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 |
226 OBD_CONNECT2_OVERSTRIPING |
228 OBD_CONNECT2_LOCK_CONVERT |
229 OBD_CONNECT2_ARCHIVE_ID_ARRAY |
231 OBD_CONNECT2_ASYNC_DISCARD;
233 #ifdef HAVE_LRU_RESIZE_SUPPORT
234 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
235 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
237 #ifdef CONFIG_FS_POSIX_ACL
238 data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK |
239 OBD_CONNECT_LARGE_ACL;
242 data->ocd_cksum_types = obd_cksum_types_supported_client();
244 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
245 /* flag mdc connection as lightweight, only used for test
246 * purpose, use with care */
247 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
249 data->ocd_ibits_known = MDS_INODELOCK_FULL;
250 data->ocd_version = LUSTRE_VERSION_CODE;
252 if (sb->s_flags & MS_RDONLY)
253 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
254 if (sbi->ll_flags & LL_SBI_USER_XATTR)
255 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
258 /* Setting this indicates we correctly support S_NOSEC (See kernel
259 * commit 9e1f1de02c2275d7172e18dc4e7c2065777611bf)
261 sb->s_flags |= MS_NOSEC;
264 if (sbi->ll_flags & LL_SBI_FLOCK)
265 sbi->ll_fop = &ll_file_operations_flock;
266 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
267 sbi->ll_fop = &ll_file_operations;
269 sbi->ll_fop = &ll_file_operations_noflock;
271 /* always ping even if server suppress_pings */
272 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
273 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
275 obd_connect_set_secctx(data);
277 #if defined(CONFIG_SECURITY)
278 data->ocd_connect_flags2 |= OBD_CONNECT2_SELINUX_POLICY;
281 data->ocd_brw_size = MD_MAX_BRW_SIZE;
283 err = obd_connect(NULL, &sbi->ll_md_exp, sbi->ll_md_obd,
284 &sbi->ll_sb_uuid, data, sbi->ll_cache);
286 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
287 "recovery, of which this client is not a "
288 "part. Please wait for recovery to complete,"
289 " abort, or time out.\n", md);
292 CERROR("cannot connect to %s: rc = %d\n", md, err);
296 sbi->ll_md_exp->exp_connect_data = *data;
298 err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
299 LUSTRE_SEQ_METADATA);
301 CERROR("%s: Can't init metadata layer FID infrastructure, "
302 "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
306 /* For mount, we only need fs info from MDT0, and also in DNE, it
307 * can make sure the client can be mounted as long as MDT0 is
309 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
310 ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
311 OBD_STATFS_FOR_MDT0);
313 GOTO(out_md_fid, err);
315 /* This needs to be after statfs to ensure connect has finished.
316 * Note that "data" does NOT contain the valid connect reply.
317 * If connecting to a 1.8 server there will be no LMV device, so
318 * we can access the MDC export directly and exp_connect_flags will
319 * be non-zero, but if accessing an upgraded 2.1 server it will
320 * have the correct flags filled in.
321 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
322 valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
323 if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
324 valid != CLIENT_CONNECT_MDT_REQD) {
327 OBD_ALLOC_WAIT(buf, PAGE_SIZE);
328 obd_connect_flags2str(buf, PAGE_SIZE,
329 valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
330 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
331 "feature(s) needed for correct operation "
332 "of this client (%s). Please upgrade "
333 "server or downgrade client.\n",
334 sbi->ll_md_exp->exp_obd->obd_name, buf);
335 OBD_FREE(buf, PAGE_SIZE);
336 GOTO(out_md_fid, err = -EPROTO);
339 size = sizeof(*data);
340 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
341 KEY_CONN_DATA, &size, data);
343 CERROR("%s: Get connect data failed: rc = %d\n",
344 sbi->ll_md_exp->exp_obd->obd_name, err);
345 GOTO(out_md_fid, err);
348 LASSERT(osfs->os_bsize);
349 sb->s_blocksize = osfs->os_bsize;
350 sb->s_blocksize_bits = log2(osfs->os_bsize);
351 sb->s_magic = LL_SUPER_MAGIC;
352 sb->s_maxbytes = MAX_LFS_FILESIZE;
353 sbi->ll_namelen = osfs->os_namelen;
354 sbi->ll_mnt.mnt = current->fs->root.mnt;
356 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
357 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
358 LCONSOLE_INFO("Disabling user_xattr feature because "
359 "it is not supported on the server\n");
360 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
363 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
365 sb->s_flags |= MS_POSIXACL;
367 sbi->ll_flags |= LL_SBI_ACL;
369 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
371 sb->s_flags &= ~MS_POSIXACL;
373 sbi->ll_flags &= ~LL_SBI_ACL;
376 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
377 sbi->ll_flags |= LL_SBI_64BIT_HASH;
379 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
380 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
382 if (obd_connect_has_secctx(data))
383 sbi->ll_flags |= LL_SBI_FILE_SECCTX;
385 if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
386 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
387 LCONSOLE_INFO("%s: disabling xattr cache due to "
388 "unknown maximum xattr size.\n", dt);
389 } else if (!sbi->ll_xattr_cache_set) {
390 /* If xattr_cache is already set (no matter 0 or 1)
391 * during processing llog, it won't be enabled here. */
392 sbi->ll_flags |= LL_SBI_XATTR_CACHE;
393 sbi->ll_xattr_cache_enabled = 1;
397 sbi->ll_dt_obd = class_name2obd(dt);
398 if (!sbi->ll_dt_obd) {
399 CERROR("DT %s: not setup or attached\n", dt);
400 GOTO(out_md_fid, err = -ENODEV);
403 /* pass client page size via ocd_grant_blkbits, the server should report
404 * back its backend blocksize for grant calculation purpose */
405 data->ocd_grant_blkbits = PAGE_SHIFT;
407 /* indicate OST features supported by this client */
408 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
409 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
410 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
411 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
412 OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
413 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
414 OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
415 OBD_CONNECT_EINPROGRESS |
416 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
417 OBD_CONNECT_LAYOUTLOCK |
418 OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
419 OBD_CONNECT_BULK_MBITS | OBD_CONNECT_SHORTIO |
420 OBD_CONNECT_FLAGS2 | OBD_CONNECT_GRANT_SHRINK;
422 /* The client currently advertises support for OBD_CONNECT_LOCKAHEAD_OLD so it
423 * can interoperate with an older version of lockahead which was released prior
424 * to landing in master. This support will be dropped when 2.13 development
425 * starts. At the point, we should not just drop the connect flag (below), we
426 * should also remove the support in the code.
428 * Removing it means a few things:
429 * 1. Remove this section here
430 * 2. Remove CEF_NONBLOCK in ll_file_lockahead()
431 * 3. Remove function exp_connect_lockahead_old
432 * 4. Remove LDLM_FL_LOCKAHEAD_OLD_RESERVED in lustre_dlm_flags.h
434 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 12, 50, 0)
435 data->ocd_connect_flags |= OBD_CONNECT_LOCKAHEAD_OLD;
438 data->ocd_connect_flags2 = OBD_CONNECT2_LOCKAHEAD;
440 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
441 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
443 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
444 * disabled by default, because it can still be enabled on the
445 * fly via /sys. As a consequence, we still need to come to an
446 * agreement on the supported algorithms at connect time
448 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
450 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
451 data->ocd_cksum_types = OBD_CKSUM_ADLER;
453 data->ocd_cksum_types = obd_cksum_types_supported_client();
455 #ifdef HAVE_LRU_RESIZE_SUPPORT
456 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
458 /* always ping even if server suppress_pings */
459 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
460 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
462 CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
463 "ocd_grant: %d\n", data->ocd_connect_flags,
464 data->ocd_version, data->ocd_grant);
466 sbi->ll_dt_obd->obd_upcall.onu_owner = &sbi->ll_lco;
467 sbi->ll_dt_obd->obd_upcall.onu_upcall = cl_ocd_update;
469 data->ocd_brw_size = DT_MAX_BRW_SIZE;
471 err = obd_connect(NULL, &sbi->ll_dt_exp, sbi->ll_dt_obd,
472 &sbi->ll_sb_uuid, data, sbi->ll_cache);
474 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
475 "recovery, of which this client is not a "
476 "part. Please wait for recovery to "
477 "complete, abort, or time out.\n", dt);
480 CERROR("%s: Cannot connect to %s: rc = %d\n",
481 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
485 sbi->ll_dt_exp->exp_connect_data = *data;
