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 sbi->ll_flags |= LL_SBI_FLOCK;
116 #ifdef HAVE_LRU_RESIZE_SUPPORT
117 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
119 sbi->ll_flags |= LL_SBI_LAZYSTATFS;
121 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
122 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
124 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
128 /* metadata statahead is enabled by default */
129 sbi->ll_sa_running_max = LL_SA_RUNNING_DEF;
130 sbi->ll_sa_max = LL_SA_RPC_DEF;
131 atomic_set(&sbi->ll_sa_total, 0);
132 atomic_set(&sbi->ll_sa_wrong, 0);
133 atomic_set(&sbi->ll_sa_running, 0);
134 atomic_set(&sbi->ll_agl_total, 0);
135 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
136 sbi->ll_flags |= LL_SBI_FAST_READ;
137 sbi->ll_flags |= LL_SBI_TINY_WRITE;
140 sbi->ll_squash.rsi_uid = 0;
141 sbi->ll_squash.rsi_gid = 0;
142 INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
143 init_rwsem(&sbi->ll_squash.rsi_sem);
145 /* Per-filesystem file heat */
146 sbi->ll_heat_decay_weight = SBI_DEFAULT_HEAT_DECAY_WEIGHT;
147 sbi->ll_heat_period_second = SBI_DEFAULT_HEAT_PERIOD_SECOND;
151 static void ll_free_sbi(struct super_block *sb)
153 struct ll_sb_info *sbi = ll_s2sbi(sb);
157 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
158 cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
159 if (sbi->ll_cache != NULL) {
160 cl_cache_decref(sbi->ll_cache);
161 sbi->ll_cache = NULL;
163 OBD_FREE(sbi, sizeof(*sbi));
168 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
169 struct vfsmount *mnt)
171 struct inode *root = NULL;
172 struct ll_sb_info *sbi = ll_s2sbi(sb);
173 struct obd_statfs *osfs = NULL;
174 struct ptlrpc_request *request = NULL;
175 struct obd_connect_data *data = NULL;
176 struct obd_uuid *uuid;
177 struct md_op_data *op_data;
178 struct lustre_md lmd;
180 int size, err, checksum;
183 sbi->ll_md_obd = class_name2obd(md);
184 if (!sbi->ll_md_obd) {
185 CERROR("MD %s: not setup or attached\n", md);
199 /* pass client page size via ocd_grant_blkbits, the server should report
200 * back its backend blocksize for grant calculation purpose */
201 data->ocd_grant_blkbits = PAGE_SHIFT;
203 /* indicate MDT features supported by this client */
204 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
205 OBD_CONNECT_ATTRFID | OBD_CONNECT_GRANT |
206 OBD_CONNECT_VERSION | OBD_CONNECT_BRW_SIZE |
207 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
208 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
209 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
210 OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 |
211 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
212 OBD_CONNECT_64BITHASH |
213 OBD_CONNECT_EINPROGRESS |
214 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
215 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS|
216 OBD_CONNECT_MAX_EASIZE |
217 OBD_CONNECT_FLOCK_DEAD |
218 OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
219 OBD_CONNECT_OPEN_BY_FID |
220 OBD_CONNECT_DIR_STRIPE |
221 OBD_CONNECT_BULK_MBITS | OBD_CONNECT_CKSUM |
222 OBD_CONNECT_SUBTREE |
223 OBD_CONNECT_MULTIMODRPCS |
224 OBD_CONNECT_GRANT_PARAM |
225 OBD_CONNECT_SHORTIO | OBD_CONNECT_FLAGS2;
227 data->ocd_connect_flags2 = OBD_CONNECT2_FLR |
228 OBD_CONNECT2_LOCK_CONVERT |
229 OBD_CONNECT2_DIR_MIGRATE |
230 OBD_CONNECT2_SUM_STATFS |
231 OBD_CONNECT2_ARCHIVE_ID_ARRAY |
234 #ifdef HAVE_LRU_RESIZE_SUPPORT
235 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
236 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
238 #ifdef CONFIG_FS_POSIX_ACL
239 data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK |
240 OBD_CONNECT_LARGE_ACL;
243 data->ocd_cksum_types = obd_cksum_types_supported_client();
245 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
246 /* flag mdc connection as lightweight, only used for test
247 * purpose, use with care */
248 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
250 data->ocd_ibits_known = MDS_INODELOCK_FULL;
251 data->ocd_version = LUSTRE_VERSION_CODE;
253 if (sb->s_flags & MS_RDONLY)
254 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
255 if (sbi->ll_flags & LL_SBI_USER_XATTR)
256 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
259 /* Setting this indicates we correctly support S_NOSEC (See kernel
260 * commit 9e1f1de02c2275d7172e18dc4e7c2065777611bf)
262 sb->s_flags |= MS_NOSEC;
265 if (sbi->ll_flags & LL_SBI_FLOCK)
266 sbi->ll_fop = &ll_file_operations_flock;
267 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
268 sbi->ll_fop = &ll_file_operations;
270 sbi->ll_fop = &ll_file_operations_noflock;
272 /* always ping even if server suppress_pings */
273 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
274 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
276 obd_connect_set_secctx(data);
278 #if defined(CONFIG_SECURITY)
279 data->ocd_connect_flags2 |= OBD_CONNECT2_SELINUX_POLICY;
282 data->ocd_brw_size = MD_MAX_BRW_SIZE;
284 err = obd_connect(NULL, &sbi->ll_md_exp, sbi->ll_md_obd,
285 &sbi->ll_sb_uuid, data, sbi->ll_cache);
287 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
288 "recovery, of which this client is not a "
289 "part. Please wait for recovery to complete,"
290 " abort, or time out.\n", md);
293 CERROR("cannot connect to %s: rc = %d\n", md, err);
297 sbi->ll_md_exp->exp_connect_data = *data;
299 err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
300 LUSTRE_SEQ_METADATA);
302 CERROR("%s: Can't init metadata layer FID infrastructure, "
303 "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
307 /* For mount, we only need fs info from MDT0, and also in DNE, it
308 * can make sure the client can be mounted as long as MDT0 is
310 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
311 ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
312 OBD_STATFS_FOR_MDT0);
314 GOTO(out_md_fid, err);
316 /* This needs to be after statfs to ensure connect has finished.
317 * Note that "data" does NOT contain the valid connect reply.
318 * If connecting to a 1.8 server there will be no LMV device, so
319 * we can access the MDC export directly and exp_connect_flags will
320 * be non-zero, but if accessing an upgraded 2.1 server it will
321 * have the correct flags filled in.
322 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
323 valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
324 if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
325 valid != CLIENT_CONNECT_MDT_REQD) {
328 OBD_ALLOC_WAIT(buf, PAGE_SIZE);
329 obd_connect_flags2str(buf, PAGE_SIZE,
330 valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
331 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
332 "feature(s) needed for correct operation "
333 "of this client (%s). Please upgrade "
334 "server or downgrade client.\n",
335 sbi->ll_md_exp->exp_obd->obd_name, buf);
336 OBD_FREE(buf, PAGE_SIZE);
337 GOTO(out_md_fid, err = -EPROTO);
340 size = sizeof(*data);
341 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
342 KEY_CONN_DATA, &size, data);
344 CERROR("%s: Get connect data failed: rc = %d\n",
345 sbi->ll_md_exp->exp_obd->obd_name, err);
346 GOTO(out_md_fid, err);
349 LASSERT(osfs->os_bsize);
350 sb->s_blocksize = osfs->os_bsize;
351 sb->s_blocksize_bits = log2(osfs->os_bsize);
352 sb->s_magic = LL_SUPER_MAGIC;
353 sb->s_maxbytes = MAX_LFS_FILESIZE;
354 sbi->ll_namelen = osfs->os_namelen;
355 sbi->ll_mnt.mnt = current->fs->root.mnt;
357 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
358 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
359 LCONSOLE_INFO("Disabling user_xattr feature because "
360 "it is not supported on the server\n");
361 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
364 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
366 sb->s_flags |= MS_POSIXACL;
368 sbi->ll_flags |= LL_SBI_ACL;
370 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
372 sb->s_flags &= ~MS_POSIXACL;
374 sbi->ll_flags &= ~LL_SBI_ACL;
377 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
378 sbi->ll_flags |= LL_SBI_64BIT_HASH;
380 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
381 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
383 if (obd_connect_has_secctx(data))
384 sbi->ll_flags |= LL_SBI_FILE_SECCTX;
386 if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
387 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
388 LCONSOLE_INFO("%s: disabling xattr cache due to "
389 "unknown maximum xattr size.\n", dt);
390 } else if (!sbi->ll_xattr_cache_set) {
391 /* If xattr_cache is already set (no matter 0 or 1)
392 * during processing llog, it won't be enabled here. */
393 sbi->ll_flags |= LL_SBI_XATTR_CACHE;
394 sbi->ll_xattr_cache_enabled = 1;
398 sbi->ll_dt_obd = class_name2obd(dt);
399 if (!sbi->ll_dt_obd) {
400 CERROR("DT %s: not setup or attached\n", dt);
401 GOTO(out_md_fid, err = -ENODEV);
404 /* pass client page size via ocd_grant_blkbits, the server should report
405 * back its backend blocksize for grant calculation purpose */
406 data->ocd_grant_blkbits = PAGE_SHIFT;
408 /* indicate OST features supported by this client */
409 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
410 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
411 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
412 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
413 OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
414 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
415 OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
416 OBD_CONNECT_EINPROGRESS |
417 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
418 OBD_CONNECT_LAYOUTLOCK |
419 OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
420 OBD_CONNECT_BULK_MBITS | OBD_CONNECT_SHORTIO |
421 OBD_CONNECT_FLAGS2 | OBD_CONNECT_GRANT_SHRINK;
423 /* The client currently advertises support for OBD_CONNECT_LOCKAHEAD_OLD so it
424 * can interoperate with an older version of lockahead which was released prior
425 * to landing in master. This support will be dropped when 2.13 development
426 * starts. At the point, we should not just drop the connect flag (below), we
427 * should also remove the support in the code.
