECT_GRANT_PARAM |
222 OBD_CONNECT_SHORTIO | OBD_CONNECT_FLAGS2;
224 data->ocd_connect_flags2 = OBD_CONNECT2_DIR_MIGRATE |
225 OBD_CONNECT2_SUM_STATFS |
227 OBD_CONNECT2_LOCK_CONVERT |
228 OBD_CONNECT2_ARCHIVE_ID_ARRAY |
231 #ifdef HAVE_LRU_RESIZE_SUPPORT
232 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
233 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
235 #ifdef CONFIG_FS_POSIX_ACL
236 data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK |
237 OBD_CONNECT_LARGE_ACL;
240 data->ocd_cksum_types = obd_cksum_types_supported_client();
242 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
243 /* flag mdc connection as lightweight, only used for test
244 * purpose, use with care */
245 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
247 data->ocd_ibits_known = MDS_INODELOCK_FULL;
248 data->ocd_version = LUSTRE_VERSION_CODE;
250 if (sb->s_flags & MS_RDONLY)
251 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
252 if (sbi->ll_flags & LL_SBI_USER_XATTR)
253 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
256 /* Setting this indicates we correctly support S_NOSEC (See kernel
257 * commit 9e1f1de02c2275d7172e18dc4e7c2065777611bf)
259 sb->s_flags |= MS_NOSEC;
262 if (sbi->ll_flags & LL_SBI_FLOCK)
263 sbi->ll_fop = &ll_file_operations_flock;
264 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
265 sbi->ll_fop = &ll_file_operations;
267 sbi->ll_fop = &ll_file_operations_noflock;
269 /* always ping even if server suppress_pings */
270 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
271 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
273 obd_connect_set_secctx(data);
275 #if defined(CONFIG_SECURITY)
276 data->ocd_connect_flags2 |= OBD_CONNECT2_SELINUX_POLICY;
279 data->ocd_brw_size = MD_MAX_BRW_SIZE;
281 err = obd_connect(NULL, &sbi->ll_md_exp, sbi->ll_md_obd,
282 &sbi->ll_sb_uuid, data, sbi->ll_cache);
284 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
285 "recovery, of which this client is not a "
286 "part. Please wait for recovery to complete,"
287 " abort, or time out.\n", md);
290 CERROR("cannot connect to %s: rc = %d\n", md, err);
294 sbi->ll_md_exp->exp_connect_data = *data;
296 err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
297 LUSTRE_SEQ_METADATA);
299 CERROR("%s: Can't init metadata layer FID infrastructure, "
300 "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
304 /* For mount, we only need fs info from MDT0, and also in DNE, it
305 * can make sure the client can be mounted as long as MDT0 is
307 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
308 ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
309 OBD_STATFS_FOR_MDT0);
311 GOTO(out_md_fid, err);
313 /* This needs to be after statfs to ensure connect has finished.
314 * Note that "data" does NOT contain the valid connect reply.
315 * If connecting to a 1.8 server there will be no LMV device, so
316 * we can access the MDC export directly and exp_connect_flags will
317 * be non-zero, but if accessing an upgraded 2.1 server it will
318 * have the correct flags filled in.
319 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
320 valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
321 if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
322 valid != CLIENT_CONNECT_MDT_REQD) {
325 OBD_ALLOC_WAIT(buf, PAGE_SIZE);
326 obd_connect_flags2str(buf, PAGE_SIZE,
327 valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
328 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
329 "feature(s) needed for correct operation "
330 "of this client (%s). Please upgrade "
331 "server or downgrade client.\n",
332 sbi->ll_md_exp->exp_obd->obd_name, buf);
333 OBD_FREE(buf, PAGE_SIZE);
334 GOTO(out_md_fid, err = -EPROTO);
337 size = sizeof(*data);
338 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
339 KEY_CONN_DATA, &size, data);
341 CERROR("%s: Get connect data failed: rc = %d\n",
342 sbi->ll_md_exp->exp_obd->obd_name, err);
343 GOTO(out_md_fid, err);
346 LASSERT(osfs->os_bsize);
347 sb->s_blocksize = osfs->os_bsize;
348 sb->s_blocksize_bits = log2(osfs->os_bsize);
349 sb->s_magic = LL_SUPER_MAGIC;
350 sb->s_maxbytes = MAX_LFS_FILESIZE;
351 sbi->ll_namelen = osfs->os_namelen;
352 sbi->ll_mnt.mnt = current->fs->root.mnt;
354 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
355 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
356 LCONSOLE_INFO("Disabling user_xattr feature because "
357 "it is not supported on the server\n");
358 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
361 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
363 sb->s_flags |= MS_POSIXACL;
365 sbi->ll_flags |= LL_SBI_ACL;
367 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
369 sb->s_flags &= ~MS_POSIXACL;
371 sbi->ll_flags &= ~LL_SBI_ACL;
374 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
375 sbi->ll_flags |= LL_SBI_64BIT_HASH;
377 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
378 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
380 if (obd_connect_has_secctx(data))
381 sbi->ll_flags |= LL_SBI_FILE_SECCTX;
383 if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
384 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
385 LCONSOLE_INFO("%s: disabling xattr cache due to "
386 "unknown maximum xattr size.\n", dt);
387 } else if (!sbi->ll_xattr_cache_set) {
388 /* If xattr_cache is already set (no matter 0 or 1)
389 * during processing llog, it won't be enabled here. */
390 sbi->ll_flags |= LL_SBI_XATTR_CACHE;
391 sbi->ll_xattr_cache_enabled = 1;
395 sbi->ll_dt_obd = class_name2obd(dt);
396 if (!sbi->ll_dt_obd) {
397 CERROR("DT %s: not setup or attached\n", dt);
398 GOTO(out_md_fid, err = -ENODEV);
401 /* pass client page size via ocd_grant_blkbits, the server should report
402 * back its backend blocksize for grant calculation purpose */
403 data->ocd_grant_blkbits = PAGE_SHIFT;
405 /* indicate OST features supported by this client */
406 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
407 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
408 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
409 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
410 OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
411 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
412 OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
413 OBD_CONNECT_EINPROGRESS |
414 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
415 OBD_CONNECT_LAYOUTLOCK |
416 OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
417 OBD_CONNECT_BULK_MBITS | OBD_CONNECT_SHORTIO |
418 OBD_CONNECT_FLAGS2 | OBD_CONNECT_GRANT_SHRINK;
420 /* The client currently advertises support for OBD_CONNECT_LOCKAHEAD_OLD so it
421 * can interoperate with an older version of lockahead which was released prior
422 * to landing in master. This support will be dropped when 2.13 development
423 * starts. At the point, we should not just drop the connect flag (below), we
424 * should also remove the support in the code.