487 /* Don't change value if it was specified in the config log */
488 if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages == -1) {
489 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
490 max_t(unsigned long, SBI_DEFAULT_READAHEAD_WHOLE_MAX,
491 (data->ocd_brw_size >> PAGE_SHIFT));
492 if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages >
493 sbi->ll_ra_info.ra_max_pages_per_file)
494 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
495 sbi->ll_ra_info.ra_max_pages_per_file;
498 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
499 LUSTRE_SEQ_METADATA);
501 CERROR("%s: Can't init data layer FID infrastructure, "
502 "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
506 mutex_lock(&sbi->ll_lco.lco_lock);
507 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
508 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
509 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
510 mutex_unlock(&sbi->ll_lco.lco_lock);
512 fid_zero(&sbi->ll_root_fid);
513 err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
516 CERROR("cannot mds_connect: rc = %d\n", err);
517 GOTO(out_lock_cn_cb, err);
519 if (!fid_is_sane(&sbi->ll_root_fid)) {
520 CERROR("%s: Invalid root fid "DFID" during mount\n",
521 sbi->ll_md_exp->exp_obd->obd_name,
522 PFID(&sbi->ll_root_fid));
523 GOTO(out_lock_cn_cb, err = -EINVAL);
525 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
527 sb->s_op = &lustre_super_operations;
528 #ifdef HAVE_XATTR_HANDLER_FLAGS
529 sb->s_xattr = ll_xattr_handlers;
531 #if THREAD_SIZE >= 8192 /*b=17630*/
532 sb->s_export_op = &lustre_export_operations;
536 * XXX: move this to after cbd setup? */
537 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
538 if (sbi->ll_flags & LL_SBI_ACL)
539 valid |= OBD_MD_FLACL;
541 OBD_ALLOC_PTR(op_data);
543 GOTO(out_lock_cn_cb, err = -ENOMEM);
545 op_data->op_fid1 = sbi->ll_root_fid;
546 op_data->op_mode = 0;
547 op_data->op_valid = valid;
549 err = md_getattr(sbi->ll_md_exp, op_data, &request);
551 OBD_FREE_PTR(op_data);
553 CERROR("%s: md_getattr failed for root: rc = %d\n",
554 sbi->ll_md_exp->exp_obd->obd_name, err);
555 GOTO(out_lock_cn_cb, err);
558 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
559 sbi->ll_md_exp, &lmd);
561 CERROR("failed to understand root inode md: rc = %d\n", err);
562 ptlrpc_req_finished(request);
563 GOTO(out_lock_cn_cb, err);
566 LASSERT(fid_is_sane(&sbi->ll_root_fid));
567 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
568 sbi->ll_flags & LL_SBI_32BIT_API),
570 md_free_lustre_md(sbi->ll_md_exp, &lmd);
571 ptlrpc_req_finished(request);
574 #ifdef CONFIG_FS_POSIX_ACL
576 posix_acl_release(lmd.posix_acl);
577 lmd.posix_acl = NULL;
580 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
582 CERROR("lustre_lite: bad iget4 for root\n");
586 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
587 if (sbi->ll_checksum_set) {
588 err = obd_set_info_async(NULL, sbi->ll_dt_exp,
589 sizeof(KEY_CHECKSUM), KEY_CHECKSUM,
590 sizeof(checksum), &checksum, NULL);
592 CERROR("%s: Set checksum failed: rc = %d\n",
593 sbi->ll_dt_exp->exp_obd->obd_name, err);
599 sb->s_root = d_make_root(root);
600 if (sb->s_root == NULL) {
602 CERROR("%s: can't make root dentry: rc = %d\n",
603 sbi->ll_fsname, err);
606 #ifdef HAVE_DCACHE_LOCK
607 sb->s_root->d_op = &ll_d_ops;
610 sbi->ll_sdev_orig = sb->s_dev;
612 /* We set sb->s_dev equal on all lustre clients in order to support
613 * NFS export clustering. NFSD requires that the FSID be the same
615 /* s_dev is also used in lt_compare() to compare two fs, but that is
616 * only a node-local comparison. */
617 uuid = obd_get_uuid(sbi->ll_md_exp);
619 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
626 if (sbi->ll_dt_obd) {
627 err = sysfs_create_link(&sbi->ll_kset.kobj,
628 &sbi->ll_dt_obd->obd_kset.kobj,
629 sbi->ll_dt_obd->obd_type->typ_name);
631 CERROR("%s: could not register %s in llite: rc = %d\n",
632 dt, sbi->ll_fsname, err);
637 if (sbi->ll_md_obd) {
638 err = sysfs_create_link(&sbi->ll_kset.kobj,
639 &sbi->ll_md_obd->obd_kset.kobj,
640 sbi->ll_md_obd->obd_type->typ_name);
642 CERROR("%s: could not register %s in llite: rc = %d\n",
643 md, sbi->ll_fsname, err);
653 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
655 obd_disconnect(sbi->ll_dt_exp);
656 sbi->ll_dt_exp = NULL;
657 sbi->ll_dt_obd = NULL;
659 obd_fid_fini(sbi->ll_md_exp->exp_obd);
661 obd_disconnect(sbi->ll_md_exp);
662 sbi->ll_md_exp = NULL;
663 sbi->ll_md_obd = NULL;
672 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
676 size = sizeof(*lmmsize);
677 rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
678 KEY_MAX_EASIZE, &size, lmmsize);
680 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
681 sbi->ll_dt_exp->exp_obd->obd_name, rc);
685 CDEBUG(D_INFO, "max LOV ea size: %d\n", *lmmsize);
688 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
689 KEY_MAX_EASIZE, &size, lmmsize);
691 CERROR("Get max mdsize error rc %d\n", rc);
693 CDEBUG(D_INFO, "max LMV ea size: %d\n", *lmmsize);
699 * Get the value of the default_easize parameter.
701 * \see client_obd::cl_default_mds_easize
703 * \param[in] sbi superblock info for this filesystem
704 * \param[out] lmmsize pointer to storage location for value
706 * \retval 0 on success
707 * \retval negative negated errno on failure
709 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
714 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
715 KEY_DEFAULT_EASIZE, &size, lmmsize);
717 CERROR("Get default mdsize error rc %d\n", rc);
723 * Set the default_easize parameter to the given value.
725 * \see client_obd::cl_default_mds_easize
727 * \param[in] sbi superblock info for this filesystem
728 * \param[in] lmmsize the size to set
730 * \retval 0 on success
731 * \retval negative negated errno on failure
733 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
737 if (lmmsize < sizeof(struct lov_mds_md) ||
738 lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
741 rc = obd_set_info_async(NULL, sbi->ll_md_exp,
742 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
743 sizeof(int), &lmmsize, NULL);
748 static void client_common_put_super(struct super_block *sb)
750 struct ll_sb_info *sbi = ll_s2sbi(sb);
755 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
756 obd_disconnect(sbi->ll_dt_exp);
757 sbi->ll_dt_exp = NULL;
759 ll_debugfs_unregister_super(sb);
761 obd_fid_fini(sbi->ll_md_exp->exp_obd);
762 obd_disconnect(sbi->ll_md_exp);
763 sbi->ll_md_exp = NULL;
768 void ll_kill_super(struct super_block *sb)
770 struct ll_sb_info *sbi;
774 if (!(sb->s_flags & MS_ACTIVE))
778 /* we need restore s_dev from changed for clustred NFS before put_super
779 * because new kernels have cached s_dev and change sb->s_dev in
780 * put_super not affected real removing devices */
782 sb->s_dev = sbi->ll_sdev_orig;
783 sbi->ll_umounting = 1;
785 /* wait running statahead threads to quit */
786 while (atomic_read(&sbi->ll_sa_running) > 0) {
787 set_current_state(TASK_UNINTERRUPTIBLE);
788 schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
795 static inline int ll_set_opt(const char *opt, char *data, int fl)
797 if (strncmp(opt, data, strlen(opt)) != 0)
803 /* non-client-specific mount options are parsed in lmd_parse */
804 static int ll_options(char *options, struct ll_sb_info *sbi)
807 char *s1 = options, *s2;
808 int *flags = &sbi->ll_flags;
814 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
817 CDEBUG(D_SUPER, "next opt=%s\n", s1);
818 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
823 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
828 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
833 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
838 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
843 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
848 tmp = ll_set_opt("context", s1, 1);
851 tmp = ll_set_opt("fscontext", s1, 1);
854 tmp = ll_set_opt("defcontext", s1, 1);
857 tmp = ll_set_opt("rootcontext", s1, 1);
860 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
865 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
871 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
874 sbi->ll_checksum_set = 1;
877 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
880 sbi->ll_checksum_set = 1;
883 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
888 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
893 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
898 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
903 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
908 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
913 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
918 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
923 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
929 s2 = strchr(s1, ',');
937 void ll_lli_init(struct ll_inode_info *lli)
939 lli->lli_inode_magic = LLI_INODE_MAGIC;
941 spin_lock_init(&lli->lli_lock);