429 * Removing it means a few things:
430 * 1. Remove this section here
431 * 2. Remove CEF_NONBLOCK in ll_file_lockahead()
432 * 3. Remove function exp_connect_lockahead_old
433 * 4. Remove LDLM_FL_LOCKAHEAD_OLD_RESERVED in lustre_dlm_flags.h
435 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 12, 50, 0)
436 data->ocd_connect_flags |= OBD_CONNECT_LOCKAHEAD_OLD;
439 data->ocd_connect_flags2 = OBD_CONNECT2_LOCKAHEAD;
441 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
442 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
444 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
445 * disabled by default, because it can still be enabled on the
446 * fly via /sys. As a consequence, we still need to come to an
447 * agreement on the supported algorithms at connect time
449 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
451 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
452 data->ocd_cksum_types = OBD_CKSUM_ADLER;
454 data->ocd_cksum_types = obd_cksum_types_supported_client();
456 #ifdef HAVE_LRU_RESIZE_SUPPORT
457 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
459 /* always ping even if server suppress_pings */
460 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
461 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
463 CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
464 "ocd_grant: %d\n", data->ocd_connect_flags,
465 data->ocd_version, data->ocd_grant);
467 sbi->ll_dt_obd->obd_upcall.onu_owner = &sbi->ll_lco;
468 sbi->ll_dt_obd->obd_upcall.onu_upcall = cl_ocd_update;
470 data->ocd_brw_size = DT_MAX_BRW_SIZE;
472 err = obd_connect(NULL, &sbi->ll_dt_exp, sbi->ll_dt_obd,
473 &sbi->ll_sb_uuid, data, sbi->ll_cache);
475 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
476 "recovery, of which this client is not a "
477 "part. Please wait for recovery to "
478 "complete, abort, or time out.\n", dt);
481 CERROR("%s: Cannot connect to %s: rc = %d\n",
482 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
486 sbi->ll_dt_exp->exp_connect_data = *data;
488 /* Don't change value if it was specified in the config log */
489 if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages == -1)
490 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
491 max_t(unsigned long, SBI_DEFAULT_READAHEAD_WHOLE_MAX,
492 (data->ocd_brw_size >> PAGE_SHIFT));
494 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
495 LUSTRE_SEQ_METADATA);
497 CERROR("%s: Can't init data layer FID infrastructure, "
498 "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
502 mutex_lock(&sbi->ll_lco.lco_lock);
503 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
504 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
505 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
506 mutex_unlock(&sbi->ll_lco.lco_lock);
508 fid_zero(&sbi->ll_root_fid);
509 err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
512 CERROR("cannot mds_connect: rc = %d\n", err);
513 GOTO(out_lock_cn_cb, err);
515 if (!fid_is_sane(&sbi->ll_root_fid)) {
516 CERROR("%s: Invalid root fid "DFID" during mount\n",
517 sbi->ll_md_exp->exp_obd->obd_name,
518 PFID(&sbi->ll_root_fid));
519 GOTO(out_lock_cn_cb, err = -EINVAL);
521 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
523 sb->s_op = &lustre_super_operations;
524 #ifdef HAVE_XATTR_HANDLER_FLAGS
525 sb->s_xattr = ll_xattr_handlers;
527 #if THREAD_SIZE >= 8192 /*b=17630*/
528 sb->s_export_op = &lustre_export_operations;
532 * XXX: move this to after cbd setup? */
533 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
534 if (sbi->ll_flags & LL_SBI_ACL)
535 valid |= OBD_MD_FLACL;
537 OBD_ALLOC_PTR(op_data);
539 GOTO(out_lock_cn_cb, err = -ENOMEM);
541 op_data->op_fid1 = sbi->ll_root_fid;
542 op_data->op_mode = 0;
543 op_data->op_valid = valid;
545 err = md_getattr(sbi->ll_md_exp, op_data, &request);
547 OBD_FREE_PTR(op_data);
549 CERROR("%s: md_getattr failed for root: rc = %d\n",
550 sbi->ll_md_exp->exp_obd->obd_name, err);
551 GOTO(out_lock_cn_cb, err);
554 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
555 sbi->ll_md_exp, &lmd);
557 CERROR("failed to understand root inode md: rc = %d\n", err);
558 ptlrpc_req_finished(request);
559 GOTO(out_lock_cn_cb, err);
562 LASSERT(fid_is_sane(&sbi->ll_root_fid));
563 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
564 sbi->ll_flags & LL_SBI_32BIT_API),
566 md_free_lustre_md(sbi->ll_md_exp, &lmd);
567 ptlrpc_req_finished(request);
570 #ifdef CONFIG_FS_POSIX_ACL
572 posix_acl_release(lmd.posix_acl);
573 lmd.posix_acl = NULL;
576 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
578 CERROR("lustre_lite: bad iget4 for root\n");
582 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
583 if (sbi->ll_checksum_set) {
584 err = obd_set_info_async(NULL, sbi->ll_dt_exp,
585 sizeof(KEY_CHECKSUM), KEY_CHECKSUM,
586 sizeof(checksum), &checksum, NULL);
588 CERROR("%s: Set checksum failed: rc = %d\n",
589 sbi->ll_dt_exp->exp_obd->obd_name, err);
595 sb->s_root = d_make_root(root);
596 if (sb->s_root == NULL) {
597 CERROR("%s: can't make root dentry\n",
598 ll_get_fsname(sb, NULL, 0));
599 GOTO(out_root, err = -ENOMEM);
601 #ifdef HAVE_DCACHE_LOCK
602 sb->s_root->d_op = &ll_d_ops;
605 sbi->ll_sdev_orig = sb->s_dev;
607 /* We set sb->s_dev equal on all lustre clients in order to support
608 * NFS export clustering. NFSD requires that the FSID be the same
610 /* s_dev is also used in lt_compare() to compare two fs, but that is
611 * only a node-local comparison. */
612 uuid = obd_get_uuid(sbi->ll_md_exp);
614 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
621 if (sbi->ll_dt_obd) {
622 err = sysfs_create_link(&sbi->ll_kset.kobj,
623 &sbi->ll_dt_obd->obd_kset.kobj,
624 sbi->ll_dt_obd->obd_type->typ_name);
626 CERROR("%s: could not register %s in llite: rc = %d\n",
627 dt, ll_get_fsname(sb, NULL, 0), err);
632 if (sbi->ll_md_obd) {
633 err = sysfs_create_link(&sbi->ll_kset.kobj,
634 &sbi->ll_md_obd->obd_kset.kobj,
635 sbi->ll_md_obd->obd_type->typ_name);
637 CERROR("%s: could not register %s in llite: rc = %d\n",
638 md, ll_get_fsname(sb, NULL, 0), err);
648 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
650 obd_disconnect(sbi->ll_dt_exp);
651 sbi->ll_dt_exp = NULL;
652 sbi->ll_dt_obd = NULL;
654 obd_fid_fini(sbi->ll_md_exp->exp_obd);
656 obd_disconnect(sbi->ll_md_exp);
657 sbi->ll_md_exp = NULL;
658 sbi->ll_md_obd = NULL;
667 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
671 size = sizeof(*lmmsize);
672 rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
673 KEY_MAX_EASIZE, &size, lmmsize);
675 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
676 sbi->ll_dt_exp->exp_obd->obd_name, rc);
680 CDEBUG(D_INFO, "max LOV ea size: %d\n", *lmmsize);
683 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
684 KEY_MAX_EASIZE, &size, lmmsize);
686 CERROR("Get max mdsize error rc %d\n", rc);
688 CDEBUG(D_INFO, "max LMV ea size: %d\n", *lmmsize);
694 * Get the value of the default_easize parameter.