426 * Removing it means a few things:
427 * 1. Remove this section here
428 * 2. Remove CEF_NONBLOCK in ll_file_lockahead()
429 * 3. Remove function exp_connect_lockahead_old
430 * 4. Remove LDLM_FL_LOCKAHEAD_OLD_RESERVED in lustre_dlm_flags.h
432 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 12, 50, 0)
433 data->ocd_connect_flags |= OBD_CONNECT_LOCKAHEAD_OLD;
436 data->ocd_connect_flags2 = OBD_CONNECT2_LOCKAHEAD;
438 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
439 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
441 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
442 * disabled by default, because it can still be enabled on the
443 * fly via /sys. As a consequence, we still need to come to an
444 * agreement on the supported algorithms at connect time
446 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
448 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
449 data->ocd_cksum_types = OBD_CKSUM_ADLER;
451 data->ocd_cksum_types = obd_cksum_types_supported_client();
453 #ifdef HAVE_LRU_RESIZE_SUPPORT
454 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
456 /* always ping even if server suppress_pings */
457 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
458 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
460 CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
461 "ocd_grant: %d\n", data->ocd_connect_flags,
462 data->ocd_version, data->ocd_grant);
464 sbi->ll_dt_obd->obd_upcall.onu_owner = &sbi->ll_lco;
465 sbi->ll_dt_obd->obd_upcall.onu_upcall = cl_ocd_update;
467 data->ocd_brw_size = DT_MAX_BRW_SIZE;
469 err = obd_connect(NULL, &sbi->ll_dt_exp, sbi->ll_dt_obd,
470 &sbi->ll_sb_uuid, data, sbi->ll_cache);
472 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
473 "recovery, of which this client is not a "
474 "part. Please wait for recovery to "
475 "complete, abort, or time out.\n", dt);
478 CERROR("%s: Cannot connect to %s: rc = %d\n",
479 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
483 sbi->ll_dt_exp->exp_connect_data = *data;
485 /* Don't change value if it was specified in the config log */
486 if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages == -1)
487 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
488 max_t(unsigned long, SBI_DEFAULT_READAHEAD_WHOLE_MAX,
489 (data->ocd_brw_size >> PAGE_SHIFT));
491 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
492 LUSTRE_SEQ_METADATA);
494 CERROR("%s: Can't init data layer FID infrastructure, "
495 "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
499 mutex_lock(&sbi->ll_lco.lco_lock);
500 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
501 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
502 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
503 mutex_unlock(&sbi->ll_lco.lco_lock);
505 fid_zero(&sbi->ll_root_fid);
506 err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
509 CERROR("cannot mds_connect: rc = %d\n", err);
510 GOTO(out_lock_cn_cb, err);
512 if (!fid_is_sane(&sbi->ll_root_fid)) {
513 CERROR("%s: Invalid root fid "DFID" during mount\n",
514 sbi->ll_md_exp->exp_obd->obd_name,
515 PFID(&sbi->ll_root_fid));
516 GOTO(out_lock_cn_cb, err = -EINVAL);
518 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
520 sb->s_op = &lustre_super_operations;
521 #ifdef HAVE_XATTR_HANDLER_FLAGS
522 sb->s_xattr = ll_xattr_handlers;
524 #if THREAD_SIZE >= 8192 /*b=17630*/
525 sb->s_export_op = &lustre_export_operations;
529 * XXX: move this to after cbd setup? */
530 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
531 if (sbi->ll_flags & LL_SBI_ACL)
532 valid |= OBD_MD_FLACL;
534 OBD_ALLOC_PTR(op_data);
536 GOTO(out_lock_cn_cb, err = -ENOMEM);
538 op_data->op_fid1 = sbi->ll_root_fid;
539 op_data->op_mode = 0;
540 op_data->op_valid = valid;
542 err = md_getattr(sbi->ll_md_exp, op_data, &request);
544 OBD_FREE_PTR(op_data);
546 CERROR("%s: md_getattr failed for root: rc = %d\n",
547 sbi->ll_md_exp->exp_obd->obd_name, err);
548 GOTO(out_lock_cn_cb, err);
551 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
552 sbi->ll_md_exp, &lmd);
554 CERROR("failed to understand root inode md: rc = %d\n", err);
555 ptlrpc_req_finished(request);
556 GOTO(out_lock_cn_cb, err);
559 LASSERT(fid_is_sane(&sbi->ll_root_fid));
560 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
561 sbi->ll_flags & LL_SBI_32BIT_API),
563 md_free_lustre_md(sbi->ll_md_exp, &lmd);
564 ptlrpc_req_finished(request);
567 #ifdef CONFIG_FS_POSIX_ACL
569 posix_acl_release(lmd.posix_acl);
570 lmd.posix_acl = NULL;
573 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
575 CERROR("lustre_lite: bad iget4 for root\n");
579 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
580 if (sbi->ll_checksum_set) {
581 err = obd_set_info_async(NULL, sbi->ll_dt_exp,
582 sizeof(KEY_CHECKSUM), KEY_CHECKSUM,
583 sizeof(checksum), &checksum, NULL);
585 CERROR("%s: Set checksum failed: rc = %d\n",
586 sbi->ll_dt_exp->exp_obd->obd_name, err);
592 sb->s_root = d_make_root(root);
593 if (sb->s_root == NULL) {
595 CERROR("%s: can't make root dentry: rc = %d\n",
596 sbi->ll_fsname, err);
599 #ifdef HAVE_DCACHE_LOCK
600 sb->s_root->d_op = &ll_d_ops;
603 sbi->ll_sdev_orig = sb->s_dev;
605 /* We set sb->s_dev equal on all lustre clients in order to support
606 * NFS export clustering. NFSD requires that the FSID be the same
608 /* s_dev is also used in lt_compare() to compare two fs, but that is
609 * only a node-local comparison. */
610 uuid = obd_get_uuid(sbi->ll_md_exp);
612 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
619 if (sbi->ll_dt_obd) {
620 err = sysfs_create_link(&sbi->ll_kset.kobj,
621 &sbi->ll_dt_obd->obd_kset.kobj,
622 sbi->ll_dt_obd->obd_type->typ_name);
624 CERROR("%s: could not register %s in llite: rc = %d\n",
625 dt, sbi->ll_fsname, err);
630 if (sbi->ll_md_obd) {
631 err = sysfs_create_link(&sbi->ll_kset.kobj,
632 &sbi->ll_md_obd->obd_kset.kobj,
633 sbi->ll_md_obd->obd_type->typ_name);
635 CERROR("%s: could not register %s in llite: rc = %d\n",
636 md, sbi->ll_fsname, err);
646 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
648 obd_disconnect(sbi->ll_dt_exp);
649 sbi->ll_dt_exp = NULL;
650 sbi->ll_dt_obd = NULL;
652 obd_fid_fini(sbi->ll_md_exp->exp_obd);
654 obd_disconnect(sbi->ll_md_exp);
655 sbi->ll_md_exp = NULL;
656 sbi->ll_md_obd = NULL;
665 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
669 size = sizeof(*lmmsize);
670 rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
671 KEY_MAX_EASIZE, &size, lmmsize);
673 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
674 sbi->ll_dt_exp->exp_obd->obd_name, rc);
678 CDEBUG(D_INFO, "max LOV ea size: %d\n", *lmmsize);
681 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
682 KEY_MAX_EASIZE, &size, lmmsize);
684 CERROR("Get max mdsize error rc %d\n", rc);
686 CDEBUG(D_INFO, "max LMV ea size: %d\n", *lmmsize);
692 * Get the value of the default_easize parameter.
694 * \see client_obd::cl_default_mds_easize
696 * \param[in] sbi superblock info for this filesystem
697 * \param[out] lmmsize pointer to storage location for value
699 * \retval 0 on success
700 * \retval negative negated errno on failure
702 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
707 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
708 KEY_DEFAULT_EASIZE, &size, lmmsize);
710 CERROR("Get default mdsize error rc %d\n", rc);
716 * Set the default_easize parameter to the given value.
718 * \see client_obd::cl_default_mds_easize
720 * \param[in] sbi superblock info for this filesystem
721 * \param[in] lmmsize the size to set
723 * \retval 0 on success
724 * \retval negative negated errno on failure
726 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
730 if (lmmsize < sizeof(struct lov_mds_md) ||
731 lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
734 rc = obd_set_info_async(NULL, sbi->ll_md_exp,
735 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
736 sizeof(int), &lmmsize, NULL);
741 static void client_common_put_super(struct super_block *sb)
743 struct ll_sb_info *sbi = ll_s2sbi(sb);
748 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
749 obd_disconnect(sbi->ll_dt_exp);
750 sbi->ll_dt_exp = NULL;
752 ll_debugfs_unregister_super(sb);
754 obd_fid_fini(sbi->ll_md_exp->exp_obd);
755 obd_disconnect(sbi->ll_md_exp);
756 sbi->ll_md_exp = NULL;
761 void ll_kill_super(struct super_block *sb)
763 struct ll_sb_info *sbi;
767 if (!(sb->s_flags & MS_ACTIVE))
771 /* we need restore s_dev from changed for clustred NFS before put_super
772 * because new kernels have cached s_dev and change sb->s_dev in
773 * put_super not affected real removing devices */
775 sb->s_dev = sbi->ll_sdev_orig;
776 sbi->ll_umounting = 1;
778 /* wait running statahead threads to quit */
779 while (atomic_read(&sbi->ll_sa_running) > 0) {
780 set_current_state(TASK_UNINTERRUPTIBLE);
781 schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
788 static inline int ll_set_opt(const char *opt, char *data, int fl)
790 if (strncmp(opt, data, strlen(opt)) != 0)
796 /* non-client-specific mount options are parsed in lmd_parse */
797 static int ll_options(char *options, struct ll_sb_info *sbi)
800 char *s1 = options, *s2;
801 int *flags = &sbi->ll_flags;
807 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
810 CDEBUG(D_SUPER, "next opt=%s\n", s1);
811 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
816 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
821 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
826 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