942 lli->lli_posix_acl = NULL;
943 /* Do not set lli_fid, it has been initialized already. */
944 fid_zero(&lli->lli_pfid);
945 lli->lli_mds_read_och = NULL;
946 lli->lli_mds_write_och = NULL;
947 lli->lli_mds_exec_och = NULL;
948 lli->lli_open_fd_read_count = 0;
949 lli->lli_open_fd_write_count = 0;
950 lli->lli_open_fd_exec_count = 0;
951 mutex_init(&lli->lli_och_mutex);
952 spin_lock_init(&lli->lli_agl_lock);
953 spin_lock_init(&lli->lli_layout_lock);
954 ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
955 lli->lli_clob = NULL;
957 init_rwsem(&lli->lli_xattrs_list_rwsem);
958 mutex_init(&lli->lli_xattrs_enq_lock);
960 LASSERT(lli->lli_vfs_inode.i_mode != 0);
961 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
962 mutex_init(&lli->lli_readdir_mutex);
963 lli->lli_opendir_key = NULL;
965 spin_lock_init(&lli->lli_sa_lock);
966 lli->lli_opendir_pid = 0;
967 lli->lli_sa_enabled = 0;
968 init_rwsem(&lli->lli_lsm_sem);
970 mutex_init(&lli->lli_size_mutex);
971 lli->lli_symlink_name = NULL;
972 init_rwsem(&lli->lli_trunc_sem);
973 range_lock_tree_init(&lli->lli_write_tree);
974 init_rwsem(&lli->lli_glimpse_sem);
975 lli->lli_glimpse_time = ktime_set(0, 0);
976 INIT_LIST_HEAD(&lli->lli_agl_list);
977 lli->lli_agl_index = 0;
978 lli->lli_async_rc = 0;
979 spin_lock_init(&lli->lli_heat_lock);
980 obd_heat_clear(lli->lli_heat_instances, OBD_HEAT_COUNT);
981 lli->lli_heat_flags = 0;
983 mutex_init(&lli->lli_layout_mutex);
984 memset(lli->lli_jobid, 0, sizeof(lli->lli_jobid));
987 #define MAX_STRING_SIZE 128
989 #ifndef HAVE_SUPER_SETUP_BDI_NAME
991 #define LSI_BDI_INITIALIZED 0x00400000
993 #ifndef HAVE_BDI_CAP_MAP_COPY
994 # define BDI_CAP_MAP_COPY 0
997 static int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
999 struct lustre_sb_info *lsi = s2lsi(sb);
1000 char buf[MAX_STRING_SIZE];
1004 err = bdi_init(&lsi->lsi_bdi);
1008 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1009 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1010 lsi->lsi_bdi.name = "lustre";
1011 va_start(args, fmt);
1012 vsnprintf(buf, MAX_STRING_SIZE, fmt, args);
1014 err = bdi_register(&lsi->lsi_bdi, NULL, "%s", buf);
1017 sb->s_bdi = &lsi->lsi_bdi;
1021 #endif /* !HAVE_SUPER_SETUP_BDI_NAME */
1023 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
1025 struct lustre_profile *lprof = NULL;
1026 struct lustre_sb_info *lsi = s2lsi(sb);
1027 struct ll_sb_info *sbi = NULL;
1028 char *dt = NULL, *md = NULL;
1029 char *profilenm = get_profile_name(sb);
1030 struct config_llog_instance *cfg;
1031 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1032 const int instlen = 16 + 2;
1033 unsigned long cfg_instance = ll_get_cfg_instance(sb);
1034 char name[MAX_STRING_SIZE];
1043 /* for ASLR, to map between cfg_instance and hashed ptr */
1044 CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1045 profilenm, cfg_instance, sb);
1047 try_module_get(THIS_MODULE);
1051 GOTO(out_free_cfg, err = -ENOMEM);
1053 /* client additional sb info */
1054 lsi->lsi_llsbi = sbi = ll_init_sbi();
1056 GOTO(out_free_cfg, err = -ENOMEM);
1058 err = ll_options(lsi->lsi_lmd->lmd_opts, sbi);
1060 GOTO(out_free_cfg, err);
1062 #ifndef HAVE_DCACHE_LOCK
1063 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1064 sb->s_d_op = &ll_d_ops;
1067 generate_random_uuid(uuid.b);
1068 snprintf(sbi->ll_sb_uuid.uuid, UUID_SIZE, "%pU", uuid.b);
1070 CDEBUG(D_CONFIG, "llite sb uuid: %s\n", sbi->ll_sb_uuid.uuid);
1073 len = strlen(profilenm);
1074 ptr = strrchr(profilenm, '-');
1075 if (ptr && (strcmp(ptr, "-client") == 0))
1078 if (len > LUSTRE_MAXFSNAME) {
1079 if (unlikely(len >= MAX_STRING_SIZE))
1080 len = MAX_STRING_SIZE - 1;
1081 strncpy(name, profilenm, len);
1083 err = -ENAMETOOLONG;
1084 CERROR("%s: fsname longer than %u characters: rc = %d\n",
1085 name, LUSTRE_MAXFSNAME, err);
1086 GOTO(out_free_cfg, err);
1088 strncpy(sbi->ll_fsname, profilenm, len);
1089 sbi->ll_fsname[len] = '\0';
1092 snprintf(name, sizeof(name), "%.*s-%016lx", len,
1093 profilenm, cfg_instance);
1095 err = super_setup_bdi_name(sb, "%s", name);
1097 GOTO(out_free_cfg, err);
1099 /* Call ll_debugfs_register_super() before lustre_process_log()
1100 * so that "llite.*.*" params can be processed correctly.
1102 err = ll_debugfs_register_super(sb, name);
1104 CERROR("%s: could not register mountpoint in llite: rc = %d\n",
1105 sbi->ll_fsname, err);
1109 /* The cfg_instance is a value unique to this super, in case some
1110 * joker tries to mount the same fs at two mount points.
1112 cfg->cfg_instance = cfg_instance;
1113 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1114 cfg->cfg_callback = class_config_llog_handler;
1115 cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1116 /* set up client obds */
1117 err = lustre_process_log(sb, profilenm, cfg);
1119 GOTO(out_debugfs, err);
1121 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1122 lprof = class_get_profile(profilenm);
1123 if (lprof == NULL) {
1124 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1125 " read from the MGS. Does that filesystem "
1126 "exist?\n", profilenm);
1127 GOTO(out_debugfs, err = -EINVAL);
1129 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1130 lprof->lp_md, lprof->lp_dt);
1132 dt_len = strlen(lprof->lp_dt) + instlen + 2;
1133 OBD_ALLOC(dt, dt_len);
1135 GOTO(out_profile, err = -ENOMEM);
1136 snprintf(dt, dt_len - 1, "%s-%016lx", lprof->lp_dt, cfg_instance);
1138 md_len = strlen(lprof->lp_md) + instlen + 2;
1139 OBD_ALLOC(md, md_len);
1141 GOTO(out_free_dt, err = -ENOMEM);
1142 snprintf(md, md_len - 1, "%s-%016lx", lprof->lp_md, cfg_instance);
1144 /* connections, registrations, sb setup */
1145 err = client_common_fill_super(sb, md, dt, mnt);
1147 GOTO(out_free_md, err);
1149 sbi->ll_client_common_fill_super_succeeded = 1;
1153 OBD_FREE(md, md_len);
1156 OBD_FREE(dt, dt_len);
1159 class_put_profile(lprof);
1162 ll_debugfs_unregister_super(sb);
1169 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1170 LCONSOLE_WARN("Mounted %s\n", profilenm);
1172 } /* ll_fill_super */
1174 void ll_put_super(struct super_block *sb)
1176 struct config_llog_instance cfg, params_cfg;
1177 struct obd_device *obd;
1178 struct lustre_sb_info *lsi = s2lsi(sb);
1179 struct ll_sb_info *sbi = ll_s2sbi(sb);
1180 char *profilenm = get_profile_name(sb);
1181 unsigned long cfg_instance = ll_get_cfg_instance(sb);
1183 int next, force = 1, rc = 0;
1187 GOTO(out_no_sbi, 0);
1189 /* Should replace instance_id with something better for ASLR */
1190 CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1191 profilenm, cfg_instance, sb);
1193 cfg.cfg_instance = cfg_instance;
1194 lustre_end_log(sb, profilenm, &cfg);
1196 params_cfg.cfg_instance = cfg_instance;
1197 lustre_end_log(sb, PARAMS_FILENAME, ¶ms_cfg);
1199 if (sbi->ll_md_exp) {
1200 obd = class_exp2obd(sbi->ll_md_exp);
1202 force = obd->obd_force;
1205 /* Wait for unstable pages to be committed to stable storage */
1207 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1208 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1209 atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1213 ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1214 if (force == 0 && rc != -EINTR)
1215 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1217 /* We need to set force before the lov_disconnect in
1218 lustre_common_put_super, since l_d cleans up osc's as well. */
1221 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1223 obd->obd_force = force;
1227 if (sbi->ll_client_common_fill_super_succeeded) {
1228 /* Only if client_common_fill_super succeeded */
1229 client_common_put_super(sb);
1233 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1234 class_manual_cleanup(obd);
1237 if (sbi->ll_flags & LL_SBI_VERBOSE)
1238 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1241 class_del_profile(profilenm);
1243 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1244 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1245 bdi_destroy(&lsi->lsi_bdi);
1246 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1251 lsi->lsi_llsbi = NULL;
1253 lustre_common_put_super(sb);
1255 cl_env_cache_purge(~0);
1257 module_put(THIS_MODULE);
1260 } /* client_put_super */
1262 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1264 struct inode *inode = NULL;
1266 /* NOTE: we depend on atomic igrab() -bzzz */
1267 lock_res_and_lock(lock);
1268 if (lock->l_resource->lr_lvb_inode) {
1269 struct ll_inode_info * lli;
1270 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1271 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1272 inode = igrab(lock->l_resource->lr_lvb_inode);
1274 inode = lock->l_resource->lr_lvb_inode;
1275 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1276 D_WARNING, lock, "lr_lvb_inode %p is "
1277 "bogus: magic %08x",
1278 lock->l_resource->lr_lvb_inode,
1279 lli->lli_inode_magic);
1283 unlock_res_and_lock(lock);
1287 void ll_dir_clear_lsm_md(struct inode *inode)
1289 struct ll_inode_info *lli = ll_i2info(inode);