696 * \see client_obd::cl_default_mds_easize
698 * \param[in] sbi superblock info for this filesystem
699 * \param[out] lmmsize pointer to storage location for value
701 * \retval 0 on success
702 * \retval negative negated errno on failure
704 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
709 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
710 KEY_DEFAULT_EASIZE, &size, lmmsize);
712 CERROR("Get default mdsize error rc %d\n", rc);
718 * Set the default_easize parameter to the given value.
720 * \see client_obd::cl_default_mds_easize
722 * \param[in] sbi superblock info for this filesystem
723 * \param[in] lmmsize the size to set
725 * \retval 0 on success
726 * \retval negative negated errno on failure
728 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
732 if (lmmsize < sizeof(struct lov_mds_md) ||
733 lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
736 rc = obd_set_info_async(NULL, sbi->ll_md_exp,
737 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
738 sizeof(int), &lmmsize, NULL);
743 static void client_common_put_super(struct super_block *sb)
745 struct ll_sb_info *sbi = ll_s2sbi(sb);
750 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
751 obd_disconnect(sbi->ll_dt_exp);
752 sbi->ll_dt_exp = NULL;
754 ll_debugfs_unregister_super(sb);
756 obd_fid_fini(sbi->ll_md_exp->exp_obd);
757 obd_disconnect(sbi->ll_md_exp);
758 sbi->ll_md_exp = NULL;
763 void ll_kill_super(struct super_block *sb)
765 struct ll_sb_info *sbi;
769 if (!(sb->s_flags & MS_ACTIVE))
773 /* we need restore s_dev from changed for clustred NFS before put_super
774 * because new kernels have cached s_dev and change sb->s_dev in
775 * put_super not affected real removing devices */
777 sb->s_dev = sbi->ll_sdev_orig;
778 sbi->ll_umounting = 1;
780 /* wait running statahead threads to quit */
781 while (atomic_read(&sbi->ll_sa_running) > 0) {
782 set_current_state(TASK_UNINTERRUPTIBLE);
783 schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
790 static inline int ll_set_opt(const char *opt, char *data, int fl)
792 if (strncmp(opt, data, strlen(opt)) != 0)
798 /* non-client-specific mount options are parsed in lmd_parse */
799 static int ll_options(char *options, struct ll_sb_info *sbi)
802 char *s1 = options, *s2;
803 int *flags = &sbi->ll_flags;
809 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
812 CDEBUG(D_SUPER, "next opt=%s\n", s1);
813 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
818 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
823 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
828 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
833 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
838 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
843 tmp = ll_set_opt("context", s1, 1);
846 tmp = ll_set_opt("fscontext", s1, 1);
849 tmp = ll_set_opt("defcontext", s1, 1);
852 tmp = ll_set_opt("rootcontext", s1, 1);
855 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
860 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
866 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
869 sbi->ll_checksum_set = 1;
872 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
875 sbi->ll_checksum_set = 1;
878 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
883 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
888 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
893 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
898 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
903 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
908 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
913 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
918 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
924 s2 = strchr(s1, ',');
932 void ll_lli_init(struct ll_inode_info *lli)
934 lli->lli_inode_magic = LLI_INODE_MAGIC;
936 spin_lock_init(&lli->lli_lock);
937 lli->lli_posix_acl = NULL;
938 /* Do not set lli_fid, it has been initialized already. */
939 fid_zero(&lli->lli_pfid);
940 lli->lli_mds_read_och = NULL;
941 lli->lli_mds_write_och = NULL;
942 lli->lli_mds_exec_och = NULL;
943 lli->lli_open_fd_read_count = 0;
944 lli->lli_open_fd_write_count = 0;
945 lli->lli_open_fd_exec_count = 0;
946 mutex_init(&lli->lli_och_mutex);
947 spin_lock_init(&lli->lli_agl_lock);
948 spin_lock_init(&lli->lli_layout_lock);
949 ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
950 lli->lli_clob = NULL;
952 init_rwsem(&lli->lli_xattrs_list_rwsem);
953 mutex_init(&lli->lli_xattrs_enq_lock);
955 LASSERT(lli->lli_vfs_inode.i_mode != 0);
956 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
957 mutex_init(&lli->lli_readdir_mutex);
958 lli->lli_opendir_key = NULL;
960 spin_lock_init(&lli->lli_sa_lock);
961 lli->lli_opendir_pid = 0;
962 lli->lli_sa_enabled = 0;
963 lli->lli_def_stripe_offset = -1;
964 init_rwsem(&lli->lli_lsm_sem);
966 mutex_init(&lli->lli_size_mutex);
967 lli->lli_symlink_name = NULL;
968 init_rwsem(&lli->lli_trunc_sem);
969 range_lock_tree_init(&lli->lli_write_tree);
970 init_rwsem(&lli->lli_glimpse_sem);
971 lli->lli_glimpse_time = ktime_set(0, 0);
972 INIT_LIST_HEAD(&lli->lli_agl_list);
973 lli->lli_agl_index = 0;
974 lli->lli_async_rc = 0;
975 spin_lock_init(&lli->lli_heat_lock);
976 obd_heat_clear(lli->lli_heat_instances, OBD_HEAT_COUNT);
977 lli->lli_heat_flags = 0;
979 mutex_init(&lli->lli_layout_mutex);
980 memset(lli->lli_jobid, 0, sizeof(lli->lli_jobid));
983 #define MAX_STRING_SIZE 128
985 #ifndef HAVE_SUPER_SETUP_BDI_NAME
987 #define LSI_BDI_INITIALIZED 0x00400000
989 #ifndef HAVE_BDI_CAP_MAP_COPY
990 # define BDI_CAP_MAP_COPY 0
993 static int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
995 struct lustre_sb_info *lsi = s2lsi(sb);
996 char buf[MAX_STRING_SIZE];
1000 err = bdi_init(&lsi->lsi_bdi);
1004 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1005 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1006 lsi->lsi_bdi.name = "lustre";
1007 va_start(args, fmt);
1008 vsnprintf(buf, MAX_STRING_SIZE, fmt, args);
1010 err = bdi_register(&lsi->lsi_bdi, NULL, "%s", buf);
1013 sb->s_bdi = &lsi->lsi_bdi;
1017 #endif /* !HAVE_SUPER_SETUP_BDI_NAME */
1019 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
1021 struct lustre_profile *lprof = NULL;
1022 struct lustre_sb_info *lsi = s2lsi(sb);
1023 struct ll_sb_info *sbi = NULL;
1024 char *dt = NULL, *md = NULL;
1025 char *profilenm = get_profile_name(sb);
1026 struct config_llog_instance *cfg;
1027 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1028 const int instlen = 16 + 2;
1029 unsigned long cfg_instance = ll_get_cfg_instance(sb);
1030 char name[MAX_STRING_SIZE];
1038 /* for ASLR, to map between cfg_instance and hashed ptr */
1039 CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1040 profilenm, cfg_instance, sb);
1042 try_module_get(THIS_MODULE);
1046 GOTO(out_free_cfg, err = -ENOMEM);
1048 /* client additional sb info */
1049 lsi->lsi_llsbi = sbi = ll_init_sbi();
1051 GOTO(out_free_cfg, err = -ENOMEM);
1053 err = ll_options(lsi->lsi_lmd->lmd_opts, sbi);
1055 GOTO(out_free_cfg, err);
1057 err = super_setup_bdi_name(sb, "lustre-%016lx", cfg_instance);
1059 GOTO(out_free_cfg, err);
1061 #ifndef HAVE_DCACHE_LOCK
1062 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1063 sb->s_d_op = &ll_d_ops;
1066 len = strlen(profilenm);
1067 ptr = strrchr(profilenm, '-');
1068 if (ptr && (strcmp(ptr, "-client") == 0))
1072 snprintf(name, MAX_STRING_SIZE, "%.*s-%016lx", len,
1073 profilenm, cfg_instance);
1075 /* Call ll_debugfs_register_super() before lustre_process_log()
1076 * so that "llite.*.*" params can be processed correctly.
1078 err = ll_debugfs_register_super(sb, name);
1080 CERROR("%s: could not register mountpoint in llite: rc = %d\n",
1081 ll_get_fsname(sb, NULL, 0), err);
1085 /* The cfg_instance is a value unique to this super, in case some
1086 * joker tries to mount the same fs at two mount points.
