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_DIR_MIGRATE |
228 OBD_CONNECT2_SUM_STATFS |
230 OBD_CONNECT2_LOCK_CONVERT |
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) {
598 CERROR("%s: can't make root dentry: rc = %d\n",
599 sbi->ll_fsname, err);
602 #ifdef HAVE_DCACHE_LOCK
603 sb->s_root->d_op = &ll_d_ops;
606 sbi->ll_sdev_orig = sb->s_dev;
608 /* We set sb->s_dev equal on all lustre clients in order to support
609 * NFS export clustering. NFSD requires that the FSID be the same
611 /* s_dev is also used in lt_compare() to compare two fs, but that is
612 * only a node-local comparison. */
613 uuid = obd_get_uuid(sbi->ll_md_exp);
615 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
622 if (sbi->ll_dt_obd) {
623 err = sysfs_create_link(&sbi->ll_kset.kobj,
624 &sbi->ll_dt_obd->obd_kset.kobj,
625 sbi->ll_dt_obd->obd_type->typ_name);
627 CERROR("%s: could not register %s in llite: rc = %d\n",
628 dt, sbi->ll_fsname, err);
633 if (sbi->ll_md_obd) {
634 err = sysfs_create_link(&sbi->ll_kset.kobj,
635 &sbi->ll_md_obd->obd_kset.kobj,
636 sbi->ll_md_obd->obd_type->typ_name);
638 CERROR("%s: could not register %s in llite: rc = %d\n",
639 md, sbi->ll_fsname, err);
649 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
651 obd_disconnect(sbi->ll_dt_exp);
652 sbi->ll_dt_exp = NULL;
653 sbi->ll_dt_obd = NULL;
655 obd_fid_fini(sbi->ll_md_exp->exp_obd);
657 obd_disconnect(sbi->ll_md_exp);
658 sbi->ll_md_exp = NULL;
659 sbi->ll_md_obd = NULL;
668 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
672 size = sizeof(*lmmsize);
673 rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
674 KEY_MAX_EASIZE, &size, lmmsize);
676 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
677 sbi->ll_dt_exp->exp_obd->obd_name, rc);
681 CDEBUG(D_INFO, "max LOV ea size: %d\n", *lmmsize);
684 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
685 KEY_MAX_EASIZE, &size, lmmsize);
687 CERROR("Get max mdsize error rc %d\n", rc);
689 CDEBUG(D_INFO, "max LMV ea size: %d\n", *lmmsize);
695 * Get the value of the default_easize parameter.
697 * \see client_obd::cl_default_mds_easize
699 * \param[in] sbi superblock info for this filesystem
700 * \param[out] lmmsize pointer to storage location for value
702 * \retval 0 on success
703 * \retval negative negated errno on failure
705 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
710 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
711 KEY_DEFAULT_EASIZE, &size, lmmsize);
713 CERROR("Get default mdsize error rc %d\n", rc);
719 * Set the default_easize parameter to the given value.
721 * \see client_obd::cl_default_mds_easize
723 * \param[in] sbi superblock info for this filesystem
724 * \param[in] lmmsize the size to set
726 * \retval 0 on success
727 * \retval negative negated errno on failure
729 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
733 if (lmmsize < sizeof(struct lov_mds_md) ||
734 lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
737 rc = obd_set_info_async(NULL, sbi->ll_md_exp,
738 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
739 sizeof(int), &lmmsize, NULL);
744 static void client_common_put_super(struct super_block *sb)
746 struct ll_sb_info *sbi = ll_s2sbi(sb);
751 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
752 obd_disconnect(sbi->ll_dt_exp);
753 sbi->ll_dt_exp = NULL;
755 ll_debugfs_unregister_super(sb);
757 obd_fid_fini(sbi->ll_md_exp->exp_obd);
758 obd_disconnect(sbi->ll_md_exp);
759 sbi->ll_md_exp = NULL;
764 void ll_kill_super(struct super_block *sb)
766 struct ll_sb_info *sbi;
770 if (!(sb->s_flags & MS_ACTIVE))
774 /* we need restore s_dev from changed for clustred NFS before put_super
775 * because new kernels have cached s_dev and change sb->s_dev in
776 * put_super not affected real removing devices */
778 sb->s_dev = sbi->ll_sdev_orig;
779 sbi->ll_umounting = 1;
781 /* wait running statahead threads to quit */
782 while (atomic_read(&sbi->ll_sa_running) > 0) {
783 set_current_state(TASK_UNINTERRUPTIBLE);
784 schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
791 static inline int ll_set_opt(const char *opt, char *data, int fl)
793 if (strncmp(opt, data, strlen(opt)) != 0)
799 /* non-client-specific mount options are parsed in lmd_parse */
800 static int ll_options(char *options, struct ll_sb_info *sbi)
803 char *s1 = options, *s2;
804 int *flags = &sbi->ll_flags;
810 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
813 CDEBUG(D_SUPER, "next opt=%s\n", s1);
814 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
819 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
824 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
829 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
834 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
839 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
844 tmp = ll_set_opt("context", s1, 1);
847 tmp = ll_set_opt("fscontext", s1, 1);
850 tmp = ll_set_opt("defcontext", s1, 1);
853 tmp = ll_set_opt("rootcontext", s1, 1);
856 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
861 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
867 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
870 sbi->ll_checksum_set = 1;
873 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
876 sbi->ll_checksum_set = 1;
879 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
884 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
889 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
894 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
899 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
904 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
909 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
914 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
919 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
925 s2 = strchr(s1, ',');
933 void ll_lli_init(struct ll_inode_info *lli)
935 lli->lli_inode_magic = LLI_INODE_MAGIC;
937 spin_lock_init(&lli->lli_lock);
938 lli->lli_posix_acl = NULL;
939 /* Do not set lli_fid, it has been initialized already. */
940 fid_zero(&lli->lli_pfid);
941 lli->lli_mds_read_och = NULL;
942 lli->lli_mds_write_och = NULL;
943 lli->lli_mds_exec_och = NULL;
944 lli->lli_open_fd_read_count = 0;
945 lli->lli_open_fd_write_count = 0;
946 lli->lli_open_fd_exec_count = 0;
947 mutex_init(&lli->lli_och_mutex);
948 spin_lock_init(&lli->lli_agl_lock);
949 spin_lock_init(&lli->lli_layout_lock);
950 ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
951 lli->lli_clob = NULL;
953 init_rwsem(&lli->lli_xattrs_list_rwsem);
954 mutex_init(&lli->lli_xattrs_enq_lock);
956 LASSERT(lli->lli_vfs_inode.i_mode != 0);
957 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
958 mutex_init(&lli->lli_readdir_mutex);
959 lli->lli_opendir_key = NULL;
961 spin_lock_init(&lli->lli_sa_lock);
962 lli->lli_opendir_pid = 0;
963 lli->lli_sa_enabled = 0;
964 lli->lli_def_stripe_offset = -1;
965 init_rwsem(&lli->lli_lsm_sem);
967 mutex_init(&lli->lli_size_mutex);
968 lli->lli_symlink_name = NULL;
969 init_rwsem(&lli->lli_trunc_sem);
970 range_lock_tree_init(&lli->lli_write_tree);
971 init_rwsem(&lli->lli_glimpse_sem);
972 lli->lli_glimpse_time = ktime_set(0, 0);
973 INIT_LIST_HEAD(&lli->lli_agl_list);
974 lli->lli_agl_index = 0;