831 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
836 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
841 tmp = ll_set_opt("context", s1, 1);
844 tmp = ll_set_opt("fscontext", s1, 1);
847 tmp = ll_set_opt("defcontext", s1, 1);
850 tmp = ll_set_opt("rootcontext", s1, 1);
853 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
858 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
864 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
867 sbi->ll_checksum_set = 1;
870 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
873 sbi->ll_checksum_set = 1;
876 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
881 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
886 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
891 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
896 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
901 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
906 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
911 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
916 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
922 s2 = strchr(s1, ',');
930 void ll_lli_init(struct ll_inode_info *lli)
932 lli->lli_inode_magic = LLI_INODE_MAGIC;
934 spin_lock_init(&lli->lli_lock);
935 lli->lli_posix_acl = NULL;
936 /* Do not set lli_fid, it has been initialized already. */
937 fid_zero(&lli->lli_pfid);
938 lli->lli_mds_read_och = NULL;
939 lli->lli_mds_write_och = NULL;
940 lli->lli_mds_exec_och = NULL;
941 lli->lli_open_fd_read_count = 0;
942 lli->lli_open_fd_write_count = 0;
943 lli->lli_open_fd_exec_count = 0;
944 mutex_init(&lli->lli_och_mutex);
945 spin_lock_init(&lli->lli_agl_lock);
946 spin_lock_init(&lli->lli_layout_lock);
947 ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
948 lli->lli_clob = NULL;
950 init_rwsem(&lli->lli_xattrs_list_rwsem);
951 mutex_init(&lli->lli_xattrs_enq_lock);
953 LASSERT(lli->lli_vfs_inode.i_mode != 0);
954 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
955 mutex_init(&lli->lli_readdir_mutex);
956 lli->lli_opendir_key = NULL;
958 spin_lock_init(&lli->lli_sa_lock);
959 lli->lli_opendir_pid = 0;
960 lli->lli_sa_enabled = 0;
961 lli->lli_def_stripe_offset = -1;
962 init_rwsem(&lli->lli_lsm_sem);
964 mutex_init(&lli->lli_size_mutex);
965 lli->lli_symlink_name = NULL;
966 init_rwsem(&lli->lli_trunc_sem);
967 range_lock_tree_init(&lli->lli_write_tree);
968 init_rwsem(&lli->lli_glimpse_sem);
969 lli->lli_glimpse_time = ktime_set(0, 0);
970 INIT_LIST_HEAD(&lli->lli_agl_list);
971 lli->lli_agl_index = 0;
972 lli->lli_async_rc = 0;
973 spin_lock_init(&lli->lli_heat_lock);
974 obd_heat_clear(lli->lli_heat_instances, OBD_HEAT_COUNT);
975 lli->lli_heat_flags = 0;
977 mutex_init(&lli->lli_layout_mutex);
978 memset(lli->lli_jobid, 0, sizeof(lli->lli_jobid));
981 #define MAX_STRING_SIZE 128
983 #ifndef HAVE_SUPER_SETUP_BDI_NAME
985 #define LSI_BDI_INITIALIZED 0x00400000
987 #ifndef HAVE_BDI_CAP_MAP_COPY
988 # define BDI_CAP_MAP_COPY 0
991 static int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
993 struct lustre_sb_info *lsi = s2lsi(sb);
994 char buf[MAX_STRING_SIZE];
998 err = bdi_init(&lsi->lsi_bdi);
1002 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1003 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1004 lsi->lsi_bdi.name = "lustre";
1005 va_start(args, fmt);
1006 vsnprintf(buf, MAX_STRING_SIZE, fmt, args);
1008 err = bdi_register(&lsi->lsi_bdi, NULL, "%s", buf);
1011 sb->s_bdi = &lsi->lsi_bdi;
1015 #endif /* !HAVE_SUPER_SETUP_BDI_NAME */
1017 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
1019 struct lustre_profile *lprof = NULL;
1020 struct lustre_sb_info *lsi = s2lsi(sb);
1021 struct ll_sb_info *sbi = NULL;
1022 char *dt = NULL, *md = NULL;
1023 char *profilenm = get_profile_name(sb);
1024 struct config_llog_instance *cfg;
1025 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1026 const int instlen = 16 + 2;
1027 unsigned long cfg_instance = ll_get_cfg_instance(sb);
1028 char name[MAX_STRING_SIZE];
1037 /* for ASLR, to map between cfg_instance and hashed ptr */
1038 CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1039 profilenm, cfg_instance, sb);
1041 try_module_get(THIS_MODULE);
1045 GOTO(out_free_cfg, err = -ENOMEM);
1047 /* client additional sb info */
1048 lsi->lsi_llsbi = sbi = ll_init_sbi();
1050 GOTO(out_free_cfg, err = -ENOMEM);
1052 err = ll_options(lsi->lsi_lmd->lmd_opts, sbi);
1054 GOTO(out_free_cfg, err);
1056 #ifndef HAVE_DCACHE_LOCK
1057 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1058 sb->s_d_op = &ll_d_ops;
1061 generate_random_uuid(uuid.b);
1062 snprintf(sbi->ll_sb_uuid.uuid, UUID_SIZE, "%pU", uuid.b);
1064 CDEBUG(D_CONFIG, "llite sb uuid: %s\n", sbi->ll_sb_uuid.uuid);
1067 len = strlen(profilenm);
1068 ptr = strrchr(profilenm, '-');
1069 if (ptr && (strcmp(ptr, "-client") == 0))
1072 if (len > LUSTRE_MAXFSNAME) {
1073 if (unlikely(len >= MAX_STRING_SIZE))
1074 len = MAX_STRING_SIZE - 1;
1075 strncpy(name, profilenm, len);
1077 err = -ENAMETOOLONG;
1078 CERROR("%s: fsname longer than %u characters: rc = %d\n",
1079 name, LUSTRE_MAXFSNAME, err);
1080 GOTO(out_free_cfg, err);
1082 strncpy(sbi->ll_fsname, profilenm, len);
1083 sbi->ll_fsname[len] = '\0';
1086 snprintf(name, sizeof(name), "%.*s-%016lx", len,
1087 profilenm, cfg_instance);
1089 err = super_setup_bdi_name(sb, "%s", name);
1091 GOTO(out_free_cfg, err);
1093 /* Call ll_debugfs_register_super() before lustre_process_log()
1094 * so that "llite.*.*" params can be processed correctly.
1096 err = ll_debugfs_register_super(sb, name);
1098 CERROR("%s: could not register mountpoint in llite: rc = %d\n",
1099 sbi->ll_fsname, err);
1103 /* The cfg_instance is a value unique to this super, in case some
1104 * joker tries to mount the same fs at two mount points.
1106 cfg->cfg_instance = cfg_instance;
1107 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1108 cfg->cfg_callback = class_config_llog_handler;
1109 cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1110 /* set up client obds */
1111 err = lustre_process_log(sb, profilenm, cfg);
1113 GOTO(out_debugfs, err);
1115 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1116 lprof = class_get_profile(profilenm);
1117 if (lprof == NULL) {
1118 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1119 " read from the MGS. Does that filesystem "
1120 "exist?\n", profilenm);
1121 GOTO(out_debugfs, err = -EINVAL);
1123 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1124 lprof->lp_md, lprof->lp_dt);
1126 dt_len = strlen(lprof->lp_dt) + instlen + 2;
1127 OBD_ALLOC(dt, dt_len);
1129 GOTO(out_profile, err = -ENOMEM);
1130 snprintf(dt, dt_len - 1, "%s-%016lx", lprof->lp_dt, cfg_instance);
1132 md_len = strlen(lprof->lp_md) + instlen + 2;
1133 OBD_ALLOC(md, md_len);
1135 GOTO(out_free_dt, err = -ENOMEM);
1136 snprintf(md, md_len - 1, "%s-%016lx", lprof->lp_md, cfg_instance);
1138 /* connections, registrations, sb setup */
1139 err = client_common_fill_super(sb, md, dt, mnt);
1141 GOTO(out_free_md, err);
1143 sbi->ll_client_common_fill_super_succeeded = 1;
1147 OBD_FREE(md, md_len);
1150 OBD_FREE(dt, dt_len);
1153 class_put_profile(lprof);
1156 ll_debugfs_unregister_super(sb);
1163 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1164 LCONSOLE_WARN("Mounted %s\n", profilenm);
1166 } /* ll_fill_super */
1168 void ll_put_super(struct super_block *sb)
1170 struct config_llog_instance cfg, params_cfg;
1171 struct obd_device *obd;
1172 struct lustre_sb_info *lsi = s2lsi(sb);
1173 struct ll_sb_info *sbi = ll_s2sbi(sb);
1174 char *profilenm = get_profile_name(sb);
1175 unsigned long cfg_instance = ll_get_cfg_instance(sb);
1177 int next, force = 1, rc = 0;
1181 GOTO(out_no_sbi, 0);
1183 /* Should replace instance_id with something better for ASLR */
1184 CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1185 profilenm, cfg_instance, sb);
1187 cfg.cfg_instance = cfg_instance;
1188 lustre_end_log(sb, profilenm, &cfg);
1190 params_cfg.cfg_instance = cfg_instance;
1191 lustre_end_log(sb, PARAMS_FILENAME, ¶ms_cfg);
1193 if (sbi->ll_md_exp) {
1194 obd = class_exp2obd(sbi->ll_md_exp);
1196 force = obd->obd_force;
1199 /* Wait for unstable pages to be committed to stable storage */
1201 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1202 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1203 atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1207 ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1208 if (force == 0 && rc != -EINTR)
1209 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1211 /* We need to set force before the lov_disconnect in
1212 lustre_common_put_super, since l_d cleans up osc's as well. */
1215 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1217 obd->obd_force = force;
1221 if (sbi->ll_client_common_fill_super_succeeded) {
1222 /* Only if client_common_fill_super succeeded */
1223 client_common_put_super(sb);
1227 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1228 class_manual_cleanup(obd);
1231 if (sbi->ll_flags & LL_SBI_VERBOSE)
1232 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1235 class_del_profile(profilenm);
1237 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1238 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1239 bdi_destroy(&lsi->lsi_bdi);
1240 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1245 lsi->lsi_llsbi = NULL;
1247 lustre_common_put_super(sb);
1249 cl_env_cache_purge(~0);
1251 module_put(THIS_MODULE);
1254 } /* client_put_super */
1256 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1258 struct inode *inode = NULL;