1088 cfg->cfg_instance = cfg_instance;
1089 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1090 cfg->cfg_callback = class_config_llog_handler;
1091 cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1092 /* set up client obds */
1093 err = lustre_process_log(sb, profilenm, cfg);
1095 GOTO(out_debugfs, err);
1097 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1098 lprof = class_get_profile(profilenm);
1099 if (lprof == NULL) {
1100 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1101 " read from the MGS. Does that filesystem "
1102 "exist?\n", profilenm);
1103 GOTO(out_debugfs, err = -EINVAL);
1105 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1106 lprof->lp_md, lprof->lp_dt);
1108 dt_len = strlen(lprof->lp_dt) + instlen + 2;
1109 OBD_ALLOC(dt, dt_len);
1111 GOTO(out_profile, err = -ENOMEM);
1112 snprintf(dt, dt_len - 1, "%s-%016lx", lprof->lp_dt, cfg_instance);
1114 md_len = strlen(lprof->lp_md) + instlen + 2;
1115 OBD_ALLOC(md, md_len);
1117 GOTO(out_free_dt, err = -ENOMEM);
1118 snprintf(md, md_len - 1, "%s-%016lx", lprof->lp_md, cfg_instance);
1120 /* connections, registrations, sb setup */
1121 err = client_common_fill_super(sb, md, dt, mnt);
1123 GOTO(out_free_md, err);
1125 sbi->ll_client_common_fill_super_succeeded = 1;
1129 OBD_FREE(md, md_len);
1132 OBD_FREE(dt, dt_len);
1135 class_put_profile(lprof);
1138 ll_debugfs_unregister_super(sb);
1145 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1146 LCONSOLE_WARN("Mounted %s\n", profilenm);
1148 } /* ll_fill_super */
1150 void ll_put_super(struct super_block *sb)
1152 struct config_llog_instance cfg, params_cfg;
1153 struct obd_device *obd;
1154 struct lustre_sb_info *lsi = s2lsi(sb);
1155 struct ll_sb_info *sbi = ll_s2sbi(sb);
1156 char *profilenm = get_profile_name(sb);
1157 unsigned long cfg_instance = ll_get_cfg_instance(sb);
1159 int next, force = 1, rc = 0;
1163 GOTO(out_no_sbi, 0);
1165 /* Should replace instance_id with something better for ASLR */
1166 CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1167 profilenm, cfg_instance, sb);
1169 cfg.cfg_instance = cfg_instance;
1170 lustre_end_log(sb, profilenm, &cfg);
1172 params_cfg.cfg_instance = cfg_instance;
1173 lustre_end_log(sb, PARAMS_FILENAME, ¶ms_cfg);
1175 if (sbi->ll_md_exp) {
1176 obd = class_exp2obd(sbi->ll_md_exp);
1178 force = obd->obd_force;
1181 /* Wait for unstable pages to be committed to stable storage */
1183 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1184 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1185 atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1189 ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1190 if (force == 0 && rc != -EINTR)
1191 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1193 /* We need to set force before the lov_disconnect in
1194 lustre_common_put_super, since l_d cleans up osc's as well. */
1197 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1199 obd->obd_force = force;
1203 if (sbi->ll_client_common_fill_super_succeeded) {
1204 /* Only if client_common_fill_super succeeded */
1205 client_common_put_super(sb);
1209 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1210 class_manual_cleanup(obd);
1213 if (sbi->ll_flags & LL_SBI_VERBOSE)
1214 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1217 class_del_profile(profilenm);
1219 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1220 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1221 bdi_destroy(&lsi->lsi_bdi);
1222 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1227 lsi->lsi_llsbi = NULL;
1229 lustre_common_put_super(sb);
1231 cl_env_cache_purge(~0);
1233 module_put(THIS_MODULE);
1236 } /* client_put_super */
1238 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1240 struct inode *inode = NULL;
1242 /* NOTE: we depend on atomic igrab() -bzzz */
1243 lock_res_and_lock(lock);
1244 if (lock->l_resource->lr_lvb_inode) {
1245 struct ll_inode_info * lli;
1246 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1247 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1248 inode = igrab(lock->l_resource->lr_lvb_inode);
1250 inode = lock->l_resource->lr_lvb_inode;
1251 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1252 D_WARNING, lock, "lr_lvb_inode %p is "
1253 "bogus: magic %08x",
1254 lock->l_resource->lr_lvb_inode,
1255 lli->lli_inode_magic);
1259 unlock_res_and_lock(lock);
1263 void ll_dir_clear_lsm_md(struct inode *inode)
1265 struct ll_inode_info *lli = ll_i2info(inode);
1267 LASSERT(S_ISDIR(inode->i_mode));
1269 if (lli->lli_lsm_md != NULL) {
1270 lmv_free_memmd(lli->lli_lsm_md);
1271 lli->lli_lsm_md = NULL;
1275 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1276 const struct lu_fid *fid,
1277 struct lustre_md *md)
1279 struct ll_sb_info *sbi = ll_s2sbi(sb);
1280 struct mdt_body *body = md->body;
1281 struct inode *inode;
1285 ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1286 inode = iget_locked(sb, ino);
1287 if (inode == NULL) {
1288 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1289 ll_get_fsname(sb, NULL, 0), PFID(fid));
1290 RETURN(ERR_PTR(-ENOENT));
1293 if (inode->i_state & I_NEW) {
1294 struct ll_inode_info *lli = ll_i2info(inode);
1295 struct lmv_stripe_md *lsm = md->lmv;
1297 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1298 (body->mbo_mode & S_IFMT);
1299 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1302 inode->i_mtime.tv_sec = 0;
1303 inode->i_atime.tv_sec = 0;
1304 inode->i_ctime.tv_sec = 0;
1307 #ifdef HAVE_BACKING_DEV_INFO
1308 /* initializing backing dev info. */
1309 inode->i_mapping->backing_dev_info =
1310 &s2lsi(inode->i_sb)->lsi_bdi;
1312 inode->i_op = &ll_dir_inode_operations;
1313 inode->i_fop = &ll_dir_operations;
1314 lli->lli_fid = *fid;
1317 LASSERT(lsm != NULL);
1318 /* master object FID */
1319 lli->lli_pfid = body->mbo_fid1;
1320 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1321 lli, PFID(fid), PFID(&lli->lli_pfid));
1322 unlock_new_inode(inode);
1328 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1331 struct lmv_stripe_md *lsm = md->lmv;
1332 struct ll_inode_info *lli = ll_i2info(inode);
1335 LASSERT(lsm != NULL);
1337 CDEBUG(D_INODE, "%s: "DFID" set dir layout:\n",
1338 ll_get_fsname(inode->i_sb, NULL, 0),
1339 PFID(&lli->lli_fid));
1340 lsm_md_dump(D_INODE, lsm);
1342 /* XXX sigh, this lsm_root initialization should be in
1343 * LMV layer, but it needs ll_iget right now, so we
1344 * put this here right now. */
1345 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1346 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1347 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1348 /* Unfortunately ll_iget will call ll_update_inode,
1349 * where the initialization of slave inode is slightly
1350 * different, so it reset lsm_md to NULL to avoid
1351 * initializing lsm for slave inode. */
1352 lsm->lsm_md_oinfo[i].lmo_root =
1353 ll_iget_anon_dir(inode->i_sb, fid, md);
1354 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1355 int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1357 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1359 iput(lsm->lsm_md_oinfo[i].lmo_root);
1360 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1366 lli->lli_lsm_md = lsm;
1371 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1373 struct ll_inode_info *lli = ll_i2info(inode);
1374 struct lmv_stripe_md *lsm = md->lmv;
1379 LASSERT(S_ISDIR(inode->i_mode));
1380 CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1381 PFID(ll_inode2fid(inode)));
1384 * no striped information from request, lustre_md from req does not
1385 * include stripeEA, see ll_md_setattr()
1391 * normally dir layout doesn't change, only take read lock to check
1392 * that to avoid blocking other MD operations.
1394 if (lli->lli_lsm_md)
1395 down_read(&lli->lli_lsm_sem);
1397 down_write(&lli->lli_lsm_sem);
1400 * if dir layout mismatch, check whether version is increased, which
1401 * means layout is changed, this happens in dir migration and lfsck.
1403 if (lli->lli_lsm_md && !lsm_md_eq(lli->lli_lsm_md, lsm)) {
1404 if (lsm->lsm_md_layout_version <=
1405 lli->lli_lsm_md->lsm_md_layout_version) {
1406 CERROR("%s: "DFID" dir layout mismatch:\n",
1407 ll_get_fsname(inode->i_sb, NULL, 0),
1408 PFID(&lli->lli_fid));
1409 lsm_md_dump(D_ERROR, lli->lli_lsm_md);
1410 lsm_md_dump(D_ERROR, lsm);
1411 GOTO(unlock, rc = -EINVAL);
1414 /* layout changed, switch to write lock */
1415 up_read(&lli->lli_lsm_sem);
1416 down_write(&lli->lli_lsm_sem);
1417 ll_dir_clear_lsm_md(inode);
1420 /* set directory layout */
1421 if (!lli->lli_lsm_md) {
1422 struct cl_attr *attr;
1424 rc = ll_init_lsm_md(inode, md);
1425 up_write(&lli->lli_lsm_sem);
1429 /* set md->lmv to NULL, so the following free lustre_md
1430 * will not free this lsm */
1434 * md_merge_attr() may take long, since lsm is already set,
1435 * switch to read lock.