1291 LASSERT(S_ISDIR(inode->i_mode));
1293 if (lli->lli_lsm_md) {
1294 lmv_free_memmd(lli->lli_lsm_md);
1295 lli->lli_lsm_md = NULL;
1298 if (lli->lli_default_lsm_md) {
1299 lmv_free_memmd(lli->lli_default_lsm_md);
1300 lli->lli_default_lsm_md = NULL;
1304 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1305 const struct lu_fid *fid,
1306 struct lustre_md *md)
1308 struct ll_sb_info *sbi = ll_s2sbi(sb);
1309 struct mdt_body *body = md->body;
1310 struct inode *inode;
1314 ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1315 inode = iget_locked(sb, ino);
1316 if (inode == NULL) {
1317 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1318 sbi->ll_fsname, PFID(fid));
1319 RETURN(ERR_PTR(-ENOENT));
1322 if (inode->i_state & I_NEW) {
1323 struct ll_inode_info *lli = ll_i2info(inode);
1324 struct lmv_stripe_md *lsm = md->lmv;
1326 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1327 (body->mbo_mode & S_IFMT);
1328 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1331 inode->i_mtime.tv_sec = 0;
1332 inode->i_atime.tv_sec = 0;
1333 inode->i_ctime.tv_sec = 0;
1336 #ifdef HAVE_BACKING_DEV_INFO
1337 /* initializing backing dev info. */
1338 inode->i_mapping->backing_dev_info =
1339 &s2lsi(inode->i_sb)->lsi_bdi;
1341 inode->i_op = &ll_dir_inode_operations;
1342 inode->i_fop = &ll_dir_operations;
1343 lli->lli_fid = *fid;
1346 LASSERT(lsm != NULL);
1347 /* master object FID */
1348 lli->lli_pfid = body->mbo_fid1;
1349 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1350 lli, PFID(fid), PFID(&lli->lli_pfid));
1351 unlock_new_inode(inode);
1357 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1360 struct lmv_stripe_md *lsm = md->lmv;
1361 struct ll_inode_info *lli = ll_i2info(inode);
1364 LASSERT(lsm != NULL);
1366 CDEBUG(D_INODE, "%s: "DFID" set dir layout:\n",
1367 ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid));
1368 lsm_md_dump(D_INODE, lsm);
1370 /* XXX sigh, this lsm_root initialization should be in
1371 * LMV layer, but it needs ll_iget right now, so we
1372 * put this here right now. */
1373 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1374 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1375 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1377 if (!fid_is_sane(fid))
1380 /* Unfortunately ll_iget will call ll_update_inode,
1381 * where the initialization of slave inode is slightly
1382 * different, so it reset lsm_md to NULL to avoid
1383 * initializing lsm for slave inode. */
1384 lsm->lsm_md_oinfo[i].lmo_root =
1385 ll_iget_anon_dir(inode->i_sb, fid, md);
1386 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1387 int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1389 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1391 iput(lsm->lsm_md_oinfo[i].lmo_root);
1392 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1398 lli->lli_lsm_md = lsm;
1403 static void ll_update_default_lsm_md(struct inode *inode, struct lustre_md *md)
1405 struct ll_inode_info *lli = ll_i2info(inode);
1407 if (!md->default_lmv) {
1408 /* clear default lsm */
1409 if (lli->lli_default_lsm_md) {
1410 down_write(&lli->lli_lsm_sem);
1411 if (lli->lli_default_lsm_md) {
1412 lmv_free_memmd(lli->lli_default_lsm_md);
1413 lli->lli_default_lsm_md = NULL;
1415 up_write(&lli->lli_lsm_sem);
1417 } else if (lli->lli_default_lsm_md) {
1418 /* update default lsm if it changes */
1419 down_read(&lli->lli_lsm_sem);
1420 if (lli->lli_default_lsm_md &&
1421 !lsm_md_eq(lli->lli_default_lsm_md, md->default_lmv)) {
1422 up_read(&lli->lli_lsm_sem);
1423 down_write(&lli->lli_lsm_sem);
1424 if (lli->lli_default_lsm_md)
1425 lmv_free_memmd(lli->lli_default_lsm_md);
1426 lli->lli_default_lsm_md = md->default_lmv;
1427 lsm_md_dump(D_INODE, md->default_lmv);
1428 md->default_lmv = NULL;
1429 up_write(&lli->lli_lsm_sem);
1431 up_read(&lli->lli_lsm_sem);
1434 /* init default lsm */
1435 down_write(&lli->lli_lsm_sem);
1436 lli->lli_default_lsm_md = md->default_lmv;
1437 lsm_md_dump(D_INODE, md->default_lmv);
1438 md->default_lmv = NULL;
1439 up_write(&lli->lli_lsm_sem);
1443 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1445 struct ll_inode_info *lli = ll_i2info(inode);
1446 struct lmv_stripe_md *lsm = md->lmv;
1451 LASSERT(S_ISDIR(inode->i_mode));
1452 CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1453 PFID(ll_inode2fid(inode)));
1455 /* update default LMV */
1456 if (md->default_lmv)
1457 ll_update_default_lsm_md(inode, md);
1460 * no striped information from request, lustre_md from req does not
1461 * include stripeEA, see ll_md_setattr()
1467 * normally dir layout doesn't change, only take read lock to check
1468 * that to avoid blocking other MD operations.
1470 if (lli->lli_lsm_md)
1471 down_read(&lli->lli_lsm_sem);
1473 down_write(&lli->lli_lsm_sem);
1476 * if dir layout mismatch, check whether version is increased, which
1477 * means layout is changed, this happens in dir migration and lfsck.
1479 * foreign LMV should not change.
1481 if (lli->lli_lsm_md &&
1482 lli->lli_lsm_md->lsm_md_magic != LMV_MAGIC_FOREIGN &&
1483 !lsm_md_eq(lli->lli_lsm_md, lsm)) {
1484 if (lsm->lsm_md_layout_version <=
1485 lli->lli_lsm_md->lsm_md_layout_version) {
1486 CERROR("%s: "DFID" dir layout mismatch:\n",
1487 ll_i2sbi(inode)->ll_fsname,
1488 PFID(&lli->lli_fid));
1489 lsm_md_dump(D_ERROR, lli->lli_lsm_md);
1490 lsm_md_dump(D_ERROR, lsm);
1491 GOTO(unlock, rc = -EINVAL);
1494 /* layout changed, switch to write lock */
1495 up_read(&lli->lli_lsm_sem);
1496 down_write(&lli->lli_lsm_sem);
1497 ll_dir_clear_lsm_md(inode);
1500 /* set directory layout */
1501 if (!lli->lli_lsm_md) {
1502 struct cl_attr *attr;
1504 if (lsm->lsm_md_magic == LMV_MAGIC_FOREIGN) {
1505 /* set md->lmv to NULL, so the following free lustre_md
1506 * will not free this lsm */
1508 lli->lli_lsm_md = lsm;
1509 up_write(&lli->lli_lsm_sem);
1513 rc = ll_init_lsm_md(inode, md);
1514 up_write(&lli->lli_lsm_sem);
1518 /* set md->lmv to NULL, so the following free lustre_md
1519 * will not free this lsm */
1523 * md_merge_attr() may take long, since lsm is already set,
1524 * switch to read lock.
1526 down_read(&lli->lli_lsm_sem);
1528 OBD_ALLOC_PTR(attr);
1530 GOTO(unlock, rc = -ENOMEM);
1532 /* validate the lsm */
1533 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1534 ll_md_blocking_ast);
1540 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1541 md->body->mbo_nlink = attr->cat_nlink;
1542 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1543 md->body->mbo_size = attr->cat_size;
1544 if (md->body->mbo_valid & OBD_MD_FLATIME)
1545 md->body->mbo_atime = attr->cat_atime;
1546 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1547 md->body->mbo_ctime = attr->cat_ctime;
1548 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1549 md->body->mbo_mtime = attr->cat_mtime;
1554 up_read(&lli->lli_lsm_sem);
1559 void ll_clear_inode(struct inode *inode)
1561 struct ll_inode_info *lli = ll_i2info(inode);
1562 struct ll_sb_info *sbi = ll_i2sbi(inode);
1565 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1566 PFID(ll_inode2fid(inode)), inode);
1568 if (S_ISDIR(inode->i_mode)) {
1569 /* these should have been cleared in ll_file_release */
1570 LASSERT(lli->lli_opendir_key == NULL);
1571 LASSERT(lli->lli_sai == NULL);
1572 LASSERT(lli->lli_opendir_pid == 0);
1575 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1577 LASSERT(!lli->lli_open_fd_write_count);
1578 LASSERT(!lli->lli_open_fd_read_count);
1579 LASSERT(!lli->lli_open_fd_exec_count);
1581 if (lli->lli_mds_write_och)
1582 ll_md_real_close(inode, FMODE_WRITE);
1583 if (lli->lli_mds_exec_och)
1584 ll_md_real_close(inode, FMODE_EXEC);
1585 if (lli->lli_mds_read_och)
1586 ll_md_real_close(inode, FMODE_READ);
1588 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1589 OBD_FREE(lli->lli_symlink_name,
1590 strlen(lli->lli_symlink_name) + 1);
1591 lli->lli_symlink_name = NULL;
1594 ll_xattr_cache_destroy(inode);
1596 #ifdef CONFIG_FS_POSIX_ACL
1597 forget_all_cached_acls(inode);
1598 if (lli->lli_posix_acl) {
1599 posix_acl_release(lli->lli_posix_acl);
1600 lli->lli_posix_acl = NULL;
1603 lli->lli_inode_magic = LLI_INODE_DEAD;
1605 if (S_ISDIR(inode->i_mode))
1606 ll_dir_clear_lsm_md(inode);
1607 else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1608 LASSERT(list_empty(&lli->lli_agl_list));
1611 * XXX This has to be done before lsm is freed below, because
1612 * cl_object still uses inode lsm.
1614 cl_inode_fini(inode);
1619 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1621 struct lustre_md md;
1622 struct inode *inode = dentry->d_inode;
1623 struct ll_sb_info *sbi = ll_i2sbi(inode);
1624 struct ptlrpc_request *request = NULL;
1628 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1629 LUSTRE_OPC_ANY, NULL);
1630 if (IS_ERR(op_data))
1631 RETURN(PTR_ERR(op_data));
1633 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1635 ptlrpc_req_finished(request);
1636 if (rc == -ENOENT) {
1638 /* Unlinked special device node? Or just a race?