975 lli->lli_async_rc = 0;
976 spin_lock_init(&lli->lli_heat_lock);
977 obd_heat_clear(lli->lli_heat_instances, OBD_HEAT_COUNT);
978 lli->lli_heat_flags = 0;
980 mutex_init(&lli->lli_layout_mutex);
981 memset(lli->lli_jobid, 0, sizeof(lli->lli_jobid));
984 #define MAX_STRING_SIZE 128
986 #ifndef HAVE_SUPER_SETUP_BDI_NAME
988 #define LSI_BDI_INITIALIZED 0x00400000
990 #ifndef HAVE_BDI_CAP_MAP_COPY
991 # define BDI_CAP_MAP_COPY 0
994 static int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
996 struct lustre_sb_info *lsi = s2lsi(sb);
997 char buf[MAX_STRING_SIZE];
1001 err = bdi_init(&lsi->lsi_bdi);
1005 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1006 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1007 lsi->lsi_bdi.name = "lustre";
1008 va_start(args, fmt);
1009 vsnprintf(buf, MAX_STRING_SIZE, fmt, args);
1011 err = bdi_register(&lsi->lsi_bdi, NULL, "%s", buf);
1014 sb->s_bdi = &lsi->lsi_bdi;
1018 #endif /* !HAVE_SUPER_SETUP_BDI_NAME */
1020 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
1022 struct lustre_profile *lprof = NULL;
1023 struct lustre_sb_info *lsi = s2lsi(sb);
1024 struct ll_sb_info *sbi = NULL;
1025 char *dt = NULL, *md = NULL;
1026 char *profilenm = get_profile_name(sb);
1027 struct config_llog_instance *cfg;
1028 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1029 const int instlen = 16 + 2;
1030 unsigned long cfg_instance = ll_get_cfg_instance(sb);
1031 char name[MAX_STRING_SIZE];
1039 /* for ASLR, to map between cfg_instance and hashed ptr */
1040 CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1041 profilenm, cfg_instance, sb);
1043 try_module_get(THIS_MODULE);
1047 GOTO(out_free_cfg, err = -ENOMEM);
1049 /* client additional sb info */
1050 lsi->lsi_llsbi = sbi = ll_init_sbi();
1052 GOTO(out_free_cfg, err = -ENOMEM);
1054 err = ll_options(lsi->lsi_lmd->lmd_opts, sbi);
1056 GOTO(out_free_cfg, err);
1058 err = super_setup_bdi_name(sb, "lustre-%016lx", cfg_instance);
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 len = strlen(profilenm);
1068 ptr = strrchr(profilenm, '-');
1069 if (ptr && (strcmp(ptr, "-client") == 0))
1072 if (len > LUSTRE_MAXFSNAME) {
1073 if (unlikely(len >= MAX_STRING_SIZE))
1074 len = MAX_STRING_SIZE - 1;
1075 strncpy(name, profilenm, len);
1077 err = -ENAMETOOLONG;
1078 CERROR("%s: fsname longer than %u characters: rc = %d\n",
1079 name, LUSTRE_MAXFSNAME, err);
1080 GOTO(out_free_cfg, err);
1082 strncpy(sbi->ll_fsname, profilenm, len);
1083 sbi->ll_fsname[len] = '\0';
1086 snprintf(name, MAX_STRING_SIZE, "%.*s-%016lx", len,
1087 profilenm, cfg_instance);
1089 /* Call ll_debugfs_register_super() before lustre_process_log()
1090 * so that "llite.*.*" params can be processed correctly.
1092 err = ll_debugfs_register_super(sb, name);
1094 CERROR("%s: could not register mountpoint in llite: rc = %d\n",
1095 sbi->ll_fsname, err);
1099 /* The cfg_instance is a value unique to this super, in case some
1100 * joker tries to mount the same fs at two mount points.
1102 cfg->cfg_instance = cfg_instance;
1103 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1104 cfg->cfg_callback = class_config_llog_handler;
1105 cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1106 /* set up client obds */
1107 err = lustre_process_log(sb, profilenm, cfg);
1109 GOTO(out_debugfs, err);
1111 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1112 lprof = class_get_profile(profilenm);
1113 if (lprof == NULL) {
1114 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1115 " read from the MGS. Does that filesystem "
1116 "exist?\n", profilenm);
1117 GOTO(out_debugfs, err = -EINVAL);
1119 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1120 lprof->lp_md, lprof->lp_dt);
1122 dt_len = strlen(lprof->lp_dt) + instlen + 2;
1123 OBD_ALLOC(dt, dt_len);
1125 GOTO(out_profile, err = -ENOMEM);
1126 snprintf(dt, dt_len - 1, "%s-%016lx", lprof->lp_dt, cfg_instance);
1128 md_len = strlen(lprof->lp_md) + instlen + 2;
1129 OBD_ALLOC(md, md_len);
1131 GOTO(out_free_dt, err = -ENOMEM);
1132 snprintf(md, md_len - 1, "%s-%016lx", lprof->lp_md, cfg_instance);
1134 /* connections, registrations, sb setup */
1135 err = client_common_fill_super(sb, md, dt, mnt);
1137 GOTO(out_free_md, err);
1139 sbi->ll_client_common_fill_super_succeeded = 1;
1143 OBD_FREE(md, md_len);
1146 OBD_FREE(dt, dt_len);
1149 class_put_profile(lprof);
1152 ll_debugfs_unregister_super(sb);
1159 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1160 LCONSOLE_WARN("Mounted %s\n", profilenm);
1162 } /* ll_fill_super */
1164 void ll_put_super(struct super_block *sb)
1166 struct config_llog_instance cfg, params_cfg;
1167 struct obd_device *obd;
1168 struct lustre_sb_info *lsi = s2lsi(sb);
1169 struct ll_sb_info *sbi = ll_s2sbi(sb);
1170 char *profilenm = get_profile_name(sb);
1171 unsigned long cfg_instance = ll_get_cfg_instance(sb);
1173 int next, force = 1, rc = 0;
1177 GOTO(out_no_sbi, 0);
1179 /* Should replace instance_id with something better for ASLR */
1180 CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1181 profilenm, cfg_instance, sb);
1183 cfg.cfg_instance = cfg_instance;
1184 lustre_end_log(sb, profilenm, &cfg);
1186 params_cfg.cfg_instance = cfg_instance;
1187 lustre_end_log(sb, PARAMS_FILENAME, ¶ms_cfg);
1189 if (sbi->ll_md_exp) {
1190 obd = class_exp2obd(sbi->ll_md_exp);
1192 force = obd->obd_force;
1195 /* Wait for unstable pages to be committed to stable storage */
1197 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1198 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1199 atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1203 ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1204 if (force == 0 && rc != -EINTR)
1205 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1207 /* We need to set force before the lov_disconnect in
1208 lustre_common_put_super, since l_d cleans up osc's as well. */
1211 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1213 obd->obd_force = force;
1217 if (sbi->ll_client_common_fill_super_succeeded) {
1218 /* Only if client_common_fill_super succeeded */
1219 client_common_put_super(sb);
1223 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1224 class_manual_cleanup(obd);
1227 if (sbi->ll_flags & LL_SBI_VERBOSE)
1228 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1231 class_del_profile(profilenm);
1233 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1234 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1235 bdi_destroy(&lsi->lsi_bdi);
1236 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1241 lsi->lsi_llsbi = NULL;
1243 lustre_common_put_super(sb);
1245 cl_env_cache_purge(~0);
1247 module_put(THIS_MODULE);
1250 } /* client_put_super */
1252 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1254 struct inode *inode = NULL;
1256 /* NOTE: we depend on atomic igrab() -bzzz */
1257 lock_res_and_lock(lock);
1258 if (lock->l_resource->lr_lvb_inode) {
1259 struct ll_inode_info * lli;
1260 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1261 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1262 inode = igrab(lock->l_resource->lr_lvb_inode);
1264 inode = lock->l_resource->lr_lvb_inode;
1265 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1266 D_WARNING, lock, "lr_lvb_inode %p is "