1437 down_read(&lli->lli_lsm_sem);
1439 OBD_ALLOC_PTR(attr);
1441 GOTO(unlock, rc = -ENOMEM);
1443 /* validate the lsm */
1444 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1445 ll_md_blocking_ast);
1451 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1452 md->body->mbo_nlink = attr->cat_nlink;
1453 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1454 md->body->mbo_size = attr->cat_size;
1455 if (md->body->mbo_valid & OBD_MD_FLATIME)
1456 md->body->mbo_atime = attr->cat_atime;
1457 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1458 md->body->mbo_ctime = attr->cat_ctime;
1459 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1460 md->body->mbo_mtime = attr->cat_mtime;
1465 up_read(&lli->lli_lsm_sem);
1470 void ll_clear_inode(struct inode *inode)
1472 struct ll_inode_info *lli = ll_i2info(inode);
1473 struct ll_sb_info *sbi = ll_i2sbi(inode);
1476 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1477 PFID(ll_inode2fid(inode)), inode);
1479 if (S_ISDIR(inode->i_mode)) {
1480 /* these should have been cleared in ll_file_release */
1481 LASSERT(lli->lli_opendir_key == NULL);
1482 LASSERT(lli->lli_sai == NULL);
1483 LASSERT(lli->lli_opendir_pid == 0);
1486 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1488 LASSERT(!lli->lli_open_fd_write_count);
1489 LASSERT(!lli->lli_open_fd_read_count);
1490 LASSERT(!lli->lli_open_fd_exec_count);
1492 if (lli->lli_mds_write_och)
1493 ll_md_real_close(inode, FMODE_WRITE);
1494 if (lli->lli_mds_exec_och)
1495 ll_md_real_close(inode, FMODE_EXEC);
1496 if (lli->lli_mds_read_och)
1497 ll_md_real_close(inode, FMODE_READ);
1499 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1500 OBD_FREE(lli->lli_symlink_name,
1501 strlen(lli->lli_symlink_name) + 1);
1502 lli->lli_symlink_name = NULL;
1505 ll_xattr_cache_destroy(inode);
1507 #ifdef CONFIG_FS_POSIX_ACL
1508 forget_all_cached_acls(inode);
1509 if (lli->lli_posix_acl) {
1510 posix_acl_release(lli->lli_posix_acl);
1511 lli->lli_posix_acl = NULL;
1514 lli->lli_inode_magic = LLI_INODE_DEAD;
1516 if (S_ISDIR(inode->i_mode))
1517 ll_dir_clear_lsm_md(inode);
1518 else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1519 LASSERT(list_empty(&lli->lli_agl_list));
1522 * XXX This has to be done before lsm is freed below, because
1523 * cl_object still uses inode lsm.
1525 cl_inode_fini(inode);
1530 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1532 struct lustre_md md;
1533 struct inode *inode = dentry->d_inode;
1534 struct ll_sb_info *sbi = ll_i2sbi(inode);
1535 struct ptlrpc_request *request = NULL;
1539 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1540 LUSTRE_OPC_ANY, NULL);
1541 if (IS_ERR(op_data))
1542 RETURN(PTR_ERR(op_data));
1544 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1546 ptlrpc_req_finished(request);
1547 if (rc == -ENOENT) {
1549 /* Unlinked special device node? Or just a race?
1550 * Pretend we done everything. */
1551 if (!S_ISREG(inode->i_mode) &&
1552 !S_ISDIR(inode->i_mode)) {
1553 ia_valid = op_data->op_attr.ia_valid;
1554 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1555 rc = simple_setattr(dentry, &op_data->op_attr);
1556 op_data->op_attr.ia_valid = ia_valid;
1558 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1559 CERROR("md_setattr fails: rc = %d\n", rc);
1564 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1565 sbi->ll_md_exp, &md);
1567 ptlrpc_req_finished(request);
1571 ia_valid = op_data->op_attr.ia_valid;
1572 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1573 * cache is not cleared yet. */
1574 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1575 if (S_ISREG(inode->i_mode))
1577 rc = simple_setattr(dentry, &op_data->op_attr);
1578 if (S_ISREG(inode->i_mode))
1579 inode_unlock(inode);
1580 op_data->op_attr.ia_valid = ia_valid;
1582 rc = ll_update_inode(inode, &md);
1583 ptlrpc_req_finished(request);
1588 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1589 * object(s) determine the file size and mtime. Otherwise, the MDS will
1590 * keep these values until such a time that objects are allocated for it.
1591 * We do the MDS operations first, as it is checking permissions for us.
1592 * We don't to the MDS RPC if there is nothing that we want to store there,
1593 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1594 * going to do an RPC anyways.
1596 * If we are doing a truncate, we will send the mtime and ctime updates
1597 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1598 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1601 * In case of HSMimport, we only set attr on MDS.
1603 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr,
1604 enum op_xvalid xvalid, bool hsm_import)
1606 struct inode *inode = dentry->d_inode;
1607 struct ll_inode_info *lli = ll_i2info(inode);
1608 struct md_op_data *op_data = NULL;
1612 CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1613 "valid %x, hsm_import %d\n",
1614 ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1615 inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1618 if (attr->ia_valid & ATTR_SIZE) {
1619 /* Check new size against VFS/VM file size limit and rlimit */
1620 rc = inode_newsize_ok(inode, attr->ia_size);
1624 /* The maximum Lustre file size is variable, based on the
1625 * OST maximum object size and number of stripes. This
1626 * needs another check in addition to the VFS check above. */
1627 if (attr->ia_size > ll_file_maxbytes(inode)) {
1628 CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
1629 PFID(&lli->lli_fid), attr->ia_size,
1630 ll_file_maxbytes(inode));
1634 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1637 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1638 if (attr->ia_valid & TIMES_SET_FLAGS) {
1639 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1640 !cfs_capable(CFS_CAP_FOWNER))
1644 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1645 if (!(xvalid & OP_XVALID_CTIME_SET) &&
1646 (attr->ia_valid & ATTR_CTIME)) {
1647 attr->ia_ctime = current_time(inode);
1648 xvalid |= OP_XVALID_CTIME_SET;
1650 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1651 (attr->ia_valid & ATTR_ATIME)) {
1652 attr->ia_atime = current_time(inode);
1653 attr->ia_valid |= ATTR_ATIME_SET;
1655 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1656 (attr->ia_valid & ATTR_MTIME)) {
1657 attr->ia_mtime = current_time(inode);
1658 attr->ia_valid |= ATTR_MTIME_SET;
1661 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1662 CDEBUG(D_INODE, "setting mtime %lld, ctime %lld, now = %lld\n",
1663 (s64)attr->ia_mtime.tv_sec, (s64)attr->ia_ctime.tv_sec,
1664 ktime_get_real_seconds());
1666 if (S_ISREG(inode->i_mode)) {
1667 if (attr->ia_valid & ATTR_SIZE)
1668 inode_dio_write_done(inode);
1669 inode_unlock(inode);
1672 /* We always do an MDS RPC, even if we're only changing the size;
1673 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1675 OBD_ALLOC_PTR(op_data);
1676 if (op_data == NULL)
1677 GOTO(out, rc = -ENOMEM);
1679 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1680 /* If we are changing file size, file content is
1681 * modified, flag it.
1683 xvalid |= OP_XVALID_OWNEROVERRIDE;
1684 op_data->op_bias |= MDS_DATA_MODIFIED;
1685 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1688 if (attr->ia_valid & ATTR_FILE) {
1689 struct ll_file_data *fd = LUSTRE_FPRIVATE(attr->ia_file);
1691 if (fd->fd_lease_och)
1692 op_data->op_bias |= MDS_TRUNC_KEEP_LEASE;
1695 op_data->op_attr = *attr;
1696 op_data->op_xvalid = xvalid;
1698 rc = ll_md_setattr(dentry, op_data);
1702 if (!S_ISREG(inode->i_mode) || hsm_import)
1705 if (attr->ia_valid & (ATTR_SIZE | ATTR_ATIME | ATTR_ATIME_SET |
1706 ATTR_MTIME | ATTR_MTIME_SET | ATTR_CTIME) ||
1707 xvalid & OP_XVALID_CTIME_SET) {
1708 /* For truncate and utimes sending attributes to OSTs, setting
1709 * mtime/atime to the past will be performed under PW [0:EOF]
1710 * extent lock (new_size:EOF for truncate). It may seem
1711 * excessive to send mtime/atime updates to OSTs when not
1712 * setting times to past, but it is necessary due to possible
1713 * time de-synchronization between MDT inode and OST objects
1715 rc = cl_setattr_ost(lli->lli_clob, attr, xvalid, 0);
1718 /* If the file was restored, it needs to set dirty flag.