1260 /* NOTE: we depend on atomic igrab() -bzzz */
1261 lock_res_and_lock(lock);
1262 if (lock->l_resource->lr_lvb_inode) {
1263 struct ll_inode_info * lli;
1264 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1265 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1266 inode = igrab(lock->l_resource->lr_lvb_inode);
1268 inode = lock->l_resource->lr_lvb_inode;
1269 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1270 D_WARNING, lock, "lr_lvb_inode %p is "
1271 "bogus: magic %08x",
1272 lock->l_resource->lr_lvb_inode,
1273 lli->lli_inode_magic);
1277 unlock_res_and_lock(lock);
1281 void ll_dir_clear_lsm_md(struct inode *inode)
1283 struct ll_inode_info *lli = ll_i2info(inode);
1285 LASSERT(S_ISDIR(inode->i_mode));
1287 if (lli->lli_lsm_md != NULL) {
1288 lmv_free_memmd(lli->lli_lsm_md);
1289 lli->lli_lsm_md = NULL;
1293 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1294 const struct lu_fid *fid,
1295 struct lustre_md *md)
1297 struct ll_sb_info *sbi = ll_s2sbi(sb);
1298 struct mdt_body *body = md->body;
1299 struct inode *inode;
1303 ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1304 inode = iget_locked(sb, ino);
1305 if (inode == NULL) {
1306 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1307 sbi->ll_fsname, PFID(fid));
1308 RETURN(ERR_PTR(-ENOENT));
1311 if (inode->i_state & I_NEW) {
1312 struct ll_inode_info *lli = ll_i2info(inode);
1313 struct lmv_stripe_md *lsm = md->lmv;
1315 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1316 (body->mbo_mode & S_IFMT);
1317 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1320 inode->i_mtime.tv_sec = 0;
1321 inode->i_atime.tv_sec = 0;
1322 inode->i_ctime.tv_sec = 0;
1325 #ifdef HAVE_BACKING_DEV_INFO
1326 /* initializing backing dev info. */
1327 inode->i_mapping->backing_dev_info =
1328 &s2lsi(inode->i_sb)->lsi_bdi;
1330 inode->i_op = &ll_dir_inode_operations;
1331 inode->i_fop = &ll_dir_operations;
1332 lli->lli_fid = *fid;
1335 LASSERT(lsm != NULL);
1336 /* master object FID */
1337 lli->lli_pfid = body->mbo_fid1;
1338 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1339 lli, PFID(fid), PFID(&lli->lli_pfid));
1340 unlock_new_inode(inode);
1346 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1349 struct lmv_stripe_md *lsm = md->lmv;
1350 struct ll_inode_info *lli = ll_i2info(inode);
1353 LASSERT(lsm != NULL);
1355 CDEBUG(D_INODE, "%s: "DFID" set dir layout:\n",
1356 ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid));
1357 lsm_md_dump(D_INODE, lsm);
1359 /* XXX sigh, this lsm_root initialization should be in
1360 * LMV layer, but it needs ll_iget right now, so we
1361 * put this here right now. */
1362 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1363 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1364 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1366 if (!fid_is_sane(fid))
1369 /* Unfortunately ll_iget will call ll_update_inode,
1370 * where the initialization of slave inode is slightly
1371 * different, so it reset lsm_md to NULL to avoid
1372 * initializing lsm for slave inode. */
1373 lsm->lsm_md_oinfo[i].lmo_root =
1374 ll_iget_anon_dir(inode->i_sb, fid, md);
1375 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1376 int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1378 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1380 iput(lsm->lsm_md_oinfo[i].lmo_root);
1381 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1387 lli->lli_lsm_md = lsm;
1392 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1394 struct ll_inode_info *lli = ll_i2info(inode);
1395 struct lmv_stripe_md *lsm = md->lmv;
1400 LASSERT(S_ISDIR(inode->i_mode));
1401 CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1402 PFID(ll_inode2fid(inode)));
1405 * no striped information from request, lustre_md from req does not
1406 * include stripeEA, see ll_md_setattr()
1412 * normally dir layout doesn't change, only take read lock to check
1413 * that to avoid blocking other MD operations.
1415 if (lli->lli_lsm_md)
1416 down_read(&lli->lli_lsm_sem);
1418 down_write(&lli->lli_lsm_sem);
1421 * if dir layout mismatch, check whether version is increased, which
1422 * means layout is changed, this happens in dir migration and lfsck.
1424 * foreign LMV should not change.
1426 if (lli->lli_lsm_md &&
1427 lli->lli_lsm_md->lsm_md_magic != LMV_MAGIC_FOREIGN &&
1428 !lsm_md_eq(lli->lli_lsm_md, lsm)) {
1429 if (lsm->lsm_md_layout_version <=
1430 lli->lli_lsm_md->lsm_md_layout_version) {
1431 CERROR("%s: "DFID" dir layout mismatch:\n",
1432 ll_i2sbi(inode)->ll_fsname,
1433 PFID(&lli->lli_fid));
1434 lsm_md_dump(D_ERROR, lli->lli_lsm_md);
1435 lsm_md_dump(D_ERROR, lsm);
1436 GOTO(unlock, rc = -EINVAL);
1439 /* layout changed, switch to write lock */
1440 up_read(&lli->lli_lsm_sem);
1441 down_write(&lli->lli_lsm_sem);
1442 ll_dir_clear_lsm_md(inode);
1445 /* set directory layout */
1446 if (!lli->lli_lsm_md) {
1447 struct cl_attr *attr;
1449 if (lsm->lsm_md_magic == LMV_MAGIC_FOREIGN) {
1450 /* set md->lmv to NULL, so the following free lustre_md
1451 * will not free this lsm */
1453 lli->lli_lsm_md = lsm;
1454 up_write(&lli->lli_lsm_sem);
1458 rc = ll_init_lsm_md(inode, md);
1459 up_write(&lli->lli_lsm_sem);
1463 /* set md->lmv to NULL, so the following free lustre_md
1464 * will not free this lsm */
1468 * md_merge_attr() may take long, since lsm is already set,
1469 * switch to read lock.
1471 down_read(&lli->lli_lsm_sem);
1473 OBD_ALLOC_PTR(attr);
1475 GOTO(unlock, rc = -ENOMEM);
1477 /* validate the lsm */
1478 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1479 ll_md_blocking_ast);
1485 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1486 md->body->mbo_nlink = attr->cat_nlink;
1487 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1488 md->body->mbo_size = attr->cat_size;
1489 if (md->body->mbo_valid & OBD_MD_FLATIME)
1490 md->body->mbo_atime = attr->cat_atime;
1491 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1492 md->body->mbo_ctime = attr->cat_ctime;
1493 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1494 md->body->mbo_mtime = attr->cat_mtime;
1499 up_read(&lli->lli_lsm_sem);
1504 void ll_clear_inode(struct inode *inode)
1506 struct ll_inode_info *lli = ll_i2info(inode);
1507 struct ll_sb_info *sbi = ll_i2sbi(inode);
1510 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1511 PFID(ll_inode2fid(inode)), inode);
1513 if (S_ISDIR(inode->i_mode)) {
1514 /* these should have been cleared in ll_file_release */
1515 LASSERT(lli->lli_opendir_key == NULL);
1516 LASSERT(lli->lli_sai == NULL);
1517 LASSERT(lli->lli_opendir_pid == 0);
1520 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1522 LASSERT(!lli->lli_open_fd_write_count);
1523 LASSERT(!lli->lli_open_fd_read_count);
1524 LASSERT(!lli->lli_open_fd_exec_count);
1526 if (lli->lli_mds_write_och)
1527 ll_md_real_close(inode, FMODE_WRITE);
1528 if (lli->lli_mds_exec_och)
1529 ll_md_real_close(inode, FMODE_EXEC);
1530 if (lli->lli_mds_read_och)
1531 ll_md_real_close(inode, FMODE_READ);
1533 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1534 OBD_FREE(lli->lli_symlink_name,
1535 strlen(lli->lli_symlink_name) + 1);
1536 lli->lli_symlink_name = NULL;
1539 ll_xattr_cache_destroy(inode);
1541 #ifdef CONFIG_FS_POSIX_ACL
1542 forget_all_cached_acls(inode);
1543 if (lli->lli_posix_acl) {
1544 posix_acl_release(lli->lli_posix_acl);
1545 lli->lli_posix_acl = NULL;
1548 lli->lli_inode_magic = LLI_INODE_DEAD;
1550 if (S_ISDIR(inode->i_mode))
1551 ll_dir_clear_lsm_md(inode);
1552 else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1553 LASSERT(list_empty(&lli->lli_agl_list));
1556 * XXX This has to be done before lsm is freed below, because
1557 * cl_object still uses inode lsm.
1559 cl_inode_fini(inode);
1564 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1566 struct lustre_md md;
1567 struct inode *inode = dentry->d_inode;
1568 struct ll_sb_info *sbi = ll_i2sbi(inode);
1569 struct ptlrpc_request *request = NULL;
1573 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1574 LUSTRE_OPC_ANY, NULL);
1575 if (IS_ERR(op_data))
1576 RETURN(PTR_ERR(op_data));
1578 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1580 ptlrpc_req_finished(request);
1581 if (rc == -ENOENT) {
1583 /* Unlinked special device node? Or just a race?
1584 * Pretend we done everything. */
1585 if (!S_ISREG(inode->i_mode) &&
1586 !S_ISDIR(inode->i_mode)) {
1587 ia_valid = op_data->op_attr.ia_valid;
1588 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1589 rc = simple_setattr(dentry, &op_data->op_attr);
1590 op_data->op_attr.ia_valid = ia_valid;
1592 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1593 CERROR("md_setattr fails: rc = %d\n", rc);
1598 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1599 sbi->ll_md_exp, &md);
1601 ptlrpc_req_finished(request);
1605 ia_valid = op_data->op_attr.ia_valid;
1606 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1607 * cache is not cleared yet. */
1608 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1609 if (S_ISREG(inode->i_mode))
1611 rc = simple_setattr(dentry, &op_data->op_attr);
1612 if (S_ISREG(inode->i_mode))
1613 inode_unlock(inode);
1614 op_data->op_attr.ia_valid = ia_valid;
1616 rc = ll_update_inode(inode, &md);
1617 ptlrpc_req_finished(request);
1622 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1623 * object(s) determine the file size and mtime. Otherwise, the MDS will
1624 * keep these values until such a time that objects are allocated for it.