1267 "bogus: magic %08x",
1268 lock->l_resource->lr_lvb_inode,
1269 lli->lli_inode_magic);
1273 unlock_res_and_lock(lock);
1277 void ll_dir_clear_lsm_md(struct inode *inode)
1279 struct ll_inode_info *lli = ll_i2info(inode);
1281 LASSERT(S_ISDIR(inode->i_mode));
1283 if (lli->lli_lsm_md != NULL) {
1284 lmv_free_memmd(lli->lli_lsm_md);
1285 lli->lli_lsm_md = NULL;
1289 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1290 const struct lu_fid *fid,
1291 struct lustre_md *md)
1293 struct ll_sb_info *sbi = ll_s2sbi(sb);
1294 struct mdt_body *body = md->body;
1295 struct inode *inode;
1299 ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1300 inode = iget_locked(sb, ino);
1301 if (inode == NULL) {
1302 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1303 sbi->ll_fsname, PFID(fid));
1304 RETURN(ERR_PTR(-ENOENT));
1307 if (inode->i_state & I_NEW) {
1308 struct ll_inode_info *lli = ll_i2info(inode);
1309 struct lmv_stripe_md *lsm = md->lmv;
1311 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1312 (body->mbo_mode & S_IFMT);
1313 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1316 inode->i_mtime.tv_sec = 0;
1317 inode->i_atime.tv_sec = 0;
1318 inode->i_ctime.tv_sec = 0;
1321 #ifdef HAVE_BACKING_DEV_INFO
1322 /* initializing backing dev info. */
1323 inode->i_mapping->backing_dev_info =
1324 &s2lsi(inode->i_sb)->lsi_bdi;
1326 inode->i_op = &ll_dir_inode_operations;
1327 inode->i_fop = &ll_dir_operations;
1328 lli->lli_fid = *fid;
1331 LASSERT(lsm != NULL);
1332 /* master object FID */
1333 lli->lli_pfid = body->mbo_fid1;
1334 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1335 lli, PFID(fid), PFID(&lli->lli_pfid));
1336 unlock_new_inode(inode);
1342 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1345 struct lmv_stripe_md *lsm = md->lmv;
1346 struct ll_inode_info *lli = ll_i2info(inode);
1349 LASSERT(lsm != NULL);
1351 CDEBUG(D_INODE, "%s: "DFID" set dir layout:\n",
1352 ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid));
1353 lsm_md_dump(D_INODE, lsm);
1355 /* XXX sigh, this lsm_root initialization should be in
1356 * LMV layer, but it needs ll_iget right now, so we
1357 * put this here right now. */
1358 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1359 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1360 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1361 /* Unfortunately ll_iget will call ll_update_inode,
1362 * where the initialization of slave inode is slightly
1363 * different, so it reset lsm_md to NULL to avoid
1364 * initializing lsm for slave inode. */
1365 lsm->lsm_md_oinfo[i].lmo_root =
1366 ll_iget_anon_dir(inode->i_sb, fid, md);
1367 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1368 int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1370 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1372 iput(lsm->lsm_md_oinfo[i].lmo_root);
1373 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1379 lli->lli_lsm_md = lsm;
1384 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1386 struct ll_inode_info *lli = ll_i2info(inode);
1387 struct lmv_stripe_md *lsm = md->lmv;
1392 LASSERT(S_ISDIR(inode->i_mode));
1393 CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1394 PFID(ll_inode2fid(inode)));
1397 * no striped information from request, lustre_md from req does not
1398 * include stripeEA, see ll_md_setattr()
1404 * normally dir layout doesn't change, only take read lock to check
1405 * that to avoid blocking other MD operations.
1407 if (lli->lli_lsm_md)
1408 down_read(&lli->lli_lsm_sem);
1410 down_write(&lli->lli_lsm_sem);
1413 * if dir layout mismatch, check whether version is increased, which
1414 * means layout is changed, this happens in dir migration and lfsck.
1416 * foreign LMV should not change.
1418 if (lli->lli_lsm_md &&
1419 lli->lli_lsm_md->lsm_md_magic != LMV_MAGIC_FOREIGN &&
1420 !lsm_md_eq(lli->lli_lsm_md, lsm)) {
1421 if (lsm->lsm_md_layout_version <=
1422 lli->lli_lsm_md->lsm_md_layout_version) {
1423 CERROR("%s: "DFID" dir layout mismatch:\n",
1424 ll_i2sbi(inode)->ll_fsname,
1425 PFID(&lli->lli_fid));
1426 lsm_md_dump(D_ERROR, lli->lli_lsm_md);
1427 lsm_md_dump(D_ERROR, lsm);
1428 GOTO(unlock, rc = -EINVAL);
1431 /* layout changed, switch to write lock */
1432 up_read(&lli->lli_lsm_sem);
1433 down_write(&lli->lli_lsm_sem);
1434 ll_dir_clear_lsm_md(inode);
1437 /* set directory layout */
1438 if (!lli->lli_lsm_md) {
1439 struct cl_attr *attr;
1441 if (lsm->lsm_md_magic == LMV_MAGIC_FOREIGN) {
1442 /* set md->lmv to NULL, so the following free lustre_md
1443 * will not free this lsm */
1445 lli->lli_lsm_md = lsm;
1446 up_write(&lli->lli_lsm_sem);
1450 rc = ll_init_lsm_md(inode, md);
1451 up_write(&lli->lli_lsm_sem);
1455 /* set md->lmv to NULL, so the following free lustre_md
1456 * will not free this lsm */
1460 * md_merge_attr() may take long, since lsm is already set,
1461 * switch to read lock.
1463 down_read(&lli->lli_lsm_sem);
1465 OBD_ALLOC_PTR(attr);
1467 GOTO(unlock, rc = -ENOMEM);
1469 /* validate the lsm */
1470 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1471 ll_md_blocking_ast);
1477 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1478 md->body->mbo_nlink = attr->cat_nlink;
1479 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1480 md->body->mbo_size = attr->cat_size;
1481 if (md->body->mbo_valid & OBD_MD_FLATIME)
1482 md->body->mbo_atime = attr->cat_atime;
1483 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1484 md->body->mbo_ctime = attr->cat_ctime;
1485 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1486 md->body->mbo_mtime = attr->cat_mtime;
1491 up_read(&lli->lli_lsm_sem);
1496 void ll_clear_inode(struct inode *inode)
1498 struct ll_inode_info *lli = ll_i2info(inode);
1499 struct ll_sb_info *sbi = ll_i2sbi(inode);
1502 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1503 PFID(ll_inode2fid(inode)), inode);
1505 if (S_ISDIR(inode->i_mode)) {
1506 /* these should have been cleared in ll_file_release */
1507 LASSERT(lli->lli_opendir_key == NULL);
1508 LASSERT(lli->lli_sai == NULL);
1509 LASSERT(lli->lli_opendir_pid == 0);
1512 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1514 LASSERT(!lli->lli_open_fd_write_count);
1515 LASSERT(!lli->lli_open_fd_read_count);
1516 LASSERT(!lli->lli_open_fd_exec_count);
1518 if (lli->lli_mds_write_och)
1519 ll_md_real_close(inode, FMODE_WRITE);
1520 if (lli->lli_mds_exec_och)
1521 ll_md_real_close(inode, FMODE_EXEC);
1522 if (lli->lli_mds_read_och)
1523 ll_md_real_close(inode, FMODE_READ);
1525 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1526 OBD_FREE(lli->lli_symlink_name,
1527 strlen(lli->lli_symlink_name) + 1);
1528 lli->lli_symlink_name = NULL;
1531 ll_xattr_cache_destroy(inode);
1533 #ifdef CONFIG_FS_POSIX_ACL
1534 forget_all_cached_acls(inode);
1535 if (lli->lli_posix_acl) {
1536 posix_acl_release(lli->lli_posix_acl);
1537 lli->lli_posix_acl = NULL;
1540 lli->lli_inode_magic = LLI_INODE_DEAD;
1542 if (S_ISDIR(inode->i_mode))
1543 ll_dir_clear_lsm_md(inode);
1544 else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1545 LASSERT(list_empty(&lli->lli_agl_list));
1548 * XXX This has to be done before lsm is freed below, because
1549 * cl_object still uses inode lsm.