1639 * Pretend we done everything. */
1640 if (!S_ISREG(inode->i_mode) &&
1641 !S_ISDIR(inode->i_mode)) {
1642 ia_valid = op_data->op_attr.ia_valid;
1643 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1644 rc = simple_setattr(dentry, &op_data->op_attr);
1645 op_data->op_attr.ia_valid = ia_valid;
1647 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1648 CERROR("md_setattr fails: rc = %d\n", rc);
1653 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1654 sbi->ll_md_exp, &md);
1656 ptlrpc_req_finished(request);
1660 ia_valid = op_data->op_attr.ia_valid;
1661 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1662 * cache is not cleared yet. */
1663 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1664 if (S_ISREG(inode->i_mode))
1666 rc = simple_setattr(dentry, &op_data->op_attr);
1667 if (S_ISREG(inode->i_mode))
1668 inode_unlock(inode);
1669 op_data->op_attr.ia_valid = ia_valid;
1671 rc = ll_update_inode(inode, &md);
1672 ptlrpc_req_finished(request);
1677 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1678 * object(s) determine the file size and mtime. Otherwise, the MDS will
1679 * keep these values until such a time that objects are allocated for it.
1680 * We do the MDS operations first, as it is checking permissions for us.
1681 * We don't to the MDS RPC if there is nothing that we want to store there,
1682 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1683 * going to do an RPC anyways.
1685 * If we are doing a truncate, we will send the mtime and ctime updates
1686 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1687 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1690 * In case of HSMimport, we only set attr on MDS.
1692 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr,
1693 enum op_xvalid xvalid, bool hsm_import)
1695 struct inode *inode = dentry->d_inode;
1696 struct ll_inode_info *lli = ll_i2info(inode);
1697 struct md_op_data *op_data = NULL;
1701 CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1702 "valid %x, hsm_import %d\n",
1703 ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid),
1704 inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1707 if (attr->ia_valid & ATTR_SIZE) {
1708 /* Check new size against VFS/VM file size limit and rlimit */
1709 rc = inode_newsize_ok(inode, attr->ia_size);
1713 /* The maximum Lustre file size is variable, based on the
1714 * OST maximum object size and number of stripes. This
1715 * needs another check in addition to the VFS check above. */
1716 if (attr->ia_size > ll_file_maxbytes(inode)) {
1717 CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
1718 PFID(&lli->lli_fid), attr->ia_size,
1719 ll_file_maxbytes(inode));
1723 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1726 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1727 if (attr->ia_valid & TIMES_SET_FLAGS) {
1728 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1729 !cfs_capable(CFS_CAP_FOWNER))
1733 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1734 if (!(xvalid & OP_XVALID_CTIME_SET) &&
1735 (attr->ia_valid & ATTR_CTIME)) {
1736 attr->ia_ctime = current_time(inode);
1737 xvalid |= OP_XVALID_CTIME_SET;
1739 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1740 (attr->ia_valid & ATTR_ATIME)) {
1741 attr->ia_atime = current_time(inode);
1742 attr->ia_valid |= ATTR_ATIME_SET;
1744 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1745 (attr->ia_valid & ATTR_MTIME)) {
1746 attr->ia_mtime = current_time(inode);
1747 attr->ia_valid |= ATTR_MTIME_SET;
1750 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1751 CDEBUG(D_INODE, "setting mtime %lld, ctime %lld, now = %lld\n",
1752 (s64)attr->ia_mtime.tv_sec, (s64)attr->ia_ctime.tv_sec,
1753 ktime_get_real_seconds());
1755 if (S_ISREG(inode->i_mode)) {
1756 if (attr->ia_valid & ATTR_SIZE)
1757 inode_dio_write_done(inode);
1758 inode_unlock(inode);
1761 /* We always do an MDS RPC, even if we're only changing the size;
1762 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1764 OBD_ALLOC_PTR(op_data);
1765 if (op_data == NULL)
1766 GOTO(out, rc = -ENOMEM);
1768 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1769 /* If we are changing file size, file content is
1770 * modified, flag it.
1772 xvalid |= OP_XVALID_OWNEROVERRIDE;
1773 op_data->op_bias |= MDS_DATA_MODIFIED;
1774 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1777 if (attr->ia_valid & ATTR_FILE) {
1778 struct ll_file_data *fd = LUSTRE_FPRIVATE(attr->ia_file);
1780 if (fd->fd_lease_och)
1781 op_data->op_bias |= MDS_TRUNC_KEEP_LEASE;
1784 op_data->op_attr = *attr;
1785 op_data->op_xvalid = xvalid;
1787 rc = ll_md_setattr(dentry, op_data);
1791 if (!S_ISREG(inode->i_mode) || hsm_import)
1794 if (attr->ia_valid & (ATTR_SIZE | ATTR_ATIME | ATTR_ATIME_SET |
1795 ATTR_MTIME | ATTR_MTIME_SET | ATTR_CTIME) ||
1796 xvalid & OP_XVALID_CTIME_SET) {
1797 /* For truncate and utimes sending attributes to OSTs, setting
1798 * mtime/atime to the past will be performed under PW [0:EOF]
1799 * extent lock (new_size:EOF for truncate). It may seem
1800 * excessive to send mtime/atime updates to OSTs when not
1801 * setting times to past, but it is necessary due to possible
1802 * time de-synchronization between MDT inode and OST objects
1804 rc = cl_setattr_ost(lli->lli_clob, attr, xvalid, 0);
1807 /* If the file was restored, it needs to set dirty flag.
1809 * We've already sent MDS_DATA_MODIFIED flag in
1810 * ll_md_setattr() for truncate. However, the MDT refuses to
1811 * set the HS_DIRTY flag on released files, so we have to set
1812 * it again if the file has been restored. Please check how
1813 * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1815 * Please notice that if the file is not released, the previous
1816 * MDS_DATA_MODIFIED has taken effect and usually
1817 * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1818 * This way we can save an RPC for common open + trunc
1820 if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1821 struct hsm_state_set hss = {
1822 .hss_valid = HSS_SETMASK,
1823 .hss_setmask = HS_DIRTY,
1827 rc2 = ll_hsm_state_set(inode, &hss);
1828 /* truncate and write can happen at the same time, so that
1829 * the file can be set modified even though the file is not
1830 * restored from released state, and ll_hsm_state_set() is
1831 * not applicable for the file, and rc2 < 0 is normal in this
1834 CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1835 PFID(ll_inode2fid(inode)), rc2);
1840 if (op_data != NULL)
1841 ll_finish_md_op_data(op_data);
1843 if (S_ISREG(inode->i_mode)) {
1845 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1846 inode_dio_wait(inode);
1847 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
1848 * flag. ll_update_inode (called from ll_md_setattr), clears
1849 * inode flags, so there is a gap where S_NOSEC is not set.
1850 * This can cause a writer to take the i_mutex unnecessarily,
1851 * but this is safe to do and should be rare. */
1852 inode_has_no_xattr(inode);
1855 ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1856 LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1861 int ll_setattr(struct dentry *de, struct iattr *attr)
1863 int mode = de->d_inode->i_mode;
1864 enum op_xvalid xvalid = 0;
1866 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1867 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1868 xvalid |= OP_XVALID_OWNEROVERRIDE;
1870 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1871 (ATTR_SIZE|ATTR_MODE)) &&
1872 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1873 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1874 !(attr->ia_mode & S_ISGID))))
1875 attr->ia_valid |= ATTR_FORCE;
1877 if ((attr->ia_valid & ATTR_MODE) &&
1879 !(attr->ia_mode & S_ISUID) &&
1880 !(attr->ia_valid & ATTR_KILL_SUID))
1881 attr->ia_valid |= ATTR_KILL_SUID;
1883 if ((attr->ia_valid & ATTR_MODE) &&
1884 ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1885 !(attr->ia_mode & S_ISGID) &&
1886 !(attr->ia_valid & ATTR_KILL_SGID))
1887 attr->ia_valid |= ATTR_KILL_SGID;
1889 return ll_setattr_raw(de, attr, xvalid, false);
1892 int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
1895 struct obd_statfs obd_osfs = { 0 };
1900 max_age = ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS;
1902 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1906 osfs->os_type = LL_SUPER_MAGIC;
1908 CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1909 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree, osfs->os_files);
1911 if (osfs->os_state & OS_STATE_SUM)
1914 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1915 flags |= OBD_STATFS_NODELAY;
1917 rc = obd_statfs(NULL, sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1918 if (rc) /* Possibly a filesystem with no OSTs. Report MDT totals. */
1921 CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1922 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1925 osfs->os_bsize = obd_osfs.os_bsize;
1926 osfs->os_blocks = obd_osfs.os_blocks;
1927 osfs->os_bfree = obd_osfs.os_bfree;
1928 osfs->os_bavail = obd_osfs.os_bavail;
1930 /* If we have _some_ OSTs, but don't have as many free objects on the
1931 * OSTs as inodes on the MDTs, reduce the reported number of inodes
1932 * to compensate, so that the "inodes in use" number is correct.
1933 * This should be kept in sync with lod_statfs() behaviour.
1935 if (obd_osfs.os_files && obd_osfs.os_ffree < osfs->os_ffree) {
1936 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1938 osfs->os_ffree = obd_osfs.os_ffree;
1944 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1946 struct super_block *sb = de->d_sb;
1947 struct obd_statfs osfs;
1948 __u64 fsid = huge_encode_dev(sb->s_dev);
1951 CDEBUG(D_VFSTRACE, "VFS Op: at %llu jiffies\n", get_jiffies_64());
1952 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1954 /* Some amount of caching on the client is allowed */
1955 rc = ll_statfs_internal(ll_s2sbi(sb), &osfs, OBD_STATFS_SUM);
1959 statfs_unpack(sfs, &osfs);
1961 /* We need to downshift for all 32-bit kernels, because we can't
1962 * tell if the kernel is being called via sys_statfs64() or not.