1625 * We do the MDS operations first, as it is checking permissions for us.
1626 * We don't to the MDS RPC if there is nothing that we want to store there,
1627 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1628 * going to do an RPC anyways.
1630 * If we are doing a truncate, we will send the mtime and ctime updates
1631 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1632 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1635 * In case of HSMimport, we only set attr on MDS.
1637 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr,
1638 enum op_xvalid xvalid, bool hsm_import)
1640 struct inode *inode = dentry->d_inode;
1641 struct ll_inode_info *lli = ll_i2info(inode);
1642 struct md_op_data *op_data = NULL;
1646 CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1647 "valid %x, hsm_import %d\n",
1648 ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid),
1649 inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1652 if (attr->ia_valid & ATTR_SIZE) {
1653 /* Check new size against VFS/VM file size limit and rlimit */
1654 rc = inode_newsize_ok(inode, attr->ia_size);
1658 /* The maximum Lustre file size is variable, based on the
1659 * OST maximum object size and number of stripes. This
1660 * needs another check in addition to the VFS check above. */
1661 if (attr->ia_size > ll_file_maxbytes(inode)) {
1662 CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
1663 PFID(&lli->lli_fid), attr->ia_size,
1664 ll_file_maxbytes(inode));
1668 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1671 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1672 if (attr->ia_valid & TIMES_SET_FLAGS) {
1673 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1674 !cfs_capable(CFS_CAP_FOWNER))
1678 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1679 if (!(xvalid & OP_XVALID_CTIME_SET) &&
1680 (attr->ia_valid & ATTR_CTIME)) {
1681 attr->ia_ctime = current_time(inode);
1682 xvalid |= OP_XVALID_CTIME_SET;
1684 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1685 (attr->ia_valid & ATTR_ATIME)) {
1686 attr->ia_atime = current_time(inode);
1687 attr->ia_valid |= ATTR_ATIME_SET;
1689 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1690 (attr->ia_valid & ATTR_MTIME)) {
1691 attr->ia_mtime = current_time(inode);
1692 attr->ia_valid |= ATTR_MTIME_SET;
1695 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1696 CDEBUG(D_INODE, "setting mtime %lld, ctime %lld, now = %lld\n",
1697 (s64)attr->ia_mtime.tv_sec, (s64)attr->ia_ctime.tv_sec,
1698 ktime_get_real_seconds());
1700 if (S_ISREG(inode->i_mode)) {
1701 if (attr->ia_valid & ATTR_SIZE)
1702 inode_dio_write_done(inode);
1703 inode_unlock(inode);
1706 /* We always do an MDS RPC, even if we're only changing the size;
1707 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1709 OBD_ALLOC_PTR(op_data);
1710 if (op_data == NULL)
1711 GOTO(out, rc = -ENOMEM);
1713 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1714 /* If we are changing file size, file content is
1715 * modified, flag it.
1717 xvalid |= OP_XVALID_OWNEROVERRIDE;
1718 op_data->op_bias |= MDS_DATA_MODIFIED;
1719 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1722 if (attr->ia_valid & ATTR_FILE) {
1723 struct ll_file_data *fd = LUSTRE_FPRIVATE(attr->ia_file);
1725 if (fd->fd_lease_och)
1726 op_data->op_bias |= MDS_TRUNC_KEEP_LEASE;
1729 op_data->op_attr = *attr;
1730 op_data->op_xvalid = xvalid;
1732 rc = ll_md_setattr(dentry, op_data);
1736 if (!S_ISREG(inode->i_mode) || hsm_import)
1739 if (attr->ia_valid & (ATTR_SIZE | ATTR_ATIME | ATTR_ATIME_SET |
1740 ATTR_MTIME | ATTR_MTIME_SET | ATTR_CTIME) ||
1741 xvalid & OP_XVALID_CTIME_SET) {
1742 /* For truncate and utimes sending attributes to OSTs, setting
1743 * mtime/atime to the past will be performed under PW [0:EOF]
1744 * extent lock (new_size:EOF for truncate). It may seem
1745 * excessive to send mtime/atime updates to OSTs when not
1746 * setting times to past, but it is necessary due to possible
1747 * time de-synchronization between MDT inode and OST objects
1749 rc = cl_setattr_ost(lli->lli_clob, attr, xvalid, 0);
1752 /* If the file was restored, it needs to set dirty flag.
1754 * We've already sent MDS_DATA_MODIFIED flag in
1755 * ll_md_setattr() for truncate. However, the MDT refuses to
1756 * set the HS_DIRTY flag on released files, so we have to set
1757 * it again if the file has been restored. Please check how
1758 * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1760 * Please notice that if the file is not released, the previous
1761 * MDS_DATA_MODIFIED has taken effect and usually
1762 * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1763 * This way we can save an RPC for common open + trunc
1765 if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1766 struct hsm_state_set hss = {
1767 .hss_valid = HSS_SETMASK,
1768 .hss_setmask = HS_DIRTY,
1772 rc2 = ll_hsm_state_set(inode, &hss);
1773 /* truncate and write can happen at the same time, so that
1774 * the file can be set modified even though the file is not
1775 * restored from released state, and ll_hsm_state_set() is
1776 * not applicable for the file, and rc2 < 0 is normal in this
1779 CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1780 PFID(ll_inode2fid(inode)), rc2);
1785 if (op_data != NULL)
1786 ll_finish_md_op_data(op_data);
1788 if (S_ISREG(inode->i_mode)) {
1790 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1791 inode_dio_wait(inode);
1792 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
1793 * flag. ll_update_inode (called from ll_md_setattr), clears
1794 * inode flags, so there is a gap where S_NOSEC is not set.
1795 * This can cause a writer to take the i_mutex unnecessarily,
1796 * but this is safe to do and should be rare. */
1797 inode_has_no_xattr(inode);
1800 ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1801 LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1806 int ll_setattr(struct dentry *de, struct iattr *attr)
1808 int mode = de->d_inode->i_mode;
1809 enum op_xvalid xvalid = 0;
1811 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1812 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1813 xvalid |= OP_XVALID_OWNEROVERRIDE;
1815 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1816 (ATTR_SIZE|ATTR_MODE)) &&
1817 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1818 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1819 !(attr->ia_mode & S_ISGID))))
1820 attr->ia_valid |= ATTR_FORCE;
1822 if ((attr->ia_valid & ATTR_MODE) &&
1824 !(attr->ia_mode & S_ISUID) &&
1825 !(attr->ia_valid & ATTR_KILL_SUID))
1826 attr->ia_valid |= ATTR_KILL_SUID;
1828 if ((attr->ia_valid & ATTR_MODE) &&
1829 ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1830 !(attr->ia_mode & S_ISGID) &&
1831 !(attr->ia_valid & ATTR_KILL_SGID))
1832 attr->ia_valid |= ATTR_KILL_SGID;
1834 return ll_setattr_raw(de, attr, xvalid, false);
1837 int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
1840 struct obd_statfs obd_osfs = { 0 };
1845 max_age = ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS;
1847 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1851 osfs->os_type = LL_SUPER_MAGIC;
1853 CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1854 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree, osfs->os_files);
1856 if (osfs->os_state & OS_STATE_SUM)
1859 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1860 flags |= OBD_STATFS_NODELAY;
1862 rc = obd_statfs(NULL, sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1863 if (rc) /* Possibly a filesystem with no OSTs. Report MDT totals. */
1866 CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1867 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1870 osfs->os_bsize = obd_osfs.os_bsize;
1871 osfs->os_blocks = obd_osfs.os_blocks;
1872 osfs->os_bfree = obd_osfs.os_bfree;
1873 osfs->os_bavail = obd_osfs.os_bavail;
1875 /* If we have _some_ OSTs, but don't have as many free objects on the
1876 * OSTs as inodes on the MDTs, reduce the reported number of inodes
1877 * to compensate, so that the "inodes in use" number is correct.
1878 * This should be kept in sync with lod_statfs() behaviour.
1880 if (obd_osfs.os_files && obd_osfs.os_ffree < osfs->os_ffree) {
1881 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1883 osfs->os_ffree = obd_osfs.os_ffree;
1889 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1891 struct super_block *sb = de->d_sb;
1892 struct obd_statfs osfs;
1893 __u64 fsid = huge_encode_dev(sb->s_dev);
1896 CDEBUG(D_VFSTRACE, "VFS Op: at %llu jiffies\n", get_jiffies_64());
1897 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1899 /* Some amount of caching on the client is allowed */
1900 rc = ll_statfs_internal(ll_s2sbi(sb), &osfs, OBD_STATFS_SUM);
1904 statfs_unpack(sfs, &osfs);
1906 /* We need to downshift for all 32-bit kernels, because we can't
1907 * tell if the kernel is being called via sys_statfs64() or not.