1551 cl_inode_fini(inode);
1556 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1558 struct lustre_md md;
1559 struct inode *inode = dentry->d_inode;
1560 struct ll_sb_info *sbi = ll_i2sbi(inode);
1561 struct ptlrpc_request *request = NULL;
1565 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1566 LUSTRE_OPC_ANY, NULL);
1567 if (IS_ERR(op_data))
1568 RETURN(PTR_ERR(op_data));
1570 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1572 ptlrpc_req_finished(request);
1573 if (rc == -ENOENT) {
1575 /* Unlinked special device node? Or just a race?
1576 * Pretend we done everything. */
1577 if (!S_ISREG(inode->i_mode) &&
1578 !S_ISDIR(inode->i_mode)) {
1579 ia_valid = op_data->op_attr.ia_valid;
1580 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1581 rc = simple_setattr(dentry, &op_data->op_attr);
1582 op_data->op_attr.ia_valid = ia_valid;
1584 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1585 CERROR("md_setattr fails: rc = %d\n", rc);
1590 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1591 sbi->ll_md_exp, &md);
1593 ptlrpc_req_finished(request);
1597 ia_valid = op_data->op_attr.ia_valid;
1598 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1599 * cache is not cleared yet. */
1600 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1601 if (S_ISREG(inode->i_mode))
1603 rc = simple_setattr(dentry, &op_data->op_attr);
1604 if (S_ISREG(inode->i_mode))
1605 inode_unlock(inode);
1606 op_data->op_attr.ia_valid = ia_valid;
1608 rc = ll_update_inode(inode, &md);
1609 ptlrpc_req_finished(request);
1614 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1615 * object(s) determine the file size and mtime. Otherwise, the MDS will
1616 * keep these values until such a time that objects are allocated for it.
1617 * We do the MDS operations first, as it is checking permissions for us.
1618 * We don't to the MDS RPC if there is nothing that we want to store there,
1619 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1620 * going to do an RPC anyways.
1622 * If we are doing a truncate, we will send the mtime and ctime updates
1623 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1624 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1627 * In case of HSMimport, we only set attr on MDS.
1629 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr,
1630 enum op_xvalid xvalid, bool hsm_import)
1632 struct inode *inode = dentry->d_inode;
1633 struct ll_inode_info *lli = ll_i2info(inode);
1634 struct md_op_data *op_data = NULL;
1638 CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1639 "valid %x, hsm_import %d\n",
1640 ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid),
1641 inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1644 if (attr->ia_valid & ATTR_SIZE) {
1645 /* Check new size against VFS/VM file size limit and rlimit */
1646 rc = inode_newsize_ok(inode, attr->ia_size);
1650 /* The maximum Lustre file size is variable, based on the
1651 * OST maximum object size and number of stripes. This
1652 * needs another check in addition to the VFS check above. */
1653 if (attr->ia_size > ll_file_maxbytes(inode)) {
1654 CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
1655 PFID(&lli->lli_fid), attr->ia_size,
1656 ll_file_maxbytes(inode));
1660 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1663 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1664 if (attr->ia_valid & TIMES_SET_FLAGS) {
1665 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1666 !cfs_capable(CFS_CAP_FOWNER))
1670 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1671 if (!(xvalid & OP_XVALID_CTIME_SET) &&
1672 (attr->ia_valid & ATTR_CTIME)) {
1673 attr->ia_ctime = current_time(inode);
1674 xvalid |= OP_XVALID_CTIME_SET;
1676 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1677 (attr->ia_valid & ATTR_ATIME)) {
1678 attr->ia_atime = current_time(inode);
1679 attr->ia_valid |= ATTR_ATIME_SET;
1681 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1682 (attr->ia_valid & ATTR_MTIME)) {
1683 attr->ia_mtime = current_time(inode);
1684 attr->ia_valid |= ATTR_MTIME_SET;
1687 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1688 CDEBUG(D_INODE, "setting mtime %lld, ctime %lld, now = %lld\n",
1689 (s64)attr->ia_mtime.tv_sec, (s64)attr->ia_ctime.tv_sec,
1690 ktime_get_real_seconds());
1692 if (S_ISREG(inode->i_mode)) {
1693 if (attr->ia_valid & ATTR_SIZE)
1694 inode_dio_write_done(inode);
1695 inode_unlock(inode);
1698 /* We always do an MDS RPC, even if we're only changing the size;
1699 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1701 OBD_ALLOC_PTR(op_data);
1702 if (op_data == NULL)
1703 GOTO(out, rc = -ENOMEM);
1705 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1706 /* If we are changing file size, file content is
1707 * modified, flag it.
1709 xvalid |= OP_XVALID_OWNEROVERRIDE;
1710 op_data->op_bias |= MDS_DATA_MODIFIED;
1711 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1714 if (attr->ia_valid & ATTR_FILE) {
1715 struct ll_file_data *fd = LUSTRE_FPRIVATE(attr->ia_file);
1717 if (fd->fd_lease_och)
1718 op_data->op_bias |= MDS_TRUNC_KEEP_LEASE;
1721 op_data->op_attr = *attr;
1722 op_data->op_xvalid = xvalid;
1724 rc = ll_md_setattr(dentry, op_data);
1728 if (!S_ISREG(inode->i_mode) || hsm_import)
1731 if (attr->ia_valid & (ATTR_SIZE | ATTR_ATIME | ATTR_ATIME_SET |
1732 ATTR_MTIME | ATTR_MTIME_SET | ATTR_CTIME) ||
1733 xvalid & OP_XVALID_CTIME_SET) {
1734 /* For truncate and utimes sending attributes to OSTs, setting
1735 * mtime/atime to the past will be performed under PW [0:EOF]
1736 * extent lock (new_size:EOF for truncate). It may seem
1737 * excessive to send mtime/atime updates to OSTs when not
1738 * setting times to past, but it is necessary due to possible
1739 * time de-synchronization between MDT inode and OST objects
1741 rc = cl_setattr_ost(lli->lli_clob, attr, xvalid, 0);
1744 /* If the file was restored, it needs to set dirty flag.
1746 * We've already sent MDS_DATA_MODIFIED flag in
1747 * ll_md_setattr() for truncate. However, the MDT refuses to
1748 * set the HS_DIRTY flag on released files, so we have to set
1749 * it again if the file has been restored. Please check how
1750 * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1752 * Please notice that if the file is not released, the previous
1753 * MDS_DATA_MODIFIED has taken effect and usually
1754 * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1755 * This way we can save an RPC for common open + trunc
1757 if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1758 struct hsm_state_set hss = {
1759 .hss_valid = HSS_SETMASK,
1760 .hss_setmask = HS_DIRTY,
1764 rc2 = ll_hsm_state_set(inode, &hss);
1765 /* truncate and write can happen at the same time, so that
1766 * the file can be set modified even though the file is not
1767 * restored from released state, and ll_hsm_state_set() is
1768 * not applicable for the file, and rc2 < 0 is normal in this
1771 CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1772 PFID(ll_inode2fid(inode)), rc2);
1777 if (op_data != NULL)
1778 ll_finish_md_op_data(op_data);
1780 if (S_ISREG(inode->i_mode)) {
1782 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1783 inode_dio_wait(inode);
1784 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
1785 * flag. ll_update_inode (called from ll_md_setattr), clears
1786 * inode flags, so there is a gap where S_NOSEC is not set.