1963 * Stop before overflowing f_bsize - in which case it is better
1964 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1965 if (sizeof(long) < 8) {
1966 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1969 osfs.os_blocks >>= 1;
1970 osfs.os_bfree >>= 1;
1971 osfs.os_bavail >>= 1;
1975 sfs->f_blocks = osfs.os_blocks;
1976 sfs->f_bfree = osfs.os_bfree;
1977 sfs->f_bavail = osfs.os_bavail;
1978 sfs->f_fsid.val[0] = (__u32)fsid;
1979 sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1983 void ll_inode_size_lock(struct inode *inode)
1985 struct ll_inode_info *lli;
1987 LASSERT(!S_ISDIR(inode->i_mode));
1989 lli = ll_i2info(inode);
1990 mutex_lock(&lli->lli_size_mutex);
1993 void ll_inode_size_unlock(struct inode *inode)
1995 struct ll_inode_info *lli;
1997 lli = ll_i2info(inode);
1998 mutex_unlock(&lli->lli_size_mutex);
2001 void ll_update_inode_flags(struct inode *inode, int ext_flags)
2003 inode->i_flags = ll_ext_to_inode_flags(ext_flags);
2004 if (ext_flags & LUSTRE_PROJINHERIT_FL)
2005 ll_file_set_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
2007 ll_file_clear_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
2010 int ll_update_inode(struct inode *inode, struct lustre_md *md)
2012 struct ll_inode_info *lli = ll_i2info(inode);
2013 struct mdt_body *body = md->body;
2014 struct ll_sb_info *sbi = ll_i2sbi(inode);
2017 if (body->mbo_valid & OBD_MD_FLEASIZE) {
2018 rc = cl_file_inode_init(inode, md);
2023 if (S_ISDIR(inode->i_mode)) {
2024 rc = ll_update_lsm_md(inode, md);
2029 #ifdef CONFIG_FS_POSIX_ACL
2030 if (body->mbo_valid & OBD_MD_FLACL) {
2031 spin_lock(&lli->lli_lock);
2032 if (lli->lli_posix_acl)
2033 posix_acl_release(lli->lli_posix_acl);
2034 lli->lli_posix_acl = md->posix_acl;
2035 spin_unlock(&lli->lli_lock);
2038 inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
2039 sbi->ll_flags & LL_SBI_32BIT_API);
2040 inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
2042 if (body->mbo_valid & OBD_MD_FLATIME) {
2043 if (body->mbo_atime > inode->i_atime.tv_sec)
2044 inode->i_atime.tv_sec = body->mbo_atime;
2045 lli->lli_atime = body->mbo_atime;
2048 if (body->mbo_valid & OBD_MD_FLMTIME) {
2049 if (body->mbo_mtime > inode->i_mtime.tv_sec) {
2051 "setting ino %lu mtime from %lld to %llu\n",
2052 inode->i_ino, (s64)inode->i_mtime.tv_sec,
2054 inode->i_mtime.tv_sec = body->mbo_mtime;
2056 lli->lli_mtime = body->mbo_mtime;
2059 if (body->mbo_valid & OBD_MD_FLCTIME) {
2060 if (body->mbo_ctime > inode->i_ctime.tv_sec)
2061 inode->i_ctime.tv_sec = body->mbo_ctime;
2062 lli->lli_ctime = body->mbo_ctime;
2065 /* Clear i_flags to remove S_NOSEC before permissions are updated */
2066 if (body->mbo_valid & OBD_MD_FLFLAGS)
2067 ll_update_inode_flags(inode, body->mbo_flags);
2068 if (body->mbo_valid & OBD_MD_FLMODE)
2069 inode->i_mode = (inode->i_mode & S_IFMT) |
2070 (body->mbo_mode & ~S_IFMT);
2072 if (body->mbo_valid & OBD_MD_FLTYPE)
2073 inode->i_mode = (inode->i_mode & ~S_IFMT) |
2074 (body->mbo_mode & S_IFMT);
2076 LASSERT(inode->i_mode != 0);
2077 if (S_ISREG(inode->i_mode))
2078 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
2079 LL_MAX_BLKSIZE_BITS);
2081 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
2083 if (body->mbo_valid & OBD_MD_FLUID)
2084 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
2085 if (body->mbo_valid & OBD_MD_FLGID)
2086 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
2087 if (body->mbo_valid & OBD_MD_FLPROJID)
2088 lli->lli_projid = body->mbo_projid;
2089 if (body->mbo_valid & OBD_MD_FLNLINK)
2090 set_nlink(inode, body->mbo_nlink);
2091 if (body->mbo_valid & OBD_MD_FLRDEV)
2092 inode->i_rdev = old_decode_dev(body->mbo_rdev);
2094 if (body->mbo_valid & OBD_MD_FLID) {
2095 /* FID shouldn't be changed! */
2096 if (fid_is_sane(&lli->lli_fid)) {
2097 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
2098 "Trying to change FID "DFID
2099 " to the "DFID", inode "DFID"(%p)\n",
2100 PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
2101 PFID(ll_inode2fid(inode)), inode);
2103 lli->lli_fid = body->mbo_fid1;
2107 LASSERT(fid_seq(&lli->lli_fid) != 0);
2109 if (body->mbo_valid & OBD_MD_FLSIZE) {
2110 i_size_write(inode, body->mbo_size);
2112 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
2113 PFID(ll_inode2fid(inode)),
2114 (unsigned long long)body->mbo_size);
2116 if (body->mbo_valid & OBD_MD_FLBLOCKS)
2117 inode->i_blocks = body->mbo_blocks;
2120 if (body->mbo_valid & OBD_MD_TSTATE) {
2121 /* Set LLIF_FILE_RESTORING if restore ongoing and
2122 * clear it when done to ensure to start again
2123 * glimpsing updated attrs
2125 if (body->mbo_t_state & MS_RESTORE)
2126 ll_file_set_flag(lli, LLIF_FILE_RESTORING);
2128 ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
2134 int ll_read_inode2(struct inode *inode, void *opaque)
2136 struct lustre_md *md = opaque;
2137 struct ll_inode_info *lli = ll_i2info(inode);
2141 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2142 PFID(&lli->lli_fid), inode);
2144 /* Core attributes from the MDS first. This is a new inode, and
2145 * the VFS doesn't zero times in the core inode so we have to do
2146 * it ourselves. They will be overwritten by either MDS or OST
2147 * attributes - we just need to make sure they aren't newer.
2149 inode->i_mtime.tv_sec = 0;
2150 inode->i_atime.tv_sec = 0;
2151 inode->i_ctime.tv_sec = 0;
2153 rc = ll_update_inode(inode, md);
2157 /* OIDEBUG(inode); */
2159 #ifdef HAVE_BACKING_DEV_INFO
2160 /* initializing backing dev info. */
2161 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2163 if (S_ISREG(inode->i_mode)) {
2164 struct ll_sb_info *sbi = ll_i2sbi(inode);
2165 inode->i_op = &ll_file_inode_operations;
2166 inode->i_fop = sbi->ll_fop;
2167 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2169 } else if (S_ISDIR(inode->i_mode)) {
2170 inode->i_op = &ll_dir_inode_operations;
2171 inode->i_fop = &ll_dir_operations;
2173 } else if (S_ISLNK(inode->i_mode)) {
2174 inode->i_op = &ll_fast_symlink_inode_operations;
2177 inode->i_op = &ll_special_inode_operations;
2179 init_special_inode(inode, inode->i_mode,
2188 void ll_delete_inode(struct inode *inode)
2190 struct ll_inode_info *lli = ll_i2info(inode);
2191 struct address_space *mapping = &inode->i_data;
2192 unsigned long nrpages;
2195 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL) {
2196 /* It is last chance to write out dirty pages,
2197 * otherwise we may lose data while umount.
2199 * If i_nlink is 0 then just discard data. This is safe because
2200 * local inode gets i_nlink 0 from server only for the last
2201 * unlink, so that file is not opened somewhere else
2203 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, inode->i_nlink ?
2204 CL_FSYNC_LOCAL : CL_FSYNC_DISCARD, 1);
2206 truncate_inode_pages_final(mapping);
2208 /* Workaround for LU-118: Note nrpages may not be totally updated when
2209 * truncate_inode_pages() returns, as there can be a page in the process
2210 * of deletion (inside __delete_from_page_cache()) in the specified
2211 * range. Thus mapping->nrpages can be non-zero when this function
2212 * returns even after truncation of the whole mapping. Only do this if
2213 * npages isn't already zero.