1908 * Stop before overflowing f_bsize - in which case it is better
1909 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1910 if (sizeof(long) < 8) {
1911 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1914 osfs.os_blocks >>= 1;
1915 osfs.os_bfree >>= 1;
1916 osfs.os_bavail >>= 1;
1920 sfs->f_blocks = osfs.os_blocks;
1921 sfs->f_bfree = osfs.os_bfree;
1922 sfs->f_bavail = osfs.os_bavail;
1923 sfs->f_fsid.val[0] = (__u32)fsid;
1924 sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1928 void ll_inode_size_lock(struct inode *inode)
1930 struct ll_inode_info *lli;
1932 LASSERT(!S_ISDIR(inode->i_mode));
1934 lli = ll_i2info(inode);
1935 mutex_lock(&lli->lli_size_mutex);
1938 void ll_inode_size_unlock(struct inode *inode)
1940 struct ll_inode_info *lli;
1942 lli = ll_i2info(inode);
1943 mutex_unlock(&lli->lli_size_mutex);
1946 void ll_update_inode_flags(struct inode *inode, int ext_flags)
1948 inode->i_flags = ll_ext_to_inode_flags(ext_flags);
1949 if (ext_flags & LUSTRE_PROJINHERIT_FL)
1950 ll_file_set_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
1952 ll_file_clear_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
1955 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1957 struct ll_inode_info *lli = ll_i2info(inode);
1958 struct mdt_body *body = md->body;
1959 struct ll_sb_info *sbi = ll_i2sbi(inode);
1962 if (body->mbo_valid & OBD_MD_FLEASIZE) {
1963 rc = cl_file_inode_init(inode, md);
1968 if (S_ISDIR(inode->i_mode)) {
1969 rc = ll_update_lsm_md(inode, md);
1974 #ifdef CONFIG_FS_POSIX_ACL
1975 if (body->mbo_valid & OBD_MD_FLACL) {
1976 spin_lock(&lli->lli_lock);
1977 if (lli->lli_posix_acl)
1978 posix_acl_release(lli->lli_posix_acl);
1979 lli->lli_posix_acl = md->posix_acl;
1980 spin_unlock(&lli->lli_lock);
1983 inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1984 sbi->ll_flags & LL_SBI_32BIT_API);
1985 inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1987 if (body->mbo_valid & OBD_MD_FLATIME) {
1988 if (body->mbo_atime > inode->i_atime.tv_sec)
1989 inode->i_atime.tv_sec = body->mbo_atime;
1990 lli->lli_atime = body->mbo_atime;
1993 if (body->mbo_valid & OBD_MD_FLMTIME) {
1994 if (body->mbo_mtime > inode->i_mtime.tv_sec) {
1996 "setting ino %lu mtime from %lld to %llu\n",
1997 inode->i_ino, (s64)inode->i_mtime.tv_sec,
1999 inode->i_mtime.tv_sec = body->mbo_mtime;
2001 lli->lli_mtime = body->mbo_mtime;
2004 if (body->mbo_valid & OBD_MD_FLCTIME) {
2005 if (body->mbo_ctime > inode->i_ctime.tv_sec)
2006 inode->i_ctime.tv_sec = body->mbo_ctime;
2007 lli->lli_ctime = body->mbo_ctime;
2010 /* Clear i_flags to remove S_NOSEC before permissions are updated */
2011 if (body->mbo_valid & OBD_MD_FLFLAGS)
2012 ll_update_inode_flags(inode, body->mbo_flags);
2013 if (body->mbo_valid & OBD_MD_FLMODE)
2014 inode->i_mode = (inode->i_mode & S_IFMT) |
2015 (body->mbo_mode & ~S_IFMT);
2017 if (body->mbo_valid & OBD_MD_FLTYPE)
2018 inode->i_mode = (inode->i_mode & ~S_IFMT) |
2019 (body->mbo_mode & S_IFMT);
2021 LASSERT(inode->i_mode != 0);
2022 if (S_ISREG(inode->i_mode))
2023 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
2024 LL_MAX_BLKSIZE_BITS);
2026 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
2028 if (body->mbo_valid & OBD_MD_FLUID)
2029 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
2030 if (body->mbo_valid & OBD_MD_FLGID)
2031 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
2032 if (body->mbo_valid & OBD_MD_FLPROJID)
2033 lli->lli_projid = body->mbo_projid;
2034 if (body->mbo_valid & OBD_MD_FLNLINK)
2035 set_nlink(inode, body->mbo_nlink);
2036 if (body->mbo_valid & OBD_MD_FLRDEV)
2037 inode->i_rdev = old_decode_dev(body->mbo_rdev);
2039 if (body->mbo_valid & OBD_MD_FLID) {
2040 /* FID shouldn't be changed! */
2041 if (fid_is_sane(&lli->lli_fid)) {
2042 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
2043 "Trying to change FID "DFID
2044 " to the "DFID", inode "DFID"(%p)\n",
2045 PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
2046 PFID(ll_inode2fid(inode)), inode);
2048 lli->lli_fid = body->mbo_fid1;
2052 LASSERT(fid_seq(&lli->lli_fid) != 0);
2054 if (body->mbo_valid & OBD_MD_FLSIZE) {
2055 i_size_write(inode, body->mbo_size);
2057 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
2058 PFID(ll_inode2fid(inode)),
2059 (unsigned long long)body->mbo_size);
2061 if (body->mbo_valid & OBD_MD_FLBLOCKS)
2062 inode->i_blocks = body->mbo_blocks;
2065 if (body->mbo_valid & OBD_MD_TSTATE) {
2066 /* Set LLIF_FILE_RESTORING if restore ongoing and
2067 * clear it when done to ensure to start again
2068 * glimpsing updated attrs
2070 if (body->mbo_t_state & MS_RESTORE)
2071 ll_file_set_flag(lli, LLIF_FILE_RESTORING);
2073 ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
2079 int ll_read_inode2(struct inode *inode, void *opaque)
2081 struct lustre_md *md = opaque;
2082 struct ll_inode_info *lli = ll_i2info(inode);
2086 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2087 PFID(&lli->lli_fid), inode);
2089 /* Core attributes from the MDS first. This is a new inode, and
2090 * the VFS doesn't zero times in the core inode so we have to do
2091 * it ourselves. They will be overwritten by either MDS or OST
2092 * attributes - we just need to make sure they aren't newer.
2094 inode->i_mtime.tv_sec = 0;
2095 inode->i_atime.tv_sec = 0;
2096 inode->i_ctime.tv_sec = 0;
2098 rc = ll_update_inode(inode, md);
2102 /* OIDEBUG(inode); */
2104 #ifdef HAVE_BACKING_DEV_INFO
2105 /* initializing backing dev info. */
2106 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2108 if (S_ISREG(inode->i_mode)) {
2109 struct ll_sb_info *sbi = ll_i2sbi(inode);
2110 inode->i_op = &ll_file_inode_operations;
2111 inode->i_fop = sbi->ll_fop;
2112 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2114 } else if (S_ISDIR(inode->i_mode)) {
2115 inode->i_op = &ll_dir_inode_operations;
2116 inode->i_fop = &ll_dir_operations;
2118 } else if (S_ISLNK(inode->i_mode)) {
2119 inode->i_op = &ll_fast_symlink_inode_operations;
2122 inode->i_op = &ll_special_inode_operations;
2124 init_special_inode(inode, inode->i_mode,
2133 void ll_delete_inode(struct inode *inode)
2135 struct ll_inode_info *lli = ll_i2info(inode);
2136 struct address_space *mapping = &inode->i_data;
2137 unsigned long nrpages;
2140 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2141 /* It is last chance to write out dirty pages,
2142 * otherwise we may lose data while umount */
2143 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
2145 truncate_inode_pages_final(mapping);
2147 /* Workaround for LU-118: Note nrpages may not be totally updated when
2148 * truncate_inode_pages() returns, as there can be a page in the process
2149 * of deletion (inside __delete_from_page_cache()) in the specified
2150 * range. Thus mapping->nrpages can be non-zero when this function
2151 * returns even after truncation of the whole mapping. Only do this if
2152 * npages isn't already zero.