1787 * This can cause a writer to take the i_mutex unnecessarily,
1788 * but this is safe to do and should be rare. */
1789 inode_has_no_xattr(inode);
1792 ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1793 LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1798 int ll_setattr(struct dentry *de, struct iattr *attr)
1800 int mode = de->d_inode->i_mode;
1801 enum op_xvalid xvalid = 0;
1803 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1804 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1805 xvalid |= OP_XVALID_OWNEROVERRIDE;
1807 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1808 (ATTR_SIZE|ATTR_MODE)) &&
1809 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1810 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1811 !(attr->ia_mode & S_ISGID))))
1812 attr->ia_valid |= ATTR_FORCE;
1814 if ((attr->ia_valid & ATTR_MODE) &&
1816 !(attr->ia_mode & S_ISUID) &&
1817 !(attr->ia_valid & ATTR_KILL_SUID))
1818 attr->ia_valid |= ATTR_KILL_SUID;
1820 if ((attr->ia_valid & ATTR_MODE) &&
1821 ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1822 !(attr->ia_mode & S_ISGID) &&
1823 !(attr->ia_valid & ATTR_KILL_SGID))
1824 attr->ia_valid |= ATTR_KILL_SGID;
1826 return ll_setattr_raw(de, attr, xvalid, false);
1829 int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
1832 struct obd_statfs obd_osfs = { 0 };
1837 max_age = ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS;
1839 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1843 osfs->os_type = LL_SUPER_MAGIC;
1845 CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1846 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree, osfs->os_files);
1848 if (osfs->os_state & OS_STATE_SUM)
1851 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1852 flags |= OBD_STATFS_NODELAY;
1854 rc = obd_statfs(NULL, sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1855 if (rc) /* Possibly a filesystem with no OSTs. Report MDT totals. */
1858 CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1859 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1862 osfs->os_bsize = obd_osfs.os_bsize;
1863 osfs->os_blocks = obd_osfs.os_blocks;
1864 osfs->os_bfree = obd_osfs.os_bfree;
1865 osfs->os_bavail = obd_osfs.os_bavail;
1867 /* If we have _some_ OSTs, but don't have as many free objects on the
1868 * OSTs as inodes on the MDTs, reduce the reported number of inodes
1869 * to compensate, so that the "inodes in use" number is correct.
1870 * This should be kept in sync with lod_statfs() behaviour.
1872 if (obd_osfs.os_files && obd_osfs.os_ffree < osfs->os_ffree) {
1873 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1875 osfs->os_ffree = obd_osfs.os_ffree;
1881 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1883 struct super_block *sb = de->d_sb;
1884 struct obd_statfs osfs;
1885 __u64 fsid = huge_encode_dev(sb->s_dev);
1888 CDEBUG(D_VFSTRACE, "VFS Op: at %llu jiffies\n", get_jiffies_64());
1889 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1891 /* Some amount of caching on the client is allowed */
1892 rc = ll_statfs_internal(ll_s2sbi(sb), &osfs, OBD_STATFS_SUM);
1896 statfs_unpack(sfs, &osfs);
1898 /* We need to downshift for all 32-bit kernels, because we can't
1899 * tell if the kernel is being called via sys_statfs64() or not.
1900 * Stop before overflowing f_bsize - in which case it is better
1901 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1902 if (sizeof(long) < 8) {
1903 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1906 osfs.os_blocks >>= 1;
1907 osfs.os_bfree >>= 1;
1908 osfs.os_bavail >>= 1;
1912 sfs->f_blocks = osfs.os_blocks;
1913 sfs->f_bfree = osfs.os_bfree;
1914 sfs->f_bavail = osfs.os_bavail;
1915 sfs->f_fsid.val[0] = (__u32)fsid;
1916 sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1920 void ll_inode_size_lock(struct inode *inode)
1922 struct ll_inode_info *lli;
1924 LASSERT(!S_ISDIR(inode->i_mode));
1926 lli = ll_i2info(inode);
1927 mutex_lock(&lli->lli_size_mutex);
1930 void ll_inode_size_unlock(struct inode *inode)
1932 struct ll_inode_info *lli;
1934 lli = ll_i2info(inode);
1935 mutex_unlock(&lli->lli_size_mutex);
1938 void ll_update_inode_flags(struct inode *inode, int ext_flags)
1940 inode->i_flags = ll_ext_to_inode_flags(ext_flags);
1941 if (ext_flags & LUSTRE_PROJINHERIT_FL)
1942 ll_file_set_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
1944 ll_file_clear_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
1947 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1949 struct ll_inode_info *lli = ll_i2info(inode);
1950 struct mdt_body *body = md->body;
1951 struct ll_sb_info *sbi = ll_i2sbi(inode);
1954 if (body->mbo_valid & OBD_MD_FLEASIZE) {
1955 rc = cl_file_inode_init(inode, md);
1960 if (S_ISDIR(inode->i_mode)) {
1961 rc = ll_update_lsm_md(inode, md);
1966 #ifdef CONFIG_FS_POSIX_ACL
1967 if (body->mbo_valid & OBD_MD_FLACL) {
1968 spin_lock(&lli->lli_lock);
1969 if (lli->lli_posix_acl)
1970 posix_acl_release(lli->lli_posix_acl);
1971 lli->lli_posix_acl = md->posix_acl;
1972 spin_unlock(&lli->lli_lock);
1975 inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1976 sbi->ll_flags & LL_SBI_32BIT_API);
1977 inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1979 if (body->mbo_valid & OBD_MD_FLATIME) {
1980 if (body->mbo_atime > inode->i_atime.tv_sec)
1981 inode->i_atime.tv_sec = body->mbo_atime;
1982 lli->lli_atime = body->mbo_atime;
1985 if (body->mbo_valid & OBD_MD_FLMTIME) {
1986 if (body->mbo_mtime > inode->i_mtime.tv_sec) {
1988 "setting ino %lu mtime from %lld to %llu\n",
1989 inode->i_ino, (s64)inode->i_mtime.tv_sec,
1991 inode->i_mtime.tv_sec = body->mbo_mtime;
1993 lli->lli_mtime = body->mbo_mtime;
1996 if (body->mbo_valid & OBD_MD_FLCTIME) {
1997 if (body->mbo_ctime > inode->i_ctime.tv_sec)
1998 inode->i_ctime.tv_sec = body->mbo_ctime;
1999 lli->lli_ctime = body->mbo_ctime;
2002 /* Clear i_flags to remove S_NOSEC before permissions are updated */
2003 if (body->mbo_valid & OBD_MD_FLFLAGS)
2004 ll_update_inode_flags(inode, body->mbo_flags);
2005 if (body->mbo_valid & OBD_MD_FLMODE)
2006 inode->i_mode = (inode->i_mode & S_IFMT) |
2007 (body->mbo_mode & ~S_IFMT);
2009 if (body->mbo_valid & OBD_MD_FLTYPE)
2010 inode->i_mode = (inode->i_mode & ~S_IFMT) |
2011 (body->mbo_mode & S_IFMT);
2013 LASSERT(inode->i_mode != 0);
2014 if (S_ISREG(inode->i_mode))
2015 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
2016 LL_MAX_BLKSIZE_BITS);
2018 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
2020 if (body->mbo_valid & OBD_MD_FLUID)
2021 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
2022 if (body->mbo_valid & OBD_MD_FLGID)
2023 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
2024 if (body->mbo_valid & OBD_MD_FLPROJID)
2025 lli->lli_projid = body->mbo_projid;
2026 if (body->mbo_valid & OBD_MD_FLNLINK)
2027 set_nlink(inode, body->mbo_nlink);
2028 if (body->mbo_valid & OBD_MD_FLRDEV)
2029 inode->i_rdev = old_decode_dev(body->mbo_rdev);
2031 if (body->mbo_valid & OBD_MD_FLID) {
2032 /* FID shouldn't be changed! */
2033 if (fid_is_sane(&lli->lli_fid)) {
2034 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
2035 "Trying to change FID "DFID
2036 " to the "DFID", inode "DFID"(%p)\n",
2037 PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
2038 PFID(ll_inode2fid(inode)), inode);
2040 lli->lli_fid = body->mbo_fid1;
2044 LASSERT(fid_seq(&lli->lli_fid) != 0);
2046 if (body->mbo_valid & OBD_MD_FLSIZE) {
2047 i_size_write(inode, body->mbo_size);
2049 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
2050 PFID(ll_inode2fid(inode)),
2051 (unsigned long long)body->mbo_size);
2053 if (body->mbo_valid & OBD_MD_FLBLOCKS)
2054 inode->i_blocks = body->mbo_blocks;
2057 if (body->mbo_valid & OBD_MD_TSTATE) {
2058 /* Set LLIF_FILE_RESTORING if restore ongoing and
2059 * clear it when done to ensure to start again
2060 * glimpsing updated attrs
2062 if (body->mbo_t_state & MS_RESTORE)
2063 ll_file_set_flag(lli, LLIF_FILE_RESTORING);
2065 ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
2071 int ll_read_inode2(struct inode *inode, void *opaque)
2073 struct lustre_md *md = opaque;
2074 struct ll_inode_info *lli = ll_i2info(inode);
2078 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2079 PFID(&lli->lli_fid), inode);
2081 /* Core attributes from the MDS first. This is a new inode, and
2082 * the VFS doesn't zero times in the core inode so we have to do
2083 * it ourselves. They will be overwritten by either MDS or OST
2084 * attributes - we just need to make sure they aren't newer.