1720 * We've already sent MDS_DATA_MODIFIED flag in
1721 * ll_md_setattr() for truncate. However, the MDT refuses to
1722 * set the HS_DIRTY flag on released files, so we have to set
1723 * it again if the file has been restored. Please check how
1724 * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1726 * Please notice that if the file is not released, the previous
1727 * MDS_DATA_MODIFIED has taken effect and usually
1728 * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1729 * This way we can save an RPC for common open + trunc
1731 if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1732 struct hsm_state_set hss = {
1733 .hss_valid = HSS_SETMASK,
1734 .hss_setmask = HS_DIRTY,
1738 rc2 = ll_hsm_state_set(inode, &hss);
1739 /* truncate and write can happen at the same time, so that
1740 * the file can be set modified even though the file is not
1741 * restored from released state, and ll_hsm_state_set() is
1742 * not applicable for the file, and rc2 < 0 is normal in this
1745 CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1746 PFID(ll_inode2fid(inode)), rc2);
1751 if (op_data != NULL)
1752 ll_finish_md_op_data(op_data);
1754 if (S_ISREG(inode->i_mode)) {
1756 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1757 inode_dio_wait(inode);
1758 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
1759 * flag. ll_update_inode (called from ll_md_setattr), clears
1760 * inode flags, so there is a gap where S_NOSEC is not set.
1761 * This can cause a writer to take the i_mutex unnecessarily,
1762 * but this is safe to do and should be rare. */
1763 inode_has_no_xattr(inode);
1766 ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1767 LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1772 int ll_setattr(struct dentry *de, struct iattr *attr)
1774 int mode = de->d_inode->i_mode;
1775 enum op_xvalid xvalid = 0;
1777 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1778 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1779 xvalid |= OP_XVALID_OWNEROVERRIDE;
1781 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1782 (ATTR_SIZE|ATTR_MODE)) &&
1783 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1784 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1785 !(attr->ia_mode & S_ISGID))))
1786 attr->ia_valid |= ATTR_FORCE;
1788 if ((attr->ia_valid & ATTR_MODE) &&
1790 !(attr->ia_mode & S_ISUID) &&
1791 !(attr->ia_valid & ATTR_KILL_SUID))
1792 attr->ia_valid |= ATTR_KILL_SUID;
1794 if ((attr->ia_valid & ATTR_MODE) &&
1795 ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1796 !(attr->ia_mode & S_ISGID) &&
1797 !(attr->ia_valid & ATTR_KILL_SGID))
1798 attr->ia_valid |= ATTR_KILL_SGID;
1800 return ll_setattr_raw(de, attr, xvalid, false);
1803 int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
1806 struct obd_statfs obd_osfs = { 0 };
1811 max_age = ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS;
1813 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1817 osfs->os_type = LL_SUPER_MAGIC;
1819 CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1820 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree, osfs->os_files);
1822 if (osfs->os_state & OS_STATE_SUM)
1825 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1826 flags |= OBD_STATFS_NODELAY;
1828 rc = obd_statfs(NULL, sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1829 if (rc) /* Possibly a filesystem with no OSTs. Report MDT totals. */
1832 CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1833 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1836 osfs->os_bsize = obd_osfs.os_bsize;
1837 osfs->os_blocks = obd_osfs.os_blocks;
1838 osfs->os_bfree = obd_osfs.os_bfree;
1839 osfs->os_bavail = obd_osfs.os_bavail;
1841 /* If we have _some_ OSTs, but don't have as many free objects on the
1842 * OSTs as inodes on the MDTs, reduce the reported number of inodes
1843 * to compensate, so that the "inodes in use" number is correct.
1844 * This should be kept in sync with lod_statfs() behaviour.
1846 if (obd_osfs.os_files && 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 > inode->i_atime.tv_sec)
1955 inode->i_atime.tv_sec = body->mbo_atime;
1956 lli->lli_atime = body->mbo_atime;
1959 if (body->mbo_valid & OBD_MD_FLMTIME) {
1960 if (body->mbo_mtime > inode->i_mtime.tv_sec) {
1962 "setting ino %lu mtime from %lld to %llu\n",
1963 inode->i_ino, (s64)inode->i_mtime.tv_sec,
1965 inode->i_mtime.tv_sec = body->mbo_mtime;
1967 lli->lli_mtime = body->mbo_mtime;
1970 if (body->mbo_valid & OBD_MD_FLCTIME) {
1971 if (body->mbo_ctime > inode->i_ctime.tv_sec)
1972 inode->i_ctime.tv_sec = body->mbo_ctime;
1973 lli->lli_ctime = body->mbo_ctime;
1976 /* Clear i_flags to remove S_NOSEC before permissions are updated */
1977 if (body->mbo_valid & OBD_MD_FLFLAGS)
1978 ll_update_inode_flags(inode, body->mbo_flags);
1979 if (body->mbo_valid & OBD_MD_FLMODE)
1980 inode->i_mode = (inode->i_mode & S_IFMT) |
1981 (body->mbo_mode & ~S_IFMT);
1983 if (body->mbo_valid & OBD_MD_FLTYPE)
1984 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1985 (body->mbo_mode & S_IFMT);
1987 LASSERT(inode->i_mode != 0);
1988 if (S_ISREG(inode->i_mode))
1989 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1990 LL_MAX_BLKSIZE_BITS);
1992 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1994 if (body->mbo_valid & OBD_MD_FLUID)
1995 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
1996 if (body->mbo_valid & OBD_MD_FLGID)
1997 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
1998 if (body->mbo_valid & OBD_MD_FLPROJID)
1999 lli->lli_projid = body->mbo_projid;
2000 if (body->mbo_valid & OBD_MD_FLNLINK)
2001 set_nlink(inode, body->mbo_nlink);
2002 if (body->mbo_valid & OBD_MD_FLRDEV)
2003 inode->i_rdev = old_decode_dev(body->mbo_rdev);
2005 if (body->mbo_valid & OBD_MD_FLID) {
2006 /* FID shouldn't be changed! */
2007 if (fid_is_sane(&lli->lli_fid)) {
2008 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
2009 "Trying to change FID "DFID
2010 " to the "DFID", inode "DFID"(%p)\n",
2011 PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
2012 PFID(ll_inode2fid(inode)), inode);
2014 lli->lli_fid = body->mbo_fid1;
2018 LASSERT(fid_seq(&lli->lli_fid) != 0);
2020 if (body->mbo_valid & OBD_MD_FLSIZE) {
2021 i_size_write(inode, body->mbo_size);
2023 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
2024 PFID(ll_inode2fid(inode)),
2025 (unsigned long long)body->mbo_size);
2027 if (body->mbo_valid & OBD_MD_FLBLOCKS)
2028 inode->i_blocks = body->mbo_blocks;
2031 if (body->mbo_valid & OBD_MD_TSTATE) {
2032 /* Set LLIF_FILE_RESTORING if restore ongoing and
2033 * clear it when done to ensure to start again
2034 * glimpsing updated attrs
2036 if (body->mbo_t_state & MS_RESTORE)
2037 ll_file_set_flag(lli, LLIF_FILE_RESTORING);
2039 ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
2045 int ll_read_inode2(struct inode *inode, void *opaque)
2047 struct lustre_md *md = opaque;
2048 struct ll_inode_info *lli = ll_i2info(inode);
2052 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2053 PFID(&lli->lli_fid), inode);
2055 /* Core attributes from the MDS first. This is a new inode, and
2056 * the VFS doesn't zero times in the core inode so we have to do
2057 * it ourselves. They will be overwritten by either MDS or OST
2058 * attributes - we just need to make sure they aren't newer.