2154 nrpages = mapping->nrpages;
2156 xa_lock_irq(&mapping->i_pages);
2157 nrpages = mapping->nrpages;
2158 xa_unlock_irq(&mapping->i_pages);
2159 } /* Workaround end */
2161 LASSERTF(nrpages == 0, "%s: inode="DFID"(%p) nrpages=%lu, "
2162 "see https://jira.whamcloud.com/browse/LU-118\n",
2163 ll_i2sbi(inode)->ll_fsname,
2164 PFID(ll_inode2fid(inode)), inode, nrpages);
2166 #ifdef HAVE_SBOPS_EVICT_INODE
2167 ll_clear_inode(inode);
2174 int ll_iocontrol(struct inode *inode, struct file *file,
2175 unsigned int cmd, unsigned long arg)
2177 struct ll_sb_info *sbi = ll_i2sbi(inode);
2178 struct ptlrpc_request *req = NULL;
2183 case FS_IOC_GETFLAGS: {
2184 struct mdt_body *body;
2185 struct md_op_data *op_data;
2187 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2188 0, 0, LUSTRE_OPC_ANY,
2190 if (IS_ERR(op_data))
2191 RETURN(PTR_ERR(op_data));
2193 op_data->op_valid = OBD_MD_FLFLAGS;
2194 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2195 ll_finish_md_op_data(op_data);
2197 CERROR("%s: failure inode "DFID": rc = %d\n",
2198 sbi->ll_md_exp->exp_obd->obd_name,
2199 PFID(ll_inode2fid(inode)), rc);
2203 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2205 flags = body->mbo_flags;
2207 ptlrpc_req_finished(req);
2209 RETURN(put_user(flags, (int __user *)arg));
2211 case FS_IOC_SETFLAGS: {
2213 struct md_op_data *op_data;
2214 struct cl_object *obj;
2215 struct fsxattr fa = { 0 };
2217 if (get_user(flags, (int __user *)arg))
2220 fa.fsx_projid = ll_i2info(inode)->lli_projid;
2221 if (flags & LUSTRE_PROJINHERIT_FL)
2222 fa.fsx_xflags = FS_XFLAG_PROJINHERIT;
2224 rc = ll_ioctl_check_project(inode, &fa);
2228 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2229 LUSTRE_OPC_ANY, NULL);
2230 if (IS_ERR(op_data))
2231 RETURN(PTR_ERR(op_data));
2233 op_data->op_attr_flags = flags;
2234 op_data->op_xvalid |= OP_XVALID_FLAGS;
2235 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2236 ll_finish_md_op_data(op_data);
2237 ptlrpc_req_finished(req);
2241 ll_update_inode_flags(inode, flags);
2243 obj = ll_i2info(inode)->lli_clob;
2247 OBD_ALLOC_PTR(attr);
2251 rc = cl_setattr_ost(obj, attr, OP_XVALID_FLAGS, flags);
2263 int ll_flush_ctx(struct inode *inode)
2265 struct ll_sb_info *sbi = ll_i2sbi(inode);
2267 CDEBUG(D_SEC, "flush context for user %d\n",
2268 from_kuid(&init_user_ns, current_uid()));
2270 obd_set_info_async(NULL, sbi->ll_md_exp,
2271 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2273 obd_set_info_async(NULL, sbi->ll_dt_exp,
2274 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2279 /* umount -f client means force down, don't save state */
2280 void ll_umount_begin(struct super_block *sb)
2282 struct ll_sb_info *sbi = ll_s2sbi(sb);
2283 struct obd_device *obd;
2284 struct obd_ioctl_data *ioc_data;
2285 struct l_wait_info lwi;
2286 wait_queue_head_t waitq;
2289 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2290 sb->s_count, atomic_read(&sb->s_active));
2292 obd = class_exp2obd(sbi->ll_md_exp);
2294 CERROR("Invalid MDC connection handle %#llx\n",
2295 sbi->ll_md_exp->exp_handle.h_cookie);
2301 obd = class_exp2obd(sbi->ll_dt_exp);
2303 CERROR("Invalid LOV connection handle %#llx\n",
2304 sbi->ll_dt_exp->exp_handle.h_cookie);
2310 OBD_ALLOC_PTR(ioc_data);
2312 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2313 sizeof *ioc_data, ioc_data, NULL);
2315 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2316 sizeof *ioc_data, ioc_data, NULL);
2318 OBD_FREE_PTR(ioc_data);
2321 /* Really, we'd like to wait until there are no requests outstanding,
2322 * and then continue. For now, we just periodically checking for vfs
2323 * to decrement mnt_cnt and hope to finish it within 10sec.
2325 init_waitqueue_head(&waitq);
2326 lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(10),
2327 cfs_time_seconds(1), NULL, NULL);
2328 l_wait_event(waitq, may_umount(sbi->ll_mnt.mnt), &lwi);
2333 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2335 struct ll_sb_info *sbi = ll_s2sbi(sb);
2336 char *profilenm = get_profile_name(sb);
2340 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2341 read_only = *flags & MS_RDONLY;
2342 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2343 sizeof(KEY_READ_ONLY),
2344 KEY_READ_ONLY, sizeof(read_only),
2347 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2348 profilenm, read_only ?
2349 "read-only" : "read-write", err);
2354 sb->s_flags |= MS_RDONLY;
2356 sb->s_flags &= ~MS_RDONLY;
2358 if (sbi->ll_flags & LL_SBI_VERBOSE)
2359 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2360 read_only ? "read-only" : "read-write");
2366 * Cleanup the open handle that is cached on MDT-side.
2368 * For open case, the client side open handling thread may hit error
2369 * after the MDT grant the open. Under such case, the client should
2370 * send close RPC to the MDT as cleanup; otherwise, the open handle
2371 * on the MDT will be leaked there until the client umount or evicted.
2373 * In further, if someone unlinked the file, because the open handle
2374 * holds the reference on such file/object, then it will block the
2375 * subsequent threads that want to locate such object via FID.
2377 * \param[in] sb super block for this file-system
2378 * \param[in] open_req pointer to the original open request
2380 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2382 struct mdt_body *body;
2383 struct md_op_data *op_data;
2384 struct ptlrpc_request *close_req = NULL;
2385 struct obd_export *exp = ll_s2sbi(sb)->ll_md_exp;
2388 body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2389 OBD_ALLOC_PTR(op_data);
2390 if (op_data == NULL) {
2391 CWARN("%s: cannot allocate op_data to release open handle for "
2392 DFID"\n", ll_s2sbi(sb)->ll_fsname, PFID(&body->mbo_fid1));
2397 op_data->op_fid1 = body->mbo_fid1;
2398 op_data->op_open_handle = body->mbo_open_handle;
2399 op_data->op_mod_time = ktime_get_real_seconds();
2400 md_close(exp, op_data, NULL, &close_req);
2401 ptlrpc_req_finished(close_req);
2402 ll_finish_md_op_data(op_data);
2407 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2408 struct super_block *sb, struct lookup_intent *it)
2410 struct ll_sb_info *sbi = NULL;
2411 struct lustre_md md = { NULL };
2415 LASSERT(*inode || sb);
2416 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2417 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2418 sbi->ll_md_exp, &md);
2423 rc = ll_update_inode(*inode, &md);
2427 LASSERT(sb != NULL);
2430 * At this point server returns to client's same fid as client
2431 * generated for creating. So using ->fid1 is okay here.
2433 if (!fid_is_sane(&md.body->mbo_fid1)) {
2434 CERROR("%s: Fid is insane "DFID"\n",
2436 PFID(&md.body->mbo_fid1));
2437 GOTO(out, rc = -EINVAL);
2440 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2441 sbi->ll_flags & LL_SBI_32BIT_API),
2443 if (IS_ERR(*inode)) {
2444 #ifdef CONFIG_FS_POSIX_ACL
2446 posix_acl_release(md.posix_acl);
2447 md.posix_acl = NULL;
2450 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2452 CERROR("new_inode -fatal: rc %d\n", rc);
2457 /* Handling piggyback layout lock.
2458 * Layout lock can be piggybacked by getattr and open request.
2459 * The lsm can be applied to inode only if it comes with a layout lock
2460 * otherwise correct layout may be overwritten, for example:
2461 * 1. proc1: mdt returns a lsm but not granting layout
2462 * 2. layout was changed by another client
2463 * 3. proc2: refresh layout and layout lock granted
2464 * 4. proc1: to apply a stale layout */
2465 if (it != NULL && it->it_lock_mode != 0) {
2466 struct lustre_handle lockh;
2467 struct ldlm_lock *lock;
2469 lockh.cookie = it->it_lock_handle;
2470 lock = ldlm_handle2lock(&lockh);
2471 LASSERT(lock != NULL);
2472 if (ldlm_has_layout(lock)) {
2473 struct cl_object_conf conf;
2475 memset(&conf, 0, sizeof(conf));
2476 conf.coc_opc = OBJECT_CONF_SET;
2477 conf.coc_inode = *inode;
2478 conf.coc_lock = lock;
2479 conf.u.coc_layout = md.layout;
2480 (void)ll_layout_conf(*inode, &conf);
2482 LDLM_LOCK_PUT(lock);
2488 /* cleanup will be done if necessary */
2489 md_free_lustre_md(sbi->ll_md_exp, &md);
2491 if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2492 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2497 int ll_obd_statfs(struct inode *inode, void __user *arg)
2499 struct ll_sb_info *sbi = NULL;
2500 struct obd_export *exp;
2502 struct obd_ioctl_data *data = NULL;
2506 if (!inode || !(sbi = ll_i2sbi(inode)))
2507 GOTO(out_statfs, rc = -EINVAL);
2509 rc = obd_ioctl_getdata(&buf, &len, arg);
2511 GOTO(out_statfs, rc);
2514 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2515 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2516 GOTO(out_statfs, rc = -EINVAL);
2518 if (data->ioc_inllen1 != sizeof(__u32) ||
2519 data->ioc_inllen2 != sizeof(__u32) ||
2520 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2521 data->ioc_plen2 != sizeof(struct obd_uuid))
2522 GOTO(out_statfs, rc = -EINVAL);
2524 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2525 if (type & LL_STATFS_LMV)
2526 exp = sbi->ll_md_exp;
2527 else if (type & LL_STATFS_LOV)
2528 exp = sbi->ll_dt_exp;
2530 GOTO(out_statfs, rc = -ENODEV);
2532 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2534 GOTO(out_statfs, rc);
2536 OBD_FREE_LARGE(buf, len);
2541 * this is normally called in ll_fini_md_op_data(), but sometimes it needs to
2542 * be called early to avoid deadlock.