2086 inode->i_mtime.tv_sec = 0;
2087 inode->i_atime.tv_sec = 0;
2088 inode->i_ctime.tv_sec = 0;
2090 rc = ll_update_inode(inode, md);
2094 /* OIDEBUG(inode); */
2096 #ifdef HAVE_BACKING_DEV_INFO
2097 /* initializing backing dev info. */
2098 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2100 if (S_ISREG(inode->i_mode)) {
2101 struct ll_sb_info *sbi = ll_i2sbi(inode);
2102 inode->i_op = &ll_file_inode_operations;
2103 inode->i_fop = sbi->ll_fop;
2104 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2106 } else if (S_ISDIR(inode->i_mode)) {
2107 inode->i_op = &ll_dir_inode_operations;
2108 inode->i_fop = &ll_dir_operations;
2110 } else if (S_ISLNK(inode->i_mode)) {
2111 inode->i_op = &ll_fast_symlink_inode_operations;
2114 inode->i_op = &ll_special_inode_operations;
2116 init_special_inode(inode, inode->i_mode,
2125 void ll_delete_inode(struct inode *inode)
2127 struct ll_inode_info *lli = ll_i2info(inode);
2128 struct address_space *mapping = &inode->i_data;
2129 unsigned long nrpages;
2132 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2133 /* It is last chance to write out dirty pages,
2134 * otherwise we may lose data while umount */
2135 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
2137 truncate_inode_pages_final(mapping);
2139 /* Workaround for LU-118: Note nrpages may not be totally updated when
2140 * truncate_inode_pages() returns, as there can be a page in the process
2141 * of deletion (inside __delete_from_page_cache()) in the specified
2142 * range. Thus mapping->nrpages can be non-zero when this function
2143 * returns even after truncation of the whole mapping. Only do this if
2144 * npages isn't already zero.
2146 nrpages = mapping->nrpages;
2148 spin_lock_irq(&mapping->tree_lock);
2149 nrpages = mapping->nrpages;
2150 spin_unlock_irq(&mapping->tree_lock);
2151 } /* Workaround end */
2153 LASSERTF(nrpages == 0, "%s: inode="DFID"(%p) nrpages=%lu, "
2154 "see https://jira.whamcloud.com/browse/LU-118\n",
2155 ll_i2sbi(inode)->ll_fsname,
2156 PFID(ll_inode2fid(inode)), inode, nrpages);
2158 #ifdef HAVE_SBOPS_EVICT_INODE
2159 ll_clear_inode(inode);
2166 int ll_iocontrol(struct inode *inode, struct file *file,
2167 unsigned int cmd, unsigned long arg)
2169 struct ll_sb_info *sbi = ll_i2sbi(inode);
2170 struct ptlrpc_request *req = NULL;
2175 case FS_IOC_GETFLAGS: {
2176 struct mdt_body *body;
2177 struct md_op_data *op_data;
2179 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2180 0, 0, LUSTRE_OPC_ANY,
2182 if (IS_ERR(op_data))
2183 RETURN(PTR_ERR(op_data));
2185 op_data->op_valid = OBD_MD_FLFLAGS;
2186 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2187 ll_finish_md_op_data(op_data);
2189 CERROR("%s: failure inode "DFID": rc = %d\n",
2190 sbi->ll_md_exp->exp_obd->obd_name,
2191 PFID(ll_inode2fid(inode)), rc);
2195 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2197 flags = body->mbo_flags;
2199 ptlrpc_req_finished(req);
2201 RETURN(put_user(flags, (int __user *)arg));
2203 case FS_IOC_SETFLAGS: {
2205 struct md_op_data *op_data;
2206 struct cl_object *obj;
2207 struct fsxattr fa = { 0 };
2209 if (get_user(flags, (int __user *)arg))
2212 fa.fsx_projid = ll_i2info(inode)->lli_projid;
2213 if (flags & LUSTRE_PROJINHERIT_FL)
2214 fa.fsx_xflags = FS_XFLAG_PROJINHERIT;
2216 rc = ll_ioctl_check_project(inode, &fa);
2220 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2221 LUSTRE_OPC_ANY, NULL);
2222 if (IS_ERR(op_data))
2223 RETURN(PTR_ERR(op_data));
2225 op_data->op_attr_flags = flags;
2226 op_data->op_xvalid |= OP_XVALID_FLAGS;
2227 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2228 ll_finish_md_op_data(op_data);
2229 ptlrpc_req_finished(req);
2233 ll_update_inode_flags(inode, flags);
2235 obj = ll_i2info(inode)->lli_clob;
2239 OBD_ALLOC_PTR(attr);
2243 rc = cl_setattr_ost(obj, attr, OP_XVALID_FLAGS, flags);
2255 int ll_flush_ctx(struct inode *inode)
2257 struct ll_sb_info *sbi = ll_i2sbi(inode);
2259 CDEBUG(D_SEC, "flush context for user %d\n",
2260 from_kuid(&init_user_ns, current_uid()));
2262 obd_set_info_async(NULL, sbi->ll_md_exp,
2263 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2265 obd_set_info_async(NULL, sbi->ll_dt_exp,
2266 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2271 /* umount -f client means force down, don't save state */
2272 void ll_umount_begin(struct super_block *sb)
2274 struct ll_sb_info *sbi = ll_s2sbi(sb);
2275 struct obd_device *obd;
2276 struct obd_ioctl_data *ioc_data;
2277 struct l_wait_info lwi;
2278 wait_queue_head_t waitq;
2281 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2282 sb->s_count, atomic_read(&sb->s_active));
2284 obd = class_exp2obd(sbi->ll_md_exp);
2286 CERROR("Invalid MDC connection handle %#llx\n",
2287 sbi->ll_md_exp->exp_handle.h_cookie);
2293 obd = class_exp2obd(sbi->ll_dt_exp);
2295 CERROR("Invalid LOV connection handle %#llx\n",
2296 sbi->ll_dt_exp->exp_handle.h_cookie);
2302 OBD_ALLOC_PTR(ioc_data);
2304 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2305 sizeof *ioc_data, ioc_data, NULL);
2307 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2308 sizeof *ioc_data, ioc_data, NULL);
2310 OBD_FREE_PTR(ioc_data);
2313 /* Really, we'd like to wait until there are no requests outstanding,
2314 * and then continue. For now, we just periodically checking for vfs
2315 * to decrement mnt_cnt and hope to finish it within 10sec.