2060 inode->i_mtime.tv_sec = 0;
2061 inode->i_atime.tv_sec = 0;
2062 inode->i_ctime.tv_sec = 0;
2064 rc = ll_update_inode(inode, md);
2068 /* OIDEBUG(inode); */
2070 #ifdef HAVE_BACKING_DEV_INFO
2071 /* initializing backing dev info. */
2072 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2074 if (S_ISREG(inode->i_mode)) {
2075 struct ll_sb_info *sbi = ll_i2sbi(inode);
2076 inode->i_op = &ll_file_inode_operations;
2077 inode->i_fop = sbi->ll_fop;
2078 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2080 } else if (S_ISDIR(inode->i_mode)) {
2081 inode->i_op = &ll_dir_inode_operations;
2082 inode->i_fop = &ll_dir_operations;
2084 } else if (S_ISLNK(inode->i_mode)) {
2085 inode->i_op = &ll_fast_symlink_inode_operations;
2088 inode->i_op = &ll_special_inode_operations;
2090 init_special_inode(inode, inode->i_mode,
2099 void ll_delete_inode(struct inode *inode)
2101 struct ll_inode_info *lli = ll_i2info(inode);
2102 struct address_space *mapping = &inode->i_data;
2103 unsigned long nrpages;
2106 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2107 /* It is last chance to write out dirty pages,
2108 * otherwise we may lose data while umount */
2109 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
2111 truncate_inode_pages_final(mapping);
2113 /* Workaround for LU-118: Note nrpages may not be totally updated when
2114 * truncate_inode_pages() returns, as there can be a page in the process
2115 * of deletion (inside __delete_from_page_cache()) in the specified
2116 * range. Thus mapping->nrpages can be non-zero when this function
2117 * returns even after truncation of the whole mapping. Only do this if
2118 * npages isn't already zero.
2120 nrpages = mapping->nrpages;
2122 spin_lock_irq(&mapping->tree_lock);
2123 nrpages = mapping->nrpages;
2124 spin_unlock_irq(&mapping->tree_lock);
2125 } /* Workaround end */
2127 LASSERTF(nrpages == 0, "%s: inode="DFID"(%p) nrpages=%lu, "
2128 "see https://jira.whamcloud.com/browse/LU-118\n",
2129 ll_get_fsname(inode->i_sb, NULL, 0),
2130 PFID(ll_inode2fid(inode)), inode, nrpages);
2132 #ifdef HAVE_SBOPS_EVICT_INODE
2133 ll_clear_inode(inode);
2140 int ll_iocontrol(struct inode *inode, struct file *file,
2141 unsigned int cmd, unsigned long arg)
2143 struct ll_sb_info *sbi = ll_i2sbi(inode);
2144 struct ptlrpc_request *req = NULL;
2149 case FS_IOC_GETFLAGS: {
2150 struct mdt_body *body;
2151 struct md_op_data *op_data;
2153 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2154 0, 0, LUSTRE_OPC_ANY,
2156 if (IS_ERR(op_data))
2157 RETURN(PTR_ERR(op_data));
2159 op_data->op_valid = OBD_MD_FLFLAGS;
2160 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2161 ll_finish_md_op_data(op_data);
2163 CERROR("%s: failure inode "DFID": rc = %d\n",
2164 sbi->ll_md_exp->exp_obd->obd_name,
2165 PFID(ll_inode2fid(inode)), rc);
2169 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2171 flags = body->mbo_flags;
2173 ptlrpc_req_finished(req);
2175 RETURN(put_user(flags, (int __user *)arg));
2177 case FS_IOC_SETFLAGS: {
2179 struct md_op_data *op_data;
2180 struct cl_object *obj;
2181 struct fsxattr fa = { 0 };
2183 if (get_user(flags, (int __user *)arg))
2186 fa.fsx_projid = ll_i2info(inode)->lli_projid;
2187 if (flags & LUSTRE_PROJINHERIT_FL)
2188 fa.fsx_xflags = FS_XFLAG_PROJINHERIT;
2190 rc = ll_ioctl_check_project(inode, &fa);
2194 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2195 LUSTRE_OPC_ANY, NULL);
2196 if (IS_ERR(op_data))
2197 RETURN(PTR_ERR(op_data));
2199 op_data->op_attr_flags = flags;
2200 op_data->op_xvalid |= OP_XVALID_FLAGS;
2201 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2202 ll_finish_md_op_data(op_data);
2203 ptlrpc_req_finished(req);
2207 ll_update_inode_flags(inode, flags);
2209 obj = ll_i2info(inode)->lli_clob;
2213 OBD_ALLOC_PTR(attr);
2217 rc = cl_setattr_ost(obj, attr, OP_XVALID_FLAGS, flags);
2229 int ll_flush_ctx(struct inode *inode)
2231 struct ll_sb_info *sbi = ll_i2sbi(inode);
2233 CDEBUG(D_SEC, "flush context for user %d\n",
2234 from_kuid(&init_user_ns, current_uid()));
2236 obd_set_info_async(NULL, sbi->ll_md_exp,
2237 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2239 obd_set_info_async(NULL, sbi->ll_dt_exp,
2240 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2245 /* umount -f client means force down, don't save state */
2246 void ll_umount_begin(struct super_block *sb)
2248 struct ll_sb_info *sbi = ll_s2sbi(sb);
2249 struct obd_device *obd;
2250 struct obd_ioctl_data *ioc_data;
2251 struct l_wait_info lwi;
2252 wait_queue_head_t waitq;
2255 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2256 sb->s_count, atomic_read(&sb->s_active));
2258 obd = class_exp2obd(sbi->ll_md_exp);
2260 CERROR("Invalid MDC connection handle %#llx\n",
2261 sbi->ll_md_exp->exp_handle.h_cookie);
2267 obd = class_exp2obd(sbi->ll_dt_exp);
2269 CERROR("Invalid LOV connection handle %#llx\n",
2270 sbi->ll_dt_exp->exp_handle.h_cookie);
2276 OBD_ALLOC_PTR(ioc_data);
2278 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2279 sizeof *ioc_data, ioc_data, NULL);
2281 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2282 sizeof *ioc_data, ioc_data, NULL);
2284 OBD_FREE_PTR(ioc_data);
2287 /* Really, we'd like to wait until there are no requests outstanding,
2288 * and then continue. For now, we just periodically checking for vfs
2289 * to decrement mnt_cnt and hope to finish it within 10sec.
2291 init_waitqueue_head(&waitq);
2292 lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(10),
2293 cfs_time_seconds(1), NULL, NULL);
2294 l_wait_event(waitq, may_umount(sbi->ll_mnt.mnt), &lwi);
2299 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2301 struct ll_sb_info *sbi = ll_s2sbi(sb);
2302 char *profilenm = get_profile_name(sb);
2306 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2307 read_only = *flags & MS_RDONLY;
2308 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2309 sizeof(KEY_READ_ONLY),
2310 KEY_READ_ONLY, sizeof(read_only),
2313 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2314 profilenm, read_only ?
2315 "read-only" : "read-write", err);
2320 sb->s_flags |= MS_RDONLY;
2322 sb->s_flags &= ~MS_RDONLY;
2324 if (sbi->ll_flags & LL_SBI_VERBOSE)
2325 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2326 read_only ? "read-only" : "read-write");
2332 * Cleanup the open handle that is cached on MDT-side.
2334 * For open case, the client side open handling thread may hit error
2335 * after the MDT grant the open. Under such case, the client should
2336 * send close RPC to the MDT as cleanup; otherwise, the open handle
2337 * on the MDT will be leaked there until the client umount or evicted.
2339 * In further, if someone unlinked the file, because the open handle
2340 * holds the reference on such file/object, then it will block the
2341 * subsequent threads that want to locate such object via FID.
2343 * \param[in] sb super block for this file-system
2344 * \param[in] open_req pointer to the original open request
2346 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2348 struct mdt_body *body;
2349 struct md_op_data *op_data;
2350 struct ptlrpc_request *close_req = NULL;
2351 struct obd_export *exp = ll_s2sbi(sb)->ll_md_exp;
2354 body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2355 OBD_ALLOC_PTR(op_data);
2356 if (op_data == NULL) {
2357 CWARN("%s: cannot allocate op_data to release open handle for "
2359 ll_get_fsname(sb, NULL, 0), PFID(&body->mbo_fid1));
2364 op_data->op_fid1 = body->mbo_fid1;
2365 op_data->op_open_handle = body->mbo_open_handle;
2366 op_data->op_mod_time = ktime_get_real_seconds();
2367 md_close(exp, op_data, NULL, &close_req);
2368 ptlrpc_req_finished(close_req);
2369 ll_finish_md_op_data(op_data);
2374 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2375 struct super_block *sb, struct lookup_intent *it)
2377 struct ll_sb_info *sbi = NULL;
2378 struct lustre_md md = { NULL };
2382 LASSERT(*inode || sb);
2383 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2384 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2385 sbi->ll_md_exp, &md);
2390 rc = ll_update_inode(*inode, &md);
2394 LASSERT(sb != NULL);
2397 * At this point server returns to client's same fid as client
2398 * generated for creating. So using ->fid1 is okay here.