2544 void ll_unlock_md_op_lsm(struct md_op_data *op_data)
2546 if (op_data->op_mea2_sem) {
2547 up_read(op_data->op_mea2_sem);
2548 op_data->op_mea2_sem = NULL;
2551 if (op_data->op_mea1_sem) {
2552 up_read(op_data->op_mea1_sem);
2553 op_data->op_mea1_sem = NULL;
2557 /* this function prepares md_op_data hint for passing it down to MD stack. */
2558 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2559 struct inode *i1, struct inode *i2,
2560 const char *name, size_t namelen,
2561 __u32 mode, __u32 opc, void *data)
2563 LASSERT(i1 != NULL);
2566 /* Do not reuse namelen for something else. */
2568 return ERR_PTR(-EINVAL);
2570 if (namelen > ll_i2sbi(i1)->ll_namelen)
2571 return ERR_PTR(-ENAMETOOLONG);
2573 if (!lu_name_is_valid_2(name, namelen))
2574 return ERR_PTR(-EINVAL);
2577 if (op_data == NULL)
2578 OBD_ALLOC_PTR(op_data);
2580 if (op_data == NULL)
2581 return ERR_PTR(-ENOMEM);
2583 ll_i2gids(op_data->op_suppgids, i1, i2);
2584 op_data->op_fid1 = *ll_inode2fid(i1);
2585 op_data->op_default_stripe_offset = -1;
2587 if (S_ISDIR(i1->i_mode)) {
2588 down_read(&ll_i2info(i1)->lli_lsm_sem);
2589 op_data->op_mea1_sem = &ll_i2info(i1)->lli_lsm_sem;
2590 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2591 if (opc == LUSTRE_OPC_MKDIR)
2592 op_data->op_default_stripe_offset =
2593 ll_i2info(i1)->lli_def_stripe_offset;
2597 op_data->op_fid2 = *ll_inode2fid(i2);
2598 if (S_ISDIR(i2->i_mode)) {
2600 down_read(&ll_i2info(i2)->lli_lsm_sem);
2601 op_data->op_mea2_sem =
2602 &ll_i2info(i2)->lli_lsm_sem;
2604 op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2607 fid_zero(&op_data->op_fid2);
2610 if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2611 op_data->op_cli_flags |= CLI_HASH64;
2613 if (ll_need_32bit_api(ll_i2sbi(i1)))
2614 op_data->op_cli_flags |= CLI_API32;
2616 op_data->op_name = name;
2617 op_data->op_namelen = namelen;
2618 op_data->op_mode = mode;
2619 op_data->op_mod_time = ktime_get_real_seconds();
2620 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2621 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2622 op_data->op_cap = cfs_curproc_cap_pack();
2623 op_data->op_mds = 0;
2624 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2625 filename_is_volatile(name, namelen, &op_data->op_mds)) {
2626 op_data->op_bias |= MDS_CREATE_VOLATILE;
2628 op_data->op_data = data;
2633 void ll_finish_md_op_data(struct md_op_data *op_data)
2635 ll_unlock_md_op_lsm(op_data);
2636 security_release_secctx(op_data->op_file_secctx,
2637 op_data->op_file_secctx_size);
2638 OBD_FREE_PTR(op_data);
2641 #ifdef HAVE_SUPEROPS_USE_DENTRY
2642 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2644 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2647 struct ll_sb_info *sbi;
2649 #ifdef HAVE_SUPEROPS_USE_DENTRY
2650 LASSERT((seq != NULL) && (dentry != NULL));
2651 sbi = ll_s2sbi(dentry->d_sb);
2653 LASSERT((seq != NULL) && (vfs != NULL));
2654 sbi = ll_s2sbi(vfs->mnt_sb);
2657 if (sbi->ll_flags & LL_SBI_NOLCK)
2658 seq_puts(seq, ",nolock");
2660 if (sbi->ll_flags & LL_SBI_FLOCK)
2661 seq_puts(seq, ",flock");
2663 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2664 seq_puts(seq, ",localflock");
2666 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2667 seq_puts(seq, ",user_xattr");
2669 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2670 seq_puts(seq, ",lazystatfs");
2672 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2673 seq_puts(seq, ",user_fid2path");
2675 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2676 seq_puts(seq, ",always_ping");
2682 * Get obd name by cmd, and copy out to user space
2684 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2686 struct ll_sb_info *sbi = ll_i2sbi(inode);
2687 struct obd_device *obd;
2690 if (cmd == OBD_IOC_GETDTNAME)
2691 obd = class_exp2obd(sbi->ll_dt_exp);
2692 else if (cmd == OBD_IOC_GETMDNAME)
2693 obd = class_exp2obd(sbi->ll_md_exp);
2700 if (copy_to_user((void __user *)arg, obd->obd_name,
2701 strlen(obd->obd_name) + 1))
2707 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2714 p.mnt = current->fs->root.mnt;
2716 path = d_path(&p, buf, bufsize);
2721 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2723 char *buf, *path = NULL;
2724 struct dentry *dentry = NULL;
2725 struct inode *inode = page->mapping->host;
2727 /* this can be called inside spin lock so use GFP_ATOMIC. */
2728 buf = (char *)__get_free_page(GFP_ATOMIC);
2730 dentry = d_find_alias(page->mapping->host);
2732 path = ll_d_path(dentry, buf, PAGE_SIZE);
2736 "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2737 "(rc %d)\n", ll_i2sbi(inode)->ll_fsname,
2738 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2739 PFID(ll_inode2fid(inode)),
2740 (path && !IS_ERR(path)) ? path : "", ioret);
2746 free_page((unsigned long)buf);
2749 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2750 struct lov_user_md **kbuf)
2752 struct lov_user_md lum;
2756 if (copy_from_user(&lum, md, sizeof(lum)))
2759 lum_size = ll_lov_user_md_size(&lum);
2763 OBD_ALLOC(*kbuf, lum_size);
2767 if (copy_from_user(*kbuf, md, lum_size) != 0) {
2768 OBD_FREE(*kbuf, lum_size);
2776 * Compute llite root squash state after a change of root squash
2777 * configuration setting or add/remove of a lnet nid
2779 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2781 struct root_squash_info *squash = &sbi->ll_squash;
2784 struct lnet_process_id id;
2786 /* Update norootsquash flag */
2787 down_write(&squash->rsi_sem);
2788 if (list_empty(&squash->rsi_nosquash_nids))
2789 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2791 /* Do not apply root squash as soon as one of our NIDs is
2792 * in the nosquash_nids list */
2795 while (LNetGetId(i++, &id) != -ENOENT) {
2796 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2798 if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2804 sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2806 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2808 up_write(&squash->rsi_sem);
2812 * Parse linkea content to extract information about a given hardlink
2814 * \param[in] ldata - Initialized linkea data
2815 * \param[in] linkno - Link identifier
2816 * \param[out] parent_fid - The entry's parent FID
2817 * \param[out] ln - Entry name destination buffer
2819 * \retval 0 on success
2820 * \retval Appropriate negative error code on failure
2822 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2823 struct lu_fid *parent_fid, struct lu_name *ln)
2829 rc = linkea_init_with_rec(ldata);
2833 if (linkno >= ldata->ld_leh->leh_reccount)
2834 /* beyond last link */
2837 linkea_first_entry(ldata);
2838 for (idx = 0; ldata->ld_lee != NULL; idx++) {
2839 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2844 linkea_next_entry(ldata);
2854 * Get parent FID and name of an identified link. Operation is performed for
2855 * a given link number, letting the caller iterate over linkno to list one or
2856 * all links of an entry.
2858 * \param[in] file - File descriptor against which to perform the operation
2859 * \param[in,out] arg - User-filled structure containing the linkno to operate
2860 * on and the available size. It is eventually filled with
2861 * the requested information or left untouched on error
2863 * \retval - 0 on success
2864 * \retval - Appropriate negative error code on failure
2866 int ll_getparent(struct file *file, struct getparent __user *arg)
2868 struct inode *inode = file_inode(file);
2869 struct linkea_data *ldata;
2870 struct lu_buf buf = LU_BUF_NULL;
2872 struct lu_fid parent_fid;
2879 if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2880 !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2883 if (get_user(name_size, &arg->gp_name_size))
2886 if (get_user(linkno, &arg->gp_linkno))
2889 if (name_size > PATH_MAX)
2892 OBD_ALLOC(ldata, sizeof(*ldata));
2896 rc = linkea_data_new(ldata, &buf);
2898 GOTO(ldata_free, rc);
2900 #ifdef HAVE_XATTR_HANDLER_FLAGS
2901 rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
2902 buf.lb_len, OBD_MD_FLXATTR);
2904 rc = ll_getxattr(file_dentry(file), XATTR_NAME_LINK, buf.lb_buf,
2906 #endif /* HAVE_XATTR_HANDLER_FLAGS */
2910 rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2914 if (ln.ln_namelen >= name_size)
2915 GOTO(lb_free, rc = -EOVERFLOW);
2917 if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2918 GOTO(lb_free, rc = -EFAULT);
2920 if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2921 GOTO(lb_free, rc = -EFAULT);
2923 if (put_user('\0', arg->gp_name + ln.ln_namelen))
2924 GOTO(lb_free, rc = -EFAULT);
2929 OBD_FREE(ldata, sizeof(*ldata));