2215 nrpages = mapping->nrpages;
2217 xa_lock_irq(&mapping->i_pages);
2218 nrpages = mapping->nrpages;
2219 xa_unlock_irq(&mapping->i_pages);
2220 } /* Workaround end */
2222 LASSERTF(nrpages == 0, "%s: inode="DFID"(%p) nrpages=%lu, "
2223 "see https://jira.whamcloud.com/browse/LU-118\n",
2224 ll_i2sbi(inode)->ll_fsname,
2225 PFID(ll_inode2fid(inode)), inode, nrpages);
2227 #ifdef HAVE_SBOPS_EVICT_INODE
2228 ll_clear_inode(inode);
2235 int ll_iocontrol(struct inode *inode, struct file *file,
2236 unsigned int cmd, unsigned long arg)
2238 struct ll_sb_info *sbi = ll_i2sbi(inode);
2239 struct ptlrpc_request *req = NULL;
2244 case FS_IOC_GETFLAGS: {
2245 struct mdt_body *body;
2246 struct md_op_data *op_data;
2248 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2249 0, 0, LUSTRE_OPC_ANY,
2251 if (IS_ERR(op_data))
2252 RETURN(PTR_ERR(op_data));
2254 op_data->op_valid = OBD_MD_FLFLAGS;
2255 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2256 ll_finish_md_op_data(op_data);
2258 CERROR("%s: failure inode "DFID": rc = %d\n",
2259 sbi->ll_md_exp->exp_obd->obd_name,
2260 PFID(ll_inode2fid(inode)), rc);
2264 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2266 flags = body->mbo_flags;
2268 ptlrpc_req_finished(req);
2270 RETURN(put_user(flags, (int __user *)arg));
2272 case FS_IOC_SETFLAGS: {
2274 struct md_op_data *op_data;
2275 struct cl_object *obj;
2276 struct fsxattr fa = { 0 };
2278 if (get_user(flags, (int __user *)arg))
2281 fa.fsx_projid = ll_i2info(inode)->lli_projid;
2282 if (flags & LUSTRE_PROJINHERIT_FL)
2283 fa.fsx_xflags = FS_XFLAG_PROJINHERIT;
2285 rc = ll_ioctl_check_project(inode, &fa);
2289 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2290 LUSTRE_OPC_ANY, NULL);
2291 if (IS_ERR(op_data))
2292 RETURN(PTR_ERR(op_data));
2294 op_data->op_attr_flags = flags;
2295 op_data->op_xvalid |= OP_XVALID_FLAGS;
2296 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2297 ll_finish_md_op_data(op_data);
2298 ptlrpc_req_finished(req);
2302 ll_update_inode_flags(inode, flags);
2304 obj = ll_i2info(inode)->lli_clob;
2308 OBD_ALLOC_PTR(attr);
2312 rc = cl_setattr_ost(obj, attr, OP_XVALID_FLAGS, flags);
2324 int ll_flush_ctx(struct inode *inode)
2326 struct ll_sb_info *sbi = ll_i2sbi(inode);
2328 CDEBUG(D_SEC, "flush context for user %d\n",
2329 from_kuid(&init_user_ns, current_uid()));
2331 obd_set_info_async(NULL, sbi->ll_md_exp,
2332 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2334 obd_set_info_async(NULL, sbi->ll_dt_exp,
2335 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2340 /* umount -f client means force down, don't save state */
2341 void ll_umount_begin(struct super_block *sb)
2343 struct ll_sb_info *sbi = ll_s2sbi(sb);
2344 struct obd_device *obd;
2345 struct obd_ioctl_data *ioc_data;
2346 struct l_wait_info lwi;
2347 wait_queue_head_t waitq;
2350 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2351 sb->s_count, atomic_read(&sb->s_active));
2353 obd = class_exp2obd(sbi->ll_md_exp);
2355 CERROR("Invalid MDC connection handle %#llx\n",
2356 sbi->ll_md_exp->exp_handle.h_cookie);
2362 obd = class_exp2obd(sbi->ll_dt_exp);
2364 CERROR("Invalid LOV connection handle %#llx\n",
2365 sbi->ll_dt_exp->exp_handle.h_cookie);
2371 OBD_ALLOC_PTR(ioc_data);
2373 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2374 sizeof *ioc_data, ioc_data, NULL);
2376 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2377 sizeof *ioc_data, ioc_data, NULL);
2379 OBD_FREE_PTR(ioc_data);
2382 /* Really, we'd like to wait until there are no requests outstanding,
2383 * and then continue. For now, we just periodically checking for vfs
2384 * to decrement mnt_cnt and hope to finish it within 10sec.
2386 init_waitqueue_head(&waitq);
2387 lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(10),
2388 cfs_time_seconds(1), NULL, NULL);
2389 l_wait_event(waitq, may_umount(sbi->ll_mnt.mnt), &lwi);
2394 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2396 struct ll_sb_info *sbi = ll_s2sbi(sb);
2397 char *profilenm = get_profile_name(sb);
2401 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2402 read_only = *flags & MS_RDONLY;
2403 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2404 sizeof(KEY_READ_ONLY),
2405 KEY_READ_ONLY, sizeof(read_only),
2408 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2409 profilenm, read_only ?
2410 "read-only" : "read-write", err);
2415 sb->s_flags |= MS_RDONLY;
2417 sb->s_flags &= ~MS_RDONLY;
2419 if (sbi->ll_flags & LL_SBI_VERBOSE)
2420 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2421 read_only ? "read-only" : "read-write");
2427 * Cleanup the open handle that is cached on MDT-side.
2429 * For open case, the client side open handling thread may hit error
2430 * after the MDT grant the open. Under such case, the client should
2431 * send close RPC to the MDT as cleanup; otherwise, the open handle
2432 * on the MDT will be leaked there until the client umount or evicted.
2434 * In further, if someone unlinked the file, because the open handle
2435 * holds the reference on such file/object, then it will block the
2436 * subsequent threads that want to locate such object via FID.
2438 * \param[in] sb super block for this file-system
2439 * \param[in] open_req pointer to the original open request
2441 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2443 struct mdt_body *body;
2444 struct md_op_data *op_data;
2445 struct ptlrpc_request *close_req = NULL;
2446 struct obd_export *exp = ll_s2sbi(sb)->ll_md_exp;
2449 body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2450 OBD_ALLOC_PTR(op_data);
2451 if (op_data == NULL) {
2452 CWARN("%s: cannot allocate op_data to release open handle for "
2453 DFID"\n", ll_s2sbi(sb)->ll_fsname, PFID(&body->mbo_fid1));
2458 op_data->op_fid1 = body->mbo_fid1;
2459 op_data->op_open_handle = body->mbo_open_handle;
2460 op_data->op_mod_time = ktime_get_real_seconds();
2461 md_close(exp, op_data, NULL, &close_req);
2462 ptlrpc_req_finished(close_req);
2463 ll_finish_md_op_data(op_data);
2468 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2469 struct super_block *sb, struct lookup_intent *it)
2471 struct ll_sb_info *sbi = NULL;
2472 struct lustre_md md = { NULL };
2473 bool default_lmv_deleted = false;
2478 LASSERT(*inode || sb);
2479 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2480 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2481 sbi->ll_md_exp, &md);
2486 * clear default_lmv only if intent_getattr reply doesn't contain it.
2487 * but it needs to be done after iget, check this early because
2488 * ll_update_lsm_md() may change md.
2490 if (it && (it->it_op & (IT_LOOKUP | IT_GETATTR)) &&
2491 S_ISDIR(md.body->mbo_mode) && !md.default_lmv)
2492 default_lmv_deleted = true;
2495 rc = ll_update_inode(*inode, &md);
2499 LASSERT(sb != NULL);
2502 * At this point server returns to client's same fid as client
2503 * generated for creating. So using ->fid1 is okay here.
2505 if (!fid_is_sane(&md.body->mbo_fid1)) {
2506 CERROR("%s: Fid is insane "DFID"\n",
2508 PFID(&md.body->mbo_fid1));
2509 GOTO(out, rc = -EINVAL);
2512 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2513 sbi->ll_flags & LL_SBI_32BIT_API),
2515 if (IS_ERR(*inode)) {
2516 #ifdef CONFIG_FS_POSIX_ACL
2518 posix_acl_release(md.posix_acl);
2519 md.posix_acl = NULL;
2522 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2524 CERROR("new_inode -fatal: rc %d\n", rc);
2529 /* Handling piggyback layout lock.
2530 * Layout lock can be piggybacked by getattr and open request.
2531 * The lsm can be applied to inode only if it comes with a layout lock
2532 * otherwise correct layout may be overwritten, for example:
2533 * 1. proc1: mdt returns a lsm but not granting layout
2534 * 2. layout was changed by another client
2535 * 3. proc2: refresh layout and layout lock granted
2536 * 4. proc1: to apply a stale layout */
2537 if (it != NULL && it->it_lock_mode != 0) {
2538 struct lustre_handle lockh;
2539 struct ldlm_lock *lock;
2541 lockh.cookie = it->it_lock_handle;
2542 lock = ldlm_handle2lock(&lockh);
2543 LASSERT(lock != NULL);
2544 if (ldlm_has_layout(lock)) {
2545 struct cl_object_conf conf;
2547 memset(&conf, 0, sizeof(conf));
2548 conf.coc_opc = OBJECT_CONF_SET;
2549 conf.coc_inode = *inode;
2550 conf.coc_lock = lock;
2551 conf.u.coc_layout = md.layout;
2552 (void)ll_layout_conf(*inode, &conf);
2554 LDLM_LOCK_PUT(lock);
2557 if (default_lmv_deleted)
2558 ll_update_default_lsm_md(*inode, &md);
2563 /* cleanup will be done if necessary */
2564 md_free_lustre_md(sbi->ll_md_exp, &md);
2566 if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2567 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2572 int ll_obd_statfs(struct inode *inode, void __user *arg)
2574 struct ll_sb_info *sbi = NULL;
2575 struct obd_export *exp;
2577 struct obd_ioctl_data *data = NULL;
2581 if (!inode || !(sbi = ll_i2sbi(inode)))
2582 GOTO(out_statfs, rc = -EINVAL);
2584 rc = obd_ioctl_getdata(&buf, &len, arg);
2586 GOTO(out_statfs, rc);
2589 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2590 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2591 GOTO(out_statfs, rc = -EINVAL);
2593 if (data->ioc_inllen1 != sizeof(__u32) ||
2594 data->ioc_inllen2 != sizeof(__u32) ||
2595 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2596 data->ioc_plen2 != sizeof(struct obd_uuid))
2597 GOTO(out_statfs, rc = -EINVAL);
2599 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2600 if (type & LL_STATFS_LMV)
2601 exp = sbi->ll_md_exp;
2602 else if (type & LL_STATFS_LOV)
2603 exp = sbi->ll_dt_exp;
2605 GOTO(out_statfs, rc = -ENODEV);
2607 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2609 GOTO(out_statfs, rc);
2611 OBD_FREE_LARGE(buf, len);
2616 * this is normally called in ll_fini_md_op_data(), but sometimes it needs to
2617 * be called early to avoid deadlock.