2400 if (!fid_is_sane(&md.body->mbo_fid1)) {
2401 CERROR("%s: Fid is insane "DFID"\n",
2402 ll_get_fsname(sb, NULL, 0),
2403 PFID(&md.body->mbo_fid1));
2404 GOTO(out, rc = -EINVAL);
2407 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2408 sbi->ll_flags & LL_SBI_32BIT_API),
2410 if (IS_ERR(*inode)) {
2411 #ifdef CONFIG_FS_POSIX_ACL
2413 posix_acl_release(md.posix_acl);
2414 md.posix_acl = NULL;
2417 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2419 CERROR("new_inode -fatal: rc %d\n", rc);
2424 /* Handling piggyback layout lock.
2425 * Layout lock can be piggybacked by getattr and open request.
2426 * The lsm can be applied to inode only if it comes with a layout lock
2427 * otherwise correct layout may be overwritten, for example:
2428 * 1. proc1: mdt returns a lsm but not granting layout
2429 * 2. layout was changed by another client
2430 * 3. proc2: refresh layout and layout lock granted
2431 * 4. proc1: to apply a stale layout */
2432 if (it != NULL && it->it_lock_mode != 0) {
2433 struct lustre_handle lockh;
2434 struct ldlm_lock *lock;
2436 lockh.cookie = it->it_lock_handle;
2437 lock = ldlm_handle2lock(&lockh);
2438 LASSERT(lock != NULL);
2439 if (ldlm_has_layout(lock)) {
2440 struct cl_object_conf conf;
2442 memset(&conf, 0, sizeof(conf));
2443 conf.coc_opc = OBJECT_CONF_SET;
2444 conf.coc_inode = *inode;
2445 conf.coc_lock = lock;
2446 conf.u.coc_layout = md.layout;
2447 (void)ll_layout_conf(*inode, &conf);
2449 LDLM_LOCK_PUT(lock);
2455 md_free_lustre_md(sbi->ll_md_exp, &md);
2458 if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2459 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2464 int ll_obd_statfs(struct inode *inode, void __user *arg)
2466 struct ll_sb_info *sbi = NULL;
2467 struct obd_export *exp;
2469 struct obd_ioctl_data *data = NULL;
2473 if (!inode || !(sbi = ll_i2sbi(inode)))
2474 GOTO(out_statfs, rc = -EINVAL);
2476 rc = obd_ioctl_getdata(&buf, &len, arg);
2478 GOTO(out_statfs, rc);
2481 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2482 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2483 GOTO(out_statfs, rc = -EINVAL);
2485 if (data->ioc_inllen1 != sizeof(__u32) ||
2486 data->ioc_inllen2 != sizeof(__u32) ||
2487 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2488 data->ioc_plen2 != sizeof(struct obd_uuid))
2489 GOTO(out_statfs, rc = -EINVAL);
2491 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2492 if (type & LL_STATFS_LMV)
2493 exp = sbi->ll_md_exp;
2494 else if (type & LL_STATFS_LOV)
2495 exp = sbi->ll_dt_exp;
2497 GOTO(out_statfs, rc = -ENODEV);
2499 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2501 GOTO(out_statfs, rc);
2503 OBD_FREE_LARGE(buf, len);
2508 * this is normally called in ll_fini_md_op_data(), but sometimes it needs to
2509 * be called early to avoid deadlock.
2511 void ll_unlock_md_op_lsm(struct md_op_data *op_data)
2513 if (op_data->op_mea2_sem) {
2514 up_read(op_data->op_mea2_sem);
2515 op_data->op_mea2_sem = NULL;
2518 if (op_data->op_mea1_sem) {
2519 up_read(op_data->op_mea1_sem);
2520 op_data->op_mea1_sem = NULL;
2524 /* this function prepares md_op_data hint for passing it down to MD stack. */
2525 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2526 struct inode *i1, struct inode *i2,
2527 const char *name, size_t namelen,
2528 __u32 mode, __u32 opc, void *data)
2530 LASSERT(i1 != NULL);
2533 /* Do not reuse namelen for something else. */
2535 return ERR_PTR(-EINVAL);
2537 if (namelen > ll_i2sbi(i1)->ll_namelen)
2538 return ERR_PTR(-ENAMETOOLONG);
2540 if (!lu_name_is_valid_2(name, namelen))
2541 return ERR_PTR(-EINVAL);
2544 if (op_data == NULL)
2545 OBD_ALLOC_PTR(op_data);
2547 if (op_data == NULL)
2548 return ERR_PTR(-ENOMEM);
2550 ll_i2gids(op_data->op_suppgids, i1, i2);
2551 op_data->op_fid1 = *ll_inode2fid(i1);
2552 op_data->op_default_stripe_offset = -1;
2554 if (S_ISDIR(i1->i_mode)) {
2555 down_read(&ll_i2info(i1)->lli_lsm_sem);
2556 op_data->op_mea1_sem = &ll_i2info(i1)->lli_lsm_sem;
2557 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2558 if (opc == LUSTRE_OPC_MKDIR)
2559 op_data->op_default_stripe_offset =
2560 ll_i2info(i1)->lli_def_stripe_offset;
2564 op_data->op_fid2 = *ll_inode2fid(i2);
2565 if (S_ISDIR(i2->i_mode)) {
2567 down_read(&ll_i2info(i2)->lli_lsm_sem);
2568 op_data->op_mea2_sem =
2569 &ll_i2info(i2)->lli_lsm_sem;
2571 op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2574 fid_zero(&op_data->op_fid2);
2577 if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2578 op_data->op_cli_flags |= CLI_HASH64;
2580 if (ll_need_32bit_api(ll_i2sbi(i1)))
2581 op_data->op_cli_flags |= CLI_API32;
2583 op_data->op_name = name;
2584 op_data->op_namelen = namelen;
2585 op_data->op_mode = mode;
2586 op_data->op_mod_time = ktime_get_real_seconds();
2587 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2588 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2589 op_data->op_cap = cfs_curproc_cap_pack();
2590 op_data->op_mds = 0;
2591 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2592 filename_is_volatile(name, namelen, &op_data->op_mds)) {
2593 op_data->op_bias |= MDS_CREATE_VOLATILE;
2595 op_data->op_data = data;
2600 void ll_finish_md_op_data(struct md_op_data *op_data)
2602 ll_unlock_md_op_lsm(op_data);
2603 security_release_secctx(op_data->op_file_secctx,
2604 op_data->op_file_secctx_size);
2605 OBD_FREE_PTR(op_data);
2608 #ifdef HAVE_SUPEROPS_USE_DENTRY
2609 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2611 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2614 struct ll_sb_info *sbi;
2616 #ifdef HAVE_SUPEROPS_USE_DENTRY
2617 LASSERT((seq != NULL) && (dentry != NULL));
2618 sbi = ll_s2sbi(dentry->d_sb);
2620 LASSERT((seq != NULL) && (vfs != NULL));
2621 sbi = ll_s2sbi(vfs->mnt_sb);
2624 if (sbi->ll_flags & LL_SBI_NOLCK)
2625 seq_puts(seq, ",nolock");
2627 if (sbi->ll_flags & LL_SBI_FLOCK)
2628 seq_puts(seq, ",flock");
2630 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2631 seq_puts(seq, ",localflock");
2633 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2634 seq_puts(seq, ",user_xattr");
2636 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2637 seq_puts(seq, ",lazystatfs");
2639 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2640 seq_puts(seq, ",user_fid2path");
2642 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2643 seq_puts(seq, ",always_ping");
2649 * Get obd name by cmd, and copy out to user space
2651 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2653 struct ll_sb_info *sbi = ll_i2sbi(inode);
2654 struct obd_device *obd;
2657 if (cmd == OBD_IOC_GETDTNAME)
2658 obd = class_exp2obd(sbi->ll_dt_exp);
2659 else if (cmd == OBD_IOC_GETMDNAME)
2660 obd = class_exp2obd(sbi->ll_md_exp);
2667 if (copy_to_user((void __user *)arg, obd->obd_name,
2668 strlen(obd->obd_name) + 1))
2675 * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2676 * fsname will be returned in this buffer; otherwise, a static buffer will be
2677 * used to store the fsname and returned to caller.
2679 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2681 static char fsname_static[MTI_NAME_MAXLEN];
2682 struct lustre_sb_info *lsi = s2lsi(sb);
2687 /* this means the caller wants to use static buffer
2688 * and it doesn't care about race. Usually this is
2689 * in error reporting path */
2690 buf = fsname_static;
2691 buflen = sizeof(fsname_static);
2694 len = strlen(lsi->lsi_lmd->lmd_profile);
2695 ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2696 if (ptr && (strcmp(ptr, "-client") == 0))
2699 if (unlikely(len >= buflen))
2701 strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
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_get_fsname(page->mapping->host->i_sb, NULL, 0),
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));