2317 init_waitqueue_head(&waitq);
2318 lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(10),
2319 cfs_time_seconds(1), NULL, NULL);
2320 l_wait_event(waitq, may_umount(sbi->ll_mnt.mnt), &lwi);
2325 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2327 struct ll_sb_info *sbi = ll_s2sbi(sb);
2328 char *profilenm = get_profile_name(sb);
2332 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2333 read_only = *flags & MS_RDONLY;
2334 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2335 sizeof(KEY_READ_ONLY),
2336 KEY_READ_ONLY, sizeof(read_only),
2339 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2340 profilenm, read_only ?
2341 "read-only" : "read-write", err);
2346 sb->s_flags |= MS_RDONLY;
2348 sb->s_flags &= ~MS_RDONLY;
2350 if (sbi->ll_flags & LL_SBI_VERBOSE)
2351 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2352 read_only ? "read-only" : "read-write");
2358 * Cleanup the open handle that is cached on MDT-side.
2360 * For open case, the client side open handling thread may hit error
2361 * after the MDT grant the open. Under such case, the client should
2362 * send close RPC to the MDT as cleanup; otherwise, the open handle
2363 * on the MDT will be leaked there until the client umount or evicted.
2365 * In further, if someone unlinked the file, because the open handle
2366 * holds the reference on such file/object, then it will block the
2367 * subsequent threads that want to locate such object via FID.
2369 * \param[in] sb super block for this file-system
2370 * \param[in] open_req pointer to the original open request
2372 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2374 struct mdt_body *body;
2375 struct md_op_data *op_data;
2376 struct ptlrpc_request *close_req = NULL;
2377 struct obd_export *exp = ll_s2sbi(sb)->ll_md_exp;
2380 body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2381 OBD_ALLOC_PTR(op_data);
2382 if (op_data == NULL) {
2383 CWARN("%s: cannot allocate op_data to release open handle for "
2384 DFID"\n", ll_s2sbi(sb)->ll_fsname, PFID(&body->mbo_fid1));
2389 op_data->op_fid1 = body->mbo_fid1;
2390 op_data->op_open_handle = body->mbo_open_handle;
2391 op_data->op_mod_time = ktime_get_real_seconds();
2392 md_close(exp, op_data, NULL, &close_req);
2393 ptlrpc_req_finished(close_req);
2394 ll_finish_md_op_data(op_data);
2399 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2400 struct super_block *sb, struct lookup_intent *it)
2402 struct ll_sb_info *sbi = NULL;
2403 struct lustre_md md = { NULL };
2407 LASSERT(*inode || sb);
2408 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2409 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2410 sbi->ll_md_exp, &md);
2415 rc = ll_update_inode(*inode, &md);
2419 LASSERT(sb != NULL);
2422 * At this point server returns to client's same fid as client
2423 * generated for creating. So using ->fid1 is okay here.
2425 if (!fid_is_sane(&md.body->mbo_fid1)) {
2426 CERROR("%s: Fid is insane "DFID"\n",
2428 PFID(&md.body->mbo_fid1));
2429 GOTO(out, rc = -EINVAL);
2432 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2433 sbi->ll_flags & LL_SBI_32BIT_API),
2435 if (IS_ERR(*inode)) {
2436 #ifdef CONFIG_FS_POSIX_ACL
2438 posix_acl_release(md.posix_acl);
2439 md.posix_acl = NULL;
2442 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2444 CERROR("new_inode -fatal: rc %d\n", rc);
2449 /* Handling piggyback layout lock.
2450 * Layout lock can be piggybacked by getattr and open request.
2451 * The lsm can be applied to inode only if it comes with a layout lock
2452 * otherwise correct layout may be overwritten, for example:
2453 * 1. proc1: mdt returns a lsm but not granting layout
2454 * 2. layout was changed by another client
2455 * 3. proc2: refresh layout and layout lock granted
2456 * 4. proc1: to apply a stale layout */
2457 if (it != NULL && it->it_lock_mode != 0) {
2458 struct lustre_handle lockh;
2459 struct ldlm_lock *lock;
2461 lockh.cookie = it->it_lock_handle;
2462 lock = ldlm_handle2lock(&lockh);
2463 LASSERT(lock != NULL);
2464 if (ldlm_has_layout(lock)) {
2465 struct cl_object_conf conf;
2467 memset(&conf, 0, sizeof(conf));
2468 conf.coc_opc = OBJECT_CONF_SET;
2469 conf.coc_inode = *inode;
2470 conf.coc_lock = lock;
2471 conf.u.coc_layout = md.layout;
2472 (void)ll_layout_conf(*inode, &conf);
2474 LDLM_LOCK_PUT(lock);
2480 /* cleanup will be done if necessary */
2481 md_free_lustre_md(sbi->ll_md_exp, &md);
2483 if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2484 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2489 int ll_obd_statfs(struct inode *inode, void __user *arg)
2491 struct ll_sb_info *sbi = NULL;
2492 struct obd_export *exp;
2494 struct obd_ioctl_data *data = NULL;
2498 if (!inode || !(sbi = ll_i2sbi(inode)))
2499 GOTO(out_statfs, rc = -EINVAL);
2501 rc = obd_ioctl_getdata(&buf, &len, arg);
2503 GOTO(out_statfs, rc);
2506 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2507 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2508 GOTO(out_statfs, rc = -EINVAL);
2510 if (data->ioc_inllen1 != sizeof(__u32) ||
2511 data->ioc_inllen2 != sizeof(__u32) ||
2512 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2513 data->ioc_plen2 != sizeof(struct obd_uuid))
2514 GOTO(out_statfs, rc = -EINVAL);
2516 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2517 if (type & LL_STATFS_LMV)
2518 exp = sbi->ll_md_exp;
2519 else if (type & LL_STATFS_LOV)
2520 exp = sbi->ll_dt_exp;
2522 GOTO(out_statfs, rc = -ENODEV);
2524 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2526 GOTO(out_statfs, rc);
2528 OBD_FREE_LARGE(buf, len);
2533 * this is normally called in ll_fini_md_op_data(), but sometimes it needs to
2534 * be called early to avoid deadlock.