2619 void ll_unlock_md_op_lsm(struct md_op_data *op_data)
2621 if (op_data->op_mea2_sem) {
2622 up_read(op_data->op_mea2_sem);
2623 op_data->op_mea2_sem = NULL;
2626 if (op_data->op_mea1_sem) {
2627 up_read(op_data->op_mea1_sem);
2628 op_data->op_mea1_sem = NULL;
2632 /* this function prepares md_op_data hint for passing it down to MD stack. */
2633 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2634 struct inode *i1, struct inode *i2,
2635 const char *name, size_t namelen,
2636 __u32 mode, enum md_op_code opc,
2639 LASSERT(i1 != NULL);
2642 /* Do not reuse namelen for something else. */
2644 return ERR_PTR(-EINVAL);
2646 if (namelen > ll_i2sbi(i1)->ll_namelen)
2647 return ERR_PTR(-ENAMETOOLONG);
2649 if (!lu_name_is_valid_2(name, namelen))
2650 return ERR_PTR(-EINVAL);
2653 if (op_data == NULL)
2654 OBD_ALLOC_PTR(op_data);
2656 if (op_data == NULL)
2657 return ERR_PTR(-ENOMEM);
2659 ll_i2gids(op_data->op_suppgids, i1, i2);
2660 op_data->op_fid1 = *ll_inode2fid(i1);
2661 op_data->op_code = opc;
2663 if (S_ISDIR(i1->i_mode)) {
2664 down_read(&ll_i2info(i1)->lli_lsm_sem);
2665 op_data->op_mea1_sem = &ll_i2info(i1)->lli_lsm_sem;
2666 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2667 op_data->op_default_mea1 = ll_i2info(i1)->lli_default_lsm_md;
2671 op_data->op_fid2 = *ll_inode2fid(i2);
2672 if (S_ISDIR(i2->i_mode)) {
2674 down_read(&ll_i2info(i2)->lli_lsm_sem);
2675 op_data->op_mea2_sem =
2676 &ll_i2info(i2)->lli_lsm_sem;
2678 op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2681 fid_zero(&op_data->op_fid2);
2684 if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2685 op_data->op_cli_flags |= CLI_HASH64;
2687 if (ll_need_32bit_api(ll_i2sbi(i1)))
2688 op_data->op_cli_flags |= CLI_API32;
2690 op_data->op_name = name;
2691 op_data->op_namelen = namelen;
2692 op_data->op_mode = mode;
2693 op_data->op_mod_time = ktime_get_real_seconds();
2694 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2695 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2696 op_data->op_cap = cfs_curproc_cap_pack();
2697 op_data->op_mds = 0;
2698 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2699 filename_is_volatile(name, namelen, &op_data->op_mds)) {
2700 op_data->op_bias |= MDS_CREATE_VOLATILE;
2702 op_data->op_data = data;
2707 void ll_finish_md_op_data(struct md_op_data *op_data)
2709 ll_unlock_md_op_lsm(op_data);
2710 security_release_secctx(op_data->op_file_secctx,
2711 op_data->op_file_secctx_size);
2712 OBD_FREE_PTR(op_data);
2715 #ifdef HAVE_SUPEROPS_USE_DENTRY
2716 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2718 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2721 struct ll_sb_info *sbi;
2723 #ifdef HAVE_SUPEROPS_USE_DENTRY
2724 LASSERT((seq != NULL) && (dentry != NULL));
2725 sbi = ll_s2sbi(dentry->d_sb);
2727 LASSERT((seq != NULL) && (vfs != NULL));
2728 sbi = ll_s2sbi(vfs->mnt_sb);
2731 if (sbi->ll_flags & LL_SBI_NOLCK)
2732 seq_puts(seq, ",nolock");
2734 if (sbi->ll_flags & LL_SBI_FLOCK)
2735 seq_puts(seq, ",flock");
2737 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2738 seq_puts(seq, ",localflock");
2740 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2741 seq_puts(seq, ",user_xattr");
2743 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2744 seq_puts(seq, ",lazystatfs");
2746 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2747 seq_puts(seq, ",user_fid2path");
2749 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2750 seq_puts(seq, ",always_ping");
2756 * Get obd name by cmd, and copy out to user space
2758 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2760 struct ll_sb_info *sbi = ll_i2sbi(inode);
2761 struct obd_device *obd;
2764 if (cmd == OBD_IOC_GETDTNAME)
2765 obd = class_exp2obd(sbi->ll_dt_exp);
2766 else if (cmd == OBD_IOC_GETMDNAME)
2767 obd = class_exp2obd(sbi->ll_md_exp);
2774 if (copy_to_user((void __user *)arg, obd->obd_name,
2775 strlen(obd->obd_name) + 1))
2781 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2788 p.mnt = current->fs->root.mnt;
2790 path = d_path(&p, buf, bufsize);
2795 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2797 char *buf, *path = NULL;
2798 struct dentry *dentry = NULL;
2799 struct inode *inode = page->mapping->host;
2801 /* this can be called inside spin lock so use GFP_ATOMIC. */
2802 buf = (char *)__get_free_page(GFP_ATOMIC);
2804 dentry = d_find_alias(page->mapping->host);
2806 path = ll_d_path(dentry, buf, PAGE_SIZE);
2810 "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2811 "(rc %d)\n", ll_i2sbi(inode)->ll_fsname,
2812 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2813 PFID(ll_inode2fid(inode)),
2814 (path && !IS_ERR(path)) ? path : "", ioret);
2820 free_page((unsigned long)buf);
2823 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2824 struct lov_user_md **kbuf)
2826 struct lov_user_md lum;
2830 if (copy_from_user(&lum, md, sizeof(lum)))
2833 lum_size = ll_lov_user_md_size(&lum);
2837 OBD_ALLOC(*kbuf, lum_size);
2841 if (copy_from_user(*kbuf, md, lum_size) != 0) {
2842 OBD_FREE(*kbuf, lum_size);
2850 * Compute llite root squash state after a change of root squash
2851 * configuration setting or add/remove of a lnet nid
2853 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2855 struct root_squash_info *squash = &sbi->ll_squash;
2858 struct lnet_process_id id;
2860 /* Update norootsquash flag */
2861 down_write(&squash->rsi_sem);
2862 if (list_empty(&squash->rsi_nosquash_nids))
2863 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2865 /* Do not apply root squash as soon as one of our NIDs is
2866 * in the nosquash_nids list */
2869 while (LNetGetId(i++, &id) != -ENOENT) {
2870 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2872 if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2878 sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2880 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2882 up_write(&squash->rsi_sem);
2886 * Parse linkea content to extract information about a given hardlink
2888 * \param[in] ldata - Initialized linkea data
2889 * \param[in] linkno - Link identifier
2890 * \param[out] parent_fid - The entry's parent FID
2891 * \param[out] ln - Entry name destination buffer
2893 * \retval 0 on success
2894 * \retval Appropriate negative error code on failure
2896 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2897 struct lu_fid *parent_fid, struct lu_name *ln)
2903 rc = linkea_init_with_rec(ldata);
2907 if (linkno >= ldata->ld_leh->leh_reccount)
2908 /* beyond last link */
2911 linkea_first_entry(ldata);
2912 for (idx = 0; ldata->ld_lee != NULL; idx++) {
2913 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2918 linkea_next_entry(ldata);
2928 * Get parent FID and name of an identified link. Operation is performed for
2929 * a given link number, letting the caller iterate over linkno to list one or
2930 * all links of an entry.
2932 * \param[in] file - File descriptor against which to perform the operation
2933 * \param[in,out] arg - User-filled structure containing the linkno to operate
2934 * on and the available size. It is eventually filled with
2935 * the requested information or left untouched on error
2937 * \retval - 0 on success
2938 * \retval - Appropriate negative error code on failure
2940 int ll_getparent(struct file *file, struct getparent __user *arg)
2942 struct inode *inode = file_inode(file);
2943 struct linkea_data *ldata;
2944 struct lu_buf buf = LU_BUF_NULL;
2946 struct lu_fid parent_fid;
2953 if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2954 !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2957 if (get_user(name_size, &arg->gp_name_size))
2960 if (get_user(linkno, &arg->gp_linkno))
2963 if (name_size > PATH_MAX)
2966 OBD_ALLOC(ldata, sizeof(*ldata));
2970 rc = linkea_data_new(ldata, &buf);
2972 GOTO(ldata_free, rc);
2974 #ifdef HAVE_XATTR_HANDLER_FLAGS
2975 rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
2976 buf.lb_len, OBD_MD_FLXATTR);
2978 rc = ll_getxattr(file_dentry(file), XATTR_NAME_LINK, buf.lb_buf,
2980 #endif /* HAVE_XATTR_HANDLER_FLAGS */
2984 rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2988 if (ln.ln_namelen >= name_size)
2989 GOTO(lb_free, rc = -EOVERFLOW);
2991 if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2992 GOTO(lb_free, rc = -EFAULT);
2994 if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2995 GOTO(lb_free, rc = -EFAULT);
2997 if (put_user('\0', arg->gp_name + ln.ln_namelen))
2998 GOTO(lb_free, rc = -EFAULT);
3003 OBD_FREE(ldata, sizeof(*ldata));