2536 void ll_unlock_md_op_lsm(struct md_op_data *op_data)
2538 if (op_data->op_mea2_sem) {
2539 up_read(op_data->op_mea2_sem);
2540 op_data->op_mea2_sem = NULL;
2543 if (op_data->op_mea1_sem) {
2544 up_read(op_data->op_mea1_sem);
2545 op_data->op_mea1_sem = NULL;
2549 /* this function prepares md_op_data hint for passing it down to MD stack. */
2550 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2551 struct inode *i1, struct inode *i2,
2552 const char *name, size_t namelen,
2553 __u32 mode, __u32 opc, void *data)
2555 LASSERT(i1 != NULL);
2558 /* Do not reuse namelen for something else. */
2560 return ERR_PTR(-EINVAL);
2562 if (namelen > ll_i2sbi(i1)->ll_namelen)
2563 return ERR_PTR(-ENAMETOOLONG);
2565 if (!lu_name_is_valid_2(name, namelen))
2566 return ERR_PTR(-EINVAL);
2569 if (op_data == NULL)
2570 OBD_ALLOC_PTR(op_data);
2572 if (op_data == NULL)
2573 return ERR_PTR(-ENOMEM);
2575 ll_i2gids(op_data->op_suppgids, i1, i2);
2576 op_data->op_fid1 = *ll_inode2fid(i1);
2577 op_data->op_default_stripe_offset = -1;
2579 if (S_ISDIR(i1->i_mode)) {
2580 down_read(&ll_i2info(i1)->lli_lsm_sem);
2581 op_data->op_mea1_sem = &ll_i2info(i1)->lli_lsm_sem;
2582 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2583 if (opc == LUSTRE_OPC_MKDIR)
2584 op_data->op_default_stripe_offset =
2585 ll_i2info(i1)->lli_def_stripe_offset;
2589 op_data->op_fid2 = *ll_inode2fid(i2);
2590 if (S_ISDIR(i2->i_mode)) {
2592 down_read(&ll_i2info(i2)->lli_lsm_sem);
2593 op_data->op_mea2_sem =
2594 &ll_i2info(i2)->lli_lsm_sem;
2596 op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2599 fid_zero(&op_data->op_fid2);
2602 if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2603 op_data->op_cli_flags |= CLI_HASH64;
2605 if (ll_need_32bit_api(ll_i2sbi(i1)))
2606 op_data->op_cli_flags |= CLI_API32;
2608 op_data->op_name = name;
2609 op_data->op_namelen = namelen;
2610 op_data->op_mode = mode;
2611 op_data->op_mod_time = ktime_get_real_seconds();
2612 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2613 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2614 op_data->op_cap = cfs_curproc_cap_pack();
2615 op_data->op_mds = 0;
2616 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2617 filename_is_volatile(name, namelen, &op_data->op_mds)) {
2618 op_data->op_bias |= MDS_CREATE_VOLATILE;
2620 op_data->op_data = data;
2625 void ll_finish_md_op_data(struct md_op_data *op_data)
2627 ll_unlock_md_op_lsm(op_data);
2628 security_release_secctx(op_data->op_file_secctx,
2629 op_data->op_file_secctx_size);
2630 OBD_FREE_PTR(op_data);
2633 #ifdef HAVE_SUPEROPS_USE_DENTRY
2634 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2636 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2639 struct ll_sb_info *sbi;
2641 #ifdef HAVE_SUPEROPS_USE_DENTRY
2642 LASSERT((seq != NULL) && (dentry != NULL));
2643 sbi = ll_s2sbi(dentry->d_sb);
2645 LASSERT((seq != NULL) && (vfs != NULL));
2646 sbi = ll_s2sbi(vfs->mnt_sb);
2649 if (sbi->ll_flags & LL_SBI_NOLCK)
2650 seq_puts(seq, ",nolock");
2652 if (sbi->ll_flags & LL_SBI_FLOCK)
2653 seq_puts(seq, ",flock");
2655 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2656 seq_puts(seq, ",localflock");
2658 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2659 seq_puts(seq, ",user_xattr");
2661 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2662 seq_puts(seq, ",lazystatfs");
2664 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2665 seq_puts(seq, ",user_fid2path");
2667 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2668 seq_puts(seq, ",always_ping");
2674 * Get obd name by cmd, and copy out to user space
2676 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2678 struct ll_sb_info *sbi = ll_i2sbi(inode);
2679 struct obd_device *obd;
2682 if (cmd == OBD_IOC_GETDTNAME)
2683 obd = class_exp2obd(sbi->ll_dt_exp);
2684 else if (cmd == OBD_IOC_GETMDNAME)
2685 obd = class_exp2obd(sbi->ll_md_exp);
2692 if (copy_to_user((void __user *)arg, obd->obd_name,
2693 strlen(obd->obd_name) + 1))
2699 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2706 p.mnt = current->fs->root.mnt;
2708 path = d_path(&p, buf, bufsize);
2713 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2715 char *buf, *path = NULL;
2716 struct dentry *dentry = NULL;
2717 struct inode *inode = page->mapping->host;
2719 /* this can be called inside spin lock so use GFP_ATOMIC. */
2720 buf = (char *)__get_free_page(GFP_ATOMIC);
2722 dentry = d_find_alias(page->mapping->host);
2724 path = ll_d_path(dentry, buf, PAGE_SIZE);
2728 "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2729 "(rc %d)\n", ll_i2sbi(inode)->ll_fsname,
2730 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2731 PFID(ll_inode2fid(inode)),
2732 (path && !IS_ERR(path)) ? path : "", ioret);
2738 free_page((unsigned long)buf);
2741 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2742 struct lov_user_md **kbuf)
2744 struct lov_user_md lum;
2748 if (copy_from_user(&lum, md, sizeof(lum)))
2751 lum_size = ll_lov_user_md_size(&lum);
2755 OBD_ALLOC(*kbuf, lum_size);
2759 if (copy_from_user(*kbuf, md, lum_size) != 0) {
2760 OBD_FREE(*kbuf, lum_size);
2768 * Compute llite root squash state after a change of root squash
2769 * configuration setting or add/remove of a lnet nid
2771 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2773 struct root_squash_info *squash = &sbi->ll_squash;
2776 struct lnet_process_id id;
2778 /* Update norootsquash flag */
2779 down_write(&squash->rsi_sem);
2780 if (list_empty(&squash->rsi_nosquash_nids))
2781 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2783 /* Do not apply root squash as soon as one of our NIDs is
2784 * in the nosquash_nids list */
2787 while (LNetGetId(i++, &id) != -ENOENT) {
2788 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2790 if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2796 sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2798 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2800 up_write(&squash->rsi_sem);
2804 * Parse linkea content to extract information about a given hardlink
2806 * \param[in] ldata - Initialized linkea data
2807 * \param[in] linkno - Link identifier
2808 * \param[out] parent_fid - The entry's parent FID
2809 * \param[out] ln - Entry name destination buffer
2811 * \retval 0 on success
2812 * \retval Appropriate negative error code on failure
2814 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2815 struct lu_fid *parent_fid, struct lu_name *ln)
2821 rc = linkea_init_with_rec(ldata);
2825 if (linkno >= ldata->ld_leh->leh_reccount)
2826 /* beyond last link */
2829 linkea_first_entry(ldata);
2830 for (idx = 0; ldata->ld_lee != NULL; idx++) {
2831 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2836 linkea_next_entry(ldata);
2846 * Get parent FID and name of an identified link. Operation is performed for
2847 * a given link number, letting the caller iterate over linkno to list one or
2848 * all links of an entry.
2850 * \param[in] file - File descriptor against which to perform the operation
2851 * \param[in,out] arg - User-filled structure containing the linkno to operate
2852 * on and the available size. It is eventually filled with
2853 * the requested information or left untouched on error
2855 * \retval - 0 on success
2856 * \retval - Appropriate negative error code on failure
2858 int ll_getparent(struct file *file, struct getparent __user *arg)
2860 struct inode *inode = file_inode(file);
2861 struct linkea_data *ldata;
2862 struct lu_buf buf = LU_BUF_NULL;
2864 struct lu_fid parent_fid;
2871 if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2872 !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2875 if (get_user(name_size, &arg->gp_name_size))
2878 if (get_user(linkno, &arg->gp_linkno))
2881 if (name_size > PATH_MAX)
2884 OBD_ALLOC(ldata, sizeof(*ldata));
2888 rc = linkea_data_new(ldata, &buf);
2890 GOTO(ldata_free, rc);
2892 #ifdef HAVE_XATTR_HANDLER_FLAGS
2893 rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
2894 buf.lb_len, OBD_MD_FLXATTR);
2896 rc = ll_getxattr(file_dentry(file), XATTR_NAME_LINK, buf.lb_buf,
2898 #endif /* HAVE_XATTR_HANDLER_FLAGS */
2902 rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2906 if (ln.ln_namelen >= name_size)
2907 GOTO(lb_free, rc = -EOVERFLOW);
2909 if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2910 GOTO(lb_free, rc = -EFAULT);
2912 if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2913 GOTO(lb_free, rc = -EFAULT);
2915 if (put_user('\0', arg->gp_name + ln.ln_namelen))
2916 GOTO(lb_free, rc = -EFAULT);
2921 OBD_FREE(ldata, sizeof(*ldata));